You are on page 1of 19

Hillsdale College

Department of Mathematics and Computer Science

Honors Thesis

Understanding Representation Theory and


Tensor Products via Young Tableaux

Author: Supervisor:
David K. Wagner Dr. David Murphy

May 5, 2009
Abstract
Representation theory is an essential part of group theory, producing numerous important results.
We introduce the fundamentals of representation theory, such as irreducible representations and
characters. We then present partitions, Young diagrams, and tableaux and discuss their basic
properties. These combinatorial tools possess several surprising connections with representation
theory. We examine these connections and demonstrate their helpfulness in analyzing representa-
tions. Additionally, we consider tensor products and multilinear algebra and the ways in which
partitions help us study tensor products, as another example of the surprising relationships between
combinatorics and other parts of mathematics. This study offers a fascinating taste of the hidden
links between seemingly disparate branches of the mathematical endeavor.

1
Acknowledgments
I would like to thank Dr. David Murphy for his patient and diligent help through the year, guiding
both my understanding and my organized presentation of this material.
I would also like to thank my other mathematics professors and teachers for their inspiration
and support through the years.

2
Contents
1 Introduction 4

2 Fundamentals of Group Theory 4

3 Representation Theory 6

4 Introduction to Multilinear Algebra 11

5 Forming New Representations from Old 12

6 Partitions and Young Tableaux 12

7 Using Young Tableaux to Find Representations 14

8 Conclusion 17

3
1 Introduction
Group theory is a vast branch of mathematics that includes many complicated ideas. Group
theory has numerous applications; for example, it helps us understand symmetry, coding theory,
and Galois theory. Because of these applications and others, group theory merits thoughtful and
detailed study.
Representation theory is one of the most important tools for studying group theory, in which we
consider representations that relate one group to another group. Representation theory allows us to
study groups by seeing their interactions with one another, and other objects, such as vector spaces,
rather than studying them on their own; linear representations allow us to use linear algebra, which
is relatively accessible. Even so, to better understand representation theory and groups, we look
to other areas of mathematics. Combinatorial methods, involving partitions and tableaux, offer
one method of better approaching and computing properties of groups and representations. In this
paper, we shall describe group theory, with a special focus on representation theory. Next, we will
introduce partitions and Young tableaux. We will then prove how these tableaux, with relevant
operations and procedures, guarantee certain results in representation and group theory, providing
a simpler, more computable method to reach these results.

2 Fundamentals of Group Theory


Before discussing tableaux and representations, we shall examine the rudiments of group theory.
We’ll begin with a straightforward example.

Example 1. Consider an equilateral triangle with vertices 1, 2, and 3.


3

1• •2
We can find the various symmetries of this triangle through rotations and reflections. Rotating
the triangle 120, 240, and 360 degrees produces three symmetries. Reflecting the triangle about
the perpendicular bisectors passing through each vertex produces three more symmetries. This
set of six symmetries, denoted D3 , has certain properties. Doing any sequence of them produces
another symmetry in D3 ; rotating 360 degrees returns the vertices to their original position; and
each movement can be undone by another rotation or reflection. For instance, rotating 240 degrees
and then rotating 120 degrees gives a 360 degree rotation, returning the vertices to their original
locations. Because the symmetries in D3 possess these properties, we call D3 a group.

Now, let’s consider the formal definition of a group.

Definition 1. Let G be a set of elements equipped with the binary operation (◦). We call G a
group if, for all g, h, k ∈ G:
1. (g ◦ h) ◦ k = g ◦ (h ◦ k),
2. There exists e ∈ G such that
g ◦ e = e ◦ g = g,
3. For each g ∈ G, there exists x ∈ G such that

g ◦ x = x ◦ g = e.

4
Now, let’s examine a few more examples of groups. First, just as D3 is the group of symmetries
of the equilateral triangle, we have D4 , the group of symmetries of the square, and D5 , the group
of symmetries of the regular pentagon. Second, let’s look at the real numbers, R. The real
numbers are a group for addition, because 0 is the identity, every element has an inverse, and the
associative property holds. However, R is not a group for multiplication, because 0 does not have
a multiplicative inverse; if we form the set R∗ , which consists of all real numbers save 0, we obtain
a group under multiplication.
Beyond the real numbers, diverse groups exist. For example, if we consider the field Z17 , we find
that these 17 residue classes are a group under addition. Under multiplication, we obtain the group
Z∗17 , which excludes the class with residue 0 so that all the elements have multiplicative inverses.
For addition, Z17 happens to be commutative, just as Z∗17 is commutative for multiplication.
We call all groups Abelian whose binary operation is commutative. Not all groups are Abelian.
For example, the set of invertible 2 × 2 matrices, U (M2 (R)), is a group under matrix multiplica-
tion. However, matrix multiplication is not commutative, so U (M2 (R)) is a non-Abelian group.
Other examples of non-Abelian groups includes D3 , D4 , and D5 , because the various rotations and
reflections in these groups do not commute.
In this paper, we will focus on a special type of group, the symmetric group.
Definition 2. For a set S, we define a permutation of S as a one-to-one and onto function from S
to itself. The set of all permutations of the set {1, 2, ..., n} with n elements is the symmetric group
Sn , whose binary operation is composition.
Let’s take a look at an example, the symmetric group S3 . The set S3 consists of all the possible
permutations of three elements. The identity element of this set is ( 11 22 33 ), where the array shows
vertically the mapping for each element in S3 . Other elements of this set include ( 11 23 32 ), ( 12 21 33 ),
( 13 22 31 ), ( 12 23 31 ), and ( 13 21 32 ). Thus, we have six total elements in our group S3 .
Our group operation for S3 is composition of functions. Thus, when we combine elements of S3 ,
we compose from right to left. For instance, ( 13 21 32 ) ( 13 22 31 ) = ( 12 21 33 ) . We determine this result by
considering each element involved in the permutations individually from right to left. For example,
1 maps to 3 in the right-hand permutation, then 3 maps to 2 in the left-hand permutation of our
product, giving 1 mapping to 2 for our product. Notice that we have produced another element of
our group by this multiplication; we expect this result because groups are closed.
Now that we’ve looked at S3 , let’s consider some more properties of symmetric groups. We’ve
already observed that each element of a symmetric group is written as an array that maps elements
to other elements. Using array notation becomes rather cumbersome; instead, mathematicians
prefer to use notation involving cycles, which map each element to another element in a chain of
relations within the set S of n elements. Formally, here’s how we define a cycle.
Definition 3. A cycle of length k is a permutation α ∈ Sn such that integers a1 , a2 , ..., ak ∈
{1, 2, ...n} exist such that α(a1 ) = a2 , α(a2 ) = a3 , ..., α(ak−1 ) = ak , α(ak ) = a1 and the remaining
n − k elements in {1, 2, 3..., n} are left unchanged by α. We write this cycle as (a1 a2 a3 ...ak ).
So, if we consider the elements of S3 again, we find ( 12 23 31 ) can be written in cycle notation as
(123), which is a cycle of length three. However, ( 12 21 33 ) has cycle notation (12) because the 3 does
not change position; this cycle is of length two.
Cycles of length two merit particular attention. These cycles are called transpositions.
To conclude our introduction to the symmetric group, let’s consider the factorization of permu-
tations into cycles, guaranteed by the Cycle Factorization Theorem.
Theorem 1 (Cycle Factorization Theorem). Every non-identity permutation is either a cycle or
can be uniquely factored (up to order) as a product of pairwise disjoint cycles.

5
Another factorization theorem involves transpositions.

Theorem 2. Any permutation can be factored as a product of transpositions. A permutation can


only be factored into an even or odd number of transpositions, never both.

The first part of this theorem can be proved easily. We simply take a cycle of length k,
(a1 a2 ...ak ), which can be written as

(a1 ak )(a1 ak−1 )...(a1 a3 )(a1 a2 ).

Returning to our symmetric group S3 , we can write the permutation (132) as a product of tranpo-
sitions: (132) = (12)(13). Similarly, (123) = (13)(12).
The rest of the theorem requires a bit more explanation. We label even permutations those
permutations that can be expressed as the product of an even number of transpositions; odd per-
mutations are those that can be expressed as a product of an odd number of transpositions.
Before we finish our introduction of group theory, we should also consider the relationships
between groups. One concept that explores these relationships is the group homomorphism.

Definition 4. A group homomorphism is a function ϕ that maps from a group G with operation
◦ to a group H with operation ∗ such that

ϕ(g ◦ k) = ϕ(g) ∗ ϕ(k)

for every g, k ∈ G.

Homomorphisms allow us to compare groups. The definition of a homomorphism requires group


structures to be similar. When two groups have identical structure, we can define a more precise
mappping, an isomorphism.

Definition 5. An isomorphism is a group homomorphism that is both one-to-one and onto.

For example, let’s consider the function ϕ : Z → Z defined by ϕ(a) = 17a. We know ϕ is an
homomorphism for addition, because, for all p, q ∈ Z, we have

ϕ(p + q) = 17(p + q) = 17p + 17q = ϕ(p) + ϕ(q).

We know this function is not onto; for instance, we cannot find a ∈ Z such that ϕ(a) = 11.
Therefore, we conclude ϕ is not an isomorphism. However, if we consider the function ̺ : Z → 17Z,
we will have an isomorphism under addition, because ̺ is both one-to-one and onto.

3 Representation Theory
Now that we have examined some fundamentals of group theory, let’s move on to representation
theory. First of all, we need some basic definitions.

Definition 6. For two vector spaces V and W over a field k, a linear transformation from V to
W is a function ϕ : V → W satisfying

ϕ(v1 + v2 ) = ϕ(v1 ) + ϕ(v2 )


ϕ(tv) = tϕ(v)

for all v1 , v2 , v ∈ V and all t ∈ k.

6
We denote the set of all such linear transformations from V to W by Homk (V, W ).
Linear representations are group homomorphisms to a general linear group, defined as follows.

Definition 7. The general linear group of a vector space V is the group of all invertible k-linear
transformations from V to V , denoted GLk (V ). ([2], p. 3)

Having defined linear transformations and general linear groups, we can define a linear repre-
sentation:

Definition 8. A linear representation of a group G is a group homomorphism ρ : G → GL(V ),


from G to the general linear group of some vector space, V . Such a representation defines a k-linear
action of G on V by
g · v = ρg (v) = ρ(g)(v).
The dimension of V is called the dimension of the representation. ([2], p. 15).

In other words, a linear representation is a homomorphism that relates a group of interest to


a known, familiar general linear group. There are many different kinds of linear representations.
Subrepresentations are linear representations that are restricted to a G-subspace W of V , where a
G-subspace is a subspace of our vector space that is closed under addition, scalar multiplication,
and the k-linear action of G. If a linear representation has only two G-subspaces, 0 and V , we say
the representation is irreducible or simple.
Let’s consider an example of a special representation, the trivial representation.

Example 2. Let ι : G → GL(k) be the one-dimensional representation of an arbitrary group, G,


such that
ι(g) = 1 for all g ∈ G.
This representation is called the trivial representation, because every element of the group is mapped
to 1. We know ι is a homomorphism, because any combination of elements of G will produce
another element in G that will still give output 1. The trivial representation is an example of a
simple representation, because its only G-subspaces are 0 and k.

Next, let’s consider some representations of the symmetric group S3 .

Example 3. There are three irreducible representations of the symmetric group S3 . First, of course,
we have the trivial representation, which maps each group element to 1. This is a simple, one-
dimensional representation. Another representation is the sign representation. This representation
maps group elements that are even permutations to +1; group elements that are odd permutations
are mapped to −1. Thus, the identity, (13)(12), and (12)(13) are all mapped to +1, while (13),
(12), and (23) are mapped to −1. This is another one-dimensional representation of S3 . Finally,
we can think of S3 as the dihedral group D3 , which consists of the symmetries of an equilateral
triangle in R2 . This representation can be viewed graphically as the movement of the plane for
each permutation of the triangle, so it is a two-dimensional representation of S3 . At this point, we
observe that the sum of the squares of the dimensions of these representations of S3 , 12 +12 +22 = 6,
is equal to the number of elements in S3 . We will comment further on this observation later.

Let’s consider a more complicated example of a representation, involving the dihedral group
D4 .

Example 4. Consider a square centered on the origin, with vertices (−1, −1), (1, −1), (1, 1), and
(−1, 1).

7
4• •3

• •
1 2
To help us analyze elements of D4 , we will label these vertices 1, 2, 3, and 4 respectively. One
possible representation is the trivial representation. As we just saw, this representation will map
every element of our group to 1. Again, the sign representation is another one-dimensional rep-
resentation. Group elements with an even number of two-cycles, such as reflections about the
horizontal and vertical axes, map to 1. In cycle notation, these reflections are written (14)(23)
and (12)(34). Conversely, elements of D4 with an odd number of two-cycles, such as the rotation
(1234) = (14)(13)(12), map to −1. There is also a two-dimensional representation of D4 , which
maps D4 to GL(R2 ). Let’s examine this representation with (1234). We can determine ρ(1234)
geometrically by looking at each vertex: for example, vertex 1 moves from quadrant III of the
coordinate plane to quadrant IV. After looking at the movement of all the vertices, we see ρ(1234)
is a 90 degree counterclockwise
 rotation of R2 . In matrix notation, this produces cos 90 − sin 90
sin 90 cos 90
0 −1
which equals 1 0 . We can find the other outputs for the elements of D4 in the same manner.

g ∈ D4 ρ(g) g ∈ D4 ρ(g) 
(1) ρ(1) = ( 10 01 )  ((12)(34)) ρ((12)(34)) = −1 0
0 1 
((14)(23)) ρ((14)(23)) = 10−1 0 ((13)(24)) ρ((13)(24)) = −1 0
0 −1
(1432) ρ(1432) = −1 0 1 (1234) ρ(1234) = 1 0 0 −1
0
0 −1
(13) ρ(13) = −1 0 (24) ρ(24) = ( 01 10 )

Finally, using our table of outputs for our two-dimensional representations,


 0 −1 we can check that
ρ is indeed a homomorphism. For example, ρ((12)(34))ρ(1234) = −1 0
0 1 1 0 = ( 0 1 ) = ρ(24)
10
and (12)(34)(1234) = (24).

We can also compare linear representations. For example, we will look at linear transformations
involving two representations:

Definition 9. Let ρ : G → GLk (V ) and σ : G → GLk (W ) be representations. We call a linear


transformation f : V → W G-equivariant or a G-homomorphism from ρ to σ, if, for each g ∈ G,
σg ◦ f = f ◦ ρg . If a G-homomorphism is a linear isomorphism, we call it a G-isomorphism and call
them G-equivalent representations. ([2], p. 17).

For instance, if we label the vertices in clockwise order in the previous example, the resulting
two-dimensional representation is equivalent, even though the matrices generated for different group
elements differ.
Another way to study linear representations involves characters. Before defining characters,
recall that the trace of an n × n matrix, denoted tr, is the sum of the diagonal entries.

Definition 10. Let G be a finite group and ρ : G → GLC (V ) be a finite dimensional C-


representation such that dimC V = n. For all g ∈ G, the character value of g in the representation
ρ is the trace of g on V . Formally, we say the character of ρ is the function χp : G → C given by

χρ (g) = tr(ρg ) = tr(g).

([2], [p. 33]).

8
Frequently, we will construct character tables, listing the different character values given by a
representation. To build these tables, we will use conjugacy classes. For a group G, the conjugacy
class of x ∈ G is the set of all elements gxg −1 ∈ G such that g ∈ G. All of the elements in
a conjugacy class will have the same character value for a given representation, because χ(g) =
χ(hgh−1 ) for all h ∈ G. Thus, in a character table, each column is a specific conjugacy class
of the group G. The rows in a character table correspond to the various characters χi of each
nonequivalent irreducible representation of G.
Before building a sample character table, there are a few helpful results about characters to
consider. First of all, we define the inner product of characters, which will help us build our
character table. For characters α and β of a group G, we define their inner product to be
1 X
(α|β)G = α(g)β(g).
|G|
g∈G

Importantly, the characters of non-equivalent irreducible representations form an orthonormal


P 2 set.
That is, (α|β)G = 0 if α 6= β and (α|α)G = 1. As a result, 1 = (α|α)G implies α (g) = |G|,
which means the sum of the squares of the dimensions of each character equals the order of the
group represented ([2], p. 41).
There are also orthogonality relationships between the columns of a character table. If we
consider any two conjugacy classes Ca and Cb , the products of the characters for each representation
must sum to zero, that is, X
χ(Ca )χ(Cb ) = 0
χ

for Ca 6= Cb . We also know that the character of the identity conjugacy class equals the dimension of
a representation α, because α(e) = tr(In ) = n. Hence the sum of the squares of the character values
of the identity conjugacy class for the irreducible representations equals the number of elements in
G.
Another consequence is that the number of conjugacy classes in G equals the number of distinct,
inequivalent irreducible representations. Since each of these representations produces a separate
set of character values for the different conjugacy classes in G, we obtain a square character table
with as many representations as conjugacy classes ([2], p. 38). Thus, our character tables contain
as many representations (rows) as conjugacy classes (columns).
We will use these facts to fill in our character table. Let’s consider some examples.

Example 5. First, we will construct a character table for the symmetric group S3 . We begin by
determining the conjugacy classes of S3 . There are three conjugacy classes in S3 : C1 contains the
identity; C2 contains (12), (13), and (23); and C3 contains (123) and (132). We note that in each
conjugacy class, all the group elements share the same cycle structure. This result is a theorem
which holds for all symmetric groups. Next, we consider the number of elements in S3 , which is six.
This means that the sum of the squares of the dimensions of each irreducible representation equals
six. Since we have three conjugacy classes, we have three irreducible representations. Consequently,
one of these is two-dimensionsal; the others are one-dimensional.
At this point, we can fill in our table rather easily. The two sets of characters from one-
dimensional representations come from the trivial representation and the sign representation. Then,
since the third representation has dimension two, the first character entry must be 2. We can use
column orthogonality to find the remaining two character values for the two-dimensional representa-
tion. For example, comparing the columns for C1 and C2 , we need (1)(1)+(1)(−1)+(2)(χ3 (C2 )) = 0,
so χ3 (C2 ) = 0. Our completed table appears as follows:

9
C1 C2 C3
χ1 1 1 1
χ2 1 −1 1
χ3 2 0 −1

We can use our row-wise orthogonality inner product to check these results. For example, the
inner product
(χ1 |χ2 ) = (1 × 1) + 3(1 × −1) + 2(1 × 1) = 0.
Each of the other inner products also yields 0, confirming our table.
If we compare these results with the representations we discussed in Example 3, we see that we
have the exact same group of three irreducible representations. We can compute a few characters
from our matrix notation for additional verification: for example, our two-dimensional representa-
tion gives ( 10 01 ) for the identity
 element, which has tr = 2, the value we obtain for C1 in our table.
Similarly, tr = 0 for −1 0 1
0 , which is the matrix representation of (12), an element of C .
2

Example 6. Now, let’s construct a character table for the representations of the group D4 . First,
we must determine the conjugacy classes of D4 . We can check that five conjugacy classes exist: C1
contains the identity; C2 contains (13) and (24); C3 contains (13)(24); C4 contains (14)(23) and
(12)(34); and C5 contains (1234) and (1432). Thus, we know that five irreducible representations
exist, giving a character table with five rows and five columns.
Since this group has eight elements and it has an irreducible representation in 2-dimensions, it
must have four one-dimensional irreducible representations, because 12 + 12 + 12 + 12 + 22 = 8.
This allows us to fill in the first column, for the identity conjugacy class, with 1, 1, 1, 1, and 2.
We know one row of characters represents the trivial representation and another row represents
the sign representation, so we can fill the first two one-dimensional rows accordingly. The trivial
representation will always give character value 1, while the sign representation gives character
value 1 or −1, depending on the number of two-cycles in the conjugacy class (not coincidentally, all
elements of a conjugacy class will always have the same cycle structure, although cycle structure
no longer determines conjugacy class, since D4 is not a symmetric group).
For our two-dimensional representation, we can use the geometric approach that we took ealier.
Using our graph of the square from Example 4, the ‘natural representation,’ we can find matrix
values for each of the conjugacy
 classes. For example, if we consider (1234) in conjugacy class C5 ,
0 −1
we obtain the matrix 1 0 which has trace 0, giving character 0. We can compute the characters
for the other conjugacy classes similarly.
Now, what about our two other mysterious 1-dimensional representations? We can use our or-
thogonality relations to easily determine their characters. Thus, we arrive at the complete character
table shown below.

C1 C2 C3 C4 C5
χ1 1 1 1 1 1
χ2 1 1 −1 1 −1
χ3 1 1 1 −1 −1
χ4 1 1 −1 −1 1
χ5 2 −2 0 0 0

Thus, we see that characters are a useful tool for understanding and dealing with representations.
Before moving further into our study of representation theory, we will introduce some basic concepts
of multilinear algebra, which will enhance our analysis of representations.

10
4 Introduction to Multilinear Algebra
Multilinear algebra expands on the properties commonly associated with linear algebra. Multilinear
algebra generally involves combinations of vector spaces.
Definition 11. A vector space V on a field k is a set of elements equipped with operations of
vector addition and scalar multiplication subject to several axioms. This means that for v, w ∈ V
and a ∈ k, v + w ∈ V and av ∈ V .
Given two vector spaces, V and W , we can form a new vector space using direct sums.
Definition 12. Given vector spaces V and W , their direct sum is V ⊕ W = {(v, w) ∈ V × W |v ∈
V, w ∈ W } with vector addition and scalar multiplication given by:
1. (v1 , w1 ) + (v2 , w2 ) = (v1 + v2 , w1 + w2 )
2. a(v1 , w1 ) = (av1 , aw1 )
Additionally, if dim(V ) = n and dim(W ) = m, then dim(V ⊕ W ) = n + m. If {e1 , ..., en } is a
basis of V and {f1 , ..., fm } is a basis of W , then {(e1 , 0), ..., (en , 0), (0, f1 ), ...(0, fm )} is a basis of
V ⊕ W.
We call a function multilinear, if it is linear for each variable it contains separately. Before look-
ing at ‘multiplying’ elements from different vector spaces, we should consider the formal definition
of a multilinear function.
Definition 13. Given k-vector spaces V1 , ..., Vr and W , we say a function
F : V1 × · · · × Vr → W
is k-multilinear if it satisfies the following equations, where vj , vj′ ∈ V and t ∈ k
1. F (v1 , ..., vk−1 , vk + vk′ , vk+1 , ..., vr ) = F (v1 , ..., vk , ..., vr ) + F (v1 , ..., vk−1 , vk′ , vk+1 , ..., vr )
2. F (v1 , ..., vk−1 , tvk , vk+1 , ..., vr ) = tF (v1 , ..., vk−1 , vk , vk+1 , ..., vr )
Importantly, we will use this definition of a multilinear function to define a tensor product,
which will be a crucial concept in our analysis of representations.
Definition 14. The tensor product of V1 , ..., Vr is a k-vector space V1 ⊗ V2 ⊗ · · · ⊗ Vr with a
corresponding function π : V1 × · · · × Vr → V1 ⊗ V2 ⊗ · · · ⊗ Vr such that for any vector space
W and any multilinear map F : V1 × · · · × Vr → W , there is a unique linear transformation
F ′ : V1 ⊗ V2 ⊗ · · · ⊗ Vr → W satisfying F ′ ◦ π = F.
What about the dimension of a tensor product? The following theorem tells us that the dimen-
sion of the tensor product is the product of the dimension of each vector space involved.
Theorem 3. If we consider finite-dimensional k-vector spaces Vk with 1 ≤ k ≤ r having basis
vk = vk,1 , ..., vk,n where the dimk Vk = nk , then V1 ⊗ · · · ⊗ Vr has a basis consisting of the vectors
v1,i1 ⊗ · · · ⊗ vr,ir = π(v1,i1 ...vr,ir ) where 1 ≤ ik ≤ nk . Thus, we find dimk (V1 ⊗ · · · ⊗ Vr ) = n1 ...nr .
Let’s consider a simple example involving tensor products of vector spaces.
Example 7. Suppose we have a vector space V with basis e1 , e2 , ..., en and a vector space W with
basis f1 , f2 , ..., fm . The tensor product V ⊗ W is a vector space of dimension n × m with basis
ei ⊗ fj for i = 1, 2, ..., n and j = 1, 2, ..., m. Now, what if take some elements of these vectors spaces
and take their tensor product? Consider
(a1 e1 + a2 e2 ) ⊗ (b1 f1 + b2 f2 ) = a1 b1 (e1 ⊗ f1 ) + a1 b2 (e1 ⊗ f2 ) + a2 b1 (e2 ⊗ f1 ) + a2 b2 (e2 ⊗ f2 )
By the properties of a tensor product, taking the tensor product of two sums of two terms gives
a sum of four terms involving tensor products. Moreover, we have effectively combined the two
vector spaces.

11
5 Forming New Representations from Old
Now that we’ve discussed tensor products and direct sums, we have the tools necessary to build
new representations from existing ones. There are several different ways to find these new repre-
sentations.
First, we can look at subrepresentations, which we introduced in Section 3. Given a represen-
tation ρ on a vector space V , ρW is a subrepresentation of ρ when W is a subspace of V closed for
the k-linear action of G.
Second, we can use direct sums to form new representations. If V and W are G-representations,
so is V ⊕ W , whose k-linear G-action is given by g · (v, w) = (g · v, g · w).
The following theorem shows how subrepresentations and direct sums help us decompose rep-
resentations.
Theorem 4. Let ρ : G → GLk (V ) be a linear representation of a finite group and let V be
non-trivial. Then, we have G-subspaces U1 , ..., Ur ⊆ V , each of which is a nontrivial irreducible
subrepresentation, such that V ∼
= U1 ⊕ ... ⊕ Ur .
Thus, we see that a G-space factors into a direct sum of G-subspaces. Finding the irreducible
summands of a G-space is one of the key problems of representation theory.
Third, we can use tensor products to build new representations. Suppose we have V1 , ..., Vr as
k-vector spaces with representations ρ1 , ..., ρr of a group G. For each g ∈ G and each j ∈ {1, ..., r},
we have a linear transformation ρj g : Vj → Vj . Then, using tensor products, we can construct the
unique linear transformation
ρ1g ⊗ · · · ⊗ ρrg : V1 ⊗ · · · ⊗ Vr → V1 ⊗ · · · ⊗ Vr
because ρ1g ⊗ · · · ⊗ ρrg : V1 ⊗ · · · ⊗ Vr → V1 ⊗ · · · ⊗ Vr is multilinear. Thus, we obtain a linear
transformation involving the product of each of the r vector spaces, in which the representations
are expressed using tensor products. ([2], p. 21).
A fourth method to create new representations involves induced representations. Induced repre-
sentations take a representation of a subgroup H of a group G and use it to create a representation
for the whole group. This is denoted IndG H (W ) and it equals CG ⊗CH W . Induced representations
involve a critical element of group theory called cosets.
Definition 15. Given a group G and a subgroup H of G, for each g ∈ G we call a right coset of
H in G the set Hg = {hg : h ∈ H}; the left coset is the set gH = {gh : h ∈ H}.
Using our definition of cosets, we can better understand how inductions work. If we begin
with a basis b1 , ..., bm of W and {g1 H, ..., gk H} is the set of cosets of H in G, then {gi H ⊗ bj |i =
1, ..., k, j = 1, ..., m} can be taken as a basis for IndG
H (W ).
Conversely, one can find the restriction of a representation. The restiction of a representation
takes a representation ρ of a group G on a vector space W and yields a representation of a subgroup
H of that group also on W by the mapping ρH : H → GLk (W ), where ρH (h) = ρ(h) for all h ∈ H.
The restriction of a representation is denoted ResG H (W ).

6 Partitions and Young Tableaux


Partitions and Young tableaux are the final building blocks for our study of representations. We
will use Young tableaux to help us better understand representation theory, tensor products, and
character theory.
Before looking at actual tableaux, we should formally define a partition.

12
Definition 16. A partition of a nonnegative integer n is sum of positive integers adding to n, in
which the order of the summands is irrelevant.
Consider the number 7. It has 15 different partitions: 7, 6 + 1, 5 + 2, 5 + 1 + 1, 4 + 3, 4 + 2 + 1,
4 + 1 + 1 + 1, 3 + 3 + 1, 3 + 2 + 2, 3 + 2 + 1 + 1, 3 + 1 + 1 + 1 + 1, 2 + 2 + 2 + 1, 2 + 2 + 1 + 1 + 1,
2 + 1 + 1 + 1 + 1 + 1, and 1 + 1 + 1 + 1 + 1 + 1 + 1.
There are numerous ways to depict partitions. We can write the numerals for each component
of a partition; we can use little dots or circles to represent each component; we can also write
them in numerical shorthand: 3 + 2 + 1 + 1 = (3, 2, 1, 1) = (3, 2, 12 ). For our purposes, we will use
Young diagrams, also known as Ferrers diagrams, to illustrate partitions, in which each component
of a partition is shown as a horizontal row of boxes. We will usually denote Young diagrams
and partitions in general with the Greek letter λ. Here is the Young diagram for the partition
3 + 2 + 1 + 1.

Once we have our Young diagrams, we can fill the boxes as we wish. We’re specifically interested
in Young tableaux, which fill the boxes with numbers, called a numbering or a filling. There are a
few simple rules for generating these tableaux:
1. The numbers must be weakly increasing across each row.

2. The numbers must be strictly increasing down each column.


Of course, there are numerous possible fillings for a Young diagram. For instance, if we return to
our partitions of 7, we can find many different fillings for the partition 3 + 2 + 1 + 1. Some of the
tableaux that we can produce on this Young diagram include the following:

1 1 1 1 2 3 1 4 7 1 2 3
2 2 2 3 4 7 4 5
3 3 7 6
4 4 9 7

We observe that there are infinitely many possibilities, because the fillings can become infinitely
large.
Of the four tableaux we just produced, the last has a special property. In it the integers 1, 2,
3, 4, 5, 6, and 7 are each used exactly once. This is an example of a standard tableau.
Definition 17. A standard tableau is a tableau with n boxes, in which the entries are the integers
from 1 to n, each occurring once.
We can find many additional standard tableaux for a given Young diagram. Returning to our
partition of 7, we find additional standard tableaux:

1 3 6 1 5 7
2 5 2 6
4 3
7 4

Another important concept is the conjugate of a Young diagram, which is formed by flipping the
diagram diagonally on the axis from lower right to upper left. Of course, this switches the number

13
of columns with the number of rows in the original Young diagram and also switches the length of
each column and row. Here we consider the conjugate of our original Young diagram example.

A= AT =

If we begin with a tableau and take the conjugate of the underlying Young diagram, the numbers
filling the tableau will also switch locations; the resulting tableau is called the transpose. For a
tableau R, we label the transpose RT . Here are some examples of a tableau and its transpose.

1 1 1 1 2 3
2 2 1 2 3 4 4 5 1 4 6 7
3 1 2 6 2 5
P = 4 PT = 1 and Q= 7 QT = 3

The example with P and P T demonstrates that the transpose of a tableau is not necessarily a
tableau; however, as Q and QT suggest, the conjugate of a standard tableau is always a standard
tableau.

7 Using Young Tableaux to Find Representations


We now use our knowledge of Young tableaux to better understand representation theory. We want
to consider 3 results specifically involving the symmetric group Sn .

1. The partitions of n correspond to the conjugacy classes of Sn and the irreducible characters
of Sn .

2. The correspondence between the number of standard tableaux of a given shape λ and the
dimension of the related irreducible representation χλ of Sn .

3. The manner in which partitions of n can be used to find irreducible subrepresentations of a


tensor product V ⊗n .

With these goals in mind, let’s begin with the correspondence between partitions of n and conjugacy
classes and characters.
To understand the relationship between partitions and conjugacy classes, let’s return to our
favorite example, S3 . For n = 3, we find three partitions with three corresponding Young diagrams.

In Example 5, we observed that there are 3 conjugacy classes of S3 . One contains the identity;
one contains the permutations (12), (13), and (23); and the other contains the permutations (123)
and (132). We also observed that these conjugacy classes each contain permutations with the same
cycle structure. We can match each of these conjugacy classes of S3 to a particular Young diagram
for a partition of n = 3. The class containing the identity matches to the diagram with 3 rows with
1 box, because the identity permutation has 3 cycles of length 1. The class containing (12) matches
up with the diagram with 2 rows, because its permutations contain 1 cycle of length 2 and 1 cycle

14
of length 1. Similarly, the conjugacy class containing (123) corresponds to the diagram with 1 row
containing 3 boxes, because its permutations contain 1 cycle of length 3. Here’s a summary of our
results.

↔ {(132), (123)} ↔ {(12), (13), (23)} ↔ {identity}


Thus, we see that in a Young diagram of a partition of n, each row of the diagram corresponds to
a cycle in the permutations of a conjugacy class of Sn . In this way, we can find the conjugacy classes
of Sn , simply by constructing the Young diagrams for all the partitions of n. This is much simpler
than determining all the conjugacy classes by analyzing the cycle structures. This correspondence
also guarantees that, for a given n, the number of partitions of n equals the number of conjugacy
classes of Sn .
If we look at S4 and the partitions of 4, the helpfulness of partitions becomes much clearer.
The symmetric group S4 has 24 elements, so sorting them into conjugacy classes is no simple task.
However, we can find the 5 Young diagrams of the partitions of 4 quite easily.

From these Young diagrams, we conclude that S4 has 5 conjugacy classes. We can also determine
the cycle structure of the elements of each class; for example, there is a class whose elements’ cycle
structure is the product of two transpositions, and a class whose cycle structure is a three-cycle
and a one-cycle. After using the Young diagrams to determine this general information about
the conjugacy classes of S4 , it is much easier to sort out the 24 elements of S4 into the 5 known
conjugacy classes.
Now let’s examine the relationship between partitions and characters. Let’s begin with a
tableaux of shape λ. For example, consider the following tableau.

1 2 3 4
5 6 7
8 9

If we look at this tableaux, we can divide the fillings into groups, based on horizontal rows. Thus we
obtain three groups: {1, 2, 3, 4}, {5, 6, 7}, {8, 9}. We call this classification a Young Tabloid. Next,
we want to look for the elements of Sn that fix the components of this tabloid. That is, we accept
permutations among the elements of each row, but there can be no interchange between rows. For
example, the 2 and 4 can switch, but not the 4 and the 5. For our tabloid, we have

Sλ = S{1,2,3,4} × S{5,6,7} × S{8,9} ,

which is a subgroup of S9 . In general, we obtain the relationship

Sλ ∼
= S4 × S3 × S2 .

In this way, we have our Young tabloid which produces a subgroup of S9 whose factors are S4 , S3 ,
and S2 .

15
Next, we induce a representation ψλ of Sn from the trivial representation of Sλ by taking the
tensor product CSn ⊗CSλ C. This tensor product gives us the C-span of cosets of Sλ in Sn with
Sn -action
h(gSλ ) = (hg)Sλ
for all h, g ∈ Sn . We know Sλ has a finite number of cosets x1 Sλ ,x2 Sλ , ..., xkSλ for x1 , x2 , ..., xk ∈
Sn . We then construct the representation ρSSnλ , given by ρSSnλ (x) = ρb(x−1 b(x−1
i xxj ) i,j , so ρ i xxj ) fills
the ij-th block of a matrix, where we let ρb(z) = 0 if z ∈
/ Sλ and ρb(z) = ρ(z) if z ∈ Sλ .
We can induce other representations of Sn . For example, we obtain φλ as an induced repre-
sentation that corresponds to the sign representation of SλT . Then, we can find χλ , which is the
unique common irreducible component of ψλ and φλ . Continuing in this manner, we can find all
the characters of the representations of Sn . Thus, we see that the partitions help us generate our
character tables. Regrettably, there is no elementary formula for generating the values of these
characters directly from the partitions.
Next, we want to study the relationship between the fillings of a particular Young diagram and
the dimensions of representations. Before doing this, we must add a few concepts to our group
theory terminology.
Definition 18. In a group G, we say that c ∈ G is idempotent if c ◦ c = c. Similarly, if ǫ is a
nonzero constant, we say c is essentially idempotent if c ◦ c = ǫc.
With these definitions in mind, let’s consider the following tableau T .

1 2 3
4 5
6 7
8
9

Let’s define two sets of permutations, R(T ) and C(T ). We’ll let R(T ) be the set of permutations
of the rows of T in Sn . For example, (132) and (45) are in R(T ), but (147) is not. Similarly, we’ll
let C(T ) be the set of permutations of the columns of T in Sn . Elements of C(T ) include (1489)
and (257). Both R(T ) and C(T P ) are subgroups ofP Sn .
Next, we construct P = p∈R(T ) p and Q = q∈C(T ) δq q, where δq equals +1 if q is an even
permutation and −1 if q is an odd permutation.
From these sums we can build cT = P Q. In the group ring Rn = CSn , the element cT = P Q is
essentially idempotent and Rn c makes an irreducible representation of Sn ([7], p. 121).
What conclusions can we draw about the fillings of a particular Young diagram? Using our
knowledge of idempotents and representations, we conclude that the number of standard tableaux
for a given Young diagram with n boxes equals the dimension of the corresponding representation
of Sn ([7], p. 127).
Let’s return to our example of S3 . For two of the Young diagrams, we obtain only one filling.

1
2
1 2 3 3

For the third diagram, we obtain two possible fillings.

1 2 1 3
3 2

16
The first two diagrams correspond to our 1-dimensional representations for S3 , the trivial repre-
sentation and the sign representation. The third diagram, with two fillings, corresponds to our two
dimensional representation.
Hence, we find the Young tableaux assist us in determining the various dimensions of our
representations of Sn . This is extremely helpful, particularly for large values of n. Of course,
this correspondence is not to be confused with the correspondence between Young diagrams and
conjugacy classes.
Finally, we want to explore the possibilities Young diagrams offer us in our study of tensor
products. Let’s begin with a vector space V . We want to look at the tensor product

V ⊗n = V ⊗ V ⊗ ... ⊗ V.

Then V ⊗n is a linear representation of Sn by

α(v1 ⊗ v2 ⊗ · · · ⊗ vn ) = vα−1 (1) ⊗ vα−1 (2) ⊗ · · · ⊗ vα−1 (n) ,

where α ∈ Sn . Most likely, this tensor product is reducible. Our objective is to use Young diagrams
to find its irreducible subrepresentations.
To accomplish this goal, we will again use our subgroups of Sn , R(T ) and C(T ), based on
our Young diagram T with n boxes. This time, we will form the product b cT = QP . Again, bcT is
essentially idempotent. If we consider the product c
b V ⊗n , we find it is an irreducible representation
L T
of Sn . Additionally, V ⊗n = cT V ⊗n when we take the sum over all the standard tableaux
T b
T of n. Thus, we can generate our tensor product by taking the direct sum of our irreducible
subrepresentations. ([7], p. 146)
We see that we can use Young diagrams with tensor products in two ways: to find the irreducible
subrepresentation and to describe the full product in terms of its irreducible components. This is
extremely helpful for studying tensor products.

8 Conclusion
It is remarkable that Young diagrams can be applied to representation theory and tensor products.
On the surface, they seem to have no relationship to representations whatsoever, yet correspon-
dences exist with characters, conjugacy classes, and dimensions. These are just the beginning of the
power of Young tableaux–there are other applications involving more complicated manipulations
of the tableaux. Similarly, there are many more applications of Young diagrams to direct sums and
tensor products.
Through this study, I have learned a great deal about the surprising intricacies within the world
of mathematics. I have seen mathematics on the small scale: defining tableaux, building individual
character tables, and sorting permutations into conjugacy classes. More importantly, I have seen
how mathematicians can use tools, such as partitions, to research complicated topics such as group
theory. These studies have reminded me of the depth and complexity of mathematics, and that my
knowledge is very limited in comparison to the whole realm of mathematics.

References
[1] Anderson, Marlow and Todd Feil. A First Course in Abstract Algebra. 2nd ed. Chapman &
Hall/CRC, Boca Raton, 2005.

17
[2] Baker, Andrew. Representations of Finite Groups. University of Glasgow. 29 December 2008.
hhttp://www.maths.gla.ac.uk/∼ajbi

[3] Cameron, Peter J. Combinatorics: topics, techniques, algorithms. Cambridge University Press,
Cambridge, 1994.

[4] Diaconis, Persi. Group representations in probability and statistics. Institute of Mathematical
Statistics Lecture Notes—Monograph Series, 11. Institute of Mathematical Statistics, Hay-
ward, CA, 1988.

[5] Fulton, William. Young tableaux. With applications to representation theory and geometry.
London Mathematical Society Student Texts, 35. Cambridge University Press, Cambridge,
1997.

[6] Grove, Larry C. Groups and characters. Pure and Applied Mathematics. John Wiley & Sons,
Inc., New York, 1997.

[7] Miller, Willard, Jr. Symmetry groups and their applications. Pure and Applied Mathematics,
Vol. 50. Academic Press, New York-London, 1972.

[8] Murnaghan, F. D. On the Representations of the Symmetric Group. Amer. J. Math. 59 (1937),
no. 3, 437–488.

18

You might also like