Introduction to Quantum Mechanics

Eva-Maria Graefe
Department of Mathematics
Imperial College London

Academic year 2023/24

Autumn Term
ii
Contents

1 Classical mechanics 1
1.1 Newton’s second law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Hamilton’s canonical equations and phase space-dynamics . . . . . . . . . . . . . . . 1
1.2.1 Hamilton’s canonical equations . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2.2 The potential and characteristics of the dynamics . . . . . . . . . . . . . . . . 2
1.3 Time evolution of general dynamical variables - Poisson brackets . . . . . . . . . . . 4
1.3.1 Properties of Poisson brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Historic excursion: The birth of quantum mechanics . . . . . . . . . . . . . . . . . . 5

2 Schrödinger dynamics 7
2.1 The Schrödinger equation and the wave function . . . . . . . . . . . . . . . . . . . . 7
2.1.1 The wave function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Probability current and conservation of probability . . . . . . . . . . . . . . . 8
2.1.3 Superposition principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.4 The Hamiltonian and the energy . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Stationary solutions: The time-independent Schrödinger equation . . . . . . . . . . . 9
2.3 The method of stationary states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 The method of stationary states (for one dimensional systems in coordinate
representation) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Example: Particle in a half box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Excursion: Mathematical background 15

3.1 States and obervables - Hilbert spaces and Hermitian operators . . . . . . . . . . . . 15
3.2 Some linear algebra and analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 The Dirac notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3.1 The dual space and the Riesz theorem . . . . . . . . . . . . . . . . . . . . . . 18
3.3.2 The Dirac notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3.3 Rules for finding the adjoint of an expression . . . . . . . . . . . . . . . . . . 19
3.3.4 The outer product of two vectors . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3.5 Completeness and resolution of identity . . . . . . . . . . . . . . . . . . . . . 20
3.4 More on Hermitian operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.5 Vector/matrix representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4 The Principles of Quantum Mechanics 23

4.1 States and observables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.1.1 Principle 1 (States): . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.1.2 Principle 2 (Observables): . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Measurement and dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2.1 Principle 3 (“Born rule”): . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2.2 Projection operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2.3 Principle 4 (“state collapse postulate”): . . . . . . . . . . . . . . . . . . . . . 25
4.2.4 Principle 5 (dynamics): . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

iii
iv CONTENTS

4.2.5 The method of stationary states in Dirac notation (for arbitrary Hilbert spaces) 26
4.3 Commutators and uncertainties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.3.1 The uncertainty relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.3.2 Commuting observables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3.3 Complete set of commuting observables . . . . . . . . . . . . . . . . . . . . . 28

5 The harmonic oscillator 31

5.1 Algebraic solution of the eigenvalue equation . . . . . . . . . . . . . . . . . . . . . . 31
5.2 Normalised eigenvectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.3 Matrix representations in the harmonic oscillator basis . . . . . . . . . . . . . . . . . 34
5.4 Harmonic oscillator basis - A simple numerical method . . . . . . . . . . . . . . . . . 35

6 Representations 37
6.1 Discrete and continuous spectra - normalisable and improper eigenvectors . . . . . . 37
6.1.1 A few comments on the Dirac δ-Distribution . . . . . . . . . . . . . . . . . . 38
6.2 Eigenvalues of the position operator . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.3 Position representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.4 Change of bases/representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

7 Spectral properties of one-dimensional quantum systems 43

7.1 Position representation of the harmonic oscillator eigenstates . . . . . . . . . . . . . 43
7.2 Properties of eigenfunctions for general one-dimensional potentials . . . . . . . . . . 45
7.2.1 Symmetric potentials - parity . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
7.3 Piecewise constant potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7.3.1 Piecewise constant potentials and boundary conditions . . . . . . . . . . . . . 47
7.3.2 The free particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7.3.3 The infinite square well potential (“Particle in a box”) . . . . . . . . . . . . . 48
7.3.4 The finite square well potential (”Particle in a finite box”) . . . . . . . . . . . 49

8 Quantum dynamics 51
8.1 Dynamics of expectation values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
8.2 Lie groups and Lie algebras in quantum dynamics . . . . . . . . . . . . . . . . . . . 54
8.2.1 A quick overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
8.2.2 Reminder: The group SU (2) . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
8.3 Dynamical Tunneling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

9 Angular momentum 59
9.1 Definition of angular momentum - classical and quantum . . . . . . . . . . . . . . . 59
9.2 Angular momentum and rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
9.2.1 Reminder: Rotations in R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
9.2.2 Action of the eiφL̂j /~ on the position operators . . . . . . . . . . . . . . . . . 60
ˆ
9.2.3 Action of the eiφJj /~ on the angular momentum operators . . . . . . . . . . . 61
9.3 Spectral properties of angular momentum . . . . . . . . . . . . . . . . . . . . . . . . 62
9.3.1 Orbital angular momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
9.3.2 Algebraic derivation of the spectral properties . . . . . . . . . . . . . . . . . . 63
9.4 Matrix representation in the “standard basis” . . . . . . . . . . . . . . . . . . . . . . 65
9.5 Quantum Spin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
9.5.1 A short history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
CONTENTS v

10 Mastery material: The Husimi distribution 69

10.1 Classical and quantum phase-space distributions . . . . . . . . . . . . . . . . . . . . 69
10.2 Coherent states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
10.3 Definition and properties of the Husimi distribution . . . . . . . . . . . . . . . . . . 70
10.3.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
10.3.2 Dynamics of the Husimi distribution for the Hermitian Harmonic oscillator . 72
10.4 SU(2) coherent states and Husimi distribution . . . . . . . . . . . . . . . . . . . . . . 72
vi CONTENTS
Chapter 1

Classical mechanics

Here we summarise the necessary background from classical mechanics, and introduce some physical
terminology that will repeatedly occur in the lecture.

1.1 Newton’s second law

A classical particle has a mass m, and its state is described by its position ~r ∈ R3 and its velocity
~v = ~r˙ = d~
r
dt or, often more conveniently, its momentum p ~ = m~v . Newton’s second law connects the
change of momentum to the external force F : ~

p~˙ = m~r¨ = F~ . (1.1)

This is a second order differential equation for the time-dependent position of a classical particle.
The solution is uniquely determined by two initial conditions, for example the position and velocity
at a given time. The solution r(t)~ is called the trajectory of the particle.
It is often convenient to study a system’s dynamics in the so-called phase space, the space of
position and momentum coordinates. A phase-space trajectory, (~r(t), p~(t)) is a curve in phase space,
that position and momentum coordinates trace as time evolves for a particular initial condition.
A phase (space) portrait is a sketch of a collection of several phase-space trajectories belonging to
different initial conditions, that summarise the overall types of trajectories.

1.2 Hamilton’s canonical equations and phase space-dynamics

1.2.1 Hamilton’s canonical equations
In many cases of interest the force can be expressed as the negative gradient of a scalar function
V (~r), the potential, i.e.,
F~ (~r) = −∇V
~ (~r). (1.2)

Such systems are referred to as conservative, as they conserve the energy.

Remark 1. Remind yourself of the definitions and the physical units of force, energy, and action.
In what follows we shall for simplicity discuss one-dimensional systems, with coordinate q ∈ R
and momentum p ∈ R. For a conservative system Newton’s second law (1.1) can then be written
as
∂V
ṗ = − , (1.3)
∂q
and by definition we have
p
q̇ = . (1.4)
m

1
2 CHAPTER 1. CLASSICAL MECHANICS

Introducing the function

p2
H(p, q) = + V (q), (1.5)
2m
the dynamical equations (1.3) and (1.4) can be rewritten as

∂H ∂H
ṗ = − and q̇ = . (1.6)
∂q ∂p

These are Hamilton’s canonical equations. The function H(p, q) is called the Hamiltonian function,
p2
its value is the total energy of a system for a given state (p, q). The term 2m is the kinetic energy,
V (q) is the potential energy. Hamiltonian dynamics conserves the total energy, i.e., dHdt = 0.
Remark 2. Hamiltonian mechanics applies more generally than in situations where p = mq̇, i.e., q
and p do not have to refer to physical coordinate and momentum in all situations (see canonically
conjugate variables below). It does only apply in situations where the total energy is conserved.
Thus, for example systems with friction or other losses cannot be described in the standard Hamil-
tonian formalism.

1.2.2 The potential and characteristics of the dynamics

Figure 1.1: Example potential.

For systems whose Hamiltonian is of the form kinetic plus potential energy, as in equation (1.5),
one can deduce basic properties of the dynamics from the potential, as sketched in figure 1.1. The
extrema of the potential correspond to fixed points (or equilibrium configurations) of the dynamics.
Minima are stable, and maxima are unstable fixed points. For a given energy a particle can only be
in a region where its total energy is larger or equal to the potential energy. These regions are called
classically allowed. They are separated by turning points from the classically forbidden regions, in
which the potential is larger than the total energy available. In a classically allowed region that is
bounded by two turning points, a particle oscillates between them. For a region that is only bound
by one turning point, the particle comes in from one side, is reflected at the turning point and then
moves away in the direction it came from, towards (negative or positive) infinity.

Exercise 1. Sketch the phase space portrait for a

particle in one dimension with Hamiltonian of the
p2
form 2m + V (q) for the two potentials sketched here
on the right.
1.2. HAMILTON’S CANONICAL EQUATIONS AND PHASE SPACE-DYNAMICS 3

Example 1 - The free particle. The free particle moves without an external potential, i.e., the
p2
Hamiltonian is given by H = 2m . Thus, the canonical equations of motion become

p
ṗ = 0, and q̇ = , (1.7)
m
p
and we find p = const, and q(t) = mt + q0 . Sketch the phase space portrait!
p p2
Remark 3. q̇ = m of course always holds for systems whose total energy is of the form H = 2m +V (q).
Remark 4. We could have also used the conservation of energy to deduce that the momentum is
conserved here.

Example 2 - The harmonic oscillator. The harmonic oscillator is a particle in a quadratic

potential, V (q) ∝ q 2 . It is a very important model system in physics for several reasons. Firstly,
it has a simple analytical solution. Secondly, and more importantly, it is a good approximation for
many systems that perform small oscillations around an equilibrium configuration. On a mathe-
matical level this approximation can be interpreted as a Taylor expansion of the actual potential
up to second order around a minimum, where the first derivative vanishes. The harmonic oscillator
is for example a rough approximation for the vibrations of a two-atomic molecule.
Let us here consider the example of a mass m on a spring for small displacements from the
equilibrium configuration. At equilibrium the spring counterbalances gravity and there is no net
force acting on the mass. If the mass is displaced slightly, the spring responds with a compensation
force which is proportional to the displacement (this is known as Hooke’s law ):

F = −Dq, (1.8)

where D is the spring constant, and q denotes the displacement from equilibrium
q (i.e. q is positive
D
when the spring is stretched). It is convenient to define a frequency ω = m . The potential is then
given by
D mω 2 2
V (q) = q 2 = q , (1.9)
2 2
and we have the Hamiltonian
p2 mω 2 2
H= + q . (1.10)
2m 2
From the conservation of energy we can deduce the phase space portrait as consisting of concentric
ellipses.
The equations of motion
p
ṗ = −mω 2 q, q̇ = , (1.11)
m
can be combined into a single second-order differential equation for q,

q̈ = −ω 2 q. (1.12)

With the initial conditions q(t = 0) = q0 and p(t = 0) = p0 the solution is given by

p0
q(t) = q0 cos(ωt) + sin(ωt) (1.13)
mω
p(t) = p0 cos(ωt) − q0 mω sin(ωt). (1.14)

These equations describe harmonic oscillations.

4 CHAPTER 1. CLASSICAL MECHANICS

1.3 Time evolution of general dynamical variables - Poisson brack-

ets
The time evolution of any dynamical variable A(p, q) is governed by the equation
dA ∂A ∂A ∂A
= q̇ + ṗ + . (1.15)
dt ∂q ∂p ∂t

Remark 5. For a quantity without explicit time dependence we have ∂A ∂t = 0. All A(p, q) are
implicitly time dependent via the time dependence of p and q. Some variables, however, might in
addition have an explicit time dependence. Take for example the potential energy in an oscillating
potential of the form V (q, t) = (q − cos(ωt))2 .
Inserting Hamilton’s canonical equations (1.6) into (1.15) we find
dA ∂A ∂H ∂A ∂H ∂A ∂A
= − + = {H, A} + , (1.16)
dt ∂q ∂p ∂p ∂q ∂t ∂t
where we have defined the Poisson bracket, {·, ·} as
∂A ∂B ∂A ∂B
{A, B} = − . (1.17)
∂p ∂q ∂q ∂p
For N -dimensional systems the Poisson bracket is defined as
N
X ∂A ∂B ∂A ∂B
{A, B} = − . (1.18)
∂pj ∂qj ∂qj ∂pj
j=1

Remark 6. Note that both this definition and the one where p and q are interchanged are used in
the literature.
A dynamical variable without explicit time dependence is conserved in time if {H, A} = 0. Such
quantities are also referred to as constants of motion. For the special case that H does not depend
on one of the canonical variables qj the corresponding pj is conserved. Then qj is called a cyclic
variable.

1.3.1 Properties of Poisson brackets

The Poisson bracket has the following properties:
(i) {A, A} = 0 for all A

(ii) Linearity in the first element: {c1 A + c2 B, C} = c1 {A, C} + c2 {B, C}

(iii) Anti-symmetry with respect to interchange of elements: {A, B} = −{B, A}

(iv) Poisson-bracket with constant vanishes: {c, A} = 0, for c constant

(v) Leibniz rule (product rule): {AB, C} = A{B, C} + {A, C}B

(vi) Jacobi identity: {A, {B, C}} + {B, {C, A}} + {C, {A, B}} = 0.
The Poisson bracket between momentum and position coordinates {pj , qk } = δjk is called the
fundamental Poisson bracket. Every pair Pj (pl , qm ), Qj (pl , qm ) that fulfils {Pj , Qj } = 1 is a pair of
canonically conjugate variables and one can use the Hamiltonian formalism where H(Pj , Qj ) is the
P2
total energy of the system, even if this is different from 2m + V (Q). In general of course this also
P
means that Q̇ 6= m .
Exercise 2. Verify that the Poisson bracket defined in (1.17) indeed fulfils all the above properties.
1.4. HISTORIC EXCURSION: THE BIRTH OF QUANTUM MECHANICS 5

1.4 Historic excursion: The birth of quantum mechanics

In the late 19th century physics was regarded as “complete”, in the sense that it was believed
that Newton’s laws, together with Maxwell’s theory of electromagnetism explained everything, and
all that was left was applying these theories to various situations. However, at the turn of the
century some experiments were under way that revealed a fundamental lack of understanding. These
included the measurement of atomic spectra, scattering experiments, the photo-electric effect, and
the study of the light emitted from hot objects. These experiments and the attempts to build
a corresponding theory led to the biggest revolution in modern physics, and the development of
quantum mechanics.
Quantum theory is often divided into old and modern quantum theory. The name old quantum
theory refers to the theory used in the initial 25 years of quantum theory, for example Bohr’s theory
of the atom, where electrons are confined to discrete orbits around the atomic nucleus, while there
is no actual theoretical reason for this phenomenon given. While old quantum theory explained
many of the experimental results, and introduced the concept of energy quantisation, there was
no deeper understanding of the phenomena, and there were a number of unsolved paradoxes. In
1924/25 several breakthroughs were made, which led to the modern formulation of quantum theory
(which, I want to mention such as not to mislead you, still has a lot of unresolved aspects, but we
shall not focus on these.).
Werner Heisenberg (who was only 24 then) came up with a strange dynamical equation for
observables (measurable quantities) that do not commute:

2πi
ȧ = (W a − aW ),
h
where a is an arbitrary observable (for example the position of a particle), and W denotes the
energy. It later turned out that the idea of non-commutative objects is of course closely related to
matrices, which were not very popular/well-known amongst physicists in the 1920’s. Heisenberg’s
formulation of quantum mechanics is sometimes referred to as matrix mechanics.
In 1926 Schrödinger came up with his famous wave equation,

h2
 
h
i ψ̇(~r, t) = − 2 ∇2 + V (~r) ψ(~r, t),
2π 8π m

where the complex function ψ(~r, t) is meant to somehow encode the position of a quantum particle,
~2
and the operator − 2m ∇2 + V (~r) represents the energy. We shall discuss this further in section 2.
The fact that Schrödinger’s and Heisenberg’s theories are actually two different formulations
of the same theory was not immediately clear. Around the same time as Schrödinger formulated
his wave-theory, Dirac developed an abstract formulation of quantum theory, which unifies both
Heisenberg and Schrödinger picture. All three of them received Nobel prizes for the development
of quantum theory (Heisenberg in 1932, Dirac and Schrödinger jointly in 1933).
The development of the theory was very exciting for everyone involved. Legend has it that in
the summer of 1925 Heisenberg was thinking hard about quantum mechanics but didn’t make much
progress. In addition he was suffering from bad hay fever, and decided to escape to Helgoland, an
island in the northern sea. There he went for long hikes, read poetry, and also finally found his
quantum theory! Later he wrote about this in his autobiography: It was about 3 o’clock at night
when the final result of the calculation lay before me. At first I was deeply shaken. I was so excited
that I could not think of sleep. So I left the house and awaited the sunrise on the top of a rock. He
later sent his results to Wolfgang Pauli, with a letter saying: Everything is still vague and unclear
to me, but it seems as if the electrons will no longer move on orbits...
6 CHAPTER 1. CLASSICAL MECHANICS
Chapter 2

Schrödinger dynamics

In this chapter we provide a first introduction to Schrödinger’s wave equation, the meaning and
interpretation of the wave function etc. Much of the content will be formalised in chapter 4.

2.1 The Schrödinger equation and the wave function

In parallel to Heisenberg’s formulation of quantum mechanics Schrödinger developed a description
in terms of what is known as the wave function. He was inspired by Einstein’s and de Broglie’s
ideas of wave-particle-duality and matter waves. In modern form this time-dependent Schrödinger
equation is written as
i~ψ̇(~r, t) = Ĥψ(~r, t), (2.1)
where the dot denotes the derivative with respect to time. Here Ĥ is the Hamiltonian, a differential
h
operator acting on the wave function ψ, and ~ = 2π , where h ≈ 6.626×10−34 Js is Planck’s constant.
The Hamiltonian encodes the total energy of the system and is often of the form

~2 2
Ĥ = − ∇ + V (~r), (2.2)
2m
∂2
where ∇2 denotes the Laplacian, i.e. in three dimensions in Cartesian coordinates, ∇2 = ∂x21
+
∂2 ∂2
∂x22
+ ∂x23
.
The first term encodes the kinetic and the second the potential energy. We will often
consider one-dimensional systems for which the Schrödinger equation with a Hamiltonian of the
form (2.2) takes the form

~2 ∂ 2
 
i~ψ̇(x, t) = − + V (x) ψ(x, t). (2.3)
2m ∂x2

Remark 7. In the context of quantum mechanics, we usually denote the spatial coordinate of one-
dimensional systems, the argument of the wave function, by x.

2.1.1 The wave function

The wave function ψ ∈ C is interpreted as a probability amplitude. The square modulus of the wave
function is a probability density. Specifically, |ψ(x, t)|2 dx is the probability of finding a quantum
particle (described
R ∞by the wave function ψ) between x and x+dx at timeRt∞(assuming ψ is normalised
to one, that is, −∞ |ψ(x, t)|2 dx = 1.). The normalisation condition −∞ |ψ(x, t)|2 dx = 1 means
that the particle is somewhere with probability one. The probability to find the particle in the
interval A = [a, b] is given by
Z b
|ψ(x)|2 dx. (2.4)
a

7
8 CHAPTER 2. SCHRÖDINGER DYNAMICS

The wave function encodes everything that one can possibly know about a quantum particle. Know-
ing ψ(x, t0 ) one can deduce ψ(x, t) for all times. That is, the time evolution in quantum mechanics
is deterministic. R∞
To interpret |ψ(x, t)|2 as a probability distribution we need −∞ |ψ(x, t)|2 dx to be finite. That
is, ψ(x, t) ∈ L2 (R) (we refer to this as normalisable or square integrable).
Remark 8. We will mostly consider wave functions that are in L2 , but the theory can be slightly
extended to include, for example, functions such as plane waves, ψ ∝ eikx . We shall occasionally
use such functions, and fill in a minimum of interpretational background when needed.
Remark 9. While it is meaningful to consider all square integrable functions as wave functions,
as they lead to a probability density, there are other motivations that might restrict the space
of admissible wave functions to certain subspaces of L2 . For example, the right-hand side of the
Schrödinger equation (2.3) is only defined for functions ψ(x) that are at least twice differentiable,
which is not the case for all square integrable functions.
Exercise 3. Consider the wave function
ψ(x, t) = Ae−λ|x| e−iωt ,
where A, λ, ω are real positive constants.
(a) Find the value of A for which ψ is normalised to one.
(b) What is the probability to find the particle described by ψ(x, t) in the interval [− λ1 , λ1 ] ?

2.1.2 Probability current and conservation of probability

To interpret a time-dependent wave function that is a solution of the time-dependent Schrödinger
equation as a probability amplitude, the overall norm should stay conserved over time. Let us now
check whether this is indeed the case. We consider the total probability
Z ∞
|ψ(x, t)|2 dx (2.5)
−∞

as a function of time, assuming the Hamiltonian is of the form (2.2). In this context we can only
consider wave functions that are at least twice differentiable. In this case, the derivative with respect
to time and the integral over the coordinate x commute, that is, we have
d ∞
Z Z ∞
2 d
|ψ(x, t)| dx = |ψ(x, t)|2 dx
dt −∞ −∞ dt
Z ∞
= ψ̇ ∗ (x, t)ψ(x, t) + ψ ∗ (x, t)ψ̇(x, t)dx, (2.6)
−∞
∗
where the asterisk denotes complex conjugation. Inserting the Schrödinger equation and its
conjugate yields
d ∞ i ∞ ~2 00 ∗
Z Z
2
|ψ(x, t)| dx = V (x)ψ ∗ (x, t)ψ(x, t) − ψ (x, t)ψ(x, t)dx
dt −∞ ~ −∞ 2m
i ∞ ~2 ∗
Z
00
− V (x)ψ ∗ (x, t)ψ(x, t) − ψ (x, t)ψ (x, t)dx
~ −∞ 2m
Z ∞
i~ 00 00
= ψ ∗ (x, t)ψ (x, t) − ψ ∗ (x, t)ψ(x, t)dx, (2.7)
2m −∞
where we have made use of the fact that the potential V (x) is a real function. Now we recognise
the integrand as the first derivative of a function which is proportional to the so-called probability
current or probability flux, defined as
~  ∗ 0 0

j(x, t) := ψ (x, t)ψ (x, t) − ψ ∗ (x, t)ψ(x, t) . (2.8)
2mi
2.2. STATIONARY SOLUTIONS: THE TIME-INDEPENDENT SCHRÖDINGER EQUATION9

We can integrate to find

Z ∞
d x=∞
|ψ(x, t)|2 dx = − j(x, t) x=−∞ .

(2.9)
dt −∞

That is, if j(x) tends to zero for x → ±∞ then the overall probability is indeed conserved in time.
This condition is fulfilled as long as ψ(x) tends to zero at infinity and its first derivative is bounded,
in this case, the overall probability is conserved in time.
Remark 10. For those who are interested in the mathematical subtleties: we shall return to the
discrepancy of the different spaces of admissible wave functions in later chapters. For those who do
not wish to dwell too deeply into subtleties, it is of little (or no) harm to take the attitude of most
physics textbooks, that any physically meaningful wave function is a square integrable function
that also fulfils all the necessary conditions to make it normalisable, make sense of both sides of the
Schrödinger equation, and conserve the norm over time.

2.1.3 Superposition principle

The Schrödinger equation is a linear differential equation. Thus, if ψ1 (x, t) and ψ2 (x, t) are solutions,
and c1 , c2 ∈ C are constants, then c1 ψ1 (x, t) + c2 ψ2 (x, t) is also a solutions (Verify this!). This
perhaps trivial sounding fact plays a central role in quantum theory. On a formal level it makes
quantum mechanics much “easier” than classical mechanics.

2.1.4 The Hamiltonian and the energy

A quantum system is characterised by its Hamiltonian Ĥ, an operator acting on the wave function,
whose expectation value, defined as
R∞ ∗
−∞ ψ (x, t)Ĥψ(x, t)dx
hĤi := R∞
2
, (and analogously for higher dimensions) (2.10)
−∞ |ψ(x, t)| dx

gives the total energy of the system.

For a Hamiltonian of the type (2.2), consisting of potential and kinetic energy, if the potential
fulfills V (x → ±∞) → ∞, all eigenvalues (eigenenergies) of the Hamiltonian, are real numbers,
and the corresponding eigenfunctions (also referred to as eigenstates) form an orthonormal basis
for all possible wave functions of the system. The eigenstates φE (~r) are the solutions of the time-
independent Schrödinger equation
ĤφE (~r) = EφE (~r). (2.11)
Here orthonormality means that
Z
φ∗E (~r)φE 0 (~r)dV = δEE 0 . (2.12)
all space

Remark 11. For potentials that are bounded (and don’t go to infinity at infinity), the situation is
more complicated. Here, typically the spectrum consists of discrete eigenvalues and a continuum
part.

2.2 Stationary solutions: The time-independent Schrödinger equa-

tion
Consider special solutions of the time-dependent Schrödinger equation of the form

ψ(x, t) = φ(x)χ(t). (2.13)

10 CHAPTER 2. SCHRÖDINGER DYNAMICS

Inserting this ansatz into the Schrödinger equation yields

χ̇(t) Ĥφ(x)
i~ = . (2.14)
χ(t) φ(x)

Since the left side of this equation is independent of x, so has to be the right side. Similarly, as the
right side is independent of t, so is the left side, and in fact, both sides have to be constant. We
call the constant E and separate the original equation into two differential equations

i~χ̇(t) = Eχ(t) (2.15)

Ĥφ(x) = Eφ(x). (2.16)

Remark 12. This method is known as separation of variables.

Equation (2.15) is easily solved by direct integration to yield χ(t) = χ(0)e−iEt/~ . Equation
(2.16) is just the time-independent Schrödinger equation, that is, the eigenvalue equation for Ĥ.
The functions φE (x) are the eigenfunctions of Ĥ to the eigenvalues E. They are also referred to as
eigenstates. We will solve some examples of this, especially in chapter 7. The eigenvalue equation
can only be solved analytically for a small number of textbook cases.
~ ∂ 2 2
Remark 13. For Hamiltonians of the form − 2m ∂x2
+ V (x), where the potential has the property
V (x → ±∞) → ∞ normalisable solutions to the time-independent Schrödinger equation only exist
for special discrete values of E. These discrete energies are the reason for the discrete atomic
spectra, that is, the distinct colours that are emitted by stimulated atoms (think for example of
fluorescent lamps).
The time-dependence of separable wave functions is rather trivial:

ψ(x, t) = χ0 e−iEt/~ φE (x). (2.17)

In particular, if the initial wave function at time zero is an eigenfunction of Ĥ we have χ0 = 1 and

ψ(x, t) = e−iEt/~ φE (x), (2.18)

that is, an initial eigenstate evolves in time only by acquiring a phase factor, which means for the
probability density
|ψ(x, t)|2 = |φE (x)|2 . (2.19)
Hence, these wave functions are called stationary states.
Linear superpositions of stationary states, however, do have a non-trivial time-dependence.
Consider for example an initial state which is a superposition of two eigenstates belonging to two
different energies E1 and E2 ,
ψ(x, 0) = aφE1 (x) + bφE2 (x).
The probability density of the time-dependent state is then given by

|ψ(x, t)|2 =|a|2 |φE1 (x)|2 + |b|2 |φE2 (x)|2

(2.20)
+ a∗ b ei(E1 −E2 )t/~ φ∗E1 (x)φE2 (x) + ab∗ ei(E2 −E1 )t/~ φE1 (x)φ∗E2 (x)

The last two terms are time-dependent and are often referred to as the interference terms. In the
special case where a, b ∈ R, and also the eigenfunctions are real valued, φEj ∈ R, this reduces to

|ψ(x, t)|2 = a2 |φE1 (x)|2 + b2 |φE2 (x)|2 + 2ab cos( E2 −E

~
1
t) φE1 (x)φE2 (x), (2.21)

that is, the probability oscillates with a frequency determined by the energy difference of the two
states in the superposition.
2.3. THE METHOD OF STATIONARY STATES 11

2.3 The method of stationary states

Consider an initial state that is an arbitrary superposition of eigenstates of the Hamiltonian
X
ψ(x, t = 0) = an φn (x), (2.22)
n

with
Ĥφn (x) = En φn (x).

Due to the linearity of the Schrödinger equation, the time-dependent state is given by
X
ψ(x, t) = an e−iEn t/~ φn (x). (2.23)
n

Remark 14. That is the superposition principle again!

Given that the eigenstates of a Hamiltonian are a basis for all possible states of a system,
and the superposition principle, the time-evolution of an arbitrary initial state can be deduced by
representing it in the eigenbasis. This method is often called the method of stationary states.

2.3.1 The method of stationary states (for one dimensional systems in coordi-
nate representation)
Given: Hamiltonian Ĥ, initial wave function ψ(x, t = 0)
Wanted: Wave function at time t, ψ(x, t)
Method:

1. Calculate the eigenvalues En and normalised eigenstates φn (x) of the Hamiltonian Ĥ.

2. Expand the initial wave function in the basis of eigenstates {φn (x)}:

X
ψ(x, 0) = an φn (x).
n
R∞ ∗
The coefficients are given by an = −∞ φn (x)ψ(x, 0)dx.

3. The time evolution is then given by

X
ψ(x, t) = an e−iEn t/~ φn (x).
n

2.4 Example: Particle in a half box

(From David Tannor’s Introduction to Quantum Mechanics)

A standard system in quantum mechanics is the so-called particle in a box. In one dimension that
is a particle moving freely on a certain interval on the real line, here we consider the interval [0, L].
~2 ∂ 2
The Hamiltonian of the system is given by Ĥ = − 2m ∂x2
+ V (x), where there is an infinite potential
outside this interval, i.e.

0, 0 ≤ x ≤ L
V (x) =
∞, otherwise.
12 CHAPTER 2. SCHRÖDINGER DYNAMICS

We shall discuss the solution of the time-

independent Schrödinger equation for this system
in chapter 7.3. For now we simply provide the re-
sult: The eigenvalues of the Hamiltonian are given
by
n2 π 2 ~2
En = , (2.24)
2mL2
and the corresponding eigenfunctions are
( q
2
sin nπx


L L , 0≤x≤L
φn (x) = (2.25)
0, otherwise.

The first 5 eigenfunctions are schematically de-

picted at the height of their corresponding ener-
gies in figure 2.1.
Figure 2.1: Eigenvalues and eigen-
Remark 15. This is analogous to the problem of functions of a particle in a box.
standing waves on a string.
Consider a quantum particle in the lowest eigenstate of a box potential of length L. Assume
now that at time t = 0 the box is suddenly expanded to the length 2L. If this is done infinitely fast,
the wavefunction is initially not altered. However, the wavefunction is now not an eigenfunction of
the new system, and thus it dynamically evolves for times t > 0.
Remark 16. Such a fast change of the external potential in an experiment is often called a quantum
quench, a method often employed to experimentally prepare a system in a state other than an
eigenstate.
We can use the method of stationary states to understand the dynamics of the wave function.
It turns out that the motion of the wave function while being rather complicated in its details,
is periodic. Let us use the method of stationary states to calculate the shortest time τ at which
ψ(x, τ ) = ψ(x, 0). For this purpose we need to expand the wave function at time zero, which is a
normalised eigenfunction of the square well of length L,
(L)
ψ(x, 0) = φ1 (x), (2.26)
(2L)
in the basis of eigenfunctions φn (x) of the square well of twice the size, that is, we write
(L)
X
φ1 (x) = an φ(2L)
n (x). (2.27)
n

To calculate the actual coefficients an we would have to calculate the integrals

Z ∞ √ Z L
∗(2L) (L) 2  nπx   πx 
an = φn (x)φ1 (x)dx = sin sin dx. (2.28)
−∞ L 0 2L L
However, to calculate τ we do not need to know all the values of an . The wave function at time t
has the form X (2L)
ψ(x, t) = an e−iEn t/~ φ(2L)
n (x). (2.29)
n
Inserting the energy values for the square well of width 2L, we have
π2 ~ 2
an e−i 8mL2 n t φ(2L)
X
ψ(x, t) = n (x). (2.30)
n

Or
2
X
ψ(x, t) = an e−iωn t φ(2L)
n (x), (2.31)
n
2.4. EXAMPLE: PARTICLE IN A HALF BOX 13

2 π ~
where we have defined ω = 8mL 2. Thus, we have ψ(x, τ ) = ψ(x, 0) at times τ for which all
−iωn2 τ 2
e = 1, which is true if all n ωτ are integer multiples of 2π. Since n are integers here, the
shortest such time τ thus fulfils
2π 16mL2
τ= = . (2.32)
ω ~π
14 CHAPTER 2. SCHRÖDINGER DYNAMICS
Chapter 3

Excursion: Mathematical background

After having gained some first impressions of quantum behaviour looking at Schrödinger’s wave
functions, we shall now move to a more abstract and versatile formulation of quantum mechan-
ics, that will allow us to identify which questions can be answered by quantum mechanics, and
which questions the theory does not “allow” us to ask. In particular in chapter 4 we will learn
five principles/postulates of quantum mechanics that form the basis of the whole theory and its
interpretations. Several mathematical concepts and notations are needed to formulate and under-
stand these principles, and in the present chapter we provide the minimum background. Let us
first, however, have a brief look into the first two postulates that we will discuss in more detail in
chapter 4 to get an idea which mathematical background is needed and why.

3.1 States and obervables - Hilbert spaces and Hermitian opera-

tors
The main ingredients for this theoretical framework of quantum mechanics are the concepts of
a state and of observables. A state encodes all the information one can have about a quantum
system. Observables are measurable quantities (such as, for example, the position of a particle
or its kinetic energy). In some ways, the same concepts appear in classical mechanics, but we so
naturally associate the mathematical objects that describe these properties with the actual physical
properties that they are seldom talked about.
Looking back at Hamilton’s classical mechanics we see that all measurable quantities, i.e. ob-
servables are given by real phase-space functions, that is, real functions of the canonical variables.
The canonical variables, in turn, are the classical state. We implicitly consider this state to be a
property of the system (for example a particle) in classical physics, and talk about the position of
an object or the momentum of a particle, in similar terms as when considering its mass, colour or
other inherent property. In quantum mechanics we will have to let go of this.
The state of a quantum system, as defined above, encodes all we can in principle measure about
a particle, but it makes no claim about measurable quantities being properties of a particle. We
have already encountered this in the last chapter where we have seen that it makes little sense in
Schrödinger’s framework of quantum mechanics to talk about a particle’s position. The only thing
we can talk about is the probability of finding a particle in a certain region at a given time. Think
this over - it is the essence of quantum mechanics!
As Philip Ball argues in his beautiful recent book ”Beyond weird”, this is neither weird nor
strange. It does, however, require a shift away from century-old ways of thinking about the nature
of things. This allows us to predict a lot of experimental observations with unprecedented accuracy,
and to design technological devices such as transistors and lasers, that would be impossible on the
basis of classical physics. A lot can be said (and over the decades a lot has been said, indeed) about
the philosophical aspects of quantum theory. Here, however is not the place to do so. Let us instead
focus on the mathematical framework and on what we can calculate with it.

15
16 CHAPTER 3. EXCURSION: MATHEMATICAL BACKGROUND

As we shall see, compared to classical mechanics, the role of position and momentum is shifted
from the state to observables. The role of the state is taken over by another mathematical object
that does not appear explicitly in classical physics. That brings us to
The first principle of quantum mechanics: The state of a quantum system is described by a
non-zero vector in a Hilbert space.
Observables are covered by
The second principle of quantum mechanics: A measurable quantity is described by a Hermi-
tian operator on the Hilbert space. Both the measurable quantities and the operators representing
them are often referred to as observables.

With this in mind in this chapter we shall review some general concepts from linear algebra
and analysis, and in particular provide a minimal background on Hilbert spaces and Hermitian
operators. We shall also learn about a very powerful notation for vectors and operators on Hilbert
spaces that has been developed by Dirac especially for quantum mechanics, aptly called the Dirac
notation, that we shall make use of in the rest of the course.

3.2 Some linear algebra and analysis

Here we summarise a few definitions from linear algebra that are essential in the abstract formulation
of quantum mechanics.
Complex vector space: A set of elements (vectors) V , forming an Abelian (i.e. commutative)
group with respect to addition u + v, and a scalar multiplication C × V → V , satisfying
(λ + µ)v = λv + µv
(λµ)v = λ(µv)
λ(u + v) = λu + λv
1v = v.
for all λ, µ ∈ C and u, v ∈ V , is called a vector space.
Remark 17. In general a vector space can be defined over any field F, where the scalars are
elements of F. Most commonly encountered in physics are vector spaces over the real or the
complex numbers.
Inner product space: An inner product space is a vector space V together with a map V ×V → C,
written as (u, v), such that for all u, v, w ∈ V and λ, µ ∈ C:
(u, v) = (v, u)∗
(u, u) ≥ 0, and (u, u) = 0 ⇔ u = 0
(u, λv + µw) = λ(u, v) + µ(u, w).

Linear (in)dependence, basis, dimension: A set of vectors {un } is said to be linearly indepen-
dent if no nontrivial linear combination of them sums to zero. If the set is on the contrary
linearly dependent,
P one can express one member of the set as a linear combination of the
others: uk = n6=k αn un .
The maximum number of linearly independent vectors in a space is called the dimension of
the space.
A maximal set of linearly independent vectors is called a basis. That is, every
P vector in the
space can be represented as a linear combination of the basis vectors: φ = n φn un . The φn
are the coefficients or components of φ in the basis {un }. One can represent φ as an N-tupel
of complex numbers, (φ1 , ... , φN ) ∈ CN , where N is the dimension of the vector space.
3.2. SOME LINEAR ALGEBRA AND ANALYSIS 17

Norm: The norm ||v|| of a vector can be defined via ||v||2 = (v, v). A vector is called normalised
if ||v|| = 1.

Orthonormality: Two vectors u, v are called orthogonal if (u, v) = 0. A set of vectors {un } is
said to be orthonormal if each vector is normalised and the vectors are pairwise orthogonal.
An orthonormal maximal set of linearly independent vectors is an orthonormal basis.

Coefficients: The coefficients of a vector φ in an orthonormal basis {un } can be found via the
orthonormality condition as φn = (un , φ).

Hilbert space: A Hilbert space H is a complex inner product space which is complete with respect
to the norm induced by the inner product.

Completeness of a vector space: A vector space V is complete if the following holds. Whenever
{un ∈ V } is a sequence of vectors such that for any positive  there exists an integer N with

||um − un || < , for all m, n > N ,

then there exists a limit vector u ∈ V , such that

||un − u|| → 0.

A finite dimensional vector space is automatically complete, i.e. every finite dimensional
vector space is a Hilbert space.

Separable Hilbert space: A Hilbert space is called separable if it possesses a countable basis. In
quantum mechanics, we deal only with separable Hilbert spaces.

Linear Operator: An operator on a vector space is a map V → V . A linear operator A satisfies

A(λu + µv) = λ(Au) + µ(Av),

for all λ, µ ∈ C, and u, v ∈ V .

A linear operator on a finite dimensional vector space can be represented by a matrix.

Bounded operators An operator A on a Hilbert space is bounded if and only if there exists a real
number c > 0, such that for all u ∈ H,

||Au|| ≤ c||u||.

Domain The domain D(A) ∈ H of an operator A is the set of vectors on which A is bounded, i.e.,
the set of vectors u ∈ H for which Au ∈ H.

Adjoint Operator: The adjoint A† of a linear operator A is defined by the relation

(A† u, v) = (u, Av), for all v ∈ D(A) and u ∈ D(A† ).

The adjoint fulfils the following properties:

(cA)† = c∗ A†
(A + B)† = A† + B †
(AB)† = B † A† .
18 CHAPTER 3. EXCURSION: MATHEMATICAL BACKGROUND

Hermiticity and Self-adjointness: A linear operator defined on a Hilbert space H is called Her-
mitian
(Au, v) = (u, Av), for all v, u ∈ D(A),
that is if A† = A on the overlap of the domains of A and A† .
For bounded operators the domain is the whole Hilbert space, and Hermiticity leads to many
useful properties. For unbounded operators, on the other hand, Hermiticity is not enough to
guarantee some crucial properties we need in quantum mechanics. Here the stronger concept
of self-adjointness is important. An operator A is self-adjoint if A = A† and D(A) = D(A† ).
For more details, see, e.g. Ballentine, Cohen Tannoudji, or Brian Hall’s “Quantum mechanics
for Mathematicians”.
Eigenvalues and eigenvectors: If an operator acting on a certain vector produces a scalar mul-
tiple of that same vector,
Au = au,
we call the vector u an eigenvector and the scalar a an eigenvalue of the operator A.
If A is an N × N matrix the eigenvalue equation only has nontrivial solutions if

det(A − a1) = 0.

This yields the characteristic polynomial of A, which is a polynomial of N th order in a.

3.3 The Dirac notation

3.3.1 The dual space and the Riesz theorem
To any linear vector space V there exists a dual space of linear functionals on V . A linear functional
F assigns a scalar F (φ) to each vector φ, such that

F (αφ + βψ) = αF (φ) + βF (ψ), (3.1)

for α, β ∈ C and φ, ψ ∈ V .
The set of linear functionals is itself a linear vector space V 0 with the definition

(F1 + F2 )(φ) = F1 (φ) + F2 (φ). (3.2)

Theorem 1 (Riesz representation theorem). A complex Hilbert space H is anti-isomorphic to its

dual space H0 . That is, there exists a one-to-one correspondence between linear functionals F and
vectors f , such that all linear functionals can be represented in the form F (φ) = (f, φ) with a fixed
vector f and arbitrary vectors φ.
We shall proof the Riesz theorem for separable Hilbert spaces here, making use of the fact that
they possess a countable orthonormal basis. Clearly every vector defines a functional via the inner
product. The proof for the opposite direction, that every functional can be represented as an inner
product with a fixed vector, works by construction:
P
Proof. Let {φn } be an orthonormal basis of H. Let ψ = n ψn φn with ψn ∈ C be an arbitrary
vector in H. Then X
F (ψ) = ψn F (φn ). (3.3)
n
P ∗
Defining the vector f = n [F (φn )]φn we see that
X
(f, ψ) = F (φn )ψn = F (ψ) (3.4)
n

for all ψ ∈ H.
3.3. THE DIRAC NOTATION 19

3.3.2 The Dirac notation

Vectors in H are called ket vectors, written as |φi, vectors in the dual space H0 (functionals) are
called bra vectors, written as hf |. In this way we have

F (φ) = (f, φ) = hf |φi, (3.5)

where the last expression is a “bra-ket” – hence the name.

Remark 18. On some level one can think of this Dirac notation simply as a notation for scalar
products, but it turns out to be useful far beyond this.
There is an anti-linear correspondence between bras and kets due to the nature of the scalar
product:
α∗ hf1 | + β ∗ hf2 | ←→ α|f1 i + β|f2 i. (3.6)
Recalling the definition of the adjoint operator † of a linear operator  we see that the bra
associated to the ket vector Â|φi is the vector hφ|† . That is, we can interpret the expression Â|φi
as “|Âφi” and the expression hφ|† as “hÂφ|”. That is, the notation hφ|Â|ψi denotes two equal,
but morally different, things:
(i) The inner product of the vector φ with the vector resulting from acting with  on the vector
ψ, i.e. “hφ|Âψi”;

(ii) The inner product of the vector resulting from acting with † onto the vector φ with the
vector ψ, i.e. “h† φ|ψi”.
The definition of the adjoint operator is expressed in Dirac notation as

hφ|Â|ψi = hψ|† |φi∗ . (3.7)

3.3.3 Rules for finding the adjoint of an expression

To find the adjoint of an expression, one needs to
• complex conjugate scalars c ↔ c∗ ,

• replace kets with bras and vice versa |φi ↔ hφ|,

• replace operators with their adjoints  ↔ † ,

• reverse the order of factors.

As an example, consider the expression

c|ψihφ|A|ψihφ|.

This expression is an operator. The adjoint operator is given by

c∗ |φihψ|† |φihψ|. (3.8)

3.3.4 The outer product of two vectors

Besides the inner product of a bra hφ| and a ket |ψi, hφ|ψi, which is a scalar, there is also an outer
product, |ψihφ|, which is a linear operator.
Exercise 4. What is the adjoint operator of |ψihφ|?
Remark 19. An operator of the form P = |φihφ| is a projection operator, as the mapping P : |ψi 7→
hφ|ψi|φi is the projection of |ψi onto |φi. We will return to a more general notion of projection
operators in the discussion of principle 4 in chapter 4.
20 CHAPTER 3. EXCURSION: MATHEMATICAL BACKGROUND

3.3.5 Completeness and resolution of identity

An orthonormal set of vectors {φn } is complete if an arbitrary |vi ∈ V can be written as
X
|vi = vn |φn i. (3.9)
n

The coefficients vn are given by

vn = hφn |vi. (3.10)
Reinserting this into the expansion (3.9) of |vi gives the identity
X
|vi = |φn ihφn |vi. (3.11)
n

This, however, is identical to the statement

!
X
|vi = |φn ihφn | |vi. (3.12)
n

Since this is valid for all vectors v ∈ V , that means that

X
|φn ihφn | = 1. (3.13)
n

This property which is fulfilled by any complete orthonormal set, is called the resolution of identity.

3.4 More on Hermitian operators

Theorem 2. All eigenvalues of a Hermitian operator are real.

Proof. Consider the (non-zero) eigenvector |ui of a Hermitian operator  = † , corresponding to
the eigenvalue a 6= 0:
Â|ui = a|ui.
The adjoint equation reads
hu|Â = hu|a∗ .
We have
hu|Â|ui = hu|a|ui = ahu|ui.
On the other hand we have  = † , and thus

hu|Â|ui = hu|† |ui = hu|a∗ |ui = a∗ hu|ui.

It follows that
ahu|ui = a∗ hu|ui. (3.14)
Since hu|ui =
6 0 it follows that
a = a∗ ∈ R.

Theorem 3. Eigenvectors of a Hermitian operator belonging to different eigenvalues are orthogonal.

Exercise 5. Prove theorem 3.

Theorem 4. The expectation values, hv|Â|vi, of Hermitian operators are real. On the other hand,
if every expectation value of an operator is real, the operator is Hermitian.
3.5. VECTOR/MATRIX REPRESENTATION 21

We will prove theorem 4 in the lecture.

Finally, without proof, we state the following important

Theorem 5. The eigenvectors of a Hermitian operator form a complete orthogonal basis set for
the Hilbert space the operator acts upon.

An operator A is called anti-Hermitian if † = −Â. If  is Hermitian then i is anti-Hermitian.

Every operator can be uniquely decomposed into a Hermitian and an anti-Hermitian part, Â =
Âh + iÂa , with †h = Âh and †a = Âa .

3.5 Vector/matrix representation

Consider a Hilbert space H with a countable orthonormal basis {|un i}. The ket vectors |ψi ∈ H
can be represented by (possibly infinite) column vectors of their coefficients in this basis:
 
a1
a2 
 
X  .. 
|ψi = an |un i, |ψi '   .  =: ~a.

n  aj 
 
..
.
P P
P scalar∗ product of two vectors |ψi =
For the n an |un i, and |φi = n bn |un i it follows from
hφ| = n hun |bn that
X X
hφ|ψi = hum |b∗m an |un i = b∗m an hum |un i
m,n m,n
X X
= b∗m an δmn = b∗n an
m,n n

= (~b∗ )T ~a .

Similarly linear operators on the Hilbert space can be represented by matrices. Consider, e.g., the
equation M̂ |ψi = |φi, that is X X
an M̂ |un i = bn |un i.
n n

Operating on this with hum | yields

X X
an hum |M̂ |un i = bn hum |un i
n n
X X
an hum |M̂ |un i = bn δmn
n n
X
Mmn an = bm ,
n

which is valid for all m. Thus,

M̂ |ψi = |φi ⇔ M~a = ~b,
with M̂ = (Mmn ) and Mmn = hum |M̂ |un i. In matrix representation Hermiticity means that
Mmn = Mnm∗ .
22 CHAPTER 3. EXCURSION: MATHEMATICAL BACKGROUND
Chapter 4

The Principles of Quantum Mechanics

In this chapter we will learn about the five principles/postulates of quantum mechanics that allow
an abstract formulation of the theory. The first two principles that we have already briefly encoun-
tered in the previous chapter, concern the description of the states of a quantum system and its
observables, that is, its measurable quantities. The third and fourth principle concern the concept
of measurements and the final one the time-evolution of quantum systems.

4.1 States and observables

A state encodes all the information one can have about a quantum system. Observables are mea-
surable quantities (such as, for example, the position of a particle or its kinetic energy).

4.1.1 Principle 1 (States):

The state of a quantum system is described by a non-zero vector in a Hilbert space.
Remark 20. The space of square-integrable functions L2 that we have dealt with in Schrödinger’s
theory of wave functions is a Hilbert space.
The superposition principle means that the sum of two states is a state, which also follows from
the vector space properties. Two vectors describe the same state if they are complex multiples of
each other. The equivalence class ψ → cψ with c ∈ C defines rays through the ”origin” of the
Hilbert space. The space of these rays is called the projective Hilbert space.
Remark 21. Some caution is needed when dealing with the concept of the projective Hilbert space;
while the states eiφ1 ψ1 and eiφ2 ψ2 are equivalent to the states ψ1 and ψ2 respectively, in general
eiφ1 ψ1 + eiφ2 ψ2 describes a different state from ψ1 + ψ2 .

4.1.2 Principle 2 (Observables):

A measurable quantity is described by a Hermitian operator on the Hilbert space. Both the mea-
surable quantities and the operators representing them are often referred to as observables.
In particular the operators corresponding to position and momentum, q̂ and p̂, fulfil the com-
mutation relation
[q̂, p̂] = i~1, (4.1)

where the commutator [·, ·] is defined as

[Â, B̂] = ÂB̂ − B̂ Â, (4.2)

and 1 denotes the identity operator.

23
24 CHAPTER 4. THE PRINCIPLES OF QUANTUM MECHANICS

4.2 Measurement and dynamics

The first two principles state that a quantum state is a vector and that measurable quantities are
represented by operators. Now we will focus on the actual measurement and how the outcome
of a measurement is related to the operator describing an observable and the state describing the
system.

4.2.1 Principle 3 (“Born rule”):

The possible outcomes of a measurement of an observable A are given by the eigenvalues of the
corresponding operator Â. The probability of obtaining an eigenvalue aj as a measurement outcome
is encoded in the state |ψi the system is in immediately before the measurement. The probability is
given by p(aj ) = hψ|P̂j |ψi, where P̂j is the projection operator onto the subspace associated to the
eigenvalue aj . For non-degenerate, discrete eigenvalues this reduces to the modulus square |ψj |2 of
the coefficient of the (normalised) wave function |ψi in the basis of eigenstates |φj i of Â:
X
|ψi = ψn |φn i, with Â|φn i = an |φn i. (4.3)
n

For continuous spectra we replace the sum with an integral: The probability to obtain a value
in the interval I in a measurement is theR integral over the modulus squared of the components of
the wave function over this interval P = I |ψ(α)|2 dα, where
Z
|ψi = dα ψ(α)|φ(α)i, with Â|φ(α)i = α|φ(α)i. (4.4)

4.2.2 Projection operator

A linear operatorPP̂ that satisfies P̂ 2 = P̂ is called a projection operator. Examples are operators
of the form P̂ = N n=1 |φn ihφn | where {|φn i} is an orthonormal set of vectors. We can easily verify
that indeed P̂ 2 = P̂ :
X X X
P̂ 2 = |φn ihφn | |φm ihφm | = |φn ihφn |φm ihφm |
n m n,m
X X (4.5)
= δmn |φn ihφn | = |φn ihφn | = P̂
n,m n

Exercise 6. Verify that the eigenvalues λ of a projection operator are either one or zero.

Example - Energies of a particle in a box of length L

2 2 2
The eigenvalues of the Hamiltonian are given by En = n2mL π ~
2 . If the state is given by |ψi =
P
n ψn |φn i with Ĥ|φn i = En |φn i we will obtain the value En in an energy measurement with
probability |ψn |2 . If the system is in the “ground state” (that is, the state with the lowest energy,
π 2 ~2
in this case E1 ), for example, we will obtain the value E1 = 2mL 2 with probability one.

Expectation values
The mean value of the outcome of many measurements of an observable A on identically prepared
systems does not have to be an eigenvalue of the operator Â. It is given by the expectation value of
 in the state |ψi:
hψ|Â|ψi
hÂi := . (4.6)
hψ|ψi
Let us verify this for the case of discrete eigenvalues: Let us denote the eigenvalues of  by an
and the corresponding eigenvectors by |φn i, that is, Â|φn i = an |φn i. The eigenvectors belonging
4.2. MEASUREMENT AND DYNAMICS 25

to different eigenvalues are automatically orthogonal, in addition we can choose them normalised.
The mean value of many measurements is clearly given by
X
hÂi = aj P (aj ), (4.7)
j

where P (aj ) denotes the probability that a measurement yields the outcome aj . Let us now verify
that this is indeed equal to the expression (4.6), using that according to principle 3 P (aj ) is given
by
P (aj ) = |hφj |ψi|2
(4.8)
= hψ|φj ihφj |ψi,

if the system is in the normalised state |ψi. Inserting expression (4.8) for the probability of the
measurement outcome aj into equation (4.7) yields
X
hÂi = aj hψ|φj ihφj |ψi
j
X (4.9)
= hψ|Â|φj ihφj |ψi.
j
P
Now we recognise the resolution of identity j |φj ihφj | = 1 and thus we have

hÂi = hψ|Â|ψi, (4.10)

as expected.

4.2.3 Principle 4 (“state collapse postulate”):

Consider a system in the quantum state |ψi on which a measurement of an observable A is performed.
If the measurement yields the outcome aj (an eigenvalue of Â), then the system is in the state
√ P̂j |ψi immediately after the measurement. Here P̂j is the projection operator onto the subspace
hψ|P̂j |ψi
belonging to the eigenvalue aj . For a non-degenerate eigenvalue with eigenvector |φj i that is P̂j =
|φj ihφj |, which means the state afetr the measurement will be the state φj i.
Remark 22. In some ways principle 4 is a logical extension of principle 3: If we were to perform
two measurements of the same observable instantly after one another, we would expect to obtain
the same outcome in both with probability one.
Remark 23. Having the state collapse postulate in mind one often uses the formulation “to find a
system in a state |φi” in quantum mechanics. Such as for example in “... calculate the probability
to find the system in the ground state.”.

4.2.4 Principle 5 (dynamics):

The time-evolution of a quantum state |ψ(t)i is governed by the Schrödinger equation

i~|ψ̇i = Ĥ|ψi, (4.11)

where the operator Ĥ is the operator corresponding to the total energy of the system.
The solution of the Schrödinger equation can formally be written as

|ψ(t)i = Û (t)|ψ(0)i, (4.12)

where Û (t) is the time-evolution operator. For time independent Hamiltonians it is given by
 
i
Û (t) = exp − Ĥt . (4.13)
~
26 CHAPTER 4. THE PRINCIPLES OF QUANTUM MECHANICS

Since the Hamiltonian Ĥ is Hermitian, the time-evolution operator Û (t) is unitary. That is,
Û −1 = Û † . (4.14)
Unitarity means that the inner product is conserved, i.e.
hψ1 (t)|ψ2 (t)i = hψ1 (0)|ψ2 (0)i, (4.15)
with |ψj (t)i = Û (t)|ψj (0)i.
Remark 24. Think of “rotations on Hilbert space”.

4.2.5 The method of stationary states in Dirac notation (for arbitrary Hilbert
spaces)
Given: Hamiltonian Ĥ, initial wave function |ψ(t = 0)i
Wanted: Wave function at time t, |ψ(t)i
Method:
1. Calculate the eigenvalues En and normalised eigenstates |φn i of the Hamiltonian Ĥ.
2. Expand the initial wave function in the basis of eigenstates {|φn i}:
X
|ψ(0)i = an |φn i.
n

The coefficients are given by an = hφn |ψ(0)i.

3. The time evolution is then given by
X
|ψ(t)i = an e−iEn t/~ |φn i.
n

4.3 Commutators and uncertainties

The commutator of two observables restricts how accurately they can be measured simultaneously.
In this chapter we shall discuss this in some detail. In this context we shall also learn about the
famous Heisenberg uncertainty principle.

4.3.1 The uncertainty relations

The outcomes of many measurements of an observable A on identically prepared systems have a
certain spread. The width of their distribution is characterised by the expectation value of the
operator (Â − hÂi)2 , that is, the variance

(∆A)2 := h(Â − hÂi)2 i

= hÂ2 − 2hÂi + hÂi2 i
(4.16)
= hÂ2 i − 2hÂi2 + hÂi2
= hÂ2 i − hÂi2 .
In quantum mechanics we refer to the standard deviation
q
∆A := hÂ2 i − hÂi2 , (4.17)
as uncertainty of A.
It follows from the principles of quantum mechanics that in general the product of the uncer-
tainties of two observables cannot be arbitrarily small. Instead, it has a lower bound that is related
to the commutator of the observables via the following
4.3. COMMUTATORS AND UNCERTAINTIES 27

Theorem 6. Consider the two Hermitian operators  and B̂, with [Â, B̂] 6= 0. The product of the
uncertainties of the observables A and B fulfils
1
∆A∆B ≥ |hĈi|, (4.18)
2

where we have defined Ĉ := i[Â, B̂].

Remark 25. The factor of i in the definition of Ĉ has been introduced such that Ĉ is Hermitian.
Proof. Let us shift the operators  and B̂ such that their mean values are zero, that is, let us
consider the operators
â := Â − hÂi, and b̂ := B̂ − hB̂i. (4.19)
These operators have the same commutator as  and B̂. That is,

[â, b̂] = −iĈ. (4.20)

Further we have
(∆A)2 = hâ2 i and (∆B)2 = hb̂2 i. (4.21)
Let us now consider the norm of the state |χi = (â − iλb̂)|ψi for λ ∈ R and an arbitrary normalised
state |ψi. We have

hχ|χi = hψ|(â + iλb̂)(â − iλb̂)|ψi

= h(â + iλb̂)(â − iλb̂)i
= hâ2 + λ2 b̂2 + iλ(b̂â − âb̂)i (4.22)
= hâ2 i + λ2 hb̂2 i + iλhiĈi
= hâ2 i + λ2 hb̂2 i − λhĈi

On the other hand, we know that hχ|χi ≥ 0. That is, we have

hâ2 i + λ2 hb̂2 i − λhĈi ≥ 0, (4.23)

for all values of λ ∈ R. Let us now evaluate for which value of λ the norm hχ|χi becomes minimal.
The location of the minimum is found where
d
hχ|χi = 2hb̂2 iλ − hĈi = 0, (4.24)
dλ
that is
hĈi
λ= . (4.25)
2hb̂2 i
(Note that this cannot be a maximum, since the function in (4.23) is a quadratic and goes to
infinity at λ → ±∞, and thus has a single minimum at a finite value of λ.) Reinserting this into
the condition hχ|χimin ≥ 0 yields

hĈi2 hĈi2
hχ|χimin = hâ2 i + hb̂2 i −
4hb̂2 i2 2hb̂2 i
(4.26)
2 hĈi2
= hâ i − ≥ 0,
4hb̂2 i

that is
hĈi2
(∆A)2 ≥ , (4.27)
4(∆B)2
28 CHAPTER 4. THE PRINCIPLES OF QUANTUM MECHANICS

or
|hĈi|
(∆A)(∆B) ≥ .
2
The special case of the uncertainty relation in theorem 6 for the case of position and momentum
operator is known as Heisenberg’s uncertainty relation:

~
∆p∆x ≥ . (4.28)
2
This means that one can never know the exact values of both position and momentum simultane-
ously.
Remark 26. The uncertainty relation gives a lower bound. There are states for which the actual
uncertainty product for position and momentum for example is much larger than ~.

4.3.2 Commuting observables

If two observables commute they are in principle simultaneously measurable, according to the
uncertainty relation. This is an aspect of the following

Theorem 7. Commuting operators [Â, B̂] = 0 posess a set of common eigenvectors.

Exercise 7. Prove theorem 7 for the case of non-degenerate eigenvalues. I.e., consider an eigen-
vector |φi of  belonging to the eigenvalue a ∈ R, that is Â|φi = a|φi, and show that |φi is also an
eigenvector of B̂ if [Â, B̂] = 0.

Remark 27. If the eigenvalue a is degenerate (that is, it is a multiple eigenvalue), we can find a
linear superposition of the corresponding eigenvectors that is an eigenvector of B̂.
Remember that if we measure an observable A, after the measurement the quantum state of
the system will have collapsed into the eigenspace of the operator  belonging to the eigenvalue
that had been measured. Now if we measure an observable B̂ directly afterwards that measurement
will project the state of the system onto an eigenspace of B̂. That means, only if  and B̂ share
eigenvectors, will the measurements be compatible, i.e. the measurement of one of the observables
won’t delete the information about the value of the first observable. Commuting operators have a set
of shared eigenvectors, so they can be measured simultaneously or without the order of measurement
changing the probabilities of the outcomes.

4.3.3 Complete set of commuting observables

Consider a set of Hermitian operators Â1,2,... on a Hilbert space H. Their simultaneous eigenvectors
form a basis of H. They are called a complete set of commuting observables if their eigenvalues
uniquely define each basis vector (up to a phase).
If an operator  has only non-degenerate eigenvalues it is already a complete set of commuting
observables on its own. If one or more eigenvalues is/are degenerate one can find a commuting op-
erator to lift, or partially lift the degeneracy, and add further such operators to lift the degeneracies
still present in this pair, and so on, until one arrives at a complete set of observables.
Let us consider a simple example for a three dimensional Hilbert space first. The operator
 
a1 0 0
 =  0 a2 0  , (4.29)
0 0 a3

with aj ∈ R, is clearly Hermitian, and indeed its eigenvectors are the standard vectors ej , which are
a basis of C3 , and could be uniquely labeled here by the eigenvalues of Â, i.e. ej is the eigenvector of
4.3. COMMUTATORS AND UNCERTAINTIES 29

 belonging to the eigenvalue aj . If, however, some of the eigenvalues were degenerate, for example
a2 = a1 , or, for clarity, let’s consider another operator B̂
 
b1 0 0
B̂ =  0 b1 0  , (4.30)
0 0 b2

the situation would be different. We can still find a set of eigenvectors that are an orthogonal basis
for C3 , for example still the standard vectors ej , or alternatively the vector e3 and two orthogonal
linear combinations of e1 and e2 . However, these basis vectors cannot be uniquely identified by
the corresponding eigenvalues of B̂, as both e1 and e2 belong to the same eigenvalue b1 . If we in
addition consider the operator  
c1 0 0
Ĉ =  0 c2 0  , (4.31)
0 0 c2

then the eigenvalues of Ĉ alone would also not be sufficient to identify the basis vectors, but those
of B̂ and Ĉ together would be. In the case here we could label the basis as

(b1 , c1 ) ↔ e1 , (b1 , c2 ) ↔ e2 , (b2 , c2 ) ↔ e3 . (4.32)

In infinite-dimensional Hilbert spaces it can be rather useful to have an abstract basis labelled
according to the eigenvalues of an important observable (for example the energy), and if this has
degenerate eigenvalues, it is useful to identify further commuting observables to uniquely label a
basis of the Hilbert space. We will come back to this in more concrete physical examples later in
the course.
30 CHAPTER 4. THE PRINCIPLES OF QUANTUM MECHANICS
Chapter 5

The harmonic oscillator

Let us now consider the quantum harmonic oscillator, the classical version of which we have studied
in 1.2. Similarly to the classical case, the Hamiltonian is given by

p̂2 1
Ĥ = + mω 2 q̂ 2 . (5.1)
2m 2

We want to solve the eigenvalue equation Ĥ|φn i = En |φn i. First of all we want to deduce the
eigenvalues.
Remark 28. One way of solving this problem would be to express the eigenvalue equation in the
~2 ∂ 2 φE (x)
Schrödinger formulation as − 2m ∂x2
+ 12 mω 2 x2 φE (x) = EφE (x), and determine for which values
of E this has bound state solutions that fulfil φE (x → ±∞) → 0. Here, however, we use another
method that does not require us to solve any differential equation, and instead uses only uses
algebraic arguments, based in the commutation relation of q̂ and p̂. .

5.1 Algebraic solution of the eigenvalue equation

Let us first introduce the dimensionless operators
r r
mω 1
Q̂ = q̂, and P̂ = p̂. (5.2)
~ mω~

They fulfil the commutation relation

1 ˆ
[Q̂, P̂ ] = [q̂, p̂] = iI. (5.3)
~
In terms of these rescaled operators the Hamiltonian takes the form

~ω  2 
Ĥ = P̂ + Q̂2 . (5.4)
2
Now we introduce the operator
1  
â = √ Q̂ + iP̂ , (5.5)
2
and its adjoint operator
1  
↠= √ Q̂ − iP̂ . (5.6)
2
These operators fulfil the commutation relation

[â, ↠] = I.
ˆ (5.7)

31
32 CHAPTER 5. THE HARMONIC OSCILLATOR

Exercise 8. Verify the commutation relation (5.7) using the commutator(5.3), and rewrite the
Hamiltonian in terms of â and ↠.
In terms of â and ↠the Hamiltonian becomes

Ĥ = ~ω(↠â + 12 ). (5.8)

That is, we need to analyse the spectrum of the Hermitian operator N̂ := ↠â.
Theorem 8. N̂ has a discrete spectrum consisting of the non-negative integers.
To prove this theorem we will need the commutators [N̂ , â] and [N̂ , ↠], which follow from
[â, ↠] = 1, as

[N̂ , â] = [↠â, â]

= [↠, â]â
(5.9)
= −[â, ↠]â
= −â

and similarly,

[N̂ , ↠] = ↠[â, ↠]

(5.10)
= ↠.

Lemma 1. The eigenvalues of N̂ are non-negative. Further, if ν is an eigenvalue of N̂ with

eigenvector |νi then â|νi is the zero vector if and only if ν = 0.
Proof. Consider an eigenstate |νi of N̂ with eigenvalue ν, that is N̂ |νi = ν|νi. Consider now the
norm of the vector â|νi. We have

||â|νi||2 ≥ 0
hν|↠â|νi ≥ 0
(5.11)
hν|N̂ |νi ≥ 0
νhν|νi ≥ 0,

from which it follows that ν ≥ 0.

Further, for ν = 0 we have that ||â|νi||2 = 0, which means that â|0i is the zero vector, we write
â|0i = 0. (Note that |0i is not the zero vector here, it is simply a vector labelled by the index 0.)
Also note that â|νi can be the zero vector for no other value of ν according to equation (5.11).
Lemma 2. If |νi is an eigenvector of N̂ corresponding to the eigenvalue ν, and â|νi =
6 0, then â|νi
is also an eigenvector of N̂ corresponding to the eigenvalue ν − 1.
Proof.
 
N̂ â|νi = N̂ â − âN̂ + âN̂ |νi
 
= [N̂ , â] + âN̂ |νi
(5.12)
 
= −â + âN̂ |νi
= â(N̂ − 1)|νi
= (ν − 1)â|νi.
5.2. NORMALISED EIGENVECTORS 33

Combining Lemma 1 and 2 we conclude that either the spectrum is empty (i.e. there are no
eigenvalues at all) or ν = 0 has to be in the spectrum, and no non-integer value of ν could possibly
be in the spectrum (to guarantee that the series ν, ν − 1, ν − 2, . . . does not continue to negative
values).

Lemma 3. If |νi is an eigenvector of N̂ corresponding to the eigenvalue ν, then ↠|νi is also an
eigenvector of N̂ corresponding to the eigenvalue ν + 1.

Exercise 9. Prove Lemma 3.

What we have shown now is that ν = 0 could be an eigenvalue of N̂ . If it is, then starting from
ν = 0 via consecutive application of ↠we thus deduce that all integers are eigenvalues of N̂ . If
ν = 0 is not an eigenvalue, the operator N̂ has no eigenvalues at all. Thus, there is a representation
of the algebra consisting of â, ↠and N̂ (and the identity) in which N̂ is an infinite-dimensional
matrix with diagonal elements given by its egenvalues, the non-negative integers. This completes
our proof of Theorem 8.
The spectrum of the harmonic oscillator (Hamiltonian) thus consists of the discrete set of num-
bers En = ~ω(n + 21 ) where n is a non-negative integer,

Ĥ|ni = ~ω(n + 21 )|ni. (5.13)

Remark 29. The quantum energies are quantised. A classical particle in a harmonic oscillator poten-
tial can have any energy that is larger than the minimum of the potential (zero, in the present case),
a quantum particle can only be in a superposition of states with discrete energies. Remembering
the quantum rules in Chapter 4, this means that every time the energy of a quantum particle is
measured, the outcome can only be one of the discrete energy eigenvalues. The expectation value
resulting from a number of measurements, of course, can be any number larger than the smallest
eigenvalue (which is larger than zero!), depending on the state of the particle (similar to the classical
case).
Remark 30. The operators â and ↠are often referred to as ladder operators, more specifically â
is the lowering and ↠the raising operator. Often one also referes to â as annihilation and ↠as
creation operator. The operator N̂ is called the number operator.

5.2 Normalised eigenvectors

We will first consider a possible degeneracy of the eigenvalues. We will show that if the ground state
is non-degenerate (which we will explicitly show in a later Chapter), then all the eigenvalues are
non-degenerate, i.e. there is a unique (up to a complex factor) eigenvector |ni for each eigenvalue
n. We shall now prove that if the eigenvalue n is non-degenerate so is the eigenvalue n + 1. Let us
prove this by contradiction.
Assume that the eigenvalue n is non-degenerate, but there are two linearly independent eigen-
states |φn+1 i and |χn+1 i belonging to the eigenvalue n + 1. Applying the lowering operator to these
two states we can produce the states |φn i = â|φn+1 i and |χn i = â|χn+1 i. Now, since the eigenvalue
n is non-degenerate these two vectors have to be multiples of each other, i.e. there exists a λ ∈ C
such that
|φn i = λ|χn i. (5.14)
Now acting on this relation with the raising operator ↠yields:

↠|φn i = ↠â|φn+1 i = N̂ |φn+1 i = λ↠|χn i = λ↠â|χn+1 i = λN̂ |χn+1 i, (5.15)

that is N̂ |φn+1 i = λN̂ |χn+1 i. But since |φn+1 i and |χn+1 i are eigenstates of N̂ with eigenvalue
(n + 1), we have
|φn+1 i = λ|χn+1 i, (5.16)
34 CHAPTER 5. THE HARMONIC OSCILLATOR

which means the two assumed to be linearly independent eigenvectors are linearly dependent after
all, and the eigenvalue n + 1 is also non-degenerate. Thus, assuming that the eigenvalue n = 0 is
non-degenerate, we have shown that the eigenvalues are non-degenerate.

We know that ↠|ni = cn |n + 1i. Let us now deduce the constants cn assuming that the
eigenvectors are normalised. We consider the norm of |n + 1i:

1 1 n+1
hn + 1|n + 1i = hn|â↠|ni = hn|N̂ + 1|ni = hn|ni. (5.17)
|cn |2 |cn |2 |cn |2
√
Thus, if |ni is normalised, so is |n + 1i for the choice cn = n + 1. That is,
√
↠|ni = n + 1|n + 1i, (5.18)

from which it follows that

1
|ni = √ â†n |0i. (5.19)
n!

Remark 31. The corresponding equation to (5.18) for â is

√
â|ni = n|n − 1i. (5.20)

5.3 Matrix representations in the harmonic oscillator basis

1. Matrix representation of ↠: From equation (5.18) it follows for the matrix elements of â†
in the harmonic oscillator basis:
√
hm|↠|ni = n + 1 δm,n+1 . (5.21)

2. Matrix representation of â: We can deduce the matrix elements for â either directly from
equation (5.20) or from the fact that â is the adjoint of ↠, and thus their matrix elements are
related by
hm|â|ni = hn|↠|mi. (5.22)

From the matrix elements of ↠we thus find the matrix elements of â as
√
hm|â|ni = n δm,n−1 (5.23)

3. Matrix representation of Ĥ: The Hamiltonian matrix is diagonal in the eigenbasis, with
diagonal elements
1
hm|Ĥ|ni = ~ω(n + ) δm,n (5.24)
2

Exercise 10. Matrix representation of x̂ and p̂.

(a) What are the matrix elements of the operators q̂ and p̂ in the basis of harmonic oscillator
eigenstates {|ni} ?

(b) Verify that these matrices fulfil the correct commutation relation.
5.4. HARMONIC OSCILLATOR BASIS - A SIMPLE NUMERICAL METHOD 35

5.4 Harmonic oscillator basis - A simple numerical method for

general potentials
The matrix representation of position and momentum operator in the harmonic oscillator basis
forms the basis of a simple numerical method to calculate eigenvalues of one-dimensional systems
with Hamiltonians that are functions of position and momentum operators. Here the matrices are
terminated at finite size, leading to an approximation. For more details on this method see H. J.
Korsch and M. Glück, Computing quantum eigenvalues made easy, Eur. J. Phys. 23, 413 (2002).

The following matlab program gives out the 8 lowest energy states of the harmonic oscillator Hamil-
tonian Ĥ = p̂2 /2 + x̂2 /2:

clear all
N=100; %matrix size

%matrices for postition and momentum operators

n=1:N-1;
k=sqrt(n);

Q=sqrt(0.5)*(diag(k,1)+diag(k,-1));
P=i*sqrt(0.5)*(diag(k,-1)-diag(k,1)):

%Hamiltonian
H=0.5*P^2+0.5*Q^2;

%eigenvalues
EigSort=sort(eig(H));
EigSort(1:8)

This programme can be modified to calculate the spectrum of many other Hamiltonians with
different potential functions. In a video I discuss the example of a quadratic potential with a small
quartic contribution V (x) = x2 + cx4 .
36 CHAPTER 5. THE HARMONIC OSCILLATOR
Chapter 6

Representations

In this chapter, we finally connect the abstract formulation of quantum mechanics expressed in the
five principles, and using the Dirac notation, back to the Schrödinger wave formulation we have
encountered in Chapter 2. The essence is that in Schrödinger’s wave mechanics, the position plays
the role of a preferred observable, and the wave function is nothing but the state of the quantum
system, expressed in the basis of eigenstates of the position operator. What complicates the matter,
is that the position operator has a continuous spectrum, and no normalisable eigenfunctions on the
Hilbert space, which leads to some technical complications when formulating the theory.

6.1 Discrete and continuous spectra - normalisable and improper

eigenvectors
Consider a linear operator  acting on a complex Hilbert space H. If there exists a pair of a complex
number λ ∈ C and a non-zero vector |φi ∈ H, such that
Â|φi = λ|φi, (6.1)
then |φi is an eigenvector of  belonging to the eigenvalue λ.
We will now follow Dirac’s argument, to argue why eigenvectors with a continuous set of eigen-
values cannot be normalisable. There are more rigorous mathematical treatments of this problem,
which are covered in advanced analysis lectures. Here we stay within the realm of quantum me-
chanics, and assume that the principles of Chapter 4 should be applicable even to an observable
with a continuous spectrum, or else such observables would not be permissible in QM. A central
idea in the principles of QM is that the eigenstates of an observable form a basis of the relevant
Hilbert space, i.e. for discrete eigenvalues λn belonging to eigenvectors |φn i, or more generally for
potentially degenerate eigenvalues, belonging to eigenspaces with projectors P̂n , an arbitrary vector
in the Hilbert space can be expanded as
X
|ψi = P̂n |ψi. (6.2)
n

Now, let us consider the case of a continuous spectrum. For simplicity, let us assume that the
eigenvalues form a continuous interval on the real line [a, b] ⊂ R. The equivalent of equation (6.2)
then states that every vector |ψi can be written as
Z b
|ψi = P̂λ |ψidλ, (6.3)
a

however, this implies that the norm of the eigenvectors cannot be finite. Let us for simplicity assume
that the spectrum is non-degenerate, and write
Z b
|ψi = hφλ |ψi|φλ idλ, (6.4)
a

37
38 CHAPTER 6. REPRESENTATIONS

If we consider a single eigenvector |φλ i we have

Z b
|φλ i = hφλ0 |φλ i|φλ0 idλ0 . (6.5)
a

Now, since  is a self-adjoint operator, the eigenvectors belonging to different eigenvalues are
orthogonal to each other, meaning that the integrand vanishes at all but one point λ = λ0 . The
only way to make the equality work is to assume that the squared norm hφλ |φλ i of an eigenvector
belonging to an eigenvalue in the continuous spectrum is infinite. Specifically, we need

hφλ0 |φλ i = δ(λ0 − λ), (6.6)

where δ(x) denotes the Dirac Delta-Distribution, which is infinite at λ = λ0 , which means that
an eigenvalue in the continuum is not strictly speaking an eigenvalue, since it does not have an
eigenvector in the Hilbert space.
Remark 32. Remember that a Hilbert space is a complete metric space, that is, it cannot have
elements with infinite norm.
Remark 33. The continuous “eigenvalues” are referred to simply as the continuous spectrum in
mathematics, and together with the discrete eigenvalues they make up the whole spectrum of a
self-adjoint operator. Sometimes they are referred to as improper or approximate eigenvalues, and
the corresponding “eigenvectors” (with infinite norm) are referred to as improper eigenvectors.
Mathematically, the continuous spectrum is defined using a generalisation of the eigenvalue
equation, that circumvents the concept of eigenvectors. The eigenvalue equation (6.1) implies that
ˆ
(Â − λI)|φi = 0, (6.7)
ˆ is not invertible when λ is an eigenvalue. Bounded Hermitian operators Â
i.e., the operator (Â−λI)
have discrete real eigenvalues (forming what is known as the point spectrum). Unbounded Hermitian
operators, however, might have an (additional) continuous spectrum, consisting of values λ ∈ R that
span a continuous interval, and for which the operator (Â − λI) ˆ −1 is defined, however, its domain
is not the whole Hilbert space. As we have seen, there are no eigenvectors in the Hilbert space,
corresponding to these values of λ, but one can extend the notion of eigenvectors to vectors with
an infinite norm in the sense explained above.
In quantum mechanics, we simply refer to the continuous spectrum as eigenvalues and use
the same bra-ket notation for the improper eigenvectors as for the normalisable ones, belonging to
discrete eigenvalues. With the only difference that for eigenvectors belonging to discrete eigenvalues
λn we have
hφn |φm i = δn,m , (6.8)
while for continuous eigenvalues this is replaced by (6.6).
Importantly, while the eigenvectors of an operator  with a purely discrete spectrum provide a
resolution of the identity as X
ˆ
|φn ihφn | = I, (6.9)
n
the eigenvectors of an operator with a purely continuous spectrum fulfil
Z b
ˆ
|φλ ihφλ |dλ = I. (6.10)
a

6.1.1 A few comments on the Dirac δ-Distribution

The δ-distribution is strictly speaking only defined inside an integral, and fulfils the important
property Z ∞
f (x)δ(x)dx = f (0), (6.11)
−∞
6.2. EIGENVALUES OF THE POSITION OPERATOR 39

for functions f (x) with compact support on R.

The δ-distribution can be expressed as the limit of various well-defined functions, for example
as the limit of a Gaussian with vanishing width,
x2
δ(x) = lim √1 e− 2σ2 , (6.12)
σ→0 σ 2π

or as the Fourier transform of a plane wave as

Z ∞
1
δ(x − α) = eik(x−α) dk. (6.13)
2π −∞

Remark 34. In equation (6.12) we have yet another abuse R ∞of notation. Writing
R ∞ δ(x) = g(x) is a
short-hand form for the slightly more rigorous statement −∞ f (x)δ(x)dx = −∞ f (x)g(x)dx.

6.2 Eigenvalues of the position operator

While there are no finite dimensional matrices q̂ and p̂ that fulfil the required commutation relation
[q̂, p̂] = i~Iˆ for position and momentum operators, we have seen in Chapter 5 that there are infinite
dimensional matrices (in the harmonic oscillator basis), with the elements
r
~ √ √

hm|q̂|ni = n δm,n−1 + n + 1 δm,n+1 , (6.14)
2mω
and r
mω~ √ √

hm|p̂|ni = i n + 1 δm,n+1 − n δm,n−1 . (6.15)
2
Let us now formally solve the eigenvalue
q equation for this infinite dimensional matrix representation
~
of q̂. Defining the length L = mω , q̂ has the matrix representation
 √ 
√0 1 √0 0 · · ·
 1 0
 √ 2 √0 · · · 
L  0
 2 0 3 
Q= √  √
 (6.16)
2 0 .. 
 0 3 0 .
.. .. .. ..
. . . .

in the harmonic oscillator basis. The eigenvalue equation

q̂|φλ i = λ|φλ i (6.17)

translates into a recursion relation for the components φn (λ) := hn|φλ i of the eigenvector in the
harmonic oscillator basis, with √
λ 2
φ1 (λ) = φ0 (λ) (6.18)
L
and √
√ √ λ 2
nφn−1 (λ) + n + 1φn+1 (λ) = φn (λ), for n > 0. (6.19)
L
That is, fixing the first component φ0 arbitrarily, we can always construct the other components
of an eigenvector, corresponding to any real eigenvalue λ. These vectors, however, will not be
normalisable. To solve the recursion relation for the eigenvector components, we recognise it as a
version of the recursion relation fulfilled by the Hermite polynomials Hn (x), which are a family of
orthogonal polynomials, fulfilling the orthogonality relation
Z +∞
2 √
Hm (x)Hn (x)e−x dx = π2n n!δm,n . (6.20)
−∞
40 CHAPTER 6. REPRESENTATIONS

They can be defined via the expression

 n
x2 /2 d 2 /2
Hn (x) = e x− e−x . (6.21)
dx
They are polynomials of degree n with alternating parity (even, odd, even, ...). Importantly, they
fulfil the recursion relation
Hn+1 (x) = 2xHn (x) − 2nHn−1 (x),
with H0 (x) = 1, and H1 (x) = 2x.
We notice the similarity to the recursion relation (6.19). Starting from (6.19), let us introduce
the rescaled components χn (λ) of the eigenvector as
√
χn (λ) = 2n n!φn (λ). (6.22)

After a little algebra we find that these fulfil the recursion relation
λ
χn+1 (λ) = 2 χn (λ) − 2nχn−1 (λ), (6.23)
L
and thus we have
λ
χn (λ) ∝ Hn ( ), (6.24)
L
and thus
φ0 (λ) λ
φn (λ) = √ Hn ( ). (6.25)
n
2 n! L
However, as we expect, these eigenvectors are not normalisable. We have
∞
X |φ0 (λ)|2
hφ(λ)|φ(λ)i = Hn2 (λ/L), (6.26)
2n n!
n=0

which, unfortunately, is a divergent series.

Of course, from the discussions in section 6.1 we already know that we expect (with the correct
“normalising” factor) hφ(λ)|φ(λ0 )i = δ(λ − λ0 ). This indeed can be seen from the properties of the
Hermite polynomials in a suitable limit. The Hermite polynomials fulfil the important relation (also
known as Mehler’s formula)
∞ c2 (x2 +y 2 )−2cxy
X Hn (x)Hn (y)  c n 1 −
=√ e 1−c2 , for − 1 < c < 1. (6.27)
n! 2 1 − c2
n=0

λ2
Choosing φ0 (λ) = 1√
π 1/4 L
e− 2L2 , we thus have
2 (λ2 +λ02 )−2cλλ0 y
1 (λ02 +λ2 ) 1 −
c
hφ(λ)|φ(λ0 )i = √ e− 2L2 lim √ e (1−c2 )L2 , (6.28)
π c→1 1 − c2
(6.29)

the overall exponent can be rewritten as

(λ02 + λ2 ) c2 (λ2 + λ02 ) − 2cλλ0 y (1 − c)(λ + λ0 )2 (1 + c)2 (λ − λ0 )2

− − = − − , (6.30)
2L2 (1 − c2 )L2 4(1 + c)L 4(1 − c2 )L2
which simplifies our expression to
0 2
1 1 −
(λ−λ )
hφ(λ)|φ(λ0 )i = √ lim √ e (1−c2 )L2 , (6.31)
π c→1 L2 1 − c2
(6.32)
6.3. POSITION REPRESENTATION 41

1−c2
Introducing the variable σ 2 = 2 we can rewrite this as
1 λ λ0 2
e− 2σ2 ( L − L )
hφ(λ)|φ(λ0 )i = lim √ , (6.33)
σ→0 σ 2πL
which is indeed one of the representations of the δ-distribution we have encountered in section 6.1.
In summary, we have found that the position operator has a continuous spectrum spanning the
whole real line x ∈ R, with corresponding (generalised) eigenvectors |xi,

q̂|xi = x|xi. (6.34)

The eigenvectors fulfil the continuous analogue of the orthonormality condition,

hx|x0 i = δ(x − x0 ), (6.35)

and form a resolution of the identity as

Z +∞
ˆ
|xihx|dx = I. (6.36)
−∞

6.3 Position representation

The wave function in position representation ψ(x) corresponds to the coefficients of the abstract
vector |ψi in the basis of the eigenfunctions of q̂, that is ψ(x) := hx|ψi, and
Z ∞
|ψi = ψ(x)|xidx. (6.37)
−∞

The squared norm of the state |ψi can thus be evaluated in position representation as
Z +∞
hψ|ψi = |ψ(x)|2 dx. (6.38)
−∞

Demanding the norm to be finite, we thus identify our Hilbert space with the space of square
integrable functions, L2 (R).
The matrix elements of the position operator in the position basis are

hx|q̂|x0 i = xδ(x − x0 ), (6.39)

which simply means that the position operator is diagonal in the basis of its eigenvectors, as ex-
pected. We can also define the position operator in the position representation directly by its action
on a square integrable function ψ(x) ∈ L2 as

q̂ : ψ(x) 7→ xψ(x). (6.40)

More formally, this is equivalent to

hx|q̂|ψi = xhx|ψi = xψ(x). (6.41)

When measuring a particle’s position,R according to principle 3, the probability of obtaining a value
b
in the interval I = [a, b] is given by a |ψ(x)|2 dx, as we have assumed in Chapter 2. That is, the
interpretation of Schrödinger’s wave function is a special case of principle 3.
With the definition of the position operator, also the momentum operator p̂ is fixed, because p̂
and q̂ have to fulfil equation (4.1). The operator
∂
p̂ : ψ(x) 7→ −i~ ψ(x) (6.42)
∂x
fulfils this condition with (6.40).
42 CHAPTER 6. REPRESENTATIONS

Exercise 11. Verify that the operators (6.42) and (6.40) indeed fulfil the commutation relation
(4.1), by applying their commutator onto a test function ψ(x) ∈ L2 .

Remark 35. One could in principle add an arbitrary function of q̂ to p̂ and still fulfil the commutation
relation (4.1). The reason why it is fine to choose this function to be zero is a little subtle, and does
not need to concern us here. If you are interested in the details, see e.g. Dirac’s textbook.
p 2
Finally, the kinetic energy is described by Ekin = 2m in classical mechanics. This is analogous
in quantum mechanics, and thus the operator of kinetic energy in the position representation is
given by
p̂2 ~2 ∂ 2
Êkin = : ψ(x) 7→ − ψ(x). (6.43)
2m 2m ∂x2
With this we are back to Schrödinger’s equation as it appears in wave mechanics, where the
p̂2
Hamiltonian for a particle in a one-dimensional potential is the operator Ĥ = 2m + V (q̂), and thus
the Schrödinger equation in position representation,

i~hx|ψ̇i = hx|Ĥ|ψi, (6.44)

explicitly reads
~2 ∂ 2
i~ψ̇(x, t) = − ψ(x, t) + V (x)ψ(x, t). (6.45)
2m ∂x2

6.4 Change of bases/representation

To change between two representations, we simply insert a resolution of the identity in the basis we
are starting from. That is, for example, if we have the coefficients ψn of a vector |ψi in the harmonic
oscillator basis, that is, ψn = hn|ψi and want to know its position representation ψ(x) = hx|ψi we
find this in the following way.
!
X
ˆ
ψ(x) = hx|ψi = hx|I|ψi = hx| |nihn| |ψi (6.46)
n
X
= hx|nihn|ψi (6.47)
n
X
= ψn φn (x), (6.48)
n

where φn (x) = hx|ni denotes the position representation of the harmonic oscillator eigenvector
belonging to the eigenvalue ~ω(n + 21 ). As a second example, if we want to translate from the
position representation ψ(x) = hx|ψi to the momentum representation ψ̃(p) = hp|ψi, where |pi
denotes the eigenstate of the momentum operator p̂ belonging to the eigenvalue p, we find

ψ̃(p) = hp|ψi (6.49)

Z +∞
= hp|xihx|ψidx (6.50)
−∞
Z +∞
= φ∗p (x)ψ(x)dx, (6.51)
−∞

where φp (x) = hx|pi denotes the position representation of the momentum eigenvector belonging
to momentum p, which you can deduce from solving the eigenvalue equation for p̂ in position
representation.
Chapter 7

Spectral properties of one-dimensional

quantum systems

In this chapter we will analyse the eigenvalues of the Hamiltonian for a number of important one-
dimensional quantum systems. That is, we solve the time-independent Schrödinger equation. Again,
~2 ∂ 2
we consider systems of the form Ĥ = − 2m ∂x2
+ V (x). Physically these Hamiltonians describe a
particle (in one-dimensional space) of mass m under the influence of an external potential V (x).
This could be, for example, an electron in an atom, where the potential is given by the interaction
with the nucleus and the other electrons (this is complicated and obviously three-dimensional and we
will not discuss details in this chapter). The particle in a box is a good approximation for electrons
in so-called quantum dots, which are fabricated semiconductor structures in which electrons are
confined. Atoms in optical potentials (where quantum optical effects are used to confine atoms with
laser setups) are another experimental system described by Hamiltonians of the form considered
here. Experimentally these systems can often be made quasi one-dimensional. Similarly atoms
can be trapped magnetically, in approximately harmonic traps described by quadratic potentials.
There are countless other examples, even outside quantum physics: The Helmholtz equation for
microwaves in flat (quasi two-dimensional) resonators is similar to the Schrödinger equation where
the potential is related to the structure of the cavity. Similar analogies exist in optical setups.

7.1 Position representation of the harmonic oscillator eigenstates

With the position representation from Chapter 6 at hand, we can now return to the harmonic
oscillator and provide the position representation of the eigenstates. In fact, we have already
calculated this along the way when we considered the eigenvalues of the position operator in section
6.2, where we deduced that
1 x2 x
hn|xi = √ e− 2L2 L Hn ( ). (7.1)
π 1/4 n
2 n! L

Thus, we have implicitly found there that

1 x2 x
φn (x) = hx|ni = √ e− 2L2 L Hn ( ). (7.2)
π 1/4 2n n! L

Nevertheless, let us, as an exercise also consider the problem in position representation picking
up at an earlier point and verifying (7.2). Let us start with deducing the position representation of
the ground state (the state belonging to the lowest eigenvalue, E0 = ~ω 2 , directly from the condition
â|0i = 0. This will also provide the direct verification of the statement that the ground state energy
is not degenerate, which we had used without proof in Chapter 5. We can deduce the functional
form of φ0 (x) from the equation â|0i = 0 in position representation.

43
44 CHAPTER 7. SPECTRAL PROPERTIES OF ONE-DIMENSIONAL QUANTUM SYSTEMS

First we remember the definition of â in terms of q̂ and p̂:

r r !
1   1 mω 1
â := √ Q̂ + iP̂ = √ q̂ + i p̂ (7.3)
2 2 ~ mω~

We recall the expressions for q̂ and p̂ in position representation,

∂
hx|q̂|φi = xφ(x), and hx|p̂|φi = −i~ φ(x), (7.4)
∂x
to find r r !
1 mω ~ ∂
hx|â|φi = √ x+ φ(x). (7.5)
2 ~ mω ∂x
Thus, the ground state wave function fulfils the condition
∂ mω
φ0 (x) = − xφ0 (x). (7.6)
∂x ~
Thus we find
mω 2
φ0 (x) = φ0 (x) = c e− 2~ x , c ∈ C. (7.7)
We can deduce the normalisation constant c by requiring the wave function to be normalised. From
Z ∞ Z ∞
mω 2
|φ0 (x)|2 dx = |c|2 e− ~ x dx = 1, (7.8)
−∞ −∞

we deduce r
2 mω
|c| = , (7.9)
π~
and thus we choose  mω 1/4
c= , (7.10)
π~
such that the ground state wave function is normalised, real, and positive. In summary, we have
found the position representation of the normalised ground state wave function of the operator N̂
(and thus the harmonic oscillator):
 mω 1/4  mω 
φ0 (x) = exp − x2 . (7.11)
π~ 2~
We can now use the abstract expression for the chain of eigenvectors starting from the ground
state in equation (5.19), to deduce the position representation of these eigenvectors, φn (x) := hx|ni.
Using expression (5.19) we can deduce all φn (x) from φ0 (x). To find the position representations of
the higher excited states, we need to apply ↠onto the ground state:
1
hx|ni = φn (x) = √ hx|(↠)n |0i. (7.12)
n!
That is,
r r !!n
1 1mω ~ ∂
φn (x) = √ √ x− φ0 (x)
n! 2
~ mω ∂x
r r !n (7.13)
1  mω 1/4 1 mω ~ ∂  mω 
=√ √ x− exp − x2 .
n! π~ 2n ~ mω ∂x 2~

Explicitly the φn (x) can be expressed in terms of Hermite polynomials Hn (x) as

s √
mω 2
φn (q) = √ Hn (q) e−q /2 , (7.14)
2n n! π~
with q = mω
p
~ x, as expected.
7.2. PROPERTIES OF EIGENFUNCTIONS FOR GENERAL ONE-DIMENSIONAL POTENTIALS45

7.2 Properties of eigenfunctions for general one-dimensional po-

tentials
There are only very few examples of Hamiltonian operators the spectrum of which can be deduced
analytically. In general, approximative or numerical methods have to be applied. Yet, a few general
remarks can be made about the solutions of time-independent Schrödinger equations of the form

~2 ∂ 2
 
− + V (x) φE (x) = EφE (x) : (7.15)
2m ∂x2

• In general there are two types of solutions of (7.15) that we can interpret in a meaningful
way in Schrödinger’s wave mechanics: The bound states, for which φ(x → ±∞) → 0, and the
continuum states, which are improper eigenfunctions (not in L2 ), belonging to eigenvalues in
the continuous part of the spectrum.

• For potentials for which V (x → ±∞) → ∞ there are only bound states (i.e. only discrete
eigenvalues - this is the quantisation part of quantum mechanics again - energies can often
only have discrete values.)

• Eigenfunctions can be chosen real valued.

Proof. Assume we have a complex solution φ(x) of the Schrödinger equation. Then also its
complex conjugate function φ∗ (x) is a solution. Thus, the linear combination φ(x) + φ∗ (x) is
a real valued solution of the Schrödinger equation.

• There are no bound states with eigenenergies smaller than the minimum of the potential.
Proof. We have
∂ 2 φE (x) 2m
2
= 2 (V (x) − E)φE (x). (7.16)
∂x ~
Now if E < Vmin (x) then the the wave function φE (x) and its second derivative would have
the same sign for all x. That is if the wave function would have to curve away from the x-axis
everywhere. Such a function however, can not be normalisable.

• The expectation value of a Hamiltonian is bounded from below by the lowest energy eigenvalue.
We will prove this in the end of the week lecture.

• The bound state eigenvalues are non-degenerate (that is, there is only one linearly independent
solution per eigenvalue).

Proof. Let us first rewrite the Schrödinger equation (7.15) in the form

φ00 (x) + k 2 (x)φ(x) = 0, (7.17)

q
where we have introduced the function k = 2m ~2
(E − V (x)). Suppose now that φ1 (x) and
φ2 (x) are two bound state eigenfunctions corresponding to the same energy E. We shall show
now that this implies that φ1 (x) and φ2 (x) are linearly dependent, i.e. they differ at most by
a multiplicative factor.
We have
φ00j (x)/φj (x) = −k(x)2 , (7.18)
for j = 1, 2. That is,
φ001 (x)/φ1 (x) = φ002 (x)/φ2 (x), (7.19)
or
φ001 (x)φ2 (x) − φ002 (x)φ1 (x) = 0. (7.20)
46 CHAPTER 7. SPECTRAL PROPERTIES OF ONE-DIMENSIONAL QUANTUM SYSTEMS

We realise that the left hand side is equal to the derivative of φ01 (x)φ2 (x) − φ02 (x)φ1 (x), that
is we have
d
φ01 (x)φ2 (x) − φ02 (x)φ1 (x) = 0.


(7.21)
dx
Integrating this yields

φ01 (x)φ2 (x) − φ02 (x)φ1 (x) = c, c ∈ C. (7.22)

Since we consider bound states we have that φj (x → ∞) → 0 while the derivative stays
bounded, and thus we deduce that c = 0. We now integrate the expression

φ01 (x)/φ1 (x) = φ02 (x)/φ2 (x), (7.23)

to find
ln φ1 (x) = ln φ2 (x) + const. (7.24)
that is
φ1 (x) ∝ φ2 (x). (7.25)

• The ground state (the state corresponding to the lowest energy eigenvalue) does not have
zeros (apart from the asymptotic ones at x → ±∞).

• The n-th excited state (the state corresponding to the (n + 1)th eigenvalue) has n zeros inside
the potential.
For a hand-wavy explanation of this see for example Barton Zwiebach’s video lecture on the
topic.

7.2.1 Symmetric potentials - parity

Consider the Hermitian operator Ŝ on the Hilbert space of square integrable functions that inter-
changes x with −x, that is, Ŝψ(x) = ψ(−x). This is called the parity operator. The eigenfunctions
of the parity operator fulfil
Ŝφs (x) = φs (−x) = sφs (x). (7.26)
On the other hand Ŝ 2 = 1, that is,

Ŝ(Ŝφs (x)) = s2 φs (x) = φs (x). (7.27)

Thus, we have s2 = 1, and therefore the only two eigenvalues of the parity operator are

s = ±1. (7.28)

The corresponding eigenfunctions are all even or odd functions φ± (−x) = ±φ(x). (The parity
operator is highly degenerate).
~2 ∂ 2
A Hamiltonian of the form Ĥ = − 2m ∂x2
+ V (x) commutes with Ŝ if V (x) = V (−x). This can
be seen as follows. Consider the action of the commutator on a test function ψ(x) ∈ L2 :

~2 ∂ 2
[Ĥ, Ŝ]ψ(x) = − ψ(−x) + V (x)ψ(−x)
2m ∂x2
~2 ∂2 (7.29)
+ ψ(−x) − V (−x)ψ(−x)
2m ∂(−x)2
= V (x)ψ(−x) − V (−x)ψ(−x),

which is zero for arbitrary wave functions ψ(x) if and only if V (−x) = V (x).
7.3. PIECEWISE CONSTANT POTENTIALS 47

According to theorem 7, if the eigenvalues of Ĥ are non-degenerate, the corresponding eigen-

functions are either odd or even. Since the eigenvalues of one-dimensional Hamiltonians of the form
V (x → ±∞) → ∞ (binding potentials) are always non-degenerate. If the potential is symmetric,
the eigenfunctions are either even or odd. We have already seen an example when we discussed the
particle in a square well in Chapter 2, there the potential was symmetric with respect to x = L2 and
the eigenfunctions were even and odd with respect to this point.
Remark 36. Unfortunately most of the statements above do not carry over to higher dimensional
systems. This concerns in particular the non-degeneracy of eigenvalues and hence the symmetry of
eigenfunctions in symmetric potentials.

7.3 Piecewise constant potentials

7.3.1 Piecewise constant potentials and boundary conditions
For the remainder of this chapter we are specifically looking for the solutions E and φE (x) of the
time-independent Schrödinger equation

~2 ∂ 2
− φE (x) + V (x)φE (x) = EφE (x), (7.30)
2m ∂x2
where the potential V (x) is piecewise constant. Rewriting this in the form

~2 ∂ 2
− φE (x) = (E − V (x))φE (x), (7.31)
2m ∂x2
emphasises that the second derivative of an eigenfunction is essentially proportional to the value of
the potential energy. Thus, at values of x at which the potential makes a finite jump, so does the
second derivative of the wave function, which means that the first derivative is continuous, while the
wave function itself is continuous and continuously differentiable. If the potential makes an infinite
jump the first derivative of the wave function is discontinuous, but the wave function itself is still
continuous.
To obtain the eigenfunctions φE (x) we can consider the different regions of constant potential
separately and patch the solutions together. Thus, we start out by determining the solutions of the
time-independent Schrödinger equation in constant potentials.

7.3.2 The free particle

The free particle is a particle moving without the influence of an external potential, that is, V (x) = 0.
However, in quantum (as well as in classical) physics, the zero point of the energy is arbitrary.
Thus, the free particle is equivalent to a particle moving under the influence of a constant potential
V (x) = V0 , with the Hamiltonian
~2 ∂ 2
Ĥ = − + V0 , (7.32)
2m ∂x2
with V0 ∈ R. The eigenvalue equation for the Hamiltonian (7.32), given by

~2 ∂ 2
− φE (x) = (E − V0 )φE (x), (7.33)
2m ∂x2
is formally solved by functions of the form

φE (x) = Aeikx + Be−ikx , (7.34)

with the wave number p

2m(E − V0 )
k= , (7.35)
~
48 CHAPTER 7. SPECTRAL PROPERTIES OF ONE-DIMENSIONAL QUANTUM SYSTEMS

for arbitrary values of E ∈ R. However, the functions (7.34) diverge at x → ±∞ for negative values
of E − V0 . For the free particle on the whole real axis, we cannot interpret them in any meaningful
way, and thus deduce that the eigenvalues of Ĥ are given by all real numbers E ≥ V0 .
Even the eigenfunctions corresponding to these energies, however, are not square integrable, and
can thus not be interpreted in the usual way as a probability amplitude to find a particle somewhere
at a given time. They are however, a basis for all physically more meaningful wave functions in L2 .
Remark 37. We can interpret a wave function of the form

ψ± (x) ∝ e±ixk , (7.36)

a so-called plane wave, as a constant probability beam travelling from minus to plus infinity or plus
to minus infinity, respectively. These wave functions are the extreme cases of Heisenberg’s uncer-
tainty principle, with accurately defined momentum, and totally unspecified position (describing a
“completely delocalised particle”). Their probability flux is given by

~ ~k
ψ ∗ ψ 0 − ψ 0∗ ψ = ± .


j= (7.37)
2mi m
They are eigenfunctions of the momentum operator with eigenvalues p = ~k. Thus, they can be
p
viewed as a stream of particles moving with constant velocity m .
Remark 38. The plane waves (7.36) fulfil the generalised orthonormality condition
Z ∞
1 0
e−ikx eik x dx = δ(k − k 0 ). (7.38)
2π −∞

Remark 39. If we have a piecewise constant potential then in the separate regions that do not
extend to infinity on both sides also the solutions (7.34) with energies E < V0 are meaningful. They
are more commonly expressed in the form

φE (x) = Ceκx + De−κx , (7.39)

with p
2m(V0 − E)
κ= . (7.40)
~
For regions that extend to x+ ∞ we need C = 0 and for regions that extend to x → −∞ it holds
D = 0.

7.3.3 The infinite square well potential (“Particle in a box”)

Let us now finally derive the eigenfunctions of a particle in an infinite square well (or “box”)
potential 
0, 0 ≤ x ≤ L
V (x) = (7.41)
∞, otherwise,
that we have already used numerous times. This system is an extreme case of a piecewise constant
potential, where the potential is zero in a region x ∈ [0, L] (the interior of the box ) and infinite
outside this region. Thus, outside the box the wave function has to be identical to zero. Inside the
box, that is for 0 ≤ x ≤ L, we have the solutions

φ(x) = Aeikx + Be−ikx , (7.42)

√
with k = 2mE
~ . We know, however, that the wave function is continuous, and thus has to fulfil the
boundary conditions
φ(0) = 0 = φ(L). (7.43)
7.3. PIECEWISE CONSTANT POTENTIALS 49

This yields the conditions

A+B =0 (7.44)
ikL −ikL
Ae + Be =0 (7.45)

for A and B. Inserting B = −A into (7.45) yields

2iA sin(kL) = 0. (7.46)

The choice A = 0 leads to a wave function that vanishes everywhere. The only physical meaningful
wave functions exist for sin(kL) = 0, which means
√
2mE
L = nπ, (7.47)
~
for integer values of n, which yields the quantised energies

π 2 ~2 2
En = n . (7.48)
2mL2
The corresponding eigenfunctions are given by
√ 
2mEn  πnx 
φn (x) = 2iA sin x = 2iA sin . (7.49)
~ L

We can now choose A such that the φn (x) are normalised and real valued. The norm of φn (x) in
dependence on A is given by
Z L Z
 πnx L
2 2
|φn (x)| = 4|A| sin2 dx
0 0 L
Z L  
2 2πnx
= 2|A| 1 − cos dx
0 L
= 2|A|2 L. (7.50)

Thus, the choice A = √−i yields the normalised and real-valued wave functions
2L
( q
2
sin nπx


L L , 0≤x≤L
φn (x) = (7.51)
0, otherwise,

as given earlier without derivation in equation (2.25).

7.3.4 The finite square well potential (”Particle in a finite box”)

The finite square well potential is similar to the
previously considered infinite one, with the dif-
ference that the potential outside the central re-
gion is non-zero but finite. It is convenient to
consider a symmetric version

0, |x| ≤ L2

V (x) = (7.52)
V0 , otherwise.

The solutions of the time-independent Schrödinger

equation in the separatepregions are of the form
aeikx + be−ikx with k = 2m(E − Vj )/~, where
50 CHAPTER 7. SPECTRAL PROPERTIES OF ONE-DIMENSIONAL QUANTUM SYSTEMS

Vj is the value of the constant potential in re-

gion j. For values E > pVj , k ∈ R this can be equivalently written as ã cos(kx) +−κx
b̃ sin(kx), while for
E < Vj we have κ = 2m(Vj − E)/~ ∈ R, and the solutions take the form ae + beκx . That is
we make the ansatz
 C1 eκx + C2 e−κx , x ≤ −L/2


φE (x) = A cos(kx) + B sin(kx), −L/2 < x ≤ L/2 (7.53)

D1 eκx + D2 e−κx , x > L/2,

√ p
with k = 2mE/~ and κ = 2m(V0 − E)/~ for the solution of the time-independent Schrödinger
equation.
Here we are looking for bound states, that is φ(x → ±∞) → 0. From the general solution (7.53)
we deduce that this is only possible for E < V0 , and that further we need C2 = 0 = D1 . Thus, the
solutions are of the form
 Ceκx , x ≤ −L/2


φE (x) = A cos(kx) + B sin(kx), −L/2 < x ≤ L/2 (7.54)


De −κx , x > L/2.
The wave function φE (x) and its first derivative with respect to x have to be continuous ev-
erywhere. This is automatically fulfilled in the separate regions, but it imposes four boundary
conditions between the different regions.
Exercise 12. Find the four boundary conditions the eigenfunctions need to fulfil. Use the fact that
the symmetry of the potential means that the eigenfunctions are either even or odd eigenfunction,
and verify that even eigenfunctions have to fulfil the conditions
Ce−κL/2 = A cos(kL/2) (7.55)
−κL/2
κCe = kA sin(kL/2), (7.56)
and odd eigenfunctions have to fulfil the conditions
Ce−κL/2 = −B sin(kL/2) (7.57)
−κL/2
κCe = kB cos(kL/2). (7.58)

These conditions can only be fulfilled simultaneously for specific values of the energy E, leading
to the quantisation conditions
κ = k tan(kL/2) (7.59)
and
κ = −k cot(kL/2), (7.60)
for even and odd eigenfunctions, respectively. We can rewrite these in terms of the energy using
the identities
r
2mV0
κ/k = −1
~2 k 2 (7.61)
√
k = 2mE/~,
to find √ ! r
2mEL V0
tan = − 1, (7.62)
2~ E
and √ ! r
2mEL V0
− cot = − 1, (7.63)
2~ E
for even and odd eigenfunctions, respectively.
These conditions have to be solved numerically or graphically to find the quantised energies.
We will discuss this further in the lecture.
Chapter 8

Quantum dynamics

In this chapter we shall turn away from the spectral problems, and focus our attention on the
dynamical features of quantum theory. The most interesting features of observable things are often
those that change in time, after all. We shall discuss dynamics of expectation values, which yields
a version of the famous Heisenberg equations of motion, which can be viewed as an alternative to
Schrödinger’s dynamical equation in quantum mechanics. We then turn to the discussion of quantum
dynamics from a Lie algebraic perspective, and finish with an example of dynamical tunneling.

8.1 Dynamics of expectation values

From the time-dependent Schrödinger equation, i~|ψ̇i = Ĥ|ψi, we can deduce a dynamical equation
for observable expectation values hÂi: We have by definition

hψ|Â|ψi
hÂi = . (8.1)
hψ|ψi

Thus, we find
˙ d
hψ|ψi
d hψ̇|Â|ψi + hψ|Â|ψi + hψ|Â|ψ̇i
hÂi = − dt hψ|Â|ψi. (8.2)
dt hψ|ψi hψ|ψi2
From the Schrödinger equation we know
i i
|ψ̇i = − Ĥ|ψi, and hψ̇| = hψ|Ĥ, (8.3)
~ ~
d
thus, the overall norm is conserved, dt hψ|ψi = 0 and we find

i ˙
d ~ hψ|Ĥ Â|ψi + hψ|Â|ψi − ~i hψ|ÂĤ|ψi
hÂi =
dt hψ|ψi (8.4)
i ∂ Â
= hĤ Â − ÂĤi + h i.
~ ∂t
Or, using the commutator,
dhÂi i ∂ Â
= h[Ĥ, Â]i + h i. (8.5)
dt ~ ∂t
This is one formulation of the famous Heisenberg equation of motion for observables, which can be
used as the foundational dynamical equation of quantum mechanics, as an alternative of Schrödinger’s
equation. Comparing the Heisenberg equation to the classical equation of motion (1.16) for a phase
space variable A(p, q) we observe a striking similarity if we associate expectation values with clas-
sical variables and interpret the commutator, or more precisely i/~ times the commutator, as a
quantum Poisson bracket.

51
52 CHAPTER 8. QUANTUM DYNAMICS

The Ehrenfest theorem

The special case of the equations of motion for the expectation values of position and momentum
for a system with Hamiltonian of the type

p̂2
Ĥ = + V̂ (q̂) (8.6)
2m
forms the basis of what is referred to as the Ehrenfest theorem, which we shall consider in what
follows. From equation (8.5) and the form of the Hamiltonian (8.6) we find

d i p̂2
hq̂i = h[ , q̂]i
dt ~ 2m
i
= h[p̂2 , q̂]i (8.7)
2~m
i
= hp̂[p̂, q̂] + [p̂, q̂]p̂i.
2~m
Recalling the commutation relation between position and momentum operator, [q̂, p̂] = i~, this
reduces to
d hp̂i
hq̂i = . (8.8)
dt m
For the momentum operator we find
d i
hp̂i = h[V̂ (q̂), p̂]i. (8.9)
dt ~
To calculate the commutator between the momentum operator and the potential, it is convenient
to use the position representation, in which we have
∂
[V̂ (q̂), p̂] = −i~[V (x), ], (8.10)
∂x
applied to a test function ψ(x),
∂ ∂
−i~ [V (x), ] ψ(x) = −i~V (x)ψ 0 (x) + i~ (V (x)ψ(x))
∂x ∂x
= −i~V (x)ψ 0 (x) + i~V 0 (x)ψ(x) + i~V (x)ψ 0 (x) (8.11)
= i~V 0 (x)ψ(x)

Translating this back to the abstract operator notation this is

dV (q̂)
[V̂ (q̂), p̂] = i~ , (8.12)
dq̂
and thus we find
d ∂ V̂
hp̂i = −h i. (8.13)
dt ∂ q̂
In summary we have obtained the set of equations

d hp̂i d ∂ V̂
hq̂i = , and hp̂i = −h i, (8.14)
dt m dt ∂ q̂
which are strikingly similar in form to the classical equations of motion
d p d ∂V
q= , and p=− , (8.15)
dt m dt ∂q
a fact that is often referred to as the Ehrenfest theorem.
8.1. DYNAMICS OF EXPECTATION VALUES 53

Notwithstanding the formal similarity, there are of course important differences between quan-
tum and classical dynamics. On the level of the Ehrenfest theorem, this is due to the fact that in
general
∂ V̂ ∂V (hq̂i)
h i=
6 . (8.16)
∂ q̂ ∂hq̂i
Consider for example the case V̂ (q̂) = q̂ 4 . Here we have

∂ V̂ (hq̂i
= 4hq̂i3 , (8.17)
∂hq̂i

while
∂ V̂
h i = 4hq̂ 3 i. (8.18)
∂ q̂
The two expressions in equation (8.16) are, however, similar for potentials that are polynomials of
q̂ of order two or lower. For more general potentials equations (8.14) in fact not even form a closed
set of equations (as the left side of the time derivative of hp̂i is not only a function of hq̂i and hp̂i).

Literary Excursion
In the following I provide a translation of a short excerpt from the novel Dirac by Dietmar Dath
(Surkamp 2006) to provide some further insight into the development of quantum mechanics, albeit
in the form of a literary supplement.
Just like a melody, which the whistling lips are faster to grasp than the conscious mind, Heisen-
berg’s mathematics returns to the focus of Dirac’s attention. Non-commutativity, this times that is
different from that times this. A pale-blue bird thrusts itself onto the strong wind just above Dirac’s
head, struggling, staggering, spinning, re-balancing, it positions itself tangentially to the air stream,
and then rises again, very fast and very high. Dirac is viewing the bird in a coordinate plane, seeing
it rotating, and fleeing upwards.
Position and velocity, where the bird is thrown towards, and how it is flapping its wings: Sky
and puddle, wind and water, bird and – fish? “Poisson”, Dirac says.
That’s it! Poisson brackets. Siméon Denis Poisson, seventeen hundred eighty one to eighteen
hundred forty. Non-commutativity... Where has he read about this? What was it? There seems to
be a connection. Or is he just grabbing for thin air? Heisenberg’s commutator: uv − vu, and the
relation between the two quantities in this formalism for phase space transforms. ...This is only the
faintest memory: Old mathematics, which might come to rescue Heisenberg’s new physics. Dirac is
leaping over the road-side ditch without a thought on what he is doing, and he is now walking right
across the muddy meadow. Every step is making slurping sounds, Dirac is in a hurry: He has to go
home, has to look it up, find out what’s going on there... Poisson, non-commutativity – everything.
Dirt splashing on his trousers, his joints cracking, he is hurrying on. It can’t be true. The
similarity of these two forms must be a coincidence, just like a rhyme in a poem or a harmony of
two voices. It would be too weird for such a coincidence to actually mean something. Already he
has a feeling that this thought, which had only just occurred to him, might be implausible, but his
excitement is growing at the same time: What if there was something to it? After a few hundred
metres of cross-country walking he is now reaching a footpath and finding himself panting for air,
staggering, dizzy: He had been running. He has left the rural area behind and is now in the town.
He is pushed forward into his thoughts by a tailwind, which is seizing him just as it had seized the
bird before him.
Finally he reaches his door, hectically searching his coat pocket for his key. As he tries to unlock
the door too fast, the key gets stuck. He is swearing under his breath. He has to get in there,
look it up, clear definitions, order! He is rattling at the lock, which finally releases the key. Dirac
immediately reinserts it and this time the door opens. Without even taking off his coat Dirac is
rushing through his room, searching his books, his lecture notes, kneeling on the floor, opening all
54 CHAPTER 8. QUANTUM DYNAMICS

the drawers.
The storm that had been raging outside earlier is revived again in this room; sheets are flying,
eyes are searching, but nothing: canonical transformations, linear mappings, differential equations,
textbooks, mathematical tables. All of this is much too fundamental. Nothing on Poisson brackets,
no hint. But he did see it. Did read it: In a book, not in a lecture note.
Finally he is sitting in the midst of the chaos he has produced, panting. He takes off his coat
and it makes him chuckle: Sunday night. All libraries closed. He runs his fingers through his hair,
as if what he is looking for might be stuck in there.
Dreadful, euphoric night: Dirac’s heart is pounding in his throat, half asleep he is imagining
himself as a little bird in the wind, throwing himself from left to right. If it really was possible to ex-
press things this way, if one could formulate a type of Hamiltonian dynamics using non-commutative
Poisson-brackets... If that was the key...
Dawn is breaking. Dirac gets up. His joints are painful. He is extraordinarily weary, as if he had
just finished some sort of hard work. After a quick superficial wash he gets dressed and leaves the
house, walking straight towards the faculty library. He is the first one there. The librarian allows
him in, shaking his head. Dirac is rushing past long rows of bookshelves, he knows his way around
here.
He takes a left turn at the big cross-bar window in front of one of the columns. There, at breast
height, is Whittaker’s “Analytical Dynamics”. Dirac quickly finds what he was looking for. The
letters are flickering, and he is finding it difficult to read. Yet, his intuition had not failed him. He
now has his own key to the latest physics...

8.2 Lie groups and Lie algebras in quantum dynamics

8.2.1 A quick overview
The time evolution in quantum mechanics is closely related to the concept of (Lie) groups. In this
section, I will give a brief summary of the most important concepts in this context.
We begin with a brief reminder of the concept of a
Group. A group is a set G of elements, together with a map ◦ : G ×G → G, which fulfil the following
three properties:

• The map is associative, i.e. (a ◦ b) ◦ c = a ◦ (b ◦ c), for all a, b, c ∈ G,

• The group contains an identity element e with e ◦ a = a ◦ e = a for all a ∈ G,

• Every element has an inverse element in the group, often denoted by a−1 , such that a ◦ a−1 =
a−1 ◦ a = e.

A representation of a group can be thought of as a realisation of group elements as linear

operators on a linear vectorspace V, that is, a map (a group homomorphism) g → D̂(g), from G to
V, which fulfils
D̂(g1 ◦ g2 ) = D̂(g1 )D̂(g2 ). (8.19)
Thus we have D̂(g −1 ) = (D̂(g))−1 , and D̂(e) = I,ˆ where Iˆ denotes the identity on V. A represen-
tation is faithful if the map g → D̂(g) is invertible (an isomorphism).
A continuous group is a group with elements that continuously depend on parameters, i.e. the
group elements are functions of a set of parameters. The set of unitary operations on a Hilbert
space together with the consecutive application of unitary operations as the group map, form a
continuous group.
A special type of continuous groups are Lie groups. Whole lectures and textbooks are devoted
to the topic of Lie groups. We will only devote a couple of paragraphs, sacrificing rigour for an
intuitive idea. Broadly speaking, an n-dimensional Lie group is a continuous group, the elements g
8.2. LIE GROUPS AND LIE ALGEBRAS IN QUANTUM DYNAMICS 55

of which can be represented as C n functions of n real parameters, with g(0) = e. Thus, Lie groups
are both groups and analytical manifolds.
If D̂(g) is a representation of a Lie group element, we write for short D̂(x) = D̂(g(x)), and
ˆ Since a Lie group element is a differentiable function of the parameters, for small values
D̂(0) = I.
of x we can expand an element into a power series with the first two terms
n
X
D̂(x) = Iˆ + i xj T̂j + . . . , (8.20)
j=1

with
∂ D̂
T̂j = −i . (8.21)
∂xj
x=0
Remark 40. The factor of i in (8.20) and (8.21) is common in physics, but is not usually used in
the mathematics literature.
Remark 41. The higher order terms in (8.20) are in general complicated.
The operators T̂j in (8.21) are called the generators of the group. Let us for simplicity consider
a one-dimensional group with representation D̂(x) with x ∈ R. When the generator T̂ is known
the group structure allows to represent group elements not only for small but for arbitrary x, since
D̂(2∆x) = D̂(∆x)D̂(∆x) = (D̂(∆x))2 and D̂(m∆x) = (D̂(∆x))m , and thus, if we define x = m∆x
we have
x
D̂(x) = (D̂(∆x))m ≈ (1 + i∆xT̂ )m = (1 + i T̂ )m −−−−→ eixT̂ . (8.22)
m m→∞
Remark 42. Here the similarity to the relation between our time-evolution operator and our Hamil-
toniann for time-independent systems should become apparent.
The n generators of an n dimensional Lie group form a Lie algebra. A Lie algebra is a vector
space G with a map [·, ·] : G × G → G, a so-called Lie bracket, that maps each ordered pair
of algebra elements to a third algebra element, and that is anti-symmetric, bilinear, and fulfils
the Jacobi identity [A, [B, C]] + [B, [C, A]] + [C, [A, B]] = 0 (just like our Poisson brackets and
commutators).
A simple example of a finite Lie algebra we encounter in quantum mechanics consists of the
position and the momentum operators and the identity, since we have
ˆ
[q̂, p̂] = i~I. (8.23)

This algebra corresponds to the group of translations in phase space.

Another example is the algebra associated to the harmonic oscillator, spanned by the identity,
â, ↠, and N̂ , with
[â, ↠] = I,
ˆ [N̂ , â] = −â, [N̂ , ↠] = ↠. (8.24)
The corresponding group consists of translations and rotations in phase space.
An important consequence for quantum mechanics is that if a Hamiltonian is an element of a
Lie algebra, that is, a linear combination of the Lie algebra elements, K̂j ,
X
Ĥ = cj K̂j , (8.25)
j

then the corresponding time evolution operator can be expressed as

P
Û = ei j αj K̂j
, (8.26)

in what is called the Magnus form, or alternatively as

Y
Û = eiβj K̂j , (8.27)
j
56 CHAPTER 8. QUANTUM DYNAMICS

in the so-called Wei Norman form. Importantly this still holds if the coefficients cj in the Hamil-
tonian are time dependent.
When considering the time evolution of expectation values (in the Heisenberg equation), we
encounter expressions of the form Û † ÂÛ . Explicitly evaluating these can be challenging unless the
Hamiltonian and the observable  are elements of a nice Lie group. In general we have

Theorem 9 (Hadamard lemma).

s2 s3
esX̂ Ŷ e−sX̂ = Ŷ + s[X̂, Ŷ ] + [X̂, [X̂, Ŷ ]] + [X̂, [X̂, [X̂, Ŷ ]]] + ..., (8.28)
2! 3!
where X̂ and Ŷ are linear operators and s ∈ C is a complex constant.

Proof. We consider f (s) := esX̂ Ŷ e−sX̂ as an operator valued function of the complex variable s,
and Taylor expand around s = 0:
X sn
fˆ(s) = fˆ(n) (0) . (8.29)
n
n!
We have
d sX̂ −sX̂
e Ŷ e = esX̂ X̂ Ŷ e−sX̂ − esX̂ Ŷ X̂e−sX̂
ds (8.30)
= esX̂ [X̂, Ŷ ]e−sX̂ ,

d2 sX̂ −sX̂
e Ŷ e = esX̂ X̂[X̂, Ŷ ]e−sX̂ − esX̂ [X̂, Ŷ ]X̂e−sX̂
ds2 (8.31)
= esX̂ [X̂, [X̂, Ŷ ]]e−sX̂ ,
and so forth. Inserting this into the Taylor expansion yields equation (8.28).

Exercise 13. Use the Hadamard lemma and the fundamental commutator to show that

eiap̂1 /~ q̂1 e−iap̂1 /~ = q̂1 + a. (8.32)

8.2.2 Reminder: The group SU (2)

A group that is of great importance in the theory of angular momentum, which we will discuss in
the next Chapter, is the group SU (2). The corresponding Lie algebra su(2) can be spanned by
three elements Jˆk , with k = 1, 2, 3 that fulfill the commutation relations
X
[Jˆj , Jˆk ] = i~ jkl Jˆl , (8.33)
l

where 
 1, j, k, l cyclic
jkl = −1, j, k, l anticyclic (8.34)
0, otherwise,


is the Levi-Civita symbol.

You have probably encountered the group SU (2) before, as the group of complex unitary 2 × 2
matrices with determinant one, i.e. the set of matrices

SU (2) = {S ∈ C2×2 |S † = S −1 and det S = 1}, (8.35)

which is a Lie group under matrix multiplication.

In the mathematics literature, the associate Lie algebra is commonly defined as the set of
matrices generating the group elements via the exponential map, which in the case of SU (2) consists
of complex anti-Hermitian 2×2 matrices with trace zero. In the physics literature we tend to include
8.3. DYNAMICAL TUNNELING 57

a factor of i into the exponent in the exponential map resulting in the algebra of complex Hermitian
2 × 2 matrices with trace 1. A basis for these is given by the matrices
     
0 1 0 −i 1 0
σ̂1 = , σ̂2 = , σ̂3 = . (8.36)
1 0 i 0 0 −1

These matrices are also known as the Pauli matrices. They fulfil the commutation relations
X
[σ̂j , σ̂k ] = 2i jkl σ̂l . (8.37)
l

SU (2) is commonly associated with complex 2 × 2 matrices. However, it has higher dimensional
representations which will be related to quantum mechanical angular momentum, as we shall see in
Chapter 9.

8.3 Dynamical Tunneling

I include the slides from the video on dynamical tunneling here.
58 CHAPTER 8. QUANTUM DYNAMICS
Chapter 9

Angular momentum

9.1 Definition of angular momentum - classical and quantum

In this chapter we move beyond one-dimensional systems and consider angular momentum. In
classical mechanics the angular momentum of a particle moving in three spatial dimensions with
spatial coordinate ~q ∈ R3 and momentum p~ ∈ R3 is given by
~ = ~q × p~,
L (9.1)
or in components
L1 = q2 p3 − q3 p2 , L2 = q3 p1 − q1 p3 , and L3 = q1 p2 − q2 p1 . (9.2)
In quantum mechanics, we can as well consider a particle in three dimensional space R3 , with
position and momentum operators q̂ = (q̂1 , q̂2 , q̂3 )T , and p̂ = (p̂1 , p̂2 , p̂3 )T , whose components fulfil
the canonical commutation relations
[q̂j , p̂k ] = i~δjk , [q̂j , q̂k ] = 0 = [p̂j , p̂k ]. (9.3)
In analogy to classical mechanics we can then define an operator associated to the orbital angular
momentum as
L̂ = q̂ × p̂, (9.4)
that is, an operator with the three components
L̂1 = q̂2 p̂3 − q̂3 p̂2
L̂2 = q̂3 p̂1 − q̂1 p̂3 (9.5)
L̂3 = q̂1 p̂2 − q̂2 p̂1 .
The components of the orbital angular momentum fulfil the commutation relations
[L̂1 , L̂2 ] = i~L3 , [L̂2 , L̂3 ] = i~L1 , [L̂3 , L̂1 ] = i~L2 , (9.6)
which are just the commutation relations of the su(2) algebra, written more compactly as
X
[L̂j , L̂k ] = i~ jkl L̂l . (9.7)
l

In quantum mechanics, we make the following definition:

Quantum angular momentum An operator with three Hermitian operator components Jˆ1,2,3
fulfilling the commutation relations
X
[Jˆj , Jˆk ] = i~ jkl Jˆl , (9.8)
l

is an angular momentum operator.

59
60 CHAPTER 9. ANGULAR MOMENTUM

The total angular momentum operator is then defined as

Jˆ2 = Jˆ12 + Jˆ22 + Jˆ32 . (9.9)

It can be easily verified that [Jˆ2 , Jˆj ] = 0.

Remark 43. In the following we shall use the letter J to denote generic quantum angular momentum,
while we reserve the letter L to denote orbital angular momentum, which can be defined as (9.5) in
some given coordinate system.

9.2 Angular momentum and rotations

The above definition of angular momentum relates it to the group SU (2), which, is a double cover
of the group SO(3), which can be represented as rotations in R3 . Intuitively it makes perfect
sense, that angular momentum is thus related to rotations. Technically speaking, the angular
momentum operators are the generators of infinitesimal rotations. Here we shall investigate an
intuitive connection between angular momentum and rotations, by showing that the time-evolution
where the angular momentum operators act as Hamiltonians, are rotations of the coordinate system.

9.2.1 Reminder: Rotations in R3

The rotations in R3 are a non-commutative group. A rotation in R3 can be described by three
parameters: the direction of the axis of rotation (two parameters) and the angle of rotation, and
can be expressed by an orthogonal (real unitary) 3×3 matrix. There are three fundamental rotations
that can be used as a basis for all other rotations, the rotations around the x, y- and z-axes, whose
matrices are given by:
   
1 0 0 cos φ 0 sin φ
Rx (φ) =  0 cos φ − sin φ  , Ry (φ) =  0 1 0 ,
0 sin φ cos φ − sin φ 0 cos φ

and  
cos φ − sin φ 0
Rz (φ) =  sin φ cos φ 0  .
0 0 1
 
x
Applying for example Rx (φ) to ~r =  y  yields:
z
   
x x
Rx (φ)  y  =  y cos φ − z sin φ  ,
z y sin φ + z cos φ

i.e. a rotation around the x-axis by the angle φ.

9.2.2 Action of the eiφL̂j /~ on the position operators

Consider a quantum system in R3 described by the wave function |ψi. We shall now show that the
time evolution with a Hamiltonian Ĥ = L̂j (orbital angular momentum) corresponds to a rotation
of the coordinate system. The time-evolved wave function is given by |ψ(t)i = e−iL̂j t/~ |ψ(0)i. Thus,
the expectation value of an observable  at time t is given by

hψ(t)|Â|ψ(t)i = hψ(0)|eiL̂j t/~ Âe−iL̂j t/~ |ψ(0)i. (9.10)

9.2. ANGULAR MOMENTUM AND ROTATIONS 61

Alternatively, we can interpret this as the expectation value in the wave function at time zero of
the operator eiL̂j t/~ Âe−iL̂j t/~ . Let us consider the expectation values of the position operators q̂1,2,3 .
Denote the expectation value of q̂ at time zero by the vector Q ~ = (Q1 , Q2 , Q3 )T ∈ R3 .
To calculate eiL̂j t/~ q̂k e−iL̂j t/~ we can use the Hadamard lemma 8.28. From the definition of the
components of the orbital angular momentum (9.5) we find the commutators
[L̂j , q̂k ] = i~jkl q̂l . (9.11)

Using these and the Hadamard lemma we can calculate the eiφL̂j /~ q̂k e−iφL̂j /~ . Let us consider the
examples eiφL̂1 /~ q̂j e−iφL̂1 /~ . Since [L̂j , q̂j ] = 0, we trivially have

eiφL̂1 /~ q̂1 e−iφL̂1 /~ = q̂1 . (9.12)

To evaluate eiφL̂1 /~ q̂2 e−iφL̂1 /~ we need to calculate the nested commutators [L̂1 , [L̂1 , [..., [L̂1 , q̂2 ]...]]].
This can be done iteratively, starting from [L̂1 , q̂2 ] = i~q̂3 :

[L̂1 , [L̂1 , q̂2 ]] = i~[L̂1 , q̂3 ] = ~2 q̂2

[L̂1 , [L̂1 , [L̂1 , q̂2 ]]] = ~2 [L̂1 , q̂2 ] = i~3 q̂3
[L̂1 , [L̂1 , [L̂1 , [L̂1 , q̂2 ]]]] = i~3 [L̂1 , q̂3 ] = ~4 q̂2
..
.
Remembering the Taylor expansions of sin and cos:
x3 x5 x7
sin(x) = x − + − + ...
3! 5! 7!
x2 x4 x6
cos(x) = 1 − + − + ...
2! 4! 6!
we realise that
eiφL̂1 /~ q̂2 e−iφL̂1 /~ = q̂2 cos(φ) − q̂3 sin(φ).
Similarly we find
eiφL̂1 /~ q̂3 e−iφL̂1 /~ = q̂3 cos(φ) + q̂2 sin(φ),
and so on. That is, we have
~
Q(t) := hψ(0)|eiL̂j t/~ q̂e−iL̂j t/~ |ψ(0)i = Rj (t)Q,
~ (9.13)
~ is the vector of expectation
where Rj (t) is the rotation matrix around the axis j by an angle t, and Q
values of the position operators at time zero.

ˆ
9.2.3 Action of the eiφJj /~ on the angular momentum operators
ˆ ˆ
Similarly to the considerations above we can use the Hadamard lemma to calculate eiφJj /~ Jˆk e−iφJj /~ :
Trivially we have
ˆ ˆ
eiφJj /~ Jˆj e−iφJj /~ = Jˆj . (9.14)
ˆ ˆ
Let us now consider the example eiφJ1 /~ Jˆ2 e−iφJ1 /~ . Again, we need to calculate the nested com-
mutators [Jˆ1 , [Jˆ1 , [..., [Jˆ1 , Jˆ2 ]...]]]. Similarly to above, this can be done iteratively, starting from
[Jˆ1 , Jˆ2 ] = i~Jˆ3 :

[Jˆ1 , [Jˆ1 , Jˆ2 ]] = i~[Jˆ1 , Jˆ3 ] = −~2 Jˆ2

[Jˆ1 , [Jˆ1 , [Jˆ1 , Jˆ2 ]]] = −~2 [Jˆ1 , Jˆ2 ] = −i~3 Jˆ3
[Jˆ1 , [Jˆ1 , [Jˆ1 , [Jˆ1 , Jˆ2 ]]]] = −i~3 [Jˆ1 , Jˆ3 ] = ~4 Jˆ2
..
.
62 CHAPTER 9. ANGULAR MOMENTUM

And thus,
ˆ ˆ
eiφJ1 /~ Jˆ2 e−iφJ1 /~ = Jˆ2 cos(φ) − Jˆ3 sin(φ).

Similarly we find
ˆ ˆ
eiφJ1 /~ Jˆ3 e−iφJ1 /~ = Jˆ3 cos(φ) + Jˆ2 sin(φ),

and so on. In summary we have

ˆ ˆ −iJˆj t/~ = Rj (t)J,
ˆ
eiJj t/~ Je (9.15)

where Rj (t) is the rotation matrix around the axis j by an angle t.

Exercise 14. Prove that Jˆ2 and Jˆ3 have the same eigenvalues.

9.3 Spectral properties of angular momentum

We now continue to determine the eigenvalues of quantum angular momentum. Similar to the
case of the harmonic oscillator, this can be done by purely algebraic methods. We start out with
some considerations about orbital angular momentum, where we shall deduce that the possible
eigenvalues of the components of orbital angular momentum are integer multiples of ~. We shall
then move on to find that the commutation relations defining angular momentum would also allow
for half-integer values. Half-integer angular momentum does indeed occur in nature, though it
cannot be related to orbital angular momentum of particles, as the latter would have to be integer
valued. Half-integer valued angular momentum appears as an internal degree of freedom, in the
spin of quantum particles. We shall briefly discuss this quantum spin in section 9.5.

9.3.1 Orbital angular momentum

Let us consider the possible eigenvalues of L̂3 = q̂1 p̂2 − q̂2 p̂1 . One can show that these will be given
by integer multiples of ~. This is done by introducing the new set of operators x̂ and ŷ, and p̂x and
p̂y , defined as

1 1
x̂ = √ (q̂1 + p̂2 ), p̂x = √ (p̂1 − q̂2 )
2 2
1 1
ŷ = √ (q̂1 − p̂2 ), p̂y = √ (p̂1 + q̂2 ),
2 2

which also fulfil the fundamental commutation relation of position and momenta, and rewriting L̂3
in terms of these operators as

1 2
1 2
p̂x + x̂2 − p̂y + ŷ 2 .


L̂3 = (9.16)
2 2

In this way, we can interpret L̂3 as the difference of two harmonic oscillators Ĥx and Ĥy , with
masses m = 1 and frequencies ω = 1. Remembering that the eigenvalues of the harmonic oscillator
p̂2
Ĥ = 2m + 12 mω 2 q̂ 2 are given by ~ω(n + 12 ), where n are the non-negative integers, we thus deduce
that the eigenvalues of L̂3 are given by

1 1
λn = ~(nx + ) − ~(ny + ) = ~(nx − ny ), (9.17)
2 2

where nj are non-negative integers. Thus, the eigenvalues of L̂3 are integer multiples of ~, if L̂3 is
associated to orbital angular momentum. The same argument can be used to deduce the eigenvalues
of L̂1 (and L̂2 ), which are also integer multiples of ~.
9.3. SPECTRAL PROPERTIES OF ANGULAR MOMENTUM 63

9.3.2 Algebraic derivation of the spectral properties

Since the three components of angular momentum do not commute, we cannot measure them
simultaneously. The operator Jˆ2 commutes with all of the components, and thus, it is meaningful
to talk of the total angular momentum and the component in one direction. We recall Theorem
7 from Section 4.3.2, stating that two commuting observables have simultaneous eigenvectors, and
consider the simultaneous eigenvectors of Jˆ2 and one selected component of the angular momentum.
In what follows we arbitrarily choose Jˆ3 :

Jˆ2 |β, mi = ~2 β|β, mi, and Jˆ3 |β, mi = ~m|β, mi, (9.18)

and ask the question: What are the possible values of β and m?
Remark 44. Due to the symmetry of the problem, the eigenvalues of Jˆ1 and Jˆ2 are the same as
those of Jˆ3 .
We will prove the following

Theorem 10. The possible values of β are given by j(j + 1) with 2j ∈ N, and for each value of β
m can take the values −j, −j + 1, . . . , j − 1, j.

We first prove

Lemma 4. For a given β we have m2 ≤ β, that is, m is bounded from above and below.

Proof. Assume that the states |β, mi are normalised to one. Consider the sum of the norms of the
states Jˆj |β, mi, which are positive by definition. That is, we have
X
hβ, m|Jˆj |β, mi = hβ, m|Jˆ2 |β, mi ≥ 0. (9.19)
j

On the other hand we have

hβ, m|Jˆ2 |β, mi = ~2 β, and hβ, m|Jˆ32 |β, mi = ~2 m2 . (9.20)

Now we don’t know what the hβ, m|Jˆ1,2

2 |β, mi are exactly, but we do know that

hβ, m|Jˆ1,2
2
|β, mi ≥ 0, (9.21)

and thus
X
hβ, m|Jˆj |β, mi = hβ, m|Jˆ2 |β, mi
j (9.22)
2 2 2
~ β≥~ m .

The remainder of the proof of Theorem 10 proceeds very similarly to the case of the harmonic
oscillator treated in section 5.1. We first introduce the set of adjoint operators

Jˆ± = Jˆ1 ± iJˆ2 . (9.23)

We will need the commutators

[Jˆ3 , Jˆ± ] = ±~J± and [Jˆ+ , Jˆ− ] = 2~Jˆ3 , (9.24)

and the identities

Jˆ− Jˆ+ = Jˆ2 − Jˆ32 − ~Jˆ3 , and Jˆ+ Jˆ− = Jˆ2 − Jˆ32 + ~Jˆ3 . (9.25)
64 CHAPTER 9. ANGULAR MOMENTUM

Exercise 15. Verify the commutation relations (9.24) and (9.25) using the defining commutator
relations (9.8).
The operators Jˆ± now play the same role as the â(†) for the harmonic oscillator. We have the
following two lemmas:
Lemma 5. If |β, mi is an eigenvector of Jˆ3 with eigenvalue ~m, then Jˆ+ |β, mi is either the zero
vector or also an eigenvector of Jˆ3 with eigenvalue ~(m + 1).
Lemma 6. If |β, mi is an eigenvector of Jˆ3 with eigenvalue ~m, then Jˆ− |β, mi is either the zero
vector or also an eigenvector of Jˆ3 with eigenvalue ~(m − 1).
Exercise 16. Proof Lemma 5 and Lemma 6.
Since m2 ≤ β there has to be a maximal value mmax = j, for which

Jˆ+ |β, ji = 0, (9.26)

and a minimal value mmin = k, for which

Jˆ− |β, ki = 0, (9.27)

and m runs in integer values from k to j. Applying Jˆ− to equation (9.26), and using that Jˆ− Jˆ+ =
Jˆ2 − Jˆ32 − ~Jˆ3 yields

Jˆ2 − Jˆ32 − ~Jˆ3 |β, ji = 0

(9.28)
= ~2 β − ~2 j 2 − ~2 j |β, ji,

where |β, ji is assumed not to be the zero vector. That is,

β = j 2 + j = j(j + 1). (9.29)

On the other hand, applying Jˆ+ to equation (9.27), and using that Jˆ+ Jˆ− = Jˆ2 − Jˆ32 + ~Jˆ3 yields

Jˆ2 − Jˆ32 + ~Jˆ3 |β, ki = 0

(9.30)
= ~2 β − ~2 k 2 + ~2 k |β, ki,

where |β, ki is not the zero vector either, and thus

β = k 2 − k = k(k − 1). (9.31)

Combining equations (9.29) and (9.31) yields

k(k − 1) = j(j + 1), (9.32)

which has the two possible solutions

k =j+1 or k = −j, (9.33)

we know, however, that k is the minimal value of m and j the maximal, that is k ≤ j, which means
we have
k = −j. (9.34)
Thus we have deduced that m runs in integer steps from −j to j, from which it follows that
2j is an integer, and the value of β is related to j by β = j(j + 1), which completes the proof of
Theorem 10.
Remark 45. The eigenstates are usually labelled by the numbers j and m, rather than β and m.
Remark 46. For any given value of j there are 2j + 1 values of m.
9.4. MATRIX REPRESENTATION IN THE “STANDARD BASIS” 65

9.4 Matrix representation in the “standard basis”

The so-called standard basis of angular momentum is given by the set of states |j, mi fulfilling

Jˆ2 |j, mi = ~2 j(j + 1)|j, mi,

(9.35)
Jˆ3 |j, mi = ~m|j, mi,

with 2j ∈ N and m running in integer steps from −j to j. We shall now deduce the representation
of the angular momentum operators in the standard basis. By definition the matrices of Jˆ2 and Jˆ3
in the standard basis are diagonal. We have

hj 0 , m0 |Jˆ2 |j, mi = ~2 j(j + 1)δj 0 j δm0 m , (9.36)

and
hj 0 , m0 |Jˆ3 |j, mi = ~mδj 0 j δm0 m , (9.37)
We can deduce the matrix elements of Jˆ1 and Jˆ2 from those of Jˆ± . We know that Jˆ+ |j, mi ∝
|j, m + 1i. We can find the proportionality constant from the normalisation condition. We have
1 ˆ 1
hj, m + 1|j, m + 1i =2
||J+ |j, mi||2 = 2 hj, m|Jˆ− Jˆ+ |j, mi
c c
1
= 2 hj, m|Jˆ − Jˆ3 − ~Jˆ3 |j, mi
2 2
c (9.38)
1 2 2 2 2


= 2 ~ j(j + 1) − ~ m − ~ m hj, m|j, mi
c
1
= 2 ~2 j(j + 1) − ~2 m(m + 1) hj, m|j, mi.


c
p
Assuming that hj, m|j, mi = 1, |j, m + 1i is normalised if c = ~ j(j + 1) − m(m + 1), that is,

Jˆ+ |j, mi = ~ j(j + 1) − m(m + 1)|j, m + 1i.

p
(9.39)

Similarly we find
Jˆ− |j, mi = ~ j(j + 1) − m(m − 1)|j, m − 1i.
p
(9.40)
That is we have for the matrix elements of Jˆ+ and Jˆ− :

hj 0 , m0 |Jˆ+ |j, mi = ~ j(j + 1) − m(m + 1)δj 0 j δm0 ,m+1 ,

p
(9.41)
hj 0 , m0 |Jˆ+ |j, mi = ~ j(j + 1) − m(m − 1)δj 0 j δm0 ,m−1 .
p

The matrix elements of Jˆ1 and Jˆ2 then follow from

1ˆ  1 ˆ 
Jˆ1 = J+ + Jˆ− , and Jˆ2 = J+ − Jˆ− . (9.42)
2 2i
1
The angular momentum matrices for the total angular momentum j = 2 are given by ŝj = ~2 σ̂j ,
where σ̂j denotes the Pauli matrices (8.36).

9.5 Quantum Spin

There is a quantum observable whose three components fulfil the angular momentum commutation
relations, which does not describe orbital angular momentum. This observable is the so-called spin
of a particle. It corresponds to an internal degree of freedom. The values of the spin of a particle
in a given direction can be integer or half integer multiples of ~. Although it does not relate to
an actual rotation or “spinning” motion of the particle, it can be transformed into orbital angular
momentum. The spin is a fundamental quantity of a quantum particle, similar to for example, its
mass. That is, the total value (i.e. the value of Ŝ 2 ) is a conserved quantity.
66 CHAPTER 9. ANGULAR MOMENTUM

Remark 47. There are two types of particles in nature, those with integer spin, the so-called bosons,
and those with half-integer spin, the so-called fermions, which behave very differently from each
other. The details of this difference could easily lead us into many more lectures and thus we will
not comment further at this point.
Remark 48. The spin degree of freedom of a spin 21 particle is described by the 2 × 2 representation
of angular momentum, i.e. the Pauli matrices. Together with the identity matrix the Pauli matrices
form a basis for all 2 × 2 matrices. That is, every 2 × 2 matrix can be written as a superposition
of Pauli-matrices, and thus every quantum system with two states can be interpreted as a spin 21
system.

Figure 9.1: Stern-Gerlach experiment, setup, and postcard with the results from Gerlach to Bohr.

9.5.1 A short history

In 1922 a now famous experiment was performed by Stern and Gerlach, which was meant to demon-
strate the quantisation of angular momentum. We now know that this was the first experimental
verification of the quantum spin, even though at the time of the experiment that was not clear,
and even the notion of quantum spin did not exist at the time. In the experiment, the setup of
which is schematically depicted in figure 9.1 on the left, silver atoms were heated in an oven and
allowed to escape and fly through an inhomogeneous magnetic field. The interaction between a
charged particle with a given angular momentum and an inhomogeneous magnetic field leads to a
force that is given by the so-called magnetic moment of the particle multiplied with the gradient
of the magnetic field. The magnetic moment is proportional to the charge and the internal angular
momentum of the particle. Thus, one expects that after crossing the magnetic field the silver atoms
will be displaced from the straight line on which they entered the magnetic field by a distance
proportional to their angular momentum. Stern and Gerlach were lucky to have chosen silver for
their experiment, as the electrons in a silver atom have total orbital angular momentum of zero,
and the total angular momentum of the atom is entirely due to the spin of the outermost electron in
the atom, which has spin 21 (that is, j = 1/2) and thus the beam split into two. Stern and Gerlach
were aware they had measured the quantisation of angular momentum, and sent a postcard to Bohr
(see figure 9.1), congratulating him on the experimental verification of his quantum theory. They
were, however, not aware that they had in fact measured the internal spin degree of the electron, a
9.5. QUANTUM SPIN 67

concept which didn’t even exist at the time. Most certainly they had not done our calculation in
Section 9.3.1, or otherwise they would have been rather surprised to find the initial beam of silver
atoms had split into an even number of beams. It took until 1927 for the Stern-Gerlach experiment
to be connected to the electron spin.
Uhlenbeck and Goudsmith proposed the idea of a spin of the electron around 1925 in their cal-
culations of atomic spectra. At approximately the same time Pauli used the idea of a double-valued
observable and introduced the Pauli-matrices, but said this had “nothing to do with Uhlenbeck’s
and Goudsmith’s spin”... Not much later Dirac introduced a relativistic version of quantum theory,
compatible with special relativity. In this theory, the spin comes out automatically.
Today the quantum spin is at the basis of many technologies, and gains ever more importance.
Spin dynamics is the basic principle behind medical techniques such as MRI, there is an ever
increasing interest in so-called spin-tronics, and the spin is the key ingredient (qubit) in quantum
information.
68 CHAPTER 9. ANGULAR MOMENTUM
Chapter 10

Mastery material: The Husimi

distribution

10.1 Classical and quantum phase-space distributions

In classical mechanics the dynamics is conveniently represented in phase space, spanned by the
classical position and momentum coordinates. Quantum mechanically, however, already the concept
of a point in phase space contradicts the uncertainty principle (∆p∆q ≤ ~2 ), which is a challenge for
the definition of a quantum phase space. However, there are ways to define quantum phase-space
densities in analogy to classical phase-space distributions (as in statistical mechanics). None of them
fulfils all the requirements we have on a classical phase-space density, ρ(q, p), which are mainly

1. A classical phase-space density is real and non-negative, i.e., ρ(q, p) ∈ R and ρ(q, p) ≤ 0 for
all q, p.

2. A classical phase-space density is normalisable, i.e., the integral

Z
ρ(q, p)dpdq

over the whole phase space has to be finite.

3. Expectation values of functions A(p, q) of position and momentum are given by phase-space
integrals Z
hAi = A(p, q)ρ(p, q)dpdq.

4. The p- and q-distributions are given by the marginals

Z
ρp (q) = ρ(p, q)dq,

and Z
ρq (p) = ρ(p, q)dp,

respectively.

There are infinitely many choices of quantum phase-space distributions that fulfil some, but never
all, of the above requirements. The ones most commonly used are the so-called Wigner and Husimi
densities. Here we give a brief overview over Husimi distributions and their properties.
The Husimi distribution is defined as a projection of a state onto coherent states of a given
harmonic oscillator. For this purpose we start with a summary of some of the properties of coherent
states.

69
70 CHAPTER 10. MASTERY MATERIAL: THE HUSIMI DISTRIBUTION

10.2 Coherent states

Coherent states, are minimum uncertainty states, and thus as close to the classical concept of a
phase-space point as possible in quantum mechanics. They can be defined in a number of equivalent
ways. One common way is as the eigenstates of the harmonic oscillator annihilation operator â:

â|zi = z|zi, (10.1)

with eigenvalues z ∈ C. In the harmonic oscillator basis |ni (with Ĥ|ni = ~ω(n + 21 )|ni, where
Ĥ = ~ω(↠â + 21 )), they take the form
|z|2 X |z|n
|zi = e− 2 √ |ni. (10.2)
n n!
Their position representation is given by
 mω 1/4 mω 2
+ ~i px
hx|zi = z(x) = e− 2~ (x−q) , (10.3)
π~
with z = √12 (q + ip). (Verify this by applying the operator â in position representation onto the
wave function (10.3)!)
The coherent states can alternatively be defined as displaced harmonic oscillator ground states
as
|zi = D̂(z)|0i, (10.4)
† ∗
where D̂ = ezâ −z â is the complex displacement operator, and |0i is the ground state of the
harmonic oscillator. The position and momentum probability distributions of this state are those of
the ground state of the harmonic oscillator, only displaced to be centred around q and p respectively.
1
Using that eÂ+B̂ = e eB̂ e− 2 [Â,B̂] , provided that [Â, [Â, B̂]] = 0 = [B̂, [Â, B̂]], we can rewrite D̂
as
|z|2 † ∗
D̂ = e− 2 ezâ e−z â , (10.5)
Applying this to the ground state of the harmonic oscillator and using that
√ √
â|ni = n|n − 1i, and ↠|ni = n + 1|n + 1i,

we can immediately verify (10.2).

Coherent states are not orthogonal to each other. The overlap |hz1 |z2 i|2 of two coherent states
is given by (verify this as an exercise)
2
|hz1 |z2 i|2 = e−|z1 −z2 | . (10.6)

The coherent states form a resolution of the identity with

d2 z
Z
|zihz| ˆ
= I,
π
where d2 z = dRe(z)dIm(z).

10.3 Definition and properties of the Husimi distribution

There is a free parameter in the definition of the Husimi distribution, determined by the frequency
and mass of the harmonic oscillator generating the respective coherent states, and one can adjust this
parameter to yield most insights into the problem at hand. In what follows we shall for simplicity
consider m = 1 = ω, such that we have in the position representation
 1/4
1 1 2 i
hx|zi = z(x) = e− 2~ (x−q) + ~ px . (10.7)
π~
10.3. DEFINITION AND PROPERTIES OF THE HUSIMI DISTRIBUTION 71

Now we identify the expectation values q and p of the coherent state with the phase-space coordi-
nates. The Husimi distribution of a quantum state |ψi is defined as
1
ρH (p, q) = |hz|ψi|2 , (10.8)
2π~
which is equivalent to the basis expansion of |ψi in the overcomplete basis of coherent states |zi,
1
with the normalisation factor 2π~ .
The Husimi distribution is by definition real and positive. It is normalised if |ψi is normalised,
i.e. Z ∞ Z ∞Z ∞
2
|ψ(x)| dx = 1 ⇒ ρH (p, q)dpdq = 1. (10.9)
−∞ −∞ −∞
The expectation values of p̂ and q̂ are given by
Z
hp̂i = pρH (p, q)dpdq, (10.10)

and Z
hq̂i = qρH (p, q)dpdq, (10.11)

but in general Z
hÂi =
6 A(p, q)ρH (p, q)dpdq, (10.12)

The position and momentum probability distributions are not given by the marginal distributions.
Nevertheless the Husimi distribution is useful to visualise quantum mechanics in phase space and
for a direct comparison to classical phase-space structures.
Exercise 17. Assume that the state |ψi is normalised to one, and show that
R
(a) ρH (p, q)dpdq = 1.
R
(b) hâi = zρH (p, q)dpdq,
(c) h↠âi = |z|2 ρH (p, q)dpdq − 1.
R

10.3.1 Examples
The Husimi distribution of a coherent state ψ(x) of the form
 1/4
1 1 2 i
ψ(x) = e− 2~ (x−q0 ) + ~ p0 x . (10.13)
π~
is given by (check!)
1 − 1 ((q−q0 )2 +(p−p0 )2 )
ρH (p, q) = e 2~ , (10.14)
2π~
that is, a Gaussian in q and p.
The Husimi distribution of a harmonic oscillator eigenstate |ni is given by (check!)
1 −|z|2 |z|2n
ρH (p, q) = |hn|zi|2 = e . (10.15)
2π~ n!
Three examples are depicted in the top panel of Figure 10.1. We observe that the Husimi distri-
butions are localised around circles in phase space. These correspond to the classical phase-space
trajectories with the corresponding energy.
PHusimi distribution of an arbitrary state |ψi expanded in the harmonic oscillator basis
The
|ψi = n cn |ni is simply given by
2
1 X z ∗n
ρH (p, q) = cn √ . (10.16)
2π~ n n!
72 CHAPTER 10. MASTERY MATERIAL: THE HUSIMI DISTRIBUTION

Figure 10.1: Husimi distributions of the ground state and second and fifth excited states of the harmonic
oscillator.

10.3.2 Dynamics of the Husimi distribution for the Hermitian Harmonic oscil-
lator
For a harmonic oscillator Hamiltonian, the Husimi distribution (10.16) evolves in time as
2
1 X ((zeiωt )∗ )n
ρ(p, q, t) = cn √ . (10.17)
2π~ n n!

That is, we have

ρ(z, t) = ρ0 (zeiωt ), (10.18)
where ρ0 denotes the initial Husimi distribution. Hence, the dynamics of the Husimi distribution
for a harmonic oscillator is a rigid rotation in phase space with frequency ω for arbitrary initial
states.

10.4 SU(2) coherent states and Husimi distribution

In analogy to the coherent states of the harmonic oscillator one can define SU (2) or angular mo-
mentum states for a given total angular momentum j as
ˆ
|ζi = N (ζ)eζ J+ |j, −ji,

where |j, −ji is an angular momentum eigenstate with

Jˆ2 |j, −ji = j(j + 1)|j, −ji, and Jˆ3 |j, −ji = −j|j, −ji,

−j
and where N (ζ) = 1 + |ζ|2 is a normalisation factor. The complex variable ζ parameterises
the spherical phase space. It can be related to spherical coordinates via the parameterisation
ζ = −e−iφ tan( 2θ ). We can then also define a Husimi distribution on the sphere as

1
ρH (θ, φ) = |hζ|ψi|2 .
4π
Exercise 18. Consider the case j = 12 .

(i) Use that Jˆ+ |j, mi = j(j + 1) − m(m + 1)|j, m + 1i to express the coherent states in the
p

standard basis | 12 , − 21 i, | 12 , 21 i.

(ii) Calculate the Husimi distributions (as functions of φ and θ) of the two basis states.