Mecânica Quântica I

Prof. Dr. Matthew Luzum

• Idioma: Português

• Site: http://matt.luzum.org/Home/MQI2023

• Nota final: média ponderada das listas de exercı́cios (10%) + os melhores 2 dos 3
exames (45% cada)

• Texto principal: Sakurai, “Mecânica Quântica Moderna”

• Outros: A.F.R de Toledo Piza, Mecânica Quântica; Shankar, “Principles of Quantum

Mechanics”; Weinberg, “Lectures on Quantum Mechanics”

• Sala: 2011 (Auditório Gleb Wataghin)

Contents
1 Structure of Quantum Mechanics 5
1.1 Hilbert space and operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 Linear Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.2 Inner Product Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.3 Gram-Schmidt procedure . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.4 Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.1.5 Generalized Uncertainty Principle . . . . . . . . . . . . . . . . . . . . 15
1.2 Application of postulates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3 Mixed states and the density operator . . . . . . . . . . . . . . . . . . . . . 21
1.3.1 Density Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2 Spatial Degrees of Freedom 24

2.1 Projection Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2 Configuration Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Translation Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1
3 Time evolution 33
3.0.1 Hamiltonian as generator of time evolution . . . . . . . . . . . . . . . 33
3.0.2 Heisenberg vs Schrödinger picture . . . . . . . . . . . . . . . . . . . . 34
3.0.3 Initial value problem for time-independent Ĥ . . . . . . . . . . . . . 36

4 Harmonic Oscilator 38
4.1 HO as an approximation to a general potential problem . . . . . . . . . . . . 38
4.2 1D HO solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

5 Path Integrals 45
5.1 Propagator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.1.1 Propagator for free particle . . . . . . . . . . . . . . . . . . . . . . . 46
5.2 Path integral in 1D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.3 Visualization, and Integration in Path Space . . . . . . . . . . . . . . . . . . 48
5.4 Stationary Action and Hamilton’s Principle . . . . . . . . . . . . . . . . . . 49
5.4.1 Stationary Phase approximation . . . . . . . . . . . . . . . . . . . . . 49
5.4.2 Stationary Phase approximation applied to the path integral . . . . . 51
5.5 Aside: Gaussian integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

6 Some topics in 1D wave mechanics 56

6.1 Degeneracies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
6.2 Reality of energy eigenfunctions . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.3 Interleaving of nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

7 Particles in Electromagnetic Fields 60

7.1 Gauge Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
7.2 Aharonov-Bohm effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

8 Angular Momentum 66
8.1 Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
8.2 Proper rotations, SO(D) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
8.2.1 O(D), SO(D), and parity . . . . . . . . . . . . . . . . . . . . . . . . . 66
8.3 Coordinate Rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
8.4 Quantum rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
8.4.1 Infinitesimal rotations . . . . . . . . . . . . . . . . . . . . . . . . . . 71
8.4.2 Spin 1/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
8.4.3 Double-valued representation . . . . . . . . . . . . . . . . . . . . . . 74

2
 8.4.4 Neutron interferometry experiment to study 2π rotations . . . . . . . 76
8.5 Euler Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
8.6 Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
8.6.1 Matrix elements and phase conventions . . . . . . . . . . . . . . . . . 83
8.6.2 Invariant, irreducible subspaces . . . . . . . . . . . . . . . . . . . . . 84
8.6.3 Matrices for angular momentum and operators . . . . . . . . . . . . . 86
8.7 Orbital Angular Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.7.1 Orbital angular momentum in spherical coordinates . . . . . . . . . . 89
8.7.2 Spherical harmonics and rotation operators . . . . . . . . . . . . . . . 93
8.8 Wigner’s formula for d matrices . . . . . . . . . . . . . . . . . . . . . . . . . 94

9 Particle in a central potential 95

9.1 Behavior at small radius . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
9.2 Free particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
9.3 Two-body central force motion . . . . . . . . . . . . . . . . . . . . . . . . . 99
9.4 Hydrogen Atom (Single electron in Coulomb potential) . . . . . . . . . . . . 103

10 Addition of angular momenta 108

10.0.1 Tensor products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
10.1 Spin + Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
10.2 Operator tensor product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
10.3 Addition of angular momenta . . . . . . . . . . . . . . . . . . . . . . . . . . 110
10.4 Simple example: 2 spin-1/2 particles . . . . . . . . . . . . . . . . . . . . . . 111
10.5 Allowed values of j and m . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
10.6 Clebsch-Gordon Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
10.7 Irreducible tensor operators and the Wigner-Eckart Theorem . . . . . . . . . 116
10.7.1 Scalar operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
10.7.2 Vector operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
10.7.3 Tensor operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
10.7.4 Irreducible tensor operators . . . . . . . . . . . . . . . . . . . . . . . 119
10.7.5 Products of tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
10.7.6 Three Ylm Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
10.7.7 Matrix Elements of Tensor Operators and the Wigner-Eckart theorem 124
10.7.8 Proof of Wigner-Eckart Theorem . . . . . . . . . . . . . . . . . . . . 125
10.7.9 Examples of Wigner-Eckart and the projection theorem . . . . . . . . 126

3
11 Approximation Methods 129
11.1 Time-independent perturbation theory (nondegenerate) . . . . . . . . . . . . 129
11.2 Time-independent perturbation theory with degeneracy . . . . . . . . . . . . 134
11.3 Fine structure of the hydrogen atom . . . . . . . . . . . . . . . . . . . . . . 139
11.4 Variational Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
11.5 Time-dependent perturbation theory . . . . . . . . . . . . . . . . . . . . . . 148
11.5.1 Dyson Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
11.5.2 Examples: Time-independent H1 . . . . . . . . . . . . . . . . . . . . 153
11.5.3 Examples: Periodic H1 . . . . . . . . . . . . . . . . . . . . . . . . . . 154

4
1 Structure of Quantum Mechanics
I assume you have already seen some quantum mechanics, and know why it is useful.
=⇒ I will skip the motivation and historical introduction, and start directly with a
mathematical definition.
I will be more abstract and careful than a typical undergraduate, but concise (omitting
mathematical details)
I list below some standard postulates of quantum mechanics (as written in the notes of
Jaffe http://web.mit.edu/8.05/handouts/jaffe1.pdf).
Afterward, I will explain in more detail the meaning of each postulate

1. At each instant t, the state of a physical system is represented by a ket |ψi in the space
of states (“Hilbert space”).

2. Every observable attribute of a physical system is described by an operator that acts

on the kets that describe the system.

3. The time evolution of an isolated quantum system preserves the normalization of the
associated ket.
=⇒ The time evolution of the state of a quantum system is described by |ψ(t)i =
Û (t, t0)|ψ(t0 )i, for some unitary operator Û .

4. The only possible result of the measurement of an observable A is one of the eigenvalues
of the corresponding operator A.

5. When a measurement of an observable A is made on a generic state |ψi, the probability

of obtaining an eigenvalue an is given by the square of the inner product of |ψi with
the eigenstate |an i, |han |ψi|2 . (“Born rule”)

6. Immediately after the measurement of an observable A has yielded a value an , the state
of the system is the normalized eigenstate |an i .

1.1 Hilbert space and operators

A quantum state is a vector in an abstract vector space (following Shankar ch 1)

1.1.1 Linear Vector Spaces

Linear vector space X on a field F:

5
Collection of “vectors” |V i, |W i, . . . ∈ X
with definition for vector sum
X × X → X:
|V i + |W i
and scalar multiplication
X × F → X:
a|V i, a ∈ F
with properties

1. Closure: |V i + |W i ∈ X (superposition)

2. Distributive:

(a) Vectors: a(|V i + |W i) = a|V i + a|W i

(b) Scalars: (a + b)|V i = a|V i + b|V i

3. Commutative: |V i + |W i = |W i + |V i

4. Associative

(a) Scalar multiplication a(b|V i) = ab|V i

(b) Addition |V i + (|W i + |Zi) = (|V i + |W i) + |Zi

5. Identity

(a) ∃ null vector |0i: |V i + |0i = |V i

(b) 1|V i = |V i

6. ∃ additive inverse: |V i + | − V i = |0i, ∀|V i

a, b, . . . = “field” F of vector space: complex numbers =⇒ complex vector space

Note that this implies

• |0i is unique

• 0|V i = |0i

• | − V i = (−1)|V i

• | − V i is unique additive inverse

6
 Linear independence
A set of vectors |ii are linearly independent when the only solution to

X
n
ai |ii = |0i (1)
i

is ai = 0, ∀i
=⇒ not possible to write any element as linear combination of others

X −ai
|ji = |ii (2)
i6=j
aj

Vector space X has dimension N if there are a maximum of N linearly independent

vectors
=⇒ N linearly independent vectors form a “basis” that spans the space X
Dimension N can be finite or infinite (countable or uncountable)
=⇒ any vector in X can be written as a linear superposition of elements of basis

X
N
|V i = vi |ii (3)
i

vi = components of |V i
vi depends on basis

1.1.2 Inner Product Spaces

Vector space with definition for inner product g on X

g : X × X → F(= C) with properties

1. hV |W i = hW |V i∗ (skew-symmetry)

2. hV |V i ≥ 0, 0 iff |V i = |0i


3. hV | a|W i + b|Zi = hV |aW + bZi = ahV |W i + bhV |Zi (linearity in ket)

=⇒ haW + bZ|V i = hV |aW + bZi∗ (4)

= a∗ hW |V i + b∗ hZ|V i (5)

(antilinear in first vector)

Definitions:

7
 vectors orthogonal if hV |W i = 0
p
Norm of vector: hV |V i = |V |
Normalized vector: |V | = 1
Orthonormal basis: complete set of orthogonal, normalized vectors
Dual spaces
Inner product can be thought of as operation between two vectors from same space
X×X→F
(with asymmetry between first and second argument)
or as operation on vectors from two different spaces
X × X∗ → F
Vector from dual space X is “bra”
E.g., hV |
Related to a corresponding ket by dual correspondence
DC
hV | ←→ |V i
In a Hilbert space, this is invertible – every ket has one corresponding bra and vice versa
Note: name comes from bra + ket = bracket
We can define the operation that takes ket to bra as “Hermitian conjugate”:

(|V i)† = hV | (6)

We can use the same notation for the inverse (bra→ket)

(hV |)† = |V i (7)

=⇒ (|V i)†† = |V i (8)

Antilinearity of first argument of scalar product implies

(a|V i + b|W i)† = a∗ hV | + b∗ hW | (9)

1.1.3 Gram-Schmidt procedure

Not all bases are orthonormal, but it’s always possible to construct one (for finite or countably
infinite dimension)

8
Consider linearly independent basis

|Ii, |IIi, |IIIi, . . . (10)

Construct orthonormal set |1i, |2i, |3i, . . .:

|Ii
|1i = p (11)
hI|Ii
=⇒ h1|1i = 1 (12)

Subtract the part of |IIi parallel to |1i:

|20 i = |IIi − |1ih1|IIi (13)

=⇒ h1|20 i = 0 (14)

Normalize to get |2i:

|20 i
|2i = p (15)
h20 |20 i

Subtract part of |IIIi parallel to |1i and |2i:

|30 i = |IIIi − |1ih1|IIIi − |2ih2|IIIi (16)

Normalize, and repeat for all vectors

Linearly independent =⇒ all basis vectors nonzero
Schwarz Inequality

|hV |W i|2 ≤ hV |V ihW |W i (17)

Proof:

9
 define

hW |V i
|Zi ≡ |V i − |W i (18)
hW |W i
=⇒ hW |Zi = 0 (19)

hZ|Zi ≥ 0 (20)
= hV |Zi (21)
hW |V ihV |W i
= hV |V i − ≥0 (22)
hW |W i
=⇒ hW |W ihV |V i ≥ hW |V ihV |W i (23)

The equality holds only when the vectors are linearly dependent |V i = a|W i for some
complex number a
We will use this to derive Heisenberg uncertainty relations

1.1.4 Operators

An operator is a mapping of a vector to another vector: X→X

Ω|V i = |V 0 i (24)

In QM, operators can be linear L or antilinear A:

L (α|V i + β|W i) = αL|V i + βL|W i (25)

A (α|V i + β|W i) = α∗ A|V i + β ∗ A|W i (26)

This semester we will only consider linear operators (in the second semester we will talk
about the time reversal operator, an antilinear operator)
Product of operators means sequential operation:

ΛΩ|V i = Λ (Ω|V i) = Λ|ΩV i (27)

In general, operators do not commute:

[Λ, Ω] ≡ ΛΩ − ΩΛ 6= 0 (28)

10
For reference the anticommutator is

{Λ, Ω} ≡ ΛΩ + ΩΛ (29)

If inverse exists:

ΩΩ−1 = Ω−1 Ω = 1 (30)

With identity operator leaving any vector unchanged:

1|V i = |V i, ∀V (31)

In infinite dimensional spaces, can have right-inverse but not left, or vice versa
Can define operation to the left on bras:

(hV |Ω)(|W i) = hV |(Ω|W i) = hV |Ω|W i (32)

Outer product
Can define linear operator from two kets |V i and |W i by action on arbitrary ket |ψi:

(|V ihW |)|ψi = |V ihW |ψi, ∀|ψi ∈ X (33)

=⇒ (hψ|V ihW |) = hψ|V ihW | (34)

Inner product hW |V i; outer product |V ihW |

Resolutions of the identity
Any state can be written in terms of a basis
X
|ψi = cn |ni (35)
n

For orthonormal basis hn|mi = δnm

hn|ψi = cn (36)
!
X
=⇒ |ψi = |nihn| |ψi (37)
n

11
Since this is true for an arbitrary state |ψi,
X
1= |nihn| (38)
n

Hermitian Conjugation
We defined Hermitian conjugate † of bras and kets. We can do the same for operators.
If Ω is a linear operator, so is Ω† , defined by

Ω† |ψi = (hψ|Ω)† (39)

Some consequences of these definitions:

hV |Ω† |W i = hW |Ω|V i∗ (40)

(Ω† )† = Ω (41)
(c1 Ω1 + c2 Ω2 )† = c∗1 Ω†1 + c∗2 Ω†2 (42)
(ΩΛ)† = Λ† Ω† (43)
(|V ihW |)† = |W ihV | (44)

Hermitian, Anti-Hermitian and Unitary Operators

Hermitian: Ω† = Ω
Anti-Hermitian: Ω† = −Ω
For Hermitian operator hψ|Ω|ψi is real.
Positive definite Hermitian operator: hψ|Ω|ψi > 0
Nonnegative definite (positive semidefinite): hψ|Ω|ψi ≥ 0
=⇒ eigenvalues are positive (nonnegative)
Unitary: ΩΩ† = Ω† Ω = 1
=⇒ Ω−1 = Ω†
Eigenvalues, Eigenkets, Spectrum of an operator
ket |ui is eigenket of operator Ω with (right) eigenvalue a if

Ω|ui = a|ui (45)

12
similarly for eigenbra hv| and (left) eigenvalue b

hv|Ω = bhv| (46)

In finite dimensions every right eigenvalue is also a left eigenvalue, but not necessarily true
in infinite dimensions Notes 1: Mathematical Formalism 15
Set of eigenvalues = “spectrum” of operator

Im E Im E

(a) (b)

Re E Re E

Im E Im
(c) (d)

Re E Re
Egnd

Fig. 1. Examples of spectra of operators. Part (a), the harmonic oscillator; part (b), the free particle; part (c), the
hydrogen atom; part (d), a finite-dimensional unitary operator.

the free particle Hamiltonian H = p2 /2m. It consists of all real energies E ≥ 0, indicated by the
a) Harmonic
thick line on theoscillator
positive real axis. This is an example of the continuous spectrum. In part (c) we
have the spectrum of the hydrogen atom. It consists of a discrete set of spots at negative energy,
2
p of1which2 accumulate
representing the bound states, an infinite number as we approach E = 0. As
H= + mω x2 (47)
2m 2above
with the free particle, there is a continuous spectrum  E = 0. Altogether, the hydrogen atom
has a mixed spectrum (discrete and continuous). Finally, 1 in part (d) we have the spectrum of a
En = ~ω n + (48)
2
unitary operator on a finite-dimensional space. Its eigenvalues are phase factors that lie on the unit
circle in the complex plane.
b) Free particle
E2 E2
2
p
H= (49)
2m
E1
E≥0 (50)

E1

13

Fig. 2. An operator A acting on ket space E has eigen- Fig. 3. In the case of a Hermitian operator, the eigenspa-
spaces En , associated with its discrete eigenvalues an . ces are orthogonal.
 c) Hydrogen atom
d) unitary operator (in finite dimensional space)
Eigenspace: space of eigenstates with particular eigenvalue
Can be 1-dimensional (nondegenerate) or degenerate.
Dimension of eigenspace: order of degeneracy
Hermitian operators
Eigenvalues real:

hu|Ω|ui − hu|Ω|ui∗ = 0 (51)

= (a − a∗ )hu|ui (52)
=⇒ a = a∗ (53)

Eigenspaces orthogonal:
Consider two eigenkets Ω|ui = a|ui, Ω|u0 i = a0 |u0 i

hu0 |Ω|ui = ahu0 |ui (54)

hu|Ω|u0 i = a0 hu|u0 i = hu0 |Ω|ui∗ (55)
=⇒ (a − a0 )hu0 |ui = 0 (56)

Hermitian operator is complete (in finite dimensions): sum of dimensionalities of eigenspaces

is dimension of the whole space
=⇒ eigenvectors orthogonal basis
Even in infinite dimensions, observables are complete Hermitian operators
Compatible observables
Two (or more) observables share a common eigenbasis iff they commute (“compatible
observables”)
Consider Hermitian operators A, B in finite dimensional Hilbert space, that commute
[A, B] = 0

A|ni = an |ni (57)

=⇒ A(B|ni) = an (B|ni) (58)

If nondegenerate: |ni is simultaneous eigenket of A and B.

If degenerate: B|ni is still in degenerate eigenspace Xn

14
 B Hermitian =⇒ has orthonormal basis in subspace Xn

B|nmi = bnm |nmi (59)

=⇒ states |nmi∀n, m form complete basis

If bnm 6= bnm0 , operator B breaks the degeneracy.
Can add more commuting observables until degeneracy completely broken (each basis ket
has unique set of eigenvalues) — complete set of commuting observables
Converse: assume ∃ a simultaneous eigenbasis

A|si = as |si (60)

B|si = bs |si (61)
=⇒ [A, B]|si = 0, ∀|si (62)
=⇒ [A, B] = 0 (63)

1.1.5 Generalized Uncertainty Principle

Normalization of kets
Each ket in the Hilbert space does not correspond to a unique physical state.
Physical state = “ray” = equivalence class of kets related by
|ψ 0 i = c|ψi =⇒ |ψi, |ψ 0 i represent the same state
Can reduce redundancy by considering normalized states:

|ψ 0 i
|ψi = p (64)
hψ 0 |ψ 0 i
=⇒ hψ|ψi = 1 (65)

Usually I will assume we have already done this

But still ambiguous by a phase: |ψi and eiφ |ψi have same normalization for any real φ,
and represent same state
Overall phase of state is not important, and will often be ignored.
Relative phase important: |ψi = |V i + |W i and |ψ 0 i = |V i + eiφ |W i represent different
states.
Measurement

15
Recall: observables represented by complete Hermitian operator A
Possible result of measurement = eigenvalue of A, an
Probability of result an :

|han |ψi|2
Pn = (66)
hψ|ψi

Total probability (assuming no degeneracy)

X 1 X
Pn = hψ|an ihan |ψi (67)
n
hψ|ψi n
!
1 X
= hψ| |an ihan | |ψi (68)
hψ|ψi n

=1 (69)

=⇒ a physical state must be normalizable: hψ|ψi = 6 ∞

=⇒ Physical Hilbert space is finite or countably infinite
Still useful to work with non-normalizable states
E.g., eigenstates of p or x in infinite space:

Z ∞
hx|xi = hp|pi = dx|e−ipx |2 = ∞ (70)
−∞

=⇒ |pi, |xi not in physical Hilbert space.

Span a larger space, which includes both normalizable and non-normalizable states
=⇒ also span physical Hilbert space
Orthonormality condition for continuous states:

hp0 |pi = δ(p − p0 ) (71)

Z
=⇒ dp|pihp| = 1 (72)

Acts as identity physical Hilbert space as well as larger space

=⇒ still valid bases for normalizable states (square integrable wave functions)

16
 ψ(x) = hx|ψi (73)
ψ(p) = hp|ψi (74)

Expectation value
= mean of many measurements of systems prepared in some state |ψi:

X
hAi ≡ an P n (75)
n
X
= an hψ|an ihan |ψi (76)
n
X
= hψ|A|an ihan |ψi (77)
n

= hψ|A|ψi (78)

Define (Hermian) operator ∆A wrt state |ψi

∆A = A − hAi (79)

dispersion (or variance):

h(∆A)2 i = hA2 i − hAi2 (80)

For eigenstate: h(∆A)2 i = (a2n − a2n ) = 0 — observable has precise value

arbitrary state: observable could take a range of values, quantified by dispersion
Uncertainty principle
For any state |ψi and two observables A and B:

1
h∆A2 ih∆B 2 i ≥ |h[A, B]i|2 (81)
4

Proof:

17
 Define

|αi = ∆A|ψi (82)

|βi = ∆B|ψi (83)

and use Schwartz inequality

hα|αihβ|βi ≥ |hα|βi|2 (84)

Left side:

hα|αi = hψ|∆A† ∆A|ψi = h(∆A)2 i (85)

hβ|βi = h(∆B)2 i (86)

Right side:

|hα|βi|2 = |hψ|∆A∆B|ψi|2 (87)

1 1
∆A∆B = [∆A, ∆B] + {∆A, ∆B} (88)
2 2

= anti-Hermitian part + Hermitian

[∆A, ∆B]† = [A, B]† = (AB − BA)† = BA − AB = −[A, B] (89)

{∆A, ∆B}† = {∆A, ∆B} (90)

=⇒ expectation of first term imaginary, second term real

1 1
|h∆A∆Bi|2 = |h[A, B]i|2 + |h{∆A, ∆B}i|2 (91)
4 4
1
≥ |h[A, B]i|2 (92)
4

Example: A = x, B = p

18
 3. Postulates Applied to the Stern-Gerlach Experiment

To answer such questions, we will analyze the Stern-Gerlach experiment, pretending that we
know nothing except the experimental results and the postulates listed above. In particular, we will
pretend that we know nothing about wave functions, spin, Pauli matrices, the Schrödinger equation,
etc. The following presentation is based closely on that given by J. J. Sakurai, Modern Quantum
Mechanics.
Suppose we are working with a beam of silver atoms, as in the original Stern-Gerlach experiment.
From a modern perspective, we know that silver atoms possess a magnetic moment µ because of their
single unpaired valence electron, so that the measured value of any component of magnetic moment is
[x, p] = i~ (93)
±µ0 , where µ0 = eh̄/2mc is a Bohr magneton (this is in Gaussian units; see Appendix A). The silver
atom has the same electronic spin and magnetic moment ~ as a free electron, but it is electrically
=⇒ ∆x∆p ≥
neutral. Charged particles are not suitable for a Stern-Gerlach experiment, because in practice
(94)
2
the ordinary Lorentz force (q/c)v×B is much larger than the force due to the magnetic moment,
F = ∇(µ · B). The original experiment of Stern and Gerlach in 1921 provided a measurement of
1.2 Application ofgoodpostulates
µ , with a result in
0 agreement with Bohr’s value. We also know that the magnetic moment
operator is proportional to the spin operator, µ = (e/mc)S, where any component of spin takes on
Consider the
theStern-Gerlach
values ±h̄/2. In the experiment (1921)
following discussion, however, we will play dumb and ignore all of this, and
instead we will work solely with the experimental results. For the same reason, we will speak in
A beam of silver atoms are sent through an inhomogeneous magnetic field, which measures
terms of measurements of magnetic moment, not spin.
a component of its magnetic moment

+ |µx +⟩ |µz +⟩
50%
Ag µx − µz |µz −⟩
50%
oven

Fig. 1. A beam of silver atoms is subjected to a measurement of µx , after which the atoms with µx = +µ0 are passed
to a second magnet which measures µz .

It is found that the x component only takes two values µx = ±µ0 .

=⇒ µ̂x has two eigenvalues
=⇒ Hilbert space is (at least) 2 dimensional
For now assume no degeneracy
=⇒ space spanned by 2 eigenkets of µ̂x (or µ̂y , µ̂z , etc.)
|µi ±i = normalized eigenket of µ̂i
(eiα |µi ±i also eigenket – must choose convention for phase of unique ket |µi ±i)
Eigenspaces orthogonal: hµi + |µi −i = 0
After first measurement, system is in (pure) state |µx +i
After second measurement, half in each state |µz ±i ≡ |±i

|µx +i = c+ |+i + c− |−i (95)

c± = h±|µx +i (96)

Probabilities for the second measurement:

Prob(µz = +µ0 ) = hµx + |+ih+|µx +i (97)

1
= |c+ |2 = (98)
2
1 iα1
=⇒ c+ = √ e (99)
2

19
And similarly,

1
c− = √ eiβ1 (100)
2
1  iα1 
=⇒ |µx +i = √ e |+i + eiβ1 |−i (101)
2

Choose overall phase of |µx +i to match that of |+i:

1 h 0
i
|µx +i0 = √ |+i + eiβ1 |−i (102)
2

Then drop primes |µx +i → |µx +i0 , β10 → β1

Doing the second measurement instead on |µx −i ensemble, also gives a 50/50 split:

1  
|µx −i = √ |+i + eiγ1 |−i (103)
2

We have fixed the phases of both |µx ±i (with respect to |+i)

Orthogonality:

1 
hµx + |µx −i = 0 = 1 + ei(−β1 +γ1 ) (104)
2
=⇒ e = −eiβ1
iγ1
(105)
1  
|µx ±i = √ |+i ± eiβ1 |−i (106)
2

Repeat the experiment with µy for the first measurement: same 50/50 split:

1  
|µy ±i = √ |+i ± eiβ2 |−i (107)
2

~ˆ:
Write operator µ

20


µˆz = µ0 |+ih+| − |−ih−| (108)


µˆx = µ0 |µx +ihµx + | − |µx −ihµx − | (109)


= µ0 e−iβ1 |+ih−| + eiβ1 |−ih+| (110)


µˆy = µ0 e−iβ2 |+ih−| + eiβ2 |−ih+| (111)
(112)

Another experiment: measure µx and then µy — still 50/50 split

1 1
=⇒ = |hµy ± |µx +i|2 = [1 ± cos(β2 − β1 )] (113)
2 2
π
=⇒ β2 = β1 ± (114)
2
iβ2 iβ1
=⇒ e = ±ie (115)

Still have one unknown phase β1 , and an unknown sign

β1 can be fixed by redefining |−i
=⇒ can make either µˆx or µˆy real in z basis (but not both!)
Choose |−i → e−iβ1 |−i:
 
µˆx = µ0 |+ih−| + |−ih+| (116)
 
µˆy = ±µ0 −i|+ih−| + i|−ih+| (117)
 
µˆz = µ0 |+ih+| + |−ih−| (118)

All phases fixed (but still have unknown sign: we didn’t specify right/left coordinate
system)
Another way of writing the operators:
! ! !
0 1 0 −i 1 0
µx = µ0 , µy = ±µ0 , µz = µ0 (119)
1 0 i 0 0 −1

1.3 Mixed states and the density operator

The quantum state coming from the oven is a thermal state

21
 =⇒ no preferred spin direction

hSx i = hSy i = hSz i = 0 (120)

=⇒ cannot be described by pure state:

In some basis (e.g., eigenkets of Sz ) an arbitrary pure state can be written as

|χi = α|+i + β|−i (121)

Properly normalized:

|α|2 + |β|2 = 1 (122)

~ is a matrix
Written in this basis, then each component of S

~ = ~ ~σ
S (123)
2

*Sx +
D E
~ =
S Sy (124)
Sz
 ∗ 
α β + β ∗α
~ 
= −iα∗ β + iβ ∗ α (125)
2
|α|2 − |β|2
~
= n̂ (126)
2

n̂ = unit vector: n̂ · n̂ = 1
=⇒ hŜi =6 0 for any pure state
Must be a mixed state.
Consider a mixture of atoms, each in a pure state defined by some n̂ above, with an equal
share of the beam at each n̂.
For each pure state, the expectation of an operator A is

hAin̂ = hSn̂ + |A|Sn̂ +i (127)

22
 A measurement on the full (mixed) ensemble is an average over angle:
Z
1
hAi = dΩhSn̂ + |A|Sn̂ +i (128)
4π

=⇒ isotropic
~ =0
e.g., hSi
In general, can have arbitrary set of pure states, with arbitrary probability
R
E.g., continuous |ψ(λ)i, with probability f (λ) with dλf (λ) = 1
Z
hAi = dλf (λ)hψ(λ)|A|ψ(λ)i (129)

P
or discrete: |ψi i, i fi = 1
X
hAi = fi hψi |A|ψi i (130)
i

Or a combination of both
Z X
hAi = dλf (λ)hψ(λ)|A|ψ(λ)i + fi hψi |A|ψi i (131)
i

with
Z X
dλf (λ) + fi = 1 (132)
i

(states arbitrary, not necessarily linearly independent or orthogonal)

1.3.1 Density Operator

The information in a mixed state be expressed in terms of a density operator ρ

23
 X
ρ= fi |ψi ihψi | (133)
i

=⇒ hAi = Tr(ρA) (134)

X
N X
= fi hj|ψi ihψi |A|ji (135)
j i
X
= fi hψi |A|ψi i (136)
i

With |ji some orthonormal basis

General: Expectations of this form can represent any possible measurement
E.g., A = |an ihan |, for any eigenstate of any operator
Also describes pure states:

ρ = |ψihψ| (137)

2 Spatial Degrees of Freedom

2.1 Projection Operators
A projection operator or projector P is an observable that satisfies

P2 = P (138)

=⇒ eigenvalues 0 or 1:

P |pi = p|pi (139)

P 2 |pi = p2 |pi = p|pi (140)
=⇒ p2 = p (141)
=⇒ p = 0, 1 (142)

P divides the Hilbert space into two orthogonal subspaces X = X0 ⊕ X1 ,

where P annihilates any vector in X0 and leaves any vector in X1 invariant
=⇒ P projects onto the space X1

24
 Consider degenerate spectrum of Hermitian operator A

A|nri = an |nri, (143)

with |nri an orthonormal basis labeled by the eigenvalue n, and some orthonormal states r
within the degenerate subspace
Can define projector onto the eigenspace Xn
X
Pn = |nrihnr| (144)
r

=⇒ Pn Pm = Pm Pn = δnm Pn (145)

Also:
X
Pn = 1 (146)
n

By considering the operation of A on each state |nri, you can see

X
A= an P n (147)
n

The measurement postulates then takes this form: the probability of measuring normal-
ized state |ψi to have value an is

Prob(A = an ) = hψ|Pn |ψi (148)

and after the measurement, the state is represented by the ket Pn |ψi
This is more general than the expressions we saw earlier assuming no degeneracy.

2.2 Configuration Space

Consider single spinless particle moving in one dimension
(or, a particle whose spin and motion in the other two dimensions is not relevant)
Assume that the position x can be measured as a continuous variable

x̂|xi = x|xi (149)

25
 Assume x̂ = complete set of commuting observables
|xi nondegenerate =⇒ |xi form orthogonal basis
Choose normalization:

hx1 |x2 i = δ(x1 − x2 ) (150)

Z ∞
=⇒ 1 = dx|xihx| (151)
−∞

Projection operator onto interval I with x0 ≤ x ≤ x1 :

Z x1
PI = dx|xihx| (152)
x0

Probability that particle in state |ψi is measured in some region x0 ≤ x ≤ x1 :

Z x1
Prob(x0 ≤ x ≤ x1 ) = hψ|PI |ψi = dxhψ|xihx|ψi (153)
x0

Define the wavefuncion

ψ(x) = hx|ψi (154)

So
Z x1
Prob(x0 ≤ x ≤ x1 ) = dx|ψ(x)|2 (155)
x0

=⇒ |ψ(x)|2 is probability density. ψ(x) an amplitude

Can invert by using resolution of the identity:
Z
|ψi = dx|xihx|ψi (156)
Z
= dx|xiψ(x) (157)

ψ(x) are set of expansion coefficients of |ψi in basis |xi.

Any other basis could be used

26
 If we know the wavefunctions, can compute inner products:
Z Z
hψ|φi = dxhψ|xihx|φi = dxψ ∗ (x)φ(x) (158)

More dimensions: Let x = 3D position vector (x, y, z), or (x1 , x2 , x3 )

Correspond to eigenvalues of vector operator x̂ = (x̂, ŷ, ẑ) = (x̂1 , x̂2 , x̂3 )
Assume experimental result: [x̂i , x̂j ] = 0
For state |xi,

x̂i |xi = xi |xi (159)

with i ∈ {1, 2, 3}
Choose normalization

hx1 |x2 i = δ 3 (x1 − x2 ) = δ(x1 − x2 )δ(y1 − y2 )δ(z1 − z2 ) (160)

A resolution of the identity is then

Z
1= d3 x|xihx| (161)

and the wavefunction is

ψ(x) = hx|ψi (162)

Z
|ψi = d3 x|xiψ(x) (163)

Probability of finding particle in volume V :

Z
d3 x|ψ(x)|2 (164)
V

Action on wavefunction:

hx|x̂|ψi = xhx|ψi (165)

=⇒ ψ(x) → xψ(x) (166)

27
2.3 Translation Operators
(back to one dimension, for simplicity)
We want to find an operator that takes all particles and moves their position by some
constant a.

T (a)|xi = |x + ai (167)

This completely defines T (a), since |xi form a complete basis

This operator changes configuration space wavefunction ψ(x) in the expected way:

6 |φi ≡ TDegrees
Notes 4: Spatial (a)|ψi of Freedom (168)
=⇒ φ(x) ≡ hx|φi = hx|T (a)|ψi (169)
Z
0 (21)0 ihx
and0 |ψi
where we have inserted a resolution of the identity,
= dxused hx|T
Eqs.
(a)|x (2), and carried out the (170)
integral. We write the result as
! " Z
T (a)ψ (x) = ψ(x0 − a),0 (23)
= dx hx|x + aihx0 |ψi (171)
where we have replaced φ by T (a)ψ. This is aZcompanion to Eq. (21), which gives the action of
translation operators on the basis kets; this gives their 0action on 0wave functions.
= dx δ(x − x − a)ψ(x0 ) (172)
There are several remarks concerning Eq. (23). First, notice the minus sign in this equation,
compared to the plus sign in Eq. (21). The minus sign is necessary to get a wave function that has
= ψ(x − a) (173)
been moved forward under the displacement operation, as illustrated in Fig. 1. To remember the
! "
signs it helps to write Eq. (22) in the form, T (a)ψ (x + a) = ψ(x) and to say, “the value of the new
Notice the
wave minus
function sign:
at the new point equals the value of the old wave function at the old point.”

T (a)

ψ(x) ψ(x − a)

x
x0 x0 + a

Fig. 1. The action of the translation operator T (a) on a wave function ψ(x).

Easily generalized to three dimensions

Another remark is that that Eq. (23) uses the translation operator T (a) in a different sense
from its original definition, because it is acting on a configuration space wave function ψ instead of
a ket |ψ⟩. As we say, Eq. (23) gives the representation of the operator T (a) on configuration space
wave functions.
Finally, we remark that many books would write Eq. (23) without the parentheses, that is, as
T (a)ψ(x) = ψ(x − a). (24)
The problem with this notation is that it is not clear what T (a) acts on. ψ is a function, and ψ(x)
is the value of that function at a point x, that is, it is a number. Does T (a) act on the function or
28
the value of the function? Obviously, it acts on the function, which is what the extra parentheses
in Eq. (23) make explicit.
The translation operator is easily generalized to three dimensions, where the displacement a is
a vector, and the translation operator is defined by
 Properties of T as defined:

T (0) = 1 (174)
T (a)T (b) = T (a + b) = T (b)T (a) (175)
T (a)−1 = T (−a) (176)
T (a)−1 = T (a)† (177)

Last property: Unitary (preserves norm)

Generators
Unitary operators can be expressed in terms of Hermitian operators called Generators,
which represent infinitessimal transformations
Consider a small translation a, and expand T in a Taylor series

dT
T (a) = T (0) + a (0) + . . . (178)
da

Define the operator

dT
k̂ = i (0) (179)
da

which is Hermitian

T (a)† = 1 + iak̂ † (180)

= T (−a) = 1 + iak̂ (181)
=⇒ k̂ = k̂ † (182)

T is exponential of k:

dT (a) T (a + ) − T (a)
= lim (183)
da →0
  
T () − 1
= lim T (a) (184)
→0 
= −ik̂T (a) (185)

With boundary condition T (0) = 1, the solution is

T (a) = e−iak̂ (186)

29
 Action of k̂

dT (a) d d
k̂|xi = i |xi =i |x + ai =i |xi (187)
da a=0 da a=0 dx

or, in configuration space, with |φi = k̂|ψi:

dT (a)
φ(x) = ihx| |φi (188)
da a=0
Z
dT (a)
= dx0 ihx| |x0 ihx0 |ψi (189)
da a=0
Z  
d
= dx0 i hx|x0 i hx0 |ψi (190)
dx0 a=0
dψ(x)
= −i (191)
dx

Translation in 3D
It’s straightforward to generalize to 3 dimensions:

∂T (a)
k̂i = i (192)
∂ai a=0
1
T (a) = e−ia·k̂ = 1 − ia · k̂ − (a · k̂)2 + . . . (193)
2
∂
k̂i |xi = i |xi (194)
∂xi
=⇒ [ki , kj ] = 0 (195)

Commutation between x and k:

∂
x̂i k̂j ψ ≡ hx|x̂i k̂j |ψi = −ixi hx|ψi (196)
∂xj
∂ ∂
k̂j x̂i ψ = −i xi hx|ψi = −iδij hx|ψi − ixi hx|ψi (197)
∂xj ∂xj
=⇒ [x̂i , k̂j ] = iδij (198)

Momentum
In classical mechanics, generator of translations is (canonical) momentum

30
 ∂L
p= (199)
∂ ẋ

=⇒ identify p̂ ∝ k̂
Need coefficient to get right dimensions. Define it as ~

p̂ = ~k̂ (200)

With this definition of p and ~, we finally have:

i
T (a) = e− ~ (a·p̂) (201)
[x̂i , x̂j ] = [p̂i , p̂j ] = 0 (202)
[x̂i , p̂j ] = i~δij (203)
p̂|xi = i~∇||xi (204)

Momentum representation
p̂ is vector of commuting Hermitian operators
=⇒ simultaneous eigenstates form basis

p̂|pi = p|pi (205)

Relate to configuration basis:

hx|p̂|pi = −i~∇hx|pi (206)

= phx|pi (207)
=⇒ ψp (x) ≡ hx|pi = Aeip·x/~ (208)

31
 Choosing the usual normalization for continuous states gives:

hp1 |p2 i = δ 3 (p1 − p2 ) (209)

Z
1
= d3 xei(p1 −p2 )·x/~ (210)
(2π~)3
Z
= d3 xhp1 |xihx|p2 i (211)
1
=⇒ hx|pi = eip·x/~ (212)
(2π~)3/2

Configuration space wavefunction and momentum space wavefunction related by Fourier

transforms

Z Z
3 d3 p
ψ(x) ≡ hx|ψi = d phx|pihp|ψi = eip·x/~ ψ(p) (213)
(2π~)3/2
Z
d3 x
ψ(p) = e−ip·x/~ ψ(x) (214)
(2π~)3/2

Action of x̂ on ψ(p)

Z
hp|x̂|ψi = d3 xhp|xihx|x̂|ψi (215)
Z
d3 x
= e−ip·x/~ xψ(x) (216)
(2π~)3/2
Z
∂ d3 x
= i~ e−ip·x/~ ψ(x) (217)
∂p (2π~)3/2
∂
18 = Spatial
Notes 4: i~ ψ(p) (218)
∂p Degrees of Freedom

Summary of x̂ and p̂ in each respective basis:

x̂i p̂i
∂
Configuration mult by xi −ih̄
∂xi
∂
Momentum ih̄ mult by pi
∂pi

Table 1. Representations of the operators x̂i and p̂i when acting on configuration and momentum space wave functions.

Finally, we summarize the representations of the operators x̂i and p̂i in the configuration and
momentum representations in Table 1.
32
15. Multiparticle Wave Functions

In a system of N particles, the positions of the individual particles are independent observables
3 Time evolution
Want to find an operator U that describes how a state changes with time:

|ψ(t)i = U (t, t0 )|ψ(t0 )i (219)

We demand that it has the properties:

U (t0 , t0 ) = 1 (220)
U (t, t0 )−1 = U (t, t0 )† (221)

(total probability must always add to 1)

Also need composition property:

U (t2 , t1 )U (t1 , t0 ) = U (t2 , t0 ) (222)

3.0.1 Hamiltonian as generator of time evolution

Consider infinitesimal time evolution from t to t + , expand in Taylor series:

U (t + , t) = 1 − iΩ(t) + . . . (223)
∂
=⇒ Ω(t) ≡ i U (t0 , t) (224)
∂t0 t0 =t

In Classical mechanics, generator of time translations is Hamiltonian, =⇒ Ĥ ∝ Ω

again we need a coefficient to give H units of energy. Choose ~:

Ĥ(t) = ~Ω(t) (225)

i
=⇒ U (t + , t) = 1 − Ĥ(t) + O(2 ) (226)
~

We observe that the coefficient ~ is the same for Ĥ as for every component of p̂. No
surprise since energy and momentum are part of a single 4-vector.
Differential equations for time evolution

33
 differentiate U :

∂U (t, t0 ) U (t + , t0 ) − U (t, t0 )
= lim (227)
∂t →0 
U (t + , t0 ) = U (t + , t)U (t, t0 ) (228)
 
∂U (t, t0 ) U (t + , t) − 1
=⇒ = lim U (t, t0 ) (229)
∂t →0 
∂U (t, t0 )
i~ = Ĥ(t)U (t, t0 ) (230)
∂t

differentiating Eq. (219) gives time evolution of a state

∂
i~ |ψ(t)i = Ĥ(t)|ψ(t)i (231)
∂t

“Schrödinger equation”
If Ĥ(t) = H (time independent), the solution is

U (t, t0 ) = e−iĤ(t−t0 )/~ (232)

Only a function of t − t0 → t:

U (t) = e−iĤt/~ (233)

[Ĥ, U (t)] = 0 (234)

In general [H(t), H(t0 )] 6= 0, and [H, U ] 6= 0.

3.0.2 Heisenberg vs Schrödinger picture

So far, we have been using the Schrödinger picture.

Equivalent is the Heisenberg picture, where time evolution is not in kets, but instead in
operators.
Heisenberg picture state vectors:

|ψH i = U (t, t0 )† |ψS (t)i (235)

=⇒ |ψH i = |ψS (t0 )i (236)

(No time dependence)

34
 Heisenberg picture operators:

AH (t) = U (t, t0 )† AS (t)U (t, t0 ) (237)

dAH ∂AS
=⇒ i~ = −U † HAS U + U † AS HU + i~U † U (238)
dt ∂t
† ∂AS
= −U| {zHU} U † AS U +U † AS U U † HU + i~U † U (239)
| {z } ∂t
HH AH l
 
∂A
= [AH , HH ] + i~ (240)
∂t H

The parentheses mean to differentiate A with respect to explicit time dependence in S

picture, then convert to H picture.
All measurable quantities can be written in terms of matrix elements hφ|A|ψi, which are
equivalent in both pictures:

hφS (t)|AS |ψS (t)i = hφH |AH (t)|ψH i (241)

When S Hamiltonian is time independent [HS , U ] = 0, and

HH = U (t)† HS U (t) = HS (242)

Mostly we will use S picture, but sometimes H picture will be useful. Other pictures can
also be useful (e.g., interaction picture).
Time evolution of Density Operator
At some time t0 we have a state described by the density operator
X
ρ̂(t0 ) = fi |ψi (t0 )ihψi (t0 )| (243)
i
X
=⇒ ρ̂(t) = fi |ψi (t)ihψi (t)| (244)
i
!
X
= U (t, t0 ) fi |ψi (t0 )ihψi (t0 )| U (t, t0 )† (245)
i

= U (t, t0 )ρ̂(t0 )U (t, t0 )† (246)

Or, as a differential equation

∂ ρ̂
i~ = [Ĥ, ρ̂] (247)
∂t

35
 Constants of Motion
If an observable A commutes with H

[A, H] = 0 (248)
dAH
=⇒ =0 (249)
dt
d
=⇒ hAi = 0 (250)
dt

where the last expression is true in any picture (e.g., Schrödinger)

E.g., energy is conserved if [H(t), H(t0 )] = 0, which is always true if the Hamiltonian is
time-independent

3.0.3 Initial value problem for time-independent Ĥ

Consider a time-independent Hamiltonian.

If energy spectrum and eigenstates are known, time dependence can be written simply:

H|ni = En |ni (251)

Expand a time dependent state in terms of energy eigenstates:

X
|ψ(t)i = cn (t)|ni (252)
n

Recall

∂
i~ |ψ(t)i = Ĥ(t)|ψ(t)i (253)
∂t
=⇒ i~ċn = En cn (254)
=⇒ cn (t) = e−iEn t/~ cn (0) (255)

where

cn (0) = hn|ψ(0)i (256)

This is equivalent to expanding time-evolution operator in energy eigenstates

36
 X
U (t) = e−iHt/~ = e−iEn t/~ |nihn| (257)
n

If state is energy eigenstate, it stays in that state. The only change is its overall phase
e−iEn t/~ — “stationary state”

37
4 Harmonic Oscilator
4.1 HO as an approximation to a general potential problem
Consider the Hamiltonian of a particle of mass m, moving in a potential V (x, t).
The quantum Hamiltonian is given by

p̂2
Ĥ = + V (x̂, t) (258)
2m

This can be generalized to multiple particles:

XN
p̂2i
Ĥ = + V (x̂1 , x̂2 , . . . , x̂N , t) (259)
i=1
2mi

States can be built out of simultaneous eigenstates of the N position operators {x̂i }

x̂i |x1 x2 . . . xN i = xi |x1 x2 . . . xN i (260)

With some approximations, this can be written as a sum of simple harmonic oscillators.
Let i label the vector component and the particle.
E.g., {x1 , x2 , x3 } = {x, y, z} of the first particle, {x4 , x5 , x6 } = {x, y, z}, etc.
The total degrees of freedom are then n = 3N (for 3 dimensions).
Change notation

mi
µi = , (261)
M

where M is an arbitrary reference mass.

Then transform the operators (from now on written with out the hat x = x̂, etc.):

1 √
xi → x0i = √ xi , pi → p0i = µi pi (262)
µi

The new operators have the same commutation relation:

[xi , pj ] = i~δij , (263)

so the physics doesn’t change. It’s just a change of units. The original units can be obtained
by rescaling after the problem has been solved.

38
 We define a transformed potential by

W (x01 , . . . , x0n ) = V (x1 , . . . , xn ) (264)

So the Hamiltonian can now be written (dropping the primes)

1 X 2
n
H= p + W (x1 , . . . , xn ) (265)
2M i=1 i

The force on the particle vanishes at points in configuration space (x10 , . . . , xn0 ) where
the first derivatives of W vanish:

∂W
(x10 , . . . , xn0 ) = 0. (266)
∂xi

This is a “critical point” of the potential.

Close to the critical point, we can approximate the potential by its Taylor series:

1 X 00
W (x1 , . . . , xn ) ' W0 + Wij (xi − xi0 )(xj − xj0 ), (267)
2 ij

where W0 is the value of W at the critical point and

00 ∂ 2W
Wij = (xi − xi0 )(xj − xj0 ) (268)
∂xi ∂xj
00
Assume a positive definite matrix Wij , =⇒ the potential energy increases as we move
away from the critical point, =⇒ the critical point is a local minimum
We plug this expression into the Hamiltonian and perform the following transformations:
Subtract a constant W0 /2 from H. Only energy differences are physically meaningful, so
adding a constant doesn’t change the physics.
Shift coordinates

xi → x0i = xi − x0i , (269)

So the minimum of W is at the origin.

39
 Transform the coordinates
X X
x0i = Rij xj , p0i = Rij pi (270)
j j

where R is the ortogonal matrix that diagonalizes W 00 . Again, this preserves the commutation
relation.
Finally, we have the approximate Hamiltonian near the critical point

Xn  2 
pi M ωi2 x2i
H= + , (271)
i=1
2M 2

where the (positive) eigenvalues of W 00 are written as M ωi2

=⇒ close to the critical point, H is a sum of n one-dimensional harmonic oscillators.
The Schrödinger equation separates, and each can be solved independently. The energy
eigenfunctions of the full system are products of energy eigenfunctions of the one-dimensional
harmonic oscillators.

4.2 1D HO solution
Let’s write one of the terms in the form

p2 mω 2 x2
H= + . (272)
2m 2

Choose units such that m = ω = ~ = 1.

I.e., define rescaled variables:

x v p E H
x̄ = q , t̄ = ωt, v̄ = q , p̄ = √ , Ē = , H̄ = , (273)
~ ~ω m~ω ~ω ~ω
mω m

etc.
Can always multiply by correct factors at the end to obtain usual units.
E.g., ψ(x) must have dimension length−1/2 , so

 mω 1/4
ψ(x) = ψ̄(x), (274)
~

etc.

40
 After transforming to the dimensionless variables and dropping the overbars, we have:

1 2

H= p̂ + x̂2 , (275)
2

and the commutation relations are

[x̂, p̂] = i (276)

Solve using Dirac’s algabraic method.

Define (non-Hermitian) operators

1 1
a = √ (x̂ + ip̂), a† = √ (x̂ − ip̂) (277)
2 2

or, inverted

1 −i
x̂ = √ (a + a† ), p̂ = √ (a − a† ) (278)
2 2

=⇒ [a, a† ] = 1 (279)

a = “lowering” or “annihilation” operator

a† = “creation” or “raising” operator.
in terms of these operators we have

1 1
H = a† a + =N+ (280)
2 2

N = “number operator”. Nonnegative definite since any expectation

hφ|N |φi ≥ 0 (281)

is the norm of a state a|φi

Find the spectrum of N (and therefore also H):

N |νi = ν|νi (282)

41
with ν ≥ 0:

hν|N |νi = ν ≥ 0 (283)

Commutation between N and a or a† :

N a = a† aa = aa† a − a = a(N − 1), (284)

N a† = a† aa† = a† a† a + a† = a† (N + 1) (285)
=⇒ N a|νi = (ν − 1)a|νi, (286)
N a† |νi = (ν + 1)a† |νi (287)

6 0, it is an eigenket with eigenvalue ν − 1

If a|νi =
†
If a |νi =
6 0, it is an eigenket with eigenvalue ν + 1
When do they vanish:

hν|a† a|νi = hν|N |νi = ν (288)

=⇒ a|νi = 0 iff ν = 0

hν|aa† |νi = hν|(N + 1)|νi = ν + 1 (289)

=⇒ a† |νi =
6 0
=⇒ ν is an integer. Otherwise, acting on an eigenstate with a would eventually give a
negative ν state, which is not possible.
=⇒ Spectrum consists of all nonnegative integers n

=⇒ H|ni = En |ni, (290)

1
En = n + (291)
2

Phase Conventions
a lowers the energy by 1 unit (~ω):

a|ni = cn |n − 1i (292)

42
The magnitude of cn is fixed by the normalization condition

hn|a† a|ni = n = hn − 1||cn |2 |n − 1i = |cn |2 (293)

It could still have an arbitrary phase

√
a|ni = eiαn n|n − 1i (294)
√
a|1i = eiα1 1|0i (295)
√
a|2i = eiα2 2|1i (296)
..
. (297)

Starting with an arbitrary |0i, we can define the other |ni so that
√
a|ni = n|n − 1i (298)

With this convention, the action of a† is

a† |ni = dn |n + 1i (299)
√
aa† |ni = (a† a + 1)|ni = (n + 1)|ni = dn a|n + 1i = dn n + 1|ni (300)
√
=⇒ dn = n + 1 (301)

Summarizing:
√
a|ni = n|n − 1i (302)
√
a† |ni = n + 1|n + 1i (303)

Eigenstates
The above implies that we can write an arbitrary state in terms of the ground state as

(a† )n
|ni = √ |0i (304)
n!

43
 First find the ground state in configuration space:

a|0i = 0 (305)
 
1 d
√ x+ ψ0 (x) = 0 (306)
2 dx
1 2
=⇒ ψ0 (x) = 1/4 e−x /2 (307)
π

Then an arbitrary energy eigenstate is obtained by raising operations

 n
1 1 d 2
ψn (x) = 1/4 √ x− e−x /2 (308)
π n!2n dx

Use the identity:

 
d 2 2 df (x)
x− ex /2 f (x) = −ex /2 (309)
dx dx
 n
1 (−1)n x2 /2 d 2
=⇒ ψn (x) = 1/4 √ e e−x (310)
π n!2n dx
1 1 2
= 1/4 √ Hn (x)e−x /2 , (311)
π n!2 n

with Hn the Hermite polynomials, satisfying the Rodriguez formula

 n
n x2 d 2
Hn (x) = (−1) e e−x (312)
dx

Restoring the original units, we have:

 mω 1/4 r
1 mω 2
ψn (x) = √ Hn ( x)e−mωx /2~ (313)
π~ n!2n ~

44
5 Path Integrals
Path Integral: alternative (but equivalent) way of formulating QM
Provides different way to think about QM, and can be useful for certain types of problems
(especially in quantum field theory, where a Lagrangian formalism is more convenient)

5.1 Propagator
Propagator K = matrix element of time evolution operator

K(x, t; x0 , t0 ) = hx|U (t, t0 )|x0 i (314)

= amplitude to find particle at position x at time t, if it was at x0 at time t0

(We work in 1D with no spin for simplicity. The generalization is straightforward).

U (t0 , t0 ) = 1 (315)
=⇒ K(x, t0 ; x0 , t0 ) = hx|x0 i = δ(x − x0 ) (316)

∂U (t, t0 )
i~ = Ĥ(t)U (t, t0 ) (317)
∂t
∂K(x, t; x0 , t0 ) ∂U (t, t0 )
=⇒ i~ = hx|i~ |x0 i (318)
∂t ∂t
= hx|Ĥ(t)U (t, t0 )|x0 i (319)
 
~2 ∂ 2
= − + V (x, t) hx|U (t, t0 )|x0 i (320)
2m ∂x2
= H(t)K(x, t; x0 , t0 ) (321)

K satisfies Schrodinger eq. for x and t, with δ-function initial condition.

I.e., if ψ(x, t0 ) = δ(x − x0 ), then ψ(x, t) = K(x, t; x0 , t0 )
For some arbitrary initial condition ψ(x, t0 ), the propagator can generate a general solu-
tion:

45
 Z
ψ(x, t) = hx|ψ(t)i = hx|U (t, t0 )|ψ(t0 )i = dx0 hx|U (t, t0 )|x0 ihx0 |ψ(t0 ) (322)
Z
= dx0 K(x, t; x0 , t0 )ψ(x0 , t0 ) (323)

K = kernel of integral transform ψ(x0 , t0 ) → ψ(x, t)

Superposition (Huygen’s principle of wave propagation): each point x0 at time t0 acts as
independent wave source, weighted by the strength of the field at point (x0 , y0 )

5.1.1 Propagator for free particle

p̂2
Ĥ = (324)
2m

Time independent =⇒ set t0 = 0 for simplicity and define U (t) = U (t, 0).

U (t) = e−itĤ/~ (325)

−itp̂2 /2m~
=⇒ K(x, x0 , t) = hx|e |x0 i (326)
Z
2
= dphx|e−itp /2m~ |pihp|x0 i (327)
Z
2
= dpe−itp /2m~ hx|pihp|x0 i (328)
Z
2 1 ipx/~ −ipx0 /~
= dpe−itp /2m~ e e (329)
2π~
Z   
dp i p2 t
= exp p(x − x0 ) − (330)
2π~ ~ 2m
r  2

m i m(x − x0 )
= exp (331)
2πi~t ~ 2t

5.2 Path integral in 1D

p2
H =T +V = + V (x) (332)
2m

46
Time independent =⇒ set t0 = 0 for simplicity and define U (t) = U (t, 0).

U (t) = e−iHt/~ (333)

=⇒ K(x, x0 , t) = hx|U (t)|x0 i (334)

Break time t into N equal intervals  = t/N

=⇒ U (t) = [U ()]N (335)

At the end, we will take N → ∞ ( → 0), holding t fixed.

U () = e−i(T +V )/~ = e−iT /~ e−iV /~ + O(2 ) (336)

 N
=⇒ U (t) = e−iT /~ e−iV /~ + O(1/N ) (337)
 N
=⇒ K(x, x0 , t) = lim hx| e−iT /~ e−iV /~ |x0 i (338)
N →∞

Insert a resolution of the identity between each factor and define xN ≡ x:

Z
K(x, x0 , t) = lim dx1 . . . dxN −1 hxN |e−iT /~ e−iV /~ |xN −1 ihxN −1 | . . . |x1 ihx1 |e−iT /~ e−iV /~ |x0 i
N →∞

(339)

Each of the N matrix elements can be evaluated by inserting a (momentum) resolution

of the identity

Z
−iT /~ −iV /~ 2
hxj+1 |e e |xj i = dphxj+1 |e−ip̂ /2m~ |pihp|e−iV (x̂)/~ |xj i (340)
Z   
dp i p2
= exp − + p(xj+1 − xj ) − V (xj ) (341)
2π~ ~ 2m
r   
m i (xj+1 − xj )2
= exp m − V (xj ) (342)
2πi~ ~ 2

The final result is a discretized version of the path integral in configuration space

47
 To visualize the integrations being performed in this integral, we observe that x0 and xN = x
are fixed parameters of the integral, being the x-values upon which K depends, whereas all the
other x’s, x1 , . . ., xN −1 , are variables of integration. Therefore we identify the sequence of numbers,
(x0 , x1 , . . . , xN ) with a discretized version of a path x(t) in configuration space with fixed endpoints
(x0 , xN = x), but with all intermediate points being variables. We think of the path x(t) as passing
through the point xj at time tj = jϵ, so that t0 = 0 and tN = t. See Fig. 3, in which the heavy line is
( )
the discretized path,with fixed Z (x,t) = (xN , tN ). The 
m N/2 endpoints (x0 , t0 ) = (x0 , 0) and
m(xj+1 − xj )2 intermediate
N −1
i X
K(x
xi , i0 ,=x,1,t). .=. , N
Nlim
− 1 are variables ofdxintegration
1 . . . dxN −1that
exptake on all values from 2 −∞−toV (x
+∞,j ) each
→∞ 2πi~ ~ 2
effectively running up and down one of the dotted lines in the j=0figure. Then as N → ∞, we obtain a
(343)
representation of the path at all values of t, and the integral turns into an integral over an infinite
space of paths x(t) in configuration space, which are constrained to satisfy given endpoints at given
5.3endtimes.
Visualization, and Integration in Path Space
x

xN = x

x0
t
t1 t2 t3 tN −3 tN −2 tN −1

t0 = 0 tN = t
Fig. 3. A space-time diagram to visualize the integrations in the discretized version of the path integral in configuration
space, Eq. (25).
xN = x and x0 are fixed. All others are integration variables, and can take on any value.
Each set {x
Notice i } if
that is we
a discretized path
set ∆t = ϵ and ∆xjx(t), starting
= xj+1 at xthe
− xj , then 0 , and passing
exponent through
in Eq. eachlike
(25) looks point
i/h̄
xj at time tj = j.
As N → ∞, we have an integral over all paths x(t).
Let ∆t =  and ∆xj = xj+1 − xj . Then the exponent looks like i/~ times a Riemann sum,

−1
"  2 #
X
N
m ∆xj
∆t − V (xj ) (344)
j=0
2 ∆t

which is an approximation to an integral:

Z "  2 # Z
t
m dx t

S[x(τ )] = dτ − V (x) = dτ L x(τ ), ẋ(τ ) (345)
0 2 dτ 0

where L is the classical Lagrangian

mẋ2
L(x, ẋ) = − V (x) (346)
2

48
 S[x(τ )] = action of path x(τ )
This inspires the compact notation for the path integral
Z  Z t 
i
K(x, x0 , t) = C d [x(τ )] exp Ldτ (347)
~ 0

5.4 Stationary Action and Hamilton’s Principle

In classical mechanics, a particle takes a path according to Hamilton’s principle

δS
=0 (348)
δx(t)

Or,
Z
δ Ldt = 0 (349)

This is exactly the condition of stationary phase of the path integral:

=⇒ in the limit ~ → 0, the integral is dominated by the classical path(s), and one
recovers the classical limit.
(Like the method of steepest descent/saddle point approximation of an integral)

5.4.1 Stationary Phase approximation

Consider 1D integral:
Z
dxeiφ(x)/κ (350)

where κ is a small parameter, and φ is real

=⇒ only a small change ∆x (O(κ)) changes phase by 2π.
=⇒ integration over rapid oscillations =⇒ small contribution to integral
However, at stationary phase point = roots of

dφ
(x̄) = 0, (351)
dx

The integrand has the same phase for a longer x-interval (∆x ∼ κ1/2 )

49
 Expand around stationary phase point x̄

y 2 00
yφ0 (x̄)
φ(x) = φ(x̄) +  + φ (x̄) + . . . (352)
2
y ≡ x − x̄ (353)

and plug into the integral:

Z Z
iφ(x)/κ iφ(x̄)/κ 2 φ00 (x̄)/2κ
dxe 'e dyeiy (354)
s
2πiκ iφ(x̄)/κ
= e (355)
φ00 (x̄)
s
2πκ iφ(x̄)/κ
= eiνπ/4 e (356)
|φ00 (x̄)|

with ν = sgn φ00 (x̄)

If there is more than one critical point xb , there is a contribution from each of them
Z s
X 2πκ iφ(x̄b )/κ
dxeiφ(x)/κ = eiνb π/4 e (357)
b
|φ00 (x̄b )|

If the integral is over more than one dimension x ≡ (x1 , . . . , xn ):

Z
dn x eiφ(x)/κ (358)

The stationary phase points are roots of

∂φ
(x̄) = 0, i = 1, . . . , n (359)
∂xi

The expansion around the critical points then takes the form

1X ∂ 2 φ(x̄)
φ(x) = φ(x̄) + yk yl (360)
2 kl ∂xk ∂xl

where again ~y = ~x − ~x̄

50
 The integral becomes
Z Z !
i X ∂ 2 φ(x̄)
dn xeiφ(x)/κ = eiφ(x̄)/κ dn y exp yk yl (361)
2κ kl ∂xk ∂xl

We can diagonalize the matrix ∂ 2 φ/∂xk ∂xl with an orthogonal transformation z = Ry.
Since det R = 1, dn y = dn z, and the integral becomes

Z Z !
i X
dn xeiφ(x)/κ = eiφ(x̄)/κ dn z exp λk zk2 , (362)
2κ k

where λk are the eigenvalues of ∂ 2 φ/∂xk ∂xl .

This is a product of 1-dimensional Gaussian integrals, which gives
Z −1/2
n iφ(x)/κ iνπ/4 n/2 ∂ 2 φ(x̄)
d xe =e (2πκ) det eiφ(x̄)/κ (363)
∂xk ∂xl

with ν = ν+ − ν− , where ν± is the number of positive/negative eigenvalues of ∂ 2 φ/∂xk ∂xl .

With more than 1 stationary phase point, we sum over them:
Z X −1/2
n iφ(x)/κ iνb π/4 n/2 ∂ 2 φ(x̄b )
d xe = e (2πκ) det eiφ(x̄b )/κ (364)
b
∂xk ∂xl

5.4.2 Stationary Phase approximation applied to the path integral

Recall the discretized version of the path integral

( N −1  )
 m N/2 Z i X m(xj+1 − xj )2
K(x0 , x, t) = lim dx1 . . . dxN −1 exp − V (xj )
N →∞ 2πi~ ~ j=0 22
(365)

The quantity φ in this case is

X
N −1  
m(xj+1 − xj )2
φ(x0 , x1 , . . . , xN ) =  − V (xj ) (366)
j=0
22

and κ = ~.

51
 We need a second derivative of φ

∂φ hm i
=  2 (2xk − xk+1 − xk−1 ) − V 0 (xk ) (367)
∂xk 
∂ 2 φ(x̄b ) m
= Qkl (368)
∂xk ∂xl 

with

2 00
Qkl = 2δkl − δk+1,l − δk−1,l − V (xk )δkl , (369)
m
k, l ∈ (1, . . . , N − 1) (370)

The stationary points are (discretized) paths x̄k that make ∂φ/∂xk = 0. I.e.,

x̄k+1 − 2x̄k + x̄k−1

m 2
= −V 0 (x̄k ), (371)


which is the discretized version of Newton’s laws:

d2 x̄(τ )
m = −V 0 (x̄), (372)
dτ 2

so that x̄(τ ) is a classical path, satisfying x̄(0) = x0 , x̄(t) = x.

Or, noting

lim φ(x̄) = S(x, x0 , t), (373)

N →∞

we can write it in the form of Hamilton’s principle

δS
=0 (374)
δx(t)

The path integral is dominated by the classical path(s).

To finish the calculation, we need the determinant and signs of eigenvalues for ∂ 2 φ/∂xk ∂xl ,
evaluated at x̄k (or in the limit, x̄(τ ).
The determinant will be multiplied by
 m N/2  m N/2
−iN π/4
=e , (375)
2πi~ 2π~

which diverges as −N/2 as N → ∞.

=⇒ determinant must diverge as −N to get finite result

52
    m N −1
∂ 2φ
det = det Qkl (376)
∂xk ∂xl 

=⇒ Qkl ∼ 1/ for finite result

Skipping the details, the result is:
 −1
m ∂ 2S
lim det Qkl = − (377)
N →∞  ∂x∂x0

We also need the number of negative eigenvalues µ. Note that the total number of
eigenvalues is the dimension of the matrix, ν+ + ν− = N − 1, so

ν = ν+ − ν− = N − 1 − 2ν− (378)

So the total prefactor is

e−iN π/4 eiνπ/4 = e−iπ/4 e−iµπ/2 (379)

µ has a finite limit for N → ∞, so the overall phase reaches a definite limit
In this limit, µ is the number of negative eigenvalues of operator

m d2 1 00

B=− − V x̄(τ ) (380)
2 dτ 2 2

For µ = 0, the action is a minimum along the classical path.

For µ > 0, the path integral gets an extra phase e−iµπ/2 .
Finally, we have:

1/2  
e−iµπ/2 ∂ 2 S i
K(x, x0 , t) = √ exp S(x, x0 , t) (381)
2πi~ ∂x∂x0 ~

If there is more than one classical path:

X e−iµb π/2 ∂ 2 Sb 1/2  

i
K(x, x0 , t) = √ exp Sb (x, x0 , t) (382)
b
2πi~ ∂x∂x0 ~

In 3 dimensions, the result is:

53
 X e−iµb π/2 1/2  
∂ 2 Sb i
K(~x, ~x0 , t) = √ det exp Sb (~x, ~x0 , t) (383)
b
2πi~ ∂~x∂~x0 ~

“Van Vleck” formula – exact if V is at most quadratic in x

V (x) = a0 + a1 x + a2 x2 (384)

I.e., free particle, particle in uniform gravitation field, harmonic oscillator. Also for uni-
form magnetic field.

5.5 Aside: Gaussian integrals

We computed Gaussian integrals several times in the previous lectures, without discussing
the details. Here are a few tips for Gaussian integrals:
In 1D. we start with a general integral of the form
Z ∞
2 √
dxe−x = π (385)
−∞

An arbitrary rescaling by a and shift by b gives

Z ∞
r
−a(x−b)2 π
dxe = (386)
−∞ a

(even for a, b complex, as long as Re(a) ≥ 0)

When you have an arbitrary linear term added, you must complete the square
Z ∞ Z ∞
−ax2 +bx 2 +bx− b2 + b2
dxe = dxe−ax 4a 4a (387)
−∞
Z−∞
∞
b 2 b2
= dxe−a(x− a ) + 4a
(388)
−∞
r
b2 π
= e 4a (389)
a

Multiplying N integrals together gives the formula for a multi-dimensional Gaussian in-

54
tegral

Z s
∞ Y
N
)2 πN
dxi e−ai (xi −bi = Q (390)
−∞ i=1 ai
Z ∞
−ai (xi −bi )2
P
= dN xe (391)
Z−∞
∞
| Λx̄
= dN x̄ex̄ (392)
−∞
r
π
= (393)
det Λ
Z ∞
|
= dN x̄R ex̄R K x̄R (394)
−∞
r
πN
= (395)
det K

with

x̄i = xi − bi (396)
Λij = ai δij (397)
x̄R = Rx (398)
K = RΛR−1 (399)

and K therefore an arbitrary symmetric, positive definite matrix.

This extends to operators in infinite-dimensional spaces
For reference, in the multidimensional case, completing the square gives
Z ∞ n
π2

N |
d x exp (−x Ax + J x) = |
1 exp J | A−1 J (400)
−∞ |A| 2

55
6 Some topics in 1D wave mechanics
These are some general results for solutions of the time-independent Schrödinger equation in
1 dimension:

~2 d2 ψ
− + V (x)ψ(x) = Eψ(x) (401)
2m dx2

6.1 Degeneracies
Consider two solutions of Eq. (401),

~2 00
− ψ + V (x)ψ(x) = Eψ(x) (402)
2m
~2 00
− φ + V (x)φ(x) = Eφ(x) (403)
2m

Multiply the first equation by φ and the second by ψ, and then subtract:

dW
φψ 00 − ψφ00 = 0 = (404)
dx

where S is the Wronskian

φ ψ
W = φψ 0 − ψφ0 = (405)
φ0 ψ 0

=⇒ W = constant.
If there is a point x0 where

ψ(x0 ) = φ(x0 ) = 0 (406)

Then W = 0 everywhere. This means

ψ0 φ0
= (407)
ψ φ
=⇒ ψ(x) = cφ(x) (408)

=⇒ energy eigenstates non-degenerate.

This includes points x0 = ±∞.
=⇒ bound states non-degenerate.

56
6.2 Reality of energy eigenfunctions
The Schrödinger equation is real. So if ψ is a solution with energy E, so is ψ ∗ . But if E is
not degenerate, they must be proportional:

ψ = cψ ∗ (409)

with |c|2 = 1, since squaring both sides gives

|ψ|2 = |c|2 |ψ|2 (410)

So

ψ = eiα ψ ∗ (411)

Now define

φ(x) = e−iα/2 ψ(x) = eiα/2 ψ(x)∗ = φ(x)∗ (412)

=⇒ any nondegenerate energy eigenfunction in 1D can be chosen to be real.

For degenerate case, can always choose linear combinations that are real.
(related to time-reversal invariance)

6.3 Interleaving of nodes

The zeros xn of an energy eigenfunction are called nodes

ψ(xn ) = 0 (413)

Consider a potential V (x) that supports bound states with energies E0 < E1 < E2 < . . ..
(We know they must be nondegenerate)
It can be shown that the state with energy En has exactly n nodes.
The nodes interleave — the nodes of ψn (x) are between the nodes of ψn+1 (x)
For example, energy eigenfunction of the harmonic oscillator:

57
 n=0
n=1 ψn (x) n=5
n = 2 n=6
n nodes. Moreover, the nodes interleave one another, that is, the nodes of ψn (x) lie between the
nodes of ψn+1 (x). This can be seen in some examples such as the harmonic oscillator. See Figs. 1
and 2. The proof of these facts may be found in Morse and Feshbach.
ψn (x) x x
n=0
n=1 ψn (x) n=5
n=2 n=6

Fig. 1. Eigenfunction ψn (x) has n nodes. Harmonic os- Fig. 2. The nodes of successive eigenfunctions interleave,
x x
cillator eigenfunctions are illustrated for n = 0, 1, 2. here illustrated for n = 5, 6. The small dots are the nodes
for n = 5, the large ones for n = 6.

Here is an application. Consider the double well oscillator, whose potential is sketched in Fig. 3.
Fig. 1. Eigenfunction ψn (x) has n nodes. Harmonic os- Fig. 2. The nodes of successive eigenfunctions interleave,
If the barrier
cillatorin the middle
eigenfunctions is high,forthen
are illustrated n = 0,tunnelling
1, 2. is small
here and
illustrated the
for n = 5,ground state
6. The small wave
dots are functions
the nodes
for n = 5, the large ones for n = 6.
in the left and right wells do not communicate with each other very much. Thus, both ψℓ (x) and
ψr (x) are nearly
Hereground states forConsider
is application:
an application. the wholedouble
potential, and there is apotential
near degeneracy.
Example symmetricthe well oscillator,
double well whose is sketched in Fig. 3.
If the barrier in the middle is high, then tunnelling is small and the ground state wave functions
in the left and right wells do not communicate with each other very much. Thus, both ψℓ (x) and
ψr (x) are nearly ground states for the whole potential, and there is a near degeneracy.
V (x)

ψℓ (x) ψr (x)
V (x)

ψℓ (x) ψr (x)

x state wave functions in the left

Fig. 3. A double well potential (heavy line). Also shown are the approximate ground
and right wells.
If barrier is high, tunneling is small, and localized ground state harmonic oscillator states
Fig. 3. A double well potential (heavy line). Also shown are the approximate ground state wave functions in the left
onand
theright
leftwells.
or right are almost ground states, and there is a near degeneracy.
As shown
Frominthe Sec. 2, however,
above, we knowthere
therecannot
can’t bebeaan exact degeneracy.
degeneracy. Now
What is the sinceground
actual the Hamiltonian
state?
commutes withAs shown in Sec. 2, however,
parity, isbyinvariant
The system Theoremunder there
1.5 the cannot be an
eigenstates
parity exact
x → −x.ofI.e., degeneracy.
H, which Now since the
are nondegenerate,
the parity Hamiltonian
operator commutes must also be
with
commutes with parity, by Theorem 1.5 the eigenstates of H, which are nondegenerate, must also be
eigenstates of parity. Since ψℓ and ψr are not eigenstates of parity, to get the energy eigenstates we
the Hamiltonian
eigenstates of parity. Since ψℓ and ψr are not eigenstates of parity, to get the energy eigenstates we
must formmust
symmetric and antisymmetric linear combinations,
form symmetric and antisymmetric linear combinations,
π̂|xi = | − xi (414)
1 1
ψ± ψ=±√= π̂|pir ±
√(ψ(ψ r ±ψψ
ℓ ℓ),
), (11) (11)
=⇒ 2 2 = | − pi (415)

which arewhich are illustrated

illustrated in Figs.in 4Figs.
and4 5. 5. =⇒
andThese [Ĥ,
These
areareπ̂] =0
eigenstates
eigenstates of parity.
of parity. (416)

So energy eigenstates are also eigenstates of parity. So the two lowest eigenstates are
symmetric and antisymmetric linear combinations

58
 1
ψ± = √ (ψr ± ψl ) (417)
4 2
Notes 6: Topics in One-Dimensional Wave Mechanics

Which are even and odd eigenstates of parity

ψ+ (x) ψ− (x)

x x

Fig. 4. Even eigenfunction in double well potential. Fig. 5. Odd eigenfunction in double well potential.

But which one of them is the ground state, and which the excited state? According to the rules
weWhich is the ground
have presented, state? Itstate
the symmetric mustinbe the4 state
Fig. has nowith
node,fewer nodes.
and must be the ground state, while
the antisymmetric state in Fig. 5 has one node and must be the first excited state.

Problems

1. A δ-function potential is often a useful model, especially in one-dimensional problems. The

potential
V (x) = −aδ(x) (12)

can be seen as the limit of a square well potential,

!
1/L, |x| ≤ L/2,
V (x) = −a × (13)
0, |x| > L/2,

as L → 0. The parameter a has dimensions of energy times distance. The potential shown is a
δ-function well potential (for a > 0), but by changing the sign of a we get a δ-function barrier.
Show that the δ-function well supports precisely one bound state. Find the energy and the
wave function.
Hint: This can be done by taking the limit of a square well solution, but the following is an easier
method.
Normally we require the wave function and its first derivative to be continuous in x in a one-
dimensional solution of the Schrödinger equation. If the first derivative had a discontinuity, it would
mean that the second derivative, which appears in the kinetic energy term, would have a δ-function
at the location of the discontinuity of the first derivative. That δ-function can only be balanced out
in the Schrödinger equation if the potential has a δ-function. However, in the present situation, that
is precisely what we have. 59
So solve the Schrödinger equation in the region x < 0 and x > 0, where V = 0, and then find
the discontinuity in ψ ′ across x = 0 by integrating over a small interval [−ϵ, ϵ].
7 Particles in Electromagnetic Fields
Consider a single particle moving in an (external) electromagnetic field.
In classical electromagnetism, the equations of motion are written in terms of E and B,
 
1
ma = q E(x, t) + v × B(x, t) , (418)
c

where v = ẋ and a = ẍ.

However, the classical Hamiltonian cannot be written in terms of E and B, but requires
the use of a scalar Φ and vector A potential:

1 ∂A
E = −∇Φ − (419)
c ∂t
B = ∇ × A. (420)

The classical Hamiltonian is then:

1 h q i2
H= p − A(x, t) + qΦ(x, t) (421)
2m c

The momentum p is the canonical momentum (the generator of translations, ∂L/∂x),

q
p = mv + A(x, t), (422)
c

and not the kinematic momentum pkin = mv.

So the first term in the Hamiltonion is the usual kinetic energy (1/2)mv 2 .
For a quantum system, we use the same expression, but interpret x and p as operators.
There is an ambiguity in the term p · A = A · p, since these are equivalent in the classical
system, but different in the quantum system since [xi , pj ] = i~δij .
We choose a prescription so that the Hamiltonian is always Hermitian:

1 
p·A→ p̂ · Â + Â · p̂ (423)
2
p̂2 q   q2
Ĥ = − p̂ · Â + Â · p̂ + A2 + qΦ (424)
2m 2mc 2mc2

The time-dependent Schrödinger equation in configuration space is then

1 h q i2 ∂ψ(x, t)
−i~∇ − A(x, t) ψ(x, t) + qΦ(x, t)ψ(x, t) = i~ (425)
2m c ∂t

60
where
h q i2 i~q i~q q2
−i~∇ − A ψ ≡ −~2 ∇2 ψ + ∇ · (Aψ) + A · ∇ψ + 2 A2 ψ (426)
c c c c

7.1 Gauge Transformations

In classical mechanics, the potentials are not unique, but are equivalent to another set of
potentials that are related by a gauge transformation

1 ∂f
Φ0 = Φ − , (427)
c ∂t
A0 = A + ∇f (428)

for any function f (x, t).

Any measureable quantity is invariant under a guage transformation (independent of f ).
E.g., E an B are invariant and can be measured directly, but Φ and A depend on the
gauge choice and can’t be measured.
In a quantum system, there is also a similar concept of gauge invariance. However, to
keep the Schrodinger equation invariant, the wavefunction also must transform under a gauge
transformation:

ψ 0 = eiqf /~c ψ (429)

So then
 q   q  −iqf /~c 0
p− A ψ = p− A e ψ (430)
c c 
q q
= e−iqf /~c p − A − ∇f ψ 0 (431)
 c c
−iqf /~c q 0
=e p− A ψ (432)
c
1  q 2 −iqf /~c 1
 q 2
p− A ψ =e p − A0 ψ (433)
2m c 2m c

61
Similarly,
   
∂ ∂
i~ − qΦ ψ = i~ − qΦ e−iqf /~c ψ 0 (434)
∂t ∂t
 
−iqf /~c ∂ q ∂f
=e i~ − qΦ + ψ0 (435)
∂t c ∂t
 
−iqf /~c ∂
=e i~ − qΦ ψ 0
0
(436)
∂t

So every term in the Schrodinger equation has an extra factor of e−iqf /~c , which cancels:

1 h q 0 i2
0 0 0 ∂ψ 0 (x, t)
−i~∇ − A (x, t) ψ (x, t) + qΦ (x, t)ψ (x, t) = i~ (437)
2m c ∂t

7.2 Aharonov-Bohm effect

We start with the following simple setup: A particle is confined to a cylindrical shell with
hard walls such that the wavefunction zero for inside the inner radius ψ(ρ ≤ ρa ) and outside
the outer radius ψ(ρ ≥ ρb ), where ρ is the radius in cylindrical coordinates
p
ρ= x2 + y 2 . (438)

The energy levels of the system can be solved in cylindrical coordinates with appropriate
boundary conditions ψ(ρa ) = ψ(ρb ) = 0.

62
 2m
∇2 ψ = Eψ (439)
−i~
 2
∂ ∂ 1 ∂
= ρ̂ + ẑ + φ̂ ψ (440)
∂ρ ∂z ρ ∂φ

Now consider the same setup, except with a uniform magnetic field enclosed by the shell
such that the magnetic field vanishes in the region of the particle ρa < ρ < ρb
Even though B = 0 in the region of the particle (i.e., the particle cannot travel to a region
of space with B 6= 0), its presence still affects the energy spectrum of the system.
We can see this by noting that the vector potential is non-zero in the region of the particle.
For ρ < ρa . we can write the vector potential as
 
Bρ
Ain = φ̂, ρ < ρa , (441)
2

which generates a uniform magnetic field B = B ẑ. The field for ρ > ρa is B = ∇ × A = 0,
but the potential must be continuous at ρ = ρa :

 
Bρ2a
Ashell = φ̂, ρ > ρa (442)
2ρ

The Schrodinger equation is modified by

ie
∇→∇− A (443)
~c

which in this case means

 
∂ ∂ ie Bρ2a
→ − (444)
∂φ ∂φ ~c 2

This causes an observable change in the energy spectrum, even though the field B vanishes
where the wavefunction is nonzero.
This is a simple bound-state example of the Aharonov-Bohm effect. Next we consider the
original form of the effect:
We still have a uniform magnetic field inside a cylinder of radius ρa , which the particle
cannot enter. However, the particle is no longer restricted to ρ < ρb .
The particle is emitted from point A on one side of the cylinder, and detected at point

63
B on the other side:

The probability to detect the particle at B is affected by interference that depends on the
presence of the magnetic field.
Instead of using the Schrodinger equation, we use the path integral.
With the addition of a magnetic field, the classical Lagrangian becomes
 2
(0) m dx e dx
L = → L = L(0) + ·A (445)
2 dt c dt

So the corresponding change in action for some infinitesimal time step tn−1 → tn is
Z
(0) (0) e tn dx
S (n, n − 1) → S (n, n − 1) + dt ·A (446)
c tn−1 dt
Z
(0) e xn
= S (n, n − 1) + A · ds, (447)
c xn−1

where ds is the differential line element along the path segment.

The total change for a given path is then

Y   Y  (0)   Z xN 
iS (0) (n, n − 1) iS (n, n − 1) ie
exp → exp exp A · ds (448)
~ ~ ~c x1
R
The contour integral A · ds is independent of the path, unless a pair of paths forms a
loop the encloses a magnetic flux.
=⇒ all paths above the cylinder get the same phase, and paths below get another phase.

64
 So we can split the path integral into two parts:
Z   Z  (0) 
iS (0) (N, 1) iS (N, 1)
D[x(t)] exp + D[x(t)] exp (449)
above ~ below ~
Z  (0)   Z xN  
iS (N, 1) ie
→ D[x(t)] exp exp A · ds (450)
above ~ ~c x1
Z  (0)   Z xN  above
iS (N, 1) ie
+ D[x(t)] exp exp A · ds (451)
below ~ ~c x1 below

The interference term (obtained from squaring the probability amplitude) depends on the
phase difference
 Z xN   Z xN  I
e e e
A · ds − A · ds = A · ds (452)
~c x1 above
~c x1 below
~c
e
= ΦB (453)
~c

where ΦB is the magnetic flux inside the cylinder.

If one changes the strength of the magnetic field, there will be a sinusoidal interference
effect in the probability of observing the particle. The period is given by a fundamental unit
of magnetic flux

2π~c
= 4.135 × 10−7 gauss − cm2 (454)
|e|

Classically, the Lorentz force is zero everywhere, but in quantum mechanics, the magnetic
flux has observable affects on the phase of the wavefunction.

65
8 Angular Momentum
8.1 Groups
Symmetry operators can be classified into groups:
Group: set + definition for combining two elements
(here: element of set = symmetry operator U , combined by sequential operation U U 0 )
Must satisfy 4 axioms:

1. Closure: U U 0 = U 00 is also element of group

2. Associativity: (U1 U2 )U3 = U1 (U2 U3 )

3. Identity: includes identity, U 1 = 1U ∀U

4. Inverse: every element has inverse, U U −1 = U −1 U = 1

Important groups in physics: “Lie groups” – continuous, differentiable.

“Abelian” group: [U, U 0 ] = 0 ∀U, U 0 .
“Non-abelian”: [U, U 0 ] 6= 0.
Operators forming Lie groups can be written in terms of generators of transformations
— “Lie algebra” defined by commutators of generators.
Different sets of symmetry operations can have same group structure =⇒ same proper-
ties. I.e., there are different “representations” of a group.
We want to find representations of the rotation operator on various possible Hilbert spaces
— the same group as rotation matrices in 3 dimensions.

8.2 Proper rotations, SO(D)

8.2.1 O(D), SO(D), and parity

Coordinate transformations that preserve scalar product

(form group O(|{z}
D ), “orthogonal” group)
dimensions

66
 xi → Rij xj , , with (455)
~x · ~y → ~x · ~y (456)
=⇒ (Rx)i (Ry)i = xi yi (457)
= Rij xj Rik yk (458)
= δjk xj yk (459)
=⇒ Rij Rik = δjk (460)
= (Rji )T Rik (461)
=⇒ RT R = 1 (462)

det(RT R) = det 1 = 1 (463)

T
Notes 11: Rotations= in
det(R ) det(R) Space
Ordinary (464)
= det(R)2 (465)
=⇒ det(R) = ±1 (466)
not form a group, since it does not contain the identity element. An improper rotation of som
Transformations with det = +1: SO(D), “special orthogonal” group
importance is
⎛
Proper rotations (continuously connected ⎞
−1 to 0 identity)
0
Transformations with det = -1 (not
P = ⎝ 0a group
−1 by0themselves
⎠ = −I, – doesn’t include identity): (16
“parity” transformation
0 0 −1
Includes reflections, parity : ~x → −~x
which inverts vectors through the origin.
We will only discuss proper rotations this semester.
For the duration of these Notes we will mostly restrict consideration to proper rotations whos
matrices8.3
belongCoordinate
to the groupRotations
SO(3), but we will return to improper rotations at a later date.

z
θ

n̂

x
Fig. 1. A proper rotation is specified by a unit vector n̂ that defines an axis, and a rotation of angle θ about that axi
The rotation follows the right-hand rule.
67

6. Rotations About a Fixed Axis; the Axis-Angle Parameterization

 Consider a proper rotation about an axis n̂ by angle θ.
A (classical) vector transforms according to the rotation matrix R(n̂, θ).
For example,
 
1 0 0
 
R(x̂, θ) = 0 cos θ − sin θ (467)
0 sin θ cos θ
 
cos θ 0 sin θ
 
R(ŷ, θ) =  0 1 0  (468)
− sin θ 0 cos θ
 
cos θ − sin θ 0
 
R(ẑ, θ) =  sin θ cos θ 0 (469)
0 0 1

Any proper rotation can be written in terms of n̂ and θ.

Rotations do not commute, unless they are around the same axis:

R(n̂, θ)R(n̂, θ0 ) = R(n̂, θ0 )R(n̂, θ) = R(n̂, θ + θ0 ) (470)

Otherwise, R(n̂, θ)R(n̂0 , θ0 ) 6= R(n̂0 , θ0 )R(n̂, θ)

We want to quantify non-commutativity by specifying the Lie algebra.
Proper rotations are continuously connected to the identity. Consider an infinitesimal
rotation

R = 1 + A + O(2 ) (471)

R is orthogonal, so A is antisymmetric:

RT R = 1 (472)
=⇒ AT + A = 0 (473)

An antisymmetric matrix can be parametrized as

68
  
0 −a3 a2
 X
3

A =  a3 0 −a1  = ai Ji (474)
i=1
−a2 a1 0

where (J1 , J2 , J3 ) is a vector of matrices

 
0 0 0
 
J1 = 0 0 −1 (475)
0 1 0
 
0 0 1
 
J2 =  0 0 0 (476)
−1 0 0
 
0 −1 0
 
J3 = 1 0 0 (477)
0 0 0
=⇒ (Ji )jk = −ijk (478)

Comparing to the full rotation matrices for x, y, and z, we can see

R(n̂, θ) ' 1 + θn̂ · J + O(θ2 ) (479)

Since rotations along the same axis commute, this can be exponentiated

R(n̂, θ) = eθn̂·J (480)

~~
= eθ·J (481)

The matrices J serve as generators of rotations. We can derive their commutation relation:

(Ji Jj )kl = ika jal = δil δkj − δij δkl (482)

=⇒ [Ji , Jj ] = ijk Jk (483)

69
8.4 Quantum rotations
We want to find the rotation operator U on a quantum state.
In order for expectation values to rotate as expected, we postulate that they form a
representation of SO(3). I.e.:
For every rotation matrix R, there is an associated quantum operator

R → U (R) (484)

As a symmetry operator, it must preserve the norm of states, and so must be unitary:

U (R)−1 = U (R)† (485)

Rotations with θ = 0 are represented by the identity

U (1) = 1 (486)

The product of rotation operators is the operator corresponding to the product of rotation
matrices:

U (R1 )U (R2 ) = U (R1 R2 ) (487)

=⇒ inverses map:

U (R−1 ) = U (R)−1 = U (R)† (488)

Actually, this is too strong — for spin-1/2 systems, there are 2 possible rotation operators
for each rotation matrix R.
E.g., rotations of θ = 0 and θ = 2π both have R = 1, but in spin-1/2 system, we have
U (θ = 0) = 1 = −U (θ = 2π). In this case, the group is actually the double cover of SO(3),
SU (2) = Spin(3)

70
8.4.1 Infinitesimal rotations

Classically we have

R(~θ) = 1 + ~θ · ~J + . . . (489)
∂R(~θ)
= 1 + θi + ... (490)
∂θi θ=0

The infinitesimal quantum operator is

∂U (~θ)
U (~θ) = 1 + θi + ... (491)
∂θi θ=0
i
= 1 − ~θ · ~J (492)
~

Similar to translations and time evolution, we defined the generators as

∂U (~θ)
Jˆk = i~ , (493)
∂θi θ=0

and identify them as angular momentum operators.

In order for the rotation operators to have the correct commutation relations, the same
algebra as the classical case must be respected:

[Ji , Jj ] = i~ijk Jk (494)

Since rotations along the same axis commute,

[n̂ · J, n̂ · J] = 0, (495)

an infinitesimal rotation can be exponentiated to obtain a finite rotation:

i ~
U (n̂, θ) = e− ~ θn̂·J (496)

~ × ~p,
Note that we define angular momentum as a generator of rotations, and not as R
which is less general.

71
8.4.2 Spin 1/2

The simplest system is a 2-dimensional Hilbert space. We want to find operators that satisfy
the commutation relations (494).
Recall that the Pauli matrices satisfy

[σi , σj ] = 2iijk σk , (497)

with
! ! !
0 1 0 −i 1 0
σx = , σy = , σz = (498)
1 0 i 0 0 −1

So we can postulate that the angular momentum operators on a spin-1/2 system are

~J = ~ ~σ ≡ S ~ (499)
2
=⇒ [Ji , Jj ] = i~ijk Jk (500)

An arbitrary rotation operator is then

θ θ
U (n̂, θ) = e−iθn̂·σ/2 = cos − i(n̂ · σ) sin (501)
2 2

(Recall homework set 1, where you proved the last expression)

Expectation values should rotate according to the usual R(n̂, θ).
Earlier we showed that the operator that measures the magnetic moment for a spin-1/2
system is

µi = µ0 σi , (502)

in the basis where µz is diagonal (up to an unknown sign for µy ).

If a quantum state is rotated |ψi → |ψ 0 i = U (n̂, θ)|ψi, the expectation value of the
magnetic moment vector should rotate as expected

h~µi → R(n̂, θ)h~µi (503)

=⇒ hψ 0 |~σ |ψ 0 i = hψ|U †~σ U |ψi (504)
= Rhψ|σ|ψi (505)

72
 Since this should be true for any state, it must be that
X
U † σi U = Rij σj (506)
j

We can check this explicitly:

First, a simple example – rotation around the z axis:

U (ẑ, θ) = e−iθẑ·σ/2 = e−iθσz /2 (507)

Consider the effect of the rotation on hσx i.

First by explicit calculation, note that

σz |±i = ±|±i (508)

so that

U † σx U = eiθσz /2 σx e−iθσz /2 (509)

= eiθσz /2 (|+ih−| + |−ih+|)e−iθσz /2 (510)
= eiθ/2 |+ih−|eiθ/2 + e−iθ/2 |−ih+|e−iθ/2 (511)
= eiθ |+ih−| + e−iθ |−ih+| (512)
= [(|+ih−| + |−ih+|) cos θ + i (|+ih−| − |−ih+|) sin θ] (513)
= σx cos θ − σy sin θ (514)

Just as expected.
We can also show it using only the commutation relations, proving that it will also be
true for higher dimensional representations. The only thing we need is the correct algebra.
(from Sakurai:)

73
 In general, we have:

   
† θ θ θ θ
U ~σ U = cos + i(n̂ · σ) sin ~σ cos − i(n̂ · σ) sin (515)
2 2 2 2
θ θ θ θ
= cos2 ~σ + i cos sin [n̂ · σ, σ] + (n̂ · σ)σ(n̂ · σ) sin2 (516)
2 2 2 2
θ θ θ θ
= cos2 ~σ + i cos sin [−2in̂ × σ] + [2n̂(n̂ · σ) − σ] sin2 (517)
2 2 2 2
= cos θ~σ + (1 − cos θ) n̂(n̂ · ~σ ) + sin θn̂ × ~σ (518)
= R(n̂, θ)~σ , (519)

where we used the identities

[n̂ · σ, σ] = −2in̂ × σ (520)

(n̂ · σ)σ(n̂ · σ) = 2n̂(n̂ · σ) − σ (521)

8.4.3 Double-valued representation

Now consider a rotation of an arbitrary state around the z axis

e−iθσz /2 |ψi = e−iθ/2 |+ih+|ψi + eiθ/2 |−ih−|ψi (522)

Notice the half-angle θ/2.

For θ = 2π, for example,

|ψi → U (ẑ, θ = 2π)|ψi = −|ψi (523)

74
 Or more generally,

U (n̂, 0) = 1, U (n̂, 2π) = −1 (524)

Does this matter? The overall phase of a wavefunction is irrelevant.

However, one can rotate part of a system, and let it interact with an unrotated state.
Consider spin precession in a magnetic field.
The Hamiltonian is given by
 
e
H=− S · B = ωSz (525)
me c

with Si the spin-1/2 representation of angular momentum operators Ji and

|e||B|
ω= (526)
me c

The time evolution operator is

U (t) = e−iHt/~ = e−iSz ωt/~ = e−iωtσz /2 (527)

= U (ẑ, θ = ωt) (528)

So moving through a magnetic field is equivalent to a rotation.

Expectation values rotate with period t = 2π/ω:

hµx it = hµx it=0 cos ωt − hµy it=0 sin ωt (529)

hµy it = hµy it=0 cos ωt + hµx it=0 sin ωt (530)
hµz it = hµz it=0 (531)

However, the states get a minus sign after time t = 2π/ω, and take twice as much time
to return to the same state

e−iθσz /2 |ψi = e−iθ/2 |+ih+|ψi + eiθ/2 |−ih−|ψi (532)

75
 So we have the periods:

2π
τprecession = (533)
ω
4π
τket = (534)
ω

8.4.4 Neutron interferometry experiment to study 2π rotations

To study this effect, one must rotate part of the system, and let the rotated and unrotated
parts interact

A beam of neutrons is split in two. One beam (A) never experiences a magnetic field,
while the other (B) experiences a uniform magnetic field for time T .
The beam in path B gets a phase change from the magnetic field

e∓iωT /2 , (535)

with frequency

gn eB
ω= , (gn ' −1.91) (536)
mp c

for the newtron with magnetic moment gn e~/hmp c (compare to the result for an electron
with magnetic moment e~/2me c).
(The upper/lower sign corresponds to the component |+i/|−i, respectively).
When the beams meet, the amplitudes have a relative phase that depends on B:

||Ai + |Bi|2 = hA|Ai + hB|Bi + 2RehA|Bi (537)

h ∓iωT i
RehA|Bi = Re Ce 2 (538)
∓ωT
= C 0 cos( + δ) (539)
2

76
 Even if we don’t know δ, we can check this by changing B and measuring the period of
the interference, which should be

4π~c
∆B = , (540)
egn λ̄l

where l is the path length.

If the state goes to itself under a rotation of 2π instead of 4π, the period would be halved.
Experiments prove the quantum prediction correct.
So rotations in spin-1/2 systems follow a representation of SU (2) (the group of unitary,
unimodular 2x2 matrices) instead of SO(3). They have the same algebra, so they are said
to be locally isomorphic.
However, the global structure of the group is different. SU (2) is the double cover of
SO(3), meaning that there is a 1 to 2 relationship, so that for every element of SO(3) there
are exactly two elements of SU (2).


U (n̂, θ),
R(n̂, θ) → (541)
U (n̂, θ + 2π) = −U (n̂, θ)

8.5 Euler Rotation

An arbitrary rotation is parameterized by 3 real numbers – e.g., two angles specifying the
axis n̂ and an angle θ. Another way to paramaterize is using Euler angles, defined by 3
successive rotations:

77
 R(α, β, γ) ≡ Rz0 (γ)Ry0 (β)Rz (α) (542)

The unprimed directions are “body” axes, defined with respect to an object that has been
rotated.
We can write them in terms of fixed axes as

Ry0 (β) = Rz (α)Ry (β)Rz−1 (α) (543)

Rz0 (γ) = Ry0 (β)Rz (γ)Ry−1
0 (β) (544)

Plugging these in gives:

Rz0 (γ)Ry0 (β)Rz (α) = Ry0 (β)Rz (γ)Ry−1

0 (β)Ry 0 (β)Rz (α) (545)
= Rz (α)Ry (β)Rz−1 (α)Rz (γ)Rz (α) (546)
= Rz (α)Ry (β)Rz (γ) (547)

We can write this rotation matrix explicitly for a spin-1/2 system:

78
 U (α, β, γ) = U (ẑ, α)U (ŷ, β)U (ẑ, γ) (548)
= e−iσ3 α/2 e−iσ2 β/2 e−iσ3 γ/2 (549)
! ! !
e−iα/2 0 cos β2 − sin β2 e−iγ/2 0
= (550)
0 eiα/2 sin β2 cos β2 0 eiγ/2
!
e−i(α+γ)/2 cos β2 −e−i(α−γ)/2 sin β2
= (551)
ei(α−γ)/2 sin β2 ei(α+γ)/2 cos β2

Later we will discuss representation theory, and see that this is a matrix element

 
(1/2) 1 0 − iJz α − iJy β − iJz γ 1
Dm0 m (α, β, γ) = j = ,m e ~ e ~ e ~ j = ,m (552)
2 2

79
8.6 Representations
Now that we have seen an explicit example of rotation operators in a 2-dimensional Hilbert
space, we want to generalize to other spaces. Let’s discuss the general aspects of representa-
tions of rotation operators.
In general, we need to find a representation of angular momentum operators, which are
Hermitian operators that satisfy

[Ji , Jj ] = i~ijk Jk (553)

Once those are found, the rotation operators are exponentials

i ~
U (n̂, θ) = e− ~ θn̂·J (554)

Or, given the rotation operator, U (n̂, θ) = U (~θ), we can find ~J by

∂U (~θ)
Ji = i~ (555)
∂θi ~θ=0

Lets define another operator that commutes with each angular momentum operator

J 2 = J12 + J22 + J32 = Ji Ji (556)

[J 2 , Jk ] = [Ji Ji , Jk ] (557)
= Ji Ji Jk − (Ji Jk Ji − Ji Jk Ji ) − Jk Ji Ji (558)
= Ji [Ji , Jk ] + [Ji , Jk ]Ji (559)
= iJi ikm Jm + iikm Jm Ji (560)
= i(ikm + mki )Ji Jm (561)
=0 (562)

=⇒ J 2 commutes with any function of ~J, including rotation operators. An operator

that commutes with all generators of a group is called a Casimir operator.
We can construct simultaneous eigenkets of J 2 and any component of ~J (but only 1
component, since the components Ji don’t commute with each other).
By convention we choose J3 .

80
 J 2 |ami = ~2 a|ami (563)
J3 |ami = ~m|ami (564)
ham|ami = 1 (565)

With this convention, a and m are dimensionless.

Note that they are also real since J 2 and J3 are Hermitian.
We also know that a ≥ 0, since J 2 is nonnegative definite:
X X 2
ham|J 2 |ami = ham|J i J i |ami = Ji |ami (566)
i i

Lets assume (J 2 , J3 ) form a complete set of commuting operators. I.e., |ami are nonde-
generate and form a basis. We will check later what happens if this isn’t true.
The eigenvalue equations only determine |ami up to a phase, but we can choose some
phase convention to define them unambiguously.
Define (non-Hermitian) ladder operators

J± = J1 ± iJ2 (567)
J±† = J∓ (568)
[J3 , J± ] = ±~J± (569)
[J+ , J− ] = 2~J3 (570)
[J 2 , J± ] = 0 (571)

They satisfy the relations

1
J 2 = (J+ J− + J− J+ ) + J32 (572)
2
J− J+ = J 2 − J3 (J3 + ~) (573)
J+ J− = J 2 − J3 (J3 − ~) (574)

They are called ladder operators (or raising and lower operators) because they change

81
the value of m:

J3 (J± |ami) = ([J3 , J± ] + J± J3 )|ami (575)

= ~(m ± 1)(J± |ami) (576)

The value of a is unchanged, since [J 2 , J± ] = 0. So

J± |ami = c± |a, m ± 1i (577)

Note that we could have c± = 0, in which case the eigenvalues are bounded by some
max/min value of m:

ham|J− J+ |ami = |J+ |ami|2 = ~2 [a − m(m + 1)] ≥ 0 (578)

ham|J+ J− |ami = |J− |ami|2 = ~2 [a − m(m − 1)] ≥ 0 (579)
=⇒ a ≥ max [m(m + 1), m(m − 1)] (580)
4 Notes 13: Representations of Rotations

max[m(m + 1), m(m − 1)]

m(m − 1) m(m + 1)

m m
−2 −1 1 2 −2 −1 1 2
m = −j m=j

Fig. 2. Function max[m(m+1), m(m−1)], with maximum

Fig. 1. Functions m(m + 1) and m(m − 1). and minimum values of m for a given value of a.

Thus, knowing the eigenvalue a of J 2 , there are bounds on the eigenvalue m of J3 . The quantity j
specifying the bounds is a function of a, and, as is clear from the figure, j ≥ 0 since a ≥ 0.
For a given m,
Fromthere is see
Fig. 2 we a limit
that theto the possible
quantities a and j are value of a. Or, conversely, given a, there is
related by
a maximum (and minimum) possible value a =of
j(j m,
+ 1).which we can label j. (20)

It turns out to be more convenient to parameterize the eigenvalues of J 2 by j instead of a, so

henceforth let us write j(j + 1) for the eigenvalue of J 2 instead of a, and let us write |jm⟩ instead
of |am⟩ for the eigenkets of J 2 and J3 . Then Eqs. (7) become
−j ≤ m ≤ j
J 2 |jm⟩ = j(j + 1)h̄2 |jm⟩,
(581)
(21)
J3 |jm⟩ = mh̄|jm⟩.
with the relationship between
We also rewrite Eqs. (16) a
andand j, this change of notation, noting that the right hand sides
(17) with
can be factored,
⟨jm|J− J+ |jm⟩ = h̄2 [j(j + 1) − m(m + 1)] = h̄2 (j − m)(j + m + 1) ≥ 0, (22)
a = j(j + 1) (582)
2 2
⟨jm|J+ J− |jm⟩ = h̄ [j(j + 1) − m(m − 1)] = h̄ (j + m)(j − m + 1) ≥ 0. (23)

Let us consider the conditions under which these two inequalities become equalities, that is, when
the matrix elements on the left hand sides vanish. For Eq. (22), we have J+ |jm⟩ = 0 if and only if
82
j−m=0 or j + m + 1 = 0, (24)

that is,
m=j or m = −j − 1. (25)
But by Eq. (19), m = −j − 1 is impossible, so we find
 We can label the states by j instead of a:

J 2 |jmi = ~2 j(j + 1)|jmi (583)

J3 |jmi = ~m|jmi (584)

Since m is bounded, it must be that

J+ |jji = 0 (585)
J− |j, −ji = 0 (586)

The min and max states are related by ladder operators, which change m by an integer,
and so

mmax − mmin = 2j = Integer (587)

So j is a nonnegative integer, or half integer

1 3
j ∈ {0, , 1, , . . .} (588)
2 2

and there are 2j + 1 states with a given j.

8.6.1 Matrix elements and phase conventions

In the simultaneously diagonalized basis, the matrix elements of J 2 and J3 are

hj 0 m0 |J 2 |jmi = ~2 j(j + 1)δj 0 j δm0 m (589)

hj 0 m0 |J3 |jmi = ~mδj 0 j δm0 m (590)

Next we determine the matrix elements of the ladder operators. We already know

J± |jmi = c±
jm |j, m ± 1i, (591)

83
and we can determine the coefficients up to a phase by squaring the state

hjm|J− J+ |jmi = |c+ 2 2 2

jm | = hjm|(J − J3 − ~J3 )|jmi (592)
 
= ~2 j(j + 1) − m2 − m = (j − m)(j + m + 1) (593)
hjm|J+ J− |jmi = |c− 2 2 2
jm | = hjm|(J − J3 + ~J3 )|jmi (594)
= (j + m)(j − m + 1) (595)

It is conventional to choose the phase so that all the c−

jm are real and positive. Since J+ J−
is nonnegative definite, this then implies that c− + +
jm cj,m−1 > 0, and all the cjm are also real and
positive.
So we have
p
J± |jmi = ~ (j ∓ m)(j ± m + 1)|j, m ± 1i (596)

The matrix elements of J1 and J2 can be determined by noting

1
J1 = (J+ + J− ) (597)
2
1
J2 = (J+ − J− ) (598)
2i

Since we chose J± to have real matrix elements, those of J1 are real and J2 are imaginary.

8.6.2 Invariant, irreducible subspaces

A subspace Xj of a Hilbert space X is “invariant” under the action of an operator if the

operator maps every vector of the subspace into another vector in the subspace (and never a
vector outside the subspace).

Ô|ψi = |ψ 0 i ∈ Xj ∀|ψi ∈ Xj (599)

For example, the eigenspaces of J 2 are invariant under any rotation (j does not change
by action of any rotation), and form invariant spaces under rotation:

Xj = span{|jmi, m = −j, . . . , +j} (600)

The entire Hilbert space is composed of the complete set of invariant subspaces Xi ,

84
 X
X= ⊕Xj = X0 ⊕ X 1 ⊕ X1 ⊕ X 3 . . . (601)
2 2
j

If a subspace itself contains a smaller subspace that is also invariant, it is reducible. If it

does not, it is irreducible.
Any invariant space can thus be decomposed into irreducible subspaces.
It turns out that the subspaces Xj are irreducible.
In the basis consisting of kets belonging to the irreducible subspaces, the rotation opera-
tors are in block diagonal form.

When each subspace is irreducible, it cannot be further reduced into smaller blocks

85
8.6.3 Matrices for angular momentum and operators

Let O = f (~J) be any function of ~J. The action one a basis state can be decomposed into
the other basis states
X
O|jmi = |j 0 m0 ihj 0 m0 |O|jmi (602)
j 0 ,m0
X
O|jmi = |jm0 ihjm0 |O|jmi (603)
m0

where we only need states of the same j, since O is block diagonal:

hj 0 m0 |O|jmi = δj 0 j hjm0 |O|jmi (604)

Some explicit examples:

J 2 is proportional to the identity:

hjm0 |J 2 |jmi = δm0 m ~2 j(j + 1) (605)

If we compute the matrices for J3 and J+ , we can easily get J− , J1 and J2 from them.
For j = 0

J3 = ~(0) (606)
J+ = ~(0) (607)

j = 12 :
!
1
0 2
J3 = ~ (608)
0 − 12
!
0 1
J+ = ~ (609)
0 0
~
=⇒ ~J = ~σ (610)
2

86
 j = 1:
 
1 0 0
 
J3 = ~ 0 0 0 (611)
0 0 −1
 √ 
0 2 0
 √ 
J+ = ~ 0 0 2 (612)
0 0 0

j = 23 :
 
3
2
0 00
 1 
0 2 0 0 
J3 = ~ 
0 0 − 1 0 
 (613)
 2 
3
0 0 0 −2
 √ 
0 3 0 0
 
0 0 2 0 
J+ = ~ 
0 0 0 √3
 (614)
 
0 0 0 0

For a rotation matrix O = U (n̂, θ), it is customary to denote matrix elements in repre-
sentation j as

(j)
Dm0 ,m (n̂, θ) = hjm0 |U (n̂, θ)|jmi (615)
X
=⇒ U (n̂, θ)|jmi = |jm0 ihjm0 |U (n̂, θ)|jmi (616)
m0
X (j)
= |jm0 iDm0 ,m (617)
m0

(D stands for the German Drehung, for “rotation”).

87
 For an arbitrary rotation, it is useful to write in terms of Euler angles:

(j)
Dm0 ,m (α, β, γ) = hjm0 |e−iαJz /~ e−iβJy /~ e−iγJz /~ |jmi (618)
X
= hjm0 |e−iαJz /~ |jm1 ihjm1 |e−iβJy /~ |jm2 ihjm2 |e−iγJz /~ |jmi (619)
m1 ,m2
0
= e−iαm −iγm hjm0 |e−iβJy /~ |jmi (620)
−iαm0 −iγm (j)
=e dmm0 (β) (621)

where d = the reduced rotation matrix, which contains the only non-trivial matrix elements.
With the phase conventions that we chose, the matrix elements of Jy are imaginary, so
the d = Real. (This is why the zyz convention is usually used in QM instead of zxz which
is common in classical mechanics.)
Examples:

d0mm0 (β) = (1) (622)

A rotation does not change a state with 0 angular momentum.

For j = 1/2, Jy = ~σy /2:
!
1/2 cos(β/2) − sin(β/2)
dmm0 (β) = cos(β/2) − iσ sin(β/2) = (623)
sin(β/2) cos(β/2)

The j − 1 result is worked out in Sakurai:

 
1
2
(1 + cos β) − sin
√β 1
2
(1 − cos β)
 2 
d1mm0 (β) = 

sin
√β
2
cos β − sin
√β
2

 (624)
1 sin
√β 1
2
(1 − cos β) 2 2
(1 + cos β)

Standard references usually have tables of the reduced rotation matrices, which can be
useful.

88
8.7 Orbital Angular Momentum
Consider a spinless particle moving in 3-dimensional space.
The angular momentum operator in this Hilbert space is called orbital angular momentum
J = L.
We can check that this is equivalent to the definition

~ = ~x × ~p
L (625)
=⇒ Li = ijk xj pk (626)

For example, the angular momentum commutation relations are satisfied:

[xi , pj ] = i~δij (627)

=⇒ [Li , Lj ] = i~ijk Lk (628)

Basis states are rotated as expected. For example, an infinitesimal z rotation:

iθ
U (ẑ, θ)|xyzi = e− ~ Lz |xyzi (629)
 
iθ
' 1 − (x̂py − ŷpx ) |xyzi (630)
~
h py px i
= 1 − i θx + i θy |xyzi (631)
~ ~
= |x − yθ, y + xθ, zi (632)

as expected.
In general,

U (R)|~xi = |R~xi (633)

8.7.1 Orbital angular momentum in spherical coordinates

Just as the general case, we can use a basis of eigenstates of L2 and Lz (labeled by l instead
of j):

L2 |lmi = ~2 l(l + 1)|lmi (634)

Lz |lmi = ~m|lmi (635)

Let’s find these operators and eigenstates in the configuration basis. The angular mo-

89
mentum operators are

   
∂ ∂ ∂ ∂
Lx = −i~ y −z = −i~ − sin φ − cot θ cos φ (636)
∂z ∂y ∂θ ∂φ
   
∂ ∂ ∂ ∂
Ly = −i~ z −x = −i~ cos φ − cot θ sin φ (637)
∂x ∂z ∂θ ∂φ
 
∂ ∂ ∂
Lz = −i~ x −y = −i~ (638)
∂y ∂x ∂φ

and the ladder operators

 
±iφ ∂ ∂
L± = Lx ± iLy = −i~e ±i − cot θ (639)
∂θ ∂φ

and the Casimir operator

1
L2 = (L+ L− + L− L+ ) + L2z (640)
2    
2 1 ∂ ∂ 1 ∂2
= −~ sin θ + (641)
sin θ ∂θ ∂θ sin2 θ ∂φ2

Note that L2 is the angular part of the Laplacian:

 
2 2 21 ∂ 2 2 ∂ψ L2
p̂ ψ = −~ ∇ ψ = −~ 2 r + 2 ψ, (642)
r ∂r ∂r r

Which is separable in radial and angular variables.

Also note that no component of L contains ∂/∂r. This is because rotations only change
the direction of vectors, and not the magnitude.
To find the eigenfunctions ψlm , we start by finding the stretched state, and then act with
the lowering operator to find the states with m < l.
The stretched state can be defined by

Lz ψll (~x) = ~lψll (~x) (643)

L+ ψll (~x) = 0. (644)

The first equation gives

∂ψll (r, θ, φ)
−i~ = ~lψll (r, θ, φ), (645)
∂φ

90
which has the general solution

ψll (r, θ, φ) = Fll (r, θ)eilφ . (646)

In general, representation of angular momentum can have j as an integer or half integer.

Since the wave function must be single valued, we see that in this case l must be an integer.
To get the dependence on θ, we use the fact that the raising operator annihilates Fll :
 
iφ ∂
L+ ψll (~x) = 0 = −i~e i − il cot θ Fll (r, θ)eilφ (647)
∂θ
∂Fll (r, θ)
=⇒ = l cot θFll (r, θ) (648)
∂θ

This has the solution

Fll (r, θ) = u(r) sinl θ (649)

So

ψll (r, θ, φ) = u(r) sinl θeilφ . (650)

Since angular momentum does not involve r, we can ignore the radial part, and study
the space of wavefunctions on a unit sphere. Each eigenfunction can be multiplied by an
arbitrary radial function, and it will still be an eigenfunction of L2 and Lz .
So we have the angular wavefunction.

Yll (θ, φ) = sinl θeilφ (651)

∝ hn̂|l, li (652)

The usual normalization for spherical harmonics are

Z
dΩ|Y |2 = 1 (653)

and so we find

r
(−1)l (2l + 1)! l ilφ
Yll (θ, φ) = l sin θe (654)
2 l! 4π

91
where we have also added the conventional phase factor (−1)l .
With the phase conventions chosen earlier for angular momentum, an arbitrary m state
is related to the stretched state by
s  j−m
(j + m)! J−
|jmi = |jji (655)
(2j)!(j − m)! ~

One can show that

 l−m  l−m
L− l ilφ eimφ d
sin θe = m sin2l θ (656)
~ sin θ d(cos θ)

So altogether we have
s  l−m
l
(−1) 2l + 1 (l + m)! einφ d
Ylm (θ, φ) = l sin2l θ (657)
2 l! 4π (l − m)! sinm θ d(cos θ)

We can write the result in terms of Legendre polynomials, defined by

(−1)l dl
Pl (x) = (1 − x2 )l (658)
2l l! dxl

If we set x = cos θ, then we can write the m = 0 result as

r
2l + 1
Yl0 (θ, φ) = Pl (cos θ) (659)
4π

The (−1) phase factor was chosen so that Yl0 (θ = 0) is real and positive (note that
Pl (1) = 1).
From m = 0 we can get the m > 0 functions via
s  m
(l − m)! L+
|lmi = |l0i (660)
(l + m)! ~
 m  l−m
L+ m m imφ d
Pl (cos θ) = (−1) sin θe Pl (cos θ) (661)
~ d(cos θ)
= (−1)m eimφ Plm (cos θ) (662)

92
where Plm are the associated Legendre functions

dm Pl (x)
Plm (x) = (1 − x2 )m/2 (663)
dxm

So altogether we have for m ≥ 0:

s
2l + 1 (l − m)! imφ
Ylm (θ, φ) = (−1)m e Plm (cos θ) (664)
4π (l + m)!

Repeating for m > 0 we get the relation

Yl,−m (θ, φ) = (−1)m Ylm (θ, φ)∗ (665)

8.7.2 Spherical harmonics and rotation operators

We can write the spherical harmonics as

Ylm (r̂) = hr̂|lmi, (666)

where |lmi is a state defined on the Hilbert space of a sphere

We can rotate the wave function:
X
(U (R)Ylm )(r̂) = Ylm (R−1 r̂) = hr̂|U (R)|lmi = hr̂|lm0 ihlm0 |U (R)|lmi (667)
m0

but the final factor is a D matrix

X
(U (R)Ylm )(r̂) = Ylm (R−1 ) = l
Ylm0 Dm 0 m (R) (668)
m0

Let’s choose r̂ → ẑ and change notation R → R−1 , so that r̂ = Rẑ.

X
l −1
Ylm (r̂) = Ylm0 (ẑ)Dm 0 m (R ) (669)
m0

If we write r̂ = (θ, φ) for the direction of interest, then a rotation then maps the z-axis
into this direction is simple in Euler angle form

93
 R(φ, θ, 0)ẑ = r̂ (670)

Also, the value of Ylm at the north pole is simple, since

Plm (cos θ = 1) = δm,0 (671)

r
2l + 1
=⇒ Ylm (ẑ) = δm,0 (672)
4π

Then we note that

j −1 j∗
Dmm 0 (U ) = [Dj (U )−1 ]mm0 = [Dj (U )† ]mm0 = Dm 0 m (U ). (673)

So finally we have

r
2l + 1 l∗
Ylm (θ, φ) = Dm0 (φ, θ, 0) (674)
4π

which shows a useful connection between the Ylm ’s and the D-matrices

8.8 Wigner’s formula for d matrices

For the general construction of Wigner’s d matrices, see Sakurai Ch. 3.9 (second edition,
starting with Schwinger’s oscillator model of angular momentum).

94
9 Particle in a central potential
Consider a spinless particle in a (3D) central potential (useful, for example when a light
particle is moving near a heavier particle)

p2
H= + V (r), (675)
2m
√ p
where the potential is only a function of r = x · x = x2 + y 2 + z 2 , and so is rotation
invariant, as is the kinetic energy term p · p = p2 .

=⇒ [L, H] = [U (R), H] = 0 (676)

=⇒ ∃ simultaneous eigenbasis of (H, L2 , Lz ).

From the derivation of spherical harmonics, we know the eigenstates have the form

ψ(r, θ, φ) = R(r)Ylm (θ, φ) (677)

By demanding that the wave function is also an eigenstate of H, we can determine R(r).

 
p2
+ V (r) ψ = Eψ (678)
2m
 
~2 2
= − ∇ + V (r) ψ (679)
2m
 
~2 1 d 2 dψ L2
=− r + ψ + V (r)ψ (680)
2m r2 dr dr 2mr2
 
~2 1 d 2 dψ ~2 l(l + 1)
=− r + ψ + V (r)ψ (681)
2m r2 dr dr 2mr2
 
~2 1 d 2 dR(r)
=⇒ − r + U (r)R(r) = ER(r), (682)
2m r2 dr dr

where the effective potential U is

~2 l(l + 1)
U (r) = + V (r) (683)
2mr2

= the centrifugal potential + the true potential

95
 We can simplify the equation by substituting

f (r) ≡ rR(r) (684)

~2 d2 f (r)
=⇒ − + U (r)f (r) = Ef (r) (685)
2m dr2

which looks like the Schrodinger equation of a particle moving in a 1D potential U (r), with
domain r > 0 (instead of −∞ ≤ x ≤ ∞).
The normalization condition is
Z Z ∞ Z
3 2 2 2
d x|ψ| = r dr|R(r)| dΩ|Ylm (θ, φ)|2 (686)
0
Z ∞
= r2 dr|R(r)|2 (687)
0
Z ∞
= |f (r)|2 (688)
0

=1 (689)

9.1 Behavior at small radius

Consider the radial wave function for r → 0. Let’s assume that the potential is 0 or finite at
r = 0, or diverges no faster than 1/r, so that

lim r2 V (r) = 0. (690)

r→0

Assume the wave function behaves as R(r) = ark for some power k, i.e.,

R(r)
lim =1 (691)
r→0 ar k

Let’s derive the allowed values of k. The Schrodinger equation at small r reads
 
~2 1 d 2 dR(r) ~2 l(l + 1)
− r + R(r) + V (r)R(r) = ER(r) (692)
2m r2 dr dr 2mr2
~2 ~2 (((
' −a k(k + 1)rk−2 + a l(l + 1)rk−2 + ( V( (r)R(r)
(((( −(ER(r) = 0, (693)
2m 2m

where the last terms vanish at least as fast as O(rk−1 ), and so are negligible in the small r
limit.

96
 So the solutions are

k(k + 1) = l(l + 1) (694)

=⇒ k = l, k = −l − 1 (695)

For l ≥ 1, the second solution k = −l − 1 is not allowed because R(r) ∼ r−l−1 has a
non-integrable singularity and is therefore not normalizable. (There is an infinite probability
near r = 0, and so it is not a valid wavefunction).
For l = 0, the solution k = −l−1 is also not allowed, because the wavefunction R(r) = a/r
would not satisfy the Schrodinger equation at r = 0:
 
21
∇ = −4πδ(x). (696)
r

So the only allowed solution is k = l, or

R(r) ∼ rl . (697)

97
 Notes 16: Central Force Motion 7

R(r)

ℓ=0

ℓ=1

ℓ=2

ℓ=3
r

Fig. 2. The radial wave function R(r) behaves as rℓ near r = 0. It lies down ever more flat against the r-axis as ℓ
Due
increases. This to the
is only the centrifugal barrier,
leading behavior particles
of R(r) with
for small larger
r; as l > 0 (i.e.,
r increases, nonzero
correction angular
terms becomemomen-
important.
tum) have smaller and smaller probability to be found near the origin.
Thus, the only possible solution of Eq. (19) is k = ℓ, and we see that the radial wave function
has the behavior
9.2 Free particle
R(r) ∼ rℓ
The simplest example is the free particle V (r) = 0.
(22)

near r = 0. H is invariant
The under both
leading behavior translations
depends andthe
only on rotations,
angularsomomentum
we can choose differentnumber
quantum complete
ℓ. This
setsrule
is a simple of commuting
that is oftenobservables.
important in practice. It means that the radial wave function R(r) lies
down more Ifandwe more
chooseflat
(pxnear
, py , pzr),=the simultaneous
0 as ℓ increaseseigenstates are and that the probability of finding
(see Fig. 2),
the particle in some small neighborhood of r = 0 goes to zero exponentially as ℓ increases. This
ψp (x) = eik·x , (698)
applies, for example, to the probability of finding an atomic electron inside the nucleus.
a plane wave with momentum p = ~k. The energy is
6. Free Particle
p2
E= (699)
2m
Let us now take the case of the free particle, V (r) = 0. In this case there is no question of the
mass of the
The object that
spectrum is creates thewith
continuous forceEfield,
> 0. since there is no force. The free particle Hamiltonian
is symmetricWe can also
under bothchoose (H, L2 , Land
translations z ) asrotations,
a completesoset of commuting
a complete set ofobservables.
commutingInobservables
this case, can
weinhave
be chosen moreeigenstates
than one way. If we choose (px , py , pz ) as the complete set, then the simultaneous
eigenstate is
ψnlm = R(r)Y lm (θ, φ),
ik·x (700)
ψ(x) = e , (23)

a plane wave with p = h̄k. The energy is given in terms of the momentum eigenvalues by

|p|2
E= . (24)
2m
98
The spectrum is continuous with E ≥ 0.
If we choose (H, L2 , Lz ) as the complete set, then we must solve the radial Schrödinger equation
where the radial wavefunction satisfies
 
~2 1 d 2 dR(r) ~2 l(l + 1)
− r + R(r) = ER(r), (701)
2m r2 dr dr 2mr2

If we define

~2 k 2
8
E =
Notes 16: Central Force Motion
(702)
2m

and the dimensionless

k by
variable ρ = kr, then the equation becomes
h̄2 k 2
E= , (25)
2m
   
and writing ρ = kr for a dimensionless radial variable, the radial Schrödinger equation becomes
1 d 2 dR l(l + 1)
ρ ! + # 1 −ℓ(ℓ + 1) $2
" R=0 (703)
ρ2 dρ ρ12 dρddρρ2 dRdρ
+ 1−
ρ 2
ρ R = 0. (26)

This is the standard differential equation for the spherical Bessel functions. These come in two
This is the types,
equation for the spherical Bessel functions j and nl (or yl )
denoted jℓ (ρ) and yℓ (ρ) (see Abramowitz and Stegun, Chapter l10).

jℓ (ρ)
yℓ (ρ)
1.0 y0 (ρ)
0.4 y1 (ρ)
j0 (ρ) y2 (ρ)
0.8
y3 (ρ)
0.2
0.6 j1 (ρ)

0.4 j2 (ρ) ρ
0.0
j3 (ρ)
0.2
-0.2
0.0 ρ

-0.2 -0.4

0 5 10 15 0 5 10 15
Fig. 3. Spherical Bessel functions jℓ (ρ) for different values Fig. 4. Spherical Bessel functions yℓ (ρ) for different val-
of ℓ. ues of ℓ.

The functions jℓ (r) are regular at the origin and so are physically acceptable solutions for the
nl diverge for r → 0Some
free particle. ∼ ρ−l−1
(nlexamples ), and
are plotted so 3,are
in Fig. not
which physically
clearly acceptable
shows the behavior solutions for the
R(r) ∼ rℓ near
r = 0. The functions yℓ (ρ) diverge at the origin and so are not acceptable free particle solutions if
free particle (but in general wavefunctions can behave like nl at large r, e.g., in scattering
the particle is able to reach the origin. They are plotted in Fig. 4. On the other hand, the y-type
problems). spherical Bessel functions are useful in scattering theory where at large distances from the origin the
particle becomes free, and it is desired to represent an arbitrary solution of the Schrödinger
l equation
jl are regular at region.
in that the origin, and
Since there is noare acceptable
attempt solutions
to extend those solutions all(note
the way the
downrto rbehavior
= 0, the at small r).
y-type solutions are acceptable (and necessary) in that case.

9.3 Two-body
7. Limitingcentral
Forms of the force
Spherical motion
Bessel Functions

Many problems involving special functions can be solved knowing only the limiting forms for
Consider two particles interacting with each other via a potential that depends only on the
large and small values of the arguments. In the case of the spherical Bessel functions, the limiting
distance between the particles V = V (|x1 − x2 |)

p21 p2
H= + 2 + V (|x1 − x2 |) (704)
2m1 2m2

99
 Again the system is invariant under both translations and rotations.
In configuration space, each momentum operator (1,2) acts on the coordinate of that
particle k:

∂
pk = −i~ = −i~∇k , k = 1, 2 (705)
∂xk

The translation operator T (a) acts on the wave function as

(T ψ)(x1 , x2 ) = ψ(x1 − a, x2 − a) (706)

So the generator of translations is the total momentum

P = p1 + p2 (707)

A translation doesn’t change the distance between the particles, so H is translation in-
variant

[H, P] = 0 (708)

It is also rotation invariant. The action of rotations is

(U (R)ψ)(x1 , x2 ) = ψ(R−1 x1 , R−1 x2 ) (709)

=⇒ L = L1 + L2 (710)

where L is the total (orbital) angular momentum of the two particles

Lk = xk × pk , k = 1, 2 (711)

The distance between particles is rotation invariant, and so

[H, L] = 0 (712)

It is useful to change coordinates

m1 x1 + m2 x2
R= (713)
M
r = x2 − x1 (714)

100
where M = m1 + m2 is the total mass of the system, R is the coordinate of the center of
mass of the system, and r is the relative position vector between the particles.
We can also define conjugate momenta

∂
P = −~ = p1 + p2 (715)
∂R
∂ m1 p2 − m2 p1
p = −~ = (716)
∂r M

The transformation (x1 , p1 ; x2 , p2 → (R, P; r, p) is a canonical transformation. I.e., it

preserves the canonical commutation relations.

[Ri , Pj ] = [ri , pj ] = i~δij (717)

[Ri , rj ] = [Ri , pj ] = [Pi , rj ] = [Pi , pj ] = 0 (718)

Note that any function of (R, P) commutes with any function of (r, p)
The Hamiltonian becomes

H = HCM + Hrel (719)

P2
HCM = (720)
2M
p2
Hrel = + V (r), (721)
2µ

with µ the reduced mass

1 1 1
= + , (722)
µ m1 m2

which is the harmonic mean of the two masses (if the mass difference is large, it is close to
the lighter mass, while if the masses are equal, µ = m/2).
HCM is a free particle Hamiltonian, which is the kinetic energy of the center of mass,
while Hrel is the kinetic energy about the center of mass plus the potential energy.
They commute

[HCM , Hrel ] = 0 (723)

because they are functions of different degrees of freedom, and they also commute with the
total Hamiltonian H.

101
 So they possess a simultaneous eigenbasis of the form

Ψ(R, r) = Φ(R)ψ(r) (724)

HCM Φ = ECM Φ (725)
Hrel ψ = Erel ψ (726)
E = ECM + Erel (727)

The center of mass motion is that of a free particle. E.g., with energy eigenfunctions

Φ(R) = eiP·R/~ , (728)

or equivalently Bessel functions and spherical harmonics (written in spherical coordinates).

The relative motion is that of a one-body problem of a particle with position r and mass
µ.
If we consider angular momentum, we find that the total angular momentum can be
written

L = LCM + Lrel , (729)

with

LCM = R × P (730)
Lrel = r × p (731)

We have

[HCM , LCM ] = 0 (732)

[Hrel , Lrel ] = 0 (733)

and so the total Hamiltonian H commutes not only with an overall rotation of the system,
but separately with rotations of the center of mass and with rotations about the center of
mass.

102
9.4 Hydrogen Atom (Single electron in Coulomb potential)
Consider a single electron moving in a Coulomb potential with charge Ze (i.e., an electron
moving around a nucleus with atomic number Z), and ignore spin degrees of freedom.

Ze2
V (x) = − (734)
r

Since it is spherically symmetric, energy eigenstates can be chosen simultaneously as

eigenstates of L2 and Lz

ψ(x) = R(r)Ylm (θ, φ) (735)

with the radial wavefunction satisfying

 
1 d 2 dR(r) 2m −2m
r − 2 U (r)R(r) = ER(r), (736)
r2 dr dr ~ ~2

where the effective potential U is

~2 l(l + 1)
U (r) = + V (r) (737)
2mr2
~2 l(l + 1) Ze2
= − (738)
2mr2 r

Recall that at small r, the radial wavefunction must behave as

R(r) ∼ rl (739)

At large r, bound states (E < 0) must decrease exponentially

d2 R(r) −2mE
2
= R(r) (740)
dr ~2
≡ κ2 R(r) (741)
=⇒ R(r) ∼ e−κr (742)

It is useful to separate the small- and large-radius behavior, and write.

R(r) = (κr)l e−κr w(κr) (743)

= ρl e−ρ w(ρ), (744)

103
where we intruduced the dimensionless radial variable ρ = κr
If we plug this into the radial Schrodinger equation, we get

d2 w dw
ρ 2
+ 2 (l + 1 − ρ) + [ρ0 − 2(l + 1)] w(ρ) (745)
dρ dρ

with
 1/2  1/2
V κZe2 2m Ze2 2mc2
ρ0 = ρ = = = Zα (746)
E E −E ~ −E

with α ∼ 1/137 the fine structure constant.

This is a specific case of Kummer’s equation

d2 F dF
x 2
+ (c − x) − aF = 0 (747)
dx dx

with

x = 2ρ (748)
c = 2(l + 1) (749)
2a = 2(l + 1) − ρ0 , (750)

whose solution is the Confluent Hypergeometric Function F (a; c; x).

So

ρ0
w(ρ) = F (l + 1 − ; 2(l + 1); 2ρ) (751)
2

The wavefunction is only normalizable if N = ρ0 /2 − (l + 1) is a non-negative integer

0, 1, 2, . . ..
We define the principle quantum number as
 1/2
ρ0 mc2
n= = Zα = 1, 2, 3 . . . (752)
2 −2E
Eg
=⇒ En = 2 (753)
n
1
Eg = − mc2 Z 2 α2 = −13.6Z 2 eV (754)
2

Given n, l has the range 0, 1, . . . , n − 1

104
 The degeneracy of each energy eigenstate is

X
n−1
degen(En ) = (2l + 1) = n2 (755)
l=0

Usually a degeneracy indicates the presence of a symmetry.

For the hydrogen atom, there is more than just the obvious SO(3) symmetry of rota-
tions, but a larger group SO(4) — the group defined by 4x4 orthogonal matrices with unit
determinant, or rotations in 4 dimensional space.
We can see this from the following.
In the classical problem of bodies in a 1/r potential, closed orbits are formed with a shape
and orientation that is described by the Lenz vector (Laplace-Runge-Lenz vector) M, which
is a constant of motion

p × L Ze2
M= − r (756)
m r

105
 Noting that for two Hermitian vector operators A and B,

(A × B)† = −B × A (757)

The corresponding Hermitian Lenz operator is

1 Ze2
M= (p × L − L × p) − r (758)
2m r

This operator commutes with the Hamiltonian

p2 Ze2
H= − (759)
2m r
[M, H] = 0, (760)

and so it is a constant of motion in the quantum system.

Other useful properties are

L·M=0=M·L (761)
2

M2 = H L2 + ~2 + Z 2 e4 (762)
m

To identify the symmetry responsible for this constant of motion, we need to know the
algebra of the generators of the symmetry.
We know that part of the symmetry is rotational symmetry, with angular momentum
generators satisfying the SO(3) algebra

[Li , Lj ] = i~ijk Lk (763)

We can calculate the other possible commutators

[Mi , Lj ] = i~ijk Mk (764)

2
[Mi , Mj ] = −i~ijk HLk (765)
m

So on the full Hilbert space, this is not a closed algebra. However, on an energy eigenspace
with energy E, we can replace H → E, and it becomes a closed algebra between the 6
generators (L, M).

106
 It simplifies if we rescale
 m 1/2
N≡ − M (766)
2E
=⇒ [Li , Lj ] = i~ijk Lk (767)
[Ni , Lj ] = i~ijk Nk (768)
[Ni , Nj ] = i~ijk Lk (769)

This is the algebra of SO(4), or rotations in 4 dimensions, of which actual rotations in

3D space is an SO(3) subgroup.

107
10 Addition of angular momenta
10.0.1 Tensor products

We can combine two Hilbert spaces with a tensor product.

For example, the Hilbert space for a multi-particle state is a tensor product of single-
particle spaces, or for a single particle including orbital and spin degrees of freedom.
Consider, for example, two spinless distinguishable particles, labeled 1 and 2, with ket
spaces X1 and X2 , each consisting of wavefunctions in 3D space:

X1 = {φ(x), particle1} (770)

X2 = {χ(x), particle2} (771)

In this case, the two Hilbert spaces are identical to each other, but we consider them two
distinct copies, with label 1 and 2.
The wavefunction for a two-particle system is another space, defined on a 6-dimensional
configuration space (x1 , x2 ):

X = {ψ(x1 , x2 )} (772)

A special case of a two-particle wave function is a product of single-particle wave functions

ψ(x1 , x2 ) = φ(x1 )χ(x2 ), (773)

but in general a wave function can not be written in this form.

However, one can build a basis of two-particle states from products of this form. Any
state can then be written as a linear combination of these states.
Consider a set of basis states for each single-particle state

X1 → {un (x)} (774)

X2 → {vn (x)} (775)

An arbitrary two-particle state can then be written as

X
ψ(x1 , x2 ) = cn,m un (x1 )vm (x2 ) (776)
n,m

We say that the full space X spanned by these product states is a tensor product of spaces

108
X1 and X2 ,

X = X 1 ⊗ X2 (777)

The tensor product of kets themselves is written

|un i ⊗ |vn i = |un i|vn i (778)

The dimension of a tensor product space is the product of the dimensions of each of the
products

dimX = (dimX1 )(dimX2 ) (779)

10.1 Spin + Space

A particle with spin can have a position in 3D as well as a spin orientation. Its Hilbert space
is a tensor product of the orbital and spin spaces

Xorb = span{|xi} (780)

Xspin = span{|mi} (781)

where x is a vector in 3D and m = −s, . . . , s.

The total Hilbert space is

Xorb ⊗ Xspin = span{|xi ⊗ |mi} (782)

= span{|x, mi} (783)

Since the states |x, mi span the space, any state can be written in that basis
XZ
|ψi = d3 x|x, mihx, m|ψi (784)
m
XZ
= d3 x|x, miψ(x, m) (785)
m

10.2 Operator tensor product

We can also define the tensor product of operators.
If A1 operators on space X1 , and A2 operates on X2 , then the tensor product A1 ⊗ A2

109
operates on tensor product states as

(A1 ⊗ A2 )(|αi1 ⊗ |βi2 ) = (A1 |αi1 ) ⊗ (A2 |βi2 ) (786)

Since any state can be written as a superposition of tensor product states, this completely
defines the tensor product A1 ⊗ A2 .
A simple case is when one operator is the identity

(A1 ⊗ 1)(|αi1 ⊗ |βi2 ) = (A1 |αi1 ) ⊗ |βi2 (787)

Normally, we just write A1 , and it is understood to mean A1 ⊗ 1 when it operates on a state

in space X1 ⊗ X2 .
Note that operators that act on disjoint subspaces always commute

[A1 ⊗ 1, 1 ⊗ A2 ] = [A1 , A2 ] = 0 (788)

10.3 Addition of angular momenta

Consider two Hilbert spaces X1 , X2 , which are acted on by angular momentum operators
J1 , J2 , respectively. (For example, orbital and spin angular momentum of a single particle
J1 = L, J2 = S.)
For each individual subspace, the space breaks up into a direct sum of irreducible sub-
spaces under rotations, each with some definite j.

X
X1 = ⊕X1j1 (789)
j1
X
X2 = ⊕X2j2 , (790)
j2

For the tensor product space X = X1 ⊗ X2 , we can define a total angular momentum
operator

J = J1 ⊗ 1 + 1 ⊗ J2 = J1 + J 2 (791)

Since operators on different spaces commute [J1 , J2 ] = 0, the total angular momentum

110
satisfies the expected commutation relation

[Ji , Jj ] = i~ijk Jk (792)

Therefore, we can decompose the full space X into a direct sum of invariant subspaces
with respect to the total angular momentum J.
The basic problem of addition of angular momenta is to find which values of j occur in
X and with what multiplicity, and to find some convenient way of constructing the standard
basis |jmi from states |j1 m1 ; j2 m2 i, with

J12 |j1 m1 ; j2 m2 i = ~2 j1 (j1 + 1)|j1 m1 ; j2 m2 i (793)

J1z |j1 m1 ; j2 m2 i = ~m1 |j1 m1 ; j2 m2 i (794)
J22 |j1 m1 ; j2 m2 i = ~2 j2 (j2 + 1)|j1 m1 ; j2 m2 i (795)
J2z |j1 m1 ; j2 m2 i = ~m2 |j1 m1 ; j2 m2 i (796)
J 2 |jmi = ~2 j(j + 1)|jmi (797)
Jz |jmi = ~m|jmi (798)

We only need to consider one irreducible subspace from each space X1 and X2 , since the
combined space is a tensor product space. That is

X
X = X1 ⊗ X2 = ⊕ (X1j1 ⊗ X2j2 ) (799)
j1 ,j2

So, given an irreducible subspace j1 and j2 , we form a (2j1 + 1)(2j2 + 1)-dimensional

product space spanned by |j1 m1 ; j2 m2 i with all possible values of {m1 , m2 }, and we would
like to decompose that space into a direct some of irreducible subspaces of total angular
momentum.

10.4 Simple example: 2 spin-1/2 particles

Consider the tensor product of 2 spin-1/2 Hilbert spaces. It is a 4-dimensional space spanned
by |m1 i ⊗ |m2 i. That is

| + +i, | + −i, | − +i, | − −i (800)

with S1z | + −i = ~m1 | + −i = +~/2| + −i, etc.

111
 These are already eigenstates of Sz = Sz1 + Sz2 :

Sz |m1 m2 i = ~(m1 + m2 )|m1 m2 i = ~m|m1 m2 i (801)

So m = m1 + m2 runs from −1, 0, 1. We can already see that the maximum total spin
will be 1, and the 4 basis states will be

|s, mi = |1, 1i = | + +i (802)

|1, −1i = | − −i (803)
|1, 0i = a| + −i + b| − +i (804)
|0, 0i = a0 | + −i + b0 | − +i (805)

We can find the correct m = 0 states by using ladder operators on |1, 1i or |1, −1i.

S− |1, 1i = (S1− + S2− )| + +i (806)

s   s  
p 1 1 1 1 1 1 1 1
= (1 + 1)(1 − 1 + 1)|1, 0i = + − + 1 | − +i + + − + 1 | + −i
2 2 2 2 2 2 2 2
(807)
1
=⇒ |1, 0i = √ (| + −i + | − +i) (808)
2

The |0, 0i state must be orthogonal to this one. Up to an arbitrary phase, this gives

1
|0, 0i = √ (| + −i − | − +i), (809)
2

Later we will choose a convention for the overall phase.

10.5 Allowed values of j and m

As before, the allowed values of m are easy to determine:

Jz |j1 m1 ; j2 m2 i = (J1z + J2z )|j1 m1 ; j2 m2 i = ~(m1 + m2 )|j1 m1 ; j2 m2 i (810)

So for known values of j1 and j2 the only possible values of m are all possible values of
m1 + m2 . That is

112
 m ∈ {−(j1 + j2 ), −(j1 + j2 ) + 1, −(j1 + j2 ) + 2, . . . (j1 + j2 )} (811)
Notes 18: Coupling Ket Spaces and Angular Momenta 11

Take, for example, j1 = 5/2 and j2 = 1. The total dimensionality is 6 × 3 = 18. The
degeneracy of each m is as follows:
m2

− 12 1
2
3
2
5
2
7
2

− 32

− 52
m1
− 72

Fig. 1. Each dot in the rectangular array stands for one vector of the uncoupled or tensor product basis, |j1 j2 m1 m2 ⟩ =
|j1 m1 ⟩|j2 m2 ⟩. The dashed lines are contours of m = m1 + m2 .

Following the logic from the previous section, there must be a j = 7/2, m = 7/2 state,
|m1 = 5/2, m2 = 1i, and the other 8 j = 7/2 states, related by the lowering operator. There’s
a second linearly independent m = 5/2 state that must belong to a j = 5/2 irreducible
subspace,
These and
12 is orthogonal
eigenvaluesNotes 18:toin
of Jz are the j =degenerate.
Coupling
general 7/2,Spaces
Ket m = 5/2 and
To state.theOnce
Angular
follow this isargument,
Momenta
subsequent found, the other
it helps
5
5toj have
= 5/2an states are Let
example. related by the
us take the case
lowering
j1 = operator. After this, there is a third m = 3/2
2 and j2 = 1, so that 2j1 + 1 = dim E1 = 6 and
2j2 +that
state 1 = dim
must E2 be
= 3.
theThus, thestate
stretch dimensionality of Etriplet.
of a j = 3/2 = E1 ⊗ SoE2 is 6 × 3 we
overall = 18. It is convenient to
have:
make a plot in the m1 -m2 plane of the basis vectors7 of the tensor
5 product
3 basis, by placing a dot at
m g(m) j= 2 j= 2 j= 2
each allowed m1 and m2 value. We 7 then obtain a rectangular array of dots, as illustrated in Fig. 1.
2 1 1
As illustrated in the figure, lines of
5 constant m = m1 + m2 are straight lines, dashed in the figure,
2 2 1 1
sloping downwards. The number 3of dots each dashed line passes through is the number of kets of
2 3 1 1 1
the uncoupled basis with a given1 m value; we see that the degeneracies of the different m values
2 3 1 1 1
range from one to three, as summarized
1 in Table 1.
−2 3 1 1 1
− 32 3 1 1 1
In particular, the stretched state
5 (the one for which (m1 , m2 ) = (j1 , j2 ), in the upper right hand
−2 2 1 1
corner in the figure) is a nondegenerate
7 eigenstate of Jz with quantum number m = 27 . But since
−2 1 1
2 2
[J , Jz ] = 0, this state is also an eigenstate of J , in accordance with Theorem 1.5. But what is
Total 18 8 6 4
the eigenvalue of J 2 , that is, what is the quantum number j? Certainly we cannot have j < 72 ,
Table 1. The first column contains m, the quantum number of Jz ; the second column contains
7 g(m), the degeneracy of
because then 3,this
m; columns would
4 and violate
5 contain a unit the
for each m ≤|jm⟩
rulevector j. of
Northe can we or
standard have j >basis
coupled 2 , because,
with given jfor
andexample,
m values. if
the stretched
Now letstate
us were |jm⟩
consider | 92 72 ⟩, then weeigenspace
the= 2-dimensional could apply
of the raising operator
Jz corresponding J+ and obtain
to quantum numberthe
9 95 9
m |=
state 2 22⟩. But space
. This thereisisspanned
no statebywith m =of 2the
the kets , asuncoupled
we see from
113 basisthe figure or table.
corresponding to (m1Therefore
, m2 ) = ( 52 ,the
0) j
and ( 32number
quantum of the stretched
, 1). Furthermore, | 27 25 ⟩ of
the ket state mustthe be j = 72basis
coupled ; thisalso
state
liesisinthe
thisstate |jm⟩
space.
77
Let=us| 2consider
2 ⟩ of the
75
coupled basis.
the ket, call But
it |x⟩,given
that this state, wetocan
is orthogonal | 2 2apply lowering
⟩ in this space. operators
Certainly |x⟩ J− istoanobtain all of
eigenket eight states
Jz with
5 2
| 72 m⟩,
eigenvalue 2 . indicated
which are And it is in
also
theanthird
eigenket of Jof
column , the
a fact that is left as an exercise (the logic is an
table.
5
 This can be summarized by the notation

5 3 5 7
⊗1= ⊕ ⊕ , (812)
2 2 2 2

which corresponds to the dimensionality count

6×3=4+6+8 (813)

In the general case, we have

j1 ⊗ j2 = |j1 − j2 | ⊕ |j2 − j2 | + 1 ⊕ . . . ⊕ j1 + j2 (814)

with consistent dimensionalities

j1 +j2
X
2j + 1 = (2j1 + 1)(2j2 + 1) (815)
j=|j1 −j2 |

10.6 Clebsch-Gordon Coefficients

The two bases are connected by a unitary matrix. Explicitly:

j1 j2
X X
|jmi = |j1 m1 ; j2 m2 ihj1 m1 ; j2 m2 |jmi (816)
m1 =−j1 m2 =−j2
j1 +j2 j
X X
|j1 m1 ; j2 m2 i = |jmihjm|j1 m1 ; j2 m2 i (817)
j=|j1 −j2 | m=−j

The expansion coefficients hj1 m1 ; j2 m2 |jmi are called “Clebsch-Gordon” coefficients.

They can be calculated as above, starting with the stretch state of largest j and lowering
with J− , then taking the stretch state of the second-largest j as orthogonal to the previous
state of the same m, and lowering, etc.
The relative phases for each fixed j and different m can be chosen as discussed in previous
lectures. The phase of the stretch state |jji is chosen so that

hjj|j1 j1 ; j2 , j − j2 i > 0 (818)

With these phase conventions, the Clebsch-Gordon coefficients are real

hjm|j1 m1 ; j2 m2 i = hj1 m1 ; j2 m2 |jmi (819)

114
 CG coefficients have the following useful properties.
Since they form a unitary matrix,

X
hjm|j 0 m0 i = hjm|j1 m1 ; j2 m2 ihj1 m1 ; j2 m2 |j 0 m0 i = δjj 0 δmm0 (820)
m1 m2
X
hj1 m1 ; j2 m2 |j1 m01 ; j2 m02 i = hj1 m1 ; j2 m2 |jmihjm|j1 m01 ; j2 m02 i = δm1 m01 δm2 m02 (821)
jm

There is the selection rule

hjm|j1 m1 ; j2 m2 i = 0 unless m = m1 + m2 (822)

They satisfy recursion relations. We derive one by applying J− = J1− + J2− to |jmi and
its resolution in terms of |j1 m1 ; j2 m2 i

1 p
J− |jmi = (j + m)(j − m + 1)|j, m − 1i (823)
~
X p
= (j1 + m1 )(j1 − m1 + 1)|j1 , m1 − 1; j2 , m2 i (824)
m1 m2

p
+ (j2 + m2 )(j2 − m2 + 1)|j1 , m1 ; j2 , m2 − 1i hj1 m1 ; j2 m2 |jmi (825)
X p
= (j1 + m1 + 1)(j1 − m1 )hj1 , m1 + 1; j2 m2 |jmi (826)
m1 m2

p
+ (j2 + m2 + 1)(j2 − m2 )hj1 m1 ; j2 , m2 + 1|jmi × |j1 m1 ; j2 m2 i (827)

Applying hj1 m01 ; j2 m02 | on the left and rearranging indices


1
hj1 m1 ; j2 m2 |j, m − 1i = p (828)
(j + m)(j − m + 1)
p
(j1 + m1 + 1)(j1 − m1 )hj1 , m1 + 1; j2 m2 |jmi (829)

p
+ (j2 + m2 + 1)(j2 − m2 )hj1 , m1 ; j2 m2 + 1|jmi (830)

115
 Repeating for J+ gives the general recursion relation

1
hj1 m1 ; j2 m2 |j, m ± 1i = p (831)
(j ∓ m)(j ± m + 1)
p
(j1 ∓ m1 + 1)(j1 ± m1 )hj1 , m1 ∓ 1; j2 m2 |jmi (832)

p
+ (j2 ∓ m2 + 1)(j2 ± m2 )hj1 , m1 ; j2 m2 ∓ 1|jmi (833)

Other useful identities are

hj1 m1 ; j2 m2 |jmi = (−1)j1 +j2 −j hj2 m2 ; j1 m1 |jmi, (834)

s
2j + 1
= (−1)j1 −j+m2 hjm; j2 , −m2 |j1 m1 i, (835)
2j1 + 1
s
2j + 1
= (−1)j2 −j+m1 hj1 , −m1 ; jm|j2 m2 i, (836)
2j2 + 1
= (−1)j1 +j2 −j hj1 , −m1 ; j2 , −m2 |j, −mi (837)

10.7 Irreducible tensor operators and the Wigner-Eckart Theorem

We can analyze operators based on how they transform under rotations, which gives valuable
information about matrix elements.
We define a rotation of an operator as follows:
A ket rotates as

|ψi → |ψ 0 i = U |ψi (838)

We define a rotated operator A → A0 such that it has the same matrix elements in the
rotated ket space. I.e.,

hψ 0 |A0 |φ0 i = hψ|A|φi (839)

= hψ|U −1 A0 U |φi (840)
=⇒ A = U −1 A0 U (841)
A0 = U AU −1 (842)

Note that A → A for a rotation of 2π, even in a half-integer representation.

116
 U −1 (θ = 2π)AU (θ = 2π) = ±1A(±1) = A (843)

10.7.1 Scalar operators

We can classify operators by how they transform under rotations.

E.g., a scalar operator K has the property K → K:

U KU † = K (844)

for all rotations.

=⇒ [U, K] = 0 (845)
=⇒ [J, K] = 0 (846)

10.7.2 Vector operators

A vector operator V is a set of (3) operators whose matrix elements transform as components
of a vector.

hψ 0 |V|ψ 0 i = Rhψ|V|ψi (847)

X
hψ 0 |Vi |ψi = hψ|Vi |ψi (848)
j
X
=⇒ U † Vi U = Rij Vj (849)
j

If we take instead take the inverse of this rotation (R → R−1 , so U → U † ), this gives the
transformation property of the operator itself:
X X
−1
U Vi U † = Vj Rij = Vj Rji (850)
j j

As in the scalar case, this implies a commutation relation with angular momentum. Con-
sider an infinitesimal rotation:

117
 i
U = 1 − θn̂ · J (851)
~
R = 1 + θn̂ · J (852)

Note that J operates on a Hilbert space, which can be any dimension 2j + 1, while J is a
3x3 matrix with components

(Ji )jk = −ijk (853)

A vector operator then satisfies

   
i i
1 − θn̂ · J V 1 + θn̂ · J = (1 − θn̂ · J) V (854)
~ ~
i
=⇒ − θ[n̂ · J, Vk ] = θni ijk Vj (855)
~
ni [Ji , Vj ] = i~ijk ni Vk (856)

Since this is true for any unit vector n̂ (in particular, n̂ = x̂, ŷ, and ẑ), we must have

[Ji , Vj ] = i~ijk Vk (857)

Any operator that satisfies this relation is a vector operator.

Note that angular momentum itself J is a vector operator.
We can also see that operators such as X and p are also vector operators, as expected.
To see this, consider the angular momentum operator in configuration space

J =L=x×p (858)

We can check that

[Li , xj ] = ikl xk [pl , xj ] = −i~ikj xk = i~ijk xk (859)

[Li , pj ] = i~ijk pk (860)

So X and p are indeed vector operators.

Combining these results, one can show that the scalar product of two vector operators

118
V · W is a scalar operator, while the cross product V × W is a vector operator.
For example, the Hamiltonian of a particle in a central potential is a scalar operator
because it is a function of the dot products p · p, x · x.

10.7.3 Tensor operators

We can continue this process and define general tensor operators, so that scalar operators
have rank 0 and vector operators have rank 1.
A rank 2 tensor operator Tij is a tensor of (3 × 3 = 9) operators with the following
transformation properties
X
U Tij U † = Tkl Rki Rlj (861)
kl

And similarly for tensors of higher rank — each index is contracted with a rotation matrix
−1
R .

X
U Tijk... U † = TU Ti0 j 0 k0 ... Ri0 i Rj 0 j Rk0 k . . . (862)
i0 j 0 k0 ...

10.7.4 Irreducible tensor operators

A simple example of a tensor operator can be constructed from vector operators

Tij = Vi Wj (863)

This (Cartesian) tensor, however, is reducible — it can be decomposed into objects that
transform among themselves under rotations

 
V·W Vi Wj − Vj Wi Vi Wj + Vj Wi V · W
Vi Wj = δij + + − δij (864)
3 2 2 3

The first term is a scalar, the second can be written as a vector product ijk (V × W )k ,
and the third contains the other 5 of the 9 components:

3×3=1+3+5 (865)

119
 Note that this count of degrees of freedom is the same as the angular momentum decom-
position into irreducible representations.

1⊗1=0⊕1⊕2 (866)

We will see that each term represents an irreducible tensor operator.

Is is useful to write such irreducible tensor operators into a spherical basis, similar to how
we use |j, mi as basis kets.
Before giving a general definition of spherical tensors, let’s look at an example.
If we write Ylm (θ, φ) as the function of a vector Ylm (n̂), we can define a spherical tensor
of rank k constructed from a vector operator V :

m=q
Tq(k) = Yl=k (V) (867)

For k = 1 we have
r r r r
2 2 z 2 (1) 2
Y10 = cos θ = = n̂z → T0 = Vz (868)
4π 4π r 4π 4π
r r  
2 x ± iy (1) 2 Vx ± iVy
Y1±1 =∓ √ → T±1 = ∓ √ (869)
4π 2r 4π 2

And k = 2, for example

r r
15 (x ± iy)2 (2) 15
Y2±2 = 2
→ T±2 = (Vx ± iVy )2 (870)
32π r 32π
(k)
These Tq are irreducible in the same way as the Ylm — operators themselves form a linear
vector space, which are transformed under rotations as defined above. Spherical tensors of
(k)
rank k (but different q) Tq form an invariant subspace under rotations.
They are defined by their rotation properties, so let’s recall how the spherical harmonics
rotate.
A direction eigenket transforms as

|n̂i → |n̂0 i = U (R)|n̂i (871)

120
 So we have

Ylm (n̂) = hn̂|l, mi (872)

X (j)
hn̂|U (R−1 )|l, mi = hn̂|l, m0 iDm0 m (R−1 ) (873)
m0
X 0 (j)
=⇒ Ylm (n̂0 ) = Ylm (n̂)Dm0 m (R−1 ) (874)
m0

(k) m=q
It is reasonable to expect, then, that the operators Tq = Yl=k (V) transform as

0
X
m
(j)∗
† m
U Yl (V)U = (V)Dmm0 (875)
l

or

X
k
(j)∗
†
U Tq(k) U = Dqq0 Tq(k) (876)
q 0 =−k

X
k
(j)
=⇒ U Tq(k) U † = Dq0 q Tq(k) (877)
q 0 =−k

which we take as the definition of an irreducible spherical tensor operator of rank k.

This is the generalization of the scalar and vector operators, which satisfy this equation
with k = 0 and 1, respectively.
Alternatively, we can again write the definition in terms of the commutator with angular
momentum by considering the infinitesimal version

    X k  
iJ · n̂θ iJ · n̂θ 0 iJ · n̂θ
1+ Tq(k) 1− = (k)
Tq hkq | 1 + |kqi (878)
~ ~ q 0 =−k
~

or

X (k)
[J · n̂, Tq(k) ] = Tq0 hkq 0 |J · n̂|kqi (879)
q0

If we take n̂ to be in the direction of ẑ, or in the (complex) directions (x̂ ± iŷ), then J · n̂
becomes Jz and J± , respectively, and we can write

121
 X (k)
[Jz , Tq(k) ] = Tq0 hkq 0 |Jz |kqi (880)
q0

= ~qTq(k) (881)
X (k)
[J± , Tq(k) ] = Tq0 hkq 0 |J± |kqi (882)
q0
p (k)
= ~ (k ∓ q)(k ± q + 1)Tq±1 (883)

10.7.5 Products of tensors

Products of irreducible tensors decompose into irreducible tensors in the same way as tensor
product kets |j1 m1 i ⊗ |j2 m2 i decompose into basis kets |jmi:

X
Xqk11 Yqk22 = Tqk hk1 q1 ; k2 q2 |kqi (884)
kq

or, equivalently
X
Tqk = Xqk11 Yqk22 hk1 q1 ; k2 q2 |kqi (885)
q1 ,q2

To prove this, we need to show that the expression on the right is indeed an irreducible
tensor operator:

X
U Tqk U † = U Xqk11 U † U Yqk11 U † hk1 q1 ; k2 q2 |kqi (886)
q1 ,q2
X
= Xqk01 Yqk0 1 Dqk01q1 Dqk02q2 hk1 q1 ; k2 q2 |kqi (887)
1 1 1 2
q1 ,q2 ,q10 ,q10

And we can replace the D matrices by following identity:

U |j1 m1 ; j2 m2 i = U1 |j1 m1 i ⊗ U2 |j2 m2 i (888)

XX j
X j1 j2
= |jm0 iDm 0 m hjm|j1 m1 ; j2 m2 i = |j1 m01 ; j1 m02 iDm 0 m Dm0 m
1 2
(889)
1 2
jm m0 m01 m02

122
 If we multiply by on the left by hj1 m001 ; j2 m002 |, we get

j1 j2
X j
Dm 00 m Dm00 m =
1 2
hj1 m001 ; j2 m002 |jmiDmm 0
0 hjm |j1 m1 ; j2 m2 i (890)
1 2
jmm0

Inserting this into the expression for the rotated spherical tensor gives
X
U Tqk U † = Xqk01 Yqk0 2 hk1 q10 ; k2 q20 |KQ0 iDQ
K
0 Q hKQ|k1 q1 ; k2 q2 ihk1 q1 ; k2 q2 |kqi (891)
1 2
q1 ,q2 ,q10 ,q20 ,K,Q,Q0

Then we can do the sums over q10 and q20 by using Eq. (885)
X
Xqk01 Yqk0 1 hk1 q10 ; k2 q20 |KQ0 i = TQK (892)
1 1
q10 q20

The sums over q1 and q2 can be done by noting the orthogonality of the Clebsch-Gordan
coefficients

X
hKQ|k1 q1 ; k2 q2 ihk1 q1 ; k2 q2 |kqi = δKk δQq (893)
q1 ,q2

So finally we have
X X
U Tqk U † = TQK0 DQ
K
0 Q δKk δQq = Tqk0 Dqk0 q , (894)
KQQ0 q0

which is the defining relation for an irreducible tensor operator.

This indeed gives us the decomposition of a rank 2 tensor from above in terms of a scalar
(k = 0), a vector (k = 1), and a symmetric, traceless tensor (k = 2), according to

X
Tqk = Vq1 Wq2 h1q1 ; 1q2 |kqi (895)
q1 q2

10.7.6 Three Ylm Formula

A useful identity for the homework is the three Ylm formula, which we can derive from the
above relation

123
 j1 j2
X j 0 0 0
Dm 0 m Dm0 m =
1 2
hj1 m1 ; j2 m2 |jmiDmm 0 hjm |j1 m1 ; j2 m2 i (896)
1 2
jmm0

along with Eq. (674):

r
2l + 1 l∗
Ylm (θ, φ) = Dm0 (φ, θ, 0) (897)
4π

which gives
X
l1 l2 l 0
Dm 10
Dm 20
= hl1 0; l2 0|lmiDmm 0 hlm |l1 m1 m2 i (898)
lmm0
r r r
4π 4π X 4π
= Yl1 m1 Yl2 m2 = hl1 0; l2 0|l0i Ylm0 hlm0 |l1 m1 m2 i (899)
2l1 + 1 2l2 + 1 lm0
2l + 1

and finally,
s
X (2l1 + 1)(2l2 + 1)
Yl1 m1 Yl2 m2 = Ylm hl0|l1 0; l2 0ihl1 m1 ; l2 m2 |lmi (900)
lm
4π(2l + 1)

If we multiply by Yl3∗m3 and integrate over angle, we get the useful formula

Z s Z
X (2l1 + 1)(2l2 + 1)
dΩYl3∗m3 Yl1 m1 Yl2 m2 = hl0|l1 0; l2 0ihl1 m1 ; l2 m2 |lmi dΩYl3∗m3 Ylm
lm
4π(2l + 1)
(901)
s
X (2l1 + 1)(2l2 + 1)
= hl0|l1 0; l2 0ihl1 m1 ; l2 m2 |lmiδl3 l δm3 m (902)
lm
4π(2l + 1)
s
(2l1 + 1)(2l2 + 1)
= hl3 0|l1 0; l2 0ihl1 m1 ; l2 m2 |l3 m3 i (903)
4π(2l3 + 1)

10.7.7 Matrix Elements of Tensor Operators and the Wigner-Eckart theorem

Now we consider matrix elements of tensor operators with respect to kets in the standard
angular momentum basis. We can derive a number of results and selections rules based only
on general principles.

124
 First, there is a simple m-selection rule

hα0 , j 0 m0 |Tq(k) |α, jmi = 0, unless m0 = q + m (904)

which can be proved using the angular momentum commutation relation for a spherical
(k)
tensor Tq

[Jz , Tq(k) ] = ~qTq(k) (905)

=⇒ 0 = hα0 , j 0 m0 |[Jz , Tq(k) ] − ~qTq(k) |α, jmi (906)
= [(m0 − m)~ − ~q] hα0 , j 0 m0 |Tq(k) |α, jmi (907)

So the matrix element can only be non zero if the first factor is zero, or m0 = q + m.
The Wigner-Eckart Theorem is

hα0 j 0 m0 |Tq(k) |αjmi = hα0 j 0 ||T (k) ||αjihj 0 m0 |jm; kqi (908)

The double bar matrix element is called the reduced matrix element and is defined by this
equation. The non-trivial aspect of the theorem is that the reduced matrix element defined
this way is independent of m0 , m0 , and q. The dependence on those quantities is contained
entirely in the Clebsch-Gordan coefficient (which is independent of α and the specific operator
in question, other than the value of k and q.
Note that some authors (like Sakurai) define the reduced matrix element instead with

hα0 j 0 ||T (k) ||αji

hα0 j 0 ||T (k) ||αji → √ (909)
2j + 1

The selection rules are now apparent from the Clebsch-Gordan coefficient. Besides the
restriction of m0 = q + m, we must have |j − k| ≤ j 0 ≤ j + k

10.7.8 Proof of Wigner-Eckart Theorem

Using a previous result

p (k)
[J± , Tq(k) ] = ~ (k ∓ q)(k ± q + 1)Tq±1 (910)
p (k)
=⇒ hα0 j 0 m0 |[J± , Tq(k) ]|αjmi = ~ (k ∓ q)(k ± q + 1)hα0 j 0 m0 |Tq±1 |αjmi (911)

125
 Acting to the left and right with J± :
p p
(j 0 ± m0 )(j 0 ∓ m0 + 1)hα0 j 0 , m0 ∓ 1|Tq(k) |αjmi = (j ∓ m)(j ± m + 1)hα0 j 0 m0 |Tq(k) |αj, m ± 1i
p (k)
+ (k ∓ q)(k ± q + 1)hα0 j 0 m0 |Tq±1 |αjmi
(912)

Compare to the recursion relation for Clebsch-Gordon coefficients

p p
(j ± m)(j ∓ m + 1)hj1 m1 ; j2 m2 |j, m ∓ 1i = (j1 ∓ m1 )(j1 ± m1 + 1)hj1 , m1 ± 1; j2 m2 |jmi
p
+ (j2 ∓ m2 )(j2 ± m2 + 1)hj1 , m1 ; j2 m2 ± 1|jmi
(913)

If we associate j 0 → j, m0 → m, j → j1 , m → m1 , k → j2 , and q → m2 , then we get two

first-order linear homogeneous equations with the same coefficients.
I.e., the two relations can be written into the form
X X
aij xj = 0, aij yj = 0 (914)
j j

with the same aij in both equations, and where the indices run over all values of m0 , m, and
q.
Therefore the solutions must be proportional, xj = cyj , for some constant c.

(k)
=⇒ hα0 j 0 m0 |Tq±1 |αjmi = Chj, m; kq ± 1|j 0 m0 i, (915)

where the proportionality constant is independent of m0 , m, and q, but can depend on the
other parameters α, α0 , k, j 0 , j:

C ≡ hα0 j 0 ||T (k) ||αji (916)

10.7.9 Examples of Wigner-Eckart and the projection theorem

Two simple examples are scalar and vector operators, k = 0, 1.

(0)
For a scalar operator, T0 = S:

hα0 j 0 m0 |S|αjmi = hα0 j 0 ||S||αjihj 0 m0 |jm; 00i = δjj 0 δmm0 hα0 j 0 ||S||αji (917)

126
 Acting with a scalar operator is like adding an angular momentum of zero.
For a vector operator, it’s like adding angular momentum 1:

hα0 j 0 m0 |Vq |αjmi = hα0 j 0 ||V ||αjihj 0 m0 |jm; 1qi (918)

So the selection rules are


±1
∆m ≡ m0 − m = q = ±1, 0; ∆j = j 0 − j = , (919)
0

and the j = 0 → j 0 = 0 transition is forbidden.

For a vector operator and j = j 0 , the Wigner-Eckart theorem takes a simple form, known
as the projection theorm

0 0 hα0 jm|J · V |αjmi

hα jm |Vq |αjmi = 2
hjm0 |Jq |jmi, (920)
~ j(j + 1)

where we define Jq in the spherical basis from the Cartesian components in the usual way

1
J0 = Jz , J±1 = ∓ √ (Jx ± iJy ), (921)
2

and the same for Vq

Proof:

hα0 jm|J · V |αjmi = hα0 jm|(J0 V0 − J+1 V−1 − J−1 V+1 )|αjmi (922)
~ p
= m~hα0 jm|V0 |αjmi + √ (j + m)(j − m + 1)hαjm − 1|V−1 |αjmi
2
~ p
−√ (j − m)(j + m + 1)hα0 jm + 1|V+1 |αjmi (923)
2
= cjm hα0 j||V ||αji (924)

The last line is given by the Wigner-Eckart theorem, since all the matrix elements are
proportional to the same reduced matrix element, and so cjm is independent of α, α0 , and V .
In addition, since J · V is a scalar operator, the coefficient is also independent of m:

cjm = cj (925)

127
 Since cj does not depende on V , we can let V → J and α0 → α, and the expression still
holds:

hαjm|J 2 |αjmi = cj hαj||J||αji (926)

We can relate the reduced matrix elements of V and J via the Wigner-Eckart theorem

hα0 jm0 |Vq |αjmi hα0 j||V ||αji

= (927)
hαjm0 |Jq |αjmi hαj||J||αji
hα0 jm|J · V |αjmi
= (928)
hαjm|J 2 |αjmi
hα0 jm|J · V |αjmi
= , (929)
~2 j(j + 1)

which proves the projection theorem.

128
11 Approximation Methods
Quantum problems can rarely be solved analytically, and so approximation methods are very
important.

11.1 Time-independent perturbation theory (nondegenerate)

We start with a time-independent Hamiltonian that can be written as

H = H0 + λH1 (930)

We assume that H0 is a solvable Hamiltonian with a known solution,

H0 |n(0) i = En(0) |n(0) i, (931)

and that the rest of the Hamiltonian λH1 has only a small effect on these energies and
eigenstates.
Specifically, we can consider the limit λ → 0, and write the solutions as a power series in
λ:

H|ni = (H0 + λH1 )|ni = En |ni (932)

with

En = En(0) + λ∆(1) 2 (2)

n + λ ∆n + . . . (933)
= En(0) + ∆n (934)
|ni = |n(0) i + λ|n(1) i + λ2 |n(2) i + . . . (935)

For now, we assume that a particular energy eigenstate |n(0) i is non-degenerate. We will
treat the case of degeneracy next.
If the series converges, we can achieve any desired accuracy by truncating at some fixed
power of λ. In practice, even if the series doesn’t converge, a very accurate result can often be
obtained at a finite order in λ. Some problems, on the other hand, cannot be approximated
this way, and are instead “non-perturbative”.
Note on convergence:
We can see some examples of this from functional analysis. An exponential can be ap-

129
proximated by the Taylor series around x = 0
∞
X
x xn
e = . (936)
n=0
n!

This series converges for all x. Adding more terms always improves the approximations.
Contrast that with the series

X ∞
1
= (−1)n x2n (937)
1 + x2 n=0

From complex analysis, we know that the presence of singularities at x = ±i means the series
only converges for |x| < 1 (even if we only consider real values of x).
Other functions have zero radius of convergence, which means the series never converges.
However, even in this case, the series can be asymptotic, such that it converges toward the
true value for a finite number of terms before diverging away from the true result.
For example, the Gamma function has an asymptotic series around its inverse argument
1/x = 0, or x → ∞:
 
−x x−1/2
√ 1 1 139 571
Γ(x) ∼ e x 2π 1 + + − − + ... (938)
12x 288x2 51840x3 2488320x4

Adding more terms to the series increases the accuracy at first (e.g., up to 14 terms for x = 2
or 20 terms for x = 3), but then begins to diverge.
Other functions cannot be represented by a series at all, and are instead non-perturbative.
For example, calculating the Taylor series of this function around x = 0 gives:

2
e−1/x = 0 + 0 + 0 + . . . (939)

We can rearrange Eq. (932) to obtain

(En0 − H0 )|ni = (λH1 − ∆n )|ni (940)

with ∆n the correction to the energy, which vanishes in the limit λ → 0

∆n ≡ En − En(0) (941)

We would like to solve this equation for |ni. This would require inverting the operator

130
(En0 − H0 ), and acting on both sides of the equation with (En0 − H0 )−1

|ni = (En0 − H0 )−1 (λH1 − ∆n )|ni + |ψ 0 i. (942)

However, we must be careful, since (En0 − H0 )−1 is not well defined on the entire Hilbert
space. Specifically, if it acts on a state with unperturbed energy En0 , or |n(0) i, it is ill-defined.
Fortunately, the right-hand side is orthogonal to |n(0) i:

hn(0) |(λH1 − ∆n )|ni = hn(0) |(H − H0 )|ni − hn(0) |(En − En(0) )|ni (943)
= hn(0) |ni(En − En(0) − En + En(0) ) (944)
=0 (945)

We instead define an operator that is the inverse (En0 − H0 )−1 for any state in this
orthogonal subspace, but annihilates |n(0) i.
First, define the complementary projection operator
X
φn = 1 − |n(0) ihn(0) | = |k (0) ihk (0) | (946)
k6=n

This projects a state into the component orthogonal to |n(0) i.

Therefore, the operator (En0 − H0 )−1 is always well-defined if it follows the projector φn .

1 X 1
φn = |k (0) ihk (0) | (947)
En0 − H0 k6=n
E 0 −H
n 0

Since the right-hand-side of Eq. (940) is already orthogonal to |n(0) i, we have

(λH1 − ∆n )|ni = φn (λH1 − ∆n )|ni (948)

So by inverting Eq. (940), we obtain an expression for energy eigenstate |ni

1
|ni = φn (λH1 − ∆n )|ni + |Ci (949)
En0 − H0

Where C is a possible solution of the homogeneous equation (like a constant of integration

when inverting a differential operator)

(En0 − H0 )|Ci = 0, (950)

131
which is set by the boundary condition

lim |ni = |n(0) i (951)

λ→0

1
=⇒ |ni = φn (λH1 − ∆n )|ni + cn (λ)|n(0) i (952)
En0 − H0

with

cn (λ) = hn(0) |ni (953)

lim cn (λ) = 1 (954)
λ→0

It is convenient to normalize |ni such that cn (λ) = hn(0) |ni = 1, instead of the usual
normalization. We can always normalize it at the end of the calculation, if we want a
properly normalized state.
So finally we have

φn
|ni = |n(0) i + (λH1 − ∆n )|ni, (955)
En0 − H0

and note that,

hn(0) |∆n |ni = ∆n = λhn(0) |H1 |ni (956)

The unknown energy and state appear on both sides of the equation, so we haven’t solved
the problem. However, in this form it is straightforward to generate a series of the form (933)
and (935), by iteratively substituting these expressions on the right side of each equation:

∆(1) (0) (0)

n = hn |H1 |n i (957)
∆(2) (0) (1)
n = hn |H1 |n i (958)
..
. (959)
∆(N
n
)
= hn(0) |H1 |n(N −1) i (960)
..
. (961)

Each order of energy shift requires knowledge of the energy eigenstate at the previous

132
order.
The corrections to the energy eigenstates can be derived by expanding in λ and matching
coefficients:

|ni = |n(0) i + λ|n(1) i + λ2 |n(2) i + . . . (962)

φn
= |n(0) i + 0 (λH1 − ∆n )|ni (963)
En − H0
φn

= |n(0) i + 0 (λH1 − λ∆(1) 2 (2) (0) (1) (2)
n − λ ∆n − . . .) |n i + λ|n i + λ|n i + . . . (964)
En − H0

Equating the first order coefficients gives:

φn φn
|n(1) i = (H1 − ∆(1) (0)
n )|n i = H1 |n(0) i, (965)
En0 − H0 En − H0
0

which then allows use to plug this in to the expression for the second order energy shift

φn
∆(2) (0)
n = hn |H1 H1 |n(0) i (966)
En0 − H0

Then we can plug in the second order energy shift to obtain the second order wavefunction
correction

φn φn φn φn
|n(2) i = H1 0 H1 |n(0) i − 0 hn(0) |H1 |n(0) i 0 H1 |n(0) i (967)
En0 − H0 En − H0 En − H0 En − H0

We can continue this iteration indefinitely to get an expression for the energy shift and
wavefunction at any order.
We can plug in the explicit definition of φn to see more clearly how this works.
Let’s define matrix elements of the perturbation H1 in the unperturbed energy basis

Vnk = hn(0) |H1 |k (0) i (968)

The energy shift is then

X |Vnk |2
∆n ≡ En − En(0) = λVnn + λ2 (0) (0)
+ ... (969)
k6=n En − Ek

133
 And the energy eigenket is
X Vkn
|ni = |n(0) i + λ |k (0) i (0) (0)
(970)
k6=n En − Ek
 
X X  |k (0) iVkl Vln X |k (0) iVnn Vkn 
+ λ2   −  2  (971)
(0) (0) (0) (0) (0) (0)
k6=n l6=n En − Ek En − El k6=n En − El

+ ... (972)

11.2 Time-independent perturbation theory with degeneracy

The above procedure must be modified when there is a degeneracy in the unperturbed energy
(0)
En .
In that case, we need to know which linear combinations of states in degenerate subspace
are energy eigenstates of the perturbed Hamiltonian, to make sense of our boundary condition

lim |ni = |n(0) i (973)

λ→0

We can also see that our formulae become ill-defined because of the terms where off-
diagonal matrix elements of the perturbation are diveded by the energy difference

Vnk
(0) (0)
. (974)
En − Ek

(0)
The correct choice is to choose H1 to be diagonal within the degenerate subspace of En .
(0)
Suppose there is a degenerate subspace of dimension g with energy ED . When the
perturbation is added, the space will typically split into g states that all have a different
energy.
If we start with some basis states |m(0) i, the perturbed energy eigenstates |li will be some
linear combination of these states in the limit λ → 0.

X
lim |li ≡ |l(0) i = hm(0) |l(0) i|m(0) i, (975)
λ→0
m∈D

where the sum is over the energy eigenkets in the degenerate subspace.
To expand in λ, let’s first rearrange the Schrodinger equation.
Let PD be the projection operator onto the degenerate subspace spanned by {|m(0) i}.

134
The projection operator onto the orthogonal subspace (all other states) is then P⊥ = 1 − PD .
That is,
X
PD = |m(0) ihm(0) | (976)
m∈D
X
P⊥ = |k (0) ihk (0) | (977)
k∈D
/

(0) (0) (0) (0)

with Em = ED and Ek 6= ED . Note that PD P⊥ = 0, and as projection operators,
PD2 = PD , and P⊥2 = P⊥ . We can also note that [H0 , PD ] = [H0 , P⊥ ] = 0.
Then the energy eigenstate |li can be separated into a parallel and orthogonal part:

|li = (PD + P⊥ )|li = PD |li + P⊥ |li (978)

We can then write the Schrodinger equation for the full energy eigentates |li as

0 = (E − H0 − λH1 )|li (979)

(0)
= (E − ED − λH1 )PD |li + (E − H0 − λH1 )P⊥ |li (980)

We can separate into two equations by projecting onto the two orthogonal subspaces:

(0)
0 = PD (E − ED − λH1 )PD |li + PD (E − H0 − λH1 )P⊥ |li (981)
(0)
= (E − ED − λPD H1 )PD |li − λPD H1 P⊥ |li (982)

and similarly

(0)
0 = P⊥ (E − ED − λH1 )PD |li + P⊥ (E − H0 − λH1 )P⊥ |li (983)
= −λP⊥ H1 PD |li − (E − H0 − λP⊥ H1 )P⊥ |li (984)

We can invert the last equation, since the operator on the right is not singular in the P⊥
subspace.

λ
P⊥ |li = P⊥ P⊥ H1 PD |li (985)
E − H0 − λP⊥ H1 P⊥

135
 If we now expand

|li = |l(0) i + λ|l(1) i + . . . (986)

we get the first order result:

λ

P⊥ |l(0) i + λ|l(1) i + . . . = P⊥ P⊥ H1 PD |l(0) i + λ|l(1) i + . . . (987)
E − H0 − λP⊥ H1 P⊥
1
=⇒ P⊥ |l(1) i = P⊥ (0) P⊥ H1 PD |l(0) i (988)
ED − H0
X 1
= (0)
|k (0) ihk (0) |H1 |l(0) i (989)
/ ED − H0
k∈D
X |k (0) iVkl
= (0) (0)
(990)
/ ED − Ek
k∈D

To calculate the parallel part PD |li, we substitute the orthogonal part into Eq. (982).
 
(0) 2 1
(E − ED − λPD H1 PD ) − λ PD H1 P⊥ P⊥ H1 PD PD |li = 0 (991)
E − H0 − λP⊥ H1 P⊥

Plugging in the zero-order eigenstate |l(0) i, we get an equation for the energies to order
λ and eigenfunctions to order zero

(0)
(E − ED − λPD H1 PD )(PD |l(0) i) = 0 (992)

This is an eigenvalue/eigenvector equation for the operator PD H1 PD , or H1 in the g × g

subspace.
The eigenvalues E (1) are just the roots of the secular equation
h i
(0)
det H̄1 − (E − ED ) = 0, (993)

where H̄1 is the matrix of P0 H1 P0 with matrix elements hm(0) |H1 |m(0) i.
Explicitly, in matrix form we have
    
V11 V12 · · · h1(0) |l(0) i h1(0) |l(0) i
    
V21 V22 · · · h2(0) |l(0) i = ∆(1) h2(0) |l(0) i (994)
   l  
.. .. . . .. ..
. . . . .

136
 (1)
The roots determine the eigenvalues ∆l , and by substituting them into the eigenvector
equation we can solve for hm(0) |l(0) i for each l up to an overall normalization constant.
So by solving the eigenvalue problem, we obtain both the first order energy shifts and the
zeroeth order eigenkets.
Notice that the zeroth-order kets we obtain for λ → 0 are just the linear combinations of
|m(0) i that diagonalize the perturbation H1 , with the diagonal elements giving the first-order
shift

(1)
∆l = hl(0) |H1 |l(0) i (995)

This is the same as the non-degenerate case, but when there is a degeneracy we must use
states |l(0) i within the degenerate subspace that diagonalize the perturbation H1 .
Now we want to find PD |li to first order in λ, starting from Eq. (991), neglecting the
higher-order term in the denominator:

 
(0) 1 2
(E − ED − λPD H1 PD ) − λ PD H1 P⊥ P⊥ H1 PD PD |li = 0 (996)
E − H0

(0)
For the g × g matrix PD H1 PD , let’s call the eigenvalues vi and the eigenvectors PD |li i.
The energies to first order are

(1) (0)
Ei = ED + λvi (997)

We assume that the degeneracy is completely resolved, so that all energy levels are dif-
(1) (1)
ferent to first order, so that Ei − Ej = λ(vi − vj ) are all non-zero.
In that case, we can use nondegenerate perturbation theory on this equation, with the λ2
term as the perterbation
X Vkn
|n(1) i = |k (0) i (0) (0)
(998)
k6=n En − Ek

(1)
X PD |lj(0) i (0) 1 (0)
=⇒ PD |li i = λ hlj |H1 P⊥ (0) P⊥ H1 |li i, (999)
j6=i
vj − vi ED − H0

137
 or more explicitly

(1)
X PD |lj(0) i X (0) 1 (0)
PD |li i = λ hlj |H1 |kiP⊥ (0) P hk|H1 |li i
(0) ⊥
(1000)
j6=i
vj − vi ED − Ek
k∈D
/

Adding this to the orthogonal component, Eq. (990), gives the total state to order λ.
We again choose the normalization

hl(0) |li = 1. (1001)

so that

(1) (2)
λhl(0) |H1 |li = ∆l = λ∆l + λ2 ∆l + . . . (1002)

The first order energy shift is as written above. The second order shift is then

(2)
∆l = hl(0) |H1 |l(1) i (1003)
= hl(0) |H1 PD |l(1) i + hl(0) |H1 P⊥ |l(1) i (1004)
= hl(0) |H1 P⊥ |l(1) i (1005)
X |Vkl |2
= (0) (0)
(1006)
/ ED − Ek
k∈D

(0)
Note that the first term is zero because PD |lj i is an eigenvector of H1 but the sum in
Eq. (1000) only includes terms j 6= i.
So the result is the same as the non-degenerate case, except the sum only includes states
outside the degenerate subspace D.
To summarize the procedure when there is a degeneracy:

1. For degeneracy of order g, construct matrix elements of the perturbation H1 in the

degenerate subspace

2. Diagonalize the resulting g × g perturbation matrix

3. The eigenvalues are the first-order energy shifts; the eigenvectors are the zeroth order
kets (which the perturbed kets approach in the limit λ → 0).

4. For higher orders, use the formulas of the corresponding nondegenerate perturbation
theory except in the summations, where we exclude all contributions from the unper-
turbed kets in the degenerate subspace D.

138
11.3 Fine structure of the hydrogen atom
We can use time-independent perturbation theory to improve our model of the hydrogen
atom, by including effects from relativity and spin on the dynamics of the electron.
The unperturbed system, which we already solved, is

p2
H0 = + V (r) (1007)
2m
p2 Ze2
= − (1008)
2m r

The energy eigenstates are eigenstates of L2 and Lz , |nlml i, with energy that only depends
on the principle quantum number n:

Eg mc2 Z 2 α2
En = 2 = − (1009)
n 2n2

The degeneracy of each energy level is n2 , due to the SO(4) symmetry of the problem.
To H0 we add fine structure corrections, which will change the energy levels. Since the
Hamiltonian will no longer respect the full symmetry of H0 , the corrections will break the
degeneracy.
We add three terms: the relativistic kinetic energy correction, the spin-orbit term, and
the Darwin term.

HFS = HRKE + HSO + HD (1010)

For the relativistic kinetic energy, recall that the energy (mass plus kinetic energy) of a
relativistic particle is

p p2 p4
m2 c4 + c2 p2 = mc2 + − + ... (1011)
2m 8m2 c2

where we have expanded the square root assuming p  mc.

The first term is the rest mass, the second is the nonrelativistic kinetic energy, and the
third is the first-order relativistic correction.

p4
HRKE = − (1012)
8m2 c2

139
 The spin-orbit term has the form

1
HSO = − µ · B0 , (1013)
2

where µ is the magnetic moment of the electron and B0 is the magnetic field in the electron
rest frame.
In a hydrogen atom, the magnetic field is due to the positive charge of the nucleus, which
is moving in the electron’s rest frame.
We can find it by performing a Lorentz transformation on the electromagnetic field in the
lab frame (where the nucleus is at rest and the magnetic field is zero) into the rest frame of
the electron.
In the lab frame, we have

1 1 1 dV
E = −∇Φ = ∇V = x, B=0 (1014)
e e r dr

If v is the velocity of the electron in the lab frame, we can perform a Lorentz transfor-
mation on the fields
Keeping only leading order terms for v  c, we find

1 1
B0 = − v × E = E × v (1015)
c c

So using p = mv, we can write

1 1 dV
B0 = L, (1016)
emc r dr

recalling that L = x × p.
Then, with

e
µ=− S, (1017)
mc

(where we approximate g ' 2), we finally have

1 1 dV
HSO = L·S (1018)
2m2 c2 r dr

The Darwin term is more complicated to understand, but describes a nonlocality in the
interaction of the electron with the electrostatic field.

140
 The result is

1 ~2
HD = ∇2 V (1019)
8 m2 c2

You can find a rough derivation in Littlejohn’s notes

http://bohr.physics.berkeley.edu/classes/221/1617/notes/finestruc.pdf
It can be derived more rigorously from a non-relativistic limit of the Dirac equation, but
here we will just use the result as practice for doing calculations in perturbation theory.
For the hydrogen atom, we have

Z
V (r) = − (1020)
r
1 dV Z
= 3 (1021)
r dr r
2
∇ V = 4πZδ(x) (1022)

To simplify the notation, we again use atomic units, in which

m=e=~=1 (1023)

So we have finally

α2 4
HRKE = − p (1024)
8
α2 2
HD = ∇V (1025)
8
π
= Zα2 δ(x) (1026)
2
α2 1 dV
HSO = L·S (1027)
2 r dr
Zα2 1
= L·S (1028)
2 r3

Each term has a factor

 2
1
2
α ' ∼ 5 × 10−5 , (1029)
137

so we can treat them as small perturbations.

Since the perturbations depend on spin, we need to consider the full Hilbert space of

141
spatial (orbital) and spin degrees of freedom

X = Xorb ⊗ Xspin (1030)

A complete set of energy eigenstates of H0 are

|nlml i ⊗ |sms i ≡ |nlml ms i (1031)

These are simultaneous eigenstates of (H0 , L2 , Lz , Sz ), which I will call the uncoupled basis.
Since the unperturbed energy levels are degenerate, we need to use degenerate perturba-
tion theory.
This means we will need to diagonalize the perturbation in the n2 -dimensional degenerate
subspace.
8
This is easier if we choose a good basis for states from the beginning.
Notes 24: Fine Structure
Note that if H1 commutes with one of the operators in the complete set of commuting
operators above, it is already diagonal with respect to that quantum number.
6. If
A we use Basis
Good the above (uncoupled)
for the basis, Perturbations
Fine Structure for example, we need to diagonalize the matrix
with Therefore
elements in analyzing the fine structure perturbation, we should look for observables that
commute with HFS . The results are summarized in Table 1. We start with HRKE . This is a purely
0 0 0
orbital operator, and is furthermore a hnlm s |H1 |nl m
scalar.l mTherefore itl m si
commutes (1032)
with L, the generator of orbital
rotations. Also, since it is purely an orbital operator, it commutes with S, which implies that it
alsoIf commutes
[Lz , H1 ] =with
0, then
J = L + S. Furthermore, since it commutes with L, S and J, it commutes
with any functions of them as well, including L2 , S 2 and J 2 . The term HD is similar; it is also a
0 0 0 0 0 0 0
= hnlm
purely0orbital l ms |(Lzwhich
operator, H1 − is
H1aLscalar.
z )|nl ml ms iδ-function
(The = (ml − can
ml )hnlm l ms |Hof
be thought 1 |nl
as m si
l mlimit
the of a (1033)
highly
concentrated, rotationally symmetric function centered on x = 0, which therefore commutes with
so 2 l =2 m0 .
L.)the matrix element
Therefore can be non-zero
HD also commutes with L,only
S, J,ifLm , S and 2
l J . However, the term HSO does not
So in general,
commute we want
with either to since
L or S, use aeither
basis purely
of commuting
spatial or observables, using
spin rotations as many
would rotate observables
one half or
thepossible
as other ofthat
the dot
commute L · S,
productwith H1and wouldwenot
. Then leave
will theneed
only dot product invariant.
to diagonalize However,
a smaller HSO
part of
does commute with J, which generates
the subspace (in this example, l and ms ). overall rotations of the system and which rotates both L and
S simultaneously.
In our case, theIt operators
also commutes
that with L2 andwith
commute S 2 , because
each part of of
thethe
commutators,
perturbation [L, L 2
are] = 0 and
summa-
2 2
[S, S ] = 0, and with J , because it commutes with J.
rized in this table.

FS Term L L2 S S2 J J2
HRKE Y Y Y Y Y Y
HD Y Y Y Y Y Y
HSO N Y N Y Y Y

Table 1. The table indicates whether the given fine structure term commutes with the given angular momentum
operator (Y = yes, N = no).

We see that HRKE and HD are diagonal in 142the uncoupled basis (29), but that HSO is not.
However, all three operators are diagonal in the eigenbasis of the operators (L2 , S 2 , J 2 , Jz ). This
suggests that we use the coupled basis for the perturbation treatment, that is, the basis in which
we have combined angular momenta according to J = L + S or ℓ ⊗ 21 , as in Notes 18. Following
 So HRKE and HD are diagonal in the uncoupled basis, but HSO is not.
All the operators are diagonal in the eigenbasis of (L2 , S 2 , J 2 , Jz ), so it is better to use
this (coupled) basis for calculation,
X
|nljmj i = |nlml ms ihlsml ms |jmj i (1034)
ml ,ms

in which case we can skip the diagonalization step of degenerate perturbation theory. The
diagonal elements of the perturbation themselves directly give us the first order energy shifts.
Further, each of the three terms is diagonal, so we can calculate the energy shifts inde-
pendently, and the total energy shift will just be the sum.
The shift due to the relativistic kinetic energy is easier to calculate in the uncoupled basis,
so we have

(1)
X X
∆RKE = hnljmj |HRKE |nljmj i = hjmj |lsml ms ihnlml ms |HRKE |nlm0l m0s ihlsm0l m0s |jmj i
ml ,ms m0l ,m0s

(1035)

Since HRKE is independent of spin, the middle matrix element is

hnlml ms |HRKE |nlm0l m0s i = δms ,m0s hnlml |HRKE |nlm0l i, (1036)

where the last matrix element is purely orbital.

By the Wigner-Eckart theorem, it is equal to δml ,m0l times a quantity that is independent
of magnetic quantum numbers, which we can write as

hnlml |HRKE |nlm0l i = δml ,m0l hnl0|HRKE |nl0i, (1037)

choosing ml = 0 for convenience.

Plugging these back in and using the orthogonality of Clebsch-Gordan coefficients,
X X
hnljmj |HRKE |nljmj i = hjmj |lsml ms ihnlml ms |HRKE |nlm0l m0s ihlsm0l m0s |jmj i
ml ,ms m0l ,m0s

(1038)
X
= hnl0|HRKE |nl0i hjmj |lsml ms ihlsml ms |jmj i (1039)
ml ,ms

= hnl0|HRKE |nl0i (1040)

143
 So the energy shift depends only on n and l.
We can calculate the final orbital matrix element by writing

α2 2 α2
HRKE = − T = − (H0 − V )2 , (1041)
2 2

with T ≡ p2 /2 the kinetic energy. Then

α2
hnljmj |HRKE |nljmj i = − hnl0|(H02 − H0 V − V H0 + V 2 )|nl0i (1042)
2
α2
= − hnl0|(En2 − 2En V + V 2 )|nl0i (1043)
2

For the hydrogen atom we have

 
1 Z
= 2 (1044)
r n
 
1 Z2
= (1045)
r2 n2 (l + 21 )

After some algebra, the final answer is

 
2 1 3 n
hnljmj |HRKE |nljmj i = (Zα) (−En ) 2 − (1046)
n 4 l + 12

So the energy correction is of order (Zα)2

Now consider the Darwin term:

π
HD = Zα2 δ(x) (1047)
2

It is also a purely spatial, scalar operator, so its matrix element also reduces in the same
way

π
hnljmj |HD |nljmj i = hnl0|HD |nl0i = Zα2 |ψnl0 (0)|2 (1048)
2
ψnlm = Rnl (r)Ylm (θ, φ) (1049)
∼ rl (1050)

So the only non-zero result is for l = 0 (s-waves).

144
 √
Using Y00 = 1/ 4π and
 3/2
Z
Rn0 (0) = 2 , (1051)
n

the final answer can be written in the form

1
hnljmj |HD |nljmj i = (Zα)2 (−En ) δl0 (1052)
n

Finally, for the spin-orbit term,

Zα2 1
HSO = L·S (1053)
2 r3

note that

1
L · S = (J 2 − L2 − S 2 ) (1054)
2
Zα2 1
=⇒ hnljmj |HSO |nljmj i = [j(j + 1) − l(l + 1) − s(s + 1)]hnljmj | 3 |nljmj i (1055)
4 r

The remaining matrix element is again a purely spatial scalar operator, which can be
reduced. The result is
 
1 Z3
= (1056)
r3 n3 l(l + 12 )(l + 1)

The expectation value diverges for l = 0, but that is also the place where the operator L
is the zero operator. We need to take the limit carefully to evaluate the expression.
If we smooth out the Coulomb singularity (for example, giving the nucleus a finite size),
the expectation of 1/r3 does not diverge, and the answer vanishes for l = 0. The same is
true in the limit of vanishing nuclear size.
Altogether, the answer for l 6= 0 is

2 1 j(j + 1) − l(l + 1) − 43
hnljmj |HSO |nljmj i = (Zα) (−En ) (1057)
2n l(l + 21 )(l + 1)

1  l(l+ 12 )(l+1) , j = l + 2
1 1
2
= (Zα) (−En ) (1058)
2n − 1 1 , j = l − 12
l(l+ )2

145
 Adding the three contributions gives
 
2 1 3 n
∆EFS = (Zα) (−En ) 2 − (1059)
n 4 j + 12
    
Z 2 1 3 n
=⇒ Enj = − 2 1 − (Zα) 2 − + ... (1060)
2n n 4 j + 12
Notes 24: Fine Structure 13
Although each correction depends on l, the total energy shift does not. Since H is a
scalar, it does not depend on mj either.
Recalling the spectroscopic notation nLj , the fine structure (and allowed electric dipole
included, one misses the fact that the total fine structure energy shifts are independent of ℓ. It is
transitions) looks like:
clear from the formula (51) that this ℓ-degeneracy persists to all orders in the expansion of the Dirac
equation in powers of α. 3d5/2
3p3/2 3d3/2
3s1/2 3p1/2

2p3/2
2s1/2 2p1/2

1s1/2

Fig. 1. Energy level diagram for hydrogen or hydrogen-like atoms, including fine structure, with allowed electric dipole
transitions indicated (Grotrian diagram). Not shown are transitions only involving small energy differences, such as
3p11.4
3/2 → 3s1/2Variational Methods
. Diagram is schematic and not to scale; in particular, the fine structure splittings are of order (Zα)2
times the separation between the levels of different n.
Perturbation theory requires the solution to a Hamiltonian H0 that is close to the desired
Hamiltonian.
The standard spectroscopic notation for the states of hydrogen-like or alkali atoms is nℓj , where
The variational
ℓ is represented method
by one of the allows
standardto symbols,
estimate s,thep, ground-state
d, f , etc. Forenergy (and
example, theexcited
groundstate
state of
energies)iswithout
hydrogen needing
the 1s1/2 any low
level. The exact solutions.
lying levels in a hydrogen-like atom are indicated schematically
We1.start by guessing a ground state solution |0̃i, which in general may not be the true
in Fig.
ground state |0i.
13. Electric Dipole Transitions in Hydrogen-Like Atoms

Figure 1 also indicates the most important electric dipole transitions in a hydrogen-like atom.
Allowed electric dipole transitions, (n′ ℓ′ j ′ m′j ) → (nℓjmj ), are those for which the matrix element

|xq |n′ ℓ′ j ′ m′j ⟩

⟨nℓjmj146 (53)

is nonzero, where xq is the component of the position operator x with respect to the spherical basis
(19.28). The operator xq is a k = 1 irreducible tensor operator, both under purely spatial rotations,
 The expectation value of the energy of this state is

h0̃|H|0̃i
H̄ = (1061)
h0̃|0̃i

which cannot be lower than the ground state energy

H̄ ≥ E0 (1062)

To prove, imagine that we know the correct energy eigenstates

H|ki = Ek |ki, (1063)

which form a complete basis, and so

∞
X
|0̃i = |kihk|0̃i (1064)
k=0
P∞ 2
k=0 |hk|0̃i| Ek
=⇒ H̄ = P ∞ (1065)
k=0 |hk|0̃i|
2
P∞
|hk|0̃i|2 (Ek − E0 )
= E0 + k=1P∞ (1066)
k=0 |hk|0̃i|
2

≥ E0 , (1067)

where E0 is the ground state energy, so (Ek − E0 ) ≤ 0, and the second term is nonnegative
definite.
So the ground state energy is a lower bound on the expected energy. If we could calculate
H̄ for all possible kets, then the ground state is the one with the smallest energy.
In practice, we can’t explore the entire Hilbert space, but we can consider a family of
states characterized by a set of parameters λ1 , λ2 , . . .

|0̃i = |0̃i(λ1 , λ2 , . . .) (1068)

=⇒ H̄ = H̄(λ1 , λ2 , . . .) (1069)

If we choose the parameters that minimize H̄

∂ H̄ ∂ H̄
= 0, = 0, ..., (1070)
∂λ1 ∂λ2

then the resulting energy is the closest to the ground state energy withing the class of trial

147
states.
If the functional form is correct, then this gives the exact ground state wavefunction.
For example, if we could guess the wavefunction of the hydrogen atom as having the form

hx|0i ∝ e−r/a , (1071)

we would find that the energy is minimized when a = a0 is the Bohr radius, and the cor-
resonding energy is the correct ground state energy −e2 /2a0 .
However, even if the functional form is not correct, it is often possible to get very close
to the correct ground state energy E0 .

11.5 Time-dependent perturbation theory

We discussed time-independent perturbation theory, where the Hamiltonian does not depend
on time, and to solve the problem we need the energy eigenvalues and eigenstates.
We can also use perturbation theory to give information about time-dependent processes.
The general problem can be written as

H = H0 + H1 (t) (1072)

where H0 is again a solvable Hamiltonian, and H1 is small, but this time H1 can depend
on time.
In this case, we want to find time-dependent transitions between states. The basic quan-
tity is the transition amplitude

hf |U (t)|ii (1073)

where U (t) is the time evolution operator for H, recalling that for time-dependent problems

U (t) 6= e−iHt/~ (1074)

Typically, |ii and |f i are two eigenstates of the unperturbed Hamiltonian (the “initial”
and “final” states).

(0) (0)
H0 |ii = Ei |ii, H0 |f i = Ef |f i, (1075)

The absolute square is the transition probability – the probability that a state |ii at time

148
ti = 0 will transition to |f i in time t.

P (i → f ) = |hf |U (t)|ii|2 (1076)

We define two different time evolution operators: the full time evolution U (t), and the
time evolution of a system with Hamiltionian H0 , U0 (t).
They satisfy the equations

∂U (t)
i~ = H(t)U (t) (1077)
∂t
∂U0 (t)
i~ = H0 U0 (t) (1078)
∂t

Since the unperturbed Hamiltonian is time-independent,

U0 (t) = e−iH0 t/~ , (1079)

but this is not true for U (t).

Recall that, instead of the Schrodinger picture that we usually use, there are other ways
to consider the time dependence in quantum mechanics.
We have already seen the Heisenberg picture, where the states are time-independent, and
all the time dependence is contained in the operators:

|ψH i = U (t)† |ψS (t)i = |ψS (t = 0)i (1080)

AH (t) = U (t)† AS (t)U (t) (1081)

In this case, it is useful to use a third option: the interaction picture:

|ψI (t)i = U0† (t)|ψS (t)i (1082)

note that the states are the same at t = 0:

|ψI (0)i = |ψS (0)i (1083)

For matrix elements to have the correct time evolution, the operators must then have the

149
relation

AI (t) = U0† AS (t)U0 (t) (1084)

AS is sometimes time-independent, but AI is always time-dependent.

If H1 is small, than most of the time dependence comes from H0 . The interaction picture
removes this fast time dependence from states, leaving only the slow time dependence due
to H1 .

11.5.1 Dyson Series

Define W (t) as the operator that evolves interaction-picture states in time:

|ψI (t)i = W (t)|ψI (0)i = W (t)|ψS (0)i (1085)

= U0 (t)† |ψS (t)i = U0 (t)† U (t)|ψS (0)i (1086)
=⇒ W (t) = U0 (t)† U (t) (1087)

We can differentiate this expression to get a differential equation for W :

∂W (t) ∂U0 (t)† ∂U (t)

i~ = i~ U (t) + U0 (t)† i~ (1088)
∂t ∂t ∂t
† †
= −U0 (t) H0 U (t) + U0 (t) HU (t) (1089)
= U0 (t)† H1 U (t) (1090)
  
† †
= U0 (t) H1 U0 (t) U0 (t) U (t) (1091)

= H1I (t)W (t) (1092)

Acting on the initial state |ψI i, we can get a version of the Schrodinger equation in the
interaction picture

∂|ψI (t)i ∂W (t)

i~ = i~ |ψI (0)i (1093)
∂t ∂t
= H1I (t)W (t)|ψI (0)i (1094)
= H1I (t)|ψI (t)i (1095)

150
 We can integrate the operator equation, with boundary condition W (0) = 1:
Z t
1
W (t) = 1 + dt0 H1I (t0 )W (t0 ) (1096)
i~ 0

This is an exact relation, but not a solution for W , since it appears on both sides.
However, assuming the perturbation is small, we can generate a power series in H1I by
iteratively substituting the entire right side of the equation into the integral:
Z t Z t Z t0
1 1
W (t) = 1 + 0 0
dt H1I (t ) + dt 0
dt00 H1I (t)H1I (t00 )W (t00 ) (1097)
i~ 0 (i~)2 0 0

This is still an exact relation, but if we ignore the last term, we have a first-order correction
to the time evolution W (t). We can plug in the first-order result into this full equation to
get a second order result, and so on to arbitrary order

Z t Z t Z t0
1 1
W (t) = 1 + 0 0
dt H1I (t ) + dt 0
dt00 H1I (t)H1I (t00 ) + . . . (1098)
i~ 0 (i~)2 0 0

Each extra term has another power of the perturbation H1 , and so hopefully gives a
smaller correction, and the series can be truncated at a finite number of terms to get an
accurate result.
This series is called the Dyson series.
As mentioned above, the usual problem that we want to solve is when the system is
initially in an eigenstate of the H0 .

H0 |ii = Ei |ii (1099)

If we act with the Dyson series on this initial state |ψI (0)i = |ψS (0)i = |ii, the state at a
later time is

|ψI (t)i = W (t)|ii (1100)

Z t Z t Z t0
1 1
= |ii + dt0 H1I (t0 )|ii + dt0 dt00 H1I (t)H1I (t00 )|ii + . . . (1101)
i~ 0 (i~)2 0 0

We can expand this solution in eigenstates of H0 :

X
|ψI (t)i = cn (t)|ni (1102)
n

151
The coefficients cn (t) are functions of time, and encode the time dependence.
Acting on the left with an energy eigenstate hn| shows that

cn (t) = hn|W (t)|ii, (1103)

So that the coefficients are the matrix elements of the time evolution operator in the
interaction picture W (t) in the unperturbed eigenbasis.
They are related to the transition amplitudes by a simple phase factor

cn (t) = hn|U0 (t)† U (t)|ii = eiEn t/~ hn|U (t)|ii, (1104)

which removes the rapid time evolution from the transition amplitude, leaving behind only
the slower evolution due to H1 .
The extra phase cancels in the probability

Pn (t) ≡ |cn (t)|2 = |hn|W (t)|ii|2 = |hn|U (t)|ii|2 (1105)

If we act on the left with hn| to the Dyson series for states |ψI (t)i, we get a series expansion
for cn (t):

cn (t) = δni + c(1) (2)

n (t) + cn (t) + . . . , (1106)

with
Z
1 t 0
c(1)
n (t) = dt hn|H1l (t0 )|ii, (1107)
i~ 0
Z t Z t0
1
(2)
cn (t) = 2
dt0 dt00 hn|H1l (t0 )H1l (t00 )|ii, (1108)
(i~) 0 0

etc.
Going back to the Schrodinger picture, we have
Z
1 t 0
c(1)
n (t) = dt hn|U0 (t0 )† H1 (t0 )U0 (t0 )|ii (1109)
i~ 0
Z
1 t 0 iωni t0
= dt e hn|H1 (t0 )|ii, (1110)
i~ 0

152
where ωni is the Einstein frequency connecting unperturbed states n and i,

En − Ei
ωni = (1111)
~

For the second order, it is useful to insert a resolution of the identity between the two
perturbations
Z t Z t0 X
1 0
c(2)
n (t) = dt dt00 hn|H1l (t0 )|kihk|H1l (t00 )|ii (1112)
(i~)2 0 0 k
Z t Z t0 X
1 0 0 00
= dt dt00 eiωnk t +iωki t hn|H1 (t0 )|kihk|H1 (t00 )|ii, (1113)
(i~)2 0 0 k

and similarly for each higher order.

Notice that the variables of integration satisfy 0 ≤ t00 ≤ t0 ≤ t. So we can think of this
calculation as a sum of possible transitions involving intermediate states |ki

|ii → |ki → |ni (1114)

each higher order in perturbation theory contains one more intermediate state.
This is the basic structure of time-dependent perturbation theory. Next we look at some
examples and applications.

11.5.2 Examples: Time-independent H1

A simple example is a time-independent H1 .

In that case the matrix elements can be taken out of the integrals, which can be done
analytically. For example, in the first order expression we have

Z t
1 0
c(1)
n (t) = hn|H1 |ii dt0 eiωni t (1115)
i~ 0
 
1 iωni t/2 sin ωni t/2
= hn|H1 |ii2e (1116)
i~ ωni

The transition probability can also be expanded in the perturbation series

Pn (t) = |cn (t)|2 = |δni + c(1) (2)

n (t) + cn (t) + . . . |
2
(1117)

Note that there are interference terms between amplitudes at different orders.

153
 If we assume we are interested in transition to a different state than the initial state n 6= i,
we have to first order

 
(1) 24 sin2 ωni t/2
Pn (t) ' |c | = 2 2
|hn|H1 |ii|2 (1118)
~ ωni

The time dependence is contained in the parentheses, which depend only on the energy
of the initial and final states, while all other aspects of the states only enter in the (time-
independent) matrix element.

11.5.3 Examples: Periodic H1

Another simple and important case is when H1 is periodic in time,

H1 (t) = Ke−iω0 t + K † eiω0 t (1119)

I refer to each of these terms as the positive frequency and negative frequency part of the
perturbation, respectively.
This applies, for example, to the interaction of spins or atoms with a classical electro-
magnetic wave.
If we insert this into the expression for the first order transition amplitude, we get
 
2 i(ωni −ω0 )t/2 sin(ωni − ω0 )t/2 i(ωni +ω0 )t/2 sin(ωni + ω0 )t/2 †
c(1)
n (t) = e hn|K|ii + e hn|K |ii
i~ ωni − ω0 ωni + ω0
(1120)

The probability depends on the magnitude of each term, plus interference terms.
However, if we are only interested in the final states to which most of the probability
goes, this is dominated by states where the denominator of one of the terms is small:

ωni ∓ ω0 ' 0, (1121)

En ' Ei ± ~ω0 (1122)

The first term (positive frequency) is resonant when the system has absorbed a quantum of
energy ~ω0 from the perturbation, whereas the second (negative frequency) term is resonant
when the system has given up a quantum of energy ~ω0 to the perturbation.
We call these two cases absorption and stimulated emission, respectively.

154
 In the case of absorption, considering only final states |ni that are near resonance (En '
Ei + ~ω0 ), the first order transition probability is
 
4 sin2 (ωni − ω0 )t/2
Pn = 2 |hn|K|ii|2 (1123)
~ (ωni − ω0 )2

For stimulated emission, we have

 
4 sin2 (ωni + ω0 )t/2
Pn = 2 |hn|K † |ii|2 (1124)
~ (ωni + ω0 )2

In both cases, and in the case of a time-independent potential, the time dependence has
a similar functional form.

155