Journal of Algebra 281 (2004) 666–678

www.elsevier.com/locate/jalgebra

Artin groups of spherical type up to isomorphism

Luis Paris
Institut de Mathématiques de Bourgogne, Université de Bourgogne, UMR 5584 du CNRS, BP 47870,
21078 Dijon cedex, France
Received 23 October 2003
Available online 11 September 2004
Communicated by Michel Broué

Abstract
We prove that two Artin groups of spherical type are isomorphic if and only if their defining
Coxeter graphs are the same.
 2004 Published by Elsevier Inc.

1. Introduction

Let S be a finite set. Recall that a Coxeter matrix over S is a matrix M = (mst )s,t ∈S
indexed by the elements of S such that mss = 1 for all s ∈ S, and mst = mt s ∈ {2, 3, 4,
. . . , +∞} for all s, t ∈ S, s = t. A Coxeter matrix M = (mst ) is usually represented by
its Coxeter graph, Γ , which is defined as follows. The set of vertices of Γ is S, two
vertices s, t are joined by an edge if mst
3, and this edge is labelled by mst if mst
4.
For s, t ∈ S and m ∈ Z
2 , we denote by w(s, t : m) the word sts . . . of length m. The Artin
group associated to Γ is defined to be the group G = GΓ presented by

 
G = S  w(s, t : mst ) = w(t, s : mst ) for s, t ∈ S, s = t and mst < +∞ .

The Coxeter group W = WΓ associated to Γ is the quotient of G by the relations s 2 = 1,

s ∈ S. We say that Γ (or G) is of spherical type if W is finite, that Γ (or G) is right-angled

E-mail address: [email protected].

0021-8693/$ – see front matter  2004 Published by Elsevier Inc.

doi:10.1016/j.jalgebra.2004.04.021
 L. Paris / Journal of Algebra 281 (2004) 666–678 667

if mst ∈ {2, +∞} for all s, t ∈ S, s = t, and that G (or W ) is irreducible if Γ is connected.
The number n = |S| is called the rank of G (or of W ).
One of the main questions in the subject is the classification of Artin groups up to
isomorphism (see [2, Question 2.14]). This problem is far from being completely solved
as Artin groups are poorly understood in general. For example, we do not know whether
all Artin groups are torsion free, and we do not know any general solution to the word
problem for these groups. The only known results concerning this classification question
are contained in a work by Brady, McCammond, Mühlherr, and Neumann [4], where the
authors determine a sort of transformation on Coxeter graphs which does not change the
isomorphism class of the associated Artin groups, and a work by Droms [16], where it
is proved that, if Γ and Ω are two right-angled Coxeter graphs whose associated Artin
groups are isomorphic, then Γ = Ω. Notice that an Artin group is biorderable if and only
if it is right-angled, hence a consequence of Droms’ result is that, if Γ is a right-angled
Coxeter graph and Ω is any Coxeter graph, and if the Artin groups associated to Γ and
Ω are isomorphic, then Γ = Ω. The fact that right-angled Artin groups are biorderable
is proved in [17]. In order to show that the remaining Artin groups are not biorderable,
one has only to observe that, if 2 < mst < +∞, then (st)mst = (ts)mst and st = ts, and
that, in a biorderable group, two distinct elements cannot have a common mth power for a
fixed m.
In this paper we answer the classification question in the restricted framework of spher-
ical type Artin groups. More precisely, we prove the following theorem.

Theorem 1.1. Let Γ and Ω be two spherical type Coxeter graphs, and let G and H be the
Artin groups associated to Γ and Ω, respectively. If G is isomorphic to H , then Γ = Ω.

Remark. I do not know whether a nonspherical type Artin group can be isomorphic to a
spherical type Artin group.

Artin groups were first introduced by Tits [24] as extensions of Coxeter groups. Later,
Brieskorn [5] gave a topological interpretation of the Artin groups of spherical type in terms
of complements of discriminantal varieties. Define a (real) reflection group of rank n to be
a finite subgroup W of GL(n, R) generated by reflections. Such a group is called essential
if there is no non-trivial subspace of Rn on which W acts trivially. Let A be the set of
reflecting hyperplanes of W , and, for H ∈ A, let HC denote the complexification of H ,
i.e., the complex hyperplane
 in Cn defined by the same equation as H . Then W acts freely
on M(W ) = C \ H ∈A HC , and, by [12], N(W ) = M(W )/W is the complement in Cn of
n

an algebraic variety, D(W ), called discriminantal variety of type W . Now, take a spherical
type Coxeter graph Γ , and consider the associated Coxeter group W = WΓ . By [13], the
group W can be represented as an essential reflection group in GL(n, R), where n = |S| is
the rank of W , and, conversely, any essential reflection group of rank n can be uniquely
obtained in this way. By [5], π1 (N(W )) is the Artin group G = GΓ associated to Γ .
So, a consequence of Theorem 1.1 is that π1 (N(W )) = π1 (Cn \ D(W )) completely
determines the reflection group W as well as the discriminantal variety D(W ).
Since the work of Brieskorn and Saito [7] and that of Deligne [15], the combinatorial
theory of spherical type Artin groups has been well studied. In particular, these groups are
668 L. Paris / Journal of Algebra 281 (2004) 666–678

know to be biautomatic (see [10,11]), and torsion-free. This last result is a direct conse-
quence of [15] and [5], it is explicitly proved in [14], and it shall be of importance in the
remainder of the paper.
The first step in the proof of Theorem 1.1 consists of calculating some invariants for
spherical type Artin groups (see Section 3). It actually happens that these invariants sepa-
rate the irreducible Artin groups of spherical type (see Proposition 5.1). Afterwards, for a
given isomorphism ϕ : G → H between spherical type Artin groups, we show that, up to
some details, ϕ sends each irreducible component of G injectively into a unique irreducible
component of H , and that both components have the same invariants. In order to do that,
we first need to show that an irreducible Artin group G cannot be decomposed as a product
of two subgroups which commute, unless one of these subgroups lies in the center of G
(see Section 4).
From now on, Γ denotes a spherical type Coxeter graph, G denotes its associated Artin
group, and W denotes its associated Coxeter group.

2. Preliminaries

We recall in this section some well-known results on Coxeter groups and Artin groups.
For a subset X of S, we denote by WX the subgroup of W generated by X, and by GX
the subgroup of G generated by X. Let ΓX be the full Coxeter subgraph of Γ whose vertex
set is X. Then WX is the Coxeter group associated to ΓX (see [3]), and GX is the Artin
group associated to ΓX (see [19] and [21]). The subgroup WX is called standard parabolic
subgroup of W , and GX is called standard parabolic subgroup of G.
For w ∈ W , we denote by lg(w) the word length of w with respect to S. The group W
has a unique element of maximal length, w0 , which satisfies w02 = 1 and w0 Sw0 = S, and
whose length is m1 + · · · + mn , where m1 , m2 , . . . , mn are the exponents of W .
The connected spherical Coxeter graphs are exactly the graphs An (n
1), Bn (n
2),
Dn (n
4), E6 , E7 , E8 , F4 , H3 , H4 , I2 (p) (p
5) represented in [3, Chapter IV, §4,
Theorem 1]. (Here we use the notation I2 (6) for the Coxeter graph G2 . We may also use
the notation I2 (3) for A2 , and I2 (4) for B2 .)
Let F : G → W be the natural epimorphism which sends s to s for all s ∈ S. This
epimorphism has a natural set-section T : W → G defined as follows. Let w ∈ W , and let
w = s1 s2 . . . sl be a reduced expression of w (i.e., l = lg(w)). Then T (w) = s1 s2 . . . sl ∈ G.
By Tits’ solution to the word problem for Coxeter groups [25], the definition of T (w) does
not depend on the choice of the reduced expression.
Define the Artin monoid associated to Γ to be the (abstract) monoid G+ presented by

 +
G+ = S  w(s, t : mst ) = w(t, s : mst ) for s = t and mst < +∞ .

By [7], the natural homomorphism G+ → G which sends s to s for all s ∈ S is injective.

Note that this fact is always true, even if Γ is not assumed to be of spherical type (see [22]).
The fundamental element of G is defined to be ∆ = T (w0 ), where w0 denotes the
element of W of maximal length. For X ⊂ S, we denote by wX the element of WX of
maximal length, and by ∆X = T (wX ) the fundamental element of GX .
 L. Paris / Journal of Algebra 281 (2004) 666–678 669

The defining relations of G+ being homogeneous, we can define two partial orders L
and R on G+ as follows:

• We set a L b if there exists c ∈ G+ such that b = ac.

• We set a R b if there exists c ∈ G+ such that b = ca.

Now, the following two propositions are a mixture of several well-known results
from [7] and [15].

Proposition 2.1.

(1) G+ is cancellative.
(2) (G+ , L ) and (G+ , R ) are lattices.
(3) {a ∈ G+ ; a L ∆} = {a ∈ G+ ; a R ∆} = T (W ).

Note that the fact that G+ is cancellative is true even if Γ is not of spherical type
(see [20]). The elements of T (W ) are called simple elements. We shall denote the lattice
operations of (G+ , L ) by ∨L and ∧L , and the lattice operations of (G+ , R ) by ∨R
and ∧R .
Define the quasi-center of G to be the subgroup QZ(G) = {a ∈ G; aSa −1 = S}.

Proposition 2.2. Assume Γ to be connected.

(1) For X ⊂ S we have

 
{s; s ∈ X} = {s; s ∈ X} = ∆X .
L R

In particular,
 
{s; s ∈ S} = {s; s ∈ S} = ∆.
L R

(2) There exists a permutation µ : S → S such that µ2 = Id and ∆s = µ(s)∆ for all s ∈ S.
(3) The quasi-center QZ(G) of G is an infinite cyclic subgroup generated by ∆.
(4) The center Z(G) of G is an infinite cyclic subgroup of G generated either by δ = ∆ if
µ = Id, or by δ = ∆2 if µ = Id.

The generator δ of Z(G) given in the above proposition shall be called the standard
generator of Z(G). Note also that the assumption “Γ is connected” is not needed in (1)
and (2). Let Γ be connected. Then µ = Id if and only if Γ is either An , n
2, or D2n+1 ,
n
2, or E6 , or I2 (2p + 1), p
2 (see [7, Section 7.2]).
Now, the following result can be found in [11].

Proposition 2.3 (Charney [11]). Each a ∈ G can be uniquely written as a = bc−1 where
b, c ∈ G+ and b ∧R c = 1.
670 L. Paris / Journal of Algebra 281 (2004) 666–678

The expression a = bc−1 of the above proposition shall be called the Charney form
of a.
An easy observation shows that, if s1 s2 . . . sl and t1 t2 . . . tl are two positive expressions
of a same element a ∈ G+ , then the sets {s1 , . . . , sl } and {t1 , . . . , tl } are equal. In partic-
ular, if a ∈ G+
X , then all the letters that appear in any positive expression of a lie in X.
A consequence of this fact is the following lemma.

Lemma 2.4. Let X be a subset of S, let a ∈ GX , and let a = bc−1 be the Charney form
of a in G. Then b, c ∈ G+
X and a = bc
−1 is the Charney form of a in G .
X

Proof. Let ∨X,R and ∧X,R denote the lattice operations of (G+ X , R ). The above observa-
tion shows that, if a R b and b ∈ G+ X , then a ∈ G +
X . This implies that b ∧X,R c = b ∧R c
for all b, c ∈ G+
X . Now, let a ∈ G X and let a = bc −1 be the Charney form of a in G . We
X
have b, c ∈ G+ +
X ⊂ G and b ∧R c = b ∧X,R c = 1, thus a = bc
−1 is also the Charney form

of a in G. 2

Corollary 2.5. Let X be a subset of S. Then GX ∩ G+ = G+

X.

Corollary 2.6. Let X be a subset of S, X = S. Then GX ∩ ∆

= {1}.

/ G+
Proof. Take s ∈ S \ X. By Proposition 2.2, we have s R ∆, thus ∆ ∈ X =
+
GX ∩ G . 2

3. Invariants

The purpose of the present section is to calculate some invariants of the spherical type
Artin groups.
The first invariant that we want to calculate is the cohomological dimension, denoted
by cd(G). We assume the reader to be familiar with this notion, and we refer to [9] for
definitions and properties. Our result is the following.

Proposition 3.1. Let n = |S| be the rank of G = GΓ . Then cd(G) = n.

Proof. Recall the spaces M(W ) and N(W ) defined in the introduction. Recall also that
π1 (N(W )) = G, that W acts freely on M(W ), and that N(W ) = M(W )/W . In particular,
π1 (M(W )) is a subgroup of π1 (N(W )) = G (it is actually the kernel of the epimorphism
F : G → W ). Finally, recall the well-known fact that, if H1 is a subgroup of a given
group H2 , then cd(H1 )  cd(H2 ).
Deligne proved in [15] that M(W ) is aspherical,
 and Brieskorn proved in [6] that
H n (M(W ), Z) is a free abelian group of rank ni=1 mi = 0, where m1 , m2 , . . . , mn are
the exponents of W , thus n  cd(π1 (M(W )))  cd(G). On the other hand, Salvetti has
constructed in [23] an aspherical CW-complex of dimension n whose fundamental group
is G, therefore cd(G)  n. 2
 L. Paris / Journal of Algebra 281 (2004) 666–678 671

The next invariant which interests us is denoted by mf(G) and is defined to be the
maximal order of a finite subgroup of G/Z(G), where Z(G) denotes the center of G. Its
calculation is based on Theorems 3.2 and 3.3 given below.
Recall the permutation µ : S → S of Proposition 2.2. This extends to an isomorphism
µ : G+ → G+ which permutes the simple elements. Actually, µ(a) = ∆a∆−1 for all
a ∈ G+ .

Theorem 3.2 (Bestvina [1]). Assume Γ to be connected. Let G = G/ ∆2

, and let H be a
finite subgroup of G. Then H is a cyclic group, and, up to conjugation, H has one of the
following two forms:

Type 1: The order of H is even, say 2p, and there exists a simple element a ∈ T (W ) such
that a p = ∆, µ(a) = a, and a generates H , where a denotes the element of G
represented by a.
Type 2: The order of H is odd, say 2p + 1, and there exists a simple element a ∈ T (W )
such that (aµ(a))(p−1)/2a = ∆ and aµ(a) generates H .

Now, recall the so-called Coxeter number h of W (see [18, Section 3.18]). Recall also
that this number is related to the length of ∆ by the following formula:

nh
= m1 + · · · + mn = lg(∆),
2

where n = |S| is the rank of G, and m1 , . . . , mn are the exponents of W .

Theorem 3.3 (Brieskorn–Saito [7]). Choose any order S = {s1 , s2 , . . . , sn } of S and write
π = s1 s2 . . . sn ∈ G. Let h be the Coxeter number of W .

(1) If µ = Id, then h is even and π h/2 = ∆.

(2) If µ = Id, then π h = ∆2 .

Now, we can calculate the invariant mf(G).

Proposition 3.4. Assume Γ to be connected, and let h be the Coxeter number of W .

(1) If µ = Id, then mf(G) = h/2.

(2) If µ = Id, then mf(G) = h.

Proof. Assume µ = Id. Let G 0 = G/Z(G) = G/ ∆

. First, observe that mf(G)
h2 by
Theorem 3.3. So, it remains to prove that mf(G)  h2 , namely, that |H |  h2 for any finite
subgroup H of G 0 .
Let H be a finite subgroup of G 0 . Consider the exact sequence

φ
1 → Z/2Z → G → G 0 → 1,
672 L. Paris / Journal of Algebra 281 (2004) 666–678

where G = G/ ∆2
, and set H  = φ −1 (H ). By Theorem 3.2, H  is a cyclic group and,

up to conjugation, H is either of Type 1 or of Type 2. The order of H  is even, say 2p,

thus H is of Type 1, and there exists a simple element a ∈ T (W ) such that a p = ∆ and
. Let a = s1 s2 . . . sr be an expression of a, and let X = {s1 , s2 , . . . , sr }. We
a generates H
have ∆ = a p ∈ GX , thus, by Corollary 2.6, X = S and r = lg(a)
|S| = n. Finally,

|
|H lg(∆) nh
h
|H | = =p= = r .
2 lg(a) 2 2

Now, assume µ = Id. Let G = G/Z(G) = G/ ∆2

. First, observe that mf(G)
h by
Theorem 3.3. So, it remains to prove that mf(G)  h, namely, that |H |  h for any finite
subgroup H of G.
Let H be a finite subgroup of G. By Theorem 3.2, H is cyclic and, up to conjugation,
H is either of Type 1 or of Type 2. Let p be the order of H . In both cases, Type 1 and
Type 2, there exists an element b ∈ G+ such that bp = ∆2 and b generates H (take b = a
if H is of Type 1, and b = aµ(a) if H is of Type 2). Let b = s1 s2 . . . sr be an expression
of b, and let X = {s1 , s2 , . . . , sr }. We have ∆2 = bp ∈ GX , thus, by Corollary 2.6, X = S
and r = lg(b)
|S| = n. It follows that

lg(∆2 ) nh
|H | = p = =  h. 2
lg(b) r

The values of the Coxeter numbers of the irreducible Coxeter groups are well-known
(see, for instance, [18, Section 3.18]). Applying Proposition 3.4 to these values, one can
easily compute the invariant mf(G) for each irreducible (spherical type) Artin group. The
result is given in Table 1.

Remark. Combining [1, Theorem 4.5], with [8, Section 3], one can actually compute all
the possible orders for a finite subgroup of G/Z(G). The maximal order suffices for our
purpose, thus we do not include this more complicate calculation in this paper.

The next invariant that we want to compute is the rank of the abelianization of G that we
denote by rkAb(G). This invariant can be easily computed using the standard presentation
of G, and the result is as follows.

Table 1
The invariant mf(G)
Dn , n
4 Dn , n
5
Γ A1 An , n
2 Bn , n
2 n even n odd E6
mf(G) 1 n+1 n n−1 2n − 2 12

I2 (p), p
6 I2 (p), p
5
Γ E7 E8 F4 H3 H4 p even p odd
mf(G) 9 15 6 5 15 p/2 p
 L. Paris / Journal of Algebra 281 (2004) 666–678 673

Proposition 3.5. Let Γ0 be the (non-labelled) graph defined by the following data:

• S is the set of vertices of Γ0 ;

• two vertices s, t are joined by an edge if mst is odd.

Then the abelianization of G is a free abelian group of rank rkAb(G), the number of
connected components of Γ0 .

The last invariant which interests us is the rank of the center of G that we denote by
rkZ(G). The following proposition is a straightforward consequence of Proposition 2.2.

Proposition 3.6. The center of G is a free abelian group of rank rkZ(G), the number of
components of Γ .

4. Irreducibility

Throughout this section, we assume that G is irreducible (namely, that Γ is connected).

Let H1 , H2 be two subgroups of G. Recall that [H1 , H2 ] denotes the subgroup of G gen-
erated by {a1−1 a2−1 a1 a2 ; a1 ∈ H1 and a2 ∈ H2 }. Our goal in this section is to show that
G cannot be expressed as G = H1 · H2 with [H1 , H2 ] = {1}, unless either H1 ⊂ Z(G) or
H2 ⊂ Z(G). This shall implies that G cannot be a non-trivial direct product.
Recall that δ denotes the standard generator of Z(G). For X ⊂ S, we denote by δX the
standard generator of Z(GX ), and, for a ∈ G, we denote by ZG (a) the centralizer of a
in G.

Lemma 4.1. Let t ∈ S be such that ΓS\{t } is connected and µ(t) = t if µ = Id. Then
ZG (δS\{t }) is generated by GS\{t } ∪ {δ} and is isomorphic to GS\{t } × δ
.

Proof. Assume first that µ = Id (in particular, δ = ∆). By [21, Theorem 5.2], ZG (δS\{t } )
is generated by GS\{t } ∪ {∆2 , ∆∆−1S\{t } }, thus ZG (δS\{t } ) is generated by GS\{t } ∪ {δ}.
Now, assume µ = Id (in particular, δ = ∆2 and µ(t) = t). By [21, Theorem 5.2],
ZG (δS\{t }) is generated by GS\{t } ∪ {∆2 , ∆∆−1 −1 −1
S\{µ(t )} ∆∆S\{t } }. Observe that ∆∆S\{µ(t )} ×
∆∆−1 2 −2
S\{t } = ∆ ∆S\{t } , thus ZG (δS\{t } ) is generated by GS\{t } ∪ {δ}.
By the above, we have an epimorphism GS\{t } × δ
→ ZG (δS\{t } ), and, by Corol-
lary 2.6, the kernel of this epimorphism is {1}. 2

Remark. It is an easy exercise to show (under the assumption that Γ is connected) that
there always exists t ∈ S such that ΓS\{t } is connected and µ(t) = t if µ = Id. This can be
done using the classification of the connected Coxeter graphs of spherical type by a case
by case inspection.

Proposition 4.2. Let H1 , H2 be two subgroups of G such that G = H1 · H2 and [H1 , H2 ] =

{1}. Then either H1 ⊂ Z(G) or H2 ⊂ Z(G). If, moreover, H1 ∩ H2 = {1}, then either
H1 = {1} and H2 = G, or H1 = G and H2 = {1}.
674 L. Paris / Journal of Algebra 281 (2004) 666–678

Proof. We argue by induction on n = |S|. If n = 1, then Γ = A1 and G = Z, and the

conclusion of the proposition is well-known.
Assume n
2. For i = 1, 2, let H i denote the subgroup of G generated by Hi ∪ {δ}. We
   
have G = H1 · H2 , [H1 , H2 ] = {1}, H1 ⊂ H 1 , and H2 ⊂ H 2 . Observe also that H
1 ∩ H2
   
must be included in the center of G, and that δ ∈ H1 ∩ H2 , thus H1 ∩ H2 = δ
. Take t ∈ S
1
such that ΓS\{t } is connected and µ(t) = t if µ = Id, write X = S \ {t}, and choose d1 ∈ H

and d2 ∈ H2 such that δX = d1 d2 .
Let a ∈ GX . Choose a1 ∈ H 1 and a2 ∈ H2 such that a = a1 a2 . We have

−1
1 = a −1 δX aδX = a1−1 d1−1 a1 d1 a2−1 d2−1 a2 d2 ,

thus

1 ∩ H
a1−1 d1−1 a1 d1 = d2−1 a2−1 d2 a2 ∈ H 2 = δ
.

Let k ∈ Z such that a1−1 d1−1 a1 d1 = δ k . Consider the homomorphism deg : G → Z which
sends s to 1 for all s ∈ S. Then
   
0 = deg a1−1 d1−1 a1 d1 = deg δ k = k lg(δ),

thus k = 0, hence a1 and d1 commute. Now, a1 and d2 also commute (since a1 ∈ H 1


and d2 ∈ H2 ), thus a1 commutes with δX = d1 d2 . By Lemma 4.1, a1 can be written as
a1 = b1 δ p1 , where b1 ∈ GX and p1 ∈ Z. Note also that b1 = a1 δ −p1 ∈ H1 , since δ ∈ H
1 ,

thus b1 ∈ GX ∩ H1 . Similarly, a2 can be written as a2 = b2 δ where b2 ∈ GX ∩ H2 and
p 2 
p2 ∈ Z. We have δ p1 +p2 = ab1−1b2−1 ∈ GX ∩ δ
= {1} (by Corollary 2.6), thus p1 + p2 = 0
and a = b1 b2 .
So, we have

1 ) · (GX ∩ H
GX = (GX ∩ H 2 ).

Moreover, by Corollary 2.6,

1 ) ∩ (GX ∩ H
(GX ∩ H 2 ) = GX ∩ δ
= {1}.

By the inductive hypothesis, it follows that, up to permutation of 1 and 2, we have

GX ∩ H 1 = GX (namely, GX ⊂ H 1 ), and GX ∩ H2 = {1}.
We turn now to show that H 2 ⊂ δ
= Z(G). Since H2 ⊂ H 2 , this shows that H2 ⊂
Z(G).
2 . Since δX ∈ GX ⊂ H
Let a ∈ H 1 , a and δX commute. By Lemma 4.1, a can be written
as a = bδ p , where b ∈ GX and p ∈ Z. Since δ ∈ H 2 , we also have b = aδ −p ∈ H 2 , thus
b ∈ GX ∩ H 2 = {1}, therefore a = δ ∈ δ
.
p

Now, assume that H1 ∩ H2 = {1}. By the above, we may suppose that H2 ⊂ Z(G) = δ
.
In particular, there exists k ∈ Z such that H2 = δ k
. Choose any order S = {s1 , . . . , sn }
of S, and write π = s1 s2 . . . sn ∈ G. Let b ∈ H1 and p ∈ Z such that π = bδ pk . Observe that
b = 1 since π is not central in G. Let h denote the Coxeter number of W . By Theorem 3.3,
 L. Paris / Journal of Algebra 281 (2004) 666–678 675

π h = bh δ phk ∈ Z(G), thus bh ∈ Z(G). Moreover, bh = 1 since G is torsion free and b = 1.

This implies that Z(H1 ) = {1}. Now, observe that Z(H1 ) ⊂ Z(G) = δ
, thus there exists
l > 0 such that Z(H1 ) = δ l
. Finally, δ lk ∈ H1 ∩ H2 = {1}, thus kl = 0, therefore k = 0
(since l = 0) and H2 = {1}. Then we also have H1 = G. 2

Proposition 4.3. Assume n = |S|
2. Let H be a subgroup of G such that G = H · δ

.
Then cd(H ) = cd(G), mf(H ) = mf(G), and rkAb(H ) = rkAb(G).

Proof. For all s ∈ S, take bs ∈ H and ps ∈ Z such that s = bs δ ps . We can and do suppose
that ps = pt if s and t are conjugate in G. Then the mapping S → H , s → bs = sδ −ps
determines a homomorphism ϕ : G → H .
We show that ϕ : G → H is injective. Observe that the mapping S → Z, s → ps deter-
mines a homomorphism η : G → Z, and that ϕ(a) = aδ −η(a) for all a ∈ G. In particular,
if a ∈ Ker ϕ, then a = δ η(a) ∈ Z(G). Choose any order S = {s1 , . . . , sn } of S, and write
π = s1 s2 . . . sn ∈ G. Note that ϕ(π) = 1, since π is not central in G, and that, by Theo-
rem 3.3, there exists k > 0 such that π k = δ. Let a ∈ Ker ϕ. Then a = δ η(a) = π kη(a) , thus
1 = ϕ(a) = ϕ(π)kη(a) . We have ϕ(π) = 1 and G is torsion free, hence η(a) = 0 (since
k > 0) and a = 1.
Now, recall that cd(H1 )  cd(H2 ) if H1 is a subgroup of a given group H2 . So,
 
cd(G) = cd ϕ(G)  cd(H )  cd(G).

The equality G = H · δ
= H · Z(G) implies that Z(H ) = Z(G) ∩ H and G/Z(G) =
H /Z(H ). In particular, we have mf(H ) = mf(G).
Let H be a group, let g be a central element in H, and let p > 0. Let G = (H × Z)/
(g, p)
. Then one can easily verify (using the Reidemeister–Schreier method, for ex-
ample) that we have exact sequences 1 → H → G → Z/pZ → 1 and 1 → Ab(H) →
Ab(G) → Z/pZ → 1, where Ab(G) (respectively Ab(H)) denotes the abelianization of G
(respectively H).
Now, recall the equality G = H · δ
. By Proposition 4.2, we have H ∩ δ
= {1}. So,
there exists p > 0 such that H ∩ δ
= δ p
. Write d = δ p ∈ H . Then d is central in H
and G (H × Z)/ (d, p)
. By the above observation, it follows that we have an exact
sequence 1 → Ab(H ) → Ab(G) → Z/pZ → 1, thus Ab(H ) is a free abelian group of
rank rkAb(G). 2

5. Proof of the main theorem

Proposition 5.1. Let Γ and Ω be two connected spherical type Coxeter graphs, and let
G and H be the Artin groups associated to Γ and Ω, respectively. If cd(G) = cd(H ),
mf(G) = mf(H ), and rkAb(G) = rkAb(H ), then Γ = Ω.

Proof. Let n and m be the numbers of vertices of Γ and Ω, respectively. By Proposi-

tion 3.1, we have n = cd(G) = cd(H ) = m.
Suppose n = m = 1. Then Γ = Ω = A1 .
676 L. Paris / Journal of Algebra 281 (2004) 666–678

Suppose n = m
3. Then one can easily verify in Table 1 that the equality mf(G) =
mf(H ) implies Γ = Ω.
Suppose n = m = 2. Let p, q
3, such that Γ = I2 (p) and Ω = I2 (q). By Propo-
sition 3.5, either rkAb(G) = rkAb(H ) = 2 and p, q are both even, or rkAb(G) =
rkAb(H ) = 1 and p, q are both odd. If p, q are both even, then, by Table 1, p2 = mf(G) =
mf(H ) = q2 , thus p = q and Γ = Ω = I2 (p). If p, q are both odd, then, by Table 1,
p = mf(G) = mf(H ) = q, thus Γ = Ω = I2 (p). 2

Corollary 5.2. Let Γ and Ω be two connected spherical type Coxeter graphs, and let G
and H be the Artin groups associated to Γ and Ω, respectively. If G is isomorphic to H ,
then Γ = Ω.

Proof of Theorem 1.1. Let Γ and Ω be two spherical type Coxeter graphs, and let G
and H be the Artin groups associated to Γ and Ω, respectively. We assume that G is
isomorphic to H and turn to prove that Γ = Ω.
Let Γ1 , . . . , Γp be the connected components of Γ , and let Ω1 , . . . , Ωq be the connected
components of Ω. For i = 1, . . . , p, we denote by Gi the Artin group associated to Γi ,
and, for j = 1, . . . , q, we denote by Hj the Artin group associated to Ωj . We have G =
G1 × G2 × · · · × Gp and H = H1 × H2 × · · · × Hq . We may and do assume that there exists
x ∈ {0, 1, . . . , p} such that Γi = A1 for i = 1, . . . , x, and Γi = A1 for i = x + 1, . . . , p. So,
G1 , . . . , Gx are non-abelian irreducible Artin groups of rank
2, and Gx+1 , . . . , Gp are all
isomorphic to Z. Similarly, we may and do assume that there exists y ∈ {0, 1, . . . , q} such
that Ωj = A1 for j = 1, . . . , y, and Ωj = A1 for j = y + 1, . . . , q. We can also assume
that x
y.
A first observation is, by Proposition 3.6, that

p = rkZ(G) = rkZ(H ) = q.

Now, fix an isomorphism ϕ : G → H . For 1  i  p, let ιi : Gi → G be the natural

embedding, for 1  j  p, let κj : H → Hj be the projection on the j th component, and,
for 1  i, j  p, let ϕij = κj ◦ ϕ ◦ ιi : Gi →Hj .
p
Let j ∈ {1, . . . , y}. Observe that Hj = i=1 ϕij (Gi ), and that [ϕij (Gi ), ϕkj (Gk )] = 1
for all i, k ∈ {1, . . . , p}, i = k. Let δjH denote the standard generator of Z(Hj ), and, for
i ∈ {1, . . . , p}, let Hij be the subgroup of Hj generated by ϕij (Gi ) ∪ {δ H }. By Proposi-
j
tion 4.2, there exists χ(j ) ∈ {1, . . . , p} such that Hj = H χ(j )j , and H
ij = Z(Hj ) = δ H

j
for i = χ(j ). Since Hj is non-abelian, χ(j ) is unique and χ(j ) ∈ {1, . . . , x}.
We turn now to show that the map χ : {1, . . . , y} → {1, . . . , x} is surjective. Since x
y,
it follows that x = y and χ is a permutation.
Let i ∈ {1, . . . , x} such that χ(j ) = i for all j ∈ {1, . . . , y}. Then ϕij (Gi ) ⊂ Z(Hj ) for
all j = 1, . . . , p, thus ϕ(Gi ) ⊂ Z(H ). This contradicts the fact that ϕ is injective and Gi is
non-abelian.
So, up to renumbering the Γi ’s, we can suppose that χ(i) = i for all i ∈ {1, . . . , x}.
We prove now that ϕii : Gi → Hi is injective for all i ∈ {1, . . . , x}. Let a ∈ Ker ϕii .
Since ϕij (a) ∈ Z(Hj ) for all j = i, we have ϕ(a) ∈ Z(H ). Since ϕ is injective, it follows
 L. Paris / Journal of Algebra 281 (2004) 666–678 677

that a ∈ Z(Gi ). Let {s1 , . . . , sr } be the set of vertices of Γi , and let π = s1 s2 . . . sr ∈ Gi .

Observe that ϕii (π) = 1 since π is not central in Gi . Let δiG be the standard generator
of Z(Gi ). By Theorem 3.3, there exists k > 0 such that π k = δiG . On the other hand, since
a ∈ Z(Gi ), there exists l ∈ Z such that a = (δiG )l = π kl . Now, 1 = ϕii (a) = ϕii (π)kl , Hi
is torsion free, and ϕii (π) = 1, thus kl = 0 and a = π kl = 1.
Let i ∈ {1, . . . , x}. Recall that ϕii : Gi → Hi is injective, and Hi = ϕii (Gi ) · δiH
, where
δiH denotes the standard generator of Hi . By Proposition 4.3, it follows that

cd(Gi ) = cd(Hi ), mf(Gi ) = mf(Hi ), rkAb(Gi ) = rkAb(Hi ),

thus, by Proposition 5.1, Γi = Ωi . Let i ∈ {x + 1, . . . , p}. Then Γi = Ωi = A1 . So,

Γ = Ω. 2

Remark. In the proof above, the homomorphism ϕii is injective but is not necessarily
surjective as we show in the following example.

Let G1 = s1 , s2 | s1 s2 s1 = s2 s1 s2
be the Artin group associated to A2 , let G2 =
Z = t
, and let G = G1 × G2 . We denote by δ = (s1 s2 )3 the standard generator of Z(G1 ).
Let ϕ : G → G be the homomorphism defined by

ϕ(s1 ) = s1 δt, ϕ(s2 ) = s2 δt, ϕ(t) = δt.

Then ϕ is an isomorphism but ϕ11 is not surjective. The inverse ϕ −1 : G → G is determined

by

ϕ −1 (s1 ) = s1 t −1 , ϕ −1 (s2 ) = s2 t −1 , ϕ −1 (t) = δ −1 t 7 .

Acknowledgments

The idea of looking at centralizers of “good” elements in the proof of Proposition 4.2
is a suggestion of Benson Farb. I am grateful to him for this clever idea as well as for all
his useful conversations. I am also grateful to Jean Michel who pointed out to me his work
with Michel Broué, and to John Crisp for so many discussions on everything concerning
this paper.

References

[1] M. Bestvina, Non-positively curved aspects of Artin groups of finite type, Geom. Topol. 3 (1999) 269–302.
[2] M. Bestvina, Questions in geometric group theory; available at http://www.math.utah.edu/~bestvina.
[3] N. Bourbaki, Groupes et algèbres de Lie, Chapitres IV, V et VI, Hermann, Paris, 1968.
[4] N. Brady, J.P. McCammond, B. Mühlherr, W.D. Neumann, Rigidity of Coxeter groups and Artin groups,
Geom. Dedicata 94 (2002) 91–109.
[5] E. Brieskorn, Die Fundamentalgruppe des Raumes der regulären Orbits einer endlichen komplexen Spiegel-
ugsgruppe, Invent. Math. 12 (1971) 57–61.
678 L. Paris / Journal of Algebra 281 (2004) 666–678

[6] E. Brieskorn, Sur les groupes de tresses, in: Séminaire Bourbaki, 24ème année (1971/72) Exp. No. 401, in:
Lecture Notes in Math., vol. 317, Springer-Verlag, Berlin, 1973, pp. 21–44.
[7] E. Brieskorn, K. Saito, Artin-Gruppen und Coxeter-Gruppen, Invent. Math. 17 (1972) 245–271.
[8] M. Broué, J. Michel, Sur certains éléments réguliers des groupes de Weyl et les variétés de Deligne–Lusztig
associées, in: Finite Reductive Groups, Luminy, 1994, in: Progr. Math., vol. 141, Birkhäuser Boston, Boston,
MA, 1997, pp. 73–139.
[9] K.S. Brown, Cohomology of Groups, Grad. Texts in Math., vol. 87, Springer-Verlag, New York, 1982.
[10] R. Charney, Artin groups of finite type are biautomatic, Math. Ann. 292 (1992) 671–683.
[11] R. Charney, Geodesic automation and growth functions for Artin groups of finite type, Math. Ann. 301
(1995) 307–324.
[12] C. Chevalley, Invariants of finite groups generated by reflections, Amer. J. Math. 77 (1955) 778–782.
[13] H.S.M. Coxeter, Discrete groups generated by reflections, Ann. of Math. 35 (1934) 588–621.
[14] P. Dehornoy, Gaussian groups are torsion free, J. Algebra 210 (1998) 291–297.
[15] P. Deligne, Les immeubles des groupes de tresses généralisés, Invent. Math. 17 (1972) 273–302.
[16] C. Droms, Isomorphisms of graph groups, Proc. Amer. Math. Soc. 100 (1987) 407–408.
[17] G. Duchamp, J.Y. Thibon, Simple orderings for free partially commutative groups, Internat. J. Algebra
Comput. 2 (1992) 351–355.
[18] J.E. Humphreys, Reflection Groups and Coxeter Groups, Cambridge Stud. Adv. Math., vol. 29, Cambridge
Univ. Press, Cambridge, 1990.
[19] H. Van der Lek, The homotopy type of complex hyperplane complements, PhD thesis, Nijmegen, 1983.
[20] J. Michel, A note on words in braid monoids, J. Algebra 215 (1999) 366–377.
[21] L. Paris, Parabolic subgroups of Artin groups, J. Algebra 196 (1997) 369–399.
[22] L. Paris, Artin monoids inject in their groups, Comment. Math. Helv. 77 (2002) 609–637.
[23] M. Salvetti, The homotopy type of Artin groups, Math. Res. Lett. 1 (1994) 565–577.
[24] J. Tits, Normalisateurs de tores. I. Groupes de Coxeter étendus, J. Algebra 4 (1966) 96–116.
[25] J. Tits, Le problème des mots dans les groupes de Coxeter, in: INDAM, Rome, 1967/68, in: Sympos. Math.,
vol. 1, Academic Press, London, 1969, pp. 175–185.