96.

Last Latexed: March 22, 2012 at 10:47

Joel A. Shapiro

a representation of arbitrary maximal weight ~ =

direct product
O

Di

qi

P i i
q ~

by reducing the

Chapter 8

8.0.1

Representations of Lie Groups

For SU(3), the positive roots are T+ , V+ , and U , of which the last two are
simple.
U = E1/2,3/2 ,
V+ = E1/2,3/2 ,

SU(3) Multiplets

so =

Let us label the simple roots

~ , i = 1, . . . , m. They and their conjugates generate (not linearly) any element of the Lie algebra. Thus any representation
can be determined by how it behaves under these roots1 .
The weights have an ordering, so any finite dimensional representation
has a greatest weight ~max . The weight ~ of a basis element is ~max if and
i ~max /(i)2 , which
only if ~ +
~ i is not a weight for each i . Define q i = 2~
i
can be arbitrary nonnegative integers. As the
~ are linearly independent
and complete, the {q i } and ~max determine each other.
Define m vectors ~ j such that
2~
i ~ j
= ij .
(i )2
These are called the fundamental weights of the Lie algebra corresponding
(as maximal weights) to a set of representations D j called the fundamental
representations
Warning: ~ i is a vector, the ith fundamental weight, while i is the ith
component of any old weight ~. There is no connection, although each index
takes on the same values 1, 2, . . . , m.
A direct product of representations of a group has weights which are
just the sum of the weights. In particular, the maximum weight of the
product is the sum of the maximum weights of the factors. Thus we can find
1
A mathematician, who calls the representation the matrices which act on the vector
space, and module that which the physicist calls the representation, would say instead the
representation is determined by how it represents these roots.

95

!
1 3
,
,
2 2

!
1
3
,
,
2
2

The corresponding fundamental weights with 2

!

1 +2

T+ = E

= E(1,0) .

~ i ~ j
= ij is solved by
(i )2

1
1 1
1
, , ~ 2 =
, .
~ =
2 2 3
2 2 3
1
2
Let us generate the representation D 2 . The qs
 areq E= 0, q = 1, so only

2
E2 | i gives a non-zero state proportional to 0, 1/ 3 . Now on this state



E
E
1

 1
3)
1 0, 1/

=
1,
E
3

,
1/2
3 is
E2 gives zero, but as 2 ~ (0,1/
1
2

( )
2
2
check

on
this
state.
not zero. Only one E1 can act, but we must still

~ 2 ( 12 , 1/2 3)
 1
= the number of E2
E2  2 , 1/2 3 gives zero so 2
(1 )2
we can use, but it is zero. So we are done, having found three basis vectors,
as there is no other way to get another state.
1 H
In flavor SU(3), these are the antiquarks. On the
2
horizontal axis, H1 = T3 is the isospin compos
nent. The highest weight state (rightmost) is the
anti-d quark |2 i, part of an isospin doublet with
1
the anti-u quark T |2 i = E1 E2 |2 i. The
-1
0
H1
antistrange quark s = E2 |2 i is an isosinglet,
u
d
T = T3 = 0. The vertical axis is generally described in terms of strangeness or hypercharge.
-1
1

Strangeness S is defined as zero for the u and d quarks and their antiparticles,
and 1 for the anti-strange quark. Baryon number B is invariant under SU(3),

618:

97

Last Latexed: March 22, 2012 at 10:47

and is defined as 1/3 the number of quarks minus the number of antiquarks,
so is 1/3 for all the antiquark states. The (strong2 ) hypercharge Y = S + B
(in the absence of charm, topness and bottomness)

is then 2/3 for s and 1/3 for u and d.

This
representation is the conjugate of the representation D 1 corresponding to the fundamental weight
1 , which is also called the defining representation for SU(3). In flavor SU(3) this is the representation of the first three quarks, with the up
quark u at the upper right, the down quark d to
its left, and the strange quark s at the bottom.

1
H2
1
-1

H1

-1

The defining representation of SU(3)

The quarks have B = 1/3. The

quark B T T3
S
Y
Q/e
electric charge is Q = T3 + Y /2
1
1
2
1
1
u
0
times the positron charge e. Then
3
2
2
3
3
the quarks have the quantum num1
1
1
1
d
2
0
13
3
2
3
bers as shown. The antiquarks have
1
all the quantum numbers (except
s
0
0 1 23 23
3
T ) reversed.
In general, if we have a representation Ta of the generators of a Lie algebra,
so that [Ta , Tb ] = icabd Td , with real structure constants cabd , then
Ta0 := Ta satisfies [Ta0 , Tb0] = [Ta , Tb ] = icabd Tc = icabd Tc0 ,
so T 0 is also a representation, called the conjugate represention to T . The
weights are the eigenvalues of T (Hi ), which are real, so ~ 0 = ~. Thus the
conjugate representation has a weight diagram which is just a parity reversed
(i.e. ~ ~) image of the original representation. The highest weight of
the conjugate representation is minus the lowest weight of the original.
The lowest weight of the defining representation for SU(3) is the d, and the d = |2 i
is the highest weight of the antiquark representation. The adjoint representation of
SU(3) is self-conjugate, that is, it is the
same (equivalent, isomorphic) representation as its conjugate.
2

To be distinguished from weak hypercharge.

98. Last Latexed: March 22, 2012 at 10:47

For another example,
q = 2, q 2 = 1, ~max
=
2~1 + ~2 = 3/2, 1/2 3 .
Then to max we can apply
E1 twice to get 2 roots
shown as , and E2 once
to get .
Now
consider
E1 on

E

= 1, 2/ 3 . p must be
zero or we would
E a root
get

vector 3/2, 7/2 3 which is
higher weight than max .

~1
2
But 2 1 2 = q p = 2 1,
( )
3
three states, shown with .

E1

max

T3

E2

!
1 3
= 3, so E1 on generates
,
2 2


E

Next consider the higher = 1, 1/ 3 , acted on by E2 The p is zero
because

E
E2 1, 1/ 3 E2 E1 |max i = E1 E2 |max i
|

{z

=0

where the first = is because different simple raising and lowering operators
commute
andthe
is because
you cant raise the highest weight. So

 = 0

1

1, 1/ 3 2 , 3/2
E

= 2, Thus E2 on 1, 1/ 3 generates the
q=2
1
two states.
!
1 1
,
are the same
Now the question arises whether the and at
2 2 3
state or not. This amounts to asking whether
E2 E1 |max i

The
adjoint
representation of
SU(3).

Joel A. Shapiro

and E1 E2 |max i

are linearly independent. You will show (problem IX.A) that they are linearly
independent,
so there are in fact two states corresponding to the weight
!
1 1
, .
2 2 3
Rather than continuing down and to the left, we will find that the symmetries of the root diagram will determine the rest.

99

100. Last Latexed: March 22, 2012 at 10:47

Given any root, there is a reflection which one can perform analogous
to eiL2 which reflects the weight vector of any representation. To see this,
consider, for any root
~ , not necessarily simple,

~ cos(t 22 )
=
~ H

618:

Last Latexed: March 22, 2012 at 10:47

[E E , Hi ] = i (E + E ) .
We now consider two cases. If ~
~ = 0,
h

~ =

~
~
~ = ~ 2
~,

~
~
~

~ + ,
2

~ |i =
~ H

Suppose we consider the state

=

|i = et (E E ) |~, Di
where ~ is an arbitrary basis state with weight vector i in an arbitrary
representation D i . The Cartan generators
~ |i = ~ He
~ t (E E ) |~, Di
~ H
~ |~, Di = ~ ~ |i
= et (E E ) ~ H
for ~
~ = 0.
~ |i =
~ t (E E ) |~, Di we use the general
To calculate
~ H
~ He
expression
n
X
t
n (A, B),
etA BetA =
n=0 n!
where n means the nth multiple commutator:
n (A, B) = [A, n1 (A, B)] .

~
~ H,
Let A = E E , B =
1 = 2 (E + E ) ,
2 = 2 [E E , E + E ]

Thus n =

~
(22 )n/2
~ H
n1
(22 ) 2 (2 ) (E + E )

Hi |i = i 2

~
i |i ,
2

~
~
~.
so has weight vector ~ 0 = ~ 2 2

(E E )

22
is a unitary transformation, so
The transformation e
it makes a 1-1 correspondence between weights of weight ~ and those of
~ . This corresponds to reflection in a plane (or hyperplane)
weight ~ 2~2~
perpendicular to
~ . Thus the weight diagram of any representation must be
symmetric under such reflections.
From the part of the (2, 1) representation we have found so far, as shown
in (a), we can reflect in the plane (line) perpendicular to 1 , to get the states
in (b). Then we reflect perpendicular to T3 to get (c), then perpendicular to
2 to get the full representation, or multiplet (d).
max

T3

E2

n even
,
n odd

~
~
~ ~ |i
~ 2

!
~
~
~ 2 2
~ ~ |i .

Each generator Hi corresponds to j = ij ,

E1

~
~ H.
= 22 [E , E ] = 22
(

~
~ = 0.

~ = 2 (E + E ) .
E E ,
~ H

0 (A, B) = B,

2
(E + E ) sin(t 22 ). Let
2

~ tA = ~
~ Now
~ |i =
~ tA |~, Di =
, so etA
~ He
H.
~ H
~ He
t = 2
2
~ |~, Di = ~
~ |i = ~ ~ |i for ~
H)
~ |i ~ H
~ = 0.
etA (~
Now any vector ~ can be written
~ tA
and etA
~ He

~ = 0,
E E , ~ H

while

Joel A. Shapiro

(a)

E1

max

T3

E2

(b)

E1

max

T3

E2

(c)

(d)

618:

8.1

Last Latexed: March 22, 2012 at 10:47

101

Tensor Methods

Consider the basis vectors e1 , e2 , e3 of the

d = e2
u = e1
defining representation of SU(3). The algebra
generators act by Ta ei = ej (Ta )j i with (Ta )j i =
1
(a )ji (and we are using the summation conven2
tion that an index appearing once upstairs and
s = e3
once downstairs in a term is understood to be
summed over).
From this point on we must be careful with upper and lower indices in
another sense than in our previous discussion. Here we are using them in the
sense of co- and contra-variant quantities, as is done in relativity.
We consider the basis vectors of the conjugate
s = e3
representation u = e1 , d = e2 , s = e3 . The generators act here with the conjugate representation
T 0 = T , so
Ta ei = ej (Ta0 )j = ej T j i = ej T ij ,
i

d = e2

u = e1

as is hermitean.
Now if we consider a tensor product, eij k = ei ej ek , the Lie algebra
generators act as a sum of pieces (like a derivative does), so
ij

Ta e

ij

=e

m
m Ta k

mj

i
k Ta m

im

j
k Ta m .

A vector ~v can be specified in terms of its components, which we give

indices to so as to contract with the basis vectors. Thus a vector in the
defining representation is
|vi = v i ei ,

Ta |vi = v i ej Ta ji =: |vi = v j ej ,

with v i = Ta ij v j .
A vector in the direct product space has coefficients with several indices
|vi = vij k eij k ,

Ta |vi = |vi = vij k eij k

where vij k = T km vij m T mi vmj k T mj vim k .

The set of all states ~v = vij k eij k , for arbitrary v, clearly form a representation,
but it is also clearly not irreducible, because the operation of the group

102. Last Latexed: March 22, 2012 at 10:47

Joel A. Shapiro

does nothing that alters the symmetry of the vs under i j). That is,
suppose we start off with a particular state vij k , and divide it into parts
vij k = sij k + aij k symmetric and antisymmetric under i j). Then sij k is
mapped into other symmetric coefficients under the group operations, while
a is mapped into other antisymmetric ones, and they dont mix, so we have
reduced the representation into two.
As an example, consider 3 3, where 3 is
the usual name for the defining representation
with q i = (1, 0). The tensor product is a 9 dimensional representation with nine basis vectors ei ej
having weights (ei )+(ej ), with a general vector
|vi = v ij ei ej . The three basis vectors ei ej
with i = j have weights which can only be composed in one way, but the ones with i 6= j have
the same weight for each pair i, j. So the weight
diagram is as shown.
d = e2

u = e1

s = e3

Dividing the space of 3 3 matrices v into symmetric ones sij and antisymmetric ones aij , we see that sij forms a six dimensional space and aij a
N
L
each
=
= 6 + 3,
three dimensional one. This divides
of which is irreducible.
There is something else left invariant by the generators of SU(3). If we
trace (i.e. set equal and sum) a lower with an upper index, the corresponding
terms in v cancel.

The trace
The simplest example is vij of 33.
P i
v = i vi is invariant under the generators,
vii = T im vim T mi vm i = 0.
Writing vij = 13 w i j +wij with w = vii , which
ensures that wij is traceless, reduces the nine 3 into the irdimensional representation 3
reducible representations 1 and 8, where 1
is the identity representation (all group elements are represented by 1, and the generators by zero), and 8, called the octet, is the
adjoint representation,

The adjoint (or octet) representation.

618:

Last Latexed: March 22, 2012 at 10:47

103

Notice that |vi := i j ei ej is actually an invariant under the group, even

though it looks like a tensor, i.e. both of its indices do transform under the
group.
If we have a representation defined by some tensor coefficients vij k which
transform in the correct manner under the group, then the trace in any upper
index paired with any lower indexX
extracts a tensor, with fewer indices, which

vij i transforms properly as a 3.

transforms appropriately: vj =
i

If we have two representations transforming properly, say uij and vij k ,

the direct product w ij`mk = uij v`m k is a tensor which transforms properly. It
can be contracted to form reduced representations, e.g. w k = w ijij k = uij vij k
is a 3. It can also have symmetries imposed to reduce it.
We will now construct the arbitrary SU(3) representation (n, m) as a
tensor product of n defining representations and m of its conjugate.
First construct (n, 0). ~max = n~1 , so |~max i = e1 e1 e1 (n of
Q
them). This state corresponds to the tensor w j1 jn = i ji ,1 which is clearly
symmetric under interchange of any two indices. How many components of
w of the full representation are there? One needs only to know how many
1s, 2s and 3s are picked to make a total of n. If you choose r 1s, there are
n r + 1 choices of how many 2s to pick, so in total there are
n
X

(n r + 1) =

r=0

n+1
X

j=

n+2
2

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

choices, so w has (n + 1)(n + 2)/2 independent components, and the (n, 0)

representation is (n + 1)(n + 2)/2 dimensional.
The same argument applies to the representation (0, m) of weight ~max =
m~2 ,
(with m factors).
|~max i = e2 e2 e2 ,
So the uk1 km corresponding to this representation is totally symmetric in all
its indices, and is a (m + 1)(m + 2)/2 dimensional representation.
The direct product of w j1 jn with uk1km can be reduced
n
w j1 jn uk1 km = vkj11j
km +

X j
r
rs

jr jn
ks Xkj1
k ,
k

where jr means leave out the jr index. The division is arranged so that v is
traceless,
X j j j
vj11k22 knm = 0.
j1

104. Last Latexed: March 22, 2012 at 10:47





Joel A. Shapiro

n+2
m+2
The
degrees of freedom in wu have had
2
2



n1+2
m1+2
degrees of freedom in X constrained out, leav2
2
ing
(n + 2)(n + 1)(m + 2(m + 1) (n + 1)n(m + 1)m

4
4
(n + 1)(m + 1)(n + m + 2)
=
2

Dim(n, m) =

for the dimension of the (n, m) representation.

We have extracted the leading irreducible representation from the product, but we have not fully reduced the product. Although v is irreducible, we
could reduce X iteratively in the same way, extracting the traces and being
left with traceless parts. Thus we find
(n, 0) (0, m) = (n, m) (n 1, m 1) (n m, 0)
if n m, or ending with (0, m n) if m > n.
We have constructed an arbitrary representation of SU(3) by products
but we have also seen that 3
is the antisymmetric part in
of 3s and 3s,
the product of two 3s. So any representation can be built of the defining
representation alone. This is a general feature of SU(n). The states, however,
will not correspond to some simple symmetry under permutations. Consider
333, the states of three quarks. The highest weight is 3 us, so (3, 0) = 10
is the symmetric part. There is also a totally antisymmetric part w ijk , but as
there is only one choice for {i, j, k} which doesnt vanish by antisymmetry,
there is only one degree of freedom here, w ijk = wijk , so this is the one
dimensional identity representation 1. The remaining 16 degrees of freedom
are in fact two octets (two 8s). We will have to show this.
N
For rank greater than 2, using i (D i )qi requires 3 or more sets of indices,
and our notational skills are not up to that. But, as we saw for SU(3), one
fundamental representation may be given by a tensor product of copies of
another. In Chapter XIII, Georgi shows that for SU(N), all fundamental
representations can be extracted from tensor products of the defining representation, with mixed symmetries.
If we start with the tensor product of k defining representations of SU(n),
we have an nk dimensional space which is not only a representation of SU(n)
but also of Sk , the permutation group on the indices. In fact, these operations

618:

Last Latexed: March 22, 2012 at 10:47

105

commute, so we may reduce the space into simultaneous representations.

So we must first learn something about representations of the permutation
group.

106. Last Latexed: March 22, 2012 at 10:47

Joel A. Shapiro