Chapter 11

Sk and Tensor Representations

(Ref: Schensted Part II)

i1 ,··· ,ik
If we have an arbitrary
 tensor withk indices W we can act on it
1 2 ··· k
with a permutation P = so
a b ··· ℓ

(P w)i1 ,i2 ,··· ,ik = w ia ,ib ,··· ,iℓ .

Consider the algebra A formed by taking arbitrary linear combinations of the

different permutations, considered as operators acting on the space of k’th
rank tensors. This algebra can be constructed for any group, particularly
finite groups, and is called the group algebra. (this is not the Lie algebra!).
Note that this sum of permutations makes sense only as operators on a vector
space. It is not the composition of permutations. Also note that as A is an
algebra1 , one can both add and multiply (by composition) elements in A.
1
Definition: An algebra consists of a vector space V over a field F , together with a
binary operation of multiplication on the set V of vectors, such that for all a ∈ F and
α, β, γ ∈ V , the following are satisfied:
1. (aα)β = a(αβ) = α(aβ)
2. (α + β)γ = αγ + βγ
3. α(β + γ) = αβ + αγ
V is an associative algebra over F if, in addition,
4 (αβ)γ = α(βγ) for all α, β, γ ∈ V .

121
122. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

The group algebra is useful because it can extract the tensors of specified
symmetry. First consider tensors of rank 2. Writing 1I = 12 (1I + (1 2)) +
1
2
(1I − (1 2)) we can extract
1
sij = (1I + (1 2)) w ij
2
1
aij = (1I − (1 2)) w ij
2
and w ij = sij + aij is a decomposition into a symmetric tensor and an anti-
symmetric tensor.
The action of the permutations commutes with the SU(n) rotations on
the tensors, so a constraint on a tensor of the form Aw = 0 for some A ∈ A,
if it holds for one state of an irreducible representation of SU(n), will hold on
all states in that representation. Thus s and a are separate representations.
Now consider a rank 3 tensor w ijk , and define
1 X
sijk = P w ijk
6 P ∈S
3

1 X
aijk = (sign P )P w ijk
6 P ∈S
3

These are the totally symmetric and totally antisymmetric parts of w, but it
is not all of w. For example, suppose w 112 = w 121 = 1, w 211 = −2, all other
components zero. Then sijk and aijk are both zero. The rest is related
to the two-dimensional representation of S3 (see homework #3, problem
1). In general, there will be operators in A associated with the different
irreducible representations of Sk , which extract the corresponding irreducible
representations of SU(n).
So we now turn to the problem of finding the irreducible representations
of Sk .

11.1 Irreducible Representations of Sk

We know in general that the number of irreducible representations is the
number of conjugacy classes. So let us begin with that.
Any element
 of Sk can
 be written as a product of disjoint cycles. For
1 2 3 4 5
example, = (1 2 3)(4 5). This factorization is unique (re-
2 3 1 5 4
618: Last Latexed: April 25, 2017 at 9:45 123

member (1 2 3) = (2 3 1)) up to the order of the factors, which commute

because they are disjoint cycles.  
1 2 ··· k
Under conjugation by P = a cycle simply has its
P1 P2 · · · Pk
elements permuted. Thus P (i j k)P −1 = (Pi Pj Pk ). This is true for products
of cycles as well. Thus two permutations whose descriptions in terms of
disjoint cycles contain the same number of cycles of each length are conjugate,
and only those are. We describe the conjugacy class of
elements describable in
as ℓi11 ℓi22 · · · . Including one-cycles
terms of disjoint cycles, Ik of length ℓk ,P
for any element left unmoved, we have m im ℓm = k.
Example: S3

permutations class


1I ∈ 13
(1 2) = (1 2)(3) ∈ (2, 1)
(2 3), (1 3) ∈ (2, 1) as well
(1 2 3), (1 3 2) ∈ (3)

There is one conjugacy class for each partition of k. A partition of an

integer k is an unordered set of positive integers, possibly with repeats, which
add to k.
Example: How many classes2 are there in S5 ?

(5); (4, 1); (3, 1, 1) = (3, 12); (3, 2); (22 , 1); (2, 13); (15 )
answer: 7.
Thus we also know that there are that many irreducible representations,
although there is not a straightforward correspondance between the repre-
sentations and the conjugacy classes.
Define a Young graph for Sk as a set of k boxes arranged, left-justified,
in rows each of which is no longer than the preceeding. The lengths of the
rows provide a partition of k. So
2
The number of partitions of n is given by the partition function of number theory, p(n).
There are other things called partition functions, especially Z of statistical mechanics,
which is different. The number-theory one, also called the integer partition function,
arises also in counting statesQin string theory. PIt has the fascinating property that p(k)
k −1 k
has the generating function ∞ k=1 (1 − x ) = k=0 p(k)x , where we say p(0) = 1.
124. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

(5)
(3, 2)
(4, 1) (2, 13 ) (15 )
2
(2 , 1)
(3, 12)

There is one irreducible representation of Sk corresponding to each Young

graph.
A Young tableau is a Young graph with the numbers 1, 2, · · · , k inserted
in the boxes in some order, for example τ = 25 31 4 .
P
For each tableau we define an element of the group algebra, Pτ = P , where
the sum is over those permutations which permute the numbers within each
row but do not move them from one row to another. Here

Pτ = [1I + (2 3) + (3 4) + (2 4) + (2 3 4) + (2 4 3)] [1I + (5 1)] ,

which includes 12 of the P

120 permutations in S5 .
We also associate Qτ = (sign P )P where the sum includes only permuta-
tions which permute numbers in the same column but don’t move numbers
from one column to another. Thus

Qτ = [1I − (2 5)] [1I − (1 3)] .

Finally we define the Young operator Yτ = Qτ Pτ .

We see that the way to get a totally symmetric rank 5 tensor is to apply
Y to an arbitrary one while you get a totally antisymmetric tensor by
applying Y , with the numbers in any order in the boxes.

The Yτ corresponding to any Young tableau τ is almost, but not quite, the
element of the group algebra we want to extract irreducible representations.
We find a related set of basis vectors in the group algebra by using the
representations of Sk . Define

ℓη X η −1
eηij = Γ (P )P,
k! P ∈S ji
k

where η is the Young graph corresponding to an irreducible representation

of Sk , and ℓη is the dimension of that representation. The sum is over all the
permutations.
618: Last Latexed: April 25, 2017 at 9:45 125

For , ℓη = 1, Γ = 1, and e11

=Y1 2 3 4 5
which is also equal to any other Young operator for a tableau in .

For , ℓη = 1, Γ = sign P , e11 = Y 1 . Other Young operators in differ

2
3
only in sign, e.g. Y 2 = sign(1 2) · Y 1 = −Y 1 .
1 2 2
3 3 3
The e’s have some marvelous properties. They form vector spaces trans-
forming as irreducible representations under Sk separately from the right and
from the left: For3 Q ∈ Sk ,
ℓη X η
ℓη X η X η

Qeηij = Γji P −1 QP = Γmi (Q) Γjm P −1 Q−1 QP
k! P k! m P
ℓη X η X η
X η
= Γ (Q) Γjm R−1 R = Γmi (Q)eηmj
k! m mi R m

where we again used the rearrangement theorem. Thus Q acts just the way
you’ld expect for a basis vector ei of representation η to transform, for each
fixed j. P
From the other side, eηij Q = m Γηjm(Q)eηim .
We say that the set eηij is a two sided ideal (or invariant subalgebra) of
the group algebra over Sk .
This gives the e’s an interesting algebra:
ℓη ℓη ′ X η
′

eηij eηmn =
′

2
Γji P −1 Γηnm P ′ −1 P P ′
(k!)
P,P ′ ∈Sk
ℓη ℓη ′ X η −1

X η′ −1

η′
= Γ ji P Γ np R Γpm (P ) R
(k!)2 P,R∈S p
k

ℓη X X
Γηij∗ (P )
′ ′
= Γηpm (P ) eηpn by unitarity
k! P ∈S p
k

= δηη′ δjm eηin by the great orthogonality theorem

We may also show that the diagonal elements eηii form a decomposition
of the identity. From the great orthogonality theorem “transposed” ,
X ℓη η ∗
δGG′ = Γij (G′ )Γηij (G)
ijη
k!
3
Note P and Q are any elements of Sk , and are not related to Pτ and Qτ defined earlier.
126. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

we can write the identity element of Sk as

X X ℓη η ∗
1I = δG−1 ,1I G = Γij (1I)Γηij (G−1 )G
G ijηG
k!
X ℓη η X η
= Γii (G−1 )G = eii
iηG
k! ηi

so X
1I = eηii
ηi

Thus the whole algebra is spanned by these two sided ideals. In particular,
the Yτ are contained in the corresponding eηij (an ℓ2η dimensional algebra).
In fact, the space spanned by eηij is also spanned by Qi sij Pj , where Qi and
Pi are the antisymmetrizers and symmetrizers of a set of standard tableaux
for η, which means tableaux in which the numbers increase left to right in
each row, and also top to bottom in each column. Thus 13 2 and 12 3 are
standard tableaux, but 31 2 , 21 3 and 23 1 are not. Here sij is the permutation
such that τi = sij τj . Each of these spaces has dimension ℓ2η , with ℓη equal to
the number of standard tableaux, so
The dimension of Γη is the number of standard tableaux of η.
Counting all possibilities is tedious, so we have a magic formula in terms
of hooks.
For each box b in a Young graph with k boxes, define the hook of b,
gb = 1 plus the number of boxes directly to the right plus the number of
boxes directly beneath. Then
k!
ℓη = Y .
gb
b

Example: In the Young graph I have placed the corresponding hooks (this is
not a Young tableau)
7 5 3 2
6 4 2 1 11!
ℓ= = 1320.
3 1 7 · 6 · 5 · 4 · 32 · 22 · 13
1
It would be hard to count this explicitly. For our more reasonable case ,
3 1 3!
1
gives ℓ = 3
= 2.
618: Last Latexed: April 25, 2017 at 9:45 127

11.2 Representations of SU(n)

We now turn to the extraction of arbitary representations of SU(n). Georgi

discusses the fundamental weights of SU(n), and shows that an arbitrary
representation can be found from a tensor product of an adequate number of
defining representations. The problem is to extract from the tensor product of
k defining representations ⊗Nk the irreducible pieces. We have seen that this
can be done by demanding that elements of the permutation group algebra
vanish. If we impose eηii w = 0 for all η and i save one, that is equivalent to
projecting out our representation

T a1 a2 ···ak = (eηii w)a1 a2 ···ak no sum on i

for one particular representation η and one basis vector i.

The different i generate equivalent representations. The different η’s,
however, each correspond to a different (inequivalent) representation of SU(n).
Before doing more formal arguments, we will do an example. Consider
three spin 12 objects, or the tensor product of three defining representations
of SU(2). We will extract from this 8 dimensional state space the piece e .
11
From the problem you did for homework (#3, problem 1),

2X e

e11 = Γ 11 P −1 P
6 P
 
1 1 1 1 1
= 1I + (1 2) − (2 3) − (1 3) − (1 2 3) − (1 3 2) ,
3 2 2 2 2

Let this act on the basis vectors which we expand as ↑= e1 , ↓= e2 .

128. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

v e v
11

1 1
↑↑↑ 3 ↑↑↑ + 3 ↑↑↑ − 16 ↑↑↑ − 61 ↑↑↑ − 61 ↑↑↑ − 16 ↑↑↑ = 0
1 1
↑↑↓ 3 ↑↑↓ + 3 ↑↑↓ − 16 ↑↓↑ − 61 ↓↑↑ − 61 ↓↑↑ − 16 ↑↓↑ = 2
3 ↑↑↓ − 31 ↑↓↑ − 13 ↓↑↑
1 1
↑↓↑ 3 ↑↓↑ + 3 ↓↑↑ − 16 ↑↑↓ − 61 ↑↓↑ − 61 ↑↑↓ − 16 ↓↑↑ = − 31 ↑↑↓ + 61 ↑↓↑ + 16 ↓↑↑
1 1
↓↑↑ 3 ↓↑↑ + 3 ↑↓↑ − 16 ↓↑↑ − 61 ↑↑↓ − 61 ↑↓↑ − 16 ↑↑↓ = − 31 ↑↑↓ + 61 ↑↓↑ + 16 ↓↑↑
1 1
↑↓↓ 3 ↑↓↓ + 3 ↓↑↓ − 16 ↑↓↓ − 61 ↓↓↑ − 61 ↓↑↓ − 16 ↓↓↑ = 1
6 ↑↓↓ + 61 ↓↑↓ − 13 ↓↓↑
1 1
↓↑↓ 3 ↓↑↓ + 3 ↑↓↓ − 16 ↓↓↑ − 61 ↓↑↓ − 61 ↓↓↑ − 16 ↑↓↓ = 1
6 ↑↓↓ + 61 ↓↑↓ − 13 ↓↓↑
1 1
↓↓↑ 3 ↓↓↑ + 3 ↓↓↑ − 16 ↓↑↓ − 61 ↑↓↓ − 61 ↑↓↓ − 16 ↓↑↓ = − 31 ↑↓↓ − 31 ↓↑↓ + 23 ↓↓↑
1 1
↓↓↓ 3 ↓↓↓ + 3 ↓↓↓ − 16 ↓↓↓ − 61 ↓↓↓ − 61 ↓↓↓ − 16 ↓↓↓ = 0

1
Notice this only results in one state of Jz = 2
and one of Jz = − 12 . So
e11 projects out a 2-dimensional s = 12 state. e22 would project out an
orthogonal spin 21 . Thus the tensor product of three spin 1/2’s is a spin
3/2 (the totally symmetric part, e11 ) and two spin 1/2 representations,
2 × 2 × 2 = 4 + 2 + 2.
Having completed this trivial but tedious example of the simple case of
SU(2) and , we are ready for some abstract reasoning.
Now we consider the general case of Nk . The basis vectors which are
mixed by the permutations are only those with the same number of indices
equal to 1, and the same number equal P to 2, etc.. Consider the subspace
with ri of the indices equal to i, with ri = k, each ri = 1, . . . , N.
This subspace S ~r is spanned by the basis vector

e = e1 ⊗ e1 · · · ⊗ e1 ⊗ e2 · · · ⊗ e2 · · · ⊗ eN · · · ⊗ eN ,
| {z } | {z } | {z }
r1 times r2 times rN times
together with all permutations P e, for P ∈ Sk . If all the indices are different,
all ri = 0 or 1, all of the permutations are inequivalent, and we get a k!
dimensional space. But if the ri ’s are not all ≤ 1, there is a subgroup P ⊂ Sk
with B e = e for
PB ∈ P. In fact, P = Sr1 × Sr2 × · · · SrN .
Let PP = B∈P B which is a element of the group algebra A. Then
while the subspace S ~r is spanned by {P e|P ∈ Sk } it is also spanned by
{P PP e|P ∈ Sk }.
618: Last Latexed: April 25, 2017 at 9:45 129

We now want to extract from S ~r the piece projected out by eηii . The
products {eηii P } for all P are just sums of multiples of eηij , for all j (by
p. 125) so we want to know the dimensionP of the space {eηij PP |j = 1, ℓη }. As
eη is a two-sided ideal, this space is h k eηik bk i, so the dimensionality depends
on the constraints on bk . If they were all independent, they would form an ℓη
dimensional space. But there are constraints. For B ∈ P, PP B = PP . Let’s
be more explicit:
X X X
eηij PP = eηin bnj = eηij PP B = bnj eηin B = Γηnm (B)eηim bnj .
n n nm

P
The eηim are linearly independent, so bnj = m bmj Γηmn (B), for B ∈ P. To
find out how many degrees of freedom survive this constraint for each j,
observe that Γηmn (B) forms a reducible representation of the subgroup P. So
we can write M
Γη (B) = U Γǫ (B) U −1
ǫ

where Γǫ are irreducible representations of P. Now if c = bU,

M M 
b = bΓη (B) = bU Γǫ (B)U −1 =⇒ c = c Γǫ (B) .

The vector c breaks up into pieces for each representation ǫ, with cǫ Γǫ (B) =
cǫ for all B ∈ P. This is possible for nonzero c only if ǫ is the identity
representation, as the representations are irreducible.
Therefore the dimensionality of the space eηii P e is the number of times,
γη , that the identity representation of P is contained in Γη .
But the number of times the representation i is contained in Γη is

1 X i∗ 1 X η
γη = ai=1I = χ (B)χη (B) = χ (B)
gP B∈P gP B∈P
Q
for i=identity, where gP is the number of elements in P, which is ri !
Example: η = . From homework, and recalling χ = Tr Γ,

χ = 2 for B = 1I
χ = 0 for B = (1 2), (1 3), or (2 3)
χ = −1 for B = (1 2 3), or (1 3 2)
130. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

Consider the space starting from e1 ⊗ e1 ⊗ e2 .

1
P = {1I, (1 2)}, gP = 2, γη = (2 + 0) = 1
2
so e11 generates only one state from the three-dimensional space S ~r .
From e1 ⊗ e1 ⊗ e1 , P = S3 , gP = 6, γη = 61 (2 − 1 − 1) = 0 so we get no state
here.
If all vectors are unequal, say e1 ⊗ e2 ⊗ e3 for SU(n > 2), P = 1I, gP = 1,
γη = 12 = 2.
For SU(N), there are N states of the form ei ⊗ ei ⊗ ei , not contributing
anything to η = . There are N(N − 1) states ei ⊗ ei ⊗ ej with i 6= j, each
contributing one state, so from these we get N(N − 1) states. There are also
N(N − 1)(N − 2)/6 states of the form ei ⊗ ej ⊗ ek , with i < j < k, each
contributing 2 states, so the dimension of is

1 N(N 2 − 1) 2 for N = 2
N(N − 1) + N(N − 1)(N − 2) = =
3 3 8 for N = 3

Let’s work another example, for SU(N). As we need the characters

for this representation, let’s take them from Schensted: χ = 3 for 1I ∈ (14 ),
[1 element]; χ = 0 for (3, 1), [8 elements]; χ = 1 for (2, 12), [6 elements];
χ = −1 for (2, 2), [3 elements] and for (4), [6 elements].
Enumerating the basis states in the various partitions, and multiplying
χ(B) by their number for those within P, we find

indices subspace P γ 14 (3, 1) (2, 12) (2, 2) (4)

η
1
 
all i ei ei ei ei S4 1 · 3+8 · 0+6 · 1+3 · (−1)+6 · (−1) =0
24
 
1
i 6= j ei ei ei ej S3 1 · 3+2 · 0+3 · 1+0 · (−1)+0 · (−1) =1
6
 
1
i<j ei ei ej ej S2 ×S2 1 · 3+0 · 0+2 · 1+1 · (−1)+0 · (−1) =1
4
 
1
i6=j<k6=i ei ei ej ek S2 1 · 3+0 · 0+1 · 1+0 · (−1)+0 · (−1) =2
2
 
1
all 6= ei ej ek eℓ (1I) 1
1 · 3+0 · 0+0 · 1+0 · (−1)+0 · (−1) =3

Rewriting this with only the results for γη to allow room for counting
index choices and states, we have
618: Last Latexed: April 25, 2017 at 9:45 131

indices subspace P γ # index choices # states

η
all i ei ei ei ei S4 0 N 0
i 6= j ei ei ei ej S3 1 N(N−1) N(N−1)
i<j ei ei ej ej S2 ×S2 1 N(N−1)/2 N(N−1)/2
i6=j<k6=i ei ei ej ek S2 2 N(N−1)(N−2)/2 N(N−1)(N−2)
N
N

all 6= ei ej ek eℓ (1I) 3 4
3 4

So the total dimensionality of for SU(N)

is
N(N − 1)
0 + N(N − 1) + + N(N − 1)(N − 2)
2
N(N − 1)(N − 2)(N − 3)
+3
4!
 
3 1 (N + 2)!
= N(N − 1) + (N − 2) + (N − 2)(N − 3) = .
2 8 8 (N − 2)!

For SU(3), N = 3, the total dimension is

5!/(8 · 1!) = 15.

We now know how to extract the irreducible representations or just to

count their dimensionality. Now it is time for magic.
The number γη of states extracted by eηii from S ~r , the space spanned by
N
P ( eri i ) by all P ∈ Sk , is given by the number of ways one can place r1
1’s, r2 2’s, · · · in the Young graph so that in each row the numbers do not
decrease, and in each column they increase. This is called a permissible
placement.
To see how this works, let’s check it out on for SU(N). For r1 = 4
there is no way to avoid two 1’s in the same column, so γ = 0.
For r1 = 3 and r2 = 1, the 2 has to be in the second row, so there is only one
way, γ = 1.
For r1 = r2 = 2, the only possibility is 12 1 2 , γ = 1.
For r1 = 2, r2 = r3 = 1, we have 13 1 2 and 21 1 3 , so γ = 2.
For r1 = r2 = r3 = r4 = 1, we have 14 2 3 , 12 3 4 , 13 2 4 , and γ = 3.
Note in our previous method, it was clear that these numbers only de-
pended on the set {ri } and not on the order. This is now not obvious.
132. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

Consider r1 = 1, r2 = 2, r4 = 1. Then we have 12 2 4 and 14 2 2 , so again γ = 2,

as for r1 = 2, r2 = r3 = 1.
Now to count the dimensionality of an irreducible representation of SU(N)
belonging to the Young graph η, we must sum, over all choices ri , the corre-
sponding γη . But for each choice of ri the γη is the number of ways of placing
the indices in the graph in a permissible fashion. So the dimension of the
full irreducible representation of SU(N) is the number of ways of placing k
integers, chosen from 1, 2, , . . . N (repeats allowed) in a permissible fashion
in η.
Example for SU(3): 12 1 , 13 1 , 12 2 , 13 2 , 12 3 , 13 3 , 23 2 , 23 3 , for a total
of 8 dimensions.
There is an easier method of finding the dimensionality. For each box,
associate the value (N + column number − row number). Then divide by the
hook of that box. The dimension of the representation is the product of
these quotients over all the boxes.

= NN−1 N +1 = N ( N3
2 − 1)
3 1
Examples: Dim 1

Dim = N N +1 N +2
N −1
4 2 1
1 = ( N + 2) !
8( N − 2) !

Note: If the first column of a graph has N boxes, the hook of each box in
column 1 is equal to the (N + column number − row number) of the last box
in that row. Thus eliminating the first row does not change the dimension.
In fact, it does not change the representation either. This is because a totally
antisymmetric tensor with N indices is invariant.
Thus in SU(2), = , as we saw in detail. It also means that for
SU(N), we needn’t consider representations with N or more rows (except
perhaps to indicate the identity representation by one column of N boxes).