Tensor operators

We have used the vector notation for three component

operators for example to express the scalar product, like
p x

= p
x
x

+p
y
y

+ p
z
z

.
Classically a vector is a quantity that under rotations
transforms like V
3
(or (
3
), i.e. if R O(3), then
V

i
=
3

j=1
R
ij
V
j
.
In quantum mechanics V is a vector operator provided
that V ) (
3
is a vector:
R
[V
i
[)
R
= [T

(R)V
i
T(R)[)
=
3

j=1
R
ij
[V
j
[),
[) H, R O(3).
Thus we must have
T

(R)V
i
T(R) =

j
R
ij
V
j
.
Thus the innitesimal rotations
T( n) = 1
iJ n
h
satisfy
_
1 +
iJ n
h
_
V
i
_
1 +
iJ n
h
_
= V
i
+
i
h
(J nV
i
V
i
J n) +O(
2
)
=

j
R
ij
V
j
or
V
i
+

ih
[V
i
, J n] =

j
R
ij
( n)V
j
.
Substituting the explicit expressions for innitesimal
rotations, like
R( z) =
_
_
_
1

2
2
0
1

2
2
0
0 0 1
_
_
_,
we get
V
x
+

ih
[V
x
, J
z
] = V
x
V
y
+O(
3
).
Handling similarly the other components we end up with
[V
i
, J
j
] = ih
ijk
V
k
.
Finite rotation
A nite rotation specied by Euler angles is accomplished
by rotating around coordinate axises, so we have to
consider such expressions as
exp
_
iJ
j

h
_
V
i
exp
_

iJ
j

h
_
.
Applying the Baker-Hausdor lemma
e
iG
Ae
iG
=
A + i[G, A] +
_
i
2

2
2!
_
[G, [G, A]] +
+
_
i
n

n
n!
_
[G, [G, [G, . . . [G, A]]] . . .] +
we end up with the commutators
[J
j
, [J
j
, [ [J
j
, V
i
] ]]].
These will be evaluated in turn into operators V
i
and V
k
(k ,= i, j).
A vector operator (V ) is dened so that it satises the
commutation relation
[V
i
, J
j
] = ih
ijk
V
k
.
We can easily see that for example p, x and J are vector
operators.
In classical mechanics a quantity which under rotations
transforms like
T
ijk
. .
n indeces

R
ii
R
jj
R
kk
T
i

,
is called a Cartesian tensor of the rank n.
Example The dyad product of the vectors U and V
T
ij
= U
i
V
j
is a tensor of rank 2.
Cartesian tensors are reducible, for example the dyad
product can be written as
U
i
V
j
=
U V
3

ij
+
(U
i
V
j
U
j
V
i
)
2
+
_
U
i
V
j
+ U
j
V
i
2

U V
3

ij
_
.
We see that the terms transform under rotations
dierently:

U V
3

ij
is invariant. There is 1 term.

(U
i
V
j
U
j
V
i
)
2
retains its antisymmetry. There are 3
terms.

_
U
i
V
j
+ U
j
V
i
2

U V
3

ij
_
retains its symmetry and
tracelessness. There are 5 terms.
We recognize that the number of terms checks and that
the partition might have something to do with the
angular momentum since the multiplicities correspond to
the multiplicities of the angular momenta l = 0, 1, 2.
We dene the spherical tensor T
(k)
q
of rank k so that the
argument n of the spherical function
Y
m
l
( n) = n[lm)
is replaced by the vector V :
T
(k)
q
= Y
m=q
l=k
(V ).
Example The spherical function Y
1
:
Y
0
1
=
_
3
4
cos =
_
3
4
z
r
T
(1)
0
=
_
3
4
V
z
Y
1
1
=
_
3
4
x iy

2r
T
(1)
1
=
_
3
4
_

V
x
iV
y

2
_
.
Similarly we could construct for example a spherical
tensor of rank 2:
Y
2
2
=
_
15
32
(x iy)
2
r
2
T
(2)
2
=
_
15
32
(V
x
iV
y
)
2
.
The tensors T
(k)
q
are irreducible, i.e. there does not exist
any proper subset
T
(k)
p
1
, T
(k)
p
2
, . . . T
(k)
q
[q = k, . . . , +k,
which would remain invariant under rotations.
Transformation of spherical tensors
Under the rotation R an eigenstate of the direction
transforms like
[ n) [ n

) = T(R)[ n).
The state vectors [lm), on the other hand, transform
under the rotation R
1
like
T(R
1
)[l, m) =

[l, m

)T
(l)
m

m
(R
1
).
So we get
Y
m
l
( n

) = n

[lm) = n[T

(R)[lm)
= n[T(R
1
)[lm) =

n[[lm

)T
(l)
m

m
(R
1
)
=

Y
m

l
( n)T
(l)
m

m
(R
1
)
=

Y
m

l
( n)T
(l)
mm

(R).
We dene a tensor operator Y
m
l
(V ) so that
T

(R)Y
m
l
(V )T(R) =

Y
m

l
(V )T
(l)

mm
(R).
Generalizing we dene: T
(k)
q
is a (2k + 1)-component
spherical tensor of rank k if and only if
T

(R)T
(k)
q
T(R) =
k

=k
T
(k)

qq
(R)T
(k)
q

or equivalently
T(R)T
(k)
q
T

(R) =

k
q

=k
T
(k)
q

q
(R)T
(k)
q
.
Under the innitesimal rotations
T( n) =
_
1
iJ n
h
_
a spherical tensor behaves thus like
_
1 +
iJ n
h
_
T
(k)
q
_
1
iJ n
h
_
=
k

=k
T
(k)
q
kq

[
_
1 +
iJ n
h
_
[kq)
=
k

=k
T
(k)
q
kq

[kq) +
k

=k
iT
(k)
q
kq

[J n[kq),
or
[J n, T
(k)
q
] =

T
(k)
q
kq

[J n[kq).
Choosing n = z and x i y we get
[J
z
, T
(k)
q
] = hqT
(k)
q
and
[J

, T
(k)
q
] = h
_
(k q)(k q + 1)T
(k)
q1
.
Example Decomposition of the dyad product.
We form spherical tensors of rank 1 from the vector
operators U and V :
U
0
= U
z
, V
0
= V
z
,
U
1
=
U
x
iU
y

2
, V
1
=
V
x
iV
y

2
.
Now
T
(0)
0
=
U V
3
=
U
+1
V
1
+ U
1
V
+1
U
0
V
0
3
,
T
(1)
q
=
(U V )
q
i

2
,
T
(2)
2
= U
1
V
1
,
T
(2)
1
=
U
1
V
0
+U
0
V
1

2
,
T
(2)
0
=
U
+1
V
1
+ 2U
0
V
0
+ U
1
V
+1

6
.
In general we have
Theorem 1 Let X
(k
1
)
q
1
and Z
(k
2
)
q
2
be irreducible spherical
tensors of rank k
1
and k
2
. Then
T
(k)
q
=

q
1

q
2
k
1
k
2
; q
1
q
2
[k
1
k
2
; kq)X
(k
1
)
q
1
Z
(k
2
)
q
q
is a (irreducible) spherical tensor of rank k.
Proof: We show that T
(k)
q
transforms like
T

(R)T
(k)
q
T(R) =
k

=k
T
(k)

qq
(R)T
(k)
q
.
Now
T

(R)T
(k)
q
T(R)
=

q
1

q
2
k
1
k
2
; q
1
q
2
[k
1
k
2
; kq)
T

(R)X
(k
1
)
q
1
T(R)T

(R)Z
(k
2
)
q
2
T(R)
=

q
1

q
2

2
k
1
k
2
; q
1
q
2
[k
1
k
2
; kq)
X
(k
1
)
q

1
T
(k
1
)
q

1
q
1
(R
1
)Z
(k
2
)
q

2
T
(k
2
)
q

2
q
2
(R
1
)
=

q
1

q
2

k
1
k
2
; q
1
q
2
[k
1
k
2
; kq)
k
1
k
2
; q

1
q

2
[k
1
k
2
; k

)
k
1
k
2
; q
1
q
2
[k
1
k
2
; k

)T
(k

)
q

q
(R
1
)X
(k
1
)
q

1
Z
(k
2
)
q

2
,
where we have substituted the Clebsch-Gordan series
expansion
T
(j
1
)
m
1
m

1
(R)T
(j
2
)
m
2
m

2
(R) =

j
1
j
2
; m
1
m
2
[j
1
j
2
; jm)
j
1
j
2
; m

1
m

2
[j
1
j
2
; jm

)T
(j)
mm
(R)
Taking into account the orthogonality of the
Clebsch-Gordan coecients

m
1
m
2
j
1
j
2
; m
1
m
2
[j
1
j
2
; jm)j
1
j
2
; m
1
m
2
[j
1
j
2
; j

)
=
jj

mm

we get
T

(R)T
(k)
q
T(R)
=

kk

qq
k
1
k
2
; q

1
q

2
[k
1
k
2
; k

)
T
(k

)
q

q
(R
1
)X
(k
1
)
q

1
Z
(k
2
)
q

2
,
which can be rewritten as
T

(R)T
(k)
q
T(R)
=

_
_

1
q

2
k
1
k
2
; q

1
q

2
[k
1
k
2
; kq

)X
(k
1
)
q

1
Z
(k
2
)
q

2
_
_
T
(k)
q

q
(R
1
)
=

T
(k)
q
T
(k)
q

q
(R
1
) =

T
(k)

qq
(R)T
(k)
q

Matrix elements of tensor operators

Theorem 2 The matrix elements of the tensor operator
T
(k)
q
satisfy

, j

[T
(k)
q
[, jm) = 0,
unless m

= q + m.
Proof: Due to the property
[J
z
, T
(k)
q
] = hqT
(k)
q
we have

, j

[[J
z
, T
(k)
q
] hqT
(k)
q
[, jm)
= [(m

m) h qh]

, j

[T
(k)
q
[, jm) = 0,
so

, j

[T
(k)
q
[, jm) = 0,
if m

,= q + m
Theorem 3 (Wigner-Eckardts theorem) The matrix
elements of a tensor operator between eigenstates of the
angular momentum satisfy the relation

, j

[T
(k)
q
[, jm) = jk; mq[jk; j

|T
(k)
|j)
_
2j + 1
,
where the reduced matrix element

|T
(k)
|j) depends
neither on the quantum numbers m, m

nor on q.
Proof: Since T
(k)
q
is a tensor operator it satises the
condition
[J

, T
(k)
q
] = h
_
(k q)(k q + 1)T
(k)
q1
,
so

, j

[[J

, T
(k)
q
][, jm)
= h
_
(k q)(k q + 1)

, j

[T
(k)
q1
[, jm).
Substituting the matrix elements of the ladder operators
we get
_
(j

)(j

+ 1)

, j

, m

1[T
(k)
q
[, jm)
=
_
(j m)(j m + 1)

, j

, m

[T
(k)
q
[, j, m1)
+
_
(k q)(k q + 1)

, j

, m

[T
(k)
q1
[, jm).
If we now substituted j

j, m

m, j j
1
, m m
1
,
k j
2
and q m
2
, we would note that the recursion
formula above is exactly like the recursion formula for the
Clebsch-Gordan coecients,
_
(j m)(j m + 1)j
1
j
2
; m
1
m
2
[j
1
j
2
; j, m1)
=
_
(j
1
m
1
+ 1)(j
1
m
1
)
j
1
j
2
; m
1
1, m
2
[j
1
j
2
; jm)
+
_
(j
2
m
2
+ 1)(j
2
m
2
)
j
1
j
2
; m
1
, m
2
1[j
1
j
2
; jm).
Both recursions are of the form

j
a
ij
x
j
= 0, or sets of
linear homogenous simultaneous equations with the same
coecients a
ij
. So we have two sets of equations

j
a
ij
x
j
= 0,

j
a
ij
y
j
= 0,
one for the matrix elements (x
i
) of the tensor operator
and the other for the Clebsch-Gordan coecients (y
i
).
These sets of equations tell that
x
j
x
k
=
y
j
y
k
j and k xed,
so x
j
= cy
j
while c is a proportionality coecient
independent of the indeces j. Thus we see that

, j

[T
(k)
q
[, jm)
= ( constant independent on m, q and m

)
jk; mq[jk; j

).
If we write the proportionality coecient like

|T
(k)
|j)
_
2j + 1
we are through.
According to the Wigner-Eckart theorem a matrix
element of a tensor operator is a product of two factors,
of which
jk; mq[jk; j

) depends only on the geometry, i.e.

on the orientation of the system with respect to the
z-axis.

|T
(k)
|j)
_
2j + 1
depends on the dynamics of the
system.
As a special case we have the projection theorem:
Theorem 4 Let
J
1
=
1

2
(J
x
iJ
y
) =
1

2
J

, J
0
= J
z
be the components of the tensor operator corresponding to
the angular momentum. Then

, jm

[V
q
[, jm) =

, jm[J V [, jm)
h
2
j(j + 1)
jm

[J
q
[jm).
Proof: Due to the expansions
T
(0)
0
=
U V
3
=
U
+1
V
1
+ U
1
V
+1
U
0
V
0
3
,
T
(1)
q
=
(U V )
q
i

2
,
T
(2)
2
= U
1
V
1
,
T
(2)
1
=
U
1
V
0
+U
0
V
1

2
,
T
(2)
0
=
U
+1
V
1
+ 2U
0
V
0
+ U
1
V
+1

6
we can write

, jm[J V [, jm)
=

, jm[(J
0
V
0
J
+1
V
1
J
1
V
+1
), jm)
= mh

, jm[V
0
[, jm)
+
h

2
_
(j + m)(j m + 1)

, j, m1[V
1
[, jm)

2
_
(j m)(j +m + 1)

, j, m + 1[V
+1
[, jm)
= c
jm

j|V |j),
where, according to the Wigner-Eckart theorem the
coecient c
jm
does not depend on ,

or V .
The coecient c
jm
does not depend either on the
quantum number m, because J V is a scalar operator,
so we can write it briey as c
j
. Because c
j
does not
depend on the operator V the above equation is valid
also when V J and

, or
, jm[J
2
[, jm) = h
2
j(j + 1) = c
j
j|J|j).
If we now apply the Wigner-Eckart theorem to the
operators V
q
and J
q
we get

, jm

[V
q
[, jm)
, jm

[J
q
[, jm)
=

j|V |j)
j|J|j)
.
for the ratios of the matrix elements. On the other hand,
the right hand side of this equation is

, jm[J V [, jm)
, jm[J
2
[, jm)
,
so

, jm

[V
q
[, jm) =

, jm[J V [, jm)
h
2
j(j + 1)
jm

[J
q
[jm)
Generalizing one can show that the reduced matrix
elements of the irreducible product T
(k)
q
of two tensor
operators, X
(k
1
)
q
1
and Z
(k
2
)
q
2
, satisfy

[[T
(k)
|j)
=

2k + 1(1)
k+j+j

_
k
1
k
2
k
j j

|X
(k
1
)
|

[Z
(k
2
)
|j).