You are on page 1of 23

Representations of S

n
1 Introduction
The ultimate goal of representation theory - at least when considered as a stan-
dalone topic - is to give a complete description of the representations of some
algebraic object, and to provide a method for calculating the properties and
invariants thereof. When the object under consideration is a nite group, it
suces to nd the irreducible representations and to give their characters. Fur-
thermore, we know that this is a nite task, since there are exactly as many
irreducible representations as there are conjugacy classes of the group. Unfortu-
nately this is an area where the theory is more dicult than the practice: while
there are many techniques for calculating the characters of any given group,
there is no general method for constructing the representations.
In the case of the symmetric groups, however, each conjugacy class of S
n
corresponds to a partition of n. We can use the combinatorial properties of
these partitions to explicitly construct the irreducible representations, and then
go on to calculate the characters and other properties. This gives us a complete
description of the representation theory for a whole family of groups at once.
This essay is essentially split in half. The rst half develops the theory
outlined above, with a focus on computation. We assigning to each partition a
diagram of boxes permuted by S
n
, and by manipulating these diagrams we prove
a series of results, including an explicit formula for the irreducible characters
and some remarkable combinatorial identities.
The emphasis in the rst half is on providing a straightforward and self-
contained proof of the main results, and we have to avoid much of the related
theory. Two omissions in this section are particularly notable. Firstly, although
the results hold for any eld with characteristic p > n, for concreteness we work
over C. Secondly, no mention is made of Schur polynomials, a collection of
symmetric functions indexed by partitions which are intimately linked to this
theory. Indeed, many of our results could be stated and proved in terms of these
functions; the reader is referred to [1].
Given the success of this theory, it is natural to wonder whether structures
related to the symmetric groups might have an equally nice representation the-
ory. In the second half of the essay we develop the theory of cellular algebras,
due to Graham and Lehrer, in which the irreducible representations of an alge-
bra can be determined from the existence of a basis with certain combinatorial
properties. In chapter 4 we introduce the Iwahori-Hecke algebra, a generalisa-
tion of the symmetric group algebra CS
n
, and demonstrate that it is cellular
so we can apply the theory from the previous chapter. We nish by applying
our new machinery to the symmetric group algebra, giving a description of the
irreducible representations over arbitrary elds.
Chapter 2 is taken from varous parts of [1] and [5]. Chapter 3 closely fol-
lows the paper of Graham and Lehrer [3], although we simplify in many places
because we work entirely over elds instead of rings. Chapter 4 is from [2].
1
2 Representations of Symmetric Groups
2.1 Young Tableaux and Irreducible Representations
This chapter examines the representations of the symmetric groups, largely
following [1].
Conjugacy classes of S
n
correspond to partitions of n. A partition is a set
of non-increasing positive integers (
1
,
2
, ...,
k
) with

i
= n. The Young
diagram corresponding to is constructed from one row of k boxes for each
element of the partition with k elements. Thus the partition (4, 3, 3, 1) of 11
elements would generate the diagram:
Conversely, any diagram with n boxes gives a unique partition; thus we will
often conate the two.
We will be considering the group of permutations of the n boxes in a given
diagram. In order to associate this group with S
n
, we need to number the boxes.
A Young tableau is a numbering of a Young diagram with the numbers 1,...,n.
There are clearly n! dierent tableaux on each Young diagram, including the
regular tableau consisting of a lexicographic numbering:
1 2 3 4
5 6 7
8 9 10
11
We write t

for the regular tableau on .


For a general nite group G there are the same number of irreducible rep-
resentations as there are conjugacy classes of G, but in general there is no way
to give an explicit bijection between the two sets. In the case of the symmetric
groups, however, each conjugacy class of S
n
corresponds to a distinct Young
diagram, via the corresponding partition. We nd that we can use each Young
diagram to generate an distinct irreducible representation of S
n
.
Given a Young diagram with n boxes, we want to let S
n
act on the set
t : t is a tableau on by permuting the boxes. In order to dene this action,
we need to label the boxes by picking an initial tableau. By symmetry, this
choice will make no dierence to the following theory, but it must be made, so
for concreteness we choose the regular tableau t

. Now a permutation in S
n
2
swapping i and j will act on any tableau by swapping those boxes labelled i and
j in the standard tableau. For example,
(12)(456)
1 2 3
4 5
6
=
2 1 3
6 4
5
(234)(56)
6 5 4
3 2
1
=
6 3 5
4 1
2
We begin by dening the subgroup A of S
n
as the subgroup permuting only
along the rows of our tableau, and B as the subgroup which permutes only
up and down the columns (notice that if we had used a dierent tableau to
dene the action then we would obtain dierent but conjugate subgroups at
this point). We then dene an element c

CS
n
as follows:
c

A,B
(sgn)
Since A and B have a trivial intersection, each term in c

is distinct, and
thus the non-zero coecients in c

are 1, depending on the sign of . In


particular, c

,= 0. The element c

is known as the Young symmetrizer of .


Worked Example We consider S
3
. There are three partitions of 3: a single
group of three, three groups of one, or a group of two and a group of one. Thus
the three Young diagrams are
1 2 3
1
2
3
1 2
3
each labelled with the regular tableau to make clear the action of S
3
on the
boxes. The rst has all of S
3
permuting the only row, and only the identity
permuting columns. Thus the Young symmetriser is
c
(3)
=

wS
3
w 1
The second has A = 1 and B = S
3
, so
c
(1,1,1)
=

wS
3
(sgnw)w
The nal diagram has 1, (12) permuting the rows while 1, (13) permutes
columns. The Young symmetriser is
c
(2,1)
= 1 + (12) (13) (12)(13)
where the minus signs come from the sign of (13).
The remainder of this section proves the following result:
Theorem 2.1. The set (CS
n
)c

: a partition of n, with S
n
acting on the
left by multiplication, is a complete set of irreducible representations of S
n
.
3
Worked Example We continue with our example of S
3
. The theorem claims
that there is one distinct representation for each of the Young symmetrisers.
The rst is
CS
3
c
(3)
= CS
3

wS
3
w = Cc
(3)
So we have a one-dimensional subspace, invariant under S
n
: the trivial rep-
resentation. The second is
CS
3
c
(1,1,1)
= Cc
(1,1,1)
This time the action of w S
3
depends on the sign of w, so this is the
alternating representation. Finally
CS
3
c
(2,1)
=

c
(2,1)
, (13)c
(2,1)
_
which is the standard two-dimensional representation of S
3
. These form a
complete set of CS
3
-modules, as predicted.
(Notice that CS
3
contains one more copy of the two dimensional submodule.
We will see at the end of this chapter how to nd a complete decomposition in
a natural way).
Proving the Theorem. The theorem consist of three claims: that (CS
n
)c

is
an irreducible submodule of CS
n
; that each dierent choice of partition produces
a distinct S
n
module; and that all irreducible S
n
modules are produced in this
way. The key to our proof will be the following technical lemma:
Lemma 2.2. An element x of CS
n
is a scalar multiple of c

if and only if for


all A, B we have x (sgn) = x
Proof. The forward implication is clear from the denition of c

. Suppose con-
versely that we have an element x of CS
n
with x (sgn) = x for all
A, B. Then we write
x =

gS
n
k
g
g (k
g
C)
This implies that
k
g
= (sgn)k
g
for all g and all A, B.
First we need to show that k
g
= 0 for any g , AB. Fix g, and suppose we can
nd a transposition t of two elements in the same row of the tableau (that is, a
transposition t in A) such that g
1
tg is in B. If we can nd such a transposition
then k
g
must be zero: note that g = (t)g(g
1
tg), so k
g
= sgn(g
1
tg)k
g
= k
g
.
We claim that such a t exists for all g , AB.
If there are two digits in the same row of T whose images under g
1
are in
the same column, then a transposition of these two digits is the required t. If
this is never the case, then g
1
must send each element in the top row of T to
a distinct column. Therefore there is an element
1
of B (that is, an element
of S
n
permuting only columns) moving the elements of the top row directly
4
downwards so that g
1
T has the same (unordered) top row as T. Thus there is
an element
1
of A such that
1
g
1
is the identity on the top row of T.
Setting g

=
1
g
1
, we repeat the same argument considering the second
row. We take care to leave the top row unchanged as we generate
2
and
2
such that
2
g

2
xes the top two rows. Repeating to the bottom, we nd

k
...
1
g
1
...
k
= e, whence g AB as required.
To nish, we need to show that each element of AB has the correct non-zero
coecient in x. But k
g
= (sgn)k
g
and in particular k

= (sgn)k
e
. Thus
x is a scalar multiple of c

as claimed.
Corollary 2.3. For all elements x of CS
n
, the product c

xc

is a scalar multiple
of c

.
In particular, c

is a scalar multiple of c

. We will need to show that this


is non-zero.
Lemma 2.4. Write c

= kc

. Then k =
n!
dimCS
n
.
Proof. We count the trace of multiplication by c

in two dierent ways. Taking


the standard basis of the group algebra, the trace is dimCS
n
multiplied by the
coecient of the identity in c

, which gives n! 1. Alternatively, taking a basis


for CS
n
and extending to a basis of the whole algebra, we nd that the trace is
(dimCS
n
) k.
We can now prove the rst of our claims: that (CS
n
)c

is irreducible.
Suppose we have a CS
n
-module decomposition (CS
n
)c

= V W. Then
c

(CS
n
)c

= Cc

,= 0 by the previous lemma. Without loss of generality


assume that c

V = Cc

. But then (CS


n
)c

= (CS
n
)c

V which is in V, so in
fact W is trivial. We conclude that (CS
n
)c

is irreducible, and write V

for this
representation.
Remark. We see at this point why the results of this section do not hold
over elds of characteristic p < n: over these elds, Lemma 2.4 does not prove
that c

,= 0, so V

is not necessarly irreducible. As noted in the introduction,


we will see in Chapter 4 that we can still make some progress in this case, but
for now we continue working over C.
To prove the other claims, we introduce the lexicographical ordering on
Young diagrams. If and are Young diagrams then we write > if
i
>
i
for the rst i at which they dier. Note that this is a total order.
We prove the following:
Lemma 2.5. For a given partition dene
a

A
b

B
(sgn)
so that c

= a

. If > then a

x b

= 0 for all x.
Proof. It suces to show that a

g b

= 0 for any g in S
n
. Up to this point, we
have used the regular tableau t

to determine how S
n
acts on the boxes of the
Young diagram . If instead we use the tableau gt

, we nd that the rows and


columns are xed by the sets gAg
1
and gBg
1
, with the rest of our contructions
altered similarly. In particular, since our claim is that a

(gb

g
1
) = 0, we can
consider the tableau gt

, so it suces to prove that a

= 0.
5
Let T, S be the tableaux on the partitions , . Since > it follows from
a pigeonhole argument that there are two digits in the same row of T and the
same column of S. Let t be the transposition of these two digits; then a

t = a

while tb

= b

, so a

= a

t
2
b

= a

as required.
Now we can prove the second of our claims. Suppose ,= . We wish to
prove that V

= V

. Assume without loss of generality that > . If the


two are isomorphic then we would expect their images under any element of
CS
n
to be isomorphic, since the isomorphism commutes with the group action.
However, we have c

,= 0 while c

= 0 by the above.
We have constructed a set of pairwise non-isomorphic irreducible CS
n
mod-
ules, one for each partition of n. It is well known that the number of ir-
reducible representations of a nite group is equal to the number of conju-
gacy classes, which for S
n
is the number of partitions; therefore the set V

:
a partition of n must be a complete set of irreducible representations.
2.2 Frobenius Formula
The aim of this section is to give some practical value to the results above by
proving a formula for the characters of the irreducible representations V

.
Frobenius Formula Let C be the conjugacy class of S
n
whose elements
have c
1
1-cycle, c
2
2-cycles, and so on. Then the value of the irreducible char-
acter of V

on C is equal to the coecient of x


1

1
+k1
x
2

2
+k2
...x
k

k
in the
polynomial
(x) (x
1
+... +x
k
)
c
1
(x
2
1
+... +x
2
k
)
c
2
...(x
n
1
+... +x
n
k
)
c
n
where (x) is the Vandermonde determinant

i<j
(x
i
x
j
).
Some preparation is needed before we can give a proof. Given a partition
of n, dene the Young subgroup S

of S
n
to be the subgroup permuting only
the rows of . That is,
S

= S

1
S

2
... S

k
S
n
We let 1

be the trivial character of this subgroup, and let

= 1

S
n
be the induced character on all of S
n
. We will use this induced character as an
intermediate step in calculating the irreducible character of V

.
The character induced from a subgroup H to a group G can be calulated
using the well-known formula:
( G)(C) =
[G[
[H[

i
[D
i
[
[C[
(D
i
)
where is a character on a subgroup H and the conjugacy class C of G is the
union of the conjugacy classes D
i
in H. When is the trivial character, this
becomes
(1 G)(C) =
[G[[C H[
[H[[C[
6
In our case, H is the Young subgroup S

, and we let the conjugacy class C


have c
1
1-cycles, c
2
2-cycles, and so on. Then we have
[G[ = n!
[S

[ =
1
!
2
!...
k
!
[C[ =
n!
1
c
1
c
1
!2
c
2
c
2
!...n
c
n
c
n
!
Finally, consider an element x of CS

, so x is in S

= S

1
...S

k
. Let the
S

i
component of x be the product of p
i1
1-cycles, p
i2
2-cycles, and so on up to
p
in
n-cycles. Then we have a collection of integers p
ij
with 1 < i < k, 1 < j < n
such that

j
jp
ij
=
i

i
p
ij
= c
j
We can then count the number of elements of C S

as
[C S

[ =

i
!
1
p
i1
p
i1
!...n
p
in
p
in
!
where the sum is over all such collections of integers p
ij
.
Now we combine these values to calculate the value of the induced character

(C) = (1

G)(C) =
n!

i
!
1
p
i1p
i1
!...n
p
inp
in
!

1
!
2
!...
k
!(
n!
1
c
1c
1
!2
c
2c
2
!...n
c
n
c
n
!
)
which simplies to

(C) =

j
c
j
!
p
1j
!...p
kj
!
where the sum is still over all such collections of integers p
ij
. This expression is
more useful to us if we notice that it is exactly the coecient of x
1

1
...x
k

k
in
the product
(x
1
+... +x
k
)
c
1
(x
2
1
+... +x
2
k
)
c
2
...(x
n
1
+... +x
n
k
)
c
n
(When expanding the product and considering how the required term can be
formed, the integer p
ij
represents the power of x
j
i
taken from the jth bracket)
Now write

for the character of the irreducible representation V

. We have
already shown that this is a complete set of irreducible characters, so we have

for some non-negative integers K

. These integers, indexed by two partitions


of n, are known as Kostka numbers.
7
Lemma 2.6. The Kostka numbers K

satisfy K

> 0 and K

= 0 for
< .
Proof. By Frobenius reciprocity we have
(1

S
n
),

) = 1

, (

))
So we are looking for one-dimensional subspaces of V

on which S

acts
trivially. Recall that we constructed c

from sets A and B which xed rows and


columns of respectively, and that we set
V

= (CS
n
)c

= (CS
n
)

B
(sgn)
If = then A is exactly the Young subgroup S

, so V

contains the
one-dimensional submodule Cc

which is xed under the (left) action of S

.
Therefore

contains the trivial representation, and thus K

> 0.
Now let < . If xc

spans a trivial subrepresentation then we would


expect a

xc

= [H

[xc

. But by Lemma 2.5 we have a

xc

= 0. Therefore
K

= 0.
Now we have some information about the multiplicity of the irreducible

in the induced character

. We are almost ready to prove Frobenius formula,


but rst we need the following identity:
Lemma 2.7 (Cauchys identity).
det
_
1
1 x
i
y
j
_
=

i<j
(x
i
x
j
)(y
i
y
j
)

i,j
(1 x
i
y
j
)
Proof. We will prove a slightly dierent form:
det
_
1
x
i
y
j
_

i,j
(x
i
y
j
) =

i<j
(x
i
x
j
)(y
i
y
j
)
To see this, note that the left side is a homogeneous polynomial divisible by

i<j
(x
i
x
j
)(y
i
y
j
), since it disappears whenever x
i
= x
j
or y
i
= y
j
for any
i ,= j. By counting degrees, we see that the left side is in fact a scalar multiple
of the right.
We have to prove that the constant of proportionality is 1. To do this, take
the product inside the determinant on the left by multiplying the ith row by

j
(x
i
y
j
). Then the (i,j)th entry in the matrix is

k=j
(x
i
y
k
). Now we take
x=y, so all the entries are zero except down the diagonal; then the determinant
is

i,j=i
(x
i
x
j
) =

i<j
(x
i
x
j
)
2
, which is equal to the right hand side.
Now dividing both sides by

i,j
(x
i
y
j
) and setting z
i
= 1/x
i
gives us the
stated form of the identity.
We are ready to prove Frobenius formula; we follow [5]. Write for the
Vandermonde determinant:
(x) =

1i<jk
(x
i
x
j
)
8
Theorem 2.8 (Frobenius Formula). The value of the irreducible character

on the conjugacy class C is equal to the coecient of x


1

1
+k1
x
2

2
+k2
...x
k

k
in the polynomial
(x) (x
1
+... +x
k
)
c
1
(x
2
1
+... +x
2
k
)
c
2
...(x
n
1
+... +x
n
k
)
c
n
Proof. Write

for the class function given by these coecients, so our claim


is that

. We will show that

is an irreducible character and that

contains

, which in turn contains

. Our rst step, therefore, is to prove


that

) = 1. We have

) =
1
n!

C
[C[

(C)
2
This is equal to the coecient of (x
1

1
+k1
...x
k

k
)(y
1

1
+k1
...y
k

k
) in:
(x)(y)

C
[C[
n!
(x
1
+...+x
k
)
c
1
...(x
n
1
+...+x
n
k
)
c
n
(y
1
+...+y
k
)
c
1
...(y
n
1
+...+y
n
k
)
c
n
= (x)(y)

m
(

j,k
x
m
j
y
m
k
)
c
m
m
c
m
c
m
!
()
Now the term we are interested in has degree k(k 1)n. Any term in
(x)(y) is of degree k(k 1), so we can replace the rest of the expression with
any other series containing the same terms of degree n, and still obtain the same
coecient. Consider the terms of degree n in

i=0
(

j,k
x
m
j
y
m
k
)
i
m
i
i!
For every collection of positive integers c
i
with

ic
i
= n there is exactly one
term

m
(

j,k
x
m
j
y
m
k
)
c
m
m
c
m
c
m
!
in this expression of degree n; but each such collection of integers corresponds
to a unique conjugacy class C. Therefore we can replace () with
(x)(y)

i=0
(

j,k
x
m
j
y
m
k
)
i
m
i
i!
= (x)(y)

m
exp
_
_

j,k
x
m
j
y
m
k
m
_
_
= (x)(y) exp
_
_

j,k
log(1 x
j
y
k
)
_
_
= (x)(y)

j,k
1
1 x
j
y
k
9
By Cauchys identity above, this is equal to
= det
_
1
1 x
i
y
j
_
= det
_

r=0
(x
i
y
j
)
r
_
Finally the term we require is symmetric in x and y, so it has coecient 1
as required.
Having shown that

is an irreducbile character, we consider how multiples


of x
1

1
+k1
x
2

2
+k2
...x
k

k
arise in the product of Frobenius formula. Such
terms are formed from the product of a term in (of degree
k(k1)
2
) and a term
in (x
1
+ ... + x
k
)
c
1
(x
2
1
+ ... + x
2
k
)
c
2
...(x
n
1
+ ... + x
n
k
)
c
n
of degree n. Every term of
degree n in the latter product has a coecient which is the value of

for some
(as we proved above by calculating

) so the required coecient

can be
written as

where L

is 1 if there is a term in which multiplies with x

1
1
...x

k
k
to give
x
1

1
+k1
x
2

2
+k2
...x
k

k
(1 depending on the sign of that term in ) and zero
if contains no such term.
Lemma 2.9. L

= 1, and L

= 0 if < .
Proof. We see that contains x
k1
1
x
k2
2
...x
0
k
, so L

= 1. However, we note
that a term in contains x
1
to the power of k-1 at most. Thus if L

is non-
zero then
1

1
. If a term in contains x
1
to the highest possible power
then it contains x
2
to the power of at most k-2, so
1
=
1
and
2

2
, and
so on. We deduce that the only

with a non-zero coecient are those with


>= .
Now we can nish our proof of Frobenius formula by showing that

.
By denition

,
L

It follows from our results about L

and K

that

appears in exactly one


non-zero term, with coecient L

> 0, so

contains

. But

) = 1,
so the two are equal and the theorem is proved.
2.3 Decomposition of CS
n
The aim of this section is to extend the basic results above by giving an explicit
decomposition of CS
n
into irreducible submodules. In order to do so, we rst use
Frobenius formula to calculate the dimension of an irreducible representation.
Our proof is adapted from [4]. We dene a standard tableau on a given
Young diagram to be tableau in which the numbers are increasing along rows
and columns.
10
Lemma 2.10. The dimension of V

is equal to the number of standard tableaux


on .
Proof. By Frobenius formula, dimV

is the coecient of x
1

1
+k1
x
2

2
+k2
...x
k

k
in
(x) (x
1
+... +x
k
)
n
We consider taking (x) and multiplying by (x
1
+...+x
k
) at each step, a total
of n times, keeping track of the terms that are produced. The key observation
is that after each multiplication, the result is an alternating function. Therefore
after each multiplicaton by (x
1
+ ... + x
k
), any term containing two variables
with the same index will have coecient zero.
Note that the term we are interested in has indices in strictly decreasing
order:
(
1
+k 1) > (
2
+k 2) > ... >
k
The only term in with strictly decreasing indices is x
k1
1
x
k2
2
...x
1
k1
. By
the observation above, the only term in (x
k1
1
x
k2
2
...x
1
k1
)(x
1
+ ... + x
k
) with
a non-zero coecient is (x
k1
1
x
k2
2
...x
1
k1
)x
1
. Multiplying by (x
1
+ ... + x
k
)
again, the two terms with non-zero coecients are (x
k1
1
x
k2
2
...x
1
k1
)x
1
x
1
and
(x
k1
1
x
k2
2
...x
1
k1
)x
1
x
2
. At each stage, the only terms which can have non-zero
coecients are those constructed by multiplying a previous non-zero term by a
variable x
i
in such a way that the new term still has strictly decreasing indices.
Every such sequence of variables containing exactly
i
copies of each x
i
will give one copy of our required term in the nal product. Furthermore the
required term cannot arise starting from a dierent term in , whose indices
begin in a dierent order, since there must be some multiplication after which
two variables have the same index and the term has coecient zero.
It follows that the coecient we require is exactly the number of such se-
quences of variables. But if we associate multiplication by x
i
with label the
next box in row i, a little thought shows that the number of such sequences
is exactly the number of ways to ll in a standard tableau on . The result is
therefore proved.
Now we can give our decomposition of CS
n
. For a given Young diagram ,
write Std() for the set of all standard tableaux on . Above, we dened c

using the regular tableau on a given Young diagram , noting that we could
have chosen any other tableau. We now let c
,T
(T Std()) be constructed
in the same way using the tableau T instead of the regular tableau. That is,
we let A
T
, B
T
be the subgroups xing rows and columns of T repectively, and
dene
c
,T
=

A
T
B
T
(sgn)
It is clear that c
,S
is conjugate to c
,T
for any tableaux T, S on the same
diagram.
Theorem 2.11.
CS
n
=

TStd()
(CS
n
)c
,T
11
Proof. We will rst show that the sum on the right is direct. We know that
(CS
n
)c
,S

= (CS
n
)c
,T
for any two tableau S and T on the same diagram,
and we proved above that submodules generated using dierent diagrams are
non-isomorphic, so by irreducibility the sum is direct at least over the dierent
diagrams:

TStd()
(CS
n
)c
,T
=

TStd()
(CS
n
)c
,T
We next prove that c
,T
c
,S
= 0 when S and T are distinct standard tableaux
on the same diagram . Suppose the rst box at which S and T dier (reading
from left to right, then top to bottom) has the digit i in S and j in T. Suppose
without loss of generality that i < j, and consider where the digit i lies in T. It
cannot lie above or to the left of j, its location in S, because S and T agree in
these boxes, and it cannot lie anywhere to the right of j because T is a standard
tableau. It must lie in a lower row, below or to the left of j. Thus there are
two digits - the digit i, and some other directly left of j in T - which lie in the
same row of S and the same column of T. Let t be the transposition of these
two digits.
Now b
,T
t = b
,T
while ta
,S
= a
,S
. Thus
b
,T
a
,S
= b
,T
t
2
a
,S
= b
,T
a
,S
so c
,T
c
,S
= 0. Therefore (CS
n
)c
,S
(CS
n
)c
,T
= 0, so the sum is direct as
required.
To nish, we recall from elementary representation theory that V

occurs
with multiplicity dimV

in the group algebra. Above we showed that dimV

=
[ Std()[, and we have one distinct copy for each standard tableau, so the sum
covers the entire group algebra and we are done.
If we count dimensions, we obtain a nice combinatorial identity:
Corollary 2.12.

[ Std()[
2
= n!, where the sum is over all partitions of n.
We will nd a use for this identity later in examining the Hecke algebras.
2.4 The Hook Length Formula
In this section we continue with CS
n
, obtaining another elegant expression for
the dimension of the irreducible representation V

.
According to Frobenius formula, dim V

is the coecient of x
1

1
+k1
...x
k

k
in
(x) (x
1
+... +x
k
)
n
With some manipulation we can show that this coecient is
dimV

=
n!
(
1
+k 1)!(
2
+k 2)!...(
k
)!

i<j
((
i
+k i) (
j
+k j))
For every box in a Young diagram we dene the hook at that point to be
the box itself together with every box directly below and directly to the right.
12
The hook length is the total number of boxes involved. In the diagram below a
hook of length 5 is shaded and each box is labelled with its hook length.
Theorem 2.13. For any ,
dimV

=
n!

(Hook length)
where the product is over all boxes in the diagram.
Examples. The hook lengths on the three partitions of 3 are
3 2 1
3
2
1
3 1
1
so the theorem states that the rst two generate representations of dimension
3!
321
= 1 and the last a representation of dimension
3!
311
= 2, which is true. The
diagram above of 11 boxes corresponds to a representation of S
11
of dimension
11!
7 5 5 4 4 3 2 2
= 1188
Proof. By our expression for dimV

, we need to prove

(Hook lengths) =
(
1
+k 1)!(
2
+k 2)!...(
k
)!

i<j
((
i
+k i) (
j
+k j))
We will prove this by showing that
(
i
+k i)!

i<j
((
i
+k i) (
j
+k j))
is the product of all hook lengths of boxes on the ith row.
Notice that each hook on the ith row passes through a unique box on the
lower edge of the diagram, and that (
i
+k i) is the length of the largest hook
on the row, corresponing to the square in the bottom left corner of the diagram.
Starting in this corner, we imagine a path which traverses the bottom edge by
moving up or right one step at a time. We set a counter h = (
i
+ k i) to
begin with, and decrease h by one with each step. We stop at the far right end
of the ith row, at which point h = 1.
After each step right, the current square is on the bottom edge of the diagram
and contributes a hook of length h to the product. Figure 1 shows the path
13
traversed to nd the hooks on the second row of an example diagram; a square
on the bottom edge is shaded, together with the hook of length 7 to which it
corresponds.
If we never have to move upwards (if we are considering the very bottom
row) then the resulting product will be (
i
+k i)!. If we do move up then the
current square is not on the bottom edge, and we have to delete this value of h
from the factorial to get the correct product. This occurs exactly when we can
t another hook of length (
j
+k j) below our current position, as in gure 2:
At these points we have h = ((
i
+k i) (
j
+k j)) for some j > i,
which proves the theorem.
We proved in the previous chapter that the dimension of V

is also the
number of standard tableaux on , so we have the remarkable identity
Corollary 2.14 (Frame-Robinson-Thrall, 1954).
n!

(Hook lengths)
= [ Std()[
At rst glance it seems that a simple proof of this identity would be easy
to nd. A wide range of proofs have indeed been found, but none of them are
as simple as we would like. A bibliography can be found in [8], in which Sagan
notes that all the simple proofs are not combinatorial, and all the combinatorial
proofs are not simple!
14
3 Cellular Algebras
In this section we develop the theory of cellular algebras due to Graham and
Lehrer. We follow their paper [3]. In the next chapter we will show that these
results apply to a generalisation of the symmetric group algebra CS
n
.
A cellular algebra is an algebra A over a eld F together with (, M, C, *)
such that:
is a nite partially ordered set;
M is a function assigning to each in a nite set M();
C is a function from

M() M() to A, sending (S,T) to an element


C
(S,T)
, such that the image of C forms an F-basis of A;
The operation * is the anti-involution in A dened by C

(S,T)
= C
(T,S)
(an
anti-involution being a map such that (ab)

= b

);
Finally we demand that for any a A the product aC
(S,T)
is, mod-
ulo lower-order terms, a linear combination of C
(R,T)
in which the coe-
cients are independent of T. Here, lower order terms means the span of
C
(U,V )
: U, V M() for some < . That is,
aC
(S,T)
=

R
r
a
(R, S)C
(R,T)
+ lower order terms
where the function r
a
, which gives the coecients in F, depends on S and
S but not on T.
Examples Let A be the ring F[x]/(x
n
) in one variable over a eld. Let be
N ordered in reverse (so 0 > 1 > ... > n) and let M(i) have just one element for
each i . Then A is cellular with basis x
i
: i , and for each polynomial a
the coecient r
a
is simply the polynomials unit coecient. If we did not insist
that be nite then F[x] would also be cellular with = 0, 1, ....
Let A be the algebra of n n matricies over a eld F, and let E
ij
be the
elementary matrix with a 1 at component (i,j) and zero elsewhere. Then E
ij

is a basis for A. Furthermore, the product AE


ij
for any matrix A is in the span
of E
kj
: 1 < k < n with coecients independent of j. Therefore this algebra
is cellular, taking = 1, ..., n and the transpose map as

.
Let , be partially ordered sets. Then is partially ordered by letting
any element of be greater than any element of . If A, B are cellular algebras
with posets , respectively then A B is a cellular algebra on the poset
, the basis vectors being (C
(S,T)
, 0) for S, T M() and (0, D
(U,V )
) for
U, V M(). Together with Wedderburns theorem, these facts show that any
semisimple algebra is cellular.
If L is a subset of then we write A(L) for the span of those basis vectors
originating from some L. We then have a rst basic result:
Lemma 3.1. If L is a poset ideal (that is, if L, < L) then A(L)
is a two-sided ideal of A.
15
Proof. The multiplication condition proves immediately that A(L) is a left ideal.
If we apply the involution

to this condition, we obtain
C
(T,S)
a

r
a
(S

, S)C
(T,S

)
+ lower order terms
which shows that A(L) is also a right ideal.
The use of cellular algebras is their application to the study of representa-
tions. Write A( ) for the span of C
(S,T)
: S, T M() for some .
Then the multiplication condition ensures that the quotient A( )/A(< ) is
an A-module. By the independence of the coecients on one of the indicies,
this module is the direct sum of [M()[ isomorphic copies of some smaller A-
module, which is called the cell module ([3]) or standard module ([6]) at .
We prove below that every irreducible A-module is a quotient of a cell module.
For each in we can dene a left A-module W

as follows: Let C
S
: S
M() be a set of basis vectors and let V

be the vector space over F generated


by this set. Let the action of A on this space be given by
aC
S
=

RM()
r
a
(R, S)C
R
This is just an explicit construction of the cell module mentioned above. We
also dene a bilinear form f on V

, which will turn out to be vital. To begin


with, consider
C
(P,Q)
C
(S,T)
The multiplication condition shows that this is a linear combination of basis
vectors C
(S

,T)
plus lower order terms, and the involution of that condition
shows that C
(P,Q)
C
(S,T)
must also be in the span of C
(P,Q

)
plus lower order
terms. By the linear independence of the basis vectors we must have
C
(P,Q)
C
(S,T)
= (Q, S)C
(P,T)
+ lower order terms
for some uniquely determined coecient . Furthermore, by the two condi-
tions above, is independent of P and T, and depends only on Q and S, so we
are justied in writing (Q, S)
Now we can dene our bilinear form f on V

, spanned by the elements C


S
,
by setting f(C
S
, C
T
) = (S, T) and extending.
16
Lemma 3.2. f(aC
S
, C
T
) = f(C
S
, a

C
T
)
Proof. The proof is just manipulation using the identities we already have:
f(aC
S
, C
T
)C
(P,Q)
=

R
r
a
(R, S)f(C
R
, C
T
)C
(P,Q)
=

R
r
a
(R, S)C
(P,R)
C
(T,Q)
=
_
aC
(S,P)
_

C
(T,Q)
= C
(P,S)
(a

C
(T,Q)
)
= C
(P,S)

R
r
a
(R, T)C
(R,Q)
=

R
r
a
(R, T)f(C
S
, C
R
)C
(P,Q)
= f(C
S
, a

C
T
)C
(P,Q)
Now dene rad() = x V

: f(x, y) = 0 for all y v

. By our lemma,
this is an A-submodule of V

. We dene W

= V

/ rad(f).
We can now state the main theorem of this section, which explains why we
are interested in cellular algebras.
Theorem 3.3 (Graham, Lehrer). The module W

is non-zero if and only if


its bilinear form f is non-zero, and W

: f ,= 0 is a complete set of pairwise


non-isomorphic irreducible representations of A.
Example. We test this on our easy example A = F[x]/(x
n
). For any i the
cell module at i is the quotient x
i
A/x
i+1
A. We calculate that f = 0 for i > 0,
so the theorem states that W
0
- the representation of A on A/(x)

= F sending
x to the zero map - is the unique irreducible representation of A. This is easily
seen to be true, since ker x is a submodule of any non-trivial representation.
A non-example. The restriction that be nite is essential, as we see from
the example of F[x]. If the theorem held then the argument above would apply
and F[x] would have only one trivial representation; however, if we let x act
as any irrational rotation on R
2
then we have an irreducible two-dimensional
represenation.
Proof of Theorem 3.3 The rst part is apparent from the denition of W

. To
prove the rest, we will need:
Lemma 3.4. If < and a is in A() and S,T are in M() then r
a
(S, T) =
0. In particular, aV

= 0.
Proof. By the condition for multiplication, aC

S,T
is in A( ), and its compo-
nent in A() is determined by the coecients r
a
(S, T). But by the involution
of the multiplication condition, aC

S,T
also lies in A( ). The result follows.
Lemma 3.5. Suppose f is not zero, and let y, z be elements of W

such that
f(y, z) ,= 0. Then A()z = W

.
17
Proof. Without loss of generality we may take f(y, z) = 1. Take x W

, and
write
x =

SM()

S
C
S
y =

TM()

T
C
T
We claim that
_
_

S,T

T
C
S,T
_
_
z = f(y, z)x
Since the right hand side is x, an arbitrary element of W

, this identity
proves the lemma. First, note that by the linearity of everything involved it
suces to prove
C
S,T
C
U
= f(C
T
, C
U
)C
S
for any S, T, U in M(). But this follows from the denitions of f and of the
A-action on V

.
Corollary 3.6. The A-module W

is irreducible.
Corollary 3.7. Let be a non-zero map from W

to W

. Then
Proof. Suppose < , and let a A(). Then (az) = a(z) = 0 for any z
in W

, by Lemma 3.4. By Lemma 3.5, the whole of of W

is of this form for


some z, so the image of is zero.
Corollary 3.8. If there is an isomorphism W to W then = .
Thus we have proved that the A-modules W

: f ,= 0 are pairwise non-


isomorphic. All that remains is to prove that every irreducible A-module is of
this form.
Lemma 3.9. Every irreducible composition factor of A is equal to W

for some

Proof. Our algebra A is the direct sum of subspaces indexed by . It is possible


to put a total order on which agrees with the partial order and thus obtain a
ltration with composition factors equal to the cell modules. Furthermore, if B
is an irreducible composition factor of a cell module V

then either B = W

or
B is a factor of radV

. Therefore it suces to prove that every factor of radV

is of the required form.


Write A

= A( ,< ). It is easy to show from the denition of f that


radV

= 0, so radV

is a composition factor of A/A

. But A

was dened
so that A/A

has a composition series with factors W

such that < . The


result then follows by induction on .
18
Remark. The Kazhdan-Lusztig basis (see [7]) is a basis for the Hecke algebra
found in 1979, indexed by a partially-ordered set and containing information
about the representation theory of H. In [3] Graham and Lehrer mention this as
a starting point, which is why the denition given here talks explicitly of basis
vectors. A more algebraic denition of cellularity is given by Konig and Xi in
[6] in terms of certain ltrations, whose composition factors correspond to our
cell modules.
19
4 Hecke Algebras
4.1 Denitions
In this section we construct and examine the Hecke algebra H
n
(q), a gener-
alisation of the group algebra CS
n
. We are able to give a complete description
of the irreducible modules in this case also, this time by showing that H
n
(q)
is a cellular algebra as dened in the previous chapter, and applying the re-
sults developed there. The Hecke algebras arise in many dierent contexts, so
this theory is widely applicable, but in this essay we reapply our results to the
simpler case of the symmetric groups in order to complete our study of their
representations.
This chapter gives considerably less detail, omitting some proofs entirely. A
full treatment can be found in [2], but here we only prove a few partial results
for illustration and leave the rest as an outline.
We begin by noticing that if we let
i
be the permutation (i, i + 1) then
the symmetric group S
n
is generated by the elements
1
, ...,
n1
satisfying the
following relations:

2
i
= 1

i

j
=
j

i
for [i j[ > 1

i

i+1

i
=
i+1

i+1
In fact, these relations give a presentation for S
n
(although we do not prove
this). In particular, any w S
n
can be written as the product of transpositions
in many dierent ways, but the corresponding products of the elements are
equal modulo the given relations.
Motivated by the above, let q be a non-zero element of C and dene the
Hecke algebra H
n
(q) to be the unital C-algebra generated by the n1 elements
T
i
satisfying the relations:
(T
i
q)(T
i
+ 1) = 0
T
i
T
j
= T
j
T
i
for [i j[ > 1
T
i
T
i+1
T
i
= T
i+1
T
i
T
i+1
If q = 1 then this is as above, so H
n
(1)

= CS
n
. We take this similarity
further by dening an element T
w
for each w S
n
by
T
w
= T
s
1
T
s
2
...T
s
k
where
s
1

s
2
...
s
k
is a reduced expression for w.
Remark on notation Groups generated by involutions
i
and relations of
the form (
i

j
)
k
ij
are known as Coxeter groups, and can be classied into
innite families A
n
, B
n
, C
n
, D
n
together with a handful of exceptional cases. In
this paper we are only concerned with the symmetric groups which form the
rst and simplest family A
n
, so the Hecke algebras we construct are known as
Hecke algebras of type A. We suppress all of this, together with the index
n and the parameter q, and just refer to our algebra as H.
With some straightforward but long-winded manipulation, it is possible to
prove
20
Theorem 4.1. The element T
w
is well-dened (does not depend on the choice
of reduced expression for w) and the set T
w
: w S
n
is a basis for H. In
particular, dimH = n!.
Proof. Omitted
While easy to dene, this basis is not cellular: the action of H by multipli-
cation on any basis vector T
w
is transitive, so the multiplication condition for
cellular bases cannot be satised. In order to dene a suitable new set of basis
elements, we will need some more notation.
Recall that for a partition we write Std() for the set of standard tableaux
and S

for the Young subgroup. We set


m

For any tableau s we let d(s) S


n
be the permutation taking the regular
tableau to s (that is, d(s)t

= s) and we dene an operation



by T

w
= T
w
1.
Now we set
m
s,t
= T

d(s)
m

T
d(t)
For any standard tableaux s and t. Now we are ready to dene our new
basis:
M = m
s,t
: s, t Std() for some
We call M the Murphy basis of H, and we claim that it is cellular.
4.2 An Outline Proof
In this section we outline a proof that M is a cellular basis. As expected, the
set will be partitions of n and M() will be standard tableaux on .
Remark The only surprising detail is that we need to order in reverse
lexicographic order. We therefore have to choose between introducing a new
order with > , or reversing the order in the denition of a cellular
algebra. We choose the latter to avoid confusing the notation. Thus we talk
about higher order terms instead of lower order terms, but the theory is
unchanged.
Theorem 4.2. The set M is a cellular basis. Specically, is the set of par-
titions of n, and for each partition there is a nite set Std() of standard
tableaux. The map C : Std() Std() H sends the pair (s, t) to m
s,t
.
Finally, the involution

sending T
w
to T
w
1 can be extended to an antiauto-
morphism on H which sends m
s,t
to m
t,s
as required.
First we have to show that M spans H. Let be the partition (1,1,...,1), so
the Young subgroup of is the identity, and recall that t

is the regular tableau


on . If w is in S
n
and if s is the tableau wt

then
m
t

,s
= T

d(t

)
m

T
d(s)
= 1 1 T
w
21
Thus it suces to show that M spans m
t

,s
: s a tableau on . Rather
than proving this directly we proceed by induction, showing that for each par-
tition the set
m
s,t
: s, t row-standard tableaux on
is spanned by M. Our goal is to prove this for = (1, 1, ..., 1), the minimal
partition under our order, and our base case is = (n) for which the result is
trivial: the unique row-standard partition on this one-row diagram is necessarily
standard, so m
s,t
is in M itself. The induction depends on the following lemma:
Lemma 4.3. Let s, t be row-standard tableaux on some partition . Then
m
s,t
=

uStd()
ut
r
u
m
s,u
+h
where h is a linear combination of terms from
m
v,w
: v,w row-standard tableaux on some >
Unfortunately the proof is long and involved, and not very enlightening, so
we omit it. Details can be found in [2], where the result is rst proved for a
special type of only just non-standard tableaux and then extended to the full
result.
If we assume this lemma then by our induction hypothesis, h will be in
the span of M. Thus m
s,t
can be written as a sum of terms m
s,u
having a
strictly standard second index u. If we then apply the involution

, swapping
the indicies, and reapply the lemma, we can write m
s,t
in the span of M and
the induction holds.
If M spans H then a count proves that it is a basis, since by the combinatorial
identity in Corollary 2.12
[M[ =

[ Std()[
2
= n! = dimH
It remains to prove the multiplication condition
am
s,t
=

u
r
a
(u, s)m
u,t
+ higher order terms
for all a in H. The Young subgroup of S
n
has a set of coset representatives
u : t

u row standard, so we can write


Hm

T
d(u)
m

: t

u row standard
_
Thus we can write
am
s,t
=
_

u row standard
r
u
T
d(u)
m

_
T
d(t)
=

u
r
u
m
u,t
for some coecients r
u
which are independent of t. But by our lemma above,
this sum can be written in the required form.
22
4.3 Applications to S
n
In chapter 1 our representation theory of S
n
was restricted to elds with charac-
teristic p > n. Our theory in this section holds over any eld F of characteristic
p, so we can remove this restriction.
When q = 1 the cell modules of H are exactly the S
n
-modules V

constructed
in section 1. In this context they are known as Specht modules. They are not
necessarily irreducible, but by the above results every irreducible FS
n
-module is
a quotient of some Specht module. Furthermore, it is possible to prove exactly
when the bilinear form f is non-zero and thus prove which Specht modules we
need:
Theorem 4.4. We say a partition is p-restricted if
i

i+1
< p for all i.
Over a eld of characteristic p, a Specht module V

has a non-zero irreducible


quotient if and only if is p-restricted.
A proof can be found in [2]. Another important question is which Specht
modules remain irreducible when restricted from Q to F
p
; the answer has only
recently been proved ([9], [10]).
References
[1] W Fulton, J Harris Representation Theory: A First Course. Springer 1991
[2] A Mathas Iwahori-Hecke Algebras and Schur Algebras of the Symmetric
Group. AMS 1999
[3] J J Graham, G I Lehrer Cellular Algebras. Inventiones Mathematicae 1996
[4] D E Littlewood The Theory of Group Characters and Matrix Representa-
tions of Groups. AMS 1940
[5] P Etingof, O Golberg, S Hensel, T Liu, A Schwendner, E Udovina, D Vain-
trob Introduction to Representation Theory. arXiv:0901.0827v3 [math.RT]
2009
[6] S Konig, C C Xi On the Structure of Cellular Algebras. Algebras and Modules
II, CMS Conference Proceedings 1996
[7] D Kazhdan, G Lusztig Representations of Coxeter Groups and Hecke Alge-
bras. Inventiones Mathematicae 1979
[8] C Sagan The Ubiquitous Young Tableau. IMA Volumes in Math. and its
Applications Vol 19, Springer-Verlag 1990
[9] M Fayers Reducible Specht Modules. J. Algebra 280 2004
[10] M Fayers Irreducible Specht Modules for Hecke Algebras of Type A. Adv.
Math. 2005
23

You might also like