On Normal Subgroups of Coxeter Groups

Generated by Standard Parabolic Subgroups

Światosław R. Gal
Wrocław University
http://www.math.uni.wroc.pl/~sgal/papers/nscg.dvi
arXiv:math/0502563v1 [math.GR] 26 Feb 2005

Abstract : We discuss one construction of nonstandard subgroups in the category of Coxeter groups.
Two formulae for the growth series of such a subgroups are given.
As an application we construct a flag simple convex polytope, whose f-polynomial has non-real roots.

Introduction
The central object of this paper is the growth series of a Coxeter group. Many geometric features
of such a group (or any group in general) reflect in properties of the growth series.
We describe in detail the normal closure of a standard parabolic in a Coxeter group W. It is again
a Coxeter group and its Coxeter presentation is given explicitly.
We give two formulae for the growth series of such a normal subgroup. The first one is given as a
specialization of a multi-variable version of the growth series of W.
The second formula works when the normal subgroup is right-angled (this can be easily checked
by analysis of the Dynkin diagram of W). The formula is based on counting special subdiagrams
of the Dynkin diagram of W.
As an application we construct a flag simple convex polytope, whose f-polynomial has non-real
roots.
The author would like to thank Tadeusz Januszkiewicz for useful discussions, Paweł Goldstein for
his extensive help with the final version of this manuscript and Mike Davis for a preliminary version
of [D2].

1. Preliminaries on Coxeter Groups

Definition. A Coxeter system (W, S) is a group W together with a set of generators S, and presen-
tation
W = hS|(st)mst = 1 for all s, t ∈ Si,
where mss = 1 (i.e. all the generators have order two) and mst = mts ∈ {2, 3, . . . , ∞} if s 6= t.
One reads (st)∞ = 1 as no relation between s and t imposed. The matrix m is called the Coxeter
matrix of W . We usually ignore S and call W a Coxeter group. The subgroup of W generated by
T ⊂ S is denoted WT . Such a subgroup is called a standard parabolic or parabolic for short.

Traditionally a Coxeter matrix is depicted in a following decorated graph called Dynkin diagram.
Its vertex set is S and any two distinct vertices are joined by an edge with label mst . By convention
• we omit an edge if mst = 2,

2000 Mathematics Subject Classification: 20F55

Partially supported by a KBN grant 2 P03A 017 25.
2 S. R. Gal

• we omit a label if mst = 3,

• we draw double edge instead that labeled by 4 and
• we draw dashed edge instead that labeled by ∞.
We will use the following convention for the Cayley graph: the group acts on the left, therefore
{g1 , g2 } is an edge provided g−1
1 g2 ∈ S. We say that an edge {g1 , g2 } is marked by the generator
−1
g1 g2 .

Theorem 1.1 [D2]. Let W be a group generated by a set of involutions S. Let CW be the Cayley
graph of (W, S). Then (W, S) is a Coxeter system if and only if for any element s ∈ S the fixpoint
set (CW )s separates CW (i.e. CW − (CW )s has two components being interchanged by s).

Proof: It is a part of Theorem 2.3.3 in [D2] (cf. [D2, Definitions 2.2.1 and 2.2.10]). 

Definition. A conjugate of a generator from S is called a reflection. The set of all reflections is
denoted by R(W).

Corollary 1.2. Reflections are intrinsically defined by a Coxeter system. Precisely, if T ⊂ S then
R(WS ) ∩ WT = R(WT ).

Proof: Assume that r ∈ R(WS ) ∩ WT . Since the action of r is conjugate to that of some generator,
the fixpoint set of r also separates CW . Then r and the neutral element e are in different components
of CWS − (CWS )r , and thus in different components of CWT − (CWT )r . Take any path joining
e and r in CWT . Such a path has to intersect the separating fixpoint set (CWT )r in a midpoint of
some edge {w, wt}. Thus rw = wt and, what follows, r = wtw−1 ∈ R(WT ). 

Definition. The length of an element w of any group with respect to a given generating set S is
denoted by ℓ(w) and is equal to the minimal length of a word in S representing w. The word is
called reduced if its length is equal to the length of the element it represents.

Lemma 1.3 [T]. If (W, S) is a Coxeter system than any word can be reduced (into any reduced
word representing the same element of W ) in a sequence of moves of the following form:
(a) removing a subword of a form ss, or
(b) replacing an alternating subword st . . . of length mst by an alternating subword ts . . . of the
same length.

Note: The move (a) is not symmetric, i.e. we do not need creation of a pair ss. In particular any
two reduced words representing the same element may be joined by moves of the form (b) only.

2. Normal closure of a standard parabolic

If mst = 2n + 1 is odd then s and t are conjugates since the relation reads then

(st)n s(st)−n = t.

If one wants to find a normal closure of a subgroup WT it is convenient to incorporate in T all the
generators conjugate to those already in T . This clarifies the assumption of the following
On Normal Subgroups of Coxeter Groups 3

Proposition 2.1. Assume that T ⊂ S is such that for any t ∈ T and s ∈ S − T the exponent mst
is even (or ∞). Then
(1) there exists a well defined homomorphism φT : W → WS−T that is an identity on WS−T and
sends WT to the identity element,
(2) the normal closure WT of WT is the kernel of φT ,
(3) WT is generated by ST , a set of all the conjugates of T with respect to WS−T ,
(4) the Cayley graph CWT of (WT , ST ) is made from the Cayley graph CW of W by collapsing
all the edges marked by the generators not in T ,
(5) (WT , ST ) is a Coxeter system.
Proof : Let t ∈ T and s 6∈ T , then the relation st . . . = ts . . ., after substituting the identity for s,
becomes trivial exactly when mst is even. Thus (1) follows. This also proves that the relation “s
and t are conjugate” is the equivalence closure of the relation “mst is odd”.
Any element w of W may be written as follows
w0 t 1 w1 . . . t n wn =
w0 t1 w−1 w0 w1 t2 (w0 w1 )−1 . . . w0 . . . wn−1 tn (w0 . . . wn−1 )−1 (w0 . . . wn )




0

where ti ∈ T and wi ∈ WS−T . If e = φT (w) = w0 . . . wn then w is expressible in ST , therefore

(2) and (3).
Let g1 u1 , g2 u2 , where gi ∈ WT and ui ∈ WS−T , be the endpoints of an edge corresponding
to some generator in T , i. e. (g1 u1 )−1 g2 u2 ∈ T . In particular e = φT (u−1 −1
1 g1 g2 u2 ) =
φT (u−1
1 u2 ). Therefore u1 = u2 . Furthermore

g−1 −1
(g2 u1 ) u−1


1 g2 = u1 (g1 u1 ) 1 ∈ ST .

On the contrary, if g−11 g2 = utu

−1
, then the edge {g1 , g2 } in CWT is covered by the edge
{g1 u, g2 u} in CW . This proves (4).
Let t ∈ ST . By Theorem 1.1 we need to show that the fixpoint set of t separates CWT . Let W + and
W − denote the two components of CW − (CW )t . Two vertices, one from W + and the other from
W − are joined by an edge precisely when they differ by t on the right. Since t is not a conjugate
of any element of WS−T all the cosets of WS−T have to lie in W + or W − . Therefore the fixpoint
set of t also separates CWT . Thus (5). 
Remark 2.2: If WS−T is not finite, then WT has in general an infinite number of generators.

3. Coxeter matrix of WT
Definition. Define the support supp w of an element w of a Coxeter group to be the set of gener-
ators that appear in its reduced presentation. By Lemma 1.3 this does not depend on the reduced
presentation.
Definition. Let T1 , T2 ⊂ S. We write T1 ⊥ T2 if mt1 t2 = 2 for any t1 ∈ T1 and t2 ∈ T1 . If
T ⊂ S we define T ⊥ = {s ∈ S: mst = 2 for all t ∈ T }.
Fix t1 , t2 ∈ T . Let w∗ be the shortest element in the double coset Wt⊥ −T wWt⊥ −T . Such an
1 2
element is unique [B, Ch. IV, §1, Ex. 3].
4 S. R. Gal

Corollary 3.1. Let τ1 = w1 t1 w−1 −1

1 and τ2 = w2 t2 w2 be two generators of WT . Then τ1 = τ2
if and only if
t1 = t2 and (w−1 ∗
1 w2 ) = 1.

If τ1 6= τ2 then we have

 mt1 t2 if t1 6= t2 and (w−1 ∗
1 w2 ) = 1 ,
mτ1 τ2 = mst1
 2 if t1 = t2 and (w−1 ∗
1 w2 ) = s ∈ S − T ,
∞ otherwise.

Proof: Set w = (w−1 ∗

1 w2 ) , so
w−1
1 w2 = η1 wη2

for some ηi ∈ Wt⊥ −T . In particular ti commutes with ηi

i

Let N be a natural number such that

(3.2) (τ1 τ2 )N = 1.

Then (3.2) reads


−1 
1 = η−1
1 w −1
1 w η
1 1 = η −1 −1
1 w 1 w t
1 1 (w −1
1 w )t
2 2 w −1
1 w 2 . . . w1 η 1
= η−1 t1 (η1 wη2 )t2 (η1 wη2 )−1 . . . η1


1
= (η−1 −1
1 t1 η1 )w(η2 t2 η2 )w
−1
...
= t1 wt2 w−1 . . .

Lemma 1.3 implies that w has length at most one, otherwise no cancellation is possible since any
reduced form of w does not start with a letter that commutes with t1 nor ends with a letter that
commutes with t2 .
The case ℓ(w) = 0 is obvious. If w is a generator, then it commutes neither with t1 nor with t2 .
The only possibility to use Lemma 1.3 is described by the claim. 

4. Multi-variable growth series.

The length function on a Coxeter group may be refined in the following way. Take a set of generators
S0 ⊂ S with the property that whenever s belongs to S0 , then all the generators conjugate to s belong
to S0 .
Lemma 1.3 implies the following

Corollary. The number (counted with multiplicities) of generators from S0 appearing in a reduced
presentation of w, and denoted ℓS0 (w), does not depend on the presentation.

Note that ℓS0 satisfies the triangle inequality, i.e. ℓS0 (w1 w2 ) ≤ ℓS0 (w1 ) + ℓS0 (w2 ).
On Normal Subgroups of Coxeter Groups 5

Let ϕ: S → I be any partition of the generating set into the sets with the above property. We call
such a partition allowable. For each w ∈ W define a monomial in indeterminate I: wϕ (x) =
Q ℓϕ−1 (x) (w)
x∈I x , where x = (x)x∈I .
P
For any subset A of W, a formal power series Aϕ (x) = w∈A wϕ (x) is called a multi-variable
growth series of A ⊂ W.
The above definition needs some finiteness assumption on (W, S). It is enough to assume that S
is finite (if not, the coefficients of the growth series may be infinite). Since we also want to study
the growth series of WT , we want to know that ST is finite. Therefore, due to Remark 2.2, we will
assume that WS−T is finite.
The notation may look a little ambiguous. However, note that Wϕ does not depend on whether


W is a standard parabolic subgroup of some bigger Coxeter group or not. Similarly WT ϕ may
denote two things: either a growth series of WT or that of WT ⊂ W. In fact these two series
coincide, as shown by the following

Proposition 4.1. Assume that T ⊂ S is as in Proposition 2.1 and ϕ−1 (x0 ) = S − T . There is a
well defined prolongation ϕT of ϕ to ST by requiring that ϕ(wtw−1 ) = ϕ(t) for any t ∈ T and
w ∈ WS−T . It is obviously allowable. Then

(WT )ϕT (x) = (WT )ϕ (x)|x0 =1 ,

where the left series is a growth series of WT and on the right that of WT ⊂ W .

Note: The right hand side is well defined, i.e. there is only a finite number of nonzero coefficients
at Jxm0 for given monomial J in variables different from x0 .
If wϕ (x) = Jxm
Q
0 then w = w0 tj1 w1 . . . wn−1 tjn wn , where ϕ(tjk ) = J and wk ∈ WS−T .
There is only a finite number of possibilities for tjk and wk (notice that we have assumed that WST
is finite).
Proof of Proposition 4.1: Take a reduced word in S representing an element from WT and write it
as

(4.2) w0 t 1 w1 . . . t n wn

where ti ∈ T and wj are subwords in S − T .

If the following word in ST (representing the same element)

w0 t1 w−1 w0 w1 t2 (w0 w1 )−1 . . . w0 . . . wn−1 tn (w0 . . . wn−1 )−1

(4.3) 0

were not reduced, one would do some cancellations in (4.3), expand the result to obtain the word
in S with smaller ℓϕ−1 (x) for some x 6= x0 than that of (4.2). This would show that (4.2) was not
reduced.
Finally (4.2) and (4.3) define the same monomial. Thus the claim. 
6 S. R. Gal

Theorem 4.4. Under the assumptions of the preceding proposition

Wϕ (x)|x0 =1 = (#WS−T ) · WT ϕT
(x).

Proof : Any element w of W decomposes uniquely as w = wT w0 where, wT ∈ WT and

w0 ∈ WS−T . We need to show that ℓϕ−1 (x) (w) = ℓϕ−1 (x) (wT ) for any x 6= x0 .
Since ℓϕ−1 (x) (w0 ) = 0, by the triangle inequality we have

ℓϕ−1 (x) (wT ) ≥ ℓϕ−1 (x) (wT w0 ) ≥ ℓϕ−1 (x) (wT w0 w−1
0 ) = ℓϕ−1 (x) (wT ).


Note: Two generators of WT conjugate in W may not be conjugate in WT , thus not every multi
variable growth series of WT is specialization of that of W.

Corollary. Assume that T ⊂ S is as in Proposition 2.1. Let ϕ: S → {x0 , x} be such that ϕ(t) = x
if and only if t ∈ T . Then the ordinary growth series of WT and WT are computed as follows:

WT (x) = Wϕ (0, x),

Wϕ (1, x)
WT (x) = ,
Wϕ (1, 0)

where Wϕ (x0 , x) is a multi-variable growth series of W associated to the allowable partition ϕ.

Proof: Wϕ (1, 0) = WS−T (1) = #WS−T .

5. Right-angled Coxeter groups and flag complexes

Definition. A nerve NW of a Coxeter system (W, S) is a simplicial complex consisting of T ⊂ S

such that the subgroup WT is finite.

Remark : Some authors define a nerve as a baricentric subdivision of the above.

Definition 5.1. A Coxeter group is said to be right-angled if mst ∈ {1, 2, ∞} for all s, t ∈ S.

If W is a right-angled Coxeter group, then WT is finite if and only if WT ′ is finite for any two-
element subset T ′ of T , therefore NW is a flag completion of its one-skeleton: T is a face of NW if
and only if its one skeleton is contained in NW . Such a simplicial complex is called a flag complex.
On the other hand, let Γ be any graph. Let S denote the set of vertices of Γ. Declare

1 if s = t,
(
mst = 2 if there is an edge joining s and t,
∞ otherwise.

Then NW is the flag completion of Γ.

On Normal Subgroups of Coxeter Groups 7

Let X be any simplicial complex. For a given function ϕ: S → I on the set of vertices one defines
an f-polynomial of X by the formula
XY
(5.2) fX,ϕ (x): = ϕ(s).
σ∈X s∈σ

Proposition 5.3 [S, Prop. 26]. Let W be an arbitrary Coxeter group and ϕ an allowable partition.
Then Wϕ (x) is a series of a rational function. Moreover, if W is infinite, then

1 X (−1)#T
= ,
Wϕ (x−1 ) (WT )ϕ|T (x)
T ⊂S

where T runs over subsets of S such that WT is finite, and x−1 = (x−1 )x∈I .
Corollary 5.4. Assume that W is a right angled Coxeter group. Since no two generators are
conjugated, any function ϕ: S → I is an allowable partition and

−1

1
(5.5) fNW ,ϕ = .
1+x Wϕ (x−1 )
Q
Proof: If WT is finite, then (WT )ϕ (x) = t∈T (1 + ϕ(t)), thus (5.5) follows form Proposition
5.3. 

6. Computation of f-polynomial
Corollary 3.1 implies the following

Corollary 6.1. WT is a right-angled Coxeter group if and only if

(1) if s 6∈ T and t ∈ T then mst ∈ {2, 4, ∞}, and
(2) if t, t′ ∈ T then mtt′ ∈ {1, 2, ∞}.

Let us describe the basic example which will serve as a model for the general case.

Definition. The Coxeter system (W, S) is of type Bk if its Dynkin diagram is the following:

sk sk−1 sk−2 s4 s3 s2 s1
(6.2) b b b ··· b b b b

Coxeter group of type Bk is the group of symmetries of a regular k-dimensional cube. Assume that
the cube has vertices with each coordinate equal to 1 or −1. In this case the generator s1 corresponds
to the reflection in the hyperplane defined by vanishing the first coordinate (s1 changes the sign of
the first coordinate). The group W{s1 } is right-angled Coxeter group generated by reflections in the
hyperplanes defined by vanishing of some coordinate. We will call such a hyperplane a coordinate
hyperplane.
The parabolic subgroup W{s2 ,...,sn } is the symmetric group of n letters. It acts by permuting
coordinates. Precisely, sk transposes (k − 1)st and kth coordinates.
8 S. R. Gal

Proposition 6.3. Let (W, {s1 , . . . , sm }) be of type Bm . If Θ is a family of n commuting conjugates

of s1 then Θ may be conjugated by an element from W{s2 ,...,sk } in such a way that Θ consists of
all conjugates of s1 in W{s1 ,...,sn } .

Proof: Family of reflections in coordinate hyperplanes is defined by the intersection of these

hyperplanes. Since all such intersections can be conjugated by permuting the coordinates, provided
they have the same dimension, the claim follows. 
Now return to the general question. Let Θ ⊂ ST be a family of commuting generators of a right-
angled group WT . The aim of the rest of the present Section is to compute the f-polynomial of
NWT in case when WT is right angled, in terms of the Coxeter matrix of (W, S) (in contrast with
Theorem 4.4, which gives the f-polynomial in terms of the growth series of W).
In order to compute the f-polynomial of its nerve we need to determine maximal families of com-
muting conjugates of generators.

Theorem 6.4. Assume that T is as in Corollary 6.1. For any commuting family Θ of conjugates
of generators there is a unique subset Σ ⊂ S of the same cardinality as Θ, such that
(1) WΣ is finite, and if for any t ∈ T ′ = T ∩ Σ the set of vertices of the connected component of
the DynkinSdiagram of WΣ containing t is denoted by Σt , then:
• Σ = t∈T ′ Σt ,
• Σt is of type Bk(t) ,
• t is the unique element of Σt ∩ T and it is the distinguished generator s1 ;
(2) there exists wΘ ∈ WS−T such that
[ [
wΘ ΘwΘ = {wtw−1 |w ∈ WΣ } = {wtw−1 |w ∈ WΣt }.
t∈T ′ t∈T ′

Lemma 6.5. If G is a finite subgroup of a Coxeter group, then it is conjugate to a subgroup of some
finite standard parabolic.

Proof : Any Coxeter group acts on its Davis complex [D1] which is CAT(0) (nonpositively curved)
by the theorem of Moussong [M]. The only stabilizers of that action are the conjugates of finite
standard parabolics.
The claim follows, since any action of a finite group on a CAT(0) complex has a fixed point [BH].

Obviously, since the set of parabolic subgroups is closed under the intersection, there exist unique
smallest parabolic containing G.
Proof of Theorem 6.4: Since a family Θ of commuting involutions generates a finite subgroup, by
Lemma 6.5 we may assume that Θ lies in a finite parabolics WS′ . Let T ′ denote a set of those
elements of T whose conjugates appear in Θ.
We claim that T ′ ⊂ S′ ∩ T and moreover that each element of Θ is conjugated to some element
of T ′ by an element of WS′ . Indeed, by Corollary 1.2 each element of Θ (being a reflection) is
conjugated inside WS′ to some generator from S′ . By Corollary 6.1 no element of T is a conjugate
of any other generator from S, thus the claim.
On Normal Subgroups of Coxeter Groups 9

By Corollary 6.1 any generator from T ′ may be connected in the Dynkin diagram to any other only
by an edge labeled by 4. Thus by the classification of finite Coxeter groups ([B]), the connected
component S′t of t ∈ T ′ (within the Dynkin diagram of WS′ ) is a graph of type Bk for some k.
Let S′0 = S′ − t∈T ′ S′t . From the definition it follows that any Coxeter group is the product
S
of its finite parabolics corresponding to the connected components of its Dynkin diagram. Thus

WS′ = WS′0 × t∈T ′ WS′t . Since each element of Θ is a conjugate of some element of T ′ , we

have Θ ⊂ t∈T ′ WS′t . Moreover, if Θt denotes those elements from Θ which are conjugated to
t ∈ T ′ then Θt ⊂ WS′t .
By Proposition 6.3 we may find elements wt ∈ WS′t such that each wt Θt w−1 t consists of all
′
conjugates
Q of t in WΣt for a suitable subset Σ t ⊂ S t . Since different W ′
St commute, setting
wΘ = t∈T ′ wt completes the proof. 
Note that Θ determines Σ but Σ determines Θ only up to conjugacy. In order to count all commuting
families conjugate to Θ we need to understand the action of WS−T on the set of families as in
Theorem 6.4 (1).
Let Σ be such a family. The stabilizer of Σ is the centralizer of Σ in WS−T , i.e. WΣ0 , where
S
⊥
Σ0 = t∈T Σt − T.
Summarizing the above we have proved

Theorem 6.6. The f-polynomial of WT reads

X Y ϕ(t)k(t)
fN W ,ϕ = [WS−T : WΣ0 ]
T k(t)!
Σ⊂S t∈T

Where the sum runs over all subsets Σ satisfying Theorem 6.4 (1).

This formula allows the following refinement. Assume that σ ⊂ τ ∈ NW . The corresponding
families Σσ and Στ satisfy Σσ τ σ
t ⊂ Σt for all t ∈ T . However if we set σ and Σ such that Σt ⊂ Σt
for all t ∈ T , then there are several τ wit the property τ ⊃ σ and Στ ⊂ Σ.
The centralizer of Σσ act on the set of such τ. thus the number of such τ equals

[WΣ0 : WΣσ0 ].

Definition. The link Lkσ of σ a simplex in a simplicial complex X is a subcomplex consisting of

all τ ∈ X such that τ∗ = σ ∪ τ ∈ X and σ ∩ τ = ∅.

Corollary 6.7. Fix a simplex σ ∈ NWT . It is the commuting family corresponding to Σσ ⊂ S.

The f-polynomial of the link of σ reads

X Y ϕ(t)k(t)−kσ (t)
fLkσ ,ϕ (t) = [WΣσ : WΣ ] .
σ
(k(t) − kσ (t))!
Σ⊃Σ t∈T
10 S. R. Gal

7. An Example.

In this section we would like to present a specific triangulation of a sphere (which is a boundary
complex of a convex polytope), such that its f-polynomial has non-real roots. Other such examples
are constructed by a different method in [G], where the reader may also find a connection of the
(former) conjecture on real-rootedness of f-polynomials of such triangulations to geometry and
combinatorics with further references.

Let W be defined by the following Dynkin diagram:

b
t
c 1 b b b b b b b
t
bc 2
(7.1)

Let T = {t1 , t2 } consist of white dots in the Diagram 7.1. Recall that WT = ker(WS → WS−T ).

Remark 7.2. NWT may be realized as a convex triangulation of a sphere, i.e. a boundary complex
of a convex polytope.

Proof : Consider the group defined by the following Dynkin diagram:

b b b b b b b b

It can be realized as a finite covolume reflection group in the hyperbolic space with the fundamental
domain D a simplex with unique ideal vertex [B, Ch. V, §4, Ex. 17]. Let D = WS0 · D. It is
a convex hyperbolic polytope with some ideal vertices. Finally let D∗ be D with ideal vertices
truncated. Is is a polytope dual to NW T .

Since D∗ may be realized as a convex polytope in the Klein model of the hyperbolic space, the
polar dual of D∗ realizes NW T as a convex triangulation of a sphere [Z, Cor. 2.14]. 

According to Theorem 6.6 we have to find all suitable triples {Σt1 , Σt2 , Σ0 }. They are listed in the
following picture. Black dots in the connected component of ti are elements of Σti . The remaining
black dots (in the middle) are the element of Σ0 .
On Normal Subgroups of Coxeter Groups 11

P
k(t)

b c
b b b b b b b b bc b b b b b b b b b bc bc b b b b b b
8 c
b bc b
b c
b c
b b b b b b b bc bc b b b b b b b bc b bc b b b b b b
7 c
b bc b
b c
b b c
b b b b b b bc b bc b b b b b b
6 b bc
b c
b b b c
b b b b b bc b b bc b b b b b
5 b b
b c
b b b b c
b b b b bc b b b bc b b b b
4 b b
b c
b b b b b c
b b b bc b b b b bc b b b
3 b b
b c
b b b b b b c
b b bc b b b b b bc b b
2 b b
b c
b b b b b b b c
b bc b b b b b b bc b
1 b b
bc b b b b b b b bc
0 b

The orders of finite Coxeter groups are well known ([B]) Substituting the above to Theorem 6.6
one obtains

2t72 t1 t82 t6 t1 t7 t72

fNW (t1 , t2 ) = 696729600 ( + )+( 2 + 2 + )
T 5040 40320 720 5040 2 · 5040
t52 t1 t62 t4 t1 t52
+( + )+( 2 + )
2 · 2 · 120 2 · 720 24 · 24 2 · 6 · 120
t3 t1 t42 t2 t1 t32
+( 2 + )+( 2 + )
6 · 192 24 · 120 2 · 1920 6 · 1920
t2 t1 t22 t1 t2 1

+( + )+( + )+
23040 2 · 51840 322560 2903040 696729600
= t1 (276480 t2 + 967680 t2 + 1451520 t2 + 1209600 t2 + 604800 t32
7 6 5 4

+ 181440 t22 + 30240 t2 + 2160) + (17280 t82 + 207360 t72

+ 483840 t62 + 483840 t52 + 241920 t42 + 60480 t32 + 6720 t22 + 240 t2 + 1).

Therefore, in particular,

fNW (t, t) = 293760 t8 + 1175040 t7 + 1935360 t6 + 1693440 t5 + 846720 t4

T

+ 241920 t3 + 36960 t2 + 2400 t + 1.

12 S. R. Gal

and
t8 − 2392 ∗ t7 + 20188 ∗ t6 − 70504 ∗ t5 + 107590 ∗ t4
1 −70504 ∗ t3 + 20188 ∗ t2 − 2392 ∗ t + 1
=
WT (t) (1 + t)8
Thus the poles of WT (·) (with 2 digit precision) are: 0.41 · 10−3 , 0.24 ± 0.16i, 0.63, 1.6, 2.9 ± 1.9i,
2.4 · 103 . This provides a first known counterexample to the Real Roots Conjecture [G].
The link L (of codimension 2) of a face
c
b b b b b b bc b b

is a triangulation of a five dimensional flag sphere. Using Corollary 6.7 we check that fL (−1/2) 6= 0.
A link K (in L) of an edge defined by
b c
b b b b bc b b b

satisfies fK (t) = (1 + 2t)4 , which is an f-polynomial of a cross polytope (and, as follows, K is a

cross polytope).
Such a pair (K, L) allows us to construct a triangulation of a five dimensional sphere which is a
counterexample to the Real Roots Conjecture [G, Th. 5.2]. Five is also the smallest dimension
when this is possible [G, Th. 3.2].

8. References
[B] N. Bourbaki, Groupes et algèbres de Lie, chapitres IV-VI, Hermann 1968,
[BH] M. R. Bridson and A. Haefliger, Metric spaces of non-positive curvature, Springer, Berlin,
1999.
[D1] M. W. Davis, Nonpositive curvature and reflection groups, Handbook of Geometric Topol-
ogy, R. Daverman and R. Sher ed., Elsevier 2001.
[D2] M. W. Davis, book in preparation
[G] S. R. Gal, Real Root Conjecture fails for five and higher dimensional spheres, to appear in
Discrete & Computational Geometry
[M] G. Moussong, Hyperbolic Coxeter groups, PhD Dissertation, Ohio State University, 1987.
[S] J. P. Serre, Cohomologie des groupes discrets, in Prospects in Mathematics, pp. 77-169,
Annal of Math. Studies No. 70, Princeton 1971
[T] J. Tits, Le problème des mots dans les groups de Coxeter in Symposia Mathematica, vol. 1,
London 1969
[Z] G. M. Ziegler, Lectures on polytopes, Graduate Texts in Mathematics, Vol. 152, Springer-
Verlag, New York 1995

Wrocław, April 2002