Ph125b Monday 5 March 2007

Spherical tensor operators and the Wigner-Eckart Theorem

[After Merzbacher 17.7]
A x , A y , A z as a set of three
Lets intuitively define a vector operator A
Hilbert-space operators whose expectation values transform under rotations like a
coordinate-space vector. Recalling that inner products between vectors are preserved
under spatial rotations, we must have
A x

A y

A z
|
|A

|
| A

where is a unit vector in the coordinate space and primed objects are the images of
unprimed objects after the rotation. If we let U R denote the Hilbert-space
representation of a spatial rotation R,
R .
| U R | ,
Since
| A x e A y e A z e
| A
x

|
| A
and
|
|A

U | ,
|U R A
R

we must have
U A
,
UR A
R
U A
.
UR A
R
The matrix R that represents a spatial rotation by clockwise angle around an axis n
is
1 n 2x cos n 2x
R

n y n x 1 cos n z sin

n x n y 1 cos n z sin n x n z 1 cos n y sin

1 n 2y cos n 2y

n z n x 1 cos n y sin n z n y 1 cos n x sin

n y n z 1 cos n x sin

1 n 2z cos n 2z

so with R we see that

U R A j U R

A i R ij .
i

To write it another way,

Ax

Ax
Rn ,

Ay

Ay

Az

Az

under the spatial rotation n , , which shows that the transformation of a vector
operator under rotation follows (or induces) a linear representation of the rotation
group.
Of course, we know what the Hilbert-space representation U R looks like as well,
from which we can infer
exp i n
.
exp i n
J A
J RA

Taking to be infinitesimal and expanding both the operator exponentials and the
rotation matrix R to first order gives the convenient relation
, n

A
J i n A
.
that is necessarily satisfied for arbitrary n by any vector operator A

This notion of a vector operator can be generalized to define spherical tensor

operators, with a vector operator being a rank-1 spherical tensor operator. We have
previously seen that the rotation group has linear representations on vector spaces of
various dimension, and the vector operators pick out the representation with
dimension 3 2j 1 (hence rank j 1). For arbitrary (non-negative) integer rank k,
we define
q
U R T k U R

T kq D qkq R.

q k
k

Here D q q R denotes the q , q matrix element of the k-dimensional linear

representation of the rotation R. Keep in mind that we have implicitly fixed some
q
underlying Hilbert space, on which T k acts and which defines the U R representation.
For rank 1, it is straightforward to explicitly compute the spherical tensor
components of an arbitrary vector operator with Cartesian components A x , A y , A z :
T 11 1 A x iA y
2
T 10 A z ,
T 11 1 A x iA y .
2
An important example of a vector operator is the position operator r for particle motion
in 3D, whose spherical tensor components are evidently

x iy
x iy
, z,
2
2

It is again convenient to consider infinitesimal rotations, now in the definition of

spherical tensor operators, yielding the necessary and sufficient condition
q
n
J , Tk

T kq

kq | n
J | kq ,

q k

which in turn can be reduced to

q
q
J z , T k qT k ,
q

k qk q 1 T k ,

k qk q 1 T k .

J , T k
J , T k

q1
q1

It should be clear that knowing the decomposition of an operator into spherical tensor
components can thus be quite useful in dealing with problems that have rotational
symmetry.

Given that spherical tensor operators have such nice transformation properties under
rotations of the coordinate space, it seems convenient to ask the following question
q
(as Merzbacher does in his book): If T k is an irreducible tensor operator, how much
information about its matrix elements in the angular momentum basis can be
inferred? As it turns out, quite a lot! The answer is summarized in the famous
Wigner-Eckart theorem (see, e.g., Merzbacher pp. 432-435 or Sakurai pp. 238-240 for
the derivation):
q
q
j m | T k | j m jkmq | jkj m j || T k || j .
If and denote the set of quantum numbers necessary in addition to j, m to fully
specify a basis state, this shows that the matrix elements of a spherical tensor
operator are proportional to Clebsch-Gordan coefficients jkmq | jkj m . The constant
q
of proportionality characterizes the aspects of T k that are not pinned down by
rotational symmetry, and is known as the reduced matrix element.

Central potentials and the radial equation

For spherically symmetric (central) potentials Vr, the 3D Schrdinger Equation with
2
p
H
Vr
2m
has stationary solutions of the form
H E r E E r,
E r R E rY ml , .
Here Y ml , are the spherical harmonics we saw previously. Since H is rotationally
symmetric it commutes with both L 2 and L z , hence we may choose the energy
eigenstates to be eigenstates of total angular momentum and L z as well. The function

R E r contains the radial behavior of , and reflects the influence of a particular

potential energy term Vr.
Using

i
p
in the position representation and the form of 2 in spherical coordinates, one can
show (Merzbacher 11.111) that the kinetic energy term
2
2
2
p
L 2 2 r2 ,
2m
r
2mr
2mr r

is the operator for orbital angular momentum. Hence the stationary

where L r p
form of the 3D Schrdinger Equation may be written
2
2
2 r 2 L 2 Vr E r E E r.
r
2mr r
2mr
Substituting in the separated form involving spherical harmonics,
2
ll 1 2
2 r2
Vr R E r E R E r,
r
2mr r
2mr 2
since Y ml , is an eigenfunction of L 2 with eigenvalue ll 1.
It is often convenient to introduce one more transformation, by defining
ur r R E r.
This now satisfies the radial equation
2
2
ll 1 2
d u2
Vr u E u,
2m dr
2mr 2
which looks like the 1D Schrdinger Equation with an extra potential energy term
ll 1 2 /2mr 2 , corresponding to a centrifugal barrier. Having made these
transformations, we are now free to think of the problem of finding eigenstates for a
central potential as being equivalent to that of solving the 1D radial equation. An
important difference however is that r (unlike x) is never negative, so we must be
careful about the boundary condition on u as r 0. Clearly we would like u0 0 for
finiteness of at the origin, but later on we will see that there are some requirements
on its functional form as well.

Solution of the radial equation for a free particle

In the absence of a real potential energy term (as opposed to the centrifugal one), it
turns out that the radial equation can be morphed into something whose solutions are
clearly Bessel functions. Starting from the original
2
ll 1 2
2 r2
Vr R E r E R E r,
r
2mr r
2mr 2
we first introduce the scaling

2mE r kr.
2

With this, we have

d 2 R 2 dR 1 ll 1
d
d 2
2

R 0.

Inserting the ansatz

R

J
,

we obtain
2
d 2 J 1 dJ 1 l 1/2
J 0,
d
d 2
2
which is Bessels equation. The solutions of this equation that are well-behaved at
r 0 are the cylindrical Bessel functions J l1/2 . Hence the radial components of the
stationary wave functions can be written
R C j l
where C is a normalization constant and j l is a spherical Bessel function
J
j l
.
2 l1/2

Thus we end up with free-particle solutions

klm r C j l kr Y ml , ,
where
2 2
E k .
2m
It should be noted that these energy eigenstates for the free particle (like plane waves)
are not normalizable, as the asymptotic form of the spherical Bessel functions is
cos l 1/2
j l
,
l

and we thus will have an integrand in the normalization integral that goes like
2 d |R| 2 d cos 2 .
As you probably know from E&M, Bessel functions are quite common in physics
and engineering so most computer mathematics packages have built-in routines to
generate them. For example in Matlab a few lines of code suffice to generate the
following plot of j l , for l 0 (black), l 1 (red), l 5 (blue) and l 10 (green):

0.8

0.6

0.4

0.2

-0.2

-0.4

10

12

14

16

18

20

A few things to note are that only l 0 has a non-zero value at the origin
(singlevaluedness of the wave-function) and that the initial maximum of j l moves
to higher as l increases (centrifugal force).

The spherical square well potential

Next we move on to the 3D version of a potential well, with Vr V 0 for r a and
Vr 0 for r a (with V 0 0). Just as in the 1D case, we must solve the radial
equation within regions of constant Vr and then match solutions at the boundary:
2
2 ll 1
2 d r 2 dR
R E V 0 R for r a,
dr
2mr 2
2mr dr
2
2 ll 1
2 d r 2 dR
R ER
for r a.
dr
2mr 2
2mr dr
Bound states of the spherical square well will have V 0 E 0.
Inside the well region r a we can use the above results with E E V 0 ,
Rr A j l

2mE V 0
r
2

for r a.

Outside the well r a we should be more careful. Since we are now dealing with a
bound state problem, we should expect the eigenstates to be normalizable. The trick
is that Bessels equation admits additional solutions besides the j l , which are
singular at the origin but may be admissible in a region that does not include r 0.
For a r , the radial solutions of choice are spherical Hankel functions of the first
kind. Recall that for bound states we have E 0 and therefore outside the well

k i 2mE
i,
2
and the corresponding solutions are
1

Rr B h l

i 2mE
r .
2

As in the 1D case we match solutions at the boundary r a. Note that we have a

total of three constraints (continuity of , continuity of , overall normalization) in two
unknowns (A and B). Hence we can match the logarithmic derivative to find the energy
spectrum and worry about normalization separately. This leads to

2 2 jl
i
jl

2 2 a
2 2 a

h l ia
1
h l ia

as an equation for the allowed discrete energy eigenvalues, where 2 2mV 0 / 2 .

Single-valuedness of the wave function

[Sakurai pp. 201-202]
In our discussion of the Schrdinger Equation for central potentials we have left the
orbital angular momentum l as a completely free parameter, but it is important to note
that orbital angular momentum is normally constrained to take on integer values only.
One way to see why this should be so is simply to look at the form of the spherical
harmonics:
d lm
2l 1 l m! expimsin m
sin 2l .
lm
4 l m!
dcos
We have previously noted that, if we ignore the normalization and -dependent terms,
we have
Y lm , expim,
where of course m l, l 1, , l 1, l . This being the case, we see that
Y lm , will have a weird sort of branch-cut unless l is constrained to be an integer.
Considering 0, 2, we have
Y lm ,
lim
expi2m,
2
Y lm , 0
which is equal to 1 if m is a half-integer. This would lead to various singularities in
quantities such as
L z i d 3 r r, , r, , .

Sakurai also points out that under a rotation about the z-axis by 2, wave-functions
should obey
Y lm ,

1 l
2 l l!

r, , r | r | exp 2i L z |

r | exp 2i L z
|

exp 2i L z | r |

r | .
Therefore the wave-function should not change if 2, but we see that this is
not the case if l is half-integer.
We can contrast this with the case of a spin- 12 degree of freedom, for which
Sz
2

0 1

and a 2 rotation about the z-axis is represented by

R z 2 exp i 2S z

exp

i 0
0

e i

e i

1.

This is in fact not an artifact but a real, observable property of spin- 12 particles (the
so-called spinor property).
Before leaving this topic, it is worth noting that one has to be careful in scenarios
with more complicated rotation symmetry, because it is actually possible for the
considerations above to lead to the requirement of half-integer l when spin degrees of
freedom are coupled to orbital degrees of freedom (an example of this is the case of a
spin- 12 particle in a cylindrical quadrupole magnetic field).