Lecture Notes

Download as pdf or txt
Download as pdf or txt
You are on page 1of 257

Group Cohomology

Sommersemester 2019
Universität Regensburg

Clara Löh

1 1 1

1
1
1
1

1 1
1 1

1
1 1 1
1 1
1

1 1 1
1

1 1 1 1

1 1 1 1

1 1 1 1

1 1 1
1

1 1
1 1 1 1

1 1
1 1

1
1

1
1

1 1 1

1
Version of July 30, 2019
[email protected]
Fakultät für Mathematik, Universität Regensburg, 93040 Regensburg
©Clara Löh, 2019
Contents

Guide to the Literature vii

0 Introduction 1

1 The basic view 5


1.1 Foundations: The group ring 6
1.1.1 The group ring 6
1.1.2 Modules over the group ring 7
1.1.3 The domain categories for group (co)homology 10
1.2 The basic definition of group (co)homology 11
1.2.1 The simplicial and the bar resolution 11
1.2.2 Group (co)homology 13
1.3 Degree 0: (Co)Invariants 18
1.4 Degree 1: Abelianisation and homomorphisms 20
1.4.1 Homology in degree 1: Abelianisation 20
1.4.2 Cohomology in degree 1: Homomorphisms 24
1.4.3 Application: Hilbert 90 25
1.5 Degree 2: Presentations and extensions 30
1.5.1 Homology in degree 2: Hopf’s formula 30
1.5.2 Cohomology in degree 2: Extensions 32
1.6 Changing the resolution 41
1.6.1 Projective resolutions 41
1.6.2 The fundamental theorem of group (co)homology 44
1.6.3 Example: Finite cyclic groups 47
1.6.4 Example: Free groups 54
iv Contents

1.7 (Co)Homology and subgroups 58


1.7.1 Restriction and (co)induction 58
1.7.2 The Shapiro lemma 60
1.7.3 Transfer 63

2 The geometric view 67


2.1 Foundations: Geometric group theory 68
2.1.1 Quasi-isometry 68
2.1.2 Amenability 72
2.2 Uniformly finite homology 75
2.2.1 Uniformly finite homology of spaces 75
2.2.2 Uniformly finite homology of groups 79
2.2.3 Application: Ponzi schemes and amenability 80
2.3 Bounded cohomology 86
2.3.1 Bounded cohomology of groups 86
2.3.2 Application: A characterisation of amenability 88
2.3.3 Application: Quasi-morphisms 90
2.3.4 Application: Stable commutator length 95

3 The derived view 99


3.1 Derived functors 100
3.1.1 Axiomatic description 100
3.1.2 A construction 102
3.1.3 The two sides of Tor 107
3.1.4 Group homology as derived functor 108
3.1.5 Group cohomology as derived functor 110
3.1.6 The derived category 111
3.2 The Hochschild-Serre spectral sequence 117
3.2.1 Terminology for spectral sequences 118
3.2.2 Classical spectral sequences 122
3.2.3 The spectral sequence of a group extension 123
3.2.4 A proof of Hopf’s formula 129
3.2.5 Universal coefficients and products 132

4 The topological view 141


4.1 Classifying spaces 142
4.1.1 The standard simplicial model 142
4.1.2 Changing the classifying space 144
4.1.3 Examples of classifying spaces 149
4.1.4 Group (co)homology via classifying spaces 153
4.2 Finiteness conditions 157
4.2.1 Cohomological dimension 158
4.2.2 Finite type 162
4.3 Application: Free actions on spheres 164
4.3.1 From sphere actions to periodic resolutions 165
4.3.2 From periodic resolutions to Sylow subgroups 167
Contents v

A Appendix A.1
A.1 Amalgamated free products A.3
A.1.1 The free group of rank 2 A.3
A.1.2 Free products of groups A.4
A.1.3 Amalgamated free products of groups A.5
A.1.4 Free groups A.6
A.2 Some homological algebra A.7
A.3 Homotopy theory of CW-complexes A.11

B Exercise Sheets B.1

C Etudes C.1

D General Information D.1

Bibliography E.1

Dictionary E.8

Symbols E.17

Index E.21
vi Contents
Guide to the Literature

This course will not follow a single source and there are many books that
cover the standard topics (all with their own advantages and disadvantages).
Therefore, you should individually compose your own favourite selection of
books.

Group Cohomology and Homological Algebra


• K.S. Brown. Cohomology of Groups, Graduate Texts in Mathemat-
ics, 82, Springer, 1982.
• S.I. Gelfand, Y.I. Manin. Methods of Homological Algebra, Springer
Monographs in Mathematics, second edition, Springer, 2003.
• R. Geoghegan. Topological Methods in Group Theory, Graduate Texts
in Mathematics, 143, Springer, 2008.
• P.J. Hilton, U. Stammbach. A Course in Homological Algebra, Graduate
Texts in Mathematics, 4, second edition, Springer, 1996.
• C. Weibel. An Introduction to Homological Algebra, Cambridge Studies
in Advanced Mathematics, 38, Cambridge University Press, 1995.

Group Theory
• C. Druţu, M. Kapovich. Geometric Group Theory, Colloquium Publi-
cations, 63, American Mathematical Society, 2018.
• C. Löh. Geometric Group Theory. An Introduction, Universitext, Sprin-
ger, 2018.
Errata: http://www.mathematik.uni-r.de/loeh/ggt book/errata.pdf
viii Guide to the Literature

• R.C. Lyndon, P.E. Schupp. Combinatorial Group Theory, Classics in


Mathematics, Springer, 2001.
• J.J. Rotman. An Introduction to the Theory of Groups, Graduate Texts
in Mathematics, 148, third edition, Springer, 1994.

Geometry and Topology


• M. Aguilar, S. Gitler, C. Prieto. Algebraic Topology from a Homotopical
Viewpoint, Springer, 2002.
• M.W. Davis. The Geometry and Topology of Coxeter Groups, Lon-
don Mathematical Society Monographs, 32, Princeton University Press,
2008.
• A. Dold. Lectures on Algebraic Topology, Springer, 1980.
• S. Friedl. Algebraic Topology I, II, III, . . . , lecture notes, Universität
Regensburg, 2016–. . . .
• A. Hatcher. Algebraic Topology, Cambridge University Press, 2002.
http://www.math.cornell.edu/∼hatcher/AT/ATpage.html
• W.S. Massey. Algebraic Topology: an Introduction, seventh edition,
Springer, 1989.
• P. May. A Concise Course in Algebraic Topology, University of Chicago
Press, 1999.
• T. tom Dieck. Algebraic Topology, European Mathematical Society,
2008.

Category Theory
• M. Brandenburg. Einführung in die Kategorientheorie: Mit ausführli-
chen Erklärungen und zahlreichen Beispielen, Springer Spektrum, 2015.
• S. MacLane. Categories for the Working Mathematician, second edition,
Springer, 1998.
• B. Pierce. Basic Category Theory for Computer Scientists, Foundations
of Computing, MIT University Press, 1991.
• E. Riehl. Category Theory in Context, Aurora: Dover Modern Math
Originals, 2016.
0
Introduction

This course provides an introduction to group cohomology and its applica-


tions. More precisely, we will investigate different views on group cohomology
and group homology from algebra, geometry, and topology as well as their
applications.

What is group cohomology?


Group cohomology and group homology are theories that convert groups
and modules over group rings (the so-called “coefficients”) into graded
groups/graded algebras. In particular, for each group G, each G-module A,
and each n ∈ N, we obtain Abelian groups

• Hn (G; A), the homology of G in degree n with coefficients in A,

• H n (G; A), the cohomology of G in degree n with coefficients in A.

More precisely, group homology is obtained by deriving the tensor product


functor over the group ring; group cohomology is obtained by deriving the
Hom-functor over the group ring. While this description is concise and valu-
able for conceptual (and some computational) reasons, it hides the versatility
of these theories and the interaction of group (co)homology with other fields.
Indeed, group (co)homology can be described from different points of view;
the most prominent ones are:

• the basic view (via the standard resolutions),

• the derived view (via derived functors),


2 0. Introduction

• the topological view (via classifying spaces),


• the geometric view (via other geometric structures).
Moreover, different choices of coefficient modules exhibit different invariance
properties.
Because all these descriptions lead to the same theory, group (co)homology
encodes many interesting connections between algebraic and geometric or
topological properties of groups.
The art is then to find coefficients and suitable degrees that are well
adapted to the target application.

Why group cohomology?

Classical applications of group (co)homology include the following:


Algebra
• generalisations of Hilbert 90 in Galois theory
• classification of group extensions with Abelian kernel; classification of
central extensions
• generalisations of the group-theoretic transfer
• generalisations of finiteness properties of groups (such as finiteness,
finite generation, finite presentability, . . . )
• computations of stable commutator length
• ...
Geometry
• characterisations of amenable groups
• rigidity properties of (non)amenable groups
• study of quasi-morphisms on groups
• differentiation between different dynamical systems of groups
• ...
Topology
• characterisation of finite groups that admit free actions on spheres
• rigidity results in topology and geometry
• ...
3

Overview of this Course

In this course, we will develop the basics of group (co)homology. We will


begin with the basic view (via standard resolutions) and group (co)homology
in low degrees. We will then pursue a more geometric direction, with a focus
on amenability. Afterwards, we will develop the derived view and discuss some
general homological techniques. Finally, we will briefly study the topological
view and the description of group (co)homology via classifying spaces and
topological group actions. I will try to keep everything rather elementary.
Basics in category theory and homological algebra will be recapitulated when
necessary.
Moreover, we will discuss some of the applications along the way. The
goal is to prepare the participants to access more advanced topics in group
(co)homology and related fields and its applications to other fields.
Study note. These lecture notes document the topics covered in the course
(as well as some additional optional material). However, these lecture notes
are not meant to repace attending the lectures or the exercise classes!
Furthermore, this course will only treat the very beginning of this vast
subject. It is therefore recommended to consult other sources (books and
research articles!) for further information on this field.
References of the form “Satz I.6.4.11”, “Satz II.2.4.33”, “Satz III.2.2.25”,
“Satz IV.2.2.4”, or “Corollary AT.1.3.25” point to the corresponding loca-
tions in the lecture notes for Linear Algebra I/II, Algebra, Commutative
Algebra, Algebraic Topology in previuos semesters:

http://www.mathematik.uni-r.de/loeh/teaching/linalg1 ws1617/lecture notes.pdf


http://www.mathematik.uni-r.de/loeh/teaching/linalg2 ss17/lecture notes.pdf
http://www.mathematik.uni-r.de/loeh/teaching/algebra ws1718/lecture notes.pdf

http://www.mathematik.uni-r.de/loeh/teaching/calgebra ss18/lecture notes.pdf


http://www.mathematik.uni-r.de/loeh/teaching/topologie1 ws1819/lecture notes.pdf

Literature exercise. Where in the math library can you find books on group
cohomology, homological algebra, geometry, algebraic topology, category the-
ory?
Convention. The set N of natural numbers contains 0. All rings are unital
and associative (but very often not commutative). We write R Mod for the
category of left R-modules.
4 0. Introduction
1
The basic view

The most straightforward definition of group (co)homology is based on the


standard resolutions of the trivial module Z over the group ring. This view
on group (co)homology already allows for interesting applications in algebra
(e.g., Galois theory and group extensions).
However, this view is not suitable for most concrete calculations. The
fundamental insight is that group (co)homology is flexible in the sense that
different resolutions will lead to the same theory. We will use this for some
simple computations and for a first discussion of the Shapiro lemma and its
consequences.
The change of resolution will be exploited systematically in the following
chapters.

Overview of this chapter.

1.1 Foundations: The group ring 6


1.2 The basic definition of group (co)homology 11
1.3 Degree 0: (Co)Invariants 18
1.4 Degree 1: Abelianisation and homomorphisms 20
1.5 Degree 2: Presentations and extensions 30
1.6 Changing the resolution 41
1.7 (Co)Homology and subgroups 58

Running example. cyclic groups, free groups


6 1. The basic view

1.1 Foundations: The group ring


The basic algebraic object in the context of group (co)homology is the group
ring (and its module category).

1.1.1 The group ring


By definition, the group ring of a group G is an extension of the ring Z with
new units coming from the group G.
Definition 1.1.1 (group ring). Let G be a group. The (integral) group ring of G
is the ring ZG (sometimes also denoted by Z[G] to avoid misunderstandings)
L
• whose underlying Abelian group is the free Z-module g∈G Z, freely
generated by G (we denote the basis element corresponding to g ∈ G
simply by g),
• and whose multiplication is the Z-linear extension of composition in G,
i.e.:

· : ZG × ZG −→ ZG
X X  XX
ag · g, bg · g 7−→ ah · bh−1 ·g · g
g∈G g∈G g∈G h∈G

(where all sums are “finite”).


Remark 1.1.2. Let G be a group. A straightforward calculation shows that
ZG is indeed a ring (check!). Moreover, ZG is unital, with multiplicative
unit 1 · e, where e denotes the neutral element of G (check!).
Example 1.1.3 (group rings).
• The group ring of “the” trivial group 1 is just Z[1] ∼
=Ring Z.
• The group ring Z[Z] of the additive group Z is isomorphic to Z[t, t−1 ],
the ring of Laurent polynomials over Z (check!).
• Let n ∈ N>0 . Then we have Z[Z/n] ∼ =Ring Z[t]/(tn − 1) (check!). In
general, the ring Z[Z/n] is not isomorphic to the subring Z[e2·π·i/n ]
of C (check!).
• In general, group rings are not commutative. In fact, a group ring ZG
is commutative if and only if the group G is Abelian (check!). Hence,
for example, the group ring Z[S3 ] of the symmetric group S3 is not
commutative.
1.1. Foundations: The group ring 7

Caveat 1.1.4 (notation in group rings). When working with elements in group
rings, some care is required. For example, the term 4 · 2 in Z[Z] might be
interpreted in the following different(!) ways:
• the product of 4 times the ring unit and 2 times the ring unit, or
• 4 times the group element 2.
We will circumvent this issue in Z[Z], by using the notation “t” for a gen-
erator of the additive group Z and viewing the infinite cyclic group Z as
multiplicative group. Using this convention, the first interpretation would be
written as 4 · 2 (which equals 8) and the second interpretation would be writ-
ten as 4 · t2 . Similarly, also in group rings over other groups, we will try to
avoid ambiguous notation.
As with every new construction, we should capture the essence of the
construction in a universal property:
Proposition 1.1.5 (group ring, universal property). Let G be a group. Then
the group ring ZG, together with the canonical inclusion map i : G −→ ZG
(as standard basis) has the following universal property: For every ring R
and every group homomorphism f : G −→ R× , there exists a unique ring
homomorphism Zf : ZG −→ R with Zf ◦ i = f .

G
f
/ R× incl 6/ R

i
∃! Zf

ZG

Proof. This is a straightforward calculation (check!).


Outlook 1.1.6 (Kaplansky conjecture). The ring structure of group rings is
not well understood in full generality. For example, the following versions of
the Kaplansky conjectures are still open: Let G be a torsion-free group.
• Then the group ring ZG is a domain (?!).
• The group ring ZG does not contain non-trivial idempotents (?!).
(I.e., if x ∈ ZG with x2 = x, then x = 1 or x = 0).
However, a positive solution is known for many special cases of groups [17,
65, 21][55, Chapter 10] (such proofs often use input from functional analysis
or geometry) and no counterexamples are known.

1.1.2 Modules over the group ring

If G is a group, then (left) ZG-modules correspond to Abelian groups that


carry a G-representation; more explicitly, if A is a (left) ZG-module, then, in
addition to the usual additive structure, we have a map · : G × A −→ A with
8 1. The basic view

e·a=a
g · (a + b) = g · a + g · b
(g · h) · a = g · (h · a)

for all g, h ∈ G and all a, b ∈ A. By default, ZG-modules mean left ZG-


modules. Because the group ring is in general non-commutative, we have to
be careful when taking tensor products. However, the group ring carries an
involution, which simplifies matters:

Remark 1.1.7 (the involution: left vs. right modules). Let G be a group. Then
taking inverses is an involution on G. Hence, if A is a left ZG-module, then

A × G −→ A
(a, g) 7−→ g −1 · a

allows to view A as a right module (check!), and vice versa. We will denote
this right ZG-module by Inv A.

Definition 1.1.8 (tensor product and Hom-modules over ZG). Let G be a group
and let A, B be left ZG-modules. Then we write

A ⊗G B := Inv(A) ⊗ZG B ∈ Ob(Z Mod)


HomG (A, B) := ZG Hom(A, B) ∈ Ob(Z Mod).

Example 1.1.9 (G-modules). Let G be a group.

• If Z is a Z-module, then the trivial G-action

G × Z −→ Z
(g, z) 7−→ z

on Z turns Z into a so-called trivial (left) ZG-module. Usually, we will


regard Z as a trivial ZG-module.

• The group ring ZG is a ZG-ZG-bimodule with respect to the action


of G on the Z-basis G of ZG by left/right translation.

• Let `∞ (G, R) := {f : G → R | supg∈G |f (g)| < ∞}. Then `∞ (G, R) is


a Z-module with respect to pointwise addition/scalar multiplication.
Moreover, `∞ (G, R) is a left ZG-module via translation:

G × `∞ (G, R) −→ `∞ (G, R)

(g, f ) 7−→ h 7→ f (h · g) .

Similarly, also `∞ (G, Z), `2 (G, R), and `2 (G, C) are (left) ZG-modules.
1.1. Foundations: The group ring 9

• If X is a topological space equipped with a continuous (left) G-action,


then this action turns the singular chain modules Cn (X) into (left)
ZG-modules.

Definition 1.1.10 (invariants, coinvariants). Let G be a group and let A be a


(left) ZG-module. Then we introduce the following Z-modules:

• The invariants of A are: AG := {a ∈ A | ∀g∈G g · a = a}.

• The coinvariants of A are: AG := A/ SpanZ {g · a − a | g ∈ G, a ∈ A}.

These definitions are compatible with ZG-linear maps (check!) and thus ex-
tend to (covariant) functors

· G: ZG Mod −→ Z Mod
· G: ZG Mod −→ Z Mod .

Study note (categories and functors). For now, we will only need very basic
notions from category theory: categories, functors, natural transformations
(Chapter IV.1). If you don’t know these terms yet, then this is not an issue.

Remark 1.1.11 (an alternative description of (co)invariants). Let G be a group


and let A be a (left) ZG-module. Then (where Z carries the trivial ZG-module
structure)

AG −→ HomG (Z, A)
a 7−→ (n 7→ n · a)
AG −→ A ⊗G Z
[a] 7−→ a ⊗ 1

are Z-isomorphisms (check!). These isomorphisms lead to natural isomor-


phisms between the invariants functor and HomG (Z, · ) as well as between
the coinvariants functor and · ⊗G Z (check!).

Study note. Which properties do invariants and coinvariants inherit from the
description as Hom- and tensor functors (Remark 1.1.11), respectively?

Example 1.1.12. Let G be a group.


P
• If G is finite, then (ZG)G ∼
=Z Z, generated by g∈G g (check!).

• If G is infinite, then (ZG)G ∼


=Z {0} and `2 (G, C)G ∼
=Z {0} (check!).

• Similarly, we have `∞ (G, Z)G ∼


=Z Z (check!).

• Moreover, (ZG)G ∼
=Z ZG ⊗G Z ∼
=Z Z.
10 1. The basic view

1.1.3 The domain categories for group (co)homology

As indicated in the introduction, group (co)homology has as arguments a


group and a module over this group. In order to work efficiently with this
setup, we introduce the categories GroupMod and GroupMod* as domain for
group homology and group cohomology, respectively.
Definition 1.1.13 (GroupMod, GroupMod* ). The category GroupMod consists
of the following data:
• objects: The class of all pairs (G, A), where G is a group and A is a
(left) ZG-module.
• morphisms: The set of morphisms from (G, A) to (H, B) is the set of
pairs (ϕ, Φ), where
• ϕ : G −→ H is a group homomorphism, and
• Φ : A −→ ϕ∗ B is a ZG-module homomorphism; here, ϕ∗ B is the
ZG-module whose underlying additive group coincides with B and
whose ZG-structure is given by

G × B −→ B
(g, b) 7−→ ϕ(g) · b.

• compositions: The composition of morphisms is defined by composing


both components separately (this is well-defined in the second compo-
nent; check!).
The category GroupMod* consists of the following data:
• objects: We set Ob(GroupMod* ) := Ob(GroupMod).
• morphisms: The set of morphisms from (G, A) to (H, B) is the set of
pairs (ϕ, Φ), where
• ϕ : G −→ H is a group homomorphism, and
• Φ : ϕ∗ B −→ A is a ZG-module homomorphism.
• compositions: The composition of morphisms is defined by ordinary
composition in the first component and reverse composition in the sec-
ond component.
The following, simple, example lies at the heart of group (co)homology.
Example 1.1.14 (invariants and coinvariants). Taking coinvariants can be
viewed as a functor GroupMod −→ Z Mod; taking invariants can be viewed as
a (contravariant) functor GroupMod* −→ Z Mod (check!).
1.2. The basic definition of group (co)homology 11

1.2 The basic definition of group (co)homology


We will now give a hands-on construction of group (co)homology. The ad-
vantage of this definition is that it immediately gives a functorial description
of group (co)homology in both arguments. The disadvantage is that it is not
immediately clear why this results in a reasonable theory of anything.

1.2.1 The simplicial and the bar resolution


The simplicial resolution of a group G is based on a simple geometric idea: We
view the elements of G as the vertices of a (potentially infinite-dimensional)
simplex and then take the simplicial chain complex of all (finite ordered)
simplices of this huge simplex. Algebraically, this can be concisely formulated
as follows (the geometric counterpart is introduced in Definition 4.1.1):
Definition 1.2.1 (the simplicial resolution). Let G be a group. Then the sim-
plicial resolution C∗ (G) of the group G is the N-indexed ZG-chain complex
defined by:
• For n ∈ N, we consider the chain module
M
Cn (G) := Z.
Gn+1

The Z-basis elements of Cn (G) are denoted as the corresponding tuples


in Gn+1 .
• For n ∈ N, the G-action on Cn (G) is induced from the diagonal action

G × Gn+1 −→ Gn+1

g, (g0 , . . . , gn ) 7−→ (g · g0 , . . . , g · gn )

on the Z-basis Gn+1 .


• For all n ∈ N>0 , we define the boundary operator ∂n as the Z-linear
map with (which turns out to be ZG-linear)

∂n : Cn (G) −→ Cn−1 (G)


n
X
Gn+1 3 (g0 , . . . , gn ) 7−→ (−1)j · (g0 , . . . , gbj , . . . , gn ).
j=0

Here, (g0 , . . . , gbj , . . . , gn ) is a shorthand for (g0 , . . . , gj−1 , gj+1 , . . . , gn ).


Moreover, we set ∂0 := 0 : C0 −→ 0.
12 1. The basic view

The simplicial augmentation is defined as the unique Z-linear map with

ε : C0 (G) = ZG −→ Z
G 3 g 7−→ 1.

Remark 1.2.2. Let G be a group. In order to check that C∗ (G) is a chain


complex, we need to verify that ∂n ◦ ∂n+1 = 0 for all n ∈ N>0 . This is a
straightforward computation (check!), as in the case of the singular chain
complex (Proposition AT.4.1.3) or the chain complex associated with a sim-
plicial set. Moreover, all ∂n are ZG-homomorphisms and we have ε ◦ ∂1 = 0.
For instance, we prove that ∂1 ◦ ∂2 = 0: It suffices to check this relation
on the Z-basis G3 of C2 (G). For all (g0 , g1 , g2 ) ∈ G2 , we have
 
∂1 ◦ ∂2 (g0 , g1 , g2 ) = ∂1 (g1 , g2 ) − (g0 , g2 ) + (g0 , g1 )
  
= ∂1 (g1 , g2 ) − ∂1 (g0 , g2 ) + ∂1 (g0 , g1 )
= g2 − g1 − g2 + g0 + g1 − g0
= 0,

as claimed.

Study note (homological algebra). For a while, we will only need basic notions
from homological algebra (chain complexes, homology of chain complexes,
tensor products, chain maps, chain homotopy); this might be a good oppor-
tunity to refresh your background on homological algebra, or to acquire it
for the first time (Chapter IV.5, Appendix AT.A.6).

While the simplicial resolution is very symmetric and straightforward, it


does have the disadvantage that we only have a canonical Z-basis of the
chain modules (which are ZG-modules), but we will want to apply ZG-
constructions to this complex. Therefore, we will introduce the bar resolution,
which comes with a canonical ZG-basis in every degree (while keeping a rea-
sonable description of the boundary operator):

Definition 1.2.3 (the bar resolution). Let G be a group. Then the bar resolu-
tion C ∗ (G) of G is the N-indexed ZG-chain complex defined by:

• For n ∈ N, we consider the ZG-module


M
C n (G) := ZG.
Gn

The ZG-basis element of C n (G) corresponding to (g1 , . . . , gn ) ∈ Gn is


denoted by [g1 | . . . |gn ]. The unique element of G0 is denoted by [].

• For all n ∈ N>0 , we define the boundary operator ∂ n as the ZG-linear


map with
1.2. The basic definition of group (co)homology 13

∂ n : C n (G) −→ C n−1 (G)


[g1 | . . . |gn ] 7−→ g1 · [g2 | . . . |gn ]
n−1
X
+ (−1)j · [g1 | . . . |gj−1 |gj · gj+1 |gj+2 | . . . |gn ]
j=1

+ (−1)n · [g1 | . . . |gn−1 ].

Remark 1.2.4 (simplicial vs. bar resolution). Let G be a group. Then the maps

Cn (G) −→ C n (G)
Gn+1
3 (g0 , . . . , gn ) 7−→ g0 · [g0−1 · g1 | . . . |gn−1
−1
· gn ] (Z-linear extension)
C n (G) −→ Cn (G)
[g1 | . . . |gn ] 7−→ (e, g1 , g1 · g2 , . . . , g1 · g2 · · · · · gn ) (ZG-linear extension)

for n ∈ N form mutually inverse isomorphisms C(G) ∼ =ZG C(G) of ZG-chain


complexes (check!); hence, both resolutions are different views on the same
thing. The bar resolution has the advantage that it is easier to read off its
coinvariants and related constructions.
A particularly convenient aspect of these constructions is that they are
compatible with group homomorphisms (without passing to the homotopy
category of the category of chain complexes):
Proposition 1.2.5 (group homomorphisms and the standard resolutions). Let
ϕ : G −→ H be a group homomorphism.
1. Then C∗ (ϕ) : C∗ (G) −→ ϕ∗ C∗ (H), defined by

Cn (ϕ) : Cn (G) −→ ϕ∗ Cn (H)



Gn+1 3 (g0 , . . . , gn ) 7−→ ϕ(g0 ), . . . , ϕ(gn )

for all n ∈ N, is a ZG-chain map.


2. Similarly, C ∗ (ϕ) : C ∗ (G) −→ ϕ∗ C ∗ (H), defined by

C n (ϕ) : C n (G) −→ ϕ∗ C n (H)


 
[g1 | · · · | gn ] 7−→ ϕ(g1 ) | · · · | ϕ(gn )

for all n ∈ N, is a ZG-chain map.


Proof. This is a straightforward calculation (check!).

1.2.2 Group (co)homology


Group cohomology is the cohomology of the invariants of the simplicial reso-
lution; group homology is the homology of the coinvariants of the simplicial
14 1. The basic view

resolution. More generally, using Hom-functors and tensor products, we can


decorate these constructions with coefficient modules.

Definition 1.2.6 (group homology). Let G be a group and let A be a (left)


ZG-module.

• Then we write C∗ (G; A) := C∗ (G)⊗G A ∈ Z Ch for the standard complex


of G with coefficients in A (Remark 1.2.7).

• For n ∈ N, we define group homology of G with coefficients in A by



 ker ∂n ⊗G idA : Cn (G; A) → Cn−1 (G; A)
Hn (G; A) := Hn C∗ (G; A) := .
im ∂n+1 ⊗G idA : Cn+1 (G; A) → Cn (G; A)

Let (ϕ, Φ) : (G, A) −→ (H, B) be a morphism in GroupMod.

• We write C∗ (ϕ; Φ) := C∗ (ϕ) ⊗ Φ for the composition

C∗ (G) ⊗G A
C∗ (ϕ)⊗G Φ
/ ϕ∗ C∗ (H) ⊗G ϕ∗ B can. proj./ C∗ (H) ⊗H B

of Z-chain maps.

• For n ∈ N, we then set (which is a Z-linear map)



Hn (ϕ; Φ) := Hn C∗ (ϕ; Φ) : Hn (G; A) −→ Hn (H; B).

We will also write H∗ (G; A) for the sequence (Hn (G; A))n∈N and H∗ (ϕ; Φ)
for the sequence (Hn (ϕ; Φ))n∈N .

Remark 1.2.7 (tensor product of a chain complex with a module). Let G be a


group and let A be a (left) ZG-module. We recall the definition of the functor

· ⊗G A : ZG Ch −→ Z Ch .

Let C∗ be a ZG-chain complex. Then C∗ ⊗G A is the Z-chain complex con-


sisting of the chain modules Cn ⊗G A and the boundary maps

∂n ⊗G idA : Cn ⊗G A −→ Cn−1 ⊗G A
x ⊗ a 7−→ ∂n (x) ⊗ a.

If f∗ : C∗ −→ D∗ is a ZG-chain map, then the sequence f∗ ⊗G idA := (fn ⊗G


idA )n∈N is a Z-chain map C∗ ⊗G A −→ D∗ ⊗G A (check!).

Example 1.2.8 (homology of the trivial group). Let 1 denote “the” trivial
group. Then Z[1] ∼
=Ring Z, and we will formulate everything in terms of Z-
modules (instead of Z[1]-modules). If A is a Z-module, then
1.2. The basic definition of group (co)homology 15
(
A if n = 0
Hn (1; A) ∼
=Z
0 if n > 0

for all n ∈ N. This follows from the simple observation that the Z-chain
complex C∗ (1) is isomorphic to the Z-chain complex (check!)

degree 2 1 0

··· /Z /Z /Z /Z
idZ 0 idZ 0

Hence, C∗ (1; A) = C∗ (1) ⊗1 A is isomorphic to

degree 2 1 0

··· /A /A /A /A
idA 0 idA 0

This chain complex has the homology stated above (check!).


Dually, group cohomology is defined in terms of cohomology of cochain
complexes. This works in the same way – the only difference between chain
complexes and cochain complexes is that the boundary maps increase the de-
gree, instead of decreasing it. We denote the categories of cochain complexes
by Ch* instead of Ch and also denote cohomology by an upper index/star
instead of a lower index/star.
Definition 1.2.9 (group cohomology). Let G be a group and let A be a (left)
ZG-module.

• Then we write C ∗ (G; A) := HomG C∗ (G), A ∈ Z Ch* for the standard
cochain complex of G with coefficients in A (Remark 1.2.10).
• For n ∈ N, we define group cohomology of G with coefficients in A by

 ker HomG (∂n+1 , A) : C n (G; A) → C n+1 (G; A)
H n (G; A) := H n C ∗ (G; A) =  .
im HomG (∂n , A) : C n−1 (G; A) → C n (G; A)

Let (ϕ, Φ) : (G, A) −→ (H, B) be a morphism in GroupMod* .


• We write C ∗ (ϕ; Φ) := HomG (C∗ (ϕ), Φ) for the composition

can. incl. / HomG (C∗ (ϕ),Φ) 


HomH C∗ (H), B) HomG ϕ∗ C∗ (H), ϕ∗ B / HomG C∗ (G), A

of Z-cochain maps.
• For n ∈ N, we then set (which is a Z-linear map)

H n (ϕ; Φ) := H n C ∗ (ϕ; Φ) : H n (H; B) −→ H n (G; A).
16 1. The basic view

We will also write H ∗ (G; A) for the sequence (H n (G; A))n∈N and H ∗ (ϕ; Φ)
for the sequence (H n (ϕ; Φ))n∈N .

Remark 1.2.10 (Hom-cochain complexes). Let G be a group and let A be a


(left) ZG-module. We recall the definition of the (contravariant) functor

HomG ( · , A) : ZG Ch −→ Z Ch* .

Let C∗ be a ZG-chain complex. Then HomG (C∗ , A) is the Z-cochain complex


consisting of the cochain modules HomG (Cn , A) and the coboundary maps

(−1)n+1 · HomG (∂n+1 , A) : HomG (Cn , A) −→ HomG (Cn+1 , A)



f 7−→ x 7→ (−1)n+1 · f (∂n+1 (x)) ;

the sign does not affect the corresponding cohomology, but will later result in
good sign conventions on (co)chain complexes. If f∗ : C∗ −→ D∗ is a ZG-chain
map, then the sequence HomG (f∗ , A) := (HomG (fn , A))n∈N is a Z-cochain
map HomG (D∗ , A) −→ HomG (C∗ , A) (check!).

Remark 1.2.11 (functoriality of group (co)homology). The functors (check!)

C∗ ( · ; · ) : GroupMod −→ Z Ch (covariant)
∗ * *
C ( · ; · ) : GroupMod −→ Z Ch (contravariant)

in combination with the (co)homology functors on (co)chain complexes yield


for each n ∈ N group (co)homology functors of the following types:

Hn : GroupMod −→ Z Mod (covariant)


n *
H : GroupMod −→ Z Mod (contravariant)

Caveat 1.2.12 (lechts und rinks). There are several, different, conventions for
group (co)homology in use. Indeed, instead of taking left modules as coeffi-
cients, one could also work with right modules. Because the group ring carries
a canonical involution (Remark 1.1.7), we can always canonically translate
between these different conventions and all conventions essentially lead to the
same theory. However, when working with formulas from the literature, one
should always carefully check which conventions are in place in the respective
sources.

Example 1.2.13 (cohomology of the trivial group). We continue the considera-


tions from Example 1.2.8: Let 1 denote “the” trivial group; then Z[1] ∼
=Ring Z.
If A is a Z-module, then
(
A if n = 0
H (1; A) ∼
n
=Z
0 if n > 0
1.2. The basic definition of group (co)homology 17

for all n ∈ N. Again, using that the Z-chain complex C∗ (1) is isomorphic to
the Z-chain complex (check!)

degree 2 1 0

··· /Z /Z /Z /Z
idZ 0 idZ 0

we obtain that C ∗ (1; A) = Hom1 (C∗ (1), A) is isomorphic to

degree 2 1 0
idA 0 idA 0
··· o Ao Ao Ao A

This cochain complex has the claimed cohomology (check!).

Study note. If you know basic algebraic topology: What do the computa-
tions from Example 1.2.8 and 1.2.13 remind you of? This is part of a deeper
connection of group (co)homology with algebraic topology. We will explore
this connection in Chapter 4.

Remark 1.2.14 (group (co)homology via bar resolution). Because the simplicial
resolution and the bar resolution are (naturally) isomorphic (Remark 1.2.4),
we can also use the bar resolution to compute group homology and group
cohomology: If G is a group and A is a (left) ZG-module, then the chain
isomorphisms from Remark 1.2.4 induce chain isomorphisms

C∗ (G; A) = C∗ (G) ⊗G A ∼
=Z Ch C ∗ (G) ⊗G A =: C ∗ (G; A)
  ∗
C (G; A) = HomG C∗ (G), A ∼

= * HomG C ∗ (G), A =: C (G; A).
Z Ch

Moreover, these isomorphisms are natural (on GroupMod and GroupMod* ,


respectively).
This is particularly helpful if we work with constant coefficients, because
the (co)invariants are easier to read off from the bar resolution than from the
simplicial resolution.

In general, computing group (co)homology from the simplicial or the bar


resolution is not feasible (the chain complexes can be huge even though the
homology might be very small). In the following sections, we will see how
this basic description of group (co)homology allows to establish connections
with more classical invariants/problems. However, for explicit computations,
we will usually rely on different descriptions of group (co)homology, which
we will develop later.
By definition, (co)homology is the quotient of the (co)cycles by the
(co)boundaries. Therefore, we should expect that group (co)homology also
carries geometric meaning. We will discuss this in more detail in Chapter 2
and Chapter 4.
18 1. The basic view

1.3 Degree 0: (Co)Invariants


As a warm-up, we compute group (co)homology in degree 0.
Theorem 1.3.1 (group (co)homology in degree 0). Let G be a group and let A
be a ZG-module. Then there are canonical isomorphisms

H0 (G; A) ∼
= Z AG
H (G; A) ∼
0
= Z AG .

More precisely, the functors H0 and H 0 are canonically naturally isomorphic


to the coinvariants functor GroupMod −→ Z Mod and the invariants func-
tor GroupMod* −→ Z Mod, respectively.
Proof. We use the bar complex to compute group (co)homology in degree 0
(Remark 1.2.14). The lower part of the bar complex looks as follows:

degree 1 0
∂1 ∂ 0 =0
C 1 (G) / C 0 (G) /0
L
G ZG ZG

[g]  / g · [] − []

In combination with Remark 1.2.4, we see that the low-degree part


of C∗ (G; A) = C∗ (G) ⊗G A is canonically naturally isomorphic to

degree 1 0
∂ 1 ⊗G idA 0
C 1 (G; A) / C 0 (G; A) /0
L 
G ZG ⊗G A ZG ⊗G A
=Z
=Z


L
G A A
 / g −1 · a − a.
[g] · a

Hence,

H0 (G; A) ∼
=Z C 0 (G; A) / im(∂ 1 ⊗G idA )

=Z A/ SpanZ {g −1 · a − a | g ∈ G, a ∈ A}
= AG .
1.3. Degree 0: (Co)Invariants 19

Dually, the low-degree part of C ∗ (G; A) = HomG (C∗ (G), A) is canonically


naturally isomorphic to

degree 1 0
1 − HomG (∂ 1 ,A) 0 0
C (G; A) o C (G; A) o 0
L 
HomG G ZG, A HomG (ZG, A)

=Z
=Z



Q
G A A

(a − g · a)g∈G o a

Therefore,

H 0 (G; A) ∼
=Z ker HomG (∂ 1 , A)
= {a ∈ A | ∀g∈G a = g · a}
= AG .

Tracing through the explicit descriptions of these isomorphisms shows that


these isomorphisms are natural on GroupMod and GroupMod* , respectively
(check!).

Study note. Write down the isomorphisms constructed in the proof of Theo-
rem 1.3.1 in terms of the simplicial resolution (instead of the bar resolution).

Example 1.3.2 (group (co)homology in degree 0). Let G be a group. Plugging


in the computations from Example 1.1.12, we obtain the following group
(co)homology in degree 0:

=Z Z and H 0 (G; Z) ∼
=Z Z G ∼
• If Z is a trivial G-module, then H0 (G; Z) ∼ =Z
G
Z = Z.

• We have H 0 (G; `∞ (G, Z)) ∼


=Z `∞ (G, Z)G ∼=Z Z (however, the computa-
tion of H0 (G; ` (G, Z)) is more delicate; Remark 2.2.18) and we have

H0 (G; ZG) ∼=Z (ZG)G ∼ =Z Z.

=Z 0 and H 0 (G; `2 (G, C)) ∼


• If G is infinite, then H 0 (G; ZG) ∼ =Z 0.

• If G is finite, then H 0 (G; ZG) ∼


=Z Z and H 0 (G; `2 (G, C)) ∼
=Z C.

In particular: Taking ZG-coefficients or `2 (G; C)-coefficients allows to char-


acterise finiteness of groups in terms of group cohomology.

In fact, group (co)homology is uniquely determined by its values in degree 0


and the fact that we can apply the dimension shifting trick (Corollary 3.1.13).
We will return to this point of view in Chapter 3.
20 1. The basic view

1.4 Degree 1: Abelianisation and homomorphisms


We will now discuss group (co)homology in degree 1. For simplicity, we will
first focus on the case of trivial coefficients. Moreover, we will indicate appli-
cations to group theory and Galois theory.

1.4.1 Homology in degree 1: Abelianisation

Theorem 1.4.1 (group homology in degree 1). Let G be a group. Then (where
Z carries the trivial G-action) there is a canonical isomorphism

H1 (G; Z) ∼
=Z Gab .

More precisely, the functor H1 ( · ; Z) : Group −→ Z Mod is canonically natu-


rally isomorphic to the Abelianisation functor.
Remark 1.4.2 (Abelianisation). Recall that for a group G, the Abelianisa-
tion Gab is the quotient group

Gab := G/[G, G].

Here, [G, G] is the commutator subgroup of G, i.e., the subgroup of G gener-


ated by the set {[x, y] | x, y ∈ G} of all commutators

[x, y] := x · y · x−1 · y −1

in G. By construction, Gab is Abelian; more precisely, the Abelianisation Gab


is the largest Abelian quotient of G. The construction of the Abelianisa-
tion is also compatible with group homomorphisms and thus defines a func-
tor Group −→ Z Mod (Chapter AT.4.5.1, Chapter III.1.3.4).
Proof of Theorem 1.4.1. Again, we use the bar complex to compute group
homology in degree 1 (Remark 1.2.14). The corresponding part of the bar
complex looks as follows:

degree 2 1 0
∂2 ∂1
C 2 (G) / C 1 (G) / C 0 (G)
L L
G2 ZG G ZG ZG

[g]  / g · [] − []

[g1 | g2 ]  / [g1 ] + g1 · [g2 ] − [g1 · g2 ]


1.4. Degree 1: Abelianisation and homomorphisms 21

In combination with Remark 1.2.4 (and the fact that G acts trivially on Z),
we see that the corresponding part of C∗ (G; Z) = C∗ (G) ⊗G Z is canonically
naturally isomorphic to

degree 2 1 0
∂ 2 ⊗G idZ ∂ 1 ⊗G idZ
C 2 (G; Z) / C 1 (G; Z) / C 0 (G; Z)

=Z

=Z

=Z


L L
G2 Z G Z Z

[g]  / g · [] − [] = 0

[g1 | g2 ]  / [g1 ] + [g2 ] − [g1 · g2 ]


 
Hence, H1 (G; Z) = C 1 (G; Z) im(∂ 2 ⊗G idZ ) . A straightforward calculation
shows that
 
C 1 (G; Z) im(∂ 2 ⊗G idZ ) −→ Gab
ϕ : [g] + im(. . . ) 7−→ g · [G, G]
[g] + im(. . . ) ←−[ g · [G, G] : ψ

are well-defined mutually inverse isomorphisms:


• The map ϕ is a well-defined Z-linear map: We consider the Z-linear
map defined by

ϕ : C 1 (G; Z) −→ Gab
[g] 7−→ g · [G, G];

this is possible because Gab is Abelian. In order to show that ϕ is well-


defined, we only need to show that ϕ maps all elements of im(∂ 2 ⊗G idZ )
to the trivial element of Gab . If g1 , g2 ∈ G, then we have in Gab that
 
ϕ ∂ 2 ([g1 | g2 ]) = ϕ [g1 ] + [g2 ] − [g1 · g2 ]
  
= ϕ [g1 ] + ϕ [g2 ] − ϕ [g1 · g2 ]
  −1
= g1 · [G, G] · g2 · [G, G] · g1 · g2 · [G, G]
= e · [G, G].

• The map ψ is a well-defined Z-linear map: The map


 
ψ : G −→ C 1 (G; Z) im(∂ 2 ⊗G idZ )
g 7−→ [g] + im(. . . )

is a group homomorphism (by the explicit description of ∂ 2 ⊗G idZ


above). Hence, the universal property of the Abelianisation shows that
22 1. The basic view

[x−1 | y −1 ]

x·y
x · y · x−1 · y −1 y
[x | y] [x | y]

Figure 1.1.: A geometric description of the well-definedness of ψ; the triangles


visualise elements in C 2 , the edges visualise elements in C 1 .

ψ induces a well-defined group homomorphism


 
ψ : Gab −→ C 1 (G; Z) im(∂ 2 ⊗G idZ )
g · [G, G] 7−→ [g] + im(. . . ).

• The maps ϕ and ψ are mutually inverse: This is clear from the con-
struction.

Moreover, ϕ and ψ are natural on Group (check!).

Remark 1.4.3 (more on the geometry of commutators). The proof of Theo-


rem 1.4.1 shows that for each commutator g = [x, y] in G with x, y ∈ G we
have that [g] ∈ im(∂ 2 ⊗G idZ ). This can also be seen geometrically as follows:
By construction, we have in C 1 (G; Z) (Figure 1.1)

[g] = [x · y · x−1 · y −1 ]
= ∂ 2 ⊗G idZ [x | y] + [x−1 | y −1 ] − [x · y | x−1 · y −1 ]

− [x | x−1 ] − [y | y −1 ] + [e | e] + [e | e] .

This observation can be used to give a more geometric proof of the well-
definedness of ψ (as we have done in the lecture).

Study note. If you know basic algebraic topology: The statement of Theo-
rem 1.4.1 should look familiar (why?). Again, this is part of a deeper connec-
tion of group (co)homology with algebraic topology (Chapter 4).

Example 1.4.4 (group homology in degree 1).

• If G is Abelian, then H1 (G; Z) ∼


=Z Gab ∼
=Z G.

• Let n ∈ N≥2 . Then (Sn )ab ∼


=Z Z/2, and so H1 (Sn ; Z) ∼
=Z Z/2.

Definition 1.4.5 (perfect group). A group G is perfect if G = [G, G].


1.4. Degree 1: Abelianisation and homomorphisms 23

Corollary 1.4.6 (homological characterisation of perfect groups). Let G be a


group. Then G is perfect if and only if H1 (G; Z) ∼
=Z 0.

Proof. Clearly, a group G is perfect if and only if Gab ∼


=Z 0. Hence, the claim
follows from Theorem 1.4.1.

Example 1.4.7 (first homology of alternating groups). Let n ∈ N≥5 . Then the
alternating group An is perfect (Satz III.1.3.13). Hence, H1 (An ; Z) ∼
=Z 0.

Outlook 1.4.8 (Thompson’s group T ). Thompson’s group T is an example of


a finitely generated infinite simple group [15]. In particular, T is perfect, and
so H1 (T ; Z) ∼
=Z 0.

Example 1.4.9 (first homology of free groups). Let S be a set and let F (S)
be the free group, freely generated by S (Appendix A.1). Then the universal
properties of free groups
L and of the Abelianisation show that the homomor-
phism FL (S)ab −→ S Z, that maps S to the standard basis of the free Z-
module S Z is an isomorphism of Z-modules. In particular, Theorem 1.4.1
shows that  M
H1 F (S); Z ∼=Z Z.
S

Moreover, the first group homology also gives a crude lower bound for the
rank of a group. Let us first recall the rank of Z-modules: If A is a Z-module,
then
(
rank of the free part of A if A is finitely generated
rkZ A =
∞ if A is not finitely generated.

For general groups, we will use the following notion of rank:

Definition 1.4.10 (rank of a group). Let G be a group. The rank of G, denoted


by rk G ∈ N∪{∞}, is the minimal number of generators needed to generate G.

Corollary 1.4.11 (homological rank estimate). Let G be a group. Then

rkZ H1 (G; Z) ≤ rk H1 (G; Z) ≤ rk G.

In particular: If rkZ H1 (G; Z) = ∞, then G is not finitely generated.

Proof. Combining Theorem 1.4.1 with the fact that the rank of a quotient
group cannot exceed the rank of the original group (check!), we obtain

rkZ H1 (G; Z) ≤ rk H1 (G; Z) = rk Gab = rk G/[G, G] ≤ rk G.

Outlook 1.4.12 (homological rank gradient estimate). In this basic form, the
estimate of Corollary 1.4.11 might not seem very impressive and, in general,
24 1. The basic view

it is very far from being optimal. Taking `2 ( · ; C)-coefficients instead of Z-


coefficients, one can prove a corresponding lower bound for the so-called rank
gradient of group [64, 29, 48]: If G is a finitely generated residually finite
group, then
(2) rk H
b1 (G) ≤ inf ,
H∈FinSub(G) [G : H]

where FinSub(G) denotes the set of all finite index subgroups of G and the left
hand side is a suitable “dimension” of H1 (G; `2 (G, C)). There is no example
known of a finitely presented residually finite group, for which this inequality
is not an equality(!).

1.4.2 Cohomology in degree 1: Homomorphisms

Theorem 1.4.13 (group cohomology in degree 1). Let G be a group and let
Z be a Z-module. Then (where Z carries the trivial G-action) there is a
canonical isomorphism

H 1 (G; Z) ∼
=Z HomGroup (G, Z).

More precisely, the functor H 1 ( · ; Z) : Group −→ Z Mod is canonically natu-


rally isomorphic to HomGroup ( · , Z) : Group −→ Z Mod.

Proof. As in the proof of the corresponding result for homology (Theo-


rem 1.4.1), we use the bar complex to compute H 1 (G; Z). In combination
with Remark 1.2.14 and the hypothesis that G acts trivially on Z, we obtain
that the corresponding part of C ∗ (G; Z) = HomG (C ∗ (G), Z) is canonically

naturally isomorphic to C (G; Z), i.e., to

2 1 0
2 HomG (∂ 2 ,Z) 1 − HomG (∂ 1 ,Z) 0
C (G; Z) o C (G; Z) o C (G; Z)
L  L 
HomG G2 ZG, Z HomG G ZG, Z HomG (ZG, Z)
=Z

=Z

=Z

L  L 
HomZ G2 Z, Z HomZ G Z, Z HomZ (Z, Z)
=Z

=Z

=Z





[g1 | g2 ] 7→ f ([g1 ]) + f ([g2 ]) − f ([g1 · g2 ]) o f


 
[g] 7→ f (∂ 1 [g]) = 0 o f

Hence, the elements in the kernel of the coboundary map HomG (∂ 2 , Z) pre-
cisely correspond to those maps G −→ Z that satisfy the homomorphism
condition. Because im(− HomG (∂ 1 , Z)) = 0, it follows that
1.4. Degree 1: Abelianisation and homomorphisms 25

H 1 (G; Z) ∼
=Z ker HomG (∂ 2 , Z) −→ HomGroup (G, Z)

f 7−→ g 7→ f ([g])

[g] 7→ f (g) ←−[ f

are well-defined, mutually inverse, isomorphisms of Z-modules. Moreover, all


these isomorphisms are natural on Group (check!).
In particular, we can apply this to torsion groups. A group G is a torsion
group if every element of G has finite order. In contrast, a group G is torsion-
free if no non-trivial element of G has finite order.
Example 1.4.14 (torsion groups).
• Every finite group is a torsion group.
Q
• The group N Z/2 is a torsion group, but infinite.
• There exist finitely generated, infinite, torsion groups [33][53, p. 111ff].
• The additive group Z is torsion-free.
Corollary 1.4.15. Let G be a torsion group and let Z be a torsion-free Abelian
group (endowed with the trivial G-action). Then H 1 (G; Z) ∼=Z 0.
Proof. By Theorem 1.4.13, we have H 1 (G; Z) ∼ =Z HomGroup (G, Z). Because
G is a torsion group and Z is torsion-free, all group homomorphisms G −→ Z
are trivial. Hence, the right hand side is the trivial Abelian group.
Example 1.4.16. By Corollary 1.4.15, we have H 1 (G; Q) ∼
=Z 0 for every finite
group G (we will generalise this in Corollary 1.7.16).

1.4.3 Application: Hilbert 90


In algebraic number theory, number theoretic properties are encoded in field
extensions. Problems on Galois extensions can be translated into problems
on Galois groups and their cohomology. Therefore, group cohomology con-
tributes to algebraic number theory in various ways. We will briefly discuss a
simple example of this type, namely the solution of polynomial equations via
radicals. Because the Galois group of a field extension acts on the extension
field, we can view the extension field as a module over the Galois group.
Theorem 1.4.17 (Hilbert 90, cohomological version). Let L | K be a finite
Galois extension of fields with Galois group G. Then

H 1 (G; L× ) ∼
=Z 0,

where the G-action on the coefficients L× is the Galois action (and we think
of the coefficients L× as a multiplicative group).
26 1. The basic view

Proof. During this proof, it will be convenient to denote the cohomology


group H 1 (G; L× ) multiplicatively (because L× should be viewed as multi-
plicative group). Moreover, in view of Remark 1.2.14, it suffices to prove

that H 1 (C (G; L× )) ∼
=Z 1.
We follow the general strategy of averaging: The group G is finite and thus
we can sum up (i.e., average) expressions over all elements of G.
1
Let f ∈ C (G; L× ) = HomG (C 1 (G), L× ) be a cocycle. The elements of G
yield group homomorphisms L× −→ L× and thus can be viewed as characters
of L. Such characters are linearly independent over L (Lemma 1.4.18) and
thus there exists an x ∈ L such that
X 
x := f 1 · [τ ] · τ (x)
τ ∈G

is non-zero, whence lies in L× . We will now prove that the ZG-homomorphism

b : C 0 (G) −→ L×
σ0 · [] −→ σ0 (x)

witnesses that f is a coboundary: Indeed, for all σ1 ∈ G, we have (where we


0
abbreviate δ := − HomG (∂1 , L× ))
0  
(δ b) 1 · [σ1 ] = b (−1) · (σ1 · [] − [])

b []
=  (multiplicative notation!)
b σ1 · []
x
= (construction of b)
σ1 (x)
x
=P  (construction of x)
τ ∈G σ1 f 1 · [τ ] · σ1 ◦ τ (x)
x
=P  (f is ZG-linear)
τ ∈G f σ1 · [τ ] · σ1 ◦ τ (x)
x
=P 1
 (f is a cocycle)
τ ∈G f ([σ1 ]) · f [σ1 ◦ τ ] · σ1 ◦ τ (x)

x · f [σ1 ]
=P  (averaging is invariant)
τ ∈G f [τ ] · τ (x)
 1
= x · f [σ1 ] · (construction of x)
 x
= f 1 · [σ1 ] .
0 0
Because f and δ (b) are ZG-linear, we obtain δ b = f . This shows that
H 1 (G; L× ) is trivial.
1.4. Degree 1: Abelianisation and homomorphisms 27

Proposition 1.4.18 (linear independence of characters). Let G be a group and


let L be a field. If n ∈ N and χ1 , . . . , χn : G −→ L× are n distinct group
homomorphisms (so-called characters), then the family (χ1 , . . . , χn ) in the
L-vector space of maps G −→ L is L-linearly independent.

Proof. We proceed by induction over n:

• The case n = 0: The empty family is always linearly independent.

• The case n = 1: Because all values of χ1 lie in L× , the function χ1 is


not the zero function and thus linearly independent.

• The induction step: Let n ∈ N>1 and let the claim already be proved for
families of size n−1. Then also (χ1 , . . . , χn ) are L-linearly independent:
Let λ1 , . . . , λn ∈ L with
n
X
λ j · χj = 0
j=1

(where the 0 on the right hand side is the zero function G −→ L). We
will now show that λ1 = · · · = λn = 0:
Because χ1 6= χn , there exists a g ∈ G with χ1 (g) 6= χn (g). Applying
the equation above to g · h with h ∈ G shows that
n
X n
X
∀h∈H 0 = λj · χj (g · h) = λj · χj (g) · χj (h).
j=1 j=1

On the other hand, multiplying the equation above by χn (g) leads to


n
X n
X
∀h∈H 0 = χn (g) · χj (h) = χn (g) · χj (h).
j=1 j=1

Subtracting the last two equations from each other results in


n−1
X 
0= λj · χj (g) − χn (g) · χj .
j=1

Because the family (χ1 , . . . , χn−1 ) is L-linearly independent by induc-


tion, we obtain that all these coefficients must be zero. In particular,

λ1 · χ1 (g) − χn (g) = 0.

By construction, χ1 (g) 6= χn (g), whence λ1 = 0. Inserting this into the


original equation shows that also
28 1. The basic view
n
X
λj · χj = 0.
j=2

Thus, applying the induction hypothesis to (χ2 , . . . , χn ), we also ob-


tain λ2 = · · · = λn = 0.
We will now apply this cohomological version to cyclic Galois extensions:
Corollary 1.4.19 (Hilbert 90). Let L | K be a finite cyclic Galois extension,
let σ be a generator of the Galois group G, and let x ∈ L. Then the following
are equivalent:
1. The norm NL|K (x) of x equals 1.
2. There exists an a ∈ L× satisfying x = a/σ(a).
Before giving the proof, we recall some terminology from Galois theory:
• A Galois extension is cyclic if its Galois group is cyclic.
• If L | K is a finite Galois extension and x ∈ L, then the norm NL|K (x) ∈
K of x is the determinant of the K-linear map L −→ L given by
multiplication with x.
• If L | K is a finite Galois extension, then [9, Kapitel 4.7]
Y
NL|K (x) = τ (x)
τ ∈Gal(L,K)

for all x ∈ L. In particular, NL|K is invariant under the action of the


Galois group Gal(L, K). For our purposes, we could also take this as a
definition of the norm.
Proof of Corollary 1.4.19. Ad 2 =⇒ 1. Let a ∈ L× with x = a/σ(a). Then

NL|K (a)
NL|K (x) =  (multiplicativity of the determinant/norm)
NL|K σ(x)
NL|K (a)
= (Galois invariance of the norm)
NL|K (a)
= 1.

Ad 1 =⇒ 2. Conversely, let us suppose that NL|K (x) = 1 and that the σ


has order n (thus G ∼
=Group Z/n). Then the ZG-linear map specified by

f : C 1 (G) −→ L×
k−1
Y
1 · [σ k ] 7−→ σ j (x)
j=0
1.4. Degree 1: Abelianisation and homomorphisms 29

1 Qn−1
is a well-defined cocycle in C (G; L× ) (because j=0 σ j (x) = NL|K (x) = 1;
check!). Hence, the cohomological version of Hilbert 90 (Theorem 1.4.17)
0 0
shows that there exists a b ∈ C (G; L× ) with δ (b) = f . Then a := b(1 · [])
satisfies
 
 0  b 1 · [] b 1 · [] a
x = f 1 · [σ] = (δ b) 1 · [σ] = = = ,
b σ · [] σ b(1 · []) σ(a)

as desired.

Remark 1.4.20 (application to number theory). One can now apply Corol-
lary 1.4.19 to roots of unity in the base field: Let L | K be a cyclic Galois
extension of degree n ∈ N and let ζ ∈ K be an n-th root of unity. Then
NL|K (ζ) = ζ n = 1 (we have ζ ∈ K and thus the determinant is just ζ dimK L ).
If σ ∈ Gal(L, K) is a generator of the Galois group of L | K, then Corol-
lary 1.4.19 provides an element a ∈ L with a = ζ · σ(a). In particular,
an ∈ Lσ = K and so X n − an ∈ K[X].
If ζ is a primitive root of unity (primitive in an algebraic closure of K)
and if charK - n, then one can show in this situation that L is the splitting
field of X n − an over K.
This observation is the base case of the inductive proof of the character-
isation of solvability by radicals in terms of solvability of the corresponding
Galois groups [9, Kapitel 6.1] (Chapter III.3.5.2).
Outlook 1.4.21 (generalisations). Clearly, it is possible to prove the classi-
cal Hilbert 90 theorem (Corollary 1.4.19) without talking about group co-
homology (for instance, we could take the same proof and apply the av-
eraging/character argument directly to the specific cocycle in the proof of
Corollary 1.4.19). The cohomological formulation has the advantage that it
is easy to guess generalisations:
• There is an additive version of Theorem 1.4.17, involving the trace
instead of the norm (which is used to complete the characterisation of
cyclic Galois extensions in positive characteristic).
• Further generalisations occur in Kummer theory (i.e., finite Galois ex-
tensions with Abelian Galois group).
• There is a version of Theorem 1.4.17 for infinite Galois extensions (in
terms of continuous group cohomology of the (profinite) Galois group).
Furthermore, also group cohomology in degree 2 plays an important role
in algebraic number theory (as Brauer groups). Therefore, computations in
group cohomology are relevant for algebraic number theory.

Literature exercise. Who first proved Corollary 1.4.19? Why is it called


“Hilbert 90”? Who first proved Theorem 1.4.17?
30 1. The basic view

1.5 Degree 2: Presentations and extensions


We now move on to group (co)homology in degree 2. In homology, we will
see a close relation with presentations of groups; dually, in cohomology, we
will be able to classify group extensions with Abelian kernel.

1.5.1 Homology in degree 2: Hopf’s formula

Theorem 1.5.1 (Hopf’s formula). Let F be a free group, let N ⊂ F be a


normal subgroup, and let G := F/N . Then there is an exact sequence

0 / H2 (G; Z) / H1 (N ; Z)G / H1 (F ; Z) / H1 (G; Z) /0

(where the homomorphisms on H1 are induced by the canonical inclusion and


projection, respectively, and G acts on H1 (N ; Z) by conjugation of represen-
tatives in F ). More explicitly,

N ∩ [F, F ]
H2 (G; Z) ∼
=Z .
[F, N ]

Here, [F, N ] denotes the subgroup of F generated by the set of commuta-


tors [x, y] with x ∈ F and y ∈ N .

Proof. The Hopf formula admits different proofs, e.g., via spectral sequences
or via classifying spaces. We will give a proof later, once we have more tools
available (Theorem 3.2.18). A proof that only uses basic homological algebra
can be found in the book by Hilton and Stammbach [42, Chapter VI.9].

Hopf’s formula has many applications in group theory; for example, it can
be used to establish an analogue of the rank estimate from Corollary 1.4.11
for the number of relations. In order to formulate this result, we briefly recall
presentations of groups in terms of generators and relations:

Remark 1.5.2 (generators and relations). Let S be a set, let F (S) be “the”
free group generated by S (Appendix A.1), and let R ⊂ F (S). Then the group
generated by S with relations R is defined as

hS | Ri := F (S)/N,

where N := hRi/F (S) ⊂ F (S) is the smallest (with respect to inclusion) nor-
mal subgroup of F (S) containing R. It is common to abuse notation and to
write hS | Ri also to refer to the presentation, not only the group.
1.5. Degree 2: Presentations and extensions 31

For example, working with the corresponding universal properties shows


that [53, Chapter 2.2]

ha | i ∼=Group Z
−1 −1
ha, b | aba b i = ∼Group Z2
∼Group Z/2
ha | a2 i =
hs, t | s2019 , t2 , tst−1 si ∼
=Group D2019 .

However, it can be proved that there is no algorithm that, given a pre-


sentation with finitely many generators and finitely many relations, decides
whether the corresponding group is trivial or not [70, Chapter 12].
Corollary 1.5.3. Let G be a finitely generated group. If G admits a finite
presentation, then H2 (G; Z) is a finitely generated Z-module.
In other words: If H2 (G; Z) is not a finitely generated Z-module, then G
does not admit a finite presentation.
Proof. If hS | Ri is a finite presentation of G, then

G∼
=Group F (S)/hRi/F (S)

and we can apply Hopf’s formula (Theorem 1.5.1) to the free group F := F (S)
and the normal subgroup N := hRi/F (S) . Hence, H2 (G; Z) is isomorphic to a
submodule of H1 (N ; Z)G . Because rkZ H1 (N ; Z)G ≤ |R| (Theorem 1.4.1 and
the fact that R normally generates N ), it follows that H2 (G; Z) is finitely
generated.
We can refine this finiteness property to a quantitative statement:
Corollary 1.5.4. Let G be a finitely presentable group and let hS | Ri be a finite
presentation of G. Then

|S| − |R| ≤ rkZ H1 (G; Z) − rkZ H2 (G; Z).

Proof. Again, we can apply Hopf’s formula (Theorem 1.5.1) to the free
group F := F (S) and the normal subgroup N := hRi/F (S) . Then the short
exact sequence in Hopf’s formula shows that

rkZ H2 (G; Z) − rkZ H1 (N ; Z)G + rkZ H1 (F ; Z) − rkZ H1 (G; Z) = 0;

this is a generalised version of the dimension formula from linear algebra,


which can, e.g., be proved by applying the exact functor Q ⊗Z · . Hence,

|S| − |R| ≤ rkZ H1 (F ; Z) − rkZ H1 (N ; Z)G (Example 1.4.9; N = hRi/F (S) )


= rkZ H1 (G; Z) − rkZ H2 (G; Z), (dimension formula)

as claimed.
32 1. The basic view

Outlook 1.5.5 (deficiency). Let G be a finitely generated group and let P (G)
denote the “set” of all presentations of G with finite generating set. The
deficiency of G is defined as

def G := max |S| − |R| hS | Ri ∈ P (G) .

Then Corollary 1.5.4 shows that

def G ≤ rkZ H1 (G; Z) − rkZ H2 (G; Z).

In particular, def G is a well-defined element of Z ∪ {−∞} (and never equal


to +∞).
Conversely, we can use group homology in higher degree to introduce fur-
ther finiteness conditions on groups, going beyond finite presentability. We
will return to the point of view in Chapter 4.2.

1.5.2 Cohomology in degree 2: Extensions


One of the most classical applications of group cohomology is the classifica-
tion of extensions through H 2 ( · ; · ).
Definition 1.5.6 (extensions). Let Q and A be groups.
• An extension of Q by A is an exact sequence in Group of the form

1 /A /G /Q /1

0 0
• Extensions 1 /A i /G π /Q / 1 and 1 / A i / G0 π / Q /1
of Q by A are called equivalent if there exists a group homomor-
phism ϕ : G −→ G0 fitting into the commutative diagram

i π
1 /A /G /Q /1

ϕ

1 /A / G0 /Q /1
i0 π0

(In this case, ϕ is already an isomorphism; check!)


Remark 1.5.7 (the conjugation action of the quotient on the kernel). Let

i π
0 /A /G /Q /1

be an extension of a group Q by an Abelian group A. This extension induces


a ZQ-module structure on A: The group G acts by conjugation on the normal
1.5. Degree 2: Presentations and extensions 33

subgroup i(A), whence on A ∼ =Group i(A). Because A is Abelian, the conjuga-


tion action of A on itself is trivial, whence the conjugation action by G on A
descends to a well-defined action of Q ∼ =Group G/i(A) on A:

Q × A −→ A
(q, a) 7−→ “g · a · g −1 ”= i−1 (g · i(a) · g −1 ), where g ∈ G with π(g) = q.

This Q-action on A yields a ZQ-module structure on A.


Equivalent extensions of Q by A lead to the same ZQ-module structure
on A (check!).

Example 1.5.8 (extensions and actions).

• Let A be an Abelian group and let Q be a group. Then the action of Q


on A induced by the product extension

0 / Acan. incl./ A × Qcan. proj./ Q /1

is the trivial action.

• In the extension

0 / A3 can. incl/ S3can. proj./ S3 /A3 /1

the non-trivial element of the quotient S3 /A3 ∼


=Group Z/2 acts by taking
inverses on A3 ∼=Group Z/3 (check!).
• The extensions

can. incl. can. proj.


0 /Z /Z × Z/2 / Z/2 /1

0 /Z 2· · / Z can. proj.
/ Z/2 /1

both lead to the trivial action of Z/2 on Z. But these extensions are
not equivalent (because the extension groups are not isomorphic).

Definition 1.5.9 (extension set). Let Q be a group and let A be a ZQ-module.


We then write E(Q, A) for the set(!) of all equivalence classes of extensions
of Q by A that induce the given ZQ-module structure on A.

Theorem 1.5.10 (classification of group extensions with Abelian kernel). Let Q


be a group and let A be a ZQ-module. Then the maps

H 2 (Q; A) −→ E(Q, A)
[f ] 7−→ [Ef : 0 → A → Gf → Q → 1]
ηE ←−[ E
34 1. The basic view

are mutually inverse bijections (the extensions Ef and the cohomology class ηE
will be specified in the proof below).
Proof. Again, we will work with the description of H 2 (Q; A) in terms of the
bar resolution (Remark 1.2.14).
We start with the map from the right hand side to the left hand side; i.e.,
we explain how an extension defines a 2-cocycle in such a way that equivalent
extensions lead to cohomologous cocycles. Let E ∈ E(Q, A) and let

i π
0 /A /G /Q /1

be an extension of Q by A that induces the given Q-action on A and that


represents E. The idea is to measure the failure of π : G −→ Q to be a split
epimorphism by a 2-cocycle:
• Choosing a section. Let s : Q −→ G be a set-theoretic section of the
map π : G −→ Q (such a section exists by the axiom of choice). Of
course, in general, we cannot expect s to be a group homomorphism.
Measuring the failure of s being a group homomorphism leads to the
map

F : Q × Q −→ A
(q1 , q2 ) 7−→ s(q1 ) · s(q2 ) · s(q1 · q2 )−1 ;

this map F indeed maps to A (because s is a section of π) and will be


the key to defining a 2-cocycle.
• Rewriting the group structure on G. Using the map F , we can recover
the group structure on G from the group structure on Q and the Q-
action on A as follows: First of all,

G −→ A × Q

g 7−→ g · s(π(g))−1 , π(g)
a · s(q) ←−[ (a, q)

are mutually inverse bijections (check!). Under these bijections, the


composition on G translates into the following composition on A × Q
(where • denotes the Q-action on A induced by the given extension):

(A × Q) × (A × Q) −→ A × Q
 
(a, q), (a0 , q 0 ) 7−→ a + q • a0 + F (q, q 0 ), q · q 0 .

Indeed, for all (a, q), (a0 , q 0 ) ∈ A × Q we have (in G)

a · s(q) · a0 · s(q 0 ) = a · q • a0 · s(q) · s(q 0 )


= a · q • a0 · F (q, q 0 ) · s(q · q 0 ),
1.5. Degree 2: Presentations and extensions 35

because the Q-action on A is given by conjugation in G.


• Constructing a cocycle. Because the composition on G is associative, a
short calculation shows that

F (q1 , q2 ) + F (q1 · q2 , q3 ) − q1 • F (q2 , q3 ) − F (q1 , q2 · q3 ) = 0

for all q1 , q2 , q3 ∈ Q (check!). Hence,

f : C 2 (Q) −→ A
q0 · [q1 | q2 ] 7−→ q0 • F (q1 , q2 )

is a cocycle in C (Q; A).
• Changing the section. Let s0 : Q −→ G be another set-theoretic section
of π and let f 0 ∈ C 2 (Q; A) be the corresponding cocycle. Because s and
s0 are sections of π, there is a function B : Q −→ A with

∀q∈Q s0 (q) = B(q) · s(q).

Then

b : C 1 (Q) −→ A
q0 · [q1 ] 7−→ q0 • B(q1 )
1
satisfies δ (b) = f 0 − f : By construction, for all q1 , q2 ∈ Q, we have

f 0 ([q1 | q2 ]) = s0 (q1 ) · s0 (q2 ) · s0 (q1 · q2 )−1 in G


−1 −1
= B(q1 ) · s(q1 ) · B(q2 ) · s(q2 ) · s(q1 · q2 ) · B(q1 · q2 ) in G

= B(q1 ) · q1 • B(q2 ) · s(q1 ) · s(q2 ) · s(q1 · q2 )−1 · B(q1 · q2 )−1 in G

= B(q1 ) + q1 • B(q2 ) + f [q1 | q2 ] − B(q1 · q2 ) in A
1  
= (δ b) [q1 | q2 ] + f [q1 | q2 ] . in A

• Changing the extension. Let

i0 / G0 π0
0 /A /Q /1

be an extension that is equivalent to the previous one. Furthermore, let


ϕ : G0 −→ G be an isomorphism witnessing that these extensions are
equivalent. If s0 : Q −→ G0 is a section of π 0 , then ϕ ◦ s0 : Q −→ G is a
section of π and the cocycles corresponding to s0 and to ϕ ◦ s0 coincide
(check!). Therefore, the previous step shows that the extension G0 leads
to the same cohomology class as the extension involving G.
Therefore, we obtain a well-defined cohomology class ηE ∈ H 2 (Q; A) out
of E.
36 1. The basic view

Conversely, let a cohomology class η ∈ H 2 (Q; A) be given. We now show


how to construct an equivalence class of extensions of Q by A out of this
cohomology class. More precisely, we construct extensions out of 2-cocycles
in such a way that cohomologous cocycles lead to equivalent extensions:
2
• A group structure out of a cocycle. Let f ∈ C (Q; A) be a cocycle and
let

F : Q × Q −→ A

(q1 , q2 ) 7−→ f [q1 | q2 ] .

Inspired by the first part of the proof, on the set A × Q, we define the
composition

(A × Q) × (A × Q) −→ A × Q
 
(a, q), (a0 , q 0 ) 7−→ a + q • a0 + F (q, q 0 ), q · q 0 .

The same calculation as above shows that f being a cocycle implies


that this composition is associative (check!).
Moreover, using the cocycle property of f once more, we see that (ε, e)
is a neutral element for this composition (check!), where

ε := −F (e, e) = −f [e | e] .

A straightforward computation shows that every element of A × Q has


an inverse element with respect to this composition and the neutral
element (ε, e) (check!). So Gf := A × Q with this composition is a
group.

• An extension out of a cocycle. The group Gf fits into the extension

if πf
0 /A / Gf /Q / 1,

where the homomorphisms (check!) are given by

if : A −→ Gf = A × Q
a 7−→ (a + ε, e)
πf : Gf = A × Q −→ Q
(a, q) 7−→ q.

• The induced action on the kernel. The map

sf : Q −→ Gf = A × Q
q 7−→ (0, q)
1.5. Degree 2: Presentations and extensions 37

is a set-theoretic section of πf . Hence, the Q-action ∗ of the above


extension on A is given by

q ∗ a = sf (q) · (a + ε, e) · sf (q)−1
= (0, q) · (a + ε, e) · (0, q)−1

= 0 + q • (a + ε) + F (q, e), q · (00 , q −1 )

= q • (a + ε) + F (q, e) + q • 00 + F (q, q −1 ), e

= q • (a + ε) + F (q, e) + ε, e ((00 , q −1 ) is inverse to (0, q))
= (q • a + ε, e) (because f is a cocycle; for [q | e | e])
= if (q • a)

for all q ∈ Q and all a ∈ A; here, we write (00 , q −1 ) := (0, q)−1 . Thus,
i π
the extension Ef := ( 0 / A f / Gf f / Q / 1 ) induces the given
Q-action on A and so represents a class in E(Q, A).
• Changing the cocycle. Similarly to the previous arguments, we see that
changing the cocycle f by a coboundary leads to an equivalent extension
(check!).
Therefore, we obtain a well-defined map H 2 (Q; A) −→ E(Q, A).
These two maps are mutually inverse:
2
• If f ∈ C (Q; A) is a cocycle, then the cocycle associated with the
extension Ef and the set-theoretic section sf : Q −→ Gf is f (check! A
is embedded via if into Gf ). Hence, η[Ef ] = [f ].

• Conversely, if 0 /A i /G π /Q / 1 is an extension that induces


the given Q-action on A, if s is a set-theoretic section of π and f ∈
2
C (Q; A) is the associated cocycle, then the group homomorphism

Gf = A × Q −→ G
(a, q) 7−→ a · s(q)

shows that Ef is equivalent to the given extension.

When viewing group cohomology as derived functor of a Hom-functor, we


will see that group cohomology can be described as Ext-functor. In fact, the
name “Ext” goes back to the above classification of extensions.
Remark 1.5.11 (trivial extensions). Let Q be a group and let A be a ZQ-
module. Then the extension corresponding (under the bijection in Theo-
rem 1.5.10) to the zero class in H 2 (Q; A) is the semi-direct product extension

0 / Acan. incl./ A o Qcan. proj./ Q /1


38 1. The basic view

with respect to the given Q-action on A (as can be seen from the definition
of the semi-direct product (Chapter III.1.1.6) and the construction of the
extension corresponding to the zero cocycle in the proof of Theorem 1.5.10).

Conversely, non-trivial extensions lead to non-trivial cohomology classes in


degree 2. For example, we can use this observation to show that H 2 (Z/n; Z) is
non-trivial for all n ∈ N≥2 , where Z/n acts trivially on Z (Exercise). Similarly,
also H 2 (Z2 ; Z) and H 2 (Homeo+ (S 1 ); Z) are non-trivial (Exercise).

Corollary 1.5.12 (cohomology of free groups in degree 2). Let F be a free group
and let A be a ZF -module. Then H 2 (F ; A) ∼=Z 0.

Proof. In view of the classification theorem (Theorem 1.5.10) and the de-
scription of the extensions corresponding to the zero class (Remark 1.5.11),
it suffices to show that every extension

i π
0 /A /G /F /1

that induces the given ZF -module structure on A is trivial (i.e., equivalent


to a semi-direct product extension).
The universal property of the free group F shows that the epimor-
phism π : G −→ F admits a group-theoretic section s : F −→ G; i.e., s is
a group homomorphism with π ◦ s = idF .
Then the isomorphism (check! see also the rewriting argument in the
proof of Theorem 1.5.10; because s is a group homomorphism, the associ-
ated map F × F −→ A is the zero map)

G −→ A o F

g 7−→ g · s(π(g))−1 , π(g)

(where the semi-direct product on the right hand side is formed with respect
to the given F -action on A) shows that the given extension is trivial.

Of course, constructing non-trivial cohomology classes out of non-trivial


extensions is a “wrong-way” application of the classification theorem; usually,
one applies it in the other direction. Indeed, once we have more computa-
tional tools available, we will be able to derive group-theoretic results from
Theorem 1.5.10.
The classification of extensions with Abelian kernel is functorial in the
following sense:

Theorem 1.5.13 (functoriality of the classification of extensions with Abelian


kernel). Let (ϕ, Φ) : (Q, A) −→ (Q0 , A0 ) be a morphism in GroupMod and
let E ∈ E(Q, A) and E 0 ∈ E(Q0 , A0 ) be represented by the extensions
/A i /G π /Q / A0 i0 / G0 π0 / Q0
0 / 1 and 0 / 1, respectively.
Then the following are equivalent:
1.5. Degree 2: Presentations and extensions 39

if πf
0 /A / Gf /Q /1

Φ ψ
 ig  πg
0 / ϕ ∗ A0 / Gg /Q /1


=
ig  πg
0 / ϕ ∗ A0 / Gg0 /Q /1

ψ0 ϕ
 
0 / A0 / Gf 0 / Q0 /1
if 0 πf 0

Figure 1.2.: Comparing extensions

e : G −→ G0 making the following


1. There exists a group homomorphism ϕ
diagram commutative:

i π
0 /A /G /Q /1

Φ e
ϕ ϕ
  
0 / A0 / G0 / Q0 /1
i0 π0


2. In H 2 Q; ϕ∗ (A0 ) , we have H 2 (idQ ; Φ)(ηE ) = H 2 (ϕ; idA0 )(ηE 0 ).

Proof. Ad 1 =⇒ 2. In this situation, we can choose set-theoretic sec-


tions s : Q −→ G and s0 : Q0 −→ G0 of π and π 0 , respectively, that sat-
isfy s0 ◦ ϕ = ϕ e ◦ s (check!). Then the explicit construction of the coho-
mology classes of these extensions in the proof of Theorem 1.5.10 shows
that H 2 (idQ ; Φ)(ηE ) = H 2 (ϕ; idA0 )(ηE 0 ) holds already at the level of associ-
ated cocycles (check!).
Ad 2 =⇒ 1. Let H 2 (idQ ; Φ)(ηE ) = H 2 (ϕ; idA0 )(ηE 0 ) in H 2 (Q; ϕ∗ A0 ). We
2 2
choose cocycles f ∈ C (Q; A) and f 0 ∈ C (Q0 ; A0 ) that represent ηE and
ηE 0 , respectively. Without loss of generality, we may assume that the given
extensions are the extensions constructed from the cocycles f and f 0 . Then

g : C 2 (Q) −→ ϕ∗ A0

[q1 | q2 ] 7−→ Φ f ([q1 | q2 ])
g 0 : C 2 (Q) −→ ϕ∗ A0

[q1 | q2 ] 7−→ f 0 [ϕ(q1 ) | ϕ(q2 )]
40 1. The basic view

are cocycles representing H 2 (idQ ; Φ)(ηE ) and H 2 (ϕ; idA0 )(ηE 0 ), respectively.
Because [g] = [g 0 ], we know that the corresponding extensions Eg and Eg0
of Q by ϕ∗ A0 are equivalent (Theorem 1.5.10). Moreover, the group homo-
morphisms (check!)

ψ : Gf = A × Q −→ ϕ∗ A0 × Q = Gg

(a, q) 7−→ a, ϕ(q)
ψ 0 : Gg0 = ϕ∗ A0 × Q −→ A0 × Q0 = Gf 0

(a0 , q) 7−→ Φ(a0 ), q

fit into the commutative diagram in Figure 1.2 (check!). Therefore, the com-
position of the middle vertical arrows is a group homomorphism ϕ e : Gf −→
Gf 0 with the desired property.

Corollary 1.5.14. Let Q be a group, let A be a ZQ-module, let E ∈ E(Q, A),


and let 0 /A i /G π /Q / 1 be an extension representing E. Then

H 2 (π; idA )(ηE ) = 0.

Proof. On the one hand, the zero class in H 2 (G; π ∗ A) is represented by


the semi-direct product extension 0 / π∗ A j / π∗ A o G q / G / 1 (Re-
mark 1.5.11). On the other hand, the group homomorphism (check!)

e : π ∗ A o G −→ G
π
(a, g) 7−→ a · g

fits into the commutative diagram

j q
0 / π∗ A / π∗ A o G /G /1

idA e
π π
  
0 /A /G
π
/Q /1
i

Therefore, Theorem 1.5.13 and Remark 1.5.11 show that

H 2 (π; idA )(ηE ) = H 2 (idQ ; idA )(ηsemi-direct product extension )


= H 2 (idQ ; idA )(0) = 0.

Outlook 1.5.15 (universal central extensions and the Schur multiplier). Let G
be a perfect group and let H := H2 (G; Z) (which we consider as trivial G-
module). Then H 2 (G; H) is isomorphic to HomZ (H, H) and the extension
of G by H corresponding to idH is the universal central extension, which is
1.6. Changing the resolution 41

initial among all central extensions of G [87, Chapter 6.9]; an extension is


central if the kernel is a central subgroup of the extension group.
If G is perfect and finite, then H2 (G; Z) is isomorphic to H 2 (G; C× ), the
Schur multiplier of G. Schur used the group H 2 (G; C× ) to study represen-
tations of G. These considerations were one of the precursors of group coho-
mology.

Outlook 1.5.16 (classification of extensions with non-Abelian kernel). More


generally, also extensions of groups by non-Abelian kernels can be classified by
means of group cohomology [12, Chapter IV.6]. However, this classification is
more delicate; for example, given a quotient group and a kernel group with an
action by the quotient, there does not necessarily exist a extension inducing
the given action (i.e., there is no non-Abelian analogue of the semi-direct
product). This classification can, e.g., be used to give efficient algorithms
testing whether finite groups of certain types are isomorphic or not [34].

1.6 Changing the resolution


Using the fundamental theorem of homological algebra, we will obtain
that group (co)homology can be computed via many (co)chain complexes.
This flexibility is the key to many computations and applications of group
(co)homology. We will first briefly review projective resolutions and recall
the proof of the fundamental theorem of homological algebra. Then, we will
use these results for some concrete computations and later to establish the
Shapiro lemma.

1.6.1 Projective resolutions


The basic building blocks in derived homological algebra are projective reso-
lutions. Projective modules are a convenient generalisation of free modules.

Definition 1.6.1 (projective module). Let R be a ring with unit. A left R-


module P is projective if it has the following lifting property: For every epi-
morphism π : B −→ C in R Mod and every R-homomorphism α : P −→ C,
there exists an R-homomorphism α e : P −→ B with π ◦ α e = α.

P
e
α
α
 
B π
/C /0

(Analogously, we define projective right modules).


42 1. The basic view

Example 1.6.2 (projective modules).

• All free modules are projective (check!). In fact, for most of our appli-
cations free modules will suffice.

• Direct summands of free modules are projective (check!).

• Direct sums of projective modules are projective.

• The Z-module Z/2 is not projective: The lifting problem

Z/2
? idZ/2
~ 
Z / Z/2 /0
Proj.

in Z Mod has no solution.

• Not every projective module is free: For example, the sections of the
Möbius strip form a projective module over the ring C(S 1 , R) of con-
tinuous functions, which is not free (Beispiel IV.3.4.12).

Moreover, we have the following equivalent characterisations of projectiv-


ity (Proposition IV.5.2.3):

Proposition 1.6.3 (characterisations of projectivity). Let R be a ring with unit


and let P be a left R-module. Then the following are equivalent:

1. The R-module P is projective.

2. The R-module P is a direct summand of a free R-module.

3. Every short exact sequence 0 −→ A −→ B −→ P −→ 0 in R Mod


admits a right split.

4. The functor R Hom(P, · ) : R Mod −→ Z Mod is exact (i.e., maps exact


sequences to exact sequences).

Definition 1.6.4 (projective resolution). Let R be a ring with unit and let M
be a left R-module. A resoluion of M by R-modules is a pair (C∗ , ε), where

• C∗ is an N-indexed chain complex of left R-modules (with boundary


operators ∂∗ )

• ε is an R-homomorphism C0 −→ M ,

• and the concatenated sequence P  ε is exact:

∂2 ∂1 ε
··· / P2 / P1 / P0 /M /0
1.6. Changing the resolution 43

(e, g1 ) (e, g2 )
(e, g1 , g2 )

g1 g2
(g1 , g2 )

Figure 1.3.: The cone operator in the singular resolution

A resolution (C∗ , ε) of M by R-modules is

• projective if for each n ∈ N, the chain module Cn is projective,

• free if for reach n ∈ N, the chain module Cn is free.

Proposition 1.6.5 (simplicial resolution and bar resolution). Let G be a group


and let ε : ZG −→ Z be the augmentation map (Definition 1.2.1). Then
(C∗ (G), ε) and (C ∗ (G), ε) are free (whence projective) resolutions of Z (with
the trivial G-action) over ZG.

Proof. As the ZG-chain complexes C∗ (G) and C ∗ (G) are isomorphic (Re-
mark 1.2.4) and as these isomorphisms are compatible with ε, it suffices to
prove that

À C n (G) is for each n ∈ N a free ZG-module and that

Á (C∗ (G), ε) is a ZG-resolution of Z.


L
Ad À. Let n ∈ N. Then C n (G) = Gn ZG is a free ZG-module by con-
struction.
Ad Á. Because C∗ (G) is a chain complex and ε ◦ ∂1 = 0, the concatena-
tion C∗ (G)  ε is a ZG-chain complex. In order to prove exactness, we will give
an explicit construction of a chain contraction over the ring Z (geometrically,
this corresponds to a cone-operation with cone point e; Figure 1.3): To this
end we consider the Z-homomorphisms given by

s−1 : Z −→ ZG = C0 (G)
1 7−→ 1

as well as, for all n ∈ N,

sn : Cn (G) −→ Cn+1 (G)


n+1
G 3 (g0 , . . . , gn ) 7−→ (e, g0 , . . . , gn ).
44 1. The basic view

Then we have

ε ◦ s−1 = idZ
s−1 ◦ ε + ∂1 ◦ s0 = idC0 (G)

and
sn−1 ◦ ∂n + ∂n+1 ◦ sn = idCn (G)
for all n ∈ N (check! Figure 1.3); i.e., (sn )n∈N∪{−1} is a chain contraction
of C∗ (G)  ε as chain complex of Z-modules (the whole point of the theory
is that this, in general, is not a chain contraction over the ring ZG). This
proves exactness.
More explicitly: By construction, ε is surjective. If c ∈ ker ε, then

c = (s−1 ◦ ε + ∂1 ◦ s0 )(c) = ∂1 s0 (c) ∈ im ∂1 .

In the same way, if n ∈ N>0 and c ∈ ker ∂n , then



c = (sn−1 ◦ ∂n + ∂n+1 ◦ sn )(c) = ∂n+1 sn (c) ∈ im ∂n+1 .

Example 1.6.6 (resolutions from topology). Let X be a path-connected CW-


complex with fundamental group G and universal covering π : X e −→ X.
Then the (involution of the) deck transformation action of G on X e turns
e e e is
the singular chain complex C∗ (X; Z) of X into a ZG-chain complex. If X
e Z), together with
contractible, then C∗ (X;

e Z) −→ Z
C0 (X;
e 3 σ 7−→ 1,
map(∆0 , X)

is a free resolution of the trivial ZG-module Z over ZG.


If we equip the universal covering X e with the lifted CW-structure from X,
then also the cellular chain complex of X e can be viewed as a free resolution
of Z over ZG. We will explore this connection in more detail in Chapter 4.
If G is a group, then the full simplex ∆(G) on the vertex set G is con-
tractible and its full simplicial chain complex coincides with C∗ (G). This gives
an alternative proof of the fact that C∗ (G) (together with the augmentation
map) is a free resolution of Z ∼ =Z H0 (•; Z) over ZG.

1.6.2 The fundamental theorem of group (co)homology


In derived homological algebra, one replaces modules by projective resolu-
tions. The fundamental observation is that every module admits a projective
resolution and that projective resolutions are unique up to chain homotopy
equivalence. In our situation (i.e., for modules over the group ring), this shows
1.6. Changing the resolution 45

that we can replace the simplicial/bar resolution by other resolutions, which


we call the fundamental theorem of group (co)homology.
Theorem 1.6.7 (fundamental theorem of homological algebra). Let R be a ring
with unit, let M and N be left R-modules, and let f : M −→ N be an R-
module homomorphism. Let (P∗ , ε) be a projective resolution of M over R
and let C∗  (γ : C0 → N ) be an exact sequence in R Mod.

1. Then f : M −→ N admits an extension fe∗  f : P∗  ε −→ C∗  γ to an


R-chain map.
2. Moreover, this extension fe∗ is unique up to R-chain homotopy.
Proof. Ad 1. We construct fe∗ inductively: The boundary operators of P∗ will
be denoted by ∂∗P , the ones of C∗ by ∂∗C . In order to keep notation simple,
we will also use the (dangerous) conventions ∂0P := ε, ∂0C := γ, and fe−1 := f .
• The base case. Because γ : C0 −→ N is surjective and P0 is projective,
there exists an R-homomorphism fe0 : P0 −→ C0 with γ ◦ fe0 = f ◦ ε:
ε
P0 /M /0

fe0 f
 
C0 γ
/N /0

• Induction step. Let n ∈ N and let us suppose that we already con-


structed an extension fe: P∗ −→ C∗ up to degree n (as R-chain
map). We then obtain an R-homomorphism fen+1 : Pn+1 −→ Cn+1
C
with ∂n+1 ◦ fen+1 = fen ◦∂n+1
P
as solution to the following lifting problem:

P
∂n+1
P
Pn+1 / im ∂n+1

fen+1 fen |im ∂ P


n+1
 
C
Cn+1 C
/ im ∂n+1 /0
∂n+1

Here, we use that fen (im ∂n+1


P C
) ⊂ im ∂n+1 = ker ∂nC , which follows from
∂nC ◦ fen ◦ ∂n+1
P
= fen−1 ◦ ∂nP ◦ ∂n+1
P
= 0.

Ad 2. Uniqueness follows from a similar construction: Let fe∗ and ge∗ be


extensions of f . We then set h−1 := 0 : M −→ C0 and, inductively, we
construct R-homomorphisms hn : Pn −→ Cn+1 satisfying the chain homotopy
equation
C
∂n+1 ◦ hn + hn−1 ◦ ∂nP = gen − fen .
for all n ∈ N.
46 1. The basic view

g0 − fe0 ) ⊂ im ∂1C because


• The base case. We have im(e

g0 − fe0 ) = f ◦ ∂0P − f ◦ ∂0P


∂0C ◦ (e (fe∗  f and ge∗  f are chain maps)
=0

and C∗  γ is exact. Hence, projectivity of P0 allows to choose h0 : P0 −→


C1 as solution to the lifting problem

P0
h0
e0 −fe0
g
| 
C1 / im ∂1C /0
∂1C

• Induction step. Let n ∈ N and let us suppose that we already con-


structed h0 , . . . , hn satisfying the chain homotopy equation. The induc-
tion hypothesis and exactness of C∗  γ shows that

gn+1 − fen+1 − hn ◦ ∂n+1


im(e P C
) ⊂ im ∂n+2 .

Therefore, projectivity of Pn+1 guarantees the existence of an R-homo-


morphism hn+1 : Pn+1 −→ Cn+2 with
C
∂n+2 ◦ hn+1 = gen+1 − fen+1 − hn ◦ ∂n+1
P
.

Corollary 1.6.8 (existence and uniqueness of projective resolutions). Let R be


a ring with unit and let M be a left R-module. Then M admits a projective
resolution over R, which is unique up to R-chain homotopy equivalence (which
is canonical up to R-chain homotopy).

Proof. Existence. We first establish that the category R Mod of leftLR-mod-


ules has enough projective objects: If M is a left R-module, then M R is
L and the map idM : M −→ M extends to an R-homomor-
a free R-module
phism εM : M R −→ M , which is surjective (by construction).
Using this fact, we can construct a projective resolution of M by induction:
We set
M
P0 := R and
M
M
Pn+1 := R
ker ∂n

∂n+1 := εker ∂n : Pn+1 −→ ker ∂n ⊂ Pn

for all n ∈ N. Then (P∗ , εM ) is a free resolution of M over R. (Moreover, this


construction is functorial on R Mod in terms of the resolved module).
1.6. Changing the resolution 47

Uniqueness follows from the usual universal property yoga (which uni-
versal property in which category?!): Let (P∗ , ε) and (P∗0 , ε0 ) be projective
resolutions of M over R. By the fundamental theorem (Theorem 1.6.7), the
R-homomorphism idM : M −→ M extends to R-chain maps

fe∗  idM : P∗  ε −→ P∗0  ε0


ge∗  idM : P∗0  ε0 −→ P∗  ε.

Moreover, the uniqueness part of the fundamental theorem shows that these
R-chain maps are unique up to R-chain homotopy and that

(fe∗  idM ) ◦ (e
g∗  idM ) 'R idP∗0  idM
g∗  idM ) ◦ (fe∗  idM ) 'R idP∗  idM .
(e

This shows the uniqueness claim.

Corollary 1.6.9 (fundamental theorem of group (co)homology I). Let G be a


group, let A be a ZG-module, and let (P∗ , ε) be a projective resolution of Z
over ZG. Then, for each n ∈ N, there exist canonical isomorphisms

Hn (G; A) ∼
=Z Hn (P∗ ⊗G A)

H (G; A) ∼
n
=Z H n HomG (P∗ , A) .

Proof. By the uniqueness of projective resolutions (Corollary 1.6.8) and be-


cause the simplicial resolution is a projective resolution (Proposition 1.6.5),
there exists a ZG-chain homotopy equivalence f : C∗ (G) −→ P∗ extend-
ing idA (which is canonical up to ZG-homotopy).
Then f ⊗G idA : C∗ (G; A) −→ P∗ ⊗G A and HomG (f, A) : HomG (P∗ , A) −→
C ∗ (G; A) are Z-(co)chain homotopy equivalences. In particular, these (co)chain
homotopy equivalences induce (canonical!) isomorphisms

Hn (f ⊗G idA ) : Hn (G; A) −→ Hn (P∗ ⊗G A)


n
 
H HomR (f, A) : H n HomG (P∗ , A) −→ H n (G; A).

of Z-modules for every n ∈ N.

The challenge now is to find resolutions over the group ring that are well
adapted to the target application.

1.6.3 Example: Finite cyclic groups


As first example, we use the freedom of changing the resolution to compute
the (co)homology of finite cyclic groups:
48 1. The basic view

t2 · e0 t · e1
t · e0
e1

e0

tn−1 · e1

tn−1 · e1

Figure 1.4.: A free Z/n-equivariant CW-structure on S 1 , leading to the res-


olution in Proposition 1.6.11

Definition 1.6.10. Let G be a group and let a ∈ ZG. Then we denote the
associated ZG-homomorphism

ZG −→ ZG
x 7−→ x · a

given by right multiplication with a by Ma .

Study note. Why do we choose right multiplication instead of the more com-
mon left multiplication?

Proposition 1.6.11 (a resolution for finite cyclic groups). Let n ∈ N>0 , let
Pn−1
G := Z/n, let t := [1] ∈ Z/n, and let N := j=0 tj ∈ ZG. Then

MN Mt−1 MN Mt−1 ε
··· / ZG / ZG / ZG / ZG /Z

is a projective resolution of Z (with trivial G-action) over ZG.

Proof. The chain modules are free ZG-modules, whence projective. Clearly,

N · (1 − t) = 0 ∈ ZG.

This shows that the above chain of ZG-homomorphisms is a chain complex.


Exactness follows from concrete computations in finite-dimensional linear
algebra (Exercise).

Remark 1.6.12 (geometric idea). Let n ∈ N≥2 . The cyclic group G := Z/n
acts freely on the circle by rotation about 2·π/n. This leads to a corresponding
free G-equivariant CW-structure on S 1 (Figure 1.4). Looking at the cellular
homology of this CW-complex results in a short exact sequence
1.6. Changing the resolution 49

η Mt−1 ε
0 / H1 (S 1 ; Z) ∼
=Z Z / ZG / ZG /Z∼
=Z H0 (S 1 ; Z) /0

of ZG-modules,
Pn−1 where ε is the usual augmentation and η is multiplication
by N := j=0 tj . Therefore, we can splice these sequences together to obtain
a long exact sequence of the form

MN Mt−1 MN Mt−1 ε
··· / ZG / ZG / ZG / ZG /Z

This is the projective resolution from Proposition 1.6.11.

Corollary 1.6.13 ((co)homology of finite cyclic groups). Let n ∈ N>0 , let G :=


Pn−1
Z/n, let t := [1] ∈ Z/n, let N := j=0 tj , and let A be a ZG-module. Then,
for all k ∈ N,


 AG if k = 0
Hk (G; A) ∼
=Z AG /N · A if k is odd


ker(N : A −→ A)/(t − 1) · A if k > 0 is even

 G
A if k = 0
k ∼
H (G; A) =Z ker(N : A −→ A)/(t − 1) · A if k is odd

 G
A /N · A if k > 0 is even.

In particular, for all k ∈ N>0 ,

Hk+2 (G; A) ∼
=Z Hk (G; A) and H k+2 (G; A) ∼
=Z H k (G; A).

Moreover, (where G acts trivially on Z),


 
Z
 if k = 0 Z
 if k = 0
Hk (G; Z) ∼
=Z Z/n if k is odd and H k (G; Z) ∼
=Z 0 if k is odd

 

0 if k > 0 is even Z/n if k > 0 is even.

Proof. In view of Theorem 1.3.1, we only need to consider (co)homology in


positive degrees. We use the fundamental theorem of group (co)homology
(Corollary 1.6.9) and the periodic projective resolution (P∗ , ε) from Proposi-
tion 1.6.11. We only treat the case of homology in detail (the computation in
cohomology works in the same way; check!). The Z-chain complex P∗ ⊗G A
is isomorphic to

degree 3 2 1 0
Mt−1 ⊗G idA Mt−1 ⊗G idA
··· / ZG ⊗G A / ZG ⊗G AMN ⊗G id/ AZG ⊗G A / ZG ⊗G A
=Z

=Z

=Z

=Z

··· /A /A Pn−1
/A /A
t−1 −1 j=0 t−j t−1 −1
50 1. The basic view

Figure 1.5.: Module (ducky) and a projective resolution (octopus), showing


the true nature of the module/ducky.

Pn−1 P
Moreover, we have j=0 t−j = g∈G g = N and (t−1 − 1) · A = (t − 1) · A.
Taking homology proves the claim.
Study note. It does not make much sense to memorise the formulas in Corol-
lary 1.6.13. Instead, it is much more efficient to memorise the projective res-
olution in Proposition 1.6.11 and how to compute the (co)homology of finite
cyclic groups from this resolution.
Remark 1.6.14. Let n ∈ N≥2 and G := Z/n. Combining the fundamental
theorem of group (co)homology (Corollary 1.6.9) and the computation of
the cohomology of finite cyclic groups (Corollary 1.6.13 has the following
consequence:
There does not exist a projective resolution (P∗ , ε) of the ZG-module Z
(with trivial G-action) satisfying Pk ∼
=R 0 for all large enough k.
In other words, projective resolutions uncover that even though both the ZG-
module Z and the group G look very tame, the module Z is “complicated”
(Figure 1.5).
As sample application of these computations, we consider a purely group-
theoretic result, namely the classification of finite p-groups with a unique
subgroup of order p (Corollary 1.6.18) and the resulting characterisation of
cyclic p-groups (Corollary 1.6.19); for simplicity, we avoid the prime 2.
Theorem 1.6.15 (classification of p-groups with a cyclic subgroup of index p).
Let p ∈ N be an odd prime. Every (finite) p-group that contains a cyclic
subgroup of index p is isomorphic to exactly one of the groups in the following
list:
A. Z/pn for some n ∈ N>0
1.6. Changing the resolution 51

B. Z/pn × Z/p for some n ∈ N>0

C. Z/pn o Z/p for some n ∈ N>1 , where the generator [1] of Z/p acts
on Z/pn by multiplication with 1 + pn−1 .

Before giving the proof, let us observe that

(1 + pn−1 )p ≡ 1 mod pn

for all primes p ∈ N and all n ∈ N>1 (divisibility of binomial coefficients


. . . check!). Hence, the groups of type C in Theorem 1.6.15 indeed exist.

Proof of Theorem 1.6.15. The groups listed in the theorem all fall into dif-
ferent isomorphism classes (we can use the cardinality, the set of orders, and
the property of being Abelian as separating invariants).
We now show that the list is complete:
Let G be a finite p-group that contains a cyclic subgroup A of index p.
Then A is a normal subgroup of G (Lemma 1.6.16). Hence, we obtain an
extension
0 / A i / G π / Z/p /1

where i : A −→ G is the inclusion. As A is cyclic (and a subgroup of a p-


group), we have A ∼ =Group Z/pn for some n ∈ N. Without loss of generality,
we may assume that n > 0 and that A = Z/pn (to simplify notation).
The idea is now to use the classification of extensions in terms of group
cohomology (Theorem 1.5.10). Therefore, we first need to understand the
Z/p-action on A induced by the extension above.
If the Z/p-action on A is trivial, then G is Abelian (because it is then
generated by two commuting elements) and thus of type A or B (by the
classification of finite Abelian groups; Satz III.1.3.1).
If the Z/p-action on A is non-trivial, then we can argue as follows: In view
of the classification of extensions (Theorem 1.5.10) and the characterisation
of trivial extensions (Remark 1.5.11), it suffices to prove the following in order
to establish that G is of type C:

À There exists a (non-trivial) element in Z/p that acts by multiplication


with 1 + pn−1 on A = Z/pn (which shows that the extension is iso-
morphic to one, where [1] ∈ Z/p acts by multiplication with 1 + pn−1
on A).

Á We have H 2 (Z/p; A) ∼
=Z 0.

Ad À. The Z/p-action of [1] ∈ Z/p on A = Z/pn is given by multiplication


by a number a ∈ Z; because the action is non-trivial, we have a 6≡ 1 mod pn .
Therefore, by Lemma 1.6.17, there exists a k ∈ {1, . . . , p − 1} with [ak ] =
[1 + pn−1 ] in Z/pn . Then the commutative diagram
52 1. The basic view

i π / Z/p
0 /A /G /0

multiplication by k

0 /A /G / Z/p /0
i k◦π

shows that we can restrict to the case that [1] acts by multiplication by 1 +
pn−1 on A.
Ad Á. By Corollary 1.6.13, we have

=Z AZ/p /hN iA ,
H 2 (Z/p; A) ∼
Pn−1 j
Pn−1
where N := j=0 [a ]= j=0 [1 + j · pn−1 ] ∈ A. On the one hand, we have

AZ/p = A[1+p
n−1
]
= [x] ∈ Z/pn pn−1 · x ≡ 0 mod pn
= p · Z/pn .

On the other hand, we have (in A) that


p−1
X h p · (p − 1) i
N= [1 + j · pn−1 ] = [p] + · pn−1 = [p],
j=0
2

and so hN iA = p · Z/pn . Therefore, H 2 (Z/p; A) ∼


=Z 0.
This shows that G is of type A, B, or C.

Lemma 1.6.16 (subgroups of prime index). Let p ∈ N be prime, let G be a


finite p-group, and let H ⊂ G be a subgroup of index p. Then H is a normal
subgroup of G.
Proof. This statement can be proved in several ways; for instance: The
group G acts by left translation on the coset space G/H. This action cor-
responds to a group homomorphism ϕ : G −→ Sym(G/H) ∼ =Group Sp into
the corresponding symmetric group. Let K := ker ϕ. We will now show that
K = H, which proves that H as a kernel is normal in G.
By construction, K ⊂ H (as can be seen by evaluation on the coset H).
Moreover,
[G : K] [G : K] | im ϕ|
[H : K] = = = .
[G : H] p p
Because G is a p-group, also | im ϕ| is a power of p. On the other hand, | im ϕ|
divides |Sp | = p!. Because p is prime, it follows that | im ϕ| = 1 or | im ϕ| = p.
The first case cannot occur (because [H : K] is an integer). Hence, | im ϕ| = p,
which implies [H : K] = p/p = 1; thus, H = K.

Lemma 1.6.17 (Z/p-actions on Z/pn ). Let p ∈ N be an odd prime, let n ∈


N>1 , and let a ∈ Z with ap ≡ 1 mod pn .
1.6. Changing the resolution 53

1. Then a ≡ 1 mod pn−1 .

2. If a 6≡ 1 mod pn , then there exists a k ∈ {1, . . . , p − 1} with ak ≡


1 + pn−1 mod pn .

Proof. The first part follows from elementary number theory (Exercise; the
little Fermat might help). The second part is a consequence of the first part
(Exercise).

Corollary 1.6.18 (classification of p-groups with a unique subgroup of order p).


Let p ∈ N be an odd prime and let G be a finite p-group that contains a unique
subgroup with p elements. Then G is cyclic.

Proof. We prove the claim by induction over |G|. By assumption G is non-


trivial and |G| is a power of p.

• Base case. If |G| = p, then G ∼


=Group Z/p, which is cyclic.
• Induction step. Let |G| > p and let us suppose that the claim is already
established for all groups with fewer elements. As p-group, G contains
a normal subgroup N of index p (Satz III.1.3.33 and Satz III.1.3.27).
Because |G| > p, the normal subgroup N is also a p-group with a unique
subgroup with p elements: The subgroup N is a non-trivial p-group and
so contains an element of order p; because N ⊂ G, the corresponding
subgroup with p elements is unique.
Hence, by induction, N is cyclic. Therefore, we can apply Theo-
rem 1.6.15 and thus G is of one of the types listed in the theorem. The
only one of these types that has a unique subgroup with p elements is
type A (which is cyclic). This shows that G is cyclic.

Corollary 1.6.19 (recognising cyclic p-groups). Let p ∈ N be an odd prime and


let G be a finite p-group. Then the following are equivalent:

1. The group G is cyclic.

2. All Abelian subgroups of G are cyclic.

Proof. The implication “1 =⇒ 2” follows from the fact that subgroups of


cyclic groups are cyclic.
Conversely, let us suppose that all Abelian subgroups of G are cyclic.
Without loss of generality, we may assume that G is non-trivial. In view of
Corollary 1.6.18, it suffices to show that G contains a unique subgroup of
size p.
By a classical result, the centre Z(G) of G is non-trivial (Satz III.1.3.33)
and thus contains an element x of order p. Let H := hxiG ⊂ G be the
corresponding subgroup.
Let K ⊂ G also be a subgroup of size p and let y ∈ K be of order p.
Because x is central in G, the subgroup
54 1. The basic view

A := hx, yiG ⊂ G

is Abelian (and a p-group with |A| ≤ p2 ). By assumption, A is cyclic,


whence A = H or A ∼=Z Z/p2 . However, Z/p2 contains only one subgroup of
size p and so K = H.

Outlook 1.6.20 (the odd even prime). For the prime 2, in principle, the same
arguments apply. However, the situation is a little bit more complicated in
the sense that more groups can occur. More precisely [12, Chapter IV.4]:
Every (finite) 2-group that contains a cyclic subgroup of index 2 is iso-
morphic to exactly one of the groups in the following list:

A. Z/2n for some n ∈ N>0

B. Z/2n × Z/2 for some n ∈ N>0

C. Z/2n o Z/2 for some n ∈ N>1 , where the generator [1] of Z/2 acts
on Z/2n by multiplication with 1 + 2n−1

D. dihedral 2-groups. Z/2n oZ/2 for some n ∈ N>2 , where the generator [1]
of Z/2 acts by multiplication by −1

E. generalised quaternion 2-groups. Let H be the quaternion algebra and


let n ∈ N>0 . Then the generalised quaternion group Q2n is the subgroup
n
of the units of H generated by eπ·i/2 and j; the group Q2n can also be
described by the presentation
n
hx, y | y 4 = 1, y 2 = x2 , y · x · y −1 = x−1 i.

F. Z/2n o Z/2 for some n ∈ N>2 , where the generator [1] of Z/2 acts
on Z/2n by multiplication by − + 2n−1 .

Inductively, one then obtains: A finite 2-group that contains a unique


subgroup with two elements is cyclic or a generalised quaternion group.

These classification results can, for instance, be used in the investigation


of free actions on spheres (Chapter 4.3).

1.6.4 Example: Free groups

As second class of examples, we compute the (co)homology of free groups.

Proposition 1.6.21 (a resolution for free groups). Let S be a set and let F be
the free group, freely generated by S. Then

0 0 L ∂ ε
··· /0 /0 /
S ZF / ZF /Z
1.6. Changing the resolution 55

is a projective resolution of Z (with trivial F -action) over ZF . Here,


M
∂: ZF −→ ZF
S
X X
xs · es 7−→ xs · (s − 1).
s∈S s∈S

Proof. Clearly, the chain modules are free (whence projective). We have (a
fact that does not only hold for free groups)

ker ε = SpanZF {s − 1 | s ∈ S} (Exercise)


= im ∂. (by definition of ∂)

Therefore, it remains to show that ∂ is injective (and this is special for free
groups).
P To this end, we will make use of the following notation: If y =
g∈G s g ∈ ZF , then we define the support of y by
y ·

supp y := {g ∈ F | yg 6= 0} ⊂ F.
P
Let x := s∈S xs · es be a non-zero element. Then

X := {y · s | s ∈ S, y ∈ supp xs } ∪ {y | s ∈ S, y ∈ supp xs } ⊂ F

is a finite and non-empty subset of F and supp ∂(x) ⊂ X. Moreover, every


element g ∈ X of maximal length (when viewing the elements of F as reduced
words in S∪S −1 ) satisfies g ∈ supp ∂x; geometrically this is related to vertices
far away from ε in the tree in Figure 1.6. We can verify this claim algebraically
as follows: Let z ∈ X be an element of maximal length. We only need to show
that z can only have one “representation” as an element of X.

• Let z = y · s with s ∈ S and y ∈ supp xs . Let w be the reduced word


over S ∪ S −1 underlying y. Then ws is reduced as well (otherwise y
would be longer than y · s and in X) and z = ws. We need to consider
two cases:
– We have z = y 0 · s0 with s0 ∈ S and y 0 ∈ supp xs0 . The same
argument as above shows that w0 s0 is also the reduced word rep-
resenting z. Hence, w = w0 (whence y = y 0 ) and s = s0 .
– We have z = y 0 with s0 ∈ S and y 0 ∈ supp xs0 . Then w0 ends
with s0−1 (because otherwise y 0 · s0 would be longer than y 0 and
in X), which cannot happen (because we know that the reduced
representative of z is ws, which ends in s and not in s0−1 .

• Let z = y with s ∈ S and y ∈ supp xs ; let w be the reduced word


over S ∪ S −1 representing y. Then w ends in s−1 (because of the max-
imality). We need to consider two cases:
56 1. The basic view

b ba

ab

ε a a2

ab−1

Figure 1.6.: A free F -equivariant CW-structure on the 4-regular tree [53, Fig-
ure 3.10]; blue/red edges are the F -orbits of 1-cells.

– We have z = y 0 · s0 with s0 ∈ S and y 0 ∈ supp xs0 . This cannot


happen (see above).
– We have z = y 0 with s0 ∈ S and y 0 ∈ supp xs0 . Then w0 ends
in s0−1 , whence s = s0 .

This shows the uniqueness of the “representation” of z.


Therefore, supp ∂(x) 6= ∅ and so ∂(x) 6= 0. Thus, ∂ is injective.

Remark 1.6.22 (geometric idea). Let F be a free group of rank 2, freely gener-
ated by a and b. Then F ∼ =Group π1 (X, x0 ), where (X, x0 ) := (S 1 , 1) ∨ (S 1 , 1).
The universal covering X e of X is homeomorphic to the (geometric realisa-
tion of the) 4-regular tree (Figure 1.6) and X e inherits a free F -equivariant
CW-structure from the “obvious” CW-structure on X (with one 0-cell and
two 1-cells). The corresponding cellular chain complex is isomorphic to

··· /0 0 /0 0 / L ZF ∂ / ZF,
S

e is contractible (as a
where ∂ is defined as in Proposition 1.6.21. Moreover, X

tree) and so H∗ (X; Z) =Z H∗ (•; Z). This shows that the above complex leads
to a resolution of Z over ZF .

Corollary 1.6.23 ((co)homology of free groups). Let S be a set, let F be the free
group freely generated by S, and let A be a ZF -module. Then, for all k ∈ N≥2 ,
1.6. Changing the resolution 57

Hk (F ; A) ∼
=Z 0 and H k (F ; A) ∼
=Z 0.

Moreover, we have (where Z carries the trivial action) for all k ∈ N


 

Z if k = 0 
 Z if k = 0
M Y
Hk (F ; Z) ∼
=Z Z if k = 1 and H k (F ; Z) ∼
=Z Z if k = 1

 

 S  S
0 if k > 0 0 if k > 0.

Proof. In view of Theorem 1.3.1, we only need to consider (co)homology


in positive degrees. We use the fundamental theorem of group (co)homology
(Corollary 1.6.9) and the projective resolution (P∗ , ε) from Proposition 1.6.21.
We only treat the case of homology in detail (the computation in cohomology
works in the same way; check!).
Because the projective resolution is trivial in degree bigger than 1, we
know that Hk (F ; A) ∼=Z Hk (P∗ ⊗F A) ∼ =Z 0 for all k ∈ N≥2 . Therefore, it
remains to compute H1 (F ; Z). The corresponding degrees of P∗ ⊗F Z are
isomorphic to (because F acts trivially on Z)

degree 2 1 0

0 ⊗F Z /L ZF ⊗F Z
∂⊗F idZ
/ ZF ⊗F Z
S

=Z
=Z

=Z


L
0 /
S Z /Z
0

Taking homology finishes the proof.


Alternatively, we could compute H1 (F ; Z) and H 1 (F ; Z) also via the meth-
ods from Chapter 1.4.1 (Example 1.4.9) and Chapter 1.4.2. The approach via
the resolution has the advantage that it, in principle, also works with non-
trivial coefficients.

In particular, this also includes a computation of the (co)homology of the


infinite cyclic group Z (which is freely generated by a single element).
As a sample application we compute the deficiency of free groups:

Corollary 1.6.24 (deficiency of free groups). Let S be a finite set. Then the
free group F (S), freely generated by S, has deficiency

def F (S) = |S|.

Proof. On the one hand, the presentation hS | i of F (S) shows that

def F (S) ≥ |S| − 0 = |S|.

On the other hand, we have


58 1. The basic view

def F (S) ≤ rkZ H1 (F (S); Z) − rkZ H2 (F (S); Z) (Corollary 1.5.4)


= |S| − 0. (Example 1.4.9, Corollary 1.6.23)

Therefore, we obtain def F (S) = |S|.

Example 1.6.25. For instance, Corollary 1.6.24 shows that:

• The group described by the presentation hx, y, z | x · y 2019 · z · xi is not


isomorphic to Z (however, we could also use the Abelianisation to see
that).

• The group described by the presentation ha1 , a2 , b1 , b2 | [a1 , b1 ] · [b1 , b2 ]i


has deficiency at least 3 and thus is not isomorphic to a free group of
rank 2 (which we could also derive from the Abelianisation, which is
isomorphic to Z4 ).
Moreover, one can show that this group has deficiency equal to 3 (Exam-
ple 4.1.28), so it is not free of rank 4 (and because of the Abelianisation
also not free of any other rank).
This group has geometric meaning: It is (isomorphic to) the fundamen-
tal group of an oriented closed connected surface of genus 2.

1.7 (Co)Homology and subgroups


We will now consider the following problem: How does the (co)homology
of a subgroup relate to the (co)homology of the ambient group? As a first
step, we will first discuss how we can convert modules of the ambient group
into modules of the given subgroup and vice versa. We then prove Shapiro’s
lemma and then discuss the transfer for finite index subgroups.

1.7.1 Restriction and (co)induction


Restricting the action of an ambient group to a subgroup yields the restriction
functor.

Definition 1.7.1 (restriction). Let G be a group and let H be a subgroup, and


let i : H −→ G be the inclusion. Then we write

ResG ∗
H := i : ZG Mod −→ ZH Mod

for the restriction functor from G to H. In other words, the restriction is


given by forgetting the action of the group elements of G that do not lie
in H.
1.7. (Co)Homology and subgroups 59

Proposition 1.7.2 (projectivity and restriction). Let G be a group and let H be


a subgroup.
1. If P is a projective ZG-module, then ResG
H P is a projective ZH-module.

2. If (P∗ , ε) is a projective resolution of Z over ZG, then (ResG G


H P∗ , ResH ε)
is a projective resolution of Z over ZH.
Proof. Ad 1. We use the characterisation of projective modules as direct
summands of free modules (Proposition 1.6.3. Because the functor ResG H is
compatible with direct sums (check!), we only need to show that ResG
H ZG is
a free ZH-module. By construction,
M
ResG ∼
H ZG =ZH ZH,
G/H

which is a free ZH-module.


Ad 2. Because restricting the module structure preserves exactness, the
second part follows from the first part.
Corollary 1.7.3. Let G be a group that contains an element of non-trivial
finite order. Then there is no projective resolution of Z (with trivial G-action)
over ZG of finite length.
Proof. Let (P∗ , ε) be a projective resolution of Z over ZG. By hypothesis,
there exists a non-trivial finite cyclic subgroup C of G. Then (ResG G
C P∗ , ResC ε)
is a projective resolution of the trivial ZC-module Z over ZC (Proposi-
tion 1.7.2). Therefore, ResG C P∗ (whence also P∗ ) does not have finite length
(Remark 1.6.14).
Conversely, we can use tensor products to extend module structures over
a subgroup to an ambient group. This leads to the induction functor.
Definition 1.7.4 (induction). Let G be a group and let H be a subgroup.
Then the induction functor from H to G is defined by

IndG
H := ZG ⊗ZH · : ZH Mod −→ ZG Mod .

Here, we equip ZG with the ZG-ZH-bimodule structure given by the G-


multiplication on G from the left and the H-multiplication on G from the
right. More explicitly: If B is a left ZH-module, then the G-action on IndG
H (B)
is given by

G × (ZG ⊗ZH B) −→ ZG ⊗ZH B


(g, x ⊗ b) 7−→ (g · x) ⊗ b.

Example 1.7.5 (induction). Let G be a group and let H ⊂ G be a subgroup.


• There is a canonical isomorphism IndG ∼
H ZH = ZG ⊗ZH ZH =ZG ZG.
60 1. The basic view

• We have IndG HLZ = ZG ⊗ZH Z ∼ =ZG Z[G/H], where the G-action


on Z[G/H] := G/H Z is the one induced by the left translation action
of G on the coset space G/H. (If H is a normal subgroup of G, this co-
incides with the left G-action on the group ring Z[G/H] of the quotient
group G/H; therefore, the ambigous notation is not a problem).
Dually, replacing the tensor product by the corresponding Hom-functor,
we can introduce the coinduction:
Definition 1.7.6 (coinduction). Let G be a group and let H be a subgroup.
Then the coinduction functor from H to G is defined by

CoindG
H := HomH (ZG, · ) : ZH Mod −→ ZG Mod .

Here, we equip ZG with the right ZG-module structure induced by right


multiplication of G on G. More explicitly: If B is a left ZH-module, then the
G-action on CoindG H (B) is given by

G × HomH (ZG, B) −→ HomH (ZH, B)



(g, f ) 7−→ x 7→ f (x · g) .

Proposition 1.7.7 ((co)induction for finite index subgroups). Let G be a group


and let H ⊂ G be a finite index subgroup. Then there is a canonical natural
isomorphism
IndG ∼ G
H (B) =ZG CoindH (B)

for all ZH-modules B.


Proof. A straightforward computation shows that the two homomorphisms

ϕ : IndG
H (B) = ZG ⊗ZH B −→ HomH (ZG, B) = CoindH (B)
G

g ⊗ b 7−→ x 7→ χH (x · g) · (x · g) · b
ψ : CoindG G
H B = HomH (ZG, B) −→ ZG ⊗ZH B = IndH (B)
X
f 7−→ g ⊗ f (g −1 )
gH∈G/H

are well-defined and ZG-linear (Exercise). Here, χH : G −→ {0, 1} denotes


the characteristic function of the subset H ⊂ G; it should be noted that the
term following χH (x · g) only makes sense if x · g ∈ H. Moreover, ϕ and ψ
are mutually inverse (Exercise).

1.7.2 The Shapiro lemma

The Shapiro lemma allows to express the (co)homology of a subgroup in


terms of the cohomology of the ambient group – for the price of changing the
coefficient module via (co)induction.
1.7. (Co)Homology and subgroups 61

Theorem 1.7.8 (Shapiro lemma). Let G be a group, let H be a subgroup, let


B be an H-module, and let n ∈ N. Then there are canonical isomorphisms

Hn (H; B) ∼
=Z Hn (G; IndG
H B)

H (H; B) ∼
n
=Z H (G; CoindG
n
H B).

More precisely: Let i : H −→ G be the incusion and let

I : B −→ ZG ⊗ZH B = IndG
H (B)
b 7−→ 1 ⊗ b
C: CoindG
H (B) = HomH (ZG, B) −→ B
f 7−→ f (1);

then I and C are ZH-homomorphisms, (i, I) : (H, B) −→ (G, IndG


H B) is a
morphism in GroupMod, and (i, C) : (H, B) −→ (G, CoindGH B) is a mor-
phism in GroupMod* . The induced homomorphisms

Hn (i; I) : Hn (H; B) −→ Hn (G; IndG


H B)

H n (i; C) : H n (G; CoindG n


H B) −→ H (H; B)

are Z-isomorphisms.

Proof. The proof for group homology basically consists of the cancellation
of ⊗G ZG as well as the fundamental theorem of group cohomology (Corol-
lary 1.6.9):
Because the restriction functor turns projective resolutions into projective
resolutions (Proposition 1.7.2), (ResG G
H C∗ (G), ResH ε) is a projective resolu-
tion of the trivial ZH-module Z over ZH. Hence, there is a canonical isomor-
phism

Hn (H; B) ∼ =Z Hn ResG H C∗ (G) ⊗H B (Corollary 1.6.9)


=Z Hn C∗ (G) ⊗G (ZG ⊗ZH B) (induced by I)
G

= Hn C∗ (G) ⊗G IndH B
= Hn (G; IndG
H B).

More precisely, C∗ (i) : C∗ (H) −→ ResGH C∗ (G) is a chain map that extends
the identity of Z. Because both sides are projective resolutions, C∗ (i) is al-
ready “the” ZH-chain homotopy equivalence. This induces the first isomor-
phism in the above computation. The second isomorphism is induced by I
(and the canonical projection from ⊗H to ⊗G ). Therefore, the whole isomor-
phism coincides with Hn (i; I).
Similarly, the version for cohomology can be proved using the cancellation
of HomG ( · , HomH (ZG, · )) (check!).
62 1. The basic view

Corollary 1.7.9 (homological characterisation of the trivial group). Let G be a


group that satisfies Hn (G; A) ∼
=Z Hn (1; A) for all ZG-modules A and all n ∈
N. Then G is trivial.
Proof. Let C be a cyclic subgroup of G. Then

C∼
=Z H1 (C; Z) (C is Abelian; Theorem 1.4.1)

=Z H1 (G; IndG Z) (Shapiro lemma; Theorem 1.7.8)
C

=Z 0 (by assumption).

As every non-trivial element of G would generate a non-trivial cyclic subgroup


of G, we obtain that G is the trivial group.
Corollary 1.7.10. Let G be a group and let H be a subgroup of finite index.
Then, for all n ∈ N, there is a canonical isomorphism

H n (G; ZG) ∼
=Z H n (H; ZH).

Proof. By Shapiro’s lemma (Theorem 1.7.8), we have

H n (H; ZH) ∼
=Z H n (G; CoindG
H ZH).

Because the subgroup H has finite index in G, we can convert the coinduction
into an induction: CoindG ∼ G
H ZH =Z IndH ZH (Proposition 1.7.7). Therefore,
we obtain

H n (H; ZH) ∼
=Z H n (G; IndG
H ZH)
∼ n
=Z H (G; ZG), (Example 1.7.5)

as claimed.
Study note. Why is the corresponding result for H∗ (G; ZG) not exciting?
Outlook 1.7.11 (geometric meaning of H 1 (G; ZG)). Let G be an infinite,
finitely generated group. Then H 1 (G; ZG) is related to the number e(G)
of ends of G. The number of ends of G is the number of path-connected
components “at infinity” of (geometric realisations of Cayley graphs of) G [53,
Chapter 8.2]. One then has [32, Theorem 13.5.5][80]:
• e(G) = 1 if and only if rkZ H 1 (G; ZG) = 0,
• e(G) = 2 if and only if rkZ H 2 (G; ZG) = 1,
• e(G) = ∞ if and only if rkZ H 2 (G; ZG) = ∞,
• and there are no other cases.
Typical examples of these three cases are Z2 , Z, and the free group of rank 2,
respectively (Exercise). The number of ends does not change when passing
to a finite index subgroup; algebraically, this is reflected in Corollary 1.7.10.
1.7. (Co)Homology and subgroups 63

1.7.3 Transfer
For subgroups of finite index, there are also “wrong-way” maps, the transfer
maps:
H∗ (H; ResG
H A)
O
transfer Hn (incl.;idA )

H∗ (G; A)
Transfer maps usually are defined through some averaging process. Firstly,
the maps going in the “right” direction are the (co)restriction maps:

Definition 1.7.12 ((co)restriction). Let G be a group, let H ⊂ G be a sub-


group, let i : H −→ G be the inclusion, and let A be a ZG-module. Then we
define the
G
• restriction map as resG ∗ ∗ ∗
H := H (i; idA ) : H (G; A) −→ H (H; ResH A);
G
• corestriction map as corG
H := H∗ (i; idA ) : H∗ (H; ResH A) −→ H∗ (G; A).

For the “wrong-way” maps, we will use the following hands-on description;
one should note that even though the maps on the chain level are only well-
defined/unique up to (co)chain homotopy, after passage to (co)homology,
we obtain well-defined, canonical homomorphisms (because (co)homology is
chain homotopy invariant).

Definition 1.7.13 (transfer). Let G be a group, let H ⊂ G be a finite index


subgroup (with inclusion i : H −→ G), let A be a ZG-module, and let n ∈ N.

• Homological transfer. The transfer map


G
trG G
H := resH : Hn (G; A) −→ Hn (H; ResH A)

is the map on homology induced by composing the averaging map with


the “canonical” chain homotopy equivalence:

/ ResG G C∗o (i)⊗H idA


C∗ (G) ⊗G A H C∗ (G) ⊗H ResH A 'Z
C∗ (H) ⊗H ResG
HA

 X
x⊗a / (g −1 · x) ⊗ (g −1 · a)
gH∈G/H

• Cohomological transfer. The transfer map


G
trG G n n
H := corH : H (H; ResH A) −→ H (G; A)
64 1. The basic view

is the map on cohomology induced by composing the “canonical”


cochain homotopy equivalence with the averaging map:

HomH (C∗ (i),A)


HomH (C∗ (H), ResG
H A)
o
'Z
HomH (ResG G
H C∗ (G), ResH A)
/ HomG (C∗ (G), A)

 X 

f / x 7→ g · f (g −1 · x) .
gH∈G/H

Remark 1.7.14 (alternative descriptions of the transfer). For every description


of group (co)homology, there is a corresponding description of the transfer
maps. For example, one can eliminate the concrete choice of projective resolu-
tion and hide the explicit averaging with the help of the fundamental theorem
(Corollary 1.6.9) and Proposition 1.7.7. Moreover, using covering theory, one
can also give a topological description of the transfer (Proposition 4.1.30).

Theorem 1.7.15 (transfer). Let G be a group, let H ⊂ G be a subgroup of


finite index, let A be a ZG-module, and let n ∈ N. Then

corG G
H ◦ resH (α) = [G : H] · α

holds for all α ∈ Hn (G; A) and all α ∈ H n (G; A).

Study note. In Theorem 1.7.15, in the homological case, resG H is the transfer
map; in the cohomological case, corGH is the transfer map. This slight abuse
of notation allows us to state properties of the transfer in a uniform way.

Proof. We prove only the statement in homology; the cohomological case


is similar. We can check this on the chain level. On the chain level, the
G
composition corG G G
H ◦ resH = corH ◦ trH is modelled (up to chain homotopy) by
(check!)

C∗ (G) ⊗G A −→ C∗ (G) ⊗G A
X X
x ⊗ a 7−→ (g −1 · x) ⊗ (g −1 · a) = x ⊗ a,
gH∈G/H gH∈G/H

which is [G : H] times the identity.

The transfer shows that rationally all finite groups look like the trivial
group:

Corollary 1.7.16. Let G be a finite group and let R be a commutative ring with
unit (with trivial G-action) in which |G| is invertible. Then, for all n ∈ N>0 :

Hn (G; R) ∼
=Z 0 and H n (G; R) ∼
=Z 0.

In particular, Hn (G; Q) ∼
=Z 0 and H n (G; Q) ∼
=Z 0 for all n ∈ N>0 .
1.7. (Co)Homology and subgroups 65

Proof. We only consider homology; the cohomological case works in the same
way. We apply the transfer (Theorem 1.7.15) to the trivial subgroup H = 1:
Then [G : H] · idHn (G;R) factors over Hn (1; R) ∼ =Z 0. Because [G : H] is
invertible in R, we obtain that also idHn (G;R) is the zero map.
Moreover, we can deduce torsion results of the following type from the
transfer formula (Theorem 1.7.15).
Example 1.7.17 (some torsion results).
• The group SL2 (Z) contains a free group of index 12 [53, Proposi-
tion 4.4.2]. Therefore, the transfer (Theorem 1.7.15) and the vanishing
of the (co)homology of free groups in higher degrees (Corollary 1.6.23)
shows that for all k ∈ N≥2 :
 
12 · Hk SL2 (Z); Z ∼ =Z 0 and 12 · H k SL2 (Z); Z ∼ =Z 0.

• The infinite dihedral group D∞ (i.e., the isometry group of Z with


respect to the standard metric on R) is isomorphic to a semi-direct
product of the form Z o Z/2 and thus contains an infinite cyclic group
of index 2 (Exercise). Therefore, for all k ∈ N≥2 :

2 · Hk (D∞ ; Z) ∼
=Z 0 and 2 · H k (D∞ ; Z) ∼
=Z 0.

Outlook 1.7.18 (classical transfer). The (co)homological transfer is a gener-


alisation of the classical, group-theoretic, transfer: Let G be a group and
let H ⊂ G be a subgroup of finite index. Then the classical transfer is the
well-defined(!) homomorphism

Gab −→ Hab
[G:H]
Y 
[g] 7−→ gk · g · R(gk · g)−1 ,
k=1

where {g1 , . . . , g[G:H] } is a set of representatives of H \G, and where R : G −→


{g1 , . . . , g[G:H] } selects for g ∈ G the representative of the coset H · g.
Under the canonical natural isomorphism between H1 ( · ; Z) and the
Abelianisation functor (Theorem 1.4.1), the transfer H1 (G; Z) −→ H1 (H; Z)
then corresponds to the classical transfer (check!).
Hence, the (co)homological transfer can be viewed as a generalisation of
the classical transfer.
Outlook 1.7.19 (Legendre symbol). Let p ∈ N be an odd prime. Then the
transfer map on H1 ( · ; Z) for the subgroup {−1, +1} of the multiplicative
group (Z/(p))× coincides with the Legendre symbol associated with p; this
is a consequence of the Gauß lemma on quadratic resdiues (Exercise).
66 1. The basic view
2
The geometric view

We will now turn to a more geometric setting, viewing groups as (geo)metric


objects. Therefore, we will first briefly introduce notions from geometric group
theory; in particular, we will consider the class of amenable groups.
We will then focus on two aspects:

• On the one hand, we will see how suitable choices of coefficients lead
to geometric invariance properties of group homology.
• On the other hand, we will modify the simplicial cochain complex in a
metric way.

In both settings, we will investigate the contrasting behaviour of amenable


groups vs. free groups.

Overview of this chapter.

2.1 Foundations: Geometric group theory 68


2.2 Uniformly finite homology 75
2.3 Bounded cohomology 86

Running example. amenable groups, free groups


68 2. The geometric view

2.1 Foundations: Geometric group theory


The aim of geometric group theory is to investigate the interplay between
geometric and algebraic properties of groups [35, 37, 39, 25, 53]. One flavour
of this theory starts by viewing groups as geometric objects.

2.1.1 Quasi-isometry
Using word metrics, we can view groups as metric spaces:

Definition 2.1.1 (word metric). Let G be a group and let S ⊂ G be a gener-


ating set. Then the word metric of G with respect to S is given by

dS : G × G −→ N ⊂ R≥0
(g, h) 7−→ min{n ∈ N | ∃s1 ,...,sn ∈S∪S −1 g −1 h = s1 · · · · · sn }.

Study note. A straightforward computation shows that the word metric on


a group G with respect to a generating set S of G indeed is a metric (check!).
What would happen if the subset S ⊂ G is not a generating set?

Example 2.1.2 (word metrics). Let n ∈ N and let S ⊂ Zn be the standard


basis. Then the word metric dS on Zn is the `1 -metric on Zn ⊂ Rn . The
word metrics on Z associated with the generating sets {2, 2019} and Z are
very different from that (the latter one even has finite diameter).

Remark 2.1.3 (visualisation of word metrics). A convenient way to visualise


word metrics is through Cayley graphs: Let G be a group and let S ⊂ G
be a generating set. Then the Cayley graph of G with respect to S is the
(undirected) graph Cay(G, S) whose

• vertex set is G and whose



• edge set is {g, g · s} {g ∈ G, s ∈ S ∪ S −1 ) \ {e} .

Then the word metric dS on G is the graph metric of Cay(G, S) on its set
of vertices (i.e., all edges have length 1 and the distance between to vertices
is the length of a shortest path) (check!). A few prototypical examples are
shown in Figure 2.1.

The word metric does depend on the chosen generating set. But all word
metrics associated with finite generating sets on a given group lead to the
same large-scale geometry. Large-scale geometry arises by relaxing the notion
of isometry by allowing for uniformly bounded error terms.
2.1. Foundations: Geometric group theory 69

b ba
[2] [1] (−2, 2) (−1, 2) (0, 2) (1, 2) (2, 2)

ab
−2 −1 0 1 2 (−2, 1) (−1, 1) (0, 1) (1, 1) (2, 1)

ε a a2
[3] [0]
ab−1
(−2, 0) (−1, 0) (0, 0) (1, 0) (2, 0)

(−2, −1) (−1, −1) (0, −1) (1, −1) (2, −1)

[4] [5]
(−2, −2) (−1, −2) (0, −2) (1, −2) (2, −2)

   
Cay Z/6, {[1]} Cay Z, {1} Cay Z2 , {(1, 0), (0, 1)} Cay ha, b | i, {a, b}

Figure 2.1.: Some Cayley graphs

Definition 2.1.4 (bilipschitz embedding/equivalence). Let (X, dX ) and (Y, dY )


be metric spaces and let f : X −→ Y be a map.

• The map f is a bilipschitz embedding if there exists a c ∈ R>0 with


1 
∀x,x0 ∈X · dX (x, x0 ) ≤ dY f (x), f (x0 ) ≤ c · dX (x, x0 ).
c

• The map f is a bilipschitz equivalence if it is a bilipschitz embedding


and if there exists a bilipschitz embedding g : Y −→ X with

f ◦ g = idY and g ◦ f = idX .

• Metric spaces are bilipschitz equivalent if there exists a bilipschitz equiv-


alence between them.

Study note. How can bilipschitz equivalences be viewed as isomorphisms in


a suitable category?

Proposition 2.1.5. Let G be a finitely generated group and let S, T ⊂ G be


finite generating sets of G. Then the identity map (G, dS ) −→ (G, dT ) is a
bilipschitz equivalence.

Proof. As S is finite, we can consider the (finite) maximum

c := max dT (e, s).


s∈S∪S −1

Let g, h ∈ G and n := dS (g, h). Then there exist s1 , . . . , sn ∈ S ∪ S −1


with g −1 · h = s1 · · · · · sn . Using the triangle inequality and the fact that the
word metric dT is left-invariant (check!), we find
70 2. The geometric view

dT (g, h) = dT (g, g · s1 · · · · · sn )
≤ dT (g, g · s1 ) + dT (g · s1 , g · s1 · s2 )
+ · · · + dT (g · s1 · · · · · sn−1 , g · s1 · · · · · sn ) (triangle inequality)
= dT (e, s1 ) + dT (e, s2 ) + · · · + dT (e, sn ) (left-invariance)
≤c·n (construction of c)
= c · dS (g, h). (definition of n)

Similarly, we can estimate dS in terms of dT .


Therefore, we can also safely talk about finitely generated groups being
bilipschitz equivalent to a metric space. However, as bilipschitz equivalences
are bijections, we will not be able to relate the geometry of finitely generated
groups to the geometry of Euclidean or hyperbolic spaces in a meaningful
way. Thus, one allows additionally a uniform additive error:
Definition 2.1.6 (quasi-isometric embedding, quasi-isometry). Let (X, dX ) and
(Y, dY ) be metric spaces and let f : X −→ Y be a map.
• The map f is a quasi-isometric embedding if there exist c ∈ R>0 and
b ∈ R≥0 with

1 
∀x,x0 ∈X · dX (x, x0 ) − b ≤ dY f (x), f (x0 ) ≤ c · dX (x, x0 ) + b.
c

• A map f 0 : X −→ Y has finite distance from f (is uniformly close to f )


if there is a c ∈ R≥0 with

∀x∈X dY f (x), f 0 (x) ≤ c.

• The map f is a quasi-isometry if it is a quasi-isometric embedding and


if there exists a quasi-isometric embedding g : Y −→ X such that f ◦ g
and g ◦ f have finite distance from idY and idX , respectively.
• The metric spaces X and Y are quasi-isometric if there exists a quasi-
isometry X −→ Y . In this case, we write X ∼QI Y .
Study note. If you know about homotopy categories: How can quasi-isometries
be viewed as isomorphisms in a suitable homotopy category?
Sometimes, it is convenient to use the following characterisation of quasi-
isometries:
Proposition 2.1.7 (alternative characterisation of quasi-isometry). Let (X, dX )
and (Y, dY ) be metric spaces. A quasi-isometric embedding f : X −→ Y is
a quasi-isometry if and only if it has quasi-dense image, i.e., if there exists
a c ∈ R>0 such that

∀y∈Y ∃x∈X dY f (x), y ≤ c.
2.1. Foundations: Geometric group theory 71

Proof. If f : X −→ Y is a quasi-isometry, then there exists a quasi-inverse g


of f . Hence, there is a c ∈ R>0 satisfying

∀y∈Y dY f ◦ g(y), y ≤ c.

Therefore, f has quasi-dense image.


For the converse implication, one applies the axiom of choice to obtain a
quasi-inverse [53, Proposition 5.1.10].

Corollary 2.1.8. Let G be a finitely generated group and let S, T ⊂ G be


finite generating sets of G. Then the identity map (G, dS ) −→ (G, dT ) is a
quasi-isometry.
Proof. Every bilipschitz equivalence is a quasi-isometry. Therefore, we can
apply Proposition 2.1.5.
Therefore, we can also safely talk about finitely generated groups being
quasi-isometric to a metric space.
Example 2.1.9.
• Let n ∈ N. The group Zn is quasi-isometric to the space (Rn , `1 -metric),
and whence also to the Euclidean space (Rn , Euclidean metric). In par-
ticular: In general, quasi-isometries are neither injective nor surjective.
• All finite groups are quasi-isometric to the trivial group (check!). More-
over, every finitely generated group quasi-isometric to the trivial group
is finite (check!).
Caveat 2.1.10 (subgroup distortion). In general, injective group homomor-
phisms between finitely generated groups are not quasi-isometric embeddings
(distances in the target group can be significantly shorter than in the domain
group [53, Exercise 6.E.6]).
One of the central goals of geometric group theory is to understand finitely
generated groups up to quasi-isometry. Therefore, it is useful to develop quasi-
isometry invariants. Important examples of quasi-isometry invariants are, for
instance [53]:
• number of ends
• growth type (and whence also: containing a nilpotent subgroup of finite
index (!) [35])
• hyperbolicity
• amenability (Remark 2.1.18, Corollary 2.2.20)
• ...
In Chapter 2.2, we will see an example of a quasi-isometry invariant extracted
from group homology; in particular, this will allow us to separate some quasi-
isometry types of finitely generated groups.
72 2. The geometric view

2.1.2 Amenability

The notion of amenability is based on (almost) invariance properties and has


applications in various fields [73][66][53, Chapter 9]. We will use the classical
definition via invariant means:
Definition 2.1.11 (amenable group). Let G be a group. The group G is
amenable if it admits a left-invariant mean. A left-invariant mean on G is an
R-linear map m : `∞ (G, R) −→ R with the following properties:
• Normalisation. We have m(1) = 1.
• Positivity. We have m(f ) ≥ 0 for all functions f ∈ `∞ (G, R) that
satisfy f ≥ 0 (pointwise).
• Left-invariance. For all g ∈ G and all f ∈ `∞ (G, R), we have

m(g · f ) = m(f ).

Example 2.1.12 (finite groups are amenable). Let G be a finite group. Then
G is amenable; this is witnessed by the invariant mean

`∞ (G, R) −→ R
1 X
f 7−→ · f (x).
|G|
x∈G

Example 2.1.13 (Abelian groups are amenable). Every Abelian group is


amenable; the existence of an invariant mean can, for example, be shown
via the Markov-Kakutani fixed point theorem [53, Proposition 9.1.3].
Moreover, there is the following, slightly more geometric, argument: For
simplicity, we will only treat the case Z. For n ∈ N let

Fn := {−n, . . . , n} ⊂ Z.

P is now for f ∈ ` (Z, R) to average the values over Fn via fn :=



The idea
1/|Fn |· x∈Fn f (x) and then to take the “limit” n → ∞; the sequence (fn )n∈N
is bounded (because f is bounded), but in general not convergent. We will
therefore need an appropriate notion of limit: Let ω be a non-principal ultra-
filter on N (such things exist in suitably rich set theory). We then consider

m : `∞ (Z, R) −→ R
1 X
f 7−→ lim · f (x).
n∈ω |Fn |
x∈Fn

The limit limn∈ω along ω allows to choose “systematically” accumulation


points in bounded sequences.
2.1. Foundations: Geometric group theory 73

A straightforward computation shows that m is normalised and positive.


Left-invariance of m follows from the following almost invariance property
of (Fn )n∈N : If g ∈ Z, then
1
lim · |Fn M g · Fn | = 0.
n→∞ |Fn |

Therefore, m is a left-invariant mean on Z, which shows that Z is amenable.


Similarly, one can argue for all finitely generated Abelian groups.
Outlook 2.1.14 (Følner sequences). Let G be a finitely generated group and
let S ⊂ G be a finite generating set of G. A Følner sequence for G is a se-
quence (Fn )n∈N of finite non-empty subsets of G with the following property:
For all r ∈ N, we have
1
lim · |∂rS Fn | = 0;
n→∞ |Fn |

for a finite set F ⊂ G, we define the r-boundary (with respect to S) by



∂rS F := x ∈ G \ F ∃f ∈F dS (x, f ) ≤ r .

Then the following are equivalent [66][53, Chapter 9.2.1]:


• The group G is amenable.
• The group G admits a Følner sequence (with respect to S).
Example 2.1.15 (non-amenability of free groups of higher rank). Free groups of
rank at least 2 are not amenable: For notational simplicity, we only consider
the case F := ha, b | i of a free group of rank 2.
Assume for a contradiction that F were amenable, i.e., that there existed
a left-invariant mean m : `∞ (F, R) −→ R on F . Let A ⊂ F be the set of
reduced words ending with a non-trivial power of a. Then

A ∪ A · a−1 = F.

Using the properties of the mean m, we obtain

1 = m(1)
= m(χA∪(A·a−1 ) )
≤ m(χA ) + m(χA·a−1 )
= m(χA ) + m(a · χA )
= m(χA ) + m(χA ) (left-invariance)
= 2 · m(χA ),

and so m(χA ) ≥ 1/2. However, the sets A, A·b and A·b2 are pairwise disjoint;
therefore,
74 2. The geometric view

1 = m(1)
≥ m(χA∪A·b∪A·b2 )
= m(χA ) + m(χA·b ) + m(χA·b2 )
= 3 · m(χA ) (left-invariance)
3
≥ ,
2
which is impossible. Hence, F is not amenable.
Proposition 2.1.16 (inheritance properties of amenable groups).
1. Subgroups of amenable groups are amenable.
2. Homomorphic images of amenable groups are amenable.
3. Let
1 /N /G /Q /1

be an extension of groups. Then G is amenable if and only if N and Q


are amenable.
In particular: Every group that contains a solvable subgroup of finite index is
amenable.
Proof. These inheritance properties can, e.g., be shown by transforming the
corresponding invariant means [53, Proposition 9.1.6].

Outlook 2.1.17 (the von Neumann problem). Our previous discussion shows
that: A group that contains a free subgroup of rank 2 is not amenable. Con-
versely, the von Neumann problem asked whether the converse also holds.
By now, it is known that the converse does not hold; i.e., there exist
non-amenable groups that do not contain a free subgroup of rank 2 [63].
However, for an early candidate of such an example (namely, Thompson’s
group F , which is known to not contain any free subgroups of rank 2), it is
still an open problem to decide whether this group is amenable or not!
The von Neumann problem had a big influence on the development of ge-
ometric group theory and measurable group theory. Moreover, it turns out
that converse statements of these type do hold in certain geometric situa-
tions [90, 30, 77].

Remark 2.1.18 (quasi-isometry invariance of amenability). Let G and H be


finitely generated groups with G ∼QI H. If G is amenable, then also H is
amenable.
One can show by direct computation that Følner sequences on G can
be converted into Følner sequences on H [53, Theorem 9.3.1]. Alternatively,
we can also invoke the homological characterisation of amenability and the
quasi-isometry invariance of uniformly finite homology (Corollary 2.2.20).
2.2. Uniformly finite homology 75

2.2 Uniformly finite homology


We will now give an example of a quasi-isometry invariant of finitely gener-
ated groups of homological origin. In general, group homology with (trivial)
Z-coefficients is not quasi-isometry invariant:

Example 2.2.1.

• The group Z/2 is quasi-isometric to the trivial group, but H1 (Z/2; Z) ∼


=Z
Z/2 ∼
6 Z0∼
= =Z H1 (1; Z).
• The group Z is quasi-isometric to the infinite dihedral group D∞ ,
but H3 (Z; Z) ∼=Z 0 and H3 (D∞ ; Z) ∼
6 Z 0 (the latter can be seen by
=
using a retraction from D∞ to Z/2; check!).

• Free groups of rank 2 contain free groups of rank 3 of finite index (in-
dex 2; Example AT.2.3.49, Corollary AT.5.3.13); therefore, free groups
of rank 2 and 3 are quasi-isometric (check!), but H1 (ha, b | i; Z) ∼6 Z
=
H1 (ha, b, c | i; Z).

In contrast, taking different coefficients changes the situation: As we will


see, group homology with `∞ ( · ; Z)-coefficients is quasi-isometry invariant
(Theorem 2.2.12). In order to prove this fact, we will show that group homol-
ogy with these coefficients admits a purely geometric description in terms of
so-called uniformly finite homology. Moreover, we will characterise amenabil-
ity in terms of uniformly finite homology.

2.2.1 Uniformly finite homology of spaces


As first step, we introduce a large-scale homology for metric spaces, namely
uniformly finite homology. For simplicity, we will restrict to the case of so-
called (countable) UDBG spaces.

Definition 2.2.2 (UDBG space). Let (X, d) be a metric space.

• The metric space (X, d) is uniformly discrete if there are uniform gaps
between all points, i.e., if

inf d(x, x0 ) x, x0 ∈ X, x 6= x0 > 0.

• The metric space (X, d) has bounded geometry if balls of fixed radius
are uniformly bounded, i.e., if

∀r∈R>0 ∃Kr ∈N ∀x∈X BrX,d (x) ≤ Kr .
76 2. The geometric view

A UDBG space is a uniformly discrete metric space with bounded geometry


that consists only of countably many points (countability is automatic if the
metric does not take the value ∞).
Example 2.2.3 (UDBG spaces).
• If G is a finitely generated group and S ⊂ G is a finite generating set,
then (G, dS ) is a UDBG space (check!).
• The Euclidean space (R2 , Euclidean metric) is neither uniformly dis-
crete nor of bounded geometry.
Uniformly finite homology is defined via the uniformly finite chain com-
plex; the basic idea is as follows: Tuples in a space can be viewed as “sim-
plices” (described by their vertices). We then form chains of such simplices.
As in all other constructions of this type, we need a finiteness condition to
make the simplicial boundary operator well-defined. In our situation, we will
allow chains with infinitely many simplices and impose the following finite-
ness conditions:
• The coefficients have to be uniformly bounded.
• The simplices have uniformly bounded size.
More precisely: In the following, a normed ring is a commutative ring with
unit with a multiplicative norm (e.g., Z or R).
Proposition and Definition 2.2.4 (uniformly finite chain complex). Let R be
a normed ring and let (X, d) be a UDBG space. Then the uniformly finite
chain complex C∗uf (X; R) of (X, d) is the N-indexed R-chain complex defined
by:
• For n ∈ N, we write Cnuf (X; R) for the R-module of all bounded func-
tions c : X n+1 −→ R with the following property: There is an r ∈ R>0
with
supp c ⊂ {x ∈ X n+1 | diam x ≤ r}.
Here, supp c := {x ∈ X n+1 | c(x) 6= 0} is the support of c and
diam x := maxj,k∈{0,...,n} d(xj , xk ) denotes the diameter of tuples in X.
The elements of Cnuf (X; R) are called uniformly finite n-chains in X
n+1
with coefficients in R; usually,
P such functions c : X −→ R are de-
noted as “sums” of the form x∈X n+1 c(x) · x.
• If n ∈ N>0 , then

∂n : Cnuf (X; R) −→ Cn−1


uf
(X; R)
X X X n
cx · x 7−→ (−1)j · cx · (x0 , . . . , x
bj , . . . , xn )
x∈X n+1 x∈X n+1 j=0
2.2. Uniformly finite homology 77

describes a well-defined R-linear map and this turns C∗uf (X; R) into an
R-chain complex.
Proof. We first have to show that the boundary operator is well-defined;
even though the notation is suggestive, as the “sums” are not necessarily
finite, this is not just a “linear extension” of the usual definition on the
standard
P basis! Therefore, we first have to correct the description of ∂n : Let
c = x∈X n+1 cx · x ∈ Cnuf (X; R). Then

X n −→ R
k
X X
y 7−→ (−1)j · cx
j=0 x∈{z∈X n+1 |(z0 ,...,b
zj ,...,zn )=y}

uf
is a well-defined chain in Cn−1 (X; R) (Exercise). This is the chain that is
meant by the (more suggestive, but less correct) description in the claim.
That ∂n ◦ ∂n+1 = 0 holds follows as in the standard argument in the
simplicial world (check! see also Remark 1.2.2).
Definition 2.2.5 (uniformly finite homlogy). Let R be a normed ring, let (X, d)
be a UDBG space, and let n ∈ N. Then, we define the uniformly finite
homology of X with coefficients in R in degree n by (where ∂0 := 0)

uf
ker ∂n : Cnuf (X; R) → Cn−1 (X; R)
Hnuf (X; R) := uf (X; R) → C uf (X; R)
 ∈ Ob(R Mod).
im ∂n+1 : Cn+1 n

Example 2.2.6 (uniformly finite homology of the point). Let R be a normed


ring, let • be “the” one-point metric space, and let n ∈ N. The standard
computation shows that
(
uf ∼R R if n = 0
Hn (•; R) =
0 if n > 0

(check!). Indeed, this is the same computation as for the homology of the
trivial group (Example 1.2.8) or the singular homology of the one-point space.
Proposition and Definition 2.2.7 (functoriality of uniformly finite homology).
Let R be a normed ring and let (X, dX ), (X, dY ) be UDBG spaces.
1. Let f : (X, dX ) −→ (Y, dY ) be a quasi-isometric embedding. Then, for
each n ∈ N,

Cnuf (f ; R) : Cnuf (X; R) −→ Cnuf (Y ; R)


X X 
cx · x 7−→ cx · f (x0 ), . . . , f (xn )
x∈X n+1 x∈X n+1

describes a well-defined R-linear map.


78 2. The geometric view

The corresponding sequence C∗uf (f ; R) is an R-chain map C∗uf (X; R) −→


C∗uf (Y ; R) and we denote the induced map on uniformly finite homology
by H∗uf (f ; R) : H∗uf (X; R) −→ H∗uf (Y ; R).
2. This construction is functorial (in the sense that it maps identities to
identities and is compatible with composition).
3. If f, f 0 : (X, dX ) −→ (Y, dY ) are uniformly close quasi-isometric em-
beddings, then C∗uf (f ; R) 'R C∗uf (f 0 ; R). In particular, H∗uf (f ; R) =
H∗uf (f 0 ; R).
Proof. Ad 1. Again, we first have
Pto convert the description in the proposition
into a proper definition. If c = x∈X n+1 cx · x ∈ Cnuf (X; R), then

Y n+1 −→ R
X
y 7−→ cx
x∈{z∈Z n+1 |(f (z0 ),...,f (zn ))=y}

is a well-defined chain in Cnuf (Y ; R) (check!). Moreover, the standard compu-


tation shows that this construction is compatible with the boundary operators
on C∗uf (X; R) and C∗uf (Y ; R), respectively (check!).
Ad 2. It is clear that C∗uf (idX ; R) is the identity map on C∗uf (X; R) and
that this construction is compatible with composition (check!).
Ad 3. As f and f 0 are uniformly close it is not hard to see that (hn )n∈N ,
given by

hn : Cnuf (X; R) −→ Cn+1


uf
(Y ; R)
X X X n

cx · x 7−→ (−1)j · cx · f (x0 ), . . . , f (xj ), f 0 (xj ), . . . , f 0 (xn )
x∈X n+1 x∈X n+1 j=0

for all n ∈ N, is an R-chain homotopy between C∗uf (f ; R) and C∗uf (f 0 ; R)


(check!). This is a quasi-geometric version of the prism construction in the
proof of homotopy invariance of singular homology (Lemma AT.4.2.2).
Corollary 2.2.8 (QI-invariance of uniformly finite homology). Let R be a normed
ring, let f : (X, dx ) −→ (Y, dY ) be a quasi-isometry between UDBG spaces,
and let n ∈ N. Then Hnuf (f ; R) : Hnuf (X; R) −→ Hnuf (Y ; R) is an R-iso-
morphism.
Proof. This is a direct consequence of functoriality of H∗uf ( · ; R) with respect
to quasi-isometric embeddings and uniform closeness (Proposition 2.2.7).
Example 2.2.9 (uniformly finite homology of finite spaces). Let (X, d) be a
finite metric space. Then X is a UDBG space and X ∼QI •. Hence, quasi-
isometry invariance of uniformly finite homology (Corollary 2.2.8) shows
that H∗uf (X; R) ∼
=R H∗uf (•; R) for every normed ring R. The latter homology
has been computed in Example 2.2.6.
2.2. Uniformly finite homology 79

2.2.2 Uniformly finite homology of groups

As second step, we relate group homology with `∞ ( · ; R)-coefficients to uni-


formly finite homology. The conversion between uniformly finite chains and
simplicial chains with function-coefficients is based on viewing the coefficients
on an orbit of a simplex as values of a function on the given group.
Remark 2.2.10 (uniformly finite homology of finitely generated groups). Let
R be a normed ring, let G be a finitely generated group, and let S, T ⊂
G be finite generating sets. Because idG : (G, dS ) −→ (G, dT ) is a quasi-
isometry (Corollary 2.1.8), Proposition 2.2.7 shows that C∗uf ((G, dS ); R) and
C∗uf ((G, dT ); R) coincide. We will simply write C∗uf (G; R) and H∗uf (G; R) for
the corresponding uniformly finite chain complex and the uniformly finite
homology of G, respectively.
Theorem 2.2.11 (group homology as uniformly finite homology [10]). Let R be
a normed ring and let G be a finitely generated group. Then the maps

Cn G; `∞ (G, R) −→ Cnuf (G; R)
X
(x0 , . . . , xn ) ⊗ f 7−→ f (y) · (y · x0 , . . . , y · xn )
y∈G

Cnuf (G; R) −→ Cn G; `∞ (G, R)
X X
cx · x 7−→ (e, z1 , . . . , zn ) ⊗ (y 7→ c(y,y·z1 ,...,y·zn ) )
x∈Gn+1 z∈Gn

are well-defined for each n ∈ N and form mutually inverse R-chain isomor-
phisms C∗uf (G; R) ←→ C∗ (G; `∞ (G, R)). In particular, there are canonical
isomorphisms Hn G; `∞ (G, R) ∼ =R Hnuf (G; R).
Proof. This follows from lengthy, but straightforward, computations (check!
do you see how the finiteness conditions correspond to each other?).
Corollary 2.2.12 (a homological QI-invariant). Let R be a normed ring, let G
and H be finitely generated groups with G ∼QI H, and let n ∈ N. Then, every
quasi-isometry G −→ H induces an isomorphism
 
Hn G; `∞ (G, R) ∼ =R Hn H; `∞ (H, R) .

Proof. If f : G −→ H is a quasi-isometry, then we have



Hn G; `∞ (G, R) ∼ =R Hnuf (G; R) (Theorem 2.2.11)
∼ uf
=R Hn (H; R) (Corollary 2.2.8 for Hnuf (f ; R))


=R Hn H; `∞ (H, R) (Theorem 2.2.11)

as claimed.
80 2. The geometric view

2.2.3 Application: Ponzi schemes and amenability

As an application, we give a homological characterisation of amenability [8],


which has applications in geometric group theory.
Theorem 2.2.13 (homological characterisation of amenability). Let G be a
finitely generated group. Then the following are equivalent:
1. The group G is amenable.

2. H0 G; `∞ (G, R) ∼ 6 R0
=

3. H0 G; `∞ (G, Z) ∼6 Z0
=
In order to prove this theorem, we use the connection with uniformly finite
homology and a closer investigation of uniformly finite homology in degree 0.
Definition 2.2.14 (fundamental class in uniformly finite P
homology). Let R be
a normed ring and let (X, d) be a UDBG space. Then x∈X 1 · x is a cycle
in C0uf (X; R) and the corresponding uniformly finite homology class is the
fundamental class of X with R-coefficients:
X 
[X]R := 1 · x ∈ H0uf (X; R).
x∈X

Theorem 2.2.15 (fundamental class in uniformly finite homology). Let (X, d)


be a UDBG space. Then the following are equivalent:
1. We have H0uf (X; Z) ∼
=Z 0.
2. We have [X]Z = 0 in H0uf (X; Z).
3. We have [X]R = 0 in H0uf (X; R).
4. We have H0uf (X; R) ∼
=R 0.
Proof. The implications 1 =⇒ 2 and 4 =⇒ 3 are clear. Moreover, 2 =⇒ 3
holds (because every integral uniformly finite boundary also is a real uni-
formly finite bonudary).
Therefore, it suffices to prove the following implications:
• Ad 2 =⇒ 1 and 2 =⇒ 4. Let [X]Z = 0 in H0uf (X; Z). Then, by the
tail lemma (Lemma 2.2.16), there exists a family (tx )x∈X of chains
(X; Z) such that ∂1 tx = 1 · x in C0uf (X; Z) for all x ∈ X and such
in C1uf P
that “ x∈X tx ” is a well-defined chain in C1uf (X; Z).
Let R be a normed ring (e.g., R
Por Z) and α ∈ufH0 (X; R). Then α is
uf

represented by some chain c = x∈X cx · x ∈ C0 (X; R); hence,


2.2. Uniformly finite homology 81
X
b := cx · tx
x∈X

describes a well-defined chain in C1uf (X; R) (check!) with ∂1 b = c.


Therefore, α = 0 ∈ H0uf (X; R).
• Ad
P 3 =⇒ 2. Let [X]R = 0 in H0 (X; R); let b ∈ C1 (X; R) with ∂1 b =
uf uf

x∈X 1·x. As in the proof of Lemma 2.2.16, we may assume that b ≥ 0.


We then consider the floored version
eb : X 2 −→ Z
 
x 7−→ (2 · K + 1) · b(x) ,

where K is defined as follows: Because b is a uniformly finite chain,


there exists an r ∈ N with diam x ≤ r for all x ∈ supp b. As X is a
UDBG space, we can define the number

K := max |BrX,d (x)| x ∈ X .

Clearly, eb is a uniformly finite chain in C1uf (X; Z).


We now set e c := ∂1eb ∈ C0uf (X; Z) and show that supp e
c = X: Let x ∈ X.
e
In (∂1 b)(x), at most 2 · K coefficients of b contribute (by construction
of K; check!). Therefore, we obtain the improved floor estimate

c(x) = (∂1eb)(x)
e
X X
= eb(y, x) − eb(x, y) (definition of ∂1eb)
y∈X y∈X
X X 
≥ (2 · K + 1) · b(y, x) − b(x, y) − 2 · K · 1 (improved floor estimate)
y∈X y∈X

= (2 · K + 1) · (∂1 b)(x) − 2 · K (definition of ∂1 b)


X
= (2 · K + 1) · 1 − 2 · K = 1, (because ∂1 b = 1 · x)
x∈X

and thus x ∈ supp e


c.
Applying the tail lemma to ec yields a family (tx )x∈X of tails in C(uf X; Z)
for all of supp e
c = X with
X  X
∂1 tx = 1 · x,
x∈X x∈X

which shows that [X]Z = 0 in H0uf (X; Z).


Lemma 2.2.16 (the tail lemma by Block and Weinberger). Let (X, d) be a
UDBG space. If c ∈ C0uf (X; Z) is a uniformly finite chain with c ≥ 0 (i.e., all
82 2. The geometric view

y5 ...
(y6 , y5 ) y6
(y5 , y4 )

y3
(y4 , y3 ) y4
(y3 , y2 )

x (y1 , y0 ) y1 (y2 , y1 ) y2

Figure 2.2.: a tail for x

coefficients of c are non-negative) and [c] = 0 in H0uf (X; Z), then there exists
a family (tx )x∈supp c of chains in C1uf (X; Z) with the following properties:

• For all x ∈ supp c, we have ∂1 tx = 1 · x in C0uf (X; Z)

• and
X X X 
tx = tx (y) · y
x∈supp c y∈X 2 x∈supp c

describes a well-defined chain in C1uf (X; Z).

P
Proof. Let b = x∈X 2 bx ·x ∈ C1uf (X; Z) with ∂1 b = c. Using ∂1 (1·(x0 , x1 )) =
−∂1 (1 · (x1 , x0 )) for all x ∈ X 2 , we may assume without loss of generality
that bx ≥ 0 for all x ∈ X 2 (check!).
We extract the tails tx out of b by (double) induction. Because X is count-
able, we can enumerate all points of supp c and treat one point of supp c at
a time.
Let x ∈ supp c; we set y0 := x. Because ∂1 b = c and b ≥ 0 (and b is
integral), there exists a y1 ∈ X \ {y0 } with

b(y1 ,y0 ) ≥ 1.

Then y1 ∈ supp(c−1·y0 +1·y1 ) and ∂1 (b−b(yP 1 ,y0 )


·(y1 , y0 )) = c−1·y0 +1·y1 .
Hence, inductively, we obtain a “tail” tx := n∈N 1 · (yn+1 , yn ) (Figure 2.2)
with tx ≤ b (pointwise). So tx indeed is a well-defined uniformly finite chain
in C1uf (X; Z). By construction, we have

∂1 tx = 1 · x, c − 1 · x ≥ 0, supp c \ {x} ⊂ supp(c − 1 · x) ⊂ supp c,


∂1 (b − tx ) = c − 1 · x, b − tx ≥ 0, supp(b − tx ) ⊂ supp b.

By induction over supp c, we can construct tails for all points in supp c. By
construction, the accumulation of these tails is contained in the non-negative
2.2. Uniformly finite homology 83

1 1 1

1
1
1
1

1 1
1 1

1
1 1 1
1 1
1

1 1 1
1

1 1 1 1

1 1 1 1

1 1 1 1

1 1 1
1

1 1
1 1 1 1

1 1
1 1

1
1

1
1

1 1 1

Figure 2.3.: a Ponzi scheme on the rooted 3-regular tree

P
uniformly finite chain b and whence their “sum” “ x∈X tx ” is a well-defined
uniformly finite chain in C1uf (X; Z) with the desired properties.

Block and Weinberger proposed the following interpretation of the vanish-


ing of the fundamental class in uniformly finite homology [8]:

Remark 2.2.17 (Ponzi schemes in UDBG spaces). Let X be a UDBG space


P 0 in H0 (X; R). Then there exists a chain b ∈ C1 (X; R)
uf uf
with [X]R =
with ∂1 b = x∈X 1 · x; without loss of generality, we may assume that all
coefficients of b are non-negative (by “swapping edges” if necessary).
We now view b as a flow of money: If (x0 , x1 ) ∈ X, then x0 sends the
amount b(x0 , x1 ) to x1 . The fact that
X
∂1 b = 1·x−0
x∈X

holds means that afterwards every point in X has something (namely 1) even
though they had nothing in the beginning (namely 0). The chain b pushes
the problem of generating enough money to infinity (in particular, the total
amount of money in this system is infinite . . . ). Figure 2.3 shows such a flow
84 2. The geometric view

for the UDBG space given by the vertices of the rooted 3-regular tree (with
the path-metric induced by the graph structure).
A famous example of real-world schemes that “worked” in a similar way
are the so-called Ponzi schemes (Exercise).
Finally, we can complete the proof of the homological characterisation of
amenability:
Proof of Theorem 2.2.13. Ad 1 =⇒ 2. Let G be amenable, i.e., G admits
a left-invariant mean m : `∞ (G, R) −→ R. We now use the mean m for a
transfer-like argument:
Let i : R −→ `∞ (G, R) be the injective R-linear map given by viewing
scalars as constant functions on G (this map is ZG-linear!). Because m is
normalised, we have m ◦ i = idR . Moreover, (idG , i) and (idG , m) are mor-
phisms in GroupMod (check!). Therefore, we obtain for every n ∈ N

Hn (idG ; m) ◦ Hn (idG ; i) = Hn (idG ; m ◦ i) = Hn (idG ; idR ) = idHn (G;R) .

In particular, H0 (idG ; i) : H0 (G; R) −→ H0 (G; `∞ (G, R)) is injective. As


H0 (G; R) ∼
=R R, we obtain H0 (G; `∞ (G, R)) 6∼=R 0.
Ad 2 ⇐⇒ 3. This equivalence follows from the interpretation in terms of
uniformly finite homology (Theorem 2.2.11) and Theorem 2.2.15.
Ad 2 =⇒ 1. Let H0 (G; `∞ (G, R)) 6∼ =R 0. By the general computation of
group homology in degree 0 (Theorem 1.3.1), this means that W 6= `∞ (G, R),
where 
W := SpanR f − g · f f ∈ `∞ (G, R) .
In oder to find a left-invariant mean on G, we will first show that 1 6∈ W
(where we take the closure with respect to the supremum norm | · |∞ on the
Banach space `∞ (G, R)); more precisely, we will show that the distance

d := inf |1 − f |∞ f ∈ W ∈ R≥0

from 1 to W equals 1: We clearly have d ≤ |1 − 0|∞ = 1 (because 0 ∈ W ).


Assume for a contradiction that there exists an f ∈ W with |1 − f |∞ < 1.
This means that ε := inf x∈G f (x) > 0.
Using the translation of Theorem 2.2.11, we therefore obtain a chain c ∈
C0uf (G; R) with c ≥ ε and [c] = 0 ∈ H0uf (G; R). Now the same argument
as in the proof of the implication 3 =⇒ 2 of Theorem 2.2.15 (applied
to the chain 1/ε · c) shows that this already implies H0 (G; `∞ (G, R)) ∼
=R
H0uf (G; R) ∼
=R 0, which contradicts our hypothesis.
Hence, d = 1. Therefore, the Hahn-Banach theorem [72, Theorem 5.16]
shows that there exists an R-linear map m : `∞ (G, R) −→ R with the follow-
ing properties:
• We have m(1) = 1 and m|W = 0.

• We have m(f ) ≤ d1 · |f |∞ = |f |∞ for all f ∈ `∞ (G, R).
2.2. Uniformly finite homology 85

The condition m|W = 0 implies that m is left-invariant. Moreover, one can


deduce from these properties that m is positive (Exercise). Hence, m is a
left-invariant mean for G and thus G is amenable.
Remark 2.2.18. Let G be a finitely generated group. Then Theorem 2.2.13
and Theorem 1.3.1 show (and the same arguments also apply to `∞ (G, R)):
• If G is amenable, then `∞ (G, Z)G ∼
6 Z 0. In fact, in many cases, this
=
group is huge [7].
• If G is non-amenable, then `∞ (G, Z)G ∼
=Z 0.
Remark 2.2.19 (Ponzi schemes on groups). Combining Theorem 2.2.13 and
Remark 2.2.17 shows: A finitely generated group admits a Ponzi scheme if and
only if it is not amenable. This is compatible with the geometric intuition that
non-amenable groups have enough branching/space on the way to “infinity”.
Corollary 2.2.20 (quasi-isometry invariance of amenability). Let G and H be
finitely generated groups with G ∼QI H. If G is amenable, then also H is
amenable.
Homological proof. This is a direct consequence of the fact that uniformly
finite homology characterises amenability (Theorem 2.2.13 and 2.2.11) and
that uniformly finite homology is quasi-isometry invariant (Corollary 2.2.8).

Outlook 2.2.21 (QI-invariance of the Hirsch rank). One can use an amenable
transfer argument in uniformly finite homology and the quasi-isometry in-
variance of uniformly finite homology to show that Zn ∼QI Zm if and only
if n = m (Exercise).
More generally, one can also use uniformly finite homology to give an
alternative proof of the fact that the Hirsch rank is a quasi-isometry invariant
of finitely generated (virtually) nilpotent groups [7].

Outlook 2.2.22 (quasi-isometries vs. bilipschitz equivalences). Let f : X −→ Y


be a quasi-isometry between UDBG spaces. Then one can show that the
following are equivalent [90]:
1. The map f is uniformly close to a bilipschitz equivalence.
2. We have H0uf (f ; Z)[X]Z = [Y ]Z in H0uf (Y ; Z).
In particular, because H0uf ( · ; Z) is trivial on non-amenable groups, every
quasi-isometry between non-amenable groups is uniformly close to a bilips-
chitz equivalence(!).
In contrast, there do exist infinite finitely generated amenable groups that
are quasi-isometric but not bilipschitz equivalent [26].

Outlook 2.2.23 (homological invariants in measured group theory). Taking suit-


able coefficients also leads to homological invariants for different notions of
“equivalence”, such as orbit/measure equivalence [29, 74, 54].
86 2. The geometric view

2.3 Bounded cohomology


Now, in a slightly different direction, we will metrically modify the definition
of the simplicial cochain complex (by introducing a boundedness condition).
This leads to bounded cohomology, which has various applications in group
theory, geometric group theory, and manifold geometry [36, 43, 44, 61, 27, 50].

2.3.1 Bounded cohomology of groups

Bounded cohomology is a functional analytic twin of ordinary group cohomol-


ogy; in order to describe its construction in terms of the simplicial resolution,
we need normed/Banach versions of the ordinary theory.

Definition 2.3.1 (Banach G-module). Let G be a group.

• A normed G-module is a (real) normed vector space V together with


an isometric (left) G-action on V .

• A Banach G-module is a (real) Banach space V together with an iso-


metric (left) G-action on V (i.e., a complete normed G-module).

• Let V and W be normed [Banach] G-modules. A morphism V −→ W


of normed [Banach] G-modules is an R-linear map ϕ : V −→ W that
– is G-equivariant, i.e., ϕ(g · x) = g · ϕ(x) for all g ∈ G and all x ∈ V ,
and that
– is bounded, i.e., there exists a C ∈ R≥0 with

∀x∈V ϕ(x) ≤ C · kxk.

(The infimum of all such bounds C is the norm kϕk of ϕ).


We write BHomG (V, W ) for the space of all morphisms V −→ W of
normed [Banach] G-modules (which is a Banach space with respect to
the operator norm as soon as W is Banach).

The category of normed and Banach G-modules is denoted by G Norm and


*
G Ban, respectively. Moreover, we write GroupBan for the Banach version
*
of GroupMod .

Example 2.3.2 (the normed simplicial resolution). Let G be a group. For n ∈ N,


we can equip CnR (G) := R ⊗Z Cn (G) with the `1 -norm associated with the
basis Gn+1 , i.e.,
2.3. Bounded cohomology 87

| · |1 : CnR (G) −→ R≥0


X X
ag · g 7−→ |ag |.
g∈Gn+1 g∈Gn+1

Then the boundary operator idR ⊗Z ∂n : CnR (G) −→ Cn−1


R
(G) is a bounded
operator (of norm at most n + 1; check!). Hence, we obtain a chain com-
plex C∗R (G) in the category of normed G-modules. Moreover, group homo-
morphisms induce chain maps on C∗R that consist of bounded operators of
norm at most 1.
In this Banach setting, we replace the Hom-functors and dual modules
with their bounded versions and then proceed as before:
Example 2.3.3 ((bounded) dual). Let G be a group and let V be a (left)
Banach G-module. Then V represents the corresponding functor

BHomG ( · , V ) : G Norm −→ Ban

to the category Ban of Banach spaces (and bounded linear maps).


Definition 2.3.4 (bounded cohomology). Let G be a group and let V be a
(left) Banach G-module.

• Then we write Cb∗ (G; V ) := BHomG C∗R (G), V ∈ Ban Ch* for the
bounded cochain complex of G with coefficients in V .
• For n ∈ N, we define bounded cohomology of G with coefficients in V
in degree n by

Hbn (G; V ) := H n Cb∗ (G; V ) ∈ R Mod .

Let (ϕ, Φ) : (G, V ) −→ (H, W ) be a morphism in GroupBan* .


• We write Cb∗ (ϕ; Φ) := BHomG (C∗R (ϕ), Φ) for the composition

can. incl. / BHomG (C∗ (ϕ),Φ)


R 
BHomH C∗R (H), W ) BHomG ϕ∗ C∗R (H), ϕ∗ W / BHomG C∗R (G), V

of (degree-wise) bounded cochain maps.


• For n ∈ N, we then set (which is R-linear)

Hbn (ϕ; Φ) := H n Cb∗ (ϕ; Φ) : Hbn (H; W ) −→ Hbn (G; V ).

Remark 2.3.5 (additional structure on bounded cohomology). By construction,


bounded cohomology in each degree is not only an R-vector space, but a
semi-normed R-vector space (with respect to the semi-norm induced by the
operator norm on the bounded cochain complex). This structure is useful in
the context of simplicial volume (Chapter AT.4.4.5) and topological rigidity
of volume [36, 50, 27].
88 2. The geometric view

Example 2.3.6 (bounded cohomology of the trivial group). Let 1 be “the”


trivial group and let V be a Banach space. Then Cb∗ (1; V ) ∼
=R Ch C ∗ (1; V )
(check!), and therefore, for all n ∈ N,
(
V if n = 0
Hb (1; V ) ∼
n
=R
0 if n > 0.

Remark 2.3.7 (bounded cohomology in degree 0). The same argument as in the
proof of the computation of group cohomology in degree 0 (Theorem 1.3.1)
shows that the functor Hb0 is canonically naturally isomorphic to the invari-
ants functor GroupBan* −→ Ban (check!).
Remark 2.3.8 (bounded cohomology in degree 1). The same argument as in
the computation of H 1 ( · ; R) (Theorem 1.4.13) shows that for each group G
we have (check!):

Hb1 (G; R) ∼
=R ϕ ∈ HomGroup (G, R) sup |ϕ(g)| < ∞ (bounded version of Theorem 1.4.13)
g∈G

=R 0 (R has no non-trivial bounded subgroups)

In particular, in general, the comparison map Hb∗ ( · ; R) −→ H ∗ ( · ; R) is not


surjective (for example H 1 (Z; R) ∼
6 R 0).
=
Bounded cohomology in degree 2 turns out to be interesting: it is related
to quasi-morphisms (Theorem 2.3.17).
Remark 2.3.9 (comparison map). The inclusion of bounded G-equivariant
bounded linear maps into G-equivariant linear maps induces a natural trans-
formation
Hb∗ ( · ; · ) =⇒ H ∗ ( · ; · )
(check!), the so-called comparison map.
In general, the comparison map is neither injective nor surjective (Corol-
lary 2.3.19, Remark 2.3.8).

2.3.2 Application: A characterisation of amenability

Theorem 2.3.10 (cohomological characterisation of amenability). Let G be a


group. Then the following are equivalent:
1. The group G is amenable.
2. For all right Banach G-modules V and all n ∈ N≥1 , we have

Hbn (G; V # ) ∼
=R 0.

3. For all right Banach G-modules V , we have Hb1 (G; V # ) ∼


=R 0.
2.3. Bounded cohomology 89

Notation 2.3.11. For a right Banach G-module V , we write V # for the Ba-
nach G-module consisting of the Banach space of bounded linear function-
als V −→ R together with the left G-action

G × V # −→ V #

(g, f ) 7−→ x 7→ f (x · g) .

Proof of Theorem 2.3.10. Ad 1 =⇒ 2. Let G be an amenable group and n ∈


N≥1 . In order to keep notation simple, we only show that Hbn (G; R) ∼ =R 0,
where G acts trivially on R. We argue by transfer:
As G is amenable, G admits a left-invariant mean m : `∞ (G, R) −→ R. We
then consider the corresponding transfer t∗ : Cb∗ (G; R) −→ C ∗ := (C∗R (G))# ,
given by
# 
tk : C k = CkR (G) −→= BHomG CkR (G), R Cbk (G; R)

f 7−→ (g0 , . . . , gk ) 7→ m x 7→ f (x · g0 , . . . , x · gk )

in each degree k; then tk is a bounded linear map (check!) that is compatible


with the simplicial coboundary operators (check!) and that satisfies

t∗ ◦ i∗ = idCb∗ (G;R)

(check!), where i∗ : Cb∗ (G; R) −→ C ∗ is the inclusion. In particular, the iden-


tity of Hbn (G; R) factors through H n (C ∗ ):

H n (t∗ ) ◦ H n (i∗ ) = H n (t∗ ◦ i∗ ) = idHbn (G;R) .

Therefore, it suffices to show that H n (C ∗ ) ∼


=R 0 holds: The standard chain
contraction for C∗ (G) (given by cones; proof of Proposition 1.6.5) induces a
chain contraction for C∗R (G), which is degreewise bounded with respect to the
`1 -norm (check!). Applying the functor · # leads to a well-defined cochain
contraction for C ∗ (check!). In particular, H n (C ∗ ) ∼
=R 0 (because n ≥ 1).
For the general case, we instead make use of the transfer map given by

BHom CkR (G), V # −→ Cbk (G; V # )

f 7−→ (g0 , . . . , gk ) 7→ v 7→ m(x 7→ f (x · g0 , . . . , x · gk )(v))

and show that the cochain complex BHom CkR (G), V # is contractible.
Ad 2 =⇒ 3. This is clear.
Ad 3 =⇒ 1. We consider the Banach G-module (check!)

V := `∞ (G, R)/C,

where C ⊂ `∞ (G, R) is the subspace of constant functions (which is isomor-


phic to R and closed).
90 2. The geometric view

The condition Hb1 (G; V # ) ∼=R 0 can be used to show that there exists a
bounded R-linear functional µ : `∞ (G, R) −→ R with µ(1) = 1 that is (left)
G-invariant (Exercise); unfortunately, it is not a priori clear that the norm
of µ is 1 (equivalently, that µ is positive).
One can then apply a decomposition argument from functional analysis to
improve this to a bounded linear functional m : `∞ (G, R) −→ R with m(1) =
1 that is positive: Similarly to the Hahn decomposition of signed measures of
finite total variation [45, Chapter 7.1.2] there is a “minimal” decomposition

µ = µ+ − µ−

into positive functionals µ+ , µ− −→ `∞ (G, R) −→ R. In addition, this de-


composition is unique in a certain sense; this uniqueness is strong enough to
show that also µ+ has to be left-invariant. Because of µ(1) = 1, we know that
µ+ 6= 0. Hence, a suitable normalisation of µ+ is a left-invariant mean on G.
The left-invariant mean m shows that G is amenable.
Caveat 2.3.12. Amenable groups cannot be characterised using bounded co-
homology only with trivial R-coefficients: There exist many examples of non-
example groups G with Hb∗ (G; R) ∼=R Hb∗ (1; R) [57, 52].
Study note (transfer). Which transfer arguments in group (co)homology do
you know by now? Collect and compare them!
The characterisations of amenability in terms of bounded cohomology
and uniformly finite homology complement each other and are actually re-
lated [10].

2.3.3 Application: Quasi-morphisms


One classical application of bounded cohomology concerns quasi-morphisms.
A quasi-morphism G −→ R on a group G is a map that almost satisfies the
homomorphism condition:
Definition 2.3.13 (quasi-morphism). Let G be a group.
• A quasi-morphism on G is a map ϕ : G −→ R such that the defect

D(ϕ) := sup ϕ(g · h) − ϕ(g) − ϕ(h)
g,h∈G

is finite.
• A quasi-morphism ϕ : G −→ R is trivial if it is uniformly close to a
homomorphism, i.e., if there exists a group homomorphism ψ : G −→ R
with
sup ϕ(g) − ψ(g) < ∞.
g∈G
2.3. Bounded cohomology 91

• A quasi-morphism ϕ : G −→ R is homogeneous, if for each g ∈ G and


each n ∈ Z we have ϕ(g n ) = n · ϕ(g).
• We write

QM(G) for the R-vector space of all quasi-morphisms on G


QM0 (G) for the R-vector space of all trivial quasi-morphisms on G
QM(G) for the R-vector space of all homogeneous quasi-morphisms on G.

One can now wonder whether every quasi-morphism (i.e., a map that al-
most satisfies the homomorphism condition) has to be trivial (i.e., uniformly
close to a homomorphism). Bounded cohomology answers this question (The-
orem 2.3.17). Homogeneous quasi-morphisms allow to express the difference
between quasi-morphisms and group homomorphisms in a simplified way:
Proposition 2.3.14 (homogenisation of quasi-morphisms). Let G be a group
and let ϕ : G −→ R be a quasi-morphism.
1. Then the following map is a homogeneous quasi-morphism on G that is
uniformly close to ϕ:

ϕ : G −→ R
1
g 7−→ lim · ϕ(g n )
n→∞ n

2. Hence, we obtain an isomorphism of R-vector spaces:

QM(G)/ QM0 (G) −→ QM(G)/ HomGroup (G, R)


[ϕ] 7−→ [ϕ]

Proof. The first part follows from Lemma 2.3.15 and a straightforward com-
putation (Exercise). The second part follows from the first part and the fact
that the homogenisation of a trivial quasi-morphism is a group homomor-
phism (Exercise).
Lemma 2.3.15 (convergence of normalised sequences). Let (an )n∈N be a se-
quence of real numbers.
1. If supn,m∈N |an+m − an − am | is finite, then the sequence (an /n)n∈N>0
converges.
2. If the sequence (an )n∈N is non-negative and an+m ≤ an + am for
all n, m ∈ N, then the sequence (an /n)n∈N>0 converges and

1 1
lim · an = inf · an .
n→∞ n n∈N>0 n

Proof. This is elementary analysis (Exercise).


92 2. The geometric view

Proposition 2.3.16 (counting quasi-morphism). Let F := ha, b | i be “the” free


group of rank 2 (described by reduced words) and let w ∈ F .

1. Then the following map is a quasi-morphism on F , the counting quasi-


morphism associated with w, where #(w, g) denotes the number of (pos-
sibly overlapping) occurrences of the word w in the reduced word g:

ψw : F −→ R
g 7−→ #(w, g) − #(w−1 , g)

2. If w = ab, then ψ w is not a group homomorphism (and hence ψw is


not a trivial quasi-morphism).

Proof. Ad 1. Let g, h ∈ F . By construction, we have

D(g, h) := ψw (g · h) − ψw (g) − ψw (h)


= #(w, g · h) − #(w−1 , g · h)
− #(w, g) + #(w−1 , g)
− #(w, h) + #(w−1 , h).

The word gh, in general, is not reduced; therefore, we have to be careful with
counting occurrences of w and w−1 at the “end” of g and the “beginning”
of h. Let r be the tail/initial part of g and h, respectively, that is deleted
when forming the product g · h; i.e., g = g 0 r and h = r−1 h0 are reduced
decompositions and g 0 h0 is the reduced word that represents g · h.
Then the only terms that can contribute to D(g, h) are occurrences of w
or w−1 that intersect with r in g or h but are not completely contained in r
(Figure 2.4); the number of these occurrences can be bounded in terms of the
length `(w) of w. More precisely: For all x, y ∈ F for which xy is reduced, we
have
#(w, xy) − #(w, x) − #(w, y) ≤ `(w).

Therefore, we obtain

D(g, h) = #(w, g 0 h0 ) − #(w−1 , g 0 h0 )
− #(w, g 0 r) + #(w−1 , g 0 r)

− #(w, r−1 h0 ) + #(w−1 , r−1 h0 )
≤ 6 · `(w)

by resolving all six reduced decompositions and the triangle inequality


(check!).
Ad 2. It suffices to prove that ψ w : F −→ R is not a group homomorphism
(in view of Proposition 2.3.14). By construction,
2.3. Bounded cohomology 93

g·h g

Figure 2.4.: bounding the defect of counting quasi-morphisms

1
ψ w (a) = lim · ψw (an ) = 0
n
n→∞
1
ψ w (b) = lim · ψw (bn ) = 0.
n→∞ n

As {a, b} is a generating set of F , if ψ w were a group homomorphism, then


ψ w = 0. However,
1  1
ψ w (ab) = lim · ψw (ab)n = lim · (n − 0) = 1 6= 0.
n→∞ n n→∞ n

Hence, ψw is not a trivial quasi-morphism.


Theorem 2.3.17 (quasi-morphisms via bounded cohomology). Let G be a
group. Then the sequence

HomGroup (G, R)
incl. / QM(G) ∆ / H 2 (G; R)comparison
/ H 2 (G; R)
b map

of R-vector spaces is exact, where ∆ is the map induced by (when describing


bounded cohomology via the bar complex)
2
∆ : QM(G) −→ C b (G; R)

ϕ 7−→ [g1 | g2 ] 7→ ϕ(g1 · g2 ) − ϕ(g1 ) − ϕ(g2 ) .

Proof. We briefly comment on the use of the bar complex: As in the situation
of ordinary cohomology, we may also describe bounded cohomology in terms
R
of the bar complex (to simplify notation). Let C ∗ (G) := R ⊗Z C ∗ (G) be the
R-version of the bar complex and let
∗ R 
C b (G; R) := BHomG C ∗ (G), R

be the corresponding bounded dual. Then the isomorphism from Remark 1.2.4
carries over to this bounded setting and shows that we can compute Hb2 (G; R)

as H 2 (C b (G; R)) (check!).
94 2. The geometric view

We now proceed in the following steps:


À The maps ∆ and ∆ are well-defined: Let ϕ : G −→ R be a quasi-
morphism. Then the map
R
Φ : C 2 (G) −→ R
[g1 | g2 ] 7−→ ϕ(g1 · g2 ) − ϕ(g1 ) − ϕ(g2 )

is bounded (with respect to the norm obtained from the `1 -norm


2
on C2R (G)) and G-equivariant (by construction); hence, Φ ∈ C b (G; R).
∗ ∗
Moreover, Φ is a cocycle in C b (G; R), because (in C (G; R); check!)
1 
Φ = δ [g] 7→ −ϕ(g)
2 1
and δ ◦ δ = 0.
Á Computation of the kernel of ∆: Clearly, HomGroup (G, R) ⊂ ker ∆. Con-
1
versely, let ϕ ∈ ker ∆, i.e., there exists a bounded cochain f ∈ C b (G; R)
1 1
with δ f = δ ([g] 7→ −ϕ(g)). This shows that

G −→ R

g 7−→ ϕ(g) + f [g]

is a group homomorphism (check!) that is uniformly close to ϕ (because


f is bounded). Therefore, ϕ is a trivial homogeneous quasi-morphism,
whence a group homomorphism G −→ R (Proposition 2.3.14).
 Computation of the image of ∆: The argument in step À shows that
the image of ∆ lies in the kernel of the comparison map (check!). Con-
2
versely, let f ∈ C b (G; R) be a bounded cocycle whose associated class
in ordinary group cohomology H 2 (G; R) is zero. Then there exists a
1
chain b ∈ C (G; R) with
1 
f = δ (b) = [g1 | g2 ] 7→ b([g1 ]) − b([g1 · g2 ]) + b([g2 ]) .

Because f is bounded, the map ϕ := (g 7→ −b([g])) : G −→ R is a


quasi-morphism and ∆(ϕ) = f . Therefore, ∆(ϕ) = [f ] in Hb2 (G; R)
(check!).
We will now apply this theorem in two directions: On the one hand, we
will use it to show that amenable groups do not admit non-trivial quasi-
morphisms. On the other hand, we will use it to exhibit non-trivial bounded
cohomology for free groups of higher rank.
Corollary 2.3.18 (quasi-morphisms on amenable groups). Let G be an amenable
group. Then all quasi-morphisms on G are trivial: QM(G) = HomGroup (G, R).
2.3. Bounded cohomology 95

Proof. By Theorem 2.3.17, QM(G)/ HomGroup (G, R) is isomorphic to the


kernel of the comparison map cG : Hb2 (G; R) −→ H 2 (G; R). Because G is
amenable, we know that Hb2 (G; R) ∼
=R 0 (Theorem 2.3.10) and so the kernel
of cG is trivial as well.

Corollary 2.3.19 (bounded cohomology of free groups). Let F be a free


group of rank at least 2. Then Hb2 (F ; R) is non-trivial and the comparison
map Hb2 (F ; R) −→ H 2 (F ; R) is not injective.

Proof. Because F retracts onto a free group of rank 2, we only need to con-
sider the case of ha, b | i (check!).
As the free group ha, b | i admits non-trivial (counting) quasi-morphisms
(Proposition 2.3.16), the description of Hb2 ( · ; R) in terms of quasi-morphisms
(Theorem 2.3.17) shows that Hb2 (ha, b | i; R) 6∼ =R 0 and that the comparison
map Hb2 (ha, b | i; R) −→ H 2 (ha, b | ; iR) is not injective.

Caveat 2.3.20. The computation in Corollary 2.3.19 shows that bounded


cohomology is, in general, not restricted by the length of projective resolu-
tions. The proof of the fundamental theorem of group cohomology (Corol-
lary 1.6.9) does not directly carry over to bounded cohomology: In general,
projective resolutions do not carry a norm (and so we cannot talk about
bounded cochains) and it is not clear that the canonical chain homotopies
would be bounded. In fact, there is a fundamental theorem for bounded co-
homology, based on a functional analytic notion of projectivity [36, 43].

Outlook 2.3.21 (higher bounded cohomology of free groups). Let F be a free


group of rank at least 2. Constructions from hyperbolic geometry in di-
mension 3 show that Hb3 (F ; R) ∼ 6 R 0 [79]. However, it is unknown whether
=
Hbn (F ; R) is non-trivial or not for n ∈ N≥4 (!).

2.3.4 Application: Stable commutator length


We will now move one step further into group theory and briefly discuss
stable commutator length.

Definition 2.3.22 ((stable) commutator length). Let G be a group and let


g ∈ [G, G].

• The commutator length of g is defined by



clG g := min n ∈ N ∃a1 ,...,an ,b1 ,...,bn ∈G [a1 , b1 ] · · · · · [an , bn ] = g ∈ N.

• The stable commutator length of g is defined by


1
sclG g := inf · clG (g n ).
n∈N>0 n
96 2. The geometric view

Remark 2.3.23 (lim vs. inf). Let G be a group and let g ∈ [G, G]. Then the
second part of Lemma 2.3.15 shows that
1
sclG g = lim · clG g n .
n→∞ n

Of course, the definition of (stable) commutator length only makes sense


because the commutator subgroup, in general, does not only consist of com-
mutators, but also of elements that can be written as products/inverses of
commutators.

Example 2.3.24 (short-cuts). Let G be a group and let a, b ∈ G. Then

[a, b]3 = [a · b · a−1 , b−1 · a · b · a−2 ] · [b−1 · a · b, b2 ]

(check! [18]). Therefore, clG [a, b]3 ≤ 2 and so


1 2
sclG [a, b] = lim · clG [a, b]n ≤ .
n→∞ n 3

Stable commutator length can be expressed in terms of quasi-morphisms


(and thus, at least implicitly, in terms of bounded cohomology):

Proposition 2.3.25 (commutator estimate for quasi-morphisms). Let G be a


group, let ϕ : G −→ R be a homogeneous quasi-morphism, and let a, b ∈ G.
Then 
ϕ [a, b] ≤ D(ϕ).

Proof. As first step, we show that ϕ(a · b · a−1 ) = ϕ(b): Because ϕ is homo-
geneous, we have

ϕ(a · b · a−1 ) − ϕ(b) = 1 · ϕ((a · b · a−1 )n ) − ϕ(bn )
n
1
= · ϕ(a · bn · a−1 ) − ϕ(bn )
n
1
≤ · ϕ(a · bn · a−1 ) − ϕ(a · bn ) − ϕ(a−1 ) + ϕ(a · bn ) + ϕ(a−1 ) − ϕ(bn )
n
1
= · ϕ(a · bn · a−1 ) − ϕ(a · bn ) − ϕ(a−1 ) + ϕ(a · bn ) − ϕ(a) − ϕ(bn )
n
1 1
≤ · ϕ(a · bn · a−1 ) − ϕ(a · bn ) − ϕ(a−1 ) + · ϕ(a · bn ) − ϕ(a) − ϕ(bn )
n n
2
≤ · D(ϕ)
n
for all n ∈ N. Taking n → ∞ shows that |ϕ(a · b · a−1 − ϕ(b)| = 0. Therefore,
2.3. Bounded cohomology 97

ϕ [a, b] = ϕ(a · b · a−1 · b−1 ) − ϕ(a · b · a−1 ) − ϕ(b−1 ) + ϕ(a · b · a−1 ) + ϕ(b−1 )

≤ D(ϕ) + ϕ(a · b · a−1 ) + ϕ(b−1 )

= D(ϕ) + ϕ(b) − ϕ(b)
= D(ϕ),

as claimed.

Theorem 2.3.26 (Bavard duality). Let G be a group and let g ∈ [G, G]. Then
(where sup ∅ = 0)

1 ϕ(g)
sclG g = · sup .
2 ϕ∈QM(G)\HomGroup (G,R) D(ϕ)

Proof. We only prove the easy estimate: Let ϕ ∈ QM(G). Then the commu-
tator estimate (Proposition 2.3.25) and a straightforward induction over the
commutator length (check!) show that

∀g∈[G,G] ϕ(g) ≤ 2 · D(ϕ) · clG g.

Therefore, for all g ∈ [G, G], we obtain (because ϕ is homogeneous)


1 1
∀n∈N>0 ϕ(g) = · ϕ(g n ) ≤ · 2 · D(ϕ) · clG g n .
n n

Taking the limit shows that ϕ(g) ≤ 2 · D(ϕ) · sclG g.
The proof of the converse estimate is more delicate and relies on topological
arguments (involving surfaces) [3, 14].

Example 2.3.27 (stable commutator length on amenable groups). If G is an


amenable group, then QM(G) = HomGroup (G, R) (Corollary 2.3.18). There-
fore, Bavard duality (Theorem 2.3.26) implies that

∀g∈[G,G] sclG g = 0.

This can be viewed as a far-reaching generalisation of the fact that the com-
mutator subgroup of an Abelian group is trivial.

Example 2.3.28 (stable commutator length on free groups). Let F := ha, b | i.


Then the homogenisation ϕ of the counting quasi-morphism associated with
the word aba−1 b−1 satisfies
 1  1
ϕ [a, b] = lim · ψaba−1 b−1 [a, b]n = lim · (n − 0) = 1.
n→∞ n n→∞ n

Moreover, ϕ is not a group homomorphism (we can use the same argument
as in the proof of Proposition 2.3.16). Therefore, Bavard duality shows that
98 2. The geometric view

1 1
sclF [a, b] ≥ · > 0.
2 D(ϕ)

Outlook 2.3.29 (values of scl). More precisely, we have in the free group F :=
ha, b | i the following results on stable commutator length [14]:
• More careful estimates show that sclF [a, b] = 1/2.

• It is known that sclF g ∈ Q≥1/2 for all g ∈ [F, F ] \ {e}. In particular,


sclF has a gap in (0, 1/2).
• There is a polynomial-time (in the length of the input word) algorithm
that computes stable commutator length on F .

• No element g ∈ [F, F ] with sclF g ∈ (1/2, 7/12) is known; so, there


might be a second scl-gap on F (this is an open problem).
In contrast, it is known that there exist finitely presented groups whose
stable commutator length function also assumes transcendental values [91].
Computations of stable commutator length in Thompson-like groups have
recently been used to manufacture closed manifolds with arbitrary (non-
negative) rational simplicial volume [41]; in particular, this shows that there
is no simplicial volume gap in dimension at least 4.
3
The derived view

Group homology and group cohomology are derived functors of tensor and
Hom-functors, respectively. We will briefly describe this aspect of group
(co)homology (leading to an axiomatic description of group homology) and
explain how computational tools from homological algebra can be used in
group (co)homology (e.g., spectral sequences).

Overview of this chapter.

3.1 Derived functors 100


3.2 The Hochschild-Serre spectral sequence 117

Running example. tensor and Hom-functors


100 3. The derived view

3.1 Derived functors


Many algebraic constructions only yield semi-exact functors (e.g., tensor
products and Hom-functors). Derived functors provide exact extensions of
such functors and therefore naturally come up in many computations.
For simplicity, in the following discussion, we will focus on derived functors
of right exact functors (and group homology); we will summarise the corre-
sponding results for left exact functors (and group cohomology) in Chap-
ter 3.1.5. Moreover, we will stay in the elementary language of homological
algebra of module categories; the refined approach via derived categories will
briefly be discussed in Chapter 3.1.6.

3.1.1 Axiomatic description


We first give an axiomatic description of derived functors. A construction
will be sketched in Chapter 3.1.2.
Definition 3.1.1 (module category). A module category is a category of the
form R Mod, where R is a ring. Similarly, a right module category is a category
of the form ModR , where R is a ring.
Study note. In order to avoid introducing the foundations of categories with
additional structures (such as additive, Abelian, exact, triangulated, . . . ) and
reproving standard homological facts about them, we will just work in the
elementary setup of module categories. If you are familiar with more abstract
setups of homological algebra, you should try to remember, which kind of
structures are needed in which step. If you are not familiar with any of these
setups, it might be a good idea to consult the literature [87, 13].
Definition 3.1.2 (right exact functor). Let F : C −→ D be a (covariant) addi-
tive functor, i.e., F is compatible with the additive structure on the Hom-sets
(given by pointwise addition). The functor F is right exact, if for every short
exact sequence 0 / A0 i / A π / A00 / 0 in C, the sequence

F (i) F (π)
F (A0 ) / F (A) / F (A00 ) /0

is exact in D.
F (i)
The functor F is exact if, in addition, also 0 / F (A0 ) / F (A) is
exact.
Example 3.1.3. Tensor product functors are (additive and) right exact, as
can be seen from the adjunction with the Hom-functor (Korollar IV.1.5.14).
3.1. Derived functors 101

Derived functors of right exact functors are a systematic “optimal” ex-


tension of the given functor turning short exact sequences into long exact
sequences (in a natural way). We formulate this universal property in terms
of homological ∂-functors:
Definition 3.1.4 (homological ∂-functor). Let C and D be module categories.
A homological ∂-functor C −→ D consists of the following data:
• a sequence (Tn : C −→ D)n∈N of additive functors,
• morphisms (∂n ∈ MorC (Tn (A00 ), Tn−1 (A0 )))n∈N , the connecting mor-
phisms, for each short exact sequence 0 / A0 i / A π / A00 /0
in C
such that for every commutative

/ A0 i π / A00
0 /A /0

f0 f f 00
  
0 / B0 /B / B 00 /0
j ϕ

diagram in C with exact rows, the corresponding diagram

∂n+1 Tn (i) Tn (π) ∂n


··· / Tn+1 (A00 ) / Tn (A0 ) / Tn (A) / Tn (A00 ) / ···

Tn+1 (f 00 ) Tn (f 0 ) Tn (f ) Tn (f 00 )
   
··· / Tn+1 (B 00 ) / Tn (B 0 ) / Tn (B) / Tn (B 00 ) / ···
∂n+1 Tn (j) Tn (ϕ) ∂n

in D is commutative and has exact rows. Here, we set T−1 := 0 and ∂0 := 0.


In order to unclutter notation, we usually just say that T∗ : C −→ D is a
homological ∂-functor (even though it consists of more data!) and the short
exact sequence is not included in the notation of the connecting morphisms.
Study note. What is a reasonable notion of natural transformations between
homological ∂-functors? When are two homological ∂-functors isomorphic?
(These notions also have to incorporate the connecting morphisms!)
Definition 3.1.5 (derived functor). Let C and D be module categories and
let F : C −→ D be a right-exact functor. A left-derived functor of F is a
homological ∂-functor L∗ : C −→ D
• together with a natural isomorphism L0 =⇒ F (i.e., L∗ extends F ),
• and the following universality property: If T∗ : C −→ D is a homological
∂-functor and τ0 : T0 =⇒ F is a natural transforation, then there is a
unique natural transformation τ∗ : T∗ =⇒ L∗ of homological ∂-functors
that extends τ0 .
102 3. The derived view

Remark 3.1.6 (uniqueness of derived functors). If F : C −→ D is a right-exact


functor between module categories, then derived functors of F (if they exist)
are unique up to canonical natural isomorphism of homological ∂-functors
(check!).
Example 3.1.7 (derived functors of exact functors). Let F : C −→ D be an
exact functor. Then

L0 := F
∀n∈N>0 Ln := 0
∀n∈N ∂n := 0

defines a (“the”) left-derived functor of F (check!).

3.1.2 A construction
In order to construct a left-derived functor of a right-exact functor, we pro-
ceed as follows:
• We replace objects by projective resolutions (i.e., by decomposition into
homologically simpler objects),
• apply the functor in question to these projective resolutions,
• and measure the failure of exactness via homology.
Theorem 3.1.8 (existence of derived functors). Let F : C −→ D be a right-
exact functor between module categories. Then there exists a left-derived func-
tor of F . (Moreover, by Remark 3.1.6, we have essential uniqueness.)
Proof. As first step, we construct the functors (Ln : C −→ D)n∈N : For
each A ∈ Ob(C), we choose a projective resolution (P∗A , εA ) of A in C (as
module category, C contains enough projectives (Proposition IV.5.2.10). Let
n ∈ N; we then define Ln : C −→ D as follows:
• On objects: For A ∈ Ob(C), we set

Ln (A) := Hn F (P∗A ) ∈ Ob(D).

• On morphisms: Let A, B ∈ Ob(C) and let f ∈ MorC (A, B). By the


fundamental theorem (Theorem 1.6.7), there exists a lift f∗  f : P∗A 
εA −→ P∗B  εB of f (which is unique up to chain homotopy), and we
set  
Ln (f ) := Hn F (f∗ ) ∈ MorD Ln (A), Ln (B) .
Because f∗ is uniquely determined up to chain homotopy and because
homology is a chain homotopy invariant, Ln (f ) does not depend on the
chosen extension f∗ .
3.1. Derived functors 103

This construction has the following properties:


À Each Ln is a functor C −→ D:
Clearly, Ln maps identity morphisms to identity morphisms (we can
choose the identity as extension). Moreover, the fundamental theorem
(Theorem 1.6.7) shows that Ln is compatible with compositions (for the
composition, we may choose compositions of extensions as extension).
Á If P ∈ Ob(C) is projective and n ∈ N>0 , then Ln (P ) ∼
=D 0:
If (Q∗ , ε) is any projective resolution of P , then the fundamental theo-
rem (Theorem 1.6.7) shows that there is a canonical isomorphism
 
Ln (P ) = Hn F (P∗P ) ∼=D Hn F (Q∗ ) .

On the other hand, as P is projective, we can choose the trivial projec-


tive resolution
idP
··· /0 /0 /P /P /0

of P . Therefore, if n ∈ N>0 , we have Ln (P ) ∼


=D 0.
 There is a canonical natural isomorphism L0 =⇒ F :
Let A, B ∈ Ob(C), let f ∈ MorC (A, B), and let f∗  f : P∗A  εA −→
P∗B  εB be an extension of f . Then, in particular, we have the following
commutative diagram in C with exact rows:

∂1A εA
P1A / P0A /A /0

f1 f0 f
  
P1B / P0B /B /0
∂1B εB

Applying F and homology leads to the following commutative diagram


in D (as F is right-exact, we know that im F (∂1A ) = ker F (εA ) and
im F (εA ) = A, and similarly for B):

“F (εA )”
L0 (A) H0 (F (P∗A )) F (P0A )/ im F (∂1A ) F (P0A )/ ker F (εA ) ∼ / F (A)
=D

L0 (f ) “F (f1 )” “F (f0 )” “F (f0 )” F (f )


    ∼ 
=D
L0 (B) H0 (F (P∗B )) F (P0B )/ im F (∂1B ) F (P0B )/ ker F (εB ) / F (B)
“F (εB )”

Hence, L0 ∼
= F.
As next step, we construct the connecting morphisms, using the horseshoe
lemma and the long exact sequence in homology (Appendix A.2):
104 3. The derived view

• Let 0 / A0 i / A π / A00 / 0 be a short exact sequence


in A. By the horseshoe lemma (Proposition A.2.3), there exists a pro-
jective resolution (Q∗ , ε) of A and chain maps i∗  i as well as π∗  π
such that
0 / P∗A0 i∗ / Q∗ π∗ / P∗A00 /0

is a degree-wise short exact sequence of chain complexes over C. Be-


00
cause P∗A consists of projective modules, in each degree, we have a split
exact sequence. Because F is additive, we hence obtain a degree-wise
short exact sequence

F (i∗ ) F (π∗ )
0 / F (P∗A0 ) / F (Q∗ ) / F (P∗A00 ) /0

of chain complexes over D. Applying homology (and independence of L∗


of the chosen projective resolutions), we therefore get a long exact ho-
mology sequence (Proposition A.2.2)

∂n+1 Ln (i) Ln (π) ∂n


··· / Ln+1 (A00 ) / Ln (A0 ) / Ln (A) / Ln (A00 ) / ···

We then define the connecting morphisms for the given short exact
sequence as the connecting morphisms in this long exact sequence.

What about naturality? The extended horseshoe lemma [87, Theo-


rem 2.4.6] allows to construct compatible projective resolutions for ev-
ery ladder of short exact sequences. Because the long exact homology
sequence is also natural, this naturality carries over to the connecting
morphisms (check!).

Hence, L∗ , together with these connecting morphisms, defines a homological


∂-functor C −→ D that extends F .
Therefore, it remains to show universality: Let T∗ : C −→ D be a ho-
mological ∂-functor and let τ0 : T0 =⇒ F be a natural transformation; be-
cause L0 ∼= F , we can instead also assume that τ0 is a natural transforma-
tion T0 =⇒ L0 (to simplify notation). We inductively extend τ0 to a natural
transformation of homological ∂-functors from T∗ to L∗ by dimension shifting
(Corollary 3.1.9 below):
Let n ∈ N and let us suppose that τ∗ is already constructed up to degree n.
We now construct τn+1 : Let A ∈ Ob(C). Then there exists a short exact
sequence
0 /K i /P π /A /0

in C with a projective P (because C has enough projectives). Because T∗ and


L∗ are homological ∂-functors and because Ln+1 (P ) ∼
=D 0 (by Á), we obtain
the following commutative diagram (the solid arrows)
3.1. Derived functors 105

T
∂n+1 Tn (i)
Tn+1 (A) / Tn (K) / Tn (P )

τn+1 (A) τn (K) τn (P )


  
0∼
=D Ln+1 (P ) / Ln+1 (A)
L
/ Ln (K) / Ln (P )
∂n+1 Ln (i)

in D with exact rows. A simple diagram chase shows that there is a unique
morphism τn+1 (A) in D that makes the left-hand square commutative
(check!).
One now has to check that τn+1 (A) indeed is compatible with the con-
necting morphisms of all short exact sequences with A as quotient and that
τn+1 is compatible with homomorphisms of modules:

à The homomoprhism τn+1 (A) is compatible with the connecting mor-


phisms ∂n+1 :

Let 0 / B0 j / B ϕ / A / 0 be a short exact sequence


in C that ends in the given A. As P is projective, a diagram chase
reveals that there exist morphisms f and f 0 in C that fit into the com-
mutative diagram

i π
0 /K /P /A /0

f0 f
 
0 / B0 /B /A /0
j ϕ

in C (check!). We now consider the diagram in Figure 3.1. The large


outer square commutes by construction of τn+1 (A). The small outer four
quadrangles commute because T∗ and L∗ are homological ∂-functors
and because τn is natural (by induction). Therefore, also the inner
square is commutative, as claimed.

Ä The homomorphisms τn+1 are compatible with morphisms:

Let f ∈ MorC (A, B) and let

i π
0 /K /P /A /0
0 /M /Q
ϕ
/B /0
j

short exact sequences in C with projectives P and Q. Similarly, as


above, we can find morphisms p and f 0 fitting into the commutative
diagram
106 3. The derived view

T
∂n+1
Tn+1 (A) / Tn (K)
Tn (f 0 )
T y
∂n+1
Tn+1 (A) / Tn (B 0 )
τn+1 (A) τn+1 (A) τn (B 0 ) τn (K)
 
Ln+1 (A) L / Ln (B 0 )
∂n+1 e
 Ln (f 0 ) 
Ln+1 (A) L
/ Ln (K)
∂n+1

Figure 3.1.: Compatibility of τn+1 (A) with ∂n+1

i π
0 /K /P /A /0

f0 p f
  
0 /M /Q
ϕ
/B /0
j

in C (check!). Considering the diagram in Figure 3.2 and arguing in-


ductively as above, we find that
L L
∂n+1 ◦ τn+1 (B) ◦ Tn+1 (f ) = ∂n+1 ◦ Ln+1 (f ) ◦ τn+1 (A).

Because Ln+1 (Q) ∼ L


=D 0, the morphism ∂n+1 : Ln+1 (B) −→ Ln (M ) is
monic, and thus we obtain

τn+1 (B) ◦ Tn+1 (f ) = Ln+1 (f ) ◦ τn+1 (A),

as desired.
This completes the proof that L∗ is a left-derived functor of F .
During the proof, we also used/established the following property (that
allows to express derived functors in higher degree by lower degrees, for the
price of changing the objects):
Corollary 3.1.9 (dimension shifting). Let F : C −→ D be a right-exact functor
between module categories and let L∗ be a/“the” left-derived functor of F .
Let A ∈ Ob(C) and let
i π
0 /K /P /A /0

be a short exact sequence in C with a projective P . Then, for all n ∈ N>0 ,


the connecting morphism of L∗ induces an isomorphism

Ln+1 (A) ∼
=D Ln (K).
3.1. Derived functors 107

Tn (f 0 )
Tn (K) / Tn (M )
f T T 8
∂n+1 ∂n+1

Tn+1 (f )
Tn+1 (A) / Tn+1 (B)
τn (K) τn+1 (A) τn+1 (B) τn (M )
 
Ln+1 (A) / Ln+1 (B)
Ln+1 (f )

L L
 x ∂n+1 ∂n+1 & 
Ln (K) / Ln (M )
Ln (f 0 )

Figure 3.2.: Naturality of τn+1

Moreover, 
L1 (A) ∼
=D ker L0 (i) : L0 (K) → L0 (A) .
Proof. By observation Á of the proof of Theorem 3.1.8, we know that Ln is
trivial on projectives, provided that n ∈ N>0 . The claim then follows from
the long exact sequence obtained from the given short exact sequence by
application of L∗ .

3.1.3 The two sides of Tor


We now apply the previous discussion to the tensor product:

Example 3.1.10 (Tor). derived functor Let R be a ring and let M ∈


Ob(ModR ). Then “the” derived functor of M ⊗R · : R Mod −→ Z Mod is
denoted by TorR∗ (M, · ). More explicitly: If A ∈ Ob(R Mod) and (P∗ , ε) is a
projective R-resolution of A, then, for all n ∈ N,

TorR ∼
n (M, A) =Z Hn (M ⊗R P∗ ).

Moreover, Tor can also be computed by resolving the first argument:

Theorem 3.1.11 (“symmetry” of Tor). Let R be a ring and let M ∈ Ob(ModR ).


If (P∗ , ε) is a projective R-resolution of M (by right R-modules), then, for
all n ∈ N, there is a natural isomorphism

TorR ∼
n (M, · ) =Z Hn (P∗ ⊗R · ).

Proof. We prove this via the universal property of derived functors; therefore,
we first introduce an appropriate homological ∂-functor:

• For n ∈ N, we set Ln := Hn (P∗ ⊗R · ) : R Mod −→ Z Mod .


108 3. The derived view

• If 0 / A0 i / A π / A00 / 0 is a short exact sequence of


left R-modules, then also the sequence

P∗ ⊗R i
0 / P∗ ⊗ R A 0 / P∗ ⊗R A P∗ ⊗R π/ P∗ ⊗R A00 /0

in Z Ch is degree-wise short exact (because all Pn are projective, whence


flat over R). Hence, we obtain an associated natural long exact homol-
ogy sequence
∂n+1 Ln (i) Ln (π) ∂n
··· / Ln+1 (A00 ) / Ln (A0 ) / Ln (A) / Ln (A00 ) / ···

We take these connecting morphisms as connecting morphisms for our


homological ∂-functor.
We now verify that L∗ with these connecting morphisms is a left-derived
functor of M ⊗R · :
• Extension of M ⊗R · : Let A ∈ Ob(R Mod). Then the right-exactness
of · ⊗R A shows (as in the proof of Theorem 3.1.8) that

L0 (A) = Hn (P∗ ⊗R A) = (P0 ⊗R A) im(∂1P∗ ⊗R A)

= (P0 ⊗R A) ker(ε ⊗R A) = ∼Z (P0 / ker ε) ⊗R A

=Z M ⊗R A.

Clearly, this isomorphism is also natural.


• Universality: We can use the same argument as in the proof of Theo-
rem 3.1.8; indeed, the argument in that proof only used the properties
of homological ∂-functors and the vanishing on projectives in higher
degree. Therefore, it suffices to establish this vanishing on projectives:
If Q is a projective left R-module, then (P∗ ⊗R Q, ε ⊗R Q) is exact
(because Q is flat over R). In particular, for all n ∈ N>0 ,

Ln (Q) = Hn (P∗ ⊗R Q) ∼
=Z 0.

Hence, L∗ is a left-derived functor of M ⊗R · . Therefore, the uniqueness of


left-derived functors yields the desired isomorphism.

3.1.4 Group homology as derived functor


We can interpret group homology as a Tor-functor:
Theorem 3.1.12 (group homology as Tor-functor). Let G be a group and let
n ∈ N. Then there is a canonical natural isomorphism

=Z TorZG
Hn (G; · ) ∼ n Z, ·
3.1. Derived functors 109

of functors ZG Mod −→ Z Mod. In particular, for every left ZG-module A, we


have:
1. If (P∗ , ε) is a projective ZG-resolution of A, then there is a canonical
isomorphism
Hn (G; A) ∼
=Z Hn (Z ⊗G P∗ ).

2. If (P∗ , ε) is a projective ZG-resolution of Z, then there is a canonical


isomorphism
Hn (G; A) ∼
=Z Hn (P∗ ⊗G A).

Proof. We start with the last claim; in fact, we already know this – this is
nothing but the fundamental theorem of group homology (Corollary 1.6.9).
Then (Inv P∗ , Inv ε) is a projective resolution of Z by right ZG-modules.
Hence, Theorem 3.1.11 shows that

Hn (G; A) ∼=Z Hn (P∗ ⊗G A) = Hn (Inv P∗ ) ⊗ZG A ∼ =Z TorZG
n (Z, A).

Therefore, Example 3.1.10 gives the first description of Hn (G; A).


Hence, one can characterise group homology concisely as “derived functor
of co-invariants”. In particular, one could use this description in terms of
Tor-functors also to give an axiomatic characterisation of group homology.
Corollary 3.1.13 (derived properties of group homology). Let G be a group.
1. Vanishing on projectives. If A is a projective (or flat) ZG-module and
n ∈ N>0 , then
Hn (G; A) ∼
=Z 0.

2. Long exact sequence. If 0 / A0 i / A π / A00 / 0 is a


short exact sequence of ZG-modules, then there is an associated (natu-
ral) long exact sequence

∂n+1
··· / Hn (G; A0 )Hn (idG ;i)
/ Hn (G; A)Hn (idG ;π)
/ Hn (G; A00 ) ∂n
/ Hn−1 (G; A0 ) / ···

3. Dimension shifting. If 0 /K i /P π /A / 0 is a short


exact sequence of ZG-modules and P is projective (or flat), then, for
each n ∈ N>0 , the connecting homomorphism induces an isomorphism

Hn+1 (G; A) ∼
=Z Hn (G; K).

Moreover, H1 (G; A) ∼
=Z ker H0 (idG ; i) : H0 (G; K) → H0 (G; A) .
=Z TorZG
Proof. By Theorem 3.1.12, H∗ (G; · ) ∼ ∗ (Z, · ). Therefore, the listed
properties follow from the corresponding properties of the derived func-
tor TorZG
∗ (Z, · ) of Z ⊗ZG · (Definition 3.1.5, proof of Theorem 3.1.8,
Corollary 3.1.9).
110 3. The derived view

Example 3.1.14. Let G be a finite group. We consider the Abelian group S 1


as ZG-module with the trivial G-action. Then S 1 ∼
=Group R/Z and we use the
short exact sequence

0 /Z /R / S1 /0

of (trivial) ZG-modules (given by the canonical inclusion and projection).


Because G is finite, we know that

Hn (G; R) ∼
=Z 0

for all n ∈ N>0 (Corollary 1.7.16). Hence, the long exact sequence in group
homology (Corollary 3.1.13) associated with the short exact coefficients se-
quence above shows that

Hn (G; S 1 ) ∼
=Z Hn−1 (G; Z)

for all n ∈ N>1 . In particular, H2 (G; S 1 ) ∼


=Z H1 (G; Z) ∼
=Z Gab .

3.1.5 Group cohomology as derived functor


Dually to homological ∂-functors, one can also introduce cohomological δ-
functors and define right-derived functors of left-exact functors via injective
resolutions.
Example 3.1.15 (Ext). Let R be a ring and let M ∈ Ob(R Mod). Then
“the” right-derived functor of R Hom(M, · ) : R Mod −→ Z Mod is denoted
by Ext∗R (M, · ). More explicitly: If A ∈ Ob(R Mod) and (I ∗ , ε) is an injective
R-resolution of A, then, for all n ∈ N,

ExtnR (M, A) ∼ =Z H n R Hom(M, I ∗ ) .

Similarly, to the case of Tor, one can also compute Ext by resolving the first
variable [87, Chapter 2.7]: If (P∗ , ε) is a projective R-resolution of M , then,
for all n ∈ N, 
ExtnR (M, A) ∼=Z H n R Hom(P∗ , A) .
As in the case of group homology, also cohomology can be described in
terms of derived functors (check!):
Theorem 3.1.16 (group cohomology as Ext-functor). Let G be a group and let
n ∈ N. Then there is a canonical natural isomorphism

H n (G; · ) ∼
=Z ExtnZG (Z, · )

of functors ZG Mod −→ Z Mod. In particular, for every left ZG-module A, we


have:
3.1. Derived functors 111

1. If (I ∗ , ε) is an injective ZG-resolution of A, then there is a canonical


isomorphism 
H n (G; A) ∼
=Z H n ZG Hom(Z, I ∗ ) .

2. If (P∗ , ε) is a projective ZG-resolution of Z, then there is a canonical


isomorphism 
H n (G; A) ∼
=Z H n ZG Hom(P∗ , A) .

Thus, group cohomology can be characterised as “derived functor of in-


variants”, which also leads to an axiomatic description of group cohomology.
Also, the analogue of Corollary 3.1.13 holds. As mentioned before, this rela-
tion between group cohomology and Ext also explains the name Ext: Second
group cohomology is related to extensions (Theorem 1.5.10).
Remark 3.1.17 (duality). The properties of group cohomology are dual to
the ones of group homology. One can establish these properties by imitating
the homological proof (and making the obvious modifications). Alternatively,
one can also pass to the more general setup of homological algebra in Abelian
categories and then apply the previous theory to the opposite category.

3.1.6 The derived category


Our definition of derived functors via homological ∂-functors and the con-
struction via resolutions has the advantage of being rather concrete, but also
the disadvantage of requiring tedious bookkeeping and working with unruly
objects. These issues become even more apparent when one starts working
with several derived functors at once or when one wants to transfer the theory
of derived functors to other settings.
A remedy is the derived category and the notion of total derived functors.
In the following, we will roughly outline this version of homological algebra
(for module categories); further details can be found in the literature [87,
46, 69]. The key idea is to postpone/avoid taking (co)homology as long as
possible.
Let us recall the computation of the value of the derived functor of a
right-exact functor F : C −→ D between module categories on A ∈ Ob(C):
• We choose a projective resolution (P∗ , ε) of A.
• We apply F to P∗ .
• We apply homology.
According to the key idea, we should avoid taking homology in the last step,
but work with F (P∗ ) instead, which leads to the following issues:
• The result F (P∗ ) will depend more on the choice of P∗ than H∗ (F (P∗ ));
it is not well-defined up to isomorphism, but only well-defined up to
canonical chain homotopy equivalence.
112 3. The derived view

• We might strive for a uniform setup for both objects in C as well as res-
olutions of objects in C. We can view A as chain complex concentrated
in degree 0 and ε : P∗ −→ A as a chain map; however, ε, in general,
will not be a chain homotopy equivalence, but only a chain map that
induces an isomorphism on the level of homology.

Therefore, the derived category of C will be defined as the category con-


structed out of the category of chain complexes over C by turning all homol-
ogy isomorphisms into isomorphisms. More precisely, one proceeds as follows:

Definition 3.1.18 (quasi-isomorphism). Let R be a ring and let C∗ and D∗ be


chain complexes of left R-modules. A quasi-isomorphism C∗ −→ D∗ is an R-
chain map f∗ : C∗ −→ D∗ such that for all n ∈ N, the induced homomorphism

Hn (f∗ ) : Hn (C∗ ) −→ Hn (D∗ ).

is an R-isomorphism.

Caveat 3.1.19 (quasi-isomorphisms need not be invertible). There exist chain


complexes C∗ and D∗ over the ring Z such that there exists a quasi-
isomorphism C∗ −→ D∗ but no quasi-isomorphism D∗ −→ C∗ (Exercise).

Example 3.1.20 (projective resolutions). Let R be a ring, let A be an R-


module and let (P∗ , ε) be a projective resolution of A over R. If we view A as
an R-chain complex concentrated in degree 0, then ε (in degree 0; together
with the 0-morphisms in all other degrees) is a quasi-isomorphism P∗ −→ A
(check!):
.. ..
. .

 
P1 0
∂1 0
 
P0 ε
/A

For chain complexes of projective modules, quasi-isomorphisms are tame:

Theorem 3.1.21 (quasi-isomorphisms between projective chain complexes). Let


R be a ring, let C∗ and D∗ be (N-indexed) chain complexes over R that
consist of projective modules. Every quasi-isomorphism C∗ −→ D∗ is a chain
homotopy equivalence.

Proof. Using mapping cones (Definition 3.1.22, Proposition 3.1.23), one only
needs to consider the case that one of the two complexes is trivial (Exercise).
In this case, the theorem can be deduced from the fundamental theorem of
homological algebra (Theorem 1.6.7) (Exercise).
3.1. Derived functors 113

Definition 3.1.22 (mapping cone). Let R be a ring and let f∗ : C∗ −→ D∗ be


a chain map of R-chain complexes. The mapping cone of f∗ is the R-chain
complex Cone∗ (f∗ ) consisting of the chain modules

Conen (f∗ ) := Cn−1 ⊕ Dn

for all n ∈ N (where C−1 := 0), equipped with the boundary operators

∂n : Conen (f∗ ) −→ Conen−1 (f∗ )


C

(x, y) 7−→ −∂n−1 (x), ∂nD (y) − fn−1 (x)

for all n ∈ N>0 .

Proposition 3.1.23 (mapping cone trick). Let R be a ring, let f∗ : C∗ −→ D∗


be a chain map of R-chain complexes. The the following are equivalent:

1. The chain map f∗ : C∗ −→ D∗ is a quasi-isomorphism.



2. For all n ∈ N, we have Hn Cone(f∗ ) ∼=R 0.

Proof. This can be extracted from a suitable long exact homology sequence
(Exercise).

Study note (mapping cones). For the boundary operator on Cone(f∗ ), several
different sign conventions are in use. Therefore, literature has to be used with
care! Of course, the mapping cone of chain maps is an algebraic imitation of
the topological mapping cone; however, the algebraic version has slightly
“better” properties (Exercise).

The derived category is now defined as the localisation of the category of


(bounded below) chain complexes at the class of all quasi-isomorphisms.
We first formalise this localisation via a universal property (similar to the
universal property of localisations of (non-commutative) rings); we will then
briefly explain why the localisation at the class of quasi-isomorphisms does
exist.

Definition 3.1.24 (localisation). Let C be a category and let S be a class


of morphisms of C. A localisation of C at S is a functor q : C −→ D to a
category D with the following universal property:

• For each f ∈ S, the morphism q(f ) is an isomorphism in D.

• If E is a category and F : C −→ E is a functor with the property


that F (s) is an isomorphism for all s ∈ S, then there exists a unique
functor Fe : D −→ E that satisfies

Fe ◦ q = F.
114 3. The derived view

F
C /E
>
q
e
F

D

(If D and q exist, then they are unique up to natural equivalence, and the
category D is usually denoted by C[S −1 ] or S −1 C.)

Example 3.1.25 (the homotopy category as localisation). Let R be a ring. Then


the homotopy category of R Ch is the category R Chh defined as follows:

• objects: We set Ob(R Chh ) := Ob(R Ch).

• morphisms: For all C∗ , D∗ ∈ Ob(R Ch), we set



MorR Chh (C∗ , D∗ ) := MorR Ch (C∗ , D∗ ) 'R .

• compositions: Induced by ordinary composition of representatives.

Together with the canonical functor R Ch −→ R Chh , the homotopy category


is a localisation of R Ch at the class of all R-chain homotopy equivalences;
this can be proved using mapping cylinders [87, Proposition 10.1.2].

In general, localisation categories do not exist and (if they exist) might be
hard to compute (for various reasons).

Theorem 3.1.26 (the derived category). Let R be a ring. Then there exists a
localisation D(R) of R Ch at the class of all quasi-isomorphisms, the derived
category of R Mod. More concretely, D(R) can be constructed as a localisa-
tion of R Chh at the class of all quasi-isomorphisms, which in turn can be
constructed by the following calculus of fractions:

• objects: Let Ob(D(R)) := Ob(R Chh ) = Ob(R Ch).

• morphisms X −→ Y : Equivalence classes of fractions, i.e., of mor-


s
phisms X o X0 / Y in R Chh with a quasi-isomorphism s. Two
s / Y and X o t X 00
such fractions X o X0 / Y are equivalent
u
if there exists a fraction X o X 000 / Y that fits into a commutative
diagram in R Chh of the form

XO 0
s
} u !
X ao X 000 /Y
=
t 
X 00
3.1. Derived functors 115

• compositions: If F : X −→ Y and G : Y −→ Z are morphisms rep-


f
resented by fractions X o
s
X0 / Y and Y o t Y 0 g / Z , respec-
tively, then the composition g ◦ f is represented by the fraction

s◦s 0 g◦f 0
Xo X 00 /Z

where s0 : X 00 −→ X 0 is a quasi-isomorphism and f 0 : X 00 −→ Y 0 is a


chain map that fit into the commutative square (in R Chh )

f0
X 00 / Y0

s0 t
 
X0 /Y
f

The localisation functor qR : R Ch −→ D(R) is the identity on objects and


turns morphisms f : X −→ Y into the corresponding fraction

idX [f ]'
Xo X /Y .

Sketch of proof. The main technical problem is the following:

When naively constructing D(R) by generators and relations (which


enforce the existence of inverses for every quasi-isomorphism) it is not
clear a priori that the morphisms between two objects will form a set
(and not only a class).

This problem can be resolved by a careful examination of the class of quasi-


isomorphisms in R Chh and a crude cardinality argument [87, Chapter 10.4].
Moreover, it should be noted that we need to work in the homotopy cat-
egory R Chh for the calculus of fractions to work (e.g., for the composition,
which involves a category-theoretic version of the Ore condition); in R Ch, the
situation would be more involved.
Then, the proof is mainly a matter of computation [87, Chapter 10.3],
analogously to the construction of localisations of non-commutative rings.

In general, concrete computations in derived categories are hard. However,


in our simple case (of derived categories of module categories), one can just
compute in the homotopy category of (bounded below) chain complexes of
projective modules [87, Theorem 10.4.8]. This again underlines the impor-
tance of projectivity.
We will now come to derived functors: The derived category, in general, will
not be an Abelian category; therefore, more general terminology is needed
to formulate exactness properties, e.g., the language of triangulated cate-
gories [87, Chapter 10.2] (we will ignore this here and we will also ignore
116 3. The derived view

the technicality that we should work with bounded below chain complexes
instead of N-indexed chain complexes).
If F : R Mod −→ S Mod is an exact functor, then it is not difficult to see
that there is a corresponding functor D(F ) : D(R) −→ D(S) (of triangulated
categories) that satisfies

D(F ) ◦ qR = qS ◦ F,

where qR : R Ch −→ D(R), qS : S Ch −→ D(S) are the localisation functors.


However, if F is not exact, this will not be possible in general. For right-
exact functors, total left-derived functors are “optimal” extensions to the
derived categories in the following sense (where the compatibility with the
triangulated structure encodes the exactness properties):

Definition 3.1.27 (total derived functor). Let R and S be rings and let
F : R Mod −→ S Mod be a right-exact functor. Then a total left-derived functor
of F is a functor L F : D(R) −→ D(S) (of triangulated categories) together
with a natural transformation τ : (L F ) ◦ qR =⇒ qS ◦ F with the following
universal property:
If G : D(R) −→ D(S) is a functor (of triangulated categories) and if σ : G◦
qR =⇒ qS ◦ F is a natural transformation, then there exists a unique natural
transformation σe : G =⇒ L F such that

∀A∈Ob(R Ch) τ (A) ◦ σ
e qR (A) = σ(A).

LF G G
D(R) / D(S) D(R) / D(S) D(R) / D(S)
O O O O O O
qR
τ
qS qR
σ
qS qR e
σ ' F
' '
R Ch
/ S Ch R Ch
/ S Ch R Ch
/ D(R)
F F qR

Here, qR : R Ch −→ D(R) and qS : S Ch −→ D(S) denote the localisation


functors.

Study note. Compare this definition with the definition of ordinary left-
derived functors! Where did the data/properties shift to in the totally de-
rived setting? Moreover, it might be interesting to look up the notion of a
Kan extension and to compare it to the definition of total derived functors.

Theorem 3.1.28 (derived functor via total derived functor [87, Corollary 10.5.7,
Remark 10.5.8]). Let R and S be rings and let F : R Mod −→ S Mod be a right-
exact functor. Then a/“the” total derived functor L F : D(R) −→ D(S) exists
and we have a canonical natural isomorphism (as functors R Mod −→ S Mod)

L∗ (F ) ∼
=S H∗ L F (qr ( · )) .

(One should note that taking homology H∗ is well-defined on D(S).)


3.2. The Hochschild-Serre spectral sequence 117

What is the benefit of the total derived functor?


• Composition and combination formulas for several derived functors are
simpler on the total derived level than on the level of homology mod-
ules [87, Chapter 10.8].
For example, it is easier to compare L(G) ◦ L(F ) with L(G ◦ F ) on the
derived categories than to compare L∗ (G) ◦ L∗ (F ) (whatever that even
means) with L∗ (G ◦ F ) (because in the latter case we would have to
unravel the intermediate homology first; Theorem 3.2.11).
• The formalism of derived categories and total derived functors covers
a wide range of situations (also in algebraic/arithmetic geometry and
homotopy theory).
In order to enjoy the full power of derived categories and derived functors
one has to pass to a more general setup (e.g., triangulated categories or
model categories) [87, Chapter 10.9][69]. Then, in particular, also the close
connection with homotopy theory of topological spaces becomes visible:
Remark 3.1.29 (a derived category in topology). We have the following rough
dictionary between homological algebra and homotopy theory of spaces:
algebra topology
module topological space
chain complex spectrum
chain homotopy homotopy of maps of spectra
quasi-isomorphism weak equivalence
projective resolution CW-spectrum
The construction of the topological derived category (as localisation at
weak equivalences) can then be performed in analogy with the algebraic
case [87, Chapter 10.9].

3.2 The Hochschild-Serre spectral sequence


Bookkeeping in homological algebra can quickly get overwhelming when more
than a single complex is involved, for example, when taking multi-step filtra-
tions of chain complexes or double complexes. A convenient computational
tool is provided by spectral sequences. We will explain the setup of spectral
sequences and how to perform computations with them. However, we will
refrain from proving any convergence results.
Why do we need any of this in group (co)homology? For example, when
trying to compute the (co)homology of an extension group in terms of the
(co)homology of the kernel and the quotient, our techniques so far are not
sufficient to give a satisfactory answer. This problem will be solved by the
Hochschild-Serre spectral sequence (Chapter 3.2.3).
118 3. The derived view

q q q

d3
1 d2
d
p p p
1 2 3
E E E

Figure 3.3.: pages of a homological spectral sequence

3.2.1 Terminology for spectral sequences


We will now explain the principle of spectral sequences. In the beginning,
spectral sequences and all the notation might seem frightening and very tech-
nical; however, with a little bit of practice one will sooner or later appreciate
their power and the challenge of tricking spectral sequences into revealing
their secrets.
We restrict our discussion to spectral sequences in the first quadrant. Ex-
tensive treatments of spectral sequences (including proofs) can for instance
be found in the books by Weibel [87], Hatcher [40], McCleary [58], and in the
lecture notes of Bauer [2].
A spectral sequence is a sequence of bigraded modules, where the next
bigraded module is obtained from the previous one by taking homology (see
also Figure 3.3 for an illustration):

Definition 3.2.1 (homological spectral sequence). A (bigraded, homological)


spectral sequence over a ring R is a sequence (E r , dr )r∈N>0 (or r ∈ N>1 ) of
bigraded R-modules (i.e., every E r is family (Epq
r
)p,q∈N of R-modules) and
r r r
R-homomorphisms d : E −→ E with the following properties:

• For every r ∈ N>0 the map dr has degree (−r, r − 1), and dr ◦ dr = 0.

• For every r ∈ N>0 there is an isomorphism


ker dr
E r+1 ∼
=R H∗ (E r , dr ) = ,
im dr
and this isomorphism is also part of the data of the spectral sequence.

The term E r is also called the r-th page of (E ∗ , d∗ ) and the isomor-
phism E r+1 ∼
=R H∗ (E r , dr ) is the r-th page-turning isomorphism.
3.2. The Hochschild-Serre spectral sequence 119

q
F0 An
n
Fp An /Fp−1 An
n−p

p n p

E

Figure 3.4.: convergence of a homological spectral sequence, schematically

Definition 3.2.2 (∞-page of a spectral sequence). Let (E r , dr )r∈N>0 be a ho-


mological spectral sequence over a ring R. Because all the (E r , dr ) reside in
the first quadrant, for every p, q ∈ N there exists an s ∈ N>0 such that the
differentials starting and ending at (p, q) are trivial in all later pages, i.e.,
such that the page-turning isomorphisms induce isomorphisms
s ∼ s+1 ∼ s+2 ∼
Epq =R Epq =R Epq =R · · · ;
∞ s
we then define Epq := Epq (for the minimal such s).

Definition 3.2.3 (degeneration of a spectral sequence). Let (E r , dr )r∈N>0 be a


homological spectral sequence over a ring R and let s ∈ N>0 . We say that this
spectral sequence degenerates at stage s if, for all r ∈ N≥s , we have dr = 0.
In particular, the page-turning isomorphisms induce isomorphisms

Es ∼ =R E s+2 ∼
=R E s+1 ∼ =R . . . and E s ∼
=R E ∞ .

Until now, nothing happened – we merely introduced some notation. The


next definition is crucial for the applications of spectral sequences; it relates
a spectral sequence to something we want to compute:

Definition 3.2.4 (convergence of a spectral sequence). Let R be a ring, let


A be an N-graded R-module, and let (Fn A)n∈N be an increasing filtration
of A that is compatible with the grading of A. We say that a spectral se-
quence (E r , dr )r over R converges to A if the following conditions are satisfied
(see also Figure 3.4):

• For all p, q ∈ N we have (with F−1 A := 0)

∞ ∼ Fp Ap+q
Epq =R .
Fp−1 Ap+q
120 3. The derived view

F1 An
0 −→ F0 An −→ F1 An −→ −→ 0
F0 An
F2 An
0 −→ F1 An −→ F2 An −→ −→ 0
F1 An
..
.
Fn An
0 −→ Fn−1 An −→ Fn An = An −→ −→ 0
Fn−1 An

Figure 3.5.: convergence of a homological spectral sequence, extensions

• The spectral sequence is exhaustive, i.e., Fn An = An for all n ∈ N.

In this case one writes


2 1
Epq =⇒ Ap+q or Epq =⇒ Ap+q .

Remark 3.2.5 (stepping through a spectral sequence). What is the typical “us-
age” of a spectral sequence? We might be interested in some graded object A
(in most cases: homology of something) for which there happens to exist a
(homological) spectral sequence (E r , dr )r∈N>1 converging to A, where the
E 2 -term is something accessible:
2
Epq =⇒ Ap+q .

Usually, one then proceeds as follows:

1. Try to compute as many of the modules of the E 2 -term as possible; in


general, the more zeroes, the better!

2. Try to prove that many of the differentials d2pq in the E 2 -term are zero,
e.g., using the degree, torsion phenomena, product structures, . . .

3. Using the results of the first two steps, try to compute as much of the
E 3 -term and the differential d3 as possible.
Fortunately, many spectral sequences degenerate (at least to a large
extent) at the E 2 -stage or the E 3 -stage!

4. Carry on like that and try to compute as much of the E ∞ -term as


possible.

5. Try to solve the extension problems arising when reconstructing A out


of E ∞ .
3.2. The Hochschild-Serre spectral sequence 121

Caveat 3.2.6. If a spectral sequence (E r , dr )r∈N>0 converges to a graded fil-


tred module A, and if we know this spectral sequence, then this does not
necessarily mean that we can actually compute A; we only obtain the quo-
tients F∗ A/F∗−1 A of the associated filtration (as depicted in Figure 3.4)! I.e.,
we still have to solve a sequence of extension problems (Figure 3.5).
In most cases, one is not able to determine the differentials (dr )r∈N>0
explicitly; however, the degrees of these differentials already tell a lot about
the spectral sequence and its long-term development, and additional external
input might provide enough information to extract non-trivial conclusions out
of a spectral sequence.
In a way, spectral sequences behave more like puzzles than like determin-
istic processes. We will explain some of the basic techniques for handling
spectral sequences below (Chapter 3.2.3).

Dually to the concept of homological spectral sequences there is also a


notion of cohomological spectral sequences:

Definition 3.2.7 (cohomological spectral sequence). A (bigraded) cohomolog-


ical spectral sequence over a ring R is a sequence (Er , dr )r∈N>0 of bigraded
R-modules Er and R-homomorphisms dr : Er −→ Er with the following prop-
erties:

• For every r ∈ N>0 the map dr has degree (r, −r + 1), and dr ◦ dr = 0.

• For every r ∈ N>0 there is an isomorphism


ker dr
Er+1 ∼
=R H∗ (Er , dr ) = ,
im dr
and this isomorphism is also part of the data of the spectral sequence.

Similar to the homological case, the ∞-page and degeneration are defined for
cohomological spectral sequences.

Definition 3.2.8 (convergence of a cohomological spectral sequence). Let R be


a ring, let A be an N-graded R-module, let A be an N-graded R-module, and
let (Fn A)n∈N be an decreasing filtration of A that is compatible with the
grading of A. We say that a cohomological spectral sequence (Er , dr )r∈N>0
converges to A if the following conditions are satisfied:

• For all p, q ∈ N we have

pq ∼ Fp Ap+q
E∞ =R
Fp+1 Ap+q

• The filtration F∗ A is exhaustive and Hausdorff, i.e., F0 A = A and


Fn+1 A = 0.
122 3. The derived view

3.2.2 Classical spectral sequences

Most spectral sequences are based on the following classical prototypes. It all
starts with the big brother of the long exact homology sequence associated
with a short exact sequence of chain complexes (Proposition A.2.2):

Theorem 3.2.9 (filtration spectral sequence [87, Chapter 5.4]). Let R be a


ring, let C∗ be an (N-indexed) R-chain complex, and let (Fp C)p∈N∪{−1} be a
canonically bounded filtration of C∗ , i.e.,

0 = F−1 C ⊂ F0 C ⊂ F1 C ⊂ · · · ⊂ C∗

is a nested chain of subcomplexes of C∗ with (Fn C)n = Cn for all n ∈ N. Then


there is a natural converging homological spectral sequence (E r , dr )r∈N≥1 with
1
Epq = Hp+q (Fp C/Fp−1 C) =⇒ Hp+q (C∗ ).

For many applications it is not relevant to know the corresponding fil-


tration on H∗ (C∗ ) or the construction of the differentials – existence and
naturality is often enough!
Applying the filtration spectral sequence to the horizontal/vertical filtra-
tion of a double complex leads to the two double complex spectral sequences:

Theorem 3.2.10 (double complex spectral sequences [87, Theorem 5.5.1]). Let
R be a ring and let C∗∗ be an (N × N-indexed) R-double complex, i.e., a fam-
h
ily (Cpq )p,q∈N of R-modules together with R-homomorphisms (∂p,q : Cp,q →
v
Cp−1,q )p,q∈N and (∂p,q : Cp,q → Cp,q−1 )p,q∈N satisfying

∂v ◦ ∂v = 0 (columns are chain complexes)


h h
∂ ◦∂ =0 (rows are chain complexes)
v h h v
∂ ◦∂ +∂ ◦∂ =0 (compatbility of rows/columns)

Then there exist two natural converging homological spectral sequences:


À 1 v
Epq = Hq (Cp,∗ , ∂p,∗ ) =⇒ Hp+q (Tot C)
Á 1 h
Epq = Hq (C∗,p , ∂∗,p ) =⇒ Hp+q (Tot C),

where Tot C is the total chain complex (check!)


Ln associated with the double
complex C∗∗ : For n ∈ N, we have (Tot C)n := j=0 Cj,n−j and the boundary
operator

(Tot C)n −→ (Tot C)n−1


x 7−→ ∂ v (x) + ∂ h (x).
3.2. The Hochschild-Serre spectral sequence 123

These double complex spectral sequences in turn can be applied to reso-


lutions of chain complexes (leading to double complexes), which results in a
spectral sequence for the derived functor of a composition:

Theorem 3.2.11 (Grothendieck spectral sequence [87, Corollary 5.8.4]). Let


F : C −→ D and G : D −→ E be right-exact functors between module cate-
gories. Moreover, we assume that F is G-acyclic, i.e., that for every projec-
tive P ∈ Ob(C) and every n ∈ N>0 , we have Ln G(F (P )) ∼ =E 0. Then there
is a converging homological spectral sequence
2

Epq = (Lp G) Lq F (A) =⇒ Lp+q (G ◦ F )(A).

In the situation of Theorem 3.2.11, the corresponding formula for total


derived functors on the derived categories is much simpler, namely [87, The-
orem 10.8.2]
L(G ◦ F ) ∼
= L(G) ◦ L(F ).

3.2.3 The spectral sequence of a group extension

Group (co)homology can be viewed as the derived functor of the (co)invariants


functor. If 1 −→ N −→ G −→ Q −→ 1 is a short exact sequence of groups,
then

G-invariants functor = (Q-invariants functor) ◦ (N -invariants functor)


G-coinvariants functor = (Q-coinvariants functor) ◦ (N -coinvariants functor)

The Grothendieck spectral sequence (Theorem 3.2.11) then translates into


the following spectral sequence (Figure 3.6):

Theorem 3.2.12 (Hochschild-Serre spectral sequence [87, Theorem 6.8.2]). Let


1 −→ N −→ G −→ Q −→ 1 be a short exact sequence of groups and let A be
a ZG-module.

1. Then there is a natural converging homological spectral sequence



2
Epq = Hp Q; Hq (N ; ResG
N A) =⇒ Hp+q (G; A);

here, Q ∼
=Group G/N acts on the coefficients H∗ (N ; ResG
N A) as described
in Proposition 3.2.13 below.

2. Similarly, there is a natural converging cohomological spectral sequence



E2pq = H p Q; H q (N ; ResG
N A) =⇒ H
p+q
(G; A).
124 3. The derived view

Hp (Q; Hq (N ; ResG
N A))

p
2
E

Figure 3.6.: the Hochschild-Serre spectral sequence, schematically

Proposition 3.2.13 (conjugation action on (co)homology). Let G be a group,


let N ⊂ G be a normal subgroup, let g ∈ G, and let A be a ZG-module.
Moreover, let

c(g) := (x 7→ g · x · g −1 , x 7→ g · x)

∈ MorGroupMod (N, ResG G
N A), (N, ResN A)
c∗ (g) := (x 7→ g −1 · x · g, x 7→ g · x)

∈ MorGroupMod* (N, ResG G
N A), (N, ResN A)

be the associated conjugation morphisms in GroupMod and GroupMod* , re-


spectively (check!).

1. Then, for all k ∈ N and all n ∈ N ,



Hk c(n) = idHk (N ;ResG
N A)
k ∗

H c (n) = idH j (N ;ResG
N A)
.

2. In particular: If Q := G/N is the corresponding quotient, then conjuga-


tion of G on N induces a well-defined action of Q on H∗ (N ; ResGN (A))
via

Q × Hk (N ; ResG G
N A) −→ Hk (N ; ResN A)

(g · N, α) 7−→ Hk c(g) (α)

and similarly on H ∗ (N ; ResG


N (A)).

Proof. This can, for instance, be proved via the fundamental theorem of
group (co)homology (Corollary 1.6.9) (Exercise).
3.2. The Hochschild-Serre spectral sequence 125

Clearly, on H1 (N ; Z) (with the trivial N -action on Z), this action is just


the conjugation action of G on Nab .

Remark 3.2.14 (naturality of the Hochschild-Serre spectral sequence). The


Hochschild-Serre spectral sequence is natural in the following sense: For sim-
plicity, let A be a Z-module on which all of the following groups act trivially.
If

1 /N /G /Q /1

f g h
  
1 / N0 / G0 / Q0 /1

is a commutative diagram of groups with exact rows, then the corresponding


induced homomorphisms on homology fit together to form a morphism

2
Epq = Hp Q; Hq (N ; A) +3 Hp+q (G; A)

Hp (h;Hq (f ;A)) Hp+q (g;A)


  
2
Epq = Hp Q0 ; Hq (N 0 ; A) +3 Hp+q (G0 ; A)

of spectral sequences (i.e., homomorphisms between the corresponding pages


of the spectral sequences that are compatible with the differentials, and such
that the map between the (r + 1)-st pages is the map induced on homology
by the map between the r-th pages).
Similarly, the cohomological Hochschild-Serre spectral sequence is natural
with respect to such morphisms.

Outlook 3.2.15 (multiplicativity of the Hochschild-Serre spectral sequence).


Let 1 −→ N −→ G −→ Q −→ 1 be a short exact sequence of groups. If A
is a ZG-algebra, then the cohomological Hochschild-Serre spectral sequence
is multiplicative in the following sense: All pages (Er )r∈N≥2 carry the struc-
ture of a differential graded algebra (i.e., they are equipped with a graded
commutative product such that the differentials satisfy the Leibniz rule) such
that

• the product on the E2 -term coincides with the so-called cup-product


on the cohomology H ∗ (Q; H ∗ (N ; ResG
N A)) (Outlook 3.2.28), and

• such that the filtration on H ∗ (G; A) induced by this spectral sequence


is also compatible with the cup-product on H ∗ (G; A).

This additional, multiplicative, information is useful in computations.

We give some simple examples to illustrate basic techniques in spectral


sequence computations:
126 3. The derived view

5 0 0 0 0 0 0
4 0 0 0 0 0 0
3Z/3 0 0 0 0 0
2 0 0 0 0 0 0
1 0 0 0 0 0 0
0 Z Z/2 0 Z/2 0 Z/2 p
0 1 2 3 4 5
E2 = E∞

Figure 3.7.: the Hochschild-Serre spectral sequence for S3

Degeneration at the E 2 -stage, trivial extension problems. We start illus-


trating the use of the Hochschild-Serre spectral sequence by computing the
homology of the symmetric group S3 :

Example 3.2.16 (the symmetric group S3 ). The symmetric group S3 fits into
a group extension

1 −→ Z/3 −→ S3 −→ Z/2 −→ 1,

where the quotient Z/2 acts on the kernel Z/3 by taking inverses. The
Hochschild-Serre spectral sequence then gives us:
2

Epq = Hp Z/2; Hq (Z/3; Z) =⇒ Hp+q (S3 ; Z),

where Z/2 acts on the coefficients H∗ (Z/3; Z) by the maps induced by tak-
ing inverses on Z/3; i.e., for k ∈ N, the group Z/2 acts by multiplication
by (−1)k+1 on H2k+1 (Z/3; Z) ∼ = Z/3 (Exercise).

1. How does the E 2 -term look like? The description of the Z/2-action
on the homology of Z/3 gives the vertical axis of the E 2 -term of the
Hochschild-Serre spectral sequence (recall that zeroth homology is given
by taking coinvariants; Theorem 1.3.1).
Of course, the horizontal axis is nothing but H∗ (Z/2; Z). In view of the
torsion results provided by the transfer (Corollary 1.7.16), we obtain

2
Epq = Hp Z/2; Hq (Z/3; Z) ∼ =Z 0

for all p, q ∈ N>0 . Therefore, the E 2 -term looks as in Figure 3.7.


3.2. The Hochschild-Serre spectral sequence 127

2. Are there non-trivial differentials? For every r ∈ N≥2 , the differential dr


of the Hochschild-Serre spectral sequence has degree (−r, r − 1); in
particular, the horizontal and the vertical component of the bidegree
have different parity. Hence, all differentials (dr )≥2 have to be trivial in
this example. In other words, the spectral sequence corresponding to the
above extension degenerates at the E 2 -stage, and therefore E ∞ ∼ =Z E 2 .
3. What about the extension problems? From the E ∞ -page of the spec-
tral sequence, for k ∈ N>0 we obtain short exact sequences of Abelian
groups of the following types:

0 −→ Hk (S3 ; Z) −→ 0 if k ≡ 0 mod 4
0 −→ Hk (S3 ; Z) −→ Z/2 −→ 0 if k ≡ 1 mod 4
0 −→ Hk (S3 ; Z) −→ 0 if k ≡ 2 mod 4
0 −→ Z/3 −→ Hk (S3 ; Z) −→ Z/2 −→ 0 if k ≡ 3 mod 4.

The classification of finitely generated Abelian groups tells us that all


these extensions have to be trivial. Therefore, we obtain



 Z if k = 0



 0 if k ≡ 0 mod 4 and k > 0

Hk (S3 ; Z) =Z Z/2 if k ≡ 1 mod 4



 0 if k ≡ 2 mod 4


Z/6 if k ≡ 3 mod 4.

Degeneration at the E 2 -stage, non-trivial extension problems. We now give


an example of a Hochschild-Serre spectral sequence that still degenerates at
the E 2 -term, but where the resulting extension problems are non-trivial:
Example 3.2.17 (the infinite dihedral group). We consider the infinite dihedral
group D∞ , which fits into an extension

1 −→ Z −→ D∞ −→ Z/2 −→ 1,

where the quotient Z/2 acts on the kernel Z by taking inverses. The Hochschild-
Serre spectral sequence then gives us:
2

Epq = Hp Z/2; Hq (Z; Z) =⇒ Hp+q (D∞ ; Z),

where Z/2 acts on the coefficients H∗ (Z; Z) by the maps induced by taking
inverses in Z; i.e., the group Z/2 acts trivially on H0 (Z; Z) ∼
=Z Z and by
multiplication by −1 on H1 (Z; Z) ∼
=Z Z (check!).
1. How does the E 2 -term look like? With help of the standard periodic
Z[Z/2]-resolution of Z (see the proof of Corollary 1.6.13) we see that the
E 2 -term of this spectral sequence has the shape depicted in Figure 3.8.
128 3. The derived view

3 0 0 0 0 0 0
2 0 0 0 0 0 0
1Z/2 0 Z/2 0 Z/2 0
0 Z Z/2 0 Z/2 0 Z/2 p
0 1 2 3 4 5
E2 = E∞

Figure 3.8.: the Hochschild-Serre spectral sequence for D∞

2. Are there non-trivial differentials? Looking at the degrees of the dif-


ferentials, we see that the differentials all start or end in 0; so, there
are no non-trivial differentials. In other words, the spectral sequence
degenerates at the E 2 -stage, and thus E ∞ ∼ =Z E 2 .
3. What about the extension problems? We obtain H0 (D∞ ; Z) ∼ =Z Z, and
Hk (D∞ ; Z) ∼
=Z 0 for all even k ∈ N>0 from the spectral sequence.
For all odd k ∈ N, the E ∞ -term of the Hochschild-Serre spectral se-
quence gives us short exact sequences of the following type:

0 −→ Z/2 −→ Hk (D∞ ; Z) −→ Z/2 −→ 0.

Now the classification of finitely generated Abelian groups shows that


=Z Z/4 or Hk (D∞ ; Z) ∼
Hk (D∞ ; Z) ∼ =Z Z/2 ⊕ Z/2.
In order to determine which of these alternatives actually happens, we
take advantage of the fact that the group extension

1 −→ Z −→ D∞ −→ Z/2 −→ 1,

we started with splits. In particular, the identity homomorphism Z/2 ∼


=Z
Hk (Z/2; Z) −→ Hk (Z/2; Z) ∼ =Z Z/2 factors through Hk (D∞ ; Z). Hence,
Hk (D∞ ; Z) cannot be isomorphic to Z/4. Therefore,

Hk (D∞ ; Z) ∼
=Z Z/2 ⊕ Z/2

for all odd k ∈ N. Alternatively, one could also use H1 (D∞ ; Z) ∼ =Z


(D∞ )ab in degree 1, and in the higher odd degrees a transfer argument
(Example 1.7.17).

This completes the computation of H∗ (D∞ ; Z) via the Hochschild-Serre spec-


tral sequence. Alternatively, one can also compute the homology of D∞ using
the decomposition D∞ ∼ =Group Z/2 ∗ Z/2 (Example 4.1.29).
3.2. The Hochschild-Serre spectral sequence 129

3.2.4 A proof of Hopf’s formula


Finally, we also have a suitable tool to give a straightforward proof of Hopf’s
formula (Theorem 1.5.1):

Theorem 3.2.18 (Hopf’s formula). Let F be a free group, let N ⊂ F be a nor-


mal subgroup, and let G := F/N . Then there is an exact sequence (in Z Mod)

0 / H2 (G; Z) / H1 (N ; Z)G / H1 (F ; Z) / H1 (G; Z) /0

(where the homomorphisms on H1 are induced by the canonical inclusion and


projection, respectively, and G acts on H1 (N ; Z) by conjugation of represen-
tatives in F ). More explicitly,

N ∩ [F, F ]
H2 (G; Z) ∼
=Z .
[F, N ]

Proof. Let us first prove the explicit description out of the exact sequence:
By the exact sequence, H2 (G; Z) is isomorphic to kernel of the homomor-
phism H1 (N ; Z)G −→ H1 (F ; Z) induced by the inclusion N −→ F . Using
the computation of group homology in degree 1 (Theorem 1.4.1), we obtain
a canonical isomorphism

H1 (N ; Z)G ∼
=Z N/[N, N ] SpanZ {g · [n] − [n] | n ∈ N, g ∈ G}

= N/[N, N ] SpanZ {[f · n · f −1 ] − [n] | n ∈ N, f ∈ F }

= N/[N, N ] SpanZ {[f · n · f −1 · n−1 ] | n ∈ N, f ∈ F }
= N/[F, N ].

Therefore, the kernel of the map H1 (N ; Z)G −→ H1 (F ; Z) induced by the


inclusion N −→ F is isomorphic to (N ∩ [F, F ])/[F, N ], as desired.
It remains to establish the four-term exact sequence: We apply the
Hochschild-Serre spectral sequence (Theorem 3.2.12) to the group extension

i π
1 /N /F /G /1

and obtain 
2
Epq = Hp G; Hq (N ; Z) =⇒ Hp+q (F ; Z).
In contrast with the previous examples, we know the “limit” of this spectral
sequence (this is just homology of a free group) and we will try to reverse
engineer the spectral sequence in order to compute H2 (G; Z):

1. How does the E 2 -term look like? As we do not know H2 (G; Z) yet, we
cannot compute all of the E 2 -page, but we obtain partial information
130 3. The derived view

3 0 0 0 0 0 0
2 0 0 0 0 0 0
1À • • • • •

0 Z Á Â • • •
p
0 1 2 3 4 5
E2

Figure 3.9.: the Hochschild-Serre spectral sequence for the proof of Hopf’s
formula

from the freeness of F : As subgroup of the free group F , also N is


a free group (Theorem AT.2.3.52). Therefore, by Corollary 1.6.23, for
all k ∈ N≥2 ,

Hk (F ; Z) ∼
=Z 0 and Hk (N ; Z) ∼
=Z 0.
2 ∼
In particular, Epq =Z 0 for all q ∈ N≥2 and so the spectral sequence
degenerates at stage 3 (because of the degrees of the differentials; Fig-
ure 3.9).
Moreover, we know that (by Theorem 1.3.1)

2
À = E10 = H1 G; H0 (N ; Z) ∼=Z H1 (G; Z)

Á = E01 = H0 G; H1 (N ; Z) ∼
2
=Z H1 (N ; Z)G

 = E20 = H1 G; H0 (N ; Z) ∼
2
=Z H2 (G; Z)

and the only potentially non-trivial differential in this range is

d220 : H2 (G; Z) ∼ 2
=Z E20 2 ∼
−→ E01 =Z H1 (N ; Z)G .

2. What do we know about the E ∞ -term? As the spectral sequence de-


generates at stage 3, we have E ∞ ∼
=Z E 3 .
• Because of H2 (F ; Z) ∼
=Z 0, we have E203 ∼ ∞ ∼
=Z E20 =Z 0.
• Convergence of the spectral sequence to H∗ (F ; Z) implies that
there is a short exact sequence

0 ∞
/ E01 / H1 (F ; Z) ∞
/ E10 / 0.

Moreover,
3.2. The Hochschild-Serre spectral sequence 131

∞ ∼ 3 ∼ 2 ∞ ∼ 3 ∼
E01 =Z E01 =Z E01 / im d220 and E10 =Z E10 2
=Z E10 .

3. Refining the information on the E 2 -term. In view of the previous step


and the transition between the pages of a spectral sequence, we have

0∼ 3 ∼
=Z E20 =Z ker d220 / im 0,

and so d220 is injective.

We can now put it all together. Thus, we obtain a four-term exact sequence

“d2 ”
0 / H2 (G; Z) 20 / H1 (N ; Z)G Ä / H1 (F ; Z) Ã / H1 (G; Z) / 0.

It remains to show that the homomorphisms on H1 are induced by the


group homomorphisms in the original extension (up to sign). Of course, if
one knows the inner workings of a concrete construction of the Hochschild-
Serre spectral sequence, then one can deduce this fact from the construction.
However, as we blackboxed the construction, we will instead rely on the
naturality of the Hochschild-Serre spectral sequence.
To this end, we consider the commutative diagram (in Group)

idF
1 /1 /F /F /1

1 idF π
  
1 /N /F
π
/G /1
i

with exact rows. By the naturality of the Hochschild-Serre spectral sequence


(Remark 3.2.14), this leads to a corresponding transformation between the
associated Hochschild-Serre spectral sequences and thus to the following com-
mutative diagram (which are parts of the four-term exact sequences):
ϕ
0∼
=Z H1 (1; Z)F / H1 (F ; Z) / H1 (F ; Z) /0

H1 (idF ;Z) H1 (π;Z)


 
H1 (F ; Z) / H1 (G; Z)
Ã

Hence, ϕ : H1 (F ; Z) −→ H1 (F ; Z) is an isomorphism (but we do not know


yet, which one). We will now show that ϕ = ± idH1 (F ;Z) , again, by naturality:
If f : F −→ F is a group automorphism, then naturality of the Hochschild-
Serre spectral sequence and the associated four-term exact sequence shows
that
132 3. The derived view

ϕ
H1 (F ; Z) / H1 (F ; Z)

H1 (f ;Z) H1 (f ;Z)
 
H1 (F ; Z) ϕ
/ H1 (F ; Z)

is a commutative diagram. Because H1 (F ; Z) ∼ =Z Fab is free Abelian (with


rank equal to the rank of F ), using the right type of automorphisms f of F
shows that ϕ is sufficiently central in the automorphism group of this free
Abelian group (check!) and thus equal to ± idH1 (F ;Z) (check!). In combination
with the previous commutative diagram, we therefore obtain à = ±H1 (π; Z).
Analogously, one can use the extension 1 −→ F −→ F −→ 1 −→ 1 given
by the identity on F to deduce that Ä is induced by ±H1 (i; Z) (check!).
Because sign changes do not affect the exactness properties, we obtain the
claimed exact sequence.

3.2.5 Universal coefficients and products

Products of two groups are a special case of extensions. Hence, we could


apply the Hochschild-Serre spectral sequence (Theorem 3.2.12) to relate the
(co)homology of the product to the (co)homology of the two factors. However,
it turns out that we can obtain more refined information if we work with
the double complex spectral sequences directly. This strategy leads to the
universal coefficient theorem and the Künneth formula. For simplicity, we
focus on the case that the base ring is a principal ideal domain (but the
theory in principle also allows to treat more general cases):

Theorem 3.2.19 (Künneth spectral sequence). Let R be a ring, let P∗ ∈


Ob(ChR ) a chain complex consisting of flat R-modules, and let A ∈ Ob(R Mod).
Then there is a natural (both in P∗ and A) converging spectral sequence

2
Epq = TorR p Hq (P∗ ), A =⇒ Hp+q (P∗ ⊗R A).

Proof. We will prove this via the double complex spectral sequences. How
do we get a double complex involved? Let (Q∗ , ε) be a projective resolution
of A over R. Then P∗ ⊗R Q∗ (by definition) is the total complex of the double
complex C∗∗ , which is given by Cp,q := Pp ⊗R Qq for all p, q ∈ N and the
horizontal/vertical boundary maps
h
∂p,q : Cp,q −→ Cp−1,q
x ⊗ y 7−→ ∂pP (x) ⊗ y
v
∂p,q : Cp,q −→ Cp,q−1
x ⊗ y 7−→ (−1)p · x ⊗ ∂qQ (y).
3.2. The Hochschild-Serre spectral sequence 133

Then we have two natural spectral sequences converging to H∗ (P∗ ⊗R Q∗ )


(Theorem 3.2.10):
À 1
Epq = Hq (Pp ⊗R Q∗ , ∂∗v ) =⇒ Hp+q (Tot C) = Hp+q (P∗ ⊗R Q∗ )
Á 1
Epq = Hq (P∗ ⊗R Qp , ∂∗h ) =⇒ Hp+q (Tot C) = Hp+q (P∗ ⊗R Q∗ )

We will now use the first spectral sequence to identify the limit as H∗ (P∗ ⊗R A)
and then the second spectral sequence to obtain the short exact sequence.

• Because Pp is flat and (Q∗ , ε) is a projective resolution of A, we have


(
À 1 ∼ ∼ Pp ⊗R A if q = 0
Epq =R Pp ⊗R Hq (Q∗ ) =R
0 if q > 0.

In particular, the first spectral sequence degenerates at stage 2, the ∞-


page is concentrated in the “q = 0”-line, and thus Hn (P∗ ⊗R Q∗ ) ∼ =R
À ∞ ∼ À 2 ∼
En,0 =R En,0 =R Hn (P∗ ⊗R A) for all n ∈ N.

• Therefore, also the second spectral sequence converges to H∗ (P∗ ⊗R A).


Moreover, because Qp is projective/flat,
1
Epq = Hq (P∗ ⊗R Qp ) ∼
=R Hq (P∗ ) ⊗R Qp

and so (by construction of TorR


∗ ( · , A))

Á 2 ∼
 
Epq =R Hp Hq (P∗ ) ⊗R Q∗ ∼ =R TorR
p Hq (P∗ ), A .

Hence, we obtain the desired natural converging spectral sequence



2
Epq = TorRp Hq (P∗ ), A =⇒ Hp+q (P∗ ⊗R A).

Corollary 3.2.20 (algebraic universal coefficient theorem). Let R be a principal


ideal domain, let P∗ ∈ Ob(ChR ) be a chain complex consisting of free R-
modules, and let A ∈ Ob(R Mod). Then, for each n ∈ N, there is a natural
short exact sequence
µ 
0 / Hn (P∗ ) ⊗R A / Hn (P∗ ⊗R A) / TorR
1 Hn−1 (P∗ ), A
/ 0.

of R-modules, where

µ : Hn (P∗ ) ⊗R A −→ Hn (P∗ ⊗R A)
[z] ⊗ a 7−→ [z ⊗ a].

The sequence splits (but not naturally); in particular,


 
Hn (P∗ ⊗R A) ∼ =R Hn (P∗ ) ⊗R A ⊕ TorR 1 Hn−1 (P∗ ), A .
134 3. The derived view

Proof. Because R is a principal ideal domain, TorR ∼


p ( · , A) =R 0 for all p ∈
N≥2 (every module admits a short projective resolution (Example IV.5.3.13;
which also works without finite generation)) and TorR ∼
0 ( · , A) =R · ⊗R A.
Therefore, the Künneth spectral sequence is concentrated on the two lines
with “p ∈ {0, 1}”, thus degenerates at stage 2, and from convergence we
obtain natural short exact sequences of the form
µP∗ ,A 
0 / Hn (P∗ ) ⊗R A / Hn (P∗ ⊗R A) / TorR
1 Hn−1 (P∗ ), A
/ 0.

As next step, we show that µP∗ ,A (up to a unit in R, which we can safely
ignore as it does not affect the kernels or images) has the claimed form
(via naturality!): We consider the chain complex Q∗ that just consists of the
module R, concentrated in degree n. Then, in the corresponding short exact
sequence for Q∗ ⊗R R we know that µQ∗ ,R : Hn (Q∗ ) ⊗R R −→ Hn (Q∗ ⊗R R)
is an isomorphism (the right outer term is trivial). Because

Hn (Q∗ ) ⊗R R ∼
=R R ⊗R R ∼
=R R and Hn (Q∗ ⊗R R) ∼
=R Hn (Q∗ ) ∼
=R R,

the homomorphism µQ∗ ,R corresponds (under these isomorphisms) to multi-


plication by a unit r ∈ R× . Let z ∈ Pn be a cycle and let a ∈ A. Then

R −→ Pn
1 7−→ z

defines an R-chain map f∗ : Q∗ −→ P∗ (as z is a cycle). Together with the


module homomorphism g : R −→ A with g(1) = a and naturality of the short
exact sequence, we obtain the commutative diagram
µQ∗ ,R
Hn (Q∗ ) ⊗R R / Hn (Q∗ ⊗R R)

Hn (f∗ )⊗R g Hn (f∗ ⊗R g)


 
Hn (P∗ ) ⊗R A µP∗ ,A
/ Hn (P∗ ⊗R A)


of R-modules. Therefore, µP∗ ,A [z] ⊗ a = r · [z ⊗ a] (and every element
of Hn (P∗ ) ⊗R A is an R-linear combination of such elementary tensors).
Hence, we may assume that µ has the shape in the statement and it remains
to show that µ admits a split. Because the chain modules of P∗ are free and R
is a principal ideal domain, there is an R-homomorphism pn : Pn −→ ker ∂n
with pn ◦ in = idker ∂n , where in : ker ∂n −→ Pn is the inclusion (check! im ∂n
is free as a submodule of Pn−1 . . . ). Then a straightforward computation
shows that the following map is a well-defined split of µ (check!):

Hn (P∗ ⊗R A) −→ Hn (P∗ ) ⊗R A
Xk  k
X  
zj ⊗ aj 7−→ pn (zj ) ⊗ aj
j=1 j=1
3.2. The Hochschild-Serre spectral sequence 135

Alternatively, one can also prove this theorem directly, by manipulating the
chain complexes and homology groups by hand [87, Chapter 3.6]. The split-
ting in Corollary 3.2.20 is not natural in P∗ [42, proof of Proposition V.2.4].
Moreover, the splitting also exists if the complex P∗ consists of flat (instead
of free) modules [42, Theorem V.2.1].
Study note. What does the universal coefficient theorem say if the principal
ideal domain is a field?! How do these proofs relate to symmetry of Tor ?
Literature exercise (Theorems for free!). Read “Theorems for free!” by Wad-
ler [86]. What does this have to do with the proof of Corollary 3.2.20?
In particular, we can use the algebraic universal coefficient theorem to
relate group homology with coefficients with trivial action to group homology
with Z-coefficients.
Corollary 3.2.21 (universal coefficients for group homology). Let G be a group,
let A be a Z-module (with trivial G-action), and let n ∈ N. Then
 
=Z Hn (G; Z) ⊗Z A ⊕ TorZ1 Hn−1 (G; Z), A .
Hn (G; A) ∼

Proof. We have
 
Hn (G; A) = Hn C∗ (G) ⊗G A ∼=Z Hn (C∗ (G) ⊗G Z) ⊗Z A .

Moreover, C∗ (G) ⊗G Z consists of free Z-modules (check!) and thus we can


apply the universal coefficient theorem (Corollary 3.2.20) to this situation
(over the ring Z).
To understand the homology of product groups, we need to upgrade the
universal coefficient theorem to tensor products of two chain complexes:
Theorem 3.2.22 (algebraic Künneth theorem). Let R be a principal ideal do-
main, let P∗ ∈ Ob(ChR ) be a chain complex consisting of flat R-modules,
and let Q∗ ∈ Ob(R Ch). Then, for each n ∈ N, there is a natural short exact
sequence
n
M n−1
M 
0 −→ Hp (P∗ ) ⊗R Hn−p (Q∗ ) −→ Hn (P∗ ⊗R Q∗ ) −→ TorR1 Hp (P∗ ), Hn−1−p (Q∗ ) −→ 0
p=0 p=0

of R-modules. Moreover, the left map is given by [z] ⊗ [w] 7→ [z ⊗ w] and the
sequence splits (but not naturally).
Sketch of proof. This can be shown similarly to the proof of the universal co-
efficient theorem (Theorem 3.2.19, Corollary 3.2.20), using so-called Cartan-
Eilenberg resolutions of Q∗ and using Corollary 3.2.20 to identify the modules
in the E 2 -term [71, Theorem 10.90].
One should be aware that using the obvious double complex related
to P∗ ⊗R Q∗ and its two spectral sequences will give the correct limit and
good first pages, but it seems hard to figure out how to pass from page to
page or what the ∞-page looks like [71, Chapter 10.10].
136 3. The derived view

Corollary 3.2.23 (Künneth theorem for group homology). Let G and H be


groups, let R be a principal ideal domain (with trivial G- and H-action),
and let n ∈ N. Then (where the R-module structure is inherited from the
coefficients)
n
M
Hn (G × H; R) ∼
=R Hp (G; R) ⊗R Hn−p (H; R)
p=0
n−1
M 
⊕ TorR
1 Hp (G; R), Hn−1−p (H; R) .
p=0

Proof. With the goal in mind to apply the algebraic Künneth theorem (The-
orem 3.2.22), we show that (C∗ (G) ⊗Z C∗ (H), ε) is a projective Z[G × H]-
resolution of Z, where

ε : C0 (G) ⊗Z C0 (H) −→ Z
x ⊗ y 7−→ εG (x) · εH (y).

As first step, one checks that

Z[G × H] ←→ ZG ⊗Z ZH
G × H 3 (g, h) ←→ g ⊗ h

describes well-defined mutually inverse ring isomorphisms (check!).


In particular, C∗ (G) ⊗Z C∗ (H) is a Z[G × H]-chain complex that consists
of free Z[G × H]-modules (because C∗ (G) and C∗ (H) consist of free modules
over ZG and ZH, respectively, the compatibility of the tensor product with
direct sums, and the previous step). Moreover, the algebraic Künneth theorem
(Theorem 3.2.22) shows that (C∗ (G) ⊗Z C∗ (H))  ε has trivial homology.
Hence, (C∗ (G) ⊗Z C∗ (G), ε) is a projective resolution of Z over Z[G × H].
Therefore, by the fundamental theorem (Corollary 1.6.9),

Hn (G × H; R) ∼ =R Hn (C∗ (G) ⊗Z C∗ (H)) ⊗G×H R .

Using the mutually inverse chain isomorphisms (R is commutative!)


  
C∗ (G) ⊗Z C∗ (G) ⊗G×H R −→ C∗ (G) ⊗G R ⊗R C∗ (H) ⊗H R
(x ⊗ y) ⊗ r 7−→ (x ⊗ r) ⊗ (y ⊗ 1),
  
C∗ (G) ⊗G R ⊗R C∗ (H) ⊗H R −→ C∗ (G) ⊗Z C∗ (G) ⊗G×H R
(x ⊗ r) ⊗ (y ⊗ s) 7−→ (x ⊗ y) ⊗ (r · s)
 
we obtain Hn (G × H; R) ∼
=R Hn C∗ (G) ⊗G R ⊗R C∗ (H) ⊗H R . Now the
algebraic Künneth theorem (Theorem 3.2.22) finishes the proof.
3.2. The Hochschild-Serre spectral sequence 137

Example 3.2.24 (homology of Z × Z). Let G := Z × Z. Then the Künneth


formula (Corollary 3.2.23) allows us to compute H∗ (G; Z) (with trivial G-
action on Z): For all n ∈ N, we have
n
M
Hn (G; Z) ∼
=Z Hp (Z; Z) ⊗Z Hn−p (Z; Z)
p=0
n−1
M 
⊕ TorZ1 Hp (Z; Z), Hn−1−p (Z; Z) .
p=0

Let us first take care of the Tor-terms: If k ∈ N, then Hk (Z; Z) is a free Z-


module (Corollary 1.6.23); therefore, all Tor-terms are trivial and we obtain:

• We have
 
H1 (G; Z) ∼
=Z H0 (Z; Z) ⊗Z H1 (Z; Z) ⊕ H0 (Z; Z) ⊗Z H1 (Z; Z) ⊕ 0

=Z Z ⊕ Z.

• Moreover,

H2 (G; Z) ∼
=Z H1 (Z; Z) ⊗Z H1 (Z; Z) ⊕ 0

=Z Z.

• For all n ∈ N≥2 , we have (because there is always a contribution by a


higher degree homology of Z, which is trivial) Hn (G; Z) ∼
=Z 0.

Alternatively, one can also directly figure out a projective resolution for the
group Z × Z (Exercise) or use Theorem 1.4.1 for degree 1 and the Hochschild-
Serre spectral sequence (Theorem 3.2.12) for degree at least 2 (check!). More-
over, we will see a topological computation (Example 4.1.17).

Example 3.2.25 (homology of Z/p × Z/p). Let p ∈ N be prime. We consider


the group G := Z/p×Z/p and compute the homology H∗ (G; Fp ) (with trivial
G-action on the field Fp ) via the Künneth formula: For each n ∈ N, we have
(Corollary 3.2.23)
n
M
Hn (G; Fp ) ∼
=Fp Hj (Z/p; Fp ) ⊗Fp Hn−j (Z/p; Fp )
j=0
n−1
M F 
⊕ Tor1p Hj (Z/p; Fp ), Hn−1−j (Z/p; Fp ) .
j=0

In view of the computation of H∗ (Z/p; Fp ) (Corollary 1.6.13) and the fact


that Fp is a field, we obtain
138 3. The derived view
n
M
Hn (G; Z) ∼
=Fp Hj (Z/p; Fp ) ⊗Fp Hn−j (Z/p; Fp ) ⊕ 0
j=0

=Fp Fp n .

This behaviour is very different from the group homology of finite cyclic
groups or S3 (which all are “periodic”).
With a little more patience, we can also compute H∗ (G; Z) (with trivial
G-action on Z) via the Künneth theorem.

Outlook 3.2.26 (cohomological universal coefficient theorem and cohomological


Künneth theorem). The previous results also have counterparts in cohomol-
ogy:

• A cohomological universal coefficient theorem (that relates cohomology


of the dual cochain complex to the dual of the homology, using Ext for
the correction terms) [23, Section VI.4.2].

• A cohomological algebraic Künneth theorem [23, Theorem VI.10.11].

• A Künneth theorem for group cohomology [87, Exercise 6.1.8] (which


is also compatible with the product structures) under a mild finiteness
assumption.

Outlook 3.2.27 (topological Künneth theorem). The universal coefficient the-


orem can, of course, also be applied to singular or cellular chain complexes
(which consist of free modules!) in algebraic topology. This leads to the
universal coefficient theorem for singular/cellular (co)homology [23, Sec-
tion VI.7.8].
If X and Y are topological spaces, then, by the Eilenberg-Zilber theo-
rem [23, Theorem VI.12.1], there are canonical natural chain homotopy equiv-
alences
C∗ (X × Y ) 'Z C∗ (X) ⊗Z C∗ (Y )
of the corresponding singular chain complexes. Therefore,

H∗ (X × Y ; R) ∼
=R H∗ C∗ (X; R) ⊗R C∗ (Y ; R)

for all rings R. If R is a principal ideal domain, we can then apply the
algebraic Künneth theorem (Theorem 3.2.22) to obtain the Künneth formula
for each n ∈ N:
n
M
Hn (X × Y ; R) ∼
=R Hp (X; R) ⊗R Hn−p (Y ; R)
p=0
Mn

⊕ TorR
1 Hp (X; R), Hn−1−p (Y ; R) .
p=0
3.2. The Hochschild-Serre spectral sequence 139

Outlook 3.2.28 (irresponsible omission: product structures). Let G be a group


and let R be a commutative ring. Then the maps

· ∪ · : C p (G; R) ⊗R C q (G; R) −→ C p+q (G; R)



f ⊗R g −→ (g0 , . . . , gp+q ) 7→ (−1)p·q · f (g0 , . . . , gp ) · g(gp , . . . , gp+q )

induce a well-defined product, the so-called cup-product,

H p (G; R) ⊗R H q (G; R) −→ H p+q (G; R)


[f ] ⊗ [g] 7−→ [f ∪ g],
L
which turns n∈N H n (G; R) into a graded ring. This multiplicative structure
is functorial with respect to group homomorphisms (check!), carries valuable
additional information, and most constructions in group cohomology are com-
patible with this product structure. For a deeper understanding of group
(co)homology this multiplicative structure is essential [12, Chapter V/VI].
140 3. The derived view
4
The topological view

Finally, we arrive at the topological view on group (co)homology. By the


fundamental theorem, we can choose our favourite projective resolution to
compute group (co)homology; in particular, we can choose projective reso-
lutions of topological origin, namely chain complexes of classifying spaces of
the group in question.
We will first briefly survey the translation of group theory into topology
via classifying spaces. Nice models of classifying spaces then allow to com-
pute group (co)homology for many groups and lead to additional inheritance
properties (e.g., for free products).
We will then discuss higher finiteness properties of groups and sketch the
role of group (co)homology for the understanding of free actions of finite
groups on spheres.

Overview of this chapter.

4.1 Classifying spaces 142


4.2 Finiteness conditions 157
4.3 Application: Free actions on spheres 164

Running example. graphs, surfaces


142 4. The topological view

4.1 Classifying spaces


Classifying spaces allow to translate group theory into topology, more pre-
cisely, into homotopy theory. In this context, it is convenient to work
with topological spaces that admit a nice cellular structure, i.e., with CW-
complexes. This will allow us to obtain a topological version of the funda-
mental theorem (Corollary 1.6.9).

4.1.1 The standard simplicial model


The standard simplicial model of a group is the topological origin of the bar
resolution, constructed from simplices spanned by group elements.
Definition 4.1.1 (standard simplicial model). Let G be a group.
• The standard simplicial model of G is the topological space BG :=
G \ EG, where EG is defined as follows: Let
G 

EG := ∆n × Gn+1 ∼,
n∈N

where “∼” is the equivalence relation generated by


 
ij (t), (g0 , . . . , gn ) ∼ t, (g0 , . . . , gbj , . . . , gn )

for all n ∈ N>0 , j ∈ {0, . . . , n}, t ∈ ∆n+1 , g0 , . . . , gn ∈ G (here,


ij : ∆n−1 −→ ∆n is the inclusion of the j-th face; Proposition AT.4.1.3).
We equip EG with the quotient topology of the disjoint union topology
(where the products Gn+1 carry the discrete topology). Moreover, we
endow EG with the (well-defined and continuous) diagonal G-action:

G × EG −→ EG
  
g, [t, (g0 , . . . , gn )] 7−→ t, (g · g0 , . . . , g · gn ) .

• If f : G −→ H is a group homomorphism, then we write (which is


well-defined and continuous)

Ef : EG −→ EH
   
t, (g0 , . . . , gn ) 7−→ t, (f (g0 ), . . . , f (gn ))

Moreover, we write Bf : BG −→ BH for the associated (well-defined


and continuous; check!) map.
In this way, we obtain a functor B : Group −→ Top (check!).
4.1. Classifying spaces 143

Caveat 4.1.2. Usually, in the literature BG often refers to any choice of a


model of a classifying space for G (as in Definition 4.1.6). However, it seems
to convenient to reserve BG for this concrete model and to make the choice
of other models explicit.
Proposition 4.1.3 (properties of EG). Let G be a group.
1. The space EG is contractible and locally path-connected.
2. The diagonal action G y EG is properly discontinuous.
Proof. Ad 1. We use a cone construction as in the proof of Proposition 1.6.5.
The well-defined and continuous (check!) map

EG × [0, 1] −→ EG
  
[t, (g0 , . . . , gn )], s 7−→ s · i0 (t) + (1 − s) · e0 , (e, g0 , . . . , gn )

shows that EG is contractible; we use that [0, 1] is compact and Hausdorff


to ensure that · × [0, 1] is compatible with quotients [23, Lemma V.2.13].
The “obvious” simplicial structure on EG can be viewed as a CW-
structure. Therefore, path-connectedness implies local path-connectedness [28,
Theorem 1.3.2] (or one can use the neighbourhoods constructed below).
Ad 2. Let x ∈ EG; we can write x = [t, (g0 , . . . , gn )] with minimal n ∈ N.
Then t lies in the (relative) interior of ∆n . Hence, there is an ε ∈ (0, 1) with

t ∈ Uε := (t0 , . . . , tn ) ∈ (ε, 1 − ε)n t0 + · · · + tn = 1 ⊂ ∆n .

Then a straightforward calculation (check!) shows that


[ [ 
U := [s, (h0 , . . . , hm )] s ∈ ∆m , (sk0 , . . . , skn ) ∈ Uε ,
m∈N≥n k0 <···<kn ∈{0,...,m}
∀j∈{0,...,m}\{k0 ,...,kn } sj ∈ [0, ε),

(hk0 , . . . , hkn ) = (g0 , . . . , gn )

is an openFneighbourhood of x in EG and with (essentially because G acts


freely on n∈N Gn+1 )

∀g∈G g · U ∩ U 6= ∅ =⇒ g = e.

Therefore, the diagonal action G y EG is properly discontinuous.


Corollary 4.1.4 (properties of BG). Let G be a group.
1. Then the canonical projection EG −→ BG is a (contractible) universal
covering of BG and there is a canonical isomorphism

ϕG : π1 (BG, x0 ) −→ G

of groups through the deck transformation action, where x0 is the point


in BG represented by e ∈ G0+1 of EG.
144 4. The topological view

2. If f : G −→ H is a group homomorphism, then

f = ϕH ◦ π1 (Bf ) ◦ ϕ−1
G .

3. In particular, π1 ◦ (B · , x0 ) is canonically naturally isomorphic to the


identity functor on Group.
Proof. This follows from the properties of the action G y EG and covering
theory (Corollary AT.2.3.39; check!).
Using Corollary 4.1.4, we can model Group in Top. In particular, every
topological invariant leads to a corresponding invariant for groups.
Outlook 4.1.5 (classifying spaces of categories). Let G be a group. The
spaces EG and BG are (fat) geometric realisations of simplicial sets, which
in turn are nerves of certain categories [31, Definition II.20] (Exercise).

4.1.2 Changing the classifying space


As in the algebraic case, the standard simplicial model tends to be rather
big. Therefore, it is desirable to be able to replace the standard simplicial
model by other spaces that are better adapted to the groups/problems under
consideration. In the topological setting, we replace
• projective chain complexes by CW-complexes, and
• resolutions by spaces with contractible universal covering
(and whence projective resolutions by classifying spaces). Basic homotopy
theory of CW-complexes is collected in Appendix A.3.
Definition 4.1.6 (classifying space). Let G be a group. A (model of a) classi-
fying space for G is a pair ((X, x0 ), ϕ), consisting of
• a path-connected pointed CW-complex (X, x0 ) with contractible uni-
versal covering,
• and a group isomorphism ϕ : π1 (X, x0 ) −→ G.
In order to minimise notational overhead, one often also simply says that X
is classifying space for G (and leaves the rest of the data implicit).
Example 4.1.7 (classifying spaces of the trivial group). The one-point space
(together with the only point as base-point and the trivial group homomor-
phism) clearly is a classifying space for the trivial group.
More generally: If (X, x0 ) is a pointed CW-complex that is (pointedly)
contractible, then ((X, x0 ), trivial homomorphism : π1 (X, x0 ) −→ 1) also is a
classifying space for 1. In particular, classifying spaces are not unique up to
homeomorphism.
4.1. Classifying spaces 145

Example 4.1.8 (classifying spaces from group actions). Let G be a group and
let G y Y be a continuous group action on a topological space Y with the
following properties:

• The action is properly discontinuous.

• The space Y is contractible.

• The space Y is locally path-connected.

Then the orbit projection p : Y −→ G \ Y onto quotient space X := G \ Y


is the universal covering of X and given x0 ∈ X and y0 ∈ p−1 (x0 ), there
is a canonical group isomorphism ϕ : π1 (X, x0 ) −→ G (through the deck
transformation action; Corollary AT.2.3.39).
In particular: If X admits a CW-structure, then ((X, x0 ), ϕ) is a classifying
space for G.

Example 4.1.9 (standard simplicial models). Let G be a group. Then the stan-
dard simplicial model BG can be canonically turned into a classifying space
for G: In view of Example 4.1.8 and Corollary 4.1.4, we only need to specify
a CW-structure on BG and compatible base-points in BG and EG.

• As CW-structure on BG, we use the CW-structure induced by the


simplicial structure on the underlying simplicial complex/set for EG
(which descends to BG; check!).

• As base-point x0 , we take the 0-cell that corresponds to e ∈ G0+1


(which we choose as covering base-point in EG).

In particular, every group admits a classifying space (and the standard sim-
plicial construction is functorial).

Theorem 4.1.10 (recognising classifying spaces). Let G be a group, let (X, x0 )


be a path-connected pointed CW-complex, and let ϕ : π1 (X, x0 ) −→ G be a
group isomorphism. Then the following are equivalent:

1. The pair ((X, x0 ), ϕ) is a classifying space for G.

2. For all n ∈ N≥2 , we have πn (X, x0 ) ∼


=Group 1.
e Z) ∼
3. For all n ∈ N≥2 , we have Hn (X; e is “the” universal
=Z 0, where X
covering space of X.

Proof. Let p : X e −→ X be the universal covering map and let x e0 ∈ Xe


with p(ex0 ) = x0 .
Ad 1 =⇒ 2. Let ((X, x0 ), ϕ) be a classifying space for G and let n ∈ N≥2 .
Then p induces an isomorphism πn (p) : πn (X, e xe0 ) −→ πn (X, x0 ) (Corol-
lary AT.2.3.25). Because the universal covering X e is contractible, we ob-
∼ e ∼
e0 ) =Group 1 (similar to Example AT.2.1.8).
tain πn (X, x0 ) =Group πn (X, x
146 4. The topological view

Ad 2 =⇒ 3. This is a consequence of the compatibility of higher homotopy


e
groups with coverings (Corollary AT.2.3.25), the simply connectedness of X,
and the Hurewicz theorem (Theorem AT.4.5.6).
Ad 3 =⇒ 1. This follows from the Whitehead theorem (Corollary A.3.6),
applied to the constant map from Xe to the one-point space.

Theorem 4.1.11 (uniqueness of classifying spaces). Let G be a group and let


((X, x0 ), ϕ) and ((Y, y0 ), ψ) be classifying spaces of G. Then there exists a
(unique up to homotopy) pointed homotopy equivalence f : (X, x0 ) −→ (Y, y0 )
with
idG = ψ ◦ π1 (f ) ◦ ϕ−1 .

Proof. It suffices to consider the case that ((X, x0 ), ϕ) = ((BG, x0 ), ϕG )


(check!). Moreover, by the Whitehead theorem (Theorem A.3.4), it suffices to
construct a weak homotopy equivalence f : BG −→ X with π1 (f ) = ψ −1 ◦ϕG .
We will construct such an f inductively, by induction over the dimension of
the “simplices” of BG:

• Dimension 0: We take

f : [∆0 3 t, g ∈ G0+1 ] 7−→ y0 .

• Dimension 1: For g0 , g1 ∈ G, we map the edge in BG defined by (g0 , g1 )


as follows to Y :
 
f : ∆1 3 t, (g0 , g1 ) 7−→ γg−1 ·g1 (t0 ),
0

where γg−1 ·g1 : [0, 1]/ ∼−→ Y is a based loop that represents the element
0

  
ψ −1 ◦ ϕG s 7→ [(s, 1 − s), (e, g0−1 · g1 )] ∗ ∈ π1 (Y, y0 ).

• Dimension 2: For g0 , g1 , g2 ∈ G, we map the triangle in BG defined


by (g0 , g1 , g2 ) to Y by choosing a pointed homotopy in (Y, y0 ) between
the loops γg−1 ·g1 ∗ γg−1 ·g2 and γg−1 ·g2 (such a homotopy exists!).
0 1 0

• Dimension ≥ 3: For n ∈ N≥3 and g0 , . . . , gn ∈ G, the already con-


structed map {[t, [g0 , . . . , gn ]] ∈ BG | t ∈ ∂∆n } −→ Y represents an
element in πn−1 (Y, y0 ) ∼ =Group 1. Therefore, we can extend our map
continuously over the whole n-simplex {[t, [g0 , . . . , gn ]] ∈ BG | t ∈ ∆n }.

By construction, the resulting map f : BG −→ X induces ψ ◦ ϕG on π1 and


isomorphisms on the (trivial) higher homotopy groups. Hence, f is a weak
equivalence.
Similarly, inductive constructions over BG × [0, 1] show uniqueness up to
homotopy (check!).
4.1. Classifying spaces 147

Remark 4.1.12 (an inductive construction of classifying spaces). A more hands-


on construction of classifying spaces than the standard simplicial model can
be performed as follows: Let hS | Ri be a presentation of a group G. Induc-
tively, we construct a CW-complex X:
W
• 1-skeleton: We take S (S 1 , 1).
W
• 2-skeleton: We attach |R| two-dimensional disks D2 to S (S 1 , 1), ac-
cording to the relations in R. This two-dimensional CW-complex is
the presentation complex of hS | Ri. However, in general, the universal
covering of the presentation complex need not be contractible.

• For n ∈ N≥2 , from the n-skeleton Xn to the (n + 1)-skeleton Xn+1 : Let


An be a generating set of πn (Xn , 1). Then, for each a ∈ An , we choose
a map γ : S n −→ Xn representing a and attach an (n + 1)-cell Dn+1
via γ on ∂Dn+1 = S n .

By the Seifert and van Kampen theorem, the fundamental group of the re-
sulting CW-complex X is isomorphic to G. Moreover, the Blakers-Massey
theorem (or a careful application of the Hurewicz theorem to the universal
covering) allow to prove that the higher homotopy groups of X are trivial.
Applying Theorem 4.1.10 shows that X can be extended to a classifying space
for G [51, Proposition 3.83].

Theorem 4.1.13 (new classifying spaces out of old). Let G and H be groups
and let ((X, x0 ), ϕ) and ((Y, y0 ), ψ) be classifying spaces for G and H, respec-
tively. Then:

1. Subgroups. If K ⊂ G is a subgroup, then “the” pointed path-connected


covering q : (Y, y0 ) −→ (X, x0 ) of X corresponding to the subgroup K
of G yields a classifying space ((Y, y0 ), ϕ ◦ π1 (q, y0 )) of K.

2. Products. The product ((X × Y, (x0 , y0 )), ϕ × ψ) is a classifying space


for G × H.

3. Free products. The wedge ((X, x0 ) ∨ (Y, y0 ), [x0 ], ϕ ∗ ψ) is a classifying


space for the free product G ∗ H.

Proof. Ad 1. The covering space (Y, y0 ) inherits a CW-structure from the


CW-structure on (X, x0 ). By the classification theorem of coverings (The-
orem AT.2.3.43), the homomorphism ϕ ◦ π1 (q, y0 ) is a group isomorphism
π1 (Y, y0 ) −→ K and the universal covering space of (Y, y0 ), which coincides
with the one of (X, x0 ), is contractible.
Ad 2. Here, X ×Y denotes the product CW-complex (which in general does
not carry the product topology of X and Y ) [28, p. x, Theorem 2.2.2][24]. By
Theorem 4.1.10, it suffices to show that all higher homotopy groups of X × Y
are trivial and that ϕ × ψ is an isomorphism on π1 . This facts on homotopy
groups follow from: The canonical projections induce group isomorphisms
148 4. The topological view

πn X × Y, (x0 , y0 ) ∼=Group πn (X, x0 ) × πn (Y, y0 )

for all n ∈ N≥1 (the argument for the usual product (Proposition AT.2.2.4)
also works for this modified product [82]).
Ad 3. We equip the wedge (Z, z0 ) := (X, x0 ) ∨ (Y, y0 ) with the combined
CW-structure of the summands. By the Seifert and van Kampen theorem
(Theorem AT.2.2.6, Example AT.2.2.10), the group homomorphism

ϕ ∗ ψ : π1 (X, x0 ) ∨ (Y, y0 ) −→ G ∗ H

is an isomorphism. Therefore, it remains to prove that the total space Ze of


the universal covering p : Ze −→ Z of (Z, z0 ) is contractible: In view of the
Whitehead theorem (Theorem 4.1.10), we only need to show that Hn (Z; e Z)
is trivial for all n ∈ N≥2 . To this end, we use a Mayer-Vietoris argument:
Let X := p−1 (X), Y := p−1 (Y ) ⊂ Z. e

• Then X ∪ Y = Z e and X ∩ Y = p−1 ([x0 ]), which is a closed discrete


e
subspace of Z.
• The path-connected components of X are total spaces of coverings of X
(check!). Moreover, each path-connected component U ⊂ X is simply
connected: Let u0 ∈ U ∩ p−1 (x0 ) (which exists). By construction, we
have the following commutative diagram in Group (where the horizontal
arrows are induced by the inclusions):

π1 (U, u0 ) e u0 )
/ π1 (Z,

π1 (p|U ) π1 (p)
 
π1 (X, x0 ) / π1 (Z, z0 )

The vertical homomorphisms are injective (by covering theory; Corol-


lary AT.2.3.25) and the lower horizontal homomorphism is injective (by
the Seifert and van Kampen theorem, see above). Hence, also the upper
e u0 ) ∼
horizontal homomorphism is injective. Because π1 (Z, =Group 1, we

obtain also π1 (U, u0 ) =Group 1.
Therefore, p|U : U −→ X is a universal covering of X and thus is con-
tractible; in particular, U has trivial reduced homology. Similarly, we
can argue for Y .
We can now apply the Mayer-Vietoris sequence (Theorem AT.3.3.2) for
singular homology He ∗ ( · ; Z) (strictly speaking, we need a cellular version or
we need to thicken up the subspaces suitably). For n ∈ N≥2 , we consider the
following fragment of the Mayer-Vietoris sequence:

∆n
Hn (X; Z) ⊕ Hn (Y ; Z) e Z)
/ Hn (Z; / Hn (X ∩ Y ; Z).
4.1. Classifying spaces 149

The outer terms all are singular homology groups in non-zero degree of spaces
all of whose path-connected components are contractible. Therefore, strong
additivity of singular homology (Proposition AT.4.1.14) and homotopy in-
variance show that these groups are trivial. Therefore, we obtain
e Z) ∼
Hn (Z; =Z 0

If one wants to avoid technical problems in the Seifert and van Kampen
argument and the Mayer-Vietoris argument, areasonable alternative space
is the “thick” wedge (X, x0 ) t [0, 1] t (Y, y0 ) (x0 ∼ 0 ∧ 1 ∼ y0 ).

Outlook 4.1.14 (what do classifying spaces classify?). Let G be a group,


equipped with the discrete topology. Then the functor [ · , BG] : Toph −→ Set
classifies G-principal bundles. More explicitly: If X is a paracompact topolog-
ical space and f : X −→ BG is a continuous map, then f ∗ p (the pullback of
the covering p : EG −→ BG) is a principal G-bundle over X (as G is discrete,
principal G-bundles are just G-covering maps). Then

[X, BG] −→ isomorphism classes of principal G-bundles over X


[f ] 7−→ [f ∗ p]

is a well-defined bijection [85, Chapter 14.4]. In other words, BG (and the


universal covering p : EG −→ BG) classifies principal G-bundles.
Outlook 4.1.15 (Eilenberg-MacLane spaces). Let G be a group. Every classi-
fying space for G is a K(G, 1)-Eilenberg-MacLane space: If n ∈ N≥1 , then a
K(G, n)-Eilenberg-MacLane space is a triple ((X, x0 ), ϕ), consisting of
• a path-connected pointed CW-complex (X, x0 ) with

∀k∈N>0 \{n} πk (X, x0 ) ∼


=Group 1,

• and a group isomorphism ϕ : πn (X, x0 ) −→ G.


For each n ∈ N≥2 and each Abelian group G, there exists a K(G, n)-
Eilenberg-MacLane space.

4.1.3 Examples of classifying spaces


We list some prominent examples of “nice” classifying spaces for groups:
Example 4.1.16 (a classifying space for Z). Clearly, the circle S 1 is a classifying
space for Z:
• The circle S 1 admits a CW-structure (e.g., with a single 0-cell and a
single 1-cell).
150 4. The topological view

Figure 4.1.: A CW-structure on the 2-torus with one 0-cell, two 1-cells, and
one 2-cell

• The fundamental group of S 1 is isomorphic to Z (Theorem AT.2.3.40).


• The universal covering space of S 1 is R, which is contractible (Exam-
ple AT.2.3.31).
Example 4.1.17 (a classifying space for Z × Z). The 2-torus S 1 × S 1 is a
classifying space for the group T := Z × Z (Example 4.1.16, Theorem 4.1.13).
We equip S 1 × S 1 with the CW-structure in Figure 4.1. Then the univer-
sal covering space R2 of S 1 × S 1 inherits the CW-structure in Figure 4.2.
The cellular chain complex of this CW-structure on R2 (with the Z2 -action
induced by the deck transformation action) is isomorphic to the following
resolution of Z by Z[T ]-modules (check!):

∂2 ∂1 ε
··· /0 / ZT / ZT ⊕ ZT / ZT /Z

Here, we use that Z[T ] ∼


=Ring Z[a, b]{an ·bm |n,m∈N} (Exercise) and the maps

∂2 : ZT −→ ZT ⊕ ZT

x 7−→ x · (1 − b), x · (a − 1)
∂1 : ZT ⊕ ZT −→ ZT
(x, y) 7−→ x · (a − 1) + y · (b − 1),

together with the usual augmentation ε : ZT −→ Z.

Example 4.1.18 (a classifying space for Z/2). The infinite-dimensional real


projective space RP ∞ := colimn→∞ RP n is a classifying space for Z/2.
This can, for instance, be seen by looking at the antipodal Z/2-action
on S ∞ := colimn→∞ S n (which satisfies the hypotheses of Theorem 4.1.10).
The corresponding cellular chain complex of S ∞ (with the CW-structure in-
duced from the standard CW-structure on RP ∞ , which in turn has a single
cell in each dimension) is Z[Z/2]-isomorphic to the periodic resolution in
Proposition 1.6.11.
Example
W 4.1.19 (classifying spaces for free groups). Let S be a [finite] set.
Then S (S 1 , 1) is a classifying space for the free group F (S) (Example 4.1.16,
4.1. Classifying spaces 151

action by a, b:

b
a

Figure 4.2.: The universal covering of the 2-torus

W
Theorem 4.1.13). The universal covering of S (S 1 , 1) is a 2 · |S|-regular tree
(check!).
NowWlet S = {a, b} with a 6= b. We consider the CW-structure on the
wedge {a,b} (S 1 , 1) that consists of a single 0-cell and two 1-cells. Then
we obtain the corresponding CW-structure on the universal covering (Fig-
ure 1.6). The cellular chain complex of this CW-structure on the 4-regular
tree then is isomorphic to the resolution from Proposition 1.6.21, as explained
in Remark 1.6.22.

Example 4.1.20 (classifying spaces for surface groups). Let g ∈ N≥2 and let


Γg := a1 , . . . , ag , b1 , . . . , bg [a1 , b1 ] · · · · · [ag , bg ] .

Let Σg be the topological space that is obtained from a regular 4g-gon by the
identifications indicated in Figure 4.3 (in fact, this is a closed surface with
g “holes”).

• A straightforward application of the Seifert and van Kampen theorem


(Theorem AT.2.2.6) shows that the fundamental group of Σg is isomor-
phic to Γg .

• On Σg , we can find a CW-structure (e.g., with one 0-cell and 2 · g


one-cells and a single 2-cell; Exercise).

• Moreover, the universal covering of Σg is contractible. This can be


proved as follows [4, Proposition B.3.1]: In the hyperbolic plane H2 ,
there exists a regular geodesic 4g-gon all of whose angles are π/(2 · g).
152 4. The topological view

a2
b2 b2

a2 a1

b1 b1
a1

Figure 4.3.: From a 4g-gon to a surface of genus g (check!)

Then the Riemannian metric of H2 induces a well-defined Riemannian


metric on Σg (via this geodesic 4g-gon). In particular, Σg admits a Rie-
mannian metric of constant sectional curvature −1. Hence, the Rieman-
nian universal covering of Σg is isometric to H2 . Because H2 ∼
=Top R2 ,
the universal covering of Σg is contractible.
Alternatively, this can also be proved via a generalisation of Theo-
rem 4.1.13 from free products to amalgamated free products (and the
fact that Γg is an amalgamated free product of two free groups over Z).

Hence, we can view Σg as a classifying space of Γg .

The case of surface groups admits the following important generalisations:

Outlook 4.1.21 (manifolds of non-positive sectional curvature). Let (M, g) be


a (non-empty) complete Riemannian manifold with secg ≤ 0 (i.e., of non-
positive sectional curvature). By the Cartan-Hadamard theorem [49, Theo-
rem 11.5], the universal covering of M is homeomorphic to Rdim M and thus
contractible. Moreover, M admits a triangulation (as smooth manifold [89]),
whence a CW-structure. Therefore, we can view M as classifying space of its
fundamental group.
Oriented closed connected surfaces of genus at least 1 fall into this class
of examples (Example 4.1.20).
In contrast, by now one knows many examples of groups that admit a
closed manifold as a classifying space but which have “exotic” properties [20]
(and which thus are far away from being non-positively curved).

Outlook 4.1.22 (classifying spaces for hyperbolic groups). Let G be a hyper-


bolic group (i.e., a finitely generated group such that G with its word-metrics
is a Gromov-hyperbolic metric space [53, Chapter 7][37]; e.g., a surface group
or a free group) and let S ⊂ G be a finite generating set.
The Rips complex Rr (G, S) of G with respect to S for the radius r ∈ R>0
is the geometric realisation of the simplicial complex on the vertex set G with
the set
4.1. Classifying spaces 153

σ ⊂ G |σ| < ∞, ∀g,h∈σ dS (g, h) ≤ r
of simplices. If G is in addition torsion-free and r is large enough, then the
quotient G \ Rr (G, S) is a classifying space for G [37, Section 2.2].

Outlook 4.1.23 (classifying spaces for one-relator groups). A one-relator group


is a group G that admits a presentation hS | ri, where r ∈ F (S) is a single
relation. Then the following holds [67, 56, 62]:

• If r ∈ F (S) is not a proper power of an element in F (S), then the


presentation complex of hS | ri is a classifying space for G = hS | ri.

• If r ∈ F (S) is a proper power of an element in F (S), then the pre-


sentation complex of hS | ri is not a classifying space for G (because
the group G contains torsion and the presentation complex is finite-
dimensional; Corollary 4.2.7). But in this situation, the group G =
hS | ri is hyperbolic (and thus we can find a classifying space through
the Rips construction; Outlook 4.1.22).

Clearly, the fundamental groups of oriented closed connected surfaces of


genus at least 1 are one-relator groups.

4.1.4 Group (co)homology via classifying spaces


Chain complexes of classifying spaces lead to projective resolutions over the
fundamental group and thus can be used to compute group (co)homology. In
the following, we will use the following notation (on topological spaces/CW-
complexes and maps between them):

• C∗ for the singular chain complex

• H∗ for singular homology

• C∗CW for the cellular chain complex (associated with H∗ ( · ; Z))

• H∗CW for cellular homology (associated with H∗ ( · ; Z))

Theorem 4.1.24 (resolutions from classifying spaces). Let G be a group, let


e −→ X be “the” universal
((X, x0 ), ϕ) be a classifying space for G, and let X
covering of X (with the CW-structure induced from the CW-structure on X).
e Z), ε) is a free resolution of Z by ZG-modules. Here,
1. Then (C∗ (X;
e Z) carries the G-action induced from the deck transformation
C∗ (X;
action (and ϕ) and

e Z) −→ Z
ε : C0 (X;
e 3 σ 7−→ 1.
map(∆0 , X)
154 4. The topological view

e Z), εCW ) is a free resolution of Z by ZG-


2. Similarly, also (C∗CW (X;
CW e
modules. Here, C∗ (X; Z) carries the G-action induced from the deck
transformation action (and ϕ) and

e Z) −→ Z
εCW : C0CW (X;
0-cells σ 7−→ 1.

Corollary 4.1.25 (group homology via classifying spaces). Let G be a group, let
A be a Z-module (with trivial G-action), and let n ∈ N.

1. Then there is a canonical isomorphism Hn (G; A) ∼


=Z Hn (BG; A).
2. If f : G −→ H is a group homomorphism, then Hn (f ; A) corresponds
under this isomorphism to Hn (Bf ; A) : Hn (BG; A) −→ Hn (BH; A).

3. If ((X, x0 ), ϕ) is a classifying space for G, then there is a canonical


isomorphism

Hn (G; A) ∼
=Z Hn (X; A) ∼
=Z HnCW (X; A).

Proof of Theorem 4.1.24. Because X is a classifying space, its universal cov-


e is contractible and so
ering X

H∗ C∗ (X;e Z) = H∗ (X;e Z) ∼
=Z H∗ (•; Z)

H∗ C (X; Z) = H (X; Z) ∼
CW e

CW e

e Z) ∼
=Z H∗ (X; =Z H∗ (•; Z).

Together with the concrete computation of singular/cellular homology in de-


gree 0 (Proposition AT.4.1.15), we see that the homology of C∗ (X; e Z)  ε and
CW e e Z), ε)
of C∗ (X; Z)  ε CW
is trivial in all degrees (check!). Therefore, (C∗ (X;
CW e
and (C∗ (X; Z), ε ) both are resolutions of Z by ZG-modules.
CW

Moreover, the chain modules of C∗ (X; e Z) and C CW (X;e Z) consist of free



Z-modules that admit Z-bases on which G-acts freely (in the singular case,
the basis of all singular simplices; in the cellular case, the basis of all open
cells). Hence the chain modules of C∗ (X; e Z) and C∗CW (X;e Z) are free ZG-
modules.

Proof of Corollary 4.1.25. Ad 1. By Theorem 4.1.24, (C∗ (EG; Z), ε) is a


projective resolution of Z over ZG. By the fundamental theorem (Corol-
lary 1.6.9), there is thus a canonical isomorphism

Hn (G; Z) ∼
=Z Hn C∗ (EG; Z) ⊗G A .

We now relate the right-hand side to BG: The lifting properties of cov-
ering maps show (because simplices are simply connected) that the chain
map C∗ (p; Z) : C∗ (EG; Z) −→ C∗ (BG; Z) induced by the (covering) projec-
tion map p induces a chain isomorphism
4.1. Classifying spaces 155

C∗ (EG; Z) ⊗G A ∼
=Z Ch C∗ (BG; Z) ⊗Z A

(because G acts trivially on A). This proves the first part.


Ad 2. This can easily be seen by comparing the standard simplicial model
construction with the simplicial resolution construction and Corollary 4.1.4
(check!).
Ad 3. This is a direct consequence of the first part and the uniqueness
of classifying spaces up to homotopy equivalence (Theorem 4.1.11). More-
over, cellular homology is known to coincide with singular homology (Theo-
rem AT.5.2.13).

Outlook 4.1.26 (group cohomology via classifying spaces). Analogously, there


is also a version of Corollary 4.1.25 for group cohomology and singular/cellular
cohomology of classifying spaces. Moreover, the corresponding isomorphisms
are also compatible with the respective cup-products [12, Chapter V].
Outlook 4.1.27 (twisted coefficients). The results of Corollary 4.1.25 also ex-
e −→ X be the universal covering of
tend to twisted coefficients: Let p : X
a path-connected CW-complex with fundamental group G and let A be a
ZG-module. Then, singular homology of X with twisted coefficients in A is
defined as 
H∗ (X; A) := H∗ C∗ (X;e Z) ⊗G A ,

e Z) with “the” deck transformation action of G.


where we equip C∗ (X;

Example 4.1.28 (surface groups). Let g ∈ N≥2 and let Γg be the surface
group of Example 4.1.20. Then we obtain from the classifying space in Ex-
ample 4.1.20 and Corollary 4.1.25 (Exercise):


Z if n=0

Z2·g if n=1
Hn (Γg ; Z) ∼
=Z


Z if n=2

0 if n>2

Together with the above presentation of Γg and Corollary 1.5.4, the deficiency
of Γg can be computed as

def Γg = 2 · g − 1.

This class of examples can also be used to show that, in general, the
transfer map is not induced by a morphism in GroupMod or GroupMod* :
There is a double sheeted covering map p : Σ3 −→ Σ2 (e.g., given by “rotation
around the middle hole of Σ3 around π”). In particular, we may view Γ3 as
subgroup of Γ2 of index 2.
We now consider the transfer map tr2 : H2 (Σ2 ; Z) −→ H2 (Σ3 ; Z) associ-
ated with this subgroup. Then
156 4. The topological view

H2 (p; Z) ◦ tr2 = 2 · idH2 (Σ2 ;Z) 6= 0.

However, if g : Σ2 −→ Σ3 is a group homomorphism, then H2 (g; Z) = 0:


Assume for a contradiction that H2 (g; Z) 6= 0. For g, there exists a contin-
uous map f : Σ2 −→ Σ3 that induces g on the level of π1 (Theorem 4.1.11,
and functoriality of the standard simplicial model). Then a Poincaré duality
argument shows that H1 (f ; Z) : H1 (Σ2 ; Z) −→ H1 (Σ3 ; Z) is surjective, which
contradicts the above computation of H1 (Σg ; Z). Therefore, H2 (g; Z) = 0.
Hence, tr2 is not induced by a group homomorphism Σ2 −→ Σ3 .
Example 4.1.29 (free products). Let G and H be groups. Then the wedge
X := BG ∨ BH yields a classifying space for the free product G ∗ H (The-
orem 4.1.13). Therefore, the Mayer-Vietoris sequence (Theorem AT.3.3.2)
shows that the inclusions BG −→ X and BH −→ X induce isomorphisms

Hn (G ∗ H; Z) ∼
=Z Hn (G; Z) ⊕ Hn (H; Z)

for all n ∈ N≥1 .


More generally, in a similar way, also amalgamated free products can be
handled [12, Chapter II.7].
Study note. One could also define group (co)homology as (co)homology of
the corresponding classifying spaces (with twisted coefficients). Try to estab-
lish as many properties of group (co)homology as possible via this approach!

Remark 4.1.30 (topological description of transfer). Let G be a group, let


H ⊂ G be a subgroup of finite index, and let n ∈ N. Then the transfer

trG
H : Hn (G; Z) −→ Hn (H; Z)

can be described topologically as follows (Figure 4.4): Let ((X, x0 ), ϕ) be


a classifying space for G and let p : (Y, y0 ) −→ (X, x0 ) be “the” path-
connected covering corresponding to the subgroup ϕ−1 (H) ⊂ π1 (X, x0 ).
Then ((Y, y0 ), ϕ ◦ π1 (p)) is a classifying space for H (Theorem 4.1.13) and
the transfer map trG H corresponds under the canonical isomorphisms from
Corollary 4.1.25 to the following homomorphism [12, Chapter III.9]:

Hn (X; Z) −→ Hn (Y ; Z)
X
k  X
k X 
aj · σj 7−→ aj · σ ,
j=1 j=1 σ∈p−1 (σj )

where p−1 (σj ) := {σ ∈ map(∆n , Y ) | p ◦ σ = σj } is the set of all p-lifts of σj ;


by covering theory, p−1 (σj ) contains exactly [G : H] elements.

Outlook 4.1.31 (bounded cohomology, topologically). The computation of


group (co)homology via classifying spaces (Corollary 4.1.25) also has a coun-
terpart in bounded cohomology: If (X, x0 ) is a sufficiently nice path-connected
4.2. Finiteness conditions 157

Y p−1 (σ)

X σ

Figure 4.4.: Transfer, topologically: lifting simplices

pointed topological space (e.g., a CW-complex), then there is a canonical iso-


morphism
 
H ∗ BHom(C∗ (X; R), R) ∼ =R H∗ π1 (X, x0 ); R ,

which is isometric with respect to the k · k∞ -semi-norm on the respective


cohomology spaces [36, 43, 44, 27].
A notable difference between this bounded setting and the ordinary setting
is that we do not need that the universal covering of X is contractible (!). The,
rough, underlying reason is that all higher homotopy groups of X are Abelian
(whence amenable) and thus do not contribute to bounded cohomology. This
fact has remarkable applications in geometric topology, e.g., in the context
of simplicial volume [36, 43, 27, 50, 75].

4.2 Finiteness conditions


When studying groups from the point of view of (co)homology, several natural
finiteness conditions/properties emerge: On the one hand, we could try to
introduce a “dimension” for groups:

• What is the maximal dimension with non-vanishing cohomology?

• How short can projective resolutions of Z over the group ring be?

• What is the minimal dimension of a classifying space?

On the other hand, we could also ask for finiteness in certain degrees:
158 4. The topological view

• Does there exist a finite classifying space? Or at least a classifying space


that has finitely many cells in many dimensions?
• Does there exist a finite projective resolution? Or at least a projective
resolution that is finitely generated in many degrees?

4.2.1 Cohomological dimension

Let us first consider the “dimension” of the group (over Z).


Definition 4.2.1 (cohomological dimension). Let G be a group. The cohomo-
logical dimension of G is defined as∗

cd G := sup n ∈ N ∃A∈Ob(ZG Mod) H n (G; A) ∼
6 Z 0 ∈ N ∪ {∞}.
=

This dimension can also be rephrased in terms of projective resolutions:


Proposition 4.2.2 (cohomological dimension, alternative descriptions). Let G
be a group. If cd G < ∞, then∗

cd G = sup n ∈ N ∃A∈Ob(ZG Mod) H n (G; A) ∼ 6 Z 0 and A is free
= À

k
= inf n ∈ N ∀A∈Ob(ZG Mod) ∀k∈N>n H (G; A) =Z 0 ∼ Á


= inf n ∈ N ∀A∈Ob(ZG Mod) H n+1 ∼
(G; A) =Z 0 Â


= inf n ∈ N there exists a projective resolution of Z over ZG

of length n . Ã

If cd G = ∞, then the descriptions Á, Â, Ã are also valid.


Proof.
• Clearly, cd G ≥ À. Conversely, let n := cd G ∈ N and let A be a
left ZG-module with H n (G; A) ∼
6 Z 0. We then consider a short exact
=
sequence
0 /K i /F π /A /0

of ZG-modules, where F is a free ZG-module (this exists; proof of


Corollary 1.6.8). Then we have the following portion of the associated
long exact cohomology sequence (for the derived functor Ext; Theo-
rem 3.1.16):
H n (idG ;π) δn
H n (G; F ) / H n (G; A) / H n+1 (G; K) ∼
=Z 0

Therefore, im H n (idG ; π) = ker δ n = H n (G; A) ∼


6 Z 0. In particular,
=
also H n (G; F ) ∼
6 Z 0.
=
• It is clear that cd G = Á (with the convention inf ∅ = ∞ and sup ∅ = 0).
∗ There is a small set-theoretic issue here; strictly speaking, the beast on the right-

hand side is not a set, and not even a class. However, the proof of the description à in
Proposition 4.2.2 can be used to show that this beast can be replaced by a set.
4.2. Finiteness conditions 159

• Moreover, Á = Â follows from dimension shifting (here, we need a coho-


mological version that is obtained by embedding modules into injective
modules [12, Chapters III.4, III.7]).

• Clearly, cd G ≤ Ã. For the converse inequality, we proceed as follows:


Let n := cd G; without loss of generality, we may assume n < ∞. Let
(P∗ , ε) be a projective resolution of Z over ZG. It then suffices to show
that K := ker ∂n ⊂ Pn is a projective ZG-module (then we can truncate
the original projective resolution by taking K in degree n instead of Pn ).
To this end, we show that K is a direct summand of Pn (which estab-
lishes that also K is projective).

∂n+2 ∂n+1 ∂n
Pn+2 / Pn+1 / Pn
O
/ Pn−1
p
∂n+1 
"
K

By assumption,

0∼
=Z H n+1 (G; K) (definition of n)


=Z H n+1 HomG (P∗ , K) (Corollary 1.6.9)
ker HomG (∂n+2 , K)
= .
im HomG (∂n+1 , K)

In particular, we can apply this to the cocycle ∂n+1 ∈ HomG (Pn+1 , K);
thus, there exists a ZG-homomorphism p : Pn −→ K with ∂n+1 =
p ◦ ∂n+1 . Because, im ∂n+1 = ker ∂n = K, we have p|K = idK , and so K
is a direct summand of Pn .

The cohomological dimension is also related to the geometric dimension,


defined in terms of classifying spaces.

Definition 4.2.3 (geometric dimension). Let G be a group. Then the geometric


dimension of G is (where the dimension of classifying spaces is the dimension
in the sense of CW-complexes)

gd G := inf dim X X is a classifying space for G ∈ N ∪ {∞}.

Proposition 4.2.4 (cohomological vs. geometric dimension). Let G be a group.


Then
cd G ≤ gd G.

Proof. We may assume that n := gd G is finite. Hence, there exists a


classifying space X for G that has dimension n (as CW-complex). Then
e Z), εCW ) is a projective resolution of Z over ZG (Theorem 4.1.24);
(C∗CW (X;
because, dim X = n, we have
160 4. The topological view

e Z) ∼
CkCW (X; =ZG 0

for all k ∈ N>n . Therefore, cd G ≤ n = gd G.


Proposition 4.2.5 (dimension, inheritance properties). Let G be a group.
1. If H ⊂ G is a subgroup, then cd H ≤ cd G and gd H ≤ gd G.
2. If H ⊂ G is a subgroup of finite index and cd G < ∞, then cd H = cd G.
3. If 1 −→ N −→ G −→ Q −→ 1 is an extension of groups, then

cd G ≤ cd N + cd Q.

4. If H is another group, then



gd(G ∗ H) = max gd G, gd H .

Proof. Ad 1. The claim about cohomological dimension follows, e.g., from


the Shapiro lemma (Theorem 1.7.8; Exercise). The claim about geometric
dimension follows from covering theory (Theorem 4.1.13).
Ad 2. By the first part, cd H ≤ cd G. For the converse inequality, we use
the characterisation of cohomological dimension via free coefficient modules
(Proposition 4.2.2): Let n := cd G ∈ N. Then there exists a free ZG-module A
with H n (G; A) 6∼
=Z 0. Because A is free and [G : H] < ∞, there is a free ZH-
module B with
A∼=ZG IndG ∼
H B =ZG CoindH B
G

(the latter isomorphism follows from Proposition 1.7.7). Therefore, by the


Shapiro lemma (Theorem 1.7.8), we have

H n (H; B) ∼
=Z H n (G; CoindG ∼ n ∼
H B) =Z H (G; A) 6=Z 0,

and so cd H ≥ n = cd G.
Ad 3. This is a standard spectral sequence argument (Exercise): We may
assume that n := cd N and m := cd Q are finite. Let A be a left ZG-module.
Then the Hochschild-Serre spectral sequence (Theorem 3.2.12) gives a con-
verging cohomological spectral sequence

E2pq = H p Q; H q (N ; ResG
N A) =⇒ H
p+q
(G; A).

Let k ∈ N>m+n and let p, q ∈ N with p + q = k. Then E2pq ∼ =Z 0. Therefore,


pq ∼
also E∞ =Z 0; i.e., all terms on the k-“diagonal” are trivial. Convergence
thus implies that H k (G; A) ∼
=Z 0. Hence, cd G ≤ n + m.
Ad 4. This follows from the fact that we can take a wedge of classifying
spaces for G and H as classifying space for G ∗ H (Theorem 4.1.13).
We also have cd(G ∗ H) = max cd G, cd H [12, Proposition VIII.2.4]; this
can, for example, be shown by arguing as in Example 4.1.29 via a Mayer-
Vietoris sequence with twisted coefficients.
4.2. Finiteness conditions 161

Example 4.2.6 (cohomological/geometric dimension).

• If G is a non-trivial finite group, then cd G = gd G = ∞ (Proposi-


tion 4.2.4 and Corollary 1.7.3).

• A group G satisfies cd G = 0 if and only if G is trivial. Clearly, the


trivial group has cohomological dimension 0. Conversely, Let cd G = 0
and let C be a cyclic subgrup of G. Then

cd C ≤ cd G = 0

and so C is trivial (by the previous example and the computation


of H 1 (Z; Z)).

• If G is a free group, then by Proposition 4.2.4 and using wedges of


circles as a classifying space (Example 4.1.19), we obtain

cd G ≤ gd G ≤ 1

Conversely, a group G satisfies cd G ≤ 1 if and only if G is free [84].

• For all n ∈ N, we have

cd Zn = gd Zn = n :

On the one hand, cd Zn ≤ gd Zn ≤ n (by Proposition 4.2.4 and taking


the n-torus as classifying space); on the other hand, cd Zn ≥ n (by the
Künneth theorem; Corollary 3.2.23).

• For all g ∈ N≥2 , the surface group Γg satisfies

cd Γg = gd Γg = 2 :

On the one hand, cd Γg ≤ gd Γg ≤ dim Σg = 2; on the other hand,


cd Γg ≥ 2 (by the computation in Example 4.1.28).
In particular, Γg does not contain a subgroup that is isomorphic to Z3 ;
in fact, geometric methods also allow to show that Γg does not even
contain Z2 [37][53, Corollary 7.5.15].

• The Heisenberg group H ⊂ SL3 (Z) satisfies cd H = gd H = 3: On the


one hand, dim H ≤ gd H ≤ 3 (because of the Heisenberg manifold;
Exercise); on the other hand, H3 (H; Z) ∼
=Z Z 6∼=Z 0 (Exercise) and
so cd H ≥ 3.

Corollary 4.2.7. Let G be a group that contains a non-trivial torsion element.


Then
gd G = cd G = ∞.
162 4. The topological view

Proof. Because G contains a non-trivial torsion element g, there is also a


non-trivial finite cyclic subgroup hgiG of G. Hence,

∞ = cdhgiG (Example 4.2.6)


≤ cd G (Proposition 4.2.5)
≤ gd G, (Proposition 4.2.4)

as claimed.
Example 4.2.8. By Corollary 4.2.7, the groups D∞ and SL2 (Z) have infinite
cohomological/geometric dimension (because they contain torsion elements),
but they contain free subgroups of finite index.
Usually, the contraposition is applied: If a group admits a finite-dimensional
classifying space, then this group must be torsion-free:
Example 4.2.9 (fundamental groups of manifolds of non-positive curvature). If
G is the fundamental group of a closed smooth manifold M that admits a Rie-
mannian of non-positive sectional curvature, then M is a finite-dimensional
classifying space for G (Outlook 4.1.21). In particular, G is torsion-free
(Corollary 4.2.7).

Caveat 4.2.10 (Eilenberg-Ganea problem). Let G be a group and let n := cd G.


If n 6= 2, then it is known that also gd G = n [12, Theorem VIII.7.1]. More-
over, if cd G = 2, then one at least has gd G ≤ 3 [12, Theorem VIII.7.1].
However, it is still an open problem if every group of cohomological dimen-
sion 2 has geometric dimension 2.
Outlook 4.2.11 (lattices). An important class of finitely generated groups
are lattices (i.e., discrete subgroups with finite covolume) of Lie groups. The
cohomological and geometric dimension can be determined in terms of the
Borel-Serre compactification of the associated symmetric space [12, Chap-
ter VIII.9].

4.2.2 Finite type


We briefly indicate degree-wise finiteness types, both geometrically and alge-
braically:
Definition 4.2.12 (type F, Fn , F∞ ). Let G be a group and let n ∈ N.
• The group G is of type F if there exists a classifying space for G that
consists of only finitely many cells (whence, finite-dimensional).
• The group G is of type Fn if there exists a classifying space for G
(possibly of infinite dimension) whose n-skeleton is finite.
4.2. Finiteness conditions 163

• The group G is of type F∞ if there exists a classifying space for G that,


in each dimension, contains only finitely many open cells.

Remark 4.2.13 (relation with classical finiteness conditions). Let G be a group.

• The concrete description of the fundamental group of CW-complexes in


terms of their 2-skeleta [83, Chapter 4.1.6f] and the inductive construc-
tion of classifying spaces from presentation complexes (Remark 4.1.12)
shows that:
– Every group G is of type F0 .
– The group G is finitely generated if and only if it is of type F1 .
– The group G is finitely presented if and only if it is of type F2 .

• If G is of type F, then clearly cd G ≤ gd G < ∞ and G is of type F∞ .

• The converse does not hold, in general: If G is finite (and non-trivial),


then G is of type F∞ (e.g., the standard simplicial model works), but
not of type F (because cd G = ∞; Corollary 4.2.7).
Moreover, the free group F (N) satisfies cd F (N) = gd F (N) = 1 < ∞,
but F (N) is not finitely generated (check!) and thus not of type F1 .

• There exist finitely presented groups that are not of type F3 [81].

Definition 4.2.14 (type FP, FPn , FP∞ [6]). Let G be a group and let n ∈ N.

• The group G is of type FP if there exists a finite-length projective


resolution of Z over ZG that consists of finitely generated ZG-modules.

• The group G is of type FPn if there exists a projective resolution (P∗ , ε)


of Z over ZG such that P0 , . . . , Pn are finitely generated ZG-modules.

• The group G is of type FP∞ if there exists a projective resolution (P∗ , ε)


of Z over ZG such that Pn is finitely generated for each n ∈ N.

Remark 4.2.15 (relation between F and FP). Let G be a group and n ∈ N. In


view of Theorem 4.1.24, we have:

• If G has type F, then it has type FP.

• If G has type Fn , then it has type FPn .

• If G has type F∞ , then it has type FP∞ .

The converse implications are more delicate: There exist groups of type FP
that are not of type F2 [5]. However, for finitely presented groups, type FPn
coincides with Fn for all n ∈ N≥3 , by work of Eilenberg-Ganea and Wall [12,
Chapter VIII.7].
164 4. The topological view

Outlook 4.2.16 (Euler characteristic). If a group G is of type F or FP, then


one can, for instance, consider its Euler characteristic
X
χ(G) := (−1)n · dimQ Hn (G; Q).
n∈N

The Euler characteristic can be used to show that certain groups are not
commensurable (Exercise).
Using the theory of von Neumann dimensions, one sees that the Euler char-
acteristic can also be computed via (co)homology with `2 -coefficients [1][55,
Theorem 1.35]. Therefore, the invariance properties of these coefficients also
carry over to the Euler characteristic; for example:

• Amenable groups of type F have Euler characteristic 0 [16][55, Corol-


lary 6.75].

• (Fundamental groups of) Oriented closed connected connected hyper-


bolic manifolds of even dimension have non-zero Euler characteris-
tic [22][55, Theorem 1.62].

• The sign of the Euler characteristic is an orbit equivalence invariant for


groups of type F [29].

Moreover, the (virtual) Euler characteristic of (arithmetic) lattices is also


related to their arithmetic properties [38, 76][12, Chapter IX].

4.3 Application: Free actions on spheres


As conclusion of this course, we use the theory of group (co)homology to
approach a classical problem in geometric topology, the space form problem:
Which manifolds have a sphere as universal covering? In terms of group
actions, one can reformulate a part of this problem as follows:

Questions 4.3.1. Which finite groups can act freely on spheres?

There are some obvious positive examples (but it seems hard to exclude
groups by elementary means):

Example 4.3.2 (free actions on spheres).

• Let n ∈ N>0 . Then the finite cyclic group Z/n acts freely on the circle S 1
(the one-dimensional sphere) by rotation around 2 · π/n.

• The generalised quaternion groups (Outlook 1.6.20) act freely on the


unit sphere of the quaternion algebra (which is S 3 ) by multiplication
in the quaternion algebra.
4.3. Application: Free actions on spheres 165

Theorem 4.3.3 (free actions on spheres [78, 59]). Let G be a finite group that
admits a free continuous action on a sphere. Then:
• If p ∈ N is an odd prime, then every p-Sylow subgroup of G is cyclic.
• The group G contains at most one element of order 2 and the 2-Sylow
subgroups of G are cyclic or isomorphic to a generalised quaternion
group.
The theorem does not mention group homology, but the proof (going
back to work by Smith and Milnor) will take advantage of group homology.
More precisely, group actions on spheres lead to periodic resolutions (Chap-
ter 4.3.1); we can then use our knowledge on homology of Abelian groups
and the recognition of cyclic groups to derive the theorem (Chapter 4.3.2).
Before giving the proof, we illustrate the power of the theorem in simple
examples.
Example 4.3.4. The group Z/2019×Z/2019 does not admit a free action on a
sphere: We have 2019 = 3·673 and the 3-Sylow subgroups of Z/2019×Z/2019
are isomorphic to Z/3 × Z/3, which is not cyclic. Therefore, Theorem 4.3.3
implies that Z/2019 × Z/2019 does not admit a free action on a sphere.
Example 4.3.5 (symmetric groups acting on spheres? [59, Corollary 2]). Let
n ∈ N≥3 . Then the symmetric group Sn does not admit a free action on a
sphere because Sn then contains at least two elements of order 2 (e.g., two
different transpositions), which is excluded by Theorem 4.3.3.
Example 4.3.6. Let G be a non-trivial finite group. Then G × G does not
admit a free action on a sphere. This can be deduced from the Sylow theorems
(Theorem III.1.3.35) and Theorem 4.3.3 (check!).

4.3.1 From sphere actions to periodic resolutions

Theorem 4.3.7 (from sphere actions to periodic resolutions). Let n ∈ N be odd


and let G y S n be a free continuous action of a finite group G on S n . Then
there exists a projective resolution (P∗ , ε) of Z over ZG that is (n+1)-periodic,
i.e., for all k ∈ N, ` ∈ N>0 , we have

Pk+n+1 = Pk and ∂`+n+1 = ∂` .

The basic idea is to consider a cellular ZG-chain complex of S n and to


use the fact that H0 (S n ; Z) ∼ =Z Z ∼ =Z Hn (S n ; Z) to splice together infinitely
many copies of the cellular chain complex, which then results in a periodic
projective resolution. However, for this to work, we need to know that the G-
action on Hn (S n ; Z) is trivial; this problem is solved by the Lefschetz number,
an “Euler characteristic” for self-maps (replacing dimensions by traces), and
the Lefschetz fixed point theorem:
166 4. The topological view

Definition 4.3.8 (Lefschetz number). Let X be a finite CW-complex and let


f : X −→ X be a continuous map. Then the Lefschetz number of f is
X 
Λ(f ) := (−1)n · trZ Hn (f ; Z) : Hn (X; Z) −→ Hn (X; Z) ∈ Z.
n∈N

Here, trZ denotes the trace on the free part of the corresponding finitely
generated Z-module.
Example 4.3.9 (Euler characteristic as Lefschetz number). If X is a finite CW-
complex, then
X X
Λ(idX ) = (−1)n · trZ Hn (idX ; Z) = (−1)n · rkZ Hn (X; Z)
n∈N n∈N
= χ(X).

Theorem 4.3.10 (Lefschetz fixed point theorem [23, Chapter VII.6]). Let X be
a finite CW-complex and let f : X −→ X be a continuous map. If f has no
fixed points, then Λ(f ) = 0.
Corollary 4.3.11 (action on top homology). Let n ∈ N and let G y S n be a
free continuous action of a group G on S n .
=Group 1 or G ∼
1. If n is even, then G ∼ =Group Z/2.
2. If n is odd, then the induced G-action on Hn (S n ; Z) is trivial.
Proof. If n = 0, then |S n | = 2 and so |G| ≤ 2. This implies that G is trivial
or isomorphic to Z/2. Therefore, in the following, we only need to consider
the case that n > 0.
Let g ∈ G be a non-trivial element and let fg : S n −→ S n be the cor-
responding homeomorphism; because the action is free, fg does not have a
fixed point. As the homology of S n is concentrated in the (distinct) degrees 0
and n, the Lefschetz fixed point theorem (Theorem 4.3.10) shows that

0 = Λ(fg )
= trZ H0 (fg ; Z) + (−1)n · trZ Hn (fg ; Z)
= 1 + (−1)n · deg fg .

We now distinguish two cases:


• If n is odd, then deg fg = 1. In particular, g acts trivially on Hn (S n ; Z).
• If n is even, then deg fg = −1. Therefore, all non-trivial elements of G
act by multiplication by −1 on Hn (S n ; Z). The multiplicativity of the
degree shows that, for all h ∈ G \ {e}, we have

deg fg·h = deg(fg ◦ fh ) = deg fg · deg fh = (−1) · (−1) = 1,

and thus g · h = e. In particular, G ∼


=Group 1 or G ∼
=Group Z/2.
4.3. Application: Free actions on spheres 167

Proof of Theorem 4.3.7. There exists a G-equivariant finite n-dimensional


CW-structure on S n such that G freely permutes the open cells (the quotient
space G \ S n is an odd(!)-dimensional closed topological n-manifold, whence
homeomorphic to an n-dimensional CW-complex [60][47, p. 107][68]). Let C∗
be the associated cellular chain complex. Because G acts freely on the open
cells, C∗ is a free ZG-chain complex that is concentrated in degrees 0, . . . , n.
Let

η: Z ∼
=Z Hn (S n ; Z) ∼
=Z ker ∂n −→ Cn
ε : C0 −→ C0 / im ∂1 =Z H0 (S n ; Z) ∼
∼ =Z Z

be the inclusion and projection, respectively. Then the doubly augmented


sequence

η ∂n ∂2 ∂1 ε
0 /Z / Cn / ··· / C1 / C0 /Z /0

is exact (the homology of S n , whence of C∗ , in degrees 1, . . . , n − 1 is trivial).


The G-action on H0 (S n ; Z) ∼ =Z Z is trivial (because continuous self-maps
of path-connected spaces induce the identity on H0 ( · ; Z)); moreover, also
the G-action on Hn (S n ; Z) ∼
=Z Z is trivial (Corollary 4.3.11). Therefore, the
spliced sequence

∂2 ∂1 η◦ε ∂n ∂2 ∂1 ε
··· / C1 / C0 / Cn / ··· / C1 / C0 /Z /0

is a free ZG-resolution of Z, which is periodic of period n + 1.

4.3.2 From periodic resolutions to Sylow subgroups

We can now complete the proof of Theorem 4.3.3 on finite groups that admit
free actions on spheres:

Proof of Theorem 4.3.3. Let n ∈ N and let G y S n be a free continuous


action. In view of Corollary 4.3.11, we may assume without loss of generality
that n is odd (as the trivial group and Z/2 clearly satisfy the conclusion of
the theorem).
Ad 1. Let p ∈ N be an odd prime and let S ⊂ G be a p-Sylow subgroup
of G. In order to show that S is cyclic, it suffices to prove that all Abelian
subgroups of S are cyclic (Corollary 1.6.19).
Let H ⊂ S be an Abelian subgroup. Assume for a contradiction that H
is not cyclic. As H is a p-group, this implies that H contains a subgroup H
that is isomorphic to Z/p × Z/p (check!). As subgroup of G also H admits
a free action on S n . Therefore, there exists a periodic projective resolution
of Z over ZH (Theorem 4.3.7); in particular, the homology H∗ (H; Fp ) has to
168 4. The topological view

be periodic. However, this contradicts the Künneth computation in Exam-


ple 3.2.25. Therefore, all Abelian subgroups of the p-Sylow group S are cyclic
and so S is cyclic.
Ad 2. Similarly, also the prime 2 can be handled. We first show that G
contains at most one element of order 2. As additional ingredient, we need
Milnor’s generalisation of the Borsuk-Ulam theorem [59, Theorem 1]:
Let T : S n −→ S n be a continuous map without fixed points that sat-
isfies T ◦ T = idS n . Then, for every continuous map f : S n −→ S n of
odd degree, there exists an x ∈ S n with

T ◦ f (x) = f ◦ T (x).

Let f, g : S n −→ S n be the homeomorphisms corresponding to elements of


order 2 in G. Then f ◦ f = idS n = g ◦ g and deg f, deg g ∈ {−1, 1}; in
particular, f and g have odd degree. By the above theorem, f ◦ g and g ◦ f
coincide at one point; because the action is free, we thus have f ◦ g = g ◦ f .
Therefore, all elements of order 2 in G commute.
As in the argument for the case of odd primes, we know that G does
not contain a subgroup isomorphic to Z/2 × Z/2. Therefore, the elements
inducing f and g must be equal, i.e., G contains at most one element of
order 2.
This implies that the 2-Sylow subgroups of G are cyclic or a generalised
quaternion group (Outlook 1.6.20).
Further information on the space form problem and on groups with peri-
odic cohomology can be found in the literature [19][12, Chapter VI.9].
More generally, investigating group actions via group (co)homology is a
recurring theme in topology, geometry, and the theory of arithmetic groups.
A
Appendix

Overview of this chapter.

A.1 Amalgamated free products A.3


A.2 Some homological algebra A.7
A.3 Homotopy theory of CW-complexes A.11
A.2 A. Appendix
A.1. Amalgamated free products A.3

A.1 Amalgamated free products


We will now briefly review some concepts from group theory that allow to
construct coproducts and pushouts of groups explicitly.

A.1.1 The free group of rank 2


We start with an explicit description of the free group of rank 2, using reduced
words [53, Chapter 3.3, Chapter 2.2]. Roughly speaking, this group is the
group generated by two different elements with the least possible relations
between these elements.

a, bb four distinct elements.


Definition A.1.1 (group of reduced words). Let a, b, b
Let W be the set of words (i.e., finite sequences) over S := {a, b, b a, bb}.

• Let n ∈ N and let x1 , . . . , xn ∈ S. The word x1 . . . xn ∈ W is reduced if

xj+1 6= xbj [
and x j+1 6= xj

holds for all j ∈ {1, . . . , n − 1}. In particular, the empty word ε is


reduced.

• We write F (a, b) for the set of all reduced words over S.

• On F (a, b), we define a composition by concatenation and reduction:

· : F (a, b) × F (a, b) −→ F (a, b)


(x1 . . . xn , xn+1 . . . xm ) 7−→ x1 . . . xn−r xn+1+r . . . xn+m .

Here,

r := max k ∈ {0, . . . , min(n, m − 1)} ∀j∈{0,...,k−1} xn−j = x\
n+1+j

∨ x[
n−j = xn+1+j .

Example A.1.2. In the situation of the previous definition, the word abb abb is
reduced; the word babab is not reduced. The elements a and b a are inverse to
each other with respect to “·”; analogously, also b and bb are inverse to each
a and bb, respectively.
other. Hence, one usually writes a−1 and b−1 instead of b

Proposition A.1.3 (free group of rank 2).

1. The set F (a, b) is a group with respect to the composition specified in


the previous definition.
A.4 A. Appendix

2. The set {a, b} is a free generating set of F (a, b), i.e., the following
universal property is satisfied:
For every group H and every map f : {a, b} −→ H, there exists a unique
group homomorphism f : F (a, b) −→ H with f |{a,b} = f .

3. In other words,
1 /Z

17→b
 
Z / F (a, b)
17→a

is a pushout in Group.

Proof. The first part follows from a straightforward computation (associativ-


ity is not obvious!) [53, Chapter 3.3].
The second part (and the third part) can be verified directly by hand
(check!).

A.1.2 Free products of groups


More generally, we can also consider the free product of a family of groups.
Again, we are looking for a group generated by the given groups with as few
relations between them as possible.

Definition
F A.1.4 (free product of groups). Let (Gi )i∈I be a family of groups;
for g ∈ i∈I (Gi \ {1}) let i(g) ∈ I be the unique index with g ∈ Gi(g) .

• A finite (possibly
F empty) sequence (s1 , . . . , sn ) with n ∈ N of non-trivial
elements of i∈I Gi is a reduced word (over the family (Gi )i∈I ), if

∀j∈{1,...,n−1} i(sj ) 6= i(sj+1 ).

• We write Fi∈I Gi for the set of all reduced words over the fam-
ily (Gi )i∈I .

• On Fi∈I Gi , we define a composition by concatenation/reduction:

· : Fi∈I Gi × Fi∈I Gi −→ Fi∈I Gi


(
 (s1 , . . . , sn−k(s,t) , tk(s,t)+1 , . . . , tm ) À
s = (s1 , . . . , sn ), t = (t1 , . . . , tm ) 7−→
(s1 , . . . , sn−k(s,t) · tk(s,t)+1 , . . . , tm ) Á

Here, k(s, t) ∈ {0, . . . , min(n, m)} is the biggest k ∈ {0, . . . , min(n, m)}
satisfying

∀j∈{1,...,k} i(sn−j+1 ) = i(tj ) ∧ sn−j+1 = t−1


j .
A.1. Amalgamated free products A.5

Case À occurs if i(sn−k(s,t) ) 6= i(tk(s,t)+1 ); case Á occurs if i(sn−k(s,t) ) =


i(tk(s,t)+1 ).

• We call Fi∈I , together with this composition, the free product of (Gi )i∈I .

The free product G := Fi∈I Gi of a family (Gi )i∈I indeed is a group (again,
associativity is non-trivial!) and the canonical inclusions Gi −→ G are group
homomorphisms.
Free products are an explicit model of coproducts of groups:

Proposition A.1.5 (coproduct of groups). Let (Gi )i∈I be a family of groups.


Then Fi∈I Gi , together with the canonical inclusions (Gi −→ Fj∈I Gj )i∈I , is
the coproduct of the family (Gi )i∈I in the category Group.

Proof. This can be shown by verifying the universal property (check!).

A.1.3 Amalgamated free products of groups


“Glueing” groups along another group leads to the amalgamated free product:

Definition A.1.6 (amalgamated free product). Let G0 , G1 , and G2 be groups


and let i1 : G0 −→ G1 as well as i2 : G0 −→ G2 be group homomorphisms.
The associated amalgamated free product of G1 and G2 over G0 is defined by

G1 ∗G0 G2 := (G1 ∗ G2 )/N,

where N ⊂ G1 ∗G2 is the smallest (with respect to inclusion) normal subgroup


of G1 ∗ G2 that contains the set {i1 (g) · i2 (g)−1 | g ∈ G0 }.

Proposition A.1.7 (pushouts of groups). Let G0 , G1 , and G2 be groups and


let i1 : G0 −→ G1 as well as i2 : G0 −→ G2 be group homomorphisms. Let
j1 : G1 −→ G1 ∗G0 G2 and j2 : G2 −→ G1 ∗G0 G2 be the homomorphisms
induced by the canonical inclusions G1 −→ G1 ∗ G2 and G2 −→ G1 ∗ G2 ,
respectively. Then

i1
G0 / G1

i2 j1
 
G2 / G1 ∗G0 G2
j2

is a pushout in Group.

Proof. This can be shown by the same argument as in the construction of


the pushout of topological spaces (Proposition AT.1.1.14), using the universal
property of the free product and of quotient groups.
A.6 A. Appendix

A.1.4 Free groups

A related generalisation of F (a, b) are general free groups; the universal prop-
erty of free groups/free generating sets is a group-theoretic version of the
universal property of bases of vector spaces.

Definition A.1.8 (free generating set, free group, rank of a free group).
• Let G be a group. A subset S ⊂ G is a free generating set of G if the
following universal property is satisfied: The group G is generated by S
and for every group H and every map f : S −→ H there exists a unique
group homomorphism f : G −→ H with f |S = f .
• A free group is a group that contains a free generating set; the cardi-
nality of such a free generating set is the rank of the free group.
Caveat A.1.9. Not every group has a free generating set! For example, the
groups Z/2 and Z2 are not free (check!).

Comparing the corresponding universal properties establishes existence of


free groups of arbitrary rank:
Proposition A.1.10 (existence of free groups). Let S be a set, let G := FS Z
be the associated free product and for every s ∈ S let is : Z −→ G be the
inclusion of the s-th summand. Then {is (1) | s ∈ S} is a free generating set
of G.
Proof. We can translate the universal property of coproducts into the uni-
versal property of free generating sets (because the building blocks are the
groups Z, which are free of rank 1) (check!).

Proposition A.1.11 (invariance of rank of free groups). Let G and G0 be free


groups with free generating sets S and S 0 , respectively. Then G and G0 are
isomorphic if and only if |S| = |S 0 |.
Proof. This can be shown, for example, by looking at homomorphisms to Z/2
and a cardinality argument [53, Exercise 2.E.12].
A.2. Some homological algebra A.7

A.2 Some homological algebra


In this appendix, we recall two basic results from homological algebra: The
long exact homology sequence and the horseshoe lemma. For simplicity, we
only consider homological algebra in module categories (instead of general
Abelian categories); in view of the Freyd-Mitchell embedding theorem, this
is not a substantial limitation.

Setup A.2.1. In the following, R will always be a (not necessarily commuta-


tive) ring with unit.

Proposition A.2.2 (algebraic long exact homology sequence). Let

i p
0 / A∗ / B∗ / C∗ /0

be a short exact sequence in R Ch (i.e., in every degree, the corresponding


sequence in R Ch is exact). Then there is a (natural) long exact sequence

∂k+1 Hk (i) Hk (p) ∂k


··· / Hk (A∗ ) / Hk (B∗ ) / Hk (C∗ ) / Hk−1 (A∗ ) / ···

This sequence is natural in the following sense: If

i∗ p∗
0 / A∗ / B∗ / C∗ /0

f∗ g∗ h∗
  
0 / A0∗ / B∗0 / C∗0 /0
i0∗ p0∗

is a commutative diagram in R Ch with exact rows, then the corresponding


ladder
∂k+1
··· / Hk (A∗ ) Hk (i∗ )/ Hk (B∗ ) Hk (p∗ )/ Hk (C∗ ) ∂k
/ Hk−1 (A∗ ) / ···

Hk (f∗ ) Hk (g∗ ) Hk (h∗ ) Hk−1 (fA )


   
··· / Hk (A0∗ ) / Hk (B∗0 ) / Hk (C∗0 ) / Hk−1 (A0∗ ) / ···
∂k+1 Hk (i0∗ ) Hk (p0∗ ) ∂k

is commutative and has exact rows.

Proof. Let k ∈ Z. We construct the connecting homomorphism

∂k : Hk (C∗ ) −→ Hk−1 (A∗ )


A.8 A. Appendix

as follows: Let γ ∈ Hk (C∗ ); let c ∈ Ck be a cycle representing γ. Because


pk : Bk −→ Ck is surjective, there is a b ∈ Bk with

pk (b) = c.

As p∗ is a chain map, we obtain pk−1 ◦ ∂kB (b) = ∂kC ◦ pk (b) = ∂kC (c) = 0; then
exactness in degree k shows that there exists an a ∈ Ak−1 with

ik−1 (a) = ∂kB (b).

In this situation, we call (a, b, c) a compatible triple for γ and we define

∂k (γ) := [a] ∈ Hk−1 (A∗ ).

Straightforward diagram chases then show (check!):

• If (a, b, c) is a compatible triple for γ, then a ∈ Ak−1 is a cycle (and so


indeed defines a class in Hk−1 (A∗ )).

• If (a, b, c) and (a0 , b0 , c0 ) are compatible triples for γ, then [a] = [a0 ]
in Hk−1 (A∗ ).

These observations show that ∂k is an R-homomorphism and that ∂k is nat-


ural (check!).
Further diagram chases then show that the resulting long sequence is exact
(even more to check . . . ).

Proposition A.2.3 (horseshoe lemma). Let

f0 f 00
0 / M0 /M / M 00 /0

be a short exact sequence in R Mod and let (P∗0 , ε0 ) and (P∗00 , ε00 ) be projective
R-resolutions of M 0 and M 00 , respectively:

.. ..
. .

 
P00 ? P000
ε0 ε00
 f 0
f 00 
0 / M0 /M / M 00 /0

 
0 0
A.2. Some homological algebra A.9

There there exists a projective R-resolution (P∗ , ε) of M and R-chain maps fe∗0 
f 0 : P∗0  ε0 −→ P∗  ε and fe∗00  f 00 : P∗  ε −→ P∗00  ε00 such that

fen
0
fen
00

0 / Pn0 / Pn / Pn00 /0

is exact in every degree n ∈ N.


Proof. On the level of modules, we define (P∗ , ε) as direct sum of the outer
resolutions and choose fe∗0 and fe∗00 as the corresponding inclusion and projec-
tion, respectively. The boundary operators are constructed as follows:
As P000 is projective and f 00 : M −→ M 00 is surjective, there exists an R-
module homomorphism εe00 : P000 −→ M with

f 00 ◦ εe00 = ε00 .

We set
ε := (f 0 ◦ ε0 ) ⊕ εe00 : P0 = P00 ⊕ P000 −→ M.
Inductively, we construct boundary operators ∂n+1 : Pn+1 −→ Pn that are
compatible with the boundary operators on P∗0 and P∗00 :
Let n ∈ N>0 and let us suppose, by induction, that ∂n is already con-
structed. A diagram chase shows that

fn00 (ker ∂n ) = ker ∂n00 = im ∂n+1


00

00
(check!). Because the module Pn+1 is projective, there exists an R-homo-
e00 00
morphism ∂n+1 : Pn+1 −→ ker ∂n that satisfies

f 00 ◦ ∂en+1
00 00
= ∂n+1 .

We then set
∂n+1 := (f 0 ◦ ∂n+1
0
) ⊕ ∂en+1
00
;
a diagram chase yields ∂n ◦ ∂n+1 = 0 (check!).
In this way, we obtain a short exact sequence

fe0 f 0 fe00 f 00
0 / P 0  ε0 / P ε / P 00  ε00 /0

in R Ch (indexed over N ∪ {−1}). Applying the long exact homology sequence


(Proposition A.2.2) shows that (P∗ , ε) is exact. Hence, (P∗ , ε) is a projective
resolution of M with the desired properties.
A.10 A. Appendix
A.3. Homotopy theory of CW-complexes A.11

A.3 Homotopy theory of CW-complexes


Let us review some terminology for CW-complexes and their basic homotopy-
theoretic properties. Roughly speaking, a CW-complex is a topological space
that is built from inductively attaching disks of increasing dimension to a set
of points.

Definition A.3.1 ((relative) CW-complex).

• Let (X, A) be a pair of spaces. A relative CW-structure on (X, A) is a


sequence
A =: X−1 ⊂ X0 ⊂ X1 ⊂ · · · ⊂ X
of subspaces of X with the following properties:
S
– We have X = n∈N Xn .
– The topology on X coincides with the colimit topology of the
system A = X−1 ⊂ X0 ⊂ X1 ⊂ . . . ; i.e., a subset U ⊂ X is open
if and only if for every n ∈ N ∪ {−1}, the intersection U ∩ Xn is
open in Xn .
– For every n ∈ N, the space Xn is obtained from Xn−1 by attaching
n-dimensional cells, i.e., there exists a set In and a pushout of the
form F n−1
In S
/ Xn−1

inclusion inclusion
F  
In Dn / Xn

in Top; here, we use the convention S −1 := ∅. Then, Xn is the


n-skeleton of X. The number |In | equals the number of path-
connected components of Xn \ Xn−1 , but the choice of pushouts
is not part of the data!

• A relative CW-complex is a pair (X, A) of spaces together with a


relative CW-structure on (X, A). If A = ∅, then X, together with
this CW-structure, is a CW-complex. Usually, we will leave the fi-
bration of the CW-structure implicit and say things like “a relative
CW-complex (X, A)” if the underlying CW-structure is clear from the
context or irrelevant.

• If (X, A) is a relative CW-complex and n ∈ N, then the path-connected


components of Xn \ Xn−1 are homeomorphic to Dn◦ (check!) and are
called open n-cells of (X, A).
A.12 A. Appendix

• If (X, A) is a relative CW-complex,


 then the dimension of (X, A) is
defined as dim(X, A) := min n ∈ N ∀m∈N≥n Xm = Xn ∈ N ∪ {∞}.

• A (relative) CW-complex is finite, if it consists of finitely many open


cells. A (relative) CW-complex is of finite type, if in each dimension, it
has only finitely many open cells.

The strange prefix “CW” refers to the “closure finiteness” condition on


cells (which can be derived from the definition above) and the “weak topol-
ogy” (i.e., the colimit topology).

Example A.3.2 (CW-structures). Examples of CW-structures on the circle, on


the sphere S 2 , on the torus S 1 × S 1 , and on RP 2 are indicated in Figure A.1.

Caveat A.3.3 (products of CW-complexes). Let X and Y be CW-complexes


and let (Zn )n∈N∪{−1} be given by Z−1 := ∅ and
[
∀n∈N Zn := Xk × Yn−k .
k∈{0,...,n}

Then, in general, (Zn )n∈N∪{−1} is no CW-structure on X × Y [24] (and


the question of when this happens is rather delicate [11]). Therefore, when
working with products of (infinite) CW-complexes, it is sometimes convenient
to pass to the category of compactly generated spaces [82].

For CW-complexes, homotopy equivalences can be characterised in the


following way:

Theorem A.3.4 (Whitehead theorem). Let X and Y be CW-complexes and


let f : X −→ Y be a continuous map. Then the following are equivalent:

1. The map f : X −→ Y is a homotopy equivalence (in Top).

2. The map f : X −→ Y is a weak equivalence, i.e., for every x0 ∈ X


and every n ∈ N the induced map πn (f ) : πn (X, x0 ) −→ πn (Y, f (x0 )) is
bijective.

3. For every CW-complex Z, the map

[Z, f ] : [Z, X] −→ [Z, Y ]


[g] 7−→ [f ◦ g]

bijective.

Sketch of proof. Ad 1 =⇒ 2. Let n ∈ N. Then, by construction, πn : Top* −→


Set is a homotopy invariant functor. Moreover, one can show that πn
also translates unpointed homotopy equivalences to bijections [85, Propo-
sition 6.2.4].
A.3. Homotopy theory of CW-complexes A.13

0-skeleton 1-skeleton 2-skeleton total space

S1

one 0-cell one 1-cell no 2-cells

S1

two 0-cells two 1-cells no 2-cells


(North-/South-arc)

S2

oe 0-cell no 1-cells one 2-cell

S2

two 0-cells two 1-cells two 2-cells


(North-/South-hemisphere)

S1 × S1

one 0-cell two 1-cells one 2-cell

RP 2
one 0-cell one 1-cell one 2-cell

Figure A.1.: Examples of CW-structures


A.14 A. Appendix

Ad 2 =⇒ 3. Because CW-complexes are built up from cells, one can prove


this implication by a careful induction [88, Chapter IV.7, Chapter V.3].
Ad 3 =⇒ 1. If 3. holds, then [ · , f ] is a natural isomorphism [ · , X] =⇒
[ · , Y ]. Then, the Yoneda lemma (Proposition AT.1.2.23) shows X ' Y .
Caveat A.3.5.
• The notion of “weak equivalence” is not an equivalence relation on the
class of topological spaces; in general, symmetry is not satisfied [88,
p. 221].
However, the Whitehead theorem shows that on the class of CW-
complexes, weak equivalence coincides with homotopy equivalence and
thus is an equivalence relation on CW-complexes.
This is similar to the notion of quasi-isomorphism for chain complexes.
• Abstract isomorphisms between homotopy groups of CW-complexes are
not sufficient to conclude that the given CW-complexes are homotopy
equivalent. It is essential that these isomorphisms are induced by a
continuous map.
For example, the spaces RP 2 × S 3 and S 2 × RP 3 (which both admit a
CW-structure) have isomorphic homotopy groups (because they have
the common covering space S 2 × S 3 ; Corollary AT.2.3.25), but they are
not homotopy equivalent (as can be seen from the (cellular) homology
in degree 5; check!).
In particular, in the simply connected case, the Whitehead theorem shows
that ordinary homology with Z-coefficients is a rather powerful tool.
Corollary A.3.6 (Whitehead theorem, simply connected case). Let X be a sim-
ply connected CW-complex. Then the following are equivalent:
1. The space X is contractible (in Top).

2. For each x0 ∈ X and each n ∈ N, we have πn (X, x0 ) = 1.
3. For each n ∈ N, we have Hn (X; Z) ∼
=Z Hn (•; Z).
Proof. Applying the Whitehead theorem (Theorem A.3.4) to the constant
map X −→ • shows that 1. and 2. are equivalent.
Moreover, because X is simply connected, the equivalence of 2. and 3. is
a consequence of the Hurewicz theorem (Corollary AT.4.5.10).
Remark A.3.7 (homotopy invariants of CW-complexes). Examples of homo-
topy invariants of CW-complexes are
• homotopy groups (in particular, the fundamental group) and
• cellular/singular homology
B
Exercise Sheets
Group Cohomology – Exercises
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 1, April 29, 2019

Exercise 1 (a trace on group rings?). Let G be a group and let


τ : ZG −→ Z
X
ag · g 7−→ ae .
g∈G

Which of the following statements are true? Justify your answer with a suitable
proof or counterexample.
1. For all a, b ∈ ZG, we have τ (a · b) = τ (b · a).
P P
2. For all a ∈ ZG, we have τ (a·a∗ ) ≥ 0, where ( g∈G ag ·g)∗ := g∈G ag−1 ·g.
Exercise 2 (standard (co)chain complexes in the literature).
1. What is “C q (Q, G)” from the following article in our notation? [I on p. 3/4]
S. Eilenberg. Topological methods in abstract algebra. Cohomology theory
of groups, Bull. Amer. Math. Soc., 55, pp. 3–37, 1949.
2. What is the “bar resolution C∗ (Γ)” from the following article in our no-
tation? [Definition and Lemma 2.1 a)]
M. Puschnigg. The Kadison-Kaplansky conjecture for word-hyperbolic
groups, Invent. Math., 149(1), pp. 153–194, 2002.
Hints. Of course, you do not need to read/understand the whole article. It
suffices to untangle the terminology and to compare it to our setup. You have
to justify your answer in your submission (e.g., by an explicit comparison).
Exercise 3 (the augmentation ideal). Let G be a group and let I(G) := ker ε
be the augmentation ideal (where ε : C0 (G) = ZG −→ Z is the augmentation
map).
1. Show that I(G) = SpanZ {g − 1 | g ∈ G}.
2. Show that I(G) = SpanZG {s − 1 | s ∈ S} holds for every generating
set S ⊂ G of G.
Exercise 4 (group rings of cyclic groups). Let n ∈ N>0 , let G := Z/n, let t :=
Pn−1
[1] ∈ G, and let N := j=0 tj ∈ ZG. For a ∈ ZG, we consider the associated
ZG-homomorphism Ma : ZG −→ ZG given by right multiplication with a.
1. Show that im MN = ker Mt−1 .
2. Show that im Mt−1 = ker MN .
Bonus problem (Kaplansky zero divisor conjecture and unique products).
1. What is the unique product property of groups?
2. Give an example of a (non-trivial) group with the unique product property
and an example of a group without the unique product property.
3. Show that the group ring ZG has no non-trivial zero divisors if G is a
group with the unique product property.

Submission before May 6, 2019, 10:00, in the mailbox


(Solutions may be submitted in English or German.)
Group Cohomology – Exercises

Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 2, May 6, 2019

Exercise 1 (induced maps in group homology). Let ϕ : G −→ H be a group


homomorphism. Which of the following statements are true? Justify your answer
with a suitable proof or counterexample.

1. If ϕ is injective, then H1 (ϕ; Z) : H1 (G; Z) −→ H1 (H; Z) is injective.


2. If ϕ is surjective, then H1 (ϕ; Z) : H1 (G; Z) −→ H1 (H; Z) is surjective.
Exercise 2 (finitary symmetric groups). Let X be a set with |X| ≥ 2 and let
FSym(X) be the group of all finitary permutations of X; a bijection f : X −→ X
is finitary if the set {x ∈ X | f (x) 6= x} is finite.

1. Compute H 1 FSym(X); Z .

2. Compute H1 FSym(X); Z .

Exercise 3 (certain groups of homeomorphisms in the literature). We consider the


following article:
J. N. Mather. The vanishing of the homology of certain groups of homeo-
morphisms, Topology, 10, pp. 297–298, 1971.

1. What is the main result of this article?


2. Name at least two further published articles whose titles contain the string
“certain groups of homeomorphisms”.
Hints. The database https://mathscinet.ams.org/mathscinet might help.
Access to Mathscinet requires a subscription; the website can be accessed
through the campus network (or using SSH tunnels via UR).
3. Bonus problem. Can you make sense of the definition of C(G) in the proof
of the lemma on p. 297?
4. Bonus problem. What is ϕ in (a) on p. 298? How do you think that this
happened?

Exercise 4 (the integral Heisenberg group). Let


  
 1 x z 
H := 0 1 y  x, y, z ∈ Z ⊂ SL(3, Z)
 
0 0 1

be the (integral) Heisenberg group.


1. Compute H1 (H; Z).

2. Show that rk H = 2.

Please turn over


Bonus problem (a perfect homeomorphism group). Let n ∈ N>0 . For a homeo-
morphism f : Rn −→ Rn , we define the support by

supp f := {x ∈ Rn | f (x) 6= x} ⊂ Rn .

We say that f has compact support if supp f is compact. Let G be the group(!) of
all homeomorphisms Rn −→ Rn with compact support. Show that G is perfect
(whence H1 (G; Z) ∼ =Z 0).
Hints. Let f ∈ G, let B be an open ball containing supp f , and let g ∈ G
with g k (B) ∩ g m (B) = ∅ for all k, m ∈ N with k 6= m (why does such a g exist?).
Then consider the following situation:

f g ◦ f ◦ g −1 g 2 ◦ f ◦ g −2 ...

B g(B) g 2 (B)

Submission before May 13, 2019, 10:00, in the mailbox


Group Cohomology – Exercises
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 3, May 13, 2019

Exercise 1 (projectivity). Let G be a group and let Z be equipped with the trivial
G-action. Which of the following statements are true? Justify your answer with
a suitable proof or counterexample.
1. If Z is a a projective ZG-module, then G is finite.
2. If G is finite, then Z is a projective ZG-module.
Exercise 2 (cohomology of Z/3). Use group extensions to show that H 2 (Z/3; Z)
contains at least three elements (where Z carries the trivial action). More pre-
cisely: Provide three extensions of Z/3 by Z that are pairwise non-equivalent.
∼Z
Exercise 3 (cohomology of Z2 ). Use group extensions to show that H 2 (Z2 ; Z) 6=
0 (where Z carries the trivial action).
Hints. The Heisenberg group (Sheet 2, Exercise 4) can serve as a middleman.
Exercise 4 (extensions in the literature). We consider the following article:
M. Bucher, R. Frigerio, T. Hartnick. A note on semi-conjugacy for circle
actions, L’Enseignement Mathématique (2), 62, pp. 317–360, 2016.
1. What is “e(ξ)” from this article (paragraph before Lemma 3.1) in our
notation? In particular, how does “cσ ” relate to our notation?
2. Give a proof of Lemma 3.1 (in our notation).
Bonus problem (cohomology of homeomorphisms on the circle). We consider the
circle S 1 := R/Z, the group G := Homeo+ (S 1 ) of orientation-preserving homeo-
morphisms of S 1 , and the subgroup

e := f ∈ Homeo+ (R) ∀x∈R f (x + 1) = f (x) + 1
G
of the orientation-preserving homeomorphisms of R. A homeomorphism R −→ R
is orientation-preserving if it is monotonically increasing. Moreover, a homeo-
morphism S 1 −→ S 1 is orientation-preserving if it preserves orientations in the
sense of linear algebra (which can be defined via suitable determinants). Let
p : R −→ R/Z = S 1 be the projection map and let
e −→ G
π: G

f 7−→ [x] 7→ p(f (x))
1. Show that π is a well-defined group homomorphism and that there is a
central extension of the form
/Z e π
0 /G /G / 1.

Hints. For topologists, surjectivity of π should be easy!


2. Show that this extension is not trivial and conclude that H 2 (G; Z) ∼
6 Z 0.
=
Hints. Torsion!

Submission before May 20, 2019, 10:00, in the mailbox


Group Cohomology – Exercises
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 4, May 20, 2019

Exercise 1 ((co)homology with group ring coefficients). Let G := Z. Which of


the following statements are true? Justify your answer with a suitable proof or
counterexample.
1. We have H1 (G; ZG) ∼=Z 0.
2. We have H 1 (G; ZG) ∼
=Z 0.
Exercise 2 ((co)homology of Z2 ). Let T := Z2 .
1. Show that ZT ∼=Ring Z[a, b]S , where S := {an · bm | n, m ∈ N}.
∂2 ∂1
2. Show that the complex · · · /0 / ZT / ZT ⊕ ZT / ZT,
where
∂2 : ZT −→ ZT ⊕ ZT

x 7−→ x · (1 − b), x · (a − 1)
∂1 : ZT ⊕ ZT −→ ZT
(x, y) 7−→ x · (a − 1) + y · (b − 1),
together with the augmentation ε : ZT −→ Z is a projective resolution of
the trivial ZT -module Z over ZT .
3. Compute H∗ (T ; Z) and H ∗ (T ; Z) (with the trivial T -action on Z).
4. Compute H 1 (T ; ZT ).
5. Bonus problem. What could be the geometric background of the above
projective resolution?
Exercise 3 (cohomology of free groups). Let F be a free group of rank 2. Show
that H 1 (F ; ZF ) is not a finitely generated Z-module.
Hints. For a free generating set {a, b} of F , it might help to consider (bn , an ) ∈
ZF × ZF , the reduction to Z[Fab ] × Z[Fab ], and the resolution from Exercise 2.
Exercise 4 (Shapiro lemma in the literature). We consider the following article:
R.G. Swan. Groups of cohomological dimension one, Journal of Algebra,
12, pp. 585–601, 1969.
1. How is cdZ defined and what does the Shapiro lemma say about cdZ ?
2. Which result does Swan use to derive Theorem B from Theorem A?
Bonus problem (some congruences for Z/p-actions on Z/pn ). Let p ∈ N be an
odd prime, let n ∈ N>1 , and let a ∈ Z with ap ≡ 1 mod pn .
1. Show that a ≡ 1 mod pn−1 .
2. Conclude that if a 6≡ 1 mod pn , there exists a k ∈ {1, . . . , p−1} with ak ≡
1 + pn−1 mod pn .

Submission before May 27, 2019, 10:00, in the mailbox


Group Cohomology – Exercises
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 5, May 27, 2019

Exercise 1 (homology of dihedral groups). For n ∈ N≥3 , let Dn denote the dihe-
dral group for n (i.e., the isometry group of a regular Euclidean n-gon). Which
of the following statements are true? Justify your answer with a suitable proof
or counterexample.
1. H2019 (D2021 ; Z/2020) ∼=Z 0
2. H2019 (D2020 ; Z/2021) ∼
=Z 0
Hints. You may use the description Dn ∼ =Group Z/n o Z/2, where [1] ∈ Z/2 acts
by (additive) inversion on Z/n (Proposition III.1.1.57).
Exercise 2 (the infinite dihedral group). The infinite dihedral group D∞ is the
isometry group of the metric space Z with respect to the metric inherited from
the standard metric on R. Let t denote the reflection at 0, let s denote the
translation by 1, and let t0 denote the reflection at 1/2 ∈ R. Solve two out of
the following four problems:
1. Show that S := {s, t} is a generating set of D∞ and that D∞ is isomorphic
to a suitable semi-direct product Z o Z/2.
2. Show that the word metric on D∞ associated with S is isometric to the
word metric on Z×Z/2 associated with the generating set {(1, 0), (0, [1])}.
3. Show that T := {t, t0 } is a generating set of D∞ .
4. Show that the word metric on D∞ associated with T is isometric to the
word metric on Z associated with the generating set {1}.
Exercise 3 (metric embedding notions in the literature). We consider:
J. Block, S. Weinberger. Aperiodic tilings, positive scalar curvature, and
amenability of spaces, J. Amer. Math. Soc., 5(4), pp. 907–918, 1992.
1. Prove the statement “Also note that a coarse quasi-isometry in the sense
of Gromov is an EPL map.” (p. 909). More precisely: Show that every
quasi-isometric embedding between metric spaces is an effectively proper
Lipschitz map (defined on p. 909).
2. Does the converse also hold? Justify your answer!
Exercise 4 ((co)induction of finite index subgroups). Let G be a group and let
H ⊂ G be a subgroup of finite index. We consider
ϕ : IndG
H (B) = ZG ⊗ZH B −→ HomH (ZG, B) = CoindH (B)
G

g ⊗ b 7−→ x 7→ χH (x · g) · (x · g) · b
ψ : CoindG G
H B = HomH (ZG, B) −→ ZG ⊗ZH B = IndH (B)
X
f 7−→ g ⊗ f (g −1 )
gH∈G/H

Show that ϕ and ψ are well-defined ZG-linear maps and that ϕ and ψ are
mutually inverse.
Please turn over
Bonus problem (Legendre symbol and transfer). Let p ∈ N be an odd prime.
1. What is the Legendre symbol associated with p ?

2. Show that the Legendre symbol “coincides” with the transfer on H1 ( · ; Z)



of the subgroup {−1, +1} of the multiplicative group Z/(p) .

Submission before June 3, 2019, 10:00, in the mailbox


Group Cohomology – Exercises

Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 6, June 3, 2019

Exercise 1 (UDBG spaces). Which of the following statements are true? Justify
your answer with a suitable proof or counterexample.
1. Every uniformly discrete metric space has bounded geometry.
2. Every metric space with bounded geometry is uniformly discrete.
Exercise 2 (uniformly finite chains). Let P
R be a normed ring with unit, let (X, d)
be a UDBG space, let n ∈ N≥1 , let c = x∈X n+1 cx · x ∈ Cnuf (X; R), and let j ∈
uf
{0, . . . , n}. Show that the following map is a well-defined chain in Cn−1 (X; R):

X n −→ R
X
y 7−→ cx .
x∈{z∈X n+1 |(z0 ,...,b
zj ,...,zn )=y}

Exercise 3 (the fundamental class in uniformly finite homology). Let (X, d) :=


PZ ∈ H0 (X; Z) be the homology class represented
uf
(Z, d{1} ) and for A ⊂ X let [A]
by the uniformly finite cycle x∈A 1 · x.

1. Show that for every finite set A ⊂ X, we have [A]Z = 0 in H0uf (X; Z).
2. Show that for each n ∈ N>0 , there exists a class αn ∈ H0uf (X; Z) that
satisfies n · αn = [X]Z in H0uf (X; Z).
Exercise 4 (means). We consider the following article:
M. Gromov. Volume and bounded cohomology, Publ. Math. IHES, 56,
pp. 5–99, 1982.

1. Where and how are amenable groups defined in this article?


2. Show that this notion of amenability is equivalent to ours; more precisely,
show that an R-linear map m : `∞ (G, R) −→ R is a left-invariant mean
on G if and only if all of the following conditions are satisfied:
• m(1) = 1
• m(g · f ) = m(f ) for all f ∈ `∞ (G, R) and all g ∈ G

• m(f ) ≤ |f |∞ for all f ∈ `∞ (G, R)

Bonus problem (Ponzi schemes).


1. Who was Charles Ponzi?
2. What is a Ponzi scheme?

Submission before June 10, 2019, 10:00, in the mailbox

As June 10 is a holiday: extended deadline: June 11, 10:00


Group Cohomology – Exercises

Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 7, June 10, 2019

Exercise 1 (uniformly finite homology of groups). Which of the following state-


ments are true? Justify your answer with a suitable proof or counterexample.

1. H0uf ha, b | a2019 , aba−1 b−1 i; Z ∼ =Z 0

2. H0uf ha, b, c, d | c2019 d2020 i; Z ∼=Z 0
Exercise 2 (Z 6∼QI Z2 ). Use uniformly finite homology H∗uf ( · ; R) to prove that
Z and Z2 are not quasi-isometric.
Hints. Amenability calls for transfer! And Sheet 4 might help.
Exercise 3 (quasi-isometry vs. bilipschitz equivalence). We consider the following
article:
K. Whyte. Amenability, bi-Lipschitz equivalence, and the von Neumann
conjecture, Duke Math. J., 99(1), pp. 93–112, 1999.

1. Give a proof of the statement “Observe that a quasi-isometry between


UDBG spaces is bilipschitz if and only if it is bijective” at the beginning
of the proof of Theorem 4.1.
2. How is this fact used in the proof of Theorem 4.1?

Exercise 4 (homology of free groups with `2 -coefficients). Let F be a free group


of rank 2. Give an explicit example of a chain b ∈ C1 (F ; `2 (F, R)) (with the
usual left ZF -module structure on `2 (F, R)) that satisfies

∂1 b = e ⊗ χ{e} ∈ C0 F ; `2 (F, R) .

Sketch this chain!


Bonus problem (measure equivalence).
1. What is measure equivalence of (countable) groups?
2. How does the definition of measure equivalence relate to the dynamical
criterion for quasi-isometry of (finitely generated) groups?
3. How does measure equivalence relate to orbit equivalence of group actions?
Hints. It is enough to cite such a result from the literature; no proof is
required.
4. Give an example of (non-trivial) coefficients for group (co)homology that
lead to measure equivalence invariants!
Hints. It is enough to cite such a result from the literature; no proof is
required.

Submission before June 17, 2019, 10:00, in the mailbox


Group Cohomology – Exercises
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 8, June 17, 2019

Exercise 1 (commutator length). Let G be a group. Which of the following state-


ments are true? Justify your answer with a suitable proof or counterexample.
1. If g ∈ [G, G] and h ∈ G, then clG (h · g · h−1 ) = clG (g).
2. If g, h ∈ [G, G], then clG (g · h) = clG g + clG h.
Exercise 2 (homogenisation). Let G be a group and let ϕ : G −→ R be a quasi-
morphism.
1. Show that the following map is a well-defined homogeneous quasi-morphism
on G that is uniformly close to ϕ (this requires, in particular, a proof of
the existence of the limit on the right-hand side):
ϕ : G −→ R
1
g 7−→ lim · ϕ(g n )
n→∞ n
2. Conclude that the following map is a well-defined isomorphism of R-vector
spaces:
QM(G)/ QM0 (G) −→ QM(G)/ HomGroup (G, R)
[ϕ] 7−→ [ϕ]

Exercise 3 (rotation number). We consider the following article:


K. Mann. Rigidity and flexibility of group actions on the circle, Handbook
of group actions, IV, pp. 705–752, Adv. Lect. Math., 41, International
Press, 2018.
The preprint version is freely available at: https://arxiv.org/abs/1510.00728
Let Ge := {f ∈ Homeo+ (R) | ∀x∈R f (x + 1) = f (x) + 1} be the group (with
respect to composition) of periodic orientation-preserving (i.e., monotonically
increasing) homeomorphisms of R (see also the bonus exercise on Sheet 3).
1. How is the rotation number erot : G e −→ R on G e defined? Why is this
definition well-defined? (You should give more details than the article . . . )
e
2. Show that erot is a homogeneous quasi-morphism on G.
Exercise 4 (means from vanishing bounded cohomology). Let G be a group, let
V := `∞ (G, R)/C, where C ⊂ `∞ (G, R) is the subspace of constant functions,
and suppose that Hb1 (G; V # ) ∼
=R 0. Show that there exists a bounded R-linear
functional µ : `∞ (G, R) −→ R with µ(1) = 1 that is left-invariant.
Bonus problem (acronyms). How do the following acronyms expand in Mathe-
matics? Where are they located?
MFO, MSRI, RIMS, IAS, BIRS, IMPAN,
IML, MPIM, INI, IHÉS, PIMS, KIAS

Submission before June 24, 2019, 10:00, in the mailbox


Group Cohomology – Exercises
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 9, June 24, 2019

Exercise 1 (second cohomology and stable commutator length). Let G be a group.


Which of the following statements are true? Justify your answer with a suitable
proof or counterexample.
∼R 0, then sclG = 0.
1. If H 2 (G; R) =
b

2. If H 2 (G; R) ∼
=R 0, then sclG = 0.
Exercise 2 (torsion groups). Let G be a torsion group (i.e., every element in G
has finite order).
1. Compute the function sclG : [G, G] −→ R and the space QM(G) (directly,
without using Bavard duality).
2. Let G, in addition, have the property that every group extension of the
form 0 −→ R −→ ? −→ G −→ 1 splits. Show that Hb2 (G; R) ∼ =R 0.
Exercise 3 (quasi-isomorphisms). Let R be a ring. A chain map f∗ : C∗ −→ D∗
in R Ch is a quasi-isomorphism if, for each n ∈ N, the induced homomor-
phism Hn (f∗ ) : Hn (C∗ ) −→ Hn (D∗ ) is an isomorphism. Show (via an example
over a suitable ring R) that if there exists a quasi-isomorphism C∗ −→ D∗ , then,
in general, there is no quasi-isomorphism D∗ −→ C∗ .
Exercise 4 (exact categories). We consider the article
T. Bühler. Exact categories, Expo. Math., 28, pp. 1–69, 2010.
Before proceeding, you should look up what an additive category is.
1. On p. 4, admissible epics are defined. Make this definition explicit.
2. What is the obscure axiom?
3. Why is it called obscure?
4. How are exact functors between exact categories defined?
Bonus problem (duality principle for semi-norms on homology). Let C∗ be a chain
complex in the category of normed R-vector spaces (and bounded linear opera-
tors) and let D∗ := BHom(C∗ , R) be the dual cochain complex. Let n ∈ N and
let α ∈ Hn (C∗ ) be represented by the cycle c ∈ Cn .
1. Show that
n 1 o

kαk = sup f ∈ Dn , δ n f = 0, f (c) = 1 .
|f |∞
Here, k · k denotes the semi-norm on Hn (C∗ ) induced by the norm on Cn
and δ ∗ is the coboundary operator of D∗ . Moreover, we set sup ∅ := 0.
Hints. Hahn-Banach!
2. Does there exist an amenable group G and a class α ∈ H2019 (G; R)
with kαk1 = 2019 ? Here, k·k1 denotes the semi-norm on H∗ (G; R) induced
by the `1 -norm on C∗R (G).

Submission before July 1, 2019, 10:00, in the mailbox


Group Cohomology – Exercises
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 10, July 1, 2019

Exercise 1 (Tor). In the following, all modules carry the trivial group action.
Which of the following statements are true? Justify your answer with a suitable
proof or counterexample.
Z[Z2 ]
1. Tor2 (Z, Z) ∼
=Z 0
Z[Z]
2. Tor1 (Z2 , Z2 ) ∼
=Z 0
Exercise 2 (algebraic mapping cones). Let R be a ring and let f∗ : C∗ −→ D∗
be a chain map of R-chain complexes. The mapping cone of f∗ is the R-chain
complex Cone∗ (f∗ ) consisting of the chain modules
Conen (f∗ ) := Cn−1 ⊕ Dn
for all n ∈ N (where C−1 := 0), equipped with the boundary operators
∂n : Conen (f∗ ) −→ Conen−1 (f∗ )
C

(x, y) 7−→ −∂n−1 (x), ∂nD (y) − fn−1 (x)
for all n ∈ N>0 . Show the mapping cone trick, i.e., that f∗ : C∗ −→ D∗ is a
quasi-isomorphism if and only if

∀n∈N Hn Cone(f∗ ) ∼ =R 0.
Hints. For the boundary operator on Cone(f∗ ), several different sign conventi-
ons are in use. Therefore, literature has to be used with care!
Exercise 3 (quasi-isomorphisms of complexes of projectives). Let R be a ring.
Prove that if C∗ and D∗ are (N-indexed) R-chain complexes that consist of
projective R-modules, then every quasi-isomorphism C∗ −→ D∗ is a chain ho-
motopy equivalence.
Hints. Mapping cone . . .
Exercise 4 (injectivity). We consider the article
N.V. Ivanov. Foundations of the theory of bounded cohomology, J. Soviet
Math., 37, pp. 1090–1114, 1987.
1. How are ordinary injective modules defined in module categories?
2. How are relatively injective Banach G-modules defined in the article?
3. What is the fundamental theorem of homological algebra in this context?
4. Use MathSciNet (https://www.ams.org/mathscinet) to find an article that
solves the problem in Remark 3.9.1.
Hints. Use the “Citations” tool!
Bonus problem (mapping cones in algebra and topology).
1. How can one relate algebraic mapping cones to topological mapping cones?
2. What are differences/similarities between the properties of algebraic map-
ping cones and topological mapping cones, respectively?

Submission before July 8, 2019, 10:00, in the mailbox


Group Cohomology – Exercises
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 11, July 8, 2019

Exercise 1 (puzzle convergence). Let R be a ring and let (E r , dr )r∈N>1 be a


homological spectral sequence that converges to a graded R-module A:
2
Epq =⇒ Ap+q .

Which of the following statements are true? Justify your answer with a suitable
proof or counterexample.
2 ∼
1. If Epq =R 0 for all p, q ∈ N with q 6= 2019, then A2020 ∼ 2
=R E1,2019 .
2 ∼
2. If Epq =R 0 for all p, q ∈ N for which p + q is odd, then A2019 ∼
=R 0.
Exercise 2 (homology of the Heisenberg group). Let H ⊂ SL(3, Z) be the integral
Heisenberg group (Sheet 2, Exercise 4). Compute Hn (H; Z) for all n ∈ N (where
H acts trivially on Z) via the Hochschild-Serre spectral sequence.
Hints. You may use the result on H1 (H; Z) from Exercise 4 on Sheet 2.
Exercise 3 (standard spectral arguments). We consider the article
M.R. Bridson, P.H. Kropholler. Dimension of elementary amenable groups,
J. Reine Angew. Math., 699, p. 217–143, 2015.
Institut Mittag-Leffler, report no. 38, 2011/2012, spring.
In the paragraph after Theorem I.5 it says that “In both cases of course the
inequalities ≤ follow at once from standard spectral sequence arguments.”
1. Which spectral sequence should be applied?
2. Carry out these “standard spectral sequence arguments.”
Hints. Cohomological dimension already appeared in Exercise 4 of Sheet 4.
Exercise 4 (the conjugation action on homology).
1. Let G be a group, let A be a ZG-module, and let g ∈ G. Moreover, let

c(g) := (x 7→ g · x · g −1 , x 7→ g · x) ∈ MorGroupMod (G, A), (G, A) .

Show that Hn (c(g)) = idHn (G;A) for all n ∈ N.


2. For each n ∈ N, compute Hn (ϕ; Z) : Hn (Z/3; Z) −→ Hn (Z/3; Z), where
ϕ : Z/3 −→ Z/3 is the group automorphism x 7→ −x.
Hints. First understand how H∗ (c(g)) can be described in terms of a projective
resolution of Z and then avoid confusion at all cost.
Bonus problem (centre kills). Let n, k ∈ N>0 and let K be a field with char K 6=
2. Show that Hk (GL(n, K); K n ) ∼
=K 0, where GL(n, K) acts on K n by matrix
multiplication.
Hints. Let elements of the centre act . . .

Submission before July 15, 2019, 10:00, in the mailbox


Group Cohomology – Exercises
Prof. Dr. C. Löh/Dr. D. Fauser/J. P. Quintanilha/J. Witzig Sheet 12, July 15, 2019

Exercise 1 (universal coefficients and Künneth). Let G be a group that satis-


fies Hn (G; Z) ∼
=Z Hn (1; Z) for all n ∈ N. Which of the following statements are
true? Justify your answer with a suitable proof or counterexample.
1. If H is a group and n ∈ N, then Hn (G × H; Z) ∼
=Z Hn (H; Z).
2. If A is a Z-module (with trivial G-action) and n ∈ N, then Hn (G; A) ∼
=Z
Hn (1; A).
Exercise 2 (topology of discrete groups). We consider the articles
G. Baumslag, E. Dyer, A. Heller. The topology of discrete groups, J. Pure
Appl. Algebra, 16(1), pp. 1–47, 1980.
C.R.F Maunder. A short proof of a theorem of Kan and Thurston. Bull.
London Math. Soc., 13(4), pp. 325–327, 1981.

1. What is a mitotic group?


2. Sketch the proof that mitotic groups have trivial homology.
3. How does the main theorem of the second paper relate to group homology?
4. How/Where is the first paper used in the second paper?

Exercise 3 (a classifying space for the Heisenberg group). Let H ⊂ SL(3, Z) be


the integral Heisenberg group (Sheet 2, Exercise 4). Show that there exists a
classifying space for H that is a compact 3-manifold.
Hints. Start with the real Heisenberg group.
Exercise 4 (surface groups). For g ∈ N≥2 , we consider the group


Γg := a1 , . . . , ag , b1 , . . . , bg [a1 , b1 ] · [a2 , b2 ] · · · · · [ag , bg ] .

1. Compute H∗ (Γg ; Z), using “the” oriented closed connected surface of ge-
nus g as classifying space (and a suitable CW-structure on it).
2. Compute the deficiency of the group Γg .
Bonus problem (classifying space of a category).
1. How is the classifying space of a (small) category defined?
2. How can one construct classifying spaces for groups out of classifying
spaces of a category?
Hints. Rough sketches suffice.
Bonus problem (lecture notes). Find as many typos/mistakes in the lecture
notes as you can!

Submission before July 22, 2019, 10:00, in the mailbox


Group Cohomology – Exercises

Prof. Dr. C. Löh/Dr. D. Fauser/J. P. Quintanilha/J. Witzig Sheet 13, July 22, 2019

The ISSSS case. The InterStellar Spectral Sequence Station (ISSSS) has been
infiltrated and taken over by a pangalactic group X of con artists. Detective
Blorx, an agent of the Cohomological Intelligence Agency, handles the case. He
collected the following evidence:

A. The suspects. Only the following groups are sufficiently powerful to be able
to infiltrate the ISSSS; here, Fn denotes the free group of rank n and Γg denotes
the surface group of genus g:
• The Free Group: F2018
• The Freer Group: F2019
• The Bi-Cycle Group: Z/4 × Z/674 × F2018
• GaoS (Group avalaible on Surfaces): Γ1010
• Torus Inc.: Z2019

B. Statements by witnesses.
• The group X was not able to carry invariant means.
• The free products (F4 × F674 ) ∗ X and (F4 × F674 ) ∗ F2018 are not com-
mensurable.
• Every finite subgroup of X acted freely on some sphere.
• The group X has no subgroup that is isomorphic to the fundamental group
of an oriented closed connected surface of genus at least 2.

C. Project Euler. Blorx hacked into the servers of the Secret Invariants Service
and found the following files on Project Euler:
Let G be a group of type F, i.e., G admits a classifying space with a finite
CW-structure. Then the Euler characteristic of G is defined as
X
χ(G) := (−1)n · dimQ Hn (G; Q).
n∈N

À In this situation, χ(G) is a well-defined integer and it equals the Euler


characteristic of any classifying space for G with finite CW-structure.
Á If H ⊂ G is a subgroup of finite index, then χ(H) = [G : H] · χ(G).
 If H is a group of type F, then χ(G × H) = χ(G) · χ(H).
à If H is a group of type F, then χ(G ∗ H) = χ(G) + χ(H) − 1.

D. Law of Commensurability. Two groups G and H are commensurable, if there


exist finite index subgroups G ⊂ G and H ⊂ H with G ∼
=Group H.

Please turn over


Exercise (4 + 8 + 4 credits). Help Blorx!
1. Establish two of the four claims of Project Euler.

2. Which group infiltrated the ISSSS? Justify your answer!


Hints. Sphere actions will be discussed in the final lecture.
?tcaretni ytilibarusnemmoC fo waL eht dna reluE tcej orP od woH
3. Help Blorx to get a promotion by working as an informant for him: Pick
your favourite field of Mathematics (e.g., algebraic number theory, Rie-
mannian geometry, geometric topology, geometric group theory, operator
algebras, functional analysis, graph theory, algebraic geometry, ergodic
theory, . . . ).
Find (in the literature or on the servers of the Secret Invariants Service)
an application of group (co)homology to that field that we did not discuss
in the lectures!
Bonus problem (fickle witness). The last witness statement was later withdrawn
by the witness and changed to the following:
• The group X is not quasi-isometric to F2018 .

Does this lead to the same conclusion?


Hints. One can (provided one has access to it) use the secret H2UF-technology.

Optional submission before July 29, 10:00, in the mailbox


B.18 B. Exercise Sheets
C
Etudes
Group Cohomology – Etudes
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 0, April 25, 2019

Exercise 1 (group rings). Let G := Z/8 and let t := [1] ∈ Z/8.


1. Is ZG a commutative ring?
P7
2. Compute (t − 1) · j=0 tj in ZG.

3. Is ZG isomorphic to Z[e2πi/8 ] ⊂ C ?
4. Is t4 − 1 a unit of ZG ?

Exercise 2 (invariants). Let G be a group. For each of the following ZG-modules,


compute the invariants.
1. R with the trivial G-action
2. map(G, R), the space of R-valued functions on G, with respect to the
G-action by right translation on G
3. `2 (G, C), the space of square-summable complex functions on G, with
respect to the G-action by right translation on G
4. A × B, where A and B are ZG-modules, with respect to the diagonal
G-action on A × B.
Exercise 3 (basic homological algebra).
1. What is the definition of chain complexes and chain maps?
2. What are typical examples?
3. What is the homology of a chain complex?
4. How can homology be computed?
5. How does all this relate to exactness?
6. What is homotopy invariance in homological algebra?
7. Why did we introduce chain complexes in Commutative Algebra or Alge-
braic Topology or Any-Other-Course?
Hints. In case you don’t know any homological algebra: Don’t panic! Basic
notions from homological algebra will also be quickly reviewed in the lectures.
Exercise 4 (basic category theory).
1. What is the definition of a category?
2. What is the definition of a functor between categories?
3. Give examples of categories and functors between them. (How) Did these
arise naturally in previous courses?
Hints. In case you don’t know any category theory: Don’t panic! Categories
and functors will also be quickly reviewed in the lectures.

no submission!
These problems will be discussed in the exercise classes in the second week.
Group Cohomology – Etudes
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 1, May 2, 2019

Exercise 1 (group (co)homology in degree 0). Let G := Z/2. For a Z-module Z,


let Z − be the left ZG-module whose underlying additive group is Z and whe-
re the non-trivial element of G acts by multiplication with −1. Compute the
following (co)homology groups:
1. H 0 (G; R− )
2. H0 (G; R− )
3. H 0 (G; Z− )
4. H0 (G; Z− )
5. H 0 (G; Z/2− )
6. H0 (G; Z/2− )
Exercise 2 (simplicial chains). Let G := Z and let t := 1 ∈ Z = G. Which of the
following elements of C1 (G) are cycles (i.e., in the kernel of ∂1 )?
1. 2019 · (t2019 , t2019 )
2. 2019 · (t, t2019 )
3. 1 · (t0 , t) + 1 · (t, t2 ) − 1 · (t0 , t2 )
Exercise 3 (free groups). Let F be the free group of rank 2, freely generated
by {a, b} (Appendix A.1). Which of the following equalities hold in F ?
1. (aba−1 )2019 = ab2019 a−1
2. aba−1 · ab2 a−1 = b3
3. [a, b]2 = [a2 , b2 ]
4. [a2 , a−1 b2 ] = ab2 a−2 b−2 a
Exercise 4 (summary). Write a summary of Chapter 1.1 (Foundations: The
group ring) and Chapter 1.2 (The basic definition of group (co)homology), kee-
ping the following questions in mind:
1. How can one work with the group ring? What are basic examples?
2. What are important examples and constructions of ZG-modules?
3. What are the domain categories for group (co)homology?
4. How are the simplicial/bar resolutions constructed? Why?
5. How is group (co)homology defined in terms of the simplicial resolution?
6. Did you check all the little things that we did not discuss in detail in the
lectures?

no submission!
Group Cohomology – Etudes
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 2, May 9, 2019

Exercise 1 (cohomology in degree 1). Compute the following cohomology groups


(with respect to the trivial action on the coefficients).
1. H 1 (Z; Z/2)
2. H 1 (Z/2; Z/3)
3. H 1 (S2019 ; Z/2019)
4. H 1 (S2020 ; Z/2020)
Exercise 2 (presentations). Which groups are described by the following presen-
tations? Use universal properties to verify your claims.
1. ha, b | ai
2. ha, b | abi
3. ha, b | ab2 i
4. ha, b | aba−1 i
5. ha, b | a3 , b2 , aba−1 b−1 i
6. ha, b | a3 , b2 , aba−2 i
Exercise 3 (extensions). Do there exist extensions of the following types?
1. 0 / Z/2 / Z/6 / Z/3 /0

2. 0 / Z/2 / S3 / Z/3 /0

3. 0 /Z /Z /Z /0

4. 0 / Z/2 /Z /Z /1

Exercise 4 (summary). Write a summary of Chapter 1.3 (Degree 0: (Co)Invari-


ants), Chapter 1.4 (Degree 1: Abelianisation and homomorphisms), and Chap-
ter 1.5 (Degree 2: Presentations and extensions) keeping the following questions
in mind:
1. How can one compute group (co)homology in low degrees?
2. What are typical examples?
3. What are typical applications?
4. What kind of finiteness obstructions can we get from group homology in
low degrees?
5. Did you check all the little things that we did not discuss in detail in the
lectures?

no submission!
Group Cohomology – Etudes

Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 3, May 16, 2019

Exercise 1 (projectivity). Which of the following modules are projective?


1. Z/2019 over Z ?

2. Q over Z ?
Q
3. N Q over Q ?
4. Z × {0} ⊂ Z × Z over Z × Z ?
Exercise 2 (homology of cyclic groups). Compute the following (co)homology
groups (where Z/2019 acts trivially on the coefficients):
1. H∗ (Z/2019; Z)
2. H∗ (Z/2019; Z/2019)
3. H∗ (Z/2019; Q)

4. H∗ (Z/3; Z/2019)
Exercise 3 (p-groups). Let p ∈ N be a prime.
1. How can one prove that p-groups are solvable?

2. Give examples of p-groups that do not contain a cyclic subgroup of index p.


Exercise 4 (example list). Start a list of groups and their (co)homology, contai-
ning the following data (as far as you know it):
• name of the group

• standard notation of the group


• standard presentations of the group
• geometric relevance of the group
• homology of the group (at least with Z-coefficients) and method of com-
putation
• cohomology of the group (at least with Z-coefficients) and method of com-
putation
• applications of this group (co)homology

Update this list during this course (don’t forget to add the examples from the
exercises!).

no submission!
Group Cohomology – Etudes

Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 4, May 23, 2019

Exercise 1 (tree navigation). Let F be a free group, freely generated by {a, b}


with a 6= b. In “the” 4-regular tree (Figure 1.6), find the vertices corresponding
to the following elements of F :

1. abba
2. aba−1 b−1
3. a−2019 b
4. ab2019

Exercise 2 ((non-)triviality?). Use transfer/functoriality to decide whether the


following homology groups are trivial or not (in each of these cases, the action
on the coefficients is trivial).

1. H2020 Z/2019 × Z/2019; Z/2018

2. H2019 Z/2018 × Z/2018; Z/2020
3. H2019 (Z × Z/2018; Z)

4. H2018 (Z × Z/2019; Q)
Exercise 3 (summary). Write a summary of Chapter 1.6 (Changing the resolu-
tion) and Chapter 1.7 ((Co)Homology and subgroups), keeping the following
questions in mind:
1. What are projective resolutions?

2. How can projective resolutions be used to compute group (co)homology?


3. What are typical examples?
4. What are applications of this approach?

5. How are the (co)homology of subgroups and ambient groups related?


6. Did you check all the little things that we did not discuss in detail in the
lectures?
Exercise 4 (save early, save often). And now for something completely different:

1. What is S.M.A.R.T.?
2. How can S.M.A.R.T. be used to make predictions about disk health?
3. Find a good backup solution for your data!

no submission!
Group Cohomology – Etudes

Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 5, May 30, 2019

Exercise 1 (quasi-isometric embeddings). We consider the metric space R with


the standard metric. Which of the following maps R −→ R are quasi-isometric
embeddings?

1. x 7→ 2019 · x + 2019
2. x 7→ x2019
3. x 7→ d2019 · xe
1
4. x 7→ x2019 +1

Exercise 2 (Cayley graphs). Sketch the following Cayley graphs!



1. Cay Z × Z/2, {(1, 0), (0, [1])}

2. Cay Z2 , {(1, 0), (1, 1), (0, 1)}

3. Cay Z2 , {(1, 2), (2, 1)}

4. Cay D2019 , {t, t0 } , where t and t0 are two different reflections
Exercise 3 (the Hahn-Banach theorem). Recall/look up the following terms and
statements from functional analysis:

1. What is a bounded linear map?


2. What is the relation between boundedness and continuity for linear maps?
3. What does the Hahn-Banach theorem say?

Exercise 4 (GAP). Figure out how to use GAP (https://www.gap-system.org/)


to compute group (co)homology!

no submission!
Group Cohomology – Etudes

Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 6, June 6, 2019

Exercise 1 (amenability). Which of the following groups are amenable?


1. Z × Z/2019

2. D∞
3. the Heisenberg group
4. the symmetric group Sym(N) of all bijections N −→ N
5. SL2 (Z)
Q
6. N R
Exercise 2 (uniformly finite chains). Which of the following “sums” describe uni-
formly finite chains in C1uf (Z; Z) ? Draw them! If they describe uniformly finite
chains: What is their boundary?
P
1. n∈Z 1 · (n, n + 1)
P
2. n∈N 1 · (n, n + 1)
P
3. n∈N 1 · (1, n)
P
4. n∈N 1 · (−n, n)
P
5. n∈N n · (n, n + 1)
P
6. n∈N 2019 · (n, n + 2019)

Exercise 3 (uniformly finite homology of finite groups). Let G be a finite group.


Compute H∗uf (G; Z) in as many ways as you can!

Exercise 4 (summary). Write a summary of Chapter 2.1 (Foundations: Geome-


tric group theory), keeping the following questions in mind:
1. What is the (geo)metric setup in geometric group theory?
2. What are typical examples?

3. What is amenability?
4. Did you check all the little things that we did not discuss in detail in the
lectures?

no submission!
Group Cohomology – Etudes

Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 7, June 13, 2019

Exercise 1 (group presentations and amenability). Which of the following groups


are amenable?
1. ha, b | a, bi
2. ha, b | aba−1 b−1 i
3. ha, b, c | c2 i
4. ha, b, c | abi
5. ha, b | a2 , b2 i
6. ha | a2019 i
2
Exercise 2 (bounded cochains). Which of the following cochains in C (Z2 ; R)
are bounded? Which of them are cocycles?
1. [x | y] 7−→ det(x y)
2. [x | y] 7−→ kxk2
3. [x | y] 7−→ 2019
1
4. [x | y] 7−→ kxk2020 +kyk2020 +1
2 2

Exercise 3 (amenable vs. free). Start a table for amenable groups and free groups,
respectively, listing their behaviour with respect to:
1. uniformly finite homology
2. bounded cohomology
3. quasi-morphisms
4. stable commutator length
Exercise 4 (summary). Write a summary of Chapter 2.2 (Uniformly finite ho-
mology) keeping the following questions in mind:
1. What is uniformly finite homology of spaces?
2. Which coefficients for group homology lead to uniformly finite homology?
3. What is the fundamental class in uniformly finite homology in degree 0 ?
4. How does all this relate to amenability?
5. Did you check all the little things that we did not discuss in detail in the
lectures?

no submission!
Group Cohomology – Etudes

Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 8, June 20, 2019

Exercise 1 (counting quasi-morphisms). Which of the following counting quasi-


morphisms on ha, b | i are homogeneous? Which are group homomorphisms?
1. ψa
2. ψa2
3. ψab
4. ψb−1
5. ψaba−1 b−1
Exercise 2 (functional analysis basics). Recall/loop up the following items from
functional analysis:
1. bounded linear operator
2. operator norm
3. Hahn-Banach theorem
4. open mapping theorem
Exercise 3 (Tor). Recall the following terminology:
1. projective module
2. projective resolution
3. construction/properties of Tor
Hints. In case you don’t know anything about Tor: Don’t panic! I will
quickly review the material in the lectures.
Exercise 4 (summary). Write a summary of Chapter 2.3 (Bounded cohomology)
keeping the following questions in mind:
1. What is bounded cohomology of groups?
2. How does bounded cohomology characterise amenability?
3. What are similarities/differences between bounded cohomology and ordi-
nary group cohomology?
4. How does bounded cohomology relate to quasi-morphisms?
5. How do quasi-morphisms relate to stable commutator length?
6. Did you check all the little things that we did not discuss in detail in the
lectures?

no submission!
Group Cohomology – Etudes

Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 9, June 27, 2019

Exercise 1 (scl). Determine whether the following stable commutator lengths


are zero or not:

1. sclZ [2019, 2018] · [2019, 2018]

2. sclha,b | i [a, b] · [a, b]

3. sclha,b | i [a · b, b]

4. sclha,b | i [a · a, a]

5. sclha,b | a2 ,b2 i [a, b] · [a, b]
Exercise 2 (Tor). Compute the following Tor-groups!
1. TorZ0 (Z/2019, Z)
2. TorZ2019 (Z/2019, Z/2019)
3. TorZ1 (Z, Z/2019)
4. TorZ1 (Z/2018, Z/2019)
5. TorZ1 (Z/2019, Z/2019)
6. TorZ1 (Z/2019, Q)
Exercise 3 (quasi-isomorphisms). For each choice C∗ and D∗ of the followi-
ng chain complexes (over the ring Z), decide whether there exists a quasi-
isomorphism C∗ −→ D∗ or not.

/Z 0 /Z
1. · · · /0 /0

/Z 1 /Z
2. · · · /0 /0

3. · · · /0 /0 /Z /0

4. · · · /0 /0 / Z/2019 /0

/Z 2019 /Z
5. · · · /0 /0

Exercise 4 (homological algebra). Recall/look up the following fundamental re-


sults of homological algebra:
1. long exact homology sequence/snake lemma
2. horseshoe lemma

no submission!
Group Cohomology – Etudes
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 10, July 4, 2019

Exercise 1 (spectral sequence). The following diagram is the second page of a


homological spectral sequence (of Z-modules; all unlabelled modules are trivial).
1. Which components of the second differential have to be trivial? Which
could be non-trivial? What about the third differential? What about all
differentials?
2. What does this imply for the ∞-page?
q

Z
Z
Z/2 Z Z
Z Z Z/2 Z p

Exercise 2 (five lemma).


1. View the situation of the five lemma as a double complex.
2. Spell out the double complex spectral sequences of this double complex
(i.e., what are the first pages? To what do they converge?, . . . ).
3. Use the double complex spectral sequences to prove the five lemma.
Exercise 3 (long exact homology sequence).
1. Spell out the filtration spectral sequence for the inclusion of a single sub-
complex of a chain complex (i.e., what is the first page? To what does it
converge?, . . . ).
2. Use the filtration spectral sequence in this situation to establish the long
exact homology sequence associated with a short exact sequence of chain
complexes.
Exercise 4 (summary). Write a summary of Chapter 3.1 (Derived functors),
keeping the following questions in mind:
1. Why does one need derived functors?
2. What are derived functors?
3. How can derived functors be constructed/computed?
4. How do derived functors relate to group (co)homology?
5. What is the derived category and what are total derived functors?
6. Did you check all the little things that we did not discuss in details in the
lectures?

no submission!
Group Cohomology – Etudes
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig Sheet 11, July 11, 2019

Exercise 1 (universal coefficients). Compute the following homology groups (whe-


re the groups act trivially on the coefficients):
1. H∗ (S3 ; F2 )
2. H∗ (S3 ; F3 )
3. H∗ (D∞ ; F2 )
4. H∗ (D∞ ; F3 )
Exercise 2 (product groups). Compute the following homology groups (where
the groups act trivially on the coefficients):
1. H∗ (Z/2019 × Z/2019; F3 )
2. H∗ (Z/3 × Z/2019; F3 )
3. H∗ (Z/3 × Z/9; F3 )
4. H∗ (Z/2019 × Z/2019; F2017 ) (be lazy!)
Exercise 3 (algebraic topology). Recall the following terminology/facts:
1. contractibility
2. covering map
3. classification of coverings
4. CW-complex
5. singular homology (and its properties)
6. cellular homology (and its properties)
Hints. In case you don’t know anything about algebraic topology: Don’t panic!
I will quickly review some basics in the lectures. However, it might still be helpful
to browse literature on algebraic topology to get a first impression.
Exercise 4 (summary). Write a summary of Chapter 3.1 (The Hochschild-Serre
spectral sequence), keeping the following questions in mind:
1. What are spectral sequences? How do they work?
2. Which spectral sequences are in your toolbox?
3. Which computational tricks do you know for spectral sequences?
4. How can spectral sequences be used in group (co)homology?
5. Did you check all the little things that we did not discuss in details in the
lectures?

no submission!
Group Cohomology – Etudes
Prof. Dr. C. Löh/Dr. D. Fauser/J. P. Quintanilha/J. Witzig Sheet 12, July 18, 2019

Exercise 1 (classifying spaces for Z). Which of the following spaces are classifying
spaces of Z (when equipped with a suitable CW-structure, etc.)?
1. S 1 × R2019
2. R2 \ {0}
3. R2 \ {0, (1, 0)}
4. R2019 \ {0}
5. S 1 ∨ D2019
6. S 1 × S 2019
Exercise 2 (two-dimensional classifying spaces). Which of the following groups
admit a classifying space of dimension 2 ? Here, Γg denotes the surface group
of genus g and Fn denotes the free group of rank n.
1. Γ2018 ∗ Γ2019 ∗ F2019
2. Γ2018 × Γ2019
3. Z2019
4. Z/2019 ∗ Γ2019
5. Γ2018 ∗ (F2019 × F2018 )
6. Z
Exercise 3 (presentation complexes). Draw the presentation complexes of the
following group presentations. Do you recognise the groups?
1. hx | x2 i
2. hx, y | xi
3. hx, y | [x, y]i
4. hx, y | x2 , y 2 i
Exercise 4 (summary). Write a summary of Chapter 4.1 (Classifying spaces)
keeping the following questions in mind:
1. What is the definition of classifying spaces for groups?
2. What is an example of a functorial construction?
3. What are typical “nice” classifying spaces?
4. Why are classifying spaces useful in group (co)homology?
5. How do classifying spaces compare to projective resolutions?
6. Did you check all the little things that we did not discuss in detail in the
lectures?

no submission!
Group Cohomology – Etudes
Prof. Dr. C. Löh/Dr. D. Fauser/J. P. Quintanilha/J. Witzig Sheet 13, July 25, 2019

Exercise 1 (cohomological dimension). Determine the cohomological dimension


of the following groups.
1. Z/2019 × F2019
2. GL2019 (F2017 )
3. GL2019 (Z)
4. Γ2017 ∗ Γ2018 ∗ Γ2019
5. Γ2017 × Γ2018 × Γ2019
6. Z2019 × F2020
Exercise 2 (sphere actions). Which of the following groups admit a free action
on some sphere?
Z/2019, Z/4 × Z/9, GL2019 (F2017 ), Z
Exercise 3 (summary). Write a summary of Chapter 4.2 (Finiteness conditions)
and Chapter 4.3 (Application: Free actions on spheres), keeping the following
questions in mind:
1. Which finiteness conditions for groups do you know?
2. How are they related?
3. What are typical examples?
4. How can they be computed?
5. Which necessary conditions do you know for finite groups to admit a free
action on a sphere? How does this relate to group (co)homology?
6. How does this work in concrete examples?
7. Did you check all the little things that we did not discuss in detail in the
lectures?
Exercise 4 (exam). Due to a tragic malfunction in space-time you end up as
examiner in an oral exam on Group Cohomology.
1. Which questions will you ask on basic notions?
2. Which questions will you ask on applications?
3. Which questions will you ask on the different views and methods of com-
putation on group (co)homology?
4. Which examples will you discuss during the exam?
Try out your questions on your fellow students!

no submission!
C.16 C. Etudes
D
General Information
Group Cohomology, SS 2019
Organisation
Prof. Dr. C. Löh/D. Fauser/J. P. Quintanilha/J. Witzig April 2019

Homepage. Information and news concerning the lectures, exercise classes, of-
fice hours, literature, as well as the exercise sheets can be found on the
course homepage and in GRIPS:

http://www.mathematik.uni-regensburg.de/loeh/teaching/grouphom ss19
https://elearning.uni-regensburg.de

Lectures. The lectures are on Mondays (10:15–12:00; M102) and on Thursdays


(10:15–12:00; M104).
Basic lecture notes will be provided, containing an overview of the
most important topics of the course. These lecture notes can be found
on the course homepage and will be updated after each lecture. Please
note that these lectures notes are not meant to replace attending the
lectures or the exercise classes!
Exercises. Homework problems will be posted on Mondays (before 10:00) on
the course homepage; submission is due one week later (before 10:00, in
the mailbox).
Each exercise sheet contains four regular exercises (4 credits each)
and more challenging bonus problems (4 credits each).
It is recommended to solve the exercises in small groups; however,
solutions need to be written up individually (otherwise, no credits will
be awarded). Solutions can be submitted alone or in teams of at most
two participants; all participants must be able to present all solutions
of their team.
The exercise classes start in the second week; in this first session,
some basics on categories, homological algebra, and free groups will be
discussed (as on the sheet Etudes 0).
In addition, we will provide etudes that will help to train elementary
techniques and terminology. These problems should ideally be easy
enough to be solved within a few minutes. Solutions are not to be
submitted and will not be graded.
Registration for the exercise classes. Please register for the exercise classes via
GRIPS:

https://elearning.uni-regensburg.de

Please register before Friday, April 26, 2019, 10:00, choosing your
preferred time slot. We will try to fill the groups respecting your pref-
erences.
The distribution will be announced at the beginning of the second
week via GRIPS.

1
Credits/Exam. This course can be used as specified in the commented list of
courses and in the module catalogue.

• Studienleistung: Successful participation in the exercise classes:


50% of the credits (of the regular exercises), presentation of a so-
lution in class, active participation
• Prüfungsleistung: Oral exam (25 minutes), by individual appoint-
ment at the end of the lecture period/during the break.

You will have to register in FlexNow for the Studienleistung and the
Prüfungsleistung (if applicable).
Further information on formalities can be found at:

http://www.uni-regensburg.de/mathematik/fakultaet/studium/studierende-und-studienanfaenger/index.html

Contact.

• If you have questions regarding the organisation of the exercise


classes, please contact Daniel Fauser or Johannes Witzig:

[email protected]
[email protected]
• If you have questions regarding the exercises, please contact your
tutor.
• If you have mathematical questions regarding the lectures, please
contact your tutor or Clara Löh.
• If you have questions concerning your curriculum or the examina-
tion regulations, please contact the student counselling offices or
the exam office:
http://www.uni-regensburg.de/mathematik/fakultaet/studium/ansprechpersonen/index.html

In many cases, also the Fachschaft can help:

http://www-cgi.uni-regensburg.de/Studentisches/FS MathePhysik/cmsms/

2
D.4 D. General Information
Bibliography

Please note that the bibliography will grow during the semester. Thus,
also the numbers of the references will change!

[1] M.F. Atiyah. Elliptic operators, discrete groups and von Neumann al-
gebras, Astérisque, 32–33, pp. 43–72, 1976. Cited on page: 164

[2] T. Bauer. Homologische Algebra und Gruppenkohomologie, lecture


notes, available online at
https://people.kth.se/∼tilmanb/pdf/homalg.pdf Cited on page: 118

[3] C. Bavard. Longueur stable des commutateurs, Enseign. Math. (2),


37(1–2), pp. 109–150, 1991. Cited on page: 97

[4] R. Benedetti, C. Petronio. Lectures on Hyperbolic Geometry, Universi-


text, Springer, 1992. Cited on page: 151

[5] M. Bestvina, N. Brady. Morse theory and finiteness properties of


groups, Invent. Math., 129(3), pp. 445–470, 1997. Cited on page: 163

[6] R. Bieri. Homological dimension of discrete groups, Queen Mary College


Mathematics Notes. Mathematics Department, Queen Mary College,
1976. Cited on page: 163

[7] M. Blank, F. Diana. Uniformly finite homology and amenable groups,


Algebr. Geom. Topol., 15(1), pp. 467–492, 2015. Cited on page: 85

[8] J. Block, S. Weinberger. Aperiodic tilings, positive scalar curvature and


amenability of spaces, J. Amer. Math. Soc., 5(4), pp. 907–918, 1992
Cited on page: 80, 83
E.2 Bibliography

[9] S. Bosch. Algebra, 6th edition, Springer, 2006. Cited on page: 28, 29
[10] J. Brodzki, G.A. Niblo, N. Wright. Pairings, duality, amenability and
bounded cohomology, J. Eur. Math. Soc., 14(5), pp. 1513–1518, 2012.
Cited on page: 79, 90
[11] A. Brooke-Taylor. Products of CW complexes, preprint, available at
arXiv: 1710.05296 [math.GN], 2017. Cited on page: A.12
[12] K.S. Brown. Cohomology of Groups, Graduate Texts in Mathemat-
ics, 82, Springer, 1982. Cited on page: 41, 54, 139, 155, 156, 159, 160,
162, 163, 164, 168
[13] T. Bühler. Exact categories, Expo. Math., 28, pp. 1–69, 2010. Cited on
page: 100
[14] D. Calegari. scl, MSJ Memoirs, 20, Mathematical Society of Japan,
2009. Cited on page: 97, 98
[15] J.W. Cannon, W.J. Floyd, W.R. Parry. Introductory notes on Richard
Thompson’s groups. Enseign. Math. (2), 42(3–4), pp. 215–256, 1996.
Cited on page: 23
[16] J. Cheeger, M. Gromov. L2 -cohomology and group cohomology, Topol-
ogy, 25(2), pp. 189–215, 1986. Cited on page: 164
[17] J.M. Cohen. Zero divisors in group rings, Comm. Algebra, 2, pp. 1–14,
1974. Cited on page: 7
[18] M. Culler. Using surfaces to solve equations in free groups, Topology,
20(2), pp. 133–145, 1981. Cited on page: 96
[19] J.F. Davis, R.J. Milgram. A survey of the spherical space form problem,
Mathematical Reports, 2, Part 2, Harwood Academic Publishers, 1985.
Cited on page: 168
[20] M.W. Davis. The Geometry and Topology of Coxeter Groups, Lon-
don Mathematical Society Monographs, 32, Princeton University Press,
2008. Cited on page: 152
[21] T. Delzant. Sur lanneau dun groupe hyperbolique, C. R. Acad. Sci.
Paris Ser. I Math., 324(4), pp. 381–384, 1997. Cited on page: 7
[22] J. Dodziuk. de Rham-Hodge theory for L2 -cohomology of infinite cov-
erings, Topology, 16(2), pp. 157–165, 1977. Cited on page: 164
[23] A. Dold. Lectures on Algebraic Topology, Springer, 1980. Cited on
page: 138, 143, 166
[24] C.H. Dowker. Topology of metric complexes. American Journal of
Mathematics, 74(3), pp. 557–577, 1952. Cited on page: 147, A.12
Bibliography E.3

[25] C. Druţu, M. Kapovich. Geometric Group Theory, with an appendix


by Bogdan Nica, volume 63 of Colloquium Publications, AMS, 2018.
Cited on page: 68
[26] Tullia Dymarz. Bilipschitz equivalence is not equivalent to quasi-
isometric equivalence for finitely generated groups, Duke Math. J.,
154(3), pp. 509–526, 2010. Cited on page: 85
[27] R. Frigerio. Bounded Cohomology of Discrete Groups, Mathematical
Surveys and Monographs, 227, AMS, 2017. Cited on page: 86, 87, 157
[28] R. Fritsch, R. Piccinini. Cellular structures in topology, Cambridge
Studies in Advanced Mathematics, 19, Cambridge University Press,
1990. Cited on page: 143, 147
[29] D. Gaboriau. Invariants l2 de relations d’équivalence et de groupes,
Publ. Math. Inst. Hautes Études Sci., 95, pp. 93–150, 2002. Cited on
page: 24, 85, 164
[30] D. Gaboriau. A measurable-group-theoretic solution to von Neumann’s
problem, Invent. Math., 177(3), pp. 533–540, 2009. Cited on page: 74
[31] S.I. Gelfand, Y.I. Manin. Methods of Homological Algebra, Springer
Monographs in Mathematics, second edition, Springer, 2003. Cited on
page: 144
[32] R. Geoghegan. Topological Methods in Group Theory, Graduate Texts
in Mathematics, 243, Springer, 2007. Cited on page: 62
[33] E.S. Golod. On nil-algebras and finitely approximable p-groups, Izv.
Akad. Nauk SSSR Ser. Mat., 28, pp. 273–276, 1964. Cited on page: 25
[34] J.A. Grochow, Y. Qiao. Algorithms for group isomorphism via group
extensions and cohomology, SIAM J. Comput., 46(4), pp. 1153–1216,
2017. Cited on page: 41
[35] M. Gromov. Groups of polynomial growth and expanding maps, with
an appendix by Jacques Tits, Publ. Math. Inst. Hautes Études Sci., 53,
pp. 53–78, 1981. Cited on page: 68, 71
[36] M. Gromov. Volume and bounded cohomology. Publ. Math. IHES, 56,
pp. 5–99, 1982. Cited on page: 86, 87, 95, 157
[37] M. Gromov. Hyperbolic groups, Essays in group theory, Math. Sci. Res.
Inst. Publ., 8, pp. 75–263, Springer, 1987. Cited on page: 68, 152, 153,
161
[38] G. Harder. A Gauss-Bonnet formula for discrete arithmetically defined
groups, Ann. Sci. École Norm. Sup., 4(4), pp. 409–455, 1971. Cited on
page: 164
E.4 Bibliography

[39] P. de la Harpe. Topics in Geometric Group Theory, Chicago University


Press, 2000. Cited on page: 68

[40] A. Hatcher. Spectral Sequences in Algebraic Topology, preliminary book,


available online at
http://www.math.cornell.edu/ hatcher/SSAT/SSATpage.html Cited on
page: 118

[41] N. Heuer, C. Löh. The spectrum of simplicial volume, preprint, 2019.


arXiv: 1904.04539 [math.GT] Cited on page: 98

[42] P.J. Hilton, U. Stammbach. A Course in Homological Algebra, Gradu-


ate Texts in Mathematics, 4, second edition, Springer, 1996. Cited on
page: 30, 135

[43] N.V. Ivanov. Foundations of the theory of bounded cohomology, J. So-


viet Math., 37, pp. 1090–1114, 1987. Cited on page: 86, 95, 157

[44] N.V. Ivanov. Notes on the bounded cohomology theory, preprint, avail-
able at arXiv:1708.05150 [math.AT], 2017. Cited on page: 86, 157

[45] V. Kadets. A course in functional analysis and measure theory, trans-


lated from the 2006 Russian edition by Andrei Iacob, Universitext,
Springer, 2018. Cited on page: 90

[46] B. Keller. Derived categories and their uses, Handbook of Algebra, 1,


pp. 671–701, Elsevier/North-Holland, 1996. Cited on page: 111

[47] R.C. Kirby, L.C. Siebenmann. Foundational essays on topological man-


ifolds, smoothings, and triangulations, with notes by J. Milnor and
M. Atiyah, Annals of Mathematics Studies, 88, Princeton University
Press, University of Tokyo Press, 1977. Cited on page: 167

[48] M. Lackenby. Expanders, rank and graphs of groups, Israel J. Math.,


146, pp. 357–370, 2005. Cited on page: 24

[49] J.M. Lee. Riemannian Manifolds. An Introduction to Curvature, Grad-


uate Texts in Mathematics, 176, Springer, 1997. Cited on page: 152

[50] C. Löh. Simplicial Volume, Bull. Man. Atl., 7–18, 2011. Cited on
page: 86, 87, 157

[51] C. Löh. Algebraische Topologie I – Elementare Homotopietheorie,


lecture notes, Universität Regensburg, 2014.
http://www.mathematik.uni-regensburg.de/loeh/teaching/topologie1 ss13/lecture notes.pdf
Cited on page: 147

[52] C. Löh. A note on bounded-cohomological dimension of discrete groups,


J. Math. Soc. Japan, 69(2), pp. 715–734, 2017. Cited on page: 90
Bibliography E.5

[53] C. Löh. Geometric Group Theory. An Introduction, Universitext,


Springer, 2018. Cited on page: 25, 31, 56, 62, 65, 68, 71, 72, 73, 74,
152, 161, A.3, A.4, A.6

[54] C. Löh. Ergodic theoretic methods in group homology. A minicourse


on L2 -Betti numbers in group theory, MSRI Summer graduate school
on Random and arithmetic structures in topology, 2019.
http://www.mathematik.uni-r.de/loeh/teaching/l2 msri/lecture notes.pdf
Cited on page: 85

[55] W. Lück. L2 -Invariants: Theory and Applications to Geometry


and K-Theory, Ergebnisse der Mathematik und ihrer Grenzgebiete,
3. Folge, 44, Springer, 2002. Cited on page: 7, 164

[56] R.C. Lyndon. Cohomology theory of groups with a single defining re-
lation, Ann. of Math. (2), 52, pp. 650–665, 1950. Cited on page: 153

[57] S. Matsumoto, S. Morita. Bounded cohomology of certain groups of


homeomorphisms, Proc. Amer. Math. Soc., 94(3), pp. 539–544, 1985.
Cited on page: 90

[58] J. McCleary. A user’s guide to spectral sequences, second edition, Cam-


bridge Studies in Advanced Mathematics, 58, Cambridge University
Press, 2001. Cited on page: 118

[59] J. Milnor. Groups which act on S n without fixed points, Amer. J.


Math., 79, pp. 623–630, 1957. Cited on page: 165, 168

[60] E.E. Moise. Affine structures in 3-manifolds. V. The triangulation the-


orem and Hauptvermutung, Ann. of Math. (2), 56, pp. 96–114, 1952.
Cited on page: 167

[61] N. Monod. Continuous Bounded Cohomology of Locally Compact


Groups, Lecture Notes in Mathematics, 1758, Springer, 2001. Cited
on page: 86

[62] B.B. Newman. Some results on one-relator groups, Bull. Amer. Math.
Soc., 74, pp. 568–571, 1968. Cited on page: 153

[63] A.Ju. Olshanskii. On the question of the existence of an invariant mean


on a group, Uspekhi Mat. Nauk, 35 (4), pp. 199–200, 1980. Cited on
page: 74

[64] D. Osin. Rank gradient and torsion groups, Bull. Lond. Math. Soc.,
43(1), pp. 10–15, 2011. Cited on page: 24

[65] D.S. Passman. Group rings, crossed products and Galois theory. CBMS
Regional Conference Series in Mathematics, 64. AMS, 1986. Cited on
page: 7
E.6 Bibliography

[66] A.L.T. Paterson. Amenability, Mathematical Surveys and Mono-


graphs, 29, AMS, 1988. Cited on page: 72, 73

[67] A. Putman. One-relator groups, notes


https://www3.nd.edu/∼andyp/notes/OneRelator.pdf Cited on page: 153

[68] F. Quinn. Ends of maps. III. Dimensions 4 and 5, J. Differential Geom.,


17(3), pp. 503–521, 1982. Cited on page: 167

[69] E. Riehl. Categorical Homotopy Theory, New Mathematical Mono-


graphs, 24, Cambridge University Press, 2015. Cited on page: 111,
117

[70] J.J. Rotman. An Introduction to the Theory of Groups, fourth edition,


Graduate Texts in Mathematics, 148, Springer, 1999. Cited on page: 31

[71] J.J. Rotman. An Introduction to Homological Algebra, second edition,


Universitext, Springer, 2009. Cited on page: 135

[72] W. Rudin. Real and Complex Analysis, third edition. McGraw-Hill


Book Co., 1987. Cited on page: 84

[73] V. Runde. Amenability, Springer Lecture Notes in Mathematics, 1774,


Springer, 2002. Cited on page: 72

[74] R. Sauer. L2 -Betti numbers of discrete measured groupoids, Internat.


J. Algebra Comput., 15(5–6), pp. 1169–1188, 2005. Cited on page: 85

[75] R. Sauer. Amenable covers, volume and L2 -Betti numbers of aspherical


manifolds, J. reine angew. Math., 636, 47–92, 2009. Cited on page: 157

[76] J.-P. Serre. Cohomologie des groupes discrets, Prospects in mathemat-


ics (Proc. Sympos., Princeton Univ., Princeton, 1970), Ann. of Math.
Studies, 70, Princeton University Press, pp. 77–169, 1971. Cited on
page: 164

[77] B. Seward. Burnside’s Problem, spanning trees and tilings, Geom.


Topol., 18(1), pp. 179–210, 2014. Cited on page: 74

[78] P.A. Smith. Permutable periodic transformations, Proc. Nat. Acad. Sci.
U.S.A., 30, pp. 105–108, 1944. Cited on page: 165

[79] T. Soma. Bounded cohomology and topologically tame Kleinian groups,


Duke Math. J., 88(2), pp. 357–370, 1997. Cited on page: 95

[80] E. Specker. Die erste Cohomologiegruppe von Überlagerungen und Ho-


motopieeigenschaften drei-dimensionaler Mannigfaltigkeiten, Comm.
Math. Helv., 23, pp. 303–333, 1949. Cited on page: 62
Bibliography E.7

[81] J.R. Stallings. A finitely presented group whose 3-dimensional integral


homology is not finitely generated, Amer. J. Math., 85, pp. 541–543,
1963. Cited on page: 163

[82] N.E. Steenrod. A convenient category of topological spaces, Michigan


Math. J., 14(2), pp. 133–152, 1967. Cited on page: 148, A.12
[83] J. Stillwell. Classical Topology and Combinatorial Group Theory, sec-
ond edition, Graduate Texts in Mathematics, 72, Springer, 1993. Cited
on page: 163

[84] R.G. Swan. Groups of cohomological dimension one, Journal of Algebra,


12, pp. 585–601, 1969. Cited on page: 161
[85] T. tom Dieck. Algebraic Topology, European Mathematical Society,
2008. Cited on page: 149, A.12

[86] P. Wadler. Theorems for free! FPCA ’89, Proceedings of the fourth in-
ternational conference on Functional programming languages and com-
puter architecture, pp. 347–359, London, 1989. Cited on page: 135
[87] C. Weibel. An Introduction to Homological Algebra, Cambridge Studies
in Advanced Mathematics, 38, Cambridge University Press, 1995. Cited
on page: 41, 100, 104, 110, 111, 114, 115, 116, 117, 118, 122, 123, 135,
138
[88] G.W. Whitehead. Elements of Homotopy Theory, Graduate Texts in
Mathematics, 61, Springer, 1978. Cited on page: A.14
[89] J.H.C. Whitehead. On C 1 -complexes, Ann. of Math. (2), 41, pp. 809–
824, 1940. Cited on page: 152
[90] K. Whyte. Amenability, bi-Lipschitz equivalence, and the von Neumann
conjecture, Duke Math. J., 99(1), pp. 93–112, 1999. Cited on page: 74,
85
[91] D. Zhuang. Irrational stable commutator length in finitely presented
groups, J. Mod. Dyn., 2(3), pp. 499–507, 2008. Cited on page: 98
Deutsch → English

A
Abbildungskegel mapping cone 113
abgeleiteter Funktor derived functor 101
amenabel, mittelbar amenable 72
Auflösung resolution 42
Augmentierung augmentation 12
azyklisch acyclic 123

B
Bar-Auflösung bar resolution 12
beschränkte Kohomologie bounded cohomology 87
bigraduiert bigraded 118
Bilipschitz-Äquivalenz bilipschitz equivalence 69
Bilipschitz-Einbettung bilipschitz embedding 69

C
CW-Komplex CW-complex A.11
CW-Struktur CW-structure A.11

D
Defekt defect 90
Defekt deficiency 32
degenerierende Spektralsequenz degenerating
spectral sequence 119
derivierte Kategorie derived category 111
derivierter Funktor derived functor 101
Doppelkomplex double complex 122
Dictionary E.9

E
Ein-Relator-Gruppe one-relator group 153
Erweiterung extension 32
Erzeugendensystem generating set 30
Erzeuger generator 30
Eulercharakteristik Euler characteristic 164
exakter Funktor exact functor 100

F
Følner-Folge Følner sequence 73
Flächengruppe surface group 151
freie Auflösung free resolution 42
freies amalgamiertes Produkt amalgamated free product A.3
freies Erzeugendensystem free generating set A.6
freies Produkt free product A.3
Fundamentalklasse fundamental class 80

G
geometrische Dimension geometric dimension 159
getwistete Koeffizienten twisted coefficients 155
gleichmäßig diskret und von beschränkter Geometrie UDBG 75
gleichmäßig endlich uniformly finite 76
gleichmässig nah uniformly close 70
Gruppenhomologie group homology 1
Gruppenkohomologie group cohomology 1
Gruppenpräsentation presentation of a group 30
Gruppenring group ring 6

H
homologische Algebra homological algebra A.7
homologischer ∂-Funktor homological ∂-functor 101
Hufeisenlemma horseshoe lemma A.8
hyperbolische Gruppe hyperbolic group 152

I
Induktion induction 59
Invarianten invariants 9
Involution involution 8

K
klassifizierender Raum classifying space 144
kohomologische Dimension cohomological dimension 158
Koinduktion coinduction 60
Koinvarianten coinvariants 9

L
Lefschetzzahl Lefschetz number 166
Lokalisierung localisation 113
E.10 Dictionary

M
Mittel mean 72
mittelbar amenable 72
Modulkategorie module category 100

N
normierter Ring normed ring 76

P
Präsentations-Komplex presentation complex 147
projektive Auflösung projective resolution 42
projektiver Modul projective module 41

Q
Quasi-Isometrie quasi-isometry 70
quasi-isometrische Einbettung quasi-isometric embedding 70
Quasiisomorphismus quasi-isomorphism 112
Quasimorphismus quasi-morphism 90
Quaternionengruppe quaternion group 54

R
Rang rank 23, A.6
rechts-exakter Funktor right exact functor 100
reduziertes Wort reduced word A.4
Restriktion restriction 58

S
Schur-Multiplikator Schur multiplier 41
Seite page 118
simpliziale Auflösung simplicial resolution 11
Skelett skeleton A.11
Spektralsequenz, Spektralfolge spectral sequence 118

T
torsionsfrei torsion-free 25
Torsionsgruppe torsion group 25
Transfer transfer 63

U
universelles Koeffiziententheorem universal coefficient theorem 133

V
Verbindungsmorphismus connecting morphism 101

W
Wort word A.3
Dictionary E.11

Wortmetrik word metric 68

Z
Zelle cell A.11
E.12 Dictionary
Dictionary E.13

English → Deutsch

A
acyclic azyklisch 123
amalgamated free product freies amalgamiertes Produkt A.3
amenable amenabel, mittelbar 72
augmentation Augmentierung 12

B
bar resolution Bar-Auflösung 12
bigraded bigraduiert 118
bilipschitz embedding Bilipschitz-Einbettung 69
bilipschitz equivalence Bilipschitz-Äquivalenz 69
bounded cohomology beschränkte Kohomologie 87

C
cell Zelle A.11
classifying space klassifizierender Raum 144
cohomological dimension kohomologische Dimension 158
coinduction Koinduktion 60
coinvariants Koinvarianten 9
connecting morphism Verbindungsmorphismus 101
CW-complex CW-Komplex A.11
CW-structure CW-Struktur A.11

D
defect Defekt 90
deficiency Defekt 32
degenerating spectral sequence degenerierende
Spektralsequenz 119
E.14 Dictionary

derived category derivierte Kategorie 111


derived functor abgeleiteter Funktor,
derivierter Funktor 101
double complex Doppelkomplex 122

E
Euler characteristic Eulercharakteristik 164
exact functor exakter Funktor 100
extension Erweiterung 32

F
Følner sequence Følner-Folge 73
free generating set freies Erzeugendensystem A.6
free product freies Produkt A.3
free resolution freie Auflösung 42
fundamental class Fundamentalklasse 80

G
generating set Erzeugendensystem 30
generator Erzeuger 30
geometric dimension geometrische Dimension 159
group cohomology Gruppenkohomologie 1
group homology Gruppenhomologie 1
group ring Gruppenring 6

H
homological ∂-functor homologischer ∂-Funktor 101
homological algebra homologische Algebra A.7
horseshoe lemma Hufeisenlemma A.8
hyperbolic group hyperbolische Gruppe 152

I
induction Induktion 59
invariants Invarianten 9
involution Involution 8

L
Lefschetz number Lefschetzzahl 166
localisation Lokalisierung 113

M
mapping cone Abbildungskegel 113
mean Mittel 72
module category Modulkategorie 100

N
normed ring normierter Ring 76
Dictionary E.15

O
one-relator group Ein-Relator-Gruppe 153

P
page Seite 118
presentation complex Präsentations-Komplex 147
presentation of a group Gruppenpräsentation 30
projective module projektiver Modul 41
projective resolution projektive Auflösung 42

Q
quasi-isometric embedding quasi-isometrische Einbettung 70
quasi-isometry Quasi-Isometrie 70
quasi-isomorphism Quasiisomorphismus 112
quasi-morphism Quasimorphismus 90
quaternion group Quaternionengruppe 54

R
rank Rang 23, A.6
reduced word reduziertes Wort A.4
resolution Auflösung 42
restriction Restriktion 58
right exact functor rechts-exakter Funktor 100

S
Schur multiplier Schur-Multiplikator 41
simplicial resolution simpliziale Auflösung 11
skeleton Skelett A.11
spectral sequence Spektralsequenz, Spektralfolge118
surface group Flächengruppe 151

T
torsion group Torsionsgruppe 25
torsion-free torsionsfrei 25
transfer Transfer 63
twisted coefficients getwistete Koeffizienten 155

U
UDBG gleichmäßig diskret und
von beschränkter Geometrie 75
uniformly close gleichmässig nah 70
uniformly finite gleichmäßig endlich 76
universal coefficient theorem universelles
Koeffiziententheorem 133

W
word Wort A.3
word metric Wortmetrik 68
E.16 Dictionary
Symbols

Symbols G Ban category of left


|·| cardinality, Banach G-modules, 86
∩ intersection of sets, BG standard simplicial
∪ union of sets, model of G, 142
t disjoint union of sets, BHomG space of morphisms of
⊂ subset relation Banach G-modules, 86
(equality is
permitted), C
V# bounded dual of V , 89
#(w, g) number of occurrences C set of complex
of the word w in g, 92 numbers,
·G G-coinvariants, 9 Cay(G, S) Cayley graph of G
·G G-invariants, 9 with respect to the
⊗G tensor product of left generating set S, 68
modules over ZG, 8 cd cohomological
F free product, A.4 dimension, 158
× cartesian product, C∗ (G) the simplicial
resolution of G, 11
C∗ (G; A) simplicial complex
A of G with coefficients
AG G-coinvariants of A, 9 in A, 14
AG G-invariants of A, 9 C ∗ (G; A) simplicial cochain
complex of G with
coefficients in A, 15
B Cb∗ (G; V ) (simplicial) bounded
Ban category of Banach cochain complex of G
spaces (and bounded with coefficients in V ,
operators), 87 87
E.18 Symbols

C∗R (G) real analogue H


of C∗ (G), 86 Hb∗ (G; V ) bounded cohomology
χ(G) Euler characteristic of G with coefficients
of G, 164 in V , 87
CoindG H coinduction functor, H∗uf ( · ; R) uniformly finite
60 homology, 77
corG
H homology: induced by Hn (G; A) group homology of G
inclusion; cohomology: with coefficients in A,
transfer, 63 14
C∗uf ( · ; R) uniformly finite chain H n (G; A) group cohomology
complex, 76, 77 of G with coefficients
in A, 15
D HomG module of
ZG-homomorphisms,
D∞ infinite dihedral
8
group, 65
Dn dihedral group of the
n-gon, B.7 I
diam diameter, 76 IndGH induction functor, 59
dS word metric for the Inv switching left to right
generating set S, 68 ZG-module structure,
8
E
e neutral element in a L
group, 6 Λ(f ) Lefschetz number
EG universal covering of of f , 166
the standard
simplicial model of G, M
142
Ma right multiplication
e(G) number of ends of G,
by a, 47
62
R Mod category of left
R-modules,
G
Γg surface group of N
genus g, 151
N set of natural
gd geometric dimension, numbers: {0, 1, 2, . . . },
159
GroupBan* the domain category
G Norm category of left
for bounded normed G-modules, 86
cohomology, 86
GroupMod domain category of
group homology, 10 P
GroupMod* domain category of P∗  ε concatenated
group cohomology, 10 sequence, 42
Symbols E.19

ϕ∗ B ZG-module structure trG


H transfer in
on B via ϕ, 10 (co)homology, 63
ψw counting trZ trace on the free part,
quasi-morphism 166
associated with w, 92
X
Q [X]R fundamental class
Q set of rational in H0uf (X; R), 80
numbers,
QM(G) space of Z
quasi-morphisms
on G, 90 Z set of integers,
QM0 (G) space of trivial ZG integral group ring
quasi-morphisms of G, 6
on G, 90
QM(G) space of homogeneous
quasi-morphisms
on G, 90

R
R set of real numbers,
ResGH restriction functor, 58
resG
H cohomology:
restricition; homology:
transfer, 63
rk rank of a group, 23
rkZ rank of Z-modules, 23
R Mod category of left
R-modules,

S
Σg oriented closed
connected surface of
genus g, 151
supp support, 76

T
Tor derived functor of the
tensor product, 107
Tot C total complex
associated with the
double complex C∗∗ ,
122
E.20 Symbols
Index

A relatively injective, B.13


Abelianisation, 20 bar resolution, 12, 43
acronyms, B.11 Bavard duality, 97
acyclic, 123 bigraded module, 118
algebraic Künneth theorem, 135 bilipschitz embedding, 69
algebraic universal coefficient the- bilipschitz equivalence, 69, 70, B.10
orem, 133 Blorx, B.16
amalgamated free product, A.3 bounded cochain complex, 87
amenable, 72 bounded cohomology, 86, 87
Abelian, 72 amenability, 88
cohomological characterisation, comparison map, 88
88 degree 0, 88
Følner sequence, 73 degree 1, 88
free group, 73 degree 2, 93
group, B.9 duality principle, B.12
homological characterisation, free groups, 95
80 quasi-morphisms, 93
inheritance, 74 topological view, 156
QI-invariance, 74, 85 trivial group, 88
quasi-morphism, 94 bounded dual, 87, 89
scl, 97 bounded geometry, 75
augmentation
simplicial, 12 C
augmentation ideal, B.2 canonically bounded filtration, 122
Cartan-Hadamard theorem, 152
B category
Banach module, 86 classifying space, 144, B.15
E.22 Index

derived, 111, 114 surface group, 151


exact, B.12 transfer, 156
for bounded cohomology, 86 uniqueness, 146
for group (co)homology, 10 cochain complex, 15
homotopy category, 114 for bounded cohomology, 87
localisation, 113 for group cohomology, 15
module, 100 cohomological dimension, 158, 159,
Cayley graph, 68 161
cell, A.11 description, 158
cell complex, see CW-complex inheritance properties, 160
central extension, 41 torsion element, 161
centre kills trick, B.14 cohomological spectral sequence, 121
certain groups of homeomorphisms, cohomology, 15
B.3 amenability, 88
chain complex classification of extensions, 33
for group (co)homology, 14, of finite cyclic groups, 49
15 of finite groups, 64
quasi-isomorphism, 112, B.12 of free groups, 56, B.6
tensor product, 14 of subgroups, 61
uniformly finite chains, 76, 77 of Homeo+ (S 1 ), B.5
character, 26 of Z, B.6
CIA (Cohomological Intelligence Agency), of Z/3, B.5
B.16 of Z2 , B.5
classical transfer, 65 torsion, 65
classification of certain p-groups, via classifying spaces, 153
50, 53 via injective resolutions, 110
classification of extensions, 33 via Ext, 110
functoriality, 38 with group ring coefficients,
classifying space, 142, 144 62, B.6
constructions, 147 cohomology of a group, see group
examples, 149 cohomology
free product, 147 coinduction, 60
Heisenberg group, B.15 coinvariants, 9
homology, 154 commensurability, B.16
hyperbolic group, 152 commutator length, 95, B.11
inductive construction, 147 comparison map, 88
manifold of non-positive sec- complex
tional curvature, 152 CW-complex, A.11
of a category, 144, B.15 Rips complex, 152
one-relator group, 153 conjecture
product, 147 Eilenberg-Ganea problem, 162
recognition, 145 Kaplansky, 7
standard simplicial model, 142, conjugation action, B.14
143, 145 on the kernel, 32
subgroup, 147 connecting morphism, 101, A.7
Index E.23

convergence of a spectral sequence, E


119, 121 Eilenberg-Ganea problem, 162
coproduct Eilenberg-MacLane space, 149
of groups, A.5 ends of a group, 62
corestriction, 63 equivalence of extensions, 32
counting quasi-morphism, 92 Euler characteristic, 164, 166, B.16
cup-product, 139
exact category, B.12
CW-complex, A.11
exact functor, 100
dimension, A.12
derived functor, 102
finite, A.12
Ext, 110
of finite type, A.12
and group cohomology, 110
product, A.12
symmetry, 110
relative, A.11
extension, 32
skeleton, A.11
classification, 33
CW-structure, A.11
conjugation action, 32
cyclic Galois extension, 28
extension problem, 120
cyclic group
(co)homology, 49, 56
recognition, 53 F
F, see type F
D filtration
defect, 90 canonically bounded, 122
deficiency, 32 spectral sequence, 122
surface group, 155, B.15 finitary symmetric group, B.3
degenerating spectral seqeunce, 119 finite CW-complex, A.12
derived category, 111, 114 finite type, A.12
derived functor, 116 Følner sequence, 73
derived functor, 100, 101 FP, see type FP
derived category, 116 free actions on spheres, 164, 165
existence, 102 free group, A.3
of exact functors, 102 (co)homology, 56
total, 116 classifying space, 150
uniqueness, 102 non-amenability, 73
dihedral group, B.7 rank, A.6
(co)homology, 65, 127 scl, 97, 98
dimension universal property, A.3
cohomological, 158 free product, A.4
CW-complex, A.12 homology, 156
geometric, 159 free resolution, 42
dimension shifting, 106 Freyd-Mitchell embedding theorem,
double complex, 122 A.7
spectral sequences, 122 functor
total complex, 122 derived, 100, 101
dual, 87, 89 exact, 100
duality principle, B.12 right exact, 100
E.24 Index

fundamental class presentation, 30


uniformly finite homology, 80, quasi-isometric, 71
B.9 rank, 23
rank gradient, 23
G surface group, 151
Galois extension torsion, 25
cyclic, 28 torsion-free, 25
Hilbert 90, 28 types, 162, 163
norm, 28 uniformly finite homology, 79
generalised quaternion group, 54, unique product, B.2
164 word metric, 68
geometric dimension, 159, 161 group (co)homology
inheritance properties, 160 basic view, 5
torsion element, 161 derived view, 99
geometric group theory, 68 domain category, 10
Gromov-hyperbolic group, see hy- geometric view, 67
perbolic group topological view, 141
Grothendieck spectral sequence, 123 via classifying space, 154, 155
group via derived functors, 108, 110
amalgamated free product, A.5 via projective resolutions, 47
amenable, 72, 73, B.9 group cohomology, 15
bilipschitz equivalent, 70 amenability, 88
bounded cohomology, 86 applications, 2
Cayley graph, 68 classification of extensions, 33
certain homeomorphisms, B.3 in degree 0, 18
classifying space, 142 in degree 1, 24
cohomology, 15 in degree 2, 33
commensurable, B.16 of finite cyclic groups, 49
coproduct, A.5 of finite groups, 64
deficiency, 32 of free groups, 56, B.6
dihedral group, B.7 of subgroups, 61
Euler characteristic, 164 of Homeo+ (S 1 ), B.5
extension, 32 of Z, B.6
finitary symmetric, B.3 of Z/3, B.5
free, A.3 of Z2 , B.5
free product, A.4 product, 139
generalised quaternion, 54 via classifying spaces, 153
Heisenberg, B.3 via injective resolutions, 110
homeomorphisms, B.5 via projective resolutions, 47
homology, 14 via Ext, 110
hyperbolic, 152 Why?, 2
infinite dihedral, 65, B.7 with group ring coefficients,
modular, 65 B.6
one-relator group, 153 group extension
perfect, 22, 23 central, 41
Index E.25

classification, 33 homology
functoriality, 38 long exact sequence, A.7
trivial, 37 of finite cyclic groups, 49
universal central, 40 of finite groups, 64
group homology, 14, B.10 of free groups, 56
amenability, 80 of subgroups, 61
conjugation action, B.14 of Z, B.6
free product, 156 of Z/p × Z/p, 137
Heisenberg group, B.14 of Z × Z, 136
in degree 0, 18 of Z2 , B.6
in degree 1, 20 of D∞ , 127
in degree 2, 30, 129 of S3 , 126
of finite cyclic groups, 49 semi-norm, B.12
of finite groups, 64 torsion, 65
of free groups, 56 uniformly finite, 77, 78
of subgroups, 61 via Tor, 108
of Z, B.6 via classifying spaces, 153, 154
of Z2 , B.6 homology of a group, see group ho-
of D∞ , 127 mology
of S3 , 126 homotopy category, 114
surface group, 155, B.15 as localisation, 114
universal coefficient theorem, Hopf’s formula, 30, 129
135 horseshoe lemma, A.8
via Tor, 108 hyperbolic group, 152
via classifying spaces, 153
via projective resolutions, 47 I
group ring, 6
of cyclic groups, B.2 induction, 59
trace, B.2 infinite dihedral group, 65, B.7
universal property, 7 injective module, B.13
invariant mean, 72
H invariants, 9
involution, 8
Heisenberg group, B.3 ISSS (InterStellar Spectral Sequence
Hilbert 90, 25, 28 Station, B.16
Hochschild-Serre spectral sequence,
123, B.14
K
examples, 125
naturality, 125 Kaplansky conjecture, 7, B.2
homeomorphism group, B.5 Künneth spectral sequence, 132
homogeneous quasi-morphism, 90, Künneth theorem
91 algebraic, 135
homogenisation, 91, B.11 cohomology, 138
homological algebra, A.7 group cohomology, 138
homological ∂-functor, 101 group homology, 136
homological spectral sequence, 118 topological, 138
E.26 Index

L of an operator, 86
lattice, 162 normed G-module, 86
Lefschetz fixed point theorem, 166 normed ring, 76
Lefschetz number, 166 number of ends, 62
left-invariant mean, 72
Legendre symbol, 65, B.7 O
library, 3 one-relator group, 153
localisation, 113 open cell, A.11
at chain homotopies, 114 orbit equivalence, B.10
at quasi-isomorphisms, 114
long exact sequence P
group homology, 109
page, 118
homology (algebraic), A.7
page-turning isomorphism, 118
Lyndon-Hochschild-Serre spectral
perfect group, 22, 23
sequence, see Lyndon-Hochschild-
periodic resolution, 165
Serre spectral sequence
Ponzi scheme, 83, B.9
presentation, 30
M
presentation complex, 147
mapping cone, 113, B.13 problem
trick, 113, B.13 amenability of Thompson’s group,
mean, 72, B.11 74
measure equivalence, B.10 Eilenberg-Ganea, 162
metric space extension, 120
bounded geometry, 75 second scl-gap, 98
UDBG, 75 space form, 164
uniformly discrete, 75 von Neumann, 74
model of classifying space, see clas- product structure, 139
sifying space projective module, 41
modular group, 65 projective resolution, 42, 45
module bar, 43
bigraded, 118 periodic, 165
coinduction, 60 quasi-isomorphism, 112
induction, 59 simplicial, 43
injective, B.13 uniqueness, 46
over ZG, 7 via classifying space, 153
projective, 41 via sphere action, 165
restriction, 58
module category, 100 Q
morphism
quasi-isometric embedding, 70, B.7
of Banach G-modules, 86
quasi-isometry, 70, 71, B.10
alternative characterisation, 70
N quasi-isometry invariant, 71
norm quasi-isomorphism, 112, B.12
in Galois theory, 28 quasi-morphism, 90, B.12
Index E.27

amenable group, 94 semi-norm on homology, 87, B.12


bounded cohomology, 93 Shapiro lemma, 61
counting, 92 simplicial augmentation, 12
defect, 90 simplicial classifying space, 142
free group, 92 simplicial resolution, 11, 43
homogeneous, 90, 91 simplicial volume, 87, 98
trivial, 90 SIS (Secret Invariants Service), B.16
quaternion group, 164 skeleton, A.11
space form problem, 164
R spectral sequence, B.14
cohomological, 121
rank convergence, 119, 121
of a Z-module, 23 degeneration, 119
of a free group, A.6 examples, 125
of a group, 23 Grothendieck, 123
rank gradient, 23 Hochschild-Serre, 123
reduced word, A.4 homological, 118
relative CW-complex, A.11 ∞-page, 119
relatively injective Banach mod- Künneth, 132
ule, B.13 of a double complex, 122
resolution, 42 of a filtration, 122
bar, 12, 43 page, 118
free, 42 stepping through, 120
periodic, 165 sphere action, 164
projective, 42, 45 stable commutator length, 95, B.12
simplicial, 11, 43 standard simplicial model, 142, 143,
topological, 44, 153 145
restriction, 58, 63 subgroup distortion, 71
right exact functor, 100 surface group, 151, 155, B.15
right module category, 100 symmetric group
ring (co)homology, 126
group ring, see group ring finitary, B.3
normed, 76
Rips complex, 152 T
rotation number, B.11
tail lemma, 81
tensor product, 107
S
of a chain complex with a mod-
scheme ule, 14
Ponzi, B.9 of chain complexes, 132
Schur multiplier, 41 of left ZG-modules, 8
scl, 98 theorem
amenable groups, 97 Bavard duality, 97
free groups, 97, 98 Cartan-Hadamard, 152
gap, 98 classification of certain p-groups,
transcendental, 98 50, 53
E.28 Index

classification of extensions, 33 type FP, 163


dimension shifting, 106
free actions on spheres, 165 U
Freyd-Mitchell, A.7
UDBG space, 75
fundamental theorem of group
uniformly close, 70
(co)homology, 47
uniformly discrete, 75
fundamental theorem of ho- uniformly finite chain complex, 76
mological algebra, 45 functoriality, 77
Hilbert 90, 25, 28 uniformly finite homology, 77, B.10
Hopf’s formula, 30, 129 degree 0, 80
horseshoe lemma, A.8 functoriality, 77, 78
Künneth theorem, 135, 136 fundamental class, 80, B.9
Lefschetz fixed point theorem, of groups, 79
166 via group homology, 79
linear independence of char- unique product groups, B.2
acters, 26 universal central extension, 40
long exact homology sequence, universal coefficient theorem, 135
A.7 algebraic, 133
Shapiro lemma, 61 cohomology, 138
tail lemma, 81 universal property
transfer, 64 free generating set, A.6
universal coefficient theorem, free group, A.3
133, 135 group ring, 7
Whitehead, A.14 localisation, 113
Tor, 107 total derived functor, 116
an group homology, 108
symmetry, 107 V
torsion group, 25, B.12
torsion-free group, 25 view
total complex, 122 basic, 5
total derived functor, 116 derived, 99
trace geometric, 67
on Z-modules, 166 topological, 141
on group ring, B.2 von Neumann problem, 74
transfer, 63, 64
bounded cohomology, 89 W
classical, 65 weak isomorphism, see quasi-isomorphism
Legendre symbol, 65 Whitehead theorem, A.12
topologically, 156 Why group cohomology?, 2
trick word
centre kills, B.14 reduced, A.3
mapping cone, 113 word metric, 68
trivial quasi-morphism, 90 visualisation, 68
twisted coefficients, 155 word-hyperbolic group, see hyper-
type F, 162, 163 bolic group
Index E.29

Z
ZG-module, 7
coinvariants, 9
Hom, 8
invariants, 9
tensor product, 8
trivial, 8

You might also like