Tensor Method in SU(n)

December 20, 2012

Contents
1 Tensor analysis in O(3) 1
1.1 Rotation in R
3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Tensor operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Transformation Law of Tensors in SU(N) 3
2.1 Invariant tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Irreducible Representations and Young Tableaux 5
3.1 Permutation symmetry and tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Young tableaux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4 Reduction of the product representations 8
Tensor Analysis
The analysis we have discussed for ol (2) and ol (3) shows that, as thr group gets larger, the elmentary
techniques used to dissect the representation structure becomes very complicate. The tensor method we will discuss
here provides a handle which is very useful for low rank representations.
1 Tensor analysis in O(3)
1.1 Rotation in R
3
Rotation: coordinate axes are xed and the physical system is undergoing a rotation. Let r
0
a
, r
b
be the components
of new and old vectors. Then we have
r
0
a
=

b
1
ab
r
b
where 1
ab
are elements of matrix which represents rotation. For example, the matrix for the rotation about .axis
is of the form,
1 =
_
_
cos 0 sin0 0
sin0 cos 0 0
0 0 1
_
_
Note that the relation between r
0
a
and r
b
is linear and homogeneous.
Important properties of transformation:
1. 1 is an orthogonal matrix,
11
T
= 1
T
1 = 1, or 1
ab
1
ac
= c
bc
, 1
ab
1
cb
= c
ac
More conviently, we will write the orthogonality relations as,
1
ab
1
cd
c
ac
= c
bd
, 1
ab
1
cd
c
bd
= c
ac
i.e. in the product of 2 rotation matrix elements, making row(or coulm) indices the same and summed over
will give Kronecker c.
1
2. The combination
!
r
2
= r
a
r
a
is invariant under rotations,
r
0
a
r
0
a
= 1
ac
1
ab
r
c
r
b
= r
b
r
b
This is the same as saying the legth of a vector is invariant under rotation. We can can generalize to the case
of 2 arbitrary vectors,
!
,
!
1 with transformation property,

0
a
= 1
ab

b
, 1
0
c
= 1
cd
1
d
Then
!

!
1 =
a
1
a
is invariant under rotation. It is more convienent to write this as
!

!
1 =
a
1
a
=
a
1
b
c
ab
which is sometimes called the contraction of indices.
3. Transfomation of the gradient operators is
0
0r
0
a
=
0
0r
c
0r
c
0r
0
a
From r
b
=
_
1
1
_
ba
r
0
a
, we get then
0
0r
0
a
=
_
1
1
_
ca
0
0r
c
Thus gradient operator transforms by
_
1
1
_
T
. However, for rotations, 1 is orthogonal,
_
1
1
_
T
= 1,
0
0r
0
a
= 1
ac
0
0r
c
i.e. 0

=
0
0r
a
tranforms the same way as r
a
.
Note that for the case the coordinate transformation is not orthogonal the corrdinates and the partial derivative
transform dierently.
1.2 Tensors
Tensors are essentially those quantities which transform in the same way as the products of vectors under the
coordinate transformation, the rotation in our case. Suppose we have two vectors, i.e. they have the transformation
properties,

a

0
a
= 1
ab

b
, 1
c
1
0
c
= 1
cd
1
d
then the product will have the transformtion,

0
a
1
0
c
= 1
ab
1
cd

b
1
d
The second rank tensors are those objects which have the same transformation properties as the product of 2
vectors, i. e.,
T
ac
T
0
ac
= (1
ab
1
cd
) T
bd
Denition of :th rank tensors (Cartesian tensors)
T
i1i2:::
T
0
i1i2:::in
= (1
i1j1
) (1
i2j2
) (1
injn
) T
j1j2jn
Note again that these transformations are linear and homogeneous which implies that
i) T
j1j2jn
= 0, )or all ,
m
then they all zero in other coordinate system.
2
1.3 Tensor operations
1. Mulplication by constants
(cT)
i1i2:::in
= cT
i1i2:::in
2. Add tensors of same rank
(T
1
+T
2
)
i1i2:::in
= (T
1
)
i1i2:::in
+ (T
2
)
i1i2:::in
3. Multiplication of 2 tensors
(oT)
i1i2:::inj1j2jm
= o
i1i2:::in
T
j1j2jm
This will give a tensor of rank which is the sum of the ranks of 2 constituent tensors.
4. Contaction
o
abc
T
ae
= o
abc
T
de
c
ad
3rd rank tensor
This will reduce the rank of tensor by 2.
5. Symmetrization
if T
ab
2nd rank tensor = T
ab
T
ba
are also 2nd rank tensors
6. Special numerical tensors
11
T
= 1, =1
ij
1
kj
= c
ik
, or 1
ij
1
kl
c
jl
= c
ik
This means that c
ij
can be treated as 2nk rank tensor. Similarly,
(det 1) -
abc
= -
ijk
1
ai
1
bj
1
ck
-
abc
a 3rd rank tensor. Useful identities for -
abc
-
ijk
-
ijl
= c
kl
, -
ijk
-
ilm
= c
jl
c
km
c
jm
c
kl
One can use these numerical tensors to construct other tensors. For example,
-
ijk

i
1
k
vector
r
i
r
j
c
ij
r
2
second rank tensor
More general notation for tensor transformation (Jackson),
r
0
a
= 1
ab
r
b
, =1
ab
=
0r
0
a
0r
b
Then we can wrtie
r
0
a
=
0r
0
a
0r
b
r
b
This can be used for more general coordinate transfomations other than the simple rotation and is used extensively
in general relativity.
2 Transformation Law of Tensors in SU(N)
The ol (:) group consists of : : unitary matrices with unit determinant. We can regard these matrices as linear
transformations in an :dimensional complex vector space C
n
. Thus any vector
c
i
= (c
1
, c
2
, c
n
)
in C
n
is mapped by an ol (:) transformation l
ij
, as
c
i
c
0
i
= l
ij
c
j
(1)
Thus c
0
i
also belong to C
n
, with ll
y
= l
y
l = 1 and with det l = 1. Clearly for any two vectors we can dene a
scalar product
(c, c) = c

i
c
i
3
which is invariant under ol (:) transformation. The transformation law for the conjugate vector is given by,
c

i
c
0
i
= l

ij
c

j
= c

j
l
y
ji
(2)
Here we see that c
i
and c

i
transform dierently under the unitary transformation. It is convenient to introduce
upper and lower indices to write
c
i
= c

i
, l
j
i
= l
ij
, l
i
j
= l

ij
Thus complex conjugation just change the lower indices to upper ones, and vice versa. In these notation, the
transformation law in Eqs(1,2) become
c
i
c
0
i
= l
j
i
c
j
and
c
i
c
0i
= l
i
j
c
j
The ol (:) invariant scalar product can be written as
(c, c) = c
i
c
i
and the unitary condition becomes
l
i
k
l
k
j
= c
i
j
(3)
where the Kronecker delta is dened as
c
i
j
= c
ij
=
_
1 if i = ,
0 otherwise
Note that in this notation the summation is always over a pair of upper and lower indices. We call this a contraction
of indices. The c
i
are the basis for the ol (:) dening representation (also called the fundamental or vector
representation and denoted as n), while the c
i
are the basis for the conjugate representation, n
*
.
Higher-rank tensors are dened as those quantities which have the same transformations properties as the direct
products of vectors. Thus tensors generally have both upper and lower indices with the transfomation law,
c
0i1i2ip
j1j2jq
=
_
l
i1
k1
l
i2
k2
l
ip
kp
__
l
l1
j1
l
l2
j2
l
lq
jq
_
c
k1k2kp
l1l2lq
They correspond to the basis for higher-dimensional representations.
2.1 Invariant tensors
The Kronecker delta and Levi-Civita symbol are invariant tensors under ol (:) transformations. They play impor-
tant role in the study of irreducible tensors.
1. From the unitarity condition of Eq(3) we immediately have
c
i
j
= l
i
k
l
l
j
c
k
l
(4)
Hence c
i
j
,even though do not change under the ol (:) transformations, behaves as if they are second rank
tensors. They can be used to contract indices of other tensor to produce a tensor of lower rank. For example,
if c
k
ij
is a 3rd rank tensor,
c
k
ij
c
0k
ij
= l
k
a
l
b
i
l
c
j
c
a
bc
the contracting with c
k
i
gives
c
0k
ij
c
k
i
= c
k
i
l
k
a
l
b
i
l
c
j
c
a
bc
= l
c
j
c
b
a
c
a
bc
where we have used Eq(4). This gives a tensor of rank 1( vector).
2. The Levi-Civita symbol is dened as the totally antisymmetric quantity,
-
i1i2in
= -
i1i2in
=
_
_
_
1 if (i
1
i
2
i
n
) is an even permutation of (1, 2, :)
1 if (i
1
i
2
i
n
) is an odd permutation of (1, 2, :)
0 otherwise
This is also an invariant tensor, because from the property of the determinant we have
(det l) -
i1i2in
= l
j1
i1
l
j2
i2
l
jn
in
-
j1j2jn
Since det l in ol (:) , -
i1i2in
can be treated as : t/ rank tensor. We can use this to change the rank of a
tensor. For example,
c
i2in
-
i1i2in
~ c
i1
is a vector.
4
3 Irreducible Representations and Young Tableaux
3.1 Permutation symmetry and tensors
Generally the tensors we have just dened are basis for reducible representation of ol (:) . To decompose them
into irreducible representations we use the following property of these tensors. The permutation of upper(or lower)
indices commutes with the ol (:) transformations, as the latter consists of product of identical l
ij
s(or l

ij
s). We
will illustrate this with a simple example. Consider the second rank tensor c
ij
whose transformation is given by
c
0
ij
= l
a
i
l
b
j
c
ab
Since ls are the same, we can relabel the indices to get
c
0
ji
= l
b
j
l
a
i
c
ba
Thus the permutation of indices in the tensor does not change the transformation law. If 1
12
is the permutation
opeator which interchanges the two indices,
1
12
c
ij
= c
ji
,
then 1
12
commutes with the group transformation,
1
12
c
0
ij
= l
a
i
l
b
j
(1
12
c
ab
)
This property can be used to decompose c
ij
as follows. First we form eigenstates of the permutation operator 1
12
by symmetrization or antisymmetrization,
o
ij
=
1
2
_
c
ij
+c
ji
_
,
ij
=
1
2
_
c
ij
c
ji
_
so that
1
12
o
ij
= o
ij
, 1
12

ij
=
ij
In group theory language, o
ij
form basis of an one-dimensional representation of the permutation group o
2
and
ij
the basis for another representation. It is clear that o
ij
, and
ij
will not mix under the ol (:) transformations,
o
0
ij
= l
a
i
l
b
j
o
ab
,
0
ij
= l
a
i
l
b
j

ab
This shows that the second rank tensor c
ij
decomposes into o
ij
, and
ij
in such a way that group transformations
never mix parts with dierent symmetries. It turns out that o
ij
, and
ij
can not be decomposed any further and
they thus form the basis of irreducible representations of ol (:) . This can be generalized to tensors of higher rank
(hence the possibility of mixed symmetries) with the result that the basis for irreducible representations of ol (:)
correspond to tensors with denite permutation symmetry among (the positions of) its indices.
3.2 Young tableaux
The task of nding irreducible tensors of an arbitrary rank ) (i.e. number of indices) involves forming a complete set
permutation operations on these indices. The problem of nding irreducible representations of the permutation group
has a complete solution in terms of the Young tableaux. They are pictorial representations of the permutation
operations of ) objects as a set of ) boxes each with an index number in it. For example, for the second rank tensors,
the symmetrization of indices i and , in o
ij
is represented by i , ; the antisymmetrization operation in
ij
is
represented by
i
,
. For the third rank tensors c
ijk
, we have
i , / for totally symmetric combination o
ijk
=
1
6
_
c
ijk
+c
jki
+c
kij
+c
jik
+c
kji
+c
ikj
_
(5)
i
,
/
, for the totally antisymmetric combination
ijk
=
1
6
_
c
ijk
+c
jki
+c
kij
c
jik
c
kji
c
ikj
_
(6)
i ,
/
, for the tensor with mixed symmetry
ij;k
=
1
4
_
c
ijk
+c
jik
c
kij
c
kji
_
(7)
5
A general Young tableau is shown in the following gure. It is an arrangement of ) boxes in rows and columns
such that the length of rows should not increase from top to bottom: )
1
_ )
2
_ )
3
and )
1
+)
2
+)
3
= ). Each
box has an index i
k
= 1, 2, :.
i
1
i
2
i
f1
i
f1+1
i
f2
.
.
.
i
:::
To this tableau we associate the tensor,
c
i1i
f
1
;
i
f
1
+1

with the following operations.

1. First apply symmetrization operators which leave invariant the indices appearing in each row.
2. Then apply antisymmetrization operators which leave indices in each column invariant.
A tableau where the index numbers do not decrease when going from left to right in a row and always increase
from top to bottom is a standard tableaux. , For example, the : = 3 mixed symmetry tensors are given here with
the respective Young tableaux.
1 1
2
2 (c
112
c
211
) ,
1 1
3
, 2 (c
113
c
311
) ,
1 2
2
(c
122
c
221
)
1 3
3
(c
133
c
331
) ,
2 3
3
(c
233
c
332
) ,
2 2
3
2 (c
223
c
322
)
1 2
3
(c
123
+c
213
c
321
c
312
) ,
1 3
2
(c
132
+c
312
c
231
c
213
)
where we have make used of the tensor
ij;k
given in Eq (7). Thus we have 8 indepedent tensors which correspond to
the 8dimensional irrep we discussed before. Note that non-standard tableaux give tensors that, by symmetrization
or antisymmetrization, either vanish or are not independent of the standard tableaux. Thus for a given pattern of
the Young tableaux the number of independent tensors is equal to the number of standard tableaux which can be
formed. It is not hard to see that this number for the simplest case of a tensor with / antisymmetric indices is
.
.
.
_

_
/,
_
:
/
_
=
:(: 1) (: / + 1)
1 2 3 /
and that for a tensor with / symmetric indices the number is

. .
_
: +/ 1
/
_
=
:(: + 1) (: +/ 1)
1 2 3 /
/
One should note thath because of antisymmetrization there are not more than : rows in any Young tableaux. Also if
there are : rows, we can use -
i1in
to contract the indices in the columns with : entries. Pictorially, we can simply
cross out anu columns with : rows without changing the transformation property of the tensor.
Fundamental theorem (See for example, Hammermesh 1963.) A tensor corresponding to the Young tableaux
of a given pattern forms the basis of an irrep of ol (:) . Moreover if we enumerate all possible Young tableaux under
the restriction that there should be no more than : 1 rows, the corresponding tensors form a complete set, in the
sense that all nite-dimensonal irreps of the group are counted only once.
We next give two formulae of the dimensionalitof irreps. If the Young tableaux is characterized by the length
of its rows ()
1
, )
2
, )
n1
) , dene the length dierences of adjacent rows as
`
1
= )
1
)
2
, `
2
= )
2
)
3
, `
n2
= )
n2
)
n1
, `
n1
= )
n1
,
6
The dimension of an ol (:) irreps will then be the number of standard tableaux for a given pattern and is given by
d (`
1
, `
2
, `
n1
) = (1 +`
1
) (1 +`
2
) (1 +`
n1
) (8)

_
1 +
`
1
+`
2
2
__
1 +
`
2
+`
3
2
_

_
1 +
`
n2
+`
n1
2
_

_
1 +
`
1
+`
2
+`
3
3
__
1 +
`
2
+`
3
+`
4
3
_

_
1 +
`
n3
+`
n2
+`
n1
3
_

_
1 +
`
1
+`
2
+ +`
n1
: 1
_
On can easily check that the special results of totally symmetric and anti-symmetric tensors are recovered from this
formaula. Here we give some simple examples.
Example 1. ol (2) group. The Young tableaux can have only one row
d (`
1
) = (1 +`
1
)
Thus `
1
= 2,. It follows that
doublet, triplet,
Example 2. ol (3) group. The Young tableaux can now have two rows and
d (`
1
, `
2
) = (1 +`
1
) (1 +`
2
)
_
1 +
`
1
+`
2
2
_
This equivalent to the other formula for the dimensionality of ol (:) representations with the identication, `
1
= j
and `
2
= . Simple cases are given below
(1, 0) 3, (2, 0) 6, (3, 0) 10,
(0, 1) 3

, (0, 2) 6

, (1, 1) 8 ,
The formula given in Eq(8) is rather cumbersome to use for large value of :; in such case the second formula is
perhaps more useful. For this we need to introduce two denitionshook length and distance to the rst box.
For any box in the tableaux, draw two perpendicular lines, in the shape of a hook, one going to the right and another
going downward. The total number of boxes that this hook passes, including the original box itself, is the hook
length (/
i
) associated with it/ box. For example,
/
i
= 3, /
i
= 1
The distance to the rst box (1
i
) is dened to be the number of steps going from the box in the upper left-handed
corner of the tableaux to the ith box with each step toward the right counted as +1 unit and each downward step
as 1 unit. For example, we have
0 1 2
1 0
2
The dimension of the ol (:) irrep associated with the Young tableaux is given by
d =

i
_
: +1
i
/
i
_
where the products are taken over all boxes in the tableaux. We illustrate this with a simple example
Young tableaux , hook lengths
3 1
1
, distance to the rst box
0 1
1
and the dimension is
d =
_
:
3
_
_
: + 1
1
__
: 1
1
_
=
:
_
:
2
1
_
3
This gives d = 8 for : = 3 as expected.
7
4 Reduction of the product representations
One of the most useful application of ol (:) irrep with Young tableaux is the decomposition of the product repre-
sentations. To nd the irrep in the product of two factors,
1. In the smaller tableaux, assign the same symbol, say a, to all boxes in the rst row, the same symbol / to all
boxes in the second row, etc.
a a a
/ / /
c c

2. Attach boxes labelled by a to the second tableaux in all possible ways subjected to the rules that no two a
0
s
appear in the same column and that the resultant graph is still a Young tableaux (i.e. the length of rows does
not increase from top to bottom and there are not more than n rows, etc.) Repeat this process with /
0
s ..etc.
3. After all symbols have been added to the tableaux, these added symbols are then read from right to letft in
the rst row, then the second row in the same order,. . . and so forth. This sequence of symbols aa//ac ,
must form a lattice permtutation. Thus, to left of any symbol there are not fewer a than / and no fewer /
than c, etc.
We consider two examples in the ol (3) group.
Example 1.
a =
a
+ a
which corresponds to
3 3 = 3

+6
Example 2.
a a
/
= 8 8
Firs step:
a
+
a
+
a
Second step:
a a
+
a
a
+
a
a
+ a
a
Third step:
a a
/
+
a a
/
+
a
a /
+
a
a
/
+
a
/
a
+ a
a /
27 10 10

8 8 1 (9)
This gives the usual result
8 8 = 1 +8 +8 +10 +10

+27
In terms of tensor analysis we have the following relations. If
n
i
~ 3,
j
~ 3

then
n
i

j
=
_
n
i

1
3
c
j
i
n
k

k
_
+
1
3
c
j
i
n
k

k
This corresponds to
3 3

= 8 +1
8
Similarly, if
n
i
~ 3, n
j
~ 3
then
n
i
n
j
=
1
2
(n
i
n
j
+n
j
n
i
) +
1
2
(n
i
n
j
n
j
n
i
)
which corresponds to
3 3 = 6 +3