You are on page 1of 4

ROTATIONS IN QUANTUM MECHANICS

TODD KARIN

Abstract. Rotations in quantum mechanics are an important yet difficult topic for physics graduate students. The following
summary comes from a graduate quantum mechanics course taught by Prof. Andreas Karch at University of Washington
(winter 2012). Most of the rest comes from the course text [2], however a few of the sections are my own ideas. This article
does not attempt to prove every result, but often the most important results can be proven in a few lines. Learning these short
proofs is essential to understanding the whole picture.

1. Rotations and Commutation Relations We postulate that the rotation operators satisfy the com-
position law that ordinary rotations matrices satisfy, namely
1.1. Finite versus Infinitesimal Rotations. A prerequi-
site is familiarity with the usual 3D rotations R about an axis D(R2 )D(R1 ) = D(R2 R1 )
by a certain angle. If we write a finite rotation in the limit 2
as the rotation angle gets small, we will get infinitesimal ro- Using (2) and keeping only terms to dφ , we find:
tation. To first order in the rotation angle, rotations about (5) [Ji , Jj ] = i~ijk Jk
different axes will commute, to second order, they will not.
In order to compute the product D2 D1 , we may be tempted
1.2. Infinitesimal Rotations in Quantum Mechanics. to use the operator identity:
Rotations affect physical systems. With every physical rota-
tion R, we associate a quantum mechanical rotation operator (6) eA eB = eA+B e[A,B]/2
D(R) which rotates quantum mechanical states:
Unfortunately the above only holds if [A, B] commutes with
(1) |αiR = D(R)|αi. A and B. Because of (5), this formula does not apply.
In order to construct the rotation operator, we first write Instead, we must use the generalization of this identity, the
down the operator D(n̂, dφ), which corresponds to an infini- Baker-Campbell-Hausdorff series. This series includes many
tesimal rotation. We know that D(n̂, dφ) must be the identity commutators, but since we know how the J commute, the
if dφ = 0 and also be linear in dφ for small rotations. We can product of two rotation operators is known in theory.
therefore write:
  1.4. SO(3), SU(2), and Euler Rotations. The group
J · n̂
(2) D(n̂, dφ) = 1 − i dφ SO(3) is the set of special orthogonal 3x3 matrices. Special
~
matrices are rotations; no inversions are allowed in the group.
viewing (2) as a definition of the angular-momentum operator The group SU(2) is the set of special unitary 2x2 matrices.
J. We call Jk a generator of rotation about the k-th axis. In an Euler rotation, we construct an arbitrary rotation
Any finite rotation can be built from infinitesimal ones: out of 3 rotations:
 N
J · n̂ φ D(α, β, γ) = Dz (α)Dy (β)Dz (γ).
(3) D(n̂, φ) = lim 1 − i
N →∞ ~ N
 
n̂ · J 2. Eigenvalues and Eigenstates of Angular
(4) = exp −iφ
~ Momentum
1.3. Rotations are a group. A group is a set together with 2.1. Commutation relations and ladder operators. Ev-
a multiplication rule that satisfy specific properties. A group ery result in this section is an algebraic result of the commuta-
contains an identity, inverses, is closed under multiplication, tion relations (5). Since [J2 , Jz ] = 0, eigenstates of J2 can be
and is associative. chosen to be simultaneous eigenstates of Jz ; call these |jmi.
The set of rotation operators D(R) form a group. To prove If we define the operators
this we need to show that the product of two rotations oper-
ators is another rotation operator. How can we compute the J± = Jx ± iJy ,
product D1 D2 ? Using (4), we suspect that if we know how and compute their action on an eigenket, we find
to multiply different components of J, then we will be able to p
multiply two rotation operators D1 and D2 . (7) J± |j, mi = (j ∓ m)(j ± m + 1)~|j, m ± 1i.
Date: March 2012.
Contact: Todd Karin. University of Washington. tkarin (at) uw (dot) edu.
1
2 TODD KARIN

In order for the ket on the right to be normalized, we must 3.2. Spherical Harmonics. Because [L2 , Lz ] = 0, we can
restrict m to (2j + 1) values: try to find simultaneous eigenkets of L2 and Lz :
m = −j, j + 1, . . . , j − 1, j. L2 |lmi =l(l + 1)~2 |lmi
Lz |lmi =m~|lmi
2.2. Representations of the Rotation Operator. We are
often interested in the matrix elements of the rotation opera- Spherical harmonics are these eigenkets in a direction basis:
tor in the Jz basis,
  hn̂|lmi = Ylm (n̂)
(j) J · n̂φ
(8) Dm0 m (α, β, γ) = hjm0 | exp −i |jmi
~ 3.3. Spherical Harmonics as Rotation Matrices. We
called the (2j + 1)-dimensional irreducible representation or can construct the ket |n̂i from by rotating |ẑi about the y-axis
wigner d-matrix. We only consider matrix elements with the by θ and then about the x-axis by φ. Denoting this rotation
same j on the left and right because rotations don’t change by R,
X
the total-j value. To see this, we use the very useful fact |n̂i = D(R)|ẑi = D(R)|lmihlm|ẑi
that the dot product of two vectors is a scalar and therefore lm
doesn’t change under rotations. For example, if L and S are Using (8),
vector operators (see sec. 5.2), then X (l)
hlm0 |n̂i = Dm0 m (R)hlm|ẑi
[L · S, D(R)] = 0, [L · S, J] = 0 m

where the second relation comes from the fact that J corre- Evaluating the spherical harmonics at ẑ, we can write
sponds to an infinitesimal rotation. Using this fact, we can
r
(l) 4π
prove that the Wigner d -matrix does not change mix j: Dm0 (α, β, γ = 0) = Y m ∗ (θ = β, φ = α).
2l + 1 l
J2 D|jmi = DJ2 |jmi = j(j + 1)~2 D|jmi which gives the connection between the rotation operators
Any matrix representation of a rotation operator D(R) can and spherical harmonics.
be brought into block-diagonal form using a change of basis.
(j) 4. Addition of Angular Momentum
The blocks are the square matrices Dm0 m (R).
(j)
The set of Dm0 m (R) with a definite j form a group.4.1. Rotation Operators. The basis states of a system with
multiple degrees of freedom can be written as the direct prod-
3. Orbital Angular Momentum uct of the basis states of each degree of freedom. For example,
if we wanted to write basis states for an electron with both a
3.1. Orbital Angular Momentum as Rotation Gener-
position and spin, we would write:
ator. The orbital angular momentum operator L = r × p
satisfies the commutation relations (9) |x, ±i = |xi ⊗ |±i
[Li , Lj ] = i~ijk Lk . To rotate this state, we apply a rotation operator in each of
Does L generate rotations? We find out by acting a candidate the subspaces:
rotation operator on a position eigenket |x, y, zi. Using the (10) D(R) = Dorb (R) ⊗ Dspin (R)
fact that linear momentum generates displacement
p · dx 4.2. Formal theory. If we want to add two angular mo-
T (dx) = 1 − i , menta J1 and J2 , we first note that the total angular mo-
~
mentum J = J1 + J2 satisfies the commutation relations
we can prove that orbital angular momentum generates rota-
tions 0 = [J2 , J21 ] = [J2 , J22 ] = [J2 , Jz ]
 
dφ Therefore we can alternately describe the states of the com-
1 − i Lz |x, y, zi = |x − ydφ, y + xdφ, zi.
~ posite system in the product basis:
The orbital angular momentum operators can be written |j1 , j2 ; m1 m2 i
in spherical coordinates:
or in the total-j basis
 
∂ ∂ |j1 , j2 ; jmi
Lx = − i~ − sin φ − cot θ cos φ
∂θ ∂φ
  By inserting a complete set of states,
∂ ∂
Ly = − i~ cos φ − cot θ sin φ X
∂θ ∂φ |j1 , j2 ; jmi = |j1 , j2 ; m1 m2 ihj1 , j2 ; m1 m2 |j1 , j2 ; jmi
∂ m1 ,m2
Lz = − i~
∂φ The numbers are called the Clebsch-Gordan coefficients.
ROTATIONS IN QUANTUM MECHANICS 3

4.3. Direct sums and products. Confusion arises over the 5. Tensor Operators
the direct product operator ⊗ because it is used to mean two
5.1. Active versus Passive Transformations. Up until
mathematically different things. On the one hand, ⊗ is used
now, we have discussed active rotations in which the state
to form product kets from two kets as in (9). In this context,
kets rotate and the “coordinate axes” don’t. Physically, this
it is a way of describing a composite system using a single
means that our measuring devices aren’t moved. Therefore,
vector space.
we conclude that quantum mechanical operators do not change
The symbol ⊗ also means the sum or product of the ba-
under active rotations.
sis vectors of two Hilbert spaces. For example, 12 ⊗ 12 stands 
for the set of basis vectors formed by multiplying the basis |αi → D|αi
vectors of the subspaces. That is: (12) Active Rotation
O→O
1 1 On the other hand, in passive rotations, the state kets
⊗ = {|+i, |−i} ⊗ {|+i, |−i}
2 2 aren’t moved while the measuring devices rotate by R.
= {| + +i, | + −i, | − +i, | − −i} 
|αi → |αi
(13) Passive Rotation
On the other hand the direct sum ⊕ of two hilbert spaces O → OR
stands for the set of basis vectors we get by simply adding
together the basis vectors of the subspaces. For example: The active and passive transformations must yield the same
physical results if the measuring devices are rotated in the op-
1 ⊕ 0 = {|11i, |10i, |1, −1i} ⊕ {|00i} posite direction as the system. Therefore:
= {|11i, |10i, |1, −1i, |00i}
(14) hα|O|αi → hα|D† (R)OD(R)|αi = hα|OR−1 |αi
Writing
and so
1 1
⊗ ≡1⊕0 OR−1 = D† (R)OD(R)
2 2
means that the sets of basis vectors on the left and right are Using the fact that D(R−1 ) = D† (R), we conclude that if the
equivalent; i.e. can be gotten from each other with a change measuring devices are rotated by R, then we should send the
of basis (the Clebsch-Gordan coefficients). It is common prac- operators to
tice to write ‘=’ instead of ‘≡’ to mean the same thing.
If we want to add together any two angular momenta j1 (15) OR = D(R)OD† (R)
and j2 , we use the formula
In retrospect, we can consider (14) either as an active rotation
(11) j1 ⊗ j2 ≡ (j1 + j1 ) ⊕ (j1 + j2 − 1) ⊕ · · · ⊕ |j1 − j2 | by R or a passive rotation by R−1 .
Except where explicitly stated, the rest of this article only
which describes which total angular momentum states are uses active rotations.
formed from adding two angular momenta.
5.2. Vector Operators. A vector in classical physics is a set
4.4. Obtaining Clebsch-Gordan coefficients. The of numbers that transform under rotations as
Clebsch-Gordan coefficients are zero unless m = m1 + m2 .
This can be seen by multiplying by 0: vj → Rij vj .

(J1 − J1z − J2z )|j1 j2 ; jmi = 0 In quantum mechanics, we define a vector operator as a set
of operators whose expectation values transform as a classical
and so vector.
hα|Vi |αi → Rij hα|Vj |αi
(m − m1 − m2 )hj1 , j2 ; m1 m2 |j1 , j2 ; jmi = 0
But also, by (14),
The Clebsch-Gordan coefficients are obtained from the fol-
lowing procedure (I will neglect to write j1 , j2 as it is under- hα|Vi |αi → hα|D† (R)Vi D(R)|αi
stood that we are adding two angular momenta.)
and so a vector operator satisfies
(1) The top state is unique:
(16) D† (R)Vi D(R) = Rij Vj .
|m1 = j1 , m2 = j2 i = |j = j1 + j2 , m = j1 + j2 i
Note that from (15) it would be wrong to consider
(2) Apply J− (eq. 7) to the left and right sides to get a tower. D† (R)V D(R) as the vector V rotated by R. Equation 16
i i
(3) Use orthogonality to move to the next tower. is a mathematical relation for vector operators in the active
This procedure is described in detail in many textbooks. picture.
4 TODD KARIN

5.3. Cartesian and Spherical Tensors. We define a carte- This is proven by the explicitly checking the rotation proper-
sian tensor of rank 2 as a set of numbers that transforms as ties of vm :
(17) Tij → Rii0 Rjj 0 Ti0 j 0 vm → U exp(−iT · α)vi
The generalization to higher order tensors is straightforward: = U exp(−iT · α)U † vm
more copies of Rkk0 = exp(−iJ · α)vm
A spherical tensor of rank-k, Tq (k) , is defined by
where the last step comes from expanding the matrix expo-
k
(k) †
X (k) (k) nential, using (21), inserting a bunch of U U † = 1, and putting
(18) D(R)Tq D (R) = Dq0 q (R)Tq0
back together. We have therefore proved that vm transforms
q 0 =−k
like a spherical vector.
Note that the rotation properties of higher rank spherical ten- For rank-2 tensors, the story is slightly more complicated.
sors are much easier to write down than for cartesian tensors. Similarly to rank-1, we expect
Writing this out in terms of infinitesimals we find an equiv-
alent definition of spherical tensors: (22) Tm1 m2 = Um1 i Um2 j Tij
might be a spherical tensor. However, it turns out that Tm1 m2
(19) [Jz , Tq(k) ] =~qTq(k)
doesn’t transform as a spherical tensor of a definite rank. Us-
(k)
p
(20) [J± , Tq(k) ] =~ (k ∓ q)(k ± +1)Tq±1 ing an analogy with angular momentum addition, we form
the spherical tensors
5.4. Relation between cartesian and spherical tensors. X
(j)
The basic definition of a rotation for a cartesian vector is (23) Tm = hjm|11m1 m2 iTm1 m2 , j = 0, 1, 2
m1 ,m1
vj → exp(−iT · α)vj
using the Clebsch-Gordan coefficients. Basically, we are
where the rotation angle α = αn̂, T · α = (Ti )jk αi and the
adding together two angular momentum 1 to get
generators of rotation are
1 ⊗ 1 = 2 ⊕ 1 ⊕ 0.
(Ti )jk = iεijk .
The Clebsch-Gordan coefficients choose the correct compo-
This is just a rewriting of the usual rule to rotate a vector
nents of the individual spin-1’s to add together. One should
with a 3 × 3 matrix of cosines and sines. On the other hand,
check that these tensors (23) do indeed transform as spheri-
for a spherical vector,
cal tensors of rank-j. Equations 22 and 23 shows us how to
vm → exp(−iJ · α)vm . construct a rank-2 spherical tensor from its cartesian compo-
I will use the convention that vi , i = 1, 2, 3 are the cartesian nents.
components of a vector and vm , with m = −1, 0, 1 are the 5.5. Wigner-Eckart Theorem. The matrix elements of
spherical components of the vector. tensor operators with respect to angular momentum eigen-
How can we get the spherical components of a vector from states satisfy
its cartesian components? I claim that there exists a unitary
hα0 j 0 ||T (k) ||αji
matrix U such that (24) hα0 j 0 m0 |Tq(k) |αjmi = hjk; mq|jk; j 0 m0 i

2j + 1
(21) Jk = UTk U†
where the reduced matrix element (the double-bar number) is
This can be proven by noting that Jz is diagonal in the Jz a number not depending on m0 , m, or q. To find it, evaluate
basis, so we are really asking for the matrix U that diagonal- the right hand side of
izes Tz . By choosing the order and phase of the eigenvectors √
0 0 (k) 0 0 0 (k) 2j + 1
correctly, U will also work for the x and y components of (21). hα j ||T ||αji = hα j m |Tq |αjmi
Using this procedure, we find that hjk; mq|jk; j 0 m0 i
0
  for some convenient choice of m , m, or q. [1]
− √12 √i2 0
U = 0 0 1 
 
References
1 i

2

2
0 [1] Kurt Gottfried and Tung-mow Yan. Quantum mechanics : funda-
A spherical vector is related to its cartesian counterpart by mentals. Springer, 2003.
[2] J. J. Sakurai and Jim J. Napolitano. Modern Quantum Mechanics
vm = Umj vj . (2nd Edition). Addison Wesley, 2 edition, July 2010.

You might also like