1

Addition of Angular Momentum

The discussion in this chapter closely follows [Shankar].

I. ADDITION OF 2 SPIN-1/2 PARTICLES

Consider a system of 2 spin-1/2 particles with spin operators S1 and S2 , whose orbital degrees
of freedom we ignore. The two-particle Hilbert space V1⊗2 is spanned by the four vectors

|s1 m1 ⟩ ⊗ |s2 m2 ⟩ ≡ |s1 m1 , s2 m2 ⟩, (1)

which obey

Si2 |s1 m1 , s2 m2 ⟩ = ℏ2 si (si + 1)|s1 m1 , s2 m2 ⟩

Siz |s1 m1 , s2 m2 ⟩ = ℏmi |s1 m1 , s2 m2 ⟩ (i = 1, 2). (2)

Let’s use the notation | + +⟩, | + −⟩, | − +⟩, | − −⟩ to denote the four states. They form the product
basis. We now try to determine the magnitude and z values for the total angular momentum
operator,

S = S1 + S2 , (3)
P
which obeys the commutation relations expected for a generator of rotations, [Si , Sj ] = iℏ k ϵijk Sk .

Our problem is to find the eigenvalues and eigenvectors of S 2 and Sz . Consider

Sz = S1z + S2z , (4)

which commutes with S12 , S12 , S1z and S2z . It is diagonal in the product basis,

Sz | + +⟩ = ℏ| + +⟩

Sz | + −⟩ = 0| + −⟩

Sz | − +⟩ = 0| − +⟩

Sz | − −⟩ = −ℏ| − −⟩, (5)

or, written as matrix,

 
1 0 0 0
 
0 0 0 0
 
Sz → ℏ 

.
 (6)
0 0 0 0
 
0 0 0 −1
 2

The eigenvalue sz = 0 is two-fold degenerate, and the eigenspace is spanned by | + −⟩ and | − +⟩.
Any linear combination of those will still result in an eigenvalue sz = 0, but this state will not have
definite values for S1z and S2z .
Consider next the operator

S 2 = (S1 + S2 ) · (S1 + S2 ) = S12 + S22 + 2S1 · S2 = S12 + S22 + 2S1z S2z + S1+ S2− + S1− S2+ . (7)

Although S 2 commutes with S12 and S22 , it does not commute with S1z and S2z . By explicit
computation,
 
2 0 0 0
 
0 1 1 0
 
2 2
S →ℏ  .
 (8)
0 1 1 0
 
0 0 0 2

The states of zero Sz , namely | + −⟩ and | − +⟩ are not eigenstates of S 2 . The following linear
combinations are however eigenstates,

| + −⟩ + | − +⟩
√ (s = 1)
2
| + −⟩ − | − +⟩
√ (s = 0). (9)
2

The allowed values for total spin are thus s = 1 and s = 0, while the allowed values of sz are ℏ, 0,
and −ℏ. The corresponding eigenstates in the product basis are

|s = 1, m = 1, s1 = 1/2, s2 = 1/2⟩ = | + +⟩
| + −⟩ + | − +⟩
|s = 1, m = 0, s1 = 1/2, s2 = 1/2⟩ = √
2
|s = 1, m = −1, s1 = 1/2, s2 = 1/2⟩ = | − −⟩
| + −⟩ − | − +⟩
|s = 0, m = 0, s1 = 1/2, s2 = 1/2⟩ = √ . (10)
2

The three s = 1 states are called triplets and the single spin-0 state is called a singlet. The
problem of addition of angular momentum is essentially a change of basis from one that diagonalizes
(S12 , S22 , S1z , S2z ) to one that diagonalizes (S 2 , Sz , S12 , S22 ). Symbolically, one writes

1/2 ⊗ 1/2 = 1 ⊕ 0. (11)

The total spin-s states have another property: they have definite symmetry under the exchange
of the two particles. The triplets are symmetric and the singlet is anti-symmetric. Now, the state
 3

vector of two identical spin-1/2 particles (these are necessarily fermions) must be antisymmetric
under the exchange of particle labels, ie under the exchange of their spin and orbital degrees of
freedom. Since for some orbital operator Ω,

|ω1 , ω2 , S⟩ = 2−1/2 [|ω1 ω2 ⟩ + |ω2 ω1 ⟩]

|ω1 , ω2 , A⟩ = 2−1/2 [|ω1 ω2 ⟩ − |ω2 ω1 ⟩] , (12)

are symmetric (S) and antisymmetric (A), we obtain wave functions that are antisymmetric by
combining the symmetric (antisymmetric) orbital wavefunction with the spin singlet (triplet).

II. GENERAL CASE

A. The problem

When adding two angular momenta J1 and J2 , what are the eigenvalues and eigenstates of J 2
and Jz where J = J1 + J2 ? One could construct the (2j1 + 1)(2j2 + 1) dimensional matrices J 2 and
Jz and diagonalize them. Now, Jz will be diagonal in the product basis itself for Jz |j1 m1 , j2 m2 ⟩ =
ℏ(m1 + m2 )|j1 m1 , j2 m2 ⟩. It will be a degenerate operator because there are many ways to build
up a total m = m1 + m2 , except in the extremal cases m = ±(j1 + j2 ). For each of the degenerate
subspaces of Jz we must choose a basis that diagonalizes J 2 (at the expense of undiagonalizing J1z
and J2z ). This can be a tedious business, but fortunately there exists a more efficient alternative
way to be described now.

As a first step we need to know all the values of j. This follows from (as a generalization of
what we found in the previous section), with j1 ≥ j2 ,

j1 ⊗ j2 = (j1 + j2 ) ⊕ (j1 + j2 − 1) ⊕ . . . ⊕ (j1 − j2 ). (13)

In other words, the total-j kets are

|jm, j1 j2 ⟩ j1 + j2 ≥ j ≥ j1 − j2 , j ≥ m ≥ −j. (14)

 4

Written in the form of an array (shown here for j1 = j2 ),

j1 + j2 j1 + j2 − 1 ... j1 − j2
|j1 + j2 , j1 + j2 ⟩
|j1 + j2 , j1 + j2 − 1⟩ |j1 + j2 − 1, j1 + j2 − 1⟩
|j1 + j2 , j1 + j2 − 2⟩ |j1 + j2 − 1, j1 + j2 − 2⟩
.. .. .. ..
. . . . (15)
|j1 + j2 , j1 − j2 ⟩ |j1 + j2 − 1, j1 − j2 ⟩ . . . |j1 − j2 , j1 − j2 ⟩
.. .. .. ..
. . . .
|j1 + j2 , −(j1 + j2 − 1)⟩ |j1 + j2 − 1, −(j1 + j2 − 1)⟩
|j1 + j2 , −(j1 + j2 )⟩

The top state in the first column, with m = j1 + j2 , can be built out of only one product ket, the
one in which both angular momenta take on maximum possible projections along the z-axis:

|j1 + j2 , j1 + j2 ⟩ = |j1 j1 , j2 , j2 ⟩. (16)

The other m states at this value of j are obtained by lowering. Let us consider going down just
one step. Since
p
J− |j1 + j2 , j1 + j2 ⟩ = ℏ 2(j1 + j2 )|j1 + j2 , j1 + j2 − 1⟩, (17)

and J− = J1− + J2− , we find that

 1/2  1/2
j1 j2
|j1 + j2 , j1 + j2 − 1⟩ = |j1 (j1 − 1), j2 j2 ⟩ + |j1 j1 , j2 (j2 − 1)⟩. (18)
j1 + j2 j1 + j2

Proceeding in this manner, we can go to the bottom of the first row (or to m = 0 and then use
special properties of Clebsch-Gordan coefficients which are discussed below for m < 0). The top
state in the second column has m = j1 + j2 − 1. There are 2 product kets eligible to enter the
linear combination, |j1 j1 , j2 (j2 − 1)⟩ and j1 (j1 − 1), j2 j2 ⟩. The combination must be normalized to
unity and orthogonal to the other state formed out of these kets, namely |j1 + j2 , j1 + j2 − 1⟩, and
by convention have real coefficients. the answer is, by inspection,
 1/2  1/2
j1 j2
|j1 + j2 − 1, j1 + j2 − 1⟩ = |j1 j1 , j2 (j2 − 1)⟩ − |j1 (j1 − 1), j2 j2 ⟩. (19)
j1 + j2 j1 + j2

The overall sign is fixed by the convention that the coefficient of the product ket with m1 = j1 be
positive. Given the top state, the rest of the second column may be obtained by lowering. The top
state in the third column, |j1 + j2 − 2, j1 + j2 − j − 2⟩ can be a superposition of three product kets.
 5

The three (real) coefficients can be determined from orthogonality to the two preceding total-j
kets of the same m and unit normalization. It is clear that there are always enough constraints
to determine the top states of each column, and once the top states are known, the rest follow by
lowering.

B. Clebsch-Gordan Coefficients

The completeness of the product kets allows us to write the total-j kets as
XX
|jm, j1 , j2 ⟩ = |j1 m1 , j2 m2 ⟩⟨j1 m1 , j2 m2 |jm, j1 j2 ⟩. (20)
m1 m2

The coefficients of the expansion

⟨j1 m1 , j2 m2 |jm, j1 j2 ⟩ = ⟨j1 m1 , j2 m2 |jm⟩ (21)

are called the Clebsch-Gordan coefficients or vector addition coefficients. Its properties are

• ⟨j1 m1 , j2 m2 |jm⟩ =
̸ 0 only if j1 − j2 ≤ j ≤ j1 + j2 . This is called the triangle inequality.

• ⟨j1 m1 , j2 m2 |jm⟩ =
̸ 0 only if m1 + m2 = m.

• By convention, they are real (but may be negative).

• ⟨j1 j1 , j2 (j − j1 )|jj⟩ is positive, by convention.

• ⟨j1 m1 , j2 m2 |jm⟩ = (−1)j1 +j2 −j ⟨j1 (−m1 ), j2 (−m2 )|j(−m)⟩.

If we assemble the CG coefficients into a matrix, we find that it is orthogonal (real and unitary),
since it relates one orthonormal basis to another. If we invert the matrix, we can write the
product kets in terms of total-j states. The coefficients in this expansion are also Clebsch-Gordan
coefficients,

⟨jm|j1 m1 , j2 m2 ⟩ = ⟨j1 m1 , j2 m2 |jm⟩∗ = ⟨j1 m1 , j2 m2 |jm⟩, (22)

because the Clebsch-Gordan coefficients are real. For example, for the 1/2 ⊗ 1/2 problem,
    
|1, 1⟩ 1 0 0 0 | + +⟩
   √ √  
 |1, 0⟩  0 1/ 2 1/ 2 0 | + −⟩
    
 =  . (23)
|1, −1⟩ 0 0 | − +⟩
    
0 1  
   √ √   
|0, 0⟩ 0 1/ 2 −1/ 2 0 | − −⟩
 6

This can be inverted,

    
| + +⟩ 1 0 0 0 |1, 1⟩
   √ √   
| + −⟩ 0 1/ 2 0 1/ 2   |1, 0⟩ 
    
 = √ √ 
  . (24)
| − +⟩ 0 1/ 2 0 −1/ 2 |1, −1⟩
   
    
| − −⟩ 0 0 1 0 |0, 0⟩

In practice, one uses Clebsch-Gordan coefficients to go both ways.

C. Addition of L and S

Since an electron has spin 1/2, its total angular momentum J = L + S can have values of
j = l ± 1/2. We wish to express the total-j states in terms of product states |lmo , sms ⟩. Since
ms = ±1/2 there will be at most two product kets,

|j = l + 1/2, m⟩ = α|l, m − 1/2; 1/2, 1/2⟩ + β|l, m + 1/2; 1/2, −1/2⟩

|j = l − 1/2, m⟩ = α′ |l, m − 1/2; 1/2, 1/2⟩ + β ′ |l, m + 1/2; 1/2, −1/2⟩ (25)

Imposing orthonormality leads to α2 + β 2 = 1, α′2 + β ′2 = 1, and αα′ + ββ ′ = 0. We need one

more constraint, say the ratio α/β, which we find by demanding that

J 2 |j = l + 1/2, m⟩ = ℏ2 (l + 1/2)(l + 3/2)|j = l + 1/2, m⟩. (26)

Writing J 2 = L2 + S 2 + 2Lz Sz + L− S+ + L+ S− we find that

l + 1/2 − m 1/2
 
β
= . (27)
α l + 1/2 + m

With the convention for the overall sign,

1 p
|j = l ± 1/2, m⟩ = [± l + 1/2 ± m|l, m − 1/2; 1/2, 1/2⟩
(2l + 1)1/2
p
+ l + 1/2 ∓ m|l, m + 1/2; 1/2, −1/2⟩]. (28)

HW. Show that P1 = 34 I + (S1 · S2 )/ℏ2 and P0 = 14 I − (S1 · S2 )/ℏ2 are (i) projector operators,
(ii) that project on the spin-1 and spin-0 spaces.

D. The modified spectroscopic notation

In the absence of spin, it is sufficient to use a single letter such as s, p, d, . . . to denote the orbital
momentum of a particle. In the presence of spin, the notation is changed as follows
 7

• Use capital letters S, P, D, . . . to indicate the value L of the orbital angular momentum.

• Append a subscript J to the right of L to indicate the j value.

• Append a superscript 2S + 1 to the left of L to indicate the multiplicity due to spin projec-
tions.

For example,

2S+1
LJ = 2 P3/2 (29)

denotes a state with l = 1, s = 1/2, j = 3/2. The ground state of He is 2S+1 L = 1 S0 .

J

III. IRREDUCIBLE TENSOR OPERATORS

A. Scalar and Vector operators

A scalar operator S transforms like a scalar under rotations, ie remains invariant under

S → S ′ = U † [R]SU [R] = S. (30)

By considering infinitesimal rotations, we deduce that [Ji , S] = 0, or [J± , S] = 0 and [Jz , S] = 0.

Examples of S are rotationally invariant Hamiltonians such as the Coulomb or isotropic oscillator
Hamiltonian.
A vector operator V was defined as a collection of three operators (Vx , Vy , Vz ) which transform
as the components of a vector in V 3 (R),
X
Vi → Vi′ = U † [R]Vi U [R] = Rij Vj , (31)
j

where R is the usual 3 × 3 rotation matrix. By replacing R by R−1 = RT in Eq.(31), we get the
equivalent form
X
U [R]Vi U † [R] = Rji Vj . (32)
j

We are summing now over the first index of R. The relation between Eq.(31) and Eq.(32) is the
following. Let |1⟩, |2⟩, and |3⟩ be basis kets in V 3 (R) and R a rotation operator on it. If |V ⟩ is
some vector with components vi = ⟨i|V ⟩, its rotated version |V ′ ⟩ = R|V ⟩ has components
X X
vi′ = ⟨i|R|V ⟩ = ⟨i|R|j⟩⟨j|V ⟩ = Rij vj . (33)
j j
 8

If we instead ask what R does to the basis, we find |i⟩ → |i′ ⟩ = R|i⟩ where
X X
|i′ ⟩ = R|i⟩ = |j⟩⟨j|R|i⟩ = Rji |j⟩. (34)
j j
−1
Since Rji = Rij we see that basis vectors and vector components transform in opposite ways.
Both definitions are of course equivalent. By considering infinitesimal rotations, we deduce from
P
Eq. 32 that vector operators obey [Vi , Jj ] = iℏ k ϵijk Vk where i, j, k = x, y, z.

B. Tensor operators

A second-rank tensor |T (2) ⟩ is an element of a direct product space V 3 (R) ⊗ V 3 (R) spanned by
the nine kets |i⟩ ⊗ |j⟩,
3 X
X 3
(2)
|T ⟩= tij |i⟩ ⊗ |j⟩. (35)
i=1 j=1

A tensor operator of rank 2 is a collection of nine operators Tij which, under Tij → U † Tij U respond
as do the tensor components tij , or, equivalently, under Tij → U Tij U † , respond as do the basis
kets |i⟩ ⊗ |j⟩. Tensors and tensor operators of rank n > 2 are defined in a similar way. We call
these tensors Cartesian tensors.
A spherical tensor operator of rank k has (2k + 1) components Tkq , q = k, (k − 1), . . . , −k which,
under Tkq → U Tkq U † respond like the angular momentum eigenkets |j = k, m = q⟩ = |kq⟩,
X (k) q′
U [R]Tkq U † [R] = Dq′ q Tk . (36)
q′
(k)
|k ′ q ′ ⟩⟨k ′ q ′ |U [R]|kq⟩ = |k ′ q ′ ⟩⟨kq ′ |U [R]|kq⟩ = Dq′ q′ |kq ′ ⟩).
P P P
(Recall that U [R]|kq⟩ = k′ q ′ q′ q′

Since the (2k + 1) kets |kq⟩ transform irreducibly, so do the operators Tkq . They are called ir-
reducible tensor operators. By considering infinitesimal rotations, we deduce that

[Jz , Tkq ] = ℏqTkq , (37)

and

[J± , Tkq ] = ℏ[(k ∓ q)(k ± q + 1)]1/2 Tkq±1 . (38)

Acting with Tkq on a state |jm⟩ adds angular momentum (k, q) to the state. This can be seen from

U [R]Tkq |jm⟩ = U [R]Tkq U † [R]U [R]|jm⟩

X (k) q′ X (j)
= Dq′ ,q Tk Dm′ ,m |jm′ ⟩
q′ m′
(k) (j) ′
Dq′ ,q Dm′ ,m Tkq |jm′ ⟩.
XX
= (39)
q′ m′
 9

We find that Tkq |jm⟩ responds to rotations like the product ket |kq⟩ ⊗ |jm⟩. An irreducible tensor
operator Tkq imparts a definite amount of angular momentum (k, q) to the state it acts on. This
allows to say

⟨α′ j ′ m′ |Tkq |αjm⟩ = 0 unless k + j ≥ j′ ≥ |k − j|, m′ = m + q. (40)

Let us consider some examples. The tensor operator of rank 0 has only one component, T00 ,
which transforms like |00⟩, ie remains invariant. Thus T00 is just a scalar operator S. The selection
rule says

⟨α′ j ′ m′ |T00 |αjm⟩ = 0 unless j = j′ , m = m′ . (41)

Consider next T1q (q = 1, 0, −1). Since a vector operator V also has three components that trans-
form irreducibly into each other, there must be some linear combination of the vector operator
components that equals T1q ,

Vx ± iVy
T1±1 = ∓ √ ≡ V1±1
2
T10 = Vz ≡ V10 . (42)

It can readily be verified that V1±1 and V10 obey the commutation relations, Eqs. 37 and 38.
Once we go beyond rank 1, it is no longer possible to express Cartesian and spherical tensors of
the same rank in terms of each other. A Cartesian tensor of rank n has 3n components, whereas
a spherical tensor of rank k has (2k + 1) components. For n = 0 and n = 1 the Cartesian tensors
happened to have the same number of components as spherical tensors of rank k = 0 and k = 1,
and also transformed irreducibly. But for tensors of higher rank this is no longer true. The tensor
T2q has five components that transform irreducibly, but the tensor Tij has nine components which
transform reducibly; ie, it is possible to form combinations of Tij such that some of them never
mix with others under rotations. There is one combination which is invariant (transforms like T00 ),
there are three combinations that transform like a T1q and five that transform like T2q . (We will see
later what they are).
We can go one step further and relate the non-vanishing matrix elements. Consider for example
the position operator in spherical form, R1q . We have, in the absence of spin (J = L),
Z  1/2
4π
⟨α2 l2 m2 |R1q |α1 l1 m1 ⟩ = Rα∗ 2 l2 (r)Ylm 2∗
(θ, ϕ)r Y1q Rα1 l1 (r)Ylm 1
(θ, ϕ)r2 drdΩ
2 3 1

4π 1/2
  Z Z
2∗
= Rα∗ 2 l2 rRα1 l1 r2 dr Ylm Y1q Ylm 1
dΩ
3 2 1

= ⟨α2 l2 ||R1 ||α1 l1 ⟩⟨l2 m2 |1q, l1 m1 ⟩ (43)

 10

where ⟨α2 l2 ||R1 ||α1 l1 ⟩, the reduced matrix element, is independent of m1 , m2 and q. Those appear
only in the Clebsch-Gordan coefficient, which is essentially the angular integral up to a factor
independent of m1 , m2 and q. This example illustrates a more general result,

⟨α2 j2 m2 |Tkq |α1 j1 m1 ⟩ = ⟨α2 j2 ||Tk ||α1 j1 ⟩⟨j2 m2 |kq, j1 m1 ⟩. (44)

This is called the Wigner-Eckart theorem. It separates the matrix element on spatial separation
(m2 , m1 and q from the rest). The former is expressed entirely in terms of the Clebsch-Gordan
coefficients.