An Eilenberg–Ganea Phenomenon for Actions with

Virtually Cyclic Stabilisers

Martin G. Fluch Ian J. Leary
Bielefeld University School of Mathematics
PO Box 100131 University of Southampton
33501 Bielefeld Southampton SO17 1BJ
Germany United Kingdom
[email protected] [email protected]

October 4, 2012

Abstract
In dimension 3 and above, Bredon cohomology gives an accurate
purely algebraic description of the minimal dimension of the classify-
ing space for actions of a group with stabilisers in any given family
of subgroups. For some Coxeter groups and the family of virtually
cyclic subgroups we show that the Bredon cohomological dimension
is 2 while the Bredon geometric dimension is 3.

1 Introduction and Preliminaries

For a discrete group G, a family of subgroups F is a non-empty collection
of subgroups of G that is closed under conjugation and taking subgroups.
If F is a family of subgroups of G then a model for EF G, the classifying
space for G-actions with stabilisers in F, is a G-CW-complex X such that
for H ≤ G, the fixed point set X H is empty if H ∈ / F and is contractible
if H ∈ F. For any G and F there is always a model for EF G and it is unique
up to equivariant homotopy.
In the case when F consists of just the trivial group, EF G is the same
thing as EG, the universal cover of an Eilenberg–Mac Lane space for G. In
the case when F is the family Ffin (G) of all finite subgroups of G (respec-
tively the family Fvc (G) of all virtually cyclic subgroups of G) we write EG
(respectively EG) for EF G. The minimal dimension of any model for EF G
is denoted by gdF G and is called the Bredon geometric dimension of G.
 Homological algebra over the group ring ZG can be used to study models
for EG, and Bredon cohomology is the natural generalisation for studying
models for EF G. In Bredon cohomology the orbit category OF G replaces
the group G. The orbit category OF G is the category with objects the G-
sets G/H with H ∈ F and G maps as morphisms. A (right) OF G-module is
then a contravariant functor from the orbit category OF G to the category
of abelian groups. In the case when F consists of just the trivial group, OF G
is a category with one object and morphism set G and OF G-modules are
the same as ZG-modules.
The category of OF G-modules is an abelian category with enough pro-
jectives. The Bredon cohomological dimension cdF G is defined to be the
projective dimension of the trivial OF G-module Z, which takes the value Z
on any object of OF G and which maps any morphism to the identity.
The derived functors of the morphism functor in the category of Bre-
don modules over OF G are denoted by Ext∗F (−, −). The Bredon coho-
mology groups of G with coefficients the OF G-module M are the abelian
groups HF∗ (G; M ) = Ext∗F (Z; M ). For details on Bredon cohomology we
refer to [12] or [9].
If the family F consists of the trivial subgroup only, then gdF G is the
minimal dimension gd G an Eilenberg–Mac Lane space for G can have. If F
is the family Ffin (G) (respectively Fvc (G)) then we use the notation gd G
(respectively gd G) for gdF G.
As in the classical case a model for EF G gives rise to a resolution of
the trivial OF G-module Z by projective OF G-modules. Therefore cdF G ≤
gdF G in general. If cdF G ≥ 3, then cdF G = gdF G. In the classical case,
that is when F = {1} consists only of the trivial subgroup, this is due to
Eilenberg–Ganea [7]. For F = Ffin (G) this was proved in [12] and this proof
generalises to arbitrary families F, cf. Theorem 0.1 in [13, p. 294]. In the
classical case it is well known that cdF G = 0 implies gdF G = 0 and for
general families this implication follows from Lemma 2.5 in [16, p. 265].
In the classical case, the statement that the cohomological and geomet-
ric dimension always agree is known as the Eilenberg–Ganea Conjecture.
Since the work of Stallings [14] and Swan [15] implies that cd G = 1 if
and only if gd G = 1, this conjecture can only be falsified by a group G
with cd G = 2 but gd G = 3.
In [1] right-angled Coxeter groups W such that cd W = 2 but gd W = 3
were exhibited. We show here that some, but not all, of these examples
have a similar property for actions with virtually cyclic stabilisers.
Main Theorem. Let (W, S) be a right-angled Coxeter system for which
the nerve L = L(W, S) is an acyclic 2-complex that cannot be embedded in

2
any contractible 2-complex.

• If W is word hyperbolic, then

cd W = 2 and gd W = 3.

• If W is not word hyperbolic, then

cd W = gd W ≥ 3.

A right angled Coxeter group W is word hyperbolic if and only if its

nerve L satisfies the so called “flag no squares condition”, cf. [4, p. 233].
By Proposition 2.1 of [5] the “flag no squares condition” puts no restriction
on the homeomorphism type of the 2-complex L (or see [1, p. 498] for an
explicit example for a suitably triangulated L). Therefore it follows from
our theorem, that the Bredon analogue of the Eilenberg–Ganea Conjecture
is false for the family of virtually cyclic subgroups.
The proof of the non-word hyperbolic case of our Main Theorem is
the easy part and is described in Section 3. The word hyperbolic case is
Theorem 6 and 7 combined.
As mentioned before, in the classical case cdF G = 1 implies gdF G = 1
by the work of Stallings and Swan. It follows from Dunwoody’s Accessibility
Theorem [6], that the same implication is true in the case that F = Ffin (G).
In the light of this one may ask, whether this implication also holds in the
case that F = Fvc (G). The first author obtained in his thesis a positive
answer for countable, torsion-free, soluble groups [9, p. 127]. In this class,
the groups G with cd G = 1 are precisely the subgroups of the rational
numbers which are not finitely generated and for these groups gd G = 1
holds. However, a general answer to this question is still open.

Acknowledgments
The first author is grateful for the support of the CRC 701 of the DFG.

2 Coxeter Groups and the Davis Complex

A Coxeter matrix is a symmetric matrix M = (mst ) indexed by a finite
set S and with entries integers or ∞ subject to the conditions that for
all s, t ∈ S

1. mst = 1 if s = t, and

3
 2. mst ≥ 2 otherwise.

Associated to a Coxeter matrix M one has the Coxeter group W given by

the presentation

W = hS | (st)mst = 1 for all s, t ∈ S with mst 6= ∞i.

The Coxeter group W is right-angled if the finite off-diagonal entries of the

Coxeter matrix are all equal to 2. The elements of S are called the funda-
mental Coxeter generators of the Coxeter group W and the pair (W, S) is
called a Coxeter system. If T ⊂ S, then WT denotes the subgroup of W
generated by T and these subgroups are called special.
The nerve L = L(W, S) of a Coxeter system (W, S) is the simplicial
complex with vertex set S and whose simplices are the non-empty sub-
sets T ⊂ S for which the special subgroup WT is finite.
Given a Coxeter system (W, S) the Davis Complex Σ = Σ(W, S) is a
contractible simplicial complex on which W acts with finite stabilisers; the
action of the fundamental generators S is by reflections. This complex has
been introduced in [3] and it can interpreted as the barycentric subdivison
of a cell complex where the cells are in bijective correspondence with the
cosets of finite special subgroups of W . This cell complex admits in a
natural way a piecewise Euclidean metric and this metric can be shown
to be CAT(0). The links of the 0-cells of this complex can be identified
with the nerve L. The full subcomplex of Σ whose vertices correspond to
the identity cosets of the finite special subgroups is denoted by K. It is a
fundamental domain of the action of W and it can be realised as the cone
of L, where L is identified with the boundary ∂K in Σ. For details see [4].
If (W, S) is a right angled Coxeter system, then its nerve is a flag com-
plex [4, p. 125]. Conversely, if we are given a finite flag complex L, then we
can construct a Coxeter system (W, S) such that L is its nerve as follows:
let S be the set of vertices of L and for s 6= t set mst = 2 if s and t are
adjacent in L and set mst = ∞ if no edge connects s and t in L.

3 The Non-Hyperbolic Case

It suffices to show that cd W ≥ 3. For this it is enough to show that W
contains a subgroup H with cd H ≥ 3. Since W is not word hyperbolic it
contains a subgroup isomorphic to Z2 [4, p. 241].
We show that EZ2 = 3 using an explicit 3-dimensional model X for
EZ2 , which was first described by Farrell. See [8] for a general construction
containing this as a special case, or see [11] for a description of X and a

4
computation of H∗ (X/Z2 ; Z) from which it follows that H 3 (X/Z2 ; Z) is a
countable direct product of copies of Z. Theorem 4.2 in [9, p. 83] states
that H 3 (X/Z2 ) ∼
= HF3vc (Z2 ) (Z2 ; Z). Hence it follows that cd Z2 = 3.

4 The Geometric Dimension in the Hyperbolic

Case
Given a Coxeter system (W, S) and a W -space X we set
[
X# = Xs
s∈S

and

X sing = {x ∈ X | Wx 6= 1}.

Clearly X # ⊂ X sing .
Lemma 1. Let K ⊂ Σ be the fundamental chamber of Σ and let s ∈ S.
Then both K and K ∪ sK are convex subsets of Σ.
Proof. For each t ∈ S the fixedpoint set Σt separates Σ into two connected
half spaces. Denote by Ht− the half space which does not intersect K and
+ +
denote by H t the complement of Ht− . Then H t is a convex subset of Σ
T +
containing K. Then K = t∈S H t is a convex subset of Σ. Finaly, K ∩ sK
is
T convex since K ∪ sK = K0 ∩ sK0 where K0 is the convex set K0 =
+
t∈S\{s} H t .

Lemma 2. Let X be a model for EW . Then X # is homotopy equivalent

to L.
Proof. Since X is W -homotopy equivalent to Σ it follows that X # is homo-
topy equivalent to Σ# . Thus it is enough that Σ# is homotopy equivalent
to L.
Let K be the fundamental chamber of Σ. Then K is complete and
compact and due to Lemma 1 also convex. Therefore, since Σ is a CAT(0)
space, there exists a retraction of Σ onto K which sends every point x ∈
Σ \ K to the unique point π(x) of K which is nearest to x, cf. [2, p.176f.].
Let K S the union of all mirrors of K, that is

K S = {x ∈ K | x ∈ K ∩ sK for some s ∈ S},

cf. [4, p. 63, p. 127]. The set K S is homotopy equivalent to L [4, p. 127].

5
 Let s ∈ S and x ∈ Σ# \ K. Let y = π(x). Then sy ∈ sK and
since K ∪ sK is convex it follows that the midpoint m of the geodesic
joining y and sy is contained in K ∪ sK. Since y and sy have the same
distance from K ∩ sK it follows that m ∈ K ∩ sK. In particular m ∈ K.
Since x ∈ Σ# it follows that d(x, y) = d(sx, sy) = d(x, sy). Since the metric
of Σ is CAT(0) it follows that d(x, m) ≤ max(d(x, y), d(x, sy)) = d(x, y).
By the uniqueness of the point π(x) it follows that m = y. Hence y ∈ K S .
It follows that the homotopy equivalence Σ ' K restricts to a homotopy
equivalence Σ# ' K S . Thus X # ' L.

Remark 3. The above lemma could be used to give a slightly different

proof of the main assertion of Proposition 4 of [1, p. 497].

Lemma 4. Let X be a model for EW . If W is word hyperbolic, then X #

is homotopy equivalent to _
L∨ S1
i∈I

where the index set I consists of all maximal infinite virtually cyclic sub-
groups of W which contain at least two non-commuting Coxeter generators.

Proof. Let Y be the model for EW which is obtained from Σ as described

in [11]. This construction yields for every maximal infinite virtually cyclic
subgroup H of W a 1-dimensional model ZH for EH together with an H-
equivariant embedding fH : ZH → Σ. We identify ZH with its image in Σ
under this embedding. Then Y is obtained by coning off the sets ZH and
extending the W -action suitably.
Since X is W -homotopy equivalent to Y it follows that X # is homotopy
equivalent to Y # . The set Y # is obtained from Σ# by coning off the
intersection Σ# ∩ZH for every maximal infinite virtually cyclic subgroup H
of W .
Let s, t ∈ S such that s, t ∈ H for some maximal infinite virtually
cyclic subgroup H of W . Then x ∈ ZH can be a common fixed point of s
and t if and only if s and t commute. In particular ZH ∩ X # can consist
of at most 2 points as a virtually cyclic subgroup of W cannot contain
more than 2 pairwise non-commuting Coxeter generators. Coning off a
singleton set of a path connected space does not change its homotopy type.
And coning off a subset of a path connected space which has two points is
homotopy equivalent to attaching a S 1 to it. Hence the claim of the lemma
follows.

Lemma 5. Let (X, A) be a CW-pair and let B be a CW-complex which

is homotopy equivalent to A. Then there exists a CW-pair (Y, B) which is

6
homotopy equivalent to (X, A) such that the cells of X \ A are dimension
wise in a 1-to-1 correspondence to the cells of Y \ B.
Proof. This follows directly from Theorem 4.1.7 in [10, p. 104].

Theorem 6. Let (W, S) be a Coxeter system with W word hyperbolic and

such that the nerve L(W, S) of this Coxeter system is an acyclic com-
plex, which is not homotopy equivalent to a subcomplex of a contractible
2-complex. Then gd W = 3.

Proof. Assume towards a contradiction that there exists a 2-dimensional

model X for EW . Then X # is homotopy equivalent to L ∨ S 1 by
W
Lemma 4. By Lemma 5 there exists a 2-dimensional CW-complex
W 1 Y which
is homotopy equivalent to X and which contains L ∨ S . In particular L
is a subcomplex of Y contradicting the assumption that L does not embed
into a contractible 2-complex. Thus gd W ≥ 3.
On the other hand, the Davis complex Σ is a model for EW and
dim Σ = dim L + 1 = 3. Since W is word hyperbolic we can elevate Σ to a
model for EW by attaching orbits of cells in dimension 2 and less, cf. [11].
Thus gd W ≤ 3 and equality holds.

5 The Cohomological Dimension

Theorem 7. Let (W, S) be a Coxeter system with W word hyperbolic and
such that the nerve L(W, S) of this Coxeter system is an acyclic com-
plex which is not homotopy equivalent to a subcomplex of a contractible
2-complex. Then cd W = 2.
Proof. Let F be the family of virtually cyclic subgroups of W . Let Z be the
submodule of the trivial OF W -module given by Z(G/H) = Z for any finite
subgroup H of W and which is 0 otherwise. The complex Σsing is acyclic
and 2-dimensional by [1] and it follows that C ∗ (Σsing ) gives a projective
resolution of Z of length 2. Thus pd Z ≤ 2.
On the other hand, if X is a model for EW , then a model Y for EW
can be obtained from X by attaching orbits of cells in dimension 2 and
less [11, Proposition 9]. It follows that C ∗ (Y, X) gives a free resolution
of Q = Z/Z of length 2. Thus pd Q ≤ 2.
Consider the short exact sequence

0→Z→Z→Q→0

of OF W -modules. Since pd Z and pd Q are bounded by 2 it follows by the

Horseshoe Lemma that pd Z ≤ 2, that is cd W ≤ 2.

7
 On the other hand, it follows from [11, Corollary 16] that the quo-
tient space EW/W has non-trivial cohomology in dimension 2, and thus
HF2 (W ; Z) must be non-trivial too, cf. Theorem 4.2 in [9, p. 83]. As a
consequence we get cd W ≥ 2 and therefore the claim follows.

References
[1] N. Brady, I. J. Leary, and B. E. A. Nucinkis, On algebraic and geometric dimen-
sions for groups with torsion, J. London Math. Soc. (2) 64 (2001), no. 2, 489–500.
MR1853466 (2002h:57007)
[2] M. R. Bridson and A. Haefliger, Metric spaces of non-positive curvature,
Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of
Mathematical Sciences], vol. 319, Springer-Verlag, Berlin, 1999. MR1744486
(2000k:53038)
[3] M. W. Davis, Groups generated by reflections and aspherical manifolds not cov-
ered by Euclidean space, Ann. of Math. (2) 117 (1983), no. 2, 293–324. MR690848
(86d:57025)
[4] , The geometry and topology of Coxeter groups, London Mathematical So-
ciety Monographs Series, vol. 32, Princeton University Press, Princeton, NJ, 2008.
MR2360474 (2008k:20091)
[5] A. N. Dranishnikov, Boundaries of Coxeter groups and simplicial complexes with
given links, J. Pure Appl. Algebra 137 (1999), no. 2, 139–151, DOI 10.1016/S0022-
4049(97)00202-8. MR1684267 (2000d:20069)
[6] M. J. Dunwoody, Accessibility and groups of cohomological dimension one, Proc.
London Math. Soc. (3) 38 (1979), no. 2, 193–215. MR531159 (80i:20024)
[7] S. Eilenberg and T. Ganea, On the Lusternik–Schnirelmann category of abstract
groups, Ann. of Math. (2) 65 (1957), 517–518. MR0085510 (19,52d)
[8] F. T. Farrell and L. E. Jones, Computations of stable pseudoisotopy spaces for aspher-
ical manifolds, Algebraic topology Poznań 1989, Lecture Notes in Math., vol. 1474,
Springer, Berlin, 1991, pp. 59–74. MR1133892 (92k:57038)
[9] M. Fluch, On Bredon (Co-)Homological Dimensions of Groups, Ph.D. thesis, Uni-
versity of Southampton (2010), available at arXiv:1009.4633v1.
[10] R. Geoghegan, Topological methods in group theory, Graduate Texts in Mathematics,
vol. 243, Springer, New York, 2008. MR2365352
[11] D. Juan–Pineda and I. J. Leary, On classifying spaces for the family of virtually
cyclic subgroups, Recent developments in algebraic topology, 2006, pp. 135–145.
MR2248975 (2007d:19001)
[12] W. Lück, Transformation groups and algebraic K-theory, Lecture Notes in Math-
ematics, vol. 1408, Springer-Verlag, Berlin, 1989. Mathematica Gottingensis.
MR1027600 (91g:57036)
[13] W. Lück and D. Meintrup, On the universal space for group actions with compact
isotropy, Geometry and topology: Aarhus (1998), 2000, pp. 293–305. MR1778113
(2001e:55023)

8
[14] J. R. Stallings, Groups of dimension 1 are locally free, Bull. Amer. Math. Soc. 74
(1968), 361–364. MR0223439 (36 #6487)
[15] R. G. Swan, Groups of cohomological dimension one, J. Algebra 12 (1969), 585–610.
MR0240177 (39 #1531)
[16] P. Symonds, The Bredon cohomology of subgroup complexes, J. Pure Appl. Algebra
199 (2005), no. 1–3, 261–298. MR2134305 (2006e:20093)