Contents

1 References 7
1.1 Textbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.1 Classics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.2 Online . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Course Plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Grading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Postulates of QM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Linear Algebra 13
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Classical States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Quantum States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.1 Infinite Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4.2 Function Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Measuring vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.1 Dual vector space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.6 Inner product and Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6.1 Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6.2 Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6.3 Basis for infinite spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6.4 Gram-Schmidt orthonormalization . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.7 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.8 Linear Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.8.1 Eigenvalues and Eigenvectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.8.2 Functions of operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.8.3 Special operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.8.4 Density Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.9 Measurement and Quantum Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.9.1 Compatible observables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.9.2 Transition Amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.9.3 Uncertainty principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.10 More vector spaces: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.10.1 Multiparticle systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.10.2 Entanglement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3 Quantum Phenomena 41
3.1 Double Slit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.1.2 The experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

1
2 CONTENTS

3.1.3 The issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3.2 Double slit as a two level system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3 Schrödinger’s cat and Wigner’s friend . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3.1 Wigner’s Friend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4 Bunching and Anti bunching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.5 Entanglement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6 EPR paradox and Bertlmann’s socks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6.1 Bell’s Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6.2 Mermin’s experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.6.3 GHZ state in QFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7 Cheshire cat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.8 Foundations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4 Schrodinger equation 53

5 Position/Momentum operators 57
5.1 Momentum operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.2 Hermiticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.3 BCH formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

6 Examples 65
6.1 Free Particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.1.1 Hilbert space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.1.2 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.2 Infinite potential well . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.3 SHO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.3.1 Energy levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.3.2 Wavefunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.3.3 Time evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.4 Other instructive examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.5 Finite potential well . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.5.1 Reflection and Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.5.2 Step potential Barrier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.5.3 Same calculations done differently . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.6 Delta function potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.6.1 Repulsive delta and phase shifts . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.6.2 Poles of the S-matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.7 Double potential wells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.7.1 Double potential well . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.8 Pöschl-Teller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.10 MIDSEM II solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

7 Symmetry 87
7.1 Symmetries in Classical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
7.1.1 Symmetries in 1D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
7.1.2 Kroning-Penney . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.1.3 Variations on this theme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
7.2 Higher dimensional problems and symmetry . . . . . . . . . . . . . . . . . . . . . . . . . 93
7.3 Central potentials and Angular Momentum . . . . . . . . . . . . . . . . . . . . . . . . . 93
7.3.1 Free particle in 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.3.2 2D SHO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
CONTENTS 3

7.4 Particle in constant electric/magnetic fields . . . . . . . . . . . . . . . . . . . . . . . . . 96

7.4.1 Constant magnetic field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
7.5 Motion in 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
7.6 Angular momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
7.6.1 Spherical Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
7.6.2 Legendre Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.7 Spin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.8 Total Angular Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.9 notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.9.1 What is an example of a representation of U(1) that is not unitary? . . . . . . . 107

8 QM in 3D 109
8.1 Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
8.2 Free particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
8.3 Infinite Potential well 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
8.4 3D SHO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
8.5 Hydrogen Atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

9 Additional reading 113

4 CONTENTS

These lecture notes are based on the lectures given at IISER Mohali as part of the course PHY302:
Quantum Mechanics I during the Odd Terms of 2023.

Course Outline

• Classical vs. quantum Mechanics, Simple 2-state QM system. Hilbert Spaces, Operators. Ob-
servables - Compatible Observables, Tensor Product Spaces, Uncertainty Relations. Position,
Momentum and Translation. Eigenvalue Problems. Emphasis on Linear Vector Spaces from a
mathematical point of view.

• Energy-time Uncertainty, and Time Evolution (Quantum Dynamics). Schroedinger, Heisenberg

and Interaction Pictures; Interpretation of Wavefunction. Quantum Particles in Potential. Har-
monic oscillator.

• Angular Momentum. Rotation in Quantum mechanics. SO(3) vs. SU (2). Spherical Harmonics.
Addition of Angular Momenta.

• Single electron atoms: Spherically symmetric potentials, spherical harmonics, Hydrogen atom
problem, solution of Schroedinger equation, energy levels and eigenfunctions, orbital angular
momentum, Hydrogenic atoms.

The next semester’s syllabus is more technical and will focus on computations. Both exact and
approximate methods of computation.

• Time independent and time dependent perturbation theory. Transitions under the action of a
perturbation acting for a finite time, Transitions under the action of a periodic perturbation.

• Review of Hydrogen atom, Fine structure, Hyperfine structure as an application of perturbation

theory to real systems. Zeeman effect, Stark effect.

• The semi-classical approach, WKB approaximation. Penetration through a potential barrier.

The variational principle, applications.

• Scattering theory: Scattering cross-section, partial waves, Yukawa and Coulomb potentials, scat-
tering by square well potential, reaction rates, mean free path, retarded potentials, Born approx-
imation.

• Relativistic quantum mechanics: Klein-Gordon equation, negative probabilities. Dirac equation,

relativistic free particle solutions, negative energy solutions, anti-particles. Zitterbewegung.

I propose to spend a lot of time on the weirdness of Quantum Mechanics in developing points 1
and 2 of the first part.

Course Outline:
Quantum mechanics marks a paradigm shift in the way we approach Natural Phenomena.
In classical physics, by which we mean, physics rooted in the Ideas of Newton, Euler, Maxwell,
Kelvin, Joule and Gauss - we describe Natural phenomena by using concepts which are directly ‘ele-
ments of reality’. Thus, in Newtonian mechanics we use balls, levers, fluids, planets, gases, liquids etc.
And additionally, a revolutionary idea called fields (electric, magnetic, gravitational).
For many purposes we can think of fields as liquids and liquids as fields. Thus, fields have weight
momentum and energy and can be compressed etc. Similarly, regarding liquids as fields allows us to
write down equations for them (Euler, Navier-Stokes etc)
Quantum Mechanics marks a radical departure from the above.
CONTENTS 5

The elements that make up the quantum description of reality have no correspondence to ‘elements
of reality’. That is to say, they do not ‘exist’ out there. Instead, they are to be solely regarded as
devices of calculation.
This opens up a new frontier of thought - namely the connection between the calculation and the
experiment. And as we shall see, this is the source of a tremendous amount of confusion - and also a
huge open area of research.
In the first part of the course, I shall focus on the elements of a quantum description of nature.
This will not take much time - but the difficulty is in wielding these elements.
We will then discuss the Born rule and the Copenhagen interpretation of quantum mechanics. This
will set up the basic elements in using Quantum Mechanics.
In physics, we are always concerned with states and how they evolve in time. Thus, we will next
turn to dynamics - given states, how do they time evolve?
The remaining two parts of the course will be entirely standard: We will simply study examples
which are technically more complicated, but conceptually nothing new.
6 CONTENTS
Chapter 1

References

Three categories:

1.1 Textbooks
• Griffiths [4]

• Sakurai, J. J

• Ballentine

• Weinberg, S. W.

• Shankar, R.

• Cohen-Tannoudji

• Landau

• Schiff

• Dicke & Wittke

• Rajasekar & Veluswamy

• Migdal,

• Tong, D.

• Thankappan

• Rajasekhar and Ponnuswami

• Liboff

• Bohm, D.

• Gallindo Pascual

7
8 CHAPTER 1. REFERENCES

1.1.1 Classics
1.1.2 Online
David Tong has very nice notes
The main two textbooks for this course are going to be
Ballentine, “Quantum Mechanics.”
Griffiths, “Quantum Mechanics.”
along with Feynman Lectures vol III.
In addition, we will use Resonance, American Journal of Physics and Physics Today for supple-
mentary material.

1.2 Course Plan

1. Vector spaces, Norms and Measurements [10/08-18/08] 5 hours

2. Operators, Compatibility and Measurements [21/08-28/08] 5 hours

3. Superpositions, Double Slit and Measurements. [29/08-01/09] 3 hours

4. Bells Inequalities, EPR, Uncertainty [04/09-09/09] 3 hours

5. Midsem exam

6. Tensor products, Entanglement, Purification. [14/09-18/09] 3 hours

7. Schrodinger equation [19/09-22/09] 3 hours

8. Position and Momentum Operators [25/09-26/09]

9. Harmonic Oscillator, Finite potential well. 12 hours

10. Midsem

11. Symmetry and Rotations. [30-10,3-11]

12. Angular momenta, Spherical Harmonics, [3-11,4-11]

13. Angular momentum addition. Tensor products [6-11, 7-11]

14. 3D problems: Free particle, 3D SHO. [9-11/11]

15. Hydrogen atom [13-11 - 17-11]

16. Helium and other approximations [20-11 to 24-11]

1.3 Grading
The grading will consist of

• Midsem 2 X 20%

• Attendance 10% You will not be allowed to attend the final exam if below 65%.
65-75% attendance: 0-10% credit.
75% and above. 10% credit.

• Endsem 50%
1.4. POSTULATES OF QM 9

Tutorials will be conducted as and when needed.

Attendance will be important for this course.
Topic hours
Intro & Postulates 10-8
finite dim V 11-8
Linear Indep & infin V & fn sp 14-8
Function spaces, Basis and Dimension 17-8
Duals 18-8
Inner prod & Norm, Cauchy-Schw & Triang 21-8
Operators & Eigenvectors 24,25-8
Adjoints, Hermitean 28-8
Compatible, CSCO, Measurements 29-8
Finite dim example 29-8
Finite examples contd 31-8
Multiparticles 1-9
Sch cat 4-9
EPR 5-9
Double Slit 7,8-9
Midsem 13-09
Direct Sum and Tensor product 1
Entanglement & Purification 1
Schrodinger equation 18-19/09
Position and Momentum Operators 21,22/09
Free particle 25,26,29/09
Potential well 3,5,6/10
SHO 9-13/10
Particle in electric/magnetic fields 16,17/10
Midsem 19-21/10
Symmetries & Wigner’s theorem 30,31/10
Bloch’s Theorem 2/11
2D problems: Band Structure of graphene 3,6,7/11
3D problems: angular momentum 9,10,13,14/11
Separation of variables 16/11
Free particle 16-17/11
Hydrogen 20-24/11
Quantum topics

• Hong–Ou–Mandel interference

• Bells Inequalitites

• Entanglement

• Purification

• EPR

• Cheshire Cat, Weak Measurements, Delayed Choice, Superoscillations

1.4 Postulates of QM
This course will present QM from a modern perspective. Therefore, we will present the postulates of
Quantum Mechanics.
10 CHAPTER 1. REFERENCES

• The states of a Quantum System are vectors (rays) in a vector space.

• Multiparticle states are assumed to be contained in the tensor product of the single particle
vector spaces (not Cartesian products).

• For identical particles, the tensor product states are to be either symmetrized (Bosons) or anti-
symmetrized (Fermions).
• The dynamics of the system is governed by the Schrodinger equation.
• Experimental quantities are determined by the bilinear scalars constructed from the vectors.

• A measurement is defined by a Hermitean operator.

• The outcome of the measurement can be any of the possible eigenvalues of the operator.
A subtle implication is that if the system is NOT in an EIGENstate - then it cannot be set
to possess the property being measured (the property is defined by the operator). For, if the
system ‘possesses’ the property, then the outcome of the experiment must definitely be a fixed
value (think ‘color’ and ‘red’) - even if the experiment to measure was not conducted.
• The effect of a measurement is to ‘collapse’ the state of the system to an eigenstate of the operator
defining the measurement.
This has the following subtle implication. All the possible states of the system are obtainable as
linear combinations of the eigenstates of the measurement operator.
Another subtler idea is that this must be true for every possible measurement that can be
performed on this system. Can it be that the eigenvectors defines only a dense subspace?
This introduces a fundamental irreversibility in the evolution of the system.
• Quantum Mechanics predicts only the relative probabilities of the occurrence of various eigen-
values.

Contrast with classical dynamics where the state of the system is precisely and reversibly determined.
Further, in classical mechanics, it is assumed that one can perform measurements without disturbing
the system.
We will learn the details of how these postulates are managed in this course.
For this part, the postulates are listed in various places in various ways.
See Cohen-Tannoudji, Ballentine, Wikipedia etc.
Bibliography

[1] P. Dennery and A. Krzywycki. Mathematics for Physicists. Dover, 1996.

[2] Yaakov Fein; et al. Quantum superposition of molecules beyond 25kda. Nature Physics,
15(12):1242–1245, 2019.
[3] R. P. Feynman. The Feynman Lectures on Physics, vol III. Narosa, Delhi, third edition edition,
2018.
[4] David J. Griffiths and Darrell F. Schroeter. Introduction to quantum mechanics. Cambridge
University Press, Cambridge ; New York, NY, third edition edition, 2018.
[5] Tongcang Li and Zhang-Qi Yin. Quantum superposition, entanglement, and state teleportation
of a microorganism on an electromechanical oscillator. Science Bulletin, 61(2):163–171, January
2016.
[6] Olaf Nairz, Markus Arndt, and Anton Zeilinger. Quantum interference experiments with large
molecules. American Journal of Physics, 71(4):319–325, 04 2003.
[7] R. L. Pavelich and F. Marsiglio. Calculation of 2D electronic band structure using matrix mechanics.
American Journal of Physics, 84(12):924–935, 12 2016.

[8] Peter Rodgers? The double-slit experiment. Physics World, 2002.

11
12 BIBLIOGRAPHY
Chapter 2

Linear Algebra

2.1 Introduction
Classical states are points in configuration/phase space i.e., points on the tangent or cotangent bun-
dles. Classical systems have trajectories - sections of these bundles. Multi-particle type systems have
Cartesian products as phase spaces, but in this latter case, the trajectories are Cartesian products of
trajectories only if the particles are not interacting.
Classical particle interactions are local - i.e., two particles have to be at the same physical location
to change their respective trajectories. It is also possible to imagine that three particles have to be at
a location simultaneously for something to happen (nuclear potentials).
Classical field theories are also modelled on locality. A classical particle at a particular location (in
phase space) will only be affected by the field value at that ‘event’. Same goes for fields.
In statistical physics, we can have more complicated ideas such as Markov processes - where the
interaction depends on the history.
This picture is drastically altered in Quantum theory. The states are vectors in a vector space -
this also permits the states to be density matrices.
Is it possible to introduce the material without actually specifying whether the state is a vector or
a density matrix. Are the foundations of QM agnostic about whether the states are vectors or density
matrices.

2.2 Classical States

By state of a system, we mean a specification of a list of ‘variables’ using which we can predict the
future states of the system over indefinite periods of time.
In doing so, we will use the ‘laws of motion’ which allow us to work out the future development of
the current state. Often, these laws of motion take the form of differential equations, although this is
by no means necessary.
The courses of Classical Mechanics and Electrodynamics will explain how this is done. In classical
mechanics, we will learn about the states of and equations for systems of point-like particles. Electro-
dynamics is the classical mechanics of extended entities - called fields. In this sense, Electrodynamics
is your first ‘Field theory’. Electrodynamics is a complicated field theory though, being relativistic as
well as being a gauge theory.
The interesting thing about these classical objects - point particles and fields - is that you can
directly experience them. To put it strikingly - you can be ‘slapped’ by both point particles and by
fields.

13
14 CHAPTER 2. LINEAR ALGEBRA

Exercise 1
Explain how you can contrive to ‘slap’ someone using electromagnetic fields?
How about a ‘sound slap’ ?

Thus, they are objects with properties and existing out there (even when no one is looking).
Specifying the list of properties allows us to pin down the object entirely to the extent that its future
states are then determined by applying the laws of Classical Mechanics. This list of properties (or
attributes) is the coordinate space of classical mechanics. Sometimes these are called generalized
coordinates. There is a lot of freedom in defining these coordinates.
Examples:
Phase space: In classical mechanics, the state of a set of point particles is completely specified
by listing the position variables and the momentum variables of each particle at an instant of time.
The position variables are specified by using a coordinate system - but any coordinate system will
be fine. The momentum variables are related to the time derivatives of the position variables. But
this need not take the simple form p = mẋ.

Exercise 2
Two interesting examples are
pϕ = mr2 sin θ ϕ̇
and
px = mẋ − qAx (x, t)
Where do these come from?

Phase space is the total space of all position and momentum values that the system can take.
We have to also specify which momentum is ‘conjugate’ to which coordinate (‘symplectic structure’).

2.3 Quantum States

The states in Quantum mechanics are vectors i.e., elements (points) of a complex vector space.
By definition, vectors can be added. Thus, Quantum states can be superposed. This brings
a variety of new consequences which will be our focus during this course.
There is no sense in ‘adding’ states in classical mechanics

Exercise 3
Construct examples illustrating the absurdity of such additions.
To answer the question, you will need to define the meaning of addition.

However, to study a system, we not only would like to know its states - but we would also like to
apply forces on the system, and make measurements of the system.
In classical mechanics, our ideas of forces are often mechanical in origin - i.e, these are ‘contact’
type. The discovery of gravitation and electromagnetism gave rise to the concept of fields and how
fields act on systems. This allowed us to avoid ‘action at a distance’ - but fields are extended things.
Since the system is a trajectory or a collection of trajectories (many body system), forces always
act at the location of the particle(s).
2.4. VECTOR SPACES 15

And the effect of the force is to change the future states in a quantifiable manner. ‘Forces’ are agents
of ‘change’. Note that no forces means no change - in particular constant velocity (which leads to a
change in position) is not considered a ‘change’. This is of course the Galilean principle of relativity?
In quantum mechanics, the notion of a trajectory does not exist - and so, we need to
learn a new language of how ‘forces’ are described and how measurements are implemented.

2.4 Vector Spaces

Recall the definition of a finite dimensional vector space.
A vector space is an Abelian group V taken together with a field F, and a specific ring homomor-
phism from F to end(V).
The point of this abstract definition is to make it clear that vectors are abstractly conceived. They
are not column (or row vectors).
To begin with, we will work with the simplest examples of vector spaces: Finite dimensional vector
spaces.
The key property of vectors which are the elements making up the vector space V, is that they can
be linearly combined to give other vectors.

α|v1 ⟩ + β|v2 ⟩ ∈ V (2.1)

Here α, β belong to the Field F . In our case, this will always be C, the space of complex numbers.
Unless otherwise indicated, assume that the vector space is defined over the Field of complex numbers.
The natural question to ask is - do these form the states of some quantum system? The answer
is yes, but approximately yes. More generally, the states of most quantum systems form Infinite
dimensional vector spaces which bring in several new features.

Exercise 4
Consider the set of ‘colors’. Do they form a vector space?
How about ‘smells’ ?
Positions of objects are not vectors in this sense.

If a vector cannot be written as a linear combination of a set of vectors, then it is said to be linearly
independent of that set.
More precisely, |v1 ⟩ and |v2 ⟩ are linearly independent if

a|v1 ⟩ + b|v2 ⟩ = 0 =⇒ a = b = 0. (2.2)

i.e., linearly dependent vectors “point in the same direction”. Note that the single equation gives a
solution for two variables a, b.

Exercise 5
Test this idea on the set of colors

Finite dimensional vector spaces (by definition) have finitely many mutually linearly independent
vectors. This number is called the Dimension of the vector space. The maximal set of linearly
independent vectors is called a Basis of that vector space.
16 CHAPTER 2. LINEAR ALGEBRA

Exercise 6
Test this out using R3 .
Try to write a given vector as at linear combination of some vectors. Is this unique?
Test this out using End(R3 ) - ie 3 × 3 matrices.

Exercise 7
Zero vector vs. Zero
|0⟩ + |v⟩ = |v⟩. (2.3)
But,

Exercise 8
The set of symmetric polynomials of degree < J in three variables with complex coefficients
forms a vector space.
How many mutually linearly independent vectors?

We have a very important theorem about vector spaces: Every vector space has a basis. Even
infinite dimensional ones.

Exercise 9
It is easy to prove this for finite vector spaces.
What is the key difficulty in the infinite case.

How can we ‘recognize’ a vector space?

Important to note, we have not discussed anything about dot products of vectors. This is an extra
structure that we impose and is done so arbitrarily.

2.4.1 Infinite Vector Spaces

In such a vector space, there are infinitely many linearly independent vectors.
Some interesting questions: How can we write uncountably many linearly independent vectors?
The reals R form an infinite dimensional vector space over Q. Does it follow that there are infinitely
many irrational numbers?
It maybe clear that there are infinitely many linearly independent vectors. Nonetheless, all reals
have a decimal expansion. So we have a countable basis?
Yes: because R is complete - every convergent sequences tends a limit which is also in R.

2.4.2 Function Spaces

The textbook of Dennery [1] is an excellent reference for this topic. In these cases, the key point to
appreciate is the notion of addition.
Are there other ways to define addition of functions which will still lead to vector spaces?
Ex. 10 — Set of functions on I = [0, 1]
2.5. MEASURING VECTORS 17

Ex. 11 — Set of continuous functions

Ex. 12 — Set of differentiable functions.
Ex. 13 — Set of functions on I that vanish at the endpoints.

Ex. 14 — What if we require f (0) = 0 and f (1) =?

Some further problems are in Griffiths etc.
There are other natural occurrences of vector spaces. For instance,
Ex. 15 — Set of solutions to linear equations with coefficients from a field F.

Ex. 16 — Set of solutions to homogeneous linear ODEs

Ex. 17 — Set of solutions to PDEs with appropriate boundary conditions. Think of physical exam-
ples.

Ex. 18 — Do physics eqns always give vector spaces?

Ex. 19 — Set of convergent Taylor series.

Exercise 20
Basis vs Hilbert Basis

These are our space of states. A particular state is then simultaneously a vector as well as a
function. When we think of it as a vector, the values the function takes are not under focus.
To get a grip on QM and its interpretation, we must understand states. Write/sketch a variety of
states.
Ex. 21 — Observe maxima minima zero crossings.

Ex. 22 — Plot |ψ|2 and compare it with ψ. Note that |ψ|2 < |ψ| always.

Ex. 23 — Plot ψ1,2 and |ψ1 + iψ2 |2 and contrast it with the individual ψ

2.5 Measuring vectors

Experiments convert vectors to numbers - at least to an interval of real numbers.
Thus, an experiment can be thought of as a function

f :V→C (2.4)

Actual experiments give real numbers - but this is clearly an easy generalization. So, if the system is
in state |ψ⟩, we will get the number f (|ψ⟩).
Lets try some simple examples of functions on vectors.

2.5.1 Dual vector space

It is important to learn the idea of a dual vector space. This becomes especially important in the
infinite situation.
Given any set (such as the vector space ), we can think of functions from that set to R or C. In the
case of a vector space, linear functions are specially important.

F (α|v1 ⟩ + β|v2 ⟩) = αF (|v1 ⟩) + βF (|v2 ⟩) (2.5)

18 CHAPTER 2. LINEAR ALGEBRA

Clearly, these are defined by specifying their values on the basis vectors ei . Note that this does not
happen for functions in general. We will denote these linear functions by the Dirac notation: ⟨F |

⟨F |(α|v⟩ + β|w⟩) = α⟨F |v⟩ + β⟨F |W ⟩ (2.6)

These exercises will help you become comfortable with Vector spaces.
Ex. 24 — Show that these linear functions themselves form a vector space. This space is called the
dual vector space of V and is denoted V ∗ .

Ex. 25 — Show that V ∗ has the same dimensions as V .

Ex. 26 — Observe that, if we are given a basis |ei ⟩ for V , there is a special basis for ⟨ej | for V ∗ . We
can construct the dual basis as
⟨ej |(|ek ⟩) = ⟨ej |ek ⟩ = δkj (2.7)

Ex. 27 — Using the dual basis, we can define the components ci of a vector |ψ⟩ = i ci |ei ⟩ in the
P
basis |ei ⟩ quite simply ci = ⟨ei |ψ⟩
Since we have already talked about function spaces, we can think of the dual spaces of these vector
spaces. One of the interesting dual vectors in a function space is the valuation map:

F (ψ(x)) = ψ(x0 )

for some choice of the point x0 in the domain. Another interesting linear functional is
Z b
F (ψ) = ψ(x′ )dx′
a

A challenging question is: how can we write a formula for the dual basis? We will discuss this after
the next two topics.

2.6 Inner product and Norm

Given a vector space and vectors, there are two ways to manufacture scalars which are of fundamental
importance.

2.6.1 Inner Product

Given two vectors |v⟩, |w⟩, the inner product is denoted

g(|v⟩, |w⟩) = ⟨v|w⟩ (2.8)

This is a complex function of the two variable vectors such that

• g(|v⟩, |w⟩ + α|u⟩) = g(|v⟩, |w, ⟩) + αg(|v⟩, |u⟩)

• g(|v⟩, |w⟩) = g(|w⟩, |v⟩)∗ with ∗

denoting complex conjugation

• The previous two imply g(|v⟩ + α|w⟩, |u⟩) = g(|v⟩, |u⟩) + α∗ g(|w⟩, |u⟩)

• g(|v⟩, |v⟩ ≥ 0

• g(|v⟩, |v⟩) = 0 iff |v⟩ = 0

2.6. INNER PRODUCT AND NORM 19

Thus, given a vector |v⟩, we can make a real number by its inner product with itself g(|v⟩, |v⟩).
Inner products are, in general, complex. This is because our Field is the field of complex numbers.

Exercise 28
Write down three inequivalent inner products on R3 .
First think about what inequivalent can mean?

Exercise 29
Suppose we think of R3 as the vector space of 3 × 3 real antisymmetric matrices, i.e, the Lie
Algebra so(3) - check that
⟨v|w⟩ = T r(v T w) (2.9)
defines an inner product. Is this different from the others?

Exercise 30
R3 is also the space of 2 × 2 Hermitean complex matrices, i.e., the Lie algebra su(2)

m = (mT )∗ (2.10)
∗ T
where denotes complex conjugation and stands for transpose.

Amusingly, we can use the inner product to define a linear functional:

⟨fg | = g(|f ⟩, .)

where the dot in place of the second vector tells us that it is a slot to be filled with a vector. The
subscript on the bra is to point out that it is being defined wrt to the inner product g. Thus we have
a correspondence between kets and bras (ie vectors and dual vectors) defined by the inner product.
Note that g(., |f ⟩) is not an element of the dual vector space. Why not?
Earlier, we had defined ⟨ej | was defined as the dual basis vector wrt to the basis |ei ⟩ by the rule
⟨ej |(|ek ⟩) = δkj . This idea
P does not use the inner product g at all. So, we can imagine defining the dual
to the ket |v⟩ as ⟨v| ≡ i ci ⟨ei | where ci are the components of |v⟩ in the basis |ei ⟩.

Exercise 31
How does this differ from the dual bra defined by using the inner product.

In Quantum mechanics, by ⟨f | we will always mean the dual vector defined by g(|f ⟩, −).
Orthogonal vectors are those whoseP inner product is zero.
The Component ci of a vector |v⟩ = i ci |ei ⟩ can be calculated using the dual basis as ci = ⟨ei |(|v⟩).
Depending on the basis chosen, the components will vary while the abstract vector remains the same.
On the other hand, we can also try to calculate the component using the inner product:
X
g(|v⟩, |ek ⟩) = ci g(|ei ⟩, |ek ⟩) (2.11)
i
20 CHAPTER 2. LINEAR ALGEBRA

which is a linear algebraic equation. On the RHS, the matrix g(|ei ⟩, |ek ⟩) is invertible. Why?
An orthonormal basis is a set of vectors which form a basis and which are pairwise orthogonal and
each of them have unit norm, wrt to the inner product. In this case, the above formula simplifies.

2.6.2 Norm
Norm is a non-negative real function defined on the vector space |||| : V → R satisfying these properties

• |||v⟩|| = 0 iff |v⟩ = |0⟩

• ||α|v⟩|| = |α||||v⟩||

• |||v⟩ + |w⟩|| ≤ |||v⟩|| + |||w⟩|| Triangle inequality.

Note that the inner product defines a norm because of the penultimate property
1
||v|| = (⟨v|v⟩) 2 (2.12)

but can we use a norm to define the inner product?

Norms and inner products are matters of choice. They are arbitrary.

Exercise 32
Since they are arbitrary, let us construct a few interesting examples.

Consider the set of differentiable functions on the unit interval. Show that
Z 1
2
||f || = dx|f (x)|2 (2.13)
−1

defines a norm.

Exercise 33
Is the set of functions of finite norm a subspace ?

The idea of the norm can be generalized by adding a weight function w(x) > 0 (also called ‘measure’)

Exercise 34
The Lp -spaces:
Z 1
2
Lp ([−1, 1]) = {f : ||f || ≡ dx|f (x)|p w(x) < ∞} (2.14)
−1
2.6. INNER PRODUCT AND NORM 21

Exercise 35
Let us consider the space of Taylor series
P∞ again on the interval I = [−1, 1] where only finitely
many coefficients are nonzero: f (x) = n=0 an xn with only finitely many an being nonzero.
We can define an inner product by postulating
X
⟨f |g⟩ = fn∗ gn (2.15)

Note that this is not the same as an inner product defined by the integral formula
Z 1
⟨f |g⟩ = f (x)∗ g(x) (2.16)
−1

Is this well-defined on this space?

An Orthonormal basis is a special kind of basis in which the basis vectors are orthogonal to each
other and have unit norm.

2.6.3 Basis for infinite spaces

For infinite vector spaces, the idea of a basis is more tricky. A definition is that a set |ek ⟩, k = 1, 2, ...
of linearly independent vectors is a basis for an infinite vector space if the only vector orthogonal to
all basis vectors is the null vector.

Exercise 36
We can check that this idea works for the finite dimensional case (and gives us the usual thing).

This does not tell us how to expand a vector or what is the meaning of such an expansion, in the
infinite case. This is because of the following:

Exercise 37
A sequence of vectors can converge in the norm - but the limit may not be in the vector space.

If we have a basis as above, we can ask whether every vector can be written as a linear combination
of the basis vectors, i.e.,
XN
|v⟩ = lim ci |ei ⟩ (2.17)
N →∞
i=1

Thus, we get a sequence |vN ⟩ for each N - and the key question seems to be whether the limit exists
and if it does, is the limit also in the same vector space that we started.
So, we have a somewhat special class of vector spaces called Hilbert space in which the limit exists
and gives a vector in the same vector space. The equality in the above equation, in the case of Hilbert
spaces, is to be understood in the following manner: the norm of the difference tends to zero in the
limit.
N
X
lim |||v⟩ − ci |ei ⟩|| → 0 (2.18)
N →∞
i=1
22 CHAPTER 2. LINEAR ALGEBRA

The tricky point that happens is that we can easily construct examples where |v⟩ is actually not
identically the same as the infinite sum, but nevertheless, the above is true. The example I have in
mind is from Fourier series:
X 2n(1 + cos(nπ))
cos(πx) = sin(nπx) (2.19)
n
(n2 − 1)π

where the Hilbert space in question is the set of square integrable function on the unit interval [0, 1]
which vanish at both ends. All the terms on the RHS vanish at the endpoints, but the LHS does not.
It is instructive to plot the LHS and RHS for different values of N . This example is relevant to the
quantum particle in the infinite potential well.

Exercise 38
The inner product satisfies a very important inequality called the Cauchy Schwarz Inequality

||a|| ||b|| > ||a.b||

where both the norm and the inner product are computed using the metric g

Exercise 39
Triangle Inequality
||a|| + ||b|| ≥ ||a + b||

The triangle inequality uses only the norm.

2.6.4 Gram-Schmidt orthonormalization

Any basis can be converted into an Orthonormal basis. This can be done within N-steps where N is
the dimension of the vector space.

2.7 Experiments
In an experiment, a system is “prepared” in a definite state. We then apply the measuring apparatus
on the state. The outcome is a number (or a boolean value).
We have studied several finite dimensional vector spaces. Can these be the states of some quantum
system? The answer, in short, seems to be yes.
These are approximate representations of quantum states of some systems provided we ignore other
aspects.
For e.g., every quantum particle has a discrete label called spin (intrinsic angular momentum) which
takes finitely many values. If we can ignore the ‘motion’ of the particle, we can then say that the states
of the particle are vectors in a finite dimensional space. This is not unrealistic, we can ‘freeze’ the
position of the particle by applying electric and magnetic fields on it (provided it is charged) - but this
does not entirely eliminate the position labels.
In the double slit, if we may imagine, by analogy with classical particles that the particles coming
out of the upper slit are labelled by |1⟩ and the particles from the lower slit are | − 1⟩ - and since
these are clearly distinguishable and there are no other possibilities, we may imagine that we have a
2d vector space of these two possibilities.
2.8. LINEAR OPERATORS 23

Exercise 40
We can develop our ideas: what can we say about the state α|1⟩ + β| − 1⟩?

What about infinite vector spaces?

In the double slit experiment, the particles are observed to arrive at the ‘screen’ which is some kind
of a detector. Further, in each such detection event, a single particle is observed - not half a particle
or more than one particle.
This suggests that we can assign a state vector labelled by the position of the particle |x⟩ (after
it has been detected). Since this position value is a real number, we have a range x ∈ [−L, L]. We
usually assume that these kets are linearly independent (and even orthogonal).
In this manner, we can identify the vector space
P in which, our initial particle, is some vector (i.e.,
a linear combination of the basis vectors |v⟩ = x cx |x⟩).
By pursuing this idea, we can also argue that the particle must also have, minimally, a y− and a
z− label since our universe appears to be 3-dimensional.
To check that we have ‘described’ the system correctly, we have to make predictions and then verify
those. This gives us confidence in our ‘modelling’ - but since there are actually infinite things to check,
we can never ‘confirm’ our ideas.
BTW, it must become clear that the states are being labelled by quantities which are related to
experiments. It is possible to use other labels for (basis) states - but then we have to work to identify
the relation between the ket labels and experimental quantities.

2.8 Linear Operators

Operators are maps/ linear functions on a vector space which give vectors as output.
One can also imagine nonlinear functions, but these are not discussed in Linear algebra and QM.
This allows us to answer the question: How can we change a vector (i.e,. a ‘state’)?
Answer: apply an operator. By definition, an operator ‘acts’ on a state and gives another state.
So operators act within a vector space. One would imagine that given any vector, the operator will
spit out another vector in the same vector space.
In finite dimensional vector spaces, an operator will act on any vector and always give another
vector. In Infinite dimensional vector spaces, usually operators have a domain of definition. Because
of this, we have to check that our ‘operator’ is well-defined. This can be a problem in that, the operator
maybe defined only on a subset (‘domain’) of the full vector space. If the subset is sufficiently big
(’dense’), then it may still be ok - but in any case, we physicists will face difficulties.

Exercise 41
Particle in a box problem:

There are ‘maps’ from a vector space to another vector space. These are not the ones we are talking
about - not even if the two vector spaces are isomorphic (no canonical isomorphisms).
How can we write down operators/maps.
For any linear operator, because of the linearity, it is enough to describe how the operators act on
a basis. If we choose any basis, then an operator can be represented in the form of a matrix . It will
act on the components of a vector in that basis.
24 CHAPTER 2. LINEAR ALGEBRA

Exercise 42
Is the operator AB represented by the product of the matrices A and B? Is the orthogonality
of the basis vectors important for this to happen?

If we have a Linear operator A on V, then we can define its action on the dual space V ∗ by using
the equation
(⟨w|A)|v⟩ = ⟨w|(A|v⟩). (2.20)
It is important to understand that the above equation actually defines how A acts on the dual space.
Note that we are using the same symbol for the operator on the dual space - this could have caused
confusion.

Exercise 43
Write down the action of A on the dual space by taking an example.

2.8.1 Eigenvalues and Eigenvectors

.
For every operator, we can search for its eigenvectors which are nontrivial solutions to the equation

A|v⟩ = λ|v⟩ (2.21)

The number λ is called the ‘eigenvalue’ of the operator A corresponding to the eigenvector |v⟩.
If we choose a basis, then the above equation becomes
X
Aij v j = λv i (2.22)
j

the lower index is the column index of course (because v is column vector with rows labelled by j).
The point of finding the eigenvectors is that we can write the operator using the eigenvectors
X
O= λi |vi ⟩⟨vi |. (2.23)
i

A theorem from mathematics tells us that there is an upper limit to the number of solutions of the
above equation (in a finite dimensional vector space). More generally, we have special vectors called
generalized eigenvectors which allows us to bring our matrix to the Jordan canonical form.
If we have a vector space equipped with an inner product (metric) g, then for every operators A
we can define another operator A† called the adjoint of A:

g(|x⟩, A|y⟩) = g(A|y⟩, |x⟩) = g(|y⟩, A† |x⟩), ∀|x⟩, |y⟩ ∈ V. (2.24)

Note that A† also acts on V.

Comment: in physics books, it is common to write an equation of this type (see Griffiths Appendix
A.5)
⟨Ay||x⟩ = ⟨y|A† |x⟩ (2.25)
where ⟨Ax| is the dual vector to the ket |Ax⟩ = A|x⟩ using the inner product. As usual, it is important
to construct a few examples - this is important to appreciate how the adjoint of an operator depends
on the choice of the inner product.
2.8. LINEAR OPERATORS 25

Exercise 44
Firstly, work out the adjoint of the product operator AB in terms of those of A and B.

Exercise 45
Consider R2 and use the inner product
   1

g1 0 v
⟨w||v⟩ ≡ w1 w2
0 g2 v2
 
a b
(with g1,2 > 0, of course). If A = , calculate its adjoint.
c d
Repeat the exercise for C2 and the standard inner product. Compare the two.
Finally, can you write an operator for which A = A† .

It is especially important to understand the idea of the adjoint for function spaces.

Exercise 46
d2
Repeat the exercise for a function space with weight function w and the operator O = dx2
(assume whatever else is needed).
Answer:
1 d2 d2 w′ d w′′
2
w(x) = 2 + 2 +
w dx dx w dx w

Exercise 47
i d
The operator w dx is Hermitean in the above vector space.

As you can appreciate, the domain of definition of an operator and its adjoint need not be the same.
This point can be emphasized most forcefully, if you add boundary conditions on the vector space. For
some additional information, refer to https://mathworld.wolfram.com/HermitianOperator.html.
These things are important when one considers (generalized) symmetries.
Normal operators form a special class of operators and are defined by the equation [A, A† ] =
AA − A† A = 0 (i.e., their commutator vanishes). From any operator M , we can construct a Normal
†

operator n = m† m but these are not the most general. We also have the product of two normal
operators is normal if they commute.
Normal operators are a very important class of operators because they satisfy the Spectral theorem
which states that Every normal operator can be diagonalized by a Unitary transformation.
This means the following - if we start with an orthonormal basis in a complex vector space V,
in which the Normal operator N is represented as a matrix Nji , then we can always find another
orthonormal basis in which N can be written as a diagonal matrix
X
N= λi δji . (2.26)
i,j
26 CHAPTER 2. LINEAR ALGEBRA

Note that this automatically has given you d-linearly independent eigenvectors for the operator N
(where d is the dimension of V). This is called a spectral decomposition .

Exercise 48
Q
Suppose νi are the eigenvalues of a normal operator N . The operator O = i (N − νi ) is
actually identically 0.
This is because the eigenvectors form a basis and the above operator annihilates every basis
vector.
We can also construct another operator called the resolvent
1
R(x) = (2.27)
N −x
which blows up whenever x coincides with an eigenvalue.
−νi )
Also, consider P (x) = i (N
Q
x−νi . This operator acts on the ket |νI ⟩ as 0 except when x = νI
Sakurai Chapter 1 exercises.

Hermitean operators are a special case of normal operators since they coincide with their adjoint
H = H † . All eigenvalues of Hermitean operators are real.

Exercise 49
You can prove this easily by assuming the spectral theorem.

This maximal set of eigenvectors can be used as a basis of the vector space. Therefore, the eigen-
vectors of a Hermitean operator form a linearly independent set (more precisely, we can always choose
them to be linearly independent).
From all this, we can get a feeling for an operator by thinking of them as being ‘made’ of their
eigenvalues and eigenvectors. When an operator acts on a vector, it ‘rescales’ the components of the
vector along the eigen-vectors by the eigenvalue. This is not entirely correct since the more general
story is that of the Jordan canonical form (Jordan blocks). If the field is not complete, then there
are nilpotent parts (tr = 2) and rotation type (tr < 2) blocks apart from the eigenvalues. If the field
is complete, then we still can have nilpotent blocks. This allows one to discuss how the vector space
decomposes into subspaces defined by the various blocks.
An object of the form |v⟩⟨u| is a linear operator. Because, we can apply it on a vector |ψ⟩ as:

|v⟩⟨u||ψ⟩ = |v⟩⟨u||ψ⟩ (2.28)

which is clearly a vector multiplied by a complex number ⟨u||ψ⟩.

Thus, suppose we have an orthonormal basis for a vector space |ei ⟩, we can write every operator
A as
X
A= Aij |ei ⟩⟨ej | (2.29)
ij

Aij is the matrix that represents the operator A in the basis |ei ⟩.
2.8. LINEAR OPERATORS 27

Exercise 50
Since |ei ⟩⟨ej | is an operator for each i, j, what is its Hermitean conjugate (adjoint) ?
Hence, verify that the condition for Hermitean operator is Aij = (Aji )∗

For a Hermitean operator H, we know that the eigenvectors form a basis. Then, we can show that
X
H= λi |λi ⟩⟨λi | (2.30)
i

If N is an operator with real eigenvalues, it does not follow that its eigenvectors will form a basis.
Only for Normal or Hermitean operators.

2.8.2 Functions of operators

For all our purposes, we can construct functions of operators F (A) by expanding the function in terms
of its Taylor series. Thus,
F2 2
F (A) = F0 + F1 A + A + ... (2.31)
2
This series will be sensible if it converges in some sense.
If the operator is Hermitean (normal), then the operator above series will always make sense as
long as the Taylor series has some radius of convergence . You can see this using the spectral theorem.
One particularly important function that is ubiquitous in QM is the exponential of an operator. In
the finite dimensional case, the exponential of an operator always makes sense. In an infinite setting,
we have to worry about it, for the same reasons as discussed before.

2.8.3 Special operators

There are other kinds of operators which are important in QM.
Positive operators: these are a special case of Hermitean operators since a positive operator P
has the form P = M † M
Bounded operators : These are operators with bounded eigenvalues.
Projection operators: These are very special Hermitean operators since the only eigenvalues of
a projection operators are 0 and 1.

2.8.4 Density Matrices

There is an amusing thing about states in a vector space. We can always think of a state as a linear
function on the space (algebra) of operators:
|v⟩ : A → ⟨v|A|v⟩. (2.32)
Note that the above equation is not linear in the vector, but it is linear in the operator. This equation is
better understood in the following manner. If we are given a state vector |v⟩, it is possible to uniquely
associate a projection operator to it:
1
ρv = |v⟩⟨v|
N
where N = ⟨v||v⟩. Then the above equation is
ρV : A → T r(ρv A) (2.33)
entirely in terms of the operators (and we dont have to talk about states). But since not every operator
A can be interpreted as being associated to a state, we need a better characterization of those operators
which are going to be thought of as states.
28 CHAPTER 2. LINEAR ALGEBRA

1 †
Ex. 51 — Written in matrix form, this will read ρv = N vv .

Ex. 52 —
ρ2V = ρV

Ex. 53 — If we rescale |v⟩ → α|v⟩, α ∈ C, ρV does not change.

Ex. 54 — The only eigenvalues of ρV are 0,1.

Ex. 55 — How many zero eigenvalues are present?

Ex. 56 — T r(ρV ) = 1
A Hermitean operator whose trace is unity is called a Density matrix. A density matrix satisfying
T r(ρ2v ) = 1 is called a Pure state. Density matrices (ρ2 = ρ) which are not pure T r(ρ2 ̸= 1) are said
to be Mixed states .

2.9 Measurement and Quantum Mechanics

We now have the necessary ingredients to explain how the vectors relate to experiments.
How can we ‘measure’ a vector ?
Colors and smells are ‘measured’ by receptors - the response of the eye to incoming light - which
translate into neural signals.
Where is the ‘vector’ property in this measurement?

Exercise 57
Bats use sonar to ‘measure’ distances and directions.

Exercise 58
Electric field is a vector. How do we measure it? Magnetic field ?

Thus, we come to the question of how can we perform measurements on quantum systems. That
is to say, take a quantum state as an input and produce
a ‘number’ as an output.
For example, if we represent our state as 1 2 3 , we can imagine that the outcome of the
experiment is the number in the first row. Or the second row. Or the sum of all the rows. Note that,
in each case, the number can be obtained by taking the inner product with some ‘basis’ vector.
However, since quantum mechanics uses complex vector spaces and our experiments measure real
numbers, the above suggestions cannot represent measurements.
Are there other possibilities?
One point that is noteworthy is that the experimental quantity cannot be linearly dependent on the
state. This is because the vector in QM is an element in a complex vector space (and the experimental
measurement is a real number).

Exercise 59
If the state is represented as a density matrix, then also. Since the density matrix is a complex
matrix in general.
2.9. MEASUREMENT AND QUANTUM MECHANICS 29
 
a
One way to get real numbers is to take absolute values. Thus, if we have a complex vector ,
b
we can consider |a|2 or |b|2 . This idea takes care of both issues: complex vector space, and nonlinear
dependence on the state.
However, we still have to understand what does this represent in the actual experiment. To put it
differently, if our experiment results in a digital readout of an electric current, we have to relate the
current to the above quantity which is obtained from the state vector.
Some of you might have noted that the numbers used above to represent the vector as row/column
vectors depend on the choice of the basis. Should the experiment depend on the choice of the basis?
In any experiment, the experimenter makes a choice of what quantity to measure. On a given
system, the experimenter can perform experiments that measure several different quantities. The
input of which quantity is being measured must be provided and ‘applied’ to the state. We can
perhaps, somehow, relate the choice of the basis to the experimenter’s choice of the quantity being
measured.
Suppose the experimenter chooses to measure a physical quantity identified by a Hermitean operator
- lets called it O - for observation.
This operator defines a basis: its eigenvectors
X
O= ωi |oi ⟩⟨oi | (2.34)
i

where the state vectors |oi ⟩ are now being labelled with the eigenvalues. Let our system be in the
state(vector) |ψ⟩ (or the density matrix ρψ ).

Born Interpretation
also called the Copenhagen Interpretation of Quantum mechanics.
Let us perform the experiment - which will ‘measure’ the operator O. According to the postulates
of QM, the end result of the measurement is that, immediately after the measurement, the system will
be found in one of the eigenstates:
ρψ → |oI ⟩⟨oI | (2.35)
commonly, one says that the system has collapsed to the I−th eigenstate. Note that this procedure is
irreversible - after the measurement all information about the |ψ⟩ has disappeared.
Further, we cannot predict which eigenstate. The only thing we can predict is the relative fre-
quency of the various eigenstates — provided the experiment is repeated many times - with identically
prepared initial states |ψ⟩. The relative probability/frequency of getting the value ωK is P given by
T r(ρV |oK ⟩⟨oK |). We can rewrite this formula in the more standard way: first expand |v⟩ = i v i |oi ⟩.
Then, the relative frequency is |v K |2 .
To complete the idea, we can normalize the state or the density matrix and get an actual probability:

|v K |2
p(|oL ⟩) = P i 2 (2.36)
i |v |

If our vector space is infinite, we might want to simply stick to the relative frequency view since the
normalization might not be sensible.
Clearly, the experimental outcome depends on the inner product structure. It is not a linear
function of the state vector.
Unlike Classical Physics, quantum physics can only predict probabilities for experiments.
Examples
Suppose the vector space is a 2D vector space - this is the space of states of a single qubit.
  
α
H= , α, β ∈ C (2.37)
β
30 CHAPTER 2. LINEAR ALGEBRA

The inner product (check) on this vector space will be

   
a c
g( , ) = a∗ c + b∗ d. (2.38)
b d

Exercise 60
 
a
What is the dual vector space? What is the dual vector of ?
b

 
i+2
Consider a state of the form . Since we have written it already using a column vector
3 +4i 
a1 0
notation, there is some operator A = whose eigenvectors are the basis vectors (1, 0) and
0 a2
(0, 1).
The system therefore has two possible values for a measurement of A (and only two). These two
values, a1 and a2 , are the eigenvalues of A.
If repeated measurement of the operator A always yielded the eigenvalue a1 corresponding to the
eigenvector (1, 0), we will say that the system has the property A because the outcome is definite every
time. Similarly for the other eigenvector.
i+2
OTOH, the state is not an eigenvector of A. Then, we say that |i + 2|2 = 5 is the relative
3 + 4i
probability that the system will be found in the state (1, 0). And |3+4i|2 = 25 is the relative probability
that the system is in state (0, 1). We can state it another way, 5 is the relative probability that the
measurement will give the value a1 .
Since there are only two possible outcomes for the experiment measuring A, the total probability
is 25 + 5 = 30 which must equal unity.
5
Thus, the absolute probability for state (1, 0) is 30 and that for state (0, 1) is 25
30 .

Exercise 61
 
3−i
Check that the same probabilities are obtained from the state . Why?
7+i
Can you write another vector which gives the same probabilities?

Consider calculating ⟨v|A|v⟩

⟨v||v⟩ . To do this calculation, we first expand |v⟩ using the eigenvectors of
A. |v⟩ = v |a1 ⟩ + v |a2 ⟩. Then, we note that A|v⟩ = a1 v 1 |a1 ⟩ + a2 v 2 |a2 ⟩. Finally, we take the inner
1 2

product again with the ket |v⟩. This gives

|v 1 |2 |v 2 |2
   
⟨v|A|v⟩
= a1 + a2 (2.39)
⟨v||v⟩ |v | + |v 2 |2
1 2 |v | + |v 2 |2
1 2

Note that here we use the orthonormality of the eigenvectors |a1,2 ⟩. Clearly, the above expression is
an expectation value.
Note that the expectation value is not an outcome. The result of any measurement is always an
eigenvalue. The expectation value is a property of the ensemble of systems and not of any particular
copy in the ensemble.  
1 i
Suppose we had an operator B = . Check that it is Hermitean wrt the usual inner product.
−i 1
2.9. MEASUREMENT AND QUANTUM MECHANICS 31

Let us think about the measurement of B. Since the possible outcomes of the measurement arethe 
1
eigenvalues, we must first calculate them. These are 0, 2. The eigenvector corresponding to 0 is
i
 
i
and that corresponding to 2 is .
1
 
i+2
The state given is clearly not an eigenvector of B. Nevertheless, measurements of B on
3 + 4i
identically prepared copies of the system can only give values 0 and 2.

Exercise 62
What is the relative frequency with which these values will occur?
Answer: First note that
     
i+2 1 i
= (3 − i) + (2 + i)
3 + 4i i 1

 
i+2
Ex. 63 — What is the dual vector corresponding to the state |ψ⟩ =
3 + 4i

Ex. 64 — Suppose the time evolution of the quantum state given before is by the operator
 
cos ωt sin ωt
U (t) = .
−sinωt cos ωt

That is to say, the state |v(t)⟩ at time t is |v(t)⟩ = U (t)|v(0)⟩.

What happens to the measurement of the operator B at different times.
 
i+2
Ex. 65 — What is the density matrix corresponding to the state
3 + 4i
 
3−i
Ex. 66 — What is the density matrix corresponding to the state .
7+i

Ex. 67 — Check that the probabilities turn out correctly using the density matrix.

Ex. 68 — Are these density matrices ‘pure’ ?

Ex. 69 — Write a mixed density matrix.

Ex. 70 — Compute the von Neumann entropy of the state given by the above density matrices

S = −Trρ log ρ (2.40)

We have already suggested that every experiment represents a basis choice. That is to say, an
experiment represents a set of projection operators.
But, experiments are not operators since many operators can have the same eigenvectors.

Exercise 71
Write down examples of operators with the same eigenvectors.
32 CHAPTER 2. LINEAR ALGEBRA

Since there are as many “rows” as basis vectors - the outcomes of the experiment can only take
as many values as the number of rows. This is the origin and meaning of the word “quantum” in
“Quantum Mechanics” - experimental outcomes can only take a few ‘quantized’ values. In contrast, in
classical physics, the experimental quantity is usually a real number because its a coordinate of phase
space.
The space of experimental outcomes vs. the dimension of the vector space.

2.9.1 Compatible observables

Because a quantum measurement ‘collapses’ the wavefunction (state) - something new can happen in
quantum mechanics.
If we first measure the operator A, we have collapsed to an eigenstate of A. Let us say, that this
is followed up by repeating the measurement A, then the postulate of QM says that we will get the
same eigenvalue for the measurement and the state will remain the same eigenstate that it collapsed
to at the first instant.
On the other hand, if we now measure a different operator B - then, our state which was an
eigenstate of A will now collapse to an eigenstate of B.
If we now repeat our measurement of the operator A, there is no more certainty about the value
we will get - since the measurement of B wipes out all information of which state of operator A the
system was in.
This is not always necessary. Suppose A and B have the same eigenvectors - then the measurement
of A also gives a definite value for an immediately successive measurement of B.
If we wait too long, the system will time evolve and it may be far away from the eigenstate of A.

Exercise 72
Following up on the previous example, can we conduct an experiment that will simultaneously
measure A and B?

Two operators with the same eigenvectors are thus said to be compatible - since the measurement
on one of them, determines values for the other.

Exercise 73
In a finite dimensional space, [A, B] = 0 is a necessary and sufficient condition for two operators
to be compatible.
We will assume this to be so even in the infinite case.

Therefore, a set of Commuting Observables is a list of operators that mutually commute with each
other. Thus, all of them are said to be simultaneously measurable.
This set is said to be complete if the number of independent eigenvectors is equal to the dimension
of the space. That is to say, every vector in the vector space can be written as a linear combination of
the simultaneous eigenvectors.
One of the goals of modern particle physics (and string theory) can thus be described as trying to
identify this set of complete commuting observables (CSCO).

2.9.2 Transition Amplitudes

Start with a normalized state |ψ⟩. Measure A. Then we measure B.
2.9. MEASUREMENT AND QUANTUM MECHANICS 33

A natural experimental question is to determine the probabilities of getting specified values for A
and B.
|⟨a1 ||ψ⟩|2 is the probability of getting the value a1 . The conditional probability of subsequently
getting b1 is |⟨b1 ||a1 ⟩|2 . Similarly, the subsequent conditional probability of getting b2 is |⟨b2 ||a1 ⟩|2 .
We can imagine calculating the probability of getting b1 when we measure operator B from the
starting state in two ways.
The first is a direct transition from the initial state to the measurement B, the measurement
postulate tells us that the answer is
|⟨b1 ||ψ⟩|2 .
OTOH, we can do this by using Bayes’ rule in the following way.
X X
p(b1 ) = p(b1 |ai )p(ai |ψ) = |⟨b1 ||ai ⟩|2 |⟨ai ||ψ⟩|2 ̸= |⟨b1 ||ψ⟩|2 (2.41)
i i

Clearly, these two answers are not the same! Why did this happen? Because we performed the
measurement A, the experiment B will give answers different from the situation when A was not
performed. We can state this as “it is not possible to interpret the experimental results for B as saying
that B occurred having gone through A.
On the other hand, the direct answer can be written in the following way:
2
|⟨b1 ||ψ⟩|2 = |⟨b1 ||a1 ⟩⟨a1 ||ψ⟩ + ⟨b1 ||a1 ⟩⟨a1 ||ψ⟩| (2.42)
Thus, the formula above can be described in the following words: “ The transition probability from
the initial state |ψ⟩ to the final state |b1 ⟩ can be computed by first computing the transition amplitude
⟨b1 ||ψ⟩ = ⟨b1 ||a1 ⟩⟨a1 ||ψ⟩ + ⟨b1 ||a1 ⟩⟨a1 ||ψ⟩ (2.43)
as a transition passing through an intermediate, complete set of states |ai ⟩ of some Hermitean operator
A. However, in this case, since the measurement of A is not being performed, we add the amplitudes
and not the probabilities.

Exercise 74
P
This happened because i |ai ⟩⟨ai | = I (since A is Hermitean)

2.9.3 Uncertainty principles

What happens if we try to measure two operators which do not commute with each other .
Griffiths: section 3.5 Sakurai problem
As mentioned before, ⟨ψ|A|ψ⟩ is the expectation value of the observable A in the state |ψ⟩ (provided
⟨ψ|ψ⟩ = 1). This expectation value is a meaningful quantity concerning experiments provided the
‘fluctuations about the mean’ are small. These are quantified by the standard deviation
2
σA = ⟨(A − ⟨A⟩)2 ⟩ = ⟨A2 ⟩ − (⟨A⟩)2 (2.44)
Suppose we have a second operator B which is compatible with A (i.e., [A, B] = 0).
Even if the first observable has a well defined mean, this does not mean anything about the second
observable. The only thing we learn from this is about our own unreasonable expectations.

Exercise 75
Construct some examples.
34 CHAPTER 2. LINEAR ALGEBRA

On the other hand, if [A, B] = iC where C is also an observable (i.e., automatically Hermitean),
then
2 2 2
4σA σB ≥ (⟨C⟩) (2.45)
This inequality is called an ‘uncertainty principle’. This name referred to the idea that since A and B
do not have simultaneous eigenstates (in general that is), then the inequality suggests that minimizing
the fluctuations in the A measurement can only occur at the cost of increasing the fluctuations in B.

Exercise 76
Write down an example of A, B and C where [A, B] = iC but A and B have a common
eigenvector |v⟩.
What implication does it have, if any?

The most important thing to realise about the inequality above is that it only applies to a collection
of systems - this is because both expectation value and standard deviation do not make sense for a
single system. So, it is meaningless to think of the inequality as saying something about the system
itself.

Exercise 77
If you toss a dice (with faces numbered 1...6) which is ‘fair’ (all the numbers have equal prob-
ability), then
the expectation value is 3.5!
This is clearly meaningless as an outcome. The measurement will never give us this value.

I seem to have conveyed the idea, then, that the state vector describes a single system. This idea
is also doubtful. I do not wish to compel this idea.

Exercise 78
Problem 3.36 Griffiths:
2 2
σB ≥ ⟨C⟩2 + ⟨D⟩2


4σA (2.46)
where D = AB + BA − 2⟨A⟩⟨B⟩

It might be clear to you that much of quantum mechanics involves finding eigenvectors of operators.
In our course, we will find eigenvectors mostly by guesswork.

2.10 More vector spaces:

We can take two vector spaces and make new ones out of those. The simplest way to describe these
new vector spaces is to exhibit a basis for them. However, it is possible to define them without using
a basis.

Direct Sum.
The Direct Sum of two vector spaces is constructed in the manner of a Cartesian product of sets.

V1 ⊕ V2 = {(v1 , v2 ), v1 ∈ V1 &v2 ∈ V2 } (2.47)

2.10. MORE VECTOR SPACES: 35

The basis for the Direct Sum is obtained by taking a disjoint union of the two bases. Thus, the
dimension of the Direct Sum will be the sum of the dimensions of the individual vector spaces.

Exercise 79
A simple example is R3 = R ⊕ R ⊕ R. When we talk about position vector - we think of the
space of positions as vectors in the above vector space.
But since positions cannot be added, we are making a mistake in our thinking.
On the other hand, velocities can be added. What is the meaning of ‘addition of velocities’ ?
This is why the set of velocities is a vector space in Classical Physics.
Can you construct a more interesting example of a direct sum.

Tensor product
For the Tensor Product of two vector spaces, the basis is constructed by taking a formal product of
the two bases. The dimension of the Tensor product vector space is the product of the dimensions of
the individual factors.
(1) (2)
V1 ⊗ V2 = Span{ ei ⊗ ej } (2.48)
The ⊗ is just a symbol (which will eventually be entirely omitted in writing), and we may imagine
that the two basis vectors have only been written adjacent to each other.
A simple example: Take a 3-dimensional Real vector space and a 2-dimensional complex vector
space with basis vectors ei and fj . We can make a 6-dimensional complex vector space by forming the
basis EI = ei ⊗ fj . Here I = 1..6 and one can map I to the pair ij in any possible way we like. Note
that this is possible because Reals form a subset of the Complex numbers.
As another example, we can take a tensor product of a function space and a finite vector space
that is of much importance. In this vector space, we can have vectors of the form:
   ′  
α α ψ̃1
ψ1 (x) ⊗ + ψ2 (x) ⊗ = ∈ Hx ⊗ H. (2.49)
β β′ ψ̃2

The first equality defines the ψ̃.

What is the form of the most general vector in this vector space?
It is sometimes useful to think of tensor products in terms of an array notation.

2.10.1 Multiparticle systems

Suppose we had two qubits. The space of states is the tensor product.
           
1 1 1 0 0 1 0 0
V = {α +β +γ +δ α, β, γ, δ ∈ C} (2.50)
0 0 0 1 1 0 1 1
The order matters: β, γ cannot be interchanged.

Exercise 80
If you consider the way the tensor product has been defined, what are the coefficients α..δ
defining the tensor product state
   
2+i 3−i
⊗ (2.51)
3 + 4i 7+i
36 CHAPTER 2. LINEAR ALGEBRA

On the other hand, we can consider the direct sum as well:

 
α    
β  α γ
V = {  =
  ⊕ α, β, γ, δ ∈ C} (2.52)
γ β δ
d
Define another system called a qutrit.
     
1 0 0
V = {α 0 + β 1 + γ 0 α, β, γ, δ ∈ C} (2.53)
0 0 1

Exercise 81
Write down the Tensor product and Direct sum of a qubit and a qutrit.

The following operators are important in these vector spaces.

The Pauli matrices
     
0 1 0 i 1 0
σ1 = σ2 = σ3 = (2.54)
1 0 −i 0 0 −1
These have unique properties
Ex. 82 —
σi2 = 1

Ex. 83 — They are the generators of the Angular momentum algebra (generators of su(2) or so(3))

σi σj − σj σi = 2iϵijk σk

Ex. 84 — They form a Clifford algebra

σi σj + σj σi = 2δij

Ex. 85 — Every Hermitean matrix in 2D can be written as a linear combination of the Pauli matrices.

Ex. 86 — Every 2 × 2 Unitary matrix can be written by using the Pauli matrices.
X
U = exp[i ai σ i ] (2.55)
i

Ex. 87 — What are their eigenvalues and eigenvectors?

In the space of qutrits, we have the following important operators
     
0 1 0 0 −i 0 1 0 0
L1 = 1 0 1 L2 =  i 0 −i L3 = 0 0 0 (2.56)
0 1 0 0 i 0 0 0 −1
upto some normalization factors.
Ex. 88 — They are also the generators of the angular momentum algebra (generators of su(2) or
so(3))
Li Lj − Lj Li = iϵijk Lk
2.10. MORE VECTOR SPACES: 37

Ex. 89 — L3i = ai Li

Ex. 90 — What are their eigenvalues and eigenvectors?

Since these are all Hermitean operators, they represent possible experiments that can be performed
on the space of qubits and qutrits.
For example, measuring σ3 = σz is usually called ‘measuring the spin’ of the qubit system.

2.10.2 Entanglement
In tensor product spaces, we can have special kinds of states which are called ‘entangled’.
If V = V1 ⊗ V2 , V1,2 are called subsystems of the system V . A state v which cannot be written as
a product of a state from V1 and state from V2 is called entangled

Exercise 91
         
α γ 1 1 0 0
⊗ is not entangled. But, + is entangled.
β δ 0 0 1 1

Entanglement makes sense only between subsystems. So, it does not make sense to talk about
entanglement without specifying the tensor product structure (ie define the subsystems).
If we have a vector space which is 24-dimensional, it can arise from many different tensor products.

Exercise 92
It will be interesting to write a few examples of vectors which are entangled in one way but not
entangled in another way.

Some Operators on tensor products naturally arise from the individual vector spaces. We can
construct operators like O1 ⊗ O2 where O1,2 acts on V1,2 respectively.
But these are not all the operators on the tensor product of course.

Exercise 93
Can you write one Hermitean operator which is not itself a tensor product? This will represent
an experiment which cannot be performed on one subsystem alone.
Ans: σx ⊗ σy + σy ⊗ σz + σz ⊗ σx which has eigenvalues −3, 1, 1, 1 and eigenvectors (−i, i, 1, 1)

When we have multiple particles, the quantum states of these are to be found in the tensor product
of the individual vector spaces. Subject to one restriction.
Quantum particles come in two types: Bosons (integer spin s) and Fermions (Half integer spin s).
The spin states of a particle of spin s form a complex vector space of dimension 2s + 1.
Suppose we have two fermionic particles that are identical - say both are electrons or both are
protons etc. Then, the only permitted states for such a pair are those which are antisymmetric under
the interchange of the two particles.
If there are more than two identical fermions, then the only states for such a collection must be
antisymmetric under the interchange of any two of them.
For example, if we have two identical spin 1/2 fermions each of whose states form a two dimensional
complex vector space (this means that we are ignoring the position of the particles) the following states
38 CHAPTER 2. LINEAR ALGEBRA
  
1 1
are not allowed etc. In fact, the only possibility is the so-called spin singlet state
0 0
     
1 0 0 1
− (2.57)
0 1 1 0

which is an entangled state of two spins.

Exercise 94
Is the state      
a α α a
− (2.58)
b β β b
which is clearly antisymmetric - the same as the one above, upto a scalar factor?

If we have a state with a proton and an electron - since they are different particles (not identical)
- there is no need to antisymmetrize the states. Thus, for an electron and proton state, the following
are all perfectly legitimate orthogonal states
   
1 1
, (2.59)
0 e 0 p
   
0 0
, (2.60)
1 e 1 p
       
1 0 0 1
+ , (2.61)
0 e 1 p 1 e 0 p
       
1 0 0 1
− (2.62)
0 e 1 p 1 e 0 p

Exercise 95
Convince yourself that these states are orthogonal.

The interesting thing about such states is that we can perform a measurement on the proton only.
Or only the electron.
Suppose we consider the last
 mentioned
 state - and measure the z-spin of the proton: this means
1 0
that we use the operator Ie ⊗ .
0 −1 p

Exercise 96
What are the eigenstates of this operator? What are the possible experimental outcomes.

If the e-p system is in one of the last mentioned states above, measuring the spin of the proton
immediately determines the spin state of the other particle (electron) which can be anywhere in the
universe! The natural question that rises is whether this is compatible with the special theory of
relativity. This also brings up a discussion of the EPR paradox.
Note that, we cannot have a state of three spin-1/2 particles which is completely antisymmetric.
2.10. MORE VECTOR SPACES: 39

Exercise 97
But suppose we also include position degrees of freedom — this means that the state space of
a single particle has an additional label
   
α
V = |x⟩ ⊗ , x ∈ R, α, β ∈ C (2.63)
β

In this case we can make a state for 3-fermions:

     
1 1 1
(|x1 ⟩ ⊗ |x2 ⟩ ⊗ |x3 ⟩ − |x1 ⟩ ⊗ |x3 ⟩ ⊗ |x2 ⟩ − |x2 ⟩ ⊗ |x1 ⟩ ⊗ |x3 ⟩ + |x2 ⟩ ⊗ |x3 ⟩ ⊗ |x1 ⟩..) ⊗
0 1 0 2 0 3
(2.64)
where the 1,2,3 subscripts indicate which particle is in that state.
Notice how its the total state that is antisymmetric. We can even make a ‘triplet state’ for a
pair of spin1/2 fermions
   
1 1
(|x1 ⟩ ⊗ |x2 ⟩ − (|x2 ⟩ ⊗ |x1 ⟩) ⊗ (2.65)
0 1 0 2

In these examples, the space part produced the antisymmetry upon exchange.

Note that        
1 0 0 1
|x1 ⟩ ⊗ |x2 ⟩ − |x2 ⟩ ⊗ |x1 ⟩ ⊗ (2.66)
0 1 1 2 1 1 0 2
is antisymmetric - but the space or the spin part is not separately symmetric or antisymmetric.
40 CHAPTER 2. LINEAR ALGEBRA
Chapter 3

Quantum Phenomena

In this chapter, we will discuss some of the strange features of Quantum Mechanics. These features
arise because QM states are vectors. Sometimes they arise because of the Measurement Postulate.

3.1 Double Slit

41
42 CHAPTER 3. QUANTUM PHENOMENA

3.1.1 Introduction
The double slit experiment is an important experiment in the History of Physics.
The optics version of this experiment, first demonstrated by Young in 1809 proved to be pivotal
in establishing that light is a wave. Huyghens principle and subsequently Maxwell’s derivation settled
the matter beyond doubt.
Yet, the corpuscular theory did not quite ever die - because it provided neat explanations of
rectilinear propagation (free particle) and reflection and refraction laws in terms of the principle of
least time (action).
In quantum mechanics this experiment played a much more important role. For long since Benjamin
Franklin, Michael Faraday Dmitri Mendeleev and Amadeo Avogadaro it was known that electric charge
was quantized. Together with the chemical stoichiometric relations, the atomic hypothesis was firmly
established and many things came to be understood rather well using the idea that the elementary
constituents of matter are ‘particles’.
Therefore it came as a great surprise that atoms and electrons exhibit both diffraction and inter-
ference when they are used as the source in a double slit experiment. Feynman (chapter I vol III) says
that this experiment is “absolutely impossible to explain in any classical way, and which has in it the
heart of quantum mechanics”.

3.1.2 The experiment

Suppose, the incoming atoms (from the silver beam of Stern and Gerlach) or the electrons of Davisson
and Germer are, following our logic, vectors in some vector space. Each electron state |ψ⟩ evolves in
time and is eventually “detected at the screen” at the position z along the screen.
Experimentally, it was found that the number count of electrons (atoms) was similar to the intensity
distribution of light in a double slit experiment. From our description of quantum mechanics, this
number count can be written as
N (x) = p(x|ψ)N (3.1)
where p(x|ψ) is the conditional probability that the electron which started out in state ψ is found
at the position x and N is the total number of electrons. In writing this formula, we have used a
frequentist interpretation of probability.
It must be mentioned that we seem to be making a rather serious assumption that every electron
or atom is produced in the same initial state ψ. If everything is quantum mechanical, then should we
not worry about whether this is even possible? The short answer is that we must worry.
The long answer is that after we learn about Hanbury-Brown Twiss effect and the Hong-Ou-Mandel
experiment, we should worry more.
Since we know that our experimental results for the electron agrees with that of light, we can write
2
2 cos[k.(⃗r − dx̂ − bẑ)] cos[k.(⃗r − dx̂ + bẑ)]
p(x|ψ) ∼ A p + p (3.2)
|(⃗r − dx̂ − bẑ)| |(⃗r − dx̂ + bẑ)|
∼ (3.3)

by using the intensity formula for light at the position x calculated from Maxwell equations. The above
expression is a demonstration of Huyghens’ principle. The second line is the geometrical approximation
calculation.
We see that the intensity as a function of z oscillates with z.

Exercise 98
What are the conditions under which the above formula is valid for light?
3.1. DOUBLE SLIT 43

Here, we have assumed that the electron or atom also arrives at the detector/screen only from the
slits and not by passing through the thickness of the screen at an arbitrary location along the screen.
Or at least, the contribution of these are negligibly small. This seems reasonable since if there were
no slits, the electrons/atoms do not arrive at the screen?
Since we are interested in relative probabilities, we do not need to worry about the amplitudes A.
This must be Born’s interpretation of the quantum wavefunction.
We may then try to understand the origin of the probability by noting that, in analogy with our
previous description in terms of state vectors, the complex number A(cos[kx .d + kz (z − b)] + cos[kx .d +
kz (z + b)]) must be the component of the initial state of the atom or electron (ie, the state at the
position of the screen just before the measurement occurs) in some basis determined by the detector.
That is why the probability is being obtained as the square of this component.

Exercise 99
Can we understand this basis?

To complete this tale, we must check that when the electron/atom is “detected” at the screen/de-
tected, it is actually the “same particle” that started out at the source. For instance, if we use electrons,
the particle detected should always have an electric charge equal to e. Or if it is an atom, then the
weight of the ‘silver’ deposited at the spot on the screen should always be an integral multiple of the
atomic mass of silver.
How can we check that this is indeed the case?
Thus, there is nothing surprising about the analysis of the data: only that we are applying it to
electrons and atoms which are definitely particles.

3.1.3 The issues

If the electrons were regarded as particles, then each electron will either appear at the first slit or the
second one. Subsequently, we can study their propagation from the slit to the screen either as waves
or as particles. Both calculations do not agree with experimental data.

Exercise 100
Particles: Each electron is emitted from the slit with a randomly directed velocity. Suppose
that all angles are equiprobable - then since dz = sec2 θdθ where θ is the angle from the slit,
the relative number of electrons at a position z is

dN z 2 + w2
∼ 2(1 + )
dz d2
where we have counted electrons from both slits.
44 CHAPTER 3. QUANTUM PHENOMENA

Exercise 101
From the wave point of view, each slit produces a diffraction pattern, and since the electrons
are independent, the intensities must be added. This leads to
dI
=
dZ
in the Fraunhofer limit.

These analyses suggest that the experimental data do imply that electrons must be thought of as
waves from the get-go. Right from their emission at the source all the way up to their absorption at
the screen/detector, they obey some kind of a wave equation similar to that of light.
The speciality of Maxwell Equations is their linearity. Thus their solutions do form a vector space.
In this sense too, the analogy with light agrees with the fundamental postulates that the states of a
quantum system are vectors.
Feynman Lectures describes a modification of the double slit where an attempt is made to detect
the path of the electron. Feynman argues that this washes out the interference pattern entirely.
The double slit is a famous experiment that is said to demonstrate the strangeness of the quantum
world. In particular, it is claimed to establish the idea that in the quantum world, there is a ‘wave-
particle’ duality. In our course, we will ignore such phrases (”wave-particle duality”) because these are
words being attached to vectors in a vector space.
The double slit experiment is counter-intuitive because the ‘particle’ arrives ‘whole’ at the screen
- i.e, the detector measures one unit of electric charge say. If, as in the wave picture, the primary
wave was giving rise to secondary waves, this suggests that the original charge is distributed among
the secondary wavelets, and it is perplexing how we detect a whole unit of charge at the detector
everytime.
This seems to prevent the possibility of interpreting the outcome of the experiment in terms of a
wave picture alone and motivates the ‘wave-particle’ duality picture of Quantum Mechanics.
Similarly, in the double slit, its natural to imagine that the particle arriving at the screen must
have passed through one slit or the other.
The current view seems to be that since we are detecting the positions of the final particles - this
precludes the idea that these particles have a definite trajectory.
To say it differently, a particle cannot have a definite trajectory as well as a definite position. If we
modify the setup so that we force the particle to have a definite trajectory, we then lose the interference
pattern [3].
In this experiment, one uses particles to produce an interference-like pattern in the number of
particles detected at various locations along a ‘screen’.

Exercise 102
Hitachi Experiment and Shot noise.

This experiment has been conducted with large molecules [6] and even with proteins [2]. Efforts
are on to conduct this experiment with bacteria [5].
The Physics World article [8] has a very nice set of references. See also https://arxiv.org/pdf/
1606.09442.pdf for a quantitative discussion along the lines of my presentation.
3.2. DOUBLE SLIT AS A TWO LEVEL SYSTEM 45

Exercise 103
Hanbury-Brown and Twiss interfence and photon bunching

3.2 Double slit as a two level system

System: 1 Qbit. Initial state |ψ⟩.
Experiment S has eigenstates |U ⟩, |L⟩ – the states are quantum particles coming out of the upper
and lower slit in a double slit experiment.
Initial state |ψ⟩ = cU |U ⟩ + cL |L⟩ — we cant say which slit it will ”go through” (went through).
At the screen, we detect the “particle” at various positions. We can think of this as having many
detectors with a label called position ‘x’. If the detector x gives output 1 - we have a particle at
that position. We need to model our detector by a Hermitean operator which depends on the label
x ∈ [−1, 1] = screen. So, let us assume that our screen operator D = c1 |d1 (x)⟩⟨d1 (x)|+c2 |d2 (x)⟩⟨d2 (x)|
- the eigenstates |di (x)⟩ are not necessarily the same as the eigenstates of S. Note that the kets depend
on the x-label, but the eigenvalues c1,2 are assumed to be 1, 0 respectively.
The state c1 means that we hear a click - ie a particle has appeared, and c2 means no click. The
experiment D counts the number of clicks.
If we perform S measurement and obtained |U ⟩, its as if we have only those particles which went
through the upper slit. Similarly for the other possibility |L⟩. Thus, the probability for a click after
the particle went through either slit U or slit L is |cu |2 |⟨d1 ||U ⟩|2 + |cL |2 |⟨d1 ||L⟩|2 . We must add the
probabilities because these are mutually exclusive events because of the measurement (Bayesian logic).
On the other hand, if we did not perform the experiment at all -
2
p(click) = |⟨d1 ||ψ⟩|2 = cU ⟨d1 ||U ⟩ + cL ⟨d1 ||L⟩ (3.4)
u 2 1 2 L 2 1 2 U L 1 1 ∗
= |c | |⟨d ||U ⟩| + |c | |⟨d ||L⟩| + 2Re(c c ⟨d ||U ⟩⟨d ||L⟩ ) (3.5)

The difference in the two answers comes from the last “interference” term.

Exercise 104
This must be equated to the optics answer – which allows us to determine the ‘amplitudes’

cU ⟨d1 ||U ⟩ = cos(k(x + L) + ky) cL ⟨d1 ||L⟩ = cos(k(x − L) + ky) (3.6)

Is this correct?

Clearly, the problem is arising because of the collapse postulate - after the measurement is per-
formed, the system “collapsed” to the eigenstate. This leads to “loss of information” - what is this
information that is lost?
Answer: the relative phase between cU,L which is precisely the interference information.
Feynman discusses this in some more detail.
A somewhat confusing point arises - did the screen with the two slits perform a measurement
(on the incoming state) or should we think that there is a transition amplitude via the upper slit
and another transition amplitude via the lower slit which should be added - or is there no difference
between the two.
Supposing we had three slits - or ten slit or many more — supposing there were infinitely many
slits - then we really have no slits at all - and the electrons/photons are directly coming to the screen.
46 CHAPTER 3. QUANTUM PHENOMENA

In this case, the addition of amplitudes clearly makes good sense. Assuming that there is nothing
special about the limit - we can conclude that even in the case of the double slit, the slits did not
perform any measurement.
On the other hand, the opaque part of the slit-screen did (nearly) stop the particles in much the
same way as the detector screen did. To model this situation, we can think of the initial state as

|ψ⟩ = (cU |U ⟩ + cL |L⟩ + m|M ⟩)

The |M ⟩ represents the possibility that the initial particle can go through the ‘middle’ of the screen S
which contains the double slits.
U L
 after the slit - our state has become |ψ⟩ = (c |U ⟩ + c |L⟩) - the screen operator S =
 Then just
1 0 0
0 1 0 is actually a projector - which when applied on the initial state produces the two-state
0 0 0
system after the slits.
We can now anticipate the next question - how come the two slits together are being represented
by the above projector and why are we not using separate projectors for each slit.
One good answer is that this explains the data.
How can we test this argument?
Two pairs of slits
In the class, we discussed the situation with two pairs of slits. The main thing to note was that
the ‘amplitude’ for detection was obtained by composition

A(x) = ⟨d1 (x)||u1 ⟩⟨u1 ||u2 ⟩ + ... (3.7)

which we ‘read’ as the amplitude for transition from the state |u2 ⟩ to the state |u1 ⟩ followed by the
transition to the detector state |d1 ⟩ which corresponded to the detection of the particle. The system
made a transition from its initial state via all the possible states |u2 ⟩, |l2 ⟩ at the screen S2 . In particular,
we noted that this happens because the sum over all intermediate states

|u1 ⟩⟨u1 | + |l1 ⟩⟨l1 | = I (3.8)

gave the identity operator.

We also discussed what happens when the number of slits in the intermediate screen S2 were
increased. In the limit n → ∞ - its as if we do not have a screen.
So we can say, the amplitude for going from the initial state to the detector |d1 ⟩ proceeds via all
intermediate states |y⟩ as
X
A(x) = ⟨d1 (x)| |y⟩⟨y| |u2 ⟩ + ... (3.9)
y

Thus, we have defined a vector space whose basis |y⟩ is labelled by position. Clearly, an infinite vector
space. These must be linearly independent - and orthogonal (Check by imagining finite number of
locations (slits) first). Clearly, we also expect the infinite sum
Z
|y⟩⟨y| = I (3.10)
y

to also evaluate to identity.

So far, nothing has explained why the states |u1 ⟩ occur ‘after’ the states |u2 ⟩. Clearly, we need to
attach one more label called ‘time’.
More fringe games
Supposing, we are using light - and in front of each slit, we place a polarizer. Say, x-polarizer in
front of the upper slit and y-polarizer in front of the lower slit.
3.2. DOUBLE SLIT AS A TWO LEVEL SYSTEM 47

Exercise 105
Fresnel and Arago showed that no fringes are obtained.

Now, suppose we are using electrons - and after the upper slit we, place a polarizer for the electron.
Such a thing can be made - because electrons are spin 1/2 particles, we can have a polarizer which
allows only +polarization or -polarization with the spin polarization being along some chosen direction
n̂. This means that the electron state after the slits takes the form of a tensor product

|U ⟩ ⊗ |+⟩ or |U ⟩ ⊗ |−⟩ (3.11)

and similarly for the other slit. It is understood that we are specifying an orthonormal basis. So, we
have a 4d-vector space. Hence, the state before the slits can be written (not normalized)

|ψ⟩ = α|U ⟩ ⊗ |+⟩ + β|U ⟩ ⊗ |−⟩γ|L⟩ ⊗ |+⟩ + δ|L⟩ ⊗ |−⟩ (3.12)

Suppose we use the + polarizer for both the upper and lower slits - then, after the slits the state will
be
α|U ⟩ ⊗ |+⟩ + γ|L⟩ ⊗ |+⟩ (3.13)
(again not normalized). The new thing is that our state has changed because of the double slits - in
particular, its norm has decreased (so loss of probability) - but this decrease in norm is meaningless.
Compare this with the earlier discussion where we introduced the |M ⟩ state which represented ‘going
through the slit-screen’.

Exercise 106
Write the screen with two +polarizers as an operator acting on the initial state.

Whether we get anything from the detector or not, depends on the detector and what it can
measure. If the detector can only detect −polarization, then the above state will not be ‘detected’
at all. This is mathematically captured by writing the detector click eigenstate |d1 ⟩ with |−⟩ in the
second factor in the tensor product.
On the other hand, if we have a detector that produces a signal regardless of which polarization
state the particle is in - the eigenstates of the detector?
Ans:
|d1 ⟩ ⊗ |+⟩, |d1 ⟩ ⊗ |−⟩, |d2 ⟩ ⊗ |+⟩, |d2 ⟩ ⊗ |−⟩,
Recall that |d1 ⟩ produces a click.
https://arxiv.org/pdf/1712.02670.pdf
Many interesting modern topics covered here by Frank Rioux
One other thing which is not discussed: a detector is supposed to produce an output ONLY when
a particle is interacting with it - the signal can depend on the state the particle is in, but if there is no
particle, the detector is not supposed to produce any outputs. How to represent this second situation
- no output for no particle. Is ‘no particle’ a state vector? What is ‘no output’ ?
R. D. Sorkin, Quantum mechanics as quantum measure theory. Mod. Phys. Lett. A 9, 3119
(1994).
Sinha, Urbasi. “The Triple Slit Experiment.” Scientific American. Jan. 2020. Print. 58-60.
Zyga, Lisa. “Physicists detect exotic looped trajectories of light in three-slit experiment.” Phys.org
Paradox about the Stern-Gerlach experiment
https://www.hrpub.org/download/20040201/UJPA-18490184.pdf
48 CHAPTER 3. QUANTUM PHENOMENA

Exercise 107
Why does the Magnetic field on the electron move it depending on its spin in the Stern-Gerlach
experiment. Check this from the Dirac equation

3.3 Schrödinger’s cat and Wigner’s friend

The Schrodinger’s cat is a thought experiment which explores the extent to which quantum states
can be assigned experimental properties. It also focuses attention on the peculiar role played by the
observer of a quantum experiment.
In this experiment, one imagines a cat which is placed in a box. The box contains an apparatus
which will randomly release a poisonous gas based on the decay of a radioactive nucleus.
The problem is to discuss the state of the system. Classically, the state of the system is clear -
either the cat is dead or the cat is alive. This is quite simple because the random event is a classical
random variable. The two states of the cat are mutually exclusive, classically.
However, the nucleus itself, being a quantum system, will evolve into a quantum superposition of
a decayed state and a un-decayed state. That is to say, under time evolution, the nucleus has a finite
amplitude to be found in a (number of) decayed state(s).
This would force the ‘entirely classical’ cat into a superposition of being dead and alive - i.e., the
cat is forced into a quantum state. The problem appears to be that it is inconceivable that the cat
can be in both ‘dead’ and ‘alive’ states at the same time.
It should be clear that the issue arises because of the classical nature of the cat. It is difficult (if
not inconceivable) to imagine that the cat did not have a definite state (either ‘alive’ or ‘dead’) just
before the measurement.
Ex. 108 — Analyse the same ‘cat set up’ by replacing the radioactive atom with a coin toss.

Ex. 109 — Analyse the same by replacing the ’cat’ with a second atom which maybe ionized by
the radioactive decay (the latter will evolve into a quantum superposition of ionized and non-ionized
states).

Ex. 110 — Analyse the same by replacing both the cat with a second atom and the random source
by a coin tosser.

Ex. 111 — Instead of a source of quantum particles falling on a double slit, if we replace the slit by
a pair of sources which emit particles (in random directions) at random times (determined by some
classical coin tosses), will there be an interference pattern on the screen?
It seems clear that, for quantum spookiness to occur, it is necessary that the cat be assigned a
wavefunction (at least by analogy with the double slit).

3.3.1 Wigner’s Friend

To highlight the issue that arises because of the classical interpretation (which is made possible only
by measurement, Wigner introduced a modification.

3.4 Bunching and Anti bunching

Hanbury-Brown and Twiss effect: Photon Bunching
Electron anti-bunching
3.5. ENTANGLEMENT 49

3.5 Entanglement
3.6 EPR paradox and Bertlmann’s socks
Bertlmann’s socks

3.6.1 Bell’s Inequalities

https://en.wikipedia.org/wiki/Bell’s_theorem
The basic idea of Bell’s inequalities is that Quantum properties (labels that define the state vector)
cannot be physical in the sense that they exist independent of measurement. Because if they did, they
must satisfy Bell’s inequality - but quantum experiments violate Bell’s inequalities copiously.
https://en.wikipedia.org/wiki/CHSH_inequality
Undergraduate lab experiment to test CHSH A photon has free will!
50 CHAPTER 3. QUANTUM PHENOMENA

3.6.2 Mermin’s experiment

We have three spin 1/2 particles which are moving apart in a plane, maybe because they are produced
i
in a decay. For each of them, we have the corresponding operators σx,y,z . The direction x is understood
to be upwards to the plane of motion. While z for each of them is in the direction of motion of that
particle. y is defined so that we get a right handed coordinate system.
Let us assume that the initial state of the three particles is
1
|GHZ⟩ = √ (|1, 1, 1⟩ − | − 1, −1, −1⟩)
2
This state has a somewhat special property: it is a simultaneous eigenstate of the three operators

A = σx1 σy2 σy3 , B = σy1 σx2 σy3 , C = σy1 σy2 σx3 (3.14)

each of which square to identity.

Since the particles are flying apart in different directions, we can perform independent measure-
ments on each of them. Further, on each of them we can choose to measure either σx,y as we like
because these operators act on different spaces and therefore commute.
Suppose we measure the x-component spin of the first, and the y-components of the remaining two.
Since the state is an eigenstate of the first operator with eigenvalue one, the product of the measured
values must be unity.
This means that the x-value of the first particle is uniquely determined by y-values of the second
and the third particles. But these latter measurements are so far away that there is no possible way
they can influence the measurement on particle 1.
Thus, we conclude that the x-spin measurement is detecting/measuring a property which already
has a definite value independent of the measurement.
This is the realism viewpoint which states that the particle 1 actually has the property called
x-component of angular momentum (which was perhaps determined at the time of decay). This is
because the value of this property is determined without even making any measurement on it. The
only way this can happen is if the particle 1 actually already possessed the property in question.
A similar argument can be applied to the x-spin components of particles 2 and 3 as well. More
remarkably, we can also apply the same ideas to conclude that the y-spin components of all particles
must also be real properties.
This means that m1x,y , m2x,y m3x,y are definite numbers which are the values we will get if we perform
measurements on the appropriate particles.
Now recall that our state was an eigenstate of the three operators in eqn:(3.14). This means that
m1x m2y m3y = 1 and similarly for the other two products.
By multiplying all the three, we get m1x m2x m3x = 1 since any m = ±1.
Now, let us consider the outcome of measuring the operator σx1 σx2 σx3 .
Using the reality of the x-spin property, and the fact that we can measure these quantities for each
without affecting the other particles, we can state that the result of measurement must be unity.
On the other hand, consider the product of all the three operators in eqn:(3.14). This product is

ABC = −σx1 σx2 σx3 (3.15)

Using the operator product discussed above, we get the value −1 since the GHZ state is an eigenstate
of the three operators with eigenvalue +1.
This impossibility shows that we cannot assume that the operators six actually had a definite value
in the GHZ state - we cannot say that in the GHZ state, the various particles have a definite property
called the x- or y-component of angular momentum.

3.6.3 GHZ state in QFT

3.7. CHESHIRE CAT 51

Bells Inequality at the Belle Experiment

Bells inequality and Quantum Gravity

3.7 Cheshire cat

This depends on the notion of ‘weak measurement’. Once you accept that, the idea in QM, is that
you can separate a quantum system from its ‘properties’. Just like the Cat in “Alice in Wonderland”
which leaves its grin behind.
https://www.nature.com/articles/nindia.2020.104
I recall that in this paper https://arxiv.org/abs/0807.0972, the author writes the electron
operator in a manner that separates its spin from its Fermi nature. It will be interesting to investigate
its connection to the Poincare group representations and the spin statistics theorem.

3.8 Foundations
There are many recent papers which discuss how the postulates of quantum mechanics can be derived
from other starting points
Masanes et al claim that the measurement postulate is actually redundant
Masanes et al claim that one can get the postulates from information theory
Gisin et al claim to derive the dynamics
Lucien Hardy claims to get the postulates from Probability theory
This famous paper discusses the reality of the state-vector
Quantum equivalence principle
Spacetime is a quantum information channel
52 CHAPTER 3. QUANTUM PHENOMENA
Chapter 4

Schrodinger equation

Having discussed the states and operators and something about measurements, we will now tackle the
question of dynamics.
Dynamics discusses evolution/change of states.
In quantum mechanics, our states are vectors (and often, we ignore the issue of normalization
because all we are interested in is the probabilities).
This means that in time, the only thing a our state vectors can do is “rotate”. If the length/norm
is fixed, then the only thing that can happen is for the “tip” of the vector to move around on a sphere.
Thus, the evolution is controlled by a “rotation” operator in the N-dimensional vector space. In
complex vector spaces such operators are called Unitary matrices (in real vector spaces, such operators
will be termed Orthogonal).
The set of Unitary matrices form a group - U (N ).
In classical mechanics, the future states of a system are determined from its current state by a
differential equation - Newtons II law.
In QM, we have another differential equation, the Schrödinger equation

d
iℏ |ψ(t)⟩ = Ĥ|yt⟩ (4.1)
dt

where Ĥ is a special Hermitean operator called the Hamiltonian. The Hamiltonian decides which
system we are talking about.
But, any Hermitean operator can be the Hamiltonian of some physical system. A key problem in
quantum mechanics is to construct the Hamiltonian for a given system. This is similar to identifying
the forces that will appear in Newton’s law. For e.g., a Hamiltonian operator for a single qbit, placed
in a magnetic field B, can be written as
3
µ X
H= B·σ = µB i σi (4.2)
2 i=1

where µ is the magnetic moment of a single spin.

Exercise 112
√
cos θ2 sin θ2 e−iϕ
   
2 +B 2
Bx y
The energy eigenvectors are and where tan θ = and tan ϕ =
sin θ2 eiϕ cos θ2 Bz
By
Bx .

53
54 CHAPTER 4. SCHRODINGER EQUATION

Our objective will be to learn to solve such equations, and then to derive experimental conclusions
from the time dependent probabilities.
Since the Hamiltonian is a Hermitean operator, it is natural to study the equation using its
eigenvectors as basis.     
ċ1 E1 0 c1
iℏ = (4.3)
ċ2 0 E2 c2
where we have specialized to 2D and denoted the eigenvalues of the Hamiltonian as E1,2 indicating
that these will be the possible energy eigenvalues of the system. c1,2 are the components of the vectors
in the “energy basis” for the vector space.

Exercise 113
How is E1,2 related to µ|B|?

The equation is easily solved.

E1,2 t
c1,2 = c1,2 (0) exp(−i ) (4.4)
ℏ
which means that our state at time t is
c1 (0) exp(−i Eℏ1 t )
 
(4.5)
c2 (0) exp(−i Eℏ2 t )
The integration constants c1,2 (0) tells us our initial state and Schrödinger equation has told us how
the state time evolves.
Before discussing the general case, let us note the following. Suppose that the initial state was
c1 (0) = 1 - this must mean that c2 (0) = 0 because of the probability interpretation.
 −iE t 
e 1
Then, at any later instant of time, our system is always found in the state . (henceforth,
0
I will forget to write ℏ which can always be fixed by checking dimensions).  
0
Thus, the conditional probability that at time t, our system is in the state given that at time
  1
1
t = 0, it was in the state is, in fact, zero.
0
Perhaps, this example is sufficient to convince you that energy eigenstates are Stationary states.
If a system starts out in an energy eigenstate, it will always be found in that same state unless some
measurement happens to change it.
This prototype problem illustrates what happens in general in a quantum system under time
evolution.

Exercise 114
 
1
Rabi Oscillations: We can ask the following question, if we start in a state which is an
0
eigenstate of the σ3 operator, what is the probability that at a later time t, it will be in the
same state.
Ans: cos2 ωt where ω = µ|B|

A more general situation arises when the Hamiltonian operator itself is time dependent. For e.g.,
µ
H = B · σ sin ωt (4.6)
2
 55

which occurs when we have a time dependent magnetic field (such as what might happen when the
spin is illuminated by light).
Again, we can go to the eigenbasis of the Hamiltonian in which the equation becomes
    
ċ E1 sin ωt 0 c1
iℏ 1 = (4.7)
ċ2 0 E2 sin ωt c2

which is quite simple.

But, we can have a more complicated Hamiltonian

H(t) = sin ωtσ1 + cos ωtσ3 (4.8)

which is complicated because the eigenvectors depend on time. In this case, we cannot find the solution
as simply.

Exercise 115
Why is it that if the eigenvectors are time dependent, the Schrödinger equation becomes difficult
to solve.

The new thing that happens is that the commutator of the Hamiltonian at two different times,
[H(t), H(t′ )] ̸= 0.
In this case, the solution of the above differential equation is vastly more complicated.
As usual, Wikipedia provides adequate amounts of references
Feynman Lectures Chapters 8 to 11 are three fantastic chapters.
Sakurai Chapter 2 problems 2, 3, 8,9,
Griffiths, Chapter 3 3.37, 38,
Interesting three level system
56 CHAPTER 4. SCHRODINGER EQUATION
Chapter 5

Position/Momentum operators

Reference: Feynman Lectures vol 3. Chapter 16

We discussed the possibility of infinite vector spaces by using the double slit experiment.
Thus, we can imagine a vector space where the basis vectors are labelled by a real variable

H = {|x⟩, x ∈ R} (5.1)

This also means that Z

dx|x⟩⟨x| = I (5.2)

by analogy with the finite dimensional vector spaces. The second equation above is called a ‘complete-
ness relation’.
Since inner products are a matter of choice, let us define

⟨x||x′ ⟩ = 0, if x ̸= x′ (5.3)
′
= N if x = x (5.4)

A state in this vector space is a linear combination

Z
|ψ⟩ = dx ψ(x)|x⟩ (5.5)

The complex function ψ(x) is called the wavefunction.

By analogy with the finite vector spaces, we can say |ψ(x)|2 is the relative probability that our
system is found in the state |x⟩.
Often, the label x for the basis vector is interpreted as a position label along the real line. In this
view, clearly the different kets represent mutually independent probability events. So we can interpret
|ψ(x1 )|2 | + |ψ(x2 )|2 as the probability that our system is either in the state |x1 ⟩ or the state |x2 ⟩.
This allows us to interpret
Z b
dx |ψ(x)|2 (5.6)
a

as the relative probability to find our system in the position range [a, b].
If these are relative probabilities, to talk about absolute probability requires us to normalize our
states so that the total probability adds up to unity.
For example, we can imagine that the domain of our system is [−L, L] so that, for every state with
wavefunction ψ(x), we require
Z L
dx |ψ(x)|2 = 1 (5.7)
−L

57
58 CHAPTER 5. POSITION/MOMENTUM OPERATORS

1
L can also be infinity. This has the consequence that the wavefunction has the dimension of L− 2 .
For a properly normalized wavefunction, |ψ(x)|2 dx is dimensionless and can be interpreted as the
absolute probability to find the system in a small range dx of states around |x⟩.
Using the inner product defined above, we can compute
Z
⟨ϕ||ψ⟩ = dx ϕ∗ (x)ψ(x). (5.8)

Given such a vector space, the next question of interest for us are Operators. What are the operators
that we can define on this vector space?
In the finite dimensional examples we studied, we were able to choose the eigenvectors of any
Hermitean operator as the basis for our vector space.
In this case, we already have a basis, so we can ask is there an Hermitean operator whose eigenstates
are |x⟩? i.e.,
X̂|x⟩ = x|x⟩ (5.9)
where the x on the RHS multiplying the ket is the real number x labelling the ket.
NB: if X̂ is such an operator, so is f (X̂) where f is an analytic function.
We will pretend that such an operator exists. Then,
Z Z
X̂|ψ⟩ = dy X̂|y⟩⟨y||ψ⟩ = dyy|y⟩ψ(y) (5.10)

i.e., the action of the operator X̂ in the language of wavefunctions is to multiply the wavefunction
ψ(x) by its argument x. We can arrive at the same result by a different manipulation using the inner
product and Hermiticity of X̂,

⟨y|X̂|ψ⟩ = g(X̂|y⟩, |ψ⟩) = g(y|y⟩, |ψ⟩) = y ∗ ⟨y||ψ⟩ = y ψ(y) (5.11)

The difficulty is that if ψ(x) is normalized to unity, it maybe that xψ(x) is not even normalizable.
Of course, there are a large number of functions ψ(x) for which xψ(x) is normalizable. On this
subset, the operator X̂ is a nice operator.
Therefore, if we require operators such as X̂ to exist, the vector space must not be too small (ie it
must allow pretty much any kind of wavefunction).
Since ψ(x) are functions, we can construct many examples. A very important example is the
Gaussian function
1 (x−a)2
ψ(x) = p √ e− 2σ2 (5.12)
σ π

Exercise 116
Plot a graph of this function. How does the shape depend on σ and on a. Is it correctly
normalized.
Normalize the state xn ψ(x).
How does one calculate the integrals that come up?

Exercise 117
Compare the graphs of ψ(x) and |ψ(x)|2

Some comments: the kets |x⟩ are idealized states - they represent the state of a quantum particle
as it exits a slit – provided the slit is really infinitesimal.
 59

Since real slits have finite widths, if a particle exits such a slit, the best we can say is that it is in
a state Z b
1
√ dxeiϕ(x) |x⟩ (5.13)
−b b
representing uniform ignorance about the location of the particle within the width of the slit.
We maybe interested in constructing a normalized state such that its probability is concentrated
at exactly one point.

Exercise 118
Dirac delta function: This object, which is definitely not a function, is defined by two
properties Z ϵ
δ(x) = 0 if x ̸= 0 dx δ(x) = 1 (5.14)
−ϵ

Note that δ(x) is very close to ⟨x||0⟩

R since the latter also vanishes everywhere except at x = 0.
Because of the Rlatter property, dx δ(x − a) f (x) = f (a). Note also that δ(x) = δ(−x).
Now, consider dy δ(x − y)δ(y − z). Using the above property, we have
Z
dy δ(x − y)δ(y − z) = δ(x − z) (5.15)

which looks like Z

dy ⟨x||y⟩⟨y||z⟩ = ⟨x||z⟩ (5.16)

We maybe tempted to identify ⟨x||y⟩ = N δ(x − y) in which case,

R ϵ we get ⟨x||x⟩ = N δ(0). If the
LHS has any finite norm, it is easy to prove that the integral −ϵ dx δ(x) = 0 - so we are forced
to conclude that the the states |x⟩ are not normalizable.
We can model the delta functions by using a limit involving the Gaussian functions
1 2
δ(x − y) = lim √ e−(x−y) (5.17)
σ→0 σ π

or even using a box function (see Feynman 16.3)

The same ideas can be generalized to two and three dimensions.

|x⟩ = |x, y, z⟩ = |x1 , x2 , x3 ⟩ (5.18)

etc. We can also imagine generalizing these ideas to coordinates other than Cartesian type.

|r, θ, ϕ⟩ or |ρ, ϕ, z⟩ (5.19)

etc.
The related questions that arise have to do with the existence of operators such as r̂, θ̂ etc. Since
ϕ is an angular coordinate, can an operator equation like ϕ̂ = ϕ̂ + 2π mean anything?
Suppose we have a system in a state |ψ⟩ - we can ask about the meaning of ⟨ψ|X̂|ψ⟩. Using the
completeness relation, we get Z
⟨ψ|X̂|ψ⟩ = dy|ψ(y)|2 y (5.20)

Since dy|ψ(y)|2 is the probability of finding the system at the position y, the above quantity is the
Expectation Value ⟨X̂⟩ of the position in the state |ψ⟩ or the mean value .
60 CHAPTER 5. POSITION/MOMENTUM OPERATORS

Note that this number does not represent anything at all about any particular system - but it is a
property of the ensemble (identical copies).

Exercise 119
(x−a)2 (x+a)2
Consider a particle with wavefunction ψ(x) = e− 2σ2 +e− 2σ2 ? Normalize the wave function
and plot it.
If you perform a position measurement on the system, where can you expect to find the particle?
What is the expectation value of the position operator? How does your answer depend on the
dimensionless ratio σa ? Why is this the correct question to focus on?
(x−a)2 (x+a)2
How do your answers change for the wavefunction ϕ(x) = e− 2σ 2 − e− 2σ 2 ?

Given that we are talking about expectation values, we can also ask if the average value or the
expectation value is well defined. This will be the case if the standard deviation is small.
The variance or standard deviation or ‘uncertainty’ in the position of a particle in the state |ψ⟩ is
defined as usual q
σX = ⟨X̂ 2 ⟩ − ⟨X̂⟩2 (5.21)

Exercise 120
What is the standard deviation in the two states of the previous problem.
What can you learn from the standard deviation about the expectation value in this case?

5.1 Momentum operator

R
Consider the state dxψ(x)|x⟩ where for ψ(x) we take the Gaussian function for instance.
This state can be interpreted as a particle state which has a high probability of being found at the
position a. In fact, in the limit σ → 0, the probability becomes zero almost everywhere else.
Now, consider the wavefunction ψ(x − A) - this represents the situation where the probability
maximum is shifted to the location a + A. We say that we have ‘translated’ the state by an amount A.
Remember that ψ(x) are the components of the state vector |ψ⟩ in the |x⟩ basis.
Let us try to express the effect of translation directly on the basis vectors so that we can figure out
the operator representing the translation.
Z Z Z Z
∂ ∂
−A ∂x
dx ψ(x − A) |x⟩ = dy ψ(y) |y + A⟩ = dx |x⟩ e ψ(x) = dy ψ(y) eA ∂y |y⟩ (5.22)

allowing us to arrive at a formal equation

∂
|y + A⟩ = eA ∂y |y⟩ = TA |y⟩ (5.23)

defining the operator TAx which is doing the translation for us. We can also write the action of the
same operator on wavefunctions
∂
ψ(x − A) = e−A ∂x ψ(x) (5.24)
∂
which defines the matrix e−A ∂x that represents the operator. Note that if ψ(x) is properly normalized,
∂
the translated wavefunction is also properly normalized. This means that the operator eA ∂y is actually
“unitary” - i.e., it preserves the length of vectors.
5.1. MOMENTUM OPERATOR 61

Of course, we have proved that our translation operator preserves norm for only those wavefunctions
which have finite norm. Since finite norm wavefunctions (and hence vectors) do form a complex vector
subspace, in this subspace, the translation operator is unitary.

Exercise 121
T−A X̂TA = X̂ − A

In 3 space dimensions, we also have TBy , TCz .

Exercise 122
These operators form an Abelian group called the group of Translations

TAx TBx = TA+B

x
, TAx TBy = TBy TAx (5.25)

etc. This group is isomorphic to the group R3 .

Since the elements of the group are labelled by real numbers A, B, C etc, we can have a notion of
continuity. And differentiability.
Consider an infinitesimal translation

∂ ∂
Tϵx = e−ϵ ∂x ≈ (1 − ϵ ) (5.26)
∂x

which indicates that infinitesimal transformations are “close to the identity.”

∂
The operator ∂x is called the “generator” of the translation along the x-direction. It is conventional
in physics to write this as
ℏ ∂
p̂x = (5.27)
i ∂x
defining a “momentum operator” p̂x .
iA
Using the momentum operators, the unitary translation operators are written U = e ℏ p̂

Exercise 123
How can verify that the generators commute [p̂x , p̂y ] = 0

Why is the operator p̂ a “momentum” operator? Answer: It generates translations, just like in
classical mechanics
Corollaries:
p̂2
• A free particle has an energy Ĥ = 2m

•
[X̂, p̂] = iℏ

which is viewed as the quantum version of the classical Poisson Brackets.

62 CHAPTER 5. POSITION/MOMENTUM OPERATORS

Exercise 124
An important part of the understanding of quantum mechanics Rinvolves the free particle.
2 2
Study the time evolution of the ‘Gaussian wave packet’ |ψ⟩ = dxe−x /2σ |x⟩ using the free
p̂2
particle Hamiltonian Ĥ = 2m .
Why does it happen that way?
What happens if we consider
Z
2 2
|ψ⟩ = dxe−(x−a) /2σ +ikx |x⟩ (5.28)

A more interesting problem could be the time dependence of

Z  
2 2 2 2
|ψ⟩ = dx e−(x+a) /2σ + e−(x−a) /2σ +ikx |x⟩ (5.29)

How does the story depend on the sign of k?

Normalizing the eigenstates of the momentum operator and the probability current

P̂ i
⟨ ⟩ (5.30)
m
can be called the probability or the particle current.
It represents the flux of particles moving along the i-direction. We can therefore normalize the
state by fixing this number to be anything we like. For instance, in the LHC particle accelerator, we
can say that the number of particles is 109 protons per second along the beam direction. If we do not
integrate over positions, then we can regard

iℏ ∗
− (ψ ∂i ψ − ψ∂i ψ ∗ ) (5.31)
2m

as the distribution of the particles along the cross-section of the beam. We can also regard the above as
the electrical current in a material, or inside a long protein molecule (such as the Haemoglobin/Chloro-
phyll)

Exercise 125
The probability current vanishes for a real wavefunction.

Thus, thinking about the difference between conductors and insulators, perhaps we can say that in
a conductor, the electron wavefunctions are not real whereas they are real in an insulator.

5.2 Hermiticity
Let us recall the condition that makes an operator Hermitean: for every state |ψ⟩, |ϕ⟩, we require
g(Aϕ, ψ) = g(ϕ, Aψ)
Using the inner product g(ϕ, ψ) = dxϕ∗ (x) ψ(x) we can verify that X̂ is Hermitean.
R
5.3. BCH FORMULAE 63

As for the momentum operator,

Z
ℏ ∂ψ(x)
g(|ϕ⟩, p̂|ψ⟩) = dxϕ∗ (x) (5.32)
i ∂x
∂ϕ∗ (x)
Z Z 
ℏ ∂ ∗
= dx (ϕ (x)ψ(x)) − dx ψ(x) (5.33)
i ∂x ∂x
Z   ∗
ℏ ∂ϕ(x)
= dx ψ(x) = g(p̂|ϕ⟩, |ψ⟩) (5.34)
i ∂x

Hence it is Hermitean provided the total derivative term in the second line evaluates to zero.

• If the range of integration is over all real numbers, we will require ψ, ϕ → 0 at infinity, so the
total derivative will evaluate to zero.
• If the integration is over a finite range [0, L] for e.g, and our wavefunctions vanish at these
endpoints, again the momentum operator will be Hermitean.

• If the integration is over a finite range [0, L] and our wavefunctions vanish are periodic ψ(x+L) =
ψ(x), again the momentum operator will be Hermitean.
An interesting exercise is to understand whether the free particle Hamiltonian is also Hermitean
when the momentum is Hermitean.

−ℏ2 ∂ 2 ψ(x)
Z
g(|ϕ⟩, Ĥ|ψ⟩) = dxϕ∗ (x) (5.35)
2m ∂x2
2
∂ϕ∗ (x) ′
Z 
−ℏ
Z
∂ ∗ ′
= dx (ϕ (x)ψ (x)) − dx ψ (x) (5.36)
2m ∂x ∂x
ℏ2
Z Z 
∂ ∗′
= (ϕ ψ) − dxϕ∗ ′′ ψ (5.37)
2m ∂x
−ℏ2 ′′ ∗
Z
= dx( ϕ ) ψ = g(Ĥ|ϕ⟩, |ψ⟩) (5.38)
2m
provided in each of the intermediate steps the total derivatives integrate to zero.

5.3 BCH formulae

1 1
eA B e−A = B + [A, B] + [A, [A, B] + [A, [A, [A, B]]] + ... (5.39)
2! 3!
BCH formula and its proof
1
eA eB = exp A + B + [A, B] + ... (5.40)
2
Trotter formula
There is this nice paper by M. Suzuki which generalizes the Trotter formula proofs in a useful
manner.
64 CHAPTER 5. POSITION/MOMENTUM OPERATORS
Chapter 6

Examples

65
66 CHAPTER 6. EXAMPLES

6.1 Free Particle

The simplest example of a quantum system, in some sense, is the free particle

p̂2
Ĥ = (6.1)
2m
Note that the Hamiltonian depends only on the momentum operator, and hence

[Ĥ, p̂] = 0 (6.2)

which means that we energy eigenstates are also momentum eigenstates.

Exercise 126
The more careful statement is that these can be chosen to be so. Since we are working in
infinite vector spaces, it is possible that there are states on which one operator acts nicely while
the other is not defined.

6.1.1 Hilbert space

Given the Hamiltonian operator, one of the first question that is usually asked is to find its energy
eigenstates.
Ĥ|ψ⟩ = E|ψ⟩ (6.3)

Exercise 127
We can write this equation in terms of wavefunctions

ℏ2 ∂ 2 ψ(x)
− = Eψ(x) (6.4)
2m ∂x2

This is an ordinary differential equation which can be solved.

ψ(x) = Aeikx + Be−ikx (6.5)

q
where k = 2mE ℏ2 for any sign of E.
Case 1: E > 0
The individual energy eigenstates are periodic ψ = eikx and not normalizable. We can fix the
normalization as mentioned at the end of the previous chapter
The energy states are doubly degenerate.
Ex. 128 — What is the spatial probability distribution for a particle in the state ψ(x) = Aeikx +
Be−ikx
Ex. 129 — What is the expectation value of the momentum for a particle in the state ψ(x) =
Aeikx + 2Be−ikx
−|B 2 |
Ans: |A|
|A|2 +|B 2 | ℏk

Ex. 130 — For a particle in the state ψ(x) = Aeikx + Be−ikx , what is the conditional probability
that if the energy is is E the momentum is k?
6.1. FREE PARTICLE 67

Case 2: E < 0 In this case, if we consider e±kx , these solutions are not even bounded let alone
normalizable.
We might imagine constructing a combination such as

ψb (x) = e−px , x ≥ 0 = epx , x ≤ 0 p > 0 (6.6)

which can be normalized in the usual way. This state has very small probability of being found far
away from x = 0 and thus might represent the wavefunction of an electron which is bound to a nucleus
for eg.
But it is not differentiable at x = 0 (why not?) - so both the Momentum and the Hamiltonian
operators cannot act on it at all.
We can make things better by “smoothing the tip” thereby making a good wavefunction; thus, the
above wavefunction can give a useful approximation of a bound state.
Since E is the eigenvalue of a Hermitean operator (i.e, the Hamiltonian) you might expect that
these states, eikx for various k and epx for various p, are orthogonal.

Exercise 131
Show that
⟨ψb |p̂|ψb ⟩ = 0 (6.7)
if the wavefunction is real ψb = ψb∗

6.1.2 Dynamics
Given an arbitrary state ψ(x), to find its time dependence, we must express it as a linear combination
of energy eigenstates X
|ψ⟩ = |E⟩⟨E||ψ⟩ (6.8)
E

Exercise 132
P
Show that this can be written in terms of the wavefunctions as ψ(x) = E cE ψE (x) where
Z
∗
γe = dxψE (x)ψ(x) (6.9)

where ψE (x) are the energy eigenfunctions.

ikx
e
R
Since ψb (x) is not an energy eigenstate, it can be expanded ψb (x) = dkc(k) √ 2π
where c(k) =
2ik
√
2πp(k2 +p2 )
.
ikx
e
The wavefunctions √ 2π
are eigenfunctions of the momentum and the Hamiltonian operators. The
normalization is determined by the completeness relation
Z
dkeikx = 2πδ(x) (6.10)

which must be memorized. This means that if we define

e−ikx
Z
ψ̃(k) = dxψ(x) √ (6.11)
2π
68 CHAPTER 6. EXAMPLES

then,
eikx
Z
ψ(x) = dk ψ̃(k) √ (6.12)
2π
2
ikx ikx−i ℏk t
e
The time evolution of √ 2π
gives the wavefunction e √2π2m .
We can now study the time evolution of the state ψb (x) defined before
e−ikx
Z Z
ℏk2 t 2ik
ψb (x, t) = dEe−iEt ψE (x)c(E) = dke−i 2m √ √ (6.13)
2πp(k 2 + p2 ) 2π
to get
ℏp2 t √
ψb (x, t) = ei 2m p e−p|x| (6.14)
which is what you might have expected. Clearly, this solution does not spread with time. Check
dimensions and normalization.
Somewhat amusingly, we can also translate the state ψb by first writing in terms of momentum
eigenstates
e−ikx e−ik(x−b)
Z Z
2ik 2ik
Tb |ψb ⟩ = dk √ √ 2 2
Tb |k⟩ = dk √ √ |k⟩ (6.15)
2π 2πp(k + p ) 2π 2πp(k 2 + p2 )
to get ψb (x − b) which is exactly what you might expect from drawing pictures.
In this procedure of expanding in terms of eigenstates a minor miracle has occurred which seems to
suggest that the Hamiltonian and the Momentum operators acted on the state directly - even though
ψb was not in the domain of definition of these operators.

6.2 Infinite potential well

In this problem, we will consider a system defined by the Hamiltonian with a potential
p̂2
Ĥ = + V (x) (6.16)
2m
where the potential is
V (x) = 0 0<x<L (6.17)
= ∞ otherwise (6.18)
This potential is an extreme idealization which is nevertheless insightful.
As is customary, we can proceed to construct the Hilbert space by finding all the eigenfunctions of
Ĥ.
We will do this by solving the differential equation and considering boundary conditions (which
will make the Hamiltonian Hermitean).
Ĥψ(x) = Eψ(x) (6.19)
We can solve this differential equation piecewise.
Since V (x) is infinite, if we consider expectation values of the energy, they will be meaningless
unless ψ(x) = 0 in those regions. This suggests boundary conditions ψ(x) = 0 x < 0 andx > L.

Exercise 133
With these boundary conditions, the Hamiltonian becomes Hermitean. Then, its eigenfunctions
will form a basis for the vector space of states (Hilbert space).
6.2. INFINITE POTENTIAL WELL 69

ℏ 2 n2 π 2
This fixes E = 2mL2 .
By plotting the first few eigenfunctions, we can make the following general observations

• The ground state has no nodes.

• The first excited state has one node and so on.. n is related to the number of nodes.

• The more ψ varies with x, the higher its energy - this is because

ℏ2 ′ 2
Z
⟨Ĥ⟩ = dx |ψ | + ⟨V (x)⟩ (6.20)
2m

This also suggests that we can decrease the energy by smoothing the wavefunction.

• The more the nodes the more the variation.

• The normalization of all the ψn (x) does not depend on n. This is a bit surprising.

• It is instructive to study the classical system.

• And compare with QM.

The theorem about number of nodes etc. can be found in the textbook on Differential equations
by Simmons in a beautiful section on Sturm-Liouville systems.

Exercise 134
In the class, the potential considered was centered around the origin.
In that case, the energy eigenvalues are

ℏ2 (n2 )
E= (6.21)
4mL2
and the corresponding wavefunctions are
nπx nπx
ψn (x) = cos sin (6.22)
2L 2L
for odd and even n respectively.
Using those, we can verify the earlier statements about nodes again.
Note that all these eigenfunctions are either odd or even under the symmetry x → −x. So there
is no degeneracy. The other function with the same eigenvalue does not satisfy the boundary
conditions.
Is it mysterious that no such symmetry was visible in the previous discussion?
70 CHAPTER 6. EXAMPLES

Exercise 135
Using the eigenfunctions, we can expand
X nπx
cos x = cn sin( ) (6.23)
n
L

Find cn .
Observe that the LHS does not satisfy the boundary conditions whereas the RHS does. To
understand how this can still be possible, plot LHS and RHS close to the walls x = 0 and
x = L.
The LHS is said to equal the RHS “almost everywhere” – meaning that that the equality can
fail at a discretely separated set of points. Therefore, the equality in the above equation must
be handled with care.

Exercise 136
Consider ψb (x) = x(1 − x). Normalize and find the time dependent wavefunction by expanding
along the energy basis. cf: Griffiths

Exercise 137
ipx
Consider the wavefunction ψb (x) = e Box[x − 1/2, 1/4] where the Box function

Box[x − b, a] = 1 |x − b| < a =0 otherwise (6.24)

Study its time evolution.

Ans: q First normalize ψb . Then, expand in terms of energy eigenfunctions ψb (x) =
P 2 nπx
RLq2 nπx
n cn L sin L . where cn = 0 L sin L ψb (x).
En t
q
The time dependent state is then given by ψb (x, t) = n cn e−i ℏ 2 nπx
P
L sin L .
Approximate the Box function, by keeping only the first N = 3 energy eigenfunctions in the
expansion. Choose some numerical values, and plot your wavefunction at various times.
How does this evolution change when you increase N .
Watch what happens at the walls.
How does this depend on b.
Such a box wavefunction has its probability concentrated at a single location. Thus, we might
want to think of this as corresponding to the classical situation of a particle at rest.
However, it is not a momentum eigenfunction. This means that it is a superposition of momen-
tum eigenfunctions. This of course means that we cannot say that the wavefunction describes
a particle at rest (i.e., momentum zero).
With time, the different energy eigenstates pick up relative phases. And the probability center
p
in position space moves with velocity m .
It is worthwhile observing that the wavepacket moves similar to the Gaussian free particle
packet as long as it is far from the walls.
6.3. SHO 71

6.3 SHO
This is the most important quantum system. One reason is that every system is approximated by a
SHO near its minima. Secondly, its an exactly solvable system. Thirdly, it appears as a building block
in the quantum theory of fields.
1 k
V (x) = mω 2 x2 = x2 (6.25)
2 2
which is the potential of a classical oscillator.
Exactly solvable systems are important because they demonstrate validity of ideas. Approximants
are only as good as the range of convergence which is almost always zero in QM.

Exercise 138
Summarise the classical physics of the oscillator.

6.3.1 Energy levels

p̂2 1
H= + mω 2 x2 (6.26)
2m 2
The range of x is now R.

Exercise 139
p, x, H are all hermitean provided ψ is square integrable.

Therefore, the vector space can be taken to be the set of square integrable complex functions on R.
We first need to construct a basis for the Hilbert space. As usual, we do this by finding energy
eigenvalues and eigenfunctions.
p2
Consider the classical Hamiltonian 2m + 12 ω 2 x2 and rescale
√
p = mωp′ (6.27)
r
1 ′
x= x (6.28)
ωm

Note that the commutation relation does not change [x′ , p′ ] = iℏ. The Hamiltonian

ω ′2 2 ω
H= (p + x′ ) = (x′ + ip′ )(x′ − ip′ ) (6.29)
2 2
1 1 1
where [x′ ] = [p′ ] = LM 2 T − 2 = [ℏ] 2 The order of the factors does not matter in CM, but in QM it
is important. This operator ordering ambiguity is resolved by Weyl-ordering which says that we must
fully symmetrise any such classical expression in QM.

ω
Ĥ = [(x′ + ip′ )(x′ − ip′ ) + (x′ − ip′ )(x′ + ip′ )] (6.30)
4
72 CHAPTER 6. EXAMPLES

Exercise 140
Define the non-Hermitean operators
1 1
â = √ (x̂′ + ip̂′ ) ↠= √ (x̂′ − ip̂′ ) (6.31)
ℏ ℏ

and show that [â, ↠] = 1 and that

1
Ĥ = ℏω(a† a + ) (6.32)
2
Show that
[a† a, a] = −a [a† a, a† ] = a† (6.33)

Because of the last equation above, Ha† |ψ⟩ = a† H|ψ⟩ + a† |ψ⟩.

If |ψ⟩ is an energy eigenstate, Ha† |ψ⟩ = (E + 1)a† |ψ⟩ i.e., a† |ψ⟩ is also an energy eigenstate with
eigenvalue (E + 1). Since a† increases the energy, it is called a raising operator.
Similarly, a|ψ⟩ is also an energy eigenstate with eigenvalue (E − 1). Since a decreases the energy,
it is called a lowering operator.

Exercise 141
Let |ϕ⟩ = a|ψ⟩ - show that
E 1
⟨ϕ||ϕ⟩ = ( − )⟨ψ||ψ⟩ (6.34)
ℏω 2
E
Since both LHS and RHS involve the norm of a state we must have ( ℏω − 12 ) > 0.

By using the lowering operator a repeatedly, we can decrease the energy arbitrarily. But the above
equation says that this is not possible. This must mean that there is a state |0⟩ such that

a|0⟩ = 0 (6.35)

Thus, the lowering operator is also sometimes called the ‘annihilation’ operator.

Exercise 142
†
If a a||ψ⟩ = λ|ψ⟩, where the integer part of λ is n−1, then |ϕ⟩ = an |ψ⟩ has eigenvalue λ−n < 0.
Consider |γ⟩ = a|ϕ⟩. It has norm,

⟨γ||γ⟩ = ⟨ϕ|a† a|ϕ⟩ = (λ − n)⟨ϕ||ϕ⟩ < 0

which is not possible.

The only way out is if a|ϕ⟩ does not exist - i.e., an |ψ⟩ = 0 = a(an−1 |ψ⟩).
6.3. SHO 73

Thus, we have found all the energy levels of the system.

1
|0⟩ E0 = ℏω
2
3
|1⟩ ∼ a† |0⟩ E1 = ℏω
2
5
|2⟩ ∼ (a† )2 |0⟩ ∼ a† |1⟩ E2 = ℏω
2
and so on... We have not used the equality sign because we assume that the LHS is normalized but
we must normalize the RHS properly to make this work.
It is a little surprising that the energies do not depend on the mass. Classically the energy will be
E = 12 mω 2 A2 where A is the amplitude of oscillation. The quantum energy levels do not possess any
property that we may call amplitude.
We can perform any calculation we like on this system without ever calculating the wavefunctions
and performing integrations.
Ex. 143 — Assume |0⟩ is normalized. Work out the norm of (a† )n |0⟩
Here is a clever trick: since [a, a† ] = 1 we can think of a = dad† or the other way round a† = − da d
. This
ℏ d
is very similar to how we think of p̂ = − i dx .
Ans: Then, ⟨0|(a)n (a† )n |0⟩ = ⟨0|( dad† )n (a† )n |0⟩ = n!. This also can be used to show the orthogonality
of |m⟩ and |n⟩ for m ̸= n.

Ex. 144 — Calculate ⟨x̂⟩, ⟨p̂⟩, ⟨x̂2 ⟩, ⟨p̂2 ⟩ for the n-th energy level.
αp̂
Ex. 145 — Consider the state |α⟩ = ei ℏ |0⟩. Write it in terms of the energy eigenfunctions.
1
Ans: We have to use the BCH formula, which for this special case becomes eA+B = eA eB e 2 [A,B] .
αp̂ α † −α † −α † α α2 †
ℏ (a −a) e ℏ a e 2ℏ2 ([a ,a])
ei ℏ |0⟩ = e ℏ (a−a ) |0⟩ = e |0⟩ = e ℏ a |0⟩ (6.36)

from which we see that

αp̂ −α2 X αn
ei ℏ |0⟩ = e 2ℏ2 √ |n⟩ (6.37)
n n!

Exercise 146
An important exercise from the view of entrance exams is to calculate σx2 = ⟨x2 ⟩ − ⟨x⟩2 and
σp2 = ⟨p2 ⟩ − ⟨p⟩2 and check that σx σp ≥ ℏ2 or something like that.

6.3.2 Wavefunctions
However, we will calculate the wavefunctions.
The wavefunction corresponding to the |0⟩ can be computed by

1 ∂
⟨x|a|0⟩ = 0 = √ (x̂′ + ℏ ′ )⟨x||0⟩ (6.38)
ℏ ∂x

Solving the differential equation,

1 x2
ψ0 (x) = ⟨x||0⟩ = √ 1 e− 2σ2 (6.39)
σπ 4
74 CHAPTER 6. EXAMPLES

where I have also fixed the normalization and written the wavefunction in terms of the original variables.
Here σ 2 = .
The first excited state can be obtained by using |1⟩ = a† |0⟩

1 ∂ x2
ψ1 (x) = √ (x̂′ − ℏ ′ )e− 2σ2 = (6.40)
ℏ ∂x

upto normalization. This can be repeated to find all the wavefunctions

x2
ψn (x) = Hn (x)e− 2σ2 (6.41)

where Hn (x) is the n-th Hermite polynomial. The following facts about the Hermite polynomials
• The n-th polynomial has degree n − 1
• The Hermite polynomials are alternately even and odd.
• The n-th polynomial has (n − 1) nodes.
• The n-th polynomial has a node between two successive nodes of the (n-1)-th polynomial.
agree with our expectations about the degeneracy and Sturm-Liouville theory.
The first few Hermite polynomials are:
• H0 = 1
• H1 = x
• H2 = 2x2 − 1
• H3 = 2x3 − 3x
Note that when x is large, the wavefunctions behave as
2
/2σ 2
ψn ∼ xn−1 e−x (6.42)

Exercise 147
which can directly read off from the Schrödinger equation

ℏ d2 mω 2
− ψ+ x ψ = nψ (6.43)
2mω dx2 2ℏ
by using the substitution ψ = eiS(x)/ℏ . It is very useful to learn such asymptotic analysis. cf:
Bender and Orszag
(S ′ )2 − ℏS ′′ + 2x2 = 2n
and if we consider the region x2 >> ℏ
mω , we can drop the second and the n-term.
q
ℏ
The length scale is mω .

Exercise 148
αp̂
What is the wavefunction corresponding to ei ℏ |0⟩?
6.4. OTHER INSTRUCTIVE EXAMPLES 75

Exercise 149
d
We are used to thinking of p̂ = ℏi dx when acting on wavefunctions.
∂
Why not use x̂ = iℏ ∂p when acting on wavefunctions which are functions ψ̃(p) of a momentum
variable so that p̂ψ̃ = pψ̃.
Check that the commutation relations are satisfied.
Find the wavefunctions of the SHO, this time in terms of momentum wavefunctions. Is there
anything interesting about your answer?

We found the wavefunction by starting with aψ(x) = 0.

Let us study the eigenvalue problem directly starting with the Hamiltonian:

d2
− ψ + x2 ψ = λψ (6.44)
dx2
x2
by assuming ψ(x) = H(x)e− 2 . (see Griffiths).
The wavefunction is not square integrable for λ not an integer.
Thus, in the previous section, when we assumed that λ was a real eigenvalue of a† a, the eigenvector
actually did not exist (because the wavefunction is not square integrable). In this sense, we did not
find an eigenvector - only a root of the characteristic polynomial.
This funda is not visible when starting with aψ = 0.

Exercise 150
There exist many Hamiltonians H which can be written as a product H = a† a of operators.

6.3.3 Time evolution

i
As usual, an energy eigenstate time evolves with a phase factor e− ℏ Ĥt .
Thus, if we express the state of our system as a linear superposition of the energy eigenstates, we
can immediately find the time evolution.

Exercise 151
αp̂
What is ⟨x̂⟩(t), ⟨p̂⟩(t), ⟨x̂ ⟩(t), ⟨p̂ ⟩(t) for the state ei
2 2 ℏ |0⟩?

Griffiths: coherent state problem.

6.4 Other instructive examples

These sections mention several other quantum systems which are very useful in building intuition
about quantum phenomena.

6.5 Finite potential well

This is an important problem since many quantum systems involve particles with short range interac-
tions for which these are good models.
76 CHAPTER 6. EXAMPLES

Examples are diodes and transistors, quantum dots, atomic traps etc. It can also be used to discuss
the photoelectric effect and work-function etc.
Let us define the potential energy symmetrically

V (x) = −V0 − L < x < L = 0else. (6.45)

Because of the symmetry the wavefunctions can be assumed to be either even or odd (ie have a definite
parity).
2
Given this potential, we have a dimensionless parameter α = mLℏ2V0 for a particle of mass m.
The formula for the n-th energy level can be written as
ℏ2
En = V0 f (α) = g(α)
mL2
where the unknown functions can also depend on the integer n.

Energy levels
We solve the Schrödinger equation piecewise. First the even states,

ψ(x) = Ae−p|x| |x| ≥ L (6.46)

= B cos qx |x| ≤ L (6.47)

with another possibility, the odd states,

ψ(x) = Ae−px x>L (6.48)

= B sin qx |x| ≤ L (6.49)
+px
= −Ae x < −L. (6.50)

Substituting into the Schrödinger equation, we find

ℏ2 p 2 ℏ2 q 2
E=− = − V0 (6.51)
2m 2m
This is possible only if E < 0 and fixes q in terms of p.
The wavefunction and its first derivative are required to be continuous.

Exercise 152
Why?

This gives us the following conditions

Ae−pL = B cos qL Ape−pL = Bq sin qL (6.52)

from which we get

q
1= tan qL (6.53)
p
We can find solutions of this equation in the following manner. Writing it as
2mV0 ℏq
2 2
− 1 = tan2 √ α.
ℏ q 2mV0

Plot the LHS and RHS as a function of x = √ ℏq for different values of α. If the two curves
2mV0
intersect, we have an energy level.
6.5. FINITE POTENTIAL WELL 77

Exercise 153
1
Check that the maximum number of bound states is α

Exercise 154
Observe that the number of nodes increases with energy. Can you show that the lowest energy
wavefunction is even.

Exercise 155
What is the probability that a particle trapped in this potential well will have momentum p?
Ans: To figure this, we have to first construct the momentum eigenstates. The wavefunctions
d
for these solve the differential equation −iℏ dx ψ = pψ. The solutions are ψl (x) = √1V eilx where
the normalization factor has been inserted by hand (V is the volume of spacetime which has,
in this case, dimension Length).
First we normalize our bound states
e−2Lp
Z
sin 2qL
dx|ψ(x)|2 = A2 + B 2 (L + )=1 (6.54)
p 2q

where B = Ae−pL sec qL.

Then, we can calculate the momentum space wavefunction ⟨p||ψ⟩ with |ψ⟩ our bound state
wavefunction.
2Ae−Lp (p cos lL + l sin lL)
Z
1
⟨l||ψ⟩ = dx √ e−ilx ψ(x) = √ (6.55)
V V (p2 + l2 )
2B
+√ (l sin lL cos qL − cos lL sin qL) (6.56)
V (l2 − q 2 )

upto overall normalization.

This gives us the relative probability of the particle in the bound state being found with
momentum l.
What is the most probable momentum? Does it agree with your expectation?

Note that the particle has vanishing probability of being at large distances. This is definitive of a
bound state.

6.5.1 Reflection and Scattering

ℏ2 q 2
For E > 0, ie 2m > V0 the wavefunctions “outside” the well become qualitatively different.

ψ(x) = eikx + Re−ikx x < −L (6.57)

ilx −ilx
= αe + βe |x| ≤ L (6.58)
ikx
= Te x > L. (6.59)
2 2 2 2
where ℏ2Ml − V0 = ℏ2M k
= E > 0 These are scattering solutions - there is nonzero probability of finding
the particle at arbitrary far off positions.
78 CHAPTER 6. EXAMPLES

Exercise 156
Calculate
ℏ dψ dψ ∗
J(x) = −i (ψ ∗ −ψ ) (6.60)
2m dx dx
ℏk
in the region to the right J(x) = 2m |T |2 and J(x) = 2m
ℏk
(1 − |R|2 ) to the left.
dJ
Note that dx = 0 which means that the current to the left should be the same as the current
to the right.
Conservation of the probability current therefore requires |R|2 + |T |2 = 1
Contrast with Exercise (165)

Using continuity and differentiability,

we get
i sin 2qa 2
R= (l − k 2 )T (6.61)
2kl
e−2ika
T = 2 2 (6.62)
cos 2la − i (k 2kl
+l )
sin 2la

and α = and β = . The transmission coefficient

p
V02 8mL2 (E + V0 )

1
= 1 + sin2
|T |2 4E(E + V0 ) ℏ

Its obvious that 0 < |T |2 < 1

Exercise 157
Observe that for x < 0, the solution is the superposition of an incident, left moving wave of the
ei(px−ωt) and a reflected right moving wave e−i(px+ωt) . This terminology is correct and comes
from considering the probability current as well.
For x > 0, we only have a transmitted wave. This is not something special, but this is the kind
of solution we have constructed

Exercise 158
This solution is quite weird since we have reflection even though there is no barrier.

Exercise 159
It is possible to construct the scattering type solutions as even/odd solutions.

Exercise 160
Transfer matrix method:
6.5. FINITE POTENTIAL WELL 79

6.5.2 Step potential Barrier

It is easy to solve this problem which is defined by the potential

V (x) = V0 |x| < L, = 0, else (6.63)

We can simply replace V0 by −V0 in the previous calculations.

Exercise 161
What happens? Are there any bound states?

6.5.3 Same calculations done differently

The even bound state wavefunctions are given by

 N e ρ(x+a) x < −a
ψ(x) = N cos µx sec µa −a ≤ x ≤ a (6.64)
N e−ρ(x−a) x>a


µ2 ρ ℏ2 ρ2
and the normalization constant N 2 = µ2 +ρ2 1+aρ where −E = 2m > 0 is the bound state energy and
ℏ2 µ2
E + V0 = 2m > 0.
The condition for a bound state is
2µρ ρ
sin 2µa = or tan µa = . (6.65)
ρ2 + µ2 µ

Analogously, the odd bound states are given by


 −N e ρ(x+a) x < −a
ψ(x) = N sin µx cosec µa −a ≤ x ≤ a (6.66)
N e−ρ(x−a) x>a


The condition for a bound state is

2µρ ρ
sin 2µa = or cot µa = . (6.67)
ρ2 + µ2 µ

The even (and real) scattering states are


 Ak (cos la − i kl sin la) e ik(x+a) + c.c. x < −a
ψk (x) = Ak 2 cos lx −a ≤ x ≤ a (6.68)
Ak (cos la − i kl sin la) e−ik(x−a) + c.c. x > a


q q
where l = 2m(E+V
ℏ2
0)
and k = 2mE
ℏ2 . Note that these are “standing” waves since ψ−k (x) = ψk (x).
For completeness, I should also construct the odd “standing waves”. This is given by

 Bk (cos la − i kl sin la) e ik(x+a) − c.c.


x < −a
ϕk (x) = Bk 2 i kl sin lx −a ≤ x ≤ a (6.69)
k −ik(x−a)
−Bk (cos la − i l sin la) e − c.c. x > a


and again ϕ−k (x) = ϕk (x) (note that ϕ−k (x) is pure imaginary) .
80 CHAPTER 6. EXAMPLES

Exercise 162
I found it an interesting exercise to show that these are all orthogonal to each other.

Exercise 163
I have not succeeded in showing that these form a complete set. This is an important problem.
This is equivalent to a proof of the spectral theorem for this Hamiltonian. This is an important
thing for chemists in the case of the 3d-Hydrogen problem - because it is a question of what
are all the necessary wavefunctions to be used in computer programs.

Exercise 164
Is there a difference between these two kinds of states apart from their normalizability.
Entanglement characterization.

6.6 Delta function potentials

These potentials are easy exercises compared to the previous.

Attractive case and bound states

V (x) = −V0 δ(x) (6.70)
We can directly find the wavefunctions, which must take the form

ψ(x) = e±ipx or e±px (6.71)

which must be continuous, but

ϵ ϵ
ℏ2
Z Z
′′
− ψ (x)dx = (E + V0 δ(x))ψ(x)dx (6.72)
2M ϵ ϵ

In the limit of ϵ → 0, we get

ℏ2
− (ψ ′ (+ϵ) − ψ ′ (−ϵ)) = V0 ψ(0) (6.73)
2M
ℏ2
For E < 0, if we take ψ(x) = Ae−p|x| because of normalizability, we get AV0 = 2M 2pA which gives us
√
V02 M
the energy eigenvalue E = − . There is only one bound state. The normalization factor A = Vℏ0 m .
2ℏ2
For E > 0, let us take ψ(x) = Aeipx for x > 0. Then, let us try ψ(x) = Beipx for x < 0. Using
ℏ2
continuity A = B, using the jump condition, we get − 2M ip(A − B) = V0 which is not possible.
So, let us try ψ(x) = Beipx + Ce−ipx for x < 0. Continuity now gives A = B + C and the first
derivative gives
2M
ipA − (Bip − Cip) = − 2 V0 A (6.74)
ℏ
iM iM V0
This gives C = − pℏ 2 V0 A and B = A(1 + pℏ2 ).

Amazingly, we again have a reflected and transmitted wave.

6.6. DELTA FUNCTION POTENTIALS 81

So, if we imagine that we have a left moving wave 1eipx for x < 0, then we have a transmitted wave
Teipx
and a reflected wave Re−ipx .
−iM V0
1 pℏ2
T = iM V0
R= (6.75)
1+ pℏ2 1 + iM V0
pℏ2

Exercise 165
2 2
We must expect that |T | + |R| = 1. Why?
Check this in any case.

6.6.1 Repulsive delta and phase shifts

V (x) = +V0 δ(x) (6.76)
ℏ2
In this case, if we try the bound state problem, we get AV0 = − 2M 2pA
which is inconsistent with the
assumption that p > 0.
Thus there are no bound states.
The scattering states can be determined by simply changing the sign of V0 in the previous calcu-
lations.

Exercise 166
Why does this work?

The thing to learn from these examples is the phase shift in the transmitted and reflected waves.
Especially the signs. An attractive potential (V < 0) leads to a positive phase shift in the transmission
while a repulsive potential (V > 0) makes the phase shift negative.
From the eqn (6.75),
T = teiϕ R = rei(ϕ+π/2) (6.77)
where tan ϕ = − M V0
pℏ2 .
The sign of the phase shift depends on the sign of V0 . If V0 < 0 then the potential is attractive,
and the phase shift in the transmitted wave is positive.
The reflected wave involves an additional π/2 phase shift.
A repulsive V0 > 0 potential produces a negative phase shift.
In the finite potential well, let us check these ideas.
The phase shift for the potential well is (6.61)
k 2 + l2
eiϕ = e−2ika+iψ tan ψ = tan 2la (6.78)
2kl
Thus,
2 2
k +l
tan ψ − tan 2ka tan 2la − tan 2ka
tan ϕ = tan(−2ka + ψ) = = 2kl 2 +l2 .
1 + tan ψ tan 2ka 1 + tan 2ka k 2kl tan 2la
In this case, is it clear that V0 > 0 implies positive phase shift√ϕ < 0? Plot graphs?
For small incident momentum k we can try to check. l ∼ V0
√
V0
2k tan 2la − 2ka
tan ϕ = √ .
1 + a V0 tan 2la
Scattering of identical particles by a 1d-Dirac delta function barrier potential: The role of statistics
82 CHAPTER 6. EXAMPLES

6.6.2 Poles of the S-matrix

When does the transmission or the reflection coefficient become infinite?
Consider the attractive Dirac delta - the denominator (6.75) vanishes when
ℏ2 p = iM V0 (6.79)
√
If we write p = E, we get the binding energy equation for the bound state in this potential.
Let us repeat this exercise with the step potential. The transmission coefficient (6.61) becomes
infinite when
2kl
= i tan 2la (6.80)
k 2 + l2
where k 2 = E and l2 = E + V0 . Again, we can get a solution if we rotate k = iq with q > 0 whence
2ql l2 + q 2
= tan 2la =⇒ = sec 2la (6.81)
l − q2
2 l2 − q 2
from which we get the bound state condition:
l2
= tan2 la (6.82)
q2
Thus, poles of the S-matrix in the Energy plane give the bound state energies.
In fact, it is also possible to calculate the wavefunctions from the reflection and transmission
matrices using the residues at the poles.

6.7 Double potential wells

Double delta
V (x) = −V0 (δ(x − a) + δ(x + a)) (6.83)
Using the symmetry, we can construct two bound states:
ψ0 = e−p|x| , A cosh px (6.84)
−p|x|
ψ1 = sign(x)e , A sinh px (6.85)
Let us find the energy difference between these two levels.
Continuity gives A cosh pa = e−pa and A sinh pa = e−pa . The derivative discontinuity gives
−pe−pa −Ap sinh pa = −V0 e−pa and −pe−pa −Ap cosh pa = −V0 e−pa respectively. Thus the eigenvalue
conditions becomes
V0
−p + V0 = −p tanh pa = 1 − tanh pa (6.86)
p
V0
−p + V0 = −p coth pa = 1 − coth pa (6.87)
p
It is possible to see that there is always one energy level from the first equation. But the second one
has a solution only if maV
ℏ2 > 1?
0

The question is:

How to find the energy difference.
This calculation is forms an important tool in the arsenal of a quantum mechanic.
Reference: Sidney Coleman: “The uses of instantons.”
Instantons are tunneling configurations which lift the degeneracy in a non-perturbative manner.
It is weird to me that the word ‘tunneling’ which is completely fictitious in this context is somehow
viewed as a satisfactory explanation for the lack of degeneracy in 1-D quantum systems (see Witten’s
lectures, in Quantum Fields and Strings, vol II). However, these words are prophetic in the sense that
Gamow was able to calculate the properties of α-decay and in particular explain the Geiger-Nuttall
law by using this “picture” of tunneling.
6.8. PÖSCHL-TELLER 83

6.7.1 Double potential well

This very important problem has potential

V (x) = 0 |x| < a (6.88)

= −V0 (x) a < |x| < L + a (6.89)
= 0 else (6.90)

The reason this is important is that the ground state is non-degenerate. Following the same steps as
in the previous problem, we can find the energies for the odd and even functions.
But, the energy difference between the ground state and the first excited state is an interesting
calculation with important funda.
What happens to bound states

6.8 Pöschl-Teller
This potential is important because its a 1-d model for typical interatomic forces. It is also important
because of its mathematical features.
1 x
V (x) = − λ(λ + 1)sech2 ( ) (6.91)
2 σ
Wolfram demos have graphs of the wavefunctions.

Exercise 167
Ramsauer-Townsend effect and reflectionless potentials.
In quantum mechanics, we have unusual things when compared with classical mechanics.
Situations in which there is perfect transmission.

The constant theme in this course has been that the comparison of Quantum Mechanics with
classical systems must always be taken with a grain of salt.

6.9 Summary
By considering 1-D quantum problems using wavefunctions, we focused on the following.

• The Schrödinger equation is now a differential equation which can be solved piecewise.
• The wavefunction is required to be continuous
• The first derivative of the wavefunction can have a jump discontinuity.
• Energy eigenstates are found by applying restrictions on the solutions of the differential equation.
These restrictions include
1. Normalizability
2. Continuity
but, we note that the continuity condition is effectively the same as making sure that the diff
eqn is solved everywhere.
• It is after we find all the energy eigenvalues (with some idea of the inner product) that we have
a vector space which defines the quantum system.
84 CHAPTER 6. EXAMPLES

• In general, the eigenvalues consist of a Discrete set together with a continuous spectrum of
energies.
• The latter are associated with wavefunctions that are not normalizable (but behave like plane
waves asymptotically).
These maybe normalized by considering the probability current instead of the total probability.
For these, only relative probability makes sense.
• The bound states are associated with the Discrete part and their wavefunctions fall off rapidly
with distance.
In this case, the asymptotic behaviour may (long-range) or may not (short range) depend on the
energy level label n.
The normalization of the wavfunctions also may or may not depend on the energy n.
• The bound state wavefunctions are characterized by the nodes. Increasing the number of nodes
increases the energy.
• The number of states below energy E is, to a good approximation, the volume in phase space
defined by the equation H < E divided by the Planck constant h (not ℏ). Roughly, we have one
quantum state per h-volume in phase space.
• In 1-D discrete symmetries do not give rise to degeneracies. This is because of “tunnelling” or
instantons which remove the degeneracy.
• Because of the previous, we can assume that our wavefunctions are actually mapped into them-
selves (upto an overall phase) under the symmetry trafo.
This is very useful in simplifying calculations.

• When considering scattering type states, the sign of the phase shift tells us the nature of the
potential.
In fact, the potential can be reconstructed from the full k dependence of the phase shift.
• Poles of the reflection and transmission coeffficents (more generally the S-matrix) in energy give
rise to bound states.
Residues at the pole are |ψb |2 of the bound state wavefunction associated to the pole energy.
• Many times, the quantum energies and wavefunctions are analytic functions of the parameters
in the Hamiltonian. Horne conjecture
When they are not analytic functions, we have quantum phase transitions.

• Did I miss something?

6.10. MIDSEM II SOLUTIONS 85

6.10 MIDSEM II solutions

R∞ x2 √
1 What is the value of the integral −∞
dxe− 2σ ? 2πσ
∗
2 For energy eigenstates, J(x) = iℏ(ψ ∗ ∂ψ ∂ψ
∂x − ψ ∂x ) does not vary with x. True or False. Explain.
The answer is True. For solution refer notes or book.
2 For a free particle with wavefunction ψ(x) = Aeikx + Be−iqx , what is the conditional probability
that if the energy is E the momentum is k? Answer: P (k|E) = 1.
P (k&E) ℏ2 k 2 ℏ2 k 2
P (k|E) = p(E) but since [H, p] = 0 P (k&E) = 0 if E ̸= 2m and = p(E) if E = 2m .

1 If the initial state of a quantum simple harmonic oscillator is |1⟩ + |3⟩, compute ⟨x̂⟩(t) Answer
zero.
q
1 Show that the equation x12 − 1 = tan(αx) has finitely many solutions for any α.
2
This equation can be rewritten αy 2 = cos2 y Since cos(0) = 1, there is at least one solution. LHS
exceeds unity when y > α. So, in the region 0 < y < α, there can be finitely many intersections
which can be seen visually from a graph.

3 What is the probability that the measurement of the momentum of a particle which is in
the ground state of the SHO yields a value p? The wavefunction for the ground state is
  41
mωx2
mω
πℏ e− 2ℏ .
  12
p2
Ans: 1
mℏωπ e− mωℏ dp

For the SHO, because of the special property of the Hamiltonian, the energy levels are easy to
  14
p2
calculate in momentum basis mωℏπ1
e− 2mωℏ .

5 It is known that at time t = 0, a particle of mass m is located somewhere in the range [−2, 2]
with a constant wavefunction.
√
Answer: ψ(p) = sin(2p)/ Lp.
If there are no forces acting on this particle, what is the relative probability that it will have
energy E at a time t = 1?
√ 2 √
Answer: ψ(t) = sin(2p)/ Lpe−ip tℏ/2m prob = (sin(2p)/ Lp)2
What is the probability that at time t = 2, it will be found in the region [−1, 1] ?
−ip2 tℏ/2m
R1
Answer: ψ(x, t) = dpe−ipx sin(2p) and −1 dx|ψ(x, t)|2
R
(2πp) e

5 Suppose the potential experienced by a particle of mass m is V (x) = V0 x < 0 and zero
otherwise.
What are the energy levels and wavefunctions of this system?
86 CHAPTER 6. EXAMPLES
Chapter 7

Symmetry

87
88 CHAPTER 7. SYMMETRY

What is a symmetry?
In physics, we regard symmetries as transformations which act on the phase space variables.
Classically, a symmetry leads to multiple solutions of the equations of motion with the same energy.
It is an organizing principle that allows us to group solutions.
If we have two such transformations, we can perform them successively. This leads to the idea of
composition of transformations which is also a transformation.
However, in general, the set of transformations may not form a group. That is to say, some
transformations may not have inverses.
For the purposes of this course, we will consider only those sets of trafos which do form a group.
Symmetries come in two types: Discrete and Continuous.
In the latter case, the trafos depend on parameters that are real variables.
In quantum mechanics, a transformation is represented by an operator - since we need to change
states into other states.
Wigner’s theorem: A symmetry of a quantum system is represented by Unitary (or anti-Unitary)
operators.
This is because unitary and anti-unitary operators preserve norm (by definition). Then probabilities
are invariant under the transformation.
And if all experimental quantities only depend on probabilities, then those are also unchanged
under the trafo. This is what we expect of a symmetry.
Definition: A symmetry of a quantum system is an operator Q that commutes with the Hamiltonian
operator H.

[H, Q] = 0

Because of this, the energy eigenstates H|En ⟩ = En |En ⟩ are also eigenstates of Q.

Exercise 168
If two Hermitean operators commute [A, B] = 0, then there exists a basis of common eigenvec-
tors.
This must be true: if |a⟩ is an
P eigenvector of A, it can be written as a linear combination of
the eigenvectors of B |a⟩ = n αn |bn ⟩. Then commutativity demands that all the |bn ⟩ must
have the same B eigenvalue (in this subspace, if A is a diagonal matrix and B is not a diagonal
matrix, they wont commute).

In general, we can have the following:

several energy levels of the Hamiltonian have the same Q eigenvalue
and
several Q eigenstates have the same energy eigenvalue.
The consequence of the former is that under time evolution, the eigenstates of Q will not, in general,
be stationary.
The consequence of the latter is that energy eigenstates need not possess a property that corresponds
to the operator Q.

Exercise 169
Write an example for the two cases above using matrices. Is it possible for both to occur in the
same quantum system.
7.1. SYMMETRIES IN CLASSICAL MECHANICS 89

In QM, the transformation is obtained from the operator Q by exponentiation. U = eiαQ where α
is a real number.

Exercise 170
U is automatically unitary - since Q is Hermitean.

Since eiαQ eiβQ = ei(α+β)Q etc, the set of operators

{eiαQ , α ∈ R} (7.1)

form an abelian group. Note that these elements may not all be distinct.
In a given Hilbert space H, we can have eiAQ = 1 while in a second Hilbert space eiBQ = 1. To
put it differently, the range of α can be different in different Quantum systems.

7.1 Symmetries in Classical Mechanics

References
Mukunda, Simon et al in Resonance.
My notes for PHY 301

7.1.1 Symmetries in 1D
Reflection symmetry
V (x) = V (−x) (7.2)
This example has already been studied.
As far as energy levels are concerned, if ψ(x) is an eigenfunction, we can expect ψ(−x) to also be
an eigenfunction with the same energy eigenvalue.
However, in 1D QM, ψ(x) = ±ψ(−x) so that degeneracies do not occur.
This is a special case of the more general possibility: ψ(−x) = eiϕ ψ(x), the probabilities computed
using the wavefunction will not change. Thus, the physics (i.e., all expectation values) will not change.
Note that ϕ must be a constant for this to be true. When the phase the wavefunction picks up is
not independent of position under a symmetry transformation, we have a gauge symmetry.
If we repeat this transformation once more,

ψ(− − x) = ψ(x) = e2iϕ ψ(x)

which implies that eiϕ = ±1.

Thus, we have even or odd wavefunctions.

Discrete translation symmetry

Periodic potentials.

V (x + a) = V (x) (7.3)
The object of the study is to show that energy eigen-functions form bands.
iαp
Recall that the finite translation as above is produced by the operator e ℏ . The translation sym-
metry means
iαp̂ −iαp̂
e ℏ H e ℏ = H.
90 CHAPTER 7. SYMMETRY

Exercise 171
iαp
The eigenstates of the translation operator e ℏ are the functions ψ(x) = eikx
Since the operator is unitary, its eigenvalues are complex numbers with unit modulus.

We then have Bloch’s theorem which states that the energy eigenstates take the form ψ(x) =
eikx u(x) where u(x + a) = u(x). Proof:
For each energy eigenvalue, the eigenfunctions can be assumed to also be eigenfunctions of the
translation operator.
In general, it must be a linear combination of such eigenfunctions of the symmetry operators. But,
in 1D energy levels are non-degenerate.
The general eigenstate of the translation operator can be written ψ(x) = eikx uk (x) where uk (x +
a) = uk (x).
Proof: ψ(x + a) = eika ψ(x) as an eigenstate. Then u(x) ≡ e−ikx ψ(x) is invariant under the lattice
translation.
Aliter:
τ (a)|θ⟩ = eiθ |θ⟩ (7.4)
can be solved by
⟨x|τ (a)|θ⟩ = ⟨x − a||θ⟩ = ψ(x − a) = ⟨x||θ⟩eiθ = eiθ ψ(x). (7.5)
We can now substitute ψ(x) in the energy eigenvalue equation

Hψ = Eψ (7.6)

and try to determine the energy levels of the system. Since the functions uk depend on k, the energy
is also likely to depend on k which is a real variable. Thus, we can have continuously varying energy
levels - i.e., bands.
Of course, it can also happen that the eigenvalue problem has eigenstates only for special values
of k, in which case, we will not have bands. Additionally, we do not know, as yet, if the energy levels
are degenerate. Or whether there are band gaps.
But there is one more curiosity: the eigenvalue of the operator Ta does not uniquely determine k.
2π
This is because eika = ei(k+ a )a. This means that two k values define the same Ta eigenvalue.
The energy levels will only depend on the values of k within the Brillouin zone. Since k and k + 2π
a
define the same ψ and u, the energy levels cannot depend on the absolute value of k but only on values
of k mod 2π 2π 2π
a . The range − a < k ≤ a is called a Brillouin zone .
Bloch’s theorem.

7.1.2 Kroning-Penney
2
p
If H = 2m + V (x), then the Schrödinger equation becomes

ℏ2 ′′
− (u − 2iku′k − k 2 uk ) + V (x)u = Euk (7.7)
2m k
P
Suppose V (x) = −V0 a n δ(x − na), can you find the energy levels?
Answer:
2ma2 V0 sin(κa)
cos(ka) = cos(κa) −
ℏ2 κa
ℏ2 κ2
where E = 2m and k is the Bloch wavenumber.
7.1. SYMMETRIES IN CLASSICAL MECHANICS 91

x Tan@xD approx xΒ
1000
100 1.5
10
1.0
1
0.1
0.5
0.01
Β
0.5 1.0 1.5 1 2 3 4 5

ℏ2 κa
We must see that we have energy bands. The RHS can be massaged a bit using tan ϕκ = 2ma2 V0 :

sin(κa − ϕκ )
cos(ka) = . (7.8)
sin ϕκ
Given that LHS is less than unity, we have solutions for all values of κ such that RHS is less than
unity.
In the range 0 < κa − ϕκ < π2 and 0 < ϕκ < π, RHS is always less than unity. Thus, we have a
continuous family solutions of the above equation for a fixed RHS.
2
Gaps between bands will occur when the RHS is greater than unity, i.e., −β = − maℏ2V0 > κa κa
2 tan( 2 )
2
This gives a range of κ for a given maℏ2V0 . If β > 0, then this inequality is met for x = π2 + ϵ until xβ ,
where xβ tan xβ = −β with xβ < π. Therefore, the bandgap is

2ℏ2 x2β ℏ2 π 2
∆E = − (7.9)
M a2 2M a2

Exercise 172
Find a good approximation to xβ .
Answer: If β < 1, then xβ < 3π 4 , so, it is in the linearish part of tan. A good approximation in
√
3 2 9+12β−3
this range is obtained from Taylor expansion β = x(x + x /3) which gives xβ = 2 .
The first figure shows the actual curve x tan x in blue against the two approximations - purple
for the above, and green for the below function.
If β > 1, then it is in the rapidly rising part of tan for which an excellent approximation
√ √
is the
6β−2 3 3(β+1)2 +π 2 +π 2 +6
Taylor expansion β = −1 + π6 (x − π2 ) − 2x−π
π
from which we get: xβ = 2π
These solutions for xβ are unique in this range and are shown in blue for 0 < β < 1 and in purple
for β > 1. These are only valid until xβ = π2 . The curve in green shows the approximation for
xβ using only the first term in the Taylor expansion. The dots are numerical solutions. The
approximations are somewhat poor for 1 < β < 2.

https://arxiv.org/pdf/2310.07920.pdf
The Kronig-Penney model however uses the finite potential well instead of the delta functions.
Technically more complicated without much newer insights.
It is useful to appreciate how energy levels which were originally degenerate can become non-
degenerate via quantum mechanical tunnelling.
To this end, let us study the Dirac delta potentials. For a single delta function potential, we have
exactly one energy level.
If we take two attractive delta’s, the energy levels are well approximated by symmetric and anti-
symmetric combinations.
92 CHAPTER 7. SYMMETRY

For the single delta-well,

ℏ2 ∂ 2 ℏ2 ∂2
H=− 2
− V0 δ(x) = 2
( 2 − δ(y)) (7.10)
2m ∂x 2ma ∂y

where a is a length scale determined by the binding energy. The wavefunction then ψ(x) ∼ √1 e−a|x| =
a
√1 e−|y| , From this we see that the probability is significant for |x| > a.
a
If we have two delta function potential wells,

ℏ2 ∂ 2
H=− − V0 aδ(x − a) − V0 aδ(x + a) (7.11)
2m ∂x2

we may imagine that the bound states are ψ(x) ∼ √1a e−a|x−a| and ψ(x) ∼ √1a e−a|x+a| , ie the two
individual bound states. Thus, we can expect that the ground state is doubly degenerate corresponding
to particles ‘trapped’ in either well.
If this is actually correct, we must expect that the Hamiltonian operator is diagonal in the basis
made of these two wavefunctions There are many scattering states also, but those will have E > 0.
Then, those are guaranteed to be orthogonal1 to any linear combination of these two states. Thus,
we can look for a basis for the bound states alone and ignore the scattering states. This is similar to
studying the bands of a conductor, while omitting those states of the electrons which allow them to
escape the conductor altogher (scattering states).
Let us write the Hamiltonian operator using these wavefunctions as basis.
 
⟨I|H|I⟩ ⟨I|H|II⟩
H= (7.12)
⟨II|H|I⟩ ⟨II|H|II⟩

where |I⟩ is the eigenstate of the first well ignoring the second and |II⟩ is the energy level of the second
well ignoring the first.
The off-diagonal elements of this matrix are given by the integral
Z
dx ψI∗ ĤψII (7.13)

which is very small because of the exponential falloffs, but nonzero. So,
 
E A
H= (7.14)
A E
2
A
which will give us the energy eigenvalues as approximately E ± 2E . This means that the true eigenstates
are a linear combination. Is it obvious that the symmetric combination will have lower energy?
2 2
a
Let us estimate the energy difference: A ≈ ℏ2m |ψ(0)|2 where 1 = a|ψ(a)|. Observe how the energy
c
−ℏ
difference depends on Planck’s constant as e . The usual quantum mechanical perturbation theory
for the energy levels (or the wavefunctions) can be interpreted as a series in ℏ.
This sort of contribution to the energy, which cannot be expanded as a Taylor series in ℏ is therefore
regarded as the signature of a non-perturbative effect.

7.1.3 Variations on this theme

Reference: Feynman Lectures vol III. Chapters 10-15.
For three delta functions - such as applicable for a cyclo-propane molecule.
For four delta functions?
For six delta functions? Benzene?
1 inspite of Bound states in the continuum
7.2. HIGHER DIMENSIONAL PROBLEMS AND SYMMETRY 93

In all these cases, the Hamiltonian is well approximated by a tri-diagonal system.

Each time we add another ’site’ potential - it contributes one more energy level and additionally,
the previous wavefunctions are rearranged slightly.
Dirac Comb minus one: Impurities and bands.
Quantum Information Meets Quantum Matter – From Quantum Entanglement to Topological
Phase in Many-Body Systems
Topological phases and quasiparticle braiding

7.2 Higher dimensional problems and symmetry

In two dimensions, we can have more possible symmetries since our wavefunctions are functions of two
variables ψ(x, y).
Lattice periodicity in 2D: V (x + a, y + b) = V (x, y).
Again we have Bloch’s theorem ψ(x, y) = eikx x+ky y uk (x, y) where uk although periodic may depend
on the vector k. Again, we have a band of energy values that depend on the vector k in a continuous
manner.
Graphite

7.3 Central potentials and Angular Momentum

In 2D we can also perform rotations

ψ(x, y) → Tθ ψ(x, y) = ψ(x cos θ + y sin θ, −x sin θ + y cos θ) = eiθLz ψ(x, y) (7.15)

The first equality defines the operator and hence the transformation. The second equality defines the
operator Lz and states that the rotation is obtained by exponentiating the operator Lz .

Exercise 173
∂ ∂
Verify that Lz = −iℏ(x ∂y− = x̂ × p̂.
y ∂x )
For quantum mechanics purposes, the important observation is that the operator Lz is Her-
mitean and therefore the operator Tθ is unitary.

We can write the above equation in polar coordinates in the plane.

ψ(ρ, ϕ) → Tθ ψ(ρ, ϕ) = ψ(ρ, ϕ + θ) = eiθLz ψ(ρ, ϕ) (7.16)

Exercise 174
∂
In this case, Lz = −iℏ ∂ϕ . Is it the same as the one before? How is the sign fixed?
94 CHAPTER 7. SYMMETRY

Exercise 175
So far, we were using operators x̂ and p̂x which acted on wavefunctions ψ(x) as xψ(x) and
−iℏψ ′ respectively.
d
It appears to be a reasonable question to talk about operators such as ϕ̂ and pϕ = −iℏ dϕ . But
this is not as simple, because there is an identification on the ϕ coordinate.
It is more natural to imagine an operator ẑ = eiϕ which is automatically takes care of the
periodicity. But, which however does not have a logarithm (recall that operators need not have
logs).
Another difficulty will be that this operator satisfies a more complicated commutation relation
with pϕ which is defined as the translation operator on ϕ.
We do not discuss these mathematical intricacies here and will simply avoid using the operator
ϕ entirely.
However, we can still talk about |ψ(ρ, ϕ)|2 dρρdϕ as the relative probability of finding our system
in the region around the point (ρ, ϕ).

Eigenfunctions of the operator Tθ will have eigenvalues which are unit complex numbers as before.
The same argument as Bloch’s theorem leads to:
ψ(ρ, ϕ) = eimϕ um (ρ) (7.17)
but, now we have an additional restriction arising from the fact that ϕ and ϕ + 2π represent the same
point in the plane. This means that m must be an integer. Note that although um (ρ) is invariant
under rotation, the function um may depend on m.
While we have not constructed the functions um (ρ) and uk (x, y) appearing in Bloch’s theorem, we
seem to have constructed eigenstates of the operators Tθ and Ta .
The advantage of such eigenfunctions is that if the Hamiltonian is rotationally invariant, we can
search for energy eigenfunctions among these kinds of states. This greatly simplifies our calculations
since the Schrödinger equation which was a PDE has now become an ODE.

7.3.1 Free particle in 2D

A first exercise is to find the energy levels of a free particle moving in 2D.
p2 +p2 1
This is an easy problem since H = x2M y , so ψk = √2πV eiK·x . The interesting question is to find
the energy levels in polar coordinates using Bloch’s theorem. First show that Schrödinger equation
becomes
1 d m2 2mE
(ρu′m ) − 2 um = − 2 um = −k 2 um (7.18)
ρ dρ ρ ℏ
Is it ok to assume E > 0? Repeat with E < 0?
Opening the derivatives out, we get
ρ2 u′′ + ρu′ + (k 2 ρ2 − m2 )u = 0 (7.19)
which is Bessel’s equation with solutions Jm (kρ) and Ym (kρ).
There are several features of interest. First we see a scale symmetry ρ → αρ, k → α1 k. For large
values of ρ, we can see that r
2 ±ikρ∓i( mπ + π )
Jm ± iYm ∼ e 2 4 (7.20)
πρ
q
The ρ1 dependence cancels the factor ρ in the integration measure to ensure orthogonality
Z Z
d2 xψn∗ ψm = dϕdρρ ψn∗ ψm (7.21)
7.3. CENTRAL POTENTIALS AND ANGULAR MOMENTUM 95

After that cancellation, we will have dϕe±i(m−n)ϕ which is the same δ-function orthogonality as in
R

the cartesian coordinates.

This is a general idea in all cases, the asymptotic behaviour for scattering type states always has
a factor which cancels those in the integration measure. For bound-states also, a similar idea is true
except that we will have exponentials e±mρ and normalizability will pick out the negative sign. Note
that the long-distance behaviour depends on the energy eikρ as expected.

Exercise 176
Near the origin of polar coordinates, we must expect a problem. This is because the polar angle
ϕ becomes ill defined at ρ = 0. To put it differently, one point, namely the origin, is represented
by (ρ = 0, ϕ) with any value for ϕ.
What should it mean for the wavefunction ψ = eimϕ um (ρ) ?
If we think of the wavefunction as a function, then for one point in the domain, it must have
a unique (complex) value, but clearly at ρ = 0, our ψ varies with ϕ. One way out is that
u(ρ = 0) = 0 so that ψ is single valued.
A second idea is that the wavefunction itself is not experimentally ever measurable. Its always
|ψ|2 d2 x which is measured and for this to be finite and sensible, it is enough if u(ρ) ∼ ρa with
a > − 21 . In this case, our wavefunction is not single valued - but probabilities are.
Clearly, this is an experimental question depending on the kind of system we are studying.

For the free particle, the Bessel functions Jm ∼ ρm and Ym ∼ ρ−m for m ̸= 0 and for m = 0 J0 ∼ 1
and Y0 ∼ log ρ. This means that we will not allow Y -type components since those always blow up and
the origin is a regular point in our space (potential energy terms which blew up at ρ = 0 would make
the origin a singular point).
So, we do not obtain any restrictions on k from either infinity or from the origin.
In summary, the energy levels of a free particle are labelled by (k, m) with wavefunctions Jm (kρ)eiϕ
2 2
k
and energy E = ℏ2M and where m is an integer and k a real number. Applying Lz to these energy
imϕ
eigenfunctions e (Jm ) = ⟨ρ, ϕ||k, m⟩, we see that we get

Lz |k, m⟩ = mℏ|k, m⟩ (7.22)

which means that we have a simultaneous eigenfunction and mℏ is the angular momentum of the state.
The energy does not depend on the m−value, and hence each energy level is infinitely degenerate.
The meaning of the degeneracy is clear: since k 2 = Kx2 + Ky2 , the energy depends only on the
magnitude of the momentum and not the direction. Then angle ϕ is however, not the direction of the
linear momentum K.

Exercise 177
We can normalize these free particle eigenstates by using the probability current, both in
Cartesian and Polar coordinates.
iℏ
Jr = (ψ ∗ ∂r ψ − ψ∂r ψ ∗ )
2M
Show that the sign of k determines whether particles are ingoing or outgoing. What about
iℏ
Jϕ = (ψ ∗ ∂ϕ ψ − ψ∂ϕ ψ ∗ )
2M r
96 CHAPTER 7. SYMMETRY

7.3.2 2D SHO
p2x + p2y 1
H= + mω 2 (x2 + y 2 ) (7.23)
2M 2
In the language of differential operators,

ℏ2
 2
∂2 ℏ2
 2
1 ∂2
 
∂ 1 2 2 2 ∂ 1 ∂ 1
H=− + + mω (x + y ) = − + + + mω 2 ρ2 (7.24)
2M ∂x2 ∂y 2 2 2M ∂ρ2 ρ ∂ρ ρ2 ∂ϕ2 2
∂
which makes it clear that Lz = −iℏ ∂ϕ commutes with H. Clearly, we have a rotation symmetry.
Energy levels and degeneracies:
H = ℏω(nx + ny + 1)
where nx = a† a and ny = b† b.
Compute the degeneracies as a function of the energy. There is a pattern here - if you type the
sequence of degeneracies that you get in The On-Line Encyclopedia of Integer Sequences, what do you
get?
Observe [H, a† b − b† a] = 0, [H, a† a] = [H, b† b] = 0. Lz ∼ [a† b − b† a, a† a]?
See Mallesh, et. al. for more information.

7.4 Particle in constant electric/magnetic fields

A very important problem as you can imagine.
The Hamiltonian for this system can be written

p2x + p2y + p2z q

H= ⃗ − qA0
+ p̂ · A (7.25)
2M m

where the scalar electromagnetic potential A0 = −E ⃗ · ⃗x for a constant eletric field and the magnetic
vector potential
A⃗ = − 1 (⃗x × B)
⃗ (7.26)
2

Exercise 178
Describe the classical motion of a charged particle in constant electric and magnetic fields.
What are the cases to consider ?
What can we expect in the quantum problem?

7.4.1 Constant magnetic field

in 2D we can set up this problem,

Exercise 179
Show that, if the magnetic field is along the z-axis, the Hamiltonian can be written

p2x + p2y qB q2 B 2 2
H= + Lz + (x + y 2 ) (7.27)
2M 2M 8M
7.4. PARTICLE IN CONSTANT ELECTRIC/MAGNETIC FIELDS 97

Note that [Lz , H] = 0. In polar coordinates, we can use simultaneous eigenfunctions of Lz and H
as ψ = u(ρ)eimϕ to get

ℏ2 1 d ℏ2 m 2 M ωc2 2
− (ρu′ ) + 2
u+ ρ u = (E − ℏωc )u (7.28)
2M ρ dρ 2M ρ 2
The remarkable result is that the energy is again independent of the Lz = mℏ eigenvalue. These are
the famous Landau levels and we have infinite degeneracy coming from m.
In some sense, since magnetic fields do not do work, the energies of a particle in a constant magnetic
field must be the same as those of the free particle. This is what we get. Except that instead of two
real numbers Kx,y , we get two integers. Of course, for the same free particle, we got one integer and
one real number when we used polar coordinates.
The same physical system is also described by another Hamiltonian (corresponding to a different
gauge choice for the vector potential). In this case, we get something different.
A couple of interesting variations that arise are electrons in a periodic potential together with a
constant magnetic field and the Integer quantum Hall effect and Fractional quantum hall effectsintegers.
Other 2D type problems which might be mathematically interesting are to compare the energy
levels of a particle moving freely on a the surface of a sphere (positive curvature), a paraboloid and a
hyperboloid (negative curvature), x2 + y 2 + z 2 = 1, x2 + y 2 = z 2 , x2 + y 2 − z 2 = ±1 respectively.
98 CHAPTER 7. SYMMETRY

7.5 Motion in 3D

In this section, we will study the motion of particles in three spatial dimensions.
Lattice problems bring in no new features.

Exercise 180
A system which has a periodic potential is necessarily a solid. To prove this, we have to show
that
• It produces Bragg peaks (easy)

• It has a nonzero “shear modulus” which is the defining property of a solid.

The shear modulus is computed, in Many particle Quantum Mechanics, from an expec-
tation value of momentum operators.
Liquids are said to ‘assume the shape of the container’. This expression can be given a precise
meaning: liquids have zero shear modulus. Liquids will also not give rise to Bragg peaks since
the presence of Bragg peaks implies nonzero shear modulus.
Supersolids give rise to Bragg peaks, but have zero shear modulus.

Many times, we have two “particles” which are moving under the effect of mutual forces. Typically
such forces can be modelled by a Potential energy which is a function of the position variables of the
two particles V (x1 , x2 ). Thus V is a function of six variables.
We may expect that it should not matter whether the pair is here or on the moon so long as their
relative separation is the same.
That is we expect the potential to be invariant under the translation x1,2 → x1,2 + a. Then the
potential simplifies to V (x1 − x2 ). Now V is a function of only 3 variables.

Exercise 181
Derive Newton’s third law. What is the associated conserved quantity?

If we have reason to expect that the relative orientation of the two “particles” does not matter,
the potential can be further simplified to a central potential V (|x1 − x2 |) = V (r). This potential is
invariant under rotation x1,2 → R(α, n̂)x1,2 . This final Central potential is a function of only one
variable r.
In 3D, we can perform rotations in three different planes.
Let us define rotation angles ϕ1,2,3 corresponding to rotations in the yz,zx,xy planes respectively.
Then, the corresponding operators will be denoted Lx,y,z respectively.
The operator was shown to be Lz = −iℏ(x∂y − y∂x ). Similarly, Lx = −iℏ(y∂z − z∂y ) and Ly =
−iℏ(z∂x − x∂z ).
7.6. ANGULAR MOMENTUM 99

Exercise 182
In this case, the rotation operators Lx,y,z commute with the Hamiltonian.

p2x + p2y + p2z

H= + V (r) (7.29)
2M
There are two things to learn here.
First, the algebraic manipulations.
Second, the geometric observation.

In operator language,
p⃗2 = −∇2 = ∆ (7.30)
defines the Laplacian operator ∆. The eigenvalues of the Laplacian operator are geometric in origin
and carry valuable information into the nature of the space on which the particles are moving. For
all spaces, the eigenvalues of the Laplacian are non-negative, making it clear that it is a Hermitean
operator (with the usual square integrable wavefunctions).
The spectrum of the Laplacian is nearly unique which makes it possible to hear the shape of a
drum.

7.6 Angular momentum

Exercise 183
Check that
[Lx , Ly ] = iℏLz [Ly , Lz ] = iℏLx [Lz , Lx ] = iℏLy (7.31)
These commutation relations define the angular momentum algebra so(3).

If we have a Hamiltonian that commutes with two of Lx,y,z , then it automatically commutes with
the third.
In spherical polar coordinates, the Hamiltonian for a particle of mass M in a potential V is

p⃗2
H=+ V (r, θ, ϕ) (7.32)
2M
 
1 2 1 2
Use classical mechanics to show that H = 2M pr + r2 L where

1
L2 = L2x + L2y + L2z = p2ϕ + p2θ (7.33)
sin2 θ
∂ ∂
But, pθ ̸= ∂θ and neither does pr = ∂r .
∂ ∂ ∂ ∂
In polar coordinates, Lz = −iℏ ∂ϕ while Lx = iℏ(sin ϕ ∂θ + cos ϕ cot θ ∂ϕ ) and Ly = iℏ(− cos ϕ ∂θ +
∂
sin ϕ cot θ ∂ϕ ) although we will rarely, if ever, require the second two operators.
Nanophotonic electron accelerator: This paper is an illustration of how little quantum mechanics
is actually used in even designing experiments about quantum systems.
If [H, Li ] = 0, we can try to look for simultaneous eigenfunctions of these. Does this mean [L2 , Li ] =
0?
100 CHAPTER 7. SYMMETRY

But, since [Li , Lj ] ̸= 0, we cannot have simultaneous eigenfunctions of all three Li . The best we
can do is to find eigenfunctions of one of the three – the number of simultaneously diagonalizable
operators in an algebra is called the rank of the algebra. so(3) is rank 1.
It is traditional to use Lz eigenfunctions eimϕ .
Thus, the eigenfunctions of the Hamiltonian by Bloch’s theorem take the form
ψE = eimϕ F (r, θ) = ⟨r, θ, ϕ||m, E⟩ (7.34)
where m is an integer, but we can do better.

Exercise 184
How did we use Bloch’s theorem? Is that all that was used?
Consider the unitary operator Tψ = eiLz ψ , its eigenfunctions ψm will have eigenvalues eiα(ψ) .
Repeated application of Tψ (and continuity) will lead us to conclude that α = kψ. In that case,
if we consider F = e−ikψ ψm , it is invariant under the action Tψ F (r, θ, ϕ) = F (r, θ, ϕ + ψ) =
F (r, θ, ϕ).
Since, this is true for any ψ, and we assume continuity: ψE = eikϕ F (r, θ). That is, F is
independent of ϕ′
In the case of the crystal, the F part was required to be only periodic because, there, ϕ was
restricted to the periodicity. Here it is independent of ϕ because ϕ was arbitrary.
But, k must be an integer, because ϕ and ϕ + 2π represent the same point (for single valuedness
of wavefunctions.)

Consider the operators, L± = Lx ± iLy . Show that

[Lz , L± ] = ±L± (7.35)
which immediately means that
Lz L± |m, E⟩ ∼ |m ± 1, E⟩ (7.36)
Thus, we can increase or decrease the eigenvalue of the Lz operator by using the ‘raising’ and ‘lowering’
operators L± . This means that if we have one eigenstate of Lz , we can make many more.
Addtionally, since [L2 , Lz ] = 0, we can imagine that our kets are simultaneous eigenkets of L2 and
Lz . Let us denote the eigenvalue L2 |β, m, E⟩ = β|β, m, E⟩ and Lz |β, m, E⟩ = mℏ|β, m, E⟩.
Now, let us compute the norm of L± |β, m, E⟩
⟨β, m, E|L∓ L± |β, m, E⟩ = ⟨β, m, E|(β 2 − ℏ2 m2 ± ℏ2 m)|β, m, E⟩ ≥ 0 (7.37)
because
1
L2 = (L+ L− + L− L+ ) + L2z = L+ L− + L2z − ℏLz = L− L+ + L2z + iℏLz . (7.38)
2
Thus, there is a largest m value and similarly a smallest m value for a given real β. This means that
if we start with a state on which L+ |j, E⟩ = 0 or L− | − j, E⟩ = 0, where j is an integer, we will not be
able to raise or lower states indefinitely.
For the highest or lowest such state, we must have L± |β, j, E⟩ = 0 respectively. But this immedi-
ately determines the eigenvalue of the ‘total angular momentum’ β 2 = ℏ2 (j 2 ± j).

7.6.1 Spherical Harmonics

Thus, our energy eigenkets take the form, |j, m, E⟩. In position language, we can write these wave-
functions:
 j−m
⟨θ, ϕ||j, m, E⟩ ∼ ⟨θ, ϕ|Lj−m
+ |j, −j, E⟩ = e iϕ
∂θ + i cot θ∂ϕ ⟨θ, ϕ||j, −j, E⟩ (7.39)
7.6. ANGULAR MOMENTUM 101

Its actually simpler to use Cartesian coordinates for this.

Observe that Lz (x + iy)m = mℏ(x + iy)m . Also,

L± = ±ℏ{−(x ± iy)∂z + z(∂x ± i∂y )} (7.40)

which both do not change the degree of the function they operate on.

Exercise 185
This is the statement that they commute with ∂r = xi ∂i which rescales the r coordinate.
What does this mean for the eigenfunctions of the Li ?

This allows us to write down the entire vector space as the space of Homogeneous polynomials in
the 3 variables x, y, z.

7.6.2 Legendre Polynomials

It is easy to see that (x + iy)j = eijϕ rj is an eigenstate of Lz . L+ acting on such a state gives zero.
We now apply L− which replaces one occurrence of z with (x − iy) and one occurrence of (x + iy)
with z.
It is perhaps clear that at each step we have decreasing powers of (x + iy) and increasing powers
of x − iy. By repeating this, we can see that at the end we will be left with (x − iy)j = e−imϕ rj .
At each order, we will always have degree homogeneous j polynomial in x,y,z. The word ‘homoge-
neous’ means that each term will have rj .
Its easy to write these down directly.
l Ylm
0 1
1 x + iy, z, x − iy
2 (x + iy)2 , (x + iy)z, z 2 , z(x − iy), (x − iy)2
3 (x + iy)3 , (x + iy)2 z, (x + iy)z 2 , (x − iy)z, ...
Note that each row is proportional to rl . We can divide each by rl to make them dimensionless.
Clearly this is relevant to understand how the probability density of these states will behave.
We then obtain the spherical harmonics Ylm (θ, ϕ) if we substitute for x, y, z in terms of polar
coordinates.

Exercise 186
Since this is quite different from standard textbooks, but much easier to remember, we must
check all the usual stuff.
But first, why am I not writing (x + iy)m (x − iy)k z (l−k−m) type terms?
Answer: (x + iy)(x − iy) = x2 + y 2 = r2 − z 2 so every such appearance can be “reduced” to
due to the appearance of r2 . The degree of the polynomial does not change.
Are these eigenstates of L2 ?

If we remove a factor of eimϕ from the spherical harmonics we get the Associated Legendre Poly-
nomials.
Ylm (θ, ϕ) = N eimϕ Plm (cos θ)
where the normalization factor will be computed below. Can you see Bloch’s theorem operating?
For e.g., in the l = 2 case if we apply L− on (x + iy)2 , we get 2z(x + iy) and again z(x + iy) gives
2
−(x2 + y 2 ) + 2z 2 = −r2 + 3z 2 . In the last term, we see the Legendre Polynomial P20 (cos θ) = 3 zr2 − 1.
102 CHAPTER 7. SYMMETRY

If we again apply L− only the z 2 will be differentiated and etc. In the table, instead of 3z 2 − 1, I have
written only z 2 . The entries in the table are not orthonormal.
1
If we take (x + iy) 2 and apply L− we get √ z ∼ cot θ which blows up on the z−axis. Is it
x+iy
normalizable on the sphere?
I will mention some important results - you should try to understand these and the ones in the
Wikipedia references in the language of QM (using bra ket notation)

• Orthogonality:
Z
∗
dθ sin θ dϕ Ylm (θ, ϕ) Yp,q (θ, ϕ) = δlp δmq N (7.41)

• Completeness:
X
∗
Ylm (θ, ϕ)Ylm (θ′ , ϕ′ ) = δ(cos θ − cos θ′ )δ(ϕ − ϕ′ ) (7.42)
lm

• Orthogonality
Z 1
dxPlm (x)Ppm (x) = δpl N (7.43)
−1

• Rodrigues formula or ladder relation

• Generating function

• This is a very important formula

1 X Pl (cos γ)
= (7.44)
|x1 − x| |x1 |l+1
l

where γ is the angle between the two vectors and we have assumed |x| < |x|1 . The more general
expression is give in Jackson, but can be derived with little effort.

•
m dm
Plm = (−1)m (1 − x2 ) 2 Pl (x) (7.45)
dxm

• In spherical harmonics point of view, m varies from −l to l in steps of 1. Is this natural? What
does m count?

• Parity

Ylm (θ, ϕ) = (−1)l Ylm (cos(π − θ), ϕ + π) (7.46)

7.6. ANGULAR MOMENTUM 103

Exercise 187
Normalize the spherical harmonics so that Y ∗ Y = 4π
R

First, we construct the highest weight Lz ψ = lψ eigenfunction ψ = sinl θeilϕ and normal-
Rπ √ Γ(l+1) 2l
(l!)2
ize them. 2π|N |2 0 sin2l θ sin θdθ = 4π The integral evaluates to π Γ( 3
+l)
= 2(2l+1)! . Thus,
q 2
(2l+1)! l
⟨θ, ϕ||l, l⟩ = 4π22l (l!)2
sin θeilϕ upto an overall phase factor (which is however quite impor-
tant). Then, we use the recursion relation to normalize the others in analogy with the SHO.
This is done as follows.

c2 ⟨l, m+1||l, m+1⟩ = ⟨l, m|L− L+ |l, m⟩ = ⟨l, m|(L2 −L2z −Lz )|l, m⟩ = (l(l+1)−m(m+1))⟨l, m||l, m⟩
(7.47)
Hence if |l, m⟩ is properly normalized, then
1
|l, m + 1⟩ = p L+ |l, m⟩ (7.48)
(l − m)(l + m + 1)

upto the phase factor which cannot be fixed thus.

We can thus compute the matrix elements:

Z p
Yln L± Ylm = δn±1,m = l(l + 1) − m(m ± 1) (7.49)

The upshot of this is the following: we have (2l + 1) dimensional matrices Lx,y,z which satisfy the
commutation relations of the angular momentum algebra.
If we now consider the (2j+1)-dimensional matrix

R = exp i(αLx + βLy + γLz ) (7.50)

it acts as a rotation on the vectors ψ which are linear combination of the spherical harmonics
X
am Ylm (θ, ϕ) (7.51)
m

all with fixed l.

a′j
   
aj
a′j−1  aj−1 
 ..  = R  ..  (7.52)
   

a′−j a−j
The meaning of “acts as a rotation” is the following:
if we consider two 3-D rotations (ie 3d matrices), R(α, β, γ) and R(α′ , β ′ , γ ′ ) - corresponding toeach
of these we have (2l+1) dimensional matrices

exp i(αLx + βLy + γLz ), exp i(α′ Lx + β ′ Ly + γ ′ Lz ) (7.53)

which satisfy the same group multiplication.

Since the product of two rotations
′′ ′′ ′′
R(α, β, γ)R(α′ , β ′ , γ ′ ) = R(α , β , γ ) (7.54)
′′ ′′ ′′
is also a rotation. Remember, α , β , γ are functions of the other angles.
104 CHAPTER 7. SYMMETRY

The magical thing is that

′′ ′′ ′′
exp{i(αLx + βLy + γLz )} exp{i(α′ Lx + β ′ Ly + γ ′ Lz )} = exp{i(α Lx + β Ly + γ Lz )} (7.55)
′′ ′′ ′′
as products of 2l + 1 dimensional matrices. That is, we get the same functions α , β , γ either by 3D
multiplication or by the above 2l + 1 matrix product.
And this occurs for every integer l.
We say that we have a representation of the SO(3) group for every l. For every group element α, β, γ,
we have a corresponding 2l + 1 matrix R(α, β, γ). This correspondence is such that group composition
is naturally matrix multiplication. Such a homomorphism is called a (2l + 1)−representation of the
rotation group SO(3).

Exercise 188
At this stage, it is possible to imagine a vast generalization.
What are all possible differentiable functions f (⃗x)) which satisfy group composition laws as
above. This is the subject of Lie theory.
The remarkable answer is that this can be fully solved. See Chevalley for the details.

7.7 Spin
In the preceding section, we saw that the rotation group has a representation in every odd integer-
dimension in terms of wavefunctions.
In this section, we will see a unique even dimensional representation of SO(3) which cannot be
interpreted in terms of orbital angular momentum i.e, wavefunctions.
Consider the Pauli matrices σi which are a set that satisfy

{σi , σj } = 2δij (7.56)

They are all traceless and therefore have eigenvalues ±1. Therefore, such matrices have to been even
dimensional (i.e., size of the matrix is an even integer). We can write them down explicitly for the 2d
case, when they are called Pauli matrices
     
0 1 0 −i 1 0
σ1 = σ2 = σ3 = (7.57)
1 0 i 0 0 −1

Using these, we can find such matrices in 4D as well- when the Kronecker delta is replaced by ηij
the Minkowski metric, they are then called Dirac matrices.
Using these Pauli matrices, we can form
i i i
S1 = − [σ2 , σ3 ] S2 = − [σ3 , σ1 ] S3 = − [σ1 , σ2 ] (7.58)
4 4 4
It is easy to check that these 2-d matrices also satisfy the angular momentum commutation relations.

Exercise 189
Compute Si . Verify that they satisfy the commutation relations.

More generally, Sij = − 4i [γi , γj ] satisfy the “rotation” algebra associated to the quadratic form ηij
for any set of matrices γi forming a Clifford algebra {γi , γj } = 2ηij .
7.8. TOTAL ANGULAR MOMENTUM 105

Thus, if we consider the 2d matrices

exp{i(αS1 + βS2 + γS3 )} (7.59)

these will also form a representation of the rotation group.

The vector space that these act on is the space of qubits that we discussed in the first part of the
course.

Exercise 190
For these ‘spins’, compute the value of L2 and the possible values of Lz .

7.8 Total Angular Momentum

Define the total angular momentum operator as

Ji = Li + Si (7.60)

where L is the orbital angular momentum Li = x[i, ∂j] and Si = σ2i form the spin 1/2 operators. Define
the operator
J 2 = L2 + S 2 (7.61)
 
1
Now, let us start with ψ = (x + iy)l and determine its properties
0

1
Jz ψ = (l + )ψ (7.62)
2
and apply the lowering/raising operators defined as

J± = Jx ± iJy (7.63)

Note that    
l−1 1 l 0
J− ψ = 2lz(x + iy) + (x + iy) (7.64)
0 1
and further
   
2 2 l−2 2 2 l−2 1 l−1 0
(J− ) ψ = (4l(l − 1)z (x + iy) + 2l(x + y )(x + iy) + 2lz(x + iy) (7.65)
0 1

etc.
On each of these states J 2 gives l(l + 1) + 21 ( 12 + 1). Is it easy to see that we get 2(l + 12 ) + 1 states?
A general state in this vector space will then take the form

l    
X 1 0
|ψ⟩ = cl Ylm + dl Ylm (7.66)
0 1
m=−l

which is a linear combination of the basis vectors. Note that the number of basis vectors in the above
1
is actually (2l + 1)2. But, by applying the lowering operator J− to Yll , we get only 2l + 2 states.
0
106 CHAPTER 7. SYMMETRY

Exercise 191
   
l−1 1 l 0
Can you check that 2lz(x + iy) −(x + iy) is orthogonal to all the states obtained
  0 1
1
by (J− )m (x + iy)l .
0

On such a vector, when we apply a general rotation, we first use Euler’s theorem
⃗
ei⃗α·J |ψ⟩ = e−iJz γ e−iJx β e−iJz α |ψ⟩ (7.67)
and then we apply the individual operators in series
l
X    
1 1 1 0
e−iJz γ e−iJx β e−iJz α |ψ⟩ = e−iJz γ e−iJx β e−iJz α cl ei(m+ 2 )α Ylm + dl ei(m− 2 )α Ylm
0 1
m=−l
(7.68)
etc. It is cumbersome to work out the effect of Jx = 21 (J+ + J− ) on the states - but it is algebraic (for
e.g., we need not worry about powers Jxm with m > 2l + 2). So one can write a computer program.

Exercise 192
A good exercise is provided by l = 1.

Exercise 193
Compute the probability current for the spherical harmonics.

7.9 notes
Andrew Winkler, “Can compact groups have non-unitary representations?”
There are three things that can “go wrong”. Maybe the vector space isn’t complex. Maybe
it’s complex, but has no sesquilinear positive form. Maybe it’s a complex Hilbert space, but the
representation still isn’t unitary.
A compact group acting as the identity map on a real vector space of odd dimension can’t possibly
be unitary, because an odd dimensional real vector space can’t carry a complex structure, and so has
no unitary maps.
Consider now the integrable complex valued functions on the real numbers. Consider the unitary
group U(1) of complex numbers of modulus 1. It can be represented on integrable functions by
multiplication. But it’s not a unitary representation, because the integrable functions don’t have a
Hilbert space structure.
In fact, if you have any unitary representation U(g) of a group, by conjugating by a non-unitary
non-intertwining map, you get a representation that is not unitary.
However, if a compact group has a representation in a real or complex Hilbert space, you can
define a new Hilbert space structure by integrating over the group, and for this new structure, the
representation is orthogonal or unitary, respectively.
There are three things that can “go wrong”. Maybe the vector space isn’t complex. Maybe
it’s complex, but has no sesquilinear positive form. Maybe it’s a complex Hilbert space, but the
representation still isn’t unitary.
7.9. NOTES 107

A compact group acting as the identity map on a real vector space of odd dimension can’t possibly
be unitary, because an odd dimensional real vector space can’t carry a complex structure, and so has
no unitary maps.
Consider now the integrable complex valued functions on the real numbers. Consider the unitary
group U(1) of complex numbers of modulus 1. It can be represented on integrable functions by
multiplication. But it’s not a unitary representation, because the integrable functions don’t have a
Hilbert space structure.
In fact, if you have any unitary representation U(g) of a group, by conjugating by a non-unitary
non-intertwining map, you get a representation that is not unitary.
However, if a compact group has a representation in a real or complex Hilbert space, you can
define a new Hilbert space structure by integrating over the group, and for this new structure, the
representation is orthogonal or unitary, respectively.

7.9.1 What is an example of a representation of U(1) that is not unitary?

Keith Ramsay,
There is a kind of uninteresting way that a representation can be made non-unitary, by conjugating
a unitary representation by a suitable non-unitary transform. In this case, if we express U(1) as angles
θ but identifying θ with θ + 2π for each θ, there is a familiar unitary representation,

T (θ)(x, y) = (x cos(θ) − y sin(θ), x sin(θ) + y cos(θ))

. We can conjugate by a very simple transformation to get the representation,

T ′ (θ)(x, y) = (x cos(θ) − y sin(θ)/2, 2x sin(θ) + y cos(θ)).

This representation is not linear because it includes transformations like (x, y) → (−y/2, 2x) that do
not preserve the norm of the vector they act on (in other words, are not unitary).
There is a standard trick that I heard referred to as the “unitary trick” (Wikipedia seems to prefer
to call it the “unitarian trick”), where you take a representation of a group and produce a new norm
for the space it acts on, such that the representation is now unitary with respect to the new norm. I
don’t remember the precise conditions required to make it work, but the example I just gave is a case
where it does work. This is why I say it is not unitary in sort of an uninteresting way. I assume first
that the norm of (x, y) is the usual one, and describe for you a non-unitary representation. But the
representation then asks me, wouldn’t I rather use one of the norms in which it is unitary? Why am
I tripping it up by using another one? It can provide me with a better one if I use the unitary trick.
Then the vector space says, why did I use the basis I used, instead of an orthonormal basis? Wouldn’t
it be nicer to look at it that way? So although the representation is not unitary, it is equivalent, as a
representation on a vector space without a norm on it, to a representation that is unitary. It is only
if I need to keep the same norm for some reason that there is a concern.
To get the unitary trick to work, one needs however to be able to do an integral involving the
representation, and as far as I know this is only helpful in cases where the representation is at least
measurable. In your case, if we don’t require continuity or measurability of the representation, we can
prove the existence of representations that are not unitary in a more interesting way (depending on
how interesting you find “pathology”).
Start by taking a Hamel or “algebraic” basis B for the reals R as a vector space over the rationals Q
that includes 2π as one of its elements. This is what requires the axiom of choice. What is means for B
to be an algebraic basis is that each real can be written in a unique way as a finite linear combination
of elements of B with rational coefficients. Take any function g : B → C that satisfies g(2π) = 0 (for
instance, let g(u) = 2 unless u = 2π) and is not identically 0. Let our new representation h have the
value h(ru)(z) = zerg(u) on the rational multiples of elements of B, and using the fact that h needs to
be a homomorphism to extend h to all reals, i.e. h(r1 u1 + ... + rk uk )(z) = zer1 g(u1 )+...+rk g(uk ) . This
is a representation on U(1) because I made sure that h(2π) is the identity.
108 CHAPTER 7. SYMMETRY

We can tell that h is discontinuous because h(r) is the identity for a dense set of values r, the
rational multiples of 2π, but h is not identically equal to the identity. It’s possible to show that h is
much worse than just discontinuous; it’s not even a measurable function.
If I choose all the values of g on the basis to be imaginary numbers, I get a unitary representation
anyway. But now there’s nothing standing in the way of my letting some of the values of g be off of
the imaginary axis, and if I do, I get an h which isn’t unitary.
Of course I haven’t actually given you an example; I’ve just indicated how the axiom of choice
implies that there is an example. It’s consistent with the ZFC axioms of set theory that an explicit
example can be given, for example if V=L (meaning that the universe of sets is Goedel’s constructible
universe L) because in L there is a formula that chooses a “first” element out of each nonempty set.
In Goedel’s L we can give you an explicit Hamel basis (each time the proof asks us to choose a new
element, take the one that comes first in Goedel’s ordering of L
) and given that, we can exhibit tons of examples. I have the impression however that no such
assumption is anywhere near being accepted as “true” or appropriate for general use. So we basically
can’t give you a specific example; at least we can’t give you an example and also prove that it is an
example (using only axioms that are widely accepted).
When people talk about representations they almost always are talking about continuous ones (or
at least measurable ones!), preferably ones that can be exhibited explicitly (without using questionable
axioms).
https://scholar.harvard.edu/files/noahmiller/files/representation-theory-quantum.pdf
Chapter 8

QM in 3D

109
110 CHAPTER 8. QM IN 3D

8.1 Potentials
2
p
H = 2M + V (r)
We can use the commuting of L2 and Lz to write the energy eigenstate as

ψE = uE (r)Ylm (8.1)

where  
1 ∂ 2 ∂ l(l + 1) 2mE
(r uE ) − + V (r) uE = − 2 uE = −k 2 uE (8.2)
r2 ∂r ∂r r 2 ℏ
Observe how the angular momentum l(l + 1) leads to a repulsive l(l+1) r2 potential. This is a centrifugal
force keeping the particle away from the origin.
It is not always that we obtain central potentials. For e.g., if we add electric and magnetic fields,
a direction is selected which breaks rotation symmetry. In atomic physics, for e.g., we have the Jahn-
Teller distortions and crystal field effects. We also have the so-called spin-orbit interactions which lead
to reduction in the symmetry.
⃗ ·S
VLS = f (r) L ⃗ (8.3)

Exercise 194
Verify that while [VLS , Li ] ̸= 0 and [VLS , Si ] ̸= 0 nevertheless

[VLS , Ji ] = 0 (8.4)

Why did this happen?

In nuclear physics, a major part of the interaction between two nucleons is the spin-spin interactions

Vs = v(r) S⃗1 · S⃗2 (8.5)

which depend on the spin states of the individual nucleons. This is relatively less important in atomic
physics although all the interesting magnetic and some interesting electromagnetic properties of ma-
terials originate from the spin interactions.

8.2 Free particle

Let us first study the free particle V (r) = 0. This gives the ODE (which is a big improvement)
′′
r2 uE + 2ru′E − (l(l + 1) + k 2 r2 )uE = 0 (8.6)

For large r, we can keep the first and last terms only, to get uE ∼ eikr . For small r,

u′′ + u′ /r − l(l + 1)u/r2 = 0 (8.7)

which is suggestive of u ∼ rl+1 . These two suggest that we should again expect a Bessel type character.
This suggests that we substitute
u(r)
J(r) = √ (8.8)
r
√
because the Bessel part J will contribute another 1/ r at long distances.
Finally, introduce a dimensionless variable x = kr to get the Bessel equation
1
x2 J ′′ − 2xJ ′ + (x2 − (l + )2 )J = 0 (8.9)
2
8.3. INFINITE POTENTIAL WELL 3D 111

and the two independent solutions of this equation are Jl+ 12 (kr) and Yl+ 21 (kr). But the latter blow up
at the origin.
So, the δ−function normalizable states of the free particle in 3D are of the form

ψ = Jl (kr)Ylm (θϕ) (8.10)

2 2
k
upto a normalization factor. These correspond to energies E = ℏ2M and are infinitely degenerate -
since l = 0, 1, 2, 3... and each l−level is (2l + 1) degenerate.
There is a Rodrigues relation for the spherical bessel functions - why?

8.3 Infinite Potential well 3D

In this case, we have a potential V (r) which restricts the particle to a finite sphere around the origin.

V (r) = 0 0≤r≤R (8.11)

= ∞ r>R (8.12)

Then, the same wavefunctions as above must be now supplemented by a vanishing boundary condition
at the r = R surface. That is to say,
Jl (kR) = 0 (8.13)
The Bessel functions Jl (x) have infinitely roots at some values of x which can be denoted xln . Then
the only energy levels are
ℏ2 x2ln
E= (8.14)
2M R2
If the potential well is a finite potential well, then we will have two cases of bound and scattering type
states. The bound states will occur only if the potential is sufficiently deep unlike in the 1D case. See
Ballentine Chapter 10 for a succinct presentation.

8.4 3D SHO
In this case,
1
V (r) = mω 2 r2 (8.15)
2
Solving in Cartesian coordinates we get
3
E = (nx + ny + nz + )ℏω (8.16)
2
and the degeneracy is the number of ways of partitioning an integer N into three integers ni such that
nx + ny + nz = N.
It isqinteresting to repeat the problem in polar coordinates. In this case, the natural length scale
ℏ
is α = mω so that we must define x = r/α. This gives

2 ′ l(l + 1) 2E
u′′ + u − u − x2 u = − u (8.17)
x x2 ℏω
For large x,
u′′ − x2 u ∼ 0 (8.18)
2
suggests that u−x /2 independent of the energy but the small x is still u ∼ xl+1 .
From this we can develop a series for H
2
u = xl He−x /2
(8.19)
112 CHAPTER 8. QM IN 3D

and analyse the series solution for H to get

3
E = (n + l + )ℏω (8.20)
2
Note that the energy depends on the orbital angular momentum l. Rotational symmetry implies that
if we have a state and we rotate it, the energy will not change. This will give us (2l+1) degeneracy
and no more.
But, we see that only the combination n + l appears which indicates more degeneracy than 2l + 1.
The explanation for this is that there is a SU (3) dynamical symmetry of the 3D SHO which is explained
in Mukunda et al.
An amusing question is whether there are ladder operators in spherical polar coordinates also.

8.5 Hydrogen Atom

Again, we may refer to Griffiths and Ballentine for some details.
The new thing about this potential is that it is a singular potential

e2
V (r) = − (8.21)
r
Q2e
where e2 = 4πε0 = [E][L]. This represents the relative potential between an electron and a proton in
ℏ2
a nucleus such as the Hydrogen atom. We note that m has dimensions EL2 . This defines an energy
scale R ∼ me4 /2ℏ2 ∼ 13.6eV
In this case, the radial equation becomes
2 ′ l(l + 1) 1 E
u′′ + u − u+ u=− u (8.22)
x x2 x R
2
ℏ
and the dimensionless variable x = r/a0 where a0 = me 2 is called the Bohr radius. Note that in this

case both E > 0 scattering states and E < 0 bound states are possible.
https://farside.ph.utexas.edu/teaching/qm/Quantum/node44.html
√E
For bound states at large r, u ∼ rn e−x R which depends on the energy E. For short distances, rl
behaviour is perhaps clear.
The bound state wavefunctions turn out to have the form

ψ = Lnl e−nr/a0 Ylm (8.23)

where the Lnl are Laguerre polynomials as you might expect (if they were not polynomials, they will
overcome the e−nr/a0 falloff).
These Lnl have increasing numbers of nodes as you might expect.
The 0 ≤ l < n range for l arises from the normalizability condition of the radial wavefunction.
The bound states have energies En = − 13.6eVn2 and a degeneracy of n2 . It is customary to include
the spin of the electron in this discussion which increases the degeneracy by a factor 2.
This degeneracy is much more than the expected 2l + 1 which arises from rotational symmetry.
Again, we have a dynamical SO(4) symmetry which is generated by the Laplace Runge vector together
with the rotations. This explains the degeneracy fully.
For the scattering states, the dynamical
√E symmetry is the Lorentz group O(3, 1). The large r be-
haviour will be at large r, u ∼ rn eix R .
https://www.mdpi.com/2073-8994/12/8/1323
Is zero a hydrogen eigenvalue?
Scattering states of Hydrogen atom
Can a normalizable function *always* be decompose into the discrete Hydrogen spectrum?
Chapter 9

Additional reading

Here I will collect some articles which can be of additional interest to motivated people
one [7]

113
Index

Angular momentum, 99 Pauli matrices, 36

Angular momentum Algebra, 36 Positive operators, 27
Average, 59 Projection operators, 27
Pure state, 28
Basis, 15
Bloch’s theorem, 90 qubit, 29
Bounded operators, 27 qutrit, 36
Brillouin zone, 90
rank, 100
Cauchy-Schwarz Inequality, 22 representation, 104
central potential, 98
Clifford Algebra, 36, 104 Spectral decomposition, 26
Component, 19 Spectral theorem, 25
Spherical harmonics, 100
Density Matrix, 28 Spin, 104
Dimension, 15 Stationary states, 54
Dirac delta, 59
Tensor Product, 35
Direct Sum, 34
Triangle Inequality, 22
Domain, 23
Dual space, 17 Uncertainty principles, 33
Eigenvalue, 24 Vector Spaces, 15
Eigenvector, 24 von Neumann entropy, 31
Expectation, 59
Weyl ordering, 71
Gram-Schmidt Orthonormalization, 22

Hermitean Operator, 26
Hilbert Space, 21

Infinite Vector Spaces, 16

Inner Product, 18

Linear Independence, 15
Linear operators, 23

matrix, 23
Mixed State, 28

Norm, 20
Normal operators, 25

Orthogonal vectors, 19
Orthonormal basis, 21

114