Topics in Algebra II: Coxeter groups

Felix Schremmer

April 26, 2023

By definition, a reflection group is a subgroup of the orthogonal group which is gen-

erated by a finite set of reflections. The concept of a Coxeter group is a purely group-
theoretic analogue of those reflection groups. While the definition of a Coxeter group is
fairly accessible, as it is a finitely presented group of a very specific shape, the theory is
rather deep and still subject to research.
Typical examples of Coxeter groups are dihedral groups and symmetric groups. While
the most important application of Coxeter groups is definitely Lie theory (coming from
Weyl groups and affine Weyl groups), the theory is also relevant for classical geometry
and knot theory.
The course will be divided into two parts. The first part will cover the fundamental
structure theory of Coxeter groups, such as Bruhat order and parabolic subgroups. In
the second part, we will turn our attention to more advanced topics related to Coxeter
groups, depending on the participants’ interests and backgrounds.
Please do not hesitate to email me any questions or suggestions for the courses con-
tents, or these lecture notes in particular. Confer to the course website for the latest
version of the lecture notes, as well as announcements of any kind.

Contents
1 Introduction 2

2 Coxeter systems 4

3 Examples of Coxeter groups 6

4 The geometric representation 9

5 Length 12

6 Structure theory of Coxeter groups 16

7 Matsumoto’s Theorem 18

1
8 Parabolic subgroups 21

9 Bruhat order 23

10 Bruhat order on parabolic quotients 26

11 The Tableau criterion 28

12 Finite Coxeter groups 30

13 Weak Order 32

14 Foldings 34

15 List of finite Coxeter groups 37

16 Iwahori-Hecke algebra 40

17 Braid group 42

18 Kazhdan-Lusztig polynomials 43

19 Kazhdan-Lusztig representations 46

20 Reflection orders 47

21 Reflection subgroups 50

22 Demazure products 53

23 Finiteness results 54

24 Conjugacy classes 56

1 Introduction
The definition of a Coxeter group is fairly abstract, so let us motivate it by discussing
the symmetric group.

Definition 1. Let n ě 0 be an integer. The symmetric group over n letters is defined

as the set of all bijective maps from the set t1, . . . , nu into itself. It becomes a group
under function composition.
For i ‰ j, we denote by pi jq P Sn the map that interchanges i and j and leaves all other
elements of t1, . . . , nu fixed, called a transposition. The transpositions si “ pi i ` 1q are
called standard transpositions. More generally, for pairwise distinct i1 , . . . , iℓ , we write

pi1 i2 ¨ ¨ ¨ iℓ q P Sn

2
for the ℓ-cycle, sending i1 to i2 etc.
Elements in f P Sn are written as a product of cycles, or in the two-row notation

1 2
ˆ ˙
¨¨¨ n
f“ .
f p1q f p2q ¨ ¨ ¨ f pnq

Our main goal for now is to show the following:

Theorem 2. The symmetric group Sn is generated by the standard transpositions s1 , . . . , sn´1 .

This is certainly well-known and a combinatorial proof is not hard to come by. How-
ever, we want to give a geometric proof of Theorem 2.
In order to give such a geometric proof, we introduce a useful representation of the
symmetric group.
Denote by V one of the following vector spaces:

V “ Rn or V “ tpv1 , . . . , vn q P Rn | v1 ` ¨ ¨ ¨ ` vn “ 0u.

We have an action of Sn on V by

f.pv1 , . . . , vn q “ pvf p1q , . . . , vf pnq q, f P Sn , pv1 , . . . , vn q P V.

For i ‰ j, denote by Hi,j the hyperplane

Hi,j “ tpv1 , . . . , vn q P V | vi “ vj u “ tv P V | pi jq.v “ vu.

Call a vector v P V regular if v does not lie on any of the hyperplanes Hi,j , i.e. if the
coordinates are pairwise distinct. Denote by V reg Ă V the set of regular vectors.
The connected components of V reg are called Weyl chambers. An example for such a
Weyl chamber is the dominant chamber

C “ tpv1 , . . . , vn q P V | v1 ă ¨ ¨ ¨ ă vn u.

It is easy to see that we get a bijective map

Sn Ñ tWeyl chambersu, f ÞÑ f C.

(pictures)
We define the length function in Sn as follows: The length ℓpf q of f P Sn is the number
of hyperplanes Hi,j such that C and f C lie on opposite sides of Hi,j . So ℓpf q “ 0 if and
only if f C “ C, which means that f is the identity map. Note that Sn permutes the set
of hyperplanes, where f Hi,j “ Hf ´1 piq,f ´1 pjq .

Lemma 3. Let f P Sn and 1 ď i ď n ´ 1 such Hi,i`1 lies between f C and C. Then

ℓpsi f q “ ℓpf q ´ 1.

3
Proof. We have

ℓpsi f q “#tHa,b | Ha,b lies between psi f qC and Cu

“#tHa,b | Ha,b lies between f C and si Cu
“#tHa,b | Ha,b lies between f C and C and Ha,b ‰ Hi,i`1 u
“ℓpf q ´ 1.

Corollary 4. An element f P Sn can be written as the product of ℓpf q standard trans-

positions, but not as a product of less than ℓpf q standard transpositions.

Proof. Induction on ℓpf q. If ℓpf q “ 0, then f must be the identity map, and there is
nothing to prove.
If ℓpf q ą 0, then f C ‰ C. There must be some hyperplane between f C and C, and
the only hyperplanes adjacent to C are those of the form Hi,i`1 . So we find hyperplane
Hi,i`1 between f C and C.
Now ℓpsi f q “ ℓpf q ´ 1 by the lemma, hence si f can be written as the product of
ℓpf q ´ 1 standard transpositions. We see that f “ si ¨ psi f q is a product of ℓpf q standard
transpositions.
For the converse, use the lemma to show that ℓpsi f q ď ℓpf q ` 1 for all f P Sn and all
simple reflections si .

Observe that Theorem 2 is an immediate consequence of that corollary.

We have proved that Sn is generated by the si . Moreover, the following equations
always hold true (and are easy to prove):

s2i “ 1, si sj “ sj si if |i ´ j| ě 2,
si si`1 si “ si`1 si si`1 .

It turns out that these are enough: Any true equation consisting of si ’s on both sides
can be derived from the above relations and the laws of group theory. By definition, this
makes Sn an example of a Coxeter group.
Exercise 5. Consider the element
1 2 3
ˆ ˙
f“ “ p1 3q P S3 .
3 2 1

What is ℓpf q? Give an explicit formula writing f as the product of ℓpf q transpositions.

2 Coxeter systems
Definition 6. A Coxeter system is a (typically finite) set S together with a map

m : S ˆ S Ñ Zě1 Y t`8u

such that mps, s1 q “ mps1 , sq for all s, s1 P S, and mps, s1 q “ 1 if and only if s “ s1 .

4
 The Coxeter group associated with a Coxeter system pS, mq is the group W with the
following presentation:
Generators: The set S.
1
Relations: For each s, s1 P S, we require pss1 qmps,s q “ 1 unless mps, s1 q “ `8.
In the example of Sn , the set S is given by ts1 , . . . , sn´1 u and the function m is
$
&1,
’ i “ j,
mpsi , sj q “ 2, |i ´ j| ě 2,
’
3, |i ´ j| “ 1.
%

We visualize the Coxeter system pS, mq by the Coxeter diagram. The nodes of the
Coxeter diagram are in one to one correspondence with the set S. We draw an edge
between s and s1 iff mps, s1 q ě 3. If mps, s1 q ě 4, we moreover label the connecting edge
by mps, s1 q. The Coxeter diagram for Sn is known as An´1 .
An´1 : ‚ ´ ‚ ´ ¨ ¨ ¨ ´ ‚.
Let us recall the concept of a finitely presented group in this specific context.
Write S ˚ for the set of words in S, i.e. finite sequences ps1 , . . . , sℓ q of arbitrary length
ℓ with si P S. An elementary reduction of a word w is the deletion of a subword of the
form
ps, s1 , . . . , s, s1 q
loooooooomoooooooon
length 2mps,s1 q

for s, s1 P S. Two words w, w1 are equivalent if there exists a sequence of words w “

w1 , . . . , wn “ w1 of words such that wi`1 is an elementary reduction of wi or vice versa.
Then W can be defined as the set of all equivalence classes of words. There is a group
structure on W given by
rws„ ¨ rw1 s„ “ rw ˝ w1 s„ ,
the circle denoting composition of words. Moreover, we have a natural map φ : S Ñ W ,
sending an element s P S to the equivalence class of the one-letter word rpsqs„ P W .
The group W is, up to unique isomorphism, uniquely determined by its universal
property:
Proposition 7. Let W 1 be any group, and ψ : S Ñ W 1 any function such that for all
s, s1 P S, we have
1
pψpsqψps1 qqmps,s q “ 1 in W 1 .

Then there exists a unique group homomorphism ψ̂ : W Ñ W 1 such that ψ “ ψ̂ ˝ φ.

φ
S W
ψ̂
ψ
W 1.

5
 It turns out that the presentation we gave for W is already minimal, i.e. that S is a
minimal generating set and the set of relations is minimal as well.
Proposition 8. The map φ : S Ñ W is injective. For each s, s1 P S, the order of
φpsqφps1 q in W is equal to mps, s1 q.
We will give a proof of this proposition later. For now, we use it as an excuse to write
s P W instead of φpsq P W .
Exercise 9. If you are not familiar with the language of finitely presented groups, it
makes sense to think a bit about these concepts:
(a) Going back to the construction of W as equivalence classes of words, why is the
multiplication well-defined, and why does it make W a group?

(b) Give a proof of Proposition 7.

Exercise 10. Show that whenever S ‰ H, there exists a subgroup H ď W of index 2
such that φpsq R H for all s P S.

3 Examples of Coxeter groups

In view of Proposition 8, we say that pW, Sq is a Coxeter group without specifying mp¨, ¨q.
Example 11. The symmetric group Sn is a Coxeter group with respect to the standard
transpositions s1 , . . . , sn´1 .
Example 12. The Dihedral group of order 2n is defined to be the group of symmetries
of the regular n-gon. e.g. D6 is the symmetry group of an equilateral triangle. In D2n ,
we find precisely n reflections. If a and b are “adjacent” reflections, then pD2n , ta, buq is
a Coxeter group.
a

Example 13. The most interesting examples of Coxeter groups come from Lie theory. If
G denotes a Lie group or a linear algebraic group, one may pick a Borel subgroup B Ă G
and denote by W Ă G the1 Weyl group. This is always a finite Coxeter group (for some
choice of S determined by B). The Bruhat decomposition
ğ
G“ BwB
wPW

is the first step towards understanding the group G (which is infinite and has non-trivial
geometry) via the finite Coxeter group W .
For a concrete example, one may choose G to be the general linear group G “ GLn ,
B the subgroup of upper triangular matrices and W the group of permutation matrices
1
W is not actually a subset of G, but in practice, one can pretty much always treat it like a subgroup.

6
(i.e. matrices whose associated linear map permutes the set of standard basis vectors).
Then W – Sn is a Coxeter group. The Bruhat decomposition in this case is a fancy way
to express Gauss’ algorithm.
Example 14. Let G be the group of orthogonal n ˆ n-matrices with n odd. The Weyl
group pW, Sq is a Coxeter group with Coxeter diagram
4
Bpn´1q{2 : ‚ ‚ ‚ ¨¨¨ .

It can be identified with the subgroup of Sn`1 consisting of those permutations w that
satisfy

wpn ` 2 ´ iq “ n ` 2 ´ wpiq, i “ 1, . . . , n ` 1.

The simple reflections consist of pi i ` 1qpn ` 1 ´ i n ` 2 ´ iq for i “ 1, . . . , pn ´ 1q{2 and

the one standard transposition interchanging pn ` 1q{2 with pn ` 3q{2.
Example 15. Let n ě 1. We define W to be the group of matrices g P GLn pZrt˘1 sq
subject to the following three conditions:

• Each row and each column of g contains only one non-zero entry.

• Each non-zero entry of g has the form tm for some m P Z.

• The sum of those exponents occurring in g is equal to zero. Equivalently, detpgq “

˘1.

Since W contains all permutation matrices, we get a natural embedding of Sn into W .

Group-theoretically, W is the semi-direct product of Sn acting on

tpµ1 , . . . , µn q P Zn | µ1 ` ¨ ¨ ¨ ` µn “ 0u.

This group W is a Coxeter group with respect to the generating set ts0 , . . . , sn´1 u. Here,
s1 , . . . , sn are the usual standard transpositions of Sn Ă W . The reflection s0 is given
by the matrix

0 ... 0 t
¨ ˛
˚ 1 0‹
˚ .. . .. ‹ .
˚ ‹
˚ . . . .‹
˝ 0 1 ‚
˚ ‹

t´1 0 . . . 0

The corresponding Coxeter diagram is a cycle with n nodes:

7
 The group W is known as the affine Weyl group of GLn . Denote by L the field of
formal Laurent series
8
! ÿ )
L “ kpptqq “ ai ti | N P Z and a‚ P k .
i“´N

Let I denote the subgroup of G “ tg P GLn pLq | detpgq P pkrrtssqˆ u consisting of those
matrices pgi,j q P Lnˆn such that

• Each gi,j lives in krrtss, i.e. has no negative powers of t occurring with non-zero
coefficient.

• The elements above the main diagonal live in tkrrtss, i.e. have no non-positive
powers of t occurring with non-zero coefficient.

The group I is known as Iwahori subgroup of G. We get the Iwahori-Bruhat decomposi-

tion
ğ
G“ IwI.
wPW

Similar decompositions exist for SLn , GLn etc.

Exercise 16. Let pW1 , S1 q and pW2 , S2 q be two Coxeter groups.

(a) Show that the product group W1 ˆ W2 can be equipped with the structure of a
Coxeter group (“is” a Coxeter group).

(b) Show that the free product group W1 ˚ W2 , i.e. the coproduct object in the category
of groups, can be equipped with the structure of a Coxeter group.

Exercise 17. Give an example of two non-isomorphic Coxeter systems whose associated
Coxeter groups are isomorphic.
Exercise 18. The Bruhat decomposition divides GL3 pkq into six subsets, as indexed by
the six elements of S3 .

• For each w P S3 , write down a list of conditions that determine whether a matrix
g P GL3 pkq lies in BwB or not.

• Determine the dimension of BwB{B as a variety over k (if k is algebraically closed)

or manifold over k (if k “ R).

Exercise 19. Give a description of the Coxeter group pD8 , ta, buq associated with the
Coxeter graph
8
‚ ‚.

8
4 The geometric representation
In this section, pick a Coxeter system pS, mq, denote the corresponding Coxeter group
by W and let φ : S Ñ W be the canonical map.
It is easy to write down elements in W , namely by expressing them as w “ φps1 q ¨ ¨ ¨ φpsn q
for some elements s1 , . . . , sn P S. If two such expressions yield the same element w P W ,
we can prove they are identical by applying the defining relations of a Coxeter group.
But how can we prove that two expressions yield distinct elements in W ?
We can answer this question by introducing a group representation of W , i.e. a suitable
map W Ñ GLpV q for a vector space V . For each pair s, s1 P S, pick a real number
ks,s1 P R subject to the following conditions:

• ks,s “ 2 whereas ks,s1 ď 0 for s ‰ s1 ,

• ks,s1 “ 0 if and only if mps, s1 q “ 2,

´ ¯2
• ks,s1 ks1 ,s “ 4 cos mps,s
π
1q if 2 ď mps, s1 q ă 8 and

• ks,s1 ks1 ,s ě 4 if mps, s1 q “ 8.

If is a canonical choice to always select

π
ks,s1 “ ´2 cos .
mps, s1 q

In some circumstances, other choices might be more appealing (e.g. if we can chose all
ks,s1 to be integers).
We define V to be the R-vector space with basis tαs | s P Su. For each s P S, we
define the linear map αs_ : V Ñ R via

αs_ pαs1 q “ ks,s1 , s1 P S.

For s P S, v P V , define the reflection of v along αs as

σs pvq :“ v ´ αs_ pvqαs .

Observe that σs defines a map in GLpV q such that pσs q2 “ id.

If v P V is any vector, we write v ě 0 if v is a Rě0 -linear combination of the basis
vectors αs , s P S. Similarly, we write v ď 0 if ´v ě 0.

Lemma 20. Consider the rank 2 case S “ ts, s1 u (the cardinality of the set S is known
as the rank of the Coxeter system) and let f : V Ñ V be the linear map defined by
f “ σs ˝ σs1 .

(a) The order of f is equal to mps, s1 q (i.e. the n-fold composition f n identity map if
and only if mps, s1 q divides n).

9
(b) Let gn : V Ñ V be described as the alternating composition

gn “ ¨ ¨ ¨ ˝ σs1 ˝ σs ˝ σs1

of n terms ending with σs1 . If n ă mps, s1 q, then gn pαs q ě 0.

Proof. Let g “ ks,s1 ks1 ,s and let B denote the ordered basis B “ p g{ks,s1 αs , g{ks1 ,s αs1 q
a a a

(resp. B “ pαs , αs1 q if mps, s1 q “ 2). With respect to the basis B, the matrix representing
f is given by

1 0 g ´ 1 ´g
ˆ ˙ˆ ˙ ˆ 2 ˙
´1 g
“ “: A.
0 1 g ´1 g ´1

The determinant of A is one and the trace is given by g 2 ´ 2.

In case mps, s1 q “ 8, we get g ě 2. It follows that the matrix A has two positive
real eigenvalues and is not conjugate to the identity matrix. In particular, f must have
infinite order.
In case mps, s1 q ă 8, the trace is further computed to be
ˆ ˙2
π ´ 1 1
¯2 1 1
4 cos ´ 2 “ eiπ{mps,s q ` e´iπ{mps,s q ´ 2 “ e2πi{mps,s q ` e´2πi{mps,s q .
mps, s1 q
1
It follows that e˘2πi{mps,s q are the two eigenvalues of A. Moreover, direct inspection in
case mps, s1 q “ 2 shows that A is always diagonalizable, even if the two eigenvalues are
both ´1. The claim on the order of f follows from this.
We now prove (b). Once can see this from Euclidean geometry, choosing a basis of the
Euclidean plane consisting of two unit length vectors with angle mps,s π
1 q between them.

Then A can be seen as the composition of the associated mirror reflections, which is a
rotation by the angle mps,s
2π
1 q . The statements can now be seen to follow from Euclidean

geometry. For these notes, we give a purely algebraic proof that leaves no special case
unchecked.
Denote the two (previously computed) eigenvalues of A by λ1 and λ2 , and define
integers
n
ÿ
an “ 1 ` pλk1 ` λk2 q.
k“1

Observe that an ě 0 whenever n ă mps, s1 q{2. We claim that An has the form
ˆ ˙
an ˚
n
A “ , 1 ď n ă mps, s1 q.
˚ ´an´1

The claim for n “ 1 is easily verified. In an inductive step, let us calculate

f n pαs1 q “ σs f˜pn´1q p´αs1 q,

10
where f˜ “ σs1 ˝ σs . Since everything is symmetric in s and s1 , we may assume that the
analogous statement on f˜pn´1q has been proved and conclude that
f˜pn´1q p´αs1 q “ an´1 p´αs1 q ` p˚qp´αs q.
Application of σs does not change the αs1 -coordinate, so that
f n pαs1 q ” ´an´1 αs1 pmod Rαs q,
proving the claim on the lower-right coefficient of An . Observe that trpAn q “ λn1 ` λn2 “
an ´ an´1 , so that the claim on the top-left coefficient follows.
In the situation of (b), observe that
#
σs1 f pn´1q{2 , n odd
gn “
f n{2 , n even.
In any case, it follows that
gn pαs q P atn{2u αs ` Rαs1
with atn{2u ě 0 whenever n ă mps, s1 q. Suppose that gn pαs q ğ 0, i.e. that we had
gn pαs q “ atn{2u αs ` cαs1 , c ă 0.
Then
σs gn pαs q “ p´atn{2u ´ cks,s1 qαs ` cαs1
has a negative coefficient for αs . Since σs gn “ gn˘1 , we immediately get a contradiction
unless n “ mps, s1 q ´ 1 and σs gn “ gn`1 . The latter condition means that n is odd,
1
by definition of gn . If mps, s1 q is even and n “ mps, s1 q ´ 1, observe that Amps,s q must
1
be diagonalizable with two eigenvalues equal to ´1, thus f mps,s q “ ´1. We conclude
σs gn pαs q “ ´αs , showing that the gn pαs q “ αs .

We return to the situation of a general Coxeter group.

Theorem 21. The map S Ñ GLpV q, s ÞÑ σs extends uniquely to a group homomorphism
σ : W Ñ GLpV q.
1
Proof. In view of (Proposition 7), we have to show that f :“ pσs σs1 qmps,s q P GLpV q is
the identity map for all s, s1 P S with mps, s1 q ă `8. We saw in the previous lemma
that f pαs q “ αs and f pαs1 q “ αs1 .
Observe that the matrix
ˆ _ ˙
αs pαs q αs_ pαs1 q
A :“
αs_1 pαs q αs_1 pαs1 q
has determinant 4 ´ ks,s1 ks1 ,s ą 0 since mps, s1 q ă `8. It follows that for each vector
v P V , we find real numbers c, d P R such that v 1 :“ v ` cαs ` dαs1 satisfies
αs_ pv 1 q “ αs_1 pv 1 q “ 0.
Thus f pv 1 q “ v 1 by definition of f . We conclude that f pvq “ v.

11
 We will later see that this representation is faithful (i.e. an injective function). This
allows us to replace (abstract) Coxeter group elements by concrete matrices.
Exercise 22. Show that σ is injective in the rank 2 case.
Exercise 23. In the situation of Lemma 20 (b), show that gn pαs1 q ď 0 for n ă mps, s1 q.
Show that gn pαs q ď 0 for mps, s1 q ď n ă 2mps, s1 q.
Exercise 24. Let σ _ : W Ñ GLpV _ q be the action of W on the dual space of V , given
by the composing σ with the natural map GLpV q Ñ GLpV _ q. Show that for s P S, the
map σ _ psq : V _ Ñ V _ is given by

σ _ psqpλq “ λ ´ λpαs qαs_ .

5 Length
We saw that the map φ : S Ñ W is injective, so let us write s instead of φpsq when
denoting group elements. The elements of s P W are called simple reflections. We saw
that each w P W can be written as

w “ s1 ¨ ¨ ¨ sn

for some elements s1 , . . . , sn P S. In this case, we say that ps1 , . . . , sn q is a word repre-
senting w. A word that has minimal length among all words representing w is called a
reduced word of w (also reduced expression and reduced decomposition). We define the
length of w to be the length of any reduced word.
We have the important triangle inequality

ℓpw1 w2 q ď ℓpw1 q ` ℓpw2 q.

Moreover, the set S can be identified with the set of elements having length 1. It follows
that |ℓpwsq ´ ℓpwq| ď 1 (and in view of Exercise 10, it must be equal to 1).

Proposition 25. Let w P W and s P S. In case ℓpwsq ě ℓpwq, we have σpwqpαs q ě 0.

If ℓpwsq ď ℓpwq, we have σpwqpαs q ď 0.

Proof. We only show the first statement, as the second one follows from reversing the
roles of w and ws (using wss “ w).
If ℓpwq “ 0, we must have w “ 1 and there is nothing to show. So let ℓpwq ě 1 and
the statement be proved for all elements of smaller length.
If there exists a reduced word of w ending with s, then ℓpwsq “ ℓpwq ´ 1, contradicting
our assumption. Hence we find a reduced word of w ending in some s1 ‰ s. Put
w1 :“ ws1 , so that

σpwqpαs q “ σpw1 qσps1 qpαs q “ σpw1 qpαs q looooooooomooooooooon

´ks1 ,s σpw1 qpαs1 q .
ě0 by induction

12
If σpw1 qpαs q ě 0, we are done. Otherwise we get ℓpw1 sq ă ℓpwq by induction. Define
w2 :“ w1 s. If ℓpw2 s1 q ă ℓpw2 q, define w3 :“ w2 s1 and so forth. After at most ℓpwq steps,
we find an element wn such that wn pαs q, wn pαs1 q ě 0 and

p¨ ¨ ¨ s1 ss1 q,
w “ wn loooomoooon ℓpwq “ ℓpwn q ` n.
n terms

If n ă mps, s1 q, then σp¨ ¨ ¨ s1 ss1 qpαs q “ cαs `dαs1 for real numbers c, d ě 0 by Lemma 20.
Hence σpwqpαs q ě 0 as claimed.
If n ě mps, s1 q, we could write

p¨ ¨ ¨ s1 ss1 sq
ws “wn looooomooooon
n`1 terms
1 1
“ wn ps ss ¨ ¨ ¨ q .
Def. Coxeter group loooomoooon
2mps,s1 q´n´1 terms

This implies

ℓpwsq ď ℓpwn q ` 2mps, s1 q ´ n ´ 1 ă ℓpwn q ` n “ ℓpwq,

contradiction.

The root system of pW, Sq is defined as

Φ “ tσpwqpαs q | w P W, s P Su Ď V.

It follows from Proposition 25 that each root is either positive (i.e. ě 0) or negative (i.e.
ď 0). Write Φ` for the set of positive roots and Φ´ for the set of negative roots.
For each w P W , we define the inversion set of w to be

invpwq “ tα P Φ` | wα P Φ´ u.

Theorem 26. The cardinality of the set invpwq is equal to ℓpwq. More explicitly, given
any reduced word w “ s1 ¨ ¨ ¨ sℓ , the roots

α1 “ σpsℓ ¨ ¨ ¨ s2 qpαs1 q, α2 “ σpsℓ ¨ ¨ ¨ s3 qpαs2 q, . . . , αℓ “ αsℓ

are pairwise distinct and invpwq “ tα1 , . . . , αℓ u.

Proof. If w “ s1 ¨ ¨ ¨ sℓ is a reduced word, then for each i P t1, . . . , ℓu, the root αi “
σpsℓ ¨ ¨ ¨ si`1 qpαsi q must be positive with

σpwqpαi q “ σps1 ¨ ¨ ¨ si´1 qp´αsi q ď 0

by Proposition 25. If it happens that αi “ αj for some i ă j, then

αsi “ σpsi`1 ¨ ¨ ¨ sℓ qpαi q “ σpsi`1 ¨ ¨ ¨ sℓ qpαj qσpsi`1 ¨ ¨ ¨ sj´1 qp´αsj q

13
contradicts the fact that si`1 ¨ ¨ ¨ sj is a reduced word. Hence the αi are pairwise distinct
and w has at least ℓpwq inversions.
Conversely, if α P Φ` satisfies wα P Φ´ , there must be an index i P t1, . . . , ℓu such
that

σpsi ¨ ¨ ¨ sℓ qpαq P Φ´ , σpsi`1 ¨ ¨ ¨ sℓ qpαq P Φ` .

It suffices to show that the only positive root β P Φ` with σpsi qpβq P Φ´ is β “ αsi . By
definition of σ, the conditions β P Φ` and σpsi qpβq P Φ´ imply that β P Rą0 αsi .
We claim that s P S and w P W are such that σpwqpαs q is of the form cαsi for some
c P Rě0 , then c “ 1, via induction on ℓpwq.
Choose some s1 with ℓpws1 q ă ℓpwq and write w as in Proposition 25 as

w “ wn p¨ ¨ ¨ s1 ss1 q,
loooomoooon σpwn qpαs q ě 0, σpwn qpαs1 q ě 0.
n terms

We may write σp¨ ¨ ¨ s1 ss1 qpαs q “ cs αs ` cs1 αs1 with cs , cs1 ě 0. Then cs σpwn qpαs q and
cs1 σpwn qpαs1 q are both ě 0 and their sum has only a non-zero αsi -coefficient. It follows
that

cs σpwn qpαs q, cs1 σpwn qpαs1 q P Rě0 αsi .

In particular, these two vectors are linearly dependent. Since αs and αs1 are not, we
get cs “ 0 or cs1 “ 0. It follows that σp¨ ¨ ¨ s1 ss1 qpαs q is already a positive multiple of
either αs or αs1 . In view of Lemma 20, it must be equal to αs or αs1 . We see that
αsi “ cσpwqpαs q is equivalent to

αsi P tcσpwn qpαs q, cσpwn qpαs1 qu.

In both cases, we conclude c “ 1 by induction. This finishes the proof.

Corollary 27. The group homomorphism σ : W Ñ GLpV q is injective.

Proof. If w lies in the kernel, invpwq “ H by definition. In view of the previous theorem,
this means ℓpwq “ 0, i.e. w “ 1.

We will in the future frequently omit the explicit map σ and simply write wα instead
of σpwqpαq.

Lemma 28. Let w1 , w2 P W . Then

ℓpw1 w2 q “ ℓpw1 q ` ℓpw2 q ´ 2#pinvpw1 q X invpw2´1 qq.

14
Proof. We calculate

# invpw1 w2 q “#tα P Φ` | w1 w2 α ď 0 and w2 α ď 0u

` #tα P Φ` | w1 w2 α ď 0 and w2 α ě 0u
“#tα P Φ` | w2 α ď 0u
´ #tα P Φ` | w1 w2 α ě 0 and w2 α ď 0u
` #tα P Φ | w1 w2 α ď 0 and w2 α ě 0u
´ #tα P Φ´ | w1 w2 α ď 0 and w2 α ě 0u
“# invpw2 q ´ #tβ P Φ´ | w1 β ě 0 and w2´1 β ě 0u
` # invpw1 q ´ tβ P Φ` | w1 β ď 0 and w2´1 β ď 0u
“ℓpw1 q ` ℓpw2 q ´ 2#pinvpw1 q X invpw2´1 qq.

We finish this section by giving a different perspective on the set of positive roots.

Definition 29. A reflection in W is any element conjugate to an element of S. The set

of reflections is denoted by T Ă W .

Proposition 30. There is a bijective map

s‚ : Φ` Ñ T

that can be evaluated as follows: If α P Φ` is written as α “ vpαs q with v P W and

s P S, then sα “ vsv ´1 .

Proof. Certainly every positive root is of the form vαs for some v P W and s P S. By
definition, every t P T is of the form vsv ´1 for some v P V and s P S. Replacing v by vs
does not change t but changes vαs to its negative, so we always can find a v such that
t “ vsv ´1 and vαs ě 0.
However, none of these representations are unique. We have to show that for all
v, v 1 P W and s, s1 P S with vαs , v 1 αs1 ě 0 that

vαs “ v 1 αs1 ðñ vsv ´1 “ v 1 s1 pv 1 q´1 .

Replacing pv, v 1 q by ppv 1 q´1 v, 1q, we may assume that v 1 “ 1. Now we evaluate for x P V

vsv ´1 x “ σpvq ˝ σs ˝ σpv ´1 qpxq “ x ´ αs_ pv ´1 xqvαs .

Observe that vαs P invpvsv ´1 q.

If vsv ´1 “ s1 is a simple reflection, it follows from Theorem 26 that vαs “ αs1 is the
only element in invpvsv ´1 q “ invps1 q.
Conversely, if vαs “ αs1 , the map vsv ´1 only changes the αs1 -coordinates of its input.
It follows that each element in invpvsv ´1 q must be a multiple of αs1 . We saw in Theo-
rem 26 that the only root in Rě0 αs1 is αs1 itself, such that invpvsv ´1 q “ tαs1 u follows.
In view of Theorem 26, we get vsv ´1 “ s1 , finishing the proof.

15
Exercise 31. Let W “ Sn be the symmetric group and S “ ts1 , . . . , sn´1 u the standard
transpositions.
(a) Show that the positive roots are given precisely by the roots of the form

αi,j`1 “ αsi ` αsi`1 ` ¨ ¨ ¨ ` αsj´1 ` αsj

for 1 ď i ď j ď n ´ 1.

(b) Show that αi,j`1 lies in invpwq if and only if wpiq ą wpj ` 1q.
Exercise 32. Let w “ s1 ¨ ¨ ¨ sn be a reduced word. Define the support of w to be the
smallest subset S 1 Ď S such that invpwq is contained in the subspace of V spanned by
S 1 . Show that

supppwq “ ts1 , . . . , sn u.

Exercise 33. Show that each t P T has a symmetric reduced expression, i.e. a reduced
expression t “ s1 ¨ ¨ ¨ sn such that sn`1´i “ si for i “ 1, . . . , n (conversely, any symmetric
expression in the above form will yield an element of T or 1).

6 Structure theory of Coxeter groups

Let W be a group and S Ď W a subset such that S generates W , and each element in
S has order 2 (in particular, 1 R S). We define a length function ℓ : W Ñ Zě0 as in
the case of Coxeter groups. Let T Ă W denote the set of all elements in W which are
conjugate to some element of S. The goal of this section is to introduce and halfway
prove the following classification result.
Theorem 34. The following are equivalent:
(a) pW, Sq is a Coxeter group, i.e. there exists a function m : S ˆ S Ñ Z Y t`8u
such that W can be identified as the Coxeter group arising from the Coxeter system
pS, mq.

(b) pW, Sq satisfies the Strong Exchange Condition: For each t P T and s1 , . . . , sn P S
such that

ℓps1 ¨ ¨ ¨ sn tq ď ℓps1 ¨ ¨ ¨ sn q,

we have

s1 ¨ ¨ ¨ sn t “ s1 ¨ ¨ ¨ si´1 si`1 ¨ ¨ ¨ sn for some i P t1, . . . , nu.

(c) pW, Sq satisfies the Weak Exchange Condition: For each reduced word w “ s1 ¨ ¨ ¨ sn P
W (i.e. s1 , . . . , sn P S and ℓpwq “ n) and each s P S such that ℓpwsq ď ℓpwq, we
have

ws “ s1 ¨ ¨ ¨ si´1 si`1 ¨ ¨ ¨ sn for some i P t1, . . . , nu.

16
(d) pW, Sq satisfies the Deletion Condition: For each non-reduced word w “ s1 ¨ ¨ ¨ sn P
W (i.e. ℓpwq ă n), there exist indices 1 ď i ă j ď n such that
w “ s1 ¨ ¨ ¨ si´1 si`1 ¨ ¨ ¨ sj´1 sj`1 ¨ ¨ ¨ sn .

We start by proving (a) ùñ (b).

Lemma 35. Let pW, Sq be a Coxeter group, w “ s1 ¨ ¨ ¨ sn a not necessarily reduced word
and α P invpwq. Then there exists an index i P t1, . . . , nu such that α “ σpsn sn´1 ¨ ¨ ¨ si`1 qpαsi q.
For any such index, we have wsα “ s1 ¨ ¨ ¨ si´1 si`1 ¨ ¨ ¨ sn and ℓpwsα q ă ℓpwq.
Proof. Since α ě 0 and wα ď 0, we find an index i such that
σpsi`1 ¨ ¨ ¨ sn qpαq ě 0, σpsi ¨ ¨ ¨ sn qpαq ď 0.
In particular, σpsi`1 ¨ ¨ ¨ sn qpαq lies in invpsi q, which we know to be tαsi u. Thus α “
σpsn ¨ ¨ ¨ si`1 qpαsi q. By definition, it follows that
sα “ sn ¨ ¨ ¨ si`1 si si`1 ¨ ¨ ¨ sn .
The claimed expression for wsα is easily verified. The final length condition does not
depend on the particular word w “ s1 ¨ ¨ ¨ sn , so we may replace it by a reduced word,
and then wsα “ s1 ¨ ¨ ¨ spi ¨ ¨ ¨ sn is an even shorter word for wsα .

Certainly, the Strong Exchange Condition implies the Weak Exchange Condition.
Lemma 36. It is always true that the Weak Exchange Condition is equivalent to the
Deletion Condition.
Proof. WEC ùñ DC: Assume that pW, Sq satisfies the Weak Exchange Condition and
that w “ s1 ¨ ¨ ¨ sn is a non-reduced word. This means that the function
f : t1, . . . , nu Ñ Z, j ÞÑ ℓps1 ¨ ¨ ¨ sj q
is not strictly increasing. In other words, we find an index such that f pjq ě f pj ` 1q, or
explicitly
ℓps1 ¨ ¨ ¨ sj sj`1 q ď ℓps1 ¨ ¨ ¨ sj q.
We choose j minimally with this property, then s1 ¨ ¨ ¨ sj is reduced. Apply the Weak
Exchange Condition to the reduced word s1 ¨ ¨ ¨ sj and the element sj`1 P S to find the
index i needed for the Deletion Condition.
DC ùñ WEC: If s1 ¨ ¨ ¨ sn is reduced but s1 ¨ ¨ ¨ sn s is not, apply the Deletion Condition
to the non-reduced word s1 ¨ ¨ ¨ sn s. It is impossible to find indices i ă j ď n such that
s1 ¨ ¨ ¨ sn s “ s1 ¨ ¨ ¨ spi ¨ ¨ ¨ spj ¨ ¨ ¨ sn s,
as this would contradict s1 ¨ ¨ ¨ sn being reduced. Thus the Deletion Condition must yield
an index i with
s1 ¨ ¨ ¨ sn s “ s1 ¨ ¨ ¨ spi ¨ ¨ ¨ sn ,
as required for the Weak Exchange Condition.

17
 We see that a Coxeter group must satisfy both Exchange Conditions as well as the
Deletion Condition. This is enough for the structure theory of Coxeter groups. The
reverse implications however are still extremely useful, as they allow us to give examples
of Coxeter groups.
Exercise 37. Using Theorem 34, prove the following result due to Deodhar [Deo89]:
Let pW, Sq be a Coxeter group and let W 1 Ď W be a subgroup generated by a set of
reflections (i.e. a subset of T ). Define

S 1 :“ tt P T X W 1 | ℓptt1 q ą ℓptq for each t1 P T X W 1 zttuu.

Show that pW 1 , S 1 q is a Coxeter group.

Exercise 38. We might define the following Extra Weak Exchange Condition:
For each reduced word w “ s1 ¨ ¨ ¨ sn and each s P S such that ℓpwsq ă ℓpwq, we have
ws “ s1 ¨ ¨ ¨ spi ¨ ¨ ¨ sn for some index i.
The only difference to the Weak Exchange Condition is that this version requires a
strict inequality ℓpwsq ă ℓpwq.
Give an example of a group W with a generating set of involutions S that is not a
Coxeter group but satisfies this Extra Weak Exchange Condition (justifying why we did
not use this extra weak version in Theorem 34)

7 Matsumoto’s Theorem
In this section, we complete the proof of Theorem 34. Let pW, Sq be a group together
with a generating set of involutions. Define the function m : S ˆ S Ñ Z Y t`8u by

mps, s1 q “ orderpss1 q,
1
i.e. mps, s1 q is the smallest positive integer such that pss1 qmps,s q “ 1 in W (or `8 if no
such integer exists).

Definition 39. Let ps1 , . . . , sn q be a word in S. We define the following transformations,

yielding a new word in S:

(a) A braid move consists of replacing a subword of the form ps, s1 , s, s1 , . . . q with length
mps, s1 q ă `8 by the word ps1 , s, s1 , s, . . . q of the same length.

(b) A nil move consists of replacing a subword of the form ps, sq by the empty word.

The following is an example of two braid moves followed by a nil move in the Coxeter
group S4 .

s1 s2 s1 s3 s2 s3 Ñ s2 s1 s2 s3 s2 s3 Ñ s2 s1 s3 s2 s3 s3 Ñ s2 s1 s3 s2 .

By definition of m and the fact that S consists of involutions, it follows that two words
related by a sequence of braid and nil moves will represent the same element of W .

18
Theorem 40 ([Mat64]). Suppose that pW, Sq satisfies the Weak Exchange Condition.
Let ps1 , . . . , sn q and ps11 , . . . , s1ℓ q be two words in S such that

w :“ s1 ¨ ¨ ¨ sn “ s11 ¨ ¨ ¨ s1ℓ P W

and ℓpwq “ ℓ (i.e. the second word is reduced, the first might not be). Then ps1 , . . . , sn q
can be transformed into ps11 , . . . , s1ℓ q by a sequence of braid moves and nil moves.

Proof. Induction on n ě 0. If n “ 0, the first word is empty, thus w “ 1 and we must

have ℓ “ 0 as well. The claim is trivially satisfied.
Let us now assume that n ą 0 and the corresponding claim has been proved for all
words in S of length ă n.
If ℓpwsn q ą ℓpwq, then s1 ¨ ¨ ¨ sn´1 “ s11 ¨ ¨ ¨ s1ℓ sn with the right-hand side reduced. By
induction, we can transform s1 ¨ ¨ ¨ sn into s11 ¨ ¨ ¨ s1ℓ sn sn using a sequence of braid moves
and nil moves (where we don’t touch the rightmost letter). Apply another nil move to
transform this into s11 ¨ ¨ ¨ s1ℓ .
Certainly, if sn “ s1ℓ , we get

s1 ¨ ¨ ¨ sn´1 “ s11 ¨ ¨ ¨ s1ℓ´1

and can apply induction immediately.

We may and will assume that ℓpwsn q ď ℓpwq with sn ‰ s1ℓ . Applying the Weak
Exchange Condition to s11 ¨ ¨ ¨ s1ℓ and sn shows that ℓpwsn q ă ℓpwq.
We consider alternating words, i.e. words in ts1ℓ , sn u of the form

p. . . , s1ℓ , sn , s1ℓ , sn , . . . q

of arbitrary length, without specifying whether the words starts/ends in sn or s1ℓ .

Pick such an alternating word a “ p. . . , s1ℓ , sn , s1ℓ , sn , . . . q of maximal length m ě 0
such that

ℓpwaq “ ℓpwq ´ m,

here, a P W denotes the element in W represented by the word a. Write the letters of a
as a “ pa1 , . . . , am q. Define am`1 to be the unique element of tsn , s1ℓ u not equal to am .
Chose a reduced word ω for wa, and denote by ωa the word obtained by composing
ω with a. By choice of a, ωa is a reduced word for w.
We saw that ℓpwam`1 q ă ℓpwq, since am`1 P ts1ℓ , sn u. Thus we may apply the Weak
Exchange Property to the reduced word ωa and the simple reflection am`1 P S. We
conclude that one of the following statements must hold true:

(1) wam`1 is represented by a word of the form ω 1 a with ω 1 being obtained from ω by
deleting one letter or

(2) wam`1 is represented by a word of the form ωa1 with a1 being obtained from a by
deleting one letter.

19
 In case (1), we would get a reduced word for w given by ω 1 aam`1 , which contradicts
the choice of m. So we must have (2).
Case (2) means that a is a reduced word but aam`1 is not. Since ℓpaam`1 q ě ℓpaq ´ 1,
the word a1 must be reduced. It follows that a1 is obtained from a by deleting the
leftmost letter a1 (as a1 cannot contain a subword of the form ss, nor can a1 end with
am`1 ). We see that

a1 ¨ ¨ ¨ am`1 “ a2 ¨ ¨ ¨ am P W,
or equivalently a1 ¨ ¨ ¨ am`1 am ¨ ¨ ¨ a2 “ 1.

It follows that mps, s1 q ď m, in particular mps, s1 q ă `8. Moreover, m ď mps, s1 q

follows since a is reduced. Hence m “ mps, s1 q. Denote ã “ pa2 , . . . , am`1 q. Then we
can perform a braid move to change between ωa and ωã, both of which are reduced
words for w.
One of these reduced words ends with sn . If, say, ωa ends with sn , then ωã ends with
sℓ and we can use induction to get
1

braid & nil moves braid move braid moves

ps1 , . . . , sn q ωa ωã ps11 , . . . , s1ℓ q.

If ωa ends with s1ℓ , then ωã ends with sn and we just have to interchange ωa with ωã in
the above picture.

From Matsumoto’s theorem, we can immediately see the remaining direction (c) ùñ
(a) of Theorem 34. The interested reader may take it as an exercise to formalize such a
proof.
Remark 41. Associated with a Coxeter system pS, mq, we can define the Braid group to
be the group presented as follows: The generating set is given by S, and the relations
are

pss 1 1
ps1 ss1 s ¨ ¨ ¨ q
ss ¨ ¨ ¨ q “ looooomooooon
looooomooooon @s, s1 P S s.th. mps, s1 q ă `8.
mps,s1 q terms mps,s1 q terms

There is a natural map from the Braid group to the associated Coxeter group, which is
of course surjective but not injective. However, Matsumoto’s theorem asserts that the
take a reduced word map is well-defined from the Coxeter group to the Braid group.
Braid groups of type An have a nice geometric interpretation, where group elements
can be understood as braids of n ` 1 strings and group multiplication being composition.
Hence the name braid move. We might talk a bit more in depth about braid groups
later, depending on time constraints and participants’ interest.
Exercise 42. Given any word w “ s1 ¨ ¨ ¨ sn , describe an algorithm to check if the word is
reduced or not based on Matsumoto’s theorem. Give a second algorithm that is based
on Proposition 25.
Exercise 43. Using the results of section 1, prove that Sn is indeed a Coxeter group.

20
8 Parabolic subgroups
Consider a Coxeter group pW, Sq.
Definition 44. For each subset I Ď S, we define WI to be the subgroup of W generated
by I.
Subgroups of this form are known as standard parabolic subgroups. By parabolic sub-
groups, one understands subgroups equal or conjugate to standard parabolic subgroups,
depending on the author.
Using Matsumoto’s theorem, it is easy to see that w P WI if and only if supppwq Ď I.
It follows that WI X WJ “ WIXJ for I, J Ď S.
Definition 45. For L, R Ď S, we define L W R Ď W to be the subset of all those w P W
that satisfy

@l P L : ℓplwq ą ℓpwq and @r P R : ℓpwrq ą ℓpwq.

If one of the sets is empty, we write

W R :“ H W R “ tw P W | ℓpwsq ą ℓpwq @s P Ru,

L
W :“ L W H “ tw P W | ℓpswq ą ℓpwq @s P Lu.

Lemma 46. Let I Ď S and define VI Ď V to be the vector space spanned by the αs for
s P I. Let w P W . Then

w P WI ðñ invpwq Ď VI
w P W I ðñ invpwq X VI “ H.

Proof. The first equivalence follows from the above remark on the support and Exer-
cise 32.
If w R W I , then αs P invpwq for some s P I. Conversely, if w P W I , then wαs ě 0 for
all s P I. Hence wα ě 0 for all α P Φ` X VI .

Corollary 47. For I Ď S, w1 P W I and w2 P WI , we have

ℓpw1 w2 q “ ℓpw1 q ` ℓpw2 q.

Proof. Indeed, observe that w2 P WI such that invpw1 q X invpw2´1 q “ H follows. Con-
clude using Lemma 28.

Proposition 48. Let L, R Ď S and w P W .

(a) The double coset

WL wWR “ twL wwR | wL P WL , wR P WR u Ď W

contains precisely one element of minimal length, denoted L wR . We have

pWL wWR q X L W R “ tL wR u.

21
(b) There exist elements wL P WL and wR P WR such that

w “ wL L wR wR and ℓpwq “ ℓpwL q ` ℓpL wR q ` ℓpwR q.

If L “ H or R “ H, then the elements wL , wR are uniquely determined.

Proof. (a) Let w1 P WL wWR be an element of minimal length. It is clear that such an
element must lie in L W R .
Pick any element w2 P pWL wWR q X L W R . We need to prove that w1 “ w2 .
Since w2 P WL w1 WR , we find elements wL P WL and wR P WR such that w2 “
wL w1 wR .
For any such elements wL , wR , we apply Corollary 47 to see
´1 ´1 ´1
ℓpw2 q ` ℓpwR q ´ ℓpwL q “ ℓpw2 wR q ´ ℓpwL q ď ℓpwL w2 wR q “ ℓpw1 q ď ℓpw2 q.

Hence ℓpwR q ď ℓpwL q. An analogous argument shows ℓpwL q ď ℓpwR q, so that

ℓpwL q “ ℓpwR q.
Among all pairs pwL , wR q P WL ˆ WR with w2 “ wL w1 wR , pick one such that
ℓpwL q “ ℓpwR q gets minimal. If this length is zero, we see w1 “ w2 .
Otherwise, we find a simple reflection s P R such that wR αs ď 0. Since w1 P W R ,
we see w1 wR αs ď 0. However, w2 P W R again such that

wL w1 wR αs “ w2 αs ě 0 ùñ β :“ ´w1 wR αs P invpwL q Ď VL .

We conclude

w2 “ wL w1 wR “ loomoon
pwL sβ q w lo
pw R sq
omo on,
PWL PWR

contradicting minimality of ℓpwR q.

(b) Among all pairs pwL , wR q P WL ˆ WR with w “ wL L wR wR , pick one such that
ℓpwR q is as small as possible. We already know from Corollary 47 that

ℓpwL L wR q “ ℓpwL q ` ℓpL wR q.

In view of Lemma 28, it suffices to show that there exists no root α P Φ` with
wL L wR α ď 0 and wR
´1
α ď 0.
Indeed, if α was such a root, then L wR α ě 0 as L wR P W R . Hence β :“ L wR α P
invpwL q Ď VL and we conclude

w “ wL L wR wR “ lo
womo L R
L soβn loooomoooon
w s α wR ,
PWL PWR

contradicting minimality.
It remains to observe that if L “ H or R “ H, we must have wL “ 1 resp. wR “ 1,
and then the other element is uniquely determined by w “ wL L wR wR .

22
Example 49. Consider the symmetric group W “ Sn . Decompose the set t1, . . . , nu into
blocks of adjacent numbers, i.e. subsets

Bk “ tbk , bk ` 1, . . . , bk`1 ´ 1u

for numbers 1 “ b1 ă ¨ ¨ ¨ ă bℓ “ n ` 1. Call a permutation w P Sn block-preserving if

wBk “ Bk for all blocks Bk .
Let I Ď ts1 , . . . , sn´1 u be the subset of those standard transpositions si “ pi i`1q such
that i and i ` 1 lie in the same block. Then WI is precisely the group of block-preserving
permutations. Note that each standard parabolic subgroup of Sn arises in this way.
The set W I consists of all permutations w P W which are block monotonic, i.e. satisfy
wpiq ă wpjq whenever i ă j lie in the same block.
Exercise 50. Let w P S5 be the permutation w “ p1 2 3qp4 5q.

(a) Determine the length and the support of w.

(b) Let L “ ts2 , s3 u and R “ ts2 , s3 , s4 u. Compute L wR and provide elements wL , wR

as in Proposition 48 (b).

Exercise 51. Let R1 , R2 Ď S. Show that W R1 YR2 “ W R1 X W R2 .

Exercise 52. Give an example of a Coxeter group pW, Sq, an element w P W and subsets
L, R Ď S such that the elements wL , wR as in Proposition 48 (b) are not unique.
Exercise 53. In the setting of a general Coxeter group pW, Sq with w P W and L, R Ď S,
prove that
L
wR “ L pwR q :“ L pH wR qH .

Exercise 54. Let L, R Ď S and w P L W R . Let R1 Ď R be the subset of those reflections

s P R with wsw´1 P L. Show that there is a bijective map

pWL ˆ WR qä
„ Ñ WL wWR ,

where we identify the pairs pwL , swR q and pwL wsw´1 , wR q for wL P WL , wR P WR and
s P R1 .
Exercise 55. Let L, R Ď S and w P W . Show that ℓpL wR q is given by the number of
roots α P Φ` zVR such that wα P Φ´ zVL .

9 Bruhat order
Let pW, Sq be a Coxeter group.

Definition 56. If w P W and t P T are such that ℓpwq ă ℓpwtq, we write w Ñ wt. We
say that w ď w1 in the Bruhat order if there is a sequence of such arrows

w “ w1 Ñ w2 Ñ ¨ ¨ ¨ Ñ wm “ w 1 .

23
 We note the following immediate properties.

Lemma 57. Let v, w P W and t P T .

(a) If v ă w then ℓpvq ă ℓpwq.

(b) The Bruhat order is a partial order.

(c) We have 1 ď w.

(d) w ă tw if and only if ℓpwq ă ℓptwq.

(e) v ď w if and only if v ´1 ď w´1 .

Example 58. Let G Ů be a Lie group, B Ă G a Borel subgroup and W the Weyl group.
We know that G “ wPW BwB.
The group G carries the structure of a topological space, so we can consider the closure
of a double coset BwB. This is naturally a union of double cosets again, so we may ask
which double cosets Bw1 B appear as subsets of BwB. It turns out that this is the case
if and only if w1 ď w in the Bruhat order. In other words,
ğ
BwB “ BwB.
w1 ďw

We give equivalent descriptions of this order.

Definition 59. A Bruhat cover is a pair pw, w1 q P W ˆ W such that w ă w1 and

ℓpw1 q “ ℓpwq ` 1. We write w Ì w1 .

Note that if pw, w1 q is a Bruhat cover, then w´1 w1 P T .

Theorem 60. Let w “ s1 ¨ ¨ ¨ sℓpwq P W be a reduced word and v P W . The following

are equivalent.

(a) v ď w in the Bruhat order.

(b) There are indices 1 ď i1 ă ¨ ¨ ¨ ă ik ď ℓpwq such that v “ si1 ¨ ¨ ¨ sik .

(c) There are indices 1 ď i1 ă ¨ ¨ ¨ ă ik ď ℓpwq such that v “ si1 ¨ ¨ ¨ sik and k “ ℓpvq.

(d) There is a sequence of Bruhat covers

v “ v0 Ì v1 Ì ¨ ¨ ¨ Ì vℓpwq´ℓpvq “ w.

Proof. (a) ùñ (b): Take a sequence

w “ w1 Ñ ¨ ¨ ¨ Ñ wm “ v

as in the definition of the Bruhat order and apply the Strong Exchange Property m ´ 1
times.

24
 (b) ùñ (c): Any word representing v P W contains a reduced subword representing
v, this follows from the Deletion Property.
(c) ùñ (d): Induction on ℓpwq ´ ℓpvq. If ℓpvq “ ℓpwq, there is nothing to show.
Otherwise, we can denote by h P t1, . . . , ℓpwqu the smallest value not equal to any of
the i1 , . . . , ik . Among all reduced subwords representing v, choose one such that this
h P t1, . . . , ℓpwqu is as large as possible. We let v 1 P W be the element obtained by
inserting si into this reduced word for v, i.e.
v 1 “ s1 ¨ ¨ ¨ si´1 sh sih ¨ ¨ ¨ sik
(using the fact ij “ j for j ă h). Let α :“ sik ¨ ¨ ¨ sih pαsh q P Φ, where αsh denotes the
simple root associated with the simple reflection sh .
If α P Φ´ , we can apply the Weak Exchange Property to write
si sih ¨ ¨ ¨ sik “ sih ¨ ¨ ¨ sij´1 sij`1 ¨ ¨ ¨ sik
for some j P th, . . . , ku. Then
v “ s1 ¨ ¨ ¨ sh sih ¨ ¨ ¨ sij´1 sij`1 ¨ ¨ ¨ siℓpwq
is a reduced word representing v starting with s1 ¨ ¨ ¨ sh , contradicting the choice of h.
If α P Φ` , we see that
vα “ s1 ¨ ¨ ¨ sh´1 pαsh q P Φ`
as s1 ¨ ¨ ¨ sℓpwq is reduced. Hence ℓpv 1 q “ ℓpvsα q ą ℓpvq. In particular, the above word for
v 1 must be reduced and v Ì v 1 . We can apply induction to finish the proof.
The final implication (d) ùñ (a) is trivial.
Corollary 61. Let v, w P W and s P S such that ws ă w.
(a) If vs ă v, then v ď w if and only if vs ď ws.
(b) If vs ą v, then v ď w if and only if v ď ws.
Proof. Pick a reduced word of w “ s1 ¨ ¨ ¨ sm ending with s and apply the subword
criterion Theorem 60 (c). Then part (b) as well as the implication “ ðù ” in (a) follow
immediately.
It remains to show the implication “ ùñ ” in (a). If v ď w, then vs ď v ď w. Apply
(b) to get vs ď ws.
Exercise 62. Let s P S and w P W . Show that s ď w if and only if s P supppwq.
Exercise 63. Give an example of a Coxeter group pW, Sq, two elements v ď w in W and
a reduced word
v “ s1 ¨ ¨ ¨ sℓpvq
that is not a subword of any reduced word for w.
Exercise 64. Show that if pWI , Iq is a standard parabolic subgroup of pW, Sq, then the
Bruhat order on WI is the restriction of the Bruhat order on W .
Exercise 65. Show that for any finite subset F Ď W , there exists some w P W with
v ď w for all v P F .

25
10 Bruhat order on parabolic quotients
Let pW, Sq be a Coxeter group and L, R Ď S be two arbitrary subsets.
Lemma 66. Let v P L W R and w P W . Then
v ď w ðñ v ď L wR .
Proof. Decompose w “ wL L wR wR as in Proposition 48 (b). Choose reduced words for
wL , L wR and wR and concatenate them to a reduced word for w.
Since v P L W R , no reduced word for v can start with an element of L nor end with an
element of R. Applying the subword criterion from Theorem 60 (c), we conclude.

Corollary 67. If v Ì w with v P L W R , then w P L W R or w “ sv for some s P L or

w “ vs for some s P R.
Proof. v ď L wR ď w. One of them must be an equality.

Proposition 68. If v, w P W R , then v ď w if and only if there exists a chain of Bruhat

covers
v “ v0 Ì v1 Ì ¨ ¨ ¨ Ì vℓpwq´ℓpvq “ w

consisting of elements in W R .
Proof. Induction on ℓpwq. If w “ 1, there is nothing to show.
Pick a simple reflection s with sw ă w. Then sw P W R .
If sv ą v, we get v ď sw Ì w by Corollary 61 and are done.
We may hence assume that v ą sv P W R . By induction, there exists some v 1 P W R
with sv ď v 1 Ì sw.
If sv 1 ă v 1 , we see v ď v 1 Ì sw Ì w and are done by induction.
If sv 1 ą v 1 , then v ď sv 1 Ì w. In case sv 1 P W R , we are again done by induction. If
however sv 1 R W R , Corollary 67 shows v 1 “ psv 1 qR . Hence v ď v 1 Ì sv 1 Ì w. We are
again done by induction.

The main result of this section is the following result, known as Deodhar’s lemma.
Lemma 69 ([Deo77]). Let tRr Ď SurPρ be a family of subsets of S and R “ rPρ Rr its
Ş
intersection.
For v, w P W , we have v R ď w if and only if v Rr ď w for each r P ρ.
Proof. Induction on ℓpwq. If ℓpwq “ 0, we get v Rr “ 1 so v P WRr for each r P ρ. In
particular, v P WR , so that v R “ 1.
Let now ℓpwq ą 0 and pick a simple reflection s with sw ă w. Let L “ tsu.
By Corollary 61, we have v R ď w if and only if L v R ď sw (using Exercise 53).
Similarly, v Rr ď w if and only if L v Rr ď sw. We summarize:
ind.
v R ď w ðñ v ď sw ðñ @r :
L R L Rr
v ď sw ðñ @r : v Rr ď w.
This finishes the induction and the proof.

26
 The two-sided analogue of Deodhar’s lemma holds true as well.

Theorem 70. Let λ, ρ be arbitrary indexing sets and pick subsets

tLℓ Ď SuℓPλ , tRr Ď SurPρ .

Define
č č
L :“ Lℓ , R :“ Rr .
ℓPλ rPρ

For v, w P W , we have L v R ď w if and only if for each pℓ, rq P λ ˆ ρ, we have

Lℓ Rr
v ď w.

Proof. The direction “ ùñ ” is trivial since Lℓ v Rr ď L v R .

Suppose now that Lℓ v Rr ď w for all ℓ, r. In view of Lemma 69, it suffices to show
L v Rr ď w for all r P ρ. So fix r P ρ for the remainder of the proof.

Apply Lemma 69 to the sets Lℓ and the elements

´ ¯´1
v Rr , w´1 P W

to see that
L Rr
v ď w ðñ @ℓ : Lℓ Rr
v ď w.

This finishes the proof.

A typical, and arguably optimal, choice for given v, w is to let

λ “ ts P S | sv ă vu, ρ “ ts P S | vs ă vu
Lℓ “ Sztℓu and Rr “ Sztru.

Then L v R “ v. The elements Lℓ v Rr are usually much smaller than v, and come with
a bit of extra structure (i.e. all reduced words starting and ending with the same fixed
simple reflections).
Exercise 71. Let L, R Ď S and v ď w. Show that L v R ď L wR .
Exercise 72. Let v, w P L W R and v ď w. Show that the following are equivalent:

• For each u P W with v ď u ď w, we have u P L W R .

• For each s P L, we have sv ę w and for each s P R, we have vs ę w.

Exercise 73. Give an example of a Coxeter group pW, Sq, subsets L, R Ď S and two
elements v, w P L W R such that v ă w, ℓpwq ´ ℓpvq ě 2 and there is no u P L W R with
v ă u ă w (explaining why we can prove Proposition 68 only for one-sided quotients).

27
11 The Tableau criterion
In this section, we focus on the example of the symmetric group W “ Sn . Recall that
the set of simple reflections is given by the standard transpositions si “ pi i ` 1q. The
set of reflections is T “ tpi jq | i ‰ ju. The set of roots can be identified with

Φ “ tei ´ ej | i ‰ ju Ď V “ tpv1 , . . . , vn q P Rn | v1 ` ¨ ¨ ¨ ` vn “ 0u.

The positive roots are those ei ´ ej where i ă j. It follows w ă wpi jq in the Bruhat
order if and only if wpiq ă wpjq for i ă j.

Lemma 74. Let i, j P t1, . . . , n ´ 1u and R :“ Sztsi u, L :“ Sztsj u. For

maxp1, i ` j ´ n ` 1q ď k ď minpi, jq ` 1,

define wk P W to be the permutation

$
’
’
’ h, h ă k,
&h ` j ´ k ` 1,
’
k ď h ď i,
wk : t1, . . . , nu Ñ t1, . . . , nu, h ÞÑ
’
’
’ h ` k ´ i ´ 1, i ` 1 ď h ď i ` j ´ k ` 1,
h ą i ` j ´ k ` 1.
’
%h,

So the permutation matrix of w is given by the block matrix

Idk´1 0 0 0
¨ ˛
˚ 0 0 Idj´k`1 0
w|V : ˚
‹
‹
˝ 0 Idi´k`1 0 0 ‚
0 0 0 Idn´i´j`k´1 .

(a) The set L W R is given by

L
W R “ twk | maxp1, i ` j ´ n ` 1q ď k ď minpi, jq ` 1u.

(b) For w P W , let k :“ 1 ` #th ď i | wphq ď ju. Then L wR “ wk .

(c) If both wk and wk`1 are defined, then wk ą wk`1 in the Bruhat order.

Proof. (a) The claim wk P L W R is easily verified, we have to show that these exhaust
this set. Note that wk “ 1 if k “ minpi, jq ` 1.
Let w P L W R zt1u. This means that whenever h1 ă h2 with wph1 q ą wph2 q, we must
have h1 ď i ă h2 and wph1 q ą j ě wph2 q.
Since w ‰ 1, we let k P t1, . . . , nu be the smallest integer such that wpkq ‰ k.
From wpt1, . . . , k ´ 1uq “ t1, . . . , k ´ 1u, we get wpkq, w´1 pkq ą k. By the above
considerations, it follows that k ď i ă w´1 pkq and wpkq ą j ě k.

28
 Since k ă wpkq ă ¨ ¨ ¨ ă wpiq, we must have w´1 phq ą i for k ď h ă wpkq. Note that
wpi ` 1q ă ¨ ¨ ¨ ă wpnq, so that

wpi ` 1q “ k, wpi ` 2q “ k ` 1, . . . , wpi ` wpkq ´ kq “ wpkq ´ 1.

Since k ă w´1 pkq ă ¨ ¨ ¨ ă w´1 pjq, we must have wphq ą j for k ď h ă w´1 pkq.
Note that w´1 pj ` 1q ă ¨ ¨ ¨ ă w´1 pnq, so that

w´1 pj ` 1q “ k, w´1 pj ` 2q “ k ` 1, . . . , w´1 pj ` w´1 pkq ´ kq “ w´1 pkq ´ 1.

Substituting wpkq “ j ` 1 and w´1 pkq “ i ` 1, we see that wphq “ wk phq for
h ď i ` j ´ k ` 1. It follows that wpt1, . . . , i ` j ´ k ` 1uq “ t1, . . . , i ` j ´ k ` 1u.
Thus wphq ą i ` j ´ k ` 1 for h ą i ` j ´ k ` 1. Since j ` 1 ě k, we get

wpi ` j ´ k ` 2q ă ¨ ¨ ¨ ă wpnq,

so that w “ wk follows.

(b) The cardinality in question does not change when multiplying by a simple reflection
in L on the left, or a simple reflection in R on the right. By (a), it suffices to show
the claim for the wk , which is easily verified.

(c) Suppose that

maxp1, i ` j ´ n ` 1q ď k ď minpi, jq.

Then wk`1 pkq “ k and wk`1 pnq “ n. Thus wk`1 ă w, where w “ wk`1 pk nq is
the product of wk`1 with the reflection pk nq. By (b), we have L wR “ wk . By
Lemma 66, we get wk`1 ď wk .

Definition 75. Let w P W and i, j P t1, . . . , nu. We define

wri, js “ #th ď i | wphq ą ju.

Theorem 76 (Tableau Criterion). Let v, w P W . Then the following are equivalent:

(a) v ď w in the Bruhat order.

(b) For all i, j P t1, . . . , nu, we have vri, js ď wri, js.

(c) For all i P t2, . . . , nu and j P t1, . . . , n ´ 1u such that

v ´1 piq ą v ´1 pi ` 1q and vpjq ą vpj ` 1q,

we have vri, js ď wri, js.

Proof. Since vr1, ‚s “ 0 “ vr‚, ns, we may restrict our attention to i ą 1 and j ă n in
(b). By the previous lemma, the condition vri, js ď wri, js is equivalent to L v R ď L wR
for R “ Sztsi´1 u, L “ Sztsj u. Now (a) ùñ (b) is Exercise 71. The implication (c)
ùñ (a) is the two-sided version of Deodhar’s lemma, Theorem 70.

29
Example 77. Consider the group W “ S4 with the permutations given as follows:

0 1 0 0 0 0 1 0
¨ ˛ ¨ ˛
˚0 0 1 0‹ ˚0 1 0 0‹
v “ p1 3 2q “ ˚
˝1 0 0 0‚, w “ p1 4 3q “ ˝0 0 0
‹ ˚ ‹
1‚
0 0 0 1 1 0 0 0

Inspecting the element v, we only have to check vri, js ď wri, js for i “ j “ 2. We

compute vr2, 2s “ 1 and wr2, 2s “ 1. Thus v ď w.
Exercise 78. Show that wk ą wk`1 by exhibiting a sequence of Bruhat arrows wk Ñ
¨ ¨ ¨ Ñ wk`1 .
Exercise 79. Let v, w P W and i P t1, . . . , nu. Let vi P Zi be the vector obtained by
sorting the sequence vp1q, . . . , vpiq so that it becomes increasing, and similarly wi . Show
that the following are equivalent:

• vri, js ď wri, js for all j P t1, . . . , nu.

• We have vi ď wi component-wise in Zi .

Conclude that v ď w if and only if vi ď wi whenever vpiq ą vpi ` 1q.

Arranging the corresponding vectors v‚ and w‚ nicely, one obtains so-called “tableaux”,
giving the name of the Tableau Criterion. These structures and their relationship to the
combinatorics and representation theory of the symmetric group are beyond the scope
of this course (unless the audience insists), so we refer to [BB05, Appendix A3], as well
as any treatment of the representation theory of symmetric groups (Wikipedia is a good
starting point).

12 Finite Coxeter groups

Let pW, Sq be a Coxeter group.

Proposition 80. Let w0 P W . Then

p@w P W : ℓpwq ď ℓpw0 qq ðñ p@s P S : w0 s ă w0 q .

There exists an element w0 satisfying these properties if and only if W is finite. If such
w0 exists, it is uniquely determined and satisfies moreover the following properties:

(a) w02 “ 1.

(b) ℓpw0 q “ #T “ #Φ` .

(c) For any w P W , we have ℓpww0 q “ ℓpw0 wq “ ℓpw0 q ´ ℓpwq.

(d) For any s P S, we have w0 sw0 P S and w0 pαs q “ ´αw0 sw0 .

(e) Any element w P W satisfies w ď w0 .

30
Proof. If w0 P W has maximal length, then ℓpw0 sq ď ℓpw0 q for all s P S, showing the
implication “ ùñ ”.
If w0 s ă w0 for all s P S, we see w0 pαq ď 0 for all α P Φ` . Thus invpw0 q “ Φ` .
Such an element is uniquely determined and satisfies invpwq Ď invpw0 q for all w P W .
In particular, we get the implication “ ðù ”.
If W0 is finite, then there trivially exists an element of maximal length. If w0 exists,
then each w P W is uniquely determined by invpwq Ď invpw0 q, and there are only finitely
many possibilities for those subsets.
Suppose now that W is finite and w0 satisfies the defining properties.
(a) Since w0´1 has the same length as w0 , it must be equal to w0 .

(b) We already calculated invpw0 q “ Φ` .

(c) The first identity follows from Lemma 28 together with the above observation
invpw´1 q Ď invpw0 q.
The other equation follows from

ℓpww0 q “ ℓppww0 q´1 q “ ℓpw0 w´1 q “ ℓpw0 q ´ ℓpw´1 q “ ℓpw0 q ´ ℓpwq.

(d) Note that

ℓpw0 sw0 q “ℓpw0 q ´ ℓpsw0 q “ ℓpw0 q ´ ℓppsw0 q´1 q

“ℓpw0 q ´ ℓpw0 sq “ ℓpsq “ 1.

Thus invpw0 sw0 q “ tαw0 sw0 u, proving w0 αs “ ˘αw0 sw0 . We already saw w0 αs ď 0.

(e) This is an easy consequence of (c): Concatenate a reduced word for w with a reduced
word for w0 w to obtain a reduced word for w0 containing that reduced word for w
as a subword.
Example 81. In the finite Coxeter group W “ Sn , the longest element is given by
w0 piq “ n ` 1 ´ i.
Definition 82. Let L, R Ď S be subsets and define
´L
W ´R :“ tw P W | sw ă w for all s P L and ws ă w for all s P Ru.

We say that L is spherical if WL is finite.

Lemma 83. Let L, R Ď S and w P W .
(a) The intersection pWL wWR q X ´L W ´R is non-empty if and only if both WL and WR
are spherical.
If both WL and WR are spherical, the double coset WL wWR contains a unique ele-
ment of maximal length, denoted ´L w´R . Moreover,

pWL wWR q X ´L W ´R “ t´L w´R u.

31
(b) Suppose that L, R are spherical. Then there are elements wL P WL , wR P WR such
that ´L w´R “ wL wwR is a length-additive product:

ℓp´L w´R q “ ℓpwL q ` ℓpwq ` ℓpwR q.

(c) Suppose that L, R are spherical. For v P W , we have

L R L R
v ď w ðñ v ď ´L w´R ðñ v ď ´L w´R .

Proof. Let us first consider the case that L is not spherical. Then W ´L “ H: Indeed,
if v P W ´L , we could write v “ v1 v2 with v1 P W L and v2 P WL . Thus v2 αs ď 0 for all
s P L, showing that L is spherical. The same argument works for R being spherical.
Assume for the remainder of the proof that both WL and WR are spherical. The
arguments are very analogous to the proof of Proposition 48, so we leave them as an
exercise to the interested reader (we won’t use these results).

Exercise 84. Let R be spherical. Show that the elements of W ´R are precisely those of
the form ww0 pRq where w P W R and w0 pRq P WR is the longest element.
Exercise 85. Show that for finite Coxeter groups of type Bn for n ě 3, we have w0 sw0 “ s
for all s P S.
Exercise 86. Let pW, Sq be a finite Coxeter group and v, w P W . Show that v ď w if
and only if w0 v ě w0 w.

13 Weak Order
Let pW, Sq be a Coxeter group.
Definition 87. Let v, w P W . We say that v is less than w in the left weak order and
write v ďL w if ℓpwq “ ℓpwv ´1 q ` ℓpvq.
We say that v is less than w in the right weak order and write v ďR w if ℓpwq “
ℓpvq ` ℓpv ´1 wq.
One has to verify that these define partial orders, but this is not too hard. So v ďR w
means that there is some reduced word for w that starts with a reduced word for v.
The inequality v ďL w implies v ď w, hence the name “weak order”. In a finite Coxeter
group, the longest element is the unique maximum with respect to both weak orders.
The concepts of left and right weak order are closely related, e.g. by passing to inverses.
In any case, they ask which pairs pw1 , w2 q P W satisfy the length additivity condition
ℓpw1 w2 q “ ℓpw1 q ` ℓpw2 q. If G is a Lie group with Borel B and Weyl group W , this
length additivity condition is equivalent to the geometric condition that the product of
Borel double cosets / Schubert cells Bw1 B ¨ Bw2 B should be a Borel double coset again,
namely Bw1 w2 B.
We get the following characterization analogous to Bruhat order.
Lemma 88. For v, w P W , the following are equivalent

32
(a) v ďR w.

(b) There is a sequence

t
1 t
2 tn
v “ v1 Ý
Ñ v2 Ý
Ñ ¨ ¨ ¨ ÝÑ vn “ w

as in the definition of Bruhat order, with the additional restriction that ti P S for
all i.

(c) There is a sequence of right weak order covers

v “ v 1 ÌR v 2 ÌR ¨ ¨ ¨ Ì R v n “ w

with vi ÌR vi`1 means vi ďR vi`1 and ℓpvi`1 q ´ ℓpvi q “ 1.

(d) We have invpv ´1 q Ď invpw´1 q.

Proof. (a) ùñ (b): Just pick a reduced word for v ´1 w to obtain the required ti .
(b) ùñ (c): Trivial.
(c) ùñ (d): We may assume n “ 2, i.e. w “ vs for a simple reflection s P S. Then
Proposition 25 shows invpw´1 q “ invpv ´1 q Y tvαs u.
(d) ùñ (a): It follows for each root α P Φ´ with vα P Φ` that w´1 vα P Φ´ . By
Lemma 28, ℓpwq “ ℓpvq ` ℓpv ´1 wq.

Of course, we get an analogous lemma for the left weak order up to passing to inverses.
Lemma 89. For v, w P W , the following are equivalent
(a) v ďL w.

(b) There is a sequence

t
1 t
2 tn
v “ v1 Ý
Ñ v2 Ý
Ñ ¨ ¨ ¨ ÝÑ vn “ w

as in the definition of Bruhat order, with the additional restriction that vi ti vi´1 P S
for all i.

(c) There is a sequence of left weak order covers

v “ v 1 ÌL v 2 ÌL ¨ ¨ ¨ Ì L v n “ w

with vi ÌL vi`1 means vi ďL vi`1 and ℓpvi`1 q ´ ℓpvi q “ 1.

(d) We have invpvq Ď invpwq.

Proposition 90. Let E Ď W be an non-empty set of elements of W . Then an infimum
E exists, i.e. the set of lower bounds

lbpEq “ tv P W | v ďL w for all w P Eu

contains a unique maximum with respect to ďL .

33
Proof. Induction on mintℓpwq | w P Eu. If 1 P E, then 1 is the only lower bound of E.
In the inductive step, let

I :“ ts P S | ws ă w for all w P Eu.

If v P lbpEq and vs ă s, then s P I. So if I “ H, we get lbpEq “ t1u and are done again.
Otherwise, pick s P I define E 1 :“ tws | w P W u. By induction, there is an infimum z
of E 1 . Then zs is an infimum of E.

Exercise 91. Show that the infimum of a set is uniquely determined. Show that if E Ď W
is a non-empty set with an upper bound, then a supremum (= smallest upper bound)
exists.
Exercise 92. Let W “ S5 , v “ p1 3 2 4 5q and w “ p1 4 2 5 3q. Show that v ęL w and
v ęR w.
Exercise 93. Let v, w P W and z P W be the infimum of tv, wu with respect to ďL .
Show that invpzq Ď invpvq X invpwq and give an example where the inclusion is strict.

14 Foldings
In this section, we introduce an important concept to create new Coxeter groups from
old ones, as well as understand automorphisms of Coxeter systems.
Let pW, Sq be a Coxeter group. Let Γ be a group that acts on S such that

mps, s1 q “ mpγ.s, γ.s1 q for all s, s1 P S and γ P Γ.

Then each γ P Γ induces a group automorphism of W that preserves S (i.e. a Coxeter

group automorphism). We denote by W Γ the set of those elements w P W with γ.w “ w
for all γ P Γ.
Lemma 94. Let w P W Γ and I :“ ts P S | ws ă wu. Then I is a spherical subset of S
that is preserved by all γ P Γ. Denoting by w0 pIq P WI the longest element, we have

w “ wI w0 pIq, wI P pW I q X pW Γ q.

Proof. Since w P W ´I , we saw already that I must be spherical. The condition γ.w “ w
implies γpIq “ I for all γ P Γ.
We know that we can always write w as a length additive product w “ wI wI for some
w P W I and wI P WI . By choice of I, we have wI “ w0 pIq. The condition γpIq “ I
I

implies γ.w0 pIq “ w0 pIq, hence w0 pIq P W Γ . We conclude wI P W Γ .

Definition 95. (a) We say that I Ď S is a spherical Γ-orbit if I spherical and of the
form I “ tγ.s | γ P Γu for some s P S.

(b) For I Ď S a spherical Γ-orbit, define

ÿ
αI :“ αs P V Γ :“ tv P V | γ.v “ v for all γ P Γu.
sPI

34
(c) Define S Γ Ď W Γ to be the set

S Γ “ tw0 pIq | I Ď S is a spherical Γ-orbit.u.

Observe that the αI for spherical Γ-orbits I form a basis of V Γ .

Lemma 96. For v P V Γ and I Ď S a spherical Γ-orbit, we have

w0 pIqv “ v ´ αI_ pvqαI P V Γ ,

where αI_ : V Γ Ñ R is a linear map satisfying αI_ pαI q “ 2.

Proof. For v P V Γ , we have v ´ w0 pIqv P VI X V Γ . The latter intersection is just RαI by

choice of I. So we get a linear form αI_ : V Γ Ñ R such that w0 pIqv “ v ´ αI_ pvqαI for
all v P V Γ .
Since αI , w0 pIqαI P V Γ X VI and w0 pIq2 “ 1, we see that w0 pIqαI “ ˘αI . It cannot
be `αI , hence αI_ pαI q “ 2.

Lemma 97. Let I1 , . . . , In be spherical Γ-orbits, w “ w0 pI1 q ¨ ¨ ¨ w0 pIn q and s P S such

that wαs ď 0. Denote by I the Γ-orbit containing s. Then I is spherical and

w “ w0 pI1 q ¨ ¨ ¨ w0 pIi´1 qw0 pIi`1 q ¨ ¨ ¨ w0 pIn qw0 pIq

for some index i P t1, . . . , nu.

Proof. We saw that I is spherical in Lemma 94. We may certainly find an index i such
that

w0 pIi`1 q ¨ ¨ ¨ w0 pIn qαs ě 0, w0 pIi q ¨ ¨ ¨ w0 pIn qαs ď 0.

Since w1 “ w0 pIi`1 q ¨ ¨ ¨ w0 pIn q is in W Γ , we see w1 αs1 P invpw0 pIi qq for all s1 P I.

Conversely, if pw1 q´1 αs1 ď 0 for some s1 P Ii , then pw1 q´1 β ď 0 for all β P invpw0 pIi qq
(using pw1 q´1 P W Γ ). This contradicts w1 αs P invpw0 pIi qq. We see pw1 q´1 αs1 ě 0
for all s1 P Ii . The same argument shows pw1 w0 pIqq´1 αs1 ď 0 for all s1 P Ii . Hence
pw1 q´1 pαs1 q P invpw0 pIqq.
Looking at minimal elements in the respective inversion sets, we see that w1 sends
tαs1 | s1 P Iu bijectively to tαs1 | s1 P Ii u. Hence Ii “ w1 Ipw1 q´1 and the claim follows.

Theorem 98. (a) pW Γ , S Γ q is a Coxeter group.

(b) The action of W Γ on V Γ is contains the geometric representation of pW Γ , S Γ q as

natural subrepresentation.

(c) For spherical Γ-orbits I1 , . . . , In , the product

w0 pI1 q ¨ ¨ ¨ w0 pIn q

is length-additive in W if and only if it is reduced in W Γ .

35
(d) W Γ inherits the Bruhat order, left weak order and right weak order from W .
Proof. For clarity, we denote the length function of W by ℓW : W Ñ Z.
(a) We saw that W Γ is generated by S Γ and by construction, each element in S Γ is an
involution. Denote by ℓW Γ : W Γ Ñ Z the corresponding length function.
It remains to verify the Weak Exchange Condition.
Let w P W Γ be written as w “ w0 pI1 q ¨ ¨ ¨ w0 pIn q for spherical Γ-orbits I1 , . . . , In with
n minimal. Let I be another spherical Γ-orbit such that ℓW Γ pww0 pIqq ď ℓW Γ pwq.
If wαs ď 0 for some s P I, we are done by the previous lemma. Otherwise, we find
ww0 pIqαs ď 0, and then the previous lemma shows

ℓW Γ pww0 pIqq ą ℓW Γ pwq,

contradiction.

(b) Let Ṽ be the vector space with a formal basis βw0 pIq where I runs through the
spherical Γ-orbits. Choose real numbers kw0 pIq,w0 pI 1 q :“ αI_ pαI 1 q. From the fact that
W Γ Ñ GLpV Γ q is a well-defined and faithful representation of W Γ , one deduces that
the numbers kw0 pIq,w0 pI 1 q as defined above satisfy the properties needed to define a
geometric representation of W Γ .
Then the natural map Ṽ ãÑ V Γ , βw0 pIq ÞÑ αI is a W Γ -equivariant injection, by
construction.

(c) If it is not length additive, it cannot be reduced for pW Γ , S Γ q using Lemma 28

for pW, Sq together with the previous Lemma. Conversely, if it is not reduced for
pW Γ , S Γ q, we find some index with w0 pI1 q ¨ ¨ ¨ w0 pIi´1 qαIi ď 0 using Theorem 26 for
pW Γ , S Γ q, and then the product cannot be length additive by Lemma 28.

(d) For v, w P W Γ , we have

ℓW pwvq “ ℓW pwq ` ℓW pvq ðñ ℓW Γ pwvq “ ℓW Γ pwq ` ℓW Γ pvq

by part (c). Hence the statements on the weak orders follow. If v ď w with respect
to the Bruhat order on W Γ , we conclude that the same holds in the Bruhat order
on W using part (c) and the word criterion for the Bruhat order.
For the converse implication, we use induction on w and compare the statements of
Corollary 61 for pW, Sq with pW Γ , S Γ q.
We call pW Γ , S Γ q the folding of pW, Sq under the action of Γ. The Coxeter diagram
of pW Γ , S Γ q can be determined using the following result:
Proposition 99. Let I, J be spherical Γ-orbits and consider the parabolic subgroup
WIYJ .
(a) WIYJ is finite if and only if the order m :“ mpw0 pIq, w0 pJqq of w0 pIqw0 pJq in W Γ
is finite.

36
(b) If m is finite, then
2ℓpw0 pI Y Jqq
m“ .
ℓpw0 pIqq ` ℓpw0 pJqq

(c) If m is finite and odd, then ℓpw0 pIqq “ ℓpw0 pJqq.

Proof. If w P WIYJ is a longest element, then w P W Γ since there can be only one
longest element. Moreover, we have wαI , wαJ ď 0. Thus w P W Γ is a longest element
as well.
If conversely w P pWIYJ qΓ is a longest element, we must have wαs ď 0 for all s P I
and all s P J. Thus w P WIYJ is a longest element as well.
Now pWIYJΓ , tw pIq, w pJquq is a dihedral group. It is finite if and only if the order of
0 0
w0 pIqw0 pJq is finite. In this case, the longest element is given by the alternating product
of m terms
w0 “ w0 pIqw0 pJqw0 pIq ¨ ¨ ¨ .
This expression must be length additive for pW, Sq by the theorem. We conclude
ℓpw0 q “ rm{2sℓpw0 pIqq ` tm{2uℓpw0 pJqq.
We may repeat the argument with I and J interchanged. So if m is odd, we must have
ℓpw0 pIqq “ ℓpw0 pJqq and in any case we obtain
m
ℓpw0 q “ pℓpw0 pIqq ` ℓpw0 pJqq .
2
This finishes the proof.
Exercise 100. Consider our standard example W “ S3 and let Γ “ t1, w0 u, acting on
pW, Sq by conjugation. Determine the Coxeter diagram for pW Γ , S Γ q.

15 List of finite Coxeter groups

It is a fundamental result in the theory of Coxeter groups that we can enumerate all
finite examples.
Definition 101. A Coxeter group pW, Sq is called reducible if there exist non-empty
subsets S1 , S2 Ď S with S “ S1 \ S2 such that s1 s2 “ s2 s1 for all s1 P S1 , s2 P S2 .
Otherwise, it is called irreducible.
If pW, Sq is reducible as above, then naturally W “ WS1 ˆ WS2 and W is finite iff both
WS1 and WS2 are. In this section, we describe all finite and irreducible Coxeter groups.
Example 102. For n ě 1, the Coxeter group associated with the diagram of type
An : ‚ ´ ‚ ´ ¨ ¨ ¨ ´ ‚

is Sn`1 . The length of w0 is given by npn`1q

2 .
For n ě 2, the diagram An has a unique non-trivial automorphism. With Γ generated
by this automorphism, S Γ consists of rn{2s elements. The quotient is of type Brn{2s .

37
Example 103. For n ě 2, consider the Coxeter diagram
4
Bn : ‚ ‚ ´ ‚ ´ ¨ ¨ ¨ ´ ‚.

The Coxeter group of signed permutations was already introduced. The length of the
longest element is n2 .
The diagram has no non-trivial automorphisms unless n “ 2, in which case there is
one non-trivial automorphism (whose folding yields A1 ).
Example 104. For n ě 4, consider the Coxeter diagram
‚

Dn : ‚ ‚ ¨¨¨ ‚ .

‚
If pW, Sq is of type Bn with s0 being the isolated node on the left (i.e. mps0 , sq P t2, 4u
for s P S), then we get a Coxeter group of type Dn by considering the subgroup of W
consisting of those elements w P W where the number of occurrences of s0 in some /
any reduced word of w is even.
The longest element of type Dn has length n2 ´ n.
There is a non-trivial automorphism interchanging the two leftmost nodes in the above
picture. The corresponding folding is of type Bn´1 . If n ě 5, this is the only non-trivial
automorphism.
For n “ 4, the automorphism group of the Coxeter diagram is the symmetric group
S3 , acting freely on the three outer nodes. If 1 ‰ Γ ď S3 fixes one of the outer nodes, the
folding is B3 . Otherwise, there are only two Γ-orbits and the resulting Coxeter group
has type G2 “ I2 p6q.
Example 105. For n “ 6, 7, 8, consider the Coxeter diagram
‚
En : .
‚ ‚ ‚ ‚ ¨¨¨ ‚
The corresponding Coxeter groups does not have any easily accessible description, but
it is finite with longest element of length 36, 63, 120 respectively.
The automorphism group is trivial unless n “ 6, in which case there is precisely
one non-trivial automorphism (in the above picture, its the left-right symmetry). The
corresponding quotient has type F4 .
Example 106. Consider the Coxeter diagram
4
F4 : ‚ ´ ‚ ‚ ´ ‚ .

The corresponding Coxeter group is finite with longest element having length 24. There
is a unique automorphism, and the resulting folding is of type I2 p8q.

38
Example 107. The Coxeter diagrams
5 5
H3 : ‚ ‚ ´‚, H4 : ‚ ‚ ´ ‚ ´‚

yield finite Coxeter groups with longest elements of length 15 resp. 60. There are no
non-trivial diagram automorphisms.
Example 108. Let m ě 3 and consider the Coxeter diagram
m
I2 pmq : ‚ ‚.

The corresponding Coxeter group is the dihedral group of order 2m, with longest element
having length m. There is a unique non-trivial diagram automorphism, whose folding
has type A1 .
Remark 109. Call a Coxeter group crystallographic 2 if it is possible to choose the con-
stants ks,s1 defining the geometric representation to be integers. This is equivalent to
mps, s1 q P t1, 2, 3, 4, 6, `8u for all s, s1 P S.
The finite crystallographic Coxeter groups are precisely the Weyl groups known from
Lie theory. The crystallographic dihedral groups are I2 p3q “ A2 , I2 p4q “ B2 and I2 p6q
which is called G2 for this reason.
If pW, Sq is such a Weyl group, there is typically only one way to choose the constants
ks,s1 P Z up to a diagram automorphism. The only exception is the case Bn , in which
case there are essentially two ways. These choices yield root systems called Bn and Cn ,
whose Weyl groups are isomorphic as Coxeter groups.
In the exercises of this section, you are allowed to use without proof that the above
list of finite and irreducible Coxeter groups is complete.
Exercise 110. Let pW, Sq be a Coxeter group of type Bn . Using the description in terms
of signed permutations, give a description of the Bruhat order similar to Theorem 76.
Exercise 111. Call a Coxeter group simply laced if mps, s1 q P t1, 2, 3u for all s, s1 P S.

(a) Show that a finite Coxeter group is a Weyl group if and only if it is the folding of a
simply laced finite Coxeter group.

(b) Look up a classification of affine Coxeter groups, e.g. [BB05, Appendix A1] or
wikipedia3 and show that each affine Coxeter group is the folding of an affine and
simply laced Coxeter group.

(c) Show that the folding of an irreducible affine Coxeter group is again affine or finite.

Exercise 112. Suppose that pW, Sq is an irreducible Coxeter group and Γ a group of
automorphisms such that pW Γ , S Γ q is of type H3 , H4 or I2 pmq for m ‰ 3, 4, 6 (i.e. one
of the finite non-Weyl groups). Show that W “ W Γ .
2
There are other definitions of this notion, e.g. [Hum90, Section 6.6].
3
https://en.wikipedia.org/wiki/Affine_root_system.

39
16 Iwahori-Hecke algebra
Remember that for a Lie group G with Iwahori B and Weyl group W , we have
ğ
G“ BwB.
wPW

The Hecke algebra of G is defined as

HpGq “ tf : G Ñ C | f pb1 gb2 q “ f pgq @g P G, b1 , b2 P Bu.

It becomes a C-algebra where we define multiplication via a convolution product

ż
pf1 f2 qpgq “ f1 pgpg 1 q´1 qf2 pg 1 q dg 1 .
G

As vector space over C, it has a basis given by the indicator functions Tw “ 1BwB for
w P W . Assume that all subsets BsB for s P S have the same volume q P C. Then one
computes
#
Tws , ws ą w,
Tw Ts “
qTws ` pq ´ 1qTw , ws ă w.

It is common to change variables T̃w “ q ´ℓpwq{2 Tw , which then satisfy

#
T̃ws , ws ą w,
T̃w T̃s “ 1{2 ´1{2
T̃ws ` pq ´ q qT̃w , ws ă w.

Personally, I like to introduce the shorthand notation Q “ q 1{2 ´ q ´1{2 . In any case,
there are many different but largely equivalent parametrizations of the Hecke algebra
around. Confer to [Bon17] for a most general treatment.
This is a bit of a toy example, really one wants to use p-adic Lie groups and compact
subgroups instead of pG, Bq. Anyway, the goal of this section is to introduce an analogue
construction for arbitrary Coxeter groups.
Definition 113. Let A be an commutative ring and q1 , q2 P A. The Hecke algebra with
equal parameters H “ HpW q associated to pW, A, q1 , q2 q is the A-algebra generated by

Tw , wPW

and relations

Tw Tw1 “Tww1 , if ℓpww1 q “ ℓpwq ` ℓpw1 q,

Ts2 “q1 T1 ` q2 Ts , if s P S.

Lemma 114. (a) If q1 P A is invertible, then each Tw P HpW q is invertible in HpW q.

If w “ s is a simple reflection, its inverse is given by

Ts´1 “ q1´1 pTs ´ q2 q.

40
(b) For any w P W and s P S, we have
# #
Tws , ws ą w Tsw , sw ą w
Tw Ts “ .Ts Tw “
q1 Tws ` q2 Tw , sw ă w. q1 Tsw ` q2 Tw , sw ă w.

Proof. (a) The formula for Ts´1 is immediately verified. In general, Tw is a product of
Tsi following a fixed reduced word of w.

(b) The claims are clear if ws ą w resp. sw ą w. If ws ă w, we get a length additive

product w “ pwsq ¨ s, hence

Tw Ts “ Tws Ts2 “ Tws pq1 T1 ` q2 Ts q “ q1 Tws ` q2 Tw .

The other part is analogous.

The following structural result is crucial for the remainder of the course.

Proposition 115. As module over A, the Hecke algebra HpW q is free over the basis
given by Tw for w P W .

For the proof of the proposition, we define the A-module M to be free over W ,

M :“
à
Avw .
wPW

For each s P S, denote by τs P GLpM q the linear automorphism defined on basis vectors
by
#
vsw , sw ą w
τs pvw q “
q1 vsw ` q2 vw , sw ă w.

Lemma 116. Let s, t P S such that mps, tq ă `8. Let τ, τ 1 be the mps, tq-fold compo-
sitions

τ “ ¨ ¨ ¨ ˝ τs ˝ τt ˝ τs
τ 1 “ ¨ ¨ ¨ ˝ τt ˝ τs ˝ τt .

Then τ “ τ 1 as endomorphisms of M .

Proof. We show τ pvw q “ τ 1 pvw q by induction on ℓpwq. Write w0 ps, tq for the longest
element of WJ where J “ ts, tu.
If sw ą w and tw ą w, we get ℓpw0 ps, tqwq “ ℓpw0 ps, tqq ` ℓpwq. Hence we directly
evaluate τ pvw q “ vw0 ps,tqw “ τ 1 pvw q.
Otherwise, we may and do assume without loss of generality that sw ă w. Let
s , t1 P ts, tu such that the two reduced decompositions of w0 ps, tq have the form
1

s1 t1 s1 ¨ ¨ ¨ sts “ w0 ps, tq “ t1 s1 t1 ¨ ¨ ¨ tst.

41
Then we evaluate
τ pvw q “τs1 ˝ τt1 ˝ ¨ ¨ ¨ ˝ τt ˝ τs pvw q “ τs1 ˝ τt1 ˝ ¨ ¨ ¨ ˝ τt pq1 vsw ` q2 vw q
“q1 τs1 ˝ τt1 ˝ ¨ ¨ ¨ ˝ τt pvsw q ` q2 τs1 ˝ τt1 ˝ ¨ ¨ ¨ ˝ τt ˝ τs pvsw q
“ q1 τs1 ˝ τt1 ˝ ¨ ¨ ¨ ˝ τt pvsw q ` q2 τt1 ˝ τs1 ˝ ¨ ¨ ¨ ˝ τs ˝ τt pvsw q.
ind

It is a simple calculation to verify that τt21 pvz q “ q1 vz ` q2 τt1 pvz q holds for all z P W .
Thus we calculate similarly
τ 1 pvw q “τ 1 τs pvsw q “ τt1 ˝ τs1 ¨ ¨ ¨ ˝ τs ˝ τt ˝ τs pvsw q
“ τt1 ˝ τt1 ˝ τs1 ¨ ¨ ¨ ˝ τs ˝ τt pvsw q
ind.
“q1 τs1 ¨ ¨ ¨ ˝ τs ˝ τt pvsw q ` q2 τt1 ˝ τs1 ¨ ¨ ¨ ˝ τs ˝ τt pvsw q.
This finishes the induction and the proof.
By Matsumoto’s theorem, we get that whenever w “ s1 ¨ ¨ ¨ sn is a reduced word, the
map
τw “ τs1 ˝ ¨ ¨ ¨ ˝ τsn
depends only on w and not the particular chosen word.
Proof of Proposition 115. We get a map from HpW q to the endomorphism algebra EndA pM q
sending Tw to τw . We get a map from EndA pM q to M sending f to f pv1 q. Composing
both maps, we get an A-linear map from HpW q to M sending Tw to vw . Hence the Tw
are linearly independent. They certainly generate HpW q as A-module.
Exercise 117. Compute the product Tw0 ¨ Tw0 for the Hecke algebras of S3 and S4 , with
A being the polynomial ring A “ Zrq1 , q2 s.
Exercise 118. Show that for all w, w1 P W , the value Tw Tw1 ´ Tww1 is divisible by q2 in
HpW q.
Exercise 119. Let w1 , w2 P W and express the product Tw1 Tw2 using the usual basis as
ÿ
Tw1 Tw2 “ fw3 Tw3 , f‚ P A “ Zrq1 , q2 s.
w3 PW

Show that if ℓpw1 q ` ℓpw2 q ` ℓpw3 q is even (resp. odd), then only even (resp. odd) powers
of q2 occur in the polynomial fw3 P ZrQs.

17 Braid group
Definition 120. Associated to a Coxeter system pS, mq, we associate the Braid group
B presented by the generators S and relations
1 1 1 1
ss ss ¨ ¨ ¨ “ slooomooon
looomooon ss s ¨ ¨ ¨ .
mps,s1 q terms mps,s1 q terms

If pS, mq is of type An´1 , we denote B by Bn .

42
 Certainly the Coxeter group W is a quotient of B, so Sn is a quotient of Bn . The
braid group Bn is infinite in general. It has the following geometric interpretation:
Consider a sequence of smooth paths in r0, ns ˆ r0, 1s

p0 :p1, 0q Ñ p0, 0q,

p1 :p1, 1q Ñ p0, 1q,
..
.
pn :p1, nq Ñ p0, nq

such that any two paths pi , pj intersect only at a finite number of points. To each
such crossing, we moreover record information on which path lies above the other one.
(picture)
Such a diagram, up to smooth stretching and moving of paths above and below each
other, can be called a braid. These form a group by composition (picture). We can
associate si P S to the braid consisting of straight paths pj : p1, jq Ñ p0, jq for j ‰ i, i`1,
pi : p1, iq Ñ p0, i ` 1q and pi`1 : p1, i ` 1q Ñ p0, iq such that pi lies above pi`1 (picture).
The resulting group of braids is isomorphic to Bn .
Each braid can be associated to a union of knots, or links, in R3 by connecting p0, iq
to p1, iq outside of the box r0, ns ˆ r0, 1s (picture). The same link may have different
braid presentations, but there is some sort of equivalence relations between braids.
So the knot theorists are interested in representations of the group Bn , say over the
complex so Bn Ñ GLn pCq. Any two generators of Bn are conjugate, so representations
with only one eigenvalue over C (e.g. one-dimensional representations) must be of the
form si ÞÑ diagpc, . . . , cq for some constant c P Cˆ .
Let us consider a representation ρ : Bn Ñ GLn pCq such that each ρpsi q is diagonaliz-
able with at most two eigenvalues, e.g. for n “ 2. By conjugation, they have the same
eigenvalues, say λ1 , λ2 . We get

pρpsi q ´ λ1 qpρpsi q ´ λ2 q “ 0,
ðñ ρpsi q2 “ pλ1 ` λ2 qρpsi q ´ λ1 λ2 .

The fact that ρ is a representation is equivalent to

ρpsi qρpsj q “ρpsj qρpsi q if |i ´ j| ě 2

ρpsi qρpsi`1 qρpsi q “ρpsi`1 qρpsi qρpsi`1 q.

If we put q1 “ ´λ1 λ2 and q2 “ λ1 ` λ2 , these representations are the same as repre-

sentations of the Hecke algebra HpSn q defined over C. We refer to the very remarkable
article [Jon87] using this correspondence to define an important invariant of links.

18 Kazhdan-Lusztig polynomials
We specialize to the Hecke algebra defined over A “ Zrq ˘1{2 s with parameters q1 “ q
and q2 “ q ´ 1. In other words, pTs ` 1qpTs ´ qq “ 0 for all s P S.

43
 Kazhdan-Lusztig construct a large class of irreducible representations of HpW q, which
we review in this section. In particular, specialized to W “ Sn and q “ 1, all irreducible
representations of the group algebra CrSn s arise from their construction.
The ring A has an automorphism induced from q 1{2 ÞÑ q ´1{2 which we denote by
a ÞÑ a.
Lemma 121. There is an automorphism of rings HpW q Ñ HpW q sending

aTw ÞÑ aTw :“ aTw´1

´1 .

Proof. It suffices to see that the Tw satisfy the defining relations of the Iwahori-Hecke
algebra. If ℓpww1 q “ ℓpwq ` ℓpw1 q, we get

Tw ¨ Tw1 “ Tw´1 ´1
´1 Tpw 1 q´1 “ pTpw 1 q´1 Tw ´1 q
´1 ´1
“ Tpww 1 q´1 “ Tww 1 .

Let now s P S. We compute

Ts ¨ Ts “ Ts´1 Ts´1 .

Observe Ts´1 “ q ´1 pTs ` 1 ´ qq so that

pTs´1 q2 “q ´2 pTs2 ` 2p1 ´ qqTs ` p1 ´ qq2 q “ q ´2 ppq ´ 1qTs ` q ` 2p1 ´ qqTs ` p1 ´ qq2 q
“q ´2 pp1 ´ qqpTs ` 1 ´ qq ` qq “ q ´1 p1 ´ qqTs´1 ` q ´1 “ q ´ 1 ¨ Ts ` q.

So the relations are satisfied for Ts ¨ Ts . This finishes the proof.

Definition 122. For w P W , we write

ÿ
Tw “ Rv,w q ´ℓpvq Tv P HpW q
vPV

for polynomials Rv,w P A, called the R-polynomial associated with v, w P W .

They satisfy the following properties.
Lemma 123. Let v, w P W and s P S.
(a) If ws ă w then
#
Rvs,ws , if vs ă v,
Rv,w “
qRvs,ws ` pq ´ 1qRv,ws , if vs ą v.

(b) If w “ 1, then
#
1, if v “ 1,
Rv,1 “
0, if v ‰ 1.

(c) If v ď w in the Bruhat order, then Rv,w is a monic polynomial of degree ℓpwq ´ ℓpvq
in Zrqs. Otherwise, Rv,w “ 0.

44
Proof. (a) We compute
ÿ ÿ
Ru,w q ´ℓpuq Tu “Tw “ Tws ¨ Ts “ Ru,ws q ´ℓpuq Tu Ts´1
uPW uPW
ÿ ÿ
“ Ru,ws q ´ℓpuq
Tus ` Ru,ws q ´ℓpuq pq ´ 1Tu ` qTus q.
uPW uPW
usău usąu

Let us compare the A-coefficients for Tv in the above expression. If vs ă v, then

the left sum does not contribute to Tv and the right sum contributes the value
Rvs,ws q ´ℓpvq Tv coming from u “ vs.
If vs ą v, then the left sum contributes qRvs,ws q ´ℓpvq Tv coming from u “ vs and the
right sum contributes pq ´ 1qRv,ws q ´ℓpvq Tv coming from u “ v.

(b) Indeed we have T1 “ T1 .

(c) Induction on ℓpwq, the inductive start for w “ 1 being clear. In the inductive step,
pick s P S such that ws ă w and let v 1 “ minpv, vsq. By the lifting property of
Bruhat order, v ď w if and only if v 1 ď ws.
So in case v ę w, we get v 1 ę ws and also v ę ws, showing that Rv,w , being a linear
combination of Rv1 ,ws and Rv,ws , is zero.
If v ď w, then by induction we see that Rv1 ,ws is a monic polynomial of degree
ℓpwsq ´ ℓpv 1 q. If vs ă v, we get Rv,w “ Rv1 ,ws and ℓpswq ´ ℓpv 1 q “ ℓpwq ´ ℓpvq,
proving the claim. Otherwise, we get

Rv,w “ qRvs,ws ` pq ´ 1qRv1 ,ws

with Rvs,ws having degree ď ℓpwsq ´ ℓpvsq “ ℓpwq ´ ℓpvq ´ 2 and Rv1 ,ws being monic
of degree ℓpwsq ´ ℓpv 1 q “ ℓpwq ´ ℓpvq ´ 1. The induction is complete.

Theorem 124. There is a unique basis tCw uwPW of the A-module HpW q satisfying the
following properties: Cw “ Cw for all w P W and
ÿ
Cw “ p´1qℓpvq`ℓpwq q ℓpwq{2´ℓpvq Pv,w Tv P HpW q
vďw

for polynomials Pv,w P A of q-degree ď 1{2pℓpwq ´ ℓpvq ´ 1q such that Pw,w “ 1.

Proof sketch. Uniqueness: Using the definition of R-polynomials, the condition Cw “ Cw
expanded as required implies for u ă w that
ÿ
q pℓpwq´ℓpuqq{2 Pu,w ´ q pℓpuq´ℓpwqq{2 Pu,w “ p´1qℓpuq`ℓpvq q ℓpvq´ℓpuq{2´ℓpwq{2 Ru,v Pv,w .
uăvďw

In an inductive sense, we may assume that the Pv,w are uniquely determined for u ă
v ď w. Now

q pℓpuq´ℓpwqq{2 Pu,w

45
is a polynomial in q ´1{2 without constant term. Applying the involution ¨ to it yields
a polynomial in q 1{2 without constant term. There are no cancellations between such
polynomials possible, so Pu,w is uniquely determined by the Pv,w for u ă v ď w.
Existence: Use the above relation to construct suitable elements Cw which are invari-
ant under the involution ¨ and have the required shape. Then the basis property is easily
verified.

Definition 125. The polynomials Pv,w P Zrqs are called Kazhdan-Lusztig polynomials.
Exercise 126. Calculate the R-polynomials for W “ S3 . With the given relation in the
above proof, calculate the Kazhdan-Lusztig polynomials.
Exercise 127. Verify the full proof of existence and uniqueness of Kazhdan-Lusztig poly-
nomials [KL79, Section 2.2].
Exercise 128. Show that Pv,w ‰ 0 whenever v ď w.

19 Kazhdan-Lusztig representations
We continue with the notation from the previous section.
Definition 129. For v ď w in W , let µpv, wq be the coefficient of q pℓpwq´ℓpvq´1q{2 in
Pv,w . If v ą w, we define µpv, wq :“ µpw, vq.
We write v ă w if v ă w, ℓpwq ´ ℓpvq is odd and µpv, wq ‰ 0.
Lemma 130. Let s P S and w P W .
(a) If sw ă w, then Ts Cw “ ´Cw .
(b) If sw ą w, then
ÿ
q ´1{2 Ts Cw “ Csw ` q 1{2 Cw ` µpz, wqCz ,
with the sum taken over all z ă w with sz ă z.
Proof reference. Cf. [KL79, Sections 2.2, 2.3].

We see that the basis elements Cw P HpW q are not invertible, unlike their Tw coun-
terparts. This is the exciting aspect from a representation theoretic point of view!
Definition 131. We define the left cell order ďL on W to be the partial order generated
by the relations v ăL w whenever the following two conditions are satisfied:
(a) v ă w or w ă v and
(b) there exists s P S with sv ă v and sw ą w.
The right cell order is defined by v ďR w if v ´1 ďL w. It has an analogue description if
we replace (b) by vs ă v and ws ą w.
The two-sided cell order is the partial order generated by all pairs v ďLR w such that
v ďL w or v ďR w.
For ‚ P tL, R, LRu, we write v „‚ w if v ď‚ w ď‚ v.

46
 It is rather straightforward from Lemma 130 to see the following:
Proposition 132. Let w P W .
(a) The left ideal HpW qCw Ď HpW q is a free A-module with basis given by all Cv for
v ďL w.
(b) The right ideal Cw HpW q Ď HpW q is a free A-module with basis given by all Cv for
v ďR w.
(c) The two-sided ideal HpW qCw HpW q Ď HpW q is a free A-module with basis given by
all Cv for v ďLR w.
Definition 133. The equivalence classes of „L , „R , „LR are called left/right/two sided
cells in W . Given such a left cell C, the left cell module for C is given by

pHpW qwq{I

for some w P W where I is the left ideal in HpW q generated by all Cv for v ăL w.
Similarly, right cells induce right cell modules and two-sided cells induce cell bimodules.
We may consider the case of a finite Coxeter group and specialize A Ñ C, q ÞÑ 1.
For the W “ Sn being the symmetric group, it follows that the left cell representa-
tions are irreducible and cover all irreducible complex representations. For general finite
Coxeter groups, these complex representations may be reducible, but contain every ir-
reducible representation as a subrepresentation.
The theory of Kazhdan-Lusztig polynomials, cells and representations is subject to
active research and many conjectures.

20 Reflection orders
The concept of reflection order appears in different places of the theory of Coxeter groups.
These orders can be used e.g. to give a convenient description of R-polynomials.
Definition 134. A reflection order ă on Φ` is a total order such that for all α, β P Φ`
with α ă β and scalars λ, µ P Rą0 satisfying λα ` µβ P Φ` , we have

α ă λα ` µβ ă β.

Example 135. For W “ Sn , the positive roots can be identified as

Φ` “ tei ´ ej | i ă ju Ď Rn .

Then the lexicographic order is a reflection order on Φ` , i.e.

pei ´ ej q ă pei1 ´ ej 1 q ðñ pi ă i1 q or pi “ i1 and j ă j 1 q.

Indeed, if α ă β are such that α ` β P Φ` , we must have α “ ei ´ ej , β “ ej ´ ek for

indices i ă j ă k. Now α ` β “ ei ´ ek sits between α and β as claimed. There are
many more reflection orders on Φ` .

47
 The first non-trivial observation of the theory of reflection orders is the following:

Proposition 136. There exists a reflection order ă for every Coxeter group pW, Sq.

ř Choose a well-ordering
Proof. ă on S and for each s P S a positive scalar λs ą 0. For
α “ sPS cs αs P Φ we define
`

ÿ 1
vpαq “ pλs cs qsPS P RSě0 , σpαq “ pvpαqq P Rą0 and ṽpαq “ vpαq P RSě0 .
σpαq

We define the order ă on Φ` by comparing the vectors ṽpαq lexicographically. Explicitly,

α ă β if and only if there exists s P S such that ṽpαqs ă ṽpβqs and

@s1 ă s : ṽpαqs1 “ ṽpβqs1 .

Observe that the value of s is uniquely determined and this defines a well-defined and
total order on Φ` . If now α ă β with s P S as above and γ “ λα ` µβ P Φ` , we get
σpγq “ λσpαq ` µσpβq. In particular

λσpαqṽpαqs ` µσpβqṽpβqs
ṽpαqs ă ṽpγqs “ ă ṽpβqs
λσpαq ` µσpβq

and for each s1 ă s, we get

λσpαqṽpαqs1 ` µσpβqṽpβqs1
ṽpαqs1 “ ṽpγqs1 “ “ ṽpβqs1 .
λσpαq ` µσpβq
This verifies the reflection order property.

For finite Coxeter groups, we have the following characterization due to Dyer [Dye93].

Lemma 137. Let W be finite with longest element w0 and let ă be a total order on Φ` .
Then the following are equivalent.

(a) The order ă defines a reflection order on Φ` .

(b) There exists a (unique) reduced word w0 “ sα1 ¨ ¨ ¨ sαn such that β1 ă ¨ ¨ ¨ ă βn , where

βi “ sαn ¨ ¨ ¨ sαi`1 pαi q.

Proof. (a) ùñ (b). Enumerate the positive roots as Φ` “ tβ1 ă ¨ ¨ ¨ ă βn u and define
for i “ 1, . . . , n the root

αi “ sβn ¨ ¨ ¨ sβi`1 pβi q P Φ.

We claim that all these roots are simple (in particular positive), via induction on n ´ i.
If i P t1, . . . , nu and the claim has been proved for all i1 ą i, note that

sβn ¨ ¨ ¨ sβi`1 “ sαi`1 ¨ ¨ ¨ sαn .

48
Moreover, we have 1 ă sβn ă ¨ ¨ ¨ ă sβn ¨ ¨ ¨ sβi`1 such that the right one of the above
words is reduced and

invpsαi`1 ¨ ¨ ¨ sαn q “ tβi`1 , . . . , βn u.

In particular, αi “ sαi`1 ¨ ¨ ¨ sαn pβi q P Φ` . If αi was not simple, we could write αi “

λ1 γ1 ` λ2 γ2 for positive scalars λ1 , λ2 ą 0 and distinct positive roots γ1 , γ2 P Φ` (e.g.
by choosing a simple root which gets sent to a negative root under sαi ). Put γ̃j “
sαn ¨ ¨ ¨ sαi`1 γj , j “ 1, 2, so that βi “ λ1 γ̃1 ` λ2 γ̃2 .
If γ̃j is positive, then the condition γj “ sαi`1 ¨ ¨ ¨ sαn γ̃j P Φ` implies γ̃j ĺ βi . Similarly,
if γ̃j P Φ´ then ´γ̃j ą βi .
In case both γ̃1 and γ̃2 are positive, they must both be ĺ βi and then the fact βi “
λ1 γ̃1 ` λ2 γ̃2 contradicts (a).
If, say, γ̃1 is positive and γ̃2 is negative, we get
1 λ2
´γ̃2 ą βi ľ γ̃1 , γ̃1 “ βi ` p´γ̃2 q,
λ1 λ1
contradiction again.
Finally, if both γ̃1 and γ̃2 are negative, we get βi P Φ´ which is absurd. This finishes
the induction.
The inductive proof shows moreover that

s α1 ¨ ¨ ¨ s αn

is a reduced word of an element sending every root in Φ` to a negative root. This is the
desired reduced word for (b).
(b) ùñ (a). If i ă j are indices such that β :“ λβi ` µβj P Φ` , note that

invpsαi ¨ ¨ ¨ sαn q “tβi , . . . , βn u

invpsαj`1 ¨ ¨ ¨ sαn q “tβj`1 , . . . , βn u.

Since both βi and βj are in invpsαi ¨ ¨ ¨ sαn q, so must be β. Similarly, β cannot be in

invpsαj`1 ¨ ¨ ¨ sαn q. Hence β P tβj`1 , . . . , βi´1 u.

Using Matsumoto’s theorem to pass between different reduced words for w0 , one can
pass from one reflection order to another using a sequence of well-defined operations, cf.
[BFP98]. A typical application of reflection orders is the following result, whose proof
we will not state.

Theorem 138 ([BB05, Theorem 5.3.4]). Let ă be a reflection order and v, w P W . Let
P be the set of all sequences pα1 , . . . , αn q of positive roots such that

• α1 ă ¨ ¨ ¨ ă αn and

• v ă vsα1 ă ¨ ¨ ¨ ă vsα1 ¨ ¨ ¨ sαn “ w.

49
Then the R-polynomial of v, w is given by
ÿ
Rv,w “ q pℓpuq´ℓpvqq{2 pq 1{2 ´ q ´1{2 qn P Zrqs.
pα1 ,...,αn qPP

Exercise 139. List the reflection orders for S3 , S4 .

Exercise 140. Assume that W is a finite group with reflection order ă and that

tα1 ă ¨ ¨ ¨ ă αn u Ă Φ`

are some positive roots such that a linear combination

λ1 α1 ` ¨ ¨ ¨ ` λn αn “ α P Φ`

is a positive root as well, where λ1 , . . . , λn P Rě0 . Prove α1 ĺ α ĺ αn .

21 Reflection subgroups
Proposition 141. Let W 1 Ď W be a reflection subgroup, i.e. such that T 1 :“ W 1 X T
generates W 1 . Define

S 1 :“ tt P T 1 | @t ‰ t1 P T 1 : ℓptt1 q ą ℓptqu.

Define Φ1 Ď Φ` to be the set of all positive roots α P Φ` such that sα P W 1 (so sα P T 1 ).

Call a root α P Φ1 simple for W 1 if for any linear combination
ÿ
α“ λn β P V, λβ P Rě0 almost all zero,
βPΦ1

we must have
#
1, α “ β,
λβ “
0, α ‰ β.

(a) pW 1 , S 1 q is a Coxeter group.

(b) A root α P Φ1 is simple for W 1 if and only if sα P S 1 . Under that identification, the
geometric representation of W 1 is isomorphic to the action of W 1 on the subspace of
V generated by the simple roots for W 1 .

Proof. W 1 generated by S 1 : For this, it suffices to see that every element t P T 1 is

a product of elements in S 1 . We show this via induction on ℓW ptq. If t P S 1 (e.g. t is
simple), there is not much to show.
Otherwise, we find t1 P T 1 with ℓW pt1 tq ă ℓW ptq. Pick a reduced word

t “ sα1 ¨ ¨ ¨ sαℓ

50
such that

invptq “ tαℓ , sαℓ pαℓ´1 q, . . . , sαℓ ¨ ¨ ¨ sα2 pα1 qu.

Writing t “ sα , we see that α must appear in this list. If it appears in position i, then
t “ sα1 ¨ ¨ ¨ sαi ¨ ¨ ¨ sα1 , which means (by reducedness) i ě pℓ ´ 1q{2. Considering t “ t´1 ,
a similar argument shows i ď pℓ ´ 1q{2 such that i “ pℓ ´ 1q{2 and t “ sα1 ¨ ¨ ¨ sαi ¨ ¨ ¨ sα1
is reduced.
We may and do assume that αj “ αℓ`1´j for all j P t1, . . . , ℓu. Write now t1 “ sβ . Up
to replacing t1 by tt1 t, we may assume that

β “ sα1 ¨ ¨ ¨ sαj´1 pαj q

for some index j ď i “ pℓ ´ 1q{2. If j “ i, we would get t “ t1 , contradiction. Hence

j ă i and

ℓW pt1 q “ℓW psα1 ¨ ¨ ¨ sαj ¨ ¨ ¨ sα1 q ď 2j ´ 1 ă ℓ,

ℓW pt1 tt1 q “ℓW psα1 ¨ ¨ ¨ sx αℓ`1´j ¨ ¨ ¨ sαℓ q ď ℓ ´ 2.
αj ¨ ¨ ¨ sαpℓ´1q{2 ¨ ¨ ¨ s{

By induction, both t1 and t1 tt1 lie in the subgroup of W 1 generated by S 1 . Hence so does
t, finishing the proof of W 1 being generated by S 1 .
Characterization of S 1 . For α P Φ1 , the condition sα P S 1 is equivalent to saying
that for all α ‰ β P Φ1 we have sα pβq P Φ` (or equivalently in Φ1 ).
Suppose first that sα P S 1 . We have to show that α is simple for W 1 . So consider any
linear combination
ÿ
α“ λβ β
βPΦ1

as above. Apply sα to write

ÿ ÿ
´α “ λβ sα pβq “ ´λα α ` λβ sα pβq.
βPΦ1 α‰βPΦ1

So pλα ´ 1qα is a Rě0 -linear combination of positive roots, showing λα ě 1. Now

ÿ
λβ β “ p1 ´ λα qα ď 0,
α‰βPΦ1

so that λβ “ 0 for all α ‰ β P Φ1 . The claim follows.

Suppose now conversely that α is simple for W 1 . If we had sα pβq P Φ´ for some
α ‰ β P Φ1 , we got

α “ β _ pαqβ ´ sα pβq

as a linear combination of two roots in Φ1 , both not equal to α. This contradict the
condition of α being simple for W 1 .

51
 Weak Exchange Condition. Let w “ t1 ¨ ¨ ¨ tn P W 1 be reduced for pW 1 , S 1 q and
α P Φ1 with ℓW 1 pwsα q ď ℓW 1 pwq.
If wα ď 0, we find an index i with

ti ¨ ¨ ¨ tn α ď 0, ti`1 ¨ ¨ ¨ tn α ě 0.

So β “ ti`1 ¨ ¨ ¨ tn α P Φ1 satisfies ti β ď 0. Since ti P S 1 , the above argument shows ti “ sβ

and we get the Weak Exchange Condition.
If wα ě 0, the same argument proves ℓW 1 pwsα q ą ℓW 1 pwq.
Geometric Representation. Define coroots α_ : V Ñ R for all α P Φ` such that
sα pvq “ v ´ α_ pvqα. This yields a faithful representation of W 1 on the subspace of V
spanned by Φ1 . We can verify the defining properties of the geometric representation by
noticing that each root in Φ1 is a Rě0 -linear combination of the simple roots for W 1 .

Definition 142. A dihedral reflection subgroup of W is a subgroup W 1 generated by

two reflections.

Lemma 143. Let ă be a total order on T . Then the following are equivalent:

(a) We get a reflection order on Φ` defined by

α ă β : ðñ sα ă sβ

(b) For every dihedral reflection subgroup W 1 Ď W with canonical generators S 1 as in

Proposition 141, we may write S 1 “ tt, t1 u such that
1 1
t ă t1 tt1 ă tt t ă ¨ ¨ ¨ ă t t t1 ă tt1 t ă t1

Proof. We may certainly write S 1 “ tsα , sβ u for some positive roots α, β which are simple
for W 1 . The reflections listed in (b) come from the roots

α, sβ pαqsα sβ pαq, . . . , sβ sα pβq, sα pβq, β.

Carefully analysing the geometric representation on dihedral groups, we see that ă

defines a reflection order on Φ1 if and only if (b) is satisfied. Of course, if (a) is satisfied,
then ă is a reflection order on Φ1 .
Assume now that (b) is satisfied. Let α, β P Φ` and λ, µ P Rą0 such that λα`µβ P Φ` .
Let W 1 be the group generated by sα and sβ . Then ă being a reflection order on Φ1
implies that α ă λα ` µβ ă β or vice versa. Hence (a) follows.

Exercise 144. Prove the claimed characterization of reflection orders on dihedral groups.

52
22 Demazure products
In this section, we specialize the Hecke algebra further to the case q “ 0. Up to slightly
optimizing on signs, we get the presentation

Tw Tw1 “Tww1 if ℓpww1 q “ ℓpwq ` ℓpw1 q,

Ts2 “Ts if s P S.

Proposition 145. Let w1 , w2 P W be two elements. Then Tw1 Tw2 has the form Tz for
some z P W . Each of the following three sets has a unique maximum with respect to
Bruhat order, which is moreover equal to z.

tw11 w2 | w11 ď w1 and ℓpw11 w2 q “ ℓpw11 q ` ℓpw2 qu,

tw1 w21 | w21 ď w2 and ℓpw1 w21 q “ ℓpw1 q ` ℓpw21 qu,
tw11 w21 | w11 ď w1 and w21 ď w2 u.

Proof. Pick a reduced word w2 “ s1 ¨ ¨ ¨ sn . Iterating the above relations n times, we see

Tw1 Tw2 “ Tw1 si1 ¨¨¨sik

for some indices 1 ď i1 ă ¨ ¨ ¨ ă ik ď n. In particular, the element z as required exists, is

uniquely determined and an element of the second and third set. A completely analogous
argument shows that z also lies in the first set.
It remains to show that every element in the third set is ď z in the Bruhat order.
We do this via induction on ℓpw2 q. If w2 “ 1, we get z “ w1 and the claim is clear.
Otherwise, pick a simple reflection s with w2 s ă w2 , and let Tw1 Tw2 s “ Tz 1 . If w11 ď w1
and w21 ď w2 , we get minpw21 , w21 sq ď w2 s and hence

w11 minpw21 , w21 sq ď z 1 ď z.

So focus on the case w21 s ă w21 . Then w11 w21 s ď z. Since zs ą z, we get w11 w21 ď z as
well. This finishes the induction and the proof.

Definition 146. For w1 , w2 P W , the element z in the above proposition is called

Demazure product of w1 , w2 and denoted w1 ˚ w2 . Note that pW, ˚q is a monoid but not
a group.
Corollary 147. In the general setting of a Hecke algebra (before the specializations),
consider the product
ÿ
Tw1 Tw2 “ fw3 Tw3 .
w3 PW

Then fw3 “ 0 unless w3 ď w1 ˚ w2 . We have

ℓpw1 q`ℓpw2 q´ℓpw1 ˚w2 q
fw1 ˚w2 ” q2 pmod q1 q.

If w3 ‰ w1 ˚ w2 , then fw3 ” 0 pmod q1 q.

53
Proof. Since the specializations q2 Ñ 1, q1 Ñ 0 yield the above 0-Hecke algebra, the
congruence statements are easily verified. Use that by the definition of the Hecke algebra,
fw3 is always a sum of monomials q1n1 q2n2 such that ℓpw1 q ` ℓpw2 q ´ ℓpw3 q “ 2n1 ` n2 .
From the definition of the Hecke algebra, it is clear that any w3 with fw3 ‰ 0 must be
of the form w3 “ w1 w21 with w21 ď w2 .

We get similar statements when expanding Tw1 Cw2 using the basis C‚ . One can show
that the Kazhdan-Lusztig polynomials Pv,w simplify to 1 under the specialization q Ñ 0.
Exercise 148. Show that if W is finite, then w0 ˚ w “ w ˚ w0 “ w0 for all w P W .
Exercise 149. Show that pw1 ˚ w2 q´1 “ w2´1 ˚ w1´1 .

23 Finiteness results
For this section, assume that S is a finite set.
Let us write
ÿ
Tw1 Tw2 “ fw3 Tw3 , fw3 P Zrq1 , q2 s
w3 PW

for a general Hecke algebra. It was an important conjecture of Lusztig, which has been
proved now, that the q2 -degree of fw3 is always bounded by a constant depending only
on pW, Sq. In this section, we demonstrate the baby version claiming that
suptℓpw1 q ` ℓpw2 q ´ ℓpw1 ˚ w2 q | w1 , w2 P W u ă `8.
Definition 150. For w P W , we define the cone type of w to be
Cpwq “ tw1 P W | ℓpww1 q “ ℓpwq ` ℓpw1 qu Ď W.
For individual w P W , the cone type Cpwq Ď W may typically be an infinite set. It
interesting to observe that many distinct group elements may have the same cone type.
We even have the following striking result.
Theorem 151. There are only finitely many different cone types for W , i.e. the set
tCpwq | w P W u Ď 2W
is finite.
We obtain the following consequences for Demazure products.
Corollary 152. For w1 , w2 P W , the value of
w1´1 pw1 ˚ w2 qw2´1 P W

depends only on the pair pCpw1 q, Cpw2´1 qq P 2W ˆ 2W . Moreover,

0 ď ℓpw1 q ` ℓpw2 q ´ ℓpw1 ˚ w2 q ď N
for some constant N P Z depending only on pW, Sq.

54
Proof. If Cpw1 q “ Cpw11 q and Cpw2´1 q “ Cppw21 q´1 q, we get

w1´1 pw1 ˚ w2 q “ maxpw̃2 ď w2 | w̃2 P Cpw1 qq “ pw11 q´1 pw11 ˚ w2 q.

A similar argument proves pw11 ˚w2 qw2´1 “ pw11 ˚w21 qpw21 q´1 . This shows the independence
claim.
We calculate
´ ¯ ´ ¯
ℓpw1 q ` ℓpw2 q ´ ℓpw1 ˚ w2 q “ ℓpw1 q ´ ℓ pw1 ˚ w2 qw2´1 ď ℓ w1´1 pw1 ˚ w2 qw2´1 .

There are only finitely many possibilities for the right-hand side.

The proof of Theorem 151 is rather involved, cf. [BB05, Chapter 4]. We can only
outline the major steps.
Definition 153. Let α, β P Φ` . We say that α dominates β if any w P W such that
wα P Φ´ satisfies wβ P Φ´ .
We say that α is humble if it only dominates itself and not other root.
Observe that simple roots are humble.
Lemma 154. Let α P Φ` be humble and s P S such that α ‰ αs . Then sα is not humble
if and only if the following conditions are both satisfied.
• αs_ pαq ă 0, or equivalently sα P α ` Rą0 αs and

• The group generated by s, sα is infinite, or equivalently

αs_ pαqα_ pαs q ě 4.

Proof. First assume that both stated conditions are satisfied. We claim that sα dom-
inates αs . Indeed, if this was not the case, we could find some w P W with wsα P
Φ´ and wαs P Φ` . Then α, αs P invpwsq, so the infinitely many roots of the form
λα ` µαs , λ, µ ą 0 are also in invpwsq. This contradicts ws having finite length.
Now assume conversely that sα is not humble. Then it dominates a root β. If β ‰ αs ,
then sβ P Φ` and α dominates sβ, contradiction. Hence sα dominates αs and no other
root. Certainly, sα would dominate every positive linear combination of αs and sα, so
αs must be the only such linear combination. In particular, α R Rą0 sα`Rą0 αs , showing
the first condition. The second one follows from analysing the dihedral group generated
by s and sα . If it is finite, we would find an element w in that group with wαs P Φ`
and wsα P Φ´ .

Corollary 155. For w P W , denote the set of humble inversions by

huminvpwq “ tβ P invpwq | β is humbleu.

If s P S satisfies ws ą w, then

huminvpwsq “ tαs u Y tsβ | β P huminvpwq and sβ is humble.u.

55
Proof. Since invpwsq “ tαs uYs invpwq, the inclusion Ď is clear. For the reverse inclusion,
pick αs ‰ β P huminvpwsq. Then sβ P invpwq and β is humble. We have to show that
sβ is humble as well. If it was not humble, we would get infinitely many roots of the
form λβ ` µαs P invpwsq for λ, µ ą 0, which is impossible.

If w, w1 P W and w1 “ s1 ¨ ¨ ¨ sn is a reduced word, we have

w1 P Cpwq ðñ @i : ℓpws1 ¨ ¨ ¨ si q ą ℓpws1 ¨ ¨ ¨ si´1 q ðñ @i : αsi R huminvpws1 ¨ ¨ ¨ si´1 q.

Using the above corollary, the last condition only depends on the set huminvpwq Ď Φ` .
So Theorem 151 follows from the following result.

Proposition 156. There are only finitely many humble roots.

For the proof of the proposition, observe that if α is a non-simple humble root, we
can find a simple root αs with sα s ă sα , or equivalently sα P α ` Ră0 αs . Then sα is
humble as well and we get a sequence
1s 2 s sn´1
α “ α1 ÝÑ α2 ÝÑ ¨ ¨ ¨ ÝÝÝÑ αn

of humble roots ending in a simple root αn . If there are infinitely many humble roots,
there exist such sequences of arbitrary length.
One has to develop the theory of humble roots a lot further to associate some in-
variants for which there are only finitely many possibilities and which are, in a sense,
monotonic with respect to such sequences. If such sequences are longer than the number
s
of invariants, we would get adjacent humble roots α ÑÝ β with identical invariants, and
can use this to derive a contradiction.
Exercise 157. Show that if W is finite, then the number of distinct cone types is equal
to the number of elements in W . Moreover, show that every root in a finite group is
humble.
Exercise 158. Show that if W is infinite, then each cone type Cpwq is an infinite subset
of W .
Exercise 159. Give an example of a Coxeter group pW, Sq and two distinct elements
w1 , w2 P W such that Cpw1 q “ Cpw2 q.

24 Conjugacy classes
The structure of conjugacy classes in Coxeter groups is of central interest.

Definition 160. Let O Ď W be a conjugacy class and w P W .

(a) We write ℓpOq :“ mintℓpwq | w P Ou and Omin for the elements in O of length ℓpOq.

(b) If s P S and ℓpswsq ď ℓpwq, we call sws an elementary cyclic shift of w and write
s
wÑ Ý sws (not to be confused with the similar, but incompatible notation used for

56
 Bruhat order). We call w1 P W a cyclic shift of w if there is a sequence of elementary
cyclic shifts
s s s
1
w “ w1 ÝÑ 2
w2 ÝÑ n
¨ ¨ ¨ ÝÑ wn`1 “ w1 .

We write w Ñ w1 to indicate this relation.

(c) If v P W satisfies ℓpwvq “ ℓpwq`ℓpvq and ℓpv ´1 wvq “ ℓpwq, we call v ´1 wv elementary
strongly conjugate to w. We say that w and w1 P W are strongly conjugate if there
is a sequence

w “ w1 , . . . , w n “ w 1

such that wi and wi`1 are strongly conjugate.

Observe that if w Ñ w1 and ℓpwq “ ℓpw1 q, then also w1 Ñ w and w1 is strongly
conjugate to w.
Theorem 161 ([Mar18]). Let w P W and O Ď W its conjugacy class.
(a) There is an element w1 P Omin such that w Ñ w1 .
(b) If w, w1 P Omin , then they are strongly conjugate.
Marquis’ proof is rather subtle, generalizing previous works of Geck-Pfeiffer for finite
Coxeter groups and He-Nie for affine ones.
Example 162. In the group Sn , the conjugacy class of an element w written in cycle
notation

w “ pi1,1 , . . . , i1,cp1q qpi2,1 , . . . , i2,cp2q q ¨ ¨ ¨ pin,1 , . . . , in,cpnq q

with cp1q ě ¨ ¨ ¨ ě cpnq is determined by the vector pcp1q, . . . , cpnqq, up to adding and
removing trivial cycles. A typical minimal length element is given by

p1, 2, 3qp5, 6qp10, 11, 12, 13q.

The conjugacy classes in S3 are given by

t1u, tp1 2q, p2 3q, p1 3qu, tp1 2 3q, p1 3 2qu.

Note that the second class O has the simple reflections as minimal length elements
Omin “ ts1 , s2 u. Certainly we do not have s1 Ñ s2 . However, the element v “ s2 s1
satisfies s1 v “ w0 , which has length 3, and v ´1 w0 “ s2 .
Definition 163. Let H “ HpW q be the Iwahori-Hecke algebra, defined over some ring
A and some q1 , q2 P A, such that q1 P Aˆ . The commutator rH, Hs of H is the A-module
generated by

rh1 , h2 s “ h1 h2 ´ h2 h1 , h1 , h2 P H.

The cocenter is the quotient A-module H{rH, Hs.

57
Lemma 164. (a) If w1 , w2 P W are strongly conjugate, then the images of Tw1 and Tw2
in the cocenter H{rH, Hs coincide.

(b) The cocenter H{rH, Hs is generated, as A-module, by the images of Tw where w P W

runs through all elements which are of minimal length in their conjugacy class.

Proof. (a) It suffices to prove the claim for elementary strongly conjugate elements. If
v P W satisfies ℓpw1 vq “ ℓpw1 q ` ℓpvq, ℓpvw2 q “ ℓpvq ` ℓpw2 q and w1 v “ vw2 , we get

Tw1 “ Tw1 v Tv´1 ” Tv´1 Tw1 v “ Tv´1 Tvw2 “ Tw2 pmod rH, Hsq.

(b) Induction on ℓpwq. If w has minimal length in its σ-conjugacy class, we are done.
Otherwise, we find by Marquis’ theorem a sequence
1s ns
w “ w1 ÝÑ ¨ ¨ ¨ ÝÑ wn`1

with ℓpw1 q “ ¨ ¨ ¨ “ ℓpwn q ą ℓpwn`1 q. By (a), we get Tw ” Twn pmod rH, Hsq. We
compute

Twn “Tsn wn`1 sn “ Tsn Twn`1 Tsn

”Ts2n Twn`1 “ pq2 Tsn ` q1 qTwn`1 “ q2 Tsn wn`1 ` q1 Twn`1 pmod rH, Hsq.

Since ℓpwn`1 q, ℓpswn`1 q ă ℓpwq, we are done by induction.

Definition 165. For a conjugacy class O Ď W , write TO P H{rH, Hs for the image of
Tw for any w P Omin .

This is well-defined by Marquis’ theorem. We saw above that these generate H{rH, Hs
as A-module.

Theorem 166 ([HN14]). Suppose that pW, Sq is spherical or affine. Then H{rH, Hs is
a free A-module with basis

tTO | O Ď W conjugacy classu.

It is an open problem whether the theorem holds for arbitrary Coxeter groups. But
at least, we have the following map:

Definition 167. Define

τ : H Ñ A,

to be the A-linear map sending Tw for w P W to

#
1, w “ 1,
τ pTw q “
0, w ‰ 1.

58
Lemma 168. We have τ phh1 q “ τ ph1 hq for all h, h1 P A. In other words, we get a
well-defined map

H{rH, Hs, h ` rH, Hs ÞÑ h.

Proof. Observe that

#
ℓpw1 q
q1 , w1 “ w2´1 ,
τ pTw1 Tw2 q “
0, w1 ‰ w2´1 .

Then both statements follow immediately.

Definition 169. If the statement of Theorem 166 holds true, we define the class poly-
nomials fw,O P A for w P W and O Ď W via the identity
ÿ
Tw ` rH, Hs “ fw,O TO P H{rH, Hs.
OĎW

Corollary 170. If W is spherical, the center of H is a free A-module with basis

ÿ ´ℓpwq
zO :“ fw,O Tw´1 q1
wPW

indexed by the conjugacy classes O Ď W .

Proof. Define the scalar product

H ˆ H Ñ A, ph, h1 q ÞÑ τ phh1 q.

Let h P H. We have

h central in H
ðñ hh1 “ h1 h for all h1 P H
ðñ τ phh1 Tw´1 q “ τ pTw´1 h1 hq for all h1 P H, w P W
ðñ τ phh1 h2 q “ τ ph2 h1 hq for all h1 , h2 P H
ðñ τ phh̃q “ 0 for all h̃ P rH, Hs.

So the center of H is the orthogonal complement of rH, Hs under this scalar product.
Thus, the composition

ZpHq Ñ H Ñ H{rH, Hs

yields an isomorphism of A-modules.

Let now h P H by central. The above calculation shows that

H{rH, Hs Ñ A, h1 ` rH, Hs ÞÑ τ phh1 q

59
is well-defined. For w P W , we compute
´ ÿ ¯
τ phTw q “τ phpTw ` rH, Hsqq “ τ h fw,O TO
O
ÿ ÿ ´ÿ ¯
“ fw,O τ phTO q “ τ pzO Tw qτ phTO q “ τ zO τ phTO qTw .
O
OĎW OĎW

Thus
ÿ
h“ τ phTO qzO .
OĎW

So the center of H, which is free with rank equal to the number of conjugacy classes in W ,
is contained in the submodule of H generated by the elements zO . This is only possible
if the center of H is equal to that submodule and the zO are linearly independent.

A peculiar class of elements, which plays an important role in different aspects of

Coxeter group theory, are the so-called straight elements.
Definition 171. An element w P W is called straight if for any n ě 1, we have

ℓpwn q “ nℓpwq.

Lemma 172. Let w P W and C P R such that for all n ě 1, we have

ℓpwn q ě nℓpwq ´ C.

(a) w is straight.

(b) Let O be the conjugacy class of w. Then Omin is given precisely by the straight
elements in O.
Proof. (a) For n, m ą 0, we get

mℓpwn q ě ℓpwmn q ě mnℓpwq ´ C.

Divide by m and take the limit m Ñ 8.

(b) If w1 “ v ´1 wv, then

ℓppw1 qn q “ ℓpv ´1 wn vq P rℓpwn q ´ 2ℓpvq, ℓpwn q ` 2ℓpvqs.

If ℓpw1 q ď ℓpwq ´ 1, we would get

ℓppw1 qn q ď nℓpw1 q ď nℓpwq ´ n “ ℓpwn q ´ n,

contradicting the above estimate. Hence w P Omin and all straight elements in O are
in Omin . If conversely w1 P Omin , we get ℓpw1 q “ ℓpwq so the above estimate shows

ℓppw1 qn q ě nℓpw1 q ´ 2ℓpvq.

Hence w1 is straight by (a).

60
Theorem 173 ([Mar18]). Let O be a straight conjugacy class, i.e. containing a straight
element. Then for any w1 , w2 P Omin , we have w1 Ñ w2 .

So straight elements are nice, but how do we find them?

Proposition 174. Let w P W and consider the iterated Demazure products

w˚,n :“ w ˚ w ˚ w ˚ ¨ ¨ ¨ ˚ w P W.

Define wn :“ pw˚,n´1 q´1 w˚,n P W .

(a) The sequence pwn qně1 stabilizes, i.e. there is an index N such that wN “ wn for all
n ě N.

(b) The element w8 :“ wN as in (a) is straight.

(c) If w1 is strongly conjugate to w, then pw1 q8 is conjugate to w8 .

Proof. (a) By definition of the Demazure product, observe that the value of wn only
depends on w and the cone type Cpw˚,n´1 q. Moreover, we have w˚,n´1 ďR w˚,n in
the right weak order. Hence

Cpw˚,1 q Ě Cpw˚,2 q Ě ¨ ¨ ¨ .

By finiteness of cone types, this sequence stabilizes (if S is infinite, consider the
parabolic subgroup generated by the support of w, which is always finite). Hence
the sequence pwn q stabilizes.

(b) By definition of the Demazure product, we get for every n ě 1 that

ℓpw˚,n`1 q “ ℓpw˚,n q ` ℓpwn q.

Thus
´ ¯
n
nℓpwN q “ ℓpw˚,N `n q ´ ℓpw˚,N q ď ℓ pw˚,N q´1 w˚,N `n “ ℓpwN q.

(c) It suffices to show this for elementary strongly conjugate elements. So let v P W
such that

ℓpwvq “ ℓpwq ` ℓpvq, ℓpvw1 q “ ℓpvq ` ℓpw1 q and vw1 “ wv.

For n ě 1, write

w˚,n ˚ v “ w˚,n vn , v ˚ pw1 q˚,n “ vn1 pw1 q˚,n .

Similar to (a), the sequences pvn qně1 and pvn1 qně1 stabilize. Observe that by asso-
ciativity of the Demazure product, we get

w˚,n ˚ v “ w˚,n´1 ˚ w ˚ v “ w˚,n´1 ˚ pwvq “ w˚,n´1 pvw1 q “ w˚,n´1 ˚ v ˚ w1 .

61
 Repeating this argument, we get
w˚,n ˚ v “ v ˚ pw1 q˚,n .
Now
” ı´1 ” ı ” ı´1 ” ı
vn´1 wn vn “ w˚,n´1 ˚ v w˚,n ˚ v “ v ˚ pw1 q˚,n´1 v ˚ pw1 q˚,n
” ı´1
“ pw1 q˚,n´1 pv 1 q´1 1 1 ˚,n
n´1 vn pw q .

For sufficiently large n, we get wn “ w8 , vn´1 “ vn , vn´1 1 “ vn1 and

” ı´1
pw1 q˚,n´1 pw1 q˚,n “ w8
1
.

We conclude
vn´1 w8 vn “ w8
1
.
This finishes the proof.

This proposition allows to associate to every σ-conjugacy class O Ď W a corresponding

straight σ-conjugacy class.
Exercise 175. Assume that S is finite. Show that
suptℓpwq ´ ℓpw8 q | w P W u ă `8.
Conclude that there are infinitely many straight conjugacy classes if W is infinite and
irreducible. Show that there is only one straight conjugacy class if W is finite.
Exercise 176. For W “ S3 and A “ Zrq1 , q2 s, write the image of Tw0 in H{rH, Hs as
A-linear combination of TO ’s.

Suggestions for presentation topics

Those who formally enrolled for the course are required to give a short presentation of
20–30 minutes on a subject related to Coxeter groups. Longer talks are possible, but
please inform me beforehand and keep it under 60 minutes.
Please try to give a motivating and instructive talk, focussing on results and appli-
cations rather than elaborate technical proofs. You can find a couple of suggestions for
topics below, or you use your own topic after confirming it is suitable for this course.
Please email me once you made a decision. Topics are first come, first serve.

The type Dn
The remaining infinite family of finite Coxeter groups is the type Dn for n ě 4. Explain
how these groups look like, how to do computations like length, descent sets and Bruhat
order. You may follow [BB05, Section 8.2], or you construct the root system of type
Dn (using any standard Lie theory reference) and define it using the usual Weyl group
construction.

62
Order automorphisms
A right weak order automorphism is a bijective map f : W Ñ W such that v ďR w if
and only if f pvq ďR f pwq. A Coxeter group automorphism is a group automorphism
f : W Ñ W with f pSq “ S. Show that, in most of the interesting cases, these two
notions agree following [BB05, Theorem 3.2.5]. Explain the situation for Bruhat order
automorphisms, using [BB05, Theorem 2.3.5].

Finite reflection groups

A finite subgroup G ď GLn pRq generated by reflections is called a finite reflection group.
It is a classical result that these are “the same as” finite Coxeter groups. Explain what
this means, e.g. following [Hum90]. You may either sketch what goes into the proof of
such a correspondence, or how to use a classification of finite Coxeter groups to obtain
a classification of regular polyhedrons (e.g. Platonic solids).

Braid groups and Jones polynomial

In a seminal paper, Jones [Jon87] introduced the so-called Jones polynomial to knot
theory, the study of smooth embeddings of the circle into R3 . Following his paper,
explain what Braid groups are, how they relate to Coxeter groups of type An and how
they relate to knot theory. Explain how to compute the Jones polynomial of a knot
and, as time permits, how Jones derives his polynomial from the representation theory
of Iwahori-Hecke algebras.

Flag varieties
Let V “ Cn . A complete flag in V is a chain of sub-vector spaces

t0u Ĺ U1 Ĺ ¨ ¨ ¨ Ĺ Un´1 Ĺ V,

so that dim Ui “ i. Let G “ GLn pCq and B Ă G the subset of upper triangular matrices.
Explain the one-to-one correspondence between complete flags in V and the flag variety
G{B, e.g. following [Bri04].
Explain why the flag variety is projective, i.e. embeds into some projective space PN pCq
as closed subset (e.g. by embedding G{B into a suitable product of Grassmannians, then
using projectivity of Grassmannians as a black box).
Define the Schubert cell BwB Ă G{B for w P Sn and show why it is isomorphic (as
variety, or manifold) to Cℓpwq . Explain some applications of flag varieties, or explain how
to show that the covering relations of Schubert cells are given by the Bruhat order.

Further topics
• Shellability of Bruhat order [BB05, Section 2.7].

• Normal forms of reduced words [BB05, Section 3.4]. (taken)

63
 • Introduction to Coxeter group computations with computer algebra (GAP, SAGE
etc.).

• Young tableaux and the symmetric group.

• Unipotent orbits and conjugacy classes. (taken)

• Coxeter matroids. (taken)

• Categorification of Kazhdan-Lusztig polynomials. (taken)

• Conjugacy classes in finite Weyl groups. (taken)

References
[BB05] A. Björner and F. Brenti. Combinatorics of Coxeter Groups. Springer-Verlag
Berlin Heidelberg, 2005.
[Bon17] C. Bonnafé. Kazhdan-Lusztig Cells with Unequal Parameters. Cham, Switzer-
land: Springer International Publishing, 2017.
[BFP98] F. Brenti, S. Fomin, and A. Postnikov. Mixed Bruhat Operators and Yang-
Baxter Equations for Weyl Groups. International Mathematics Research No-
tices 8 (1998), pp. 419–441.
[Bri04] M. Brion. Lectures on the geometry of flag varieties. arXiv (2004). eprint:
math/0410240.
[Deo77] V. V. Deodhar. Some Characterizations of Bruhat Ordering on a Coxeter
Group and Determination of the Relative Möbius Function. Inventiones Math-
ematicae 39 (1977), p. 187.
[Deo89] V. V. Deodhar. A note on subgroups generated by reflections in Coxeter
groups. Arch. Math. (Basel) (1989).
[Dye93] M. J. Dyer. Hecke algebras and shellings of Bruhat intervals. Compositio
Mathematica 89.1 (1993), pp. 91–115.
[HN14] X. He and S. Nie. Minimal length elements of extended affine Weyl groups.
Compositio Mathematica 150.11 (2014), pp. 1903–1927.
[Hum90] J. E. Humphreys. Reflection Groups and Coxeter Groups. Cambridge, Eng-
land, UK: Cambridge University Press, 1990.
[Jon87] V. F. R. Jones. Hecke Algebra Representations of Braid Groups and Link
Polynomials on JSTOR. Ann. Of Math. 126.2 (1987), pp. 335–388.
[KL79] D. Kazhdan and G. Lusztig. Representations of Coxeter groups and Hecke
algebras. Invent. Math. 53.2 (1979), pp. 165–184.
[Mar18] T. Marquis. Cyclically reduced elements in Coxeter groups. arXiv (2018).
eprint: 1812.02543.

64
[Mat64] H. Matsumoto. Générateurs et relations des groupes de Weyl généralisés.
Comptes Rendus Hebodomadaires Des Seances De L’Academie Des Sciences
258.13 (1964), p. 3419.

65