REPRESENTATION THEORY OF THE SYMMETRIC GROUP

MARK WILDON

Recommended reading: [7] G. D. James, Representation theory of the

symmetric groups, Springer Lecture Notes in Mathematics 692, Springer
(1980).
Acknowledgement. I thank Darij Grinberg for corrections to numer-
ous errors in the original version of these notes, and also for supplying
many useful clarifications and suggestions for extra content. I thank
Bakhshinder Samra for further corrections. Of course I take full respon-
sibility for all the remaining errors.

1. B ACKGROUND DEFINITIONS
Let Sn denote the symmetric group of degree n. Let N denote the set
of natural numbers {1, 2, 3, . . .} and let N0 = N ∪ {0}.
In these notes all maps will be written on the right. For example,
the composition of the cycles (12) and (123) in the symmetric group S3
is (12)(123) = (13), and the image of 1 under the permutation (12) is
1(12) = 2. This convention agrees with James’ lecture notes, and means
that compositions of maps can be read from left to right.
A LGEBRAIC DEFINITIONS . Let G be a finite group and let R be a com-
mutative ring. The group ring RG is defined to be the free R-module
with basis { g : g ∈ G } and multiplication defined by linear extension
of the multiplication in G. So given ∑h∈G αh h ∈ RG and ∑k∈G β k k ∈ RG
we set
    
∑ αh h ∑ β k k = ∑ αh β k hk = ∑ ∑ αh β h−1 g g.
h∈ H k∈G h∈ H g∈ G h∈ H
k∈K

An RG-module is an abelian group V together with a bilinear map

V × RG → V, which we shall write as (v, r ) 7→ vr, such that
(i) if r, s ∈ RG then v(rs) = (vr )s for all v ∈ V;
(ii) v idG = v for all v ∈ V.
If V is an RG-module then a subgroup W of V is an RG-module if and
only if the map W × RG → V has image inside W. In this case we say
that W is an RG-submodule of V.

Exercise: By looking at the action of α idG for α ∈ R, show that V has

the structure of an R-module.
Date: First term 2011/12 (revised October 2013).
2

Most of the time R will be a field. In this case the exercise above says
that V is an F-vector space. We shall also sometimes use ZG-modules:
these will always be free as Z-modules. All modules in these notes will
have finite rank / dimension.
Exercise:
(a) Let F be a field and let V be an FG-module. Show that if gρ :
V → V is the map defined by v( gρ) = vg then ρ is a group homomor-
phism from G into the group GL(V ) of invertible F-linear transforma-
tions of V.
(b) Show conversely that if V is an F-vector space and ρ : G → GL(V )
is a group homomorphism (so V is a representation of G) then setting
v(αg) = αv( gρ) for α ∈ F and g ∈ G, and extending linearly to a general
element of FG, gives V the structure of an FG-module.

In practice, it is usually most convenient to specify an FG-module V

by defining V as an F-vector space, and then defining an action of G on
V by invertible F-linear transformations. (There is then a unique way
to extend the actions of F and G on V so that the map V × FG → V is
linear in its second component, and so bilinear.)
For example, the trivial FG-module is the F-vector space F on which
each g ∈ G acts as the identity map.

Remark 1.1. The definition of an FG-module is more technical than the

definition of a representation of G, but, as the exercise shows, the two
notions are equivalent. Module can be more convenient to work with,
because there is less notation, and we can use results from ring theory
without any translation. The language of representations is preferable
if we want to have an explicit map ρ : G → GL(V ).
A similar situation arises when we have a group G acting on a set Ω.
Here one can choose between the equivalent languages of G-actions
(writing ωg for the image of ω ∈ Ω under g ∈ G) and permutation rep-
resentations (using a homomorphism from G into the symmetric group
on Ω). Again, both have their advantages.

The easiest example of a non-trivial FSn -module is given in the next

example.

Example 1.2. Let F be a field and let V = he1 , . . . , en i F be an n-dimensional

vector space over F. We make V into an FSn -module by defining
(1) ei g = eig for i ∈ {1, . . . , n} and g ∈ Sn .
By the remark after the exercise above, this suffices to give V the struc-
ture of an FSn -module. The corresponding representation ρ : Sn →
 3

GL(V ) represents each g ∈ Sn as a permutation matrix. For example if

n = 3, g = (12) and h = (123) then
0 1 0 0 1 0
   

gρ = 1 0 0 and hρ = 0 0 1 .
0 0 1 1 0 0
Note that these matrices act on the right, on row vectors.

C OMBINATORIAL DEFINITIONS . A partition of n ∈ N0 is a non-increasing

sequence of natural numbers whose sum is n. For example, the 7 parti-
tions of 5 are (5), (4, 1), (3, 2), (3, 1, 1), (2, 2, 1), (2, 1, 1, 1) and (1, 1, 1, 1, 1).
The numbers making up a partition are its parts. We shall use powers
to indicate multiplicities of parts: for example (2, 1, 1, 1) = (2, 13 ).
It is often useful to represent a partition by a rectangular array of
boxes, called its Young diagram1. For example, the Young diagram of
(4, 2, 1) is

Remark 1.3. It is well-known that the partitions of n label the conju-

gacy classes of Sn via cycle types. For example, the conjugacy class of
S5 labelled by (2, 2, 1) contains (12)(34). We shall see later that the ir-
reducible CSn -modules are also canonically labelled by the partitions
of n.

Let λ = (λ1 , . . . , λk ) be a partition of n ∈ N. A λ-tableau is an assign-

ment of the numbers {1, . . . , n} to the boxes of the Young diagram of λ,
so that each number is used exactly once. For example,
5 6 3 2
7 1
4
is a (4, 2, 1)-tableau. More formally, if we fix a numbering of the boxes
of λ, then we may regard a λ-tableau t as a function
t : {1, . . . , n} → {1, . . . , n}
such that it = j if and only if box i contains j.
1Young diagrams are named after the Reverend Alfred Young: see
www-history.mcs.st-andrews.ac.uk/Biographies/Young_Alfred.html. In a
letter sent to Frobenius, Young wrote:
I am delighted to find someone else really interested in the matter.
The worst of modern mathematics is that it is now so extensive
that one finds there is only about one person in the universe really
interested in what you are . . .
4

If s and t are λ-tableaux then s and t are said to be row-equivalent if

for each i ∈ {1, . . . , k }, the sets of numbers in the ith row of s and t are
equal. For example,
5 6 3 2 2 3 5 6
7 1 and 1 7
4 4
are row-equivalent.
A λ-tabloid is a row-equivalence class of λ-tableaux. We denote the λ-
tabloid corresponding to the λ-tableau t by {t}. Tabloids are drawn by
omitting the vertical lines from a representative tableau. For example if
t is the first tableau above then
5 6 3 2 2 3 5 6
{t} = 7 1 = 1 7 = ....
4 4

A λ-tableau is said to be row standard, if its rows are increasing when

read from left to right, column standard, if its columns are increasing
when read from top to bottom, and standard, if it is both row standard
and column standard.

Example 1.4. There are 6! = 720 distinct (4, 2)-tableaux, 15 distinct

(4, 2)-tabloids and 9 distinct standard (4, 2)-tableaux.
A (4, 2)-tabloid is determined by its second row, so the (4, 2)-tabloids
are in bijection with 2-subsets of {1, 2, 3, 4, 5, 6}. For example, the 2-
subset {2, 4} corresponds to the tabloid
1 3 5 6.
2 4
The 9 standard (4, 2)-tableaux,
1 3 5 6 , 1 3 4 6 , 1 3 4 5 , 1 2 5 6 , ...
2 4 2 5 2 6 3 4
will turn out to index a basis of an irreducible 9-dimensional CS6 -module.

2. Y OUNG PERMUTATION MODULES AND S PECHT MODULES

Throughout this section let n ∈ N and let R be a commutative ring.
Let λ be a partition of n. The symmetric group Sn acts on λ-tableaux
in an obvious way. For example, if g = (235) ∈ S5 then
1 2 5 6 g = 1 3 2 6
3 4 5 4
since 1g = 1, 2g = 3, 5g = 2, 6g = 6, and so on. With the definition
of tableau as function, if t : {1, . . . , n} → {1, . . . , n} is a λ-tableau and
 5

i ∈ {1, . . . , n} is the number of a box, then i (tg) = (it) g. Either way, it

is clear that the action of Sn on λ-tableaux is regular, i.e. the stabiliser of
any given tableau is the trivial group.

Lemma 2.1. Let λ be a partition of n. There is a well-defined action of Sn

on the set of λ-tabloids defined by {t} g = {tg} where {t} is a λ-tableau and
g ∈ Sn .

Definition 2.2. Let λ be a partition of n. The Young permutation module

MR λ , defined over R, is the RS -permutation module with free R-basis
n
{t} : {t} a λ-tabloid .


Usually we will write Mλ rather than MR λ since the ground ring (or,

more usually, field) will be clear from the context.

Example 2.3. We continue to work with RSn -modules.

(1) If λ = (n) then there is a unique λ-tabloid, namely 1 2 · · · n
and so M(n) ∼
= R, the trivial RSn -module.
(2) If λ = (n − 1, 1) then there are n distinct (n − 1, 1)-tabloids, with
representative tableaux

2 3 ... n
t1 =
1
1 3 ... n
t2 =
2
..
.
1 2 ... n−1
tn =
n

The top row of a two-row tabloid can be deduced from the sec-
ond row so we may write

{ t1 } = 1 , { t2 } = 2 , ..., {tn } = n .

It is clear that {ti } g = {tig } for each g ∈ Sn . Thus M(n−1,1)

affords the natural permutation representation of Sn .
(3) More generally, let 1 ≤ k < n and let λ = (n − k, k). Ignor-
ing the first row of a (n − k, k)-tabloid gives a bijection between
λ-tabloids and k-subsets of {1, . . . , n}. Hence M(n−k,k) is isomor-
phic to the RSn -permutation module of Sn acting on k-subsets of
{1, . . . , n}.
6

(4) Let 0 ≤ k < n and let λ = (n − k, 1k ). Ignoring the first row

of a (n − k, 1k )-tabloid gives a bijection between λ-tabloids and
k
k-tuples of distinct elements of Sn . Correspondingly M(n−k,1 )
is the permutation module of RSn acting on k-tuples of distinct
n
elements from {1, . . . , n}. Taking k = n − 1, it follows that M(1 )
is isomorphic to the group ring RSn as a right RSn -module.

Definition 2.4. Let λ be a partition of n and let t be a λ-tableau. The

column group C (t) consists of all permutations which preserve setwise
the columns of t. The signed column-sum bt is the element of the group
algebra FSn defined by
bt = ∑ h sgn(h).
h∈C (t)

The polytabloid (also called a λ-polytabloid, if the partition needs to be

emphasised) corresponding to t is defined by
e ( t ) = { t } bt .
Note that while the polytabloid e(t) is a linear combination of tabloids,
it depends on the tableau t, and not just on the tabloid {t}. For example
if λ = (2, 1) then

e 1 2 = 1 2 − 3 2 = 1 2 − 2 3
 
3 3 1 3 1

e 2 1 = 2 1 − 3 1 = 1 2 − 1 3
 
3 3 2 3 2
In general the polytabloid e(t) is a sum of |C (t)| distinct tabloids. In
particular, this shows that e(t) 6= 0, no matter what field we work over.
We observe that Sn permutes the set of all λ-polytabloids. If t is a
λ-tableau and g ∈ Sn , then
(2) e(t) g = {t}bt g = {t} gg−1 bt g = {tg}btg = e(tg).
where we have used that C (tg) = C (t) g , and so btg = g−1 bt g. (Note
that if h, g ∈ Sn then we set h g = g−1 hg.) One special case of (2) worth
noting is that
(3) e(t) g = sgn( g)e(t) if g ∈ C (t).

Definition 2.5. Let λ be a partition of n. The Specht module SλR , defined

over R, is the submodule of MR λ spanned by all the λ-polytabloids.

Again, we shall usually write Sλ rather than SλR . By (2) above and
the previous remark, Sλ is a well-defined non-zero RSn -module. It also
follows from (2) that Sλ is cyclic, generated by any single polytabloid.
 7

Example 2.6. Let n ∈ N and let F be a field. All modules in this example
are modules for FSn .
(A) Take λ = (n). Then while there are n! distinct (n)-tableaux, each
gives the same polytabloid, namely
e( 1 2 ... n ) = 1 2 ... n .

Hence S(n) = M(n) ∼

= F.
(B) Let n ≥ 2, let λ = (n − 1, 1) and let t1 , . . . , tn be the (n − 1, 1)-
tableaux in Example 2.3 above. With the same notation (i.e. omit-
ting the redundant top row in a tabloid), we have
e ( t1 ) = 1 − 2 , e ( t2 ) = 2 − 1 , . . . , e ( t n ) = n − 1 .
Therefore
S(n−1,1) = i − 1 :2≤i≤n .



It is easy to see that these vectors are linearly independent, so

S(n−1,1) is (n − 1)-dimensional. Problem 5 on the first problem
sheet shows that S(n−1,1) is irreducible if the characteristic of F
is zero or coprime to n.
(C) Take λ = (1n ) and let t be the (1n )-tableau whose single column
has entries 1, 2, . . . , n from top to bottom. Then C (t) = Sn , and
so by (3) above we have
e(tg) = e(t) g = e(t) sgn( g)
for all g ∈ Sn . Hence if s is any (1n )-tableau then e(s) = ±e(t)
n
and S(1 ) is isomorphic to the 1-dimensional sign representation
of Sn .

Example 2.7. Example 5.2 in James’ notes on S(3,2) is well worth study-
ing. He shows that S(3,2) is the subspace of M(3,2) given by imposing
‘all reasonable conditions’ on a general sum of tabloids. Omitting the
redundant top row in a (3, 2)-tabloid, the ‘reasonable conditions’ on

∑ αij i j ∈ M(3,2) ,
1≤ i < j ≤5

where αij ∈ F, are

(i) ∑1≤i< j≤5 αij = 0
(ii) ∑1≤ j≤5 αij = 0 for each i ∈ {1, . . . , 5}, where we set αkk = 0 and
αij = α ji if i > j.

Remark 2.8. There are at least three equivalent definitions of Specht

modules (four, counting the determinantal version of polytabloids be-
low):
8

(1) Definition 2.5 above, which defines Sλ as the subspace of Mλ

spanned by all λ-polytabloids: in §6 below we will prove the
Standard Basis Theorem which states that
{e(t) : t is a standard λ-tableau}
is a basis for Sλ .
(2) The description by generators and Garnir relations, which de-
fines Sλ as a quotient of FSn . (See §6 below.)
(3) As the subspace of Mλ given by imposing all reasonable condi-
tions on λ-tabloids: see Theorem 17.18 in James’ notes.

Remark 2.9. Some historical comments may helpful to motivate the

definition of Specht modules. First of all, it’s worth mentioning that
the irreducible characters of the symmetric groups were constructed by
Frobenius in 1904, long before Specht’s construction in 1935 of Specht
modules. (See W. Specht, Die irreduziblen Darstellungen der Symmetrischen
Gruppe, Math. Z. 39 (1935), no. 1, 696–711. Specht credits a 1907 paper
by his supervisor I. Schur for many of the ideas.) Frobenius’ charac-
ter formula is naturally expressed in terms of symmetric polynomials,
and Specht constructed his modules inside the polynomial ring in n
commuting variables. This would also have been a familiar object from
invariant theory.
To give a small example, let R = F [ x1 , x2 , x3 , x4 ]. By Specht’s defi-
nition S(3,1) is the submodule of R spanned linearly by the ‘Spechtian
polytabloids’
  x0 x0     
a b c

c =  1a d1   xb0   xc0  = xd − x a .
d x x a d
(In general c(t) is a product of Vandermonde determinants correspond-
ing to the columns of the tableau t.) The map sending xk to k gives an
isomorphism with the version of S(3,1) defined above. To give another
example,

 x0a xc0   x0 x0 
  
c a b =  1 1   b1 d1  = ( xc − x a )( xd − xb ),
c d x a xc xb xd

is a FS4 -generator of S(2,2) . Later we will see the Garnir relations which
can express a general polytabloid as a sum of polytabloids e(t) where t
is standard. In Specht’s setup, these express (specializations of) deter-
minantal identities that were probably well known to algebraists of his
time. For example, in S(2,2) we have, in James’ notation,

e 2 1
=e 1 2
−e 1 3
.
3 4 3 4 2 4
 9

In Specht’s language, this expresses the determinantal identity

1 11 1 1 11 1 1 11 1
        
 x2 x3   x1 x4  =  x1 x3   x2 x4  −  x1 x2   x3 x4  ,
        

which is a specialization of the Plücker relation

 α γ  A C   A γ  α C   α A γ C 
        
 β δ   B D =  B δ   β D +  β B   δ D .
        

One feature of Specht’s construction is that it makes it very easy to

prove that EndFSn Sλ = F, provided char F 6= 2. (This result also holds
in many cases when char F = 2, but is harder to prove). For example,
let φ : S(2,2) → S(2,2) be an endomorphism and let

x=c 1 2
.
3 4
We must have ( xφ)(13) = − xφ and xφ(24) = − xφ, so xφ is divisible,
in R, by x1 − x3 and x2 − x4 . Hence xφ = αx for some α ∈ F. Over
the complex numbers, a module is irreducible if and only if it has trivial
endomorphism ring, and so Specht’s construction leads to a quick proof
of the irreducibility of Specht modules.
Finally, it is worth noting that while Specht worked only over the
complex numbers, one of the nicest features of his construction is that
it gives modules that are defined in a uniform way over all rings.

3. J AMES ’ S UBMODULE T HEOREM

Definition 3.1. Let F be a field and let λ be a partition of n ∈ N. We
define a symmetric bilinear form h , i on MFλ by linear extension of
(
1 if {s} = {t}
h{s}, {t}i =
0 otherwise.

Therefore given x = ∑ αs {s} and y = ∑ β t {t} ∈ Mλ , where the sums

are over distinct tabloids, we have
h x, yi = ∑ αs β t h{s}, {t}i = ∑ αu β u .
s, t u

When F = C it is most natural to define the form so that it is Hermitian,

i.e. with the notation as above, h x, yi = ∑t αt β t .
If U ⊆ Mλ we shall write U ⊥ for the orthogonal space to U, i.e.
U ⊥ = {v ∈ Mλ : h x, vi = 0 for all x ∈ U }.
It is useful to note that the form h , i is Sn -invariant, i.e.
h xg, ygi = h x, yi
10

for all x, y ∈ Mλ and g ∈ Sn . (This follows easily from the case when
x and y are tabloids.) One consequence of this invariance is that if U ⊆
Mλ is an FSn -submodule, and v ∈ U ⊥ , g ∈ Sn then
vg, x = v, xg−1 = 0




for all x ∈ U. Hence U ⊥ is also an FSn -submodule of Mλ .

(2,2)
Example 3.2. Let F be a field of characteristic p. By definition S F is
spanned linearly by the polytabloids e(t) where t is a (2, 2)-tableau. But
by (3) just before Definition 2.5, e(t)h = sgn(h)e(t) for any h ∈ C (t), so
we need only take those t that are column standard. Hence if

x=e 1 3
, y=e 1 2
, z=e 1 2

2 4 3 4 4 3
then S(2,2) = h x, y, zi. By writing out x, y and z as linear combinations
of tabloids, one can check that z = − x + y, and that x and y are linearly
independent. (This relation is a special case of the relations mentioned
in Remark 2.9 above.) Therefore
S(2,2) = h x, yi F .
Consider the restriction of h , i to S(2,2) . The Gram matrix of this form
is, with respect to the basis x, y,
h x, x i h x, yi 4 2
   
G= = .
hy, x i hy, yi 2 4
If p = 2 we see that the restriction of h , i to S(2,2) is identically zero. If
p = 3 then
1 −1
 
G=
−1 1
and so
S(2,2) ∩ (S(2,2) )⊥ = h x + yi .
Calculation shows that, when p = 3,

x+y = 1 2 + 1 3 + 1 4 + 2 3 + 2 4 + 3 4
3 4 2 4 2 3 1 4 1 3 1 2
is the sum of all distinct (2, 2)-tabloids. Therefore h x + yi affords the
trivial representation of S4 .

Exercise:
(2,2)
(1) Let F be a field of characteristic 3. Show that S F / h x + yi affords
the sign representation of S4 , defined over F.
(2) Let V4 = h(12)(34), (13)(24)i ≤ S4 . Show that, over any field, the
kernel of the representation homomorphism S4 → S(2,2) is V4 , and that,
 11

thought of as a representation of S4 /V4 ∼

= S3 , there is an isomorphism
S ( 2,2 ) ∼ ( 2,1 ) .
=S

We now give the critical combinatorial lemma needed to prove James’

Submodule Theorem. James states this in more general form (see his
Lemma 4.6), but the version below suffices for this section.

Lemma 3.3. Let λ be a partition of n. If u and t are λ-tableaux then either

ubt = 0, or there exists h ∈ C (t) such that {uh} = {t} and ubt = ±e(t).

Proof. Suppose that ubt 6= 0. Let i and j be distinct numbers that appear
in the same row of u. If these numbers also appear in the same column
of t then (i, j) ∈ C (t) and so
ubt = u 1 − (i, j) ∑ g = 0


g

where the sum is over a set of right-coset representatives for the sub-
group h(i, j)i of C (t). Therefore:
Any two distinct numbers that appear in the same row of u
must lie in different columns of t.
Suppose that λ has k parts. Using the fact just proved, we may choose
h1 ∈ C (t) such that u and th1 have the same set of entries in their first
rows. Now choose h2 so that h2 fixes the first rows of u and th1 , and u
and th1 h2 have the same set of entries in their second rows. Repeat until
h1 , h2 , . . . , hk−1 have been chosen. Then u and th1 . . . hk−1 have the same
set of entries in all their rows. Hence if h = h1 . . . hk−1 then {u} = {th}
and
{u}bt = {t}hbt = ±{t}bt = ±e(t). 

Observe that Lemma 3.3 immediately implies that if λ is a partition

of n and t is a λ-tableau then
(4) Mλ bt = he(t)i.
We use this in the proof of the next theorem.

Theorem 3.4 (James’ Submodule Theorem). Let F be a field and let λ be a

partition of n ∈ N. If U is an FSn -submodule of Mλ then either U ⊇ Sλ or
U ⊆ (Sλ )⊥ .

Proof. Suppose that U is not contained in (Sλ )⊥ . Since Sλ is spanned

by the λ-polytabloids, there exists v ∈ U and a λ-tableau t such that
hv, e(t)i 6= 0. Using the Sn -invariance of h , i we have
hvbt , {t}i = hv, {t}bt i = hv, e(t)i 6= 0.
12

Using the remark after Lemma 3.3, that Mλ bt = he(t)i, it follows that
vbt = αe(t) for some non-zero α ∈ F. Hence e(t) ∈ U. We saw after Def-
inition 2.5 that Sλ is cyclic, generated by any single polytabloid. Hence
Sλ ⊆ U, as required. 

Corollary 3.5. Let F be a field of characteristic zero and let λ be a partition of

n ∈ N. Then SλF is irreducible.
Proof. Let U be a non-zero submodule of SλF . Since U is, in particular,
a submodule of MFλ , it follows from James’ Submodule Theorem that
either U ⊇ Sλ or U ⊆ Sλ ∩ (SλF )⊥ . Hence either U = Sλ , or U ⊆
Sλ ∩ (SλF )⊥ . It suffices to prove that Sλ ∩ (SλF )⊥ = 0.
If F = Q or F = R, this is clear, because then the bilinear form h , i is
an inner product, and so if V is any subspace of Mλ then V ∩ V ⊥ = 0.
If F = C the same argument holds, provided we define the form to be a
Hermitian inner product.
To deal with the general case (or to deal with F = C while using a
bilinear form rather than a Hermitian form), James argues as follows:
since the polytabloids span Sλ , we may choose, as in Example 3.2, a
basis of Sλ consisting of polytabloids. Let G be the Gram matrix of h , i
restricted to Sλ , with respect to this basis. The entries of G lie in Z, so
the rank of G does not depend on F.2 Therefore h , i is non-degenerate
when restricted to Sλ , and so Sλ ∩ (Sλ )⊥ = 0. 

4. H OMOMORPHISMS
Definition 4.1. Let n ∈ N and let λ and µ be partitions of n, with k and
` parts, respectively. We say that λ dominates µ, and write λ  µ, if
j j
∑ λi ≥ ∑ µi
i =1 i =1
for all j ≤ min(k, `).
Remarks:
(A) The relation  defines a partial order on the set of partitions of
n. It is not a total order. For example, (4, 1, 1) and (3, 3) are
incomparable.
2Generally if A is a d × d-matrix over a field F then there exist invertible d × d
matrices P and Q over F, and r ∈ N0 such that PAQ is equal to the matrix
Ir 0
 

0 0
where Ir is the r × r-identity matrix and 0 indicates a zero matrix of the appropri-
ate dimensions. It follows that if K is an extension field of F then the rank of A,
thought of either as a matrix over F, or as a matrix over K, is r.
 13

(B) Let n ∈ N and let F be a field. For each partition λ of n, let

J (λ) ∈ Matn ( F ) be a nilpotent matrix of Jordan type λ. For
instance, we may take J (λ) = diag( Bλ1 , . . . , Bλk ) where
0 1 0 ··· 0
 
0 0 1 · · · 0
Bd =  ... ... ... . . . ... 
 
.
 
0 0 0 · · · 1
 

0 0 0 · · · 0 d×d

Define N (λ) = J (λ)GLn ( F) . Then J (λ) is the set of all n × n nilpo-

tent matrices over F whose Jordan type is λ. Then
N (λ) ⊇ N (µ) ⇐⇒ λ  µ.
Here N (λ) means the closure of N (λ) in the Euclidean topology
if F = R or F = C, or in the Zariski topology otherwise. (The
two closures agree in the cases F = R and F = C.) For example,
the closure of
a b 0 0
    
2
N (2) = : a + bc = 0 \
c −a 0 0
contains the zero matrix, which is the unique element of N (1, 1).
(C) The dominance order appears in many other settings. The Gale–
Ryser Theorem is an important example in algebraic combina-
torics. To state this result we need some preliminary definitions.
The conjugate partition µ0 to a partition µ is defined by
µ0j = |{i : µi ≥ j}|.

The Young diagram of µ0 is obtained by reflecting the Young

diagram of µ. For example, (4, 2)0 = (2, 2, 1, 1), as seen below.

0
= .

If B is a k × `-matrix then the sequence of row-sums of B is

` `
∑ B1j , . . . , ∑ Bkj


j =1 j =1

and the sequence of column-sums of B is

k k
∑ Bi1 , . . . , ∑ Bi` .


i =1 i =1

We say that B is zero-one if each entry of B is either 0 or 1.

14

Theorem (Gale–Ryser). Let λ and µ be partitions of n. There is

a zero-one matrix with row-sums λ and column-sums µ0 if and
only λ  µ.
The necessity of the condition is fairly easily seen. Suppose
that A is a zero-one matrix with row-sums λ and column-sums
µ0 . There are λ1 1s in the first row of A so at least λ1 of the
column-sums are non-zero. Hence λ1 ≤ µ1 . There are λ1 + λ2
1s in the first two rows of A and there are µ1 − µ2 columns that
can contain at most a single 1, and µ2 that can contain two 1s.
Hence
λ1 + λ2 ≤ (µ1 − µ2 ) + 2µ2 = µ1 + µ2 .
Repeating this argument shows that λ  µ. See Problem Sheet 3
for a proof using representation theory that the condition is also
sufficient.
The next lemma is a more general version of Lemma 3.3.

Lemma 4.2. Let n ∈ N and let λ and µ be partitions of n where λ has k parts.
Suppose that u is a µ-tableau and t is a λ-tableau such that {u}bt 6= 0. Then
λ  µ and there exists h ∈ C (t) such that, for each i ∈ {1, . . . , k }, all the
numbers in the first i rows of u lie in the first i rows of th.
Proof. Suppose that there exists h ∈ C (t) with the claimed properties.
Then, for each i ∈ {1, . . . , k }, we see that there are µ1 + · · · + µi numbers
in the first i rows of u, and these must all appear as one of the λ1 + · · · +
λi numbers in the first i rows of th. Hence λ  µ.
To construct such an h ∈ C (t) we use the argument in Lemma 3.3,
that any two numbers in the same row of u must appear in different
columns of t.
Let h1 ∈ C (t) to be the permutation moving the numbers in the first
row of u into the first row of t. Suppose inductively we have hi−1 such
that all the numbers in the first i − 1 rows of u lie in the first i − 1 rows
of thi−1 . Consider the numbers in row i of u. These all lie in different
columns of thi−1 . Take gi ∈ C (thi−1 ) = C (t) to be the permutation
which moves these numbers as high as possible in thi−1 , while fixing
the entries of the first i − 1 rows of u. There is at most one entry in
each column of t that has to be moved up by gi , so the entries of row i
of u lie in rows 1, . . . , i in thi−1 gi . So we may take hi = hi−1 gi and
h = h1 . . . h k . 

We are now ready to prove the key result on homomorphisms.

Theorem 4.3. Let λ and µ be partitions of n. If there is a non-zero homo-

morphism θ : Sλ → Mµ of FSn -modules that extends to a homomorphism
θe : Mλ → Mµ then λ  µ.
 15

Proof. Since Sλ is spanned linearly by the polytabloids e(t) where t is a

λ-tableau, there exists a λ-tableau t such that e(t)θ 6= 0. Suppose that
{t}θe = ∑ α{u} { u }
{u}

where the sum is over all µ-tabloids {u}. Then, since e(t) = {t}bt and
e t 6= 0, there exists a µ-tabloid {u} such that {u}bt 6= 0. Now apply
{t}θb
Lemma 4.2. 
The technical condition in Theorem 4.3 on extensions of homomor-
phisms may fail to hold when F has characteristic two. The smallest
example is the unique homomorphism

= S(1,1) → M(2)
F∼
which fails to extend to a homomorphism M(1,1) → M(2) . (The unique
such homomorphism has S(1,1) ∼ = F in its kernel.) When F has charac-
teristic zero, there are no such problems.

Corollary 4.4. Let F be a field of characteristic zero. If SλF ∼

= SF then λ = µ.
µ

Proof. Since F has characteristic zero, the FSn -modules are completely
reducible (by Maschke’s Theorem). Hence MFλ = SλF ⊕ C for some com-
plementary FSn -module C. If SλF ∼ = SF then there is an injective FSn -
µ

module homomorphism θ : SλF → MF . We extend θ to an FSn -module

µ

homomorphism θe : MFλ → MF by setting vθe = 0 for v ∈ C. By Theo-

µ

rem 4.3 we have λ  µ. By symmetry µ  λ, so λ = µ. 

The following example gives a phenomenon related to the failure of

the extension condition in Theorem 4.3 when F has characteristic two.

Example 4.5. Over F2 there is an isomorphism

S(5,1,1) ∼
= S(5,2) ⊕ S(7) .
Hence there is a non-zero homomorphism from S(5,1,1) to S(5,2) . See
Question 1 on Sheet 2 for a proof of this decomposition.

If G is any finite group then the number of complex irreducible repre-

sentations of G is equal to the number of conjugacy classes of G. There
are |Par(n)| conjugacy classes in Sn , so the Specht modules SC λ form a

complete set of non-isomorphic irreducible CSn -modules, as λ varies

over the partitions of n. (In fact this holds over any field of characteris-
tic zero.)
Over C any module is completely reducible. So Theorem 4.3 implies
that MC is a direct sum of Specht modules labelled by partitions λ such
µ
16

that λ  µ. For example, it follows from Questions 2 and 3 on the second

problem sheet that if n ∈ N then
(n−r,r ) ∼ (n−r,r ) (n−r +1,r −1) (n−1,1) (n)
MC = SC ⊕ SC ⊕ · · · ⊕ SC ⊕ SC
(1n )
for any r such that 2r ≤ n. To give another example, since MC is
isomorphic to the regular representation of CSn we have
(1n ) ∼
MC (dim Sλ )SC .
M
λ
=
λ`n

5. C ONSTRUCTION OF SIMPLE MODULES IN CHARACTERISTIC p

Overview: Let F be a field. By James’ Submodule Theorem, if U is a
proper FSn -submodule of Sλ then U ⊆ Sλ ∩ (Sλ )⊥ . Hence if
⊥
Sλ /(Sλ ∩ (Sλ ) )
is non-zero then it is a simple FSn -module. When F has characteristic
0 we saw in Corollary 3.5 that Sλ ∩ (Sλ )⊥ = 0 for all partitions λ. To
describe the situation in prime characteristic p we need the following
definition.
Definition 5.1. Let λ be a partition of n with exactly ai parts of length i
for each i ∈ {1, 2, . . . , n}. Given p ∈ N, we say that λ is p-regular if
ai < p for all i.
In this section and the next we shall prove the following theorem,
mainly following James’ proof in §10 and §11 of [7].
Theorem. Let F be a field of prime characteristic p. If λ is a partition
of n then Sλ 6⊆ (Sλ )⊥ if and only if λ is p-regular, and
n Sλ o
: λ a p-regular partition of n
Sλ ∩ (Sλ )⊥
is a complete set of non-isomorphic simple FSn -modules.

Remark: It is reasonable to ask for a heuristic reason for this theorem.

• If G is any finite group then the number of isomorphism classes
of irreducible CG-modules is equal to the number of conjugacy
classes of G. A deeper result of Brauer states that if F is a split-
ting field for G (i.e. every FG module is absolutely irreducible, in
the sense defined at the end of this section), then the number of
isomorphism classes of irreducible FG-modules is equal to the
number of conjugacy classes of G of elements of order coprime
to p. By Proposition 5.11 below, the number of p-regular parti-
tions is equal to the number of irreducible FG-modules. Thus
 17

the set of modules specified in the theorem above does at least

have the right size (assuming they are all non-zero).
• One consequence of the theorem is that every simple FSn -module
appears as the unique top composition of some Specht module.
A similar (but nicer) situation occurs for modules for a quasi-
hereditary algebra: If A is a quasi-hereditary algebra then there
is a canonical set of standard modules for A such that any pro-
jective A-module is filtered by standard modules, and the top
composition factors of the standard modules form a complete
set of non-isomorphic simple A-modules. See [3] for an intro-
duction to this area.
• An important example of quasi-hereditary algebras are the Schur
algebras. The module category of S(n, r ), defined over an infi-
nite field F, is equivalent the category of polynomial modules
of degree r for the general linear group GLn ( F ). (See [6] for an
introduction.) For any n ∈ N there is a Schur algebra A and
an idempotent e ∈ A such that FSn = eAe. This means that
the symmetric group algebra FSn inherits some of the desirable
properties of quasi-hereditary algebras.
• Cellular algebras generalize quasi-hereditary algebras in a way
modelled on symmetric group algebras. For an introduction,
see [9]. An analogous version of the theorem above holds for
the simple modules of any cellular algebra.

Example 5.2. Let F have prime characteristic p.

(2,2)
(1) In Example 3.2 we saw that the restriction of h , i to S F is zero
if and only if F has characteristic 2. Correspondingly, (2, 2) is p-regular
if and only if p > 2.
(2) Let λ = (2, 1, 1). Let
2 1 1 2 1 3 1 4
t1 = 3 , t2 = 3 , t3 = 2 , t4 = 2
4 4 4 3
be the four column standard (2, 1, 1)-tableaux. Reasoning as in Exam-
ple 3.2 we see that if F is any field then
(2,1,1)
SF = he(t1 ), e(t2 ), e(t3 ), e(t4 )i .
These polytabloids are not linearly independent: one can check that
e ( t1 ) = e ( t2 ) − e ( t3 ) + e ( t4 ).
However e(t2 ), e(t3 ), e(t4 ) are linearly independent, because each in-
volves a tabloid not appearing in the other two. For instance, since 1
18

and 2 appear in the first column of both t3 and t4 , the polytabloid e(t2 )
is the unique polytabloid having
1 2
3
4
as a summand. Each polytabloid e(ti ) has 6 summands, so he(ti ), e(ti )i =
6 for each i. To calculate he(t2 ), e(t3 )i we argue that any common tabloid
must have 2 and 3 in its first row. Calculation shows that
3 2 3 2
e ( t2 ) = − 1 + 4 + ··· ,
4 1
2 3 2 3
e ( t3 ) = − 1 + 4 + ···
4 1
so he(t2 ), e(t3 )i = 2. Similar calculations show that he(t3 ), e(t4 )i = 2
(2,1,1)
and he(t2 ), e(t4 )i = −2. Hence the matrix for h , i on S F is
6 2 −2
 
 2 6 2 .
−2 2 6
It is clear this matrix is zero if and only if p = 2. Correspondingly
(2, 1, 1) is p-regular if and only if p > 2. In fact the determinant of the
matrix is 128, and so when p > 2 the matrix always has full rank. Hence
(2,1,1)
S(2,1,1) ∩ (S(2,1,1) )⊥ = 0 and S F is irreducible in these cases.
It is an instructive exercise to generalize Example 5.2(2) to the parti-
tions (2, 1n−2 ). (Question 6 on Sheet 1 is relevant.)

Remark on integral Specht modules: Work over F2 . We saw above that

the bilinear form h , i is identically zero. If we divide each entry in the
(2,1,1)
matrix by 2 we obtain a symmetric, S4 -invariant bilinear form on SF2
with matrix
1 1 1
 
1 1 1 .
1 1 1
The radical S(2,1,1) ∩ (S(2,1,1) )⊥ of this new form is
V = he(t1 ) − e(t2 ), e(t1 ) − e(t3 )i .
= S(3,1) /S(3,1) ∩
One can check that V is a simple F2 S4 -module: in fact V ∼
(S(3,1) )⊥ . There is also an isomorphism V ∼
= S(2,2) .
This idea is usually applied to integral Specht modules. Given an
integral Specht module SZλ and a prime p we can define a filtration:
SZλ = U0 ⊇ U1 ⊇ U2 ⊇ . . .
 19

where
Ui = {u ∈ SZλ : pi | hu, vi for all v ∈ SZλ }.
If we then reduce mod p, by quotienting out by pSZλ , we get a filtration
of the Specht module SλF . In this example we have

U1 = e(t1 ) − e(t2 ), e(t1 ) − e(t3 ), 2e(t1 ) Z



and
U1 ∼
(2,1,1)
=V
2SZ
where V is the simple F2 S4 -module defined above.
The Jantzen–Schaper formula gives some information about the sim-
ple modules that appear. See [1] for one account. In [4] Fayers used the
Jantzen–Schaper formula to determine all the Schaper layers for Specht
modules labelled by hook partitions (n − r, 1r ) and two-row partitions.

The following two lemmas will imply that Sλ 6⊆ (Sλ )⊥ if and only
if λ is p-regular. The lemmas make sense for Specht modules defined
over any field but are strongest for Specht modules defined over Z.

Lemma 5.3. Let λ be a partition with exactly a j parts of size j for each j ∈
{1, . . . , n}. If t and t0 are λ-tableaux then he(t), e(t0 )i is a multiple of ∏nj=1 a j !.

Proof. We shall say that two λ-tabloids {u} and {v} are Foulkes equiva-
lent if it is possible to obtain {v} from {u} by reordering the rows of u.
(This term is not standard.) Let
T = {u} : {u} is a summand in both e(t) and e(t0 ) .


If {u} ∈ T and {v} is Foulkes equivalent to {u} then there exist permu-
tations h ∈ C (t) and h0 ∈ C (t0 ) which permute the entries in the rows
of {u} as blocks for their action, such that
(i) {u} h = {u} h0 = {v},
(ii) sgn h = sgn h0 .
(Here (ii) holds because we can swap any two rows of length r in {u}
by a product of r disjoint transpositions in either C (t) or C (t0 ).) Hence
T is a union of Foulkes equivalence classes. Moreover, if {u} ∈ T then
the contribution from {u} to the inner product he(t), e(t0 )i, namely
h{u}, e(t)ih{u}, e(t0 )i,
depends only on the Foulkes class containing u. Since all classes have
size ∏nj=1 a j !, it follows that he(t), e(t0 )i is a multiple of ∏nj=1 a j !. 
20

Example 5.4. Let

1 2 3 1 4 7 1 5 9 1 6 8
t= 4 5 6 t0 = 2 5 8 u= 2 6 7 v= 2 4 9 .
7 8 9 3 6 9 3 4 8 3 5 7
One Foulkes class of tabloids common to both e(t) and e(t0 ) has repre-
sentative {u} since

{t}(47)(25)(39) = {u} = {t0 }(456)(798).

The Foulkes equivalence class of {u} consists of the six tabloids

 
 1 5 9 1 5 9 2 6 7 2 6 7 3 4 8 3 4 8 
2 6 7 , 3 4 8 , 1 5 9 , 3 4 8 , 1 5 9 , 2 6 7 .
3 4 8 2 6 7 3 4 8 1 5 9 2 6 7 1 5 9
 

For example, to swap the first two rows in {u} we may either apply
(17)(25)(69) ∈ C (t) or (12)(56)(79) ∈ C (t0 ). As claimed in the proof,
these permutations both have sign −1. Since {u} appears with sign −1
in e(t) and sign +1 in e(t0 ), and the contribution from the Foulkes class
of {u} to he(t), e(t0 )i is −6.
Since
{t}(285)(369) = {v} = {t0 }(465)(789)
there is another Foulkes equivalence class of tabloids common to both
e(t) and e(t0 ), namely
 
 1 6 8 1 6 8 2 4 9 2 4 9 3 5 7 3 5 7 
2 4 9 , 3 5 7 , 1 6 8 , 3 5 7 , 1 6 8 , 2 4 9 .
3 5 7 2 4 9 3 5 7 1 6 8 2 4 9 1 6 8
 

giving a contribution to he(t), e(t0 )i of +6.

If {w} is a tabloid appearing in both e(t) and e(t0 ) then the entries in
each row w appear in different columns of t, and in different columns
of t0 . Since the columns of t0 are the rows of t, an equivalent condi-
tion is that the rows of w form three disjoint transversals for the rows
and columns of t. There are six such transversals, namely {1, 5, 9},
{4, 8, 3}, {7, 2, 6}, {1, 8, 6}, {4, 2, 9} and {7, 5, 3}; these transversals ad-
mit a unique partition into two parts each containing three disjoint sets,
namely
n o
{1, 5, 9}, {4, 8, 3}, {7, 2, 6} , {1, 8, 6}, {4, 2, 9}, {7, 5, 3} .


The two parts correspond to the tabloids {u} and {v} already seen.
Hence he(t), e(t0 )i = −6 + 6 = 0.
 21

Lemma 5.5. Let λ be a partition of n having exactly a j parts of size j for each
j ∈ {1, . . . , n}. Let t be a λ-tableau and let t? be the λ-tableau obtained from
t by reversing each row. Then
n
he(t), e(t? )i = ∏ ( a j !) j .
j =1

Note that {t} = {t? }. The following example should clarify the rela-
tionship between e(t) and e(t? ), and will show some of the key ideas in
the proof of the lemma. (This example is a trivial variation on the one
given by James in his proof: see [7, Lemma 10.4].)

Example 5.6. Let λ = (4, 3, 3, 2) and let t and t? be as shown below.

1 2 3 4 4 3 2 1
t= 5 6 7 t? = 7 6 5
8 9 10 10 9 8
11 12 12 11

Suppose that {u} is a summand in both e(t) and e(t? ). So {u} =

{t}h = {t? }h? for some h ∈ C (t) and h? ∈ C (t? ). Working down the
first column of t we see that
(1) 1h? is in row 1 of t? h? . Hence 1h? = 1h = 1;
(2) 5h? and 8h? are in rows 1, 2 or 3 of t? h? , and in rows 2, 3 or 4 of
th. Hence {5, 8} h = {5, 8} h? = {5, 8};
(3) 11h is in row 4 of th. Hence 11h = 11h? = 11.
The diagram below shows the positions of the elements in the sets
{1}, {5, 8}, {11} that are known to be permuted by h and h? in the
tableaux t and t? .
1 2 3 4 4 3 2 1
5 6 7 7 6 5
8 9 10 10 9 8
11 12 12 11

Observe that if we remove the first column of t and the final entries in
the rows of t? we can repeat this argument, showing that h and h? also
permute the elements in the sets {2}, {6, 9}, {12}, and {3}, {7, 10}, and
finally {4}. It is now clear that h = h? .
Conversely since t = t? , we clearly have {t}k = {t? }k for all

k ∈ C (t) ∩ C (t? ) = h(5, 8), (6, 9), (7, 10)i.

Hence the tabloids that appear as summands in both e(t) and e(t? ) are
precisely the tabloids {t}k for k ∈ C (t) ∩ C (t? ) and so he(t), e(t? )i = 2!3 ,
as stated by Lemma 5.5.
22

We now prove Lemma 5.5. Suppose that the tabloid {th} = {t? h? } is
a summand of both
e(t) = ∑ {t}h sgn(h),
h∈C (t)

e(t? ) = ∑ {t? }h? sgn(h? ).

h? ∈C (t? )

Claim: h = h? and if i ∈ {1, . . . , n} and ih = ih? = j then i and j lie in

rows of equal length of λ.
Proof of claim. We shall prove the claim by induction on the number of
columns of the Young diagram of λ.
Let α1 , . . . , α` be the sequence of distinct part sizes in λ in decreasing
order of size. Suppose that there are mi parts of length αi . For each
i ∈ {1, . . . , `} let Xi be the set of entries in the first column of t that lie in
rows of length αi . Working down the first column of t (or equivalently,
the ends of the rows of t? ), we see that Xi h? = Xi and Xi h = Xi for each
i. Hence the restrictions of h and h? to X1 ∪ · · · ∪ X` agree. Moreover,
the restricted permutations permute entries within rows of equal length
as required by the claim.
Let s be the tableau obtained by removing the first column from t and
let s? be the tableau obtained by removing the final entry in each row of
t? . By induction the claim holds for the restrictions of h and h? to s and
s? . Hence the claim holds for h and h? in their action on {1, . . . , n}. 
It follows from the first statement in the claim that the set of tabloids
{u} such that {u} is a summand in both e(t) and e(t? ) is exactly
{t}k : k ∈ C (t) ∩ C (t? ) .


Hence
he(t), e(t? )i = ∑ h{t}k sgn(k), {t? }k sgn(k)i
k∈C (t)∩C (t? )
= |C (t) ∩ C (t? )|.
If k ∈ C (t) then k ∈ C (t? ) if and only if for each i ∈ {1, . . . , n}, the rows
of t (or equivalent, the rows of t? ) containing i and ik have the same
length. Therefore
n
C (t) ∩ C (t? ) = ∏ Hj
j =1
where Hj is the subgroup of C (t) ∩ that permutes the entries
C (t? )
within the rows of t of length j. Since λ has a j parts of length j, the
subgroup Hj has order ( a j !) j . Therefore
n
|C (t) ∩ C (t? )| = ∏ ( a j !) j .
j =1
 23

This completes the proof of Lemma 5.5.

Theorem 5.7. Let F be a field of prime characteristic p and let λ be a partition.
Then SλF 6⊆ (SλF )⊥ if and only if λ is p-regular.
Proof. As usual, we suppose that λ has exactly a j parts of length j for
each j ∈ {1, 2, . . . , n}.
By Lemma 5.3, if t and t0 are any λ-tableaux, then we have
n
e ( t ), e ( t 0 ) = α ∏ a j !


j =1

for some α ∈ F. By definition, SλF is the F-linear span of all λ-polytabloids,

so if λ is not p-regular then Sλ ⊆ (Sλ )⊥ .
Conversely, suppose that λ is p-regular. Let t be any λ-tableau and
let t? be the tableau obtained by reversing the rows of t. By Lemma 5.5
we have
n
he(t), e(t? )i = ∏ ( a j )! j 6= 0
j =1

Hence the restriction of h , i to Sλ is non-zero, and so Sλ 6⊆ (Sλ )⊥ . 

To show that the irreducible modules coming from the previous the-
orem are non-isomorphic we need the following result. In its proof, we
shall on two occasions use the corollary of Lemma 3.3, that if t is any
λ-tableau, then
Mλ bt = he(t)i .

Theorem 5.8. Let F be a field of prime characteristic p. Suppose that λ and

µ are partitions of n and that λ is p-regular. Let V be a submodule of Mµ .
If there exists a non-zero FSn -module homomorphism θ : Sλ → Mµ /V then
λ  µ. Moreover, if λ = µ and t is a λ-tableau, then e(t)θ ∈ he(t) + V i.
Proof. As usual, we suppose that λ has exactly a j parts of size j for each
j ∈ {1, 2, . . . , n}. Let t be a λ-tableau and let t? denote the tableau ob-
tained from t by reversing each of its rows. Since
e(t? )bt ∈ Mλ bt = he(t)i ,
we have e(t? )bt = αe(t) for some α ∈ F. Using that he(t), {t}i = 1 we
can determine α as follows:
n
α = he(t? )bt , {t}i = he(t? ), {t}bt i = he(t? ), e(t)i = ∏( a j !) j
j =1

where the first equality uses the Sn -invariance of h , i, and the final
step uses Lemma 5.5. In particular, α 6= 0. Since e(t) generates Sλ and θ
is non-zero we have e(t)θ 6= 0. Therefore
e(t? )θbt = e(t? )bt θ = αe(t)θ 6= 0.
24

Note that
e(t? )θbt ∈ ( Mµ /V )bt .
Hence Mµ bt 6= 0, and so there exists a µ-tabloid {u} such that {u}bt 6=
0. By Lemma 3.3 we get λ  µ. Moreover if λ = µ then we have Mµ bt =
he(t)i and so
e(t? )θbt ∈ he(t) + V i .
Since αe(t)θ = e(t? )θbt , it follows that e(t)θ ∈ he(t) + V i. 
Remark: it is worth comparing this theorem with Theorem 4.3. In The-
orem 5.8 we assume that λ is p-regular, whereas in Theorem 4.3 we
make no assumption on λ, but instead assume that the FSn -module
homomorphism θ : Sλ → Mµ extends to a non-zero homomorphism
θ̂ : Mλ → Mµ . In either case, the conclusion is the same, that λ  µ.

Corollary 5.9. Let F be a field of prime characteristic p and let λ and µ be

p-regular partitions of n ∈ N. If
Sλ ∼ Sµ
=
Sλ ∩ (Sλ )⊥ Sµ ∩ (Sµ )⊥
then λ = µ. Moreover,
 Sλ 
∼
EndFSn = F.
Sλ ∩ (Sλ )⊥
Proof. Let
Sλ Sµ
φ: →
Sλ ∩ (Sλ )⊥ Sµ ∩ (Sµ )⊥
be a non-zero FSn -module homomorphism. Since both modules in-
volved are simple, φ is an isomorphism. We may lift φ to a non-zero
map θ : Sλ → Mµ /V where V = Sµ ∩ (Sµ )⊥ by taking the composition
of the maps
Sλ φ Sµ
Sλ → → → Mµ /V.
Sλ ∩ (Sλ )⊥ Sµ ∩ (Sµ )⊥
Therefore, by Theorem 5.8, we have λ  µ. By symmetry we also have
µ  λ and so λ = µ. Hence, by Theorem 5.8 every λ-tableau t satisfies
e(t)θ = γ(e(t) + V )
for some γ ∈ F. Since e(t) generates Sλ /Sλ ∩ (Sλ )⊥ , this implies that
φ = γ id. 

Let F be a field of prime characteristic p. For λ a p-regular partition,

let D λ = Sλ /Sλ ∩ (Sλ )⊥ . By the remark using the submodule theorem
made at the start of this section, Theorem 5.7, and Corollary 5.9, the D λ
are non-isomorphic simple FSn -modules.
 25

It remains to show that any simple FSn -module is isomorphic to one

of the D λ . This can be proved using the first remark following the state-
ment of the main theorem at the start of this section, but it takes some
work to show that F is a splitting field for Sn . So instead we shall fol-
low a remark of James indicating an alternative approach using another
theorem of Brauer (taken from [2, 82.6]).

To state this theorem we need two definitions. Recall that an FG-

module U is said to be absolutely irreducible if U ⊗ F K is irreducible for
every extension field K : F. A conjugacy class is said to be p-regular if
its elements have order coprime to p.

Theorem (Brauer). Let F be a field and let G be a finite group. The

number of isomorphism classes of absolutely irreducible FG-modules
is at most the number of conjugacy classes of G of p-regular elements.

To prove that every absolutely irreducible FSn -module is isomorphic

to one of the D λ , it suffices to prove:
(a) each D λ is absolutely irreducible,
(b) the number of p-regular partitions of n is equal to the number of
partitions of n with no part divisible by p.

Part (a) follows from Corollary 5.9 and the next lemma.

Lemma 5.10. Let G be a finite group, let F be a field and let U be an irreducible
FG-module such that EndF (U ) ∼ = F. Then U is absolutely irreducible.3
Proof. Let K : F be an extension field of F. We must show that U ⊗ F
K is an irreducible KG-module. Since EndFG (U ) ∼ = F it follows from
Jacobson’s Density Theorem that the image of the action map FG →
EndF (U ) is all of EndF (U ). Hence the image of the map

KG → EndK (U ⊗ F K )

is also all of EndK (U ⊗ F K ). Therefore KG acts transitively on the non-

zero vectors in U ⊗ F K and so U ⊗ F K is an irreducible KG-module. 

Part (b) is proved in the next proposition.

Proposition 5.11. Let p ≥ 2. The number of p-regular partitions of n is equal

to the number of partitions of n with no part divisible by p.

3The converse result, that if U is absolutely irreducible then End (U ) ∼ F also

FG =
holds, but we do not need this here.
26

Proof. The generating function for p-regular partitions is

F(x) = ∏(1 + xi + x2i + · · · + x( p−1)i );

i ≥1

a partition with ai parts of size i corresponds to multiplying out the

product by choosing ia from the ith term. to a partition with a parts of
size i. Hence
1 − xip 1
F(x) = ∏ i
= ∏ 1 − xi
i ≥1 1 − x i ≥1
p6 | i

which is the generating function for partitions with no part divisible

by p. 

We note that this argument does not rule out the possibility that there
are further irreducible FSn -modules that are not absolutely irreducible.

6. S TANDARD B ASIS T HEOREM

Definition 6.1. If t is a standard tableau then we say that the corre-
sponding polytabloid e(t) is standard.

The object of this section is to prove the following theorem giving a

basis for SλF over any field F.

Theorem 6.2 (Standard Basis Theorem). Let λ be a partition of n. If F is a

field then the standard λ-polytabloids form a basis for Sλ .

Remarks:
(1) In Example 2.6(B) the basis we found of S(n−1,1) is in fact the
basis of standard (n − 1, 1)-polytabloids. In Example 3.2 we saw
that the Standard Basis Theorem holds for S(2,2) .

(2) An important consequence of the Standard Basis Theorem is

that if λ is a partition and F is a field then the dimension of SλF is
equal to the number of standard λ-tableaux, and so is indepen-
dent of the field F. The Hook-Formula (see [7, Chapter 20] or the
reference in (3) in the suggestions for further reading below) is a
remarkable combinatorial formula for this number.

(3) Another corollary of the Standard Basis Theorem is the isomor-

V (n−1,1) ∼ (n−r,1r )
phism r S F =S F over any field F: see Question 6 on
Problem Sheet 3.
 27

6.1. Linear independence. To show that the standard polytabloids are

linearly independent we need the following order on tabloids.

Definition 6.3. Let λ be a partition of n. Given distinct λ-tabloids {t}

and {s}, we define
the greatest number appearing in a different row
{t} > {s} ⇐⇒
in {t} than {s} lies in a lower row of {t} than {s}
It is straightforward to show that > is a total order on the set of all
λ-tabloids for any partition λ.
Working downwards from n down to 1, one sees if λ = (λ1 , . . . , λk )
then the greatest λ-tabloid has n, n − 1, . . . , n − λk + 1 in its bottom row,
n − λk , . . . , n − λk − λk−1 + 1 in its second from bottom row, and so on.
So the elements in its ith row are precisely
{λ1 + · · · + λi−1 + 1, . . . , λ1 + · · · + λi }.
For example, the greatest (4, 3, 2)-tabloid is
1 2 3 4
5 6 7 .
8 9

Exercise: Show that the 10 distinct (3, 2)-tabloids are, in increasing order
3 4 5 < 2 4 5 < 1 4 5 < 2 3 5 < 1 3 5
1 2 1 3 2 3 1 4 2 4

< 1 2 5 < 2 3 4 < 1 3 4 < 1 2 4 < 1 2 3.

3 4 1 5 2 5 3 5 4 5
See Problem Sheet 3, Question 9 for a connection with the colexico-
graphic order on subsets of N.

Lemma 6.4. Let λ be a partition of n and let s be a column standard λ-tableau.

If h is a non-identity element of C (s) then {s} > {sh}.
Proof. Let m = max{ j ∈ {1, 2, . . . , n} : jh 6= j}. Suppose that `h = m.
Note that since h ∈ C (s), we have that ` and m lie in the same column
of s. Suppose that ` lies in row r (`) of s and m lies in row r (m) of s.
Since s is column standard, r (`) < r (m). Therefore the greatest number
that appears in a different row in {s} and {sh}, namely m, appears in a
lower row in {s} than in {sh}. 

Proposition 6.5. Let λ be a partition of n. Working over a field F, the set

{e(s) : s a standard λ-tableau}
is F-linearly independent.
28

Proof. Suppose, for a contradiction, that

∑ αt e(t) = 0
t

where the sum is over all standard λ-tableaux t in a non-trivial linear

dependency. (So αt 6= 0 for all t in this sum.) Among these λ-tableaux,
let tmax be the one such that {t} is greatest in the order >. Now suppose
that s is a standard λ-tableau such that αs 6= 0. Since s and tmax are row
standard, either s = tmax or s 6= tmax and so {tmax } > {s}. Moreover, if
h is a non-identity element of C (s) then, by Lemma 6.4, we have
{tmax } ≥ {s} > {sh}.
Hence
0= ∑ αt e(t) = αtmax {tmax } + y
t

where y ∈ MFλis an F-linear combination of tabloids {u} such that

{tmax } > {u}. This contradicts the linear independence of the distinct
λ-tabloids. 

6.2. Garnir relations. The ‘spanning’ part of the proof of Theorem 6.2
is best presented as a result about integral Specht modules. Recall from
Definition 2.2 that MZλ is the ZSn -permutation module with free Z-basis
given by the set of λ-tabloids. By definition SZλ is the ZSn -submodule of
MZλ spanned by all the λ-polytabloids.

Definition 6.6. Let λ be a partition of n and let i ∈ N. Let t be a λ-

tableau. Let X be a subset of the entries in column i of t and let Y be a
subset of the entries in column i + 1 of t. Let SX , SY and SX ∪Y denote
the full symmetric groups on X, Y and X ∪ Y, respectively. Let
c
S X ∪Y = (SX × SY ) g j
[

j =1

where the union is disjoint. We say that the element

c
∑ gj sgn( gj ) ∈ ZSn
j =1

is a Garnir element for X and Y.

Note that there is an arbitrary choice of coset representatives involved

in this definition. We will see that the choice made is irrelevant in all
our applications of Garnir elements.
 29

Example 6.7. Let λ = (2, 1) and let t = 2 1 . Taking X = {2, 3} and

3
Y = {1} we see that a set of coset representatives for SX × SY in SX ∪Y
is 1, (12) and (13). Hence
GX,Y = 1 − (12) − (13)
is a Garnir element for X and Y. Observe that

e(t) GX,Y = 2 1 − 3 1 1 − (12) − (13)

 

3 2

= 2 1 − 3 1 − 1 2 + 3 2 − 2 3 + 1 3
3 2 3 1 1 2
= 0.

This example is a special case of a more general result that leads to an

algorithm for writing an arbitrary polytabloid as an integral linear com-
bination of standard polytabloids. In the following theorem, λ0 denotes
the conjugate of the partition λ, as defined in Remark C on page 13.

Theorem 6.8. Let λ be a partition of n and let t be a λ-tableau. Let i ∈ N, let

X be a subset of the entries in column i of t, and let Y be a subset of the entries
in column i + 1 of t. Let GX,Y ∈ ZSn be a Garnir element for X and Y. If
| X | + |Y | > λi0 then
e(t) GX,Y = 0.
Proof. Let
G X ∪Y = ∑ g sgn( g).
g ∈ S X ∪Y
It is clear that
  
G X ∪Y = ∑ h sgn(h) ∑ k sgn(k) GX,Y .
h∈SX k∈SY

Moreover, since e(t)h sgn(h) = e(t) and e(t)k sgn(k ) = e(t) for all h ∈
SX and k ∈ SY we have
e(t) GX ∪Y = | X |!|Y |!e(t) GX,Y .
Therefore, since we work over Z and e(t) = ∑h∈C(t) {th} sgn(h), it will
suffice to show that
{th} GX ∪Y = 0
for each h ∈ C (t). If h ∈ C (t) then, since | X | + |Y | > λi0 , there exist
x ∈ X and y ∈ Y such that x and y lie in the same row of th. Now
d
{th} GX ∪Y = {th}(1 − ( xy)) ∑ k j sgn(k j ) = 0
j =1

where k1 , . . . , k d is a set of coset representatives for h( xy)i inside SX ∪Y .

The theorem follows. 
30

6.3. The standard polytabloids span SZλ . Let λ be a partition. If t is a

λ-tableau and h ∈ C (t) then e(th) = e(t) sgn(h) = ±e(t). So it is clear
that
SZλ = he(t) : t a column standard λ-tableaui .
To show that in fact SZλ is spanned by the standard tableaux we need an
analogous definition to Definition 6.3.

Definition 6.9. Let λ be a partition. Given distinct column standard

λ-tableaux t and s, we define
the greatest number appearing in a different
t  s ⇐⇒ column in t than s lies in a column further to
the right in t than s.

Note that this is the opposite order to that used by James on page 30
of [7]. See Question 7 on Sheet 3 for one reason for preferring the order
as defined above.
The column standard (2, 2)-tableaux are, in increasing order
3 1 ≺ 2 1 ≺ 1 2 ≺ 2 1 ≺ 1 2 ≺ 1 3
4 2 4 3 4 3 3 4 3 4 2 4
For instance, since we compare on the largest number in a different
column, any tableau with 4 in the first column comes before any tableau
with 4 in the second column.

Lemma 6.10. Let λ be a partition and let s be a column standard λ-tableau.

Then e(s) ∈ SZλ is a Z-linear combination of standard λ-polytabloids.
Proof. We may suppose that s is not standard. By induction we may
assume that if t is column standard and t  s then e(t) is a Z-linear
combination of standard λ-polytabloids.
Since s is not standard there is some row, say row q, that is not in-
creasing. Thus there exists i < λ1 such that the entries in column i of s
in rows q up to λi0 are
x q < x q +1 < · · · < x λ 0 ,
i

the entries in rows 1 up to q of column i + 1 of s are

y1 < y2 < · · · < y q ,
and xq > yq . Set X = { xq , xq+1 , . . . , xλ0 } and Y = {y1 , y2 , . . . , yq }. Let
i
GX,Y be a Garnir element corresponding to the sets X and Y. We may
suppose that GX,Y = ∑kj=1 g j sgn( g j ) where g j is a product of transposi-
tions swapping a subset of Y with a subset of X. Let g1 be the identity,
corresponding to the case where both subsets are empty.
 31

By Theorem 6.8 we have e(s) GX,Y = 0, so

d
e(s) = − ∑ sgn( g j )e(sg j ).
j =2

It is possible that some of the sg j are not column standard. (We can
assume that the entries of sg j in the positions occupied by X and Y
are increasing, but not anything stronger than this: see Example 6.11
below.) For each j ∈ {2, . . . , d} let u j be the column standard tableau
whose columns agree setwise with sg j . Then
k
e(s) = ∑ ±e(u j )
j =2

for some appropriate choice of signs. It therefore suffices to show that

u j  s for each j ∈ {2, . . . , d}.
Since (SX × SY ) g j 6= SX × SY we have X ∩ Yg j 6= ∅. By choice of the
g j , the effect of applying g j to s is to move the elements in X ∩ Yg j from
X to Y. Let xmax be the greatest element of X ∩ Yg j . Since g j fixes all
elements not in X ∪ Y, and if x ∈ X and y ∈ Y then x > y, we see that
xmax is the largest element that appears in a different column in s and
sg j , and that in sg j , xmax lies in column j + 1. It follows that u j  s, as
required. 

Example 6.11. Let t = 2 1 . Following the proof of Lemma 6.10 we

4 3
might take X = {4} and Y = {1, 3}. A Garnir element for these sets is

GX,Y = 1 − (14) − (34),

so by Theorem 6.8 we have

2 1 2 4 2 1
     
e =e +e
4 3 1 3 3 4

= e 1 3 +e 2 1 .
   
2 4 3 4
Note that by replacing (14) with the alternative coset representative

(134) ∈ (SX × SY )(14) = {(14), (13)(14)}

we would get e 2 1 (134) = e 2 3

   
which has its second col-
4 3 1 4
umn in increasing order. But in both cases the first column must have 1
in its second row, so we cannot immediately obtain a column standard
tableau.
32

The second tableau in the sum is still not standard, so we repeat the
argument with X = {2, 3} and Y = {1}. This leads to

2 1 1 2
   
e =e .
4 3 3 4
This can be seen (in an easier way) working directly from the definition
of polytabloids. (See Question 8 on Sheet 3.)

Remarks:
(1) There are other relations that can be used to write an arbitrary
λ-polytabloid as a linear combination of standard polytabloids.
Fulton’s quadratic relations (see [5, §7.4]) are often easier to use
in practice than the Garnir relations.
(2) It follows from [8, Proposition 4.1] that if λ is a partition and t is
any column standard λ-tableau then
e(t) = e(t̄) + x
where t̄ is the row standard λ-tableau obtained from t by rear-
ranging its rows into increasing order, and x is a Z-linear combi-
nation of polytabloids e(s) such that s is standard and t  s. (By
Question 4 on Sheet 1, the tableau t̄ is in fact standard.)

We are now ready to prove the Standard Basis Theorem. Let F be a

field and let λ be a partition of n. By Proposition 6.5 the standard λ-
polytabloids are linearly independent over F. By the argument at the
start of §6.3, it suffices to show that if t is a column-standard λ-tableau
then e(t) is an F-linear combination of standard λ-polytabloids.
Thinking of e(t) as an element in SZλ , Lemma 6.10 implies that

e(t) = ∑ cs e(s)
s

where cs ∈ Z and the sum is over all standard λ-tableaux s. But the
same equation holds if we think of e(s) as an element of SλF , and regard
the cs as elements of F, so we are done.

We remark that the Standard Basis Theorem also holds for integral
Specht modules: linear independence over Z is clear from the argument
in Proposition 6.5, and the harder part, that the standard λ-polytabloids
span SZλ , follows from Lemma 6.10.
This remark has the following technical consequence. Let p be prime.
The tensor product sends the inclusion SZλ → MZλ to a homomorphism
of F p Sn -modules
SZλ ⊗Z F p → MZλ ⊗Z F p .
 33

It is clear that the codomain is isomorphic to MFλp as an F p Sn -module.

There is an obvious homomorphism of F p Sn -modules
SZλ ⊗F p Z → SFλp

defined by e(t) ⊗ α 7→ αe(t). It follows from the Standard Basis Theo-

rem that this map is an isomorphism. So the induced map is simply the
expected inclusion of SλF → MFλ . Therefore, there is an isomorphism of
ZSn -modules:
SZλ ∼ λ
= SZ ⊗ F p ∼
= SFλp .
pSZλ

Example 6.12. To see that there is something non-trivial going on here,

consider the ZS2 -permutation module MZ with Z-basis e1 , e2 and its Z-
submodule UZ = he1 + 2e2 , e2 + 2e1 i. Let MF3 be the F3 S2 -module with
F3 -basis v1 , v2 . The surjective map
MZ ⊗ Z F 3 → MF 3
defines by ei ⊗ α 7→ αvi restricts to a map
UZ ⊗Z F3 → MF3
with image he1 + 2e2 , e2 + 2e1 i = he1 + 2e2 i. So in this case the map
induced by the inclusion UZ → MZ is not injective.
We end with an immediate corollary of the Standard Basis Theorem.
Corollary 6.13. Let λ be a partition of n ∈ N. The dimension of SλF is equal
to the number of standard λ-tableaux. It is therefore independent of the field F.

7. S OME SUGGESTIONS FOR FURTHER READING

(1) James’ proof of the branching rule for representations of sym-
metric groups: see §9 of his book.
(2) Semistandard homomorphisms between Young permutation mod-
ules: see §13 of his book. (This leads to a combinatorial descrip-
λ as a summand of M µ for λ  µ.)
tion of the multiplicity of SC C
(3) Hook formula for the degreees of the characters of Sn : one ele-
gant proof was given by Greene, Nijenhius and Wilf in Adv. in
Math. 31 (1979) 104–109.
(4) Symmetric functions are closely related to the representation the-
ory of the symmetric group. Sagan’s book: The Symmetric Group,
Graduate Texts in Mathematics 203, Springer 2001 has an acces-
sible introduction.
34

(5) The representations of symmetric groups labelled by hook parti-

tions (n − r, 1r ) are often useful as a source of examples. A nice
account of their properties (including the result in Question 6)
was given by Hamernik in Specht modules and the radical of the
group ring over the symmetric group γ p , Comm. Algebra. 4 (1976)
435–476.

R EFERENCES
[1] B ENSON , D. J. Some remarks on the decomposition numbers for the symmet-
ric groups In The Arcata Conference on Representations of Finite Groups, Proc.
Symp. Pure Math. AMS 47, Part 1 (1987), pp. 381–394. Available from http:
//www.abdn.ac.uk/~mth192/html/archive/benson.html.
[2] C URTIS , C. AND R EINER , I. Representation theory of finite groups and asso-
ciative algebras. AMS Chelsea 1962.
[3] D LAB , V., AND R INGEL , C. The module theoretic approach to quasi-hereditary
algebras. In Representations of Algebras and Related Topics (1992), vol. 168 of
London Mathematical Society Lecture Note Series, Cambridge University Press,
pp. 200–224.
[4] FAYERS , M. On the structure of Specht modules. Journal of the London Math.
Soc. (Series 2) 67, (2003), 85–102.
[5] F ULTON , W. Young Tableaux, volume 35 of London Mathematical Society Student
Texts. Cambridge University Press, 1997.
[6] G REEN , J. A. Polynomial representations of GLn , vol. 830 of Lecture Notes in Math-
ematics. Springer-Verlag, Berlin, 1980.
[7] J AMES , G. D. The representation theory of the symmetric groups, vol. 682 of Lecture
Notes in Mathematics. Springer, Berlin, 1978.
[8] W ILDON , M. Vertices of Specht modules and defect groups of the symmetric
group. J. Alg. 323, (2010), 2243–2256.
[9] X I , C. Cellular algebras, Advanced School and Conference on Representation
Theory and Related Topics (9 - 27 January 2006). Available from http://www.
math.jussieu.fr/~keller/ictp2006/lecturenotes/xi.pdf.