Автор: Félix Y.   Halperin S.   Thomas J.-C.  

Теги: rational homotopy theory   homotopy  

ISBN: 0-387-95068-0

Год: 2001

Текст
                    Graduate Texts in Mathematics
1 Takeuti/Zaring. Introduction to
Axiomatic Set Theory. 2nd ed.
2 Oxtoby. Measure and Category. 2nd ed.
3 Schaefer. Topological Vector Spaces.
2nd ed.
4 Hilton/Stammbach. A Course in
Homological Algebra. 2nd ed.
5 Mac Lane. Categories for the Working
Mathematician. 2nd ed.
6 Hughes/Piper. Projective Planes.
7 Serre. A Course in Arithmetic.
8 Takeuti/Zaring. Axiomatic Set Theory.
9 Humphreys, introduction to Lie Algebras
and Representation Theory.
10 Cohen. A Course in Simple Homotopy
Theory.
11 Conway. Functions of One Complex
Variable I. 2nd ed.
12 Beals. Advanced Mathematical Analysis.
13 Anderson/Fuller. Rings and Categories
of Modules. 2nd ed.
14 Golubitsky/Guillemin. Stable Mappings
and Their Singularities.
15 Berberian. Lectures in Functional
Analysis and Operator Theory.
16 Winter. The Structure of Fields.
17 Rosenblatt. Random Processes. 2nd ed.
18 Halmos. Measure Theory.
19 Halmos. A Hubert Space Problem Book.
2nd ed.
20 Husemoller. Fibre Bundles. 3rd ed.
21 Humphreys. Linear Algebraic Groups.
22 Barnes/Mack. An Algebraic Introduction
to Mathematical Logic.
23 Greub. Linear Algebra. 4th ed.
24 Holmes. Geometric Functional Analysis
and Its Applications.
25 Hewitt/Stromberg. Real and Abstract
Analysis.
26 Manes. Algebraic Theories.
27 Kelley. General Topology.
28 Zariski/Samuel. Commutative Algebra.
Vol.1.
29 Zariskj/Samuel. Commutative Algebra.
Vol.11.
30 Jacobson. Lectures in Abstract Algebra I.
Basic Concepts.
31 Jacobson. Lectures in Abstract Algebra II.
Linear Algebra.
32 Jacobson. Lectures in Abstract Algebra
III. Theory of Fields and Galois Theory.
33 Hirsch. Differential Topology.
34 Spitzer. Principles of Random Walk.
2nd ed.
35 Alexander/Wermer. Several Complex
Variables and Banach Algebras. 3rd ed.
3 6 Kelley/Namioka et al. Linear Topologi cal
Spaces.
37 Monk. Mathematical Logic.
38 Grauert/Fritzsche. Several Complex
Variables.
39 Arveson. An Invitation to C*-Algebras.
40 Kemeny/Snell/Knapp. Denumerable
Markov Chains. 2nd ed.
41 Apostol. Modular Functions and
Dirichlet Series in Number Theory.
2nd ed.
42 Serre. Linear Representations of Finite
Groups.
43 Gillman/Jerjson. Rings of Continuous
Functions.
44 Kendig. Elementary Algebraic Geometry.
45 Loeve. Probability Theory I. 4th ed.
46 Loeve. Probability Theory II. 4th ed.
47 Moise. Geometric Topology in
Dimensions 2 and 3.
48 Sachs/Wu. General Relativity for
Mathematicians.
49 Gruenberg/Weir. Linear Geometry.
2nd ed.
50 Edwards. Fermat's Last Theorem.
51 Kxingenberg A Course in Differential
Geometry.
52 Hartshorne. Algebraic Geometry.
53 M.wiN. A Course in Mathematical Logic.
54 Graver/Watkjns. Combinatorics with
Emphasis on the Theory of Graphs.
55 Brown/Pearcy. Introduction to Operator
Theory I: Elements of Functional Analysis.
56 Massey. Algebraic Topology: An
Introduction.
57 Crowell/Fox. Introduction to Knot
Theory.
58 Koblitz. p-adic Numbers, p-adic
Analysis, and Zeta-Functions. 2nd ed.
59 Lang. Cyclotomic Fields.
60 Arnold. Mathematical Methods in
Classical Mechanics. 2nd ed.
61 Whttehead. Elements of Homotopy
Theory.
62 Kargapolov/Merlzjakov. Fundamentals
of the Theory of Groups.
63 Bollobas. Graph Theory.
64 Edwards. Fourier Series. Vol. I. 2nd ed.
(continued after index)


Yves Felix Stephen Halperin Jean-Claude Thomas Rational Homotopy Theory
Yves Felix Institut Mathematiques Universite de Louvain La Neuve 2 chemin du Cyclotron Louvain-la-Neuve, B-1348 Belgium Jean-Claude Thomas Faculte des Sciences Universite d'Angers 2 bd Lavoisier Angers 49045 France Editorial Board S. Axler Mathematics Department San Francisco State University San Francisco, CA 94132 USA Stephen Halperin College of Computer, Mathematical, and Physical Science University of Maryland 3400 A.V. Williams Building College Park, MD 20742-3281 USA F.W. Gehring Mathematics Department East Hall University of Michigan Ann Arbor, MI 48109 USA K.A. Ribet Mathematics Department University of California at Berkeley Berkeley, CA 94720-3840 USA Mathematics Subject Classification B000): 55-01, 55P62 Library of Congress Cataloging-in-Publication Data Felix, Y. (Yves) Rational homotopy theory / Yves Felix, Stephen Halperin, Jean-Claude Thomas p. cm. — (Graduate texts in mathematics ; 205) Includes bibliographical references and index. ISBN 0-387-95068-0 (alk. paper) 1. Homotopy theory I. Halperin, Stephen. II. Thomas, J-C. (Jean-Claude) III. Title. IV. Series. QA612.7 .F46 2000 514'24—dc21 00-041913 Printed on acid-free paper. © 2001 Springer-Verlag New York, Inc. All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer-Verlag New York, Inc., 175 Fifth Avenue, New York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use of general descriptive names, trade names, trademarks, etf., in this publication, even if the former are not especially identified, is not to be taken as a sign that such names, as understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone. Production managed by Timothy Taylor; manufacturing supervised by Jacqui Ashri. Camera-ready copy prepared from the authors' PDF files. Printed and bound by R.R. Donnelley and Sons, Harrisonburg, VA. Printed in the United States of America. 987654321 ISBN 0-387-95068-0 SPIN 10770738 - · -"-?.. m»„, YnrW Rerlin Heidelberg
to Agnes Danielle Janet
Introduction Homotopy theory is the study of the invariants and properties of topological spaces X and continuous maps / that depend only on the homotopy type of the space and the homotopy class of the map. (We recall that two continuous maps /, g : X —> Y are homotopic (/ ~ g) if there is a continuous map F : Xxl -,—> Y such that F(x,0) = f(x) and F(x, 1) = g(x). Two topological spaces X and ? have the same homotopy type if there are continuous maps X ( ) Y such that 9 fg ~ idy and gf ~ ???.) The classical examples of such invariants are the singular homology groups Hi(X) and the homotopy groups ??(?"), the latter consisting of the homotopy classes of maps (Sn,*) —>· (X,x0). Invariants such as these play an essential role in the geometric and analytic behavior of spaces and maps. The groups H{(X) and ??(?), ? > 2, are abelian and hence can be rational- ized to the vector spaces Hi(X;Q) and ??(?) ® Q. Rational homotopy theory begins with the discovery by Sullivan in the 1960's of an underlying geomet- ric construction: simply connected topological spaces and continuous maps be- tween them can themselves be rationalized to topological spaces Xq and to maps /q : Xq —»· Yq, such that H.(XQ) = H*(X;Q) and n,(XQ) = ?*(?) ® Q. The rational homotopy type of a CW complex X is the homotopy type of Xq and the rational homotopy class of / : X —>· Y is the homotopy class of /q : Xq —»· 1q, and rational homotopy theory is then the study of properties that depend only on the rational homotopy type of a space or the rational homotopy class of a map. Rational homotopy theory has the disadvantage of discarding a considerable amount of information. For example, the homotopy groups of the sphere S2 are non-zero in infinitely many degrees whereas its rational homotopy groups vanish in all degrees above 3. By contrast, rational homotopy theory has the advantage of being remarkably computational. For example, there is not even a conjectural description of all the homotopy groups of any simply connected finite CW complex, whereas for many of these the rational groups can be explicitly determined. And while rational homotopy theory is indeed simpler than ordinary homotopy theory, it is exactly this simplicity that makes it possible to address (if not always to solve) a number of fundamental questions. This is illustrated by two early successes: • (Vigue-Sullivan [152]) // M is a simply connected compact riemannian manifold whose rational cohomology algebra requires at least two generators then its free loop space has unbounded homology and hence (Gromoll-Meyer [73],) M has infinitely many geometrically distinct closed geodesies. • (Allday-Halperin [3]) // an r torus acts freely on a homogeneous space G/H (G and H compact Lie groups) then ? <rankG-rankii ,
viii Introduction as well as by the list of open problems in the final section of this monograph. The computational power of rational homotopy theory is due to the discovery by Quillen [135] and by Sullivan [144] of an explicit algebraic formulation. In each case the rational homotopy type of a topological space is the same as the isomorphism class of its algebraic model and the rational homotopy type of a continuous map is the same as the algebraic homotopy class of the correspond- ing morphism between models. These models make the rational homology and homotopy of a space transparent. They also (in principle, always, and in prac- tice, sometimes) enable the calculation of other homotopy invariants such as the cup product in cohomology, the Whitehead product in homotopy and rational Lusternik-Schnirelmann category. In its initial phase research in rational homotopy theory focused on the identi- fication of rational homotopy invariants in terms of these models. These included the homotopy Lie algebra (the translation of the Whitehead product to the homo- topy groups of the loop space ??" under the isomorphism tt*+i(X) = ?*(?.?)). LS category and cone length. Since then, however, work has concentrated on the properties of these in- variants, and has uncovered some truly remarkable, and previously unsuspected phenomena. For example • If X is an ?-dimensional simply connected finite CW complex, then either its rational homotopy groups vanish in degrees > 2n, or else they grow exponentially. • Moreover, in the second case any interval (k, k + n) contains an integer i such that iti(X) 0 Q ? 0. • Again in the second case the sum of all the solvable ideals in the homotopy Lie algebra is a finite dimensional ideal R, and dim ?even 5; cat -^Q · • Again in the second case for all elements a G iieVf,n(Q.X) ® <Q> of sufficiently high degree there is some ? G ?,.(??')®(?! such that the iterated Lie brackets [a, [a,..., [?, ?] ¦ ¦ · ]] are all non-zero. • Finally, rational LS category satisfies the product formula cat(X<Q ? Yq) = catXQ -f catYq , in sharp contrast with what happens in the 'non-rational' case. The first bullet divides all simply connected finite CW complexes X into two groups: the rationally elliptic spaces whose rational homotopy is finite dimen- sional, and the rationally hyperbolic spaces whose rational homotopy grows ex- ponentially. Moreover, because H*(fiX;Q) is the universal enveloping algebra ¦ v . —j„j T .„ „1n.oKra Tiv _ ?jftX") ®Q, it follows from the first two bullets
Rational Homotopy Theory ix that whether X is rationally elliptic or rationally hyperbolic can be determined from the numbers b{ = dim Hi(flX;Q), 1 < i < 3n — 3, where ? = dim .Y. Rationally elliptic spaces include Lie groups, homogeneous spaces, manifolds supporting a codimension one action and Dupin hypersurfaces (for the last two see [77]). However, the 'generic' finite CW complex is rationally hyperbolic. The theory of Sullivan replaces spaces with algebraic models, and it is exten- sive calculations and experimentation with these models that has led to much of the progress summarized in these results. More recently the fundamental article of Anick [11] has made impossible to extend these techniques for finite CW com- plexes to coefficients ? (-^-,..., ?- ) with only finitely many primes invested, and thereby to obtain analogous results for H*(UX; Wp) for large primes p. Moreover, the rational results originally obtained via Sullivan models often suggest possible extensions beyond the rational realm. An example is the 'depth theorem' origi- nally proved in [54] via Sullivan models and established in this monograph (§35) topologically for any coefficients. This extension makes it possible to generalize many of the results on loop space homology to completely arbitrary coefficients. However, for reasons of space and simplicity, in this monograph we have re- stricted ourselves to rational homotopy theory itself. Thus our monograph has three main objectives: • To provide a coherent, self-contained, reasonably complete and usable de- scription of the tools and techniques of rational homotopy theory. • To provide an account of many of the main structural theorems with proofs that are often new and/or considerably simplified from the original versions in the literature. • To illustrate both the use of the technology, and the consequences of the theorems in a rich variety of examples. We have written this monograph for graduate students who have already en- countered the fundamental group and singular homology, although our hope is that the results described will be accessible to interested mathematicians in other parts of the subject and that our rational homotopy colleagues may also find it useful. To help keep the text more accessible we have adopted a number of simplifying strategies: - coefficients are usually restricted to fields k of characteristic zero. - topological spaces are usually restricted to be simply connected. - Sullivan models for spaces (and their properties) are derived first and only- then extended to the more general case of fibrations, rather than being deduced from the latter as a special case. - complex diagrams and proofs by diagram chase are almost always avoided.
x Introduction Of course this has meant, in particular, that theorems and technology are not always established in the greatest possible generality, but the resulting saving in technical complexity is considerable. It should also be emphasized that this is a monograph about topological spaces. This is important, because the models themselves at the core of the subject are strictly algebraic and indeed we have been careful to define them and establish their properties in purely algebraic terms. The reader who needs the machin- ery for application in other contexts (for instance local commutative algebra) will find it presented here. However the examples and applications throughout are drawn largely from topology, and we have not hesitated to use geometric constructions and techniques when this seemed a simpler and more intuitive approach. The algebraic models are, however, at the heart of the material we are pre- senting. They axe all graded objects with a differential as well as an algebraic structure (algebra, Lie algebra, module, ... ). and this reflects an understanding that emerged during the 1960!s. Previously objects with a differential had often been thought of as merely a mechanism to compute homology; we now know that they carry a homotopy theory which is much richer than the homology. For example, if X is a simply connected CW complex of finite type then the work of Adams [1] shows that the homotopy type of the cochain algebra C*(X) is sufficient to calculate the loop space homology H„(UX) which, on the other hand, cannot be computed from the cohomology algebra H*(X). This algebraic homotopy theory is introduced in [134] and studied extensively in [20]. In this monograph there are three differential graded categories that are im- portant: (i) modules over a differential graded algebra (dga), (R,d). (ii) commutative cochain algebras, (iii) differential graded Lie algebras (dgl's). In each case both the algebraic structure and the differential carry information, and in each case there is a fundamental modelling construction which associates to an object A in the category a morphism ? : M —> A such that ?(?) is an isomorphism (? is called a quasi-isomorphism) and such that the algebraic structure in M is, in some sense "free". These models (the cofibrant objects of [134]) are the exact analogue of a free resolution of an arbitrary module over a ring. In our three cases above we find, respectively: (i) A semi-free resolution of a module over (R, d) which is, in particular a m"in]pv of free A-modules.
Rational Homotopy Theory xi (ii) A Sullivan model of a commutative cochain algebra which is a quasi- isomorphism from a commutative cochain algebra that, in particular, is free as a commutative graded algebra. (These cochain algebras are called Sullivan algebras.) (iii) A free Lie model of a dgl, which is a quasi-isomorphism from a dgl that is free as a graded Lie algebra. These models are the main algebraic tools of the subject. The combination of this technology with its application to topological spaces constitutes a formidable body of material. To assist the reader in dealing with this we have divided the monograph into forty sections grouped into six Parts. Each section presents a single aspect of the subject organized into a number of distinct topics, and described in an introduction at the start of the section. The table of contents lists both the titles of the sections and of the individual topics within them. Reading through the table of contents and scanning the introductions to the sections should give the reader an excellent idea of the contents. Here we present an overview of the six Parts, indicating some of the highlights and the role of each Part within the book. Part I: Homotopy Theory, Resolutions for Fibrations and P-local Spaces. This Part is a self-contained short course in homotopy theory. In particular, §0 is merely a summary of definitions and notation from general topology, while §3 is the analogue for (graded) algebra. The text proper begins with the basic geometric objects, CW complexes and fibrations in §1 and §2, and culminates with the rationalization in §9 of a topological space. Since CW complexes and fibrations are often absent from an introductory course in algebraic topology we present their basic properties for the convenience of the reader. In particular, we construct a CW model for any topological space and establish Whitehead's homotopy lifting theorem, since this is the exact geometric analogue, and the motivating example, for the algebraic models referred to above. Then, in §6, we introduce the first of these algebraic models: the semifree resolution of a module over a differential graded algebra. These resolutions are of key importance throughout the text. Now modules over a dga arise naturally in topology in at least two contexts: • /// : X —)¦ Y is a continuous map then the singular cochain algebra C*(X) is a module over C*(Y) via C*(f). • If ? ? G —)¦ X is the action of a topological monoid then the singular chains Ct(X) are a module over the chain algebra C*(G). In §7 we consider the first case when / is a fibration, and use a semifree resolution to compute the cohomology of the fibre (when Y is simply connected
xii Introduction with homology of finite type). In §8 we consider the second case when the action is that of a principal G-fibration X —> Y and use a semifree resolution to compute H*(Y). Both these results are due essentially to J.C. Moore. The second result turns out to give an easy, fast and spectral-sequence-free proof of the Whitehead-Serre theorem that for a continuous map / : X —> Y between simply connected spaces and for Jk C Q, Ht (/; Ik) is an isomorphism if and only if 7r»(/) (g> h is an isomorphism. We have therefore included this as an interesting application, especially as the theorem itself is fundamental to the rationalization of spaces constructed in §9. Aside from these results it is in Part I that we establish the notation and conventions that will be used throughout (particularly in §0—§5) and state the theorems in homotopy theory we will need to quote. Since it turned out that with the definitions and statements in place the proofs could also be included at very little additional cost in space, we indulged ourselves (and perhaps the reader) and included these as well. Part II: Sullivan Models This Part is the core of the monograph, in which we identify the rational homotopy theory of simply connected spaces with the homotopy theory of com- mutative cochain algebras. This occurs in three steps: • The construction in §10 of Sullivan's functor from topological spaces X to commutative cochain algebras Api(X), which satisfies C*(X) ~ Api(X). • The construction in §12 of the Sullivan model (AV, d) A (A, d) for any commutative cochain algebra satisfying H°(A,d) = k. (Here, fol- lowing Sullivan ([144]), and the rest of the rational homotopy literature, AV denotes the free commutative graded algebra on V.) • The construction in §I7 of Sullivan's realization functor which converts a Sullivan algebra, (AV, d), (simply connected and of finite type) into a rational topological space |AV, d| such that (AV, d) is a Sullivan model for APL(\AV,d\). Along the way we show that these functors define bijections: rational homotopy types  =» f isomorphism classes of of spaces J 1 minimal Sullivan algebras and , . , f s ( homotopy classes of homotopy classes of I s I , . · ? , . , > ¦<—*- < maps between minimal maps between rational spaces I Inn· ? ?. ; ( Su livan algebras
Rational Homotopy Theory · x"i where we restrict to spaces and cochain algebras that are simply connected with cohomology of finite type. Sullivan's functor Api was motivated by the classical commutative cochain algebra Adr{M) of smooth differential forms on a manifold. In §11 we review the construction of Adr(M) and prove Sullivan's result that Adr(M) is quasi- isomorphic to Api(M;R). This implies (§12) that they have the same Sullivan model. The rest of Part II is devoted to the technology of Sullivan algebras, and to geometric applications. We construct models of adjunction spaces, identity the generating space V" of a Sullivan model with the dual of the rational homotopy groups and identity the quadratic part of the differential with the dual of the Whitehead product. Here the constructions are in §13 but some of the proofs are deferred to §15. In §14 we construct relative Sullivan algebras and decompose any Sullivan algebra as the tensor product of a minimal and a contractible Sullivan algebra. In §15 we use relative Sullivan algebras to model fibrations and show (applying the result from §7) that the Sullivan fibre of the model is a Sullivan model for the fibre. Finally, in §16 this material is applied to the structure of the homology algebra H*{UX;]k) of the loop space of X. Part III: Graded Differential Algebra (Continued). In §3 we were careful to limit ourselves to those algebraic constructions needed in Parts I and II. Now we need more: the bar construction of a cochain algebra, spectral sequences (finally, we held off as long as possible!) and some elementary homological algebra. Part IV: Lie Models In Part I we introduced the first of our algebraic categories (modules over a dga), in Part II we focused on commutative cochain algebras and now we introduce and study the third category, differential graded Lie algebras. In §21 we introduce graded Lie algebras and their universal enveloping algebras and exhibit the two fundamental examples in this monograph: the homotopy Lie algebra L\ — ?»(??) (g> ifc of a simply connected topological space, and the homotopy Lie algebra L of a minimal Sullivan algebra (i\V, d). The latter vector space is defined by Lk = Horn [Vk+l, k) with Lie bracket given by the quadratic part of d. Moreover, if (AV,d) is the Sullivan model for X then Lx s L. In §22 we construct the free Lie models for a dgl, (L, d). We also construct (in §22 and §23) the classical homotopy equivalences (L,d)->C*(L,d) and {A, d) <~ ?{A4) between the categories of dgl's (with L = L>! of finite type) and commutative cochain algebras (with simply connected cohomology of finite type). In particular a Lie model for a free topological space X is a free Lie model of ?(\v,d)i where {AV, d) is a Sullivan model for X.
xiv Introduction Given a dgl (L, d) that is free as a Lie algebra on generators ?; of degree n; we show in §24 how to construct a CW complex X with a single (m + l)-cell for each V{. and whose free Lie model is exactly (L.d). This provides a much more geometric approach to the passage algebra —> topology then the realization functor in §17. Finally, §24 and §25 are devoted to Majewski's theorem [119] that if (L, d) is a free Lie model for X then there is a chain algebra quasi-isomorphism U(L, d) -^4 C*(QX;Jk) which preserves the diagonals up to dga homotopy. Part V: Rational Lusternik-Schnirelmann Category The LS category, catX, of a topological space X is the smallest number m (or infinity) such that X can be covered by m + 1 open sets each of which is contractible in X. In particular: • caXX is an invariant of the homotopy type of X. • // cat X = m then the product of any m + 1 cohomology classes of X is zero. • If X is a CW complex then catX < dimX but the inequality may be oo strict: indeed for the wedge of spheres X = V Sl we have dim X = oo and cat* = 1. The rational LS category, cato-X\ of X is the LS category of a rational CW complex in the rational homotopy type of X. Part V begins with the presentation in §27 of the main properties of LS cat- egory for 'ordinary' topological spaces. We have included this material here for the convenience of the reader and because, to our knowledge, much of it is not available outside the original articles scattered through the research literature. We then turn to rational LS category (§28) and its calculation in terms of Sullivan models (§29). A key point is the Mapping Theorem: Given a continuous map f : X —)¦ Y between simply connected spaces, then ?*(/) ® Q injective =? cat0 X < cato Y ¦ In particular, the Postnikov fibres in a Postnikov decomposition of a simply connected finite CW complex all have finite rational LS category. (The integral analogue is totally false!). A second key result is Hess' theorem (Meat = cat), which is the main step in the proof of the product formula cat^Q ? Yq = cat^Q + cati^ in §30. Finally, in §31 we prove a beautiful theorem of Jessup which gives circumstances under which the rational LS category of a fibre must be strictly less than that of the total space of a fibration. The "?, ?" theorem described at the start of this introduction is an immediate corollary.
Rational Homotopy Theory xv Part VI: The Rational Dichotomy: Elliptic and Hyperbolic Spaces AND Other Applications In this Part we use rational homotopy theory to derive the results referred to at the start of this introduction (and others) on the structure of ?*(???; k), when X is a simply connected finite CW complex. These are outlined in the introductions to the sections, and we leave it to the reader to check there, rather than repeating them here. As the overview above makes evident, this monograph makes no pretense of being a complete account of rational homotopy theory, and indeed important aspects have been omitted. For example we do not treat the iterated integrals approach of Chen ([37], [79], [145]) and therefore have not been able to include the deep applications to algebraic geometry of Hain and others (e.g. [80], [81], [101]). Equivariant rational homotopy theory as developed by Triantafillou and others ([151]) is another omission, as is any serious effort to treat the non-simply connected case, even though at least nilpotent spaces are covered by Sullivan's original theory. We have not described the Sullivan-Haefliger model ([144]. [78]) for the section space of a fibration even in the simpler case of mapping space, except for the simple example of the free loop space Xs , nor have we included the Sullivan-Barge classification ([144], [18]) of closed manifolds up to rational homotopy type. And we have not given Lemaire's construction [108] of a finite CW complex whose homotopy Lie algebra is not finitely generated as a Lie algebra. Moreover, this monograph does not pursue the connections outside or beyond rational homotopy theory. Such connections include the algebraic homotopy the- ory developed by Baues [20] following Quillen's homotopical algebra [134]. There is no mention in the text (except in the problems at the end) of Anick's extension of the theory to coefficients with only finitely many primes inverted ([11]) and its application to loop space homology,and there is equally no mention of how the results in Part VI generalize to arbitrary coefficients [56]. And finally, we have not dealt with the interaction with the homological study of local commutative rings [14] that has been so significantly exploited by Avramov and others. We regret that limitations of time and energy (as well as our publisher's insis- tence on limiting the number of pages!) have made it necessary simply to refer the reader to the literature for these important aspects of the subject, in the hope that what is presented here will make that task an easier one. In the last twenty five years a number of monographs have appeared that presented various parts of rational homotopy theory. These include Algebres Connexes et Homologie des Espaces de Lacets by Lemaire [109], On PL de Rham Theory and Rational Homotopy Type by Bousfield and Gugenheim ([30]), Theorie Homotopique des Formes Differentielles by Lehmann ([107]), Rational Homotopy Theory and Differential Forms by Griffiths and Morgan ([72]), Homo- topie Rationnelle: Modeles de Chen, Quillen, Sullivan by Tanre ([145]), Lectures on Minimal Models by Halperin [82], La Dichotomie Elliptique - Hyperbolique en Homotopic Rationnelle by Felix ([50]), and Homotopy Theory and Models
xvi Introduction by Aubry ([12]). Our hope is that the present work will complement the real contribution these make to the subject. This monograph brings together the work of many researchers, accomplished as a co-operative effort for the most part over the last thirty years. A clear account of this history is provided in [91], and we would like merely to indicate here a few of the high points. First and foremost we want to stress our individual and collective appreciation to Daniel Lehmann, who led the development of the rational homotopy group at Lille that provided the milieu in which all three of us became involved in the subject. Secondly, we want to emphasize the importance of the memoir [109] by Lemaire and of the article [21] by Baues and Lemaire which have formed the foundation for the use of Lie models. In this context the mini-conference held at Louvain in 1979 played a key role. It brought together the two approaches (Lie and Sullivan) and crystalized the questions around LS category that proved essential in subsequent work. Another mini-conference in Bonn in 1981 (organized jointly by Baues and the second author) led to a trip to Sofia to meet Avramov and the intensification of the infusion into rational homotopy of the intuition from local algebra begun by Anick, Avramov, Lofwall and Roos. This monograph was conceived of in 1992 and in the intervening eight years we have benefited from the advice and suggestions of countless colleagues and students. It is a particular pleasure to acknowledge the contributions of Cornea, Dupont, Hess, Jessup, Lambrechts and Murillo, who have all worked with us as students or postdocs and have all beaten problems we could not solve. We also wish to thank Peter Bubenik whose careful reading uncovered an unbelievable number of mistakes, both typographical and mathematical. The actual writing and rewriting have been a team effort accomplished by the three of us working together in intensive sessions at sites that rotated through the campuses at which we have held faculty positions: the Universite Catholique de Louvain (Felix), the University of Toronto and the University of Maryland (Halperin) and the Universite de Lille 1 and the Universite d'Angers (Thomas). The Fields Institute for Research in Mathematical Sciences in Toronto provided us with a common home during the spring of 1996, and the Centre International pour Mathematiques Pures et Appliquees, hosted and organized a two week sum- mer school at Breil where we presented a first version of the text. Our granting agencies (Centre National de Recherche Scientifique, FNRS, National Sciences and Engineering Research Council of Canada, North Atlantic Treaty Organiza- tion) all provided essential financial support. To all of these organizations we express our appreciation. Above all, however, we wish to express our deep gratitude and appreciation to Lucile Lo, who converted thousands of pages of handwritten manuscript to beautifully formatted final product with a speed, accuracy, intelligence and good humor that are unparalleled in our collective experience.
Rational Homotopy Theory xvii And finally, we wish to express our appreciation to our families who have lived through this experience, and most especially to Agnes, Danielle and Janet for their constant nurturing and support, and to whom we gratefully dedicate this book. Yves Felix Steve Halperin Jean-Claude Thomas
Contents Introduction Table of Examples I Homotopy Theory, Resolutions for Fibrations, and ?- ?? cal Spaces 0 Topological spaces 1 1 CW complexes, homotopy groups and cofibrations 4 (a) CW complexes 4 (b) Homotopy groups 10 (c) Weak homotopy type 12 (d) Cofibrations and NDR pairs 15 (e) Adjunction spaces 18 (f) Cones, suspensions, joins and smashes 20 2 Fibrations and topological monoids 23 (a) Fibrations 24 (b) Topological monoids and G-fibrations 28 (c) The homotopy fibre and the holonomy action 30 (d) Fibre bundles and principal bundles 32 (e) Associated bundles, classifying spaces, the Borel construction and the holonomy fibration 36 3 Graded (differential) algebra 40 (a) Graded modules and complexes 40 (b) Graded algebras 43 (c) Differential graded algebras 46 (d) Graded coalgebras 47 (e) When Jfe is a field 48 4 Singular chains, homology and Eilenberg-MacLane spaces 51 (a) Basic definitions, (normalized) singular chains 52 (b) Topological products, tensor products and the dgc, C*(X;Jfe). ... 53 (c) Pairs, excision, homotopy and the Hurewicz homomorphism .... 56 (d) Weak homotopy equivalences 58 (e) Cellular homology and the Hurewicz theorem 59 (f) Eilenberg-MacLane spaces 62 5 The cochain algebra Cm(X;k) 65
xx CONTENTS 6 (R, d)—modules and semifree resolutions 68 (a) Semifree models 68 (b) Quasi-isomorphism theorems 72 7 Semifree cochain models of a fibration 77 8 Semifree chain models of a G-fibration 88 (a) The chain algebra of a topological monoid 88 (b) Semifree chain models 89 (c) The quasi-isomorphism theorem 92 (d) The Whitehead-Serre theorem 94 9 P-local and rational spaces 102 (a) ¦P-local spaces 102 (b) Localization 107 (c) Rational homotopy type 110 II Sullivan Models 10 Commutative cochain algebras for spaces and simplicial sets 115 (a) Simplicial sets and simplicial cochain algebras 116 (b) The construction of A{K) 118 (c) The simplicial commutative cochain algebra -???, and Api(X) . . 121 (d) The simplicial cochain algebra Cpl, and the main theorem .... 124 (e) Integration and the de Rham theorem 128 11 Smooth Differential Forms 131 (a) Smooth manifolds 131 (b) Smooth differential forms 132 (c) Smooth singular simplices 133 (b) (d) The weak equivalence ADr(M) ~ APL(M;R) 134 12 Sullivan models 138 (a) Sullivan algebras and models: constructions and examples ..... 140 (b) Homotopy in Sullivan algebras 148 (c) Quasi-isomorphisms, Sullivan representatives, uniqueness of mini- mal models and formal spaces 152 (d) Computational examples 156 (e) Differential forms and geometric examples 160 13 Adjunction spaces, homotopy groups and Whitehead products 165 (a) Morphisms and quasi-isomorphisms 166 (b) Adjunction spaces 168 (c) Homotopy groups 171 (d) Cell attachments 173
Rational Homotopy Theory xxi (e) Whitehead product and the quadratic part of the differential ... 175 14 Relative Sullivan algebras 181 (a) The semifree property, existence of models and homotopy 182 (b) Minimal Sullivan models 186 15 Fibrations, homotopy groups and Lie group actions 195 (a) Models of fibrations 195 (b) Loops on spheres. Eilenberg-MacLane spaces and spherical fibrations200 (c) Pullbacks and maps of fibrations 203 (d) Homotopy groups 208 (e) The long exact homotopy sequence 213 (f) Principal bundles, homogeneous spaces and Lie group actions . . . 216 16 The loop space homology algebra 223 (a) The loop space homology algebra 224 (b) The minimal Sullivan model of the path space fibration 226 (c) The rational product decomposition of UX 228 (d) The primitive subspace of H*{UX\k) 230 (e) Whitehead products, commutators and the algebra structure of H*{9.X;k) 232 17 Spatial realization 237 (a) The Milnor realization of a simplicial set 238 (b) Products and fibre bundles 243 (c) The Sullivan realization of a commutative cochain algebra 247 (d) The spatial realization of a Sullivan algebra 249 (e) Morphisms and continuous maps 255 (f) Integration, chain complexes and products 256 III Graded Differential Algebra (continued) 18 Spectral sequences 260 (a) Bigraded modules and spectral sequences 260 (b) Filtered differential modules 261 (c) Convergence 263 (d) Tensor products and extra structure 265 19 The bar and cobar constructions 268 20 Projective resolutions of graded modules 273 (a) Projective resolutions 273 (b) Graded Ext and Tor 275 (c) Projective dimension .'! 278 (d) Semifree resolutions 278
xxii CONTENTS IV Lie Models 21 Graded (differential) Lie algebras and Hopf algebras 283 (a) Universal enveloping algebras 285 (b) Graded Hopf algebras 288 (c) Free graded Lie algebras 289 (d) The homotopy Lie algebra of a topological space 292 (e) The homotopy Lie algebra of a minimal Sullivan algebra 294 (f) Differential graded Lie algebras and differential graded Hopf algebras296 22 The Quillen functors C. and C 299 (a) Graded coalgebras 299 (b) The construction of C»(L) and of C*(L;M) 301 (c) The properties of C.(L;UL) 302 (d) The quasi-isomorphism C*{L) ~ ) BUL 305 (e) The construction C{C,d) 306 (f) Free Lie models , . 309 23 The commutative cochain algebra, C*{L,dL) 313 (a) The constructions C*(L,di), and ?(/t,<f) 313 (b) The homotopy Lie algebra and the Milnor-Moore spectral sequence 317 (c) Cohomology with coefficients 319 24 Lie models for topological spaces and CW complexes 322 (a) Free Lie models of topological spaces 324 (b) Homotopy and homology in a Lie model 325 c) Suspensions and wedges of spheres 326 (d) Lie models for adjunction spaces 328 (e) CW complexes and chain Lie algebras 331 (f) Examples 331 (g) Lie model for a homotopy fibre 334 25 Chain Lie algebras and topological groups 337 (a) The topological group, \TL\ 337 (b) The principal fibre bundle, 338 (c) \TL\ as a model for the topological monoid, QX 340 (d) Morphisms of chain Lie algebras and the holonomy action 341 26 The dg Hopf algebra C*{QX) 343 (a) Dga homotopy 344 (b) The dg Hopf algebra C» (UX) and the statement of the theorem . 346 (c) The chain algebra quasi-isomorphism ? : (ULv,d) 347 id) The proof of Theorem 26.5 349
Rational Homotopy Theory xxiii V Rational Lusternik Schnirelmann Category 27 Lusternik-Schnirelmann category 351 (a) LS category of spaces and maps 352 (b) Ganea's fibre-cofibre construction 355 (c) Ganea spaces and LS category 357 (d) Cone-length and LS category: Ganea's theorem 359 (e) Cone-length and LS category: Cornea's theorem 361 (f) Cup-length, c(X;Jk) and Toomer's invariant, e(A';Jc) 366 28 Rational LS category and rational cone-length 370 (a) Rational LS category 371 (b) Rational cone-length 372 (c) The mapping theorem 375 (d) Gottlieb groups 377 29 LS category of Sullivan algebras 381 (a) The rational cone-length of spaces and the product length of models 382 (b) The LS category of a Sullivan algebra 384 (c) The mapping theorem for Sullivan algebras 389 (d) Gottlieb elements 392 (e) Hess' theorem 393 (f) The model of (AV,d) -> {AV/A>mV,d) 396 (g) The Milnor-Moore spectral sequence and Ginsburg's theorem . . . 399 (h) The invariants meat and e for (AV, d)-modules 401 30 Rational LS category of products and fibrations 406 (a) Rational LS category of products 406 (b) Rational LS category of fibrations 408 (c) The mapping theorem for a fibre inclusion 411 31 The homotopy Lie algebra and the holonomy representation 415 (a) The holonomy representation for a Sullivan model 418 (b) Local nilpotence and local conilpotence 420 (c) Jessup's theorem 424 (d) Proof of Jessup's theorem 425 (e) Examples 430 (f) Iterated Lie brackets 432 VI The Rational Dichotomy: Elliptic and Hyperbolic Spaces and Other Applications 32 Elliptic spaces 434
xxiv CONTENTS (a) Pure Sullivan algebras 435 (b) Characterization of elliptic Sullivan algebras 438 (c) Exponents and formal dimension 441 (d) Euler-Poincare characteristic 444 (e) Rationally elliptic topological spaces 447 (f) Decomposability of the loop spaces of rationally elliptic spaces . . 449 33 Growth of Rational Homotopy Groups 452 (a) Exponential growth of rational homotopy groups 453 (b) Spaces whose rational homology is finite dimensional 455 (c) Loop space homology 458 34 The Hochschild-Serre spectral sequence 464 (a) Horn, Ext, tensor and Tor for i7L-modules 465 (b) The Hochschild-Serre spectral sequence 467 (c) Coefficients in UL 469 35 Grade and depth for fibres and loop spaces 474 (a) Complexes of finite length '. 475 (b) nr-spaces and C*(n7)-modules 476 (c) The Milnor resolution oik 478 (d) The grade theorem for a homotopy fibre 481 (e) The depth of if, (??) 486 (f) The depth of UL 486 (g) The depth theorem for Sullivan algebras 487 36 Lie algebras of finite depth 492 (a) Depth and grade 493 (b) Solvable Lie algebras and the radical 495 (c) Noetherian enveloping algebras 496 (d) Locally nilpotent elements 497 (e) Examples 497 37 Cell Attachments 501 (a) The homology of the homotopy fibre, X xy PY 502 (b) Whitehead products and G-fibrations 502 (c) Inert element 503 (d) The homotopy Lie algebra of a spherical 2-cone 505 (e) Presentations of graded Lie algebras 507 (f) The Lofwall-Roos example 509 38 Poincare Duality 511 (b) Properties of Poincare duality 511 (b) Elliptic spaces 512 (c) LS category 513 (c\) Inert elements 513
Rational Homotopy Theory *xv 39 Seventeen Open Problems 516 References 521 Index 531
Table of Examples §1. CW complexes, homotopy groups and cofibrations (a) 1. Complex projective space 2. Wedges 3. Products 4. Quotients and suspensions ?. Cubes and spheres 6. Adjunction spaces 7. Mapping cylinders §2. Fibrations and topological monoids (a) 1. Products 2. Covering projections 3. Fibre products and pullbacks (b) 1. Path space fibrations §3. Graded (differential) algebra (b) 1. Change of algebra 2. Free modules 3. Tensor product of graded algebras 4. Tensor algebra 5. Commutative graded algebras 6. Free commutative graded algebras (c) 1. Horn* (M,N) 2. Tensor products 3. Direct products 4. Fibre products §9. "P-local and rational spaces (a) 1. The infinite telescope S™ (b) 1. Cellular localization §10. Commutative cochain algebras for spaces and simplicial set (c) 1. APL(pt) §11. Smooth differential forms (a) Smooth manifolds: 1. W1 2. Open subsets 3. Products
xxviii Table of Examples §12. Sullivan models (a) 1. The spheres, Sk 2. Products of topological spaces 3. //-spaces 4. A cochain algebra (AV, d) that is not a Sullivan algebra 5. (AV,d),lmdcA+V-A+V 6. Topological spaces with finite homology admit finite dimensional commutative models 7. (A(a, b, x. y, z),d), da = db = 0, dx = a2, dy = ab, dz = b2 (b) 1. Null homotopic morphisms into (^4,0) are constant (c) 1. Wedges (d) 1. (AV, d) = A(x,y,z;dx = dy = 0, dz = xy), deg ? = 3, deg y = 5 2. Automorphisms of the model of S3 ? S3 ? Ss ? S6 3. (AV,d) = ?(?, ?, b;da = dx = 0,db = ax), dega = 2, dega: = 3 4. A(x,y,z,u;dx = dy = dz = 0,du = xm + ym - zm), deg a: = degy = degz = 2 5. Two morphisms that homotopy commute, but are not homotopic to commuting morphisms 6. Sullivan algebras homogeneous with respect to wordlength 7. Sullivan models for cochain algebras (H, 0) with trivial multipli- cation (e) 1. The cochain algebra Aor(G)g of right invariant forms on G, and its minimal model 2. Nilmanifolds 3. Symmetric spaces are formal 4. Compact Kahler manifolds are formal 5. Symplectic manifolds need not be formal §13. Adjunction spaces, homotopy groups and Whitehead products (e) 1. The even spheres S2n 2. S3 V S3 U/ (?>g II ?>?) §14. Relative Sullivan algebras (b) 1. The acyclic closure (B <8> AV, d) 2. A commutative model for a Sullivan fibre §15. Fibrations, homotopy groups and Lie group actions (b) 1. The model of the loop space ClSk, k>2 2. The model of an Eilenberg-MacLane space 3. The rational homotopy type of fc(Z,n) ? Qr^Vipriral fibrations
Rational Homotopy Theory xxix 5. Complex projective spaces (c) 1. The free loop space Xs (d) 1. Rational homotopy group of spheres 2. (A(eo,ei,x),dx = eoei) 3. The Quilleri plus construction 4. Smooth manifolds and smooth maps (f) 1. Homogeneous spaces 2. Borel construction; application to free actions 3. Matrix Lie groups 4. A fibre bundle with fibre S3 x SUC) that is not principal 5. Non-existence of free actions §16. The loop space homology algebra (a) 1. The homology algebra if. (US2k+1 ; Jfe), k > 1 (e) 1. The homology algebra if, (USn+i ; Jfe) §17. Spatial realization (a) 1. |?[?]| = ?" 2. Cones 3. Extendable simplicial sets (b) 1. A simplicial construction of the Eilenberg MacLane space K(k, 1) (c) 1. Realization of products 2. Contractible Sullivan algebras have contractible realizations §18. Spectral sequences (b) 1. Semifree modules 2. Relative Sullivan algebras 3. The wordlength filtration of a Sullivan algebra 4. Bigradations (d) 1. Filtered differential graded algebras §20. Projective resolutions of graded modules (b) 1. The bar construction (d) 1. The Eilenberg-Moore spectral sequence §21. Graded (differential) Lie algebras and Hopf algebras 1. Graded algebras 2. The adjoint representation 3. Derivations 4. Products 5. The tensor product .4 <8> L
xxx Table of Examples (a) 1. Abelian Lie algebras §22. The commutative cochain algebra C*(L,cIl) (a) 1. Graded Lie algebras 2. Free graded Lie algebras 3. Minimal Lie models of minimal Sullivan algebras 4. Sullivan algebras (AW,d) for which HQk(AW,d) = 0, 1 < k < ? (b) 1. C*(L,dL) 2. Topological spaces X §24. Lie models for topological spaces and CW complexes. 1. The free Lie model of a sphere (c) 1. If H+(X; Q) is concentrated in odd degrees then D has the ratio- nal homotopy type of a wedge of spheres. (f) 1. (Ly,0) is the minimal Lie model of a wedge of spheres 2. The coproduct of Lie models is a Lie model for a wedge 3. The direct sum of Lie models is a Lie model of the product 4. A Lie model for S3 ? So3 \J[*,[O,0]W)W D8 5. Lie models for CP°° and CP" 6. A Lie model for (CP2 V S3) \J[o>0]w Ds 7. Coformal spaces 8. Minimal free chain Lie model §27. Lusternik-Schnirelmann category (e) 1. The LS category of the Ganea spaces PnX §28. Rational LS category and rational cone-length (a) 1. cato I V^a ) = max{cato(Xa)) (c) 1. Postnikov fibres 2. Free loop spaces have infinite rational category (d) 1. G-spaces 2. The holonomy fibration 3. Loop spaces §29. LS category of Sullivan algebras (b) 1. A space X satisfying c0X < e0X 2. A space X satisfying e^X < cat0 X 3. A space X satisfying cat0 X < do X 4. Formal spaces 5 Coformal spaces '•?, ? „At]n v _ yodd anddim
Rational Homotopy Theory xxxi (c) 1. e ((AV, d) ® (kW, d)) = e(AV, d) + e{AW, d) 2. cat0 X - e0X can be arbitrary large (d) 1. A non-trivial Gottlieb element (e) 1. Field extension preserves category 2. Rational category of smooth manifolds §30. Rational LS category of products and fibrations (b) 1. The relative Sullivan algebra (i\.(x,y),dy = xr+1) —> (? (?, y, u, v),dy = xr+1, dv = un+1 — x) (c) 1. Fibrations with cat X = 0 2. Fibrations with cat X = 1 3. Fibrations with cat0 X = max(cat0 Y, cat0 F) §31. The homotopy Lie algebra and the holonomy representation (b) 1. The holonomy representation 2. The adjoint representation (e) 1. The fibration ? : S4m+3 —> MPm 2. The fibration associated with S3 V S3 -^ S3 3. Derivations in commutative cochain algebras 4. A fibration X —> CPn with fibre S3 and cat0 X = 2 (f) 1. A space X with eoX = 2 and cat0X = oo §32. Elliptic spaces (b) 1. A(a2,X3,u3,b4,v5,w7] da = dx = 0, du = ?2, db = ??, di» = ?& — ux, dw = b2 — vx) 2. Adding variables of high degree 3. Algebraic closure of Jfc is necessary in Proposition 32.3 4. ?-colourable graphs (e) 1. Simply connected finite ??-spaces are rationally elliptic 2. Simply connected homogeneous spaces G/K are rationally elliptic 3. Free torus actions §33. Growth of rational homotopy groups (c) 1. Wedges of spheres 2. X = S3VS3 3. X = S3VS32U[a[a>?]w]wDs 4. X = T{S3,S3,S3) 5. X = YVZ §34. The Hochschild-Serre spectral sequence (a) 1. Torul(lk,]k) = s (//[/,/]) 2. The representation of L/I in Tor^Jc, Jfc)
xxxii Table of Ex §36. Lie algebra of finite depth (e) 1. Depth = 0 2. Free products have depth 1 3. ivy 4. Products 5. X=S*aVS!l)[a[a>b]w]wD& 6- CP°°/CPn 7. eo(X) = 2; cato(JQ = depth Lx = 3 8. L = Der>0 Ly, where F is a finite dimensional vector dimerision at least 3
Part I Homotopy Theory, Resolutions for Fibrations, and P-local Spaces
0 Topological spaces In this section we establish the notation and conventions for topological spaces that will obtain throughout the monograph. A k-space (not to be confused with the symbol k used to denote coefficient ring) is a Hausdorff topological space ?' such that Aclis closed if and only if ? ? C is closed in C for all compact subspaces C C X. If X is any Hausdorff topological space then Xk is the set X equipped with the associated k-space topology: A is closed in Xk if and only if ? ? C is closed in C for all compact subspaces C oi X. It is easy to see that Xk is a fc-space, Xk -^ X is a continuous bijection and X = Xk if and only if X itself is a fc-space. Moreover, all metric spaces are fc-spaces. A continuous map / : X —» Y is proper if f~1(C) is compact whenever C is. It is an easy but important observation that: ? proper continuous bijection between k-spaces is a homeomorphism. Henceforth in this book we shall restrict attention to k-spaces. Consistent with this convention we need to modify some (but not all) of the standard topological constructions as follows [44]. • Subspaces. If .A is a subset of a fc-space X then we assign to A the &-space topology associated with the 'ordinary' subspace topology. • Products. If {Xa} is a family of fc-spaces we assign to the set theoretic product Y[Xa a the fc-space topology associated to the ordinary product topology. If ? is any fc-space then a map / : ? —» f] Xa is continuous if and only if each 'component' a fa : ? —> Xa is continuous. Note also that if X is locally compact and Y is a fc-space then the fc-space topology in ? ? ? is just the ordinary topology. • Quotients. A quotient space F of I is a surjection ? : X —t Y such that U C Y is open if and only if p~l(U) is open. We only consider quotients Y such that Y is Hausdorff and X is a fc-space; in this case Y is automatically a fc-space. The product pxp' : X xX1 —» ? ? ?' of two such quotient maps is itself a quotient map. (This follows easily from a lemma of Whitehead [44] which states that ? ? id : ? ? ? —> ? ? ? is a quotient map if ? is locally compact.) • Mapping spaces. If Ar and Y are topological spaces then Yx (as a set) is the set of continuous maps from X to Y. The compact-open topology in this set is the topology whose open sets are the arbitrary unions of finite intersections of subsets of the form U with U open in Y and C compact in X. As usual we assign to Yx the k- space topology associated with the compact-open topology. Then for fe-spaces
2 0 Topological spaces ?, ?, ? the exponential law asserts that a homeomorphism ZYxX -=» (Zy )x is given by / *-? F, with F (?) (y) — /(y, ?). In particular, the evaluation map Yx ? X —>¦ ?, (g,x) *-? g(x), is continuous. The following conventions from topology will be used without further refer- ence. • The length of ? G Rk is denoted by ||u||. • The unit interval I — [0,1]. • The unit ?-cube In = / ? · · · ? / C ?". Thus the boundary, dln, consists of the points ? G /" with some x,· G {0,1}. • The ?-dimensional disk Dn C W1 is defined by Dn = {x G W1 \\x\\ < 1}. • The ?-dimensional sphere Sn C Rn+1 is defined by 5" = {x G E"+1 ||z|| = 1}. Thus 5" = dDn+1 is the boundary of the (n + l)-disk. • A pair of topological spaces, (X, A) is a space X and a subspace A C X (with the &-space topology described above). A map of pairs, ? : {?,?) -> (?, ?) is a continuous map ? ? : ? ->¦ Y restricting to ? a : A ->· ?. If ? is another space then (?',?) ?? = (? ??,? ??). • Given continuous maps X ·—> Y ^- ? the fibre product X xy ? C X x ? is the subspace of points (x, z) such that f(x) = 5B). Projection defines a commutative square X xy ? —- ? ?—r-? and any pair of continuous maps ? : W ¦—> ?, -? : W —>¦ ? such that ?? — gu defines a continuous map (?, ?) :W —>¦ ? ?? ?. A based (or pointed) space is a pair (X,x0); x0 e X is the basepoint. A based map is a map </? : (?,??) -> (y, yo). The disjoint union of spaces Xa is denoted by ]J Xa. The we^e of based Q spaces (Xa,xa) is the based space \/QXa = Uxa/U.{x<*}- Based maps "i : (X.xn) —> {Z,z0) and t/>(y, j/0) —* (Z,z0) define a map (?, ?) : A" V
Homotopy Theory 3 • Suppose given a pair (Z,B) and a continuous map f : ? -> X. We denote by X U/ ? the quotient space of A" JJ ? obtained by identifying b ~ f(b),b G ?. It is called the adjunction space obtained by attaching ? to X along /. • Given A C X the based space {X/A, [A]) is obtained by identifying the points of .4 to a single point [A], and giving X/A the quotient topology. Thus X/A = * U/ X, where f : A -l· * is the constant map. • The suspension, ??, of a based space (A', xo) is the based space A" x 7/(A"x{0,l}U{zo} ?/)· • Continuous maps f,g : X -> Y are homotopic (/ ~ 5) if there is a continu- ous map F : ? ? I ^ Y such that F(x, 0) = /(x) and F(x, 1) = c/(x), ? G A. F is a homoEopy from / to g. Being homotopic is an equivalence relation; the equivalence class of / is its homotopy class and the set of homotopy classes of maps is denoted by [X, Y]. • A homotopy equivalence is a continuous map / : X ¦—> Y such that for some continuous map g : Y ¦—> I we have fg ~ id ? and gf ~ id ?. In this case we write / : A -=> Y and we say g is a homotopy inverse for /. If there is a homotopy equivalence from A to y then A' and Y have the some homotopy type and we write A ~ y. If the constant map A ·—> pi is a homotopy equivalence then X is contractible. • A tosec? homotopy between /, g : (A, x0) —> (Y,yo) is a continuous map F : (?,??) ? ? -> (Y,y0) such that F(x,0) = /(x) and F(x,l) = g(x)· This is an equivalence relation and the class of / is its based homotopy class. The set of based homotopy classes is denoted by [(A,xo), (Y, j/o)]· A tosec? homotopy equivalence is a continuous map / : (A, x0) ·—> (Y, J/o) such that for some continuous map g : (Y, j/o) ¦—> (?',??), fg and 5/ are , respectively, based homotopic to idy and id ?. • Two continuous maps /, g : X -> Y which restrict to the same map ? in a subspace A C A are homotopic rel .4 if there is a homotopy ? from / to g such that H(a,t) = </?(a) for all ? G .4, t G /; H is a homotopy rel .4. • A subspace .<4C—>A is a retract of A if there is a map ? : A" ->· .4 (the retraction) such that pi = id.4. It is a strong deformation retract if also ip ~ idx rel A.
1 CW complexes, homotopy groups and cofibra- tions We begin this monograph by introducing the main objects of study in homo- topy theory: CW complexes and their homotopy groups. The topological spaces (manifolds, polyhedra, real algebraic varieties, ... ) that arise in geometric con- texts are all CW complexes. Homotopy groups are groups ??(?), ? > 1, defined for any topological space, and abelian for ? > 2. A weak homotopy equivalence is a continuous map / such that 7?*(/) is an isomorphism, and we establish two important results of JHC Whitehead: • There is a weak homotopy equivalence from a CW complex to any topolog- ical space, and • A continuous map from a CW complex lifts, up to homotopy, through a weak homotopy .equivalence. It follows that two CW complexes connected by a chain of weak homotopy equiv- alences have the same homotopy type. This section is organized into the following topics (We note however that, aside from the definition of suspension, topics (d) - (f) will not be needed until Part V): (a) CW complexes. (b) Homotopy groups. (c) Weak homotopy type. (d) Cofibrations and NDR pairs. (e) Adjunction spaces. (f) Cones, suspensions, joins and smashes. (a) CW complexes. Let / : U Sq —> X be a continuous map from a disjoint union of (n — 1) spheres S^ to a topological space X. We denote by Y = X U/ ([} D%) the quotient space obtained from the disjoint union X]J(]J D?) by identifying ? e 52 with f(x) eX. The image el of ?? in Y is called an ?-cell and this construction is called attaching ?-cells to X along an attaching map /. Thus Y = X U (UQ el). The map F? : ?>? —* Y is called the characteristic map of lU"" nM- K" ^finition it restricts to / : S" —> X.
Homotopy Theory 5 A filtered space is a topological space X together with an increasing sequence X_1 c Xo C X\ C of closed subspaces such that X = \J Xn and X has the weak topology determined by the Xn- n>-\ We are now ready to introduce CW complexes, and we do so in the following sequence of basic definitions: • Let .4 be a topological space (necessarily a fc-space by our convention). A relative CW complex is a pair (X, A) in which X — (J Xn is a filtered n>-\ space, A = X-\, and there are specified identifications and maps \aelo and /?'· ? S2^Xn> Xn+l = *n U/ ( ? D%+1 ) , ? > 0. ? / ?"? is called the n~skeleton of (X, A). A CW complex is a relative CW complex of the form (?, ?). • If X is a CW complex and X = ??, some n, then X is ?-dimensional or finite dimensional. If X has finitely many ?-cells for each n, then A' has finite type. • A based CW complex is a pair (AT, Xo) with ,Y a CW complex and x0 e Xo- • A cellular map f : (X, A) —>¦ (Y, B) between relative CW complexes is a continuous map such that f : Xn —> Yn, n> —1. • A subcomplex of (X, A) is a pair (Y, A) in which Y is a subspace of X that is a union of A and of cells of X. Note that (Y,A) and (JY, 7) are then also relative CW complexes. We begin with some elementary remarks about the topology of CW complexes. Let (Ar, A) be a relative CW complex with characteristic maps Fa : ?>? —>¦ Xn. Then f : X —>¦ ? is continuous if and only if /| and / ? F? are continuous for all ? and a. Next, if xa G Dl - S2~l we may identify (?? - {z«}^) = is'a~l x @,l],S?-^x {1}). Fix such an xa for each cell, and write Xn = Xn-i U/ (\laD2), where / = {fa} : 7J 52 —> Xn_i is the attaching map. There is then an automatic procedure for extending a subset ? C Xn-i to V c Xn: set
6 1 CW complexes, homotopy groups and coflbrations If ? is open in Xn-\ then V is open in Xn. Iterating this procedure ad infinitum produces U C X such that each U ? Xk is open; i.e., U is open in X. Clearly U ? Xn = 0, and we call U the canonical extension of 0 away from the xa. Proposition 1.1 Let (X, A) be a relative CW complex. Then (i) X is Hausdorff, and hence automatically a k-space. (ii) Every compact subset of X is contained in the union of A and finitely many cells. (Hi) If A = 0 then X has a universal covering space. (iv) If A is normal so is X. proof: (i) If x,y G Xn — Xn-i (n > —1) then there are disjoint open neigh- bourhoods O{x),O(y) C Xn — Xn-i- The canonical extensions to open sets U(x),U(y) C X are disjoint. The general case is proved in the same way. (ii) Let C C X be compact and choose xa G Fa (D? - SJ) so xa G C if C ? Fa (?>2 - 52) ? 0. For each ? extend F0 (d* - S^) c Xn to an open set Up C X, as described above and in the same way extend A =.X_i to an open set Ua· This defines an open cover of X and xa is in Ua and in no other Up. Since C is compact CC^U Uai U · · · U Uate. It follows that only ? (Hi) For each ?-cell, ^?(?)^ —52) is an open subspace of Xn — Xn-i· Extend this to an open subset U? of X away from the origins of the other cells. It is easy to see that any loop in U? is homotopic to the constant loop, and that the U° cover X. Hence the standard construction [68] provides a universal cover. (iv) Suppose C and C are disjoint closed subspaces of X. We define a continuous function h : X —» / such that /iL· = 0 and h\c, = 1. Indeed assume by induction that h is constructed in Xn. (Start the induction with ? = 0 using the fact that A is normal.) If Fa : ?>?+1 —>¦ Xn+i is the characteristic map of an (n + 1)—cell with attaching map fa : Sn —> X define ha : F~\C U C) U 5" —> / by ha = 0 in F^iC), ha = 1 in F-l{C) and ha = ho fa\n Sn. Since jDq+1 is normal we may extend ha to a continuous function /?? : D?+1 ¦—> / (Tietze extension theorem [REF]). The ha extend the construction to h ;In+1 ·—> I. D Example 1 Complex protective space. CPn is the space of complex lines through the origin in C"+1. S2n+l is the unit sphere in C14. Thus assigning to ? G S2n+1 the complex line Car, we define a map / : 52ri+1 -> CPn. The reader is invited to check that CPn+l = CPn Of e2n+2. This exhibits CPn as a CW complex with one cell in each even dimension 2/c, 0 < k < n. The CW complex CP°° is obtained as the union,
Homotopy Theory 7 Example 2 Wedges. The wedge of based CW complexes, (??,??), is a CW complex whose n- skeleton is the wedge of the n-skeleta of the Xa. ? Example 3 Products- Let X and Y be CW complexes. Denote the cells of X and Y respectively by e? and /™, and let ?? : D? —> X and -? ? : D'? -> Y be the characteristic maps for e? and /?*. The product of the CW complexes X and Y is the space ? ? >' with the following CW structure. The cells of dimension k are the products e? ? fjp with ? + ?? = k, and characteristic maps DT Dn ? Dr ? ? ?. The attaching map of e? ? /^ is the restriction of the previous map to dDn+m = (dDn ? Dm)U(Dn xdDm). If X or Y is a finite CW complex then it is compact, and so the topology in Ar ? y is the ordinary product topology. ? Example 4 Quotients and suspensions. The quotient X/A for a relative CW complex (X, A) is a CW complex whose cells are the cells of X together with the additional 0-cell, [A]. In particular, the suspension of a based CW complex is again a based CW complex. ? Example 5 Cubes and spheres. The unit interval is a CW complex with three cells: {0}, {1} and /. Thus the cube /" inherits a (product) CW7 structure, and dln is a subcomplex. Note that the CW complex In/dln coincides, as a CW complex, with the suspension of /"-»/a/"-1. On the other hand, assign to 5" the basepoint * = A,0,0,... ,0). The map ? : S" x / —y Sn, represented by the picture induces a based homeomorphism ES" -A 5". Identify I/dl = S1 via the map t ?» e2wit. Then we identify inductively 5" = '1-1) = In/dln.
1 CW complexes, homotopy groups and cofibrations This exhibits 5" as the CW complex *Uen. There is also a homeomorphism dP Sn, ? > 1, defined by translating In+1 in W+l to centre it at the origin, and then projecting the boundary onto 5" (explicitly, ? ?-> (x — a)/\\x — a\\, where a = (\,..., |)). The corresponding base point of dln+l is (l, |,..., |). , |)). ? Example 6 Adjunction spaces. Suppose ? is a subcomplex of a CW complex ? and f : ? -> A" is a cellular map into a third CW complex. The adjunction space X U/ ? is a CW complex whose cells are the cells of ?' together with the cells of the relative complex (?,?). ? Example 7 Mapping cylinders. Any continuous map ? : X —> Y between CW complexes may be converted (up to homotopy) to the 'inclusion of a subcomplex' as follows: First, using Theorem 1.2 immediately below, replace ? by a homotopic cellular map ?. Then attach ? ? / to Y along ? ? {0} via the map ? to obtain the mapping cylinder, ? ?? ? ? /: Since ? is cellular, ? ?_|? {? ? /) is a CW complex (Example 6). Moreover it contains Y as a strong deformation retract (push down the cylinder) and the inclusion of X as the subcomplex ? ? {1} is homotopic through the cylinder to V>. ? Next we turn to properties useful in homotopy theory. Since (Dn xI,Dnx {0} U S71-1 ? /) S (?>" ? /, jD" x {0}), any continuous map from Dn ? {0} U Sn'1 ? / automatically extends to Dn ? /. Now suppose {X, A) is a relative CW complex, / : X -> Y is continuous and ?:^4?/ —»Y is a homotopy from f\A to ?. Then (by induction on the skeleta) /U ? : ? ? {0} U ? ? 7 -» Y extends to a homotopy X xl —t Y from / to a map g. This homotopy extension property for relative CW complexes is important in the proof of
Homotopy Theory 9 Theorem 1.2 (Cellular approximation) [160] Any continuous map f : (X, A) —i (Y, B) between relative CW complexes is homotopic rel A to a cellular map. proof: Step (i) Linear approximation. Suppose ? : ? -> Rk is a continuous map from a finite ?-dimensional CW complex. Write Zr = Zr_i U/r and let ??,? be the origin (centre) of Dra. Any point in Dra has the form tv with ? G Sq, and 0 < t < 1. The linear approximation of ? is the linear map ? : ? -> Rk defined inductively by: ? = ? in Zo, and = Wfr(v) + A - tu G An ?-dimensional flat in Efc is a subset of the form ? + W where ? e Rk and W C Efc is an ?-dimensional subspace. An obvious induction shows that Im# is contained in a finite union of ?-dimensional flats. It is also contained in any convex set C such that C D Im ?. Finally, let ? > 0 and suppose for all cells era that \\?? — v?(orja)|| < ?, ? G era. Since ||0(it;)-y>(it;)|| < t||0/r(t;) - <pfr{v)\\ + t\\<pfr(v) - <p(or,a)\\ + \\?(??>?) - ?(??)\\ it follows by induction that \\?? — ??\\ < 2??, ? e ?. Step (ii) The case (X,A) = (Dn,Sn~~L). Since Dn is com- pact, / : Dn -> Y has its image in some finite subcomplex ? D B. If dim ? < ? there is nothing to prove. Otherwise write ? — ?' U er for some subcomplex Z' D ? and some r > n. We construct g ~ /rel S" so that g : Dn -> Z'U/3(Dr-{z}) for some ? in the interior of Dr. But Sr~1 is a strong deformation retract of Dr — {z} and so Z' is a strong deformation retract of Z' Up (Dr — {z}). Hence we can find g' : Dn -> Z' with g' ~ 5 rel 5". After finitely many steps we have /' ~ /rel 5"-1 and /' : Dn -»¦ yn. It remains to construct 5. For this we may as well suppose that Y = BU? Dr, r>n. Identify (.D^S") = {In,dln). The iVth subdivision of /" is obtained by dividing / into ?^ equal subintervals, regarding these as the 1-cells of a CW complex structure on / and giving /" the product CW complex structure. This is illustrated for ? = 2 by The cells of /" will be called little cubes.
10 1 CW complexes, homotopy groups and cofibrations By setting ||6|| = 1, b G ?, we extend the standard length function in Dr to Y -> /. Choose ? so large that for any little cube ??, \\fxi \\ — \\fx2 < y^, ??,?? G ??. The union of the little cubes satisfying ||/x|| > |, ? G ?? subcomplex L of/", and dln C L. Similarly the little cubes satisfying ||/rc|| < |, ? G ?? are a subcomplex K, and In = L U K. Regard f\K as a map into the disk D C ?>r of radius |. Since ?> is convex the linear approximation of /L is a continuous map ? : ? -i D. Choose h : Y -> / so that /i(y) = 0 if \\y\\ > | and h{y) = 1 if ||y|| < |. Define ? : In ? / -> y by _ ^ ? ?/?(/?)?? + A - This is well defined because /i(/x) = 0 if a: G L. Put 5B:) = ?(?, 1). Next recall that Im ? is contained in a finite union of ?-dimensional flats in W. Since ? < r there is a point ? G Dr - ???? such that ||z|| < ^. If ? € ? then ||pz|| = ||/z|| > |. If ? G A" and ?(/?) # 1 then ||/?|| > \. As"noted above in (i), \\?? - fx\\ <ki = i- Hence II^H > H/^ll - th(fx)\\8x - fx\\.> ±. If ? ? ? and h(fx) = 1 then gx = ??. In each case gx ? ? and so ? $. Im g. Step (in) The general case. We construct a sequence of maps /„ : X -> Y such that /0 = /, and for ? > 1, fn(Xn-i) C Vn_i, and In ~ fn-i rel Xn-2· Indeed given /„ we use Step (ii) in each ?-cell of X to construct a homotopy ? rel Xn-i, fr°m fn to a map g : Xn -> Yn. Use the homotopy extension property to extend $ U fn : Xn ? I U ? ? {0} -? ? to a. map ?? : ? ? I -*· y and put /n+i(z) = ??(^; 1)- Finally, let ?? < ?? < · · · < ?? < · · · be a sequence increasing from 0 = ?? to 1. Let Hn : ? ? [??,??+?] -> Y be a map such that Hn{x,en) = fn{x), ??{?,??+?) = /n+i(z) and Hn(x,t) = /n(z), ? e Xn-i- Define ??:???->1' by Then ? is a homotopy rel A from / to a cellular map. (b) Homotopy groups. For any based space (X, xo), we write to denote the set of based homotopy classes of continuous maps / : E", *) -> (X,x0). When ? = 0, ???(?",??) is the set of path components of X, pointed by the path component of x$. By convention this basepoint is denoted by 0. If ??(-?,??) = {0} then X is called path connected. For ? > 1, ??{?,?0) is a group. Multiplication is defined via the identification _~,_i nn ¦„ T?„nrr,n]o R ahnve; jf [f^[^ ? ???(?,?0) are represented by
Homotopy Theory 11 f,g : (ES",*) —> (?,??) then the product [/] * [g] is represented by / * g : (ES71,*) —> {?,??), where f(x,2t) 0<t < i , i <i < The constant map c : Sn -> xo represents the identity of ??(??,??). A continuous map ? : (X, xo) —>· (V, t/o) induces the group homomorphisms 7rn(</?) : 7r„(,Y,x0) —^ 7rn(y,y0) given by ??{?)[?] = [? ° /]; ???(^) depends only on the based homotopy class of ?. Definition The groups ???(?,??) are the homotopy groups of (X,xq). If (X, *) and (Y, *) are based topological spaces, then the projections of ? ? ? on X and on ? define group isomorphisms: ??(? ??,*) -^ ??(?, *) ? ??(?, *). We shall often identify these groups via this isomorphism. The identification 5" = In/dln of Example 5 identifies maps E",*) -> (?,??) with maps {In,dln) —> (X,x0), and then U*9Kti,...,tn) - < , . ? - . When ? > 1 we could use the first coordinate t\ instead of tn to define a sec- ond product, f*g. Suppose ? > 1. Then a simple check shows that there are homotopics rel dln: f * g ~ (f*c). * (&g) = (f*c)i(c*g)~f*g, and 9 * / ~ (c*g) * (f*c) = (c*f)*(g*c)~f*g. Hence ??{?,??) is abelian for ? > 2. We recall Hopfs calculation of ni(Sn), i < n; the first assertion is a corollary of the cellular approximation Theorem 1.2. Theorem 1.3 [159] (i) For i < n, ¦ni(Sn) - 0. (ii) For ? > 1, ??E?) - ?, with 1 G ? represented by the identity map of Sn. a Given a path w : I -> X and a map / : (Sn,*) —> (X,w@)) we obtain a map / · w : (Sn, *) —4 (X,w(l)) as illustrated in the picture collapse right hand circles to poinis
12 ? ??? complexes, noraoiopy groups ana conorauons This correspondence defines an isomorphism nn(X,w@)) ^4 ??(?,??A)), and this isomorphism depends only on the homotopy class of torel {0,1}, cf. [159]. Thus if X is path connected the groups ??(?, ?), ? S X, are all isomorphic. If they vanish for 1 < ? < r then X is r-connected (when r = 1, X is called simply connected). When X is simply connected the isomorphisms ??(?,?) = irn(X,y) are independent of the choice of path, and we write simply ??(?)- (c) Weak homotopy type. A continuous map./ : Y —> ? is a weak homotopy equivalence if ??(/) and each 7T„(/) : *n{Y,y) —> *n(Z,f{y)), y G ?, ? > 1, are bijections. Two spaces X and Y have the same weak homotopy type if they are connected by a chain of weak homotopy equivalences X 4— Z@) —>·¦¦<— Z(n) —»· y. A cellular model for a topological space ? is a CW complex X, together with a weak homotopy equivalence / : X —> Y. Theorem 1.4 (Cellular models theorem) [160J (i) Every space Y has a cellular model f : X —> Y. (ii) If f : X' —> Y is a second cellular model then there is a homotopy equiv- alence g : X -=> X' such that /'oj~/. To prove this theorem one needs the fundamental Lemma 1.5 (Whitehead lifting lemma) Suppose given a (not necessarily com- mutative) diagram A-*-^ Y f together with a with a homotopy H : Ax I —> ? from ipi to /?. Assume (X, A) is a relative .CW complex and f is a weak homotopy equivalence. Then ? and ? can be extended respectively to a map ? : X —> Y and a homotopy ?: ? ? I —> ? from ? to f ? ?. proof: As in the proof of Step (iii) in Theorem 1.2 it is enough, by induction on the cellular structure, to consider the case that A = Xn and X = Xn+i- Then working one cell at a time reduces us to the case that A = Sn, X = Dn+1 and i is the standard inclusion. In this case ? : (Sn,*) —> (Y, y0) and / ? ? ~ ^1 ~
Homotopy Theory 13 the constant map. Since / is a weak homotopy equivalence, ? itself is homotopic to the constant map via a homotopy H'. This' produces the map ? : Sn x / U Dn+1 ? {0,1} -> (?, /(y0)), described in the following picture (for ? = 1) Since Sn ? IUDn+1 ? {0,1} is homeomorphic to 5n+1 and since / is a weak homo- topy equivalence, there is a map r : (Sn+1,*) —> (l',j/o) such that ??+?(/)[?] = [?]. Recall the homeomorphism ?5? -=> Sn+1. This identifies r as a map r : En ? 7,5" x {0,1}) —»· (y,y0). Define ? : Dn+1 —»· y to be the map Then the map ? : Sn ? I U Dn+1 ? {0,1} -—>· ? given by ? ? may be redrawn as
14 1 CW complexes, homotopy groups and cofibrations Hence ? represents [?] *[f ??] = 0, and so it extends to ? : Dn+1 ? I —> Z, as desired. ? Corollary 1.6 If X is a CW complex and f : Y —> ? is a weak homotopy equivalence then composition with f induces a bisection /# : [X, Y] —> [X, Z] of homotopy classes of maps. Similarly, if (?,??) is a pointed CW complex then /# : [(X,xo),(Y,yo)} —»· [(X,xo),(zJ{yo))] is a bisection. proof: The lifting theorem, applied with the relative CW complex (?,?), shows that /# is surjective. Regard / as a CW complex with /o = {0,1} and h — I. Then the lifting theorem applied with the relative CW complex (? ? ?, ? ? {0,1}) shows that /# is injective. A slight variation of this argu- ment gives the pointed case. ? Corollary 1.7 (i) A weak homotopy equivalence between CW complexes is a homotopy equiv- alence. (ii) If (X, A) is a relative CW complex and A has the homotopy type of a CW complex then X has the homotopy type of a CW complex. proof of Theorem 1.4: First we have to show that any space Y has a cellular model. It is sufficient to consider the case Y is path connected. We construct f :X —*Y with X the union of subcomplexes X° C X1 C · · · C Xn C · · · and fn the restriction of / to Xn. Fix a basepoint yo ?Y and choose /° : (X°, *) = VQ,n S? —> {Y,yo) so that ??(/°) is surjective for all i. Assume fn~l : I"'1 —>· Y is constructed so that also ?^/") is injective for i < n. Since Sn = *Uen, it follows from the cellular approximation theorem that 7?? ((In-')n) —»· ???(?) is surjective. Thus we may add (n + l)-cells {e?+1} to (Xn~1)n to kill ker7Tn (/n-1), and then extend over these cells to produce fn : Xn = I11 U (y?e%+1^J —> Y. The cellular approximation theorem then imDlies that ?{ (/") is injective for i < n.
Homotopy Theory 15 The uniqueness (up to homotopy) of a cellular model follows at once from the Whitehead lifting lemma and its corollaries. ? (d) Cofibrations and NDR pairs. Suppose A is a subspace of a topological space ?'. The pair (X,A) is a cofibration if for any continuous map f : X —> ? ? homotopy ? : A x I —> Y starting at f\A always extends to a homotopy ? ? I —> Y starting at /. Lemma 1.8 (A', A) is a cofibration if and only if ? ? {0} U A x I is a retract of ? ? I. proof: A continuous map / : X —> Y and a homotopy H : A x I starting at f\A define a map (/, H) : X x {0} U ? ? I —>· Y. If r : ? ? I —»· ? ? {0} UA ? I is a retraction then (/, H) or : ? ? I —> Y extends H and starts at /. Conversely, if (A", A) is a cofibration take Y = ? ? {0} U Ax I and (/, ?) the identity map. An extension of ? starting at / is a retraction r : ? ? I —> ? ? {0} U ? ? ?.? Again suppose A is a subspace of a topological space X. Definition (i) The pair (X, A) is a DR pair if A is a strong deformation retract of X and A = h'1 @) for some continuous function h : X —> I. (ii) The pair (A, A) is an NDR pair if for some open U C X there are continuous maps H :U ? I —> X and h : X —> I such that: • A = ?-^?) and U D h-l([0,e)), some ? > 0. • ? is a homotopy rel A from the inclusion of U in X to a retraction U —> A. (iii) A well-based space is a an NDR pair (X,x0). Proposition 1.9 (i) If (A, A) and (Y, B) are NDR pairs then (? ? ?, ? ? ? U ? ? ?) is an NDR pair. (ii) A relative CW complex (A, A) is an NDR pair. proof: (i) Let H, h, U be as in the definition for the NDR pair (A, A) and let A,fc, V be analogous maps and open set for (Y, B). Define ? : ? ? ? —> I by t{x,y) = inf (h{x),k(y)). Choose a continuous function a : (?, [?,?/2], [?, 1]) -—>· (? 1,0) and define ?' : ? ? I —»· ?, ?' : Y x I —»· Y by ??? ? - S H(x,a(hx)t) , x { ' ' ~ \ ? , h hx > ? and
16 1 CW complexes, homotopy groups and cofibrations Ky ' \ y , ky > ? . Define L:A'x7xMArx Y by ({^t),K'(y,t)) , k(y)<h(x) L(x.,y,t)=l )) Set ?/' = h'1 ([0,?/2)), V = k'1 ([0,?/2)) and W = U'xYuXx V. Then L, t, W exhibit (?' ? ?, ? ? YuX ? ?) as an NDR pair (where we use ?/2 instead ???). (it) Define a continuous function h : X —> I by induction on the skeleta, as follows. Set h = 0 in A. If h is defined in Xn~i and /Q : 5" —>· X„_i is the attaching map of an n-cell ?>" then let || || be the length function in ?>? and extend h to Xn by Then A = /i-1@) and ? ([0,1/4]) is contained in the canonical extension of A to an open set U in X (away from the origins oa of the cells ?>"). We show now that A is a strong deformation retract of U. Put Un = U ? Xn, and note that t/0 = A. Define a retraction rn : Un —> Un-i by rn{x) = /?(?/||?||), ? S UHD2, and define Kn : Un ? I —> Un by JiTntz, i) = A - t + t/\\x\\) ?., ? G U ? D^. Thus ??? exhibits C/n_i as a strong deformation retract of Un. Put rfc_n = rfc ? · · · ? rn : Un —> Uk-i· Then a retraction r : U —> A is defined by rx = ri>n(x), ? G Un. Set 7n = hj^y, ^ and turn ATn into a map iin : Un ? In —> Un via the obvious map Un,;^,^) —> G,0,1). Define H : U ? / —> U by setting (for ? G f/n) [ iffc (rfc+1>n(z),i) , t G 4, 1 < k < ? . Then ? exhibits A as a strong deformation retract of U. ? It turns out that the condition ?NDR pair' is only slightly stronger than the condition 'cofibration'. Proposition 1.10 The following conditions are equivalent on a topological pair
Homotopy Theory 17 (i) A is closed in X and (X, A) is a cofibration. (ii) (X, A) is an NDR pair, (in) (X x 7, X x {0} U A x 7) is a DR pair. proof: (i) => (ii): Let r = (rx, 77) : ? ? I —> X x{0}uAx I be ? retraction (Lemma 1.8). Define h : X —»· 7 by ?(?) = sup {i-r/(z,i) | t G 7}. Set [/ = /i-1 ([0,1)), and set H = rx : U ? 7 —»· X. (mJ => CinJ; Let h,U,e and ii be as in the definition of NDR pairs. Choose a continuous function ? : (/, [0, ?/2], [?, 1]) —> [1,1,0) and set ? = ah : X —> 7. Define A':Xx/-^Iby ? 7i(z,(y)z)i) , xGI/ K{x,t) = < \ ? , hx > ? . Define A; : X —»· 7 by fcx = inf (fAz,l). Then ? : Ar1 ([0,1)) x {1} —»· ^4· Define a homotopy ? : (? ? 7) ? 7 —>· ? ? 7 by , t < kx { (K(x,s), t-skx) , t>kx. It is straightforward to see that ? is a homotopy rel ? ? {0} U A x I from idxxi to a retraction onto ? ? {0} U A x I. (in) =$> (i); This is Lemma 1.8. ? Proposition 1.11 Suppose {X,A) is an NDR pair. Then the inclusion i : A —> X is a homotopy equivalence if and only if (X,A) is a DR pair. In this case (? ? ?,? ? {0,1} U A x 7) is a DR pair. proof: If (X, A) is a DR pair then i is certainly a homotopy equivalence. Sup- pose that i is a homotopy equivalence. Choose ? : X —^ A so that ?? ~ id a· Extend the homotopy to a homotopy ? ? I —> A from ? to a map r : X —> A. Clearly ri = id a- Now, since i is a homotopy equivalence, there is a homotopy ? : ? ? I —> X from idx to ir. Define .FT : X xIx{0}uX x{l}xIUAxIxI -^ X by setting and H(irx,2-2t) , \ <t < 1 , = irx , Ji(o,2(l-5)i) , 0<i<i K(a,t,s) = < , It is easy to construct a homeomorphism
18 1 CW complexes, homotopy groups and cofibrations (/x/,/x {O}U{1} ?/) {???,?: Thus {? ? ? ? 7, ?" ? ? ? {0} U ? ? {1} ? ? U ? ? 7 ? 7) S (? ? 7 ? 7, ?" ? 7 ? {0} U ? ? 7 ? 7), which is a DR pair by Proposition 1.10. In particular K extends to a map ? : ? ? ? ? I —> X, and K(x, t, 1) is a homotopy rel A from idx to ir. Thus (X, A) is a DR pair. Finally, since (X, A) is a DR pair, clearly A x {1} and A x {0} are strong deformation retracts of ? ? {0} and X x{l}, and so A x I is a strong deformation retract of ?' ? {0,1} U A x 7. Since the composite Ax I —> ? ? {0,1} UA x 7 —> ? ? I is also a homotopy equivalence it follows that the inclusion ? ? {0,1} U A x I —> ? ? I is a homotopy equivalence as well. Thus (? ? 7, ? ? {0,1} U A x 7) is a DR pair by the first assertion of the proposition. ? (e) Adjunction spaces. If (X,A) is an NDR pair and f : A —> Y is a continuous map then it is immediate that (Y \Jf X, Y) is also an NDR pair. (Note that a continuous function h : X —> I with /i-1@) = A can be used to separate the points of Y from the points of X — .4, so Y U/ X is Hausdorff.) Lemma 1.12 If (X, A) is an NDR pair and /o,/i : A continuous maps then Y U/o X — Y Uf1 X. Y are homotopic proof: Choose a homotopy H : Axl —> Y from j? to f\. DenoteXx{i}UAx7 by Bt C X x I. Proposition 1.10 implies that each Bt is a strong deformation retract of ?" ? 7. Hence Y Uh Bo and Y Uh ?? are strong deformation retracts of Y UH (? ? 7). But Y Uh Bo = 1' U/o A and Y \JH ?? = Y Ufl A. D Next, consider a commutative diagram y ~* -?. *" ji. of continuous maps. Theorem 1.13 If i and i' are the inclusions of NDR pairs and ?$??,?? an ?? are homotopy equivalences then ? = (??,??) ¦ Y U/ X —> Y' ur X'
Homotopy Theory 19 is a homotopy equivalence too. The proof will rely frequently on the following obvious remark. Lemma 1.14 If(B, C) is a DR pair and h : D —> W is a continuous map from a closed subspace D C C then (W Uh B, W U^ C) is a DR pair. In particular, the inclusion W Uh C <^-> W \Jh ? is a homotopy equivalence. ? proof of Theorem 1.13: Identify (A',.4) with (X,A) ? {1} C X x 7. Then (? ? I, X) is a DR pair (trivially) and {? ? ?, ? ? {0} U A x 7) is a DR pair by Proposition 1.10. Moreover / is identified as a map / : A x {1} —> Y and by Lemma 1.14, Y U/ (X x {0} U A x 7) —> V U/ (? ? /) f- ? U/ X are homotopy equivalences. Denote X \J Y by ? and </?? JJ </?y by </?z, and define g : A x {0,1} —> ? by p(o,0) = io and g{a, 1) = /a. Then Y Uf (? ? {0} U.4x7) = Z Us (.4 ? I). It is thus sufficient to show that (??,?? x id) : ? Us {A x I) —> Z' Uff- (.4' ? 7) is a homotopy equivalence. Since this map factors as ? Up {A x I) —> Z' Uvzg {A x I) —> Z' LV (-4' x /) it is sufficient to consider the two special cases: either A = A1 and ? a — id or else ? — ?' and ? ? = id. Case 1: ? = ?' and ?? = id: Regard ? a as a map ? ? {0} —> A' and set ? = ?' ??? (? ? I). Define a retraction r : ? —> A' by r(a, t) — </>>?<2, and set gB = g> ? (r x ii/) : ? x {0,1} —> ? . If C c ? let 5c : C x {0,1} —>· Z be the restriction of g ?. Thus the inclusion of C in ? and the retraction r define an inclusion and retraction ? Usc (Cxi) ^ ? UgB (???)?? Uff. (.4' ? 7) . Now suppose {B, C) is a DR pair. Then so is {? ? ?, ? ? {0,1} U C x 7), by Proposition .1.11. Since ? USc (C ? 7) = Z UgB (B x {0,1} U C x 7) it follows triat je is a homotopy equivalence (Lemma 1.14). In particular {B, A') is a DR pair and, clearly, qja1 — id. Thus ? is a homotopy equhralence. Moreo\rer, if we identify A = A x {1} C B then the inclusion is a homotopy equivalence
20 1 CW complexes, homotopy groups and cofibrations because ? a is. Thus (B,A) is a DR pair (Proposition 1.11) and Ja is a homotopy equivalence. Finally, (???,?? x id) = QJ.a and so it is a homotopy equivalence too. Case 2: A = A' and ? a = id: Denote (AxI, Ax{0,1}) by (B,C). Thus g : C —> ? and we have to show that if ?? : ? —> ?' is a homotopy equivalence so is (??, id) : ZUgB —> ?' ???9 ?. Suppose first that ? — ?' and id ? ~ ?? via a homotopy ? : ? ? ? —> ?. Set ? = ? ? (g ? id) : C ? ? —> ? and set ? = (?, idBxi) ¦¦ (ZUg ?) ? I —> ? \JH (? ? I) . In the proof of Lemma 1.12 we observed that the inclusions b ?—> (?,?) and b ?—> (b, 1) of ? in ? ? I induce homotopy equivalences j0 : ? \Jg ? —> ? U# (B x I) and ji : ? ???9 ? —> ? U// (? ? I). A quick check shows that ? is a homotopy from jo to ji ? (ipziid). Hence (? ?-.id) is a homotopy equivalence. Now suppose ?? is any homotopy equivalence and choose a homotopy inverse ?? '¦ ?' —> ? : ???? ~ id ?1 and ???? ~ id z- Form the sequence ? \Jg ? —> ? Upzs ? —> ? ^¦????9 ? —> ? {->??????9 ? , where 70 = (? ? Ad), 7? = {'? ?, id) and 72 = (? ?, id). The argument above shows that 7170 and 7271 are both homotopy equh'a- lences; i.e., they have homotopy inverses a and ?. Thus ??? ~ /572717??: ~ 7ou, and so 7i(/5727i7ou) ~ l\?li ~ 7i7ou ~ id ? and (/5727i7ouOi ~ idz· · Thus 71 is a homotopy equivalence. Hence so is 70 = (??, id). ? Corollary If (X, A) is an NDR pair and A is contractible then the quotient map q : X —> XjA is a homotopy equivalence. proof: Identify q as the map (c, id) : A\JaA X —> pt Uc X, where c : A —> pt, and note that c is a homotopy equivalence by hypothesis. ? (f) Cones, suspensions, joins and smashes. The cone, CX, on a topological space X is the space ? ? IjX ? {0}, and the point [? ? {0}] is· called the cone point. We usually denote the point (x,t) by tx, so that Ox is the cone point (any ? ? X). This identifies X as the subspace X x {1} and clearly (CX,X) is an NDR pair. Recall that the suspension of a based topological space (X, xq) is the based space ?? = (?? ?)/? ? {0,1} U {x0} xl = CX/X U {x0} x I . Any continuous map / : (X, x0) —> (Y, yo) suspends to the map ?/ : ?? —> ?? induced from / ? idi. Note that if (X, x0) is well based then (? ? ?, ? ? {0.1}U (??\ ? I) is an NDR pair (Proposition 1.9). Hence ?? is well based. Moreover, if
Homotopy Theory 21 (F, y0) is also well based then Theorem 1.13 implies that homotopy equivalences suspend to homotopy equivalences. The join of two topological spaces X and F is the subspace ?' * Y C CX x CY of points of the form (tx, A - t)y), tel. (Thus X * Y is the union of intervals joining ? ? X to y 6 Y.) Notice that a homeomorphism {CX x F) UXxy (? ? CY) -=> X * Y A.15) is given by (iz,y) ?—> (far, (l- f) y) and (z,iy) ?—> ((l- f ) ?, |y). More generally, the ?-fold join ?? * · · · * Xn of topological spaces X; is the subspace of CX\ ? · · · x CXn of points (iiXi,.... inxn) satisfying ??; = 1. Note the ob\dous identifications (?? * X2) * Xz = Xi * X-2 * X3 = Xi * (X2 * X3) · Next, recall that the wedge X V Y of based spaces (X,xq) and (V, yo) is the subspace X x {yo} U {^o} x F of ?' ? F. The smash product of A" and F is the space ? ? F = ? ? ?/? V ? . In particular ?? coincides with ? ? S1 and so XAEF^XAFAS1 = ?(? ? F) . A.16) Proposition 1.17 // (X,x0) o,nd (F, yo) are well based spaces then there is a homotopy equivalence, X * F —t ?(? ? F). proof: Embed / as Ix0 and Iyo in CX and CY. Identify CX U/ CF as the subspace of X * F of points (tx, A — t)y) such that either a; = x0 or y = yo· Now (CX U/ CF)/J = (CX/J) V (CF/J). Since I is contractible we can apply the Corollary to Theorem 1.13 to obtain CX U/ CY ~ (CX U/ CF) // ~ CX V CY ~ {pi} . Hence (by the same Corollary) the quotient map X * F —> X * Y/CX U/ CF is a homotopy equivalence. But if q : X x F ? ? —> ? * F is the quotient map (x, y, t) ?—>· (tx, A - i)y) then q-1 (CX U/ CF) is just (X V F) ? I. Hence c induces a homeomorphism ?(? ? F) -=> ? * F/CX U/ CF. D Exercises !· Prove that F(X,x0) is contractible and that X and Xi0'1) have the same homotopy type.
22 1 CW complexes, homotopy groups and cofibrations 2. Let ?, ? and ? be based spaces. Prove that (XvY)aZ =* (XAZ) V(V ??), (? ? y) V ? = (? V ?) ? (y V ?) and (? ? y) ? ? ~ ? ? (y ? ?). 3. Prove that ? * y is homeomorphic to the inverse image of the set {(s.,t) 6 [0,1]2 ,5 + t = 1} under the mapping CX ? CX -» [0.. If, ([z,s], [y,i]) ^ (s,i). 4. By considering the mapping Em+1 ? ??+1 ? [0,1] -> Em+n+2 ,{x,y,t) i-> (x cos ^, y sin f ), prove that 5n * Sm = 5m+n+1. 5. Let / : S3 -> CP1 = S2 be the map defined by f{zuz2) = [21,23]· Prove that there exists a homeomorphism from the cofibre of / onto CP2. 6. Let (?,??) and (Y,yo) be based spaces. Prove that the inclusion ?' V y c—? ? ? I' is a cofibration whose cofibre has the same homotopy type as ? ? ? and that ?(? ? y) has the same homotopy type as ??' V ?? ??(?'? ?).
2 Fibrations and topological monoids A continuous map ? : X ->· Y has the lifting property with respect to a pair of topological spaces (Z. A) if for every commutative diagram of the form .4 -J-+ X 9 there exists a continuous map k : ? -^ X such that pk = g and ki = f. A surjective continuous map ? : X -» Y is a 5erre fibration if it has the lifting property with respect to (? ? ?, ? ? {0}) for all CW complexes Z, and a fibration if it has the lifting property with respect to (? ? ?, ? ? {0}) for all topological spaces Z. Here ?' is called the total space, ? the projection and Y the base of the fibration. The space Xy = p~xB/) C X is called the fibre at y. If Y has a fixed basepoint y0 we often denote the fibre at y0 by F and denote the (Serre) fibration by F -> X A Y. A fibre-preserving map between (Serre) fibrations is a commutative square of continuous maps X' —j~^ X Y' Y in which p' and ? are (Serre) fibrations. In this section we first establish a long exact sequence connecting the homotopy groups of the fibre, total and base spaces of a fibration. Next, we construct the Moore loop space UX of a based topological space (X,*). This is a topological monoid acting on the contractible path space, PX, which identifies UX as the fibre of the path space fibration PX —> X. This construction generalizes. Given a continuous map / : X —> Y there is a homotopy equivalence Xxy MY -^» X which converts / into a fibration whose fibre Xxy PY admits a natural action of ??. The projection ? ? ? ?? —> X is then also a fibration (the holonomy fibration) with fibre ?? given by the action. Finally, we consider principal bundles with fibre a topological group and con- struct their classifying spaces. This section is organized into the following topics: (a) Fibrations . (b) Topological monoids and G-fibrations. (c) The homotopy fibre and the holonomy action.
24 2 Fibrations and topological monoids (d) Fibre bundles and principal bundles. (e) Associated bundles, classifying spaces, the Borel construction and the holonomy fibration. (a) Fibrations. We begin with some examples. Example 1 Products. The projection F x F —> Y is a fibration (the trivial fibration). ? Example 2 Covering projections. These are fibrations [68]. ? Example 3 Fibre products and pullbacks. Given continuous maps A —> Y ^- X, recall that the fibre product A xy X C A x X is the subspace of pairs {a,x) such that f(a) = p(x). It fits into the commutative pullback diagram ??? ? A X Y in which ? a and g are projection on the first and second factors. If ? : X -» Y is a (Serre) fibration then so is ? a '· A x ? ? -» .4. It is called the pullback fibration. When / : A -» Y is the inclusion of a subspace then A x y ? = ? (.4) and the fibration ? a is called the restriction of the original fibration to A. ? Proposition 2.1 (i) A fibration ? : X -» Y has the lifting property with respect to any DR pair (Z,A). In particular, if (W,B) is any NDR pair then ? has the lifting property with respect to (W ? I, IV x {0} Li ? ? I). (ii) A Serre fibration ? : X -» Y has the lifting property with respect to any relative CW complex (Z, A) for which the inclusion A -t ? is a weak homotopy equivalence. In particular, if{W,B) is any relative CW complex then ? has the lifting property with respect to (W ? I, W ? {0} Li ? ? I). proof: [157] In both cases we suppose given a commutathre square Y
Homotopy Theory 25 and we have to construct k : ? -> X. (i) Let r : ? -> A be a retraction and let iT : ? ? ? -> ? be a homotopy rel A from ir to idz- Choose a continuous map h : ? -> I such that A = h'1 @). Define a continuous map if' : ? ? ? —? ? by H'{z,t) = H{z,t/h{z)) , t<h(z) ? , t > h{z). Then gH1 (z,0) = gr(z) = pfr(z). Thus there is a continuous map ? : Zxl -? ? such that pK = gH' and K(z, 0) = /?). Set k(z) = ? (?, /?(?)). For the second assertion apply Proposition 1.9. (ii) The Whitehead lifting lemma 1.5 provides a retraction r : ? -» ? and a homotopy rel ?, ? : ? ? I -» ?, from ir to idz· Assume iv : ??_? ? 7 -> JV is defined so that K{a,t).= /(a), a ? A, t ? I, K(z,0) = /r(z), ? ? Zn-U and pii = gH. We extend it to ? : Zn ? I -> ? as follows. We have to fill in the dotted arrows in the Klqxid)\Jfr Write Zn = Zn_! U9 commutative solid arrow diagrams (S2'1 ? 7) U {Dl x {0}) Z ? ? Y But this is possible since ? : ? —» ? is a Serre fibration and since {Dn ? 7,5?-1 x/UU"x {0}) = (?>n ? I,Dn ? {0})· This constructs a ho- motopy ? : ? ? I -> X. Set fc(z) = AT (?, 1). D Fix a Serre fibration X in which (X,Xq) is a path connected based space, y0 = pxc, and F = Xyo. The homotopy groups of F, ?' and Y are tightly related, and this is expressed through the natural connecting homomorphism which we now describe. Let g : En,*) —> (Y,yo) be a continuous map and, as in Example 5, §1, identify ?5"-1 = 5" via the map ? : S" ? ? —> Sn. Form the commutative
26 diagram 2 Fibrations and topological monoids S" X {0} U {*} X I constant map to ? x I sn Y Since ? is a Serre fibration, Proposition 2.1 asserts that gQ can be lifted through ? to a continuous map ? : Sn~l x / —> X, which then restricts to a map Proposition 2.2 [93] With the notation above the correspondence g ~* h de- fines natural set maps dn : ??(?,yo) —> nn^i(F,xQ). When ? > 2, i/iese are group homomorphisms, and fit into a long exact sequence (the long exact homo- topy sequence) When ? = 1, 9? : 7??(?) factors to give a bijection proof: Put W = 5n ? J and ? = 5n ? {0} U {*} ? I. Then the inclusion W x {0,l}U?x/ C—>· W x I is a weak homotopy equivalence. Hence it follows from Proposition 2.1(ii) that the based homotopy class of h is independent of the choice of H and depends only on the based homotopy class of g. Recall from §1 that maps (Sn,*) —> (X,Xq) may be identified with maps (In,dln) —> (X,x0), and that, when ? > 1, composition along the first coor- dinate defines a second product, fig, such that fig ~ / * g rel dln. But it is immediate from the definition that when ? > 2, dn[f*g] = [dnf]*[dng]\ hence <9n is a group homomorphism. The exactness of the sequence and the assertion about d\ are straight forward consequences of the definition and Proposition 2.1(ii). D Next let A Xy X X Y
Homotopy Theory 27 be a pullback diagram (i.e., q and gx are the projections on ,4 and X). Recall that if g is the inclusion of a subspace .4 C Y then gx is the inclusion of Proposition 2.3 Suppose in the diagram above that ? is a fibration. (i) If g is the inclusion of a DR pair (resp. an NDR pair), (Y, A) then (X, XA) is also a DR pair (resp. on NDR pair). (ii) If g is a homotopy equivalence so is gx. proof: (i) Suppose (Y,A) is a DR pair. Let H :Y x / —> Y be a homotopy rel A from idy to a retraction r : Y —> A. Let h : Y —> I be a continuous function such that h~l@) = A and define a homotopy H' : X x I —> Y by j H(px,t/h(px)) , t<h(px) . ? (x.t) = < . v v ; | px , t> h(px) . Lift this to a homotopy K' : ? ? J —>· X starting at id ?. Define ? : X xl —> X by ? ??(?,?) , ?</?(??) . (X' j \ K'{x,h{px)) , t>h{px) . Then ? exhibits X^ as a strong deformation retract of X. Thus ? together with k = hop exhibits (X,Xa) as a DR pair. An identical argument shows that if {Y,A) is an NDR pair then so is (X,Xa)· (ii) Regard g as a map ? ? {0} —> Y, and let r : (? ? J) Us Y —> Y be the retraction sending (a, t) to ga. Denote (A x /) Us Y by ? and form the pullback diagram ZXyX —?±-+ ? Now identify A = A x {1} C Z. Then the inclusion of .4 in ? is a homotopy equivalence, because g is, while the inclusion of Y in ? is obviously a homotopy equivalence. Since (Z,A) and (Z,l') are clearly NDR pairs, Proposition 1.11 asserts they are NR pairs. Now assertion (i) of this proposition implies that (? ???,? xyX) and (? ???,? ?? X) are DR pairs. But Y xy X = X and ?? ·. ? ?? ? —>. ? is a ]eft inverse for the inclusion. Hence ? ? is a homotopy equivalence. Moreover, gx factors as A x y ? —> ? ? y X -^ A", and so it is a homotopy equivalence too. ?
28 2 Fibrations and topological monoids Finally, let be a fibre preserving map between Serre fibrations with path connected total spaces. Then from Proposition 2.2 and the five lemma (see Lemma 3.1) we deduce Proposition 2.4 Let y' € Y' be any basepoint. If any two of the maps f, g and fy> : X', —\ Xgy> are weak homotopy equivalences, then so is the third. ? (b) Topological monoids and G-fibrations. A topological monoid is a topological space G equipped with a continuous, associative multiplication ? : G x G —> G and a two sided identity e € G. A (right) action of G on a space ? is a continuous map ? ? G —> ?, (?, g) *-> ? ¦ g, such that ? ¦ e = ? and (? ¦ g\) · gi = ? · (gig?), for ? € ? and g,g\-,g2 ? G. Suppose ?' ? G' —> P' is a right action of a second topological monoid G'. Then an equivariant map consists of a morphism 7 : G —> G' of topological monoids and a continuous map / : X —> X' such that /(?·?) = f(x) -7@). If / and 7 are weak homotopy equivalences then (/, 7) is called an equivariant weak equivalence. Definition A G-(Serre) fibration is a (Serre) fibration ? : ? —y X together with a right action ? ? G —> ? such that (i) p(z ¦ g) = pz, ? G P, g G G, and (ii) For each ? G ? the map Az : G —> Ppz sending g t-> ? ¦ g is a weak homotopy equivalence. Remark 1 If ? is path connected and some Az is a weak homotopy equivalence then so is every Aw, u> € P. Indeed, join ? and u> by a path ? : I —> P. Then we have a map of fibrations I xG / , f(t,g) = (?,?(?) -g). / Let ft be the restriction of / to the fibres at t. Then Proposition 2.3 asserts that /o is a weak homotopy equivalence ¦$=$> f is a weak homotopy equivalence <?=^ fi is a weak homotopy equivalence
Homotopy Theory 29 But /o = Az and /i = Aw. Remark 2 If / : Y —> X is a continuous map the pullback ? ? ? ? —> ? of a G- (Serre) fibration is again a G-(Serre) fibration with action (y, z)-g = (y, zg), y€Y,z€P,g€G. Example 1 Path space fibrations. As usual, XY denotes the space of all continuous maps Y -> X. Injparticular, let P(X,x0) C X1 be the subspace of paths ending at xo, and let Cl(X,x0) C P(X,:ro) be the subspace of paths beginning and ending at xo· Then it follows easily from the exponential law that ?(?',??) > P(X,xo) —^ ? , ? : 7 —> 7@), is a fibration, and that PX is contractible. Note also that a continuous multi- plication in QX is defined by f 7Bi) 0 < t < i G,?) ?—»7*?, G*?)(?) = < , , ?> ' ' w ;w [ ?B?-1) I <i < 1 . This is 'homotopy associative', but not associative. These constructions are the classical version of the path space fibration and the loop space for X. It turns out to be highly convenient to use instead Moore's version of the path space fibration and loop space, since the latter is a genuine topological monoid. Thus a Moore path in X is a pair G, ?) in which 7 : [0,00) -> X is a continuous map, i € [0,oo), and j(t) = ?(?) for t > L The path starts at 7@), ends at j(?) and has length t. The space MX C Xt0·00) ? [0, oo) of all Moore paths is the free Moore path space on X. The Moore path space P(X,xo) C MX is the subspace of all Moore paths ending at Xq and the Moore loop space ?(?,??) is the subspace of all Moore paths starting and ending at xo· As above, we usually abuse notation and write simply QX and PX. We may also abuse notation and denote G, ?) simply by 7. From the exponential law it follows easily [157] that if ?" is path connected then UX —* PX —^-> X pG, t) = 7@) is a fibration and that PX is contractible. This is called the Moore path space fibration for (X,x0). Note that QX and PX can be identified as the subspaces of ?.? and PX of paths of length one, and that these inclusions are homotopy equivalences. If 7 € PX and ? € ?? are paths of lengths ? and r we define their product to be the path 7 * ? of length ? + r given by 7@, 0<i<^
30 2 Fibrations and topological monoids This product, restricted to QX ? QX, makes QX into a topological monoid with identity the constant loop of length 0 at zo· It is called the Moore loop space of X at xo· The original map, PX ? QX —> PX is a continuous right action of the monoid QX on the contractible space PX. Note that p(j * ?) = py, 7 e PX, ? 6 QX. Moreover ??' is the fibre (PX)Xo and the map Ae : QX —> (PX)Xo is just the identity, and so a weak homotopy equivalence. Thus it follows from Remark 1 that the Moore path space fibration PX —> X is an ??-fibration. We shall often abuse language and refer to PX and QX simply as the path space and loop space of X. Finally, if / : (?',??) —* (Y, 2/o) is any continuous map then ?/ : 7 1—> / ? 7 defines a morphism of topological monoids, ?/ : QX —»· QY. In particular, suppose F -? ? -^» ? is a fibration with path connected base, Y and path connected fibre, F. Choose basepoints yo € Y, xq € X and zq € F so that F = Xyo and j(.zo) = xo· It follows easily from the exponential law and Proposition 2.1 that QX -? ?? is a fibration with fibre ?^. Moreover the multiplication in QX restricts to an action of QF on QX, and so the pair QX % QY , QXxQF^ QX (c) The homotopy fibre and the holonomy fibration. Let / : X —> Y be a continuous map between path connected spaces and let <7o,<7i ·" MY —> Y denote the maps G, t) 1—> 7@) and G, t) 1—> j(?). Let cy be the path of length 0 at y. Form the commutative diagram X xY MY · X xy MY is the fibre product with respect to , where / and go • j(x) = (x,cfx) It is easy to verify that j is a homotopy equivalence and g is a fibration: we say the diagram converts f into the fibration q. Fix a basepoint y0 ? Y. The fibre of q at y0 is I x}- PY. It is called the homotopy fibre of f. Clearly the action of QY on PY defined above defines an action of QY on the homotopy fibre. Moreover the diagram PY , ? = projection on the first factor,
Homotopy Theory 31 exhibits ? as a pull back HF-fibration. It is called the holonomy fibration asso- ciated with the continuous map f : X —> Y. Let ? : ? ?? MY -t X be the projection; it is a homotopy equivalence inverse to j. The commutative diagram ?' xYPY ? ? ? MY ? ? identifies ? with i and / with q, 'up to homotopy'. Thus the long exact homotopy sequence for the fibration q translates to a long exact sequence irn(X ?? PY) ???(?') ???(?) xY PY) Finally, write F = f~1(yo)· Then j restricts to an inclusion jo : F —> X xY PY. Proposition 2.5 (i) If ? is a fibration then (? ? ? PY,F) is a DR pair. In particular, jo is a homotopy equivalence. (ii) If ? is a Serre fibration then jo is a weak homotopy equivalence. proof: (i) Define a commutative square of continuous maps X xY PY ? {0} X xY PY ? [?,??) X Y by setting ?(?, -y, t) = ~f(t). Lift ? to a continuous map ? : ? ?? PY ? [0, oo) X so that ?? = ? and ?(?,7,0) = ?. Then a homotopy ? : X xy PY x I —> ? ?? PY is given by where ?? is the length of 7 and jt is the path of length A - t)?7 given by jt(s) = ?{?? + s). Clearly ii is a homotopy rel F from the identity to a retraction X xy PY —> F. Finally, if h : X x Y PY —> / is denned by h(x, 7) = inf A, L{) then /?-1@) = F. CuJ This follows immediately from Proposition 2.3 since j is a homotopy equivalence. D
32 2 Fibrations and topological monoids On the other hand, we can exhibit the homotopy fibre Ff of / : X —> Y as the analogue of a homogeneous space. Let P'X C MX be the subspace of Moore paths G,^) that begin at the basepoint xq : 7@) = xo- Multiplication of paths defines a left action of ?? on P'X and the projection ? : P'X xxFf —»· Ff 2G,2) .—>¦ ? is a (left) ??-fibration. Let c_x and cy denote the constant paths of length zero at the basepoints xo and 2/0 = fxo- Then QX is the actual fibre of ? at the basepoint (??,??) of F/ and the fibre inclusion is given by ? : ?? —> P'X xx Ff, ? : 7 ·—»¦ G * Cx,zo,Cy). Thus we may regard F/ as the 'quotient' of P'X ? ? Ff by the topological monoid ??. Now recall that Ff = X xy PF- Thus P'X xx Ff = P'X xx ? ? ? ??. Define a continuous map ?-.P'XxxFf—^QY byr?G)x)w) We show that ? has the following properties: • 77 is a homotopy equivalence. • 77G · z) = (?/h * ryz, 7 e ??, ? € P'X ?? F/. • The diagram ?? 3- P'X ?? Ff commutes. Thus 'up to homotopy equivalence' the ??-spaces P'X ?? Ff and QY coincide and so F/ may be thought of as the quotient of ? ? by ??. Of the three properties listed for ? the last two are immediate consequences of the definition. For the first, notice that P'X x* X xy PY = P'X xY PY and that the projection on the first factor is a fibration P'X xy PY —>· P'X, with fibre ??. Since P'X is contractible the inclusion ? : ?? —> P'X ? y PY of the fibre is a homotopy equivalence. But ?? = identity and so ? is a homotopy equivalence too. (d) Fibre bundles and principal bundles. A product bundle with fibre ? is a. continuous map ? : X —^ Y for which there is a homeomorphism g : ? ? ? —? ? such that pg(y, ?) = y. A fibre bundle with
Homotopy Theory 33 fibre ? is a continuous map ? : X —> Y such that Y is covered by open sets ?? and the restrictions pi : p~l(Oi) —> O{ are product bundles with fibre Z. If ? : X —y Y is a fibre bundle and / : A —> Y is any continuous map then the pullback ? ? ? ? —> A is also a fibre bundle. Proposition 2.6 A fibre bundle is a Serre fibration. proof: First note that if a continuous surjection ? : X —» Y has the lift- ing property with respect to (/" ? /,/" ? {0})?>0, then it is a Serre fibra- tion. Indeed, suppose given a relative CW complex (W, A) and a homotopy g : Wxl —»· Y. Note that (Dn ? /.S" x IU Dn ? {0}) S (/« ? /,/« ? {0}). Thus if / : W x {0} U ? ? ? —> ? satisfies pf = g we may extend /, one cell at a time, to a lift of g. Suppose now that ? : X -> Y is a fibre bundle and we want to lift g : In+1 —> Y through ? extending f : In ? {0} —> X. Use the pullback over g to reduce to the case Y — 7n+1, g = identity. Recall (§1) that a subdivision of / gives a product cellular structure for /n+1. Choose the subdivision sufficiently fine that each 'little cell' is contained in an open set over which the fibre bundle is trivial. Then the restriction to each cell is a Serre fibration and the construction of the lift g, one cell at a time, is immediate. ? Now consider the following situation: • ? : X —> y is a continuous map. • Y is a CW complex with characteristic maps Fa : D% —> Yn- • The pullbacks D% x ?? —> D% (all a, n) are product bundles with common fibre Z. Proposition 2.7 With the hypotheses above ? : X —y Y is a fibre bundle with fibre Z. In particular it is a Serre fibration. proof: For each fc, the characteristic maps define a homeomorphism U {d\ - S*) A yfc - yfc_!. Thus the restriction ? : p~l{Yk ~ Yk-i) —> Yk — Yk-i is a product bundle over the open subset Yk — Yk-i of Yk. Suppose now by induction that we have extended this to an open set U C Yn-i and a homeomorphism h\j : U x Z —i p~l{U) such that phu(u,z) — u. We shall extend U to -an open subset V C Yn and hu to a homeomorphism hv : V ? ? -=> ? (F) such that phv(v,z) = v. For simplicity put D = [] D% and 5 = U^ and write Yn - Yn-\ U/ D a ., a where / : 5 —>¦ Yn_l is the attaching map. Then identify D% - {oa} = S^ x @,1], with {oa} the origin of D% and 52 identified with 5J" x {1}. Thus
34 2 Fibrations and topological monoids D - UM = S ? @,1] and V = U U, (f~l(U) x @,1]) is an open set of Yn extending U. Next we undertake the extension of hu to hv- The hypothesis above provides a homeomorphism ? : D ? ? —l· D ?? X-, compatible with the projections on D. Moreover hu pulls back to the homeomorphism hf : f~l(U)xZ -^ /~1(??)??? given by hf(w,z) = (w,h(fw,z)). Thus ?~? ? hf is a self homeomorphism of f~l(U) ? ? of the form A0,2) 1—> (w,ip(w,z)). Extend this to the homeomor- phism Of : rl{U) x @,1] ? ? A {rl(U) ? @,1]) xY X given by 9f(w,t,z) = ? ((w,t),ip(w,z)). The homeomorphisms /i?/ and ?/ are compatible and so define a unique set theoretic bijection hv making the following diagram commute: (U U rl{U)x{Q,l})xZ{I^a{UxYX) ? {{r\U) ? @,1}) ?? ?) qxid V ? ? p-\V) in which q is the obvious quotient map and q' is projection on X. Since hu and Qf are homeomorphisms we need only verify that q x id and q' are quotient maps to conclude that hv is a homeomorphism too. Since g is a quotient map so is the product q ? id. To consider q' write A = Fif'1^) x @,1])- Then U and A are closed in V and F = {/ U A. Since <?' = id : U ?? ? -=» ?~?(?/) it is enough to show that q' ¦ (/-1(i/) x @,1]) ?? X —> p~1(A) is a quotient map. Because we work with fc-spaces this reduces to the assertion: if ? C Kcomp*ct C p~1(A) and if (q')~l(B) is closed, then ? is closed. But the compact set p{K) is covered by finitely cells Fai (?";), 1 < i < N. Put ? C = \J F~^(pK). Thus C and hence C ?? ? are compact. Since p(K) C A, il C C f~l{U) ? @,1]. Thus (g1)^) n (C x^ ^) is compact. A simple check shows that ? - q' {{q'Y^i?) nC ?? ?). Thus ? is compact, and so closed in p~1(A). This completes the proof that q' is a quotient map and hence shows that hy is indeed a homeomorphism. This inductive construction leads to a subset ? C Y and a bijection /? : ? ? ? —>· ?'1 (?) with the following properties: for ? > fc, On ¦= ? ? yn is open in Yn. ^ " ^ " ™~1(O..) is a homeomorphism.
Homotopy Theory 35 Thus ? is open in Y. Moreover, since any compact subspace of Y is contained in some Yn (Proposition 1.1 (ii)) it follows that C C ? ? ? is compact if and only if h(C) is compact and that h : C -=> h(C)· This implies (because we work in the category of fc-spaces) that h is a homeomorphism. ? We turn now to the definition of topological groups and principal (fibre) bun- dles, which are important special cases of topological monoids G and G-Serre fibrations. A topological group is a topological monoid G such that the monoid is a group and the map g \—> g~l is continuous. If the topological group G acts continu- ously on a topological space X then X is the disjoint union of the orbits x-G, and the orbit space X/G is the set of orbits equipped with the quotient topology. (We only consider cases where X/G is Hausdorff.) Let G be a topological group. A principal G-bundle is a continuous map ? : X —> Y together with an action of G on X such that Y is covered by open sets Oa and there are continuous maps ?? : Oa —> X with the following properties: • ??? = id • A homeomorphism Oa ? G —i p~l(Oa) is given by (y,g) ?—> cra(y) · 9- In this case ? : X —» Y is a fibre bundle with typical fibre G, the orbits of G are the fibres of ? and ? induces a homeomorphism X/G ^> Y. In particular, ? : X —> Y is a G-Serre fibration. Now consider the following situation: • ? : X —y Y is a continuous map and A' x G —> X is a continuous action of a topological group G. • Y is a CW complex with characteristic maps Fa : D? —> Yn. • There are continuous maps ?? : D% —> D% xy X (all ?,?) such that ?? has the form ??(??) = (?),??(??)) and is a homeomorphism. Proposition 2.8 . With the hypotheses above ? : X —> Y is a principal G- bundle. Proof: It follows from Proposition 2.7 that ? : X —>¦ Y is a fibre bundle. Thus Y is covered by open sets Ui for which there are continuous maps r,· : ^ *¦ X such that ??{ = id. We have only to check that the continuous maps "-i :Ui ? G —> P~l{U{), (u, g) ?—> Ti(u) ¦ g, are homeomorphisms.
36 2 Fibrations and topological monoids It is immediate from the hypotheses above that hi is a bijection. Thus it is sufficient to show that /^'(C) is compact for any compact subspace C C p~l(Ui). Since p(C) is covered by finitely many cells (Proposition l.l(ii)) this too follows easily from the hypotheses. ? (e) Associated bundles, classifying spaces, the Borel construction and the holonomy fibration. Suppose given a principal G-bundle ? : X —> Y and an action of G on a third topological space Z. Then G acts diagonally on ? ? ? : (?,?) · g = (xg,zg). Consider the continuous maps ? ??? (X xZ)/G^Y where q is the map of orbit spaces induced by the projection ? ? ? —> X. It is an easy exercise to check that ? is the projection of a principal G-bundle and that q is the projection of a fibre bundle with typical fibre Z: this fibre bundle is called the fibre bundle associated to the principal bundle via the action of G on Z. Central to the study of principal G-bundles is Milnor's universal G-bundle [125] pG:EG^ BG constructed as follows. Recall (§l(f)) that CG = (G ? I)/G ? {0} is the cone on G, and that the nth join, G*n, is the subspace of CG ? ¦ ¦ ¦ x CG of points ((g0, ??), · · ·, (9n, tn)) such that ??? = 1. Thus G*n C G*(n+1) (inclusion opposite the base point of CG). Set EG = (J G*n, equipped with the weak topology determined by the G*n. ? A continuous action of G in EG is given by ((go,to),...,(gn,tn)) ¦ g = ((gog,t0),¦¦-,(gng,tn)). Set BG = EG/G and let pG : EG —> BG be the quotient map. It is an easy exercise to verify that • pG : Eg —> BG is a principal G-bundle. • every continuous map from a compact space to Eg is homotopic to a constant map. In particular, nt(EG) = 0. The space BG is called the classifying space of G. Now associated with an •arbitrary action of G on a topological space X are two important constructions: the Borel construction and the loop construction. The Borel construction is the fibre bundle with typical fibre X, associated to the universal bundle via the action of G on
Homotopy Theory 37 Proposition 2.9 If ? : X homotopy equivalence Xg — Y is a principal G-bundle then there is a weak Y. proof: The associated fibre bundle with fibre Eq has the form q' : XG = (EG x X) IG -i- Y. Since ?*(??) = 0 and since this is a Serre fibration, the long exact homotopy sequence shows that q' is a weak homotopy equivalence. ? Consider a principal G-bundle ? : X —> Y whose base y is a CW complex. Because q' : Xg —> Y is a weak homotopy equivalence there is a map ?' : Y —> Xg such that q'a' <~ id (Whitehead Lifting Lemma 1.4). Because q' is a Serre fibration we can lift the homotopy starting at a' to obtain a homotopy ?' ~ a : Y —> Xg with q'a = id. This identifies Y as a subspace of Xq- Restrict the principal G-bundle Eq xX —> Xg to a principal bundle ? —> Y. Projection Eg ? X —> X restricts to a map ? —> X of principal bundles covering the identity map of Y. This map is therefore a homeomorphism, which we use to identify ? = X. Thus the diagram X Y EGxX Eg Bg exhibits the original principal bundle as a pullback of Milnor's universal bundle (whence the name). An extension of this argument shows that the pullback construction defines a bijection isomorphism classes of principal G-bundles over Y thereby explaining the terminology 'classifying space' for Bg- Finally, suppose a topological monoid G acts on a path connected space X with basepoint x0. Then, as described in §2(c), the continuous map a : G —> X, a : 9 ?—> xo ¦ p, can be converted into the fibration G ?? ? ? —>¦ G ? ? MX —> ?. Moreover, this is homotopy equivalent to GxxPX X where ? is the projection of an ??-fibration. Now we show that G ? ? ?X is a topological monoid and that ? is a morphism of topological monoids. Thus X may be thought of as a sort of 'generalized
38 2 Fibrations and topological monoids homogeneous space'. To construct the multiplication on G xj PX let G act on MX by (w ¦ g)(t) = w(t) -g,w 6 MX, g eG. Then set (9, w) ¦ (g',w') = {gg, (w ¦ g') * w'), where, as usual, * denotes composition of paths. The pair (ec, cXo) is a two sided identity. We call the monoid G ? .? PX the loop construction at x0 corresponding to the action of G on X. In particular, suppose given an arbitrary principal G fibre bundle ? -^ ? , ZxG -^ ? in which ?»(?) = 0. Then morphisms of topological monoids are defined by 7E, u;) = g and 7'(<7, to) = ps ? «j . Proposition 2.10 7 anei 7' are weak homotopy equivalences, so that G and VtB are weakly equivalent topological monoids. proof: Define an action of G ? ? PZ on PZ by setting u-(g.w) = (u-g) * w. Then ? : PZ —>¦ B, n(w) = pB(u>@)), is a G xz PZ-fibration. It fits in the diagram of fibrations ? +-1— PZ —^ PB where f(w) = w@) and f'(w) — pB out. Moreover / and /' restrict to 7 and 7' respectively in the fibres. Since ?» vanishes on ?, ? ? and PB, it follows from the long exact homotopy sequence that 7 and 7' are weak homotopy equivalences.? Observe that in the proof of Proposition 2.10 that G,7) and (f',j') are equiv- ariant weak equivalences. Again let ? : ? —> ? be the principal G-bundle described just before Propo- sition 2.10. If ? : Y —> ? is any continuous map we can form the pullback principal bundle ? : ? ?? ? -^ ? , (? ?? ?) ? G -^ Y ?? ? and also the holonomy fibration (§2(c)), n:Yx.nPB—>Y , (Y xB PB) ? UB -^ Y xB PB .
Homotopy Theory 39 The diagram of Proposition 2.10 pulls back to the diagram of Serre fibrations ? Xe ? . YxBf Y x? PZ YXBf' - Y x? PB Y which identifies ? ? ? / and ? ? ? /' as equivariant weak equivalences. This extends Proposition 2.10 to Proposition 2.11 The pullback fibre bundle ? : ? ?? ? —? ? and the holon- omy fibration of ? are connected by equivariant weak equivalences of fibrations: Y xBZ Y xB PB G Exercises 1. Let F be the homotopy fiber of the inclusion i : X VY -? X xY. Prove that F ~ ?.? * ?.? and that Qi is a QF- fibration with a cross section. Assume that A' and Y are simply connected CW complexes. Prove that Q(X V Y) is weakly homotopy equivalent to the product ?.? ? ?1' ? UF. Deduce that if Ar (resp. Y) is r (resp. s) connected then nk(X V Y) = irk(X) © 7rfc(y) for k < r + s — 2. 2. Let X be a locally compact space and .4 c—> X a closed cofibration. Prove that the canonical map Yx -t YA is a fibration. 3. Prove that ? : ?1 -> ? ? ? = ?91, f ?- (/(?), /(I)) is a fibration. Prove that q : Bs -? ?, q(j) = 7@) is the pull-back fibration of ? : ?1 -> ? ? ? along the diagonal map ? -> ? ? ?. Deduce that nk(Bs') = nk(B) ? 7rfc+1 E), k > 1. 4. Let / : Sn -? Sn be a homeomorphism of degree —1 and ? the quotient space of the product Sn ? [0,1] by the relation (a-,0) ~ (f(x), 1). Prove that the composite Sn ? [0,1] -> [0,1] -> S1 induces a fibration ? : ? ^ S^. For ? > 2, compute 7rfe(?) and the action of US1 on the homotopy fibre Fp ~ Sn. 5. Prove that if X and y are simply connected CW complexes then the homotopy fibre of the pinch map X V Y -> X is the half-smash product ? ? ?.?/ * ? ??.
3 Graded (differential) algebra In this section we work over an arbitrary commutative ground ring k, except in (e). Thus module, linear, bilinear, ... will mean k-module, k-linear, k- bilinear, ... and the functors ????^(—, — ) and — ®jj; — will be denoted simply by Hom( —, —) and ~ <g> —. The object of this section is to establish the basic definitions and conventions of graded algebra upon which the rest of this book will rely. It is organized into the topics: (a) Graded modules and complexes. (b) Graded algebras. (c) Differential graded algebras. (d) Graded coalgebras. (e) When k is a field. (a) Graded modules and complexes. A graded module Vy is a family {Vi};6z of modules. By 'abuse of language' we say that an element ? 6 V{ is an element of V of degree i, and we write |u| — degu = i. We say that V is concentrated in degrees i 6 / if Vj· = 0, i ? I , and, by abuse of notation, we write V = {Vi}ieI. In particular we write Veven = {V2i}iez and Fodd = {V2i+1 }iez- Notice that we make frequent use of the Koszul sign convention that when two symbols of degrees k and i are permuted the result is multiplied by (—l)ke. The standard notions for non-graded modules carry over to the graded context: • A submodule V C V is a graded module {V/} with V{ C V;. • The quotient V/V of a module by a submodule is the family {Vi/V-}. • The direct sum ©aV(a) is the family {®aV(a)i}. In particular if V(i) is the graded module defined by 0 , otherwise then V = ? V(i). We shall therefore sometimes abuse notation and write The direct product f] V(a) is the family {]7 V(a)i}. The direct product of a a V and W is written V ? W.
Homotopy Theory 41 • A graded module, V, is free if each Vi is a free module. In this case the disjoint union of bases of the Fj is called a basis for V. • A linear map f : V —> W of degree i (between graded modules) is a family of linear maps fj : Vj —> Wj+i- It determines graded submodules ker/ C V and Im/ C W : (ker/),· = ker/,· and (Im/)j = Im/j_i. We denote by Hom(V, VI7) the graded module whose elements of degree i are the linear maps of degree i. If / : V —> V and g : W —> W are linear maps then Hom(/,<?) : Horn(F, IF) —>¦ Hom(V,W) is the linear map ? ? >· (_l)deg/)^ • The module Hom(F, Je) dual to F will be denoted by T/jt, and fi = Hom(/, Jfc) : W" —>¦ Fc will denote the dual of the linear map /. • Suppose fa : V(a) —> ? are linear maps of degree zero. The fibre product, ( ? )z(V(a)), is the submodule of f] V(Q) consisting of the elements a61 a6l {va} satisfying fa(va) — f?(v?), ?, ? El. The fibre product of V and W is written V ? ? W. • The tensor product V <g> W of graded modules is defined by (V ® H7)! = $y+fc=i Fy ® VFfc. If / : V —> V and 0 : W —>¦ VF' are linear maps of degrees ? and g, then f ® g :V ®W —> V <8> W is the linear map: (/ ® </)(« ® w) = (-l)de" de8"/(«) ® ?/(?) of degree ? + q. If / : F —> V we frequently (but not always) simplify / ® idw to/®W:V®W-+V® W. • A p-linear map is a map ? : V"(l) ? ··· ? V(p) —> W of the form a(vi ,...,Vp) = ?(vi ® · · · ® vp), where ? : V\ ® · · · ® Fp —^ IF is a linear map. A bilinear map V ? W —>¦ Jfc is called a pairing and is often written u, u; ?—>· {?,?). It determines (and is determined by) the linear map ? : V —? W^ given by {??)?? = {v,u>). In particular, if it is a field and F" and IF" are the dual graded vector spaces, then Fo ® W* is the subspace of (V ® IF1K given by (v'®w',v®w) = (-l)de&w'de&v(v',v)(w'1w). In addition, • The suspension of a graded module V is the graded module sV defined by (sV)i = Vi-i. If ? e ?,_! the corresponding element in (sVr), is denoted by sv. • We shall use the classical convention
42 3 Graded (differential) algebra to avoid negative degrees. Thus the degree of an element will depend on the context, depending on whether we are using upper or lower degrees. For example, if / : V —> W has degree -1 with respect to lower degrees, (/ : V;· —>¦ Wj-i), then it has degree +1 (/ : V* —>¦ W/i+1) with respect to upper degrees. In this section 'degree' will always mean 'lower degree' unless otherwise specified. • We use the notation V>k = {Vi}i>k and V>k = {Vi}i>k- The graded sub- modules V<k, V<k, V>k, F-*, V<k, V^k are defined analogously. More- over we write V+ = {?%>0 and V+ = {Vi}i>0 . A sequence M —> ? -^ Q of linear maps is exact if kerg = Im/. If also / is injective and g is surjective then 0 —? M —? ? —> Q —? 0 is a short exact sequence. A routine check establishes Lemma 3.1 (Five lemma) Suppose given a commutative row-exact diagram of linear maps of graded modules (i) If ctL is surjective and a m and &P are injective then ?/? is injective. (ii) If auf and ap are surjective and q.q is injective then a^ is surjective. (Hi) If OiL is surjective, ac is injective and am and cap are isomorphisms then ? ? is an isomorphism. D A differential in a graded module M = {A/;}igz is a linear map d : M —? ? of degree -1 such that d2 = 0, and the pair (M,d) is called a complex. The elements of ker d are cycles, the elements of Im d are boundaries and the quotient graded module H(M,d) = kerd/lmd is the homology of M. We often simplify the notation to H{M). A morphism of complexes ? : (M, d) —> (?, d) is a linear map ? : M —y N of degree zero satisfying ?? = ??. It induces ? (?) : H (M) —> H (?). If H (?) is an isomorphism, ? is called a quasi-isomorphism and we write ? : M -^4 ??. Two morphisms ? and ? are homotopic if there is a linear map h : M —>¦ ?? of degree 1 such that ?~? = hd + dh. We write ? ~ ? : {M,d) —>¦ (iV,d) and call h a chain homotopy. A chain equivalence ? : (M, d) —>¦ (?, d) is a morphism such that there is a second morphism ? : (N,d) —> (M,d) satisfying ?? ~ id^ and ?? ~ idm- Since ? {?) and ? (?) are then inverse isomorphisms, a chain equivalence is a miasi-isomorphism.
Homotopy Theory 43 If (M, d) and (TV, d) are complexes, then so are Hom(jl/, N) and ? ® ??: d(f) = df - (-l)des//d, / 6 Hom(il/, N), and d(m ®n) = dm®n + (- l)des mm®dn, me ?, ? 6 ?. Note that in particular the differential in Hom(il/, lz) is the negative of the dual of the differential in M. Suppose / is a cycle in Hom(M, iV); i.e. df = (-l)de&ffd. Consistent with the notation above we define a linear map H(f) : H(M) ->· H(N) by H(f)[z] — [f(z)], where [ ] denotes 'homology class'. Lemma 3.2 ^4 necessary and sufficient condition for H(f) to be an isomor- phism (of degree i) is that for each (??,?) ? ? ? ? satisfying dm = 0 and /(m) = dn there exist {m',n') 6 ?/xiV satisfying dm' = m and dn' — n — f(m'). proof: Suppose the condition holds. If dm = 0 and H(f)[m] = 0 then /(m) = dn; hence m = dm' and [m] = 0. If [n] 6 H(N) then dn = 0 = /@) and so there exist (m'.n') with dm' = m and dn' — ? - f(m'); i.e. [n] = H(f)[m']. Suppose H(f) is an isomorphism. Given (m,n) as in the lemma we have H(f)[™] = 0· Hence m = dm". Now ? - f(m") is cycle; hence [n - f(m")] = H(f)[z] = [/(«)]. Thus m = d(m"+z) and n-/(m" + 2) = tin'. Set m' = m"+z. D A chain complex is a complex (M,d) with A/ = {?/?}?>?; a cochain complex is a complex (M,d) with ?? = {?·/?}?>0. In the latter case d has (upper) degree 1, the elements of kerei are cocycles, the elements of Im ci are coboundaries and H(M, d) is the cohomology. The suspension of a complex (il/, d) is the complex ?(?/, d) — (sM, d) defined by sdx = —dsx. A short exact sequence of morphisms of complexes, 0 —> (il/, d) —i (N, d) —>¦ (Q, ei) —^ 0 induces a Zong exaci homology sequence, , H.{M) W> Hl(N) -^ Hi(Q) ? tfi-i(M) ->·.·, defined for all i, in which the connecting homomorphism d is defined in the usual way: if ? 6 Q represents [z] 6 Hi(Q) and if /3n = ? then 3[s] is represented by the unique cycle ? 6 M such that ax = dn. (b) Graded algebras. A graded algebra is a graded module R together with an associative linear map of degree zero, R®R —> R, x®y ?—> xy, that has an identity 1 6 Ro. Thus for ^1 x,y-, ? 6 R (xy)z = x{yz) and \x = ? = xl. We regard k as a graded algebra concentrated in degree 0. A morphism ? : R —> S of graded algebras is a linear map of degree zero such that <+>{xy) = ?(?)?(?) and <p(l) = 1.
44 3 Graded (differential) algebra An augmentation for a graded algebra R is a morphism ? : R —> L· oi graded algebras. A derivation of degree it is a linear map ? : R —> R of a degree & such that e{xy) = {9x)y + {-l)ka*zxx{9y). Let fibea graded algebra. A (left) R-module is a graded module M together with a linear map of degree zero R® M —> ?, ? ® m ?—> im, such that x(ym) = (xy)m and Im — m for all ?, y G i? and m 6 M. Right modules are defined analogously. An R-linear map f : M —> ? of degree k is a linear map of degree k such that /(im) = (-l)deg/ degIx/(m), xeRy me M. These form a graded submodule ?????(?, ?) of the graded module Hom(M,iV). The tensor product M' ®n M of a right A-module M' and a left A-module M is the quotient module M' ®n M - (M' ® M)/(m'x ®n-m' ® xn), where we have divided by the submodule spanned by elements of the form m'x ® ? - m' ® xn, m' 6 M', m ? ?, ? 6 R. We use m' ® m to denote the obvious element in M' <g> M\ its image in M' ®r M, is denoted by m' ®r m. Let R be a graded algebra. A left ideal I in R is a graded submodule such that xy € I for ? 6 R and y ? I. Right ideals, (two sided) ideals and subalgebras are defined analogously. Note that subalgebras must contain 1. The quotient R/I of R by an ideal / is a graded algebra. Example 1 Change of algebra. A morphism ? : R —> S of graded algebras makes 5 into a left (and right) A-module ? · s = tp(x)s or s · ? = s<p(x), ? € R, s 6 5. If M is an .R-module then S ®h M is an 5-module via s · (s' ®r m) = ss' ®r m. D Example 2 Free modules. Let R be a graded algebra. An A-module M is free if M S R ® V, with V a free graded module. A basis {va} for F is a basis for the free A-module M. If ? : R —> 5 is a morphism of graded algebras then 5 ®n M = S ®R (R <g> V) = S ® Vr is a free S-module with the same basis. If M is a free A-module with basis {va} and if iV is any i?-module, then the choice of elements na 6 Nk+^^ determines a unique A-linear map / : M —> ? of degree k with f(va) = na. ? Example 3 Tensor product of graded algebras.
Homotopy Theory 45 If R. S are graded algebras then R ® S denotes the graded algebra with mul- tiplication (x ® y)(x' <8> y') = (-l)de" d^x'xx' ? yy'. Example 4 Tensor algebra. For any free graded module V, the tensor algebra TV is defined by oc TV = 0 Tqv Tqv = y ®···®?. Multiplication is given by ? · ? = ? <g> b. Note that q is no? the degree: elements v\ ® · · · ® vq 6 TqV have degree = ? degu; and word length q. If {uj} is a basis of V we may write TV = ?{{?{}). Any linear map of degree zero from V to a graded algebra R extends to a unique morphism of graded algebras, TV —>¦ R. Any degree k linear map y —>¦ TV extends to a unique derivation of TV. ? Example 5 Commutative graded algebras. A graded algebra /I is commutative if x,y ? A . When i 6 1 this condition implies that x2 = 0 if .? has odd degree. If A is a commutative graded algebra, then a left yl-module, M, is automatically a right .4-module, via mx = (-l)desmdesxzm . If ? is a second ?-module then Horn a (M, iV) and ?/ «gi^ iV are .4-modules via (x/)(m) = ? ¦ /(m) = (-l)desx dee//(xm) and x(m ®? ?) = xm <8u ? = (-i)desx deg771m ®.4 in, ? e A, m 6 ?, ? 6 ? . If -4 —> ?, .4 —^ C are morphisms of commutative graded algebras then ? ® C is also commutative and the kernel of the surjection ? ® C —> ? <8u C is an ideal. Thus ? ®A C is also a commutative graded algebra. D Example 6 Free commutative graded algebras. Suppose Jk contains |. Let V be a free graded module. The elements v®w — (_l)deg« dezwWQV (v^w 6 y·} generate an ideal / C TV. The quotient graded algebra AV = TV/I is called the free commutative graded algebra on V. If {iij} is a basis of V we may write AV = ?({?,}). The algebra AV has the following properties:
46 3 Graded (differential) algebra (i) ? I·'' is graded commutative; in particular, the square of an element of odd degree in AV is zero. (ii) There is a unique isomorphism A(V®W) = AV<S>AW which is the identity in V and in W. (iii) If V is free on a single basis element {v} then a basis of AV is given by: 1 ? if deg ? is odd 1 ? ?2 ?3 if deg ? is even. (iv) A linear map of degree zero from y to a commutative graded algebra A extends to a unique morphism of graded algebras, Ay —> A. A linear map y —> Ay of degree k extends to a unique derivation in Ay. (v) Suppose ? : AV -4 A is a morphism of graded algebras and ? and ?' are derivations respectively in Ay and in A. If ??? = ?'??, ? 6 V, then ?? = ?'?. ?? (vi) Ay = ? ?9 y, where ?9 y is the linear span of the elements v\ ?···??M, 9=0 Vi & V] these elements have degree = ?* deg v, and word length q. (c) Differential graded algebras. A differential graded algebra (dga for short) is a graded algebra R together with a differential in R that is a derivation. In this situation kerd is a subalgebra and Imd is an ideal in kerd. The homology algebra H(R,d) of a differential graded algebra (R.,d) is the graded algebra H(R,d) = kerd/Imd. An augmentation is a morphism ? : (R, d) -4 it. A morphism of differential graded algebras / : (R, d) —> E, d) is a morphism of graded algebras satisfying fd = df. It induces a morphism H(f) : H(R) —> H(S) of graded algebras. If (R, d) and E, d) are dga's then so is (R, d) ? E, d) with the differential described in (a) and the multiplication given in Example 3 of (b). This is called the tensor product dga. A chain algebra is a dga {R,d) with R = {Rn}n>0. A cochain algebra is a dga (R, d) with R = {Rn}n>o- In the latter case H(R,d) is the cohomology algebra of (R,d). A (left) module over a differential graded algebra (R, d) is an A-module, ?/, together with a differential d in M satisfying d(x ¦ m) = dx ¦ m + (-l)degrx · dm, ? ? R,m ? M.
Homotopy Theory 47 Then H(M) is an H(R)-modu\e via [x] ¦ [m] = [x · m]. A morphism of (left) modules over a dga (R,d), is a morphism / : (M, d) —i (N, d) of graded i2-modules satisfying df = fd. The induced map H(f) : H(M) —> H(N) is then a morphism of H(R)-modu\es. If (M, d) and (JV, d) are left (R, demodules, then Hom/j(M, N) is a subcomplex of Hom(M, ?1'). In the same way, if (M'.d) is a right (R, d)-module then the differential d defined on M' <g> AI induces a differential in M' ®r M so that M' ®r M is naturally a complex. Example 1 Hom#(M, M). If (M,d) is any (R, d)-module, then Hom^(M,M) is a differential graded algebra with multiplication defined by the composition of maps. ? Example 2 Tensor products. If (A,d) -> E, d) and (A,d) -> (C,d) are morphisms of commutative dga's then E ®A C,d) is a commutative dga. D Example 3 Direct products. If (A, d) and (A',d) are dga's then the direct product (A, d) ? (.4', d) is the dga (A x ^4',d) given by (a,a') · (??,??) = (???,?'??) and d(a,a') = (da,da'). The direct product, Y\a(A(a),d), of a family of dga's is defined in the same way. ? Example 4 Fibre products. If / : (A,d) -> E,d) and /' : (A',d) -> E,d) are dga morphisms then (A xb A',d) is a sub dga of (.4, d) x (A',d). It is called the fibre product of (^4,d) and (^4',d). The fibre product of a family of dga's (^4(a),d) with respect to dga morphisms fa : (A(a),d) -> (B,d) is the sub dga ((Y\a)BA(a),d) of ne(^("),d)· D (d) Graded coalgebras. A graded coalgebra C is a graded module C together with two linear maps of degree zero: a comultiplication ? : C —> C®C and an augmentation ?: C —>¦ h suchthat (A®id)A = (id®A)A and (id<S>e)A = (e®id)A = idc- A morphism ? : C —^ C of graded coalgebras is a linear map of degree zero such that (<p® ?) ? = ?'? and ? = ?'?. A graded (left) comodule over a graded coalgebra C is a graded vector space M together with a linear map Am : C —> C ® M of degree zero such that (? ® id) Am = (id<8> Am) Am and (e ® idM)C = id^. A graded coalgebra is co-commutative if ?? = ? where r : C <g> C —>¦ C <g> C is the involution ? <g> 6 ?—> (-i)deea dee66 ® a.
48 3 Graded (differential) algebra A graded coalgebra is coaugmented by the choice of an element 1 € Co such that e(l) = 1 and ?A) = 1®1. Given such a choice the relations above imply that for ? € kere, ?? - (? ® 1 + 1 ® a) € ker ? ® ker ? . An element, a, in a coaugmented graded coalgebra is called primitive if ? 6 ker c and ?? = ?<8>1+1®?. Primitive elements form a graded submodule of ker ?, and a morphism of coaugmented graded coalgebras sends primitive elements to primitive elements. A coderivation of degree k in a graded coalgebra C is a linear map ? : C —> C of degree k such that ?0 = (? ® id + id ® 0)? and ?? - 0. A differential graded coalgebra (dgc for short) is a graded coalgebra C together with a differential that is a coderivation in C. If C is a graded coalgebra, then Hom(C, Ik) is a graded algebra whose multi- plication is defined by (f-g)(c) = (f®g)(Ac), f,geHom(C,k), c6C and with identity the map ? : C —> h. If (C, d) is a differential graded coalgebra, then C* = Hom(C, Jfc) is a differential graded algebra. Remark Henceforth we may occasionally suppress the differential from the no- tation, writing M, R,... for a differential graded module (Ai, d), a dga (R, d),... (e) When Iz is a field. Recall that complexes (M,d) and (N,d) determine the complexes ?®?? and Hom(A/,JV) with d(m <g> n) = dm <g> ? + (-l)desmm ® dn and df = do f - (_l)deg/yoc^ Natural linear maps H(M)®H(N)-+H(M®N) and H (Rom(M,N)) —»¦ Horn (?(??),?(?)) are given by [z] ® [w] ·—> [z ® w] and [/] ·—> H(f). (Recall that H(f)[z] = [Hz)])· Proposition 3.3. If ]k is a ?e/d these natural maps are isomorphisms: H(M) ® if(iV) = if(M ® N) and ? (Eom(M, TV)) = Horn (H(M), H(N)). proof: This is a straightforward exercise using the fact that any complex M can be written M = Im d © H @ C with d : C -=> Im d and d - 0 in H. ? In particular, suppose (C,d) is a dgc over a field Jk. Using the first iso- morphism of Proposition 3.3 we may write ?(?) : H(C) -> H(C) ® H(C). Together with ?(?) this makes H(C) into a graded coalgebra. The dual algebra,
Homotopy Theory 49 Hom(J?(C), Jk) is just the homology algebra of the dga, Hom(C, k), as follows again from Proposition 3.3. If k is a field we say a graded vector space M = {Mi} has finite type if each Mi is finite dimensional. If also ?/ is concentrated in finitely many degrees then M is finite dimensional and dim M = ? dim M,. i When M has finite type its Hubert series is the formal series However if M = {M'} we write M*(z) = ?](dimMl)zl. When the context is i clear we simply write M(z). If M is finite dimensional then its Euler-Poincare characteristic is the integer Xm defined by ?? = ?(-1I dim M, = dim A'Ieven - dimModd = A/*(-l) . If M is equipped with a differential then it is an easy exercise to verify that Xm = Xh(m) · C-4) A graded algebra R (or a dga) has finite type if each Ri is finite dimensional; i.e. if it has finite type as a graded vector space. Similarly a graded R-module of finite type is an A-module M such that each Mi is finite dimensional. By contrast R is finitely generated if it is generated by a finite set of elements (i.e. if there is a surjection ?(??5... ,Vk) —> R). Similarly M is a finitely generated e R module if every m ? ? can be expressed as m = ^Z Xirrii, Xi 6 R, where mi,... me is a fixed set of elements of M. Exercises 1. A complex M is acyclic if Hi(M) = 0, i = 0.1,... . Prove that if M is an acyclic A-free chain complex then icLm ~ 0. Let M and ? be graded chain complexes. Under what conditions is the complex Horn (M, N) acyclic? 2. Let (?,???) and (N,dN) be chain complexes and / 6 Hom°(A/,N). Prove that if M is an i2-free module then / is a chain equivalence if and only if the chain complex (C,d) defined by Ck = Mk-\ ® Nk, d(m,n) = (-dMm,f(m) +dNn) is acyclic. 3. Consider the free commutative differential graded algebra (A(x, y, z),d) with deg ? = deg y = deg ? = 1 and dx = yz ,dy = zx ,dz = xy . Prove that the map ? '- (??,?) -> (A(x,y,z).d) defined by ?(?) — xyz is a quasi-isomorphism.
50 3 Graded (differential) algebra 4. Let TV be the tensor algebra on the graded vector space V. Define ? : TV —»¦ TV ® TV , ? : TV —»¦ A, by ?(?? ® vo <8>... ® un) = 53"=0(?? ® ?2 ® ... ® ^) ® (ui+i <8> vi+2 ® ... ® vn) e(l) = 1 and ?(?? <S> vo <S> ¦¦¦ <S> vn) = 0 ?? ? > I . Prove that (T(V), ?, c) is a non cocommutative graded coalgebra. 5. Let (M,<1m) and (N,d^) be chain complexes. Prove that if M is an Et- iree chain complex then the complex Hom (??. ??) is isomorphic to the complex Hom {M, R) ® JV. 6. Let (M,cIm) be a chain complex. Prove that if M is of finite type then there exists a Hubert series Q(z) such that M(z) = H(M)(z) + A + z)Q(z).
4 Singular chains, homology and Eilenberg- MacLane spaces As usual, we work over an arbitrary commutative ground ring h. This section is a quick review of singular homolog}·. Distinct from the usual approach, however, is our use of normalized singular chains, so as to arrange that the chains on a point vanish in positive degrees. We also take great care with the comparison between C,(X xl') and C*(X) eg C*(Y). using the clas- sical Alexander-Whitney and Eilenberg-Zilber equivalences because these have important associativity and compatibility properties on which we will subse- quently need to rely. Beyond the standard material of elementary singular homology we: • show that a weak homotopy equivalence induces an isomorphism in homol- ogy- • construct the cellular chain complex of a CW complex and show that it is chain equivalent to the singular chain complex. • prove the Hurewicz theorem identifying the first non-vanishing homotopy and homology groups of a space (except when ?\ ? 0, in which case Finally, we construct the Eilenberg-MacLane spaces ?(?,?) for a group ?? (abelian if ? > 2) and show that if ?' is an (n — l)-connected CW complex then [?,?(?,?)] = Horn (??(.?), ?), the isomorphism being given by [/] ?—> ???(/). In fact this theorem generalizes to the famous Theorem ([E]) For any CW complex X a bisection [?, ?(?, ?)] -=4 ??(?: ?) is given by [f] ?—> Hn(f)t,, ? denoting the fundamental class of Hn (?(?.?):?). However we shall not need this result, and so it is not included in the text. This section is organized into the following topics: (a) Basic definitions, (normalized) singular chains. (b) Topological products, tensor products and the dgc. Ct(X;L·). (c) Pairs, excision, homotopy and the Hurewicz homomorphism. (d) Weak homotopy equivalences. (e) Cellular homology and the Hurewicz theorem. (f) Eilenberg-MacLane spaces
52 4 Singular chains, homology and Eilenberg-MacLane spaces (a) Basic definitions, (normalized) singular chains. Recall that a singular n-simplex in a space X is a continuous map ? : ?? —> ? , (? ? where ? > 0 azid ?? = \ ? t{e{ \ 0 < U < 1, ? it = 1 ? is the convex hull of ? ? J the standard basis e0, - - -, en of En+1. If X C Em is a convex subset then any sequence xc,, · · -1 ?? € X determines the linear simplex {x0 ... xn) ¦ An —> ?, ????? h-> ?????. Two important examples of linear simplices are the i-th face inclusion of ??, A,· = (eo---ei---en) : ??-1 —> ??, ("means omit) defined for ? > 1 and 0 < i < n; and the j-th degeneracy of ??, defined for ? > 0 and 0 < j < ?. The image of Ai is called the i-th face of ??. Denote by Sn(X) the set of singular n-simplices on a space X (n > 0), and by Sn(<p) : Sn(X) r Sn(Y), Sn(<p) : ? h-+ ? ? ? , the set map induced from a continuous map ? : ? —» ?. The set maps di : Sn+i(X) -> Sn(X), ? ?-)· ? ? A; and Sj : Sn(X) -> Sn+i(X), ? ?-> ? ? qj are called the face and degeneracy maps; they commute with the set maps 5»(y). A straightforward calculation shows that SiSj = Sj+iSi, , i <j; ( Sj^di ,i<j D-1) djSj = < id , i = j,j + 1 I Sjdi-i , i > 7 + 1. The free it-module with basis Sn(X) will be denoted CSn(X; Ik), and the sin- gular chain complex ?? X is the chain complex CS,(X;Jfc) = {CSn(X;Jk)}n>0, with differential d — ?(—1-)???. Its homology, denoted H,(X;Jk), is the singular i homology of X. An ? + 1 simplex of the form Sir (r 6 Sn(X), 0 < i < n) is called degenerate and the degenerate simplices span a submodule DSn+i(X) of C5n+i(„Y;Jfc). It follows easily (cf. [118]) from D.1) that DS,(X) is a subchain complex of CS*{X;k) and that H(DS,(X)) = 0. We will use the quotient chain complex: C»(X; Bt) = CS*(X', Jk)/ DS,(X) it is called the normalized singular chain complex of X. The following properties follow immediately from the definitions and the re- marks above:
Homotopy Theory 53 • Cn(X]fc) is a free module with the non-degenerate n-simplices as basis. • The surjection CS*(X;]k) —>¦ C.(X;Jc) is a quasi-isomorphism. Thus we may (and do) use this to identify H.(X;li) = H (C.(X;lc)). D.2) • C,(pt) — h, concentrated in degree zero. (The last property follows because ?? —> pt is degenerate for ? > 1.) A continuous map / : X —> Y induces the morphism C,(f) : C,(X:,L·) —> C.(Y; k) defined by C.(/)a = foa; we write #.(/) for H(&.(/)). In particular, for the constant map ?' —> pt we may identify C.(const) = ? : C„(X;k) —> k. This is called the augmentation of C»(X;k). Remark CS.(-) vs C.(-). Most presentations of singular homology use CS.(-) since it is simpler to describe, and gives the same answer. For us, however, the property C*({pt}; Ik) = L· turns out to be really useful. The constructions that follow are all made first at the level of 5«(-Y), then bi- linear extension to CS»(X;Jk) and then (by factoring) in C*(X;Jc). We shall describe only the first step, leaving the rest to the reader. (b) Topological products, tensor products and the dgc, C*(X;k). We recall now how topological products are modelled by tensor products of chain complexes. As in §3, — ® — denotes — 0* —. First note that CS,(X; ]k) ® CS*(Y;]k) is a free graded module with basis the tensor products ? ® r of singular simplices in X and Y. Moreover C,(X:k) ? C,{Y;h) is the quotient chain complex obtained by dividing by those ? ® r with at least one of ?, ? degenerate. Thus, as remarked above, constructions in CS*(—) pass to C(—) by factoring. There are two important chain complex maps that model topological products by tensor products. The first is the Alexander-Whitney map AW : C._(X ? Y; k) —»¦ C.(X; k) ® C.{Y\ Ik). defined as follows: If (?, r) : ?? -> ?' ? ? is any singular simplex on ? ? ? then ? AW (?, ?) = Y^ao(e0,...,ek)®To (ek,-..,en). ?·=0 The second is the Eilenberg-Zilber map EZ : C.{X;h) ®C.{Y;h) —»¦ C,(X ? Y;k) ,
54 4 Singular chains, homology and Eilenberg-MacLane spaces whose construction is based on a combinatorial description of the product ?? ? ?9 of two standard simplices with vertices vo,...,vp and wo,...,wq. When ? = q = 1 this reduces to the decomposition of / ? / into two triangles: ?' = ((vo,wo),(vo,wi),(vi,wi)) In general ?? ? ?9 decomposes as the union of all the linear (p + ^)-simplices ? of the form ? = ((va@),W?@) ),- ¦ ¦ ???(?)???(?))? ¦ ¦(va(p+q),W?(p+q))), D-3) where for each i either a(i + 1) = a(i) + 1 i + 1) = ?(i) + 1 or else Putn(w)= ? [?(i + l)-?(i)][a(j + l)-a(j)}. 0<i<j<p+q The idea of EZ is that if ? 6 SP(X) and r 6 5,(?) are respectively a p- simplex and a ^-simplex then the basis element ? ® r in Cp(Jf) ? C9CK) should correspond to the map ? ? r : ?? ? ?9 -> ? ? }'. Thus while the fact that ?? ? ?9 decomposes as the union of the ? motivates us, we do not actually need it. We simply write down the chain where the sum is over all ? of the form D.3). Notice that if ? or ? is degenerate then so is CS,(a x r)(cp>9), and so it represents zero in C*(X ? ?). Thus we may define EZ by Straightforward (but often tedious) calculations establish the properties below of AW and EZ. Let px : ? ? ? -> Ar and py : ?" ? F -> 7 be the projections. We identify X = ? ? {pi} and p-Y with id ? const. : Ar x F —> ? ? {??}. On the algebraic side we identify C.(A; A) ® C.({pi}; Jb) = C.(X; Jb) ® Jfc = C.(A; A).
Homotopy Theory 55 • [118] AW and ? ? are morphisms of chain complexes, natural in X and in Y. D.4) • In particular. (id ® ?) ? AW = Ct(px) and (? ® id) ? AW = C.(py); while D.5) C*(px) oEZ = id® ? andC,(pY) ? EZ = ? When X = {pt}, AW and EZ both reduce to the identity map ofC*(Y; Jk). A similar assertion holds when Y — {pt}. D.6) [118] Associativity For any three spaces ?,?,?, (AW ® id) ? AW = (id ® AW) ? AW, and D.7) EZ ? (EZ ®id) - EZ ? (id® EZ). as maps between C.(X xYxZ; k) and C.(X; Jk) ® Ct(Y; k) ® C,(Z; k). [118] Interchange of factors For topological spaces X,Y and graded mod- ules M,N define ? : ? ? ? —> ? ? X by 9(x,y) = (y, x) and 0aig : M ®N —> N®M by 0aig(m ® n) = (-l)desm des"n ® m. Then C,@)oEZ = ?Zo0alg : C.(X;k)®C,(Y;k) —> C,(Y xX;k). D.8) [49] Compatibility The following diagram commutes C.(XxX';k)®C.(YxY';k) > C.(X;k)®C.(Y;k)®C.(X';k)®C.(Y';k) EZ EZ®EZ C.{XxX'xYxY'\k) > C.(XxY;k)®C.(X'xY';k) A\VoC.(idxexid) ,. , D.yj where ? = zc?<8>oaig ® id. Finally we have Proposition 4.10 AW and EZ are inverse chain equivalences. In fact, AW ? EZ = id and EZ ? AW is naturally homotopic to the identity. -proof; The first assertion is a simple computation, depending on the fact that we have divided by the degenerate simplices. For the second, we have to construct h : Cn(X ? Y; k) —> Cn+1(X ? Y; k), natural in A' and }', such that EZoAW- id = dh + hd. We may set h = 0 in Co (? ? ?). Suppose for some ? > 1 that h is constructed for i < n. Let ?1?? : ?? —>?"??" be the diagonal, regarded as a singular ?-simplex. Then ? = (EZ ? AW)(Atop) - ?1?? - hd(iltop) is a cycle in Cn(An ? ??; k). Since ?? ? ?? is contractible we may find cn+1 G Cn+\ (?? ?
56 4 Singular chains, homology and Eilenberg-MacLane spaces ??; A-) so that dcn+\ = ?. Now for any ?-simplex (?, ?) : ?? —> ? ? ? define /?(?,?) = C.(axr)(cn+1). ? The Alexander-Whitney map defines a natural differential graded coalgebra structure in C,(X;k) via the topological diagonal, ???? : X —> ? ? ?, ? 4 (x,x). The comultiplication is given by A = AW ? C.(Atop) : C.(X;k) —»· C.(X;k)®C.(X;k). Thus for each ? G Sn(X), ? ?(?) = AW (?, ?) = ]??? < e0, ·. · ,efc > ®?? < e*.·,... ,en > . The augmentation is defined by ? = C,{const.) : C,(X;k) —> k = C,({pf}; *). The associativity of AW and its compatibility with ? (cf. (b) above) imply that (?,(?;*),?,?) is a dgc. Moreover any ? € X may be regarded as a zero-simplex in Co(X;k) and. clearly ?? = ?®?. Thus 1 i-j· ? defines a co-augmentation in the dgc, C,(X\ Jk). (c) Pairs, excision, homotopy and the Hurewicz homomorphism. For A C ?' we put C.(X,A;k) = C4Xik)/C,(A;ls)i it is it-free on the non-degenerate simplices of X whose image is not in .4. Its homology, H, (X, A; Ik) is the ordinary relative singular homolog)'. For ? C A C X we have the short exact sequence 0-+C.(A,B;le) —> C,(X,B;k) —>C,(X,A;k) —> 0 induced by the obvious inclusions; it leads to a long exact homology sequence >Hi(A,B;k) -^Hi(X,B;k) -^Hi(X,A;k) ? Hi-i(A,B;k) —>... D.11) d is called the connecting homomorphism. Since H,(Y;k) = ?,{\\?;??) this reduces to a more familiar long exact sequence when ? = ?. If W C .4 C X has the property that the closure of W is contained in the interior of .4, then the excision property [142] states that inclusion induces an isomorphism H,{X - W,A - W; Jk) ? H.(X,A;k). The morphisms EZ and AW of (b) above factor to give (relative) Eilenberg- Zilber and Alexander-Whitney morphisms EZ C. (X, A; k) ® C. (Y; k) } » C. (X x ?, ? ? ? ; k) AW
Homotopy Theory 57 satisfying AWoEZ = id and EZoAW ~ id. Thus these are quasi-isomorphisms. Suppose now that ? : A' x / —> Y is a homotopy from ?0 to ?\. Define h : Ci{X;k) —»¦ Ci+1(Y;k), i > 0, by ?(?) = (-?)'?(?) ? ??(? ® id/), where the identity map of /, id/, is regarded as a singular 1-simplex. Then C.(<p1)-C.(<p0)=dh + hd; i.e., ? is a homotopy from C,(<po) to ?,(?\). In particular ?*(??) = ?*(?0). As an example, let <9?? be the boundary of ??, and regard the identity map ?? -^ ?? as a singular simplex. It represents a cycle, zn e G?(??,<9?"; Jfc), whose homology class will be denoted by [??]. Lemma 4.12 For ? > 0, ?* (??, <9??; Ik) is a free module concentrated in de- gree ? with single basis element [??] Hf(An,dAn;k) = 1c ¦ [An]. proof: We may suppose ? > 0. Regard ?"" = Im(e0,..., e„_i) as one of the faces of ?? and let L be the union of the other faces. Then L is contractible to en. Hence H,(L;&) ? H.(An;Jk), and the long exact sequence for ? C L C ?? implies that #.(??, L; h) = 0. Thus the long exact sequence for L C <9?? C ?? provides an isomorphism 3:?4??,3??;&) A H,-.i(dAnt L; k). On the other hand, shrinking the last coordinate shows that the inclusion (??~1,???~1) —> C?? - {en},L - {en}) is a homotopy equivalence. This yields H.{dAn,L\k) \?^.?? H,(dA" - {en},L -{en};*) - ~ /f. (??~?, 8?11"'; Jfc) Combined with the isomorphism above this gives an isomorphism H,(An,dAn;]k) 9?H.-1(An-1,dAn-1\k), which carries [??] to (—l)"^"]. The lemma follows by induction on n. ? It follows from Lemma 4.12 and the long exact sequence for ? c <9?? C ?? that for ? > 1, H,(dAn;]k) is a free .fc-module on two generators dAn] D.13) where [eo] is the homology class of the 0-simplex eo and [<9??] is the homology class of the cycle zn_! = Yl(-\)i{eQ...ei...en). Since (??,???) is homeo- morphic to (Dn,Sn~1) we conclude | JL, ? .V-, U inn en—1. n-\ _ ) ¦"*·' l ~ n tii{L) ,O ,Jh,) — < _ , ' [0, otherwise and D.14) 0, otherwise.
58 4 Singular chains, homology and Eilenberg-MacLane spaces We use a different calculation to fix a generator in Hm(Sm;Z) = Z, m > 1. Recall that Im denotes the m-cube. In Example ?, §1 we constructed homeo- morphisms Im/dlm A Sm and dlm+1 A 5m , and so D.14) implies that Hm(Im,drn;Z) ?? Z. Choose generators [/m] G Hm(Im,dIm\Z) as follows. For m = 1 let [/] G Hi(I,dI;Z) be the class rep- resented by the identity map ? of / (i ; t >->¦ t), which is indeed a relative cycle. For m > 1 let [/m] be the class represented by ? ? {? <g> · · · <g> ?). Then [Im] is a generator of Hm(Im.,dIm;Z) and 9[/m+1] is a generator of Hm(dlm+1 ;Z). A straightforward calculation shows that the homeomorphisms above send [/m] and d[Pn+l} to the same generator, [Sm] G Hm(Sm:Z). Definition [Sm] is called the fundamental class of Sm class of Sm. Next, recall that Hm(<p) ¦ Hm(Sm;]k) —> Hm{X;h) depends only on the homotopy class of tp : Sm —> X. Thus set maps hurx : Tim(X, x0) —> Hm(X; Ik), m > 1, are defined by /iurx[y] = ifm((p)[5'71]. A straightforward computation shows that these are group homomorphisms; and it is immediate from the definition that they are natural; if / : (X,Xo) —> (Y,yo) is continuous then = hury- °?*(/)· Definition The Hurewicz homomorphism for X is the homomorphism (d) Weak homotopy equivalences. We shall rely heavily on Theorem 4.15 If ? : A' —>¦ Y is a weak homotopy equivalence then C»(y) : C»(X;Jk) —> C*(Y; Jk) is a quasi-isomorphism. proof: Recall the face maps Aj = (e0 ... et... en) : ??-1 —> ?". We first observe that for each ? > 0 and each ? : ?11 —> Y we can associate ?' : ?? —> ? and a homotopy ?? : ?? ? / —> Y from ? to ? ? ?' such that ?' ? Xi = (? ? Aj)' and ?? ? (?* ? id) - ????,., 0 < i < ? D.16) Indeed we proceed by induction on n. Conditions D.16) define ?' on <9?? and ?? on 3?? ? /. Now the extension to ?' and ?? is just the Whitehead lifting lemma l.o.
Homotopy Theory 59 Define a linear map / : CSm{Y\k) —> CSm(X;Jk) by ? ^ ?'. The conditions above imply that / is a chain map and that CSm(ip) ? / — id = hd + dh, where h(a) = CS.($G)oEZ(a®idI). A second application of the Whitehead lemma gives, for each r : ?" —> X, a homotopy ?? : ?? ? / —> X from (? ? ?)' to ? such that ?? ? (\t x id) = ????,, 0 < ? < ?. This then implies that / ? CSm{ip) — id — dk + kd. Hence CS*(ip) is a chain equivalence and ?.(?) is a quasi-isomorphism. ? (e) Cellular homology and the Hurewicz theorem. Let (A', .4) be a relative CW complex with ?-skeleton Xn, so that Xn = A'n_i U/n ( 1I-DS )- Let oa be the origin (centre) of D%. Put U = Xn - Ui0») and ? = U (?? - {oa}). Since Dn- {0} = Sn~l ? @,1] ~ Sn~l x {1}, it follows ? that the inclusions A'n_i —> U and U^a —> ? are homotopy equivalences. ? Thus in the diagram C.(X„,U) C.(.Y„ -.?,-,^-^-,) c where C»(—) stands for C«(—;Jk), the horizontal arrows on the left are quasi- isomorphisms, while the horizontal arrows on the right are quasi-isomorphisms by excision. Hence the characteristic map induces an isomorphism in homology: By D.14) we may identify H,(Xn,Xn-i;lk) as a free module concentrated in degree ? with basis {cQ} indexed by the ?-cells of X. This implies via an easy •induction that H,(Xk,Xr;lk) is concentrated in degrees i 6 (r,k\. Since any singular simplex of Ar has compact image it lies in some Xk, and so any element of C,(X;Js) is in some C,{Xk\ le)- It follows that Hi(XtXr-tl:)=0t i<r. D.17) Denote the free module. Hm(Xn,Xn-\\h) by Cn. Associated with the inclu- sions Arn_i c Xn C Arn+1 is a long exact homology sequence, (cf. D.11)) with connecting homomorphism d : Cn+i —> Cn. An easy computation shows that ooo = 0. Thus (C = {Cn}n>o,<9) is a chain complex. It is called the cellular chain complex for the relative CW complex, (Ar, .4). Definition A cellular chain model for (X, A) is a morphism m: (C,d) -*Cm(X,A;]k)
60 4 Singular chains, homology and Eilenberg-MacLane spaces between the cellular and normalized singular chain complexes, restricting to mor- phisms m(n) : (C<n,d) —> C.(Xn,A;k) and such that the induced morphisms rh{n) : (Cn,0) = (C<n/C<n-i,d) —> C.(Xn,*n-i;*) induce the identity map Cn = H»{Xn. Xn-i\ k). When ,4 = 0, m : (C,d) —> C*(X]k) is called a cellular model for the CW complex X. Theorem 4.18 (Cellular chain models) Every relative CW complex (X, A) has a cellular chain model m : (C.,9) —> C*(X,A;k), and m and each m(n) are always quasi-isomorphisms. proof: We construct the morphisms m(ri) inductively and observe in passing that they are quasi-isomorphisms. Suppose by induction that m(n) is constructed; we extend it to m(n + 1) as follows. Represent a basis element ca G Cn+\ by a cycle za G Cn+i(Xn+i,Xni&), and lift za to a chain wa € Cn+\(Xn+i', k)¦ Then dwa is an ?-cycle in C,(Xn;k). Since m(n) is a quasi-isomorphism there is a cycle ua G Cn and a chain aa G Cn+i(Xn',k) such that m(n)(va) = dwa +daa- Since H(rh(n)) is the identity, it is immediate that va = dca. Extend m(n) to a mor- phism m(n + 1) by setting m(n + l)(ca) = wa + aa. Since wa + aa also projects to za, rh(n + l)(ca) = za and H(fh(n + l))ca = [za] = ca, as desired. Finally, since H(m(n)) and ?(??(? + 1)) are isomorphisms so is H(m(n + 1)) by the five lemma C.1). The sequence m(ri) so constructed defines a cellular model m : (C„, <9) —> C, (X, A; Jk). Moreover for any cellular model we have, in the same way, that each m(n) is a quasi-isomorphism. Formula D.17) identifies Hi(m(n)) with Hi(m) for i < n; hence m is a quasi-isomorphism. ? Remark Suppose ? : (X, A) -> (Y, B) is a cellular map of relative CW complexes, so that it restricts to maps ip(n,k) : (Xn,Xk) -> (Yn,Yk)· Put fn = ??(?(?,? ~ 1)) : Hn(Xn,Xn-i; *) -> ??{??,??-?-?)- It is immediate from the definition of d as a connecting homomorphism that / = {fn}: (C?,d) -» (CY,d) is a morphism from the cellular chain complex of (X, A) to the cellular chain complex of (Y, B). Now suppose that mx : (C?,d) -^> C,(X,A;k) and mY : (C?,d) -=* C*(Y,B;k) are cellular chain models. We construct linear maps hn : C* -> Cn+i{Yn;k) so that mY f -C, (?)??? = dh + hd : (Cx, d) -> C, (Y, B; Je); i.e., mYf is chain homotopic to Ct(<p)mx.
Homotopy Theory 61 In fact, set /i_i = 0 and assume hi constructed for i < ? so that the equation above holds in Cxn. Let ca be a basis element of Cx+l. Then d(myf - C,(<p)mx - hd)ca = (mYf - C. (?)??? - dh - hd)dca = 0. Thus za = (mYf — Ct{tp)mx - hd)ca is a cycle in Cn+1 (Yn+X\]k). Now hdca G h(Cx) C Cn+1 (Yn; k). It follows that za projects to the cycle (my f-C,(tp)mx)c0 in Cn+\ (??+?, ^n,' -fc)· It is immediate from the definitions that this relative cycle is a boundary, so that za = dxa+ya, xa G Cn+2(Kn+1;$;), ya G Cn+i(Yn; A·). Now ?/? is an (n + 1)- cycle in Yn. Since i/n-n^n;!:) = 0 by the Cellular chain models theorem, ya = dwa, some wa G Cn+2(Yn+\]E:). Define hn+i by setting hn+l(Ca) = Wa- Theorem 4.19 (Hurewicz) Let (?,??) be a pointed topological space for which ???(?,??) =0, i <r. (i) If r = 0 (i.e., X is path connected), then the homomorphism hurx : ?? (X, x0) —> i/j (Ar; Z) is surjective and its kernel is the subgroup of ?? generated by the commu- tators ??? ?~?. (ii) Ifr>\, then Hi(X;Z) = 0, Ki <r and hurx : ??+? (Ar, x0) —> i/r+1 (?"; ?) is an isomorphism. proof: The proof of Theorem 1.4 provides a weak homotopy equivalence ?: (Y,yo) —» (Ar,x0) in which ? is a CW complex, YT = y0 and ?'r+1 = Va5^+1. Thus Theorem 4.15 asserts that Ht (?) is an isomorphism. By the naturality of the Hurewicz ho- momorphism it is sufficient to prove the theorem for Y. Moreover, again by naturality, it is sufficient to prove the theorem for the finite subcomplexes of Y; i.e. we may assume 1' itself is finite. We first consider the case r > 1 and observe that the Cellular chain models theorem above shows that Hi(Y:Z) = 0, 1 < i < r. Moreover, since 5r+1 = *Uer+1; l\Sa+1 is a CW complex with Yr+1 as its (r + 1) skeleton and the
62 4 Singular chains, homology and ????????^-????^??? spaces next smallest cells of dimension 2r + 2. Thus (since r > 1) there are no r + 2- cells, and so the Cellular approximation theorem 1.1 implies that 7rr+1(l'r+1) -=4 Kr+i(Y\Sa+l). By Hopf's theorem 1.3, this identifies ???+?(??+?) as a free abelian ? group with basis {[??]} where ?? : S?+1 —> Yr+i is the inclusion. Hence l/,_ l'y \ » TJ /¦y · ? \ is an isomorphism, hur' denoting the Hurewicz homomorphism for Yr+\. Let ? : ir+i —> Y be the inclusion. Since H(Y,Yr+i;Z) vanishes in degrees k < r + 1 (cf D.17)), Hr+i(X) is surjective. Now the commutative diagram 7??+?(??+?)-^??+1(??+1;?) shows that Ziury is surjective. Write Yr+i = Yr+\ U (U^e^+2) with attaching maps //? : S?+l —> }'?+?· Let (C, <9) be the cellular chain complex for Y. If C? ? Cr+o is the basis element corresponding to e^+2, then a straightforward computation, using the definition of d as a connecting homomorphism, shows that dc? = Hr+\(f?)[S?+1]. Thus it follows from the Cellular chain models theorem that the kernel of Hr+i(\) is spanned by the classes Hr+i(f?)[Sp+i]. Now suppose g : (ST+1,*) —> (Y>yo) represents an element in kerhury. By the Cellular approximation theorem, g is based homotopic to h : Sr+1 —> Yr+i; i.e., [g] = ??+1(?)[/?] and 0 = Hr+1(X) hur'[h}. Thus hur'[h] = ? hHr+i(f?)[Sr0+1] = Y^k? hur'[f?]. ? ? Since hur' is an isomorphism, But f? is the attaching map for a cell in Yr+o, and so 7rr+i(A)[/^] = 0. It follows that [g] = ??+?(?)[/?] = 0 and so hur is injective. In the case r = 0, Yi = Va5^ and the van Kampen theorem [68] implies that ??A?) is generated by the circles S?. These same circles form a basis of the free abelian group ??(??;?), and it follows that ker hur' is generated by the commutators. Now simple modification of the proof for r > 1 completes the argument. D (f) Eilenberg-MacLane spaces. An Eilenberg-MacLane space of type (?,?), ? > 1, consists of
Homotopy Theory 63 • A path connected, based space (X,Xo) such that ??(?.??) = 0, i ? ?, together with • A specified isomorphism of ??(?,?0) with a given group ?. If X is a CW complex it is called cellular Eilenberg-MacLane space. Eilenberg- MacLane spaces are often denoted by ?'(?,?). Here we shall need two simple properties. Proposition 4.20 Suppose given an (n—1)-connected based CW complex (X, x0), an Eilenberg-MacLane space (?(??, ?), *) and a homomorphism ? : ??(?', Xo) —> ?. Then there is a unique homotopy class of maps such that nn(g) = ?. proof: The proof of the Cellular models theorem 1.4 shows that (X,xq) has the weak homotopy of a CW complex with a single 0-cell, no cells in dimension i, 1 < i < n, and all cells attached by based maps (Sk,*) —> (Xk,xo)· Since weak homotopy equivalences between CW complexes are homotopy equivalences (Corollary 1.7) we may assume (?,??) itself satisfies this condition. Thus Xn = VaS2, and Xn+1 = Va52 U, (Up Dn0+l). Now 5J represents a class ?? e ??(?, xq); choose gn : (VaS2, xo) -> (?(?, ?), *) so that gn restricted to S? represents a(-ya). If ? : Xn —> X is the inclusion then Trn(gn) = ????(?), and it follows that gn ° f '¦ (S?,*) —> (?(?,?),*) is null homotopic. This permits us to extend gn to a map gn+\ : Xn+i —> ?(?. ?). Extend gn+\ to the rest of X inductively over the skeleta, using the fact that ??(?(?,?)) = 0, i > n. Given a second such map h we use the fact that ??(/?) = Kn(g) to conclude that hn ~ gn : (Xn,x0) —> (?(?,?),*). Extend the homotopy ? inductively over the skeleta: if ? is defined in ?'?.· ? / and Dk+l is a (k + 1) disk attached by Sk —->· Xk then ? yields a map Sk ? / —-> ?(?,?). Using h in Dk+l ? {0} and g in ?>fc+1 ? {1} we extend to a map d(Dk+1 ? /) -» ?(?,?). Since k > n> ^fc+i (?(?, ?)) = 0 and this map extends to all of Dk+1 ? /, thereby extending ? to Xk+i x I. ? Proposition 4-21 Suppose ? is an abelian group. Then there exists ? ?(?,?) and any two have the same weak homotopy type. proof: Suppose ? > 2. Write ? as the quotient of a free abelian group on generators ga divided by relations r0. The proof of the Hurewicz theorem D.19) identifies ??(??52) = ?? Z9a- Let f3 : E?,*) —> VaS2 represent r?. Then the proof of the Hurewicz theorem also identifies Kn(VaSSLI{fg} UpD0+1) = ?· Put Yn+1 = Vu52 U{fg} \}? ?>3+1· Now create Yn+2 by adding (n + 2) - cells to ^n+i to kill 7Tn+1(yn+1) and continue this process inductively, creating Yk+\
64 4 Singular chains, homology and Eilenberg-MacLane spaces by adding (k +1)- cells to Yjt to kill ?^??· The Cellular approximation theorem 1.1 will show that Y is a ?(?,?). For ? — 1 we simply note that, since path spaces are contractible, the long exact homotopy sequence B.2) for the path space fibration implies that ?,?(?, 2) is a ?(??, 1). Let (?',??) be a cellular model for an Eilenberg-MacLane space (Y, y0) and suppose ?(?,?) is any Eilenberg-MacLane space of the same type. Proposition 4.20 gives a map g : (?.,??) —> ? (?,?) such that irn(g) is the identity map of ?·. Thus ? is a weak homotopy equivalence. D Remark When ? = 1, ? (?, l)'s exist for any group ?, and are constructed in the same way: ?? is the quotient of a free group on generators ga by relations rp, and adding corresponding 2-cells to VS^ gives a space with ?? = ??. For details see [68].) We will not carry this out since such K(ir, l)'s do not arise in this text. Exercises 1. Compute the homology with ? coefficients of the space obtained from the sphere 5" by attaching an (n + l)-cell along a continuous map of degree ? and another (n + l)-cell along a continuous map of degree q with ? and q relatively prime. 2. Suppose that Ik is a field. Compute the algebra H*(SP x5«x Sr;Ik) for 1 <p<q<r. 3. Let X be a finite complex and Ik a field. Prove that dimii»(Ar, Ik) < co. Using §3-exercise6, prove that ?(?) = ^ (-l)dim ? . Compute ?(?') when X = Sn ,RPn ,CPn,T-. 4. Let X(n) be obtained by attaching to a space ?" cells of dimension n + 2 and higher to kill off the homotopy groups of A" in dimensions above n. Prove that the homotopy fibre of the canonical map ?^?+1> -> A"(n) is an Eilenberg-MacLane space ? (??+1 (?),? + 1). 5. Prove that the cap product, CP(X; Ik)®Cg(X; Ik) -> Cq-p(X; Ik) , /<gc ^ /? c = (l®f)AW(c) induces a natural of C*{X; Jfc)-module structure on C*(X;Ik).
5 The cochain algebra C*(X]k) As usual, we work over an arbitrary commutative ground ring, k. Recall from §3 that given two complexes (M,d) and (N,d) we can form the complex (Hom(M,N),d). with d(f) = df - (-l)dezffd. In particular, the nor- malized singular cochain complex of a topological space X is the cochain complex C*(X;k) = Eom(C*(X;k),k). Thus Cn(X;k) = Horn (Cn(X;k),k) and d{f) = -(-\)^&ffd. In particular, an element / 6 Cn(X; k) may be thought of as a set theoretic function / : Sn{X) —> k vanishing on the degenerate simplices, and n+l (d(f))(a) = Y{-l)d*zf+i+lf{ao<e(u...,ei,...,en+l >). i=0 The augmentation ? : C0(X;k) —> k may be regarded as an element 1 6 ..C°(X;k). Recall from §4(b) that the Alexander-Whitney comultiplication in C,{X) k) is defined by ? = AWoC*{Atop) ¦ C.(X; k) —> C,{X; k)®C*(X; k). Dually, as described in §3(d), C*(X; k) is a cochain algebra whose natural asso- ciative multiplication, the cup product, is defined by = (-l)k("-Vf(ao(e0,...,ek))g(ao(ek,...,en)), f ? Ck(X;k), g ? Cn~k(X;k), ? € Sn(X). Definition The cochain algebra C*(X;k) is called the normalized singular cochain algebra of X. The cohomology algebra H (C* (X; $:)) is denoted H* (X; k) and called the singular cohomology of X. If ? : X —> Y is a continuous map then we put ?*(?) = Horn (C, (?); k) : C*(Y;k) —> C*(X;k). Just as ?,(?) is a morphism of differential graded coalgebras so C*(<p) is a cochain algebra morphism, and ?*(?) = ? (?*(?)) is a morphism of graded algebras. If .4 c X then C*(X; k) —> C*(A;k) is surjective because (as graded mod- ules) C.(A;Jk) is free on a subset of a basis of C,(X;k). The kernel of this restriction, C*(X, A;k), is an ideal in C*(X;k) and the short exact sequence 0 —>· C*(X,A;k) —> C*(X;k) —> C'(A;le) -4 0 gives rise to a long exact cohomology sequence. Note that C*(X,A;k) con- sists of the functions S*(X) —> k that vanish on degenerate simplices and on simplices in A. Thus we may identify the cochain complexes C*(X,A;k) and
66 5 The cochain algebra C*(X;k) Horn (C»(X, A; k),k). We call C*(X,A;k) the complex of normalized relative singular cochains; H*(X,A;k) = H (C*(X, A; He)) is the relative singular coho- mology. Left and right multiplication by C*{X;k) make C*(X,A;k) into a left and right C*(A',-l;)-module. Thus H*(X,A;k) is a left and right ii*(Ar,-ifc)-module. Let / ? Cn(X,A;]k) and ? € Cn(X,A;k) be respectively a cocycle and a cycle. Then f(z) depends only on the cohomology class [/] and the homologv class [z\. Define a :H*{X,A:k) —> Horn (Hm(X, A; k),k) E.1) by (a[/]) [z] = f(z). This defines a pairing between H* and H» that we denote by ([/], [z]}, in line with the convention in §3(a). Next, let ? : (?, ?) ? Y —> (X, A) and q : ? ? ? —> Y be the projections. Since ?* ((?,?) ? Y;k) is a right H*(X ? Y; J:)-module, a linear map ? : ?*(?, .4; Jfc) ® H'(Y; k) —> ?* ((?., ?) ? Y; k) E.2) is defined by ? ([/] ® [g]) = ii*(p)[/] · H*{q)\g]. Proposition 5.3 (i) If k is a field then a is always an isomorphism and so { , } is non-degenerate. In particular, H*(X,A;k) has finite type if and only if H*(X,A;k) does. (ii) If k is a field and at least one of H„(X,A;k), Ht(Y;k) has finite type then ? is an isomorphism. proof: (i) This is just the assertion of Proposition 3.3 that "H" commutes with "Horn". (ii) As follows from §4(b), the Alexander-Whitney map is a quasi- isomorphism C*((X,A) xY;k) A C*(X,A;k) ®C*(Y;k). Since homology commutes with tensor products (Proposition 3.3) we obtain an isomorphism if, ((?, ?) ? Y; k) A H, (X, A; k) ® ?, (?; k). Because we have supposed one of H,(X, A; k), H*(Y; it) to have finite type, we may identify H*(X,A;k) ® H'(Y;k) as the dual of H,(X,A;k) ®H*(Y;k), and ? as the isomorphism dual to the isomorphism above. D Finally, if A C ? C X then the short exact sequence 0 —>· C*{X,B;k) —> C*(X,A;k) —> C*{B,A;k) —>¦ 0 leads to a long exact cohomology sequence dual to D.11).
Homotopy Theory 67 Exercises 1. a) Prove that if A C A' and ? C A' then the cup product C(X;Ik) ? C*(X;Ik) -> C"(X;Ik) restricts to a cup product C*(X, A; Ik)®C*(X,B;Ik) -> C*(X,AUB;Ik) and thus induces a cup product H*(X, A, Ik) ®H*(X,B; Ik) -> H*(X,AUB;Ik). b) Prove that if .4 and ? are contractible open sets such that A' = Al)B then for any ? > 0 and q > 0, U : #P(A; a) ® if«(A'; ft) -> Hp+q{X, Ik) is zero c) Prove that a suspension (reduced or not) has a trivial cohomology algebra. d) Prove that CP2 and S2 V S4 do not have the same homotopy type. 2. Suppose that A is a field Prove that the morphism ?, defined in 5.2, is an isomorphism of algebras when ^1 = 0 Compute the algebra H*(X;Ik) when. X = S1 ? S1 ? ... ? S1 (?-times), A' =7vZ, A' = (S3 V S7) ? E5 V S9), X = F ? ?. 3. Prove that, for a 6 tt6(CP2 V S2), the spaces ??? = CF2 V S2 Ua O7 have the same cohomology algebra. 4. Prove that ii dim Hi (X; Ik) = oo, then H1(X:Ik) is uncountable.
6 (R, (i)-modules and semifree resolutions As usual, we work over an arbitrary commutative ground ring, Ic. Thus graded module means graded Ik-module, linear means It-linear, and Hom(—, —) and — ® — stand for ?????.(—, —) and — <8>?· — · Our focus in this section is on modules (M, d) over a dga (R, d) as defined in §3. While the emphasis is on left (.ft, demodules the results apply verbatim to right modules. In the classical context of modules M over a ring R (no differentials) it is standard to construct a quasi-isomorphism (P*,d) -^ (M,0) from a chain complex of free .ft-modules to M; these are the free resolutions of M. Free resolutions were generalized by Avramov and Halperin[16] to modules over a dga (R,d). Their analogue of the complexes F» are the semifree (R,d)- modules defined below, which have these two important properties: • Any (R,d)-module admits a quasi-isomorphism from an (R,d)-semifree module. • Any morphism from an (R, d)-semifree module lifts (up to homotopy) through a quasi-isomorphism. Thus (cf. introduction to §1) semifree modules over (.ft, d) are the exact analogues of CW complexes. This section is organized into the following topics: (a) Semifree models. (b) Quasi-isomorphism theorems. (a) Semifree models. We begin with some basic definitions. Recall the tensor product ? ®r M and the module of A-linear maps ?????(?, ?') defined in §3(b) for any right A-module ? and left A-modules M and M', If (N,d),(M,d) and (M',d) are (R, ^-modules then (cf. §3(c)) r M,d) is a quotient complex of (N, d) ® (??, d), and • (HorriR(M,M'),d) is a subcomplex of Hom(M, M')- A morphism ? : (M,d) —i- {M',d) is an A-linear map of degree zero, sat- isfying ?? = ??. Two morphisms, ?, ?, are homotopic if ? — ? = ?? + Bd for some A-linear map ?;? is a homotopy and we write ? ~? ·?. Thus mor- phisms are precisely the cycles of degree zero in UomR(M,M') and two mor- phisms are homotopic if and only if they represent the same homology class in #(H(MM')) In particular an equivalence of (.ft, demodules is a morphism ? : (M.d) (M',d) such that for some morphism ? : (M',d) —> (M,d) we have ?? ~?
Homotopy Theory 69 and ?? ~A idj[j'. Thus an equivalence is a quasi-isomorphism. The morphism ? is called an inverse equivalence. In general, a cycle of degree A; in HomA( A/, M') is an R - linear map / satisfying dof= (-l)de^/od. It induces the if(#)-iinear map H(f) : H(M) —> H(M') defined by H(f)[z] = [f(z)]. Clearly H(f) depends only on the homology class, [/], of / and [/] !->¦ H(f) defines a canonical linear map H(HomR(M, M')) —> HomH{R)(H(M),H(M')). Finally if W is a graded module then unless otherwise specified R <8> W will denote the A-module defined by ?·F<8>??>) = ab^w. When w ?-» l®w is injective we identify W with the image 1 <g> W C R <8> W. Definition A left (R, d)-module (M, d) is semifree if it is the union of an in- creasing sequence M@) C M(l) C ¦ · · C M(k) C F.1) of sub (i?,d)-modules such that ?/@) and each M{k)/M(k — 1) are i?-free on a basis of cycles (cf. §3). Such an increasing sequence is called a semifree filtration of (M..d). A semifree resolution of an (R, d)-module (Q, d) is an (R. d)-semifree module (?/, d) together with a quasi-isomorphism m : (M, d) A (Q, d) of (R, d)-modules. Semifree resolutions play the same role for modules over dga's that ordinary free resolutions do in the ungraded case. Moreover they generalize the classical case: if ungraded objects are regarded as graded objects concentrated in degree zero then free resolutions are examples of semifree resolutions. Semifree resolutions also play a quite analogous role to that of CW complexes within the category of topological spaces, as will be illustrated shortly by re- sults that mimic several of the theorems in §1. First, however, we make some preliminary observations. Remark Suppose {M(k)} is a semifree filtration of (M,d). Then Jl/@) and each M(k)/M(k- 1) have the form (R,d) <g> (Z{k).O) where Z{k) is a free k- module. Thus the. surjections M(k) —> R <g> Z(k) split: M{k) = M{k-l)©{R®Z{k)), and d:Z(k) —> M(k-1). In particular, if we forget the differentials., M = R ® ( ® Z(k) ) is a free R- , , \A.-=0 / module. Lemma 6.2 // (R,d) —> (S.d) is a morphism of dga's, and if (M,d) is a semifree (R.d)-module then (S®RM,d) is (S,d)-semifree.
70 6 (R; d)-modules and semifree resolutions proof: Let {M(k)} be a semifree filtration for (M.d). From the formula in Remark 6.1 we deduce S ®r M(k) = S®r M(k - 1) © (S ® Z(k))., d : Z(k) —> 5 ®R M(k - 1). This exhibits {5 ®r M(k)} as a semifree filtration for E ®r M.d). D Lemma 6.3 Suppose an (R.d)-module (M,d) is the union of an increasing se- quence M@) C ?/A) C · · · of submodules such that M@) and each M(k)/M(k- 1) is (R,d)-semifree. Then (M,d) itself is semifree. proof: Put M(—1) = 0. In the same way as in the Remark we may write oo M(k) = M(k- ~ with Z(k, C) a free graded Jc-module and d : Z(k, ?) —> M(k - 1) © I R ® Thus M is A-free on the union {za} of the bases of the l:-modules Z(k, ?), with Z(k, ?) a free graded ic-module and d:Z(k,?) — Define an increasing family W@) C W(l) C · · ¦ of free ic-modules inductively as follows: W@) is spanned by the za for which dza — 0 and W(m) is spanned by the zo for which dza e R ¦ W(m - 1). Then {R ¦ W(m)} will be a semifree filtration of (M,d), provided that each za is in some W(m). Write (i,j) < (k,?) if i < k or if i = k and j < I. If za G Z(k,t) then dza = ????? with ?? e R and ?? e Z(i,j), some (i,j) < (A,?). We may assume by induction that each such ?? is in some W(m.?). Put m = max^m^. Then dza e H7(m) and so 2a e I-F(m + 1). D Let ? : (?, d) —> (Q,d) be a morphism of (i?,d)-modules, and for any third (i?,d)-module, (M,d), denote by ) : HomA(A/,P) —> HomA(M,Q) the morphism of complexes defined by ? ?-> ? ? ?. Proposition 6.4 Suppose (M,d) is semifree and ? is a quasi-isomorphism
Homotopy Theory 71 (i) HomA(M, 77) is a quasi-isomorphism. (ii) Given a diagram of morphisms of (R,d)-modules, (Rd) F.5) (M,d) -+¦ (Q,d) , there is a unique homotopy class of morphisms ? : (M,d) —> (P,d) such that ? ? ? ~A ?. (ni) A quasi-isomorphism between semifree (R,d)-modules is an equivalence. proof: (i) As remarked in Lemma 3.2 it is sufficient to show that given / e HomA(M,?) and g € UomR(M,Q) satisfying d(f) = 0 and 77 ? f — d(g) we can find /' e HomR(M,P) and g' e HomR(M,Q) satisfying d(f') = f and d(g') = 77 ? /' - g. Let r = deg /. Choose a semifree filtration {M(k)} of M, put M(—1) = 0 and, as in Remark 6.1 write M{k) = M(k - 1) ? (R ® Z(k)) where Z(k) is Jfc-free and d : Z(k) —> M(k - 1). We construct /' and g' inductively by extending from M{k — 1) to M(k). For this let {za} be a basis of Z(k), put pa = f(za) - {-l)rf'(dza) and put qa = g(za) - (-l)rg'(dza). By hypothesis the equations d(f') - f and d(g') =770 / - g are satisfied in M(k — 1). It follows after a short calculation that dpa = 0 and ?(??) = dqa. Since 77 is a quasi-isomorphism there are (by Lemma 3.2) elements p'a e ? and q'a € Q such that dp'a = pa and dq'a = ?{?'?) ~ Qa· Now extend /' and g' to A-linear maps in M(k) by putting f'(za) — p'a and 9'{za) = q'a. (ii) As remarked near the start of this section, ff0(Homfl(-. -)) is the set of homotopy classes of morphisms. Thus by (i), composition with 77 induces a bijection from homotopy classes of morphisms (?/, d) —> (P,d) to homotopy classes of morphisms (M,d) —> (Q,d). (Hi) Since (Q,d) is semifree we can find a morphism ? : (Q,d) —> (P,d) such that 77? ~^ idc. In particular, 7]?? ~^ ????? = ????. Since (?, d) is semifree it follows from the uniqueness in (ii) that ?? ~^ idp. G Remark Notice that Proposition 6.4 (ii) is the analogue of the Whitehead lifting lemma 1.5. As in §1 we use this to prove an analogue of the Cellular models theorem 1.4. Proposition 6.6 (V Every (R,d)-module (Q,d) has a semifree resolution m : (M,d) -^> (Q,d).
72 6 (R, cu)-modules and semifree resolutions (ii) If m' : (M',d) -=·> (Q,d) is a second semifree resolution then there is an equivalence of (R, d)-modules a : (M',d) —>· (M,d) such that m ? a ~h m!. proof: (i) Let V@) be a free graded $>module whose basis is a system of generators of the A-module of cocycles of Q. Set M@) = (R, d) ® (V@), 0). The inclusion V@) —> Q defines a morphism g@) : (M@),d) —> (Q,d). Clearly H(g@)) is surjective. We now construct an increasing sequence of morphisms g(k) : {M{k),d) —> (Q,d). If g(k - 1) is defined, let V{k) be the free Jfe- module whose basis va in degree i is in one to one correspondence with cocycles wQ representing a system of generators of kerii,-_i(p(fc — 1)). Set M(k) = M(k — 1) ? (R ® V(k)), and put duQ = wa. Denote g(k - 1) simply by g. Since g{wa) is a coboundary, 5(u;a) = d(xa) we put g(k)(va) = xa. Finally, set ?/ = \jM(k). k (ii) Proposition 6.4 (ii) asserts the existence of a morphism ? : (M',d) —> (M,d) such that ?? ? a ~^ in'. Thus H(m) ? ? (a) = H(m') and so ? (a) is the isomorphism H(m') ? H(m)~l. In other words, ? is a quasi-isomorphism. Hence, by Proposition 6.4 (iii) it is an equivalence. D (b) Quasi-isomorphism theorems. A basic property satisfied by semifree modules is 'preservation of quasi-iso morphy' under the operations Horn and <8>. First we set some notation: suppose f : M —> M' and g : ?' —> ? are ii-linear maps between left A-modules and define HomA(/,p) :HomA(M',P') —>HomA(M,P) by ? ^ (-l)degideg/5 ? ? ?/. Then, if h : Q —> Q' is an A-linear map between right il-modules, define h®Rf: Q®RM -4 Q'®RM' by / If /, g and h commute with the differentials then Hom^(/, g) and h (g>? / are morphisms of complexes. Proposition 6.7 Suppose (M,d) and (M',d) are (R,d)-semifree. (i) If f and g are quasi-isomorphisms then so is Momn{f,g). (ii) If f and h are quasi-isomorphisms then so is h (8>h /. proof: (i) Since (M,d) and (M',d) are semifree, Proposition 6.4 (iii) asserts that / is an equivalence. Thus there is an inverse equivalence /' : (M',d) —> (M, d) and ii-linear maps ? : M —-> M and ?' : M' —» M' such that f'of- idM = ?? + ?? and / ? /' - idM, = dB' + ffd. F.8) Apply Homfl(—,idp>) to these formulae to conclude that Romn(f,idp') is an equivalence. Since (M, d) is semifree. Hom^(idA/, g) is just the quasi-isomorphism
Homotopy Theory 73 Eomfi(M,g) of Proposition 6.4 (i). Thus Horrifl(/,g) = Eomn(idM,g) ° ?????(/, idp<) is a quasi-isomorphism. (ii) Applying idc ®r — to the formulae F.8) above we deduce that idq, ®r f is an equivalence. Since h®R f = (idci ®r /) ? (h ®r id m) it is sufficient to show that h ®r idM is a quasi-isomorphism. Let M(k) be a semifree filtration; it is enough to show that each h ®r idM(k) is a quasi-isomorphism, and we do this by induction on k. Indeed, since M(k)/M(k - 1) = (R, d) ® (Z(k), 0) with Z(k) a it-free graded module it is obvious that /?<8>?^?/(?:)/?/(?:-?) is a quasi-isomorphism. Moreover, as ^-modules M{k) = M(k - 1) ? {R ® Z{k)) — cf. the Remark in §6(a). Thus the commutative diagram of complexes, 0 ^ Q ®h M(k - 1) 0 >~Q'®(i7^T) F.9) is row exact. Now the Five lemma 3.1. applied to the resulting map of long exact homology sequences, gives the inductive step. ? Remark 1 In Proposition 6.7 (ii) the identical argument shows that h ®h / is a quasi-isomorphism if (Q,d) and (Q',d) are semifree and (M.d) and (M',d) are unrestricted. As it turns out, a somewhat more general result will be required. For this, fix a dga morphism ?: {R,d) —> (S,d). Thus any left (or right) (S, d)-module, (N, d) becomes an (R, d)-module via x-n = ?(?)?, ? € R, ? € ?.- Now suppose given the following left and right modules over (R, d) and (S,d) and i?-linear maps f,g and h of complexes: • / : (M,d) —> (M',d); M is left over R and M' is left over 5. • g : (?1,d) —» (?,d); ?' is left over S and ? is left over R. • h : (Q,d) —> (Q',d); Q is right over i? and Q' is right over 5. Define Hom,p(/,0) : Homs(M', P') -» HomA(M, ?) by ? .-> (-l)desi d^fgo^o f and h®pf: Q®RM —> Q' ® s M' by q®Rm^ h{q) ®s f(m). Theorem 6.10 Suppose (M, d) is (R, d) -semifree and (AI1, d) is E, d) - semifree.
74 6 (R, d)—modules and semifree resolutions (i) //</>,/ and g are quasi-isomorphisms, so is Homv,(/, g). (ii) If ?, h and f are quasi-isomorphisms, so is h®vf. proof: As observed in Lemma 6.2, (S ®R M,d) is E, <i)-semifree. Moreover, since (M,d) is (R. d)-semifree <p®Rid:(R®RM,d) —> (S®RM,d) is a quasi-isomorphism, by Proposition 6.7 (ii). On the other hand, / extends to the morphism of E, d)-modules /' :(S®RM,d) —>(M',d), f'(y®Rm) = yf(m), y e S,m e M. Clearly f = f ° (<p <8>h id) and so /' is also a quasi-isomorphism. (i) Note that HomsE ®R M, -) = HomR(M, —) and that Hom,^/,^) is the composite of the morphisms uomR{idM,g) : HomA(Ai,i") —>HomA(M,P) and uoms(f',idp.) : Homs(M',P') —> Homs(S ®R M,P'). In view of our remarks above, Proposition 6.7 (i) implies that both these mor- phisms are quasi-isomorphisms. (ii) Observe that Q'®s{S®RM) = Q'®RM and that h®vf is the composite of idQ' ®s f ·· Q' ®s (S ®R M) —>· Q' ®s M' with h ®h idM : Q ®R M —> Q' ®R M. Now apply Proposition 6.7 (ii). D Remark 2 The argument in 6.10 (ii) shows that h ®? f is also a quasi- isomorphism if (Q;d) and (Q',d) are semifree and {M,d) and (M',d) are unre- stricted — cf. Remark 1, above. Finally, we establish a partial converse to Theorem 6.10 for chain algebras. Recall that a chain algebra is a dga, (R,d) and that R = R>0- The surjection -Ro —> Hq{R), x ·->¦ [x\ may be regarded as a morphism eR:{R,d) ->(Ho(R),0) natural with respect to chain algebra morphisms. Suppose given • a chain algebra quasi-isomorphism ? : {R,d) —> (R',d).
omotopy Theory 75 • a right (i?,<2)-semifree module (N,d) and a right (R\ d)-semifree module (N',d), both concentrated in degrees > 0. • a morphism / : (N,d) —> {N',d) as above (i.e., f(nx) — /(?)?(?), ? e ?, ? ? R). ien we may construct the morphism of chain complexes / ®„ ?0{?) : {N,d) ®R Ho(R) -> (N',d) ®A. Ho(R'). F.11) ieorem 6.12 With the hypotheses above: f is a quasi-isomorphism -?=?- / ®? ?0(?) is a quasi-isomorphism. oof: We have only to prove <== since the reverse implication is just Re- irk 2. Define a decreasing sequence of differential ideals RDI1 D J1 D /2 D · · · D /n D Jn D · · · setting {0, A: < ? - 1 [ 0, k <n-l (Im d)*, jfc = ? - 1 and (Jn)k = < (kerd)fc, fc = ? At, A: > ? { Rkl k > n. lese constructions are clearly natural. By inspection, H{In/Jn) = 0. Since (N,d) is semifree, ? ®r {0} -=> iV ®? /Jn. Thus In/J ^ n/Jn ^> ice both sides have zero homology. Also by inspection, n/jn+l\ _\ Hn(R), k = ? \ fjn/j k \ 0, otherwise. lus we may write ? ®r Jn/In+l = (N ®rH0(R)) ®ho(R) Hn{R)- Now ®R H0{R) —y ?' ®R' Hq{R') is a quasi-isomorphism of semifree H0(R)- )dules. Tensoring this with the isomorphism Hn(R) -^» Hn(R') produces a asi-isomorphism (Theorem 6.10) and so N®r Jn/In+l ^N'®R(jy/(I')n+1. >w an obvious induction via the five lemma shows that TV ®r R/r A N> g,A, R'/(I')n, ? > 1.
76 6 {R, d)—modules and semifree resolutions But since ? - N>0, the modules ? and ? ®r R/In+3 coincide in degrees < ? + 1. Thus their homology coincides in degrees < n. It follows that H{f) is an isomorphism. n Exercises 1. Let V and V" be two copies of the graded vector space V and (R <g> V,d) a semifree module over the cochain algebra {R,d). Prove that there is a (R,d)- semifree module of the form (R <g> (V ? V" ? sV),D) such that (i) the natural inclusions ? , i" : {R®V, d) <^->(i?<g> (V ? V"? sV), D) defined by i'x = ?', i"x = x" are quasi-isomorphisms, (ii) Dsv = 1 <8> v' + 1 <8> v" + S(dv), ? e V where S : R®V -) R® sV denotes the unique extension of ? ·-» sv to an A-linear map of degree +1. Deduce that two morphisms of i?-modules f,g : (R<&V,d) —>¦ (M,d) are homotopic if and only if there exits a morphism of A-modules F : (R® (V ? V" ® sV),D) -» (M, d) such that F o i' = f and F o i" = g. 2. Prove that the two morphisms of (Q[x],0)-modules f,g : Q[x] ® (QaeQo) -» Q[s] ® Qz defined by: deg ? = 4, deg a = 2, deg 6 = 5 = deg z, da = 0 = <2z, db = xa, f(a) = ^(a) = 0, f(b) = ? and g(b) = 0, induce the same map in homology but are not homotopic. 3. Let Zfc be a field and (T(V),d) be a chain algebra over Ik. Construct a contractible semifree (T(F), demodule of the form (T(V) <g> {Ik ? sV),D) with ' Dsv - ? ® 1 e TV ® sV. 4. Assume that in diagram F.5) the morphim ? is onto. Prove that there is a unique morphism ? such that ?? = ?. 5. Let f,g : ? -> M and ? : M —>¦ iV be morphisms of A-modules. Prove that if ? is semifree and ? is a quasi-isomorphism, then / ~# g if and only if ?°? -R ?°9- 6. Let M and IV be chain complexes over R = {R, 0) and assume that M and H{M) are A-free modules. Prove that H{M ® N) = ii(M) 7. Let X be a finite CW complex. Prove that the cellular chain complex of X, C?, is a semifree Z-module and that for any space Y the quasi-isomorphism C? 4 Ct{X) extends to a quasi-isomorphism C? ® Ct{Y) 4 Ct{X) ® C,{Y). 8. Let ? : {R, d) -^ Ik an augmented cochain algebra and {P, d) an {R, d)-semifree module. Consider the surjective cochain map {P,d) -> {Ik ®r P,d). If R° — Ik and d = 0 we say that ? is a minimal semifree module. Prove that if R1 =0 then any {R,d)-module M, with the property that for some r0, Mr = 0, for all r < r0, admits a minimal semifree resolution ? = {R <g> V, d) 4 M and that a quasi-isomorphism between two minimal semifree resolutions is an isomorphism.
7 Semifree cochain models of a fibration As usual, we work over an arbitrary commutative ground ring, Jk. We simplify notation in this section by writing C*(—) instead of C*(—;Jk) and H*(—) instead of H*(—;Jk). Thus (cf. §5) a continuous map f ¦ X —> Y induces a morphism of cochain algebras C* (/) : C* (X) <— C* (?), which makes C*{X) into a left (resp. right) C*(l')-module via a- b = C*(f)aUb (resp. b ¦ a = b U C*(/)a). We shall also simplify notation by denoting the tensor product over C*(Y) of C*(y )-linear maps by ? <8> ? instead of ? ®?·(?) ?- Let F be the fibre of a fibration ? : X -> Y, as defined in §2. The purpose of this section is to show, under mild hypotheses, how to compute the cohomology of F from the left C*(Y)-module, C*(X). The answer is surprisingly simple: if ??? : (??,?) -?» C*(X) is a C*(y)-semifree resolution, then as graded vector spaces. (This will be proved if Jk is a field, y is simply connected and one of Ht(F;Jk) and Ht{Y;Jk) has finite type.) Thus for the rest of this section we fix the following: • A fibration ? : X —>¦ Y with simply connected base y and fibre inclusion j : F —>¦ X at a basepoint y0 e Y. G.1) • A morphism of left C*(Y)-modules, ??? : (??,?) —> C*(X), in which (MY,d) s (C*(Y) ® W,d) is C*(y)-semifree. To state the main theorem we also require the following conventions and con- structions dealing with pullbacks. Observe that ?? and C*(Y) <8>c-(y) ?? are identified by the inverse isomorphisms ? t-> 1 <8> ? and ? - ? <—? ? <8> ?. Now a continuous map ? -, ? —> y determines three constructions associated with G.1): • the pullback fibration ? ? : ? ? = ? ?? ? —> ?. • the C* (?)-semifree module C*(Z) <S>c(Y) ??· • a morphism of left C* (Z) -modules, mz : C*(Z) ®c-(Y) My —> C*(Xz) . (This is really an abuse of notation, since these constructions depend on ? rather than on Z.) The first two constructions are self evident. To construct mz consider the diagrams C*(XZ) < ^ C*(X) I ? and c-(irz) c-(-r) G.2) C(Z) +—- C'{Y)
78 7 Semifree cochain models of a fibration in which the left diagram is just the pullback diagram of §2(a). Then mz : a (g, v ?—). ?·?*(?)???(?),? G C*(Z),v G ??, is the unique C*(Z)-module morphism making the following diagram commute: C*(XZ) < ^ C*(X) I my G.3) C*(Y) ®c(Y) ??, (Here as above we have identified ?? = C*(Y) <S>c~(y) ??). When ? = {yo} the map ? in G.2) is just the inclusion j : F —> X of the fibre. Moreover C*(y0) = Je and C*(ip) is the augmentation ? : C*(Y) —> k corresponding to yo- Thus C*(y0) <8>c-(Y) ?? = k <8>c-(Y) ?? = k <8>c-(Y) (C'(Y) <g> W,d) = (W,d)» and diagram G.3) reduces to G.4) Our main result in this section is Theorem 7.5 Suppose ?? : X —> Y and ??? : ?? —> C*(X) are as described in G.1). Assume k is a field and that at least one of the graded k-vector spaces H*{Y), H*{F) has finite type. Then ??? is a quasi-isomorphism => ??,? is a quasi-isomorphism . Remarks 1 At the end of this section we shall extend Theorem 7.5 to more general cochain algebra functors (Theorem 7.10). 2 The reverse implication in Theorem 7.5 is also true (cf. Exer- cise 1) 1 3 Theorem 7.5 is essentially due to J.C. Moore, and also follows easily from a theorem of E. Brown [31]. W. Dwyer [47] has considerably weakened the hypothesis of simple connectivity on Y. We shall prove Theorem 7.5 using an inductive procedure involving relative cochains. First we set some further notation. If i ¦. A —> Y is the inclusion of a subspace we denote C*(Y, A; Je) simply by C*(Y, A); it is the kernel of the surjection C*(i) : C*(Y) —> C*(A). In this case the pullback XA is the subspace ? (A) of X, and m^ and ??? fit into the commutative row-exact diagram 0 >¦ C*(X,XA) 2 >· C'{X) ^ C*(XA) ^ 0 ? My_+C*(Y) <g> MY _^C'{A) (g> ?? C'(Y) 0®id C'(Y) C'(i)®id C"(Y)
Homotopy Theory 79 where ? and ? are the obvious inclusions. Indeed the right square commutes by definition, and hence ??? ? (? ® id) factors uniquely through ? to define ???,?- The exactness of the lower row follows because ?? is free as a module over the graded algebra C*(Y). Next, suppose A contains j/o and that ? : (?, ?,?0) —> (Y, A,yo) is a con- tinuous map. In this case we have a pullback map ? : (??,??) —> (?,? ?)- As above we may construct first ?? ? ¦ ? ? — C*{Z) ®c(Y) ?? —> C(Xz) and then ??,?,?- It is immediate from the definitions that the following diagram commutes: C* (Xz, XB ) ?^~ C' (X, XA ) r,A G.6) C(Z,B) ®c-(Y) MY < —C{Y,A) ®C-(Y) My, where we have identified C*{Z,B)®C-(Z){C"{Z) ®C-(Y) ??) - C*{Z,B)®C-(Y) ??. Moreover the fibre of ?? ? at z0 is just {z0} ? F and ? restricts to the iden- tification {z0} x F = F. Thus it is immediate from the definitions that rnz = rnY: (W,d) —>¦ Cm(F). . G.7) Finally we introduce the notation of k-regular : a morphism ? of cochain complexes is k-regular if ??{?) is an isomorphism for i < k and injective for Suppose now that we are in the situation of Theorem 7.5 and, further that: • for some A c V, (Y, A) is a relative CW complex. • y0 ? A and all the attaching maps have the form (S",*) —> (Yn-i>yo)', i.e. they are based maps. • all the cells of (Y, A) have dimension at least two. • at least one of the graded vector spaces H*(Y,A), H*(F) has finite type. • W = {W*}i>o and in ?? = (C'(Y) ® W,d), d:Wn —)¦ Cm(Y) ® W<n, ? > 0. Note that the last condition exhibits ?? as C" (Y)-semifree. Proposition 7.8 Under the hypotheses above, my is k-regular =? ???^ is (k + 2)-regular. proof: We begin by establishing the proposition in four special cases. Case 1: (Y A) = (Vuer D?, VuerS^), some ? > 2.
80 7 Semifree cochain models of a fibration Let C*(Y,A) ®m C*{X) denote the tensor product of these two cochain com- plexes (over k) and let C*{Y, ?)®& ?? denote the tensor product of the cochain complexes C*(Y,A) and ??. We have used the notation - <8>? - for emphasis. Define a commutative square of morphisms of cochain complexes. id&rriY C*(Y,A)®kMY as follows: ?(? ®b) = C*(ir)a U b, and ? is the canonical surjection — ®k > - ®c-(Y) -¦ By Proposition 3.3 we may identify ? (id ® ???) = id ® ? (???). Since H(C*(Y,A)) = H*(Y,A) is concentrated in degrees ? > 2 (cf. §4(e)) it suffices to prove that ? and ? are quasi-isomorphisms and that ??? is fc-regular. A homotopy (Y, yo) x / —> (Y, yo) from the identity to the constant map lifts to a homotopy (X, F) ? I -^ (X, F) from the identity to a map ? : X —> F. More precisely, ? is a homotopy id ~ jg. The restriction of ? to F ? I is a homotopy id ~ gj; i.e. ? is a homotopy inverse for j. Define ? — (? ?,? a) '¦ (?, ??) —> (?,?) ? F by ??(?) = (??,??). By the long exact homotopy sequence ?? and ?? are weak homotopy equivalences, and so CF) is a quasi-isomorphism. On the other hand, a commutative row-exact diagram 0 —*- C ((?,?) ? F) ^ C'(Y ? F) ^ C'(A ? F) ^ 0 h 0 -+ C*(Y,A) ®k C'(F) -^ C'(Y) ®k C'(F) -^ C'(A) ®k C'(F) -^ 0 is defined by Xi(a®b) = C*(-KL)a\jC*(nR)b, irL and irR denoting the projection of - ? F on the left and right factors. By hypothesis either H'(Y,A) or H*(F) has finite type. In the former case (?,?) has finitely many cells and H*(A) is finite dimensional. Thus in either case Proposition 5.3 asserts that the ?* are quasi-isomorphisms. Now observe that C*F) ? ?3 = ? ? (??®?*(?)). Since ? is a homotopy equivalence ?*(?) is a quasi-isomorphism. Hence so is ?. To see that ? is a quasi-isomorphism we observe that because Y is contractible, 1; —> C*(Y) is a quasi-isomorphism. Since J: is a field ?? is trivially Jc-semifree as well as C"*(F)-semifree. Hence Theorem 6.10 (ii) asserts that ? : -®kMY —> — ®c-{Y) ?? is a quasi-isomorphism. Finally, consider the commutative diagram C* ( ?) ) C* ( F) m C'(Y)
Homotopy Theory 81 Here C"(j) is a quasi-isomorphism because j is a homotopy equivalence, while ? and ? <8> id are quasi-isomorphisms because Y is contractible and ?? is semifree. Since fhy is fc-regular (by hypothesis), ??? is /c-regular too. Case 2: Y = .4U|Je2- a Because cells are attached by based maps the characteristic map has the form ? ¦ (Va^^Va^r1) -»· (Y,A). Put ? = \JaD2 and ? = Va-ST1. and consider the pullback (?,?) Let oa be the centre of ?>? and put ? = ? - \\{oa} and U = Y - U{oe}. a a The inclusions ? c—> ? and A c—> U are then homotopy equivalences and so, as in §4(e), an excision argument shows that C*(V0 is a quasi-isomorphism. Since MY is semifree, ?'{?) <g> id : C'(Y,A) ®c-(Y) My —> C'{Z,B) ®C-(Y) My is also a quasi-isomorphism. Moreover the inclusions ?? c—> Xo and Xa c—^ Xu are weak homotopy equivalences so that, again by excision, C*(tp) is a quasi- isomorphism. Consider diagram G.6). The horizontal arrows are quasi-isomorphisms so we need only show mz,B is {k + 2)-regular. As remarked in G.7) fnz = ???. Since H*(Z,B) = H"(Y, A) this conclusion now follows from Case 1. Case 3: (Y, A) has only finitely many cells. Let Yn be the ?-skeleton of (Y, ^4), denote ??? simply by Xn and put m(n, s) = mYntYs : C*(yn,ys) ®c-(Y) MY —)¦ C'(Xn,Xs). By Case 2, each m{n + l,n) is (k + 2)-regular. Induction on ? via the Five lemma 3.1 and the row exact diagram m(n+l,n) m(n+l,O) shows that each m(n, 0) is (k + 2)-regular. Since (Y, A) is finite ???,? — m(n, 0) for some n. Hence ???,? is (k + 2)-regular. Case 4: HS(Y, A) = 0, s < k + 3. Here we show ???,? is {k + 2)-regular by proving that Hs{C*{Y,A)®c.iY)MY) = 0, s < k + Z and Hs {C*{X,XA)) = 0, s < k + 2. For this recall that MY = {C*{Y) <g> W,d) with W = {W'J^o and d : Wi —> C*(Y) ® ?^<\ This gives C*(y,/1) ®c-(r) My = (C*(Y,A) ® W,d) with d :
82 7 Semifree cochain models of a fibration Wi __,. Cm{Y,A) ® W<1. Since if5 {C*{Y,A)) = HS{Y,A) = 0, s < Jfc + 3, the exactness of H (Cm(Y, A) ® W<*) —>¦ H (C*(Y, A) ® T<F^) —s· ii (C*(F, A)) ® W* implies via induction on i that #s (C*(Y, A) ® T'F^') = 0. s < A; + 3, all i. Hence iP(C*(F,A) ® W) = 0, s < A; + 3. Next observe that a non-zero class a in H*(X,Xa) restricts to a non-zero class in H*{Xz,Xa) for some finite subcomplex {Z,A) C (Y,A). (Indeed (cf. §5) ? is non-zero on some cycle ? in C*{X,XA) which, for compactness reasons, is necessarily in some C*{Xz,Xa)·) Thus to prove the second assertion it is suf- ficient to show that the restriction maps Hs (C*(X,XA)) —> Hs (C*{Xz,Xa)) are zero for s < k + 2 and (Z, -4) finite. Fix Z. We first extend it to a larger finite subcomplex (Z[1],A) C (Y,A) so that the inclusion ? : {?,A) —> (Z[l],.4) satisfies ??(?) = 0, s < k + 2. (This implies ??(?) = 0, s < k + 2). Indeed the Cellular chain models theorem 4.18 shows that Ht(Z,A) has a finite basis. Let ui,...,% be cycles representing the basis elements of degree < A: + 2. The hypothesis H-k+-(Y,A) = 0 implies H<k+2(Y,A) = 0; hence uA = db\ in C*(Y,A). By compactness 6? G C*(Z\,A) ? for some finite subcomplex (??,?) D (Z,A). Put Z[l] = \J Zx. ?=? Finally, iterate this construction to give an increasing sequence (Z,A) C (Z[l], A) C · · · C {Z[?], A) c · · · of finite subcomplexes such that the restriction maps Hs (Z[?\, A) —> Hs (Z{? - 1], .4) are zero for s < k + 2. For any subcomplex (?, A) C (Y, -4) recall that (C*(?, .4) ® W, d) = C*(T,A) ® (C*(Y)®W,d). In particular, suppose ? G C-m (? [1],.4) ®W is a cocycle of degrees < A:+2. We remark that (C*(r?) ® id) ? is cohomologous to a cocycle in C-m+1(Z, A) ®W. Indeed, writing ? = ?™ + $>m+i + ¦ ¦¦ with ?^ G C" (Z[l],.4) <g> W, and recalling that d : Wj —> C*(—) ® W<j, we deduce from ?? = 0 that (d ® ??/)??? = 0. Thus ?,? = ??? ® wa with daQ = 0. Since W is concentrated in non-negative degrees m < s < A: + 2. Thus ?*(?)[??] = 0 and ?*(?)?? = dba. Evidently {?*{?) ® id) ? - d (Eba ® wQ) G C^m+I (Z, A) ® W. Put (?,?) = (Z[fc + 3],A) and let ? : (?, A) —> (?,/I) be the inclusion. If ? G C* (?, ?) ® PF is a cocycle of degree s < A: + 2, it follows now that {?*(?) ® id) ? is cohomologous to a cocycle in C^h+2{Z, A)®W. Since s < A;+2 and W = {W*}^, this cocycle is zero; i.e. Hs (?*(?) ® id) = 0, s < A; + 2. Consider the commutative diagram H'(XZ,XA) i H'(XT,XA) |Hs(mr..4) Hs {C*{Z,A) ® IF) < iis (C*(r.^
Homotopy Theory 83 The vertical arrows are isomorphisms (Case 3) for s < k+2. Thus Hs(Xt, Xa) —> Hs{Xz,X\) = 0, s < k + 2, and so, a fortiori, is the composite HS(X,XA) —> We now prove the proposition in general. Let (C,, d) denote the cellular chain complex (§4(e)) for Y: Cn is free on a basis va corresponding to the n-cells {e™}aeTn °f Y ¦ Since J: is a field we may choose subsets fCn C Jn C Tn such that Cn = (kero)n ? ? kva = (Imo)n © ? ???. qgac„ oejn Let yn be the ?-skeleton of the relative CW complex (Y:A), and define a se- quence of subcomplexes ¦ ¦ · (S(n),A) C (T(n),A) C (S(n + 1), .4) C · · ¦ by 'S(n) = yn_x U ( U el ) and T(n) = rn_x U ( |J ena ) . Using the Cellular chain models theorem 4.18 we see that H*(T(n),S(n))=Hn(Y,A), H*(S(n + l),T(n)) = 0 and Hi{Y:S{k + 4)) = 0, i<k + 3. By Cases 2 and 4, the morphisms mr(n)iS(n), ??^^^??,) and myiSn+4) are all (k + 2)-regular. Since 5A) = .4 (because (Y,A) has no 1-cells or 0-cells) the same argument as in Case 3 shows ???,? is {k + 2)-regular. ? proof of Theorem 7.5: Choose a weak homotopy equivalence ? : (Z,zq) —> (Y,yo) from a CW complex ? (Theorem 1.4). The construction in the proof of 1.4 shows that ? may be chosen with a single 0-cell, zq, no 1-cells and with all cells attached by based maps. Let ? : Xz —> X be the pullback map. It, too; must be a weak homotopy equivalence. Thus in the commutative diagram (cf. G.3)) C*(Xz) < C {?) C*(X) C*{Z) ®C.(Y] MY < C*(Y) ®c-(Y) My , both my and the horizontal arrows are quasi-isomorphisms; hence ?? ? is a quasi- isomorphism too. Since H*(Z) S H*(Y), the pullback fibration satisfies the hypotheses of 7.5. Thus since mz — my (cf. 7.7) it is sufficient to prove the theorem for mz : C*(Z) <8>c-(y) My —> C*(XZ) and mz; i.e. we may assume Y is a CW complex as described above. Put H = H*(F;k) and let ? : (H,0) —> C*(F) be a (not necessarily mul- tiplicative) morphism of cochain complexes satisfying ?{?) = id. We con- struct now a C*(y)-semifree module (C*(Y) ®H,d) and a quasi-isomorphism
84 7 Semifree cochain models of a fibration ?? : (C*(Y)®H,d) —> C*{X) of C*(y)-modules such that d:Hn ^C*(Y)®H<n, all ? and ?? = ? : (H,0) -^ C*(F). Note that the first condition will imply that (C*(Y) ® H,d) is semifree. Suppose d and ?? are constructed in C*(Y) ® H<n. Consider the row-exact commutative diagram 0 ^ c*(X, F) ^ C*(X) *- C*{F) *- 0 0 ^(C*(Y,y0) ®H<n,d) ^ (C*(Y)®H<n,d) ^(H<n,0) ^0. Since ?? = ? is (? — l)-regular by construction, Proposition 7.8 asserts that ??,?? is (n + l)-regular. Let {7^} be a basis of Hn and lift the cocycies 6jk to elements xk G Cn(X). Then dxk is a cocycle in Cn+1(X,F) and so dxk = ?Y,yo(uk) +dvk for some cocycle uk G C(Y,y0) ®H<n and some vk G Cn(X,F). Extend d and ?? by setting d^k = uk and ??{??) = Xk — vk. In this way (C*(Y) ®H,d) -^ C*{X) is constructed. Since ?? is the quasi-isomorphism ?, Proposition 7.8 asserts that ??^0 is a quasi-isomorphism too. Hence, by the five lemma, so is ??. Finally, consider the diagram (C*(Y)®H,d) Since ?? is semifree we may (Proposition 6.4 (ii)) find a C*(y)-module mor- phism ? : ?? —> (C(Y) ® ii,d) and a C*(y)-linear map h : MY —> C*(X) such that — my = dh + hd. In particular ?(??) ? ?{?) = ?(???) and 1/ is a quasi-isomorphism. It follows by Proposition 6.7 (ii) that u factors to give a quasi-isomorphism ? : Jfe <g>c-(Y) MY —> J: ®c.(y) (C*(F) ® ii,d). Similarly, since ? is C*(Y)- linear, a linear map h : h ®c(Y) ?? —> C*(F) is defined by the commutative diagram C*(F) h ®C(Y) ?? < My
Homotopy Theory 85 It is immediate from this construction that ?,?? — ??,? = dh + hd. Hence ?(???) = ?(??) ??(?>) is an isomorphism. ? We finish this section by extending Theorem 7.5 to a somewhat more general setting. We have worked till now with the square c'(j) k < C*(Y). Now suppose more generally that -fi» ??? . ? is any commutative square of cochain algebra morphisms. A quasi-isomorphism V -=> V(l) from ? to a second such square D(l) consists of cochain algebra quasi-isomorphisms ?:?-=>?A), 7:?-=>?A) and a : A-=> A(l) that define a commutative cube connecting T> to T>A). In particular, we shall say that the squares T>c and T> above are weakly equivalent if they are connected by a finite chain of quasi-isomorphisms of the form V ^ V{1) A- X>B) ^ ¦¦¦^¦Vc . Now consider the square V. Left multiplication by ?F), b G ? makes ? into a left B-module for which (Proposition 6.6) there will be a semifree resolution m ? '· Mb ^> E. From this we may, as in G.4), define a morphism m b of complexes by the commutative diagram .4 <-$— ? *jme G.9) k ®b Mb < ;— ? (8>b Mb . Theorem 7.10 Suppose k is a field, ? : X —> Y is a fibration with Y simply connected and that one of the graded Ik-vector spaces H*(Y), H'(F) has finite type. If T> is a square weakly equivalent to Vc and if tub · Mb -^ ? is a B-semifree resolution then rnB · -& ®b Mb ? quasi-isomorphism.
86 7 Semifree cochain models of a fibration proof: Suppose given a quasi-isomorphism V —> V{\) as above. Then we have the commutative diagram ?A) ? ®B MB ~ ) -B(l) i?®d which identifies m as a quasi-isomorphism. From this we deduce that m = a o ifiB · & <S>b MB —> A(l). On the other hand, suppose m(l) : M(l) -=> E(l) is any B(l)-semifree res- olution. Then, exactly as in the last part of the proof of Theorem 7.5, we can find a quasi-isomorphism ? : M(l) —> B(l) ®b Mb of B(l) modules such that m ? ? is ??(l)-homotopic to m(l). Since ? is a quasi-isomorphism so is ? : ]k ®b(i) M(l) —> M ®b Mb· Moreover, exactly as at the end of the proof of Theorem 7.5, H{M) ? H(u) = H(m(l)). Altogether we conclude that tiib is a quasi-isomorphism <=$¦ misa quasi-isomorphism -?=> ??A) is a quasi-isomorphism. Consider the chain connecting V to T>c, and recall that the semifree model for T>c is just ??? : ?? —> C*(X). Thus the argument above, repeated along the chain, shows that tub is a quasi-isomorphism if and only if ??? is. But ?? is a quasi-isomorphism by Theorem 7.5. ? Exercises 1. Prove the converse of Theorem 7.5. 2. Let ? : X -> Y be a fibration with simply connected base. Assume Ik is a field and that H*{F;Ik) has finite type. Let (R,d) -> C*(Y,Ik) be a quasi- isomorphism of cochain algebras. Prove that the fibration ? admits a minimal semifree resolution (§ 6-exercise 7) of the form (#<8> ?, d) with ? — H*(F; Ik) and d(Hk) C R®H<k. Write, for any ? G ?, ?? = ??? + ... + ??? + ... with ??? G {Rn ® ?). Prove that this sum is finite and if ^? = ... = ??? = 0 that ??+1? is a cocycle in (R® H,d). A class ? is transgressive if there exists ? such that ??? = ... = ??? = 0 that ???+1? ? 0. Prove that this definition does not depend on the choice of the minimal resolution and that the set ? of transgressive classes is a vector space, so that ? ?-» [dn+1 ?] defines a linear map Tn 4 Hn+1(Y). A fibration is totally transgressive if H*(F,Ik) = T. Characterize the minimal resolutions of totally transgressive fibrations. 3. In exercise 2, assume F = Sn. Prove that the fibration admits a minimal semifree resolution of the form (R®Ax/(x7),d) with dx G Rn+1. The cohomology class ? = [dx] G Hn+1{X;Ik) is called the Euler class of the fibration. If ? is odd, prove that we have a short exact sequence 0 -> (R, d) -> {R® Ax, d)-
Homotopy Theory 87 Ikx, d (8> id) —» 0 with associated long exact sequence ( Gysin exact sequence) 4. In exercise 2, assume Y = Sn. Prove that this fibration admits a resolution of the form (?6/F),0) -> (Ab/(b'2) ® H,d) 4 (tf,0) with ??? = b ® ?(?), ? G ff, deg b = ?. Deduce that there is a short exact sequence 0 —» (J7c6 <8> if, 0) —» (/\b/(b2) <8> H, 0) -> (if, 0) -> 0 with associated long exact sequence ( Wang exact sequence) ... -> iP(F) -> ^+?(?) ^ Hi+n(F) -> if'+1(F) -> .... 5. Prove that the homotopy fibre of the canonical projection S2 VS1 -> 51 is an infinite bouquet of 2 dimensional spheres. Does the conclusion of theorem 7.5 hold? 6. Let F ? ? ? ? be a fibration as in exercise 2. Assuming that ??;(?) is an isomorphism for i < n, deduce that H{{p) is an isomorphism for i < n. 7. Let (X,A) be a pointed pair. Construct, for each ? > 1, a natural map h(.x,A) '¦ ??(?,?,??) —? Hn(X,A;Z) such that (hA,h.x,h(xtA)) maps the ho- motopy long exact sequence into the homology long exact sequence of the pair (X, A). Deduce Whitehead's theorem from exercise 6: Let / : X —» Y be a continuous map with .Y and Y path connected. If (??^./) <gi ifc is an isomorphism for k < ? and is surjective for k = ? then Hk{f; Ik) is an isomorphism for k < ? and is surjective for k = n.
8 Semifree chain models of a G-fibration As usual, we work over an arbitrary commutative ground ring, h. For simplicity, we denote C»{—\k) simply by C*(—) and H^(—\k) simply by Ht(—). In §2(b) we defined the action of a topological monoid, G and the notion of a G-fibration. Here we show how multiplication in G makes C*(G) into a chain algebra, and how an action of G in a space ? makes Ct{P) into a C*{G)~ module. Then, in the case of G-fibrations ? A I we show (Theorem 8.3) that: if (M,d) -?» C*(P) is a C«(G)-semifree resolution then there is an induced quasi-isomorphism (M,d) ®c.(G) ¦& —> C,(X). Notice that this is the exact analogue of Theorem 7.5. Using Theorem 8.3 we then give a geometric application of the isomorphism theorems of §6. Finally we apply this to provide an elementary proof of the Whitehead-Serre theorem (Theorem 8.6): // k is a subring of Q and if ? : X -> Y is a continuous map between simply connected spaces then ?\»{?) ®z h is an isomorphism if and only H*(ip;]k) is an isomorphism. This section is divided into the following topics: (a) The chain algebra of a topological monoid. (b) Semifree chain models. (c) The quasi-isomorphism theorem. (d) The Whitehead-Serre theorem. (a) The chain algebra of a topological monoid. Suppose G is a topological monoid. To make Cm(G) into a chain algebra we use the Eilenberg-Zilber chain equivalences introduced in §4(b). If ? is the multiplication in G then we assign to C»(G) the multiplication C,{G)®C,{G) It is associative because ? ? and ? are. The identity 1 € C»{G) is the 0-simplex at the identity e € G. By C*(G)-module we shall mean a module (M, d) over the chain algebra, C*(G). The constant map G —> pt may be regarded as a map of monoids, and so the augmentation e:Cm(G;k) —> Jfc (8.1) is a dga morphism. In particular it makes Jk into a C«(G; Jt)-module. In the same way a right action ?? : ? ? G —> ? induces the morphism CJun) ? ?? : C.(P;k) ® C*{G;k) —> C.(F;Jfc), which makes C*(P;k) into
Homotopy Theory 89 a right C»(G; Jc)-module. In the case of a G-Serre fibration ? : ? —> X the commutative diagram ? ? {pt} ——> ?" translates (cf. D.5)) to the algebraic formula C.(p)(u · a) = e(a)C.(p)u, u € C.(P), a € C,(G). Thus a morphism, m : M —> C*(P), of right C« (G)-modules induces a unique commutative diagram (?), ?? = x®c.(G) 1, a: € M . (8.2) of chain complexes. When m = id : C*(P) —> C*(P) we write ? = C»(p) : (G)^ —> C»(X). Otherwise we will usually denote ? by m : M®c.(G) (b) Semifree chain models. In general, C,(p) is not a quasi-isomorphism. However, if we replace C»{P) by a C« (G)-semifree resolution, we have Theorem 8.3 Suppose ? -^ X is a G-Serre fibration and m : M -=> C»(P) is a C»(G)-semifree resolution. Then m: M ®c.{G) * —> C*(X) is ? quasi-isomorphism. The main step in the proof of Theorem 8.3 is the construction of a certain 'geometric' C"*(G)-semifree resolution of C*(P) when X is a CW complex. This is a simple generalization of the cellular chain models of §4(e). We first carry out this construction, and then prove Theorem 8.3. Thus we suppose X is a CW complex with ?-skeleton Xn. Recall from §4(e) that a cellular chain model for X is a quasi-isomorphism q : (C, d) -%¦ C*(X) that restricts to quasi-isomorphisms qn : (C<n,<9) -^» C*(Xn), ? > 0. In particular In induces an isomorphism Cn -^» ??(??,??_?), and this identifies Cn as a free Jfc-module on a basis {cQ} corresponding to the ?-cells of X. Moreover d is identified with a certain connecting homomorphism. Denote by ? : Pn —> Xn the G-fibration obtained by restricting ? to Pn =
90 8 Semifree chain models of a G'-fibration Definition A cellular chain model for the G-fibration ? : ? —> X is a quasi- isomorphism of right C,(G)-modules of the form m:(C®C.(G),d) 4C,(P), restricting to quasi isomorphisms m(n) : (C<n ® C*(G), d) -^> C*(Pn), and such that m : (C,d) = (C®C.(G),d) ®C.(G) * —> C.(A') is a cellular chain model for X. Remark Since C,{G) is concentrated in non-negative degrees necessarily d : Cn ® 1 —> C<n <8>O,(G). This exhibits m : (C®C.(G),cQ —> C.(F) as a C,(G)-semifree resolution. Proposition 8.4 Every G-Serre fibration whose base is a CW complex has a cellular chain model. proof: Denote the fibration by ? : ? —> X and adopt the notation above, so that the ?-cells of X form a basis, cQ, of Hn(Xn,Xn-i). We show first that Hm(Pn,Pn-i) is a free H«(G)-module on a basis aQ € Hn(Pn, Pn-i) satisfying H,(p)aQ = cQ. Write Xn = Xn-i U fljes) and let / : fIJ^2,IJ^2 (Xn,Xn-i) be the characteristic map. Write {D,S) = ( U-D^US^ ) and \ Q Q / use / to pull the fibration ? —? ? back to the G-fibration PD = DxxP^ D, (y,z)-g = (y,zg), (y,z) € F/j, g € G. Let Ps be the restriction of PD to S, and let ? : (Fd,Fs) —> (Fn,Fn_i) be the pullback map. We first observe that H,(f) and ? „(?) are isomorphisms. Indeed, if oQ is the centre of DJ put ? = D - \\{oa}, U = Xn - ]J{oQ}, a a Po = p?(O) and Pc/ = p-^C/). The inclusions S^—>O and ?'?_? <=—>*/ are homotopy equivalences, and hence P5 —>¦ Po and Pn-i c—> Pu are weak homo- topy equivalences (Proposition 2.3). Exactly as in §4(e) it follows by excision that H,(f) and ?*(?) are isomorphisms. Choose a contracting homotopy Dxl—> D from the map ]\ D%—> \J {o?} to the identity. Choose points zQ € p~jj{oa) and lift this homotopy to a map ? : D ? I —> PD, starting at the map ]} #2 —> LI {^2}- Denote ii(-, 1) by ? : D —> PD. Then pD ? ? — id. Let D x G —> D be the trivial G-fibration (with action (x,g) ¦ gx — {x,ggi)) and define (D,S)xG * >- (PD,PS) (D,S)
Homotopy Theory 91 by ip(x,g) = ?(?) ¦ g. By the very definition of G-Serre fibrations ? restricts to weak homotopy equivalences in the fibres. Hence (Proposition 2.3) ? restricts to a weak homotopy equivalences D ? G —> Pp and S x G —> Ps- It follows that ? „(?) is an isomorphism. Because Hm{D,S) = Hn(D,S) is a free module (on the ?-cells of X), there is a chain equivalence Hn(D,S) -^ C*{D,S). Thus the composite Hn(D,S)®Cm(G) ^ C((D,S)xG) defines an isomorphism ?(?) : Hn(D,S) <g> i/,(G) —> i/»(Pn,Pn_i). The asso- ciativity of the Eilenberg-Zilber map implies that ^Z above is a morphism of C»(G)-modules. Hence ?(?) is an isomorphism of H*{G) modules. Let aa € Hn(Pn,Pn-i) be the image of X(n)(cQ <8> 1). Then Hn(Pn,Pn-i) is a free right //„(G)-module with basis {aa}, and Hm(p)aQ = cQ. We now construct the semifree resolution m : (C'<8> C,(G),d) —>¦ C*(P) by an inductive process. Assume the quasi-isomorphism m(n — 1) : (C<n_i ® C*(G),d) -^> C»(Pn_i) is defined. Let 6Q € Cn(Pn) be an element that projects to a cycle zQ € Cn(Pn)-Pn-i) representing aa. Then d6Q is a cycle in C,(Pn-i). Choose a cycle ua € C<n_i ® C*(G) and an element wQ € Cn(i?n-i) so that m(n — \)(ua) = d6Q + dwQ. (Recall H(m(n — 1)) is an isomorphism.) Extend d and m(n — 1) to C<n <8> C»{G) by setting d(cQ ® 1) = «Q and m(n)(cQ ® 1) = bQ + wa. The map C<n/C<n <S> C,(G) —> C*(P„,P„_i) induced from m(n) produces, in homology, the morphism I 0iccQ 1 <S> H,{G) —> H,(Pn,Pn-i) of V a J #»(G)-modules given by cQ i-» aa. This is the isomorphism ?(?). Thus by the Five lemma 3.1, m(n) is a quasi-isomorphism. This completes the inductive construction of m : (C <8> C,{G),d) —>¦ C*(P). Since any singular chain in ? lives in a compact subspace it must, in particular, be contained in some Pn. It follows from this, and the fact that each m(n) is a quasi-isomorphism, that m is a quasi-isomorphism too. Finally, by construction, m(cQ <8> 1) = 6Q + wQ projects to a cycle 2Q € Cr»(Pn,pn_1) such that C,(p)zQ represents cQ. This is precisely the condition that m : (C, d) —> C.(X) be a cellular chain model. D proof of Theorem 8.3: We are given a G-fibration ? A X and a C,{G)- semifree resolution m : M A C»(P). Let 5 : Y —> .Y be a weak homotopy
92 8 Semifree chain models of a G—fibration equivalence from a CW complex Y (Theorem 1.4) and form the pullback Y*xp * . ? ? ? is a map of G-fibrations and a weak homotopy equivalence (Proposition 2.3). Let ? : (C ® Ct(G),d) -^» C*{Y x.v P) be a cellular chain model for our G- fibration; its existence is guaranteed by Proposition 8.4. Recall (diagram (8.2)) that ? factors to define ? : (C ® C, Since (C <8> C»(G),d) is C,(G)-semifree we may, in the diagram M (C®C.(G),d) _?_^?(?), lift Ct(ip) ? ? through m to obtain a morphism ? : (C 0 C,(G), d) —> M and a C.(G)-linear map h : C ® C'.(G) —> C(P) such that mi/ - 0»{?)? = dh + hd (apply Proposition 6.4 (ii)). Since m, C»{ip) and ? are quasi-isomorphisms so is v. Thus (cf. Proposition 6.7 (ii)) is a quasi-isomorphism. Finally, since h is C« (G)-linear we can form h:[C®C, Evidently fhv-C»{g)]i - dh + hd. Since i>, C*{g) and /x are quasi-isomorphisms, so is m : M ®c,(G) & —> C»(X). ? (c) The quasi-isomorphism theorem. Suppose given • a morphism a : G —> G' of topological monoids, and
Homotopy Theory 93 • a commutative square of continuous maps where ? is a G-fibration, p' is a G'-fibration and ?(??) = ??(?)?(?), ? € P,aeG. Then C»{a) is a chain algebra morphism. Thus it makes any C«(G')-module {eg. Ct(P')) into a C« (G)-module. Moreover Ct(<p) is a morphism of C*(G)- modules: C.M(u · *) = [C.{<p)u] ¦ [Cm(a)x], u € C.(P),x e C(G). Theorem 8.5 W^ii/i ?/ie notation above, assume G is path connected and C(a) is a quasi-isomorphism. Then C*(<p) is a quasi-isomorphism <=? 0»{?) is a quasi-isomorphism. proof: The proof of Proposition 6.6 shows that C.(P) and C.(P') admit C*(G) and C»(G') semifree resolutions m:(M,d) A C.(P) and m' : (M',d) -^ C.(P') with M and M' concentrated in non-negative degrees. Since (M, d) is semifree and m' is a quasi-isomorphism we can find / : (M, d) —> (M', d) and O : M —> C.(P') such that / is a morphism of C«(G)-modules, O is C«(G)-linear and m'/ = C,(ip)m + ?? + ?? (Proposition 6.4 (ii)). Recall the notation of Theorem 6.12, which we wish to apply here, noting that Hq (Ct(G)) = Ho (G.(G')) = Jk. Diagram (8.2) and its analogue for the fibration ? yield the following diagram of chain complex morphisms: C.{X) > C.(X'). CM This diagram may not commute. However, because ? is C«(G)-linear it factors to give a linear map ? : M ®c.(G) k -+ C»{X') and m'(/®c.(Q) it) = C»{i>)m +
94 8 Semifree chain models of a G-fibration ?? + ??. Thus C*(tp) is a quasi-isomorphism if and only if / is and C*{ip) is a quasi-isomorphism if and only if / ®c*(a) & is. Now apply Theorem 6.12. D Theorem 8.5 has the following important corollary. Consider a fibre preserving map between Serre fibrations ? and p': X f X' and let fy : Xy —> X'gy be the restriction of / to a map between fibres. Corollary to Theorem 8.5 Assume Y and Y' are simply connected and X and X' are path connected. If C+(Clg) is a quasi-isomorphism then Cr(f) is a quasi-isomorphism <?=> C»(fy) is a quasi-isomorphism. proof: The construction in §2(c) of the holonomy fibration is natural. Thus there are commutative diagrams X PY fxPg ' X1 xY. FY' and X f X' X XYPY with ? and ?' weak homotopy equivalences. Since Y and Y' are simply connected, UY and ?.?' are path connected. Now apply Theorem 8.5. to the first diagram. D (d) The Whitehead-Serre theorem. Theorem 8.6 (Whitehead-Serre) Suppose Jk is a subring of Q and g : X -> Y is a continuous map between simply connected spaces. Then the following assertions are equivalent: (i) TT*(g) <8>z -& is an isomorphism, (ii) Ht[g\]k) is an isomorphism. (Hi) H*(Qg;]k) is an isomorphism.
Homotopy Theory 95 Remark Here ?, (g) ®z k denotes the homomorphism ?* (g) ®? idjk : ?» (?') <8>? k —> irt(Y)®zk. We emphasize the role of k by using the full notation C« (—;k) and #«(—; k). When k = ? the theorem reduces to Whitehead's original result: g is a weak homotopy equivalence if and only if Ht(g; Z) is an isomorphism. Lemma 8.7 Suppose k C Q, X is an (r — l)-connected space and either r > 2 or r = 1 and tti(X) is abelian. Then the Hurewicz homomorphism defines an isomorphism *r(X)®Zk = Hr(X;k). proof: By Theorem 4.19 the Hurewicz homomorphism is an isomorphism nr(X) -=> Hr(X;Z). Thus it defines an isomorphism nr(X) <S>z ¦& -^> Hr(X;Z) ®z Ik. On the other hand the identification C*(.,Y;Z)<g>z.& = C»(X;1;) defines a map ii,(.Y;Z) <8>z & -> H,(X;]k). Because Je C Q the operation -®zA preserves exactness, and it follows that this map too is an isomorphism. ? We shall make frequent use of Eilenberg-MacLane spaces (§4(f)). (The reader may wish to review §4(f), since we shall often use there properties established without explicit reference.) Finally, we shall also rely on the Proposition 8.8 Suppose h C Q and X is an Eilenberg-MacLane space of type (?r, ?), ? > 1, with ? abelian. Then ?®?* = 0 «*¦ H,{X;k) = H,{pt;k). proof: Since ir®zl: = Hn(X; k) we have only to prove that if ? <S>z k = 0 then Ht(X;k) = Ht(pt:k). Case 1: ? = 1 and it is cyclic. Since ? <g>z A = 0, there is a short exact sequence where ? is a multiplication by an integer k invertible in k. Let / : K(Z,2) ?(?,2) be a map such that ???(/) = C and convert / to a fibration The long exact homotopy sequence identifies F as an Eilenberg-MacLane space of type (Z5 2) and identifies n2(i) with the homomorphism ? : ? -> ?. Now consider the commutative square UE ??, 2)
96 8 Semifree chain models of a G-fibration as a map of QF-fibrations, where the right hand side is the loop space fibration described in §2(b). The long exact homotopy sequences for the path space fi- brations identify QF and ?.? as Eilenberg-MacLane spaces of type (Z,l) and identify ??^??) with ?. Thus Hi(O.i; k) = ???(??) ®? k is an isomorphism. Moreover, because the universal cover, E, of S1 is contractible it follows that S1 is an Eilenberg-MacLane space of type (Z, 1). Since all Eilenberg-MacLane spaces of the same type have the same homology, Hi(Q.E;k) = Hi(O.F;lc) = 0,i > 2, and so Ht(Cli,]k) is an isomorphism. We can therefore apply Theorem 8.5 to the map of OF-fibrations above to conclude that ?»{??(??,2)\?) - H*{pt\k). But ??{?,2) is an Eilenberg- MacLane space of type (?, 1) and hence has the weak homotopy type of X, whence H.(X;k) = H«{pt\k). Case 2: ? is cyclic. We induct on n, the case ? = 1 having been established in Case 1. Consider as a map from an {e}-fibration to the ?.?-path space fibration on X; {e} denoting the 1-point monoid. By the induction hypothesis, H*({e};k) -^ Ht(QX,k), since CtX is an Eilenberg-MacLane space of type (??,? — 1). Ap- ply Theorem 8.5 to conclude that H,(pt;h) -=> H,(X;h). Case 3: The general case. It is easy to modify the construction in §4(f) of cellular Eilenberg-MacLane spaces to construct a ?{?, ?) that is the union of sub CW complexes ?(??, ?) as ? a runs through the finitely generated subgroups of ?, and such that K(nQ,n) —> ?"(??, ?) induces the inclusion ?? —? ?. We may also assume that for any ??, ao there is a ? such that ?(???,?) D ?(???,?), i — 1,2. Since — ®z h is exact, each ?? ®z & = 0. If we can show that each ?,(?(??; ?); fc) ~ H,(pt; Jfc) it will automatically follow (eg via cellular chains) that ?*(?(?,?); Je) = Ht(pt;Jk). Since X has the weak homotopy type of K(n,n) this will complete the proof. We may therefore as well assume that X is one of the ?(??,?); i.e., that ? is finitely generated. Then ? is the finite direct sum ?? ? ... ? Fm of cyclic groups and each ?* <g>z h = 0. Thus X has the weak homotopy type of the product Y\^LiK(Ti,n), and, by Case 2, there are chain equivalences fc -=> C» (K(Ti, n); Jk). Since the tensor product of chain equivalences is a chain equiv- alence, the Eilenberg-Zilber chain equivalence gives
Homotopy Theory ThusHt(X;k) = H,{pt-k). 97 D proof of Theorem 8.6: Apply the Cellular models theorem 1.4 and the Whitehead lifting lemma 1.5 to obtain a based homotopy commutative diagram X' hx X Y' Y in which X' and Y' are CW complexes and hx and hy are weak homotopy equivalences. Since ?/?? and Clhy are also weak homotopy equivalences it is enough to prove the theorem for g': i.e., we may assume .Y and Y are CW complexes, which we now do. Our first step is to reduce to the case that T\2{g) is surjective. Let ? = Ko(Y)/\miT2{g) and let ? = K(T,2). Use Proposition 4.20 to find a continuous map q : Y —> ? such that ^(g) is the quotient homomorphism. Thus T\i{qg) — 0 and so qg is based homotopic to the constant map. In §2(c) we showed how to convert q to a fibration q via the commutative diagram Y a ? ?? ?? in which ? is a homotopy equivalence. Since q Xg is homotopic to the constant map and q is a fibration we may lift the homotopy to a homotopy Xg ~ g\ with 5i mapping X into the fibre ? ?? PK- By the long exact homotopy sequence, "E1) is surjective, and we claim it is sufficient to prove the theorem for g\. In fact the claim follows from the assertion that if i : ? ?? PK —> ? ?? ? ? is the inclusion then ??*(?) ® Jk, H,(i;L·) and Ht(Cli,k) are all isomorphisms, which we see as follows. The Hurewicz theorem 4.19 identifies H2(g) = ^2E) = Ti(Qg) = Hi(Clg). Hence under any of the three assertions of the theorem, "E) ®z Jk is an· isomorphism. Since — ®z k is exact ? ®? h = 0, and so ?*(?) ®z Jk is an isomorphism. Moreover Proposition 8.8 asserts that ?»(?;3? = Jk = Ht(QK,k). Apply the Corollary to Theorem 8.5 to the fibre preserving map Y xKPK Y xK MK pt ?
98 8 Semifree chain models of a G-fibration to conclude that Ht(i;k) is an isomorphism. On the other hand, 7r2(c) is sur- jective since 7r2(c) is. Thus ??(?) is injective and ? ?? PK is simply con- nected. Looping the diagram above gives a map of ?,(? ? ? Pii)-fibrations. Since ?(?" ?? ??) is path connected Theorem 8.5 can be applied to deduce that H,(Qi;k) is an isomorphism. In view of this discussion we may and do henceforth assume that iio(g) is surjective. We first establish that assertions (i) and (ii) are equivalent. Convert g into the fibration ?: ? xy MY —>Y as described in §2(c). The fibre of ? is A' x y PY and the homotopy equivalence ? -=> ? ?? MY identifies g with p. Thus the long exact homotopy sequence shows that ? ?? PY is simply connected. Moreover, since — <g>z k is exact, TTr,(g) ® k is an isomorphism On the other hand (cf. §2(c)), X xy PY ?,(? ?? PY) ® it = 0. PY X 5 ~Y is a morphism of Qy-fibrations. Thus Theorem 8.5 asserts that Hm(g,k) is an isomorphism 4=> H,(X ? ? PY: k) = H,(jpt;k), Let F —> X Xy PY be a cellular model. We have only to show that 7T,(F)<8)ZJt = 0 <=> H*(F;k) = H.(pt;k). (8.9) Define a sequence {Fr}r>i of r-connected CW complexes, beginning with F\ = F, as follows. Given Fr; put Kr+\ = K(nr+i(Fr), r + 1) and choose a continuous map so that 7??+?@?) is the identity isomorphism. Then let Fr+1 be a cellular model for the homotopy fibre Fr XKr+l PKr+\ of 6r. As described in §2(c) we have a morphism of Qiir+1-fibrations, Fr XKr+1 ???? (8.10) Fr ???
Homotopy Theory 99 Moreover, the long exact homotopy sequence has the form ... ? nn(Fr xKr+1 PKr+l) ^H *n(Fr) ^H ??(??+1) ? .... It follows that Fr x-Kr+l PKr+\ is indeed (r + l)-connected and that ??(??) is an isomorphism for ? > r + 2. Thus we have isomorphisms 7rr+1(Fr) A 7rr+1(Fr_i) A ... A 7rr+1(F)„ r > 1. Suppose now that H»(F;L·) = H,(pt;]k). We show by induction on r that H*(Fr;]fc) = H,{pt;L·), noting that this is true by hypothesis for r = 1. If this holds for some r then ?rr+1(Fr) ®zi = Hr+i{Fr\L·) = 0 and so Proposition 8.8 asserts that H*{Kr+i\]k) = H,(pt;]k). Hence ??(??;1?) is an isomorphism. Now Theorem 8.5, applied to the diagram (8.10), asserts that H,(Or,]k) is an isomorphism, whence H.(Fr+1;k) S H.(FrxKr+lPKr+i) as H.(PKr+1;k) = H.(pt;k). Since this is true for every r, the homotopy group isomorphisms above show that 7rr+i(F) ®z k = nr+i(Fr) ®z k = Hr+1(Fr]]k) = 0, r > 1, as desired. Conversely, suppose ttj(F) ®z J: = 0, i > 2. Then the homotopy group isomorphisms above show that ni(Fr) <S>z & = 0 for all i > 2 and r > 1. In particular, 7rr+1(Fr ?? Jk) = 0. On the other hand, if r > 1 then QKr+i is an Eilenberg-MacLane space of type Grr+1(F),r + 1). Hence (Proposition 8.8) H*(QKr+i;k) = H.(pt;k) for r > 1. Consider the diagram Fr xKr+1 PKr+x -U. Fr xKr+1 PKr+l id. Fr *Kr+l PKr+l Fr as a morphism from an {e}-fibration to an Qiir+1-fibration, {e} denoting the 1- point monoid. Apply Theorem 8.5 to conclude that H»(pr; 3c) is an isomorphism, ? > 1. This gives a sequence of isomorphisms A H.(Fr+l;k) A H.(Fr;k) A - · · A ff.(Fi; Jfc) = H.(F;k). Since Fr is r-connected it follows that H.(F;k) = H,{pt;k). This completes the proof that assertions (i) and (ii) are equivalent. To see that (ii) and (iii) are equivalent we recall that we may assume ito(g) is surjective
100 8 Semifree chain models of a G-fibration and we use the notation and constructions above. Theorem 8.5, applied to PX -^- PY ? 9 X implies that if H.(Qg;k) is an isomorphism so is Ht(g;k). Conversely, suppose Ht(g;k) is an isomorphism. Let Fg = X xy PY denote the homotopy fibre of g. Then, as shown above, Ht (Fg; k) = Jk and ?, (g) ®z k is an isomorphism. On the other hand, in §2(c) we constructed an ??-fibration q : P'X xxFg —>· Fg such that the fibre inclusion j : ?.? —> P'X xxFs was identified with Qg by a homotopy equivalence. Thus tt«(j) ®? k is an isomorphism by hypothesis and we have only to show that H» (j; k) is an isomorphism. But the commutative diagram QX 3- P'X *xFg {pt} is a map of ??-fibrations and the map {pt} —> Fg induces an isomorphism in homology. Now Theorem 8.5 asserts that H,(j;k) is an isomorphism too. D Exercises 1. Let Jfc[G] = {%2geG^g g. Afl € Ik}, where \g takes the value zero except on a finite number of elements of G, be the group ring of a group G. Construct a quasi-isomorphism (Jk[G],0) 4 C*(G;Ik)). If ? -> X is a G-fibration (for instance a finite covering) prove that there exists a quasi-isomorphism of Ik[G]- modules Ik[G] ® C,{X\Ik) 4 C.(P;Ik). 2. Let ? € CiiS1;^) be defined by o{t) = e2i7rt. Prove that ? is a cycle such that ?2 = 0. Construct a quasi-isomorphism of algebras (Ax, 0) —> 3. Let ? be the torus S1 ? S1 ? ... ? S1 (? times). Deduce from exercise 2 a quasi-isomorphism of algebras ? : (?(«?, ?2,..., xn), 0) -> G, (T). Assume that T acts from the right on a topological space X. Prove that C*(X; Ik) inherits, via ?, a left (?(??,?2, -.,??), 0)-module structure. 4. Let F -> X -> Y be a fibration with Y 1-connected and consider the associ- ated ny-fibration UY -> F -> X. Prove that if H,(F;Ik) is a free Ht(UY;Ik)- module, then Ht{X;Ik) = Ik ®?.(?^;?·) H.(F;Ik).
Homotopy Theory 101 5. Prove that the spaces ?' = UPm ? 5" and Y = Sm x RPn have the same homotopy groups but not the same homology groups. Does this contradict the Whitehead theorem? 6. Let ? -> X be a G-fibration with ?' 1-connected. Prove that if (T(V),d) 4 C, (ClG) is a quasi-isomorphism of chain algebras then there is a weak equivalence C*{X) 4 HomT(v)(T(F) ® M, Ik), and describe the semifree module TV <g> M.
9 P-local and rational spaces Fix a set ? of prime numbers and let 1 be the set of integers k relatively prime to the elements of P. An abelian group ? is V-local if multiplication by k in ? is an isomorphism for all k € ?. The rational numbers m/k, k € ?, are a subring Jk C Q, so ? is P-local if and only if it is a A-module. By convention, when ? = ? we take it = Q — in this case ? is a rational vector space. Throughout this section our ground ring Jk will be the subring associated in this way with the set of primes P. If 0 -> ?' -> ? -> ??" -> 0 is a short exact sequence of abelian groups, then it' and ?" are P-local if and only if ? is P-local. For any abelian group ?, ? ®?& is P-local. Moreover, if ? is P-local then ? -> ? ®% k is an isomorphism. For general ? this homomorphism ? —> ? ®z Jk , 7 ?—» 7 ® 1 is called its V-localization. The idea of P-local spaces and localization of spaces, introduced by Sullivan in [143], is a topological analogue. As will be shown in Theorem 9.3, if X is a simply connected space then tt.CX") is P-local <=*· Ht(X,pt;Z) is P-local. This motivates the Definition A simply connected space X is a V-local space if it satisfies these equivalent conditions. When L· = Q (i.e., V = ?) ? is called a rational space. This section is organized into the topics: (a) P-local spaces. (b) Localization. (c) Rational homotopy type. (a) V—local spaces. Before turning to the proof of Theorem 9.3, we introduce an important class of P-local spaces: the (relative) CW-p-complexes , beginning with the classical motivating example: the infinite telescope Sj,. Let hi, ko,... be the positive integers relatively prime to ? (i.e.; the denomi- nators of Jk). Put
Homotopy Theory 103 where DJ+1 is attached by a map 5" -> Sf_x V S? representing [Sf^] - kj[S?]. For ? = 1 this is illustrated by the picture ? hi x k-2 x k3 in which the (j — l)st right hand circle is attached to the jth left hand circle by the map e2"u h-» e2lzikit, and the bottom solid lines are identified to a single point. (This explains the terminology: 'telescope'.) The space S? is called the V-local ?-sphere and the space Dp'1 - S? ? I/S? ? {0} is called the V-local (n + l)-disk. When Jk = Q these spaces are called the rational ?-sphere Sq and the rational (n + l)-disk, D^+1. The inclusion of the initial sphere 5J -> S? extends to Dn+1 -> ?>? and this inclusion is called the natural inclusion of the initial disk and sphere. (This is the simplest example of 'P-localization of topological spaces, defined and constructed later in this section.) Now we compute the homolog}' of Sj,, showing that ? i = 0 Jt i = n 0 otherwise with the generator 1 G Hn(Sj,;Z) represented by the initial sphere [5J]. For this, let X(r) C S? be the sub complex V 5f Uh ( JJ Dn-+l). It contains S? as a strong deformation retract: simply collapse the finite telescope X(r) onto this terminal sphere. Hence H,(X(r);Z) vanishes in degrees i ? ?,?, and so #»(Sp;Z) vanishes in these degrees as well. The inclusion X(r) C X(r + 1) sends [S?] to A;r+1[5^+1]. Thus an isomorphism Hn(S?;Z) = Jk is given by mapping each [5?] to ( ? Ay·). When ? > 2, S? is simply connected since its cells, aside from the base point, all have dimension ? or ? + 1 (apply the Cellular approximation theorem 1.2). Thus S?,n > 1, is indeed a "P-local space. When ? = 1,5^ is an Eilenberg-MacLane space of type (A-, 1). Indeed any map ¦j -> S\> has image in some X(r), by compactness. Since X(r) ~ S1 it follows that wmEj,) = 0, m > 2 and ???E^) is abelian. Thus tt^SJ,) = Hi(S$>) = Jk.
104 9 "?—local and rational spaces Relative CW-p complexes are built by assembling "P-local disks via attaching maps from "P-local spheres in exactly the way classical relative CW complexes are constructed from ordinary disks. More precisely, Definition A relative CW-p complex is a topological pair (X,A) together with oo a filtration A' = |J X(n) of X by closed subspaces ?(?) such that n=-l • X(i) = X(o) = X(-i) = A, and A is 1-connected. • For ? > 1, X(n+i) is presented as Xin+1) = lWU/n(Uo^), (9.1) a with /„ : LJ ? Sv,a -» x(n) a cellular map. The D^ are called the V- local (n + l)-cells of (X, A). If A = {pi} then X is a CW-p complex. When ]k — Q we refer to (relative) CWq . Remark Since S? is a subcomplex of the CW complex D^+1 and since /„ is cellular, the adjunction space X(n) U./„ ( LI Q D^+l ) carries an induced structure as a relative CW complex with respect to A. This makes {X, A) into a relative CW complex in which each X(n) is a subcomplex. Since (Dv+l,Sj,) has only (n + l)-cells and (n + 2)-cells, it follows that Xn C X(n) C Xn+i , ? > -1, where Xn is the ?-skeleton of the relative CW complex (A', A). We call X(n) the ?-skeleton of the CW-p-structure on (X, A). In particular (X,A) has no 0-cells or 1-cells and so every circle in X is homotopic to a circle in the 1-connected space A; i.e., X is simply connected. Finally, as follows from the calculation above of i7»(S?;Z), ?,{??+? ,S?;Z) is a free l:-module concentrated in degree n. It follows by excision that each Ht(X(n+i),?(?);?) is "P-local. The long exact homology sequences now shows that H,(X,pt;Z) is "P-local: thus X is a P-local space. We turn now to the statement and proof of Theorem 9.3. This will make frequent use (without reference) of the following elementary remark. Lemma 9.2 Let ?p be the prime field of characteristic p. Then for all pairs of spaces (X,A): H.(X,A;Z) is V - local ?=> H.(X,A;Fp) = 0, p$V. In particular H.(X,pt;Z) is V- local ?=? H.(X;Fp) = H.(pt;Fp), ? $ V.
Homotopy Theory 105 proof: Consider the long exact homology sequence associated with the short exact sequence 0 -> C.(X,A;Z) -^ C.(X,A;Z) —»· C,(X,A;?P) —»· 0. It shows that multiplication by ? in H,(X,A;Z) is an isomorphism precisely ,(A',^;Fp) = 0. ? Theorem 9.3 Let X be a simply connected topological space. Then the follow- ing conditions are equivalent: (i) ??,(?) is V-local, (ii) H.(X,pt)Z) is ?-local. (Hi) H,(QX,pt',Z) is V-local. proof: This is essentially identical to the proof of Theorem 8.6. We begin, as in that Theorem, with the case of Eilenberg-MacLane spaces, where we follow closely the steps of the proof of Proposition 8.8: Lemma 9.4 If X is an Eilenberg-MacLane space of type (?, ?), ? > 1, then ? is V-local =» H, (X; Fp ) = H, (pt; Fp ), ? ? V . proof: Suppose first that ? = 1. Reduce to the case ? is finitely generated exactly as in Case 3 in the proof of Theorem 8.6. Since Jk is a principal ideal domain, finitely generated ^-modules are the finite direct sums of cyclic Jk- modules. Hence in this case X has the weak homotopy type of a finite product of spaces K(Ti, 1) with ?; a cyclic J;-module. Exactly as in the proof of Case 3 in (8.6) we are reduced to the case ? is a cyclic ^-module. If ? = k then X has the weak homotopy type of Sj> and Ht(X,pt; Z) is V- local by the calculation above of the homology of 5^>. If ?? ?. k then there is a short exact sequence of ^-modules of the form 0—>¦&—>¦!;—»-?—»-?. Exactly as in Case 1 of Proposition 8.8, this leads to a morphism of QF-fibrations of the form ?.? ?{ ) UE ?? J. pt
106 9 ?7—local and rational spaces in which both UF and QE are Eilenberg-MacLane spaces of type (k, 1). Thus as observed above both QF and ClE have the weak homotopy type of Sp. Hence H,(QF;Fp) = Ht(QE;Fp) = Ht(pt,Fp). Thus Theorem 8.5 asserts that Ht(pt;?p) A Hm{UK(n,2);Fp). Since ??'(??,2) has the weak homotopy type of X, the Lemma holds for X. This establishes Lemma 9.4 for Eilenberg-MacLane spaces of type (?, 1). The case when X has type (??,?), ? > 1, follows by induction on n; apply Theorem 8.5 to the morphism of ??-fibrations PX pt > X. We now return to Theorem 9.3, whose proof mimics the proof of formula (8.9), with X replacing F and Fp replacing Jk. Thus we construct a sequence of spaces {Xr}r>i starting with a cellular model ?? -> X, and a sequence of maps er:Xr^Kr+u Kr+1 = K(nr+1(Xr),r + l). The space Xr is r-connected, ??+?(??) is the identity isomorphism and Xr+i is a cellular model for Xr *Kr+1 PKr+i. From Qr we also obtain the morphism Xr XKr+1 PKr+i PKr+l (9.5) Xr — " Kr+\ of Q/<"r+i-fibrations and, as in the proof of (8.8), sequences of isomorphisms nr+i(Xr) ^> TTr+l(Xr-l) ^> ^> 7Tr+l(^)· Suppose H*(X,pt;Z) is TMocal. Then also if»(Zr,pi;Z) is TMocal for all r, by induction. Indeed if this holds for some r then ??+1(??) is TMocal and, by (9.4), H,(Kr+l;Fp) = H,(pt;Fp), p?V. Apply Theorem 8.5 to (9.5) to conclude that H,(§r;Fp) is an isomorphism, ? ? V. Hence H*(Xr+l,pt;Fp) = 0, ? <? V, and H,{Xr+upt\ Z) is P-local. In particular each ???+1 (Xr) is "P-local and so ?*(?) is 'P-local. Conversely if ?» (?) is V-local the homotopy group isomorphisms above im- nlv that each ???+1(?'?) is TMocal. Hence, by Lemma 9.4, H,(ClKr+i;Fp) =
Homotopy Theory H*(pt;Fp) for r > 0 and ? g V. Regard Xr *Kr+i PKr+l - Xr 107 Xr as a morphism from an {e}-fibration to an ???+? fibration. Conclude from Theorem 8.5 that iJ»(pr;Fp) is an isomorphism for ? (? V. Since Xr is re- connected, the sequence of isomorphisms implies that H.(X;FP) = Hm(pt;Fp),p ? V\ i.e., H.(X,pt;Z) is TMocal. This shows that (i) <?=> (ii). Suppose (iii) holds. Let j : {xo} -> X be the inclusion of a basepoint and apply Theorem 8.5 to the map Pj : P{xo] —> PX of path space fibrations to conclude that ii»(j;Fp) is an isomorphism for ? ? V. Thus (iii) => (ii). Conversely, suppose (i) and (ii) hold. Recall from §2(c) that there is a fibration and a homotopy equivalence Y ~ X\ which identifies q with ?\. Since X\ ? ?2 ??? is 2-connected and ^i{X\ ??2 ??) 5? ??(??), i > 3 these groups are, in particular, 'P-local and we may apply (i) «-» (ii) to deduce that ?. (?(?? ??2 PK2);FP) = H*(pt;Fp) , ? i V . Now note that iiq is an ?(?? ??2 -PA)-fibration. We shall apply Theorem 8.5 to 9 {pt} {pt} UY Lemma 9.4 asserts that if»(/;Fp) is an isomorphism for ? ? V. Hence so is H.faFp) and H,(UX;FP) S H.(ClXi;Fp) S H.(iiY;Fp) Si H.(pt;Fp), ? i V.U (b) Localization. Suppose ? : X —y Y is a continuous map between simply connected topologi- cal spaces, and that Y is TMocal. Then ?»(?) is a Jfc-module and ?,(?) extends uniquely to a morphism ?.(?)
108 9 "?—local and rational spaces Definition A V-localization of a simply connected space X is a map ? : X —? X-p to a simply connected P-local space X-p such that ? induces an isomorphism When h = Q this is called a rationalization and is denoted by X —> X<q. Theorem 9.6 A continuous map ? : X —> Y between simply connected spaces is a V-localization if and only if Y is V-local and ?*(?;?) is an isomorphism. proof: Since Y is P-local, 7r»(F) = ?*(?) ®z Ik and the homomorphism ?*(?) <8>z & —> ?.(?) is just ? »{?) <8>z Jt. Thus Theorem 9.6 is a special case of the Whitehead-Serre Theorem 8.6. ? Corollary The inclusion Sn —> SC, of the initial sphere in the telescope is a V -localization. ? Localizations always exist, and have important properties. We first describe this with a cellular construction that gives the true geometric flavour of the meaning of P-localization. Example Cellular localization. Suppose (X, A) is a relative CW complex in which Xy = Xo = A and A is a simply connected P-local space. Thus the Cellular approximation the- orem implies that X is simply connected. We shall use the natural inclusion (Dn+1,Sn) —>· (?>p+1,S?) to construct a P-localization of X of the form such that • (X-p, A) is a relative CW-p complex. • The P-local ?-cells of (X-p, A) are in 1 — 1 correspondence with the n-cells of (X,A),n> 2. • ? restricts to the identity in A, and to maps ?? : Xn —>· (X-p)^ satisfying ??+1 = ?? U ( IJ ??) : ?? U/n ( U D2+1) -> (??)(?) Us„ ( U D^), ? ? ? ' ?? denoting the standard inclusion D^+l —> D^. • In particular, ? is the inclusion of a sub relative CW complex. In fact, suppose inductively that (Xv)(n) and ?? are constructed, and let fa : S? —» A'n be the attaching map for some (n + l)-cell of (X,A). As noted in Remark 9.1, (??>)(?) is P-local; thus ipnfa represents an element in the lk- module ???((?-?){?)).
Homotopy Theory 109 We use this to extend ??/? to a cellular map ga : Spa —> (AV)(n)· Recall CO OO that S? = V S? Uh (UZi D]+1). Define ga : V ?>" —> (A»(n) so that pQ i=0 i=0 restricts to </?n/a in S? and restricts to a representative of (?[=? ^j)^/«*] in S", r > 1. (Here the kj are the integers prime to V used to define the telescope). Since h attaches ?>"+1 by a map representing [S"_j] — kr[S"}. gah is homotopically constant and so ga extends to all of S%>. Use the Cellular approximation theorem 1.2 to replace ga by a cellular map, homotopic rel SJ to the original. Set gn = Uaga. Define (A»(n+1) = (A»(n) USn (IIQ ?>?"?) and define ??+? by the formula above. This completes the construction of ? : (A'. A) —> (AV) A). To see that ? is a P-localization, it is enough to prove ?*(?\ k) is an isomorphism (Theorem 9.6). This follows at once from the construction and the fact that the standard inclusions induce isomorphisms ?? ( r\n + l en. il-i _5L ET ( n^+l Cn . n~\ ? Theorem 9.7 (i) For each simply connected space X there is a relative CW complex (?-?,?) with no zero-cells and no one-cells such that the inclusion ? : X —> X-p is a V-localizatioin. (ii) If (?-?,?) is as in (i) then any continuous map f from X to a simply connected V-local space ? extends to a map g : X-p —> Z. If g' : X-p —> ? extends f : X —> ? then any homotopy from f to f extends to a homotopy from g to g'. (iii) In particular, the V-localizations of (i) are unique up to homotopy equiv- alence rel X. proof: (i) Let ? : Y -> A' be a weak homotopy equivalence from a CW complex Y such that Yi = Yo = {pt}. Let j : Y -» Y-p be a cellular ^-localization as described in the Example above. Put X-p = XU-? (? ? /) ?,· YVi where we have identified (y,0) with 4>(y) and (y,l) with j(y). Then (X-p. X) is a relative CW complex with no zero- or one-cells. Since A is simply connected the Cellular approximation theorem implies that so is X-p. Next use excision and the fact that ?*(?\?) is an isomorphism to deduce H,(X-p, Y-p; Z) Si H.(X U^, Y x /, Y x {1}; Z) = 0. Thus i/.(A>,pi;Z) Si H.(Yp,pt;Z). Since Y-p is TMocal so is Xv.
HO 9 ?7—local and rational spaces Finally, since j is a "P-localization we may use excision to deduce that H.{Xv,X;k) = Hm(Yp,Y;Jk) = 0. Hence the inclusion ? : X -> X-p satisfies ??{?;??) is an isomorphism; i.e., ? is a ^-localization. (ii) Let (ZU/ X-p, Z) be the relative CW complex obtained by identify- ing ? ~ f(x),x € X- Since ? is V-local the construction of the Example above gives a ^-localization a:(ZL)fX-p,Z) -> ((Z Uf XV)V,Z). In particular, ?*(?,?) is an isomorphism. On the other hand, ?*(?\ Jk) is an isomorphism, and hence H*(ZL)f X-p, Z; Jk) S H*{Xv,X;k) = 0. Thus H*(Z,Jk) A H,(Z Uf ??;?) and the composite inclusion i : ? —)¦ (? U/ Xv)v also satisfies: Hm(i;]k) is an isomorphism. By the Whitehead-Serre theorem 8.6, 7r»(i) ® J: is an isomorphism. Since ? and (Z U/ Xp)p are 'P-local they satisfy ??»(—) = 7r»(—) (g! $:. Hence i is a weak homotopy equivalence. Now the Whitehead lifting lemma 1.5 can be applied to ? U/ Xv ) (Z U/ .Yp)p to obtain a retraction r : Z U/ Xp -> ?. Define ? to be the composite X-p -> ZU/IpA Z. Finally, suppose p' : Xp -> ? restricts to /' ·. X -> ? and ?: XxI-^Zsa homotopy from / to /'. Consider the map V1 = @,?,0')= (Xpx{0})UIx/U(IPx{l})->Z. An easy calculation like those above shows that the inclusion (X-p ? {0}) U ? ? lu (A> x {1}) -» ??? ? J is a 7^-localization. Hence ? extends to a homotopy from g to g'. ? (c) Rational homotopy type. We specialize now to the case Jk = Q and V = ?. Consider a continuous map /: X->Y between simply connected spaces. It follows from the Whitehead-Serre theorem 8.6 that the following conditions are equivalent: • vr»(/) ® Q is an isomorphism. • .#*(/; <Q>) is an isomorphism.
Homotopy Theory 111 • H*(f;Q) is an isomorphism. In this case / is called a rational homotopy equivalence . Theorem 9.8 implies that the rationalizations Xq of a simply connected space all have the same weak homotopy type and that the weak homotopy type of Xq depends only on the weak homotopy type of X. Definition The weak homotopy type of Xq is the rational homotopy type of X. Theorem 9.7 also gives a second, equivalent description of rational homotopy type that does not explicitly involve rationalization. Proposition 9.8 Simply connected spaces X and Y have the same rational homotopy type if and only if there is a chain of rational homotopy equivalences X <— Z@) —>···«— Z(k) —> Y. In particular if X and Y are CW complexes then they have the same rational homotopy type if and only if Xq ~ Yq. proof: The first assertion is immediate from Theorem 9.7. The second follows immediately from the Whitehead lifting lemma 1.5, because (Xq, X) and (Yq, Y) are relative CW complexes. ? Note: Rational homotopy theory is the study of properties of spaces and maps that depend only on rational homotopy type; i.e.. are invariant under rational homotopy equivalence. A rational cellular model for a simply connected space Y is a rational homotopy equivalence ? : X -> Y from a CW complex X such that ?? = Xo = {pt}- We complete this section by observing, for simply connected spaces Y, that • H*(Y;Q) has finite type <=> Y is rationally modelled (9.9) by a CW complex of finite type • H,(Y;Q) is finite dimensional <?> Y is rationally modelled (9.10) and concentrated in by a finite TV-dimensional degrees < ? CW complex Both these observations are corollaries of Theorem 9.11 Every simply connected space Y is rationally modelled by a CW complex X for which the differential in the integral cellular chain complex is identically zero.
112 9 "?—local and rational spaces Corollary Ht(X;Z) is a free graded Z-module. proof of 9.11: With the aid of cellular models (Theorem 1.4) we reduce to the case y is a CW complex, Yo = Y\ = {pi}, and all cells are attached by based maps (Sn,*) -> (Yn,pt). Let (C»,<9) denote the cellular chain complex for Y and use the same symbol to denote an ?-cell of Y and the corresponding basis element of Cn ¦ Choose ?-cells a" and 6" so that in the rational chain complex (C» <g! Q. d), Cn®Q = (kerd)n@0Qa? = (Im d)n ® 0 Qa? ? ? <Qb]. i i j Define subcomplexes W(n) C Z(n) C Yn by W(n) = Yn^U(\Jaf) and Z(n) = W(n)u(\Jb]). i j Since d : ©Qof —*¦ (Im d)n-i the Cellular chain models theorem 4.18 asserts that H,(W(n),Z(n - 1);Q) = 0. Thus the inclusion ? : ? (? - 1) -> W(n) gives an isomorphism of rational homology, and the Whitehead-Serre theorem 8.6 shows that ?, (?) <g> Q is an isomorphism. In particular, since the cells b^ are attached by maps /_,· : (S",*) -» Yn-\ C W(n), there are maps g^ : (Sn~1,*) -» (Z(n — 1), *) and non-zero integers rj so that 7??-?(?)[&] = rjlfj]. We now construct ? : X -> Y inductively so that ? restricts to rational homotopy equivalences ?? : Xn -> Z(n). Begin with X1 = Xo = {pt} and ?? : ?? -» Yi = {pi}. Suppose ??-? ¦ Xn-\ ->· ^(" — 1) is constructed. Then there are maps hj : (S",*) -> (Xn_1;pi) and non-zero integers Sj such that 7??_?(???_?)[^] = Sj[gj}. Set ? = {hj} : \fj S]1 -> Xn_i, and set Choose maps oj : (Sn~1,*) -> (S", *) such that [??] = s^r^S71'1] in *n-i(Sn-\*). Thus in ?^???^?),*) we have [t follows that fjOj is based homotopic to </?n_i/ij; i.e. there are maps ?, : S-1 x [|, 1] -> W(„), with j *["· ^ = ^
Homotopy Theory Then write Z(n) = W{n)Uf{\jD?), f = {/;} : \/S?-1 -> Yn-U 5 and extend ??-? to ?? by setting $j(v)t)eW(n), veS?-\ ?<?<1, Put X'n = Xn-i Uh (V-5" x [5,1])· Then {Xn,X'n) is a relative CW complex whose cells are the disks DJ(^) of radius ^. An immediate cellu- lar chains argument (Theorem 4.18 and the following Remark) shows that ?? induces a homology isomorphism H,(Xn,X'n;Q) A H+(Z(n), W(n);Q). But Xn-i is a deformation retract of X'n-i, an<^ tne maps ? : Z(n — 1) -> It7(n) and (/?n-i ^ -Xn-i —> Z(n— 1) induce isomorphisms of rational homology- Hence which gives ?.{??) ¦¦ H.{Xn,ty A ii.(Z(n);Q) as desired. Finally, observe that for any n. Hn(Z(n): Q) -=> Hn{Z{n),Z{n-1); Q). by an immediate cellular chains argument. Hence iin(Xn;Q) ^» iin(Xn,Xn_i;Q). If (Cx, O) denotes the integral cellular chain complex for X then this implies that (C<nld) ® Q -> (^;0) (S) Q yields an isomorphism of homology in degree n. This implies <9 : C% -> C^Lj is zero, again, for all ?. ? Exercises (All spaces are supposed 1-connected.) 1. Determine X(p) for ? = 2,3 or 0 and X = Sn ?? Dn+1 with ? : S" -> S" a continuous map of degree 6 (resp. 5). Same question with the fibre of the collapsing map X -> S"+1. 2. A finite complex X is universal if for any rational equivalence ? : A —> ? and continuous map f : X -> ? there exists a rational equivalence ? : X -> X and a continuous map g : X -> A such that ? ? g ~ / ? ?. Let X and F be finite complexes and / € [Xq, Yq\. Prove that if X is universal then there exists g 6 [X, Y] and a rational equivalence a : X -> X such that gq ~ / ? aQ. 3. Let Pbea set of prime numbers, V C V, and define the ring k' by ifc' = {??/?: | fc is relatively prime to the elements of V'}- Consider two localization maps / : X -> Xv and /' : Y -> Yv> with 7>' c V- Prove that if / : X -> F !S a continuous map then there exists a map j\^> : X-p —> Yp», unique up to homotopy. such that //_/< ? / = /' ? / and the following diagrams commute:
114 9 ?7—local and rational spaces Kn(X)®Ik' n"-^id Kn{Y)®Ik' Hn{X)®Ik' H"Md Hn{Y)®Ik' s*| |~ =1 l^ nn(Xv)®Ik> ?^>" ??(??.) Hn(Xv;Ik') "^ Hn(Yv.) 4. Let V be a set of prime numbers and the ring Ik be as defined at the beginning of this section. A continuous map / : X -> Y is a V-equivalence if for any ? > 2, iTn(X)®Bi *"-—> irn(Y)®Ik is an isomorphism. Prove that / is a ^-equivalence if and only if f-p is a homotopy equivalence. When Y is supposed 2-connected, prove that this is also equivalent to the following assertion : ?/ : CIX —> QY is a ^-equivalence. Remark: It is not true that the existence of a p-equivalence f : X —> Y for every prime ? implies that ? ~ ?. This motivates the introduction of the genus of a space X, i.e. the set of [Y] such that for any prime p, Yp ~ Xp. 5. Let / : X -> Y be a rational homotopy equivalence between finite CW complexes and V be the set of all primes. Prove that there exist p\ ,??,...,?^ € V such that Xp ~ Yp where ? = V — {p\,pi, ¦¦¦•.Pk)-
Part II Sullivan Models
10 Commutative cochain algebras for spaces and simplicial sets In this section the ground ring is a field k of characteristic zero. Recall from §5 that C*(X; k) denotes the cochain algebra of normalized singu- lar cochains on a topological space X. This algebra is almost never commutative, although it is homotopy commutative. Now however, because k is a field of characteristic zero, it turns out that we may replace C*(X; k) by a genuinely commutative cochain algebra. More pre- cisely, we introduce a naturally defined commutative cochain algebra Apl (X; k), and natural cochain algebra quasi-isomorphisms C*(X;k) ^ D(X) <?- APL(X;k), where D(X) is a third natural cochain algebra. The construction of ApL(X',k) , due to Sullivan [144], is inspired from C°° differential forms, while reflecting the combinatorial nature of how the singular simplices of X fit together. The functor X^ APL(X;k) serves as the fundamental bridge which we use to transfer problems from topology to algebra. It is important to note that this functor is contravariant; i.e., it reverses arrows. The quasi-isomorphisms above will be constructed in Theorem 10-9. They define a natural isomorphism of graded algebras, H*(X;k) = H(APL(X;k)), A0.1) and we shall always identify these two algebras via this particular isomorphism. Thus for any continuous map, f, we identify H*(f;k) = H(ApL(f;k)). Recall now from Proposition 9.8 that two simply connected spaces X and Y have the same rational homotopy type if and only if there is a chain of rational homotopy equivalences X <— Z@) —>···¦«— Z(k) —> Y. There is an analogous notion for commutative cochain algebras: two commuta- tive cochain algebras (.4, d) and (B, d) are weakly equivalent if they are connected by a chain (A,d) ^ (C@),d) ^ ··. -?> (C(k),d) ?- (B,d) of quasi-isomorphisms of commutative cochain algebras. Such a chain is called a weak equivalence between (A,d) and (B,d). Next recall from §9(c) that a continuous map f : X -> Y between simply connected spaces is a rational equivalence if and only if H*(f;Q) is an isomor- phism. Since k is a field of characteristic zero and we identify H(ApL(f;k))
116 10 Commutative cochain algebras for spaces and simplicial sets with H*(f;]k) it follows that f is a rational equivalence if and only if ApLJ(f;Jk) is a quasi-isomorphism. In particular if X and Y have the same rational homotopy type then Api (X; Jk) and Api(Y;lk) are weakly equivalent, and so Apl(—;Jk) defines a map rational homotopy types ? f weak equivalence classes of ? j [ commutative cochain algebras. J In §17 we show that if h = Q and if we restrict to simply connected spaces X and commutative cochain algebras (A,d) such that H*(X:Q) and H(A,d) have finite type, then this correspondence is a bijection. This fundamental result of Quillen-Sullivan reduces rational homotopy theory to the study of commutative cochain algebras. As noted in the Introduction, our purpose in this text is to show how this result may be applied to extract topological information from the algebra. Since homotopy information is carried by the weak equivalence class of ApL(X;Jk) we may (and frequently will) replace it by a simpler, weakly equiv- alent, commutative cochain algebra: such a cochain algebra will be called a commutative model for X. More formally we make the Definition: A commutative cochain algebra model for a space X (or simply a commutative model for X) is a commutative cochain algebra (A,d) together with a weak equivalence Henceforth we shall fix is and, for the sake of simplicity suppress Is from the notation, writing Api(-) for Api(-;Js) and C*(-) for C*{-;h). The rest of this section is spent on constructing APi{X) and establishing its relation to C*{X). The section covers six topics: (a) Simplicial sets and simplicial cochain algebras. (b) The construction A(K). (c) The simplicial commutative cochain algebra ApL, and (d) The simplicial cochain algebra Cpl, and the main theorem. (e) Integration and the de Rham theorem. (a) Simplicial sets and simplicial cochain algebras. A simplicial object ? with values in a category C is a sequence {Kn}n>o objects in C, together with C-morphisms * ?. D< i < ? + 1 and ss : Kn -> Kn+1, 0 < j < n.
Sullivan Models 117 satisfying the identities , ? , i , i , i <;; <i; = Ji >; + + 1 1. id A simplicial morphism f : L —> ? between two such simplicial objects is a sequence of C-morphisms ?? : Ln -> Kn commuting with the 9* and Sj. A simplicial set ? is a simplicial object in the category of sets. Thus it consists of a sequence of sets {Kn}n>o together with set maps di,Sj satisfying the relations A0.2). The motivating example is; of course, the simplicial set S*(X) = {Sn(X)}n>o of singular simplices ? : ?" —> X on a topological space A', with face and degeneracy maps are defined in §4(a). This is a functor, with a continuous map / inducing the morphism S*(f) : ? \—> f ? ?, ? € Sn(X). A simplicial cochain algebra A is a simplicial object in the category of cochain algebras: it consists of a sequence of cochain algebras {An}n>o with appro- priate face and degeneracy morphisms. Similarly simplicial cochain complexes, simplicial vector spaces ... are simplicial objects in the category of cochain com- .plexes, vector spaces, Thus a morphism of simplicial cochain complexes, ? : D -* E, is a sequence ?? : Dn -> En of morphisms of cochain complexes, compatible with the face and degeneracy maps. If each ?? is a quasi-isomorphism we call ? a quasi-isomorphism of simplicial cochain algebras. Let ? be any simplicial set. By analogy with St(X) the elements ? € Kn are called n-simplices and ? is non-degenerate if it is not of the form ? = sjt for some j and some r € Kn^\. The subset of non-degenerate n-simplices is denoted by NKn. It follows from the relations A0.2) that for each ?· > 0 a subsimplicial set K(k) C-K is given as follows: for ? < k, K(k)n = Kn and for ? > k, K(k)n = {sjt | 0 < j < ? - 1, r€ /f (fc)n-i}· The simplicial set K(k) is called the k-skeleton of K, and is characterized by ??? , n<k ? \nlk. The dimension of ? is the greatest k (or oo) such that NKk ? ?, and this is also the least k such that K(k) = K. Finally, we use the notation of §4(a) to introduce a sequence of important sub- simplicial sets ?[?] C St (?"), ? > 0. Let e0,... , en be the vertices of ?". Then ^¦[n]k c Sk(An) consists of the linear fc-simplices of the form ? = {eio · · ¦ eik) with 0 < i0 < · · · < ik < n. Thus ? is non-degenerate if and only if i0 < · · - < i*. In particular A[k] is fc-dimensional and the identity map cn = {e0 ¦ ¦ ¦ e„> : ?" -> ?" •s the unique non-degenerate ?-simplex; cn is called the fundamental simplex of
118 10 Commutative cochain algebras for spaces and simplicial sets ?[?]. The (n— 1)-skeleton of ?[?] is denoted by dA[n] and is called the boundary of ?[?]. Lemma 10.3 //if is any simplicial set then any ? € ??? determines a unique simplicial set map ?» : ? [?] -> ? such that a»(cn) = ?. proof: The verification using A0.2) is straightforward, but the reader may also refer to [122]. ? A simplicial object A in a category C is called extendable if for any ? > 1 and any 1 C {0,...,n}, the following condition holds: given ?* € An_i,i € 2, and satisfying 9???· = 97·_??? ,i<j, there exists an element ? e An such that ?* = oj-?, i € 2. (b) The construction A(K). Let J\~ be a simplicial set, and let A = {An}n>0 be a simplicial cochain complex or a simplicial cochain algebra. Then A(K) = {Ap(K)}p>0 is the "ordinary" cochain complex (or cochain algebra) defined as follows: • AP(K) is the set of simplicial set morphisms from ? to Ap. Thus an element ? G AP(K) is a mapping that assigns to each n-simplex ? e Kn (n > 0) an element ?? € (Ap)n, such that ??;? = ???? and • Addition, scalar multiplication and the differential are given by (? + ?)? = ?? + ??, (?·?)? = ?·?? and • If A is a simplicial cochain algebra multiplication in A(K) is given by (?·?)? = ??·??. • If ? : ? -> L is a morphism of simplicial sets then A(ip) : A(K) < A(L) is the morphism of cochain complexes (or cochain algebras) defined by (?(?)?)? = ???. • If ? : A -> ? is a morphism of simplicial cochain complexes (or simplicial cochain algebras then ?{?) : A(K) -> B(K) is the morphism defined by (?(?)?)? = ?{??).
Sullivan Models 119 • When X is a topological space we write A(X) for A(S*(X)). • If ? € Kn then ? ?-»· ??.defines a morphism A(K) —> An called restriction to ?. Remark Note that the construction A(K) is functorial in A and contrafunc- torial in K. Proposition 10.4 Let A be a simplicial cochain algebra. (i) For n> 0 an isomorphism A(A[n\) —=1—» An of cochain algebras is given by ? ?—> $c„> where cn is the fundamental simplex of ?[?]. (ii) If A is extendable and L C ? is an inclusion of simplicial sets, then A(K) -> A(L) is surjective. proof: (i) The definitions show that ? ?—> ??? is a morphism of cochain algebras. Since ?.?(?[?]) consists of the simplicial set maps ?[?] -> Ap, Lemma 10.3 asserts that this morphism is a bijection. (ii) Given ? € A(L) we give an inductive procedure for constructing ? € A(K) that restricts to ?. In fact, suppose we have found elements ??, ? € Kk, k < ? such that ?? = ??, ? ? Lk; ?9:? = ?{??; <eSjT = Sj-?,-, r e Km, m < ? - 1. Then define ??,? € ?? as follows. If ? € Ln, ?? = ??. If ? has the form Sjt, ? e Kn-i, use A0.2) to see that 5_;?? is independent of the choice of j and r and put ?? = Sj$r. Finally, if ? € Kn — Ln is non-degenerate note that di($QjC) = dj-i($dia)A < j- Since A is extendable we may therefore choose ?? so that ?*?? = ??,.?, 0 < i < ?. In all cases the equations above will be satisfied, as follows easily from A0.2). D If A is an extendable simplicial cochain algebra and L C ? is an inclusion of simplicial sets we denote by A(K, L) the kernel of the surjective morphism A{K) -> A(L). Thus A(K, L) is a differential ideal in A(K) and 0 -> A(K,L) -> A{K) -> A(L) -> 0 is a short exact sequence of cochain complexes, natural with respect to (K, L) and with respect to the simplicial cochain algebra A. A key step in the proof of Theorem 10.9 is the following Proposition 10.5 Suppose ? : D -> ? is a morphism of simplicial cochain complexes. Assume that
120 10 Commutative cochain algebras for spaces and simplicial sets (i) H{?n) : H(Dn) -> H (En) is an isomorphism, ? > 0. (ii) D and ? are extendable. Then for all simplicial sets K, ?(?(?)) : H(D(K)) -> H(E(K)) is an isomorphism. First we establish a preliminary lemma. Define a:A(K(n),K(n~l)) > JJ A(A[n],A[n - 1]) ,n>0, by setting a : ? ?-)· {^4(?,)?}?6^//<·?, where ?» : ? [?] -+ ? is the unique simplicial map (Lemma 10.3) such that a»(cn) = ?. Lemma 10.6 If ? is a simplicial set and A is an extendable simplicial cochain complex then a is an isomorphism, natural in A and in K. proof: If ?(?) = 0 then ? vanishes on all the non-degenerate simplices of ? (?) and hence ? = 0. Conversely, given a family {?? € ?(?[?],?[? — 1])}?&???, we recall first that (Proposition 10.4 (i)) A(A[n],A[n-l])cA(A[n}) = An. This identifies the ?? as elements of An satisfying 9??? = 0, 0 < i < n. Define ? € A(K(n),K(n — 1)) by the three conditions: {?? =0 , ? G Km, m < ? ?? =?? , ????? ? s ja = Sj-?? , all i, ?. Clearly ?? = {??}. D proof of 10.5: For any inclusion L C M of simplicial sets we have the row exact (Proposition 10.4 (ii)) commutative diagram 0 D(M,L) — D(M) D(L) >- 0 9(M,L) ?{?) e(L) E(M, L) — E(M) E(L)
Sullivan Models 121 Hence, if any two of the vertical arrows is a quasi-isomorphism, so is the third. Moreover Proposition 10.4 (i), identifies 0(?[?]) : ?>(?[?]) -> E(A[n\) with ?? : Dn -> ?1,,. Hence ?(?[?\) is a quasi-isomorphism, ? > 0. We now use induction on ? to show that for any simplicial set ?, ?(?(?)) is a quasi-isomorphism. Indeed, this is vacuous for ? — —1. Suppose it holds for some ? - 1. From Lemma 10.6 and the remarks above it follows that ?{—) is a quasi-isomorphism when (—) is in turn given by: dA[n], (?[?], dA[n}), (K(n), K(n - 1)) and K{n). This completes the inductive step. Along the way we have also shown that each ?(?(?),?(? — 1)) is a quasi-isomorphism. Finally, let ? be any simplicial set. Given ? e D(K) and ? € E(K) such that ?? = 0 and 0(??)? = d*, we shall find ? e D(K) and ? € E(K) such that ? = du and ? = ?(?)? + dF. This implies at once that ?(?) is a quasi- isomorphism (Lemma 3.2). To find ? and ? we construct inductively a sequence ?? e D (K, K(n — 1)) and ?? e ? (?, ? (? - 1)) such that and ? - ? (e(K)iii + dTi) ? ? (?, ? (ri)). Indeed, set ?_? = ?_? = 0. If ?; and I\- are constructed for i < n, set ?' = ? - ? dn{ and ?' = ? - ? (?(?)??? + dTi). Then ?' restricts to ?" € D(K(n),K(n-l)), ? restricts^ ?" € ?(?(?),?(?-1)), d$" = 0 and ?(?)?" = d*". Our remarks above show that ?(?(?),?(? - 1)) is a quasi-isomorphism. Hence we can find ?" € D (K(n),K(n — 1)) and ?" € E(K(n),K(n - 1)) such that ?" = du" and ?" - ?{?)?." + ??". Now close the induction by using the extendability of D and ? to find ?? € D (K, K(n — 1)) and ?? 6 E(K,K(n - 1)) which restrict respectively to ?" and ?". Finally, the sequence {??} satisfies (??)? = 0, ? > dim ?. Thus we may define ? e D(K) and ? € E(K) by ?? = ?(??)? and ?? = ?(??)?. Clearly ? ? ? = dU and ? = 0(??)? + dr. D (c) The simplicial commutative cochain algebra Apl, and Apl(X). The first step in constructing a functor from topological spaces to commuta- tive cochain algebras is the construction of the simplicial commutative cochain algebra, APL. For this consider the free graded commutative algebra ?(?0, · · · , tn, y0,... , yn) ~"cf· §3, Example 6.- in which the basis elements U have degree zero and the
122 10 Commutative cochain algebras for spaces and simplicial sets basis elements yj have degree 1. Thus this algebra is the tensor product of the polynomial algebra in the variables U with the exterior algebra in the variables yj. A unique derivation in this algebra is specified by U t-> yt and yj t-> 0. It preserves the ideal /„ generated by the two elements ]T? U - 1 and J2o Vj- Now define APL = {(APL)n}n>0 by: • The cochain algebra (APL)n is given by . _ A(t0,... ,tn,y0,... ,yn) {APL)n - (sti_ijEyi) ' dti = yi and dyj = 0. • The face and degeneracy morphisms are the unique cochain algebra mor- phisms di : (APL)n+i -> (APL)n and Sj : {APL)n -> (APL)n+i satisfying {tk ,k<i ( tk ,k <j 0 ,k = i and sj : tk ?—> < tk + tk+i ,k = j [ Notice that the inclusions U -> (APL)n,yj -> (APl)ti extend to an isomorphism of cochain algebras, (?(*?,... ,tn,yu... ,yn),d) A (APL)n. Remark 1 The elements of (APl)ti are called polynomial differential forms with coefficients in h, for the following reason. When Ac! (e.g. Jk = Q) the algebra (APL)°n = ?[??,...,??] /(Eif-1) is the Jk- (eg. rational) subalgebra of the smooth functions C°°(A") generated by the restrictions U of the coordinate functions of R"+1. This identifies {APl)ti as a sub-cochain algebra of the classical cochain algebra, ????(??) of real C°° differential forms on ?"; moreover, Adr(^") = C°°(An) ®(?? )? (APL)n. Note that this also identifies di = ADR(Xi) and Sj = Adr(qj), where the Aj are the face inclusions and the ?3- are the degeneracies for the simplices ?" as defined in §4(a). These ideas will be further developed in §11. Remark 2 Each APPL, p > 0, is itself a simplicial vector space and APl = {-4pL} >0, equipped with the obvious simplicial multiplication and differential.
Sullivan Models 123 Lemma 10.7 (i) (APL)o = k-\. (n) H ((APL)n) = ?c ¦ 1, n>0. (Hi) Each APPL is extendable. proof: (i) This is immediate from the definition. (ii) The isomorphism above identifies (ApL)n as the tensor product ® A(if)yt). Since d(i*) =kt*-1yi if follows that H(A(ti,yi)) = *· 1. But #(-) commutes with tensor products (Proposition 3.3) and so H((Apl)ti) = ic · 1 as well. (in) Suppose given X C {0,... ,n} and elements ?* € (^???,)?-?, ? € I, satisfying ^?^- = dj-\$i,i < j. Beginning with ?_? = 0, we induc- tively construct elements *r € (Apl)ti, —1 < ? < n, such that: oj*r = ?{, Hi el and ? < r. We may suppose ??_? constructed. If r ^ 1 set ?? = ??_?; otherwise proceed as follows. Embed the polynomial algebra {A0PL)n in its field of fractions F and let B° C F be the subalgebra generated by (A°PL)n and the element ?^-. Setting d ( y^- ) = ,^^? defines a cochain algebra ? = B°®{AopL)n(ApL)n=B°®A(dtu...,dtn) which contains (/lpz,)n· Moreover a morphism ? : {ApL)n-\ -> ? ?? cochain algebras is given by and ?(???) = ap{ti). On the other hand, we may extend dr to a morphism ? -> (.4??,)?_? by setting dr (ji-) = 1. Clearly dr ? ? = id. Every element of ? has the form -: -^? for some ? > 0 and some A ir) * e {ApL,)n- In particular we may write A - ??)??(?? - ar*r_0 = ?, some ? € (APi)n. Now, for i 6 X and i < r, we have 0;(?? - ar*r_0 = 3?_?(?{ - af*r_!) = 0.
124 10 Commutative co chain algebras for spaces and simplicial sets Hence ?? — <9???_? is in the ideal generated by U and dti in (/lpz,)n-i, and it follows that ? is in the ideal generated by U and dU in (ApL)n. In particular, <9;? = 0, i e 1 and i < r. On the other hand, since ??? = id, we also have 9?? = ?? - <9???_?. Thus 9;(? + ??-?) = $j for ; € X and j < r. This closes the induction. D Because the graded algebras (???,)? are commutative, the construction Apl( ) defined in (b) assigns to each simplicial set ? a commutative cochain algebra, Apl(K), and to each map / of simplicial sets a morphism, APl(/) of commu- tative cochain algebras. Since Apl is extendable, with every pair L C ? of simplicial sets is associated the short exact sequence 0 —> APL(K,L) —> APL(K) —> >ipL(L) —> 0. For topological spaces X and continuous maps / we apply this construction to the simplicial set S»(X) and to ?¦(/). This defines a contravariant functor from spaces to commutative cochain algebras, which will be denoted X~+APL{X) and / - APL(f). In particular, associated with a subspace Y is the short exact sequence 0 -> APL(X, Y) -> .4PL(.Y) -> APL(K) -> 0. When it is necessary to indicate the coefficient field explicitly we shall write APL(X]Js), APL(K;Je) and APL(f;k) for APL(X),APL(K) and APL(f). An element of APL{X) is a function assigning to each singular ?-simplex of X a polynomial p-form on ?", ? > 0, compatible with the face and degeneracy maps. This motivates the following terminology: The commutative cochain algebra, APl(X), is the cochain algebra of polynomial differential forms on X with coefficients in k. Example 1 When X = {pi}, then S*(X) = ?[0]. It follows that APL{X) = (APLH = Je-t APL(pt) = M. Thus an inclusion j : pt —> Y induces an augmentation ? = APl (j) : APl (Y) —> A. (d) The simplicial cochain algebra Cpl, and the main theorem- Recall from §5 that with every topological space X is associated to its singular cochain algebra, C*(X; 3c). Recall that we simplify notation and write C* (X) for C*(X; &). Now the construction oiC'(X) depends only on the singular simplices of X and their face and degeneracy maps; i.e., it depends only on the simplicial set St(X). As such it generalizes to any simplicial set ? to give the cochain algebra C*{K), also written C*(K; k) if we need to emphasize coefficients. More
Sullivan Models 125 • C*(K) = {Cp(K)}p>0\ • CP(K) consists of the set maps Kp —> fc vanishing on degenerate simplices. • For / ? CP(K), g ? Cq(K) the product is given by (/ U g){a) = (-l)p"f(dp+1 ¦ ¦ ¦ dp+qa) · g(dodo ¦ ¦ ¦ doa), ? G Kp+q. • The differential, <2, is given by P+l (df)(a) = ?(-1)?+?+1/(?4?) ,? G Kp+I, f G CP{K). i=0 Observe that, by definition, C*(X) — C*(S*(X)) for topological spaces X. Next observe that a simplicial cochain algebra Cpl can be defined using the simplicial sets ?[?] C 5.(?") introduced in (a). Indeed since the face inclusions and degeneracy maps for the ?" have the form Xi = (eo,---ei---en+i):An^An+1 and Pi = <e0,··· ,eheh-·· ,en) : An+1 -> ??, (cf. §4(a)) it follows that 5.(X{) and S*(pj) restrict to simplicial maps [Xi] : ?[?] -> ?[? + 1] and [Pj] : ?[? + 1] -> ?[?]. Thus we define Cpl — {(CpL)n}n>o by • (Cpl)? is the cochain algebra C*(A[n]). • The face and degeneracy morphisms are the C*([Xi]) and C*([pj]). Finally, let ® ApL = {(CpL)n ® (ApL)n\ 9j <8> <%,· Sj ® Sj } be the tensor product simplicial cochain algebra (where (Cpl)u ® (Apl)u is the tensor product cochain algebra described in §3(c)). Morphisms Cpl —> Cpl ® Apl i— Apl are defined by 7 t-> 7 0 1 and ?? 1 0 ?. Thus, for any simplicial set ? they determine the natural cochain algebra morphisms CPL(K) -> (CPL ®APL)(K) ^~ ApL(K). Our main result, which now follows, is (together with its proof) due to Chris Watkiss [155] following the idea in Weil's proof [156] that de Rham cohomology and singular cohomology are isomorphic.
126 10 Commutative cochain algebras for spaces and simplicial sets Theorem 10.9 [155] Let ? be a simplicial set. Then (i) There is a natural isomorphism Cpi(K) -^ C*(K) of cochain algebras. (ii) The natural morphisms of cochain algebras, Cpl(K) -> {CPL®APL){K) <- APL(K) are quasi-isomorphisms. Substituting S*(X) = ? in Theorem 10.9 we obtain Corollary 10.10 For topological spaces X there are natural quasi-isomorphisms of cochain algebras. C*(X) —?-> (CPL®ApL){X)^^—ApL{X) D This gives the isomorphisms H*(X) = H(APL(X)) promised in A0.1). For the proof of Theorem 10.9 we require two lemmas. Lemma 10.11 There are natural isomorphisms Cpl{K) —> C*(K). proof: Each 7 G CPL(K), ? > 0 determines the element / G CP(K) given by /(?) = 7<r(cp) , ? G Kp, where cp is the fundamental simplex of ?[?]. It follows, by a straightforward calculation from the definitions and A0.2), that the correspondence 7 1—> f is a cochain algebra morphism. To show it is injective, assume 7 1—> 0. The elements a ? ?[?]? are the linear simplices of the form {eio,... ,e,· ) : ?? -> ?" with i0 < ... < ip, and these can all be written as composites of face and degeneracy maps. It follows that for any ? > 0 and any r G Kn, Gr)(a) = Gr)(or O Cp) = Jroa(Cp) = 0. Hence 7 = 0 and our morphism is injective. On the other hand, by Lemma 10.3 any ? e Kn determines a unique simplicial map ?. : ? [?] -> ? such that ?.(??) = ?. Thus, if / G CP(K), we may define 7 € CPL(K) by ?? = ??(?»)(/). Clearly 7 ?-> / and our morphism is surjective. D Lemma 10.12 (i) H(CPL)n = A = H((CPL)n 0 (APL)n), ? > 0. (ii) CpL is extendable. ^ ?. .?^,? ;R extendable.
Sullivan Models 127 -proof: (i) The first assertion is the classical calculation of ?* (?[?]) which is left to the reader; the second then follows from Lemma 10.7 (ii) and the fact that H( —) commutes with tensor products (§3(e)). (ii) The map [?*] : ? ?-> ?* ? ? is an isomorphism of the simplicial set ?[? — 1] onto a subsimplicial set F(i) C ?[?]. Thus it identifies a p-cochain fi G Cp(A[n - 1]) with a set map // : F(i)p -> Je. Given a sequence /f,i G X C {0, ...,n} of such cochains, the condition difj — dj-ifi, i < j, implies that f[ and /j restrict to the same function in F(i)p ? F(j)p. Thus {/?} defines a function from (J F(i)p to Ik, which then trivially extends to a function / in ?[?]?, vanishing on degenerate simplices. Thus / is an element in Cp(A[n]) such that dif = fi,i el (iii) Suppose given elements ?* G <??(?[? - 1]) 0 (^L)n_1} ielc {0,... ,n}, satisfying ???^ = ?,-??;, i < j. Write ?* = ? fia ® ??? and define ?;·(?, ? r) = (-l)w ^ /??(?)???> ???[?- l]p. Now for each ? G ?[?]? let ?? ? I be the set of indices i such that ? ? F(i)p. It is immediate from the definition that the elements ?^·(?) G (ApL)n-i satisfy Since ???, is extendable (Lemma 10.7 (iii)) we may find ?? G (ApL)n such that ?,·?? = ?;.(?), i G ??. Finally, identify <7?(?[?]H(^??/)? with the set of functions ?[?]? ->· (^PL)n, just as above, and define ? G Cp(A[n]) 0 (-4pL)n by ? (?) = ?? ,?€?[?]?. It is then immediate that (di 0 9?)? = ??, as desired. ? proof of Theorem 10.9: The first assertion is Lemma 10.11. The second follows by applying Proposition 10.5 to the morphisms 0c ¦¦ Cpl —> CPL 0 Apl and ?? : APL —> Cpl 0 .4pL given by 7 h-> 70I and ? ?-> 10?. Indeed all three simplicial cochain algebras are extendable (Lemma 10.7 and Lemma 10.12). Moreover, again by the same two lemmas, for each ? > 0, H((CPL)n) = H((APL)n) = H((CpL®APL)n) = h ¦ 1. This trivially implies that ?((??)?) and ?((??)?) are isomorphisms. Thus the hypotheses of A0.5) are indeed satisfied. D Finally, let L c ? be a sub simplicial set. Then the quasi-isomorphisms of Theorem 10.9 restrict to quasi-isomorphism C'(K,L) -^ (CpL®APL)(K,L) ^ APL(K,L).
128 10 Commutative cochain algebras for spaces and simplicial sets Thus the two short exact sequences 0 _> C*(K,L) —> C*(K) —> C*(L) —> 0 and 0 -» APL(tf,L) —> APL{K) —> yipi(L) -* 0 are connected by a chain (of length two) of quasi-isomorphisms. In particular the canonical isomorphisms of cohomology define an isomorphism of long exact cohomology sequences, Of course ? and L may be replaced here by a topological space X and a subspace A. (e) Integration and the de Rham theorem. If ? is a simplicial set then Apl(K) is a commutative cochain algebra and C*(K) is not, which explains why in Theorem 10.9 we need to connect them by the chain of quasi-isomorphisms C*(K) -=> · «=- Apl(K). In this topic we show how 'integration' provides a natural direct quasi-isomorphism of cochain complexes The quasi-isomorphism § was first constructed in the 1930's in the context of smooth manifolds and ordinary differential forms. In that setting the fact that § commutes with the differentials is precisely Stokes' theorem: ?? = ?, •/? Jqa and the fact .that H{§) is an isomorphism was conjectured by E. Cartan and proved by de Rham [43]. Now recall the constructions in §10(c) and (d) of the simplicial cochain algebras Apl and Cpl- We shall construct a quasi-isomorphism of simplicial cochain com- plexes, § : Apl —> Cpl (as defined in §10(a)) and set §K = §{K) : Apl(K) —>· Cpl(K) = C*(K). There are a number of calculations (mostly omitted) which are usually easy consequences of the classical integration by parts formula, ¦<???4) To define § : Api —> Cpl we need to define a sequence of cochain complex morphisms " " ? -»<rnr)„ = C*(A[n]) , n>0,
Sullivan Models 129 compatible with the face and degeneracy maps. Recall that (-4pz,)n = ?(??...., tn, dtu...,dtn). Define a linear map Jn : {AplI^ —> -fe by setting f / ? ? ? \ ?1 ?1'*1 ? ( ? ?1 ?,2 · · · i„" dt ? ? · · · ? dtn 1 = / dti dt-2 ¦ ¦ ¦ ? \ ' Jo Jo ? ] \^---tk^dtn Jo Now (cf. §10(a)) let ? = (eio ¦ --eik) be a fc-simplex of ?"; i.e. 0 < i0 < h < ¦ ' ' < h < ?. Then ? determines the morphism ?* : (Apl)ti —> (-'Ipl)/: given by Us >-> tj,l <j <k, and tm ·-> 0 if ?? ^ iOl ·.., ?·· Define fn : (APL)n —> (Cpi)n by setting (Note that if ? is degenerate then ?*? = 0 and so [§n ?) (?) = 0, as required.) Theorem 10.15 (i) ? = {<fn} : Apl —> Cpl is a quasi-isomorphism of simplicial cochain complexes. (ii) For each simplicial set ? and topological space X the linear maps §K = §{K) and §? — §{X) are natural quasi-isomorphisms §K:APL{K)-^C*{K) and §? : APL(X) -> C*(X) of cochain complexes. proof: A quasi-isomorphism of extendable simplicial cochain complexes in- duces a quasi-isomorphism of cochain complexes when applied to any simplicial set (Proposition 10.5). Since C*(—) = Cpl(-) and since Apl and Cpl are ex- tendable (Lemmas 10.11, 10.7 and 10.12) the assertion (ii) of the theorem follows from assertion (i). To prove (i) we show first that each § commutes with the differentials, which is essentially Stokes1 theorem. For this it is sufficient to verify that ]T(-ir+i+1 / ??? , ? G (APLrn-1 , ,·_? Jn-1
130 10 Commutative cochain algebras for spaces and simplicial sets where ?{ = (e0 ¦ · -ej-iej+i ¦¦¦en) : ??"^—* ?". By linearity it is enough to consider ? of the form i* ' · · · i*ndtx ? · · · dtj ¦ ¦ ¦ ? dtn (dtj is deleted). But in this case the formula is a straightforward calculation via A0.14). Next the compatibility of the §n with the face and degeneracy morphisms follows from the equation {§rAPL{a)$) (r) = (-1)^ [?*?*? = (-1)!?^ [(??)*? J k Jk valid for all simplicial maps Ak -^ ?? -^» ?" and all ? G (-4pl)? . Finally, to show that the §n are quasi-isomorphisms notice that jfn(l) = 1 and recall that H ((APL)n) = Jk = H ((CPL)n), ? > 0 (Lemmas 10.7 and 10.12). D Remark Multiplication defines a morphism multn : (CPL)n 0 (CPL)n —> (CPL)n of cochain complexes (but not of algebras), and so mult = {multn} is a morphism of simplicial cochain complexes. Thus we have the commutative diagram Cpl \ _ / ,? = mult ? (id <g><f) , of simplicial cochain complexes, which translates to the obvious diagram when applied to any simplicial set or topological space. D Exercises 1. Prove that Apl(*) = Ik. 2. Give an explicit formula for each <9,· : (APL)i -> (APLH = Ik as defined in (c). 3. Let ? be a simplicial set, and A, B simplicial cochain algebras. Prove that if H(A) = Ik = H(B) and if A and ? are extendable then there exists a sequence of quasi-isomorphisms A(K) -+ A(K) 0 B(K) <- B(K). 4. Prove that A = {(An,dn)}n>0 ,9; ,sj is a simplicial algebra where An = ?(??,??,...,??) 0A(?/o,2/i,..-,2/n) with dx{ = y{ and ii k <i ( xk if k < i if ?; = 2 and Sj(xfc) = < xkXk+i ii k = i if k >i I xk+l if k > i
11 Smooth Differential Forms In this section the ground ring is M. The construction Api{—;lk) of polynomial differential forms in §10 was sug- gested by the classical cochain algebra Adr(M) of smooth differential forms on a smooth manifold M. In this section we review the construction of Adr{M) and establish a chain of quasi-isomorphisms of commutative cochain algebras. This implies (§12) that Adr(M) and Apl(M; M have the same minimal Sullivan algebras and hence that many rational homo- topy invariants (e.g. dim7rfc(M) ®Q, k > 2 and the rational LS category of ?/) can be computed directly from Adr(M). This section is organized into the following topics: (a) Smooth manifolds. (b) Smooth differential forms. (c) Smooth singular simplices. (d) The weak equivalence Adr(M) S APl(M;R). (a) Smooth manifolds. A topological ?-manifold M is a second countable metrizeable topological space in which each point has an open neighbourhood homeomorphic to an open subset of M.n. Recall that a map from an open subset of Mn to W1 is infinitely differentiable or smooth if all its partial derivatives of all orders exist. A smooth atlas (Ua,ua)aex for M is a covering of M by open sets Ua together with homeomorphisms ua of Ua onto an open subset of W1, and such that for each ?,? : UaU?1 : U?{Ua ? Up) -^ ua(Ua ? U?) is smooth. Each (Ua,ua) is called a chart in the atlas. A smooth ?-manifold is a topological ?-manifold together with a smooth atlas. Example 1 W1. The identity map of W1 identifies W1 as a smooth manifold. D Example 2 Open subsets. Suppose (Ua,ua)a?Z is a smooth atlas for a smooth manifold M. If ? is an open subset of M then (ua ? ?, ua L n) is a smooth atlas for 0, and thus identities ? as a smooth manifold. D
132 11 Smooth Differential Forms Example 3 Products. If (Ua,ua)a?j and (Vp,V?)?ej are respectively smooth atlases for a smooth ?-manifold M and a smooth fc-manifold ? then (Ua x V,3, ,ua ? ? ?) identifies ? ? ? as a smooth n + k manifold. ? Suppose (Ua,Ua)a?i and {Ve,V?)?&j are respectively smooth atlases for a smooth ?-manifold ?/ and a smooth fc-manifold TV. A smooth map f : M —> ? is a continuous map such that each V?fu^1 : uQ (Ua Of (V?)) —> ve(V?) is smooth. The composite of smooth maps is smooth, and a diffeomorphism is a smooth map admitting a smooth inverse. Finally a smooth function is a smooth map M —> M and pointwise addition and multiplication makes these into a commutative algebra, C°°(M). For example, the smooth functions ula : Ua —> M defined by uQ = (u^,... .u^) are called local coordinates in M. As another example, if ? G <9open c M then there is always a smooth function f : M —> [0,1] which vanishes outside O. and is identically 1 in a neighbourhood of a; [69, Prop. VIII, sec.1.8]. Such a function is called a localizing function for ? G O. Two smooth maps f.g : M —> ? are smoothly homotopic if there is a smooth map F : ? ? ? —> ? such that F(-, 0) = / and F(-, 1) = g. For example, the identity map of Kn is smoothly homotopic to the constant map Rn —^ {0} [69, Example 1, sec. 1.10]. (b) Smooth differential forms. Let M be a smooth ?-manifold with smooth atlas {(Ua,ua)}. Every smooth map u : M —> ? determines the morphism C°°(u) : C°°(M) <— C°°(N) given Definition The tangent space TXM at ? ? M consists of the linear maps ? : C°°(M) —> M such that ?(fg) = ^fg{x) + /(?)??. If u : M —> ? is smooth then Txu : TXM —> TUXN is the linear map given by (?????(/) = ? (C°°(u)/). If ? is an open subset of W1 then a canonical isomorphism Rn -^U TXO is given by h ?—? ?/?, where = lim^ [69, Prop. 1, sec. 3.3]. This isomorphism maps the standard basis of W1 to the partial derivatives d/dxl. Moreover, if i : ? —> M is the inclusion of an open subset then each Txi is an isomorphism (use localizing functions) and, finally, if F : M —> ? is a diffeomorphism then each TXF is an isomorphism. Thus altogether we obtain isomorphisms txm Atxuq ATUaX{uaua) Am71, xeuQ . A1.1) The basis of TXM corresponding to the standard basis of W1 will be denoted by ¦?-, for the obvious reason.
Sullivan Models 133 Now suppose / G C°°(M). The gradient of f at ? e M is the linear map (df)x : TXM —> M given by ?/?(?) = ?f. The gradients (dul)x,..., (du?)x are the basis of (???)* = Horn (Tx il/. M) dual to the the basis ^fr, and dfx = Y,^-{x){du>a)x. A1.2) A differential p-form on il/ is a family ? = {?? G \p(TxM)s}xeM. For ? € Ua we have and ? is called a smooth differential p-form if the A,·,...,·,, are smooth functions in Ua. The vector space of smooth differential p-forms on M is denoted by APDR{M) and pointwise wedge multiplication makes Adr(M) = {ApDR(M)}p>0 into a commutative graded algebra. Note that A°DR(M) = C°°(M) and that Adr(M) vanishes in degrees > n. Formula A1.2) shows that df = {{df)x} is a smooth 1-form on il/; it is called the gradient of /. Extend d to AppR{M), ? > 1 as follows. For ? G APDR(M) write ? = ? ??,...???/??^ ? · · · ? du<? in Ua and define c/? G .4^ (il/) by c/? = EcfAij...^ ? du^1 ? · · · ? du^ . It is straightforward to verify that this definition is independent of the choice of chart, and that d is a derivation of degree 1 in Adr(M)· Moreover, for / G C°°(M) we have cPf = d fe$rdula) = ? ? ?L d< ? du{a = 0 because of the symmetry of partial derivatives and the skew symmetry of the wedge product. It follows that d2 — 0 in Adr(M) and so Adr(M) becomes a commutative cochain algebra. Definition The cohomology algebra ? (Aor(M)) is called the de Rham co- homology of il/. Finally, if F : M —> ? is smooth then it is easy to see that C°° (F) extends to a unique morphism Adr(F) : Adr(M) <— Adr(N) of cochain algebras (use localizing functions). (c) Smooth singular simplices. Recall that the standard simplex ?* C Rk+1 consists of the points ? = (to,...,tk) such that each ?; > 0 and Eti = 1. A smooth singular k-simplex m a smooth manifold M is a continuous map ? : Ak —> M that extends to a smooth map from some open neighbourhood of Ak in Rk+1. The smooth singu- lar simplices form a sub simplicial set. 5^ (il/), of the simplicial set 5* (il/) of all angular simplices (§10(a)). The correspond cochain algebra (§10(d)) is denoted
134 11 Smooth Differential Forms by C^(M) = C* Ef (Af);R). It F : M —* ? is smooth then S*(F) and C*(F) restrict (factor) to morphisms S^°(F) and C^(F). Next, we introduce smooth differential forms on the standard simplices, ?*, and the simplicial cochain algebra Adr referred to brief!)' in §10(c). Note that k the condition J] U = 1 defines an affine fc-space Fk C R+1 which is, in particular ? a smooth fc-manifold. Set Tx(Ak) = Tx(Fk), ? € Ak. Then a smooth differential p-form on ?* is a family ? = {?? G ApTx(Aky} ? ^k that extends to a smooth differential p~ form on Fk ; as with smooth manifolds we denote the space of smooth differential p-forms on ?* by ^^(?*). It is easy to see that the kernel of AjjR(Fk) —> Ar>ft(Ak) is preserved by d, so that A?>R(Ak) inherits a differential. Now let A; and Qj denote the face inclusions and degeneracies for the sim- plices Ak, as defined in §4(a). Then set {AoR)k = ????(??') and observe that {(ADR)n}n>Oi ADR(\i), Adr(bj) is a simplicial cochain algebra, which we de- note simply"by Ajjr. As observed in Remark 1 of §10(c), there is a natural inclusion APL( :R) of simplicial cochain algebras. Moreover, straightforward calculations establish Lemma 11.3 (i) (ADR)o = M. (ii) H((ADR)n) = R, n> 0. (Hi) Each APDR is extendable. (d) The weak equivalence ADR{M) ~ APL(M;R). Let M be a smooth manifold, and recall that S^°(M) denotes the simplicial set of smooth singular simplices. In particular, we may apply the construction of §10(b) with the simplicial cochain algebra Adr to obtain the cochain algebra Adr E~(M)) whose elements are the families ? = {?? 6 -4??(?'?'}?€5??(?) compatible with the face and degeneracy operators. On the other hand, any ? ? S^°(Ad) determines the morphism Adr{ct) ¦ Adr(M) —> ???(?|?|). Thus a natural morphism <*M : Adr(M) -> ADR (S? is defined by aM : ? "—> {^?»?(?)?}?€5?(?). Moreover, recall that APL(M;R) = ~APL(St(M);R). Thus the inclusion SZ°{M) —-)· S,{M) and APL(;R) —> ADR define natural morphisms
Sullivan Models 135 Theorem 11.4 The morphisms olm,&m and 1m are all quasi-isomorphisms. In particular, Adr(M) is weakly equivalent to Apl(M;R). For the proof of Theorem 11.4 we consider first an arbitrary natural trans- formation ? : A —> ? between functors from smooth ?-manifolds to cochain algebras. We suppose, however, that (i) H(A(mn)) = m = H(B(mn)). (ii) If U,V are open in M and ??,?? and ???? are quasi-isomorphisms, then so is (iii) If ? = UOi is the disjoint union of open sets Oi then ?? = ? ^O; : Lemma 11.5 With the hypotheses above, ? ? is a quasi-isomorphism for all smooth ?-manifolds M. proof: An i-basis for M is a family of open sets V\ c M, closed under finite intersection, and such that any open subset of M is the union of some of the V\. Given such an i-basis it is possible to write M = ? U W where ? = i W = U Wj and each O, and Wj is a finite union of elements of the i-basis. If j each ??? is a quasi-isomorphism it follows by induction on ? that each ??? ?···??? is a quasi-isomorphism and hence that ?0, ??? and 9onw are too. Thus in this case ?? is a quasi-isomorphism. Now suppose U is open in Mn. A standard cube in Mn is an open set V of the form (oi,6i) x ··· x (an,bn) and each ?? is a quasi-isomorphism by (i) above. But the standard cubes contained in U are an i-basis for U and so ?? is a quasi- isomorphism. But the open subsets of M diffeomorphic to open subsets of W1 are an i-basis for M and so ? ? is a quasi-isomorphism. D proof of Theorem 11.4: (i) Pm is a quasi-isomorphism. Lemmas 10 and 11.4 assert that Apr,(—;R) —> A^r is a quasi-isomorphism of extendable simplicial cochain algebras. Thus Proposition 10.5 asserts that ?iu is a quasi-isomorphism. (ii) im is a quasi-isomorphism. The inclusion S^°(M) —> S*(M) induces Qm ¦ C*{M) —> C^(M), and Theorem 10.9 identifies ?{??) and ?(??). On the other hand if /, g : M —> ? are smoothly homotopic then the ho- motopy C»(/) — C*(g) = dh + hd defined in §4(a) restricts to a homotopy be- tween Cf (/) and C^°(g). Thus since the identity and constant maps in Mn are smoothly homotopic we conclude that H (C*(Mn)) = R = H (C^(!n)). Next, if U and V are open subsets of M then the barycentric subdivision ar- gument [121] that shows that C,(U) + Cr(V) A C,(U U V) also shows that C?(U) + C?(V) -=? C?(U U V). Thus a long exact homology sequence argu- ment shows that if C™(U) —> C,(U), Cf (V) —> C,(V) and C?(U ? V) —>
136 11 Smooth Differential Forms C,(U ? V) are all quasi-isomorphisms then so is C?°(U U V) —> C,{U ? V). Dually, if 7i/,7v and 71/nV are quasi-isomorphisms then so is 7c/uV· Finally, if ? = Jj Oi then clearly 70 = ??7?;· Thus Lemma 11.5 asserts that I i 7?? is a quasi-isomorphism for all M. (iii) ?·? is a quasi-isomorphism. First note that ? (AorO^11)) = M (clas- sical Poincare lemma — [69, Example 1, sec 5.5]) while H (ADr E~(Mn))) = H(C?°(Rn)) = M as we showed in (ii) above. Next observe that if U and V* are open in M then the difference of restriction morphisms defines a short exact sequence 0 -> ADR(U UV)^ ADr(U) © ADR(V) -> ADr(U ? V) —> 0 [69, Lemma 1, sec. 5.4]. Similarly we have the short exact sequence 0 -> ADR (S?(U) U S? and a long exact cohomology sequence argument shows that the composite ADr(U U V) -> Adr (S?(U U F)) -> ADR (S?(U) U 5,° is a quasi-isomorphism if ??,?? and aunv are. Since Proposition 10.5 identifies ? {Adr{K)) with if (C*(K)) for any simpli- cial set ? and since as in (ii) above ? (C™(U) + C?°(V)) A if (Cf(i/ U V)) it follows that ADR E»°°(i/ U V)) —> AD? E~(?/) U 5~(F)) is also a quasi- isomorphism. Hence so is ??/??· Finally, if ? = ]JOi then ao = ???<· Thus Lemma 11.5 states that a m is i i a quasi-isomorphism for all M. D Exercises 1. Let M be a differential manifold with a fixed open cover 11 = {i/o, C/i, ·.·, i/m}. We denote by jVp the set of sequences / = (io,i\,...,ip) such that 0 < io < ii < ... < ip < m and Ui := t/io ? t/^ ? ... ? Uip ? 0. Define dL : Mp+l -> Afp by 9/(io,ii,...,ip+i) = (??,?,···,?,··?«?+?) ^d denote by pf : Adr(Ul) -> the restriction map induced by an inclusion Uk C Ul- We set: (il), d" : Cp'q(IX) -> Cp'q- p+i Prove that (C*(il) ,d! + d") is a cochain algebra with the product defined by: if ? = (io, h, --,? + ?') € ??+?' , and r — ?·. ?. 7;„) ? 7^ /' = (i i 4-1 ... i + ')
Sullivan Models 137 Is this product commutative? Prove that if each Ui is contractible in M then the graded algebra H(C*(ii) ,d! + d") is isomorphic to H*(M; R). 2. Let M be a connected ?-manifold and take a nicely imbedded ?-disk D in M. Denote by A'(M) the subalgebra of differential forms on M vanishing on the disk D. Prove that there is a sequence of quasi-isomorphisms A(M) <^-> A'{M) c^> ADR{M), where A(M) denotes a connected subalgebra of .4'(il/). 3. Let M and ? be compact connected oriented ?-manifolds. We define the connected sum of M and JV, ?#??, as follows : take a nicely imbedded n-disk D in M and in N, remove their interiors, and paste the boundaries together via an orientation reversing homeomorphism. We use the notation of exercise 2) and denote by um ? A(M) (resp. ?/? ? A(N)) a cocycle representing the orientation class of M (resp. of N). Prove that there is a quasi-isomorphism: {[Adr{M) ©r A(N)) ? uM, d) -» ADR(M#N), with d(u) = um — ?/?· Deduce that the cohomology ring H*(M#N;Ik) is the direct sum of H*(M: Ik) and H*(N: Ik) with the units and the orientation classes identified.
12 Sullivan models In this section the ground ring is an arbitrary field ic of characteristic zero. Having constructed the functor Apl ¦ topological spaces ~» commutative cochain algebras in §10, and the functor Adr: smooth manifolds ~» commutative cochain algebras in §11, we focus now on the study of commutative cochain algebras themselves. Here the principal role is played by the Sullivan cochain algebras (or Sullivan algebras for short) which we introduce now: Definition A Sullivan algebra is a commutative cochain algebra of the form (AV,d), where • V = {Vp}p>i and, as usual, AV denotes the free graded commutative algebra on V; oo • V = U V(k), where V@) C F(l) C ··· is an increasing sequence of k=0 graded subspaces such that d = 0 in V@) and d : V(k) —» AV(k - 1), k > 1. The second condition is called the nilpotence condition on d. It can be restated as: d preserves each AV(k), and there exist graded subspaces Vf. C V(k) such that AV(k) = AV(k - 1) <g> AVk, with d:Vk-> AV(k - 1). Observe that a Sullivan algebra is completely described by the vector space V and the linear operator, d. By contrast, for a general commutative cochain algebra (A, dA), the 'non-linear' multiplicative structure of A is also impor- tant. Nonetheless, if H°(A) = k then we shall show that there always exists a quasi-isomorphism from a Sullivan algebra to (A,d). This applies in particu- lar to Apl(X) for X a path connected topological space, since H° (Apl(X)) = H°(X;k) = lk(d. A0.1)). Definitions 1 A Sullivan model for a commutative cochain algebra (A, d) is a quasi-isomorphism ??: (AV,d) —* (A,d) from a Sullivan algebra (AV, d). 2 If X is a path connected topological space then a Sullivan model for APL(X), m: (AV,d) ^> APL(X), is called a Sullivan model for X. 3 A Sullivan algebra (or model), (AF, d) is called minimal if imdC A+V -A+V.
Sullivan Models 139 We shall frequently abuse language and refer simply to (AV, d) as the Sullivan model for (A,d) or for X. If H°(A) = Ik then (A,d) always has a minimal Sullivan model, and this is uniquely determined up to isomorphism. This will be shown here in the simply connected case, and in general in §14. In §14 it will be shown that every Sullivan model (or algebra) is the tensor product of its unique minimal model with a Sullivan algebra of the form (A(U ? SU), ?), where ? : U -^ OU. Sullivan algebras of this form are called contractible. Sullivan models for topological spaces X are, among all the commutative mod- els, the ones that provide the key to unlocking the rational homotopy properties of X. For example, if (AV,d) is a Sullivan model then (cf. A0.1)), as with any commutative model, H(AV,d) ? H*(X;]k). However, if (AV, d) is minimal there is also a natural isomorphism provided that X is simply connected and has rational homology of finite type. (The isomorphism is established in §15, based on a construction in §13.) Recall further from §10 that if simply connected topological spaces X and Y have the same rational homotopy type, then Apl(X) and Apl(Y) are weakly equivalent. Since minimal models are unique up to isomorphism, this implies that Apl(X) and Apl(Y) have isomorphic minimal models: the isomorphism class of a minimal model of X is an invariant of its rational homotopy type. In §17 we shall show that this defines a bijection, rational homotopy ? ~ j isomorphism classes of ? types J \ minimal Sullivan algebras over Q J where on the left we restrict to simply connected spaces with rational homology of finite type, and on the right to Sullivan algebras (AV, d) with V1 = 0 and each V* finite dimensional. Sullivan models also provide good descriptions of continuous maps, and of the relation of homotopy. Indeed, let A(t, dt) be the free commutative graded algebra on the basis {t, dt} with deg t = 0, deg dt = 1, and let d be the differential sending t ?-»· dt. As noted in §10, H (A(t,dt)) = Jk. Define augmentations ??,?? :?(?,?) —> A by ??(?) = 0, ??(?) = 1. Definition Two morphisms ??,?? : (AV,ci) —> (A,d) from a Sullivan alge- bra to an arbitrary commutative cochain algebra are homotopic if there is a morphism ? : (AV, d) —> (.4, d) <g> (A(i, dt), d) such that (id-Si)$ = ??, i = 0,1. Here ? is called a homotopy from ?0 to ?, and we write ?? ~ ?-
12 Sullivan models 140 We shall see that homotopy is an equivalence relation. Moreover, suppose mx : (AV.eQ —> ApL(-X) and my : (ATV,d) —> APL(y) are Sullivan models defined over Q, and that / : A' —> Y is a continuous map. Then it turns out that there is a unique homotopy class of morphisms ? : (AW, d) —> (AV, ci) such that ???? ~ ???(/)???: ? is called a Sullivan representative for /. Furthermore, the homotopy class of ? depends only on the homotopy class of /. If X and Y are rational spaces (cf. §9) with rational homology of finite type, then we shall show in §17 that / \—> ? defines a bijection homotopy classes of ? 2 ? homotopy classes of maps X—>Y j \ morphisms (A W,d)—> (AV,d) Finally, and perhaps of most importance, Sullivan algebras and models provide an effective computational approach to rational homotopy theory. In this and the next section we shall emphasize that approach, complementing the basic propositions with a range of examples. The introduction of relative Sullivan algebras, and the proofs of a number of theorems, follow in §14 and §15. This section, then, is organized into the following topics: (a) Sullivan algebras and models: constructions and examples (b) Homotopy in Sullivan algebras. (c) Quasi-isomorphisms, Sullivan representatives, uniqueness of minimal mod- els and formal spaces. (d) Computational examples. (e) Differential forms and geometric examples. (a) Sullivan algebras and models: constructions and examples. We begin by recalling notation and basic facts associated with free commuta- tive graded algebras AV — cf. §3(b), Example 6. These will be used without further reference. • AV = symmetric algebra (Feven) <g> exterior algebra (Vodd). The subalge- bras A (V^p), A (V>"),... are denoted AV^, AV>",.... • // {va} or vi, vq, ... is a basis for V we write A({va}) or ?(?? ,vo,...) for AV. • A9V is the linear span of elements of the form vi ? · · · ? vq, Vi 6 V. Elements in AqV have wordlength q. • AV = 0A'F and we write A^qV = ? A'V and A+V = A^V. Q i>q • // V = ? Vx then AV =
Sullivan Models 141 • Any linear map of degree zero from V to a commutative graded algebra A extends to a unique graded algebra morphism AV —> A. • Any linear map of degree k (k 6 Z) from V to AV extends to a unique derivation of degree k in AV. In particular the differential in a Sullivan algebra (AV, d) decomposes uniquely as the sum d — do + di + d-i + · · · of derivations di raising the wordlength by i. The derivation do is called the linear part of d. Our first step is to show the existence of Sullivan models: Proposition 12.1 Any commutative cochain algebra (A,d) satisfying H°(A) = ]k has a Sullivan model m:(AV,d) -?> (A,d). proof: We construct this so that V is the direct sum of graded subspaces Vj., /fc-l \ k > 0 with d = 0 in Vo and d : Vk —> A 0 V{ . Choose m0 : (AFo,0) —> \i-0 / (^4, d) so that if(mo) : Vo ? ?+(?). Since H°(A) = k, H(m0) is surjective. ( ( k \ \ Suppose m0 has been extended tom^ : ( ? I 0 Vi I ,d I —> (A,d). Let za \ \i=o J J ( k \ be cocycles in ? ? Vf such that [za\ is a basis for ker ?(??^). Let V'^+i \?=? / be a graded space with basis {va} in 1-1 correspondence with the za, and with fk+l \ aegva = aegza — 1. Extend d to a derivation in ? I 0 Vj I by setting dva — za. \i=o J Since d has odd degree, d'2 is a derivation. Since drva = dza = 0, d2 = 0. Since H(mk)[za] — 0, mkZa = daa, aa 6 .4. Extend m^ to a graded algebra A+i \ morphism mt+i : ? I 0 V{ ) —> A by setting mt+it;Q = ??. Then mk+idv^ = \i=0 / «mt+ifai and so mt+1d = dm,k+i. oo This completes the construction of m : (AV", d) —> (A,d) with V = 0 \'\ and m |y^= m/.. Since m ?? = m0, and H(m0) is surjective, i/(m) is sur- ( k \ jective as well. If H(m)[z] = 0 then, since ? is necessarily in some A 0 V-t , \t=o / = 0. By construction, ? is a coboundary in A ( 0 Vi ). Thus H(m) \i=0 / is an isomorphism.
142 12 Sullivan models We show next by induction on k that 14 is concentrated in degrees > 1. This is certainly true for k = 0, because Vo = H+(A). Assume it true for V{, i < k. ( k \ Any element in ? ? Vj of degree 1 then has the form \i=0 / ? = vo + l· Vk, Vi 6 Vi . ( \ Thus if dv = 0 then dvk € d ( ? ? Vi . By construction, this implies V i=o / vk = 0- Repeating this argument we find ? = v0 and H(mk)[vo] = i/(mo)[i>o] ^ 0, unless v0 = 0. Thus kerH{rrik) vanishes in degree 1; i.e., it is concentrated in degrees > 2. It follows that \\+\ is concentrated in degrees > 1. Finally, the nilpotence condition on d is built into the construction. ? Example 1 The spheres, Sk. Recall that in §4(c) we defined the fundamental class [Sk] € Hk{Sk;Z). This determines a unique class ? € Hk (ApL(Sk)) such that {?, [Sk]) = 1. and 1, ? is a basis for ? (Api(Sk)). Let ? be a representing cocycle for ?. Now if k is odd then a minimal Sullivan model for Sk is given by m : (?(?),?) A APL(Sk), Zj? Indeed, since k is odd, 1 and e are a basis for the exterior algebra A(e). Suppose, on the other hand, that k is even. We may still define m : (A(e), 0) —> ApL{Sk) by: dege = k, me = ?. But now, because dege is even, A(e) has as basis 1, e, e2, e3,... and this morphism is not a quasi-isomorphism. However, ?2 is certainly a coboundary. Write ?2 = dty and extend m to m:(A(e,e'),eQ —> APL(Sk) by setting dege' = 2k — 1. de' — e~ and me' = ?. A simple computation shows that l,e represents a basis of H (A(e,e'),d). Thus this is a minimal model for 5*. Finally, observe that quasi-isomorphisms (Ae,0) —» (H'(Sk),0) , k odd and (A(e,e'),ci) -=> (H*(Sk),0), k even are given by e h> ?, e'i-> 0. Example 2 Products of topological spaces. Suppose ??? : (AV, d) —> Apl(X) and ??? : (AW,d) —> APi{Y) are Sullivan models for path connected topological spaces X and Y. Assume further that the rational homology of one of these spaces has finite type. Let px : ? ? ? -*· ? and ?? : ? ? y -» ? be the projections. Then APL(px) ¦ APl(py) Apl(Y) —> APl(X ? ?) is a quasi-isomorphism of cochain algebras.
Sullivan Models 143 In fact, ???(??) ¦ ???(??) is clearly a morphism of graded vector spaces commuting with the differentials. It is a morphism of algebras because Api (X x Y) is commutative. To see that it is a quasi-isomorphism use Corollary 10.10 to identify the induced map of cohomology with the map H*(X; k) ® H*(Y; k) —> H*(X ? Y; k) given by ? ®? >-» H*(px)aU?*(??)?. But Propo«ition 5.3(ii) asserts that this map is an isomorphism. Since Api(px) ¦ Api(pY) is a quasi-isomorphism so is ???-???: (AV, d) ® (AW, d) A APL(X ? ?) , where (m ? ¦ ???)(? ® b) = ??^(??)???? ¦ ApL(pY)b. This exhibits (AV,d) ® (AW, d) as a Sullivan model for ? ? F. Observe that if (AV, d) and (AW, d) are minimal models then so is their tensor product. ? Example 3 ?-spaces have minimal Sullivan models of the form (AV. 0). An ?-space is a based topological space (X, *) together with a continuous map ? : X x X —> X such that the self maps ? \-? ?(?, *) and ? t-t ?(*, ?) of X are homotopic to the identity. We establish a result of Hopf: • If X is a path connected ?-space such that H*(X; k) has finite type then H*(X;k) is a free commutative graded algebra. To see this, observe first that because H,(X; Ik) has finite type, ?*(?) can be identified as a morphism of graded algebras, ?*(?) : H'(X; k) —> H*{X; k) ® H*(X; k). Moreover, the conditions ?(?, *) ~ id, ?(*, x) ~ id imply that for h 6 H+(X; k), ?*(?)? = /?®1 + ? + 1®/?, some ? 6 H+(X; k) ® H+(X; k). Now choose a graded space V C H+(X; k) so that H+(X: k) = V®H+(X; k) ¦ H+(X; k). The inclusion extends to a morphism ? : AV —> H*(X; k) of graded algebras and an obvious induction on degree shows that ? is surjective. Suppose by induction that ? is injective in AV<n, and let ? : H*(X; k) —> H*(X; k)lip(AV<n) be the (linear) quotient map. The general element in AV-n can be written as a finite sum w = J2 v^v!;2 ¦ ¦ -^"?^^...^, where the fcl kr. ati—fcr 6 y\.V<n, the Vi are linearly independent elements in V" and ki = 1 or 0 if ? is odd. Then the component of (? ® ??)?*(?)??? in (Im ?)" ® H*(X; k) is given by ,klt...,kr
144 12 Sullivan models By construction, the nvt are linearly independent. Hence if ??? = 0 then <p(llkiVi1---vt1-1---v$*ak1...kr) = 0· By induction on degw, ?^??1 ¦ ¦ ¦ ?*' · · •Urrafe1...tr = 0 for each i. Thus each a^...^ — 0 unless ki — ¦ ¦ · = fcr = 0. Then w 6 AV<n and ??? = 0, whence tu = 0 by induction. It follows that ? is injective in AV~n. Thus, by induction, ? is injective and the proof of Hopfs result is complete: <p:AV-=> H'(X;k). Finally; let ui) 6 Apl{X) be cocycles representing the cohomology classes Uj. The correspondence vi t-t iui defines a linear map V —> Apl(X) which extends to a unique morphism 7?i : (AV.O) —> Api(X). Since ? is an isomorphism it follows that m is a quasi-isomorphism: m:(AV,0) - is a minimal Sullivan model for the ii-space X. ? Example 4 A cochain algebra (AV, d) that is not a Sullivan algebra. Consider the cochain algebra (A,d) = (A(vi,v2.,v3),d), deguj = 1, with dvx = v2v3, dv2 = v3vi, and dv% = vivo- Here (A,d) is not a Sullivan algebra. (If it were, it would have to have a cocycle of degree 1). The cocycles 1 and v^vov^ represent a basis for H (A), and so it has a minimal model m : (?(??),?) -% (AV,d), degm = 3, m(w) = viv-2v3. ? Example 5 In contrast with Example 4> o.ny cochain algebra of the form (A,d) = (AV,d), V = V^2, imd CA+V-A+V is automatically a minimal Sullivan algebra. Indeed, in this case necessarily (A+V ¦ A+V)k+1 C AV-11'1 and so for degree reasons alone, d: Vk —> AV-k~l. This exhibits (AV, d) as a Sullivan algebra. ? Again, suppose (A,d) is a commutative cochain algebra. In §14 we shall show that if H°(A) = Jk then (A,d) has a (unique) minimal Sullivan model; this follows from a more general result about relative Sullivan algebras. However, if also H1 (A) = 0 then there is a simple inductive construction for a minimal model. We carry this out here, and prove uniqueness in this context in (c). Thus suppose given (A,d) with H°(A) = k and Hl(A) = 0. • Choose m2 : (AV2,0) —> (A,d) so that H'2(m2) : V2 A H2(A). Note that Hl(m2) is an isomorphism because HX(A) — 0 and that H3(m2) is injective because (AV2K = 0.
Sullivan Models 145 • Supposing that mk : (AV-k,d) —> (A,d) is constructed, we extend to mk+\ : (AV'-fc+1,d) —> (A,d) by the following procedure. Choose cocycles aa ? Ak+1 and z? € (Al''-fc)H2 so that Hk+1 (A) = Im Hk+1 (mk) ? 0 A- · [??] and ker Hk+2{mk) = 01· [zd] . In particular, mkz? — db?, some 6jj G A. Let Ft+1 be a vector space (in degree k + 1) with basis {v'Q,v'?} in 1-1 cor- respondence with the elements {aa}, {z?}. Write AV'^+1 = AV"^fc ® AVk+l. Extend d and mj, respectively, to a derivation in AF-t+1 and to a morphism mjfc+i ·' AV-fc+1 —> A of graded algebras, by setting dv'a = ?,, di$ = ? ? and m^ = aa, ???'? = 6/j. Since d has degree 1. d2 is a derivation. By construction, dr = 0 in V"fc+1 and in AV-fc. Thus d2 = 0. In the same way m^+id = dmk+i in Ffc+1 and AV-k, and so .Proposition 12.2 Suppose (A,d) is a commutative cochain algebra such that H°(A) = k and H^A) = 0. Then (i) The morphism m : (AV, d) —> (A, d) constructed above is a viinimal Sulli- van model. (ii) Ifris the least integer greater than zero such that Hr(A) ? 0, then V1 = 0, 1 < i < r and Hr(m) : Vr A Hr{A). (Hi) If dim Hk (A) < cc; k > 1, then dim Vk <oo, k>l. proof: (i) Since d : Vk+l —> AV'-fc, this exhibits (AV,d) as a Sullivan alge- bra. More, (AV<k)k+2 c A+V^k-A+V^k, and so (AV, d) is minimal. It remains to show m is a quasi-isomorphism, and this follows at once from the assertion, iri, \ - ? an isomorphism for i < k , ^ _ .,_ „. Hl{mk) is { . . f . - ,k>2, A2.3) v ' \ mjective for ? = A; + 1 - v ' which we prove by induction on k. For k = 2 A2.3) was observed at the start of the construction above. Suppose A2.3) holds for some k. Since mt+i extends mk. Im H(mk) C Im H(mk+\). Thus iP(mfc+1) is surjective for ? < k by induction and surjective for i = k + 1 by construction. To show Hl(mk+i) is injective for i < k + 2 let [2] be a cohomology class in kertfl(mfc+1), some ? < k + 2. We have to show [z] = 0. If deg[^] < A: or if
146 12 Sullivan models deg[z] = k + 2 then ? 6 AF-fc and [?] 6 keriP(mfc)- Thus [?] = 0 by induction if deg[z] < k and by construction if deg[z] = k + 2. Suppose deg[2] = fc + 1. Then ? = ????'? + ??0?0' + ??, some u> 6 AV^k, where we use the notation from the construction above. Since dz = 0, ????? = —dw and ? ?/3 [2/3] = 0 in H(AV~k). By construction, this implies that each ?/3 = 0. But then dw = 0 and ???[??] = Hk+1(mk)[w\. Again by construction, each Xa = 0. Hence ? = w and so [z] 6 ker Hk+l(mk). By induction, [z] = 0. (ii) This is immediate from the construction. (Hi) Since V2 = ii(yi), it is finite dimensional. Suppose by in- duction that V{ is finite dimensional, i < k. Then AV-k has finite type and, in particular, keriifc+2(mt) is finite dimensional. Since also Hk+1(A) is finite dimensional it is immediate from the construction that dimF*-'+1 < 00. ? Corollary Let X be a simply connected topological space such that each Hi(X;Q) is finite dimensional. Then X has a minimal Sullivan model such that V = {V1}i>o and eac^ V1 ™ finite dimensional. proof: The Hurewicz theorem 4.19 shows that Hx{X;lk) = 0, and Hk{X;Jk) is the dual of the finite dimensional vector space Hk(X',Q) ®q 1; (cf. §3(e) and Proposition 5.3). Thus H° (APL(X)) = A, H1 (APL(X)) = 0 and each Hk (Apl(X)) is finite dimensional. Now apply Proposition 12.2. ? Example 6 Simply connected topological spaces X with finite dimensional ho- mology admit finite dimensional commutative models. Suppose X is a simply connected topological space such that H*(X; Q) is finite dimensional (e.g. a simply connected finite CW complex — cf. Theorem 4.18). Then X has a minimal model in which V = {Vl}i>2 and each V* is finite dimensional. In general, AV will not be finite dimensional, as is already shown by the even spheres 52r (cf. Example 1)· There is, however, ? '?on-free' finite dimensional commutative model for X, constructed as follows. Put ?? = max{i | Hl(X; A~) ?- 0}. Write (AV)nx = ? ? (Imd)"-* ? C, where H © (lmd)nx = {kerd)nx; thus H A Hn*{AV,d) a Hn*{X;]k). Choose a graded subspace / c (AV,d) so that /* = 0, Ar < nx - rk _ an
Sullivan Models 147 Since V = {^/z}f>9> I is an ideal. It is immediate from the construction that / is preserved by d and that H(I,d) = 0 (because Hl(AV,d) = 0, i > ??). Thus the quotient map V:(AV,d)^((AV)/I,d) is a quasi-isomorphism, and so ((AV)/I.d) is a finite-dimensional commutative model for A'. Note that (AV)/I vanishes in degrees k > nx and that [(AV)/I]nx = Example 7 The minimal Sullivan algebra (A(a,b,x,y,z),d), where da = db — 0, dx = a~, dy = ab, dz = b2 and deg ? = deg 6 = 2 and deg ? = deg y = deg ? — 3. Here, the cohomology algebra ? has as basis 1, a=[o], ? = [b], ?-= [ay — bx], S = [by — az], ? = [aby — b2x\. Note that ?? = ? = ?j, and that all other products of basis elements in H+ are zero. We can now use the procedure above to construct a minimal model for the cochain algebra (H,0). This will have the form m : (AV, d) -=+ (H,0), beginning with V*2 : v\ dv\ = 0 mv\ = a vo dv2 = 0 mi'2 = ? V3 : u\ du\ = v'l mui = 0 U3 duz = v\ muz = 0 Note that necessarily m(v\U2 — V2U1) = 0 = m(v2U2 — V1U3). Thus we need to add yA '¦ x\ dx\ — v\Uo — V2U1 mx\ = 0 Xo dXo = V0U0 ~ V\U3 ???2 = 0, and V5 : yi . dyi = 0 myi = 7 2/2 ?/2 = 0 myo = ?. The process turns out (but we can not yet prove this) to continue without end. Observe that this provides two distinct minimal Sullivan algebras with the same cohomology algebra- D
148 12 Sullivan models (b) Homotopy in Sullivan algebras. The results in this topic flow from two basic observations. First, consider a diagram of commutative cochain algebra morphisms (A,d) ~ ? ? 0W,d) >{C,d) til in which ? is a surjective quasi-isomorphism, and (AV,d) is a Sullivan algebra. Lemma 12.4 (Lifting lemma) There is a morphism ? : (AV, d) —> (A, d) such that ?? = ? (? is a lift of ? through ?). proof: We may suppose V is the increasing union of graded subspaces V(k), k>0 such that V(k) = V(k - 1) ? Vk and d : Vk —> AV(k - 1). Assume ? is constructed in V(k — 1) and let va be a basis of Vk- Then ???? is defined and d(ipdva) = ?(?2??) = 0. Furthermore, ????? = ???? — ????. Since ? is a surjective quasi-isomorphism we can find aa G .4 so that daa = ipdva and ??? = ???. Extend ? by setting ??? = ??. ? Second, with a graded vector space U = {Ul}i>0 associate the commutative cochain algebra (E(U),6), defined by E(U) = A(U © 6U) and <5 : U A 5U. It is augmented by ? : (E(U),5) —> it, where e(U) = 0- Note that if U = {Uz}i>i then this is a Sullivan algebra. Cochain algebras of this form are called contractible. We recall a remark already made in §12: Lemma 12.5 ? : (E(U),S) —> A· is a quasi-isomorphism; i.e. H(E(U),6) = Jfc. proof: Let {ua} be a basis for U. A direct calculation (using chark = 0 if degua is even) shows that ? (A(ua,dua)) = Jc. But E(U) = (g) A(ua, oua) and, a since h is a field, homology commutes with tensor products (Proposition 3.3).? As a direct consequence of this lemma we obtain: The surjective trick: // (A, d) is any commutative cochain algebra, then the iAontiht of A extends uniquely to a surjective morphism ? : (?(?),?) —> (A,d).
Sullivan Models 149 Thus any morphism ? : (?, d) as (B,d) in which the inclusion ? : b tive. (A, d) of commutative cochain algebras factors b <g) 1 is a quasi-isomorphism and ? · ? is surjec- Recall that A(i, dt) denotes the special case of (E(U),6) with U = ht and with degi = 0. It is precisely the cochain algebra (???)?, of polynomial differential forms on the standard 1-simplex (§10 (c)). The augmentations ?0, ?i : A(i, dt) —> Ic, co(i) = 0, ci(i) = 1, correspond to the inclusions of the endpoints. As defined in the introduction to this section, two morphisms ??, ?? : (A V, d) —i (.4, d) from a Sullivan algebra are homotopic if there is a morphism ? : (?? d) —> (A, d) ® ?(?, dt) such that (??·??)? = ??, i = 0,1. Given that Api reverses arrows, and that A(t,dt) is the algebra of polynomial differential forms on the standard 1-simplex. this is the obvious analogue of a topological homotopy X <— ? ? /. In fact, Apl 'preserves homotopy' in the following sense. Suppose given con- tinuous maps /o,/i : X -> Y and a morphism ip : (AV.d) —> Api(Yr) from a Sullivan algebra (AV.d). Proposition 12.6 (AV,d) -> APL(X). If f0 ~ fx : X -> Y then APL(f0)ip ~ APL{fx)yj proof: Identify A(i,dt) as a subcochain algebra of Apl(I), by mapping t t-> u e A°PL(I), where u restricts to 0 at {0} and to 1 at {1}. Denote by j0, j\ : X —> X x / the inclusions at the endpoints and by px : ? ? I —> X and ?1 : X y. I —> I the projections. Then APL(X)®A(t,dt) (id -??,?? ·??) APL(px)APL(p') APL{X) X ApL(X) (AplUo),Apl(Ji)) Apl{X x /) is a commutative diagram of cochain algebra morphisms. Since H (ApL(px)) = H*(px;]k) is an isomorphism, Apl(px) · APL(pJ) is a quasi-isomorphism. We now 'make it surjective'. Let U C APL (? ? /) be the kernel of (ApL(j0),Apl(Ji)). The inclusion of U extends to a unique cochain algebra morphism
150 12 Sullivan models Extend the diagram above to the commutative diagram APL(X)®A(t,dt)®(E(U),S) (id ·??·?,»<?·??·?) APL{X) ? APL(X) AFL(px)ApL(p')g ApL(XxI) Here ^4??,(?? )·???,(??)'9 is, obviously, surjective, and it follows from Lemma 12.5 that it is a quasi-isomorphism too. Let ? : X x / —> Y be a homotopy from /o to f\. Use Lemma 12.4 to lift ???(?)? : (AV,d) —> Apl{X ? /) through the surjective quasi-isomorphism Apl(pX) ¦ Api{p!) ¦ Q. This produces a morphism ? : (AV, d) —> Api(X) <g> A(t, dt) ® E(i/). Then set ? = (id <S> id <g> ?)?; it is the desired homotopy from ApL(fo)^ to As with homotopy of continuous maps, we have Proposition 12.7 Homotopy is an equivalence relation in the set of morphisms ? : (AV, d) —> (A,d) from a Sullivan algebra. proof: Refiexivity is obvious: A homotopy ? : (AV, d) —> (A,d)<S>A(t,dt) from ? to ? is given by ?(?) = ?? <g> 1. Symmetry is also easy: the automorphism t t-> 1 — t of ?(?,???) interchanges ?0 and ??. For transitivity, suppose ? : (AV,cO —> (A,d) <8> ?^,??) and ? : (AV.cQ —> (A,d) ® A(t2,dt2) are homotopics from ?? to ?\ and from ?? to ??. Then ? and ? define a morphism (?,?) :(AV,d) —> [(A,d)®A(t1,dt1)]xA[(A,d)®A(t2,dt2)] = (?,?)®[?(^,??) ???(?2,?2)], where the second fibre product is with respect to the augmentations ?? : ?(??, dt\) A-, ??(??) = 1, and ?0 : A(t2,dU) —> k, ?0(??) = 0. Now consider the morphism, Mp'>·> = ^v> ? y,^^ > A(tudt!) xikA(t2,dt2)
Sullivan Models 151 given by t0 h-> A-??,1-?2), ?? ^ (<i,0), t2 *-* @,i2). Using Proposition 10.4(i) we may identify it with restriction APL(A[2])-+APL(L), where L C ?[2] is the subsimplicial set with non-degenerate simplices (e0), (ei), (e2), (eo,ei) and (??,??). Schematically L is given by the solid lines in the diagram Since Apl is extendable (Lemma 10.7(iii)) this morphism is surjective. Direct computation shows that ? ((.4pz,J) = ? (Apl(L)) = h. Apply the Lifting Lemma 12.4 to lift (?, ?) to a morphism n:(AV,d)-*(A,d)®(APLJ. Define ? : (Apl).2 —> A(t,dt) by to >-* I — t, ti *-* 0, ?2 ?-> t; i.e., ? is the face operator corresponding to (eo,e2) : ?[1] —> ?[2]. Then (??®?)? is a homotopy from ?? to(/J. D Notation The set of homotopy classes of morphisms (AV, d) —> (A, d) will be denoted by [(AV, d), (A, d)\. The homotopy class of a morphism ? will be denoted by [?]. D Example 1 Null homotopic morphisms into (A, 0) are constant. Let (AV, d) be a minimal Sullivan algebra and let (.4, 0) be any commutative cochain algebra with zero differential. The constant morphism ? : (AV, d) —> (?,?) is defined by e(V) = 0. We observe that for any morphism ? : (AV,d) —> U,0), ? ~ ? <^=4> ? = ? . In fact, suppose ? : (AV, d) —> A <g> A(i, dt) is a homotopy from ? to ? and write V = (J ^(*) with V(-l) = 0 and d : V(k) —> A^V(k - 1). Assume k>-\ by induction that ? : V(k - 1) —> A ® ?? ® di. Since di ? di = 0 it follows that ? (A>2V(A; - 1)) = 0. Choose ? G V(k). Then ?(??) = ?((??) = 0. Hence ?? e (.4®1)?(?®??®???). Since 0 = ?? = (id ®??)?? it follows that ?? e A <g> ?? ® di. In other words, Im ? C A® At® dt and ^ = (^(8>??)? = ?. ?
152 12 Sullivan models Definition The linear part of a morphism ? : {AV,d) —> {AW,d) between Sullivan algebras is the linear map ?)? : V —> W defined by: ?? - Qipv 6 A--W, ? e V. Note that Q(<p) commutes with the linear parts of the differentials: Q{ip)d0 = Proposition 12.8 (i) If ?? ~ ?? : (AV, d) —> (.4, d) are homotopic morphisms from an arbi- trary Sullivan algebra then ?{?0) = ? {??). (ii) If ?? ~ ?? : (AV, d) —> {AW, d) are homotopic morphisms between minimal Sullivan algebras, and if H1{AV.d) = 0, then Qip0 = ???- proof: (i) Any element in A{t,dt) can be uniquely written as u = ?0 + ??? + X2dt + ? + dy, where ? and y are in the ideal / C A{t) generated by <A - t) and Xi e 1·. If ? is a homotopy from ?0 to ??, define a linear map h : AV —> A of degree —1 by ?(?) = ??{?) + {??{?) - ??{?)) t - {-l)de*zh{z)dt + ?, where ? G A ® (/ ? d{I)). A simple calculation gives ?? - ?0 = dh + hd. (ii) We first observe that V'1 = 0. Write V = \JV{k), with d : k V{k) —> A^2V{k - 1). If V^Jfe - 1) = 0 then this implies that d = 0 in Vl{k). But lmd C A-2V, and so no element of V1(A;) can be a coboundary (except zero). Since H1{AV,d) = 0 this implies F1^") = 0. The assertion V'1 = 0 follows by induction. Now let ? : {AV,d) —> {AW,d)®A{t,dt) be a homotopy from ?0 to ??. Since V = V^2, it follows that ? : V —> A+W ® A{t,dt). Hence ? : A^*V —> A-k\V <g> A{t,dt). In particular, it induces a linear map ? : ?+?'/?^? —> (A+W/A^W) ® A{t,dt). By construction, ? is a map of cochain complexes. However, because {AV, d) and (AH7,d) are minimal, the differential vanishes in the quotients A+/A-2. Since A+V = V ? A^2V and A+W = W © A^2W we may therefore identify ? as a linear map of cochain complexes The space of cocycles in_{W, 0)®?(?, dt) has the form (IF <g> k · \)©{W <8> A(i)di)· It follows that (id ·??)? = {id ·??)?; i.e., Q(v?0) = <3(vi). D (c) Quasi-isomorphisms, Sullivan representatives, uniqueness of min- _ j„i., oriri fnrtnai spaces.
Sullivan Models 153 Suppose ? : (AW,d) —> (AV,d) is a morphism between Sullivan algebras. If ? is a homotopy between ??,?\ : (??',??) —> (A,d) then ?? : ??? ~ ???. Thus we can define ?* : [(AV, d), (A, d)} —> [(AW, d), (A, d)] by ?* [?] = [??]. Similarly, if ? : (A, d) —> (C, d) is a morphism of commutative cochain alge- bras then (? ® ??)? : ??0 ~ ???. Thus we can define ?# : [(AV, d),(A,d)}-+ [(AV., d),(C,d)\ by ??#[?] = [??]. Now suppose given commutath'e cochain algebra morphisms (A,d) •l< (AV, d) > (C, d) ? in which ? is a (not necessarily surjective) quasi-isomorphism and (AV, d) is a Sullivan algebra. The Lifting lemma extends to the fundamental Proposition 12.9 There is a unique homotopy class of morphisms ? : (AV. d) - (A, d) such that ?? ~ ?. Thus is a bijection. proof: We pro\re the proposition first under the additional hypothesis that ? is surjective. In this case the Lifting lemma 12.4 asserts that ? lifts to a morphism ? : (AV,d) —> (A,d) such that ?? = ?. Thus ry# is certainly surjecth'e. To show ?# is injecthre, suppress the differentials from the notation (for sim- plicity). Use the morphisms C ® A(t,dt) Udc'?°' ldc'^\ CxC ^JL ? ? ? to construct a fibre product, and observe (routine verification) that id, id ? ·?0, idA -??) : A ® A(t,dt) —> [C <8> A(t,dt)] xCxC (A x .4) is a surjecth'e quasi-isomorphism of cochain algebras. Now suppose 70,71 : AV —> A are cochain algebra morphisms, and that ? is a homotopy from 7770 to 7771. Lift the morphism (?,7?,7?) : ?^ —> [C® Mt,dt)] xcxC (A x A) through the surjecth'e quasi-isomorphism (? <g> id, idA -?o, idA -<=i)- This defines a morphism ? : AV —> A <g> A(t,dt), and it is immediate from the construction that ? :7? ^7l.
154 12 Sullivan models Finally, consider the general case, where ? may not be surjective. Let (E(C),6) be the acyclic cochain algebra of Lemma 12.5. Since E(C) = A(C © 5C) and since ? : C -^ OC, a surjecth'e morphism of graded algebras, g : E(C) -> C, is defined by c h-> c, 6c h-> dc. By inspection this is a morphism of cochain algebras. Consider the morphisms (A,d) ^^ (A,d) ® (E(C),6) ^> (C,d). ? Since H (E(C)) = Jk. id -? and ? ¦ g are surjective quasi-isomorphisms. Thus, by the argument abo\'e, (id -?)# and (? ¦ ?)? are isomorphisms. Since (id -?)? = id, ?# is the bijection immerse to (id ·?)#. Hence ?# = (??·?)#??# is also abijection. ? Suppose a:(A,d) —>(A',d) is an arbitrary morphism of commutative cochain algebras that satisfy H°(—) = k. Let m : (AV,d) A (A,d) and m' : (AV',d) —> (,4',?? be Sullivan models. By Proposition 12.9 there is a unique homotopy class of morphisms ? : (AV, d) —> (AV, d) such that ??'? ~ am. Definition A morphism ? : (AV,d) —> (AV',d) such that ??'? ~ am is called a Sullivan representative for a. If / : X -> Y is a continuous map then a Sullivan representative of Apl(J) is called a Sullivan representative of f It is follows at once from Proposition 12.6 that Sullivan representatives of homotopic maps are homotopic morphisms. As a second application of Proposition 12.9 we deduce the uniqueness of min- imal models in the simply connected case. Proposition 12.10 (i) A quasi-isomorphism between minimal Sullivan algebras is an isomorphism if both cohomology algebras vanish in degree one. (ii) If(A,d) is a commutative cochain algebra and H°(A) = Je, H1 (A) = 0, then the minimal models of (A, d) are all isomorphic. proof: (i) Let ? : (AV, d) -^ (AW, d) be a quasi-isomorphism between minimal Sullivan algebras, and suppose H1(AV,d) = 0 = H1(AW,d) Proposition 12 9 then yields a morphism ? : (AW.d) —> (AV,d) such that ?? ~ id. Thus ''·""'' -'1'' nr"i (aeain by Proposition 12.9) it follows that ?? ~ id.
Sullivan Models 155 Now apply Proposition 12.8(ii) to conclude that Q(ip) and Q{il>) are inverse isomorphisms of V" with W. Since Q(ip) is surjective we ha\-e Wk C ???? + AH7-'. It follows by induction that Wk and hence ??·'"-?" are contained in Imtp. Thus ? is surjective. Use the Lifting lemma 12.4 to choose ? so that ?? = id. Then ? is injective. But also ??? = ?, so that ?? ~ id. Now the argument just given shows that ? is surjective as well. Thus ? is an isomorphism and, since ?? = id, ? is the inverse isomorphism. (ii) Suppose (AV,d) -^ (A,d) ^— (AV',d) are minimal Sullivan mod- els. Proposition 12.9 provides a morphism ? : (AV, d) —> (AV',d) such that ??'? ~ m. Thus ?(??')?(?) = H(m) — Proposition 12.8(i) — and so ? is a quasi-isomorphism. Now part (i) asserts ? is an isomorphism. D Remark In §14, Proposition 12.9 will be extended to the non-simply-connected case. Example 1 Wedges. Since ApL(pt) = Jk (§10(d)) the inclusion of a point j : ? —> X induces an augmentation :Apl(X) —>*. Let (??,??) be based CW complexes, so that ? : Apl(Xo) -> ^ are augmented cochain algebras. Denote by jQ '· (Xa,xa) —> (\ZQXa,x) the different inclusions into the wedge. Thus ? ?-> {Apl(Jq)^} defines a morphism ApL{\/aXa) >[[[) (APL(Xa)) \ a / ft to the fibre product over Jk of the ApL(Xa). This morphism, which is obviously surjective, induces the analogue H'{\JaXa;K) »· ? ? (H*(Xa,Jk)), \ a / ik in cohomology and a cellular chains argument (§4(e)) shows this is an isomor- phism. Thus the first morphism is a quasi-isomorphism: the fibre product of augmented commutative models for the XQ is an augmented commutative model for V Xa. Now suppose ma : (AVa,d) A APL{Xa) are Sullivan models. Then the quasi-isomorphism ?
15t> identifies ( []) (AVa,d) as a commutative model for the wedge, \JQXa· 1° particular, a Sullivan model for ( ? ) (AVQ, d) is a Sullh'an model for the wedge. Ik Recall (§10) that two commutative cochain algebras (.4, d) and (A'.d) are weakly equivalent if they are connected by a chain of quasi-isomorphisms. It follows from Proposition 12.9 that a Sullivan model m : (AV, d) -^ (^4., d) lifts through such a chain to yield a Sullivan model m' : (AV.d) -^ (-4',d). /fence (.4. d) and (A', d) are weakly equivalent if and only if they have the same Sullivan models. (A common Sullivan model provides an obvious weak equivalence.) A particularly important case of weak equivalence is identified in the Definition A commutative cochain algebra (.4, d) satisfying H°(A) = k is for- mal if it is weakly equivalent to the cochain algebra (H(A), 0). A path connected topological space, ?', is formal if Apl(X; Q) is a formal cochain algebra. Thus (A d) and -Y are formal if and only if their minimal Sullivan models can be computed directly from their cohomology algebras; i.e, if these models are a 'formal consequence' of the cohomology. The quasi-isomorphisms at the end of Example 1, part (a) of this section, exhibit the spheres as formal spaces. Example 2 of §12(a) shows that the product of formal spaces is formal if one of them has rational homology of finite type, and the example above shows that the wedge of formal spaces is formal. Suppose now that X has rational homology of finite type. Then the equality {ApL,k)n = {ApL&)n® Jk defines an inclusion Apl(X;Q) ®qk —> Ap?,(X;k) of cochain algebras. Since the hypothesis implies that H*(X;Q) ®q Jk -=> H*{X;3c), this is a quasi-isomorphism. Thus if (AV,d) is a rational Sullivan model for X then (AV, d) ®q Jk is a Sullivan model for X over k. In particular, if X is formal then Api(X; k) is formal. In fact the converse is true: Theorem [144] [86] If APL(X;k) is formal for some extension field k D Q then X is a formal space. We shall not reproduce the somewhat technical proof, and we shall not often need to apply the theorem except in the case of geometric examples arising from C°° differential forms. (d) Computational examples. By using Sullivan algebras and computing homotopy classes of morphisms it is possible to illustrate interesting phenomena and, sometimes, to completely list all homotopy classes. Since Sullivan algebras (of finite type and with H1 = 0)
Sullivan Models 157 are always the models of simply connected spaces (§15), such phenomena can always be realized geometrically. We shall present Sullivan algebras in the concise form (AV, d) = ?(??, v2.... ; dv\ =,...) and we shall only indicate the degree of a generator Vi when it is not implicit in the formula for dv{. Example 1 (AV,d) = A(x,y,z;dx = dy = 0, dz = xy), deg a; = 3, degy = 5. A basis for the cohomology H — H(AV,d) is provided by 1, [a:], [y], [xz], [yz] and [xyz], and the only non-trivial products are given by [x] · [yz] = [xyz] = ~[y}[xz\. Consider all the possible morphisms ? : {AV, d) —>¦ (H, 0). For degree reasons, we must have ?? — ? [?], tpy = ?[y], and ?? = 0, some ?, ? € Jk. These morphisms induce distinct maps in cohomology, and hence represent dis- tinct homotopy classes. Thus this provides a complete list of the elements in [(AV,d), (H,0)]. Note that none of these morphisms is a quasi-isomorphism so that (as with Example 7 in (a)), (AV, d) and (H, 0) are not weakly equivalent: (AV,d) is not formal. ? Example 2 Automorphisms of the model of S3 x S3 x S5 x S6. Note that the tensor product of commutative models (Sullivan or otherwise) is a commutative model of the topological product, as follows from §12(a). Thus S3 ? S3 ? S° ? S6 is formal. In Example 1 of part (a) in this section we computed the minimal Sullivan models of spheres. This shows that the minimal model of S3 x S3 x S5 ? Se has the form (AV,d) = A(x,y,z,a,u;dx = dy = dz = da = 0, du = ?2), with degar = degy = 3, degz = 5, dega = 6. Define an automorphism, ?, of this model by: ?? = ?, i/py = y, ?? = ?. ?? = ?, ??? = u + xyz. Then ?(?) = id and Q(<p) = id but ? is not homotopic to the identity. Indeed, suppose ?? ~ ?? : (AV, d) —)¦ (AV, d) via a homotopy, ?. For degree reasons, ?? = a <g> fx (t) + xy <g> /2(i) + ? ® fs(t)dt. Similarly, ?? = u <g> gx (t) + xyz ® g2(t) + za® gz(t). From ??? = 0 deduce /i, f? € Jk. If ?$? = a it follows that fx = 1 and /2 = 0. In this case the equation d<&u = (??J implies that Pi = 1, and dg-2 = 0. Thus go € k and ??? — ???? = ???. In particular, ? is not homotopic to the identity. D Example 3 (AV, d) = A(a, x, b; da = dx - 0, db = ax), deg ? = 2, deg ? = 3. Let / be the ideal generated by a2, xb2, ab2 and 63. It is preserved by d, so that a commutative cochain algebra (A, d) is defined by .4 = AV/I. A basis for ? (A) is given by 1, [a], [a:], [ab] and [xb]. In particular, H+(A) ¦ H+{A) = 0, and so ? (A) ^ H*(S2 VS3VS6V S7;Jk). However (A,d) and #*(S2 V 53 V S6 V S7; it) are not weakly equivalent.
158 12 Sullivan models Indeed if they were, a minimal model for (A, d) would also be a minimal model for H(A, d). The construction process of Proposition 12.2 produces a minimal model (AW, d) -=> (A, d) that is given in degrees < 5 by ?(?, u, x, b; da = 0, du = a2,dx = 0, db = ax). But any morphism from (AW,d) —> (H(A,d),0) sends 6 t-4 0 for degree reasons, and hence kills the cohomology class [ab]. ? Example 4 A(x,y,z,u;dx = dy = dz = 0,du = xm + ym — zm), degar = degy = degz = 2, m > 3, h = Q. A morphism (A(a;,y, z, «),d) —>· (??, 0), dega = 2, is given by ? ? ??, y M· ??, ? ?· ??, where Am + ?™ = ?7" and ?, ?, ? G Q. By Fermat's last theorem, one of ?, ?, ? is zero. This illustrates how quickly number theoretic questions intervene in these computations. ? Example 5 [66] Two morphisms that homotopy commute, but are not homo- topic to commuting morphisms. Define a minimal Sullivan algebra (AV,d) by specifying the differential in a basis of V as follows: V3 V5 Ve V7 Vn ??,?2,?? y ? ? u>i,u>2 dxi = 0 dy = ?2?3 dz = 0 dv = yx-i dw\ = z2, dw2 Define two automorphisms ?, ? : (AV, d) —» (AV, d) by requiring ?? = ? + ?\?2, ???? = w\ + 2tu2, ? = identity on the other generators, and ¦??? = x\ + xz, i\)Wi = ui2 — zy, ? = identity on the other generators. Then a homotopy ? : (AV, d) —> (AV, d) <g> A(i, dt) from ?? to ?? is given by = ?? + x3, ?? = ? + ???2 + X3X2 + ydt, = ?? 4- 2tu2 - lyzt 4- 2xiudi 4- 2x3vtdt, ?\ + xz)vdt, and ? = identity on the other generators. (This is a longish, but easy computa- tion.) On the other hand, ???~?' then ? (?1) = ? (?). This implies that ?'?? = ?,·. Hence 0 = d(<p—<p')y. There are no cocycles in degree 5, so ?'y = ipy = y. Again, because ? (?1) = ? (?) we have ?'? = ?? + ??2?3, some ? € Je. Similarly, if ?' ~ ? then ?'Xi = ???, ?'y = ?? and ?'? = ?? + ??2?3, some ? € Jk. A computation now shows that ?'?'? ? ?'?'?. The reader is challenged to show that H(AV, d) is a finite dimensional algebra, with top cohomology class of degree 49. Example 6 [57] Sullivan algebras homogeneous with respect to word length.
Sullivan Models 159 The differential in a Sullivan algebra (AV, d) is said to be homogeneous of degree k with respect to word length if d : V —> Ak+1V. Thus if we let ffp,Q c Hv+q{AV) be the subspace represented by cocycles in APV then Hn = 0 H™ and H™ · Hr's C Hp+r^+s. p+q=n On the other hand, suppose a graded vector space V is presented as a di- oo oo rect sum of graded vector spaces: V = ? Vm. Then AV = 0 (AV)m, m=0 m=0 where ???···??? € (AV)m if Vi € Vm. and Era, = m. Thus if (AV,d) is a Sullivan algebra and d : Vm —» (AF)m_i, m > 0, then a 'lower grad- oo ing' Hr(AV, d) = 0 H^AV, d) is induced in the cohomology algebra. Again, 771=0 Now fix k > 1 and any graded vector space ? of the form ? = {??}{>2· We shall construct a minimal Sullivan algebra (AV, d) homogeneous with respect to word length of degree k and with the following properties: • V is a grad?d vector space and V = {Vl}i>2- oo • V = 0 Vm and d : Vm -> {Ak+1V)m_r 771=0 • Vo = ? and the inclusion induces an isomorphism ? -^ H1'*(AV). = 0. For this, set Vo = Z, and d = 0 in Z. Next, construct <i and Vm induc- tively so that d : Vm+i A {Ak+1V<m)m nkerd. Then H(d) : Vm+i -=> i?4+1>* (AV<m,d). Now the first three properties above are immediate, as is the fact that Hk+1>*(AV,d) = 0. It remains to see that H>k+1'*(AV, d) = 0. This is proved inductively as follows. Suppose v\,..., vr € V satisfy dvi = 0 and dv{ € A(v\,..., Vi_i). Then we can divide by vi,... ,vr to obtain a quotient Sullivan algebra (AW, d): here we may identify W with any complement of Jkv\ ? · · · 0 lkvT in V. We call (AW, d) an r-quotient, and show that: • For any r-quotient (AW,d), H^k+l>*(AW,d) = 0. A2.11) In fact suppose A2.11) is false. Then there is a least degree ? in which it fails and a_least r = ro for which it fails in that degree. Thus for any s-quotient (KZ, d) and for any ? > k + 1 we have: Hp'Q(AZ,d) = 0 iip + q <n or if p + q = n and s < ?0. A2.12) Now suppose (AW, d) is any r0-quotient and ? is a cocycle of degree ? in APW for some ? > k + 1. We shall show ? is a coboundary, thereby contradicting the hypothesis that A2.11) fails.
160 12 Sullivan models Suppose first rc, > 1. Then there is an (r0 — 1) quotient of the form (Av®AW, ?) projecting to (AW,d). In particular, ?A <g> ?) = vu, some ? € Av <g> ? VF. Now i/ degv ? odd then 5A ? ?) = ? <g> ? with ? G ??-F. Moreover ??? = 0, deg? < 7? and ? € ?-?+1 W. Our hypothesis above implies that ? = ??? and so ?A ® ? + ? ® ?) = 0. On the other hand, ?/ degv is even then ?<5? = 0. Thus ? is a cocycle of degree < ? and word-length > k + 1 in an (r0 - l)-quotient. Hence ? = oSP and EA ® ? - ??) = 0. In either case ? lifts to a cocycle of degree il and word-length ? in an (ro - l)-quotient. Our hypothesis implies that this cocycle is a coboundary; hence ? is a coboundary in {AW, d). Finally, suppose r0 = 0. In this case (AW,d) = (AV,d) and J2*+1'*(AV) = 0 by construction. Thus we may suppose ? > k + 2. Now for s > 1 any s-quotient (AZ, d) is the projection of an (s — l)-quotient of the form (Av <S> ??,?). We shall show that if the image of ? in (AZ, d) is a coboundary then so is its image in (Av <S> ??, ?). Since (clearly) ? maps to zero in some g-quotient it will follow that ? itself is a coboundary. Let ? be the image of ? in Av <g> AZ. Its image in AZ has the form d~T and so ? — ?A <g> ?) = vCt. If ? has odd degree we may write ? - ?A ® ?) = ? ® ?; here <?? = 0 and ? is a cocycle of word-length ? — 1>& + Iinan s-quotient. Since deg? < ? we have ? = du' and ? = ?A <g> ? — ? <g> ?'), //? /ias euen degree then ?? = 0 and ? is a cocycle in an (s — l)-quotient of word-length > k + 1 and degree < n. Thus ? = ??.' and ? = ?A <g> ? + ??'). ? Example 7 Sullivan models for cochain algebras (H, 0) wii/i trivial multiplica- tion. Let H = h® H-2 be a graded algebra with trivial multiplication: H+ · ii+ = 0. Regard if as a cochain algebra with zero differential. In Example 6 we constructed word-length homogeneous Sullivan algebras. Here we consider the case k = 1 and ? = H+. Then the construction of Exam- ple 6 gives a Sullivan model of the form (AV, d) with V = H+®V?®¦¦¦®Vm®¦¦¦ d : Vm -> (A2F)m_i , and H^iAV, d) = H+ and H^2>*(AV, d) = 0 . It follows that dh'iding by A-2V and by V>\ defines a quasi-isomorphism (AV, d) ^ (H, 0) which exhibits (AV,d) as the minimal Sullivan model for (H, 0). ? (e) Differential forms and geometric examples. In this topic we take Jk = R. Let M be a smooth manifold. In §11 we showed that Adr(M) was weakly eauivalent to Apl(M;R). Thus the Sullivan models of M are identified with the
Sullivan Models 161 Sullivan models of Adr(M). This represents a significant strengthening of de Rham's theorem, which asserts that H*(M;R) = H (ADr(M)). In particular suppose M is simply connected and has finite dimensional ratio- nal homology, and let (AV.d) be a minimal Sullivan model for Adji(M). Since this is also a model for Apl(M; M) we may conclude (taking for granted results announced in §12 and §13 and to be pro\red in §14 and §15) that • H(AV,d) = H*(AI;R) as graded algebras. • Whitehead products may be computed from the quadratic part of the differential, d. Finally, with these hypotheses Adr(AI) is connected to Apl(M;Q)®R by quasi- isomorphisms of commutative cochain algebras (§11). It follows that (AV, d) = (AW, d) <g> K, where (AW, d) is a rational minimal Sullivan model for M. Since geometric restrictions on M (e.g. M admits an Einstein metric) are often expressed in terms of differential forms, it is natural to expect that these could lead to information about the model and hence to information about the homotopy groups of the manifold. Unfortunately results of this nature have, so far, proved to be elusive. Now suppose a Lie group G acts on AI, and let Aor(M)g denote the sub cochain algebra of differential forms invariant under translation by ? € G. A theorem of E. Cartan [32], [70], asserts that if G is compact and connected, then the inclusion Adr(AI)g —> Adr(AI) is a quasi-isomorphism, and so Adr(M)g is also a commutative model for AI. Example 1 The cochain algebra Adr(G)g of right invariant forms on G, and its minimal model. Let g be the Lie algebra of the group G, identified with the Lie algebra of right invariant vector fields on G. Regard g" = Hom(g; K) as a graded vector space concentrated in degree 1. Let G act on itself by right translation. Then Adr(G)g — (Ag'.d), and the formula for the exterior derivative gives (??;??) = (?[??)) ^ Because g*1 generates Ag" and because d is purely quadratic this formula deter- mines d. Although AgE is a free commutative graded algebra, the cochain algebra (Ag3, d) does not, in general, satisfy the nilpotence condition required for Sullivan alge- bras. In fact, it is easy to see that (Agc, d) is a Sullivan algebra if and only if g is a nilpotent Lie algebra. On the other hand. G is always homeomorphic to the product of a compact Lie group with some W1 [32]. In particular (as with any compact manifold, 143], [69]) Ht(G;R) is a finite dimensional vector space. Thus the theorem of
162 12 Sullivan models Hopf (Example 3, §12(a)) shows that H*(G;R) is an exterior algebra on a finite dimensional vector space ? concentrated in odd degrees. This gives a minimal Sullivan model (??,?) ^ ADR(G), which contrasts with the morphism (Ags, d) —> Adr(G). ? Example 2 Nilmanifolds. Let G be a nilpotent connected Lie group and let ? be a discrete sub-group of G such that the quotient space X — G/T is compact. The covering projection ? : G —> X induces an isomorphism AdrX — (ApRG)r (= the complex of right ?-invariant differential forms on X). Thus we may regard {AdrG)g as a subcochain algebra of AqRX. In [130], K. Nomizu proved that this inclusion is a quasi-isomorphism. In this case the Lie algebra g of G is nilpotent and so (Ag5,c() is a minimal model of X over the real numbers. Obviously, d = 0 if and only if the Lie algebra g is abelian. In this case X = 51 ? · ¦ · ? S1. In particular, X is formal. The converse is true. Indeed, assume that there exists a quasi-isomorphism (?(??,?·2,... ,Xk),d) -^> H*(X;M). Then ? is obviously surjective, and since the product X\X2 · · · xk is a cocycle which cannot be a coboundary, ? is injective. This in turn implies that d = 0. This result has been proved in many different manners: [23], [39], [50], [53], [88], [114]. ? Example 3 Symmetric spaces are formal. Suppose ? is an involution of a compact connected Lie group G, and that ? is the connected component of the identity in the subgroup of elements fixed by ?. Then G/K is called a symmetric space of compact type. By Cartan's theorem, ADR(G/K)G -^ADR(G/K). An argument of E. Cartan shows that the differential in Adr{G/K)g is zero, thereby exhibiting G/K as a formal space. It is interesting to note that this is still the only proof available of the formality of G/K. Cartan's argument runs as follows: observe that r induces an involution ? of G/K and that ADR(a) restricts to an involution of Adh{G/K)g. Since the ac- tion of G on G/K is transitive. Adr{G/K)G may be identified with asubalgebra of AT?{G/K) = A(g/E)\ Let ?' :TG—>TG and ?' : TG/? —> TG/K denote the derivatives of r and ?. Since ? is an involution and t = {h G g | ?'h = 1}, it follows that ?' = - id in g/L Hence ADR{o) = {-\)Hd in APDR(G/K)G. Since Adr{o) commutes with d, d = 0. D Example 4 (Deligne, Griffiths, Morgan and Sullivan [42]) Compact Kahler manifolds are foimal. Suppose M is a complex manifold with operator J : TX(M) —> TX(M) satisfying J2 = — 1, ? e M. Let ( , ) be a Riemannian metric such that (JC, J77) = (?,?). Define ? G A2DR(M) by ?(?,?) = (JC,??). ?/ is called a Kahler manifold if ( , ) can be chosen so dw = 0.
Sullivan Models 163 Given a Kahler manifold, put dc = J~ldJ : ADR{M) —> ADR(M). Then ((fcJ = 0 and dcd = ddc. In [42] the authors show that d : keidc —> Imf and that the obvious inclusion and surjection Adh(M) A (kerdc,d) -A (ker cf/ Im dc, 0) are quasi-isomorphisms. This exhibits M as formal. Here, again, the only known proof of formality for Kahler manifolds proceeds via analysis using Adr[M). ? Example 5 Symplectic manifolds need not be formal. A symplectic manifold is a 2n-manifold M equipped with ? € A?DR(M) sat- isfying ?? = 0 and ?" ^ 0, ? € ?·/. Thus Kahler manifolds are symplectic. Now let X = Nz\ ? be the manifold constructed out of the 3x3 upper triangular matrices as described in Example 2. The Lie algebra of ? has the basis h\,h.2,h.2, with hz central and [??,/12] = ^3- Thus the minimal Sullivan model constructed in Example 2 has the form (A(x, y, z),d) with dx = dy = 0, dz — ? Ay and x,y,z of degree 1. Note that this is not formal (as is already observed in [107]; p261) since the algebra H*{X; E) is not generated by ?1 (?; ?). (See also Benson and Gordon [23], Cordero, Fernandez and Gray [39], Felix [50], Felix and Halperin [53], Lupton and Oprea [114], Hasegawa [88]) Put ? = ? ? ?. Then the inclusion (A(x,y,z),d) ® (A(x',y',z'),d) A ADR{M) is a minimal Sullivan model for M. Hence, as above, M is not formal. Let ? be the image in A2DR{M) of the cocycle ? Ax' + yAz + y' Az'. Since ?3 ? 0 and ?3 is left ? ? iV-invariant it follows that ?3 is not zero at any point of M. Hence (?,?) is a non-formal symplectic manifold. ' ? Recently I.K. Babenko and I.A. Taimanov, [17], have constructed for any ? > 5 infinitely many pairwise non homotopy equivalent non formal simply connected symplectic manifolds of dimension 2n. In fact, as proved, by S.A. Merkulov, [124], a symplectic 2n-manifold M is formal if the de Rham co- homology of M satisfies the hard Lefschetz condition: i.e. the cup product [??] : Hn~k(M; 1) —> Hn+k(M; R) is an isomorphism for any k < n. Exercises 1· Let (AV, d) be a Sullivan model of S2 V S3. Prove that there exists a quasi- isomorphism ? : (AV,d) -> H*(S2 V S3;Q). (Cf. d)-example 7). Determine the restriction of d to Vn for ? < 6. 2. Prove that if X and Y are formal spaces so are X V ?, ? ? ?, ? ? ? and ^ * Y. Determine the minimal Sullivan models of Ss ? CP6, S8 A CP6 and of S8*CP6.
164 12 Sullivan models 3. Let (A,(Ia) be any commutative graded algebra and (f\Z,d) -> H(A,Aa) a bigraded model as defined in f )-example 7. Prove that there exist a differential D and a quasi-isomorphism ? : (AZ,D) ~> (Ad.4.) such that D = Dx + Do + ... + Dt + ... with DiZ1; C (/\Z)^+l and Dx = d. {see [86].) 4. Construct a surjective quasi-isomorphism with duo; = 0 = du^ji dxi = v-> + wo. dxz = v4 + W4 + vow?-. dx5 = v^wo + dx7 = u4uL, dy2 = 0, dy4 = 0, Jz5 = 2/2By4 - y|) and dz7 = y4(y^ - y4). 5. Let (A,d) be a commutative differential algebra and for ? > 1 denote by A(n) the subalgebra defined by ? Ak nk<n A(n)h = < dAn if k = ? + 1 . [ 0 if jfe > ? + 2 Pro\re that if (,4, d) is formal so is (A(n).d). Let (AV,d) be a Sullivan model of a 1-connected CW complex. Construct a commutative model of the n-skeleton of X. (Cf. a)-example 6). 6. Let (A, d) be a commutative differential graded algebra and o2· € Ani, i = 0,1,2, be cocyclessuch that dai2 = ???2 ,da2s = a2a3 . Prove that (-l)degaia1a23 - a\2a3 is a cocycle; its cohomology class (which depends on the choice of ??2 and ?23) is called a Massey product. Construct a non zero Massey product in the cohomology of the minimal model (? {ae,a\o,bi\,bi5,b\o) ,d) such that dbn = ag , dbi$ = a6 a10 , dbi9 = aj0. Using the bigraded model of d)-example 7 prove that a formal space does not admit non trivial Massey products. 7. Let (AV, d) be a Sullivan minimal model and ? ,? : (aV,d) —> (A,d) two morphisms of commutative differential graded algebras. Assume that for any ? > 1, the restictions of ? and ? to (AV-n.d) are homotopic Pro\^e that ? and ?/> are homotopic. There are no phantom maps in rational homotopy theory! 8. Let M and ? be compact connected oriented ? -dimensional manifolds. Prove that if M and TV are formal so is the connected sum M#N (see §ll-exercise 3). Compute the mimimal model of (S3 ? 53)#CF3 up to dimension 12.
13 Adjunction spaces, honiotopy groups and Whitehead products In this section, the ground ring is a field 3c of characteristic zero. In applying Sullivan models it is important to be able to compute directly the models of geometric constructions from models for the spaces used in the construction. For example, in §12 we saw that the tensor product of models was a model of the topological product, while in §14 we will see how to model fibrations. A topological pair (Z, F) and a continuous map /:!*—> ?" determine the topological pair (?" U/ Z, X) in which A' U/ ? is the adjunction space obtained by attaching ? to X along /. These define the commutative square Y X ? Sz X U/ ? A3.1) Analogously, if (Cd) A are morphisms of commutative cochain algebras then the fibre product (C ? ? .4, d) fits in the commutative fibre product square (B;d) (C,d) (A,d) A3.2) (C xs ,4, d), and ?? is surjective if ? is. Note that in C.1) we have Z/Y = X U/ Z/X while in C.2) we have keri/> = ker^c- If (Z,Y) is a relative CW complex then the first equality implies (by the Cellular chain models theorem 4.18) that H, (Z, Y\ Jk) A H* (X of Z, X; k). Our first objective in this section is to show that • Suppose ? and ? are Sullivan representatives for f and for i and one of ?, ? is surjective. IfH.(Z,Y;k) A H»{X U/ Z;k) then [C xB A,d) is a commutative model for X U/ Z. Thus a Sullivan model for the adjunction space may be computed directly from the Sullivan representatives ? and ? of / and i, provided that ? is surjective. We shall then construct particular commutative models for the geometric con- structions: cone attachments, cell attachments and suspensions.
166 13 Adjunction spaces, homotopy groups Next, suppose (AV,d) is a minimal Sullivan model for a simply connected topological space X. Our second objective in this section is to construct a natural linear map vx : V —> Horn (vr.(*),*) and then to use the model of a cell attachment to show that: • ?? transforms the quadratic part of the differential in AV to the dual of the Whitehead product in ?*(?). Later (§15) we shall show that under mild hypotheses ?/? is an isomorphism, so that this transformation is in fact an identification. This section is organized into the following topics: (a) Morphisms and quasi-isomorphisms. (b) Adjunction spaces. (c) Homotopy groups. (d) Cell attachments. (e) Whitehead products and the quadratic part of the differential. (a) Morphisms and quasi-isomorphisms. Morphisms (C,d) -A (B,d) into the fibre product square - (A,d) of commutative cochain algebras fit (B,d) -. (A,d) V : and a commutative diagram (C,d) 1 (C,d) (C xBA,d) , (B,d) (A,d) extends by G,0) : (C xB A,d) —>· (C xB' A',d) to a commutative cube con- necting the corresponding fibre squares. We shall require Lemma 13.3 If ?-, ?, a are quasi-isomorphisms and if one of ?, ? and one of ?',?' are surjective then G,0) : (C xB A,d) —> (C xB· A',d) is a quasi isomorphism.
Sullivan Models 167 proof: If ? and ?' are both surjectu'e then a restricts to a quasi-isomorphism ker?/> -=» keri//. Now use the row exact diagram 0 0 kerip —~ C ? ? A C ker0' - C xB- ?' — C 0 0 to conclude that G, a) is an isomorphism. The other case to consider is that ?' and ? are surjective. As in §12(b) choose a surjective commutative cochain algebra morphism ? : (?, ?) —> {B',d) with H(E) = k, and consider C -4— B' -^- A' ?'? ? The argument above shows that G, ??) : (C xb A, d) —> (C ??· (?' ? ?), d) is a quasi-isomorphism. Since ?' is surjective the same argument shows that (id, X) : (C Xb' A',d) —> (C Xb1 (A' <g> E),d) is a quasi-isomorphism. Hence so is G, a). ? In general a commutative cube of morphisms of commutative cochain algebras will be regarded as a morphism ?* —>¦ V from the top square to the bottom one. If the vertical arrows are all quasi-isomorphisms this is called a quasi-isomorphism of squares and written ? -—* V. Two commutative squares connected by a chain · -=*· · «=- · · · · · <^L • of quasi-isomorphisms are called weakly equivalent (as already defined at the end of §7). Thus the conclusion of Lemma 13.3 asserts that G,0) defines a Quasi-isomorphism between the two fibre product squares.
168 13 Adjunction spaces, homotopy groups Next consider a homotopy commutative diagram (AV,d) -*- (B,d) -?¦ (AW,d) (C,d) -p (G,d) -viF.d) of morphisms of commutative cochain algebras, in which (AV,d) and (AW,d) are Sullivan algebras. Lemma 13.4 Suppose ?, ?, y are quasi-isomorphisms and that one of ? and ? and one of ? and ? are surjective. Then the fibre squares corresponding to ?, ? and to ?, ? are weakly equivalent. proof: For definiteness take ? to be surjective. Suppose first that both ? and ? are surjective. Then we construct morphisms a' ~ a and 7' ~ 7 such that replacing ? by a' and 7 by 7' makes the diagram above commutath-e. Thus in this case the Lemma follows from Lemma 13.3. To construct a', let ? : (AW,d) —> (G,d) <8> A(?, dt) be a homotopy from ?? to 77a: ?0$ = ?"? and ?? ? = ??, as described in §12(b). Use ?\ to form the fibre product G ® ?(?, dt) ? ? ?. Then, in the diagram of cochain algebra morphisms F <g> ?(?, dt) (MV, d) ——-*¦ G ? ?(?, dt)xGF , (?,?) the vertical arrow is a surjective quasi-isomorphism. Use the Lifting lemma 12.4 to lift (?,?) to ? : (AW,d) —> F®A(i,df), and set a' = ????· The morphism 7' is constructed in the same way. Finally suppose ? (for definiteness) is not surjective. Use the surjective trick (§12(b)) to extend ? to a surjection ?-? : (C,d) <S> (?,?) —> (G,d), with H(E, ?) = Jk. Then the fibre square for ? ¦ ? and ? is weakly equivalent to the fibre square for ?, ? by Lemma 13.3 and weakly equivalent to the fibre square for ?, ? by the argument above. D (b) Adjunction spaces. Suppose that i : Y —> ? is the inclusion of a topological pair (?, ?), and that
Sullivan Models 169 / : Y —> X is any continuous map. Then we ha\re the commutative square Y — ? ApL(Y) - ApL(Z) f fz , and the V : commu- Ar —- Ar Uf ? tative -4??,(?) :— APL(X Uf Z) model The square V determines the morphism (ApL(ix),APL(fz)) : APL(X Uf Z) -> APL(X) xApLiY} APL(Z) which (together with the identities) defines a morphism from V to the fibre product square. Proposition 13.5 If H,(Z,Y;k) -=> H,(X Uf Z,X;k) then the morphism (ApL(i.x),ApL(fz)) is a quasi-isomorphism. Thus the fibre product is a com- mutative model for the adjunction space. proof: Since H,(Z,Y:k) -=» H,(X U/ Z,X;k) it follows that APL(X Uf ?,?) -^> ???,(?,?). Thus in the row exact diagram 0 ^ Apl (X Uf Z, X) ^ APL (X Uf Z) ^ APL(X) ^ 0 0 >~APL(Z,Y) ^Apl(X) xApl(Y) Apl{Z) * APL(X) ^0 t"he central arrow is a quasi-isomorphism. ? Proposition 13.5 replaces the geometric adjunction space by the algebraic fibre product. Using Lemma 13.4 we can now pass to Sullivan models. Thus suppose ?, ? and ? are path connected, that mx : (AV, d) —> APL(X) , m : (AU, d) —> APL(Y) and mz : (AW, d) —> APL(Z) are Sullivan models and that (AV, d) A (AU, d) ? (AW, d) are Sullivan representatives for / and for i. Proposition 13.6 If Ht(Z,Y;k) A H,(X U/ Z,X;Jk) and if one of ?,-? is surjective then V is weakly equivalent to the fibre product square (AU,d)* (AW,d) (AV,d) * (AV xau AW,d) .
170 13 Adjunction spaces, homotopy groups In particular, (AV x,\u AW,d) is a commutative model for X U/ Z. proof: This is an immediate translation of Lemma 13.4, given Proposition 13.5. ? Now consider the very important special case of adjunction spaces given by attaching a cone. Recall that the cone on a topological. space Y is the space CY = (Y x I)/(Y x {0}). We identify Y as the subspace ? ? {1} C CY. Then, given any continuous map / : Y —> X we attach the cone CY to X along / to form X LJ/ CY. Now, suppose X is path connected. We shall use a different technique to construct a commutative model for XU/CY. Let ma : (AWa,d) ^> ApL{Ya) be Sullivan models for the path components Ya of Y, and put (B,d) = Y\(AWa,d). a Then l\APL(Ya) = APL(Y) and m = l\ma : (B,d) -=» APL(Y) is a quasi- a a isomorphism. Let ?? : (AV, d) —> (AWa,d) be Sullivan representatives for fa — f \y Wltn respect to a Sullivan model ??? : (AV,d) —> Api(X), and put y(( Finally, as usual, let ??^? ·' A(t,dt) —> k be the morphisms 11—> 0,1. Thus kere0 is the ideal A+{t,dt). Use the morphisms (AV, d) A (B, d) 4^- [k @[B® A+(t, dt)} , d) A3.7) to construct the fibre product AVxB (Jk®[B ® A+(t,dt)]). Proposition 13.8 The fibre product square for A3.7) is weakly equivalent to the commutative adjunction space for XUfCY. In particular the cochain algebra AV ??(?;?[?® A+(t,dt)]) is a commutative model for X U/ CY. proof: Use Proposition 13.5 to identify APl(X) Xapl(Y) Apl(CY) as a com- mutative model for Xo/ CY. Let ix denote the inclusion y ?—> (y, 1) of Y in CY and in ? ? / and also of {1} in /, and notice that the chain of quasi-isomorphisms Apl{CY.pt) A APL (? ??,?? {0}) <?- ApL(y) ? ApL (/; {?}) is compatible with the surjections Apl(ii) and id<8>ApL(ii). Regard the iden- tity of / as a 1-simplex. Restriction to this simplex is a quasi-isomorphism ApL(I, {0}) —>¦ A+(i,di) which converts id®APL{i\) to ??®??. Since APL(CY) = k © ApL{CY,pt) we obtain a chain of quasi-isomorphisms connecting APL(X) Xapl(Y)api<{CY) with Apl(X) xApL(Y) {k ®[APL(Y) ® A+(t,dt)])- On the other hand, there are homotopics ????? ~ ApL(fa)mx. A choice of
Sullivan Models 171 these defines a commutative diagram APL(X) - ^ (.W\d) ApL(Y) -.— ????.??)®?(?,<??)) ?—.. APL(Y) -?- (B.d) and this connects -4Pi,(X) xApl{y) {k @ [APL(Y) <g> A+(?,d*)]) to AV xB (J: ? [? <g> ?+(?,d?)]) by a chain of quasi-isomorphisms too. ? As a special case of Proposition 13.8 we prove Proposition 13.9 The suspension. ??, of a well-based topological space (Y, yo) is formal, and satisfies H+{EY;Jk) ¦ ?+(??;&) = 0. proof: Apply Proposition 13.8 to the case of the constant map / : Y —> {pt} to obtain a commutative model for {pt}ofCY of the form kx?(k © [? ® A+(t, dt)]). But this is just the cochain algebra Jk © [B <8> J], where J = kerei ? ?+(?, dt). Now J is the ideal generated by t(l — t) and dt and thus the inclusion kdt —> J is a quasi-isomorphism. It follows that Jk®[B <S> dt] includes quasi-isomorphically in k ? [? <8> J]. Let ? C ? ® dt be any subspace of cocycles mapping isomor- ~phically to H(B<S>dt). Since (B®dt) -{B®dt) C B®(dtAdt) = 0, the inclusion (k ? ?, 0) —>· Je © [5 <g> J] is a cochain algebra quasi-isomorphism, exhibiting k© ? as a. commutative model for {pt} U/ CY in which ? ¦ ? = 0. Finally, {pt} U/ CF = CY/{Y x {1}), which is homotopy equivalent to ?? since (Y,yo) is well-based (§l(d)). ? (c) Homotopy groups. Suppose / : (Y, *) —> (X, *) is a continuous map between simply connected topological spaces. Recall that a choice of minimal Sullivan models, mx : (AV,d) A APL{X) and my. (???, d) A >1FL(}') determines a unique homotopy class of morphisms ?} : (XV,d) —> (AW,d) such that ApL(f)mx = myipf. Moreover, Proposition 12.8(ii) asserts that these Sullivan representatives of/ all have the same linear part: Q(ff) is independent of the choice of ?/. This will therefore be denoted by QU) ¦ V -+ W. Note that (Proposition 12.6) Q(f) only depends on the homotopy class of /. We use this construction to define a natural pairing, ( ,· ) between V and ?*{?), depending only on the choice of ??? : (AV,d) —> Apl(X). First, recall the minimal Sullivan models m^. : (A(e),0) —> ApiJ(Sh), k odd. and rrik : (A(e,e'),de' = e2) —> ApL(Sh), k even, as constructed in Example 1,
172 13 Adjunction spaces, homotopy groups §12(a). Because H*(Sk:lk) is so simple, the morphisms mk are determined up to homotopy by the condition (H*(mk)[e\, [Sk]} = 1, where [Sk] is the fundamental class defined in §4(c) and ( , ) denotes the pairing between cohomology and homology (§5). Now suppose a ? 7?^(?') is represented by ? : (Sk, *) —> (X, *). Then Q(a) : Vk —> Ik ¦ e depends only on a and the choice of the morphism ??? : (AV, d) —> Define the pairing <—,-): F x ??,(?) —>· k A3.10) by the equations Q(a)v , ? ? Vk, (vi d\e ?= 1 0 , deg ? -?- deg a . It is immediate from the definition that for / : (Y, *) —> (?", *) as above, (Q(f)v;?) = (?;?.(/)?), ? € V, ? G ??.(?). Lemma 13.11 The map (—;—} is bilinear. proof: The linearity in V is immediate from the definition. To show linearity in ?*(?), consider the space SkvSk. Let io,ii be the inclusions of the left and right spheres and let j : Efc, *) —> (Sk V Sk, *) satisfy [j] = [i0] + [i{\ in irk(Sk V 5*). The linear maps Q(io), Q(i\) and Q(j) can be computed as follows. Let ???,??? be the basis for Hk(Sk V Sk;3e) defined by (u;o,^(to)[5fc]) = 1 = (w1,H.(i-l)[Sk]) and (???,??,^?)^]) = 0 = (w1,H,(i0)[Sk}). Then a minimal Sullivan model for Sk V Sk is given by where eo,e\ are cocycles of degree k, the remaining generators have greater degree, and H(m)[e\] = ??, ? = 0,1. Thus a Sulli\ran representative ?0 for i0 satisfies ?(??)[??] = [e] and ?(??)[??] = 0. Since the models have no coboundaries in degree k this implies </?oeo = e and </?oei = 0; i.e. Q(io)eo = e and Q(i'o)ei = 0. In the same way it follows that Q(h)e\ = 0 and Q(ii)ei = e and Q(j)e0 = e = Q{j)e\. Now let ??,?? represent ??,?? € 7rfc(X). Then cto + ui is represented by We just computed Q(io),Q(ii) and Q(j). Since (??,??) ? i0 = a0 and (??,??) ? ii = ??, it is immediate from these computations that Q(ao,a.i)v = {u;ao)eo + (v\a\)ei, and so Q(a)v = ((?;?0} + (?;?·1))?, as desired. D
Sullivan Models 173 Lemma 13.11 shows that (—",—) induces a natural linear map vx : V —> Homs (?.(?'), ?) , ? ·-* (?; ->. In §15 we shall show that if H*(X;Q) has finite type then this map is a linear isomorphism. Finally, recall that m ? : (AV'x,d) —> Apl{X) denotes the minimal Sullivan model of our simply connected topological space X. On the one hand we have the Hurewicz homomorphism hurx : ??,(?') —> H,(X:Z). On the other, since Imd c ?--?"?, division by ?-2?? defines a linear map ? : H+(AVX) —> \'?. It is immediate from the definition that [z] G H+(\\\.d) = (H(mx)[z],hurx{a)) J^.^ Thus the diagram H+(AVX) H{'2x)-~ Hm(X; Horn (?,(?'), ?) commutes, where hur\ is the dual of hurx. (d) Cell attachments. Fix a continuous map a:{Sn,*)—>(X,*), n>l, into a simply connected topolog'ical space A", representing q € ??(?). The space A"Ua Dn+1 is the space obtained by attaching an {n + \)-cell to X along a. Now choose a minimal Sullivan model mx:(AV,d)-^APL(X). Our objective is to use this, and the pairing (—; —> : V ? ??,(?') —> A, to construct a commutative model for A' Ua Dn+1. If ? = 1 then A" Ua D2 ~ A" V S'2, because ? will be based homotopic to the constant map. In this case the discussion of wedges in §12(c) shows that A"Ua D2 has a commutative model of the form (??·* ? ku.d) in which deg u = 2, (AV.d) is a sub cochain algebra, u ¦ A+V = 0 = u2, and du = 0. For ? > 2 define a commutative cochain algebra (??? © ku,da) by. • degti = ? + 1. • AV is a subalgebra and u · A+V = 0 = u2. • dau = 0 and dav — dv + (v; a)u, ? € V.
174 13 Adjunction spaces, homotopy groups Note that da? = dz if ? € ?-2F because da is a derivation. Proposition 13.12 The cochain algebra (AV@Jcu,da) is a commutative model forXUaDn+1. proof: We apply Proposition 13.8, noting that Dn+1 = CSn. This gives a com- mutative model for A' Ua Dn+1 of the form A F x.\w (¦& ? [A+W <8> ?+(?, dt)]), where (AW,d) is the minimal Sullivan model for Sn (Example 1, §12(a)). The quasi-isomorphism (AW,d) —> (H(Sn),0) then defines a quasi-isomorphism AV xaw (Jfc © [A+W ® A+(t,dt)]) ? AF x//(Sn) (?* ? [#+(Sn) ® A+(i,df)D· These constructions are made using a Sullivan representative c^a : (AF, d) —>¦ (??'?,??) for a. Let ^ be the composite of ?? with m : (AW,d) -%¦ H(Sn) and define a linear inclusion ? of A F into the fibre product by ?A) = 1 and ?? = (?,^? ® ?), ? G A+V . Now H+(Sn) = Jk[e], where [e] is dual to the fundamental class [Sn]. Hence ?? = (?,?) if ? € A^2F or if ? € Vfe,fc ? ?. Moreover, for ? e Fn, ?? = (?, (v;a)[e] ® t) . It follows at once that ? is an algebra morphism and that it can be extended to a cochain algebra morphism (AV ® Jfcu, dQ) —* ?V ? wEn ) (A © [H+ En) <8> ?+ (i, dt)] ) by defining u ?—? ( — l)n[e] <8> dt. It is a trivial verification that this is a quasi- isomorphism. ? Remark In the situation of Proposition 13.12, consider the commutative square of continuous maps, X\JaDn+1 ^ A" A3.13) Dn+l " : 5" This diagram determines the row exact diagram > APL{X ue ?>n+1, A') ». APL{X UQ Dn+1) > APL(X) > 0 APL(Dn+\Sn) >¦ APL(Dn+1) >APL(Sn) >0 A3 14)
Sullivan Models 175 The proofs of Propositions 13.8 and 13.12 show that this is connected by a chain of quasi-isomorphisms to the diagram 0 in which: ¦ku· ku (AV@ku,dQ) I 3 i k ® kw ? kdw (AT7,· 0 o, A3.15) Xu = dw, ?u — dw, ?? = (?; a)w, and (gw, [Sn]} = 1. (e) The Whitehead product and the quadratic part of the differen- tial. Let X be a path connected topological space. The Whitehead product is a map [ , ]\V : nk(X) x ??(?) —> nk+n_i{X). We define it now and then show how, if A' is simply connected, to read it off from the differential in a minimal Sullivan model. Recall that in Example 5, §1, homeomorphisms Ik/dlk -^ Sk and dlh+1 -zz± Sk are specified for k > 1. Regard the first homeomorphism as a continuous map ak : (Ik,dlk) —> (Sk,*). Thus akxan:{lk+n,dlh+n,yo) -» {Sk ? Sn,Sh V 5", *) , y0 = A,1,. · ·, 1). Use the second homeomorphism to identify ak x an L/fc+n a^ a continuous map The homotopy class [ak,n] € ?·?;+?_1E/: V 5") is called the universal (k,n)- Whitehead product. (This is actually an abuse of language. The map aktH must be precomposed with an orientation-preserving rotation sending * to xq. However any two such relations are based homotopic, so the based homotopy class of the composite depends only on aktJl.) When k = ? = 1 the picture
176 13 Adjunction spaces, homotopy groups identifies [aiti] as the commutator [io]" [ii]—1 [*o][ii]- Definition The Whitehead product of 70 G 7ik(X) and 71 G ??(?) is the homotopy class [70,7i]w G Kk+n-i(X) represented by the map [cccyhv : Sk+n~l -^ Sk V5 where cq : Sk —> X represents 70 and C\ : Sn —> X represents 7X. Next, consider a minimal Sullivan algebra (AV,d). The restriction of d to V decomposes as the sum of linear maps a, : V —> Az+lV, i > 1. Each a,· extends uniquely to a derivation d, of AV. and di increases wordlength by i. Moreover d decomposes as the sum d = dx + dz H of the derivations di. Clearly, d\ raises wordlength by 2 and d2 — d\ raises wordlength by at least 3. Since d2 = 0 this implies d\ = 0; i.e. • (AV,di) is a minimal Sullivan algebra. Definition di is called the quadratic part of d. Recall from §12(b) that the linear part of a morphism ? : (AV, d) —> {AW, d) of Sullivan algebras is the linear map Q{ip) : V —> W defined by ?? — Q(<p)v G A-2W. It extends uniquely to a morphism of graded algebras, AQ(<p) : AV —> AW. If (AV, d) and (AW,d) are minimal then the same calculation as for d\ above shows that (AQ(ip)) d\ = dx (AQ(ip)); i.e., • AQ{<p) : (AV, di) —> (AW,di) is a dga morphism between minimal Sullivan algebras. Finally, suppose that X is a simply connected topological space, with minimal Sullivan model mx :(AVx,d)^APL{X). Define a trilinear map by ?,7?> = (^7i>(^7o>+(-l)gdeg'0(y;7o>K7i>, j ?' ,v, 7o» 7i G ?,(?; Proposition 13.16 The Wliitehead product in ?, (A") is dual to the quadratic part of the differential of (AVx,d). More precisely, ? G VY , (di«i7o,7i> = i-D^-'^hclijw), 70 € irk(X) , 71 G 7??(?) .
Sullivan Models 177 proof: For the sake of simplicity we denote the universal Whitehead product [ak,n] simply by a G irk+n-i(Sk V Sn). If c0 : E*,*) —> (X..*) and cx : (Sn, *) —> (X, *) represent 70 and 7!, then put c = (co:ci) : Sk V 5" —> X, so that [70,71]w = it«(c)q. Hence (?;[7?,7?]??) = (Q(c)w;a·). Moreover, if iQ : Sk —> Sk V Sn and ii : 5" —> Sk V 5" are the inclusions, then (di«;To,7i> = (dit'iMcHto], ?<#?]> = (diQ(c)v; [io],[»i]> · It is thus sufficient to prove the proposition for X = Sk V 5" and 70 = [z'o]*and 7i = [ti]· For this we require the minimal Sullivan model of 5* V 5", at least up to degree k + n — 1. First, let m0 : (AW0,d) -%¦ APr,(Sh) and mi : (A\Vi,d) -%¦ APL(Sn) be the minimal Sullivan models of Example 1, §12(a). Thus (AW0,d) = (A(eo,...),d), (AWi,d) = (A(ei,...),d) and (H(mo)[eo},[Sk}} \= 1 = (H(mi)[ei],[Sn]). Let p0 = (id,const.) : Sk V 5" -> 5fc and pi = (const., id) : Sk V 5" —>· 5" denote the projections. Use the construction process of Propo- sition 12.2 to extend the morphism (???(??)??,?) · (???(??)???) : (?Wo,ci) ® (???'?, d) —»· ApL (Sk V 5") to a Sullivan model , y,,y2f... ), d) i V 5n), in which: dx = eo <8> ex and degy; > degx, i = 1,2,... . Denote this model simply by m : (AV,d) —* APL(Sk V 5"). Next write ? = at,„, so that SkxSn = (Sk\zSn)Ua(IkxIn) . The commutative square Sk ? Sn = (Sk ? 5") Ua (Ik ? 7") «(Po'Pl) Sk ? 5" 7 X 7" " : ek+n—l 3 is precisely A3.13), where ak and an are the maps defined at the start of this topic. In particular, Proposition 13.12 asserts that (AV © ku,da) is a commuta- tive model for Sk x 5", and provides an explicit isomorphism of H{AV ? fcu, dQ) with77*Efc ? 5"; it). We shall use this isomorphism to evaluate the classes [eoei] and [u] on a certain homology class, [z]. The identity map of 7 is a singular 1-simplex l G Ci G; ?). The chain iuk = EZ{l®---®l) g Ck{Ik;Z) projects to a relative cycle representing \Ik] ? Hk{Ik,dIk;Z), as was observed in §4(c). Moreover, because
178 13 Adjunction spaces, homotopy groups at- : dlk —> {pt}, and because C*({pt};Z) = ?-pt, it follows that C*(ak)wic is a cycle in Ck(Sk\Z). Define a cycle ? G C*(Sk x 5n;Z) by ? = EZ {C*{ak)wk <g> C»(a„)u;„) = C,(ait x a„)tufc+n. As observed in §4(c), the cycle C,(ak)wk represents [Sk]. Since (§4(b)) EZ and AW7 induce inverse isomorphisms of homology, a quick computation gives <[eo ® ei], [*]> = <# (mo)[eo] ® # (m,)[ei], [5*] <8> [5n]> = (-l)*n. On the other hand, it follows from diagram A3.15) that H*(ak x dn)[u] = ?*?, where: [u] is regarded as a relative cohomology class in H*(Sk ? Sn, Sk V5n; Je), ? G Hn+k~1(Sn+k~i;]k) is dual to [S"*] and d* is the connecting homor- phism in cohomology for the pair (Ik ? In,Sn+k~1). Hence {[u],[z]) = ([u],H.(ak x an)[Ik+n}) = (d*u,[Ik+n}} = (-l)k+n(u,d[Ik+n}) . But, as observed in §4(c), d[Ik+n] = [5*4""-1]. Thus ([u],[z]) = (-l)A+n. Since dax = ec,e\ + (x;a)u (cf. §13(d)) it follows that [????] + (s;q)[u] = 0. Evaluate this equation at [z] to find (-l)kn + (-l)k+n(x;a) = 0: i.e. In particular it follows that {dix; [to], [*i]> - (eo ® er, [to], [t"i]> = (- and Unless k = ? is even, y*4"" = Jca;, and this completes the proof. In the remaining case, e'o and e\ both have degree k + n — 1. However poa : Sk+n~1 —>¦ 5* factors as S*4""-1 —»¦ 7fc ? /n —». 5fc ? 5n —> Sk, and hence is homotopic to the constant map. Thus (e'0;a) = (Q{po)e'o;a) = (e'0;Q(p0)Q(a)[idSn+,-i]) = 0 and (dieo;[<o],[<i]> = (eg-,[io],[ii]>=0. The same holds for e[, and so the proposition follows in this case as well. ? Example 1 The even spheres S2n. Let ? ? ?2?E2?) be represented by the identity map, and recall that the minimal Sullivan model of S2n has the form (A(e,e;), de' = e2). It is clear that (e;a) = 1, and so {e';[a,a]w) = -(e2;a,a)= -2. In particular, [?,?]?? is not a torsion class in 7r4n_iE2n). D
Sullivan Models 179 Example 2 S3 V S3 U/ (?>g II D?). Let ??,?? : S3 —i S3 V 53 denote the inclusions of the left and right hand spheres, and put a{ = [at·] S ??3E3 V S3). Attach Z>g and Df to S3 V S3 by the maps [ao, [ao>ai]w]w and [??, [ai,ao]w]w The same argument as in Proposition 13.12 shows that the resulting space has a commutative model of the form (AV ? Jhuo ? Jku\,D) where uoui = Uq = u\ = uqA+V = Ui A+V" = 0. Duo —'Du\ = 0, andDv = dv+(v;[ao,[ao,a\]\v]w)uo + (v;[an,[an,ao]w]w)ui. Of course deg u0 = 8 = degU\. Now (AV, d) = (A(eo, ei, x, yo, y\, z\, z-z, ¦ ¦. ), d) with dx = eoei ^yo = eo%, dy\ = e\x and deg z; > 9, i = 1,2,... . From Proposition 13.16 it follows that Dy0 = eox + uo, Dy\ = e\X + U\ and D = d on the other basis elements of V. Moreover, it follows from diagram 13.15 that uq and u\ represent the coho- mology classes dual to the cells Dq and Df, so that [1], [eo], [ei], [uq] and [ui] is a basis for H(AV ? Iiuq ? lku\, D). This implies that a quasi-isomorphism (AV ® kilo + Jkui, D) -=> (A(eo,ei,x)/eoeix, d) is given by e0 ·->· eo, e\ ·->· ei, s ·->· x, y,· ·->· 0, z,· >-)¦ 0, uo ·->· —eox, andui >-)¦ —e\x. Thus (A(eo,ei,x)/eoeix, d) is a commutative model for 53 V 53 U (Do LI-0!)· We may use this to calculate the minimal Sullivan model of 53V53U(D^ IJ Df), which has the form (A(eo,ei,x,w,wo,wi,...), dw = e^eix, dwo = eow, dw\ = e\w,... ). Note that (w;[a.?]w) = 0 for all homotopy classes ? and ? in ?. (S3VS3V(D8IJD?)). D Exercises 1. Let / : (A',*) -> (Y, *) be a continuous map between simply connected topological spaces. Determine a Sullivan representative of ?/ : ?? -> ??. 2. Prove that the cofibre of the natural inclusion CP2 c—^CP5 and the space 56V58V510 have the same Sullivan minimal model. 3. Let (AV, d) be the minimal model of S2 V S2. Determine the elements of V-5 which are dual to Whitehead products. 4. Let AT be a simply connected ?-space. Prove that [a,?]w = 0 for every ? 5- Let X be a simply connected space and ? € ??(?), ? G ??(?). a) Prove that if ? : nq(X) —> irq+i(X) denotes the morphism induced by the suspension map [S",X] -*¦ [5«+1,??'] then E([a,?]w) = 0.
180 13 Adjunction spaces, homotopy groups b) Prove that if hurx : ?, (A') -4 Hg(X) denotes the Hurewicz homomorphism then hur([a,?]w) = 0. 6. Let X be a 1-connected space. Prove that if X is formal then the Hurewicz homomorphism hurx induces a surjective map ? (?, (A") ® Q)a —>¦ H"(X;Q). 7. Consider the quotient algebra ? = A(x\ ,X2,X3,X4 with deg X!=deg x2=deg x3=3 and deg i4 =5. Prove that the Sullivan minimal model (AV",d) of (H ,0) is given in low degrees by the following table. (The reader is invited to complete the table for K4-15.) y2Xi , 2/3^4 y3X2 ,2/3^3 y 1X3X4 y 2 y 2/i 1X2X3X4 \X0X4 \\X3X4 xxx2 \x\xz 11 3:33:3 V11 ??? Vg V8 Vr Ve K3 V, V, ?; Assuming that (aV, d) is the minimal model of a space A', express y\, z\ and u in terms of Whitehead products of A- Define D : Vo © Vx © Vo —> AV by Dx{ = 0,Dyj = dyj,Dzk = dzk,k ? 1 and Dz\ = y\X\ + X3X4. Prove, by induction on the lower degree, that D extends to a differential on AV such that Du = ???? —2/2- Determine the minimal model of (AV, D) in low degrees. 8. Compute the minimal model of the space A' = (S% V Sf) U[a,[a,b}w}w e8 (in low degrees).
14 Relative Sullivan algebras In this section the ground ring is an arbitrary field Ik of characteristic zero. In §12 we showed how to model commutative cochain algebras (C, d) by Sulli- van algebras, (AV, d). Now any cochain algebra comes equipped with the specific morphism Iz —> (C, d), 1 ?—>· 1. In this section we introduce relative Sullivan al- gebras (B ® AV, d) and use them to model general morphisms (B,d) —¦» (C, d) of commutative cochain algebras. We shall then prove existence and unique- ness theorems in this setting. This will, in particular, eliminate the hypothesis H1 = 0 that was frequently supposed in §12. Definition A relative Sullivan algebra is a commutative cochain algebra of the form (B ® AV, d) where • (B, d) = {B ® 1, d) is a sub cochain algebra, and H°(B) = lz. • V = U V(k), where V@) C V'(l) C · · · is an increasing sequence of graded subspaces such that d : V@) —¦> ? and d : V(k) —¦> ? ® AV(k - 1)., k > 1. The third condition is called the nilpotence condition on d. It can be restated as follows: if we put V(—1) = 0 and write V{k) = V(k - 1) <g> Vk for some graded subspace Vk then ? ® AV(k) = [B® AV(k - 1)] ® AVk, and d : Vk —> ? ® AV(k - 1), fc > 0. Note that we identify ? = B (g) 1 and AV = 1 ® AV. We shall follow this convention throughout the rest of this monograph. However, while (B. d) is a sub cochain algebra it will almost always be the case that the differential does not preserve AV. The sub cochain algebra (B,d) is called the ease algebra of (B®AV,d). A Sullivan algebra is just a relative Sullivan algebra with ? = h. On the other hand if ? : (B, d) —> Ik is any augmentation then applying 3c cEDb — to (B ® AV, d) yields a Sullivan algebra (AV, d). the Sullivan fibre at ?. Now. generalizing the Sullivan models of §12, we consider morphisms of com- mutative cochain algebras ? : (B, d) —> (C, d) such that H°(B) - k and make the definition A Sullivan model for v? is a quasi-isomorphism of cochain algebras m : (B ® AV, d) A (C, d)
182 14 Relative Sullivan algebras such that (B®AV, d) is a relative Sullivan algebra with base (B, d) and m\B = ?. If f : X —> Y is a continuous map then a Sullivan model for Api (/) is called a Sullivan model for /. In the case of the morphism A* —> (A, d) this definition reduces to the defi- nition of a Sullivan model of (A,d). As in that case, we shall frequently abuse language and refer simply to (B <2> AV, d) as the Sullivan model of ?. Also as in the case of Sullivan algebras, the minimal relative Sullivan algebras play a distinguished role: Definition A relative Sullivan algebra (? ® AV.d) is minimal if Imd C B+ <g> AV + B® A^2V. A minimal Sullivan model for ? : (B,d) —> (C,d) is a Sullivan model (? ® AV,d) -=> (C,d) such that (B <g> AV,d) is minimal. Suppose ? : (B, d) —> (C, d) has a Sullivan model m : (B <8> AV, d) -=» (C, d). By definition, H°(B) = h and the isomorphism if (m) identifies ?(?) with ?(?), where ? is the inclusion of ? in B® AV. Since V is concentrated in degrees > 1, it follows that H°(C) = k and ??(?) is injective. Conversely, if H°(B) = H°(C) = k and if ??(?) is injective we shall show that ? has a minimal Sullivan model, determined uniquely up to isomorphism. This depends on another result we shall also establish: any relative Sullivan algebra is the tensor product of a contractible Sullivan algebra and a minimal relative Sullivan algebra with the same base. The importance of relative Sullivan algebras resides in the fact that they provide good models for fibrations. Indeed, suppose ? : X —> Y is a fibra- tion with fibre F and suppose (APL(Y) <g> AV, d) is a relative Sullivan model for APl{p) : Apl{Y) —¦> APl(X)- In §15 we shall show that if Y is sim- ply connected and if one of H?{F;k), Hm(Y;k) has finite type then (AV,d) = &®APL(Y) {Apl{Y) <8>AV, d) is a Sullivan model for F. It is from this that we deduce that the morphism defined in §13(c) is an isomorphism: for minimal Sullivan models (AV,d) of suitable topological spaces X. This section is organized into the following topics: (a) The semifree property, existence of models and homotopy. (b) Minimal Sullivan models. (a) The semifree property, existence of models and homotopy. Let (B ® AV,d) be a relative Sullivan algebra. Multiplication by ? makes (B (g> AV,d) into a left (B, d)-modu\e (cf. §6) and we shall make frequent use of
Sullivan Models 183 Lemma 14.1 (? ® AV, d) is (?, d)-semifree. proof: Write V = \J V(k) as in the definition, and set V(-l) = 0. Write V(k) = V(k - 1) © Vk, and simplify notation by writing ? <g> AV(k) = B{k). Then B(k) = B(k - 1) ® AV* and d:Vk-+ B(k - 1). Hence Assume by induction that (B(k - l),d) is (B, d)-semifree. Then this equation identifies the quotient on the left as (B, d)-semifree for each ? > 1. It follows from Lemma 6.3 that each (B(k),d) and (B(k),d) / (B(k - l),d) are (?,«*)- semifree. Now a second application of Lemma 6.3 shows that (B cg> AV, d) = \Jk (B(k), d) is (B, d)-semifree too. ? A second useful fact is the 'preservation under pushout' of relative Sullivan algebras. Suppose (B <g> AV, d) is a relative Sullivan algebra and ? : (B, d) —> (?', d) is a morphism of commutative cochain algebras with H°(B') = h. Then, immediately from the definition, the cochain algebra (B1, d) ®(B,d) (B ® AV, d) = (B' ® AV, d) is a relative Sullivan algebra with base algebra (B',d). It is called the pushout of (B ® AV, d) along -?. Notice that associated with the pushout is the commutative diagram of cochain algebra morphisms (B,d) (B®AV,d) (B',d) (B'®AV,d). To see that -? ig> id commutes with the differentials, identify ? ® AV = ? ®? (B ® AV) and observe that ' ? ® idjiv =?®? {idB®\v) :B®B{B® AV) —> B' ®B (B ® AV). Now Proposition 6.7(iii) asserts that —®b{B ® AV, d) preserves quasi-isomorphisms because (B <S> AV, d) is (B, d)-semifree. There follows Lemma 14.2 If ? is a quasi-isomorphism so is ? Cg> id : (B <S> AV, d) —> {B'®AV,d). ' D We can now establish the existence of Sullivan models for morphisms: Proposition 14.3 A morphism ? : (B,d) —> (C,d) of commutative cochain algebras has a Sullivan model if H°(B) = k = H°(C) and ??{?) is injective.
184 14 Relative Sullivan algebras proof: Choose a graded subspace Bx C ? so that (BrH = A, {B^1 ®d(B°) = B1 and (Bj)" = B'\ ? > 2. Clearly (i?i, d) is a sub cochain algebra and the inclusion ? : (??, d) —¦» (?, d) is a quasi-isomorphism. In particular the restriction ?\ : (B\,d) —l· (C,d) of ? satisfies: ^(??) is injective. Now because (??)° = Ik the argument of Proposition 12.1 applies verbatim to show the existence of a Sullivan model mi : (B\ ® AV, d) -^> (C, d) for c?x. Thus a commutative diagram of cochain algebra morphisms is given by {B,d)®{Bud){Bx®AV.,d) AV, d) in which ,7B) = 1 ®?? a. This may be rewritten as {B®A\'\d) Lemma 14,2 and the preceding remarks show that ? and (B ® AV, d) is a Sullivan algebra; hence m : Sullivan model for ?. is a quasi-isomorphism AV,d) -^ (C,d) is a ? Finally, we extend the lifting lemma to the relative case and define relative homotopy. To begin, suppose given a commutative diagram of morphisms of commutative cochain algebras (B,d) (? ® AV, d) (A,d) (C,d) in which i is the base inclusion of a relative Sullivan algebra and 77 is a surjective quasi-isomorphism. An argument identical to that of Lemma 12.4 establishes Lemma 14.4 There is a morphism ? : (B(&AV,d) (io extends a) and ?? = ? (? lifts ?). (A,d) such that ?? = ? 0
Sullivan Models 185 We come now to the notation of relative homotopy. Suppose ??,??: (B®AV,d) —)· (A,d) are two morphisms of commutative cochain algebras, in which (B ® AV, d) is a relative Sullivan algebra and ?? and ?? restrict to the same morphism ? : (B,d) —>· (A,d). Generalizing the definition in §12(b) we have Definition ?? and ?? are homotopic rel ? (?0 ~ ?? rel ?) if there is a mor- phism ? : (? ® AK d) —> (.4, d) ® A(i, dt) such that (????)? = ??, (id ??)? = ?? and ?(?) = ? F) ® 1, b e B. The morphism ? is called a relative homotopy from ?? to ??, The identical argument of-Proposition 12.7 shows that homotopy rel ? is an equivalence relation in the set of morphisms ? : (? ® AV, d) —> (A, d) that restrict to a given ? ·, (B, d) —¦» (.4,ef). The homotopy class of ? is denoted by [?] and the set of homotopy classes is denoted by [(B ® AV, d), (A. d)]a. Notice that ? ; (B, d) —¦» (.4, d) makes (.4, d) into a (B, d)-module via b· a = a(b)a. Now suppose ??·,?\ '¦ (B ® AV, d) —¦» (-4, d) both restrict to ?. Lemma 14.5 -??c/? ~ ^? rel i? ?/ien ??~^>\ = hd+dh, where h : B®AV —>· .4 zs ? B-linear map of degree —1. In particular, ?(?0) = ?(??). proof: Let ? : (? ® AV,d) —¦> (.4, d) ® A(i,di) be a homotopy rel ? from <^0 to ??. As in the proof of Proposition 12.8, define h : ? ® AK —¦> .4 by ?B) = pot*) + (^ (z) - c^oW) ? + (-l)degi/iB)cii + ?, where ? € A®(I + dI), I C A(f) denoting the ideal generated by i(l - i). Then, because ?? = ??, we obtain ?0 — ?? = dh + hd. Moreover, since ? restricts to ? in ? it follows that ?(??) = ?(?)?(?), be ?. Hence h{bz) = (-l)dezbb-h(z); i.e., h is i?-linear. D Finally, suppose that (B,d) (A,d) (? ® ??, d) (C, d) is a commutative square of morphisms of commutative cochain algebra, in which • i is the base inclusion of a relative Sullivan algebra, and • ? is a quasi-isomorphism.
186 14 Relative Sullivan algebras Note that we do not require that ? be surjective. The proof of Proposition 12.9 also goes over verbatim to give Proposition 14.6 (Lifting lemma) There is a unique homotopy class rel ? of morphisms ? : (? <%> AV, d) —¦» (A,d) such that ip\R=a and ??^-???? ?. (b) Minimal Sullivan models. Here we establish three basic results. • Any relative Sullivan algebra is the tensor product of a minimal relative Sullivan algebra and a contractible algebra (Theorem 14-9). • A quasi-isomorphism between minimal relative Sullivan algebras is an iso- morphism, provided it restricts to an isomorphism of the base algebras (Theorem I4.II). • Minimal Sullivan models of cochain algebras and of morphisms exist, and are unique up to isomorphism (Theorem 14-12). For ordinary Sullivan algebras these results extend Proposition 12.2 and several results in §12(c) to the case where Hl{—) is not necessarily zero. Before beginning the proofs, we make two small but useful observations. Con- sider first an arbitrary commutative graded algebra of the form ? = Jk&lB'}^, and a graded vector space V = {V'}i>i. Let W C (B ® AV)+ be a graded sub- space such that (B ® AV)+ = W © A^-V ? (B+ ® AV). The inclusions of ? and W extend uniquely to a morphism ? : ? <g> AW —> ? ® AV of graded algebras. Lemma 14.7 The morphism a : B ® AW —¦» ? <S> AV is an isomorphism. proof: It follows from the hypothesis that Vn C Wn + ? ® AV<n, n > 1, and an obvious induction then gives that ? is surjective. Choose ? : V —¦> ? <S> AW so that ?? = id, and extend id? and a to the morphism r = id ? ·? ·' B®AV —> ? 0 AW. Then ?? = id. Now suppose by induction that ? and r restrict to inverse isomorphisms be- tween B®AW<n and B®AV<n. Let w € Wn. Then tw = w'+ u with w' 6 Wn and «e?+® AW<n. Hence w = ???? = aw' + au = w' + au. In other words, w - w' = au e W ? (B+ ® AV<n) = 0 .
Sullivan Models 187 By induction, ? is injective in B<8)AV<n. Hence w = w' and u = 0: i.e., tw — w. This shows that ?? = id in Wn: which closes the induction. ? Next, suppose (B ® AV, d) is any relative Sullivan algebra. Choose a graded vector space fi,c?so that {??)° = k, (??I ? d(B°) = B1 and {??? = B'\ i > 2. Then B\ is preserved by d, (Bud) is a sub cochain algebra and the inclusion in (B, d) is a quasi-isomorphism. Lemma 14.8 T/iere zs ? relative Sullivan algebra {B\ ® AV, d') and an isomor- phism, ?:{?® AV, d') = (B, d) ®[Bl,d) (#i ® AV, d') A (B ® AV, d), restricting to the identity in B. proof: Write F = (J V(jfc) with AV(Jfe), d) = (? ® AV*(A: - 1) ® AV'fc, d) and d : V* —¦> ? ® AV(A: — 1). Assume by induction we have constructed (Bi ®AV(k- l),d') and - 1), d') A (? ® AV(Jb - 1), d). Lemma 14.2, applied to the inclusion (??,?) —>· (B,d) shows that the inclusion (?i (g) AV(a; - l),d') —¦> (? ® AV(A; — l),d') is a quasi-isomorphism. Hence if {va} is a basis of V^ there are d'-cocj-cles za G ?? ® ? F (A: - 1) and elements ya e B ® AV{k — 1) such that di>Q = ?B? + d'ya). Extend d' and ? to Vit by setting d'va = za and ??? = va - ayQ. By construction, ? is a cochain algebra morphism. By induction, it restricts to an isomorphism of B®AV(k— 1) onto itself. Denote the inverse of the restriction by r, extend rto?® AV(k) by setting ??? ~ va + ya, and verify that r and ? are inverse isomorphisms of B (g) AV(k). D Theorem 14.9 Let (B<g>AV, d) be a relative Sullivan algebra. Then the identity of ? extends to an isomorphism of cochain algebras, (B ® AW, d') ® (A(U ? dU),d) ? (? ® AV, d), m ly/izc/i (?® AH7,d') ? ? minimal relative Sullivan algebra and (A(U ? dU),d) is contractible. proof: First consider the case ? = k © {?'};>!. The decomposition (B ® AV)+ = (B+ ® AV) ? A^2V ? V
Igg 14 Relative Sullivan algebras defines a differential d0 : V —¦» Vr by the condition dv-dov G (B+®AV)©A-2V, y c. y? Write V as the increasing union of graded subspaces V(k) such that d : vr(fc) —» ? ® AV(k - 1) and note that d - d0 : V(fc)n —» ? ® ? (V(Jfe - l)-n) . Choose an increasing sequence D"@) C U(l) C · · · C V of graded subspaces such that U(k) e (kerdo ? V(fc)) = V(k). (Since t/(fc - 1) ? (kerd0 ? V(fc)) = 0. i/(fc - 1) can be enlarged to the direct summand U(k).) Set U = \Jk U(k). Then U ? kerdo = V and so we may write V = t/ ? d0U ? IK d0 : U A do I/, d0 : W —>· 0. This construction has the following useful property: for elements u, u\ G U and -u; G H7, u\ + dou + w G K(fe) => m G V(k). A4.10) Indeed, we may write u\ + dou + w = u' + z' with u' G U(k) and z' G kerdo- Thus u? — u' = z' — dou - w G U ? kerd0 = 0, and u\ = u'. Now consider the subspace U ? dU ? W C ? ® AV. Apply Lemma 14.7 to conclude that this inclusion, together with ???-, extends to an isomorphism of graded algebras, ? : ? (g) AU ® Ad?7 ® AH7 A?® AV. Denote the contractible algebra AU ® Adi7 simply by ?, and let d = ?~??? be the induced differential in ? <g> ?" ® AH7. Use ? to identify the two cochain algebras and so regard I7 as a subspace ?? ? <& ? <E> AW. Our next step is to exhibit the inclusion (B, d) ® (?, d) —> (B ® ? ® AW, d) as the inclusion of a relative Sullivan algebra. Define an increasing sequence of graded subspaces W@) C W(l) c · - - C W by W@) = {w e W \ dw e ? <S> E} and W(t) = {w e\V \dw e B®E® AW(i - 1)}, ? > 1. Set ? = \J( W{i). It is enough to show ? = W. Assume V<n C ? ® ? ® AZ. We show first by induction on k that V(k)n CB&E&AZ. I21 deed if u G V@)" we have dv G ?, so that do ? = 0 &?? ? = dou + w, some u G Un~x, w G II7". As observed above, (do - d)u G ? ® AV<n, and hence (do - d)tx G ? ®E® AZ. It follows that d(dou) = d(do - d)u G ? ® ? ® AZ.
Sullivan Models 189 But dv G B and so dw = dv - dd^u G B ® ? ® AZ. In particular, we must have dw 6 B®E®AW(?), some ?, and so w G IF(?+ 1). On the other hand, we have just seen that (do-d)u C B®E®AZ. Since du ? ? by definition, it follows that dou? B®E®AZ. Thus ? -dou + w 6 5® ?® AZ; i.e., F@)n C B®E®AZ. Suppose V(k— 1)" C i?<g>.E® AZ. Let ? 6 V(A:)n and write ? — U\ +dou + w, with ui € Un, u e Un~l and ? G TV". As observed in A4.10), then ux G lr(fc). Hence (d - do)ux C. ? ® i\V{k - \)^n C ? ® ? ® AZ. The same argument as just above for k = 0 shows that (d - do)dou = d(d0 - d)u S ? ® ? ® AZ. Hence dw = (d- do)w ? ? ® ? ® AZ and so, as in the case k — 0, it follows that ? 6 Z. As in that case, dou e B&E&AZ, while Ux e ? by definition. This shows that V(k)n C B®E®AZ. It follows by induction on k that Vn C B®E®AZ, and so a second induction on ? shows that V C ? ® ? ® AZ. This implies that ? ® ? ® AZ is all of ? ® E® AW, and so^= f-F, as desired. Finally, let ? : ? —-> Jfc be the augmentation. Then id ·? : (B,d) «8> (?,d) —-> (B,d) is a quasi-isomorphism. Hence it defines a suijective quasi-isomorphism id -? ® id : (? ® ? ® AW, d) —-*· (? ® AH': d')> as described in Lemma 14.2. In particular, (B ® AW, d') is a relative Sullivan algebra and, since d : II' —->¦ B+®E® AW + {B®E® AW)+ ¦ (? ® ? ® AW)+, it follows that d' : W —>· 5+® ??^??^??; i.e., (B®AW,d') is minimal. Apply the Lifting lemma 14.4 to extend id ? to a morphism ? : (? ® AW, d') —> (B ® ? ® AIF, d) such that (id -c ® id)^ = id. Then ?-w—w G ?®J5+<8>AVF. This implies (Lemma 14.7) that an isomorphism (B ® AW, d') ® (A(U © dU)) ? (? ® AV, d) is given by ? ® ? ·->¦ ?(<^? · ?). This completes the proof of.the theorem in the case B° = k. Finally, suppose (B ® AV, d) is a general relative Sullivan algebra. Write (B<S>AV,d) = (fi,d) ®(Blid) Ei ® AV",d), as in Lemma 14.8, with {B^0 = k. By what we have just proved, (Bi ®AV,d) s (?? ®AW,d')®A(U@dU). Thus (? ® AV, d)9i(B® AW, d') ® A(U ? dU). D Next, consider a cochain algebra quasi-isomorphism ? : (?' ® AV',d) ? (? ® AV, d) between minimal relative Sullivan algebras.
190 14 Relative Sullivan algebras Theorem 14.11 If ? restricts to an isomorphism t)q : B' —*¦ ? then ? itself is an isomorphism. proof: Consider the diagram (B' ®AV',d) (B,d) (B®AV,d) We apply an argument of Gomez-Tato [65] to extend ??? to a morphism 7 : (B <g> AV, d) —¦» (B' ® AV, d) such that 777 = id. By the very definition of minimality, d : V —¦» B+ ® AV ? A^2V. Thus if F = UA, v(fc) with F(Jfe) = V(k -1) ? Vfc and d : Vk —¦> ? ® AF(jfc -1) it follows that d : V? —¦> 5 ® AV<n ® A (V(k - 1)"). Thus to construct 7 it is enough to assume it has been defined in .4 = ? ® AV<n ® A (V(fc — 1)") and to extend it to A®AV?. Observe that ? factors to give a quasi-isomorphism of cochain complexes, B' ® AV ? B®AV (we are dividing by graded subspaces, not ideals). The cochain complex ? ® AVI A contains no elements of degree ? — 1, and hence no coboundaries of degree n. Thus every cocycle of degree n is the image, under 77, of a cocycle of degree ? in (B' ® AV')/j(A). In particular, if {uQ} is a basis of Vfcn then there are elements xa G ?' ® AV' and elements aQ,a^ G ? such that rya;Q = uQ + aa and dxa = 7(a'a). Hence dva = ??(?? — ???) = a'a — daa. Extend 7 to V? by putting ??? = ?? — ??<*· This completes the construction of 7. Since 7/7 = id, 7 is also a quasi- isomorphism. Now the same argument applied to 7 gives a morphism ? : (?' <8) AV',d) —¦> (B (g) AV,d) such that 7? = z'd. Thus 7 is both injective and surjective; i.e., it is an isomorphism. Since 7/7 = id, ? is the isomorphism 7. D Finally, suppose ?: (B,d) —)· (C,d) is a morphism of commutative cochain algebras such that H°{B) = h = H°(C) and ??{?) is injective.
Sullivan Models 191 Theorem 14.12 The morphism ? has a minimal Sullivan model m:(B®AV,d) -=>(C,d). If m! : (B <g>AV,ef) -^» (C,d) is a second minimal Sullivan model for ? then there is an isomorphism a:(B® AV, d) A (B ® AV', d) restricting to id?, and such that m'a ~ mrel B. proof: In Proposition 14.3 we showed ? had a Sullivan model, and in Theo- rem 14.9 we showed that this is the tensor product of a contractible algebra and a minimal relative Sullivan algebra. Thus ? has a minimal Sullivan model. Given two such models we may apply Proposition 14.6 to the diagram (B,d) (B®AV',d) (B®AV,d) = - (C,d) to extend id? to a morphism ? : (? <g> AV, d) —¦» (? <g> AV, d) such that m'a ~ mrel B. Now Theorem 14.11 asserts that a is an isomorphism. D Corollary Any commutative cochain algebra (A, d) satisfying H°(A) = .& and any path connected topological space X have a unique minimal Sullivan model. Finally, let ? : (AV, d) —y (AW, d) be an arbitrary morphism between Sullivan algebras. Recall that d0 : V—> V and do : W —> W denote the linear parts of the differentials (§12(a)) and that the linear part of ? is a morphism of complexes Q(<p):(V,do)->(W,do). Proposition 14.13 If ? : (AV,d) —> (AW,d) is a morphism of Sullivan alge- bras then ? is a quasi-isomorphism if and only if H(Q(ip)) is an isomorphism. proof: ' WriteJAV, d) = (AV, d) ® (E, d) and (AW, d) = (AW, d) ® (F, d) with (AV, d) and (AW, d) minimal Sullivan algebras and (E, d) and (F, d) contractible Sullivan algebras. Let ? be the composite (AV, d) —> (AV, d) -A (AW, d) —t (AW,d). Then we may identify ?(?) = ?(?) and H(Q(tp)) = Q(^). But by Theorem 14.11, ? is a quasi-isomorphism if and only if it is an isomorphism. If -0 is an isomorphism so is B(?). Conversely, suppose <^(?) is an isomorphism, then ¦? induces isomorphisms A^mV/A>mV -=> A^mW/A>mW. By induction it in- duces isomorphisms iW/A>mV A AW/A>mW. Since (AV)k = (AV/A>mV)k for m > k it follows that ? is an isomorphism. D
192 14 Relative Sullivan algebras Example 1 The acyclic closure, (B®AV,d). Suppose (B,d) is an augmented commutative cochain algebra such that H°(B) = k and Hl{B) - 0. A minimal Sullivan model (B <g> AV,d) -=> Jc for the augmentation ? : (B,d) —> k is called an acyclic closure for (?,d). We show that the quotient differential d in AV is zero. Form the relative Sullivan algebra (B ® AV ®B ? ® AV, d) = E®AV®AV, d). The inclusions of the left and right tensorands are the base inclusions ??,?? : (? ® AV, d) —> (? ® ? F ® AF, d) of relative Sullivan algebras. Hence ? ®\0 - and ?<8>?? ~ are surjective quasi-isomorphisms from (??®AV®AV, d) to (AV, d). Let ? : (AV, d) -=* (B <8> AV ® AV', d) be a quasi-isomorphism such that {e ®Aj —) ? a = id (Lemma 12.4). Then (? ®a0 -) ? ? is a quasi-isomorphism of the minimal Sullivan algebra (AV, d). Hence (Theorem 14.11) it is an isomorphism. Denote by ? the inverse isomorphism, and consider the composite morphism ? : (AV,d) A (?®AV®AV,d) g®'d®id) (AV,d)®(AV,d) ^^4 (AV,d)®(AV,d) . By construction, for ? ? A+V we have ?? = ?®\-{-?+1®?, some ? G ?+ V ® A+V. Hence for any cocycle z, ?(?)[?] - [?] ® 1 - 1 ® [z] In Example 3, §12(a), we showed that a certain algebra H*(X; ]k) was a free graded commutative algebra using by the existence of a morphism H*(X; 3c) —>¦ H*(X; k) ®H*(X; k) satisfying this condition. That argument now shows that H(AV, d) is a free graded commutative algebra. AW. Define a quasi-isomorphism ? : {AW, 0) -=> (AV, d) by sending a basis of W to representing cocycles in AV. Then ? is an isomorphism (Theorem 14.11). D Example 2 A commutative model for a Sullivan fibre. Let ? : {B,d) —> (-4,d) be a morphism of commutative cochain algebras both of which satisfy H°{-) = k and &{—) = 0. Extend ? to a minimal Sullivan model m : {B ® AW, d) -?» (A, d) and recall that (AW, d) = k ®B {B ® AW, d) is the Sullivan fibre of ? at an augmentation ? : ? —> le. We construct a commutative model for (AW,d) as follows. Let (?®AV,d) be an acyclic closure for {B,d). Since — ®# {M,d) preserves quasi-isomorphisms for any (?,d)-semifree module {M,d) — cf. Proposition 6.7 — it follows that {A ® AV, d) ^HL (jB ? AV) ®B {B 0 AW) -^4 (AW, d) are quasi-isomorphisms; i.e. (A ® AV, d) = ,4 ®b {? ® ?V, d)
Sullivan Models 193 is a commutative model for the Sullivan fibre (AW, d). ? Exercises 1. Let ? : (B<2)AV,d) ->· (B®AW,d) be a morphism of relative Sullivan algebras which restricts to the identity on ? and let (B®AW, d) ^-T1 (.4, d) ? (.4', d') be morphisms of commutative differential graded algebras. Assume that ? and ? are quasi-isomorphisms and that ?? , ?\ restrict to the same morphism (B. d) —>· {A,d). Prove that the following assertions are equivalent: (i) v?o ~ 9i rel B, (ii) ??$ ~ ?^?? rel B, (iii) ??0 ~ ??\ rel B. 2. Let (AV,d) be a Sullivan minimal algebra. a) Fix an automorphism ? of (AV,d) such that for each x G (AV,d) there is some ? (depending on x) such that (? — idAv)n(x) = 0. Prove that ? = (? - idAy)n is a derivation of (AV,d) , homogeneous of degree 0 ?—' ? k-\ which commutes with d and that for each x G (AV. d) there is some ? (depending on x) such that ??(?) = 0. OO . b) Prove that ? = - Y^ —?? is a linear map which commutes with d and that ?? = ?? = ? — id. c) Prove that (Au ® AV. D) with u of degree 1 and Dv = u ® #(?) + 1 ® dw, is a minimal model of the kernel of e.\ · id — e0 · ? ¦ A(t, dt) ® (AV, d) ->· (AV. d). 3. Let (? ® AV, (i) be a relative Sullivan algebra and consider the contractible cochain algebra E(sV) = AsV <g> AV where V1 = V\ a) Prove that (B ® AV. ii) ® 5(sV) = (? ® A(V ? sV @ V), D) is a relative Sullivan model. b) Prove that the natural inclusion ?0 : (B®AV,d) c—>(B®A(V@sV@V),D) is a quasi-isomorphism. c) Consider the degree -1 derivation S defined in (B ® A(V © sV ? V),D) by S{x) = sx,x GV and S{B) = S{sV) = S{V) = 0. Prove that ? = SD + DS ~ ?n is a degree 0 derivation and that e6 — > —- is an automorphism of differential . n! n=0 graded algebras. OO Qj d) Prove that ?? = eoAo satisfies ?? (?) = ? + ? + V] — —E ? d(x)). i=o (^ + 1)! e) Let <^o , ?? : (? ® AV. d)^>(A. dA) be morphisms of commutative differential graded algebras which restrict to the same morphism (B, d) ->· (.4, dA). We write / « <7 if there exists a morphism of differential graded algebras ? : (? ? A(V ® sV ? V),Z)) ->· (.4,6?^) such that ??? = (/?o and ??? = ??. Prove that « is an equivalence relation and that ?? ~ ?\ implies that ?(?0) = ?{??).
194 14 Relative Sulliva f) Assume ? — Ik — H°(A). Prove that ?0 m ?? if and only if ?0 this result extend to the case H°(B) = Ik = ?°(?)?
15 Fibrations, homotopy groups and Lie group actions In this section the ground ring is an arbitrary field fc of characteristic zero. In this section we see how relative Sullivan algebras model fibrations. In par- ticular, if / : ?" —> Y is a continuous map with homotopy fibre F we construct a Sullivan model for F directly from the morphism Api(f): Apl(Y) —> Apl(X), provided Y is simply connected with rational homology of finite type. We use this to construct Sullivan models for many more interesting spaces. We are also able, at last, to establish the isomorphism (promised in §12 and §13(c)) between the generators of a minimal model and the dual of the homotopy groups. Finally we apply Sullivan models to the study of principal bundles and group actions of a path connected topological group G. Our main focus is on the case when H*.(G;]k) is finite dimensional, which includes all connected Lie groups. In particular, we use Milnor's universal bundle (§2) to obtain a simple form for the Sullivan model of any principal bundle. We also consider models for group actions and see, for example, rational homotopy reasons why spaces may not support free actions of groups such as 53 x SUC). Much of the material was originally developed by H. Cartan, Koszul and Weil in the context of smooth principal bundles, with the aid of principal connections and the curvature tensor. This was described in three lectures [33] [34] [102] given in Brussels in 1949 and provided one of the major clues that led to Sul- livan's introduction of minimal models. Indeed, the main result announced in Koszul's. lecture is the construction of (what we now call) a Sullivan model for a homogeneous space. This section is organized into the following topics: (a) Models of fibrations. (b) Loops on spheres, Eilenberg-MacLane spaces and spherical fibrations. (c) Pullbacks and maps of fibrations. (d) Homotopy groups. (e) The long exact homotopy sequence. (f) Principal bundles, homogeneous spaces and Lie group actions. (a) Models of fibrations. Consider a Serre fibration of path connected spaces ? : X —> Y,
196 15 Fibrations, homotopy groups and Lie group actions whose fibres are also path connected. Let j : F —> X be the inclusion of the fibre at y0 6 Y· Then APl converts the diagram ? _i_4 ? ApL(F) 4^ APL{X) to I Mpi.(p) A5.1) yo >Y A i—-—APL(Y) . where ? is the augmentation corresponding to yo- Observe that H1 (Apl(p)) is injective. Indeed, since F is path connected it follows from the long exact homotopy sequence that ??? (?) is surjective (Propo- sition 2.2). Hence H\{p:Z) is surjective (Theorem 4.19). But Hi A'; k) = Hi (Y; Q) <8>q k = Hi {Y; Z) ®z k , and so Hi{p\h) is also surjective. Thus the dual map, ?1 (APi(p)) = Hl{p;h) is injective (Proposition 5.3(i)). Since H1 (APl(p)) is injective, Proposition 14.3 asserts the existence of a Sullivan model for p, m : (APL(Y) ® XV,d) -^ APL{X). (In fact (Theorem 14.12) there is even a minimal Sullivan model for p.) The augmentation ? : ???(?) —y h defines a quotient Sullivan algebra (AV,d) = k ®apl(y) {Apl{Y) ® AF,d), which is called the fibre of the model at yo- Since APl(j)APl(p) reduces to ? in Apl(Y), APL(j)m factors over ? ¦ id to give the commutative diagram of cochain algebra morphisms APL{F) . AplU) APL{X) A5.2) (AV,d) ' ^.d (APL(Y)®AV,d) We show now that, under mild hypotheses, m is a quasi-isomorphism. Thus in this case m : (AV, d) -=> APl (F) is a Sullivan model for F: the fibre of a model is a model of the fibre. Theorem 15.3 Suppose Y is simply connected and one of the graded spaces H*(Y;k), Hm{F;k) has finite type. Then m: (AV,5) —> APL(F) is a auasi-isomorphism.
Sullivan Models 197 proof: Suppose first that ? is a fibration. We wish to apply Theorem 7.10, with diagram A-5.2) corresponding to diagram G.9). For this we need to verify two things: first, m has to be an .4p?,(F)-semifree resolution and second, diagram A5.1) has to be weakly equivalent to the corresponding diagram with C*( —) replacing Api(-). But the first assertion is just Lemma 14.1 and the second follows from the natural cochain algebra quasi-isomorphisms C*(—) -^ · <^- Apl(-) of Corollary 10.10. Thus we may apply Theorem 7.10 and it asserts precisely that mis a quasi-isomorphism. Now suppose only that p. is a Serre fibration. In the diagram constructed in §2(c), X Y X ? ? MY , ?? = (?, const, path at px), ? is a map from a Serre fibration to a fibration. Since ? is a homotopy equiva- lence it restricts to a weak homotopy equivalence ? : F -=·» ? ? ? ?? (Proposi- tion 2.5(?)). Moreover, .4??,(?) is asurjective quasi-isomorphism. Apply the Lifting lemma 14.4 to the diagram APL(Y) ApL(q\ APL(X xY MY) ApL(X) (APL(Y)®AV,d) Apl(X) to construct the quasi-isomorphism ? extending Api{q). Then ? : (AV,d) —> Apl(X xy PY) is a quasi-isomorphism by the argument for fibrations, and m is the quasi-isomorphism Apl(X)u. ? Remark In [82] the theorem is proved under the weaker hypothesis that ?? {?) acts locally nilpotently in each Hi(F; Jk). It is in fact possible to show that this hypothesis and the hypothesis that one of H,{Y\ Je), H*(F; le) has finite type are both necessary for the conclusion of the theorem to hold. ? More generally, suppose that f:X Y is an arbitrary continuous map from a path connected topological space X to a simply connected topological space Y. Suppose further that H*(Y;lk) has finite type. Since Y is simply connected if1 (!';.&) = 0 and H1 (APL(f)) is injective. Thus / has a Sullivan model m : (APL(Y) ® AV,d) A APL(X),
198 15 Fibrations, homotopy groups and Lie group action; as above. In this case the argument in the proof of Theorem 15.3 shows that th< fibre (AV,d) at yo is a Sullivan model of the komotopy fibre of f. Return to the situation of the Serre fibration ? : X —> Y with which thi; topic began, and recall the notation of A5.1). Theorem 15.3 establishes a fun damental property for relative Sullivan models (Apl(Y) ®AV,d) -^ Apl(X) It is, however, often useful to replace ApL(Y) by a Sullivan model, and this i: done as follows. ^ Choose a Sullivan model my : (AVy,d) -^> Apl(Y). Since Vy = {VY}i>: by definition, there is a unique augmentation ? : (AVy,d) —> k. Construct ? commutative diagram of cochain algebra morphisms APL(Y) ApL(p] > APL(X) AplU) > APL(F) A5.< (AVy,d) (AVy®AV,d) g .d » (AV,d) by requiring that • i is the inclusion of a relative Sullivan algebra, and • m is a Sullivan model for the composite ApL{p)my. Then Apl(j)tti factors over ? ¦ id to yield m. As in Theorem 15.3, we now restrict to the case that Y is simply connectee and one of H*(F;Jk), H*(Y;]k) has finite type. With these hypotheses we have Proposition 15.5 The three morphisms my, m and m in A5.4) are a^ Sul- livan models. proof: (i) The morphism my. This is a Sullivan model by hypothesis. (ii) The morphism m. This is a quasi-isomorphism by construction Thus we have only to exhibit (AVy <g> AV. ci) as a Sullivan algebra. Put W = Vy © V and define an increasing sequence of subspaces 0 = W(—l) C W@) C • · · W by setting W{i + 1) = {w € W | dw 6 AW(?)}. It suffices to show that W = \JtW(?). Now Vy = Ufc VY(k) and V = \Jk V(k) with d : Vy(k) —> AVY(k - 1) and d : V(k) —> AVy ®AV(fc-l). Thus VY{k) c W(k). If V(k-l) c {J(W{i) and ? e V(k) then since dv ? AVY <8> AV(k — 1), dv is in some AW(t). Hence ? is in some W(e+l) and V(k) C \JW{i) as well. (iii) The morphism m. Write APL{Y) ®(Avv,d) (AVV ® AV,d) = (APL(Y) ® AV,d) . This is a relative Sullivan algebra, and m factors as (AVy®AV,d) ^^ {ApL(Y)®AV,d) ^PL(p)m) APL(X).
Sullivan Models 199 But my.<8> id is a quasi-isomorphism (Lemma 14.2). Hence so is Api(p) ¦ m. Now apply Theorem 15.3. ? Since the morphism ApL(p)my has a minimal Sullivan model (Theorem 14.12) the relative Sullivan algebra (AVy,d) —> (AVy <8> AV,d) may be taken to be minimal. In this case (AV,d) is minimal and m : (AV, d) -=» Apl(F) is the minimal· model of F. Thus Proposition 15.5 has the: Corollary Suppose (AVy,d) is a Sullivan model for Y and (AV, d) is the mini- mal Sullivan model for F. Then X has a Sullivan model of the form (AVy<g>AV, d) in which (AVy,d) is a sub cochain algebra and dv — dv G A+Vy <g> AV", ? G V.D Note: (AVy <8> AV, d) is minimal as a relative Sullivan algebra, but it is easy to make examples in which it is not minimal as a Sullivan algebra. Proposition 15.5 has an important converse. Again suppose ? : X —> Y is the Serre fibration with which this topic began. Let my : (AVy,d) -=* Apl(Y) be a Sullivan model, but now suppose given • a relative Sullivan algebra (AVy,d) —> (AVy <g> AW,d). • a cochain algebra morphism ? : (AVy <g> AW,d) —> Apl(X) that restricts to ApLip)my in (AVy,d). Then, as above, ApLU)n factors over ? · id to yield ? : (AW, d) —> Apl(F). As in Theorem 15.3 we suppose now that Y is simply connected and that one of H,(F;k),H*(Y;R) has finite type. Proposition 15.6 If ? is a quasi-isomorphism then so is n, i.e., ? : (AVy ® AW,d) —> APL(X) is a Sullivan model for X. proof: In the proof of Proposition 15.5 we showed that (AVy <g> AW, d) was a Sullivan algebra. Thus it suffices to show that ? is a quasi-isomorphism. Let m : (AVy <g> AV,d) -=» APL(X) be the Sullivan model of diagram A5.4). Use Proposition 14.6 to construct a morphism of cochain algebras ? : (AVy ® AW,d) —> (AVy ® AV,d) such that ? = id in AVy and ??? ~ ? rel (AVy,d) . Now apply k ®AVy - to obtain a morphism ? : (AW,d) —> (AV,d) such that ??? ~ n. Thus ?(??)?(?) = H(n). Since H(m) is an isomorphism (Proposition 15.5) so is ?(?). Put I(k) = (A^kVy ® AW,d) and J(k) = (A^Vy <8>AV,d). Then ? restricts to maps I(k) —» J(k). The induced maps I(k)/I(k + 1) —> J(k)/J(k + 1) have the form id®Tp: (AkVy,d0)®(AW,d) —> (A*Vy,d0) ® (AV,d),
200 15 Fibrations, homotopy groups and Lie group actions where do '¦ AVy —> AVy is the 'linear part' of d. Thus these maps are all quasi-isomorphisms. An obvious induction on k now shows that ? induces quasi-isomorphisms 0(k) : (AVY ® AW,d)/I(k) -=» (AVY ® AV,d)/J(k) for k > 1. Since Vy = {VY}i:>1 by the definition of a Sullivan algebra, I(k) and J(k) are concentrated in degrees > k. Thus we may identify ??(?) = Hl(9(k)) for i < k, and so ? is a quasi-isomorphism too. Since ??? ~ n, so is n. ? (b) Loops on spheres, Eilenberg-MacLane spaces and spherical fibra- tions. We apply the results of §15(a) to compute explicit Sullivan models in some important, but elementary, examples. Example 1 The model of the loop space QSk , k > 2. Let ? : PSk —> Sk be the path space fibration, with fibre QSk. Ifk is odd the minimal Sullivan model for Sk has the form ??$ : (A(e),0) -^ ^^E^). Define m : (A(e,ti),du = e) —> APL{PSk) by me = ApL(p)mse and mu = ?, where ? is any cochain satisfying d$ = -^pl(p)ttis&· (Since PSk is contractible any cocycle of positive degree is a coboundary.) By inspection, m is a quasi-isomorphism. Hence it follows from Proposition 15.5 that m factors to yield a minimal Sullivan model Thus H'(QSk; Jk) is the polynomial algebra on a class [u] of even degree k — 1. // k is even then the minimal Sullivan model for Sk has the form m$ : (A(e,e'),de' = e2) —> ApL(Sk). In this case ApL{p)ms extends to a quasi- isomorphism m : (A(e,e,u,u'), du = e, du' - e' - eu) —» APL(PSk). Thus a minimal Sullivan model for ClSk is given by m : (?(??,??'),?) -=> -4??(?5?). In particular H'(USk;]k) is the tensor product of the exterior algebra on the class [u] and the polynomial algebra on the class [u1]. Here deg[u] = k - 1 and deg[u'] = 2k - 2. Note that whether k is even or odd the Hubert series (§3(e)) of H*(QSk;Jk) is given by oo „ ' 1 — Zk~1 '
Sullivan Models 2°1 Example 2 The model of an Eilenberg-MacLane space . Let X be an Eilenberg-MacLane space of type (?, ?), ? > 1. Thus (cf. §4(f)) Wi(X) = 0 for i ? ? and there is a specified isomorphism ??(?) = ?. Assume ? is abelian (this is automatic for ? > 1) and set Vn = ????(?, k). The Hurewicz theorem 4.19 asserts that Hi(X;Z) = 0, 1 < i < ? and that the Hurewicz map is a natural isomorphism ? —> Hn(X; Z). This extends to an isomorphism ? ®z k -=» Hn{X;Jk), which dualizes to an isomorphism Vn A- H"(X; k). Now suppose ? ®% k is finite dimensional. We shall show that the minimal Sullivan model of X is given by m: (AVn,0) -=> APL(X) where m induces the isomorphism Vn = Hn(X\k) above. This is clearly equiv- alent to the classical assertion: H*(X;k) is the exterior algebra on Hn{X;k) if ? is odd and is the polynomial algebra on Hn(X;k) if ? is even. The proof is by induction on n, beginning with ? = 1. Let a\,...,ar G ?? represent a basis of ? ®z k. These elements define a homomorphism a : ? ? • · · ? ? —> ?. By Propositions 4.20 and 4.21 there is a continuous map / : K{Zr, 2) —> ?(?,2) such that ??2(/) = a. Hence ?»(/) ®21· is an isomorphism, and so 7?*(/) ®zQ is an isomorphism. Now apply Theorem 8-6 to conclude that -ff*(H/;Q) is an isomorphism, whence also H*(uf;k) is an isomorphism. But QK(n,2) is an Eilenberg-MacLane space of type (?, 1) and ??"(??,2) is an Eilenberg-MacLane space of type (Zr, 1). Thus if ?' is any Eilenberg- MacLane space of type (?, 1), X has the weak homotopy type of ??'(?,2), by Proposition 4.21. Similarly S1 ? · · · ? 51 has the weak homotopy type of QK(Zr,2). It follows that H*(X;k) ? H'iS1 ? · · · ? S1; A); i.e. is the exterior algebra on r generators in degree one. Now assume ? > 2 and that our result is established for ? — I. Let A' be an Eilenberg-MacLane space of type (?, ?) with ? <g>z k finite dimensional. Then, as above, UX is an Eilenberg-MacLane space of type (?, ? - 1). By induction UX has a Sullivan model of the form (At/",!)) with Un~l = HomzGr,$:). In particular, H,(ClX;k) has finite type. Let (AW, d) be a minimal Sullivan model for X. Since H,(CtX;k) has finite type we may apply Proposition 15.6 to obtain a quasi-isomorphism of the form (AW ® Wn-\d) A APL(PX). Moreover, since Hi(X;Z) = 0, 1 < i < n, it follows that ?^?-,?) = 0, 1 < i < n. This implies (Proposition 12.2) that Wi = 0, 1 < k < n. By minimality, d = Qin Wn. On the other hand, the quasi-isomorphism just above shows that H(AW^AUn-1,d) = k. Thus (AW ® AUn~\d) is a contractible Sullivan algebra. It follows that d : Un~l A W and so W - Wn. D Example 3 The rational homotopy type of K(Z,n).
202 15 Fibrations, homotopy groups and Lie group actions Let K{Z,n) denote an Eilenberg-MacLane space of type (Z,n), and let an : Sn —i K(Z,n) represent a generator of ?? (?(?, ?)) = ?. Then by the Hurewicz theorem 4.19, Hn{an;Z) : Hn(Sn;Z) A Hn{K(ix,n);Z). Hence Hn(an;Q) is also an isomorphism. Moreover, for ? > 2, ??? : USn —> CiK{Z,n) = K{Z,n— 1) and ??_?(???) is an isomorphism, by the long exact homotopy sequence applied to the path space fibrations. Hence, as above, Hn~1(Clan;Q) is an isomorphism. The computations of Examples 1 and 2 above now show that H* (a2n+i;Q) and H* (Qa2n+i;Q) are isomorphisms, and so the Whitehead-Serre theorem 8.6 asserts that: a2n+i : S2n+1 —» K(Z,2n+l) and ??2?+1 : ?52?+1 —> K(Z,2n) are rational homotopy equivalences. The reader is cautioned, however, that these maps are far from 'integral' homo- topy equivalences. This illustrates the difference in complexity between integral and rational homotopy theory. ? Example 4 Spherical fibrations. A spherical fibration is a fibration ? : X -> Y whose fibre has the homotopy type of a sphere Sk. Suppose given such a fibration with simply connected base Y. If k is odd then the minimal model of Sk has the form (A(e),0). Hence we can apply Theorem 15.3 to obtain a model for ? of the form (Apl(Y) ® A(e),d) A APL(X), de = ? e APL(Y). If k is even then the model of Sk has the form (A(e,e'),de' =¦ e2). Thus ? has a model of the form (Apl(Y) <g> A(e.e'),d) with de G ???{?) and de1 = e2 + ? <8> e 4- b for some elements o,6 € Apl(Y). The condition d2e' = 0 now implies that 2de = -da. Replace e by e + ^a to obtain a model in which de = 0 and de' = e2 + z, some ? € Apl{Y)· In summary, ^.^^(p) has a model of the form (APL(Y)®A(e,e'),d) ^ APL(X), * = ° de' = e2 + ?, ? G ^??(?). Note that in both cases ? is a cocycle in Apl (V) whose cohomology class [z] is determined by the fibration. In particular this class is zero if and only if Apl (X) is weakly equivalent to Apl(Y ? Sk). However in the case of even spheres an easy calculation shows that H*(X;&) S H*(Y;k) ® H'(Sk) as H*(Y;k)-modules. Finally suppose the spherical fibration arises as the unit sphere bundle of a vector bundle ? : ? —> Y of rank k + 1. (For facts about vector bundles and characteristic classes the reader is referred to [128] and [69] [70].) If k is odd the class [2] is the Euler class of ?, essentially by definition. If k is even then [2] = \p2kfe), where ?2?;(?) denotes the 2&th-Pontrjagin class of ?. In fact, we can use ? : X -> Y to pull the vector bundle ? back to
Sullivan Models 203 a vector bundle ?? over ?, and ?? is the direct sum of a trivial line bundle and a vector bundle ? of rank k. Let ? denote the Euler class of 77. Then ?2 = p2k(v) = H*(p)p-2h{0· Moreover, if 5* is the fibre of ? : Ar -» Y at y, then 77 restricts to the tangent bundle of S?. Hence, by a result of Hopf, (?, [Sy]) = 2. On the other hand, it follows directly from the model that H*(p)[z] = [e]2 with ([e],[S?]) = 1, and that this condition determines [z] uniquely. Hence Example 5 Complex projective spaces. The inclusions S1 C S2 C · · · C Sr C · · · define a CW complex S°° = \JSr, r and it follows from the Cellular approximation theorem 1.2 that 7rrE°°) = ??E?+1) = 0, r > 0. Regard 52n+1 as the unit sphere in Cn. Complex multi- plication defines a free action of S1 on each 52n+1 with orbit space the complex projective space CPn. Similarly S1 acts on S°° with orbit space CP°°; and the 51-bundle 52n+1 —> <CPn is the restriction of the S^bundle S°° —> CP°°. From the long exact homotopy sequence deduce that CP°° ~ K(Z,2) and hence that i/*(CP°°;Q) = Au, with degu - 2 (Example 2, above). Similarly, 7r«(CPn) <8> Q = Qu © Qx with degu = 2 and degx = 2n + 1. Since CPn is a 2n-dimensional CW complex it has no cohomology in degree 2n + 2. Thus· its minimal Sullivan model must have the form A.(u,x;dx = un+l). In particular, CPn is formal with cohomology algebra Au/un+1. ? (c) Pullbacks and maps of fibrations. Suppose given any commutative square of continuous maps ? —s— x ?-?-y in which p and q are Serre fibrations, ? and X are path connected and ^4 and Y are simply ?onnected. Choose basepoints ao and yo so that /(ao) = yo and denote by g : F —> F the restriction of g to a map from the fibre of q at ao to the fibre of ? at yo·· Finally, assume that H,(—\ h) has finite type for one of F, A and for one of F, Y. Choose Sullivan models ??? : (AVy,d) —> Apl(Y) and nA : (AWA,d) -=> APL(A), and let ?: (AVY,d) —? (AWA,d) be a Sullivan representative for /; i.e., ??? ~ APL(f)mY. The Sullivan models extend (A5.4) and Proposition 15.5) to commutative diagrams
204 15 Fibrations, homotopy groups and Lie group actions APL(Y) APL(X) APL(F) and (.\Vy,d) —- (AVY®AV,d) —^ (AV,d) ApL(A) -^^l APL(Z) - APL(F) A5.7) (AWA,d) — (AWA®AW,d) —* (AW,d) in which all the vertical arrows are Sullivan models. Suppose first that we have been able to choose ? so that ??? = ApL(f)my . Then the pushout construction of §14(a) yields the morphism ? = ApL{q)nA-APL{g)m : (AWA,d) ®(Aiv,<0 (AVV ® AV'.d) —> APL(Z) . This may be written as ? : (AWA <8> AV,d) —> APl(Z), and fits in the commu- tative diagram Apl(A) Apl(Z) Apl(F) {AW? ® ??', d) {AV,d) Thus we may apply Proposition 15.6 to deduce Proposition 15.8 If H*(g) : H*(F;k) —» H*(F;lc) is an isomorphism (in particular if ? —> A is the pullback of X —> Y), then ? is a Sullivan model for ?. ? Now consider the more general situation where ??? ~ Api(f)my. Here we shall extend ? to a commutative diagram (AVY,d) (AW®AV,< ?-id (AV,d) A5.9) (AWAid) —- {AWA®AW,d) —^ (AW,d)
Sullivan Models 205 in which ? and ? are Sullivan representatives for g and for g; i.e. ?? ~ ApL(g)m and ?~? ~ ^4pl(9)t?x. Denote by ??,?? : — ® A(t,dt) —> — the morphisms (id ¦ Eq) : ? 4 0 and (id · ??) : ? h-> 1. Let be the fibre product with respect to the morphisms ApL(Z) ® A(i,di) -?2» APL(Z) A- (ATF.4 ® AIF,d) and let 0L and 0fi denote the projections of the fibre product on the left and right factors. Now ?? is surjective and ker ?? = keroo. Hence H(kerQR) = 0 and gR is a quasi-isomorphism. It follows that sqql = ??? is a quasi-isomorphism and therefore so is ql . Let ? : (AV'y,d) —> ApL(A)<8>A(t, dt) beahomotopy from ??? to ApL(f)my. Then we have the commutative diagram (AVy,d) — PL q———>- [Apl(Z) ® A(i, dt)] x.iPL(z) (AI-V.4 ® AW, d s,eL A5.10) (AVY ® AV, d) — ¦ Since ^L is a quasi-isomorphism so is ?^?1. Moreover, if ? G Apl(Z) then ?oB ® t) = 0 and so (z ® i, 0) is an element in the fibre product. Now S\QL{z ® *>0) = E\(z ® i) = z, which shows that ?ie>l is surjective. By the Lifting lemma 14.4 there is a morphism ? = (?,?) : (AVy ® AV,d) —»· [.4PL(Z) ® A(i,di)] x^pl(z) (ATV4 ® ??-'.d) extending (??/,(?)?,-0) and such that ?!^Lr = .4p?,(g)m. In particular ??? = ?? and ?? ? = ??^? = ApL(g)m. Thus ? is a homotopy from ?? to ApL(g)m and <^ : (AVy ® AV, d) —>· (?1?'".4 ® ???, d) is a Sullivan representative for g. Moreover ? extends ? and so the left hand square of A5.9) commutes. Finally, set ? = &®??: (AV, d) —> (AW, d). By construction, the right hand square of A5.9) commutes. Moreover, just as m and ? factor to give m and ? so ? factors to define a morphism ? : (AV,d) —)¦ APL(F) © A(t,dt) such that ??? = nTp and ??? = Api(g)fn. Thus nTp ~ .4pL(^)m and Tp is a Sullivan representative for g. This completes the construction of A5.8).
206 15 Fibrations, homotopy groups and Lie group actions Remark As noted in §15(a), the models in A5.7) may be chosen so that {AVY,d) —> (AVY <g> AV,d) and (AWA,d) —> {AWA ® AW,d) are minimal relative Sullivan algebras. In this case (id ·?) is a quasi-isomorphism between minimal relative Sullivan algebras. Thus Theorem 14.11 asserts that id -? : (AWA ? AV, d) A (hWA ® AW, d) is an isomorphism of cochain algebras. Example 1 The free loop space Xs1. Let Ar be a simply connected topological space with rational homology of finite type. The free loop space, Xs , is the topological space of all continuous maps S1 —> X. We may identify these as the continuous maps f : I —> X such that /@) = /(I) and this defines an inclusion i : Xs —> X1. Moreover xs' { . ?? P(g) = 9@), ?(?) = (?,?) ? —^ ? ? ? is a pullback diagram of fibrations. Finally, the constant map I —> pt defines a homotopy equivalence X1 <— Xpt = X that converts the diagonal ? to q. We may now apply the results above to compute a Sullivan model for X5 as follows. Let (AV, d) be a minimal Sullivan model for X. Then multiplication ? : (AV, d) ® (AV, d) —> (AV, d) is a Sullivan representative for ? as follows from Example 2, §12(a). Convert this to a relative Sullivan algebra (AV ® AV ® AW, d) -=> (AV, d) and then by Proposition 15.8, (AV,d) ®??®?? (AV ® AV ® AW,d) = (AV <g> KW,d) is a Sullivan model for Xs . In this case we can carry out these computations explicitly. Define V by Vk = Vk+1 and consider the acyclic Sullivan algebra (E(V),d) = AV ® AdV constructed in §12(b). In (AV, d) ® (E(V),d) define derivations s of degree -1 and ? of degree zero by sv = v, sv = sdv = 0, and ? = sd + ds. (Here the isomorphism V = Vk+l is denoted by ? —> ?.) Then an isomorphism ? : AV ® AV ® AV A (AV, d) ® (E(V),d) oo is given by ?(? ® 1 <g> 1) = ?, ?A ® ? ® 1) = ? ^ and ?A ® 1 ® ?) = ?. ? " _ Now ? is a derivation satisfying ?? = ?? in AV ® E(V). It follows that oo n e& = ? 7?" is an automorphism of this graded differential algebra. Hence ?(\® n=0 dv ® 1) = ??A ® ? ® 1). This implies that the inclusion, ? : (KV, d) ® (AV, d) —> (AV ® AV ® AV, d)
Sullivan Models 207 is a relative Sullivan algebra. The quasi-isomorphism (AV ® AV ® AV, d) ^> (AV,d) ® (E(V),d) -=» (AV, d) converts ? into multiplication in AV, and so ? is a Sullivan model for the multiplication morphism. Thus, as described above, (AV ® AV, d) = {AV, d) ®??®\? (? V ® AV ® AV, d) is a Sullivan model for Xs . It remains to compute d. Observe that s = 0 in V and dV and hence that s2 = 0 since s'2 is a derivation. It follows that = v + dv + ^2 n=l and hence that s' = ?~??? is the derivation in AV ® AF ® AV given by ?'(?® 1® 1) = ?'A®?® 1) = 1® 1®? and s'( Thus in AV ® AV ® AV, ( v + ?^ n=l Finally, let s" be the derivation in AV ® AV given by s"v = ? and s"v = 0. Then an immediate computation from this formula gives the differential d in AV ® A V as: dv = dv and dv = —s"dv . This defines the Sullivan model for Xs explicitly in terms of (AV, d). ?
208 15 Fibrations, homotopy groups and Lie group actions (d) Homotopy groups. Suppose / : X —> Y is a continuous map between simply connected spaces, and let mx : {AVx,d) —> APL(X) and mY : {AVy,d) —> APL(Y) be minimal Sullivan models. Let </?/ : (AVY,d) —> (AVx,d) be a Sullivan representative for /; i.e., ????/ ~ ???(/)???. Its linear part, QU) : VY -»¦ Fx is independent of the choice of </?/ (§13(c)). In §13(c) we introduced the bilinear map defined by Q(a)v = (?; a)e, where a: Sk —> X represents a and (A(e,... ), d) is the minimal model of Sk. This determines the linear map ux :VX —>Homz (?*(*),*) given by (uxv)(a) = (v;a). Given / : X —> Y, we have (Q(f)v;a) = (?;?*(/)?), ? ? Vy, a G ?*(?). It follows that ?? is a natural transforma- tion: vx ? Q(f) = Homz (?.(/), ?) ? i/y. The minimal Sullivan models approach to rational homotopy theory is suc- cessful because ?? identifies the generating space of the model with the dual of the homotopy groups of the space: Theorem 15.11 Suppose X is simply connected and H»{X; h) has finite type. Then the bilinear map ?? ? ?*(?) —> Jk is non-degenerate. Equivalently, is an isomorphism. Remark 1 It follows from Theorem 15.11 that if H*(X;<QI) has finite type so does 7T,(X)®Q, since in this case the Sullivan model has finite type. Conversely, if ?» (?') ®Q has finite type then the procedure for constructing a rational cellular model Xq (§9) shows that #«(Ar;<Q>) has finite type. Indeed if we know by induction that the r-skeleton of Xq is finite we can conclude that it has rational homology (and hence rational homotopy) of finite type. Thus the (r+l)-skeleton is constructed by adjoining finitely many rational cells. Remark 2 Theorem 14.11 asserts that a morphism ? between minimal mod- els is an isomorphism if and only if it is a quasi-isomorphism. Since ? is an isomorphism if and only if its linear part Q((p) is an isomorphism we recover the
Sullivan Models 209 -rational Whitehead-Serre theorem for continuous maps / : A' —> Y; ??,(/) ® Q is an isomorphism if and only if H*(f;Q) is an isomorphism. proof of Theorem 15.11: Fix k > 2. To show that ?/? : V? -=> Homz G?*(?"), Jfc) we let r be the least integer such that ??(?) ? ?, and argue by induction on k — r. If k = r then the Hurewicz homomorphism is an isomorphism ^r(A') —> Hr{X\1) (Theorem 4.19). On the other hand, since H'l{X;k) = 0 for 1 < i < r — 1 the minimal model for X satisfies Vx = 0, 1 < i < r — 1 (Proposi- tion 12.2). Thus H(mx) : V{ A Hr(X;]k). It is immediate from the defini- tion that these isomorphisms identify ?? with the isomorphism Hr(X;]k) —> Homz (HT(X; Z), Ik). Thus when k = r, ? ? is an isomorphism. Suppose now that k > r. Observe first that if f : X —> Y is a weak homotopy equivalence then H*(f; Jk) is an isomorphism. It follows that a Sullivan represen- tative ?/ is a quasi-isomorphism and hence an isomorphism (Theorem 14.11). Thus Q(f), as well as ?«(/), are isomorphisms. Thus by naturality, we may replace A" by any space of the same weak homotopy type. In particular, we may suppose A" is a CW complex. Let ? be an Eilenberg- MacLane space of type (nr(X),r). Choose a continuous map g : X —> ? such that nr(g) = identity (Proposition 4.21). Factor g (as in §2(c)) in the form ? ?? ? ? with ? a homotopy equivalence and ? a fibration. Again, use naturality to replace X by ?" ?? ??. Thus we may assume there is a fibration ? : X —> ? such that ?r{p) is an isomorphism. In Example 2 of §15(b) we showed that the minimal model of ? has the form (AFr,0) -=> APLiK). In particular, H*iK;Jk) has finite type and so Theorem 15.3 and its consequences apply to the fibration p. Now let j : F —> X be the inclusion of the fibre of p. By the long exact homotopy sequence (Proposition 2.2), 7rr(F) = 0. Because itiiF) = 0, 1 < i < r, a minimal model of F has the form mp '¦ (AV>,d) A APL{F) with VlF = 0, 1 < i < r (Proposition 12.2). Thus dia-
210 15 Fibrations, homotopy groups and Lie group actions gram A5.4) for ? becomes APL{K) ApL(p). ApL(X) AplU)> ApL(F) {AVr®AVF,d) —r L ?-id Since Vp is concentrated in degrees > r the Sullivan algebra (AVr ® AVp,d) is necessarily minimal. Lift ??? to a quasi-isomorphism ? : (AVx,d) -=> (AV ® AVp,d) such that ??? ~ ??? and conclude from Theorem 14.11 that ? is an isomorphism. Since H*(X; Ik) has finite type so does AVx (Proposition 12.2(iii)). Hence so does AVp and, a fortiori, H*(F;lk). Finally ???(? ¦ ??)? = ApLU)mip ~ ApL(j)mx. This identifies (? · ??)? as a Sullivan representative for j. In particular, Qk(j) is the isomorphism Qk(e ¦ id)Qk((p). The long exact homotopy sequence for ? shows that nk{j) is also an isomorphism. Thus ?? and vp are identified in degree k by naturality, and ?? is an isomorphism in degree k. D Recall (§13(c)) that the projection A+Vx —> Vx induces a linear map ? : H+(AVx) —> Vx. From the commutative diagram at the end of §13(c) we deduce the Corollary The isomorphisms ?? : Vx —> Homz (n*(X),k) and ?(???) : ? (AVx) ^» H*(X;h) identify ? with the dual of the Hurewicz homomorphism. a Example 1 Rational homotopy groups of spheres. Let i G 7fn(Sn) be the class represented by the identity map of Sn. Then _ ( Q-i ,n = 2/c+l | 0 , otherwise and {Q-i ,n = 2k Q'[t,t]w ,n = 4k-l 0 , otherwise. Indeed in the first case the minimal model is (A(e),0) and (e;i) = 1. In the second the minimal model is (A(e,e'), de' — e2). Again (e;t) = 1 while Propo- sition 13.16 gives (e';[t,i]w) = -(e2;i,i) = -2. Now apply Theorem 15.11 D Example 2 The model (?(??,??,?), dx = eoe-i), k = Q. In Example 2 of §13(e) we considered the space X = (S3 V S3) U/?? U D\) where the two 8-cells were attached respectively by [a0, [ao,ai]w\w and
Sullivan Models 211 [??, [ai,ao]w]vj/·· The minimal Sullivan model of ?' was shown to have the form (AV,d) = (A(eo,ei,x,w,wo,---),d), where dx = eoei, dw = eoe\x and the Wi have higher degree. Thus a linear function V10 —> Q is defined by w h-> 1. By Theorem 15.11 there is a continuous map b : S10 —> X such that Q(b)w — Ae, some non-zero ? G Q. Put Y = X Uf, Dn = (S3 V S3) U/ (D% II Df) u6 D11. According to §13(d) the cochain algebra (AV ? ku,d?) is a commutative model for Y, where ? = [b] G ????(?). A straightforward calculation shows that a quasi-isomorphism (AV ©Jku,d?) A (A(eo,ei,x),d) is given byuH —A~1eoeix, m 4 0, «ij 4 0, i > 0. Thus (A(eo,e\,x),d) is the minimal Sullivan model for Y. In particular, ,n = 3 Q ,n = 5 0 , otherwise. Observe that the inclusion (A(eo,ei,x),d) —> (A(eo,ei,x,tu,... ),d) is a Sulli- van representative for the inclusion i : X —> Y. This shows that Q(i) is injective and so the dual map ?, (i) ® Q : ?» (X) ® Q —»¦ ?,(?) ® Q is surjective. Finally, we show that the homotopy fibre, F, of i has the same minimal Sullivan model as a wedge of spheres, and compute its cohomology. Extend (A(eo,ei,x),d) to the contractible Sullivan algebra (A(eo,ei,z,t;o,t>i,y),d) by setting dvo = eo, dv\ = e\ and dy = ? + eoVi. It follows from A4.6) that Q®A(eo,e,,x) (AV,d) is a Sullivan model for F. Form the pushout (A,d) = (A(eo,e1}x,vo,vi,y),d) ®A{eo,e],x) (AV,d) as described in §14(a). On the one hand, (A,d) may be regarded as a relative Sullivan algebra (A(eo,ei,x,vo,vuy),d) —> (A(eo,...,y) ® A(w,wo,...),d). Since ? : (A(eo,... ,y),d) —> Q is a quasi-isomorphism, so is ? ® — : (.4, d) (A(u>, w0,... ), d). Thus (?, d) is a Sullivan model for F. On the other hand, (.4, d) may be regarded as a relative Sullivan algebra (AV,d) ^ (AV ® A(vo,vuy),d).
212 15 Fibrations, homotopy groups and Lie group actions Now apply the quasi-isomorphism (AV,d) —> (A(eo,ei,x)/e§e\X,d) to obtain a quasi-isomorphism (A,d) ^ (C,d) = ([A(eo,ei,x)/eoeix]®A(wo,fi,i/),c0· Since H (A(e0, · · ·, y), d) = Q, the short exact sequence 0 —> ((Qeoeix) ® A{vo.,Vi,y),O) —> (A(eo,ei,x,vo,vi,y),d) —> (C,d) —» 0 allows us to compute H(C), which is isomorphic to H*(F;Q). In particular, the Hubert series of H*(F; <Q>) is given by More precisely, the cocycles eixv^v^y ? C, k > 1, ? > 0, m > 0 represent a basis of H+(C). Since these cocycles multiply each other to zero it follows that H+(C) ¦ H+{C) — 0. Moreover these cocycles define a quasi-isomorphism (H(C),0) —> (C,d) of cochain algebras. Hence (H(C),Q) is also a commutative model for F, while by Proposition 3.4 it is also a commutative model for a wedge of spheres. ? Example 3 The Quillen plus construction . Let X be a path connected topological space whose fundamental group, ?\ {?) is finitely generated and such that every element in ix\ (X) is a product of com- mutators. Then ??(?;?) = 0, by the Hurewicz theorem 4.19. Adjoin to X finitely many two cells e?,...,e^ to kill a set of generators of ?\(?). The Cellular approximation theorem 1.2 implies that ?\(?) —> ??\{? U LJe?) is surjective. Hence X U (Uef ) ^s simply connected. On the i \ i / other hand, since iii(X;Z) = 0, the long exact homology sequence for the pair ( X U '( U e? j ,X\ shows that there are homology classes ??,..., an ? \ \ i / / H2 I JU[Jef;ZJ that project to the classes [e?],..., [e^] in the relative ho- mology. Since XU ( |Je? j is simply connected the Hurewicz Theorem implies that the ai can be represented by maps en : S2 —> X U ( \J e\ J. Attach three cells by these maps to create the topological space Y = X U ? (J ef J U I (J e| J. This is called the Quillen plus construction on X. This construction has two important properties. Firstly Y is simply connected. Secondly the inclusion X —> Y induces an isomorphism of homology, as follows
Sullivan Models 213 immediately from the long exact homology sequence for (Y,X). In particular in the case X is an Eilenberg-MacLane space (and hence has no higher homotopy groups) the higher homotopy groups of Y are invariants of the group ?\ (X); its algebraic K-groups. Now suppose the rational homology H*(X;Q) has finite type. Then we can construct the minimal Sullivan model ??? : (AV,d) ^> Apl(X) and, as in Proposition 12.2, V will be a graded vector space of finite type. However, unlike the situation in Theorem 15.11, the graded vector space V may have little to do with ?, (X). On the other hand, the inclusion i : X —> Y induces a quasi-isomorphism Apl(i) : Apl(Y) —> Apl(X)- Hence a Sullivan representative is an isomor- phism of minimal models (Theorem 14.11), which exist by Theorem 14.12. Thus Theorem 15.11 provides an isomorphism ?? :V -=>HomzGr.(y),Q) - In particular the rational .fiT-groups of ?? (?) may be computed directly from Apl{X)- When X is a smooth manifold we may also use Adr(X) as observed in §12(e). ? Example 4 Smooth manifolds and smooth maps. Let X be a smooth manifold. In §11 we showed that Adr(X) is connected by natural quasi-isomorphisms to Apl(X;R). Thus a minimal Sullivan model (AV,d) ^> Adr(X) is a Sullivan model for X (over E). In particular, if X is simply connected and H*(X; W) has finite type (e.g. if X is compact) then there is an isomorphism vx :l/-=>HomzGr.(X),R) (Theorem 15.11). This shows that the real homotopy groups, as well as the cohomology algebra, may be computed from Adr(X)- Now let ? : X —> Y be a smooth map. In the same way, a Sullivan represen- tative for Adr(<p) is a Sullivan representative for ?. Moreover, if Y is simply connected and has real homology of finite type and if m : (ADR(Y) ® AW,d) ^ ADR{X) is a Sullivan model for Adr(<p), then we can apply Theorem 15.3 to identify (AW, d) as a Sullivan model for the homotopy fibre of ?. (This, of course, need not be a manifold !) D (e) The long exact homotopy sequence . Let ? : X —> y be a Serre fibration of path connected spaces, and with path connected fibre j : F —> X. Assume that Y is simply connected and that all spaces have homology #„(-; k) of finite type.
214 15 Fibrations, homotopy groups and Lie group actions In §15(a) we constructed the diagram, labelled A5.4): (AVY,d) —r~ (AVY®AV,d) -— {AV,d) in which all three vertical morphisms are Sullivan models. Moreover we can (and do here) take (AVy,d) and (AV,d) to be minimal. However, the central Sullivan algebra, (A(VY @V),d) need not be minimal. Let do be the linear part of d: it is the differential in V ? Vy defined by dx - dox G ?-2(V ? VY), ? ? Vo ? VY. Then i and id -? restrict to a short exact sequence of cochain complexes, 0—> (VV,0) ? (VY©V,do) -^ (V,0) —»¦ 0. A5.12) This construction is natural in the following sense. Given a map of Serre fibrations, consider the Sullivan representatives constructed in A5.8). Their linear parts define a morphism of the short exact sequences A5.12) and of the corresponding long exact cohomology sequences. Recall now from Theorem 14.9 (applied with ? = k) that {A{VY ® V),d) 9* (AW, d) ® A(U ? dU) with (AW, d) a minimal Sullivan algebra and A(U ® dU) contractible. This defines a quasi-isomorphism ? : (AW,d) -=> (A(VY ©V),d), and ??? = m\ : (??^, d) -=* Api (X) is its minimal Sullivan model. Moreover (Proposition 14.13), the linear part of ? induces an isomorphism W -^f H(VY ? V,do). Use this to replace H(Vy ? V, do) by W in the long exact cohomology sequence of A5.12), which then takes the form Proposition 15.13 Suppose X and F are also simply connected. Then the isomorphisms vY, vx and vF of Theorem 15.11 identify the long exact cohomol- ogy sequence above, up to sign, with the dual of the long exact homotopy sequence of the Serre fibration ? : X —>¦ Y. proof: The linear maps VY ->¦ W and W -> V are identified with Homz (?*(?), k) and Homz (?,(j),k), as follows easily from the definitions. To check that the two connecting homomorphisms are identified we will verify that (dov;a) = (-I)k+I(v;d*a), veVk, ae nk+1(Y). Fix k and a, and let ? : Sk+1 —> Y represent a. Recall the path space fibration ? : PSk+1 —> Sk+1. Since PSk+1 is contractible, const ~ ap. Cover ^;c hnmntnnv with a homotopy PSk+1 ? I —> X from the constant map to some
Sullivan Models 215 map b : PSk+1 —> X. Thus pb = ap and so b restricts to a map c : CiSk+1 —> F. By naturality, it is sufficient to establish our formula for the fibration PSk+1 —> Sk+1 and a = [idsk+i]. In this case, as we saw in Example 1, §12(a) (AVy,d) = (A(e),0) or (A(e, e'), de' = e~), depending on whether k + 1 is odd or even. Moreover, as shown in Example 1, §15(b), (AV,d) = (A(u),0) or (A(u,u'),O) and the differential in (AV>- ® AV, d) is given by du = e and du' = e' — eu. In this case u is a basis for Vk and dou = e, so that we are reduced to proving Recall the continuous map ? : Sk ? I —> Sk+1 whose picture is given in Example 5, §1. Regard ? as a based homotopy from the constant map to itself and lift this through ? : PSk+1 —> Sk+1 to a based homotopy § : Sk ? I —> PSk+1 from the constant map to a map h : Sk —> QSk+1. Then [h] = d*[idSk+i]. Let zk € Ct(Sk;Z) be a cycle representing [Sk] — cf. §4(d). If ? is the identity map of /, regarded as a singular 1-simplex then C*F)EZ(Zk ® ?) is a cycle representing [5fc+1]. Set w = C»(e)EZ(zk ® t)· A quick computa- tion shows that C*(p)w is the cycle representing [5fc+1] and that dw is a cy- cle in C*(USk+1;Z) representing H*(h)[Sk]. Thus w projects to a relative cycle w G Cm{PSk+1tUSk+1;Z). Moreover C7.(p) : Cm(PSk+ltUSk+1;Z) —> C*(Sk+1,pt;Z) and we have, by construction that H.(p)[w] = [Sk+1] and d.[w] = H*(h)[Sk]; here 9« also denotes the connecting homomorphism in homology for the pair (psk+l,nsk+l). On the other hand, the quasi-isomorphisms ? APL(PSk+1,nsk+1) «- APL(PSk+1) —- ApL(nsk+1) ·- ? ¦1= (A+VV®AV,<i) -*¦ {AVy ®AV,d) (AV,d) identify the connecting homomorphisms in the two long exact cohomology se- quences. Moreover (cf. A0.13)) the upper long exact sequence is, up to sign, dual to the long exact singular homology sequence. It follows that (H(m)[e ® 1], [w]) = (H*(p)H(mY)[e], [w]) = (H(mY)[e], Finally, since du = e ® 1, (H{m)[e® 1],[UJ]) = (d*H(m)[u],[w}) = (-l)k+1(H(m)[u],Hk(h)[Sk]) . _ But the Hurewicz homomorphism converts the duality between V and ?« to the duality between cohomology and homology. Thus (H(mY)[e],[Sk+1]) = (e;[ids^})
216 15 Fibrations, homotopy groups and Lie group actions and (H(m)[u], Hk(h)[Sk}) = (u; [A]> = (u; o,[id5*+.]>. ? Return to the setup described at the start of this topic and recall that a linear map C : H+{AV,d) -4 V is defined by ?[?] - ? G A^2F, ? a cocycle in A+F. When F is simply connected this is dual to the Hurewicz homomorphism, hurp, (§15(d)) and so cfoC is dual to hur? d„. We need this without the restriction that F be simply connected: Proposition 15.14 The linear maps doC and hurp d» are dual, up to sign, if F is path connected and Y is simply connected. proof: Let ? € (AF)* be a d-cocycle and let ? G ???+1 (Y) be represented by ? : Sk+1 —»?. We show that (????·,?) = (-l)k+l(H(m){zlhurFd*a). Use the pullback of the fibration via a, and naturality, to reduce to the case Y - Sk+1 and a= idsfc+i. Then (AVY,d) = (A(e,...),cQ and dz = ???? = Xe, for some ? G Jk. Now the argument used in 15.13 for the path space fibration (with the last paragraph suppressed) applies verbatim to give the proposition. D (f) Principal bundles, homogeneous spaces and Lie group actions. Let G be a path connected topological group with finite dimensional rational homology. (This holds for all Lie groups.) As observed in Example 3, §12(a), the minimal Sullivan model of G has the form where Pg is a finite dimensional graded vector space concentrated in odd degrees. Let xi,...,xr be a basis of Pg with degree Xi = 2lt + 1. Since G is weakly equivalent to QBg (Proposition 2.10), there are maps f-L : S2li+1 —> G such that {H(m.G)xj,H*(fi)[S2ei+1}} = 5i:j (apply Proposition 15.14). Thus we recover a theorem of Serre: multiplication in G defines a rational homotopy equivalence Now consider the universal principal bundle pg ¦' Eg —> Bq constructed in §2(d). As described in §15(e) it determines a long exact sequence connecting the generating spaces for minimal Sullivan models of Bg, Eg and G- But H*{EG) = k and so W = 0. Thus d0 ¦" PG A VX+l. It follows that VBo is
Sullivan Models 217 finite dimensional and concentrated in even degrees. Thus AVbg is concentrated in even degrees and so the differential must be zero. We have thus established Proposition 15.15 The minimal Sullivan model for Bq has the form where V?G = Pq+1 . In particular, H*(BG) is the finitely generated polynomial algebra, AVbg- d Since G is weakly equivalent to QBg, Bq is simply connected. Since (AV?G, 0) is the Sullivan model of Bo-, it follows from Theorem 15.11 that V?G — Homz GT«(i?G),¦&)· Thus tt«(Sg) ® Q is finite dimensional and concentrated in even degrees and, for degree reasons, all the rational Whitehead products for ? g vanish. Let yx,..., yr be the basis of Vbg corresponding to the basis x\,.. .,xr of Pg, and define a contractible Sullivan algebra (AVbg <8> APg,cI) by setting dxi = yt and dyi = 0. Since H*(Eg) = Ik, there is a commutative diagram of morphisms, Apl(Bg) Apl{pg) ' APL{EG) - APL(G) (AVBg,0) (AVBG®APG,d) defined by setting ???{ = ?$, where ?$ G Apl{Eg) is anjr element satisfying ??? = ApL(pG)rriBGyi- Here m is a quasi-isomorphism by inspection and so this is the special case of diagram A5.4) for the universal bundle. In particular (Proposition 15.5), misa quasi-isomorphism. Thus we may take tug = ??. Finally, suppose ? : X —> Y is any principal G-bundle over a simply connected CW complex, Y. (The following could be extended to the non-simply connected case.) This bundle is the pullback of the universal bundle pg ¦ Eg —> Bg over a map / : Y —> Bg as shown in §2(e). The cohomology classes in Im H*(f) are called the characteristic classes of the fibre bundle. Regard the basis yi, ¦ ¦ ¦ ,yr of Vbg as cohomology classes of Bg, so that H*(f)y\,..., H*(f)yr are charac- teristic classes. Then a Sullivan representative for / is characterized by the condition: ?/y-i is a cocycle in AVy representing H*U) 1 < i < r. It defines the relative Sullivan algebra (AVY,d) —> (AVY ® APG, d)., dxx = ?^ ;
218 15 Fibrations, homotopy groups and Lie group actions which is just the pushout (AVy,d) ®(AvBG,d) (AV?G ®APG,d). Thus the Re- marks in §15(c) give a commutative diagram APL(Y) - Apl{p) - APL(X) - APL{G) (AVY,d) (AVY®APG,d) (???,?). This identifies (AVy ® APG,d) as a Sullivan model for X. Example 1 Homogeneous spaces. Let ? C G be a closed subgroup of a connected Lie group G. Right multi- plication by ? is an action on G and the projection ? : G —> G/K onto the orbit space is the projection of a principal if-bundle [70]. The space (in fact, a smooth manifold) G/K is called a homogeneous space. Define a right action of ? on ?? x G by setting (?, ?) · b = (xb,b~xa). The inclusion j : ? —> G defines an obvious continuous map E(j) : ?? —> Eg- The formula (?, ?) h* (E(j)x) ¦ a defines a continuous map ?? xG —> Eg which factors to yield a continuous map / : (?? x G)/K —> Eg- Thus we obtain a commutative diagram K x G)jK BK BG in which q and B(j) are the maps of orbit spaces corresponding respectively to the projection ?? x G —> ?? and to E(j). As in §2(e), g is the projection of a principal G-bundle which the diagram exhibits as the pullback of the universal bundle over B(j). Suppose now that ? is connected too and use Sullivan models for ? ? and Bq to identify H*(B(j)) : as a Sullivan representative for B(j). If yi,..., yr is a basis for V?G corresponding to a basis xl,...,xr of Per then the discussion above exhibits {AVBk ® APG, d), dxi = H* as a Sullivan model for (?? x G)jK. On the other hand the projection ?? xG —> G induces (as in Proposition 2.9) a weak homotopy equivalence q' : (?? x G)/K —> G/K. Thus G/K and (V.v ? G)/K have isomorphic Sullivan models and we obtain
Sullivan Models 219 Proposition 15.16 The Sullivan algebra (AV?K ®APo,d) defined by dx{ = H* (B(j)) y{ and d = 0 in Vbk is a Sullivan model for G/K. ? Example 2 The Borel construction; application to free actions. Let G be a connected Lie group. An action of G on a space X determines the fibre bundle q : XG —> BG with fibre X, as described in §2(e). Now fibre bundles are Serre fibrations (Propo- sition 2.6) and Bg is simply connected. Thus we may apply Proposition 15.5 to this Serre fibration to obtain APL(BG) Apl(9) > APL{XG) AplU) ' APL(X) ,0) - (AVBg®AVx,D) - (AVx,d) in which ??? is a minimal Sullivan model for X. Thus the Borel construction, Xg, has a Sullivan model of the form (AVBg ® ???,?) above. G Proposition 15.17 If a compact connected Lie group acts smoothly and freely on a manifold X then the Sullivan algebra (AVbg ® ???,?) is a Sullivan model for the orbit space XjG. In particular, ?{ (AVBa ® AVX,D) = 0, i > dim X — dimG. proof: In this case the projection X —> X/G is the projection of a principal G bundle [REF] and Proposition 2.9 provides a weak homotopy equivalence q' : Xg —> X/G. This implies the first assertion. Since X/G is a manifold and dim X/G = dim X — dim G, the second assertion follows. D Example 3 Matrix Lie groups. Compact matrix Lie groups SO{n) C M(n; 1), SU(n) C M(n; C), Q(n) C M(n; M) are defined as follows, where M, C, and H are the reals, complex numbers and quaternions. Both C and H have a conjugations: a + ?i = a — ?i and ? + ?i + -yj + 5k = a - ?i - jj - Sk. Then SO(n) = {A | A1 = A~X and detA = 1} SU(n) = {???* = A~! and det.4 = Q(n) =
220 15 Fibrations, homotopy groups and Lie group actions The linear action of SO(n) in Rn restricts to a transitive action on 5n-1 which identifies S0(n)/S0(n - 1) -=> S'1. Similar remarks apply in the other cases, to give principal bundles: S0(n - 1) —> SO(n) —> S71'1, SU(n - 1) —> SU(n) —> S2" and In particular 50B) ~ S1, SUB) s S3 and Q(l) = S7. On the other hand, for any connected Lie group G with Sullivan model (APg-.O) we have Pg — ^*{G) ® k as graded vector spaces (since G is weakly- equivalent to QBg)· Thus the long exact homotopy sequences for the Serre fi- brations give an easy inductive calculation of the Sullivan models of the classical compact groups: 50B??. + 1) S0Bn) SU(n) Q(n) : ?(??,.. : ?(??,.. : A(x-2,.. : ?(??,.. ¦,??) .,??-?,? , ¦,??) ¦,??) ???" ??. — Ai UCg ? 2 ** ? ? (? ?* · 4 7 degxi = 2i- d.6?C ? ' — 4z 1. 1, 1. 1. degx'n Example 4 /4 fibre bundle with fibre S3 x SUC) that is not principal The model for SUC) is A(x2,x3) with degx2 = 3 and degx3 = 5. Thus the model for its classifying space is A(y-2,y3) with degt/2 = 4 and degj/3 = 6. These elements are dual to maps Si -^» BSuC) and 56 —>¦ BSUCy Consider the continuous map / : S3 ? S3 —> E3 ? 53)/E3 V S3) = S6 ? ?5?/C) and use it to pull the universal bundle back to a principal 5f/C)-bundle ? : X —»¦ S3 x S3. Let g : ? -»¦ S3 be the composite of ? with projection on the left factor. It is a relatively easy exercise to exhibit ? as a smooth fibre bundle, and a theorem of Ehresmann then asserts that g is a smooth fibre bundle too. (It is trivial that q is a Serre fibration.) The fibre of q is the compact Lie group S3 x SUC). Now the discussion above gives a model for the principal bundle p, from which we find that q is represented by (Au,0) >¦ (A(u,V,X2,X3),d) with degu = degu = 3; du = dv - dx2 = 0 and dx3 = uv. This is not the model of a principal bundle and so q is not principal. In Example 3, §14(d), we constructed a space Y = S3 V S3 U D8 U D8 U D10· By inspection S3 x Y has the same Sullivan model as X. It is, in fact, easy to see that X and 53 x Y have the same rational homotopy type. ?
Sullivan Models 221 Example 5 Non-existence of free actions. Consider the manifold X of Example 4. We observed there that a certain bundle q : X —>¦ S3 with fibre S3 x SUC) was not principal. Now we establish a stronger assertion: X does not admit any free smooth S3 x SU(S) action. In fact we show more: X has no free smooth S3 x S3 action. Indeed let G = S3 x S3. For any G action the Borel construction has a Sullivan model of the form with dega,· = 4, and Dai = 0. For degree reasons Dx-$ = dx3 = uv and Du,Dv G ?(??,??). Thus 0 = D2x3 = Du®v-Dv®u, whence Du = Dv = 0. We also have Dxo, G ?(??, 02) and these calculations now show that A(ai,ai)/(Dx2) is a subalgebra of H*(Xg)· This subalgebra cannot be finite dimensional and hence (Proposition 15.17) the action cannot be free. ? Exercises 1 Compute a relative Sullivan model for: a) the Hopf map 7 : Sr -> S4, b) the composite S3 x S4 -> S3 ? S4 S S7 4 S4. 2 Consider a Serre fibration of path connected spaces F ? ? -> ?. Prove the following assertions: a) If Y = S3VS4VS5 and H*(F;Q) = ?(?1??2, ....,xk)/(xi,x'i ...,xl) with each X{ of even degree then X ~q Y x F. -.b) If if*(F; Q) = A(x)/(xfc) with x of even degree and k > 2 then the mor- phism ii*(i) : H*(X;Q) -> H*(F;Q) is onto and if'(Ar;Q) is isomorphic as an if*(F;Q)-module to the free module H*(Y;Q) ®H*(F;Q). 3. Consider for ? > 1, the group of unitary matrices: ^4 G U(n) if A = .d^1. (A 0 Let ?/(&) c—y U(n + k) be the inclusion A ^ n r a) Using the fibration U(n - l)->J7(n) -)· U(n)/U(n - 1) = 52"-1 prove that f/(n) admits a minimal model of the form (?(?1,0:3, ...,a;2n-i),0) with deg Zi = i. Compute the minimal model of the classifying space BU(n). b) Using the fibration S2* -> f/(n + Jb)/J7(ife) -> J7(n + k)/U(k - 1) prove, by induction on n, that H*(U(n + k)/U(k));Q) = A(x2fc+ljx2Jt+3,-,a:2n+2i-i). 4. Let F be the homotopy fiber of the inclusion of the ?-skeleton into the classifying space X = BU(n). It is known that H*(F;Q) is the cohomology of the "Lie algebra of formal vector fields on n-variables". Compute this cohomology.
222 15 Fibrations, homotopy groups and Lie group actions 5. Recall that for ? > 1, the nth-Postnikov approximation fn : X -» Xn of a topological space X is defined by the following properties : ?^(/?) is an isomor- phism for k < ? and nk(Xn) = 0 for i > ? + 1. Construct a relative minimal model for fn. Deduce that a Sullivan minimal model of the homotopy fibre of fn is quasi-isomorphic to the quotient differential graded algebra ifc<g>Ay<n (AV,d). 6. Let X be a 1-connected finite CW complex and ? > 1 and let ? be a Sullivan representative of a continuous map / : K(Z, 2ri) —> X. Prove that ? is homotopic to the trivial map.
16 The loop space homology algebra In this section the ground ring is an arbitrary field ic of characteristic zero. In this section we consider based topological spaces, (X,x0), and we shall frequently assume that X is simply connected and Ht(X;k) has finite type, A6.1) even though a number of the results can, with additional work, be established for a larger class of spaces. We shall also simplify notation and write 7?.??®* = 7?.(?')®?*. As described in §2, composition of paths defines a continuous associative mul- tiplication ? : ?? ? ?? —> ?? in the (Moore) loop space ?? and makes the path space fibration PX —> X into an ??-fibration. The identity element of ?? is the constant path eo of length zero at Xo, and we shall always use eo as the basepoint of ??. The connecting homomorphism for the path space fibration is an isomorphism 0. :?.(?) -=>??._?(??), since PX is contractible. In particular, ? ? (??) is abelian. The geometric product ? : ?? ? ?? —> ?? makes C*(QX;Jt) into a chain algebra via the Eilenberg-Zilber morphism EZ, as described in §8(a). The mul- tiplication in the homology algebra, Hm(QX;Je), is given by ? · ? = ?*{?)?{??){?® ?), ?, ? G ?^??-k) , and the graded algebra H»(UX;Je) is called the loop space homology algebra of X. In this section we establish some strong structural results both for the coho- mology algebra H*(QX;]k) and the homology algebra H»(flX; h) for any space X satisfying A6.1). When X is a finite complex, however, its loop space ho- mology has many more remarkable and beautiful properties. Developing these is one of the main objectives of Part V of this text. The structural results to be shown here are summarized by: • H*(QX;Je) is a free graded commutative algebra. • The Hurewicz homomorphism is an isomorphism onto the primitive space for ??. The graded algebra ?» (??;3?) can be computed explicitly from the graded vector space ?, (X) <g> h and the Whitehead product map w :G?.(?)®?)®(?.(?')®?) —> ?.(?) ® Jfc.
224 16 The loop space homology algebra The first assertion is due (essentially) to Hopf and the last two to Cartan- Serre-Milnor-Moore. This section is organized into the following topics: (a) The loop space homology algebra. (b) The minimal Sullivan model of the path space fibration. (c) The rational product decomposition of UX. (d) The primitive subspace of H»(QX;]k). (e) Whitehead products, commutators and the algebra structure of ?,(??; Jk). (a) The loop space homology algebra. Let (Y, y0) be a based path connected space. Recall (§3(b)) that the tensor product of graded algebras A and ? is the graded algebra A <8> ? with (a <g> b)(ai <8>&i) = (-l)de86de6aiua1 ®bh. Recall also (§3 (e)) that since A is a field, homology commutes with tensor products. Lemma 16.2 (i) If f : (X,xq) —> (Y,Vo) is continuous then Hr(uf;]k) is a morphism of graded algebras. (it) H(EZ) : ?.(??;]?)®?.(??;*) -=»¦ ?.(?? ? ??; A) is an isomorphism of graded algebras. proof: The first assertion follows from the naturality of EZ and the fact that ?/ is a morphism of topological monoids. The second assertion is a trivial consequence of formula D.8) in §4(b). D Next, recall from §4(b) that the topological diagonal Atop : Y -> ? ? ?, y ¦->¦ (y^y)j induces the Alexander-Whitney diagonal ? = AW ? C.(Atop) : C*(Y; A) —* C.(K; A) ® C.(Y; A). Since homology commutes with tensor products, ii»(A; A) is a linear map H.(A;k) :Ht(Y;k) —>¦ H.(Y; For simplicity we abbreviate #.(?; A) to ? (A). In §4(b) we observed that (C*(Y;&),A) is a differential graded coalgebra, co-augmented by any y G Y. Passing to homology, we see that H (A) makes Ht(Y; A) into a graded coalgebra, co-augmented by any [y] ? H0(Y;Jk). If Y is
Sullivan Models 225 path connected these homology classes all coincide and we denote them by 1. In this case we have for all ? G H+(Y;k) that H(A)a - (a <g> 1 + 1 <g> a) G H+ (Y; k) ® H+ (Y; k) , as observed in §3(d). In particular, the element ? is primitive if H(A)a = ? <8>1 + 1<8>? and the primitive elements form a graded subspace Pt(Y;k) C H+(Y;k), the primitive subspace of H*(Y: k). Remark Note that the product (§5) in the cohomology algebra H*(Y;]k) is dual to the comultiplication ? (A). Lemma 16.3 If Y = UX thenH(A) : H.(ilX;k) —» H.(nX;k)®H.(nX;k) is a morphism of graded algebras. proof: Observe that ?? ? ?? is a topological monoid with component-wise multiplication, and that ???? is a morphism of topological monoids. It follows that Ht(Atop) : Ht(QX;k) —> Ht(UX x UX;k) is a morphism of graded algebras. On the other hand H(AW) = H(EZ)~l as shown in Proposition 4.10. Since H(EZ) is a morphism of graded algebras (Lemma 16.2) so is ? (AW). Hence so is ? (A). ? Example 1 The homology algebra H9(QS2k+l;k) , k > 1. Let / : S2k —* QS2k+1 satisfy [/] = dt[idS2k+,}. Then the Hurewicz theo- rem 4.19 asserts that H* (f)[S2k] is a non-zero homology class ? G H2k(^S2k+1 ; k) The inclusion of ? extends to a unique morphism from the polynomial algebra h[a}. We show this is an isomorphism: In fact [S2k] is (trivially) primitive in Ht(S2k-]k) and so ? is primitive in H*(QS2k+1;Jk). Since ? (A) is an algebra morphism, and since k is even, H(A)an = (a <g> 1 + 1 <g> a)n = an <g> 1 + ???~? <g> ? + · · · + 1 <g> ?", ? > 1. If ?" ? 0 then it follows that H(A)an ? 0 and so a" / 0. Hence our morphism is injective. But in Example 1, §15(b) we computed the Hubert series diH'(nS*k+1;k) to be (l - z2k)~l. Thus A 2 2rk = dim k[a]i. 0 , otherwise It follows that the morphism above is an isomorphism. At the end of §16(d) we establish the same result for Ht (QS2k ; Jfe). ?
226 16 The loop space homology algebra (b) The minimal Sullivan model of the path space fibration . Suppose (?,??) is a based topological space satisfying A6.1) and let ??? : (AVx,d) —> Apl(X) be a minimal Sullivan model for X. Then the path space, PX, has a Sullivan model of the form m : (AVx ®AV,d) -=> ApL(PX). It factors to give a minimal Sullivan model rn : (AV, d) —> Api(QX) for the loop space CIX (the Corollary to Proposition 15.5 applied to the path space fibration PX —y X). As in §15(e) recall that the linear part of d is the differential d0 in Vx@V defined by requiring (d — dQ)x ? A-2(Vx © V). Because of the minimality of (AVx,d) and (AV,d) we have d0 = 0 in Vx and d0 : V —» Vx. Now observe that d0 : V A VX. Indeed, (AVx <g> AV, d) is the tensor product of a contractible algebra and a minimal model for PX (Theorem 14.9). Because PX is contractible the minimal model is trivial and (AVx <8> AV, d) itself is contractible. This implies H(Vx ® V, d0) = 0 and hence that d0 : V A Vx. Next, recall that Vx is a graded vector space of finite type (Proposition 12.2). Hence so is H"(UX;Jc) = H(AV~d). Since UX is an ii-space it follows (Exam- ple 3, §l2(a)) that the differential d is zero, m: (AV,0) -=*???.(??). Thus d : A(VX®V) —» A+V*- <g>AV. In particular, H*(QX; k) is the free graded commutative algebra AV. Moreover, if 2fa ,2k2,... and 2ix + 1,2?2 + 1,... are the degrees of a basis of V then the Hubert series for H*(tlX;]k) has the form Next observe that the action of UX on PX is a map of fibrations, PX ? ?? 9—~ PX X X idx where pL denotes projection on the left factor, PX. The tensor product of Sullivan models is a Sullivan model for the topological product (Example 2, §12(a)) and so m-m: (AVX <g> AV,d) <g> (AV,0) -=> APL(PX x UX)
Sullivan Models 227 is a Sullivan model. The restriction of g to the fibres at x0 is the multiplication ? : ?.? ? ?.? —y ??. Thus the construction of diagram A5.9) gives a commutative diagram (AVx,d) (AVx®AV,d) (AV,0) (AVx,d) —- {AVX <g> AV, d) ® (AV, 0) (AV, 0) <g> (AV, 0) in which ? is a Sullivan representative for ?. In particular, (??·??)? ~ ???(?)??, and so H(m) : AV -^ H*(QX; h) identifies ? with the dual of the multiplication map for the graded algebra ?*(??;&). The fact that 1 6 ?0(??;&) is the identity element translates to ^?-(?<8>1 + 1<8>?) G A+F<g>A+F, ? e A+V. A6.5) Recall next (§13(e)) that the quadratic part of the differential d in AVx is the derivation d\ determined by the two conditions di-.Vx —> A2VX and d-dx:Vx—> A*ZVX. In §13(e) we showed how to compute di from Whitehead products. Here we show how to compute di using the map ?. Let ? €V and, as in A6.5) above, write ?? = ? <g> 1 + ??, ® ?, + 1 <g> ?, with ??,?, G A+V. Recall that ? : A+V —y V is the projection corresponding to the decomposition A+V = F © ?-2 V and that d0 : V -^ Vx. Proposition 16.6 The quadratic part of the differential in AVx is given by dxdov = ?(-!)*«¦i(doC*i) ? (doCVi). proof: Write dv = dov + ? u, ® ^ + ? + ? with u, € V*·, Uj € V, ? e ? i and ? € A^3(Vx ? F). Since d2v = 0 the component of cPu in A2(Vx ? V) is zero, and this fact translates to On the other hand, since ??? = (d <8> ??)??, the components of tpdv and {d®id)ipv in Vx®\®V also coincide. Use the following facts to compute these components:
228 16 The loop space homology algebra • Im d C i\+Vx <8> AV · Im(^? - id ®?) C ?+ Vx <g> AV <g> AV • ?= id in kVx · TpVi - 1 ® ?? 6 A+V ® AV. This gives the equation Euj(8>li8>Vi = ?^???????&???, whence J](-l)desUiUiA do^i = -?(-l)deg<i>i(doC$i) ? (doC*.·)· D t (c) The rational product decomposition of ??". We turn the first part of the algebra above into geometry and provide a geo- metric proof of A6.4) by identif3ring ??' as (rationally) a product of odd spheres and loop spaces of odd spheres. First we need to make some remarks about infinite products. Suppose (Xa:*)acj is a family of based topological spaces. Let ?(??, ... ,ar) C Y\Xa a be the subspace of points (xa) such that xQ = * if ? ? ??7...,??; clearly ? (??,..., ??) — ? ?? ? · · · ? X<*r ¦ Let ? ?? be the union of the ?(??,..., ??) ? as {??,..., ??} ranges over all finite subsets of J and give it the weak topology determined by the X(cti, ¦ ¦ ¦ ,??). This space is called the weak product of the (Xa,*)- Any compact subspace of f[ Xa is contained in some X(a\,..., ??), so ]1 Xa a a is indeed a fc-space. Moreover, if each Xa is a CW complex with * 6 (Xa)o then so is each X(a\,. ¦ ¦ ,??), and this makes f[ Xa into a based CW complex. a It also follows that and so if each Xa is an Eilenberg-MacLane space K(Va;n) then Y[Xa is a a K(@Va,n). Finally, given a topological monoid G with unit element e and a continuous based maps fa : (Xa, *) —y (G, e) we define (/?) : ? ?? -> G, (fa)(Xa) = T\UM ¦ a a Now we return to the study of ??. First, suppose (X, *) is an arbitrary simply connected based topological space. Every based map ? : Sk+1 —y X determines two maps: ?? ·. USk+1 —* ?? and ? : Sk —> ??, where [?] = 3,[?] is the image of [a] in ?.(??) under the connecting homomorphism for the path space fibration. Moreover, if id : Sh+1 —» Sk+1 is the identity then a ~ uaold by the naturality of <9». Let bt : S2?{+'2 —» X and a,· : S2k'+1 —* X represent a basis of ?, (X) <g> Q. Multiplication of the b, and the ??^· defines a continuous map / = Fi) · (?^) : ?{52'·+1 x UjUS2^1 -> ??.
Sullivan Models 229 Proposition 16.7 Both ??,(/) <8> k and Ht(f;k) are isomorphisms. proof: Since 3c is a field of characteristic zero it is sufficient to prove this for Je = Q. Recall (Example 1, §15(d)) that 7r,(S2n+1) <8> Q = Q · [td52-+i]. It is immediate from this and the construction that ?»(/) <8> Q is an isomorphism. If ??' is simply connected the Whitehead-Serre theorem 8.6 asserts that H*(f;Q) is an isomorphism. To prove this when ??' is connected we may replace X by any other space with the same weak homotopy type. Let ? be an Eilenberg-MacLane space of type (?2(?"), 2), and choose a cellular model X' for X (§l(a)). Choose a map ? : X' —> ? such that ?-2(?) = identity (Proposition 4.20) and convert ? to a fibration (§2(c)). In this way we reduce to the case that there is a fibration q : X —> ? whose fibre, F, is 2-connected. Write the infinite product above as ? ? S} ) x ?, where ? is the weak product \ i / "of the odd spheres of higher dimension and the loop spaces ?52*?+1. Then we may construct / so that it restricts to /i : ? —y ClF with ?»(/?) ® Q also an isomorphism. By the Whitehead-Serre Theorem 8.6, i2,(/i;Q) is an isomorphism. On the other hand, Clq : ??' —> UK is an ?^-?^?????? (§2(b)). Let /2 be the restriction of / to f] 5| and put g = {iiq)fc -YlSj —> ClK. Then define a i i map of ?^-?^^????? f[S}xQF 5 *?? by setting h(x,y) = f(x) ¦ y. Since h ? (id x/i) = /, it is sufficient to prove that H*(h;Q) is an isomorphism. This will follow from Theorem 8.5 if H*(g;Q) is an ~ / \ / ? isomorphism. But we can write fl sl = ?' ??,? = UK ? ? ?,2 I. Hence i V ? / \ i J g ~ ?^?, where 7ro(<?i) = ^1E). In particular, ^2E1) ® Q is an isomorphism, and so the Whitehead-Serre theorem asserts that Ht(Qgl;<Q) is an isomorphism too. D In summary: the map / is a rational homotopy equivalence. Thus loop spaces have the rational homotopy type of a weak product of odd spheres and loop spaces on odd spheres. Moreover Proposition 10.7 has the following Corollary Let X be any simply connected topological space. Then there is a rational homotopy equivalence of the form ? ?? -»¦ ??'
230 16 The loop space homology algebra where ?,(Kn) ®Q is concentrated in degree n. ? We now suppose that H*(X;Q) has finite type. Then (Remark 1 following Theorem 15.11) ?*(?) ® Q has finite type and there are finitely many odd spheres (or loop spaces on odd spheres) of any given dimension in the weak product of Proposition 16.7. Thus a simple modification of the argument for finite products (Example 2, §12(a)) shows that the minimal Sullivan model for QX is the tensor product of the minimal Sullivan models for the S2fi+1 and the QS2k} + 1. In particular the form of a minimal Sullivan model for ??' and formula A6.4) could equally well have been deduced from Proposition 16.7. Next, define homology classes /?,· and ay in H*(QX;k) by setting ?{ = Ht(bi)[S2ei+1] and ay = tf.(ay)[S2H A homology class in H*(UX; ic) is called distinguished if it can be written in the form h < ¦¦¦ <ip, ni,...,nq G N, with ?, q > 0. (If ? = 0 there are no /Vs in the expression, and if q — 0 there are no ay's. When ? = q = 0 there is a single class 7 = 1.) Proposition 16.8 The distinguished homology classes ??? ?ip -?^1 ?"9 are a basis of the graded vector space Ht(QX;]k). proof: Recall that oy ~_ (Qa.j)Jd, where id : S2ki —> CiS2hj+1 represents d.[idsns+i]. Denote Ht(Id)[S2k^] simply by [5,-] G ?.(?52*?+1; Jfe). The classes 1, [Sj], [Sj]2, ¦.. are a basis for Ht(ilS-k:i+l ; h), as we showed in the Example at the start of §16. Denote [52<i+1] simply by [Si], so that 1 and [5,] are a basis for H»{S2ti+l;k). Then (Eilenberg-Zilber) the classes are a basis for Hm ? s2Ii+1 x ? ??2*^1; ? . This basis is mapped by H.(f) \i 3 J onto the set of distinguished homology classes in H*(QX;lk). ? (d) The primitive subspace of H*(QX;Jk). Again suppose X is simply connected with rational homology of finite type. Recall from §16(a) that H(m) : AV ? ?*(??;&), and that if [z],[w] G H*(nX;k) and ? G H+(QX;k) then ([z)U[w],a) = {[z]®[w],H(A)a). Thus ? is primitive if and only if {[z] U [w], a) = 0 for all [z], [w] G ?+(??; Se), i.e., if and only if {H{fn)(A^2V),a) = 0. There follows
Sullivan Models 231 Lemma 16.9 A non-degenerate pairing between V and P*(QX:k) is given by (v,a)>—> (?(??)?,?). ? Next note that if / : Y —> ? is a continuous map between path connected topological spaces, then #*(/) restricts to a map P*(Y;k) —> Pr(Z;k), be- cause the Alexander-Whitney diagonal is natural. Moreover, [Sk] € Hk(Sk;]k) is primitive because Hi(Sk;Jk) = 0, 1 < i < k — 1. Hence if a : Sk -> Y then hurY([a\) = H*(a)[Sk] is a primitive element in H*(Y;k). Thus we may regard the Hurewicz homomorphism as a linear map 7r.(y) —>P.(y;Jfc). Recall that the topological space X we are considering is simply connected and has homology of finite type (hypothesis A6.1)). Theorem 16.10 (Cartan-Serre) The Hurewicz homomorphism extends to an isomorphism proof: We begin with three simple observations. (i) hur : it* (S2e+1 ) <g> Jfe A P,(S2/+1; Jfc), i > 0. Indeed, the Hurewicz theorem 4.19 asserts this is an isomorphism in degrees < 2^+1. The homology of S2e+i vanishes in higher degrees D.14) as does ?»E2?+1) ® Jfc (Example 1, §15(d)). (ii) hur : 7T*(OS2fc+1) ® Jfc ? ??(?52?+1; Jfc), fc > 1. Since 7r*(OS2fc+1) ® A S ?*+?E2?:+1) ® h we conclude, as in (i) that this graded vector space is concentrated in degree 2k. Thus the Hurewicz Theorem 4.19 asserts that hur : ?*(?52?'+1) ®?? ?2*(?52*+1;.?:). In the Example at the start of §16 we saw that H*(QS2k+l;Ic) = k[a} with dega = 2k, and that ?(?)?? = ? [ ¦ ) a* ® an~l- This shows that an is not primitive for ? > 2, i=o V z / so that ?.(?52*+1;?) = ?2*(?52?+1; Jfc). (Hi) If Y and ? are path connected topological spaces, then ?7??) = ??G)8?»(?) and P*(Y ? Z;k) = P*{Y;k) 0 P*{Z;k). The first assertion is immediate from the definition of ?*. For the second, recall that Eilenberg-Zilber induces an isomorphism H*(Y; k)®H*(Z; k) -=*¦ H*(Y ?
232 16 The loop space homology algebra Z;k). It follows easily from formula D.9), §4(b) that this is an isomorphism of graded coalgebras. Let ? : H*(—;k) —> k be the augmentation induced by the constant map to a point. Thus (id -??) ® {?? · id) : [H.(Y; k) ® H.(Z;k)} ® [Hr(Y; k) ® H*(Z;k)} —> H.(Y;k)®H.(Z;k). But if 7 € Hr(Y;k) <S> Hr(Z;k) and if ?® denotes the comultiplication in the tensor product coalgebra then clearly (id -??) ® (?? ¦ id) ? ?® G) = y. Hence if 7 is primitive then 7 € (if+(F;J;) ® 1) ? A ® ii+(Z;A)) and so 7 € Now recall the continuous map / : ? S'2ei+1 ? ? fiS2fc'+1 —* ?? » j of Proposition 16.7 and denote this, for simplicity, by / : Y —> (IX. The three observations above imply that hur :?.(?)® Jfe Apt(y;i). Since ?*(/) ® A is an isomorphism and since the isomorphism H,(f;k) neces- sarily restricts to an isomorphism P*(Y; k) -=+ ?*(??; k), the theorem follows. (e) Whitehead products, commutators, and the algebra structure of H.(flX;k). Again suppose X is simply connected with rational homology of finite type. The commutator of two elements ? and r in a graded algebra is the element [?, r] defined by [?,?] = ??- (-1)?^??^???. Recall (§16(c)) that as a graded vector space, H*(QX;k) has a distinguished basis consisting of the elements of the form ii < ¦ ¦ ¦ < ip, ni,... ,nq € ?. Thus the algebra structure of ??(?.?; k) would be completely determined if we knew the commutators [?u?i·], [?uocj] and [uj.uj·.].
Sullivan Models 233 (Since deg?i is odd, ?f = \[?u?i].) We shall express these simply and explicitly in terms of Whitehead products (§13(e)), using the linear map ? = hur od* : ?*(A') —>¦ #*_?(??; Jfe). Proposition 16.11 J/70 € iTk+i(X) and 71 € ??+?(?) then proof: In Proposition 16.6 we showed that the quadratic part of the differential in ??? was given by di d0 ? = ?^(-1)?^^(?????) ? (doC*i), « € V, i where ^u = ? ® 1 + ???- ® ?? + 1 <g> ?. Since d0C is dual to ? (Proposition 15.14) and since ? is dual to the multiplication in ?*(??; Je), we have {?(??)?,[???,???}) = {?(??)?, (o )?? ® ii(m)*i ^?··<#(??)?;, 07o)(ii(m)*<J On the other hand, we may apply Proposition 13.16 where we calculated the Whitehead product to obtain (H(m)v; 0[7o,7iM = (-l)k+n+1 (dov; [7o, The Proposition follows from these formulae, given the duality between V and P»(nX;ic) established in Lemma 16.9. D Finally, suppose given ?*(?) <8> h together with the Whitehead products. De- fine a graded vector space ?? by setting (Lx)k = 7r^+1(A') ® k and denote by
234 16 The loop space homology algebra sx € ?*(?) (g) k the element corresponding to ? € ??. In the tensor algebra TLx let / be the (two sided) ideal generated by elements of the form (-l)^***-1 ([sx,sy]w). A6.12) Theorem 16.13 The Hurewicz homomorphism extends uniquely to an isomor- phism of graded algebras proof: First observe that an identification ?,? = ir*(iiX) ® Je is specified by identifying ? with dtsx, ? € Lx. This identifies hurcix as a linear map Lx —> H*(QX;]k), which then automatically extends to a morphism TLx —> H*(QX;]k) of graded algebras. Proposition 16.11 asserts precisely that the el- ements A6.12) are in the kernel of this morphism. Hence it factors over the quotient TLx —> (TLx)/I to define a morphism of graded algebras ? : (TLx)/I —> H.(ClX]k). We have to show that ? is an isomorphism. As in §16(b) let bt : S2ei+2 —» X and a.j : S2k>+1 —» X represent a basis of Tt*(X)®k. Put yi = s^] and Xj = s" [a,·]. Thus {yi,Xj} is a basis of Lx. Recall that TSLX = Lx <g) · ¦ ¦ ® Lx (s factors). Let Fs C (TLX)/I be the image of T~sLx; elements in Fs have filtration degree s. By the very definition of 7, if z,w,? Lx then zw - (-l)des-deswwz has filtration length 1. In particular if ? has odd degree then z2 = \{zz + zz) has filtration degree 1. An obvious induction on filtration degree now shows that every element in (TLx)/1 is a linear combination of the elements i\ < ¦ ¦ ¦ < ip Vix yip' ?7? *J * ii < " " " < h nt G N. (Show that the subset with ? + ??^ < s spans Fs.) The morphism ? maps these elements to the distinguished basis of ?*(??; &) established in Proposi- tion 16.8. Hence the elements above must be a basis of (TLx)/I and ? must be an isomorphism. D Example 1 The homology algebra H,(USn+1;3z). Let a: Sn —> QSn+1 represent d.[idSn+i], and let ? = hur[a] = H*(a)[Sn] ? Hn(CtSn;h). We show that the inclusion of ? extends to an isomorphism k[a] A H,(QSn+1 ; Jfe)
Sullivan Models 235 of graded algebras. If ? is odd this has already been done in the example at the start of §16. Suppose ? = 2k. Let ? = [ids2"] € n2k(S2k). Then n*(S2k) <g> Jk is two dimensional, with basis given by t and [i,i]w (Example 1, §15(d)). Apply The- orem 16.13 to obtain an isomorphism T(x,y)l (x2 - \y) A H,(US2k;]k). Then note that T(x,y)/ (x2 — \y) = T(x) = k[x], and that this isomorphism sends i^a. D Exercises 1. Consider the graded vector space V — {Vi}i>o and its graded dual W = Horn (K,Q). Let ??)9 denote the subgroup of permutations ? of the set {1,2,..., ? + q} such that: ?? < ?2 < · · · < ?? and ? (? + 1) < ? (? + 2) < · · · < ? (? + q) . Define a product (called the shuffle product) on TW by: [lui <g> W2 ® ¦ ¦ ¦ ® Wp] € TPW , [w[ ® iiig ® · · · ® I«,] E/'_1V ??/) ??) ill ??) · · · ?? 7/) I V L) \Wcl^>Wc2^ <& Wa(p+q)\ ?€??,, if 1 < ?? < ? Here so denotes the graded signature. Prove that the coproduct defined on TV in §3-exercise 4 dualizes the product * on TW. 2. Determine the algebra H,(QCPn;Q) for ? > 2. 3. Let X be a finite product of simply connected Eilenberg-MacLane spaces. Determine the algebra H,(ClX; Q). 4. Let X be a simply connected homogeneous space. Prove that all triple Whitehead products are trivial in ?*(?) 5. Determine, up through dimension 5, the Sullivan minimal model of X = sl v Si U[ai[MMH, D6. Does the subalgebra H.(U(S2 V S2)); Q) generate the algebra H, (QX;Q)? 6. Prove that, up through dimension 5, the Sullivan minimal model of S2\/S2\/S5 is given by ? ? = f\(x1,x2,X3,yi,y2,y3, zu z2,....) with V — {V/1} and Xi,x2 € Km *3 € I/05, dVp C (AK)?!/, ctyi = x? .d^ = XiX2,dy3 = x\, dzx = xlV2 - x2Ui ,dz2 = Xiy2 — x2y2 (see §13-exercise 4). Compute the Whitehead products
236 16 The loop space homology [a,?}w = 0 for every ? € ?2E2 V S2 V S5) <g> Q and ? € ?3E2 V S2 V Let D : F -» ? ? be a linear map such that D - d{Vi) C (AF)<i_2. ? D extends to a differential on AV (cf. exercise 5). 7. Let V = Ikx ? Iky be the graded vector space generated by the ? € Vp and y € F,, p,c > 2. Prove that ((ifc ® V) ®T(sV),d), with by cte = 0 = dy and for saisa^.-.sak € jTfe(sV), d(sa\sa2---sak) — &\ ® is a r(sF)-semifree resolution of the field ife. Deduce that isomorphic to the graded vector space T(sV).
17 Spatial realization In \17(a), the ground ring is an arbitrary commutative ring, k. In the rest of §i7, k is a field of characteristic zero. Occasionally we specify k = Q. We began Part II with the construction, in §10, of the (contravariant) functor Apl from topological spaces to commutative cochain algebras. Now, at the conclusion of Part II, we reverse the process, and construct the (contravariant) spatial realization functor | | : commutative cochain algebras ~-> CW complexes. Recall that a Sullivan algebra (AV, d) is simply connected of finite type if V = {Vp}p>o and each Vp is finite dimensional. When k = Q the spatial realization functor has the following important properties: • Any simply connected rational Sullivan algebra of finite type is a Sullivan model of its spatial realization: (AV,d) -^ APL(\AV:d\) , and this spatial realization is a simply connected rational topological space (Theorem 17.10). • Any morphism ? : (AV, d) —> (AW, d) of simply connected rational Sulli- van algebras of finite type is a Sullivan representative of its spatial realiza- tion \?\ : \AV,d\ <— \AW,d\, A7.14). • Two morphisms ?, ? : (AV, d) —> (AW, d) between simply connected ra- tional Sullivan algebras of finite type are homotopic if and only if the con- tinuous maps \?\ and \?\ are homotopic (Proposition 17.13 and §12(c)). • Let ? : (AV, d) —> (AW, d) be a rational Sullivan representative for a continuous map / : X —> Y between simply connected CW complexes with rational homology of finite type. Then there is a homotopy commutative diagram X i Y hy \AW,d\ \AV,d\ \v\ in which ?*(/?.?) <S> Q and 7r*(/i>') ® Q are isomorphisms (Theorem 17.15). From these properties it is easy to deduce (and we leave this to the reader) the bijections .. , , . ? ( isomorphism classes of ? rational homotopy | ^ I ¦ ¦ ,?. n- 17 I r —>¦ \ minimal Sullivan algebras > 1 J 1 over and
238 17 Spatial realization {homotopy classes of ^j ( homotopy classes of I continuous maps of > -^ < morphisms of Sullivan algebras > rational spaces J [ over Q J promised at the start of §12. (Note that for these bijections we restrict to simply connected CW complexes with rational homology of finite type and to simply connected Sullivan algebras of finite type.) The spatial realization functor, | |, is constructed as the composite of two others. The first is Sullivan's simplicial realization functor [144] { ) : commutative cochain algebras -w simplicial sets , which is the adjoint of Apl(-), and the second is Milnor's realization functor ([126]) | | : simplicial sets -w CW complexes . In describing Milnor's functor we shall only sketch some of the proofs, referring to May's elegant exposition [122] for details. We complete this section by using integration to define a natural quasi-isomorphis C.{AV,d) -=>¦ Kom((AV,d),]k) , compatible with geometric products. This section is organized into the following topics: (a) The Milnor realization of a simplicial set. (b) Products and fibre bundles. (c) The Sullivan realization of a commutative cochain algebra. (d) The spatial realization of a Sullivan algebra. (e) Morphisms and continuous maps. (f) Integration, chain complexes and products. (a) The Milnor realization of a simplicial set. Recall the definition of a simplicial set from §10(a), with face and degeneracy maps di and Sj. Recall also that in §4(a) we introduced the geometric face and de- generacy maps ?, = (e0 · · · ?, · · · en) : ?" —» ?" and q, = (e0 · · · e^e,· · - · e„) : ?"+! —> ??. Now let ? be a simplicial set. Give each Kn the discrete topology. Then the Milnor realization of ? is the topological space \K\=
Sullivan Models 239 where ~ is the equivalence relation generated by the relations diO ? ? ~ ? x XiX , ? € Kn+l, ? € ?" and Sj-? ? ? ~ ? x ?jX , ? € ifn, x € ??+1 . This construction is functorial: if / : ? —> L is a morphism of simplicial sets then / = {fn : Kn —> Ln} and the continuous maps fn ? id : Kn ? ?" —> Ln ? ?" factor to define the continuous map \f\:\K\->\L\. Recall that the face and degeneracy maps of a simplicial set satisfy the com- mutation relations A0.2). A simplex ? € Kn is degenerate if ? = SjT, some r € AVi-i! otherwise it is non-degenerate. The set of non-degenerate n-simplices is denoted NKn (cf. §10(a)). Recall further that dAn = {j\i(An-1) and put ?" = ?" - dAn. i Lemma 17.1 [122] The quotient map ?? ¦ \\Kn ? ?" —> \K\ restricts to a ? bijection qK : U NKn ? ?" -> \K\ . proof: Denote qk simply by q. If ? ? ? € Kn ? ?" choose the least k such that q(a x x) = q(r x y) for some (r, y) € iv^ ? ?*1. Then r cannot be degenerate (if ? r = Sjcj then r x y ~ ? ? ^y) and y € ? (if y = AjZ then ? x y ~ dir x ?). Hence q is surjective. A tedious computation using the commutation formulae A0.2) shows that q is injective. D Suppose L C ? is a sub-simplicial set. If ? € Ln and ? = sjt, ? € -f^n-i then r = djSj-r = (9,-? e ??-?· Thus it follows from Lemma 17.1 that \L\ C \K\. In particular, recall from §10(a) that the n-skeleton of ? is the sub simplicial set K(n) determined by NKk k < ? Thus \K(n)\ C\K\. Proposition 17.2 [122] The Milnor realization \K\ of a simplicial set ? is a CW complex with n-skeleton \K(n)\ and ?-cells the non-degenerate n-simplices ? € NKn. The attaching map for ? is the restriction of qK to {?} ? dAn. proof: Let ? € Kn. The quotient map q : ]\Kn ? ?" —» \K\ satisfies q(a ? ? ???) = q(dia ? ?) and it follows that q restricts to a continuous map qa :
240 17 Spatial realization {?} ? duJ1 —> \?{?— 1)|. Thus, because of Lemma 17.1. q induces a continuous bijection q(n):\K(n-l)\U{qa) ( ]J {?} ? ?" J -> \K(n)\ . A straightforward check shows that the q(n) are proper, and hence homeomor- phisms. Thus the q(n) define a continuous bijection to \K\ from a CW complex with the desired properties. This is also proper, and hence a homeomorphism. D Example 1 |?[?]| = ?". In §10(a) we introduced the simplicial set ?[?] whose fc-simplices are the linear maps (et0 · · -eik) : Ak —» ?", 0 < i0 < ¦ · ¦ < ? < ? . The identity map of ?" is thus an ?-simplex ? of ?[?], and it is straightforward to check that qA[n) restricts to a homeomorphism {i} ? ?" -^ |?[?]|. D Example 2 Cones. The 'simplicial point' is the unique simplicial set with a single fc-simplex cn in each dimension. Now suppose ? is any simplicial set. We extend the face and degeneracy maps in {Kn ]J{cn}}n>0 to the sequence of sets {CK)n = Kn U H Kk X {Cn.u.,} U {Cn} \fc=l / by setting (for ? € Kk)'· di{a,cT) = < ? ,if r = 0 and i = A: + 1. ^ (?, cr_i) , otherwise. and (?, cr+i ) , otherwise. This defines a simplicial set CK, the cone on ?. It follows now from Lemma 17.1 that \CK\ = \K\ ? ?/\?\ ? {1}. In particular, • for any simplicial set ? the realization \CK\ of the cone on ? is a con- tractible CW complex. D Example 3 Extendable simplicial sets. Recall from §10(a) that a simplicial set ? is extendable if for any ? > 1 and any X C {0,. ..,n} the following condition holds: given ?* € A"n-i5 i € J and satisfying <9*?,· = ?,-??*, i < j, then there exists ? € Kn such that 9?·? = ?,·, i € I. We observe now that
Sullivan Models 241 • If K is an extendable simplidal set then \K\ is contractible. Indeed, extendable simplicial sets ? have the property that if ? is sub sim- plicial set of a simplicial set L then any morphism ? : ? —> ? extends to a morphism L —> K. (This was proved in Proposition 10.4(ii) for simplicial cochain complexes, but the argument applies verbatim to simplicial sets.) In particular the identity of ? extends to a morphism ? : CK —> K. Thus id\K\ = \?\ ? |?| where ? : ? —> CK is the inclusion. Since \CK\ is contractible, id\K\ ~ constant map; i.e. \K\ is contractible. ? Let ? be a simplicial set. Just as we defined the cochain algebra C*(K) in §10(d) so we now introduce the differential graded coalgebra, C,(K), defined as follows: • C,{K) = {Cn(K)}n>0 and Cn{K) is the free lr-module on Kn divided by the submodule spanned by the degenerate simplices. • The comultiplication ? in C*(K) is given by ? ?? = ^ dpdp+i ¦ ¦ ¦ dna <g> d0 ¦ ¦ ¦ doa , ? € Kn . • The differential d is given by ? da = Y^{-l)pdpo , ? € Kn . p=0 The reader will immediately see that this generalizes the construction in §4(a): if X is a topological space then C»(X) = C*(S*(X)). Moreover C*{K) is the cochain algebra Iiom(C*(K), k), exactly as in the case of topological spaces. If ? is any simplicial set then the continuous map qx of Lemma 17.1 restricts to continuous maps qa : {?} ? ?" —> \K\, ? € ??. This defines an inclusion of simplicial sets, ?? : ?-* S.{\K\) , ??':?^??? where S*(—) denotes the simplicial set of singular simplices defined in §10(a). Observe now that ?? maps the non-degenerate simplices of ? bijectively to singular simplices of \K\ that are the characteristic maps for the cells. This identifies ?*(??) as a cellular chain model in the sense of Theorem 4.18. In particular, ?*(??) is an isomorphism. Note as well that ?.(??) : C.(K)-* C.(\K\) is an inclusion, and identifies C*(K) as a sub dgc of C*(\K\).
242 17 Spatial realization On the other hand, Let X be any topological space, and define a continuous map U Sn(X) x ?" —)¦ X by sending ? ? y ?-4 o(y). This factors over the ? quotient map qs.(X) to produce the continuous map sx:\S.(X)\-*X. It is an interesting but reasonable exercise [122] to show that sx is always a weak homotopy equivalence. We shall limit ourselves here to the easy case that X is a simply connected CW complex. Proposition 17.3 (i) If X is a simply connected CW complex then s ? is a homotopy equivalence. (ii) If ? is a simplicial set and \K\ is simply connected then \??\ is a homotopy equivalence. proof: (i) For simplicity put Y = \S*(X)\ and ? = CS.(X) : S,(X) —> 5,A'). We show first that Y is simply connected. Indeed, since X is connected any two points x, y € X are connected by a singular 1-simplex / : (/,0,1) —> (X,x,y). Thus any two 0-cells ?(?) and ?(?/) in Y are connected by the l-cell ?(/); i.e., ? is path connected. If g : A,0,1) —> (X, y, z) we define i{2t) ^<?<?· Similarly, f~l (t) = /(I -1). The homotopy class of / rel {0,1} is denoted by [/] and an associative composition of homotopy classes is defined by [/]*[<?] = [/*<?], — cf. §1. Fix a 0-cell, ???, as a base point of Y. The Cellular approximation theorem 1.2 asserts that any loop in Y at ??0 can be deformed into the 1-skeleton ??. The easy part of the Van Kampen theorem asserts that any element 7 6 ??(??,???) can be written in the form [?/?]*1 * · · · * [?/?]*1 where the ?/j are 1-cells in Y. There are obvious singular 2-simplices and // \5 ? > X const / * g
Sullivan Models 243 and ?? and ?? are singular 2-simplices in Y that provide homotopies ?/ ~ (?/)-1 rel {0,1} and ?/*?9 ~ ?(/ * 5) rel {0,1}. Let 7 = tf1 *¦¦¦* f±\ where we have removed the (necessary) brackets for simplicity. Then / is a loop in A" at x0 and our observations imply 7 = [?/]. Since AT is simply connected there is a singular 2-simplex const ? > X const and ?? is a homotopy from ?/ to the constant loop. Hence ??(?') = {0}. Next recall that /7»(??·;?) is an isomorphism for any simplicial set K. In particular, ?,(?;?) is an isomorphism. But the continuous map sx : Y —> X induces S*(sx) : S,(Y) —> S*(X), and it is immediate from the definitions that ? : S»(X) —> S*(Y) satisfies 5»(??)? = id. Thus H,(sx;Z) is an isomorphism. Since a homology isomorphism between simply connected spaces is a weak homo- topy equivalence (Theorem 8.6) and a weak homotopy equivalence between CW complexes is a homotopy equivalence (Corollary 1.7), assertion (i) is established. (ii) Assertion (i) states that s\k\ : |5»(|/sT|)| —> \K\ is a homotopy equivalence. It is immediate from the definition that s^ ? \??\ = id. Thus |^| is a homotopy equivalence. D (b) Products and fibre bundles. The product of two simplicial sets ? and L is the simplicial set ? ? L = {{Kn ? Ln},di ? di,Sj ? Sj). If ? -^> ? <— L are morphisms of simplicial sets then the fibre product ? ? ? L C ? ? L is the sub simplicial set defined by (K xE L)n = {(?,?) | ?? = ??}. Now the projections qk ¦ ? ? L —> ? and ql : ? ? L —> L are morphisms of simplicial sets. Set f = (\QK\,\QL\)'-\KxL\-*\K\x\L\. Proposition 17.4 (i) The continuous map f is a natural homeomorphism \K ? L\ —* \K\ ? \L\. (ii) Furthermore, f restricts to a homeomorphism \K Xe L\ -^ \K\ ?|?| \L\. proof: We first introduce some notation. Given a surjective map ? : {?,.,.,?} —> {?,...,*} such that 0 = ?@) < ··· < ?(?) = k we put
244 17 Spatial realization Qa — (eQ(o) " ' " ea(n)) '· ?" —> Ak. Then ?? is a composite of degeneracy maps: ?? = ??} ? · · · ? Qin_h, and in any simplicial set ? the map sa = Sin_k ? · · · ? s,a : R'k —> A« depends only on a and not on the choice of decomposition (use the commutation formulae A0.2)). Moreover the map qx of Lemma 17.1 identifies saa ? y with ? ? Qay. We now turn to the actual proof. (i) It is immediate that / is proper and so we only need to show it is bijective. We do so via Lemma 17.1. Any simplex in (? ? L)n has the form ? = (saa, spr) with ? € NKp and r € NKg and it is degenerate precisely if sa = Sisa', and S? = SiS?' for some {. This is equivalent to: a(i) = a(i + 1) and ?(i) = ?(i + 1). Thus the non-degenerate simplices are the (sao, spr) with a(i) + ?(i) < a(i + l) + /0(i + l), all*. Now / : {?} ? ?" —> ({?} ? ??) x ({?} ? ??), where ? = ?(?,?) is the ? (? ?) ? ° dimension of ?. We denote this restriction by fa>? : ?" —'—> ?? x Aq- We have to show that each fa? is injective and that (a'0)), A7.5) ?,? where the ?, ? run over all the non-decreasing sequences such that ? = (saa, S?r) is non-degenerate. Regard ?? ? ?? as a subset of W+1 ? W+1. Then fQi0 is the linear map defined by e, ?-> (eQ(t). e^(,)), 0 < i < ?(?,?). Since either a(i) < a(i + 1) or ?(i) < ?(i + 1) it is easy to see that {ea(Q),e?@)) ,..., {ea{n),e?{n)) are linearly independent in W+q+2. Hence each fa? is injective. Formula A7.5) is a straightforward computation. (ii) By definition / restricts to a continuous injection fs '¦ \K ? ? L\ —> \K\ X|?| \L\ and this is proper because \K Xe L\ is closed in \K ? L\. We have only to show that /# is surjective. Suppose that ? = (saa, S?r) is a non-degenerate ?-simplex in ? ? L (as de- o scribed in the proof of (i)) and that y G ??· Then fqKxL&xy) = (??'(? x gay), qh(T x Q?V))· Suppose this point is in \K\ X\e\ \L\· Then ??(?? ? Qay) = qE^ x Q?y) ¦ Write ?? = sa>o' and ?? — S?>r'', where ?' and ?' are non-degenerate in E. The equation above gives ? ? But ?? '· ?" —> ?? and similarly for Q?, ??< and Q?<. Thus Lemma 17.1 ? implies that ?' - ?' and ga'Qay - Q?'Q?y. Now y - ^ f,-et· and the condition ? ? y G ?" means each ij > 0. It follows that the equation Qa<QQy — Q?'Q?y implies
245 Sullivan Models Qa'Qa = Q?'Q?-, whence sasa' = S?Sp·. Thus ipsaa = saSa'&' = S?S?>r' = i.e. ? € {? xE L)n and gxxz,(^ x y) € \K xE L\- Recall from §2(d) that a fibre bundle with fibre ? is a continuous map ? : X —> Y such that for some open cover {Ua} of Y we have: p~l{Ua) — Ua x Z, compatibly with the projection on Ua. There is a useful simplicial analogue of this "local product structure." Let L be a simplicial set. As we observed in §10(a) any simplex ? € Ln determines a unique simplicial map ?* : ?[?] —> L such that ?* (id^*) = ?. Thus given a morphism of simplicial sets ? : ? —> L and ? € Ln, we can form the fibre product ?[?] ? ?, ?. We define a simplicial fibre bundle with fibre F to be a morphism ? : ? —> L of simplicial sets such that for each ? > 0 and each ? € Ln there is a commutative diagram ?[?] x F ^ ?[?] ??, ? in which <?>? is an isomorphism of simplicial sets. Remark Note that a simplicial fibre bundle is not a simplicial object in the category of fibre bundles! Proposition 17.6 If ? : ? —> L is a simplicial fibre bundle with fibre F then \?\ : \K\ —> \L\ is a fibre bundle with fibre \F\. proof: Recall from §17(a) that the cells of \L\ are the non-degenerate simplices ? of L and that the characteristic maps are the maps qa : {?} ? ?" —> \L\. It is immediate from the definitions (cf. Example 1 of §17(a)) that Thus the pullback of |?| to'an ?-cell is just the fibre product |?[?]| Xj^j \K\ defined with respect to |?,| : |?[?]| —> \L\ and \?\ : \K\ —> \L\. Now apply Proposition 17.4 to translate the simplicial diagrams above to the commutative topological diagrams \A[n}\
246 17 Spatial realization These show that \?\ pulls back to a product bundle with fibre |F| over every cell. Thus Proposition 2.7 asserts that \?\ is a fibre bundle with fibre \F\. ? Example 1 A simplicial construction of the Eilenberg-MacLane space K(k, 1). Recall that k is our ground field of characteristic zero. Here we think of it simply as an abelian group under addition. In §10(c) we introduced the simplicial cochain algebra Apt = {{Api)n}. Thus each APPL is a simplicial vector space. Let d be the differential in APL; then each (kerd)p is a sub simplicial vector space of APPL. Put ? = (kerdI. We shall show that • \Z\ is an Eilenberg-MacLane space, K(k,l). For this let [k] be the simplicial vector space given by [k]n — k, di — id, Sj = id. Since H((APL)n,d) = k,n> 0, (Lemma 10.7), 0 —>· [k] -)/????-»0 is a short exact sequence of simplicial vector spaces. Let ?» : ?[?] —> ? be the simplicial map determined by any fixed ? € Zn. We show that there is an isomorphism ??:?[?] ? [*] S ?[?] xZA0PL, coherent with the projections on ?[?]. For this, fix r € (Apl)u satisfying dr = ?. If ? G ?[?]& and ? € [k]k = k, define ??(?,?) = (?,?*(?) + ?). It is immediate from the definition that ?? is a simplicial isomorphism. We may now apply Proposition 17.6 to conclude that |d| : \APL\ -> \Z\ A7.7) is a fibre bundle with fibre |[ifc]|. The elements of [k]o (= k) are the only non- degenerate simplices, and this identifies \[k]\ = k, equipped with the discrete topology. (For example, if k = M. or Q, this is not the standard topology!) Thus \d\ is a covering projection with discrete fibre, k. Since A°PL is extendable (Lemma 10.7), I^J is contractible (Example 3, §17(a)). Finally, note that addition in each (Apl)^ defines a simplicial morphism A°PL ? APL —> A°PL, which restricts to a : A°PL ? [k] —> APL. This realizes to a continuous map \ct\:\A°PL\xh-*\A°PL\ . In the same way addition defines a simplicial morphism [k] ? [k] —> [k], which realizes to ordinary addition in k. Thus |a| defines an action of the additive group of Jfc on \APL\ and this identifies k with the group of covering transfor- mations of the covering projection A7.7). Since \APL\ is contractible it follows that JcS7Ti(|Z|) and ??>2(|?|) = 0. Thus \Z\ is a K(k,l)· It is useful to identify the isomorphism k -^ ???(|?|) explicitly. For this fix as basepoints of \Z\ and l^lp^l the 0-cells zo and ao corresponding to the zero
Sullivan Models 247 elements of Zo and (APL)°. Note also that each ? € Jt = {???)°0 determines a 0-cell ax in 1.4^ |. Now recall (§10(c)) that (APLI = \(tudti). For each ? G k the element Xdti € Z\ defines a 1-cell in \Z\ that is a loop g\ at zq. Its lift is the 1-cell in \APL\ corresponding to ??? € {???)? and this is a path /? from uq to ??. Thus • The isomorphism h -^ ?? (|Z|,zo) is given by ? ?-> [g\\. ? (c) The Sullivan realization of a commutative cochain algebra. Recall again from §10(c) the simplicial cochain algebra (ApL,di,Sj). It may be thought of as a sort of 'generalized point'. From this perspective the Sullivan realization of a commutative cochain algebra (A, d) is analogous to the construc- tion of an algebraic variety Vr from a noetherian commutative algebra R. The points of Vr are just the algebra morphisms R —)¦ Ik. Analogously, we make the Definition The Sullivan realization is the contravariant functor (A, d) n^ (.4, d) from commutative cochain algebras to simplicial sets, given by: • The n-simplices of (A,d) are the dga morphisms ? : (A, d) —> {Api)n. • The face and degeneracy operators are given by d{O = d{ ? ? and s ja = Sj ? ?. • If ? : (A, d) —> (?, d) is a morphism of commutative cochain algebras then {?) : (A,d) <— (B,d) is the simplicial morphism given by (?)(?) — ? ? ?. ?€(?,?)?- Denote by DGA(—, —) and by Simpl(—, —) the set of morphisms between two dga's, and between two simplicial sets. (Thus <^4, <i>„ = DGA ((A,d), {Api)n).) Recall from §10(b) and §10(c) that if if is a simplicial set then Api(K) = Simp^if, Apl) is the cochain algebra of simplicial morphisms from ? to Api- Thus a natural bijection OGA((A,d),APL(K))^Simp\(K,(A,d)) , ? >—> f , A7.8) is defined by requiring /(?)(?) = ?{?){?) , a 6 ?, ? 6 Kn, ? > 0 . This establishes ???(—) and ( ) as adjoint functors between commutative cochain algebras and simplicial sets. In particular, adjoint to the identity of (A, d) is the canonical dga morphism VA:(A,d)->APL(A,d) . Combining Sullivan's functor with Milnor's realization gives the fundamental
248 17 Spatial realization Definition The spatial realization of a commutative cochain algebra (A, d) is the CW complex \A,d\ = \(A,d)\. The spatial realization of a morphism ? : (A,d) —» (B,d) is the continuous map \<p\ — \{?)\· Example 1 Products. Let {A,d) and [B, d) be commutative cochain algebras. A pair of morphisms ? : (A,d) —> {Api)n and ? : (B,d) —> {APi)n extend uniquely to the mor- phism ?"? : (A,d) <g> (B, d) —> (???)?· Thus restriction to A and restriction to ? defines an isomorphism of simplicial sets ((A, d) ® {B, d)) A (A, d) ? (?, d) . On the other hand, Milnor's realization converts simplicial products to topo- logical products (Proposition 17.4). Thus we obtain a natural homeomorphism \(A,d)®(B,d)\^>\A,d\x\B,d\ . ? Example 2 Contractible Sullivan algebras have contractible realizations. Suppose A = (A(U @dU),d) is a contractible Sullivan algebra, and let {ua} be a basis of U. An element of (A)n is a morphism ? : A —> (Apl)u, and so an isomorphism is given by ? ?—> {???}. By Lemma 10.7 each (Apl)\u\ is extendable, and hence so is the product (a trivial exercise). It follows (Example 3, §17(a)) that \A\ is contractible. ? More generally, we shall see now that spatial realization converts relative Sul- livan algebras (§14) to fibre bundles with geometric fibre the realization of the Sullivan fibre. (This is the obverse of the theorem in §15 that in the Sullivan model of a fibration the fibre of the model is a model of the fibre.) More precisely, fix a relative Sullivan algebra \:{B,d) —>(?®AV,d) such that Imd C (jB+®AV)e(A-®AF). Dividing by B+®W defines a Sullivan algebra (AV,d). Proposition 17.9 The continuous map |?| : \B (8) AV, d\ —> \B,d\ is a fibre bundle with fibre \AV, d\. proof: In view of Proposition 17.6 it is sufficient to identify (?) : E®AF, d) —> /R ?\ as a simplicial fibre bundle with fibre {AV, d). In other words we need to
Sullivan Models 249 show that for any ? € (B,d)n the corresponding fibre product (with respect to ?, : ? [?] —> (B,d)) satisfies ?[?] x{BA) (? ® ? F, d) 9? ? [?] ? <AV, d) , compatibly with the projections on ? [?]. Now ? is a dga morphism (B, d) —> (APL)n. Moreover, for any ? € A[n]k, q is a linear map ?*-' —> ?" and ?»(?) = APL(a) ? ?, as follows at once from the definitions. Thus the &-simplices in the fibre product above are the pairs (a, r) where ? € A[n]k and ? : {? <g> KV, d) —> {Api)k satisfy r ? ? = ???(?)?. Let {{???)? <g>AF,d) = (.4pz,)n <S>b (B <g> AV'.d) be the pushout relative Sul- livan algebra (§14(a)) defined via ?. Since (APL)n = .4Pi (?[?]) (§10(c)) there is a unique simplicial morphism, ?[?] —>· ((-4pz,)n), adjoint to the identity of (-4pi)n· Our observations just above identify ?[?] X(Bid) <?®AV,d> = ?[?] compatibly with the projections on ?[?]. On the other hand, ? ((APL)n) = k (Lemma 10.7). Thus the argument of Lemma 14.8, with k —> (APi)n replacing (Bi,d) —> (B,d), shows that the identity of (Apl)u extends to a dga isomor- phism (APL)n®(AV,d) = ((APL)n®AV.d). This identifies the right hand fibre product as (?[?] xpPl)n) ((APL)n)) x (AV,d) = A[n] ? (AV,d). ? (d) The spatial realization of a Sullivan algebra. Fix a Sullivan algebra (AV, d). By definition. V = {Vp}p>i and so there is a unique augmentation ? : (AV,d) —> k, which is then the unique 0-simplex in (AV, d) and the unique 0-cell in \AV, d\. Denote all these by ?. In §17(a) we introduced the inclusion of simplicial sets ? = ?(??,?) '¦ (?F, d) —> 5» (|AV,d|). Thus ???(?) is surjective (Proposition 10.4). As described there -for integral coefficients, ?*(?;1?) is an isomorphism, and so ???(?) is a quasi- isomorphism. Thus in the diagram Apl((, (AV,d) > APL((AV,d)) we may (Lemma 12.4) lift ?^?,?) through the surjective quasi-isomorphism ApL^) to obtain a canonical homotopy class of dga morphisms m(AV,d) ¦ (AV,d) -> APL (\AV,d\) . We also want to compare Hom^V, Jc) with the homotopy groups ?, (\.\V, d\). To do so we shall rely on the notation, constructions and observations of §13(c), often without explicit reference, as well as to much of §12, which we also take for granted here. Thus if ? : (Sn,x0) —> (\AV,d\,e) represents a G ?? (\AV,d\)
250 17 Spatial realization then Apl (a) ° m^v,d) lifts to a morphism ?? from (AV, d) to the minimal model of Sn, which has the form (A(e),0) or (A(e, e'),de' — e2) with dege = n. Furthermore, ?? restricts to a linear map Vn —> he which depends only on a, and hence determines a map ( ; ) :Vn ??? (\AV, d\) —>¦ k , ??? = (?; a)e . These maps are linear in Vn and, if ? > 2, additive in ?? (|AV, d\). Thus the maps C« :7rrt(|AV,rf|) -»· HomA(^A) , ??(?)(?) = (-1)>;?> , are morphisms of abelian groups if ? > 2. Theorem 17.10 Let (AV, d) be a Sullivan algebra such that Hl(AV,d) = 0 and each Hp(AV,d) is finite dimensional. Then (i) \AV,d\ is simply connected and ?? : 7rn(|AV,d|) -^ Romjk(V.It) is an isomorphism, ? > 2. (ii) If 3c = Q then m(Av,d) 25 ? quasi-isomorphism, m{AV,d) : (AV,d) -^ APL (|AV,d|) . When Jc = Q then Theorem 17.10 exhibits any simply connected Sullivan algebra of finite type as the Sullivan model of a CW complex. proof of Theorem 17.10: We first reduce to the case (AV,d) is minimal. Use Theorem 14.9 to write (AV,d) = (AW,d) ® (A(U ®dU),d) with (AW,d) minimal. Then |AV,d| is the product of \AW, d\ and a contractible CW complex (Examples 1 and 2, §17(c)). It is thus sufficient to prove the theorem for (AW, d). We may therefore assume that (AV, d) itself is minimal. This implies that V = {Vp}p>2 and that each Vp is finite dimensional. We shall rely on the following observation. Suppose (AU,d) —> (AU <g>AV,d) is a minimal relative Sullivan algebra in which (AU, d) is itself a minimal Sullivan algebra. As in Proposition 17.9 this realizes to a fibre bundle \AU (8) AV,d\ -^ \AU, d\ with fibre |AV, d\. In particular, |?| is a Serre fibration (Proposition 2.6). Let dt : ?, (\AU,d\) —> ?«_? (\AV, d\) be the connecting homeomorphism. Next, by adjointness we have the commutative diagram (AU,d) > (At/®AV,d) " (AV,d) APL ((AU,d)) — ApL ({AU <8) AV,d» — APL ((AV,d))
Sullivan Models 251 Use the Lifting lemma 14.4 to choose m(Ai/id), m.(,\u®AV,d) and m(AVj) so that the diagram (AU,d) (AU <g> AV, d) (AV, d) ApL(\AU,d\) — APL(\AU®AV\,d) — AF commutes. Now the argument of Proposition 15.13 establishes that (???;?) = (-?)?+?(?;3,?) , ? G Vn, a 6 nn+1(\AU\) . This translates to the commutative diagram nn+l(\AU,d\) d' ' 7rn(\AV,d\) A7.11) (Un+\k) —f ?????(??,^, in which d*0 is the dual of d0 : (d*of) (?) = (-l)n+1 f (???) ,? G Vn. proof of (i): We turn now to the proof of (i), which is accomplished in stages. Step 1: If deg ? = 1 then |?(?),0| is an Eilenberg-M acLane space K(k, 1), and ?\ is an isomorphism. Recall that (A(v),0)n consists of the morphisms ? : (?(?),?) —> (APL)n. Thus ?? G (kerd)Jj. Since the morphism ? is determined by ??, an isomorphism of simplicial sets (?(?),?) ^» (kerdI is given by ? ?-*· ??. Thus our assertion follows at once from the Example of §17(b). Step 2: If deg ? — ?, ? > 2, then |?(?),0| is an Eilenberg-M acLane space K(k,n), and ?? is an isomorphism. Consider diagram A7.11) specialized to the case of the contractible Sullivan algebra (?(?, w), dw = ?), regarded as the relative Sullivan algebra (?(?), ?) —> (?(?,??),?). Then dp is an isomorphism by inspection, O« is an isomorphism because |A(u, w),d\ is contractible (Example 2, §17(c)) and Cn-i -" ??_! (|?(??),0|) —> Rom(kw,k) = k is an isomorphism by induction. (Start the induction with Step 1!) Step 3: Assertion (i) holds if V is finite dimensional. We establish this by induction on dim V, the case dim V — 1 being Step 2. Write V = U © Jkv so that (AU, d) is a sub-Sullivan algebra and ?? G AU.
252 17 Spatial realization Thus (AU, d) —y (AU ®A(v),d) is a relative Sullivan algebra. Since (AV,d) is minimal, dQ = 0. By induction on dim V, ?, : fr»(|Ai/,d|) —> Hom(U,k) is an isomorphism. By Step 2, so is ?, : ??»(|?(?),0|) —> Hom(kv,k). Thus diagram 17.11 shows that d. = 0 : nm(\AU,d\) —> ?^^?^,?]). Now (by naturality) we have the row exact commutative diagram 0 — ??.(\?(?),0\) - 7T,(|AV',d|) 7r.(|At/,d|) — 0 0 — Eom(kv,Jc) ????(?',*) Eom(U,k) —~ 0 and it follows that the central arrow is an isomorphism. Step 4: Assertion (i) is true in general. Fix ? and let U = Vr^n+1. Thus we have a fibration |AV,d| —>· \AU,d\ with fibre |AV">n+1,dj. Since (Api,)r is concentrated in degrees < r it follows that for r < ? + 1 the only morphism (AV>T1+1,d) —> (Apl)j. is the augmentation (AV>n+i,d) —>· k. Thus |AV>n+1,rf| has a single zero cell, and no r-cells for 1 < r < ? + 1. It follows that ?4 (|AV>n+1,d|) = 0, 1 < r < n+ 1, and hence 7rn(|AV,d|) -^> ¦nn(\AU,d\). Moreover Un = V and ?? : nn(\AU,d\) -=»· Hom(t/n,A) is an iso- morphism by Step 3. Hence ?? : 7rn(|AV,d|) —> Hom(Vn, k) is an isomorphism. proof of (ii): It follows from (i) that |AF, d\ is simply connected and that each 7Tn|AV, d\ is a finite dimensional rational vector space, and that ?* : nr(\AV, d\) -^ Homij(T/r, Q) is an isomorphism. We again proceed in stages. Step 1: V = Vn. Here 7r,(|AV, 0|) is concentrated in degree ? and so |AV,d| is an Eilenberg- MacLane space of type ?(?,?) with ?? = HomQ(V,Q). In particular, the Hurewicz homomorphism is an isomorphism, hur : 7rn(|AV, 0|) -^ Hn(\AV, 0|;Z) by Theorem 4.19. It follows that Hn(\AV,0\;Z) is a rational vector space. Alto- gether then, since V" is finite dimensional, we obtain V = HomQ(HomQ(F, Q),Q) S Eomz(Hn(\AV..0\;Z),Q) = Hn(\AV,0\:Q) . Back-checking through the definitions identifies this isomorphism as the linear map Hn (m(Al/0)) : Vn —>· Hn (.4PL(|AF, 0|)). Now it follows from Example 2 of §15(b) that m(AVi0) : (AVn;0) —> -4pL(|AVrn,0|) is a quasi-isomorphism. Step 2: V = V^n.
Sullivan Models 253 Here we argue by induction on n, the case ? = 2 being covered by Step 1. Put U = V<n. The naturality of 77 : (AV,d) —> APL((AV,d)) means that (AV,d) (AU ®AVn,d) APL{{AU,d)) APL((AV,d)) commutes. Thus it is easy to arrange that (AU,d) (AU®AVn,d) APL(\AU,d\) APL(\AV,d\) APL((AV",O)) (AFn,0) APL(\AVn,0\) commutes as well. In this diagram m^i/,^) is a quasi-isomorphism (induction on n) as is rn(\vn,o) (Step 1). Thus Proposition 15.6 gives a quasi-isomorphism m : (AU ® AVn, D) -=* APL(\AV, d\) which extends m^\utd) and induces m(AK",o)· By the Lifting lemma 14.4 we can extend the identity of (AU, d) to a morphism ? : (AU <g> AVn,d) —> (AU ® AVn, D) such that ??? ~ m{AV4) rel (AU, d). Then (? - id) : Vn —>¦ At/, which implies that v? is an isomorphism and ??^?,?) is a quasi-isomorphism. Step 3; The general case. For the general case we fix any ? and show that Hn (m\v,d) is an isomor- phism. As observed in Step 4 of (i), \AV>n+l,d\ is (n+l)-connected. Thus it has a Sullivan model of the form (AW>n+l,d), while \AV^n+l,d\ has (AV^n+1,d) as its Sullivan model. By Proposition 15.6, |AV, d\ has a Sullivan model of the form (??/<"+?(8)??^>?+1,?>) and hence Hn (\AV^n+1,d\;Q) A Hn (\AV,d\;Q). This identifies Hn (??,(??,?)) with Hn (??^??<?+?^), which is an isomorphism by Step 2. ' D When It — Q and (AV, d) is a Sullivan algebra satisfying the hypotheses of Theorem 17.10 then the quasi-isomorphism ) ^ APL(\AV,d\) is called a canonical Sullivan model. Now let X be any simply connected CW complex with rational homology of finite type, and let mx : (AW, d) -"'
254 17 Spatial realization be a minimal Sullivan model. Since Apl(X) = Apl E«(JY)), the adjointness formula A7.8) produces a natural simplicial map -yx : S.(X)-+(AW,d) , adjoint to ?? ?. On the other hand in Proposition 17.3 we established a natural homotopy equivalence s ? : \S*(X)\ —> X- Let tx be the inverse homotopy equivalence (defined uniquely up to homotopy) and set hx = tx : X —» \AW,d\ . Theorem 17.12 With the notation and hypotheses above, (i) The diagram APL(AW,d) .A"-{iw APL\AW,d\ ApL{hx) . APL(X) (AW,d) is homotopy commutative. (ii) If Ik = Q then all the morphisms in the diagram are quasi-isomorphisms. In particular, hx is a rationalization of X (§9(b)). proof: (i) The left hand triangle commutes by construction. Next observe that S,(sx) °?s.(X) = id : S*(X) —>· 5,(Z). Since txsx ~ id]Sx\ we can apply Proposition 12.6 to conclude that ?-PL ° ? ?-PL ? ApL(tx) ? ? for any morphism ? : (AW.d) —>· APL (\S.(X)\). Thus ApL(hx) ? m{AW4) ~ ApL (is-(X)) ° apl {\lx\) the last equality following from the adjointness of •yx and ???. (ii) According to Theorem 17.10, m^w,d) is a quasi-isomorphism. Hence (h). In §17(a) we obser\'ed that ?*(??) is a quasi-isomorphism for so is
Sullivan Models 255 any simplicial set K. In particular Apl (?(?\?,?)) iS a quasi-isomorphism. Hence so is ?(?\?,?)· Finally since \AW, d\ is a simply connected rational space and since Ht(hx;Q) is an isomorphism (because Api(hx) is a quasi-isomorphism) it follows that hx is a rationalization (Theorem 9.6). ? (e) Morphisms and continuous maps. In §12(c) we observed that homotopic continuous maps /o ~ /i : X —> Y have homotopic Sullivan representatives. In the reverse direction we prove Proposition 17.13 Iftp0 ~ ?\ : (AV,d) —> (AW,d) are homotopic morphisms between Sullivan algebras then \??\~\??\:\???,?\-+\??,?\. proof: A homotopy from ?0 to ?? is a morphism ? : (AV, d) —> (AW. d) <g> A(t,dt) such that (id®Si)<b = ??, where ?,(?) = i. Since realization preserves products (the Example of §17(c)) we obtain a continuous map |?| : \AW, d\ ? |A(i,di)l —>· |AV,d| such \?\(?,?{) = \??\{?). But the identity map of A(t,dt) is a 1-cell in |?(?,dt)\ joining e0 to ??, and so \??\ ~ \??\. G Now suppose ? : (AV,d) —> (AW,d) is a morphism of Sullivan algebras. By adjointness — cf. A7.8) — APL({AW,d)) , where ? continues to denote the adjoints of id(Av,<f) and id(t\w,d)- It follows that (AW, d) Av,.d, A7.14) APL(\AV,d\) ^-— APL(\AW,d\) is homotopy commutative. Thus if ? is a morphism of simply connected rational minimal Sullivan algebras of finite type then A7.14) exhibits ? as a Sullivan representative of \<p\. Next, suppose / : X —> Y is a continuous map between simply connected CW complexes with rational homology of finite type. Let ?? ? : (AW, d) —> APi(X) and ??? : (AV,d) —> Apl(Y) be minimal rational Sullivan models, and recall the continuous maps hx and hy defined at the end of §17(d).
256 17 Spatial realization Theorem 17.15 With the hypothesis and notation above, let ? (AW,d) be a Sullivan representative for f. Then the diagram (AV, d) X S Y \AW,d\ \AV,d\ is homotopy commutative. proof: In the notation at the end of §17(d), ?.? and ty are homotopy inverses to sx and sy. Since fsx = sy|5,(/)| it follows that tyf ~ |S,(/)|?.v· But hx = \jx\ ? tx, and so it is sufficient to prove that \??\ ? |S*(/)| ~ \?\ ? \jx\. Because ? is a Sullivan representative of / there is a cochain algebra mor- phism ? : (AV,d) —> APL{X) ® A{t,dt) such that (id®eo)$ - ApL(f)my and (id <8>?i)i' = ????. Compose this with the obvious morphism Apl(X) ® A(t,dt) —> Apl{S*{X) ? ?[1]) and take the adjoint morphism ? : S*(X) ? ?[1] —>· (AV,d) via the adjoint relation G.8). Then \?\ : brS,(/)| ~ \???\- U (f) Integration, chain complexes and products. Suppose given a Sullivan algebra (AV, d) and recall the natural morphisms of cochain complexes (AV,d) A APL(AV,d) ? C*{AV,d) defined respectively in §17(c) and in §10(e). We denote by C(AV,d) the chain complex Hom(AV, h). Then the composite § ? ? dualizes to a natural morphism of chain complexes, via the obvious pair- (since C(AV,d) includes naturally in Hom(C*(AV,d),. ing)· Proposition 17.16 Suppose (AV, d) is a Sullivan algebra in which each V" is finite dimensional and either V = V-2 or else d preserves V. Then J^ is a quasi-isomorphism. proof: Theorem 10.15(ii) asserts that / is a quasi-isomorphism. On the other hand, so is any canonical Sullivan model, m^v.d)· (If V - ^-2 this is Theo- rem 17.10. Ifd: V —y V a trivial modification of the proof gives the same result.) Since ? = Apl{Q o"i.(Av,d) it follows that ? is a quasi-isomorphism too (§17(d)). Thus the composite dualizes to a quasi-isomorphism Horn (C*(AV, d),lk) -^>
Sullivan Models 257 Finally, if C is any chain complex there is a natural morphism C —> Hom(Hom(C, Jc),Jz). If C itself has finite type this is an isomorphism. If H{C) has finite type then this is a quasi-isomorphism because H (Hom(—, k)) — Horn (#(-),*). In the case of (AV,d) we have H*(AV,d) = H(AV,d) and so H,(AV,d) has finite type. Thus since C*{AV,d) = Hom(C.<AV,d),lfc) the in- clusion ? : C»(AV,d) —> Horn (C*(AV,d),]k) is a quasi-isomorphism too. Hence so is the composite Jt — Horn (^ ? 77,1c) ? ?. ? The quasi-isomorphisms §? : Apl(K) —> C*(K) are not a morphisms of cochain algebras. In compensation, we show that the quasi-isomorphism ft is 'compatible' with geometric products. For this we recall the definition (§17(b)) of the product ? ? L = {Kn ? Ln, di x di, Sj ? Sj} of simplicial sets. The simplicial diagonal ?# : ? —> ? ? ? is the morphism ? \—>¦ (?,?). Now in §17(a) we introduced the dgc, C,(K) and identified it as a sub dgc of C»{\K\) via the inclusion ?*(??)- In this setting, the Eilenberg-Zilber and Alexander-Whitney morphisms restrict to morphisms c„(uq ® c.(L) ^^> c»(ir ? L) -^ c.(ur) and the comultiplication in C»(K) is just AI-VOC^A^). Clearly AWoEZ = id and tedious but straightforward computations show, as in Proposition 4.10 that EZ ? AW ~ id. Now suppose (AV, d) and {AW, d) are Sullivan algebras. A natural inclusion C(.\v,d) ® C(AW,d) —)¦ ^(??,??)®(???,??) is given by (? ®?)(? <g> ?) = (—l)degi> ??8??(?N(?), ?€ C(Av.d) , o € C(AW,<f); moreover this inclusion is an isomorphism if either AV or AW has finite type. Recall also from the Example of §17(c) that we identify ((AV, d) ® {AW,d)) = (AV,d) ? {AW,d). Proposition 17.17 If {AV, d) and (AW,d) are Sullivan algebras then the dia- gram ? (AV, d) ® ? (ATV, d) ^— ? ((AV.. d) ® (AIV, d)) commuies. proof: First consider arbitrary simplicial sets ? and L and let ?? : ? ? L —y ? and pL : ? ? L —)¦ L be the projections. If ? G Apl(K) and ? G .4/»L(L) we abbreviate Apl(jpk)$ AApl^Pl)^ to ? ? ?. We begin by showing by induction onp + q that:
258 17 Spatial realization • For any ? € APPL(K), ? € A9PL(L), ? € Kk and ? € Le, A7.18) Indeed, if ? + q = 0 then ? and ? are \'ertices, ??(? <g> ?) = (?, ?) and the left hand side is just ?(?)?(?). This is (trivially) the same as the right hand side. Suppose ? + q = ? and that the proposition is pnwed for ? — 1. Recall from §10(a) that ? and ? determine simplicial maps ?» : ? [?] —>¦ /<" and ?» : ? [g] —)¦ L. By naturality we may use these to reduce to the case ? = ? [?], ? = A[q]. Thus .4PL(A') = (APL)P and APL(L) = (APL)g, as observed in Proposition 10.4(i). In particular there are polynomials f(t\,..., tp) and g(ti,... ,tq) such that ? = /dti A ¦ ¦ · ? dtp and ? = gdt\ A ¦ ¦ ¦ A dtq. Now either ? > 0 or q > 0. In either case ?? = ??? = 0. If ? > 0 clearly ? = dF and ? ? ? = ??(? ? ?), some ? G (^??,)^. Since §KxL commutes with the differentials, for any c C. Cn(K ? L) we have ) (de) . Lr ? ? Since EZ also commutes with differentials we may substitute ??(? <S> ?) for c and use induction to complete the proof. If ? = 0 and q > 0 write ? = du and use the same argument. ( A second proof may be constructed starting from the observation // f(t)g(s)dtds = ( / f(t)dt) ( / g(s)ds). This idea, while JJxxY \Jx / \Jy J more intuitive, is technically more complicated to implement.) Now specialize to the case ? = (AV,d) and L = (AW,d), and observe that ? € W ? ® *) = *7(AV,d)* X ^(AW'.d)*, ? "' Thus for ? e K,r€ L, ??(? ® ?) J (? ® ?) = ±( fKxL(vlAV,d)* ?® / ? J (?® ?) . Exercises 1. Let (AV, d) be a Sullivan minimal model with dim H*(AV, d) < oo and F1 = 0. Show that (AV,d) is the Sullivan minimal model of a 1-connected finite CW complex.
Sullivan Models 259 2. Prove that the geometric realization of the relative Sullivan model (Ax2 <8> Ax3;dx3 = x'l) is the rationalization of the principal fibration S3 —> S2 —> K(Z,2). 3. Let (?(?3,?5),0) <g> (A(v3,v'3,v5,V7,v'7,U9,V9,w9,...) ,d) be the Sullivan min- imal model of the product (S3 x S5) x (S3 V S3) . Compute the differential d for the given generators. Consider the relative Sullivan model (A(x3,y5) ® ?(?3,?3,?5,?;7,?;7,??9,?9, wg, ...),D) with ? = d up to degree 8, Dug = dug, Dvq = dvg and Dwg = dw9 + xsv5, ... . Prove that the geometric real- ization of this model is rationally equivalent to the total space of a fibration (S3 V S3) —> ? —> (S3 x 5°) with non trivial connecting morphism. If a and b are generators of 7r3E3 V S3) compute [[?, 4. Prove that if ? : (A,dA) —> (B,d?) is a quasi-isomomorphism then ?*(\?\) : H*(\A,d/i|) —> H*(\(B,cIb\) is an isomorphism. 5. Let ? : (At, 0) —>· A(x,y,z) be as in §3-exercise 3. Is \?\ an homotopy equivalence? 6. Let (AV, d) be a Sullivan minimal model and ? an element of degree ? . Prove that [(Ax,0),(AV,d)j = Hn(AV,d). Deduce that if X is a space then [X,K(Q,n)}=Hn(X;Q).
Part III Graded Differential Algebra (continued)
18 Spectral sequences The ground ring in this section is an arbitrary commutative ring ]k, except in topic (d) where h, is a field. Although we have managed to provide the reader with relatively simple 'spec- tral sequence-free' proofs in Parts I and II, subsequent material necessarily uses spectral sequences in a fundamental way. We therefore introduce them here, but limit ourselves to the strict minimum that will be required. In particular, we consider only the spectral sequence of a filtered object, and even there make no attempt to present results in their most general form. A spectral sequence is, in particular, a sequence of differential graded modules (El,d') such that H(El) = El+l. They can be used to 'approximate' the homol- ogy of a complicated differential module and in this (and other), contexts play an essential role in homological calculations. In this section we provide a brief and elementary introduction, including only material that will be required in the following sections. In particular, much of what is presented here holds in considerably greater generality. This section is divided into the following topics. (a) Bigraded modules and spectral sequences. (b) Filtered differential modules. (c) Convergence. (d) Tensor products and extra structure. (a) Bigraded modules and spectral sequences. Recall (§3) that a graded module M is a family {M{};6z of modules. Anal- ogously a bigraded module is a family M = {jV/i,j}({j)ezxz of modules indexed by pairs of integers (i,j). The basic constructions of §3 (e.g. Horn and <8>) carry over verbatim to the bigraded case. In particular a linear map / : M —> ? of bidegree (p, q) is a family of linear maps / : Mij —> Ni+pj+g and a differential is a linear map d : M —y M such that d2 = 0 and d has bidegree (—i, i - 1) for some i. The homology, H(M) = kerd/lmd, is again a bigraded module. Associated with a bigraded module M is the graded module Tot(M) defined by Tot(M)fc = ? Mitj and a differential in M induces an ordinary differential in Tot(M). Finally, as in the singly graded case, we raise and lower indices by setting MiJ = Af-f.-j . We can now define spectral sequences.
Graded Differential Algebra 261 Definition A homology spectral sequence starting at Es is a sequence (Er,dr,ar)r>s in which ET = {Ep q] is a bigraded module, dr is a differential in Er of bidegree (—r, r — 1) and ?? : H(Er) -A· ??+1 is an isomorphism of bigraded modules. A cohomology spectral sequence is a sequence (Er,dr,ar), r > s, with ET — {Ep>q}, dr a differential of bidegree (r, -r + 1) and ?? : H{Er) A ?r+1. Thus raising degrees converts a homology spectral sequence into a cohomology spectral sequence. We shall frequently suppress the ?? from the notation and refer to the spectral sequence (Er,dr) or (Er,dr). In calculating with (cohomology) spectral sequences it is frequently useful to represent them by diagrams of the form r, q - r + 1) A morphism of spectral sequences is a sequence of linear maps ?^ : E^ —> E^ of bidegree zero, commuting with the differentials, and such that ?? identifies ? (?^) with <^r+1\ (or ?^) : E(r) —> E(r) such that ?? identifies ? (?(?)) With ?(?+1))· (b) Filtered differential modules. A filtered (graded) module is a graded module, M, together with an increasing sequence d----CFpcFp+1C·-- ,p<=Z of submodules, called the filtration of M. (If M = {Mn} we write Fp = F_p and so the filtration has the form 5 : - · ? Fp D Fp+1 D ¦ ¦ · .) We may also write FPM instead of Fp if necessary to avoid confusion. The filtered module (M, ?) determines the associated bigraded module, CM, given by >P+4 = (Fp/Fp+l) V+Q
262 18 Spectral sequences Here p is called the filtration degree and q is the complementary degree. A linear map ? : M —> ? between filtered modules is filtration preserving if it sends each FP(M) to FP(N). In this case it induces Qtp : QM —> QN in the obvious way. A filtered differential module (M, d, 5) is a filtered module (M, 5) together with a filtration-preserving differential in M. Example 1 Semifree modules. Suppose (M, d) is a semifree (R, d)-module. Then (by definition, cf. §6) M admits a filtration 0 C M@) C M(l) C · · · such that (QM,Qd)^(R,d)®(V,0) . D Example 2 Relative Sullivan algebras. Let (B ® AV, d) be a relative Sullivan algebra (§14). It is filtered by the degree ofB: ? = 0,1,2,... and the associated bigraded module is given by (C(B ® AV), Gd) = (?, 0) ® (AV,d) . Example 3 The word-length filtration of a Sullivan algebra. Filter a Sullivan algebra (AV,d) by its word-length (cf. §12): ? > 0 . The associated bigraded module is then given by where do is the linear part of d : do : V —> V and d — do : V —> A-2V. D Suppose (M, d, $) is a filtered (graded) differential module. Then (QM, Cd) is a differential bigraded module in which Qd is a 'first approximation' to d. The spectral sequence associated with (M, d, ?) is a spectral sequence beginning with (GM, Qd) and, as we shall see in (c), 'converging' in many cases to H(M, d). We shall define the cohomology spectral sequence for M = {M*}i6z; the homology spectral sequence for M = {Mj}i6z is obtained simply by lowering indices. To begin, denote the filtration of M by · · · D Fp D Fp+I D ¦¦¦ . Note that each Fp is a graded module: Fp = {(Fp)n}neZ. For each integer r > 0 define graded modules Zp and Dp by Zp = {x€ Fp\dx?Fp+r} and Dp = Fp ? dFp~r .
Graded Differential Algebra 263 Evidently D% C D{'C ¦·' C Dv C ··- C Z*r C ¦¦ ¦ C Z* C ?? . Now define bigraded modules Er, r > 0, by letting Ef>q be the component of Z?/ (Zpr±{ + Dpr_^ in degree ? + q: It is immediate from the definition that d factors to define a differential dr in Er of bidegree (r, 1 —r). Moreover, the inclusions Z^+1 c—> Zp induce isomorphisms of bigraded modules Er+i A H(Er, dr) . Note as well that Definition The spectral sequence (Er, dr) is called the spectral sequence of the filtered differential module (A/, d, J). Note that this construction is natural: if ? is a morphism of filtered differential modules it induces a unique morphism ??(?) of spectral sequences such that ??(?) = ??. Example 4 Bigradations. There is one particular situation that arises frequently in practice. We suppose given a graded differential module (M, D) in which • Each Mr is given as a direct sum Mr = ? Mp'q. p+q=r • The differential d is a direct sum D = ?, A with Dt : ??? —? Mp+i'q~i+1. i>0 Here a canonical filtration is given by Fk(M)r = 0 Mp'r~p and this leads as p>k just described to a spectral sequence. Moreover from D2 = 0 we deduce D$ = 0, DXDQ + DODX = 0 and D'{ = -DqDz + D2D0. In particular, Do is itself a differential and Dx induces a differential ?(??) in H(M,D0). Now a straightforward verification shows that the first two terms of the spectral sequence are given by (Eo,do) = (M,D0) and (Eudx) = (H(M, A>), H(DX)) . ? (c) Convergence. Suppose (Er,dr) is a cohomology spectral sequence. Fix a pair (p, q). If the maps
264 18 Spectral sequences are both zero then H™ (Er, dr) = E™ and ?? : ?™ A ???. In this case we write Epg = ?y+i· ^he spectral sequence (Er,dr) is called convergent if for each (p, q) there is an integer r(p, q) such that In this case the bigraded module E™ is defined by E™ = E™, r > r(p, q). The spectral sequence collapses at Er if di = 0, i > r. In this case E^ is defined and Er = Eco. An important example of convergence arises in the case of first quadrant spec- tral sequences, which we now define. A first quadrant cohomology spectral sequence is one in which each ET = {E%'q}p>0 q>0. These spectral sequences are convergent with E™ = E™, r > max(p, q + ?). Analogously, a first quadrant (cohomology) filtration in a graded module M is a filtration {FPM} such that F°M = M and FPM = {(FpM)"}n>p . Notice that these conditions imply that M = {M"}„>0. Let (M, d, ?) be a filtered differential (graded) module. Define a filtration in H{M) by FP(H(M)) = Im (H(FpM) —»· H{M)) . If the associated spectral sequence is convergent and if there is a natural isomor- phism Eoo = QH(M) then we say the spectral sequence is convergent to H(M). Proposition 18.1 Suppose (M, d, #) is a cochain complex with a first quadrant filtration. Then the associated spectral sequence is first quadrant,and converges to H(M). proof: The first assertion is immediate since Eq = QM is necessarily concen- trated in non-negative bidegrees. For the second assertion let Zpq be the component of degree ? + q in the graded module Zp. If r > q + 1 then d : Z™ —»· (Fp+rM)p+q+1 = 0 . Thus Zp'q = kerdr\(FpM)p+" and, similarly, ??±\>"-1 = kerdn (FP+1M)P+Q . Moreover, if r > ? then Dpr_x = d{M) ? FPM. This identifies p'" = zp>y (zprt\>q-1 + (^_!)P+9) = gp>*H(M). a
Graded Differential Algebra 265 Proposition 18.2 (Comparison) Let ? : (M,d,$) —>· (N,d,$) be a mor- phism of filtered cochain complexes with first quadrant filtrations. If some ??(?) is a quasi-isomorphism then ? {?) is an isomorphism of filtered modules. In particular, ? is a quasi-isomorphism. Lemma 18.3 Suppose ? : (M, 5) —> (-W>5) is a morphism of filtered graded modules with first quadrant (cohomology) filtrations. IfC(ip) is injective (resp. sur- jective) then so is ?. proof: Suppose Q(ip) is injective. If 0 ? x 6 M" there is a greatest ? such that ? € FPM, because (Fn+1M)n = 0. Thus ? represents a non-zero element [x] € Qp'n~p(M) and so 0 ? Q(rp)[x] = [??]. In particular ibx ? 0. Suppose G(ip) is surjective. Then for each p, ip(FpM) + FP+1N = FPN. It follows by induction on q that ? (FPM) + Fp+qN = FPN, q > 0. Fix ? and choose q so p + q > n. Then (Fp+qN)n = 0. It follows that ? (FPM) = FPN. ? proof of 18.2: Since E{+\ (?) is identified with ? (?{(?)) it follows that Er+i(<p) is an isomorphism. By induction ???(?) is an isomorphism for m > r+ 1. Hence ???{?) is an isomorphism. By Proposition 18.1 this is identified with CH(<p). Now apply the Lemma above (and its proof) with ? = ?(?). ? (d) Tensor products and extra structure. In this topic we restrict to the case Jk is a field. Suppose (M,d,3) and (?,?,'?) are filtered differential (graded) modules. The tensor product is the differential graded module (M,d) <g> (N, d) equipped with the filtration FP(M <g> N) = + F{M <g> FjN i+j=p (note: the sum on the right is NOT direct). It is straightforward to identify that if QM or QN is ic-projective then G{M®N) =GM®GN as bigraded modules, compatible with the differentials. Now we observe there are natural identifications {Er{M®N),dr) s (Er(M),dr)®(Er(N),dr), r > 0 , compatible with the isomorphisms Er+i = H(Er,dr). In fact, when r = 0 this is just the remark above. For r > 1 it follows by an inductive application of the fact that homology commutes with tensor products (Proposition 3.3).
266 18 Spectral sequences Example 1 Filtered differential graded algebras. Suppose (A,d) is a dga equipped with a filtration {Fp} such that Fp is pre- served by d and satisfies Fp ¦ Fq C Fp+q. Then multiplication (A,d)®(A,d) —» (A,d) is filtration preserving and hence determines a morphism of spectral sequences (Er,dr)®(Er,dr) —>(Er,dr) . This identifies each (Er,dr) as a differential graded algebra with H(Er) — Er+\ as graded algebras: we say (Er,dr) is a spectral sequence of graded algebras. If the spectral sequence is convergent then E^ is a graded algebra. In the case of a first quadrant filtration the isomorphism E^ — GH{A) is an isomorphism of algebras. However the graded algebra TotQH(A) may not be isomorphic with ? (A) ! Exercises 1. Let (AV,d) be a Sullivan model. Consider the filtration FP(AV) = A-PV of example 3 and the associated first quadrant spectral sequence of algebras: Ep'q = (ApH(V,d0))p+g =» Hp+o{AV,d) . Prove that if (aV,d) is a 1-connected minimal model then E2 = H(AV,d\). When (AV,d) is the minimal model of CPn, ? > 3, compute the differentials and the product in E^· 2. Let (B,d) -» (B ® AV,d) -» (aV,2) be a relative Sullivan model and consider the spectral sequence of algebras defined in example 2. Prove that E\'q = (Bp <S> Hq(AV,d),di), and determine di using the short exact sequence 0 -»· Fp+1/Fp+2 -»· Fp/Fp+2 -»· Bp ® AV ->· 0. Prove that if Bl = 0. then 3. Let (AU ® AV, d) be a relative Sullivan model such that dim V < oo and such that the algebra H(AU,d) is finitely generated. Denote by at- for i = 1.2..... ? (resp. ??) a linear basis of the space (H+(AU)/H+(aU) ¦ H+(AU))'V" (resp. of the space V). Prove that if dvi = cq for some cocycle ai in AU representing a·;, then dim H(AU <g> ??·7, d) < oo. 4. Let F -> X -> F be a fibration as in theorem 15.3. Construct a spectral sequence (the rational Serre spectral sequence): P a) Prove that H(j) : H*(X;Q) -> H*{F;Q) is onto if and only if the ra- tional Serre spectral sequence collapses at the J52-term and thus H*(X;Q) — H*(Y;Q) ®ii*(F;Q) as graded vector spaces. b) Prove that if H*(F; Ik) = Ax with ? of odd degree then H,{X,Ik) is isomorphic to ??(?, Ik)/(dx) ? sd's XA where A denotes the annihilator of dx in
Graded Differential Algebra 267 5. Show that the filtration of a CW complex X by its skeleton induces a first quadrant spectral sequence which collapses at the E-i-term. 6. Suppose that ??2? = 0 unless q = 0 or q = n. Show that E^n — Ker dn+\. Deduce the short exact sequence 0 _> E&n -» E*>n dnA' ??+?+1'° -> ??+?+?'° -> 0. 7. Let F -4 I -> y be a fibration with 1-connected base Y¦ Suppose further that H'(Y;Q) = 0 for 1 < i < ? and Hj(F;Q) = 0 for 0 < j < q. Prove that there is a long exact sequence 0 -» H1 (X; Q) H-J Hl (F: Q) -*· H2 (Y; Q) H^4 H2 (X; Q) -»· ¦ · · j. 8. Consider the Serre fibration G/iiT -> 5iiT -4 BG as defined in §15 f, example 1. Deduce from exercise 4b that H*(BK;Q) is a finitely generated H*(BG;Q)- module via H'(Bj;<Qf).
19 The bar and cobar constructions In this section the ground ring is an arbitrary commutative ring k. As usual - <g> — means — <8>/t —. The bar and cobar constructions are functors augmented ? co-augmented differential graded algebras differential graded coalgebras and co-augmented ? augmented differential graded coalgebras differential graded algebras. They were introduced, respectively, by Eilenberg-MacLane [48] who showed that B(Z[T]) ~ C, (K(T, 1)) and by Adams [1] who showed that UC»{X) ~ <7,(?,?). Here we give the constructions, and establish those few properties that will be essential in the sequel. We begin with some conventions (cf. §3). The suspension sV of a graded module is defined by (sV)k = Vk-i and the correspondence is denoted by sv <-* v, ? € V. Similarly {s'1V)k = Vk+i and s~xv <-> v. Recall that the tensor algebra on a graded module V is the graded algebra TV = 0 TkV with T°V = Me and TkV = V ® - - - ® V (k times). The multipli- cation TkV <g> TeV —> Tk+iV is just the tensor product, and TV is augmented by ? : V -> 0. The tensor coalgebra on V is defined analogously. It coincides with TV as a graded module, so to avoid confusion we denote its elements by [v\\ ¦ ¦ ¦ \vk] instead of by v\ <8> · · · <8> Vk ¦ The diagonal in the tensor coalgebra is defined by • M = [?>?| · · · \vk] ® 1 + 2_j[vi\ ¦ ¦ ¦ \vi] ® [vi+i\- ¦ ¦ \vk] + 1 ® [wi| · · · \vk]. The tensor coalgebra is augmented by ? and co-augmented by it = T°V. Note that the diagonal is not compatible with the tensor algebra structure. Now suppose ? : {A, dA) —> Ik is an augmented dga and denote kere by /. Definition The bar construction on (A, dA) is the co-augmented differential graded coalgebra ? A defined as follows: • As a co-augmented graded coalgebra ? A is the tensor coalgebra T{sl) on si. • The differential in ? A is the sum d = d0 + di of the coderivations given by k do([sai\---\sak]) = -^(-l)ni [scn\ ¦ ¦ ¦ \sdAa.i\ · · · \sak] i=l
Graded Differential Algebra 269 and ? dl([sa])=O ) k di([sai\---\sak]) = ? (-l)ni[«ai| · · · |sui-iui| · · · \sak]. { i=2 Here nt- = ? deg saj. j<i Notice that do reflects the differential d^ in A while d\ is defined using the multiplication. To see that d2 = 0 note that dg = 0 because d\ = 0, dod\ +dido = 0 because dA is a derivation in A and dj = 0 because A is associative. We shall adopt the notation ? ? = ? BkA with BkA = Tk(sl). The notation k lBA' is an abuse; we should really write B(A,d) or ?(?,?,?), and we will in fact write B(A, d) when there are several possible differentials in A. More generally, if (M, d) is a left (A, d^-module (cf. §3) then the bar construc- tion on {A,dA) with coefficients in (M,d) is the complex B{A;M) = ? A <g> M with differential d = do + d\ where k do[sai\ ¦ ¦ ¦ \sak]m = - ? (-!)"''[s°iI · · · \sdAai\ ¦ ¦ ¦ \sak]m -(-l)^+i [sai\...\sak}dm and k d\[sa-i| · · ¦ \sak]m = ? (-l)ni[sai| · · · ????-??,? ¦ ¦ ¦ \sak]m i=2 + (-l)n*+> [sai| · · · \sak-i]ak · m. Of course dorn = dm, dim = 0 and di[sa]m = ( —l)dega+1a · m. We also adopt the convention: Bk{A;M) = BkA ® ?, and elements in BkA or Bk(A;M) have wordlength k. Note as well that if we consider ic as an (^4, d^-module via ? then EA, d) = (B(A;k),d). Henceforth we shall denote our differential graded algebras by (^4, d) since it will always be clear from the context whether "d" refers to the differential in .4 or in B(A;M). There are two situations in which it is easy to compute ? (?(?; ?)). The first arises when .4 is a tensor algebra. Thus suppose A = TV and either V = {K-}i>o or {l/i}i>2- Let dy : V —> V be the linear part of the differential in A, so that d — dv '¦ V —». T^2V, and define d:sV-^ sV by dsv = —sdyv.
270 19 The bar and cobar constructions Now here / = T-^V. Define a morphism of complexes ?: (B(TV,d),d) —> (sV®Jk,d) by the conditions: ?? = 1, g[sv] = sv, g[s(vi ® · · · ® vi)} = 0, t > 2 Proposition 19.1 With the notation above, ? is a quasi-isomorphism. proof: Since ? is surjective we need only show ii(ker ?) = 0. Define h : B(TV) —> ker? by h{\) = 0 and J [sv\···] —»?, ? G F ^: { v2 ® ¦ ¦ ¦ ® ve)\ ¦ ¦ ¦}, e>2. A quick calculation shows that hdi +d\h = id in ker ?. Since h raises wordlength and do preserves it, it follows that (id —hd — dh) increases wordlength in ker ?. Similarly, id -hd- dh preserves degrees. Let ? 6 ker ? be a cycle of degree n. Then (id —hd — dh)nz has degree ? and wordlength at least ? + 1. However, our hypothesis on V implies that elements in B^n+l(TV) have degree at least ? + 1. Thus (id -hd - dh)nz = 0. Since id -hd - dh)kz = 0 this gives (id -dh)nz = 0; i.e. is a boundary. D The second situation where we can compute is when we take (A, d) as a left module over itself, via multiplication. This yields the complex ? {A; A) =BA®A. Note that B(A;A) is a right (A,d)-module via multiplication on the right and that B(A;M) = ? (?; ?) ®? ?. Proposition 19.2 (i) The augmentations in ? A and A define a quasi-isomorphism, ? <8> ? : ? (A; A) -% Jfe. (ii) If k is a field then ? (A; A) is a semifree right (A, d)-module. Thus ?® ? is an (A, d)-semifree resolution of k.
Graded Differential Algebra 271 proof: (i) Define h : ? (A; A) —> ker(e ? ?) by [sail ¦ ¦¦\sa.k\a ?—> (-l)e[sa.i\ ¦ ¦ ¦ \sa.k\s(a - ??)]?, k > 1 where e = deg[sax| ¦ · · |sat]a. An easy computation gives id = dh + hd in ker(c ® ?), so that H(ker(e ? ?)) = 0. f«j Since ? (A; A) is the increasing union of the submodules B-kA<g>A, it is sufficient to show that (Lemma 6.3) the quotient modules (B-kA/B<kA) <g>.4 are (A, d)-semifree. These quotients may be identified as (Tksl,do) <8> (-4, d). Thus the inclusions kerdo® (A,d) c—>(TksI,do)®{A, d) exhibit these quotients as semifree. ? The cobar construction is a precise dual analogue of the bar construction. Suppose (C, d) is a co-augmented differential graded coalgebra with comultipli- cation ? : C —> C ® C (cf. §3(d)). Let C be the kernel of the augmentation e:C —»· Jk. Then (cf. §3(d)) Ac - (c ® 1 + 1 ® c) G C ® C , ceC. Define ? : C —^ C <8 C by Ac = Ac - c ® 1 - 1 ® c. This map is called the reduced co-multiplication and is also coassociative: (A®id)A — (id ®?)? : C —> C ® C ® "C. Definition The cobar construction on (C, d) is the augmented differential graded algebra QC defined as follows: • As an augmented graded algebra UC is the tensor algebra T(s~1C) on The differential in UC is the sum d = do + d\ of the derivations determined by do(s~1x) - -s'^x, ? e C and dxis-'x) = ^(-l^s^Xi ® s-'yi, xeC, i where Ax — Y^Xi®yi. Exercises 1. Let / : (A,d,e) —> (A',d',e') be a morphism of augmented chain algebras, then the formula Bf([sai\sa{...\sak}) = [sfal\sfa2\...\sfak], k > 0, defines a morphism of graded differential coalgebras Bf : ? ? -> ? A'. Prove that Bf is a quasi-isomorphism if and only if / is a quasi-isomorphism.
272 19 The bar and cobar constructions 2. Let (A,d,e) be a differential graded augmented algebra. Prove that the nat- ural morphism of differential graded algebras QBA —>· .4 is a quasi-isomorphism. 3. Let (A, d, e) be a 1-connected augmented cochain algebra and denote by B*A the graded dual of the bar construction ? A. Prove that B*A is a chain algebra. Prove that if A is of finite type then QA* 9* B*A. Let (A,d) = (AV,d) be a minimal model and di : V —>· V <8> V be the quadratic part of the differential d. Prove that V = sH+(B*A) and that di is dual to the multiplication in H+B*A. 4. Let (A,d,e) and B*A be as in exercise 3 and denote by (/, a) the pairing B*A <8> ? A -> Ik. Prove that the action defined by / · [sai|sa2|...|safc]a = , [sai|sa2|...|sat·]) [sai+1|sa2|-.-|safc], k > 0, makes B(A,A) into a right B*A-module and that this structure is compatible with the right ^4-module structure. Show that the natural map B*A -> EndA(B(A, A)) is a quasi- isomorphism.
20 Projective resolutions of graded modules In this section the ground ring is an arbitrary commutative ring, Jk- Our objective is to set up the classical homological functors Ext and Tor for modules over a graded algebra; of course this parallels (and includes) the standard construction in the non-graded context. Our presentation is terse, and the reader seeking a more detailed explanation is referred to [118]. The reader may also wish to review the definitions in §3(a) and §3(b). Let A be a graded algebra (over ic). A chain complex ({A/*,,}, d) of A-modules is a sequence r\ d ? r d f. r d d ¦% r d 0 <— A/o,* <— Mi,* <—¦¦¦<— A/*,, <— ¦ ¦ ¦ in which: • Each A//.,» is an .4-module, with ?^,? the component of A/*,» of degree k + i. • Each d is an ?-linear map of degree —1; i.e.. d : Mkj —> ?/^-?,?. • dr = 0. Notice that (cf. §18(a)) (Tot(A/),d) is an (.4,0)-module as defined in §3(c); the difference is that here we keep track of the bigrading. A chain map of bidegree (p, q) between two such chain complexes is a family of ^-linear maps ? : M*,. —> Nk+P,m , k > 0 , of degree ? + q, and satisfying dtp = ( — l)p+qtpd. Note that the homology, H{M*^,d) of a chain complex of ,4-modules is itself a family {//*,„} of graded ?-modules and that the chain map ? induces -4-linear maps Hk,» (?) : Hk,, (A/) —> Hk+P,. (N) of degree p + q. This section is divided into the following topics: (a) Projective resolutions. (b) Graded Ext and Tor. (c) Projective dimension. (d) Semifree resolutions. (a) Projective resolutions. Let A be a graded algebra. An /1-module ? is projective if ? ? Q is .4-free for some second ,4-module Q (cf. §3(b)). In this case any /1-linear map (of any degree) ? : ? —> M lifts through any surjective .4-linear map ? : ? —> M. Indeed, extend ? to ? ? Q by setting ? = 0 in Q, let {vQ} be a basis of ? © Q and define ? : ? ? Q —> ? by avQ = nQ where nQ G ? satisfies ??? = ava. Denote A>0 by A+ (cf. §3).
274 20 Projective resolutions of graded modules Remark 1 If A ~ k ® A+ then any A-projective ? concentrated in degrees > m, some m G Z, is ?-free. In fact write ? ? Q = V <g> A for some ?-free module V. Apply — ®.4 k to obtain (P ®a k) © (Q <8U &) = ^- Thus ? = ? <8u Je is Jc-projective and splits back into P. This inclusion extends uniquely to an ^4-linear map ? : ? ® .4 —>· P. But clearly ? = ?? ? ? · .4+ = ?? · ? + ? · ?+. Iterating gives ? = ? ? · ? + ? -(?+ ?+ ). Choosing ? — m + 1 factors A+ we see that Pn = {?? · A)n\ i-e., ? is surjective. Choose an ^.-linear map ? : ? —> ? ® ? so ?? = id. Then ??{?) © (? ® A+) = ? ® A and it follows as with ? that r is surjective. Hence r is an isomorphism, and ? is /1-free. A projective (free) resolution of an /1-module, M is a chain complex (P, d) = ({Pk,*},d) of projective (free) /i-modules together with a morphism of degree zero, ? : .Po,* —> M such that is exact. Remark 2 Think of M = {MOj,} as a trivial chain complex of .4-modules. Then a projective (free) resolution is just a quasi-isomorphism ? : {P,d) —> (M,0) of bidegree @,0). Remark 3 Suppose ? : (P,d) —> (M,0) is a free resolution. Regard ? as a quasi-isomorphism (Tot(P),d) -=> (M, 0). Filtering Tot(P) by the submodules / k \ Tot I 0 jPj * I then exhibits this as an (.4,0)-semifree resolution of (M, 0). ' / Analogous to Proposition 6.6(i) we have Lemma 20.1 Every A-module M has a free resolution. proof: Choose an .4-linear surjection of degree zero, ? : Po.« —> M, from a free .4-module POi«. If <— Po,* <— ¦¦¦ <— Pfc,« is constructed choose an .4-linear surjection d of bidegree (-1,0) from a free .4-module Pfc+i,, onto kerd C Pfc,*- D Next, suppose given the diagram (N,d) ? (P,d) . (M,d)
Graded Differential Algebra 275 in which the objects are all chain complexes of -4-modules and ? and ? are morphisms respectively of bidegrees (p.q) and (??,??). Analogous to Proposition 6.4(ii) we have Lemma 20.2 Suppose the A-modules Pfc,« are projective and ? is a surjective quasi-isomorphism. Put r = (p + q) — (m + ?). Then (i) There is a morphism ? : (P,d) —> (N,d) of bidegree (p — m,q — n), and such that ?? = a. (ii) If ? is a second such morphism then ? — ? = dj + ( — l)rjd, for some ?-linear map 7 : ? —> ker ? of bidegree (p — m + 1, q — 71). proof: (i) Suppose ? constructed in P;it, i < k. Since Pfc,« is projective there is an .4-linear map ?1 : Pki. —> ? such that ??' = a. Clearly d?' - {-l)r?d sends Pk,, into the cycles of ker ?. Since, by hypothesis, ? (ker ?) = 0, it follows that d : ker ? —> cycles(ker?) is surjectrve. Choose ?" : Pfc-, —> ker ? so that d?" = d?' - (-l)r?d and set ? = ?' - ?" in Pfc,». (ii) Note that ? - ? : (P, d) —> (ker?,d). If 7 is constructed in Pi>; i < k then ?-?-(-\??? sends Pfc,, into the cycles of ker ?. Again, d : ker ? —> cycles(ker?) is surjective, which permits us to construct 7 : P*^, —> ker ? so that d-y = ? - ? - (-ly-yd. ' D (b) Graded Ext and Tor. Fix a graded algebra, ^4, and -4-modules M and ? (both left, or both right), and choose a projective resolution ? : (P = {Pfc»},(i) -=> (M,0). Then HomJ4(Pfci«,iV) is the graded module of .4-linear maps (cf. §3(b)), and 0—>EomA(P0i,,N) ?????(??,?') A--- B0-3) is the cochain complex of graded modules defined by O(f) = —( — l)de&ffd. We use the standard convention to raise degrees, and denote the graded module HornA(Pfc,.,N) by Horn*1*(P,iV): Eom^(P,N) = Hom,(Pt,t,iV)_(Hi) . Thus the cochain complex B0.3) has the form (Hom.4 (?, ?),?) with <5 of bidegree A,0) and its homology is a bigraded A--module, ?*'* (Hom.4(P, ?),?). - If ?'.: (P',d) -=> (?'/,?) is a second projective resolution then Lemma 20.2(i) provides us with a morphism ? : (P,d) —> (P',d) such that ?'? = ?, while it follows from Lemma 20.2(ii) that ? (HomA(?,N)) is independent of the choice of ?. Reversing the roles of (P, d) and (P', d) yields a morphism ?' : (P', d) —> (P,d) and Lemma 20.2(ii) implies that ? ?' ~^ idP> while ?'? ~? idP. Thus ? (Hom.4(/3,iV)) and ? (HornaW, ?)) are canonical inverse isomorphisms, and we use them to identify H'>' (EomA(P,N)) with H*·* (RornA(P',N)). Thus we may make the
276 20 Projective resolutions of graded modules Definition For each k > 0, Ext A(M,N) is the graded module Hk'' (RomA{P,N),o). (Where necessary we write Ext^(A/,iV) = Hk<e; recall that this is the component of degree k + i.) Notice that this construction is functorial. Indeed, if ? : M' —> M and ? : ? —> ?' are morphisms of ,4-modules then (Lemma 20.2) ? lifts to a morphism ? : (P',d) —> (P,d) between projective resolutions for M' and M and HomA(u,/3) is a morphism of cochain complexes inducing the canonical linear maps ExtA{a,?) : ExtkA(M,N) —>ExtkA(M',N') . In the same way, if L is any ,4-module (left if M is right, right if M is left) then 0 <— Po,» ®? L i Pi,« 0.4 L i · · · B0.4) is a chain complex of graded modules (? ®a L = {Pk,, ®a L},d) whose homol- ogy is a bigraded module independent of the choice of resolution. Definition For each k > 0, Tor?(M, L) is the graded module Hk,* (P ®a L, d). (Where necessary we write TorJ??(M, L) = ?^,?\ this is the component of degree Notice that this construction too is functorial. If ? : M' —> M and 7 : L' then lift ? to a morphism, a, of projective resolutions and define We turn now to the elementary properties of these functors. Observe first that if ? : (?, d) —> (M, 0) is a projective resolution then the surjection ? : Po,* —> M induces natural isomorphisms Ext?t (M, N) = YiomA (M, N) and Toro (M, L) = M ®A L . B0.5) Next, suppose 0 —> M' A M ? M" —> 0 is a short exact sequence of j4~linear maps. (We do not require that ? and ? have degree zero.) Let ?' : (P',d) -% (M',0) be a projective resolution. It is trivial to modify the construction in the proof of Lemma 20.1 to obtain a projective resolution ? : (?, d) -=> (M, 0) of the following form: each Pkt, = P'k ,©P^, each Pi,» is projective, and the inclusion ? : P' C ? commutes with the differentials and satisfies ?\ = ?'.
Graded Differential Algebra 277 In this case the quotient (?/?', d) is the chain complex of projective modules ?",. The induced row exact diagram ? 0 (P\d) (M',0) (P,d) (A/, 0) {P/P',d) —- 0 (A/", 0) 0 exhibits ?" as a projective resolution of M" (pass to the long exact homology sequences and apply the Five lemma 3.1). Note that (forgetting differentials) ? = ?' ? ?/?' as families of .4-modules. Hence if ? is any yl-module, 0 <— EomA(P',N) <— EomA(P,N) <— HomA(P/P',N) <— 0 is a short exact sequence of cochain complexes of graded modules. Passing to homology gives the long exact sequence Ext*+1| ExtkA(M',N) <— ExtkA(M,N) <— ExtA(M",N) Similarly we obtain the long exact sequence B0.6) B0.7) Finally, if Q is a projective .A-module then Hom>i(Q, —) and Q ®a — preserve short exact sequences (this is immediate from the definition). From this it follows that Honru(P, —) and ? ®a — convert short exact sequences 0 —> N' —> ? —> ?" —> 0 and 0 —> V —> L —> L" —> 0 into short exact sequences of complexes, whence the long exact sequences —> ExtkA{M, N') -^ Ext^(A/, N) —> ????4(?/, ?") ? ????4+x(A/, N') and B0.8) Torj?(M, L) f,L") d. ¦k-i(M,L') . B0.9) Example 1 The bar construction. Suppose k is a field and .4 -^> k is an augmented algebra, with no differential. Then the bar construction with coefficients in A (§19) has the form
278 20 Projective resolutions of graded modules and Proposition 19.2 exhibits this as an .4-free resolution of Ik. But for any left A-module ? we have B(A; N) = ? (?; ?) ?? ?. It follows that TorA(lk,N) =H(B(A;N)) . (c) Projective dimension. As usual, fix a graded algebra, A. The projective dimension of an /1-module M, denoted proj dim ? (?), is the least k (or oo) such that M admits a projective resolution of the form M <— Po,. <— Pi,. < <— ?*,. <— 0 . Proposition 20.10 For any A-module, M, proj dim a (M) = sup I k | Ext* (M, —) is non-zero \ . proof: Denote the right hand side of the equation by p(M). If M has a projec- tive resolution of the form above then we can use it to compute Ext4(M, —) and it follows at once that Ext^(M, — ) = 0, ? > k. Hence p(M) < proj dim^M). Conversely, suppose p(M) is finite and let ? : ({Pk,*},d) —> (M,0) be any projective resolution. For simplicity write p = p(M). Let ? C Pp,» be the submodule of cycles (Z = kerd). Then the differential, d, maps ??+?,« onto Z; denote this linear map by / : ??+?,« —> ?. In particular / G ?????+1·*(?, Z) and 5f = fod = d2 = 0. Since ????+1(?, -) 0 by hypothesis, / must satisfy / = god for some g : PPi« —> Z. In other words, for ? 6 PP-fi,«, dx = f(x) = g(dx). Thus g restricts to the identity in ? and PPi« = ? ® kerg. In particular kerp is projective and M <— Po,, 4 <H- Pp_!,» A- kerg <— 0 is a projective resolution, whence proj dim^M) < p(M). ? (d) Semifree resolutions. Now let (A, dA)besi differential graded algebra and let m : (P, d) -% (?/, d) be a semifree resolution of an arbitrary right (A, rf/i)-module, (M,d). As observed in the Remark at the start of §6(a) (with completely different notation!) we may use a semifree filtration of (P, d) to write oo P = V®A, V = ffi V(k) ,
Graded Differential Algebra 279 where each V(k) is a free graded (Jk—) module and the semifree filtration is given by P(k) = 0 V(i) ® -4. In particular d : V(k) —> P(k - 1). i=0 Now m restricts to m@) : V@) ® (A,dA) —> (M,d) . Moreover the filtration gives rise to a homology spectral sequence (§18) which we denote by (?\d!) .>Q. Clearly E°k = V(k) ® (.4,d^) and so El = H (E°,d°) = V(k) ® H{A) because V(k) is .fc-free. Thus we obtain the sequence H{M) <H{rn{0)) V@) ® H{A) ?- V{\) ® H (A) < in which all but the first term form a chain complex of H(A)-modu\es. Definition If this sequence is exact, and hence a free resolution of the H(A)- module H{M) then m : (P,d) -=» (M,d) is called an Eilenberg-Moore resolution of (Af,d). Proposition 20.11 Each (M,d) has an Eilenberg-Moore resolution. More precisely, any H(A)-free resolution of H(M) appears as the E1-term of some semifree resolution of (M,d). proof: Fix a free resolution H(M) A V@) ® H (A) A V(l) ® H (A) 4 . oo Set V = 0 V(k) and ? = V <g> A. We define the differential, d and the map m k=0 in each V(k) by induction on k. Set d = 0 in V@) and, if {va} is a basis of V@), define m@) : V@) —> M by requiring m@)va to be a cycle representing gva. Then H(m@)) = ? and so H(m@)) is surjective. Now d = id®dA in K@) ® A. Let {f^} be a basis of V{\) and let ?? be a eycle in V@) ® A such that dvp = [z0]. Then [mz?] = gdv? = 0 G H{M); i.e., ???? = da:^, some a:^ G ?·/. Extend m and d to A®V{\) by setting dv? = ? ? and ??? ? = ??. ? Write ?(?) = 0 V(i) ® ?, ? > 0, and suppose inductively that m and d are ;=o extended to some P{k), k > 1. Write d = ?] d, with d0 = fd®d^ and d, an i=0 /1-linear map sending each V(i) to F(?—i)(g).4, i > 1. Thus dxdo + dodx = 0 and our inductive assumption is that H{d\) = d : V(i) ®H(A) —> V(? — 1) ® H(A). To extend m and d to V(k+ 1) we consider a do-cycle ? G V(k) ®A such that d[z] = 0, [2] denoting the class in V(k) ® H(A), and we show that • There exist w G P(k — 1) and ? G ?/ sue/i that d(z — w) = 0 . , and m(z — w) = dx. *· The construction proceeds via the following steps.
280 20 Projective resolutions of graded modules Step 1: Since d[z] = 0, dYz = dQz' (z' G V(k-1)®A) and d(z-z') G P{k-2). Step 2: Suppose for some y' G P(k-l) that d{z-y') = J2ui with Uj € V(i)®A ? and 1 < ? < k — 2. Then Edu, = 0, whence doue = 0 and d\Ut = —doue-i· This gives d[u(] = 0. By exactness [u(] = d[u'} for some do-cycle u', and so ut = d1ul + dou", with u' ?V{?+l)®A and u" G V{?) ® A. Altogether we have d(z -y' -u' - u") G P{i - 1). Iterating this procedure we find y G P(k - 1) so that d(z - y) G V@) ® A . Step 3: The class [d(z - y)] G V@) ® H(A) may be non zero, but ?[?{? - y)] is represented by the boundary dm(z - y) and so g[d{z - y)) = 0. By exactness [d(z-y)] = d[yx] for some <20-cycle yi e V(l) ® A Thus for some y0 G V@) ® .4 we have d(z - y) = dyi + dy0 . Step 4·' The cycle ? — y — y\ — yo is mapped by m to a cycle, which is necessarily of the form m(z0) + dx for some cycle z0 e V@) ® A. Set w = y + yx + y0 + z0. This completes the proof of B0.12). Now let {^a} be a basis of V(k + 1), write dv\ = [z\] and note that d[z\] = d2v\ = 0. By B0.12) there are elements w\ € P(k — 1) and x\ G M such that d(z\ — u>\) = 0 and m(z\ — w\) — dx\. Extend d and mtoV{k+\)®Aby setting dv\ = z\ — W\ and mv\ = x\. This completes the construction of d and ??. Clearly m is a morphism of (^4,dJ4)-modules, (P, d) is semifree and the u^-term of the spectral sequence is as desired. Finally, since H(m@)) is surjective, H(m) certainly is. To show H(m) is injective let ? be a class represented by a cycle ? G P(k). Put ? = ???, Zi G V(i) ® A. Then dozk = 0 and d[zk) = 0. By exactness [zk] = d[w'] and so Zk = diw' + dQw" for some do-cycle w' G V(k + 1) ® A, and some w" G V(A;) ® .4. It follows that ? - d(w' + w") G P(k - 1). In summary, any class ? in H(P) is represented by a cycle z0 G V@) ® A. If H(m)a = 0 then ?[?0] = 0, [?0] = 9[2{] and z0 = dB{ + ?'?); i.e. ? = 0. D Example 1 The Eilenberg-Moore spectral sequence. Suppose (M, d) is a right (^4, dA)-modu\e with a semifree resolution (P, d) -=> (M, d) as constructed in Proposition 20.11. If (iV, d) is any left (yl,dj4)~module then the filtration {P(k)} defines the filtration P@) ®A ? C P(l) ®A ? C · · · C P{k) ®A ? C · · · of ? ®A N. This leads to the (fundamental) Eilenberg-Moore spectral sequence, whose E1- term is just {{V(k) ® H(A)},d) ®h(a) H(N). The left hand tensorand is an
Graded Differential Algebra 281 i/(,4)-free resolution of H(M). Thus the I?2-term of the Eilenberg-Moore spec- tral sequence is given by El. = Torf(A) (H(M),H(N)) . Under simple hypotheses the spectral sequence converges to H(P<S>a N). As shown in §6, this homology is a functor, independent of the choice of semifree resolution. It is called the differential tor, Diff Tor'1 (A/, N) and so the Eilenberg- Moore spectral sequence converges (usually) from ????{?^(?(?),?(?)) => Diff ????(?,?) . Exercises 1. Let (?, ?) be an augmented graded algebra and A =ker ?. Using §20 b, example 1, prove that Torf (Ik, Ik) = A/AA. 2. Let (A, e) be a graded augmented algebra. Compute the projective dimension of the trivial yl-module Ik when a) A = T(V), b) A = AF,dimF < oo. c) A = Ik[x]/(xn). 3. Let F -» X -» Y be a fibration with Y 1-connected and let M be a semifree C*(F; .^-resolution of C*{X;Ik). Prove that C*(P(Y,yo);Ik)) ®c-(Y;ik) M is a resolution of C*(F;Ik). Deduce the Eilenberg-Moore spectral sequence of a fibration 4. Let F -> X -> Y be a fibration with 1-connected base and let Ik be a field of characteristic zero. Assuming that the Serre spectral sequence of the fibration collapses at the -Eo-term, prove that the graded algebra H*(F,Ik) is isomorphic to the quotient algebra Ik ®H-(Y-,ik) H*(X, Ik). 5. Let ? —» X be a G-fibration as in theorem 8.3. Using the spectral sequence associated to a semifree resolution construct a spectral sequence TorH-{G;rk)(Ik,Hm(P,Ik)) =» H*{X,Ik). 6. Express Theorem 7.5 in terms of differential Tor. 7. Let X = K(G, 1) with universal cover X. Deduce from §8-exercise 1 that H*(X;Ik) 9i TorA'(G](ifc,ifc) and that H*(X,Ik) 2i Ext?[?](Ik, Ik). As an appli- cation, compute Ext#.[G] (Ik, Ik) when G is a free group. 8. a) Let ? and F be Z-modules. Prove that Tor^.2(E, F) = 0. b) Let M and ? be chain complexes over Z. Deduce the following short exact sequences from the Eilenberg-Moore spectral sequence: ®H.(N))n -> Hn(M®N) -> ep+9=n-iTorf(#p(M),ii?(iV)) _> 0.
282 20 Projective resolutions of grad c) Let X and Y be topological spaces. Prove the existence of Kunneth exact sequence: 0 -> (H*(X) ? H.{Y))n -> Hn{X ? ?) ¦
Part IV Lie Models
21 Graded (differential) Lie algebras and Hopf algebras The ground ring in this section is a field k of characteristic zero. S^ denotes the permutation group on k symbols. This section introduces graded Lie algebras together with associated construc- tions such as the universal enveloping algebra. The main example for us is the rational homotopy Lie algebra of a topological space, defined in (c), whose uni- versal enveloping algebra is the loop space homology algebra. We begin with the Definition A graded Lie algebra, L, is a graded vector space L = {L,}ieZ together with a linear map of degree zero, L® L —> L, ? ®y ·-> [x,y] satisfying (i) [x,y] = -(-l)dezxdezy{y,x). (antisymmetry) (ii) [x,b,2]] = [[x,y],z] + (-l)dezxdezy{y,[x,z]]. (Jacobi identity) The product [ , ] is called the Lie bracket. A morphism of graded Lie algebras is a Lie bracket-preserving linear map of degree zero. If ? and F are graded subspaces of L then [E, F] denotes the graded subspace of linear combinations of elements of the form [x,y], ? 6 ?, y 6 F. In particular a sub Lie algebra (resp. an ideal) ? C L is a graded subspace such that [E, E] c ? (resp., [L,E] C E). In either case the restriction of the bracket makes ? into a graded Lie algebra. If ? is an ideal then there is a unique graded Lie algebra structure in L/E for which the quotient map L —> L/E is a morphism of graded Lie algebras. Given a subset 5 C L the sub Lie algebra (or ideal) generated by S is the intersection of all the sub Lie algebras (or ideals) containing 5. The subspace [L, L] is an ideal, called the derived sub Lie algebra, and L is abelian if [L, L) — 0; i.e. if [x,y] — 0 for all x,y 6 L. Example 1 Graded algebras. Recall that graded algebras A are associative by definition (§3(b)). A Lie bracket (called the commutator) is defined in A by [x,y] =xy-{-l)dezxdezyyx with the Jacobi identity following from the associath'ity. Note that this Lie algebra is abelian if and only if A is commutative. ? Let L be a graded Lie algebra. A (left) L-module is a graded vector space V together with a degree zero linear map L<g>V —> V, ? <8> ? >->¦ ? · ?, such that [x,y).v = x.(y >?) - (-l)de*x de*yy -(? >?) . A right L-module is a graded \'ector space V together with a degree zero linear <dx^V'X such that v[x,y] = (v<x)-y- (_l)uegi teiii(v.y).x. ???
284 21 Graded (differential) Lie algebras and Hopf algebras is a left L-module the induced linear map ? : L —> Hom(V, V) is a morphism of graded Lie algebras, where Hom(V", V) is given the commutator Lie bracket of Example 1. This is called a left representation of L in V. A morphism of L-modules a : V —> W is a linear map of degree zero such that a(x ·?) — ? · ??, ? G L, ? G V. Submodules and quotient modules are defined in the obvious way and the tensor product of two L-modules is the L-module Vr <g> W given by x-{v®w) = x-v®w + (-l)degxdeguv®x-u), ? € L, ? 6 V, w 6 IV. Note that there is no tensor product analogue for modules over an associative algebra. Example 2 The adjoint representation. If L is a graded Lie algebra then a representation ad : L —> Hom(L, L) is defined by (adx)(y) = [x,y], x,y 6 L. This is called the adjoint representation, and (by the Jacobi identity) makes L into an L-module. The submodules are precisely the ideals in L. D Example 3 Derivations. Suppose first that A is a graded algebra. The graded space Der A of deriva- tions of A (§3(b)) is a graded Lie algebra under the commutator [a, ?] = a? — (_l)dega deg??a_ jn other words, Der .4 is a sub Lie algebra of Hom(yl, A) with the Lie bracket of Example 1. Now suppose L is a graded Lie algebra. A derivation of L of degree & is a linear map 9 : L —» L of degree k such that ?[?, y] = [??, y) + (-l)kde^x[x, 9y]. The derivations of L form a graded sub Lie algebra DerL C Hom(L, L), again with respect to the commutator. Note that the Jacobi identity states precisely that each ad a: is a derivation of L. Thus ad : L —» Der L. D Example 4 Products. The product of two graded Lie algebras ? and L is the direct sum, ? © L, with Lie bracket f In particular for ? G ?, y G L we have [?, y) = 0 in ? ? L. Example 5 The tensor product, A ® L. Suppose L is a graded Lie algebra and .4 is a commutative graded algebra. Then A <g> L is a graded Lie algebra with Lie bracket [o ® X, b ® y] = (-l)des* des*ab ® [z, y]. D
Lie Models 285 The rest of this section is organized into the following topics: (a) Universal enveloping algebras. (b) Graded Hopf algebras. (c) Free graded Lie algebras. (d) The homotopy Lie algebra of a topological space. (e) The homotopy Lie algebra of a minimal Sullivan algebra. (f) Differential graded Lie algebras and differential graded Hopf algebras. (a) Universal enveloping algebras. Let L be a graded Lie algebra and TL the tensor algebra on the graded vector space, L. Let / be the ideal in the (associative) graded algebra TL generated by the elements of the form x®y-(-l)dezx dezyy®x-[x,y\, x,y e L. The graded algebra TL/I is called the universal enveloping algebra of L and is denoted by UL. Observe that the inclusion L —> TL gives a linear map ? : L —> UL and that ? is a morphism of Lie algebras with respect to the commutator in UL. Conversely, let a : L —> A be any linear map into a graded algebra which is a morphism of Lie algebras with respect to the commutator in A, then the extension of a to an algebra morphism TL —> A annihilates / and so factors to yield a unique morphism of graded algebras ? : UL —> A such that ?? = a. In particular a left representation a of L in V determines an algebra morphism ? : UL —> Hom(V, V) such that ?? = a. This identifies left L-modules with left modules (in the classical sense) over the universal enveloping algebra. Similarly right L-modules are just right UL-modules. In the same way if ? : ? —> L is a morphism of graded Lie algebras then the composite ? -^> L -^» UL determines a morphism of graded algebras ??-.UE —> UL such that Uip ? ? = ?? ?. Example 1 Abelian Lie algebras. If L is abelian then UL = TL/(x®y- (-l)desx dee»y ®x); i.e., UL is the free commutative graded algebra, AL, defined in Example 6 of §3(b). D
286 21 Graded (differential) Lie algebras and Hopf algebras In general the algebra UL is more complex than AL. Nonetheless a funda- mental theorem asserts that UL and AL are isomorphic as graded vector spaces. The first form of the theorem is by an explicit comparison of bases. For this we fix a well ordered basis of L : {xa}aej- Define an admissible sequence M of length k to be a finite (or empty) sequence of indices ai,...,ctk such that &i < a2 < " · " < ak and such that ??· occurs with multiplicity one if xai has odd degree. Define the corresponding admissible ?-monomial to be the element XjX} = ??? ? · · · ? xak G AL (or ?? = 1). Then the admissible ?-monomials are a basis of AL. Next define the admissible [/-monomials to be the elements u<p = 1 and um = {????)¦¦¦ (ixak.) <? UL. Theorem 21.1 (Poincare-Birkoff-Witt) Let L be a graded Lie algebra. Then (i) The admissible U-monomials are a basis of UL. (ii) In particular, the linear map ? : L —> UL is an inclusion and extends to an isomorphism of graded vector spaces AL -^ UL. proof: [139] If M = ??,..., ?* is admissible then so is ? = ?2,..., ak and we write M = ???. We show first that an L-module structure in AL is defined by the conditions xa · ?? = xa and, if ? = ???, {xaM if a < ?? or a — ai and deg?a is even. \{xa,Xa\-XN if ? = ?? and degza is odd. [??,??,]·** + (-l)dee*edegIei*a1 -Xa'XN if » > »1 · Indeed suppose by (transfinite) induction that xa> · xm1 is defined if length M' < length M or if length M' = length M and a' < a. Assume further that xa' ·??· is a linear combination of monomials of length < length M' + 1. Then the right hand side of the formula above is defined, and is a linear combi- nation of monomials of length < length M + 1. By induction xa :Xm is defined for all a and M. To show that this makes AL into an L-module we write (?,?,?) = [x«,X?\-xM-x«-X?>XM+{-l)de%Xa deeX0X?'X*'XM and then verify that (?, ?, A4) = 0 for all ?, ? and M. For this we may clearly suppose ? < a and induct, as before, on length M and on ?. It is immediate from the defining formula that {?, ?,?) = 0. Now suppose by induction that (?',?',?1) = 0 if length M' < length M or if length M' = length M and ?' < ?. Write ? = ???, so that xm = Xai · xn- We proceed in four steps. (i) If ? < »? or ? = ai and degz/? is even then ??'?? = ??? and the defining formula gives (?, ?, ?) = 0.
Lie Models 287 (ii) If ?? — ? = a and if deg xp is odd then by our inductive hypothesis on length M, (?,?,?) = [??,??] ·?? ·?? - 2?? -(?? -?? ·??) = [??,??]·?? ·?? - ??'[??,??]·?? = [[??,??],??]·??. The Jacobi identity implies 3[[??,??],??] = 0, whence (?,?,?) = 0. (Hi) If ?; = ? < a then xa · ? ? is a linear combinations of monomials xm> with either length M' < length M or else ?' = ?iN' with ?? > a. Thus (i) and the inductive hypothesis on length give ? ? ·??·?? ·?? = -y [? ?, ??] ·?? ·?^. Now a simple calculation using the Jacobi identity and induction on length yields (?, ?,?) = 0. (iv) Suppose ? > ai. Use induction on length and the Jacobi identity to verify that (?, ?, M) is a linear combination of expressions of the form (?, ai, M') and (?, ??, M') with length M' < length M. Now the inductive hypothesis implies {?, ?,?) =0. This completes the proof that the defining formula above makes AL into an L- module. The action of L extends to an algebra morphism UL —> Hom(AL,AL) and the linear map UL —> AL defined by a ·-*· a · 1 sends um to xm. Since the ?? are a basis for AL the um are linearly independent. On the other hand, UL is generated as an algebra by t(L) and hence is linearly spanned by monomials of the form i(xai) t(^at) where no restriction is placed on the order of the ??·. However, by construction, we have t(x)t(y) — (-l)desx dezyi{y)L{x) = t[x,y] and i(xJ - \l([x,x\) if \x\ is odd. It follows that any monomial can be written as a linear combination of an admissible monomial um and monomials of shorter length. Induction on monomial length now shows that every element in UL is a linear combination of the um; i.e. the um are a basis for UL. ? Henceforth we shall identify L with its image in UL and drop the notation ? '. If ? is a permutation of k letters and if ??,..., Xk are elements in a graded vector space V then in AkV we have ?? ? · · · ? xk = ???^ ? · · · ? ??(^, where ? = ±1 and depends only on ? and the degrees of the X{. Where the context is clear we abuse notation and denote this sign simply by ??. Recall that 5/t denotes the permutation group on k letters. The following is a useful variant of the Poincare-Birkoff-Witt Theorem: Proposition 21.2 If L is any graded Lie algebra then a natural linear isomor- phism of graded vector spaces, 7 : AL ? UL is given by 7 (?? ? · · · ? xk) = ^ ? ????{\) ' ' · Xa(k) · S
288 21 Graded (differential) Lie algebras and Hopf algebras proof: We adopt the notation of Theorem 21.1 and its proof. Let UL(k) denote the subspace spanned by monomials in L of length < k. The last part of the proof of 21.1 establishes in fact that the admissible monomials of length < k are a basis of UL(k). Moreover, if xai ¦ ¦ -xak is an admissible monomial in UL(k) and if ? is any permutation then xai ¦ ¦ ¦ xah — ???????) ¦ ¦ ¦ ???{1:) ? UL(k — 1). Hence for any admissible sequence M of length k, j{xm) — u-\j G UL(k — 1). Since the um with M of length k represent a basis of UL(k)/UL(k — 1) while the xm are a basis of AkL, it follows that 7 is an isomorphism. D Corollary If ? Q L is a sub Lie algebra of a graded Lie algebra L then UL is a free left (or right) UE-module. proof: Write L = E@F (as graded vector spaces) and identify AL = Ai?® AF. Define 7' : AE <g> AF —>· UL by 7'(? ® b) = 7@) · 7F). The same calculation as in the Proposition shows that 7' — 7 : AkL —> UL(k — 1). Hence 7' induces iso- morphisms AkL JZ^ UL(k)/UL(k— 1), which implies that 7' is an isomorphism. Since 7(A?) = UE C UL it follows that multiplication in UL defines an isomorphism UE®^(AF) -^> UL. This exhibits UL as the free left C/fJ-module on 7(Ai*1). The same argument shows it is a free right U^-module. D (b) Graded Hopf algebras. A graded Hopf algebra is a graded vector space G which is simultaneously a graded algebra and (§3(d)) a graded coalgebra (with comultiplication ? : G —> G®G and augmentation ? : G —> Je), such that both ? and ? are morphisms of graded algebras. Note that the multiplication in G is automatically a morphism of coalgebras and that 1 6 Go provides the coaugmentation. The Hopf algebra is cocommutative if ?? = ? where r(a <g> b) — (—l)des° deg&e 0 a. Morphisms of graded Hopf algebras are linear maps of degree zero that preserve all the additional structure. Suppose G is a Hopf algebra and recall (§3(d)) that an element a: 6 G is primitive if ?? = x®l + l®x. If ? and y are primitive then so is the commutator bracket [x,y]: A([x,y]) = A{xy-(-l)dezxdezyyx) = i\xi\y - (-l)de8* de%yAyAx = [x ® 1 + 1 <8> x,y <8> 1 + 1 <8> y] = [x, y] <g> 1 + 1 ® [x, y) . Thus the space P*(G) of primitive elements is a graded Lie algebra with respect to the commutator bracket. Conversely, if L is a graded Lie algebra then UL is naturally endowed with the structure of a cocommutative graded Hopf algebra. To define the diagonal we recall first that if ? and L are two graded Lie algebras then elements of ? commute with those of L in ? ? L (Example 4). It follows that UE and UL commute inside U(E © L) and so the inclusions extend to morphism of
Lie Models 289 graded algebras UE ®UL —» U(E ? L). On the other hand the Lie morphism ? @ L —-> [/?¦ <g> UL given by (z, y) >-> ? <g> 1 + 1 <g> y extends to an algebra morphism U(E ? L) —> UE <g> UL. Since these two morphisms reduce to the identity in ? and L they are inverse to each other: U(E ®L) = UE <g> f/L as graded algebras. When ? = L the diagonal ? : L —> L ? L, ? : ? ?-*· (?, ?) extends to C/i : f/L —>· f/(L ? L), which we may now write as ?-.UL—iUL® UL. Similarly, ? : UL —> k is obtained from the trivial Lie morphism L —> 0. These are both algebra morphisms by construction. From (? ? ??)? = (id ? ?)? we deduce (? <g> id)?. = (id <g> ?)?. The identity (? <g> ??)? = id = (??®?)? is even easier. Thus (UL, ?. ?) is a graded Hopf algebra. To see that it is cocommutative let ? : L ? L —> L ? L be the involution (x,y) *-l· (y, x). Then ?? = ? so that (?/?)? = ?. But a trivial check shows that (Ua)(a <g> b) = (-l)desQ deebb <g> a. Note also that the isomorphism 7 of Proposition 21.2 is an isomorphism of graded coalgebras. Proposition 21.3 The inclusion L —> UL is an isomorphism of L onto the graded Lie algebra of primitive elements in UL. proof: It is immediate from the definition of UL that the inclusion is a mor- phism of Lie algebras. To see that it is an isomorphism onto P*(UL) define a Hopf algebra structure in AL with diagonal ?? the unique algebra morphism given by ?? (?) = x <8> 1 + 1 <8> ?, x G L. Since ? is an algebra morphism a very short computation shows that the linear isomorphism 7 : AL —> UL of Proposition 21.2 is an isomorphism of coalgebras. Thus it is sufficient to prove that L is the primitive subspace of AL. Fix a basis {xa} of L. Since ?? is an algebra morphism, ?? (?*1, ? · · · ? x^r) can be expanded easily. In particular it contains a term of the form k\xla <g> (ij'~' ? · · · ? ?^.) and this term cannot appear in the diagonal of any other monomial in the {xa}- Thus a linear com- bination of monomials is primitive if and only if they all have length 1; i.e., all primitive elements in AL are in L. ? (c) Free graded Lie algebras. Recall that TV denotes the tensor algebra on a graded vector space V. This is a graded Lie algebra with the commutator bracket. The sub Lie algebra generated by V is called the free graded Lie algebra on V and is denoted by Ly · This is justified by the appropriate universal property. A linear map V —> L of degree zero, L a graded Lie algebra, extends uniquely to an algebra morphism TV —> UL. Now L is a sub Lie algebra of UL (Theorem 21.1 (ii)) and so this extension restricts to a map hv —> L which is necessarily a morphism of graded Lie algebras. Note as well that the inclusion Ly —> TV extends to an algebra morphism ULy —> TV. On the other hand, the inclusions V c—> Ly c—>· C/Ly extend to
290 21 Graded (differential) Lie algebras and Hopf algebras a morphism TV —> ULv ¦ Since V generates Ly as a Lie algebra it generates ULv as an algebra. Since these two morphisms reduce to the identity in V they are inverse isomorphisms: TV=ULv ¦ Now as usual we say elements in TkV have wordlength k. Analogously, we say that an element in Ly has bracket length k if it is a linear combination of elements of the form [??,..., [vk-i, ?*] · · · ]. Since Ly is generated by V it is the sum of its subspaces of bracket length k. Since these spaces are contained in TkV we may deduce that • Ly = ? (Ly DTkV). it>l • ? G Ly has bracket length k if and only if ? G Ly ? TkV. Now let L be any graded Lie algebra and choose any graded subspace V C L such that L = V ©[L,L]. Extend the inclusion of V to a morphism ? : Ly —> L of graded Lie algebras. Proposition 21.4 If L = {L{}i;>1 then ? is surjective. Moreover the following three conditions are equivalent: (i) ? is an isomorphism, (ii) L is free. (Hi) projdinn/ii*) = 1 (cf. \20(c)). proof: Put ? = a(Ly); it is the sub Lie algebra generated by V. Clearly ? + [L, L] = L. Since [E, E] C E, substitution gives ? + [L, [L, L]] = L. Iterating this process gives ? + [L,[L,...{L,L]...,]J = L for any k. Since L = {Li}i:>1 any obvious degree argument gives ? = L. It remains to prove the equivalence of (i), (ii) and (iii). Clearly (i) =>¦ (ii). If (ii) holds then write L = L\v and UL = TW. Let sW be the suspension of W : (sW)i — Wi-i. Then the free resolution 0 <— Jfc <— TW ^- sW ® TW +— 0 ,
Lie Models 291 d(sw®a) =w<g>a, shows that proj dirndl;) = 1. To show (iii) =?- (i), suppose proj dime//,($;) = 1. Because L = {-?'i}i>1, k has a free resolution of the form Of- k^-UL^- Phm< with Pi,» = {Pi,i}i>0- Now the proof of Proposition 20.10 provides a projective resolution of the form with ? C Pi,». Since ? is projective it has the form ? = W ® UL, by the Remark at the start of 20(a). Thus d embeds W (with a shift of degrees) as a graded subspace W C UL and it is a simple exercise to see that UL = TW. Next observe that {UL)+ = V® {UL)+ -{UL)+ . In fact since ? is surjective it follows that (UL)+ = V + (UL)+ -(UL)+. On the other hand, denote the abelian Lie algebra L/[L,L] simply by F. Its universal enveloping algebra is just AF and the composite map UL —> AF —> AF/A-2F kills (UL) + -{UL)+ and sends V isomorphically to F. Hence Vn(UL)+ -{UL)+ = 0. Now filter by the wordlength in TW to see that for any subspace V com- plementing (UL)+ -(UL)+ in (UL)+ we have UL = TV. In particular we may choose V = V. Then the identification UL = f/Ly is an isomorphism of Hopf algebras and hence restricts to an isomorphism L = Ly of primitive Lie algebras , which is inverse to ?. ? Corollary IfV — {Vi}i>1 then any sub Lie algebra L C Ly is free. proof: As noted in the proof of (ii) =$¦ (iii) in the Proposition, k has a C/Ly- free resolution of length 2. Since C/Ly is a free C/L-module, (Corollary to Propo- sition 21.2) this is a free C/L-resolution and proj dime//,($;) = 1. ? Example Free products. Suppose {L(a)}acj is a family of graded Lie algebras. The free product. U L(a), is the graded Lie algebra L defined as follows: write V = ®L(a) and a a denote by ia: L(a) —> V the inclusion. Let / C Ly be the ideal generated by the elements ia[x,y] — [iax,iay], x,y € L(a), a G J. Then \J L(a) = Ly//. a Note that ia induces a morphism ja : L(a) —> \J L(a) of graded Lie al- a gebras, and that any family of graded Lie algebra morphisms ipa : L(a) —> ? determine a unique morphism ? : \J L(a) —> ? such that ?? = ? oja, a G J. a In particular there is a morphism ?? : [] L(u) —> L(a) such that gaja = id a a"d Q?ja = 0. ? ? 0. Thus each ja is an inclusion.
292 21 Graded (differential) Lie algebras and Hopf algebras Next suppose A(a) = Jfc © A(a) are graded associative algebras, with A(a) — ( ? an ideal. We define the free product, ]J A(a), and show that U I \J L(a) I = a \ a J U UL(a). Indeed, let J be the set of all finite sequences a\,..., aq of index ? elements such that aj ? chj+i. Set with the product of ?,? ® · · · ® aq ? -A(a:i) <g> · · · <g> --4(a?) and ?>i <g> · · · <g> ?>r 6 A(a[) ® · · · <g> A(a'r) given by ?? <g> · · · <g> ???>? <g> 60 <8> · · · <8> &r if »? = <*? and by ai ;8> · · · ® aq ® h <g> · · · <g> ?>r otherwise. Clearly morphisms ·?? : A(a) —> ? of graded associative algebras extend uniquely to a morphism JJ A(a) —> B. a Use this, and the analogous universal property for ?J L(a) to construct inverse ? isomorphisms between U ]\ L(a) and \J UL(a). a a Finally, suppose A = h®A and ? = ic©? are associative graded algebras and that F» —> h and Qr —> h are respectively an ?-free and a Z?-free resolution beginning with the augmentations Pq = A —> k and Q\ — ? —> Jk. Note that AUB is ^4-free on the direct sum of the subspaces B®A®··· and h. It follows that _ P>\ ®a (? ? ?) -=> A ®A {A U B) is an A II 5-free resolution. Since A II 5 = A~ ®A (A II5) ? ? ®B {A II B) we obtain an yl II Z?-free resolution of k of the form [?>? ®? (^4 U ?) ? Q>i ®b (A U B)} —> ? U ? —> Jfc . This yields the formulae Tori*UB(Jfc1 -) = Torf (k, -) ? Torf (Jfc, -) , i > 1 and Ext^UB(i:,-) = Ext^(l;, -) ® ExtB(J:,-) , i > 1 . These apply, in particular, to case that A and 5 are universal enveloping alge- bras. (d) The homotopy Lie algebra of a topological space. Let A" be a simply connected topological space. In §16 we defined the loop space homology algebra, Ht(UX:k). Moreover, the Alexander-Whitney diago- nal H(&) : Hr(flX;k) —> Hr(ClX;k) ® Hr(flX; k) is a morphism of graded algebras, as was observed in Lemma 16.3. The map ClX —> pt is trivially product preserving and so the canonical augmentation
Lie Models 293 ? : Hr(UX;k) —> k is also a morphism of graded algebras. In other words, the Alexander-Whitney diagonal and the canonical augmentation make Hr(QX;Jk) into a graded Hopf algebra. On the other hand, recall that since PX is contractible, the connecting ho- momorphism for the path space fibration is an isomorphism, dr : ?» (?) -^> ?,_?(?,?). Thus we may transfer the Whitehead product [,}w ???,(?) (§13(e)) to a bracket [ , ] in ?, (??) by setting [?,?] = (-l)de*a+1dr {[d:la,d:l?)w) , ?,? G ??,(??) . Theorem 21.5 (Milnor-Moore) [127] If X is a simply connected topological space then (i) The bracket above makes ?, (??) <8> Se into a graded Lie algebra (to be denoted by Lx). (ii) The Hurewicz homomorphism for ?? is an isomorphism of Lx onto the Lie algebra P,(flX;Jk) of primitive elements in H*(QX;Se). (iii) The Hurewicz homomorphism extends to an isomorphism of graded Hopf algebras, ULx -=> proof: Consider first the case that X has rational homology of finite type. The Cartan-Serre Theorem 16.10 asserts that hur : ?,(??'") <g> Se -^ Pr(UX:Se) is a linear isomorphism. Hence a unique Lie structure is induced in ?,(??) (8>ic such that hur is an isomorphism of Lie algebras. Proposition 16.11 asserts precisely that for ?,? G ?»(??), [hura,hur?] = (_i)^ga+i hurdr ([?^?,?^?]^ . Thus the induced Lie structure in ?, (??") ® Se is given by the required bracket. This proves (i) and (ii). For assertion (iii) note that, by the definition of universal enveloping algebras, hur extends uniquely to a morphism of graded algebras ? : ULx —> ?*(??; Je). Theorem 16.13 states precisely that this is an isomorphism. Since L\ is primitive in ULx (Proposition 21.3) the algebra morphisms (? ® ?)? and ?(?)? agree in Lx. But Lx generates the algebra ULx. It follows that (? ® ?)? = ?(?)?; i.e., ? is an isomorphism of Hopf algebras. Now consider the general case. A weak homotopy equivalence X' —> X from a CW complex X' (Theorem 1.4) will induce a weak homotopy equivalence ??"' —> ?? (long· exact homotopy sequence) and hence a homology isomor- phism H»(QX';Je) -=> H*{UX\Se) (Theorem 4.15). Thus, given the proof of Theorem 1.4, we may suppose X is a CW complex with a single 0-cell and
294 21 Graded (differential) Lie algebras and Hopf algebras no 1-cells. Then any singular chain c G C*(X;k) and any continuous map / : 5" —> A' will satisfy c € C*(Y;k) and f(Sn) C Y for some finite sub- complex Y C X. If / : ? —> ?? is a continuous map from a compact set then all the loops f(y), y G ? have length bounded above by some I. Moreover {/(?/) @ j 0 < i < ^, y G iv} is a compact subset of X and hence contained in some finite complex Y. Thus any singular chain c G C, (??; le) and any contin- uous map f : Sn —> V.X will satisfy c G C*(UY;]k) and f(Sn) C QY for some finite subcomplex Y C X- Since the theorem has been proved for finite subcomplexes Y C A it follows now in general. D Definition The graded Lie algebra Lx = (?,(???) ® ic, [ , ]) is called the homotopy Lie algebra of X with coefficients in Ik. When h = Q it is called the rational homotopy Lie algebra of X. Observe that all the constructions above are functorial: if / : X —> Y is a continuous map between spaces satisfying the hypotheses of Theorem 21.5 then ??,(?/) <S> 1; is a morphism of graded Lie algebras and hur is a natural isomorphism. (e) The homotopy Lie algebra of a minimal Sullivan algebra. Let (AV, d) be a minimal Sullivan algebra, as defined in §12. Thus the differ- ential may be written as an infinite sum d = d\ + g?2 + " ¦ of derivations, with dk raising wordlength by k. As observed in §13(e), (AV,d\) is itself a minimal Sullivan algebra. Define a graded vector space L by requiring that sL = Hom(V, A) , where as usual the suspension sL is defined by (sL)k = Lk-\- Thus a pairing ( ; ) : V x sL —> A is defined by (v;sx) = (-l)degusx(u). Extend this to (k + 1)-linear maps AkV ? sLx ¦¦¦ ? sL —» A by setting (Vi A ¦¦¦ AVk;SXk,-·· ,SXl) = V" ??(??(?)-,8??) ¦ ¦ ¦ (va(k)-,SXk), where as usual Sk is the permutation group on k symbols and ??(?) ?- - -???^) = Definition A pair of dual bases for V and for L consists of a basis (vi) for V and a basis (xj) for L such that 1 if i = j 0 otherwise.
Lie Models 295 We observe now that L inherits a Lie bracket [ , ] from d\. Indeed, a bilinear map [ , ] : L ? L —> L is uniquely determined by the formula (v;s[x,y]) = (-l)de^+1{dlV;sx,sy) , x,yeL,veV. The relation ? ? w = (-l)deeu aegww ? ? ieacjs at once to [x,i/] = -(-l)degxdegl'b,x], and an easy computation gives (d?u; sx, sy,sz) = (-l)d**v(v,s[x, [y, z]] - s[[x,y],z] - (-I)«1*«» de*vs[y, [x, z]]). Thus the Jacobi identity is implied by (indeed is equivalent to) the relation Definition The Lie algebra L is called the homotopy Lie algebra of the Sullivan algebra (AV,d). This construction is functorial. If ? : (AV, d) —> (AW, d) is a morphism of minimal Sullivan algebras then recall (§12(b)) that Q(ip) : V —> W is the linear part of ? : ?? — Q(tp)v € A-2W. It is immediate that the algebra morphism kQ{ip) ¦¦ AV —> AW satisfies AQ(ip) ? di = dx ? AQ(ip). Now let L and ? denote respectively the homotopy Lie algebras of (AV, d) and (AW, d). Then Q(ip) dualizes to a Lie algebra morphism ? : L <— ?, {?,???) = (Q(tp)v;sx) , as follows from the compatibility of AQ(tp) and d\. Finally, suppose X is a simply connected topological space with rational ho- mology of finite type. In §21 (d) we defined the homotopy Lie algebra Lx = 7?»(??) ® h. On the other hand, X has a minimal Sullivan model m: (AV,d) —> APL(X) with its own homotopy Lie algebra L. Now, by definition, sL = Hom(V, Jc). Identify sLx = nr(X) ® k by setting sa = — (—l)dega571a, where 5, : ?*(?) ^» ?,-^??) is the connecting homo- morphism for the path space fibration. In §13(c) we used the quasi-isomorphism, m, to define a bilinear map ( ; ) : V ? ?*(?) —> Jk (Lemma 13.11). Define a linear map ? : ?,(?) ® Jfe —>· Horn(V, A) by (??)? = (-l)de^a(v;a). Theorem 21.6 The linear map ? : Lx > L defined by ? (sa) = saa, a € Lx, is an isomorphism of graded Lie ?/
296 21 Graded (differential) Lie algebras and Hopf algebras proof: Theorem 15.11 implies that ? is an isomorphism of graded vector spaces. Hence so is ?. To check that ? preserves Lie brackets, recall that if ?,/3 € Lx then their Lie bracket, as defined in §21 (d), satisfies d^ [] d On the other hand Proposition 13.16 states that (v,[sa,s?\w) = (-l)deea+dee0+1 (d^sa^s?), ? € V . Thus the Lie bracket in Lx satisfies {dxv;sa,s?) = (-l)d**0+1(v;s[a,?\) , which is the defining condition for the Lie bracket in L. D (f ) Differential graded Lie algebras and differential graded Hopf alge- bras. A differential graded Lie algebra (dgl for short) is a graded Lie algebra equipped with a differential d satisfying d[x. y] = [dx, y] + (—l)deg x[x, dy]. Morphisms, sub- algebras and ideals have the obvious meaning in the differential context. A chain Lie algebra is a dgl in which L = {Lj}z>o- If Lo = 0, L is a connected chain Lie algebra. A (left) module for a dgl, (L, d) is a complex (V, d) together with a (left) representation of L in V such that d(x ·?) = dx · ? + (—l)deg:rx · dv. Let (L,d) be a dgl. Extend d uniquely to a derivation of square zero in the tensor algebra TL and observe that this preserves the ideal I generated by the elements ? ® y — (-l)de6x de&yy ® ? - [?,y]. Hence the universal enveloping algebra UL = TL/I inherits a differential d, and with it the structure of a dif- ferential graded algebra. The differential graded algebra (UL,d) is the universal enveloping algebra of the dgl, (L,d) and is often denoted by U(L,d). Note that (L, d)-modules are precisely the modules over the dga, U(L, d) and so can be treated with the techniques of §6. Note also that if ? is a morphism of dgl's then ? ? is a morphism of dga's. As with dga's, the homology H(L) of a dgl inherits the structure of a graded Lie algebra as follows: if ? and w are cycles in L representing homology classes a and ?, then [?, ?] is the class represented by the cycle [2,11;]. If ? is a dgl morphism and if (V, d) is an (L,d)-module then ? (?) is a morphism of graded Lie algebras and [2] ·\?] = [? · ?] makes H(V) into an if (L)-module. Finally, let (L, d) be a dgl and consider the inclusion ? : (L, d) —> U(L, d) as a dgl morphism. Thus H(l) : H(L) —> H(UL) is a Lie morphism with respect to the commutator bracket in the graded algebra H(UL). In particular, it extends to a morphism of graded algebras, UH(L) —> H(UL).
Lie Models 297 Theorem 21.7 (i) The morphism UH(L) —> H(UL) is an isomorphism. (ii) A dgl-morphism ? is a quasi-isomorphism if and only if ?? is a quasi- isomorphism. proof: (i) Recall the isomorphism 7 : AL —> UL of Proposition 21.2. Extend d to a derivation in AL; then 7 is a morphism of complexes. Write L — ? © V @W where d = 0 in ? and d : V —> W. Then, as is noted in Lemma 12.5, H(A(V © W)) = Jfe and so (??, 0) ^> (AL, d). Identify ? = H(L) and observe that ?? A UH(L) —» H{UL) coincides 7 with the isomorphism ?? ? ? (AL) -^-> H(UL). Thus UH(L) A H{UL). ?(?) (ii) The isomorphism of (i) identifies H(Utp) with UH(tp). But 7 iden- tifies AH (?) with UH(ip) and so ? (?) is an isomorphism if and only if UH(ip) is. D Finally, we define differential graded Hopf algebras: these are simply graded Hopf algebras together with a differential compatible with all the algebraic struc- ture. It is immediate from the definition that U(L, d) is a differential graded Hopf algebra. In §26 we shall see that so as Cm(UX;Jk) and establish an equivalence (up to homotopy) between C*(fLY;l:) and a suitable U(L,d). Exercises 1. Suppose ? G L and consider x2 G UL. Prove that x2 G L. Is it true that xk € L for k > 3? 2. Let L be the quotient of the free Lie algebra L(a, 6) by the ideal generated by the element [a, [a, b]] G L(a, 6). Prove that if ? and 6 are of even degree then a2b + ab2 = 2a[a,b] in UL. 3. Let L = V 0 L" be the sum of two graded Lie algebras. Prove that UL = UV ®UL". 4. Let F ~> ? ? ? be a fibration with 1-connected base and fibre. Prove that if ?»(?) ® (Q> is onto then the graded algebra H*(UX,Q) is isomorphic to H*(VY,Q) ® H.(CIF,Q). 5. Compute the homotopy Lie algebra Lx when A' = CPn, ? > 2, X = SpVSq, X = CPp V CP? for p, q > 2, or X = CP°° V CP°°. Deduce if,(??,«© in each case.
298 21 Graded (differential) Lie algebras and Hopf algebras 6. Let (A,d, e) be a differential graded augmented algebra. Deduce from §19- exercise 1 that if A is commutative, then B(A) is a commutative differential graded Hopf algebra with the shuffle product defined in §16-exercise 1 and the usual coproduct. Determine the Hopf algebra B(Ik[x]/(x2)) when the degree of x is even. 7. Let (A, e) be a graded augmented algebra. Prove that if A is commutative, then TorA(Ik, Ik) is a graded Hopf algebra. 8. Prove that the adjoint representation (example 2) extends to a left represen- tation of L in AL and that the morphism 7 of Proposition 2 is L-linear. Deduce that UL is the direct sum of the L-modules j(AkL), k = 0.1.... 9. Let 0 -> L' ~> L ~> L" ~> 0 be a short exact sequence of graded Lie algebras. Define the Hopf kernel of Uf : UL -* UL" by HK(f) = {x G UL, (id ? /)(?? - ? <g> 1) = 0)}. Prove that HK(f) is isomorphic to UL' and that UL = UL' ®UL", 10. Let Ik be a commutative ring with unit and ? be the least prime, if any. which is not a unit in Ik. Prove that (p — 1)! is a unit in Ik. Let (L,d) be a differential graded Lie algebra such that Li = 0 for i < r. Prove that the natural inclusion (L,d) -> (UL,d) induces injective maps Hn(L,d) -> Hn(UL,d) for ? < rp. 11. Let A' be a simply connected finite type CW complex. Supposing that dimii»@X;Q) < 00 and that H+(X;Q) ? 0. Prove that X has the rational homotopy type of a finite product of Eilenberg-MacLane spaces of the form Ar(Q,2n). 12. Let V be an ?-dimensional Q-vector space concentrated in degree 2. Using Poincare-Birkhoff-Witt, compute the dimension of the brackets of length 2r in hv, for r > 1.
22 The Quillen functors C* and C In this section the ground ring is a field h of characteristic zero. A coalgebra of the form C = h ® {Cr!}z>2 will be called one-connected. Such coalgebras are co-augmented by k. The object of this section is to construct Quillen's functors one-connected r , , . c . connected chain cocommutative ; > ,. . , , . . *~^ Lie algebras chain coalgebras °- and to establish the natural quasi-isomorphisms C(C,(L,'dL)) A (L,dL) and C. (C(C,d)) ?- (C,d). In §24 we shall see how this correspondence exhibits chain Lie algebras as an alternative description for rational homotopy theory. The passage from (L, ul) to C,(L, ul) is essentially inverse to the construction of the homotopy Lie algebra of a minimal Sullivan algebra given in §21 (d) and we make this remark precise in §23. This section is organized into the following topics: (a) Graded coalgebras. (b) The construction of C*(L) and of C*(L; M). (c) The properties of C*(L; UL). (d) The quasi-isomorphism C*(L) ^» BUL. (e) The construction of C(C, d). (f) Free Lie models. (a) Graded coalgebras. Recall (§3(d), §19) that a graded coalgebra C is equipped with a comulti- plication ? : C —> C <g> C and an augmentation ? : C —> h and that a co-augmentation is an inclusion k c—> C so that ?A) = 1 and ?A) = 1 ® 1. For such a coalgebra we write C = kere, so that C = k® C._ _ _ As described in §19, the reduced comultiplication ? : C —> C ® C is defined by Ac = ?? — c ® 1 — 1 ® c. Its kernel is the graded subspace of primitive ejements. Now set ?^0^ = id^, ?^ = ? and define the nth reduced diagonal ?(") = ' (? <g> id <g> ¦ · ¦ <g> id) ? A(n-V :C—>-C®---®C(n + l factors). We say C is primitively cogenerated if C = |JkerA(n). Notice that if C = k®C>0 then ? C is automatically primitively cogenerated.
300 22 The Quillen functors Cm and C Finally, recall that C is cocommutative if ?? = ? where ? : C ®C —> C ® C sends a ® b ·-> (-l)dee Q dee 66 ® a. The main example, for us, of such graded coalgebras is the coalgebra AV whose comultiplication ? is the unique morphism of graded algebras such that A(v) = ? ® 1 + 1 ® ?. ? G V". It is augmented by ? : \+V —> 0, 1 m· 1 and co-augmented by fo = A°V. It is trivially cocommutative. It is also easy to see that kerii*") = A-nV, and so AV is primitively cogenerated. Among the primitively cogenerated, cocommutative coalgebras (AV, d) has an important universal property. Let ? : A+V —> V be the surjection defined by Lemma 22.1 Suppose C = Jk ® C is a primitively cogenerated cocommutative graded coalgebra. Then any linear map of degree zero, f : U —> V lifts to a unique morphism of graded coalgebras, ? : C —> AV such that ?? |^- = /. proof: Define fW : C ® - · - ® C —> Ak V by /(Cl) ? · · · ? f{ck). Recall that A(o^ = id ? and define ? by ?A) = 1 and ceC. (Since C is primitively cogenerated, this is a finite sum.) To verify that ? is a coalgebra morphism, write A^k~^c = 5Zcf <g> · · · <g> a c?. Co-commutativity implies that ?^'^ = JZ ^Oi) ® ' ' ' ® c"(fc)j ^or eacn ? permutation, ?. Co-associativity implies that ?(?) = (?(?) ® ?(?)) ? ? for all p,q such that ? + q = k — 1. Finally, for any vi G V, ? · · · ? ?*) = (vi ® 1 + 1 ® ui ) ? · · · ? (vk ® 1 + 1 <8> ? k = E^ITF^II ? ±Mi)a'"aMp)®)M) p=0 ^ V ^' CTgSfc Combining these facts in a straightforward calculation gives (? ® ?) A = ??. To prove uniqueness observe that any morphism maps kerA^1^ into V. Hence if ? is a second morphism then for c € kerA^1^, ?? = ??? = f(c) = ??? = ??. Suppose ? and ? agree in kerA^") and let c G kerA^n+1^. By co-associativity, Ac G kerA(n) ® kerA(n). Thus Aipc = (? g> ?)?? = (? ® ?)Ac = ???. This implies that (? — ?)? = (?? — ??)? = 0. ? An analogue of Lemma 22.1 also holds for coderivations. Indeed, suppose g : AkV —> V is a linear map of some arbitrary degree, and k > 1. Define
Lie Models 301 ?9 : AV —> AV by 9g(vi A •••Avn) = ^2 ±9^1, A ¦¦¦ Avilc) Avi A-¦¦uil---vik---Avn. (Here means deleted and ± is the sign given by v\ A ¦ ¦ ¦ A vn = ±vi1 A · · · ? Vik A v\ A ¦ ¦ ¦ ujj · · · Vik ¦ ¦ ¦ A vn.) Note that ?9 decreases wordlength by k — 1. Lemma 22.2 ?9 is a coderivation in AV. It is the unique coderivation that extends g and decreases wordlength by k — 1. proof: The coderivation property is a simple calculation. Uniqueness is proved in the same way as in Lemma 22.1. D (b) The construction of C,(L) and of C,(L;M). Let (L,cIl) be a differential graded Lie algebra. The differential and the Lie bracket determine coderivations in AsL given explicitly by (Lemma 22.2) k do(sx\ A ¦ ·· ? sxk) = — ^(—l)nisxi ? · · · ? sd^x,- ? - · · ? sx/., and di (sxiA- ¦ -Asxk) = ]? (-l)degIi+1(-l)"ijs[xi, Xj]As.Ti · · ¦ satt-·· sxj · · -Asxk. (Heren,· = ? degsxj, and sxiA- ¦ -Asxk = (—l)nijsxiAs sxk. The symbol "means 'deleted'.) Since d0 and d\ are coderivations of odd degree, g?o, dQd\ + dido and d\ are also coderivations, decreasing wordlength by 0, 1 and 2 respectively. Moreover dl(sx) = sd\x = 0; (dodi + dido)(sx A sy) = (-l)de^s {dL[x,y] - [dLx,y] - {-l)d^*[x,dLy}) = 0; d\{sx AsyA sz) = (-l)d^y+^s ([x> [y, z]] _ [[x,y],z] - (-l)degx degy^ [XjZjj) = 0 . Thus by the uniqueness assertion in Lemma 22.2, all three coderivations are zero. Hi other words, (AsL, d = do + dy) is a differential graded coalgebra. Definition The Cartan-Eilenberg-Chevalley construction on a dgl, (L, g^) is the differential graded coalgebra C.(L,dL) = We often abuse notation and write simply C(L) for C*(L,di).
302 22 The Quillen functors C, and C This construction is functorial: if ? : (E,d?) —> (L,cIl) is a dgl morphism then C9 (?) is the dgc morphism ?? : As ? —> AsL, with Tp(sx) = ???. Remarks 1 The construction C,(L) is a cocommutative co-augmented and primitively cogenerated coalgebra. 2 For the construction for Lie algebras see Chevalley-Eilenberg [38] and Cartan-Eilenberg [35]: for the differential graded case see Quillen ([135], Appendix B). 3 The reader may have noticed two apparent coincidences: the first between the construction above and the homotopy Lie algebra of a Sullivan model (§21(d)); the second between the construction above and those in §19. These are not coincidences. The first will be explained in §23 and the second in §22(d), below. D Next, we extend the construction above to the case with coefficients in a graded (L,c??)-module, (M,d). Denote the differential in C(L) by d= d0 +d~i. Define a complex (C*(L; M),d = do + d\) as follows: • C*(L; M) — AsL <g> M as a graded vector space. • d0 = d0 ® id + id ®d. • di = d\ <8> id +?, where k 9(sx\ ? · · · ? sxk <8> m) = /,(—l)niS?i ? · · · sx?· · · · ? sx/. ® ?,· ·?? k k and m = ? (dee*i + 1) + ? (degxi + lXdegx,· + 1). (Note that, as always, the tensor product of linear maps is defined by (/ <8> 5)(a® b) = (-l)d**'d**af(a)®g(b).) As in the case of C*(L) we have dl = dod\ + dido — d\ = 0. Indeed ?% = 0 because d2 = 0, dido + dodi = 0 because d(x-m) = dLX-m + (—l)degxx-dm and d\ = Q because [x,y]-m = x-y-m- (-l)degz degi'y ·? ·??. (c) The properties of C,(L; [/?,). Recall that (UL,d) = U(L,dL) is the universal enveloping algebra of the dgl (L,di) — cf. §21(e). Multiplication from the left makes UL into a left t/L-module, and hence into an L-module. Form the complex C*(L;UL) = (AsL <g> UL,d), as described in (b). Proposition 22.3 The inclusion k —> (AsL®UL,d) is a quasi-isomorphism. Proposition 22.4 Right multiplication makes (AsL®UL, d) into a right semifre (UL,d)-module.
Lie Models 303 proof of 22.3: In the proof of Proposition 21.2 we introduced the subspaces UL(n) C UL spanned linearly by the monomials ?? ¦ --xk, x; € L, k < n. We showed there that a linear isomorphism 7 : AL —> UL is given by x\ ? · · · Axk ?-4 ^r ? ????(?) ' ' · xa(k) > and that it induces isomorphisms AkL -^ UL(k)/UL(k — ? 1)- Now write AsL ® UL as the increasing union of the graded subspaces Ff. = ? A-lsL <g> UL{j). As usual, identify AsL <g> AL = A(sL © L). Then ^*(sL ? L) A Ffc, Jfe > 0, and so induces isomorphisms Ah(sL ®L) Next observe that the differential, d, preserves the spaces F/.. Moreover, be- cause s[xi,Xj] has lower wordlength than sxi A sxj the corresponding terms in the formula for d (see (b), above) disappear in Fk/Fk-i. Also, ?-7B/1 ? ··· ? y le) — j(x ? yi ? · · · ? i/fc) € UL(k). These comments imply that id ®j is an isomorphism of complexes (Ak(sL®L),S) A (Fk/Fk^,d) where ? is the derivation in the algebra A(sL ? L) specified by 6x = dx and Ssx = (-l)degx+1x - sdx, ? e L. Trivially, H(sL ® L,S) = 0. Choose a basis of sL ® L of the form {???, iua}. Then A(sL ? L, d) = (g) A(uQ, i«Q) and so ? (A(sL ? L), i) = Jk; i.e. ? ii (Ak(sL ? L),i) = 0, k > 1. Because of the isomorphism above, ii (Fk/Fk-i,d) = 0, k > 1. It follows by induction on k that ii (Fk/F0,d) = 0, k > I, and so if ((AsL ® C/L)/Jk, d) = 0, as desired. D proof of 22.4: It is immediate from the definitions that multiplication from the right makes C*{L; UL) = AsL ® UL into a right (UL, d)-module. Moreover „the subspaces M{k) = A-ksL 0 UL define an increasing family of submodules, and (M(k)/M(k - l),d) = (AksL,d0) ® (UL,d). The two step filtration (kerd0) ® (UL,d) C (AksL,d0) ® (UL,d) exhibits (M(k)/M(k - 1), d) as (UL, d)-semifree. Now Lemma 6.3 asserts that C,(L; UL) is (UL, d)-semifree. ? Notice that the construction Ct(L; UL) is functorial: if ? : (?, d) —i (L, d) is a morphism of dgl's then ?*(?) ® ?? : (AsE ® UE, d) —> (AsL ® UL, d)
304 22 The Quillen functors C, and C is a morphism of (UE, d)-modules, where UE acts on UL from the right via ? ?. Moreover, because of the acyclicity (Proposition 22.3), C*(ip) ®Uip is automati- cally a quasi-isomorphism. Recall from Theorem 21.7 that ? is a quasi-isomorphism if and only if ? ? is. Using the construction above we prove Proposition 22.5 (i) If ? is a quasi-isomorphism then so is C*(<p), (it) If ? = {?{};>() and L = {L,}i>o then ? is a quasi-isomorphism if and only if C, (?) is a quasi-isomorphism. proof: (i) Here Uip is a quasi-isomorphism (Theorem 21.7) and C,(ip) ®Utp is a quasi-isomorphism because of Proposition 22.3. Since the modules Ct (E; UE) and C,(L; UL) are semifree, Theorem 6.10 (ii) asserts that (C«(y>) <8> Uip) ®?? k ¦¦ (As? <g> UE) <%UE Jk —> (AsL ® UL) is a quasi-isomorphism. But this is precisely C,(?). (ii) Conversely, suppose C,((/j) is a quasi-isomorphism. Consider (UL, d) as a left (UE, d)-module via multiplication on the left by (Uip)a, a 6 UE. Then C.(tp)®id:C.(E;UL) ^Ct(L;UL) is a morphism of right semifree (UL, d)-modules. Moreover (??(?) <g> id) ®ul Ik is the quasi-isomorphism C* ((/>). Now Theorem 6.12 asserts (because ? and L are concentrated in strictly positive degrees) that ?»(?) ? id is a quasi-isomorphism. Hence H(C*(E;UL)) = Jk. Now choose a (UE.d)-semifree resolution of left (U?, d)-modu\es, a : (M,d) ^ (UL,d), with M = {Mi}i>0. Since Ct(E;UE) is (UE,d)-semifree, C*(E; UE)®ue — preserves quasi-isomorphisms. Furthermore, since M is semi- free, — <8>ue A/ preserves quasi-isomorphisms. Thus k ®?? ? 4^- C*(E; UE) ®?? ? ^^ C*(E; UE) ®UE UL = C.(E; UL); i.e. H(k ®ue M) = k. Define ? : (UE,d) —> (M,d) by ?? = a-m, where m is a cycle in Mo such that H(a)[m] = 1. The calculation H (He ®?? ?) = k shows that k <%>?? ? is a quasi-isomorphism, and hence so is ?, again by Theorem 6.12. Moreover, H(a) is an isomorphism by construction. Thus H(U<p) = H(a) ? ?(?) is an isomorphism. Now Theorem 21.7 asserts that ? is a quasi-isomorphism. D We now return to the construction C,(L\UL). Recall from (a) above that AsL is a cocommutative graded coalgebra and from §21 (e) that (UL,d) is a differential graded Hopf algebra.
Lie Models 305 Proposition 22.6 (i) The tensor product of the diagonals in AsL and in UL makes C,(L;UL) into a differential graded coalgebra. (ii) The linear isomorphism 7 : AL -=» UL of Proposition 21.2 is an isomor- phism of coalgebras. Hence C» (L; UL) = AsL <g> AL as a graded coalgebras. (in) The module action (AsL®UL,d)®(UL, d) —> (AsL®ULxd) is a morphism of graded coalgebras. (iv) The quotient map — <S>ul k '¦ (AsL <g> UL, d) —> (AsL, d) is a morphism of graded coalgebras. proof: The tensor product of the diagonals in AsL and in UL certainly makes AsL <8> UL into a graded coalgebra. Moreover the isomorphism 7 : AL -=> UL of Proposition 21.2 is an isomorphism of graded coalgebras, as follows from the relation Aul(x\ •••in) = (x'i <8> 1 + 1 ®xi)---(xfc ® 1 + l<g>zjt) , x* € L . To show that the differential is a coderivation denote the differentials in AsL = C« (L) by do and d\ ; they are both coderivations. Since d is a coderivation in UL (§21(e)) it follows that do = do® id + id®d is a coderivation. Moreover, recall that d\ = d\ <g> 1 + ? where d\ is a coderivation and 9(sx\ ? · · · ? sxk <?> a) = ? ±sxi ¦ ¦ ¦ sii ¦ ¦ ¦ sxk <S> Xid. Since Am is an algebra morphism. Am(x{a) = (xi ® 1 + 1 ® Xj)A(/?O. This implies (after a short calculation) that ? is a coderivation. The remaining assertions are self-evident. D "(d) The quasi-isomorphism C,(L) -^ BUL. As we observed earlier, the constructions Ct(L) and Ct(L:M) bear a strong similarity to the bar constructions of §19. We can now make this precise. Let (L,d) be a dgl and denote by BUL the bar construction on the dga (UL.^d). As in §19, B(UL:UL) denotes the bar construction with coefficients in UL, where UL acts on itself via left multiplication. Now BUL is the tensor coalgebra TsUL on the suspension of the augmentation ideal HZ c UL. The inclusion L C UL (§21 (a)) has its image in UL; hence we may regard sL as a subspace of sUL. Proposition 22.7 A natural dgc quasi-isomorphism A : C.(L) ^ BUL is given by A : sxi ?- · -Asx^. ?-» ^ ?^[sxCT(i)| · · · \sxa(ic)], where sxi ?- · -Asx/t = S
306 22 The Quillen functors C, and ? proof: A straightforward, if tedious, calculation verifies that A is a morphism of dgc's. Similarly. A ® id : (AsL ® UL, d) —> (BUL ® ?/L, d) is a morphism of (UL, d)-modules. According to Proposition 19.2, H(BUL <g> t/L) = 1; and (BUL®UL, d) is (?/L, d)-semifree. According to Propositions 22.3 and 22.4, (AsL <g> UL, d) has the same two properties. Because H (BUL <g> ?/L) = Jk = H (AsL <g> ?/L), A <g> zc? is necessarily a quasi- isomorphism. Because both are (UL, d)-semifree, Proposition 6.7 (ii) asserts that (A <g> id) ®ul & is a quasi-isomorphism. But, trivially, (A <g> id) ®ul Jk = X. D. Remark Proposition 22.7 replaces the huge complex T(s~1UL) by the rel- atively small complex AsL, which is a major computational advance. In the classical case of ungraded Lie algebras without differentials, the smaller complex in fact came first, as the dual of the left invariant forms on a Lie group. The bar construction and the equivalence of Proposition 22.7 appear later in [118] and [35]. (e) The construction L(C,d). Recall that the free graded Lie algebra Ly is the sub Lie algebra of TV gener- ated by V" (§21(c)) and that TV is the universal enveloping algebra of Ly. Thus a dgl of the form (Ly, d) will be called a free dgl Note that this is a serious abuse of language: (Ly ,d) is almost never a free object in the category of dgl's. Recall first some basic facts about free Lie algebras Ly (cf. §21(c)). First, Ly is the direct sum of the subspaces Ly of bracket length i, and Lv = V. Second, any linear map / : V —> L (L a graded Lie algebra) extends uniquely to an algebra morphism TV —> UL, which then restricts to a Lie morphism Ly —> L, the unique extension of /. Finally any linear map g : V —> Ly extends uniquely to a derivation ? of TV which then restricts to the derivation of Ly uniquely extending g. Let (Ly, d) be a free dgl. The linear part of the differential d is the differential dy '¦ V —> V defined by dv — dyv 6 0 Ly . It suspends to a differential d in sV : dsv = —sdyv. Now consider the composite ? : C*(Ly ) = AsLy —> shy ?!" —> sV ? ]k, where we have first annihilated A-2sLy and then s(Ljr ). Proposition 22.8 The linear map ? is a natural quasi-isomorphism of com- plexes, (C.(Lv),d) A (sV ®k,d).
Lie Models 307 proof: It is immediate from the definitions that ? commutes with the dif- ferentials. Moreover, since Uhy = TV, there is an analogous morphism ?' : (BUhv,d) —> (sV(BJk,d) constructed in §19 just before Proposition 19.1. More- over, Proposition 19.1 asserts that ?' is a quasi-isomorphism. It is immediate that the quasi-isomorphism of Proposition 22.7, A : C» (Lv ) -^> BUhy, satisfies ?'? = ?. Thus ? is a quasi-isomorphism. D A prime (but by no means the only) way of constructing free dgl's is through Quillen's functor, C, which is the analogue here of the cobar construction. The functor C is constructed as follows. Let (C, d) = (U, d) ®Jk be any co-augmented dgc, which is cocommutative. Form the cobar construction, Q.C — Ts~xC, as described in §19. The differential has the form d = do + d\ with do preserving s~1C and di : s~1C —> s~lC ® s~xC. Moreover, because C is cocommutative, we can express d\ (s~1c) as a sum of commutators in the tensor algebra Ts~lC. Indeed, write Ac = ?>j ® bt. Then also Ac = ?(-i)deeQi deg&,o. ^ o. jjence (_l)(dega,-l)(degfc-l)s-l&i This shows that di : s~xC —> ^s-iU c ?C~??). Because d\ is, in particular, a Lie derivation it preserves Ls_i^. Hence so does d; i.e., (Ls_ic, d) is a dgl with universal enveloping algebra flC This construction is obviously functorial. Definition The dgl (hs-i-g,d) will be called the Quillen consti-uction on the co-augmented cocommutative dgc, (C, d) and will be denoted by C(C,d). Theorem 22.9 Suppose (L = {Li}i>i,d) is a connected chain Lie algebra and (C = k ? C>2,d) is a cocommutative dgc. Then there are natural quasi- isomorphisms ? : (C, d) ^ C.C(C, d) and ? : CC.(L, d) A (L,d) of dgc's (respectively, of dgl's). proof: (i) Existence of ?. Write C = C>->. Thus ?(C,d) = i-s-ic anc* = AsLs_i^. Thus by Proposition 22.8 we have a quasi-isomorphism of com- plexes,
308 22 The Quillen functors C, and C Now ss~lC = C, and a quick inspection shows that d is the original differential in C: However, g is not a coalgebra morphism. To obtain one, apply Lemma 22.1 to the inclusion / : C = ss-^ C sL^c C noting that C is trivially primitively cogenerated because C = {Ci}i>2· This gives a unique morphism ? : C —> AsLs_i ^ of coalgebras such that (? — f){C) C A-2. We show first that ? is a dgc morphism, and then that it is a quasi- isomorphism. For simplicity, write L = Ls_! ^ and denote the differentials in L by d = do + ?? with dos~1c = —s~ldc and d\ reflecting the diagonal. Then write ? = ??? where ?? : C —> iVsL is the component of wordlength i. Let ? : A+sL —> sL be the projection with kernel A-2sL. Now observe that ?(?? — dip) — 0 or, equivalently, that ??? = ???? + dnp2- Indeed, ???\? = do(ss~1c) = —sds~1c. In defining ?(C,d) we gave an explicit formula for dsc. Thus if, Ac = ? c; ® c'v then On the other hand, ?\?? = ss^ and (cf. Lemma 22.1) dupoc = ?? I ^J2ss~1ci A ss~ic'i j It follows that ??? = Put ? = ?? — ??. We have just seen that fo = 0. The coderivation property implies that (? ® ? + ? <S> 0)? = ??, where ? denotes both reduced diagonals. By coassociativity, ? : kerA^"^ —> kerA(n~^ ®kerA("\ where ?(?) is the nth reduced diagonal as defined in §22(a). Assume by induction that ? vanishes on kerAt"-1). Then for c € kerA(n\A0c = 0 and 9c = ??? = 0. It follows by induction that 0 = 0; i.e. ?? = ??. Finally, note that ?? = id. Since ? is a quasi-isomorphism, so is ?. _ (it) Existence of ?. Put C = C+{L) so that C = A+sL. Define ? : s~1C —> L by setting a(s~lsx) = ? and aCs'^a) = 0, ? 6 A-2sL. Then since Ls_ic is a free Lie algebra., ? extends to a unique Lie morphism -? : i-s-ic —> L. As in (i) we have to check that ? commutes with the differentials and that ?(?) is an isomorphism. The linear map ? = ?? - ?? satisfies 0[a,b] = [??,??] + (~l)desa[ipa,0b], a, b 6 Ls_!^. Since this Lie algebra is generated by s~*C it is enough to check that ? vanishes there. This is a straightforward computation from the definition of the differentials, as in (i).
Lie Models 309 Let ? : Ct(L) ^> C»Ls_!c be the dgc quasi-isomorphism constructed in part (i) of the theorem (for general cocommutative dgc's). A quick computation shows that 0*{?)? = id. Hence Cm(ip) is a quasi-isomorphism, and then so is ? by Proposition 22.5. ? Corollary Suppose a : (C,d) —> (C',d) is a morphism between dgc's satisfying the hypotheses of the theorem. Then a is a quasi-isomorphism if and only if C(a) proof: The theorem identifies H(a) with ? (CtC(a)). Now apply Proposi- tion 22.5. D (f) Free Lie models. In §22(e) we introduced the free dgl's (L\.-,d). Thus (Ly,d) is a connected chain Lie algebra precisely when V — {Vi}z>1. These play the same role within the category of connected chain Lie algebras that Sullivan algebras play in the category of commutative cochain algebras, and for the same reason: the under- lying algebraic object is free. In particular we make the Definition A free model of a connected chain Lie algebra (L, d) is a dgl quasi- isomorphism of the form m : (Lv, d) ^ (L, d) with V = {Vi}4>1. There is a simpler approach, however, which extends any morphism 7 : (Lh' , d) —1 (L, d) of connected chain Lie algebras to a free model: (? is the obvious inclusion) (Lw@v,d) ~ (L,d) ?\ ? B2·10) (Uv,d) Indeed, suppose for some r > 0 that Hi(j) is an isomorphism for i < r and surjective for i = r. Extend 7 to 7' : (Lu'©v,.+1, d) —> (L,d) by requiring that
310 22 The Quillen functors C, and ? • The elements dv, «el^!,,, are cycles in hw representing all the classes in • The elements in V"+i are cycles and are mapped by 7 to cycles representing all the elements in coker #,.+1G). Then ??(?') is an isomorphism for i < r and surjective for i = r + 1. Iterate this procedure to construct B2.10). D Free Lie models of the form B2.10) are analogues of the relative Sullivan models of §14. This motivates the Definition A free Lie extension is a morphism ? : (Liv, d) —> (Lwev, d) of free connected chain Lie algebras, in which ? is the obvious inclusion. Suppose now given a commutative diagram (Lw,d) ^^ (E,d) ? (Uv&v,d) > (L,d) of morphisms of connected chain Lie algebras. Assume ? is a free extension and ? is a surjective quasi-isomorphism. Proposition 22.11 With these hypotheses a extends to a morphism ? : (Lw®v,d) —> (E,d) such that ?? = ?. proof: Identical with that of Lemma 12.4, except that the induction is in the degree of the elements in W. ? Suppose next that (Ly, d) is a free connected chain Lie algebra. Then d decomposes uniquely as the sum of derivations d; : V —> L^+1 . The differential d0 : V —> V is the linear part of d, as defined in (e) above. Similarly, if ? : (Lw, d) —> (Ly, d) is a morphism of free connected chain Lie algebras then the linear part of ? is the chain complex morphism defined by ? - ?0 : W —* L^-2). Proposition 22.12 If ? : (h\v,d) —> (Lv,d) is a morphism of free connected chain Lie algebras then ? is a quasi-isomorphism <=$¦ ?? is a quasi-isomorphism.
Lie Models 311 proof: According to Proposition 22.5 and Proposition 22.8 respectively, ? is a quasi-isomorphism <=> C, (?) is a quasi-isomorphism <?=>¦ ?? is a quasi-isomorphism. Definition A free connected chain Lie algebra (Ly, d) is minimal if the linear part do of the differential is zero. In this case a quasi-isomorphism m : (Lv,d) -=* (L,d) is called a minimal free Lie model. Theorem 22.13 Every connected chain algebra (L,d) admits a minimal free Lie model m:(Lv,d) A (L,d) and (Ly, d) is unique up to isomorphism. proof:" Theorem 22.9 provides a surjective free model, which we write as Decompose W = V ®U ® doll with d0 = 0 in V and do : U ^> doll. Next, let / c Liy be the Lie ideal generated by U and by dU. Thus / is preserved by d and (Liy //, d) is a quotient connected chain Lie algebra. Denote the quotient map by ? : (Lw, d) —> (h\v //, d). It is an elementary exercise in algebra to verify that the inclusions of V, dU and U into Liy extend to an isomorphism ? (V ®dU®U) -=^> TW (filter by the ideals T-k). Hence Lw = Ly © / and ? restricts to an isomorphism Ly —> Liv//. Use this to identify (Ljy //,d) as a free chain Lie algebra (Ly, J). It is straightforward to check that the linear part of ? is the linear map ?? : W —> V which is the identity on V and zero in U and doU. Since dO?Oi; = ?>odou = 0 it follows that (Ly,d) is minimal. Moreover ?(?0) is an isomorphism and hence ? is a quasi-isomorphism (Proposition 22.12). Finally, Proposition 22.11 allows us to lift id^v through ? to obtain ? : (Ly,d) —> (h\v, d) such that ?? = id. Then ?? = ?? : (Ly,d) ^> (L,d) is a minimal Lie model. For uniqueness, let m' : (Ly,<f) -^> (L,d) be any other minimal Lie model. Lift m' through the surjective quasi-isomorphism ?, and compose with ?. This gives a quasi-isomorphism
312 22 The Quillen functors C, and ? between minimal free connected chain Lie algebras. Now Proposition 22.12 asserts that ?(??) is an isomorphism. But d0 and d'o are zero, so ?? itself is an isomorphism. Now apply Proposition 22.12 to ? : (Lv,0) —> (Lv,O) to conclude that ? is an isomorphism. D Notice that the last paragraph of the proof of Theorem 22.13 in fact establishes Proposition 22.14 A quasi-isomorphism between minimal free connected chain Lie algebras is an isomorphism. D Exercises 1. Prove that if deg ? = 2n - 1, ? > 1, then ? : ? (?) -4 ?? <g> T(u) defined by ip(v2k+1) = ? <8> uk and <p{v2k) = 1 ® uk is an isomorphism of coalgebras (see §3-exercise 4). Deduce that Ht(US2n;Q) = H*(Sn;Q) ® Ht(US4n~1;q). 2. Let ?,~? : (Lv ,d) -4 (L,d) be two morphisms of differential graded algebras. A degree /c-linear map ? : Lv -4 L is an (?, ?)-derivation if it satisties ?([?, y]) = [??,???] + (-l)d'g ?[??, Oy]. We write ? ~ ? if there exists an (?, ^-derivation of degree 1 satisfying [?, d] = ? - ?. Prove that ~ is an equivalence relation.
23 The commutative cochain algebra, C*(L,dL) In this section the ground ring is a field k of characteristic zero. In particular, we write #„(-) and H*{-) for Hm(-;Jk) and #·(-; Jfc). In §22 we constructed the dgc C*{L,cIl) associated with a dgl, (L,di). Here we dualize that construction to obtain a commutative differential graded algebra, C'(L,dL). When L is a chain Lie algebra, C*(L,di,) is a cochain algebra, and so the full Sullivan machinery of Part II may be applied. In particular we can use chain Lie algebras (L,di,) to model topological spaces A" by requiring that C*(L,dL) be a model for Apl{X). This is carried out in §24. This section is organized into the topics (a) The constructions C*(L, cIl) and ?{A,d)· (b) The homotopy Lie algebra and the Milnor-Moore spectral sequence. (c) Cohomology with coefficients. (a) The constructions C*{L,di), and C^^y Again, let (L,di,) be a dgl. Dual to the construction C»(L, c2l) we define the differential graded algebra, C*(L,dL), by C*(L,dL) =Eom(C,(L,dL),Jk) as described in §3(d). Thus multiplication and the differential are given by and Df)() (l)d*'/(fc) ''* ??? (When the differential in L is clear from the context we may abuse notation and simply write C*(L) and C»(L), as in §22.) This is entirely analogous to the construction in §5 of the singular cochain algebra on a space X as the dual of the singular chain coalgebra. However here we have the crucial fact that C*(L) is a commutative dga, because C»(L) is cocommutative. Suppose now that (L,dL) is a connected chain Lie algebra. Then C»(L) = AsL = k © {Ci}i>o. It follows that C*(L) is a cochain algebra and so the machinery of Sullivan models (Part II) can be applied. We shall see now that if L is a connected chain Lie algebra and if each Li is finite dimensional then C'(L) is, itself, a Sullivan algebra (§12). Indeed, the decomposition AsL = ®AksL defines a surjection AsL —> sL k (not compatible with the differential in AsL), which dualizes to an inclusion (sL)& = Hom(sL, Jfc) <—> C*(L) .
314 23 The commutative cochain algebra, C*(L,cIl) Since C*(L) is commutative this extends uniquely to a morphism of graded algebras Lemma 23.1 // (L,d) is a connected chain Lie algebra and each Li is finite dimensional then ? : A(sLY A C*{L) is an isomorphism of graded algebras, which exhibits C*(L) as a Sullivan algebra. proof: Let y, = sx{ be a basis for sL and let vj be the dual basis for (sL^ : (vj,yi) = o{j. live (sL)> and ? € Ap(sL)* then (???,?/i, A---Ayip+1) = (?®?,?^?? ? · · · Ayip+l)) , and C= E(-l)de6^ de6*(v,yi.)(<e,vilA-~Vi.-~Ayip+l). It follows that (?*1 ? ¦ ¦ · ? ?*- ,j/?» ? · ¦ · ? yf1) = ??! ¦ · ¦ knl (where ki < 1 if |vi| is odd) and that vf1 ?· · ·???" evaluates all other monomials to zero. Since char h = 0 this implies that ? is an isomorphism, because C»(L) is finite dimensional in each degree. Now ? identifies C*(L) as a cochain algebra of the form (AV, d) with V = {]/p}p>2. Such a cochain algebra is always a Sullivan algebra. Indeed define V@) CVA) C ··· by V@) = VDkevd and V[k) = VDd'1 (AV(k-1)). Since (AVK = F3, d(F2) C F3 and V2 C F(l) dj^ffc). Suppose V^" cU^(fe). ?· ?- For ? € F", civ = w + ? with w 6 Fn+1 and ? 6 AV-n~l. Thus B? = -?? 6 ? ( (J V{k) ) and so ? € (J F(A;). But then also ? ? (J F(A;) and it follows that V = \JV(k). a k If F and W are graded vector spaces of finite type then an isomorphism W = Vi defines a pairing ( ; ) : W ? V —> L· which in turn determines an isomorphism V = WK In the case we say V and W are dual graded vector spaces with pairing ( ; ). Proposition 23.2 Suppose (L.dL) is a connected chain Lie algebra of finite type and each Li is finite dimensional. Then (i) C*(L, d[^) — (AV,d) and V and sL are dual graded vector spaces. (ii) d = do+di is the sum of its linear and quadratic parts (§12(a) and §13(e)), and
Lie Models 315 (dov;sx) = (-l)de^v(v;sdLx) and (dlV;sxAsy) = (-l)de*v+l(v;s[x,y]) . Conversely, suppose (AV, d) is an arbitrary Sullivan algebra such thatd = do + d\ and V = {Vp}p>2 with each Vp finite dimensional. Then a connected chain Lie algebra {L,di) is determined uniquely by the condition (AV,d) = C*(L,di). proof: The first assertion follows from Lemma 23.1 and the definition of the differential in C»(L,c2l). For the second assertion let L be the desuspension of Hom(F, k) and use the formulae above to define d^ : L —> L and [, ] : L ? L —> L. The equation d2 = 0 reduces to dj = 0, dodi + dido = 0 and ds = 0. These translate respectively to: [ , ] is a Lie bracket (§21(e)), d^ is a Lie derivation and di = 0. D Finally suppose that (.4, d) is a commutative cochain algebra in which ,4 = Jk(B A-2 and each A1 is finite dimensional. Let (C,dc) = Hom(J4, Jfc), recalling that dc is the negative dual of d (§3(a)). Because .4 has finite type, the multi- plication in A dualizes to a comultiplication in C, which makes (C,dc) into a cocommutative differential graded coalgebra with dual cochain algebra (A,d). Apply the functor C of §22 to this dgc and, abusing notation, denote the result by ?{A,dy- ?(A,d) = ?(C,dc) ¦ Note that C(A,d) ls a connected chain Lie algebra, finite dimensional in each degree. Thus the quasi-isomorphism ? of Theorem 22.9 dualizes to a qu'asi- isomorphism of commutative cochain algebras thereby exhibiting C* (C^d)) BS a functorial Sullivan model of (A,d). Example 1 Graded Lie algebras. Suppose L = {Li}i>i is a graded Lie algebra and each Li is finite dimensional. Then Proposition 23.2 reduces to in which sL and V are dual graded vector spaces and d\ is purely quadratic. Conversely, any commutative cochain algebra of the form (?V, di ) with V = { Vp}p>2> each Vp finite dimensional and d\ purely quadratic determines a graded Lie algebraL by the requirement that (AV,d\) = C*(L,0). Next notice that Propositions 22.3 and 22.4 identify C*(L;UL) as an exact sequence 0 <— k^UL^- sL®UL^- A'2 4 of right t/L-modules. In other words (cf. §20(a)) this is a free f/L-resolution of the trivial f/L-module, Jk. On the other hand, it is immediate from the definitions that C*(L) = HomuL (C*(L;UL),]k), and so ).. a
316 23 The commutative cochain algebra, C*(L,Al) Example 2 Free graded Lie algebras. Suppose ? = {?'i}!>2 is a graded vector space of finite type and (H, 0) is the commutative cochain algebra with zero differential defined by H = h@E and E-E = 0 . The dual graded coalgebra has the form C — k@C with ? = 0 : C —>~C& C. Thus the differential in ?(c,o) is zero (§22(e)). In other words, if W is the graded vector space defined by W,· = Horn (?'+?,?) then ?(H,0) = ?(C,0) = CUv j 0) . Thus in this case dualizing Theorem 22.9 provides a cochain algebra quasi- isomorphism and C*.(Lw, 0) is a minimal Sullivan algebra with purely quadratic differential. In other words, this is the minimal Sullivan model of (H,0), and so we recover the construction of Example 7, §12(d). ? Example 3 Minimal Lie models of minimal Sullivan algebras. Suppose (AW,d) is a minimal Sullivan algebra and that W = {W!}->, is a graded vector space of finite type. Let (Ly ,9) be a minimal Lie model of ?(AW,d) (Theorem 22.13). Then by Proposition 22.5 the quasi-isomorphism ? : (Ly ,9) -=> ?(AW,d) induces a quasi-isomorphism C*(a) which then dualizes to a quasi-isomorphism C*(a) : C* (Ly ,9) <=- C* (?(\w,d))· On the other hand, as remarked above, we have a quasi-isomorphism C* (?(Aw,d)) -^ (AW,d). This has a homotopy inverse (AW,d) -^ C* (?(\w,d)) (Proposition 12.9), and the composite ?: (AW,d) A C*(Ly,9) exhibits (AW,d) as a minimal model for C* (Ly,9): z/(Ly,9) is a minimal Lie model for (AW,d) then (AW, d) is a minimal Sullivan model for C* (Ly ,9). Recall further that C* (Ly,9) is itself a Sullivan algebra (AZ,d) with ? = Horn (sLy, Jc). Thus dividing by A~2W and A-2Z yields a commutative diagram H+(AW,d) —^-+ H+(C*(U>,d)) < W =—- Horn (sH (Lv, 9), jfe) , where Q(tp) is the linear part of ? and is a quasi-isomorphism because ? is (Proposition 14.13). On the other hand, Ly = V ? [Ly, Ly ] and Im 9 C [Ly, Ly ]. Thus dividing by [Ly,Ly] is a surjective chain map ? : (Ly,9) —> (^»0). The quasi-isomorphism
Lie Models 317 C>o (Lv,<9) -=» sV of Proposition 22.8 converts the inclusion shy —> C, (Ly) into ??. Its dual therefore converts the dual of ?? into the surjection A+Z —> ?. Thus it identifies Horn (sH(?), k) with the map ?' in the diagram above, and we obtain the commutative diagram H+(AW,d) ^— Hom(sV,Jfe) Horn (sH (Ly ) , Jfe) >2 Example 4 Sullivan algebras (AW,d) for which H2k(AW,d) =0, 1 < k < n. Here we consider minimal Sullivan algebras (AW,d) such that W = { is a graded vector space of finite type. The surjection (A+W,d) —> (W,0) with kernel A^2W induces a linear map ? = {?? : HP(AW) —> Wp}. We shall use Lie models to show that H'2k(AW,d) = 0, 1 < k < ? => Cp is injective , ? < 2n + 1 . In fact, let (Ly,<9) be a minimal Lie model for ?(Atv,d) (Theorem 22.13). Then (Example 3), H+(AW,d) = Eom(sV,Je) as graded vector spaces. Thus Vofc-i = 0, 1 < k < ? and L\/<2n is concentrated in even degrees. For degree reasons the differential in Ly<2n is then zero. Recall from Example 3 that ? : (Ly,<9) —> (KO) is the surjection with kernel [Ly,Ly]. Since d vanishes in degrees < 2n, ^G7) is surjective for i < 2n. It follows that Horn (sH(?),&) is injective in degrees ? < 2n + 1. Example 3 identifies this dual with ?; i.e. ?? is injective, ? < 2n + 1. D (b) The homotopy Lie algebra and the Milnor-Moore spectral se- quence. Consider an arbitrary minimal Sullivan algebra (AW,D) such that W = {Wp}p>2 and each Wp is finite dimensional. Then D = Di + Do ? with Di : W —> Ai+1W. Associated with (AW, D) is its homotopy Lie algebra ? (as defined in §21(e)), and it is immediate from that definition that (AIF,?>i) =C*(E) (where ? is regarded as a dgl with zero differential). Indeed, as observed in Proposition 23.2, this relation characterizes E. Next suppose (L,cil) is a connected chain Lie algebra with each L/ finite dimensional. Let m: (AW,D) ^ C'(L,dL) be a minimal Sullivan model and let ? be the homotopy Lie algebra of (AIT7, D). As in Proposition 23.2, write C*(L,Bi,) = (AV,d0 + di) with V dual to sL. Let Q{m) : W —> V be the linear part of m (§12(b)). Then d0Q(m) = 0.
318 23 The commutative cochain algebra, C*{L.d,L) Proposition 23.3 With the hypotheses above, Q(m) induces an isomorphism W -^ H(V, do). Its desuspended dual is an isomorphism of graded Lie algebras, H(L) A E . proof: Use Theorem 14.9 and Theorem 14.11 to extend m to an isomorphism of the form (AW, D) ® (A(U © SU), ?) A (AV, d0 + dx) with ? : U -^ OU. It follows that Q(m) induces an isomorphism AH{Q{m)) : (AW,D{) A (AH(V,do),H(dx)) . Now V is dual to sL and do is dual (up to sign and suspension) to c2l. This identifies H(V,d0) as the dual of sH(L) and (AH(V,do),H(di)) as C (H(L),0). Since (AW,Di) = C*(E,0), the 'desuspended' dual of H(Q(m)) is an isomor- phism from H(L) to the homotopy Lie algebra E. D Remark The effect of replacing C*(L) by (AW,D) is to get rid of the linear part do and to convert d\ to Dx = H(d\). However we 'pay' for this simplification through the addition of (possibly infinitely many) higher order terms Di, ?>3,... in the sum D = Finally, consider a general Sullivan algebra (AV,d) in which d = do + di • · ·, with di : V —> Ai+lV. Filter (AV, d) by the decreasing sequence of ideals FP = (AV)+ (AV)+ , p>l, ? factors and set F° = AV. Then (independently of the choice of generating space V) Fp = A-PV, and so this is called the wordlength filtration. It determines a first quadrant spectral sequence (E{,di): the Milnor-Moore spectral sequence of the Sullivan algebra. Any morphism ? : (AV, d) —> (AV',d') of Sullivan algebras is automatically filtration preserving and so induces a morphism ??(?) of Milnor- Moore spectral sequences. It is immediate from the definition that (Eo,do) = (AV,d0) and ?0(?) = AQ{ip) where Q(ip) : V —> V is the linear part of ? (§12(b)). Next, let m: (AW,D) ^ (AV,d) be a minimal Sullivan model (Theorem 14.12) and suppose W = {Wp} and has finite type. It follows from Theorem 14.9 and Theorem 14.11 that m extends to an isomorphism (AW, D) ® (A(U © OU), ?) -=> (AV, d)
Lie Models 319 with ? : U -^ 5U. Since this isomorphism and its inverse preserve wordlength filtrations it induces an isomorphism of Milnor-Moore spectral sequences. In particular, if ? is the homotopy Lie algebra for (AW, D) then (E0,dQ) ? (AW,0)®(A(U®SU),S) , and (E^dy) .? (AW,Dl) = C*(E) . Thus the Milnor-Moore spectral sequence converges from E2 = ExtUE(]k,]k) =*> H(AW,D) . Example 1 C*(L,dL). Let L be a connected chain Lie algebra with each Li finite dimensional. Then the homotopy Lie algebra of C*(L, <2l) is just H(L) (Proposition 23.3) and so the Milnor-Moore spectral sequence for C*(L, c2l) converges from E2 = Extc/^i) (A, k) => HC* (L, dL) . ? Example 2 Topological spaces X. Suppose (AW,D) is the minimal Sullivan model for a simply connected topo- logical space X with rational homology of finite type. Then ? (AW, D) = H*(X;]k), H*(QX;]k) = ULx and ?? is isomorphic to the homotopy Lie alge- bra of (AW,D). (This follows, respectively, from Corollary 10.10, Theorem 21.5 and Proposition 21.6.) Thus the Milnor-Moore spectral sequence for a minimal Sullivam model of X converges from E2 = ExtH.{ii (c) Cohomology with coefficients. Suppose given a graded Lie algebra L and a right L-module M; i.e., M is a right t/X-module. We say this defines a right representation of L in M. Denote the action of ? € L on ? € M by ? · ?. Since (cf. §3(d) and Propositions 22.4 and 22.6) C*(L;UL) is a C,(L)-comodule and a right t/L-module and these structures are compatible, C*(L; M) = Homc/L (C*(L- UL), M) is a C*(L)-module. Since C*(L\UL) is a f/L-projective resolution of h (Exam- ple 1, §23(a)) we may identify H (C*(L;M)) - Extudk M) . B3.4)
320 23 The commutative cochain algebra, C*(L,di) The cohomology of this complex is called the Lie algebra cohomology of L with coefficients in M'. Now suppose that L = {Li}i>1 and that each Li is finite dimensional. Then each Ck{L) is finite dimensional. Suppose further that M = {Ml}i>p some ? ? ? and use the convention Ml = M~i to write M = {??;}?<_?. Thus if a : C»(L) —> M is a linear map, it will vanish on C>_|a|_p. It follows that there are natural isomorphisms of graded vector spaces C*(L;M) = RomUL (C.(L)®UL,M) = Eom (C.(L)tM) (93 ?) = C*(L)®M . Write C*{L) = (AV,d), where V is dual to sL and (du; sz ? sy) = (-l)deg?/+1(v; s[x,y]) (Example 1, §23(a)). Then B3.5) identifies C*(L;M) as the (AF,d)-module given by C'(L;M) = (AV®M,d) with the (AV, d)-modu\e structure simply multiplication on the left. Thus the differential is determined by its restriction d : M —> V <8> M. If we identify V <8> M = Hom(sL,M) via (v ® z: sx) = {-I)de&z^d^x+I'>{v,sx)z then this restriction is given by (dz;sx) = (-i)d*e*+deexz.x , z?M, xeL, B3.6) as follows immediately from the definition of the differential in C,(L;—) at the end of§22(b). Conversely suppose given a (AV, d)-modu\e of the form (AV <8> N, d) in which d-.N^r V® N. Then B3.6) defines a bilinear map ? ? L —> ? and it follows easily from the equation d2 = 0 that this is a right representation of L in TV. By construction, if ? = {??}?>? then Exercises 1. Let M be a L-module and assume that L and M are graded vector spaces of finite type. Prove that the graded dual M* is also an L-module and that there exists a quasi-isomorphism (C*(L; M))* —> C*(L;M#). 2. Let L = L\ © L2 be the direct sum of two differential graded Lie algebras. Prove that C"\l) = C*(LX) ® C*(L2).
Lie Models 321 3. Let (L(V),d) be a free differential graded Lie algebra. Denote by (L(V),d) and (L(V"),d) two copies of (L(V),d). Prove that there is a quasi-isomorphism ? : (h(V © V" © sV),D) -+ (L(V),d) with ?{?') = ?(?") = ? for any ? 6 V, D(sv) -?' + ?" € L^2(V ? V" © sV), and </>(sF) = 0. 4. Two morphisms of differential graded Lie algebras /,g : (L(V),d) ->· (L, <i) are homotopic, / ~ p, if there is a morphism of differential graded Lie algebras F : (L(V © V" © sV),D) -»· (L,d) satisfying F(u') = /(u) and F(v") = g{v). Prove that the homotopy relation is an equivalence relation. 5. Prove that two morphisms of differential graded Lie algebras/, g : (L(V),d) ->· (L,d) are homotopic in the sense of exercise 4 if and only if they are equivalent in the sense of §22, exercise 2. 6. Assume / ~ g : (L(V),d) -+ (L,d). Prove that C(f) ~ C'E). 7. Consider the diagram of differential graded Lie algebras (Lud,) (h(V),d) A (L2,d2) where V = {V^};>0 and dim Vi < co. Prove that if ? is a quasi-isomorphism then there exists a morphism ? : (L(V),d) ->· (Li,di) such that ?? ~ ?. 8. Let h : (L(F),d) -»· (L(W')>cQ. and /,5 : (L(H7),^) -»· (^,«0 be morphisms of differential graded Lie algebras. Assume that ? is a quasi-isomorphism and that fh ~ gh. Prove that / ~ g. 9. Let /, g : (AV, d) —>¦ (A, d^) be morphisms of commutative differential graded algebras. Suppose that / ~ g. Prove that ?(/) 10. Let L be a finite dimensional graded Lie algebra concentrated in even degrees. Prove that there exists an integer ? such that dim HnC*(L, Ik) = 1 and dim HkC(L, Ik) = 0, k > n.
24 Lie models for topological spaces and CW complexes The ground ring in this section is a field Jc of characteristic zero. The fundamental bridge from topological to algebra used in this book is Sul- livan's functor topological spaces — >· commutative cochain algebras , constructed in §10. Thus an algebraic object is a model for a topological space X if it is associated with the cochain algebra Apl(X)· For example (§12) a Sullivan model for X is, by definition, a Sullivan algebra (AVx,d) together with a quasi-isomorphism m:(AVx,d)-Z>APL(X) . Analogously, we make the Definition Let X be a simply connected topological space with rational ho- mology of finite type. A Lie model for X is a connected chain Lie algebra (L, di,) of finite type equipped with a dga quasi-isomorphism m:C*(L,dL) A APL(X) . If L = Lv is a free graded Lie algebra we say (L, <2?,) is a free Lie model for X. If ? : C*(E,dE) -^> Apl{Y) is a Lie model for a second topological space Y then a Lie representative for a continuous map / : X —> y is a dgl morphism <p:(L,dL)—f (E,dB) such that mC*(ip) ~ Api(f)n (as defined at the start of §12). Example 1 The free Lie model of a sphere. In the tensor algebra T{v) on a single generator, v, the free Lie subalgebra h(v) is given by lev if deg ? = 2n hv @ h[v,v\ if deg?; = 2n + 1. Thus the cochain algebra C*(L(w)) is given, respectively, by (A(e),0) , dege = 2n + 1 (?(?,?'), de' = e2) ,dege = 2n + 2. The reader will recognize these as the minimal Sullivan models for spheres constructed in Example 1, §12(a). In other words, we have for all ? > 1 a quasi- isomorphism C*(L(u)) -?* APL(Sn+1) , deg ? = ? ,
Lie Models 323 which exhibits (L(i;),0) as a minimal free Lie model for Sn+1. D As with Sullivan models we often abuse language and refer simply to {L,<1l) as a Lie model for X. Lie models provide a description of rational homotopy theory that is different from but equivalent to that provided by Sullivan models. In particular, if we restrict to simply connected topological spaces with rational homology of finite type then • Every space X has a minimal free Lie model, unique up to isomorphism, and every continuous map has a Lie representative. • Every connected chain Lie algebra, (L,cIl) of finite type, and defined over Q, is the Lie model of a simply connected CW complex, unique up to ra- tional homotopy equivalence. If (L, A1) is a Lie model for X then there is a natural isomorphism H(L) -^» 7?,(??) <8>z k of graded Lie algebras (§24(b)), which is in a certain sense 'dual' to the isomorphism H(AVx) -^ H*{X;k) in the context of Sullivan algebras. The first isomorphism suspends to an isomorphism sH{L) ? ?.??®* , which (up to sign) converts the bracket in H(L) to the Whitehead product in ?» (?) ®k. Finally, given any free rational chain Lie algebra (Ly, d) (connected of finite type) we shall in (d) below, construct a CW complex X for which (L\.·, d) is a Lie model. Here • For each ? > 2, the ?-cells D? of X correspond to a basis {va} of Vn-i. • The sub dgl (Ly<n ,d) is a Lie model for the ?-skeleton Xn. • The homology classes [dva] e ? (hv<n_1) correspond to the homotopy classes [fa] e ??_?(,??_?) <8> ifc of the attaching maps fa : S?~l —> Xn-\- In this case (Ly, d) will be called a cellular Lie model for X. Throughout this section we shall use the properties of Sullivan algebras (§12, §14), often without explicit reference. For example, suppose (L,dL) is a Lie model for a space ?' and (E,dE) is a connected chain Lie algebra of finite type joined to (L, A1) by a sequence of dgl quasi-isomorphisms. Then the Sullivan algebras C*(L,dL) and C*(E,dE) are joined by quasi-isomorphisms of commu- tative dga's and hence suitable application of 'lifting up to homotopy' exhibits (E, d?) as a Lie model for X. Similarly, suppose (A, d.4) is a commutative model for X with A° = k, A1 = 0 and each A* of finite dimension. Then the dga quasi-isomorphism C*C(AtdA) ~> {A,cLa) lifts to a dga quasi-isomorphism C*?(A,dA) ~^ Apl(X), which exhibits ?(A,dA) as a k^e model for X. This section is organized into the following tOpiCS
324 24 Lie models for topological spaces and CW complexes (a) Free Lie models of topological spaces. (b) Homotopy and homology in a Lie model. (c) Suspensions and wedges of spheres. (d) Lie models for adjunction spaces. (e) CW complexes and chain Lie algebras. (f) Examples. (g) Lie model of a homotopy fibre. (a) Free Lie models of topological spaces. Let X and Y be simply connected topological spaces with rational homology of finite type. The next proposition establishes the first two bullets in the in- troduction to this section. Notice however that the 'existence of a CW complex modelled by a chain Lie algebra', but not the uniqueness is established directly in §24(d). Proposition 24.1 (i) The space X has a minimal free Lie model (Lv,d), unique up to isomor- phism, and any continuous map f : X —> Y has a Lie representative. (ii) If Ji = Q, any connected chain Lie algebra {L,Al) of finite type is the Lie model of a simply connected CW complex, unique up to rational homotopy equivalence. proof: (i) The construction in §12(a) provides minimal Sullivan models m ? : (AVx,d) -^ Apl(X) with Vx = {V'?} >0 and each V? finite dimensional, and ??? : (AVV,d) —> Apl(Y) with the same properties. Apply the Quillen construction (§23(b)) to obtain natural quasi-isomorphisms rx : C* (C[/iVx,d)) -^ (AKy.cQ an ?? : C* (C[AVy,d)) ^ (AVY,d) . Composed with ???, this exhibits ?(Avx,d) and ?(AW,d) as free Lie models of X and Y. Next, let ? : (i\Vy ,d) —> (AVx,d) be a Sullivan representative for /. Then ??? = ???*(?^), which exhibits ?? as a Lie representative for /. Finally, the argument in the proof of Theorem 22.13 now provides a surjective dgl quasi-isomorphism ? : C^Vx>d) —> (Lv,d) onto a minimal free Lie chain algebra, which is connected of finite type because C(\vx,d) 1S- This exhibits (Ly, d) as a minimal free model for X.
Lie Models 325 For uniqueness, suppose m' : C*(Lw,d) —> Apl(X) is an arbitrary minimal free Lie model. Lift the quasi-isomorphism C*(hy ,d) —> Apl(X) (up to homo- topy) through m! to a quasi-isomorphism C*(hv,d) —> C*(L\y,d). This yields a chain of dgl quasi-isomorphisms (LW,d) 4— Cc(Lw,d) —> ?c-(Lv,d) —> (^V,d) . Use Proposition 22.11 to invert the first, and Proposition 22.14 to conclude that the resulting quasi-isomorphism (Ljv,<2) —> 0W,<2) is an isomorphism. The same argument constructs a dgl morphism between minimal free Lie models that is a Lie representative for /. (ii) Apply Theorem 17.10 to obtain mc-(L,dL) -C*(L,dL) ^APL(\C*(L,dL)\) , thereby exhibiting {L,dL.) as a Lie model for \C*(L,dt)\. If C*(L,dL) is a Sul- livan model for a second simply connected CW complex X then Theorem 17.12 identifies \C*(L,di)\ as a rationalization of ?. ? (b) Homotopy and homology in a Lie model. Fix a simply connected topological space X with rational homology of fi- nite type, choose a Lie model (L,di) for X, and choose a minimal Sullivan model (AV\,d) for C"{L,di,). We then have specified cochain algebra quasi- isomorphisms (AVx,d) -?* C'(L,dL) -2* APL(X) ¦m q whose composite is a minimal Sullivan model for X. Now the homotopy Lie algebra L\ is just ?,(???) <8> He with a certain Lie bracket, and· in Theorem 21.6 we showed that it was naturally isomorphic to the homotopy Lie algebra of the minimal Sullivan model (AVx,d). Further, in Proposition 23.3 we used the quasi-isomorphism m to identify this homotopy Lie algebra with H(L,di). Together these provide an isomorphism of graded Lie algebras, aL ¦¦ H(L,dL) ? ?,(??') ® h . B4.2) As in §21(e) we identify ?»(„?)®!: as the suspension of 7r»(fil)® Jk by writing sa = —(—l)desad~ia, a G ?»(??) <8> Jk, where <9» is the connecting homomor- phism for the path space fibration. Then 0l suspends to an isomorphism tl :sH(L) -^TTm(X)<S>]k . Now the Lie bracket in ?,(??') <8> ]k, as defined in §21(d), is given by [?,?] = (-l)de^+1o, [dZla,dZl?]w , ?,? G ??.(??') , where [ , ]w denotes the Whitehead product. Since ??, is an isomorphism of Lie algebras we deduce that " TLs[a,?] = (-l)dee* [rLsa,TLs?]w , ?,? G H(L) .
326 24 Lie models for topological spaces and CW complexes Next we observe that a free Lie model (Ly, d) for X also encodes the homol- ogy of X. Indeed the morphism C*(Ly,d) -^ Api(X) induces a cohomology isomorphism, which dualizes to an isomorphism i/(C«(Ly,d)) 4^- H*(X;k). Let d : sV —> sV be the suspension of the linear part dy : V —> V of the differential d in Ly. Then Proposition 22.8 provides a quasi-isomorphism C»(Ly .d) -—> (sV ® lc,d). Altogether then we obtain an isomorphism sH(V,dy) ® k = H*(X;k) . B4.3) In particular, if (Ly, d) is minimal then Hm(X; Je) = sV ® k. Finally, [Ly, Ly ] is the ideal in Ly of elements of bracket length at least two. Thus Ly = V©[Ly,Ly] and [Ly,Ly] is preserved by d, so division by [Ly,Ly] is a surjective linear map 77 : (Ly,d) —> (V,dv)· As in Example 3, §23(a), we have Proposition 24.4 With the hypotheses above the diagram sH(Lv,d) ~ - ?,(?) <S> Jfc hurx sH(V,dv) = H+{X;k) commutes, and identifies sH{q) with the Hurewicz homomorphism hurx. proof: The quasi-isomorphism C«(Ly ) —> sV®Jc of Proposition 22.8 converts the inclusion s(Ly,d) —> C«(Ly) into 577. Thus we have to show that sH(Lv,d) n.(X)®Je hurx HC*(LV) H.(X;k) commutes. But this is precisely the dual of the commutative diagram at the end of§13(c). ? (c) Suspensions and wedges of spheres. In this topic k = Q. We show that a suspension is, rationally, a wedge of spheres. More precisely, we establish the Theorem 24.5 The following conditions on a simply connected topological space X are equivalent (i) The rational Hurewicz homomorphism hurx: ?,(?) <8>Q —> H+(X;<Qi) is surjective.
Lie Models 327 (ii) There is a rational homotopy equivalence of the form V Sna —> X (each na > 2). (Hi) There is a well based, path connected space Y and a rational homotopy equivalence of the form ?? —> X. (iv) The rational homotopy Lie algebra ?,? is a free graded Lie algebra. Before proving the theorem we consider the special case that X = V Sn" with each na > 2, and let ia : Sn<* —> X be the inclusion. The classes [ia] G ??? (?) are mapped by hurx to a basis of H+(X;Q). Hence they are linearly indepen- dent and are mapped by the connecting homomorphism to linearly independent elements wa in ?*(??) <S>z Q = L\. Let W be the linear span of the wa. Proposition 24.6 With the hypotheses and notation immediately above (i) kerhurx = s[Lx,Lx}. (ii) The inclusion of W extends to an isomorphism Lu·· —> Lx. proof: (i) It is clearly sufficient to prove this for a finite wedge of spheres. But then Ht(X; Q) has finite type and X is a suspension. Thus (Proposition 13.9) X has a commutative model of the form k©H with zero differential and ? ? — 0. Now CikQH is a Lie model for X and clearly has the form Cjk®ti — (Ly ,0). In particular, Lx = H(hv ) = Ly. Moreover, Proposition 24.4 identifies hurx with the surjection S7/ : shy —> sV, whose kernel is precisely s[Ly,Ly]. (ii) It follows from (i) that Lx = W ? [Lx,Lx]. Since Lx is free Proposition 21.4 states precisely that h\v —> Lx. Q proof of Theorem 24.5: If (i) holds then we can choose based continuous maps fa : Sn" —> X so that the homology classes H»(fa) [Sna] are a basis of H+\X;Q). Thus the map / : \/aSna —> X defined by the fa induces an isomorphism of rational homology. Hence / is a rational homotopy equivalence (Theorem 8.6) and (i) => (ii). Clearly (ii) => (iii). Suppose (iii) holds. Use the Cellular models theorem 1.4 to reduce to the case y is a CW complex with a single 0-cell. Moreover, since L-zY = Lx we may take X = ??. Now for any finite subcomplex ? C Y we know from Proposition 13.9 that ?? has a commutative model of the form Jk©H with zero differential and ? · ? = 0. Hence ?ft©/r is a Lie model for ?? and clearly this has the form (L\v, 0). Thus LZ^H(LW) =LW. Choose now a graded subspace V C Lx so that V ? [Lx,Lx] = Lx and extend the inclusion of V to a surjection ? : Ly —> Lx (Proposition 21.4). Let a e ker ?. Then atlw for SOme finite dimensional subspace W C V. Moreover
328 24 Lie models for topological spaces and CW complexes by choosing a sufficiently large finite subcomplex ? C Y we can arrange that W C L-?z and that the morphism L\v —> Lsz sends a to zero. But clearly W ? [Lt.z,L-?z\ = 0. Thus, since L-zz is free, Proposition 21.4 implies that Lu-' —> L-?z is injective. Hence ? = 0, and ? is an isomorphism. It follows that Lx is free; i.e., (iii) => (iv). It remains to show that (iv) => (i). We show (iv) => (ii), observing that (i) is a trivial consequence of (ii). If Lx = Lv choose based continuous maps fa : Sn" —> X such that the homotopy classes [fa] G ???(?) are mapped by the connecting homomorphism d, to a basis va of V C ?? (= 7?,(??") ®Q). The fa define a continuous map / : \JaSn° —> X. On the other hand, write S = \/aSna and let ia ¦ Sn° —> S be the inclusion. Then by Proposition 24.6, the elements wa = d.[ia] are the basis for a subspace W C Ls such that Ls = Lw. Since 7?.(?/) : wa ?—> va it follows that 7?*(?/) ® Q is an isomorphism and / is a rational homotopy equivalence. ? Example (Baues [19]) If H+(X;Q} is concentrated in odd degrees then X has the rational homotopy type of a wedge of spheres. Here we consider simply connected spaces A' with rational homotopy type of finite type. Let (Ly, d) be a rational minimal Lie model for X (Proposition 24.1) then d restricts to zero in V and so sV - H+(X;Q), by B4.3). Since H+(X;Q) is concentrated in odd degrees, V = Veven and thus Ly is concentrated in even degrees too. For degree reasons, then, the differential in Ly is zero. Now the Lie algebra isomorphism B4.2) shows that ?,(??') <g)Q is a free graded Lie algebra, and the conclusion follows from Theorem 24.5. ? (d) Lie models for adjunction spaces. Consider an adjunction space where: (i) X is simply connected with rational homology of finite type, (ii) f={fa:(Sn",*) -> (A>0)}and (iii) the cells Dn° + l are all of dimension > 2, with finitely many in any given dimension. Suppose that m : C*(Lv,d) —> APL(X) is a free Lie model for ?\ We shall construct a free Lie model for Y. In fact, in §24(b) we constructed the isomorphism tl : sH(Lv) ? ?»(?) <g> k
Lie Models 329 Thus the classes [fa] G ???(?) determine classes s[za] = T~l[fa] e sH(Lv) represented by cycles za G Lv- Let IV be a graded vector space with basis {wa} and degwa = na. Then we can extend L\/ to a chain Lie algebra Lv$h·' = L(V © W) by defining dwa = za. Theorem 24.7 The chain Lie algebra (hv@w,d) is a Lie model for Y. proof: Decompose C*(Ly,<2) as the tensor product of a minimal Sullivan al- gebra (AVx,d) and a contractible Sullivan algebra. This exhibits (AVx,d) as a minimal Sullivan model for A". Let U be a graded vector space with basis {ua} and degua = na + 1. Then it is shown in Proposition 13.12 that a commutative model .4 = (AVX © U, dA) for Y is given by setting: • AVx is a sub algebra. • U'A+ =0, and dA(U) =0. • dAv = dv + ?(?; [fa])ua , ? G Vx , a where we have already identified ?*(?') <S> Jk with Hom(Vx,]k) as in Theo- rem 15.11. (In fact, the proof in 13.12 is for a single cell, so that there U = 3cu. but the argument in the general case is identical.) We proceed now in four steps. Step (i): Description of Ca ¦ Let (C,dc) be the dgc dual to (AVx,d), so that C(\vx,d) = {Hs~lQ,dc)- Denote Hom(i7, Je) by U*, with dual basis u*a given by {u$,u*a) = ???. Then the graded coalgebra dual to .4 is just C ®U*, and the elements of U* are primitive: Ku*a = 0- The differential, ?, in Hom(.4, A·) reduces to dc in C and satisfies 5u*a = (— l)des3iQcQ, where ca G C is the primitive cycle given by (v,ca) = (v;[fa}) and (A*2Vx,ca) = 0 . Thus CA = (Hs-^Qs-^^^a) with dA restricting to dc in L(s-1C) and Step (it): Identification of the za. The cycles ca G C = Hom(AVjv, A·) restrict to the linear functions ca G Hom(Vx,k) which we have identified with [fa] G ??? (?) ® Jk. On the other hand, the decomposition of C*(Ly,d) as (AVx,d) <g> E. with ? contractible, defines a surjective dga quasi-isomorphism which is left inverse to the inclusion. The linear part. Qo, of ? therefore dualizes to an inclusion, ?0 : (Eom(Vx,k),0) —» s(Lv,d) ,
330 24 Lie models for topological spaces and CW complexes and ?(??) is inverse to the isomorphism r. Thus T~l[fa] = H(uo)(ca) = [oocq] and we may take Step (iii): . The dgl quasi-isomorphism CA —> (L(^ ? W),d). The quasi-isomorphism ? of Step (ii) dualizes to a dgc quasi-isomorphism ? : (C,dc) —> C*(Ly,d) and hence defines the dgl quasi-isomorph'ism ? : C(C,dc) -^H ?(C.(W)) A (Lv,d) , where ? is the quasi-isomorphism of Theorem 22.9. Moreover C(C,dc) = L{s~1O) and ?C~1??) = s~1luoc0i = za, as follows from the definition of -?. Extend ? to a morphism of graded Lie algebras by requiring ??^ = (~l)desCawa. By Step (i), ? a is a dgl morphism from ?.4 to (L(F Q\V),d), as defined before the statement of the theorem. Since ? is a quasi-isomorphism, so is its linear part: s~xC —> V, as we showed in Proposition 22.12. By construction, the linear part of ?4 sends s~lU* isomor- phically to W. Hence the linear part of ?,4 is a quasi-isomorphism, and a sec- ond application of Proposition 22.12 establishes ?,4 as a dgl quasi-isomorphism, Step (iv): (L(V ? W),d) is a Lie model for Y. Since (A, dA) is a commutative model for Y the quasi-isomorphism C*(CA) -^> (A, d) identifies C*(?A) as a Sullivan model for Y and (?A,d) as a Lie model for y. Since (?A,d) ~ (L(V ? W),d) the latter is a Lie model for Y as well. ? Remark In the setup at the start of §24(d) let j : X —*· Y denote the inclusion. It follows from the Remark after Proposition 13.12 that the surjection (A,dA) —> (AVx,d) is connected by a chain of commutative dga quasi-isomorphisms to Apl(j). A tedious chase through the proof of Theo- rem 24.7 will therefore establish a homotopy commutative diagram C*(Lv) « C*(A) C- APL(X) . APL(Y) , where the left hand vertical arrow is the model with which we started and ? (Lv,d) —> (LvQ\v,d) is the inclusion.
Lie Models 331 (e) CW complexes and chain Lie algebras. Suppose X is a connected CW complex with no 1-cells and finitely many cells in each dimension. Then we can use Theorem 24.7 to construct a Lie model (Ly, d) for X such that: 'each (Ly<n , d) is a Lie model for Xn, a basis {va} of Vn corresponds to the (n + l)-cells D^+l of X, and s[dva] G sH(Lv<n ) corresponds to the class [fa] of the attaching map fa : S? —> Xn under the isomorphism tl : sH (Lv<n ) ? ??,(???) ® Q of §24(b). We say that (Ly, d) is a cellular Lie model for X. Conversely, let (Ly ,d) be a connected free chain Lie algebra of finite type defined over Q. Then we can use Theorem 24-7 to construct a CW complex X for which (Lv,d) is a cellular Lie model. Indeed, suppose the ?-skeleton, Xn is constructed and (Ly<n_,, d) is identified as a Lie model for Xn. Then the isomorphism 77, identifies Choose a basis {va} of Vn so the classes s[dva] correspond to elements in irn(Xn); i.e. are represented by continuous maps fa : Sn —> Xn, and set Xn+1 = XnU{fa} Now Theorem 24.7 identifies (Ly<n ,d) as a Lie model for Xn+\. This provides an inductive construction of X. By the Remark following The- orem 24.7 there is a homotopy commutative diagram of commutative dga's, - - - - C (Ly<n , d) ' C"{X) C* (Ly<„ , d) - - - I I -\ \- ¦¦¦ - APL(Xn) APL(Xn+1) - Since C*(Ly,d) —> C* (Ly<n ,d) is an isomorphism in degrees < ? + 1, it is straightforward to construct a quasi-isomorphism C*(Ly ,d) ^* Apl(X). Thus (Ly,d) is a cellular model for X. (f) Examples. Example 1 (Ly, 0) is the minimal Lie model of a wedge of spheres. In fact Theorem 24.7 asserts that for any V = {Vi}{>1 of finite type with basis RJ, (Ly,0) is a Lie model of the space X = \JaSn°+1 = pt\Jf ( \J Dn°+1 ), V Of / deguQ = na. Note that this also follows from §24(c). ? Example 2 The free product of Lie models is a Lie model for a wedge, \JaXa. Suppose Xa are simply connected spaces and X = \JaXa has rational homol- ogy of finite type. Let (LV(a), da) be a Lie model for Xa.
332 24 Lie models for topological spaces and CW complexes Recall the free product defined in §21(c) and note that \J (Ly(a),da) is the free dgl L^^.d , in which d restricts to da in each V(a). Simplify the notation (LV(a),da) to (La,da)- Then C*(La,da) is a Sullivan model for Xa. Now the Example of §12(c) exhibits the fibre product ? (C*(La,da)) as a Ik a commutative model for X. On the other hand the inclusions (La,da) —> {J (La, da) induce a dga morphism Of C*(u (?*> da)) -> ? ik (C*(La, da)) - V or Ja We show that this is a quasi-isomorphism, thereby exhibiting JJ (La,da) as a Q Lie model for \/aXa. To see that this is a quasi-isomorphism note that it is the dual of the horizontal arrow in in which the slant arrows are the quasi-isomorphisms of Proposition 22.8. Finally, observe that the homology of a free product is the free product of the homologies, so that ?. (?\/???) ® Q = h(u U'(a),da) = U ??*?) · \ ? J a Example 3 The direct sum of Lie models is a Lie model of the product. Again let (La,da) be Lie models for simply connected spaces Xa such that X = Y\ Xa has rational homology of finite type. Then only finitely many Xa a will have rational homology degrees less than any given ? and so we may suppose only finitely many La have elements in degrees < n. It follows that C* (g)(La,da)) =CaC*(La,da) ..
Lie Models 333 which exhibits 0(?a,da) as a Lie model for X (Example 2, §12(a)). ? Of Example 4 A Lie model for S3 V S3 U(ct>(ct^]iv]vv. D8. Let ?,? G ti3(S3 V S3) be the elements represented by S3 and S3 respectively. Then (cf. Example 1) a Lie model for S3 V S3 is just (L(v,w),Q) with degu = degw = 2 and u,u> corresponding to ?,?. Moreover, the isomorphism 7i\,(S3 V S3) ®Q s sL(u, u>) identifies [a, [a, ?]w]w with 5 [?. [?, ??]], as is shown in §24(b). Hence by Theorem 24.7, (L(u,u;)u).du = [u[u,u>]]) is ? Lie model for S3 V S3. U[Q>[QKj,r] D8. ? Example 5 Lie models for CP°° and CPn. The space CP°° is a K(Z, 2) and so its minimal Sullivan model is A = (?(?). 0) with dega = 2 — cf. §15(b), Example 2. Thus CA is a Lie model for CP00. This Lie model has the form L(i»i ,??,??, ¦ ¦ ¦ ) with ?,· the desuspended dual of a!. (Thus degvi = 2i — 1). In the coalgebra C dual to .4 let c; be the element dual to a*. Then Ac^ = ^Z ci ® ci· Thus the formula in §22(e) shows that the i+j=k differential in CA is given by dVk = 2 2^ ^'^J · Since H(CA,d) = 7r»(CP°°) ® Q it follows that , ? = 2 Hn(CA,d) = 0 , otherwise, a fact that may not be immediately obvious from the formula for d. Notice that the construction of a CW-complex for this Lie model (§24(e)) has one cell in each even degree 2n and no cells of odd degree. The 2n-skeleton has the rational homotopy type of CPn, with Lie model C(v\,... ,vn) and the same differential. In this model ? [vhvj] is a cycle whose suspended homology class in i+j=n+l 7T2n+i (CPn) ® Q is not a Whitehead product. ? Example 6 A Lie model for (CP2 V S3) U(Q)?]H, D8. As we saw in Example 5·, a Lie model for CP2 is just (L(ux ,vo),d) with dvo = |[ui,ui]. Thus [v\,vo\ is a cycle and the homology class of s[ui,uo] corresponds to a class a G tt5(CP2) <g> Q. Let ? G ^(S3) be the fundamental class. Then as in Example 4, Theorem 24.7 shows that ,v-2,w,u); dvo = -[vi,v2}, dw - 0, du =
334 24 Lie models for topological spaces and CW complexes is a Lie model for (CP'2 V S3) U[a,?]w D8. ? Example 7 Coformal spaces. A simply connected topological space X with rational homology of finite type is called coformal (sometimes ?-formal in the literature) if it has a Lie model (L, 0) with zero differential. In this case L is the homotopy Lie algebra Lx (§21(d), (e)); i.e., the rational homotopy type of X can be formally deduced from its homotopy Lie algebra. (Note the parallel with formal spaces as defined in §12(c).) If X is coformal with Lie model (L, 0) then C*(L, 0) is a Sullivan model for X of the form (AVx,di) with d\ : ?? —> ?.2\'?. Conversely any Sullivan model of this form can be written as C*(L, 0) — cf. §23(a): X is coformal if and only if it has a 'purely quadratic ' minimal Sullivan model. Finally, recall from Example 1 in §23(a) that H(C*(L,0)) = ExtUL(lk,lk). But ULX = H*(nX;]h), as we saw in Theorem 21.5. Thus X coformal => H*(X; k) = ExtH^Qx.k) (k, k). ? Example 8 Minimal free chain Lie algebras. Suppose (Lv, d) is a minimal connected free chain Lie algebra of finite type, defined over Q and let X be a CW complex for which (Ly ,d) is a cellular Lie model (§24(e)). Then • sV ? Q = H*(X; Q), by the last comment in §24(b). • A basis of V is in 1-1 correspondence with the cells of positive dimension of X, which also form a basis of the reduced cellular chain complex of X. The conclusion is that the differential in the rational cellular chain complex of X must be zero, and hence the differential in the integral cellular chain complex is zero too. In other words: H*(X;Z) is a free Z-module with basis corresponding to a basis of Q® sV. ? (g) Lie model for a homotopy fibre. Let 0 —»¦ (/, d7) —»¦ (L, dL) A> (A-, dK) —». 0 be a short exact sequence of differential graded Lie algebras (connected and of finite type). By taking cochain algebras (see Section 23(a)), we obtain a relative Sullivan algebra (see Section 14) C*{K,dK) ^4 C*(L,dL) -^C*(I,dj) . Proposition 24.8 When ? is a Lie representative for a continuous map f : X —> Y between simply connected spaces, then (I, dj) is a Lie model for the homotopy fibre of f.
Lie Models 335 proof: By hypothesis there is a homotopy commutative diagram C'(K,dK) ^ APL(Y) C'(p) C*(L,dL) 22—+ APL{X) . B4.9) Now let H be the homotopy from ???€*(?) to ApL{f)my. Then in the com- mutative diagram C*{K,dK) —2— APL(X)®A(t,dt) C'(p) C*{L,dL) we may lift ?? ? through id ®?o because C*{p) is the inclusion of a relative Sulli- van algebra (Proposition 14.4). This provides a homotopy ??? ~ m'x such that m'xC*(p) = ApL(f)my. In other words we may suppose that B4.9) commutes exactly. Now Proposition 15.5 identifies C*(I, dj) as a Sullivan model for the homotopy fibre, F, of /, and hence by definition identifies (/, dj) as a Lie model for F. ? Exercises 1. Determine the minimal Lie models of the spaces: CP"#CP" , S3 x S3 x S3 , CP4/S2 , CP5/S'2 . 2. Show that if X is a simply connected CW complex and Ht(flX; Q) is a tensor algebra, then X has the rational homotopy type of a wedge of spheres. 3. Let (AV, d) -> (A, 0) be a Sullivan minimal model with V1 = 0, and let ? be of even degree and ? : (A <g> ??,?) —>· ? = (A ® Ax/1,0) a projection whose kernel is the ideal / C A <S> A+x. a) Prove that there is a quasi-isomorphism (A®Z, 0) —>· (^<8>?(x) ??, D) where ? = (kerd ? ?) <S> Ax. b) Suppose that the continuous map f : X -*¦ Y admits ? as commutative model, and let F be the homotopy fibre of /. Prove that F admits a model of the form (.4 © ?) ?(??,?) E(V) and thus that F has the rational homotopy type of a wedge of spheres. c) Prove that there exists a short exact sequence of graded Lie algebras 0 _> L(V) -> ?,(??-) ? q ?'^/] ?,(??) ® 0 -> 0.
336 24 Lie models for topological spaces and CW complexes 4. Let H — A(xi,... ;Xn)/I, where the rr;'s have even degree and where the ideal / is generated by monomials. Prove that H = ((··· {/\X\)/I\ <g> ??2)/-^2 ® - · - ® Axn)/In , where each Ip is an ideal in A{xi,X2,- ¦ ¦ ,xp) contained in the ideal generated by xp. Let X be a formal space such that H*(X;Ql·} = H. Deduce from exercise 3 that there is a finite sequence of extensions 0-»L(Vi) ->?i -» (??,... ,xn) ->0 0 ->L(V2) -> L-2 ->Li -> 0 0 -» L(Fn) -> ?.(??') ® Q -> Ln_! -» 0 s where (xi,... , a;n) denotes the abelian Lie algebra on the variables ? ?,??,- ¦ ¦ :xn- 5. Let X and y be simply connected CW complexes of finite type. Prove that Lxvy — LxlJLy. Suppose that ? G (Lx)even and y G (Ly)even. Prove that L(x,y) C LXvy.
25 Chain Lie algebras and topological groups In this section the ground ring is Q. In this section we pass from algebra to CW complexes using the spatial realiza- tion functor | | of §17(c). It will often be convenient to abbreviate the notation \A,d\ simply to \A\. We shall also abbreviate \C*(L,cLl)\ simply to \C"L\. Let {L,cLl) be an arbitrary connected chain Lie algebra of finite type. Our principal objective is to • Convert the universal enveloping algebra {UL,di) to a topological group \TL\. • Construct a contractible CW complex \C*(L;FL)\ and a principal fibre bun- \C*(L;FL)\ -^ \C*L\ , \C*{L;TL)\ x \TL\ -> \C*{L;TL)\ , thereby identifying \C*L\ as a classifying space for \TL\. In particular, if (L, ci/,) is a Lie model for a simply connected topological space X then we will deduce a chain of rational equivalences of topological monoids from ?? to |??|. This section is divided into the following topics (a) The topological group \TL\. (b) The principal fibre bundle \C*{L;TL)\ A- \C*L\. (c) \TL\ as a model for the topological monoid, ?.?. (d) Morphisms of chain Lie algebras and the holonomy action. Throughout we shall work with the spatial realization functor (§17(c)) from commutative cochain algebras to CW complexes. We recall that it converts the tensor product of cochain algebras to the topological product of spaces (Example of§17(c)). (a) The topological group, \TL\. Recall that the universal enveloping algebra, (UL, d) is a dg Hopf algebra (§21) which is cocommutative as a coalgebra. Thus its dual (TL, ??) = Horn {UL, Q) is also a dg Hopf algebra with commutative multiplication ?? and comultiplication Af dual, respectively, to the comultiplication and multiplication in UL. Since TL is commutative, both ?? and ?? are dga morphisms. Proposition 25.1 The CW complex \TL\ is a topological group, with multipli- cation |??| and topological diagonal \??\.
338 25 Chain Lie algebras and topological groups proof: The coassociativity of ?? implies that |??| : \TL\ ? |??| -> |FL| is an associative product. The single 0-cell of |FL| is the augmentation e : TL —> Jc, and its compatibility with ?? implies that e G |FL| is a two-sided identity. To construct an inverse we first construct a distinguished involution ? in UL. Let ULopp be the dg Hopf algebra defined as follows: as a differential graded coalgebra, ULopp = UL. However the product in ULopp is given by a-b = (_l)dega deg6ouj where ba is the product in UL. Define a : L —*· ULopp by ax = —x. Then a[z,?/] = ax-ay — (— l)desx de%yay -ax, and so a extends to a dga morphism v.UL—y ULopp. We may regard ? as a self morphism of the dgc, UL. Clearly v2 = id, and so ? is an automorphism satisfying v(ab) = (—l)desa ae%bba. Observe next that the composite coincides with UL -^> k -^—y UL. In fact, if ? G L then ?? = mult(a; ® 1 — 1 <g> ?) = 0. Moreover if ?? = ??* ® a[ and ?6 = ?bj <S> b'j then ?{(?) = 22l±aibj{vb'j)va'i = Y^±ai((pb)i/a'i . Hence if <pb = 0 then <p(ab) = 0 and so ? = 0 in I7L+. Let v* = Hom(f,wfe) be the dual of v. It is an automorphism of the cochain algebra (FL,Si). Apply | | to the dual of the formula above for ? to see that \i/*\ is an inverse in |FL|, which is therefore a topological group. Finally, note that in any commutative cochain algebra (A, d) the multiplication ?A realizes to the topological diagonal in \A,d\, as follows immediately from the definitions. ? (b) The principal fibre bundle, \C*(L;FL)\ A- \C*L\. Recall the acyclic construction C*(L; UL) considered in §22(c). It is, in partic- ular, a cocommutative differential graded coalgebra of the form (A.sL <g> UL, d). Moreover, as we remarked in §22(c), the projection ? : (AsL<S)UL,d) —> (AsL, d) and the right module action a : (AsL <g> UL, d) <g> (UL, d) —> (AsL <g> UL, d) are both dgc morphisms. Apply the realization functor to the cochain algebra C*(L; TL) dual to (AsL® UL,d), to construct the CW complex |C*(L,-FL)|. (Note that the differential in C*(L;TL) = (C*(L) (&TL,d) is not the tensor product differential, and so |C*(L;FL)| is not the product of \C*L\ and |FL|!). The duals q* and a* of the morphisms above realize to continuous.maps \C*(L;TL)\ ^h \C*L\ and \C*(L;TL)\ ? \TL\ —*· \C*(L;TL)\ .
Lie Models 339 Proposition 25.2 The construction above is a principal \FL\-fibre bundle whose total space. |C*(L,-FL)|, is a contractible CW complex. proof: Step 1: C*(L) is a Sullivan algebra and the inclusion ofC*(L) in C*(L;YL) is a relative Sullivan algebra. The first assertion is Lemma 23.1. For the second recall the isomorphism 7 : AL -^ UL of Proposition 21.2. It is obvious that 7 preserves differentials and the comultiplication. Thus, dually, (FL,??) = {AL*,—d*L) as graded algebras, L* denoting Hom(L, <Q) and d*L the dual of cLl- It now follows exactly as in Proposition 23.1 that C*{L;TL) = (C*(L) <8>AL*,d) is a relative Sullivan algebra with respect to the inclusion of C*(L). Step 2: \C*(L;FL)\ is a contractible CW complex. By Step 1, C*{L;TL) is itself a Sullivan algebra of the form (AV,d) with V = (sL <g> 1)* 6A® L)*. The linear part of the differential, do : V —*· V, is dual to the restriction of the differential in C,(L; UL) to sL ® 1 ? 1 <g> L. Since d(sx <S> 1) = -sdLx <g> 1 + (-l)de6*+ii ® x and d{\ ® x) = 1 ® d/,? it follows that (sL ® 1) ? A <g> L) = (sL ® 1) © rf(sL ® 1). Dually, ? = U ? cfei/ with d0 : U A doi7. From §14(b) we deduce that C{L;TL) = A(U ® dU) with d : U A dC/. Now let {ua} be a basis of U. A dga morphism ? : A(U ? dU) —> (Apl)u is specified by the arbitrary choice of elements tpua G (ApL)„esUa. This identifies (A(U@dU)) as the extendable simplicial set UApll- Thus \C*{L;TL)\ = \A{U® a dU)\ is a contractible CW complex (Example 3, §17(a)). Step 3: pL : |C*(L,-FL)| —)· \C*L\ is a principal \TL\-fibre bundle. Recall that C"(L;TL) = (C*(L) ®TL,d). An ?-simplex of \C'L\ is a dga morphism ? : C*(L) —> (Api,)n. As in the proof of Proposition 17.9, the pullback of pl over ? has the form ?" X](Apc)n\ \{APL)n ®C*(L) (C*(L) ® TL, d) I . If we can exhibit this as ?? ? |FL|, compatibly with the projection on ?" and the action of |FL|, then Proposition 2.8 will apply and show that pi is a principal bundle. Denote {APL)n by A and (APL)n ®C-(L) {C*{L)®TL,d) by (A <g> TL,d). Note that the dual, a*, of the Z7L-action a induces an obvious morphism ? : (A <g> TL,d) -+ (A® TL,d) ® (FL,6L). Since ? {A) = Q, the argument of Lemma 14.8 with Q —> A replacing B-i —> ? shows that id a extends to an isomorphism V : (A <&FL,d) -=> (A, d) ® (FL,6l)- Combine this with the augmentation, e, of (TL, 6L) to define {??®?)? : {A®TL,d) —)· (A,d). The composite <p:{A® TL, d)A(A® ??, d) ® {TL, ol) i!^!^!i (A, j) 0 (??> Sl)
340 25 Chain Lie algebras and topological groups reduces to the identity in .4 and satisfies ?A<?>?) — 1<?>? € ?+®??,, ? G TL, Hence ? is an isomorphism. It clearly converts ? into id ®??- Thus \?\ : |.4| ? |FL| -=» |,4 ®FL.d\ compatibly with the projection on |.4| and the action of |??|. ? Notice that Proposition 25.2 identifies \C*L\ as weakly homotopy equivalent to the classifying space (§2(e)), B^l\- In fact (cf. §2(e)) the principal bundle ? : \C*(L;TL)\ —> \C*L\ pulls back from the universal bundle E^l[ via a map / : \C'L\ —> B{rL{. Since it, (\C*(L;TL)\) = 0 = ??. (E[rL\), / is a weak homotopy equivalence. (c) \TL\ as a model for the topological monoid, ?.?. Suppose now that (L,dL) is a Lie model for a simply connected CW complex. X. In particular we have a dga quasi-isomorphism which in turn (Theorem 17.12) defines a rationalization ?? : X —> \C*L\. Thus ?/?.? : ?? —> Q\C*L\ is a morphism of topological monoids and also induces an isomorphism of rational homotopy and homology (Theorem 8.6). We call such a morphism a rational monoid equivalence. On the other hand suppose G is any topological group and ? -^> ? is any principal G-bundle with 7r,(Z) = 0. Then from Proposition 2.10 we obtain weak equivalences of topological monoids GP-GxzPZAfi5, a<—i(a,u/)i—>p°w. and corresponding weak equivalences of Serre fibrations PZ PB Thus here we have the chain of rational monoid equivalences |??| <=- |??| X|c-(L;rL)i P\C*{L-TL)\ A il\C"L\ ?^ B5.3) thereby exhibiting |FL| as a 'topological model' for ??'". We also have the corresponding (rational) equivalences of Serre fibrations \C'(L\rL)\ P\C'(L;rL)\ \C'L\ B5.4)
Lie Models 341 (d) Morphisms of chain Lie algebras and the holonomy action. Suppose ? : (L, di) —>· (E, d?) is a morphism of connected chain Lie algebras of finite type. Then U<p : (UL,dL) —> (UE,d?) is a morphism of dg Hopf algebras. This is, in particular, a representation of {L,di) in (UE,d?), and therefore determines the chain complex (C*(L;UE),d) as described in §22(b). This example has many of the additional properties of C*(L; UL). It, too, is a cocommutative differential graded coalgebra and a right (UE,d?)-module, with the module action a dgc morphism. Moreover, C»(</>) : Cm(L,di) —> ?*(?,??) extends to the morphism of dgc's and ?7i?-modules: C.(</>; id) : C.(?; UE) —> C.(E; UE) . Now apply the spatial realization functor to the commutative cochain algebras (and morphisms of same) dual to these constructions. This gives the commuta- tive diagram of continuous maps \C*(L;FE)\ |C*(y;id)| - \C*{E;TE)\ \C*L\ _ . \C'E\ in which: ? is a principal |F?7J—fibre bundle (exactly as in Proposition 25.2) and \?*(?; id)\ is a map of |r\E|-bundles. In particular, this exhibits ? as the pullback of pe via |C"V|- On the other hand, the continuous map \C*ip\ determines a holonomy fibra- tion as described in §2(c), which is just the pullback via \?*?\ of the Q\C*E\ fibration P\CE\ —> \CE\. We may therefore apply Proposition 2.11 to obtain equivariant weak equivalences of Serre fibrations \C*(L;rE)\ = ? -Z-~ \C*L\ X|C-E| P\C'E\ , |??| ?- ? ^ U\C*E\ \C'L\ B5.5) connecting the principal \TE\-fibre bundle ? with the holonomy fibration for \cy\. Finally, suppose / : X —> Y is a continuous map between simply connected CW complexes with rational homology of finite type. Assume further that (L,di,) and (?,??) are Lie models for Ar and for Y and that ? : (L,di) —>
342 25 Chain Lie algebras and topological groups makes the diagram APL(Y) APL(X) B5.6) C\E,dE) ?'(?) C*(L,dL) homotopy commute in the sense of §12(b). In analogy with §25(c) we observe that the principal \TE\-fibre bundle ? : \C*(L;TE)\ —> \C*L\ is equivariantly rationally equivalent to the holonomy fibration ? ? ? PY —> ?. In fact, the diagram B5.6) produces the homotopy commutative diagram X Y hx hY \C*L\ \C*E\ (cf. §17(d)) in which hx and hy are rationalizations. Regard the homotopy as a map ? from X to the space of Moore paths in \C*E\ of length 1, such that (??)@) = \?*?\???(?) and (??)A) = hyf(x). Then an equivariant rational equivalence h:XxYPY-+\CmL\x]C'E\P\CmE\ B5.7) is given by h(x,w) — (???,?(?) *hy o»). Combined with B?.?) above this provides the desired equivariant rational equivalence. ?
26 The dg Hopf algebra Cm(QX) In this section the ground ring is Q; as usual we simplify notation by writing C.(-) and H.(-) for C.(-;Q) and H.(—,Q). Suppose given a free Lie model (Ly,d) for a simply connected topological space X with rational homology of finite type. A natural isomorphism of graded Hopf algebras, H(ULv,d)^H,(ClX) B6.1) is then described as follows. First, H{Uhv) = UH(LV), by Theorem 21.7. Next, H(Ly) is identified with the homotopy Lie algebra L\ by B4.2). Finally, the graded Hopf algebras ULx and ?*(??) are identified by the Milnor-Moore theorem 21.5. The main objective of this section is to show (Theorem 26.5) that the isomor- phism B6.1) is induced from a chain algebra quasi-isomorphism Q:(ULv,d) -=>C.(ilA") which commutes with the comultiplications up to dga homotopy (to be defined in §26(a)). This is a result of Majewski [119]. Its significance is due to a theorem of Anick [11] who showed directly the existence of a unique quasi-isomorphism class of free chain Lie algebras (L, d) admitting such a quasi-isomorphism. However, these were potentially different from the Lie models constructed via Sullivan's functor Apl as described here. Majewski's result shows that they coincide. Our proof of Majewski''s theorem is different from his, which takes a more abstract approach (permitting him to prove that other forms of introducing Lie models also produce the same answer). The idea of the proof here is as follows. In §25(c) we connected UX to a certain topological group, |FLv|, by rational monoid equivalences. This'reduces the problem to constructing an appropriate quasi-isomorphism (ULv,d) -=» C,|rLy|. For this we use the integration map /.of§17(f). Finally, at the end of this section, we consider continuous maps / : X —> Y and prove an analogous theorem about the holonomy fibration. This section is divided into the following topics: (a) Dga homotopy. (b) The dg Hopf algebra C*(QX) and the statement of the theorem. (c) The chain algebra quasi-isomorphism ? : (t/Ly,d) -=> C^IXy,^). (d) The proof of Theorem 26.5. In carrying out these steps we shall make frequent use of the Alexander- Whitney and Eilenberg-Ziller chain equivalences (-) and EZ : <?»^
344 26 The dg Hopf algebra C.{ilX) defined in §4(b) for topological spaces. As observed in §17(f), these equivalences are defined for all simplicial sets, and we shall use them as such. In particular we shall quote the important properties D.4)-D.10) as if they had been stated for simplicial sets, since they hold in thus wider context. (a) Dga homotopy. Suppose given a chain algebra of the form (TV, d). Given two graded algebra morphisms ?,?' : TV —> A we say a linear map ? : TV —> A is a (?,?1)- derivation if <&(xy) = ?? Vy + (-l)degxdeg<IV:z;-3>y , x,yeTV . Any linear map V —> A extends uniquely to a (?, <?>')-derivation. Two chain algebra morphisms ?. ?' : (TV,d) —> (A,d) are dga homotopic if ? — ?' = ?? + ?? for some (?, ?')-derivation ?, in which case we write ? ~ ?'. The derivation ? is called a dga homotopy. Note that both sides of such an equation are themselves (?, <//)-derivations. Thus in constructing a dga homotopy it is sufficient to verify that this equation holds in V. Dga homotopy appears to be quite distinct from Sullivan's homotopy described in §12(b). In fact the two are closely related, but we shall not pursue this here. The following lifting lemma parallels all the others. Lemma 26.2 Suppose ? : (?, d) -=> (.4, d) is a chain algebra quasi-isomorphism Then (i) For any chain algebra morphism ? : (TV, d) —> (A, d) there is a morphism ? : (TV,d) —> (B,d) such that ? ~ ??. (ii) If ?,?' : (TV,d) —> (B,d) are morphisms satisfying ?? ~ ??' then ? ~ ?'. proof: (i) We construct ? and a homotopy ? : ? ~ ?? by induction. Suppose they are defined in T(V<n) and choose a basis {va} for Vn. Then the cycles ???? satisfy ?(????) = ???? — ????? = ?(??? — ????). Since 77 is a quasi- isomorphism there are elements ba G ? and aa G A such that dba = ???? and ?? = ???? — ???? + daa. Extend ? and ? to Vn by setting ??? = ba and ??? — uq- Then ? extends uniquely to an algebra morphism from TV<n and ?? = ?? because this relation holds in V<n. Similar ? extends uniquely to a (?, rcT/O-derivation from TV<n to ? and ? — ?? = ?? + ?? because this relation holds in V<n- (ii) By hypothesis there is an (??, ??')-derivation ? such that ?? — ??' = B?+?B, and we seek to construct a (?, VO-derivation ? such that ?—?' = d* + *i!d. Assume ? constructed in V<n so that ?? — ? = dE — Ed for some
Lie Models 345 (??, 77^')-derivation ?. Again let {va} be a basis of Vn. Then (?-?')?? - is a cycle za in B, and ??? = ???? - (?^ - ?)??? = ?(??? - Edva). Choose ba G 5 and aa ? A so that zQ = <2bQ and ?? = ??? - ?B?? + daa. Then extend ? and ? by setting #uQ = ba and ??? = aa. D Lemma 26.3 5uj9j[?ose ? : (TW,d) morphisms of chain algebras. (i) Ififo ~ ?? then ?(?0) = ? (??), (ii) Ififo ~ ?? then ??0 ~ ??! and ??? ~ ?\?. (iii) Dga homotopy is an equivalence relation. proof: (i) This is immediate from the equation ?? — ?\ = ?? + ??. (ii) If ? is a dga homotopy from ?? to ?? .then ?? is a dga homotopy from ??? to ??? and ?? is a dga homotopy from ??? to ?\?. (iii) Reflexivity: The zero map is a dga homotopy from ?0 to ?0. Symmetry: First construct a new dga (/, d) as follows. Let V and F" be graded vector spaces isomorphic with V and let sV be the suspension of V with suspension isomorphism ? ?—> sv. Put / = T(V ? F" ? sV) and let i', i" : TV —> I be the algebra inclusions extending the linear isomorphisms V s V',V". Let 5 : TV —->· / be the (i',i")-derivation determined by 5u = su. Finally, let d be the derivation of / determined by di'v = i'dv, di"v = i"dv and dsv = i'v — i"v — Sdv, ? ? V. Next, verify in sequence the equations di' = i'd, di" = i"d, i' — i" = dS + Sd and d2 = 0, noting that it is sufficient to check each on the appropriate generating subspace. In particular (/, d) is a chain algebra, and i' and i" are chain algebra morphisms. Now suppose ? is a dga homotopy from ?? to ?\. Define a dga morphism ? : (I,d) —>¦ (A,d) by ?,?' = ?0, ?" = ?? and Clsv = ??. We show that i" ~ I and conclude that ?? = Qi" ~ Qi' = ??· For this define a dga morphism ? : (I, d) —> (TV, d) by ??' = id = ??" and ??? = 0. If i' is a quasi-isomorphism then so is ? and hence Lemma 26.2 will give i" ~ i'. It remains to show that i' is a quasi-isomorphism. Let W = (i' — i")(V). Then / = T(V ©W ® sV). Moreover the ideal I generated by W and sV is preserved by d. Since / = TV ?? we need only show H(l) = 0. For this use multiplication in / to write I = TV' <g> (W ® sV) ® /. Define h : X —>¦ X by ,x = i'v-i"v,
346 26 The dg Hopf algebra C.(QX) Then a straightforward computation shows that hd + dh— id sends TV <g> W <8>I into TV ®sV <g> / and sends TV' <g> sy <g> / to zero. It follows that #(I) = 0 and i' is a quasi-isomorphism. Transitivity: Extend (I,d) to a chain algebra (J,d) as follows. Set J = T(V © V" © V" © sV © sV) where V" is a third copy of F and sV is a second copy of the suspension of V. Regard i', i" and 5 as maps into J, let i'" : TV —> J be the algebra inclusion extending the isomorphism V = V" and extend s : V -^ sV to an (i",i'")-derivation 5 : TV —>¦ J. Define d in J by requiring ?"'" = i'"d and dsu = ?"? - i'"v - ISdv, ? G I7· Now suppose </>0 ~ ?? ¦ (TV,d) —> (A,d) and ?? ~ ?-2 : (TV,d) —> (A,d) are dga homotopic by homotopies ? and ?'. Define a chain algebra morphism ? : (J,d) —> U»rf) by ?'?' = <^0, ?'*" = </>i, ?*'" = <^2, IIsu = ?? and ??? = ?'?. Define a chain algebra morphism ? : (J, d) —> (TV,d) by ??' = 7??" = ??'" = id and 7rsu = ttsu = 0, ? ? V. As in the proof of symmetry, ? is a quasi-isomorphism and so Lemma 26.2 implies that i' ~ i"'. It follows that </50 = ??' ~ II»'" = <y52- ? (b) The dg Hopf algebra C*(Q.X) and the statement of the theorem. Let (X, xq) be a based topological space, and recall (§2(b)) that UX is the topological monoid of Moore loops at xo· Multiplication UX ? UX —> UX is given by composition of loops and the identity is the constant loop c at xq. Now suppose G is any topological monoid with multiplication, ?, and identity, c. Then, as observed in the introduction to §16 for G = UX, y"aig = ?,(?) ? ? ? : Ct(G) ® C,(C?) —> C,(G) makes C»{G) into a chain algebra. On the other hand (§4(b)) the Alexander- Whitney diagonal Aalg = AW ? C.(Atop) : C,{G) -* C,(G) ® C.(C) makes C,(G) into a differential graded coalgebra, where Atop denotes the topo- logical diagonal sending y i-> (y. y). Proposition 26.4 The maps /iaig and Aaig mate C*(G) into a differential graded Hopf algebra with identity and augmentation given by h = Ct({c}) —> C.(G) and C,(G) —> C,(pi) = A · proof: This is a straightforward calculation using properties D.4)-D.9). In particular, the fact that Aaig is a morphism of chain algebras follows from the compatibility D.9) of AW and EZ. ?
Lie Models 347 On the other hand, suppose (Ly,d) is a free Lie model for a simply con- nected topological space X with rational homology of finite type (cf. §24). Then U{Lv,d) is a differential graded Hopf algebra of the form (TV,d), whose co- multiplication is the chain algebra morphism Av : (TV, d) —» (TV, d) <g> (TV, d) defined by Ayv = ? ® 1 + 1 ® ?, ? G V* (§21(c), (f)). Our main theorem reads Theorem 26.5 Let X be a simply connected topological space with rational homology of finite type. The choice of a free Lie model (Ly, d) for X determines a natural homotopy class of chain algebra quasi-isomorphisms e-.U(Lv,d) -^C.(QX) such that (? <8> ?)?? and Aaig ? ? are dga homotopic. Moreover ?(?) is the isomorphism B6.1). Remark Suppose (Ly, d) is a cellular Lie model (§24(c)) for a simply connected CW complex X with rational homology of finite type. The quasi-isomorphism of Theorem 26.5 has the form (TV,d) 4?(??) , because t/Ly = TV (§21(c)). Moreover V is free on a basis corresponding to the cells of A'. This is therefore a special case of the classical Adams-Hilton model [2]. This model, and the de Rham algebra of differential forms on a manifold, were the first uses of differential graded algebras as a tool in homotopy theory. ? (c) The chain algebra quasi-isomorphism ? : (ULv,d) -=> C,(FLy ,SL). Fix an arbitrary free connected chain Lie algebra (Ly, d) of finite type. In §25(a) we constructed a topological group |rLy,5jr| as follows. First, we let (TLv,Sl) be the commutative graded Hopf algebra dual to [/Ly. Thus the multiplication ??· and the comultiplication Ar are respectively dual to the co- multiplication Au and the multiplication ?? in [/Ly. Then we applied the spatial realization functor of §17(c) to obtain |FLy,i| = |(FLy,?i,)| with multi- plication |??| and topological diagonal \??·\ (cf. Proposition 25.1). Since (Ly,d) is fixed in this discussion we shall abbreviate the dg Hopf algebra (FLy,^z,) to ? so that our topological group is simply denoted by |?|. As observed in Proposition 26.4, C, |?| is a dg Hopf algebra with multiplication C»(|Ar|) oEZ and comultiplication AW ? C„(|/zrl)· However, included in C,\T\ is the sub dgc C»(T), as defined in §17(a). Since ? ? and AW are defined in the wider context of simplicial sets (§17(f)), C*(T) is a sub dg Hopf algebra of C,|r|, with multiplication /iaig = C(Ar) ° EZ and comultiplication Aa)g = Now recall from §17(f) the natural morphism of chain complexes /e : C*{AV, d) —> C(\v,d), dual to the integration operator § defined in §10(e). In the case of (FLy, 5l) it takes the form
348 since ? is the dual of ULv ¦ 26 The dg Hopf algebra ?.(??') Proposition 26.6 The map Jt : C*(T) —» {ULv ,d) is a chain algebra quasi- isomorphism. proof: Proposition 17.16 asserts that J9 is a quasi-isomorphism. Proposi- tion 17.17 asserts that C, (?) ® C, (?) /-®/-v ULV ® ULv commutes. But the multiplication |??| in |?| is the realization of the dual of the multiplication ?? in ULv. Thus for ?,? e (?), /.? - r = JmC*(&r)EZ(a ® r) = ??^??(? ® r) = /. ? · /, r . which is Recall now (§21(c)) that the universal enveloping algebra (C/Lv,d) is, as a chain algebra, the tensor algebra (TV, d). Thus we may apply Lemma 26.2 to the chain algebra quasi-isomorphism Jr : C»(r) —> (TV,d) to obtain a chain algebra quasi-isomorphism uniquely determined up to dga homotopy by the requirement that ft ? ? ~ id. Lemma 26.7 The quasi-isomorphism ? commutes with the comultiplications in ULv and in C»(F) up to dga homotopy. proof: We have to show that (? ® ?)?? ~ AW ? ?(?1??) ? ? . Now here ?1?? is just (??), where ?? is the dual of ?[/. Consider the (non- commutative) diagram
Lie Models 349 are dga morphisms, as follows easily from property D.7) and D.8) for EZ and from property D.9) for AW. Moreover (Proposition 4.10) these are quasi- isomorphisms, and AW ? EZ — id. In particular, AW ? C, (??) ? ? = Aaig ?? = AW ? EZ ? Aaig ? ? . Since AW is a chain algebra quasi-isomorphism we may apply Lemma 26.2(ii) to conclude that C»(/ir) ° ? ~ ? ? ? Aaig ? ?. Finally, note that the right hand square in the diagram above commutes by naturality. Thus Au ~ Au ? Jt ? ? ~ /? ? <7,(??) ? ? ~ /? ? ?? ? Aaig ? ? . The diagram in the proof of Proposition 26.6 allows us to replace J oEZ by Au ~ /, <8> /, ? Aaig ? ? . Thus, since ft ? ~ id, L ® /. ° o ® o ° At/ ~ At/ ~ /? ® /? ? Aaig ? (9 . But /t is a quasi-isomorphism. Apply Lemma 26.2 to obtain 1 ? Au ~ Aaig OO. D (d) The proof of Theorem 26.5. We are given a free Lie model (Ly, d) for a simply connected topological space X, which means that we have specified a cochain algebra quasi-isomorphism m:C*(Lv,d) ^ APL{X) . For simplicity of notation we shall denote (Lv,d) simply by L. Suppose first X is a CW complex. Recall that ? = (FL, ?) denotes the dg Hopf algebra Hom(UL,]k). Apply C, to the chain B5.3) of rational monoid equivalences to obtain the chain ?^ B6.8) of dg Hopf algebra quasi-isomorphisms. Next, recall that UL = TV, and use Lemma 26.2 to lift the quasi-isomorphism of §26(c) through this chain to obtain a chain algebra quasi-isomorphism ?: (UL,d) A
350 26 The dg Hopf algebra C.(iLY) It commutes up to dga homotopy with the comultiplications because ? does (Lemma 26.7) and because the quasi-isomorphisms in B6.8) commute with the comultiplications. Moreover ? is uniquely determined up to dga homotopy and hence, so is ? (Lemma 26.2). Since J is natural and the homotopy class of h ? is natural in X (Theorem 17.15) so is the homotopy class of ?. Now we have to show that ?(?) coincides with the isomorphism B6.1). We may clearly suppose X = \C*L\, h ? = id and m:C*L^> APL\C*L\ is one of the canonical homotopy class of morphisms (specified in §17(d)). Since H{UL) = UH{L) it is enough to show these isomorphisms agree on [v] for any cycle ? e L. Thus by naturality we are reduced to the case L = (L(u),0) is the free Lie algebra on a single generator of degree n. In this case the isomorphism B6.1) sends ? to the unique homology class corresponding to [sv] under the isomorphisms Hn (Q\C*L\) #- Hn+1 (P\C*L\, n\C*L\) ? Hn+l {\C*L\) . But H(Q) has the same effect (use the fibrations B5.4) and a simple calculation in C*{L;TL)). D Exercise Determine a chain algebra quasi-isomorphism ? : (TV, d) —> C» (fLY) when X is a wedge of spheres 5"', n* > 2. Deduce from this the Hopf algebra structure of Ht(QX;Q).
Part V Rational Lusternik Schnirelmann Category
27 Lusternik-Schnirelmann category In this section the ground ring is an arbitrary commutative ring k, unless oth- erwise specified. A subspace ? of a topological space X is contractible in X if the inclusion i : ? —y X is homotopic to a constant map ? —> xq. Definition The LS category of X, denoted cat A", is the least integer m (or oo) such that A' is the union of m + 1 open subsets ?/z, each contractible in X. Remark LS category was originally introduced in [116], where it was shown that cat X + 1 is a lower bound for the number of critical points of any smooth function on a closed manifold X. The original definition differs from the one above by 1; however the definition above has become standard in homotopy theory because, clearly, cat A = 0 if and only if A" is contractible. Note that it follows that catS" = 1, ? >0 . Moreover if xa G Xa has a neighbourhood contractible to xa in Xa then cat ( V Xa ) = max {cat Xa } \a J a as is immediate from the definition. This section is devoted to some of the basic geometric properties of LS cat- egory. (For other presentations the reader is referred to [96], [62] and [40].) For example, in §27(a) we shall prove that cat A' < d/r for an d-dimensional (r — l)-connected CW complex, and in §27(f) we shall see that if cohomol- ogy classes a; e H+{X;k) satisfy ?? U · · · U an ? 0, then cat A > n. Now CPn is a 1-connected CW complex of dimension 2n whose rational cohomol- ogy algebra has the form k[a)/an+l where ? G H2(CPn;Q). It follows that ? < catCP" < 2n/2; i.e. catCP" =n . An identical argument shows that the LS category of any simply connected symplectic manifold A" (Example 5, §12(e)) is half the dimension. However, our focus in this section is on a number of useful characterisations of the inequality cat A" < m, including: • deformation of the diagonal of Am+1 into the fat wedge (Whitehead). • existence of a cross-section of the mth Ganea fibration (Ganea). • retraction of A" from an m-cone (Ganea). • A V ?? is an m-cone (Cornea).
352 27 Lusternik-Schnirelmann category LS category is an invariant of homotopy type, but it is particularly difficult to compute. For this reason it is useful to approximate it by other invariants. Thus we introduce the cup-length, c(X;R) and Toomer's invariant, e(X;k), which are more algebraic in nature, as well as the geometric cone-length cl X which is the least m (or oo) such that X has the homotopy type of an m-cone. We shall show that c(A';A) <e(X;k) <catA <clA . In this section we shall make heavy use of cones and suspensions. Thus we recall (§ 1 (f)) that the cone on a topological space A' is the space CX = A" x I/X x {0}, and that the points (x, t) are usually denoted by tx, 0 < t < 1, with Ox the cone point. This identifies I = Ix {1} asa subspace of CX. If g : A —>· X is any continuous map the adjunction space A Us CA is called the cofibre of g. Next define the reduced cone on a based space (A, xo) to be the space CX = CX/Ixq. Then the suspension of X is the based space ? A = CA/A. If (A, x0) is well-based then (CA, Ix0) is an NDR pair, the quotient map CX —> ~CX is a homotopy equivalence (Corollary to Theorem 1.13) and (CA,.t0) and ?? are well-based. If, in addition, A is path connected a simple van Kampen argu- ment shows that ?? is simply connected. Furthermore the long exact homology sequence associated with the pair (CA, A) gives rise to natural isomorphisms ?{(??; Ik) A Hi-i(X; Jfe) , i > 2 . We shall use all these facts freely without further reference. This section is organized into the following topics: (a) LS category of spaces and maps. (b) Ganea's fibre-cofibre construction. (c) Ganea spaces and LS category. (d) Cone-length and LS category: Ganea's theorem. (e) Cone-length and LS category: Cornea's theorem. (f) Cup-length, c(A; Jfc) and Toomer's invariant, e(A; k). (a) LS category of spaces and maps. The definition of LS category extends from spaces to continuous maps as follows: Definition The LS category of a continuous map f : X —> Y, denoted cat/, is the least integer m (or oo) such that A is the union of m + 1 open sets Ui for which the restriction of / to each [7; is homotopic to a constant map Ui —> j/j.
Rational LS-Category 353 Note that cat ? = cat id ? so that this is a generalisation of the LS category of a space. We say a topological space X is a homotopy retract of a topological space Y if there are continuous maps X —> Y -^ X such that gf ~ id ?. Lemma 27.1 Suppose f : X —> Y is a continuous map. (i) If f ~ f then cat/' = cat/. (ii) If g :Y —>.Z is continuous then cat gf < min(catc?, cat/). (iii) If g (resp. f) is a homotopy equivalence then cat g f = cat/ (resp. cat gf = cat^j. fiw,) cat / < min (cat X, cat Y). (v) If X is a homotopy retract of Y then cat A" < catY. proof: (i) and (ii) are trivial consequences of the definitions. For (iii) suppose g' is a homotopy inverse for g. Then g' gf ~ / and so cat/ = c&tg'gf < cat gf < cat/. Similarly if/' is a homotopy inverse for / then cat g = cat gf f < cat gf < cat g. (iv) follows from (iii) applied to idy of and / ? id ? and (v) is immediate from (i), (ii) and (iv). D As an immediate consequence of Lemma 27.1 (v) we deduce Proposition 27.2 If X and Y have the same homotopy type then cat A' = cat ?. ? Lemma 27.3 For any continuous map g : A —> X, cat(A Ug CA) < cat X + 1 . proof: Let cat AT = m and put ? = [? ? {0}] e CA. Then A' Ug CA - {a} is an open subset of A Ug CA containing A as a deformation retract. Hence TO cat (X Ug CA — {a}) = m and A Ug CA — {?} = (J Ui with Ui open and con- o tractible in A Ug CA. Since CA — A is open and contractible in A" Ug CA it follows that cat(A Ug CA) < m + 1. ? Next consider a based topological space (A, x) and denote its (m + l)-fold product by Am+1 = Ax- · -x A. The/?? wedge, Tm+lX C Am+1 is the subspace given by Tm+1(A) = {(x0,.. .,xm) e A"m+1 | some x{ = x). The diagonal, Ax : X —> Xm+l, is the continuous map ?? : ? ?—)¦ (?, ?,..., ?). Definition The Whitehead category of X, denoted Wh cat A", is the least in- teger m (or oo) such that ?? is homotopic to a map / : X —> Tm+1(X).
354 27 Lusternik-Schnirelmann category Proposition 27.4 [158] Suppose (X, x) is path connected (i) If X is normal then Wheat.Y < catX. (ii) If ? is contained in a subspace U that is open and contractible in X then cat ?' < Wh cat X. proof: (i) Let m = cat X so that X = (J Ui with Ui open and contractible in ?". ? Because X is normal there are subspaces At C Oi C B{ C Ui with Ai, Bi closed and Oi open and X = \JAi, and there are continuous functions hi : X —> I such that hi I . ? 1 and hi \v „ ? 0. Let Hi : Ui y. I —> X be a homotopy from the inclusion of Ui to a constant map. Because X is path connected we may suppose Hi(—, 1) : Ui —> x. Define Ki-.X x I—>X by j ? , ? ? X - Ki(x,t) = j Then I<i(-, 1) : >li —> ? and so ? : ? ? / —> .Ym+1 , Uf(z, i) = (UfoU i)>..., Km(x, t)) is a homotopy from ? to a map X —» Tm+1(X). (ii) Let m = Whcat(.Y) and let K(x,t) = (K0(x, t),... ,Km(x,t)) be a homotopy from ? to a map / = (/0,.-.,/m) : A' —> Tm+1(X). Set [7, = f^l{U). Then X = (J^i· Moreover ?"; is a homotopy from the inclusion of ? Ui to the map /; : Ui —> U. Since U is contractible in X it follows that Ui is contractible in X too. D Corollary //A' is a path connected CW complex then cat A" = Wh cat X . proof: By Proposition 1.1 (iv) and (?), ? satisfies the two additional hypothe- ses of Proposition 27.4. G Recall next that a topological space X is c-connected if Ki(X) = 0, 0 < i < q. Proposition 27.5 Suppose X is an (r — \)-connected CW complex of dimen- sion d (some r > I). Then cat.Y < d/r .
Rational LS-Category 355 proof: It follows from the construction of Theorem 1,4 that there is a weak homotopy equivalence g : Y -=» X where Y is a CW complex whose (r — 1)- skeleton is a single 0-cell, yo· By Corollary 1.7 this is a homotopy equivalence, and hence has a homotopy inverse / : A" -^ Y. Let m be the integer part of d/r, and recall from Example 3, §l(a) that ym+1 is a CW complex whose fc-cells are the products ?>?° ? · · · ? D^ of cells in Y with ?&? = k. Since d < (m + l)r, the (i-skeleton of Ym+1 Ts contained in Tm+1(Y). In particular, a cellular approximation (Theorem 1.2) of the map ?/ = (/,...,/): A' —» Ym+1 is a map h : X —» rm+1(F) such that ?/ ~ h. Finally, since gf ~ id, g ? · · · ? g ? ?/ ~ ??, ?? the diagonal of A'. Thus ?? ~ (g ? ¦ ¦ ¦ ? g)h : ? —> Tm+1(X). By the Corollary to Proposition 27.4, catX <m. D (b) Ganea's fibre-cofibre construction. Suppose (A", xo) is a based space and let j : F —> ? be the inclusion of the fibre at xq of a fibration p:E—>X. Extend ? to the continuous map (cf. §l(f)) PC : ? Uj CF —> A" by setting pc(CF) = x0. Then (§2(c)) convert pc to the fibration p : ?/ —> A defined by ??' = (E Uj CF) ? ? Ai A and ?'B,7) = -?(?), ? the length of the path 7. According to Proposition 2.5, a homotopy equivalence A : ? Uj CF ^=> 2?' is given by Az = (z,cPc2), cx denoting the path of length zero at x. Note that p'X = pc- Precomposing A with the inclusion of ? gives the commutative diagram which is called Ganea's fibre-cofibre construction. It is obviously functorial with respect to maps of fibrations. Now consider the special case that ? : ? —> X is itself obtained by converting a continuous map / : ? —> X into a fibration, so that E = Z xx MX and F = ? xx PX .
356 27 Lusternik-Schnirelmann category Thus a holonomy action is defined on F as well as on the fibre F' of p'. Recall further from §l(f) that the join F * QX is defined by F * ?? = (Fx CQX) UFxnX (CF ? ??) . Right multiplication in ?? and the holonomy action in F define a diagonal action of QX on F * ??((??,?) · g = (ug,vg)). More generally an ??"-space is a topological space equipped with a right QX action and a map of ??-spaces is a continuous map that commutes with the actions. Proposition 27.6 (Ganea [63]) (i) The fibre F' of p' at x0 has the homotopy type of F * QX. (ii) If ? is obtained as above by converting a map f to a fibration then there is a homotopy equivalence F' -^ F*QX which is also a map of QX-spaces. proof: (i) By construction, F' = (E Uj- CF) xx PX = (Exx PX) U,- {CF ? ??) , i denoting the inclusion of Fx QX in ? ? ? PX. Now in the proof of Proposition 2,5 (with different notation) we constructed a continuous map ? : ? ?x PX ? [0, ??) —> ? such that ??(?, ¦y, t) = j(t) and ?(?,7,0) = ?. For 7 € QX let 7 be the loop of length ?7 given by 7(?) = ?{??—t). Define ? : F ? ?? ? I —> ? ?? ? ? by setting where 7 has length ?? and ?[ is the path of length ?? given by j't(s) = 7 (A - ?)?? + s). Then ? factors to give a map ? : F ? CQX —» ? xx PX. Moreover, by construction K(y,j, 1) = (y',7) where y' = $(y, 7, ?7) e F. Thus it extends to the map (K, /) : (F ? CQX) UFxnx (CF ? QX) —» (? x* PX) U; (CF ? ??) = F' , where f(ty,j) = (ty',j) is obviously a homotopy equivalence. But formula A.15) identifies (F ? 6??) UF*ax (CF x_QX) as F * ??. Moreover, if cXo denotes the loop of length 0 at ar0 then ? restricts to the inclusion F x {cXo} ? {1} —> ? ? ? PX, and this inclusion is a homotopy equivalence by Proposition 2.5. Thus ? is a homotopy equivalence and hence Theorem 1.13 asserts that so is (K,f). (ii) As above. F' = (? ? ? ?X) U,· {CF x ??), with ?? acting by multiplication on the right. Observe that composition of paths defines a homo- topy equivalence MX ? ? ? ? -=» PX. This extends to a homotopy equivalence
Rational LS-Category 357 ? xx PX = ? ?? MX ?? PX -=> ? ?? PX = F and hence (Theorem 1.13) to a homotopy equivalence ?-.F'^FUa (CF ? ??) , where a : F ? ?? —> F is the action of ??. If 7 G ?? has length ? define 7' e ?? to be the loop of the same length satisfying ¦y'(t) = ?{? — t). Then 7 1—> 7' is a homeomorphism and the maps 7 1—> 7*7', 7 1—> 7' ·7 are homotopic to the constant map. It follows that a homotopy equivalence m : CFxuX —> CFxuX is defined by (y, 7) 1—> A/7,7). Denote by q0 : F ? ?? —> F and q\ : F ? ?? —> ?? the projections. Then qi ·?? = a and so (Theorem 1.13) id Urn : F Ua {CF ? ??) ^ F Ugo {CF ? ??) is a homotopy equivalence. By inspection the homotopy equivalence {id Urn) ? ? : F' -?» FUgo {CF x ??) converts the action of ?? in F' to the diagonal action in CF ? ??. Since F Ugo {CF ? X) = F Ugo (/ ? F ? ??) UCl UX = F * ?? , the proof is complete. D (c) Ganea spaces and LS category. Fix a based path connected topological space, {X, x0). Start with the path space fibration, and iterate the fibre-cofibre construction to produce a sequence of fibrations PX PiX PmX The space PmX is called the mth Ganea space for X and pm is called the mth Ganea fibration. We set P0X = PX, Po = ? and we denote the inclusion of the fibre of pn at x0 by jm : Fm —» .PmAr. Thus Fo = ?? and (Proposition 27.6) Note also that if / : {Y, yo) —> {X, xo) is a continuous map of based path con- nected spaces then the functoriality of the Ganea construction gives commutative diagrams PmY Pmf PmX Y - X , m > 0 ,
358 27 Lusternik-Schnirelmann category in which Pof = Pf. Notice that cat PmX < m . B7.7) Indeed, since PmX — Pm-iX Ujm_j CFm_i we have = cat (Pm-iX Ujm_, CFm_i) < catP^A" + 1 , by Proposition 27.2 and Lemma 27.3 respectively. Since P0X = PX is con- tractible, B7.7) follows by induction. Proposition 27.8 (Ganea [62]) The following conditions are equivalent on a continuous map f :Y —> X from a normal space Y : (i) f = pma for some continuous ? : Y —> PmX¦ (ii) f ~ pma for some continuous ? : Y —> PmX- (Hi) cat f <m. proof: (i) <i=?> (ii): Suppose q : ? —> X is any fibration and h,? '· W —» E, hx : W —> X are arbitrary continuous maps such that qhs ~ ??. Lift the homotopy starting at Jie to obtain Jie ~ h with qh = hx. In particular (i) <=$> (ii). (H) => (Hi): If (ii) holds apply Lemma 27.1 to conclude cat/ = catpma < catpm < catPmX. Since catPmX < m by B7.7) it follows that cat/ < m. (Hi) => (ii): Denote the constant map by cy : Y —» {^o}. If cat/ = 0 then / ~ cy, which certainly factors through po- Suppose cat/ < m, some m > 1. Then Y = \J Ui, where the Ui are open and f\u_ ~ cy\v , Since Y is i=0 ' ' normal there are open subspaces Vo,V\ CY such that Y = VbU Vi and such that m-l the closures A and ? of Vo and Vi satisfy A C U Ui and ? C Um. Denote /I and cy\A by /a and ca- Clearly cat/? < m — 1. Since A is closed it is normal and so we may suppose by induction (because (i) <i=?> (ii)) that Ja = Pm-??? for some ? a '· A —> Pm-\X, On the other hand, since ? C Um there is a homotopy HB : ? ? / -^ X from f\B to cY\B. Now construct a continuous map ?:??{0}?(??)???)???{1}^ Pm-\X ??{0}
Rational LS-Category 359 as follows. First, set ?\??<0-, = ??. Next, recall (§l(f)) that the points in CFm_i are denoted by tz, t e I, z e Fm-i· Let v0 be the cone point: vQ = Oz for all ? e Fm_i. Lift Hb to a homotopy ? : AnB ? I —> Pm-\X from ? a \AnB to a map -4??-> Fm^1. Define ?\4nBx/ by ? H(y,2t) , 0<i< I . <?< Finally, set </?(? ? {1}) = v0. Since Y is normal there is a continuous function h : Y —> I such that h(Y — \\) = 0 and h(Y — Vo) = 1. Then y <—> (y,h(y)) defines a continuous map a:F->Ax{O}U(An?x/)UBx{l}. Set ? = ???:?—> Pm-iX U CFm_, ^-^ PmX . It remains to show that pma ~ /. Note that pm(CFm-i) = xo· Thus pm(p extends to the continuous map ? :? ? {0} U ? ? ? —> X given by ip(y, 0) = fy and, for y G B, f HB(y,2t) , 0<?<| The obvious deformation of y ? {0} U ? ? I to ? ? {0} provides a homotopy from ????? = "?a to /. D Corollary // X is a normal topological space then cat X < m if and only if there is a continuous map ? : X —» PmX such that pmv = idx· ? (d) Cone-length and LS category: Ganea's theorem. The iterated suspensions ???? of a based topological space X are defined inductively by ?°? = X and Y.kX = ???^?) , k > 1 . They appear in the Definition An ?-cone is a based topological space (?,??) presented in the form {po} = PoCPi C---CPn = P where Pk+1 = Pk Uhk ??*?· , k > 0 , for a sequence of well-based spaces (Yk,yit) and continuous maps hk : {EkYk,yit) — (Pk,xo)· The (Yk,yk) are the constituent spaces of the cone and the hk are the attaching maps.
360 27 Lusternik-Schnirelmann category Remark 1 A simple van Kampen argument shows that if (P,xo) is an n- cone with path-connected constituent spaces then each Pk (including ? itself) is simply connected. Remark 2 Note that Pi = ???, and so the 1-cones are simply the suspensions. Remark 3 An ?-cone is automatically well-based since each (??'?^,?'?^ is an NDR pair. In §l(a) we introduced a homeomorphism ?5* = Sk+1. In the same way we have CSk = Dk+l. Thus an ?-dimensional CW complex X with a single 0-cell, no 1-cells and based attaching maps hk '¦ Vq-^o —* Xk 1S a special case of an n-cone: xk+l = xk uhk Definition The cone-length, cLY, of a topological space X is the least ? (or oo) such that X has the homotopy type of an n-cone. In this topic we establish two main results. Proposition 27.9 (Ganea [63]) If(X,xo) is well-based then each Ganea space PnX has the homotopy type of an n-cone with constituent spaces Yk — (QX)Ak+1. In particular, if X is simply connected so is each PmX, Theorem 27.10 (Ganea [62]) If X is normal then cat AT < m <==>· ? is a homotopy retract of an m-cone. In particular, cat.Y <clAr . Remark A recent result of Dupont shows that, even rationally, this inequality can be strict; he constructs a rational CW complex X with catA7" = 3 and clA' = 4. proof of Proposition 27.9: Recall that pn : PnX —> X denotes the nth Ganea fibration (§27(c)) and that the fibre inclusion at xq is denoted by jn : Fn —> PnX- Let cI0 denote the loop of length zero at xo· Step 1: (?.?, cI0) is well-based. Choose a continuous h : X —> I, an open set xq G U C X and a homotopy rel xo, H : U x I —> X from the inclusion of U to the constant map U —> Xq, such that h~l(G) = xq and /?-1([0. ?]) C U. Define a continuous function k :
Rational LS-Category 361 ?? —> / by ?G) = ?7 + sup {?G*) | ? G [0, oo)}, ?7 denoting the length of 7. Let V C ?? be the open set of loops of length < 1 that lie entirely in U. If 7 G V let 7t be the loop of length it given by i^s),») , 0<i<§. , f*r , 0<f< 7 \ x \ < t < 1 . ' \ B - 2?)?7 , I < ? < 1 . Then k,V and ? : (-y,t) 1—> {it,it) exhibit (??,??0) as well-based. Step 2: The smash products (??)?* = ?? ? · · · ? ?? (k times) are well-based and ?"(??)??+1 ~ Fn. proof: If (.4, uq) and (?, 60) are well-based then (.4 ? ?,? ? {b0} U {a0} x B) is an NDR pair (Proposition 1.9). Hence ,4 ? ? is well-based. It follows by induction starting with Step 1 that (??)?? is well-based. On the other hand, Fn ~ Fn^i *CIX, as follows from Proposition 27.6. Since Fo = QX and since for well-based spaces A * ? ~ ?(? ? ?) ~ ? ? ?? we have (by induction) that Fn ~ (??)*<?+1) = (??)*(??)*? ~ ??*?"-1 ((??)??) ~ ?? ((??)??+1) . Step 3: ??? has the homotopy type of an ?-cone with constituent spaces Yk = (??)?*+1. Recall from §27(c) that P0X = PX ~ {pt} and that PnX ~ Pn-iX Ujn_^ CFn_lt Choose h : ?"-1(??)?? ^ Fn^ (Step 2) and (by induction) / : Pn-iX -^ Z, where ? is an (n—l)-cone with constituent spaces Yk = (??'')??"+1. Set g = fjn-ih. Then Theorem 1.13 asserts that the maps are homotopy equivalences. Moreover, since ??-1(??)?? is well-based ?7?"-1(??)?? ^ ???-1(??)?? and so, finally ??-?? Ujn_, CFn^ ~ ? Uh ???-'(??)??, for some continuous h : ?"-1(??)?? —» ?. Since the homo- topy type of an adjunction space only depends on the homotopy class of the adjoining map (Lemma 1.12) we may take h to be a based map. ? proof of Theorem 27.10: Reduce to the case of well-based spaces (X, xo) by replacing X by X Uz [0,1] and setting x0 — 1. If X is a homotopy retract of an m-cone ? then catX < cat ? (Lemma 27.1(v)) and cat ? < m (induction using Lemma 27.3). Conversely, if cat X < m then X is a retract of PmX (Corollary to Proposition 27.8) which has the homotopy type of an m-cone (Proposition 27.9). The final assertion of the Theorem follows immediately from this equivalence.u (e) Cone-length and LS category: Cornea's theorem. The purpose of this topic is to establish Cornea's remarkable strengthening of Theorem 27.10:
362 27 Lusternik-Schnirelmann category Theorem 27.11 (Cornea [40]) If X is a normal topological space then cat ?' < m 4=> X V ?,? has the homotopy type of an m-cone for some m — 1 connected space Y. Remark In fact, Cornea shows that there is an m-cone of the form X V Emlr' and deduces that cl A' < cat X + 1. We will not prove this refinement here since rationally it is immediate and since it requires a result from homotopy theory not included in the text. In proving Theorem 27.11 we shall need to identify spaces as having the ho- motopy type of ?-cones. For this we shall reply heavily on the following. Basic Facts: • // (Y, A) is an NDR pair then the homotopy type of ? U/ Y depends only on the homotopy class of f : A —> ? (Lemma 1.12). • // (?',?1) is a second NDR pair and if f : A' —> Z' then compatible homo- topy equivalences (?,?,?) —>· (?1, Y', A') induce a homotopy equivalence ? U.4 ? ^ ?' LU- Y' (Theorem 1.13). • // (Y, A) is an NDR pair and A is contractible then the quotient map Y —> Y/A is a homotopy equivalence (Corollary to Theorem 1.13). Now suppose q:E—>B is a fibration with fibre inclusion j : F —> ? at some base point b0 G ?. The key step in the proof of Theorem 27.11 is Proposition 27.12 Suppose ? has the homotopy type of an ?-cone with con- stituent spaces Y/.. Then ? U,· CF has the homotopy type of an ?-cone whose kth constituent space has the homotopy type of (Yjt ? F) U CF. proof: Step 1: Reduction to the case that ? is an n-cone. Let g : ? —> ? be ? homotopy equivalence from an ?-cone ? with constituent spaces Yk· Then (Z, zq) is well-based. Since ? is path connected we may replace g with a homotopic based map; i.e. we may assume g : (?,??) —> (?,bo). In the pullback diagram ZxBE ^ ? ?
Rational LS-Category 363 gs is a homotopy equivalence because g is (Proposition 2.3). By our second 'Basic Fact' above. (? ? ? ?) U,· CF -=> ? U,- CF; i.e. we may assume ? = ? is an ?-cone with constituent spaces Y^. Write ? = W U/ CA where (W, wo) is an (n - l)-cone with constituent spaces Yk, k < n, (A,a0) = (?'??,??) and / : (A,a0) —» (??,??). Let g ; ?,,t.·· —> PF be the restriction of the fibration q to W. Step 2. There is a homotopy equivalence of the form Ew U9 (CA xF) ^> ? in which ? : A x F —> ?\? restricts to j in {ao} x F. Let (?, /) : (CA, A) —> (B, W) be the canonical map, and use it to pull the original fibration back to a pair of fibrations (CA,A) xBE (?,?? ) (CA, A) (E,EW) (B,W), noting that (??, Ie) is projection on (E, E\y) and (?, ? a) is projection on (CA, A). The fibre of ? at ao is just f — {ao} x j : {a0} x F —» CA ? ? ?. Since CA is contractible, Proposition 2.3 asserts that f is a homotopy equivalence. Now let pL : CA x F —» CA and pR : ? ? ? F —» F be the projections, and consider the (non-commutative) diagram CAxF j'p" CA-x.bE We construct a new map h : CA x F —> CA ? ? ? such that ph — pL, h\r }xF = j' and both h and its restriction hA : A x F —> ? ? ? ? are homotopy equivalences. For this we recall (Proposition 2.1) that a fibration has the lifting property with respect to any DR pair and in particular with respect to a pair of the form (? ? ?, ? ? {0} Ul" x /) where (Y, Y') is an NDR pair. We also recall (Proposition 2.3) that the pullback of a fibration over an NDR pair is an NDR pair. _ _ Choose a based homotopy ? : CA ? I —> CA from the constant map CA —> {ao} to id-cA. Lift ? ? (pL ? id/) to a homotopy rel {ao} x F starting at j'pR. Define h to be the restriction of this homotopy to CA x F ? {1}. Then ph = pL and so h restricts to h a. : A x F —> A X? F. Also h restricts to j' in {ao} x F and so h is a homotopy equivalence.
364 27 Lusternik-Schnirelmann category It remains to show hA is a homotopy equivalence. As above, modify an arbi- trary homotopy inverse of h to obtain a homotopy inverse h' satisfying pLh' = p. Let p' : CA x I —> CA be the projection. If If is a homotopy from hti to idcAXBE, then there is a homotopy rel CAx?E ? {0,1} from pK to p'(p ? idj) because CA is contractible. Lift this to obtain ? ~ K' rel CA ? ? ? ? {0,1}. Then ?' is a homotopy from hh' to "%???? and pK' = p'(pxidj). In particular K' restricts to a homotopy hAh'A ~ idAXBB. Similarly, hAh,A ~ idAxF. Now set ? = fEhA : Ax F —> Ew. Since (trivially) ? = Ew U/E (U.4 x??) our second 'Basic Fact' gives (id, h) : Ew ?? (CA xf)A Ew USb (CA xBE) = E . Step 3: Completion of the proof of the proposition. Recall the notation established at the end of Step 1. Since the map h of Step 2 restricts to f in {ao} x F it follows that ? = f?hA restricts to j : {ao} xF —y E. Extend ? by idcF to a map g : Ax FU {ao} x CF —> Ew U,· CF. By our second 'Basic Fact' we may adjoin CF to both sides of the homotopy equivalence of Step 2 and still have a homotopy equivalence. This may be written in the form (Ew Uj CF) Ug (CA x F U {aQ} ? CF) ^ ? us CF . By induction Ew Uy CF has the homotopy type of an (n — l)-cone, (D,do) with constituent spaces of the homotopy type of Yk x F U CF, k < ? — 1. Use the second 'Basic Fact' to obtain D Ug' (CA x F U {ao} ? CF) ~ ? Uj CF , for a suitable map g' : ? ? FU {a0} ? CF —> D. Choose a well-based space F' of the same homotopy type of F (e.g. F' = FUy [0,1]). Then CAxFU {a0} xCF~CAxF'u {a0} x CF' ~CAx F'/{aQ} x F' . Thus our 'Basic Facts' give a homotopy equivalence of the form D Ufc (CA ? F'/{a0} x F') ~ ? Uj CF , where k : A x F'/{a0} x F' —> D is a based map. Finally, recall (§l(f)) that the points of CA are denoted by ??, ? € ?, ? € I. Define a homeomorphism CA x F'/{a0} x F' A C (A x F'/{a0} x F') by (ta,y) ?—> t(a,y). Dividing by A x F'/{ao} x F' on the left yields a home- omorphism ?? ? F'/{a0} ? F' ^?(?? F'/{a0} x F'). Thus D Uk (CA ? F'/{a0} xF') = D UkC (A x F'/{a0} x F1) and A x F'/{a0} xF' = ?"-1^..! ? F'/{a0} x F' S ?"" (Yn^ x F'/{a0} x F1) .
Rational LS-Category 365 Since ??-? ? F'/{a0} xF' ~Yn-XxF'U CF' ~ Yn^ xFuCF, the inductive step is complete. D proof of Theorem 27.11: If A" V ?1' has the homotopy type of an m-cone then X is a homotopy retract of an m-cone and cat X < m (Theorem 27.10). Conversely, suppose cat ?" < m. Then there is a continuous map ? : X —> PmX such that pma = id ? (Corollary to Proposition 27.8). Convert ? to the fibration q : X Xpmx M(PmX) —> PmX as described in §2(c). Denote X xpmx M(PmX) by ? and denote the inclusion of the fibre of q by j : F —> ?. Since PmX has the homotopy type of an m-cone (Proposition 27.9) it follows that ? Uj CF has the homotopy type of an m-cone too (Proposition 27.12). On the other hand, as observed in §2(c) there is a homotopy equivalence A : A" -^» ? such that q\ = ?. Let A' be a homotopy inverse for A. Then \pmq ~ ApmaA' = AA' ~ id?. Hence j ~ \pmqj, which is a constant map. By our first basic fact "EUj CF ~ EVCF/F. But CF/F ~ ?? for any well pointed space Y homotopy equivalent to F. Thus A' V ?? ~ ? V ?? ~ ? Uj CF , and ? Uj CF has the homotopy type of an m-cone. Finally, it is relatively easy to show that F ~ flFm. However, here we simply notice that since pma = id ? it follows that 7r,(PmA') = ?* (A") (Bn*(Fm). More- over 7r,(g) is an isomorphism of ?*(?) onto the subgroup identified with ??,(?'). It follows that the connecting homomorphism for the fibration q maps K*(Fm) isomorphically onto tt^^F). Since Fm ~ Em(OAT)Am+1 and ??' is path con- nected we may conclude that Fm is simply connected and that Hi{Fm) = 0, 1 < i < m. By the Hurewicz theorem 4.19, Fm is m-connected. Hence F and Y are m — 1 connected. ? Again, consider a fibration q: ? —> ? with fibre inclusion j : F —>· ? at some base point bo G B. From Proposi- tion 27.12 we deduce Proposition 27.13 If ? is normal then d(E Uj CF) < cl ? and cat(? U,· CF) < cat ? . proof: The first assertion is immediate from Proposition 27.12. For the second, let cat? = m. Then ? is a homotopy retract of an m-cone ? (Theorem 27.10). If ? -U ? -^? ? satisfy rf ~ idB then the homotopy lifts to a
366 27 Lusternik-Schnirelmann category h ~ ids such that qh = rf. Thus we have maps of fibrations ? ? (fg,fi). f ? which exhibit ? U CF as a homotopy retract of (? ? ? ?) U CF. Since P is an m-cone so is (P xB E) U CF, by Proposition 27.12, and so cat? U CF < m. D Example 1 ITie LS category of the Ganea spaces PnX. Suppose X is a normal topological space with catX = m, m < oo. We show that in if ? < m m if ? > m. In fact, since Pn+xX ~ PnX U CFn it follows that catPn+iX < catPnX + 1 (Lemma 27.3). On the other hand, X is a retract of PmX (Corollary to Proposition 27.8) and thus catPmA" > catX = m. Since P0X is contractible catPo^ = 0 and now the first inequality implies catPnX = ?, ? <m. On the other hand, Proposition 27.13 gives ca.tPnX < catA", all n. Thus since X is a retract of PnX, ? > m we have catX < catPnX < catA", ? > m.D (f) Cup-length, c(X; k) and Toomer's invariant, e(X;jk). The cup-length, c(X; Jk), of a path connected topological space X is the great- est integer ? such that there are cohomology classes a\,...,an G ?+{?;$?) such that ?? U · · · U an ? 0. Toomer's invariant, e(X; Ik) is the least integer ? for which there is a continuous map / : ? —> X from an ?-cone ? such that H*(f; ]k) is injective. (This invariant was introduced in a slightly different form by Toomer in [149].) Proposition 27.14 If X is a path connected normal space then for any coef- ficient ring fc, c{X;k) < e(X-]k) < catX < cLY . proof: To show c(X;Jk) < e(X;Jk) it is enough to show that c(Z\lk) < ? for ?-cones Z. Write ? = ? Uh CA for some (n — 1) cone Y, choose classes a0,. -., an € H+(Z; ]k) and let ? = ???· · Uan_i- Recall (§5) that the inclusion C*(Z,Y;k) —> C*(Z;lk) induces a morphism A : H*(Z,Y;Jk) —> H*(Z;Jk) of H*(Z;Jk)-modules. Now by induction ? restricts to zero in H*(Y;k) and so ? = ??', some ?' € ?* (?, Y; M). Hence ? U an = ?(?' U an).
Rational LS-Category 367 On the other hand, by excision the obvious map (?, h) : (CA, A) —» (?, ?) induces an isomorphism ?*(?,?) : H*(Z,Y;Jk) -=» H*(CA, A;Jk). Moreover ?*(?,?)? U an) = H*(<p,h)?' U ?*(?)?? = 0, because CA is contractible. Thus 0' U an = 0 and hence so is 0 U an = \{0' U an). This proves that c(Z;M) <n. The remaining inequalities are trivial consequences of Theorem 27.10 (where, indeed, the third inequality is stated). D Next recall the Ganea fibrations pn : PnX —> X introduced in §27(c). Proposition 27.15 If (X.x0) is well-based and normal then e(X;Jk) is the least integer m (or oo) such that H*(pm;Jk) is injective. proof: Clearly e(X; Jk) is less than or equal to this least integer, because PnX has the homotopy type of an ?-cone (Proposition 27.9). On the other hand, if e(X;Jk) = m choose a map / : ? —> X from an m-cone such that H*(f;Jk) is injective. We can suppose / is a based map because ? is well-based. (Replace / by a homotopic map if necessary.) Recall (§27(c)) that / induces continuous maps Pmf : PmZ —> PmX such that Pm °Pmf = }°Pm- Moreover, since ? is an m-cone cat ? < m (Theorem 27.10) and so there is a continuous map ? : ? —> PmZ such that p^ ? ? = id ?- Then / = / °Pmocr = Pmo pmf ° ?. Since H*(f; k) is injective so is H* (p?; Jk). ? Next, consider a continuous map between topological spaces, and recall that it induces maps PnY ^-^ PnX I* ? 7 . ? between the Ganea fibrations. If ? * (/; k) is an isomorphism then clearly c( Y ; .&) = c(JY; h). However, we also have Proposition 27.16 Suppose X and Y are normal and simply connected, and let Jk cQ. If H*(f;Jk) is an isomorphism then e(Y;Jk) = e(X;k). Remark There can be no analogue of Proposition 27.16 for LS category. In- deed, the inclusion of the base point, xo, in the space X below is a weak homotopy equivalence:
368 27 Lusternik-Schnirelmann category X However, no neighbourhood of x0 is contractible in X and so cat X = oo! The main step in the proof of Proposition 27.16 is Lemma 27.17 Suppose Jk C Q and H*(f;Jk) is an isomorphism. Then H*(Pmf;Jk) is an isomorphism for m > 1. In particular if f is a weak ho- motopy equivalence so is each Pmf. proof: Suppose first that if /1.4 : A' —> A and hg '· ?1 —> ? are continuous maps between path connected spaces, and that Ht(fiA',lk) and Ht(hB',lk) are isomorphisms. Then the natural homeomorphism (CA ? ?) XaxB {A x CB) = .4 * ? of §l(f) implies that H,(Iia * ??; $:) is an isomorphism. Let Fmf : FmY —> FmX denote the restriction of Pmf to the fibres. Proposi- tion 27.6 provides homotopy equivalences Fm(—) ~ Fm_i(—) *?(—) which, with a little care, can be chosen to identify Fmf with Fm_i/ * Qf up to homotopy. The Whitehead-Serre theorem 8.6 asserts that H*(uf;]k) is an isomorphism. Hence so is each H*(Fmf; ]k). A simple van Kampen argument shows that the join of path connected spaces is simply connected. In particular, it follows from Theorem 8.6 that each Tr*(Fmf)® Jk, m > 1, is an isomorphism. Hence PmX is simply connected, m > 1 and 7T«(-Pm/) ® Jfe is an isomorphism (long exact homotopy sequence). A second application of Theorem 8.6 gives that Ht(Pmf;k) is an isomorphism. The final assertion of the lemma follows immediately from the same theorem. D proof of Proposition 27.16: If g : W —» ? is a continuous map and if H*(g; k) is an isomorphism then C«(p; ]k) is a chain equivalence (i.e., has a chain inverse) and so C*(g;lk) is a quasi-isomorphism too. In particular. H*(f;]k) and H*(Pmf\Jk) are isomorphisms. Now use the construction X Ux [0,1] to reduce to the case / is a based map between well-based spaces. Then apply Proposition 27.15. D Exercises 1. Let X and ? be based path connected spaces, and suppose catZ < m. Prove that every continuous map f : ? —> X factors up to homotopy through Pm(X).
Rational LS-Category 369 2. Let X be a simply connected CW complex and Y = X U^ en. Suppose that cat X = m and ? < 2m. Prove that csXY < csXX. 3. Prove that P^X) * ???. 4. A co-H-space is a space X together with a continuous map V : ?" -4 X V X such that (idx V *) ? V ~ idx and (* V ???) ? V ~ idx. Prove that a based CW complex X is a co-H-space if and only if catX < 1. 5. Let X be a finite CW complex. Prove that cat0X = cat,Yp, where ? is the complement of a finite set in the set of all prime numbers (cf. §9-exercise 5).
28 Rational LS category and rational cone-length In this section the ground ring is Q. We begin by recalling some basic facts from §8 and §1 that will be used without further reference; then we introduce cato, cl<j and e<j as (geometric) invariants and finally outline the main results of the section, which treats rational category from a geometric perspective. Firstly, if / is a continuous map between simply connected topological spaces then H, (/; Q) is an isomorphism if and only if ?» (/) <g> Q is an isomorphism (Theorem 8.6). In this case / is a rational homotopy equivalence. Two spaces X and Y have the same rational homotopy type if they are connected by a chain of rational homotopy equivalences in alternating directions (Proposition 9.8), in which case we write X ~q Y. A simply connected space is rational if its homotopy groups (or, equivalently, its integral homology groups) are rational vector spaces. For each simply con- nected space X there is a relative CW complex {Xq, X), unique up to homotopy type rel A' such that A'q is a rational space and A' —> Xq is a rational homo- topy equivalence (§9(b)). We call such an Xq a rationalization of X. If X is a CW complex then so is A'q. Finally if g : X —> Y is any continuous map into a simply connected rational space Y then g extends (uniquely up to homotopy rel X) to a map pQ : A'q —>· Y (Theorem 9.7). Definition Let X be a simply connected topological space. 1 The rational LS category, cat0 X, is the least integer m such that Ar ~q Y and cat y = m. 2 The rational cone-length, cl0 A, is the least integer ? such that X ~q Y and cl(F) = n. 3 The rational Toomer invariant, eQX, is the least integer r such that X ~q Y and e(Y; Q) = r. 4 The rational cup length, cqX, is the cup length of H*(X;Q). Note that, by definition, these are invariants of rational homotopy type. Among the main results of this section we establish • For simply connected CW complexes X, cato A = cat A'q, cl0 X = cI(Aq) and e0X = e(A;Q). • For any simply connected space X, e0X < cato Ar < cl0 A" < cat0 X + 1 . • (Mapping Theorem) If a continuous map f : A' —> Y between simply connected spaces satisfies: Ti*(f) ® Q is injective, then cato A < cato Y¦ This section is organized into the following topics: (a) Rational LS category.
Rational LS-Category 371 (b) Rational cone-length. (c) The mapping theorem. (d) Gottlieb groups. (a) Rational LS category. Here we prove Proposition 28.1 If X is a simply connected CW complex, then (i) catoA' = cat.YQ. (ii) e0X = e(X;Q) = e(XQ;Q). First we need Lemma 28.2 Let X be a simply connected CW complex. (i) If f : X —» Y is a weak homotopy equivalence then cat X < caXY. (ii) catA'Q < catX. proof: (i) Put cat F = ??. Then cat/ < m (Lemma 27.1). Hence / factors asp^off for some continuous ? : X —> PmY (Proposition 27.8). Since Pmf : PmX —> PmY is also a weak homotopy equivalence (Lemma 27.17) we may lift ? through Pmf to construct a continuous ? : X —> PmX such that Pmf °? ~ ?. Then / ? ?^ ? r ~ / and so pj? ? r ~ id ? (Lemma 1.4). It follows (Proposition 27.8) that catAr < m. (ii) If cat ?' = m then there is a map ? : ?' —> PmX such that ?* ? ? = id (Proposition 27.8). This extends to aq : Xq —>· (PmX)q, and (Pm)Q°*-Q~U/. On the other hand, let j :'X —> Xq be the inclusion. A simple calculation (using Fm ~ ???(??)???+? of Step 2 in Proposition 27.9) shows that Pm (Xq) is a rational space. Hence Pm(j) extends to a continuous map Pm(j)Q ¦ Pm(X)Q —> and Pm" °Pm(j)Q ~ {Pm)n· Since Xq is a CW complex it follows that ~ id and cat Xq < m (Proposition 27.8). D proof of Proposition 28.1: Suppose X ~q Y. Choose a weak homotopy equivalence ? —> Y from a CW complex Z. Then Xq and Zq have the same weak homotopy type. But these are CW complexes and so they have the same homotopy type. Hence cat^Q = cat^Q < cat ? < cat F. On the other hand, Proposition 27.14 states that e(Y; Q) = e(Z; Q) and Propo- sition 27.15 identifies this as the least integer m such that H*(p^l;Q) is injective. Rationalizing PmZ and Z, we can replace p^ with (^)„. But it follows from
372 28 Rational LS category and rational cone-length the proof of Lemma 28.2(ii) that if H* ( (p^) Q; <Q>) is injective so is H* (p^ ; Q) . Thus e(XQ; Q) = e(ZQ; Q) < e(Z; Q) = e(Y; Q). D Example 1 cat0 [\/ Xa) = max{cat0Xa}. Vu / a Here the Xa are taken simply connected. We may suppose the Xa are rational CW complexes and then the equality follows from Proposition 28.1 and the analogous equality for cat (introduction to §27). D (b) Rational cone-length. As a special case of the ?-cones denned in §27(d) we define spherical n-cones to be based spaces (P.po) presented as {po} = Po C Pi C · · · C Pn with each Pk+i — Pk Uhk CSk, where Sk has the form Sk = V Sr°+k+1 , ra > 0 . a€Jk (In particular any composite of ? +1 Steenrod operations vanishes in a spherical ?-cone. Thus, although ECP°° is a 1-cone, it is not a spherical ?-cone for any n.) Notice that each Pk in a spherical ?-cone is automatically simply connected. Proposition 28.3 (i) If X is simply connected then clo X is the least integer ? such that there is a rational homotopy equivalence f : ? —» X from a spherical ?-cone, P. (ii) If X is a simply connected CW complex then clo X = Lemma 28.4 If ? : X —>· Q is a rational homotopy equivalence to a simply connected ?—cone, Q, then there is a rational homotopy equivalence f : ? —» ?' from a spherical r-cone, P. proof: Write Q = Qr D · · · D Qo = {q0} with Qk+i = Qk U5fc CT.kYk as in §27(d). Sublemma We may suppose each Qk is simply connected. proof of sublemma: It is sufficient (van Kampen) that Yq be path connected. Now Qi is simply connected because Q is, and so Hi(Q\;Z) = 0 (Theorem 4.19). It follows that #?(??) : ?1(??1;?) —» ??(??0;?) is surjective, since this may be identified with the connecting homomorphism iI2(<2i,Qo;Z) —> Hi(Qo\Z). Next, recall that Yq and Y\ are well-based with base-points yo and y\. Let yoa and yi? be base-points for the other path components of Yq and Y\. Then the interval through yoa becomes a circle in ?^? representing a homology class ??, and ?\(???',?) = 0???, again by van Kampen and Theorem 4.19. Similarly
Rational LS-Category 373 ??(???;?) = @Zw0 where W? is represented by the interval through yl0. In 0 particular, we may choose a subset J of path components of Yl such that Now for each yOa adjoin an interval Ia to Yo by attaching {0} to yo and {1} to you· This gives Yo = V U ( ? Ia ). Similarly, construct \\ = ?? U ? ? ??\. Va J \0eJ ¦ J Then Yo is path connected, so ??0 is simply connected and pi extends to a continuous map ?? : ??? —> ??0. Set Q = ??? Up, ???? U ??'2?2 U · · · U ???~???-\. Then Q is simply connected and a straight forward calculation shows that the inclusion Q —> Q induces an isomorphism of rational homology. Thus we may replace Q by Q; i.e., we may assume Yo is path connected, and each Qk is simply connected. D We now return to the proof of the lemma. Observe that it is sufficient to consider the case that ? is a fibration. Set Xk = Q~1(Qk)- Since ?«(?) ®Q is an isomorphism the fibre, F, of ? satisfies it» (F) <g> Q = 0. Hence the restrictions Qk ¦ Xk —>· Qk satisfy ?*{?^ ® Q is an isomorphism. Suppose by induction on k that there is a continuous map ? : ? —>· Xk with ? homotopy equivalent to a spherical k-cone, Pk and 7r»(p)®Q an isomorphism. Again we may take ? to be a fibration. Theorem 24.5 provides a rational homo- topy equivalence h : Sk —> ?*?* with Sk = V Sk+1+r". Let ha = h Then the map C5t+1+r° -^- CEfcFfc —» Qk+i lifts through ?fc+1 to a based map Ha ¦ C5fc+1+r» —» ??+?, and this restricts to a map OQ : 5A+1+r° —> XA. Since 7r*(p)<g>Qis an isomorphism, there is a positive integer ma with mQ[oQ] G Im7r»(p). Moreover, without loss of generality we may suppose ma = 1· (Simply replace h by a new map h' so that [/iQ] = ma[ha]; then relabel /?' as h.) Then, since ? is a fibration, ?? lifts to a based map ?? : Sk+1+ra —> Z, and we may construct ? = (?, {Ha}) : ? U{(To} UV5fc+1+r» -* Xk+i . a Since ?, (?fcp) ® Q is an isomorphism so is H* ^kg; Q). Now an easy homology calculation shows that the composite of ? with ?^\ is a rational homotopy equivalence. Hence 7r„(p) ® Q is an isomorphism, and the lemma follows by induction. ? proof of Proposition 28.3: (i) Given a rational homotopy equivalence / : ? —> X from a spherical ?-cone we have clo(Ar) < cl(P) < n. Conversely, suppose clo(X) = n. Then there is a chain of rational homotopy equivalences ? _> ...<_ Q
374 28 Rational LS category and rational cone-length with Q an ?-cone. Apply Lemma 28.4 a finite number of times to obtain a rational homotopy equivalence / : ? —> X with ? a spherical n-cone. (ii) By definition cl0 A' < cI(A'q). Moreover, if Y ~q X and cl(F) = ? then (Lemma 28.4) there is a rational homotopy equivalence ? —> Y from a spherical ?-cone. Present ? as {p0} C · - · C Pn = ? with Pfc+Z = Pk U CY.kYk, where Yk is a wedge of spheres. Then there is an obvious weak homotopy equivalence {po} U CT(n)Q U · · · U CE" (rn_i)Q -> Pq . Denote the ?-cone on the left by P'. Since each {Pk+i,Pk) is a relative CW complex and since Pq —> Xq is a weak homotopy equivalence a direct application of Lemma 1.4 shows that P' —> Xq is a homotopy equivalence. Thus cl(AQ) < ?. It follows that cl(.YQ) < cl(F) for all Y ~q X; i.e. cl(AQ) = cl0 X. a Theorem 28.5 If X is a simply connected topological space then (i) e0X < cato X < cl0 ?' < cat0 A' + 1. (ii) If e0X = 1 then e0X = cato X = clo X = 1 and X has the rational homotopy type of a wedge of spheres. (Hi) If X is a CW complex then cato AT < m if and only if Xq is a homotopy retract of an m-cone. proof: (i) Since the invariants are invariants of rational homotopy type we may suppose A is a CW complex, so that eoAT = e(AQ;Q), cato A' = catXQ and clo X = cI(Aq). (Proposition 28.1 and 28.3). Now the first two inequalities of (i) follow from Proposition 27.14. For the last inequality and the proof of (iii), let catoAr = m. We lose no generality in supposing A a rational CW complex. Then A" ~ Xq and cat A" = m (Proposition 28.1). Thus the mth Ganea fibration pm ¦ PmX —> X admits a cross-section ? (Corollary to Proposition 27.8). Convert ? to a fibration ? —> PmX with fibre F, and ? ~ A". In the proof of Theorem 27.11 we showed that ? U CF ~ X V ?? , where Y is a well-pointed space homotopy equivalent to F. Hence A ~ (E U CF) U ???. On the other hand, we also showed in the proof of 27.11 that k*(F) = 7r*+i(Fm), Fm the fibre of pm. But Fm ~ Em (UXAm+1) by Step 2 in the proof of Proposition 27.9. Since the smash of path connected well-based spaces is simply connected (trivial) it follows that F and Y are m-connected. Finally, Proposition 27.9 identifies P?nX as homotopy equivalent to an m-cone and so EUCF also has this property (Proposition 27.13). Since X ~ (EUCF) ????, it follows that cl0 A < cLY < m + 1.
Rational LS-Category 375 Note that we have identified A' as a retract of PmX which is homotopy equiv- alent to an m-cone. Finally to prove (ii) suppose e0X = 1. Then the first Ganea fibration, ??.? —>· X, is surjective in rational homology. Since there is a rational ho- motopy equivalence from a wedge of spheres to ??.? (Theorem 24.5) the ratio- nal Hurewicz homomorphism for X is surjective and X itself has the rational homotopy type of a wedge of spheres (Theorem 24.5). D Remark As observed in §27, Theorem 28.5(i) is a special case of a theorem of Cornea. (c) The mapping theorem. The usefulness of rational LS category in rational homotopy theory is in large part due to Theorem 28.6 (Mapping theorem) Let f : X —)¦ Y be a continuous map between simply connected topological spaces. If 7?, (/) <g> Q is injective then cato X < cato Y · proof: We lose no generality in assuming X and Y are rational CW complexes. Let cato Y = m and convert / to the fibration g : ? = ? ? ? MY —> ? as de- scribed in §2(c). Then inclusion of the fibre ? ? ? PY induces zero in homotopy, because nt(g) is injective. The homotopy equivalence ? —> X converts this inclusion to a fibration ? : ? ? ? PY —> X and so ?, (?) = 0 and the fibre inclusion j : QY —» ? ?? PY satisfies: n*(j) is surjective. Decompose the rational vector spaces ?,,(??') as ?^ © ?'?, with tt,(j) : ?'? ^» ??(? xY PY). Let K'n and ?'? be cellular Eilenberg-MacLane spaces of types {?'?,?) and (??^',?), respectively. Apply the Corollary to Proposition 16.7 to obtain a weak homotopy equivalence Then setting ?' = f\ K'n we see that ?? : ?' —> ? ? ? PY is a weak homotopy ? equivalence. Now let A : ? ? ? PY —>· ? be the inclusion and set ? = (id, Cjtp) : ? l)Xjtp CK' —> El)x C(X ? ? PY) . By Proposition 27.13, cat (E U C(X ?? PY)) < m. Moreover, since Xj is the constant map, E{J\jv CK' = EV(CK'/K'), and this space is simply connected. Thus ? is a weak homotopy equivalence, by an obvious homology calculation.
376 28 Rational LS category and rational cone-length Since ? V (CK' jK') has the homotopy type of a CW complex we conclude from Lemma 28.2 that cato X < cat X < cat (E V (CK'/K1)) < cat (E Ux C(X ? ? PY)) < m . D Corollary /// is as in Theorem 28.1 and X is a CW complex then cat/jj = proof: Let cat/Q = m and let pm : Pm —» Yq be the mth Ganea fibration for Yq. Then /q = pma for some continuous map ? : Xq —> Pm (Proposition 27.8). In particular, ??,(?) <g> Q is injective and so m > cato Pm > cato Xq = cat Xq (Proposition 28.1). But also m = cat/Q < catXq by Lemma 27.1. ? Example 1 Postnikov fibres. Let X be a simply connected CW complex and suppose ??{?) = 0, i < r. From Proposition 4.20 we obtain a fibration X -?*· ? (??(?), r) such that ???(?) is the identity. If ? is a CW complex mapping by a weak homotopy equivalence to the fibre of ? then the composite ? —>· X satisfies: ?^(/) is an isomorphism for i > r + 1. In this way we obtain a sequence of maps __^ ??+\ 9j^ , , . __^ ?3 _?^ ?2 _ ? such that Xn+l is an ?-connected CW complex and ni(gn) is an isomorphism for i > ? + 1. The Xn are called Postnikov fibres of X. Notice that we can apply the Mapping theorem to this sequence to obtain ¦ · · < cat0 Xn <¦¦· < cat0 X2 < cat0 X . In particular, if X has finite rational LS category, so do all its Postnikov fibres. D Example 2 Free loop spaces have infinite rational category. Let X be a topological space. The free loop space of X is the space Xs of all continuous maps S1 —> X (§0). We show that • If X is two-connected and if H+(X;Q) ? 0 then cato(X5 ) = oo .
Rational LS-Category 377 (Better results can be proved with a little more work.) Indeed, let e e 51 be a basepoint. Evaluation at e defines a fibration ? : Xs —> X whose fibre at a basepoint x0 € X is the loop space UX. The map s : X —» Xs which associates to ? the constant loop 51 —> ? is a cross-section for p, and it follows that the inclusions j : ?.? —> Xs and s : X —> Xs1 are injective in homotopy. Suppose cato(Xsl) = m < oo. Since X is 2-connected, ??' and „Ysl are simply connected. Thus Theorem 28.6 asserts that if cato (Xs1) = m < oo, cato X < m and cat0 ?? < m. Suppose ft : 52n;+1 —> X and ga : S2mi —> X represent linearly independent elements in 7r„(X) <g> Q. Then, as in Proposi- tion 16.7, ?/; ? Hgj : ?]?52??+1 ? ?5"'^ —> ?? is injective in ratio- i j i j nal homotopy, and so the category of this product is bounded by m. Since H* (Q52ni+1;C) is a polynomial algebra (Example 1, §15(b)) it follows that TTodd(-X) <8> Q = 0. and dim 7reVen(X) ® Q is finite. Now X has a Sullivan model of the form AVeven and so its cohomology is a polynomial algebra, contradicting cato X < oo. D (d) Gottlieb groups. Recall (Theorem 1.4) that any based topological space (X,x0) admits a weak homotopy equivalence from a based CW complex (Y, yo), called a cellular model for X. In particular an element a € ??(?) is representable by a unique based homotopy class of continuous maps / : E",*) —> (Y,yo). Definition a ? ??(?) is a Gottlieb element for X if (/, id) : Sn V Y —> Y extends to a continuous map ?/ : Sn ? ? —i Y. (Note that this condition is independent of the choices of / and of cellular model.) Remark Gottlieb elements were introduced by Gottlieb in [67]; however, not with this name! Given a second Gottlieb element ? 6 ??(?) represented by g : Sr —> Y we note that restricts to (f,g, idY) in 5" V Sr V Y. More generally, if /f represents a Gottlieb element in ???; (X) then (ft,..., fk, idy) : S V · · · V S"* V Y —> Y extends to a continuous map Sni x · · · x Snk x Y —> Y . Next observe that the Gottlieb elements form a group. In fact if /, g : En, *) —> (Y, y0) then [/] + [g] is represented by En, *) ? En V 5n, *) -^4 (Y, yo) as described in the proof of Lemma 13.6. Since (f,g,idy) extends to a map S"xSnxr-> Y, \(ftg)j, idy) extends to 5" ? ?. The fact that the inverse of a Gottlieb element is a Gottlieb element is obvious.
378 28 Rational LS category and rational cone-length Definition The group of Gottlieb elements in ??(?) is called the nth Gottlieb group of X and is denoted by Gn(X)· (It is an invariant of weak homotopy type.) If X is simply connected the group Gn(X^) of Gottlieb elements in Xq is called the nth rational Gottlieb group of X and will be denoted by G®(X). -Note that G^(X) is a rational subspace of ???{?) <8> Q and contains Gn(X) <g> Q. This containment may be strict (see Exercise 1). Recall from §21 (d) the definition of the rational homotopy Lie algebra Lx = n»(UX) g> <Q> = 7t,(UXq). By definition, the Lie bracket in Lx corresponds up to sign to the Whitehead product in ?,(?) <g> Q under the isomorphism (Lx). ~?.+?( Proposition 28.7 If a € (Lx)n corresponds to a Gottlieb element inirn+i(X)<g> Q then [a,?}=0, ?eLx. proof: Let a € ??+1 (X) <8> Q be the Gottlieb element corresponding to a and let a be represented by a map / : (S"+1, *) —? {?, *). Without loss of generality assume X is a rational CW complex. Then for any g : (Sk, *) —? (?, *) the map (f,9) '¦ Sn+1 V 5* —> X extends to a map Sn+1 ? Sh —> X, which shows that the Whitehead product of ? = [/] and [g] is zero (§13(d)). D Example 1 G-spaces. Suppose a topological monoid G acts on a topological space X with basepoint #?· Restricting the action to {#?} x G defines a continuous map j : G —> X and, essentially by definition, lmicn(j)CGn(X) · In particular, nn(G) = Gn(G). ? Example 2 The holonomy fibration. Suppose / : X —> Y is a continuous map and, as in §l(c) construct the holonomy fibration ? ? ??? —? X with fibre ?,?. Since the inclusion i : ?.? —? X xy PY of the fibre extends to an action of UY it follows as in Example 1 that Imirn(i) CGn(X xY PY) , n>l. Now suppose / is itself a Serre fibration with fibre inclusion j : F —)¦ X. Then there is a weak homotopy equivalence between ? ? ? PY and F which identifies 1????(?) with ker7rn(i). It follows that keritnU) CGn(F) , n>l. ? Proposition 28.8 Suppose X is a simply connected topological space of finite rational LS category. Then
Rational LS-Category 379 (i) G®(X) is concentrated in odd degrees, and (ii) dimG®(X) < cato A'. proof: (i) We may assume A^ itself is a simply connected rational CW complex, and hence that Gm(X) = G®(X)· Suppose first that / : S2k —> X represents a non-zero Gottlieb element, and that X is BA; - l)-connected. By Theorem 4.19, ¦#2Jfc(A';<Q>) = 7it(A") ® Q = ^-(AT)· Thus there is a cohomology class w € H-k(X; Q) such that H*(f)w ? 0. Extend (/,...,/, idx) to a map ? : S2k ? • · · ? S7k ? ? —? X and observe that ?*{?)?? = Hm(f)w <g> 1 <g> · · · <g> 1 + • · · + 1 g» · · · g» H'm(f)w <g>l + l<g>---<g>l<g>i/;. If we have used ?-factors 5-fc then ?*{?)??? = {?*{?)??)? ? 0. It follows that ?? ^ 0 for all ? and hence cat0X = oo. Now suppose X is only simply connected, and let X2k —> X be its 2kth Postnikov fibre. If / : S'k —> A' represents a non-zero element of Gok{X) then we may assume / : S2k —l· X~k. Apply the argument of Lemma 1.5 to the diagram s2kvx2k —{Lid) ' x2k s-k ? ? ? to fill in the dotted arrow making the upper triangle commute. This identifies / as representing a non-zero element of G2k(X2k)- But now the Mapping theorem 28.6 gives cat0 X > cato X2k = oo. (ii) Again we may suppose A' is a simply connected rational CW com- plex. Thus G®(X) = Gm(X)· If Qi,...,Qr are linearly independent elements of Gm(X) then; as at the start of this topic we can extend representatives fi : Sn: —? ? to a map ? : Sni x · · · x 5n- x X —> X . Let g be the restriction of ? to Sni ? ·· · ? Sn". Then by (i) each n; is odd and so ??.E";) <g> Q = 7rn,.Eni) <S> Q = Q (Example 1, §15(d)). It follows that ?*{?) <8> Q is injective and so the Mapping theorem 28.6 asserts that cato A" > catotS ? · · · ? S·). But the rational cup length of Sni x · · · ? Snr is r and so catoEni ? · · · ? Sn') > r. Thus cato Jf > dimG^(A'). ? CorolLary Suppose j : F —l· X is the fibre inclusion of a Serre fibration X —> Y, and that F is simply connected. If cat0 F < oo then ker 7r„(y) ® Q is concentrated in odd degrees and dim (ker ??, (j) <g> Q) < cat0 F. proof: By Example 2, above, ker7r,(i) ® Q C G,(F) ® Q C G®(F). ?
380 28 Rational LS category and rational cone-length Example 3 Loop spaces. Suppose ?? is a simply connected loop space. We show that cat0 ?? < oo <=» ?*(?.?\ Q) = AV , with V = Vodd and dim V < oo . In fact if cat0 UY < oo then ?,(?.?) <g> Q = G,(UY) <g> Q (Example 1) and hence is finite dimensional and concentrated in odd degrees. Thus ?,(?) <g>Q is finite dimensional and §16(b) asserts that UY has a Sullivan model of the form (AV, 0) with V = ?, (UY) ? Q. " Conversely if H*(UY;Q) = AV as above then (AF,0) is a Sullivan model for ?.? and ?,(??) (8>Q is a finite dimensional vector space concentrated in odd degrees. If fi : S2n;+1 —? UY represent a basis then multiplication in QY defines ? a rational homotopy equivalence ? 52??+1 —> CIY and so cato ?? = r < oo. ? Exercises 1. Prove that if ? € GnX, then [?, ?] = 0 for any ? e nkX. 2. Let X = S3 V (Vn>45n) Un>4 (on[s3,S"}Dn+3). Prove that ?3(?) = ?, G3(X) = 0, and G^(X) = Q. 3. Let / : Sn -> X be a continuous map such that [/] € Gn(X), and ifn(/;Z)([Sn]) ? 0. Prove that X has the same homotopy type as ? ? Sn. 4. Let / : X -> F be a continuous map, and suppose that ?,(/) <8>Q is injective. Prove that cato/ = cato-X"· 5. Let f ¦ X —> Y be a continuous map such that H*(f\Q) is injective. Prove that eo(JO > eo(Y).
29 LS category of Sullivan algebras In this section the ground ring is an arbitrary field k of characteristic zero. Let X be a simply connected topological space with rational homology of finite type. Then the rational homotopy type of X is completely determined by its minimal Sullivan model (§12, §17) or by any Lie model (§24). It is thus not un- expected that we should be able to compute cl0 JY, cat0 X and eQX directly from these models. Recall also that a commutative model for X is any commutative cochain algebra connected to Apl(X) by a chain of quasi-isomorphisms (§10). Thus (A, d) is a commutative model for X if and only if its minimal Sullivan model coincides with that of X. Now we introduce the following Definition Suppose A = A-0 is a graded algebra with A0 = Ic. The prod- uct length of A, denoted by nil A, is the greatest integer ? (or oo) such that .4+ A+ ? 0 (? factors). Then we prove (§29(a)) a result of Cornea's [40]: • clo X < ? 4=> nil A < ? for some rational com- mutative model (A,d) of X. Note that H(A,d) = H*(X;.Q) and so the product length of ? (A) is just the cup length c(X;Q) as defined in §27(f). Thus cl0 X = ? corresponds to product length ? in a suitable cochain algebra model, while c(JY;Q) = ? corresponds to product length ? in the cohomology algebra. Next, suppose (AV,d) is any Sullivan algebra (§12), and m > 1. Recall that (AV,'d) is filtered by ideals (A>mV, d) of elements of word length > m. The surjections gm : (AV, d) —> (AV/A>mV,d) extend to relative Sullivan models of the form (AV,d) —^-* (AV®AZ(m),d) (AV/A>mV,d) , as described in §14(a). Definition The LS category, cat(AV,d), is the least integer m (or oo) such that there is a cochain algebra morphism nm .· (AV ® AZ(m),d) —> (AV, d) such that nmXm = idAv- The Toomer invariant, e(AV,d), is the least integer r (or oo) such that ?(??) is injective. In §29(b) we then establish
382 29 LS category of Sullivan algebras • // (AV, d) is a rational Sullivan model for X then cato A' = cat(AV. d) and e0X = e(AV, d) . We shall deduce this as a corollary of Cornea's theorem, although the result itself is older (with a different proof). Then we shall use this to give easy examples in which eoX < cato A". Notice that these results indicate a close relation between the topological filtration P0X c · · · C PmX C · · · and the algebraic filtration AV D A+V D---D A>mV D · · -. The second result above reduces the computation of cato X to the existence (or non-existence) of retractions (AV <g> AZ(m),d) —> (AV,d). Constructing such morphisms, however, is not easy, because they are required to be algebra morphisms. Fortunately, a magnificent theorem of Hess [90] about any Sullivan algebra over k comes to our rescue (§29(e)): • cat(AV, d) is the least integer m such that there is a morphism of (AV,d)- modules, irm : (AV <g> AZ(m),d) —> (AV,d), such that Trm\m = id\v- Finally, in §29(g) we shall extend the notion of e(AV, d) to e(M,d), for any (AV, d)-module (M,d). In particular, we may consider the module (AV,d)i = Homft(AV,*) with (a · f)(b) = (-l)de*ade*ff(ab), a,b G AV, / G (AV)*. If V is a graded vector space of finite type we use Hess's theorem to prove the recent result of Felix, Halperin and Lemaire [55]: This section is organized into the following topics: (a) The rational cone-length of spaces and the product length of models. (b) The LS category of a Sullivan algebra. (c) The mapping theorem for Sullivan algebras. (d) Gottlieb elements. (e) Hess' theorem. (f) The model of (AV,d) —> (AV/A>mV,d). (g) The Milnor-Moore spectral sequence and Ginsburg's theorem, (h) The invariants meat and e for (AV, d)-modules. (a) The rational cone-length of spaces and the product length of mod els. Our central result provides an algebraic description .of cone-length. It reads Theorem 29.1 Let Ik = Q. The following conditions are equivalent for simply connected topological space X with rational homology of finite type:
Rational LS-Category 383 (i) do A' < ?. (ii) rii\A<n for some commutative model (A, d) for X with A0 — Q. (Hi) X has a free Lie model (hy,d) with an increasing filtration 0 = V@) C V(l) C - - - C V(n) = V such that for each i, d : V(i) —> Ly (,_!). (iv) There is a rational homotopy equivalence ? —y X from a spherical n-cone whose constituent spaces are wedges of spheres with finitely many spheres in each dimension. The proof of the theorem requires one small technicality. Lemma 29.2 Suppose (A,d) is a commutative cochain algebra (over any k of characteristic zero) such that A° = Ik, H1(A) — 0 and each H'(A) has finite dimension. Then there is a subcochain algebra (B, d) such that; Bl = 0, each Bl is finite dimensional and d : B+ —> B+ · B+. proof: Choose A C A so that A1 = 0, ?2 ? ?(??) = A2 and A* = A\ i > 3. Then (A,d) ^> (A, d) and so we may suppose A1 — 0. Suppose next that A' is finite dimensional for i < n, and let A be the sub- algebra generated by A<n and (Im<2)n. Then A is a sub cochain algebra and .each A1 is finite dimensional. Now write .4" = d~1(An+1) ? Un and choose C CAso that (i) ? = A1, i < n, (ii) Cn = ??(??+1), (iii) Cn+l D An+l and Cn+1 ? dUn = An+l and (iv) Cl = Ai,i>n + 2. Then (C, d) is a sub cochain algebra including quasi-isomorphically in (A, d) and Cl is finite dimensional, i < n. We now construct a decreasing sequence of quasi-isomorphic inclusions of sub cochain algebras A D .4B) D .4C) D ¦¦¦ as follows: if A(k) is constructed so that A'(k) is finite dimensional, i < k, then A(k + 1) is a subcochain algebra including quasi-isomorphically in A(k) such that Al(k + 1) = Al(k), i < k and such that among all these Ak+i(k + l) has minimal dimension (necessarily finite by the argument above). Set ? = ?^(^) and note that Bl = Al(k), i < k, k so that (B, d) -=> (A,d). Note as well that for any proper sub cochain algebra (B, d) C (B, d) the inclusion is not a quasi-isomorphism. Let ? : ? —> k®B+ /B+ · B+ be the projection and let Q{d) be the differential in B+/B+ -B+. If kerQ(d) - ? ? imQ(d) then C^ © ii) is a sub cochain algebra including quasi-isomorphically in B. Thus ?~*(& ? ?) = B and H = B+/B+-B+;\.e.,Q(d) = 0. ? proof of Theorem 29.1: (i) => (ii). If cl0 X = 1 then ? ~ ?? for a well- based space Y, and the desired conclusion is exactly Proposition 13.9. For ? > 1 we may suppose X = Y U/ CZ where clo Y < ? — 1 and ? and Y are path connected. Let my : (AVy,d) —> Apl(Y) be a Sullivan model and extend
384 29 LS category of Sullivan algebras ApL(f)m,Y to a Sullivan model mz : (AVy ? AV,d) -=> APL(Z). Then the inclusion A : AVy —> AVy <g> AV is a Sullivan representative for /. As usual, let A(i,eft) be the free commutative cochain algebra on genera- tors t and dt of degrees zero and one and let 6\ : A(i, dt) —l· Q be the aug- mentation sending t >-> 1. The morphisms —> -f—— define a fibre product AVy ??vy®av (Q ? [AVV ® AV ® ?+(?, dt)}), and Proposition 13.8 identifies this as a commutative model for X. By induction there is a quasi-isomorphism of commutative cochain algebras (AVY,d) —> (B,d), where 5° = Q and m\B < ? - 1. Apply 5 <8>??? - to obtain a quasi-isomorphism (AVy- ® AV,d) ^i (B <g> AV,d) as described in Lemma 14.2. This identifies ? xb®av (Q®[B ® AV ® A+(t,dt)]) as a com- mutative model for X. This contains (A,d) — ? ?#®?? (Q@[B+®AV<g> A+(t,dt)j ? (AV <8> dt)) as a sub cochain algebra. Since the inclusion of dt in A+(i, dt) ? ker ei is a quasi-isomorphism, so is the inclusion of (A, d) in the fibre product. Thus (.4, d) is a commutative model for X satisfying nil A < n. (ii) => (in). If (ii) holds we may use Lemma 29.2 to find a commutative model (B, d) for X such that B° = Q, B1 = 0, each Bl is finite dimensional, d : B+ —> B+ -B+ and also nil ? < ?. Then ?(?M) is a Lie model for X (§23(a)). Now C(B,d) =U- withV = s-1Hom(B+,Q).'Put/(fc) = B+ B+ (fc+1 factors) and dualize the sequence 0 *- 5+//(I) <— B+/IB) < <— B+/I(n) = 5+ to obtain a filtration of V. The facts that 5+ -I(k) C /(& + 1) and d : I(k) —> I(k + 1) imply that the filtration on V has the stated properties, as follows immediately from the definition of the differential in C{B,d)- (iii) => (iv). We may clearly suppose X is a CW complex. It follows at once from Theorem 24.7 that (Lv, d) is the Lie model of a space of the form P. This means that X and ? have the same rational homotopy type, and so there is a weak homotopy equivalence Xq -=> Pq. Now apply Lemma 28.4 to the composite X -+Pq. (iv) => (i). This is obvious. D (b) The LS category of a Sullivan algebra. We begin by proving Proposition 29.3 (i) Let (AV, d) be a Sullivan algebra. Then cat(AVr, d) < m if and only if there
Rational LS-Category is a diagram of commutative cochain algebras (AV, d) (A,d) (C,d) 385 (B,d) in which ?a is a quasi-isomorphism and nil ? < m. (ii) If(AV, d) and (AW, d) are quasi-isomorphic Sullivan algebras then cat(AV, d) cat(AW,d). proof: (i) Recall the diagram at the start of this section. If cat(AV, d) < m, that diagram, together with the retraction (AV ® AZ(m),d) —i (AV,d) satisfies the Proposition. Conversely, given such a diagram note that ?? factors to yield 7 : (AV/A>mV,d) —> (B,d) with jgm = ??. In the diagram (AV,d) (AV®AZ(m),d) (A,d) (B,d) we may extend ? to a morphism ? : (AV <8> AZ(m),d) - ?? ~ 7Cmrel (AV,d). Finally, in the commutative diagram (A, d) such that (AV, d) (AV ®AZ(m),d) (AV,d) (C,d) we may extend id\v to a retraction irm : (AV g> AZ(m),d) —> (AV,d). (ii) This is a trivial consequence of (i). ? Corollary 1 Suppose for some ? > r > 1 that the non-zero elements of H+(AV,d) are concentrated in degrees i with r < i < n. Then ca.t(AV,d) < n/r .
386 29 LS category of Sullivan algebras proof: It is easy to construct a quasi-isomorphism (AV, d) -=» (B,d) in which B+ is concentrated in degrees i, r < i < n. Then nil ? < n/r. Apply the Proposition. D Corollary 2 m\H(AV,d) < e(AV,d) < cat(AV,d) < niL4 for any commuta- tive model (A, d) of(AV.id). In particular, if H{AV, d) is finite dimensional then cat(AK,d) < max{i|iT(AV, d)^0}. proof: The inequalities are immediate, the first two from the definitions and the third from the Proposition. Finally, if H(AV, d) is zero in degrees > ? then H(I) = 0 where / is the ideal defined by / = (AV)>n ? Un, with Un chosen so (AV)n = Un ® (kerd)n. Thus (AV,d) —> (AV/I,d) is a quasi-isomorphism and cat(AV, d) < nil(AV/7) < n. D Next we establish Proposition 29.4 Suppose (AV, d) is a rational Sullivan model for a simply connected space X with rational homology of finite type. Then cat(AV, d) = cat0 X and e(AV, d) = e0X . proof: We may suppose X is a rational CW complex, so that cat0 X = cat X (Proposition 28.1). Recall the quasi-isomorphism Cm : (AV <g> AZ(m),d) -=> (AF/A>ml/,d) defined at the start of this section. The Sullivan algebra (AF ® AZ(m),d) is a Sullivan model for the realization Y = \AV ® AZ(m),d\, as we showed in Theorem 17.10. Thus (AV/A>mF,d) is a commutative model for Y, which implies that clo^) < m (Theorem 29.1). Since Y is a rational CW complex, Y has the homotopy type of an m-cone (Proposition 28.3). Suppose cat(AV,d) = m. Then there is a morphism 7rm : (AV®AZ(m),d) —> (AV, d) such that nmXm = id. Thus \Xm\ \irm\ = id\Av,d\ and |AF,d| is a retract of Y. Since |AV, d\ ~Q X (Theorem 17.12) it follows that cat0 X < cat |AV, d\ < m. Conversely, if cat0 X = m then cat X — m and X is a homotopy retract of an m-cone ? (Theorem 27.10). Let (AV,d) A (AVP,d) ? (AV,d) be Sullivan representatives respectively for the retraction and the inclusion. Then ?? is a quasi-isomorphism and hence an isomorphism (Theorem 14.11). Moreover, Theorem 29.1 provides a quasi-isomorphism ? : (AVp,d) -=> (A,d) with nil.4 < m. Thus ca.t(AV,d) < m (Proposition 29.3). Finally, since e0X = e(X;Q) this is the least integer, r, such that there is a continuous map / of X into an r-cone such that Hr(f;Q) is injective (Proposition 28.1). Now a simplified version of the argument above shows that e(AV,d) = e0X. ? Corollary Suppose (AV, d) is any rational Sullivan algebra with V = {V1} .>0 a graded vector space of finite type. Then the spatial realization (§17) \AV,d\ is
Rational LS-Category 387 ? rational space satisfying cat IA V, d\ =cat(AV,d) . proof: According to Theorem 17.10, |AK, d\ is a simply connected rational topological space with (AV, d) as a Sullivan model. Thus cat |AV, d\ = cat0 |AV, d\ ¦¦ cat(AV, d), by Proposition 28.1 and Proposition 29.4 D Example 1 A space X satisfying cqX < e0X. Let X be a simply connected space with Sullivan model (A(x,y, z),d) where dega: = 3 = degy, dx = dy = 0 and dz = xy. Then the cohomology algebra H*(X;Q) has 1, [x], [y], [xz], [yz] and [^[yz] as basis, and so c0X = 2 . On the other hand, nil (A(x, y,z)) = 3 and so 3 > cloA' > cat0X > e0X. Finally, xyz G A3(x,y,z) represents a non-trivial cohomology class, so e0X > 3. Thus e0X = cat0 X = cl0 X = 3 . D Example 2 (Lemaire-Sigrist [111]) A space X satisfying e0X < cat0A'. Consider the commutative cochain algebra (A, d) given by A = ?(?, y, t)/(x4, xy, xt) , dx = dy — 0, dt = ?3 , with dega; = 2, degy = 3 and degi = 5. Evidently nil(A,d) =3 . Thus if X is a simply connected space with (.4, d) as commutative model then cato X < cl0 X = 3 . A vector space basis for A+ is given by x,x2,x3,y,t,yt and so a vector space basis for H+{A) is given by [x], [x]2, [y], [yi\. In particular. H(A) is concentrated in degrees < 7. A minimal Sullivan model (AV, d) for (A, d) has the form m : ? (?, y,z,t,...) —> A with mx = x, my = y, mz = 0, rat = t and dz = xy; the remaining generators all have degree at least 7. In particular A^3V = hzx2 ® (A-3F)>?. Since zx2 is not a cocycle no cohomology class can be represented by a cocycle in A-3V: = e(AV,d) = 2. Finally, we show that cat(AV, d) > 2, thereby establishing do (JO = cat0 (?') = cat(AV, d) = 3 .
388 29 LS category of Sullivan algebras Thus in particular, eo{X) < cato (A'). In fact, let ? : (AV,d) —> (AV/A>2V,d) be the projection. If cat(AF, d) < 2 then there is a morphism of graded algebras ? : H (AV/A>2V) —)¦ if(AF) such that ????(?) = id. Now t becomes a cocycle in AF/A>2K and so we would have [yt+x2z] ^ [y][t] hA 0. Since [yt+x2z] »—> [yi] in #(,4), [yi + ?2 ?) ? 0 and this contradiction proves cat(AV, d) > 2. ? Example 3 {Dupont [45]) A space X satisfying cato X < clo X. Let (L, d) be the free differential graded Lie algebra L(a, b, x, y) with deg a = 3, dege = 5, deg a; = 7, degy = 11 and da = db — 0, dx = [a,a], dy = [a,x] + [b,b]. Let (L', d) = (L(a', 6', x', y'M) be a copy of L and extend L(a, 6, x, y, a', 6', x', y') to the free dgl L(a, 6, a;, y, a', b', x', y', a", u, u', 6", w) by requiring da" = [a,a'}, du = [a,6'], du' = [a',b], db" = [b,b'} and dv = [[a, [b,b]}, 4[a',y'] + [?',?']] - 7 , where 7 G L(a,b,a' ,b',a",u,u',b") has bracket length 5 and satisfies dy = 4[[a,[b,b}},[a',[b',b'}]]. In [45] Dupont shows that if X is a simply connected space with Lie model (L, d) then clo X = 4 and cat0 X = 3 . Since the argument is lengthy and difficult, we do not reproduce it here. D Example 4 Formal spaces. Let X be a simply connected topological space with rational homology of finite type. If X is formal then co(X) = eo(Ar) = cato X = cl0 X . In fact, since X is formal H*(X) is a commutative model for X. Thus co(X) = nil if* (A') > clo X, by Theorem 29.1. The reverse inequalities are established in Theorem 28.5. D Example 5 Coformal spaces. Recall that a simply connected topological space X with rational homology of finite type is coformal if it has a Lie model of the form (L, 0) — Example 7, §24 (f). Equivalently X has a purely quadratic Sullivan model (AV,d\). We show for coformal spaces that e0p0 =cato(X) = doX . In fact, suppose eo(A') = r. Choose a vector space complement S for kerdi in ArV. Then / = S @ A>rF is an acyclic ideal and (AV/I, d) is a commutative model for X. Since m\{AV/I,d) = r we have (Theorem 29.1) that r > c\0X. The reverse inequalities are always true, as in Example 1. ?
Rational LS-Category 389 Example 6 Minimal Sullivan algebras (AV, d) with V = yodd ??? dim y < oo. If (AF, d) is as in the title of the example, let r = dim F. Then ? F = ? A'F, dimArF = 1 and the elements in ArV are cocycles and not coboundaries. Thus e(AV,d)=r = nil(AV,d) and hence e(\V,d) = cat(AF,d) = r. D Example 7 (AV, d) = A(a, 6, a;, y, z) with dx = a2, dy = b2 and dz = ab. In this example we take dega = 2, degb = 2, but any even degrees would do. Hence (AV, d) -=> Aa/a2 ® Ab/b2 ® Az, and this commutative model (v4,d) has nil A = 3. A non trivial cohomology class is represented by ab ? — xb2 and so e(AV, d) > 3. It follows that e(AV, d) = cat(AV, d) = 3. ? (c) The mapping theorem for Sullivan algebras. The Mapping theorem 2S.6 for topological spaces (§28(c)) has an analogue for Sullivan algebras. In the case Jk = Q and the spaces and algebras have finite dimensional cohomology in each degree Theorem 28.6 follows from the theorem for Sullivan algebras, which reads: Theorem 29.5 Suppose ? : (AV, d) —> (AW, d) is a morphism of minimal Sullivan algebras with V = {Vl}i>2 and W = {H/Z}i>2. (i) If ? is surjective then cat(AF.d) > cat(AW,d). (ii) If there are r elements of odd degree, W\,..., wr € W such that AW = Im ? · A(wi ,...,wr) then cat(AF, d) + r > cat (AW, d). proof: (i) Extend ? to a quasi-isomorphism ? : (AV <g> AZ, d) —> (AH^.d) from a minimal relative Sullivan algebra, and let ? : (AV <g> AZ, d) —> (AZ, d) be the surjection obtained by setting V = 0. Choose a quasi-isomorphism ? : (AW,d) ^ (AV ®AZ,d) such that ?? = id (Lemma 12.4). We first establish • The differential d in AZ is zero. B9.6) and • ?(?+\?) CA+V&AZ. B9.7) Indeed let (AV<8)AV, D) -^ Ik be a minimal Sullivan model for the augmenta- tion ? F —> 3z. Thus (§14(a)) ? extends to a quasi-isomorphism (AF® AF)®av (AV ® AZ) A (AZ,d). Moreover ??®? : (AV ® AV) ®av (AV ® AZ)_-=> (AV ® AV) ®av APV and so (AZ, d) ~ AW ®av (AV ® AV) = (AW ® AV, ?). The linear part, ?0, of ? is the map Q(ip)D0 : V —> W, where Q(ip) is the linear part of ?. Now (AF ® AF, D) is contractible, because it is a Sullh'an algebra quasi- isomorphic to k (Theorem 14.9). Thus Do : V -=> V and so ?0 is surjective. Write V = Fo ? Vi with Fo = kerJ0- Then recall (Example of §14(&)) that
390 29 LS category of Sullivan algebras the quotient differential in AV is zero. Thus a composite morphism 7 : (AW <S> ??, ?) —)¦ (AV, 0) —> (AV0,0) is defined by setting first W7 and then Fi to zero. By construction H(Q(j)) is an isomorphism and hence 7 is a quasi-isomorphism (Proposition 14.13). This shows that (AVo,0) is a minimal Sullivan model for (AW ® AV,6). It follows that (AZ,d) 9* (AFo,0), which proves B9.6). For the proof of B9.7) let ? : (AV,d) —> (AV®AZ,d) be the inclusion. Then ?? = ? = ??? and so ? ~ ?/></? (Proposition 12.9). Thus ?? ~ ?rcy</? : (AV, d) —> (AZ, 0). But ?? is the augmentation in AV and so the Example in §12(b) asserts that g\ = ???. In other words, gip(A+W) = ???(?+\?) = 0, which proves B9.7). Finally, we suppose cat(AV,d) = m and deduce that cat(AW,d) < m. Let (AK, d) - Am > A V ® AZ(m) (AV, d) AV/A>mV,d) be as described at the start of the section. Apply — augmentation ideals of this diagram to obtain (\AV]+®AZ,d) ({AV ® AZ(m)}+ ® AZ,d) (AV ® ??, d) to the ([AV}+®AZ,d ([AV/A>mV]+®AZ,d), where the vertical arrow is a quasi-isomorphism because (AV®AZ, d) is (AV, d)- semifree (Lemma 14.1 and Proposition 6.7). Now B9.7) states that ? factors as (AW, d) —> (Jb ? [AV]+ ® AZ, d) —> (AV <g> AZ, d). Thus it also factors as (AW,d) (Ik ®{AV ® AZ(m)}+ ® AZ,d) (AV®AZ,d) (? ? [AV/A>m V]+ «8» AZ, d) Since the lower cochain algebra has product length m this is exactly the situation envisaged in Proposition 29.3, which therefore asserts that c&t(AW,d) < m. (ii) This is proved by induction on r, the case r = 0 being precisely assertion (i). Assume deg^i < · · · < degu;r. Set degun = p and put (AZ, d) = (AW<p,d)®{AV<p,d)(AV,d) = (AW<P®AV^P,d). Then ? factors as (AV,d) ? (AZ, d) —> (AW,d) and ? is surjective. It follows that ????' · A(w\,... ,wT) = AW and (by assertion (i)) that cat(AV,d) > cat(AZ,d).
Rational LS-Category. 391 Thus it is sufficient to prove assertion (ii) for ?'; i.e., without loss of generality we may and do assume that ? is an isomorphism from AV<P to AW<P. Since V and W are concentrated in degrees > 2, and since (AW,d) is minimal, dwx is a cocycle in AM/<P. Extend ? to a morphism ? : (AV ® Au,d) —> (AW, d) by setting du = ip~1dwi and ??. = w\. By induction, cat(AU',d) < cat(AV ® Au, d) + r - 1. It remains to prove cat(AV ® Au, d) < cat(AV, d) + 1. For this, apply - ®av (AV ® Au,d) to the diagram of Proposition 29.3(i), with nil ? = cat(AV, d). Now ? ®av (AV ® Au) = ? ® Au. Since degu is odd, nil (B ®av7 (AV ® Au)) = nil ? + 1. Since — ®av" (AV & Au, d) preserves quasi- isomorphisms it follows immediately from Proposition 29.3(i) that cat(AV ® Au, d) < nil ? + 1 < cat(AV, d) + l. D Example 1 e ((AV,d) ® (AW, d)) = e(AK, d) + e(A\V, d). Let (AV,d) and (AH7, d) be any minimal Sullivan algebras. Then e(AV, d) is the least integer r such that (AV, d) —> (AV/A>rV, d) is injective in cohomology. Thus it is the greatest integer r such that A V-r contains a cocycle ? representing a non-zero class in H(AV). Let s = e(AW, d) and w G A-SIV be a cocycle representing a non-zero class in H (AW). Then v®w G A^T+S(V®W) and [v®w] ? 0, so e ((AV, d) ® (AIV, d)) > r + s. Conversely the surjection AV® AW —> AV/A>rV® AW/A>SW is injective in cohomology, and its kernel contains A>r+s(V®W). Thus e ((AV, d) ® (AW, d)) < r + s; i.e. e ((AV, d) ® (AW, cQ) = e(AV, d) + e(AW, d). ? Example 2 cat0 ?' — e0 X can be arbitrarily large. Let X be the space of Example 2, §29(b). Its minimal Sullivan model (AV, d) satisfies cat(AV, d) = 3 and e(AV, d) = 2, and has the form A(x,y, z,t,u,v,...) with dx = dy = 0, dz = xy, dt = x3, du = zy, dv = z2 - 2ux and with dega; = 2, degy = 3, degz = 4, degi = 5, degu = 6 and degi> = 7. Set u and y to zero to define a surjection (A(x, y, z, t, u, v),d) —> (A(x, z, t, v), d) with dx = dz = 0, dt = x3 and dv = z2; denote this quotient Sullivan algebra by (AW,d). Then H(AW,d) is concentrated in degrees < 8 while the remaining elements of V have degrees at least 8. It follows that the surjection above extends to a surjective morphism ? : (AV,d) —> (AW,d). Taking tensor products we obtain surjective morphisms ®?? : (AV,d)9n —» (AW,d)®n . The cocycle zx2 in A3 W represents a non trivial cohomology class. Thus e(AW,d) > 3 and e((AW,d)®n) > 3n (Example 1). Now apply the Mapping theorem 29.5 to obtain cat(AV,d)®n > ca.t(AW,d)®n > e(AW,d) > 3n .
392 29 LS category of Sullivan algebras On the other hand, again by Example 1, e(A V, d)®n = ? e(AV, d)=2n . Thus if X" = ?' x · · · x ?' (n factors), cato(A'n) > 3n and eo(Xn) = In . (In fact, it is not too hard to show that cato(A'n) = 3n.) D (d) Gottlieb elements. A Gottlieb element of degree ? for a minimal Sullivan algebra (AV, d) is a linear map / : Vn —> k that extends to a derivation ? of (AV,d), that is, a derivation of AV satisfying ?? = (~1)???. The Gottlieb elements form a graded subspace G.(AV,d) C Hom(V, Jfe). Now suppose X is a simply connected topological space with rational homology of finite type. Recall that the choice of a rational minimal Sullivan model m : (AV,d) -=» Apl(X) determines an isomorphism ?? : V A HomQ (tt»(A'q),Q) , as follows from Theorem 15.11. Proposition 29.8 The dual of ?? restricts to an isomorphism (§28(d)) (i) G.(AV,d) 4 (it) If (AW, d) is any minimal Sullivan algebra with W = {VFP}P>2 and ca.t(AW, a < m then G» (AH7, d) is concentrated in odd degrees, and dim G, (AW, d) < proof: (i) Write H*(Sn) - h @ he, where e G Hn{Sn) is dual to the funda- mental class of 5". Any continuous map g : Sn ? Xq —> Xq is represented by a morphism ? : (AV, d) —> H*(Sn) ® (AV,d) and conversely, any such morphism represents a continuous map (Theorem 17.15). The maps g such that g correspond to the morphisms ? of the form id in which case ? is automatically a derivation of degree —? in (AV, d). Thus / : Vn —> Q corresponds to a Gottlieb element of Xq if and only if f?Gn(AV,d). (ii) Suppose 0^/6 Gok(AW,d), and extend it to a derivation ? of (AW,d). Then ? = 0 in AW<lk and so dividing by A+W<2k · AW we obtain a derivation ? in (AW-2k,d) restricting to / in W2k. Choose w e W'2k so that
Rational LS-Category 393 O(w) = f(w) = 1. Clearly dw_= 0 and Ok(wk) = kl Thus wk cannot be a coboundary and m\H(AW-2k,d) = oo. Now apply the Mapping theorem 29.5 to conclude ca.t(AW,d) > cat (AW^2k,d) > nil H (AW^'2k, J) = oo. On the other hand if 0 ? f G G2n+\(AW,d), extend / to a derivation ? of (AW,d) and define a morphism ? : (AW,d) —> (Au,0) <g> (AW,d) of Sullivan algebras by ? : ? ?—> 1 <g> ? + u ® #? (here degu = 2n + 1). If </?!,...,</?r correspond to Gottlieb elements /i, ·.., /r of odd degree then form the composite morphism 0- ..oVl :(AW,d) —>(A(ui,...,ur),0)®(AW,d) —> (A(Ul,... ,ur),0) If the fi are linearly independent then this morphism is surjective and so cat(At-F, d) > cat (A(«i,..., ur), 0) = r. D Example 1 .4 non-trivial Gottlieb element. Let (j4, d) be the commutative cochain algebra defined by A = A(a, b, x, y) /abxy with dega = dego = dega; = 3 and dy = abx. The Sullivan model for (A,d) has the form (AV,d) — A(a,b,x,y,z,...) with dz = abxy. We show that the map f : y 1—> 1 is ? Gottlieb element for (AV, d). In fact extend / to a derivation ? of (AV, d) by first setting ?? = yx. Then note that elements of higher degree in V have degree at least 15, while H(AV, d) is concentrated in degrees < 11. Thus ? extends automatically to the rest of V so as to satisfy ?? + ?? = 0. On the other hand, suppose (AV, d) is a Sullivan model for a simply connected topological space X and let X —> Y be any fibration with fibre S5. It follows from §15(a) and a simple calculation that the inclusion S5 —> X is zero in rational homotopy. In particular the Gottlieb element / does not arise as the fibre of an S5-fibration. D (e) Hess' theorem. Fix a minimal Sullivan algebra (AV,d) such that V = {Vi}i>2 an<^ fix an integer n. Simplify the diagram in the introduction to (AV,d) in particular ? is a minimal Sullivan model for ?. Our goal is Theorem 29.9 (Hess — [90]) Assume there is a morphism ?:(??®??,?) —>(AV,d)
394 29 LS category of Sullivan algebras of (AV,d)-modules such that ?? = id. Then there is a morphism ?' : (AV <g> AZ,d) —> (AV, d) of cochain algebras such that ?'? = id; i.e., cat(AV,d) < m . Remark Theorem 29.9 holds even if V1 ? 0 and is so stated and proved in [90]. We limit ourselves to the case V1 = 0 because the proof is then simpler. D The proof of Theorem 29.9 requires a technical result about the structure of (AV®AZ, d) which we state as Proposition 29.10 immediately below. The proof, however, will be deferred to the next topic as only the statement is required for the theorem. Proposition 29.10 The Sullivan model (AV ®AZ, d) can be chosen so that Q(Z) = 0, ? = ? ?? and p>0 ? ? (i) d: Zp —> 1®(?*?)?_?? ? A>^m-^V ® Zp-q \®A>m+^m-^V, ? > 0. L,=o J (it) The quotient differential, d, in AZ satisfies d: Zp A (A2Z) nkerd, ? > 1, n > 1 . (Hi) The inclusion of (AV ® (it ? Zo),d) in (AV <g> AZ, d) is a quasi-isomorphism (Here (AZ), is the grading induced by the grading Zt.) proof of Theorem 29.9: Regard (AV ®(lk®Zo),d) as a (AV, d)-bimodule by writing U?AV®(]k®Z0) . Then define a product in AV ® (Ik© Zo) by setting A simple calculation shows that this product is associative and makes (AV <g> (Jfc ? Z0),d) into a commutative cochain algebra. We denote this cochain algebra by (A,d). It is immediate that ? : (AV, d) —> (A,d) is a morphism of cochain algebras as is the restriction 77^ : (A, d) —> (AV, d) of ?. We prove the theorem by constructing a morphism ? : (AV ® AZ, d) —> (A, d) of cochain algebras such that ?? = X. Then ?' = ?)?? '¦ (AV® AZ, d) —> (AV, d) is the desired retraction.
Rational LS-Category 395 First observe (Proposition 29.10 (iii)) that ? restricts to a surjective quasi- isomorphism (AV ®(k© ZQ),d) —> (AV/A>mV,d). Thus its kernel, /,. satisfies ? (I, d) = 0. Moreover, since ?(?0) = 0,1 = A>mV ? (AV ® Zo). Now we construct ? inductively to satisfy (? + ??.??) : Zp—> I , p>0. First, since Zo c I we may set ?? = ?, ? ? Zo. Suppose ? is constructed in AV ® ? (?<? ? Z<n), and simplify notation by writing W = Z<p @Z<n. Set U = (A2IF)P_! ? (?+V ® Wp) ? (A^mV ® W<p) ? A>mV . Then Proposition 29.10 (i) asserts that dZp C U. On the other hand, a trivial calculation (using the fact that ? preserves products and restricts to the identity in AV) shows that (? + ????)? C /. Thus for ? G Zp , (? + ????)?? is a cocycle in /. Since H(I) — 0 there is a linear map h : ?™ —> I such that (? + ????)?? = dhz. Extend ? to an algebra morphism ? : AV ® A(\V ? Zp) —> ? by setting ?? = hz !p[T}Ahz, ? G Zp. The equations and ????? = ——????? ,z G Z" p+ 1 ? p+ are immediate from the definitions and together give ??? = ???, ? ? Zp. Finally ? + ???? = ? ·' ? ? —> I. This completes the inductive step, and the proof of the theorem. ? Corollary cat(AV, d) < m if and only if there is a diagram of (AV,d)-module morphisms (AV,d) (P,d) (AV,d) (M,d) in which ?? = id, 7 is ? quasi-isomorphism and A>mV · M = 0. proof: The definition of cat(AV, d) < m provides such a diagram. Conversely given such a diagram we may (in the notation of Theorem 29.9) factor 70 as (AV,d) A (AV/A>mV,d) A (M,d). Then, because (AV ® AZ,d) is semifree, ?? lifts through the morphism 7 to yield a morphism of (AV, d)-modules ? : (AV ® AZ,d) —> (P,d) such that ?? ~ ?? (as morphisms of (AV,^-modules). Then ??\ ~ ??? = ? ? = ??. Since 7 is a quasi-isomorphism, ?? ~ a (Proposi- tion 6.4(ii)).
396 29 LS category of Sullivan algebras Set ? = ?? : (AV <g> AZ, d) —> (AV, d). Then ?? = ??? ~ ?? = »dAy. Now iiG?A)[l] = [1] and so ???A) = 1. Since 7?? is AV-linear. 7?? = id,\V. By Theorem 29.9 cat(AV, d) < m. D Example 1 Field extension preserves category. Let (AV,d) be a minimal Sullivan algebra with V = {V*} ·>9? and let ? be a field extension of k. Then (AV, d) <g> ? is a minimal Sullivan aTgebra over K. We show now that cat(AV, d) = cat ((AV, d) <g> K) . Suppose cat ((AV, d) <8> K) = m and let (AV, d) <g> ? 2 ^ (p, rf) E. - (?V, d) <g> ? (M, d) be morphisms of (AV, d) <8> K. modules as in the Corollary to Theorem 29.9. In particular, A>mV ¦ M = 0 and ?? = id. Let ? : k —> ? the inclusion and let ? : ? —> k be a it-linear map such that ?? = id. Then the identity of (AV, d) factors as (AV, d) -24 (P, d) -^ (AV, d) , and 7 is also a quasi-isomorphism of (AV, d)-modules. Thus the Corollary to Theorem 29.9 asserts that cat(AV, d) < m. Conversely, if cat(AV, d) = m we can apply — <8> ? to the diagram of the Corollary to conclude cat ((AV, d) <g> K) < m. D Example 2 Rational category of smooth manifolds. Let M be a simply connected smooth manifold with rational cohomology of finite type, and let (AW, d) —y Aor(M) be a minimal Sullivan model for the cochain algebra of C°° differential forms on M. We observe that cato M = cat(AT-F,d) . In fact, as we saw in §11, Adr(M) is connected by quasi-isomorphisms of commutative cochain algebras to Api(M; Q) <g> M. Thus if (AV. d) is a rational minimal Sullivan model for M then (AW, d) = (AV. d) <8> M. It follows now from Example 1 (and Proposition 29.4) that cat0 M = cato(AV, d) = caX(AW,d). D (f) The model of (AV,d) —» (AV/A>mV,d). Recall that Proposition 29.10 describes the structure of a Sullivan model ? : (AV®AZ, d) ^> (AV/A>m V, d) of the surjection ?, where (AV, d) is a minimal Sullivan algebra and V = {V1} .>o. Here we provide the proof, beginning with two preliminary steps.
Rational LS-Category 397 Step 1 The quadratic part of the differential. Let d\, be the quadratic part of the differential in AV. Construct a (AV, de- module, (AV® (k®M),di) by requiring that d\ : Mn —» A>mV ? (A+V <g> M<n) and that a = #(^) : ?" ? ? (A>mV ? (A+V ® M<n)) . Since F = {Vl}.>o any element of degree ? in A>mV ? (A+V <g> M) has the form ? = ? + y + ~z with ? G Mn, y G A+V <g> M<n~l and ? G A>mV. If ?? = 0 then ? is in the kernel of the map a above and so ? = 0. Then d(x + y) = 0 and so by construction, ? + y is a coboundary in A>mF ? (AF <8> AiT<n). It follows that ? (A>mF ? (AV <8> M), di) = 0. In particular, a quasi-isomorphism of (AV, d\ )-modules ?: (AV®(Jfc®M),di) A (AV/A>mV,d1) is defined by C(l) = 1 and ?(?) = 0. Bigrade AV by putting (AV)P'* = APV. Then di is homogeneous of bide- gree A,0). The inductive procedure above defines a bigradation in M : M = ? Mp'* so that d\ is homogeneous of bidegree A,0) with respect to the in- p>m duced bigradation in AV <g> (Jk ? ?), denoted by [AV <g> (jfe ? M)]p'q. We call ? the filter degree. The key observation is that ?>??·* = 0; i.e., Lemma 29.11 M = Mm·*. proof: As in the Example in §14(b) extend (AV, di) to a minimal relative Sullivan algebra (AV <g> AV, ?>)_with H(AV <g> AV, D) = Jk. As shown in that Example, D : V —> A+V <8> AV, and the linear part of D is an isomorphism V -^ V of degree 1. Filter (AV <g> AV, D) by the ideals A-PV <g> AV. In the resulting spectral sequence the differential in Eq is zero and so the isi-term is a Sullivan algebra (A V<8>A V, ?) extending (A V, d\ ). As with D the linear part of ? is an isomorphism V A V. It follows that (AV®AV, ?) is contractible and hence H(AV®AV, ?) = L· Note that ? is homogeneous of bidegree A,0) with respect to the bigradation (AV <g> AV)p·* = ApV <g> AV. Next recall (Proposition 6.7) that if (?,?) is any semifree (AV, di)-module then —<8>avN preserves quasi-isomorphisms. Thus we obtain quasi-isomorphisms (AV/A>mV <g> AV, ?) A [AV <g> (Jfe ? ?)] <8>?? [?V <g> AV] —> (k ? ?, 0) . It follows that k ? ? and H (AV/A>mV ® AV,5) are isomorphic as bigraded spaces. Finally, use the short exact sequence 0 —> (A>mV® AV,o) —> (AVtg)AV,J) —> (AV/A>mV®AV,5) —> 0
398 29 LS category of Sullivan algebras to obtain an isomorphism H+ (AV/A>mV <g> ??,?) A H (A>mV ®AV,o) of bidegree A,0). The source is concentrated in bidegrees (p, *) with ? <m and the target in bidegrees (q, *) with q > m. It follows that H+ (AV/A>mV ® AV, ?) is concentrated in bidegrees (m, *). D Step 2 A (AV,d)-semifree resolution for (AV/A>mV,d). The differential d in AV satisfies: (d - dx) : (AV)*·· —» (AV)^+2'*, since (AV)P·* is just ApV. We shall now extend (AV,d) to a (AV, d)-module (AV ® (it ? M),d) such that d-dx: [AV ® (Jfe ? M)]Pt' —» [AV ® (Jfe ? M)]^p+2'* . For this it is sufficient to suppose d constructed in M<n and to extend it to Mn. Observe first that any d-cocycle ? of degree n + 2 in [AV ® (Jfe © jif<n)]-m+3'* is the d-coboundary of an element in [AV ® (Jfe ? M<n)]-m+2>*. In fact, write ? = n+2 ? Xii where x; has filter degree r, some r > m+ 3. Then dz = 0 implies d\xr — i=r 0. Since H (AV ® (Jk ? ?), di) is concentrated in filter degrees < m it follows that xr = ???' with x' of filter degree r—1. In particular, x' G (A-2V ® ?)"+1? n+2 AV c AV®(Jfc©M<n). Now z—dx' = ? Vi witn 2/? of filter degree i. Continue i=r+l in this way to obtain ? = du with u G [AV (8) (Jfe ? M<n)]-m+2'*. We now extend d to M" as follows. If tu G M" then diu; G AV ® (Jk ? M<n), because V = {Vl}.>o. Thus ddity is a d-cocycle of degree ? + 2 in [AV®'(Jfe©M<n)]-m+3f·, because M = M··. This implies ddxw = du for some u G [AV <g> (Jk ? M<n)]-m+2'* as we observed just above. Extend d by setting dw = d\w — u as w runs through a basis of Mn. This completes the construction of (AV ® (Jk ? M), d). Since d(M) C A>mV© (AV ® M) a morphism C : (AV ® (Jfe ? M),d) —» (AV/A>mV, d) of (AV,d)-modules is defined by ? = g and ?(M) = 0. (Thus if we forget AV differentials ? coincides with the morphism ? of Step 1). In particular filtering by [AV ® (Jk ? ?)]-?·* and by A^pV/(A>mV). we obtain from ? a morphism of spectral sequences which at the J?i-term is just ??(?) = ? : (AV ® (Jfe ? M),dx) -> (AV/A>mV,d!) . Since ? is a quasi-isomorphism so is ?: ? is ? (AV, d) semifree resolution of (AV/A>mV,d). Finally, the same argument as used in the observation at the start of the construction of d gives
Rational LS-Category 399 Lemma 29.12 If w ? [AV <g> (k ? M)]>r'" is a d-cocycle and if r > m then w = du for some u G [AV <8> (k ? M)]-r'*. D We now turn to the proof of Proposition 29.10: Set Zo = A/, so that the (AV, d)-module of Step 2 is written (AV <g> (k © Zo),d). Extend d and the quasi-isomorphism ? of Step 2 (uniquely) to a morphism C : (AV <8> AZ0,d) —» (AV/A>m V, d) of cochain algebras. Next suppose by induction that this is extended to a morphism C·. (AV®A(Z<p(BZ<n),d) -> (AV/A>mV,d) of cochain algebras such that ? (?<? ? Z<p) = 0 and conditions (i) and (ii) of the Proposition -are satisfied. Simplify notation by writing W = Z<p ? Z<" and extend the quotient Sullivan algebra (AW, d) to a quadratic Sullivan algebra {A(W ? Z?),d) by requiring that d : Z? A (AHV)^ A herd. Let z G Z?. Then ddz = ? + y with ? G 0 A>9(m-1)V ® (?2^)^^! and y G ??^+^'?? <g> Wp-q-i . q<P q<P The component of d(x + y) in AV <8> A3H^ is just (id ? d)x. Since d(x + y) — 0 so is (id <g> d)x = 0. By (ii) of the Proposition, ? = (id <8> d)x' for some x' G 0A>9(m-l)y0Iy . P It follows that ? = d(dz—x') is a cocycle in AV®(k@W). Thus its component in AV <8> IV is an (id® d)-cocycle. By (ii) of the Proposition this component is in AV <8> Zq. Thus a careful inspection shows that ? G (?>1+?(??)?® ?? ©A>m+1+p(m~1)V . Apply Lemma 29.12 to conclude that ? = <2? with ? G (A>p(m-^V <8> Zo) @ i\>m+p{m-\)y Extend d by setting dz = d? - ?' — ? as ? runs through a basis of Zp. Extend ? by setting ? {Zp) = 0. This completes the inductive step and hence the construction of (AV <8> AZ, d) and of ?. To check that ? is a quasi-isomorphism it is sufficient to establish (iii): the inclusion of (AV <g> (k ? Zo), d) in (AV <g> AZ, d) is a quasi-isomorphism, because the restriction of ? to (AV <g> (it ? Zo), d) is a quasi-isomorphism (Step 2). Filter both sides by the submodules A-p V®— to obtain a morphism of spectral sequences with ?O-term the inclusion (AV 0 (k ? Zo), 0) —> (AV <g> AZ, id<g>d). It is thus sufficient to show that k ? Zo ^ (AZ, d). But in view of (ii) this is precisely the assertion in Example 7 of §12(d). D (g) The Milnor-Moore spectral sequence and Ginsburg's theorem.
400 29 LS category of Sullivan algebras Let (AV, d) be a minimal Sullivan algebra with V = {V*} . Filtering by the ideals Fp = A-PV yields a spectral sequence, the Milnor-Moore spectral sequence for (AV, d). This spectral sequence was introduced in §23(b) where it was shown that (Eo,do) = (AV,0) and where L is the homotopy Lie algebra of (AV, d). Moreover Eqo is just the asso- ciated bigraded algebra of H(AV,d): Eg _ FpH(AV)/Fp+lH(AV) , where FPH(AV) is the image of H(A^pV,d) in H(AV,d). On the other hand, e(AV, d) is the largest integer ? such that the image of H(A~rV,d) in H(AV,d) is non-zero. We restate this as Proposition 29.13 The Toomer invariant e(AV, d) is the largest integer r such that E^m(AV,d) ? 0. In particular, ??* (AV, d) = 0 p> cat(AV, d) . Finally we establish the Sullivan algebra version of a theorem of Ginsburg [64], for which a simple proof was later given by Ganea [62]. Theorem 29.14 Suppose (AV, d) is a minimal Sullivan algebra and V = {Vl}i>0. If cat(AV,d) = m then the Milnor-Moore spectral sequence collapses at some Ee, ?<m + l; i.e. Em+i = Eoo· proof: Recall the (AV,d)-semifree resolutions C(Jfe) = (AV ? (Jfe ? M(Jfe)), d) ^> (AV/A>kV, d) constructed in Step 2 of §29(f). We denote AV <& (k @ M(k)) simply by Q(k). Recall that Q(k) is bigraded with AW = (AV)P'* and M(k) = M(k)k-* (Lemma 29.11). Moreover, it follows from the construction that d : Q(k)P'* —» <2(a;)^p+1'*. Our first step is to construct a commutative diagram of (AV, c?)-modules of the form (Q(k + m),d) —i— AV/A>m+kV (Q(m),d)
Rational LS-Category 401 such that ? : Q(k + m)p'* —> Q(m)-P'* for all ? and ? is the identity on AV. (Here ?A) = 1 !) For this it is sufficient to construct ? in M(k + ??) and we suppose by induction this is done in M(k + m)<n. Let ? G M(k +m)n. Then ?? G [AV <g> (Jk ? ?/(A: + m)<n)]>fc+m'*. Thus ^dz is defined and ??? G Q(m)>k+m'*. By Lemma 29.12 we may write ??? — au with u G <2(m)-fc+m"\ Extend ? by setting y?z = u. This completes the construction of ?. Since ? and ? preserve nitrations ?? (M(k + m)) = 0 = ?? (M(k + m)) and thus the diagram commutes. Now suppose caX(AV,a) < m. Then there is a morphism ? : (Q(m),a) —> (AV, ?) of (AV, d)-modules extending the identity on (AV, ?). In particular, ? : Q(m)p·* —> A>p~mF. It follows that ?(?)?? vanishes in M(k + m) and thus the diagram Q(m + k) ? Q(m) gC(m+k) AV commutes. Finally let z G ??*'*, some i > m. Then 2 is represented by an element w G A-fcy such that aw 6 A-*+fy. To show ?(? = 0 it is sufficient to find a cocycle w' G A-kV such that ty - w' G A>Ay. But since aw ? A>k+mV, ?(?? + k)w is a cocycle in AV/A>k+mV. Write ?(?? + k)w — C(m + k)u for some cocycle u G Q(m + k). Then g(k)w = ??(?? + k)w = ??(?? + k)u = ?(^????. Thus the cocycle w' = ???, has the desired properties, since g(k)w' = g(k)w. This shows ?( - 0, ? > m + 1 and so ??? = 2?m+i. ? (h) The invariants meat and e for (AV, d)-modules. Fix a minimal Sullivan algebra (AV,a). This topic relies heavily on §6, often without explicit reference. In particular morphism means morphism of (AV, ?)- module and homotopy is the relation defined at the start of §6(a). For any (AV, d)-module, (?, ?), we may then make the following constructions: • a semifree resolution (?, ?) -^ (?, ?). • the surjections g(k) : (P, a) —» (P/A>kV -?,?). • homotopy commutative diagrams of (AV, c?)-module morphisms, (P(k),d) (P,d) (P/A>kV-P,a) , where c(k) is a semifree resolution of (P/A>kV *P,d).
402 29 LS category of Sullivan algebras These constructions are unique 'up to quasi-isomorphism'. Indeed, if (Q, d) -=* (M, d) is a second (AV, d)-semifree resolution then there is a quasi-isomorphism a : (Q, d) -=» (P, d) of (AV, d)-modules. Moreover id®Ava : AV/A>kV®AV Q —» AK/A^V ®Av ? is also a quasi-isomorphism, and this is just the quotient map a(k) : Q/A>kV · Q —» P/A>kV-P. Thus a(fc) lifts to a quasi-isomorphism ?(k) : (Q(k),d) —» (P(&), d) such that Q(k)?(k) ~ a(A;)C(A;). Automatically then ?(*)? ~ ?(k)\(k). In particular, the following definition is independent of the choice of semifree resolution and constructions above: Definition (i) The module category of (M,d), mcat(M, d), is the least integer m for which there is a morphism ? : (?(m), d) —> P(d) such that ?\(??) ~ id. (ii) The Toomer invariant of (M,d), e(M,d), is the least integer r such that H(g(r)) is injective. Proposition 29.15 For a minimal Sullivan algebra (AV,d) and any (AV,d)- module, (M,d): (i) e(M,d) < mcat(M,d). (ii) mcat(M,d) < mcat(AV,d). (Hi) If V = {^*}i>2 ???? mcat(AF,d) = cat(A7,d). proof: (i) This is immediate from the definitions. (ii) Suppose mcat(AV, d) = k and construct a homotopy commutative diagram (AV,d) * ~ (N,d) (AV/A>kV,d) and a morphism ? : (N,d) —> (AV, d), such that ? is a semifree resolution and ?\ ~ id. Let (P, d) be a semifree resolution for (M,d), and apply — ®AV ? to this diagram. This yields (P,d) ^ (N®AVP,d) ?- (P,d) Q (P/A>kV-P,d) with 77'?' ~ id. It follows that mcat(M, d) < k.
Rational LS-Category 403 (iii) Let C : (AV(g>AZ,d) —> (AV/A>mV,d) be a minimal Sullivan model for the surjection ?. It is in particular a (AV, d)-semifree resolution. If cat(AV,d) = m there is a morphism ? : (AV <g> AZ,d) —» (AV,d) of Sullivan algebras restricting to the identity in (AV,d). In particular, ? is AV-linear and so mcat(AV, d) < m. Conversely, if mcat(AV, d) = m then there is a morphism ? : (AV<g> ??, d) —> (AV, d) of (AV, d)-modules such that ?? ~ id, X denoting the obvious inclusion. But this implies ?(?\) = id and so 7??A) = 1. Since ?? is AV-linear, ?? = id. Now Hess' theorem 29.9.asserts that cat(AV, d) < m. D Finally, suppose (M,d) is a (AV,d)-module. The dual (AV,d)-module, Hom(M, L·), of J;-linear functions is defined by df = -(-l)desffod and (?·/)(?) = (-1)??6? for / G Hom(M, Jt), ? G AV and ? G M. Theorem 29.16 [55] Let (AV, d) be a minimal Sullivan algebra such that V — {} and has finite type. Then cat(AV, d) = e (Hom(AV, Jfe), d) . Lemma 29.17 If a quasi-isomorphism a : (Q,d) —> (N,d) of (AV, d) -modules factors as (Q,d) —> (iV',d) —> (iV,d), un?/i A>rV-iV' = 0, i/ien mcat(Q, d) < r . proof: We lose no generality in assuring (Q,d) seniifree. It follows from our hypotheses that a factors as (Q,d) -^4 (Q/A>rV-Q,d) A (N,d). This yields a homotopy commutative diagram (Q,d) in which Q(r) is a semifree resolution. Since ??? ~ ?g(r) = a and 'q is a quasi-isomorphism it follows that 7?? ~ id. Thus mcat(Q,d) <r. ? proof of Theorem 29.16: Denote (Hom(AV,J:), d) by (M,d). The finite type restriction implies that (AV, d) = (Hom(M, Jk),d). Since cat(AV, d) = mcat(AV, d) > e(M, d) — Proposition 29.15 — we have only to show that e(M, d) > mcat(AV, d) .
404 29 LS category of Sullivan algebras Let ? : (P,d) -^ (M,d) be a semifree resolution. Since (Hom(M,]k),d) = (AV, d) this dualizes to a semifree resolution Hom(</?, k) : (AV, d) -=» Hom(P, k). Let ? = Hom(y?,A)l G Hom(P, jfc). Now suppose e(M,d) = r. Then the surjection ? : (P,d) —» (P/A>rV -P,d) is injective in homology. It follows that the dual, Hom(p, A), induces a surjection ? (Hom (P/\>rV · P, A)) H· H (Hom(P, Jfe)) . In particular there is a cocycle / G Hom (P/A>TV · P, k) such that [f ? ?] = [?]. The quasi-isomorphism AV ^» Hom(P, It) is given by ? ?—> ?·?. Thus it factors as (AV,d) A (Hom (P/A>rV-P,k),d) Hom<>e'*\ (Eom(P,k),d) where ?? = ? · /. Now for any g G Hom (P/A>rF · P, k), x G P/A>rV ¦ ? and ? GA>rFwehave(*-p)(x) = ±g($-x) =0. Thus A>rF- Hom (P/A>rV-P,k = 0 and Lemma 29.17 implies that mcat(AV, d) < r. D Corollary Let X be a simply connected topological space with rational homol- ogy of finite type, and rational minimal Sullivan algebra (AV,d). Then proof: Apply Proposition 29.4.
Rational LS-Category 405 Exercises 1. Suppose that the homogeneous space M = SpE)/SUE) admits a minimal model of the form (A(a,b,x,y,z),d) with deg ? = 6, deg b — 10, deg ? = 11, deg y = 15, deg ? = 19, da = db — 0, dx = a2, dy = ab and dz = b2. Compute 2. Let F be a coformal space and X = ??? en. Supposing catF — ? > 1. Prove that catX < catF. 3. Using 24.5, 24.7 and 29.5, prove that any sub Lie algebra of a free graded Lie algebra is free. 4. Let X — (CP2 V S2). Denote by ?, ? and 7 generators, respectively, of 7T5(CP2), 7T2(S2) and vr2(CP2). Consider the spaces Y = X U[a>0] e7 and ? = CP2 ? S3. Using minimal models prove that the map / : CP2 V53 -> F which restricts to the identity on CP2 and represents the class [?, y] when restricted to S3, extends, once localized, to a continuous map / : Zq -» Yq such that ?»/ is injective. Prove that catoF = 3. Deduce that there is no mapping theorem for the invariants eo and for the rational cup length.
30 Rational LS category of products and fibra- tions In this section the ground ring is an arbitrary field h of characteristic zero. Lusternik-Schnirelmann category is not well-behaved on products and fibra- tions. For example, a simple argument shows that for topological spaces Y and max(cat Y, cat Z) < cat(F xZ)< cat Y + cat ? , and inequalities are sharp: for each one there is a non-trivial example where the inequality is an equality. For general fibrations ? : X —> Y with fibre F, moreover, the best that can be asserted is 0 < cat A" < (cat Y + l)(cat F + 1) - 1 . This section analyzes the behaviour of the rational category of products and fibrations. For products cat0 behaves well: we establish the recent result with Lemaire [55]: Y and ? are simply connected with rational homology of finite type then cato(F ? Z) = cato Y + cat0 ? . (The case ? = Sn had been established earlier by Hess [90] and Jessup [98].) In the case of fibrations of simply connected spaces with rational homology of finite type, the inequalities above remain sharp if cat is replaced by cato- With reasonable assumptions, however, they can be improved. For example, if F is an odd sphere then cato A' < cato Y + 1 , while if cato F < oo then cato A' > cato F - dim d, (?????(^) <8> Q) , <9» ® Q : vr,(F) ® Q —> n.-i(F) <8> Q denoting the connecting homomorphism. Finally, we strengthen the Mapping theorems 28.6 and 29.5 for the special case of a fibre inclusion. This section is organized into the following topics: (a) Rational LS category of products. (b) Rational LS category of fibrations. (c) The mapping theorem for a fibre inclusion. (a) Rational LS category of products. We begin with the classical Proposition 30.1 If Y and ? are normal topological spaces then max(cat K, cat Z) < cat(F ? ?) < cat Y + cat ? .
Rational LS-Category 407 proof: The first inequality follows from the fact that Y and ? are retracts of ? ? ? (Lemma 27.1). For the second, suppose cat Y = m and cat ? = ?. Then Y and ? are respectively retreats of an m-cone P' and an ?-cone Q' (Theorem 27.10). We may write ?' ~ ? = {p0} C Pi C - - - C Pm with Pk+l = Pk Ufk CAk, and Q' ~ Q = {q0} C Qi C · - - C Qn with Qe+l = Qe Ugt CBt. It is clearly sufficient to prove cat(P x Q) < m + ?. Set(PxQ)r = U PkxQe. Then (PxQ)r-(P xQ)r-i is the disjoint union k+t=.r of the contractible open subsets (Pk - Pk-i) x (Qr-k ~ Qr-k-i) of (? ? Q)r; thus it is contractible in (? ? Q)r- Let pk 6 Pk — Pk-i and qe ? Qe — Qe-i be ? the vertices of the cones CAk and CBt and set U = (? ? Q)r — [J (pk,qr-k)· k=0 Since (? ? Q)r = ((? ? Q)r - (? ? Q)r_i) U U it follows that cat(P ? Q)r < catU + 1. On the other hand, U deformation retracts onto (? ? Q)r-\ with deformation ? : U x I —> U given explicitly as follows: If (p, q) 6 (? ? Q)r-i then H(p,q,t) = (p,q). If ? = (?, s) 6 CAk and q = (b,s') 6 CBr-k then s + s' > 0 and we set Thus t/ ~ (PxE)r_! and cat U = cat(PxQ)r_!. Now we havecat(PxQ)r < cat(P ? Q)r-i + 1 and hence cat(P ? Q) <n + m. D The following classical example shows that the inequality of Proposition 30.1 can be strict. Fix a prime ? and let Mn{p) = Sn U/n CSn where /„ : 5" —> 5" induces multiplication by ? in homology. Clearly we may take fn+\ to be the suspension ?/? of/„ and so Mn+1(p) ^ ???(?). In particular, catMn(p) = 1, n> 2. Now if c is a second, different prime the inclusion i : Mm(p) V Mn(q) —> Mm(p) ? Mn(q) induces a homology isomorphism (easy calculation). Thus the Whitehead-Serre theorem 8.6 asserts that i is a weak homotopy equivalence. But these spaces are CW complexes and so i is a homotopy equivalence (Corol- lary 1.7). Thus cat(Mm(p) ? Mn{q)) = cat (Mm(p) V Mn(q)) = 1 . Recently, Iwase [95] has constructed a 2-cell complex Y such that cat(K x 5") = cat Y = 2 , thereby providing a considerably more dramatic example. Rationally, however, we have
408 30 Rational LS category of products and fibrations Theorem 30.2 [55] (i) If Y and ? are simply connected topological spaces with rational homology of finite type then cato(F ? Z) = cato Y + cat0 ? . (ii) If (AV,d) and (AW,d) are minimal Sullivan algebras with V = {Vz\ and W — {W%} i>2 9r<ided vector spaces of finite type then cat ((AV, d) ® (AW, d)) = cat(AV, d) + cat(AW, d) . proof: The category of the rational Sullivan minimal models for ?, ? and ? ? ? is just the rational LS category of ?, ? and ? ? ? (Proposition 29.4). Moreover the models for Y and ? satisfy the conditions of (ii) (Proposition 12.2) and their tensor product is the model for F ? ? (Example 2, §12(a)). Thus (i) follows from (ii) when Hz = Q. For (ii) we first let (M,d) and (N, d) be respectively any (AV, d)-module and any (??47, d)-module and notice that (M, d) ® (TV, d) is a (AV, d) ® (AW, d) in the obvious way: (? ® ?) -(m ® n) = (-l)dee* deem<? . m ® ? · ?. We show that e ((M, d) ® (N, d)) = e(M, d) + e(N, d) ; C0.3) the theorem then follows from Theorem 29.16 which asserts that cat(—) = e(Hom(—,&)) for Sullivan algebras satisfying the hypotheses above. It remain to prove C0.3). We may suppose (Ai, d) and (N,d) are respec- tively (AV, d)- and (AW, d)-semifree; in this case the tensor product is obvi- ously (AV, d) ® (AW, d)-semifree. If e(M, d) = m and e(N, d) = ? then the surjection ? : (M,d) ® (N,d) —> (M/A>mV-M,d) ® (N/A>nW · N, d) is in- jective in cohomology. Since kex-? D A>m+n(V ? TV)-(M ® N) it follows that e((M,d) ® (N,d)) < m + n. On the other hand, if u 6 A^mV-M and ? e A>nW · ? are cocycles representing non-trivial cohomology classes in H(M) and H(N) then u ® ? 6 A^m+n(V ? TV) -(M ® N) represents a non-trivial co- homology class, so e ((M, d) ® (iV, d)) > m + n. D (b) Rational LS category of fibrations. In this topic we consider both topological spaces and Sullivan algebras. Thus ? ? ? C0.4) will always denote a fibration with fibre F in which ?, ? and F are simply connected. Similarly, (AV, d) —> (AV ® AW, d) C0.5) will always denote a minimal relative Sullivan algebra in which (AV, d) is itself a minimal Sullivan algebra and V and W are concentrated in degrees > 2. We
Rational LS-Category 409 recall from Theorem 15.3 that if k = Q and C0.5) is a Sullivan model for C0.4) then the Sullivan fibre (AW, d) is a Sullivan model for F and that the connecting homomorphism d* <8> Q : vr,(F) <g> Q —>¦ ?*_!(^) ® Q is dual up to sign to the linear part of the differential do : W —> V (Proposition 15.13). Finally, in this case Proposition 29.4 asserts that cat(AF, d) = cat0 Y, cat(AV®AW, d) = cat0 X and cat (AW, d) = cat0 F. These results and notation will be used without further reference throughout this topic. Example 1 The relative Sullivan algebra (A(x, y), dy = xr+1) —> (A(x, y, u, ?), dy = xr+1,dv = un+1 -x). In this example we take ? and u to be cocycles respectively of degrees 2(n +1) and 2. The quasi- isomorphism ? (?, y) —> (Ax/xr+1,d) shows that H (A(x,y),d) is a commutative model for (A(x,y),d). Thus (Corollary to Proposition 29.3) cat(A(x,y),d) = nil (Ax/xr+1) = r. Similarly (A(u,v),d) -^> Au/un+l and (A(x,y,u,v),d) -^ (Au/u(r+1)(n+1),0). Thus cat(A(x,y,u,v),d) = (n + l)(r + l) - 1 = (cat (A(x, y), d) + 1) (cat (A(u, ?), d) + l) - 1 . Proposition 30.6 The fibration ? : X —> Y C0.4) satisfies catX < (caty + l)(catF+l) -1 and this inequality is best possible, even for rational spaces. proof: Let cat F = m so that Y is the union of m + 1 open sets Ua each contractible in A'. The inclusion \a of p~1(Ua) in A" is then homotopic to a map P~l{Ua) —> F M- A' and it follows that catAa < catj < cat F (Lemma 27.1). Thus ?~?(???) is the union of (n + 1) open sets each contractible in A', and so catX + 1 < (catF + l)(catF + l). To see that this inequality can be sharp recall from Proposition 17.9 that spatial realization | | converts a relative Sullivan algebra to a Serre fibration. Thus in Example 1, \A(x, y, z, u, v),d\ —> \A(x,y),d\ is a Serre fibration with fibre |?(??, ?), d\. Denote \A(x, y), d\ by Y and convert this to a fibration A" —> Y with A" ~ \A(x,y,u,v),d\ and fibre F ~ \A(u,v),d\. Now by the Corollary to Proposition 29.4 the LS category of the realization is the category of the model. Thus Example 1 translates to = (r + l)(n + l)-l = (catY + l)(catF + 1) - 1 . ? Under reasonable hypotheses it is possible to improve considerably on Propo- sition 30.6 for rational LS category. First we have
410 30 Rational LS category of products and fibrations Proposition 30.7 (?) If dim W is finite and W is concentrated in odd degrees then cat(AV ® AW, d) < cat(AV, d) + dim W = cat(AV, d) + cat(AW, d) . (ii) If ir*(F) ® Q is finite dimensional and concentrated in odd degrees and if Y has rational homology of finite type then cato X < cato Y + dim ?,(F) ® Q = cat0 Y + cat0 F . proof: (i) The inequality is just the Mapping theorem 29.5(ii) applied to the inclusion (AV, d) —> (AV ® AW, d). The equality dim W = cat(AW, d) is Exam- ple 6, §29(b). (ii) is an immediate translation of (i). ? Proposition 30.8 (Jessup [98]) (i) IfH(AV®AW,d) ^% H(AW,d) is surjective then cat(AV ® AW, d) > cat(AV, d) + nil H (AW, d) . (ii) If H*(X;Q) —> H*(F;Q) is surjective and if Y and F have rational ho- mology of finite type then cato X > cat0 Y + c0F . proof: Put H = H(AW, d) and choose a linear map ? : (?, 0) —> (AV®AW, d) such that ?(?)?(?) = id. Then ? extends to a morphism id·? : (AV,d) ® (H,0) —> (AV®AW,d)of (AV, d)-modules. It preserves the filiations (AV)^p® ? and (AV)-P ® AW and the resulting morphism of spectral sequences is an isomorphism of .&>-terms. Thus id · ? is a quasi-isomorphism. It follows that there is also a quasi-isomorphism ? : (AV ® AW, d) —> (AV ® H, d) of (AV, d)- modules because (AV ® AW, d) is also (AV, d)-semifree. Suppose nil ? (AW, d) = r and let ? 6 ? be the product of r cohomology classes in H+. Each of these is the image of a cohomology class in H+(AV ® AW, d). The product of representing cocycles is thus a cocycle ? 6 AV ® AW such that ?? — 1 ® ? 6 A+V ® H<deguJ. Now define AV-linear maps a : (AV,d) —> (AV ® AW, d) and (id® /) : (AV,d) ® ? —> (AV,d) by setting ?(?) = ? ·? and requiring /(u>) = 1- Then (id® ^?? = 1 and hence (id ® f) ? ? ? ? = ic
Rational LS-Category 411 On the other hand, write AV <g> AW = AZ, with ? = V ? W. Since ? is the product of r cocycles of positive degree, ? 6 A-rZ. Thus for any m, ? induces a morphism ? : (AV/A>mV,d) —> (AZ/A>m+rZ,d). This leads to the diagram (AV, d) \v (Pv,d) (AZ,d) (Pz,d) in which ?? and ?? are respectively (AV, d)- and (AZ, d)-semifree resolutions and ??, ?? and /? are respectively 'homotopy lifts' of ??, Qz and ???· Thus ????? ~ CzAzu: as (AF, d)-linear maps; since ?? is a quasi-isomorphism, /3?^ ~ ?^? (Proposition 6.4(i)). Finally, observe that since [?] maps to the non zero cohomology class ?, [?] ? 0 and cat(AZ, d) > e(AZ, d) > r. Suppose cat(AZ, d) = m + r. Then there is a morphism ? : (Pz,d) —> (AZ,d) of (AZ, d)-modules such that ??? = id. Set ? = (id ® ?)??? : (Pv,d) —> (AV, d). Then in the diagram (AV,d) (AV/A>mV,d) we have ??? = (id® /)????? ~ (id® /)????& = (id® /)?? = id\v- Thus the Corollary to Theorem 29.9 asserts that cat(AV, d) < m, i.e. cat(AZ, d) > cat(AV,d)+n\\H(AWj). This establishes (i), and (ii) is an immediate translation. D (c) The mapping theorem for a fibre inclusion. In the case of a fibre inclusion the Mapping theorems 28.6 and 29.5 can be strengthened as follows.
412 30 Rational LS category of products and fibrations Proposition 30.9 (i) Ifca,t(AW,d) < oo then d0 = 0 in Weven. Conversely, if do = 0 in Weven, then dim Im d0 < cat(A W, d) < cat(AF <g> AW, d) + dim Im d0 . (ii) Suppose Y and F have rational homology of finite type. If cat0 F < oo then 9. ®Q = 0 in ????(?) <g>Q. Conversely, if d* ®Q = 0 m 7rodd(r) ®Q, dim Im(9* <g> Q) < cat0 -F < dim Im(9* <g> <Q>) + cat0 X . proof: Divide (AV®AW, d) by A-'2V®W to obtain a quotient cochain algebra ((k <8> V) <8> AW,<5). If {ui} is a basis of F then E = id <8> d + ??* <g> oi and the #i are derivations of (AW, d). Thus if ni = deg#* then ?? restricts to a Gottlieb element fi : Wni —> k. Now suppose dou>i = Vi, 1 < i < r. Then fi(wi) = 1 and fj(wi) = Q, j ? i- Thus Proposition 29.8 (ii) asserts that if cat (AW, d) < co then each ?, is odd and r < cat(AW,J). It follows that d0 = 0 in Weven and dim Im do < catiAW, d). The second inequality of (i) is just the Mapping theorem 29.5(ii). (ii) is an immediate translation. D Example 1 Fibrations with catX = 0. If cat X = 0 then X is contractible. Modify the constant map PY —> pt —> Y to a fibre preserving map PY —> X, which is (automatically) a homotopy equivalence and so induces a weak homotopy equivalence flY —> F. This map is then a rational homotopy equivalence and so cat0 ?.? = cat0 F. Thus we have two possibilities (Example 3, §28(d)): either cat0 F = oo or else ?,(?7) <g>Q is finite dimensional and concentrated in odd degrees. In the latter case, 7r„(F) <8> Q is finite dimensional and concentrated in even degrees; thus Y has a Sullivan model of the form (AFeven,0). In particular, cato(F) = oo: cat X = 0 ==$¦ max (cato Y, cat0 F) = oo . D Example 2 Fibrations with cat X = 1. We construct examples of fibrations with catX = cato X — 1, and cato ^ = ? and ? < cato F < ? + 1 , for any ? > 1. For this, let pt : Sj —> Sf be a copy of the Hopf fibration, with ? ? ? fibre Sf, 1 < i < n. Convert the composite map V Sj —> ? 5/ —> ? ^ into i=l i=l i=l a fibration ? : X —> Y with fibre F : Y = f[ S* and ? ~ V 5Z7. In particular cat X — cato X = 1, and cat0 Y = cat ? = ?.
Rational LS-Category 413 Now ?. E*) ® Q = Qea ® Qe'a with degea = 4 and deg< = 7 (Example 1, §15(d)) and e'a is in the image of ?,(??) <g> Q by construction. Thus each e'a G Im ?* (?) <g> Q. On the other hand ?' is 6-connected and so <9* <8> Q is a injective in tt4(F) <g> Q. It follows that ker(<9* <g> Q) = ?7(?) <g> Q and Proposition 30.9(ii) asserts that ? < cat0 F < ? + 1. D When the fibration ? : X —> Y has a cross-section we obtain a lower bound for cato X: Proposition 30.10 // the fibration ? : X —> Y has cross-section then cato X > max (cat0 Y, cat0 F) = cato(F V F). proof: Recall from the Example in §28(a) that max (cato Y, cato F) = cato(Y V F). Let s : Y —> X be the cross-section: ps = ???. Thus exhibits F as a retract of X so cato Y < cato X and it exhibits vr*(p) as surjective, so the inclusion j : F —> Y induces an injection 7r*(j) ® (Q> of rational homotopy groups. By the Mapping theorem 28.6, cato F < cat0 X as well. D The next example shows that Proposition 30.10 is best possible. Example 3 Fibrations with cato X = max (cato Y, cato F). For any simply connected topological spaces Y and ? convert the map (idy, const.) : Y V ? —> Y into a fibration ? : X —> Y with A'~KV Z, as described in §2(c). The inclusion Y —> ? \? ? then defines a cross-section s of this fibration; in particular the fibre F is simply connected. Moreover, the inclusion ? —> ? ? ? defines an inclusion i : ? —> F. Let j -. F —> X be the inclusion. Since the fibration has a cross-section, ?» (j) <8> Q is injective. Since ji is homotopic to the inclusion ? —> Y V ?, ?» (ji) ® Q is injective. Hence ?» (i) <g>Q is also injective and the Mapping theorem 28.6 asserts that cato ? < cato F < cat0 X - Thus cato X = max (cato Y, cato Z) = max (cato Y, cato F). D Exercises 1. Let F ? X A F be a fibration in which ??, F and F are simply connected. Prove that cat A' < (cat i + 1) ¦ (cat ? + 1) — 1. 2. Let X be a simply connected space and let ? be a prime number. Consider the Moore space Mn(p) = Sn Up en+1. Prove that if X is a rational space, the inclusion X V Mn(p) -> ? ? ??(?) is a homotopy equivalence. Supposing that tf+(X;Q) ^ 0. Prove that cat(X ? ??{?)) ? cat(X) + cat(Mn(p)). 3. Let F ? ? ->· F be a fibration in which A', F and F are simply con- nected. Suppose that dimH>n(F;Q) = 0, and that there exists an element ? g Hn(X;Q) such that ??(?)(?) f 0. Prove that cato(I) > cato(F) +1.
414 30 Rational LS category of products and fibration 4. Let / : X ->· Y be a continuous map between simply connected CW complexe of finite type. Suppose that K2n(f) ® Q is injective for ? > 2. Prove that cato/ < catoX < cato/ + dim (ker7rodd(/)
31 The homotopy Lie algebra and the holonomy representation In this section the ground ring is an arbitrary field h of characteristic zero. Recall from §21 (d) that the homotopy Lie algebra, Lx, of a simply connected topological space X is the graded vector space ?,(??) <8> k, equipped with a Lie bracket defined via the Whitehead product in ?,(?). In particular, (cf. Example 2, §21) each ? 6 Lx determines the linear transformation adz : Lx —> Lx, given by ad ? : y >—> [?, y] , y 6 Lx . In this section (Theorem 31.17) we show that • //cato X < co then for each non-zero ? 6 (Lx)even of sufficiently large de- gree there is some y 6 Lx such that the iterated Lie brackets [x, [x, [x,... [x, y]... ]]] are all non-zero. A linear transformation ? : V —> V in a graded vector space is called locally nilpotent if for each ? 6 V there is an integer n(v) such that ??^? = 0, and an element ? in a graded Lie algebra is called an Engel element if adz is locally nilpotent. Thus we may restate the result above as: z/cato X < co then the non- zero Engel elements in (Lx)even are concentrated in finitely many degrees. In particular, if cato X < co and if (dim Lx)even is infinite then Lx is not abelian. We shall use this to show cato X — co in an example in which eo(X) = 2 and the Milnor-Moore spectral sequence for X collapses at the Ez-term. The theorem above follows as a special case of a theorem of Jessup [99], which will be our principal objective in this section. For this we consider a Serre fibration p:X —> Y with fibre F at a basepoint yo 6 Y, and such that Y is simply connected with rational hoinology of finite type. Associated with this fibration is a right repre- sentation (§23(c)) of the homotopy Lie algebra LY in H*(F;jk), the holonomy representation for this fibration. This representation, which is an important invariant of the fibration, is con- structed as follows from the space ? ?? ?? of §2(c). First, the right ac- tion of UY on X xY PY makes H* (X xY PY;Jk) into a right ?*(??;??- module as described in §8(a). Second, there is a natural weak homotopy equiv- alence j : F —> ? ?? ?? (Proposition 2.5) and we use the isomorphism H*(j;fc) *? identify H9(F;3c) as a right H,(ilY;Je)-modu\e. Finally, since Hm(QY;Jk) = ULY (Milnor-Moore Theorem 21.5) the action of H*(QY;k) in Hr(F; Jk) restricts to a right representation of Ly. Definition This representation of Ly in H, (F; Ik) is the holonomy represen- tation for the fibration ? : X —> Y.
416 31 The homotopy Lie algebra and the holonomy representation Observe now that any ? 6 Ly determines two linear transformations: ad ? : LY —> LY and hi ? : H*(F; k) —> #.(F; A) , where hi ? is simply the restriction of the holonomy representation to x. Jes- sup's theorem asserts that if X and F are also simply connected with rational homology of finite type then: • // ?, (?) ® Jk is surjective and if there are r linearly independent elements a?? 6 (-i'y)even such that adz* and hlxt· are both locally nilpotent, then cato X > cato F + r . Jessup's theorem is proved by a careful analysis of the holonomy representation in terms of the Sullivan model of the fibration ? : X —> Y. For this we first introduce the holonomy representation for an arbitrary minimal relative Sullivan algebra (AV, d) —> (AV ® AW, d) in which V = {V*}i>9 and W = {^'}.>9 are graded vector spaces of finite type and (AV, d) is itself a minimal Sullivan algebra. Recall that the homotopy Lie algebra L of (AV, d) is the graded Lie algebra defined by V = Hom(sL, Jfc) and (AV, di ) = C* (L) , where d\ is the quadratic part of the differential d in AV (§21(e) and Example 1, §23(a)). Now filter (AV ® AW, d) by the ideals F? = A^PV ® AW to obtain a first quadrant spectral sequence (??,??). Its So-term is just (??,??) = (AV ® AW, id ® d) , d denoting the quotient differential in AW, and so the .??-term, (?\,?\), is a (AV, di)-module of the form (??, ??) = (AV®H(AW,d),Si) in which <5i : H(AW,d) —* V ® H(AW,d). Thus (cf. B3.6)) a right representa- tion of the homotopy Lie algebra L in H(AW,d) is defined by Definition This representation of L in H (AW, d) is the holonomy representa- tion for the relative Sullivan algebra (AV, d) —>· (AV ® AW, d). Now suppose the relative Sullivan algebra (AV ® AW, d) is a model for the fibration ? : X —>· Y, as in A5.4). This determines a quasi-isomorphism (AW,d) -=> APL(F), and so identifies H(AW,d) = H*(F;k). On the other hand, Theorem 21.6 identifies L with Ly , so that the holonomy representation for the fibration is a right representation of L in H*(F; k). In Theorem 31.3 we show that
Rational LS-Category 417 • The holonomy representations of L in H(AW,d) and in H*(F;k) are dual up to sign. This permits the translation of hypotheses on the holonomy representation of Ly in H*(F;jk) into conditions on the differential in the Sullivan model (AV ® AW,d) where they can be applied, in particular, to prove Jessup's theorem. An important aspect of the holonomy representation for any relative Sullivan algebra (AV, d) —> (AV ® AW, d) is that it can be expressed in terms of deriva- tions of the quotient Sullivan algebra (AW, d), (a derivation ? in a differential graded algebra (A,d^) is a derivation in A such that 6dA = (—l)degedA6). More precisely, if Vi is a basis of V, derivations #* of (AW, d) are defined by d(\ ® ?) - 1 ® ?? - Y^ vi ® ??? 6 A.-2V ® AW , ? 6 AW . i Thus by definition the spectral sequence differential ?? : ? (AW, S) —> V <8> H(AW,d) is given by <5?[?] = ? ?, ® ?(??)[?]. It follows that if xj is the dual i basis of L ((vi; sxj) = 1 or 0 as i = j or i ? j), then a-Xi = (-l)deeXi{deea+1)H(ei)a , ? 6 H(AW,d) . C1.1) Formula C1.1) expresses the holonomy representation for the relative Sullivan algebra in terms of the derivations ?(??). However, the derivations ?% themselves carry more information. For example, if V = fov and deg ? is odd then the entire differential in Av <g> AW is given by dv = 0 , and d(l <g> ?) = 1 <g> ?? + ? <g> ?? , ? 6 ??47 . In particular, any derivation <?' of (AW, d) with deg ?' = 1 — deg ? determines a relative Sullivan algebra Av —> (Av <8> AW, d') by the formula d! = l®d + v®6'. Note as well that (in general) the holonomy representation is characterized by ¦ (AV®H(AW,d),S1) = C* (L;H(AW,d)) (cf. §23(c)). In particular, the spectral sequence above converges from E2 = ExtUL (Je, H (AW, d)) =*· H(AV ® AW, d) . This section is organized into the following topics (a) The holonomy representation for a Sullivan model. (b) Local nilpotence and local conilpotence. (c) Jessup's theorem. (d) Proof of Jessup's theorem. (e) Examples.
418 31 The homotopy Lie algebra and the holonomy representation (f) Iterated Lie brackets. (a) The holonomy representation for a Sullivan model. Consider a Serre fibration ? : X —> Y with fibre inclusion j : F —> X, and such that F, X and Y are simply connected and have rational homology of finite type. Let ? : (AV, d) —> (AV <8> AW, d) be a minimal relative Sullivan algebra that is a minimal Sullivan model for the fibration in the sense of A5.4); in particular we have the commutative diagram APL(Y) (AV,d) Apl(p) Apl(X) Apl(F) C1.2) (AV ® AW, d) (AW, d) , in which ??? and m are minimal Sullivan models and m is a Sullivan model. Use H(m) to identify H(AW,d) = H*(F;]k) = Horn (//.(F; *),*), and use Theorem 21.5 to identify the homotopy Lie algebra Ly for Y with the homotopy Lie algebra L for (AV, d). Recall the two holonomy representations of L, one in H(AW,d) and one in H^(F;k), defined in the introduction to this section. Theorem 31.3 These two representations are dual (up to sign): proof: As in §16(b) we may apply the construction in §15(a) to the path space fibration q : PY —> Y to obtain a commutative diagram APL(Y) APL(PY) APL(QY) (AV, d) (AV ®AV,d) - (AV,0) of Sullivan models. Now consider the pullback diagram of iiF-fibrations X xY PY * PY f X = Y ¦
Rational LS-Category 419 Diagram C1.2) provides a Sullivan model m : (AV ® AW, d) -=> APL(X) and a Sullivan representative for ?, ? : (AV, d) —> (AV ® AW,d), such that mA = ???,{?)???- We may therefore, as in §15(a), construct t = APL(f)m-ApL(g)n:(AV®AW,d)®Kv(h.V®AV,d)-*ApL(XxYPY). Moreover, Proposition 15.8 asserts that ? is a Sullivan model of the form ?: (AV ® AW ® AV, d) -=> i4PL(X x (Note that AV ® AW and AV ® AV are sub cochain algebras.) Next consider the composite Serre fibration ? ? / ·. ? ? ? ?? —y Y. The fibre at yo is F x CIY and the inclusion of the fibre is the continuous map ? : F ? CIY —» ? ?? ?? , ? : (?, ?) ·—+ U'^) ·7 , ? e F, y e ?? . A straightforward check shows that in this case diagram A5.4) has the form APL(Y) Ap^°f) APL(X ????) ^^ APL(F ? ??) ¦n C1.4) (AV,d) (AV ® AW ® AV, d) — (AW,d) ® (AV,0). Let ii be a cocycle in (AW,d) representing ? 6 H*(F;Jc). If ? : AW —> V ® AW is the map defined by ?? - 1 ® ?? - ?? 6 A^2V ® AW, then by definition ?(?) = ?? and a-z = (-l)desu+des*(tf@)a;5:r) . C1.5) On the other hand, the weak homotopy equivalence j : F —> ? ? ? ?? is represented by the surjective quasi-isomorphism ? · id ·? : (AVigiAW^AV, d) —> (AW, d). Let w G AV <g> AW ® AV be a cocycle that maps to u. Then it follows from diagram C1.4) that for ? G H*(F;lk), (?,?-?) = {H*(a)H*(j)-la,?®hur(x)) = (H(e-id)[w],?®hur(x)) . But by construction we may write w = 1 ® u ® 1 + w\ ® 1 + y^l ® Uj ® Vj + ? where W! G V®AW, m G AW, Vi G V and ? G A^2(Ve V)-A(V® AW® AV). From dw = 0 we deduce that dui = 0 and hence that (?,?.?) = (_l)Wegz T([Ui},?)([Vi},hur(x)) .
420 31 The homotopy Lie algebra and the holonomy representation ex Furthermore, if d0 : V -^> V is the linear part of the differential in AV <g> AV we also deduce from dw = 0 that Qu + (id®d)wi + ?(~l)degUi desVid0Vi <g>Uf = 0. i Thus C1.5) becomes (a-x,?) = - Finally, recall that L = Ly = ?*(?.?) <g> k and that sLy is identified with ?*(?) <8> k via sz = — (— l)de&xd~1x, where 9, is the connecting homomor- phism of the path space fibration (§21(e)). Thus the formula in the proof of Proposition 15.13 becomes (d0Vi;sx) = (v^x) = ([vi], hur(x)), where (?,?·?) = d3d ' Again suppose ? : X —> Y is a Serre fibration with fibre F, and with A', Y and F simply connected and ha\ring rational homology of finite type. The holonomy representation of the homotopy Lie algebra Ly in the homology H*(F; h) of the fibre is an important measure of the non-trh'iality of the fibration: the holonomy representation for the product fibration is trh'ial. More generally we ha\'e Proposition 31.6 Let ? : X —> Y be a Serre fibration with fibre F, and sup- pose Y is simply connected with rational homology of finite type. IfH*(X; k) —> H*{F]k) is surjective then the holonomy representation of Ly in H,:(F;Ji) is trivial. proof: Let (AV <g> AW, d) be a minimal Sullivan model for the fibration as in A5.4). Then for any cohomology class ? G H (AW, d) there is a d-cocycle ? G AV <g> AW of the form ? = 1 <g> ?? + *? ? such that ?; G AiV <g> AW and ?? is a- <i-cocycle representing a. As in the introduction, choose a basis V{ of V and define derivations ?{ of (AW, d) by d(l <g>?) - 1 ® ?? - ??{ ®?{? G A^2V <g> AW. Then (formula C1.1)) the holonomy representation in H(AW, d) is given by ? · Xi = ±?(??)?, where ?? is the dual basis of the homotopy Lie algebra L of (AV, d). Here we have dty = 0 and so ??{ <g> ^?? = -(id <8> rf)*i. Thus each ?(?{) = 0 and the holonomy representation in H*(F;Jk) is trivial (Theorem 31.3). D (b) Local nilpotence and local conilpotence. Recall that a linear map 7 : M —> M in a graded vector space is locally nilpotent if e\'ery ? G M is in the kernel of some power 7fc of 7. We shall make frequent use of the dual notion: Lemma 31.7 Suppose ? : M —> M is a linear map of non-zero degree in a graded vector space M = {Mp} eZ of finite type. Then the following conditions are equivalent: (i) The dual linear map in Hom(M, k) is locally nilpotent.
Rational LS-Category 421 (ii) For each ? there is an integer k(p) such that Mp ? lmok^ = 0. (in) If (wk)k>o is an infinite sequence of elements of M such that aWk+i = Wk, k>0 then wk = 0, k > 0. oo (iv) ? Im ?* = 0. k=0 Definition A linear transformation of non-zero degree in a graded \rector space of finite type is called locally conilpotent if it satisfies the conditions of Lemma 31.7. proof of Lemma 31.7: (i) ·?=?- (ii): Denote the linear map dual to ? by /. Then (M, fk (Hom(M, Jfc)p)> = (Mp ? Im ?*, Hom(M, k)p) . Thus if (ii) holds then fk^ = 0 in Hom(M, Jk)p, pGZ, and / is locally nilpotent. Conversely, if (i) holds then /fc(p) = 0 in Hom(A/, fo)p for some k(p), because this vector space is finite dimensional. Thus Mp ? Imak^ = 0 and (ii) holds. (ii) <=>¦ (Hi): Clearly (ii) => (iii). If (ii) fails then for some ? and each jfe, Mp ? lmak ? 0. Set Ak = (?^1) (?" - {0}). Then Ao ?- ?? ?- A2 ?— · · · is an infinite sequence of linear maps between non-void affine spaces. For dimension reasons the sequences .4^ D ?(.4^+1) D ?2 (Ak+2) D ··· must stabilize at some integer K(k) : ?? (Ak+n) = ??^ (Ak+K^) for ? > K(k). Put Ek = ??^ (Ak+K(k))· Then for ? large enough a(Ek+1) =?(?? (Ak+l+n)) = ??+1 (Ak+n+i) = Ek - Thus we may inductively construct elements Wk G Ek so that aWk+i = Wk- Moreover, since Ak = (crk+l) (Mp — {0}) no element in Ak is zero, so u>k ? 0, Jfe>0. (ii) ^=^ (iv): Clearly (ii) =?· (iv) and the reverse implication follows from decreasing sequence Mp D Mp ? Im ? D Mp ? Im ?2 D ¦ ¦ ¦ and the fact that Mp is finite dimensional. ? Example 1 The holonomy representation. Suppose (AV, d) —> (AV <g> AW, ci) is a relative Sullivan algebra as in the introduction to this section and let (vi) be a basis of V. Let L be the homotopy Lie algebra of (AV, d) and denote by hi' ?, ? G L, the holonomy representation of L in H(AW,d). Recall from the introduction that derivations ?-? in (AV, d) axe determined by the equations d(l ® ?) - ( 1 ® ?? + ? Vi ® ??? J G A^2V <g> AW , ? G AW . Formula C1.1) states that HFi) coincides (up to sign) with hl'xi, where (??) is the basis of L dual to the basis (v{). Thus Lemma 31.7 asserts that the folknving conditions are equivalent:
422 31 The homotopy Lie algebra and the holonomy representation • hi' Xi is locally conilpotent. • For each ? there is an integer k(jp) such that Hp(AW, d) Dim ?@i)h^ — 0. If (ak)k>o is an infinite sequence of elements of H(AW,d) such that H{9i)oik+l = ak, k > 0, then ak = 0, k > 0. D Example 2 The adjoint representation. Let L be the homotopy Lie algebra of a minimal Sullivan algebra (AV, d) in which V = V-2 is a graded \-ector space of finite type. Each ? G L determines the linear function V —> k given by ? \—>¦ (v;sx). Extend this to a derivation, ??, in AV and define a derivation ?? in (AV, d) by setting ?? = ??? — (-1)?^???? d. Because (AV, d) is minimal ?? preserves A+V. Define fx : V —> V by requiring fx is the linear part of ??. We show now that (fxv;sy) = (-l)d^xde^(v;s[x,y}) , ? G V, y G L . This formula exhibits fx as the dual (up to sign) of ad x. In particular, fx is locally conilpotent if and only if ad ? is locally nilpotent. For the proof of C1.8) note that fx = ???? - (-l)desVxnxdi, where dx is the quadratic part of d. Then use §21(e) to compute (fxv;sy) = -(-l)de^^xd1v;sy) = (-l)d We shall need one more observation about local conilpotence: its good be- haviour with respect to spectral sequences. Indeed, suppose ? : (M, d) —> (M, d) is a linear transformation of a graded vector space of finite type and suppose fur- ther that both ? and d preserve a filtration of M of the form M = F°M DFlAID---D FPM D ¦ ¦ ¦ in which f]FpM = 0. This determines a cohomology spectral sequence (Ei,d{) ? and a morphism ??(?) : (Ei,di) —> (Ei,di). Lemma 31.8 If some Ei(a) is locally conilpotent then so is ?(?). proof: If ?(?) is not locally conilpotent then there is an infinite sequence of non-zero classes ak G H (M), such that H(a)ak+i = ak, k > 0. Let p(k) be
Rational LS-Category 423 the greatest integer such that ak has a representing cocycle in Fp(fc)il/. Since p(k) > p(k + 1) > · · · there is a p and a ko such that p(k) = p, k > ko. Let Zk be the affine space of cocycles in Fp representing ak, k > ko. Then ?? (Zk+n) 3 cn+1 (Z/j+n+i) D · · · is a decreasing sequence of finite dimensional affine subspaces and hence Ak = f] ?? (Zk+n) is non void subspace of Zk- Exactly ? as in Lemma 31.7 it follows that a(Ak+i) = Ak, and so there is a sequence of cocycles zk G Ak such that ? Zk+i = Zk and Zk represents a^. Since ? = p{k) and Zk G Fp, Zk represents a non-zero class [zk] G Ef'*. But clearly Ei(a)[zk+i] = [zk], k > ko, and so if ? (?) is not locally conilpotent then Ei(a) is not locally conilpotent either. D Examples 1 and 2 allow us to prove a technical lemma that is important for the proof of Jessup's theorem, and in §33. Consider a minimal Sullivan algebra of the form (Av ® AU ® AW, D) in which • ? has odd degree and Dv = 0. • (Av ® AU, d) —> (Av ® AU ® AW, D) is a relative Sullivan algebra. Then a derivation, ?, of the quotient. Sullivan algebra (??7 <g> AW, d) is defined by D(l <g> ?) = 1 ® ~?? + ? <g> ??. Next, let L be the homotopy Lie algebra of (Av <g> AU,D) and denote by ad and by hi' the adjoint representation of L, and the holonomy representation of L in H(AW,d), where d is the quotient differential in AW. Define ? G L by (?; sx) = 1 and (U; sx) — 0. Then we have Lemma 31.9 With the notation and hypotheses above suppose ad ? is locally nilpotent and hi' ? is locally conilpotent. Then ?(?) is locally conilpotent. proof: First observe that ? preserves AU and also A+U, because (Av®AU, D) is a minimal Sullivan algebra. Next, let ? be the derivation in Av ® AU defined by ??? = (w;sx), w G jkv ? U; thus ?? = 1 and rj(U) = 0, and thus setting ? = ?·? - (-1)?^^?? we have ?A <g> ?) = 1 <g> ?? , ? G AU . Define / : U —>· U by the condition ?-f-.U—i- A^2U. Then Example 2 states that / is dual (up to sign) to ad x. In particular / is locally conilpotent. Next, filter AU® AW by the ideals Fp - A^PU®AW. Since ? preserves A+U it presents this filtration and in the corresponding spectral sequence we have ?? (?) =ef® id + id ®?(??) ¦ AU ® H (AW, d)^AU® H(AW, d) , where #/ is the derivation in AU extending /. Now ?(??) is locally conilpotent by hypothesis, and we have just observed that so is /, because ad ? is locally nilpotent. Given a positive integer n choose ? so that for k > 0: U^n ? fk (U^N) = 0 and H^n(AW, d) ? HFv)k (H^N(AW, d)) = 0 .
424 31 The homotopy Lie algebra and the holonomy representation Then the formula above for E\ (?) implies that [APU ® H(AW, d)] nDE1 @)* [.\PU ® H(AW, d)] ^ip+1)N = ? , jfe > 0 . Hence ?\(?) is locally conilpotent and thus so is ?(?), by Lemma 31.8. D (c) Jessup's theorem. Consider a fibration V.X-+Y of simply connected topological spaces with simply connected fibre F and such that Y and F (and hence X) have rational homology of finite type. Let <9, <g> Jc : ?*(?) ® He —> tt»-i(F) ® Ik denote the connecting homomorphism and let hi denote the holonomy representation of the homotopy Lie algebra Ly in H* (F; ]k). Theorem 31.10 (Jessup [99]) Suppose in the fibration ? : X —> Y that d, ® k = 0. Assume there are r linearly independent elements ? ? (Ly)even such that both ad ? and hi ? are locally nilpotent. Then cato X > cato F + r . We shall deduce Theorem 31.10 from its translation into Sullivan algebras. For this we consider a minimal relative Sullivan algebra (AV, d) —»· (AV ® AW, d) in which V = V-2 and W = W-2 are graded vector spaces of finite type, and (AV, d) is itself a minimal Sullivan algebra. Choose a basis v\, v2, ¦ ¦ - of V such that degu; < degUj+i and define derivations O,· in the quotient Sullivan algebra (AW, d) by requiring d(l ® ?) - 1 ® ?? - ? Vi ® ??? e A^2V ® AW , ? € AW . i Then let (x{) be the dual basis of the homotopy Lie algebra L of (AV,d). Theorem 31.11 Suppose in the relative Sullivan algebra (AV, d) —> (AV ® AW, d) that the linear part of the differential in AV ® AW is zero. Assume that in the basis of L there are r elements Xi of even degree such that ad ?? is locally nilpotent and ?(?{) is locally conilpotent. Then cat(AK® AW, d) > cat(AW, d) + r . We show first that Theorem 31.11 does imply Theorem 31.10. Indeed, choose (AV, d) —> (AV ® AW, d) to be a rational Sullh'an model for the fibration ? :
Rational LS-Category 425 X —> Y as described in A5.4). Then (AV, d) is a minimal Sullh'an model for Y and L = Ly. Thus the linear part of the differential in AV ® AW is zero, because it is dual to <9, <g> k (Proposition 15.13). Suppose there are r linearly independent elements ? G Leven such that ad ? and hi ? are locally nilpotent. Choose a basis Vi,v%,... ?? V such that degux < degu2 < · · · and so that these r elements appear as a subset ???,...,??? of the dual basis (x{) oiL. Then adz*,, is locally nilpotent. Moreover, H(9iu) is dual (up to sign) to hlzit,, by formula C1.1) and Theorem 31.3. Thus H(9it/) is locally conilpotent. This sho\vs that the minimal Sullivan model of ? : X —> Y satisfies the hypotheses of Theorem 31.11. Now let (AU1, d) and (AW1, d) be minimal rational Sullivan models for X and for F. Then (AU',d) ® k and (AW',d) ® k are minimal Sullh'an models (over k) for X and F. It follows that cat(AV<g>Aiy, d) = cat(AC/', d) = cato X and cat(AtV, d) = cat(AW', d) = cat0 F (Example 1, §29(e) and Proposition 29.4). Thus Theorem 31.10 follows from Theorem 31.11. The proof of Theorem 31.11 will occupy the next two topics. First, in §31(d), we shall consider the special case that V = kv and degu is odd, so that AV is the exterior algebra Av. Then in §31 (e) we shall use induction and the mapping theorem to prove the theorem in general. (d) Proof of Jessup's theorem. Consider first a minimal Sullh'an algebra of the form (Av <g> AW, d) in which deg ? is odd. Thus the differential is described by: dv = 0 and d(l ® ?) = 1 ® ?? + ? ® ?? , ? G AW , C1.12) where ? is a derivation in the quotient Sullh'an algebra (AW,d). In this case Theorem 31.11 reduces to Proposition 31.13 If ?(?) is locally conilpotent then cat (Au ® AW, d) > cat (AW, d) + 1 . proof: Letcat(Au®AW,d) = m + l. The ideals ?® (AW,d) and v®(A>mW,d) are in particular (Av®AW, d)-modules. Let ? : (Q, d) -=> v® (AW/A>mW, d) be a (Av ® AW, d)-semifree resolution for the quotient module. Define lq G H(Q) by ?(?)??> = [?® 1]. Next, observe that Q extends to the (Av®AW, d)-semifree module (Q®N,d) denned as follows: iV is a graded space with basis (an)o<n<oo and degan = n(degv — 1), the module action is ? -(? ® an) = ?? ® an for ? G Av ® AW, ? G Q, and the differential is gh'en by d(z ® ao) = dz ® oq and d(z ® an) = dz® an+vz® ??-? , ? > 1 We identify Q with Q ® ao via ? <—> ? ® ao-
426 31 The homotopy Lie algebra and the holonomy representation The proof of the proposition then consists of the following four steps. Step 1: Construction of a morphism of (Av ® AW, d)-modules 7 : (Q, d) —> (Av ® AW, d) such that H(j)lq = [v ® 1]. Step 2: Extension of 7 to a morphism ? : (Q ® N,d) —> (Av ® AW, d). Step 3: Construction of a homotopy from ? to a morphism ? with image in v®AW. Step 4: Proof that caX(AW, d) <m. We now carry out this program. Step 1: Construction of a morphism of (?? ® AW, d)-modules 7 : (Q,d) —> (Av ® AW, d) such that H(j)lq = [v ® 1]. Let I(m + 1) = A>m+1 (v®W) <zAv® AW and let ? : (?, d) ^> (Av ® AW/ I(m + \),d) be a (Av ® AW,d)-semifree resolution. Define iP e H(P,d) by: H(C)ip = [1]. Since cat(Au ® AW, d) = m + 1 there is a morphism 77 : (P, d) —> (Av ® AW, d) of (Av ® AW, d)-modules such that ?(??) = [1]. (This is just the definition of cat(Au <g> AW, d) if we take for (P, d) the minimal Sullivan model of the surjection Av <g> AW —> Av <8> APF/7(m + 1)-) Next observe that the inclusion of the ideal v®(AW, d) in (Av®AW, d) factors to define a morphism ? : ? ® (AW/A>mW, d) —»· (At; ® AW/I(m +\),d) of (Av ® AW,d)-modules. Lift ? to a morphism ?' : (Q,d) —> (P,d) such that ??' ~ ^. Finally set j = ??' : (Q,d) —»· (At; ® AW,d). Since ?(?)?(?')?? = ?(?)?(?)??> = ?(?)[?® 1] = [?] · 1 = ?^@^ = #(?([?]·*?) we have H(tp')iQ = [?] ¦ Lp. Thus H(j)lq = [v® 1], Step 2: Extension of 7 to a : (Q ® N,d) ^ (?? ® AW, d). Extend f to 6v : (Q®N, d) —> v® (AW/A>mW, d) by setting ??(?>®??) = 0, ? > 1. In the diagram (Q ® N, d) ? ® (AW/A>mW, S) we may construct the morphism r so that ?? ~ ^ (Proposition 6.4(i)). Moreover it is immediate from the proof of this proposition that we may suppose r restricts to the identity in Q, because ?/v and ? coincide in Q. Thus ? = ? ? ?' ? ? : (Q ® ?, d) —> (?? ® AW, d)
Rational LS-Category 427 restricts to 7 in (Q,d). Step 3: Construction of a homotopy from a to a morphism ? with image in ? ® AW. It is in this Step that we apply the hypothesis that ?(?) is locally conilpotent. More precisely, we establish Lemma 31.14 Suppose (xn, zn)n>o is an infinite sequence of pairs of elements xn,zn 6 AW satisfying: dzn = 0 and ???+? = zn + dxn , ? > 0 . Then there is an infinite sequence of elements yn ? AW such that dyn = zn and 9yn+i = yn + xn , ? > 0 . proof: Observe first that each [zn] G f| Im HF)k = 0, since [zn] = ?(?)[??+1] = k ?(?J[??+2] = ¦ ¦ ¦ ¦ Thus the affine spaces An = d~l(zn) are non-void and finite dimensional. Affine maps —> An+i —> An —> > Ao are defined by an(y) = 6y — xn , y G An+\. For each ? this gives the infinite decreasing sequence of affine spaces An DlmanD Im(an ? ??+?) D · · · D Im(an ? · · · ? an+k) D · · · · For dimension reasons this sequence must stabilize at some k(n) : Im(an ? · · · ? cxn+p) =Im(ano.--an+fc(n))forp > k(n). Put En = Im(ano· · -oan+fe(n)). Then an(En+i) = En since, for ? sufficiently large, an(En+i) = an(Iman+i ? ··· ? ctn+p) = En- Thus we may choose an infinite sequence of elements yn G En such that an+iyn+i = yn,n> 0. Thus 0(yn+i) = yn + Xn- Since En C An = d~l(zn) we also have dyn = zn, ? > 0. ? We now complete Step 3 by constructing a homotopy ? from ? to a morphism ? satisfying the stronger condition ?(Q®ao)Cv®AW and ?(Q®an)=0, n>\. C1.15) For this, write Q = Av <g> APF <g> M, where M decomposes as the direct sum 00 ? = ? Mfc such that d(M0) = 0 and d(Mk+i) C Av <g> ??-?7 <g> M<fc. Denote fc=0 and Av <g> APF ® M<fc respectively by Qt and Q<t· Now any linear map g^ : Mk®N —> Av®AW extends uniquely to a Av®AW- linear map Gk ¦ Q <8> ? —>· ?? ® ??-?7 such that Gk{Mi ® N) = 0, ? # A;. A
428 31 The homotopy Lie algebra and the holonomy representation sequence of linear maps, gk : Mk ® ? —> ?? ® AW, determines in this way the oo Av <g> A W-linear map G = ? Gk ¦ Q ® ? —> ??® \W. fc=0 Suppose now by induction that gi, i < k of degree —1 are constructed so that a' = a - [ ? dGi + Gid] satisfies condition C1.15) in Q<k ® N. We shall \i<k J construct g/, so that a- ^ dGi +Gid I satisfies C0.15) in Q<k ® N. If (gk) \i<k ) f °° is the infinite sequence constructed in this way then ? = a — ^ dGi + Gi \i=0 f satisfies C1.15) because ? restricts to ? — [? dGi + Gid I in Q<k ® N. Vi=o / It remains to construct gk, given a'. Define ?? ® AM^-linear maps fn,qn : Q<k —> AW by the equations ?'(?®??) = /?? + ?(????) , ?>0. The condition da = ad then implies dfn = fnd and #/n+i - fn = dqn+i + qn+id , n>0. Recall now that we suppose a' = 0 in Q<fc<8>an, ? > 1 and that a' (Q<k ? ao) C ? ® AW. Since d : Mk —> Q<k it follows from this that fn°d : Mk —> 0 , ? > 0 and qn ? d : Mk —> 0 , ? > 1 . Thus if ? G Mk the equations above reduce to dfnZ = 0 , ? > 0 and ?/?+?? — fnz = dqn+iz , ? > 0 . We may thus apply Lemma 31.14 with zn = fnz and xn = qn+iz to find a sequence of elements yn € AW such that dyn = Zn and 0yn+i = yn + ?? , ? > 0 . Now define ^ : Mfc ® ? —> AW by setting gk(z ® an) = yn where (yn) is a sequence as above for z, and ? runs through a basis of Mk- Then d<7fc(z <g> an) = /„? and 9gk(z ® an+1) = 5fcB <g> an) + ???+?2, ? > 0, ? G Mk- Moreo\'er [a' - (dGk + Gkd)} (z®an) = fnz+v(qnz)-dgk(z®an)-v9gk(z®an)+vgk(z®an--l) where ?_? =0. Thus the right hand side is zero for ? > 1 and reduces to v(qkZ — 9gk(z ® oo)) if ? = 0. Since a' and ? coincide in Q<k ® ?, the inducth'e step is complete. Step 4: Proof that cat(AJ?, d) < m.
Rational LS-Category 429 Define a (?W,d)-module structure in (Q ® N,d) by setting $-(z®an) = JT(-l)P( P + U )(???)?®??+?, ?????, zGQ . p=0 \ ? / (Since ? decreases degrees this series is a finite sum). By Step 3, ? : (Q®N, d) —> v®(AW, d) is a morphism of (Av®AW, d)-modules, vanishing on Q®an, ? > 1. It follows that ? is a morphism of (AW, <i)-modules. Next, let zq be a cocycle in Q representing the class lq of Step 1 and define a morphism ? : v®(AW, d) —> (Q®N, d) of [AW, d)-modules by setting ?(?®?) = (-1)***? .ZQm We shall show that ?? = id. Since ? and ? are AW-linear it is enough to show that ?\(? ®\) = ? ® 1. But ?\(? ®\) — tv®l, some t G 3c, for reasons of degree. Moreover ?\(? ® 1) = ?zQ differs from azc by a coboundary in Av <g> AW, because ? ~ ? (Step 3). Since ? restricts to 7 in Q (Step 2) and since H(-j)lq = [v ® 1] (Step 1) it follows that t[v ® 1] = [v ® 1] in H(Av ® AW, d). Since (?? ® AW,d) is a minimal Sullivan algebra ? ® 1 cannot be a coboundary; i.e. [v ® 1] ? 0 and t = 1. Thus ?\(? ®\)-?®\ and ?? = id. Finally, recall the (Av ® AW, (i)-quasi-isomorphism ? : (Q, d) ~ ? ® (AW/A>mW, S) defined at the start of the proof of the proposition. It extends to the quasi- isomorphism ? ® id : (Q ® N, d) —> ? ® (AW/A>mW, d) ® (N, 0) since ?(vQ) = ? · ?(?)) = 0. Moreover we can make ? ® (AW/A>mW, d) ® (N, 0) into a (AW, d)-module , and ? into a morphism of (AW, d)-modules, by setting ? -(? ® ? ® an) = (-l)dee* X;(-l)p ( P + n ) (<???) ? ® an+p . P=o \ ? / Thus altogether we have the diagram of morphisms of (AW, d)-modules ? ® (AW, d) (Q ® N, d) ^ ^ ? ® (AW, d) ~ ?.®id v ® (AW/A>mW, d) ® (N, 0) , with ?? = id. Since Av ® AW is a minimal Sullivan algebra, imdCl® A^2W + v® A+W. Thus ? preserves A+W. It follows that with respect to the module action just defined, W · (v ® AW/A>mW ® N) C ? ® A+W/A>mW ® ? .
430 31 The homotopy Lie algebra and the holonomy representation Hence A>mW · (v ® AW/A>mW ® N) = 0. Now the Corollary to Hess' theo- rem 29.9 asserts that cat(ATV, d) <m. D proof of Theorem 31.11: Recall that V is equipped with a basis (vt). Di- viding by the elements v\,... , u;_i gives a quotient Sullivan algebra of the form (AVi ® AW,Di) in which Vi,Vi+i,... project to a basis of Vj·. Moreover the Mapping theorem 29.5 asserts that cat(AV, ® AW, D{) > cat(AFi+1 ® AW, Di+i) > > cat(AW, d) . Thus Theorem 31.11 follows from the assertion: If Xi G Leven and if adxj is locally nilpotent and ?(?{) is locally conilpotent, then cat(AVj ® AW, Di) > cat(AVi+i ® AW, Di+1) + 1 . To verify this, simplify notation by writing Vi — v, Vi+1 = U, #* = ?? and X{ = x. Denote D{ by D and Di+l by ~D. Thus the Sullivan algebra (AVi ® AW, D{) has the form (Av ® AU ® AW, D). The homotopy Lie algebra of (?? ® ??7, D) is the subspace I C L spanned by the Xj, j > i. In particular adx : / —> I is the restriction of ad X{ : L —> L and hence is locally nilpotent. Since ? has even degree, ? has odd degree. Thus the differential in ?? ® ??7 ® AH'* is gi\7en by Dv = 0 and D(\ ® ?) = 1 ® ?)? + ? ® ?? , ? G AU ® Aiy , where O is a derivation in (??/???7, D). Lemma 31.9 asserts that ?(?) is locally conilpotent, and thus Proposition 31.13 applies to give cal(Av ® AU® AW, D) > cai(AU®AW,~D) + l. D (e) Examples. Example 1 The fibration ? : S4m+3 —l· MPm. The unit sphere S3 of the quaternions El acts freely by right multiplication on the unit sphere S4m+3 of Hm+1 (= M4m+4). This is the action of a classical principal S3 bundle (§2(a)) p : S4m+3 —> MPm, with base space the quaternionic projective m-space. The long exact rational homotopy sequence for this fibration reduces to ir4rn+3(S4m+3)®Q^K4m+3(MPm)®® and ?4 (MPm)®Q A n3(S3)®Q . Thus dim ?, (MPm) ® Q < oo and dimwodd (HPm) ® <Q> = 1. Moreo\'er, the holonomy representation is automatically locally nilpotent (but not trh'ial) since the fibre has finite rational homology. Thus all the hypotheses but one of Theorem 31.10 are satisfied: <9, ® Q ? 0. Since cat0 S4m+3 < cat0 S3 + dim ???? (MPm ) ® Q the conclusion of the theorem fails too, which shows that the hypothesis ?, ®Q = 0 is necessary. (Note, howe\rer that it intervenes only at the end, in Step 4, of the proof of Proposition 31.13). D
Rational LS-Category 431 Example 2 The fibration associated with S3 V S3 —> S3. Convert projection from S3 V S3 to the first factor to a fibration ? : X —> S3 with X ~ S3\JS3-- Since ? is not a rational homolog}' equh'alence the fibre, F, has non-tri vial rational homology and so cato F >1 = cat0 „Y, and the conclusion of Theorem 31.10 fails. Here all the hypotheses hold, however, except for the hypothesis that the holonomy representation is locally nilpotent. This must therefore fail, and the example shows it is a necessary hypothesis. D Example 3 Derivations in commutative cochain algebras. Suppose ? is a derivation of even (negative) degree in a commutative cochain algebra (A, dA) such that A0 = h and Hl{A) — 0. Then we may construct the commutative cochain algebra (Av <g> A, d), with degu + deg# = 1 by setting dv = 0 and da — dAa + ? <g> ??, a G A. Now let (AW, d) be a minimal model of (A, d). Then (Av®A, d) has a Sullh'an model of the form (Av ® AW, d) with ?? = 1 <g> ?? + ? <g> ?'?. If (?? <g> AW, d) is not a minimal Sullivan algebra then ?' restricts to a non-zero Gottlieb element jydegt;-i __j. A and Proposition 29.8(_ii) asserts that cat(AW,d) = oo. On the other hand, the isomorphism H(AW,d) = H(A,dA) identifies ?(?') with ?(?). Thus if ?) Im H(9)k — 0 we can apply Proposition 31.13 to conclude that k f < catiAu <g> AW, d) — 1 , if (?? ® AM7, d) is minimal. cat(AW, d) is < I oo , otherwise. An interesting special case of the above is an arbitrary Jfc[0]-module M con- centrated in degrees > 2, from which we construct (A, cU) by setting ^1 = k®M, dA = 0 and M ¦ M = 0. In this case if ? is locally conilpotent then caA(Av®AW,d) = 2 . In fact since (AW,d) ~ (A,0) and niL4 = 1 it follows that cat(AW,d) = 1 (Proposition 29.3) and hence cat(Au ® AW, d) > 2. On the other hand, (Av <g> AW, d) -=> (Av <g> A, d) and nil (Au ®A) = 2, so cat(Au ® AH7, d) < 2. D Example 4 ? fibration X —> CPm ????? ^ere 53 and cat0 X = 2. Let S1 act on 52m+1 by complex multiplication: S1 is the unit circle of C and 52m+1 is the unit sphere in <Cm+1. This is the action of a principal S^bundle, 52m+1 —> €Pm (§2(d)). For m = 1 we have the action of S1 on S3 and the associated bundle (§2(e)) p:X = S3 xSi 52m+1 —> CPm is a Serre fibration with fibre S2. Since ?, (CPm)®Q is concentrated in degrees 2 and 2m+1 it follows for degree reasons that the rational connecting homomorphism is zero. Since the fibre has
432 31 The homotopy Lie algebra and the holonomy representation finite dimensional homology the holonomy representation is by locally nilpotent transformations. Thus Theorem 31.10 applies and gives cato X > cat0 53 + l = 2. But \ve may also regard X as the total space of a Serre fibration X —> S2 with fibre 52m+1, so that cat0Ar < 2 (Proposition 30.7). Altogether we have cato X = 2. ¦ D (f) Iterated Lie brackets. Let Lx be the homotopy Lie algebra of a simply connected topological space X with rational homology of finite type. Apply Theorem 31.10 to the fibration ? -?4 ? to obtain Theorem 31.16 If cato X = m there are at most m linearly independent ele- ments Xi € (Lx)even such that adxi is locally nilpotent. In particular if cato X is finite then for each non-zero ? € (Lx)even of suf- ficiently large degree there is some y € Lx such that the iterated Lie brackets [x, [x[x,..., [x,y] ¦¦¦]]] are all non-zero. Similarly, if L is the homotopy Lie algebra of a minimal Sullivan algebra (AV, d) with V = V-2 a graded vector space of finite type, then Theorem 31.11 applied to the relathre Sullivan algebra (AV, d) -^-> (AV, d) ghres Theorem 31.17 i/cat(AV, d) = m there are at most m linearly independent elements x^ € Leven such that ad X{ is locally nilpotent. Example 1 A space with e0X = 2 and cato X = oo. In Example 6, §12(d), we constructed a minimal Sullivan model (AV,d) such that d : V —> A3V and every cocycle in A-3V is a coboundary. Moreover V = V-2 and has finite type and it is immediate from the construction that we may choose Vodd to be infinite dimensional. Because every cocycle in A-3V is a coboundary, e(AV, d) < 2. Because there are no coboundaries in A2V, e(AV,d) = 2. Because d: AkV —> Afc+3V the Milnor-Moore spectral sequence collapses at the ^-term. On the other hand, the homotopy Lie algebra is abelian, because the quadratic part of d is zero. Thus Theorem 31.18 shows that cat(AV,d) = oo. In particular, if X is the geometric realization of (AV,d) (§17) then cat0 X = cat (Ay, d) = oo and e0X = e(AV, d) = 2 . D Finally recall that the construction of (AV,d) begins \vith a graded subspace ? C V such that d{Z) = 0, and that V = ? ® W with d injective in W. Let / be the ideal A-2V © W, and let (A,d) be the sub cochain algebra given by A = ??2*?. The inclusion (A,d) —> (? ? lk,d) is a quasi-isomorphism. k Moreover, dividing by (A+J and by a complement of kerd in A2V defines a quasi-isomorphism (A, d) -^ ? (A) and shows H+(A) · H+(A) = 0.
Rational LS-Gategory 433 The surjection (AV, d) —> (AV/I,d) is a commutative representative for a continuous map q : X —> \J Sna, where Sna corresponds to a basis element of a ? with degree na. Moreover (/ 0 it) is a commutative model for the cofibre of q as follows from a simple calculation using the Remark after Proposition 13.6. In view of the observations above / ? it is the commutative model of a wedge of spheres (Example of §12(a)) and so we have constructed a cofibration Exercises 1. Let F -> X -> S2n+1 be a fibration. Suppose that F has the homotopy type of a simply connected CW complex of finite type and that Ht(QS2n+1) acts trivially on H*(F). Show that the induced map Ht(F;Q) -> H*(X;Q) is injective. 2. Let / : X = (S3 x S3) V (S5 x S5) -> S3 be the continuous map whose restriction to S3 x S3 is projection onto the first factor, and that restricts to the trivial map on S5 x S5. Denote by F the homotopy fibre of /. Prove that catX = cato-F = 2. Prove that H*(F;Z) is not a free ii,(fiS3)-module, nor a locally nilpotent one. 3. Let ? : X -> 52n+1 be a fibration with simply connected fibre. If n2n+i(S2n+1) acts locally nilpotently on H*(F;k) and if H*(X;k) is finite dimensional, prove that H*{F;k) is finite dimensional. 4. We consider the rational fibration V™=1Sq -> A" -> 5??+1 whose commutative model is given by (?? <g> (?,??,... ,an),d) with at-ay = 0 and d(di) = ???-?- Prove that: adx.(az·) = ??·+? ,... ,adx(an) = 0. Using Jessup's result, prove that cat0A/' = 2. 5. Prove that the fibration F -? ? -i S2n+1 , ? > 1, admits a Lie model that is a short exact sequence of differential graded Lie algebras 0 -> L(r(x) <8> MO -»· L(Qa; © W) -»· L(x) -»· 0. Interpret the holonomy operation as multiplication by x. 6. Denote by ?? the subspace of Lx generated by the Engels elements of 7?.? ® Q. Prove that ^(X) C ?*·. Construct an example with GQ(X) ? Ex ? Lx. 7. Let F 4 ? -> ? be a fibration with dim ?»(?) (g) Q < oo and ker^»(j) <8> Qjcvcn _ q Suppose that H,((W;Q acts locally nilpotently in H,(F;Q), and prove that
Part VI The Rational Dichotomy: Elliptic and Hyperbolic Spaces and Other Applications
32 Elliptic spaces In this section the ground ring is an arbitrary field k of characteristic zero. A simply connected topological space X is called rationally elliptic if it satisfies the two conditions: dim#„(X;Q) < oo and dimir*{X) ® Q < oo . By analogy a minimal Sullwan algebra (AV,d) is elliptic if both H(AV, d) and V are finite dimensional. If (AV, d) is a minimal Sullivan model for a simply connected space X then H*(AV,d) = H*(X;k) and V = Romz(n,(X),k) (Theorem 10.1 and Theorem 15.11). Thus X is rationally elliptic if and only if its minimal Sullivan models are elliptic. Rationally elliptic spaces occur naturally; for instance the classical homoge- neous spaces G/K of differential geometry are rationally elliptic. Howe\'er, the 'generic' space with finite homology is not rationally elliptic; for example, in this section we shall see that the homotopy groups of rationally elliptic spaces satisfy very stringent restrictions. Indeed, suppose X is rationally elliptic and let ?? be the maximum degree such that Hnx (X;Q) ? 0. In Theorem 32.6 we shall show that if (xt) is a basis of 7TOdd(-20 ® Q and if (yj) is a basis for 7reven(J\f) <g> Q then • ? degXi < 2?? — 1 and ?deg?/7· < ??, so that, in particular, • tti(X) ® Q = 0 , i > 2nx and dim^(A') ® Q < nx. We also obtain a formula for ??: Vi - 1)· i j Finally, in Proposition 32.10 we shall show that • dim^dd(X)®Q-dim^ven(X)®Q> 0 and J2(-iy dimH^X;^) > 0, i with equality holding on the left if and only if it fails on the right. These various inequalities have interesting geometric properties. For example • If X is rationally elliptic then dim TToddpf) ® Q < cat0 X , and (Allday-Halperin [3])
The Rational Dichotomy 435 • // an r-torus acts smoothly and freely on a rationally elliptic closed mani- fold X then r < dim 7rodd(A') ® Q - dim 7reven(A') <g> Q . In particular, if X — G IK with G and ? compact simply connected Lie groups then r < rank G — rank if . Most of these results are obtained by reduction to the case of pure Sullivan algebras (AV, d); these are the finitely generated Sullh'an algebras in which d = 0 in Veven and d : Vodd —> AVeven. Cochain algebras of this special form were introduced by H. Cartan in [33] for the computation of the cohomology of homogeneous spaces; in retrospect these can be seen to be the first examples of Sullivan algebras. The properties of pure Sullivan algebras as established in (a) and (d) below are largely due to Koszul [102], and indeed they are often referred to as Koszul complexes. This section is organized as follows: (a) Pure Sullivan algebras. (b) Characterizations of elliptic Sullh'an algebras. (c) Exponents and formal dimension. (d) Euler-Poincare characteristic. (e) Rationally elliptic topological spaces. (f ) Decomposability of the loop spaces of rationally elliptic topological spaces. (a) Pure Sullivan algebras. Given a Sullivan algebra (AV,d), write Veven = Q and Vodd = P, so that Ay = AQ <g> AP. Recall that (AV, d) is called pure if V is finite dimensional and if d = 0 in Q and d(P) C AQ. If (AV, d) is a pure Sullivan algebra, then d is homogeneous of degree — 1 with respect to wordlength in P. 0 <— AQ <— \Q ® ? < 4— AQ ® i\kP < . Thus if Hk(AV,d) is the subspace representable by cocyles in AQ®A*P we have H(AV,d) = ® Hk(AV,d)i this will be called the lower grading of H(AV,d). k Note that d(AQ ® P) — AQ - d(P) is the ideal in the polynomial algebra AQ generated by d(P). Thus H0(AV,d) = AQ/AQ-d(P) . Proposition 32.1 If (AV, d) is apure Sullivan algebra then H(AV,d) is finite dimensional if and only if Ho(AV; d) is finite
436 32 Elliptic spaces proof: Since AV is a finitely generated module over the (noetherian) polyno- mial algebra AQ any submodule is also finitely generated. Since d{AQ) — 0, kerd is a AQ-submodule of AV; hence it is finitely generated. Thus H(AV, d) is finitely generated as a module over the subalgebra H0(AV,d). The Proposition follows. D Consider again a pure Sullivan algebra (AV, d) as above and suppose 2??,..., 2aq are the degrees of a basis (yj) of Q and 2bx - 1,..., 2bp - 1 are the degrees of a basis (xi) of P. (Thus dim Q = q and dim ? = p). Proposition 32.2 Suppose H(AV,d) is finite dimensional and let r and ? be the maximum integers such that Hn(AV,d) ? 0 and HT(AV,d) ? 0. Then (i) Hn(AV,d) is a one-dimensional subspace of Hr(AV,d). (ii) r = dim ? — dim Q; in particular, dim ? > dim Q and equality holds if and only if H(AV,d) = H0(AV,d). (in) ? = ?B^-1)- i i proof: Define a graded \'ector space Q by Q = Ql+1 and denote the cor- respondence between elements by y «-» y. Extend (AV, d) to a pure Sullivan algebra (AQ ® A(P <g> Q), d) by setting dy = y, y G Q. This can be written as (AQ <g> AQ <g> ??, d) and (AQ <g> AQ, d) is contractible (§12(b)). Thus division by Q and Q defines a quasi-isomorphism (AQ ® AQ <g> AP,d) ^> (??, 0), whence #sm (AQ «8» ?? «8» AQ, d) = (AsP)m . Let k be the largest integer such that Hr(AV,d) ? 0. It is easy to construct an ideal / C AQ <g> ?? such that: / is preserved by d, H (I, d) = 0, / = ? / ? s (AQ <g> ?5?) and / D (AQ <g> A5P)^ if s > r or if s = r and I > k. Let (A, d) be the quotient cochain algebra (AQ <g> AP/I.d) and note that A = ®AS, where 5 As is the image of AQ <g> A5P, and that A+ -A* C A? ¦ A* + A>r = 0. By Lemma 14.2 the quasi-isomorphism ? : (AQ ® AP,d) —> (A,d) extends to a quasi-isomorphism (? ® id) : (AQ ® ?? ® AQ,d) ^> (A ® AQ,d) , where dy = ?dy. Thus if iiis(^4 ® AQ) is the space represented by cocycles in ? Ai ® AJ'Q, we have fism(^ ® AQ) ^ H™(AQ ® ?? ® AQ) = (AsP)m . Now let ? G if be a cocycle representing a non-zero cohomology class. Since A+ ¦ A*t = 0, and since dy~i G A+, a®y~iA---Ayq is a cocycle in ? <g> AQ. It
The Rational Dichotomy 437 cannot be a coboundary, since (A®AQ)r+q+1 = 0. Thus it represents a non-zero cohomology class of lower degree r + q, which is obviously the maximum lower degree possible. Since dim ? = ? this gives an inclusion H${A) ® ft ? · · · ? yq C Hr+q(A ® AQ) ? APP . In particular r + q = ? and r = p- q = dim ? — dim Q. Since dim APP = 1 we have dim H^(A) — 1 and, if x\,..., xp is a basis of P, fc = Finally, let ? be the largest degree such that HN(A ® AQ) ? 0 and let J C AQ ® ?? be an ideal of the form (AQ ® AP)>n © Jn chosen so that H(J,d) = 0. Set ? = AQ ® AP/J. Then the identical argument gives /i ? · · · ? yq C ^^E ® AQ) S APP . This shows that ? — k, which completes the proof. ? Finally, let Q be a finite dimensional graded vector space concentrated in even degrees. A regular sequence in the polynomial algebra AQ is a sequence of elements «i,...,uTO in A+Q such that u\ is not a zero divisor in AQ and (for i > 2) the image of Ui is not a zero divisor in the quotient graded algebra On the other hand any sequence ulr..,um of elements in A+Q determines the pure Sullivan algebra (AQ ® ?(??,... ,xm), d) defined by dxi = Ui and we have Proposition 32.3 The sequence U\,... ,um is regular if and only if H+(AQ®A(xu...,xm),d) = 0. proof: We show first that if H+(AQ<8>A(xi,... ,xm-i)) = 0 and if ui,..., um_i is a regular sequence then H+(AQ<8>A(xi,.. ¦ ,xm),d) = 0 if and only if ui,... ,um is a regular sequence. Denote by / the ideal in AQ generated by «i,..., um-i· Under the hypotheses above a quasi-isomorphism ? : (AQ ® A(a^i,... ,a;m_i), d) —> AQ/I is defined by: ??? — 0 and, for a G AQ, ?? is the image of ? in AQ/1. Then (Lemma 14.2) ?®??: (AQ <8> ?(??,... ,xm),d) -=> (AQ/I <g> A:rm, d) is a quasi-isomorphism, where dxm = ??-m- Let ? C AQ/I be the subspace of elements u such that u · ipum = 0. Then H(AQ/I <g> Axm,d) = AQ/(u\....,um) © (? ® xm). Since the isomorphism ?(?®??) preserves lower degrees (clearly) we have Ar>2(AQ®A(a;i,..., xm), d) = 0 and Hi(AQ® A(x\,... ,xm)) — K®xm- Since ? = 0 if and only if ui,... ,um is a regular sequence, this is equivalent to H+ (AQ ® A(^!,... ,xm)) = 0. Finally, if m, ¦ ¦ ¦ ,um is a regular sequence so is «i,... ,um_i and so by in- duction H+(AQ® A{xi,...,xm-\)) = 0. Thus H+(AQ ® A(xu ... ,xm)) = 0.
438 32 Elliptic spaces Conversely, suppose H+(AQ ® A(:ri,.. .,xm)) = 0. For any k, let ? G AQ <g> Ak(xi,. .. ,xm-i) be a cocycle representing a non-zero class of least degree in Hk(AQ ® A(:ri,... , ?m_i)). A simple calculation shows ? is not a coboundary in AQ® A(xi,...,xm)· Thus k = 0 and H+(AQ <g> ?^?,... ,xm_i)) = 0. By induction, ui,... ,um-i is a regular sequence and so by the argument above, so is«i,...,um. D (b) Characterization of elliptic Sullivan algebras. Suppose (AV, d) is a Sullivan algebra in which V is finite dimensional. Write yodd = ? and yeven _ q^ so that AV = AQ <g> AP is the tensor product of the polynomial algebra, AQ and the exterior algebra, AP. Then bigrade AV by setting (AQ ® AkP)n = (AV)k+n>~k and filter AV by setting F"(AV) = (AV)^P'*. Since the differential increases degree by one and decreases the wordlength in ? by at most one it preserves the filtration. Thus (AV, d) is a filtered cochain algebra, with which there is then a naturally associated spectral sequence of cochain algebras, called the odd spectral sequence for the Sullivan algebra. It is immediate from the definition that the first term of the odd spectral sequence has the form where da is characterized by: dff(Q) =0 ,?? :P —>AQ and d - d„ : ? —> AQ ® A+P . Thus (AV, da) is a pure Sullivan algebra and (in the notation of §32(a)) E™ = Hp_+qq{AV,da) . We call (AV, da) the pure Sullivan algebra associated with (AV, d). Note also that Fp((AV)n) = 0, ? > n + dimP. Thus the odd spectral sequence is convergent to H(AV, d). Proposition 32.4 Let (AV, d) be a minimal Sullivan algebra in which V is finite dimensional and V = V-2. Then the following conditions are equivalent: (i) aixnH{AV,d„) < oo. (ii) dimH(AV,d) < oo. (Hi) cat(AV,d) < oo. proof: Since the odd spectral sequence converges from H(AV, da) to #(AV, d) the implication (i) =?· (ii) is immediate, while (ii) =$» (iii) follows from Corol- lary 1 to Proposition 29.3. To prove (iii) =? (i) let ??,.. .,xn be a basis of V such that degzx < deg^2 < ··· and dxi G A(x\,...,Xi-{). We show first by induction on i that if dega^^ is even then for some JVj, xi ' = da^i.
The Rational Dichotomy 439 For this, divide by the ideal /* generated by x\,... ,Xi-\ to obtain a quotient Sullivan algebra (AV,d). Then Xi,...,xn project to a basis (xj) of V, and dxi = 0. On the other hand, by the Mapping theorem 29.5 (AV, d) has finite category. Thus there is an element ? G A(xi,... ,xn) such that d? = xf (some N), where ? is the image of ? in AV. Since deg ? is odd we have ? = ?? + ?3+· · · with ?^· ? AjVodd ® AVeven. It follows that (djj, = ??, which implies that Define elements yj G V by setting yj = Xj if dega^j is even and yj = 0 if degxj is odd. Since ?? G Vodd ® AVeven it follows that ???? - ?? = j G AVeven. The induction hypothesis gives yj > = da^j, j < i. Since daClj = 0 follows that some power of Xi is a d^-coboundary. This closes the induction. As in §32(a) put Q = Veven and ? = Vodd. We have just shown that for a basis yj of Q some power y · J is a da-coboundary. It follows that Hq(AQ ® AP,dc) = AQ/AQ -da(P) is finite dimensional. Hence H(AQ <g> AP,dc) is also finite dimensional (Proposition 32.1). D Example 1 ?(?2,^3,U3,64,^5,^7; da = dx = 0, du = a2, db = ax, dv = ab — ux, dw = b2 — vx). Here subscripts denote degrees. The differential da is given by daa = dab = dax = 0, dau = a2, dav — ab, daw = b2. Thus in H(AV,da) we have [a]2 = [b]2 - 0 and so (AV, d) is elliptic. D Example 2 Adding variables of high degree. Suppose (AV", d) is a minimal Sullivan algebra in which V — V-2 and V is finite dimensional. It is an interesting question of Anick whether, given an arbitrary positive integer n, there is an elliptic Sullivan algebra of the form (AV" ® AW, d) with W concentrated in degrees > n. For example, consider the Sullivan algebra (subscripts denote degrees) ?(?2, b2, X3, y3,z3, C4; da = db = 0, dx — a2, dy = ab, dz = b2, de = ya — xb) . It is not elliptic because dac = 0 and no power of c is a d<r-coboundary. On the other hand, if we add odd degree variables u, ? and w with du - cna - ncn~1xy, dv = cn+1b - (n + \)uyb - (n + \)cnxz + (n+ \)uaz and dw = c2n+2 - 2(n + l)cn+1uy - 2(n + \)vua + 2(n + l)vckx then we obtain an elliptic Sullivan algebra, since daw = c2n+2. ? Again let (AV, d) be a minimal Sullivan algebra and denote by k[z] the poly- nomial algebra in a single variable, 2, of degree 2. (Note Jk[z] = Az, but we wish to emphasize that this is a polynomial algebra.) Proposition 32.5 Suppose V is finite dimensional and V = V-2, and suppose Jk is algebraically closed. Then (AV, d) is elliptic if and only if every morphism ? : (AV, d) —> (J;[z],0) is trivial.
440 32 Elliptic spaces proof: Suppose (AV, d) is elliptic and let x\,..., xn be a basis of V such that dxi G ?(?!,... ,?;_i). Given </? : (AV, d) —» (jfo[z], 0), suppose by induction that ??? = 0, j < i. Then ? factors to give a morphism ? : (A(x{,... , ?n),d) —> (]k[z],0) from the quotient Sullivan model. If degzj is odd then ??? = 0. If deg ?* is even then, since this quotient model has finite category (Theorem 29.5), xf = ??, some JV. Hence (???)? = ?(?^) = ??? = 0. It follows that ??? = 0. Thus ? is trh-ial. Conversely, suppose (AV, d) is not elliptic. Then by Proposition 32.4, H{AV, dc) is infinite dimensional, and hence so is H0(AV,da), by Proposition 32.1. This algebra has the form A(j/i,... ,yq)/I where / is generated by polynomials /,·, 1 < i < p. Since it is infinite dimensional there is some yj such that y*j $ I for all A;. Extend the polynomial algebra A(y1;... ,yg) to a polynomial algebra AB/1, ¦ · ·, Vq, u) with deg u = - deg yj. Then the ideal generated by A,..., /p and by pjU — 1 is properly contained in A(y\,... ,yq,u). Thus the Hilbert Nullstellensatz asserts the existence of scalars h,... ,tq, t G Jk such that each fi(h, ¦ ¦ ¦ ,tq) = 0 and such that tjt = 1. In particular, tj ? 0, and a non- trivial morphism ? : A(yi,... ,yq)/I —> h[z\ is given by yz- 1—> t{zai, where degyi = 2at. Finally, let J C AV be the ideal generated by Fodd and by Imd. Then AV/J = AVeven/AVeven-da(Vodd) = A(yu...,yg)/I, and a non-trivial mor- phism (AV,d) —> fo[z] is obtained by composing (AV,d) —> (AV/J,0) A k[z\. D Example 3 Algebraic closure of k is necessary in Proposition 32.3. Indeed if Jk = Q the Sullivan algebra A(a2.&2)?3;cte = a2 + b2) admits no non-trivial morphism to Q[z], since we would have a \—> az, b <—>¦ ?? with ?, ? G Q satisfying a2 + ?2 = 0. D Example 4 (Lechuga and Murillo,[106]) ?-colourable graphs. To each finite graph with vertices vj and edges e% and for each integer ? > 2 associate a pure Sullivan algebra (AQ <g> ??, d) as follows: Q is concentrated in degree 2 and has a basis yj corresponding to the vertices Vj\ ? is concentrated in degree 2n — 3 and has a basis Xi corresponding to the edges e*; finally dxi = ^Ty^y"-, where yk and y^ correspond to the endpoints of e*. The graph is ?-colourable if each vertex can be assigned one of ? distinct colours so that vertices connected by an edge have different colours and, as slKnvn in [121]: A finite connected graph is n-colourable 4=^ dim H(AQ ® ??, d) = 00. Indeed we lose no generality in assuming k is algebraically closed. If the graph is n-colourable identify the colours with the distinct nth roots of unity wa and note by wQ(j) the colour of the vertex Vj. If wa and W? are distinct nth roots of unity then ^,w^w^~s = w^ — W?/wa — W? = 0, and so a non-trivial morphism (AQ ® ??, d) —> k[z] is given by yj 1—> vja(j)z and Xi 1—> 0.
The Rational Dichotomy 441 Conversely, given a non-trivial morphism ? : (AQ ® ??, d) —> ]k[z] note that if yic and ye correspond to the vertices of ei then 0 = ip{dxi){ipyk — (pye) = s&yk - vw) = (</>?/fc)n - {???)?· Since the graph is connected 5 there is a single scalar ? such that (y>yj)n = Xzn, and since some c>yj ? ?, ? ? 0. Choose an nth root of ?, ?, and define wa(j) by ipyj = wa^\z. This n-colours the graph. D (c) Exponents and formal dimension. Fix an elliptic Sullivan algebra (?V, d) and bases y\,..., yq of Veven and ??,. ..,???? Vodd. The sequences ??,..., aq and bi,...bp defined by deg yj = 2aj and degZj = 26Z- — 1 are independent (up to permutation) of the choice of basis. Definition The integers ??,..., ?, and b\,..., bp are respectively the even and the odd exponents of (AV, d). The formal dimension of (AV, d) is the largest integer ? (or oo) such that ??(??,?)?0. In this topic we prove Theorem 32.6 (Friedlander-Halperin [61] Suppose (AV, d) is an elliptic Sulli- van algebra with formal dimension ? and even and odd exponents ??,..., aq and bi,... ,bp. Then ? g g (ii) 2_\2aj < n- ? (iii) y^Bbi - 1) <2n- 1. (iv) dimyeven < dimV01111 < cat(Ay,d). Corollary 1 // (AV, d) is an elliptic Sullivan algebra of formal dimension ? then V is concentrated in degrees < 2n — 1, and at most one basis element ofV can have degree (necessarily odd) > n. Corollary 2 If (AV, d) is an elliptic Sullivan algebra of formal dimension ? then dim y < ?. proof: We have ? = ? Bbt - 1) - ? BaJ - 1) > ? B&t - 1) + Q > ? + V, i=l j=l i=l where ? = dim Vodd and q = dim Veven. ?
442 32 Elliptic spaces The proof of Theorem 32.6 depends on a reduction to the pure case (Lemma 32.7) and a characterization of exponents (Proposition 32.9). Consider then an elliptic Sullivan algebra (AV, d). By Proposition 32.4 the associated pure Sullivan algebra (AV, ??) also has finite dimensional cohomology. Lemma 32.7 (AV, da) and (AV, d) have the same formal dimension. proof: We argue by induction on dimV. Write (AV, d) as a relative Sullivan algebra (Av ® AW, d) in which V = hv ? W and ? is an element in V of minimal degree. The Mapping theorem 29.5 asserts that the quotient Sullivan algebra (AW,d) has finite category; hence it too is elliptic (Proposition 32.4). Next observe that if m is the formal dimension of (AW, d) and ? is the formal dimension of (AV, d) then {m + deg ? if deg ? is odd , , C2·8) m — deg ? + 1 if deg ? is even. Indeed if deg ? is odd (and therefore > 3) this follows from the long exact coho- mology sequence associated with 0 —> ? <8> (AW, d) —» (Av <8> AW, d) —» (AW, d) —» 0 . If deg ? is even extend (AV, d) to (AV <8> Av,d) by setting dv = v, and use the long exact cohomology sequence associated with 0 —» (AV, d) —» (AV <g> ??, d) —» (AV, d) <g> ? —> 0 to conclude that formal dimension (AV®Av,d) = n+degu = n+degf —1. Then write (AV®Av,d) = (Av<S>Av<S>AW, d) and observe that because H(Av<S)Av,d) = Jk the surjection (AV <8> ??, d) —> (AW, d) is a quasi-isomorphism. This proves C2.8). The same argument applies to the relative Sullivan algebra (Av <8> AW, da), and here the quotient Sullivan algebra (AW, da) is just the pure Sullivan algebra associated with (AW,d). By induction (AW,d) and (AW, dc) have the same formal dimension. Thus formula C2.8) for (AV, d) and for (AV, da) shows these have the same formal dimension too. ? Our next step is to derive the fundamental property enjoyed by the exponents of an elliptic Sullivan algebra. Proposition 32.9 (Priedlander-Halperin [61]) Suppose a\,..., aq and b\,..., bp are the even and odd exponents of an elliptic Sullivan algebra (AV, d). Then for any integer s A < s < q) and any subsequence a^,...,ajs of even exponents, there are at least s odd exponents bi that can be written in the form
The Rational Dichotomy 443 where the k\ are non negative integers (depending on bi) and Y^k\ > 2. ? proof: Since the associated pure Sullivan algebra is also elliptic (Proposi- tion 32.4) we lose no generality in assuming that (AV, d) is itself pure. Re-index the even exponents so that the subsequence in question is ??,..., as. Let (yy) and (??) be bases of Veven and Vodd chosen so degyy = 2ay and degXi = 2bi — 1. Divide by the yy, j > s, to obtain a pure elliptic Sullivan algebra of the form (A(yi,... ,ys) ® A{xu ... ,xp),d). Next, renumber the basis xt so that dxt ^ 0, 1 < i < r and dxi = 0, i > r. Then our pure Sullivan algebra has the form (A(yi,... , ys) <g> A(a;1,... ,xr),d) <g> {h(xr+i,... ,Xp), 0) and so (A(yi,... , ys) <g> A(a;i,... ,xr),d) is elliptic. By Propo- sition 32.2 (ii), r>s. On the other hand, for i < r, dxi is a non-zero polynomial in the yy with no linear term, and so it contains a term of the form yf' · · ·y*5 with Y^k\ > 2. f \ Thus 2bi = degXi + 1 = ? kx degyA = 2 ?=? \?=? Remark In [61] it is shown via a difficult argument from algebraic geometry that any pair of sequences ax,..., a, and b\,...,bp satisfying the conclusion of Proposition 32.9 are the even and odd exponents of an elliptic Sullivan algebra. Thus this condition precisely characterizes the possible degrees of a basis for V in an elliptic Sullivan algebra (Ay, d). D proof of Theorem 32.6: In view of Lemma 32.7 we may suppose (AV, d) is pure. Thus (i) is just Proposition 32.2 (iii). Now order the exponents so that ci > a2 > ¦ ¦ ¦ > aq and b\ > b% > ¦ ¦ ¦ > bq. At least s of the bi are non- trivial non-negative integral combinations of a\,...,as (Proposition 32.9). In particular, at least s of the bi satisfy bi > 2as. This implies that bs > 2as. Thus by(i) Similarly, ? = ? and so 2n - 1 > ? {2bi - 1). i=l This proves (i)-(iii). Proposition 32.2 (ii) asserts that dim Veven < which is the first inequality of (iv). Finally, the homotopy Lie algebra L of (Al7, d) is finite dimensional, since sL is dual to V. Thus adz is locally nilpotent for all ? ? L and so Theorem 31.17 implies that cat(AV, d) > dimLeven (= dimVodd). D
444 32 Elliptic spaces (d) Euler-Poincare characteristic. Recall (§3(e)) that the Euler-Poincare characteristic xm of a finite dimensional graded vector space is defined by ?? = ?(-1)' dim Mi = dim Aieven-dim Modd. If M is equipped with a differential, d, then xm = XH(M,d), as follows from elementary linear algebra. The purpose of this topic is to prove Proposition 32.10 Let ? be the Euler-Poincare characteristic of the cohomol- ogy of an elliptic Sullivan algebra (AV,d). Then ? > 0 and dim Fodd - dim V'even > 0 . Moreover, the following conditions are equivalent: ?) X > 0- (ii) H(AV,d) is concentrated in even degrees. (Hi) H(AV, d) is the quotient A(y\,..., yq)/(u\ , - -., uq) of a polynomial algebra in variables (yj) of even degree by an ideal generated by a regular sequence (iv) (AV,d) is isomorphic to a pure Sullivan algebra (AQ <25 AP,d) in which Q z= Qeven, ? z= podd and d maps a basis of ? to a regular sequence in AQ. (v) dim y°dd - dim Veven = 0. The proof of the Proposition relies in part on another invariant of pure Sullivan algebras, introduced originally by Koszul in [102]. If (AQ <8> AP,d) is a pure Sullivan algebra with Q = Qeven and ? = Podd then its Koszul-Poincare series is the formal series U(z) given by oo U(z) = S^ \rzr , ?? = ^(-l)fc dim(AQ ® AkP)r~k . Lemma 32.11 If 2ai,..., 2aq are the degrees of a basis (yj) of Q and if 2b\ — 1,..., 2bp — 1 are the degrees of a basis (xt·) of ? then ? (i-*26') U(z) = ?? . ? (?-*2?0
The Rational Dichotomy 445 proof: Write AQ <g> ?? = AV. Clearly U = UAV does not depend on the differential, and ?Av®aiv = ^av 'U.\w- Since AV = ??/? (8) - · · (8) Ayq q$ ??? <g> • - · <g> ??? and since (trivially) Uhyi (z) = _ z2a. and UAxi (z) = l- z2bi , the lemma follows. ? Now while the Koszul-Poincare series U(z) = ]T Xrzr for (AQ&AP, d) does not ? depend on the differential, it nonetheless gives important numerical information about H(AQ <g> AP,d). Indeed, since elements of AQ <g> AkP have odd (even) degree if k is odd (even), ?? is just the Euler-Poincare characteristic of Cr = ?(?<2 <8> AkP)r~k. Moreover, because d decreases wordlength in ? by 1 and k increases degree by 1 it follows that d preserves each Cr. Thus ?? is also the Euler-Poincare characteristic of H(Cr,d): where Hk{AQ ? ??) is as denned in §32(a). Now suppose H(AQ <g> AP, d) is finite dimensional. Then H(Cr) = 0 (and hence ?? = 0) for sufficiently large r; i.e., U(z) is a polynomial. Moreover, since AQ <g> AP = 0Cr, the Euler-Poincare characteristic ? of H(AQ <g> AP, d) is given by x = ? xmc) = ? ?' = w(!) - C212) proof of Proposition 32.10: Recall the odd spectral sequence (Ei,di) (§32(b)) for (AV,d), whose i?i-term is just the cohomology H(AV,da) of the associated pure Sullivan algebra. Since this is finite dimensional (Proposition 32.4) and since Ei+\ = H(Ei,di) the Euler-Poincare characteristics satisfy ??(??,??) = ??? = Xe2 = ¦¦¦ - Since ??'"* = AFeven <g> AkVodd it vanishes unless 0 < k < dimV01111 and so Em = Ei, i > diml/odd. But ?,? is the associated bigraded vector space for H(AV, d) and we conclude that XH(AV,d) =XH(\v,d„) ¦ C2.13) On the other hand, if 2??,.... 2aq are the degrees of a basis for yeven and if 2by - 1,..., 2bp - 1 are the degrees for a basis of Vodd then Lemma 32.11 asserts that the Koszul-Poincare series U for (AV, ??) satisfies
446 32 Elliptic spaces Since U{z) is a polynomial and since ? = 1 occurs as a root with multiplic- ity 1 in 1 — zm it follows that ? > q and that ? = 1 occurs as a root of U(z) with multiplicity ? — q. Since ? = dim V01" and ? = dimVeven we have shown that dimFodd > dimVeven (another proof is given in Theorem 32.6), that XH(AV,d) = XH(AV,d„) = W(l) = 0 if dimVodd > dimVeven and that if dim Vodd = dim Veven = q then XH(AV,d) = W(l) = ? bj H a,- > 0 C2.14) This proves the first assertion of the Proposition and the equivalence (i) <=> (v). Now clearly (iii) =? (ii) =? (i), and Proposition 32.3 asserts that (iv) => (iii). It remains to show that (i) => (iv). If (i) holds then, since (i) => (v), dimVeven = dim Vodd. Since H(AV,dc) is finite dimensional (Proposition 32.4), Proposition 32.2(ii) asserts that H+(AV,da) = 0. It follows (Proposition 32.3) that if (??) is a basis of Vodd then (daX{) is a regular sequence in AVeven and that H(AV,da) = H0{AV,da) = AV^en/{daxu... ,daxq). In particular, H(AV, da) is evenly graded and so the differentials dx, d2,. ¦ ¦ in the odd spectral sequence all vanish. In other words H(AV,da) is the bigraded vector space associated with H(AV, d). Let y\,..., yq be a basis for yeven. Since dayi = 0 there are elements ?? e Ayeven <8> A+Vodd such that d(yi + ?^) = 0. Put y[ = yi + ?? and note that (AV,d) = A{y[,...,y'q)® AVodd. Thus without loss of generality we may assume d — 0 in Veven. Finally let xx,... ,xq be a basis of Vodd such that degzi < · · · < dega;, and suppose by induction that for i < k we have found elements ?? 6 AVeven <g> j\>3yodd such tfrat ?(?? + ?{) = daX{. Replace x-x by x-t + *i to reduce to the case dxi = daxu i < k. Then d = da in Ayeven (g) ?(??,... ,xk_i). Since ????,... ,daXk~i is a regular sequence, Proposition 32.3 implies that daxk — dxk is a c^-coboundary in AVeven <g> A+Vodd and hence it has the form da^k, some ? a: G AVeven <8> A^3yodd. By construction d^k = da^>k, and so ^?* = Now set x[ =xi + ^i. Then (AF,d) = A(y[,... ,y'q,x\,... ,x'q), with dj/^ = 0 and dx\,..., cte'g a regular sequence in A(yi,..., y'q). D Corollary // dim Veven = dim yodd = q then ? A-^) ^2dimHr(AV,d)zr = ? ?2 ? / ? and XH(AV,d) = dimH{AV,d) = ? h ? i=i I j=\
The Rational Dichotomy 447 proof: Since H(AV,d) *? H(AV,da) = H0(AV,dc) we have in the Koszul- Poincare series U for (AV, da) that Ar = ??-1)* dimHk~k(AV>d*) = dimHr(AV,d) . k Apply Lemma 32.11. G (e) Rationally elliptic topological spaces. Let X be a simply connected topological space with finite dimensional ra- tional homology. The formal dimension ?? of ?' is the maximum integer such that ??? (X; Q) ? 0, and its Euler-Poincare characteristic is the integer Xx = ?{-1)???????(?;<0?). Recall that X is rationally elliptic if the graded ? vector space ?*(?) <8> Q is also finite dimensional. In this case we set ??(?) = dmiTTevenPO ® Q - dim7roddP0 <g> Q , and we let 2??,...,2?? and 2bi — 1,... ,2bp - 1 be the degrees of a basis of ?»(X) (8) <Q>, so that ? = dim 7TOdd(X) ® Q and ? = dim ????? (X) ® Q. Since the minimal Sullivan model, (AF, d), for X satisfies , d) = ?*{X; Q) , F = ??. (X) ® Q and cat(AV, d) = cat0 X we may translate Theorem 32.6 and Proposition 32.10 and their corollaries as follows: Theorem 32.15 If X is a rationally elliptic space then, in the notation above, (i) ?? (ii) ??>? 2aj. (Hi) 2nx-l> ?=? (iv) dim7reven(X) ® Q < dim7rOdd(-Y) ® Q < cat0 X- In particular ?»(X)<g>Q is concentrated in degrees < 2?? — 1 and dim tt»(X)®Q < Proposition 32.16 If X is a rationally elliptic space then in the notation above XX > 0 and X7r(X) < 0 . Moreover, the following conditions are equivalent:
448 32 Elliptic spaces (i) Xx > 0- (ii) H*(X;Q) is the quotient of a polynomial algebra in q variables of even degree by an ideal generated by a regular sequence of length q. (Hi) dim 7reven(X) <g> Q = dim ????(?) <g> Q. // these conditions hold then ) ^2dimHk(X;Q)zk = -± and Xx = -j k ? (?-*2*') ? Example 1 Simply connected finite ?-spaces are rationally elliptic. If G is as in the title then its Sullivan model is an exterior algebra on a graded vector space Pg of finite dimension concentrated in odd degrees, and has zero differential (Example 3, §12(a)). The dimension of Pq is called the rank of G.O Example 2 Simply connected compact homogeneous spaces GjK are rationally elliptic. Proposition 15.16 asserts that these spaces have a Sullivan model of the form {AVbk ® APG,d), where d = 0 in AVbk, Vbk *s concentrated in even degrees, d{PG) C AVbk and Pg and Vbk are finite dimensional. Note that this is a pure Sullivan algebra. In this example we have ??(?/?) = dimV?K — dim Pc = dim ?? — dim Pc, (even though the Sullivan algebra may not be minimal). Thus = rank(K) - rank(G) .
The Rational Dichotomy 449 Example 3 (Allday-Halperin [3]) Free torus actions. Suppose an r-torus ? = S1 ? · · · ? S1 (r factors) acts smoothly and freely on a simply connected compact smooth manifold M. Then the projection M —> M/T onto the orbit space is a smooth principal bundle. Hence there is a classifying map M/T —> BT whose homotopy fibre is homotopy equivalent to M (§2(e)). Now assume M is rationally elliptic. Since BT = CP°° ? · · · ? CP°° its homotopy groups are concentrated in degree 2, and since M/T is compact its homology is finite dimensional. Thus M/T is rationally elliptic. It is immediate from the long exact homotopy sequence that ??(?/?) = ??(?) + ????) = ??{?) + r . On the other hand, Theorem 32.15 asserts that ??(?/?) < 0 and so we conclude that r < -XM ; i-e., — Xm is an upper bound for the dimension of a free torus acting on M. In the case of homogeneous spaces G/K (Example 2, above) this shows that rank G — rank ? is an upper bound for the dimension of a torus acting freely on G/K. a (f) Decomposability of the loop spaces of rationally elliptic spaces. Let X be a simply connected CW complex. Recall the localization X-p (§9(b)) at a set of primes V, obtained by 'inverting the primes not in V.' Here we prove Theorem 32.17 (McGibbon-Wilkerson [117]) If the integral homology H.{X\ Z) is ? finitely generated abelian group and if X is rationally elliptic then invert- ing some finite set of primes gives a localization X-p such that there is a weak homotopy equivalence 1—? oh· ? ?—? /?'2?. ¦ 1 Corollary If X is a rationally elliptic finite CW complex then H*(UX; Z) has p-torsion for only finitely many primes p. Remark By contrast Anick [9] and Avramov [15] construct simply connected finite CW complexes X such that Hm(UX; Z) has torsion of all orders. proof of Theorem 32.7: In the course of the proof we shall use a little 'inte- gral' homotopy theory not derived in this monograph. The proof is by induction on dim7r»(X)<8>Q and for purposes of the induction we prove the theorem under the slightly more general hypothesis that for some finite set of primes Pi,-..,pn, H*(X;R) is a finitely generated module over R = ? (±,..., j- J cQ. In this case we may replace X by a CW-p-complex
450 32 Elliptic spaces (V = primes of R) with cells in finitely many dimensions. In the proof we shall often extend R (and further localize X) by inverting finitely many additional primes, and then continue to denote the result by R (and by X). We shall also denote by Sk the localization 5^,. We shall frequently rely on the following observation. Suppose given continu- ous maps X —> ? <^— ? in which g is a rational homotopy equivalence and each Hk(Y;Z) and Hk(Z;Z) are finitely generated A-modules. Then after inverting finitely many more primes we can find localizations X-p· —^ Yp· ^— Z-p· and a map h : X-p· —> Z-p· such that g-p* h ~ f-p<. Indeed, we may assume Y and ? are CW-p-complexes of finite type and that / maps X into a finite sub CW-p- complex of Y. Invert finitely many primes to make g a homology isomorphism between large skeleta and then compose h with the homotopy inverse- Note as well that by inverting finitely many primes we may arrange that X is (r— l)-connected and that HT(X\ Z) is a non-zero free i2-module. We distinguish two cases: Case 1: r is odd. Since Hr{X; Z) is a non-zero free A-module there are continuous maps Sr -A- X -^ K(R,r) such that Tir(gf) is an isomorphism (Theorem 4.19 and Propo- sition 4.20). Since r is odd the composite is a rational homotopy equivalence (Example 3, §15(b)). Thus by inverting finitely more primes we can construct h : X —-> Sr such that hf ~ id. Let i : F —> X be the 'inclusion' of the homotopy fibre of h. Then (cf. §2(c)) there is a principal fiSr-fibration ?5? —> F —> X. Since ([2]) ??»(?5?,· R) is the polynomial algebra R[ar-\) and H*(X;R) is finitely generated, a Serre spectral sequence argument shows that H*(F;R) is noetherian over R[ar~i] and hence has torsion at only finitely many primes. By inverting these we can arrange that H*(F;R) is torsion free. On the other hand the Mapping theorem 28.6 asserts that cato F < oo, and hence (Proposition 32.4) F is rationally elliptic. Thus H*(F;Q) is finite dimensional and H*(F;R) is a finitely generated A-module. But dim^(F) ® <Q> < dim7r»(X) <g> <Q> and so the theorem holds for F by induction. Since fli · ?/ : ?.? ? ?5? —> UX is a weak homotopy equivalence, the theorem holds for X as well. Case 2: r is even, r — 2q. As in the previous case we have Sr —> X -^ K(R, r) with nr(gf) an isomor- phism. Recall the definition of the special unitary group SU(q). As in Example 3, §15(f) there is an action of SU(q) on 52?-1 and this gives an associated fibre bundle ? : ESU{q) xSU(g) S2^1 —»· BSU{q) with fibre S2* (§2(e)). The calculations referred to in Example 3, §15(f) show that BSU(q)<Q ~ q Y\ KBi — 1,Q) and that the connecting homomorphism ?» for ? maps a ba- sis element for K2qBSU(q)Q to 5q9-1. Thus by inverting finitely many primes we obtain from g : X —> K(R,r) a map h : X —> BSU(q)-p such that
The Rational Dichotomy 451 d,[hf] = [S2"-1]. Use h to pull the spherical fibration back to a fibration S2' -^ ? —> X whose connecting homomorphism maps [/] to [S2']. This implies that dim ?» (E) <g> Q < dim ?» (X) <g> Q while a Serre spectral sequence argument shows that if» (E; R) is a finitely generated A-module. Thus by induction, the theorem holds for E. Finally, as in §2(c) we have a fibration ?? —> ?? —> S2' which, after finitely many primes are inverted, admits a cross-section. Thus multiplication in CiX defines a weak homotopy equivalence ?? ? S2q~1 -^ ???. ? Exercises 1. Determine all the elliptic spaces X satisfying dimL^ < 3. 2. Let X be an elliptic space and denote by ? the maximal length of a nonzero Whitehead bracket. Prove that cato^ > n/2.
33 Growth of Rational Homotopy Groups In this section the ground ring h is an arbitrary field of characteristic zero. Suppose that X is a simply connected topological space with rational homology of finite type. In this section we describe the implications for ?» (X) ® Q of the hypothesis that X has finite rational category. If dim ?,(?) <g>Q is finite then X is rationally elliptic (Proposition 32.4) and its properties are described in §32. Thus the focus here is on spaces X such that dim7r»(X) <g> Q = oo. Definition A simply connected topological space X with rational homology of finite type is called rationally hyperbolic if cato X < oo and dim7r»(A'')<g>(Q) = oo. The justification for the terminology lies in our first main result: k • If X is rationally hyperbolic then ? diinKi(X) <8>Q grows exponentially in i=2 k. Note that this defines a dichotomy: • Simply connected spaces X with rational homology of finite type and finite rational category are either: - rationally elliptic, with ?*(?) <8> Q finite dimensional, or else - rationally hyperbolic, with ?,(?) <g) Q growing exponentially. The result above leaves open the possibility of large intervals [r, s] such that Ki(X) <8> Q = 0, r < i < s. However, this cannot happen: it follows from Theorem 31.16 that • If X is rationally hyperbolic then there are integers ? and d such that Q^O , k>0. This assertion can be considerably improved in the case that H*(X; Q) has finite dimension. In this case we let ?? be the formal dimension of ?; ? ? is the largest integer such that Hnx (X; Q) ^ 0. Then we show that • If X has finite dimensional rational homology and formal dimension ?? then either - X is rationally elliptic and iTi(X) <8> Q = 0, i > 2??, or else - X is rationally hyperbolic and for each k > 1, Ki{X) <8> Q ? 0, some i G {k,k + ??). In particular, if X is rationally hyperbolic the non-zero rational homotopy groups of X occur at intervals of at most ? ? — 1. This section is organized as follows: (a) Exponential growth of rational homotopy groups.
The Rational Dichotomy 453 (b) Spaces whose rational homology is finite dimensional. (c) Loop space homology. (a) Exponential growth of rational homotopy groups. Proposition 33.1 Let X be a simply connected topological space with rational homology of finite type. If X is rationally hyperbolic then (i) dim????{?) <g> <Q> = oo. (ii) For some integers ? and d: iIN+kd{X) ® Q ? 0, k > 0. proof: Let {AV, d) be a minimal Sullivan model for A'. Then cat(AV,d) = cato A < oo. For any ? > 2,.divide by V<n to obtain a quotient Sullivan al- gebra (AV-n,d), which also has finite category (Mapping theorem 29.5). Since dim7Tp(A") ® <Q> = dim Vp, ? > 1, it follows that V~n ? 0. If V-n were concen- trated in even degrees we would have d = 0 and H(AV-n,d) = AV-n would be a polynomial algebra, contradicting cat(AV-n,d) < oo. It follows that Vodd is infinite dimensional, which proves (i). Let L be the homotopy Lie algebra of (AV, d). Since Lk is dual to Vk+1, Leven is infinite dimensional. Thus according to Theorem 31.17 for some non-zero ? G Leven and some ? G L the iterated Lie brackets (ad a)k? are all non-zero. This proves (ii). ? Theorem 33.2 Let X be a simply connected topological space with rational homology of finite type. If X is rational hyperbolic then for some C > 1 and some positive integer K: k ^2 dim ???(?) ® Q > Ck , k> ? . ?=0 proof: Let (AV,d) be a minimal Sullivan model for X. As in the proof of Proposition 33.1 we have to show that because cat(AF, d) < oo and dim V = oo, ? dim V1 > Ck, k > ? for suitable C and K. Set m = cat(AV, d). Now for any integer k dividing by V<k gives a quotient minimal Sullivan alge- bra which we can write as (AVffc·2*-2! ® AV^2*, d), where V^ = {Vl}k<i<r By minimality d — 0 in l/[*>'2*-2]. Moreover, this quotient Sullivan algebra has category < m (Mapping theorem 29.5) and hence the product of m +1 cohomol- ogy classes is always zero (Corollary to Proposition 29.3). Thus in particular,
454 33 Growth of Rational Homotopy Groups Define ? : V^2* —» AFt*·2*) by requiring that d = a + ? with Im/3 in the ideal generated by V^2*-1. If ?? is the component of ? in ArVl*'2*~2] then we have m+l Am+lVr[*,2*-2] = ? up (V5:2*-Ij .Am+ Since A^'*'2*' is concentrated in degrees i e \pk,2pk - 2p] it follows that 2k-2 Now for each integer k set A(fc) = ^ dim V*. Denote 2m+1 (m + l)!m by c i=k and use Proposition 33.1 (ii) to find an integer JV such that \(k) >c3m , k>N. Then it follows from the inequality above that if k > N, v ' *->' p=2 This in turn implies that for some ? G [2, m + 1], In particular, \{pk - 1) > A(fc) and \(pk - lI^-1 > L J Iterating this process yields a sequence of integers No < Ni < ¦ ¦ ¦ such that No = ? and 7Vi+i = piiVj - 1 (some pt G [2, m + 1]), and such that L cm J But since Ni+1 > |^ we have ? ? < i g (§)* = &. Thus cS ^ < 0 ' i-0 and it follows that Now for any i > 1 and for Ni <r < Ni+i we have r < (m 4- lJ7Vi_i and
The Rational Dichotomy 455 In other words, setting C - \^?\ (m+1Jjv we have ? dim Vn > Cr, r > ?. ? L Ct" J n=0 (b) Spaces whose rational homology is finite dimensional. Theorem 33.3 Suppose X is a simply connected topological space with finite dimensional rational homology and formal dimension ??. Then either tti(X) ® Q = 0, i > 2nx or else for each k > 1, ??(?) ® Q ? 0 for some i G (k,k + nx). proof: If dim n*(X)®Q is finite then Theorem 32.15 asserts that Ki(X)®Q = 0, i > 2??. Thus if X were a counterexample to the theorem there would be integers k > 1 and ? > ? ? + k such that ?{(?) ® Q = 0, k < i < i, and ne(X) <8> Q is non-zero. It follows (Theorem 15.11) that a minimal Sullivan model for X (over any field Jk of characteristic zero) has the form (AU <g> AW, d) in which U = U-k and W = W-* are non-zero graded vector spaces and We ? 0. Moreover, H^AU<g> AW,d) ^ H^X; jfe) = 0, i > ??. We shall show that this is impossible. Denote by (AW, d) the quotient Sullivan algebra and let b be a variable of arbitrary degree. The main step in the proof is Lemma 33.4 In the situation of the proof of Theorem 33.3 suppose given a morphism of the form g : (AU,d) —> (??,?). Then the identity in Ab extends to an isomorphism of cochain algebras Ab ®au (AU ? AW, d) = (Ab, 0) ® (AW, d) . proof: Given any Sullivan algebra (AV, d), define V by Vk = Vk~1 and denote corresponding elements by ? <—> ?. Define a commutative cochain algebra (AV ®(Jc®V),S\ as follows: • Multiplication by elements in AV is just multiplication on the left in AV <g> V), and V-V = 0. • ? = d + s, where s is the derivation defined by sv = ? and sv = 0, and where ?(? <g> 1) = ?? <g> 1 and dv = -sdv. This construction is natural: if ? : (AV, d) —> (AZ,d) is a morphism of Sullivan algebras, extend it to ? : (AV ®(lk®V),s) —-> (?? ® (? ® ?), ?\ by setting ?? = sipv. Moreover, division by V defines a natural surjective cochain algebra morphism ? : (AV ®(&®?),?) —> (AV, d). In particular, denote the cochain algebra (AC/ ® (k ? U), ? ) simply by (?, ?). The key observation is then that there is a relative Sullivan algebra
456 33 Growth of Rational Homotopy Groups such that applying AU ®a — yields the original relative Sullivan algebra AU —> AU ® AW. The derivations d in .4 and in AU ® AW extend uniquely to a derivation d in A® AW = A® au (AU ® AW). Now extend the derivation s in .4 to a derivation s in A ® AW by constructing its restriction to W as a linear map W —> AU ® U ® AW such that dsw = -sdw, w ? W. Then sd + ds = 0, and s2 = 0 because s will automatically vanish in AU ® C/ <8> AW. Thus J = d + s is a differential. Suppose by induction that s is defined in W<r. If w ? WT, then by construc- tion dsdw = —sd2w = 0, and so sdw is a d-cocycle in AU ® C/ <8> AW. Since r > ? > k + ?? it follows that degsdw > k + ?? + 2. On the other hand, U is concentrated in degrees <k + l and H(AU ® AW, ci) is concentrated in degrees < ??. Hence //'(AC/ <8> f/ <g> AW, ci) = 0, ?" > k + nx + 2, and sdw must be a d-coboundary. Thus we may extend s to WT so dsiu = —sdw, w ? Wr. This constructs ?. Next, given ? : (AC/, d) —-> (Ab, 0) extend it to ? : (?, ?) —+ [Ab ® (Jk © ieb),s\ as described above. Define the quotient cochain algebra (?,?) by IAb ® (Jk ® Jkb), degoiseven. ?6 ® (ic ? 3cb)/bb, deg ? is odd. Set E ® AW,5) = 5 ®? (? <8> AW,5). By inspection H(B,8) = A, and so the quotient map ? : (B ® AW, ?) —> (AW, d) is a quasi isomorphism (Lemma 14.2). Choose a morphism ? : (AW,d) —> (B ® AW, ?) such that ¦?? = id (Lifting lemma 12.4); then multiplication defines an isomorphism id-? : (?,?) <g> (AW,d) -=» (B <S> AW,?), as follows by filtering using the de- gree of B. Finally, dividing by b gives a morphism (?,?) —> (Ab,0) and (?6 ???/ (AC/ <8> AW, d) ^ Ab®BB ®A (A ® AW, ?) s (??, 0) ® (AW, d) . ? We now return to the proof of Theorem 33.3, where we show that • Both ? (AW, d) and H(AU,d) are finite dimensional. C3.5) Suppose C3.5) is proved. Filter AU® AW by setting FP(AU®AW) = (AU)-P® AW and observe that the ?2-term of the corresponding cohomology spectral sequence is given by E%'Q = Hp (AU ,d)® Hq (AW,d). If ? and q are the maximum degrees in which H(AU,d) and H(AW,d) are non-zero then for obvious degree reasons d,- = 0 in Ef'\ i > 2 and E?'9nlmd; = 0, i >2. It follows that ?™ ? 0 and hence that HP+Q(AU <g>AW,d) ? 0. But (AW,d) is a non-trivial minimal Sullivan model and W = W-e with ? > k + nx > ??. This contradicts the hypothesis that HX(AU ® AW, d) = 0, i > ? ?, and establishes the theorem. Now we have to prove C3.5). Fix a basis u\,... ,ur of U such that degtti < deg u-2 < · · · and let (AC/; ® AW, Di) be the quotient Sullivan algebra obtained by
The Rational Dichotomy 457 dividing by m,..., Uj_i. We show first by induction that each H (AUi ® AW, A) is finite dimensional so that, in particular, H(AW,d) is finite dimensional. Indeed suppose H(AUi<S>AW, Di) is finite dimensional. If degu* is even extend this to the Sullivan algebra (AUi ® AW ® Ab, Di) by setting ?? = ui; and note that this is quasi-isomorphic to (AC/j+i <g> AW,Di+1). Thus the long exact cohomology sequence associated with 0 —»· (AUi®AW,Di) —»· (AUi®AW®Ab,Di) —»· (?[/;®??··', ?><) ® (Jfco, 0) —»· 0 shows that //(At/j+i ®AW,Di+i) is finite dimensional. On the other hand, if deguj is odd write ?>*(]. <8>?) = l®.Dj+i<i> + Uj<g>o<i>. ? G AC/j+i <g>AW. Let ? be the maximum degree such that HN (AUi® AW, DJ ? 0. If ? is a .Dj+i-cocycle representing [?] ? ? (AUi+i® AW, Di+i) and ifdeg? > ? then U{ ® ? is a .Dj-cocycle and hence of the form Dj(uj <g> ? +1 <8> ??). It follows that .Dj+??? = 0 and ?(?)[??] = [?]. This procedure constructs an infinite sequence [?] «— ?(?)[??] 4 . But division by (uj), j ? i defines a surjection (AC/, d) —> (Auj,0). According to Lemma 33.4, Auj ®AU (AU ® AW,d) ^ (Aui,0) ® (APF, J). It is immediate that if xi,... ,xr is the dual basis of the homotopy Lie algebra L of (AU,d) then the holonomy representation hi' of L in H(AW, d) satisfies hl'(xi) = 0 (Introduction to §31). Since L is finite dimensional adz,- is trivially nilpotent. Hence (Lemma 31.9) ?(?) is locally conilpotent and [?] = 0; i.e. H>N(AUi+l ® AW,Di+i) = 0. This shows that H(AUi+l ® AW,Di+i) is finite dimensional. It remains to show that H(AU,d) is finite dimensional and for this we lose no generality in assuming Jk algebraically closed. Consider an arbitrary morphism ? : (AU,d) —> (??,?) where dego = 2. It extends to a morphism ? : (AU ® AW, d) —> (Ab,O) defined as follows: use Lemma 33.4 to identify Ab®au (AU ® AH7, d) as (??,?) <g> (AW,d) and let ? be the composite (AU ® AW, d) —»· Ab ®au (AU ® AW, d) —»· (?6,0) . Suppose now by induction that ??,\ = ¦ ¦ ¦ = nui-i = 0. If deguj is odd then nui = 0. Otherwise factor ? to give a morphism ? : (AUi ® AW, Di) —> Ab. Since ? (AUi ® AW,Di) is finite dimensional and since D{Ui = 0, it follows that uf = Dfi, some N. Hence 7r(uf ) = n(uf) = W(Diu) = 0; and it follows that ¦nui = 0. Thus ? = 0 in U and the only morphism (AU, d) —> (Ab,0) is the trivial one. Now Proposition 32.3 asserts that H(AU,d) is finite dimensional. This completes the proof of C2.5) and hence of the theorem. D Proposition 33.6 (Lambrechts [104]) Suppose X is as in Theorem 33.3. Then for r sufficiently large, d\mH*(X;< ;=r+i v '
458 33 Growth of Rational Homotopy Groups proof: Let (AV, d) be the minimal Sullivan model of X and set ? = dimiT(X;Q) = dim H {AV, d). If the inequality of the Proposition fails for some r > nx +1 we show by induction that for 0 < k < ?? — 1, (AV, d) can be decomposed as a relative Sullivan algebra of the form (AUk®AWk,dk) in which Wk = (Wk)-r, (WkY = 0, r+1 < i < r + k and ?+??—1 Y^ ? dim Wi < dim Wk . Since Vk = irk(X) ® Q, this can be achieved for k = 0 by setting Uk = V<r. For the inductive step, let w ? (Wk)r+k+l and write dw = aw + ?w with aw ? ) ® W? and ?w ? AUk. Since cPw = 0 it follows that aw ? kerd® W?. Let k1 H (a) : (Wk)r+k+1 —> ffiAV)®^ be the induced map and write ???? = ?® ?' where ? is the smallest subspace such that lmH(a) C H(AV) ® Z. Then by replacing a basis Wi of Wr+/:+1 by elements of the form w[ = Wi + ?{ with ?,· ? AUk ® Z' we may arrange that ? : Wr+k+1 —> AC/* ® (Jfe ? ?). Set C/fc+i = Uk ? ? ® (Wk)r+k+1 and set VKfc+1 = ?' ® Wj%+k+2. Since (clearly) iVdim(Wfc)r+/:+1 > dim ? it follows that ' ? ? dimW* < dim ?'. This i=r+k+2 closes the induction. For k = ?? — 1 we thus obtain a decomposition (AV, d) = (AU <g> AVF, d) in which W = W-r, Wr ? 0 and Wl = 0, r + 1 < i < nx + r - 1. Extend a non- zero linear function / : H/r —> ?c to a derivation 0 in (AC/ <g> AWr, d) by putting O = 0 in U. Suppose ? is further extended to (AU <g> AW^m,d). \i w ? Wm+l then 6dw is a cocycle of degree at least ?? + 1, and hence is a coboundary in AV. Thus we can extend ? to Wm+1 so that Odw = duw, w € Wm+l. This exhibits / : Wr —> Ik as a non-zero Gottlieb element (§29(d)) for (AV, d). Since cat(AVr, d) is finite there are such elements in only finitely many degrees r (Proposition 29.8 (ii)), and hence the inequality of this Proposition holds for sufficiently large r. ? (c) Loop space homology. For any topological space X with rational homology of finite type, the Poincare oo series for X is the formal power series ?? = X)dimfin(X;Q)zn. In particular, ? if X is simply connected the Milnor-Moore theorem 21.5 states that the loop space homology of X is isomorphic to the universal enveloping algebra of the homotopy Lie algebra Lx : H*(UX;Q) = ULx. Denote by ? (or by r{(X) when the dependence on X is not clear from the context) the integers ?? = dim ?» (?.?) <g> Q = Then the Poincare-Birkoff-Witt theorem 21.1 identifies Pqx as the formal power
The Rational Dichotomy 459 ?(? + *2?+1?+1 ??? = ^ - C3.7) (Compare with the formula in §32(d) for ?? for certain rationally elliptic spaces!) Note that C3.7) provides algorithms for computing the integers dim Hi(flX; Q). 1 < i < ? from the integers dim7Tj+i(X) ® Q, 1 < i < ? and for computing the integers dim7ri+i(X) ® Q, 1 < i < ? from the integers dimHi(ttX;Q), 1 < i < N. Proposition 33.8 Suppose fT(X;Q) = 0, i > nx. Then the integers dimHi(ClX;Q), 1 < t < 3(?,? — 1) determine whether X is rationally elliptic or rationally hyperbolic. proof: Theorem 33.3 asserts that X is rationally elliptic if and only if ttj (X) <g> Q = 0, 2?? < j < 3?? — 1. This only requires the calculation of ru 2?.? — 1 < i < 3nx - 3. D Proposition 33.9 (i) If X is rationally elliptic then there are constants 0 < A < ? such that ? ??? < ? dim Ht(IIX; Q) < Bnr , ? > 1 , where r = (ii) If X is rationally hyperbolic and if Hl(X;Q) = 0, i > ?? then there are constants C > 1 and ? such that k+2(nx-l) ? dimHi{nX;Q) > Ck , k>K. i=k+l proof: The first assertion is immediate from C3.7). For the second let (AV, d) be the minimal Sullivan model of X. An element ? of V of least degree is a cocycle of degree ? < ??. If ? is even then vp is a coboundary for some ? such that (p — l)n < ??. Thus whether ? is even or odd it follows that V has a non-zero element of odd degree 2k + 1 < 2?? — 1, and so Lx has an element 00 of even degree 2k < 2?? - 2. By C3.7), ??? - 1Jzak ? riz* is a power series i=2 with positive coefficients. Now apply Theorem 33.2. ? The radius of convergence R of a power series Y^anzn is the least upper bound of the nonnegative numbers r such that X) |an|rn converges, and is given
460 33 Growth of Rational Homotopy Groups by A = limsup lunl1/". Recall that X is supposed simply connected with rational homology of finite type. Proposition 33.10 The formal power series Pqx and ? rnzn have the same radius of convergence, R. Moreover (i) R = 1 if X is rationally elliptic and R < 1 if X is rationally hyperbolic. (ii) If X is rationally hyperbolic and if ??(?;^ = 0, { > ?? then R < ? < 1 for some constant ? depending only on ??. proof: Write Y^anzn <C Y^bnzn if an < bn for all n. Since it follows by C3.7) that X)rn2n is convergent if and only if Pqx is convergent. If X is rationally elliptic the products in C3.7) are finite and R = 1. If X is rationally hyperbolic let cato X = m. The proof of Theorem 33.2 shows that given ? sufficiently large there is an infinite sequence of integers Ni such that ? = No, Ni < Ni+i <(m + l)iV; and /2N-2 j=N J where cm is a constant depending only on m. It is also shown that for ? suffi- ciently large the right hand side is larger than one. It follows that lim sup r\[n > 1. Moreover, Theorem 33.3 states that if i/l'(X;Q) = 0, i > nx, then when ? = 2n.\;e3x the right hand side is at least 21/N, from which assertion (ii) follows at once. ? Example 1 Wedges of spheres- Suppose X = V 5n>+1 with rii > 1 and at most finitely many spheres of a fixed i degree. Then the homotopy Lie algebra is a free Lie algebra Ly on a graded vector space with basis (v{) and degUj = ni (Example 1, §24(e)). But ULv is the tensor algebra TV (§21(c)) and thus ?-?>?; i Example 2 X = S3 V S3.
The Rational Dichotomy 461 As in Example 1, PQX = ^^ and hence Ht(UX;Q) is concentrated in even degrees. Thus formula C3.7) becomes l-2z2 where r2n = dim7r2nQ(S3 V S3) <8> Q- Take logs of both sides to obtain y2nk ?? ? k ? k Equating the coefficients of z2N gives J^ v^da = 2N. d\N The Mobius function ?(?) is defined by: ?(?) = 1, (—l)r or 0 as ? is 1, a product of r distinct primes or divisible by a prime squared. Elementary number theory gives as a precise formula for the dimension of ?2^?(?-?) ® Q· As observed above, ) 0. D Example 3 X = S? V Sf U[a[a<?]w]w D8. Here ? and ? are the elements of ?3 (S? V Sf) represented by Sf and S|, and [—,— ]w is the Whitehead product. According to Example 4, §24(f) X has a Lie model of the form (L,d) = (L(v,w, u),d) with degu = degu; = 2, deg u = 7, du = [?, [?, w]} and dv = dw = 0. In degrees < 8 the cochain algebra C*(L,d) coincides with (A(x3,y3, z5,r7, S7,as),d) where x,y,z,r,s and a are dual (up to sign) respectively to v, w, [v, w], [w, [w, v]], [v, [v, w]] and u, and where dz = xy and dr = yz, ds = xz — a and da = 0. Dividing by elements of degree > 8, and by r and y ? gives a commutative model for X of the form (A,d) = (A(x,y,z),d)/(yz). The quotient map q : X —> Sf is represented by the inclusion (Ax, 0) —> {A,d). Extend this to a quasi-isomorphism from a relative Sullivan algebra, (Ax <g> AV,d) -=> (A,d). Since {A,d) is Ax-semifree this induces a quasi- isomorphism k <8>?? (Ax <g> AF, d) -^> Jfe <8>?? (A,d) — Proposition 6.7 (ii). But -& ®Ax {Ax <8> AV, d) = (AV, d) is a Sullivan model for the homotopy fibre F of q (Theorem 15.3), while k <g>Al (.4,d) = (A(y,z)/yz,0). This shows that F=Q53V55. Let j : F —> X be the map corresponding to the fibre inclusion of the fibration associated with q : X —> Sf and let i : Sf —> X be the inclusion. Since qi = id, ?? · Clj : USf x ?.? —> ?? is a weak homotopy equivalence. In particular, (cf. Example 2) D
462 33 Growth of Rational Homotopy Groups Example 4 X = T{S3,S3,S3). Recall that S3 is a CW complex with a 0-cell and a 3-cell; thus the product S3 x S3 x. S3 has 8 cells consisting of a 0-cell, the three spheres, three 6-cells attached by Whitehead products and a 9-cell. The 6-skeleton of S3 x S3 x S3 is called the fat wedge T(S3,S3,S3). The inclusion q : X —> S3y.S3y. S3 can be represented by ? : (A(x, y,z),0) —> (A, d) where A is a commutative model concentrated in degrees < 6. It follows that A = A(x,y,z)/xyz. If we extend ? to a quasi-isomorphism (A(x,y.z) <8> AV,d) —> (A,d) from a relative Sullivan algebra then, as in Example 3, the quotient Sullivan algebra (AV, d) is a Sullivan model for the homotopy fibre F of q. But there are quasi- isomorphisms (AV, d) ~ (A(x, y,z) <g> AV <8> A(x,y, z),d) = (? ® A(x,y,z),d), where dx = x, dy = y and dz = z. Consider the short exact sequence 0 —> Ikxyz ® A(x,y,z) —> A(x,y,z) <g> A(x, y,z) —> A <g> A(x, y,z) —> 0 . Since the central term is contractible it follows that every cohomology class in H (A ® A(x, y,z)) is represented by a cocycle in A2(x,y,z) ® A(x,y, z). The product of two such cocycles is zero and it follows that F is formal and thus has the rational homotopy type of a wedge of spheres (Proposition 13.12 and Theorem 24.5). The long exact cohomology sequence also shows that Pp = A_fT3K ¦ Thus (Example 1) \-??/{\-?2K A- Since n*(q) is (trivially) surjective it follows that 1 (l-22K-27 ' It follows in particular that dim7r3(X) ® Q = 3, dim7rg(X) <8> Q = 1 and that ???(.?) <8> Q = 0, 4 < i < 7. This interval of four consecutive zero rational homotopy groups is the maximum allowed by Theorem 33.3, since H{(X; Q) — 0, i > 6. ? Example 5 X = Y V ?. Suppose Y and ? are simply connected CW complexes. Let j : F —> ??/ ? be the 'inclusion' of the homotopy fibre of the inclusion q:Y\l ? —> ? ? ?. The construction of F (§2(c)) identifies it as F - ?.? ? PZ [j PY ? ??. Since QYxQZ the pairs (??,?,?) and (CQ.Y, ?.?) are homotopy equivalent (elementary), and similarly for ? we have F ~ QF ? CQZ (J CQF ? ?? = ?? * ?? ~ ?(?? ? ??) , QY'xQZ
The Rational Dichotomy 463 (A.15) and Proposition 1.17 — note that (?,?,??) and (?,?,??) are well based by Step 1 of Proposition 27.9). This shows that F has the rational homotopy type of a wedge of spheres (Theorem 24.5). NowsinceF~?(ri:KAQZ),PFB) = z(PqY-1)(Pqz-1). Thus (Example 1) since ?, (q) is clearly surjective, and ? - (??? - \){??? - ?) "? ? - (??? - \){??? - ?) · Simplifying gives the classical formula Exercises 1. Let X be a simply connected CW complex of finite type. Suppose that ??(?) <g> Q = 0 for r > n, and HP(X;Q) = 0 for ? < ? < 2n + 1. Prove that H?{X;Q) =0forp>n. 2. Denote by R the radius of convergence of the series ?)n dimi/n(QX;QJn. If X is a hyperbolic space prove that for large r and for any ? > 0, we have where m = 3. Let /?? be the radius of convergence of ?*(??; Q)[z] and P(z) the Poincare series of H*(X;Q). Suppose that dimH*(X;Q) < oo. Prove that Rx > min{|2|, ? ? 0 and ? - ? (?) + 1 = 0}.
34 The Hochschild-Serre spectral sequence In this section the ground ring is an arbitrary field h of characteristic zero. In this section we fix a graded Lie algebra, L = {L{}ieZ, with universal en- veloping algebra UL. If M is a left (right) [/L-module then restriction of the action of UL to L defines a left (right) representation of L in M, and this correspondence is a bijection (§21(a)). We use this to identify L-modules as UL-modules and conversely. In particular, the trivial representation of L in ic defines a canonical (left and right) UL-module structure in Ic. Now suppose / C L is an ideal and ? is a right L-module. Restriction to / makes ? into a right /-module, and so the graded vector spaces Ext^/(A, N) are defined (§20(b)) via any [//-projective resolution of Ik. It turns out (§34(a)) that the L-module structure in ? makes each Ext?)/(Jc, N) into a right L/7-module. Thus if M is any right L//-module the graded vector spaces are defined. On the other hand, the quotient map L —> L/I makes M into a right L- module. The main objective of this section is to construct a spectral sequence, due to Hochschild and Serre [92], which converges from E%'9 = ExtpUL/I(M,ExtqUI(fc,N)) => Ext?+9(M,jV) . The case ? = UL will be of particular importance in the applications. The key notion in this section is that of a chain complex of UL-modules. Recall from the introduction to §20 that this is a complex of the form 0 <- POf. 4- Plt. 4- in which each P^» is a (graded) UL-module with (Pi,*)j = Pi,j-i, and d is homogeneous of bidegree (—1,0). We shall adopt the Notation convention: A chain complex of C/L-modules will be denoted by P, = {Pi}, with Pi denoting the UL-module Pit*. If ?? is a UL-module then Hom?/?,(P*; N) denotes the complex 0 —»· EomUL (Po, ?) ? Homf/L (?, ?) ? - · · of graded vector spaces with ? homogeneous of degree 1 (bidegree A,0)). This section is organized into the following topics (a) Horn, Ext, tensor and Tor for C/L-modules. (b) The Hochschild-Serre spectral sequence. (c) Coefficients in UL.
The Rational Dichotomy 465 (a) Horn, Ext, tensor and Tor for C/L—modules. Let M and ? be right L-modules. A natural right L-module structure in Hom(M, N) is then defined by (f-x)(m) = (-1)*** d^(f(m).x- f(m.x)) , { ? Hom(MJV), Clearly for any ideal / C L, ?????^?,?) = {/ e Eom(M,N)\f-x = 0,x ? /} . Thus Hom{//(M, N) is a sub L-module in which / acts trivially; i.e., it is an L//-module In particular, let ?, -^ ? be a C/L-projective resolution (§20(a)). The Corollary to Proposition 21.2 implies that P. is also C/Z-projective, and so Exti//(M. N) = H (Hom?//(P.. N)). But the remarks above identify Hom?//(P,, iV) as a complex of L//-modules. which endows H (Homf//(P«,iV)) with a right L//-module structure. Moreover, given a second such resolution Q* —>· M. the unique homotopy class of L-linear quasi-isomorphisms Q* -^ P. induc- ing the identity in M induce L//-linear quasi-isomorphisms Homi//(P«,./V) ^=> Homi//(Q.,iV). Thus the identification H (RomUI(P,,N)) = H (Homy/iQ.,^)) is an identification of L//-modules and so this endows each Ext?)/(il/, N) with the structure of right L/7-module. Definition The L//-module structure above is called the canonical LjI-struc- ture in Extui(M,N). Again consider two right L-modules M and N. As with Hom(M, N), the tensor product M <g> ? inherits a right L-module structure: (m®n)-x = (-l)degndegIm-x(8)n + m(8)n-x , ? ? L, m ? ?, ? € ? . As usual, let (M®N) · L denote the subspace spanned by the elements (m<8>n) · x. Lemma 34.1 If M or ? is UL-free then M <g> ? is UL-free. proof: Suppose ? is C/L-free on a basis aa. Since M <g> iV = 0 M <g> (?^ · C/L) it is sufficient to prove that M <g> UL is C/L-free. Let Fp C C/L be the linear span of elements of the form x^ x;9, ?,„ ? L, q < p. If m G ?/ and a ? Fp then (m<8>l)-a — m®fl? M <g> Fp_i. It follows that M ® UL is C/L-free on a basis m\ <8> 1 where m\ is any it-basis of M. The proof when ?/ is C/L-free is identical. ? Next observe that a left L-module structure in ? is defined by x.n = -(-l)degldegnn-x . n?N,x?L.
466 34 The Hochschild-Serre spectral sequence Moreover, (M ®N) · L is the kernel of the surjection M®N —> M®ULN and of the surjection M®N —> (M® N) ®ul &¦ This provides a natural identification M ®UL ? - (M ® N) ®UL h . C4.2) Furthermore, if / C L is an ideal then (M <g> N) · I is an L-module and so the quotient ? ®?? ? — (? ® ?) ®ui 3k inherits a natural structure of right L//-module. Recall (§3(a)) that we denote Hom(-, Jfc) simply by (-)K Lemma 34.3 Suppose M and ? are right L-modules and that I c L is an ideal. (i) Each Tor^ (M,N) is naturally a right L/I-module. (ii) There are natural isomorphisms ????(?, ?) S Tor^/(M®iV, k) ofL/I- modules. (in) There are natural isomorphisms Ext^/(M, iVs) = Tor^//(M, iV)" of L/I- modules. proof: Let P, —> Mbea i/L-projective resolution (§20(a)). It is i//-projective by the Corollary to Proposition 21.2. (i) Note that ?????(?, ?) is the homology of the complex P, ®VI ? of right L/7-modules. (ii) Note that Pt®N^M®Nisa, C/L-projective resolution (Lemma 34.1) and hence that TorUI(M®N, k) = ? ((P. ® N) ®VI k). But (?, ® N)®UI& = P. ®ui N as L/7-modules. (iii) Write Tor^(M, N)* = {Tor^(M, N)s} and (P. ®ui N)* = {(Pk ®VI iV)« But clearly Horn (M ®ui N, Je) = Homf//(M, Ns), and since Ik is a field it follows that Tor^M, Nf = H ((P. ®ui N)s) = H (Hom?//(P., iVd)) = Ext^(M, iV«) . Example 1 Tor^^k, k) = s (//[/, /]). Again let I C L be an ideal. Denote by [/, I] the ideal in L which is the linear span of the elements [y, z], y,z G /. If ? G / denote by (x) the image of sx in the suspension s (//[/,/]). We shall establish an isomorphism Torf7 (A, A) = 5 (//[/,/]) and show that the representation of L/I in Tor^/(ic,i:) corresponds under this isomorphism to the representation
The Rational Dichotomy 467 In fact, in §22(b) we constructed a Ul-ivee resolution P» (Propositions 22.3 and 22.4) of k explicitly, of the form k<—UI 4- sI®UI 4- A2 si ®UI 4 . Apply - <8>ui k and note from the definition that in this quotient d(sx A sy) = (-l)degsIs[x,y]. Thus Tor^i*, Jb) = sI/dA2sI = s (//[/, /]) . Exactly the same way we have a UL-kee resolution Q* of k of the form Jfc <— t/L A sL®UL^- A2 4 and the inclusion of P« in Q* is a quasi-isomorphism. Let ? G / and y G L. Then the formula for d in §22(b) gives d{sx A sy ® 1) = (-l)degSI+degsysx ®y±sy®x + (-l)degsxs[x, y] ® 1 . Applying ®mk kills sy ® x, while sx <S> y = (sx <S> 1) · y. Thus in homology we have (x)-y.= (-i)degi([x,y]) , xel,yeL. ? Example 2 The representationof L/I in Torj^ (k,k). As in Example 1 the inclusion of P, in Q* is a quasi-isomorphism of Ul-kee resolutions of k, and in particular applying — ®ui k gives a quasi-isomorphism (As/, d) -=> Qt ®ui k. Let a G A9s/ be a cycle (representing a class ? € Torj^A, A) and let y € L. Then the formulae of §22(b) show that in Q. <S>ui k, where ? · y is defined by (SX! ? · · · ? SX,) · y = (_l)deg(sxiA-Asx,) J-> (_l)deg j/ ???(???+??.-??3:,M?? ? . . . ? s[x,-,y]A ··· ? sx,. It follows that the representation of L/I in ???1^1 (k, k) is given by (b) The Hochschild-Serre spectral sequence. In this topic we construct the spectral sequence. Fix an ideal I C L and a right L//-module M and a right L-module N. For any graded vector space S an isomorphism ? : Hom(M <s> S, N) -% Horn (M, HomE, iV))
468 34 The Hochschild-S erre spectral sequence is given by (pf(m)(s) = f(m <g> s), f G Hom(M <g> S,N). Lemma 34.4 If S is a right L-module then ? restricts to an isomorphism EomUL(M ®S,N) A Homw// (M, Hom^S, ?)) . proof: It is immediate that ? is L-linear. Thus if / G Hom?/?,(M ® 5, iV) we have f -x = 0, ? e L and so (</>/) ·? = 0. Thus y?/ is L-linear. For ? e I and m G M it follows that (?/)(??)·? = ?/(??·?) = 0. Thus ?/(??) is /- linear; i.e. ? : ??????,(? ® 5, ?) —> Hom^// (M, Hom^S, ?*)). Conversely if / G Hom(M ® 5, iV) and if </?/ G EomUL/I (M, HomujiS, N)) then ?/ is L-linear and hence so is /. D Now we construct the spectral sequence. Choose a UL/I-free resolution em '¦ Pt -^> M and a t/L-free resolution ? : Q* —> Jk. Denote the tensor product of these two complexes by (J2., d) = (P., d) <S> (Q*,d) : Rk = 0 Pp® Qg. (Note as in §20(a) that each Pp and Qq is itself a graded vector space!) Observe that • Each Rk is a free UL-module (Lemma 34.1). • em <S> ? : R» -^> M is a UL-free resolution (because ik is a field). Now Ext?/L(M, N) is the homology of the complex HomuLIR*, N) and this is in fact a bigraded complex with (A.,iV)= 0 Homf/L (Pp ® Qq, N) . Refer to A; as the total homological degree to distinguish it from the 'normal' degree inherent in the fact that ? and Rh are each graded vector spaces. Filter the complex EomUL(Rt, N) by the subspaces 0 Homt/L (Pi <S>Q*,N). i>p Use this and the total homological degree to obtain a first quadrant cohomology spectral sequence, convergent to Ext{/i,(M, N) and with ??'9 = EomUL{Pp®Qq,N) . Lemma 34.4 identifies HomUL (Pp®Qg,N) = Hom^// (Pp,Hom?//(Q9,iV)). The differential do is induced from the differential in Qt. Thus because each Pp is [/L/7-projective we have E['q = = Homl/L//(Ppi#«(Homl//(Q*liV))) = Homi/L//(Pp,Ext9w(l;,iV)) ,
The Rational Dichotomy 469 where Ext^ (A, iV) is equipped with its canonical right L//-module structure. The differential, di, is then induced from the differential in P., so that E™ = Ext* L// (M, Ext^A, N)) . It is immediate that this spectral sequence is independent of the choice of reso- lutions P, and Qt and so we may make the definition: The spectral sequence above, converging from Ext^L// (M, Ext^A, N)) to Ext?^9(M, N) is called the Hochschild-Serre spectral sequence. Remark If / acts trivially in ? then the u^-term of the Hochschild-Serre spectral sequence may be identified as where Tor^A, A) has the UL/I structure of Lemma 34.3(i). In fact in this case ??'" = Homi/L (PP®Qq,N) = HomUL/I(Pp®ULQq,N) = HomUL/I(Pp <S> (k ®ui Qq), This identifies E™ = Honu/i// (PP ® Tor^i*, A), N) with L/I acting diagonally in ?, ® Tor^A, A). By Lemma 34.1 the differential in P. makes P. ® Tor^(A, A) into a UL/1-ivee resolution of M ® Tor^(A, A). This gives E™ = ExtP L/I (M ® Tor^(A, ?), ?) , as desired. (c) Coefficients in UL. In this topic we suppose that the graded Lie algebra L satisfies: L = L>i and each Li is finite dimensional. It follows from this that UL is a graded vector space of finite type. Let / C L be an ideal and choose a graded subspace W C UL so that mul- tiplication in UL defines an isomorphism W <S> UI J=L> UL of right UI modules (Corollary to Proposition 21.2). The proof of the Corollary shows that the surjec- tion UL —> UL/I restricts to a linear isomorphism W -^ UL/I. In particular, this endows W with the structure of right L//-module.
470 34 The Hochschild-Serre spectral sequence Now let M be any graded vector space of the form M = {Mi}i>Q. Since each Wk is finite dimensional a natural isomorphism Wk ® Hom(M, UI) A Horn (M, Wk ® ?//) is given by u> ® <? ?—> h, with h(m) = w ® ^(m). Thus an isomorphism Hom(M, ?/L) A JJ W'fc ® Hom(M, UI) k is defined by / ?—> (fk)·. where for m G ?, fk(m) is the component of f(m) in Wk ®UI. (Note that given any m G M and any sequence /^ G W\ ®Hom(M, UI) of degree ? we have ?-(??) G ??^ ® UIdegm+p_k. Thus ?-(??) = 0, k > degm + p and the sum ? /fc(m) is finite.) Now suppose M is a right L-module. The isomorphism above then restricts to an isomorphism ? : Homt//(M, UL) A JJ Wfc ® Homy/(M, UI) . A: The infinite product contains the direct sum 0 Wk ® Homi//(M, C//) as a sub- space, and this is just W ® Homf//(M, C//). Recall from §34(a) that Homer/ (?/, I7L) is a right L/Z-module. Thus \\Wk® k Homer/ (M, UI) inherits a right L//-module structure via ?, while W has a right L//-module structure through the isomorphism W J=L> UL/I. Note that the inclusion W<g>Homer/(MrUI) —> fl wh ®Homer/(M, I7/) iden- *.· tifies each VFfc ® Homer/(M, C//) as a subspace of the infinite product. A simple calculation gives Lemma 34.5 If ? € L/I, w G Wk and f G Homer/(M, UI) then (w®f)-x- (-l)d*sf a^xw -x®fe W<k ® Homer/(M, UI) . Now let ? : ?» ^> k be a C/L-projective resolution of ic. Recall (§34(a)) that the canonical L//-module structure in Ext^r/(i:, UL) is obtained by identify- ing Extui(k,UL) = H (Homer/(F«,C/L)) and using the L//-module structure induced from that in Homer/(-P*, UL). As above, identify Homer/(F.,C/L) = Y[wk ® Homer/(F., C//) and hence identify
The Rational Dichotomy 471 Lemma 34.5 then implies that for ? G L/I, w G Wk and a G ExtqUI(k,UI) we have (w ® ?) · ? - (-l)desQ d^xw -x ® a G W<fc ® Ext^Jfc, C//) . C4.6) It follows that W ® Exty/(A', UI) is an L//-submodule of the infinite product. Proposition 34.7 (i) W ® Ext^Jb, UI) is a free UL/I-module on 1 ® Ext^Jfe, UI). (ii) If either UL/I or Ex.X1qUI(k,UI) is finite dimensional then W ® Ext^i*, UI) = Extg^Jfc, UL) . proof: Denote Extfj^k, UI) by Eq. (i) An L//-linear map ? from the free L//-module Eg ® [/I,// to M^ ® given by , ? e Eq, aeUL/I . (We identify C/L// = IV.) It follows from formula C4.6) that ?(? ® ?) - (_1)??8? degoa (g, ? G Pi/<dega ® ?:«. This implies that 0 is an isomorphism, (ii) If either UL/I or Eq is finite dimensional then = J| W/t ® ?« = Ext^Jfc, C/L) . Recall again, that L is a graded Lie algebra satisfying L = {?t}f>1 and each Li is finite dimensional. Suppose further that • / C L is an ideal and that ? is an L-module. Write UL = W ® C// as above. Then (C/L)8 = (C//)8 ® PF» and Tor^ (TV, (t/L)*) = Tor^ (iV, {Ulf) ® W» . C4.8) The surjection I7L —>· t/L// restricts to an isomorphism W -^ UL/I, which dualizes to (UL/I)* -=> Ws and so makes W* into an L/I module. Thus the left hand side of C4.8) is an L//-module because ? is an L-module (Lemma 34.3) and the right hand side is an L//-module via the action in W". However, C4.8) is usually NOT an isomorphism of L/7-modules. Even so, it is often possible to construct an isomorphism that does preserve the L/I- module structure, as the next proposition shows. Recall (Lemma 34.3) that each TorJ^jV, A) is an L/J-module.
472 34 The Hochschild-Serre spectral sequence Proposition 34.9 With the hypotheses and notation above assume the Lie algebra L/I is finitely generated and that a · UL/I is finite dimensional for each a G Tor^!(N, He). Then there is an isomorphism of U L /1-modules Tor^ (TV, (ULf) Si Tor^ (N, {Ulf) ® (UL/lf where L/I acts on the right via the action in (UL/lf. In particular, (Jfc, TorJ" (N, (ULf)) = Tor^' (Jk, (UL/I)*) ® Tor^ (N, (Ulf) . Lemma 34.10 With the hypotheses of Proposition 34.9, ? · UL/I is finite di- mensional for all ? G Tor^ (N, (ULf). proof: Write M = (ULf. Then M = M<0 is the union of the submodules M>-p. It is thus sufficient to prove the lemma for ? e Tor^ (N, M>~p). Con- sider the exact sequence TorJ" (N, M>_p) -> Tor^ (N, M>_p) -> Tor^ (?, ??_?) , where UL acts trivially in M_p. By hypothesis the lemma holds for this trivial C/L-module. Hence for ? e Tor^ (Ar,M>_p) and some ? > 0, ?-(UL/I)>n c Image Tor^ (N,M>-P). Because L/I is a finitely generated Lie algebra each (UL/I)>n is a finitely generated t/L//-module. Hence so is ? -(UL/I)>n. By induction on ? it follows that ? -(UL/I)>n is finite dimensional. Hence so is ? · UL/I. a proof of Proposition 34.9: Dualize the inclusion UI —> UL to a surjection (ULf —> (Ulf of [//-modules. This induces a linear map / : Tor^ (TV, (ULf) -> Tor^ (N, (Ulf) , and C4.8) shows that / is surjective. Right multiplication in UL/I makes Uom(UL/I, —) into a left (and thus right) UL/I-modu\e. Define a morphism F : TorJ" (TV, (ULf) —> Horn (UL/?,???^1 (N, (Ulf)) of UL/I-modules by setting F(a)(a) = f(a-a), a € TorJ" ((N, (ULf), a G UL/I. Recall that an inclusion ? : TorJ" (N, (Ulf) ® (UL/lf —> Horn (UL/I, TorJ" (iV, (C//)»)) of C/L//-modules is given by ?(? ® g)(a) — {g.a)a. the image of ? consists of the linear maps ? : UL/I —> ???^1 (?,(Ulf) such that for some ?(?), ?
The Rational Dichotomy 473 vanishes on (UL/I)>n, y Thus Lemma 34.10 implies that Im F C Im ? and we may regard F as a ?/?,/7-linear map into Tor^ (N, {Ulf) ® {UL/lf. Finally, identify {UL/If = Wi and use C4.8) to write F as a map F : ???^1 (?, {Ulf) ® W" —> Tor, (TV, This F may not be the identity, but a straightforward calculation shows that (F - id) : — ® Wtn —> - <S> W>_n> ? > 0. Thus F is an isomorphism. ? Exercise Suppose / is an ideal in a graded Lie algebra L = L>\ and that each Li is finite dimensional. Prove that: (a) If Ext^iJb, UI) = 0, j < q, then Ext{,L(A, UL) = 0 , j < q . (b) If in addition Ext{//(A, UI) is finite dimensional for q < j < ? + q + 1 and if Ext\jL/I(&, UL/1) = 0 for i < ? then k JO ,k < p+ q ExtUL(lk, UL) = \ /n-o>T?„*9 (IL- TTT\ b-n-t-n \ih, ? ? ) ,?. — ? ? q ·
35 Grade and depth for fibres and loop spaces In this section (unless otherwise specified) the ground ring k is an arbitrary principal ideal domain. Recall from §20 the functors Ext^(-,-) defined for right modules over a graded algebra A, and the definition of projective dimension of an yl-module. Definition 1. The grade, grade^(M), of a right rl-module is the least integer k such that ExtkA(M, ?) ? 0. (If Ext*A(M,A) = 0 we say grade^(M) = oo.) The projective grade, projgradej4(M), is the least integer k (or oo) such that ExtkA(M,?) ? 0 for some yl-projective module P. 2. If A = {Ai}i>0 and Ao = k then the depth of .4, depth A, is the grade of the trivial yl-module k. The global dimension of A, gldimyl is the greatest integer k (or oo) such that ExtA(k, -) ? 0. The main theorem of this section provides a connection between LS category and the homological notion of grade. Recall from §2(c) that if f:X—>Y is a continuous map then a holonomy action of the loop space ? F is determined in the homotopy fibre F of /. This makes Ht(F) into a right Ht(UY)-modu\e. We shall prove: • If X is normal and (Y,yo) is well based, and if Ht(F) and ??(??) are k-free then projgrade^(fJy) Ht(F) < cat/ . Moreover, if equality holds then H*(F) = cat/ = projdim^ny) H*(F) . When each Hi(F) and Hj(QF) have finite k bases we will replace projgrade by grade in this theorem. The special case that / = id : X —> X is sufficiently important that we restate the theorem for it: • If X is path connected, well based and normal and if Ht(QX) is k-free with a finite basis in each degree then depthH*(nX) <catX . // equality holds then depth ?.(??) = catX = gldimH.(ClX). The depth theorem for topological spaces was originally deduced, for the case that k = Q and Ht(X) has finite type, from a theorem on Sullivan algebras established in a joint paper with Jacobsson and Lofwall [54]. Subsequently it was extended for spaces to Jc = Fp (with H»(X) still of finite type) in a joint
The Rational Dichotomy 475 paper with Lemaire [85] using a different but similar approach. However the theorem for Sullivan models in [54] is for a coefficient field of any characteristic, and is applied as such to the Ext-algebra of a local commutative ring. The proof given here (for the more general grade theorem) is different in form, although based on the same underlying idea. We shall, however, also sketch the proof of the Sullivan algebra theorem in characteristic zero, since it provides an interesting application of the material in §29: • If L is the homotopy Lie algebra of a minimal Sullivan algebra (AV, d) and if V = {V*} is a graded vector space of finite type then depth UL < cat(AV, d) < gl dim UL . Moreover, if depth UL = cat(AV, d) then cat(A V, d) = gl dim UL. Note: When (AV, d) is the Sullivan model of a simply connected space X, then UL = ?*(??) and cat0X = ca,t(AV,d), (Theorem 21.5, Proposition 29.4) and so the two results coincide. This section is organized into the following topics: (a) Complexes of finite length. (b) fiy-spaces and C,(fiF)-modules. (c) The Milnor resolution of Jk. (d) The grade theorem for a homotopy fibre. (e) The depth of H.(flX). (f) The depth of UL. (g) The depth theorem for Sullivan algebras. Both the grade theorem and the Sullivan algebra theorem have their roots in an elementary theorem about ? (Hom^(F,, Q»)), where A is a graded algebra, P, is an yl-projective resolution of an ^4-module M and Q» is a complex of free yl-modules of finite length. Topics (b)-(e) are then devoted to the grade theorem, which is proved in a topological setting with no reference to Sullivan algebras or graded Lie algebras. This material can be read with only Part I, Part III and §27 as background. Topics (f) and (g) deal with the Sullivan algebra result. (a) Complexes of finite length. Let A be a graded algebra (over A). As in §34 we denote by F, = {F;} a chain complex of yl-modules Pi = Pi,* of the form 0 *-Po,. ?¦ Pi,. *
476 35 Grade and depth for fibres and loop spaces in which (Fj,.) ¦ = Pij-i- If Q* is asecond such chain complex then Hoiru(P., Q* will denote the bigraded complex of ic-modules given by ????(?.,?.){,. =\{RomA{Pj,Qj+i) , i with df = dof- (-l)dezff ? d. Key to the proof of the grade theorem is Lemma 35.1 Suppose P. -—> M is an A~projective resolution of an A-module M and suppose Q* = {Qi}o<i<m iS a complex of free A-modules. Then Hi^ (Hom^(P,, Q.)) = 0 , i> m- proj grade^ M . proof: Set Q't = {Qi}o<i<m-i· Because the Pi are /1-projective the sequence 0—>Hom.4(P.,Qt) —>Homy4(P.,g.) —> HomA(F., Qm) —> 0 is exact. Since Qm is ?-free, i/;,. (Hom^(P., Qm)) = Ext™'*(M,Qm) = 0, i > m — projgrade^ M. By induction on m, //j,, (Hom^iP,, Q',)) = 0, i > m - proj grades M. The lemma follows. D Suppose next that an ?-free module ? has an A-basis xa with only finitely many elements in each (ordinary) degree and that A and ? are concentrated in degrees > 0. It is then immediate that EomA(-,N) = Y[xa.EomA(-,A) , a whence Thus we have Lemma 35.2 Suppose in the situation of Lemma 35.1 that A and Qi are con- centrated in nonnegative degrees, and that each Qi has an ?-basis with finitely many elements in each degree. Then Hi,. (Honu(P., Q*)) = 0 , i>m- grade^ M . (b) fir-spaces and C.(fir)-modules. Let (Y, y0) be a based path connected topological space. Multiplication in the loop space ?.? makes it into a topological monoid (§2(b)). A topological space ?' equipped with a right ??-action will be called an QY-space and a map
The Rational Dichotomy 477 of ??-spaces is a continuous map that preserves the action. If X and ? are QF-spaces then UY acts diagonally on ? ? ? via (x, z) · 7 = (x · 7, ? · 7). Next recall (§8(a)) that multiplication in ?? makes ?*(??) into a chain algebra via the Eilenberg-Zilber equivalence, c.(nr) ® c.(nr) ^> c.(nr ? ??) c'(mult)> c.(ny). In the same way, if Ar is any fiF-space then the action defines a C,(fii')-module structure in C»(X). For example, the constant map ?? —y pt defines a chain algebra morphism C*(QY) —> k, and makes k into a C,(fiF)-module, because C,(pt) = k — cf. D.2). Now consider the Alexander-Whitney comultiplication ? : c.(nr) —» c.(nr) ® c.(nr) introduced in §4(b). It follows directly from the compatibility D.9) of the Eilenberg-Zilber and Alexander-Whitney equivalences that ? is a morphism of chain algebras, and so it makes ?*(??) into a differential graded Hopf algebra (as also described in §26(b)). In particular, if (M,d) and (N,d) are any two C„(fir)-modules then (M,d) ® (N,d) is a C»(QY) ® Ct(UY)-modu\e in the obvious way, and this action composed with ? makes (M, d) ® (N, d) into a C.(CIY)-module. We call this the diagonal action of C*(QY). Now suppose X and ? are fiF-spaces, and let ?,? act diagonally on ? ? ?. A simple calculation with formula D.9) gives Lemma 35.3 The Alexander-Whitney equivalence AW :C.(IxZL C»(X) ® C.(Z) is ? quasi-morphism of ?»{??)-modules, with respect to the topological and al- gebraic diagonal actions. ? Next, we pass to homology. For any complexes (M,d) and (N,d) a natural map H(M, d) ® H(N, d) —>H ((M, d) ® (iV, d)) is defined by sending [z] <S> [w] 1—y [? <S> w] for any cycles ? and w in M and iV. Moreover, if M and H(M) are ic-free (or if iV and H(N) are A-free) then this is an isomorphism. Indeed, because d(M) C M and k is a principal ideal domain, Im d is jfc-free. Split the short exact sequences 0 —» ker d —» M —» d(Af) —> 0 and 0 —» Im d —» ker d —> ff(M) —> 0 to write (M,d) s (f/'(M),O)e0(Amu,JK;dmu). Since H(M)'\s ic-free we obtain H((M, d) ® (iV, d)) = H{H{M)® (N, d)) = ff(M) ® ff(JV). In particular the map if(mult)>
478 35 Grade and depth for fibres and loop spaces makes ?*(??) into a graded algebra and, if X is an ??-space, then the map > ? {Ct{ makes Ht(X) into a right f/,(fiF)-module. Now suppose #.(??) is Jc-free. Then we may use ?*(??) ® ?*(??) = ?(?*(??) ® ?7.(??)) to identify #.(?) as a morphism ff.(A) : ?.(??) —> of graded algebras; in other words, ?*(??) is a graded Hopf algebra. Thus if X and ? are ??-spaces then H*(CIY) acts diagonally on H*(X) ® H*{Z) and, moreover, the morphism H*{X) ®HM(Z) —> ? (C*(X) ® C*(Z)) is a mor- phism of f/,(fiF)-modules. In particular if Ht(X) is ic-free then this provides isomorphisms Ht{X)®H*{Z) ^ H{Ct{X)®C»{Z)) <—-—i/»(X ? ?) H(AW) of ff.(nr)-modules. (c) The Milnor resolution of .fc. Let (F, yo) be a well based path connected topological space. For topological groups G, Milnor [125] constructed the universal bundle EG —> BG by putting EG = G*°°, the infinite join of G with itself (§2(e)). Here we consider the topological monoid ?,? and, when ?*(??) is It free, we use the filtration of (??)*°° by the subspaces (??)*? to construct an Eilenberg-Moore resolution of the C*(fiF)~module, Ik (§20(d)), which we call the Milnor resolution. Recall from §l(f) that the join X * ? of topological spaces is the space (? ? CZ) U(CX x Z). If X and ? are fiy spaces then ?? acts diagonally on ?*? in the obvious way and the inclusion X —> ? ? {0} in X * ? is a map of ?? spaces. In particular, starting with the action of ?? on itself by right multiplication we obtain a diagonal action in each of the joins (??)*? = (??)<?~^ * ??. The inclusions ?? —> (??)*2 —> »· (??)*? —> are maps of fiy-spaces and, as in §2(e), we set (??)*°° = 1](??)*? with the ? weak topology determined by the (??)*?. Lemma 35.4 Let A C X be an inclusion of ??-spaces and give (?, ?) ? ?? the diagonal action, where ?? acts by right multiplication on ??. If H*(X,A) is It-free then ?. ((?,?) ? ??) is Hm(QY)-free. proof: If 7 6 ?? is a loop of length ? let 7' be the loop of length ? given by 7'(f) = ?(? — t), 0 < t < ?. Then 7 1—> 77' and 7 1—> ?'? are homotopically constant maps. Thus the map f-.????^????, /(x,7) = (x-7,7)
The Rational Dichotomy 479 is a homotopy equivalence with homotopy inverse (x, 7) 1—> (? «7', 7). Denote by (X ® ??)? and (? ? ??)? the ??-spaces in which ?? acts respectively diagonally and by right multiplication on ??. The isomorphism (cf. §35(b)) H.(X,A) <8> ?.(??) s H. ((X, A) ? ??)? identifies this latter as the free H*(QY)-module with basis a J;-basis of H*(X,A). On the other hand, / : (? ? QY)r —> (? ? ??)& is a map of ??-spaces and Ht(f) is therefore an isomorphism of i/«(fiF)-modules. G Lemma 35.5 (i) (??)*°° has the weak homotopy type of a point, (ii) ?/?.(??) is k-free so is each ?* ((??)*?). proof: (i) This is an easy exercise since ?? is well-based (Step 1 of Propo- sition 27.9) and so (??)*<?+1) ~ ?(??*? ? ??) of §l(f), and (V (ii) When X and ? have ic-free homology the isomorphism Hm(X) <8> H.(Z) A H{C*{X) ®-C.(Z)) -=> H.(X x Z) identifies H.(X ? ?) as Jfc-free. Now use the formula (?>')*? ~ ?^^ Henceforth we assume ?*(??) is k-free. We are ready to construct the Milnor resolution. Define JHxee graded modules V(n), ? > 0 by setting V@) = Vb(O) = k and V{n) = H* (C(QY)*n , (??)*?) , ? > 1 . Set V = 0V(n). The Milnor resolution will have the form (V <g> C*(QY),d) ? with d : V(n) —> V(< n) ® C*(UY) and C*(QY) acting by multiplication from the right. We construct it by simultaneously constructing the differential d and a commutative diagram ··· V(<n)®C*(QY) ··· in which the vertical arrows are quasi-isomorphisms of C«(fiF)-modules. ' Indeed, set ?@) = id and then suppose by induction that ?(? — 1) and the differential in V(< n) ®C*(UY) are constructed- Use excision and Lemma 35.3 to identify
480 35 Grade and depth for fibres and loop spaces ?, ((??')·(?+1),(????) = H {{C{C\YYn,{UYYn) ? UY) as the free H.CiY)- module V(n)®H*(UY). Next, let va be a basis of V(n) and choose elements za e C, ((??)*(?+1)) such that • za maps to a cycle in C* ((??)*(?+1),(??)*?) representing vQ, and • dza = ?(? — l)uQ for some cycle ua e V(< n) ® ?7« Extend the differential to V(< n) ® C,(?F) by setting dvQ = uQ and extend ?(? — 1) to ?(?) by setting ?(?)?? = zQ. Then <^(n) induces a quotient chain map ?(?) : V(n)®Ct(UY) —> C, ((??)^^1),(??)*?), and i/((^(n)) is the identity, by construction. By the five lemma 3.1, ?(?) is a quasi-isomorphism. Thus the morphisms ?{?) define a quasi-isomorphism ? : (V ® C«(?l/'),d) -^> C, ((??)0), while the constant map (??)*00 —)¦ pi is a map of ??-?????? and a weak homotopy equivalence (Lemma 35.5). Composing yields a quasi- isomorphism of C ^F)-modules (F®C,(fir),(iLl, C5.6) which we call the Milnor resolution of k. Filter this semifree resolution by the submodules (V(< n) ® Cm(UY),d) to obtain a spectral sequence whose ??-term has the form 0 ^-Jc ^-H*(UY) < ^-V(n)®H*(UY) ^- . C5.7) Proposition 35.8 The sequence C5.7) is an Ht(QY)-free resolution of Jk; i.e., the Milnor resolution is an Eilenberg-Moore resolution (%20(d)). proof: We need only show C5.7) is exact. Filter C, ((??)*°°) by the submod- ules C, ((??)*?). Then the quasi-isomorphism ? : V®C*(QY) —> C, ((?>')*°°) preserves filtrations. In the construction of ? we showed that each ?(?(?)) was an isomorphism. This means precisely that ? induces an isomorphism between the i?1-terms of the spectral sequences. It is thus sufficient to prove that k <— Hr(QY) A ¦¦¦?-! is exact. Suppose by induction that 0 <_ jfe <_ ?.(??) A · · · ?- ?* is exact. A simple spectral sequence argument then shows that any d\ -cycle, a, in
The Rational Dichotomy 481 ?. ((??)*?,(??)*(?-1)) is the image of some class ? € Ht ((??)*?). But (?.?)*? is contractible in (??)*(?+1). Thus a representing cycle, z, for ? has the form ? = dw for some w € C« ((??)*(?+1)). In particular, u> represents a class 7 € #, ((??)*(?+1\ (??)*?) and di7 = <*· This closes the induction and completes the proof. D (d) The grade theorem for a homotopy fibre. Fix a continuous map / : X -> Y from a normal topological space X to a path connected, well based topological space (Y,y0). Convert / into the fibration ? : ? ?? MY —> Y, whose fibre F = X xY PY is the homotopy fibre of / (§2(c)). The action of ?.? on F then makes H.(F) into an Hm(QY)-mod\ue (§35(bj). Theorem 35.9 (Grade theorem) With the hypotheses and notation above sup- pose H»(F) andHm(UY) are k-free. Then (i) projgradeH.(nK) H*(F) < cat/. (ii) If equality holds in (i) then *(F) = cat/ = proof: Since proj grade < proj dim by definition, both assertions are vacuous unless cat / is finite. Set cat / = m. The proof of the theorem is in five steps. Step 1: We construct a map of UY-spaces, where UY acts diagonally on (QY)*^m+^ as described in §35(b)). For this recall the construction in §27(c) of the Ganea fibrations pm : PmY —> Y, with fibre Fm at y0. According to Proposition 27.8 (with the roles of X and Y reversed) there is a continuous map ?0 : X —> PmY such that pmao = f ¦ Moreover, by construction PmY is obtained by convertingPm-iYuCFm-i —> Y into a fibration and so there is a homotopy at from ?? to a map ?\ : X —> Pm-\Y U CFm_i. Define a continuous map 9'.XxYMY—>PmY = (Pm-iY U CFm_x ) ? ? MY as follows: if (x,w) € ? ?? MY and w is a path of length ? then 9(x,w) = (???,??1), where w' is the path of length i + 1 given by w'(t) = pmai_t(x), 0 < t < 1 and w'(t) = w(t — 1), t > 1. Observe that pm9 = p, and so ? restricts to a continuous map to a continuous map
482 35 Grade and depth for fibres and loop spaces By construction, 9f is an QY-map. On the other hand, Proposition 27.6(ii) provides a map Fm —> Fm-.x * ?.? which is a map of QF-spaces and a homotopy equivalence. Iteration of this con- struction produces an ?,?-map Fm —> (fi7)"(m+1>, because Fo = QY. Compo- sition with ? F produces the desired hF : F ^l Step 2: We construct an algebraic model for the continuous map h = (id, hF):F^F ^) Choose an Eilenberg-Moore resolution (Proposition 20.11), a : (R,d) —t C.(F), for the C,(fir)-module C*(F) described in §35(b). Then recall the Mil- nor resolution ? : (V ? C*(UY),d) A C. ((??)*°°) of §35(c) and denote the obvious inclusions by Am : (V(< m) ® C*(UY),d) —+ (V ® C.(UY),d). The algebraic model of h we wish to construct will be a morphism of C«(fiF)-modules, where C*(UY) acts diagonally on the target, and it will satisfy: (Am <g> ??)? : (R,d) —> C,(F) ® (V ® C,(fiF),d) zs ? quasi-isomorphism. In fact, since (A, d) is semifree we may lift morphisms from (R, d) (up to homotopy) through quasi-isomorphisms (Proposition 6.4(ii)). Apply this remark to the diagram to construct ?. (Note that AW is a morphism of C«(fiy)-modules by Lemma 35.3. To show (id®\m)%lj is a quasi-isomorphism, let jm : (CiY)*(m+1'> —> (??)*00 be the inclusion. Then (id x jmOi = (id,jmhF) :F^Fx (??)*°° is a weak homotopy equivalence, because (??)*°° —> pt is (Lemma 35.5(i)). But (idxXm)ip is connected (up to homotopy) to the quasi-isomorphism C« ((id ? jm)h) by quasi-isomorphisms, and hence is one itself.
The Rational Dichotomy 483 Step 3: The associated H*(IIY)-free resolutions of Ht(F). In the Eilenberg-Moore resolution (R,d) -=> C*(F) write R = W <8>C*(QY) oo where: W is the direct sum W = 0 W(p) of ic-free modules, d : W(p) —> p=0 W(< p)®C.(nr) and the filtration {W(< ?) ® ?.(??)} is the Eilenberg-Moore filtration of R. This defines a homology spectral sequence (Ei,di) whose E1- term by definition gives an ?(??)~(??? resolution of H*(F) of the form H.(F) <— W@) ® ?.(??) ?- W(l) ? ?* We denote this resolution by P. -?> Jf.(F) with (as in §35(a)) Pp = Next, filter C(F) ® (V (g)C,(fir),d) by the submodules C,(F) ® T^(< ?) ® C,(??). This produces a second homology spectral sequence whose E1-term gives a complex of u«(fiF)-modules of the form, Ht{F) <— Hm(F)®V@)®H.(nY) ?- H.(F) ®VA) ® H.(QY) ? , since H.(F) and ?.(??) are it-free (cf. §35(b)). Now ?.(??) acts diagonally in (H.(F) <S> V(p))®H*(QY) and this is a free #»(i7y)-module on a it-basis of H*(F) ® V(p) (§35(b) and Lemma 35.4). Moreover, since the Milnor resolution is an Eilenberg-Moore resolution (Proposition 35.8) it follows that this sequence is exact; i.e. it is also an H*(QY)-iree resolution of H*(F). We denote it by with Qp = H.(F) ® V(p) ® ?.(??). Step 4-' Set projgrade^^yj H*(F) — r. Then ? is homotopic (as a morphism of C*(QY)-modules) to a morphism ? such that ?: W(p) ^> Ct(F) ®V(< ? + m - r) ® C*(UY), p>0 . Filter C„(F)(g>V(< m)®C*(UY) by the submodules C. (f)® V{< p) ®C. (??) to produce a spectral sequence whose E1-term is just (Q<m,di), and denote all spectral sequences by (E\dl). k Now write ? as the infinite sum ? = J2 ?? of the C»(fiy)-linear maps ?? —oo defined by
484 35 Grade and depth for fibres and loop spaces (Necessarily k < m.) Then ????? = i>kdo and H(ipk) is an Ht(QY)-lineax map that raises the resolution degree by exactly k. We may now apply Lemma 35.1. It asserts that if k > m-projgradeH-(Qr) H*(F) (= m-r) then H(ipk) = ???+??} for some H*(ClY)-linear map ? : ?« —> Qt that raises resolution degree by k + 1. Choose C„(ttr)-linear maps ?' : W(p) ® C*(QY) —> C,(F) ® V(p + k + 1) ® C*(QY), ? > 0, so that 6>'(it; ® 1) is a cycle representing 9(w ® 1). Then do9' + 9'd0 = 0 and hence ?' = ? - (??' + 9'd) also raises filtration degree by at most k : ?' = ??'??· i<k By construction, H(tp'k) = 0 : P« —> Q<m- Choose C«(fiF)-linear maps ?" :W(p)®C.(ttY) ^ C4F)®V(p + k)®C*(UY), ? > 0, so that do9"(w®l) = ip'k(w®l), w €W. Set ?" = ?'-(??" + ?"?). By construction, ???" + ?"?? = ^ fc-l and so ?" = ? ?", while clearly -?" ~ -?. Remark If the ic-modules Hi(F) and /?^(??) have finite bases for each i and j then this will also be true of each [H*(F) ®V(< m)]k, as follows from the definition of V in §35(c). In this case we may use Lemma 35.2 in the proof of Step 4 to find ? ~ ? with ? : W(p) —»· C,(F) ® V(p + m - ?) ® ?(??) , ? > 0 , where ? = gradeH_(QK) H.(F). Step 5: Completion of the proof of Theorem 35.9. Suppose projgrade^^Qyj H.(F) > cat/ and, as above, write m = cat/ and r = pro] gradefj^QY^ H.(F). Since r >m Step 4 provides a filtration preserving morphism of C.(Oy)-modules, ? : (W ® C*(UY),d) ^ C.(F) ® (V(< m) Thus ? induces a morphism ??(?) of spectral sequences which, at the i?1-level has the form (in the notation of Step 3). Compose ? with the inclusion id®Xm of Step 2 to obtain a filtration preserving morphism ?' : (W ® C.(nr),d) —»· (C.(F) <8> (F ® Ct(QY),d)) , which then yields a morphism ?'(?') of spectral sequences. The E1 -terms of these spectral sequences are respectively the resolutions (?^,?1) and (Q«,^1) of Step 3; in particular it follows from Step 3 that in both cases E2 = -Eg». In
The Rational Dichotomy 485 each case ??, = H,(F) and so the inclusions W@) <8>G,(??) —> W®C,{QY) and C,{F) e V@) <8> ?7,(??) —»· C.(F) ® F ® G.(?1') induce isomorphisms ?"o,* —¦> H(—) which identify ?2(?') — ?^,(?') = ? (?1). But 77' = (id ®AmO7 ~ (id® Am)t/>, which is a quasi-isomorphism by Step 2. Hence ?2(?') is an isomorphism and is a quasi-isomorphism of resolutions. Suppose first that projgrade^^y) H*(F) > cat/. Then Step 4 would allow us to choose ? so that ? = ? Vk- It would follow that ??{?) = 0 and so fc<0 Ex{q') — E1(id<8>\m) ? ?1 (?) = 0 as well. Since ??(?') is a quasi-isomorphism this is absurd; i.e., H.(F) < cat/ . Now suppose projgradeH(QK) H*(F) = cat/, and let JV be any right H,(QY)- module. Since E1^') is a quasi-isomorphism of resolutions, it induces a quasi- isomorphism Hominy)(P., ??) <=- Homw.(ny)(C,,iV) . It follows that ?1 (?) induces surjections H"'" (EomHAnY)(Pt,N)) «- H™ {UomHm(nY)(Q<m,N)) , where (as in §35(a)) the left degree "p" is the resolution degree. Since Q<m is concentrated in resolution degree < m and since ? is concen- trated in resolution degree zero it follows that Hp'* (Hominy)(?,,./?)) = 0, p> m. But HP'* (EomHAnY)(Pt>N)) = Extp {H.{F),N), since P. is an H.{QY)-free reso- lution of H* (F) (cf. §20(b)). Thus ???^"(??) (Ht(F),-) = 0. Since, by assump- tion, projgradeH>(Qr)u,(F) = m it follows that ???^(??) (H*(F),-) ? 0. Thus projdim/i^ny) (Hm(F)) = m, as desired. This completes the proof. ? Corollary: Assume, in addition to the hypotheses of Theorem 35.9, that the Jz-modules H{(F) and Hj(QY) have finite bases for each i and j. Then gradeH.(QK) H.(F) < cat / and if equality holds then also cat/ = projdimjy^ny) H*{F). proof: The Remark at the end of Step 4 in the proof of the theorem allows us to replace proj grade by grade in Step 5. ?
486 35 Grade and depth for fibres and loop spaces (e) The depth of ?.(??). Theorem 35.10 If (X,x0) is a normal path connected topological space and if each Hi(QX) is Ik-free on a finite basis then depth?.(??) < catX . // equality holds then also cat.Y = g\dimHt(QX). proof: Replace X by a well based space of the same homotopy type by adjoin- ing an interval to X at the base point x0. Then apply the Corollary above to Theorem 35.9 to / = idx. D Corollary If h is a field of characteristic zero and if X is simply connected then depth ULx < cato X, where Lx is the homotopy Lie algebra. proof: Replace X by the rationalization (§9) Xq and note (Theorem 21.5 and Proposition 28.1) that ULX = H*(ilX;Jc) = ?*(??®;1?) and that cat0 X = ? (f) The depth of UL. In this topic Jk is an arbitrary field of characteristic zero. Suppose L — {Li)i>\ is a graded Lie algebra, with universal enveloping al- gebra UL. In §22(a) and §22(b) we constructed C*(L;UL) and established its properties, where L is regarded as a differential graded Lie algebra with zero differential. This gives, in particular, a UL-iree resolution of Ik of the form k*—UL<±-sL®UL4 4- AnsL ®UL < This resolution, which plays the key role in this and the next topic, will be denoted by P. ^> Ik . By definition (§20(b)), Ext^L(A, -) = if"·· (Hom^F,, -)). Proposition 35.11 Let L = {?;};>1 be a graded Lie algebra with each Li finite dimensional. (i) gldim UL is the largest integer ? (as ooj such that Exi^L(Jc, k) ? 0. (ii) depth UL < gl dim UL. proof: (i) Clearly ? < gl dim UL, since ExtyL(A, Ik) ? 0. On the other hand, write F, = C*(L) ? UL. Then ExtUL(Jc,k) = H(HomUL(P*,]k)) = ? {C*(LN) = ? {C*{L)Y ,
The Rational Dichotomy 487 because k is a field. It follows that H(C*(L)) is concentrated in homological degrees < n. Now choose a graded subspace ? C AnsL so that E®d (An+1sL) = AnsL. Observe that (A<nsL 0 E) <8> UL is automatically a subcomplex of P«, and that the inclusion defines a morphism of the spectral sequences derived from the fil- trations Fp(—) = (—) •(UL)>p. At the ?7°-term the differentials are just d <8> id in C*(L) <g>UL and so this morphism is a quasi-isomorphism of ?7°-terms. It fol- lows that the inclusion of (A<nsL © E) ®UL in P» is itself a quasi-isomorphism. This (A<nsL © E) <g> UL is a t/L-free resolution of 1; and so Ext?2(&> -) = 0; i.e., gldim UL < n. (ii) Let gldim UL = n. Then Ext^1 (¦&> ~) = 0. Apply B.omUL(P*, -) to the short exact sequence 0 —> (UL)+ —> UL —> 3: —> 0 and pass to homology to obtain an exact sequence Ext&i (A, t/L) -»¦ Ext&L(A, A) -»¦ Ext^1 (*, (t/^)+) · Thus ???^?,(?, I7L) surjects onto Ext^?,(A, A), which is non-zero by (i), and so depth UL < n. ? (g) The depth theorem for Sullivan algebras. As in the previous topic, in this topic Ik is an arbitrary field of characteristic zero. Fix a minimal Sullivan algebra (AV,d) such that V = {^*}i>2 is a graded vector space of finite type, and let L be the homotopy Lie algebra of (AV, d) as defined in §21 (e). Recall that cat(AV, d) is defined in the introduction to §29. As in the previous topic the complex P» = C*(L; UL) and associated constructions will play a key role. In particular, let d\ denote the quadratic part of the differential in AV : d\ : V —> A2V and d-di : V —> A^3V. From the definition of the Lie bracket in §21(e). and Proposition 23.2, and Example 1 in §23(a) we obtain ) = C*(L) = EomUL(P*,]k) , with A?V = HomUL(Pp,]k). Thus ExtpUL(k,]k) = Hp'm(hV,d{) , with the left degree "p" corresponding to wordlength in AV. Proposition 35.12 cat(AV, d) < gl dim UL. proof: Let ? = gldim?/L; according to Proposition 35.11 it is the largest integer such that Ext?L(A,A) ? 0. Define an ideal / C AV by setting / = A>nV ? In. where ? @ (kerdx)"·* = AnV. Then H(I,di) = 0. Filter / by wordlength and use the associated spectral sequence to deduce that H(I,d) — 0. Conclude from Corollary 2 to Proposition 29.2 that cat(AV, d) < nil(AV//) < ? because (Av, d) —>. (Av//, d) is a quasi-isomorphism. D
488 35 Grade and depth for fibres and loop spaces Theorem 35.13 [54] // L is the homotopy Lie algebra of a minimal Sullivan algebra (AV, d), and ifV= {V'}i>o is a graded vector space of finite type, then depth UL < cat(AV, d) < gl dim UL . Moreover if depth UL = cat(A V, d) then cat(AV, d) = gl dim UL. proof: In view of Propositions 35.11 and 35.12 it is sufficient to prove that depth UL > cat(AV,d) =*> cat(AV,d) > gldimt/L , and for this we may suppose cat(AV, d) is a finite integer m. Recall first that in §29(f) the surjection (AV,d) —> (AV/A>mV,d) is extended to a (AV", d)-semifree resolution of the form ? : (AV ® (Af © A), d) -^> (AV/A>m, d) , and with the following three properties: • The differential di in AV extends to a differential ?\ in AF ® (? ? Je) characterized by - ?? : M —> (V (g) ?) ? Am+1 V, and d - ^ : M —> (A^2V ® M) © >+2 - AV (g) (M ? Je) is a (AV, d^-module. • The surjection (AV,di) —> (AV/A>mV,d1) extends to a (AV,dx)-semifree resolution ? : (AV ? (? ? Jfe),Ai) ^> (A\7A>mV, di) such that ?(?) = 0. • Because cat(AV, d) = m there is a morphism (retraction) ? : (AV ® (M © A), d) —»· (AV, d) of (AV, d)-modules such that ?A) = 1. Now bigrade AV and the tensor product by setting (AV)P>* = APV and (AV ® (M © Jk))p'* = (Ap-mV®M) ? ApV. Then filter by setting Fp(-) = (—)-Pi*. Thus d\ (or ?\) increases filtration degree by 1 while d — di (or d — ?\) increases it by at least 2. On the other hand, ? need not preserve the filtration degree. However ? can decrease the filtration degree by at most m and so it can oo be written as the infinite sum ? = ? ??? some k < m, of the AV-linear maps i=-k ?? characterized by: • ?? increases filtration degree by exactly i.
The Rational Dichotomy ,o„ It is immediate that ???-k ~ ?-kai = 0. Simplify notation by writing ? = ? ? ic. Then ^_fc is a cycle of bidegree (-k,0) in the bigraded com- plex ?.???\?(?? ® ?, AV), (defined with respect to the differentials di and ??). We show below that Hp>*(EomAV(AV®N,AV)) =0 , p<0. C5.14) This implies that if k > 0 then ?-k = ??? + ??? for some AF-linear map ?: AV ® iV —> AV that decreases filtration degree by exactly k + 1. Thus ^ — (do + od) is a retraction which decreases filtration degree by at most k — 1. Proceeding in this way we find a retraction ? that preserves filtrations. But then ? = ? ?{ and ?0 : (AV ® ?,??) —> (AV.di) satisfies 770A) = 1. This {=0 implies that ?(?0) is surjective. Since HP>*(AV ® ?,??) -=-» #p'* (Ay/A>mV')d1) = 0 , ? > m , it follows that Ext??(Jfe,Jfe) = i/>m'*(AF)d1) = 0; i.e., cat(AV,d) = m > gl dim ?/L. ? remains to prove C5.14)- We shall work entirely in the category of bigraded complexes; in particular, Horn will always mean bigraded Horn and j{ will mean the bigraded dual: (<?!,„)?'* = C^q. Now regard each ApsL ® UL as a left L-module by setting x-a= -(-l)desidegaa.x , x€ L, a€ ApsL®UL . Denote the differential in each C*(L;ApsL® UL) by di and denote the maps C. (L;ApsL ® t/L) -^4 C, (L; Ap~lsL ® by 92. Then d = di +d2 is a differential of bidegree (-1,0) in the bigraded module C„(Z,)<g>C„(L)(g>t/L, where the bigrading is given by (-)p>, = 0 Ci®Cj®UL. As in §22(b) this complex-is in fact a differential graded coalgebra. Moreover, if ? is the comultiplication in C*(L) then an easy calculation shows that the composite C*(L) A C*(L) <g> C*(L) c—^ C,(L) <g> C,(L) ® C/L commutes with the differentials and is in fact a dgc quasi-isomorphism. Dualizing we find that multiplication in (AV, dx ) extends to a cochain algebra quasi-isomorphism of the form (AV®AV®(UL)Kdi) ^> (AV,di) , where (AV)P-* = APV and {AV ® AV ® (t/L)tt)Pl* = ? AlV ® tiV ® (C/L)8. This is a quasi-isomorphism of (AV, di )-semifree modules and so we may apply — to obtain a quasi-isomorphism (AV®AVlA>mV®(UL)\di) A (AV/A>mV,di) .
490 35 Grade and depth for fibres and loop spaces Lift this to a quasi-isomorphism into (AV <g> ?,??) and thereby reduce C5.14) to the assertion HP'' (HoniAV (AK ® AV/A>mV ® (?/L)«, AV)) = 0 , ? < 0 . C5.15) Next observe that AV/A>mV ® (I7L)» = (A^msL® ?7L)*. Thus there are canonical identifications of bigraded vector spaces Honuv {AV ® AV/i\>mV ® (Z7L)', ??) = Horn (AV7A>mF ® (?/L)s, AV) = Horn ((A^msL ®UL)\ (AsL)") = Horn (AsL, (A^msL ® E/L)M) = Horn (AsL, A$msL ® i/L) = Hornet (AsL ® ?/L, A-msL ® UL) , where we may identity (A-msL ® I7L) with A-msL ® I7L because this graded vector space has finite type. An arduous but straightforward calculation shows that this defines an isomorphism of complexes HomAv (AV ® AV/A>mV ® {UL)*',AV) = KomUL(P*,P<m) . Since AsL is a graded vector space of finite type Lemma 35.2 (applied with A = UL and M = Jfc) gives Hp'* (EomUL(Pt, P<m)) = 0 , ? < depth UL-m . Since depth ?/L is supposed > m this proves C5.15) and the theorem. ? Exercises 1. Let X be a simply connected CW complex of finite type. Prove that depth ?. (?E3 V X); Q) = 1. 2. Let X be a simply connected finite CW complex. Prove that depth H*(QX; Ik) for any field Ik. 3. Let X be a simply connected formal space with H*(X\Q) ~ Q[a, ?>]/ (?4,?4,?2?2) and deg ? = 2, deg ? = 2. Prove that cat0^ = 4. Consider the map /:54VS4-+ ^(Q2,4) (resp. g : ^(Q2,2) -* ^(dj2,4)) corresponding to the fundamental classes of the spheres (resp. to the squares of generators of //¦•(/^(Q2,2))). Convert g into a fibration and show that X has the rational homotopy type of the pullback of g along /. Xo ? E4v54)o 4
The Rational Dichotomy 491 Compute the rational homotopy Lie algebra of X using the homotopy fibre of h. Prove that the dimension of the centre of Lx is 2, and that the depth of Ht(QX;Q) is 3.
36 Lie algebras of finite depth In this section the ground ring it is a field of characteristic ? 2,3. Throughout the section we adopt the convention: • L is a graded Lie algebra such that L — {Lj}j>i and each Li is finite dimensional. ^ ' ' Recall from the start of §35 that the depth of the universal enveloping algebra, UL, is the least integer m (or oo) such that Ex.t™L(k,UL) ? 0. Here we shall slightly abuse language with the Definition The depth of L is the depth of its universal enveloping algebra, UL. In §35 (Corollary to Theorem 35.10) we showed that if A" is a simply connected topological space and if each H((QX) is finite dimensional then depth Lx < cat0 X , where Lx is the homotopy Lie algebra of X. Thus spaces of finite rational category have homotopy Lie algebras of finite depth. Here we study graded Lie algebras L of finite depth m and show that • The sum of all the solvable ideals in L is finite dimensional. • UL is right noetherian if and only if L is finite dimensional. • There are at most m linearly elements Xi in Leven such that ad x\ is locally nilpotent. As just observed these results apply to the homotopy Lie algebra of a space of finite rational category. The reader should also note that the third result is a considerable strengthening of Theorem 31.16. This section is organized into the following topics: (a) Depth and grade. (b) Solvable Lie algebras and the radical. (c) Noetherian enveloping algebras. (d) Locally nilpotent elements. (e) Examples.
The Rational Dichotomy 493 (a) Depth and grade. Proposition 36.2 (i) If L is the direct sum of ideals I and J then depth L = depth / + depth J . (ii) If L is the infinite direct sum of non-zero ideals then depth L = oo. proof: (i) Because of C6.1) we may identify Ext^L (k, UL) with Tor^1(k,UL*)*, ? > 0 (Lemma 34.3(iii)). Thus depth L is the least integer m such that Tor^A, ULS) ? 0. On the other hand, if P« and Q« are respectively VI- and UJ-iree resolu- tions of k then P, ® Q, is a VI ® U J-free resolution of A. Since UL = UI®UJ this gives = H((P.®Q.)®UL(UIi®UJt)) = H (P. ®ui UI*) ® //(Q, ®i/j t/J») Assertion (i) follows. (ii) Here there is no ? G UL such that (UL)+ ¦ a = 0 and so ExtyL (A, UL) = H.omm(k,UL) = 0. Thus depthL > 1. But then for each ? > 1 we may write L as the direct sum of ? ideals L(i), each of which is itself an infinite direct ? sum. Thus by (i), depth L = ? depth L(i) > n. Hence depth L = oo. ? Next, recall from Lemma 34.3(i) that if / C L is an ideal then ?????(%:, Jk) is naturally a right ?/L//-module. Proposition 36.3 Let I C L be an ideal, (i) depth/ < depthL. (ii) If a-UL/1 is finite dimensional for all a G Tor^I(k,k), q < depthL, then depth L/I < depth L. (Hi) If the hypothesis of (ii) holds and the Lie algebras L/I is finitely generated then depth / + depth L/I — depth L. proof: (i) The Hochschild-Serre spectral sequence (§34(b)) converges from ??'9 = E^tpUL/I(k,E-KtqUI{k,UL)) to Extj^A-, UL). Since UL is UI-ivee
494 36 Lie algebras of finite depth it follows that Ext^jfc.t/L) = 0, q < depth/. Hence ExtrUL(Jk,UL) = 0, r < depth I and depth L > depth I. (ii) Let m = depthL. We show first that Ext™L(k,UL/I) ? 0. In fact, filter UL by the ?/L-modules Fp = UL-(UI)>p. Use the duality of Ext2rL(Jb,i/L) and Tor^jfc, (UL)*) to deduce that for 0 ? a G Ext™L(Jfc, UL) there is a maximum ? such that ? is in the image of Ext™L(k,Fp). It follows that a maps to a non-zero element in Ext™L (k, Fp/Fp+1). But Fp/Fp+1 is just i/L// <g> (UI)P as a t/L-module, and so Ext%L(k, UL/I) ? 0. Next observe that the Remark in §34(b) identifies the u^-term in the Hochschild Serre spectral sequence for Ext™L(k,UL/I) as E™ = ExtpUL/I(ToT%I{jk,Jk),UL/l) . Thus, for some ? + q = m, this term is non-zero and so (cf §35) gradec/t// (TorJ^Jfc, Jfc)) < ? < depth L . It is thus sufficient to prove that depth L/I < gradeUL/J(M) for any non-zero L//-module M for which ? · UL/I is finite dimensional, a G M. Write M as the increasing union of sub L//-modules M(i) C M(i+ 1), i > 0, with L/I acting trivially on each M(i)/M(i — 1). As in §20(b) there is a UL/1- free resolution P, of M such that P« is the increasing union of free resolutions P,(z) of M(i), and also P»(z)/P,(z - 1) is a free resolution of M(i)/M(i - 1). Suppose now that Extp/L/j(k, UL/I) = 0. If ? G uomUL/j (Pp, UL/I) is a co- cycle vanishing on Pp(i—1) then the induced cocycle in Horrim/j (Pp(i)/Pp(i — 1) , UL/I) is a coboundary by hypothesis. Thus for some b G HomUL/j (Pp_i, UL/I) vanishing on Pp_i(z —1) we have: a — db vanishes on Pp(i). This argument shows that ? itself is a coboundary and hence that ExtpJL,I(M,UL/I) = 0. Thus gradeULjj(M) > depth L/I, as desired. (iii) Recall that the Hochschild-Serre spectral sequence (§34(b)) con- verges from Ext?rL// (Jc.Ext^Jc, UL)) to Ext^9(&, UL). The duality observa- tions in §34(b) identify the E2-tevm with the dual of Tor^L// (k, Tot (k, (i/L)")) and this coincidences with Tor^L// (k, (E/L//L) ® TorJ" (Jfc, ([//)'), by Propo- sition 34.11, provided that q < depth L. Since some ?1?'9 ? 0 for ? + q — depth L we may conclude that depth L > depth L/I + depth /. Suppose inequality held and put r = depth L/I and s = depth/. Then E?s ? 0 and ??+i's~i+1 = 0 = jE?-Ji'+j-1. This would imply ?oes 7e 0 contradicting depth L>r + s. ?
The Rational Dichotomy 495 (b) Solvable Lie algebras and the radical. A graded Lie algebra L is called solvable if some ?/?+1) = 0, where L(n) is the ideal defined inductively by L<°> = L and Hn+^ = [tin\ L™] , ? > 0 . The solvlength of L is the largest ? such that l/n) ? 0. Recall that we restrict attention to those L satisfying L = {Lj}i>i and with each L{ finite dimensional. Theorem 36.4 The graded Lie algebra L is solvable and of finite depth if and only if L is finite dimensional. In this case EyLt*UL{k,UL) is one dimensional, and depth L — dim Leven . proof: Suppose L is solvable and has finite depth. The ideal [L, L] has finite depth (Proposition 36.3(i)). and so by induction on solvlength, [L, L] is finite dimensional. In particular, for some k, L>k is an abelian ideal, also of finite depth. Write L>* = @JkxQ and note that each kxa is an ideal in L>*. Since a L>k has finite depth it cannot be an infinite direct sum. (Proposition 36.2(ii)) and hence L>* is finite dimensional. Thus so is L. Conversely, suppose L is finite dimensional. Since L — L>i, L is trivially solvable. Moreover, a non-zero element ? in L of maximal degree is central. Set kx = I and put 1 if deg ? is even 0 if deg ? is odd. Since UI is either the exterior or polynomial algebra on kx, a simple direct calculation gives Extpj(k, UI) = 0 and dim Ext^Jk, UI) = 1 . Thus Proposition 34.7 asserts that there are isomorphisms of ?/L//-modules, ULI I iiq = e 0 otherwise. In particular the Hochschild-Serre spectral sequence collapses at Eo, and ExtrUL(k,ULJ?Extru-L'/J(k,ULII) . It follows by induction on dim L that Ext*UL{k, UL) is one dimensional and that depth L — dim LeVen· n Definition The radical of a graded Lie algebra L is the sum of all the solvable ideals of L.
496 36 Lie algebras of finite depth Theorem 36.5 [54] If L satisfying C6.1) has finite depth then its radical, R, is finite dimensional and dimi?even < depth L. proof: Every solvable ideal I C L satisfies dim/even < depth L, by Theorem 36.4. Choose / so that dim 7even is maximized. For any solvable ideal J, I + J is solvable, and hence JeVen C /even· It follows that i?even = /even and so. for some k, R>k is concentrated in odd degrees. Thus R>k is abelian and R itself is solvable. Now Theorem 36.4 asserts that Ext^fi(ic, UR) is one-dimensional and con- centrated in * = dimi?even· It follows that the isomorphism Extyfi(.fc, f/L) = Ext*UR(k, UR)®UL/R identifies the left hand side with UL/R as a L/R-modu\e. Thus the ?2-term of the Hochschild-Serre spectral sequence converging from ExtUL/R (k,ExtgUR(k,UL)) to Extffj?(k,UL) is given by ,q = dimR , otherwise. In particular, the spectral sequence collapses at Eo, and depth L = depth L/R + dimi?even. D (c) Noetherian enveloping algebras. A graded algebra A is right noetherian if any right A-submodule of A is finitely generated. In this case any submodule of a finitely generated right module is finitely generated. Theorem 36.6 Suppose L is a graded Lie algebra satisfying C6.1). If L has finite depth then UL is right noetherian <==> dim L is finite. Remark This Theorem was originally proved for homotopy Lie algebras of spaces of finite category by Bogvard and Halperin [28]. proof: Suppose UL is right noetherian. Then any submodule of a finitely generated right [/L-module is finitely generated. In particular k admits a UL~ free resolution P, such that each Pp is UL-free on a finite basis, and this implies that each Tor^L(i:, k) is finite dimensional. Next, note that any sub Lie algebra ? C L is also right noetherian. Indeed (Corollary to Proposition 21.2) multiplication in UL defines an isomorphism UE ® V ^> UL for some subspace V C UL. If M C UE is any subspace then M · UL = M · UE <S> V. It follows that a minimal set of generators for the UE module M · UE is also a minimal set of generators for M · UL as a ?/L-module. In particular this is finite and so any right ?/i?-submodule of UE is finitely generated; i.e., UE is right noetherian.
The Rational Dichotomy 497 Consider the ideal / = L>n, some ? > 0. Since / is right noetherian each Tor^/(Ik, Ik) is finite dimensional and hence Proposition 36.3(ii) asserts that depth L > depth L/I. Since L/I is finite dimensional Theorem 36.5 asserts that dim(L//)even < depth L. This holds for I = L>n, any n, and so Leven is finite dimensional and, for large n, L>n is concentrated in odd degrees. Then L>n is abelian and L is solvable, hence finite dimensional (Theorem 36.5). Conversely, if dim L is finite write L = ? ? kx where / is an ideal and ? is a non-zero element of minimal degree. By induction on dim L we may suppose UI to be right noetherian. Use multiplication in UL to define an isomorphism UI ® Ax A UL. If dega; is odd then UI <g> Ax is generated as a U/-module by 1 and ? and any sub ?/L-module is finitely generated even as a U /-module. Suppose dega; is even. If M C UL is a right f/L-submodule define subspaces S(k) C UI as follows: ? G S(k) if and only if axk + ? ???% e M f°r some elements Oj G UI. i<k Then S(k) is a right U /-submodule. Since S(k) C S(k + 1) C · · · it follows that for some n, S(n) = \J S(k), and we can find a finite subset of |J S(k) which k k contains a set of generators for each S(k), k < n. As in the classical case of polynomial algebras it is straightforward to use these to construct a finite set of generators for M as a [/L-module. Thus UL is right noetherian. ? (d) Locally nilpotent elements. An element ? G L is locally nilpotent or an Engel element (cf. introduction to §31) if for all y e.L there is an n(y) for which (adx)n^y = 0. Theorem 36.8 A graded Lie algebra L satisfying C6.1) and of depth m con- tains at most m linearly independent Engel elements of even degree. proof: We show that if L = / ? li, with / an ideal and ? a non-zero Engel element of even degree, then depth / < depth L. (The theorem follows from this by an obvious argument.) To establish this assertion note first that since ad a; is locally nilpotent, x acts locally nilpotently in each Tor^7(k,k), as follows directly from Example 2, §34(a). Thus Proposition 36.3(iii) asserts that depth / + depth kx = depth L and the conclusion follows from depth kx = 1 (because ? has even degree). ? (e) Examples. Recall that graded Lie algebras L satisfy L = L>\ and each Lj is finite dimen- sional.
498 36 Lie algebras of finite depth Example 1 Depth = 0. A graded Lie algebra L has depth0 if and only if HomUL(k,UL) ? 0; i.e.., if and only if ? · UL+ = 0 for some non-zero ? G UL. It follows at once from the Poincare Birkoff Witt theorem 21.1 that this occurs if and only if L is finite dimensional and concentrated in odd degrees. D Example 2 Free products have depth 1. Let ? and L be graded Lie algebras and recall the free product EUL defined in §21(c). We shall show that depth ? II L - 1. Indeed, choose free resolutions of the form A V{2) ®UEA V{1) ®UEAuE->R, and ? W{2) ® ?/? A W{1) ®UL^UL—>k. Then as described in §21(c) there is a U(EU L)-free resolution of k of the form [VB) © WB)] ® U(E II L) A [V(l) © W(l)\ ® U(E II L) A U(E II L) -» Jfc in which the restriction of d to V(i) and W(i) is given by the resolutions above. Choose non-zero elements ? G ? and y G L and define a U(E U L)-linear map / : [V(l) ® W(l)] ® U(E U L) —> U(E U L) by f(v) = (dv)-yx and f(w) = (dw) -xy. Trivial calculations show that / ? d = 0 and that / is not a coboundary. ? Example 3 XV Y. Let X and Y be simply connected spaces with rational homotopy of finite type. In Example 2 of §24(f) we observed that LXvy = LxULy. Thus depth Lxvy = 1. On the other hand, cato(X VK) = max(cat0 X, cat0 Y) as follows from the remarks at the start of §27. Thus the difference cato-depth can be arbitrarily large. ? Example 4 Products. If ? and L are graded Lie algebras then depth (? ? L) = depth ? + depth L as observed in Proposition 36.2. Since Lxxy = Lx®Ly we have that depth Lxxy = depth Lx + depth Ly in analogy with cato(^ ? Y) = cat0 X + cat0 Y (Theo- rem 30.2) D Example 5 X = S\ V 563 U[a[a,6]w]w ?>8· This CW complex was first discussed in Example 2, §13(d) and subsequently in Example 4, §24(f) and in Example 3, §33(c). In §33(c) we showed that the homotopy fibre of the retraction X —> S% was rationally S$ V S5, and that the fibre inclusion ? : S% V 55 —> X restricted to 55 represented [a, b]w-
The Rational Dichotomy 499 Recall that Lx is the rational homotopy Lie algebra of X. If ?, ? G (LxH correspond to a,b G ??3(?) then it follows easily that Lx = L(a,/3)/[a, [a,/3]f. As an immediate consequence we see that ? is an Engel element in Lx. Write Lx = ha® /, where I = k? ? {Lx)>r The argument proving The- orem 36.8 shows that depth/ < depthL. Since also depth/ > 0 (Example 1, above) we have depth Lx > 2. On the other hand, X is a 2-cone and so cat A' < 2 (Theorem 27.10). Since depth Lx < cat0X (Theorem 35.10) we conclude depth Lx = cat0 X = cat X = 2 . D Example 6 CP°° /CPn. Because the inclusion i : CPn —> CP°° induces the surjection Ax —> Ax/xn+1 in cohomology (deg a: = 2), the cohomology algebra of CP°° /CPn is just Q © xn+1 -Ax. Moreover, CP°°/CPn is 2n+1 connected and so ?2?+2(€?°°/???) = H2n+2{CPO°l<CPn) = ?. This defines a continuous map / : CP°°/CPn —> K(Z, 2n + 2). Let g : F —» CP°° /CPn be the homotopy fibre. Next, observe that a Sullivan representative for i is also a Sullivan representa- tive for the surjection Ax —> Ax/xn+l, (dega; = 2). Use Lemma 13.3 and 13.4 to deduce that A = k @ xn+1Ax is a commutative model for CP°°/CPn. Extend the inclusion Axn+1 —> A to a minimal Sullivan model ? : (Axn+1 <8s AV,d) A (A,0). This is a Sullivan model for CP°°/CPn and the inclusion of Axn+1 is a Sullivan representative for /. Thus the quotient Sullivan algebra (AV,d) is a minimal Sullivan model for F. Moreover since A is Aa:n+1-free on the basis l,xn+2,... ,x2n+1 it follows that ? induces a quasi-isomorphism 2n+l ? : (? V, d) -=> Jfc ? 0 kxl - i=n+2 Thus F is a formal space in which cup-products vanish and hence F has the 2n+l rational homotopy type of V ^2l (Theorem 24.5). i=n+2 Let L be the rational homotopy Lie algebra of CP°° /CPn. The discussion above establishes the short exact sequence 0 —-> L(qi ,..., q„) —>· L —>· Q? —>· 0 with deg/3 = 2n + 1 and dega, = 2n + 2z + 1. Now Proposition 36.3(iii) asserts that depth L = depth Q? + depth L(qi ,..., ??) = 1 . D Example 7 eo(A:) = 2; cato(A') = depth Lx = 3. Consider the commutative rational cochain algebra (^4, d) = (?(?, b, {xn}n>i)/ I, d) with deg? = deg b = 2, degxn = 3n — 1, / the ideal generated by a2, b2 and A-3({xn}n>i), and dx\ = 0 and dxn+\ = abxn. This is a commutative model for
500 36 Lie algebras of finite depth some rational topological space X (§17). The inclusion A(a,b)/ar,b'2 —> (A,d) defines a fibration Xq —> Sq ? Sq whose homotopy fibre is the infinite wedge F = V S«" (same argument as in Example 6). n>l Consider the inclusion Sq V Sq —> Sq ? Sq and pull this fibration back to a fibration Y —> Sq V Sq. A commutative model for Y is just ?(?. 6)/?-2?, b ®A(a,b)/a2,b2 A (same argument again), and this is just the tensor product H(SqV Sq) ® H(F). It follows that Y ~q (S? ? Sq) ? F. Since S^vS^—iS^x SQ is surjective in homotopy this implies that Lx is the direct sum of the ideals Lp and ?s?xs?· Now Proposition 36.2(i) asserts that depth Lx = depth LF + depth L5? + depth L5? = 3 . On the other hand, X has a commutative model with product length 3 and so Theorem 29.1 states that cat AT < clA < 3. Since depth X < cat AT we have depth X = cat A = c\X = 3. Finally, if (AV, d) is a minimal Sullivan model for X then any cocycle of wordlength > 3 in AV will map to such a cocycle in A, and these are all coboundaries. It follows that e(A") < 2. If e(X) = 1 then cat A" = 1 (Theorem 28.5(ii)). Thus e(X) = 2. D Example 8 L = Der>o Lv, where V is a finite dimensional vector space of dimension at least 3. Write V = ]kv® W where ? G Veven unless Feven = 0. Then there are infinitely many linearly independent elements ? in L(W)even and each one determines ?? G DerLy by ??(?) = ? and 6X(W) = 0. It is straightforward to verify that each ?? is an Engel element in L, and so depth L = oc. ? Exercises 1. Let A" be a 1-connected finite CW complex and let Y be an r-connected hyperbolic space such that dim X + 1 < r. Prove that the space of continuous maps Map(A", Y), endowed with the compact-open topology, is a 1-connected space. Prove that the evaluation map, M&p(X,Y) —> X, f ?—> f(xo), is a fibration which admits a section. We denote by Mapt(X,Y) the fibre of this fibration. Prove that depth Mapt(X, V) = depth Map(X, Y) — oc, so that catMap(X,y) = oc. 2. Prove that the radical of the graded Lie algebra L = L(o, b)/[a[a, b]], with dego = 2 = degb is trivial. ? 3. Let / : f\ S3 —>¦ S3N be the natural pinch map. Convert the projection 2=1 S3VS3N —> S3N into a fibration ? —>· S3N and consider the pull back fibration ? X —> ?[ S3 along /. Compute the radical of ?*(??) ?=?
37 Cell Attachments In this section the ground ring is Q except that in the first topic it is any com- mutative ring. Consider a cellular map / : Sn —> X between simply connected CW complexes (recall from Theorem 1.2 that any continuous map is homotopic to a cellular map). The first objective of this section is to study the relationship between the rational homotopy Lie algebra of X and of the CW complex Y = XUf Dn+1 obtained by attaching an (n + l)-cell to X along /. We shall assume / is not a rational homotopy equivalence, so that Y is not a rational point. Convert the inclusion i '¦ X —>¦ Y into the fibration q : ? ? ? MY —>¦ Y, as defined in §2(c). The fibre of P, F = ? ???? is the homotopy fibre of i, and the holonomy action of Q.Y on F is simply right multiplication in PY. The inclusion oi X xy PY inlxy MY is homotopic to the projection ? : X xy PY —>¦ X, which is an fiy-fibration — the holonomy fibration for q. Note that because / is null homotopic in Y it lifts to a map g : Sn —>¦ ? ? ? PY. Thus we have the commutative diagram XxY PY = F Y = XUf Dn+1. Let [5n] denote the fundamental class in Hn(Sn) (any coefficients). Our first step is to prove • H+(F) is the free ?'.(??)-module on the element Hn(g)[Sn]. We use this to give several equivalent conditions for ?, (i) <g> Q to be surjective, in which case there is an explicit formula connecting the Hubert series for ULx = Ht(QX;<Q) and ULY = H*(UY;Q). Next we consider the case where X is a wedge of spheres and we attach a family of cells: Y = ( \/Sa ) U/ ( [J Dni+l ), via a map / = {/,· : 5n; —»¦ X). For these spherical 2-cones we derive the explicit formula for the Hubert series of ULy originally established by Anick [6]. In [140] Serre posed the question: If Y is a simply connected finite CW com- plex is the Hubert series of ULy rational? The original negative answer was given by Anick [6]. Here we construct an example due to Lofwall and Roos.
502 37 Cell Attachments It is a finite spherical 2-cone with the additional property that Ly contains an infinite dimensional abelian Lie algebra (although it cannot contain an infinite dimensional abelian ideal by Theorem 36.5). This section is organized into the following topics: (a) The homology of the homotopy fibre, ? ?? PY. (b) Whitehead products and G-fibrations. (c) Inert elements. (d) The homotopy Lie algebra of a spherical 2-cone. (e) Presentations of graded Lie algebras. (f) The Lofwall-Roos example. (a) The homology of the homotopy fibre, ? ?? PY. The path space fibration PY —> Y restricts to the holonomy fibration X xY PY —> X. In Proposition 8.4 we constructed a cellular model for C*(X ????) of the form (C ® C. (??), d) A C. (? ? ? PY) , where each Ck was free on the ft-cells of X. A slight modification of the con- struction extends this to a cellular model of the form This induces a quasi-isomorphism ) A C»(PY,X xY PY) of C*(O.Y)-modu\es. The connecting homomorphism is an isomorphism of degree -1 of H*(UY)-modu\es from H» (PY, ? ? ? ?? ) to H+{X ?? ??) and it follows that H+(X ?? PY) is the free H*(UY)-modu\e on a single class of degree n. Moreover, it is immediate from the construction that this class is H*(g)[Sn]. This establishes Proposition 37.1 H+(X ?? PY) is the free H*(QY)-module on the single basis element ?*(g)[Sn]. D (b) Whitehead products and G—fibrations. Let ? : ? —> ? be a G-fibration with fibre G at the basepoint * C B, where G is any topological monoid. Given continuous maps ? : En+1,*) —> (B,*) and b : Em,*) —> (E,*) we construct a continuous map c: En+m,*)—>(?,*)
The Rational Dichotomy 503 as follows. Lift ? to ? : (?>?+1,5?,*) —> (E,G,*) and regard b as a map (Dm,Sm-1) —»- (?,*). Then set ? b(x)-a(y) ,(x,y)€DmxSn c{x, y) - j a^ ( ^ ^ e 5m_! ? ?)n+1 Now a, 6, e represent classes ? € ???+?(?), /3 G nm(E) and 7 € 7rn+m(jE), and the restriction <9a : 5n —>¦ G of o represents the image «9,? of ? under the connecting homomorphism <9, : ?,(?) —> ?»-?(?). Finally, recall that the action of G on ? determines an action of H»{G) on H*{F) (§8). Proposition 37.2 ([S-T]) With the notation above, 1»' and hur-y = hur ? · hur d*a . proof: The first assertion is immediate from the definition of the Whitehead product (§13(e)). For the second, observe that c factors over the surjection d(DmxDn+1) —>· (SmxSn)UDm+1 todefinec : (SmxSn)UDm+1 —>· E. More- over, hur-y = H*(c) [d(Dm ? Dn+1)] = H»{c) ([Sm] ® [5n]) = hur?-hura, where [ ] denotes fundamental class. D (c) Inert elements. We maintain the notation established at the start of this section. Recall also that Lz = ?, (Q.Z) ® Q denotes the homotopy Lie algebra of a simply connected space Z, and that the connecting homomorphism of the path space fibration is an isomorphism ?,+1(?) <g> Q -—> Lz- In particular, let ? € (Lf)u-i and ? € (?,?)?_! correspond to [g] and to [/]. Then ?,(??)? = ? and so ?,(??)? = 0. Theorem 37.3 The following conditions on [/] are equivalent: (i) F has the rational homotopy type of a wedge of at least two spheres, (ii) ?, (??) : Lx —> Ly is surjective. (in) ?,(??) is surjective, and its kernel is the Lie ideal I generated by ?. (iv) The ideal I generated by ? is a free Lie algebra, and the quotient I/[I, I] is a free ULx/I module on the single generator, ?. Remark An element ? in a graded Lie algebra L is called inert if the Lie ideal / it generates is a free Lie algebra and if also //[/, /] is a free UL/I module on the single generator a. Thus Theorem 37.2 asserts that Lx —> Ly is surjective if and only if w is inert.
504 37 Cell Attachments proof of Theorem 37.3: Recall that we use rational coefficients throughout, so that #,(-) = H»{ ;Q). First observe that (i) ? (ii). Indeed, if (i) holds then Lp is a free Lie algebra on at least two generators (Theorem 24.5) and hence has no central elements. But in §28 (Example 2 combined with Proposition 28.7) we showed that elements in the kernel of k*{F) —> ?*(?) correspond to central elements in Lp. Thus Lp —> Lx is injective and Lx —> Ly is surjective. Conversely, if Lx —> Ly is surjective then it follows from Proposition 37.2 that the subspace hur[g] · ULy is in the image of the Hurewicz homomorphism, hur : 7r,(F) 0Q-> H.(F). Since hur[g] -ULY = H+(F) (Proposition 37.1) it follows that Im hur = H+(F). Now Theorem 24.5 asserts that F has the rational homotopy type of a wedge of spheres (at least two because Y is non trivial). Next observe that if (i) and (ii) hold then ?,(??) : Lp -^ kein*(Qi). Since / C ker7r,(Sz) (because ?,(??)? = 0) it is, in this case, a sub Lie algebra of a free Lie algebra. By the Corollary to Proposition 21.4 / itself is a free Lie algebra. This shows that under any of the hypotheses (i)-(iv), / itself is a free Lie algebra. Factor ?,(??) to define a morphism ? : Lx/I —y Ly and note that / C n*(Qp)Lp (= ker^(fiz))· It is immediate from the long exact homotopy sequence that; ? is an isomorphism in degrees < r <==? ?,(??) : (Lp)<r -^> I<r . C7.4) Moreover, since / is a free Lie algebra we may choose a wedge of spheres 5 = y Sha+1 such that Ls = I, and since / C Imn*(Qp) we may construct a continuous map h:\JSk*+1 —>F a such that ?, (??) ? ?, (?/i) is the inclusion of / in Lx. The main step in the proof of the theorem is Lemma 37.5 If any of the conditions (i)-(iv) of Theorem 37.2 hold then ? is an isomorphism. proof: Suppose ? is an isomorphism in degrees < r. Then C7.4) shows that ?.(?) : 7r<rE)®Q A K<r(F)®Q. Extend h to h' : S' = SvV 5[+1 V\/ 5J+2 so * i that ?»(/?') is a rational isomorphism in degrees < r + 1 and rationally surjective in degree r + 2. Then (proof of Theorem 8.6 or via an easy argument using Theorem 7.5) H»(h!) is an isomorphism in degrees < r + 1. Let ? e 7rr+i(F) <g> Q correspond to ? e (Lf)»·· Then hur ? = hur[g] -? for some ? e (ULy)r+i-n. Since ? is an isomorphism in this degree we can write ? = ??' for some ff e ULX/I. Thus Proposition 37.2 asserts that hur ? = hur ?'
The Rational Dichotomy 505 where ?»(?)/3' is a linear combination of iterated Whitehead products starting with [/]. However, the correspondence ?»(A')<g>Q = (L.y)»_! converts Whitehead products to Lie brackets (Proposition 16.11). Thus if a' € (Lp)r corresponds to ?' then ?»(??)?' € /. Moreover, the image of ?»(??)?' in //[/,/] is just ? ·?'. On the other hand, since ? — ?'?. ker hur and since ??+? (?/) <g> Q and Hr+i (h1) are isomorphisms it follows that ? — ?' = ??+?(/?'O, where 7 e ??+?E') <g> Q is in the kernel of the Hurewicz homomorphism for 5'. Hence ? — ?' = ??+?(?O, where 7 € ??+1 (,5)<8>Q corresponds to an element 7 € [L5, L5]. Thus ??,(??)(? — We are ready to show that ??+? is an isomorphism. Indeed, ?, (??)? = ?»(??)?' + ?»(??)(? — ?') € /, as just shown, and so K:,(Clp)(Lp)r = Ir. On the other hand if (i), (ii) or (iii) hold then ?,(??) ". Lp —>¦ Lx is automatically injective (because (i) => (ii)). Suppose (iv) holds and that ?.(??)? = 0. As just shown ?»(??)(?' — ?) € [/,/] and ?,(??)?' represents ?-?' in //[/,/]. Since ULx/I acts freely on ? in //[/,/] it follows that & — 0 and so hur ? = hurlgj-??' = 0. This implies (as above) that ? € nr+i(h) (??+1E) <g> Q) and hence that ? € ?,»(?/?)?,5. Since ?»(??) is injective in this subspace, ? = 0; i.e. ?»(??) : (LF)r+i —> (Lx)r+1 is injective. D We now return to the proof of Theorem 32.3. The Lemma asserts that if (ii) holds then ? is an isomorphism. This implies at once that (ii) => (iii). Moreover since ? is an isomorphism if (iv) holds then in this case ?. (??) : Lp —i- I and hence ?,(??,) : Ls —>¦ Lp is an isomorphism, and h is a rational homotopy equivalence. Thus (iv) =$¦ (i). Finally if (iii) holds then again h : 5 —>¦ F is a rational homotopy equivalence. This defines a degree 1 isomorphism Ls/[Ls,Ls] -=> H+(F). Now Ls = I and it follows from Proposition 37.2 that the action of Lx/I in //[/, /] corresponds to the action of Lx/I in H+(F). Since the latter is a free ULx/I module on a single generator (Proposition 37.1) so is the former. Thus (iii) => (iv). D (d) The homotopy Lie algebra of a spherical 2-cone. Consider a continuous map f:Z = V5m"+1 —>X = VSna+1 ? a between two simply connected wedges of spheres with rational homology of finite type; and write Y = X U/ CZ. Then (Theorem 24.5) Y has a free Lie model of the form (Lv<sw,d) in which: sV = H+(X), dV = 0 and W = H+{Z) and d:W—>lv- Assign a bigrading to this differential graded Lie algebra by setting Vp = VQ^P and Wq = Wi,9_i. Then d has bidegree (—1,0). In particular the homotopy Lie 00 algebra Ly = H (Lv&w,d) inherits a bidegree: LY = 0(Ly){ ». This exhibits i=0
506 37 Cell Attachments Ly as the direct sum LY = L ? / of the sub Lie algebra L = (LyH,, and the ideal / = (Ly)+i*. Proposition 37.6 With the hypotheses above, I is a free graded Lie algebra. proof: Denote h\/®w simply by L. Then o^Lx,,) is an ideal in Lo., and so ? = L+,, © d(Li,,) is an ideal, in L with quotient h/E = (L,0). Now apply the cochain construction of §23 to obtain a relative Sullivan algebra Cm(L,0) —>· Cm(JL,d) with Sullivan fibre C"(E,d). Recall that C"(L) = A(sL)% which we bigrade by setting (sLp,,)9 = [(sL)lt]p+1'9. Thus the differential in C*(L) has bidegree A,0). Now let (E',d) C (E,d) be the sub Lie algebra given by @ if ? = 0 kerd if ? = 1 Ep ifp>2. The inclusion (E',d) —>¦ (E,d) is a quasi-isomorphism and so (Proposition 22.5) C*(E,d) -=> C*(E',d). In particular it follows that: H+ (C*(E,d)) is concentrated in bidegrees (p,q) with p>2. C7.7) On the other hand, H+ (C*(L, d)) ? ? ((sV ? sTV)8,d), by Proposition 22.8. Hence H+ (C*(L,d)) is concentrated in bidegrees (p,q) with ? = 1,2. It follows that there is a quasi-isomorphism C*(L, d) -^ (^4,d) of bigraded commutative cochain algebras in which A = Q © .41'* © /I2'*. Finally, we can extend C*(L,0) to the acyclic relative Sullivan algebra C* (L, (UL)^) by dualizing the construction of §23(c). Then we have quasi- isomorphisms C*{E) A C*(L) ®c*(L) C* (L, (UL)*) -»¦ ^ ®C.(L) C* (L, (UL)*) . Now v4<8>c-(L)C* (L, (ULf) = v4®(?/L)8 is concentrated in bidegrees (p,*) with ? < 2, because (L)9 is concentrated in bidegrees @, *). It follows that H (C*(E)) is concentrated in bidegrees (p, *) with ? < 2. Comparing this with C7.7) we see that H+ (C*(E,d)) is concentrated in bidegrees B,*). This implies that there is a quasi-isomorphism C*(E, d) -=» (k ? if2'*, ?) with multiplication in H2'* trivial. It follows that / = H(E,d) ^ H(CC(E,d)) = L(s-1ii2'*) is a free graded Lie algebra (use Theorem 22.9). D Recall from §3(e) that the Hubert series for a graded vector space V>0 is the oo formal power series V(z) = J^dimVnzn. Similarly, if H = H^° then H{z) = ?
The Rational Dichotomy 507 Proposition 37.8 (Anick [6]) With the notation preceding Proposition 37.6, ULY(z)-* = A + z)UL(z)-1 -(z- H+{Z){z) + H+{X){z)) . Corollary If X and ? are finite wedges then ULy(z) is rational if and only if UL(z) is rational. proof of Proposition 37.8: If Co <— C\ <— ·¦· <— Cn is a finite dimen- sional chain complex then ?(—l)p dim HP(C) = ?(—l)pdimCp, as follows by a trivial calculation. Hence if Co,* <— Ci,, <— ··· <— Cn,« is a finite complex of graded vector spaces of finite type with differential of bidegree (—1,0) then ?(-1)? dim Cp,q = ?(-1)? dimHp,,, q > 0. This implies that Apply this (in the notation of the proof of Proposition 37.6) to the complex (A ® (ULf) = A* ®UL to obtain [A2>*(z) - ???'·(?) + ?2 ?0'* (?)] UL(z) = [s S Os) + ?2] , where / = L5. On the other hand since C*(L, d) -=> (A, d) it follows that H+(A) is the dual of the homology of the complex {sVf —»¦ {sWf . Thus = H2'*{A){z) - ???'·{?){?) + z2H°>*(A)(z) = (sW)(z) -z(sV)(z) + z2 = H+{Z){z)-zH+{X){z) + z2 . Finally, ULY(z) = UI(z)UL(z) and UI(z) = TS(z) = A - S(z))-1. A short calculation completes the proof. ? (e) Presentations of graded Lie algebras. Let L = L>i be a graded Lie algebra. We may always write L = Ly /J where J C [Ly ,Ly] is an ideal in the free graded Lie algebra Ly. In this case a basis (vQ) of F is a minimal set of generators of L. Similarly the Lie bracket defines a representation of Ly in J. Decompose J = R® J · ?/Ly. Then a basis (r/j) of R is a minimal set of generators for the ideal J. The representation L = h(vQ)/(r3) is called a minimal presentation of L.
508 37 Cell Attachments Proposition 37.9 With the notation above (s2 = double suspension), (i) sV = Tor"L(Q,Q). (ii) s2i?STor^(Q,Q). proof: (i) Indeed V = U//[Ly,L;/] = L/[L,L] and so sV = Torf1"(Q,Q) by Example 1, §34(a). (ii) Consider the Hochschild-Serre spectral sequence converging from TorJ^ (Q,Tor^J(Q,Q)) to Tor^v(Q,Q). Again by Example 1 of §34(a) we may identify TorfJ(Q, Q) = s (J/[J,J]) with the representation of UL induced by Lie bracket in J. It follows that El, = Tor%L (Q,TorfJ(Q,Q)) ~ s (J/[J, J] ®UL Q) ~ sR . On the other hand since Lv is free we have ???^? (Q, Q) = sV and Tor^v (Q> Q) = 0 — cf. Proposition 21.4. Since Effi = Tor^ (Q, Q) = sV it follows that the differential d2 in the spectral sequence satisfies d2 : E2fl = To#L(Q, Q) A E2tl ^ sR . D Again let L = L(uQ)/(r/j) be a minimal presentation of a graded Lie algebra L = L>\. Then without loss of generality we may suppose the rp are integral combinations of Lie brackets of the vQ. Put deguQ = nQ and degr/j = ???. Then Lv is the homotopy Lie algebra of the space X = V 5na+1 and the r? define a a continuous map / : ? = \fSm0+1 —>X = VS"Q+1 · ? a As in §37(d) set Y = X U/ CZ. Then as described there the homotopy Lie algebra Ly is the direct sum of a sub Lie algebra and an ideal /, and the sub Lie algebra in question is exactly L = L(vQ)/(r?) . Thus we may reinterpret Proposition 37.9 as H+ (X) = Torf1"(Q, Q) and sH+{Z) ^ Tor^L(Q, Q) . In particular, L is finitely presented if there are only finitely many generator va and finitely many relations rp and in this case Y = X U/ CZ is a finite CW complex. Since L is a sub Lie algebra of Ly we deduce: Theorem 37.10 Let L = L>, be a finitely presented Lie algebra. Then there is a finite simply connected CW complex Y = X U/ CZ whose homotopy Lie algebra Ly is the direct sum Ly = L@I
The Rational Dichotomy 509 of L as a sub Lie algebra and an ideal I. Moreover I is free as a graded Lie algebra and ULY(z)-1 = A + z)UL(z)-1 -(z- H+(Z)(z) + H+{X){z)) . (f) The Lofwall-Roos example. Here we construct a finite CW complex Y, due to Lofwall and Roos [112], such that the Hubert series ULy is irrational and also Ly contains an infinite dimensional abelian Lie algebra. To begin, note that the cohomology algebra H = H ((S'2 V S2) ? (S2 V S'2)) has the form [?(?, b)/a2, ab, b2] ® [A(x, y)/x2, xy, y2] where a,b,x,y are classes of degree 2 corresponding to the four copies of S2. Put V = ? kvi with deg vi = 2 and define a relative Sullivan algebra (H ® AV, d). by setting dv2 - 0, dv3 = (b + y)vs + ax and dvi = (b + (-l)i+1j/)^'-i) i > 4- Now the minimal Sullivan model for ? is the cochain algebra (AW,d) = C* (LE,6)) <g> C* (L(x,j/)), whose differential is the purely quadratic operator given by the Lie bracket. The quasi-isomorphism (AW,d) —> (-#,0) extends to a wordlength preserving quasi-isomorphism of the form (AW <g> AV, D) —> (H <g> AV, d). Thus it is also a morphism (AW <g> AV, D2) —> (H ® AV, d), where D2 is the quadratic part of D. However, D = D? in AW, and the quotient differential ~D in AV satisfies ~D = d = 0, so that ~D = ~D2 = 0. It follows that (AW®AV,D2) -^(H®AV,d). Let H^ (AW®AV, D-2) be the cohomology represented by cocyles of wordlength p. Then H^(AW ® AV,D2) = H^(H ® AV,d) and a short calculation shows that bases for H^(H ® AV,d) and H^(H ® AF,d) are given by. H^ : a,b,x,y,V2 H^ : ax,ay,bx,by,av2,xv2,v\ . On the other hand (Example 1, §23(a)) a graded Lie algebra L is determined by the condition C*(L) = (AW®AV,D2) . Moreover, H^ = Ext^L(Q,Q) and H& = Ext^L(Q,Q) are dual to TorfL (Q, Q) and Tor^L(Q,Q). Thus (Proposition 37.9) L is finitely presented with five gen- erators in degree 1 and seven relations in degree 2. Let
510 37 Cell Attachments be the corresponding CW complex described in Theorem 37.10. Then L is a sub Lie algebra of Ly and ULY{z)'1 = A + z)UL(z)-1 - (z - 7z3 + 5z2) . Finally, since C*{L) = (AW ® AV,D3) we have (sL)» = W ? F with Lie bracket dual to D-i- Since .?>2 = 0 in AV it follows that there is a short exact sequence 0—> I —*L —>· L(a,b) ? L(x,j/) —>0 where / is an abelian ideal in L with basis U2,U3, · · · , and degu^ = A; — 1. In particular • Ly contains the infinite dimensional abelian sub Lie algebra ? Qufc. k • ULy1 is the irrational power series Exercises 1. Let i : X —>· F = X U Dn+1. Prove that: (a) If i : CP2 —>¦ CP3 denotes the natural inclusion then dimkerTr, (??) — dim Coker ?,(??) = 1. (b) If i : 53 V 53 —>· 53 ? 53 denotes the natural inclusion then ?, (fii) is surjective. (c) If i : Xe—>X V 5n, ? > 2, denotes the natural inclusion then ?, (??) is not surjective and dim Coker ?,(??') — oo. 2. Let X be a simply connected topological space with rational homology of finite type. Using Theorem 29.1, prove that if cat0 X = 2 then cl0 X = 2. 3. Let F be the homology fibre of the inclusion i : X c—> X U Dn+1 = Y and let d : UY —> f be the inclusion {*} ? UY <^-> F xy PF. Prove that if H,d ? 0 then depth H, (QY) = 1.
38 Poincare Duality In this section the ground ring k is an arbitrary field of characteristic zero. Suppose A = {v4!}o<?<n is a finite dimensional commutative graded algebra such that A0 = k, and suppose ? a is an element in the dual space (An)a. Define bilinear maps ( , ) : An~p ? Ap —> Jk by (a,b) =uA(a-b) , a € An~b , b e Ap . Definition .4 is a Poincare duality algebra with fundamental class ? a if the bilinear maps ( , ) are all non-degenerate scalar products. In this case ? is called the formal dimension of A. If M is a compact simply connected manifold then its cohomology algebra H*{M) satisfies Poincare duality [68] and this is the principal condition that a simply connected space must satisfy in order to have the rational homotopy type of a compact manifold [18]. In this section we consider the rational homotopy theory of simply connected topological spaces X whose cohomology satisfies Poincare duality. The main results are: • If X is rationally elliptic then H*(X) satisfies Poincare duality. • If H*(X) satisfies Poincare duality then eo(X) = cato(-^)· ·¦ Suppose X satisfies Poincare duality with formal dimension ? and write X = ? U/ Dn so that Hn(Dn,Sn~1) A Hn{X). Then the element ? ? (Lx)n-2 corresponding to [/] € ??-?(?) is inert. This section is organized into the following topics: (a) Properties of Poincare duality. (b) Elliptic spaces. (c) LS category. (d) Inert elements. (a) Properties of Poincare duality. Let A be a finite dimensional commutative graded algebra of the form k ? {¦^!}?<?<? and suppose ? a G An. The bilinear maps ( , ) defined above deter- mine linear maps ? : An~p —> {??? by 9(a)(b)=uA(a-b) = (a,b) . Since (.<4p)tt = (As)~p, we may regard oasa linear map A —> A^ of degree —n.
512 38 Poincare Duality Lemma 38.1 (i) A satisfies Poincare duality if and only if ? is an isomorphism. (ii) If ? = L·® {Bl}i<i<m is a second finite dimensional commutative graded algebra and if A <g> ? satisfies Poincare duality then so do A and B. (Hi) If A is equipped with a filtration and the associated graded algebra satisfies Poincare duality then so does A. proof: (i) is by definition, (ii) follows from ??®? = #.4 ®?? and the fact that a tensor product is an isomorphism if and only if both tensorands are. (iii) is an obvious calculation. D Next consider a commutative differential graded algebra (.4, d), and note that (A*,dl·) is a chain complex. Lemma 38.2 (i) If A is a Poincare duality algebra and with fundamental class ? ? € (??? and if Su a = 0, then H (A) is a Poincare duality algebra with fundamental class [? a}· (ii) Suppose ? € Aeven is a cocycle and define (A®Au,d) by du = ?. If H {A) is a Poincare duality algebra, then H(A(&Au, d) is a Poincare duality algebra. proof: (i) Because euA = 0, BAd = (~l)nSeA and ??(?) = ? (? a) is an isomorphism. (ii) This is a simple exercise left to the reader. D (b) Elliptic spaces. Proposition 38.3 If (AV, d) is an elliptic Sullivan algebra (introduction to §32) then H(AV,d) is a Poincare duality algebra. Its formal dimension is ? = X^degXj — X3(degy_; — 1), where (??) is a basis of Vodd and (yj) is a basis of i ? ¦(/¦even proof: Recall the odd spectral sequence defined in §32(b). Its first term is (AV,da) with tk(I/even) = 0 and ??(???) c Veven. Proposition 32.4 asserts that H(AV,da) is finite dimensional. Choose r sufficiently large that each yj = d^j and extend (AV, da) to (AV <g> AU,da) by assigning U the basis (uj) and setting daUj = j/J. Then (AV <g> (\ 1 \ AU,da) -?» ®Ayj/uJ ®AVodd,do- and it follows from Lemma 38.2(ii) \[j J / that H(AV <g> AU, ??) satisfies Poincare duality. On the other hand, H(AV <g>
The Rational Dichotomy 513 AU,da) = ?(??,??) <8>A(ui - ??,... ,u, - ?,) and so Lemma 38.1(ii) asserts that H(AV, da) satisfies Poincare duality. Finally, Theorem 32.6 shows that ? = X^degZj — J2(degyj — 1) is the top * i degree in which both (AV,da) and (AV,d) have non-vanishing cohomology. Since H(AV,da) is a Poincare duality algebra, Hn(AV,da) — k[z] and so this class must survive through the spectral sequence. Thus if Ep is the p-th term of the spectral sequence then (E%)* = ku\ and ?*?? - 0. By Lemma 38.2(i), each Ep is a Poincare duality algebra. Hence the bigraded algebra associated to H(AV, d) is a Poincare duality algebra and thus so is H(AV,d) (Lemma 38.1(iii)). ? (c) LS category. Theorem 38.4 Let X be a simply connected topological space and let (AV,d) be a minimal simply connected Sullivan algebra such that H(X) and H(AV,d) are Poincare duality algebras. Then eo(X) = catoX and e(AV,d) = cat(AV,d) . Corollary If X and (AV, d) are elliptic then eo(X) = cato.Y and e(AF, d) = cat(AV,d). proof of Theorem 38.4: Choose ? € (AV)& so that d*z = 0 and ? represents a fundamental class of (AV,d). Define ? : (AV,d) —> ((AF)",^) by setting ??(?) = ?(? ? ?), ?, ? € AV. Then ? is a linear map of (AV,d)-modules and ?(?) is an isomorphism because H(AV, d) satisfies Poincare duality. On the other hand in §29(h) we introduced the invariants meat and e for (AV, c?)-modules (M,d) in terms of a semifree resolution of (M,d). In particu- lar, these invariants coincide for quasi-isomorphic modules. Thus according to Theorem 29.16, e(AV,d) = e ((AVy,d*) = cat(AV,d) . Finally, the assertion for X follows from this via the minimal Sullivan model for X. D (d) Inert elements. Theorem 38.5 (Halperin-Lemaire [85]) Suppose X = ZU/Dn is ? simply con- nected space such that H*(X) is a Poincare duality algebra of formal dimension n, and Hn{Dn,Sn) -=> Hn{X). If the algebra H*(X) is not generated by a single element, then the element a € (Lz)n-2 corresponding to [/] € 7rn_i(Z) is inert. proof: Let (AV,d) be a minimal Sullivan model for X and extend it to a minimal relative Sullivan algebra (AV <g> AV, d) such that H[AV ® AV*, d) = Jfc.
514 38 Poincare Duality Thus the linear part of d will be an isomorphism V -^> V, which we denote by V ? > V. Because X is simply connected, V = V^2 and d = 0 in V2. Thus V2 = H2(AV,d). Write (AV)n = Jn ? ??, where ? is a cocycle representing a ba- sis element of Hn(AV,d), and Jn D (Imc?)n. Poincare duality implies that multiplication ?2 ? Hn~2 —> Hn is non-degenerate. It follows that there is a subspace Jn~2 C (AF)" such that Jn~2 ? (kerc*)" = (AV)n~2 and such that V2 · Jn~2 C Jn. Now extend these subspaces to a differential ideal (J,d) C (AV,d) by setting Jk = 0, k < ?-2, Jn~l = (AVO1'1 and Jk = (AV)k, k > n. Again by Poincare duality, ??"'(?7, d) = 0 and so it follows that (AV, d) —> (AV/J, d) is a quasi-isomorphism. Denote AV/J simply by A and denote the image of ? by ?. Then a Sullivan representative for (A, d) —> (?/??, d) is also a Sullivan representative for the inclusion i : ? —> X. Thus if we define (.4 ? ku,d) by setting du — ? and u · A+ = 0 the composite (AV,d) -=» (A,d) —> (?? hu,d) is also equivalent to a Sullivan representative for i. Extend this composite to a quasi-isomorphism (AV <g> AW, d) -^» (? ? Jku, d) from a relative Sullivan algebra (AV ® AW, d). Then Proposition 15.5 identifies the quotient Sullivan algebra (AW, d) as a Sullivan model for the homotopy fibre of i. Thus by Theorem 37.3 we have only to show that (AW, d) is a Sullivan model of a wedge of spheres. Now, as observed in Lemma 14.2, we have quasi-isomorphisms (AW, d) ^ (AV <8) AW, d) ®?? (AV <g> AV, d) = (AV <8) AW <8) AV, d) ¦=> ((A<5>]ku)®AV,d) . Suppose now that the least integer r such that Vr ? 0 is odd. In this case we shall construct a quasi-isomorphism (B,0) -=> ((^4 0 fcu) <&AV, d) with B+ 'B+ = 0. By the Example of §12(c), (B,0) is a commutative model for a wedge of spheres, and hence (AW, d) will be the Sullivan model for the wedge. For this observe that Hr(AV,d) = Vr = Ar. Choose ? ? 0 in Ar and (by Poincare duality) a cocycle ? € An~r such that vz = ?. Then A+ = Mv ? /, where / is the differential ideal in A of elements a satisfying za — 0. Extend ? to a basis ?, ??,..., Ok, · · · of AV with degOk < degufc+i and define ? : AV —i AV by 6(vsa) = ^fj-a, a € ?(??,... ,vk,...). Now dv = v. Suppose dvq e 7_(g) AF for 1 < q < k. Then dvk = ? <g> ? -I- w, ? € ?(?,.. .ojfc-i), w e I (g) AV, and d(vk - 0?) el® AV. JReplace vk by vk — ?? and continue in this way to arrange that dvk € / <g> AF for all i < k. Let -B+ c(i9 ku) (8> ? F consist of the elements of the form u <g> ? - ? <g> ??, ? € AF. Since ?« = ? and /? = 0 it follows that d(B+) = 0 and B+ · B+ = 0. Since ii(yl <g> AF) = A it is immediate that the inclusion of ? = h 0 B+ in AF is a quasi-isomorphism, as desired.
The Rational Dichotomy 515 We now deal with the general case. Let I be the least odd integer such that Ve ? 0. We prove the theorem by induction on ? — ? dimV\ The case 0<i<i ? = 0 is the case above when the least positive integer r such that V ? 0 is itself odd. For the inductive step we may take r even and, as above, write A+ = hv ? / where ? is a non zero element of degree r, ? is a cocycle such that vz = ? and I = {ae A\az = 0}. Now write V = kv ® V\. Since ? has odd degree H(AV <g> Av, d) also satisfies Poincare duality, with formal dimension ? + r — 1 (Lemma 38.2(ii)). Moreover we have quasi-isomorphisms (AV, d) #- (?V <g> AC, d) -^ (A <g> ??, d) <^ (Ax, d) where Ai C A <8> ?? is the subalgebra i: ? (/ (g) ??) ©?;(?®?). On the one hand this implies that H(A,d) is a Poincare duality algebra with top cohomology class represented by ? (g) v. On the other hand, it allows us to assume that the differential in (A <8> AV, d) sends V\ into Af <g> AV^. Since the theorem is true by the induction hypothesis for (AVi,<i) and since ?(??®?) = ?®? and (u®d)(Af) = 0 it follows that (B,d) = ([yli © J:(u ® ?)] <8> AVi,<i) is a commutative model for a wedge of spheres. Finally, since (??, d) -^> (yl <g> ??, c?) it follows that ([?? ? (Au <8> ??)] (8) AVx, d) -^((ie Au) <8> AV, d) . This source algebra can be written as ? = B\ ® (u — ?® ?)???. By construction d(u — ? ® v) = ? — ? = 0 and (u — ? (g) ?)?+ = 0. It follows that (u — ? <g> ?)??? consists entirely of cocycles that multiply Z?+ to zero. Thus since {B\, d) is a commutative model for a wedge of spheres so is (B, d), and the theorem follows by induction. ? Exercises 1. Let / : M —> ? be a continuous map between 1-connected n-dimensional compact manifolds. Suppose that Hn(f;Q) ? 0. Show that cato(M) > cato(iV). 2. Let / : M —> ? be a smooth map between 1-connected compact manifolds. Prove that if either (a) / is a fibration whose cohomology Serre spectral sequence collapses at the ?2-term, or else (b) or else, / is a locally trivial bundle whose fibre F satisfies x(F) ? 0, then cato M > cato N.
516 39 Seventeen Open Problems 39 Seventeen Open Problems We close this monograph with seventeen of our favourite problems, all of which remain open as this goes to press, and many of which are decades old. Other compilations of problems can be found in [46] and [147]. 1. Suppose F is a rationally elliptic space with non-zero Euler-Poincare charac- teristic, and F —> ? —> ? is a Serre fibration of simply connected spaces. Does the (rational) Serre spectral sequence always collapse at E-2 ? A positive answer was conjectured by Halperin in 1976. The cohomology of such spaces F has the form H = ?(??,... ,xn)/(fi, ¦ ¦ ¦ > /?), where xi,...,xn have even degrees and f\,..., fn is a regular sequence in the polynomial algebra ?(??,..., xn). Thus an equivalent formulation of the problem is: For graded algebras of the form ? is zero the only derivation of neg- ative degree? Positive answers to this question have been given in the cases that: all the Xi have the same degree ([161]), ? < 2 ([146]), ? < 3 ([113]). if F is a homoge- neous space ([141]), and if each polynomial fi is homogeneous with respect to wordlength ([120]). In [132] it is shown that the answer is positive in the generic case. Essentially the same question has arisen in the context of deformation of sin- gularities ([154], [36]) where attention is restricted to the case that fi = J?· for some single polynomial /. 2. Suppose an r-torus acts continuously with finite isotropy groups on a closed simply connected manifold M. Is it true that dimH*(M;Q) >2r? The torus rank conjecture asserts that the answer to this question is positive. It can be easily reformulated in terms of Sullivan models as follows: Suppose a simply connected Sullivan algebra A(V, d) whose cohomol- ogy satisfies Poincare duality occurs as the Sullivan fibre of a relative Sullivan algebra of the form (?(??,..., xr) <8> AV, d) with deg ?? = 2, 1 < i < r. If ? (A(xi,..., xr) <g> AV) is finite dimensional, is it true that dim H(AV, d)>2r? The torus, rank conjecture has been established for homogeneous spaces ([83]), for homology Kahler manifolds ([4]), for r < 3 ([5]) and for cohomologically symplectic manifolds satisfying the hard Lefschetz theorem ([115]). It implies that if L = L>i is an evenly graded finite dimensional Lie algebra over Q, then dimExtUL(Q.~Q) >2dimcentreA'). 3. If X is any finite simply connected CW complex and if ? > dimX is there a rationally elliptic CW complex Y such that X ~q Yn, Yn the ?-skeleton ofY?
The Rational Dichotomy 517 This question has been attributed by Anick as a conjecture. The answer is obviously affirmative if k<n(X)®Q is concentrated in odd degrees and a positive answer is given in [106] for the case ? dim7r2;(-Y) <8> Q = 1. 2i<N 4. // X is a rationally hyperbolic finite simply connected CW complex does Lx contain a free Lie algebra on two generators? A positive answer to this question would provide an 'explanation' for the exponential growth of Lx, and a positive answer was conjectured (separately) by Avramov and Felix in 1981. The case cato X = 2 was settled in [58] and the far more general case that depth Lx = 1 was settled positively in [29]. The case when A' is formal and H*{X) is evenly graded and generated by at most three elements is settled positively in [13]. 5. Suppose X is a simply connected space with rational homology of finite type. If Heven(X;Q) and the image of the Hurewicz homomorphism are both finite dimensional does it follow that H*(X;Q) is finite dimensional? This is trivially true for formal spaces, because in this case a generating space for the algebra H*(X;Q) is dual to the image of the Hurewicz homomorphism. If Heven{X; Q) = 0 then Baues' theorem [19] states that X is rationally a wedge of spheres, which implies a positive answer in this case too. This question has been referred to in the literature as the Omnibus Conjec- ture', the origin of the terminology being a false proof produced by the second author during a particularly slow (omnibus) train ride from Leuven to Louvain in 1981. 6. Suppose X is a simply connected rationally hyperbolic finite CW complex of dimension n. Are there numbers ? > 0 and C > 1 such that s+n-l ? dim7TiPO<8>Q> XCS , all s > 1 ? i-s+l A positive answer to this question was conjectured by the authors in the early 1980's. The gap theorem (Theorem 33.3) establishes the weaker (!) inequality with XCS replaced by 1. An important case where there is a positive answer is established by Lambrechts in [104]. When X is rationally hyperbolic the exponential growth of Theorem 33.2 implies that the Hilbert series ?/,(??)(?) has radius of convergence ? < 1. Lambrechts gives a positive answer to this problem when the singularities of H„ (??) (?) on the boundary circle of radius ? are all poles. There are many examples of such spaces, and no examples where this condition is known to fail. 7. Is the radical R C Lx of a simply connected finite CW complex X of dimension n, concentrated in degrees < 2(n — 1), and is dimR < n?
518 39 Seventeen Open Problems When X is elliptic all of L\ is concentrated in degrees < 2(n — 1), and dimLx < 2dim(jLx)even < catX < n, which motivated the authors to con- jecture a positive answer to this question in 1982. A weaker question asks: Are all rational Gottlieb elements of X of degree < 2n — 1 ? and in the same vein, If X is rationally hyperbolic is there an a € (Lx)even of degree < n—l such that ad ? is not locally nilpotent? The authors believe that the answer to both questions should be yes. 8. Do the rational homotopy types of the configuration spaces of a simply con- nected compact manifold M depend only on the rational homotopy type of M? The configuration space of k points in M is the subspace of points (??,..., xk) e ? ? · · · ? ? such that Xi ? Xj for i ? j. In [22] it is shown that the ratio- nal homology of the configuration space depends only on the rational homotopy type of M and in [103] and [150] a positive answer is given for smooth complex projective varieties. 9. Let X be a finite simply connected CW complex. Is X rationally elliptic if and only if for each prime number ? some pr annihilates all the p-primary torsion ???,(?)? A positive answer to this question was conjectured by J.C. Moore in the 1970's, and the Moore conjecture has been a focus of research in unstable homotopy ever since. In [117] it is shown that rationally elliptic spaces satisfy the Moore condition for all but finitely many primes. In [10] this is shown for 2-cones. The full Moore conjecture is known for spheres ([148] and [97]). A survey of progress as of 1988 is given in [138]. 10. Let X be a finite simply connected CW complex. Is it true that either H*(UX;Z) has ? torsion for all but finitely many primes ? or else that Ht(QX;Z) has ? torsion for only finitely primes p? Note that if two finite simply connected CW complexes have the same rational homotopy type then they have the same homotopy type after inverting only finitely many primes, so that the question above depends only on the rational homotopy type of X. Moreover, by using 'integral' Sullivan models it is possible to compute H* (QX; %·(?)) for all but finitely many primes ? directly from a finite dimensional commutative rational model for X. The question has a positive answer for rationally elliptic spaces X since [117] shows that then ?* (??;?) has p-torsion for only finitely many primes.
The Rational Dichotomy 519 11. If X is a simply connected, rationally hyperbolic, finite CW complex do the betti numbers bi = dim Hi(QXs ; Q) of the free loop space grow exponentially? A positive answer to this question was conjectured by Vigue in [153], where it is proved when ?' is a wedge of spheres or a manifold of LS category 2. A positive answer is given in [105] for non-trivial connected sums of manifolds. This conjecture is motivated by the fact that the fibration UX —>· Xs1 —>· X admits a section and the fact that H*(QX;Q) grows exponentially. However there is a cautionary example in [153] of a rationally elliptic X such that ?/*(??'; Q) grows like a cubic while Ht (Xs1 ; Q) grows only like a quadratic. The original interest in the question arises from the fact that the betti num- bers of Xs can be used to give a lower bound estimate for the number of geometrically distinct closed geodesies on X as a function of their length [74]. 12. Are all simply connected compact riemannian manifolds M with non-negative sectional curvature rationally elliptic? A positive answer to this question has been attributed to Bott as a conjecture, and in [24] a result is established in this direction. A positive answer in general would solve the Chern-Hopf conjecture that the Euler-Poincare characteristic is non-negative and the Gromov conjecture [75] that dimH*(M;Q) < 2n, ? = dim M. It would also shown that the rationally hyperbolic manifold of [89] with a metric of positive Ricci curvature did not admit a metric of non-negative sectional curvature. In a somewhat different direction Paternain [133] conjectures a positive answer to the question: Are all simply connected compact riemannian manifolds with com- pletely integrable geodesic flows rationally elliptic? 13. Let X be a rationally hyperbolic finite simply connected CW complex. Is it true for some a in the homotopy Lie algebra of ? autx (X) that ad ? is not locally nilpotent? Here auti(X) is the monoid of continuous maps f : X —> X homotopic to the identity, and the homotopy Lie algebra in question is just ?» (auti (X)) ® Q with the Samelson product. This question is posed by Salvatore in [136]. Note that if X is rationally elliptic the homotopy Lie algebra ?, (autx (X)) ®Q is finite dimensional. Thus this question provides another (conjectural) characterization of the elliptic-hyperbolic dichotomy. A second question dealing with the rational homotopy type of ? auti (X) was raised by Schlessinger in 1976: Does every simply connected space Y have the rational homotopy type of some ? auti (X)?
520 39 Seventeen Open Problems 14. Suppose given i : X —> Y = X U/ Dn+l and assume H*(Y;Q) is a finite dimensional algebra not generated by a single element. If f is not inert does coker ?, (i) <g> Q grow exponentially? 15. Suppose ?, ? and ? are simply connected rational CW complexes with ho- mology of finite type. Is it true that XxZ~XxY implies ? ~Y? This question has a positive answer when ?, ? and ? are formal spaces ([25], [27])· 16. Let X be a simply connected rationally hyperbolic CW complex with homotopy Lie algebra Lx. Let ?? = dimX and set k+??-? k+nx-l rk= ? dim(Lx)i and mk = ]T dim (LX/[LX, Lx))i . i-k+1 i=k+l Does nik/rk —> 0 as k —> oo? This question is motivated by the theorems that show that Lx has 'lots' of non-vanishing Lie brackets and by computer experiments that suggest that [Lx, Lx] is large. A positive answer would assert that in sufficiently large degrees [Lx,Lx] was arbitrarily close to all of Lx. A stronger version of the question asks Is lim (mk/rkI/h =0? fc-+oo and this is closely related to a problem in [105]: If (AV, d) is the Sullivan minimal model for X does the Hilbert series for AV have a strictly smaller radius of convergence than the Hilbert series for H(AV, d2) = Ext^* (Q, Q) ? 17. Find an explicit number ? € @,1) that is not the radius of convergence of the Hilbert series Ht(UX;Q)(z) of the loop space homology of a finite simply connected CW complex. There are, up to homotopy type, only countably many simply connected finite CW complexes. Thus only countably many numbers, ?, appear as the radius of convergence of H*(UX;Q)(z) in the question above. However the series for Sn+1 ? Sn+1 V · · · V 5n+1 (r copies) is A - rzn)~l with radius of convergence (l/rI/". Thus the numbers ? appearing as radii of convergence form a dense subset of [0,1]. Moreover these numbers ? can be transcendental as is shown by an example in [8].
References [1] J.F. Adams, On the cobar construction. Proc. Nat. Acad. Sei USA 42 A956), 409-412. [2] J.F. Adams and P.J. Hilton, On the chain algebra of a loop space. Com- ment. Math. Helvetia 30 A956), 305-330. [3] C. Allday and S. Halperin, Lie group actions on spaces of finite rank Quart J. Math. Oxford B) 29 A978), 69-76. [4] C. Allday and V. Puppe, Bounds on the torus rank, in Transformation groups, Poznam 1985. Lecture Notes in Mathematics 1217, Springer-\;erlag A986), 1-10. [5] C. Allday and V. Puppe, Cohomological Methods in Transformation Groups. Cambridge Studies in Advanced Mathematics 32. Cambridge Uni- versity Press A993). [6] D. Anick, A counterexample to a conjecture of Serre. Annais of Math. 115 A982), 1-33, and 116 A982), 661. [7] D. Anick, Non-commutative graded algebras and their Hubert series. Jour- nai of Aigebra 78 A982), 120-140. [8] D. Anick, Diophantine equations, Hubert series and Undecidable spaces. Annais of Main. 122 A985), 87-112. [9] D. Anick, A loop space whose homology has torsion of all orders. Pacific J. Math. 123 A986), 257-262. [10] D. Anick, Homotopy exponents for spaces of category two. in Algebraic Topology - Rational Homotopy, Lecture Notes in Math. 1318 A988), 24- 52. [11] D. Anick, Hopf algebras up to homotopy. J. of Amer. Math. Soc. 2 A989), 417-453.· [12] M. Aubry, Homotopy Theory and Models. DMV Seminar Band 24, Birkhauser, 1995. [13] L. Avramov, Free Lie subalgebras of the cohomology of local rings. Trans. Amer. math. Soc. 270 A982), 589-608. [14] L. Avramov, Local Algebra and Rational Homotopy. In iiomotopie Algebrique ei Algebre Locale. Asterisque 113-114, Societe Mathematique de France A984), 15-43. [15] L. Avramov, Torsion in loop space homology. Topology 25 A986), 155-157.
522 REFERENCES [16] L. Avramov and S. Halperin, Through the looking glass; a dictionary be- tween rational homotopy theory and local algebra, in Algebra, Algebraic Topology and their Interactions, Stockholm 1983 Lecture Notes in Mathe- matics 1183, Springer Verlag 1986, 1-27. [17] I.K. Babenko and LA. Taimanov, On non formal simply connected sym- plectic manifolds, Preprint series, math.SG: 9811167v3, 17 Dec. 1998. [18] J. Barge, Structures differentiates sur les types d'homotopie rationelle simplement connexes. Ann. Scient. Ec. Norm Sup. 9 A976), 1-33. [19] J.H. Baues, Rational homotopietypen. Manuscripta Math. 20 A977), 119- 131. [20] H. Baues, Algebraic Homotopy. Cambridge University Press, 1989. [21] H. Baues and J.M. Lemaire, Minimal models in homotopy theory. Math. Ann. 225 A977), 219-242. [22] M. Bendersky and S. Gitler, The cohomology of certain function spaces. Trans. Amer. Math. Soc. 326 A991), 423-440. [23] C Benson and C.S. Gordon, Kahler and symplectic structure on nilmani- folds. Topology 27 A988), 513-518. [24] M. Berger and R. Bott, Sur les varietes a courbure strictement positive. Topology 1 A962), 301-311. [25] R. Body and R. Douglas, Rational homotopy and unique factorization. Paci?cJ. of Math. 75 A978), 909-914. [26] R. Body and R. Douglas, Unique factorization of rational homotopy types. Paci?c J. of Math. 90 A980), 21-26. [27] R. Body, M. Mimura, H. Shiga and D. Sullivan, p-universal spaces and rational homotopy types. Comment. Math. Helv. 73 A998), 427-442. [28] R. Bogvad and S. Halperin, On a conjecture of Roos. Lecture Notes in Math 1183 A986), 120-127. [29] R. Bogvad and C. Jacobsson, Graded Lie algebras of depth one. Manuscripta Math. 66 A989), 153-159. [30] A.K. Bousfield and V.K.A.M. Gugenheim, On PL de Rham theory and rational homotopy type. Memoirs of the Amer. Math. Soc. 179 A976) [31] E. Brown, Twisted torsion product, I. Annals of Math. 69 A959), 223-242. [32] E. Cartan, La topologie des groupes de Lie. Actualites Scientifiques et industrielles, 358. Paris Hermann, 1936.
Rational Homotopy Theory 523 [33] H. Cartan, Notions d'algebre differentielle; applications aux groupes de Lie, aux varietes ou opere un groupe de Lie. Colloque de Topologie (Espaces Fibres) Bruxelles A950), Liege et Paris A951), 15-27. [34] H. Cartan, La transgression dans un groupe de Lie et dans un espace fibre principal. Colloque de Topologie (Espaces Fibres) Bruxelles A950), Liege et Paris A951), 57-71. [35] H. Cartan and S. Eilenberg, Homological Algebra. Princeton Universty Press, 1956. [36] H. Chen, Nonexistence of negative weight derivations on graded Artin al- gebras: A conjecture of Halperin. J. Algebra 216 A999), 1-12. [37] K.T. Chen, Iterated integrals of differential forms and loop space homology. Annals of Math. 97 A973), 1033-1035. [38] C. Chevalley and S. Eilenberg, Cohomology theory of Lie groups and Lie algebras. Trans. Amer. Math. Soc. 63 A948), 83-124. [39] L.A. Cordero, M. Fernandez and A. Gray, Symplectic manifolds with no Kahler structure. Topology 25 A986), 375-380. [40] O. Cornea, Cone length and LS category. Topology 33 A994), 95-111. [41] O. Cornea, There is just one cone length. Trans. Amer. Math. Soc. 344 A994), 835-848. [42] P. Deligne, P. Griffiths, J. Morgan and D. Sullivan, Real homotopy theory of Kahler manifolds. Invent. Math. 29 A975), 245-274. [43] G. de Rham, Sur l'analysis situs des varietes a n dimensions. Journal de Mathematiques Pures et Appliquees 10 A931), 115-200. [44] J. Dugundji, TopoJogy. Allyn and Bacon, Boston, 1965. [45] N. Dupont, A couterexample to the Lemaire-Sigrist conjecture, Topology 38 A999), 189-196. [46] N. Dupont, Problems and Conjectures in rational homotopy theory: A survey. Expo. Math. 12 A994) 323-352. [47] W.G. Dwyer, Strong convergence of the Eilenberg-Moore spectral se- quence. Topology 13 A974), 255-265. [48] S. Eilenberg and S. MacLane, On the group ?(?,?) I. Annals of Math. 58 A953), 55-106. [49] S. Eilenberg and J.C. Moore, Homology and fibrations I : cotensor product and its derived fonctors. Comment. Math. Helvetia 40 A986^ 199-236.
524 REFERENCES [50] Y. Felix, Caracteristique d'Euler homotopique. Note aux Comptes-Rendus de l'Academie des Sciences de Paris 291 A980), 303-306. [51] Y. Felix, La Dichotomie Elliptique-Hyperbolique en Homotopie Ra- tionnelle. Asterisque 176, Soc. Math. France, 1989. [52] Y. Felix et S. Halperin, Rational LS category and its applications. Trans. Amer. Math. Soc. 273 A982), 1-38. [53] Y. Felix and S. Halperin, Formal spaces with finite dimensional rational homotopy. Trans. Amer. Math. Soc. 270 A982), 575-588. [54] Y. Felix, S. Halperin, C. Jacobsson, C. Lofwall and J.-C. Thomas, The radical of the homotopy Lie algebra. Amer. J. of Math. 110 A988), 301- 322. [55] Y. Felix, S. Halperin and J.M. Lemaire, The rational L.S. category of products and of Poincare duality complexes. Topology 37 A998), 749-756. [56] Y. Felix, S. Halperin, J.-M. Lemaire and J.-C. Thomas, Mod ? loop space homology. Inventiones Math. 95 A989), 247-262. [57] Y. Felix, S. Halperin and J.-C. Thomas, L.S. categorie et suite spectrale de Milnor-Moore (une nuit dans le train). Bull. Soc. Math. France 111 A983), 89-96. [58] Y. Felix, S. Halperin and J.-C. Thomas, Sur l'homotopie des espaces de categorie 2. Math. Scand. 55 A984), 216-228. [59] Y. Felix, S. Halperin and J.-C. Thomas, Loop space homology of spaces of LS category one and two. Math. AnnaTen 287 A990), 377-386. [60] Y. Felix and J.-C. Thomas, Sur la structure des espaces de L.S. categorie deux. Illinois J. Math. 30 A986), 175-190. [61] J. Friedlander and S. Halperin, An arithmetic characterization of the ra- tional homotopy groups of certain spaces. Inventiones Math. 53 A979), 117-133. [62] T. Ganea, Lusternik-Schnirelmann category and strong category. Illinois J. Math. 11 A967), 417-427. [63] T. Ganea, Induced fibrations and cofibrations. Trans. Amer. Math. Soc. 127 A967), 442-459. [64] M. Ginsburg, On the Lusternik-Schnirelmann category. Annab of Math. 77 A963), 538-551. [65] A. Gomez Tato, Modeles minimaux resolubles. Journal of Pure and Applied Algebra 85 A993), 43-56.
Rational Homotopy Theory 525 [66] A. Gomez-Tato, S. Halperin and D. Tanre, Non simply connected homo- topy theory. Preprint . [67] D. Gottlieb, Evaluation subgroups of homotopy groups. Amer, j ot Math 91 A969), 729-756. [68] M. Greenberg, Lectures on Algebraic Topology. W.A. Benjamin A966) [69] W. Greub, S. Halperin and R. Vanstone, Connections, Curvature and Co- ijomoJogy I. Academic Press A972). [70] W. Greub, S. Halperin and R. Vanstone, Connections, Curvature and Co- homoiogy I. Academic Press A973). [71] W. Greub, S. Halperin and R. Vanstone, Connections, Curvature and Co- homology M. Academic Press A976). [72] P.A. Griffiths and J.W. Morgan, Rational Homotopy Theory and Differ- ential Forms, Progress in Mathematics, 16, Birkhauser, 1981. [73] D. Gromoll and W. Meyer, Periodic geodesies on a compact riemannian manifold. J. Differentia] Geometry 3 A969), 493-510. [74] M. Gromov, Homotopical effects of dilatations. J. Differentia] Geometry 13 A978), 303-310. [75] M. Gromov, Curvature, Diameter and Betti numbers. Comment. Math. Helvetia 56 A981), 179-195. [76] K. Grove and S. Halperin, Contributions of rational homotopy theory to global problems in geometry. PufaJ. I.H.E.S. 56 A982), 379-385. [77] K. Grove and S. Halperin, Dupin hypersurfaces, group actions and the double mapping cylinder. J. Differential Geometry 26 A987), 429-459. [78] A. Haefliger, Rational homotopy of the space of sections of a nilpotent bundle. Trans. Amer. Math. Soc. 273 A982), 609-620. [79] R. Hain, Jnterated integrais and Jiomotopy Periods. Memoirs of the Amer. Math. Soc. 291 A984). [80] R. Hain, The geometry of the mixed Hodge structure on the fundamental group. In Algebraic Geometry, Bowdoin, PSPM, 46 A987), 247-282. [81] R. Hain, The de Rham homotopy theory of complex algebraic varieties. K-theory 1 A987), 271-324 and 481-497. [82] S. Halperin, Lectures on Minimal models. Me'moires de la societe Mathematiques de France 230 A983).
526 REFERENCES [83] S. Halperin, Rational homotopy and torus actions. In Aspects of topology, Mem. H. Dowker, London Math. Soc. Lect. Note Ser. 93 A985), 293-306. [84] S. Halperin and J.-M. Lemaire, Suites inertes dans les algebres de Lie graduees. Math. Scand. 61 A987), 39-67. [85] S. Halperin and J.-M. Lemaire, Notions of category in differential algebra. Lecture Notes in Mathematics, 1318, A988), 138-154. [86] S. Halperin and J.D. Stasheff, Obstructions to homotopy equivalences. Adv. in Math. 32 A979), 233-279. [87] S. Halperin et M. A^igue-Poirrier, The homology of a free loop space. Pacific J. of Math. 147 A991), 311-324. [88] K. Hasegawa, Minimal models of nilmanifolds. Proc. Amer. Math. Soc. 106 A989), 65-71. [89] M. Hernandez-Andrade, A class of compact manifolds with positive Ricci curvature. Proc. Symp. Pure Math. 27 A975), 73-87. [90] K. Hess, A proof of Ganea conjecture for rational spaces. Topology 30 A991), 205-214. [91] K. Hess, A history of rational homotopy theory. In History of Topology, Ed. I.M. James, Elsevier Sciences, 1999. [92] G. Hochschild and J.P. Serre, Cohomology of Lie algebras. Annals of Math. 57 A953), 591-603. [93] W. Hurewicz, Beitrage zur Topologie der Deformationen I-I W. Proc. Koninkl. Akad. Wetenschap Amsterdam. 38 A935), 111-119, 521-528, and 39 A936), 117-126, 215-224. [94] K. Iwasawa, On some types of topological groups. Ann. of Math. 50 A949), 507-748. [95] N. Iwase, Ganea's conjecture on Lusternik-Schnirelmann category. Bull. London Math. Soc. 30 A998), 623-634. [96] I. James, On category in the sense of Lusternik-Schnirelmann. Topology 17 A978), 331-348. [97] I. James, On the suspension triad of a sphere. Ann. of Math. 63 A956), 407-429. [98] B. Jessup, Rational L.-S. category and a conjecture of Ganea. Journal of Pure and Applied Algebra 65 A990), 57-67. [99] B. Jessup, Holonomy nilpotent fibrations and rational Lusternik- Schnirelmann category. Topology 34 A995), 759-770.
Rational Homotopy Theory 527 [100] ?. Jessup and A. Murillo-Mas, Approximating rational spaces with elliptic complexes and conjecture of Anick. Pacific J. Math. 181 A997), 269-280. [101] R. Kaenders, On De Rhasn Homotopy Theory for Plane Algebraic Curves and their Singularities. Thesis, Nijmegen, 1997. [102] J. L. Koszul, Sur un type d'algebre differentielle en rapport avec la trans- gression. Colloque de Topologie (Espaces Fibres) Bruxelles 1950, Liege et Paris A951), 73-81. [103] I. Kriz, On the rational homotopy type of configuration spaces. Annals of Mathematics 139 A994), 227-237. [104] P. Lambrechts. Analytic properties of Poincare series of spaces. Topology 37 A998), 1363-1370. [105] P. Lambrechts, Croissance des nombres de Betti des espaces de lacets, These Louvain-La-Neuve A995) [106] L. Lechuga and A. Murillo, Complexity in rational homotopy. Topology 39 A999), 89-94. [107] D. Lehmann, Theorie homotopique des formes differentielles. Asterisque 45, Soc. Math. France, 1977. [108] J.-M. Lemaire, A ?nite complex whose rational homotopy is not finitely generated. Lecture Notes in Mathematics 196 A971), 114-120. [109] J.-M. Lemaire, Algebres connexes et homologie des espaces de lacets. Lec- ture Notes in Math. 422, Springer-Verlag, 1974. [110] J.-M. Lemaire, "Autopsie d'une meurtre" dans l'homologie d'une algebre de chaines. Ann. Scient. Ec. Norm. Sup. 11 A978), 93-100. [Ill] J.M. Lemaire and F. Sigrist, Sur les invariants d'homotopie rationnelle lies a la L.S. categorie. Comment. Math. Helvetia 56 A981), 103-122. [112] C. Lofwall and J.E. Roos. Cohomologie des algebres de Lie graduees et series de Poincare-Betti non rationelles. Note aux Comptes-Rendus Acad. Sei. Paris 29?'A98?), 733-736. [113] G. Lupton, Note on a conjecture of Stephen Halperin's. Springer Lecture Notes in Math. 1440 A990), 148-163. [114] G. Lupton and J.F. Oprea, Symplectic manifolds and formality. Journal Pure and Applied Algebra. 91 A994) 193-207. [115] G. Lupton and J. Oprea, Cohomologically symplectic spaces: toral actions and the Gottlieb group. Trans. Anm. Math. Soc. 347 A005), 961-288.
528 REFERENCES [116] L. Lusternik and L. Schnirelmann, Methodes topologiques dans les problemes variationeis. Actualites scientifiques et industrielles, 188, Paris, Hermann, 1934. [117] C. McGibbon and C. Wilkerson, Loop space of finite complexes at large primes. Proc. Amer. Math. Soc. 96 A986), 698-702. [118] S. MacLane, Homology, Springer-Verlag, Berlin and New York, 1975. [119] M. Majewski, A proof of the Baues-Lemaire conjecture in rational homo- topy theory. Rend. Circ. Mat. Palermo. Suppl. 30 A993), 113-123. [120] M. Markl, Towards one conjecture on collapsing of Serre spectral sequence, in Geometry and physics, Proc. 9th Winter Sch., Srni/Czech. 1989, Suppl. Rend. Circ. Mat. Palermo, 22 A990), 151-159. [121] C.R.F. Maunder, Algebraic Topology. Van Nostrand Reinhold Company, London,1970. [122] J.P. May, Simplicial object in Algebraic Topology. University of Chicago Press, 1967. [123] W. Meier, Rational universal fibrations and flag manifolds. Math. Ann. 258 A982), 329-340. [124] S.A. Merkulov, Formality of canonical symplectic complexes and Frobenius manifold. Preprint series, math. SG:9805072, 15 May 1998. [125] J. Milnor, Construction of the universal bundles, I and I. Annals of Math. 63 A956), 272-284 and 430-436. [126] J. Milnor, The geometric realization of a semisimplicial complex. Annals of Math. 65, A957) 357-352. [127] J. Milnor and J.C. Moore, On the struture of Hopf algebras. Annals of Math. 81 A965), 211-264. [128] J. Milnor and J.D. Stasheff, Characteristic classes. Princeton Univ. Press, 1974. [129] J. Neisendorfer and P. Selick, Some examples of spaces with or without exponents, in Current trends in Algebraic Topology, CMS Conf. Proc. 2 A982), 343-357. [130] K. Nomizu, On the cohomology of compact homogeneous spaces of nilpo- tent Lie groups. Annals of Math. 59 A954), 531-538. [131] J. Oprea, Gottlieb Groups, Group Actions, Fixed Points and Rational Homotopy. Lecture Notes series 29, Research Institute of Math. Seoul Na- tional University A995).
Rational Homotopy Theory 52g [132] S. Papadima et L. Paunescu, Reduced weighted complete intersection and derivations. Journal of Algebra 183 A996), 595-604. [133] G. Paternain, On the topology of manifolds with completely integrable geodesic flows. Ergodic Theory Dyn. Syst. 12 A992), 109-122. [134] D. Quillen, Homotopical algebra. Lecture Notes in Math. 43, Springer- Verlag, 1967. [135] D. Quillen, Rational homotopy theory. Annals of Math. 90 A969), 205- 295. [136] P. Salvatore, Rational homotopy nilpotency of self-equivalences. Topoiogy Appl, 77 A997), 37-50. [137] H. Scheerer and D. Tanre, Fibrations a la Ganea. Bull. Belg. Math. Soc. 4 A997), 333-353. [138] P. Selick, Moore conjectures. In Algebraic Topology - Rational Homotopy, Lecture Notes in Math. 1318 A988), 219-227. [139] J.P. Serre, Lie algebras and Lie Groups. New-York, Bejamin, 1965. [140] J.P. Serre, Algebre Locale - Multipliates. Lecture Notes in Math. 11, Springer-Verlag, 1965. [141] H. Shiga and M. Tezuka, Rational fibrations, Homogeneous spaces with positive Euler characteristics and Jacobians. Ann. Inst. Fourier 37 A987), 81-106. [142] E.H. Spanier, Algebraic Topology. McGraw Hill, Academic Press, 1966. [143] D. Sullivan, Geometric Topology- 1 : Localization, Periodicity, and Galois Symmetry. MIT Press, 1970. [144] D. Sullivan, Infinitesimal computations in topology. Publ. I.H.E.S. 47 A977) 269-331. [145] D. Tanre, Homotopie Rationnelle : Modeles de Chen, Quillen, Sullivan. Lecture Notes in Mathematics 1025, Springer-Verlag, 1983. [146] J.C. Thomas, Rational homotopy of Serre fibrations. Ann. Inst. Fourier 331 A981), 71-90. [147] J.C. Thomas, Homologie de l'espace des lacets, problemes et questions. Journal of Pure and Applied Algebra 91 A994), 355-379. [148] H. Toda, On the double suspension E2. J. Inst. Polytech. Osaka City Univ. A7 A956), 103-145.
530 REFERENCES [149] G. Toomer, Lusternik-Schnirelmann category and the Milnor-Moore spec- tral sequence. Math. Zeitschrift 138 A974), 123-143. [150] ?. Totaro, Configuration spaces of algebraic varieties. Topology 35 A996), 1057-1067. [151] G. Triantafillou, Equivariant Rational Homotopy Theory. In Equivariant homotopy and cohomology theory, C.B.R.M. 91 A996), 27-32. [152] M. Vigue-Poirrier et D. Sullivan, The homology theory of the closed geodesic problem. Journal of Differential Geometry 11 A976), 633-644. [153] M. Vigue-Poirrier, Homotopie rationnelle et nombre de geodesiques fermees. Ann. Scient. Ecole Norm. Sup. 17 A984), 413-431. [154] J.M. Wahl, Derivations, automorphisms and deformations of quasi- homogeneous singularities. In Proc. Symp. Pure Math. 40 A983), 613-624. [155] C. Watkiss, Cohomology of principal bundles in semisimplicial theory. The- sis, Toronto A974). [156] A. Weil, Sur les theoremes de de Rham. Comment. Math. Helv. 26 A952), 119-145. [157] G. Whitehead, Elements of homotopy theory. Graduate Text in Math. 61 Sprinberg-Verlag, New York Inc., 1978. [158] G. Whitehead, The homology suspension. Colloque de Topologie Algebrique, Lou\'ain A957), 89-95. [159] J.C.H. Whitehead, On adding relations to homotopy groups. Ann. of Math. 42 A941), 409-428. [160] J.C.H. Whitehead, Combinatorial homotopy I and I. Bull. Amer. Math. Soc 55 A949), 213-245 and 453-496. [161] O. Zariski and P. Samuel, Commutative Algebra. Graduate Texts in Math. 28, Springer-Verlag, 1975.
Index Alexander-Whitney map 53, 477 Bar construction 268, 277 of an enveloping algebra 305 Borel construction 36 Minimal Sullivan model for Bq 217 Sullivan model for the Borel con- struction 219 Bundle Associated fibre bundle 36 Fibre bundle 32 Milnor's universal G-bundle 36 Principal G-bundle 35 CW complex 4 CW ^-complex 102 Cellular approximation theorem 9 Cellular chain model 60 Cellular chain model for a fibra- tion 90 Cellular chain model of a space 59 Cellular map 5 Cellular models theorem 12 Product of 7 Quotient of 7 Relative CW complex 5 Subcomplex 5 Wedge of CW complexes 7 Whitehead lifting lemma 12 Cartan-Eilenberg-Che\ralley construc- tion on a dgl (see chain coalgebra on a dgl) 301 Cartan-Serre Theorem 231 Cell attachment 173 Sullivan model 173 Cellular approximation theorem 9 Cellular chain model 59, 60 Chain algebra 46 Chain algebra on a loop space 343 Chain coalgebra Chain coalgebra on a dgl 301 Chain coalgebra on a dgl, C. (L) 301 Chain complex 43 C,(L;UL) 302,305 Cellular chain complex 59 Chain complex of .4-modules 273 Chain complex of t/X-modules 464 Chain complex on a dgl 305 Normalized singular chain com- plex 52 of finite length 475 Cobar construction 271 Cochain algebra 46 C*{X;k) 65 Cochain algebra APL(X) 121 Cochain algebra on a dgl 313, 314 Cochain complex 43 Normalized singular cochain com- plex 65 Cofibration 15 Cohomology De Rham Theorem 128 Lie algebra cohomology 320 Relative singular cohomology 66 Singular cohomology 65 Commutathre model 116 Complex projective space 6 Lie model 333 Cone length 360, 362 Rational cone length 370, 372, 374,382,388 Cup product 65 Cup-length 366 Rational cup length 370, 387 De Rham Theorem 128 Depth of a Lie algebra 492, 493 Depth of an algebra 474, 486 Differential graded algebra 46
532 INDEX dga Homotopy 344 Direct product 47 Fibre product 47 Tensor product 47 Differential graded coalgebra 48 Differential graded Hopf algebra 297 Differential graded Lie algebra 296 Module 296 Differential model on a dga Semifree model 183 Differential module on a dga 46 Equivalence 68 Homotopy of maps 68 Lifting property 70 Semifree module 69, 262 Dupont example 388 Eilenberg-MacLane space 62 Simplicial construction 246 Sullivan model 201 Euler-Poincare characteristic 49, 444 Existence of models 182 Exponents 441 Ext 275, 465 Fibration 23 G-(Serre) fibration 28 Lifting property 23, 24 Path space fibration 29 Serre fibration 23, 33 Fibre of the model 196 Formal dimension 441 Ganea spaces 357, 366 Ganea fibres 357, 479 Ganea's fibre-cofibre construction 355 Ginsburg's Theorem 399 Global dimension 474 Gottlieb groups 377, 378, 392 Grade 474 Projective grade 474, 476, 481 Graded algebra 43 Commutative 45 Free commutative 45 Tensor algebra 45 Tensor product 44 Graded coalgebra 47, 299 Coderivation 301 Primitively cogenerated 299 Graded Hopf algebra 288 Primitwe element 288 Graded Lie algebra 283 Abelian Lie algebra 283, 285 Adjoint representation 284, 422 Derivation 284 Derived sub Lie algebra 283 Free graded Lie algebra 289, 316 Lie algebra cohomology 320 Lie bracket 283 Module 283 Morphism 283 Product 284 Representation 284 Universal em'eloping algebra 285 Graded module 40 Suspension 41 Tensor product 41 Graphs (ra-colourable) 440 Hess' Theorem 393 Hubert series 49 Holonomy fibration 30, 342, 378 Holonomy representation for a fibration 415, 418 for a relative Sullh'an algebra 416, 418 Homogeneous space 218, 435 Sullivan model 219 Homology Singular homology 52 Homotopy Based homotopy 3 Homotopy rel 3 Homotopy equivalence 3, 242 Weak homotopy equivalence 12, 14, 58 Homotopy fibre 30, 196, 334 Homotopy groups 10, 208, 452 Cartan-Serre Theorem 231 Long exact sequence 26, 213 Rational homotopy groups 453 Homotopy Lie algebra of a space 292, 294, 295, 325, 432
Rational Homotopy Theory 533 of a Sullivan algebra 295, 317, 475 of a suspension 326 Hurewicz homomorphism 58, 210, 231, 234, 293, 326 Iwase example 407 Jessup's Theorem 424 Kahler manifold 162 Koszul-Poincare series 444 Lie group 216 Matrix Lie group 219 Right invariant forms on a Lie group 161 Sullivan model for a Lie group 161 Sullivan model for the orbit space X/G 219 Lie model for a space 322 Cellular Lie model 323, 331 Free Lie model 322, 324, 383 Lie model for a wedge 331 Lie model of a wedge of spheres 331 Lie model of the product 332 Lie models for adjunction spaces 328 Lie model of a dgl 309 Free Lie extension 310 Minimal Lie model 311 Minimal Lie model for a Sulli- van algebra 316 Localization 108 TMocal group 102 "P-local space 102 CW-p-complex 102 Cellular localization 108 Relative CW-p complex 104 Loop space 29, 340 ??-spaces 476 Chain algebra on a loop space 88, 343, 347 Decomposability 449 LS category 380 Rational product decomposition 228, 229 Loop space homology Basis 230 Primitive subspace 230 Loop space homology algebra Milnor Moore Theorem 293 Loop space homology Ht{UX;k) Depth 486 Loop space homology algebra H* (QX 224, 232, 458 Lusternik-Schnirelmann category (of algebras and modules) of a differential module 401 of a Sullivan algebra 381, 384, 488 Lusternik-Schnirelmann category (of spaces) 351, 362, 406, 486 of a continuous map 352, 481 of free loop spaces 376 of loop spaces 380 of Postnikov fibres 376 Rational LS category 386, 387, 388 Rational LS category of a space 370, 371, 374, 432 Rational LS category of smooth manifolds 396 Whitehead LS category 353 Mapping Theorem 375, 389, 411 Milnor-Moore Theorem 293 Nilmanifolds 162 Poincare-Birkoff-Witt Theorem 286, 287 Polynomial differential forms 122 Product length, nil 381, 382 Projective dimension 278, 474, 481 Quasi-isomorphisms 152 Quillen construction C(C,d) 306, 307 Quillen plus construction 212 Radius of com'ergence 459 Rational elliptic space 434 Rational hyperbolic space 452 Rational space 102 Rational cellular model 111 Rational homotopy equivalence mOQQ
534 INDEX Rational homotopy type 111 Rationalization 108, 386 Regular sequence 437 Resolution i/,(fiy)-free resolution of Ht(F) 483 Eilenberg-Moore resolution 279, 480 Projective (free) resolution 274 Semifree resolution 69, 71, 278 The Milnor resolution of Ic 478. 480 Simplex Degeneracy map 52 Degenerate 52 Face map 52 Linear simplex 52 Singular n-simplex 52 Simplicial morphism 117 Simplicial object 116 Extendable simplicial object 118, 240 Milnor realization 238 Simplicial cochain algebra 117 Simplicial set 117 Spatial realization 248, 386 Homotopy groups 250 of a relath'e Sullh'an model 248 Spectral sequence 260, 261 Comparison Theorem 265 Eilenberg-Moore spectral sequence 280 Filtered module 261, 266 Hochschild-Serre spectral sequence 464 Milnor-Moore spectral sequence 317, 318, 319, 399 Odd spectral sequence 438 Sphere 2, 7, 11 :P-local n-sphere 102 Fundamental class 58 Homology algebra H.(Q.Sn;k) 225, 234 Homology Lie algebra of a wedge of spheres 326 Lie model of a sphere 322 Lie model of a wedge of spheres 331 Loop space homology of a fat wedge 462 Loop space homology of a wedge of spheres 460 Model of the loop space Q.Sk 200 Rational homotopy groups of spheres 210 Sullivan model for a sphere 142 Whitehead product 178 Spherical fibration 202 Sullivan algebra 138 Pure Sullh'an algebra 435 Relative Sullivan algebra 181, 262 Sullh'an model for a commutath'e cochain algebra 138, 141 Formality 156 Homotopy of maps 139, 148.150, 153 Lifting lemma 148, 184, 186 Linear part of a morphism 152, 295 Linear part of the differential 141 Minimal Sullivan model 154,186 Sullivan model for a map Fibre 181, 192 Model of a pullback fibration 204 Model of fibrations 195 Model of spherical fibrations 202 Model of the path space fibra- tion 226 Sullivan model for a space 138, 141 Contractible Sullivan model 139 Minimal Sullivan model 138, 146, ' 154 Model for a cell attachment 173 Model for a product 142 Model for a suspension 171 Model for a wedge 155 Model for an if-space 143 Models for smooth manifolds 213
Rational Homotopy Theory 535 Quadratic part of the differen- tial 175, 176 The homotopy groups 171, 208 Sullivan realization functor 247 Sullivan representative for a map 154 Symmetric spaces 162 Symplectic manifolds 163 Tensor algebra 268 Tensor coalgebra 268 The set of homotopy classes 151 Toomer's invariant 366, 367 of a Sullivan algebra 381, 386, 391, 400 of a differential module 401 Rational Toomer invariant 370, 374, 386, 387, 432 Topological group 35 Classifying space 36 Orbit space for a group action 35 Topological monoid 28 Action of 28 Chain algebra on 88 Topological space Ji-space 143 fc-space 1 ra-cone 359 Adjunction space 3, 8, 328 Cell attachment 4 Cofibre 352 Conformai space 334, 388 Compact-open topology 1 CW complex 4 Filtered space 5 Formal space 156, 162, 163, 388 Homotopy fibre 30, 78 Join 21 Mapping cylinders 8 Mapping spaces 1 Moore path space 29 NDR-pair 15, 20 Product of 1, 7 Quotient of 1, 7 Reduced cone 352 Retract 3,15 Smash product 21 Suspension 3, 7, 20, 359 Wedge of spaces 2, 7, 462 Well-based space 15 The construction A(K) 118 The simplicial cochain algebra 124 Tor 275, 465 Differential Tor 281 Torus action 435 Uniqueness of minimal models 152 Universal enveloping algebra 285,486 of a differential graded Lie alge- bra 296 Poincare-Birkoff-Witt Theorem 286, 287 Weak homotopy type 12, 14 Whitehead lifting lemma 12 Whitehead product 175, 176, 293 Whitehead-Serre Theorem 94 Wordlength 140 in a tensor algebra 290 in Sullivan model 158, 262, 318 232,