Wigner–Eckart Theorem

The Wigner–Eckart theorem concerns the matrix elements of scalar, vector, and tensor
operators between states |α, j, mi of definite angular momentum — or rather, definite Jz
and definite J2 . Most generally, the theorem gives us

α0 , j 0 , m0 Âindices |α, j, mi = α0 , j 0 | Â ||α, ji × a known function of(j, m, j 0 , m0 , indices)

(1)
where ‘indices’ denotes 0, 1, or more vector indices of a scalar, vector, or tensor operator
Âindices , and hα0 , j 0 || Â ||α, ji is the reduced matrix element which does not depend on m0 , m,
or the indices of Â. As to the ‘known function’ in eq. (1), I shall spell it out in terms of the
Clebbsch–Gordan coefficients later in these notes.

The Wigner–Eckart theorem drastically reduces the number of independent matrix ele-
ments one may need to calculate the hard way. For example, consider the transition in which
the outer-most electron of an aluminum atom moves from the excited L = 2 state down to
the ground L = 1 state while emitting a photon. The transition rate here depends on the
matrix element of the electric dipole operator d̂, between the two states,

ground, L0 = 1, m0L dˆi |excited, L = 2, mL i (2)

Alas, the initial state here can be any one of 5 degenerate states with L = 2 (but different
mL ); likewise, the final state can be any one of 3 degenerate ground states with L0 = 1; and
the electric dipole vector has 3 components dˆi , so altogether there are 5 × 3 × 3 = 45 matrix
elements for the transition in question. For simplicity I ignore the spin here; otherwise we
would need even more matrix elements. But even “only” 45 matrix elements are way too
much work to calculate as explicit integrals. But thanks to the Wigner–Eckart theorem
we do not need to work quite this hard. Instead, it’s enough to calculate any one non-
vanishing matrix element (2) the hard way, and the eq. (1) yields the reduced matrix element
hground, L0 = 1|| dˆ||excited, L = 2i and hence all 45 matrix elements (2) for the transition
in question.

1
Scalar Case

Let’s start with the matrix elements of a scalar operator Â. The scalar operators com-
mute with Ĵ and Ĵ2 , so the state  |α, j, mi has exactly the same j and m as the original
|α, j, mi state. Consequently,

α0 , j 0 , m0 Â |α, j, mi = 0 unless j 0 = j and m0 = m. (3)

Moreover, since the rotation operators R̂(n, φ) mix up states with different m’s, the matrix
element here does not depend on m0 = m (as long as m0 = m), thus

α0 , j 0 , m0 Â |α, j, mi = δj 0 ,j δm0 ,m × function of(α0 , α, j, but not m). (4)

Altogether, there is a selection rule — the matrix element of a scalar operator vanishes unless
j 0 = j, — and for any given α0 , α, and j 0 = j, all the (2j + 1)2 matrix elements corresponding
to different m and m0 are related via eq. (4).

Finally, the function of α0 , α, and j 0 = j on the RHS of eq. (4) is called the reduced matrix
element hα0 , j|| Â ||α, ji of the scalar operator Â. Or rather, in some conventions it’s called
the reduced matrix element, while in other conventions — like in the Sakurai’s textbook —
√
there is an extra factor of 2j + 1, thus

hα0 , j|| Â ||α, ji

α0 , j 0 , m0 Â |α, j, mi = δj 0 ,j δm0 ,m × √ . (5)
2j + 1

Vector Case

Next, consider the matrix elements of the vector operators V̂. The Cartesian components
V̂x,y,z of a vector operator obey [Jˆi , V̂j ] = ih̄i,j,k V̂k , but for our purposes it’s more convenient
to use the so-called spherical components

+V̂x − iV̂y −V̂x − iV̂y

V̂−1 = √ , V̂0 = V̂z , V̂+1 = √ , (6)
2 2
inspired by the ` = 1 spherical harmonics,
r r r
4π +x − iy 4π 4π −x − iy
rY1,−1 (θ, φ) = √ , rY1,0 (θ, φ) = z, rY1,+1 (θ, φ) = √ .
3 2 3 3 2
(7)
The commutators of the spherical components V̂µ (µ = −1, 0, +1) with the angular momen-

2
tum operators Ĵ exactly parallels the action of the Ĵ operators on the |j, mi states with
j = 1:

[Jˆz , V̂µ ] = µh̄V̂µ , Jˆz |j = 1, mi = mh̄ |j = 1, mi ,

[Jˆ± , V̂±1 ] = 0, Jˆ± |j = 1, m = ±1i = 0,
√ √
[Jˆ± , V̂0 ] = 2h̄V̂±1 , Jˆ± |j = 1, m = 0i = 2h̄ |j = 1, m = ±1i ,
√ √
[Jˆ± , V̂∓1 ] = 2h̄V̂0 , Jˆ± |j = 1, m = ∓1i = 2h̄ |j = 1, m = 0i
(8)
(The proof is a part of your homework set#12). Consequently, for any unit vector n

X
[(n · Ĵ), V̂µ ] = V̂µ0 × j = 1, µ0 (n · Ĵ) |j = 1, µi (9)
µ0

and hence for any finite rotation R(φ, n)

(j=1)
X
R̂(φ, n)V̂µ R̂† (φ, n) = V̂µ0 × Dµ0 ,µ (φ, n). (10)
µ0

(Again, the proof is a part of the homework set#12).

Next, consider the vector operator V̂µ acting on states with definite j and m, V̂µ |α, j, mi.
When we rotate this state, we get

R̂(φ, n)V̂µ |α, j, mi = R̂(φ, n)V̂µ R̂† (φ, n) × R̂(φ, n) |α, j, mi

  !
(1) (j)
X X
=  V̂µ0 Dµ0 ,µ (φ, n) × α, j, m0 Dm0 ,m (φ, n)
(11)
µ0 m0
(1) (j)
X
= V̂µ0 α, j, m0 × Dµ0 ,µ (φ, n)Dm0 ,m (φ, n) .
µ0 ,m0

This transformation law under finite rotation looks exactly like transformation of two simul-
taneously rotated sets degrees of freedom, in multiplets |j1 , m1 , j2 , m2 i where for the case at
hand j1 = 1, m1 = µ, j2 = j, and m2 = m. Consequently, from the net angular momentum
point of view the state V̂µ |α, j, mi is a linear combination of states with mnet = µ + m
while jnet runs from |j − 1| to j + 1 by 1. Moreover, the coefficient of any particular jnet

3
component of this state is is the usual Clebbsch–Gordan coefficient C(jnet , mnet |1, µ, j, m)
for adding angular momenta j1 = 1 and j2 = j. Although once we get the non-rotational
degrees of freedom represented by α’s into account, we get a more general formula

j+1 X
X
V̂µ |α, j, mi = α0 , j 0 , m0 = µ + m × C(j 0 , m0 |1, µ, j, m) ×
j 0 =|j−1| α0

× function(α0 , j 0 , α, j but not m0 , µ, m).

(12)
Or in terms of the matrix elements of the vector operator V̂,

C(j 0 , m0 |1, µ, j, m)
α0 , j 0 , m0 V̂µ |α, j, mi = √ 0 × α0 , j 0 | V̂ ||α, ji , (13)
2j + 1

where the reduced matrix element hα0 , j 0 || V̂ ||α, ji depends on the multiplets (α, j) and
(α0 , j 0 ) and on the vector operator V̂ but does not depend on the multiplet members m or
m0 of the specific component µ of the vector operator.

Also, the Clebbsch–Gordan coefficient in eq. (13) enforces the selection rules: the matrix
elements of vector operators vanish unless |j − 1| ≤ j 0 ≤ j + 1 — or equivalently |j 0 − j| =
(0 or 1) and j 0 + j ≥ 1, — and also m0 = µ + m.

Spherical Tensors

Similar to the scalar and vector operators, there are Wigner–Eckart theorems for the
tensor operators T̂ij , T̂ijk , etc., with any numbers of indices. However, such tensor form
reducible multiplets of the rotation group SO(3), so they need to be reorganized into ir-
reducible multiplet, and then each irreducible multiplet re-expressed in terms of spherical
rather than Cartesian components.

For example, the irreducible parts of a general two-index tensor Tij are:
P
1. The trace S = tr(T ) = Tii (implicit i ), which is a scalar.

2. The antisymmetric part TijA = 21 (Tij − Tji ) = −TjiA , which is equivalent to a vector
Vi = 12 ijk Tjk
A = 1 T , T A =  V .
2 ijk jk ij ijk k

4
 3. The traceless symmetric part,

TijS = 1
2 (Tij + Tji ) − 1
3 δij tr(T ), TijS = +TjiS , tr(T S ) = 0. (14)

For the higher-rank tensors, their irreducible parts are also scalars, vectors, and traceless
symmetric tensors of various ranks. For example, a general 3-index tensor Tijk decomposes
into a 3-index traceless totally-symmetric tensor, 2 different 2-index traceless symmetric
tensors, 3 different vectors, and one scalar.

Next, consider the spherical harmonics or rather f`,m (x) = r` Y`,m (θ, φ). In terms of
Cartesian coordinates x, y, z, each f`,m is a homogeneous polynomial of degree `, hence

f`,m (x) = r` Y`,m (θ, φ) = T i1 ,i2 ,...,i` xi1 xi2 · · · xi` (15)

for a totally symmetric traceless tensor T i1 ,i2 ,...,i` , where the tracelessness follows from

∇2 f`,m (x) = `(` − 1)T j,j,i3 ,...,i` xi3 · · · xi` = 0. (16)

`+2

Moreover, a rank-` (i.e., `-index) totally symmetric tensor in 3d has 2 = 21 (` + 1)(` + 2)
1
independent components, while the zero trace condition amounts to 2 (` − 2 + 1)(` − 2 + 2)
linear constraints, so a traceless symmetric tensor has only

1 1
2 (` + 1)(` + 2) − 2 (` − 2 + 1)(` − 2 + 2) = 2` + 1 (17)

independent components. Consequently, eq. (15) provides a one-to-one map between the
Cartesian components T i1 ,...,i` and the spherical components µ = −`, . . . , +`. For example,
for ` = 2
r
8π 2
r Y2,±2 (θ, φ) = 12 (∓x − iy)2 ,
15
r
8π 2 (18)
r Y2,±1 (θ, φ) = (∓x − iy)z,
15
r
8π 2
q
1
2z 2 − x2 − y 2 ,


r Y2,0 (θ, φ) = 6
15

5
so for any traceless symmetric 2-index tensor Tij we define its spherical components as

(2) 1
T±2 = 2 (Txx − Tyy ) ± iTxy ,
(2)
T±1 = ∓Txz − iTyz , (19)
q
(2) 1


T0 = 6 2Tzz − Txx − Tyy .

The bottom line of this exercise is that any ordinary tensor is equivalent to one or
(`)
several spherical tensors Tm which transform under rotations like the spherical harmonics
Y`,m (θ, φ), namely
X ∗
(`) (`) (`)
R : Tm → Dm,m0 (R) × Tm0 , (20)
m0

similarly to

Y`,m (R ◦ n) = hR ◦ n|`, mi = hn| R̂† (R) = R̂(R−1 ) |`, mi

(`)
X
= n|`, m0 × Dm0 ,m (R−1 )
m0
X †  (21)
0 (`)
= n|`, m × D (R)
m0 ,m
m0
X ∗
(`)
= Dm,m0 (R) Y`,m0 (n).
m0

Tensor Operators
(k)
A spherical tensor operator T̂ (k) has 2k + 1 components T̂µ which transform into each
other under rotations according to

X ∗
(k) (k)
R̂† (R)T̂ (k) R̂(R) = Dµ,µ0 (R) × T̂µ0 , (22)
µ0

or in reverse direction

(k) (k)
X
R̂(R)T̂ (k) R̂† (R) = T̂µ0 × Dµ0 ,µ (R), (23)
µ0

6
exactly as the quantum states in a (j = k) multiplet,

(k)
X
R̂(R) |j = k, m = µi = j = k, m = µ0 × Dµ0 ,µ (R). (24)
µ0

By the Baker–Hausdorff lemma, the transformation laws (23) are equivalent to the commu-
tation relations with the angular momenta,

(k) (k)
Jˆz , T̂µ
 
= µh̄ T̂µ ,
(25)
(k) (k)
Jˆ± , T̂µ
  p
= (k ∓ µ)(k + 1 ± µ)h̄ T̂µ±1 ,

similarly to the action of the Ĵ operators on |j, mi states in a (j = k) multiplet,

Jˆz |j = k, m = µi = µh̄ |j = k, m = µi ,
(26)
Jˆ± |j = k, m = µi =
p
(k ∓ µ)(k + 1 ± µ)h̄ |j = k, m = µ ± 1i .

(k)
When the spherical tensor operators T̂µ act on the |α, j, mi states, the resulting states
transform under rotations R according to

(k) (k)
R̂(R)T̂µ |α, j, mi = R̂(R)T̂µ R̂† × R̂ |α, j, mi
 
X (k) (k)  
(j)
=  T̂µ0 Dµ0 ,µ (R) × α, j, m0 Dm0 ,m (R)
(27)
µ0
(k) (k) (j)
X
= T̂µ0 α, j, m0 × Dµ0 ,µ (R) Dm0 ,m (R),
µ0 ,m0

which works exactly like a (j1 = k) × (j2 = j) multiplet of two separate (but simultaneously
rotated) sets of degrees of freedom. Similar to what we had in eq. (11) for the vector
operators, this reducible multiplet splits into irreducible multiplets of definite jnet ranging
from |j − k| to j + k, and the decomposition of states with definite µ and m into states with
definite jnet and mnet = µ + m follows the Clebbsch–Gordan coefficients
C(jnet , mnet |j1 = k, m1 = µ, j2 = j, m2 = m). Allowing for an overall (µ, m)-independent

7
coefficient — as well as changing α → α0 — due to a non-trivial action of the tensor operator
T̂ (k) , we end up with

j+k X
(k)
X
T̂µ |α, j, mi = α0 , j 0 , m0 = µ + m × C(j 0 , m0 |k, µ, j, m) ×
j 0 =|j−k| α0

× function(α0 , j 0 , α, j but not m0 , µ, m).

(28)
Or in terms of matrix elements of the spherical tensor operator,

(k) C(j 0 , m0 |k, µ, j, m)

α0 , j 0 , m0 T̂µ |α, j, mi = √ 0 × α0 , j 0 | T̂ (k) ||α, ji , (29)
2j + 1

where the reduced matrix element hα0 , j 0 || T̂ (k) ||α, ji depends on the multiplets (α, j) and
(α0 , j 0 ) and on the vector operator V̂ but does not depend on the multiplet members m or
m0 of the specific component µ of the tensor operator.

Implicit in the Clebbsch–Gordan coefficient in eq. (29) is the selection rule for the matrix
elements of the spherical tensor operator: The matrix element vanishes unless m0 = µ + m
while j, j 0 , and k obey the triangle rule,

j + j 0 ≥ k, j + k ≥ j 0, j 0 + k ≥ j. (30)

Equation (29) is the general form of the Wigner–Eckart theorem, the scalar operators
and the vector operators we have discussed earlier in these notes are simply special cases of
k = 0 for the scalars and k = 1 for the vectors.

Projection Theorem

The Wigner–Eckart theorem for the vector operators has a very useful corollary for their
matrix elements between states with the same j 0 = j:

hα0 , j, m| (Ĵ · V̂) |α, j, mi

α0 , j, m0 V̂i |α, j, mi = 2 × j, m0 Jˆi |j, mi . (31)
h̄ j(j + 1)

This formula is known as the projection theorem because for α0 = α and |ψi =
P
m cm |α, j, mi

8
eq. (31) becomes
hψ| (Ĵ · V̂) |ψi
hψ| V̂ |ψi = hψ| Ĵ |ψi , (32)
hψ| Ĵ2 |ψi

i.e., the expectation value of the vector V is the projection of V onto the direction of the
angular momentum.

Proof: By the Wigner–Eckart theorem

α0 , j, m0 V̂ i |α, j, mi = α0 , j| V̂ ||α, ji × Cm
i
0 ,m (j) (33)

i 0 √
where the Cm 0 ,m (j) factor combines the Clebbsch C(j, m |1, µ, j, m), the 1/ 2j + 1 factor,
and the translation from the spherical basis Vµ for the vectors to the Cartesian basis Vx,y,z .
Eq. (33) apples to matrix elements of any kind of a vector operator, including the angular
momentum Ĵ itself, thus

j, m0 Jˆi |j, mi = hj|| Ĵ ||ji × Cm

i
0 ,m (j) (34)

i
for exactly the same matrices Cm 0 ,m as in eq. (33).

Next, for the dot product Ĵ · V̂ we have

α0 , j, m (Ĵ · V̂) |α, j, mi = α0 , j, m Jˆi V̂ i |α, j, mi

hh inserting a complete basis of states between Jˆi and V̂ i ii
X
= α0 , j, m Jˆi α00 , j 0 , m0 × α00 , j 0 , m0 V̂ i |α, j, mi
α00 ,j 0 ,m0
hh where the first factor vanishes uness α00 = α0 and j 0 = j ii
X
= hj, m| Jˆi j, m0 × α0 , j, m0 V̂ i |α, j, mi
m0 only
X
i 0 i
= hj|| Ĵ ||ji Cm,m0 (j) × α , j| V̂ ||α, ji × Cm0 ,m (j)

m0
X
0 i i
= α , j| V̂ ||α, ji × hj|| Ĵ ||ji × Cm,m0 (j) Cm0 ,m (j),

m0
(35)

9
and in exactly similar fashion

X
hj, m| Ĵ2 |j, mi = hj|| Ĵ ||ji × hj|| Ĵ ||ji × i
Cm,m i
0 (j) Cm0 ,m (j) . (36)
m0

Consequently, taking the ratio of the last two formulae, we arrive at

hα0 , j, m| (Ĵ · V̂) |α, j, mi hα0 , j|| V̂ ||α, ji

= . (37)
hj, m| Ĵ2 |j, mi hj|| Ĵ ||ji

Finally, taking the ratio of eqs. (33) and (34), we get

hα0 , j, m0 | V̂i |α, j, mi hα0 , j|| V̂ ||α, ji hα0 , j, m| (Ĵ · V̂) |α, j, mi
= = , (38)
hj, m0 | Jˆi |j, mi hj|| Ĵ ||ji hj, m| Ĵ2 |j, mi

and therefore

hα0 , j, m| (Ĵ · V̂) |α, j, mi

α0 , j, m0 V̂i |α, j, mi = × j, m0 Jˆi |j, mi . (39)
hj, m| Ĵ2 |j, mi = h̄2 j(j + 1)

Quod erat demonstrandum.

Example: The magnetic moment of an electron is approximately

~µ̂ = −e (L̂ + 2Ŝ), (40)

2mc

so the magnetic moment operator for a whole atom is

~µ̂ −e

net = L̂net + 2Ŝnet . (41)
2mc

Now consider the expectation value of this operator in a state of definite net orbital angular
momentum L, definite net spin S, and definite net angular momentum J. The mJ quantum

10
number can have any value, or we can have a linear combination of states with different mJ .
By the projection theorem, in such a state
D E D E
~µ̂ · Ĵ D E
D E
~µ̂ = −e (L̂ + 2Ŝ) · Ĵ D E
D E Ĵ = D E Ĵ , (42)
Ĵ2 2mc Ĵ2

or in short,
~µ̂ = − ge Ĵ
D E D E
(43)
2mc
where
D E
(L̂ + 2Ŝ) · Ĵ
g = D E (44)
Ĵ2

is the gyromagnetic factor of the atom.

To evaluate eq. (44), we notice that

(L̂ + 2Ŝ) · Ĵ = (Ĵ + Ŝ) · Ĵ = Ĵ2 + Ŝ · Ĵ (45)

and

2Ŝ · Ĵ = Ĵ2 + Ŝ2 − (Ĵ − Ŝ = L̂)2 , (46)

thus
3 2 1 2 1 2
(L̂ + 2Ŝ) · Ĵ = 2 Ĵ + 2 Ŝ − 2 L̂ . (47)

Consequently, in a state with definite values of L, S, and J,

D E
3 2 1 2 1 2
(L̂ + 2Ŝ) · Ĵ 2 h̄ j(j + 1) + 2 h̄ S(S + 1) − 2 h̄ L(L + 1) (48)

and therefore
3 S(S + 1) − L(L + 1)
g = + . (49)
2 2J(J + 1)

For example, an oxygen atom in its ground state has L = 1, S = 1, J = 2, and hence g = 23 .

11