On the equivariant cohomology of

homogeneous spaces
Jeffrey D. Carlson
This book is dedicated to my parents and brother,

who have continued to believe in me

through it all.
Acknowledgments

This monograph originated as my dissertation at Tufts University. The first person who deserves
credit for making this possible is my advisor, Loring W. Tu, whose friendship and supervision
made all this possible.
I am grateful to many mathematicians for corresponding with me, contributing ideas, cor-
rections, moral support, professional advice, citations, and even LATEXadvice. All of the follow-
ing contributed, in ways great or small to this effort: Moon Duchin, Oliver Goertsches, Yael
Karshon, Julianna Tymoczko, Jason DeVito, Larry Smith, Joel Wolf, Paul Baum, George Leger,
Marek Mitros, Jay Taylor, Chris O’Donnell, Chi-Kwong Fok, Mathew Wolak, Michael Atiyah,
Steve Halperin, Zach Himes, George McNinch, Fulton Gonzalez, Svjetlana Terzić, Christopher
Allday, Andreas Arvanitoyeorgos, Jun-Hou Fung, and an anonymous editor. My mathematical
formation was due in part to the kind interventions of various mentors throughout the years be-
fore Tufts. Among these I count Bob Cornell and Ed Siegfried from Milton Academy and James
Cummings from CMU.
This dissertation would also never have been completed without the many people from Tufts
who made my time in grad school successful. These include Genevieve Walsh, Christoph Börgers,
Lenore Feigenbaum, Boris Hasselblatt, Kim Ruane, James Adler, Ellen Goldstein, Aaron Brown,
Erin Munro, Emily Stark, Yulan Qing, Erica Waite, Andy Eisenberg, Charlie Cunningham, Burns
Healy, Brendan Foley, Lise Chlebak, Andrew Sánchez, Kasia Sierzputowska, Jürgen Frikel, and
Nathan Scheinmann.
Finally, people I met in real life who also kept me going include Lisa Zhang (张艺), Nancy
Leung, Huda Shalhoub, Kathleen Reilly, Jef Guarante, Claire Weigand, Jason Walen, and Cristina
Izuno. I’ve already dedicated this work to my immediate family, Deborah (Z) Kasik, Dan Carlson,
and Drew Carlson. I also wanted to thank my grandmother Elisabeth Kasik, who loved me
unconditionally and shaped me in a lot of ways as a person, and my childhood pets Mocha and
Simon, but unfortunately I no longer can. You are missed.

iii
Contents

1 Bundles, actions, and orbits 5

1.1 The category of G-actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 Freedom and efficacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 The category of G-orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Tubes, slices, and G–CW complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Fixed point sets of actions on homogeneous spaces . . . . . . . . . . . . . . . . . . . 10

2 Classifying spaces 13
2.1 The weak contractibility of EG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 A construction of EG for G a compact Lie group . . . . . . . . . . . . . . . . . . . . . 15
2.3 Milnor’s functorial construction of EG . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Cohomology theories 21
3.1 Desiderata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Candidate theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.1 The cohomology of the naive quotient . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.2 Bredon cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.3 Borel equivariant cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 The Atiyah–Hirzebruch–Leray–Serre spectral sequence . . . . . . . . . . . . . . . . . 25
3.3.1 The transgression in the Serre spectral sequence . . . . . . . . . . . . . . . . . 29
3.4 A natural lemma on bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4 The cohomology of complete flag manifolds 34

4.1 The cohomology of a flag manifold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2 The acyclicity of G{NG pTq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

5 First properties of equivariant cohomology 38

5.1 Values on the orbit category and Mayer–Vietoris . . . . . . . . . . . . . . . . . . . . . 38
5.2 The equivariant Künneth theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.3 The standard example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.4 Weyl-invariants and the restricted action a maximal torus . . . . . . . . . . . . . . . 49

6 The cohomology of Lie groups and classifying spaces 52

6.1 The cohomology of the classical groups . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.1.1 Complex and quaternionic unitary groups . . . . . . . . . . . . . . . . . . . . 53
6.1.2 Real difficulties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.2 The rational cohomology of Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . 64

iv
Contents v

6.3 The Serre spectral sequence of S1 Ñ ES1 Ñ BS1 . . . . . . . . . . . . . . . . . . . . . 70

6.4 The Serre spectral sequence of T Ñ ET Ñ BT . . . . . . . . . . . . . . . . . . . . . . 72
6.5 The Koszul complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.6 The Serre spectral sequence of G Ñ EG Ñ BG . . . . . . . . . . . . . . . . . . . . . . 78
6.6.1 Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.6.2 Two proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.6.3 Complements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.7 Characteristic classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.8 Maps of classifying spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.8.1 Maps of classifying spaces of tori . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.8.2 Maps of classifying spaces of compact, connected Lie groups . . . . . . . . . 89

7 The cohomology of homogeneous spaces 92

7.1 The Borel–Cartan machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
7.1.1 The fiber sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
7.1.2 The spectral sequence map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.1.3 Cartan’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
7.2 The structure of the Cartan algebra, I . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
7.3 Cohomology computations, I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.3.1 Cohomology-surjective pairs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.3.2 Pairs of equal rank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.4 The structure of the Cartan algebra, II: formal pairs . . . . . . . . . . . . . . . . . . . 111
7.5 The cohomology of G{S1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

8 Equivariant formality 121

8.1 Equivariant formality of group actions . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.2 The Borel localization theorem and equivariant formality . . . . . . . . . . . . . . . . 122
8.2.1 The localization theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
8.2.2 Equivariant formality of torus actions and fixed points . . . . . . . . . . . . . 125
8.3 Equivariant formality and Koszul duality . . . . . . . . . . . . . . . . . . . . . . . . . 126

9 Equivariant formality of isotropy actions 127

9.1 Reduction of isotropy-formality to a maximal torus . . . . . . . . . . . . . . . . . . . 127
9.2 Isotropic torus actions and fixed points . . . . . . . . . . . . . . . . . . . . . . . . . . 130
9.3 Earlier work on equivariant formal isotropy actions . . . . . . . . . . . . . . . . . . . 131
9.3.1 Isotropy-formality of equal-rank and generalized symmetric pairs . . . . . . 131
9.3.2 The criterion of Shiga and Takahashi . . . . . . . . . . . . . . . . . . . . . . . 132

10 Equivariant formality of isotropic torus actions 135

10.1 Survey of original work on isotropy-formality . . . . . . . . . . . . . . . . . . . . . . 135
10.2 The equivariant cohomology of isotropy-formal pairs . . . . . . . . . . . . . . . . . . 138
10.3 Lifting to the universal compact cover . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
10.4 Reduction to a semisimple group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
10.5 Equivariant formality of isotropic circle actions . . . . . . . . . . . . . . . . . . . . . . 143
10.6 Reflectibility of circles: reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
10.7 Reflected circles in simple groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Contents vi

A Algebraic background 151

A.1 Commutative algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
A.2 Chain complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
A.2.1 Polynomials and numbers associated to a graded module . . . . . . . . . . . 153
A.3 Commutative and differential graded algebras . . . . . . . . . . . . . . . . . . . . . . 154
A.3.1 Commutative graded algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
A.3.2 Bigraded algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
A.3.3 Differential graded algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
A.3.4 The algebraic Künneth theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 159
A.3.5 Poincaré duality algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
A.4 Splittings and formality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
A.5 Filtrations and spectral sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
A.5.1 Filtered differential groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
A.5.2 The spectral sequence of an exact couple . . . . . . . . . . . . . . . . . . . . . 166
A.5.3 The spectral sequence of a filtered differential graded algebra . . . . . . . . . 167
A.5.4 Fundamental results on spectral sequences . . . . . . . . . . . . . . . . . . . . 168
A.5.5 The transgression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

B Topological background 172

B.1 Topological structures of interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
B.1.1 Cell complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
B.1.2 Fiber bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
B.1.3 Principal bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
B.2 Algebraic topology grab bag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
B.3 Covers and transfer isomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
B.4 The structure of compact Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
B.4.1 The maximal torus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
B.4.2 The root and weight lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

C Classical results on the cheap 182

C.1 Equivariant extensions of characteristic classes . . . . . . . . . . . . . . . . . . . . . . 182
C.2 The Berline–Vergne/Atiyah–Bott localization theorem . . . . . . . . . . . . . . . . . 183
C.3 The hammer applied . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

D Borel’s proof of Chevalley’s theorem 186

D.1 Sheaf cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
D.2 The Leray spectral sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
D.3 Borel’s proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

E The original classification of reflected circles in simple groups 194

E.1 Theoretical reflections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
E.1.1 The case all circles are reflected . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
E.1.2 The cases of SUpnq and Spinp4n ` 2q . . . . . . . . . . . . . . . . . . . . . . . . 195
E.1.3 E6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
E.2 The GAP verification of Proposition E.1.7 . . . . . . . . . . . . . . . . . . . . . . . . . 200
Contents vii

Bibliography 203
List of Tables

10.1.6 Reflected lines in simple Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . 136

10.1.11 The classification for circles in Upnq . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

List of Figures

2.3.1 Some low-dimensional joins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.3.5 Nullhomotoping a curve γ in X ˚ Y to lift into A . . . . . . . . . . . . . . . . . . . . 18
5.3.2 A schematic of pS2 qS1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.3.5 Homotoping a representative of π2 X1 into the end-caps t˘1u ˆ CP8 . . . . . . . . 48
6.1.2 The Serre spectral sequence of Upnq Ñ Upn ` 1q Ñ S2n`1 . . . . . . . . . . . . . . . 54
6.1.7 The differential dn´1 in the Serre spectral sequence of Sn´2 Ñ V2 pRn q Ñ Sn´1 . . . 56
6.1.8 The reflection of e1 through vK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6.1.12 The Serre spectral sequence of Sn´j Ñ Vj pRn q Ñ Vj´1 pRn q over F2 . . . . . . . . . . 60
6.3.2 The potentially nonzero region in the Serre spectral sequence of S1 Ñ ES1 Ñ BS1 70
6.3.3 The first few subcomplexes of E2 in the Serre spectral sequence of S1 Ñ ES1 Ñ BS1 71
6.3.4 Isomorphism classes in E2 of the Serre spectral sequence of S1 Ñ ES1 Ñ BS1 . . . 71
6.5.1 The potentially nonzero region in the Serre spectral sequence of S3 Ñ ES3 Ñ BS3 73
7.1.2 The differential of the algebra C “ APL pBq b H ˚ pGq as defined on primitives . . . 96
7.3.3 The Serre spectral sequence of Up4q Ñ Up4qSpp2q Ñ BSpp2q . . . . . . . . . . . . . . 104
A basis for the E8 page of the Serre spectral sequence 3
7.3.16 ` of Up3q˘ Ñ Up3qT3 Ñ BT . 110
7.5.4 The Serre spectral sequence of Spp1q ˆ Up2q Ñ Spp1q ˆ Up2q S Ñ BS . . . . . . . 118
10.7.5 The graph involutions of An . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
10.7.9 The graph involution of D2n`1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
10.7.13 The graph involution of E6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
A.5.10 The differentials to and from E54,3 leave the first quadrant . . . . . . . . . . . . . . . 169
A.5.12 An edge map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

viii
On the equivariant cohomology of homogeneous spaces
 1 Introduction

2 It is now abundantly clear that the spectral sequence is one of the fundamental algebraic
3 structures needed for dealing with topological problems.

4 —William Massey, 1955

5 Now we illustrate the advantages of commutative multiplication in a fibration formula. This

6 is the [ . . . ] analogue of the Chevalley–Hirsch–Koszul formula for principal Lie group bundles
7 which was current in 1950 and ignored later in topology. The evident power and simplicity of
8 the CHK formula helped prompt me to the present theory after Armand Borel kindly explained
9 it to me in 1970.

10 —Dennis Sullivan, 1977 [Sul77]

11 This monograph is broadly about the cohomology of homogeneous spaces. It evolved from
12 a dissertation whose purpose was to explore the satisfiability and consequences of a techni-
13 cal condition on Borel equivariant cohomology called equivariant formality, which originated in
14 Borel’s seminar on transformation groups [BBF+60] and has since the work of Goresky, Kottwitz,
15 and MacPherson [GKM98] also been known to have interesting applications in symplectic and
16 algebraic geometry.
17 Briefly, given a continuous action of a topological group K on a space X, one constructs the
18 Borel equivariant cohomology HK˚ pXq, a graded ring depending on the action as well as the
19 topology of the spaces K and X, and says the action is equivariantly formal if a certain natural ring
20 map HK˚ pXq ÝÑ H ˚ pXq to the singular cohomology of the space X is surjective. In the case G is
21 a Lie group and K a closed subgroup of full rank in G (i.e., K contains a maximal torus of G),
22 one gets much mileage already out of this condition from its applicability to the isotropy action of
23 left multiplication by K on the right quotient G{K, which for example allows one to compute an
24 integral of a K-invariant function as the sum of values of a related function over the finite fixed
25 point set pG{KqK . Thus it is natural to ask when the same isotropy action is equivariantly formal
26 if K is not of full rank.
27 To address this question, we must define and discuss equivariant cohomology and discuss
28 some of its commonplaces, but at this level of generality the subject is already well served by
29 the monograph [GS99] of Guillemin and Sternberg and by the forthcoming text [BTarb] of Bott
30 and Tu. On the other hand, in the case where X is a homogeneous space G{K, the smaller Cartan
31 model for equivariant cohomology can be concisely constructed from algebraic considerations
32 and the differential-geometric background required to do so in full generality can be avoided:
33 the algebras HK˚ pG{Kq and H ˚ pG{Kq can both be computed from the Serre spectral sequence one
34 encounters in a second course in algebraic topology (for instance the classic book [BT82] of Bott

2
 List of Figures 3

35 and Tu). In particular, in a broad swath of cases, whether the ring map HK˚ pG{Kq ÝÑ H ˚ pG{Kq in
36 question is surjective can be determined entirely in terms of the latter ring and the finite group
37 π0 NG pKq of symmetries k ÞÝÑ gkg´1 of K induced by conjugation in the larger group. Thus in
38 this case the main tool in attacking the question of equivariant formality is really the cohomology
39 of a homogeneous space, with some additional input from classical invariant theory.
40 Given the basic importance of homogeneous spaces in smooth geometry—they are precisely
41 orbits of a transitive action—, the importance of cohomology as an algebro-topological invariant,
42 and that computing the rings H ˚ pG{Kq was an active research topic from the late 1930s to the
43 early 1970s, one might expect the full story on them to be well understood and easy to read
44 about; but the author found this not to be the case. It became therefore a major goal to present the
45 machinery for these computations simply and efficiently, and this presentation is in the author’s
46 view one of the major justifications for the publication of this work. To explain our insistence on
47 presenting yet another version of this old if insufficiently publicized story, some discussion of
48 existing expositions is in order.
49 The primary literature predominantly dates to a movement clustered around Henri Cartan
50 from 1949–53, is presented rather telegraphically, is littered with references to results whose
51 proofs were never published, and is reliant on an early version of sheaf theory which is now vir-
52 tually forgotten if not actively suppressed. There are long surveys by André and by Rashevskii
53 [And62; Ras69] summarizing the results of this school and filling in the details, but the main sec-
54 ondary sources in English are the books of Onishchik and of Greub–Halperin– Vanstone [Oni94;
55 GHV76].
56 Onishchik is relatively concise at under 300 pages, and surprisingly difficult to lay hands
57 on. With a view toward classifying pairs pG, Kq of compact, connected Lie groups with respect
58 to the diffeomorphism, homeomorphism, or homotopy type of G{K, it develops Lie theory, a
59 real version of Sullivan’s rational homotopy theory, the Weil algebra, the theory of symmetric
60 invariants, and the Cartan model. The Weil algebra appears ex nihilo, without reference to the
61 connection and curvature forms associated to a principal bundle which were Weil’s motivation.
62 It is a notable feature of this work that through clever use of filtration arguments, it manages to
63 avoid invoking spectral sequences altogether.
64 The book of Greub, Halperin, and Vanstone, on the other hand, comprises nearly 600 pages.
65 It develops the necessary background in great generality, arriving at target results on the coho-
66 mology of homogeneous spaces in Ch. XI, on page 457. The development is unfortunately in an
67 earlier language and notation than that now current (see, e.g., P-spaces and P-algebras in the early
68 chapters, the notational distinction between P and IP, and the intuitively pleasing but ill-fated
Ž
69 use of S for a symmetric algebra), which is now better developed in the language of rational
70 homotopy theory for which the later work [FHT01] of Halperin has become a standard reference.
71 The generality of the formulations results in an ouroboric presentation style where the desired
72 results at the end are notational permutations of results obtained toward the beginning.
73 By way of contrast, in this monograph the goal is to arrive along a quasigeodesic path at
74 Chevalley’s and Cartan’s respective theorems on the cohomology of principal bundles G Ñ P Ñ
75 B and homogeneous spaces G{K. The Serre spectral sequence is summarized, but assumed, Lie
76 theory is developed to the minimum necessary extent, and the results are seen to follow for es-
77 sentially algebraic reasons from the presence of the multiplication on a Lie group. Some language
78 of rational homotopy theory is used, but the theory itself lurks silently in the background, only
79 making an explicit appearance is in the introduction of the algebra of polynomial differential
 List of Figures 4

80 forms, which allows one to circumvent

81 (i) the assumption of a manifold structure on B,

82 (ii) the bulk of Lie theory, and

83 (iii) the development of sheaf theory.

84 This general calculation gives an algebraic expression for H ˚ pG{Kq as either the cohomology of a
85 certain finitely-generated
` differential
˘ graded algebra structure on H ˚ pBKq b H ˚ pGq or as the Tor
˚,˚
86 functor Tor H˚ pBGq Q, H ˚ pBKq . These versions of the computation are easily interchangeable and
87 arise from the fact that the Serre spectral sequence of G Ñ G{K Ñ BK and the Eilenberg–Moore
88 spectral sequence of G{K Ñ BK Ñ BG contain essentially the same information. The proofs
89 presented of the main results are based spiritually on the originals.
90 To explicitly compute the cohomology ring of a given specific homogeneous space G{K, one
91 needs to understand a map of polynomial rings ρ : H ˚ pBGq ÝÑ H ˚ pBKq. Useful expressions for
92 this map arise almost exclusively from the Chern–Weil and Chevalley restriction isomorphisms
93 identifying the real vector space extension ρ bQ R with the map of invariant polynomial rings
94 RrtG sWG ÝÑ RrtK sWK , where tG is the Lie algebra of a maximal torus of G and WG the Weyl
95 group of G. The problem is then converted into one of understanding invariant polynomials
96 on a vector space and how they restrict on a subspace. Threaded through the abstract account,
97 we develop enough of this theory to enable us to give completely concrete calculations for the
98 examples when G and K are classical groups. This development having been completed, in the
99 last chapters we return to the question which was the subject of the author’s thesis, that of when
100 the map HK˚ pG{Kq ÝÑ H ˚ pG{Kq is surjective.
101 While the initial motivating question may be a technical one of relatively specialist interest,
102 the cohomology of a homogeneous space should not be. This exposition presents a direct, his-
103 torically honest account, demonstrates the essential simplicity of the determination of H ˚ pG{Kq,
104 and offers motivation for the study of rational homotopy theory (which can reobtain some of
105 these results as corollaries from standard theorems) without building up the entire edifice of this
106 general theory, already well-developed in Felix et al. [FHT01]. We require throughout only basic
107 commutative algebra, Lie theory, and algebraic topology, much less than that contained in the
108 respective books of Atiyah–Macdonald, Bröcker–tom Dieck, and Hatcher [AM69; BtD85; Hat02],
109 and all of which is summarized in the appendices, in order that the whole be legible to a second-
110 year graduate student. The author believes the result to be the most accessible available account
111 of some essential material which should be better known.
112 Chapter 1
113 Bundles, actions, and orbits

114 The main body of this work will be in understanding functors from group actions to abelian
115 groups, so in this chapter we recount some material on continuous group actions that, though
116 standard, is not covered in a typical graduate algebra, topology, or Lie theory course and so
117 seemed to deserve formal demarcation from the algebraic and topological background relegated
118 to the appendices.

119 The category of G-actions

120 Let G be a topological group. The category G-Top of G-actions is specified as follows:

121 • Objects are pairs pX, Aq P Top of topological spaces, such that X is equipped with a contin-
122 uous G-action that restricts to an action on A. One says X (as well as A) is a G-space.

123 • Morphisms are G-equivariant maps (briefly G-maps): continuous maps of pairs f : pX, Aq ÝÑ
124 pY, Bq such that f pgxq “ g f pxq for all g P G and x P X.

125 The objects are really actions, but we will abusively refer to them as spaces whenever convenient.
126 As with Top, we will consider the class of individual G-spaces and G-maps between such as a
127 full subcategory via the canonical inclusion X ÞÝÑ pX, ∅q.
128 A right G-space and a left are essentially the same: the anti-isomorphism g ÞÑ g´1 : G Ñ G
129 converts a right action to a left action (and vice versa) via gx :“ xg´1 . We will only worry about
130 the difference in the event a space admits both a left and a right action.
131 There are two canonical functors to Top, the forgetful functor that takes an action G ñ pX, Aq
132 and returns the pair pX, Aq P Top, disregarding the action, and the orbit-space functor pX, Aq ÞÑ
133 pX{G, A{Gq.
134 The product and coproduct in G-Top have the topological product and disjoint union as
135 underlying spaces. The action on the coproduct is the union of the actions; the intended action
136 on the product X ˆ Y is the diagonal action given by g ¨ px, yq :“ pgx, gyq, so that the projections
137 X Ð X ˆ Y Ñ Y are also G-maps. The orbit space of the diagonal action is the mixing space

X ˆ Y :“ X ˆ Y pg´1 x, yq „ px, gyq.

L
G

138 If we write the action on X instead as a right action, one has the more aesthetically pleasing
139 identification pxg, yq „ px, gyq.1 The application of the orbit-space functor ´{G to the projection
1 One could say g commutes with comma.

5
 Chapter 1. Bundles, actions, and orbits 6

140 diagram X Ð X ˆ Y Ñ Y yields a mixing diagram

Xo XˆY /Y

  
X{G o XˆY / Y{G.
G

141 We always let G act trivially on the closed unit interval I “ r0, 1s. A G-homotopy between two
142 G-maps f , f 1 : X ÝÑ Y is a G-map F : X ˆ I ÝÑ Y such that Fp´, 0q “ f and Fp´, 1q “ f 1 . This
143 is the proper notion of homotopy within the category G-Top, and has the important feature that
144 application of the orbit-space functor takes a G-homotopy X ˆ I ÝÑ Y of G-maps to a homotopy
pX{Gq ˆ I ÝÑ Y{G. In particular, a pair p f : X Ð 1 1 1
145 ÝÝ
ÝÝÝÑÝ Y : f q of G-maps such that f ˝ f and f ˝ f
146 are G-homotopic to idX and idY (respectively) descend to a homotopy equivalence X{G » Y{G.
147 We end this section with the observation that mixing with the translation action of G on itself
148 changes nothing.

149 Lemma 1.1.1 ([Bre72, Prop. II.2.2, p. 73]). Let Γ be a topological group and X a right Γ-space. Then
150 X ˆ Γ « X.
Γ
«
Proof. The composite X Ý Ñ X ˆ t1u ãÑ X ˆ Γ  pX ˆ Γq{Γ is clearly continuous. It is surjective
because the diagonal Γ-orbit of any point px, γq P X ˆ Γ contains the point pxγ´1 , 1q, and injective
because px, 1q “ py, 1q ¨ γ “ pyγ, γq only if γ “ 1 and x “ y. To see its inverse is continuous,
consider the composition

X ˆ Γ ÝÑ X ˆ Γ ÝÑ X,
px, γq ÞÝÑ px, γ´1 q ÞÝÑ xγ´1 .

151 This map is constant on diagonal Γ-orbits, since pxδ, γδq ÞÝÑ xδ ¨ δ´1 γ “ xγ´1 , so it descends to
152 a continuous map pX ˆ Γq{Γ ÝÑ X, the inverse to the map X ÝÑ pX ˆ Γq{Γ described above.

153 Freedom and efficacy

154 We will have to deal with fixed point sets of isotropy actions starting in Section 9.2. If a group G
155 acts on a topological space X, we denote the fixed point set of the action by X G .
156 A G-action G ˆ X ÝÑ X is said to be effective (or faithful) if only the neutral element 1 P G
157 acts trivially: that is, one has the implication

@x P X pgx “ xq ùñ g “ 1.

158 Equivalently, an action is effective just if it curries to an injective homomorphism G ÝÑ Homeo X.

159 A G-action G ˆ X ÝÑ X is said to be free if only the neutral element 1 P G fixes any point: that
160 is, one has the implication
Dx P X pgx “ xq ùñ g “ 1.
161 These conditions can be pleasingly restated in terms of stabilizers thus:
Ş
162 • an action is effective when xPX Gx “ t1u;
Ť
163 • an action is free when xPX Gx “ t1u.
 Chapter 1. Bundles, actions, and orbits 7

164 We will write G-Free for the full subcategory of G-Top whose objects are free G-actions. Contained
165 in G-Free is the full subcategory of principal G-bundles and G-maps (automatically bundle maps),
166 as discussed in Appendix B.1.3.
167 In Section 3.2.3, we will define a construction that from any G-action produces a free G-action,
168 and use it to define an equivariant cohomology theory.

169 The category of G-orbits

170 The action on a G-space X induces an orbit decomposition: one has
ž
X“ Gx
GxPX{G

as sets where the orbit Gx is defined as tgx : x P Xu and the inclusion Gx ãÝÝÑ X is continuous.
The orbits have a fairly tightly constrained structure: the pointwise stabilizers or isotropy sub-
groups Stabpxq “ Gx :“ tg P G : gx “ xu of points x P X are closed subgroups if X is T1 ,2 and the
orbit–stabilizer theorem induces homeomorphisms
„
G{ Stabpxq ÝÑ Gx :
g Stabpxq ÞÝÑ gx,

171 where G{ Stabpxq is the left coset space with the quotient topology. Such a space is called a
172 homogeneous space for G. These maps are clearly continuous, and will certainly be open as well
173 if G is compact and X Hausdorff, which they are in our cases of interest. We mark this proposition
174 for later use.

175 Proposition 1.2.1. Let G be a locally compact, Hausdorff, second countable topological group acting
176 continuously on a locally compact Hausdorff space X, and Stabpxq the stabilizer in G of x P X. Then the
177 group action induces a homeomorphism
«
G{ Stabpxq ÝÑ Gx.

178 Proof. The argument above suffices for a compact Lie group acting on a Hausdorff space. See
179 Garrett [Gar10, Prop. 6.0.2] for a proof of the statement with the weaker hypotheses.

It follows that an understanding of the full subcategory G-Orbit of right quotients of G by

closed subgroups is an important component of understanding G-Top. Write SubpGq for the cat-
„
egory of closed subgroups of G, with morphisms inclusions and isomorphisms H ÝÑ gHg´1 in-
duced by inner automorphisms of G. Then the correspondence K ÞÝÑ G{K induces an equivalence
„
SubpGqop ÝÑ G-Orbit of categories: an inclusion H ãÝÝÑ K corresponds to a G-equivariant quo-
tient map G{K ÝÝ G{H; and if H “ gKg´1 , then r g : x ÞÝÑ xg descends to a G-homeomorphism
„
G{H ÝÑ G{K :
xH ÞÝÑ xgK.
2 It will be in all cases we care about. The proof is that the inverse image of txu under the action G ˆ X ÝÑ X and

the set G ˆ txu are both closed if txu is.

 Chapter 1. Bundles, actions, and orbits 8

180 Tubes, slices, and G–CW complexes

181 Given the orbit decomposition, one can ask how much of the structure of G-Top is determined
182 by G-Orbit (or equivalently by SubpGq), and the answer, in a sense, is all of it, at least if G is a
183 compact Lie group. We reproduce here the topological consequences of this analytic structure,
184 suppressing most actual analysis. Most of this material can be found in the useful Appendix B
185 to the book by Ginzburg et al. [GGK02]. Much of the exposition in this section is inspired by a
186 manuscript textbook of Raoul Bott and Loring Tu [BTara].
187 Let G be a compact Lie group acting smoothly on a manifold M, and x P M. Then K “ Stabpxq
188 is a closed subgroup of G, hence a Lie group, and the coset space G{K is a manifold as noted
189 in Theorem B.4.3, diffeomorphic to the orbit Gx. Moreover, orbits admit equivariant tubular
190 neighborhoods [Kos53].

191 Theorem 1.3.1 (Equivariant tubular neighborhood theorem (Koszul)). Let G be a compact Lie group,
192 M a smooth G-manifold, and x P M a point. Then there exists a G-invariant open neighborhood of Gx in
193 M.

194 Sketch of proof. ş Given a Riemannian metric x´, ´y1 on M and Haar measure dg on G, the aver-
195 age xv, wy :“ G xg˚ v, g˚ wy1 dg defines G-invariant Riemannian metric on M.
( This induces a G-
196 invariant metric d on M. The neighborhood Bε pGxq “ y P M : dpGx, yq ă ε is G-invariant.

197 This statement can and should be strengthened.

198 Proposition 1.3.2 (Koszul). For small enough ε, the tubular neighborhood of Theorem 1.3.1 is G-
199 isomorphic to the normal bundle.

200 Sketch of proof. The G-invariant metric defines a norm } ´ } on TM which allows us to canonically
201 embed the normal bundle ν to Gx as the orthogonal complement of TpGxq in pTMq|Gx . It is a
202 well-known feature of the exponential map that each point gx P Gx, there exists ε ą 0 such that
203 expgx is defined on the ball Bε p0q Ĺ Tx M. Since Gx is compact, there then exists ε small enough
204 that for all x and all v P pTMq|Gx with }v} ă ε, the map exp : pv P Tgx Mq ÞÝÑ expgx v is defined.
205 The exponential of M, when restricted to TpGxq, is the exponential for Gx and so maps into Gx,
206 so to obtain a potential homeomorphism from some bundle to a tubular neighborhood of Gx,
207 we should restrict our exponential
“ ‰ orthogonal complement ν to TpGxq, which also has the
to the
208 correct total dimension dim pTMq|Gx ´ dim Gx “ dim M. If we decrease ε sufficiently, then by
209 the nonequivariant tubular neighborhood theorem, exp is a G-equivariant diffeomorphism from
210 νε :“ tv P ν : }v} ă εu to Bε pGxq.

211 Finally, if we write νx for the disk Tx M X νε , we can rewrite νε as a disk bundle.
„
212 Proposition 1.3.3. In this notation, there is a G-equivariant diffeomorphism G ˆ νx ÝÑ νε .
Gx

Proof. Under the action

µ : G ˆ νx ÝÑ νε ,
pg, vq ÞÝÑ g˚ v,

213 each g P G takes νx diffeomorphically onto pνε qgx , since g P Diff M is itself smooth and invertible
214 and we constructed the norm on ν to be G-invariant. The fiber of µ over a fixed v0 P pνε qy0 is the
 Chapter 1. Bundles, actions, and orbits 9

215 collection of pairs pg, vq, pg1 , v1 q, etc., with g˚ v “ v0 . We must have gx “ g1 x “ y0 for any two
216 such pairs, so k “ g´1 g1 P Gx , or g1 “ gk and then

g˚ v˚ “ v0 “ g˚1 v1 “ g˚ k˚ v1

217 implies pg1 , v1 q “ pgk, k´1

˚ vq by bijectivity
` of g˘˚ . So each fiber is an orbit of the diagonal action
218
´1
Gx ñ G ˆ νx given by k ¨ pg, vq “ g˚ pk˚ q , k˚ v of Gx . Quotienting out this action, µ descends to a
„
219 diffeomorphism ϕ : G ˆGx νx ÝÑ νε , which is left G-equivariant: ϕrg1 g, vs “ g˚1 g˚ v “ g˚1 ϕrg, vs.

220 Given a point x P M, if there is a Gx -invariant superset X Q x such that GX is an open

221 neighborhood of x satisfying GX « G ˆGx X, then X is called a slice. We have just seen that a
222 small normal disk νx is such a slice, so the result is also traditionally called the slice theorem. It is
223 a subtle theorem of Sören Illman that the in addition to these slices, smooth G-manifolds admit
224 another very nice kind of G-equivariant decomposition.

225 Definition 1.3.4 (Matumoto). A G–CW complex is a topological space X admitting a filtration
226 pX n q as follows:

• X 0 “ KPSubpGq α pG{Kq ˆ Dα0

š š
227

228 • Given X n , there are G-equivariant maps ϕα,K : pG{Kq ˆ Sαn ÝÑ X n , where K runs over SubpGq
229 and α over some arbitrary index set, compiling into a map
ž ž
ϕ“ ϕα,K : pG{Kq ˆ Sαn ÝÑ X n
α,K α,K

230 such that N

n n`1
š
X n`1
“ X > α,K pG{Kq ˆ D α
ϕpsq „ s.

X n with the weak topology.

Ť
231 • X“

232 Write en for the interior of a disk D n . Then the various pieces G{K ˆ en making up X, called G-
233 cells, are comprised of open n-balls gK ˆ en , called the n-cells of the complex. We write Celln pXq
234 for the set of n-cells of X. A G–CW pair pX, Aq is a G–CW complex X and subcomplex A.

235 Note that G acts cellularly on a G–CW complex X in the sense that if σ P Celln pXq, then so also
236 is gσ in Celln pXq, and given such a cell σ, every point x P σ has the same stabilizer Stabpxq “:
237 Stab σ under the defining G-action. Setting G “ 1 to be the trivial group in the definition recovers
238 CW-complexes in the traditional Whitehead sense; because an orbit is a “point” in the equivariant
239 sense, this definition is a natural generalization. The following is then plausible.

240 Proposition 1.3.5 (Matumoto [Mat71b]). Let X be a G–CW complex. Then the orbit space X{G inherits
a canonical CW structure whose n-cells are σ{G : σ P CellnG pAq .
(
241

242 Theorem 1.3.6 (Illman [Ill83]). Let G be a compact Lie group and M a smooth manifold, possibly with
243 boundary. Then M admits a G–CW structure.

244 Thus G–CW complexes are a natural class of topological group actions to consider from the
245 point of view of homotopy theory.
 Chapter 1. Bundles, actions, and orbits 10

246 Historical remarks 1.3.7. The result Theorem 1.3.6 has an interesting history. It was Takao Matu-
247 moto [Mat71a, Prop. 4.4] who first considered G–CW structures and observed that they existed
248 for compact Lie groups G and smooth, closed G-manifolds. His proof relies on an earlier result of
249 Chung-Tao Yang [Yan63], which in turn relies on a result of Stewart Cairns [Cai41] whose proof
250 was never published, and which unfortunately contradicts a valid 1970 result of Larry Sieben-
251 mann [Sie77, p. 312]. The contradiction was apparently discovered by John Mather around 1976
252 and relayed to Sören Illman by Katsuo Kawakubo; Illman reported this error in 1978 [Ill78], and
253 Andrei Verona was able to repair Yang’s proof in 1979 [Ver79]. Illman was then able to regain
254 Matumoto’s result in 1983 [Ill83, Thm. 7.1], twenty years after Yang’s proof and a full forty-two
255 after Cairns’s statement.

256 Fixed point sets of actions on homogeneous spaces

257 In order to localize integrals over and study cohomology rings of fixed point sets of actions on
258 homogeneous spaces, as we shall do in Chapter 9 and Appendix C, it helps to understand what
259 these sets are.
260 We initially focus on describing these fixed point sets in terms of normalizers. Several of
261 these lemmas are due to Oliver Goertsches and Sam Noshari, while the author does not know
262 citations for the others. The proofs (admittedly trivial) are of the author’s own design unless
263 noted otherwise.

264 Lemma 1.4.1. Let G be a group, H and K subgroups acting respectively by left and right multiplication.
265 The fixed point set pG{Kq H is tgK P G{K : g´1 Hg ď Ku.

266 Proof. We have the chain of equivalences of set containments

gK P pG{Kq H ðñ HgK “ gK ðñ g´1 HgK “ K ðñ g´1 Hg ď K ðñ H ď gKg´1 .

267 The last condition is clearly independent of the representative g of gK P G{K.

268 This set is easier to describe in the event of a torus action.

269 Lemma 1.4.2 ([Goe12, Lem. 4.3]). Let G be a compact, connected Lie group, K a closed, connected
270 subgroup, and S the maximal torus of K. The fixed point set pG{KqS is NG pSqK{K, which is homeomorphic
271 to the coset space NG pSq{NK pSq.

272 Proof (after Goertsches). From Lemma 1.4.2 we know

gK P pG{KqS ðñ g´1 Sg ď K.

273 By Theorem B.4.9, the latter condition holds if and only if there is k P K such that k´1 g´1 Sgk “ S,
274 in which case gk P NG pSq and g P NG pSqK. Thus pG{KqS “ NG pSqK{K ď G{K.
275 Now, there is a natural continuous map NG pSq  NG pSqK  NG pSqK{K, under which an
276 element n P NG pSq is sent to K if and only if n P K, so the “kernel” is NG pSq X K and one has a
277 continuous bijection
NG pSq N pSq X K ÐÑ NG pSqK K.
L L
G

278 Since these are compact Hausdorff spaces, this is a homeomorphism. But the denominator on
279 the left is NK pSq by definition.
 Chapter 1. Bundles, actions, and orbits 11

280 This is even simpler if the isotropy subgroup is a torus.

281 Corollary 1.4.3. Let G be a compact, connected Lie group containing a torus S. Then

pG{SqS “ NG pSq{S

282 Proof. Take K “ S in Lemma 1.4.2.

283 The Weyl group also comes up in this context as well, for the following reason.

284 Lemma 1.4.4. Let G be a compact Lie group and K a closed subgroup containing a maximal torus T. The
285 set of fixed points of the left multiplication action of T on the right coset space G{K is in natural bijection
286 with the coset space WG {WK , where WG “ NG pTq{T is the Weyl group of G. In particular, WG is the fixed
287 point set of the left action of T on G{T.

288 Proof. By Lemma 1.4.2, we have pG{KqT « NG pSq{NK pSq, so by what is essentially the third
289 isomorphism theorem,
NG pTq WG
pG{KqT « ÐÑ .
NK pTq WK
290 Lemma 1.4.1 obviously implies many fixed point sets are empty.

291 Corollary 1.4.5. Let G be a compact Lie group and K and H subgroups. The set of fixed points of the left
292 multiplication action of H on the right coset space G{K is empty unless H is conjugate in G to a subgroup
293 of K. In particular, if rk H ą rk K, then pG{Kq H “ ∅.

294 Now that we have at least a philosophical understanding of these fixed point sets, we try
295 to develop as well an understanding of the number and homeomorphism types of their com-
296 ponents. These components will turn out to be quotients of centralizers, and their cohomology
297 rings computable by techniques we will develop in Section 7.3.1. One needs the following useful
298 counting lemma.

299 Proposition 1.4.6 (Noshari [GN15, Proposition 2.3]). Let Γ be compact Lie group and H a closed
300 subgroup. Then the number of components of Γ{H is given by
ˇ |π0 Γ| ¨ ˇπ0 pH X Γ0 qˇ
ˇ ˇ
ˇ
ˇπ0 pΓ{Hqˇ “ ,
|π0 H|
301 where Γ0 is the identity component of Γ.

302 Proof. Consider the intermediate group Γ0 H, which is some union of components of Γ. One has
303 a bundle
Γ0 H H ÝÑ Γ H ÝÑ Γ Γ H ,
L L L
0

Γ
ˇ ˇ
304 the fiber of which has only one component, since 0 is connected, so to find ˇπ0 pΓ{Hqˇ is to find
the index rΓ : Γ0 Hs “ ˇπ0 pΓ{Γ0 Hqˇ, which is finite because rΓ : Γ0 s already is, Γ being compact.
ˇ ˇ
305

306 Moreover, since indices are transitive,

|π0 Γ| “ rΓ : Γ0 s “ rΓ : Γ0 HsrΓ0 H : Γ0 s “ ˇπ0 pΓ{Hqˇ ¨ rΓ0 H : Γ0 s.

ˇ ˇ

307 For π0 H, one has an analogous decomposition

|π0 H| “ rH : H X Γ0 s ˇπ0 pH X Γ0 qˇ,

ˇ ˇ
 Chapter 1. Bundles, actions, and orbits 12

308 linked to the previous one by the “second isomorphism theorem” bijection rΓ0 H : Γ0 s “ rH :
309 H X Γ0 s. Rearranging, one finds

|π0 Γ| |π0 H|
ˇπ0 pΓ{Hqˇ “ rΓ0 H : Γ0 s “ rH : H X Γ0 s “ ˇπ0 pH X Γ0 qˇ ,
ˇ ˇ ˇ ˇ

ˇ ˇ
310 and isolating ˇπ0 pΓ{Hqˇ on one side yields the result.

311 In what follows, let G be a compact, connected Lie group, K a closed, connected subgroup,
312 and S a maximal torus of K. Write N “ NG pSq and write Z “ ZG pSq for the centralizer. Recall
313 that π0 N denotes the component group of N. Then pG{KqS « N{NK pSq is naturally expressed in
314 terms of other known objects.

Lemma 1.4.7. Conjugation induces maps

N ÝÑ Aut S;
π0 N Aut S.

315 The kernel of the first is the centralizer Z, which is consequently the identity component of the normalizer
316 N.

Proof. Conjugation yields a continuous natural homomorphism

N ÝÑ Aut S,
` ˘
n ÞÝÑ x ÞÝÑ nxn´1 .

317 The kernel of this map is definitionally Z, for conjugation by n P N fixes S pointwise if and only
318 if n commutes with all elements of S. By the first isomorphism theorem, we obtain an injection
319 N{Z Aut S.
320 Since G is connected, the centralizer Z of S is connected, being the union of those maximal
321 tori of G that contain S. As Z is connected and Aut S discrete, it follows Z is the connected
322 component of the identity in N.

323 Corollary 1.4.8 (Noshari). Let G be a compact, connected Lie group, K a closed, connected subgroup, and
324 S the maximal torus of K. Then the fixed point set pG{KqS « NG pSq{NK pSq has |π0 N|{|WK | connected
325 components, each homeomorphic to Z{S.

326 Proof. To get the number of components, apply Proposition 1.4.6 with Γ “ NG pSq and H “ NK pSq,
327 noting N0 “ Z by Lemma 1.4.7 and that NK pSq X ZG pSq “ ZK pSq “ S is path-connected because
328 S is a maximal torus of K.
329 Since NG pSq acts transitively, the components will be mutually homeomorphic, and the iden-
330 tity component is
ZG pSqNK pSq N pSq « ZG pSq Z pSq X N pSq “ Z{S
L L
K G K

331 by the second isomorphism theorem.

332 Corollary 1.4.9. Let G be a compact, connected Lie group and S a torus in G. Then the fixed point set
333 pG{SqS « N{S has |π0 N| connected components, each homeomorphic to Z{S.
334 Chapter 2
335 Classifying spaces

336 In this section, we carry out the construction of the universal principal G-bundle EG Ñ BG,
337 which we use essentially as a tool to convert actions into closely related free actions. The existence
338 of this bundle is more important than the details of its construction in almost everything that
339 follows, but we will at one key point (Definition 9.1.3) use the fact that EG admits commuting
340 right and left actions of G, and to justify this fact we provide the construction.

341 The weak contractibility of EG

342 The original purpose of the universal principal G-bundle EG Ñ BG was to be a principal G-
343 bundle such that all others G Ñ E Ñ B arose as pullbacks. For that reason, the resulting map
344 B ÝÑ BG of base spaces classifies the bundle E Ñ B, and is called the classifying map of the
345 bundle, and BG is called the classifying space of a principal G-bundle.
346 The fact that EG is weakly contractible—which is much of why we care about the universal
347 bundle—turns out to be a consequence of that demand. In this subsection, we explain the rele-
348 vance of this demand. It will simplify the argument to know that all maps of principal G-bundles
349 are pullbacks.

350 Proposition 2.1.1. Consider a principal G-bundle map

P /E

 f 
X / B.

351 The pullback bundle f ˚ E Ñ X is isomorphic to P Ñ X as a principal G-bundle.

352 Proof. Recall from earlier that the total space f ˚ E “ X is the pullback in Top of the diagram
353 X Ñ B Ð E. Since P also admits a map to such a diagram, there is a continuous map P Ñ f ˚ E
354 commutatively filling in
P / f ˚E /* E

 f 
X / B.

355 For any x P X, by assumption, the maps of fibers P|x Ñ E| f pxq Ð p f ˚ Eq|x are G-equivariant
356 homeomorphisms, so P Ñ f ˚ E is a bijective G-map. To see its inverse is continuous, it is enough

13
 Chapter 2. Classifying spaces 14

357 to restrict attention to an open U Ď X trivializing both P and f ˚ E, so we need only show the
358 inverse of a continuous G-bijection ϕ filling in the diagram

/UˆG
ϕ
UˆG

 
U
` ˘
359 is continuous. By commutativity,
` ˘ we may write ϕpx, 1q “ x, ψpxq` for a continuous
˘ ψ : U Ñ G,
360 so that ϕpx, gq “ x, ψpxqg by equivariance. Then ϕ´1 px, gq “ x, ψpxq´1 g , and since ψ and
361 g ÞÑ g´1 are continuous, so is ϕ´1 .

362 Thus the EG Ñ BG we seek just needs to be a final object in the category of principal G-
363 bundles. Recall that Top admits CW approximations, so that up to homotopy, we may assume
364 the base space of our principal G-bundle P Ñ X is a CW complex. Then X is built one level at a
365 time from a discrete set X 0 of vertices by gluing disks Dαn`1 to the n-skeleton X n along attaching
366 maps ϕα : BDαn`1 « Sn ÝÑ X n , so we can view P as being constructed inductively from principal
367 G-bundles over these attached cells.
368 We require one intuitively plausible lemma, which we will not prove.

369 Lemma 2.1.2 ([Ste51, Cor. 11.6, p. 53]). Let B be a contractible, paracompact Hausdorff space and E Ñ B
370 an F-bundle for some fiber F. Then E is isomorphic as an F-bundle to B ˆ F.

371 By the lemma, principal G-bundles over disks are trivial, so P|X n`1 is the identification space of
372 P|X n with some bundles Dαn`1 ˆ G Ñ Dαn`1 , the identifications given by G-maps Sαn ˆ G ÝÑ P|X n .
373 The task of constructing a G-map P ÝÑ EG can now be undertaken one cell at a time. To start,
374 P|X0 is a disjoint union of copies of G, and any homeomorphic map of these to fibers of EG Ñ BG
375 will work. Suppose inductively that a G-map P|X n ÝÑ EG has been built, and we want to extend
376 this to the space P|X n Y pD n`1 ˆ Gq, where D n`1 ˆ G is attached by a G-map Sn ˆ G ÝÑ P|X n . We
377 can do anything we want over the interior of D n`1 , and we know what must happen over P|X n ,
378 so our only constraint is the composition of the preexisting G-map and the attaching map,

ψ : Sn ˆ G ÝÑ P|X n ÝÑ EG.

379 Thus the task is really to extend an arbitrary G-map Sn ˆ G ÝÑ EG over the interior of D n`1 ˆ G:

D n`1 ˆ
O G

? &
Sn ˆ G / EG.
ψ

380 But a G-map ψ r : D n`1 ˆ G ÝÑ EG is uniquely determined by its restriction to the standard section
381 D n`1 r gq “ ψpx,
ˆ t1u since ψpx, r 1qg, so it is necessary and sufficient to extend the restriction
382
n
S ÝÑ EG to a map D n`1 ÝÑ EG. If it is possible to do so, then restrictions of the latter map
383 to concentric spheres of decreasing radius form a nullhomotopy of the map Sn ÝÑ EG, so the
384 condition finally turns out to be that πn pEGq “ 0.

385 Proposition 2.1.3. A principal G-bundle EG ÝÑ BG is universal just if π˚ pEGq “ 0.

 Chapter 2. Classifying spaces 15

386 Thus the collapse EG ÝÑ pt of the total space is a weak homotopy equivalence, and so if EG
387 is a CW complex, then it is actually contractible by Whitehead’s theorem B.2.7. Now seems as
388 good a time as any to derive a corollary we will use repeatedly later.
389 Corollary 2.1.4. If G is a path-connected group, then BG is simply-connected.
390 Proof. The long exact homotopy sequence Theorem B.2.4 of G Ñ EG Ñ BG contains subse-
391 quences
0 “ πn`1 pEGq ÝÑ πn`1 pBGq ÝÑ πn pGq ÝÑ πn pEGq “ 0,
392 yielding isomorphisms πn`1 pBGq – πn pGq for all n, and in particular for n “ 0.

393 A construction of EG for G a compact Lie group

394 Example 2.2.1. Embedding Cn ãÝÝÑ Cn`1 as Cn ˆ t0u, the direct union is the countable direct sum
À ś
395 C8 “ N C, which can be seen as the subspace of the countable direct product N C such that
396 all but finitely many coordinates are 0. Within C8 lies the unit 8-sphere
ÿ
S8 :“ t~z P C8 : z2j “ 1u.
397 Write C8 8 8 ˆ 8
ˆ :“ C zt0u. The scalar multiplication of C on C restricts to a free action of C on Cˆ
398 and of S1 on S8 , with the same orbit space
8 1
CP8 :“ C8 ˆ
ˆ {C « S {S ,

399 called infinite complex projective space. The fiber space S8 Ñ CP8 can be seen as the increasing
400 union of restrictions S2n´1 Ñ CPn´1 , where we conceive S2n´1 as S8 X Cn . Each CPn admits an
401 open cover by contractible affines, so these restrictions are all principal S1 -bundles, and S8 Ñ
402 CP8 is as well.
403 We claim this bundle satisfies the requirements to be ES1 Ñ BS1 . Because S8 is the union of
404 the unit spheres S2n´1 Ĺ Cn , by a compactness argument, any map Sm ÝÑ S8 must lie inside
405 some sufficiently large Sn , and πm Sn “ 0 for m ă n. Thus S8 is weakly contractible. There is a
406 natural CW structure on S8 where two hemispheres D n attach to each Sn´1 to form Sn , so we
407 know from Whitehead’s theorem S8 is contractible, but in fact, it is possible to see so directly as
408 well.
409 Proposition 2.2.2. The unit 8-sphere S8 is contractible.
` ˘
410 Proof. There is a homotopy taking the subspace S1 :“ S8 X t0u ˆ C8 « S8 with first coordinate
411 zero to the point e1 “ p1,~0q, given by
ht p~zq :“ psin tqe1 ` pcos tq~z;
412 this is just a renormalization of the straight-line homotopy. Now it will be enough to find a
413 homotopy from S8 to S1 . Write s : ~z ÞÝÑ p0, ~zq for the shift homeomorphism. One’s first inclination
414 is to take
f t p~zq “ p1 ´ tq~z ` t ¨ sp~zq.
415 If we can show f t pS8 q avoids ~0 P C8 , then the renormalization fˆt :“ f t {| f t | will suit our purposes.
416 Now
` note any ~z P C8 has a last nonzero coordinate zn , so the nth and pn ` 1qst coordinates
417 p1 ´ tqzn , tzn q of f t p~zq will never simultaneously be zero, and the linear maps f t P EndC C8 are
418 injective. Thus fˆt is an isotopy.
 Chapter 2. Classifying spaces 16

419 Example 2.2.3. Replacing C with the quaternions H (respectively, the reals R) and S1 with Spp1q «
420 S3 (resp., Op1q « S0 – Z{2), one finds a universal Spp1q-bundle ESpp1q Ñ BSpp1q is

S3 ÝÑ S8 ÝÑ HP8

421 and a universal Op1q-bundle EOp1q Ñ BOp1q is

S0 ÝÑ S8 ÝÑ RP8 .

422 Any closed subgroup K ď G acts freely on EG by a restriction of the G-action, so one has
423 a natural map EG ÝÑ EG{K with fiber K. It is intuitively plausible that this is also a fiber
424 bundle, and this is actually the case in the event G is a Lie group: by Theorem B.4.3, G ÝÝ G{K
«
425 is a principal K-bundle, and the local trivializations φ : pEGq|U ÝÑ U ˆ G of EG Ñ BG and
«
426 G|V ÝÑ V ˆ K of G Ñ G{K combine to yield local trivializations φ´1 pU ˆ G|V q ÝÑ U ˆ V ˆ K
427 making EG Ñ EG{K a principal K-bundle, so that EG can serve as EK and EG{K as BK.
428 To make use of this observation, we can use the classic result Theorem B.4.7, due to Peter
429 and Weyl, that every compact Lie group has a faithful finite-dimensional unitary representation.
430 Thus, if we can find EUpnq, we will have bundles EG Ñ BG for all compact Lie groups G. Here
431 is one construction.
432 Example 2.2.4. The infinity-sphere S8 can be seen as the collection of orthonormal 1-frames in
433 C8 and CP8 as the space of 1-dimensional vector subspaces of C8 . Analogously, one can form
434 the infinite complex Stiefel manifolds Vn pC8 q of orthonormal n-frames in C8 , which is to say,
435 sequences pv1 , . . . , vn q of n mutually orthogonal vectors of length one, topologized as a subspace
ś
436 of n S8 , and the infinite complex Grassmannian Gn pC8 q of n-planes in C8 . Just as S8 projects
437 onto CP8 , so does each Vn pC8 q project onto Gn pC8 q through the span map pv1 , . . . , vn q ÞÝÑ
ř
438 Cv j . The unitary group Upnq acts freely on Vn pC8 q; if one considers an element of S8 as an
439 infinite vertical vector, or a 8 ˆ 1 matrix, then an element of Vn pC8 q can be seen as an 8 ˆ n
440 matrix, and right multiplication by an n ˆ n matrix in Upnq produces another 8 ˆ n matrix
441 spanning the same column space, so that the fiber of the span map Vn pC8 q ÝÑ Gn pC8 q is
442 homeomorphic to Upnq. With a little work, it can be seen that Upnq Ñ Vn pC8 q Ñ Gn pC8 q is a
443 fiber bundle.
444 Moreover, an analogue of the contraction of S8 in Example 2.2.1 shows Vn pC8 q to be con-
445 tractible: the idea is to first conduct `the isotopy˘ fˆt of S8 consecutively n times, taking S8 into
446 t0un ˆ S8 and hence Vn pC8 q into V`n t0un ˆ C˘8 , and then use a renormalized straight-line ho-
447 motopy generalizing ht to take Vn t0un ˆ C8 to the identity matrix In P Cnˆn Ĺ C8ˆn , rep-
448 resenting the standard basis of the subspace Cn ă C8 . Write gt for the resulting homotopy
449 Vn pC8 q ˆ I ÝÑ C8ˆn . In the same way that our first guess for S8 failed to have image strictly
450 unit-length, this map gt , while it preserves linear independence, does not preserve orthogonal-
451 ity. But if we postcompose to gt the Gram–Schmidt orthonormalization procedure, which is a
452 well-defined projection

tn-tuples of linearly independent vectors in C8 u ÝÑ Vn pC8 q,

453 we achieve the desired homotopy.

454 One analogously finds that Vn pR8 q Ñ Gn pR8 q and Vn pH8 q Ñ Gn pH8 q respectively satisfy
455 the hypotheses for EOpnq Ñ BOpnq and ESppnq Ñ BSppnq. The double cover Vn pR8 q{SOpnq “:
456 Grn pR8 q of Gn pR8 q, the oriented Grassmannian consisting of all oriented n-planes in R8 , is a
457 BSOpnq.
 Chapter 2. Classifying spaces 17

458 Milnor’s functorial construction of EG

459 These constructions, though pleasing, are ad hoc. In 1955, John Milnor [Mil56] found a functorial
460 construction of EG Ñ BG that works for any topological group G, not even assumed Hausdorff,
461 which we will briefly summarize.
462 To lay the groundwork, the join X ˚ Y of two topological spaces X and Y is the quotient of the
463 product X ˆ Y ˆ I with an interval by identifications px, y, 0q „ px, y1 , 0q and px, y, 1q „ px1 , y, 1q
464 for all x, x1 P X and all y, y1 P Y. We may think of this as an pX ˆ Yq-bundle over I that has been
465 collapsed to X over 0 and to Y over 1, and consider X and Y to be included as these particular
466 end-subspaces.

Figure 2.3.1: Some low-dimensional joins

I*I S0 * S0 S1 * S0

467 Examples 2.3.2. The join I ˚ I of two intervals is a 3-simplex ∆3 , the join S0 ˚ S0 is a circle S1 , and
468 the join S1 ˚ S0 is a 2-sphere S2 .
469 It is not hard to see that generally X ˚ pt is the cone CX on X and, as in the examples
470 above, X ˚ S0 is the suspension SX of X, so the process of iteratively joining points generates the
471 simplices ∆n and that of iteratively joining copies of S0 yields spheres Sn .
472 In particular, the join S1 ˚ S1 is S3 , which can also be seen more geometrically. If one views S3
473 as the unit sphere in C2 , it is “foliated” by

Tr :“ pz cos r, w sin rq : z, w P S1 ,
(

474 for r P r0, π{2s, which are tori S1 ˆ S1 for r P p0, π{2q and circles for r P t0, π{2u: the S1 factor
475 corresponding to the w-coordinate collapses at r “ 0 and the S1 corresponding to the z-coordinate
476 collapses at r “ π{2. Further foliating the toral Tr into circles with constant ratio z{w P S1 yields
477 the Hopf fibration S1 Ñ S3 Ñ S2 .
478 The key feature of joins is that they are more connected than their factor spaces, as one already
479 sees in the sphere examples above.
480 Lemma 2.3.3. Let X, Y ‰ ∅ be nonempty spaces. Then the join X ˚ Y is path-connected.
481 Proof. Fix a basepoint x0 “ rx0 , y, 0s P X Ĺ X ˚ Y. From a given rx,“ y, ts P X ˚ Y,‰ we may trace a
482 path to y P Y simply by increasing t to 1, then follow the path t ÞÑ x0 , y, p1 ´ tq back to x0 .

483 Lemma 2.3.4. Let X be a path-connected space and Y nonempty. Then the join X ˚ Y is simply-connected.
484 Proof. The projection t : X ˆ Y ˆ I ÝÝ I induces a projection X ˚ Y ÝÝ I we also call t. The subset
485 V :“ t´1 r0, 1q deformation retracts onto X by uniformly decreasing t to 0, and X in turn contracts,
486 within X ˚ Y, to any fixed y0 P Y via f t pxq “ rx, y0 , ts, so V is contractible in X ˚ Y.
487 To show a loop γ in X ˚ Y is nullhomotopic, it therefore suffices to homotope it into V. We
488 could always do so if X ˚ Y deformation retracted into V, but it does not. The partial quotient

A :“ X ˆ Y ˆ I px, y, 0q „ px, y1 , 0q,

L
 Chapter 2. Classifying spaces 18

489 where the X factor over t “ 1 is not quotiented out, does however deformation retract into
490 r “ t´1 r0, 1q Ĺ A, so if γ can be lifted to A, it can be homotoped into V.
V

Figure 2.3.5: Nullhomotoping a curve γ in X ˚ Y to lift into A

Ñ Ñ Ñ
paq pbq pcq

X *Y A A A

491 Such an attempt to lift a γ to A is represented by arrow paq in Figure 2.3.5. If we write γ in
492 terms of the coordinates x, y, t, then t ˝ γ is everywhere defined, but x ˝ γ is indeterminate when
493 t ˝ γ “ 1 and y ˝ γ when t ˝ γ “ 0. Lifting to γ to A amounts precisely to continuously extending
494 x ˝ γ to the whole domain S1 . This may initially be impossible: because the X-coordinate is
495 collapsed at t “ 1, it may be that γ is continuous at a point z P pt ˝ γq´1 t1u even though the left
496 and right limits px ˝ γqpz ` 0q and px ˝ γqpz ´ 0q are not equal. But we can homotope γ into a
497 curve that does lift into A, as follows, and then we will be done.
498 Let us : I ÝÑ I be ` a nondecreasing
˘ homotopy fixing t0, 1u and such that u1 r1{2, 1s “ t1u.1
499 Then hs : px, y, tq ÞÝÑ x, y, us ptq is a homotopy of X ˚ Y collapsing t´1 r1{2, 1s onto Y, so that γ is
500 homotoped to a loop γ1 with t ˝ γ1 “ 1 on the set C “ tz P S1 : pt ˝ γqpzq ě 1{2u. This transition is
501 represented, in A, by the arrow pbq in Figure 2.3.5.
502 Thus x ˝ γ1 is a continuous function on S1 zC, and we need to extend it to a continuous function
503 x : S1 ÝÑ X. We should plainly take x1 “ x ˝ γ on BC “ tz P S1 : pt ˝ γqpzq “ 1{2u, so it remains to
1

504 define x1 on the open set CzBC Ĺ S1 , which is a union of open intervals pz0 , z1 q. But to extend x1
505 over pz0 , z1 q is to find a path in X connecting x1 pz0 q and x1 pz1 q, which we may always do because
506 X is path-connected. This extension is represented by arrow pcq in Figure 2.3.5.

507 Definition 2.3.6. A nonempty space X is p´1q-connected, and, for each n P N, is n-connected if
508 π j pXq “ 0 for all j ď n.

509 Lemma 2.3.7. If X is m-connected and Y is n-connected, then X ˚ Y is pm ` n ` 2q-connected.

(
510 Proof. The previous lemmas prove the cases pm, nq P p´1, ´1q, p0, ´1q ; now assume n ě 0.
511 Applying the Mayer–Vietoris sequence in singular homology to the expected cover tU, Vu, where
512 U “ t´1 r0, 2{3q » X and V “ t´1 p1{3, 1s » Y and U X V “ t´1 p1{3, 2{3q » X ˆ Y, one recovers exact
513 fragments
HN`1 pXq ‘ HN`1 pYq ÝÑ HN`1 pX ˚ Yq ÝÑ HN pX ˆ Yq.
514 The first map is induced by the inclusions X, Y ÝÑ X ˚ Y, both of which are nullhomotopic since
515 X and Y are nonempty, so HN`1 pX ˚ Yq injects into HN pX ˆ Yq.
516 Recall that the Künneth theorem in homology Theorem B.2.2 yields a group isomorphism
à ` ˘ à ` ˘
HN pX ˆ Yq – Hj pXq b HN´j pYq ‘ TorZ1 H j pXq, H N´j´1 pYq .
0ďjďN 0ďjďN

1 Explicitly, we can let us : I ÝÑ I be the straight-line homotopy between the identity map u0 and the map
u1 ptq “ mint1, 2tu.
 Chapter 2. Classifying spaces 19

517 These terms can be nonzero only if j ě m ` 1 and N ´ j ě n ` 1, so adding inequalities, HN pX ˆ Yq

518 can be nonzero if and only if N ě m ` n ` 2. Equivalently, HN pX ˆ Yq “ 0 for N ď m ` n ` 1.
519 It follows that Hj pX ˚ Yq “ 0 for j ď m ` n ` 2. Since π1 pXq “ 0 by the previous lemma, X ˚ Y
520 is pm ` n ` 2q-connected by the Hurewicz Theorem B.2.6.

521 It follows by induction that the n-fold iterated join ˚n X is pn ´ 1q-connected. Including ˚n X
522 as the first factor of ˚n`1 X “ p˚n Xq ˚ X, we can form the direct limit
n
EX :“ lim
ÝÑ ˚ X.
523 Because for all n we have EX « p˚n`1 Xq ˚ EX, it follows that every πn pEXq “ 0. Note that the E
524 construction is functorial:
“ÝÝÝÑa continuous
‰ map ψ : X ÝÑ Y induces a continuous map Eψ : EX ÝÑ
525 ~
EY taking r~x j , ts ÞÝÑ ψpx j q, t j .~
526 Now let G be a topological group. To construct a G-action on EG, we first provide a different
527 description of it. For any topological space X, write CX for the cone on X, the quotient of the
528 product X ˆ I obtained by pinching X ˆ t0u to a point. Then X ˚ Y can be seen as the subspace
529 of CX ˆ CY consisting of elements rx, t1 , y, t2 s such that t1 ` t2 “ 1 and X as the subspace where
530 t2 “ 0. Similarly,( the triple join X ˚ Y ˚ Z can be seen as rx, t1 , y, t2 , z, t3 s P CX ˆ CY ˆ CZ :
531 t1 ` t2 ` t3 “ 1 , and X ˚ Y as the subspace where t3 “ 0, and the infinite join EG can be seen as
!` ˘ ź ÿ )
rg j , t j s jPN P CG : only finitely many t j ‰ 0 and tj “ 1 .
N

532 Write these elements briefly as e “ r~g j ,~ts. With this topology, each coordinate function t j : EG ÝÑ
533 r0, 1s and restriction t´1
j p0, 1s ÝÑ G of a “coordinate” g j is continuous. Then a free, continuous
534 right action of G on EG is given by

r~g j ,~ts ¨ g :“ rÝ
gÑ
j g,~ts.

535 Set BG :“ EG{G, with the quotient topology.

536 We still must show EG Ñ BG is a fiber bundle. Much like projective space, EG admits an
537 open cover by sets Uj “ t´1
j p0, 1s. On U j , the g j -coordinate is well-defined and continuous, so

φj “ pp, g j q : Uj ÝÑ ppUj q ˆ G
` ˘
538 is a continuous bijection. Its inverse φ´1
j is also continuous since it is given by ppeq, g ÞÝÑ
` ˘ `
539 eg´1 ´1 ´1 ´1
j g. Where defined, φi ˝ φ j is given by ppeq, g ÞÝÑ φi peg j gq “ ppeq, gi g j gq. The transition
540 function g ÞÝÑ gi g´1 j g is clearly continuous on Ui X U j , so EG Ñ BG is a principal G-bundle.
541 The classifying space construction B is also functorial, because if ψ : G ÝÑ H is a continuous
542 homomorphism, Eψ is fiber-preserving—
“ÝÝÝÝÑ ‰ “ÝÝÝÑ ‰
Eψ r~g j ,~ts ¨ g “ EψrÝgÑ
` ˘ ` ˘
j g,~ts “ ψpg j gq,~t “ ψpg j q,~t ¨ ψpgq “ Eψ r~
g j ,~ts ¨ ψpgq

543 —so that Eψ descends to a well-defined continuous map Bψ : BG ÝÑ BH.

544 Remark 2.3.8. The most technically demanding part of this proof, Lemma 2.3.4, can be circum-
545 vented if one does not care about a hard bound on the number of joins required to achieve
546 n-connectedness: if X is path-connected, then the middle part t´1 p0, 1q in X ˚ X is also path-
547 connected, so the Seifert–van Kampen theorem applies, which is a much easier proof. We hope
 Chapter 2. Classifying spaces 20

548 our proof of this lemma is fairly readily apprehended; the Milnor original (pp. 431–432) is es-
549 sentially the explicit statement of a homotopy coupled with a remark that the formula is easily
550 verified to be well defined and to provide the needed contraction.
551 Historical remarks 2.3.9. The notation for EG and BG descends from a proud historical precedent.
π
552 The way to denote a bundle F Ñ E Ñ B equipped with a local trivialization with transition
553 functions taking values in G ď HomeopFq, as late as the 1960s [Ste51; BH58; BH59; BH60], was a
554 quintuple pE, B, F, p, Gq, with the last two entries often omitted. This arrowless notation requires
555 one to always remember which object lives in which position, but does have the benefit that if a
556 bundle is named ξ, it has canonically associated with it an entourage of ready-named objects

pEξ , Bξ , Fξ , πξ , Gξ q “ ξ.

557 The standard name for the universal principal G-bundle under this convention is, naturally
558 enough,
pEG , BG , G, πG , Gq.
559 In subsequent decades, perhaps as the functorial nature of E : G ÞÑ EG and B : G ÞÑ BG is
560 embraced, one can see the subscripts of EG and BG gradually move up until one has the EG ÝÑ
561 BG of modern day.
562 Chapter 3
563 Cohomology theories

564 It would be an understatement to say singular cohomology has proven useful in topology. Given
565 that all symmetry arises as the manifestation of group actions, it would seem fruitful to determine
566 if a similar, equivariant theory might be constructed that takes as input not topological spaces, but
567 continuous group actions. Such constructions have been undertaken. In this chapter, we expound
568 characteristics of singular cohomology we would like equivariant cohomology to share, and then
569 construct a candidate theory.

570 Desiderata
571 Ordinary cohomology is a functor H ˚ : Top ÝÑ k-CGA whose underlying functor to graded
ś n
572 k-vector spaces decomposes as a product H of functors H n : Top ÝÑ k-Mod satisfying the
573 Eilenberg–Steenrod axioms:
574 • H ˚ pptq “ H 0 pptq – k.

575 • Homotopic maps f , g : pX, Aq ÝÑ pY, Bq induce equal ring homomorphisms H ˚ f “ H ˚ g.

576 • If pX, Aq P Top and A is a deformation retract of some neighborhood of itself in X, then the
577 map H ˚ pX{A, A{Aq ÝÑ H ˚ pX, Aq induced by the quotient map pX, Aq ÝÑ pX{A, A{Aq “
578 pX{A, ptq is an isomorphism.

579 • There is a natural transformation δ : H ˚ pX, Aq ÝÑ H ˚`1 pAq between the graded k-Mod-
580 valued functors pX, Aq ÞÑ H ˚ pX, Aq and pX, Aq ÞÑ H ˚ pAq, fitting into the exact triangle

H˚ A o HC ˚ X

δ

H ˚ pX, Aq,

581 where the other maps are the ring homomorphisms induced by the inclusions pA, ∅q ãÑ
582 pX, ∅q ãÑ pX, Aq.

583 • Given a collection pXα , Aα q of pairs and pX, Aq their disjoint union, the inclusions pXα , Aα q ãÝÝÑ
584 pX, Aq induce a ring isomorphism
„
ź
H ˚ pX, Aq ÝÑ H ˚ pXα , Aα q.
α

21
 Chapter 3. Cohomology theories 22

585 A functor satisfying these axioms is called an ordinary multiplicative cohomology theory;
586 the word multiplicative is here because we ask the functor take values in graded rings rather than
587 just graded groups Any space X admits a unique map X ÝÑ pt, and any nonempty space X
588 retracts onto a point x P X, meaning pt Ñ txu ãÑ X Ñ pt is the identity. Such a retraction induces
589 algebra maps k “ H 0 pptq Ñ H ˚ X Ñ k such that the composition k Ñ k is the identity, so that the
590 maps k Ñ H ˚ X are injections. Because the map X Ñ pt is unique, given any map X Ñ Y, the
591 canonical map k Ñ H ˚ X is the composition k Ñ H ˚ Y Ñ H ˚ X, so this map k Ñ H ˚ is a bit of
592 extra structure to cohomology; for this reason, we call k the coefficient ring.
593 It can be shown that all cohomology theories agree on CW-complexes. Since any topologi-
594 cal space admits a CW approximation with the same weak homotopy type and weak homotopy
595 equivalences induce isomorphisms in cohomology (Theorem B.2.8), these axioms determine or-
596 dinary cohomology so far as homotopy theory cares.
597 Let G be a topological group. Then a G-equivariant cohomology theory E˚ with coefficients in
598 a commutative ring k is defined analogously, except the “ordinary” demand E˚ pptq “ E0 pptq “ k
599 is relaxed and the source category is taken to be G-Top.
600 Explicitly, we want a functor E˚ : G-Top ÝÑ E˚ -CGA satisfying the following axioms:

601 • G-homotopic maps f , g : pX, Aq ÝÑ pY, Bq induce equal ring homomorphisms E˚ f “ E˚ g.

602 • If pX, Aq P G-Top and A is a G-equivariant deformation retract of some neighborhood

603 of itself in X, then the map E˚ pX{A, A{Aq ÝÑ E˚ pX, Aq induced by the quotient map
604 pX, Aq ÝÑ pX{A, A{Aq “ pX{A, ptq is an isomorphism.

605 • There is a natural transformation δ : E˚ pX, Aq ÝÑ E˚`1 pAq between the graded k-Mod-
606 valued functors pX, Aq ÞÑ E˚ pX, Aq and pX, Aq ÞÑ E˚ pAq, fitting into the exact triangle

E˚ A o ED ˚ X

δ

E˚ pX, Aq,

607 where the other maps are the ring homomorphisms induced by the inclusions pA, ∅q ãÑ
608 pX, ∅q ãÑ pX, Aq.

609 • Given a collection pXα , Aα q of pairs and pX, Aq their disjoint union, the inclusions pXα , Aα q ãÝÝÑ
610 pX, Aq induce a ring isomorphism
„
ź
E˚ pX, Aq ÝÑ E˚ pXα , Aα q.
α

611 Candidate theories

612 In this section, we moot some possible equivariant cohomology theories to study before settling
613 on Borel equivariant cohomology for the remainder of the work. Again let G be a topological
614 group.
 Chapter 3. Cohomology theories 23

615 The cohomology of the naive quotient

Regular cohomology already does most of what we want, so a first idea might be to precompose
it with the orbit functor ´{G, yielding

E˚ : G-Top ÝÑ k-CGA,
X ÞÝÑ H ˚ pX{Gq.

616 This E˚ is a G-equivariant cohomology theory essentially because H ˚ is a cohomology theory;

617 the satisfaction of the axioms is immediately inherited from H ˚ and the fact that the maps in the
618 source category are G-equivariant. In addition, one has E˚ pptq “ H ˚ ppt{Gq “ H ˚ pptq “ k, so this
619 is an ordinary G-equivariant cohomology theory.
620 Unfortunately, this E˚ discards much useful information. For one thing, any nonempty G-
621 orbit has trivial G-equivariant cohomology under this theory. For another arguable failing, in
622 the standard example of G “ S1 acting on X “ S2 by rotation about a fixed axis, one has
623 X{G “ S2 {S1 « r´1, 1s » pt, so that E˚ pS2 q “ k. While it is informative that this action has trivial
624 E˚ -cohomology, one cannot help but feel some information has been lost in translation.

625 Bredon cohomology

626 N.B.: This subsection is just a sketch, included only because it seemed irresponsible not to note that “our”
627 equivariant cohomology is not that of primary interest in equivariant homotopy theory; the only use of this
628 material later in the dissertation is the cool but inessential (for us) equivariant Atiyah–Hirzebruch spectral
629 sequence of Proposition 3.3.12. Thus this subsection can be safely skipped by the uninclined.
630 To motivate the construction homotopy-theoretically informative theory, suppose first that
631 we are given a G-equivariant cohomology E˚ . As a G-action decomposes set-wise as a union of
632 orbits Gx « G{ StabG pxq, a first step to understanding E˚ might be to understand where it takes
633 the translation actions G ñ G{K for closed K ď G, or in other words, its restriction to the orbit
634 category G-Orbit. This restriction is a contravariant functor G-Orbit ÝÑ k-CGA or, alternately, a
635 covariant functor
M : SubpGq ÝÑ k-CGA
636 from the category of closed subgroups of G. Call such a functor a system of coefficients. If X is
637 a G–CW complex, so decomposed into G-cells D n ˆ pG{Kq with trivial G-action on D n , which
638 in turn equivariantly deformation retract to G{K, this system of coefficients should in principle
639 completely determine E˚ . In fact, the equivariant Mayer–Vietoris sequence, a consequence of the
640 axioms for a G-equivariant cohomology theory, implies the following theorem.

641 Proposition 3.2.1 ([AP93, Thm. 1.1.3, p. 8]). Let G be a compact Lie group and let E˚ and 1E˚ be
642 G-equivariant cohomology theories in the sense of Section 3.1, and suppose there exists a natural transfor-
643 mation η : E˚ ÝÑ 1E˚ which is an isomorphism when restricted to the orbit category G-Orbit. Then η is a
644 natural isomorphism.

645 The idea of Bredon cohomology is to start with a system of coefficients M concentrated in
646 degree 0 and from it produce a (unique) ordinary G-equivariant cohomology theory, “ordinary”
647 in the sense that E˚ pG{Kq “ E0 pG{Kq “ MpG{Kq for K P SubpGq, and unique in the sense of
648 the theorem above. The word “ordinary” is chosen because this is an equivariant version of the
649 dimension axiom: the orbits G{K are the equivariant version of points.
 Chapter 3. Cohomology theories 24

650 Recall from Section 1.2 that the maps of SubpGq are the inclusions and the G-conjugacies,
651 corresponding in the dual orbit category G-Orbit to equivariant quotient maps and to equivariant
652 homeomorphisms G{gKg´1 ÝÑ G{K given by xgKg´1 ÞÝÑ xgK, so that besides the groups
653 MpG{Kq, the functor M carries the data of maps MpG{Kq ÝÑ MpG{Hq for G ě H ě K and maps
654 MpG{Kq ÝÑ MpG{gKg´1 q which maps we will call, with acceptable ambiguity, Mpc g q.
655 Recall that a G–CW complex X is composed of G-cells en ˆ pG{Kq which are themselves unions
656 of cells en , the collection of which is called Celln pXq. If σ P Celln pXq, say σ “ en ˆ gK Ĺ en ˆ pG{Kq,
657 then the isotropy subgroup K “ Stab x is constant across points x P σ, and we can call this group
658 Stab σ. We define the G-equivariant cellular cochain group with coefficients in M to be
! ž ˇ )
n
CG pX; Mq :“ ϕ : Celln pXq ÝÑ MpG{Kq ˇ ϕpσq P MpG{ Stab σq, ϕpgσq “ Mpc g qϕpσq ,
ˇ
KPSubpGq

659 functions taking each n-cell to the coefficient group corresponding to the common orbit type of
660 its points, and equivariant with respect to the defining action of G on CellpXq and the conjuga-
661 tion action of G on M. The definition makes sense because Stab gσ “ gpStab σqg´1 . The relative
662 equivariant chains CGn pX, A; Mq are the subgroup vanishing on Celln pAq.
663 It follows from the definition that such a cochain ϕ is determined uniquely by its value at
664 one σ in each orbit Gσ, and in particular if there is one orbit Ge0 « G{K, one gets CG0 pG{K; Mq –

665 MpG{Kq as hoped. The differential is defined in tight analogy with the definition of the nonequiv-
666 ariant cellular coboundary map, but we will stop without getting any more explicit, satisfied that
667 with more work these groups yield an ordinary G-equivariant cohomology theory HGn pX, A; Mq.
668 Historical remarks 3.2.2. This construction is due to Bredon in the event that G is discrete. The
669 general definition for compact groups is apparently due to Matumoto [Mat73]. Bredon cohomol-
670 ogy is the equivariant cohomology theory most commonly used in equivariant homotopy theory.
671 Definitions nowadays tend to involve equivariant spectra and more homotopy-theoretic technol-
672 ogy than we use here, and replace the spheres Sαn in our definition Definition 1.3.4 of G–CW
673 complexes by representation-spheres, one-point compactifications of vector spaces carrying repre-
674 sentations of G. The author knows very little about this subject and dares not explore it further
675 here, but given that it is the main flavor of cohomology in modern equivariant homotopy theory,
676 it seemed it should at least be mentioned.

677 Borel equivariant cohomology

678 Having surveyed equivariant cohomology theories we are not interested in at present, for various
679 reasons, we settle on the variant we will care about.
680 Recall from Chapter 2 that given any topological group G, there exists a principal G-bundle
681 G Ñ EG Ñ BG such that EG is contractible, called a universal principal G-bundle. Given a left
682 G-space X, the product space EG ˆ X, equipped with the diagonal action, is another G-space
683 homotopy equivalent to X, but the new action is free since

pe, xq “ g ¨ pe, xq “ peg´1 , gxq ùñ e “ eg´1

684 and the action G ñ EG is free. Call X ÞÝÑ EG ˆ X the freeing functor; we can view it as a
685 projection G-Top ÝÑ G-Free. Now we can consider the orbit space of this new, free action. The
 Chapter 3. Cohomology theories 25

686 mixing diagram (Section 1.1) for EG and X is

EG o EG ˆ X /X

  
BG o EG ˆ X / X{G.
G

687 The new space on the bottom,

XG :“ EG ˆ X “ EG ˆ X peg, xq „ pe, gxq,

L
G

688 is called the homotopy quotient of X by G or the Borel construction [BBF+60, Def. IV.3.1, p. 52].
689 This space can be viewed as a sort of completion of X{G, and its ordinary cohomology will be,
690 at least nominally, the subject of most of our researches. Note

ptG “ EG ˆ pt « EG{G “ BG.

G

691 Definition 3.2.3. (Borel [BBF+60, IV.3.3, p. 53]) The singular cohomology H ˚ pXG ; kq of the homo-
692 topy quotient XG is the (Borel) equivariant cohomology HG˚ pX; kq of the action of G on X. The
693 diagonal G-orbit of px, eq P X ˆ EG is denoted rx, esG . The equivariant cohomology of a point is
694 HG˚ :“ HG˚ pptq “ H ˚ pBGq.

695 As suggested by the fact that the freeing functor can be seen as a projection, the homotopy
696 quotient construction does not differ essentially from the regular quotient construction if the
697 original action was free.

698 Proposition 3.2.4. Let G act freely on a CW complex M. Then the homotopy quotient MG is homotopy
699 equvalent to the orbit space M{G.

700 Proof. The map ρ : re, xs ÞÑ Gx from MG Ñ M{G has fiber EG{ Stabpxq in general. If G acts freely
701 on M, then all fibers are EG. Since EG is contractible, the long exact homotopy sequence of the
702 bundle EG Ñ MG Ñ M{G shows ρ is a weak homotopy equivalence. By Whitehead’s theorem, ρ
703 is a homotopy equivalence.

As pointed out in (B.1), given a G-space X, the mixing construction yields an X-bundle X Ñ
XG Ñ BG associated to the principal G-bundle G Ñ EG Ñ BG. Explicitly, the projection is

XG ÝÑ BG,
re, xsG ÞÝÑ eG.

704 Definition 3.2.5. The bundle X Ñ XG Ñ BG described above is the Borel fibration of the action
705 of G on X.

706 The Atiyah–Hirzebruch–Leray–Serre spectral sequence

707 Now that we have discussed fiber bundles and cohomology theories, we discuss the behavior of
708 cohomology theories, including equivariant cohomology, on bundles.
 Chapter 3. Cohomology theories 26

709 Notation 3.3.1. Where possible, we write the total space of a fiber bundle as E, but in situations
710 when we also have a cohomology theory E˚ and a spectral sequence pEr‚,‚ q to deal with, we
711 regretfully name the total space X.
π
712 Given a fiber bundle F Ñ X Ñ B with B a CW complex, let B p be the p-skeleton of B. Then
713 pB p q is an increasing topological filtration of B and pX p q :“ pπ ´1 B p q an increasing topological
714 filtration of X. Define X p “ B p “ ∅ for p ă 0. Suppose we are also given a multiplicative
715 cohomology theory E˚ . Then associated to each pair pX, X p q is a long exact sequence E˚ pX, X p q Ñ
716 E˚ X Ñ E˚ X p . Because X p Ď X p`1 , each map E˚ X ÝÑ E˚ X p factors through E˚ X p`1 , so the
717 topological filtration pX p q leads to a decreasing algebraic filtration

Fp E˚ X “ kerpE˚ X ÝÑ E˚ X p´1 q

718 of E˚ X. The occurrence of both p and p ´ q here might be jarring, but this index shift is necessary
719 to assure that F0 E˚ X “ E˚ X. Assume for convenience that π1 B acts trivially on E˚ pFq. Then
720 turning the crank of the associated filtration spectral sequence of Corollary A.5.4, one arrives at
721 the following.

Theorem 3.3.2 (Atiyah–Hirzebruch–Leray–Serre [AH61, Rmk. 2.2]). Let F Ñ X Ñ B be a fiber

bundle and E˚ a multiplicative cohomology theory such that π1 B acts trivially on E˚ pFq. There exists a
right half-plane spectral sequence pEr , dr q with
p,q
E2 “ H p B; Eq pFq ,
` ˘
p,q
E8 “ gr p E p`q pXq.

722 This is a simultaneous generalization of the following spectral sequences.

Theorem 3.3.3 (Atiyah–Hirzebruch [AH61, Thm. 2.1]). Let E˚ be a multiplicative cohomology theory
and X a topological space. There exists a right half-plane spectral sequence pEr , dr q with
p,q
E2 “ H p X; Eq pptq ,
` ˘
p,q
E8 “ gr p E p`q pXq.

723 Proof. Take F “ pt and X “ B in Theorem 3.3.2.

Theorem 3.3.4 (Serre [McC01, Theorem 5.2, p. 135]). Let F Ñ E Ñ B be a fibration such that π1 B
acts trivially on H ˚ pFq. There exists a first-quadrant spectral sequence pEr , dr q with
p,q
E2 “ H p B; H q pFq ,
` ˘
p,q
E8 “ gr p H p`q pEq.

724 If H ˚ pFq is a free k-module (for example, if k is a field), we may also write E2 – H ˚ pBq b H ˚ pFq. Further,
725 this construction is functorial in that a map of bundles induces a map of spectral sequences.

726 Proof. Take E˚ “ H ˚ (and write E for X) in Theorem 3.3.2 and apply Theorem B.2.1.

727 Definition 3.3.5. The spectral sequence of Theorem 3.3.4 is the Serre spectral sequence.

728 Critically, this formulation applies to principal bundles.

 Chapter 3. Cohomology theories 27

729 Proposition 3.3.6. Let G be a path-connected group. If G Ñ E Ñ B is a principal G-bundle, then π1 B

730 acts trivially on H ˚ pGq.
731 Proof. Because the transition functions are given by right multiplication by elements of G, and
732 this action of G on H ˚ pGq is trivial since G is path-connected, it follows the action of π1 B on
733 H ˚ pGq is trivial.

734 It is important to us to be able to identify the maps in cohomology induced by fiber inclusion
735 and projection to the base.
i π
736 Proposition 3.3.7. Let F Ñ E Ñ B be a fibration such that π1 B acts trivially on H ˚ pFq. The fiber pro-
0,‚
737 jection i˚ : H ˚ pEq ÝÑ H ˚ pFq is realized by the left-column map E8 ‚,‚
 E8 ãÑ E20,‚ in Theorem 3.3.4: to
738 wit, we can write
„ „
gr‚ H ˚ pEq ÝÑ E8 ‚,‚
ÝÝ E8 0,‚
ãÝÝÑ E20,‚ ÝÑ H ˚ pFq.
739 Likewise, the base lift π ˚ : H ˚ pBq ÝÑ H ˚ pEq is realized by the bottom-row map E2‚,0  E8
‚,0 ‚,‚
ãÑ E8 :
„ „
H ˚ pBq ÝÑ E2‚,0 ÝÝ E8
‚,0 ‚,‚
ãÝÝÑ E8 ÝÑ gr‚ H ˚ pEq.

740 Proof [McC01, p. 147]. We have a commutative square

F F / pt

 
F
i /E π /B

 
pt /B B

741 where each column (and row) is a fibration, with the original fibration in the middle column, and
742 the maps between columns are fiber-preserving. These maps induce maps of spectral sequences,
743 which we can denote as
F
Er ÐÝ Er ÐÝ B Er .
744 The middle spectral sequence is the Serre spectral sequence ` of the
˘ original fibration, while F Er
745 is that of F Ñ F Ñ pt, which collapses at F E2 “ H ˚ pt; H ˚ ˚ B
` pFq˚ “ ˘H pFq,˚and Er is that of
746
B ˚
pt Ñ B Ñ B, which also collapses instantly, at E2 “ H B; H pptq “ H pBq.` On E2 pages, ˘
747 the`induced ˘maps are E2 pi˚ q : E2 ÝÑ F E2 , which is the left-column projection H ˚ B; H ˚ pFq ÝÑ
748 H 0 `B; H ˚ pFq˘ – H ˚ pFq, and E2 pπ ˚ q : B E2 ÝÑ E2 , which is the bottom-row inclusion H ˚ pBq ÝÑ
749 H ˚ B; H 0 pFq , the maps we would like to descend to the maps i˚ “ gr‚ i˚ and π ˚ “ gr‚ π ˚ on
750 E8 pages. The maps between E8 pages are

H ˚ pFq
O
„

gr‚ i˚ gr‚ i˚
FE o gr‚ H ˚ pEq o BE
2 O 2
„

H ˚ pBq,

751 by the fact that the isomorphism of final page E8 with gr‚ H ˚ pEq is natural. But that shows that
752 these maps descend from the E2 column and row maps as claimed.
 Chapter 3. Cohomology theories 28

π
753 Remark 3.3.8. In the event the fibration F Ñ E Ñ B is in fact a fiber bundle, as it will be in all
754 cases that actually concern us, the Serre spectral sequence is isomorphic from E2 on to the Leray
755 spectral sequence of the map π, which we will introduce in Appendix D.2 to complete our account
756 of Borel’s original 1953 proof of Theorem 7.1.12.
757 We have stated Serre’s theorem for singular simplicial cohomology, but he initially stated it
758 for singular cubical homology and cohomology, and it goes through essentially unchanged for
759 Alexander–Spanier cohomology, Čech cohomology, or cohomology with APL -cochains as we will
760 use in Theorem 7.1.3.

761 Corollary 3.3.9. Let F Ñ E Ñ B be a fibration such that the action of π1 B on H ˚ pFq is trivial and
762 H ˚ pFq is a free k-module. Then the fiber inclusion F ãÝÝÑ E is surjective in cohomology if and only if the
763 spectral sequence of the bundle collapses at E2 .

764 Proof. Recall from the last remark that the fiber projection H ˚ pEq ÝÑ H ˚ pFq can be realized as
0,‚
765 E8  E8 ãÑ E20,‚ . This map is surjective if and only if E8
0,‚
“ E20,‚ , which in turn means that
766 E20,‚ “ E2 X ker d2 “ E3 X ker d3 and so on: all differentials vanish on the left column H ˚ pFq.
767 But all differentials from the left column vanish if and only if the sequence collapses at E2 . The
768 “if” implication is clear. For the “only if,” note that by our assumptions, E2 – H ˚ pBq bk H ˚ pFq,
769 and d2 vanishes on H ˚ pBq by lacunary considerations, so since d2 is an antiderivation, d2 “ 0.
770 But this means E3 – H ˚ pBq b H ˚ pFq, and since d3 vanishes on H ˚ pFq “ E30,‚ and H ˚ pBq, one has
771 d3 “ 0. By induction, E8 – E2 .

772 Corollary 3.3.10. Let F Ñ E Ñ B be a fibration such that the action of π1 B on H ˚ pFq is trivial and
773 H ˚ pFq is a free k-module. Suppose further that F and B are of finite type. Then the Poincaré polynomials
774 satisfy
ppEq ď ppBqppFq,
775 (in the sense that each coefficient of ppBqppFq ´ ppEq is nonnegative) with equality if and only if the fiber
776 inclusion F ãÝÝÑ E is surjective in cohomology.

777 Proof. We have E2 “ H ˚ pBq b H ˚ pFq in the Serre spectral sequence of F Ñ E Ñ B, so ppE2 q “
p,q
778 ppBqppFq. The rank of each Er , and hence the Poincaré polynomial, can only decrease by E8 ,
779 and it can only fail to decrease if E2 – E8 ; that is the case if and only if H ˚ pEq ÝÝ H ˚ pFq, by
780 Corollary 3.3.9.

781 Corollary 3.3.11. Let F Ñ E Ñ B be a fibration such that the action of π1 B on H ˚ pFq is trivial and
782 H ˚ pBq and H ˚ pFq are both concentrated in even degrees. Then the spectral sequence collapses at E2 .

783 Proof. If H ˚ pBq and H ˚ pFq are both concentrated in even degrees, then so is the tensor product
784 E2 “ H ˚ pBq b H ˚ pFq concentrated in even total degree. Since the differentials dr increase total
785 degree by 1, mapping from even diagonals to odd and vice versa, they must all be trivial, so the
786 sequence collapses at E2 .

787 The Serre spectral sequence allows a vast generalization of the covering result Proposi-
788 tion B.3.5.

789 Proposition 3.3.12. Let F Ñ E Ñ B be a fiber bundle such that the action of π1 B on H ˚ pFq is trivial
790 and h‚ pBq and h‚ pFq are finite. Then the Euler characteristics of these spaces satisfy χpEq “ χpFqχpBq.
 Chapter 3. Cohomology theories 29

791 Proof. Consider E2 “ H ˚ pBq b H ˚ pFq as a single complex with degpH p B b H q Fq “ p ` q. With
792 this grading, χpE2 q “ χpBq ¨ χpFq. By repeated application of Proposition A.2.1, one finds

χpE2 q “ χpE3 q “ ¨ ¨ ¨ “ χpE8 q “ χpEq.

793 Finally, the construction that we undertook with G–CW complexes makes available a spectral
794 sequence due to Matumoto that expresses equivariant cohomology in terms of singular cohomol-
795 ogy.
796 Theorem 3.3.13 (Equivariant Atiyah–Hirzebruch [Mat73]). Let G be a compact Lie group and E˚ a
797 G-equivariant cohomology theory with values in k-CGA, and pX, Aq a G–CW pair. Then there exists a
798 spectral sequence of k-CGAs, functorial in pX, Aq, such that
p,q
799 • E1 – E p`q pX p , X p´1 Y Aq,
‚,q
800 • E2 – HG˚ pX, A; Eq q, the Bredon cohomology with M “ Eq coefficients, as discussed in Sec-
801 tion 3.2.2,
p,q
802 • E8 “ gr p E p`q pX, Aq.
` ˘
803 Here the filtration on E˚ pX, Aq is given by F p E˚ pX, Aq “ ker E˚ pX, Aq Ñ E˚ pX p´1 , Aq .
804 Remarks 3.3.14. (a) Although we will also have occasion to invoke the spectral sequence of a
805 filtered DGA again in Section 7.1.2, from here on out, “spectral sequence” simpliciter will connote
806 the cohomological Serre spectral sequence of a bundle. This construct will be deployed with
807 sufficient frequency that we allow ourselves also to abbreviate it SSS.
808 (b) These spectral sequences apply more generally in the instance that π1 B fails to act trivially
809 on E˚ pFq, with the concession that the coefficients E˚ pFq must instead be taken to be a sheaf of
810 groups or, at the most concrete, viewed as a krπ1 Bs-module.

811 The transgression in the Serre spectral sequence

812 We will make extensive use of the transgression in the Serre spectral sequence of a bundle in
813 what follows. On the E2 level, an edge homomorphism dr takes (a submodule of) H r´1 pFq to (a
814 quotient of) H r pBq, but it seems worth stating more explicitly what this means on the cochain
815 level, so we put forth here two slightly more topologically explicit descriptions.
i π
816 Proposition 3.3.15. Let F Ñ E Ñ B be a fiber bundle such that the action of π1 B on H ˚ pFq is trivial.
q,0
817 An element rzs P H q´1 pFq (Definition A.5.13) transgresses to the image of rbs P H q pBq in Eq if and only
818 if there exists c P C q´1 pEq in the singular cochain group such that i˚ c “ z and δc “ π ˚ b. This is the
819 picture:
i˚ / Z q´1 pFq  0/ z
C q´1 pEq _c
δ
τ
t  
x
Z q pBq / Z q pEq, b / δc.
π˚

820 Sketch of proof. Recall that the Serre spectral sequence is the filtration spectral sequence associated
821 to the filtration of the simplicial cochain algebra C˚ pEq given by

Fp C˚ pEq “ ker C˚ pEq ÝÑ C˚ pπ ´1 B p´1 q ,

` ˘
 Chapter 3. Cohomology theories 30

822 the kernels of the restriction maps to the subbundles π ´1 B p of E lying over the p-skeleta B p of the
823 base space B.1 The elements of the complement Fp H ˚ pEqzFp´1 H ˚ pEq, then, are classes represented
824 by cocycles which vanish over the p-skeleton of B but not over its p-cells. Morally, these classes
825 are carried by the portion of E lying over the p-cells of B. Elements of F0 H ˚ pEq are thus already
826 carried by π ´1 B0 « B0 ˆ F, and arise from H ˚ pFq, while elements in Fp C p pEqzFp`1 C p pEq, on the
827 other hand, are those p-cycles that are only nontrivial on singular chains with all p “directions”
828 arising from the base B, and thus come from p-cycles in the base. At the E2 level, classes in
p,0
829 H ˚ pEq are arbitrary, while a class in Fp H p pEq is represented by something in the square E2 –
830 H p pFq b H 0 pFq – H p pFq.
831 Let z P Z q´1 pFq be a cocycle in the fiber. It extends (say by zero) to a cochain c P C q pEq, which
832 need not be a cocycle. Because the differentials dr in the Serre spectral sequence arise from the
833 singular coboundary map δ through the filtration, dr pcq is defined in Er if and only if δ pushes
834 c forward by r steps in the filtration. Thus in this case δ takes a “fiber” cocycle c to a “base”
835 cocycle. ` ˘
836 If δc survives to Eq , then in E2 , it is represented by a class in Fq H ˚ pBq b H ˚ pFq “ H q pBq b
837 H 0 pFq – H q pBq, so δc is in the image of π ˚ : Z q pBq ÝÑ Z q pEq.

838 There is another way of explaining this which may be more illuminating and doesn’t require
i π
839 us to work at the cochain level. Recall from Theorem B.2.4 that associated to a bundle F Ñ E Ñ B
840 is an exact triangle of homotopy groups
deg ´1
π˚ pFq ÝÑ π˚ pEq ÝÑ π˚ pBq ÝÝÝÝÑ π˚ pFq.

841 Thus there is a degree-shifting map linking the homotopy groups of the base and fiber. Viewing
842 F “ E|pt as a specific fiber over a point pt P B, this sequence can be seen to arise from a long
843 exact sequence of relative homotopy groups associated to the pair pE, Fq:
deg ´1
π˚ pFq ÝÑ π˚ pEq ÝÑ π˚ pE, Fq ÝÝÝÝÑ π˚ pFq.

844 Because E is locally trivial, a map of pairs pD n , Sn´1 q ÝÑ pB, ptq can be lifted to a map pD n , Sn´1 q ÝÑ
845 pE, Fq, and taking homotopy classes, a bit of work shows π˚ pE, Fq – π˚ pB, ptq in such a way that
846 the two long exact sequences can be identified. The long exact sequence of a pair
deg `1
H ˚ pFq ÝÝÝÝÑ H ˚ pE, Fq ÝÑ H ˚ pEq ÝÑ H ˚ pFq.

847 is one of the Eilenberg–Steenrod axioms, but it no longer will do in general to substitute H r ˚ pBq “
848 H ˚ pB, b0 q for H ˚ pE, Fq. If it did, we would have the desired degree-shifting cohomological map
849 linking the base and the fiber. Nevertheless, π is a map of pairs pE, Fq ÝÑ pB, b0 q, so one has map
850 of long exact sequences
˚
¨¨¨ / H q´1 pFq δ / H q pE, Fq / H q pEq i / H q pFq / ¨¨¨
O O O O
π˚ π˚

¨¨¨ / H q pptq / H q pB, ptq „ / H q pBq / H q pptq / ¨¨¨

851 Proposition 3.3.16. The transgression is given by the composition pπ ˚ q´1 ˝ δ where defined.
1 Again, the surprising p ´ 1 is to ensure that F0 C˚ pEq “ C˚ pEq.
 Chapter 3. Cohomology theories 31

852 Proof. If z P Z q´1 pFq is a cocycle representing rzs P H q´1 pFq, then δrzs P H q pE, Fq is by definition
853 the class of δc for any cochain c P C q pEq such that i˚ c “ z. Such a class may or may not be the
854 image under π ˚ of some rbs P H q pB, b0 q “ H q pBq, but if it is, then the elements pz, c, bq satisfy
855 exactly the specification put forth in Proposition 3.3.15.

856 Thus the transgressed classes in H q´1 pFq can be seen as the images of the connecting homo-
857 morphism η “ pπ ´1 q˚ ˝ δ in a fictitious long exact sequence
η i˚
H ˚ pFq ÝÑ H ˚ pBq ÝÑ H ˚ pEq ÝÑ H ˚ pFq

858 of a bundle corresponding to the long exact sequence of homotopy groups. The transgressive
859 elements are morally those for which such a sequence holds.
860 Remark 3.3.17. A real proof of this characterization of the transgression in the Serre spectral
861 sequence can be found in McCleary [McC01, Thm. 6.6, p. 186].
862 There is an analogous Serre spectral sequence of a bundle in homology, whose differentials are
863 of degree p´r, r ´ 1q, and a (partially defined) transgression Hr pBq ÝÑ Hr´1 pBq. Dually to our
864 definition in cohomology, the transgressed elements of Hq F are images of transgressive elements
865 of Hq`1 B under an incompletely-defined map τ˚ in a fictitious long exact sequence
τ˚
H˚ pBq ÝÑ H˚ pFq ÝÑ H˚ pEq ÝÑ H˚ pBq.

866 Because the Hurewicz homomorphism π˚ pX, Aq ÝÑ H˚ pX, Aq from homotopy groups to
867 homology groups discussed in Theorem B.2.6 is natural, it pieces together into a map from the
868 homotopy long exact sequence of a pair pE, Fq to the homology long exact sequence of that pair.
869 It follows from the existence of this map of long exact sequences and the long exact homotopy
870 sequence of a bundle (Theorem B.2.4) that everything in the image of the Hurewicz map π˚ F ÝÑ
871 H˚ F is the image of the transgression in every fibration with fiber F, a fact we will have cause to
872 comment on again in Section 6.6. Moreover, the cohomology transgression τ : H q´1 pFq ÝÑ H q pBq
873 and the homology transgression τ˚ : Hq pBq ÝÑ Hq´1 pFq are dual [Ral].

874 A natural lemma on bundles

875 In this section, we use the Serre spectral sequence to prove a lemma on cohomology of bundles
876 we will use repeatedly to good effect. It seems analogous to the Leray–Hirsch theorem that if
877 F Ñ E Ñ B is a bundle such that H ˚ pEq Ñ H ˚ pFq is surjective, then H ˚ pEq – H ˚ pBq b H ˚ pFq as an
878 H ˚ pBq-module. There is a proof by Larry Smith [Smi67, Cor. 4.4, p. 88] using the Eilenberg–Moore
879 spectral sequence as well as the SSS, but the author found the result independently through the
880 following proof, so it has pride of place.
881 Let F be a topological space and ξ 0 : E0 Ñ B0 an F-bundle. From the category of F-bundles
882 and F-bundle maps, we can form a slice category F-Bun{ξ 0 of F-bundles over ξ 0 as follows. An
883 object of F-Bun{ξ 0 is an F-bundle ξ equipped with a bundle map ξ Ñ ξ 0 ; a morphism between
884 objects ξ 1 Ñ ξ 0 and ξ Ñ ξ 0 is a bundle map ξ 1 Ñ ξ making the expected triangle commute. Such
 Chapter 3. Cohomology theories 32

885 a map entails the following commuting prism:

f1
E1 /E /+ E0
h f

ξ1 ξ ξ0 (3.1)
  f¯ 
B1
h̄ /B /3 B0 .
f1

Note that the maps between total spaces yield two functors

F-Bun{ξ 0 ÝÑ H ˚ pE0 q-CGA :

pE Ñ Bq ÞÝÑ H ˚ pEq;
pE Ñ Bq ÞÝÑ HpBq b H˚ pB0 q H ˚ pE0 q.

886 If H ˚ pE0 q ÝÑ H ˚ pF0 q is surjective, we claim these functors are naturally isomorphic.

887 Theorem 3.4.1. Let ξ 0 : E0 Ñ B0 be an F-bundle such that the fiber inclusion F ãÝÝÑ E0 is H ˚ -surjective,
888 such that H ˚ pFq is a free k-module, and such that π1 B0 acts trivially on H ˚ pFq. Then the fiber inclusions
889 of all F-bundles over ξ 0 are H ˚ -surjective, and there is a natural ring isomorphism
„
H ˚ pEq ÐÝ H ˚ pBq b H˚ pB0 q H ˚ pE0 q

890 of functors F-Bun{ξ 0 ÝÑ H ˚ pE0 q-CGA. Diagrammatically, the commutative diagram (3.1) gives rise to

h˚
H ˚ pE1 q o H ˚ pEq
O O
„

H ˚ pB1 q b H˚ pB0 q H ˚ pE0 q o H ˚ pBq b H˚ pB0 q H ˚ pE0 q.

h̄˚ b id

891 Verbally, if a fiber inclusion is surjective in cohomology, then cohomology takes pullbacks to
892 pushouts.

893 Proof. By the definition of a bundle map, the fiber inclusion F ãÝÝÑ E0 factors as F ãÑ E Ñ
894 E0 , so the assumed surjectivity of H ˚ pE0 q Ñ H ˚ pEq Ñ H ˚ pFq implies surjectivity of the factor
895 H ˚ pEq ÝÑ H ˚ pFq.
896 Because of these surjections, the spectral sequences of these bundles stabilize at their E2 pages
897 by Corollary 3.3.9. Applying H ˚ to the right square of the assemblage (3.1) yields

f˚ f¯˚ b id
H ˚ pEq o H ˚ pE0 q H ˚ pBq b H ˚ pFq o H ˚ pB0 q b H ˚ pFq
O O O O
ξ˚ ξ 0˚ id b 1 id b 1
which manifests on the E2 page as
H ˚ pBq o H ˚ pB0 q, H ˚ pBq o H ˚ pB0 q.
f¯˚ f¯˚
 Chapter 3. Cohomology theories 33

The commutativity of the left square means there is an induced map of rings

H ˚ pBq b H˚ pB0 q H ˚ pE0 q ÝÑ H ˚ pEq,

bbx ÞÝÑ ξ ˚ pbq f ˚ pxq,

898 whose E2 manifestation is the canonical H ˚ pBq-module isomorphism

“ ‰ „
H ˚ pBq b H˚ pB0 q H ˚ pB0 q b H ˚ pFq ÝÑ H ˚ pBq b H ˚ pFq.

899 Since this E2 map is a bijection, the ring map is an H ˚ pE0 q-algebra isomorphism.
900 For naturality, note that the ring map h˚ : H ˚ pEq ÝÑ H ˚ pE1 q is completely determined its
901 restrictions to its tensor-factors H ˚ pBq and H ˚ pE0 q. The left square and top triangle of (3.1) imply
902 the commutativity of the squares

h˚ h˚
H ˚ pE1 q o H ˚ pEq H ˚ pE1 q o H ˚ pEq
O O O O
pξ 1 q˚ ξ˚ p f 1 q˚ f˚

H ˚ pB1 q o H ˚ pBq, H ˚ pE0 q H ˚ pE0 q,

h̄˚

903 so that these factor maps are respectively h̄˚ : H ˚ pBq ÝÑ H ˚ pB1 q and id H˚ pE0 q .
904 Chapter 4
905 The cohomology of complete flag manifolds

906 The algebraic relation between a compact group and its maximal torus informs all discussion of
907 invariant subalgebras going forward, and is epistemologically prior to much of our discussion
908 of the cohomology of homogeneous spaces, being treated with sui generis methods that do not
909 apply in the general case.
910 The quotient G{T of a compact, connected Lie group by its maximal torus T, called a com-
911 plete flag manifold, was among the first homogeneous spaces other than groups and symmetric
912 spaces whose cohomology was understood. This material will be cited at least in Section 5.2,
913 Section 7.3.2, and Theorem B.4.9, and does not fit well anywhere else, so we propound it now. It
914 is fundamental, and but for the discussion of the Serre spectral sequence in Theorem 3.3.4, could
915 have gone earlier.

916 The cohomology of a flag manifold

917 The cornerstone result is the following.

918 Theorem 4.1.1. Let G be a compact, connected Lie group and T a maximal torus in G. Then the cohomol-
919 ogy of H ˚ pG{Tq is concentrated in even dimensions.

920 Proof sketch 1. Associated to G is a complexified Lie group GC which is a complex manifold, and
921 which contains a Borel subgroup B, a complex Lie group containing T and such that

GC {B « G{T.

922 Thus G{T admits a complex manifold structure and hence a CW structure with even-dimensional
923 cells.

924 We will use consequences of this theorem in such a critical way in Section 5.2, Theorem B.4.9,
925 and Section 7.3.2 that we would feel somewhat guilty only sketching this proof, so we repro-
926 duce Borel’s original 1950 proof. This argument was first published somewhat telegraphically
927 in Leray’s contribution [Ler51] to the 1950 Bruxelles Colloque, and is elaborated in Borel’s the-
928 sis [Bor53]. It invokes two facts we shall not prove about invariant differential forms, which are
929 these.

930 Proposition 4.1.2. Suppose a compact Lie group G acts on a manifold M. Then every cohomology class in
931 H ˚ pM; Rq is represented by a G-invariant differential form ω, which is determined uniquely by its value
932 ωx P ΛTx˚ M, an alternating multlinear form on the tangent space of one point x of M.

34
 Chapter 4. The cohomology of complete flag manifolds 35

933 Proposition 4.1.3. Let G be a compact Lie group and K a closed subgroup. The alternating multlinear
934 form ωx P Λpg{kq_ representing a G-invariant form ω P ΩpG{Kq is invariant under the action Ad˚ K of
935 K induced by the conjugation action on K on G.

936 Borel’s proof of Theorem 4.1.1. By Theorem B.2.1, we may use R coefficients. Write ` “ rk G and
937 n “ dim G ´ rk G. We prove the result by a double induction on ` and n. If ` “ 0, then G is
938 discrete, and we are done. Inductively suppose we have proven the result for all groups of rank
939 ` ´ 1. If n “ 0, then rk G “ dim G, so G “ T is a torus and we are done.
940 Now suppose inductively we have proven the result for ` and n ´ 1. Note that without loss
941 of generality, by Theorem B.4.4, G can be taken to be of the form A ˆ K with A a torus and K
942 simply-connected. Since A is a factor of the maximal torus T of G, one has G{T “ K{pT X Kq,
943 and rk K “ rk G ´ rk A, so we are done by induction on ` unless rk A “ 0 and G “ K is simply-
944 connected.
945 So assume G is simply-connected. We claim there exists an element x P G such that x R
946 ZpGq and x2 P ZpGq. To see this, let x P G, and recall from Proposition B.4.11 that ZpGq is the
947 intersection of all maximal tori in G, so in a given maximal torus T one can always find y with
n
948 y2 “ x, but if these lay in all tori for all n, then ZpGq would fail to be discrete. Let K be the
949 identity component of the centralizer ZG pxq of x. Because x lies in the maximal torus T of G, we
950 know rk K “ rk G, and because x R ZpGq, the dimension dim ZG pxq “ dim K is strictly less than
951 dim G. Thus H ˚ pK{Tq is evenly graded by the inductive assumption.
952 The tangent space g{k “ T1K pG{Kq to the identity coset 1K in G{K can be identified with an
953 orthogonal complement kK to k in g in such a way that the isotropy action of K on T1K pG{Kq
954 corresponds to the adjoint action of K on kK .
955 By Proposition 4.1.2, each de Rham cohomology class on G{K contains a left G-invariant
956 element, which is then determined by its restriction to T1K pG{Kq – kK . Such a restriction is, by
957 Proposition 4.1.3, an alternating pAd Kq-invariant multilinear form on kK . Because x2 is central,
958 Adpxq P GLpgq is an involution; thus g splits as the 1-eigenspace k and an orthogonal p´1q-
959 eigenspace, which must be kK . Since Adpxq acts as multiplication by ´1 on kK , a nonzero Adpxq-
960 invariant alternating form on kK can only have even degree. As x P K, it follows we must have
961 H ˚ pG{Kq concentrated in even degree.
962 Now we can apply the Serre spectral sequence to K{T Ñ G{T Ñ G{K. Both H ˚ pK{Tq and
963
˚
H pG{Kq are evenly-graded, so by Theorem 3.3.4, so also is G{T. In fact, by Corollary 3.3.11, the
964 spectral sequence collapses at E2 and H ˚ pG{Tq – H ˚ pG{Kq b H ˚ pK{Tq as an H ˚ pK{Tq-module.
965

966 Corollary 4.1.4. Let G be a compact, connected Lie group and T a maximal torus in G. Then the Euler
967 characteristic of χpG{Tq is positive.

968 The acyclicity of G{NG pTq

969 In this section we prove another result whose importance will not immediately be clear, but
970 which recurs in Section 5.4 and Section 9.1.

971 Proposition 4.2.1. Let G be a compact, connected Lie group, T a maximal torus in G, and N “ NG pTq
972 the normalizer. Then dim G{N is even and G{N is Q-acyclic:

H ˚ pG{N; Qq “ H 0 pG{N; Qq – Q.
 Chapter 4. The cohomology of complete flag manifolds 36

973 Proof [MT00, Thm. 3.14, p. 159]. The torus T acts on G{N on the left, fixing the identity coset 1N
974 (since T ď N); we claim this is the only such fixed point. Indeed, let t P T be a topological gener-
975 ator. If an element gN P G{N is fixed under multiplication by t, it is fixed under multiplication
976 by all powers of t, and thus, by continuity, by all of T, so that TgN “ gN, or g´1 Tg ď N. Since T
977 is a connected component of N and 1 “ g´1 1g P T, it then follows g´1 Tg “ T, or g P N, so that
978 gN “ 1N is the lone fixed point.
979 Let dim nK “ m. Because T fixes 1N, it induces a T-action on the tangent space g{n “
980 T1N pG{Nq to G{N at the identity coset 1N, which can also be seen as the normal subspace
981 nK ă g orthogonal to n in the tangent space to G at 1. Because T acts by isometries on nK , it
982 leaves invariant ε-spheres Sm´1 about the origin 0 P nK . The exponential exp : nK ÝÑ G{N will
983 map a sufficiently small sphere isometrically and T-equivariantly into G{N, and this T-invariant
984 image sphere Sm´1 divides G{N into a T-invariant disk D m and a T-invariant complement M.
985 Since T is path connected, the map `t is homotopic to the identity, so χp`t q “ χpidq on both Sm´1
986 and M. As only 1N P G{N is fixed by multiplication by T, and this point lies in the interior of
987 D m , it follows `t acts without fixed points on Sm´1 and M. By the Lefschetz fixed point theorem
988 Theorem B.2.11, then,
χpMq “ χpSm´1 q “ 0.
989 It follows m is even. Note that by excision H ˚ pG{N, Mq – H ˚ pD m , Sm´1 q – H
r ˚ pSm q, so that
990 the relative Euler characteristic χpG{N, Mq is p´1qm “ 1. The long exact sequence of the pair
991 pG{N, Mq then gives
χpG{Nq “ χpMq ` χpG{N, Mq “ 0 ` 1 “ 1.
992 As G{T Ñ G{N is a finite cover with fiber W “ N{T and H odd pG{Tq “ 0 by Theorem 4.1.1, it
993 follows from Proposition B.3.1 that

H odd pG{Nq – H odd pG{TqW “ 0.

994 Thus h‚ pG{Nq “ χpG{Nq “ 1, so it must be that H ˚ pG{Nq “ H 0 pG{Nq – Q.

995 We have the following useful corollary, which will reemerge much later in Proposition C.3.3
996 as a consequence of the Berline–Vergne/Atiyah–Bott localization theorem.

997 Corollary 4.2.2 (Weil). Let G be a compact, connected Lie group, T a maximal torus in G, and W the
998 Weyl group of G. Then
χpG{Tq “ |W|.

999 Proof. Since G{T ÝÑ G{N is a |W|-sheeted covering and χpG{Nq “ 1 by Proposition 4.2.1, it
1000 follows from Proposition B.3.5 that

χpG{Tq “ χpG{Nq ¨ |W| “ |W|.

1001 This means in a homogeneous space G{K, one can for cohomological purposes replace K with
1002 the normalizer of its maximal torus.

1003 Corollary 4.2.3. Let G be a compact, connected Lie group, K a closed, connected subgroup of lesser rank,
1004 S a maximal torus of K, and N “ NK pSq the normalizer of this torus in K. Then the natural projection
1005 G{N ÝÑ G{K induces a ring isomorphism
„
H ˚ pG{Kq ÝÑ H ˚ pG{Nq.
 Chapter 4. The cohomology of complete flag manifolds 37

1006 Proof. There is a fiber bundle K{N Ñ G{N Ñ G{K, whose fiber K{N is acyclic by Proposi-
1007 tion 4.2.1, so π1 pG{Kq acts trivially on H ˚ pK{Nq “ H 0 pK{Nq – Q, and the Serre spectral sequence
1008 of this bundle collapses instantly at the E2 page, yielding an H ˚ pG{Kq-module isomorphism
gr‚ H ˚ pG{Nq “ H ˚ pG{Kq b Q – H ˚ pG{Kq.
1009 Because the bigraded algebra H ˚ pG{Nq is concentrated in the bottom row, the associated graded
1010 construction leaves it unchanged, so this is a ring isomorphism.

1011 There is also the following further result, later generalized by Chevalley.
1012 Corollary 4.2.4 (Leray). The ring H ˚ pG{Tq is isomorphic to the regular representation of the Weyl group
1013 W.
` ˘
Proof. One characterization of the regular representation W Aut QrWs of a group W is
through the character w ÞÝÑ tr w|QrWs of the representation: a representation L is W-isomorphic
to the regular representation just if
#
|W| w “ 1,
tr w| L “
0 w ‰ 1.
1014 Consider the standard right action1 of W “ NG pTq on G{T given by gT ¨ nT :“ gnT. Since
gnT “ gT ðñ nT “ g´1 gT “ T ðñ n P T,
1015 no element of W other than the identity has any fixed points. Now, this right action induces
1016 an representation of W in H ˚ pG{Tq. For w ‰ 1, since there are no w-fixed points, w has Lef-
1017 schetz number χpwq “ 0; but since H ˚ pG{Tq is evenly graded by Theorem 4.1.1, this means that
1018 tr w| H˚ pG{Tq “ 0. On the other hand, χp1q “ χpG{Tq “ |W| by Corollary 4.2.2.

1019 We also can show that the Euler characteristic of a generic compact homogeneous space is
1020 zero.
1021 Corollary 4.2.5. Let G be a compact, connected Lie group and K a closed, connected subgroup of lesser
1022 rank. Then χpG{Kq “ 0.
1023 Proof. Let S be a maximal torus of K and T be a maximal torus of G containing S. Then we
1024 have a fiber bundle T{S Ñ G{S Ñ G{T. Since the base is simply-connected, it follows from
1025 Proposition 3.3.12 that
χpG{Sq “ χpG{TqχpT{Sq “ χpG{Tq ¨ 0,
1026 this last since a torus T{S is a product of circles and χpS1 q “ 1 ´ 1 “ 0. Let N “ NK pSq be the
1027 normalizer in K of its maximal torus S. Since N Ñ S is a covering with fiber WK , so also is
1028 G{S Ñ G{N, so by Proposition B.3.5,
L
χpG{Nq “ χpG{Sq |WK | “ 0.
1029 Now by Corollary 4.2.3 we have χpG{Kq “ χpG{Nq “ 0.

1030 Historical remarks 4.2.6. The Euler characteristic dichotomy that χpG{Kq ą 0 or “ 0 depending as
1031 rk G “ rk K or rk G ą rk K is due to Hopf and Samelson [HS40, p. 248].
1 N.B. The proof of this result in [MT00, Prop. VII.3.25, p. 399] is wrong, as it tries to use the left multiplication
action.
1032 Chapter 5
1033 First properties of equivariant cohomology

1034 In this chapter, G is a connected Lie group.

1035 Values on the orbit category and Mayer–Vietoris

1036 We have shown the importance of the orbit category G-Orbit in understanding G-actions, so the
1037 first order of business in understanding equivariant cohomology should be to know the values
1038 it takes on this category. It will be enough to understand the cohomology ring H ˚ pBKq of the
1039 classifying space BK, as we will come to in Section 6.6. We will primarily care about the case the
1040 coefficient ring k is Q, but for some statements we can get away with k “ Z.
1041 Proposition 5.1.1. Let G be a Lie group and K a closed subgroup. Then for any coefficient ring k there is
1042 an isomorphism
HG˚ pG{K; kq “ H ˚ pBK; kq.
1043 Proof. The proof is topological and nearly formal. The equivariant cohomology HG˚ pG{Kq is the
1044 singular cohomology of the homotopy quotient pG{KqG “ EG ˆG G{K, which is homeomorphic
1045 to EG{K, taking P “ EG in the following lemma. But EG{K “ BK for G Lie and K closed, by the
1046 observation in Corollary 2.1.4.

Lemma 5.1.2 ([BTara, Prop. 4.5]). Let G be a Lie group and K a closed subgroup, and suppose P ÝÑ B
is a principal bundle. Then there is a homeomorphism
P ˆ G{K ÝÑ P{K,
G
rp, gKs ÞÝÑ pgK.
Proof. The continuity and well-definedness of the map follow from the fact the right action
P ˆ G ÝÑ P,
pp, gq ÞÝÑ pg,
(
is continuous and constant on each diagonal orbit rp, gsG “ ppx, x´1 gq : x P G . It thus descends,
by definition, to a continuous map ϕ : P ˆG G ÝÑ P, and since this map is right K-equivariant
in that ϕrp, gksG “ ϕrp, gsG ¨ k for k P K, it descends further to a well-defined, continuous map
P ˆG G{K ÝÑ P{K. The map is bijective and because an inverse is given by the map
P{K ÝÑ P ˆ G{K,
G
pK ÞÝÑ rp, KsG .

38
 Chapter 5. First properties of equivariant cohomology 39

1047 It is not immediate this inverse is continuous, but it clearly would be if we could factor it through
1048 a continuous map P{K ÝÑ P ˆ tKu Ď P ˆ G{K taking pK ÞÝÑ pp, 1Kq. Though this is not generally
1049 possible, it is enough for continuity to be able to do it locally, and this is equivalent to finding
1050 local sections of P ÝÝ P{K. But this we can do: since P is assumed a G-bundle and since G is a
1051 K-bundle by Theorem B.4.3, it follows P is an K-bundle, and so locally trivial.

1052 Now that we know what HG˚ does on the orbit category, one hopes to be able to use the
1053 decomposition of a compact G-space as a G–CW complex, a union of disks D n ˆ G{K along G-
1054 equivariant maps, for K and n varying, to understand what it does to a G-space in general, a
1055 sort of “cellular equivariant cohomology.” And indeed HG˚ pD n ˆ G{Hq “ HG˚ pG{Hq – H ˚ pBHq,
1056 so these building blocks behave as simply as one could hope for.
1057 This approach has mostly theoretical utility, because G–CW decompositions seem to be hard
1058 to come by in the wild, but an understanding of the cohomology of a union is still valuable.

1059 Proposition 5.1.3 (Equivariant Mayer–Vietoris). Let U and V be G-invariant subsets of a G-space X
1060 and suppose U Y V “ X. Then there exists a long exact sequence

¨ ¨ ¨ ÝÑ HGn´1 pU X Vq ÝÑ HGn pXq ÝÑ HGn pVq ‘ HGn pVq ÝÑ HGn pU X Vq ÝÑ HGn`1 pXq ÝÑ ¨ ¨ ¨ .

1061 Proof. Since U and V are G-invariant, so is U X V. The decomposition U Y V “ X translates on

1062 taking homotopy quotients to a decomposition UG Y VG “ XG such that UG X VG “ pU X VqG ,
1063 and the result follows on applying the Mayer–Vietoris sequence in singular cohomology.

1064 The equivariant Künneth theorem

1065 Recall from Section 1.1 that if G acts on spaces X and Y, there is a natural diagonal product
1066 action on X ˆ Y. In ordinary cohomology, at least with coefficients in a field k, one has H ˚ pX ˆ
1067 Yq – H ˚ X b H ˚ Y, and in less favorable circumstances, one still has at least a Künneth spectral
1068 sequence as mentioned in Remark A.5.6. With more severe restrictions, something similar holds
1069 in equivariant cohomology.

1070 Theorem 5.2.1 (Equivariant Künneth [KV93]). Suppose a topological group G acts continuously on
1071 spaces X and Y in such a way that the fiber restriction HG˚ pY; kq ÝÑ H ˚ pY; kq arising from the Borel
1072 fibration is surjective, with H ˚ pY; kq a free k-module.1 Then, for the same coefficient ring k, there exists an
1073 isomorphism
HG˚ pX ˆ Yq – HG˚ pXq b HG˚ pYq.
˚
HG

1074 This will follow from some diagram equivalences and the recurrent bundle lemma Theo-
1075 rem 3.4.1.

1076 Lemma 5.2.2. The following diagram commutes, where every column and row form a bundle, the bundles
1 If this fiber restriction is surjective, it is said that the G-action on Y is equivariantly formal; we will have much more
to say about this condition in later chapters, predominantly for k “ Q.
 Chapter 5. First properties of equivariant cohomology 40

1077 over BG are Borel fibrations, and the central object pX ˆ YqG is the pullback XG ˆBG ˆYG :

Y Y

 
X / pX ˆ YqG / YG (5.1)

 
X / XG / BG.

Proof. The commutativity of the entire diagram will be clear if we show that we can replace
pX ˆ YqG with the pullback XG ˆBG ˆYG in such a way that the lower-right square becomes a
pullback square. Let ∆EG :“ pe, eq P EG ˆ EGu be the diagonal. Now
` ˘ (
XG ˆ YG “ re, xs, re1 , ys P XG ˆ YG : Dg P G peg “ e1 q
BG

(N
« pe, x, eg, yq P EG ˆ X ˆ EG ˆ Y
peh, x, egh1 , yq „ pe, hx, e, gh1 yq

(N
pe, e, x, yq P ∆EG ˆ X ˆ Y
«
peg, eg, x, yq „ pe, e, gx, gyq,

1078 the second homeomorphism because under the relation imposed on EG ˆ X ˆ EG ˆ Y, one can
1079 always find a representative pe, x, e, yq, and restricted to such representatives, the relation imposed
1080 by h and h1 is just the relation imposed on ∆EG ˆ X ˆ Y in the third term. On the other hand,
1081 since ∆EG ˆ X ˆ Y is G-equivariantly homeomorphic to EG ˆ X ˆ Y, we finally see pX ˆ YqG «
1082 XG ˆBG YG . Explicitly, the homeomorphism is given by
` ˘
re, x, ys ÞÝÑ re, xs, re, ys .

1083 The maps to XG and YG take these elements to re, xs and re, ys respectively, so the diagram
1084 commutes.

1085 Corollary 5.2.3. Let G be a Lie group and G and K closed subgroups. Then one has a homeomorphism

pG{H ˆ G{KqG « BH ˆ BK
BG

1086 Proof. Putting X “ G{H and Y “ G{K in Lemma 5.2.2, one finds pG{H ˆ G{KqG « pG{HqG ˆ pG{KqG ,
BG
1087 but from Lemma 5.1.2 one also knows pG{HqG « BH and pG{KqG « BK.

1088 The bundle lemma Theorem 3.4.1 and Lemma 5.2.2 make the equivariant Künneth theorem
1089 immediate.
 Chapter 5. First properties of equivariant cohomology 41

1090 Proof of Theorem 5.2.1. Taking the cohomology of diagram (5.1) yields the commutative diagram

H ˚ pYq H ˚ pYq
O O

H ˚ pXq o HG˚ pX ˆ Yq o HG˚ pYq (5.2)

O O

H ˚ pXq o HG˚ pXq o HG˚ .

1091 By assumption, the top-right vertical map HG˚ pYq ÝÑ H ˚ pYq is surjective, so by the bundle lemma
1092 Theorem 3.4.1, it follows HG˚ pX ˆ Yq – HG˚ pXq b HG˚ pYq is surjective.

1093 We state here a corollary about equivariant cohomology of homogeneous spaces, once we
1094 prove another bundle equivalence.
1095 Proposition 5.2.4. Let G be a Lie group and H and K closed subgroups. Then one has homeomorphisms

pG{HqK « BK ˆBG BH « pG{Kq H .

Proof. By symmetry, it will be enough to prove the first homeomorphism. A candidate map is

ψ : pG{HqK “ EG ˆK G{H ÝÑ BK ˆBG BH,

re, gHsK ÞÝÑ peK, egHq.

One suspects that this map must in fact yield an isomorphism of G{H-bundles over BK, and
indeed it does. To define an inverse, start with the homeomorphism

EG ˆBG EG ÝÑ EG ˆ G :
pe, egq ÞÝÑ pe, gq,

and follow it with the projections EG ˆ G Ñ EG ˆ G{H Ñ EG ˆK G{H to get

φ : EG ˆBG EG ÝÑ EG ˆK G{H “ pG{HqK ,

pe, egq ÞÝÑ re, gHsK .

1096 To see φ descends to a well defined inverse BK ˆBG BH ÝÑ pG{HqK to ψ, let a point peK, egHq P
1097 BK ˆBG BH be given and consider two points in its preimage EG ˆBG EG. One natural choice
1098
`is pe, egq, and ˘ others are all pek, eghq for some k P K and h P H. We can rewrite pek, eghq as
the
1099 ek, pekqpk´1 ghq . Now
` ˘
φ ek, pekqpk´1 gqh “ rek, k´1 ghHsK “ re, gHsK “ φpe, egq,

1100 so φ descends to a continuous inverse as claimed.

«
1101 Remark 5.2.5. For the sake of completeness, we note the composite homeomorphism pG{HqK ÝÑ
1102 pG{Kq H is given by
re, gHsK ÞÝÑ peK, egHq ÞÝÑ reg, g´1 Ks H .
1103 This map does not preserve any particular bundle structure, but does yield an isomorphism
1104 HK˚ pG{Hq – HH
˚ pG{Kq for any coefficient ring.
 Chapter 5. First properties of equivariant cohomology 42

1105 Corollary 5.2.6 ([KV93, Prop. 68, p. 161]). Let G be a compact Lie group, K a subgroup of full rank,
1106 and H another closed subgroup. Then
˚
HH pG{Kq – HK˚ pG{Hq – HH
˚
b HK˚
˚
HG

1107 as rings.

1108 Proof. Take X “ G{H and Y “ G{K in the diagram (5.2). The diagram becomes

H ˚ pG{Kq H ˚ pG{Kq
O O

H ˚ pG{Hq o HG˚ G{H ˆ G{K o

` ˘
HK˚
O O

H ˚ pG{Hq o ˚ o
HH HG˚ .

1109 From Corollary 5.2.3 and Proposition 5.2.4, we know

pG{Kq H « pG{HqK « BH ˆ BK « pG{H ˆ G{KqG .

BG

1110 Thus the first two rings in the statement of this corollary are always isomorphic. We will prove
1111 in Section 7.3.2 that the map HK˚ ÝÑ H ˚ pG{Kq is surjective, so Theorem 5.2.1 will apply.

1112 Corollary 5.2.7. Let G be a compact Lie group and K a connected, closed subgroup of equal rank. Then

HK˚ pG{Kq – HK˚ b HK˚

˚
HG

1113 as an HK˚ -algebra.

1114 Corollary 5.2.8 (Leray, 1950). Let G be a compact Lie group and K a closed, connected subgroup of equal
1115 rank. Then
H ˚ pG{K; Qq – Q b HK˚
˚
HG

1116 as an HK˚ -algebra.

1117 Proof. Take H “ 1 in Corollary 5.2.6.

1118 Remarks 5.2.9. We seem to have obtained Corollary 5.2.8 without any real effort, but this ease is
1119 illusory: it depends on knowing HK˚ ÝÑ HpG{Kq is surjective, an important fact which we will
1120 only demonstrate later as Theorem 7.3.11, this time relying directly on Theorem 4.1.1, the fact
1121 that the cohomology of a flag manifold G{T is concentrated in even dimensions.
1122 For a more explicit presentation of the computation Corollary 5.2.7 in terms of character-
1123 istic classes—which initially inspired the author’s rediscovery of the work in this section—see
1124 Tu [Tu10]. Subsequent to reading this Tu paper, the author rediscovered Proposition 5.2.4 and
1125 Corollary 5.2.6 independently of outside enlightenment, without understanding the equivariant
 Chapter 5. First properties of equivariant cohomology 43

1126 Künneth theorem applied and before realizing the seeming ubiquity of the bundle lemma Theo-
1127 rem 3.4.1. Related statements appear throughout the literature, though; see for example Prop. 3.9
1128 in Goertsches and Töben [GT10a] and Kumar and Vergne [KV93, Prop. 68, p. 161]. Kumar and
1129 Vergne prove one of the isomorphisms in Corollary 5.2.6, using the Cartan model absent bundle-
1130 theoretic considerations; it may simply be that they do not care about the other isomorphism.
1131 Wilhelm Singhof [Sin93, Proposition (2.3)] also comes very close to stating Proposition 5.2.4,
1132 in a rather different context. Singhof’s work was on biquotients: if G is a group with subgroups
1133 H and K, then the product group K ˆ H acts on G by
pk, hq ¨ g :“ hgk´1 ,
1134 and the biquotient K{G{H is the orbit space of G under this action. If G is a Lie group and the
1135 action of H ˆ K is free, the biquotient K{G{H is a smooth manifold. The set of diffeomorphism
1136 classes of such biquotients is strictly larger than that of homogeneous spaces; for example, the
1137 exotic 7-spheres can be shown to arise in this fashion.
1138 In our case, G will be compact. When the action of H ˆ K is free, K{G{H is homotopy equiv-
1139 alent to our pG{Kq H « pG{HqK « BK ˆBG BH by Proposition 3.2.4, and in the event K is of full
1140 rank, we have H ˚ pK{G{Hq – HK˚ b H˚ HH ˚ by our result. Singhof showed that if K ˆ H acts freely
G
1141 on G, then we have the this isomorphism even in the more general case rk K ` rk H “ rk G, us-
1142 ing an induction argument on tori to limit the possibly nonzero regions of the Eilenberg–Moore
1143 spectral sequence associated to the homotopy pullback square
K{G{H / BH

 
BK / BG;

1144 this spectral sequence is known by the work of many (Singhof cites Hans Munkholm [Mun74])
1145 to in this case collapse at E2 “ Tor‚,‚
H˚
pHK˚ , HH
˚ q, and in our case this is an algebra isomorphism.
G
1146 The calculation here that HK˚ pG{Hq – HK˚ b H˚ HH ˚ , on the other hand, holds regardless of how
G
1147 big H is and irrespective of whether the action is free.
1148 As Singhof only uses the homotopy commutativity of the pullback, it follows that his proof
1149 applies equally well to the calculation of HK˚ pG{Hq when rk K ` rk H “ rk G and the subgroup
1150 K ¨ H of G contains a maximal torus of G, independently of whether or not K ˆ H acts freely on
1151 G. The Eilenberg–Moore spectral sequence is really the proper tool for talking about cohomology
1152 of pullbacks, but we have resisted including this generalization in this chapter because it seemed
1153 excessive to also have to include an exposition of the Eilenberg–Moore spectral sequence at this
1154 date.

1155 The standard example

1156 In this section, we compute the equivariant cohomology of the standard motivating action and
1157 show it to be equivariantly formal. This result follows more quickly from the equivariant localiza-
1158 tion theorem Corollary 8.2.5, but this however requires an account of maps HT˚ ÝÑ HS˚ between
1159 cohomology rings of classifying spaces which we only develop in Section 6.8.1.
1160 Write ρ1 for the rotation action S1 ñ S2 on the standard unit sphere fixing the z-axis, and ρq
1161 for the action “sped up by a factor of q,” so that eiθ P S1 acts as rotation by qθ radians. We write
1162 the ρq -equivariant cohomology of S2 as Hρ˚q pS2 q.
 Chapter 5. First properties of equivariant cohomology 44

1163 Proposition 5.3.1. One has

Hρ˚1 pS2 q – Zrx, ys{pxyq, deg x “ deg y “ 2.
1164 Proof. The orbits of each ρq are the latitudes Cz for fixed z, which are circles for z P p´1, 1q
1165 and the poles for z “ ˘1. Thus discs U “ z´1 r´1, 1{2q and V “ z´1 p´1{2, 1s are invariant, as is
1166 their intersection U X V, which is the annulus z´1 p´1{2, 1{2q. Because the deformation retractions
1167 of U to the south pole z “ ´1 and V to the north pole z “ 1 are ρq -equivariant, it follows
1168 Hρ˚q pUq – Hρ˚q pVq – Hρ˚q pptq. Similarly, since U X V deformation retracts equivariantly to the
1169
1 , we have H ˚ pU X Vq – H ˚ pS1 q. It follows we should try to understand the
equatorial circle Seq ρq ρq eq
1170 homotopy quotients pSeq 1 q and pt .
ρq ρq
1171
1 with the complex unit circle S1 ,
Identifying Seq
N
S 8 ˆ S1
1
pSeq qρq « eq e P S8 , ζ P S1 , s P Seq
1
.
peζ, sq „ pe, ζ q sq,

The identification relation „ on S8 ˆ S1 yields pe, sq „ pes {q , 1q for each qth root of s. Because
1
1172

1173 these qth roots of s differ multiplicatively by elements of the group µq of qth roots of unity, pXq qz
1174 is homeomorphic to the quotient space S8 {µq , which is is an infinite lens space Lq . Note that L2
1175 is the real projective space RP8 , while L1 is just S8 again.
1176 For z P t˘1u, on the other hand, the orbit Cz is a pole S1 acts on trivially, and
N
S8 ˆ pt 8 1 8
ptρq “ peζ, ptq „ pe, ptq « S {S “ CP .
1177 So Xq can be seen as the union
` 8 ˘ ` ˘ ` ˘
CP ˆ t´1u Y Lq ˆ r´1, 1s Y CP8 ˆ t1u
obtained from the disjoint union of these parts by gluing Lq ˆ t1u to CP8 ˆ t1u and Lq ˆ t´1u to
CP8 ˆ t´1u along the map
αq : Lq ÝÑ CP8 ,
eµq ÞÝÑ eS1 .
1178 The q “ 1 case of our adjunction space construction shows that X1 is obtained from S8 ˆ r´1, 1s
1179 by gluing on two copies of CP8 to the ends with the canonical projection S8 ÝÑ CP8 as
1180 attaching map. Here is a picture of X1 ; the middle part is drawn thin because S8 is contractible.

Figure 5.3.2: A schematic of pS2 qρ1

1181 Since S8 is contractible, as we demonstrated in Proposition 2.2.2, the attaching map S8 ÝÑ

1182 CP8 is nullhomotopic, so we may homotope X1 inside itself onto the subspace
CP8 Y p´1, 1q Y CP8 .
 Chapter 5. First properties of equivariant cohomology 45

1183 Since p´1, 1q is contractible, it follows that X is homotopy equivalent to a wedge CP8 _ CP8 .
1184 From the fact that CP8 “ BS1 , we will show in Section 6.3 as an elementary application of the
1185 Serre spectral sequence that H ˚ pCP8 q – Zrxs, where deg x “ 2. It then follows from the wedge
1186 axiom2 that
Zrxs ˆ Zrys
Hρ˚1 pS2 q “ H ˚ pX1 q – H ˚ pCP8 _ CP8 q “ – Zrx, ys{pxyq,
p1, 0q „ p0, 1q

1187 where x, y P H 2 are the first Chern classes of the component CP8 ’s.
1188 But we promised to see this with the Mayer–Vietoris sequence. Note that H r ˚ pS8 q “ 0 since
1189 S8 is contractible. For n “ 0, since everything is connected, the Mayer–Vietoris sequence yields
∆ ´
1190 the short exact sequence 0 Ñ Z ÝÑ Z2 ÝÑ Z Ñ 0, and for n ě 1, it gives the fragment

0 Ñ Hρn1 pS2 q ÝÑ H n pCP8 q ‘ H n pCP8 q Ñ 0.

1191 Thus the map Hρ˚1 pS2 q ÝÑ H ˚ pCP8 q ˆ H ˚ pCP8 q is injective. Because the factor maps Hρ˚1 pS2 q ÝÑ
1192 H ˚ pCP8 q are ring homomorphisms induced by the inclusions U, V ãÝÝÑ S2 , it follows Hρn1 pS2 q ÝÑ
1193 H ˚ pCP8 q ˆ H ˚ pCP8 q “ Zrxs ˆ Zrys is an injective ring homomorphism, surjective in degrees ě 0.
1194 In particular, its image contains p1, 1q, px, 0q, p0, yq, and a dimension argument shows these are
1195 algebra generators for the image. It is not hard to see this ring is isomorphic to Zrx, ys{pxyq.

Corollary 5.3.3. One has

Hρ˚q pS2 q – Zrs1 , s2 , ts ps1 s2 , s1 ` s2 ´ qtq,

L

– Zrs, ts ps2 ´ qstq,

L
|s j | “ |s| “ |t| “ 2.

1196 Proof. To determine Hρ˚q pS2 q for |q| ą 1 we make use of the map X1 ÝÝ Xq which is the identity on
1197 the end caps CP8 and is the quotient map S8 ÝÑ Lq on the slices z “ const arising from latitudes
1198 of S2 . Now consider the Mayer–Vietoris sequences of the open covers of X1 and of Xq given by
1199 z´1 r´1, 1{2q and z´1 p´1{2, 1s. These open sets deformation retract respectively onto the end caps
1200 z´1 t´1u and z´1 t1u, which are homeomorphic to CP8 for both X1 and Xq . The intersection of
1201 these open sets is z´1 p´1{2, 1{2q, which deformation retracts onto the slice z´1 t0u. This slice is an
1202 S8 for X1 and is Lq for Xq . Thus the map of Mayer–Vietoris sequences corresponding to these
1203 covers and the map X1 ÝÑ X1 is as follows:

H 2k´1 pS8 q / H 2k pX1 q / H 2k pCP8 q‘2 / H 2k pS8 q

O O O

H 2k´1 pLq q / H 2k pXq q / H 2k pCP8 q‘2 / H 2k pLq q / H 2k`1 pXq q / H 2k`1 pCP8 q‘2 .

1204 We have already calculated that H ˚ pX1 q “ Zrx, ys{pxyq, and it follows from the Serre spectral
1205 sequence of the circle bundle Lq ÝÝ CP8 that H ˚ pLq q – Zrus{pquq, where u is the pullback of a
2The wedge axiom on a cohomology theory, due to Milnor, is that H r ˚ pŽ X α q – ś Hr ˚ pXα q. This is a theorem
α α
under our axioms, but is equivalent to the Eilenberg–Steenrod disjoint union axiom granted the other axioms, and is
sometimes substituted for it.
 Chapter 5. First properties of equivariant cohomology 46

1206 generator x P H 2 pCP8 q. Filling in these known values yields the diagram

0 / Ztx k , yk u Ztx k , yk u /0
O

f
0 / H 2k pXq q / Ztx k , yk u / pZ{qquk / H 2k`1 pXq q / 0.

1207 The map marked f is the difference of the two maps H 2k pCP8 q ÝÑ H 2k pLq q induced by including
1208 z´1 p´1{2, 1{2q into z´1 r´1, 1{2q and into z´1 p´1{2, 1s. Since these inclusions are both homotopic to
1209 the attaching map Lq Ñ CP8 adjoining an endcap CP8 to Xq , it follows that f px k q “ f p´yk q “ uk ,
1210 so H 2k pXq q “ ker f is a free Z-module generated by qx k , qyk , and x k ` yk . (One of the generators
1211 qx k and qyk is redundant, but there is no symmetric way to select just one.)
Since f is surjective, it follows from exactness of the bottom row that H 2k`1 pXq q “ 0, so
H pXq q is concentrated in even degree and the ring map H ˚ pXq q ÝÑ H ˚ pX1 q “ Zrx, ys{pxyq
˚

induced by X1 ÝÑ Xq is injective. Since xy “ 0 in H ˚ pX1 q, we have the relations

px ` yq j “ x j ` y j ,
qxpx ` yq j´1 “ qx j ,
qypx ` yq j´1 “ qy j ,

so that the image of H ˚ pXq q in H ˚ pX1 q is the subring generated by qx, qy, x ` y, or, since the
relation qpx ` yq ´ qx “ qy holds in Zrx, ys{pxyq, by just x ` y and qx. Since xy “ 0, it is clear that
within the image of H ˚ pXq q, the following relations hold:

qx ¨ qy “ 0,
qx ` qy “ qpx ` yq,
pqxq2 “ q ¨ pqxq ¨ px ` yq,

1212 Because each H 2k pXq q is free of rank two, a dimension argument shows there are no other re-
1213 lations. Writing s1 , s2 for the respective preimages of qx, qy and t for the preimage of x ` y, the
1214 result follows.

1215 We note in passing that q “ 0 yields Hρ˚0 pS2 q “ H ˚ pBS1 ˆ S2 q – Zrus b H ˚ pS2 q. We claim that
1216 the fiber restriction maps Hρ˚q pS2 q ÝÑ H ˚ pS2 q are surjective for all q.

Proposition 5.3.4. The actions ρq of S1 on S2 are equivariantly formal. The fiber projection Hρ˚q pS2 q ÝÑ
H ˚ pS2 q can be represented as

Zrs1 , s2 , ts ps s , s ` s ´ qtq ÝÑ ∆rss pdeg s “ 2q :

L
1 2 1 2
t ÞÝÑ 0,
s1 , s2 ÞÝÑ s.

1217 First proof. As S2 Ñ Xq Ñ CP8 is a fiber bundle over a connected space and the cohomology
1218 rings of the fiber and base are concentrated in even degree, by Corollary 3.3.11 the Serre spectral
1219 sequence collapses at E2 – H ˚ pCP8 q b H ˚ pS2 q, so that in particular the map E8 ÝÑ H ˚ pS2 q is
 Chapter 5. First properties of equivariant cohomology 47

1220 surjective; but this is just the map H ˚ pXq q ÝÑ H ˚ pS2 q induced by the fiber inclusion, so the action
1221 ρq is equivariantly formal.
1222 Because rk H 2 pXq q “ 2, it is spanned by two generators, one of which maps to the funda-
1223 mental class in H 2 pS2 q and the other of which generates the image of H 2 pCP8 q ÝÑ H 2 pXq q.
1224 Because the diagonal S1 -action on S8 ˆ S2 induced by ρq is free on the end-cap S8 ˆ tz “ 1u,
1225 the projection map Xq ÝÑ CP8 “ BS1 of the Borel fibration can be seen as a retraction to either
1226 CP8 end-cap. The injection H 2 pXq q ÝÑ H 2 pCP8 q‘2 in the Mayer–Vietoris sequence above is in-
1227 duced by the inclusions of the open sets U, V in the cover, or equivalently by the inclusions of
1228 the end-caps. The commutative diagram

9 Xq

+ ""
CP8 > CP8 / CP8 ,

1229 where the horizontal map identifies the two copies of CP8 , induces

H 2 pXq q
g
„
v
Ztx, yu “ H 2 pCP8 q2 o x`y ÑÝÞ u
H 2 pCP8 q “ Ztuu,

1230 so the image of H ˚ pCP8 q ÝÑ H ˚ pXq q is the subring Zrx ` ys “ Zrts, and the map H ˚ pXq q ÝÑ
1231 H ˚ pS2 q induced by the fiber inclusion S2 ãÝÝÑ Xq must take t ÞÝÑ 0 and s1 , s2 ÞÝÑ rS2 s.

1232 Second, geometric proof for q “ 1. Because each object in the bundle S2 Ñ X1 Ñ CP8 is simply
„
1233 connected, the Hurewicz theorem gives natural isomorphisms π2 Ý Ñ H2 in this sequence, and
1234 because H 2 is free and H1 “ 0, the sequence

H 2 pS2 q ÐÝ H 2 pX1 q ÐÝ H 2 pCP8 q

1235 is dual to the sequence

π2 pS2 q ÝÑ π2 pX1 q ÝÑ π2 pCP8 q.3
1236 Since H 2 pS2 q ÐÝ H 2 pXq q is dual to π2 pS2 q ÝÑ π2 pXq q, it will be enough to understand the
1237 homotopy class of a fiber inclusion σ : S2 ÝÑ X1 , which factors as
«
S2 ÝÑ te0 u ˆ S2 ãÝÝÑ S8 ˆ S2 ÝÝ X1 .
{S1

1238 Pick a point e0 P S1 Ĺ S3 Ĺ S8 ; for example, we can set e0 “ 1 P C Ĺ C8 . Then

1 2L 3 2L
im σ “ S ˆ S S1 Ĺ S ˆ S S1 Ĺ X1 .
3 In fact, although we do not need this fact, this last sequence is short exact. For of course π pS2 q “ 0. Now CP8
1
admits a decomposition as a union of one cell in each even dimension. Its 2-skeleton is S2 , and the attaching map
of the 4-cell e4 is the Hopf map Be4 “ S3 ÝÑ S2 . Any map S3 ÝÑ CP8 can be homotoped by cellularity into the
2-skeleton S2 . Since π3 pS2 q is generated by the Hopf map, which is coned off by e4 , it follows π3 pCP8 q “ 0 as well.
 Chapter 5. First properties of equivariant cohomology 48

1239 Over z “ ˘1, then, σ1 is just the inclusion of the poles of S2 into CP8 as the class r1,~0s. The other
1240 latitudes σpCz q are taken, under the homeomorphism pS8 ˆ Cz q{ „ ÝÑ S8 given by re, ss ÞÝÑ es,
1241 to
σpCz q « S1 Ĺ S3 Ĺ S8 .
1242 We can conceive this S1 as the unit circle in the first C factor in C8 and S3 as the unit sphere in
1243 the first two factors C2 . It is harmless to think of the open subset z´1 p´1{2, 1{2q as p´1{2, 1{2q ˆ S8 .
1244 Because π1 pS3 q “ 0, the restriction of σ to a loop S1 « C0 ÝÑ z´1 t0u « S8 is nullhomotopic
1245 in S3 Ĺ S8 . Using such a nullhomotopy in S3 , we can pinch the image under σ of the equator;
1246 this is homotopy (a) of Figure 5.3.5. Then we can expand this “contracted segment” in such a
1247 way that σpCz q is a circle for z R r´1{2, 1{2s, but the image σpCz q is the point j “ p0, 1q P S3 for
1248 z P r´1{2, 1{2s; this is homotopy (b) of Figure 5.3.5. Thus the image of σ is now two spheres σ´1
1249 and σ1 connected by a path.

Figure 5.3.5: Homotoping a representative of π2 X1 into the end-caps t˘1u ˆ CP8

1250 Now we can continuously deform σ to a singular sphere σ1 such that the interval of z-values
1251 such that σ1 pCz q « pt is all of p´1, 1q, completing the homotopy (c) in Figure 5.3.5. The image
1252 of σ1 will be the union of the interval p´1, 1q ˆ tju and of two spheres, σ´1
1 and σ1 comprising
1
1253
1
embedded CP ’s in the end-caps CP . 8

To see that this singular sphere σ1 is what we want, view σpCz q, for each ´1 ă z ă ´1{2, as
a subset of S3 . The process of crushing σ1 down to σ11 « CP1 , for each fixed z-coordinate circle
σpCz q, viewed as a subset of S3 , is just the canonical projection

S3 ÝÝ CP1 :
pζ 1 , ζ 2 q ÞÝÑ rζ 1 , ζ 2 s “ pζ 1 , ζ 2 q ¨ S1 .

1254 The point σpC´1 q is the class r1, 0s P CP2 Ĺ CP8 , and the point σpC´1{2 q is the quaternion j “
1255 p0, 1q P S3 , so σ1 pC´1 q “ r1, 0s and σ1 pC´1{2 q “ r0, 1s. The images σ1 pCz q Ĺ CP1 for ´1 ă z ă ´1{2
1256 are circles interpolating between these two poles, so σ´1 1 ultimately can be seen as a degree-1
1257
2 2 1
map S ÝÑ S “ CP Ĺ CP . Similarly, one has σ pC1{2 q “ r0, 1s and σ1 pC1 q “ r1, 0s, so σ11 is a
8 1

1258 degree-p´1q map S2 ÝÑ S2 .

„ „
The Mayer–Vietoris sequence yields an isomorphism π2 pCP8 q‘2 Ý Ñ H2 pCP8 q‘2 ÝÑ H2 pX1 q.
If we write x˚ , y˚ P π2 pCP8 q‘2 for the dual generators to the generators x, y in H 2 pX1 q, and v˚ P
π2 pS2 q for the dual to a generator v P H 2 pS2 q, then we have just shown σ˚ v˚ “ x˚ ´ y˚ P π2 pX1 q.
Thus

σ˚ xv˚ “ xσ˚ v˚ “ 1,
σ˚ yv˚ “ yσ˚ v˚ “ ´1,

1259 so σ˚ x “ v and σ˚ y “ ´v.

 Chapter 5. First properties of equivariant cohomology 49

1260 Weyl-invariants and the restricted action a maximal torus

1261 In Appendix B.4, we pointed that the maximal torus of a compact, connected Lie group and its
1262 Weyl group carry much of its algebraic structure. In this section, we show something similar
1263 holds for the equivariant cohomology of an action of a compact, connected Lie group K and the
1264 restricted action by that group’s maximal torus S. To do so, we use Theorem 4.1.1 and the result
1265 of Section 6.4, which we will prove later. Later, in Section 9.1, we will use the characterization in
1266 this section to obtain Theorem 9.1.4, one of this dissertation’s major original results.
1267 To start, we state a natural enhancement of the motivating observation Proposition 3.2.4 about
1268 free homotopy quotients.
1269 Lemma 5.4.1. Let K be a group, S a subgroup, and X and Y free K-spaces admitting a K-equivariant map
1270 X ÝÑ Y. Then these diagrams commute:

XS / XK XK
» / X{K
»

»
   
X{S / X{K, YK
» / Y{K;

1271 so up to homotopy, XK ÝÑ YK is equivalent to X{K ÝÑ Y{K and XS ÝÑ XK to X{S ÝÑ X{K.

1272 In this statement, the horizontal maps in the first square are the “further quotient” maps
1273 rx, esS ÞÝÑ rx, esK : XˆEK
S ÝÝ XˆEK
K and xS ÞÝÑ xK : X{S ÝÝ X{K.
1274 Definition 5.4.2. In the rest of this section, we let K be a compact, connected Lie group, S a
1275 maximal torus, N “ NK pSq the normalizer of S in K, and W “ N{S the Weyl group of K.
1276 Write K-Top for the category of topological spaces with continuous K-actions and K-equivariant
1277 continuous maps, K-Free for the full subcategory of free K-actions, Q-CGA for the category of (ho-
1278 momorphisms between) graded commutative Q-algebras, and HS˚ -CGA for subcategory of graded
1279 commutative HS˚ -algebras. Recall that the “freeing” operation X ÞÑ X ˆ EK of Section 3.2.3 taking
1280 a right K-action to the diagonal action is a functor K-Top Ñ K-Free.
1281 Observation 5.4.3. Suppose K acts on the right on a space X. Then W acts on the right on the
1282 orbit space X{S by xS ¨ nS “ xnS, and so on the cohomology H ˚ pX{Sq. Given a K-equivariant
1283 map X ÝÑ Y, the induced map X{S ÝÑ Y{S is W-equivariant, so the map H ˚ pX{Sq ÐÝ H ˚ pY{Sq
1284 is as well.
1285 Lemma 5.4.4. Suppose a finite group W acts on spaces X and Y and there is a W-equivariant continuous
1286 map X ÝÑ Y inducing a surjection H ˚ pXqÝÝ H ˚ pYq. Then the map H ˚ pXqW ÐÝ H ˚ pYqW is also
ϕ
1287 surjective.
1288 Proof. The restriction to elements b P H ˚ pYqW has image in H ˚ pXqW by W-equivariance: if w ¨ b “
1289 b for all w P W, then w ¨ ϕpbq “ ϕpw ¨ bq “ ϕpbq is invariant as well.
1290 To see the restriction is surjective, let a P H ˚ pXqW . Then it has a preimage b P H ˚ pYq, not a
1 ř
1291 priori W-invariant. However, the W-average b̄ “ |W| wPW w ¨ b certainly is, and by equivariance,
1292 ϕpb̄q “ ā. Since a was assumed invariant, this average is just a again.

1293 Lemma 5.4.5 (Leray, 1950). There is a natural isomorphism

„
H ˚ pX{Kq ÝÑ H ˚ pX{SqW

1294 of functors pK-Freeqop ÝÑ Q-CGA.

 Chapter 5. First properties of equivariant cohomology 50

1295 Proof. The quotient map X{S ÝÑ X{K factors as

X{S ÝÑ X{N ÝÑ X{K.

1296 The factor X{S ÝÑ X{N is a regular covering with fiber W, which induces by Proposition B.3.1
„
1297 an isomorphism H ˚ pX{Nq Ý Ñ H ˚ pX{SqW . The fiber of the factor X{N Ñ X{K is K{N, and
1298 H ˚ pK{Nq – H ˚ pK{Sq by Corollary 4.2.3.
1299 Naturality follows because the diagram

X / X{S / X{N / X{K

   
Y / Y{S / Y{N / Y{K

1300 commutes and because, by Observation 5.4.3, the map X{S ÝÑ Y{S is W-equivariant.

1301 This lemma makes available a natural phrasing of an important, well-known result [Hsi75,
1302 Prop. III.1, p. 31].

Corollary 5.4.6. Let K be a compact, connected Lie group with maximal torus S. Then there are natural
isomorphisms of functors pK-Topqop ÝÑ HK˚ -CGA and pK-Topqop ÝÑ HS˚ -CGA respectively taking
„
HK˚ pXq ÝÑ HS˚ pXqW ,
„
HS˚ b HK˚ pXq ÝÑ HS˚ pXq.
HK˚

1303 Proof. Note that X ÞÝÑ XS is the composition of the freeing functor X ÞÝÑ EK ˆ X of Defini-
1304 tion 5.4.2 and the orbit-space functor Z ÞÝÑ Z{S. Since the diagonal action on EK ˆ X is free, the
1305 first isomorphism follows by Lemma 5.4.4.
1306 Also note that ξ 0 : BS ÝÑ BK is a pK{Sq-bundle. Because H ˚ pK{Sq is evenly-graded by The-
1307 orem 4.1.1 and HS˚ is evenly-graded by the result of Section 6.4, the E2 page of the spectral
1308 sequence associated to ξ 0 is concentrated in even rows and columns, meaning it collapses by
1309 Corollary 3.3.11 and so the fiber inclusion K{S ãÝÝÑ BS is surjective on cohomology by Corol-
1310 lary 3.3.9.
1311 Recall from the beginning of Section 3.4 the category F-Bun{ξ 0 of bundles over ξ 0 . The con-
1312 struction p´qSãÑK : X ÞÝÑ pXS Ñ XK q is a functor K-Top ÝÑ F-Bun{ξ 0 ; the bundle map XS ÝÑ BS
1313 comes from the Borel fibration. Now the second isomorphism follows by Theorem 3.4.1.

1314 Corollary 5.4.7. Let K be a compact, connected Lie group with maximal torus S and Weyl group W. Then
1315 H ˚ pBKq – H ˚ pBSqW .

1316 Proof. Take X “ pt in Corollary 5.4.6.

1317 Remarks 5.4.8. (a) The results Lemma 5.4.5 and Corollary 5.4.6 are classical and very well known,
1318 except that the naturality of these isomorphisms is never stated. This naturality is the key feature
1319 that allowed us to use them to discover our original proof of Theorem 9.1.4, one of the main re-
1320 sults of this work; knowing only that these isomorphisms exist abstractly without understanding
1321 the relation between them does not suffice to prove the theorem.
 Chapter 5. First properties of equivariant cohomology 51

(b) Lemma 5.4.5 can fail if there exist elements of H ˚ pX{S; kq annihilated by scalar multiplication
by |W|. For example, consider the action of G “ t˘1u Ĺ Rˆ by scalar multiplication on X “ S8 Ĺ
R8 . Then X{G « RP8 , and the maximal torus T is trivial, so WG “ G, and X{T “ X “ S8 again.
With Z coefficients, one finds

H ˚ pX{G; Zq – Zrc1 s{p2c1 q, deg c1 “ 2,

H ˚ pX{T; ZqWG “ H 0 pS8 ; ZqG “ Z.

Similarly, with F2 coefficients,

H ˚ pX{G; F2 q – F2 rw1 s, deg w1 “ 1,

H ˚ pX{T; F2 qWG “ H 0 pS8 ; F2 qG “ F2 .

1322 Historical remarks 5.4.9. Leray had proved a version of Lemma 5.4.5 for classical G [Ler49b] already
1323 in 1949, and proved the general version in his Colloque paper [Ler51, Thm. 2.2]. The author is
1324 indebted to Borel [Bor98] for guiding him to these references.
1325 Chapter 6
1326 The cohomology of Lie groups and classifying
1327 spaces

1328 In this chapter, we develop enough of the theory of the cohomology of Lie groups and homo-
1329 geneous spaces to justify the theorems we use in the final few chapters and the preparations
1330 we have made in the preceding ones. This beautiful story seems to be rarely taught nowadays,
1331 so we take this opportunity to be a bit more discursive than otherwise we might. We start out
1332 with k “ Z, retreating shamelessly to k “ Q when torsion rears its head. Importantly, though, if
1333 we were willing to deal with such complications, we would not have to retreat; this is in contrast
1334 with the original Lie-algebraic methods with which this theory is typically developed, which rely
1335 essentially upon R-algebra structures and destroy torsion off the bat.
1336 The rational cohomology of a compact Lie group G is as simple as anyone has any right
1337 to expect, and this simplicity can be seen as caused either by the multiplcation on G or by the
1338 existence of invariant differential forms (again a consequence of the multiplication). The Serre
1339 spectral sequence will allow us to compute the rational cohomology of the classical groups, a
1340 major achievement in the 1930s, in a few pages.
1341 The Serre spectral sequence of G Ñ EG Ñ BG will allow us to compute the cohomology of
1342 the classifying spaces BG. This computation can be seen (perhaps ahistorically) as inspiring the
1343 definition of the Koszul complex, and through it, the definition of Lie algebra cohomology.
1344 Moreover, the Serre spectral sequence of G Ñ EG Ñ BG induces a machine, invented by
1345 Borel in his thesis, for computing the cohomology of homogeneous spaces G{K. This machine
1346 also inspires our definition of the Cartan algebra, another means to compute the cohomology of
1347 a homogeneous space which simultaneously is the motivating example behind the Cartan model
1348 for equivariant cohomology.
1349 The Cartan algebra was one of the motivating examples behind the definition of minimal
1350 models, which developed into a central tool of rational homotopy theory in the late 1960s. We
1351 use one tool from rational homotopy theory, the algebra of polynomial differential forms, to
1352 update Borel’s 1953 proof that the Cartan algebra computes the cohomology of a homogeneous
1353 space.
1354 We will cite general references for this material throughout the chapter, and diligently recount
1355 historical origins when we know them. Proofs, however, unless explicitly noted otherwise, have
1356 been dredged from the author’s own memories or created anew.
1357 The innovation in our presentation of this chapter is that we are able to present the Cartan
1358 algebra and its application in algebraic terms with essentially no use of the Lie algebra of G, of
1359 the Lie derivative, or of connections, and without developing rational homotopy theory. Though

52
 Chapter 6. The cohomology of Lie groups and classifying spaces 53

1360 many sources cover this material in more or less detail [Colloque; And62; Ras69; GHV76; Oni94],
1361 all of them rely on Lie-algebraic methods. Rational homotopy theoretic proofs of Cartan’s theo-
1362 rem can be found in texts [FHT01; FOT08], as an application of a much more of a general theory
1363 we for lack of space do not develop here. In fact, Cartan’s theorem was an early instance of and
1364 an inspiration for such methods, as discussed for example by Hess [Hes99].

1365 The cohomology of the classical groups

1366 Using the Serre spectral sequence, we can recover the main results of the 1930s on the cohomol-
1367 ogy of Lie groups in a few pages. We start with Z coefficients, then abscond away to simpler
1368 rings when it makes life easier.

1369 Complex and quaternionic unitary groups

1370 Note that Upnq acts by isometries on Cn , so that it preserves the unit sphere S2n´1 . If we view
1371 this action as a left action on the space Cnˆ1 of column vectors, the first column of an element
1372 g of Upnq determines where it takes the standard first basis vector e1 “ p1,~0qJ P S2n´1 , so the
1373 stabilizer of e1 is the subgroup « ff
1 ~0
~0J Upn ´ 1q
1374 of elements with first column e1 , which we will identify with Upn ´ 1q. Since the first vector
1375 of g P Upnq can be any element of S2n´1 , the action of Upnq on S2n´1 is transitive, so the orbit–
1376 stabilizer theorem yields a diffeomorphism Upnq{Upn ´ 1q – S2n´1 , which is in fact a fiber bundle

Upn ´ 1q ÝÑ Upnq ÝÑ S2n´1 .

1377 Similarly, the action of Sppnq on Hn , preserving the unit sphere S4n´1 , gives rise to a fiber bundle

Sppn ´ 1q ÝÑ Sppnq ÝÑ S4n´1 ,

and the action of Opnq on Rn , preserving Sn´1 , gives rise to bundles

Opn ´ 1q ÝÑ Opnq ÝÑ Sn´1 ,

SOpn ´ 1q ÝÑ SOpnq ÝÑ Sn´1 .

1378 The SSSs of these bundles allow us to recover the cohomology of the classical groups.
1379 Proposition 6.1.1. The integral cohomology of the unitary group Upnq is given by

H ˚ Upnq; Z – Λrz1 , z3 , . . . , z2n´1 s, deg z j “ j.

` ˘

1380 This can be seen as saying that in the SSSs of the bundles (right angles down) in the diagram

Up1q / Up2q / Up3q / ¨¨¨ / Upnq / Upn ` 1q

(6.1)
„

     
S1 S3 S5 ¨¨¨ S2n´1 S2n`1 ,
1381 the simplest possible thing happens, and the cohomology of each object is the tensor product of
1382 those of the objects to the left of it and below it.
 Chapter 6. The cohomology of Lie groups and classifying spaces 54

˘ proof starts with the case Up1q – S , so that H pS q – Λrz1 s. Inductively assume
1383 Proof. The 1 ˚ 1

H Upnq – Λrz1 , z3 , . . . , z2n´1 s as claimed. We have a fiber bundle

`
1384

Upnq ÝÑ Upn ` 1q ÝÑ S2n`1 ,

1385 where the cohomology of the fiber and base are known, so the impulse is to use Theorem 3.3.4.
1386 Since the cohomology of the fiber is free abelian by assumption, the E2 page is given by
E2‚,0 b E20,‚ “ Λru2n`1 s b Λrz1 , z3 , . . . , z2n´1 s,
1387 and the sequence is concentrated in columns 0 and 2n ` 1. Since the bidegree of the differential
1388 dr is pr, 1 ´ rq, the only differential that could conceivably be nonzero is d “ d2n`1 , of bidegree
1389 p2n ` 1, ´2nq.

Figure 6.1.2: The Serre spectral sequence of Upnq Ñ Upn ` 1q Ñ S2n`1

.
.
.

2n´1 z2n´1 uz2n´1

.
.
.

5 z5 uz5

0
z1 z3 uz1 z3

3 z3 uz3

1 z1 0 uz1

0 1 0 u

0 ¨¨¨ 2n ` 1

0,q ` ˘
1390 But this d sends the square E2n`1 “ H q Upnq in the leftmost column into the fourth quadrant,
1391 so dz j “ 0 for all j. Because d satisfies the product rule and sends all generators of E2n`1 into the
1392 fourth quadrant, it follows d “ 0. Thus E2 “ E8 “ Λrz1 , z3 , . . . , z2n´1
` , u2n`1 s.˘
1393 A priori, this is only the the associated graded algebra of H ˚ Upn ` 1q , but since E8 is an
1394 exterior algebra, by Proposition A.5.11, there is no extension problem.

1395 The same proof, applied to the bundles Sppn ´ 1q Ñ Sppnq Ñ S4n´1 and starting with Spp1q «
1396 S3 , yields the cohomology of the symplectic groups.
1397 The diagram associated to this induction is

Spp1q / Spp2q / Spp3q / ¨¨¨ / Sppnq / Sppn ` 1q

(6.2)
„

     
S3 S7 S11 ¨¨¨ S4n´1 S4n`3 .
1398 Proposition 6.1.3. The integral cohomology of the symplectic group Sppnq is given by
H ˚ Sppnq; Z – Λrz3 , z7 , . . . , z4n´1 s, deg z j “ j.
` ˘
 Chapter 6. The cohomology of Lie groups and classifying spaces 55

1399 The cohomology of the special unitary groups is closely related to that of the unitary groups.

1400 Proposition 6.1.4. The integral cohomology of the special unitary group SUpnq is given by

H ˚ SUpnq; Z – Λrz3 , . . . , z2n´1 s, deg z j “ j.

` ˘

1401 Proof. The determinant map yields a split short exact sequence
det
1 Ñ SUpnq ãÝÝÑ Upnq ÝÝ S1 Ñ 1; (6.3)

1402 a splitting is given by z ÞÝÑ diagpz,~1q. This semidirect product structure means Upnq is topologi-
1403 cally a product SUpnq ˆ S1 , and it follows from the Künneth theorem B.2.2 that

H ˚ SUpnq – H Upnq H ˚ pS1 q “ Λrz1 , z3 , . . . , z2n´1 s pz1 q “ Λrz3 , . . . , z2n´1 s.

` ˘ ˚
` ˘LL L

1404 The information we have accumulated makes it easy to cheaply acquire as well the cohomol-
1405 ogy the complex and quaternionic Stiefel manifolds: the idea is just, in the diagram (6.1), to stop
1406 before one gets to Up1q.

1407 Proposition 6.1.5. The integral cohomology of the complex Stiefel manifolds Vj pCn q “ Upnq{Upn ´ jq is

H ˚ Vj pCn q; Z “ Λrz2pn´jq`1 , . . . , z2n´3 , z2n´1 s.

` ˘

1408 The integral cohomology of the quaternionic Stiefel manifolds Vj pHn q “ Sppnq{Sppn ´ jq is given by

H ˚ Vj pHn q; Z “ Λrz4pn´jq`3 , . . . , z4n´5 , z4n´1 s.

` ˘

1409 Proof. The spectral sequences of the bundles (6.1) dealt with in Proposition 6.1.1 all collapsed at
1410 the E2 page, so that in particular the maps H ˚ Upnq ÝÑ H ˚ Upn ´ 1q are surjective and the iterated
1411 map H ˚ Upnq ÝÑ H ˚ Upn ´ jq is surjective by induction: explicitly, it is the projection

Λrz1 , z3 , . . . , z2pn´jq´1 s b Λrz2pn´jq`1 , . . . , z2n´1 s ÝÝ Λrz1 , z3 , . . . , z2pn´jq´1 s,

1412 with kernel pz1 , z3 , . . . , z2pn´jq´1 q the extension of the augmentation ideal of the second factor.
1413 One has more or less definitionally the fiber bundle

Upn ´ jq ÝÑ Upnq ÝÑ Vj pCn q, (6.4)

1414 whose SSS collapses at E2 by Section 7.3.1 since we have just shown the fiber projection is sur-
1415 jective. Thus the base pullback H ˚ Vj pCn q ÝÑ H ˚ Upnq is injective and H ˚ Vj pCn q is an exterior
1416 subalgebra of H ˚ Upnq whose augmentation ideal extends to the kernel pz2pn´jq`1 , . . . , z2n´1 q of
1417 the fiber projection. We see H ˚ Vj pCn q can only be as claimed.
1418 The proof for H ˚ Vj pHn q is entirely analogous.

1419 Real difficulties

1420 The degeneration of spectral sequences that occurs for unitary and symplectic fails for the orthog-
1421 onal groups, because in the analogue of the iterated fiber decompostion (6.1) of the orthogonal
1422 groups, one encounters spheres of adjacent dimension, which could lead to nontrivial differ-
1423 entials. Indeed, this does lead to rather complicated 2-torsion, so we pass to simpler coefficient
 Chapter 6. The cohomology of Lie groups and classifying spaces 56

1424 rings. Even with this simplification, there seems to be a certain unavoidable difficulty in handling
1425 H ˚ SOpnq, forcing case distinctions and a rather explicit calculation of a map of homotopy groups.
1426 The proofs here are, in the author’s own opinion, cleaner and more scrutable than those in the
1427 source material, but he would not claim they make an easy read. The reader could be forgiven
1428 for skipping this section and resuming at Section 6.2, but it seemed only right to say what could
1429 be explained about H ˚ SOpnq with the tools already at hand.
1430 To proceed, we require on a lemma [MT00, Cor. 3.13, p. 121] about the cohomology of a Stiefel
1431 manifold V2 pRn q. The proof here is a hybrid of Mimura and Toda’s and that in online notes by
1432 Bruner, Catanzaro, and May [BCM]. Recall our notational conventions from Appendix A.3.2.

1433 Lemma 6.1.6. The real Stiefel manifold V “ V2 pRn q (for n ě 4) has
#
Z n even,
Hn´2 pVq “
Z{2 n odd.

1434 Proof. If we define V2 pRn q :“ SOpnq{SOpn ´ 2q as the set of pairs of orthogonal elements of Sn´1 ,
1435 or equivalently n ˆ 2 matrices with orthonormal columns, then projection q to the first column is
1436 the projection of a bundle
q
Sn´2 ÝÑ V2 pRn q ÝÑ Sn´1 .
1437 The associated Serre spectral sequence is as in Figure 6.1.7, and it is clear the lone potentially
d
1438 nonzero differential is H n´2 pSn´2 q ÝÑ H n´1 pSn´1 q.

Figure 6.1.7: The differential dn´1 in the Serre spectral sequence of Sn´2 Ñ V2 pRn q Ñ Sn´1

n´2 z uz

..
.
..
.

0 1 u

0 ¨¨¨ n´1

1439 In particular, we have H j pVq “ 0 for j ă n ´ 2, and Hj pVq “ 0 as well by the universal coefficient
1440 Theorem B.2.1. Since we have assumed n ě 4, it follows from the long exact homotopy sequence
1441 of the bundle (Theorem B.2.4) that V is simply-connected, so by the Hurewicz Theorem B.2.6,
1442 πn´2 pVq – Hn´2 pVq, and we can concern ourselves with this group instead. The long exact
1443 homotopy sequence of Theorem B.2.4 contains the subsequence
B
πn´1 pSn´1 q ÝÑ looooomooooon
looooomooooon πn´2 pSn´2 q ÝÑ πn´2 pVq ÝÑ looooomooooon
πn´2 pSn´1 q,
–Z –Z 0
 Chapter 6. The cohomology of Lie groups and classifying spaces 57

1444 showing πn´2 pVq – Z{ im B, so our task is now to identify im B. Since πn´1 pSn´1 q is cyclic, it is
1445 enough to know what B does to a generator.
1446 Recall that the long exact homotopy sequence of the bundle Sn´2 Ñ V Ñ Sn´1 is derived
1447 from the long exact homotopy sequence of the pair pV, Sn´2 q through the isomorphism induced
1448 by the map of pairs q : pV, Sn´2 q ÝÑ pSn´1 , ptq, as follows:

πn´1 pV, Sn´2 q

6
q˚ B

„
 )
¨¨¨ / πn´1 pVq πn´1 pSn´1 , ptq πn´2 pSn´2 q / ¨¨¨ .
5

„
q˚ (  B
πn´1 pSn´1 q

1449 The top B takes the class represented by a map of pairs ι : pD n´1 , Sn´2 q ÝÑ pV, Sn´2 q to the homo-
1450 topy class of the restriction ι æ Sn´2 . Since the vertical maps are isomorphisms, such an ι will rep-
1451 resent a generator just if qι : pD n´1 , Sn´2 q ÝÑ pSn´1 , ptq represents a generator of πn´1 pSn´1 , ptq.
1452 We turn to constructing this ι.
1453 It will be convenient to consider V “ V2 pRn q as a quotient Opnq{Opn ´ 2q.1 Write p2 : Opnq ÝÑ
1454 V2 pRn q for the natural projection of a matrix to the first two columns, realizing this quotient
1455 description, and p1 : Opnq ÝÑ Sn´1 for projection to the first column alone. Note that p1 “ qp2
1456 and that p1 can be seen as the evaluation map g ÞÝÑ ge1 taking an automorphism g P Opnq ă
1457 AutR pRn q to its value at the standard basis vector e1 “ p1,~0qJ of Rn . The preimage of e1 under p1
1458 is the stabilizer
` Stabpe
˘ 1q ă Opnq, a block-diagonal t1u ˆ Opn ´ 1q which we write as Opn ´ 1q. The
1459 image p2 Opn ´ 1q « S n´2 of this subgroup in V is the fiber of the bundle Sn´2 Ñ V Ñ Sn´1
1460 over e1 . This diagram summarizes:
p1
hkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkikkkkkkkkkkkkkkkkkkkkkkkkkkkkkkj
Opnq, Opn ´ 1q ÝÑ pV, Sn´2 q ÝÑ pSn´1 , e1 q .
` ˘
p2 q

1461 The map ι arises from the natural map Sn´1 ÝÑ Opnq taking a unit vector v to the reflection
1462 rv : Rn ÝÑ Rn through the hyperplane vK ă Rn orthogonal to v. Write r : D n´1 ÝÑ Opnq for the
1463 restriction of this map to the northern hemisphere D n´1 “ tv P Sn´1 : v ¨ e1 ě 0u of Sn´1 . Note
1464 that the composition p1 r takes v ÞÝÑ p1 prv q “ rv pe1 q. If Sn´2 “ BD n´1 is the equator, made up of
1465 those unit vectors perpendicular to e1 , then we claim r takes Sn´2 to Opn ´ 1q:
r `
pD n´1 , Sn´2 q ÝÑ Opnq, Opn ´ 1q .
˘

1466 To see this, note that if v P Sn´2 , so that v is perpendicular to e1 , then e1 is in the hyperplane vK
1467 fixed by rv , so pp1 rqpvq “ rv pe1 q “ e1 . That means the first column of rv is e1 , so that rv P Opn ´ 1q.
1468 We set ι “ p2 r : pD n´1 , Sn´2 q ÝÑ pV, Sn´2 q. To see ι represents a generator of πn´1 pV, Sn´2 q,
1469 we show
qι “ qp2 r “ p1 r : pD n´1 , Sn´2 q ÝÑ pSn´1 , e1 q
1470 represents a generator of πn´1 pSn´1 , e1 q by demonstrating it takes the interior D n´1 zSn´2 home-
1471 omorphically onto Sn´1 zte1 u. Let v P D n´1 zSn´2 . If v “ e1 , then re1 pe1 q “ ´e1 , and otherwise v
1 We introduced it as SOpnq{SOpn ´ 2q, but this is the same; any g P Opnq extending an orthonormal 2-frame
pv, wq P V2 pRn q can be made into an element of SOpnq by multiplying the last column by ˘1.
 Chapter 6. The cohomology of Lie groups and classifying spaces 58

1472 and e1 together span a 2-plane, which cuts Sn´1 in a circle and vK in a line, and pp1 rqpvq “ rv pe1 q
1473 lies in this plane; see Figure 6.1.8. Since p1 r preserves these circles, it is be enough to show that
1474 the restriction of p1 r to each open upper semicircle is injective, but this ` is the case
˘ because if
1475 the nonzero angle θ “ >pe1 , vq lies in the interval p´ {2, {2q, then > rv pe1 q, ´e1 “ 2θ lies in
π π

1476 p´π, 0q Y p0, πq.

Figure 6.1.8: The reflection of e1 through vK

e1
v

π
––θ θ
2

π
––θ
2

2θ v
rv (e1)
– e1

1477 Now, since ι represents a generator of πn´1 pV, Sn´2 q, the restriction χ “ pι æ Sn´2 q : Sn´2 ÝÑ
1478 Sn´2 represents a generator of im B. Write Sn´3 Ĺ Sn´2 for the set of those unit vectors v perpen-
1479 dicular to both e1 and e2 . For such a v, the reflection
( rv will leave the first two coordinates of an
1480
n n´3
element of R invariant, so χpS q “ pe1 , e2 q P V. Since rv “ r´v , the same argument as for p1 r
1481 shows that χ takes the interiors of both north and south hemispheres homeomorphically onto
1482 Sn´2 zte2 u, so restrictions to these hemispheres are maps

τ˘ : pD n´2 , Sn´2 q ÝÑ pSn´2 , e2 q

representing generators of πn´2 pSn´2 , e2 q – πn´2 pSn´2 q such that rχs “ rτ` s ` rτ´ s. These gener-
ators are closely related: if

α : Sn´2 ÝÑ Sn´2 ,
v ÞÝÑ ´v,

1483 is the antipodal map, then we have τ´ “ τ` ˝ α. Since α`is the composition
˘ of n ´ 1 reflections of
1484 Rn´1 , it is of degree p´1qn´1 , so that χ represents sn :“ 1 ` p´1qn´1 times the generator rτ` s of
1485 πn´2 pSn´2 q.
1486 Since seven “ 2 and sodd “ 0, the group Hn´2 pVq “ πn´2 pVq – Z{sn Z is as claimed.

1487 Remark 6.1.9. Since V2 pRn q is the set of pairs pv, wq with v P Sn´1 and w K v, it can be seen as the
1488 set of unit vectors in the tangent bundle TSn´1 . This is a Sn´2 -bundle associated to a principal
1489 SOpn ´ 1q-bundle, and it can be shown that the image of the element 1 of the fiber cohomology
1490 group Z “ H n´2 pSn´2 q in the base cohomology group H n´1 pSn´1 q “ Z is the Euler class of this
1491 bundle (see Section 6.7); the fact that this number alternates between zero and two can be seen
1492 as a reflection of the fact that the Euler characteristics (Appendix A.3.2) of spheres obey the rule
1493 χpSn q “ 1 ` p´1qn .
 Chapter 6. The cohomology of Lie groups and classifying spaces 59

1494 Corollary 6.1.10. The nonzero integral cohomology groups of the real Stiefel manifold V “ V2 pRn q are
# #
Z n even, Z n even,
H 0 pVq – H 2n´3 pVq – Z, H n´2 pVq “ H n´1 pVq “
0 n odd, Z{2 n odd.

d
1495 In particular, the differential H n´2 pSn´2 q ÝÑ H n´1 pSn´1 q shown in Figure 6.1.7) is zero if n is even and
1496 multiplication by 2 if n is odd. The mod 2 cohomology ring of V is

H ˚ pV; F2 q – Λrvn´2 , vn´1 s

1497 Proof. If n is even, we have πn´2 pVq “ Hn´2 pVq infinite cyclic from Lemma 6.1.6, so by universal
1498 coefficients, H n´1 pVq is also free abelian, and it follows d “ 0 and H n´2 pVq – Z.
1499 If n is odd, we have Z{2 – πn´2 pVq “ Hn´2 pVq, so by universal coefficients, H n´2 pVq “ 0 and
1500 H n´1 pVq is the sum of Z{2 and a free abelian group. But H n´1 pVq is cyclic, since it is coker d, so
1501 we have H n´1 pVq – Z{2.
1502 As for the modulo 2 case, we have 2 ” 0 pmod 2q, so the map d is always zero and the SSS
1503 collapses. There is no extension problem simply by a dimension count.

1504 The main point of this argument, for us, is that the map d is trivial for n even and an isomor-
1505 phism over Zr 21 s if n is odd. In the mod 2 case, these differentials are all zero, so we can induct up
1506 with spheres rather than V2 pRn qs, but we do have an extension problem because exterior algebras
1507 are not free CGAs in characteristic 2.

1508 Corollary 6.1.11. The mod 2 cohomology ring of V “ Vj pRn q has a simple system vn´1 , . . . , vn´j of
1509 generators (see Definition A.3.3), where deg vi “ i. That is,

H ˚ pV; F2 q “ ∆rvn´1 , vn´2 , . . . , vn´j s.

1510 Proof. We fix n and prove the result by induction on j P r1, ns. For j “ 1, the result is just
1511 H ˚ pSn´1 q “ Λrvn´1 s. Suppose by induction the result holds for Vj´1 pRn q and the Serre spectral
1512 sequence of Sn´pj´1q Ñ Vj´1 pRn q Ñ Vj´2 pRn q collapses at E2 . Then the E2 page of the Serre
1513 spectral sequence of Sn´j Ñ Vj pRn q ÝÑ Vj´1 pRn q is (additively)

E2 “ ∆rvn´1 , . . . , vn´pj´1q s b ∆rvn´j´1 s,

1514 so the induction will go through if and only if E2 “ E8 in this spectral sequence as well. The only
1515 potentially nontrivial differential is dn´pj´1q , which vanishes on the base ∆rvn´1 , . . . , vn´pj´1q s and
1516 so is determined by the map

dn´j`1
H n´j´1 pSn´j´1 q ÝÝÝÝÑ H n´j Vj pRn q
` ˘

1517 indicated in Figure 6.1.12.

 Chapter 6. The cohomology of Lie groups and classifying spaces 60

Figure 6.1.12: The Serre spectral sequence of Sn´j Ñ Vj pRn q Ñ Vj´1 pRn q over F2

n´j vn´j

0?
..
.

0 1 ¨¨¨ ¨¨¨ vn´j`1 ¨¨¨ vn´1

0 n´j`1 n´1

1518 To see this map is zero, we identify it with the analogous differential in the Serre spectral se-
1519 quence of Sn´j Ñ V2 pRn`2´j q Ñ Sn`1´j , which we already know to be zero by Corollary 6.1.10.
1520 To do that, consider the following commutative diagram:

Sn´j Sn´j

 
V2 pRn`2´j q / Vj pRn q / Vj´2 pRn q

 
Sn`1´j / Vj´1 pRn q / Vj´2 pRn q.

1521 Each row and column is a bundle, and the bundle projections are of the form “consider the first
1522 few vectors”; for example, the map Vj pRn q Ñ Vj´2 pRn q simply forgets the last two vectors of a
1523 j-frame on Rn , and the fiber over a pj ´ 2q-frame is the set of 2-frames orthogonal to those j ´ 2
1524 vectors in Rn , and so is a V2 pRn´j`2 q.
1525 The map of columns induces a map pψr q of spectral sequences from pEr , dr q to the spectral
1526 sequence p1Er , 1dr q of the left column, which collapses at 1E2 . The bottom row is the` bundle whose
˘
1527 Serre spectral sequence we inductively assumed collapses, so ψn`1´j : H n`1´j Vj´1 pRn q ÝÑ
1528 H n`1´j pSn`1´j q is an isomorphism. The relation

0 “ 1dn`1´j ψn`1´j “ ψn`1´j dn`1´j

1529 then ensures dn`1´j “ 0 and we have collapse.

1530 Taking j “ n ´ 1 yields the result we really were after.

Corollary 6.1.13. The mod 2 cohomology ring of the special orthogonal group SOpnq has a simple system
v1 , . . . , vn´1 of generators:

H ˚ SOpmq; F2 “ ∆rv1 , v2 , . . . , vn´1 s,

` ˘

` ˘
1531 where F2 tvn´1 u is the image of H n´1 pSn´1 q ÝÑ H n´1 SOpnq .
 Chapter 6. The cohomology of Lie groups and classifying spaces 61

1532 Remark 6.1.14. We used the induction Sn´j Ñ Vj pRn q Ñ Vj´1 pRn q to pick up the cohomology of
1533 the Stiefel manifolds along the way to that of SOpnq. We could also have inducted the other way,
1534 using
SOp2q / SOp3q / SOp4q / ¨¨¨ / SOpnq / SOpn ` 1q

„
     
S1 S2 S3 ¨¨¨ Sn´1 Sn ,
` ˘
1535 in analogy with (6.1). Then the task is to show that the differential H n´1 SOpnq ÝÑ H n pSn q is
1536 zero. We can still use the collapse of the Serre spectral sequence of Sn´1 Ñ V2 pRn`1 q Ñ Sn to do
1537 this; the relevant bundle map is
SOpnq / Sn´1

 
SOpn ` 1q / V2 pRn`1 q

 
Sn Sn .
1538 The induction is substantially subtler over Z or even over k “ Zr 12 s, because the differentials
1539 no longer must be trivial. We can use the real Stiefel manifolds V2 pRn q – SOpnq{SOpn ´ 2q as
1540 building blocks now, though, the same way we used spheres before:

¨¨¨ / SOpn ´ 4q / SOpn ´ 2q / SOpnq

(6.5)
  
V2 pRn´4 q V2 pRn´2 q V2 pRn q.

Proposition 6.1.15. Let 2n ` 1 ě 3 be an odd integer and 2j ă 2n ` 1 an even integer. Then taking
coefficients in k “ Zr 12 s, we have

H ˚ SOp2n ` 1q – Λrz3 , z7 , . . . , z4n´1 s, deg z4i´1 “ 4i ´ 1.

` ˘

˚
` ˘M
M
H ˚ V2j pR2n`1 q – H SOp2n ` 1q H ˚`SOp2n ´ 2j ` 1q˘ – Λrz4pn´jq`3 , . . . , z4n´1 s.
` ˘

Proof. By Corollary 6.1.10, we have H ˚ V2 pR2j`1 q “ Λrz4j´1 s, so the objects in (6.5)

` ˘
1541
` have˘ the
˚ Sppnq . The
1542 same cohomology
` as ˘those in (6.2) which yielded the same structure (over Z) for H
1543 result for H ˚ Vj pRn q follows as in Proposition 6.1.5.

1544 To recover V2j´1 pR2n q, consider the map of bundles

V2j´2 pR2n´1 q V2j´2 pR2n´1 q

 
V2j´1 pR2n q / V2j pR2n`1 q / S2n

 
S2n´1 / V2 pR2n`1 q / S2n .
 Chapter 6. The cohomology of Lie groups and classifying spaces 62

1545 The Serre spectral sequence of the middle column collapses at E2 by an elaboration of our calcu-
1546 lation above.2 Thus we can use the bundle lemma Theorem 3.4.1 to conclude

H ˚ V2j´1 pR2n´1 q – Λre2n´1 s b Λrz4pn´jq`3 , . . . , z4n´1 s “ Λre2n´1 s b Λrz4pn´jq`3 , . . . , z4n´5 s.

` ˘
Λrz4n´1 s
` ˘
1547 Taking n “ j, we recover H ˚ SOp2nq .

1548 Proposition 6.1.16. Let 2n ě 2 be an even integer and 2j ´ 1 ă 2n odd. Then taking coefficients in
1549 k “ Zr 21 s,
H ˚ V2j´1 pR2n q – Λre2n´1 s b Λrz4pn´jq`3 , . . . , z4n´5 s,
` ˘

1550 where deg zi “ i and deg e2n´1 “ 2n ´ 1. In particular,

H ˚ SOp2nq – Λre2n´1 s b Λrz3 , . . . , z4n´5 s.

` ˘

1551 We can state the result for SOpmq more uniformly as follows:

1552 Corollary 6.1.17. Over k “ Zr 21 s, the cohomology ring of SOpmq is

#
Λrz3 , z7 , . . . , z4n´5 s b Λre2n´1 s, m “ 2n,
H SOpmq; Zr 21 s “
˚
` ˘
Λrz3 , z7 , . . . , z4n´5 s b Λrz4n´1 s, m “ 2n ` 1,
` ˘
1553 where k ¨ e2n´1 is the image of H 2n´1 pS2n´1 q ÝÑ H 2n´1 SOp2nq .

1554 To get the cases V` pRm q where ` ” m pmod 2q, we can use the Serre spectral sequence of

Sm´` ÝÑ V` pRm q ÝÑ V`´1 pRm q.

as we did in Corollary 6.1.11. The E2 page is H ˚ V`´1 pRm q b ∆rsm´` s, and the only
` ˘
1555
` potentially
˘
1556 nonzero differential, dm´``1 , is determined by a map d : H m´ ` pS m´ ` q ÝÑ H m´ ` `1 V pR m q . By
` ˘ ` ´1
1557 the last two propositions, the ring H ˚ V`´1 pRm q is an exterior algebra on generators of degree
1558 at least 2m ´ 2` ` 3 if m is odd, and at least m ´ 1 if m is even. In the former case, d is zero
1559 by lacunary considerations. In the latter, ` ě 2, since ` is of the same parity as m, so we have
1560 m ´ ` ` 1 ď m ` 1, with equality if and only if ` “ 2. Thus, if ` ą 2, then d “ 0 by lacunary
1561 considerations, and if ` “ 2, then we showed d “ 0 in Corollary 6.1.10. So no matter what, the
1562 sequence collapses at E2 , and then by Proposition A.4.4, we have

H ˚ V` pRm q – H ˚ V`´1 pRm q b ∆rsm´` s.

` ˘ ` ˘

2 The relevant bundle map is this:

SOp2n ´ 2j ` 1q / SOp2n ´ 1q / V2j´2 pR2n´1 q

 
SOp2n ´ 2j ` 1q / SOp2n ` 1q / V2j pR2n`1 q.

By Proposition 6.1.15, both rows yield tensor decompositions in cohomology and the fiber inclusion SOp2n ´ 1q ÝÑ
SOp2n ` 1q is surjective in cohomology with kernel pz4n´1 q, which is in the image of H ˚ V2j pR2n`1 q, so the same holds
of the right-hand map we are interested in.
 Chapter 6. The cohomology of Lie groups and classifying spaces 63

1563 To compile these cases into one statement, we introduce some notation. Let S be a free k-
1564 module or basis thereof and ϕ a proposition whose truth or falsehood is easily verifiable. We
1565 write #
ΛrSs if ϕ is true,
Λ tS : ϕu “
“ ‰
k otherwise.
1566 Then, gathering cases and doing some arithmetic on indices, we arrive at the following.
1567 Proposition 6.1.18 ([BCM, Thm. 2.5]). The cohomology of the real Stiefel manifold V` pRm q with coeffi-
1568 cients in k “ Zr 21 s is given by

H ˚ V` pRm q – Λrz4j´1 : 2m ´ 2` ` 1 ď 4j ´ 1 ď 2m ´ 3s b Λrem´1 : m evens b ∆rsm´` : m ´ ` evens.

` ˘

1569 Remark 6.1.19. The author found the useful notation for abbreviating case distinctions in Propo-
1570 sition 6.1.18 in the notes by Bruner, Catanzaro, and May [BCM]. Both Mimura and Toda [MT00,
1571 Thm. III.3.14, p. 121] and Félix, Oprea, and Tanré [FOT08, Prop. 1.89, p. 84] have misprints in
1572 their statements of the result Proposition 6.1.18 where the (even, even) case is omitted and an-
1573 other case repeated twice with different results. For example, Mimura and Toda list two noniso-
1574 morphic rings for the case (odd, odd). For those keeping score, the misprint in [FOT08] is also
1575 nonisomorphic to the misprint in [MT00].
1576 It is standard to discuss along with SOpnq its simply-connected double cover Spinpnq.
1577 Proposition 6.1.20. The cohomology of Spinpnq for n ě 2 satisfies

H ˚ Spinpnq; Zr 21 s – H ˚ SOpnq; Zr 21 s .
` ˘ ` ˘

1578 Proof. Since π : Spinpnq ÝÑ SOpnq is a connected double cover and 2 is invertible, the isomor-
1579 phism follows immediately from Corollary B.3.2.

1580 Finally, we will relate without proof the multiplicative structure of H ˚ SOpnq and H ˚ Spinpnq
1581 with F2 coefficients. The standard proofs invoke Steenrod squares, a technology we have not
1582 developed here and which we have made the decision (somewhat arbitrarily, given the length of
1583 this thesis) lies too far afield.
1584 Proposition 6.1.21. The mod 2 cohomology of SOpnq for n ě 2 is given by
` ˘
H ˚ SOpnq; F2 “ F2 rv1 , . . . , vn´1 s{a,

1585 where the ideal a is generated by the relations

#
v2i , 2i ă n,
v2i ”
0, 2i ě n.

1586 Shedding excess generators, we can write

` ˘
H ˚ SOpnq; F2 “ F2 rv1 , v3 , . . . , vtn{2u´1 s{b,
rn{is
1587 where b is the truncation ideal pvi q generated by the least powers of vi of degree exceeding n ´ 1.
1588 The mod 2 cohomology of Spinpnq admits a simple system of generators containing an element z of
1589 degree 2rlog2 ns ´ 1 and generators v j for each j P r1, n ´ 1s which is not a power of 2:

H ˚ Spinpnq; F2 “ ∆ z2rlog2 ns ´1 , v j : 1 ď j ă n, j ‰ 2r .
` ˘ “ ‰
 Chapter 6. The cohomology of Lie groups and classifying spaces 64

1590
`Hopf’s theorem
˘ 6.2.3, to come, allows one to be more specific about the ring structure of
1591 H˚ Spinpnq; F2 , but the description is disappointingly complicated. A simpler description can
1592 be obtained for the countable-dimensional group

Spin :“ lim
ÝÑ Spinpnq,
1593 where the colimit is taken along the unique maps Spinpnq  Spinpn ` 1q lifting the composition
1594 Spinpnq  SOpnq ãÑ SOpn ` 1q of the covering map with the canonical inclusion, which exist
1595 because the spinor groups are simply-connected. As by construction the diagrams

Spinpnq / / Spinpn ` 1q

 & 
 / SOpn ` 1q
SOpnq
Ť
1596 commute, Spin can be seen as a simply-connected double covering of SO :“ SOpnq.

1597 Theorem 6.1.22 ([BCM, Thm. 6.10, p. 55]). The mod 2 cohomology ring of Spin is given by

H ˚ pSpin; F2 q “ F2 rv2n`1 : n ě 1s

1598 and that of SO by

H ˚ pSO; F2 q “ F2 rv2n`1 : n ě 0s,
1599 the map H ˚ SO ÝÑ H ˚ Spin induced by Spin  SO being the obvious surjection.

1600 Historical remarks 6.1.23. The lemma 6.1.6 is due to Eduard Stiefel [Sti41], also the namesake of the
1601 Stiefel manifolds and the Stiefel–Whitney classes. A comprehensive account of this material, also
1602 including explicit computations for the cohomology of the exceptional groups, can be found in
1603 the much-recommended book of Mimura and Toda [MT00]. As an indication ` of the
˘ nontriviality
1604 of computing H ˚ SOpnq, we point out that while the cohomology ring H ˚ SOpnq; k for k any field
1605 follows immediately from what we have done in this section and extracting the additive structure
1606 of the integral cohomology is not hard afterward, the recovery of the integral cohomology ring from
1607 this data seems to only have been completed 1989 [Pit91].

1608 The rational cohomology of Lie groups

1609 All the cohomology rings of classical Lie groups, over sufficiently simple coefficient rings k,
1610 become exterior algebras, and one might wonder whether this holds over Lie groups in general.
1611 It has been known since 1941 that it does, due to work of Heinz Hopf exploiting a natural
1612 algebraic structure in the (co)homology of a topological group, a development that essentially
1613 reduced the study of Lie group cohomology to obtaining torsion information and collating it
1614 back into integral cohomology.
1615 We begin by isolating the essential feature of topological groups for our purposes.

1616 Definition 6.2.1. An H-space3 is a topological space G equipped with a continuous product map
1617 µ : G ˆ G ÝÑ G containing an element e P G neutral up to homotopy: we demand g ÞÝÑ µpe, gq
1618 and g ÞÝÑ µpg, eq be homotopic to idG .
3 The choice of H, due to Serre, is in honor of Heinz Hopf.
 Chapter 6. The cohomology of Lie groups and classifying spaces 65

1619 Such a map induces a coproduct in cohomology, the composition

H ˚ pµq
H ˚ pGq ÝÑ H ˚ pG ˆ Gq ÝÑ H ˚ pGq b H ˚ pGq,

1620 where the second map arises through the Künneth theorem. We denote the coproduct by µ˚ .
1621 Because H ˚ pµq and the Künneth map are maps of graded k-algebras, it follows µ˚ is a graded
algebra homomorphism, and that if x P H n pGq, then µ˚ pxq P
À j
1622 H pGq b H n´j pGq.
1623 Suppose as well that G is connected. We know µp´, eq » idG ; diagrammatically, this is the
1624 homotopy-commutative triangle below on the left, and taking cohomology whilst being casual
1625 about Künneth maps yields the commutative diagram on the right.

H ˚ piq
G
« / G ˆ teu i / GˆG H ˚ pGq ko
„
H ˚ pGq b H 0 pGq o H ˚ pGq b H ˚ pGq
O
µ H ˚ pµq
id  id
)
G H ˚ pGq

1626 This means the component of µ˚ pxq lying in H n pGq b H 0 pGq is x b 1. The same argument run
1627 with the identity µpe, ´q » idG yields the component 1 b x in H 0 pGq b H n pGq. So
ÿ
µ˚ pxq “ 1 b x ` x b 1 ` pdeg ě 1q b pdeg ě 1q.

1628 Recall that the cup product ! : H ˚ pGq ˆ H ˚ pGq ÝÑ H ˚ pGq is induced in a similar way by the
1629 diagonal map ∆ : G ÝÑ G ˆ G taking g ÞÝÑ pg, gq; to wit, it can be understood as the composition

∆˚
H ˚ pGq b H ˚ pGq ÝÑ H ˚ pG ˆ Gq ÝÑ H ˚ pGq.

1630 As ∆ and µ admit some relations on a topological level, we recover some cohomological identities.
ś4 ś2
1631 Trivially `but importantly,
˘ µ ˆ µ is a map G ÝÑ G taking the quadruple px, y, x, yq to
1632 the pair µpx, yq, µpx, yq “ p∆ ˝ µqpx, yq. If we write τ : G ˆ G ÝÑ G ˆ G for the transposition
1633 switching the coordinates, then px, y, x, yq “ pid ˆτ ˆ idqpx, x, y, yq “ pid ˆτ ˆ idqp∆ ˆ ∆qpx, yq, so

∆ ˝ µ “ pµ ˆ µq ˝ pid ˆτ ˆ idq ˝ p∆ ˆ ∆q. (6.6)

1634 Taking the cohomology of (6.6), being casual with Künneth maps again, and recalling the sign
1635 conventions for a tensor product of CGAs, one finds that for all homogeneous a, b P H ˚ pGq,

µ˚ pabq “ µ˚ paqµ˚ pbq,

1636 so that µ˚ : H ˚ pGq ÝÑ H ˚ pGq b H ˚ pGq is a ring homomorphism. All this inspires the following
1637 definition.

1638 Definition 6.2.2. A Hopf algebra over k is a graded (not necessarily associative) k-algebra A such
1639 that A0 – k, equipped with an algebra homomorphism µ˚ : A ÝÑ A bk A such that
` ˘
µ˚ paq ” 1 b a ` a b 1 mod A rb A r
k

1640 for each homogeneous a P A. (Here A r C A is the augmentation ideal À Ai – A{A0 of elements
iě1
1641 of positive degree, as defined in Appendix A.3.)
 Chapter 6. The cohomology of Lie groups and classifying spaces 66

1642 What we have shown is that, given an H-space G, its cohomology ring H ˚ pGq is naturally a
1643 commutative, associative Hopf algebra. The presence of the coproduct imposes severe constraints
1644 on the algebra structure, especially with regard to algebra generators. Here is Hopf’s powerful
1645 structure theorem.
1646 Theorem 6.2.3 (Hopf, char k “ 0: Hopf’s theorem [Hop41, Satz I, p. 23] ; Borel, char k ą 0). Let k
1647 be a field and A a commutative, associative Hopf algebra over k such that dimk An is finite for all n. As
1648 an algebra,
1649 • if char k “ 0, then A is a free k-CGA,

1650 • if char k “ 2, then A – ΛV for an oddly-graded vector space V,

1651 • if char k “ p ą 2, then A is the tensor product of a free k-CGA and truncated symmetric algebras
j
1652 krαs{pα p q, where α is even-dimensional.
Proof [Hat02, Prop. 3C.4, p. 285]. We prove the result for char k “ 0 by induction on the number
n of algebra generators, starting with n “ 0 so the result is trivial. Inductively suppose we have
shown the result for n generators and A is generated by n ` 1. Order these algebra generators
x1 , . . . , xn , y by weakly increasing degree, and let A1 be the subalgebra generated by x1 , . . . , xn .
This is actually a Hopf subalgebra, for µ˚ px j q “ 1 b x j ` x j b 1 ` pdeg ă |y|q, so the last term
cannot involve y, and must lie in A1 . Since µ˚ is an algebra homomorphism, we must have
µ˚ pA1 q ď A1 b A1 . Because A is a CGA generated by A1 and x, there is a surjective k-algebra
homomorphism

A1 b Λrys ÝÝ A if |y| is odd,

1
A b Srys ÝÝ A if |y| is even.

1653 To see A is free, it is enough to prove these maps are injective.

1654 If |y| is odd, suppose a ` by “ 0 in A, where a, b P A1 . Then 0 “ µ˚ pa ` byq P A b A projects
1655 under A b A ÝÝ A b pA {{ A1 q to

0 “ a b 1 ` pb b 1qpy b 1 ` 1 b yq “ paloo` onq b 1 ` b b y “ b b y.

moby
0

1656 This can only be zero if b is, but then 0 “ a ` 0y, so a “ 0 and our relation was trivial.
If |x| is even, we instead have to deal with a potential nontrivial relation a j y j “ 0 with
ř
1657

1658 a j P A1 . Assume the degree in y of this polynomial is minimal among nontrivial relations, and
1659 consider the image of 0 “ µ˚ a j y j P A b A under the projection to A b A{p A
ř r1 , y2 q – A b ∆rys.
1660
2
Remembering that the a j and y map to 0 in the second tensor factor, we see the image is
ÿ ÿ ÿ
0 “ pa j b 1qpy b 1 ` 1 b yq j “ a j y j b 1 ` a j y j´1 b y.
loomoon
0

a j y j´1 “ 0, contradicting the minimality of our relation a j y j “ 0.

ř ř
1661 We must then have

1662 Corollary 6.2.4. Let G be a compact, connected Lie group. Then H ˚ pG; Qq is an exterior algebra.
1663 Proof. We already know H ˚ pGq is a free k-CGA, say on V. If V contained any even-degree ele-
1664 ments, then by the theorem, H n pGq would be nontrivial for arbitrarily large n; but it cannot be,
1665 because G is a finite-dimensional CW complex. So V is oddly graded and H ˚ pGq – ΛV.
 Chapter 6. The cohomology of Lie groups and classifying spaces 67

1666 Corollary 6.2.5. Let G be a Lie group and G Ñ E Ñ B a principal G-bundle and suppose H ˚ pEq ÝÑ
1667 H ˚ pGq surjects and k is a field of characteristic zero. Then there exists a k-CGA isomorphism

H ˚ pEq – H ˚ pBq b H ˚ pEq.

1668 Proof. By Corollary 3.3.9, one has an H ˚ pBq-module isomorphism H ˚ pEq – H ˚ pBq b H ˚ pGq. By
1669 Corollary 6.2.4, H ˚ pGq is a free k-CGA, so by Proposition A.4.4, a lifting of H ˚ pEq ÝÝ H ˚ pGq
„
1670 induces a ring isomorphism H ˚ pBq b H ˚ pGq ÝÑ H ˚ pEq.

1671 We can do a bit better in identifying the generators of H ˚ pGq.

1672 Definition 6.2.6. We call x primitive if µ˚ pxq “ 1 b x ` x b 1. Write

PA “ tx P A : x is primitiveu

for the primitive subspace and grade this space by PrA “ PA X Ar . Note that the only primitive
in A0 – k can be the identity so that P0 A “ 0 and PA is contained in the augmentation ideal A.
r
˚ ˚
If A “ H pGq is the cohomology ring of an H-space G, we abbreviate PG :“ PH pGq. Another
way to phrase the definition is to say that PA is the kernel of the k-linear homomorphism

ψ : A ÝÑ A b A,
ψ
x ÞÝÑ µ˚ pxq ´ p1 b x ` x b 1q.

1673 There is a natural k-linear composite map

PpAq ãÝÝÑ A
r ÝÝ A{
r ArA
r “: QpAq

1674 linking primitives and indecomposables (for which, see Appendix A.3.2), which is an isomor-
1675 phism in the case we care about.

1676 Proposition 6.2.7 (Milnor–Moore). Let A be a commutative, cocommutative Hopf algebra finitely gen-
„
1677 erated as an algebra over a field k. Then this canonical map takes PpAq ÝÑ QpAq. In particular, A is
1678 generated by primitive elements.

1679 Proof. The strong statement is more than we need, but we will prove the result in the case A is a
1680 coassociative Hopf algebra over a field k of characteristic ‰ 2 with underlying algebra an exterior
1681 algebra, following Mimura and Toda [MT00, p. 369] for injectivity; this weaker version is due to
1682 Hopf and Samelson.
„
1683 Write A “ ΛV, for V an oddly-graded vector space. That V ÝÑ QpAq is clear, so we just need
1684 to show V can be chosen such that PpAq “ V. Pick a basis X of V. By anticommutativity, a basis
1685 of ΛV is given by monomials y “ x1 x2 ¨ ¨ ¨ xn with xi P X of weakly increasing degree. If n ą 1,
1686 then we have
ź ź` ˘ ÿ
µ˚ pyq “ µ˚ pxi q “ x1 b 1 ` 1 b xi ` p¨ ¨ ¨ q “ 1 b y ` rx1 b x2 ¨ ¨ ¨ xn s ` a b b,

1687 where none of the terms a b b have a P Qx1 . It follows the term x1 b x2 ¨ ¨ ¨ xn doesn’t cancel, and
1688 thus µ˚ pyq ‰ y b 1 ` 1 b y, so PpAq ď V.
1689 For the other containment, we induct on dim V. Assume the result is proved for n, and that
1690 dim V “ n ` 1. Arrange a homogeneous basis x1 , . . . , xn , y of V in weakly increasing degree.
 Chapter 6. The cohomology of Lie groups and classifying spaces 68

1691 By induction, V 1 “ Qtx1 , . . . , xn u, where we may choose x j primitive, and it remains to show
1692 y is. Since each x j is primitive, we have µ˚ px j q ď Λrx j s b Λrx j s for each j, so the coproduct µ˚
1693 descends to a coproduct µ˚ on ΛV {{ Λrx j s, and since this is an exterior algebra on n generators,
1694 by induction, we have µ˚ pyq “ 1 b y ` y b 1 in this quotient, so back in ΛV b ΛV, the difference
1695 ψpyq :“ µ˚ pyq ´ p1 b y ` y b 1q lies in the ideal px j b 1, 1 b x j q. Varying j, we see ψpyq lies in the the
ś
1696 intersection of all these ideals. If we write x I :“ iPI xi , this intersection ideal is that generated
1697 by the tensor products x I b x J such that I > J “ t1, . . . , nu is a partition. In fact, since by definition
1698 ψpyq P A r b A,
r it lies in the ideal generated by x I b x J with neither I nor J empty. We are then
řn
1699 done unless |y| “ i“1 |xi |, so assume this equality holds. Then since ψpyq is homogeneous and
1700 the generating elements x I b x J already have the right degree, we can write
ÿ
ψpyq “ a I,J x I b x J
I>J“t1,...,nu

1701 for some scalars a I,J P k.

1702 The fact that pµ˚ b idqµ˚ “ pid b µ˚ qµ˚ , the coassociativity of A, follows for H ˚ pGq from the
1703 associativity of the multiplication on G. It is not hard to see this is equivalent to the condition
1704 pψ b idqψ “ pid b ψqψ. Applying this equation to y we obtain
ÿ ÿ
a I,J ψpx I q b x J “ a I,J x I b ψpx J q,

1705 where the sum runs over partitions I > J “ t1, . . . , nu with I ‰ ∅ ‰ J. These equations expand to
ÿ ÿ ÿ ÿ
a I,J x I1 b x I2 b x J “ a I,J x I b x J1 b x J2 ,
I1 ,I2 J1 ,J2

1706 where I > J “ t1, . . . , nu as before and in the sums on either side, one has I1 > I2 “ I and J1 > J2 “ J,
1707 and I, J, I1 , I2 , J1 , J2 ‰ ∅. Fix a partition I1 > I2 > J “ t1, . . . , nu. The coefficients of x I1 b x I2 b x J
1708 on the left-hand side and the right, which must consequently be equal, are a I,J and a I1 ,I2 >J . These
1709 equalities show all a I,J are equal to some single scalar a P k, so
ÿ
ψpyq “ a x I b x J “ aψpx1 ¨ ¨ ¨ xn q,
I,J‰∅

1710 or ψpy ´ ax1 ¨ ¨ ¨ xn q “ 0. Thus x1 , . . . , xn , y ´ ax1 ¨ ¨ ¨ xn is a set of primitive generators of A.

1711 Remark 6.2.8. An analogous result holds in characteristic 2 with the weaker assumption on A that
1712 it not necessarily be an exterior algebra, but merely admit a simple system of generators (see
1713 Definition A.3.3). The proof is correspondingly much more difficult.
1714 We will later need as well the fact that a map of H-spaces induces a map of primitives in
1715 cohomology.

1716 Proposition 6.2.9. Let φ : K ÝÑ G be a homomorphism of H-spaces. Then the map φ˚ : H ˚ pGq ÝÑ
1717 H ˚ pKq in cohomology takes PG ÝÑ PK.

1718 Proof. To ask φ be a homomorphism is, by definition, to require µG ˝ pφ ˆ φq and φ ˝ µK be

1719 homotopic maps K ˆ K ÝÑ G. In cohomology, then, if z P PG is primitive, we have

µ˚K φ˚ z “ pφ˚ b φ˚ qµ˚G z “ pφ˚ b φ˚ qp1 b z ` z b 1q “ 1 b φ˚ z ` φ˚ z b 1.

 Chapter 6. The cohomology of Lie groups and classifying spaces 69

1720 There is a further theorem determining dim PG.

1721 Theorem 6.2.10 (Hopf [Hop40, p. 119]). Let G be a compact, connected Lie group and T a maximal
1722 torus. Then the total Betti number h‚ pGq “ 2dim T .

1723 Proof [Sam52]. By the preceding theorem, H ˚ pG; Qq is an exterior algebra, so from Appendix A.3.2
1724 we see h‚ pGq “ 2l for some l P N. To see that l “ dim T, consider the squaring map s : g ÞÝÑ g2
1725 on G. Since s “ µ ˝ ∆, it follows that for a primitive a P H ˚ pGq one has

s˚ a “ ∆˚ µ˚ a “ ∆˚ p1 b a ` a b 1q “ 1 ! a ` a ! 1 “ 2a,

1726 so if rGs P H dim G pGq is the fundamental class, the product of l independent primitives, one
1727 has s˚ rGs “ 2l rGs. Thus the degree of s is 2l . On the other hand, restricting to the abelian
1728 subgroup T – pR{Zqdim T , it is easy to see the s-preimage of a generic element of T contains
1729 2dim T points, which, since s is orientation-preserving, should each be counted with multiplicity
1730 1. By a standard theorem on degree [Hat02, Ex. 3.3.8, p. 258] we then know 2dim T “ deg s “ 2l ,
1731 so l “ dim T.

1732 The only reason we cite Samelson’s recounting rather than Hopf’s proof is that the report
1733 from which the former is taken is already in English. These results also let us obtain a classical
1734 topological fact usually proven through other means.

1735 Corollary 6.2.11 ([BtD85, Prop. V.(5.13), p. 225]). The second homotopy group π2 G of a compact Lie
1736 group G is trivial.

1737 Proof. The universal compact cover Gr of G (see Theorem B.4.4) satisfies π2 G
r – π2 G by the long
1738 exact homotopy sequence of a bundle Theorem B.2.4, and G r – A ˆ K for A a torus and K
1739 simply connected. Using the long exact homotopy sequence of the short exact sequence Zn Ñ
1740 Rn Ñ T n , one sees π2 A “ 0, and since π1 K “ 0, successively applying the Hurewicz theorem,
1741 the universal coefficient theorem, and Hopf’s theorem, one finds π2 K – H2 K – H 2 K “ 0, so
1742 π2 G
r – π2 A ˆ π2 K “ 0.

1743 Remark 6.2.12. The multiplication on a Lie group G induces a product on H˚ pG; Qq, the Pontrjagin
1744 product, making it a Hopf algebra as well, the homology ring, which is dual to H ˚ pG; Qq. It is this
1745 ring that Hopf originally discovered the structure of, though the way he put it was that the
ś 2n j ´1
1746 homology ring of G was isomorphic to that of a product S of odd-dimensional spheres.
1747 Serre found later [FHT01, p. 216] that this was actually due to a rational homotopy equivalence:
there is a map S2n j ´1 ÝÑ G inducing isomorphisms
ś
1748

´ź ¯
„
π˚ S2n j ´1 b Q ÝÑ π˚ pGq b Q

1749 on rational homotopy groups. Because the rational Hurewicz map

´ź ¯ ´ź ¯
π˚ S2n j ´1 b Q ÝÑ H˚ S2n j ´1 ; Q

Q ¨ rS2n j ´1 s of the fundamental classes of the

À
1750 is an isomorphism when restricted to the span
1751 factor spheres, the image of the Hurewicz map π˚ pGq b Q ÝÑ H˚ pG; Qq contains the homological
1752 primitives P˚ pGq “ PH˚ pGq. By Remark 3.3.17, then, these primitives are in the image of the
1753 transgression in the homological Serre spectral sequence of any G-bundle.
 Chapter 6. The cohomology of Lie groups and classifying spaces 70

1754 The Serre spectral sequence of S1 Ñ ES1 Ñ BS1

1755 The ideological mainspring of all the spectral sequence calculations we will do in the rest of this
1756 document is a sequence that is only two pages page long, the Serre sequence of the universal
1757 principal circle bundle S1 Ñ ES1 Ñ BS1 .4 We use our knowledge of H ˚ pS1 q and H ˚ pES1 q to work
1758 out H ˚ pBS1 q.

1759 Proposition 6.3.1. The cohomology of BS1 “ CP8 is given by

H ˚ pCP8 q – Zrus, deg u “ 2.

1760 Proof. By Proposition 3.3.6, π1 BS1 acts trivially on H ˚ pS1 q, so we can use untwisted coefficients
1761 in Theorem [Link] Thus we can write
p,q
E2 “ H p BS1 ; H q pS1 ; Zq
` ˘

1762 As the total space ES1 is contractible, its cohomology ring H ˚ pES1 q is that of a point, a lone Z in
1763 dimension zero, and the associated graded ring E8 again Z because the filtration is trivial.
1764 The cohomology H ˚ pS1 q is an exterior algebra Λrz1 s, where z1 P H 1 pS1 q is the fundamental
1765 class, so in particular it is a graded free abelian group, and
p,q
E2 – H p pBS1 q b H 1 pS1 q.

1766 Since the second factor is nonzero only for q P t0, 1u, the entire sequence is concentrated in these
1767 two rows.

Figure 6.3.2: The potentially nonzero region in the Serre spectral sequence of S1 Ñ ES1 Ñ BS1

.
.
.

1 ¨¨¨

0 ¨¨¨

0 2 4 6 8 10 12 ¨¨¨

1768 Thus d “ d2 is the only differential between nonzero rows, so E3 “ E8 “ Z and d must kill
‚,q
1769 everything else in E2 . Because the rows E2 “ 0 except for q P t0, 1u and d decreases q by 1, the
1770 complex pE2 , dq breaks, for each p P Z, into short complexes
p,1 p`2,0
0 Ñ E2 ÝÑ E2 Ñ 0.

1771 Because the SSS is concentrated in the first quadrant, all groups in the short complex are defini-
1772 tionally zero for p ă ´2. For p “ ´2, we have the very short complex

0 Ñ E20,0 Ñ 0,
4 We earlier, in Section 2.2, identified S8 Ñ CP8 as a model, but the calculation actually does not require this
topological knowledge.
5 In fact, from the homotopy long exact sequence of S1 Ñ ES1 Ñ BS1 , it follows that π BS1 – Z is its only
2
nonzero homotopy group, so CP8 » BS1 is an Eilenberg–Mac Lane space KpZ, 2q. In particular, BS1 is in particular
simply-connected.
 Chapter 6. The cohomology of Lie groups and classifying spaces 71

1773 red in Figure 6.3.3, witnessing the apotheosis of E20,0 – Z to H 0 pES1 q “ E8 . This in fact happens
1774 for any SSS where the fiber and base are path-connected, and must happen, since H 0 “ Z for all
1775 three spaces.
1776 For p “ ´1, we have the very short sequence
0 Ñ E21,0 Ñ 0,
1777 green in Figure 6.3.3. The middle object must zero because otherwise it would survive to E3 “ E8 ,
1778 which would mean H 1 pES1 q ‰ 0. (Then again, we already knew this because BS1 is simply-
1779 connected and H 0 is always free abelian, so that the universal coefficient theorem B.2.1 yields
1780 H 1 pBS1 q – H1 pBS1 q – π1 pBS1 qab “ 0.)

Figure 6.3.3: The first few subcomplexes of E2 in the Serre spectral sequence of S1 Ñ ES1 Ñ BS1

0 0 E20,1

Z E21,0 E22,0

0 0 0

1781 For p ě 0, the total degrees p ` 1 and p ` 2 are positive, so that both groups in the short
1782 complex must die in E3 . The only way this can happen is if the d linking them is both injective
1783 and surjective, so an isomorphism: that is,
p,1 p`2,0
E2 – E2 for all p ě 0.
1784 The first occurrence of this, for p “ 0, is blue in Figure 6.3.3. On the other hand, the simple fact
1785 that H 0 pS1 q – Z – H 1 pS1 q as abstract groups implies, on tensoring with H p pBS1 q, that likewise
p,0 p,1
E2 – E2 .
1786 Assembling these isomorphisms, all groups in even columns p “ 0, 2, 4, . . . (red in Figure 6.3.4),
1787 and all groups in odd columns (green) are isomorphic. The base cases E20,0 “ H 0 pES1 q “ Z and
1788 E21,0 “ π1 BS1 “ 0 then determine all the other entries: zero in odd columns and Z in even.
p,q
Figure 6.3.4: The partitioning by isomorphism class of groups E2 in the Serre spectral sequence
of S1 Ñ ES1 Ñ BS1

1 Z ¨¨¨

0 Z 0 Z ¨¨¨

0 1 2 3 4 5 6 7 ¨¨¨
 Chapter 6. The cohomology of Lie groups and classifying spaces 72

1789 Reading off the bottom row E2‚,0 – H ˚ pBS1 q b H 0 pS1 q – H ˚ pBS1 q, we find the cohomology
1790 groups of BS1 “ CP8 are #
Z n even,
H n pCP8 q “
0 n odd.
„
1791 Recall that the differential d “ d2 was an antiderivation restricting to an isomorphism H 1 pS1 q ÝÑ
1792 H pBS2 q. If we write u “ dz P H 2 pBS2 q for the image of the fundamental class of S1 , then since
2

1793 du “ 0, applying the product rule yields

dpuk`1 zq “ pk ` 1q ljhn
du ¨uk z ` uk`1 ¨ ljhn
dz “ uk`2
0 u

„
1794 for k ě 0. Since this d is an isomorphism E22k,1 ÝÑ E22k`2,0 and z and u are nonzero, it follows by
1795 induction that uk generates H 2k pCP8 q for all k.

1796 We could more easily have found the graded group structure of H ˚ pCP8 q through cellu-
1797 lar cohomology after pushing down the increasing union S8 “ S1 Y S3 Y S5 Y ¨ ¨ ¨ to a strictly
1798 even-dimensional CW structure CP8 “ e0 Y e2 Y e4 Y ¨ ¨ ¨ , but the spectral sequence also makes
1799 computing the ring structure almost trivial.
1800 For later reference, note that, topology aside, the calculation we just made is a manifestation
1801 of the following algebraic fact. Define B to be the graded ring Zrus, where deg u “ 2, and assign
1802 it the trivial differential. Let A be the graded ring B b Λrzs, where deg z “ 1. Make A a Z-CDGA
1803 extending pB, 0q by assigning as differential the unique antiderivation d that vanishes on 0 and
1804 satisfies
dz “ u.
1805 Then pA, dq is acyclic: H 0 pAq “ Z and H n pAq “ 0 for n ą 0. The reason we were able to deduce
1806 H ˚ pCP8 q “ Zrus is that Zrus is the unique B that makes an A “ B b Λrzs constructed as above
1807 acyclic.

1808 The Serre spectral sequence of T Ñ ET Ñ BT

śn
1809 The circle is the one-dimensional case of the torus T n “ S1 . By the Künneth theorem, one has
n n
H ˚ pT n q – Λrzs “ Λrz1 , . . . , zn s “ ΛH 1 pT n q,
â â
H ˚ pS1 q “

1810 where z j is the fundamental class of the jth factor circle and H 1 pT n q “ Ztz1 , . . . , zn u is the primitive
1811 subspace as discussed in Proposition 6.2.7.
1812 To understand H ˚ pBTq, there are at least two options. The first is an analysis analogous to,
1813 but more intricate than, that in the last section: one sees easily d2 : H 1 pTq ÝÑ H 2 pBTq must be
1814 an isomorphism and then puts more work into showing that means d2 is injective on the entire
1815 first column E20,‚ – H ˚ pTq and that E3 “ E8 “ Z. The second invokes the functoriality of the
1816 universal principal bundle construction G ÞÝÑ pG Ñ EG Ñ BGq to make the problem trivial. As
 Chapter 6. The cohomology of Lie groups and classifying spaces 73

1817 the functors E and B preserve products, one has the bundle isomorphism
„ /ś 1
T S

 ś  1
ET
„ / ES

 „ /
ś  1
BT BS ,

so that BT “ n CP8 and H ˚ pBTq “

ś Â
1818 Zru j s – Zru1 , . . . , un s.
1819 The bundle isomorphism in fact induces a Künneth isomorphism of SSSs, so that
n `
Sru j s b Λrz j s – Sr~us b Λr~zs,
â ˘
E2 “
j“1

1820 with differential d2 the unique antiderivation taking z j ÞÑ u j for each j (and hence annihilating
1821 Sr~us). Thus ´ ¯
Sr~us b Λr~zs, z j ÞÑ u j
1822 is another example of an acyclic CDGA. We will investigate the natural algebraic generalization
1823 of this phenomenon in the next section.

1824 The Koszul complex

1825 In the spectral sequences of universal bundles T Ñ ET Ñ BT, the cohomology H ˚ pTq of the
1826 fiber is an exterior algebra and the cohomology HT˚ of the base is a polynomial algebra on the
1827 same number of generators, and the algebra generators of fiber and base cancel one another in a
1828 one-to-one fashion in the spectral sequence.
1829 For another such example, consider the Lie group Spp1q. Recall that his group can be seen as
1830 the multiplicative subgroup of quaternions of norm 1, and hence is a homeomorphic 3-sphere,
and that one can take ESpp1q “ S4n´1 “ S8 and BSpp1q “ HP8 . Now, E2 page of the Serre
Ť
1831

1832 spectral sequence of the universal bundle

Spp1q ÝÑ ESpp1q ÝÑ BSpp1q
is thus E2‚,0 b E20,‚ – H ˚ BSpp1q b H ˚ pS3 q. As with the spectral sequence of S1 Ñ ES1 Ñ BS1 ,
` ˘
1833

1834 then, there are only two nonzero rows, now the 0th and the 3rd , so the only nontrivial differential
1835 can be d4 .

Figure 6.5.1: The potentially nonzero region in the Serre spectral sequence of S3 Ñ ES3 Ñ BS3
.
.
.

3 ¨¨¨

¨¨¨

¨¨¨

0 Z 0 0 0 ¨¨¨

0 2 4 6 8 10 12 ¨¨¨
 Chapter 6. The cohomology of Lie groups and classifying spaces 74

1836 Because E8 “ E5 “ Z is trivial, all the differentials d4 in and out of the other shaded boxes
1837 must be diffeomorphisms. Since the E2 “ E4 page is a tensor product, the two entries in each
1838 column must be the same, so as in the S1 case, one has isomorphisms
p,0 p`4,0
H p pBS3 q “ E2 – E2 – H p`4 pBS3 q

1839 for each p. We know H 0 pBS3 q “ Z and since there are no nonzero differentials to or from the
1840 next three boxes, these are zero. If we write z for a generator of H 3 pS3 q and q “ d4 z P H 4 pBS3 q,
1841 then as in the S1 case, we find d4 pzqn q “ qn`1 , so that finally H ˚ pBS3 q “ Zrqs. This example is
1842 very tightly analogous to the S1 example: in particular, the E2 “ E4 page was of the form

Zrqs b Λrzs, |z| “ 3, |q| “ 4,

1843 and there was only one nonzero differential, d4

1844 This example and the torus examples share the property of being tensor products of a very
1845 simple kind of spectral sequence, and we claim that for all compact, connected Lie groups G, the
1846 spectral sequence of G Ñ EG Ñ BG is such a tensor product. To facilitate future reference, we
1847 axiomatize this situation.

1848 Definition 6.5.2. Let Λrvs be the exterior k-algebra on one element v of odd degree ` and Srdvs
1849 the symmetric algebra on one element dv of degree ` ` 1. Then

Krvs “ Srdvs b Λrvs

1850 is a k-algebra. The exterior factor Λz

ˇ naturally
ˇ has a grading defined by |1| “ 0 and |v| “ `, and
1851 Srdvs inherits the natural grading ˇpdvqn ˇ “ np` ` 1q, so Krvs is bigraded by

Krvs p,q :“ Srdvs p b Λrvsq

1852 and singly graded by total degree:

Srdvs p b Λrvsq .
à à
Krvsn :“ Krvs p,q “
n“p`q n“p`q

1853 The map d : v ÞÝÑ dv uniquely defines a derivation of (total) degree one on Krvs since dpdvq “ 0,
1854 explicitly given by
d v ¨ pdvqn “ pdvqn`1 .
` ˘

1855 It should be clear from the introductory motivation that the cohomology of Krvs with respect
1856 to d is trivial. More complicated examples arise from tensoring these primordial contractible
1857 complexes.
À
1858 Definition 6.5.3. Let V “ ją0 V2j´1 be a positively- and oddly-graded free graded k-module.
1859 The grading on V induces a grading on ΛV making it a free CGA. Let ΣV “ V‚´1 be the suspen-
1860 sion, the even regrading of V achieved by moving each graded level up one degree: symbolically,
1861 pΣVq j :“ Vj´1 .6 There is a naturally induced grading on the symmetric algebra SΣV, making it a
1862 free CGA.
1863 Let KV :“ SΣV b ΛV. As an algebra, this is just the tensor product of the Krvα s for any basis
1864 vα of V. Because S1 rΣVs ‘ Λ1 rVs generates KV as a k-algebra, to characterize a derivation on KV,
6 The notation is meant to suggest the suspension ΣX of a topological space X, which satisfies H n`1 pΣXq – H n pXq.
 Chapter 6. The cohomology of Lie groups and classifying spaces 75

1865 it is enough to describe it on this submodule. One natural derivation is that which restricts on
1866 Λ1 V to the defining isomorphism
„ „
d “ Σ : Λ1 rVs Ý Ñ S1 rΣVs
ÑVÝ

1867 of ungraded free k-modules; consequently, dS1 rΣVs “ 0 and hence dpSΣVq “ 0. This is just the
1868 sum of the differentials on the Krvα s, and so is the natural differential on the tensor product.
1869 It is called the Koszul differential and the complex pKV, dq is the Koszul complex associated
1870 to V. It admits a natural bigrading

pKVq p,q :“ pSΣVq p b pΛVqq .

1871 additively extending the gradings on V and ΣV. In addition to the associated single grading
À
1872 KVn :“ n“p`q KVp,q , there is also another useful grading, setting
n
K n rVs :“ S j rΣVs b Λn´j rVs
à
j“0

1873 the submodule of KV spanned by products of n generators. This grading of KV, called the
1874 multiplicative grading, induces a grading of the cohomology of KV such that H n pKVq is the
1875 image of the cocycles in K n rVs.

1876 Just as with the Serre spectral sequence of T Ñ ET Ñ BT, we expect this cohomology to be
1877 trivial.

1878 Proposition 6.5.4 (Koszul). Let V be a free k-module and KV the Koszul complex. If k is of characteristic
1879 zero and contains Q, or if V is of finite rank, then KV is acyclic.

1880 First proof [Car51, Thm. 1]. Assume that Q ď k, so that all naturals n ě 1 are invertible in k.
„
1881 The inverse isomorphism h “ d´1 : S1 rΣVs ÝÑ Λ1 rVs extends uniquely, just as d does, to an
1882 antiderivation of KV of degree ´1. We claim it is a chain homotopy of pKV, dq.
1883 The composition dh is the projection K1 rVs Ñ S1 rVs and hd the projection K1 rVs Ñ Λ1 rVs,
1884 so hd ` dh “ id on K1 rVs. Inductively assume that also L “ dh ` hd “ n id on K n rVs and write
1885 a decomposable (e.g., basis) element of K n`1 rVs as ab, for a P K1 rVs and b P K n rVs. Then by the
1886 product rule, the base case, and the inductive assumption,

Lpabq “ pLaqb ` aLpbq “ ab ` nab “ pn ` 1qab,

1887 concluding the induction.

1888 For any n-cocycle a we then have na “ phd ` dhqa “ dha, so each d-cocycle is a coboundary
1889 for n ě 1. Thus HpKVq “ H 0 pKVq – k.

1890 This argument same argument incidentally also shows the h-cohomology of KV is trivial.
À
1891 Second proof. Assume V is of finite rank over k. Find a k-basis v j of V, so that V “ kv j and
ΣV “
À
1892 kdv j . Then we have algebra isomorphisms

KV “ SΣV b ΛV – S kdv j b Λ Srdv j s b Λrv j s “

“à ‰ “à ‰ â ` ˘ â
kv j – Krv j s,

1893 and this also holds on the level of CDGAs, because a derivation on KV is determined by its
„
1894 restriction to Λ1 rVs ÝÑ S1 rΣVs, and both the coproduct differential induced by the Koszul
 Chapter 6. The cohomology of Lie groups and classifying spaces 76

1895 differential on each Krv j s and the original Koszul differential d on KV restrict on Λ1 rVs to the
„
1896 canonical linear isomorphism Λ1 rVs ÝÑ S1 rΣVs. As everything in sight is a free k-module, the
1897 simplest version of the algebraic Künneth formula Corollary A.3.10 holds, and

Hd j Krv j s – kb rkk V – k.
â ˚` ˘
Hd˚ pKVq –
j

1898 We cite here an algebraic lemma for later use.

1899 Proposition 6.5.5. Let k be a field and d : V ÝÑ W a k-linear map of finite-dimensional k–vector spaces.
1900 Extend d uniquely to an antiderivation D on ΛV b SW annihilating SW. Then resulting k-CDGA admits
1901 a factorization as

pΛV b SW, dq – Λrker ds, 0 b Krcoim ds b Srcoker ds, 0 .

` ˘ ` ˘

1902 Proof. Because V ‘ W admits a vector space decomposition

pker d ‘ coim dq ‘ pim d ‘ coker dq,

1903 ΛV b SW admits an algebra decomposition

Λrker ds b Λrcoim ds b Srim ds b Srcoker ds

` ˘

1904 The Λrker ds and Srcoker ds factors are annihilated by D and intersect im D trivially, while by the
„
1905 definition of image and coimage, d induces the linear isomorphism coim d ÝÑ im d of the first
1906 isomorphism theorem, so that the factor Λrcoim ds b Srim ds is the Koszul algebra Krcoim ds.

1907 The Koszul complex, which makes its first appearance in thesis work of Koszul dealing with
1908 Lie algebra cohomology, which had recently been defined by Chevalley and Eilenberg, was soon
1909 discovered to have important uses in commutative algebra. Here is a more general definition.

1910 Definition 6.5.6. Let A be a unital commutative ring over k. Given a sequence ~a “ pa j q jPJ of
elements of A, we can form an abstract free k-CGA Λrz j s jPJ “ jPJ Λrz j s and the tensor algebra
Â
1911

K A~a :“ Λrz j s jPJ bk A,

À
1912 and make K A~a a CDGA by extending the k-linear map jPJ k ¨ z j ÝÑ A given by z j ÞÑ a j to an
1913 antiderivation d. We grade K A~a by multiplicative degree in the exterior factor, so that

A ~a :“ Λ rz j s jPJ b A,
n
K´n

1914 and deg d “ 1, and call this grading the resolution grading. The k-CDGA pK A~a, dq is the Koszul
1915 complex associated to the sequence ~a.
1916 Given an A-module M, the tensor product module

K A p~a, Mq :“ K A~a b M “ Λrz j s b A b M – Λ p rz j s b M,

` ˘
A k A k

inherits a differential, vanishing on M, given by

dpz j b 1q “ 1 b a j j P J,

1917 and the resulting chain complex is again called a Koszul complex.
 Chapter 6. The cohomology of Lie groups and classifying spaces 77

1918 Koszul complexes K A p~a, Mq being defined by sequence of ring elements, their potential acyclic-
1919 ity is related to properties of this sequence.
1920 Definition 6.5.7. Let A be a unital commutative ring over k. A finite or countable sequence pa j q
1921 of elements of A is called a regular sequence if for each n, the image of an is not a zero-divisor
1922 in the quotient ring A{pa1 , . . . , an´1 q. Given an A-module M, the same sequence is called M-
1923 regular (or an M-sequence) if each an annihilates no nonzero elements of the quotient module
1924 M{pa1 , . . . , an´1 qM. An ideal a E A is called a regular ideal if it can be generated by a regular
1925 sequence.
1926 Regular sequences do not normally remain regular under permutation, but do if all elements
1927 lie in the Jacobson radical of A, and in particular if A is a local ring and the elements a j are
1928 non-units [Eis95, Cor. 17.2, p. 426].
1929 Proposition 6.5.8. Let A be a connected CGA and a j elements of A;
r then the sequence pa j q is regular just
1930 if each permutation is.
1931 Since we really care only about cohomology rings, order in a regular sequence shall never be
1932 an issue for us. The connection between Koszul complexes and regular sequences is the following.
1933 Proposition 6.5.9 ([Ser00, IV.A.2, Prop. 3, p. 54]). Given a Noetherian commutative ring A, a sequence
1934 ~a of elements of the Jacobson radical of A, and a finitely-generated A-module M, the following conditions
1935 are equivalent:
` ˘
1936 1. H ´n K A p~a, Mq “ 0 for n ě 1;
` ˘
1937 2. H ´1 K A p~a, Mq “ 0;

1938 3. the sequence ~a is M-regular.

1939 The last relevant fact about Koszul complexes is that they compute Tor.
1940 Proposition 6.5.10. Let A “ Sr~as be a free commutative k-CGA generated by a sequence ~a of elements of
1941 even degree, and let B be an A-CGA. Then the Koszul complex K A p~a, Bq associated to ~a computes Tor, in
1942 that
H ´p pK A~a b A Bq – Tor A
p pk, Bq, p ě 0.

1943 Proof. The base ring k is an A-algebra in a natural way via A ÝÝ A{ A r – k. Since the generators
1944 are independent, by Proposition 6.5.9, the Koszul complex pK A~a, dq is acyclic, with

H ˚ pK A~aq “ H 0 pK A~aq – kr~as{p~aq – k.

1945 It follows that K‚A~a, with the resolution grading from Definition 6.5.6, is an A-module resolution
1946 of k, so that the ´pth cohomology of the sequence
0
¨ ¨ ¨ ÝÑ K´2 ´1
A ~a b A B ÝÑ K A ~a b A B ÝÑ K A~a b A B ÝÑ 0

1947 computes Tor A

p pk, Bq.

1948 Note that in fact Tor‚A pk, Bq is a bigraded CGA. The product descends from the product on
Λrz j s bk B, and the second component of the grading from the grading
À q
1949 B on B. We set
´p,q
p pk, B q “ H Λrz j s b B .
Tor A pk, Bq “ Tor A q p q
` ˘
k
 Chapter 6. The cohomology of Lie groups and classifying spaces 78

1950 Historical remarks 6.5.11. Regular sequences were introduced by Serre in 1955 as E-sequences [Bor67,
1951 p. 93], and this terminology apparently hung on for quite a while [Bau68, Def. 3.4]. Smith [Smi67,
1952 p. 79] uses ESP-sequence and calls a graded ideal generated by such a sequence a Borel ideal.

1953 The Serre spectral sequence of G Ñ EG Ñ BG

1954 “. . . the behavior of this spectral sequence . . . is a bit like an Elizabethan drama, full of action,
1955 in which the business of each character is to kill at least one other character, so that at the end
1956 of the play one has the stage strewn with corpses and only one actor left alive (namely the
1957 one who has to speak the last few lines).”7 —J. F. Adams

1958 Statements
1959 We have found H ˚ pBTq for all tori and by Corollary 5.4.7, we know that H ˚ pBG; Qq can be
1960 viewed as the Weyl-invariant subring H ˚ pBT; QqW , so theoretically, we understand H ˚ pBGq now.
1961 In practice, and especially if one wants to understand the torsion—something we will eventually
1962 punt on—there is more work to be done.
1963 In the torus computation, the algebra generators H 1 pTq “ PH ˚ pTq of H ˚ pTq (the primitives,
1964 as defined in Section 6.2) and H 2 pBTq – QH ˚ pBTq of H ˚ pBTq (the indecomposables, as defined
1965 in Appendix A.3.2) were linked bijectively by nontrivial differentials and were annihilated, and
1966 the algebraic repercussions of this bijection sufficed to force E8 “ Z. To work with merely
1967 generators greatly simplifies any computation, so one might hope that such a pattern holds as
1968 well for nonabelian groups. The proof of this result is due to Borel in his thesis [Bor53]. Our
1969 somewhat modernized version is based somewhat unfaithfully on the treatments contained in
1970 Mimura and Toda [MT00, p. 379–80] and Hatcher [Hat04, Thm. 1.34].
1971 The ultimate goal is the following, to be borne in mind as we regress further and further into
1972 the algebraic abstraction required for its proof in the next subsection. The transgression in the
1973 Serre spectral sequence is described in Proposition 3.3.15 and will return again in the proof of
1974 Theorem 7.1.3.

1975 Theorem 6.6.1 (Borel [Bor53, Théorème 19.1].). Let G be a compact, connected Lie group and let k be a
1976 ring (such as a field of characteristic zero) such that H ˚ pG; kq – ΛPG is an exterior algebra on odd-degree
1977 generators, (by Proposition 6.2.7, these are the primitives). Then H ˚ pBG; kq – SrτPGs is a polynomial
1978 ring on generators τPG – ΣPG of degree one greater, given by the transgressions of PG.

1979 In all this, it is to be remembered that the transgression on H p pGq is really a only a map from
0,p 0,p p`1,0 p`1,0
1980 a submodule of E p`1 ď E2 – H p pGq to E p`1 , which is a quotient of E2 “ H p`1 pBGq, so that
1981 when we lift this maps to E2 , what we get is for each p a relation τ Ď H p pGq ˆ H p`1 pBGq, rather
1982 than a map, and what it retains of the homomorphism d p`1 is additivity: if pz j , y j q are finitely
ř ř
1983 many elements of τ, then so also is p z j , y j q. Despite the imprecision, it is useful notationally
1984 and psychologically to write τ as a map in the event that the precise lift to E2 is irrelevant, and
1985 we engage in this abuse already in the statement of Theorem 6.6.1 above.
1986 That said, an precise rephrasing of Borel’s result can be obtained as follows. Writing QpBGq
1987 for the space of indecomposables (defined in Appendix A.3.2), and noting that we have a well-
7 This deservedly popular description arises in a description [Ada76] of the behavior of the Adams spectral se-

quence.
 Chapter 6. The cohomology of Lie groups and classifying spaces 79

defined isomorphism ΛPG – H ˚ pGq and an isomorphism H ˚ pBGq – S QpBGq only defined up
“ ‰
1988

1989 to some arbitrary lifting, the transgression in the spectral sequence of G Ñ EG Ñ BG nevertheless
1990 descend to well-defined isomorphisms
„
P p G ÝÑ Q p`1 pBGq

1991 summing to the isomorphism8

„
τ : PG ÝÑ QpBGq.
1992 Setting V “ PG and constructing the Koszul complex KV, this τ uniquely extends uniquely
1993 to the Koszul differential. Because H ˚ pBGq is free on QpBGq, on the level of CGAs, we recover

r2 “ H ˚ pBGq b H ˚ pGq “ KV
E

1994 and can consider τ as an antiderivation Er2 ÝÑ E r2 , sometimes called a choice of transgression,
1995 which we will use extensively in Chapter 7. By construction, it satisfies the following proposition.
1996

1997 Proposition 6.6.2. A choice of transgression τ lifts the edge homomorphisms d˜r in the sense that for each
1998 r ě 0, the following diagram commutes:

H ˚ pGq
τ / H ˚ pBGq
O

O 
rr0,r´1
E
„ / rrr,0 .
E
d˜r

1999 As the differences produced by starting with a different choice of transgression turn out to
2000 be immaterial, we will henceforth identify QpBGq with a graded subspace of H ˚ pBGq, consider
2001 it done, and not speak of it again. We also need one corollary about the original, unlifted trans-
2002 gression to prove Cartan’s theorem later in Theorem 7.1.3 and Theorem 7.1.12.
π
2003 Corollary 6.6.3 (Borel). Let G Ñ E Ñ B be a principal G-bundle classified by χ : B ÝÑ BG. Write τ
2004 for the transgression of the universal bundle G Ñ EG Ñ BG, In the spectral sequence of π, each primitive
2005 z P PH ˚ pGq transgresses to χ˚ τz.

2006 Proof. This follows from the existence of the bundle map from G Ñ E Ñ B to G Ñ EG Ñ BG,
2007 which induces a spectral sequence map as in Theorem 3.3.4 intertwining the edge homomor-
2008 phisms.

2009 Two proofs

2010 We provide two proofs of Borel’s key Theorem 6.6.1 on classifying spaces. The first is an imme-
2011 diate application of the following algebraic result to the Serre spectral sequence of the universal
2012 bundle G Ñ EG Ñ BG. It invokes the notion of transgression discussed in Appendix A.5.5.

2013 Theorem 6.6.4 (Borel [Bor53, Théorème 13.1].). Let k be a commutative ring and P an oddly-graded
2014 free k-module. Suppose pEr , dr q is a spectral sequence of bigraded k-algebras such that
8 I owe this description to Paul Baum’s thesis [Bau62, p. 3.3].
 Chapter 6. The cohomology of Lie groups and classifying spaces 80

2015 • E2 admits a tensor decomposition E2‚,0 b E20,‚ with E20,‚ – ΛP the exterior algebra on P and
0,0
2016 • the final page E8 “ E8 – k is trivial.

2017 Then all of P transgresses and E2‚,0 – SrτPs is the symmetric algebra on these transgressions.

2018 This in turn is a consequence of the following more general theorem involving simple system
2019 of generators as discussed in Definition A.3.3.

2020 Theorem 6.6.5 (Borel transgression theorem). Let k be a commutative ring. Suppose pEr , dr q is a
p,q
spectral sequence of bigraded k-DGAs with E2 – E2‚,0 b E20,‚ , that p`qďn`2 E8 “ E8
0,0
À
2021 “ k and that
0,‚
2022 there exists a free k-module P ă E8 oddly graded in degrees ď n such that the induced k-linear map

ΛP ÝÑ E20,‚

2023 is bijective in degrees ď n and injective in degree n ` 1. Then all of P transgresses, and if we write
2024 Q “ τP ă E2‚,0 , the ring map
SQ ÝÑ E2‚,0
2025 is bijective in degrees ď n ` 1 and injective in degree n ` 2.
2026 Taking the conjunction of these statements over all n, if E20,‚ – ΛP for P oddly-graded and E8 “ k,
2027 then all of P transgresses and E2‚,0 – SQ for Q “ τP.

2028 Remarks 6.6.6. Two details of this proof deserve comment.

2029 (a) The bijectivity of the linear map ΛP ÝÑ E20,‚ says precisely that the latter admits some ordered
2030 basis of P as a simple system of generators; recall that one writes then E20,‚ “ ∆P. No stronger
2031 assumption is made on the ring structure. If k is of characteristic not equal to 2, then it turns out
2032 (Corollary 6.6.8) P must be oddly-graded for the hypothesis to hold true, but in characteristic 2,
2033 this is not the case, and there is no requirement in this case that an element v P P satisfy v2 “ 0
2034 in ∆P, so the map will not generally be a ring homomorphism.
2035 (b) What we really want is the uniform statement that E20,‚ “ ∆P implies E2‚,0 “ SrτPs, and one
2036 might wonder why we bother with the fiddly degree bounds in the statement. The answer is
2037 that the weaker hypothesis of the degree-bounded statement turns out to be important to the
2038 induction step of the proof, where we only have this incomplete information to work with.

2039 Proof of Theorem 6.6.5 [MT00, p. 379–80]. The proof is an induction on n. In the n “ 0 case, since
0,0
2040 E8 “ k, we have P “ 0 and the conclusion of the theorem holds trivially. Now assume the
2041 theorem holds for n.
2042 [This refactorization of the proof is wrong and must be refactored yet again.]
2043 If n ` 1 is even, then since we have assumed the generators P to be of odd degree, there is
2044 nothing new to prove about transgressions, and we have only to prove that the map SQ ÝÑ E2‚,0 ,
2045 which we assumed to be bijective in degrees ď n ` 1 and injective in degree n ` 2, is in fact also
2046 surjective in degree n ` 2. But by assumption, the square r0, n ` 1s ˆ r0, n ` 1s in E2 “ E2‚,0 b
2047 E20,‚ is isomorphic to that in SQ b ∆P, and the differentials are completely determined by the
„
2048 transgression, which is a bijection P ÝÑ Q. Since all differentials landing in the box pn ` 2, 0q arise
2049 from within this determined square, the combined image of the differentials in E2n`2,0 descends
2050 from the polynomials in τP, namely SQ. Thus, if pSQqn`2 ÝÑ E2n`2,0 were not surjective, some
2051 elements of E2n`2,0 would never be cancelled by any differential, and therefore would survive to
À p,q
2052 E8 , contrary to the hypothesis the triangle p`qďn`2 E8 is trivial.
 Chapter 6. The cohomology of Lie groups and classifying spaces 81

2053 If n ` 1 is odd, there may be a free k-module Pn`1 of new generators of degree n ` 1 to deal
2054 with. Write Păn for the generators of lesser degree. By hypothesis,

ΛP ÝÑ E20,‚

2055 is bijective in degrees ď n ` 1 and injective in degree n ` 2. By the induction hypothesis for n, the
2056 module Păn transgresses to an evenly-graded free k-module Qăn`1 “ τPăn such that the map of
2057 graded rings
ψ : SrQăn`1 s ÝÑ E2‚,0
2058 is bijective in degrees ď n ` 1 and injective in degree n ` 2. Lift the cokernel of the degree-pn ` 2q
2059 graded component of ψ back to a free k-submodule Qn`2 ă E2n`2,0 . Writing Q “ Qăn`1 ‘ Qn`2 ,
2060 the map ψ naturally extends to a ring map

ψ1 : SQ ÝÑ E2‚,0

2061 bijective in degrees ď n ` 2, which is also injective in degree n ` 3 simply because the domain
2062 is evenly-graded. From the observation that the rectangle r0, n ` 2s ˆ r0, ns in E2 “ E2‚,0 b E20,‚
2063 is isomorphic to the tensor product SQ b ∆P, since the differentials on ∆Păn are completely
„
2064 determined by the transgressions already described, it follows dn`2 must map Pn`1 ÝÑ Qn`2
0,n`1 n`2,0
2065 or else a nontrivial subquotient of one of the two squares En`2 and En`2 would survive to
À p,q
2066 E8 , contrary to the assumption that the triangle p`qďn`2 E8 is trivial. This concludes the
2067 induction.

2068 There also is is a variant which produces the same conclusion, not assuming that P is oddly-
2069 graded, but instead that it transgresses, which yields an alternate, more topological proof of
2070 Theorem 6.6.1.

2071 Theorem 6.6.7 (Borel little transgression theorem). Let k be a commutative ring. Suppose pEr , dr q is a
2072 spectral sequence of bigraded k-algebras such that

2073 • E2 admits a tensor decomposition E2‚,0 b E20,‚ ,

2074 • the k-algebra E20,‚ – ∆pzα q is free as a k-module and admits a simple system of generators zα ,

2075 • these zα transgress in the spectral sequence, and

0,0
2076 • the final page E8 “ E8 – k is trivial.

2077 Then E2‚,0 – Srτzα s is the symmetric algebra on the transgressions of the zα .

2078 This in particular applies to the case ΛP “ ∆P to yield Theorem 6.6.1, once we know the
2079 odd-degree generators P in that spectral sequence transgress.

2080 Alternate proof of Theorem 6.6.1. Considering the homology Serre spectral sequence of the universal
2081 bundle G Ñ EG Ñ BG, Remark 6.2.12 shows the homological primitives PH˚ pGq ă H˚ pGq are
2082 all in the image of the transgression. Because H˚ pGq – H ˚ pGq and H˚ pBGq – H ˚ pBGq on the
2083 level of graded vector spaces and the homological and cohomological transgressions are dual
2084 (Remark 3.3.17), this means all elements of PG transgress in the cohomological Serre spectral
2085 sequence. Thus, by Theorem 6.6.7, we have H ˚ pBGq – SrτPs.
 Chapter 6. The cohomology of Lie groups and classifying spaces 82

2086 We already know an acyclic algebra of the form ∆P b SΣP, the Koszul complex KP “ ΛP b
2087 SΣP of the previous section 6.5, and the strategy behind the proof of Theorem 6.6.7 will be to
2088 construct a map of spectral sequences that shows E2 – KP as a bigraded SΣP-module. Recall that
2089 we constructed KP by analogy with the Serre spectral sequence of S1 Ñ ES1 Ñ BS1 , and our tack
2090 in constructing the desired spectral sequence, following Zeeman, will be to reverse that process.
2091 We equip KP with the bigrading pKPq p,q by total degree induced additively from the grading of
2092 P.

2093 Proof of Theorem 6.6.7 [Zee58],[McC01, Thm. 3.27, p. 85]. Select a k-basis vα of P and for each vα
|v |`1,0
2094 lift the transgression d|vα |`1 vα to an element τvα of E2 α . Since ∆P admits a basis comprised of
2095 ordered monomials on the vα , this k-linear map τ : P ÝÑ E2‚,0 extends uniquely to a differential
2096 on ∆P b E2‚,0 vanishing on E2‚,0 , which we will again call τ. If we equip pE2 , τq with the horizontal
2097 filtration, then by construction, the associated filtration spectral sequence for r ě 2 is just pEr , dr q
2098 again.
We construct a map of bigraded k-modules λ : pKP, dq ÝÑ pE2 , d2 q uniquely by extending the
map given on the multiplicative generators S1 ΣP ‘ Λ1 P by

λ : SΣP b ΛP ÝÑ E2‚,0 b ∆P,

1 b vα ÞÝÑ 1 b vα ,
dvα b 1 ÞÝÑ τvα b 1.

2099 and of course fixing the unity. To see this is a cochain map, one need only check on generators,
2100 the differentials vanish on SΣP and E2‚,0 , and τλv` α “ Kτvα K“ λdvα by construction.
˘ Because this
2101 map preserves the bigrading, it induces a map λr : p Er , dr q ÝÑ pEr , dr q of filtration spectral
2102 sequences, where the source is the spectral sequence of KP.
2103 Because SΣP is a free k-CGA in any characteristic and we extended λ multiplicatively from
2104 a map on the generators ΣP, the row restriction λ : SΣP ÝÑ E2‚,0 is a ring homomorphism. The
2105 column restriction λ : ΛP ÝÑ ∆P is a linear isomorphism, because both ∆P and ΛP admit a
2106 k-basis of ordered monomials in the vα . (If the characteristic of k is not 2, then ∆P “ ΛP, so this
2107 column map is a ring isomorphism, but it need not be in characteristic 2, because then it is not
2108 required that v2α “ 0 in ∆P.)
2109 Now, the limiting map λ8 : K E8 ÝÑ E8 is by construction the identity map on k, and as we
2110 noted, the map λ2 : K E20,‚ “ ΛP ÝÑ ∆P “ E20,‚ is a linear isomorphism, so the Zeeman–Moore
2111 comparison theorem A.5.7 applies to tell us λ‚,0 ‚,0
2 : SΣP ÝÑ E2 is a linear isomorphism. But we
2112 observed this is a ring homomorphism, so E2‚,0 – SrτPs as a ring.

2113 Complements
2114 The rest of this section is devoted to related results of Borel which we will not however have need
2115 of in the sequel. For example, there is also a dual result whose proof falls out of what we have
2116 already done.
2117 Corollary 6.6.8 (Borel [Bor54, Thm. 6.1, p. 297]). Let k be a commutative ring and Q – ktyα u an
2118 evenly-graded free k-module. Suppose pEr , dr q is a spectral sequence of bigraded k-algebras such that
2119 • E2 admits a tensor decomposition SQ b E20,‚ , with E2 – SQ the symmetric algebra on Q and
0,0
2120 • the final page E8 “ E8 – k is trivial.
 Chapter 6. The cohomology of Lie groups and classifying spaces 83

2121 Then E20,‚ – ∆rvα s admits a simple system of transgressive generators vα such that τvα “ yα .

2122 Proof. Since E8 – k, each yα must eventually be annihilated by some differential. There can be
2123 no generators of degree 1 since all differentials out of this box are zero. The generators Q2 of
2124 degree 2 can only be annihilated by elements E20,1 . Since d2 is a differential, it follows E3‚,0 –
2125 SQ{pQ2 q – SrQě3 s is the subalgebra on generators of degree three or more. Assume inductively
2126 that Qr survives in Er . Since it doesn’t survive to E8 , it must be annihilated by the elements Er0,2 ,
2127 which are hence all transgressive. Inductively continuing this way, each element of Q must be
2128 killed by a transgressive element of E20,‚ .
2129 Now define a cochain map λ : SQ b Λrvα s ÝÑ E2 as in the proof of Theorem 6.6.5. Applying
2130 the Zeeman–Moore comparison theorem A.5.7 again, one sees the restriction Λrvα s ÝÑ ∆rvα s ď
2131 E20,‚ must a linear isomorphism, so that E20,‚ “ ∆rvα s.

2132 Combining the two, one has the following.

2133 Corollary 6.6.9. Let k be a commutative ring. Suppose pEr , dr q is a spectral sequence of bigraded k-algebras
2134 such that

2135 • E2 admits a tensor decomposition E2‚,0 b E20,‚ ,

0,0
2136 • the final page E8 “ E8 – k is trivial.

2137 Then the following are equivalent:

2138 • The k-algebra E20,‚ – ∆rzα s admits a simple system of transgressive generators zα .

2139 • The k-algebra E2‚,0 – Sryα s is a symmetric algebra on generators yα .

2140 If the statements hold, the zα and yα are related by τzα “ yα .

À p,q
2141 Remarks 6.6.10. (a) In fact, there is a strengthening requiring only that the triangle p`qďn E8 – k
2142 is trivial, the map from ∆rzα s is a bijection in degrees ď n ´ 1 and an injection in degree n and
2143 that the map from Sryα s is a bijection in degrees ď n and an injection in degree n ` 1, as in the
2144 proof of Theorem 6.6.5.
2145 (b) The full strength version of Corollary 6.6.9 reflects a sort of duality between symmetric al-
2146 gebras and symmetric algebras, called Koszul duality. This topic in algebra is responsible for the
2147 criterion Theorem 8.3.1 for equivariant formality which we will encounter much later.
2148 To round out this subsection we include without proof some other finite-characteristic results
2149 and conclude with some historical remarks. As regards the applicability of the little transgres-
2150 sion theorem 6.6.7 in characteristic ‰ ` 2, it is not˘ universal. Borel found the following example.
2151 Recall that the cohomology rings H ˚ Spinpnq; F2 do admit simple systems of generators (Propo-
2152 sition 6.1.21).
` ˘
˚ Spinpnq; F . These are all transgressive
2153 Theorem 6.6.11. Consider a simple system
` of generators
˘ for H 2
2154 if and only if n ď 9. Accordingly, H ˚ BSpinpnq; F2 is a polynomial ring if and only if n ď 9.

2155 Specifically, as described in Proposition 6.1.21, one has

H ˚ Spinp10q; F2 – ∆rv3 , v5 , v6 , v7 , v9 , z15 s,

` ˘
 Chapter 6. The cohomology of Lie groups and classifying spaces 84

2156 but there is an element u of degree 15, congruent to z15 modulo decomposables, which has
2157 d10 pu b 1q “ d10 pv9` b 1q ¨ p1 b v˘6 q. The nontransgression of this u is related to the failure of the
2158 homology ring H˚ Spinp10q; F2 to be an exterior algebra, as described in Example A.3.5.
2159 Nevertheless, in the universal bundle for the limiting group Spin “ lim ÝÑ Spinpnq, all generators
2160 transgress again. One has then the following corollary of Theorem 6.1.22.

2161 Theorem 6.6.12 ([BCM, Thm. 6.10, p. 55]). The mod 2 cohomology ring of BSpin is given by

H ˚ pBSpin; F2 q “ F2 rw j : j ‰ 2` ` 1s

2162 and that of BSO by

H ˚ pBSO; F2 q “ F2 rw j s,
2163 the map H ˚ BSO ÝÝ H ˚ BSpin induced by Spin  SO being the obvious surjection. The transgressions
2164 are given, for j odd, by
`
τpv2j q “ w2` j`1 .

2165 Borel also found a complement in characteristic not equal to 2, showing even-dimensional
2166 spheres (other than S0 ) can’t show up as factors in the fiber of a bundle with contractible total
2167 space.

2168 Theorem 6.6.13. Let k be a ring of characteristic not equal to 2. Suppose pEr , dr q is a spectral sequence of
2169 bigraded k-algebras such that

2170 • E2 admits a tensor decomposition E2‚,0 b E20,‚ such that E20,‚ “ ∆rpα s for a simple system ppα q of
2171 generators with p2α “ 0 and
0,0
2172 • the final page E8 “ E8 – k is trivial.

2173 Then all of the pα are of odd degree.

2174 Explicitly, in this case, we have in the hypothesis that

E20,‚ “ ∆rpα s “ Λ p j : |pα | odd b

â
Srpα s pp2 q
“ ‰ L
α
|pα |even

2175 and in the conclusion that ∆rpα s “ ΛP for P oddly graded. Clearly, then, if one wanted to
2176 generalize the “simple system of generators” to even-degree generators in characteristic p ą 2,
2177 asking that they be nilsquare would not be the way to go. Postnikov would find the proper
2178 strategy in 1966 to generalize Theorem 6.6.7 to odd characteristic.

2179 Definition 6.6.14 ([Pos66, p. 36]). Let p be an odd prime. A graded commutative F p -algebra F is
2180 said to admit a p-simple system of generators pzα , y β qαPA,βPB , where the zα are of odd degree and
2181 the y β even, if F is spanned as an F p –vector space by the basis of ordered monomials

zα1 ¨ ¨ ¨ zαm y`β11 ¨ ¨ ¨ y`βnn ,

2182 where the indices αi and β j are strictly increasing and the exponents ` j ď p ´ 1.

2183 Theorem 6.6.15 ([Pos66, p. 36]). Let p be an odd prime. Suppose pEr , dr q is a spectral sequence of
2184 bigraded F p -algebras such that
 Chapter 6. The cohomology of Lie groups and classifying spaces 85

2185 • E2 admits a tensor decomposition E2‚,0 b E20,‚ such that E20,‚ admits a p-simple system pzα , y β q of
2186 transgressive generators, the zα being of odd degree and the y β even,
0,0
2187 • the final page E8 “ E8 – F p is trivial.
2188 Then E2‚,0 is the free commutative algebra on the transgressions
p´1
xα :“ τzα , v β :“ τy β , u β :“ τpv β b y β q.

2189 Explicitly, E2‚,0 – F p rxα s b F p ru β s b Λrv β s

p´1
2190 Sketch of proof. Assume first the elements τy β b y β transgress. Then we will be able to find the
2191 proper bigraded comparison complex admitting a chain map to pE2 , τq, where τ is a choice of
2192 transgression in E2 , and the proof will proceed exactly as the proof of Theorem 6.6.7.
2193 For the odd generators zα , one retains the bigraded Koszul complex Krzα s as before, but for
2194 each even generator y β one introduces a tensor-factor (not a DGA)
p´1
Λrv̄ β s b F p rū β s b Frȳ β s pȳ p´1 q, dy β “ v̄ β , dpv̄ β b ȳ β q “ ū β
` ˘ L
β

2195 bigraded with the expected degrees with v β and u β in the bottom row and y β in the left column.
2196 Collecting all of these, the assignment zα ÞÑ zα , ȳ β ÞÑ y β , v̄ β ÞÑ v β , ū β ÞÑ u β is by definition a
2197 chain map, and restricts to a ring map F p rτzα , ū β s b Λrv̄ β s ÝÑ E2‚,0 because on this subdomain it
2198 is defined by unique extension from free CGA generators.
p´1
2199 That the τy β b y β must transgress is an induction like that in Theorem 6.6.5. To prove it for
2200 maxβ |y β | “ n ` 1, inductively assume it for degrees ď n and as well that E2‚,0 agrees up to degree
2201 ppn ` 1q ` 1 with the free CGA on
` ˘ ` ˘
xα , vβ for |y β | ď n ` 1 , uβ for |y β | ď n ´ 1 .
2202 It follows from
“ this assumption
‰ and
“ the eventual triviality
‰ of E8 – F p that on the page Epn`1qpp´1q`1 ,
2203 the rectangle 0, pn ` 1qp ` 1 ˆ 0, pn ` 1qpp ´ 1q ´ 1 is trivial, the cancellations being due solely
p´1
2204 to the elements in the induction hypothesis. This means that the differentials of the τy β b y β
n`1,pn`1qpp´1q
2205 for |y β | “ n ` 1 which land in this rectangle must be trivial, and so the map τ : Epn`1qpp´1q`1 ÝÑ
pn`1qp`2,0 p´1
2206 Epn`1qpp´1q`1 must be an isomorphism, showing all the new τy β b y β also transgress.

2207 Historical remarks 6.6.16. Borel’s proof of Theorem 6.6.7 and Theorem 6.6.5 was more subtle than
2208 ours in some ways. For one, the Zeeman–Moore theorem was not available to him, so he did not
2209 construct a comparison map, but explicitly, inductively, and through careful bookkeeping ruled
2210 out the possibility of H ˚ pBGq being anything other than a polynomial ring, keeping track at the
2211 same time of what elements of ΛP transgressed and ultimately determining them to be only the
2212 primitives P themselves.
2213 More historically remarkably, in determining H ˚ pBGq Borel did not have access to BG itself. In
2214 1952, it was only known in general that n-universal principal bundles Epn, Gq ÝÑ Bpn, Gq existed
for each n P N with πi Epn, ˚
2215
˚
` ˘ Gq “ 0 for i ď n. Borel’s H pBGq is actually defined as the inverse limit
2216 of the rings H Bpn, Gq , known cohomology rings of already-existing manifolds. Resultingly, for
2217 numerous
` topological
˘ applications in which we cavalierly deploy BG, Borel must instead invoke
2218 H ˚ Bpn, Gq for n sufficiently large.
2219 While we imagine the alternate proof of Theorem 6.6.1 following Theorem 6.6.7 was known,
2220 we know no reference.
 Chapter 6. The cohomology of Lie groups and classifying spaces 86

2221 Characteristic classes

2222 Borel’s Theorem 6.6.1, the mod 2 addendum Section 6.6.2.(a), and knowledge of the cohomol-
2223 ogy rings of classical groups from Section 6.1 make instantly available a great deal of algebraic
2224 information about classifying spaces.
Corollary 6.7.1. Let k “ Zr1{2s. The cohomology rings of the classifying spaces of the classical groups are
` ˘
H ˚ BOpnq; F2 – F2 rw1 , . . . , wn s, deg w j “ j,
˚
` ˘
H BSOpnq; F2 – F2 rw2 , . . . , wn s, deg w j “ j,
˚
` ˘
H BUpnq; Z – Zrc1 , . . . , cn s, deg c j “ 2j,
` ˘
H ˚ BSUpnq; Z – Zrc2 , . . . , cn s, deg c j “ 2j,
˚
` ˘
H BSppnq; Z – Zrq1 , . . . , qn s, deg q j “ 4j,
˚
` ˘
H BSOp2n ` 1q; k – krp1 , . . . , pn´1 , pn s, deg p j “ 4j,
` ˘
H ˚ BSOp2nq; k – krp1 , . . . , pn´1 , es, deg p j “ 4j, deg e “ 2n.

2225 Definition 6.7.2. The w j in the preceding corollary are the Stiefel–Whitney classes, the c j the
2226 Chern classes, the q j the symplectic Pontrjagin classes, the p j the Pontrjagin classes, and e the
2227 Euler class.
2228 Remark 6.7.3. For G P tU, Sp, SOu, the inclusions Gpnq ãÑ Gpn ` 1q preserve ` c j , q j , p j respec-
˘
2229 tively for j ď n and annihilate c , q , p , with the exception that H ˚ BSOp2n ` 1q ÝÑ
` ˘ n`1 n`1 n`1
2230 H ˚ BSOp2nq takes pn ÞÝÑ e2 .
2231 The Pontrjagin classes and Euler class as ` described˘ above are ` actually integral
˘ in that they
2232 are in the image of the canonical map H ˚ BSOpmq; Z ÝÑ H ˚ BSOpmq; Zr1{2s . These classes
2233 carry certain
` well-known ˘ relations.
` For ˘example, the inclusion Upnq SOp2nq induces a
2234 map H 2n BSOp2nq; Z ÝÑ H 2n BUpnq; Z carrying e ÞÝÑ c , and mod-2 coefficient reduction
` ˘ ` ˘ n
2235 H n BSOpnq; Z ÝÑ H n BSOpnq; F2 takes e ÞÝÑ wn .
2236 All of these rings can also be calculated independently from the result Corollary 5.4.7 that
2237 H pBGq – H ˚ pBTqW and an understanding of the Weyl group action on BT. For example, the
˚

2238 existence of the Euler class can be seen as a result of the fact that WSOp2n`1q “ t˘1un ¸ Sn and
2239 WSOp2nq is the subgroup St˘1un ¸ Sn , where St˘1un ă t˘1un is the index-two subgroup whose
2240 elements contain an even number of ´1 entries. The product e “ t1 ¨ ¨ ¨ tn P Zrt1 , .`. . , tn s is invariant ˘
2241 under St˘1un but not under all of t˘1un , and as a result does not occur in H ˚ BSOp2n ` 1q ; its
2242 square pn “ t21 ¨ ¨ ¨ t2n is however invariant under the larger group’s action.
2243 The cohomology classes of Definition 6.7.2, elements of a cohomology ring BG only known
2244 after 1955 to globally exist, are abstract manifestations of objects associated to vector bundles
2245 which were defined in the 1930s and early 1940s by their namesakes.9
2246 Definition 6.7.4. Let E ÝÑ B be a principal G-bundle and χ : B ÝÑ BG a classifying map. Given
2247 c P H ˚ pBGq, its pullback χ˚ pcq P H ˚ pBq is written c˚ pEq and called a characteristic class of
2248 E ÝÑ B.
2249 These characteristic classes are functorial invariants of principal bundles: because the uni-
2250 versal bundle is terminal, a map of bundles induces a commutative triangle of maps of base
2251 spaces.
9 With the obvious exception of the Euler class.
 Chapter 6. The cohomology of Lie groups and classifying spaces 87

2252 Proposition 6.7.5. Let E Ñ B be a principal G-bundle, let f : B1 Ñ B be a continuous map, and let
2253 c P H ˚ pBGq. Then the pullback bundle f ˚ E satisfies

cp f ˚ Eq “ f ˚ cpEq P H ˚ pBq.
ξ
2254 Given a vector bundle F Ñ V Ñ B with transition functions in a linear group G, there is an
2255 associated principal G-bundle G Ñ P Ñ B as described in Appendix B.1.3, and one can associate
2256 to V Ñ B the characteristic classes of P Ñ B,

cpVq :“ cpPq,

2257 calling them the characteristic classes of the vector bundle V Ñ B. For example

2258 • if ξ : V Ñ B is a quaternionic vector bundle it defines symplectic classes q j pξq P H 4j pB; Zq,

2259 • if ξ is a complex vector bundle one has Chern classes c j pξq P H 2j pB; Zq,

2260 • if ξ is a real vector bundle one has Pontrjagin classes p j pξq P H 4j pB; Zq and Stiefel–Whitney
2261 classes w j pξq P H j pB; F2 q, and

2262 • if ξ is an orientable vector bundle with fiber F “ Rn , it has an Euler class epξq P H n pB; Zq,
2263 and the first Stiefel–Whitney class w1 can be shown to vanish.

2264 A smooth manifold M determines a tangent bundle TM Ñ M, which thus defines a charac-
2265 teristic class
cpMq :“ cpTMq P H ˚ pMq
2266 for each characteristic class c of the tangent bundle. For example, we can equip TM with a
2267 Riemmannian or Hermitian metric to reduce its structure group to Opnq or Upnq, so all smooth
2268 manifolds carry Pontjagin and Stiefel–Whitney classes, orientable smooth manifolds carry an
2269 Euler class epMq P H top pMq, and almost complex manifolds carry Chern classes.
2270 These classes turn out to be well-defined invariants of the topological manifold underlying M
2271 in that they are independent of the chosen metrics and smooth or almost complex structures. To
2272 see at least that the metrics are irrelevant, one way to proceed is to note that the Gram–Schmidt
2273 construction can be seen as a product decomposition [BT82, Ex. 6.5(a)]

SLpn, Rq “ SOpnq ¨ F,

2274 where F is the contractible space of positive-definite symmetric matrices. If we consider ESOpnq
2275 to be ESLpn, Rq, which is valid, as discussed in Section 2.2, since SOpnq and SLpn, Rq are Lie
2276 groups, the former closed in the latter, then taking quotients yields the bundle

F ÝÑ BSLpn, Rq ÝÑ BSOpnq,

2277 with fiber F contractible, so that BSLpn, Rq » BSOpnq. Similar homotopy equivalences hold for
2278 other classifying spaces of linear groups, so one can dispense with the metrics at the negligible
2279 cost of viewing the characteristic classes instead as arising in BGLpn; Fq or BSLpn; Fq for F P
2280 tR, C, Hu.
2281 Assume now M is compact and oriented. A characteristic class c in H top pM; Zq – Z is then
2282 some integer multiple n ¨ rMs˚ of the cohomological fundamental class rMs˚ ; alternately, evalu-
2283 ation of c against the homological fundamental class rMs yields an integer n. These integers are
 Chapter 6. The cohomology of Lie groups and classifying spaces 88

2284 called characteristic numbers of the manifold, and the data given by characteristic numbers for
2285 a real manifold can be seen as the composition
˘ χ˚ „
H n BSOpnq; Z ÝÑ H n pM; Zq ÝÑ Z,
`

2286 where χ : M ÝÑ BSOpnq is the characteristic map of the associated principal SOpnq-bundle.
2287 The Pontrjagin numbers are the images under this composition of the degree-n level of the
2288 subring Zrp1 , . . . , pn s, and the Euler characteristic can be seen as the image of e:

2289 Theorem 6.7.6. Let M be a smooth, compact, oriented n-manifold. Then the Euler class e P H n pM; Zq
2290 and cohomological fundamental class rMs˚ P H n pM; Zq and the Euler characteristic χpMq P Z satisfy the
2291 relation
e “ χpMq ¨ rMs˚ .

2292 This the reason behind the nomenclature Euler class. This equivalence also yields an out-
2293 landishly complicated way of seeing the Euler characteristic of an odd-dimensional closed man-
2294 ifold is zero.
2295 In Appendix C, we will use equivariant cohomology to describe simple conditions under
2296 which characteristic numbers vanish.

2297 Maps of classifying spaces

2298 The machine for computing H ˚ pG{Kq depends critically on an understanding of the map

ρ˚ “ pBiq˚ : H ˚ pBGq ÝÑ H ˚ pBKq

2299 induced by the inclusion i : K ãÝÝÑ G; this understanding (in what by now should be starting to
2300 seem like a familiar theme) is also due to Borel [Bor53, §28].

2301 Maps of classifying spaces of tori

2302 To start, let i : S ãÝÝÑ T be an inclusion of tori. By functoriality, or else by taking ES “ ET and
2303 representing BS ÝÑ BT as the “further quotient” map ET{S ÝÝ ET{T, we have a bundle map

S /T
i

 
ES
» / ET

 
BS
Bi / BT,

rr , d˜r q ÝÑ pEr , dr q between the spectral sequences of the bundles.

` ˘
2304 which induces a map ψr : pE
„
2305 Because these sequences both collapse on the third page, ψr is just an isomorphism H 0 pETq ÝÑ
2306 H 0 pESq “ Z for r ě 3, so we may as well drop page subscripts and consider the lone interesting
2307 map ψ “ ψ2 , which by Theorem 3.3.4 is

ψ “ pBiq˚ b i˚ : H ˚ pBTq b H ˚ pTq ÝÑ H ˚ pBSq b H ˚ pSq.

 Chapter 6. The cohomology of Lie groups and classifying spaces 89

2308 Because, by the definition of a chain map, we have dψ “ ψd,˜ and, as we have just seen, d : H 1 pSq ÝÑ
2309 H 2 pBSq and d˜: H 1 pTq ÝÑ H 2 pBTq are group isomorphisms, we have the commutative square
i˚
H 1 pSq o H 1 pTq
(6.7)

„
 pBiq˚

H 2 pBSq o H 2 pBTq.

2310 Thus i˚ : H 1 pTq ÝÑ H 1 pSq is conjugate through the transgression isomorphisms to pBiq˚ : H 2 pBTq ÝÑ
2311 H 2 pBSq. Since H 2 pBTq generates H ˚ pBTq as an algebra, and pBiq˚ is a ring homomorphism, this
2312 means pBiq˚ is determined uniquely by i˚ . This i˚ , in turn, is described by i in a transparent way.
2313 It is dual to the map i˚ : H1 S ÝÑ H1 T, or equivalently to the map π1 piq.
In the case most critical for us later, S will just be a circle, which we will identify with the
standard complex unit circle S1 ă Cˆ . Similarly identify T with pS1 qn . Then i : S ãÝÝÑ T can be
written as
i : S1 ÝÑ pS1 qn ,
z ÞÝÑ pz a1 , . . . , z an q,
2314 where the exponent vector ~a P Zn is a list of integers with greatest common divisor 1, so that i
2315 is injective.10 If x j P π1 pTq “ H1 pTq is the fundamental class of the jth factor circle and y P H1 pSq
2316 the fundamental class of S, then ÿ
i˚ : y ÞÝÑ aj xj.
2317 Let px˚j q be the dual basis for H 1 pTq and y˚ P H 1 pSq the cohomological fundamental class. Then
2318 the dual map i˚ : H 1 pTq ÝÝ H 1 pSq in cohomology takes x˚j ÞÝÑ a j y˚ since
` ˘ `ÿ ˘
pi˚ x˚j qy˚ “ x˚j i˚ y “ x˚j a` x` “ a j .

2319 Put another way, the matrix of i˚ is the transpose of the matrix of i˚ . Write s “ dy P H 2 pBSq and
2320 u j “ d2 x˚j P H 2 pBTq so that H ˚ pBSq “ Zrss and H ˚ pBTq “ Zr~us. Then, the square above implies
2321 that pBiq˚ pu j q “ a j s, so that if pp~uq P Zr~us is any polynomial,

pBiq˚ pp~uq “ ppa1 s, . . . an sq “ ppa1 , . . . , an qsdeg p .

2322 Maps of classifying spaces of compact, connected Lie groups

2323 Let K ãÝÝÑ G be an inclusion of compact, connected Lie groups. If S is a maximal torus of K, then
2324 there exists a maximal torus T of G containing S. Through the functoriality of the classifying
2325 space functor B and cohomology, this square of inclusions gives rise to two further commutative
2326 squares:
 i / T Bi / BT
pBiq˚
H ˚ pBSq o H ˚ pBTq
_
S _ BS O O
ùñ ùñ
   
K  /G BK / BG ? ?
H ˚ pBKq o
ρ
H ˚ pBGq.
ρ˚

10 This vector ~a is only well-defined up to the choice of identifications S – S1 and T – pS1 qn , but will suffice for our
later applications.
 Chapter 6. The cohomology of Lie groups and classifying spaces 90

2327 The vertical maps in the last square are inclusions by Corollary 5.4.7. Thus ρ˚ can be computed
2328 as the composition
„ pBiq˚
H ˚ pBGq ÝÑ H ˚ pBTqWG ÝÝÝÑ H ˚ pBSq;
2329 it follows from the commutativity of the square that the image lies in H ˚ pBSqWK – H ˚ pBKq.
2330 Example 6.8.1. Let G “ Up4q and K “ Spp2q,” identified
ı as a subgroup of G through the injective
2ˆ2 α ´ β̄
2331 ring map H C given by α ` jβ ÞÝÑ β ᾱ . A standard maximal torus for G is given by
2332 the subgroup T “ Up1q4 of diagonal unitary matrices, which meets K in the subgroup
! )
S “ diagpz, z̄, w, w̄q P Up1q4 : z, w P S1 .

2333 With respect to the expected basis of H1 pTq and the fundamental classes of the factor circles
2334 w “ 1 and z “ 1 of S, and the dual bases in H 1 , the maps H1 pSq ÝÑ H1 pTq and H 1 pSq ÐÝ H 1 pTq
2335 are given respectively by
» fi
1 0
„ 
—´1 0ffi 1 ´1 0 0
—
– 0
ffi and .
1fl 0 0 1 ´1
0 ´1

2336 By the commutative square (6.7), the second matrix also represents H 2 pBSq ÐÝ H 2 pBTq with
2337 respect to the transgressed bases t1 , t2 , t3 , t4 of H 2 pBTq and s1 , s2 of H 2 pBSq.
2338 The Weyl group of Up4q is the symmetric group S4 on four letters ` acting
˘ on T and hence
2339 BT by permutation of the four coordinates. It follows that when H ˚ Up4q is conceived as the
2340 invariant subring H ˚ pBTqS4 of H ˚ pBTq, it is generated by the elementary symmetric polynomials
2341 σ1 , σ2 , σ3 , σ4 in t1 , t2 , t3 , t4 , lying in respective degrees 2, 4, 6, 8. These are the first four Chern classes
2342 cj.
The Weyl group of Spp2q is the group t˘1u2 ¸ S2 , acting on H 1 pSq and hence on H 2 pBSq “
Qts`1 , s2 u by ˘ negating and/or switching the two coordinates. It follows the invariant subring
WSpp2q
˚
H BSpp2q – H pBSq ˚ is generated by q1 “ s21 ` s22 and q2 “ ps1 s2 q2 . These are the first two
4
symplectic Pontrjagin classes. The generators c j exhibit the following properties under H ˚ pBTqS ãÝÝÑ
H ˚ pBTq ÝÑ H ˚ pBSq:

c1 “ t1 ` t2 ` t3 ` t4 ÞÝÑ ps1 ´ s1 q ` ps2 ´ s2 q “ 0,

1` 1`
c2 “ σ12 ´ σ1 pt21 , t22 , t23 , t24 q ÞÝÑ 0 ´ ps21 ` s21 ` s22 ` s22 q “ ´ps21 ` s22 q “ ´q1 ,
˘ ˘
2 2
c3 “ pt1 ` t2 qt3 t4 ` t1 t2 pt3 ` t4 q ÞÝÑ p0 ¨ ´s22 q ` p´s21 ¨ 0q “ 0,
c4 “ t1 t2 t3 t4 ÞÝÑ s21 s22 “ q2 .
` ˘ ` ˘
2343 That is, H ˚ BUp4q ÝÑ H ˚ BSpp2q ` ˘is surjective,
` a ˘fact we will later be able to see as a conse-
2344
˚
quence of the surjectivity of H Up4q ÝÑ H Spp2q . ˚

2345 Example 6.8.2. Let G “ Spp2q and K “ S “ SOp2q, identified as a subgroup of G through the
2346 standard inclusion R ãÝÝÑ H. One maximal torus T of Spp2q containing S is that generated by S
2347 and "„  *
1 cos θ i sin θ
S “ : θ P r0, 2πs .
i sin θ cos θ
 Chapter 6. The cohomology of Lie groups and classifying spaces 91

With respect to the basis rSs, rS1 s of H1 pTq, and the

“ ‰ dual “bases‰ in H 1 , the maps H1 pSq ÝÑ H1 pTq
and H 1 pSq ÐÝ H 1 pTq are given respectively by 10 and 1 0 . By (6.7) again the second matrix
also represents H 2 pBSq ÐÝ H`
2 pBTq with respect to the transgressed bases t , t of H 2 pBTq and t
˘ 1 2 1
of H 2 pBSq. Generators for H BSpp2q are q1 , q2 as in Example 6.8.1, and we have

q1 “ t21 ` t22 ÞÝÑ t21 ,

q2 “ t21 t22 ÞÝÑ t21 ¨ 0 “ 0.

2348 This is an instance of a general result about the map ρ˚ in the event G is semisimple and S a
2349 circle, which we will later use in determining the rings H ˚ pG{S1 q.

2350 Lemma 6.8.3. Let K be a semisimple Lie group containing a circle S. The image of HK˚ ÝÑ HS˚ – Qrss
2351 contains s2 P HS4 .

2352 Proof. Let T be a maximal torus of K containing S, so that HK˚ ÝÑ HS˚ factors as HTW ãÑ HT˚ Ñ HS˚ ,
2353 where W is the Weyl group of K, by the results of Section 6.8.2. Identifiying HT˚ “ Qru1 , . . . , un s
2354 and HS˚ “ Qrss, by Section 6.8.2, the exponents a j of the inclusion S1 ãÑ T – pS1 qn determine the
2355 map HT4 ÝÑ HS4 according to the formula ρ˚ qp~uq “ qp~aqs2 .
2356 The elements of HT4 can be seen as quadratic forms on the vector space π1 T b Q – H1 pT; Qq –
2357 H2 pBT; Qq. So to show the map pHT4 qW ÝÑ HS4 is surjective regardless of the choice of S, it suffices
2358 to find a W-invariant definite bilinear form on π1 T b Q. But the Killing form Bp´, ´q : k ˆ k ÝÑ
2359 R is Ad-invariant and negative definite by Proposition B.4.12, so the restriction to t ˆ t is W-
2360 invariant, and the restriction to the diagonal is a W-invariant, negative definite quadratic form B
2361 on t. But π1 T embeds in t in a canonical way as the kernel of exp : t ÝÝ T, so the restriction of B
2362 to π1 T b Q corresponds to an element of H4K – pHT4 qW whose Q-span surjects onto HS4 .

2363 Historical remarks 6.8.4. The choice of notation ρ˚ for this important map follows historical prece-
2364 dent dating back to the heroic era of large tuples described in Historical remarks 2.3.9. Borel and
2365 later Hirzebruch canonically assigned the name ρpK, Gq to the map BK ÝÑ BG induced by an
2366 inclusion K ãÝÝÑ G and ρ˚ pK, Gq to the resulting map H ˚ pBGq ÝÑ H ˚ pBKq in cohomology.
2367 Chapter 7
2368 The cohomology of homogeneous spaces

2369 In this chapter, and again starting in Chapter 9, we discuss properties of a specific type of compact
2370 homogenous space G{K in terms of the transitively acting group G and the isotropy subgroup K.
2371 Now seems like a good time to formalize this setup.

2372 Definition. Let G be a compact, connected Lie group, and K a closed, connected subgroup. In
2373 this situation we call pG, Kq a compact pair of Lie groups.

2374 Our discussion will really be about properties of such pairs. Associated to a compact pair
2375 pG, Kq are three fiber bundles. The first, K Ñ G Ñ G{K, follows from Theorem B.4.3. The second
2376 is the Borel fibration G Ñ GK Ñ BK, which is a principal G-bundle. The third is the fibration
2377 G{K Ñ BK Ñ BG, where the projection Bi : BK ÝÑ BG can be seen as the “further quotient”
2378 map EG{K ÝÝ EG{G. Substituting the homotopy quotient GK for G{K when convenient, we can
2379 then (with some manipulation, to be described in Section 7.1.1) see that each three consecutive
2380 terms of the sequence
i j χ Bi
K ÝÑ G ÝÑ G{K ÝÑ BK ÝÑ BG (7.1)
2381 form a bundle up to homotopy. This section is devoted to a general discussion of the implications
2382 of this fiber sequence in the resulting cohomology sequence

i˚ j˚ χ˚ ρ˚
H ˚ pKq ÐÝ H ˚ pGq ÐÝ H ˚ pG{Kq ÐÝ HK˚ ÐÝ HG˚ . (7.2)

2383 It is a curious historical coincidence that the study of the cohomology of homogeneous spaces
2384 seems to break into three basic periods, the first studying the Leray spectral sequence of the first
2385 three terms, the second studying the Serre spectral sequence of the second three terms, and last
2386 studying the Eilenberg–Moore spectral sequence of the last three terms. It is the second period
2387 characterization that we employ in what follows, but all these maps will have some relevance for
2388 us.

2389 The Borel–Cartan machine

2390 We begin by introducing the diagrams and devices that will carry out all our computations.

92
 Chapter 7. The cohomology of homogeneous spaces 93

2391 The fiber sequence

2392 The five terms of (7.1), up to homotopy, form the labeled subdiagram in the following diagram
2393 of bundle maps, where the columns are bundles.

K /G
i
G
j
  
EK / GK / EG (7.3)
χ
  
/ BG
ρ
BK BK

2394 Here the middle row should be seen as

EG ˆK K ãÝÝÑ EG ˆK G ÝÝ EG ˆG G,

2395 the outer terms being homeomorphic to EG “ EK by Lemma 1.1.1, and the fiber inclusions from
2396 the preceding row given by g ÞÝÑ re0 , gs. The first and last columns are universal bundles and the
2397 second column is the Borel fibration. It is clear that j ˝ i and ρ ˝ χ are nullhomtotopic because they
2398 factor through EG. The classifying map ρ : BK ÝÑ BG is explicitly given by ρ “ Bi : eK ÞÝÑ eG.
While it is tempting at this point to recall that GK » G{K and identify (7.3) with (7.1), we
have not actually checked that this homotopy equivalence preserves any maps. The map j above
is easily seen to not give us any problems but the map χ, given by re, gsK ÞÝÑ eK, alarmingly
disagrees with the composition
» κ
GK ÝÑ G{K ÝÑ BK :
re, gsK ÞÝÑ Kg ÞÝÑ Kge0 ,

2399 where κ is the fiber over the point Ge0 P BG of the quotient map ρ : BK ÝÑ BG.
2400 As we do need this χ to discuss the cohomology of H ˚ pG{Kq, it becomes essential to repair
2401 this discrepancy without destroying the identification of j with π : G  G{K. At the end of the
2402 day, we will show the following diagram commutes:

G
π j
π∆
tq w  &
G{K » 0 EG ˆ G{K o » EG ˆ G ˆ EG »
/ G ˆ EG
K K

κ λ χ
$   
BK o » EG ˆ EG »
/ BK,
K

2403 where all horizontal maps are homotopy equivalences. The reader may enjoy working out the
2404 maps for herself before reading the explanation to come. Alternately, those willing to take it on
2405 faith that everything works can just skip directly to Section 7.1.2.
To get the left two triangles of the diagram, follow Mimura and Toda [MT00, pp. 90–1], in
replacing G Ñ G{K Ñ BK with a bundle. G Ñ∆ EG ˆ G{K Ñ BK. Indeed, the homeomorphism
π κ π λ
 Chapter 7. The cohomology of homogeneous spaces 94

EG ˆG G{K « EG{K “ BK given by re, gKsG ÞÝÑ egK makes the diagonal map
π∆ : G ÝÑ EG ˆ G{K :
g ÞÝÑ pe0 g´1 , gKq,
2406 a natural substitution for π. Viewing π∆ as the orbit inclusion for a G-action and writing λ : EG ˆ
2407 G{K ÝÑ EG ˆG G{K « BK for the quotient map, both the top and bottom triangle on the left of
2408 the diagram then commute, where the horizontal homotopy equivalences are the factor inclusion
2409 and projection.
For the rest of the diagram, write GK “ G ˆK EG, and note EG ˆ G{K and GK “ G ˆK EG
both are quotients of a single space:
EG ˆ G{K ÐÝ EG ˆ G ˆ EG ÝÑ G ˆ EG :
` K ˘ K
pe, xKq ÑÝÞ e, rx, e1 sK ÞÝÑ rx, e1 sK ,
2410 where both projections are homotopy equivalences because their fibers are EG. Further, these
2411 three spaces admit free G-actions such that these projections are equivariant, given by
` ˘
peg´1 , gxKq ÑÝÞ eg´1 , rgx, e1 sK ÞÝÑ rgx, e1 sK .
2412 Note that the inclusion of a G-orbit on the left is π∆ and that on the right is j : G ÝÑ GK , giving
2413 us the top two triangles of the diagram.
2414 The bottom two squares arise from quotienting out these G-actions.
EG ˆ G{K o EG ˆ G ˆ EG / G ˆ EG
K K

  
λ EG ˆ G{K o EG ˆ G ˆ EG / pG{Gq ˆ EG χ
G G K K

« « «

   
EG{K o EG ˆ EG / K {EG
K
2415 Here all horizontal maps are homotopy equivalences, the first triple of vertical maps projects out
2416 the G-action, the second set of vertical maps are all homeomorphisms, and λ and χ are indicated,
2417 concluding the proof.

2418 The spectral sequence map

2419 The Borel approach ([Bor53, §22]) to understanding the cohomology of H ˚ pG{Kq depends on the
2420 G-bundle map between the second two bundles,
G G
j
 
GK / EG (7.4)
χ
 
BK / BG.
ρ
 Chapter 7. The cohomology of homogeneous spaces 95

2421 This bundle map induces a map from the spectral sequence pE rr , d˜r q of the universal bundle,
2422 which we now completely understand, to the spectral sequence pEr , dr q of the Borel fibration,
2423 which we do not. As GK » G{K, the latter sequence converges to H ˚ pG{Kq. We write

rr , d˜r q
pψr q : pEr , dr q ÐÝ pE

2424 for this map of spectral sequences. Recall from Appendix A.5.2 that these maps ψr : E
rr ÝÑ Er are
2425 DGA homomorphisms, meaning dr ˝ ψr “ ψr ˝ d˜r , and each descends from that on the previous
2426 page, meaning ψr`1 “ H ˚ pψr q. The map ψ2 : E2 ÐÝ E
r2 between second pages is

ρ˚ b id : H ˚ pBKq b H ˚ pGq ÐÝ H ˚ pBGq b H ˚ pGq,

„
r0,‚ ÝÑ
2427 where id H˚ pGq is the isomorphism E 2 E20,‚ of the leftmost columns and ρ˚ “ pBiq˚ : H ˚ pBKq ÐÝ
2428 H ˚ pBGq is the map E r‚,0 Ñ E‚,0 of bottom rows.
2 2
2429 It is a consequence of the following lemma that the map ρ˚ at least largely determines
2430 H ˚ pG{Kq.

2431 Proposition 7.1.1. Let G be a compact, connected Lie group whose primitive subspace PG ă H ˚ pGq is
2432 concentrated in degree ď q ´ 1. Then if G Ñ E Ñ B is a principal G-bundle, its SSS collapses at Eq`1 .

2433 Proof. Recall that the spectral sequence pE rr , d˜r q of the universal G-bundle collapses at E rq`1 “
2434 E
r8 “ Q. Because G Ñ E Ñ B is principal, it admits a bundle map to the universal bundle,
2435 as in (7.4) inducing a spectral sequence map pψr q : pE rr , d˜r q ÝÑ pEr , dr q, which is a cochain map,
2436 meaning dr ψr “ ψr d˜r . Thus the edge maps dr : Er 0,r´1
ÝÑ Err,0 all vanish for r ą q. Now, the dr also
2437 vanish on the bottom row Er‚,0 by lacunary considerations, and are antiderivations with respect to
2438 an algebra structure on Er descending from that of E2 “ H ˚ pBq b H ˚ pGq, so they vanish entirely
2439 for r ą q.

2440 In particular, since the edge homomorphisms of the universal bundle spectral sequence
rr , d˜r q are determined entirely composition by an isomorphism τ : PG ÝÑ „
2441 pE QpBGq restricting
2442 the transgression, it follows much of the structure of pEr , dr q is determined by the composition
2443 ρ˚ ˝ τ. In fact, in the next subsection we will show that this composition itself yields a differential
2444 d on E2 , the Cartan differential, such that H ˚ pE2 , dq – H ˚ pG{Kq and pEr , dr q is the filtration spectral
2445 sequence associated to the filtered DGA pE2 , dq, equipped with the horizontal filtration induced
2446 from HK˚ .

2447 Cartan’s theorem

2448 In this subsection, we prove Cartan’s theorem that the complex described above actually deter-
2449 mines H ˚ pG{Kq completely. To do so, we will have to briefly invoke a cochain-level description of
2450 the situation, and rather than use singular cochains, we compute cohomology with Dennis Sulli-
2451 van’s rational algebra APL of polynomial differential forms. This object plays an essential role in
2452 rational homotopy theory [FHT01, p. 121], but for our purposes, all ` we need
˘ to know about APL
2453
˚ ˚
is that it is a contravariant functor Top ÝÑ Q-CDGA such that H APL pXq – H pX; Qq for CW
2454 complexes X. That APL pXq is itself already a CGA and not merely a CGA up to homotopy is the
2455 key feature.
 Chapter 7. The cohomology of homogeneous spaces 96

2456 Temporarily taking a step back from homogeneous spaces, consider the universal bundle
2457 G Ñ EG Ñ BG. Lifting indecomposables, which is possible by Proposition A.4.3 since H ˚ pBGq is
2458 a free CGA, the transgression yields a map
„
PpGq ÝÑ QpBGq ãÝÝÑ H ˚ pBGq,
τ

2459 Since H ˚ pBGq is also a free CGA, there exists a CGA section i˚ : H ˚ pBGq ÝÑ APL pBGq, so we can
2460 lift τ to i˚ τ : PH ˚ pGq ÝÑ APL pBGq.
π
2461 Now consider a principal G-bundle G Ñ E Ñ B. This bundle is classified by some map
2462 χ : B ÝÑ BG, inducing a ring map χ˚ : APL pBGq ÝÑ APL pBq, and we can form the composition

χ˚ i˚ τ : PpGq Ñ H ˚ pBGq Ñ APL pBGq Ñ APL pBq

2463 Because H ˚ pGq “ ΛPpGq is a free CGA, we can extend this lifted transgression uniquely to an
2464 antiderivation on
C :“ APL pBq b H ˚ pGq
2465 which we will again call χ˚ i˚ τ and which vanishes on APL pBq. Similarly, the differential d B of
2466 APL pBq extends uniquely to an antiderivation on C annihilating Q b H ˚ pGq, which we again
2467 call d B . We consider C as a Q-CDGA with respect to the unique differential dC :“ d B ` χ˚ i˚ τ
2468 extending both d B and χ˚ i˚ τ. See Figure 7.1.2.

Figure 7.1.2: The differential of the algebra C “ APL pBq b H ˚ pGq as defined on primitives

..
.

5 P5 pGq

3 P3 pGq χ˚ i ˚ τ

χ˚ i ˚ τ

1 P1 pGq χ˚ i ˚ τ

0 A0PL pBq A1PL pBq A2PL pBq A3PL pBq A4PL pBq A5PL pBq A6PL pBq ¨ ¨ ¨
dB dB dB dB dB dB dB

0 1 2 3 4 5 6 ¨¨¨

Let pz` q be a basis of PpGq and set β ` “ pχ˚ i˚ τqz` for each `, so that we have

dC pα b 1q “ d B α b 1, α P APL pBq;
dC p1 b z` q “ β ` b 1.
` ˘ ` ˘
2469 The cochain maps APL pBq, d B Ñ pC, dC q Ñ H ˚ pGq, 0 induce ring homomorphisms H ˚ pBq Ñ
2470 H ˚ pCq Ñ H ˚ pGq.
 Chapter 7. The cohomology of homogeneous spaces 97

j π
2471 Theorem 7.1.3 (Chevalley [Car51][Kos51][Bor53, Thm 24.1, 25.1] ). Let G Ñ E Ñ B be a principal
„
2472 G-bundle, and let pC, dC q be as above. Then there exists an isomorphism λ˚ : H ˚ pC, dC q ÝÑ H ˚ pEq
2473 making the following diagram commute:

H ˚ pCq
>

!
H ˚ pBq λ˚ H ˚ pGq.
=

π˚  j˚
H ˚ pEq

2474 Proof. We want to construct a cochain map λ : C ÝÑ APL pEq into the algebra of polynomial
2475 differential forms on E (any CDGA calculating H ˚ pEq would do), which we will then show to be
2476 a quasi-isomorphism by showing it induces an isomorphism between later`pages of˘ the associated
2477 filtration spectral sequences. The spectral sequence corresponding to H ˚ APL pEq – H ˚ pEq will
j π
2478 be the SSS pEr , dr q of G Ñ E Ñ B with respect to APL cochains.
Note that by construction and by Corollary 6.6.3 a primitive z` P H r´1 pGq transgresses in
Er to dr rz` s “ rβ ` s. By the description in Proposition 3.3.15 of the transgression in the Serre
spectral sequence, this means there exists a form γ` P APL pEq such that rj˚ γ` s “ z` P H ˚ pGq and
d E γ` “ π ˚ β ` P APL pEq. Define λ on algebra generators by

λ : APL pBq b H ˚ pGq ÝÑ APL pEq,

αb1 ÞÝÑ π ˚ α,
1 b z` ÞÝÑ γ` .

Then λ is a cochain map by construction, for following through the formulas on generators,

d E λpα b 1q “ d E π ˚ α “ π ˚ d B α “ λdC pα b 1q;

d E λp1 b z` q “ d E γ` “ π ˚ β ` “ λdC p1 b z` q.

2479 Filter B by its p-skeleta B p , and E by the preimages π ´1 B p of these, and filter C and APL pEq
2480 by
i
Fp APL pEq “ ker APL pEq ÝÑ APL pπ ´1 B p´1 q .
à ` ˘
Fp C “ APL pBq b H ˚ pGq,
iěp

2481 Then λ preserves filtration degree for elements of H ˚ pBq, which is enough to see that it sends
2482 Fp C ÝÑ Fp APL pEq.
2483 Write pEr , dr q still for the spectral sequence of the filtration on APL pEq and p1Er , 1dr q for that
2484 of the filtration on C. The former is just the SSS of G Ñ E Ñ B using APL cochains (Theo-
2485 rem 3.3.4,Proposition 3.3.6),
E2 “ H ˚ pBq b H ˚ pGq.
2486 On the other hand, following through the reasoning in Corollary A.5.5 in this case, 1E0 is the
2487 associated graded algebra gr C – C, and 1d0 is the differential induced by dC “ d B ` χ˚ i˚ τ.
2488 Since χ˚ i˚ τ is induced by the transgression τ, it has filtration degree ě 2 on all elements it fails
2489 to annihilate outright, and so vanishes under the associated graded algebra construction, and
2490 likewise d B adds one to the filtration degree, so 1d0 “ 0 and 1E1 “ 1E0 – C. Thus 1d1 “ d B and
1
E2 – H ˚ pBq b H ˚ pGq – E2 .
 Chapter 7. The cohomology of homogeneous spaces 98

2491 Now that we know these pages can both be identified with H ˚ pBq b H ˚ pGq in a natural way, it
2492 remains to show λ2 : 1E2 ÝÑ E2 becomes the identity map under these identifications. But this is
2493 also the case by construction: the base elements α P APL pBq b 1 and λpα b 1q “ π ˚ α P APL pEq both
2494 become rαs b 1 in 1E2 “ E2 and the fiber elements 1 b z` P 1 b H ˚ pGq and λp1 b z` q “ γ` P APL pEq
2495 each become 1 b rj˚ γ` s “ 1 b z` .
2496 Since λ2 is a cochain isomorphism, it follows from the general principle Proposition A.5.8
2497 that ` ˘
λ˚ “ H ˚ pλq : H ˚ APL pBq b H ˚ pGq, dC ÝÑ H ˚ pEq
2498 is an isomorphism.

2499 Remark 7.1.4. We are fairly committed to a classical viewpoint in this work, but those with some
2500 grounding in rational homotopy theory might note that pSQpBKq b ΛPG, dq is a Sullivan model.
2501 The algebra C “ APL pBq b H ˚ pGq, although simpler than APL pEq, still involves the algebra
2502 APL pBq of polynomial forms on the base B, which though graded-commutative and hence simpler
2503 than the algebra of singular cochains on B, is still typically a large ring (if B is a CW complex of
2504 positive dimension, then dimQ APL pBq ě 2ℵ0 ), which we would rather replace with H ˚ pBq.
2505 The E2 page of the filtration spectral sequence associated to the filtration induced from the
2506 “horizontal” grading on APL pBq is the algebra we want, namely H ˚ pBq“b H ˚ pGq‰ equipped with
2507 the differential d2 vanishing on H ˚ pBq and sending z P PG to pχ˚ τqz “ pχ˚ i˚ τqz P H |z|`1 pBq.

2508 Definition 7.1.5. The algebra C “ H ˚ pBq b H ˚ pGq equipped with the antiderivation d extending

τ χ˚
PpGq Ñ QpBGq ãÑ H ˚ pBGq Ñ H ˚ pBq

2509 is the Cartan algebra of the principal bundle G Ñ E Ñ B.

2510 Remark 7.1.6. Observe that the Cartan algebra of a principal bundle G Ñ E Ñ B is the Koszul
2511 complex (Definition 6.5.6) of a sequence ~a in H ˚ pBq of images of generators of H ˚ pBGq
“ under‰
2512
˚ ˚ ˚ ˚
the characteristic map χ : H pBGq ÝÑ H pBq. This follows because indeed H pBGq “ S QpBGq
„
2513 by Borel’s Theorem 6.6.1 and ΛPG b SQpBGq, equipped with τ : PG ÝÑ QpBGq, is the Koszul
2514 complex of PG. In particular, one has the following isomorphism.

2515 Proposition 7.1.7. Let G Ñ E Ñ B be a principal bundle and C its Cartan algebra. Then there is an
2516 isomorphism ` ˘
H ˚ pCq – Tor‚,‚
H˚
Q, H ˚ pBq .
G

2517 Proof. By Remark 7.1.6, C is the Koszul complex of the map ` χ˚ : H ˚˘pBGq ÝÑ H ˚ pBq, and by
2518 Proposition 6.5.10, the cohomology of this complex is Tor‚,‚
H˚
Q, H ˚ pBq .
G
` ˚ pGq, d
˘
2519 We` would like to find ˘ a zig-zag of quasi-isomorphisms linking A PL pBq b H C with
2520 C “ H ˚`pBq b H ˚ pGq, d ˘.1 Recall from Definition A.4.6 that in this case the space B and the
2521 complex APL pBq, d APL pBq are both called formal.

2522 Proposition 7.1.8. If B of a principal bundle G Ñ E Ñ B is formal, then the Cartan algebra of Defini-
2523 tion 7.1.5 computes the cohomology H ˚ E of the total space.
` ˘ ` ˘
1 There exists a single quasi-isomorphism H ˚ pBGq, 0 ÝÑ APL pBGq, d APL pBGq , but for general B, a chain of
quasi-isomorphisms is required.
 Chapter 7. The cohomology of homogeneous spaces 99

2524 Proof. This is an application of the algebraic

` ˚lemma
˘ A.4.7 to the zig-zag of quasi-isomorphisms
2525
˚
connecting pA, dq “ APL pBq and H pAq “ H pBq, 0 . In the lemma, let V “ PpGq and ξ : PpGq ÝÑ
2526 APL pBq a lifting of
τ
PpGq ÝÑ QpBGq ãÝÝÑ H ˚ pBGq ÝÑ H ˚ pBq.

2527 The ring endomorphism ψ in the above proof can actually be seen to be an automorphism by
2528 a filtration argument; if one filters by B-degree, then ψ induces the identity map on the associated
2529 graded algebras.

2530 Corollary 7.1.9. Let B be a generalized symmetric space in the sense of Definition A.4.9 and G Ñ E Ñ B
2531 a principal G-bundle over B. Then the Cartan algebra calculates H ˚ pEq.

2532 Proof. By Example A.4.10, a generalized symmetric space is formal, so Proposition 7.1.8 applies.
2533

2534 Corollary 7.1.10 (Koszul, [Kos51]). Let pG, Kq be a pair such that G{K is a symmetric space. Then the
2535 Cartan algebra of K Ñ G Ñ G{K calculates H ˚ pGq.

2536 The case of critical interest to us, of course, is the Borel fibration G ÝÑ GK ÝÑ BK.

2537 Definition 7.1.11. The Cartan algebra of the Borel fibration G ÝÑ GK ÝÑ BK, given by C “
2538 H ˚ pBKq b H ˚ pGq equipped with antiderivation d extending ρ˚ ˝ τ : PpGq ÝÑ QpBGq ÝÑ H ˚ pBKq,
2539 is the Cartan algebra of the pair pG, Kq.

2540 The key theorem, due to Cartan, is that the Cartan algebra of a compact pair pG, Kq does
2541 indeed compute H ˚ pG{Kq.

2542 Theorem 7.1.12 (Cartan, `[Car51, Thm. 5, p.˘ 216][Bor53, Thm. 25.2]). Given a compact pair pG, Kq,
„
2543 there is an isomorphism H ˚ H ˚ pBKq b H ˚ pGq ÝÑ H ˚ pG{Kq making the following diagram commute:
` ˘
H ˚ H ˚ pBKq b H ˚ pGq
=

H ˚ pBKq H ˚ pGq (7.5)

>

χ˚ !  j˚
H ˚ pG{Kq
“ ‰
2544 Proof. Because H ˚ pBKq – S QpBGq is a free CGA, it is formal and Proposition 7.1.8 applies.

2545 Remark 7.1.13. One might object that the inclusion of the hypothesis that the base B be formal is
2546 heavy-handed. Unfortunately, not all manifolds are formal, and if B is not, the Cartan algebra of a
2547 bundle can fail to compute the cohomology of the total space. For an example of this phenomenon,
2548 see Section 3 of Baum and Smith [BS67, p. 178].

2549 Corollary 7.1.14. There is an isomorphism

` ˘
H ˚ pG{Kq – Tor‚,‚ ˚
H ˚ pBGq Q, H pBKq .
` ˘
2550 Proof. By Theorem 7.1.12 and Proposition 7.1.7, H ˚ pG{Kq – H ˚ pCq – Tor‚,‚ ˚
H ˚ pBGq Q, H pBq .
 Chapter 7. The cohomology of homogeneous spaces 100

2551 Remark 7.1.15. If we set K “ G, this statement makes precise our motivating claim in the introduc-
2552 tion to Section 6.5 that the differentials in the SSS of the universal bundle G Ñ EG Ñ BG filter an
2553 antiderivation τ extending the transgression which can be seen as the “one true differential.” In
2554 the same way, the SSS of the Borel fibration G Ñ GK Ñ BK filters the differential on the Cartan
2555 algebra. This does not make this SSS, which we have already exploited to such effect, any less
2556 valuable: we will see examples in the next section where the Cartan algebra is unpleasantly com-
2557 plicated and it behooves us to look at the associated graded algebra E8 “ gr H ˚ pG{Kq instead.
2558 Moreover, in precisely the complement of this “bad” case, the associated graded construction is
2559 an isomorphism, so that the SSS of the Borel fibration calculates H ˚ pG{Kq on the algebra level.
2560 Rather than one description being more powerful, it is the equivalence of these two descriptions
2561 that turns out to be critical.
2562 Remark 7.1.16. It is only fair to say at one point why we insist so fervently that K be connected.
2563 The main issue is that if K is not connected, then BK will not be simply-connected, and the
2564 Serre spectral sequence of the Borel fibration is calculated with local coefficients. One can still
2565 say some things, for if K0 ă K is the identity component, then BK0 ÝÑ BK and G{K0 ÝÑ G{K
2566 are finite coverings, so if |π0 K| is invertible in k, one can embed H ˚ pG{Kq as the π0 K-invariants
2567 of H ˚ pG{K0 q by Proposition B.3.1 and likewise HK˚ as the π0 K-invariants of HK˚0 .
2568 That G be connected, on the other hand, is not a real restriction if we insist K be connected,
2569 for then K will lie in the identity component G0 of G and G{K will factor homeomorphically as
2570 π0 G ˆ G0 {K, a finite disjoint union of copies of G0 {K.
2571 Historical remarks 7.1.17. The original, unpublished statement of Chevalley’s theorem, as best the
2572 author can tell, applied to the de Rham cohomology of a smooth principal G-bundle with com-
2573 pact total space. This statement is cited by Cartan and Koszul both (without proof) in the Colloque
2574 proceedings. Borel’s generalization of this result, as proved in his thesis, removes the smoothness
2575 hypotheses by relying, instead of on forms, on an object of Leray’s creation known as a couverture,
2576 which was superseded so quickly and so thoroughly by the ring of global sections of a fine sheaf
2577 of R-CDGAs that it never acquired an English translation. Borel’s statement of the result still
2578 requires compactness of the base because it relies on (what is essentially) sheaf cohomology with
2579 compact supports and a result of Cartan which in modern terms can be interpreted as saying a
2580 resolution of the constant sheaf R on a paracompact Hausdorff space by a fine sheaf of R-CDGAs
2581 always exists. Neither the principal bundle G Ñ EG Ñ BG nor a Q-CDGA model of cohomology
2582 was available to Borel at the time, so in his statement [Bor53, Thm 24.1] of Chevalley’s theorem,
2583 our H ˚ pBq is replaced with (essentially, again) a fine resolution B of the real constant sheaf on B.
2584 As we have noted in Historical remarks 6.6.16, Borel did not actually have BK available, so his
2585 proof was slightly complicated by the need to invoke n-universal K-bundles Epn, Kq ÝÑ Bpn, Kq
2586 for n sufficiently large. Borel’s proof also incorporated not the Serre spectral sequence as we
2587 did, but the Leray spectral sequence, applied simultaneously to an early formulation of a sheaf
2588 and a couverture. We will reproduce a less drastic modernization of Borel’s original argument in
2589 Appendix D, and delve slightly further there into the meaning of the Leray spectral sequence,
2590 fine sheaves, and couvertures.

2591 The structure of the Cartan algebra, I

2592 The Cartan algebra makes a few results on H ˚ pG{Kq easy which would require more sophisti-
2593 cation if attacked with the map of spectral sequences that was the subject of Section 7.1.2. We
 Chapter 7. The cohomology of homogeneous spaces 101

2594 reproduce here the important bundle diagram (7.4) whose spectral sequence the Cartan algebra
2595 encodes.
G G
j
 
GK / EG

χ
 
BK / BG.
ρ

2596 One important subobject of the Cartan algebra is related to the image of the map j˚ : H ˚ pG{Kq ÝÑ
2597 H ˚ pGq induced by j : G ÝÝ G{K » GK .
2598 Definition 7.2.1. The image of j˚ : H ˚ pG{Kq ÝÑ H ˚ pGq is called the Samelson subring of H ˚ pGq.
2599 It meets the primitives PG ď H ˚ pGq in the Samelson subspace P. p

2600 The importance of the Samelson subspace is that in fact it generates im j˚ .

2601 Proposition 7.2.2. The Samelson subring is the exterior algebra Λ P.
p

2602 Proof. At the end of the spectral sequence pEr , dr q of the Borel fibration G Ñ GK Ñ BK, the
‚,‚ 0,‚
2603 projection E8 ÝÝ E8 to the first column is the “associated graded” map gr j˚ : gr H ˚ pG{Kq ÝÑ
0,‚ p it will be enough to show each Er0,‚
2604 H pGq (Proposition 3.3.7), so im j˚ – E8
˚ . To show this is Λ P,
0,‚
2605 is an exterior algebra on a subspace of PG. Note that E2 of the first page is H ˚ pGq “ ΛPG and
2606 assume inductively that Er “ ΛPr for some Pr ď PG.
2607 In the spectral sequence pE rr , d˜r q of the universal bundle, the subspace Pr´1 H ˚ pGq ă H ˚ pGq
2608 survives to Err , after which it disappears because of the edge isomorphism
„ rr,0
d˜r : Pr´1 pGq ÝÑ Er .

2609 From the cochain equations

ψr ˝ d˜r “ dr ˝ ψr (7.6)
2610 which arise from the bundle map (7.4) to the universal bundle, it follows inductively that Pr´1 H ˚ pGq
2611 cannot vanish before Er`1 , because for each q ă r one has
dq z “ dq ψq z “ ψq d˜q z “ ψq 0 “ 0.

2612 On the other hand, the differential dr annihilates Pr X Er0,ďr´2 because it sends it into the fourth
2613 quadrant. If we write V “ Pr X Er0, ďr´1 and V K for the subspace of Pr spanned by elements of
2614 degree other than r ´ 1, then since ΛPr – ΛV K b ΛV, we need only show that kerpdr æ ΛVq “
2615 ΛpV X ker dr q. But that follows from Proposition 6.5.5.
` ˘
2616 r K is the augmentation ideal of H ˚ , then one has Pp “ d´1 H
Proposition 7.2.3. If H r K ¨ im d .
K
0,‚
2617 Proof. Let z P Pr´1 G, and suppose dr z “ 0 in Er , so that z survives to E8 . Then it must be that
2618 the lift dz P E2 lies in the kernel of E2‚,0 ÝÝ Er‚,0 , which is the ideal generated by the lifts to
2619 E2 of the images of previous edge maps di : Ei0,i´1 ÝÑ Eii,0 . Since these edge differentials lift to
2620 those components of d which advance the horizontal filtration by fewer than r steps, it follows
2621 dz P H r K ¨ im d.
2622 On the other hand, if dz P H r K ¨ im d and dr z “ 0 in im r, then the im d components can
2623 only arise from earlier differentials, as later differentials dr`n send z past dz in the horizontal
2624 filtration.
 Chapter 7. The cohomology of homogeneous spaces 102

2625 The Samelson subring is in fact a tensor factor of H ˚ pG{Kq.

2626 Definition 7.2.4. Let pG, Kq be a compact pair. We write Pq :“ PG{ P,

p and call this the Samelson
2627 complement; the notation is supposed to indicate its complementarity to P.
p

2628 Proposition 7.2.5. The Cartan algebra admits a coproduct decomposition

pHK˚ b ΛPG, dq – HK˚ b Λ P,

`
q dq b pΛ P,
p 0q.

2629 The proof is just what one would naively hope; we paraphrase from Greub et al [GHV76,
2630 3.15 Thm. V, p. 116].

2631 Proof. Choose some Q-linear section

Pp ÝÑ ker d ď HK˚ b ΛPG

j˚
of the column projection ker d  H ˚ pG{Kq Ñ H ˚ pGq. This section extends uniquely to a ring
injection f : Λ Pp ÝÑ ker d which we can extend further to a ring map

pHK˚ b Λ Pq
q b Λ Pp ÝÑ H ˚ b ΛPG
K
pa b žq b ẑ ÞÝÑ pa b žq ¨ f pẑq.

2632 This ring map is also a cochain map, since it`is the ˘ identity on the first tensor-factor of its domain
2633 and since for ẑ P Λ P we have 0 “ dp f ẑq “ f 0pẑq .
p
2634 It remains to see f is bijective. Note that f is the identity on HK b Λ Pq and that given an
2635 element z P P, p since f is defined to be a section of the projection to the leftmost column, we
` ˘
2636 have f pzq ” 1 b z mod H r ˚ . Thus f preserves the the horizontal filtration induced by the
K
À p
2637 filtration Fp HK˚ “ ˚
iěp HK on the base HK and induces an isomorphism gr‚ f on associated
2638 graded algebras. By Proposition A.5.8, f is an isomorphism.

2639 Corollary 7.2.6. Let pG, Kq be a compact pair. Then there exists a tensor decomposition

H ˚ pG{Kq – H ˚ pHK˚ b Λ P,
q dq b Λ P,
p

2640 where the subring Λ Pp “ im j˚ ď H ˚ pGq is induced from the projection j : G ÝÑ G{K.

2641 We write the first factor as J.

2642 Corollary 7.2.7. The factor J satisfies Poincaré duality.

2643 Proof. Since G{K is a compact manifold, H ˚ pG{Kq is a PDA by Theorem A.3.12, and the exterior
2644 algebra Λ Pp is a PDA, so by Proposition A.3.14, so also must be the remaining factor J.

2645 The same way that im j˚ admits a description as the leftmost column of E8 for the SSS of
‚,0
2646 G Ñ GK Ñ BK, so also the image of χ˚ admits a description as the bottom row E8 .

2647 Definition 7.2.8. The map χ˚ : HK˚ ÝÑ H ˚ pG{Kq is traditionally called the characteristic map and
im χ˚ – HK˚ {{ HG˚ the characteristic subring of the pair pG, Kq. The factor J “ H ˚ HK˚ b Λ P,
` ˘
2648 q d of
2649 H ˚ pG{Kq in the decomposition Corollary 7.2.6 is called the characteristic factor.
 Chapter 7. The cohomology of homogeneous spaces 103

2650 The name characteristic subring arises because, up to homotopy, the classifiying map G{K ÝÑ
2651 BK of the principal K-bundle K Ñ G Ñ G{K is the projection χ : GK ÝÑ BK of the Borel fibration
2652 (see (7.1)), and the characteristic classes of the former K-bundle bundle lie in im χ˚ . The charac-
2653 teristic factor is so called because HK˚ ãÝÝÑ HK˚ b H ˚ pGq factors through HK˚ b Λ P,
q making clear the
2654 following containment.

2655 Proposition 7.2.9. The characteristic ring im χ˚ is contained in the characteristic factor J.

2656 The cohomology sequence (7.2) is coexact at HK˚ , yielding the following pleasing description
2657 of the characteristic subring.

2658 Proposition 7.2.10. The characteristic subring is given by im χ˚ – HK˚ {{ HG˚ .

2659 Proof. The bottom row HK˚ lies in the kernel of the Cartan differential dC , and meets the image
„
2660 im dC in the ideal j generated by ρ˚ pim τq. Since τ : PpGq ÝÑ QpBGq ` surjects
˘ onto generators of
2661 HG˚ , it follows that the ideal j which is the kernel of HK˚ ÝÑ H ˚ HK˚ b H ˚ pGq is generated by the
2662 image ρ˚ H r G of the augmentation ideal, so this image is H ˚ {pρ˚ H
K
r G q “ H ˚ {{ H ˚ , the ring-theoretic
K G
2663 cokernel. By the commutativity of the diagram (7.5), this image subalgebra corresponds to im χ˚
2664 in H ˚ pG{Kq.

2665 This information is already enough to compute H ˚ pG{Kq in many cases of interest.

2666 Cohomology computations, I

2667 Lest we miss the trees for the forest in fleshing out our general description of the Cartan algebra,
2668 we take a detour to describe the cohomology of two popular classes of homogeneous spaces G{K,
2669 namely those for which H ˚ pGq Ñ H ˚ pKq is surjective and those for which rk G “ rk K.

2670 Cohomology-surjective pairs

2671 The map (7.4) of spectral sequences lets us easily reobtain Hans Samelson’s classic theorem
2672 that H ˚ pGq – H ˚ pKq b H ˚ pG{Kq whenever H ˚ pGq ÝÝ H ˚ pKq. Pictorially, this means the Serre
2673 spectral sequence of G Ñ GK Ñ BK looks like that of Up4q Ñ Up4qSpp2q Ñ BSpp2q, as pictured in
2674 Figure 7.3.3; for now, just look at the E8 page, on the right.

2675 Definition 7.3.1. If pG, Kq is a compact pair such that K ãÑ G induces a surjection H ˚ pGq ÝÑ
2676 H ˚ pKq in cohomology, we call pG, Kq a cohomology-surjective pair.

2677 Theorem 7.3.2 (Samelson [Sam41, Satz VI(b), p. 1134]). Suppose that pG, Kq is a cohomology-surjective
2678 pair. Then

2679 1. ρ˚ : HG˚ ÝÑ HK˚ is surjective,

2680 2. χ˚ : HK˚ ÝÑ H ˚ pG{Kq is trivial,

2681 3. the Samelson subspace Pp is complementary to PpKq in PpGq,

2682 4. H ˚ pG{Kq is the exterior algebra Λ Pp – ΛPpGq {{ ΛPpKq, and

2683 5. H ˚ pGq – H ˚ pKq b H ˚ pG{Kq.

Figure 7.3.3: The Serre spectral sequence of Up4q Ñ Up4qSpp2q Ñ BSpp2q; nonzero differentials (shown) send ˆ ÞÑ ˝, whereas ‚s survive to the next page

z1357 ˆ ˆ ˆ ˆ ˆ

15 z357 ˆ ˆ ˆ ˆ ˆ

13 z157 ˝ ˝ ‚ ˝ ˝ 13 z1 z5 z7 ˆ

z57 ˝ ˝ ‚ ˝ ˝ z5 z7 ˆ

11 z137 ˆ ˆ ˆ ˆ ˆ 11

z37 ˆ ˆ ˆ ˆ ˆ

9 z135 ˆ ˆ ˆ ˆ ˆ 9
z35 ˆ ˆ ˆ ˆ ˆ
z17 ˝ ˝ ‚ ˝ ˝ z1 z7 ˆ

7 z7 ˝ ˝ ‚ ˝ ˝ 7 z7 ˆ

z15 ˝ ˝ ‚ ˝ ˝ z1 z5 6 z15
˝

5 z5 ˝ ˝ ‚ ˝ ˝ 5 z5 5 z5
˝

z13 ˆ ˆ ˆ ˆ ˆ

3 z3 ˆ ˆ ˆ ˆ ˆ 3

1 z1 ˝ ˝ ‚ ˝ ˝ 1 z1 ˝ 1 z1

0 1 q1 q21 q2 q31 q1 q2 0 1 q2 0 1

E4 0 4 8 12 ¨¨¨ E8 0 8 ¨¨¨ E8 0
 Chapter 7. The cohomology of homogeneous spaces 105

řn ř`
2684 6. If the Poincaré polynomials of PG and PK are respectively ppPGq “ j“1 td j and ppPKq “ j“1 td j ,
then ppG{Kq “ nj“``1 p1 ` td j q.
ś
2685

2686 Proof. By Proposition 6.2.9 the fact i : K ãÝÝÑ G is a group homomorphism implies i˚ : H ˚ pGq ÝÝ
2687 H ˚ pKq takes the primitives PpGq ÝÑ PpKq. Because we have assumed ˚
“ i surjective, it follows
i PpGq “ PpKq and because i is a ring homomorphism that ker i – Λ PpGq{PpKq .
˚ ˚ ˚
‰
2688

2689 The outer columns of (7.3) are a bundle map between the universal principal K- and G-
2690 bundles, inducing a map of Serre sequences interleaving the transgressions. Restricting to prim-
2691 itives, one has the commutative diagram

i˚
PpKq o o PpGq

„
τK τG (7.7)
 
QpBKq o QpBGq,
Qρ˚

2692 which implies that Qpρ˚ qQpBGq “ QpBKq and hence that ρ˚ : H ˚ pBGq ÝÑ H ˚ pBKq ` is also sur-
2693 jective. It˘ follows from the triviality of χ˚ ˝ ρ˚ that the characteristic subring im χ˚ : HK˚ ÝÑ
2694 H ˚ pG{Kq is Q.
2695 If we embed PpKq  PpGq by taking a section of i˚ , we see from the transgression square (7.7)
2696 that the complement of PpKq is annihilated by ρ˚ ˝ τG , so that the Samelson subspace Pp ď PpGq
2697 is a complement to PpKq, or Pp – PpGq{PpKq.
2698 Because ρ˚ ˝ τ ends PpKq onto QpBKq and annihilates P, p we have a ring factorization of
2699 E2 – H ˚ pBKq b H ˚ pGq as
H pBKq b H ˚ pKq b Λ P,
“ ˚ ‰
p

2700 which respects the transgression in that all differentials are trivial on P,
p and the left tensor factor
2701 is the beginning of the filtration spectral sequence corresponding to the Koszul complex on
0,‚
2702 QpBKq (cf. Proposition 6.6.2). It follows E8 “ E8 – Λ P.
p Thus we can identify the short coexact
2703
˚
sequence H pKq Ð ˚
H pGq Ð ˚
H pG{Kq with
˚i ˚j

0 Ð ΛPpKq ÐÝ Λ PpKq ‘ Pp ÐÝ Λ Pp Ð 0;
“ ‰

2704 the tensor factorization is valid simply because by Proposition A.4.3 the free CGA ΛPpKq is
2705 projective.
2706 The result on Poincaré polynomials follows from the statements in Appendix A.3.2, since
ppΛPGq “ nj“1 p1 ` tdi q and ppΛPKq “ `j“1 p1 ` tdi q.
ś ś
2707

2708 Remarks 7.3.4. (a) With the benefit of hindsight, our calculations of the cohomology rings of SUpnq
2709 in Proposition 6.1.4 and of Vj pCn q and Vj pHn q in Proposition 6.1.5 can all be seen to be of this
2710 form.
2711 (b) The Samelson isomorphism H ˚ pGq – H ˚ pG{Kq b H ˚ pKq also follows directly from Corol-
2712 lary 6.2.5 independent of any consideration of classifying spaces.

2713 Proposition 7.3.5 ([Car51, 1˝ , p. 69][Bor53, Corollaire, p. 179]). Let i : K ãÝÝÑ G be an inclusion of
2714 compact, connected Lie groups. Then ρ˚ : HG˚ ÝÑ HK˚ is surjective if and only if i˚ : H ˚ pGq ÝÑ H ˚ pKq
2715 is.
 Chapter 7. The cohomology of homogeneous spaces 106

2716 Proof. This follows immediately from the commutative square (7.7) in the proof of Theorem 7.3.2
2717 since the vertical maps are isomorphisms.

2718 Most of these conditions are clearly equivalent. In fact, a weaker dimension condition on
2719 H ˚ pG{Kq is equivalent to cohomology-surjectivity.

2720 Proposition 7.3.6 ([GHV76, Thm. 10.19.X,(6) p. 466]). Let pG, Kq be a compact pair. One has

h‚ pGq ď h‚ pG{Kq ¨ h‚ pKq,

2721 with equality if and only if pG, Kq is cohomology-surjective.

2722 Proof [GHV76, Cor. to Thm. 3.18.V, p. 125]. This follows from Corollary 3.3.10 as applied to the
2723 Serre spectral sequence of K Ñ G Ñ G{K, evaluating the Poincaré polynomials at t “ 1.
` ˘ ` ˘
˚ BUp4q ÝÑ H ˚ BSpp2q is surjective. From
Example 7.3.7. Recall from Example 6.8.1 ` that
˘ H ` ˘
Proposition 7.3.5, we see as well that H ˚ Up4q ÝÑ H ˚ Spp2q , as promised. We had

c1 ÞÝÑ 0,
c2 ÞÝÑ ´q1 ,
c3 ÞÝÑ 0,
c4 ÞÝÑ q2 ,

2724 so in the primitive subspace PUp4q “ Qtz1 , z3 , z5 , z7 u we have PSpp2q “ Qz3 ‘ Qz7 and Pp “
2725 Qz1 ‘ Qz5 . It follows from Section 7.3.1 that

H ˚ Up4q{Spp2q – Λrz1 , z5 s, deg z j “ j.

` ˘

2726 The resulting spectral sequence, Figure 7.3.3, appears complicated, but this complexity is only
2727 apparent. Staring closely at the picture, one sees that Λ Pp “ Λrz1 , z5 s is a tensor-factor, to which
2728 nothing ever happens, and the massive simplifications after the 4th and 8th pages just witness
2729 that the Koszul complexes Krz3 s and Krz7 s are other tensor-factors.
2730 Alternately, not bothering with the picture, the transgression in the universal principal Up4q-
2731 bundle takes z1 ÞÝÑ c1 and z5 ÞÝÑ c3 , this means that Λ Pp “ Λrz1 , z5 s splits off in the Cartan
2732 algebra immediately, and Srq1 , q2 s b Λrz3 , z7 s is a Koszul complex, so acyclic.
2733 A little more work shows that HUp2nq˚ ˚
ÝÑ HSppnq is surjective for all n with kernel the odd
2734 Chern classes, and it follows

H ˚ Up2nq{Sppnq – Λrz1 , . . . , z4n´3 s, deg z j “ j.

` ˘

2735 As an example application of Samelson’s theorem, we prove a result which will be of use to
2736 us later in investigating equivariant formality of isotropy actions.

2737 Lemma 7.3.8. Let S be a torus in a compact, connected Lie group G and Z “ ZG pSq its centralizer in Z.
2738 The cohomology of Z decomposes as

H ˚ pZq – H ˚ pSq b H ˚ pZ{Sq.

2739 Consequently, H ˚ pZ{Sq is an exterior algebra on rk G ´ rk S generators and h‚ pZ{Sq “ 2rk G´rk S .
 Chapter 7. The cohomology of homogeneous spaces 107

2740 Proof. By Theorem 7.3.2, it will be enough to show the inclusion S ãÝÝÑ Z surjects in cohomology.
2741 Since S is normal in Z, the quotient Z{S is another Lie group, so π2 pZ{Sq “ 0 by Corollary 6.2.11
2742 and in the long exact homotopy sequence (Theorem B.2.4) of the bundle S Ñ Z Ñ Z{S we find the
2743 fragment 0 “ π2 pZ{Sq Ñ π1 S Ñ π1 Z. Since S and Z are topological groups, their fundamental
2744 groups are abelian by Proposition B.4.2 and hence isomorphic to their first homology groups
2745 by Proposition B.2.5, so H1 pS; Zq ÝÑ H1 pZ; Zq is injective. It follows from Theorem B.2.1 that
2746 H1 pS; Qq ÝÑ H1 pZ; Qq is injective, and, dualizing, that H 1 pZ; Qq ÝÑ H 1 pS; Qq is surjective. Since
2747 H 1 pSq generates H ˚ pSq, it must be that H ˚ pZq ÝÑ H ˚ pSq is surjective as well.
2748 The statement on Betti number follows because Z must have the same rank as G, since S is
2749 contained in some maximal torus of G by Theorem B.4.9.

2750 Historical remarks 7.3.9. Proposition 7.3.5 was first proven by Cartan [Car51, 1˝ , p. 69][Bor53,
2751 Corollaire, p. 179].
2752 A surjection H ˚ pGq ÝÑ H ˚ pKq in cohomology corresponds dually to an injection H˚ pKq ÝÑ
2753 H˚ pGq in homology, and it was this condition Hans Samelson researched in the work in which the
2754 tensor decomposition (5.) above was first proven [Sam41]. It has since been said that K is totally
2755 nonhomologous to zero in G. Samelson said the Isotropiegruppe U nicht homolog in der Gruppe G ist
2756 or U  0, the letter U for Untergruppe (our K), and showed if the fundamental class rKs P H˚ pKq
2757 did not become zero in H˚ pGq, then H˚ pKq ÝÑ H˚ pGq: the fundamental class rKs P H˚ pK; Qq –
2758 ΛpPKq˚ is the product of a set of algebra generators, so if ρ˚ rKs ‰ 0 in H˚ pGq, then ρ˚ is injective.
2759 The “totally” is redundant and sometimes dropped for this reason.
2760 When the cohomology ring rather than the homology ring became the primary actor, later
2761 expositors (e.g. [GHV76]) named the condition, by analogy, totally noncohomologous to zero, even
2762 though that name taken literally would imply the surjection H ˚ pGq ÝÝ H ˚ pKq should be injec-
2763 tive. These conditions have been abbreviated variously TNHZ, TNCZ, and n.c.z. For safety’s sake,
2764 in dealing with this situation we will always simply say a map surjects in cohomology.

2765 Pairs of equal rank

2766 We recast some of the results from Chapter 4 in this framework.

2767 Definition 7.3.10. A compact pG, Kq is an equal-rank pair if rk G “ rk K.

2768 Theorem 7.3.11 (Leray). Let pG, Kq be an equal-rank pair. Then

2769 1. ρ˚ : HG˚ ÝÑ HK˚ is injective,

2770 2. χ˚ : HK˚ ÝÑ H ˚ pG{Kq is surjective,

2771 3. the Samelson subspace Pp is trivial,

2772
‚,0
4. H ˚ pG{Kq – E8 is HK˚ {{ HG˚ – HTWK {{ HTWG .
řn 2g j ´1
2773 5. If the Poincaré polynomials of PpGq and PpKq are respectively ppPGq “ j“1 t j and ppPGq “
řn 2k j ´1
2774
j“1 t j , then the Poincaré polynomial of G{K is

n
ppBKq ź 1 ´ t2k j
ppG{Kq “ “ . (7.8)
ppBGq 1 ´ t2gj
j“1
 Chapter 7. The cohomology of homogeneous spaces 108

2775 Proof. The inclusion K ãÑ G induces an injection of Weyl groups WK  WG . and in turn an
2776 inclusion HTWG ãÑ HTWK ãÑ HT˚ of Weyl invariants. Recalling from Corollary 5.4.7 that HG˚ – HTWG ,
2777 this means HG˚  HK˚  HT˚ and in particular ρ˚ : HG˚ ÝÑ HK˚ is injective.2 Since the transgression
„
2778 τ : PG ÝÑ QpBGq is also injective, the composition ρ˚ ˝ τ : PG ÝÑ HK˚ is as well, so its kernel Pp
2779 is 0. The injectivity of ρ˚ combined with the fact im χ˚ – HK˚ {{ HG˚ means HK˚ – HG˚ b im χ˚ as an
2780 HG˚ -module, so the Cartan algebra H ˚ pBKq b H ˚ pGq factors as
` ˘
pim χ˚ , 0q b HG˚ b H ˚ pGq, d .

2781 Since the second term is a Koszul complex, which has trivial cohomology by Proposition 6.5.4,
2782 we have H ˚ pG{Kq – im χ˚ “ HK˚ {{ HG˚ by the Künneth theorem.
2783 As far as Poincaré polynomials are concerned, note first that we are implicitly assuming the
2784 results of Section 6.2 that H ˚ pGq and H ˚ pKq are exterior algebras ‰ to conclude QpBGq –
“ in order
2785 ΣPG is spanned by generators of degree 2g j , and H ˚ pBGq “ S QpBGq is a polynomial ring on
2786 these generators. so by the results of Appendix A.3.2, we have
ź 1 ź 1
ppBGq “ and ppBKq “ .
j
1 ´ t2gj j
1 ´ t2k j

2787 The HG˚ -module isomorphism HK˚ – HG˚ b H ˚ pG{Kq reduces on the the level of graded vector
2788 spaces to
ppBKq “ ppBGq ¨ ppG{Kq.
Multiplying through by ppBGq´1 “ j p1 ´ t2gj q yields the claimed formula.
ś
2789

2790 Corollary 7.3.12 (Leray, [Bor53, Prop. 29.2, p. 201]). Let G be a compact, connected Lie group and T a
2791 maximal torus. Then the characteristic map χ˚ : H ˚ pBTq ÝÑ H ˚ pG{Tq is surjective, and if the Poincaré
2g ´1
polynomial of PpGq is ppPGq “ nj“1 t j j , then
ř
2792

n
ź 1 ´ t2
ppG{Tq “ .
j“1
1 ´ t2gj

2793 We have also a converse.

2794 Proposition 7.3.13. If H ˚ pG{Kq is concentrated in even degrees, then K and G are of equal rank.

2795 Proof. If H ˚ pG{Kq is concentrated in even degrees, then the Euler characteristic χpG{Kq ą 0.
2796 Thus the result follows from Corollary 4.2.5; if we had rk K ă rk G, then we would also have
2797 χpG{Kq “ 0.
2We have proved this from abstract results about invariants, but these maps arise from the cohomology of the base
spaces in the sequence
GT / GK / EG

  
BT / BK / BG,
of principal G-bundles maps, where the maps of total spaces can be conceived as “further quotient” maps among
quotients of EG ˆ G.
 Chapter 7. The cohomology of homogeneous spaces 109

2798 This result also admits a purely algebraic proof involving commutative algebra and the Samel-
2799 son subspace.

2800 Corollary 7.3.14 (Borel [Bor53, Corollaire, p. 168]). Suppose pG, Kq is a pair of compact, connected
2801 Lie groups such that the characteristic homomorphism χ˚ : HK˚ ÝÑ H ˚ pG{Kq is surjective. Then for every
2802 principal G-bundle G Ñ E Ñ B, the fiber inclusion of the quotient bundle G{K ÝÑ E{K Ñ B is
2803 surjective in cohomology.

2804 Proof. The principal bundle G Ñ E Ñ B is classified by a map B ÝÑ BG, inducing a bundle
2805 map to the universal bundle G Ñ EG Ñ BG. Taking the right quotient of the total spaces of both
2806 bundles by K yields a bundle map
G{K G{K
f χ
 
E{K
h / BK

 
B / BG.

2807 But the existence of this map puts us in the situation of Theorem 3.4.1, so one has H ˚ pE{Kq ÝÑ
2808 H ˚ pG{Kq surjective, and moreover

H ˚ pE{Kq – H ˚ pBq b H˚ HK˚ .

G

2809 Theorem II, sec. 10.7, Theorem VII, sec. 3.17, and the Corollary to Proposition V, sec. 3.18.
` ˘
2810 Example 7.3.15. Consider the pair Upnq, T n . The Weyl group WUpnq is the symmetric group Sn
2811 acting on HT˚ “ Qrt1 , . . . , tn s by permuting the generators t j P H 2 pBTq, so HUpnq ˚ “ Qrc1 , . . . , cn s is
2812 generated by the elementary symmetric polynomials c j “ σj p~tq. It follows that the cohomology of
2813 the complex flag manifold Upnq{T n is

H ˚ Upnq{T n – Qrt1 , . . . , tn s{pc1 , . . . , cn q,

` ˘

2814 with Poincaré polynomial

N
p1 ´ tq n
n
“ 1p1 ` tqp1 ` t ` t2 q ¨ ¨ ¨ p1 ` t ` t2 ` ¨ ¨ ¨ ` tn´1 q,
` ˘
p Upnq{T “ śn j
j“1 p1 ´ tq

2815 which, evaluated at t “ 1, yields rational dimension n! “ |Sn | “ |WUpnq |. We will see this is no
2816 coincidence.
2817 If we take n “ 2, then
N
Up2q
Up2q{T 2 “ 2 1
Up1q ˆ Up1q « Gp1, C q “ CP « S ,
2

2818 so we know what to expect. Indeed, c1 “ t1 ` t2 and c2 “ t1 t2 in HT˚ “ Qrt1 , t2 s, so

N
˚
` 2
˘ ˚ ˚ Q rt1 , t2 s 2
H Up2q{T – HT2 {{ HUp2q “ pt ` t , t t q – Qrt1 s{pt1 q
1 2 1 2

2819 as predicted.
 Chapter 7. The cohomology of homogeneous spaces 110

For a less trivial example, take n “ 3, so that c1 “ t1 ` t2 ` t3 and c2 “ t1 t2 ` pt1 ` t2 qt3 and
c3 “ t1 t2 t3 . Since we are setting each c j ” 0, we can eliminate out the generator t3 ” ´pt1 ` t2 q
and know 0 ” c2 ” t1 t2 ´ pt1 ` t2 q2 “ ´pt21 ` t22 ` t1 t2 q. Simplifying c3 ” 0 yields t1 t22 ` t21 t2 ” 0, so
N
˚
` 3
˘ Q rt1 , t2 s
H Up3q{T – pt2 ` t2 ` t t , t2 t ` t t2 q.
1 2 1 2 1 2 1 2

2820 See Figure 7.3.16.

Figure 7.3.16: A basis for the E8 page of the Serre spectral sequence of Up3q Ñ Up3qT3 Ñ BT 3

0 1 t1 , t2 t21 , t22 t31 ` t32

0 2 4 6
` ˘
2821 Example 7.3.17. Consider the pair Sppnq, Sppkq ˆ Sppn ´ kq , yielding as quotient the quaternionic
2822 Grassmannian Gpk, Hn q. The Weyl group WSppnq is the signed perrmution group t˘1un ¸ Sn : in
2823 the semidirect product, Sn acts by permuting the entries of t˘1un , and WSppnq acts on HT˚ “
2824 Qrt1 , . . . , tn s by permuting and negating the generators t j P H 2 pBTq, so HSppnq
˚ “ Qrq1 , . . . , qn s
2825 is generated by the elementary symmetric polynomials q j “ σj pt21 , . . . , t2n q in the squares t2j P
2826 H 4 pBTq. The Weyl group WSppkqˆSppn´kq “ WSppkq ˆ WSppn´kq permutes the subrings Qrt1 , . . . , tk s
2827 and Qrtk`1 , . . . , tn s separately, so
N
WSppkq WSppn´kq
˚
`
H Gpk, H q – n
˘ Q rt 1 , . . . , t k s b Q rt k`1 , . . . , t n s
pq1 , . . . , qn q

2828 We will calculate explicitly what happens if n “ 5 and k “ 3. For convenience, set u j “ t2j . The
2829
˚
numerator ring HSpp3q ˚
b HSpp2q is the polynomial subring Qrr1 , r2 , r3 , s1 , s2 s of Qru1 , u2 , u3 , u4 , u5 s
2830 generated by the five generators on the left, and the denominator is the ideal generated by the
2831 elements on the right:

r1 “ u1 ` u2 ` u3 , q1 “ r1 ` s1 ,
r2 “ u2 pu1 , u2 , u3 q, q2 “ r1 s1 ` r2 ` s2 ,
r3 “ u1 u2 u3 , q3 “ r3 ` r2 s1 ` r1 s2 ,
s1 “ u4 ` u5 , q3 “ r3 s1 ` r2 s2 ,
s2 “ u4 u5 ; q5 “ r3 s2 .

2832 Imposing the congruences generated by setting each q j ” 0 and crunching relations a few times
2833 yields
N
˚
` 5
H Gp3, H q –
˘ Q rr 1 , r 2 s deg r1 “ 4, deg r2 “ 8.
pr14 ´ r12 r2 ´ r22 , 2r13 r2 ` 3r1 r22 q,

2834 Historical remarks 7.3.18. Leray’s determination of H ˚ pG{Tq dates back to 1946 in the event G is a
2835 compact, connected, classical simple group [Ler46b]. By 1949, he only requires that the universal
2836 compact cover (see Theorem B.4.4) G r of G contain no exceptional factors [Ler49a]. His original
 Chapter 7. The cohomology of homogeneous spaces 111

2837 statement of Theorem 7.3.11 requires no exceptional group to occur as factors of the universal
2838 compact cover G r of G, but allows K to be any closed subgroup, not necessarily connected, of
2839 equal rank. His additional requirement on G is removed by the time of his contribution [Ler51]
2840 to the 1950 Brussels Colloque de Topologie. The formula (7.8) was first conjectured by Guy Hirsch
2841 and is hence traditionally called the Hirsch formula. According to Dieudonné [Die09, p. 448],
2842 Cartan and Koszul obtained this result independently around the same time. The initial proof
2843 that H ˚ pG{Tq is the regular representation of WG also dates to Leray in the Bruxelles conference;
2844 he had earlier shown in [Ler49a] the same result holds if G is finitely covered by a product of
2845 classical groups.

2846 The structure of the Cartan algebra, II: formal pairs

2847 Returning to our discussion of homogeneous spaces, let pG, Kq be a compact pair and consider
2848 the Cartan algebra HK˚ b H ˚ pGq with differential d induced by ρ˚ ˝ τ.
2849 Recall that if the Samelson subspace Pp ď H ˚ pGq is the subspace of the primitives of G where
2850 d vanishes and Pq “˘ PG{ Pp is the Samelson complement, we defined the characteristic factor to be
J :“ H ˚ HK˚ b Λ P,
`
2851 q d and found a tensor decomposition (Corollary 7.2.6)

H ˚ pG{Kq – J b Λ P.
p

2852 One would like in a similar way to be able to tensor-factor out the characteristic subring im χ˚
2853 from J, but this is not generally possible. The best we are able to do in this regard is the following.

2854 Proposition 7.4.1. The characteristic ring im χ˚ is simultaneously a subring and quotient ring of the
2855 characteristic factor J “ H ˚ pHK˚ b Λ Pq.
q

2856 Proof. Since the image of d meets HK˚ in ρ˚ HG˚ , the composite projection

HK˚ b H ˚ pGq ÝÝ HK˚ ÝÝ HK˚ {{ HG˚ “ im χ˚

2857 descends in cohomology to a homomorphism H ˚ pG{Kq ÝÝ im χ˚ split by the defining inclusion
2858 im χ˚ ãÝÝÑ H ˚ pG{Kq.

2859 In this section, we explore the propitious case in which the characteristic subring im χ˚ is the
2860 characteristic factor J.

2861 Definition 7.4.2. If H ˚ pG{Kq – im χ˚ b Λ P,

p we call pG, Kq a formal pair (traditionally, such a pair
2862 is called a Cartan pair).

2863 Example 7.4.3. Suppose pG, Kq is a cohomology-surjective pair. Then, by Theorem 7.3.2, the char-
2864 acteristic factor J is trivial.
2865 Example 7.4.4. Suppose pG, Kq is an equal-rank pair. Then, by Theorem 7.3.11, the Samelson sub-
2866 ring Λ Pp is trivial and the characteristic factor J is the characteristic ring im χ˚ .
2867 One can see formal pairs as the smallest class of cases that contains both these extreme
2868 examples. Another way of seeing it is this: the first interesting page of the Serre spectral sequence
2869 of the Borel fibration G Ñ GK Ñ BK is E2 “ E2‚,0 b E20,‚ – HK˚ b H ˚ pGq, a coproduct of CGAs,
2870 with one tensor-factor each arising from the base and the fiber of the fibration. In our examples
 Chapter 7. The cohomology of homogeneous spaces 112

2871 in Section 7.3.2 and Section 7.3.1, this tensor-product structure persisted throughout the entire
2872 sequence, in that the decomposition Er “ Er‚,0 b Er0,‚ continued to hold, and
0,‚
“ HK˚ {{ HG˚ q b Λ Pp
‚,0
`
E8 “ E8 b E8

2873 was the tensor product of the characteristic subring im χ˚ and the Samelson subring Λ P. p 3 For a
2874 representative example, see Figure 7.5.4. This is also the optimal situation from a purely numer-
2875 ical perspective, because, in particular, the tensor decomposition yields a factorization
‚,0 0,‚
ppG{Kq “ ppE8 q ¨ ppE8 q, (7.9)

2876 of Poincaré polynomials and in particular, setting the formal variable t “ 1, a factorization
0,‚
h‚ pG{Kq “ dimQ E8
‚,0
¨ dimQ E8 .

2877 We will expound a number of properties of and equivalent characterizations of the formal
2878 pair condition, in the process justifying the nomenclature. The very fact that there are so many
2879 ways of stating this property should be a further argument, were one needed, for the naturality
2880 of the concept.
2881 But first we introduce an important bound on the dimension of the Samelson subspace.

2882 Definition 7.4.5. The deficiency of a compact pair pG, Kq is the integer

dfpG, Kq :“ rk G ´ rk K ´ dim P.
p

2883 Proposition 7.4.6. The deficiency is a natural number. That is, for any compact pair pG, Kq, one has

dim PG ´ dim PK ě dim P.

p

2884 Proof [Bau68, Lem. 3.7, p. 26]. Since Pq ‘ Pp “ PG by definition, it is enough to show dim Pq ě
2885 dim PK. This can be shown“through ‰ Poincaré polynomials. We may view HK˚ as an algebra over
2886 q ˚ ˚
the polynomial ring A “ S τp Pq by restricting ρ : HG ÝÑ HK˚ . If we lift a basis of HK˚ {{ HG˚ “
2887 HK˚ {{ A back to HK˚ , this basis spans HK˚ as an A-module (typically with some redundancy; we do
2888 not expect HK˚ to be a free A-module). Thus ppHK˚ {{ HG˚ q ¨ ppAq ě ppHK˚ q (in that each coefficient
2889 of tn on the left is at least its counterpart on the right), or dividing through,

ppHK˚ q
ppHK˚ {{ HG˚ q ě .
ppAq

2890 Both the numerator and denominator on the right-hand side are products of factors 1 ´ tn , by
2891 (A.1). There are dim PK such factors in the numerator and dim Pq in the denominator, so if we
2892 had dim PK ą dim P, q the rational function ppH ˚ q{ppAq would have a pole at t “ 1, but this is
K
2893 impossible because it is majorized by the polynomial ppHK˚ {{ HG˚ q.

2894 Theorem 7.4.7 ([Oni94, Thm. 12.2, p. 211]). Let pG, Kq be a compact pair. The following conditions are
2895 equivalent:

2896 1. pG, Kq is a formal pair.

3We concede that in those examples, it was the tensor product of precisely one of those factors—there are historical
reasons why those cases were studied first.
 Chapter 7. The cohomology of homogeneous spaces 113

χ˚
2897 2. The kernel pim ρr˚ q of the characteristic map HK˚ ÝÑ H ˚ pG{Kq is a regular ideal in the sense of
2898 Definition 6.5.6.
χ˚ j˚
2899 3. The sequence HK˚ ÝÑ H ˚ pG{Kq ÝÑ H ˚ G is coexact.
2900 4. The characteristic factor J in the decomposition H ˚ pG{Kq – J b Λ Pp is evenly-graded.
2901 5. The deficiency dfpG, Kq “ dim PG ´ dim PK ´ dim Pp is zero.
2902 Proof. We always have H ˚ pG{Kq – J b Λ P,
p so the task is to prove the remaining conditions are
2903 equivalent to the statement J “ im χ˚ .
2904 1 ðñ 2. If we singly grade the CDGA C “ H ˚ b Λ P, q by
K

¨ ¨ ¨ ÝÑ HK˚ b Λ P ÝÑ HK˚ b Λ1 Pq ÝÑ HK˚ Ñ 0,

2q
(7.10)
2905 where the differential d vanishes on HK˚ and is induced by
ρ ˚
„
Pq ãÝÝÑ PG ÝÑ QpBGq ÝÑ HK˚
τ

2906 then J “ im χ˚ “ HK˚ {{ HG˚ if and only if H ˚ pCq “ H 0 pCq. But if we write ~x for a basis of
2907 q ď H ˚ , then C is the Koszul complex K ˚ p~x, H ˚ q of Definition 6.5.6. The commutative algebra
τp Pq G HG K
2908 result Proposition 6.5.9 then states this Koszul complex is acyclic if and only if the sequence is
2909 regular.
2910 1 ùñ 3. By the definition Definition 7.2.1 of the Samelson subring, j˚ factors as H ˚ pG{Kq ÝÝ
2911 Λ Pp ãÝÝÑ H ˚ G, so one can replace H ˚ G by Λ Pp in the coexact sequence above. Once we factor out
2912 Λ P,
p the new claim is that the sequence
χ˚
HK˚ ÝÑ J Ñ Q
2913 is coexact, or that every class of positive degree in J has a representative in HK˚ b Λ Pq lying in the
2914 r ˚ q. But this is clearly the case if χ surjects onto J.
ideal χp H K
2915 3 ùñ 1. Assume every class of positive degree in J admits a representative in the ideal p H r ˚ q of
K
2916 H ˚ b Λ P.
K
q Then the quadruple
r ˚ C A :“ H ˚ ,
a :“ H V :“ Q “ J0 ă M :“ J
K K

2917 satisfies M “ aM ` V, so the corollary A.1.4 of Nakayama’s lemma yields J “ M “ AV “ A ¨ 1,

2918 meaning χ˚ is surjective.
2919 1 ùñ 4. This is clear since im χ˚ “ HK˚ {{ HG˚ inherits an even grading from HK˚ .
2920 4 ùñ 2. If J is evenly graded, then H 1 of the Koszul complex C of (7.10) above must be zero
2921 because Pq ď PG is oddly-graded. But by Proposition 6.5.9, this also means J “ H ˚ pCq “ H 0 pCq “
2922 HK˚ {{ HG˚ “ im χ˚ .
2923 2 ðñ 5 [Oni94, p. 144]. Write y1 , . . . , yn for a basis of QpBKq and b1 , . . . , b` for a basis of
2924 τp Pq
q ď Sryi s. Note that we know that dfpG, Kq ě 0 in any event by Proposition 7.4.6, and if
2925 dfpG, Kq “ 0, then dim Pq “ dim PK.
2926 Working over k “ Q or C, the ring kryi s{pb j q is finite-dimensional as a k-module, so the variety
2927 V “ Vpb1 , . . . , b` q Ď kn is zero-dimensional. By a result of algebraic geometry [VA67, Ch. 16], the
2928 sequence pb j q is regular if and only each component of V is pn ´ `q-dimensional. Thus pb j q is
2929 regular if and only if rk K “ n “ ` “ dim P. q
 Chapter 7. The cohomology of homogeneous spaces 114

2930 The justification for our choice of terminology is the following result:

2931 Theorem 7.4.8 ([Oni94, p. 145][GHV76, Thm. [Link]]). A compact pair pG, Kq is formal if and
2932 only if its Cartan algebra is formal in the sense of Definition A.4.6.

Proof [GHV76, Thm. [Link], Thm. [Link], Thm. [Link]]. For the forward direction, one al-
ways has an algebra map

λ : pHK˚ b ΛPG, dq ÝÑ pHK˚ {{ HG˚ q b Λ P,

` ˘
p0 ,
` ˘
a b 1 ÞÝÑ a ` pim
Č ρ˚ q b 1,
` ˘
1 b z ÞÝÑ 1 b z ` p Pq
q ,

2933 which is in fact a DGA homomorphism since dp1 b Pq q is contained in im ρ˚ . If pG, Kq is a formal
2934 pair, so that H pG{Kq – pHK {{ HG q b Λ P, then λ is a quasi-isomorphism, so the Cartan algebra
˚ ˚ ˚ p
2935 pHK˚ b ΛPG, dq is formal.
2936 For the other direction, the strategy is to show the sequence

χ j˚
HK˚ ÝÑ H ˚ pG{Kq ÝÑ H ˚ G

2937 is coexact, this being one of the equivalent formulations in Theorem 7.4.7. Start by noting the
2938 Cartan CDGA C “ pHK˚ b H ˚ G, dq is quasi-isomorphic to APL pG{Kq by Theorem 7.1.12, and that
2939 by the assumption that G{K is formal there also exists a zigzag of quasi-isomorphisms connecting
2940 C with H ˚ pCq – H ˚ pG{Kq as equipped with the zero differential. Proposition 7.1.8 then allows
2941 us to connect a CDGA` structure on H ˚˘pG{Kq b H ˚ K via a zig-zag of quasi-isomorphisms to the
2942 Chevalley algebra APL pG{Kq b H ˚ K, d of the bundle K Ñ G Ñ G{K, which calculates H ˚ G by
2943 Theorem 7.1.3.
2944 Since this zigzag connects the subalgebra APL pG{Kq of APL pG{Kq b H ˚ K with the factor
2945
˚
H pCq – H ˚ pG{Kq ˘of H ˚ pCq b H ˚ K, when we take cohomology, we obtain an isomorphism
` „
2946 H ˚ H ˚ pG{Kq b H ˚ K ÝÑ H ˚ G such that the following triangle commutes:

H ˚ pG{Kq
j˚

u %
„ / H ˚ G.
` ˘
H ˚ H ˚ pG{Kq b H ˚ K

2947 Thus we can identify these two maps, the left being induced by the obvious inclusion H ˚ pG{Kq b
2948 Q ãÑ H ˚ pG{Kq b H ˚ K and the right by the quotient map j : G Ñ G{K.
2949 To show the sequence is coexact, it remains to show the common kernel of these maps is the
2950 r ˚ q in H ˚ pG{Kq. But the differential in the algebra on the bottom left of the
ideal generated by χp H K
2951 triangle is induced by the composition

„ χ˚
PK ÝÑ QpBKq ãÑ HK˚ ÝÑ H ˚ pG{Kq.
` ˘
2952 It follows that the image of H ˚ pG{Kq in H ˚ H ˚ pG{Kq b H ˚ K is the quotient of H ˚ pG{Kq b Q by
2953 the image of the generators of HK˚ , so the kernel is the ideal in H ˚ pG{Kq generated by χ˚ Hr K as
2954 claimed.
 Chapter 7. The cohomology of homogeneous spaces 115

2955 Proposition 7.4.9. Let pG, Kq be a formal pair of Lie groups. If the Poincaré polynomials of the Samelson
2956 subspace P,
p the Samelson complement P, q and the primitive space PK are given respectively by

rk G´rk
ÿ K ÿK
rk ÿK
rk
pp Pq
p “ td j , pp Pq
q “ tc j , ppPKq “ tk j ,
j“1 `“1 `“1

2957 then the Poincaré polynomial of G{K is

rk G´rk
ź K źK
rk
p ¨ `ppBKq ˘ “ 1 ´ tc j `1
ppG{Kq “ ppΛ Pq p1 ` td j q ¨ k j `1
p SrΣ Ps `“1 1 ´ t
q
j“1

2958 and its total Betti number is

rk K śrk K
2rk G ź c` ` 1 pc ` 1q rk G´rk K
‚
h pG{Kq “ rk K ¨ “ `“1 ` 2 .
2 k `1 |WK |
`“1 `

2959 Proof. Given the equations (7.9) and (A.1), all that really remains to be shown is that ppHK˚ {{ HG˚ q “
` ˘
2960 ppBKq{p SrΣ Psq as claimed. But Theorem 7.4.7, the generators of im ρ˚ form a regular sequence
2961 of rk K elements of HK˚ of degrees c j ` 1. These generators are thus algebraically independent and
2962 form a polynomial subalgebra S – SrΣ Ps q of H ˚ such that H ˚ is a free S-module. The result then
K K
2963 follows from Proposition A.3.8.

2964 Proposition 7.4.10 ([Oni94, Rmk., p. 212]). Suppose pG, Kq is a compact pair and S a maximal torus of
2965 K. Then pG, Kq is a formal pair if and only if pG, Sq is.

2966 Proof. This follows from Corollary 5.4.6, with X “ G. Write W for the Weyl group of K. If pG, Sq is
2967 formal, then HS˚ pGq “ H ˚ pG{Sq – pHS˚ {{ HG˚ q b Λ P.
p Since the W-action on H ˚ pGq descends from
2968 the K-action, which is trivial since K is path connected, the action of W on HS˚ pGq affects only the
2969 bottom row HS˚ {{ HG˚ , and we have
´ LL ¯W ´ ¯
˚ ˚ W LL ` ˚ LL ˘
H ˚ pG{Kq “ HK˚ pGq – HS˚ pGqW – HS H ˚ b Λ Pp – pHS q H b Λ Pp – HK H ˚ b Λ P.
˚ p
G G G

2970 On the other hand, if pG, Kq is formal, so that HK˚ pGq – pHK˚ {{ HG˚ q b Λ P,
p then

˚ LL ˚ LL
H ˚ pG{Sq – HS˚ b H ˚ pG{Kq – HS˚ b HK H ˚ b Λ Pp – HS H ˚ b Λ P.
p
˚
HK ˚ G HK G

2971 Proposition 7.4.11. Suppose pG, Kq is a compact pair such that G{K is a symmetric space. Then pG, Kq
2972 is a formal pair.

2973 Proof. We have already stated in Example A.4.10 that a symmetric space G{K is formal, but here
2974 is an actual proof. This venerable argument, essentially due to Élie Cartan, would actually turn
2975 into a proof G{K is geometrically formal with the mere addition of a proof the representing forms
2976 we find are in fact harmonic.
2977 Recall from Proposition 4.1.2 that elements of H ˚ pG{K; Rq are all represented by G-invariant
2978 forms on G{K, which are determined by their values at the identity coset, which are elements
2979 of the exterior algebra Λpg{kq_ , and further from Proposition 4.1.3 that G-invariance on ΩpG{Kq
2980 translates to pAd˚ Kq-invariance in Λpg{kq_ . Thus all elements of H ˚ pG{Kq are represented by
 Chapter 7. The cohomology of homogeneous spaces 116

˘K
elements of Λpg{kq_ . Let θ P Aut G be the involution fixing K, so that g admits a decomposition
`
2981

2982 g “ k ‘ p with k the Lie algebra of K and p the p´1q-eigenspace of θ˚ P Aut g. Then g{k – p, so
2983 H ˚ pG{Kq is represented by elements of Λrp_ sK .
2984 We claim that all these elements correspond to closed differential forms. Indeed, because
2985 θ is a Lie group automorphism, the induced map θ ˚ on ΩpG{Kq commutes with the exterior
2986 derivative d, and so the same holds true of the differentials induced on Λpg{kq_ – Λrp_ s. Now,
2987 since θ˚ acts as ´id on p, its dual θ ˚ acts as ´id on p_ and so acts as p´1q` ¨ id on the summand
2988 Λ` rp_ s spanned by wedge products of ` elements of p_ . Let ω be one such element. Then, since
2989 d ˝ θ ˚ “ θ ˚ ˝ d, we have
p´1q``1 dω “ θ ˚ dω “ dθ ˚ ω “ p´1q` dω,
2990 so dω “ 0. Thus all elements of Λrp_ sK ΩpG{Kq are closed, so G{K is a formal space.

2991 Remarks 7.4.12. Though the formality condition on pairs pG, Kq is convenient, is natural, has
2992 many equivalent formulations, is guaranteed by several commonly studied sufficient conditions,
2993 and is invariant under the act of replacing the isotropy group K with its maximal torus S, there
2994 still seems to be no simpler way of determining formality of a randomly given pair pG, Kq than
2995 carefully examining the image of the map ρ˚ : HG˚ ÝÑ HS˚ , and our knowledge has arguably not
2996 improved in any major way since regular sequences were introduced into commutative algebra
2997 in the mid-1950s. Indeed, it seems computing the map ρ˚ is an NP-hard problem [Ama13, Sec. 1].
2998 The deficiency first appears in Paul Baum’s 1962 doctoral dissertation [Bau62], where it is
2999 shown inter alia that if k “ Z or k is any field and H ˚ pG; kq and H ˚ pK; kq are exterior algebras and
3000 the analogue of the deficiency with k coefficients satisfies dfpG, Kq ď 2, then the Eilenberg–Moore
3001 spectral sequence of G{K Ñ BK Ñ BG collapses at E2 “ Tor H˚ pk, HG˚ q. The deficiency thus links
K
3002 our account with the Eilenberg–Moore spectral sequence analysis of the cohomology of homoge-
3003 neous spaces. This deficiency is actually an invariant of the homogeneous space G{K and not just
3004 of the compact pair pG, Kq according to a theorem of Arkadi Onishchik; see Onishchik [Oni72].
3005 Historical remarks 7.4.13. What we call a formal pair is traditionally called a Cartan pair (as seen,
3006 e.g., in the standard reference by Greub et al. [GHV76, p. 431]). The condition already arises in
3007 Cartan’s classic transgression paper in the Colloque [Car51, (3) on p. 70], so the attribution is just,
3008 but the name is made inconvenient by the vast prolificacy of the Cartans: pursuant to the work
3009 of Cartan père on symmetric spaces, the pair pk, pq of ˘1-eigenspaces of the Lie algebra g induced
3010 by an involutive Lie group automorphism θ : G Ñ G is also called a Cartan decomposition or a
3011 Cartan pair. (This writer spent an embarrassingly long time in finally convincing himself these
3012 two concepts of “Cartan pair” were entirely unrelated.)
3013 The formal pair condition also appears in the (Russian-language) writings of Doan Kuin’,
3014 where—at least as the translator would have it—K is said to be in the normal condition in G.
3015 This locution did not catch on. We hope that despite the existence of standard terminology, the
3016 semantic overload placed on the word regular, and the possible confusion of formality per se with
3017 equivariant formality, this section has made the case that these terms are natural, justified, and
3018 preferable.
3019 The proof of Theorem 7.4.8 is due to Steve Halperin, and in fact is the first result he proved
3020 as a graduate student. The first published proof was in Greub et al.
 Chapter 7. The cohomology of homogeneous spaces 117

3021 The cohomology of G{S1

3022 In order to obtain Theorem 10.1.8, arguably the main result of this thesis, we needed a grasp on
3023 the cohomology rings H ˚ pG{S; Qq of homogeneous spaces G{S for G compact connected and S a
3024 circle. We found the following dichotomy; these are the only two options because dimQ H 1 pSq “
3025 1.

3026 Proposition 7.5.1. Let G be a compact, connected Lie group and S a circle subgroup. Then the rational
3027 cohomology ring H ˚ pG{Sq has one of the following forms.

3028 1. If H 1 pGq ÝÑ H 1 pSq is surjective, then there is z1 P H 1 pGq such that

˚
H ˚ pG{Sq – H pGq pz1 q.
L

3029 In terms of total Betti number, h‚ pGq “ 21 h‚ pG{Sq.

3030 2. If H 1 pGq ÝÑ H 1 pSq is zero, there are z3 P H 3 pGq and s P H 2 pG{Sq such that

H ˚ pGq Qrss
H ˚ pG{Sq – b 2 .
pz3 q ps q

3031 In terms of total Betti number, h‚ pGq “ h‚ pG{Sq.

3032 This turns out to be a trivial generalization of long-known results. General statements on
3033 the cohomology of a homogeneous space were already available to Jean Leray in 1946, the year
3034 after his release from prison [Mil00, sec. 3, item (4)]. In the second of his four Comptes Rendus
3035 announcements from that year [Ler46a, bottom of p. 1421], he states the following result.4

3036 Theorem 7.5.2 (Leray, 1946). Let G be a compact, simply-connected, Lie group and S a closed, one-
3037 parameter subgroup [viz. a circle]. Then there exist an n P N, a primitive element z2n`1 P H 2n`1 pGq, and
3038 a nonzero s P H 2 pG{Sq such that

H ˚ pGq Qrss
H ˚ pG{Sq – b n`1
pz2n`1 q ps q

3039 The following year, Jean-Louis Koszul published a note [Kos47b, p. 478, display] in the
3040 Comptes Rendus regarding Poincaré polynomials for these spaces.

3041 Theorem 7.5.3 (Koszul, 1947). Let G be a semisimple Lie group and S a circular subgroup. Then the
3042 Poincaré polynomials (in the indeterminate t) of G{S and G are related by

1 ` t2
ppG{Sq “ ppGq .
1 ` t3
3043 This result implies that in fact n “ 1 in Leray’s theorem. This enhanced version of Leray’s
3044 result follows from Proposition 7.5.1 simply because H 1 pGq – HG2 “ 0 for semisimple groups.
3045 We will rely on the part of the proposition on total Betti numbers later to prove Theorem 10.1.8.
3046 The author is unaware of any published proof of the Leray and Koszul results, which is part of
3047 the motivation for including a proof of Proposition 7.5.1 here.
4 See also Borel [Bor98, par. 12]; only owing to Borel’s summary are we confident “compact Lie group” is the
contextually-correct interpretation of Leray’s groupe bicompact.
 Chapter 7. The cohomology of homogeneous spaces 118

3048 Before doing so, we illustrate the result with a representative example. Let S be a circle
3049 contained in the first factor Spp1q of the product group G “ Spp1q ˆ Up2q. The cohomology of G
3050 is the exterior algebra

H ˚ pGq “ Λrq3 , z1 , z3 s, deg z1 “ 1, deg z3 “ deg q3 “ 3,

3051 and that of BS is

HS˚ “ Qrss, deg s “ 2.
3052 The spectral sequence pEr , dr q associated to G Ñ GS Ñ BS is as follows. Its E2 page is the
3053 tensor product HS˚ b H ˚ pGq. Because the map H 1 pGq ÝÑ H 1 pSq is zero, the differential d2 is zero,
3054 and d3 is zero for lacunary reasons, so E4 “ E2 . The differential d4 annihilates each of s, z1 , z3
3055 and takes q3 ÞÝÑ s2 .
` ˘
Figure 7.5.4: The Serre spectral sequence of Spp1q ˆ Up2q Ñ Spp1q ˆ Up2q S Ñ BS

z1 z3 q3 sz1 z3 q3 s2 z1 z3 q3 ¨¨¨

6 z3 q3 sz3 q3 s2 z3 q3 ¨¨¨

z1 q3 sz1 q3 z1 z3 sz1 z3
z1 z3 sz1 z3 s2 z1 z3 s2 z1 q3 ¨¨¨
3 z3 sz3
3 z3 q3 sz3 sq3 s2 z3 s2 q3 ¨¨¨

z1 sz1 s2 z1 ¨¨¨ z1 sz1

0 1 s s2 ¨¨¨ 0 1 s

E4 0 2 4 ¨¨¨ E8 0 2

3056 Because d4 is an antiderivation, its kernel is the subalgebra Qrss b Λrz1 , z3 s and its image the
3057 ideal ps2 q in that subalgebra. Elements mapped to a nonzero element by d4 are marked as blue
3058 in the diagram and elements in the image in red; the vector space spanned by these elements
3059 vanishes in E5 . Thus E5 “ ∆rss b Λrz1 , z3 s, where ∆rss “ Qrss{ps2 q – H ˚ S2 . For lacunary reasons,
3060 E5 “ E8 . In fact,
G{S “ Spp1q S ˆ Up2q « S2 ˆ Up2q,
L

3061 so this tensor decomposition was not unexpected.

3062 This example has all the features of the general case; the pair is always formal, and either it
3063 is cohomology-surjective or else d4 is a nontrivial differential taking some z3 ÞÝÑ s2 P HS4 , which
3064 then collapses the sequence at E5 . If H 1 pGq ‰ 0, then the exterior subalgebra of H ˚ pGq generated
3065 by H 1 pGq, an isomorphic H ˚ pAq, is in the Samelson subring, and can be split off before running
3066 the spectral sequence; the factoring out of this subalgebra is the algebraic analogue of the product
3067 decomposition Proposition 10.4.1 of G{S.

3068 Lemma 7.5.5. A compact pair pG, S1 q is formal.

 Chapter 7. The cohomology of homogeneous spaces 119

3069 Proof. Consider the map ρ˚ : HG˚ ÝÑ HS˚ in the sequence

ρ˚ χ˚
HG˚ ÝÑ HS˚ ÝÑ H ˚ pG{Sq.

3070 Because ρ˚ is a homomorphism of graded rings and HS˚ – Qrss is a polynomial ring in one
3071 r S q of χ˚ is generated by a single homogeneous element and hence is
variable, the cokernel pρ˚ H
3072
n
a regular ideal ps q for some n. By Theorem 7.4.7, it follows pG, Sq is a formal pair.

3073 Proof of Proposition 7.5.1. If H 1 pGq ÝÝ H 1 pSq, then Samelson’s Corollary 6.2.5 applies and yields
3074 the result, so assume instead this map is zero. By Lemma 7.5.5, pG, Sq is a formal pair, so
˚ LL
H ˚ pG{Sq – HS H ˚ b Λ Pp
G

„
3075 with dim Pp “ rk G ´ rk S “ rk G ´ 1 and dim Pq “ 1. It follows that ρ˚ ˝ τ takes Pq ÝÑ Qs` for
3076 some `, yielding Leray’s theorem. To obtain Koszul’s, it remains to show ` “ 2.
3077 By Proposition B.3.4, we may replace G with its universal compact cover A ˆ K, where A is
3078 a torus and K simply-connected, and S with the identity component of its lift in this cover. If
3079 H 1 pGq ÝÑ H 1 pSq is trivial, then because H ˚ pAq is generated by H 1 pAq, it follows H ˚ pAq ď Λ Pp
3080 splits out of the Cartan algebra, so we may as well assume G “ K is semisimple.
3081 We now return to the map of spectral sequences described in Section 7.1.2. Recall the differ-
3082 entials in the spectral sequence pEr , dr q of the Borel fibration K Ñ KS Ñ BS vanish on HS˚ and are
3083 otherwise completely determined by by the composition

ρ˚ ˝ τ : PK ÝÑ HK˚ ÝÑ HS˚ .

3084 Because K is semisimple, H 1 pKq “ 0, so it follows HK2 “ 0 as well by Borel’s calculation from
3085 Section 6.6 of the spectral sequence of K Ñ EK Ñ BK. The edge homomorphisms d2 and d3 then
3086 must be zero, so
E4 “ E2 “ HS˚ b H ˚ pKq
3087 and the first potentially nontrivial differential is
„
d4 : H 3 pKq Ñ HK4 ÝÑ HS4 .

3088 By Lemma 6.8.3, this is surjective, so dz “ ρ˚ τz “ s2 for some z P P3 pKq. Thus pim
Č ρ˚ q is generated
3089
2
by s as claimed, concluding the proof.

3090 Historical remarks 7.5.6. Our presentation in this chapter of Cartan algebra computation of the
3091 cohomology ring H ˚ pG{K; Qq of a homogeneous space G{K introduced what we believe to be the
3092 least possible algebraic overhead, very little analysis at all, and only hints of rational homotopy
3093 theory. That said, such a presentation is dishonest as an origin story. This work was originally cast
3094 in Lie-algebraic terms, with the transgression we have been so casual about explicitly determined
3095 by a connection and induced from a CDGA called the Weil algebra. The Weil algebra, as an
3096 algebra, is the Koszul complex of Section 6.5, but is outfitted with a different differential which
3097 incorporates the adjoint action of the Lie algebra of G. It does this to emulate the behavior of
3098 connection and curvature forms determined by a connection on a principal bundle, and these in
3099 turn arise due to a desire to understand the cohomology of the total space of a principal bundle in
3100 terms of forms arising from pullback in its base. Thus it is an algebraic model of the cohomology
 Chapter 7. The cohomology of homogeneous spaces 120

3101 of EG Ñ BG and the homotopy quotient that predates the general discovery of these objects. The
3102 story of understanding the cohomology of the base of a bundle through invariant forms starts
3103 with the work of Élie Cartan in the early 1900s and continues through the work of Henri Cartan
3104 and his school (Koszul, Borel, and for a time Leray, with major unpublished contributions by
3105 Chevalley and Weil) in the late 1940s and early 1950s.
3106 The main and classical source for these developments is the conference proceedings [Col-
3107 loque] to the 1950 Colloque de Topologie (espaces fibrés), held in Bruxelles, with contributions by
3108 Beno Eckmann, Heinz Hopf, Guy Hirsch, Koszul, Leray, and Cartan. The second of the two pa-
3109 pers by Cartan in this volume, “La transgression dans un groupe de Lie et dans un espace fibré
3110 principal” [Car51], promulgates in Lie-algebraic terms what we have called the Cartan algebra
3111 and was directly responsible for the institution of the Cartan model of equivariant cohomology, a
3112 full ten years before the Borel model gained currency. The classic sketched proof of the equivariant
3113 de Rham theorem showing the equivalence between these two models of equivariant cohomology
3114 is also contained in this terse paper.
3115 There is no shortage of secondary sources for the work of this school [And62; Ras69; GHV76;
3116 Oni94], especially as it applies to the Cartan model of equivariant cohomology [GS99; GLS96;
3117 GGK02], so the author can only hope his in recasting these results in terms of elementary algebra
3118 over the rationals is of independent interest to some person other than himself.
3119 Chapter 8
3120 Equivariant formality

3121 Equivariant formality of group actions

3122 The main motivation of this document is to discuss the equivariant formality of certain Lie group
3123 actions. It is past time we defined the concept.
3124 Definition 8.1.1 ([GKM98]). The action of a topological group G on a space X is said to be
3125 equivariantly formal if the fiber inclusion X ãÝÝÑ XG in the Borel fibration X Ñ XG Ñ BG
3126 surjects in cohomology.
3127 The computational utility of this condition is that it implies the HG˚ -module structure on
3128 HG˚ pXq is as simple as one could hope [GGK02, Lemma C.24, p. 208]. Looking back, one can also
3129 see it is the condition on Y in Theorem 5.2.1.
3130 Proposition 8.1.2. Let a topological group G act on a topological space X such that H ˚ pX; kq is a free
3131 k-module of finite type. The following conditions are equivalent:
3132 1. The action of G on X is equivariantly formal.

3133 2. The SSS of the Borel fibration X Ñ XG Ñ BG collapses at E2 .

3134 3. The equivariant cohomology HG˚ pXq is isomorphic to HG˚ b H ˚ pXq as an HG˚ -module.

3135 4. The HG˚ -rank of HG˚ pXq is h‚ pXq.

3136 Proof. We prove a cycle of implications.
3137 1 ùñ 2. If H ˚ pXG q ÝÑ H ˚ pXq is surjective, then the SSS of the Borel fibration collapses at E2 by
3138 Corollary 3.3.9.
3139 2 ùñ 3. If the SSS of the Borel fibration collapses at E2 , then

gr‚ HG˚ pXq “ E8 “ E2 “ HG˚ b H ˚ pXq

3140 by Theorem 3.3.4, and HG˚ pXq – gr‚ HG˚ pXq as an HG˚ -module (regardless of how badly the asso-
3141 ciated graded construction modifies the multiplication as a whole).
3142 3 ùñ 4. This is trivial.
3143 4 ùñ 1. We prove the contrapositive. Suppose that HG˚ pXq ÝÑ H ˚ pXq fails to be surjective.
3144 Considering the SSS pEr , dr q of the Borel fibration, this nonsurjective map can be considered as
3145 the composition
E8 ÝÝ E8 0,‚
ãÝÝÑ E20,‚ ,

121
 Chapter 8. Equivariant formality 122

0,‚
3146 so the containment E8 ă E20,‚ must be strict; there is some z P E20,‚ – k b H ˚ pXq which is not
3147 in E8 . Since this z generates a cyclic HG˚ -module summand HG˚ b k ¨ z of E2 , it follows then that
3148 rk H˚ E8 ă rk H˚ E2 “ h‚ pXq.
G G

3149 In fact, equivariant formality of an action of a compact, connected Lie group depends only
3150 on the restricted action of its maximal torus.

3151 Lemma 8.1.3 ([GGK02, Prop. C.26, p. 207]). If K is a compact, connected Lie group and S a maximal
3152 torus, and K acts on a space X, then the action of K is equivariantly formal if and only if the restricted
3153 action of S is.

3154 Proof. We have an X-bundle map between Borel fibrations,

3155

XS / XK

 
BS / BK,

3156 so if the action of K is equivariantly formal, then by Theorem 3.4.1, the fiber inclusion X ãÝÝÑ XS
3157 is H ˚ -surjective as well.
3158 Now suppose that the action of S is equivariantly formal, and consider the homomorphism
3159 of spectral sequences induced by the X-bundle homomorphism above. The map HK˚ ÝÑ HS˚ is an
3160 inclusion of Weyl group invariants by Lemma 5.4.5, so the map on E2 pages is injective, and the
3161 differential for the E2 page of X Ñ XK Ñ BK is the restriction of that for X Ñ XS Ñ BS. But the
3162 latter differential is zero, by assumption, so the former is as well. By induction on page number,
3163 the sequence for X Ñ XK Ñ BK collapses at E2 as well, so H ˚ pXK q ÝÑ H ˚ pXq is surjective.

3164 Remark 8.1.4. In fact, that S is a maximal torus was inconsequential to the “only if” direction of the
3165 theorem: any restriction of an equivariantly formal action to a subgroup remains equivariantly
3166 formal.

3167 The Borel localization theorem and equivariant formality

3168 In Section 5.3, we computed the equivariant cohomology of a rotation action on S2 by embedding
3169 this ring into the equivariant cohomology of the fixed point set and showed these actions to be
3170 equivariantly formal. The method of our computation turns out to be an instance of a much more
3171 general pattern: for torus actions on compact manifolds, much of the structure of equivariant
3172 cohomology is determined by the fixed point set. The reduction of the previous section should
3173 also count as some evidence that a focus on torus actions might be worthwhile. (In fact, we only
3174 neglected to prove this important theorem in Chapter 5 because we had not yet developed the
3175 theory of maps HT˚ ÝÑ HS˚ which we need here.)

3176 The localization theorem

3177 Let T be a torus and M a compact T-manifold. The fixed point set M T is compact and T-
3178 invariant—and so itself a T-manifold—and the inclusion M T ãÝÝÑ M induces a restriction HT˚ pMq ÝÑ
 Chapter 8. Equivariant formality 123

3179 HT˚ pM T q in equivariant cohomology. This latter ring is the simplest one could hope: it is the sin-
3180 gular cohomology of the homotopy quotient pM T qT , which, since the action T ñ M T is trivial,
3181 is
ET
ET ˆ M T « ˆ M T “ BT ˆ M T .
T T
3182 Then the traditional singular Künneth isomorphism, even with Z coefficients, yields

HT˚ pM T q “ HT˚ b H ˚ pM T q, (8.1)

3183 so HT˚ pM T q is a free HT˚ -module

3184 Remark 8.2.1. This is the “largest” HT˚ pMq could be: the Serre spectral sequence of the Borel
3185 fibration M Ñ MT Ñ BT shows that HT˚ pMq is a subquotient of HT˚ b H ˚ pMq as an HT˚ -module.
3186 Both HT˚ pMq and HT˚ pM T q are HT˚ -algebras, the latter well-understood, so there is a natural
3187 hope one could understand the latter in terms of the former.

3188 Theorem 8.2.2 (Atiyah–Borel [BBF+60, Prop. IV.3.6, p. 54][Hsi75, Thm. III.1, p. 40]). Let T be a
3189 torus and M a compact T-manifold. Denote by M T the fixed point set of the action of T on M. There exists
3190 a nonzero element f P HT˚ such that the HT˚ -algebra homomorphism

HT˚ pM; Zq ÝÑ HT˚ pM T ; Zq

3191 induced in T-equivariant cohomology by restriction becomes an isomorphism upon inversion of f . That is
3192 to say, there exists a nonzero f P HT˚ for which the induced map

HT˚ pM; Zq f ÝÑ HT˚ pM T ; Zq f

3193 of algebras over pHT˚ q f “ HT˚ r f ´1 s is an isomorphism.

3194 Proof. The proof will proceeds induction on a T-invariant open cover, using the equivariant
3195 Mayer–Vietoris sequence. By the equivariant tubular neighborhood Theorem 1.3.1, each orbit
Tx admits a T-invariant open neighborhood Vx in M. Each x P M T is an orbit, so U :“ xPMT Vx
Ť
3196

3197 is a T-invariant open neighborhood of M T . Since M is compact, U and finitely many other Vx , call
3198 them Vx j , suffice to cover M. By the definition of a tubular neighborhood, U equivariantly defor-
3199 mation retracts to M T and each Vx j to Tx j , so one has HT˚ pUq – HT˚ pM T q and HT˚ pVj q – HT˚ pTx j q.
3200 We now have only to apply the equivariant Mayer–Vietoris sequence Proposition 5.1.3 to this
3201 cover tU, Vj u j to recover HT˚ pMq.
3202 We understand the ring HT˚ pM T q exactly as well as we understand H ˚ pM T q, so it remains to
3203 understand the HT˚ pTx j q. Fix a j and write S “ S j “ Stabpx j q ă T. We know Tx j « T{S from
3204 Proposition 1.2.1 and HT˚ pT{Sq – H ˚ pBSq from Proposition 5.1.1. As a subgroup of a torus, S is
3205 constrained to be of the form S0 ˆ F, where the torus S0 is its identity component and F is a finite
3206 abelian subgroup of T. Writing BS “ ET{S, we can view BS0 as a finite covering space of BS:

F ÝÑ BS0 ÝÑ BS.

3207 If m “ m j “ |F|, and we set k “ Zr m1 s, then by Corollary B.3.2, the map H ˚ pBS; kq ÝÑ H ˚ pBS0 ; kq
3208 is an isomorphism. Thus, if we replace Z with Zr m1 s, we can replace S with S0 . For convenience,
1
3209 we invert all the m j in our coefficient ring now, replacing Z with k “ Zr m1 ¨¨¨m n
s. The short exact
3210 sequence
0 Ñ S Ñ T Ñ T{S Ñ 0
 Chapter 8. Equivariant formality 124

3211 of tori splits, by Proposition A.4.2, so that T – S ˆ T{S. By the Künneth Theorem B.2.2, then,
3212 H ˚ pBSq is a tensor factor of H ˚ pBTq, so there is a factor projection H ˚ pBTq ÝÝ H ˚ pBSq inducing
3213 a HT˚ -algebra structure on H ˚ pBSq. The kernel of the projection
` is generated
˘ by the rk T ´ rk S ą 0
3214
2 ˚
elements of HT that generate the polynomial subring H BpT{Sq . Any g P HT{S 2 acts as multi-
3215
˚ ˚
plication by 0 on HS “ H pBSq, so by the discussion in Appendix A.1, if we invert g we get
3216 HS˚ rg´1 s “ 0.
3217 Thus, for each x j , there is an element g j P HT˚ annihilating ˚ ˚
` Ť ˘HT pVj q – HT pTx j q – H pBS j q. We
3218 claim the least common multiple of the g j annihilates HT ˚ Vj . Indeed, suppose inductively for
Ťn´1
3219 W “ j“1 that a nonzero element h P HT annihilates HT pWq. The restriction map HT˚ pWq ÝÑ
˚ ˚

3220 HT˚ pW X Vh q is an HT˚ -algebra homomorphism, so h also annihilates HT˚ pW X Vn q, and lcmph, gn q
3221 annihilates this ring, HT˚ pWq, and HT˚ pVn q. The equivariant Mayer–Vietoris sequence of the pair
3222 pW, Vn q contains exact fragments

HT˚´1 pW X Vn q ÝÑ HT˚ pW Y Vn q ÝÑ HT˚ pWq ‘ HT˚ pVn q,

3223 and by Lemma A.1.1, since lcmph, gn q annihilates the outer terms, it also annihilates the inner
Ť
3224 term. Thus if V “ j Vj , the nonzero element lcmpg j q P HT˚ annihilates HT˚ pVq. Now consider the
3225 equivariant Mayer–Vietoris sequence of the cover tU, Vu of M:

HT˚´1 pU X Vq ÝÑ HT˚ pMq ÝÑ HT˚ pUq ‘ HT˚ pVq ÝÑ HT˚ pU X Vq.

3226 On inverting g “ lcmpg j q, we obtain an exact sequence

0 ÝÑ HT˚ pMqrg´1 s ÝÑ HT˚ pUqrg´1 s ‘ 0 ÝÑ 0

3227 where, as noted above, HT˚ pUq “ HT˚ pM T q. To get this isomorphism, we inverted 1{m1 . . . mn so to
3228 obtain the theorem with coefficients in Z, we take f “ lcmpm j g j q.

3229 Example 8.2.3. In our examples from Section 5.3, the rotation actions ρq : S1 ñ S2 , the fixed point
1
3230 set pS2 qS was the doubleton 0
` 0 ˘S containing the north and south poles. By Corollary 8.2.5, we have
3231
˚ 2 ˚
rk H˚ Hρ1 pS q “ rkZ H S ; Z “ 2.
S1

3232 If we go ahead and all the nonzero elements of HT˚ , we obtain a simpler statement.

3233 Notation 8.2.4. Recall that we write h‚ pXq “ dimQ H ˚ pX; Qq for the total Betti number. If T is a
3234 torus, we write p ˚ – Qpu1 , . . . , u` q for the field of fractions of H ˚ – Zru1 , . . . , u` s and H
H p ˚ pXq for
T T T
3235 the localization p ˚ ˚
HT b HT pXq.
HT˚

3236 Corollary 8.2.5. Let T be a torus and M a compact T-manifold. Then

p ˚ pM T q – H
p ˚ pMq – H
H p ˚ b H ˚ pM T q.
T T T
Q

3237 Proof. The first isomorphism is immediate from Theorem 8.2.2 on further localization, while the
3238 p ˚.
second follows from tensoring (8.1) with H T

3239 Corollary 8.2.6. Let T be a torus and M a compact T-manifold. Then

rk H˚ HT˚ pMq “ h‚ pM T q.
T
 Chapter 8. Equivariant formality 125

3240 Proof. The rank of a finitely generated HT˚ -module N is the same as the dimension over H p ˚ of
T
3241 Hp ˚ b ˚ N. Taking N “ H ˚ pMq, and using Corollary 8.2.5 and the fact pM T qT « BT ˆ M T , we get
T HT T
3242 the chain of equations.

rk H˚ HT˚ pMq “ dim Hp ˚ H p ˚ pM T q “ h‚ pM T q.

p ˚ pMq “ dim p ˚ H
T T H T
T T

3243 Historical remarks 8.2.7. The main result of this section is a result is often called the “Borel lo-
3244 calization theorem,” though Borel himself proved it only for T “ S1 . In Hsiang Wu-Yi’s (項武義
3245 ) standard text [Hsi75, p. 39], it is only called a “localization theorem of Borel–Atiyah–Segal
3246 type,” so the author asked online for an origin story [Bee]. The result was apparently first stated
3247 by Atiyah in unpublished 1965 Warwick lecture notes which this writer does not know how
3248 to obtain. Atiyah (personal communication) confirms that the result in equivariant cohomology
3249 for higher-dimensional tori is probably originally due to him, but warns that history is never
3250 straightforward and the published record is only part of the story. The Atiyah–Segal completion
3251 theory in equivariant K-theory was inspired by this result.
3252 The theorem in cohomology has been substantially generalized since its discovery, both in
3253 replacing T with more general groups (at the cost of replacing M T with a related set—see for
3254 example Pedroza–Tu [PT07])—and in also considering strata of n-dimensional orbits, M T being
3255 the 0-dimensional case–see for instance Goertsches–Töben [GT10b] and Franz–Puppe [FP07].

3256 Equivariant formality of torus actions and fixed points

3257 We can use localization to obtain a simple numerical criterion for equivariant formality of torus
3258 actions. We will depend on the results of Appendix B.3 and Section 8.2 to such an extent that
3259 uniform statements would be impossible in the rest of this document if we did not, as we always
3260 will from here on out, take coefficient ring k “ Q unless explicitly stated otherwise. We denote
3261 our tori from now on by S instead of T unless they are maximal in a larger Lie group under
3262 consideration; in the cases of interest to us, they will not be.

3263 Lemma 8.2.8 ([GGK02, Cor. C.27]). Let S be a torus and M a compact S-manifold. Then the action of S
3264 on M is equivariantly formal if and only if

h‚ pMq “ h‚ pMS q.

3265 Proof. By Proposition 8.1.2, the action of S on M is equivariantly formal if and only if h‚ pMq “
3266 rk H˚ HS˚ pMq, but by Corollary 8.2.6, h‚ pMq “ rk H˚ HS˚ pMq “ h‚ pMS q.
S S

3267 In fact, one inequality in this lemma holds regardless of whether the action is equivariantly
3268 formal.

3269 Lemma 8.2.9 (Borel [BBF+60, IV 5.5, p. 62]). Let S be a torus and M a compact S-manifold. Then

h‚ pMq ě h‚ pMS q. (8.2)

3270 Proof [GGK02, Lemma C.24]. Consider the spectral sequence pEr q of the Borel fibration M Ñ
3271 MS Ñ BS. Since the E2 page is HS˚ b H ˚ pMq, we have rk H˚ E2 “ h‚ pMq. Because E8 is a subquo-
S
3272 tient of E2 , we know
rk H˚ E2 ě rk H˚ E8 ;
S S
 Chapter 8. Equivariant formality 126

3273 rank at most stays the same as the sequence unwinds. Now E8 “ gr‚ HS˚ pMq is isomorphic to
3274 HS˚ pMq as an HS˚ -module, so that

h‚ pMq “ rk H˚ E2 ě rk H˚ E8 “ rk H˚ HS˚ pMq.

S S S

3275 But by Corollary 8.2.5,

rk H˚ HS˚ pMq “ h‚ pMS q.
S

3276 Remark 8.2.10. Though we cite Lemma 8.2.8 to Ginzburg et al. [GGK02], the result as stated there
3277 is over-optimistic, as can be seen from examples of Franz and Puppe [FP08, Sec. 5]. It becomes
3278 true if “torsion-free” in the statement in [GGK02] is everywhere replaced by “free.”

3279 Equivariant formality and Koszul duality

3280 In the 1998 paper [GKM98] in which they introduce the term equivariant formality, Goresky, Kot-
3281 twitz, and MacPherson also state several sufficient and several equivalent conditions. Often these
3282 conditions are stated in much greater generality; for example, they also apply to equivariant sheaf
3283 cohomology and intersection cohomology theories. Among these equivalent conditions, one of
3284 the most striking is the following.

3285 Theorem 8.3.1 ([GKM98, Prop. 9.3]). Let K be a compact, connected Lie group and X an K-space. Then
3286 H ˚ pX; Qq is an algebra over the homology ring H˚ pK; Qq. The action of K on X is equivariantly formal if
3287 and only if the singular cochain algebra C˚ pX; Qq is quasi-isomorphic as an H˚ pKq-DGA to H ˚ pX; Qq and
3288 further H˚ pKq acts trivially on H ˚ pXq in the sense that H˚ pKq b H ˚ pXq ÝÑ H ˚ pXq factors through
3289 H0 pKq b H ˚ pXq.

3290 While we do not know how to make good use of this cochain-level homology action, even
3291 in the concrete case of the isotropy action of S on G{S discussed in the next two chapters, our
3292 discussion would be incomplete without its mention.
3293 Chapter 9
3294 Equivariant formality of isotropy actions

3295 As we first stated in Section 1.2, in the event G is a Lie group and K a closed subgroup, the orbit
3296 space G{K of the right K-action is a homogeneous space. The left action of the isotropy subgoup
3297 K on this space given by k ¨ gK “ pkgqK is the isotropy action. Recall from the beginning of
3298 Chapter 7 that when G is a compact, connected Lie group and K a closed, connected subgroup,
3299 we say pG, Kq is a compact pair of Lie groups.
3300 Definition 9.0.1. A compact pair pG, Kq is said to be isotropy-formal if the isotropy action of K
3301 on G{K is equivariantly formal in the sense of Definition 8.1.1.
3302 To motivate discussion of the isotropy action, note that K is essentially the largest subgroup
3303 of G which can act equivariant formality on G{K.
3304 Proposition 9.0.2. Let pG, Kq be a compact connected pair and H a closed subgroup of G. If the natural
3305 action of H on G{K, obtained from restricting the defining G-action on G{K, is equviariantly formal, then
3306 the maximal torus TH of H is conjugate in G to a subtorus of K.
3307 Proof. By Lemma 8.1.3, if H acts equivariant formality on G{K, then so does TH . By Lemma 8.2.8,
3308 this occurs if and only if G{K and the fixed point set pG{KqTH have equal total Betti number.
3309 In particular, the total Betti number of the fixed point set must be positive. But Corollary 1.4.5
3310 shows that pG{KqTH is nonempty if and only if TH is conjugate in G to a subgroup of K.

3311 The present chapter, with the exception of Section 9.1, will lay the groundwork for the theory
3312 of isotropy-formal actions and summarize its condition as of April 2014. A few epicycles on the
3313 Shiga–Takahashi criterion devised by the author are included. The following chapter will recount
3314 the author’s original contributions.

3315 Reduction of isotropy-formality to a maximal torus

3316 Let pG, Kq be a compact connected pair and S a maximal torus of K. From Lemma 8.1.3, we know
3317 pG, Kq is isotropy-formal if and only if the restricted action of S on G{K is also equivariantly
3318 formal. What may be a surprise—or may seem perfectly natural, but in any event was not known
3319 before last April—is that the other instance of K can also be replaced by S: that is, K acts equiv-
3320 ariant formality on G{K if and only if K acts equivariant formality on G{K, which is to say, the
3321 pair pG, Sq is isotropy-formal.
3322 To prove the result, we will need to simultaneously consider left and right actions and homo-
3323 topy quotients. From our construction of the Milnor EK in Section 2.3, we know that EK admits
3324 as well as the natural right action a natural left action.

127
 Chapter 9. Equivariant formality of isotropy actions 128

3325 Definition 9.1.1. As with a left action, given a right action of a group K on a space X, there is a
3326 diagonal action of K on X ˆ EK given by px, eq ¨ k “ pxk, k´1 eq, and a right homotopy quotient

X ˆ EK
XK :“ .
pxk, eq „ px, keq

3327 We denote the orbit of px, eq under K by rx, esK P XK . We temporarily denote by K X the left
3328 homotopy quotient defined in Section 3.2.3. We still use the left homotopy quotient to define
3329 equivariant cohomology:
HK˚ pXq :“ H ˚ pK Xq.

3330 Also as with a left action, given a right action of K on X, there is a map XK ÝÝ X{K,
3331 functorial in K and in X, which is a weak homotopy equivalence if the action is free, and satisfies
3332 the statements in Lemma 5.4.1.
3333 Remark 9.1.2. Although the notation is inconsistent with that we used previously, under which
3334 XK was a left homotopy quotient, all statements about homotopy quotients are equivalent for the
3335 left and the right actions. ` ˘ ` ˘
3336 The key point of this equivalence is the following. Let CK “ K ˆ r0, 1s { K ˆ t0u be the cone
on K. The Milnor EK can be viewed as that subspace of the countably infinite product 8
ś
3337
ř8 n“1 CK
3338 populated by lists of pairs ptn , k n q such that n“1 tn “ 1 and only finitely many tn ‰ 0. The
3339 group K acts diagonally both on the left and on the right of EK by k ¨ ptn , k n q “ ptn , kk n q and
3340 ptn , k n q ¨ k “ ptn , k n kq respectively. There is a natural self-homeomorphism of EK given by

e “ ptn , k n q ÞÝÑ ptn , k´1 ´1

n q “: e ,

3341 which takes ke ÞÝÑ e´1 k´1 . This homeomorphism allows us to exchange the right actions we have
3342 used heretoforth for left actions as often as we wish, and also descends to a homeomorphism
«
K{EK ÝÑ EK{K “ BK.

3343 It is only for matters related to the next definition that distinguishing between left and right
3344 actions becomes important, and it is for this definition we have temporarily abandoned our
3345 previous notation.

3346 Definition 9.1.3. If X admits both a left S-action and a right K-action, then the right homotopy
3347 quotient XK admits a left S-action and the left homotopy quotient S X admits a right K-action,
3348 and we can form the homotopy biquotient

ES ˆ X ˆ EK
S XK :“ « S pXK q « pS XqK ,
pe1 s, x, ke2 q „ pe1 , sxk, e2 q

3349 whose elements are denoted S re1 , x, e2 sK .

3350 With the new definitions we can prove the result.

3351 Theorem 9.1.4. Let pG, Kq be a compact pair and S a maximal torus of K. The pair pG, Kq is isotropy-
3352 formal if and only if pG, Sq is.
 Chapter 9. Equivariant formality of isotropy actions 129

3353 Proof. By Lemma 8.1.3, it is enough to show that K acts equivariant formality on G{S if and only
3354 if it does on G{K.
3355 For the forward direction, assume K acts equivariant formality on G{S. Since K acts freely by
3356 right multiplication on G and K G, by Observation 5.4.3, the Weyl group W “ NK pSq{S of K acts
3357 on H ˚ pG{Sq and HK˚ pG{Sq, and
H ˚ pG{Sq ÐÝ HK˚ pG{Sq
3358 is W-equivariant. Because, by assumption, this map is surjective, by Lemma 5.4.4, the restriction
3359 H ˚ pG{SqW ÐÝ HK˚ pG{SqW to subrings of W-invariants is also surjective; but by the naturality
3360 statement in Lemma 5.4.5, this surjection is equivalent to the map

H ˚ pG{Kq ÐÝ HK˚ pG{Kq

3361 induced by the fiber inclusion G{K ãÝÝÑ K G{K.

3362 For the reverse direction, assume the fiber inclusion G{K ãÝÝÑ K G{K is H ˚ -surjective. Since
3363 the right K-actions on G and GK are free, by Lemma 5.4.1, we may replace the fiber inclusion
3364 with κ : GK ãÝÝÑ K GK . Write ξ 0 : BS Ñ BK again. Applying the right homotopy-quotient func-
3365 tor p´qSãÑK of Corollary 5.4.6 to the fiber inclusion G ãÝÝÑ GK of the Borel fibration yields the
3366 following map in F-Bun{ξ 0 :

GS 
λ / K GS / / BS rg, e2 sS  / K re1 , g, e2 sS  / Se2
_
_ _

     
 / K GK / / BK, rg, e2 sK  / K re1 , g, e2 sK  / Ke2 ,
GK κ

3367 where the inclusions in the left square are the fiber inclusions in the Borel fibrations in question.1
3368 Applying Theorem 3.4.1 to this diagram, the upper-left map λ˚ , which we wish to show is
3369 surjective, is equivalent to

κ ˚ b id HS˚ : H ˚ pK GK q b HK˚ HS˚ ÝÑ H ˚ pGK q b HK˚ HS˚ .

3370 But by assumption, κ ˚ is surjective, and ´ b H˚ HS˚ is right exact.

K

3371 Remarks 9.1.5. (a) The original proof of the “if” implication in Theorem 9.1.4 relied on the en-
3372 hanced form of Corollary 5.4.6 stated in Theorem B.4.9, which can be used to show the map
3373 λ˚ is equivalent to κ ˚ b id H˚ . We insist upon the naturality statements because without them,
S
3374 Theorem 9.1.4 would no longer follow.
(b) One might wonder if the passage to the notationally cumbersome homotopy biquotient in this
proof is avoidable, but our proof relies crucially on the notion of being a bundle over BS Ñ BK,
and the necessary maps do not seem to exist from G{S Ñ G{K. Indeed, consider the horn

G{S ÐÝ G S ÝÑ BS,
gS ÑÝÞ rg, esS ÞÝÑ Se.

3375 A map G{S ÝÑ BS completing this horn to a triangle would have to take gS ÞÝÑ Se for all
3376 g P G and e P ES, an infeasibly tall order. Morally, since the map GS ÝÑ G{S quotients ES out
1N.B.: The rows are not themselves bundles: the maps from the total spaces to the common base space BK of the
Borel fibrations GS Ñ K GS Ñ BK and GK Ñ K GK Ñ BK are given by K re1 , g, e2 sS ÞÝÑ e1 K and K re1 , g, e2 sK ÞÝÑ e1 K.
 Chapter 9. Equivariant formality of isotropy actions 130

3377 of GS “ G ˆS ES, while GS ÝÑ BS quotients out G, the quotients G{S and BS have “nothing
3378 left in common” to construct a nontrivial map from. One can still use Lemma 5.4.1 to get the
3379 necessary right HK˚ - and HS˚ -algebra structures on H ˚ pG{Sq and HK˚ pG{Sq, and then reason using
3380 Corollary 5.4.6 on right tensor factors, but this subterfuge seems more circuitous and less honest
3381 (and less natural, no pun intended), than simply invoking homotopy biquotients.
3382 (c) That said, we shall not need biquotients again until Section 10.2, so until and after that section,
3383 the reader can consider anything denoted XK to be the left or the right homotopy quotient as she
3384 or he pleases.

3385 Isotropic torus actions and fixed points

3386 A key component of the proofs of our results in Chapter 10, is the manifestation in the case of an
3387 isotropy action of the fixed-point characterization Lemma 8.2.8, which recall:
3388 Lemma 8.2.8 ([GGK02, Cor. C.27]). Let S be a torus and M a compact S-manifold. Then the action of S
3389 on M is equivariantly formal if and only if

h‚ pMq “ h‚ pMS q.

3390 Let pG, Kq be a compact pair with and S a maximal torus. of K To understand isotropy-
3391 formality of pG, Sq, we should understand the fixed point set pG{SqS . Let N “ NG pSq be the
3392 normalizer of S and Z “ ZG pSq its centralizer. Recall from the beginning of Appendix B that π0 N
3393 denotes the component group of N.
3394 Now recall Lemma 7.3.8 from Section 7.3.1.
3395 Lemma 7.3.8. Let S be a torus in a compact, connected Lie group G and Z “ ZG pSq its centralizer in Z.
3396 The cohomology of Z decomposes as

H ˚ pZq – H ˚ pSq b H ˚ pZ{Sq.

3397 Consequently, H ˚ pZ{Sq is an exterior algebra on rk G ´ rk S generators and h‚ pZ{Sq “ 2rk G´rk S .
3398 This result makes available a useful dimension computation.
3399 Corollary 9.2.1 (Goertsches–Noshari, 2014 [GN15, Prop. 3.1]). Let pG, Kq be a compact pair with S a
3400 maximal torus of K. Then
˘ |π0 N| rk G´rk S
h‚ pG{KqS “
`
¨2
|WK |
3401 Proof. We know from Corollary 1.4.8 that N{S has |π0 N|{|WK components, each homeomorphic
3402 to Z{S and from Lemma 7.3.8 that h‚ pZ{Sq “ 2rk G´rk S .

3403 Proposition 9.2.2 (Goertsches–Noshari, 2014 [GN15, Prop. 3.2]). Let pG, Kq be a compact pair with
3404 S a maximal torus of K. The action of S on G{K is equivariantly formal if and only if
|π0 N| rk G´rk S
h‚ pG{Sq ď ¨2 . (9.1)
|WK |
` ˘
3405 Proof. By Lemma 8.2.8, the action is equivariantly formal if and only if h‚ pG{Kq “ h‚ pG{KqS ,
3406 and we always have the ě direction by Lemma 8.2.9, so we only need the reverse inequality.
|π0 N| rk G´rk S
3407 From Corollary 9.2.1, the right-hand side is h‚ pG{Sq “ ¨2 .
|WK |
 Chapter 9. Equivariant formality of isotropy actions 131

3408 Corollary 9.2.3. Let pG, Sq be a compact pair with S a torus. Then pG, Sq is isotropy-formal if and only if

h‚ pG{Sq ď |π0 N| ¨ 2rk G´rk S . (9.2)

3409 Earlier work on equivariant formal isotropy actions

3410 The groundwork in equivariant cohomology and cohomology of homogeneous spaces being laid,
3411 we in this section summarize the state of knowledge on isotropy-formality as of April 2014.

3412 Isotropy-formality of equal-rank and generalized symmetric pairs

3413 It is classical that a generalized flag manifold has equivariantly formal isotropy action [Bri98,
3414 Prop. 1].
3415 Proposition 9.3.1. An equal-rank pair pG, Kq is isotropy-formal.
3416 Proof. Consider the spectral sequence of the Borel fibration G{K Ñ pG{KqK Ñ BK. Recall from
3417 Theorem 6.6.1 and Theorem 7.3.11 that the rings HK˚ and H ˚ pG{Kq are both concentrated in even
3418 degree. By Corollary 3.3.11, the spectral sequence collapses at E2 , so by Proposition 8.1.2, the
3419 action is equivariantly formal.

3420 Recall (Definition 7.3.1) that pG, Kq is said to be cohomology-surjective if H ˚ pGq ÝÑ H ˚ pKq is
3421 surjective.
3422 Proposition 9.3.2 ([Shi96, Cor. 4.2, p. 180]). A cohomology-surjective pair pG, Kq is isotropy-formal.
3423 In fact, cohomology-surjectivity admits a characterization in terms of isotropy-formality, due
3424 to Oliver Goertsches and Sam Noshari.
3425 Proposition 9.3.3 (Goertsches–Noshari, 2014 [GN15, Prop. 3.3]). Let pG, Kq be a compact pair. Then
3426 the following are equivalent:
3427 1. pG, Kq is isotropy-formal and NG pSq{NK pSq is connected.

3428 2. pG, Kq is a cohomology-surjective.

3429 Proof. Let S be a maximal torus of K. If pG, Kq is cohomology-surjective, then by Theorem 7.3.2,
3430 we have H ˚ pG{Kq – ΛPG {{ ΛPK, so
ˇ ˇ
ˇπ0 NG pSqˇ
rk G´rk K ˇ 2rk G´rk K “ h‚ pG{KqS ,
‚
` ˘
h pG{Kq “ 2 ď ˇˇ
π0 NK pSqˇ
3431 by Corollary 9.2.1, ˇ which ˇimplies
ˇ byˇ Proposition 9.2.2 that pG, Kq is isotropy-formal and by
3432 Lemma 8.2.9 that ˇπ0 NG pSqˇˇ{ˇπ0 NK pSq
ˇ “ 1.
ˇ
ˇ ˇ
3433 On the other hand, if π0 NG pKq “ ˇπ0 NG pSqˇ, then by Proposition 9.2.2 again, isotropy-
ˇ ˇ
3434 formality is just the demand h‚ pG{Kq “ 2rk G´rk K , which happens if and only if pG, Kq is cohomology-
3435 surjective, by Proposition 7.3.6.

3436 Recall from Definition A.4.9 that a generalized symmetric pair pG, Kq is a compact pair such that
3437 K is the identity component of the fixed point set of some finite-order continuous automorphism
3438 θ P Aut G, and that pG, Kq is a symmetric pair in the event θ 2 “ id. Goertsches proved in 2011 that
3439 such symmetric pairs are isotropy-formal.
 Chapter 9. Equivariant formality of isotropy actions 132

3440 Theorem 9.3.4 (Goertsches [Goe12, Sec. 1, Theorem]). All symmetric pairs pG, Kq are isotropy-formal.
3441 Noshari, in his master’s thesis work, generalized the result to generalized symmetric pairs.
3442 Theorem 9.3.5 (Goertsches–Noshari [GN14, Thm. 5.6]). All generalized symmetric pairs pG, Kq are
3443 isotropy-formal.
3444 It is worth summarizing the proofs of these results, as they comprise a substantial portion of
3445 the existing literature on isotropy-formality.
3446 The original proof for symmetric pairs [Goe12] breaks symmetric pairs into equal-rank pairs,
3447 split-rank pairs, and outer symmetric pairs. The equal-rank case follows from Proposition 9.3.1;
3448 split-rank pairs turn out to be cohomology-surjective, so that Proposition 9.3.2 applies; and the
3449 outer symmetric pairs form a small, completely classified set of exceptional cases which are
3450 checked individually.
3451 The proof [GN14] for generalized symmetric pairs pG, Kq proceeds on related but distinct
3452 lines. Let S be a maximal torus of K. The analysis of the outer symmetric pairs pG, Kq in the
3453 earlier Goertsches paper was abetted by a subdivision of the Cartan subalgebra s of K into
3454 polyhedral cones called compartments, which are analogous to Weyl chambers of WG but instead
3455 are permuted simply transitively by π0 NG pKq – NG pKq{ZG pKq. There exists a closed, connected
3456 group H, the folded subgroup, sharing a maximal torus S with K, intermediate between K and G
3457 in that K ď H ď G, and such that the compartments in s are the Weyl chambers for H. The group
3458 H is called “folded” because, on the Lie algebra level, its Dynkin diagram ∆ H is a quotient of ∆G
3459 by a graph automorphism.
3460 This folded pair pG, Hq is again a generalized symmetric pair, and the earlier arguments about
3461 compartments are now reflected in the replacement of the isotropy subgroup K with this folded
3462 H. Noshari shows that the folded pair pG, Hq is in fact cohomology-surjective, so that Proposi-
3463 tion 9.3.2 or Proposition 9.3.3 applies and pG, Hq is both isotropy-formal and (by Theorem 7.3.2)
3464 formal. Noshari has an argument, subsumed by our Theorem 9.1.4, that if pG, Hq is formal and
3465 K shares a maximal torus with H, then pG, Hq is isotropy-formal if and only if pG, Kq is, so this
3466 shows the initial generalized symmetric pair was isotropy-formal.
3467 Breaking news. Personal communication with Oliver Goertsches informs us that work into this
3468 question proceeds apace and the finite-order restriction on the automorphism θ has been by-
3469 passed. The state of the art is now thus:
3470 Theorem 9.3.6 (Goertsches–Noshari [GN15]). Let pG, Kq be a compact pair such that K “ pG θ q0 is
3471 the identity component of the fixed point set of a continuous group automorphism θ of G. Then pG, Kq is
3472 isotropy-formal and formal.
3473 The new proof goes much more briefly through a reduction to irreducible such pairs pG, Kq
3474 and a pre-existing classification of all such cases.

3475 The criterion of Shiga and Takahashi

3476 On a related note, in 1996, Hiroo Shiga took up the question of isotropy-formality [Shi96], and
3477 working in the Cartan model, discovered the following sufficient condition.
3478 Theorem 9.3.7 (Shiga). Let pG, Kq be a compact pair and N “ NG pKq the normalizer. Suppose
3479 1. pG, Kq is formal, and
 Chapter 9. Equivariant formality of isotropy actions 133

3480 2. the map H ˚ pG{Kq N ãÑ H ˚ pG{Kq Ñ H ˚ pGq induced by G ÝÝ G{K is injective.

3481 Then pG, Kq is isotropy-formal.

3482 We have discussed the notion of a formal pair at some length in Section 7.4, but the other
3483 condition, that H ˚ pG{Kq N ÝÑ H ˚ pGq be injective, is less transparent. Fortunately, Shiga provided
3484 more easily interpreted equivalent conditions.

3485 Definition 9.3.8. Let pG, Kq be a compact pair with normalizer N “ NG pKq. We say the pair pG, Kq
3486 is invariant-surjective if the map HG˚ ÝÑ HKN of polynomial invariants is a surjection.

3487 Then Theorem 9.3.7 admits the following equivalent restatement.

3488 Proposition 9.3.9 (Shiga). If a compact pair pG, Kq is formal and invariant-surjective, then it is isotropy-
3489 formal.

3490 Shiga actually includes his statement of this result the condition that pG, Kq be formal, but
3491 by Theorem 7.3.2, this demand is redundant. In a later-written (but earlier-published) technical
3492 report [ST95] coauthored with Hideo Takahashi, Shiga also provides a partial converse to this
3493 result.

3494 Theorem 9.3.10 (Shiga–Takahashi). Let pG, Sq be a formal pair with S a torus. Then pG, Sq is isotropy-
3495 formal if and only if it is invariant-surjective.

3496 Shiga and Takahashi actually also require S to contain regular elements invoke regular ele-
3497 ments to ensure the centralizer Z “ ZG pSq is a maximal torus of G, so that Z{S is a torus and
3498 h‚ pZ{Sq “ 2rk G´rk S , but only this dimensional condition is used in the proof, and by Lemma 7.3.8,
3499 it goes through whether or not S contains regular elements. Shiga informs us (personal communca-
3500 tion) he is aware the regularity condition is redundant.
3501 In fact, we noticed the S in the theorem need not be a torus either.

3502 Corollary 9.3.11. Let pG, Kq be a formal pair. Then pG, Kq is isotropy-formal if and only if it is invariant-
3503 surjective.

3504 Proof. Let S be a maximal torus of K. Then pG, Kq is formal, isotropy-formal, or invariant-
3505 surjective if and only if pG, Sq is, by, respectively, Proposition 7.4.10, Theorem 9.1.4, and the
3506 following lemma, so the result follows from Theorem 9.3.10.

3507 Lemma 9.3.12. The map HG˚ ÝÑ pHK˚ q NG pKq is surjective if and only if the map HG˚ ÝÝ pHS˚ q NG pSq is.

3508 Proof. Retreating to real coefficients everywhere, there is a natural isomorphism HS˚ – Rrss, and it
3509 will be enough to show that under this identification, the image of pHK˚ q NG pKq in Rrss is Rrss NG pSq .
3510 Given g P NG pKq, since gKg´1 “ K, it follows gSg´1 is a maximal torus of K, so since all
3511 maximal tori of K are conjugate, there exists k P K such that kgSpkgq´1 “ S, or in other words,
3512 kg P NG pSq. Thus NG pKq “ K ¨ NG pSq as subgroups of G.
„
3513 There is a canonical isomorphism RrksK ÝÑ HK˚ , the Chern–Weil isomorphism (see, e.g.,
3514 [GGK02, p. 209]) so we can write
˘ N pKq
pHK˚ q NG pKq – RrksK G “ RrksK¨NG pKq “ RrksK¨NG pSq .
`
 Chapter 9. Equivariant formality of isotropy actions 134

3515 The Chevalley restriction theorem (see, e.g., [GGK02, Prop. C.12, p. 200]) states that restriction of
„
3516 invariant polynomials from k to s yields an isomorphism RrksK Ý Ñ RrssWK . Since WK “ NK pSq{S
3517 injects into NG pSq{S, we then have
˘ N pSq
pHK˚ q NG pKq – RrksWK G – Rrss NG pSq – pHS˚ q NG pSq .
`

3518 Theorem 9.3.13. Let pG, Kq be a compact pair. Then pG, Kq is isotropy-formal if and only if it is invariant-
3519 surjective and formal.

3520 Remark 9.3.14. The Shiga–Takahashi criterion initially appeared to us to be an inapplicable curios-
3521 ity, its proof sui generis and honestly a bit impenetrable, but as we have seen (Theorem 9.3.13), the
3522 criterion is actually very powerful, and in particular, we still do not know of any isotropy-formal
3523 pairs pG, Kq that are not formal. With proper interpretation our result Theorem 10.1.8 can be seen
3524 as a consequence.
3525 Chapter 10
3526 Equivariant formality of isotropic torus actions

3527 This chapter forms the core of our original work on equivariant formality of isotropy actions. We
3528 begin with a summary.

3529 Survey of original work on isotropy-formality

3530 Independent of any determination of which pairs are isotropy-formal, we are able to determine
3531 the equivariant cohomology of any isotropy-formal and formal pair.
Λ
`
3532 Theorem 10.1.1. Let pG, Kq be an isotropy-formal and formal pair. If we write p – im H ˚ pG{Kq ÝÑ
P
˘
3533 H ˚ pGq for the Samelson subring, the equivariant cohomology of the isotropy action of K on G{K is given
3534 as a left HK˚ -algebra by
H ˚ pG{Kq – H ˚ b H ˚ b Λ P.
` ˘
p
K K ˚
K
HG

3535 The surjection HK˚ pG{Kq ÝÝ H ˚ pG{Kq is then given by

„
pε b idq b id : HK˚ b HK˚ b Λ Pp ÝÑ k b HK˚ b Λ Pp ÝÑ pHK˚ {{ HG˚ q b Λ P,
` ˘ ` ˘
p
˚ ˚
HG HG

3536 where ε : HK˚ ÝÑ k is the augmentation.

3537 The reduction of an isotropy action to a toral action was sufficiently important to our discus-
3538 sion of the literature that we were forced to state it earlier, but we reproduce it here.
3539 Theorem 9.1.4. Let pG, Kq be a compact pair and S a maximal torus of K. The pair pG, Kq is isotropy-
3540 formal if and only if pG, Sq is.
3541 This result reduces the study of isotropy-formality to understanding embeddings of tori in
3542 Lie groups, an already more feasible-looking endeavor. Further, the question reduces to the case
3543 the commutator subgroup K is simply-connected.
3544 Theorem 10.1.2. Let pG, Sq be a compact pair with S a torus. If G r is a finite central covering of G and Sr
3545 the identity component of the preimage of S in G,
r then pG, Sq is isotropy-formal if and only if pG,r Sq
r is.

3546 The proof is in Section 10.3. In fact, the question nearly reduces to the case G itself is simply-
3547 connected.
3548 Theorem 10.1.3. Let pG, Sq be a compact pair with S a torus, K the commutator subgroup of G, and
3549 S1 “ pS X Kq0 the identity component of its intersection with S. Then the pair pG, Sq is isotropy-formal if
3550 and only if both

135
 Chapter 10. Equivariant formality of isotropic torus actions 136

3551 1. pK, S1 q is isotropy-formal and

3552 2. any element of K which normalizes S1 normalizes all of S.

3553 The proof is in Section 10.4. These reductions having been achieved, we completely determine,
3554 as a proof of concept, for precisely which circle subgroups S1 of G the pair pG, S1 q is isotropy-
3555 formal.

3556 Definition 10.1.4. Let pG, Sq be a compact pair, with S a torus. Then S is said to be reflected in G
3557 if there exists g P G normalizing S and such that gsg´1 “ s´1 for all s P S.

3558 Put another way, a torus is reflected if conjugation by some g P G acts as inversion on S.

3559 Algorithm 10.1.5. Let pG, Sq be a compact pair with S a circle. The following steps determine
3560 whether pG, Sq is isotropy-formal.

3561 1. Check whether S lies in the commutator subgroup K of G, a semisimple group. If not, then
3562 pG, Sq is isotropy-formal.

3563 2. If S is contained in K, continue. We may assume by Theorem 10.1.2 that K is a product

3564 of simple Lie groups K j as listed in Proposition B.4.5. Find the images S j of S under the
3565 compositions S ãÑ K  K j .

3566 3. For each K j , determine from Table 10.1.6 whether S j is reflected in K j .

3567 4. If each S j is reflected in K j , then pG, Sq is isotropy-formal.

3568 5. If some S j is not reflected in K j , then pG, Sq is not isotropy-formal.

Table 10.1.6: Reflected lines in simple Lie algebras

Type of K The circle S in K is reflected . . .

An if the exponent multiset J satisfies J “ ´J (see Remarks 10.1.7(b)).
Bn always.
Cn always.
3569
D2n always.
D2n`1 if S is contained in a D2n subgroup.
G2 always.
F4 always.
E6 if S is contained in a D4 subgroup.
E7 always.
E8 always.

3570 Remarks 10.1.7. (a) To ask a circle S lie in the commutator subgroup K is equivalent to asking the
3571 image of π1 S ÝÑ π1 G be infinite or the image of H 1 pGq ÝÑ H 1 pSq be trivial (Proposition 10.5.2
3572 and Remark 10.5.3).
 Chapter 10. Equivariant formality of isotropic torus actions 137

3573 (b) The exponent multiset J is the sequence of integers a1 , . . . , an P Z, considered without order,
3574 but with multiplicity, such that the injection S1 pS1 qn realizing S as a circular subgroup
3575 of a maximal torus of Upnq (the diagonal subgroup, say) is given by z ÞÝÑ pz a1 , . . . , z an q. We
3576 write ´J for the multiset t´a j u1ďjďn whose entries are the opposites of those of J; that is to say,
3577 for each a P Z, the element ´a occurs in ´J with the same multiplicity that a occurs in J. See
3578 Proposition E.1.2.
3579 (c) The demand a circle S in a group of type D2n`1 lie within a D2n subgroup means the Lie
3580 algebra s of S is conjugate into the subspace sop2q‘2n ‘ t1u‘2 of the standard (block diagonal)
3581 infinitesimal maximal torus sop2q‘2n`1 . See Proposition E.1.3.
3582 (d) The manifestation of the condition that a circle S ď E6 lying within a given maximal torus
3583 T 6 of E6 also be contained in a D4 subgroup is somewhat complicated and is the subject of
3584 Appendix E.1.3. One succinct statement is Proposition E.1.14.
3585 The succinct statement Algorithm 10.1.5 is the concatenation of Theorem 10.1.8, Proposi-
3586 tion 10.5.2, Proposition 10.6.3, and Proposition 10.6.4. The construction of Table 10.1.6 is taken up
3587 in Section 10.7. The constituent Theorem 10.1.8 in particular is important enough to be stated in
3588 this introduction.
3589 Theorem 10.1.8. Let G be a compact, connected Lie group and S a circular subgroup of G. There are the
3590 following three mutually exclusive cases.
3591 1. The inclusion S ãÝÝÑ G surjects in cohomology and S is not reflected in G.

3592 2. The inclusion S ãÝÝÑ G is trivial in cohomology and

3593 2a. S is reflected in G.

3594 2b. S is not reflected in G.

3595 Only in the case 2b is pG, Sq not isotropy-formal.

3596 Here are two sample consequences.
3597 Proposition 10.1.9 (anonymous referee, 2015). Let pG, Kq be a compact pair, where K – SUp2q –
3598 Spp1q or else K – SOp3q. Then pG, Kq is isotropy-formal.
3599 Proof. These K are precisely those of type A1 , whose maximal torus is a circle S. This circle S is
3600 reflected in K because the Weyl group WA1 – t˘1u. It follows that S is also reflected in G, so by
3601 Theorem 10.1.8, the pair pG, Sq is isotropy-formal. By Theorem 9.1.4, pG, Kq is isotropy-formal as
3602 well.
` ˘
3603 Example 10.1.10. If S is a circle in the unitary group Upnq, then Upnq, S is or is not isotropy-
3604 formal as indicated in Table 10.1.11.

Table 10.1.11: The classification for circles in Upnq

` ˘
Embedding of S Is Upnq, S isotropy-formal?
S ę SUpnq Yes
S ď SUpnq and J “ ´J Yes
S ď SUpnq and J ‰ ´J No
 Chapter 10. Equivariant formality of isotropic torus actions 138

3605 The equivariant cohomology of isotropy-formal pairs

3606 In this section, we prove the following statement.

Λ
`
3607 Theorem 10.1.1. Let pG, Kq be an isotropy-formal and formal pair. If we write p – im H ˚ pG{Kq ÝÑ
P
˘
3608 H ˚ pGq for the Samelson subring, the equivariant cohomology of the isotropy action of K on G{K is given
3609 as a left HK˚ -algebra by
HK˚ pG{Kq – HK˚ b HK˚ b Λ P.
` ˘
p
˚
HG

3610 The surjection HK˚ pG{Kq ÝÝ H ˚ pG{Kq is then given by

„
pε b idq b id : HK˚ b HK˚ b Λ Pp ÝÑ k b HK˚ b Λ Pp ÝÑ pHK˚ {{ HG˚ q b Λ P,
` ˘ ` ˘
p
˚ ˚
HG HG

3611 where ε : HK˚ ÝÑ k is the augmentation.

3612 This is a consequence of a computational technique for the cohomology of biquotients, due
3613 to Jost-Hinrich Eschenburg [Esc92] (and modeled on Borel’s proof of Cartan’s theorem as recre-
3614 ated in Section 7.1.3), and refined by Vitali Kapovitch [Kap], who constructs a DGA which the
3615 Serre spectral sequence of Eschenburg’s two-sided Borel fibration can be seen to be the filtration
3616 spectral sequence of.
Let U ď G ˆ G be a closed subgroup. It acts on G by a restriction of the pG ˆ Gq-action

pG ˆ Gq ˆ G ÝÑ G,
pg, hq ¨ x :“ gxh´1 .

3617 The homotopy quotient GU , also written G {{ U, can be viewed as the biquotient

EG ˆ G ˆ EG
.
peg, x, he1 q „ pe, gxh, e1 q

3618 The U-equivariant projection EG ˆ G ˆ EG ÝÑ EG ˆ EG descends to a map G {{ U ÝÝ BU :“

3619 pEG ˆ EGq{U which is the projection in a non-principal G-bundle. This projection descends fur-
3620 ther to a map G {{ U ÝÑ BG, where the target BG can be more explicitly specified as

EG ˆ EG EG ˆ EG
“ .
∆G peg, e1 q „ pe, ge1 q

3621 This quotient is a BG because EG ˆ EG is contractible and the G-action quotiented out is free.
3622 There is yet a further quotient BG ˆ BG “ pEG{Gq ˆ pG{EGq, and the map BG ÝÑ BG ˆ BG
3623 is the projection in another non-principal G-bundle (Eschenburg calls this the reference bundle).
3624 These bundles are related by a bundle map

G G

 
G {{ U / BG (10.1)

 
BU / BG ˆ BG,
σ
 Chapter 10. Equivariant formality of isotropic torus actions 139

3625 where the base map σ can be understood as BpU ãÝÝÑ G ˆ Gq.
3626 As with the bundle map (7.4) inducing the spectral sequence map we used to understand the
3627 ordinary cohomology H ˚ pG{Kq, the bundle map suggests that to understand H ˚ pG {{ Uq it will
3628 suffice to understand the Serre spectral sequence of the reference bundle G Ñ BG Ñ BG2 . This
3629 sequence is relatively simple to understand.
3630 Proposition 10.2.1 (Eschenburg, [Esc92, Prop., p. 157]). In the Serre spectral sequence of the bundle
3631 G Ñ BG Ñ BG2 all primitives transgress and the transgression takes the generators z P PG to τz b
3632 1 ´ 1 b τz P HG˚ b HG˚ , where τz P QpBGq is the transgression of z in the Serre spectral sequence of the
3633 universal bundle G Ñ EG Ñ BG as discussed in Section 6.6.
3634 The composition d : z ÞÝÑ σ˚ pτz b 1 ´ 1 b τzq extends to a differential on H ˚ pBUq b H ˚ pGq,
3635 which as in Theorem 7.1.3, calculates the cohomology H ˚ pG {{ Uq.
3636 Theorem 10.2.2 ([Kap, Prop. 1]). Let G be a compact, connected Lie group and U a closed, connected
3637 subgroup of G ˆ G. The cohomology of G {{ U is given by
` ˘
H ˚ pG {{ Uq – H ˚ H ˚ pBUq b H ˚ pGq, d
3638 where d is the unique antiderivation extending
τ id b1´1bid σ˚
PG Ñ QpBGq ãÑ HG˚ ÝÝÝÝÝÝÝÑ HG˚ b HG˚ Ñ H ˚ pBUq.
3639 Sketch of proof. Although neither of the G-bundles in (10.1) is principal, all that is necessary for
3640 the proof of Theorem 7.1.3 to go through is that a such a bundle map exists and all primitives
3641 z` P PG transgress in the Serre spectral sequence of the reference bundle. The new complex C is
3642 APL pBUq b H ˚ pGq, equipped with the differential d “ d BU ` σ˚ i˚ pτ b 1 ´ 1 b τq, where i˚ : HG˚ ÝÑ
3643 APL pBG ˆ BGq is a quasi-isomorphism as provided by Example A.4.8 since BG ˆ BG » BpG2 q.
3644 Since U is a compact, connected Lie group, H ˚ pBUq is a polynomial algebra by Borel’s The-
3645 orem
` 6.6.1, and˘ so BU
` is˚ formal˘ by Example A.4.8. By Proposition 7.1.8, then, we can replace
3646 APL pBUq, d BU with H pBUq, 0 in the Cartan algebra.

3647 Remark 10.2.3. If U is a product H ˆ K for some subgroups H, K ď G, then G {{ pH ˆ Kq “ H GK is

3648 just the biquotient of Definition 9.1.3, and BU » BH ˆ BK.

3649 Proof of Theorem 10.1.1. The result is an application of Theorem 10.2.2 in the case U “ K ˆ K. Thus
` ˘
HK˚ pG{Kq – H ˚ pK GK q – H ˚ HK˚ b HK˚ b H ˚ pGq, d .
3650 Here d is the unique antiderivation vanishing on HK˚ b HK˚ and taking PG ÝÑ HK˚ b HK via
z ÞÝÑ ρ˚ τz b 1 ´ 1 b ρ˚ τz,
3651 where ρ˚ “ pBiq˚ : HK˚ ÝÑ HG˚ is induced functorially by the inclusion i : K ãÝÝÑ G.
3652 From the quotient map GK » EK ˆ GK ÝÑ EK ˆ GK one has induced following bundle map
K
3653 relating the one- and two-sided Borel fibrations:
G G
j
 
GK / K GK (10.2)
κ
χ
 
BK / BK ˆ BK;
 Chapter 10. Equivariant formality of isotropic torus actions 140

3654 the base map is homotopic to a factor inclusion pt ˆ BK ãÝÝÑ BK ˆ BK and consequently the
3655 resulting map in cohomology is the factor` ˚ projection HK˚ b HK˚  k b HK˚ – HK˚ .
3656 Since pG, Kq is formal, H pG{Kq – HK {{ HG q b Λ P,
˚ ˚ p where Λ Pp – im j˚ and H ˚ {{ H ˚ – im χ˚ .
K G
3657 By assumption κ ˚ is surjective, so

impj˚ ˝ κ ˚ q “ im HK˚ pGK q ÝÑ H ˚ pGq – Λ P.

` ˘
p

3658 By the coproduct decomposition Proposition 7.2.5, it remains only to understand the other factor,
3659 the cohomology of C “ H ˚ pHK˚ b HK˚ b Λ P,q dq, where Pq – PG{ Pp is the Samelson complement.
3660 Since pG, Kq is formal, by Theorem 7.4.7, the kernel of χ˚ : HK˚ ÝÑ HK˚ {{ HG˚ is a regular
3661 ideal with kernel freely generated by the images ρ˚ τz` of a basis z` of P, q which form a regular
3662 sequence in HK . It follows that ρ τz` b 1 ´ 1 b ρ τz` forms a regular sequence in HK˚ b HK˚ and
˚ ˚ ˚

3663 by Theorem 7.4.7 again,

HK˚ b HK˚
H ˚ pCq – – HK˚ b HK˚ .
pρ˚ τz` b 1 ´ 1 b ρ˚ τz` q ˚
HG

3664 As corollaries, we obtain particularly simple expressions for popular classes.

3665 Corollary 10.2.4. Suppose pG, Kq is a cohomology-surjective pair. Then

HK˚ pG{Kq – HK˚ b H ˚ pG{Kq

3666 as an HK˚ -algebra.

3667 Proof. Because H ˚ pG{Kq ÝÑ H ˚ pGq is injective, we have H ˚ pG{Kq “ Λ P,
p and because H ˚ ÝÑ H ˚
G K
3668 is surjective, we have HK˚ b HK˚ – HK˚ b HK˚ – HK˚ .
˚
HG HK˚

3669 Corollary 5.2.7. Let G be a compact Lie group and K a connected, closed subgroup of equal rank. Then

HK˚ pG{Kq – HK˚ b HK˚

˚
HG

3670 as an HK˚ -algebra.

3671 New proof. Because Λ Pp – k, we have HK˚ pG{Kq – HK˚ b HK˚ .
˚
HG

3672 Lifting to the universal compact cover

3673 Recall the structure theorem Theorem B.4.4 for compact Lie groups, which states, inter alia, that a
3674 compact, connected Lie group G admits a finite central extension π : G
r ÝÑ G which is the direct
3675 product of a semisimple, simply-connected Lie group K and a torus A: if the fiber of π is F, we
3676 may write
r “ A ˆ K F.
L
G – G{F
3677 This G
r is a universal compact cover.
3678 In determining which toral isotropy actions are equivariantly formal, we would like to replace
3679 G with Gr and the torus S in G with the identity component Sr “ π ´1 pSq0 of its preimage in G,
r a
3680
´1
torus of equal rank. Note that π pSq “ F S, and recall this cohomological lifting lemma proven
r
3681 in Appendix B.
 Chapter 10. Equivariant formality of isotropic torus actions 141

3682 Proposition B.3.4. Let G be a compact connected Lie group and K a closed, connected subgroup, let G r be
3683 the universal compact cover of G (see Theorem B.4.4), and Kr the identity component of the preimage of K
3684 in G,
r and let F be the kernel of p : G
r ÝÑ G. If |F{pF X Kq|
r is invertible in k, then

H ˚ pG{Kq – H ˚ pG{
r Kq.
r

3685 As total Betti number is unchanged under the substitution pG, Sq ÞÝÑ pG,
r Sq,
r we want to see
3686 r “ N r pSq
the same is true of normalizer components. Write N G
r and N “ NG pSq for the normalizers
3687 of the tori.

3688 Proposition 10.3.1. Under the foregoing assumptions, the projection π : G

r ÝÑ G induces an isomor-
„
3689 phism π0 N
r ÝÑ π0 N.

3690 Proof. Since π is a homomorphism, it sends N r ÝÑ N, inducing the claimed map π0 N r ÝÑ π0 N.

3691 To see this map is injective, suppose wrPN r is such that w “ πpwq r induces the identity on S.
3692
´1 ´1
Then for all s P S we have wsw s “ 1, so since π is a homomorphism, w rr r´1r
sw s´1 P F for all
3693 s P S.
r r Since rs ÞÝÑ w
rr r´1r
sw s´1 is a continuous function Sr ÝÑ F to a discrete space, sending 1 ÞÑ 1,
3694 w
r must centralize S.r
3695 To see the map is surjective, given w P N, let w
r be any preimage in G. r Since π is a homomor-
3696 phism, conjugation by w ´1
r must take Sr into π pSq, and since it fixes 1, it must in fact take Sr ÝÑ S,
r
3697 so that w
r P N.
r

3698 These facts in hand, we conclude the proof of Theorem 10.1.2.

3699 Theorem 10.1.2. Let pG, Sq be a compact pair with S a torus. If G r is a finite central covering of G and Sr
3700 the identity component of the preimage of S in G, then pG, Sq is isotropy-formal if and only if pG,
r r Sq
r is.

3701 Proof. We know from Proposition 9.2.2 that pG, Sq is isotropy-formal if and only if

h‚ pG{Sq “ ˇπ0 NG pSq ˇ2rk G´rk S .

ˇ ` ˘ˇ

3702
r “ rk G and rk Sr “ rk S; from Proposition B.3.4, we know h‚ pG{
But evidently rk G r “ h‚ pG{Sq;
r Sq
` ˘ ` ˘
3703 and from Proposition 10.3.1, we know π0 NGr pSqr ÐÑ π0 NG pSq .

3704 In what follows, we can therefore replace G with its universal compact cover G r “ A ˆ K. For
3705 later, when we specialize to circles, we note the following corollary of Proposition 10.3.1.

3706 Corollary 10.3.2. Under these hypotheses, the torus S is reflected in G just if Sr is reflected in G.
r

3707 Reduction to a semisimple group

In this section, G is the product of a torus A and a simply-connected Lie group K. Let S be a
torus in G and

K “ K{pS X Kq,
G “ G{S,
KS
A “ cokerpS Ñ Aq.

3708 The trivial bundle K Ñ G Ñ A induces a bundle K Ñ G Ñ AKS we claim is also trivial.
 Chapter 10. Equivariant formality of isotropic torus actions 142

3709 Proposition 10.4.1. Given the above hypotheses, G « K ˆ AKS .

3710 Proof. The kernel of the composition S ãÑ G  A is S X K, so there is an isomorphism

„
S :“ S{pS X Kq ÝÑ impS Ñ Aq “: AkS

3711 of tori. Factoring out S X K, we have

KˆA KˆA
G “ «
S S
3712 by the third isomorphism theorem. As A is a torus, the projection A ÝÝ AKS splits, by Proposi-
3713 tion A.4.2, so that A may be factored as A – AKS ˆ AkS in the display, and

KˆA K ˆ AkS KˆS

« ˆ AKS « ˆ AKS
S S S

3714 But pK ˆ Sq{S « K by Lemma 1.1.1.

3715 Now we can carry through the claimed (near-)reduction to the semisimple case.

3716 Theorem 10.1.3. Let pG, Sq be a compact pair with S a torus, K the commutator subgroup of G, and
3717 S1 “ pS X Kq0 the identity component of its intersection with S. Then the pair pG, Sq is isotropy-formal if
3718 and only if both

3719 1. pK, S1 q is isotropy-formal and

3720 2. any element of K which normalizes S1 normalizes all of S.

3721 Proof. By Proposition 10.3.1, we may as well assume that G “ A ˆ K, so that N “ NG pSq “ A ˆ
3722 NK pSq. Note further that because K is normal in G, any group
` element
˘ normalizing S also must
3723 normalize S . Write N “ NK pS q. Thus we have π0 N – π0 NK pSq ď π0 N 1 . From Lemma 8.2.9 as
1 1 1

3724 applied to the pair pK, S1 q and Corollary 9.2.1, we have

1 1
h‚ pK{S1 q ě |π0 N 1 | 2rk K´rk S ě |π0 N| 2rk K´rk S . (10.3)

3725 Write AKS “ cokerpS Ñ Aq and AkS “ impS Ñ Aq, so that rk A “ rk AkS ` rk AKS and
3726 rk S “ rk AkS ` rk S1 . Then we know

rk G ´ rk S “ prk A ` rk Kq ´ prk AkS ` rk S1 q “ rk AKS ` rk K ´ rk S1 .

KS KS
3727 From Proposition 10.4.1, we have h‚ pG{Sq “ 2rk A h‚ pK{S1 q, so multiplying (10.3) by 2rk A yields

h‚ pG{Sq ě |π0 N 1 | 2rk G´rk S ě |π0 N| 2rk G´rk S . (10.4)

3728 Corollary 9.2.3 states that pG, Sq is isotropy-formal if and only if the inequalities (10.4) are in fact
3729 equalities, which is equivalent to (10.3) being equalities. But by Corollary 9.2.3 again, this can
3730 only happen if pK, S1 q is isotropy-formal and additionally π0 pN 1 q – π0 pNq.
 Chapter 10. Equivariant formality of isotropic torus actions 143

3731 Remarks 10.4.2. (a) There do exist cases where the inequality |π0 N| ď |π0 N 1 | is strict, making
3732 Theorem 10.1.3 something less than a full reduction to the semisimple case. For instance, let
3733 G “ A ˆ K “ S1 ˆ SUp2q2 , pick a circle S1 in SUp2q, let T be the maximal torus A ˆ S1 ˆ S1 of G,
3734 and let
S “ pz; w, zw´1 q : z, w P S1
(

3735 be a rank-two subtorus of T, so that

S1 “ p1; w, w´1 q : w P S1 .
(

3736 Then |π0 N 1 | “ 2, the nontrivial element of π0 N 1 acting on T as pz; t, uq ÞÝÑ pz; t´1 , u´1 q since A is
3737 central. But this map takes pz; w, zw´1 q to pz; w´1 , z´1 wq, which is not in S unless z “ z´1 “ ˘1,
3738 so π0 N “ 1.
3739 (b) Write T 1 for a maximal torus of K containing impS Ñ Kq. The demand each element w P N 1
3740 normalize S is equivalent to the demand that for each pa, tq P S ď A ˆ T 1 , we have pa, wtw´1 q P S,
3741 so that wtw´1 t´1 P S1 “ S X K for all t P impS Ñ T 1 q. Equivalently, the induced action of N 1
3742 on impS Ñ T 1 q{S1 is trivial. It is not clear, however, this reformulation is more enlightening or
3743 applicable than the original.

3744 Equivariant formality of isotropic circle actions

3745 It is now our goal to demonstrate the statements Algorithm 10.1.5 and Table 10.1.6 regarding
3746 equivariant formality of circle actions.
3747 We first make some preliminary remarks on π0 N. In general, continuous automorphisms of
3748 a torus S – pR{Zqr “ Rr {Zr are induced by linear isometries of Rr fixing the integer lattice Zr ,
3749 which correspond to elements
( of GLpr, Zq, so that Aut S – GLpr, Zq. When S is a circle, r “ 1
3750 and GLp1, Zq “ r˘1s , so by Lemma 1.4.7 the component group π0 N must be trivial or be
3751 generated by the involution s ÞÝÑ s´1 . Thus to determine |π0 N| in this case, it will suffice to
3752 determine whether there is any element g P G such that g´1 sg “ s´1 for all s P S. Recall from
3753 Definition 10.1.4 that a circle is said to be reflected if there is some such element g. We summarize:
` ˘
3754 Proposition 10.5.1. If pG, Sq is a compact pair and S a circle, then the cardinality of π0 NG pSq is 2 if S
3755 is reflected in G and 1 otherwise.
3756 A complete analysis of when a circle is reflected is conducted in Section 10.6. To be able to
3757 apply Corollary 9.2.3, it remains to understand h‚ pG{Sq, which was a consequence of the theorem
3758 Proposition 7.5.1 of Leray and Koszul.
3759 Proposition 7.5.1. Let G be a compact, connected Lie group and S a circle subgroup. Then the rational
3760 cohomology ring H ˚ pG{Sq has one of the following forms.
3761 1. If H 1 pGq ÝÑ H 1 pSq is surjective, then there is z1 P H 1 pGq such that
˚
H ˚ pG{Sq – H pGq pz1 q.
L

3762 In terms of total Betti number, h‚ pGq “ 21 h‚ pG{Sq.

3763 2. If H 1 pGq ÝÑ H 1 pSq is zero, there are z3 P H 3 pGq and s P H 2 pG{Sq such that
H ˚ pGq Qrss
H ˚ pG{Sq – b 2 .
pz3 q ps q
3764 In terms of total Betti number, h‚ pGq “ h‚ pG{Sq.
 Chapter 10. Equivariant formality of isotropic torus actions 144

3765 Now we can prove Theorem 10.1.8.

3766 Theorem 10.1.8. Let G be a compact, connected Lie group and S a circular subgroup of G. There are the
3767 following three mutually exclusive cases.

3768 1. The inclusion S ãÝÝÑ G surjects in cohomology and S is not reflected in G.

3769 2. The inclusion S ãÝÝÑ G is trivial in cohomology and

3770 2a. S is reflected in G.

3771 2b. S is not reflected in G.

3772 Only in the case 2b is pG, Sq not isotropy-formal.

3773 Proof. Recall (Corollary 9.2.3) that h‚ pN{Sq “ |π0 N| 2rk G´rk S and that pG, Sq is isotropy-formal
3774 if and only if h‚ pG{Sq “ h‚ pN{Sq. Proposition 10.5.1 imposes the constraint |π0 N| P t1, 2u, and
Proposition 7.5.1 the constraint h‚ pG{Sq P 12 h‚ pGq, h‚ pGq . By Lemma 8.2.9, it is impossible that
(
3775

3776 both h‚ pG{Sq “ 12 h‚ pGq and |π0 N| “ 2, so there are only the following three cases.

3777 1. We have h‚ pG{Sq “ 12 h‚ pGq and |π0 N| “ 1. Equivariant formality is achieved.

3778 2. We have h‚ pG{Sq “ h‚ pGq, and

3779 2a. |π0 N| “ 2. Equivariant formality is achieved.

3780 2b. |π0 N| “ 1. Equivariant formality is not achieved.

3781 It is convenient to be able to express the conditions on the map H 1 pGq ÝÑ H 1 pSq in terms of
3782 the intersection of S with the commutator subgroup K.

3783 Proposition 10.5.2. Let G be a compact, connected Lie group and S a toral subgroup. The inclusion
3784 S ãÝÝÑ G is trivial in cohomology if and only if S is contained in the commutator subgroup K.

3785 Proof. Note that being contained in the commutator subgroup is invariant under taking covers,
3786 and recall from Section 10.3 that the rank of H 1 pGq ÝÑ H 1 pSq is as well, so we may assume
3787 G “ A ˆ K with A a torus and K simply-connected. Note that S ď K just when the composition
3788 S Ñ G Ñ A is trivial.
3789 If S Ñ G Ñ A is trivial, then H 1 pAq Ñ H 1 pGq Ñ H 1 pSq is trivial. Since π1 K “ H1 pK; Zq is
3790 zero, so also are H1 pKq and H 1 pKq, meaning H 1 pAq – H 1 pGq by the Künneth theorem. Thus in
3791 this case H 1 pGq ÝÑ H 1 pSq is also trivial.
3792 On the other hand, if S Ñ G Ñ A is not trivial, its image is a circle in A, possibly traversed
3793 multiple times, so that S ÝÑ A induces a nonzero map H1 pSq ÝÑ H1 pAq. Dualizing, H 1 pAq ÝÑ
3794 H 1 pSq is nonzero.

3795 Remark 10.5.3. The torus S is actually contained in the commutator subgroup K of G if and only if
3796 its image in A “ G{K is trivial. Since H ˚ pA; Qq – H ˚ pG; Qq by the lifting lemma Proposition B.3.4
3797 and the topological Künneth Theorem B.2.2, it follows that this occurs if and only if H ˚ pG; Qq ÝÑ
3798 H ˚ pS; Qq is trivial. Dualizing, that happens if and only if H˚ pS; Qq ÝÑ H˚ pG; Qq is trivial. Because
3799 H˚ pS; Qq is an exterior algebra on generators H1 pS; Qq, that is equivalent to H1 pS; Qq ÝÑ H1 pG; Qq
3800 being trivial, or the image of H1 pS; Zq ÝÑ H1 pG; Zq being torsion, or equivalently, the kernel
 Chapter 10. Equivariant formality of isotropic torus actions 145

3801 having full rank. Since S and G are groups, their fundamental groups are abelian, so this is
3802 actually the map π1 S ÝÑ π1 G. Thus one has the statement

S ď K ðñ impπ1 S ÝÑ π1 Gq is finite.

3803 In the event S – S1 , the statement is that S lies in K if and only if π1 S ÝÑ π1 G fails to be injective,
3804 and conversely, that π1 S ÝÑ π1 G is injective if and only if S does not lie in K.
3805 Remark 10.5.4. The results on circles can be seen as a consequence of the (slightly enhanced)
3806 Shiga–Takahashi criterion Theorem 9.3.10, though only the sufficiency direction follows from the
3807 published version.
3808 Let pG, Sq be a compact pair, S a circle, and N “ NG pSq. The pair is formal by Lemma 7.5.5,
3809 so pG, Sq is isotropy-formal if and only if HG˚ ÝÑ HSN is surjective by Corollary 9.3.11. Note
3810 that |N| ď 2 by Proposition 10.5.1. If |N| “ 1, then HSN “ HS˚ , so Shiga–Takahashi says that
3811 pG, Sq is isotropy formal if and only if HG˚ ÝÑ HS˚ is surjective, which happens if and only
3812 H ˚ pGq ÝÑ H ˚ pSq is surjective by Proposition 7.3.5, the cohomology-surjective case.
3813 If |N| “ 2, on the other hand so that S is reflected in G, then HSN is the fixed point of HS˚ “ Qrss
3814 under the involution s ÞÝÑ ´s, which is the subring Qrs2 s, and the image of HG˚ ÝÑ HSN contains
3815 this subring by Lemma 6.8.3.
3816 The reflected circles still must be determined in some ` way.˘ Shiga himself [Shi96, pp. 80–
3817 82] carries through this program in the special case SUpnq, S1 , showing both necessity and
3818 sufficiency for his conditions in this case and reassuringly arriving at the same result as we did.

3819 Reflectibility of circles: reduction

3820 In the next two sections, we continue the classification implied by Algorithm 10.1.5 by determin-
3821 ing when circles are reflected in a compact Lie groups G. This section reduces reflectibility in
3822 stages to the case G is a semisimple group, then a simply-connected group, and finally a simple
3823 group. First note that elements of NG pSq reflecting S, if they exist, can be represented by elements
3824 of the Weyl group WG “ NG pTq{T of G.
3825 Lemma 10.6.1 ([Bou05, Ex. IX.2.4, p. 391][DW98, Lem. 9.7, p. 20]). Let G be a compact, connected
3826 Lie group, T a maximal torus, and S a subtorus. Given an automorphism of S induced by conjugation by
3827 a normalizing element n P NG pSq, there exists an element w P NG pTq inducing the same automorphism.
3828 Proof. Conjugation by n stabilizes the centralizer Z :“ ZG pSq of S, for given z P Z and s P S, since
3829 nsn´1 P S by normality, we have

pnzn´1 qspnzn´1 q´1 “ nzpn´1 snqz´1 n´1 “ npn´1 snqn´1 “ s.

3830 T is a maximal torus of Z, so also must nTn´1 be, by Theorem B.4.9. All maximal tori of Z are
3831 conjugate, so there exists z P Z such that znTn´1 z´1 “ T, or zn P NG pTq. Since z P Z centralizes
3832 S and nSn´1 “ S, conjugation by w “ zn induces the same automorphism of S as n does.

3833 Write Fix NG pTq pSq for the set of elements of NG pTq fixing S pointwise. This association N
3834 NG pTq is not a function, but if n Þ w1 and n Þ w2 , then conjugation by w1´1 w2 fixes S pointwise.
3835 so N NG pTq descends to a well-defined homomorphism
NG pTq WG
N ÝÑ –
ZG pSq X NG pTq FixWG pSq
 Chapter 10. Equivariant formality of isotropic torus actions 146

3836 with kernel Z; thus, to determine |π0 N|, we need only survey Weyl group elements. We state this
3837 as a proposition.

3838 Proposition 10.6.2. Let G be a compact, connected Lie group and S a torus in G. Write N “ NG pSq for
3839 its normalizer and Z “ ZG pSq for its centralizer. Then the conjugation action of G induces an injection
3840 π0 N – N{Z  WG {FixWG pSq. In particular, S is reflected in G if and only if conjugation by some
3841 element of the Weyl group WG of G induces s ÞÝÑ s´1 on S.

3842 Note that for the purposes of the previous definition, S need not be a circle. Now, however
3843 assume S is a circle in G. From Proposition 10.3.1 and Corollary 10.3.2, we may assume G is the
3844 product A ˆ K of a torus A and a simply-connected Lie group K.

3845 Proposition 10.6.3. Let G be a compact, connected Lie group and S a toral subgroup. Then S is reflected
3846 in G if and only if it is reflected in the commutator subgroup K.

3847 Proof. Assuming G “ A ˆ K, from Theorem 10.1.8 and Proposition 10.5.2, we know that S is not
3848 reflected unless it is contained in K. Further, since the conjugation action of A is trivial, if S is
3849 reflected, it is also reflected in K.

3850 Reflectibility of a circle in a simply-connected group in turn depends only on simple factors,
3851 for the fact Lemma B.4.10 that Weyl group of a product is the product of the Weyl groups of the
3852 factors then translates into the expected statement about reflectibility in products.

3853 Proposition 10.6.4. Let K be a product of Lie groups K j . If S is a toral subgroup of K and S j , for each j,
3854 its image under the projection K ÝÝ K j , then S is reflected in K just if each S j is reflected in K j .

3855 Proof. Write Tj for maximal tori of K j containing S j . If S is reflected in K, then there is some
ś ś
3856 element pw j q P W “ WK j such that pw j q reflects S. Since the projections Tj ÝÑ Tj are W-
3857 equivariant, it follows each S j is reflected by w j in K j .
3858 Now suppose each S j is reflected in K j . Then each S j is reflected by some w j P Wj , so the list
ś ś
3859 w “ pw j q reflects the torus S j in K; and S is a subtorus of S j .

3860 So, to determine reflectibility of a given circle in a compact, connected Lie group, we need
3861 only lift it to the universal cover, verify that it intersects any toral component trivially, and check
3862 whether or not its projection to each simple factor is reflected.
3863 We will make repeated use of the following.

3864 Observation 10.6.5. Let G be a compact, connected Lie group and H a closed, connected sub-
3865 group. If a torus S of H is reflected in H, it is also in G.

3866 Reflected circles in simple groups

3867 In this section we let G be a simple Lie group and classify reflected circles S in G. If we fix
3868 a maximal torus T of G, then by Theorem B.4.9, some conjugate gSg´1 is contained in T, and
3869 conjugation x ÞÝÑ gxg´1 takes NG pSq to NG pgSg´1 q, so that S is reflected in G if and only if its
3870 conjugate gSg´1 is. So from now on, we can just assume S is in our chosen maximal torus T.

3871 Proposition 10.7.1. Let G be a simple Lie group and S a circular subgroup. Then S is reflected in G if
3872 and only if it is reflected by the longest word w0 in the Weyl group W of G.
 Chapter 10. Equivariant formality of isotropic torus actions 147

3873 Proof. The circle S determines a line s in the Lie algebra t of T. Arbitrarily orienting s, we can
3874 write it as the union s` Y ´s` of a “positive” and “negative” ray. By Proposition B.4.18, the
3875 nonzero points of s` lie in a unique closed Weyl chamber C, and by Definition B.4.17 there exists
3876 a system ∆ “ tα1 , . . . , αdim T u of simple roots such that C is the “positive” closed Weyl chamber in
3877 t where all α j ě 0. It then follows that ´s` lies in the negative closed Weyl chamber ´C, where
3878 all α j ď 0. The Weyl group acts simply transitively on Weyl chambers by Proposition B.4.18, so
3879 only the longest word w0 P W sends ∆ ÝÑ ´∆ and hence C ÝÑ ´C, and there will exist an
3880 element w P W taking s` ÝÑ ´s` if and only if w0 does so.

3881 Chi-Kwong Fok (personal communication) pointed out this fact as well a representation-
3882 theoretic consequence.

3883 Proposition 10.7.2. Let G be a simple Lie group and S a circular subgroup. Then S is reflected in G if
3884 and only if the irreducible representation of G determined by S is self-dual.

3885 Proof. Identifying t with t_ through the Killing form, (the image of) s` meets the weight lattice
3886 of G and determines a minimal dominant weight λ; and S is reflected if and only if some Weyl
3887 group element takes λ to ´λ. Again, if any Weyl element does so, then w0 will, so S is reflected
3888 if and only if λ “ ´w0 λ. The representation dual to the irreducible representation with highest
3889 weight λ is that with highest weight ´w0 λ.

3890 The construction of Table 10.1.6 now proceeds case by case through the Killing–Cartan clas-
3891 sification Proposition B.4.5.

3892 Proposition 10.7.3. Let T be a maximal torus in a simple Lie group G whose type is one of

Bn , Cn , D2n , G2 , F4 , E7 , E8 .

3893 Then S is reflected in G.

3894 First proof. If G is a simple Lie group of one of these types, its Weyl group W is a Coxeter group
3895 of corresponding type. For Coxeter groups of these types, the longest word w0 acts as ´id on the
3896 vector space (here t) carrying the defining representation of W [Kan01, Lem. 27-2, p. 283][Hum92,
3897 p. 82], so t is reflected.

3898 Second proof. It is known that central involutions of a Weyl group act as X ÞÝÑ ´X on the Lie
3899 algebra t of a maximal torus [DW01, Thm. 1.8] and that the center of W is isomorphic to Z{2 for
3900 the Weyl groups of types Bn , Cn , D2n , G2 , F4 , E7 , and E8 [DW01, Rmk. 1.9].

3901 In the remaining cases, the longest word w0 P W does not act as ´id on t, so more work is
3902 required.

3903 Proposition 10.7.4. Let G be a simple Lie group with trivial center (viz., of type An , D2n`1 , or E6 ) and
3904 let S be a circle in G. Then S is reflected in G if and only if there is some w P W such that w ¨ s lies in
3905 the fixed point subalgebra tθ of the Cartan subalgebra under an automorphism θ P Autptq induced by a
3906 nontrivial diagram automorphism of the Dynkin diagram of G.

3907 Proof. From Proposition 10.7.1 we know S is reflected if and only if s is reflected by w0 . Note that
3908 this occurs if and only if s is fixed pointwise by the nontrivial automorphism ´w0 P Autptq. This
3909 automorphism ´w0 stabilizes but does not fix the positive closed Weyl chamber C, and so cannot
 Chapter 10. Equivariant formality of isotropic torus actions 148

Figure 10.7.5: The graph involutions of An

α1 α2 α3 α` α1 α2 α`´1
T T T ... T T T ... T α`

   ...    ... 
α 2` α2`´1 α2`´2 α``1 α2`´1 α2`´2 α``1

3910 be induced from W, which acts simply transitively on Weyl chambers. Likewise, any extension
3911 of ´w0 to an automorphism of g is an outer automorphism in that it is not in the image of Ad G.
3912 But it is known [FH91, Prop. D.40, p. 498] that all outer automorphisms of g are induced by a
3913 graph automorphism of the Dynkin diagram Γ of G, in the sense that

Aut g
Out g :“ – Aut Γ.
Ad G
3914 It thus remains only to understand the fixed point subalgebras of diagram automorphisms
3915 for Lie algebras of type An , D2n`1 , and E6 .

3916 Proposition 10.7.6. In a Lie algebra of type An , a point v P t_ ă Rn`1 of the dual Cartan algebra is fixed
3917 by an automorphism of the Dynkin diagram if and only if some reversing the coordinates of v yields ´v.

3918 Proof. The Dynkin diagram of An and its only automorphism θ are depicted in Figure 10.7.5;
3919 the map θ acts on simple roots of An by exchanging α j ÐÑ αn´j , as in the figure. The fixed
3920 point subspace of t_ – Rn is spanned by the vectors α j ` αn´j so it contains those vectors v1 “
3921 ra1 ¨ ¨ ¨ an s P Rn for which a j “ an´j . These α j are usually identified with e j ´ e j`1 P Rn`1 , where
3922 pe j q1ďjďn`1 is the standard basis on Rn`1 .1 This embedding v1 ÞÝÑ v P Rn`1 takes
“ ‰
ra1 a2 ¨ ¨ ¨ an´1 an s ÞÝÑ a1 pa2 ´ a1 q ¨ ¨ ¨ pan ´ an´1 q ´ an ,

3923 so the symmetry requirement a j “ an´j translates into the antisymmetry condition v j “ ´vn`1´j
3924 on the coordinates of v.

3925 Corollary 10.7.7. Let K “ SUpnq and let S be a circular subgroup. Then S is reflected if and only if the
3926 exponent multiset2 J of the inclusion S ãÝÝÑ T satisfies J “ ´J.

3927 For example, r´1 0 1s P R3 meets this condition and r2 1 ´ 3s does not.
„
3928 Proof. Using the W-equivariant isomorphism t_ ÝÑ t discussed in Corollary B.4.16, we may
3929 identify tθ with pt_ qθ ă Rn`1 . The Weyl group WAn “ Sn`1 acts on Rn`1 by permutation of coor-
3930 dinates, so given v P t, there exists a w P Sn`1 such that w ¨ v P tθ if and only if some permutation
3931 of the entries of ´v yields v. The result then follows immediately from Proposition 10.7.6 and
3932 Proposition 10.7.4.
1 Thus t_ is usually considered as the subspace of elements whose coordinates sum to zero in the Cartan algebra
Rn`1 for Upn ` 1q.
2 A multiset is like a sequence, except elements are not ordered, but are counted with multiplicity. Other collections

often considered as multisets are roots of a polynomial and eigenvalues of an operator.

 Chapter 10. Equivariant formality of isotropic torus actions 149

Figure 10.7.9: The graph involution of D2n`1

α2 n
α1 α2 α2n´1 T
...

α2n`1

3933 Remarks 10.7.8. (a) The root subsystems of A2` and A2`´1 fixed by θ are respectively of types Bn
3934 and Cn . The former corresponds on the group level to the inclusion SOp2` ` 1q ãÝÝÑ SUp2` ` 1q
3935 and the latter to the Sppnq embedded in SUp2nq via the block map on coordinates induced by
3936 the ring injection H C2ˆ2 . These subgroups are fixed points of involutive automorphisms of
3937 SUpnq yielding the classical symmetric spaces SUpnq{SOpnq and SUp2nq{Sppnq.
3938 (b) The self-duality criterion Proposition 10.7.2 rears its head here as follows. The representation
3939 τ of S on Cn given by restricting the defining representation of SUpnq to S is a direct sum
Àn
3940 ρba j of tensor powers of the defining representation ρ of S1 on C. The dual representation
j“1 À
3941 τ “ nj“1 ρbp´a j q , will be isomorphic to τ just if J “ ´J.
_

3942 Proposition 10.7.10. In a Lie algebra of type D2n`1 , a point v P t_ ă R2n`1 of the dual Cartan algebra
3943 is fixed by an automorphism of the Dynkin diagram if and only if the last coordinate of v is zero.

3944 Proof. The Dynkin diagram of D2n`1 and its lone graph automorphism θ are shown in Fig-
3945 ure 10.7.9. This θ fixes all simple roots except α2n and α2n`1 , which it exchanges. The fixed point
3946 subspace of pt_ qθ is spanned by the vectors tα j u jă2n and by α2n ` α2n`1 . The roots α j for j ď 2n
3947 are usually identified with e j ´ e j`1 P R2n`1 , and a2n`1 with en ` en`1 , where pe j q1ďjďn`1 is the
3948 standard basis on R2n`1 . The image of the composite embedding pt_ qθ ãÑ t_ Ñ R2n`1 is R2n ˆ t0u
3949 since α2n ` α2n`1 “ 2e2n .

3950 Corollary 10.7.11. Let S be a circle in Spinp4n ` 2q. Then S is reflected in Spinp4n ` 2q if and only if it
3951 is conjugate into a Spinp4nq subgroup.
„
3952 Proof. Using the W-equivariant isomorphism t_ ÝÑ t discussed in Corollary B.4.16, we may
3953 identify pt_qθ with tθ “ R2n ˆ t0u ă R2n`1 . The Weyl group WD2n`1 “ t˘1u2n ¸ S2n`1 acts on
3954 R2n`1 by permutating its coordinates and inverting an even number of them, so given v P t, there
3955 exists a w P WD2n`1 such that w ¨ v P tθ if and only if one coordinate is zero. The result then follows
3956 immediately from Proposition 10.7.10 and Proposition 10.7.4.

3957 Remark 10.7.12. The sublattice of a D2n`1 lattice fixed by θ is of type B2n and corresponds to a
3958 SOp4nq subgroup of SOp4n ` 2q. This subgroup is the fixed point set of an involutive automor-
3959 phism of SOp4n ` 2q yielding the classical symmetric space SOp4n ` 2q{SOp4nq “ V2 pR4n`2 q.
 Chapter 10. Equivariant formality of isotropic torus actions 150

Figure 10.7.13: The graph involution of E6

α1 α2
T T α3 α4

 
α6 α5

3960 Proposition 10.7.14. In a Lie algebra of type E6 , a point v P t_ of the dual Cartan algebra is fixed by an
3961 automorphism of the Dynkin diagram if and only if it lies in a certain F4 sublattice.

3962 Proof. The Dynkin diagram of E6 and its only automorphism θ are depicted in Figure 10.7.13;
3963 the map θ acts on simple roots of E6 as indicated in the picture, and pt_ qθ is spanned by ∆ “
3964 tα1 ` α6 , α2 ` α5 , α3 , α4 u. By assumption, we have αi ¨ α j “ ´2|αi ||α j | for adjacent αi , α j and αi ¨ α j “ 0
3965 otherwise, so ∆ is a simple root system of type F4 with a1 ` α6 and α2 ` α5 long and α3 and α4
3966 short.

3967 Proposition 10.7.15. A circular subgroup S of E6 or its universal cover E

r6 is reflected just if it is conjugate
3968 into a Spinp8q subgroup.
„
3969 Proof. Using the W-equivariant isomorphism t_ ÝÑ t discussed in Corollary B.4.16, we may
3970 identify pt_ qθ with tθ . It follows immediately from Proposition 10.7.14 and Proposition 10.7.4 that
3971 the tangent spaces s ă t to reflected circles S are precisely those such that there exists w P WE6
3972 such that w ¨ s ă tθ . Now because pt_qθ is spanned by an F4 sublattice of the E6 root lattice, its
3973 dual tθ is tangent to the maximal torus T 4 of an F4 subgroup. In the classic series of inclusions
3974 Spinp8q ă F4 ă E6 , the first two share a maximal torus T 4 , so tθ is actually tangent to the maximal
3975 torus of a Spinp8q.

3976 Remark 10.7.16. We had a much more intricate proof of the classification in Table 10.1.6 before it
3977 was pointed out to us by Chi-Kwong Fok (personal communication) and Jay Taylor [Tay] that the
3978 longest word of the Weyl group yields a simpler (no pun intended) classification. This original
3979 proof is reproduced for comparison in Appendix E.
3980 Included in particular are detailed original proofs that there are precisely forty-five such T 4
3981 in a given maximal torus T 6 of E6 . containing all and only reflected circles. A reader who finds
3982 our current justification that pt_qθ corresponds to an F4 subgroup of E6 sketchy will find there
3983 a much more detailed explanation why this must be the case and why in fact tθ must be are
3984 tangent to maximal tori of Spinp8q subgroups. We will also show there that each of the associated
3985 t4 are spanned by four mutually orthogonal coroots in the E6 coroot lattice. As a consequence,
3986 we accidentally demonstrate the well-known fact that WF4 injects into WE6 with index forty-five.
3987 Finally, we will provide an explicit description of the 135 mutually orthogonal bases of roots of
3988 pt6 q_ which span the forty-five reflected subspaces pt4 q_ .
3989 Appendix A
3990 Algebraic background

3991 In this appendix we gather a ragtag assortment of algebraic preliminaries. Notationally, in all
3992 that follows we denote containment of an algebraic substructure by “ď,” containment of an ideal
3993 by “E,” isomorphism by “–,” and bijection by “Ø.” The restriction of a map f : A ÝÑ B to a
3994 subset U Ď A is written f æ U.

3995 Commutative algebra

3996 We will only need a very little commutative algebra, but it will include the following. Let A be
3997 an ungraded commutative ring and B a unital A-algebra. We say an element a P A annihilates B
3998 if a1 “ 0 in B. Given any a P A we may localize, or invert a in A, to form a new A-algebra
A a :“ Ara´1 s – Arxs{pax ´ 1q.
3999 The class of x in A a , called a´1 , serves as an inverse to a in Ara´1 s; if Ara´1 s “ 0 is the zero ring,
4000 this still holds for uninteresting reasons.
4001 We define localization of an A-module M by
Ma :“ Mra´1 s :“ A a b M.
A

4002 Localization is a exact functor in that it takes an exact sequence of A-modules to an exact se-
4003 quence of Ara´1 s-modules.
4004 The localization Mra´1 s turns out to be trivial if and only if each element m P M is annihilated
4005 by some power an of M ([AM69, Ex. 3.1, p. 43]); the idea is that then m “ x n an m “ x n 0 “ 0 in
4006 Mra´1 s. We say that M is a-torsion in this instance. For a unital A-algebra B to be a-torsion, it
4007 is necessary and sufficient that a power of a annihilate the unity 1 P B, and then Bra´1 s “ 0. We
4008 have the following useful lemma.
4009 Lemma A.1.1. Let A be a ring, a an element of A, and
MÑNÑP
4010 an exact sequence of A-module homomorphisms. Then if M and P are a-torsion, so also is N.
4011 Proof. Since localization is an exact functor, the localized sequence
Mra´1 s ÝÑ Nra´1 s ÝÑ loomoon
loomoon Pra´1 s,
0 0

4012 is also exact, meaning Nra´1 s is zero.

151
 Appendix A. Algebraic background 152

4013 We will also need an corollary of Nakayama’s lemma and a version of the Krull intersection
4014 theorem.

4015 Lemma A.1.2 (Nakayama’s lemma; [AM69, Cor. 2.7, p. 22]). Let A be a commutative ring, M a
4016 finitely generated A-module, N a submodule of M, and a E A an ideal contained in the Jacobson radical.
4017 If M “ aM ` N, then M “ N.

4018 Proposition A.1.3 ([AM69, Cor. 10.19, p. 110]). Let A be a Noetherian ring, a an ideal contained in its
Jacobson radical, and M a finitely-generated A-module. Then 8 n
Ş
4019
n“0 a M “ 0.

4020 Corollary A.1.4 ([GHV76, Lemma 2.8.I, p. 62]). Let k be a field ring and A “ krx1 , . . . , xn s a poly-
4021 nomial ring in finitely many indeterminates, and write a “ px1 , . . . , xn q C A for the ideal of positive-
4022 degree polynomials. Let M be a finitely-generated A-module and V a k-submodule of M, and suppose
4023 M “ aM ` V. Then M “ AV.

4024 This is just an application of Nakayama’s lemma A.1.2 to the case N “ AV.

Alternate proof. Iteratively substituting the entire left-hand side of M “ aM ` V in for the occur-
rence of M on the same-side, one inductively finds

M “ aM ` V
“ a2 M ` aV ` V
¨¨¨
n
ÿ
“ an`1 M ` a j V.
j“0

Ş8 nM
ř8 n V,
4025 Intersecting all right-hand sides yields M “ n“0 a ` n“0 a but by Proposition A.1.3,
Ş n
4026
n m M is zero.

4027 Chain complexes

4028 Our cohomology theories will always take coefficients in an ungraded, commutative ring k with
4029 unity; usually, k will be Q or R. The category of k-modules and k-module homomorphisms
4030 is denoted k-Mod. A differential group is a pair pA, dq, where A P k-Mod is a k-module and
4031 d P Endk A, the differential, is a nilsquare endomorphism, so that the composition d2 :“ d ˝ d “ 0
4032 is the constant map to the zero element of A. A homomorphism f : pA, dq Ñ pB, δq in the category
4033 of differential groups, a group homomorphism f : A Ñ B such that f d “ δ f .
À
4034 A Z-graded k-module is an A P k-Mod admitting a direct sum decomposition A “ nPZ An .
4035 An element a P A is homogeneous if there exists some integer |a| “ deg a, the degree of A, such
4036 that a P Adeg a . We blur the distinction between 0 P An and 0 P A, and leave the degree of the
4037 latter indeterminate. A k-module homomorphism f : A ÝÑ B between graded k-modules is said
4038 to be a graded k-module homomorphism of degree n “ deg f if

deg f paq “ n ` deg a “ deg f ` deg a

4039 for all homogeneous a P A. We let gr-k-Mod be the category of graded k-modules and graded
4040 k-module homomorphisms.
 Appendix A. Algebraic background 153

4041 A cochain complex pA, dq is a differential group such that A P gr-k-Mod and additionally d
4042 is of degree 1. A homomorphism of cochain complexes, as described in the first paragraph of the
4043 subsection, is then called a cochain map. 1 We write d æ An “: dn . A map f : pA, dq Ñ pB, δq of
4044 cochain complexes is a cochain map of differential groups that is additionally a graded map of
4045 degree 0, so that f An ď Bn . We let k-Ch denote the category of cochain complexes and cochain
4046 maps of k-modules,
4047 The cohomology HpA, dq of a differential k-module pA, dq is the quotient pker dq{pim dq, which
4048 makes sense because d2 “ 0. We also write this as Hd pAq. The differential group is exact if
4049 Hd pAq “ 0. A cochain map f : pA, dq Ñ pB, δq induces a homomorphism f ˚ : HpA, dq Ñ HpB, δq
4050 of k-modules. If this map is an isomorphism, then one says f is a quasi-isomorphism.
4051 If A is a chain complex, then HpA, dq is graded by

H n pA, dq :“ H ˚ pA, dqn :“ ker dn { im dn´1 .

4052 Then a (graded) cochain map induces a map of graded modules, so cohomology is a functor
4053 k-Ch ÝÑ gr-k-Mod. A chain complex pA, dq is said to be acyclic if H ˚ pA, dq “ H 0 pA, dq “ k,
4054 meaning H n pA, dq “ 0 for n ‰ 0.
4055 We will say a map A ÝÑ B of differential groups surjects in cohomology or is H ˚ -surjective
4056 if it induces a surjection H ˚ pAq ÝÝ H ˚ pBq. In the opposite extreme case, that the map H ˚ pAq ÝÑ
4057 H ˚ pBq is zero in dimensions ě 1 and is the isomorphism H 0 pAq ÝÑ H 0 pBq in dimension 0, we
4058 call this map trivial, and say the map X ÝÑ Y is trivial in cohomology. If A ÝÑ B is the map
4059 f ˚ : H ˚ pYq ÝÑ H ˚ pXq in cohomology induced by a continuous map f : X ÝÑ Y, then we likewise
4060 say f is surjective in cohomology or trivial in cohomology if f ˚ is.

4061 Polynomials and numbers associated to a graded module

4062 A graded k-module A is said to be of finite type if each graded component An has finite k-rank.
4063 Given a graded k-module A of finite type, we define the Poincaré polynomial of A to be the
4064 formal rational function ÿ
ppAq :“ prkk An qtn .
nPZ
ř
4065 The sum ppXq|t“1 “ rkk An is the total rank or total Betti number of A. If the total Betti
4066 number of A is finite, then when we evaluate at t “ ´1 instead, we get the Euler characteristic
χpAq :“ ppXq|t“´1 “ p´1qn rkk An ; otherwise the Euler characteristic is undefined.
ř
4067

4068 Given a chain complex, Euler characteristic is preserved under cohomology: one has the
4069 following corollary of the fundamental rank–nullity theorem of linear algebra, as applied to the
4070 differential d.

4071 Proposition A.2.1. Let pA, dq be a chain complex over k of finite total Betti number. Then
` ˘
χpAq “ χ H ˚ pA, dq .
1 A chain complex is a differential group pA, dq with A P gr-k-Mod and deg d “ ´1; a homomorphism of chain
complexes is a chain map. Chain and cochain complexes are mirror images of each other under the reindexing
An “ A´n , and we will usually focus our attention on cochain complexes.
 Appendix A. Algebraic background 154

4072 Commutative and differential graded algebras

4073 A cohomology ring is a commutative graded algebra, and it is defined as the cohomology of a
4074 chain complex which is itself a graded algebra. We set out some commonplaces of these objects.

4075 Commutative graded algebras

4076 A cohomology ring A will be a graded commutative k-algebra. This means A is a graded k-
4077 module, and additionally the product is such that

Am ¨ An ď Am`n ;

4078 and for all homogeneous elements a, b P A, one has

ba “ p´1q|a||b| ab.

4079 Mostly, these rings will actually be N-graded, so that An “ 0 for n ă 0, and the absolute co-
4080 homology rings H ˚ pXq (as opposed to relative cohomology rings H ˚ pX, Yq) will be unital, so
4081 that the map x ÞÑ x ¨ 1 embeds k  A0 ãÑ A and the k-algebra structure can be seen as the
4082 restriction of the ring multiplication A ˆ A ÝÑ A. We will call these k-CGAs for short, and the
4083 category of graded commutative k-algebras and degree-preserving k-algebra homomorphisms
4084 will be written k-CGA.
4085 The cohomology theories of interest to us will also be algebras over CGAs E˚ , or E˚ -algebras.
4086 For us an E˚ -algebra structure on a CGA A is a k-bilinear map E˚ ˆ A ÝÑ A that “adds degrees”
4087 in that
˚
Em ¨ An ď Am`n .
4088 For absolute (as opposed to relative) cohomology theories, as we shall deal in almost exclu-
4089 sively, these maps will arise from the multiplication of A via precomposition with k-algebra
4090 homomorphisms E˚ ÝÑ A, just as the action of k often is. The associated morphisms in the
4091 category E˚ -CGA are those that preserve the multiplication; for absolute cohomologies, such a
4092 map f : A ÝÑ B is precisely one where the composition E˚ Ñ A Ñ B equals the structure map
4093 E˚ Ñ B.
4094 The product in k-CGA is the ring product A ˆ B, graded by pA ˆ Bqn “ An ˆ Bn . The coproduct
4095 is the graded tensor product: this is A bk B as a group, with the grading
à
pA b Bqn “ A` b Bm
k ``m“n k

4096 and the commutation rule p1 b bqpa b 1q “ p´1q|a||b| a b b for a P A|a| and b P B|b| . As often
4097 as feasible, we suppress ring subscripts on tensor signs, and in elements, we omit the tensor
4098 signs themselves, letting a b b “: ab, so that for example we recover the reassuring expression
4099 ba “ p´1q|a|b| ab.
Given a graded unital k-algebra A with a preferred basis pa j q of A0 ‰ 0, the map

„ `
A0 ÝÑ kta j u ÝÑ k,
ÿ ÿ
γ j a j ÞÝÑ γj
 Appendix A. Algebraic background 155

4100 induces a natural ring homomorphism A ÝÝ A0 ÝÑ k called the augmentation. Its kernel A r is
4101 called the augmentation ideal; the notation is in analogy with reduced cohomology.2 If A0 – k,
4102 we say A is connected; the terminology is because the singular cohomology of a connected space
À
4103 satisfies this condition. In this case, the augmentation ideal is ně1 An .
4104 Given a degree-zero homomorphism f : A ÝÑ B of graded k-algebras, write
` ˘
B {{ A :“ B{ f p Aq
r .

4105 This is the right conception of cokernel for maps between cohomology rings: one wants the 0-
4106 graded component to stay the same and the rest of the image of f to vanish. If A is a graded
4107 subalgebra of B, then one wants to think of

0 Ñ A Ñ B Ñ A {{ B Ñ 0

4108 as a “short exact sequence” of rings, but of course this doesn’t make sense: the sequence A Ñ
4109 B Ñ C of k-modules is exact at B if impA Ñ Bq “ kerpB Ñ Cq, but the image of a ring map is a
4110 ring, while the kernel is an ideal, a different type of algebraic object. The compromise solution is
4111 the following definition.

4112 Definition A.3.1. A sequence A Ñ B Ñ C of homomorphisms of unital k-algebras is said to be

4113 coexact at B if ` ˘
kerpB Ñ Cq “ imp A
r Ñ Bq
r .

4114 One should think of this as the ring-theoretic substitute for exactness in sequences of groups.

4115 Example A.3.2. Let A be a graded k-subalgebra of a graded k-algebra B. Then 0 Ñ A Ñ B Ñ

4116 A {{ B Ñ 0 is a short coexact sequence, by design. If A and C are k-algebras, free as k-modules (in
4117 the applications we care most about, k “ Q), then taking B “ A b C, we see the sequence

0 Ñ A ÝÑ A b C ÝÑ C Ñ 0

4118 is short coexact.

4119 Bigraded algebras

4120 Some k-algebras A we will encounter will have a bigrading:
à p,q
A “ A‚,‚ “ A
p,qPZ

4121 in such a way that the bidegrees pp, qq add under multiplication:

Ai,j ¨ A p,q ď Ai`p,j`q .

4122 We conventionally visualize such a ring as a grid in the xy-plane, with the pth column

A p,‚ “
à p,q
A
q

2 r ˚ is the kernel of the augmentation in

Industry standard seems to be A, but I have resisted this because H
cohomology and I am used to overbar notation referring to quotients.
 Appendix A. Algebraic background 156

4123 residing in the strip p ď x ď p ` 1 and the qth row

à p,q
A‚,q “ A
p

4124 residing in the strip q ď y ď q ` 1. For us, such gradings will always reside in the first quadrant:
4125 we demand pp, qq P N ˆ N.
4126 A linear map f : A Ñ B of bigraded algebras is said to have bidegree bidegp f q “ pp, qq if
f pAi,j q ď Bi`p,j`q . The associated singly-graded k-algebra of a bigraded algebra is A‚ “ n An ,
À
4127

graded by An :“ p,q
À
4128
p`q“n A , and a bigraded algebra will be said to be commutative if this
4129 associated singly-graded algebra and called the total complex, is a CGA. A differential bi-
4130 graded algebra pA, dq is a bigraded algebra such that d is an antideriviation on the associated
4131 singly-graded algebra A‚ of degree 1. We make no further demands initially as to how d inter-
4132 acts with the bigrading, but note that since dAn ď An`1 , one has for each bidegree pi, jq that
dAi,j ď
À i``,j`1´`
4133
`A , and composing with projections to Ai``,j`1´` , one obtains component
4134 maps d` : Ai,j ÝÑ Ai``,j`1´` of bidegree p`, 1 ´ `q such that
ÿ
d“ d` .
` PZ

4135 Free graded algebras

4136 Suppose that char k ‰ 2. As with modules, there are free objects in the category of k-CGAs, which
4137 have the following description. Given a free graded k-module V if we separate it into even- and
4138 odd-degree factors Veven and Vodd , then the free graded commutative k-algebra on V is the graded
4139 tensor product
SVeven bk ΛVodd
of the symmetric algebra SVeven on the even-degree generators and the exterior algebra ΛVodd on
the odd-degree generators. Given k-bases ~t “ pt1 , . . . , tm q of Veven and ~z “ pz1 , . . . , zn q of Vodd , we
also write these as

Sr~ts :“ SVeven ;
Λr~zs :“ ΛVodd .

4140 Write
∆rzm s :“ kt1, zm u,
4141 for the unique rank–two unital k-algebra with elements of degrees zero and m, which is the
4142 cohomology of an m-sphere. This is Λrzm s for m odd and Srzm s{pz2m q for m even.
4143 In the event char k “ 2, the graded commutativity relation xy “ p´1q|x| |y| yx, or equivalently
4144 xy ˘ yx “ 0, forces genuine commutativity xy “ yx for all elements since 1 “ ´1 in k. Thus a free
4145 k-CGA is a symmetric algebra SV in characteristic 2, independent of the grading on V. Algebras
4146 which merely resemble ΛV still play an important role in characteristic two.

4147 Definition A.3.3. Let k be a commutative ring. A k-algebra A (not assumed graded commutative),
4148 free as a k-module, is said to have a simple system of generators V “ pv1 , . . . , vn , . . .q if a k-basis
4149 for A is given by the monomials

v j1 ¨ ¨ ¨ v j` , j1 ă ¨ ¨ ¨ ă j` ,
 Appendix A. Algebraic background 157

4150 where each generator occurs at most once. If A has a simple system of generators, we write

A “: ∆V “: ∆rv1 , . . . , vn , . . .s

4151 despite the fact that this description does not specify A up to algebra isomorphism.
4152 Example A.3.4. The exterior algebra Λrz1 , . . . , zn s admits z1 , . . . , zn as a simple system of genera-
4153 tors.
4154 This is of course the motivating case. The multiplication need not be anticommutative, as one
4155 can see from the following example.
4156 Example A.3.5. Borel [Bor54, Théorème 16.4] found that the mod 2 homology ring of Spinp10q is
4157 given by
H˚ Spinp10q; F2 “ ∆rv3 , v5 , v6 , v7 , v9 , v15 s,
` ˘

4158 where all v2j “ 0 and all pairs of v j commute except for pv6 , v9 q, which instead satisfies

v6 v9 “ v9 v6 ` v15 .

4159 Polynomial rings also afford examples

4160 Example A.3.6. The polynomial ring krxs admits x, x2 , x4 , x8 , . . . as a simple system of generators,
4161 as consequence of the binary representation for natural numbers.
4162 Example A.3.7. The property of admitting a simple system of generators is preserved under tensor
i j
4163 product (e.g., krx, ys admits x2 y2 for i ` j ą 0) so in fact all free CGAs are examples.

4164 Poincaré polynomials for free algebras

4165 In most cases we care about, the Poincaré polynomial will applied to a nonnegatively-graded k-
4166 CGA of finite type. The Poincaré polynomial is a homomorphism gr-k-Mod ÝÑ krts in the sense
4167 that
ppA ˆ Bq “ ppAq ` ppBq, ppA b Bq “ ppAq ¨ ppBq.
4168 Usually the CGA in question will be the cohomology ring H ˚ pX; kq of a space, and we will write
˘ ÿ
rkk H n pX; kqtn .
`
ppXq :“ p H ˚ pX; kq “
nPN

The individual ranks hk pXq :“ rkk H k pX; kq are called the Betti numbers of X; the associated total
rank ppXq|t“1 “ hn pXq is called the total Betti number of the space and denoted h‚ pXq. The
ř

Euler characteristic ppXq|t“´1 “ p´1qn hn pXq of H ˚ pX; kq is called the Euler characteristic of the
ř

space, and written χpXq. If we write heven pXq “ h2n pXq and hodd pXq “ h2n`1 pXq, then
ř ř

h‚ pXq ` χpXq “ 2 ¨ heven pXq;

h‚ pXq ´ χpXq “ 2 ¨ hodd pXq

4169 Free CGAs behave pleasantly under Poincaré polynomial because pp´q is multiplicative. If
deg x “ n is odd, then p Λrxs “ 1 ` tn . Thus given an exterior algebra ΛV on an oddly-graded
` ˘
4170

free k-module V of finite type, with Poincaré polynomial ppVq “ tn j (it is okay if certain n j
ř
4171

4172 occur more than once), the tensor rule yields

ź
ppΛVq “ p1 ` tn j q.
 Appendix A. Algebraic background 158

4173 Likewise, if deg x “ n is even, then Srxs “ krxs is spanned by 1, x, x2 , . . ., so

` ˘ ÿ jn 1
p Srxs “ t “ .
1 ´ tn
jPN

Given a symmetric algebra SV on an evenly-graded free k-module V of finite type, ppVq “ tn j ,

ř
4174

4175 then, the tensor rule yields

ź 1
ppSVq “ . (A.1)
1 ´ tn j
4176 Proposition A.3.8. Let k be a field, V be a positively-graded k–vector space, SV the symmetric algebra,
4177 and W a graded vector subspace of SV such that the subalgebra A it generates is a free CGA and SV is a
4178 free A-module. Then
ppSVq
ppSV {{ Aq “ .
ppSWq
4179 Proof. Let pqα q be a homogeneous A-basis for SV. Then pqα b 1q forms a graded basis for SV {{ A “
4180 SV b Ar k, so on the level of graded k-modules, one has SV – A bk ktqα b 1u – A b pSV {{ Aq.
4181 Taking Poincaré polynomials and dividing through by ppSV {{ Aq gives the result.

4182 This sort of quotient will become relevant to us in Section 7.4, where it will be found that
4183 an important subring of the cohomology ring H ˚ pG{K; Qq, of a compact homogeneous space,
4184 namely the image of the characteristic map χ˚ : H ˚ pBK; Qq ÝÑ H ˚ pG{K; Qq, is frequently of this
4185 form.

4186 Indecomposables
4187 The indecomposable elements of A are, informally, those of positive degree that cannot be written
4188 as sums of products of lower-degree elements; the idea is to find an analogy for irreducible
4189 polynomials for rings with more complex ideal structure. Recalling that The most convenient
4190 definition turns out to be this: the module of indecomposables is the k-module

QpAq :“ A{
r ArA
r–A
rbk
A

4191 where A r is the augmentation ideal and the denominator denoted A rAr is understood to be the
4192 module spanned by products ab for a, b P A r of positive-degree elements. Under this definition we
4193 see Q is functorial, since a graded homomorphism A ÝÑ B takes A r ÝÑ B r and hence A
rA r ÝÑ BrB.
r
4194 If A is a free k-module, then so is QpAq, so the k-module surjection A r ÝÝ QpAq splits, and
4195 we can consider QpAq (in a badly noncanonical way) as a k-submodule of algebra generators for
4196 A. Because it satisfies a product rule, an (anti)derivation d on A, like a ring homomorphism, is
4197 uniquely determined by its values on such a lifted QpAq, so a linear map on QpAq determines at
4198 most one antiderivation of A. .

4199 Differential graded algebras

4200 A chain complex pA, dq concentrated in nonnegative degree such that A is also a graded ring
4201 satisfying the product rule
dpabq “ da ¨ b ` p´1q|a| a ¨ db
 Appendix A. Algebraic background 159

4202 for homogeneous elements a, b is a differential graded algebra (or k-DGA). A differential d on a
4203 graded ring satisfying this condition is called an antiderivation. A derivation, on the other hand,
4204 satisfies
dpabq “ da ¨ b ` a ¨ db.
4205 An (anti)derivation on a unital k-algebra satisfies d1 “ 0 and hence dpk ¨ 1q “ 0. A morphism of
4206 DGAs is a k-algebra map that is simultaneously a cochain map. If A was a k-CGA, then we say
4207 pA, dq is a commutative differential graded algebra (henceforth k-CDGA).
4208 The kernel Z “ ker d of an (anti)derivation d is a subalgebra, because d is additive and
4209 because if da “ db “ 0, then dpabq “ pdaqb ˘ apdbq “ 0. The image B “ im d is an ideal of
4210 Z “ ker d, because if b “ da P B and c P Z, then b P Z and dpacq “ pdaqc ` apdcq “ bc.
4211 The product in the category of DGAs is the graded ring direct product A ˆ B, equipped with
4212 the differential dpa, bq :“ pda, dbq. The coproduct is the same tensor product A b B as for CGAs,
k
4213 equipped with the unique antiderivation given by

dpa b bq “ da b b ` p´1q|a| a b db

4214 on pure tensors. If we omit the tensor signs, this gives back, formally, the same product rule.

4215 The algebraic Künneth theorem

4216 It is trivial that a product of DGAs remains a product on taking cohomology. In an ideal world,
4217 the same would remain true of coproducts, and this ideal world is achieved in the event one of
4218 the DGAs lacks torsion.

4219 Theorem A.3.9. Let k be a principal ideal domain and suppose A and C are graded differential k-modules,
4220 each free as a graded k-module. Then

H n pA b Cq – Tor1k H j`1 pAq, H n´j pCq .

à ` j
H pAq b H n´j pCq ‘
˘ à ` ˘
k 0ďjďn 0ďjďn

4221 Proof. Write Zn “ kerpdn : An ÝÑ An`1 q and Bn “ impdn´1 : An´1 ÝÑ An q. Then one has a short
4222 exact sequence
0 Ñ Z ÝÑ A ÝÑ B‚`1 Ñ 0
4223 of complexes where the differentials on Z and B‚`1 are 0. Since we have assumed C is flat, on
4224 tensoring these complexes with C, we obtain a short exact sequence

0 Ñ Z b C ÝÑ A b C ÝÑ B‚`1 b C Ñ 0

4225 of complexes, where the differentials on Z‚ b C and B‚`1 b C are both id A bdC and the differential
4226 on A‚ b C is the expected d A b idC ˘ id A b dC . Write i‚ : B‚ ÝÑ Z‚ for the inclusion; then it is
4227 not hard to see the the connecting map in the long exact sequence in cohomology is the map
4228 pi b idC q˚ : B‚ b H ˚ pCq ÝÑ Z‚ b H ˚ pCq induced by i b idC . Thus we get a short exact sequence

0 Ñ cokerpi b idC q˚ ÝÑ H ˚ pA b Cq ÝÑ kerpi‚`1 b idC q˚ Ñ 0.

4229 Because 0 Ñ B`‚`1 ÝÑ Z‚`1 ÝÑ H ‚`1 pAq Ñ 0 is exact, the first term is H ˚ pAq bk H ˚ pBq and
the last is Tor1k H ˚`1 pAq, H ˚ pCq . Resorting summands to gather equal total degrees yields the
˘
4230

4231 statement of the theorem.

 Appendix A. Algebraic background 160

4232 In particular, one has the following.

4233 Corollary A.3.10. Let A and C be k-DGAs free as k-modules and such that H ˚ pCq is flat over k. Then

H ˚ pA b Cq – H ˚ pAq b H ˚ pCq
k k

4234 as k-algebras.

4235 Proof. The hypotheses precisely ensure the Tor1k term vanishes.

4236 Note that it more than suffices k be a field.

4237 Poincaré duality algebras

4238 The real cohomology ring of a compact manifold exhibits an important phenomenon which we
4239 generalize to an arbitary CGA.

4240 Definition A.3.11. Let A be a k-CGA, free as a k-module. Suppose there exists a maximum n P N
4241 such that An ‰ 0, that An – k, and that for all j P r0, ns the natural pairing

A j ˆ An´j ÝÑ An

4242 obtained by restricting the multiplication of A is nondegenerate. Then we call A a Poincaré

4243 duality algebra (or PDA) and a nonzero element of An a fundamental class for A, which we
4244 write as rAs. If we fix a homogeneous basis pv j q of A, we can define a linear map a ÞÝÑ a˚ on A
4245 by setting v˚j :“ vn´j whenever v j vn´j “ rAs and extending linearly. Such a linear map is called a
4246 duality map.

4247 Theorem A.3.12 (Poincaré; [BT82, I.(5.4), p. 44]). If M is a compact manifold, the real singular coho-
4248 mology ring H ˚ pM; Rq is a PDA.

4249 Example A.3.13. Let V be a finitely generated, oddly-graded free k-module. Then the exterior
4250 algebra ΛV is a Poincaré duality algebra with fundamental class given by the product of a basis
4251 of V.
4252 Poincaré duality is a severe restriction on the structure of a ring, with powerful consequences,
4253 and it is inherited by tensor-factors.

4254 Proposition A.3.14. Let A and B be k-CGAs, free as k-modules, and suppose B is a PDA. Then A b B
4255 exhibits Poincaré duality just if A does.

4256 Sketch of proof. If A and B are PDAs with duals given by a ÞÝÑ a˚ and b ÞÝÑ b˚ , then a b b ÞÝÑ
4257 a˚ b b˚ is easily seen to be a duality on A b B up to sign. If, on the other hand, b ÞÝÑ b˚ is a
4258 duality on A and a b b ÞÝÑ a b b is a duality on A b B, then for any homogeneous a P A one has
4259 a b 1 “ a˚ b rBs for some a˚ P A, and a ÞÝÑ a˚ is a duality on A.
 Appendix A. Algebraic background 161

4260 Splittings and formality

4261 An epimorphism A ÝÝ B is said to split if there exists a monomorphism B A, called a
4262 section, such that the composition B Ñ A Ñ B is the identity on B. This section is virtually never
4263 canonical, but it is frequently useful to be able to lift the structure of B back into A, in however
4264 haphazard a manner.
4265 Surjective homomorphisms onto free objects always split in categories whose objects carry a
4266 group structure (we always assume the axiom of choice), and we use this simple fact repeatedly.

4267 Proposition A.4.1. Let π : A ÝÝ F be a surjection in gr-k-Mod and suppose F is free. Then π splits.

4268 Proof. Let S be a k-basis for F and for each s P S pick a preimage as P π ´1 tsu. This assignment
4269 extends to the needed section.

4270 Restricting to the case everything lies in one graded component, one obtains the result in
4271 k-Mod. Specializing to the category S1 -Mod of modules over S1 – R{Z (which are projective
4272 limits of tori, hence called pro-tori), one obtains the following useful statement.

4273 Proposition A.4.2. Any exact sequence 0 Ñ A ÝÑ B ÝÑ C Ñ 0 of tori splits: we can write B –
4274 A ‘ σpCq as an internal direct sum of topological groups for some suitable section σ : C B of the
4275 projection to C.

Alternate proof. Any short exact sequence of free abelian groups splits, and the functors

A ÞÝÑ A bZ R{Z,
π1 pT, 1q ÑÝÞ T

4276 furnish an equivalence of categories between finitely generated free abelian groups and tori.

4277 We will also need to apply this principle to CGAs.

4278 Proposition A.4.3. Let F be a free k-CGA and π : A ÝÝ F a surjective k-CGA homomorphism. Then
4279 there exists a section i : F A of π.

4280 Proof. Suppose F is free on the graded k-module V. Since V is free as a graded module, there
4281 exists a section i : V A of π over V by Proposition A.4.1. As π is a ring homomorphism,
4282 the subalgebra A generated in A by iV projects back onto F under π. Were A1 not itself a free
1

4283 k-CGA, there would be some relation between homogeneous elements of A1 other than those
4284 ensured by the CGA axioms, and it would not be possible for π| A1 to be surjective, so there is no
4285 such relation. Thus π| A1 is a CGA isomorphism; now extend i to be its inverse.

4286 When we deal with principal bundles, the following simple proposition will be useful.

4287 Proposition A.4.4. Let 0 Ñ A ÝÑ B ÝÑ F Ñ 0 be a coexact sequence of k-CGA maps with F free and
4288 B of finite type. Suppose further that for each degree n we have rkk Bn “ rkk pA b Fqn . Then B – A b F.
k
 Appendix A. Algebraic background 162

4289 Proof. The projection B ÝÑ F splits by Proposition A.4.3, and together with B,
r the lift of Fr
4290 generates A as an algebra, so there is a commutative diagram

A> b F
`

A > F,

 ~
B

4291 of ring maps with the vertical map surjective. If this vertical map failed to also be injective, the
4292 rank assumption would fail, so it is an isomorphism.

4293 Here is an example application of such a splitting.

4294 Example A.4.5. We will show in Section 6.2 that the cohomology ring H ˚ pG; Rq of a compact,
4295 connected Lie group is an exterior algebra. If we compute this cohomology using the de Rham
4296 complex ΩpGq, an R-CDGA, and write Z˚ pGq “ ker d for the ring of closed forms, then Z˚ pGq is
4297 an R-CGA, so the natural projection Z ÝÝ H ˚ pG; Rq admits a section, and we can embed

H ˚ pGq Z˚ pGq ãÝÝÑ ΩpGq

4298 as a subalgebra.
4299 One can of course always find representative forms, but the ability to select them to form a
4300 subring on the nose, rather than up to homotopy, is rather special, and this circumstance is of
4301 sufficient utility that we here extract a formal statement of this behavior for later use (no pun
4302 intended).

4303 Definition A.4.6. A differential graded k-algebra pA, dq is said to be formal if there exists a zig-
4304 zag of k-DGA quasi-isomorphisms

pB1 , d1 q ¨¨¨ pA, dq

` x ˘ $ } " z
H ˚ pAq, 0 pB2 , d2 q pBn , dn q
` ˘
4305 connecting H ˚ pAq, 0 and pA, dq. A path-connected topological space X is said to be formal if
4306 there exists a formal Q-DGA with cohomology H ˚ pX; Qq.

4307 We will need later the following lemma, adapted from the material in Section 3.7 of Greub et
4308 al. [GHV76, pp. 147–152] regarding formal CDGAs.

4309 Lemma A.4.7. Suppose pBi , di qi“0n is a zig-zag of quasi-isomorphic k-CDGAs as depicted in Defini-
4310 tion A.4.6, that F “ SQ b ΛP is a free k-CGA on a strictly-positive graded subspace V “ Q ‘ P and
4311 that we are given a k-linear map ξ 0 : V ÝÑ ZpB0 q “ ker d0 increasing degree by one. Extend ξ 0 uniquely
4312 to a derivation on B0 b F vanishing on B0 and define a new derivation on B0 b F by δ0 “ ξ ` d0 . Then
4313 there exist k-CDGA structures pBi b F, δi q extending the δi such that the rings H ˚ pBi b Fq are isomorphic
„
4314 through isomorphisms which preserve the images H ˚ pB0 q ÝÑ H ˚ pBi q ÝÑ H ˚ pBi b Fq
 Appendix A. Algebraic background 163

4315 In particular, if pA, dq is formal, then there exists a k-CDGA structure on H ˚ pAq b F with isomorphic
4316 cohomology to that of pA b F, ξq and such that the triangle

H ˚ pAq

w ˘ &
„ / H ˚ pA b Fq
`
H ˚ H ˚ pAq b F

4317 commutes.

4318 Proof. To guarantee the second condition, that the quasi-isomorphisms to be defined among the
4319 pBi b F, δi q preserve the image of H ˚ pB0 q, we stipulate at the beginning all the quasi-isomorphisms
4320 we construct must restrict on the bases Bi to the original quasi-isomorphisms. Now inductively
4321 suppose the construction has been established up to B “ Bi and that the differential δ “ δi on
4322 B b F is a derivation of degree 1. Write C “ Bi`1 . There are two cases for the induction step,
4323 quasi-isomorphisms κ : pB, dq ÝÑ pC, dC q or λ : pB, dq ÐÝ pC, dC q.
4324 In the former case, using the assumed differential δ on B b F and the fact that κ is a cochain
4325 map, we extend
δ
B κ
V ÝÑ ZpBq ÝÑ ZpCq,
4326 uniquely to a derivation ξ C on C b F. Then δC :“ dC ` ξ C is again a derivation of degree one
4327 because δ is. The map κ b id : B b F ÝÑ C b F is a ring map because κ was, and a cochain map
4328 because it is so on generators.
4329 In the latter case, pick a homogeneous basis pvq of V. Since λ : pB, dq ÐÝ pC, dC q is a quasi-
4330 isomorphism, for each v there is a unique class in H ˚ pCq mapping onto rδvs P H ˚ pBq under
4331 H ˚ pλq, and we may choose an element ξ C v P ZpCq representing this class. Since λξ C v and δv are
4332 cohomologous, we can then write
` ˘
λξ C v “ δv ` dαpvq “ δ pv ` αpvq

4333 for some elements αpvq P B. These maps of the basis extend k-linearly to α : V ÝÑ B and
4334 ξ C : V ÝÑ ZpCq. Uniquely extend ξ C to a derivation on C b F and define δC “ dC ` ξ C . An
4335 extension ψ : C b F ÝÑ B b F of λ to a ring map is determined uniquely by its restriction to V,
4336 and for this extension to also be a cochain map pC b F, δC q ÝÑ pB b F, δq, it is necessary and
4337 sufficient to demand that for v P V one have
` ˘
δψpvq “ ψδC pvq “ ψ ξ C v “ λξ C v “ δ v ` αpvq .
l jh n

PC

4338 But we can achieve this by just letting ψpvq “ v ` αpvq on V.

4339 It remains to see κ b idF and ψ are quasi-isomorphisms; we do this for ψ, the other case
4340 being slightly simpler without the complication of α. Filter B b F and C b F “horizontally” with
4341 respect to the degree of B- and C-tensor components respectively. Then it is clear that both δ and
4342 δC increase filtration degree and that ψ preserves filtration degree since the filtration degrees of
4343 v and v ` αpvq are equal for v P V, so ψ induces a map of filtration spectral sequences. Since
4344 deg αpvq “ 1, the element v ` αpvq becomes just v in the associated graded algebra, so the map
4345 of E0 pages is just λ b id : C b F ÝÑ B b F. Since elements of the generating space V ď F are
4346 sent forward at least two degrees in the filtration by δC , we find E1 “ E0 in both sequences
 Appendix A. Algebraic background 164

4347 and the map of E2 pages is H ˚ pλq b id : H ˚ pBq b F ÝÑ H ˚ pCq b F, which by assumption is an
4348 isomorphism. By Proposition A.5.8 in the next subsection, then, ψ is a quasi-isomorphism.

4349 Example A.4.8. We will find in Section 6.6 that for a compact, connected Lie group G, the co-
4350 homology of its classifying space BG (see Appendix B.1.3 for the definition and Chapter 2 for
4351 the construction) is a symmetric algebra, hence a free CGA. There exists a Q-CDGA computing
4352 rational singular cohomology,
` called
˘ APL [FHT01, Ch. 10], and it follows there exists a Q-CDGA
4353
˚
map H pBGq ‚
Z APL pBGq witnessing the formality of BG.
4354 Now seems as good a place as any to define generalized symmetric spaces, which we will
4355 invoke occasionally later.

4356 Definition A.4.9. Let G be a connected Lie group and θ P Aut G a smooth automorphism of
4357 finite order. Then the fixed point set G θ is a closed subgroup of G. Let K be a subgroup of G θ
4358 containing its identity component pG θ q0 . Then G{K is called a generalized symmetric space. In
4359 the event θ is an involution, G{K is a symmetric space. If in addition G and K are compact and
4360 connected, we call pG, Kq a (generalized) symmetric pair.

4361 The generalized symmetric spaces and especially the symmetric spaces per se form a com-
4362 pletely classified system of examples which have been intensively studied since the early 1900s.
4363 Example A.4.10. Élie Cartan demonstrated that symmetric spaces G{K are formal, in fact showing
4364 that the collection of harmonic forms on a symmetric space forms a subring of the differential
4365 forms ΩpG{Kq consisting of one element from each class in H ˚ pG{Kq. We will produce a version
4366 of this proof in Proposition 7.4.11.
4367 Svjetlana Terzić [Ter01] and independently Zofia Stepie
˛ ń [Ste02] have also shown that compact
4368 generalized symmetric spaces G{K with isotropy group K connected are formal. It is not, however,
4369 the case that wedge products of harmonic forms on such spaces are again harmonic (that such
4370 should happen is called geometric formality); see Terzić’s later joint article with Dieter Kotschick
4371 [KT03].

4372 Filtrations and spectral sequences

4373 This section comprises a few words, without proofs, about the spectral sequence associated to a
4374 filtered differential group.

4375 Filtered differential groups

4376 Consider the total order Z as a category in the opposite of the standard way, so that there exists
4377 a (unique) arrow n Ñ m just when n ě m, and write SubpAq for the category of k-submodules of
4378 A and inclusions therebetween.
4379 A filtered group is a pair pA, F‚ q, where A is a k-module and F‚ : Z Ñ SubpAq is a functor with
4380 F0 “ A. In other words, this is an infinite descending sequence

¨ ¨ ¨ “ F´1 “ F0 “ A ě F1 ě F2 ě ¨ ¨ ¨ .
 Appendix A. Algebraic background 165

4381 We also write Fn “ Fn A.3 One can recompile this information into a Z-graded group
à à
F‚ A :“ Fr A
rPZ
`À ˘
4382 and the inclusions Fn`1 A ãÝÝÑ Fn A into an injective endomorphism i P Endk F‚ A of degree
4383 ´1 which stabilizes to an isomorphism in nonpositive degrees. Viewing i as a direct system
4384 and taking the direct limit recovers the original filtered group, so we may regard pA, F‚ q and
À
4385 p F‚ A, iq as equivalent objects and denote both, slightly abusively, by pA, iq. Say a filtration is
Ş
4386 complete if pPZ Fp A “ 0, and finite if Fp A “ 0 for p sufficiently large. The group
à
gr‚ A :“ coker i‚ “ Fr A{Fr`1 A
rě0

4387 is the associated graded group of pA, iq. A map f : A ÝÑ B is said to preserve filtrations pA, iq and
4388 pB, ιq if f pFp Aq ď Fp B (or equivalently f ˝ i “ ι ˝ f ). We write such a map as f : pA, iq ÝÑ pB, ιq.
4389 Such a map induces an associated graded map gr‚ f : gr‚ A ÝÑ gr‚ B. We have the following
4390 recurring result on such maps.

4391 Proposition A.5.1. Let f : pA, iq ÝÑ pB, ιq be a filtration-preserving cochain map of filtered groups.
4392 Suppose that both filtrations are finite. Then if gr‚ f is an isomorphism, so also must be f itself.

4393 Proof. Fix a filtration degree p sufficiently large that Fp`1 A “ 0 “ Fp`1 B. We have a map

0 / Fp`1 A / Fp A / gr A /0
p
„

f f gr‚ f
  
0 / Fp`1 B / Fp B / gr B /0
p

4394 of short exact sequences; by the five lemma, it follows Fp f : Fp A ÝÑ Fp B is an isomorphism. This
„
4395 begins a decreasing induction on p, which terminates in f : A ÝÑ B when p “ 0.

We can also define filtered graded groups pA‚ , iq. These are simply direct sums A‚ “ nPZ An
À
4396

of filtered groups pAn , in q is a filtered group equipped with the filtration Fp A‚ “ n Fp An . Such
À
4397

4398 an object is said to be finite in each degree if each filtration F‚ An is finite. For maps of graded
4399 filtered groups, applying Proposition A.5.1 individually in each degree, one finds the following.

4400 Corollary A.5.2. Let f : pA‚ , iq ÝÑ pB‚ , ιq be a filtration-preserving cochain map of filtered graded
4401 groups. Suppose that both filtrations are finite in each degree. Then if gr‚ f is an isomorphism, so also
4402 must be f itself.

4403 A filtered differential group is a triple pA, d, iq such that pA, dq is a differential group, pA, iq a
4404 filtered group, and d preserves the filtration, inducing restricted differentials Fn d P Endk Fn A. (It
À À
4405 is also true that then i is a cochain map p F‚ A, dq Ñ p F‚ A, dq.) A homomorphism of filtered
4406 differential groups is a cochain map commuting with the filtration.
3 Ť
In general usage, filtrations pFn Aq are not required to stabilize in negative degrees or to have nPZ Fn A “ A, and
Ť
filtrations such that n Fn A “ A are called exhaustive. Since we will never have cause to use such a general filtration,
we include these more restrictive hypotheses in our definition off the bat.
 Appendix A. Algebraic background 166

4407 Given a filtered differential group pA, d,˜ ı̃q, the differential d˜ descends to a differential d¯ on
4408 gr‚ A, inducing a short exact sequence of differential groups

ı̃ ̃
0 Ñ A ÝÑ A ÝÝ gr‚ A Ñ 0.

4409 This induces a triangular exact sequence

HpAq
i / HpAq
Z

k j

E

4410 of cohomology groups, where E “ Hd¯pgr‚ Aq. Such a triangle is traditionally called an exact
4411 couple. Here i and j are induced from ı̃ and ̃ respectively, and k arises from the snake lemma
4412 and takes ra ` ı̃As ÞÑ rı̃´1 das. If we set d “ jk, then d2 “ jpkjqk “ 0, so d is a differential on E. The
4413 original filtration ı̃ descends to a filtration Fn H ˚ pAq “ in H ˚ pAq “ Hd˚˜ pim ı̃n q.
4414 Note that a map of filtered differential groups induces a map of short exact sequences and a
4415 map of exact couples in cohomology (a triangular prism), and so in particular a map of differen-
4416 tial groups on the E components. We have deg i “ deg j “ 0 and deg k “ 1.

4417 The spectral sequence of an exact couple

4418 There is a functor
i1
A0Y
i / A0 A1Y / A1

ùñ
k j k1 j1
 
E0 , E1 ,
4419 taking an exact couple to the derived couple whose objects are A1 “ iA0 and E1 “ HpE0 , d0 q, and
4420 whose maps are given by i1 “ pi æ iA0 q and j1 : ia ÞÝÑ rjas and k1 : res ÞÝÑ ke.
4421 This process can be iterated, and the sequence pEr , dr q of differential groups so derived is
4422 called the spectral sequence of the exact couple. Each Er is traditionally called a page. Note
that pA0 , iq is a filtered group with Fr A0 “ Ar “ ir A0 . Write A8 for the intersection rě0 Ar “
Ş
4423
r r
Ş
4424
rě0 i A0 . If for each a P A0 , the sequence pi aq eventually stabilizes, then i8 “ i æ A8 is injective,
4425 and fits into a short exact sequence

i8
A8Z / A8

0 j8

E8 ;

4426 that is, E8 “ coker i8 , and we may think as this E8 as the limiting behavior of the pages pEr q
4427 and call it the limiting page. If Er – E8 for some finite r, we say the sequence collapses at Er .
4428 To see how E8 arises, recall that Er`1 “ pker dr q{pim dr q, where ker dr and im dr are in
4429 turn subgroups of Er “ pker dr´1 q{pim dr´1 q. Thus, under the quotient map π : ker dr´1 ÝÝ
 Appendix A. Algebraic background 167

4430 Er , the groups ker dr and im dr lift to subgroups π ´1 im dr ď π ´1 ker dr of ker dr´1 , each con-
4431 taining im dr´1 . By the third isomorphism theorem, we still have pπ ´1 ker dr q{pπ ´1 im dr q –
4432 pker dr q{pim dr q “ Er`1 . Iteratively pulling back all the way to E0 , we get a sequence of subgroups

B0 ď B1 ď B2 ď B3 ď ¨ ¨ ¨ ď Z3 ď Z2 ď Z1 ď Z0
Ş Ť
4433 of E0 such that Er`1 “ Zr {Br . Now we can unambiguously declare Z8 :“ Zr and B8 :“ Br ,
4434 and take E8 “ Z8 {B8 .
rr , d˜r q ÝÑ pEr , dr q
` ˘
4435 A homomorphism of spectral sequences is a sequence ψr : pE rěn
of cochain
4436 maps of differential groups such that each ψr`1 for r ě n is induced by ψr , which is to say
4437 ψr`1 “ H ˚ pψr q. From the remark at the end of Appendix A.5.1, it follows that a map of filtered
4438 differential groups induces a map of exact couples and iteratively a map of spectral sequences.

4439 The spectral sequence of a filtered differential graded algebra

4440 Now let us consider a filtered cochain complex pA‚ , ı̃, dq. This is an object which is simultaneously
4441 a cochain complex pA‚ , dq, a filtered differential group, and a filtered graded group. Explicitly,
4442 one has nilsquare module homomorphisms d : An ÝÑ An`1 , and filtration F‚ An of each graded
4443 component such that d preserves each filtrand Fp A‚ .
4444 In the resulting exact couple HpAq Ñ HpAq Ñ HpEq, one has i : H n pAq ÝÑ H n pAq and
4445 j : H n pAq ÝÑ H n pEq of degree zero, but k : H n pEq ÝÑ H n`1 pEq of degree 1.

4446 Theorem A.5.3 (Koszul). Let pA, d, iq be a filtered chain complex. Then in the spectral sequence associated
i
4447 to the exact couple 0 Ñ A ÝÑ A ÝÑ gr A Ñ 0, one has

4448 • pE0 , d0 q “ pgr‚ A, gr‚ dq,

4449 • E1 – H ˚ pgr‚ Aq,

4450 • E8 – gr‚ H ˚ pAq.

4451 We call this the filtration spectral sequence of the filtered chain complex pA, d, iq. All pages inherit a
p,q p,q
4452 bigrading Er induced from the grading E0 “ gr p Aq of E0 “ gr‚ A, and the differential dr is of bidegree
p,q
4453 pr, 1 ´ rq. This is first-quadrant spectral sequence in that Er “ 0 if p ă 0 or q ă 0.

4454 Given a bigraded differential algebra A, a natural decreasing filtration, the horizontal filtra-
4455 tion, is given by à i,‚
Fp A :“ A .
iěp

If in the decomposition d “ `PZ d` of Appendix A.3.2 one has d` “ 0 for ` ă 0, then d is

ř
4456

4457 filtration-preserving, and we call pA, d, iq a filtered differential bigraded algebra. In this case,
4458 the theorem applied to pA, d, iq yields a spectral sequence pEr , dr q with E0 – A again and the
4459 differentials dr induced from the components dr .

4460 Corollary A.5.4. Let pA, d, iq be a filtered, nonnegatively-bigraded DGA. Then in the spectral sequence
4461 associated to the horizontal filtration one has

4462 • pE0 , d0 q – pA, d0 q,

• E1 – pPN H ˚ pA p,‚ , d0 q, d1 “ Hd˚0 pd1 q,

À
4463
 Appendix A. Algebraic background 168

4464 • E2 – Hd˚1 Hd˚0 pAq,

4465 • E8 – gr‚ H ˚ pAq.

4466 In one key situation in which we apply this spectral sequence, we are able to say even more
4467 about E2 .

4468 Corollary A.5.5. If pA, d, iq is a filtered, nonnegatively-bigraded DGA such that A “ A‚,‚ is a tensor
4469 product A‚,0 b A0,‚ , and i is the horizontal filtration, then d0 is induced by a map A0,‚ ÝÑ A0,‚ and d1
4470 by a map A‚,0 ÝÑ A‚,0 , and we have

4471 • E0 – A, d0 “ id b d0 ,

4472 • E1 – A‚,0 b Hd˚0 pA0,‚ q, d1 “ d1 b id,

4473 • E2 – Hd˚1 pA‚,0 q b Hd˚0 pA0,‚ q,

4474 • E8 – gr‚ H ˚ pAq.

4475 In Section 3.3, after we have discussed fiber bundles, we will introduce the Serre spectral
4476 sequence, which will have also have a tensor structure on E2 , for more topological reasons.
4477 Remark A.5.6. The algebraic Künneth Theorem A.3.9 of this chapter and the universal coefficient
4478 Theorem B.2.1 of the next both generalize to filtration spectral sequences if we do not assume
4479 that the modules in question are not free over the base ring k or that k is not a principal ideal
4480 domain.

4481 Fundamental results on spectral sequences

4482 A common way to understand the cohomology ring of a filtered DGA is to engage in wishful
4483 thinking: one finds another spectral sequence one would like to approximate that of the DGA in
4484 question, contrives a map between the idealized sequence and the actual sequence, and shows it
4485 yields an isomorphism on a late enough page. The theoretical justification behind this chicanery
4486 has at most two steps.

4487 Theorem A.5.7 (Zeeman–Moore comparison theorem, [McC01, Thm. 3.26, p. 82]). Let pψr q : p1Er , 1dr q ÝÑ
4488 pEr , dr q be a map of bigraded spectral sequences of k-modules such that E2 – E2‚,0 b E20,‚ and 1E2 –
4489
1E‚,0 b 1E0,‚ decompose as tensor products. Then any two of the following three conditions imply the third:
2 2

4490 • ψ2‚,0 : E2‚,0 ÝÑ 1E‚,0

2 is an isomorphism,

4491 • ψ20,‚ : E20,‚ ÝÑ 1E0,‚

2 is an isomorphism,

4492 • ψ8 : E8 ÝÑ 1E8 is an isomorphism.

4493 Given an isomorphism of E2 pages or E8 pages then shows that the inducing map of DGAs
4494 was a quasi-isomorphism.

4495 Proposition A.5.8. Let f : A ÝÑ B be a map of filtered DGAs and pψr q : p1Er , 1dr q ÝÑ pEr , dr q the
4496 associated map of filtration spectral sequences. Suppose that both filtrations are finite in each degree (as
4497 defined in Appendix A.5.1) If ψr is an isomorphism for any r ě 0, then f ˚ : H ˚ pAq ÝÑ H ˚ pBq is an
4498 isomorphism.
 Appendix A. Algebraic background 169

4499 Proof. If any ψr is an isomorphism, then since 1dr ψr “ ψr dr , it follows that all later ψr and ψ8 are
4500 isomorphisms. By Corollary A.5.4, ψ8 is the isomorphism gr‚ f ˚ : gr‚ H ˚ pAq ÝÑ gr‚ H ˚ pBq. For
4501 any given total degree n, we can apply Corollary A.5.2 to the map ψ8 n : gr H n pAq ÝÑ gr H n pBq
‚ ‚
4502 to conclude H n p f q is an isomorphism.

4503 A much simpler fact that comes up frequently is the following:

4504 Proposition A.5.9. Let pEr , dr q be a first-quadrant spectral sequence. If p ă r and q ă r ´ 1, then
p,q p,q
4505 Er “ E8 .
p´r,q`r´1
4506 Proof. Because the bidegree of dr is pr, 1 ´ rq, the domain Er of the component of dr with
p,q p`r,q`1´r
4507 codomain Er lies in the second quadrant, and the codomain Er of the component of
p,q
4508 dr with domain Er lies in the fourth quadrant. See Figure A.5.10. Since these quadrants are
p,q p,q p,q
4509 inhabited only by zero groups, the differentials in and out of Er are zero, so Er “ Er`1 . All
4510 later differentials out of this square must also be zero for the same reason.

Figure A.5.10: The differentials to and from E54,3 leave the first quadrant

7
6
5
4
3 E54,3
2
1
0
0 1 2 3 4 5 6 7 8 9

4511 This genre of reasoning, that something must stabilize at a certain page—or vanish before
4512 a certain page, lest it survive to E8 —goes by the trade name of “lacunary considerations.”
4513 One uses such considerations as frequently as possible because they are usually far simpler
4514 than actually computing differentials. Occasionally this kind of spatial reasoning enables one to
4515 understand what happens in a spectral sequence without having done any algebra at all.

4516 Proposition A.5.11 ([McC01, Example 1.K, p. 25]). Let pA, dq be a filtered differential bigraded k-
4517 algebra, free as a k-module. If E8 is a free k-CGA, then E8 – H ˚ pA, dq as a k-CGA.

4518 This is analogous to the statement of Proposition A.4.3, except here the “quotient” object
4519 is actually the associated graded algebra, and there is not a priori an algebra map H ˚ pAq ÝÑ
4520 gr‚ H ˚ pAq “ E8 .

4521 Proof. Consider the indecomposables QpE8 q as a bigraded k-submodule of algebra generators
4522 of E8 . Because E8 “ gr‚ H ˚ pAq and H ˚ pAq are isomorphic on the level of bigraded k-modules,
4523 we can map QpE8 q to H ˚ pAq in such a way as to preserve the bidegree. This yields a bigraded
4524 module injection QpE8 q H ˚ pAq. Since E8 is the free CGA on QpE8 q, this injection extends
 Appendix A. Algebraic background 170

4525 uniquely to the filtration-preserving CGA map f : E8 ÝÑ H ˚ pAq in the diagram below (the
4526 bottom row will be explained).
QpE 8 q
_ $
 $
E8 / H ˚ pAq
f

„
 
E8 / E8 .
gr‚ f

4527 Since f lifts generators of gr‚ E8 to H ˚ pAq in a filtration-preserving manner and gr‚ E8 “ E8 ,
4528 the map gr‚ f : E8 ÝÑ E8 is the identity on QpE8 q by the identity map. Since gr‚ f is a ring
4529 map, it is then the identity map on E8 (and a fortiori an isomorphism) so an application of
4530 Corollary A.5.2 shows the original f : E8 ÝÑ H ˚ pAq is a CGA isomorphism.

4531 The transgression

4532 Early on in the history of bigraded spectral sequences of the form discussed above, it was noticed
4533 that the edge maps dr : Er0,r´1 ÝÑ Err,0 from the left column to the bottom row have a special
4534 importance.

Figure A.5.12: An edge map

5 E60,5
4
3 d6
2
1
0 E66,0
0 1 2 3 4 5 6

4535 Definition A.5.13 (Koszul, 1950 [Kos50, Sec. 18]). Let pEr , dr q be the filtration spectral sequence
4536 of a filtered DGA pA, dq. If z P E20,r´1 is in the kernel of each d p for p ă r, so that dr z P Er0,r is
4537 defined (that is, if z survives long enough to be in the domain of an edge homomorphism), then
4538 z is said to transgress. The dotted map τ in the diagram

Er0,r´1  / E0,r´1
2
dr

τ
u / / Err,0 ,
E2r,0

4539 viewed as a k-linear map from a submodule of E20,r´1 to a quotient module of E2r,0 , is the trans-
4540 gression.
 Appendix A. Algebraic background 171

4541 Proposition A.5.14. Let pEr , dr q be the filtration spectral sequence of a filtered DGA pA, dq. An element
4542 z P E20,r´1 transgresses to τz P E2r,0 if and only if there exists y P A which represents z in gr0 A “
4543 E20,r´1 “ H r´1 pgr0 A, d0 q and such that dy P A represents τz in E2r,0 “ H 0 pgrr A, d0 q.

4544 Historical remarks A.5.15. According to the concluding notes in Greub et al. [GHV76], instances of
4545 transgressions were first identified by Shiing-Shen Chern [Che46] and Guy Hirsch [Hir48] before
4546 Koszul observed the pattern and coined the term “transgression” in his thesis work.
4547 The filtration spectral sequence is first described in Koszul’s Comptes Rendus note [Kos47a],
4548 and is extracted from Leray’s earlier work as described in a 1946 Comptes Rendus notice [Ler46a].
4549 Koszul was the first other person to work through and understand Leray’s post-war topological
4550 output, and was the chief instigator of the simplifications that made spectral sequences accessible
4551 to the rest of the mathematical community [Mil00].
4552 Appendix B
4553 Topological background

4554 In this appendix we define some relevant topological categories and state some well-known re-
4555 sults in algebraic topology and Lie theory. We will take homotopy groups and singular homology
4556 and cohomology groups as known concepts, and cite basic results in algebraic topology without
4557 proof, but will restate that the 0th homotopy set π0 X of a space X is its set of path-components,
4558 which inherits a group structure if X is a group. We denote homotopy equivalences by “»,”
4559 homeomorphisms by “«,” and Lie group isomorphisms by “–.” If a group G acts on a space X
˝
4560 via φ : G ˆ X ÝÑ X, we write φ : G ñ X. The interior of a manifold M with boundary is M and
4561 its boundary is BM. The complement of a set A Ď B is BzA.

4562 Topological structures of interest

4563 Let Top be the category whose objects are pairs pX, Aq of topological spaces, A closed in X,
4564 with morphisms pX, Aq ÝÑ pY, Bq those continuous maps f : X ÝÑ Y such that f pAq Ď B. The
4565 category whose objects are individual topological spaces and morphisms continuous maps is
4566 included as a full category through the inclusion X ÞÝÑ pX, ∅q, where ∅ is the empty space.

4567 Cell complexes

4568 A CW complex is a topological space X equipped with a decomposition into a union of disks
4569 of increasing dimension. Less elliptically, such an X must admit a filtration pX n q into n-skeleta
4570 meeting the following conditions:

4571 • The 0-skeleton X 0 is a discrete space.

4572 • Given the n-skeleton X n , index a collection of distinct pn ` 1q-disks as pDαn`1 qαPA . From each
4573 boundary Sαn , let a continuous map ϕα : Sαn ÝÑ X n , the attaching map, be given. These maps
n n n`1 is defined to be the quotient space
š
4574 assemble into a map ϕ : αPA Sα ÝÑ X , and X
N
X n > αPA Dαn`1
š
s „ ϕpsq

4575 of the disjoint union: we’ve identified the boundaries of the Dαn`1 with their images in X n .

• We let X “ nPN X n be the colimit, with the direct limit topology. This amounts to saying
Ť
4576

4577 U Ď X is open just if each U X X n is open in X n .

172
 Appendix B. Topological background 173

4578 A map f : X ÝÑ Y between two CW complexes is said to be cellular if it respects the skeleta:
4579 f : X n ÝÑ Y n for all n.
4580 Write CW for the subcategory of Top whose objects are CW pairs consisting of a CW complex
4581 X and closed subcomplex A, and whose maps pX, Aq ÝÑ pY, Bq are required to be cellular,
4582 meaning both X ÝÑ Y and the restriction A ÝÑ B are cellular. The category CW is a homotopy-
4583 theoretic skeleton of Top in the sense that given any pX, Aq P Top there exists pX,
r Aq
r P CW and
4584 a weak homotopy equivalence pX, r Aq
r ÝÑ pX, Aq in Top. This map (or pX, Aq itself) is called a
4585 CW approximation [Hat02, Example 4.15, p. 353]. Moreover, any map of pairs is the same up to
4586 homotopy as a map between CW complexes: given a map pX, Aq ÝÑ pY, Bq of pairs there exists
4587 a map between CW approximations making the following square commute up to homotopy:

pX,
r Aq
r / pX, Aq

 
pY,
r Bq
r / pY, Bq.

4588 Although CW is unstable under the formation of mapping spaces, with judicious use of CW
4589 approximations, we may basically assume every space that follows is a CW complex.

4590 Fiber bundles

4591 A fiber space with is a continuous surjection E ÝÑ B such that for each b P B, we have h´1 tbu « F
4592 for some fixed space F, the fiber. Each h´1 tbu is also called a fiber, E is the total space, and B
4593 the base. We abbreviate this assemblage as F Ñ E Ñ B. Given two fiber spaces p : E Ñ B and
4594 p1 : E1 Ñ B1 , a map h : E ÝÑ E1 of total spaces is fiber-preserving if it sends fibers into fibers.
4595 Equivalently, there is a map h̄ of bases making the following diagram commute:

E
h / E1

p p1
 
B
h̄ / B1 .
(
4596 Then hp´1 tbu Ď pp1 q´1 h̄pbq for all b P B. Fiber spaces with fiber F and fiber-preserving maps
4597 form a category whose isomorphisms are fiber-preserving homeomorphisms.
4598 A fiber space p : E Ñ B with fiber F is a fiber bundle, or an F-bundle (or locally trivial), if

4599 • the base B admits an open cover of sets U such that there are fiber space isomorphisms
«
4600 φU : p´1 pUq ÝÑ U ˆ F, called (local) trivializations, and

• these trivializations are compatible in the sense that given two overlapping trivializing
opens U and V, the transition functions gU,V defined by the composite homeomorphism

V φ´1 φU
φU,V : pU X Vq ˆ F ÝÑ p´1 pU X Vq ÝÑ pU X Vq ˆ F
` ˘
px, f q ÞÝÑ x, gU,V pxqp f q ,

4601 are continuous maps U X V ÝÑ Homeo F. Morally, different coordinatizations of the same
4602 trivial subbundle should differ continuously.
 Appendix B. Topological background 174

p
4603 Given a fiber space F Ñ E Ñ B and an subset U Ď B, the restriction E|U is the F-bundle
4604 pp æ Uq : p´1 pUq Ñ U. This generalizes to the following construction. Given a continuous map
4605 h : X ÝÑ B (for restrictions, an inclusion), we can construct a pullback space h˚ E Ñ X with fiber
4606 F fitting into the commutative square

h “ pr2
/E
r
h˚ E
h˚ p :“ pr1 p
 
X
h / B,

4607 where the new total space is

(
f ˚ E “ X ˆ E :“ px, eq P X ˆ E : hpxq “ ppeq Ĺ X ˆ E
B

4608 and the new maps the restrictions of the factor projections from X ˆ E. This total space is called
4609 the fiber product, and (with the maps), it is the pullback of the diagram X Ñ B Ð E in Top.1 If
4610 E Ñ B was an F-bundle, so also is h˚ E Ñ X: given a local trivialization
«
φ “ pp, ρq : p´1 U ÝÑ U ˆ F,

4611 a trivialization of the pullback ph˚ Eq|h´1 pUq is given by

` ˘
idX ˆ ρ : px, eq ÞÝÑ x, ρpeq ,

4612 and such sets h´1 pUq cover X. The resulting bundle is a pullback bundle.
4613 If F, E, B are all smooth manifolds and the fiber inclusion, projection, and transition functions
4614 are all C8 , we say F Ñ E Ñ B is a smooth bundle. One can similarly define holomorphic and
4615 algebraic bundles, but smooth and merely continuous bundles are all we shall work with.

4616 Principal bundles

4617 Suppose we are given a fiber bundle F Ñ E Ñ B admitting trivializations pφU qUPU , such that
4618 each transition function gU,V takes values in some subgroup G of the group Homeo F of self-
4619 homeomorphisms of the fiber. As G is a topological group, its multiplication is continuous, and
4620 left multiplication ` g by any element of g P G is a self-homeomorphism of G. In this way the
4621 transition function values gU,V pxq P G can be viewed as elements of Homeo G, and we can form
š
4622 a G-bundle G Ñ P Ñ B by starting with the disjoint union UPU U ˆ G and gluing the pieces
4623 by the relations ` ˘
px, gq „ x, gU,V pxq ¨ g
4624 for all nonempty intersections U X V of sets in U and all x P U X V and g P G.
4625 The disjoint union we started with admits a global right G-action pu, gq ¨ g1 “ pu, gg1 q, which
4626 descends to a right G-action on P since the transition functions act on the left of the fibers G.
4627 This right action is simply transitive on each fiber. We call a G-bundle admitting a right G-action
4628 acting simply transitively on each fiber a principal G-bundle; this motivating bundle G Ñ P Ñ B
4629 is one such.
1 This notation X ˆ E, now universal, is due to Paul Baum [Smi67, p. 68].
B
 Appendix B. Topological background 175

4630 We can recover the original F Ñ E Ñ B from G Ñ P Ñ B and the map ψ : G ÝÑ Homeo F by
4631 a pushout construction:
š
PˆF UˆGˆF
E « P ˆ F :“ ` ˘ ` ˘«` ˘UPU ` ˘ (B.1)
G rx, gs, f „ rx, 1sψpgq f x, gU,V pxqg, f „ px, g, f q „ x, 1, ψpgq f

4632 Verbally, this can be seen as extracting the G-valued transition functions from a principal G-
4633 bundle and applying them to F instead of G. For this reason, the bundles E Ñ B and P Ñ B are
4634 said to be associated. Because this correspondence is reversible, principal G-bundles essentially
4635 carry all information about fiber bundles.
4636 Further, in Chapter 2, we construct a universal principal G-bundle EG Ñ BG that every principal
4637 G-bundle is a pullback of. Given such a bundle, a space F, and a homomorphism ψ : G ÝÑ
4638 Homeo F, it follows the the associated F-bundle EG ˆG F Ñ BG is universal for F-bundles with
4639 transition functions in ψpGq; for example, EGLpn, Rq ˆ Rn Ñ BGLpn, Rq is a universal vector
GLpn,Rq
4640 bundle.

4641 Algebraic topology grab bag

4642 This section is just a collection of useful algebro-topological results we will need later, presented
4643 without much in the way of motivation.
4644 The algebraic Künneth Theorem A.3.9 has at least two major topological repercussions.

4645 Theorem B.2.1 (Universal coefficients [Hat02, Thms. 3.2, 3.A.3, pp. 195, 264]). Let X be a topological
4646 space and k a principal ideal domain. For each n P N one has the following short exact sequences of abelian
4647 groups:
0 Ñ Hn pX; Zq b k ÝÑ Hn pX; kq ÝÑ Tor1k Hn´1 pX; Zq, k Ñ 0,
` ˘
Z
4648
Ext1k Hn´1 pX; Zq, k ÝÑ H n pX; kq ÝÑ HomZ Hn pX; Zq, k Ñ 0.
` ˘ ` ˘
0Ñ

4649 Proof. The homology sequence follows from Theorem A.3.9 by taking C “ C0 “ k and A “ C‚ pXq
4650 the singular chain complex, taking into account the differentials go in the opposite
` direction
˘
4651 expected. The cohomology sequence arises from `taking C ˘“ k and A “ HomZ C‚ pXq, Z the
4652 singular cochain complex, noting A bZ k – HomZ C‚ pXq, k .

Theorem B.2.2 (Topological Künneth [Hat02, Thm. 3B.6, 3.21][Mas91, Thm. 11.2, p. 346]). Let X
and Y be topological spaces and k an abelian group. Suppose H ˚ pXq is of finite type. Then for each n P N
one has the following split short exact sequences of abelian groups:

Tor1k Hj pX; kq, Hn´j´1 pY; kq Ñ 0;

à ` ˘ à ` ˘
0Ñ Hj pXq b Hn´j pYq ÝÑHn pX ˆ Y; kq ÝÑ
0ďjďn 0ďjďn

H pX; Zq b H pY; kq ÝÑH n pX ˆ Y; kq ÝÑ

j n
à ` ˘ à ` j n`1´j
˘
0Ñ TorZ
1 H pX; Zq, H pY; kq Ñ 0.
0ďjďn 0ďjďn`1

4653 When one of the rings H ˚ pX; kq or H ˚ pY; kq is free as a k-module, the Ext and Tor terms
4654 disappear and these isomorphisms assume a product form

H ˚ pX ˆ Yq – H ˚ X b H ˚ Y.

4655 One also obtains the following relation between integral homology and cohomology.
 Appendix B. Topological background 176

4656 Proposition B.2.3. Let X be a topological space. The torsion subgroups and torsion-free quotients of the
4657 singular homology and cohomology groups H˚ pX; Zq and H ˚ pX; Zq satisfy

H n pX; Zq – Hn pX; Zqfree ‘ Hn´1 pX; Zqtors

4658 We will use fiber bundles frequently, and need a criterion for determining when the funda-
4659 mental groups of their base spaces are trivial.

4660 Theorem B.2.4 ([Hat02, Thm. 4.3]). Let F Ñ E Ñ B be a fiber bundle. Then there is associated a long
4661 exact sequence of homotopy groups

¨ ¨ ¨ ÝÑ π2 pFq ÝÑ π2 pEq ÝÑ π2 pBq ÝÑ π1 pFq ÝÑ π1 pEq ÝÑ π1 pBq ÝÑ π0 pFq ÝÑ π0 pEq Ñ 0.

4662 There are important but subtle relations between the homology and homotopy groups.

4663 Proposition B.2.5. The first singular homology group of a space X is the abelianization of its fundamental
4664 group: H1 pX; Zq – π1 pXqab .

Theorem B.2.6. Let X be a simply-connected topological space and let n ą 1 be the least natural number
such that πn X is nontrivial. Then the same n is also minimal such that Hn X is nontrivial, and the natural
Hurewicz map

πn X ÝÑ Hn X,
rσ : S ÝÑ Xs ÞÝÑ σ˚ rSn s,
n

4665 taking the homotopy class of a map from a sphere to the pushforward of the fundamental class, is an
4666 isomorphism.

4667 The homotopy groups completely determine homotopy type in the following sense.

4668 Theorem B.2.7 (Whitehead [Hat02, Thm. 4.5, p. 346]). Let f : X ÝÑ Y be a map of CW complexes
„
4669 such that πn f : πn X ÝÑ πn Y is an isomorphism for all n ě 0 (a weak homotopy equivalence). Then
4670 f is a homotopy equivalence.

4671 Theorem B.2.8 (Whitehead [Hat02, Thm. 4.21, p. 356]). ] Let f : X ÝÑ Y be a weak homotopy
„
4672 equivalence of topological spaces. Then H n f : H n Y ÝÑ H n X is an isomorphism for all n.

4673 We will also need the Lefschetz fixed point theorem. Note that if X is of finite type, the natural
4674 maps H n pX; Zq  H n pX; Zqfree  H n pX; Qq carry a Z-basis of the free Z-module H n pX; Zqfree to
4675 a Q-basis of H n pX; Qq.

4676 Definition B.2.9. Let f : X Ñ X be a continuous self-map of a topological space X of finite

4677 type. Then associated endomorphisms H n p f q P AutQ H n pX; Qq are defined for each n ě 0. The
4678 Lefschetz number ÿ
χp f q :“ p´1qn tr H n p f q
ně0

4679 is the alternating sum of these traces, where each trace is taken with respect to a basis of H n pX; Qq
4680 inherited from H n pX; Zqfree .

4681 Since the trace of the identity map of a vector space is just the dimension of that space and
4682 H n pidX q “ id H n pX;Qq one immediately has the following.
 Appendix B. Topological background 177

4683 Proposition B.2.10. Let f : X ÝÑ X be a continuous self-map of a topological space of finite type. Then
4684 the Lefschetz number of the identity map idX is the Euler characteristic of X:

χpXq “ χpidX q.

4685 The more interesting fact about the Lefschetz number is the Lefschetz fixed point theorem.

4686 Theorem B.2.11 (Lefschetz, [Hat02, Thm. 2C.3, p. 179]). Let X be a topological space which is a
4687 deformation retract of a simplicial complex and f : X ÝÑ X a continuous map without fixed points. Then
4688 the Lefschetz number χp f q is 0.

4689 Covers and transfer isomorphisms

4690 In this section, we leverage a standard result on the cohomology of covers to a statement we use
4691 later about the cohomology of homogeneous spaces.

4692 Proposition B.3.1 ([Hat02, Prop. 3G.1]). Let F be a finite group acting by homeomorphisms on a space
4693 X, so that p : X ÝÑ X{F is a finite covering. Suppose |F| is invertible in k. Then the map

p˚ : H ˚ pX{F; kq ÝÑ H ˚ pX; kq

4694 is an injection with image the invariant subring H ˚ pX; kqF .

4695 Proof. Since simplices ∆n are simply-connected, each singular simplex σ : ∆n ÝÑ X{F lifts to a
σ : ∆n ÝÑ X. The map τ : σ ÞÝÑ f PF f ˝ r
ř
4696 singular simplex r σ summing over all such lifts then
4697 induces a transfer map τ : Cn pX{Fq ÝÑ Cn pXq of singular chain groups. For each lift f r σ we
4698
˚
have pp f σq “ σ, so p ˝ τ “ |F| ¨ id on Cn pX{Fq. Dualizing yields a cochain map τ : C pX; kq ÝÑ
r n

4699 C n pX{F; kq such that τ ˚ ˝ p˚ “ |F| ¨ id on C n pX; kq, so the same holds in H ˚ pX; kq.
4700 If we demand |F| be a unit in k, then τ ˚ ˝ p˚ is an isomorphism, so p˚ is injective. Since
4701 p ˝ f “ p for all f P F, it follows im p˚ is contained in the invariant subring H ˚ pX; kqF . On the
ř ř
4702 other hand, since τ ˝ p sends r σ ÞÝÑ f PF f ˝ r σ, it follows p˚ τ ˚ α “ f PF f ˚ α for all α P H ˚ pX; kq.
4703 In particular, if α P H ˚ pX; kqF is F-invariant, then p˚ τ ˚ α “ |F|α, so p˚ surjects onto HpX; kqF .

4704 Corollary B.3.2. In the situation of Proposition B.3.1, suppose the action of F on X is the restriction of a
4705 continuous action of a path-connected group Γ on X. Then

H ˚ pX{F; kq – H ˚ pX; kq

4706 Proof. Let f P F. Since Γ is path-connected, the left translation γ ÞÝÑ f γ on Γ is homotopic to the
4707 identity. It follows f acts trivially on H ˚ X. Thus pH ˚ XqF – H ˚ X for any coefficient ring.

4708 Proposition B.3.3. Let Γ be a path-connected group, H0 a connected subgroup, and F a finite central
4709 subgroup of Γ. Write F0 “ F X H0 and suppose |F{F0 | is invertible in k. Then

H ˚ pΓ{FH0 q – H ˚ pΓ{H0 q.

4710 Proof. The space Γ{FH0 is the quotient of Γ{H0 by the left action of F{F0 given by f F0 ¨ γH0 “
4711 γ f H0 , which is well defined because F is central in Γ. But F{F0 is a subgroup of the path-
4712 connected group Γ{F0 , so the result follows from Corollary B.3.2.
 Appendix B. Topological background 178

4713 Proposition B.3.4. Let G be a compact connected Lie group and K a closed, connected subgroup, let G r be
4714 the universal compact cover of G (see Theorem B.4.4), and Kr the identity component of the preimage of K
4715 in G,
r and let F be the kernel of p : G
r ÝÑ G. If |F{pF X Kq|
r is invertible in k, then

H ˚ pG{Kq – H ˚ pG{
r Kq.
r

4716
r “ Γ and H0 “ K
Proof. By Theorem B.4.4, F is central, so taking G r in the statement of Proposi-
4717 tion B.3.3 we have Γ{H0 “ G{
r K r and Γ{FH0 « G{ppKq r “ G{K and the result follows..

4718 The two preceding lifting results are too simple not to have been known, yet the author knows
4719 no reference.

4720 Proposition B.3.5. Let F Ñ X Ñ B be a finite-sheeted covering. If either of the Euler characteristics
4721 χpXq, χpBq is finite, then so is the other, and χpXq “ χpBq ¨ |W|.

4722 Proof sketch. Taking a CW approximation, we may assume X and B to be CW complexes and
4723 X Ñ B cellular. Each cell of B is covered by |F| cells in X, so the result follows from cellular
4724 homology.

4725 The structure of compact Lie groups

4726 In this section, we record—without much explanation, interstitial verbiage, or real comprehensibility—
4727 the background we require on compact Lie groups. Dwyer and Wilkinson [DW98] develop this
4728 material in an atypical algebro-topological manner concordant with the approach adopted here.
4729 Bröcker and tom Dieck [BtD85] is another standard reference.

4730 Proposition B.4.1. There exists a smooth map exp : g ÝÑ G, the exponential, which is surjective if G
4731 is compact and connected, which restricts to a homomorphism on the preimage of any connected abelian
4732 subgroup (in particular, on any line), and whose inverse in a neighborhood of 1 P G serves as a smooth
4733 chart.

4734 Proposition B.4.2 ([Wik14]). The fundamental group of a topological group is abelian.

4735 Theorem B.4.3 ([War71, Thm. 3.58, p. 120][GGK02, Prop. B.18, p. 179]). Let G be a Lie group and K
4736 a closed subgroup. Then G{K is a manifold and K Ñ G Ñ G{K a principal K-bundle.

4737 One of the main structure theorems for compact Lie groups is the following.

4738 Theorem B.4.4 ([HM06, Thm. 2.19, p. 207]). Every compact, connected Lie group G admits a finite
4739 central extension
0 Ñ F ÝÑ G r ÝÑ G Ñ 0

4740 such that G

r is the direct product of a compact, simply-connected Lie group K and a torus A. Thus

G – AˆK
M
.
F

4741
r the universal compact cover of G; it is uniquely determined up to isomorphism.2
We call G
2 This is arguably a misnomer; this object cannot be initial in that we can always cover the toral factor A with
another torus, and in particular the fiber F is not uniquely determined by this characterization.
 Appendix B. Topological background 179

4742 Proposition B.4.5 (Élie Cartan–Wilhelm Killing). Every simply-connected Lie group K is a direct
4743 product of finitely many simple groups, groups whose maximal proper normal subgroups are finite. A
4744 simply-connected simple group is one of the following:

SUpnq, Sppnq, Spinpnq, G2 , F4 , E

r6 , E
r7 , E8 ,

4745 with the exception of Spinp1q “ Op1q and Spinp4q “ SUp2q ˆ SUp2q; the three infinite families comprise
4746 the simply-connected classical groups and the last five the exceptional groups.

4747 We will not explain the exceptional groups, but the groups Spinpnq are double covers of SOpnq
4748 for n ě 3 (when π1 SOpnq – Z{2) and Spinp2q “ SOp2q – S1 .
4749 A compact group finitely covered by a simply-connected group is called semisimple.

4750 Proposition B.4.6. Let G be a compact, semisimple Lie group. Then H 1 pG; Qq “ 0.

4751 Proof. It is an artifact of our definitions that G admits a simply-connected finite cover G.
r By the
4752
1
universal coefficient theorem Theorem B.2.1, we have H pG; Qq – H1 pG; Qq – H1 pG; Zq b Q, and
r r r
4753 by Proposition B.4.2 and Proposition B.2.5 we know H1 pG; r Zq – π1 G,
r which we have assumed to
4754 be a finite group.

4755 A classification-type result in the opposite direction is that all compact Lie groups can be seen
4756 as closed subgroups of GLpn, Cq.

4757 Theorem B.4.7 (Fritz Peter–Hermann Weyl [BtD85, Thm. III.4.1, p. 136]). Every compact Lie group
4758 G admits a faithful representation.

4759 This is a corollary of the Peter–Weyl theorem, and implies in particular that every compact Lie
4760 group embeds as a closed subgroup of Upnq for n sufficiently large.

4761 The maximal torus

4762 A real torus is a Lie group smoothly isomorphic to the direct product of finitely many copies
4763 of the complex circle group S1 – Up1q; for us tori are always considered as Lie groups. A one-
4764 dimensional torus is a circle. Much of the structure of the structure of compact, connected Lie
4765 groups arises due to tori they contain. The centralizer ZG pKq of a subgroup K of a group G is the
4766 set of g P G such that gkg´1 “ k for each k P K.

4767 Lemma B.4.8. Any torus T contains a topological generator, an element generating a dense subgroup.

4768 Sketch of proof. Any element of R` none of whose coordinates is a rational multiple of any other
4769 will project to such an element in pR{Zq` – T.

4770 Theorem B.4.9. Let G be a compact, connected Lie group. Every torus S of G is contained in a torus T
4771 which is properly contained in no other torus; such a T is called a maximal torus of G. Every element lies
4772 in some maximal torus, each maximal torus is its own centralizer in G, and all maximal tori are conjugate
4773 in G.

4774 Given a group G, and a subgroup K of G, we write NG pKq for the normalizer of K in G, the
4775 set of elements g P G such that gKg´1 “ K. The Weyl group WG of G is defined to be the quotient
4776 group NG pTq{T, the collection of nontrivial symmetries of T induced by conjugation. It is always
4777 a finite reflection group.
 Appendix B. Topological background 180

ś
4778 Lemma B.4.10. Let G “ Gj be a product of finitely many connected Lie groups Gj . Then the Weyl
ś
4779 group of G is the direct product WpGj q.

4780 Proof. This is an immediate consequence of the fact the functor pG, Kq ÞÝÑ NG pKq preserves
4781 products.

4782 Proposition B.4.11. Let G be a connected, compact Lie group. Then the center ZpGq is the intersection
4783 of all maximal tori in G.

4784 Proof [DW98, Prop. 7.1]. Any element of G lies in some conjugate gTg´1 of any given maximal
4785 torus T of G, so if z P G, there exists t P T with z “ gtg´1 , or equivalently t “ g´1 zg. If z P ZpGq
4786 is central, then g´1 zg “ z, so z P T. So a central element lies in every maximal torus.
4787 On the other hand, if x P G fails to lie in some maximal torus T, it has some conjugate in T,
4788 say gxg´1 ‰ g, meaning gx ‰ xg.

4789 The root and weight lattices

4790 This section contains a terse recounting of some standard results in Lie theory we use in Sec-
4791 tion 10.7 and Appendix E.1.3 to complete the classification of equivariantly formal isotropic
4792 circle actions.

4793 Proposition B.4.12 ([BtD85, Prop. V.(5.13), p. 214]). On the Lie algebra g of G there exists a symmetric
4794 bilinear form Bp´, ´q, the Killing form, which is invariant under adjoint action of G. This form is
4795 negative definite if G is compact.

4796 Sketch of proof. Each element v P g defines a linear endomorphism ad v P End g by pad vqw :“
4797 rv, ws. Once a basis of g is selected, a trace is well defined, and one sets Bpv, wq :“ trpad v ˝
4798 ad wq.

4799 Let G be a compact Lie group and T a maximal torus. The adjoint action of T on g yields, on
4800 tensoring with C, a representation T ÝÑ AutC pg b Cq. Because T is abelian and C algebraically
4801 closed, this representation decomposes as a direct sum of joint eigenspaces gϑ given by

gϑ “ tv P g : t ¨ v “ ϑptqv for each t P T u

4802 for some continuous homomorphisms

ϑ : T ÝÑ S1 .

4803 On the Lie algebra level, these homomorphisms correspond to roots

α : t ÝÑ R.

4804 such that ϑpexp vq “ e2πiαpvq for all v P t. We think of these roots α as lying in the dual vector
4805 space t_ :“ HomR pt, Rq. They form a finite subset Φ Ĺ t_ , the root system of G, spanning the root
4806 lattice QpGq :“ ZΦ ă t_ . A generic half-space H ` of t_ contains precisely half these roots, and
4807 determines a notion of positivity according to which the roots Φ` :“ Φ X H ` are the positive
4808 roots. An N-basis for Φ` comprises a set ∆ of simple roots for G, and a set of simple roots forms
4809 an R-basis for t_ .
 Appendix B. Topological background 181

The negative of the Killing form defines an inner product on g which we will simply denote
by v ¨ w. This inner product defines on t the Killing isomorphism
„
κ : t ÝÑ t_ ,
v ÞÝÑ pw ÞÑ v ¨ wq.

4810 of t with its dual [BtD85, p. 194], which is W-equivariant with respect to the standard action
4811 described in Proposition B.4.15. In particular, associated to each root α P t_ is a coroot α_ “
2 ´1 _
4812
α¨α κ pαq P t. The Z-span of these is the coroot lattice Q pGq ă t.
4813 The roots are completely determinative of simply-connected Lie groups.

4814 Proposition B.4.13. A connected, compact Lie group G is characterized uniquely by its maximal torus,
4815 its center, and its roots.

4816 The coroot lattice is contained in the integer lattice ΛpTq :“ kerpexp : t Ñ Tq – π1 T as the
4817 kernel [BtD85, p. 223] of a canonical surjection ΛpTq ÝÝ π1 G. We will need roots and coroots
4818 just for the following three results.

4819 Proposition B.4.14. If a Lie group G is simply-connected, then

Q_pGq “ ΛpTq.

Proposition B.4.15. The faithful representation of the Weyl group WG on t induced by the adjoint repre-
sentation and the dual representation on t_ are generated by the root reflections

sα : v ÞÝÑ v ´ 2αpvqα_ for v P t,

α¨β
sα : β ÞÝÑ β ´ 2 α for β P t_
α¨α
4820 in simple roots α P ∆.

4821 Corollary B.4.16. If the roots α P ΦpGq all satisfy α ¨ α “ 2, as when G P tAn , E6 , D4 u, then Killing
„
4822 isomorphism κ : t ÝÑ t_ takes Q_pGq isomorphically onto QpGq.

4823 The roots and the Weyl group determine a canonical decompositions of t and t_ .

4824 Definition B.4.17. Let ∆ “ tα1 , . . . , αrk G u be a collection of simple roots for a Lie group G, and
4825 let pε j q P t˘urk G be a sequence of signs. Any intersection of the half-planes α´1 `
j pε j R q is a Weyl
Ş
4826 chamber and its closure j pε j α j q´1 r0, 8q is a closed Weyl chamber. The chamber corresponding
to all positive signs, j α´1
Ş `
4827
j pR q, is the positive or fundamental Weyl chamber, and likewise for
4828 the closed Weyl chambers. Since given any root α, its opposite ´α is also a root, for any Weyl
4829 chamber C̊, there is a system of simple roots ∆ such that C̊ is the positive Weyl chamber.

4830 Proposition B.4.18 ([Ada69, Thm. 5.13, Cor. 5.16]). Let G be a compact, connected Lie group and W
4831 its Weyl group. The Weyl group W acts simply transitively on the set of Weyl chambers, and given any
4832 closed Weyl chamber C, each orbit of the adjoint action of W on tzt0u meets C in precisely one point.
4833 Appendix C
4834 Classical results on the cheap

4835 Every story has a beginning, however humble. In this appendix, which originated this work, we
4836 apply the Atiyah–Bott/Berline–Vergne localization theorem in a calculated manner to achieve
4837 simple proofs of classical results. There was no natural place to include these results earlier,
4838 as it would not have made sense to do so before developing (1) the cohomology ring H ˚ pBTq,
4839 which led into the cohomology of homogeneous spaces (Chapter 6), or (2) the Borel localiza-
4840 tion theorem, which led naturally into the discussion of equivariant formality of torus actions
4841 (Chapter 10); still, the author felt it should be included.

4842 Equivariant extensions of characteristic classes

4843 All characteristic classes (Section 6.7) admit equivariant extensions.

4844 Lemma C.1.1. Let G be a topological group, M a G-space, E Ñ M a G-equivariant vector bundle, and
4845 cpEq P H ˚ pMq a characteristic class. There exists an equivariant extension cG pEq P HG˚ pMq.

4846 Proof. The mixing construction of Section 1.1 yields a vector bundle EG Ñ MG . Fix a point
4847 e0 P EG; then i : x ÞÑ re0 , xs is the fiber inclusion of the Borel fibration M Ñ MG Ñ BG, and the
4848 original bundle E Ñ M is the pullback of the induced bundle EG Ñ MG under i:

E / EG / / BG

  
M / MG / / BG.
i

4849 The class

cG pEq :“ cpEG q P HG˚ pMq
4850 pulls back to cpEq by naturality:

i˚ cpEG q “ cpi˚ EG q “ cpEq.

4851 In particular, this holds for the Euler class of an oriented vector bundle: given an oriented
4852 vector bundle ν : E Ñ M over a G-manifold M, with fiber of dimension n, the Euler class
4853 epνq P H n pMq as defined in Section 6.7 has an equivariant extension eG pνq P HGn pMq, called
4854 the equivariant Euler class.

182
 Appendix C. Classical results on the cheap 183

4855 The Berline–Vergne/Atiyah–Bott localization theorem

4856 In this section we provide a brief statement of a simple version of the Atiyah–Bott formulation
4857 of the Berline–Vergne/Atiyah–Bott localization theorem, which suffices to recover some classical
4858 resutls.
4859 Recall from (8.1) that if T acts trivially on a manifold N, then HT˚ pNq – HT˚ b H ˚ pNq as an
4860 HT˚ -algebra. As HT˚ is a polynomial ring Zru1 , . . . , udim T s by (Section 6.4), one can view elements
4861 of HT˚ pNq as polynomials with coefficients in HpNq:
HT˚ pNq – H ˚ pNqru1 , . . . , udim T s.
4862 The classical evaluation map (or Kronecker pairing)
H n pNq ˆ Hn pNq ÝÑ Z,
4863 induced by evaluating a singular cochain on a chain, then extends to an evaluation map
HT˚ pNq ˆ H˚ pNq ÝÑ Zru1 , . . . , udim T s
given by extending additively the maps
` n ˘
H pNqru1 , . . . , udim T s ˆ Hn pNq ÝÑ Zru1 , . . . , udim T s,
pa ¨ u I , xq ÞÝÑ apxq ¨ u I .
4864 Rather than the traditional angle-bracket pairing, we denote evaluation simply by juxtaposition.
4865 If we extend the coefficient ring HT˚ to its localization H
p ˚ – Qpu1 , . . . , udim T q, then one similarly
T
4866 has a pairing
Hp ˚ pNq ˆ H˚ pN; Qq ÝÑ Qpu1 , . . . , udim T q,
T
4867 which obeys the following powerful and influential theorem, the “hammer” spoken of in the
4868 introduction.
4869 Theorem C.2.1 (Berline–Vergne/Atiyah–Bott localization [BV82][AB84]). Let M be a compact, ori-
4870 ented manifold admitting an action by a torus T. Suppose a P H dim M pM; Qq is a cohomology class
4871 admitting a T-equivariant extension a T P HTdim M pM; Qq. For each connected component N of the fixed
4872 point set M T , write i N for the inclusion N ãÝÝÑ M, write νN for the normal bundle to N in M, with
4873 orientation induced from M, write e T pνN q for the T-equivariant Euler class of νN , and denote homological
4874 fundamental classes with brackets. Then
ÿ i˚ a T
N
arMs “ T pν q
rNs.
T
e N
NĎM
4875 Remarks C.2.2. (a) The left-hand side of the display lies in HT˚ , whilst the terms of the right-hand
4876 p ˚ ; it is part of the theorem that the sum of these
side a priori lie only in the field of fractions H T
4877 rational functions is a polynomial.
4878 (b) Evaluation against rNs annihilates all of HT˚ pNq except H dim N pNq b HT˚ , so the theorem says
4879 something rather subtle about the nature of the “lower-degree” coefficients of an equivariant
4880 extension.
4881 (c) This is a weak version of the Atiyah–Bott formulation of the theorem: the full statement instead
4882 uses cohomological pushforwards, on elements a P HG pM; Qq on both sides of the equation. The
4883 Berline–Vergne formulation uses the Cartan model, replaces evaluation with integration, and
4884 evaluates the resulting equivariant forms at a vector X P t. Given how exclusively this dissertation
4885 has relied on the Borel model, it made sense to opt for the less analytic formulation here.
 Appendix C. Classical results on the cheap 184

4886 The hammer applied

4887 As promised, in this section we reobtain some classical theorems by application of Berline–
4888 Vergne/Atiyah–Bott localization.

4889 Proposition C.3.1 (Hopf–Samelson, [HS40, pp. 240–251]). Let G be a compact Lie group and H a
4890 subgroup of lesser rank. Then the Pontrjagin numbers and Euler characteristic of G{H vanish.

4891 Proof. Let T be a maximal torus in G and consider the equivariant cohomology HT˚ pG{Hq under
4892 the left action of T. By Lemma C.1.1, the characteristic classes of the tangent bundle E “ TpG{Hq
4893 have equivariant extensions c T pEq.
4894 For a characteristic class c P H top pG{H; Rq, with corresponding characteristic number crG{Hs,
4895 by Theorem C.2.1,
ÿ i˚N c T
crG{Hs “ rNs
T
e T pνN q
NĎpG{Hq

4896 which by Corollary 1.4.5 is an empty sum.

4897 The localization formula also allows us a simple way to calculate Euler characteristics in the
4898 event a space admits a torus action with isolated fixed points.

4899 Proposition C.3.2. Let T be be a torus and M a compact T-manifold with isolated fixed points. Then the
4900 Euler characteristic χpMq is |M T |, the number of fixed points.

4901 Proof [Mei06, Example 9.5]. Write νp for the normal bundle to a singleton tpu Ď M, and i p : tpu ãÑ
4902 M for the inclusion. By Theorem C.2.1,

ÿ i˚p e T pTMq
χpMq “ epTMqrMs “ .
e T pνp q
pPM T

4903 But i˚p TM – Tp M “ νp (since t0u “ Tp tpu ď Tp M), so by naturality,

i˚ e T pTMq “ e T i˚ pTMq “ e T pνp q,

` ˘

4904 and we arrive at the less intimidating formula

ÿ
χpMq “ 1.
pPM T

4905 This formula provides a slightly different explanation for the observation that χpG{Hq “ 0
4906 when H is not of full rank in G, a result usually obtained by use of the Serre spectral sequence,
4907 since by Corollary 1.4.5, the action of a maximal torus T of G on G{H has no fixed point. One
4908 can also obtain this result from an application of
4909 Proposition C.3.2 also allows us to cheaply reclaim a familiar result on homogeneous spaces,
4910 using the fixed point set of the left action of T on G{H calculated in Lemma 1.4.4.

4911 Proposition C.3.3 (Leray). Let G be a compact Lie group and H a subgroup containing a maximal torus
4912 T. Then
χpG{Hq “ |WG | { |WH |,
4913 where WG “ NG pTq{T is the Weyl group of G. In particular, χpG{Tq “ |WG |.
 Appendix C. Classical results on the cheap 185

4914 Proof. Use Proposition C.3.2, counting fixed points with Lemma 1.4.4.

4915 Historical remarks C.3.4. It was this writer’s advisor Loring W. Tu who suggested to him Propo-
4916 sition C.3.1 might be true and might admit of an equivariant proof. As far as we are aware, this
4917 proof is new; however, given that our initial literature search failed to uncover that this result was
4918 over seventy years old, the reader could be forgiven for imagining the equivariant proof might
4919 also not be original. The equivariant proof of Proposition C.3.2 was also independently obtained,
4920 despite this proof being so standard as to be found in introductory expository accounts [Mei06].
4921 In fact, the author seems to recall something stronger than Proposition C.3.1 holds, although
4922 he cannot recall the reference: the tangent bundle of such a homogeneous space is stably par-
4923 allelizable. One can find the analogous observations for Chern classes of stable almost complex
4924 structures of homogeneous spaces G{S, for S a torus, not necessarily maximal, in the last paper
4925 of the Borel–Hirzebruch trilogy on characteristic classes of homogeneous spaces [BH58; BH59;
4926 BH60].
4927 According to Dieudonné’s history [Die09], the result χpG{Tq “ |WG | was first proven by Weil
4928 in 1935 [Wei35, p. 520] and rediscovered by again by Hopf and Samelson [p. 251][HS40] (they
4929 write W “ S and σ “ |W|). In fact, they prove the full Proposition C.3.3 at the end of this paper.
4930 Appendix D
4931 Borel’s proof of Chevalley’s theorem

4932 We will require a very little standard material on sheaves and sheaf cohomology to proceed
4933 ([War71, Ch. 5], [ET14, Sec. 2]).

4934 Sheaf cohomology

4935 Taking as known the concepts of sheaf, constant sheaf, fine sheaf, acyclic sheaf, resolution, and
4936 sheaf cohomology, let k be a principal ideal domain and k the constant sheaf in the rest of this
4937 subsection.

4938 Definition D.1.1. Let C ‚ be any acyclic resolution of the constant sheaf k on a paracompact
4939 topological space X. Let A be any other sheaf on X. Then the sheaf cohomology H ˚ pX; A q is the
4940 cohomology of the complex

0 Ñ pC 0 b A qpXq ÝÑ pC 1 b A qpXq ÝÑ pC 2 b A qpXq ÝÑ ¨ ¨ ¨

of k-modules of global sections. In particular, the sheaf cohomology H ˚ pX; kq is H ˚ C ‚ pXq .

` ˘
4941

4942 Proposition D.1.2. Let X be a topological space homotopy equivalent to a finite CW complex. Then the
4943 singular cohomology and sheaf cohomology rings

H ˚ pX; kq – H ˚ pX; kq

4944 are isomorphic.

4945 Definition D.1.1 holds if the resolution F‚ is a fine sheaf of R-DGAs on a paracompact space
4946 X, and it is possible to find these.

4947 Proposition D.1.3 (Cartan). Let X be a paracompact space. Then there exists a fine sheaf of R-CDGAs
4948 resolving the constant sheaf R.

4949 Proposition D.1.4. Let F be a sheaf of k-DGAs and G a fine sheaf of k-DGAs on a paracompact topo-
4950 logical space X. Then the sheaf tensor product F b G is again fine.

4951 This has the following somewhat surprising corollary.

4952 Corollary D.1.5. Let F be a sheaf of torsion-free k-DGAs and G a fine sheaf of k-DGAs on a paracompact
4953 topological space X. Then the canonical sheaf map G ÝÑ F b G induces isomorphisms between the
4954 cohomology rings of the DGAs G pXq and pF b G qpXq with the expected differentials.

186
 Appendix D. Borel’s proof of Chevalley’s theorem 187

4955 We will need to apply this result only in a rather specific case.
4956 Example D.1.6. Let π : E ÝÑ B be a smooth fiber bundle with compact total space. Let Ω E be the
4957 sheaf de Rham complex on E, assigning to each open U Ď E the de Rham complex ΩpUq; this is
4958 a fine sheaf. Let Ω B be the sheaf de Rham complex on the base space B, and π ˚ Ω B its pullback
4959 to E: given U Ď B, one assigns
pπ ˚ Ω B qpUq “ Ω B πpUq .1
` ˘

4960 Now we can formally construct the tensor product C :“ π ˚ Ω B b Ω E , which will be another fine
4961 sheaf of R-DGAs on E by Corollary D.1.5. It follows the de Rham cohomology H ˚ pE; Rq is the
4962 cohomology of the complex
A “ C pEq “ π ˚ Ω B bR Ω E pEq.
` ˘

4963 This is disorienting, because it looks much as if we are claiming the cohomology of the
4964 complex A1 “ π ˚ ΩpBq b ΩpEq is H ˚ pEq, whereas by the Künneth theorem Corollary A.3.10 and
4965 the fact that π ˚ is injective on forms (so that π ˚ ΩpBq – ΩpBq), we should have

H ˚ pA1 q “ H ˚ pBq b H ˚ pEq.

Our escape is that the sheaf C is radically different from the presheaf U ÞÝÑ Ω πpUq b ΩpUq
` ˘
4966

4967 on E it is associated to, so that A and A1 are far from being quasi-isomorphic.
4968 Since a sheaf is determined completely by its stalks, it suffices to recall that at some (hence
4969 any) x P E, the stalk is
π Ω B b Ω E |x :“ Ω B |πpxq b Ω E |x .
` ˚ ˘

So a form τ b ω P C pUq “ Ω B πpUq b Ω E pUq becomes zero in the stalk if and only if τ “ 0 on
` ˘
4970

4971 some sufficiently small neighborhood of πpxq or else ω “ 0 on a neighborhood of x. Since global
4972 sections of C are pieced together from stalks, a tensor τ b ω of forms represents zero if and only
4973 if τ and ω are never simultaneously nonzero.
4974 Analogously, one has the following.

Proposition D.1.7. Let π : E ÝÑ B be a smooth fiber bundle with compact total space, B be a fine sheaf
of R-DGAs on B, and E a fine sheaf of R-DGAs on E. Then

pπ ˚ B b E qpEq – pπ BqpBq b E pEq

˚ L´ (¯
τ b ω : supp b X πpsupp eq “ ∅ .

4975 We will need only two more results on sheaf cohomology.

4976 Proposition D.1.8. Let F be a sheaf of k-DGAs and G an acyclic sheaf of k-DGAs on topological space
4977 X. Then the natural sheaf map F ÝÑ F b G induces isomorphisms in sheaf cohomology.

4978 Definition D.1.9. Let X be a topological space and F a sheaf of k-modules on X. Suppose C ‚ is
4979 a resolution of k, so that
0 Ñ k ÝÑ C 0 ÝÑ C 1 ÝÑ C 2 ÝÑ ¨ ¨ ¨
4980 is an exact sequence of sheaves. Then the cohomology sheaf H p pF q is the sheaf given by the pth
4981 cohomology of the complex of sheaves

0 Ñ C 0 b F ÝÑ C 1 b F ÝÑ C 2 b F ÝÑ ¨ ¨ ¨ ;
1 This simplification of the usual definition suffices because π is an open map: since πpUq is itself open, the direct

limit limVĚπpUq Ω B pVq is just Ω B πpUq again.

` ˘
ÝÑ
 Appendix D. Borel’s proof of Chevalley’s theorem 188

4982 proceeding stalkwise, one has

N
kerpC p |x b F |x ÝÑ C p`1 |x b F |x q
H pF qx “
p
impC p´1 |x b F |x ÝÑ C p |x b F |x q.

4983 Historical remarks D.1.10. Cartan’s original proof of the existence of a fine resolution of R-DGAs
4984 was never published, part of a veritable rash of non-publication that afflicted the work of this
4985 school and made this exposition a bit difficult more difficult than otherwise it might have been.

4986 The Leray spectral sequence

4987 We paraphrase Borel’s 1951 ETH exposition of the Leray spectral sequence [Bor51, Exposé VII-3].
4988 Let f : X ÝÑ Y be a map of spaces, with Y paracompact, and let F be a fine sheaf of k-DGAs
4989 on X and G a fine sheaf of k-DGAs on Y. Then the pushforward f ˚ F is again a sheaf on Y, and
4990 the sheaf G b f ˚ F is again fine, so that the sheaf cohomology H ˚ pX; f ˚ G b F q is just H ˚ pX; kq.
4991 Now F and f ˚ G are k-DGAs, so f ˚ G b F admits a natural bigrading

pG b f ˚ F q` “ G p b f˚F q ,
à
p`q“`

4992 and one can filter it by p:

Fp pG b f ˚ F q : U ÞÝÑ pG i b f ˚ F ‚ qpUq.
à
iěp

4993 We now restrict to global sections and consider the filtration spectral sequence of the horizontal
4994 filtration
Fp pG b f ˚ F qpYq :“ pG b f ˚ F qpYq.
à i
iěp

4995 as described in Corollary A.5.5.

4996 To make this sequence usable requires a bit of unpacking. The global sections of G b f ˚ F and
4997 of f ˚ G b F are both given by
O
pG b f ˚ F qpYq – p f ˚ G b F qpXq – G pYq b F pXq ´ (¯ (D.1)
τ b ω : f ´1 psupp τq b supp ω “ ∅

4998 as described in Proposition D.1.7. Since f ˚ G b F is a fine sheaf on X, we have H ˚ pX; f ˚ G b F q –

p,q
4999 H ˚ pX; kq by Proposition D.1.2 and hence E8 – gr p H p`q pX; kq, where associated graded modules
5000 are taken with respect to the filtration

Fp p f ˚ G b F qpXq :“ p f G b F qpYq.
à ˚ i
iěp

5001 on p f ˚ G b F qpXq lifting the filtration on pG b f ˚ F qpYq.

5002 Because G is already graded, the associated graded construction changes nothing on the
5003 algebra level, so
E0 “ pG b f ˚ F qpYq
5004 again. The differential d0 is induced by the differential dF of F and is zero on F , so we have
p,q
E1 “ G p b H q p f ˚ F q pYq.
` ˘
 Appendix D. Borel’s proof of Chevalley’s theorem 189

5005 These are global sections of a product sheaf, as in Proposition D.1.7; support of an element of
5006 H q p f ˚ F qpYq is determined by which stalks it vanishes on. The differential d1 is zero on H q p f ˚ F q
5007 and extends the differential dG of G . Recall from Definition D.1.1 that since G is acyclic, sheaf
5008 cohomology on Y with coefficients in any sheaf A is given by
H ˚ pY; A q “ H ˚ pG ‚ b A qpYq; dG b idA ;
` ˘

p,q
5009 in particular one finds E2 is the sheaf cohomology
p,q
E2 pYq – H p Y; H q p f ˚ F q .
` ˘

5010 To understand this last better, note that the pushforward f ˚ F is the sheaf whose stalk at y P Y is
5011 the direct limit of F pUq over neighborhoods U of f ´1 tyu, so
p f ˚ F q|y “ H ˚ f ´1 tyu; k .
` ˘

5012 Now, cohomology with coefficients in f ˚ F and in G b f ˚ F are the same by Corollary D.1.5, and
5013 we can write by abuse of notation
´ ˘¯
p,q
E2 pYq “ H p Y; H q f ´1 tyu .
`

5014 Essentially, this is the cohomology of Y with coefficients varying over the cohomology of the
5015 fibers. This spectral sequence pEr , dr q is the Leray spectral sequence of the map f : X ÝÑ Y.
5016 In the event f : X ÝÑ Y was a bundle, the fibers are homeomorphic, so the cohomology
5017 groups H ˚ pFq of individual fibers are isomorphic, related to one another by isomorphisms
5018 γ˚ : H ˚ pE|γp0q q ÝÑ H ˚ pE|γp1q q induced by lifting paths γ : r0, 1s ÝÑ Y in the base to homeo-
5019 morphisms between fibers. Thus the Leray spectral sequence of a bundle projection π : E ÝÑ B
5020 is the Serre spectral sequence of the bundle F Ñ E Ñ B, at least after the E2 page. Just as with the
5021 Serre spectral sequence, if the coefficients are trivial—and critically for us, if F “ G is a group and
5022 the bundle is principal, by Proposition 3.3.6—the coefficient sheaf in E2 can be seen as constant,
5023 and then if H ˚ pGq is a free k-module, we have E2 “ H ˚ pBq b H ˚ pGq.
5024 We summarize this discussion:
5025 Theorem D.2.1 (Leray). Let f : X ÝÑ Y be a map of spaces, with Y paracompact. Let F be a fine sheaf of
5026 k-DGAs on X and G a fine sheaf of k-DGAs on Y. Equip the sheaf G b f ˚ F with the horizontal filtration
5027 induced by the grading of G . Then associated to the map f is a spectral sequence of k-DGAs satisfying
p,q
5028 • E0 – pG p b f ˚ F q qpYq,
p,q
• E2 – H p Y; H q p f ˚ F q ,
` ˘
5029

p,q
5030 • E8 – gr p H p`q pX; kq.
5031 In the event f : X ÝÝ Y is the projection of a fiber bundle, this sequence agrees with the Serre spectral
f
5032 sequence after E2 ; if further G Ñ X Ñ Y is a principal G-bundle for G a topological group, then
E2 – H ˚ pYq b H ˚ pGq.
5033 Remark D.2.2. It is because the Serre spectral sequence is a very special instance of the Leray spec-
5034 tral sequence that it is often called the Leray–Serre spectral sequence. The Leray spectral sequence
5035 of a fibration a priori contains strictly more information than the Serre spectral sequence; it is
5036 precisely because the entire Leray spectral sequence contains such a surfeit of data that Serre’s
5037 paring it down to a spectral sequence of singular cohomology groups in the event of a fibration
5038 revolutionized homotopy theory.
 Appendix D. Borel’s proof of Chevalley’s theorem 190

5039 Borel’s proof

5040 In this section, we provide a proof of Chevalley’s theorem close to Borel’s original. Most of it is
5041 in the setup; once the relevant DGAs are defined, the quasi-isomorphisms are nearly immediate.
π
5042 Let k “ R, let G be a compact, connected Lie group, and let G Ñ E Ñ B be a smooth principal
5043 G-bundle. Write P “ PG for the space of primitives of H ˚ pGq “ H ˚ pG; Rq, so that H ˚ pGq – ΛP.
5044 Fix a transgression
„
τ : P ÝÑ QH ˚ pBGq H ˚ pBGq.
5045 As π : E Ñ B is a principal G-bundle, there is a classifying map χ : B ÝÑ BG. Let rx j s be a basis
5046 of P and rb j s “ χ˚ τpx j q P H ˚ pBq for each j.
5047 Let B be an fine sheaf of R-CDGAs resolving the constant sheaf R on B, as guaranteed by
5048 Proposition D.1.3 and likewise E be a fine sheaf of R-CDGAs resolving the constant sheaf R on
5049 E, so that by Definition D.1.1 and Proposition D.1.2,

H ˚ pBq – H ˚ BpBq ; H ˚ pEq – H ˚ E pEq .

` ˘ ` ˘

5050 We can pull B back to a sheaf π ˚ B on E and then the tensor product π ˚ B b E is another fine
5051 sheaf on E. If we set C “ pπ ˚ B b E qpEq with the expected differential, then by Corollary D.1.5,

H ˚ pCq “ H ˚ pπ ˚ B b E qpEq – H ˚ pEq

` ˘

5052 as well. Recall from Proposition D.1.7 that C can be seen as the quotient of pBq b E pEq by the
5053 ideal n spanned by elements of empty support; accepting
5054 By Theorem 6.6.5, the classes rx j s P PG are universally transgressive, which in particular
5055 means in this instance they transgress in the filtration spectral sequence pEr , dr q of C as filtered
5056 by
C p :“ pπ B b E qpEq.
à ˚ i
iěp

5057 By Theorem D.2.1, this is a version of the Leray spectral sequence of π : E ÝÑ B, which from
5058 E2 – H ˚ pBq b H ˚ pGq on, is isomorphic to the Serre spectral sequence of this bundle. Thus, as
5059 discussed in Proposition 3.3.15, the transgression of the primitive classes rx j s P PG means there
5060 exist elements c j P C such that dC c j “ π ˚ b j b 1 pmod nq,
5061 These transgressive cochains allow us to compile a simpler model C1 of H ˚ pEq as in the
5062 previously cited version Theorem 7.1.3 of Theorem D.3.1. As ΛP is a free CGA, we can lift it
5063 to a subalgebra Λtx j u of pπ ˚ B b E qpEq generated by global sections x j of π ˚ B b E . Let C1 “
5064 BpBq b Λtx j u, with differential the unique antiderivation dC1 satisfying

dC1 pb b 1q “ dB b b 1, dC1 p1 b x j q “ b j b 1

5065 and filtration

pC1 q p :“ B i pBq b H ˚ pGq.
à
iěp

Then the map

λ : C1 ÝÑ C :
b b 1 ÞÝÑ π ˚ b b 1,
1 b rx j s ÞÝÑ c j

5066 is a filtration-preserving DGA homomorphism, which we hope to show is a quasi-isomorphism.

 Appendix D. Borel’s proof of Chevalley’s theorem 191

5067 Theorem D.3.1 (Chevalley). This map λ is a quasi-isomorphism completing a commutative diagram
H ˚ pC1 q
=

!
H ˚ pGq λ˚ H ˚ pBq.
=

! 
H ˚ pEq
5068 Proof (Borel). Apply the filtration spectral sequence of (Corollary A.5.5) to both DGAs and the
5069 map λ˚ . The spectral sequence pEr , dr q of C is the Leray spectral sequence of π : E ÝÑ B, as
5070 discussed in Appendix D.2, by the identification (D.1). Write p1Er , 1dr q for the spectral sequence of
5071 C1 . The 0th page is the associated graded algebra of the filtration:
1 p,‚
E0 “ B p pBq b H ˚ pGq.
5072 Since deg x j ě 1, we have deg b j ě 2, so dC1 increases the filtration degree of each element of
5073 H ˚ pGq by at least 2, and the filtration degree of elements of BpBq by 1. Thus no image of dC1
5074 survives the associated graded procedure, so 1d0 “ 0 and
1
E1 “ 1E0 “ BpBq b H ˚ pGq.
5075 The differential 1d1 still sends generators of H ˚ pGq at least two filtration degrees forward, but acts
5076 as dB on BpBq, so that 1d1 “ δB b id H˚ pGq and
1
E2 – H ˚ pBq b H ˚ pGq.
5077 Thus 1E2 – E2 ; it just remains to see the map λ2 : 1E2 ÝÑ E2 itself is such an isomorphism
5078 in a manner making the diagram commute. But 1 b rx j s P C1 and 1 b x j pmod nq P C both
5079 become 1 b rx j s in H ˚ pBq b H ˚ pGq, and b b 1 P C1 and b b 1 pmod nq P C both become rbs b 1 in
5080 H ˚ pBq b H ˚ pGq.

5081 Historical remarks D.3.2. The proof presented above is in terms of a historically late formulation of
5082 Leray’s technology; there were several such accounts, of gradually improving comprehensibility.
5083 The entirety of the account that follows is derived from the work of Borel expositing Leray’s
5084 work, both in 1951 and 1997 [Bor51; Bor98].
5085 Leray’s original motivation for the topological edifice he erected seems to have been the de
5086 Rham complex. This is an R-CGA of forms supported on various subsets, yielding a complex
5087 which Poincaré already had shown to be trivial on Euclidean subsets, but which collate together
5088 nonetheless to contain global information about a manifold, as conjectured by Élie Cartan and
5089 proven by Georges de Rham in his thesis. Recall that the support of forms ω, τ on a manifold M
5090 satisfies these axioms:
supppτ ` ωq Ď supp τ Y supp ω; supppτ ^ ωq Ď supp τ X supp ω; supppα ¨ ωq Ď supp ω;
supp dω Ď supppωq; supp 0 “ ∅; supp 1 “ M.
Leray’s idea, beautiful in its audacity, is to equip a topological space X with a complex (complexe
concrete) K of “forms on a space,” equipped with a support function
K ÝÑ tclosed subsets of Xu,
k ÞÝÑ |k|,
 Appendix D. Borel’s proof of Chevalley’s theorem 192

5091 satisfying the same axioms as differential forms despite the absence of any native notion of
5092 smoothness, and despite elements of k not being functions in a real sense. As a purely algebraic
5093 object, a complex is a CGA over a commutative coefficient ring (which we will write as A to
5094 allow k P K); only the support function imparts any topological content.
5095 With this setup, and some further definitions, Leray is able to reprove a good amount of
5096 existing algebraic topology as of 1945, proving that the cohomology of certain types of complexes
5097 recovers Hopf’s and Samelson’s theorems on Lie groups, the Lefschetz fixed-point theorem, the
5098 Brouwer fixed-point theorem, invariance of domain, Poincaré duality, and Alexander duality.
5099 Building up couvertures (defined below) associated to nerves of a cofinal sequence of closed covers
5100 of a topological space, Leray can show this cohomology is isomorphic to Čech cohomology on
5101 compact Hausdorff spaces X.
5102 Here are some of those further definitions. If F Ď X is a closed subset, the intersection F.K is
5103 defined to be the quotient (
F.K :“ K{ k P K : |k| X F “ ∅
5104 with support function k ÞÝÑ |k| X F, and there is a natural restriction homomorphism K ÝÑ F.K.
5105 If F “ txu is a singleton, one writes xK; these are the germs of forms if K “ ΩpMq is the de Rham
5106 complex. The system of such restrictions F ÞÝÑ F.K becomes a sheaf (faisceau) under Leray’s later
5107 (1946) definition, which should be contrasted with the modern definition depending on an open
5108 cover; Leray’s definition arises because he is at this point interested in cohomology with compact
5109 supports on a locally compact space. The tensor product K b K1 of two complexes is assigned
5110 supports
|z| :“ tx P X : the image of z is nonzero under K b K1 ÝÑ xK b xK1 u,
5111 where the maps determining the support of z are the tensor products of the restriction maps
5112 discussed previously. The intersection K M K1 is then given by
1L
K M K1 :“ K b K z P K b K1 : |z| “ ∅ .
(

5113 Our Proposition D.1.7 can be seen as a later sheaf-theoretic redaction of this definition.
5114 An A-complexe is fine (fin) if every finite cover pUj q of X admits a partition of unity, which is
5115 a set of A-endomorphisms ϕ j : K ÝÑ K such that
ÿ
supp ϕ j pkq Ď Uj X supp k, ϕ j pkq “ k

5116 for all k P K. An A-complex is a couverture if its stalks are acyclic: i.e., if H ˚ pxKq “ H 0 pxKq – A
5117 for every x P X. Leray’s original cohomology on a normal space is that of the union of all
5118 couvertures.
5119 The closest thing to a fine couverture in modern language seems to be the space of global
5120 sections of a fine resolution of the constant sheaf. The intersection K M F of a sheaf and a complex
5121 is as expected, and one can take coefficients in an A-module M merely by forming the tensor
5122 product K b M or K M F b M, where the elements of A b M are viewed as having global support.
5123 A complex K (for example, a couverture) can be pushed forward under proper maps f : X ÝÑ Y
5124 and pulled back under any map g : Z ÝÑ Y: as ` algebraic
˘ structures
` ˘ f K – K – g´1 K, but the
5125
´1
supports of k in f K and g K are respectively f |k| and g |k| .´1

5126 By the time of Borel’s 1951 lectures on Leray’s work [Bor51], a sheaf (Borel credits this defini-
5127 tion to Lazard) has become the espace étalé associated to a presheaf satisfying the gluing axioms,
5128 so essentially the modern, open-set definition. The statement of the Leray spectral sequence in
5129 these lecture notes is as follows.
 Appendix D. Borel’s proof of Chevalley’s theorem 193

5130 Theorem D.3.3 (Leray). Let f : X ÝÑ Y be a continuous map, K and L fine A-couvertures, M an
5131 A-algebra, and F the sheaf associated to f pK b Mq. Then there exists a spectral sequence in which
` ˘ ` ˘
E0 “ gr f ´1 pLq M K b M , E1 “ L M HpFq, E2 “ H L M HpFq ,

5132 (d0 is the derivation induced by that of K, and d1 the derivation induced by that of L) and which terminates
5133 in the associated graded algebra of H ˚ pX, Mq, compatibly filtered.
5134 Appendix E
5135 The original classification of reflected circles in
5136 simple groups

5137 The construction of Table 10.1.6 proceeds case by case through the Killing–Cartan classification
5138 Proposition B.4.5. We do most of this theoretically, and then computationally verify a key detail
5139 in the E6 case.

5140 Theoretical reflections

5141 The case all circles are reflected
5142 It is known that if one exists, a central involution of the Weyl group acts as X ÞÝÑ ´X on the Lie
5143 algebra t of a maximal torus [DW01, Theorem 1.8] and further, that if there does exist a central
5144 involution, it is represented by the longest word w0 P W. We could therefore simply note that the
5145 center of W is Z{2 for the Weyl groups of types Bn , Cn , D2n , G2 , F4 , E7 , and E8 [Kan01, Lem. 27-2,
5146 p. 283], but we can also argue directly.

5147 Proposition E.1.1. Let K be a finite quotient of any of the simply-connected classical simple Lie groups
5148 Spinp2n ` 1q, Sppnq, and Spinp4nq, or the simply-connected exceptional Lie groups of types G2 , F4 , E7 ,
5149 and E8 , and S a maximal torus of K. Then S is reflected in K.
„ 
0 1
5150 Proof. It will be conventient to write J :“ P Op2q. Here are the cases.
1 0

5151 • Spinp2n
` ` 1q. In ˘SOp2n ` 1q, conjugations
` by the˘ orthogonal block-diagonal matrices g1 “
5152 diag J, . . . , J, r1s (and g´1 “ diag J, . . . , J, r´1s reflect the block-diagonal maximal torus
5153 T “ SOp2qn ˆ r1s , and one of g˘1 is in SOp2n ` 1q. Since Spinp2n ` 1q is a double cover of
5154 SOp2n ` 1q, the maximal torus T r of Spinp2n ` 1q is reflected by Corollary 10.3.2.

5155 • Sppnq. Choose as maximal torus the subgroup T “ Up1qn ă Spp1qn of diagonal elements
5156 with coordinates in the t1, iu-plane of H. Then conjugation by the diagonal matrix diagpj, ¨ ¨ ¨ , jq “
5157 j ¨ I P Sppnq reflects T.

5158 • Spinp4nq. Conjugation by the block-diagonal matrix diagpJ, . . . , Jq P SOp4nq reflects each
5159 element of the block-diagonal maximal torus T “ SOp2q2n . Since Spinp4nq is a double cover
5160 of SOp4nq, the maximal torus Tr of Spinp4nq is reflected by Corollary 10.3.2.

194
 Appendix E. The original classification of reflected circles in simple groups 195

5161 • G2 . The roots are the vertices of a nested pair of regular hexagons in the plane t˚ – t. The
5162 Weyl group is the dihedral group D6 of order 12, which contains rotation by π.

5163 • F4 . Recall [MT00, Theorem 7.4(1), p. 357] that F4 acts transitively on the octonionic projec-
5164 tive plane OP2 with point-stabilizer Spinp9q. A maximal torus T 4 of Spinp9q is reflected in
5165 Spinp9q, so by Observation 10.6.5, it is also reflected in F4 . But T 4 is also maximal in F4 .

5166 • E7 . Recall [Wol67, p. 285] that E7 admits a local product subgroup H – SOp12q ¨ SUp2q.
5167 Since this subgroup is of rank 7, it contains a maximal torus of E7 . By Observation 10.6.5,
5168 it suffices to know this maximal torus is reflected in H. Because H is finitely covered by
5169 the direct product SOp12q ˆ SUp2q, it is enough, by Corollary 10.3.2, to see that a maximal
5170 torus of SOp12q ˆ SUp2q is reflected. Finally, by Proposition 10.6.4, it is enough to know the
5171 maximal tori of SOp12q and SUp2q – Spp1q are reflected; but we proved so above.

5172 • E8 . Recall [Wol67, p. 285] that E8 has a subgroup SOp16q of rank 8. Let T 8 be a maximal
5173 torus of E8 lying in this SOp16q. We showed above that T 8 is reflected in SOp16q, so by
5174 Observation 10.6.5, it also is reflected in E8 .

5175 The cases of SUpnq and Spinp4n ` 2q

5176 In the remaining cases, there is no Weyl element reflecting the entire maximal torus, and we have
5177 to distinguish between circles.

5178 Proposition E.1.2. Let K “ SUpnq and let S be a circular subgroup. Then S is reflected if and only if the
5179 exponent multiset J of the inclusion S ãÝÝÑ T satisfies J “ ´J.

5180 Proof. Let T be the diagonal maximal torus Up1qn X SUpnq of SUpnq. Then S is conjugate into T,
5181 and we may replace S with this conjugate, which is the image of a map z ÞÝÑ diagpz a1 , . . . , z an q,
ř
5182 where the a j P Z have greatest common divisor 1 and a j “ 0. Write J “ ta1 , . . . , an u for the
5183 multiset of exponents characterizing this embedding of S.1
5184 The Weyl group of SUpnq is the symmetric group Sn , acting on T through ` permutation˘ of
5185 coordinates, so S is reflected if and only if there exists σ P Sn such that diag z aσp1q , . . . , z aσpnq “
5186 diagpz´a1 , . . . , z´an q for all z P S1 . This is the case if and only if J and ´J are equal as multisets.

5187 Proposition E.1.3. Let S be a circle in Spinp4n ` 2q. Then S is reflected in Spinp4n ` 2q if and only if it
5188 is conjugate into a Spinp4nq subgroup.

5189 Proof. By Proposition 10.3.1, it is equivalent to work in SOp4n ` 2q, where the equivalent state-
5190 ment is this: a circle S `in SOp4n ` 2q,
˘ is reflected if and only if it is conjugate into the block-
5191 diagonal subgroup diag SOp4nq, 1, 1 of SOp4n ` 2q.
5192 For sufficiency, if S ă T is contained in some SOp4nq ˆ t1u2 subgroup, then conjugation by a
5193 block-diagonal matrix with 2n blocks J and one 2 ˆ 2 block I induces s ÞÝÑ s´1 on S.
1 That is, as with the collection of roots of a polynomial, elements are not ordered, but are counted with multiplicity.
Though the sequence pa1 , . . . , an q is not uniquely determined by S, the pair of multisets tJ, ´Ju is; equivalently, J is
well-defined up to sign.
 Appendix E. The original classification of reflected circles in simple groups 196

For necessity, if S ă T is not contained in any SOp4nq subgroup, there are no 0 exponents a j
in the exponent multiset Jpφq “ ta1 , . . . , an u of the embedding
„ „
φ : S1 ÝÑ S ãÝÝÑ SOp2q2n`1 ÝÑ pS1 q2n`1 ,
z ÞÝÑ pz a1 , . . . , z an q,

so all exponents a j are either negative or positive. If there are n` pφq positive exponents and n´ pφq
negative exponents, then since n` pφq ` n´ pφq “ 2n ` 1 is odd, the numbers n` pφq, n´ pφq are of
differing parity. By Proposition 10.6.2, any reflection of S must be induced by the Weyl group
W “ WD2n`1 “ H2n ¸ S2n`1 , where H2n ă t˘1u2n`1 is the hyperplane defined by 2n`1
ś
j“1 ε j “ 1
and S2n`1 acts on T and on the normal factor H2n by permuting entries. Thus an even number of
exponent signs in Jpw ˝ φq differ from the corresponding signs in Jpφq, so any w P W preserves
the parities of n` and n´ :

n` pw ˝ φq ” n` pφq pmod 2q;

n´ pw ˝ φq ” n´ pφq pmod 2q.

5194 If w were to reflect S, however, we would need Jpw ˝ φq “ ´Jpφq, so that n` pw ˝ φq ” n´ pφq and
5195 n´ pw ˝ φq ” n` pφq pmod 2q, which is impossible since n` pφq and n´ pφq are of different parity.
5196 Thus S cannot be reflected in SOp4n ` 2q.

5197 E6
5198 We come to the final and most interesting case. In this section we allow ourselves to freely quote
5199 non-introductory material on reflection groups from the literature. Recall from Table 10.1.6 the
5200 desired conclusion:

5201 Proposition E.1.4. A circular subgroup S of E6 or its universal cover E

r6 is reflected just if it is conjugate
5202 into a Spinp8q subgroup.

5203 To motivate this conclusion, consider the famous chain of inclusions

SUp2q ă SUp3q ă G2 ă Spinp8q ă Spinp9q ă F4 ă E6 ă E7 ă E8 ,

5204 or more specifically the subchain Spinp8q ă F4 ă E6 . The three-fold universal cover E r6 ÝÝ E6
5205 restricts to covers of F4 and Spinp8q. Since these groups are simply-connected, they are isomorphic
5206 to the identity components of their preimages, so we get a lifted chain Spinp8q ă F4 ă E r6 .
5207
4
Because a maximal torus T of Spinp8q is reflected in Spinp8q, it is also reflected in E6 and E6 , by
r
5208 Observation 10.6.5 and Corollary 10.3.2:

5209 Proposition E.1.5. Any circular subgroup S of E6 or E

r6 contained in a Spinp8q subgroup is reflected.

5210 To prove Proposition E.1.4, we must show this sufficient condition is also necessary. Fix a
5211 maximal torus T 6 of E
r6 . We create a collection of 4-spaces in the dual pt6 q_ to the Lie algebra
5212 of T which contains all reflected subspaces of pt6 q_ , transport those spaces to t6 through the
6

5213 Killing isomorphism, then show these subspaces are the Lie algebras of maximal tori of Spinp8q
5214 subgroups of Er6 .
 Appendix E. The original classification of reflected circles in simple groups 197

5215 Definition E.1.6. Let XLV be the collection of subspaces RΞ of pt6 q_ spanned by quadruples Ξ of
r6 . Write κ for the W-equivariant isomorphism t6 ÝÑ „
5216 mutually orthogonal roots of E pt6 q_ between
5217 the adjoint and coadjoint representations of E r6 induced through the Killing form, as described in
5218 Proposition B.4.12.

Proposition E.1.7. The union wPWE tv P t6 : w ¨ v “ ´vu of all p´1q-eigenspaces of the adjoint action
Ť
5219
Ť 6
5220 of WE6 q is XLV.

Proof. Any collection tα j u Ă pt6 q_ of jointly orthogonal roots is reflected by the product wα j P W
ś
5221

5222 of commuting root reflections corresponding to those roots. Because κ is W-equivariant, each
5223 κ ´1 pRΞq P κ ´1 pXLVq is contained in the p´1q-eigenspace of some Weyl element.
5224 For the other direction, if v P t is a nonzero vector reflected by an element w P W of order
5225 2 ¨ m, where m is odd, then wm ¨ v “ ´v as well. So all reflected lines are reflected by elements of
k

5226 even-power order.

5227 Each involution w P W can be expressed as a product of orthogonal root reflections [Car72,
5228 Lem. 5, p. 5], and it will become clear later in this section (or alternately from the results of
5229 Appendix E.2) that the largest such collection is of cardinality 4, so all p´1q-eigenspaces of invo-
Ť
5230 lutions in W are contained in XLV.
5231 We now need only show the p´1q-eigenspace of each element of orders 4, 8, . . . in W is also
Ť
5232 in XLV. Given a pair w, w1 P W whose p´1q-eigenspaces satisfy V´1 pwq ď V´1 pw1 q, then for
5233 all g P W we have V´1 pgwg´1 q “ g ¨ V´1 pwq ď g ¨ V´1 pw1 q “ V´1 pgw1 g´1 q as well, so we only
5234 need to check for one element in each conjugacy class of even-power order elements whether its
Ť
5235 p´1q-eigenspace is contained in XLV.
5236 One may verify this using the GAP 4 program comprising Appendix E.2, which returns a
5237 positive result on a mid-2011 iMac with 4 gigabytes of RAM in approximately 1.5 seconds.

5238 It remains to show elements of XLV are tangent to Spinp8q subgroups. Because it will turn
5239 out (Proposition E.1.13) that |XLV| “ 45, it is impractical to demonstrate individually for each RΞ
5240 that it arises from a Spinp8q, so we will show (Proposition E.1.8) that W acts transitively on XLV,
5241 and then it will suffice to prove a single κ ´1 pRΞq is tangent to the maximal torus of a Spinp8q
5242 subgroup.
5243 But to do any of this, we need to be more explicit about what these spaces and roots are.
5244 Though the space pt6 q_ is six-dimensional, it standard to identify it not with R6 but rather with a
5245 six-dimensional subspace of R8 “ R5 ˆ R3 , the larger space conceived of as a Cartan algebra for
5246 E8 . A standard system of simple roots for E6 , already invoked in the program in Appendix E.2,
5247 is [CCN+85, p. 26][Bou68, Planche V, p. 260]

∆ :“ 1
2p1, 1, 1, 1, 1; 1, 1, 1q,
´p1, 1, 0, 0, 0; 0, 0, 0q,
p1,´1, 0, 0, 0; 0, 0, 0q,
p0, 1,´1, 0, 0; 0, 0, 0q,
p0, 0, 1,´1, 0; 0, 0, 0q,(
p0, 0, 0, 1,´1; 0, 0, 0q .

5248 These simple roots ∆ span the six-dimensional subspace

pt6 q_ “ R5 ˆ t0u3 ‘ R ¨ p1, 1, 1, 1, 1; 1, 1, 1q

` ˘
 Appendix E. The original classification of reflected circles in simple groups 198

5249 and generate the system Φ of 72 roots obtained by permuting the first five coordinates of
γ12 “ p1, 1, 0, 0, 0; 0, 0, 0q,
δ12 “ p1,´1, 0, 0, 0; 0, 0, 0q,
1
ε1 “ 2 p1,´1,´1,´1,´1; 1, 1, 1q, (E.1)
1
ζ “ 2 p1, 1, 1, 1, 1; 1, 1, 1q,
1
η12 “ 2 p´1,´1, 1, 1, 1; 1, 1, 1q,
5250 and multiplying the resulting elements by ˘1. Among these we declare the 36 positive roots to
5251 be2
Φ` :“ γab , δab , ηab 1ďaăbď5 Y ε a 1ďaď5 Y tζu.
( (

5252 Now we can show the elements of Ξ are conjugate.

5253 Proposition E.1.8. The Weyl group WE6 acts transitively on XLV.
5254 This follows immediately from the following stronger statement. (Cf. Carter [Car72, Lemma 11.(i),
5255 p. 14], where transitivity is proved for triples.)
5256 Lemma E.1.9. The Weyl group WE6 acts simply transitively on the set of ordered quadruples of mutually
5257 orthogonal roots in pt6 q_ .
5258 Proof. Given any ordered quadruple pα, β, γ, δq in Φ, we find a unique element of WE6 sending it
5259 to pζ, ´ε 1 , δ23 , δ45 q.
5260 Since W acts transitively and isometrically on Φ, there exists a w P W such that wα “ ζ, and
5261 w ¨ tβ, γ, δu is contained in the set ΦpA5 q “ t˘ε a u1ďaď5 Y t˘δab u1ďaăbď5 of roots orthogonal to ζ.
5262 Amongst these, a spanning sequence is ∆pA5 q “ p´ε 1 , δ12 , δ23 , δ34 , δ45 q. Each entry has inner
5263 product ´1 with adjacent entries and is orthogonal to the others, so ΦpA5 q is a root system of
5264 type A5 , and root reflections in ΦpA5 q generate a Weyl subgroup WpA5 q ă WE6 fixing ζ. As
5265 WpA5 q acts transitively on ΦpA5 q, there is a w1 P WpA5 q such that w1 wβ “ ´ε 1 . Since w1 is an
5266 isometry, the roots w1 w ¨ tγ, δu are contained in the set ΦpA3 q Ĺ ΦpA5 q “ t˘δab u1ďaăbď5 of roots
5267 orthogonal to tζ, ´ε 1 u.
5268 Amongst these, a spanning sequence is ∆pA3 q “ pδ23 , δ34 , δ45 q, so ΦpA3 q is a root system of
5269 type A3 , and root reflections in ΦpA3 q generate a Weyl subgroup WpA3 q ă WpA5 q fixing pζ, ´ε 1 q.
5270 As WpA3 q acts transitively on ΦpA3 q, there is a w2 P WpA3 q such that w2 w1 wγ “ δ23 . Since w2
5271 is an isometry, the root w2 w1 wδ is contained in the set ΦpA1 q “ t˘δ45 u of roots orthogonal to
5272 tζ, ´ε 1 , δ23 u, a system of type A1 .
5273 The root reflection in ΦpA1 q generates a Weyl subgroup WpA1 q ă WpA3 q fixing pζ, ´ε 1 , δ23 q.
5274 As WpA1 q acts transitively on ΦpA1 q, there is exists a w3 P WpA1 q such that w3 w2 w1 wδ “ δ45 .
5275 Thus the composition w3 w2 w1 w does what we want to the quadruple pα, β, γ, δq. That it is
5276 the unique Weyl group element doing so follows from a sequence of applications of the orbit–
5277 stabilizer theorem. Since WE6 has order 51,840 and transitively permutes 72 roots, it follows
5278 Stabpζq has order 720. Since Stabpζq transitively permutes the 30 roots ΦpA5 q, it follows that the
5279 ordered pair stabilizer Stabpζ, ´ε 1 q has order 24. Since this stabilizer transitively permutes the
5280 12 roots ΦpA3 q, it follows the triplet-stabilizer Stabpζ, ´ε 1 , δ23 q has order 2. And this stabilizer
5281 acts simply transitively on ΦpA1 q “ t˘δ0 u, so Stabpζ, ´ε 1 , δ23 , δ45 q is trivial and WE6 acts simply
5282 transitively on ordered quadruples of mutually orthogonal roots.
2 Here, to be clear, in δ the 1 appears in the ath coordinate and the ´1 in the bth , but we will let η “ η and γ “
ab ab ba ab
γba for notational convenience. For example, δ24 “ ´δ42 “ p0, 1, 0, ´1, 0; 0; 0, 0q and η43 “ η34 “ 21 p1, 1, ´1, ´1, 1; 1, 1, 1q
and ε 5 “ 21 p´1, ´1, ´1, ´1, 1; 1, 1, 1q.
 Appendix E. The original classification of reflected circles in simple groups 199

5283 Now we show such a Φ X RΞ is a D4 root system.

5284 Lemma E.1.10. The D4 root subsystems of Φ are precisely tΦ X RΞ : RΞ P XLVu.

5285 Proof. A root system of type D4 is spanned by four mutually orthogonal roots, so a D4 subsystem
5286 of Φ must be Φ X RΞ for some RΞ P XLV.
5287 On the other hand, let Ξ “ tγ12 , δ12 , γ34 , δ34 u; then RΞ P XLV is generic by Proposition E.1.8.
5288 The quadruple ∆ “ pγ12 , γ34 , δ34 , ´γ23 q in Φ X RΞ spans RΞ, its first three entries are mutually
5289 orthogonal, and γ23 ¨ γ12 “ γ23 ¨ γ34 “ γ23 ¨ δ34 “ ´1, so Φ X RΞ is a root system of type D4 .

5290 Now we can finally show XLV does have something to do with Spinp8q.

5291 Proposition E.1.11. Given a maximal torus T 6 of E r6 , the duals in t6 to the spaces XLV in pt6 q_ are
5292 precisely the Lie algebras t of those maximal tori T of Spinp8q subgroups such that T 4 is contained in
6 4

5293 T6.

5294 Proof. Recalling the discussion of integer and coroot lattices in Appendix B.4.2, consider a Spinp8q
5295 r6 whose maximal torus T 4 lies in T 6 . The inclusion T 4 ãÝÝÑ T 6 induces inclusions t4 ãÝÝÑ t6 and
in E
5296 ΛpT 4 q ãÝÝÑ ΛpT 6 q.3 Since Spinp8q and Er6 are both simply-connected, their coroot lattices are their
` ˘
5297 integer lattices by Proposition B.4.14, so this last inclusion can also be written Q_ Spinp8q ãÝÝÑ
?
5298 Q_ pEr6 q. Selecting all elements of length 2 in these lattices respectively yields root systems
5299 Φ_ pSpinp8qq Ĺ Φ_ pE r6 q of types D4 and E6 within t4 and t6 . Because all roots of both groups are
5300 long, by Corollary B.4.16 the Killing isomorphism κ takes these systems respectively to a root
5301 system ΦpD4 q of type D4 and the root system Φ of E r6 described above. By Lemma E.1.10, the
5302 span of ΦpD4 q is Φ X RΞ for some Ξ P XLV. That is, κ pt4 q P XLV.

5303 The transitivity of the action on orthogonal quadruples makes it relatively simple to enumer-
5304 ate XLV.

5305 Lemma E.1.12. A vector space RΞ P XLV is spanned by precisely three distinct sets of four mutually
5306 orthogonal roots.

5307 Proof. Let Ξ “ tγ12 , δ12 , γ34 , δ34 u again. The positive roots Φ` X RΞ are ˇtγab , δab : 1ˇ ď a ă b ď 4u.
5308 Among these, the three roots orthogonal to δab are tγab , γcd , δcd u, where ˇta, b, c, duˇ “ 4 and c ă d,
5309 and likewise the three roots orthogonal to γab are tδab , γcd , δcd u, so the spanning
( quadruples are
5310 determined by the partitions of t1, 2, 3, 4u into pairs of pairs ta, bu, tc, du , of which there are
5311 three.

5312 Now we can conclude the enumeration.

5313 Proposition E.1.13. We have |XLV| “ 45.

5314 Proof. The Weyl group WE6 acts simply transitively on theˇ set of ordered
ˇ quadruples pα, β, γ, δq
5315 ˇ 4
of mutually orthogonal roots, by Proposition E.1.8, and t˘1u ¸ S4 distinct such quadruples
ˇ
5316 correspond
ˇ ˇ ˇ to anyˇ given set tα, β, γ, δu of four mutually orthogonal positive roots, so there are
5317 ˇWE ˇ { ˇt˘1u4 ¸ S4 ˇ “ 51,840{384 “ 135 such sets. By Lemma E.1.12, each such 4-space admits
6
5318 three distinct unordered bases of positive roots, so there are 135/3 = 45 distinct such spaces.
3 To tie this back to circles, the exponential of each line s in t6 containing a point of ΛpT 6 q is a circle in T 6 , so the
set of circles in T 6 can be identified with the projectivization PΛpT 6 q.
 Appendix E. The original classification of reflected circles in simple groups 200

5319 To summarize, we have proven the following.

5320 Proposition E.1.14. If a circular subgroup S lies within a maximal torus T 6 of E6 , then S is reflected just
5321 if it is contained one of forty-five Weyl-conjugate maximal tori lying in T 6 which are tangent spaces of
5322 Spinp8q subgroups of E6

5323 Remarks E.1.15. (a) Proposition

ˇ ˇ E.1.13 also ˇ that the stabilizer of any element of XLV in WE6
ˇ shows
5324 is a group of order ˇWE6 ˇ{45 “ 1,152 “ ˇWpF4 qˇ. If we write N “ NE6 pT 4 q and Z “ ZE6 pT 4 q, then
5325 by the proof of Observation 10.6.5 we have the chain of injections

N WE6
WpF4 q ,
Z FixWE6 pT 4 q

5326 and we have in essence just shown the map WpF4 q  N{Z is actually an isomorphism and
5327 FixWE6 pT 4 q is trivial, so that WpF4 q injects into WE6 as the set of elements normalizing T 4 . The
5328 author is advised this result can be understood from Carter’s book [Car85, Sec. 13.3].
(b) We originally found the 135 maximal mutually orthogonal sets of positive roots through brute
force, then checked with Python we had accounted for them all. Explicitly, in terms of (E.1), they
are

p60q tε a , ηab , γac , δde u, where |ta, b, c, d, eu| “ 5 and d ă e,

p30q tηab , ηcd , γac , γbd u, where |ta, b, c, du| “ 4,
p15q tηab , ηcd , δab , δcd u, where |ta, b, c, du| “ 4 and a ă b and c ă d,
p15q tγab , γcd , δab , δcd u, where |ta, b, c, du| “ 4 and a ă b and c ă d,
p15q tζ, ε a , δbc , δde u, where |ta, b, c, d, eu| “ 5 and b ă c and d ă e.

5329 The GAP verification of Proposition E.1.7

# Supply a standard system of simple roots for E_6.
E6_simple_roots := [
[1,-1,-1,-1,-1,1,1,1]/2,
-[1,1,0,0,0,0,0,0],
[1,-1,0,0,0,0,0,0],
[0,1,-1,0,0,0,0,0],
[0,0,1,-1,0,0,0,0],
[0,0,0,1,-1,0,0,0]
];;

# Simple reflections generate the Weyl group of E_6. Check the Weyl group is the right size.
simple_refl := List(E6_simple_roots,ReflectionMat);; W := Group(simple_refl); Size(W);

# Find its conjugacy classes. How big are they?

conj_raw := ConjugacyClasses(W);; List(conj_raw,x->Size(x));

# There are no elements of order 16.

Filtered(conj_raw,c->Order(Representative(c))=16);
Appendix E. The original classification of reflected circles in simple groups 201

# We find all conjugacy classes of elements of orders 2, 4, and 8.

conj_cl248 := [];;
for expon in [1..3] do
conj_cl248[expon]:= Filtered(conj_raw,c->Order(Representative(c))=2^expon);;
od;

# How big are these conjugacy classes of elements of orders 2, 4, and 8?

List(conj_cl248,x->List(x,y->Size(y)));

# There are four conjugacy classes of involutions.

# We know these are all products of four or fewer orthogonal root reflections.

# This function finds the (-1)-eigenspace of a matrix.

reflected_space := function(matrix)
local e;
for e in Eigenspaces(Rationals,matrix) do
if matrix*Representative(e) = -Representative(e) then
return e;
fi;
od;
return [];
end;;

# This function finds a reflected space of some element in a conjugacy class.

refl_sp_of_rep := class->reflected_space(Representative(class));;

# This function orders a set by its function values.

ord := fcn->(set->Concatenation(List(Set(set,fcn),x->Filtered(set,v->fcn(v)=x))));

# We reorder the conjugacy classes of involutions so that the fourth one is the
# one with the highest-dimensional reflected spaces.
conj_cl248[1] :=
ord(x->Dimension(refl_sp_of_rep(x)))(conj_cl248[1]);;

# The largest (-1)-eigenspaces of involutions have dimension 4.

# We know these contain all (-1)-eigenspaces of other involutions.
# We generate all these (45) four-dimensional (-1)-eigenspaces.
invol_refl_spaces := List(conj_cl248[1][4],x->reflected_space(x));;

# We generate one reflecting space for each conjugacy class of elements of

# of orders 2, 4, and 8.
rep248_refl_spaces := Concatenation(List(conj_cl248,y->List(y,x->refl_sp_of_rep(x))));;

# This function checks if a vector space is contained in such an eigenspace.

Appendix E. The original classification of reflected circles in simple groups 202

subset_invol_refl_space := function(vector_space)
local V, answer;
answer := "No.";
for V in invol_refl_spaces do
if IsSubset(V,vector_space) then
answer := "Yes.";
fi;
od;
return answer;
end;;

# We run through representatives of conjugacy classes of elements of orders 4 and 8

# and check whether their (-1)-eigenspaces are contained in a four-dimensional one.
for class in [5..Size(rep248_refl_spaces)] do;
Print("Class ");
Print(class);
Print(": ");
Print(subset_invol_refl_space(rep248_refl_spaces[class]));
Print("\n");
od;

# For every class, a "Yes" shows up, demonstrating the four-dimensional (-1)-eigenspaces
# of involutions in fact contain all (-1)-eigenspaces.
5330 Bibliography

5331 [AB84] M. F. Atiyah and R. Bott, “The moment map and equivariant cohomology,” Topology,
5332 vol. 23, no. 1, pp. 1–28, 1984.
5333 [Ada69] J. F. Adams, Lectures on Lie groups. University of Chicago Press, 1969.
5334 [Ada76] J. F. Adams, “The work of H. Cartan in its relation with homotopy theory,” Astérisque,
5335 vol. 32–33, pp. 29–41, 1976.
5336 [AH61] M. F. Atiyah and F. Hirzebruch, “Vector bundles and homogeneous spaces,” in Dif-
5337 ferential geometry, ser. Proc. Symp. Pure Math. AMS, vol. III, Providence, RI, 1961,
5338 pp. 7–38.
5339 [AM69] M. F. Atiyah and I. G. Macdonald, Introduction to Commutative Algebra. 1969.
5340 [Ama13] M. Amann, “Non-formal homogeneous spaces,” Math. Z., vol. 274, no. 3-4, pp. 1299–
5341 1325, 2013.
5342 [And62] M. André, “Cohomologie des algèbres différentielles où opère une algèbre de Lie,”
5343 Tôhoku Math. J. (2), vol. 14, no. 3, pp. 263–311, 1962.
5344 [AP93] C. Allday and V. Puppe, Cohomological methods in transformation groups, ser. Cam-
5345 bridge Stud. Adv. Math. Cambridge Univ. Press, 1993, vol. 32.
5346 [Bau62] P. F. Baum, “Cohomology of homogeneous spaces,” PhD thesis, Princeton, Dec. 1962.
5347 [Online]. Available: [Link]
5348 homogeneous_spaces%28Baum_dissertation_ORIGINAL%[Link]?dl=0.
5349 [Bau68] P. F. Baum, “On the cohomology of homogeneous spaces,” Topology, vol. 7, no. 1,
5350 pp. 15–38, 1968.
5351 [BBF+60] A. Borel, G. Bredon, E. E. Floyd, D. Montgomery, and R. Palais, Seminar on transfor-
5352 mation groups, ser. Ann. of Math. Stud. 46. Princeton University Press, 1960.
5353 [BCM] R. Bruner, M. Catanzaro, and J. P. May, “Characteristic classes,” Unpublished notes,
5354 [Online]. Available: [Link]
5355 [Bee] C. Beenakker, Citation: earliest incidence of the Borel localization theorem, MathOverflow.
5356 [Online]. Available: [Link]
5357 [BH58] A. Borel and F. Hirzebruch, “Characteristic classes and homogeneous spaces, I,”
5358 Amer. J. Math., pp. 458–538, 1958.
5359 [BH59] ——, “Characteristic classes and homogeneous spaces, II,” Amer. J. Math., pp. 315–
5360 382, 1959.
5361 [BH60] ——, “Characteristic classes and homogeneous spaces, III,” Amer. J. Math., pp. 491–
5362 504, 1960.

203
 Bibliography 204

5363 [Bor51] A. Borel, Cohomologie des espaces Iocalement compacts d’aprés J. Leray, ser. Lecture Notes
5364 in Math. Springer-Verlag, 1951, vol. 2, Sém. de Top. alg., FTH.
5365 [Bor53] ——, “Sur la cohomologie des espaces fibrés principaux et des espaces homogènes
5366 de groupes de Lie compacts,” Ann. of Math. (2), vol. 57, no. 1, pp. 115–207, 1953,
5367 issn: 0003-486X.
5368 [Bor54] ——, “Sur l’homologie et la cohomologie des groupes de Lie compacts connexes,”
5369 American Journal of Mathematics, pp. 273–342, 1954.
5370 [Bor67] ——, Topics in the homology theory of fibre bundles: lectures given at the University of
5371 Chicago, 1954, ser. Lecture Notes in Math. Springer-Verlag, 1967, vol. 36.
5372 [Bor98] ——, “Jean Leray and algebraic topology,” in Jean Leray, Selected Papers: Oeuvres
5373 Scientifiques, A. Borel, Ed., vol. 1, Springer and Soc. Math. France, 1998, pp. 1–21.
5374 [Bou05] N. Bourbaki, Lie Groups and Lie Algebras: Chapters 7–9, trans. by A. Pressley, ser.
5375 Elements of Mathematics. Springer, 2005.
5376 [Bou68] ——, Groupes et algèbres de Lie, chap. IV–VI. 1968.
5377 [Bre72] G. E. Bredon, Introduction to compact transformation groups. Academic press, 1972,
5378 vol. 46.
5379 [Bri98] M. Brion, “Equivariant cohomology and equivariant intersection theory,” in Repre-
5380 sentation theories and algebraic geometry, Springer, 1998, pp. 1–37.
5381 [BS67] P. Baum and L. Smith, “The real cohomology of differentiable fibre bundles,” Com-
5382 ment. Math. Helv., vol. 42, no. 1, pp. 171–179, 1967.
5383 [BT82] R. Bott and L. W. Tu, Differential forms in algebraic topology. Springer, 1982, vol. 82.
5384 [BTara] ——, Elements of Equivariant Cohomology, ser. Monographs in Mathematics. Springer,
5385 to appear.
5386 [BTarb] ——, Elements of equivariant cohomology. Springer, to appear.
5387 [BtD85] T. Bröcker and T. tom Dieck, Representations of compact Lie groups, ser. Grad. Texts in
5388 Math. Springer, 1985, vol. 98.
5389 [BV82] N. Berline and M. Vergne, “Classes caractéristiques équivariantes. Formule de lo-
5390 calisation en cohomologie équivariante,” C. R. Acad. Sci. Paris, vol. 295, pp. 539–541,
5391 1982.
5392 [Cai41] S. S. Cairns, “Triangulated manifolds and differentiable manifolds,” in Lectures in
5393 topology, 1941, pp. 143–157.
5394 [Car51] H. Cartan, “La transgression dans un groupe de Lie et dans un espace fibré princi-
5395 pal,” in Colloque de topologie (espace fibrés), Bruxelles 1950, Centre belge de Recherches
5396 mathématiques, Liège/Paris: Georges Thone/Masson et companie, 1951, pp. 57–71.
5397 [Car72] R. W. Carter, “Conjugacy classes in the Weyl group,” Compositio Mathematica, vol.
5398 25, no. 1, pp. 1–59, 1972. [Online]. Available: [Link]
5399 1972__25_1_1_0.
5400 [Car85] R. W. Carter, Finite groups of Lie type: Conjugacy classes and complex characters. Wiley,
5401 1985.
 Bibliography 205

5402 [CCN+85] J. H. Conway, R. T. Curtis, S. P. Norton, R. A. Parker, and R. A. Wilson, Atlas of finite
5403 groups. Oxford: Clarendon Press, 1985.
5404 [Che46] S.-S. Chern, “Characteristic classes of Hermitian manifolds,” Ann. of Math., pp. 85–
5405 121, 1946.
5406 [Colloque] Colloque de topologie (espace fibrés), Bruxelles 1950, Centre belge de Recherches mathé-
5407 matiques, Liège/Paris: Georges Thone/Masson et companie, 1951.
5408 [Die09] J. Dieudonné, A history of algebraic and differential topology, 1900–1960. Springer, 2009.
5409 [DW01] W. G. Dwyer and J. C. W. Wilkerson, “Centers and Coxeter elements,” Contemp.
5410 Math., vol. 271, pp. 53–76, 2001.
5411 [DW98] ——, “The elementary geometric structure of compact Lie groups,” Bull. London
5412 Math. Soc., vol. 30, no. 4, pp. 337–364, 1998.
5413 [Eis95] D. Eisenbud, Commutative algebra with a view toward algebraic geometry, ser. Grad.
5414 Texts in Math. Springer, 1995, vol. 150.
5415 [Esc92] J.-H. Eschenburg, “Cohomology of biquotients,” Manuscripta Math., vol. 75, no. 2,
5416 pp. 151–166, 1992, issn: 0025-2611. doi: 10.1007/BF02567078. [Online]. Available:
5417 [Link]
5418 [ET14] F. El Zein and L. W. Tu, “From sheaf cohomology to the algebraic de Rham theorem,”
5419 in Hodge Theory, ser. Mathematical Notes, E. Cattani, F. E. Zein, P. A. Griffiths, and
5420 L. D. Tráng, Eds., vol. 49, Princeton University Press, 2014, ch. 2, pp. 70–121.
5421 [FH91] W. Fulton and J. Harris, Representation theory, ser. Grad. Texts in Math. New York:
5422 Springer, 1991, vol. 129.
5423 [FHT01] Y. Félix, S. Halperin, and J.-C. Thomas, Rational homotopy theory, ser. Grad. Texts in
5424 Math. Springer, 2001, vol. 205.
5425 [FOT08] Y. Félix, J. Oprea, and D. Tanré, Algebraic models in geometry, ser. Oxford Graduate
5426 Texts in Mathematics. Oxford: Oxford University Press, 2008, vol. 17.
5427 [FP07] M. Franz and V. Puppe, “Exact cohomology sequences with integral coefficients for
5428 torus actions,” Transformation Groups, vol. 12, no. 1, pp. 65–76, 2007.
5429 [FP08] ——, “Freeness of equivariant cohomology and mutants of compactified representa-
5430 tions,” in Toric Topology (Osaka, 2006), M. H. et al., Ed., ser. Contemp. Math. Vol. 460,
5431 Providence, RI: Amer. Math. Soc., 2008, pp. 87–98.
5432 [Gar10] P. Garrett, “Solenoids,” A section from electronically published course notes on
5433 modular forms, Sep. 2010, [Online]. Available: [Link]
5434 m/mfms/notes/02_solenoids.pdf.
5435 [GGK02] V. L. Ginzburg, V. Guillemin, and Y. Karshon, Moment maps, cobordisms, and Hamil-
5436 tonian group actions, ser. Math. Surveys Monogr. Providence, RI: Amer. Math. Soc.,
5437 2002, vol. 98. [Online]. Available: [Link]
5438 monograph/[Link].
5439 [GHV76] W. H. Greub, S. Halperin, and R. Vanstone, Connections, curvature, and cohomology,
5440 vol. III: Cohomology of principal bundles and homogeneous spaces. Academic Press, 1976.
5441 [GKM98] M. Goresky, R. Kottwitz, and R. MacPherson, “Equivariant cohomology, Koszul du-
5442 ality, and the localization theorem,” Invent. Math., vol. 131, no. 1, pp. 25–83, 1998.
 Bibliography 206

5443 [GLS96] V. Guillemin, E. Lerman, and S. Sternberg, Symplectic fibrations and multiplicity dia-
5444 grams. Cambridge University Press, 1996.
5445 [GN14] O. Goertsches and S. H. Noshari. (May 2014). On the equivariant cohomology of
5446 isotropy actions on generalized symmetric spaces. arXiv: 1405.2655v1 [[Link]].
5447 [GN15] ——, (Apr. 2015). On the equivariant cohomology of isotropy actions on homoge-
5448 neous spaces defined by lie group automorphisms. arXiv: 1405.2655v2 [[Link]].
5449 [Goe12] O. Goertsches, “The equivariant cohomology of isotropy actions on symmetric spaces,”
5450 Doc. Math., vol. 17, pp. 79–94, 2012.
5451 [GS99] V. W. Guillemin and S. Sternberg, Supersymmetry and equivariant de Rham theory, ser.
5452 Mathematics Past and Present. New York: Springer, 1999, vol. 2.
5453 [GT10a] O. Goertsches and D. Töben. (Apr. 2010). Equivariant basic cohomology of Rieman-
5454 nian foliations. arXiv: 1004.1043 [[Link]].
5455 [GT10b] ——, “Torus actions whose equivariant cohomology is Cohen–Macaulay,” J. Topol.,
5456 2010.
5457 [Hat02] A. Hatcher, Algebraic topology. Cambridge Univ. Press, 2002.
5458 [Hat04] ——, “Spectral sequences in algebraic topology,” Unpublished book project, available
5459 on [Link] math. cornell. edu/hatcher/SSAT/SSATpage. html, 2004.
5460 [Hes99] K. Hess, “A history of rational homotopy theory,” in History of topology, I. M. James,
5461 Ed., Elsevier, 1999.
5462 [Hir48] G. Hirsch, “Un isomorphisme attaché aux structures fibrées,” C. R. Acad. Sci. Paris,
5463 vol. 227, pp. 1328–1330, 1948.
5464 [HM06] K. H. Hofmann and S. A. Morris, The Structure of Compact Groups: A Primer for
5465 Students — a Handbook for the Expert, 2nd revised and augmented edition, ser. De
5466 Gruyter studies in mathematics. Walter de Gruyter, 2006, vol. 25.
5467 [Hop40] H. Hopf, “Über den Rang geschlossener liescher Gruppen,” Comment. Math. Helv.,
5468 vol. 13, no. 1, pp. 119–143, 1940.
5469 [Hop41] ——, “Über eie Topologie der Gruppen-Mannigfaltigkeiten und ihre Verallgemeinerun-
5470 gen,” German, Ann. of Math., Second Series, vol. 42, no. 1, pp. 22–52, Jan. 1941, issn:
5471 0003486X. [Online]. Available: [Link]
5472 [HS40] H. Hopf and H. Samelson, “Ein Satz über die Wirkungsräume geschlossener liescher
5473 Gruppen,” Comment. Math. Helv., vol. 13, no. 1, pp. 240–251, 1940.
5474 [Hsi75] W.-Y. Hsiang, Cohomology theory of topological transformation groups. Springer, 1975.
5475 [Hum92] J. E. Humphreys, Reflection groups and Coxeter groups, ser. Cambridge Stud. Adv.
5476 Math. Cambridge Univ. Press, 1992, vol. 29.
5477 [Ill78] S. Illman, “Smooth equivariant triangulations of G-manifolds for G a finite group,”
5478 Math. Ann., vol. 233, no. 3, pp. 199–220, 1978.
5479 [Ill83] ——, “The equivariant triangulation theorem for actions of compact Lie groups,”
5480 Math. Ann., vol. 262, no. 4, pp. 487–501, 1983.
5481 [Kan01] R. Kane, Reflection groups and invariant theory, ser. C. M. S. Books in Mathematics.
5482 Springer, 2001, vol. 5.
 Bibliography 207

5483 [Kap] V. Kapovitch, “A note on rational homotopy of biquotients,” [Online]. Available:

5484 [Link]
5485 [Kos47a] J.-L. Koszul, “Sur les opérateurs de dérivation dans un anneau,” C. R. Acad. Sci.
5486 Paris, vol. 225, no. 4, pp. 217–219, 1947.
5487 [Kos47b] ——, “Sur l’homologie des espaces homogènes,” C. R. Acad. Sci. Paris, vol. 225,
5488 pp. 477–479, Sep. 1947.
5489 [Kos50] ——, “Homologie et cohomologie des algèbres de Lie,” Bulletin de la Société Mathé-
5490 matique de France, vol. 78, pp. 65–127, 1950.
5491 [Kos51] ——, “Sur un type d’algebres differentielles en rapport avec la transgression,” in
5492 Colloque de topologie (espace fibrés), Bruxelles 1950, Centre belge de Recherches mathé-
5493 matiques, Liège/Paris: Georges Thone/Masson et companie, 1951, pp. 73–81.
5494 [Kos53] ——, “Sur certains groupes de transformations de Lie,” in Colloque de topologie (espace
5495 fibrés), Bruxelles 1950, Centre belge de Recherches mathématiques, vol. 52, Liège/Paris:
5496 Georges Thone/Masson et companie, 1953, pp. 137–141.
5497 [KT03] D. Kotschick and S. Terzić, “On formality of generalized symmetric spaces,” in Math-
5498 ematical Proceedings of the Cambridge Philosophical Society, vol. 134, Cambridge Univ.
5499 Press, 2003, pp. 491–505.
5500 [KV93] S. Kumar and M. Vergne, “Equivariant cohomology with generalized coefficients,”
5501 Astérisque, vol. 215, pp. 109–204, 1993.
5502 [Ler46a] J. Leray, “Structure de l’anneau d’homologie d’une représentation,” C. R. Acad. Sci.
5503 Paris, vol. 222, pp. 1419–1422, Jul. 1946. [Online]. Available: [Link]
5504 fr/ark:/12148/bpt6k31740/[Link].
5505 [Ler46b] ——, “Sur l’anneau d’homologie de l’espace homogène quotient d’un groupe clos
5506 par un sous-groupe abélien, connexe, maximum,” C. R. Acad. Sci. Paris, vol. 229,
5507 pp. 412–415, 1946.
5508 [Ler49a] ——, “Détermination, dans les cas non exceptionnels, de l’anneau de cohomologie
5509 de l’espace homogène quotient d’un groupe de Lie compact par un sous-groupe de
5510 même rang,” C. R. Acad. Sci. Paris, vol. 228, no. 25, pp. 1902–1904, 1949.
5511 [Ler49b] ——, “Sur l’anneau de cohomologie des espaces homogenés,” C. R. Acad. Sci. Paris,
5512 vol. 229, no. 4, pp. 281–283, 1949.
5513 [Ler51] ——, “Sur l’homologie des groupes de Lie, des espaces homogènes et des espaces
5514 fibrés principaux,” in Colloque de topologie (espace fibrés), Bruxelles 1950, Centre belge
5515 de Recherches mathématiques, Liège/Paris: Georges Thone/Masson et companie,
5516 1951, pp. 101–115.
5517 [Mas91] W. S. Massey, A basic course in algebraic topology, ser. Grad. Texts in Math. Springer,
5518 1991, vol. 127.
5519 [Mat71a] T. Matumoto, “Equivariant K-theory and Fredholm operators,” 1971.
5520 [Mat71b] ——, “On G-CW complexes and a theorem of J. H. C. Whitehead,” J. Fac. Sci. Univ.
5521 Tokyo Sect. 1A, vol. 18, no. 2, pp. 363–374, Dec. 1971. [Online]. Available: http://
5522 [Link]/dspace/bitstream/2261/6106/1/jfs180205.
5523 pdf.
 Bibliography 208

5524 [Mat73] ——, “Equivariant cohomology theories on G-CW complexes,” Osaka Journal of Math-
5525 ematics, vol. 10, no. 1, pp. 51–68, 1973.
5526 [McC01] J. McCleary, A user’s guide to spectral sequences, ser. Cambridge Stud. Adv. Math.
5527 Cambridge: Cambridge Univ. Press, 2001, vol. 58.
5528 [Mei06] E. Meinrenken, “Equivariant cohomology and the Cartan model,” in Encyclopedia of
5529 mathematical physics. Oxford: Academic Press/Elsevier Science, 2006. [Online]. Avail-
5530 able: [Link]
5531 [Mil00] H. Miller, “Leray in Oflag XVIIA: the origins of sheaf theory, sheaf cohomology, and
5532 spectral sequences,” Gazette des mathématiciens, no. 84 suppl. Pp. 17–34, Apr. 2000.
5533 [Online]. Available: [Link]
5534 [Mil56] J. Milnor, “Construction of universal bundles, II,” Ann. of Math., vol. 63, no. 3,
5535 pp. 430–436, May 1956. [Online]. Available: [Link]
5536 [Link].
5537 [MT00] M. Mimura and H. Toda, Topology of Lie groups, I and II, ser. Transl. Math. Monogr.
5538 Providence, RI: Amer. Math. Soc., 2000, vol. 91.
5539 [Mun74] H. J. Munkholm, “The Eilenberg–Moore spectral sequence and strongly homotopy
5540 multiplicative maps,” J. Pure Appl. Algebra, vol. 5, no. 1, pp. 1–50, 1974.
5541 [Oni72] A. L. Oniščik [Onishchik], “On certain topological invariants of homogeneous spaces,”
5542 Mat. Zametki, vol. 12, pp. 761–768, 1972, [in Russian].
5543 [Oni94] A. L. Onishchik, Topology of transitive transformation groups. Johann Ambrosius Barth,
5544 1994.
5545 [Pit91] H. V. Pittie, “The integral homology and cohomology rings of SO(n) and Spin(n),”
5546 J. Pure Appl. Algebra, vol. 73, no. 2, pp. 105–153, 1991.
5547 [Pos66] M. M. Postnikov, “On Cartan’s theorem,” Russian Mathematical Surveys, vol. 21, no.
5548 4, pp. 25–36, 1966.
5549 [PT07] A. Pedroza and L. W. Tu, “On the localization formula in equivariant cohomology,”
5550 Topology and its Applications, vol. 154, no. 7, pp. 1493–1501, 2007.
5551 [Ral] Ralph, Transgression maps in group cohomology and group homology / duality of spectral
5552 sequences, MathOverflow. [Online]. Available: [Link]
5553 [Ras69] P. K. Rashevskii, “The real cohomology of homogeneous spaces,” Russian Mathemat-
5554 ical Surveys, vol. 24, no. 3, p. 23, 1969.
5555 [Sam41] H. Samelson, “Beitrage zur Topologie der Gruppen-Mannigfaltigkeiten,” Ann. of
5556 Math., vol. 42, no. 1, pp. 1091–1137, Jan. 1941. [Online]. Available: http : / / www .
5557 [Link]/stable/1970463.
5558 [Sam52] ——, “Topology of Lie groups,” Bull. Amer. Math. Soc., vol. 58, no. 1, pp. 2–37, 1952.
5559 [Ser00] J.-P. Serre, Local algebra, trans. by C. Chin, ser. Monographs in Mathematics. Springer,
5560 2000.
5561 [Shi96] H. Shiga, “Equivariant de Rham cohomology of homogeneous spaces,” J. Pure Appl.
5562 Algebra, vol. 106, no. 2, pp. 173–183, 1996.
 Bibliography 209

5563 [Sie77] L. C. Siebenmann, “Topological manifolds,” in Foundational Essays on Topological

5564 Manifolds Smoothings, and Triangulations, R. C. Kirby and L. C. Siebenmann, Eds.,
5565 vol. 88, Princeton: Princeton University Press and University of Tokyo Press, 1977,
5566 ch. Annex C, pp. 307–337.
5567 [Sin93] W. Singhof, “On the topology of double coset manifolds,” Math. Ann., vol. 297, no.
5568 1, pp. 133–146, 1993.
5569 [Smi67] L. Smith, “Homological algebra and the Eilenberg–Moore spectral sequence,” Trans.
5570 Amer. Math. Soc., pp. 58–93, 1967.
5571 [ST95] H. Shiga and H. Takahashi, “Remarks on equivariant cohomology of homogeneous
5572 spaces,” Technological University of Nagaoka, Technical report 17, May 1995, pp. 7–
5573 9. [Online]. Available: [Link]
5574 [Link].
5575 [Ste02] Z. Stepień, “On formality of a class of compact homogeneous spaces,” Geom. Dedi-
5576 cata, vol. 93, no. 1, pp. 37–45, 2002.
5577 [Ste51] N. E. Steenrod, The topology of fibre bundles, ser. Princeton Mathematical Series. Prince-
5578 ton University Press, 1951, vol. 14.
5579 [Sti41] E. Stiefel, “Über eine Beziehung zwischen geschlossenen Lie’schen Gruppen und
5580 diskontinuierlichen bewegungsgruppen euklidischer räume und ihre anwendung
5581 auf die aufzählung der einfachen Lie’schen gruppen,” German, Comment. Math.
5582 Helv., vol. 14, no. 1, pp. 350–380, 1941.
5583 [Sul77] D. Sullivan, “Infinitesimal computations in topology,” Publications Mathématiques de
5584 l’IHÉS, vol. 47, no. 1, pp. 269–331, 1977.
5585 [Tay] J. Taylor, Is this characterization of p´1q-eigenspaces of the Weyl group of E6 known?
5586 MathOverflow answer. [Online]. Available: [Link]
5587 [Ter01] S. Terzić, “Cohomology with real coefficients of generalized symmetric spaces,” Fun-
5588 dam. Prikl. Mat, vol. 7, no. 1, pp. 131–157, 2001, in Russian.
5589 [Tu10] L. W. Tu, “Computing characteristic numbers using fixed points,” in A celebration of
5590 the mathematical legacy of Raoul Bott, P. R. Kotiuga, Ed., ser. C. R. M. Proceedings and
5591 Lecture Notes, Centre de Recherches Mathématiques Montréal, vol. 50, Amer. Math.
5592 Soc., 2010, pp. 185–206.
5593 [VA67] B. L. Van der Waerden and E. Artin, Algebra II. Springer, 1967.
5594 [Ver79] A. Verona, “Triangulation of stratified fibre bundles,” Manuscripta Math., vol. 30, no.
5595 4, pp. 425–445, 1979.
5596 [War71] F. W. Warner, Foundations of differentiable manifolds and Lie groups, ser. Grad. Texts in
5597 Math. Springer, 1971, vol. 94.
5598 [Wei35] A. Weil, “Demonstration topologique d’un théoreme fondamental de Cartan,” C. R.
5599 Acad. Sci. Paris, vol. 200, pp. 518–520, 1935.
5600 [Wik14] Wikipedia, Eckmann–Hilton argument — Wikipedia, The Free Encyclopedia, [accessed
5601 18-March-2015], 2014. [Online]. Available: http : / / en . wikipedia . org / w / index .
5602 php?title=Eckmann--Hilton_argument&oldid=631784085.
5603 [Wol67] J. A. Wolf, Spaces of constant curvature. New York: McGraw–Hill, 1967.
 Bibliography 210

5604 [Yan63] C.-T. Yang, “The triangulability of the orbit space of a differentiable transformation
5605 group,” Bull. Amer. Math. Soc., vol. 69, no. 3, pp. 405–408, 1963.
5606 [Zee58] E. C. Zeeman, “A note on a theorem of Armand Borel,” in Mathematical Proceedings
5607 of the Cambridge Philosophical Society, Cambridge Univ. Press, vol. 54, 1958, pp. 396–
5608 398.