Quantum Mechanics Lecture Notes.

Selected Chapters
S. Levit
Department of Condensed Matter Physics
Weizmann Institute of Science, Rehovot, Israel
[Link]@[Link]
April 11, 2023

Abstract
These are extended lecture notes of the quantum mechanics course which I am
teaching in the Weizmann Institute of Science physics program. They cover the
topics listed below. The first four chapters are posted here. Their content is detailed
on the next page. The other chapters are planned to be added in the coming months.
1. Motion in External Electromagnetic Field. Gauge Fields in Quantum Me-
chanics.
2. Quantum Mechanics of Electromagnetic Field
3. Photon-Matter Interactions
4. Quantization of the Schrödinger Field (The Second Quantization)
5. Open Systems. Density Matrix
6. Adiabatic Theory. The Berry Phase. The Born - Oppenheimer Approximation
7. Mean Field Approaches for Many Body Systems - Fermions and Bosons

1
Contents

1 Motion in External Electromagnetic Field. Gauge Fields in Quantum

Mechanics 5
1.1 Electromagnetic Potentials. The Hamiltonian . . . . . . . . . . . . . . . . 5
1.1.1 Electromagnetic potentials in classical physics . . . . . . . . . . . 5
1.1.2 Classical Hamiltonian and equations of motion . . . . . . . . . . . 7
1.2 Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 The orbital part . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.2 The spin magnetic moment . . . . . . . . . . . . . . . . . . . . . . 8
1.2.3 The Schrödinger equaion . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Gauge Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.1 Gauge transformations in quantum mechanics . . . . . . . . . . . . 10
1.3.2 Gauge symmetry vs gauge invariance . . . . . . . . . . . . . . . . . 11
1.3.3 The Gauge Principle – symmetry dictates
interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Electric Current Density. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.1 The orbital part . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.2 The spin contribution . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.3 Convective, diamagnetic and spin parts of the current . . . . . . . 15
1.5 Motion in a Uniform Electric Field . . . . . . . . . . . . . . . . . . . . . . 15
1.5.1 Static gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.5.2 Linear potential - the Airy function . . . . . . . . . . . . . . . . . 16
1.5.3 Time dependent gauge . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5.4 Translations in uniform E. Symmetries in the presence of gauge
fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.6 Motion in a Uniform Magnetic Field . . . . . . . . . . . . . . . . . . . . . 20
1.6.1 Classical motion. The guiding centers . . . . . . . . . . . . . . . . 20
1.6.2 Landau levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.6.3 Degeneracy of Landau levels and space symmetries . . . . . . . . 31
1.7 The Aharonov - Bohm Effect . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.7.1 Local and non local gauge invariant quantities
H . . . . . . . . . . . 32
1.7.2 Quantum mechanics ”feels” non zero C A · dr even if E = B = 0
on and near the contour C . . . . . . . . . . . . . . . . . . . . . . . 33
1.7.3 ”Gauging out” the AB flux. Periodic dependence on its value . . . 34
1.7.4 Example of the AB flux . . . . . . . . . . . . . . . . . . . . . . . . 35

2
Version of April 11, 2023 3

1.7.5 Multiply connected regions. Homotopy . . . . . . . . . . . . . . . 41

1.8 Magnetic Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.8.1 The g–factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.8.2 Atoms in a magnetic field . . . . . . . . . . . . . . . . . . . . . . . 43
1.8.3 The Zeeman effect . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.9 Time Reversal in Magnetic Field. Kramers Degeneracy . . . . . . . . . . . 47
1.10 Path Integrals with the External Electromagnetic Field . . . . . . . . . . 49
1.11 Dirac Magnetic Monopoles . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
1.11.1 Multivalued wave functions. Non integrable phases . . . . . . . . 51
1.11.2 Magnetic monopoles . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1.11.3 Angular momentum and rotational symmetry in the presence of a
monopole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
1.12 Non Abelian Gauge Fields . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2 Quantum Mechanics of Electromagnetic Field 64

2.1 Simple System First - Quantum Mechanics of a Guitar String . . . . . . . 64
2.1.1 Classical string . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.1.2 Quantum description of the string . . . . . . . . . . . . . . . . . . 66
2.1.3 Reminder - normal modes of vibrations . . . . . . . . . . . . . . . 69
2.1.4 String as a collection of decoupled oscillators . . . . . . . . . . . . 70
2.2 Quantization of Traveling Waves. Closed String . . . . . . . . . . . . . . . 75
2.2.1 Expansion in traveling waves . . . . . . . . . . . . . . . . . . . . . 76
2.2.2 Quantum mechanics of the traveling waves . . . . . . . . . . . . . 80
2.3 Quantization of the EM Field . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.3.1 Hamilton form of the Maxwell equations . . . . . . . . . . . . . . . 83
2.3.2 Canonical quantization . . . . . . . . . . . . . . . . . . . . . . . . 86
2.3.3 Gauge invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.4 Photons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.4.1 Field oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.4.2 Photon momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
2.4.3 Common states of light . . . . . . . . . . . . . . . . . . . . . . . . 100
2.4.4 Photon angular momentum and spin . . . . . . . . . . . . . . . . . 105
2.4.5 Photon parity and photon statistics . . . . . . . . . . . . . . . . . 107
2.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
2.5.1 Details of the standing to traveling waves transformation . . . . . 108
2.5.2 Details of the coherent states . . . . . . . . . . . . . . . . . . . . . 109
2.5.3 More on the EM field momentum . . . . . . . . . . . . . . . . . . 112
2.5.4 More on the EM field angular momentum . . . . . . . . . . . . . . 113

3 Photon-Matter Interactions 116

3.1 Interaction Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.1.1 Separating the interaction terms . . . . . . . . . . . . . . . . . . . 116
3.1.2 Adding spin and external fields . . . . . . . . . . . . . . . . . . . . 117
3.1.3 Disentangling radiation from the matter degrees of freedom . . . . 118
3.1.4 Resulting insights . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3.2 Emission and Absorption of Photons . . . . . . . . . . . . . . . . . . . . . 122
Version of April 11, 2023 4

3.2.1 Paradigm of spontaneous emission of radiation - discrete matter

level coupled to a photon continuum . . . . . . . . . . . . . . . . . 123
3.2.2 Photon emission rate . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.2.3 Long wavelength approximation . . . . . . . . . . . . . . . . . . . 133
3.2.4 Electric dipole emission . . . . . . . . . . . . . . . . . . . . . . . . 134
3.2.5 Angular momentum and parity selection rules . . . . . . . . . . . . 136
3.2.6 ”Forbidden” (higher multipole) transitions . . . . . . . . . . . . . . 145
3.2.7 Induced photon emission . . . . . . . . . . . . . . . . . . . . . . . . 155
3.2.8 Photon absorption . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
3.3 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
3.3.1 Discrete level coupled to continuum . . . . . . . . . . . . . . . . . 157
3.3.2 The ĤI3 part of the Hamiltonian and the parity transformation . 160

4 Quantization of the Schrödinger Field - The Second Quantization 162

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.2 Free Schrödinger field. Quantization . . . . . . . . . . . . . . . . . . . . . 163
4.2.1 Separating the real and imaginary parts . . . . . . . . . . . . . . . 163
4.2.2 Back to the complex valued field . . . . . . . . . . . . . . . . . . . 165
4.2.3 The Hamiltonian of the free Schrödinger field . . . . . . . . . . . 166
4.2.4 The eigenstates. Field quanta are free non relativistic particles . . 166
4.2.5 Momentum and angular momentum . . . . . . . . . . . . . . . . . 169
4.3 Adding external potential . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4.3.1 The Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4.3.2 The eigenstates. Field quanta are particles in the external potential 173
4.3.3 Working with the field operators . . . . . . . . . . . . . . . . . . . 177
4.4 Wave functions. Operators. Comparison with the first quantized description179
4.4.1 Wave functions in the second quantization . . . . . . . . . . . . . . 179
4.4.2 Operators in the second quantization . . . . . . . . . . . . . . . . . 182
4.4.3 Second quantization via commutators describes identical bosons . 186
4.4.4 Self interacting Schrödinger field. . . . . . . . . . . . . . . . . . . . 187
4.5 Fermions – another alternative of the second quantization . . . . . . . . . 190
4.5.1 Quantization via anticommutators . . . . . . . . . . . . . . . . . . 190
4.5.2 Fermions in external potential . . . . . . . . . . . . . . . . . . . . . 191
4.5.3 Relation to the first quantization . . . . . . . . . . . . . . . . . . . 197
4.5.4 Interacting fermions . . . . . . . . . . . . . . . . . . . . . . . . . . 199
4.6 The Fock space. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
4.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
4.7.1 Bosons - reviewing the properties of â’ s and â+ ’s . . . . . . . . . 202
4.7.2 Bosons - wave function normalization . . . . . . . . . . . . . . . . 204
4.7.3 Bosons - calculating Kop |Φ⟩ . . . . . . . . . . . . . . . . . . . . . 205
4.7.4 Fermions - calculating Fop |Φ⟩ . . . . . . . . . . . . . . . . . . . . 206
Chapter 1

Motion in External
Electromagnetic Field. Gauge
Fields in Quantum Mechanics

Electromagnetic potentials A(r, t) and A0 (r, t) appear in classical physics as auxiliary

quantities which are introduced in order to simplify the form and solutions of the Maxwell
equations, cf., Chapter 10 in Ref. [8]. The fact that they are not uniquely defined and can
be changed without affecting any physical results by a transformation bearing a strange
name of ”gauge” seems to be rather an annoying nuisance than a fundamental symmetry
of nature.
This state of affairs undergoes drastic revision when quantum mechanical description
is attempted. We do not know how to formulate such a description in the presence of the
electromagnetic field without making an essential use of the electromagnetic potentials.
Moreover the invariance under the gauge transformations becomes a profound symmetry
of our world which lies at the origin of all known interactions. Because of this the fields
which carry these interactions are termed gauge fields.
The problem of the quantum mechanical motion in an external electromagnetic field
provides the simplest setup in which one encounters some of the strange and beautiful
phenomena appearing as a result of the symbiosis of gauge fields and quantum mechanics.
Note. I have changed from CGS to SI units in Sections 1-8. The rest is in
CGS.

1.1 Electromagnetic Potentials. The Hamiltonian

1.1.1 Electromagnetic potentials in classical physics
Let us begin by briefly recalling how the electromagnetic potentials are introduced. Clas-
sical electromagnetic field is described by two vector fields E(r,t) and B(r,t). In the
present chapter these fields will be considered as external, i.e. produced by sources (elec-

5
Version of April 11, 2023 6

tric charges and currents) which dynamically are not a part of the physical system under
consideration and are not effected by it. This means that the back reaction of the system
on the sources of the field is negligible. Although in such circumstances E and B should
be regarded as controlled externally by charge and current distributions ρ(r, t) and j(r, t)
of the sources via
ρ 1 ∂E
∇·E= , ∇ × B = µ0 j + 2
ϵ0 c ∂t
they can not be taken as completely arbitrary. Indeed irrespective of the configuration of
ρ and j these fields must satisfy the homogeneous pair of Maxwell equations
∂B
∇·B=0 , ∇×E=− (1.1)
∂t
at every point in space and time. In order to have these equations automatically satisfied
the familiar vector and scalar potentials A(r,t) and A0 (r,t) are introduced1 . This is done
by noticing that the first of the equations above means that B must be a curl of a vector
field A(r, t). Using this in the second equation gives
∂A
∇ × (E + )=0
∂t
restricting the combination E+∂A/∂t to be a gradient of a scalar field. One has therefore
∂A
E = − − ∇A0 ,
∂t
B = ∇×A (1.2)
Unlike the field strengths E and B, the electromagnetic potentials can be regarded as
unrestricted so that any A(r, t) and A0 (r, t) can be realised by the poper choice of the
external charge and current distributions.
The use of the electromagnetic potentials however presents another problem. They
are not unique since the gauge transformation
A′ (r,t) = A(r,t) + ∇χ(r,t)
∂χ(r,t)
A0′ (r,t) = A0 (r,t) − (1.3)
∂t
with an arbitrary function χ(r,t) leaves E and B invariant. As was already mentioned
above this invariance, called the gauge invariance, has profound consequences in quantum
mechanical systems and will be discussed at length below. At the moment we just notice
that because of it only three among the four functions A and A0 are independent. In
general one combination of the four functions can be eliminated by a suitably chosen
gauge transformation. For instance choosing
Z t
χ(r,t) = A0 (r, t′ ) dt′
t0

(with arbitrary t0 ) eliminates A0 and leaves A(r,t) as the only independent degrees of
freedom of the electromagnetic field.
1 Although we use ”relativistic” notation for A we use ”non relativistic” terminology and call it a
0
scalar potential
Version of April 11, 2023 7

1.1.2 Classical Hamiltonian and equations of motion

Classical non relativistic equation of motion for a particle with electric charge q and mass
m in a given electromagnetic field is obtained by using the Lorenz force in the Newton
law
d2 r
 
dr
m 2 = qE + q ×B . (1.4)
dt dt
In order to obtain the quantum mechanical description one can follow either the canonical
or the path integral quantization procedures. We will start with the former. We first
determine the classical canonical variables and the classical Hamiltonian function of the
problem.
The above equation is in terms of coordinates r(t) and velocities v(t) = dr/dt so it is
is most convenient to start by determining the Lagrangian of the system. This is
1
L(r, v, t) = mv2 + qA · v − qA0 (r) . (1.5)
2
Indeed have
d ∂L dv dA dv ∂A drj ∂A
=m +q =m +q +q
dt ∂v dt dt dt ∂rj dt ∂t
and
∂L ∂ ∂A0
= q (A · v) − q
∂r ∂r ∂r
In components
dvi ∂Ai ∂Ai ∂Aj ∂A0
m +q vj + q =q vj − q
dt ∂rj ∂t ∂ri ∂ri
So have    
dvi ∂Ai ∂A0 ∂Aj ∂Ai
m =q − − +q − vj
dt ∂t ∂ri ∂ri ∂rj
which is the Newton equation (1.4). Indeed recalling Eq.(1.2) one sees that the fist term
is qE, while the last term can be transformed as

ϵijk vj Bk = (v × B)i

where we used the antisymmetric symbol ϵijk 1 to write vector products, e.g

(C × D)i = ϵijk Cj Dk , Ci Dj − Cj Di = ϵijk (C × D)k .

These two equalities are related by a useful identity

ϵijk ϵij ′ k′ = δjj ′ δkk′ − δjk′ δj ′ k .

The above calculations show that the canonical momentum is

∂L dr
p= = m + qA(r) . (1.6)
∂v dt
1 The Levi–Civita symbol ϵ
ijk is defined by ϵ123 = 1 and the antisymmetry property under interchange
of any indices, ϵijk = −ϵjik = −ϵikj , etc. ϵijk does not change under cyclic permutations ϵijk = ϵkij = ....
.
Version of April 11, 2023 8

which expresses perhaps the most unusual aspect of the motion in the EM field - the fact
that p ̸= mv. In the literature one often meets the term ”kinetic momentum” referring
to the familiar mv.
Expressing v(p, r) = (p − qA(r))/m and using

H =p·v−L

with the above v(p, r) we find the Hamiltonian function

1 2
H(p, r) = (p − qA(r,t)) + qA0 (r,t) , (1.7)
2m
It is not difficult (and not surprising) to show that with this H(p, r) the equation of
motion (1.4) is equivalent to the two Hamilton equations
dr ∂H dp ∂H
= , =−
dt ∂p dt ∂r

1.2 Quantization
1.2.1 The orbital part
Having established the form of H we follow the canonical quantization procedure and
consider the Schrödinger equation with the Hamiltonian operator which is obtained by
replacing r and p in H by the operators rop = r and pop = −iℏ∇,
1 2
Hop = [−iℏ∇ − qA(r, t)] + qA0 (r, t) . (1.8)
2m

1.2.2 The spin magnetic moment

Experimental evidence shows that this Hamiltonian is capable of describing only particles
which do not carry spin. It must be modified when the spin degrees of freedom are
present. This should not be too surprising since already in classical physics the energy
of a spinning charged particle receives an additional contribution apart from the orbital
motion. This contribution arises from the interaction with the magnetic field B of a
localized distribution of electric current j(r) which a spinning charge creates. For a ”point
like” particle, i.e. a particle the size of which is much smaller than the scale over which
B(r) changes, the corresponding energy is

Espinning charge = −µ · B(r)

where µ is the magnetic moment of the current, cf., Chapter 5 of the Ref. [8],
1
Z
µ= d3 r r × j(r) .
2
For composite particles the total current is a sum over internal components
X
j(r) = qa va δ(r − ra )
a
Version of April 11, 2023 9

each with its dynamics so the calculation of µ is in general not an easy task. However if
all the components have an equal charge to mass ratio q1 /m1 = q2 /m2 = ... = q/m the
magnetic moment can be written as
1X q X q
µ= qa (ra × va ) = ra × ma va = L. (1.9)
2 a 2m a 2m

Experimental data as well as theoretical considerations (cf., Section 1.8 below) indicate
that for elementary particles like electrons this classical linear relation between µ and the
angular momentum of a system holds also between the corresponding quantum mechani-
cal operators of the magnetic moment µop and the spin sop . However the proportionality
coefficient in general does not coincide with the classical value. To emphasize this differ-
ence it is conventional (for charged particles) to write the relation between the operators
µ and s as
q
µop = g sop (1.10)
2m
with q - the particle charge and g - dimensionless coefficient called the gyromagnetic factor
or for short the g-factor. Theoretical methods which allow to determine g and examples
of their applications are considered in Section 1.8.

1.2.3 The Schrödinger equaion

Adding the term −µop ·B to the Hamiltonian operator (1.8) one can write the Hamiltonian
for an elementary particle with a spin in an external EM field as
1 2 q
(−iℏ∇ − qA) + qA0 − g
Hop = sop · B . (1.11)
2m 2m
and the corresponding Schrödinger equation
 
∂ψ 1 2 q
iℏ = (−iℏ∇ − qA) + qA0 − g sop · B ψ (1.12)
∂t 2m 2m
where
ψ = ψ(r, σ; t)
is a function of space and spin variables r and σ.
In writing out the square in this equation one should not forget that the operator
pop = −iℏ∇ in general does not commute with the vector A which is a function of
coordinates. Since pop · A − A · pop = −iℏ∇ · A, one can write
1 2 ℏ2 2 iℏq q2 2
(−iℏ∇ − qA) = − ∇ + (∇ · A + 2A · ∇) + A .
2m 2m 2m 2m
The operators pop and A commute if ∇ · A = 0. This happens e.g., for A = (B × r)/2
which is a possible choice of A in a particular case of a uniform magnetic field2 .
2 Verifying

1 1 1
(∇ × A)i = ϵijk ∇j ϵklm Bl xm = (δil δjm − δim δjl )δjm Bl = (3Bi − Bi ) = Bi
2 2 2
∇·A = ∇i Ai = ∇i ϵijk Bj xk = ϵijk Bj δik = 0 (1.13)
Version of April 11, 2023 10

In the following sections we will examine various properties of the equation (1.12) and
will present its solutions for some particular simple choices of the electric and magnetic
fields.

1.3 Gauge Invariance

1.3.1 Gauge transformations in quantum mechanics
The electromagnetic field enters the classical and quantum equations (1.4) and (1.12)
via very different sets of variables. The classical equation depends on the physically
measurable variables E and B of the field whereas in the Schrödinger equation the field
enters via non uniquely defined and seemingly auxiliary objects A and A0 . This is not
an accident. At present no formulation of quantum mechanics exists which does not
explicitly use the electromagnetic potentials. Schrödinger and Heisenberg pictures require
the Hamiltonian while the path integral quantization uses the Lagrangian (cf., below,
Section 1.10) and both objects can not be written without A and A0 . Since the potentials
are not uniquely defined and can be changed by a gauge transformation one must address
the question of how unambiguous physical results are obtained in such a situation.
Unlike in classical mechanics where gauge transformations do not change the equa-
tions of motion the Schrödinger equation (1.12) and therefore also its solutions ψ(r,t) are
transformed in a non trivial way3 . It is not difficult to find how the transformation of
ψ(r,t) is related to the transformation of the potentials. For this we notice that A and
A0 enter the equation only in the combinations
∂
(−iℏ∇ − qA) and (iℏ − qA0 ) .
∂t
Thus if ψ(r,t) is a solution for a particular choice of A and A0 then
h q i
ψ(r,t) = exp −i χ(r,t) ψ ′ (r,t) ≡ S(r,t)ψ ′ (r,t) (1.14)
ℏ
satisfies

≡ (−iℏ∇ − qA) ψ = S(r,t) −iℏ∇ − qA′ ψ ′ = −iℏS(r,t)D′ ψ ′ ,

−iℏDψ
   
∂ ∂
iℏD0 ψ ≡ iℏ − qA0 ψ = S(r,t) iℏ − qA0′ ψ ′ = iℏS(r,t)D0′ ψ ′ (1.15)
∂t ∂t

and therefore solves the Schrödinger equation for the transformed potentials (1.3) (please
note the primed D′ and D0′ on the right hand side of the expressions above).
We see that the classical concept of the gauge transformation undergoes a general-
ization in quantum mechanics. Now not only the potentials which describe the electro-
magnetic field but also the wave functions describing the material particles must change
simultaneously according to the rules (1.3) and (1.14). This change is local, i.e. it is dif-
ferent for different points in space and time. One often emphasizes this aspect by calling
3 In this and many of the following sections the dependence of ψ(r, σ; t) on the spin variable σ will not

be of interest and will be suppressed for brevity.

Version of April 11, 2023 11

the transformation given by Eqs. (1.3), (1.14) a local gauge transformation to distinguish
it from a global transformation in which the wave function is multiplied by a constant
phase factor.
It is seen that the combinations Dψ and D0 ψ defined in (1.15) transform under a
local gauge transformation in a particularly simple way – i.e. as if it were a global
transformation. These combinations are called gauge covariant derivatives in theories with
gauge fields. The way to introduce the electromagnetic field in the dynamical equations
by replacing the ordinary derivatives ∂/∂t and ∂/∂r by the gauge covariant combinations
D0 , D is known as minimal coupling.

1.3.2 Gauge symmetry vs gauge invariance

We may now ask a question as to whether the classical gauge invariance also holds in
quantum mechanics, namely whether the result of any measurement is invariant under
gauge transformations which now include also the local transformation (1.14) of the wave
function. It is an empirical fact that the answer to this question is positive. Moreover
it is also clear that this invariance known as the local gauge invariance is a profound
fundamental symmetry of the quantum mechanical description in the presence of gauge
fields.
It is important to note that this symmetry does not mean that the wave functions
must be invariant. Like with other fundamental symmetries, e.g. the invariance with
respect to translations and rotations, the gauge symmetry means that the wave functions
transform in a particular way given by Eq. (1.14), i.e they form a representation of the
corresponding group of transformations.
Here, however, the similarity ends. Unlike other symmetries the gauge symmetry
demands that the observable quantities must not be effected by the gauge transformations
and therefore must be ”gauge scalars”, i.e. depend on gauge invariant combinations
of ψ, A and A0 . No ”gauge vectors”, ”gauge tensors”, etc, are ever observed. The
origin of this difference can only be understood when the full quantum dynamics of the
electromagnetic field and its coupling to matter are discussed.
We conclude this section by noting that explicit appearance of the electromagnetic
potentials in the equations of quantum mechanics makes the gauge invariance a very subtle
symmetry. Its consequences and generalizations are important aspects of the modern
physics. We will make a special point in this chapter to illustrate some of the related
physical ideas and results.

1.3.3 The Gauge Principle – symmetry dictates

interactions
In the previous section we started with the known transformation properties of the poten-
tials and then on the basis of the special manner in which they entered the Schrödinger
equations – i.e. in combinations D and D0 , – derived the required transformation proper-
ties of the wave functions which were necessary in order to keep the Schrödinger equation
form-invariant.
Imagine now that we reverse this derivation in the following manner. Let us begin by
Version of April 11, 2023 12

considering the free Schrödinger equation

iℏ∂t ψ = −ℏ2 ∇2 ψ/2m .

This equation is obviously invariant under the global gauge transformations i.e. the trans-
formations (1.14) with a constant χ independent of (r,t). This global gauge invariance
is a fundamental feature of the Schrödinger equation. One of its notable consequences is
the conservation of the integral dr ψ ∗ (r,t)ψ(r,t). This integral is the total probability
R

or, when multiplied by e, the total electric charge. The relation of its conservation to the
global gauge invariance is not intuitively obvious but can be rigorously derived by the
applications of arguments of the Noether theorem to the Schrödinger field.
Now let us see what happens if one demands that the nature should be invariant not
only under the global but also under local gauge transformations, i.e. with the (r,t)-
dependent phase χ in Eq. (1.14). It is obvious that the free Schrödinger equation will not
satisfy this demand since its derivatives will act on the local phase producing additional
terms with ∇χ and ∂χ/∂t. With the hindsight of the previous section we can however
write a more general Schrödinger equation which will be locally gauge invariant.
In order to compensate for the derivatives ∇χ and ∂χ/∂t and eliminate them from
the transformed Schrödinger equation we must
(a) ”postulate” the existence of a field described by the potentials A and A0 ,
(b) replace the ordinary derivatives ∂/∂t and ∇ in the equation by the gauge covariant
combinations D0 , D and
(c) require that the potentials transform according to Eq. (1.3) simultaneously with
the transformation (1.14) of the wave functions.
The demand of the local gauge invariance is thus turned into a powerful heuristic
principle – The Gauge Principle, which, had we not known about the electromagnetic
field, led us to ”discover” its existence and the way it must appear in the Schrödinger
equation.
Of course the last, spin-dependent term in (1.12) would not be deduced in such a
procedure and should be justified separately. The need for this separate discussion of the
spin interaction with the electromagnetic field disappears when a fully relativistic theory
of elementary particles is considered, cf. Section 8.1 in Ref. [2] or Chapter 3 in Ref. [3].
Moreover it can be shown that the entire Maxwell electrodynamics is fully consistent with
the The Gauge Principle supplemented by very general requirements of the time-space
translational invariance and the Lorenz invariance.
It also turns out that the fields responsible for all other known interactions, i.e. weak,
strong and gravitational are consistent with The Gauge Principle in a similar way. Namely
for every known interaction there exist a a global symmetry of a non interacting theory
which becomes a local symmetry after the interaction is introduced. The potentials de-
scribing the interaction are the compensating gauge potentials which are necessary to
introduce in order to satisfy this demand are the fields of the fundamental interactions.
Thus The Gauge Principle essentially means that Symmetry Dictates Interactions. The
Gauge Principle for general relativity for example means that the theory is invariant un-
der local Lorenz transformations. In Section 1.12 below we consider an example of how a
so called non abelian gauge field appears as a result of the demand that the Schrödinger
equation is invariant under local non abelian transformations.
Version of April 11, 2023 13

1.4 Electric Current Density.

1.4.1 The orbital part
Let us derive the quantum mechanical expression for the current density of charged
particles. We will start by considering the continuity equation for the charge density
ρ(r,t) = qψ ∗ (r,t)ψ(r,t). Multiplying the Schrödinger equation (1.12) on the left by ψ ∗
and its complex conjugate by ψ and subtracting one obtains in a standard way that
∂ρ/∂t + ∇ · j = 0 with the current density

q
jorbital (r) = [ψ(r)(iℏ∇ − qA(r))ψ ∗ (r) + ψ ∗ (r)(−iℏ∇ − qA(r))ψ(r)] . (1.16)
2m
This expression is the expectation value < ψ|jop (r)|ψ > of the operator
1
jop (r) = [qvop δ(r − rop ) + δ(r − rop ) qvop ] (1.17)
2
of the current density due to orbital motion with the velocity

vop = [pop − qA(rop )]/m .

This operator is just what is obtained from the classical expression ρ(r, t)v(t) = qδ(r −
r(t))v(t) for a point particle by replacing the classical quantities r(t) and v(t) with the
corresponding operators rop and pop and symmetrizing the final expression in order to
make it hermitian.

1.4.2 The spin contribution

The missing feature in the above expression for the current is the absence of the contribu-
tion from the spin of the particle. This is the reason we have added to it the index orbital.
As we have already discussed a spinning charged particle creates a local distribution of
electric current at its location and one should expect to find an appropriate term in the
current density in addition to the contribution of the orbital motion. We have missed this
term because as we will see in a moment it is in the form of a rotor of a vector (a so called
solenoidal term) and therefore can not be seen in the continuity equation which depends
only upon the divergence of the current.
In order to correct our result let us consider a physical system of charges {qa } placed in
positions {ra } and put it under the influence of an external electric field E(r). We start
classically and consider a time interval dt during which these charges move distances
dra = va dt. As a result their total energy is changed by
X Z X dra
dW = qa E(ra ) · dra = dt dr[ qa δ(r − ra )] · E(r)
a a
dt

The expression in the square brackets here is the total current density flowing in the
system, so that
dW
Z
= dr j(r) · E(r) . (1.18)
dt
Version of April 11, 2023 14

We assume that this relation holds also for quantum mechanical expectation values.
Considering for simplicity one particle and let us form the expectation value of the Hamil-
tonian (1.11)
Z
W = drψ ∗ Hψ
 
1 q
Z
∗ 2
= drψ (−iℏ∇ − qA) + qA0 − g sop · B ψ . (1.19)
2m 2m
We also have
dW ∂H
=< ψ| |ψ > . (1.20)
dt ∂t
This relation (sometimes called the Feynman–Hellmann theorem) is valid since the term
< ∂ψ/∂t|H|ψ > + < ψ|H|∂ψ/∂t > vanishes on account of the Schrödinger equation
iℏ ∂ψ/∂t = Hψ.
In order to find the time derivative of the Hamiltonian we note that it depends on time
only via the time dependence of the potentials A, A0 . Part of this time dependence is
not physical and is related to the time dependent gauge transformations of A and A0 . In
order to avoid this fake time dependence we fix the gauge by choosing A0 = 0. This choice
does not fix the potentials completely but the only freedom left is time independent gauge
transformations, i.e. Eq.(1.3) with time independent χ(r). With this choice we have that
∂H δH ∂A(r,t)
Z
= dr . (1.21)
∂t δA(r,t) ∂t
Using (1.2) with A0 = 0 and Eq. (1.18) we obtain the general relation for the electric
current
δH
j(r,t) = − < ψ| |ψ > . (1.22)
δA(r,t)
Varying H with respect to A and using B = ∇ × A we obtain
q2
  
iqℏ
Z
< ψ|δH|ψ > = dr ψ ∗ (∇ · δA + δA · ∇) + δA · A ψ (1.23)
2m m
q o
−g (ψ ∗ sop ψ) · (∇ × δA)
2m
Integrating by parts in the first term, using the identity
a · ∇ × b = −∇ · (a × b) + b · (∇ × a)
for the last term in this expression and assuming that the surface terms vanish we obtain
the following expression for the current
iqℏ q2 q
j(r) = (ψ∇ψ ∗ − ψ ∗ ∇ψ) − Aψψ ∗ + g ∇ × (ψ ∗ sop ψ) (1.24)
2m m 2m
The first two terms are just the ”orbital” current already obtained earlier from the con-
tinuity equation. The last, ”solenoidal” term is the spin contribution which has the
appearance of the classical relation between the current and the magnetic moment
jspin (r) = ∇ × (gq/2m)s = ∇ × µ .
Version of April 11, 2023 15

1.4.3 Convective, diamagnetic and spin parts of the current

The first term in the expression (1.24) for the current j is called the convection current
and coincides with the usual expression for the current density in the absence of the
electromagnetic field. It is not gauge invariant without the second term which is called
the diamagnetic current. The third, spin term in j is obviously gauge invariant by itself.
In elementary quantum mechanics one develops certain intuition about currents asso-
ciated with given wave functions. In particular one is used to the fact that non vanishing
current density does not appear if the wave function is real, that the current is related to
the local complex phase of ψ, etc. This intuition is founded entirely on the first term in
Eq. (1.24) and could be misleading in the presence of electromagnetic field. In this case
one finds for instance a non vanishing orbital current density

jorbital (r,t) = (q 2 /m)Aψ 2 (r) (for ψ − real)

for a real wave function. Of course the freedom of local gauge transformations (1.14)
makes the phase of ψ and the difference between real and complex wave functions into
something which depends on the choice the gauge and therefore unphysical.

1.5 Motion in a Uniform Electric Field

Already such a simple problem as the motion of a charged particle in a constant uniform
electric field E exhibits peculiarities of gauge fields in quantum mechanics. Classically
everything is simple. The particle moves with the constant acceleration q | E | /m in
the direction of the field and has a constant, determined by initial conditions velocity
perpendicular to this direction. In quantum mechanics one may have differently looking
descriptions depending on which of the many (i.e. continuous number of) possible choices
of A and A0 is made leading to the same constant E and B = 0. Of course the gauge
invariance will assure that all physical quantities are independent of the gauge choice but
in actual calculations it may require some efforts to see the connections.

1.5.1 Static gauge

We will explore in some detail two gauge choices, the simplest and most familiar gauge
A = 0, A0 = −E · r and another, time-dependent gauge A = −Et, A0 = 0. In the former
case the time and the coordinate variables are separable in the Schrödinger equation

ℏ2 2
 
∂ψ(r, t)
iℏ = − ∇ − qE · r ψ(r, t) , (1.25)
∂t 2m

and moreover also separable are the coordinates parallel and perpendicular to E. Choosing
the x axis parallel to E and denoting by subscript ⊥ vectors which are perpendicular to
E one can write the stationary solution as

ℏ2 k2⊥
  
it
ψ(r, t) = ϕε (x) exp (ik⊥ · r⊥ ) exp − ε+ , (1.26)
ℏ 2m
Version of April 11, 2023 16

where ε and ϕε (x) are the eigenenergies and the corresponding eigenfunctions of the
motion parallel to x. They satisfy the one dimensional Schrödinger equation
ℏ2 d 2
 
− − F x ϕε = εϕε (1.27)
2m dx2
where we denoted F = q | E |.
In the equation for ϕε (x) the behavior of the potential −F x at infinite values of x
is such that the energy levels ε form a continuous spectrum of values from −∞ to +∞.
They should correspond to motion which is bounded from x = −∞ but unbounded in the
direction x → +∞ . The wave functions must vanish in the region of large and negative
x and therefore the energy levels are non degenerate. Indeed if there were two solutions
ϕ1 (x) and ϕ2 (x) for the same ε then
1 d2 ϕ1 2m 1 d2 ϕ2
= (ε + F x) = (1.28)
ϕ1 dx2 ℏ2 ϕ2 dx2
so that the Wronskian w = ϕ1 (dϕ2 /dx) − ϕ2 (dϕ1 /dx) = const. The condition that wave
functions vanish at x = −∞ means that w = 0 leading to ϕ1 = const ϕ2 i.e. the two
solutions would in fact coincide.

1.5.2 Linear potential - the Airy function

The simplest way to solve the equation for ϕε is to consider it in the momentum repre-
sentation. Inserting the expansion
Z ∞
dp
ϕε (x) = √ aε (p) eipx/ℏ (1.29)
−∞ 2πℏ
in the equation for ϕε (x) we easily obtain
 2 
p ∂
− iℏF aε (p) = εaε (p) . (1.30)
2m ∂p
Integrating this first order equation we find
p3
  
i
aε (p) = const exp εp − . (1.31)
ℏF 6m
The constant in front of this expression must be determined by normalization. Choosing
e.g., to normalize aε (p) on the delta function in ε
Z ∞ Z ∞  
i
dp a∗ε′ (p) aε (p) = |C|2 dp exp (ε − ε′ ) p = δ(ε − ε′ ) (1.32)
−∞ −∞ ℏF
√
we obtain C = 1/ 2πℏF . The wave functions in the position representation are
Z ∞
p3
  
dp i 1 α
ϕε (x) = √ exp px + εp − = √ Ai [−α(x + ε/F )] ,
−∞ 2πℏ F ℏ F 6m π F
(1.33)
Version of April 11, 2023 17

Figure 1.1: Airy function Ai(x)

where we denoted α = (2mF/ℏ2 )1/3 and introduced the notation

Z ∞   3 
1 u
Ai(ξ) = √ du exp i + ξu . (1.34)
4π −∞ 3

The function Ai(ξ) defined by this integral is called the Airy function. We will explore
some of its properties below, cf., also Ref. [10]. The graph of the Airy function is shown
in Fig. 1.1.
Eq. (1.33) together with (1.26) furnish the general solution of the Schrödinger equation
(1.25).
It is instructive to explore the asymptotic behavior of the wave function ϕε (x) for
x → ±∞. This can be found by using the saddle point approximation
cf., [Link] breth/classes/AM568/lect/[Link]

in order to evaluate the integral in (1.33). Differentiating the exponent in the integrand
we obtain that the stationary value p0 of p must satisfy

p20
ε= − Fx . (1.35)
2m
This is the classical – energy momentum relation in the potential −F x. It is an example
of a typical ”cleverness” of the saddle point method – when a phase of a rapidly oscillating
integral depends on external physical parameters (the coordinate x in (1.33)) the saddle
point condition frequently has a transparent physically significance. The difference with
the classical physics is that p0 does not have to be real. Only for x > −ε/F , i.e. in the
region where the classical motion is allowed, p0 is real, but it is pure imaginary in the
classically forbidden region x < −ε/F . In both cases there are two solutions corresponding
Version of April 11, 2023 18

to the two signs in the square root

p
p0 (x) = 2m(ε + F x) . (1.36)
According to the rules of the saddle point approximation both saddle point solutions
should be retained in the real case while only the saddle point with decaying exponential
should be admitted in the imaginary case. We thus find for (1.33)
r  3 
2m p0 (x) π
ϕε (x) ≈ cos − , for x → ∞
πℏp0 3ℏF m 4
s
|p0 (x)|3
 
1 2m
ϕε (x) ≈ exp − , for x → −∞ . (1.37)
2 πℏ|p0 | 3ℏF m

We will see in the section devoted to the semiclassical limit that these expressions cor-
respond to the semiclassical approximation for wave functions. As required the wave
function decays exponentially in the classically forbidden region E < −F x. In the classi-
cally allowed region the positive and negative momenta p0 with equal amplitudes coexist
for a stationary quantum mechanical state producing the interference cosine with the
argument which can be written as
p30 (x) 1 1 x
Z
≡ W (x) = p0 (x′ ) dx′ (1.38)
3ℏF m ℏ ℏ −E/F
in terms of the classical action W . The classical momentum determines the local wave
length λ = 2πℏ(dW/dx)−1 which decreases with increasing x in accordance with the
uniform classical acceleration in the direction of the field and the de Broglie relation.

1.5.3 Time dependent gauge

Let us now examine how this problem looks in another gauge A = −Et, A0 = 0. We use
the gauge transformation (1.14) with χ = −E · rt in the equation (1.25) and obtain
ψ(r,t) = exp(iqE · rt/ℏ)ψ ′ (r,t) , (1.39)
∂ψ ′ (r,t) 1
iℏ = (−iℏ∇ + qEt)2 ψ ′ (r,t) ,
∂t 2m
where we denoted by ψ ′ (r,t) the transformed wave function.
A simple solution of this equation is a plane wave and we obtain for ψ ′ (r,t)
′
ψk (r,t) = Ak (t) exp[ik · r] (1.40)
with the time dependent amplitude Ak (t) satisfying
dAk (t) 1
iℏ = (ℏk + qEt)2 Ak (t) . (1.41)
dt 2m
Integrating we find
i ℏ2 k2⊥

 
1 3
Ak (t) = C0 exp − t + (p + F t) , (1.42)
ℏ 2m 6mF
Version of April 11, 2023 19

where C0 is an arbitrary constant, p = ℏkx and we used the same notation for k⊥ and F
as in the previous section.
We note that with this solution the wave function before the gauge transformation
(1.39) is
ψk (r,t) = Ak (t) exp[i(k + qEt/ℏ) · r] (1.43)
which of course is a solution of the Schrödinger equation (1.25) in the static gauge. The
set of these time dependent solutions with different k’s is identical to the set (1.26) of
stationary solutions as far as the motion in r⊥ is concerned. However in the direction of
the field the sets look quite different and the point to note here is that in different gauges
the same problem may have a very different appearance.
Of course mathematically both sets are equivalent and one can easily show that each
can be expressed as a linear combination of the other. The time dependent solution is
closer to the classical intuition of the accelerated motion under a constant force.
It is instructive in this simple problem to compare the calculations of the currents for
′
two solutions ψk (r,t), Eq. (1.43), and the corresponding transformed one ψk (r,t), Eq.
(1.40). One will get different results with the two solutions

ℏk + qEt ℏk
q |C0 |2 and q |C0 |2
m m
for the convective part of the current given by the fist part of Eq.(1.24). But this differ-
ence is ”counterbalanced” by the different second diamagnetic term −(q 2 /m)A|ψ|2 in the
current expression. It is zero in the static gauge but is
Et
q2 |C0 |2
m
in the time dependent gauge. The end results is of course the same expression as it must
be for the gauge invariant quantity.

1.5.4 Translations in uniform E. Symmetries in the presence of

gauge fields
Physics in a constant electric field must be invariant under translations of coordinates
r → r + a with a constant vector a. Applying this transformation in the Schrödinger
equation (1.25) one at first finds that it is not invariant – the term −E · a is added to the
Hamiltonian. This term however can be removed if one simultaneously performs a gauge
transformation of the wave function

ψ(r,t) → exp (−iqE · at/ℏ) ψ(r + a, t) = exp [ia · (−iℏ∇ − qEt)/ℏ] ψ(r,t) (1.44)

The Schrödinger equation is invariant under this combined transformation which must be
therefore adopted as the definition of the translation in the present case. One can call it
”electric translation” in analogy with the modified ”magnetic translations” in a uniform
magnetic field, cf. Section 1.6.3 below.
This feature of modification of the standard symmetry transformations by additional
gauge transformations is quite typical for theories with gauge fields. It accounts for the
Version of April 11, 2023 20

fact that changing the coordinate system may also effect the gauge choice and care must
be taken to return to the original gauge. The generator of the infinitesimal translations
in the expression above is obviously

gop = pop − qEt . (1.45)

The symmetry means that it must be conserved and indeed one finds that

dgop /dt = ∂gop /∂t + (i/ℏ)[H , gop ] = 0

with H as it appears in the right hand side of (1.25).

We will find below two other examples of the gauge field modifications of the symme-
tries – translational and rotational invariance in a uniform magnetic field (Section 1.6.3)
and rotational invariance in the field of a magnetic monopole (Section 1.11.3).

1.6 Motion in a Uniform Magnetic Field

We will now consider the quantum mechanical motion of a charged particle in a uniform
external magnetic field B which is constant in magnitude and direction over the entire
space. For convenience we present the discussion for electrons, i.e. we take the value of
the charge
q = −e .

1.6.1 Classical motion. The guiding centers

It is instructive to recall first the classical solutions of the problem. The classical equation
of motion is
m dv/dt = −e(v × B) .
Let us choose the direction of the z axis parallel to B. Then the motion along z is free,

m dvz /dt = 0 , z = vz t + z0

with constant vz and z0 determined by initial conditions.

The equations for the x and y components are
dvx
m = −eBvy ,
dt
dvy
m = eBvx . (1.46)
dt
An important observation to be made here is that these Newton equations for the velocities
of motion in a plane perpendicular to B have the same formal appearance as the Hamilton
equations of a one dimensional oscillator with vx and vy formally proportional to the
respective coordinate and momentum of the oscillator. The solution of these equations is
”harmonic motion” in the ”velocity space” with the frequency

ωc = eB/m
Version of April 11, 2023 21

called the cyclotron frequency and

vx = v cos(ωc t + α) , vy = v sin(ωc t + α) , (1.47)

so that the trajectory in the (x,y) plane is

v vy
x = sin(ωc t + α) + x0 = + x0 ,
ωc ωc
v vx
y = − cos(ωc t + α) + y0 = − + y0 (1.48)
ωc ωc
Here v, α, x0 and y0 are constants of the motion the values of which are fixed by the initial
conditions x(t0 ), y(t0 ), vx (t0 ), vy (t0 ) at some initial time t0 .
The physical meaning of these constants is the following. The above solution describes
a circle with the radius v/ωc . The position of the centre of the circle is given by the
coordinates x0 and y0 which are therefore conventionally called the coordinates of the
guiding center, cf., Fig. 1.2 below. The value of v also determines the energy

m(vx2 + vy2 )/2 = mv 2 /2

of the motion in the (x, y) plane4 . This energy is independent of where on the x, y plane
the orbit is situated, i.e. is independent of the values of x0 and y0 .
Using the terminology of quantum mechanics we can say that the above circular motion
is degenerate - all circles with the same radius v/ωc have the same energy. This degeneracy
is characterized by different values of the guiding centre coordinates x0 and y0 so one can
say that the classical motion is ∞2 degenerate. As we will see in the next section in
quantum mechanics the motion is ”only” ∞ degenerate. It is important to observe that
the expressions of the guiding centre coordinates as given by resolving (1.48)
vy vx
x0 = x − , y0 = y + (1.49)
ωc ωc
are constants of the motion
dx0 dy0
= =0
dt dt
This fact, which is trivial in classical mechanics will play a very important role in the
quantum mechanical treatment of the problem.

1.6.2 Landau levels

The quantum mechanics of this problem was first worked out by Landau and the corre-
sponding solution is known as Landau levels.
4 The conservation of this quantity is trivially ”discovered” by multiplying the two equations (1.46)

respectively by vx and vy and adding

Version of April 11, 2023 22

The eigenenergies
The quantum Hamiltonian of a particle without spin in this case is

1 2 mv2op
Hop = (−iℏ∇ + eA(r)) = (1.50)
2m 2
where the vector potential must be chosen such that B = ∇ × A is a constant vector
parallel to z. With simple choices of A it is possible to find explicit solutions of the
corresponding Schrödinger equation as we will discuss in detail below. At the moment
however we prefer to proceed in a more general manner and show that many features of
the solution can be anticipated on the basis of simple considerations which are useful to
follow in order to gain a better understanding of the physics of the problem.
We start by considering the commutators of the components

v̂i = (−iℏ∇i + eAi )/m

of the velocity operators which enter the Hamiltonian (1.50). They are easily calculated,
 
ieℏ ieℏ ∂Ak ∂Aj eℏ
[v̂j , v̂k ] = − 2 {[∇j , Ak ] + [Aj , ∇k ]} = − 2 − = −i 2 ϵjkl Bl . (1.51)
m m ∂xj ∂xk m

These non vanishing commutators show that for a general magnetic field B(r) one can
not have definite values simultaneously for all 3 components of the velocity.
In our particular case of a constant B along the z axis only the commutator [v̂x , v̂y ]
is not zero. This means that v̂z commutes with Hop , Eq. (1.50). Since moreover one can
choose Az = 0 and Ax and Ay to be functions of only (x, y) one has

m 2 p̂2 ∂
v̂x + v̂y2 + z , p̂z = −iℏ


Hop =
2 2m ∂z
so that the parts of Hop depending on (x, y) and on z are separable. The z–dependent
part of the wave function must be a plane wave exp(ikz z) with kz = mvz /ℏ describing
quantum free motion in accordance with the classical case.
The part of Hop describing the motion in the (x, y) plane,
m 2
v̂ + v̂y2


hop ≡ (1.52)
2 x
is proportional to the sum of squares of the operators v̂x and v̂y with a constant commu-
tator
[v̂x , v̂y ] = −i(eℏB/m2 ) .
This suggests to define rescaled variables

p̂ξ = mv̂x , ξˆ = (m/eB)v̂y

with the canonical commutator

ˆ = −iℏ
[p̂ξ , ξ]
Version of April 11, 2023 23

in terms of which the operator hop takes the form

p̂2ξ (eB)2 ˆ2 p̂2ξ mωc2 ˆ2

hop = + ξ = + ξ
2m 2m 2m 2
of the Hamiltonian of a one dimensional oscillator with mass m and frequency ωc in
accordance with the character of the corresponding classical motion. The spectrum of the
oscillator is well known and adding it to the free motion eigenvalues of the p̂2z /2m term
we obtain the eigenvalues of Hop as

ℏ2 kz2
 
1
E(n, kz ) = ℏωc n + + , n = 0, 1, 2, .... (1.53)
2 2m

We have succeeded to obtain the eigenvalues of the Hamiltonian (1.50) on the basis
of the commutation relations without solving the Schrödinger equation. There however
remains a problem. The eigenvalues depend only on two quantum numbers n and kz
whereas dealing with three degrees of freedom one must find three quantum numbers
which characterise the eigenfunctions of Hop .

Degeneracy of the Landau levels. Quantum guiding centers

The independence of E(n, kz ) on the third quantum number means that the energy levels
of the problem are degenerate and we will presently determine the reason and the nature
of this degeneracy. For this purpose let us consider the quantum mechanical operators
corresponding to the guiding center coordinates, Eq. (1.49),
v̂y v̂x
x̂0 = x − , ŷ0 = y + . (1.54)
ωc ωc
We easily find that they both commute with all the components of the velocity operators,

[x̂0 , v̂i ] = [ŷ0 , v̂i ] = 0 (1.55)

Indeed, e.g.
1 1 ℏ
[x̂0 , v̂x ] = [x, v̂x ] − [v̂y , vˆx ] = [x, p̂x ] − i = 0 ; [x̂0 , v̂y ] = [x, v̂y ] = 0 , etc
ωc m m
Therefore x̂0 and ŷ0 commute with Hop , i.e. are conserved as in the classical case. As
in the classical treatment the energy of the motion is independent of these quantities.
However we find that their commutator is not zero. Indeed using (1.55)

[x̂0 , ŷ0 ] = [x̂0 , y] = −[v̂y , y]/ωc = −[p̂y , y]/mωc = i(ℏ/mωc ) (1.56)

This relation is commonly written as

[x̂0 , ŷ0 ] = iℓ2 (1.57)

where the constant

ℓ = (ℏ/eB)1/2
Version of April 11, 2023 24

is called the magnetic length.

The non vanishing commutator between x̂0 and ŷ0 means that they can not both
have simultaneously definite values and moreover the constant value of the commutator
shows that like the velocity operators above, their properties are similar to a canonical
coordinate–momentum pair. Only one of the two can be specified and since it is conserved
its eigenvalues should provide the missing quantum number which we are looking for in
order to characterize the degenerate eigenfunctions belonging to the same eigenenergy
E(n, kz ). In fact the existence of the pair of non commuting conserved operators is the
cause of the degeneracy of E(n, kz ). If we choose the states of the system to be eigen-
functions of, say, x̂0 operator, acting on one of them with ŷ0 will produce a different state
with the same energy. As we will show below there is a deep relation between the proper-
ties of the operators x̂0 and ŷ0 and the basic symmetry of the system – the translational
invariance.

The eigenfunctions
From the commutation relations Eq. (1.57) it follows that for the eigenstates with definite
x0 the values of y0 are completely undetermined so that the position of the center of
the quantized cyclotron orbit will have equal probability to be found at any point along
the line with the given x0 . To see this explicitly we now turn to the solutions of the
Schrödinger equation which have definite values of x0 . We need to choose first the gauge
for the vector potential A. The explicit forms of x̂0 ,
v̂y
x̂0 ≡ x − = x − (1/eB)p̂y − Ay /B
ωc
and of
mv̂x = p̂x + eAx
suggest the following convenient choice

Ax = 0 , Ay = Bx , Az = 0 → B = (0, 0, B) (1.58)

for which
x̂0 = iℓ2 ∂/∂y , mv̂x = −iℏ∂/∂x
and the Hamiltonian
p̂2x 1 2 p̂2
Hop = + (p̂y + eBx) + z . (1.59)
2m 2m 2m
The eigenfunctions of x̂0 and p̂z have the form
2 1 1
ψ(r) = const ϕ(x) e−ix0 y/ℓ eikz z , const = p √ (1.60)
Ly Lz

with yet undetermined ϕ(x). For convenience we have assumed that the motion in the
y and the z directions is limited by large but finite intervals Ly and Lz with periodic
boundary conditions.
Version of April 11, 2023 25

Inserting in the Schrödinger equation Hop ψ = Eψ and separating the variables we

obtain  2
mωc2

p̂x 2
+ (x − x0 ) ϕ(x) = εϕ(x) , (1.61)
2m 2
where we denoted ε = E − (ℏkz )2 /2m. This is the equation of a harmonic oscillator
centered around the eigennvalue of x0 . As anticipated the eigenenergies are given by
(1.53) and are independent of x0 . The eigenfunctions are

ϕn,x0 (x) = χn (x − x0 ) , (1.62)

where χn (x) are the normalized eigenfunctions of harmonic oscillator

  14
1 1
exp −x2 /ℓ2 Hn [x/ℓ ]
 
χn (x) = √ (1.63)
πℓ2 n
2 n!
2 2
with Hn (x) = (−1)n ex (dn /dxn )e−x – the Hermite polynomials. The first few functions
χn (x) are
 14 1 √
x2 x2
     
1 1 4 2x
χ0 (x) = exp − 2 , χ1 (x) = exp − , (1.64)
πℓ2 2ℓ πℓ2 ℓ 2ℓ2
 14  2
x2
   
1 x
χ2 (x) = − 1 exp − , etc .
4πℓ2 ℓ2 2ℓ2

Imposing periodic boundary conditions in the y direction we find that x0 in Eq.(1.60)

takes discrete values separated by distances ∆x0 = 2πℓ2 /Ly . We thus have one state per
area Ly ∆x0 = 2πℓ2 in the x-y plane. The dependence of the wave functions on y via
the plane wave phase means that the probability to find a particle is independent of this
coordinate. It also seems to suggest that like in the z - direction there is a free motion
also in the y-direction. This however is not corrects as it is based on the experience in
situations in which there was no gauge field present. In this case the wavefunction’s phase
is gauge dependent so to evaluate what motions it describes one must form gauge invariant
observables. We will do this below by calculating the current density components with
physically interesting results.
In the x direction the state is centered around the value x0 . Its extension can be
determined, using e.g., the equipartition property of the oscillator meaning that the aver-
2 2
age potential
p energy is one half
p of the total energy, mωc ⟨(x − x0 ) ⟩/2 = ℏωc (n + 1/2)/2.
2
This gives ⟨(x − x0 ) ⟩ = ℓ n + 1/2. Each degenerate energy level can thus p pictorially
be viewed as a two dimensional plane filled with overlapping (for Ly ≫ 2πℓ/ n + 1/2)
”strips” occupied by individual quantum states parallel to the y axis representing quan-
tized cyclotron orbits uniformly ”smeared” along every strip. This picture repeats itself
for every
p n and kz with the radius of the orbits, i.e. the thickness of the strips growing
as ℓ n + 1/2. The smearing of the orbits is the result of the Heisenberg–like uncertainty
relation between the guiding center coordinates x0 and y0 .
The degenerate energy levels which we have just described are called Landau levels.
Choosing y0 to have defined values will lead to the same picture of Landau levels but
Version of April 11, 2023 26

with the strips parallel to the x axis. It is amusing to consider what happens if more
complicated functions of x0 and y0 are chosen to have defined values. Suppose we fix
x20 + y02 . Then the strips in the picture above will have the shape of concentric circles
around the origin. Choosing fixed x20 /a2 + y02 /b2 with some constants a and b will lead to
strips of elliptic shapes, while fixing the function (x0 y0 + y0 x0 )/2 (symmetrized to make
the corresponding operator hermitian) will result in a hyperbolic shape of the strips, etc.
In Fig.1.2 we illustrate some of these cases. Of course all the choices above are equiv-
alent as long as the degeneracy remains but some may be singled out if a perturbation
removing this degeneracy is added to the Hamiltonian.
Version of April 11, 2023 27

Figure 1.2: Schematic illustration of a single classical cyclotron orbit and how it gets
”smeared” in quantum mechanical description. The upper figure shows the cases of x0
fixed (an orbit is smeared in the y-direction) or y0 fixed (orbit is smeared in the x-
direction). In the lower figure x20 +y02 is fixed - an orbit is smeared along the corresponding
circle
Version of April 11, 2023 28

When choosing the eigenvalues of the operator ŷ0 instead of x̂0 for the characterization
of the degenerate wave functions it should be more convenient to choose the gauge Ax =
−By, Ay = Az = 0 in which ŷ0 is just −iℓ2 ∂/∂x. For a combination x̂20 + ŷ02 the symmetric
choice Ax = −By/2, Ay = Bx/2, Az = 0 is the most appropriate. In the literature it may
sometimes seem that the choice of the gauge determines which combination of x̂0 and ŷ0
will be diagonal. Our remark here is meant to clarify the correct order of choices .
The square of the magnetic length ℓ appearing in the commutator of the guiding
center coordinates plays the role of the ”Planck constant” for these variables. Therefore
the analogue of uncertainty relation ∆x0 ∆y0 ≥ ℓ2 /2 must hold. We also recall from the
statistical physics that in the semiclassical picture quantum states ”occupy” phase space
volume ∆p∆q = 2πℏ. Here we may expect an analogous situation that a single state
of a degenerate Landau level ”occupies” an area in the plane of (x0 , y0 ) which is 2πℓ2 .
And indeed we have seen this in the particular case of the solutions (1.60). The physical
meaning of this minimal area is simple and profound – the magnetic flux through this
area is ratio of universal world constants
ℏ 2πℏ h
B · 2πℓ2 = 2πB = =
eB e e
Such magnetic flux has a special notation
2πℏ h
Φ0 = = (1.65)
e e
and a special name - magnetic flux quantum. We will meet this quantity a number of times
in these notes (cf., below). Let us stress that its name doesn’t mean that the magnetic
flux in such problems is quantized. Rather, as we see here and will be seen below certain
physical features get repeated with Φ0 as a period.
As can be seen from the above discussion the density of single states in a degenerate
Landau level is the inverse of 2πℓ2 which is independent of n and of the way we choose
to classify the degeneracy. The mnemonic rule of ”one state per one flux quantum” is
something which is encountered in many quantum mechanical problems in the presence
of magnetic field and is therefore well worth remembering.

Currents and edge currents

Individual states in a Landau level carry a non vanishing current density. Apart from
an obvious contribution from the free motion in the z–direction one also finds current
distribution in the x-y plane. Qualitatively one expects that in this plane the quantum
mechanically smeared cyclotron orbits with one fixed guiding center coordinate should
combine to give opposite currents parallel to and concentrated on the edges of the strip
occupied by the state and have zero current on the midline of the strip.
We easily find for the states Eq. (1.60) using Eq. (1.24) for the current

jx = 0 , jy (x) = eωc (x0 − x)ρ(x) , jz (x) = e(ℏkz /m)ρ(x) (1.66)

where we denoted the particle density

ρ(x) = ϕ2 (x)/Ly Lz
Version of April 11, 2023 29

The appearance of the lengths Ly and Lz is related to the (standard) normalization of

the wavefunction (1.60) to one particle.
The wave function ϕ2 (x) as given by any of the solutions Eq. (1.62) is concentrated
in a symmetric strip around x0 which means that the current density jy (x) has an anti-
symmetric profile with respect to x = x0 . Because of this antisymmetry the total current
Z
Iy = jy (x) dx (1.67)

flowing in the y-directions, i.e. along the state ψn,x0 in the x-y plane is zero for these
states.
If one adds a constant electric field parallel to the x–axis one can still find exact wave
functions (cf., homework problems or tutorial). The current density profile of these wave
functions will change from antisymmetric to asymmetric and the total current in the y
direction will not be zero.
Another interesting non zero current carrying Landau states appear at the edges of
the x-y plane. Let us assume that an additional potential V (x) with the shape shown in
Fig. 1.3 is added to the Eq. (1.61)
 2
mωc2

p̂x 2
+ (x − x0 ) + V (x) ϕ(x) = εϕ(x) , (1.68)
2m 2

This potential simulates the edges of the sample in the x direction.

Figure 1.3: Potential simulating edges in the x-y plane

It is instructive to examine how the combined potential

mωc2 2
U (x) = (x − x0 ) + V (x)
2
changes when plotted for different guiding center coordinate values x0 relative to the
positions of the potential walls representing the edges. The shape of U (x) getting more
 Version of April 11, 2023 30

narrow for the values of x0 near and ”inside” the edges indicates that eigenenergies ϵn will
break the degeneracy of the Landau levels in such a waypthat they become rising functions
ϵn (x0 ) for such values of x0 . Recalling the width ℓ n + 1/2 which the unperturbed
Landau levels occupy we can approximate for low n values and the potential V (x) slowly
varying on the magnetic length ℓ scale as
V (x) ≈ V (x0 )
1 (1.69)
ϵn (x0 ) ≈ ℏωc (n + ) + V (x0 )
2
In this approximation the modified Landau levels similar to the unperturbed ones form
an ”equidistant ladder” with each step having the shape of V (x0 ).
The asymmetric shape of the combined U (x) for x0 near the edges means that the
resulting eigenfunctions ϕn,x0 (x) will not depend on x0 via x − x0 as in Eq. (1.62) and
will not have the harmonic oscillator symmetry around x0 as in the unperturbed Landau
states. This in turn means that the current density jy (x) along the edge will not have
an antisymmetric profile with respect to x = x0 and therefore the total current flowing
in the x-y plane for such states near the edges will not be zero. Such currents are called
”edge currents”. They correspond to the skipping classical orbits near potential walls,
move in opposite directions on the opposite edges and play important role in explaining
the Quantum Hall Effect, cf., Ref. [13].
Let us express the energy of a given state ϕn,x0 using
hop (x0 )ϕn,x0 = ϵn (x0 )ϕn,x0 → ϵn (x0 ) = ⟨ϕn,x0 |hop (x0 )|ϕn,x0 ⟩
with hop (x0 ) denoting the Hamiltonian operator in the left hand side of Eq. (1.68). Using
the Feynman-Hellman theorem5 one obtains
∂ϵn (x0 ) ∂hop
= ⟨ϕn,x0 | |ϕn,x0 ⟩ =
∂x0 ∂x0
Z (1.70)
= mωc2 ⟨ϕn,x0 |(x − x0 )|ϕn,x0 ⟩ = mωc2 dx(x − x0 ) ϕ2n,x0 (x)

Comparing with the expression for the current density jy (x) in Eq. (1.66) and ignoring
for convenience the z direction we find the relation
∂ϵn (x0 ) mωc Ly
= Iy (n, x0 ) (1.71)
∂x0 e
5 The theorem relates the derivative of the eigenenergy with respect to a parameter to the expectation
value of the derivative of the Hamiltonian with respect to that parameter. The proof is straightforward
∂ϵn (x0 ) ∂
= ⟨ϕn,x0 |hop (x0 )|ϕn,x0 ⟩ =
∂x0 ∂x0
∂ϕn,x0 ∂ϕn,x0 ∂hop
=⟨ |hop (x0 )|ϕn,x0 ⟩ + ⟨ϕn,x0 |hop (x0 )| ⟩ + ⟨ϕn,x0 | |ϕn,x0 ⟩ =
∂x0 ∂x0 ∂x0
 
∂ϕn,x0 ∂ϕn,x0 ∂hop ∂hop
= ϵn (x0 ) ⟨ |ϕn,x0 ⟩ + ⟨ϕn,x0 | ⟩ + ⟨ϕn,x0 | |ϕn,x0 ⟩ = ⟨ϕn,x0 | |ϕn,x0 ⟩
∂x0 ∂x0 ∂x0 ∂x0
where it was used that
∂ ∂ϕn,x0 ∂ϕn,x0
⟨ϕn,x0 |ϕn,x0 ⟩ = 0 = ⟨ |ϕn,x0 ⟩ + ⟨ϕn,x0 | ⟩
∂x0 ∂x0 ∂x0
Version of April 11, 2023 31

where Iy (n, x0 ) is the total current of a single particle in the ψn,x0 state. Referring to
Eq. (1.69) with V (x) as shown in Fig. 1.3 one sees clearly where the edge currents are
expected, their magnitude and direction.

1.6.3 Degeneracy of Landau levels and space symmetries

Conservation laws are always results of symmetries and the existence of the conserved
operators x̂0 , ŷ0 and v̂z is not an exception. They are related to the basic symmetry of
the motion in a uniform field – invariance under translations. This invariance is however
not explicit in the Hamiltonian (1.50) which changes under the translation r to r + a with
an arbitrary constant vector a. We have already encountered a similar phenomenon in
the simpler case of a uniform electric field. Also here the the Hamiltonian (1.50) remains
invariant if simultaneously with the proper translation one performs a suitably chosen
gauge transformation. The conserved quantities should be the appropriate generators of
these combined transformations.
To see this in detail we observe that after a proper translation the Schrödinger equa-
tion with the Hamiltonian (1.50) has the same form but with the different vector potential
A(r + a). For a constant B however the difference A(r + a) − A(r) is a gauge transfor-
mation, i.e. it is a gradient of a scalar function. It will be sufficient to show this for an
infinitesimal a for which we have A(r + a) ≈ A(r) + (a · ∇)A(r). The last term is

ai ∂i Aj = ai (∂i Aj − ∂j Ai ) + ai ∂j Ai = ai (ϵijk Bk + ∂j Ai )

and for a constant B it is indeed a gradient

∂j α with α = ai (ϵijk xj Bk + Ai ) = a · [r × B + A(r)] (1.72)

It can be removed from Hop by a gauge transformation of the wave function in addition
to the proper translation. The symmetry transformation is therefore
 
ie
ψ(r) → 1 + a · [r × B + A(r)] (1 + ia · pop /ℏ)ψ(r) (infinitesimal a) , (1.73)
ℏ

where we have used the proper translation operator exp(ia · pop /ℏ) for infinitesimal a to
write ψ(r + a) in terms of ψ(r).
The combined transformation (1.73) is what should be called translation in the pres-
ence of a magnetic field (the term ”magnetic translation” is sometimes used). The gener-
ators of this transformation are read off the linear term in a found after multiplying the
brackets in Eq.(1.73). They are

pop + e[A(r) + r × B] = mvop + eB(r × ez ) . (1.74)

For translations along the z axis this is just mvz whereas for the translations along the
x and y axes we obtain respectively eB ŷ0 and −eB x̂0 in terms of the operators of the
guiding center coordinates.
It should now become intuitively clear why these operators do not commute. We
expect that the result of translating the wave function parallel to x and then parallel to
y should not be the same as translating it in the opposite order. The difference should
Version of April 11, 2023 32

be related to the Aharonov-Bohm phase (see Section 1.7 below for its definition) induced
by the flux of the magnetic field through the rectangle obtained in the course of these
reversed order translations. Let us see how it happens. Transporting a wave function by
an infinitesimal ∆x followed by ∆y and then by −∆x and −∆y respectively one indeed
obtains keeping the terms up to a 2nd order in the translations ∆x and ∆y
1 1
(1 − i∆yKy − (∆y)2 Ky2 )(1 − i∆xKx − (∆x)2 Kx2 ) ×
2 2
1 1
×(1 + i∆yKy − (∆y)2 Ky2 )(1 + i∆xKx − (∆x)2 Kx2 )ψ(r) =
2 2
= (1 + ∆x∆y[Kx , Ky ])ψ(r) = [1 + 2πi(∆Φ/Φ0 )] ψ(r) (1.75)

where we denoted by ℏKx and ℏKy the corresponding vector components of the operator
of translations (1.74) and ∆Φ = B∆x∆y – is the flux through the rectangle. Since the first
non vanishing term in the expression above apart of unity was quadratic and proportional
to ∆x∆y it was necessary to keep the quadratic terms in the operators of each translation.
A similar discussion concerning the generalization of transformations and their gener-
ators can be worked out for another symmetry of the problems – the rotational symmetry
around the direction of the magnetic field B. We will leave this for homework or tutorials.

1.7 The Aharonov - Bohm Effect

1.7.1 Local and non local gauge invariant quantities
We have emphasized in Section 1.3.3 that all observable quantities in a theory with a
gauge field are gauge invariant. Perhaps the simplest such quantities are the electric
and magnetic fields and the particle density ρ(r,t) = |ψ(r, t)|2 . In the expression for
the electric current density considered in the previous section we encountered another set
involving the derivatives of ψ – the combinations ψ ∗ (r,t)Dψ(r,t) to which we can also add
their time dependent partner ψ ∗ (r,t)D0 ψ(r,t). These combinations are gauge invariant
due to the simple transformation properties of the gauge covariant derivatives (1.15).
A distinct feature of all these invariants is that they depend on ψ, A and A0 and
their first derivatives at
H the same space–time point, i.e. they are local. Consider however
a circulation integral A · dr taken around some closed contour drawn in space. By
the Stockes theorem this integral is equal to the flux of B through the contour and is
therefore gauge invariant. This is an example of a non local gauge invariant quantity. In
the
H following sections we will discuss situations in which the non trivial dependence on
A · dr leads to unexpected quantum mechanical effects which are collectively known
as the Aharonov–Bohm effect, Ref. [6]. The sensitivity of the quantum theory to non
local gauge invariants can be traced to essential non locality of the quantum mechanical
description — eigenvalues and expectation values of various physical quantities such as
energy, angular momentum, etc., depend on what happens with the wave function in the
entire configuration space of the system.
Concluding this section we mention that in addition to the circulation of the vector po-
tential another type of non local gauge invariants appears in certain physical applications.
Version of April 11, 2023 33

These are bi-local quantities of the type

Z r′
∗ ′
ψ (r ) exp[i(e/ℏ) A(r′′ ) · dr′′ ]ψ(r, t) .
r
Under gauge transformations the exponential and the wave functions produce phase fac-
tors which cancel each other. Quantities like this are often met in the field theoretical
context and recently in certain many body problems.
H
1.7.2 Quantum mechanics ”feels” non zero C
A · dr even if E =
B = 0 on and near the contour C
Let us consider a region of space in which Hlocal invariant quantities E and B are zero but
in which contours can be found for which A · dr does not vanish. A simple example is a
region outside of a long thin tube with impenetrable walls and a non zero magnetic field
concentrated inside and running parallel to the tube, Fig.1.4

Figure 1.4: Example of the Aharonov-Bohm flux Φ in an impenetrable tube and H a closed
contour encircling it in the region with zero E and B. The non zero circulation C A·dr =
Φ around such contours has no effect in classical description of charged particle motion
(trajectories) in this region but produces observable effects in its quantum mechanics
(wave functions)
H
Non zero circulation integrals A · dr are obtained for the integration contours which
wind around the tube. Since by assumption B = ∇×A = 0 outside the tube the details of
a particular contour are of no importance except for the number of times it winds around
the tube and the direction of this winding. Denoting this number by n one can write
I
A · dr = nΦ , n = 0, ±1, ... (1.76)
C
Version of April 11, 2023 34

Here Φ denotes the magnitude of the total flux of the magnetic field in the tube. The
circulation integrals outside the tube depend only on Φ and not to the details of the
magnetic field distribution. One conventionally refers to such a tube as a solenoid and to
such an isolated magnetic flux Φ as the Aharonov–Bohm flux (AB flux for brevity).

1.7.3 ”Gauging out” the AB flux. Periodic dependence on its

value
Classically the free motion of a particle in the outside region is not influenced by the
presence of the field inside the tube. At first sight one may reach a similar conclusion in
the quantum mechanical description. Indeed to write the Schrödinger equation one needs
to determine first the electromagnetic potentials. Since B = ∇ × A = 0 in the outside
region one must have that A must be a gradient of some scalar,

A = ∇ξ(r) (1.77)

With such A (and E = 0 ) it may appear that in the corresponding Schrödinger equation
∂ψ 1 2
iℏ = (−iℏ∇ + eA(r)) ψ
∂t 2m
one could remove the eA term by a gauge transformation
e
ψ(r, t) = ψ ′ (r, t) exp[−i ξ(r)]
ℏ
with Z r Z r
ξ(r) = ξ0 + A(r′ ) · dr′ = ξ0 + ∇ξ(r′ ) · dr′
r0 r0
and ξ0 some constant.
The problem however with this elimination of A from the Schrödinger equation is that
in the presence of the AB flux Φ the scalar function ξ(r) in Eq. (1.77) is not single valued.
It is a multivalued function as can be seen in the following way. To have the required value
of the AB flux the function ξ(r) must change by Φ when ”taken (followed) continuously”
along a contour C around the solenoid in the positive direction
I I Z rf
Φ= A · dr = ∇ξ(r) · dr = ∇ξ · dr = ξ(rf ) − ξ(ri ) with rf = ri . (1.78)
C C ri
Thus at every r in the region outside the solenoid the function ξ(r) has many (infinity)
of values differing by nΦ with (positive or negative) integer n.
Given this the transformed ψ ′ (r, t),
e
ψ ′ (r, t) = ψ(r, t) exp[i ξ(r)]
ℏ
will also be multivalued - its phase will change by
e Φ
∆ξ = Φ = 2π (1.79)
ℏ Φ0
Version of April 11, 2023 35

when ”taken continuously” around the solenoid.

To understand what the demand of such a particular non single valuedness of the wave
function produces let us consider a specific example of the angular momentum. Assuming
the z-axis along the solenoid and the z component L̂z we have for its eigenfunctions

L̂z ψ(ϕ) = ℏνψ(ϕ) → ψ(ϕ) = const eiνϕ

In the usual case, i.e. in the absence of the AB flux one applies the condition ψ(ϕ+2nπ) =
ψ(ϕ), i.e. the condition of single valuedness of ψ(ϕ) which leads to the usual integer
quantization
ν = M , M = 0, ±1. ± 2, ...
For the multivalued function condition Eq. (1.79) we have

eiν(ϕ+2nπ) =eiνϕ ei2πnΦ/Φ0 → 2nπ(ν − Φ/Φ0 ) = 2nπM →

(1.80)
→ ν = M + Φ/Φ0

This shows that despite our ”gauging out” of the vector potential A = ∇ξ its gauge
invariant content, i.e. the AB flux Φ in Eq. (1.78), if not zero modifies the physics via
the resulting multivalued wave function condition Eq. (1.79). We will see below that this
modifications is (not surprisingly) identical to the straightforward solution with such a
vector potential.
As an important additional observation we note that when Φ = nΦ0 there is no
effect! The transformed ψ ′ remains single valued and such AB flux is non observable
”from outside”. This observation is probably one of the advantages of the multivalued
wave function formulation. It also means that the Aharonov-Bohm effects have periodic
dependence on the magnitude Φ of the AB flux with the period of the flux quantum Φ0 .
One can see this in the dependence of the eigenvalues ν on Φ, Eq. (1.80). They change
from integer to non integer with the period Φ0 . We will also see this periodicity in the
examples considered in the next section and provide a more general point of view in
Section 1.7.5.
Another important observation is the following. The view of the Aharonov–Bohm
effect as a modification of the condition that the wave function repeats itself as it is taken
around a solenoid stresses that in order to ”feel” this modification the wave function must
extend all around the solenoid. Otherwise there will be no observable consequences of
the Aharonov–Bohm flux. Below we will consider an example of a ring pierced by the
Aharonov–Bohm flux with a particle localized on a finite sector of the ring. There is no
Aharonov–Bohm effect in this case.

1.7.4 Example of the AB flux

Assume that the solenoid with the AB flux Φ is placed along the z-axis. A possible simple
choice for the vector potential outside such a solenoid is
Φ Φ
A = ∇ξ(r) , with ξ(r) = arctan(y/x) ≡ ϕ (1.81)
2π 2π
Version of April 11, 2023 36

where ϕ is the azimuthal angle. Recalling the expression of the gradient in cylindrical
coordinates r, ϕ, z
∂ 1 ∂ ∂
∇ = er + eϕ + ez
∂r r ∂ϕ ∂z
one finds
Φ
Aϕ = ; Ar = Az = 0 (1.82)
2πr
and therefore the circulation integral outside the solenoid along a circular contour in a
plane perpendicular to the solenoid
I Z 2π
A · dr = Aϕ rdϕ = Φ
0

Since ∇ × A = B = 0 outside the solenoid one can deform the above circular contour
without changing the integral as long as the new contour has ”the same topology” - i.e.
encircles the flux once in the same direction. One can also change the particular A in
(1.81) by Hadding a single valued function to ξ without influencing B = 0 or circulation
integrals A · dr outside the solenoid. We observe that the dependence of the outside
vector potential on the magnetic field is via the flux Φ irrespective of a particular radial
dependence of B inside the solenoid.
To have a convenient example of the AB flux one can think of B = B(r)ez with a
constant B inside and zero outside. With this magnetic field one can write for all r’s,

A = (Br/2) eϕ inside the solenoid ,

Φ
A = (Br02 )/2r eϕ = eϕ outside the solenoid , (1.83)
2πr
where r0 is the radius of the solenoid. Since Φ = B(πr02 ), this expression for the outside
region is the same as (1.81).

The Hamiltonian and the spectrum

The Hamiltonian with the vector potential (1.82) outside the AB flux has a simple form
in cylindrical coordinates. Using p = pr er + pϕ eϕ + pz ez and A = Aϕ eϕ we have

p2r 1 2 p2
H= + (pϕ + eAϕ ) + z + U (r) , (1.84)
2m 2m 2m
where we disregarded the spin degrees of freedom and added U(r) - the potential which
should account for the impenetrable walls of the solenoid. Note that in our notation here
pϕ is a projection of p on eϕ and is related to the z-projection of the angular momentum
as
(r × p)z = rpϕ = Lz
Compared to the situation without the magnetic flux the Hamiltonian (1.84) is modified by
the presence of the potential Aϕ in the centrifugal term which depends on the combination
1 1
pϕ + eAϕ = (Lz + eBr02 /2) = (Lz + eΦ/2π)
r r
Version of April 11, 2023 37

In classical mechanics one could absorb the constant eΦ/2π into Lz and completely elimi-
nate Aϕ from the equations of motion. However in quantum mechanics this freedom does
nor exist since Lz becomes an operator Lz = −iℏ ∂/∂ϕ which has discrete eigenvalues
ℏM (M – integer). The eigenvalues’ selection follows from the requirement that the wave
function is single valued which imposes the periodic boundary conditions

ψ(r, ϕ, z) = ψ(r, ϕ + 2π, z).

The eigenvalues of the angular part Lz +eΦ/2π in the expression for pϕ +eAϕ are therefore

ℏ(M + eΦ/2πℏ) ≡ ℏ(M + Φ/Φ0 ) , Φ0 = h/e

which is identical with what was obtained in Eq. (1.80) of our discussion of the effect of the
multivalued wave function condition obtained after ”gauging out” the vector potential.
In the Schrödinger equation Hop ψ = Eψ one can separate the z-part and use

1
ψ(r, ϕ) = R(r) √ eiM ϕ
2π
to write the radial part of the equation as
 2
ℏ2 (M + Φ/Φ0 )2

p̂r
+ + U (r) R(r) = ε R(r)
2m 2mr2

with ε the corresponding part of the total energy E.

The above change of the spectrum of the centrifugal part of the potential is a manifes-
tation of the Aharonov-Bohm effect in this example. A classically unobservable magnetic
flux inside an impenetrable solenoid causes an observable effect in the outside region when
the problem is treated quantum mechanically. The dependence on the magnitude Φ of
the flux exhibits periodicity with magnetic flux quantum Φ0 as a period.

Thin ring solution

Let us see how this happens in a simple model of a thin ring. To construct this model we
add to the Hamiltonian (1.84) a potential V (r, z) constraining the motion in the r and
z directions to a very narrow ring region. It is the simplest to choose V (r, z) as zero for
|r − a| ≤ b, |z| ≤ b and infinite otherwise. This gives a ring of thickness b with radius a
lying in the z = 0 plane. For a very small b the radial coordinate in the second term in
(1.84) can be set to the fixed radius a and the motion in the azimuthal direction becomes
decoupled from r. The Hamiltonian of this motion is just
 2
1 ∂ eΦ
Hϕ = −iℏ + (1.85)
2ma2 ∂ϕ 2π

with eigenfunctions
1
ψM (ϕ) = √ exp(iM ϕ) , M = 0, ±1, ±2, ...
2π
Version of April 11, 2023 38

and the corresponding eigenvalues

2
ℏ2

Φ
EM = M+ . (1.86)
2ma2 Φ0

The energies of the motion in the r and the z directions in this approximation are inde-
pendent of Φ and we will not be concerned with them.
In Fig. 1.5 we plot the dependence of the energy levels on the magnetic flux which shows
the Φ0 periodicity of the Aharonov–Bohm effect. An analysis which we do not reproduce
here shows that if there is a weak additional potential V (ϕ) acting on a particle on the
ring the behavior of the levels will follow the pattern of the solid lines in Fig. 1.5 which
retain the same periodicity, Ref.[12].

Figure 1.5: Schematic diagram of the electron energy levels as a function of the flux
Φ/Φ0 in a one-dimensional ring encircling the flux, Ref.[12]. Solid and dashed curves,
respectively, are for the ring with and without weak additional potential V (ϕ) acting on
a particle on the ring, Ref.[12].

Consider now a case of a strong potential V (ϕ), so strong that the particle is localized
in a finite sector of the ring as opposed to the free motion around the entire circumference
of the ring as in (1.85). A simple such V (ϕ) is a potential ”well” V (ϕ) = 0 for 0 < ϕ <
ϕ0 < 2π and infinite outside this interval. The eigenfunctions in this case are zero except
in the interval with zero potential where they are easily found to be
r    
2 Φ πnϕ
ψn (ϕ) = exp i ϕ sin , n = 1, 2, 3, ... (1.87)
ϕ0 Φ0 ϕ0
Version of April 11, 2023 39

The dependence on the flux enters in the phase of these functions but the corresponding
eigenenergies do not depend on it at all,

ℏ2 π 2 n 2
En = . (1.88)
2ma2 ϕ20

In Fig. 1.5 they would be represented by horizontal straight lines giving a trivial limiting
case of the general periodic dependence on Φ referred to above. Here we have an example
in which the localization of the eigenfunctions on a part of the ring leads to the disap-
pearance of the the Aharonov–Bohm effect (the Φ dependent phase is the same for all
solutions and is therefore not observable in this case). As we have already stressed, in
order to have a sensitivity to the Aharonov – Bohm flux the wave function must have a
”tail” extending all around the flux.

AB effect in quantum interference and scattering off the AB flux

Here we will briefly consider two additional manifestations of the AB effect.
2-slit with AB flux
The understanding that Aharonov–Bohm flux modifies the phase of the wave function
leads to an intuitive way of describing the Aharonov – Bohm effect as the change of
the interference of the quantum mechanical waves as they propagate on each side of the
solenoid. Let us consider the classic 2-slit experiment as depicted in Fig.1.6.

Figure 1.6: Double slit interference in the presence of the Aharonov-Bohm flux Φ

Electrons pass from a point source through a wall with 2 narrow slits and fall on a
screen behind it, cf., Ref. [8]. As long as it is not detected which slit the electrons pass
through, they produce an interference pattern according to the phase difference for paths
going via each of the slits.
If the Aharonov–Bohm solenoid is placed behind the wall between the slits this phase
Version of April 11, 2023 40

difference will change by the amount

Z 
e e 2πΦ
Z I
∆β = A · dr − A · dr = A · dr = (1.89)
ℏ 1 2 ℏ Φ0

where as before Φ is the flux through the solenoid and the subscripts 1 and 2 denote
integrals along the two trajectories in Fig. 1.6. For a position y on the screen (measured
from its centre) the phase difference√between waves from the two slits in the absence of
the solenoid is β = k∆L where k = 2mE/ℏ is the wave number and ∆L – the difference
in the paths the waves travel from the slits to the screen.
For a distance b from the slits to the screen and for y, d << b one can approximate
∆L = (y/b)d where d is the distance between the slits. Therefore a given phase difference
β will be found at y = (β/kd)b. The additional phase difference ∆β due to the Aharonov
– Bohm flux will result in a shift in the interference pattern by the amount
∆β 2πb Φ
∆y = b= (1.90)
kd kd Φ0
Version of April 11, 2023 41

Scattering off the AB flux

Yet another way to see the phase difference between the waves which pass on different
sides of the solenoid is to consider a scattering of a plane wave from it. This was discussed
in the original paper by Aharonov and Bohm, Ref. [6]. For the vanishing magnetic flux
one finds a standard picture of a cylindrical wave scattered from the solenoid with the
amplitude which falls like r−1 superimposed on the initial plane wave. For non zero Φ the
wavefronts in the region ”down stream” and far away from the solenoid form a pattern
of two flat fronts shifted with respect to each other in an abrupt, almost discontinuous
fashion along the line stretching from the solenoid in the direction of the propagation
of the original plane wave. The magnitude of the shift is given by the phase difference
2πΦ/Φ0 divided by the wave number k. We refer reader to Ref. [6] for the details of this
discussion.

1.7.5 Multiply connected regions. Homotopy

Certain features of the results obtained in examples above are quite general in nature.
In any region with zero E and B Eqs. (1.2) imply that the vector potential must be the
gradientRof a (time-independent)
R function, A(r) = ∇ξ(r) and that A0 = const. The
integral A · dr = ∇ξ · dr is equal to the difference between the values of ξ at the initial
and the final points of the contour of the integration so that it must be zero for a closed
contour unless (a) the function ξ is not single valued and (b) the contour takes ξ from one
of its branches to another. This can not happen in simply connected regions, i.e. such
in which all closed contours are
H contractable to a point. Since continuous deformations
of the contour in the integral ∇ξ · dr can not change its value all such integrals will be
zero for contractable contours. Equivalently stated, a regular function like ξ(r) must be
single valued in a simply connected region.
Consider however multiply connected regions. These are regions where one can find
closed contours which can not be contracted to a point without crossing the boundaries.
The impenetrable solenoid and the ring discussed above are examples of such regions.
The contours around the ”tube” of the solenoid or around the ”hole” R of the ring can
not be contracted to zero. Non zero values for the circulation integral A · dr are to be
expected for such contours and actually occur when there is a magnetic flux through the
excluded regions. We note in passing that not every shape of excluded region will lead
to the existence of non contractable contours. Excluded cavity of a spherical shape for
instance will not. Its existence will create uncontractable closed surfaces but not curves
and will be relevant for considerations of e.g. non vanishing surface integrals of a vector
field with zero divergence.
All possible closed curves in a multiply connected region can be divided into classes
of curves which can be contracted into each other. Such classes are called homotopy
classes of curves, cf., [8]. Among all homotopy classes one can define a complete set
of elementary classes of curves Ck out of which every other non elementary class can
be obtained by multiple traverses of curves belonging to the elementary classes. For a
solenoid there is one elementary class of curves encircling the solenoid once in, say, a
clockwise direction and one in the counter clockwise direction. Clearly the changes of ξ
on closed curves within each elementary class must be the same since the curves can be
continuously deformed into each other. For different elementary classes however they in
Version of April 11, 2023 42

general will be different reflecting possible different values of the Aharonov–Bohm fluxes
through different excluded regions or their different signs.
Let us apply these considerations to a general system of charged particles in a multiply
connected field free region, cf. Ref. [9]. Their Schrödinger equation is

H[pa − qa A(ra ), ra ]ψ({ra }) = Eψ({ra }) (1.91)

where we assumed a general Hamiltonian depending on the momenta pa = −iℏ∇a and

coordinates ra of the particles with charges qa , a = 1, 2, ..., N . Since in the field free region
the vector potential A is ”pure gauge”, A = ∇ξ, we can apply a gauge transformation
N
X
ψ({ra }) = exp[i qa ξ(ra )/ℏ]ψ ′ ({ra }) (1.92)
a=1

and find that ψ ′ satisfies

H[pa , ra ]ψ ′ ({ra }) = Eψ ′ ({ra }) (1.93)

with the Hamiltonian in which the potential A was ”gauged out”. Since ψ is single
valued and since going around any elementary non Hcontractable contour Ck increases
ξ by the corresponding Aharonov–Bohm flux Φk = Ck A · dr , we must demand that
ψ ′ ({ra }) is multiplied by the factor exp(−iqa Φk /ℏ) when the particle a is brought around
Ck . Thus the boundary conditions are different from the case of zero fluxes and one
should expect that the energy levels will depend on the values of Φk . Since charges of all
particles are multiples of the elementary electronic charge e the change in the boundary
conditions is the same for the fluxes Φk which differ by multiples of the flux quantum
Φ0 . This periodicity should occur in the solutions ψ ′ and therefore in the set of energy
levels obtained from Eq.(1.92). Physical quantities which are determined by this set
must therefore exhibit this periodicity. This conclusion as well as the entire set of the
preceding arguments are very general and based solely on the fundamental principles of
gauge invariance, requirement of single valued wave functions and the elementary nature
of the electric charge e.

1.8 Magnetic Moments

1.8.1 The g–factors
We now return to the relation (1.10) between the magnetic moment and the spin operators.
It is customary to quote the numerical value of the magnetic moment of a particle as equal
to the maximum value of its projection, i.e. the value µz = g(q/2mc)sz for sz = s. In
this Section we will discuss the dimensionless gyromagnetic ratio g in this relation called
the g-factor.
For elementary particles g is determined by relativistic quantum mechanical wave
equations. E.g. for the electron the Dirac equation gives g = 2, i.e. twice the classical
value. Unlike orbital angular momentum or the spin of a composite particle the spin of
an elementary particle has a fixed value and therefore its magnetic moment is fixed and
Version of April 11, 2023 43

must be regarded as one of the characteristics of the particle like its charge, mass, etc.
The electron magnetic moment (spin 1/2) is to a good approximation given by the Dirac
value6
|e|ℏ
µ0 = . (1.94)
2mc
This quantity is called the Bohr magneton and serves as a convenient unit in which
magnetic moments are measured in atomic physics. Its numerical value is 5.79 · 10−9
eV/Gauss.
In nuclear physics a more appropriate unit is the nuclear magneton defined as in
(1.94) but with the mass of the proton used for m. Experimentally measured value for
the magnetic moment of the proton is 2.793 nuclear magnetons meaning that the g-factor
is 5.586. For neutrons the values are –1.913 and –3.826 respectively. The deviation of
these g-factors from the corresponding Dirac values g = 2 and g = 0 was among the
first experimental indications that protons and neutrons are not elementary but rather
composite particles. In general the calculation of the g-factors for composite particles
requires the knowledge of the intrinsic dynamics, i.e. the wave function of the elementary
constituents, their spins, etc. We will consider examples of such calculations below.

1.8.2 Atoms in a magnetic field

The Hamiltonian
Consider an atom placed in a uniform magnetic field. Assuming fixed heavy nucleus
atomic electrons are described by the Hamiltonian7
1 X e 2 e X
H= pa + A(ra ) + U (ra ) + B· sa , (1.95)
2m a c mc a

where we denoted by ra , pa and sa the coordinates, momenta and spin operators of the
electrons and included in U (ra ) the interaction of the electrons with the atomic nucleus
as well as their Coulomb interaction with each other. We used q = −e for electrons and
for simplicity disregarded the nuclear spin.
Choosing the vector potential in the form A = (r × B)/2 for which ∇ · A = 0 we can
write the Hamiltonian in the form
e X e2 X 2 e X
H = H0 + B· (ra × pa ) + (B × ra ) + B · sa (1.96)
2mc a
8mc2 a mc a
e2 X
= H0 + µ0 (L + 2S) · B + (B × ra )2 , (1.97)
8mc2 a

where H0 is the Hamiltonian in the

Pabsence of the magnetic
P field, µ0 is the Bohr magneton
and we used the expressions L = a (ra × pa ) and S = a sa for the total orbital angular
6 Small deviations from this value are very accurately described in Quantum Electrodynamics by the

effects of the interaction with the surrounding cloud of virtual photons and electron–positron pairs.
7 In this and the following Sections we use CGS units
Version of April 11, 2023 44

momentum and spin. The terms in H which depend on the magnetic field can be written
as −µ · B with the operator of the magnetic moment

e2 X 2
µ = −µ0 (L + 2S) + [r B − ra (ra · B)] . (1.98)
8mc2 a a

The first term in this expression is independent of B and can be considered as the
operator of the intrinsic magnetic moment of the atom which exists in the absence of the
field. It is a sum of the orbital and the spin contributions in which the latter enters with
twice as large coefficient. It is crucial to observe that because of this non classical Dirac
value of the spin g–factor the intrinsic magnetic moment is not parallel to the system
total angular momentum J = L + S. As we will presently see this is the main reason why
in general the atomic g–factors do not have the universal classical value g = 1 but depend
on the state of the atom.
The second term in µ depends on B and must be regarded as the operator of P the mag-
netic moment which is induced by the magnetic field. Its magnitude −(e2 /8m2 c2 ) j Iij Bj
is proportional to the moment of inertia Iij = a m(ra,i ra,j − δij r2a ) which of is one of
P
the manifestations of the Larmor theorem.

Treating the B dependent terms perturbatively. LS and jj couplings

Exact diagonalization of the Hamiltonian (1.97) is not feasible even when the solutions in
the absence of the magnetic field are known. The standard way of treating this problem
is to use the perturbation theory with respect to the B–dependent terms. Let us start
with the linear term in (1.97). Because of the rotational symmetry the states of the atom
are characterized by the eigenvalues J(J + 1) of J2 and for non zero J are multiplets of
degenerate states which can be labeled by one of the projections of J. One must therefore
use degenerate perturbation theory and to lowest order diagonalize the perturbation H1 =
µ0 (L + 2S) · B in the subspace of each multiplet. The magnetic field breaks the rotational
symmetry and removes the multiplet degeneracies. The remaining axial symmetry of
rotations around the direction of B indicates that within each multiplet of the degenerate
states the correct combinations which diagonalize H1 are the eigenstates of the projection
of J on B. The energy shift of these states with respect to the unperturbed value is simply
the expectation value of H1 ,

∆E = µ0 B < α; J, M |Lz + 2Sz |α; J, M >= µ0 B < α; J, M |Jz + Sz |α; J, M > , (1.99)

where we have chosen the z-axis along the direction of B and denoted by α the additional
quantum numbers apart of J and its projection M which are needed in order to specify
an atomic state.
According to the Wigner–Eckart theorem, cf. Ref. [11], the matrix element of a compo-
nent of any vector operator between states of a multiplet with a given J is proportional to
the same matrix element of the same component of the operator J with the proportionality
constant which is independent of M. We can therefore write

∆E = µ0 gα,J B < α; J, M |Jz |α; J, M >= µ0 gα,J BM , (1.100)

Version of April 11, 2023 45

where the yet undetermined proportionality constant gα,J obviously represents the g–
factor of the atomic state. Finding explicit expression for gα,J requires further information
about the structure of |α, J, M > and can only be made in certain limiting cases.
If the interactions in atoms were the ordinary Coulomb forces the total orbital and
spin angular momenta and their projections ML and MS would be separately conserved
and in this case
∆E = µ0 B(ML + MS ) . (1.101)
In reality, however relativistic effects are important and produce the so called fine structure
of atomic levels. The main relativistic P effect turns out to be the presence in the atomic
Hamiltonian H0 of the spin–orbit term a Vso (|ra |) la · sa with Vso (r) proportional to r−1
times the derivative with respect to r of the atomic potential −Ze2 /r. When this term
is relatively weak (as happens for most atomic states) one can treat it as a perturbation
and diagonalize it separately within each degenerate multiplet of (2L + 1)(2S + 1) states
with given L and S.
The result is what is called the fine splitting of the multiplet into closely lying states
which have definite values of J. In this zero order of the perturbation treatment they are
linear combinations of the unperturbed wave functions with same values of L and S but
different ML and MS . Formally these zeroth order atomic states are written as
X
|n; L, S; J, M >= < L, ML ; S, MS |L, S; J, M > |n, L, ML > |S, MS >
ML +MS =M

and are referred to as states of the ”LS – coupling” scheme. By n we denoted here the
remaining quantum numbers for the orbital motion and the coefficients in the sum are
the standard Clebsh-Gordan coefficients for coupling of two angular momenta.
In the opposite extreme case of the strong spin–orbit interaction one can not talk about
separate conservation of the orbital and spin angular momenta. Individual electrons must
be characterized by their total angular momenta j which must be combined to produce
the total J. Such a scheme of constructing the zeroth order wave functions is called the
”jj – coupling”. This extreme limit is rarely found in atoms but plays a central role in
nuclear spectroscopy.

Lande formula
For states with LS – coupling a general expression for the g-factors called the Lande
formula can be derived,
J(J + 1) − L(L + 1) + S(S + 1)
g =1+ . (1.102)
2J(J + 1)
This is found as follows. As was already mentioned the Wigner – Eckart theorem gives
< S >= const· < J > (1.103)
where we use the angular brackets to denote averages with respect to the state |n; L, S; J, M >
. Since the operator J commutes with L and S it does not change the quantum numbers
of this state so we can write
< S · J >= const < J · J >
Version of April 11, 2023 46

with the same constant. Using < J · J >= J(J + 1) have

<S·J>
< Sz >= const M = M (1.104)
J(J + 1)

Using L · L = (J − S)2 = J · J + S · S − 2J · S and the properties of the LS – coupling

state we find that
1
< S · J >= [J(J + 1) − L(L + 1) + S(S + 1)] (1.105)
2
Collecting the results in Eq. (1.99) we obtain that ∆E is in the form (1.100) with gα,J
given by the Lande expression (1.102). As usual with the results of the perturbation
theory this formula is valid when ∆E are small as compared to the intervals between the
unperturbed atomic energy levels. In the present case these are the intervals due to the
fine structure splitting.

1.8.3 The Zeeman effect

The general phenomenon of the energy splitting of atomic levels in magnetic field is called
the Zeeman effect. The Lande formula gives the classical value g = 1 in the case S = 0
and the Dirac value g = 2 when L = 0. Historically the measured deviations of g from
the classical value 1 were termed the anomalous Zeeman effect. In the case when the
magnetic field is so intense that µ0 B is larger than the intervals of the fine structure the
energy splittings ∆E deviate from the predictions of the Lande formula. This is called
the Pashen – Back effect. We will not discuss the details of it.
Let us now turn to the last term in the Hamiltonian (1.97) which is quadratic and
describes as we already mentioned the interaction of induced magnetic moment with
the field B. This interaction is sometimes called diamagnetic to distinguish it from the
linear term which is called the paramagnetic interaction . The relative magnitude of
these two terms can be estimated as (e/ℏc) r2 B ∼ 4 · 106 (r/cm)2 B/Gauss and one
finds that for typical magnetic fields in laboratory the diamagnetic term is negligible
if r has atomic dimensions. However when an atomic state has zero spin and orbital
angular momentum (L = S = 0), the linear term does not effect the energy levels in any
order of the perturbation since it has vanishing matrix elements. In this case the entire
effect is determined by the quadratic term. In first order of the perturbation theory the
corresponding energy shift is

e2 X
∆E = < (ra × B)2 > , (1.106)
8mc2 a

where the average is with respect to a (non degenerate) state with L = S = 0. Since
< (ra × B)2 >= B 2 < ra2 sin2 θa > and since the wave function of a state with L = S = 0
is spherically symmetric one can average first over the angle and obtain < (ra × B)2 >=
2B 2 < ra2 > /3 where we used
Z
< sin θa >= sin2 θa 2π cos θa dθa /4π = 2/3
2
Version of April 11, 2023 47

Therefore
e2 X
∆E = 2
B2 < ra2 > (1.107)
12mc a

Having in mind the general expression −µ · B we see that the change of the induced
magnetic moment with the field in this case is negative which means that such a state is
diamagnetic.

1.9 Time Reversal in Magnetic Field. Kramers De-

generacy
In the absence of magnetic field and for spinless particles the Schrödinger equation with
a time independent Hamiltonian is invariant under the substitution t → −t provided one
also changes ψ → ψ ∗ . One adopts

ψ(r,t) → T ψ(r,t) ≡ ψ ∗ (r, −t) (1.108)

as a definition of the time reversal transformation in this case. Magnetic field and the
particle spin require modifications of this definition. Since magnetic field acts also on the
spin variables it is natural to discuss them together.
Even time independent magnetic field breaks the time reversal symmetry. This is
already known in classical physics. The equation of motion (1.4) is time reversal invariant
for any static E if B = 0. For non vanishing B(r) this symmetry is lost but one observes
that the equation retains its form if together with the sign change of t one changes the
sign of the magnetic field. Thus all solutions r(t) found in a given E(r) and B(r) must
have ”partners” in the form r(−t) in a related problem with E(r) and −B(r). Of course
one must take care in defining properly matched initial conditions for related solutions,
i.e. impose time reversed initial velocities. One easily understands why the sign of B
must be reversed – this is consistent with Maxwell equations which relate B to external
currents which change their direction under time reversal. Similar arguments make it
clear why E should stay the same.
Let us now turn to quantum mechanics in a static electromagnetic field. We first notice
that changing the sign of t and of B without changing E simply means that A → −A
together with t → −t. Transforming also ψ(r, t) → ψ ∗ (r, −t) in the Schrödinger equation
(1.12) we see that such a combined transformation leaves invariant all the terms in the
equation except for the last, spin dependent term which becomes geB · s∗ ψ ∗ /2mc rather
than −geB · sψ ∗ /2mc. By analogy with the orbital angular momentum one needs the
reversal of the sign of the spin operators and the complex conjugation does not accomplish
this. Indeed recalling the standard representation of the spin operators in terms of the
Pauli matrices,
     
ℏ 0 1 ℏ 0 −i ℏ 1 0
sx = sy = sz =
2 1 0 2 i 0 2 0 −1

one sees that (in this particular representation, in which sz is diagonal) the complex
conjugation causes only s∗x = sx , s∗y = −sy and sz = s∗z . Hence one must together with
Version of April 11, 2023 48

the complex conjugation also change the sign of sx and sz without changing sy . This
can be accomplished by the rotation by the angle π around the y–axis in the ”space” of
the spin variables. Simce s is the operator of infinitesimal rotations in this space such
a rotation is achieved by the operator exp [ iπsy /ℏ]. Accordingly, we generalize the time
reversal transformation of the wave function for particles with spin as

ψ(r,t) → T ψ(r,t) ≡ exp [ iπsy /ℏ]ψ ∗ (r, −t) , (1.109)

which must be supplemented with the sign change of A and B in the presence of the
magnetic field. Now of course all the terms in Eq.(1.12) will transform correctly. We
note that the transformation (1.109) (as well as the incomplete (1.108)) is antilinear, i.e.
T (αψ1 + βψ2 ) = α∗ T ψ1 + β ∗ T ψ2 and antiunitary, i.e. < T ψ|T ϕ >=< ψ|ϕ >∗ .
The transformation properties of any (possibly time dependent) quantum mechanical
operator Oop under time reversal are determined by considering
∗
T (Oop ψ) = exp [ iπsy /ℏ] Oop (−t)ψ ∗ (r, −t)

and comparing with T (Oop ψ) = (T Oop T −1 )(T ψ). We thus find

T r T −1 = r , T p T −1 = −p , T s T −1 = −s , (1.110)

in line with the intuition.

It is important to remember that the explicit form of the time reversal operator as
given above was derived in the particular representation, i.e. in the coordinate represen-
tation and diagonal spin projection sz . It is in general not valid in other representations,
but can be derived following the rules of transformations between representations. E.
g., a plane wave exp(ik · r) in the coordinate representation becomes δ(p − ℏk) in the
momentum representation for which the complex conjugation is obviously not producing
what expected under the time reversal – the change of the sign of k. Using the relation
between the coordinate and momentum representations we obtain
Z
< p | T ψ >= dr < p | r >< r |ψ(−t) >∗ =< −p | ψ(−t) >∗ , (1.111)

where we ignored the spin and used < p | r >=< −p | r >∗ . It is seen that the time
reversal in momentum representation is a combined action of the complex conjugation
and the change of sign of the momenta — not surprising.
Let us return to physical systems without external magnetic field. Their Hamiltonians
are symmetric under time reversal, [H, T ] = 0. For an eigenstate ψn of H this gives
HT ψn = T Hψn = En ψn which means that ψn and T ψn have the same energy. There are
two strong results which follow from this fact :
• for spinless particles non degenerate eigenstates of H can always be chosen to be
real and

• eigenstates with half–integer total spin are always at least doubly degenerate. This
degeneracy is called the Kramers degeneracy.
Version of April 11, 2023 49

To prove the first result we note that since for spinless particles T ψn (r) = ψn∗ (r) and
since by the assumption En is not degenerate the function ψn (r) must coincide with ψn∗ (r)
up to a constant independent of r. For normalized wave functions this is at most a phase
factor which is inessential for any physical results and can be disregarded.
In order to prove the second result consider an eigenstate wave function ψn = ψαjm
and its time reversed partner T ψαjm , where we denoted by jm the total spin of the system
and its projection and by α all other quantum numbers. Should these functions represent
the same state as in the spinless case they would be related as T ψαjm = Cψαjm with
some complex constant C. Applying T once again we would get T 2 ψαjm = |C|2 ψαjm .
But on the other hand T 2 = exp [ 2iπsy /ℏ] which gives (−1)2j when applied to ψαjm , cf.,
Problem 1. This can not be equal to a positive |C|2 for half–integer j. We are therefore
led to conclude that ψαjm and T ψαjm must correspond to different states for half–integer
j which means that the corresponding eigenvalue En is at least doubly degenerate. This
Kramers degeneracy means, for instance, that for a system with odd number of electrons
the energy levels will always be at least twofold degenerate even if it is placed in any,
however complicated, electric field

1.10 Path Integrals with the External Electromagnetic

Field
The derivation of the path integral quantization of a particle in the presence of the elec-
tromagnetic field follows the standard route. The propagator

K(rf , tf ; ri , ti ) ≡ ⟨rf |e−iHop (tf −ti ) |ri ⟩

is represented as a multiple integral

Z Z Z
K(rf , tf ; ri , ti ) = lim 3
d rN d rN −1 . . . d3 r1 K(rf , tf ; rN , tN ) ×
3
N →∞
× . . . × K(r1 , t1 ; ri , ti ) , (1.112)

over infinitesimal propagators which should be calculated for the Hamiltonian operator
given by Eq. (1.8) (we do not consider the spin dependent term – such terms require
special treatment in the path integral formulation). Based on the experience with path
integrals one should expect that the propagator K(r, t + ϵ; r′ , t) over an infinitesimal time
interval ϵ is expressed as
 m 3/2 i
exp{ ϵL[(r + r′ )/2, (r − r′ )/ϵ]}
2πiℏϵ ℏ
in terms of the classical Lagrangian L(r, v) given by Eq. (1.5). An explicit calculation
indeed shows that Z
ψ(r, t + ϵ) = dr′ K(r, t + ϵ; r′ , t) ψ(r′ , t) (1.113)
Version of April 11, 2023 50

reproduces the Schrödinger equation with the infinitesimal propagator given by

( "  2 #
iϵ m r − r′ r + r′
 m 3/2 
′
K(r, t + ϵ; r , t) = exp − eA0 +
2πiℏϵ ℏ 2 ϵ 2
r + r′
 
ie
+ (r − r′ ) · A (1.114)
ℏc 2

The details of this calculation are rather cumbersome and will not be reproduced here.
They can be found in Ref. [4].
Using the expression for the infinitesimal propagator in the multiple integral for
K(rf , tf ; ri , ti ) we find after combining the product of the exponentials into a exponential
of a sum and using the continuous notation
r(tZf )=rf  t 
 i Z2 
mv2
e

K(rf , tf ; ri , ti ) = D[r(t)] exp dt − eA0 (r) + A(r) · v (1.115)
ℏ 2 c 
r(ti )=ri t1

where as usual the definition of D[r(t)] includes the product of N d3 ri ’s each multiplied
by the pre-exponential factors from Eq. (1.114).
The first two terms in Eq. (1.115) are the usual kinetic and potential energies but the
last term is a new feature of this path integral. It mixes coordinates and velocities but its
linear dependence on v is special. Making the replacement v · dt = dr the contribution of
this term for every path in the path integration can be written as

Zrf
 
ie
exp  A[r(t)] · dr .
ℏc
ri
This phase factor depends on the path but not on the velocity of propagation along it.
If one considers a difference of these phases between two arbitrary paths one can write
it as the circulation of A along a closed path which is obtained by traversing from ri
to
H rf alongR one path and Rthen back to ri along the other. Using the Stokes theorem
A · dr = ∇ × A · dS = B · dS one can write this phase difference as
 
Φ
exp 2π i , (1.116)
Φ0

where Φ is the flux of the magnetic field through the closed contour defined by the two
paths and Φ0 = hc/e is the magnetic flux quantum already familiar from our discussions
of the Aharonov-Bohm effect.
We would like to point out an important subtlety related to the appearance of terms
like A(r) · v in the path integration. One will get different answers depending on whether
A(r) is evaluated at (r + r′ )/2, at r, at r′ or somewhere in between in the infinitesimal
propagator (1.114). This ambiguity is known as the Ito ambiguity and is discussed in
detail in Ref. [4]. It is shown there that the correct prescription is to take A as it is
written in Eq.(1.114), i.e. at a midpoint. This is sometimes referred to as the mid–point
Version of April 11, 2023 51

rule. Only with this rule the correct Schrödinger equation is reproduced. The mid–point
rule is important for a term A · v and not for the conventional potential term eA0 (r).
This is because of the different powers of ϵ, i.e. ϵ0 and ϵ1 which multiply the discretized
version of A · v and eA0 respectively in the expression for the infinitesimal propagator.
As is shown in standard discussions of the path √ integrals the typical distances between
propagation points obey the estimate |r − r′ | ∼ ϵ. Changes of this order of magnitude in
the argument of eA0 [(r + r′ )/2] combined with ϵ1 in front of it will contribute a negligible
difference of the order ∼ ϵ3/2 . The same change in A[(r + r′ )/2] combined with the term
r − r′ ∼ ϵ1/2 which multiplies it contributes O(ϵ1 ) which can not be neglected.
Let us now examine how the gauge transformations effect the path integral (1.115).
Performing a gauge transformation (1.3) of the potentials A and A0 adds in the action
the term proportional to
Z tf   Z tf
dr ∂χ(r,t) dχ(r,t)
dt · ∇χ(r,t) + = dt = χ(rf , tf ) − χ(ri , ti ) , (1.117)
ti dt ∂t ti dt

where the last equality holds because of the mid–point rule of the discretization of the
integral and gives the result which is the same for all paths. Using this we find that under
the gauge transformation the propagator changes as

K ′ (rf , tf ; ri , ti ) = exp[ieχ(rf , tf )/ℏc]K(rf , tf ; ri , ti ) exp[−ieχ(ri , ti )/ℏc] . (1.118)

This of course is of the same origin as the change of the phase of the wave function (1.14).
The phase change of K(rf , tf ; ri , ti ) depends only on the initial and the final coordinates.
The phase difference between different paths is strictly gauge invariant.
In a uniform electric and magnetic field the Lagrangian (1.5) is a quadratic function
of the coordinates and velocities and the path integral in this case is of the Gaussian type
and can be evaluated exactly (cf., Problem 2 at the end of the Chapter).

1.11 Dirac Magnetic Monopoles

1.11.1 Multivalued wave functions. Non integrable phases
An instructive discussion related in a surprising way to the general issue of the gauge
transformations arises when one examines in depth the requirement that the solutions
of the Schrödinger equation must be single valued. This requirement is usually imposed
as natural and is the main reason for finding the standard quantized values of physical
quantities such as the energy, the angular momentum, etc. Following the discussion by
Dirac 8 let us try to see what happens if this requirement is removed.
Of course one still must obtain unambiguous results for quantities which have direct
physical meaning. This certainly means that the amplitude of the wave function must
be single valued since its square is a physical density function. The phase of the wave
function on the other hand does not have to have a unique value at a particular point
so in general the wave function can be written as ψ(r,t) = ϕ(r,t) exp[iβ] with ϕ(r,t) the
ordinary single valued complex function and all multivaluedness residing in the properties
8 In this section we draw freely on the original paper of P.A.M. Dirac, Ref. [14].
Version of April 11, 2023 52

of the phase β. A useful way to characterize this multivaluedness is to consider how

β changes when one goes along some curve connecting two points in space–time. Since
ψ(r,t) satisfies the Schrödinger equation it must be continuous and therefore it is natural
to assume that β must have a definite derivative almost at every point (r,t). We will
discuss later the points where this does not happen.
The change of β along a curve
P R which does not pass through such singular points can
be expressed by the integral µ κµ dxµ taken along this curve with κi = ∂β(r,t)/∂xi
and κ0 = ∂β(r,t)/∂t. Since κµ in general do not satisfy the conditions of integrability
∂κµ /∂xν = ∂κν /∂xµ the value of this integral depends on the curve and in particular the
total change in the phase β need not vanish when the integral is calculated round a closed
curve. The values of such circulation integrals for all imaginable closed curves completely
characterize the multivalued properties of the non–integrable phase β.
We now show that in order to have unambiguous results for physical quantities any
such circulation integral must be the same for all the wave functions. Indeed probabil-
ities
R ∗ to measure physical quantities are given by squares of moduli of overlap integrals
ψm ψn d3 r with different wave functions ψm and ψn . In order that any such integral will
have a definite modulus the integrand, although it need not have a definite phase at each
point, must have a definite phase difference between any two points. Thus the change of
∗
phase of ψm ψn round a closed curve must vanish. This requires that the change in phase
in ψn round a closed curve shall be equal to that in ψm and since ψm is arbitrary it must
be a universal value for a given curve for all wave functions.
This result means that without loss of generality the possible non integrable phase
factor exp(iβ) in the wave function may be taken as universal for all wave functions. Let
us now consider the Schrödinger equation for ψ. Since
 
∂ iβ ∂
−iℏ ψ = e −iℏ + ℏκx ϕ (1.119)
∂x ∂x
with similar relations for the y, z and t derivatives one obtains that the single valued part
ϕ of the general wave function ψ satisfies the Schrödinger equation with gauge potentials
which are proportional to the derivatives of the non integrable phase β. In the most
common case these would have to be identified with the electromagnetic potentials
A = (ℏc/e)κ , A0 = −(ℏ/e)κ0 . (1.120)
We therefore conclude that multivalued wave functions need not be considered in quantum
mechanical description since they are equivalent to single valued wave functions in the
presence of an external gauge field.
Although this conclusion is certainly valid there are two ambiguities which remain in
the above discussion. The first is related to the Aharonov–Bohm effect and can occur in
multiply connected regions such as the inside of a ring as was already discussed in Section
1.7 above. In this case even for a vanishing electromagnetic field inside the region one
can not in general assume that the wave function must be single valued. If one can find
non contractable closed curves in the region one must first classify these curves according
to different homotopy classes as in Section 1.7. One may then assign an arbitrary but
fixed phase factor exp(iβk ) for every elementary homotopy class Ck and demand that only
solutions of the Schrödinger equation which change their phase by these assigned factors
are allowed.
Version of April 11, 2023 53

Intuitively one can interpret this situation by thinking about a multiply connected
region as a region with ”holes”. Even when the electromagnetic field vanishes inside the
region one can still have arbitrary magnetic fluxes ”in the holes”. These fluxes will give
rise to Aharonov - Bohm phases for closed curves surrounding the ”holes” provided these
curves can not be continuously deformed to a point. Hence assigning different sets of
phase factors exp(iβk ) for elementary classes of curves corresponds to assuming different
distributions of Aharonov-Bohm fluxes exp(2πiΦk /Φ0 ) in the ”holes”.
There exists another important ambiguity in the discussion of possible appearance of
non integrable phases in quantum mechanics. This was first observed by Dirac and is
related to the fact that although in the absence of the electromagnetic field in a singly
connected region the factor exp(iβ) can be taken as single valued the phase β itself may
change by an arbitrary integer multiple of 2π. Allowing for such changes requires a recon-
sideration of the connection between the derivatives κ of the non integrable phase β and
the electromagnetic potentials and leads to a new physical phenomenon – a possible exis-
tence of magnetic monopoles with quantized charges. We will now discuss this fascinating
subject.

1.11.2 Magnetic monopoles

The Maxwell equation ∇ · B = 0 means that there are no sources of the magnetic field, i.e.
that the magnetic charges do not exist in nature. However nothing conceptually wrong
should occur in the classical theory if one assumes a non zero ∇ · B = 4π σ with σ – the
density of magnetic charges. In fact the theory would be more symmetrical in this case
since a symmetry under the so called duality transformation E → B, B → −E would
then exist if one simultaneously exchanges the magnetic and the electric charges. The
non zero ∇ · B poses however a problem in quantum theory where the canonical or path
integral quantization in the presence of a magnetic field require an explicit introduction
of the vector potential via B = ∇ × A. Without this relation one is not able to define
the Hamiltonian or the Lagrangian of the theory but it is valid only for divergenceless B.
Let us analyze this problem more closely and consider a hypothetical point–like particle,
called magnetic monopole, which carries a magnetic charge g. In its presence
r − r0
∇ · B = 4πgδ(r − r0 ) , B = g , (1.121)
|r − r0 |3

where r0 denotes the position of the monopole and g is its magnetic charge.
For all points in space apart from an infinitesimal vicinity of r0 we have a divergence-
less B and can write B(r) = ∇ × A(r). Although correct locally the function A(r) is not
single valued. This is seen by considering the integral form of the relation B = ∇ × A,
i.e. the Stokes theorem, Z I
B · dS = A · dr
S C
where C is some closed curve in space and the integral on the left hand side is over
an arbitrary surface S with C as its boundary. Such an integral – the flux of B – is
however not unique in the present case. It does not change for all surfaces which can
be continuously deformed into each other without crossing the position of the monopole
Version of April 11, 2023 54

but once the surface crosses r0 the flux changes. The difference between the fluxes for
surfaces ”on both sides” of the monopole is equal to the total flux through the closed
surface which these two surfaces form. Integrating (1.121) over the volume inside this
surface and using the Gauss theorem one finds that this flux is equal to 4π g. The non
zero ∇ · B thus effects the definition of A globally and not just near r0 . Using the Stokes
theorem with continuously changed contour C as a way of continuous definition of the
relation between the functions A(r) and B(r) we will find two ”branches” of this relation
depending on ”which side” of r0 we choose the surface S.
There is a number of ways of overcoming this difficulty. Historically the first was
suggested by P. M. Dirac, Ref. [14]. We will follow a more modern way of presenting this
approach. The idea is somewhat similar to what is done in the theory of multivalued
analytic functions, i.e. to introduce a branch cut extending from a branch point. Viewing
the position of the magnetic monopole as analogous to such a branch point one can avoid
the ambiguity in the use of the Stokes theorem for determining the relation between A(r)
and B(r) if together with r0 a thin tube extending from it to infinity (or to another,
oppositely charged monopole) is excluded from the space. By excluding we mean that the
surface S for the contour C can never be chosen such that it is pierced by the tube. This
uniquely defines ”the side” of the monopole which one should choose to draw the surface
in the Stokes formula. One can thus assure the single valuedness of the B(r) → A(r)
relation everywhere in space apart from the inside of the excluded tube. We can make
the tube as thin as we like and send it in any direction.
Let us illustrate this discussion and consider an example of a monopole placed at the
origin and let us choose the excluded tube along a positive z axis. It is easy to verify that
the vector potential the curl of which gives the magnetic field (1.121) everywhere except
on the positive z can be chosen as

g (1 + cos θ)
Ar = Aθ = 0 , A ϕ = − (1.122)
r sin θ
where Ar , Aθ and Aϕ are spherical components of A and θ and ϕ are the conventional
polar and azimuthal angles. On the z axis this potential does not reproduce the field
(1.121) of the monopole but rather gives a singular magnetic field directed towards the
monopole and carrying a flux 4π g. The total effective magnetic field represented by the
curl of (1.122) is therefore
r
Bef f = g − g δ(x) δ(y) ez . (1.123)
r3
where θ(z) denote the step function.
We could choose another vector potential

g (1 − cos θ)
A′r = A′θ = 0 , A′ϕ = (1.124)
r sin θ
which also gives the required magnetic field (1.121) but with the excluded tube along the
negative z axis. The corresponding ”effective” field again has a singular component along
this tube in the direction of the monopole. The flux along the tube is equal to the total
flux of the first component of Bef f , i.e. the flux of the monopole field.
Version of April 11, 2023 55

Figure 1.7: Magnetic field of Dirac monopole including the singular string. Note that the
string is plotted as curved, which is allowed, but in the examples in the text it was chosen
to be a straight line along the positive z axis for simplicity, Eq. (1.123).

It is easy to understand now the logic behind the excluded tube construction. The
magnetic flux along the tube ”feeds” the radially directed field of the monopole so that
the resulting ”effective” field is divergenceless, ∇ · Bef f = 0 and can be represented as a
curl of a vector potential like the examples (1.122) and (1.124) above.
At this point a crucial question arises. We have replaced the desired magnetic field of
the monopole by the effective field with the flux tube. How does one make sure that this
modification has not changed the physics of the problem? Since the entire construction
was invented for quantum mechanics we must worry if the presence of the flux tube added
to the field of the monopole influences the solutions of the Schrödinger equation. In fact
we know that such a flux tube does have a global influence in the form of the Aharonov–
Bohm effect. It is also clear how to avoid this effect and make the flux tube unobservable
at large distances. One must demand that the flux carried by the tube is equal to an
integer multiple of the magnetic flux quanta,
4π g = nΦ0 = 2πnℏc/e
This imposes a quantization condition on the possible values of g,
1
eg = nℏc , n = ±1, ±2, . . . (1.125)
2
This relation is called the Dirac quantization condition. The unobservable flux tube car-
rying integer number of magnetic flux quanta is called the Dirac string. The entire
construction which we just described is called the Dirac monopole, cf., Fig. 1.7
The quantization condition (1.125) for the monopole charge implies that if there exists
a magnetic monopole anywhere in the universe all electric charges will be quantized:
e = n(ℏc/2g). Note that this quantization condition has an explicit dependence on the
Plank constant and therefore on the quantum theory.
Version of April 11, 2023 56

Experimental search for the presence of the magnetic monopoles in nature has so far
given negative results. We note that since g = (137/2) e the force between two monopoles
is (137/2)2 ∼= 4692 larger than between two electrons. This may mean perhaps that all
the monopoles in nature are tightly bound in pairs of opposite sign. In order to decide
whether this is true one needs to know the masses of the monopoles about which the
theory above gives no information.
In recent years another theory of magnetic monopoles was suggested by t ’Hooft and
Polyakov, Ref. [16], in the context of the so called non abelian gauge theories with broken
symmetry. This theory predicts that the mass of the monopoles should be very large.
Viewed from large distances both Dirac and non abelian monopoles should look exactly
the same and our discussion of quantum mechanics in the field of monopoles is expected
to remain valid at such distances.
There is another way to introduce magnetic monopoles in quantum mechanics which
avoids the appearance of the singular Dirac string. It was proposed by Wu and Yang,
Ref. [15], and adopts an approach of sections similar to what is done by mapmakers when
they map the spherical surface of the earth onto a plane map. A single map would
obviously have a singularity at one point. Indeed imagine a rubber sheet with rectangular
coordinate grid on it and try to wrap it around the globe. In order to avoid the singularity
of a single map two maps are introduced, one for say a northern hemisphere and one for
the southern. The two maps together form a singularity–free mapping of the earth. In
order to be able to pass smoothly from one map to another one should let each to cover
more than its own hemisphere so that an overlap is created in the region of the equator. In
this overlapping region the coordinates of both maps must be in one to one correspondence
for identical points of the globe surface. In a similar way singularity-free vector potential
can be found for a magnetic monopole.
The emerging formulation is in essence the so called fiber bundle formulation of gauge
fields in quantum mechanics. We will not go into this here referring the reader to the
literature.

1.11.3 Angular momentum and rotational symmetry in the pres-

ence of a monopole
Although the magnetic field of the monopole is spherically symmetric it should be in-
tuitively clear that the Lorenz force acting perpendicular to the velocity of a moving
particle will not conserve the ordinary expression r × mv for the angular momentum
relative to the origin. Take, e.g., a particle which starts along a planar circular orbit
around the monopole. The magnetic field will deflect it away from the plane changing
the initial r × mv. Using the equation of motion one can calculate the rate of change of
this expression
d d eg d  eg r 
(r × mv) = r × m v = 3 (r × (v × r)) = . (1.126)
dt dt cr dt c r
This suggests that we can define the total angular momentum as
 eg r  eh ri
J = r × mv − =r×p− r×A+g , (1.127)
c r c r
Version of April 11, 2023 57

so that it is conserved.
The appearance of unusual terms in the expressions of conserved quantities in the
presence of electromagnetic field should already be familiar from the expressions of mo-
menta in uniform electric and magnetic fields, Eqs. (1.45) and (1.74). In addition to the
generators of the symmetry one must include the generator of the gauge transformation
which is needed to keep the formulation in the ”same” gauge. One can see this explicitly
by considering infinitesimal rotation r → r′ = r + δϕ × r and correspondingly

A(r) → A′ (r) = A(r − δϕ × r) + δϕ × A(r) → A(r) + δϕ × A(r) − [(δϕ × r) · ∇] A(r)

For A(r) which corresponds to the spherically symmetric magnetic field of the magnetic
monopole the last term in this expression must be a gauge transformation, i.e. equal to
a gradient of a scalar function, ∇ξ(r). One finds
 r
ξ(r) = −δϕ · r × A(r) + g ,
r
The transformation of the wave function under rotation is therefore
i e
[1 + δϕ · (r × p)] [1 + i ξ(r)] ψ(r) .
ℏ ℏc
In the brackets of the expression for ξ one finds just the two terms which must be added
(after multiplication by e/c) to the canonical r × p in order to obtain the conserved
Eq.(1.127).
There exists another, quite different interpretation of the last term in the expression
(1.127). It is the angular momentum contained in the electromagnetic field which exists in
the space surrounding the moving particle and the fixed monopole. Using the expression
for B of the monopole and E = e(r − r0 )/4π|r − r0 |3 for the electric field of the particle
at r0 one indeed finds

egr0 r2 (r̂ cos θ − r̂0 )

Z Z
Lem = d3 r r × (E × B) = d3 r 3 2 (1.128)
4π r (r + r02 − 2 r r0 cos θ)3/2

where θ is the angle between r and r0 . Straightforward evaluation of the integral gives
−eg r0 /|r0 | , which for eg = ℏc/2 gives the last term in (1.127). The physical picture
behind this interpretation of the additional term in L is obscure to the present author.

1.12 Non Abelian Gauge Fields

In Section 1.3.3 we discussed how the existence of the electromagnetic field could be
predicted by demanding that a global symmetry of the free Schrödinger equation becomes
local, i.e. by ”gauging” this symmetry. We give now an example of what happens when a
more complicated non abelian symmetry is gauged leading to a concept of a non abelian
gauge field, Ref. [17]. Let us assume that particles in our theory in addition to spin carry
another discrete intrinsic dynamical variable τ which we will tentatively call pseudospin
and which may take two values, τ = 1, 2. In analogy with the spin variables the wave
functions will now carry an additional index τ so that | ψτ |2 integrated and summed over
Version of April 11, 2023 58

all other variables ( r, spin, etc.) gives the probability to measure this particular value of
τ . We also have to introduce operators which act on the variables τ and in terms of which
we shall represent all observable quantities involving this variable. These operators must
be hermitian 2×2 matrices. One can write any such operator Oαβ as a linear combination
of a unit and Pauli matrices with real coefficients
3
(a)
X
O = O0 δαβ + Oa ταβ , (1.129)
a=1
     
(1) 0 1 (2) 0 −i (3) 1 0
τ = ,τ = ,τ = ,
1 0 i 0 0 −1
since the Pauli matrices together with a unit matrix represent a complete set for expanding
any 2 × 2 matrix cf., Density Matrix chapter in the notes of this course.
Now let us consider a free Schrödinger equation

∂ ℏ2 2
iℏ ψτ (r,t) = − ∇ ψτ (r,t)
∂t 2m
for a particle with theP
pseudospin. This equation is obviously invariant under a linear
transformation ψ ′τ = τ ′ Sτ τ ′ ψτ ′ with an arbitrary matrix S. In order to preserve the
normalization of ψ the matrix S must be unitary. Its general form therefore must be an
imaginary exponential of an arbitrary 2 × 2 hermitian matrix and therefore can be written
in terms of the Pauli matrices as
!
X
(a)
S = exp iφ0 + i φa τ . (1.130)
a

with arbitrary real φ ’s. For simplicity we will limit ourselves to the transformations
with φ0 = 0 which is equivalent to imposing the condition det S = 1 on the allowed
matrices S. We say that the Schrödinger equation is invariant under the global SU(2)
transformations , i.e. under the transformations which belong to the group SU(2) of all
unitary 2 × 2 matrices with unit determinant. This group is non abelian – two arbitrary
SU(2) matrices in general do not commute. Let us now employ The Gauge Principle of
Section 1.3.3 for this SU(2) symmetry and demand that our theory must be not only
globally but also locally invariant under the above SU(2) transformations. This means
that transformations with matrices S having their parameters ϕa as arbitrary functions of
r and t should leave the Schrödinger equation invariant. The way to achieve the invariance
under such local gauge transformations is to introduce the gauge field potentials which
will compensate for the derivatives of S when the transformed

ψ ′ (r,t) = S(r,t)ψ(r,t) (1.131)

is inserted in the Schrödinger equation. Since the derivatives of the matrix S are ob-
viously also matrices the compensating potentials should be matrices. Since there are
four derivatives ∂/∂t and ∇ = (∂/∂x, ∂/∂y, ∂/∂z) in the Schrödinger equation one must
Version of April 11, 2023 59

introduce four such matrix compensating potentials A(µ) , µ = 0, 1, 2, 3, 4. They can be

represented as linear combinations
3
X τ (a) τ
A(µ) (r,t) = A(µ)
a (r,t) = A(µ) · , (1.132)
a=1
2 2

where we employed obvious short hand notation for the sum of products of arbitrary real
(µ)
functions Aa (r,t) and Pauli matrices τ (a) and introduced the factor 1/2 to follow the
conventional definitions in this field. In the fixed basis of τ (a) ’s to represent a matrix A(µ)
(µ)
is equivalent to giving three functions Aa .
In analogy with electromagnetism we introduce now gauge covariant derivatives
∂ ig τ
D0 = + A(0) · ,
∂t ℏ 2
ig τ
D = ∇− A· , (1.133)
ℏc 2
and use them in the Schrödinger equation in place of the ordinary derivatives ,

ℏ2 2
iℏD0 ψ(r,t) = − D ψ(r,t) . (1.134)
2m
The constant g introduced here is analogous to the electric charge e in the electromag-
netism. It determines the strength of the coupling of the particle described by this equa-
(µ)
tion to the non abelian gauge fields Aa . In order to achieve the invariance of the equation
under the local gauge transformations (1.131) we demand that D0 ψ and Dψ have the same
transformation properties as ψ itself, i.e.
   
∂ ig τ ∂ ig τ
D0 ′ ψ ′ ≡ + A(0) ′ · ψ ′ = SD0 ψ ≡ S + A(0) · ψ (1.135)
∂t ℏ 2 ∂t ℏ 2
   
ig τ ig τ
D ′ψ ′ ≡ ∇ − A ′ · ψ ′ = SDψ ≡ S ∇ − A · ψ . (1.136)
ℏc 2 ℏc 2

This is obviously a sufficient condition for the invariance of Eq.(1.134) and determines
the transformation properties of the gauge potentials
 
τ τ i ∂S(r,t) −1
A(µ)′ (r,t) · = S(r,t)A(µ) (r,t) · S −1 (r,t) − S (r,t) . (1.137)
2 2 g ∂x(µ)

We see that under a gauge transformation each matrix gauge potential is locally ”rotated”
at every space–time point by the gauge transformation matrix S(r,t) and at the same time
it is shifted by an amount which depends on the corresponding derivative of S(r,t). This
expression as well as the relations above are valid for any unitary group SU(N) with the
appropriate generalization of the transformation matrix S and the Pauli matrices τ (a) . For
the abelian group U(1) we will obviously recover the known Schrödinger equation and the
gauge potentials of the electromagnetic field. In general there will be d × (N 2 − 1) gauge
potentials with d = 4 – the dimensionality of the space–time and (N 2 −1) – the number of
Version of April 11, 2023 60

the independent generators of the group SU(N). The gauge freedom expressed by (1.137)
means that in general only (d − 1) × (N 2 − 1) combinations of the gauge potentials are
independent. We finally remark that among the gauge fields known in nature the unified
electromagnetic and the weak interactions (often called electroweak) are described by
U (1) × SU (2) and the strong interactions by SU(3) non abelian gauge potentials. The
intrinsic quantum numbers for these interactions (analog of what we called pseudospin)
are the standard electric charge and the so called weak isospin and the color respectively.
Since as already mentioned the gravitational field is also a gauge field we have all four
basic interactions described by the gauge fields.

Problems
1. Consider the spin part of the time reversal operator U = exp[i2π sy /ℏ].
• Show that U = -1 when applied to the wave function of a spin 1/2 particle.
• Components ψjm of a spin wave function with a general spin j can be consid-
ered as far as their transformation properties are concerned as suitably chosen
components of the wave function of a system of 2j spin 1/2 particles. Use this
to prove that U ψjm = (−1)2j ψjm .
2. Find the propagator in a uniform electric and magnetic fields by evaluating the
appropriate Gaussian path integrals, cf., [5]

3. Electrons are confined to move in a plane (x, y) and are placed in a uniform magnetic
field perpendicular to the plane.
(a) Consider two different gauges choices a) Ax = −By , Ay = Az = 0 and b)
Ax = − 12 By , Ay = 12 Bx , Az = 0. electron eigenfunctions calculated in Which
combinations of the guiding center coordinates do they diagonalize? How are
these two sets of eigenfunctions related? Calculate and explain the behaviour
of the current density in each of the above cases. What is the total current?
How the current will change in the case a) above if one adds a uniform electric
field along the x direction? along the y direction? What is the total current
now? In which direction does it flow? In case b) above assume that a very
thin solenoid with magnetic flux Φ is added along the z axis (at x = y = 0).
What and how will it influence? Consider your answer for various values of the
Φ and see if there are special values of ϕ.
In addition they are su a potential U = 21 αx2 .
a) How the Landau levels are changed by this potential? What is the current density
in a single state of a Landau level as compared to the case with U = 0?
b) Now repeat for U = 12 α(x2 + y 2 ).

2. Show that the operators xo and yo of the guiding center coordinates are generators
of the translations in the presence of the uniform magnetic field.
Version of April 11, 2023 61

3. Consider wave functions of Landau levels with definite values of xo . How are
they related to the similar wave functions with yo ? Consider now wave functions
with definite x2o + y02 . Assume that a very thin solenoid with magnetic flux ϕ?
4. Consider quantum mechanics in a strong magnetic field.
[Link] projection on a lowest Landau level (LLL). Show that the eigenstates of x0
and y0 are exactly equivalent to eigenstates of p and x in a one dimensional quantum
mechanics.
[Link] semiclassical approximation in the limit of ℓ → 0.
c. Find semiclassical energy levels of two interacting electrons in 2 dimensions in a
strong magnetic field. Discuss also the case of oppositely charged particles (say an
electron - positron system or electron-hole system in a solid state).
“
Bibliography

[1] D. J. Griffith, Introduction to Electrodynamics, 4th edition, Cambridge University

Press.
[2] Ta-Pei Cheng and Ling-Fong Li, Gauge theory of elementary particle physics,
Clarendon Press, Oxford, 1984. This and the following reference are fairly advanced
quantum field theory courses which however contain several chapters which could
be read with a profit even by a less advanced student.
[3] Lewis H. Ryder, Quantum Field Theory, Cambridge University Press, 1985.
[4] L. S. Schulman, Techniques and Applications of Path Integration, Wiley, New York,
1981. Section 4 and 5 contain a pedagogical discussion of the subtleties of path
integration with vector potentials and of the related ”Ito Integral”. Another version
of the derivation is found in L.S. Schulman, Selected Topics in Path Integration,
in Lectures on Path Integration: Trieste 1991, H.A. Cerdeira et al, Eds.,World
Scientific, Singapore.
[5] Hagen Kleinert, Path Integrals in Quantum Mechanics, Statistics, and Polymer
Physics, World Scientific, Singapore, 1989. This volume contains several sections
with discussions of path integrals in the presence of magnetic field.
[6] Y. Aharonov and D. Bohm, Phys. Rev., 115 (1959) 484 – this is the original publi-
cation in which phenomena later called Aharonov–Bohm effect were described. The
effect was first experimentally seen by R. G. Chambers, Phys. Rev. Lett. 5 (1060)
3. A very lucid discussion of it appears in T. T. Wu and C. N. Yang, Phys. Rev.
D12 (1975) 3845.
[7] Our discussion of the Aharonov-Bohm version of the two slit experiment is based on
R. P. Feynman, R. B. Leighton and N. Sands, The Feynman Lectures on Physics,
vol.2, Sec. 15.5, Reading, Massachusetts: Addison-Wesley Publishing, 1964.
[8] Palash B. Pal, A Physicist’s Introduction to Algebraic Structures, Ch. 22, Cam-
bridge University Press, 2019.
[9] F. Bloch, Phys. Rev., 2 (1970) 109. Extremely clear and beautifully argumented ”old
master’s” paper exploring in the context of the Josephson effect in a superconducting
ring the most general consequences of quantum mechanics in multiply connected
regions in the presence of Aharonov–Bohm fluxes.

62
Version of April 11, 2023 63

[10] Properties of the Airy function can be found in, e.g. M. Abramovitz and I. A.
Stegun, Handbook of Mathematical Functions. Washington, DC: National Bureau
of Standards.
[11] J. J. Sakurai, Modern Quantum Mechanics (Revised Edition) , Addison-Wesley, p.
239

[12] Ho-Fai Cheung et al, Phys. Rev. B37 (1988) 6050

[13] B. I. Halperin, Phys. Rev. B25 (1982) 2185
[14] The first paper on magnetic monopoles, P. A. M. Dirac, Proc. Roy. Soc. A133
(1931) 60 was in fact an investigation of a more general subject of justification of
the demand that wave functions are single valued. It is strikingly modern, highly
readable and is reprinted in a more accessible Monopoles in Quantum Field Theory,
Proceedings of the Monopole Meeting, Trieste, Italy, December 1981, N. S. Craigie,
P Goddard and W. Nahm, Eds., World Scientific. The continuation of the first paper
followed in P. A. M. Dirac, Phys. Rev. 74 (1948) 817.

[15] T.T. Wu and C.N. Yang, Nucl. Phys., B107 (1976) 365
[16] G. tH́ooft, Nucl. Phys., B79 (1974) 276; A.M. Polyakov, JETP Letters, 20 (1974)
194.
[17] C. N. Yang and R. L. Mills, Phys. Rev., 96 (1954) 191. This is the original paper
in which the non abelian generalization of the electromagnetism, later called the
Yang–Mills field, was first proposed. In Chapter 3.5 of Ref. [3] a pedagogical review
of the Yang–Mills fields is presented.
Chapter 2

Quantum Mechanics of
Electromagnetic Field

In this Chapter we will show how the quantum mechanical description of one or several
particles is extended to the quantization of electromagnetic field. In contrast to particles
which are described by the coordinates of their positions ra , a = 1, ..., N (N- the number
of particles) the electromagnetic field is described by the configuration of the electric and
magnetic fields E(r) and B(r). In order to learn how such extended systems are treated in
quantum mechanics we shall start with a much simpler system - that of a one dimensional
string.

2.1 Simple System First - Quantum Mechanics of a

Guitar String
2.1.1 Classical string
We consider a string depicted in Fig. 2.1

64
Version of April 11, 2023 65

Figure 2.1: Configurations of a guitar string. Denoting the abscissa of the figure (the
equilibrium position of the string) by x and the ordinate (the string deviations from the
equilibrium) by ϕ the string configurations are described by a function ϕ(x).

Classically its general configuration is conveniently described by a function ϕ(x) which

determines the deviations ϕ of the string from the equilibrium position ϕ = 0 at every
point of the axis x. For simplicity we shall assume that the ends of the string are fixed
at x = 0 and x = L
ϕ(0) = 0 , ϕ(L) = 0 . (2.1)
In the following section we will extend our discussion to a more relevant example of a
string with periodic boundary conditions - the so called closed string.
We will assume that classically the string is described by a simple linear wave equation
∂ 2 ϕ(x, t) ∂ 2 ϕ(x, t)
2
= v2 (2.2)
∂t ∂x2
where v has dimensionality of velocity and is actually the phase (as well as group) velocity
of the waves
ϕ(x, t) = A sin[kx ± ω(k)t + α] .
These are solutions of the wave equation as can be easily verified by direct substitution.
Here A and α are arbitrary constant amplitude and phase and the dispersion relation is
ω(k) = vk . (2.3)
It will be very useful to view the function ϕ(x) as a collection of the coordinates describing
the ”position” of the string. To emphasize this one might think of ϕ(x) as a set {ϕx }
indicating that x is actually an index numbering different coordinates. To make it even
more precise the x variable can be discretized and ϕ(x) reduced to N + 1 variables as
follows
ϕx ≡ ϕ(x = n∆x) , ∆x = L/N , n = 0, 1, ..., N
Version of April 11, 2023 66

Formally one should at the end let N → ∞, ∆x → 0 but in practice it is enough to

have ∆x much smaller than the smallest wave length Λ of the waves which one intends to
consider. The physical reasons behind the cutoff Λ may actually be the requirement that
it is much larger than the microscopic length scales related to say the distances between
the constituents of which the string is built.

2.1.2 Quantum description of the string

The wave functional
Our goal is to quantize the classical string as described above. We shall use the the
straightforward generalization of the canonical quantization procedure for system with
few degrees of freedom like one or several particles. This means that instead of having
a definite ϕ(x) describing the string configuration we must assume that for each ϕ(x)
there is a probability amplitude Ψ[ϕ(x)] which contains all the (quantum) information
about the string. In particular |Ψ[ϕ(x)]|2 gives the probability density to find a particular
configuration ϕ(x).
Mathematically Ψ[ϕ(x)] represents a correspondence between the set of all functions
ϕ(x) subject to the conditions Eq. (2.1) and a set of complex numbers Ψ. Such a corre-
spondence is called a functional. Examples of functionals should be familiar to the reader
already from classical mechanics where the classical action S[q(t)] is a functional of the
trajectories q(t).
The functional Ψ[ϕ(x)] is called the wave functional. Using the approximate discretized
form ϕx of the functions ϕ(x) the functional Ψ[ϕ(x)] can actually be viewed as a function
of N − 1 variables {ϕx=n∆x }. The variables ϕ0 and ϕx=N ∆x are fixed to 0 to comply with
Eq. (2.1).
Classically string dynamics is described by the time dependence ϕ(x, t) as governed by
the equation (2.2). Quantum mechanical time evolution should be described by the time
dependence of the wave functional Ψ[ϕ(x), t]. What governs this time evolution? Contin-
uing the analogy with the few degrees of freedom system this should be the Schrödinger
equation
∂Ψ[ϕ(x), t]
iℏ = Hop Ψ[ϕ(x), t] (2.4)
∂t
with Hop the Hamiltonian operator of the string. We will now determine this operator
following the standard route.

The string Hamiltonian

We will start by finding the classical Hamiltonian function of the string. For this we shall
rewrite the string equation (2.2) in the Hamiltonian form. It is useful to note that this
equation represents a set of coupled Newton equations for the string coordinates ϕx . This
can be seen by rewriting it in a discretized form

d2 ϕx v2
= (ϕx+∆x − 2ϕx + ϕx−∆x ) (2.5)
dt2 ∆x2
Version of April 11, 2023 67

where we used the discretized form of the second derivative

∂ 2 ϕ(x)
 
1 ϕx+∆x − ϕx ϕx − ϕx−∆x
→ −
∂x2 ∆x ∆x ∆x

We will rewrite the second order in time wave (Newton) equation of the string as a
pair of first order equations

∂ϕ(x, t) ∂π(x, t) ∂ 2 ϕ(x, t)

= π(x, t) , = v2 (2.6)
∂t ∂t ∂x2
where (as usual) the first equation is actually the definition of the momenta. As a next
step let us introduce the following functional
Z L " 2 #
v 2 ∂ϕ(x)

1 2
H[π(x), ϕ(x)] = dx π (x) + (2.7)
0 2 2 ∂x

Using it we can write the pair (2.6) as

∂ϕ(x, t) δH[π(x, t), ϕ(x, t)] ∂π(x, t) δH[π(x, t), ϕ(x, t)]
= , =− . (2.8)
∂t δπ(x, t) ∂t δϕ(x, t)

Here the notation δ/δπ(x, t) and δ/δϕ(x, t) stands for the variational derivative (see below)
with respect to π(x, t) and ϕ(x, t) respectively. We now show that the above equations
are indeed equivalent to the pair (2.6) recalling in passing how the variational derivatives
are defined and calculated.
We let the functions π(x) and ϕ(x) in the functional (2.7) to have infinitesimal varia-
tions δπ(x) and δϕ(x). The corresponding variation δH due to this is

δH ≡ H[π(x) + δπ(x), ϕ(x) + δϕ(x)] − H[π(x), ϕ(x)] =

Z L  
2 ∂ϕ(x) ∂δϕ(x)
= dx π(x)δπ(x) + v + higher order terms
0 ∂x ∂x
Z L 
∂ 2 ϕ(x)

= dx π(x)δπ(x) − v 2 δϕ(x) + higher order terms
0 ∂x2

where we used integration by parts in the second term.

In analogy with the relation of the differential dF of a function of many variables
F (q1 , q2 , ..., qN ) and its partial derivatives
N
X ∂F
dF = dqn
n=1
∂qn

the functional derivatives of H[π(x, t), ϕ(x, t)] are by definition the functions which mul-
tiply δπ(x, t) and δϕ(x, t) respectively in the expression for the variation δH,

δH[π(x), ϕ(x)] δH[π(x), ϕ(x)] ∂ 2 ϕ(x)

= π(x) , = −v 2 (2.9)
δπ(x) δϕ(x) ∂x2
Version of April 11, 2023 68

Inserting these relations into Eq. (2.8) we see that they indeed reproduce Eq. (2.6).
Equations (2.8) have the Hamiltonian form with H[π(x), ϕ(x)] as the Hamiltonian
and π(x), ϕ(x) as the momenta and coordinates. It should perhaps be more clear if for
a moment we use the notation πx and ϕx instead of π(x) and ϕ(x) treating x as a label.
The equations (2.8) in these notations are
∂ϕx (t) ∂H[πx (t), ϕx (t)] ∂πx (t) δH[πx (t), ϕx (t)]
= , =− .
∂t ∂πx (t) ∂t ∂ϕx (t)

Basic quantum operators for the string

We shall now proceed to define the quantum mechanical operator Hop . We will do this
by first determining what are the operators corresponding to π(x) and ϕ(x) and then
replacing with them the latter in the expression (2.7) for the classical Hamiltonian.
Since in our formulation ϕ(x) are the coordinates of the string the corresponding
operator ϕop (x) should be just the operator of multiplication by ϕ(x), i.e. its action on
an arbitrary wave functional is
ϕop (x)Ψ[ϕ(x′ )] = ϕ(x)Ψ[ϕ(x′ )] (2.10)
To avoid confusion we use different arguments of ϕ’s in the operator and in Ψ. This would
perhaps be easier to understand if (again momentarily) we shall switch to the notation
ϕx′ instead of ϕ(x′ ). Then the functional Ψ[ϕ(x′ )] is just a function Ψ({ϕx′ }) of the set of
all ϕx′ variables. The action of the operator ϕ̂x , i.e. the operator of the x − th component
the coordinates of the string is just a multiplication by ϕx with this particular x. Note
that in order to avoid the double subscript we here used ϕ̂ to denote the operator.
In the same way we can determine the operator corresponding to the momentum π(x).
In the ”simplified” notations it should be π̂x = −iℏ∂/∂ϕx which means that in terms of
the functional derivatives it is
δ
πop (x)Ψ[ϕ(x′ )] = −iℏ Ψ[ϕ(x′ )] (2.11)
δϕ(x)
We note that the commutator of the basic operators is
[ϕop (x), πop (y)] = iℏδ(x − y) (2.12)
This can verified by acting with the commutator on an arbitrary wave functional
 
′ δ ′ δ ′
[ϕop (x), πop (y)]Ψ[ϕ(x )] = −iℏ ϕ(x) Ψ[ϕ(x )] − ϕ(x)Ψ[ϕ(x )] =
δϕ(y) δϕ(y)
 
δ ′ δϕ(x) ′ δ ′
= −iℏ ϕ(x) Ψ[ϕ(x )] − Ψ[ϕ(x )] − ϕ(x) Ψ[ϕ(x )] =
δϕ(y) δϕ(y) δϕ(y)
= iℏδ(x − y)Ψ[ϕ(x′ )]
Substituting the operators ϕop (x) and πop (x) in the Hamiltonian (2.7) we obtain
Z L " 2 #
v 2 ∂ϕop (x)

1 2
Hop = dx πop (x) + (2.13)
0 2 2 ∂x
Version of April 11, 2023 69

Armed with this explicit form of the Hamiltonian operator of our system we can pro-
ceed to solve the Schrödinger equation (2.4). Since the Hamiltonian is time independent
it will be sufficient to solve the stationary equation

Hop Ψ = EΨ (2.14)

Knowing all its solutions will allow to find the most general solution of (2.4).
The string Hamiltonian operator (2.7) may look formidable but is actually quite simple
because of its quadratic dependence on the coordinates and momenta. This of course is
a direct consequence of the linearity of the string equation (2.2). As is seen from the
discretized form (2.5) such equations describe coupled harmonic oscillators. The standard
way of solving such problems is to make a transformation to normal modes.

2.1.3 Reminder - normal modes of vibrations

Let us recall how the transformation to normal modes is done in the general context
represented by the set of N coupled equations
N
X
ml q̈l = − kln qn , l = 1, ..., N (2.15)
n=1

with masses ml and N by N symmetric matrix of elastic constants kln = knl . For
simplicity we shall assume in the following that all the masses are equal m1 = ... = mN =
m. The Hamiltonian of this problem is the standard sum of the kinetic and potential
energies  
N N
1  X X
H= p2l + mkln ql qn  (2.16)
2m
l=1 l,n=1

Let us try the following solution of the equations (2.15)

ql = Re(Cl eiωt ), l = 1, ..., N

where Re stands for real part and Cl ’s are constants. This form assumes that all the
degrees of freedom vibrate with the same frequency. Inserting this into the equations
(2.15) we obtain
XN
(kln − mω 2 δln )Cn = 0 (2.17)
n=1

where we remind that we set for simplicity all mi = m. To have a non trivial solution
one must demand that
det(kln − mω 2 δln ) = 0 (2.18)
which shows that ω 2 is an eigenvalue of the matrix kln /m which in turns means that in
general one will have N such solutions which will have ω 2 > 0 provided kln is positive
definite.
Let us denote by ων and {Cnν } the set of N solutions of Eq.(2.17). The symmetry of
kij assures orthogonality of the eigenvectors {Cnν }’s with different eigenvalues ων . For a
Version of April 11, 2023 70

degenerate case i.e. if some ωµ = ων one has a freedom to choose {Cnµ } and {Cnν } to
ensure that orthogonality holds also in this case. We also note that since the equations
(2.17) are homogeneous at least one of the components in a given vector {Cnν } is arbitrary
and can be used to set normalization of {Cnν }’s to unity. We thus have orthonormality
N
X
Cnµ Cnν = δµν
n=1

The N vectors {Cnν }each with N components form an N × N matrix. The orthonormality
conditions (together with completeness which we do not discuss) mean that this matrix is
orthogonal (unitary for complex Cn ’s). Let us use it to make the transformation to new
coordinates
XN
qn = Cnν Qν (2.19)
ν=1
Inserting this in Eqs. (2.15) (with mi = m) and using Eqs. (2.17) one obtains
N N N
[mClν Q¨ν + mClν [Q¨ν + ων2 Qν ] = 0
X X X
kln Cnν Qν ] = (2.20)
ν=1 n=1 ν=1

Due to orthogonality of C’s one finds that the equations for Qν ’s are decoupled. Indeed
multiplying by Clµ and summing over l one obtains

Q¨µ + ωµ2 Qµ = 0 , µ = 1, ..., N . (2.21)

The transformation (2.19) from the original coordinates ql to the new Qν is called the
transformation to normal modes and Qν – the normal mode coordinates.
How does the Hamiltonian look in the new coordinates? To answer this we need to
add to (2.19) also the transformation to the corresponding normal modes momenta. Since
in our case pl = mq̇l it is clear that p’s transform like q’s
N
X
pn = Cnν Pν (2.22)
ν=1

Inserting this and (2.19) into the Hamiltonian (2.16) we obtain using the orthonormality
of Cnµ ’s and equations (2.17)
∞
1X 2
H= [P + ωµ2 Q2µ ] (2.23)
2 µ=1 µ

where for simplicity we have set m = 1. In normal mode variables the Hamiltonian is just
a collection of independent oscillators.

2.1.4 String as a collection of decoupled oscillators

Normal modes of the guitar string
We will now use the technique described in the previous subsection to transform the
string Hamiltonian to a collection of independent oscillators. We are looking for the
Version of April 11, 2023 71

analog of the transformation (2.19) from the string coordinates ϕ(x) to the normal modes
coordinates. Since x here plays the role of the index n in qn the analog of the matrix Cnν
of the transformation should be functions of x defined on the interval 0 ≤ x ≤ L. We
denote the set of these functions by un (x) and write
X
ϕ(x) = uν (x)Qν , uν (0) = uν (L) = 0 (2.24)
ν

where we indicated that uν (x) should vanish at the end points of the string to assure the
boundary conditions (2.1).
By comparing the string equation (2.2) and the coupled oscillators equation (2.15) we
see that the role of the coupling matrix kij is played by −v 2 ∂ 2 /∂x2 so the functions un (x)
must satisfy (cf., Eq. (2.17))
∂2
−v 2 uν (x) = ω 2 uν (x) , uν (0) = uν (L) = 0 (2.25)
∂x2
which has orthonormal eigenfunction solutions
r
2 πν
uν (x) = sin kν x , kν = , ν = 1, 2, .... (2.26)
L L
with eigenvalues
ων = vkν (2.27)
We have fixed the coefficients in uν (x)’s so that these functions are normalized.
To conclude - the functions (2.26) represent the configurations of the string normal
modes in which all the points of the string oscillate with the same frequency ων which
depends on the wave number k of the mode, Eq. (2.27). We remind that in general the
relation ω = ω(k) of the frequency upon the wave vector is called the dispersion relation.
It is the most important characteristic of linear waves.

Quantum mechanics of string normal modes

Using uν (x)’s we can transform the Hamiltonian operator (2.13) of the string to a sum of
independent oscillators. We view the operators ϕop (x) and πop (x) as functions of x and
expand
∞ r ∞ r
X 2 X 2
ϕop (x) = sin kν x Q̂ν , πop (x) = sin kν x P̂ν (2.28)
ν=1
L ν=1
L

Since we expand operator valued functions the coefficients Q̂ν and P̂ν here are operators
which we denoted by hats above to avoid double subscripts.
There are important relations which these operators must satisfy in order to preserve
the canonical commutation relations (2.12) between ϕop (x) and πop (x). These operators
must themselves be canonical, i.e. they must obey

[Q̂µ , P̂ν ] = iℏδµν , [Q̂µ , Q̂ν ] = [P̂µ , P̂ν ] = 0 (2.29)

This can be verified in one of the two ways. We can insert the expansions (2.28)
in [ϕop (x), πop (y)]. Using the first commutator above and the completeness relation
Version of April 11, 2023 72

P
ν uν (x)uν (y) = δ(x − y) we will obtain that (2.12) is indeed satisfied. The other
two commutators simply assure that ϕop (x) and πop (x) commute at different points. The
other way is to ”invert” (2.28)
Z L r Z L r
2 2
Q̂ν = ϕop (x) sin kν x dx , P̂ν = πop (x) sin kν x dx (2.30)
0 L 0 L
and calculate the needed commutators. Incidentally the above relations also demonstrate
how the operators Q̂ν and P̂ν should act on wave functionals Ψ[ϕ(x)].
The commutation relations (2.29) mean that Q̂µ and P̂µ are respectively coordinate
and momentum operators of the normal modes of the string. As can be seen from (2.28)
classically they are coordinates and momenta representing the amplitudes and their ve-
locities of all the harmonic standing waves which the string can support.
Inserting the above expansions in (2.13) and using the orthonormality property of the
set (2.26) we obtain
∞
1X 2
Hop = (P̂ + ων2 Q̂2ν ) (2.31)
2 ν=1 ν
The Hamiltonian operator is reduced to a sum of terms each representing simple Harmonic
oscillator with unit mass and frequency ων . It is important to observe that the underlying
waves on the elastic string can only be seen in the dependence of ω of the oscillators on
the corresponding wave vectors encoded in the dispersion relation (2.27).
Classically the simple form (2.31) of the Hamiltonian in terms of the normal modes’
dynamical variables suggests to switch the string description from ϕ(x, t), π(x, t) to the
set {Qν (t), Pν (t)}. Quantum mechanically we note that the relation (2.10) and the first
of (2.30) implies that the operators Q̂ν are simple multiplication operators

Q̂ν Ψ[ϕ(x)] = Qν Ψ[ϕ(x)]

Following the commutation relations (2.29) the canonically conjugate operators P̂ν can
be taken as
∂
P̂ν Ψ[ϕ(x)] = −iℏ Ψ[ϕ(x)]
∂Qν
This suggests to switch to the description in which wave functionals Ψ[ϕ(x)] are viewed
as functions of (formally infinite number of) the variables Qν

Ψ[ϕ(x)] → Ψ({Qν })

We now note that the terms in the sum representing Hop , Eq. (2.31) commute between
themselves on account of the last pair of commutators in Eq. (2.29). This means that
the eigenfunctions of Hop are products of the eigenfunctions of all individual terms in the
sum and the corresponding eigenvalues are sums of individual eigenvalues.
∞ ∞
X 1 X
E{Nν } = ℏων (Nν + ) = Eground state + ℏων Nν , Nν = 0, 1, 2, ...
ν=1
2 ν=1
∞ ∞
(2.32)
X ℏων Y p
Eground state = , Ψ{Nν } ({Qν }) = ψNν (βν Qν ) , βν = ων /ℏ
ν=1
2 ν=1
Version of April 11, 2023 73

The eigenfunctions ψN are the well known harmonic oscillator eigenfunctions

1  ω 1/4 2
ψN (y) = √ e−y /2 HN (y) (2.33)
2 N ! πℏ
N

where HN (y) denotes N-th order Hermite polynomial. In Fig. 2.2 graphs of several of
these functions are shown.

Figure 2.2: Energy levels and corresponding wave functions of harmonic oscillator. The
energy levels are ”equidistant”, separated by equal energy intervals

The energies (2.32) exhibit the most important result of our discussion of string quan-
tization - that it can be viewed as a collection of independent quanta with energies
ℏων = ℏvkν . This is a consequence of two general features - string is a linear dynamical
system and therefore a collection of normal modes oscillators while the quantum energy
levels of oscillators are ”equidistant”, i.e. separated by equal energy intervals ℏω.
In the next Section we will consider a closed string which can support not only normal
modes in a form of standing waves like the present fixed end string but also traveling
waves. We will show that the corresponding energy quanta of such modes carry mechanical
momentum and could therefore be considered as particles.

String ground state. The Casimir effect

The ground state energy in (2.32) is formally a sum of infinitely many ”zero point motion”
terms. This is an ”ultraviolet” infinity related to the formal possibility to have waves with
kν → ∞, i.e. vanishingly small wavelengths λν = 2π/kν . In practice of course the simple
description given by Eq. (2.2) ceases to be valid at atomic scales and should be replaced
by a more elaborate model. As a (much more practical) alternative one could introduce
Version of April 11, 2023 74

a cutoff kcutof f for ”allowed” normal modes in the model and limit the validity of the
model (2.2) to scales ≫ λcutof f = 2π/kcutof f .
Denoting by νc the largest integer ν corresponding to the kcutof f = πνc /L the ground
state wavefunction is a product of νc Gaussians
νc
Y
Ψground state ({Qν }) = ψ0 (βν Qν ) =
ν=1
νc  c ν
! (2.34)
Y ων 1/4 X
= exp − ων Q2ν /2ℏ
ν=1
πℏ ν=1

which express the ”zero point” fluctuations of the quantum string which is not at rest
even in its lowest energy state.
As we will see in the forthcoming sections the ground state of the EM field is expected
to exhibit similar zero point fluctuations of the fields in its ground state which is the
vacuum of the theory. Is it possible to observe these vacuum fluctuations? In a 1948
famous paper Ref. [1] Casimir proposed a way to do this using what has become known
as a Casimir effect. We will now explain its principle idea in the simple example of the
guitar string ground state.
Let us consider what will happen with the quantum guitar string if we ”fret” it, i.e.
press with an imaginary finger hard at some position x = d so that the string will not
vibrate at this point, Fig. 2.3. Obviously this changes the normal modes of the string by
excluding the modes which do not vanish at x = d. This means that the ground state
energy density will change. In fact the new normal modes will consist of two families with
ων′ = vkν = vπν/d and ων′′ = vπν/(L − d). The corresponding ground state energy will
correspondingly consist of two parts
νc  
X ℏvπ ν ν
E0 (d) = + (2.35)
ν=1
2 d L−d

It is clear that for d = L/2 both parts are equal while for d < L/2 (d > L/2) the first
term, i.e. the energy of the narrower (wider) part is smaller (larger) than the second term.
The finite cutoff frequency
P ∼ νc in the above expression ”regularizes” the (ultraviolet)
divergence of the sum ν ν. To eliminate νc from the final result one must ”renormal-
ize” it which can be done, cf., Ref. [2], by calculating E0 (d) relative to the symmetric
configuration at d = L/2 with the result1
 
πℏv 1 1 4
∆E0 (d) ≡ E0 (d) − E0 (L/2) = − + −
24 d L − d L
Pνc P∞
1 The common way of calculating is to use a ”soft” cutoff, i.e. to replace e.g. ν=1 ν by ν=1 νe−ν/νc ,
calculate the last sum for νc → ∞ using
∞ ∞
X ∂ X −αν ∂ 1 1 1
νe−αν = − e = −→ − + ...
ν=1
∂α ν=1 ∂α 1 − eα α→0 α2 12

with α = 1/νc
Version of April 11, 2023 75

It is seen that ∆E0 (d) is symmetric with respect to d → L−d and decreases monotonically
as d → 0 and d → L as
πℏv 1 πℏv 1
∆E0 (d)|d≪L/2 → − , ∆E0 (d)|L−d≪L/2 → − (2.36)
24 d 24 L − d
The dependence on d means that the function E0 (d) can be considered as the potential
energy of the separation point of the string at x = d and that there is a force

∂E0 (d)
F (d) = −
∂d
acting on what causes the separation between the two parts of the string (the imaginary
fretting finger). This force ”tries to drive” the separation towards the end points of the
string. A simple physical intuition behind this force is the imbalance of the ground state
fluctuations radiation pressure on both sides of the separation point x = d when d ̸= L/2.
One must be aware however that things are more delicate as the sign of the force depends
on the type of boundary conditions assumed at x = d. For details cf., Ref. [2].
The force F (d) is called the Casimir force and its appearance is a manifestation of a
Casimir effect. We will return to this effect below in the context of the vacuum fluctuations
of the quantized EM field.

Figure 2.3: Guitar string fretted at x = d and its (schematically drawn) normal modes

2.2 Quantization of Traveling Waves. Closed String

The normal modes of a string were described in the previous section are standing waves
as can be most clearly seen by considering the expansions (2.28). Classically Q̂ν ’s and P̂ν
are functions Qν (t), Pν (t) each depending harmonically on time with frequency ων . Thus
each term in (2.28) is a standing wave ∼ Qν (t) sin kν x and ∼ Pν (t) sin kν x.
Version of April 11, 2023 76

2.2.1 Expansion in traveling waves

Periodic boundary conditions
The standing wave solution of the equations (2.25) defining the normal modes is a con-
sequence of the fixed ends boundary conditions (2.1) for the guitar string. These were
reflected in the boundary conditions uν (0) = uν (L) = 0 in the normal modes equations
Eqs. (2.25). In this section we will explore a more interesting and practical situation when
the normal modes are traveling waves. This is realized if one assumes periodic boundary
conditions for a string, i.e. for every x

ϕ(x, t) = ϕ(x + L, t) (2.37)

This effectively means that such string does not have ends, i.e. it is closed and equivalent
to a circle. Note in passing that differentiating the periodicity condition with respect to
x shows that also the derivatives ∂ϕ/∂x, ∂ 2 ϕ/∂x2 , etc of ϕ are periodic. According to the
string equation (2.2) this means that so are the time derivatives.
The periodic boundary conditions for ϕ(x) are translated into conditions

uν (x) = uν (x + L)

replacing the fixed ends conditions in the normal modes equation (2.25). The solutions
are now coming as an infinite set of degenerate pairs each with the same frequency
r r
2 2 2πν
sin kν x , cos kν x , kν = , ων = vkν , ν = 1, 2, ... (2.38)
L L L
This degeneracy is ”compensated” by the the values of kν being at twice larger intervals
∆k = 2π/L than in the standing wave solutions (2.26) with ∆k = π/L. In real space this
means that the normal modes of the closed string have integer number L/λν = ν of the
wavelengths λν = 2π/kν = L/ν over the string length L rather than integer number of
half wavelengths λν = 2L/ν as it was in the fixed ends string case.
Solutions belonging to different frequencies are automatically orthogonal and we chose
them to be orthogonal also within each degenerate pair. Here is a helpful integral
Z L Z L
1
sin kν x cos kµ xdx = [sin(kν + kµ )x + sin(kν − kµ )x] dx =
0 0 2
 
1 1 L 1 L
=− cos(kν + kµ )x + cos(kν − kµ )x =0
2 kν + kµ 0 kν − kµ 0

We also normalize them as in the fixed ends case.

As always with degeneracies the choice above is of course not unique. Another useful
possibility is
1 2πν
√ eikν x , kν = , ων = v|kν | , ν = ±1, ±2, ... (2.39)
L L
again with degenerate in frequency orthogonal pairs. We will see the results of such a
choice below, cf., Eq.(2.64). In the following sections we will use the freedom in specifying
the degeneracy of normal modes in a closed string to find the traveling waves expansion.
Version of April 11, 2023 77

Let us also note that a non vibrating constant solution u0 (x) = const exists with
ω = 0. This means that the string configuration in this mode is constant independent of
x and has linear time dependence ϕ(x, t) = at + b. It describes a uniformly moving string
and plays an important role in the so called super string theory. It will not be of interest
to us and will not be included in our considerations.

Trying the simplest expansion

The most natural expansion using the above degenerate modes would be
r ∞
2X
ϕ(x, t) = [Q1,ν (t) sin kν x + Q2,ν (t) cos kν x]
L ν=1
r ∞ (2.40)
2X
π(x, t) = [P1,ν (t) sin kν x + P2,ν (t) cos kν x]
L ν=1

with two independent sets of amplitudes {Q1,ν (t), P1,ν (t)} and {Q2,ν (t), P2,ν (t)} for the
two degenerate modes. Inserting this in the string equation and projecting each mode
leads to the decoupled equations

Q̈i,ν + ων2 Qi,ν = 0 ; i = 1, 2 ; ν = 1, 2, ... (2.41)

and the corresponding Hamiltonian

∞
X X 1 2
+ ων Q2i,ν


H= Hi,ν , Hi,ν = Pi,ν
i=1,2 ν=1
2

The general solution

Qi,ν (t) = Qi,ν (0) cos ων t + Pi,ν (0)/ων sin ων t

(2.42)
Pi,ν (t) = −ων Qν (0) sin ων t + Pν (0) cos ων t

with arbitrary initial conditions Qi,ν (0) and Pi,ν (0) = Q̇i,ν (0) shows however that the
expansion above will be in terms of standing or traveling waves depending on these con-
ditions.

Transforming to new normal modes variables

To obtain an expansion in traveling waves let us use the degeneracy of the two modes at
every kν and do the following transformation
1 1
Q1,ν = √ (Qk − Q−k ) , P1,ν = √ (Pk − P−k )
2 2
(2.43)
1 ω
Q2,ν = √ (Pk + P−k ) , P2,ν = − √k (Qk + Q−k )
ωk 2 2
to new variables Q±k , P±k . Note that to simplify notations we dropped the subscript
ν in the right hand side and denoted accordingly ωk = v|k|. We note that the above
Version of April 11, 2023 78

transformation mixes coordinates and momenta. In the Appendix we show that this
transformation is canonical.
Substituting (2.43) in the expansion (2.40) we obtain
r  
1X Pk (t)
ϕ(x, t) = Qk (t) sin kx + cos kx
L v|k|
k
r (2.44)
1X 2πν
π(x, t) = [Pk (t) sin kx − v|k|Qk (t) cos kx] , k = , ν = ±1, ±2, ...
L L
k

where we have combined together the sums over Qk , Pk and Q−k , P−k modes by extending
the sums to include the negative values of k. We show in Appendix that this is the desired
expansion in traveling waves - waves with positive and negative k’s moving in opposite
directions. The hamiltonian in the new variables has the sum of independent oscillators
form
1X 2
Pk + ωk2 Q2k


H= (2.45)
2
k

with the sum extending over both positive and negarive k’s. The key point to note is that
compared to (2.40) this is achieved in the expansion (2.44) by making the amplitudes of
the second degenerate mode not independent but proportional to the canonical conjugate
of the amplitudes of the first mode and extending the the sum to the negative k’s.

Inverting the transformation

We
p remark that p inverting (2.44) requires some care. The simplest is to recall that
2/L sin kx and 2/L cos kx form orthonormal set for positive k > 0 and use the ex-
pansion (2.40) together with the relations (2.43). This means that it is the following
combinations of Qk and Pk which are simple projections
r Z L
1 2
√ (Qk − Q−k ) = dx ϕ(x) sin kx
2 L 0
r Z L
1 2
√ (Pk − P−k ) = dx π(x) sin kx
2 L
r Z0 L (2.46)
1 2
√ (Pk + P−k ) = dx ϕ(x) cos kx
ωk 2 L 0
r Z L
ωk 2
− √ (Qk + Q−k ) = dx π(x) cos kx
2 L 0

from which the expressions for each of the Qk and Pk follow by a simple calculation.
Z L
1 1
Qk = √ dx[ϕ(x) sin kx − π(x) cos kx]
L 0 ω k
Z L (2.47)
1
Pk = √ dx[π(x) sin kx + ωk ϕ(x) cos kx]
L 0
Version of April 11, 2023 79

The physics of the new variables

Let us write the expressions for the terms in the expansions (2.44) as a single function.
For this purpose let us transform
1
q
Qk = Ck cos αk , Pk = −ωk Ck sin αk , Ck = Pk2 + ωk2 Q2k (2.48)
ωk
This gives
Pk
Qk sin kx + cos kx = Ck sin(kx − αk )
ωk (2.49)
Pk sin kx − ωk Qk cos kx = −ωk Ck cos(kx − αk )

Since the amplitude Ck is proportional to the square root of the energy of the mode it is
a constant of the motion for the mode time oscillations. It is not difficult to show that
the phase αk (t) is just
αk (t) = ωk t + αk (0)
Indeed writing p
Ck = 2Ik /ωk
one finds that Eq.(2.48) is essentially a canonical transformation from Pk , Qk to the
action-angle variables Ik , αk for a harmonic oscillator, cf., Ref.[3], with
1
Ik = (P 2 + ωk2 Q2k ) , tan α = −Pk /ωk Qk
2ωk k (2.50)
I˙k = 0 , α̇k = ωk

Mechanical momentum of the string

Apart of the Hamiltonian an important quantity describing the physics of a string is its
mechanical momentum (please do not confuse this P with the canonical Pk ’s)
Z L
∂
P=− π(x, t) ϕ(x, t)dx (2.51)
0 ∂x
It is conserved by the string equations of motion (2.2) as can be seen from the following
calculation. Defining the density of P
∂
P(x, t) = −π(x, t) ϕ(x, t)
∂x
we have
∂P(x, t) ∂π(x, t) ∂ϕ(x, t) ∂ 2 ϕ(x, t)
=− − π(x, t)
∂t ∂t ∂x ∂t∂x
2
∂ ϕ(x, t) ∂ϕ(x, t) ∂π(x, t)
= −v 2 − π(x, t)
∂x2 ∂x ∂x
"  2 #
1 ∂ 2 ∂ϕ(x, t) 2
=− v + π (x, t)
2 ∂x ∂x
Version of April 11, 2023 80

This is one dimensional analogue of the continuity equation which connects the time
derivative of P(x, t) and the space derivative of the density of the Hamiltonian
"  2 #
1 2 ∂ϕ(x, t)
H(x, t) = π (x, t) + v 2
2 ∂x

Integrating this and using the boundary conditions (must use periodic not fixed ends) we
obtain the conservation law
Z L
∂P ∂
= P(x, t)dx = 0 (2.52)
∂t ∂t 0

2.2.2 Quantum mechanics of the traveling waves

The basic operators. String Hamiltonian and momentum
As in the case of the fixed ends string the closed string is quantized by introducing wave
functionals Ψ[ϕ(x), t] for the string coordinates and the operators ϕop (x) = ϕ(x) and
πop (x) = −iℏδ/δϕ(x). The Hamiltonian operator is the same given by Eq.(2.13) since the
string equations are the same. What is different are the boundary conditions which led
to modified (degenerate) normal modes and the corresponding expansions (2.44).
Using these expansion for the operators
r  
1X 1
ϕop (x) = sin kx Q̂k + cos kx P̂k
L v|k|
k
r
1 Xh i
πop (x) = sin kx P̂k − v|k| cos kx Q̂k (2.53)
L
k

we are led to the canonical commutators for the traveling waves amplitudes

[Q̂k , P̂k′ ] = iℏδkk′ , [Q̂k , Q̂k′ ] = [P̂k , P̂k′ ] = 0 (2.54)

These of course follow from the basic commutators (2.12) and the expressions (2.47). The
Hamiltonian operator in terms of Q̂k ’s and P̂k ’s has the same form of decoupled oscillators
1 Xh 2 i
Hop = P̂k + ωk2 Q̂2k (2.55)
2
k

Quantum mechanically the string momentum P, Eq. (2.51), becomes an operator

1 L
   
∂ ∂
Z
Pop = − dx πop (x) ϕop (x)dx + ϕop (x) πop (x) (2.56)
2 0 ∂x ∂x

As usual with products of non commuting operators, here πop (x) and ϕop (x), one must
use a symmetrized expression.
The operator Pop is the generator of translations

ϕop (x) → ϕop (x + a) , πop (x) → πop (x + a)

Version of April 11, 2023 81

Indeed using the basic commutators (2.12) one can easily verify that
∂ ∂
[Pop , ϕop (x)] = −iℏ ϕop (x) , [Pop , πop (x)] = −iℏ πop (x) (2.57)
∂x ∂x
as it should be for the generator of translations.
Inserting the expansions (2.53) in the momentum Pop we obtain

1 X Lh i k′
Z 
Pop = − sin kx P̂k − v|k| cos kx Q̂k k ′ cos k ′ x Q̂k′ − sin k ′
x P̂ k dx
′
L ′ 0 v|k ′ |
kk
  X
1X k 2 k 1h 2 i
= P̂k + vk|k|Q̂2k = P̂k + ωk2 Q̂2k
2 v|k| ωk 2
k k

The eigenstates. Energies and momenta of traveling waves quanta

The traveling waves Hamiltonian (2.55) has the same decouple normal modes oscillators
form as the one for the standing waves (2.31) so formally its solutions have the same form
as (2.32)
X
E{Nk } = Eground state + ℏωk Nk , Nk = 0, 1, 2, ....
k
Y p (2.58)
Ψ{Nk } ({Qk }) = ψNk (βk Qk ) , βk = ωk /ℏ
k

with familiar harmonic oscillator eigenfunctions ψN (βQ), Eq. (2.33). There are however
important differences.
Since ωk = v|k| the traveling waves energy quanta ϵk = ℏωk are doubly degenerate with
respect to the sign (direction) of k. Even more profound is that these quanta also carry
momentum. Indeed comparing the expression (2.58) for the string momentum Pop with
the Hamiltonian Hop one observes that Pop has the same eigenfunctions (not surprising)
with the eigenvalues2 X
P{Nk } = ℏkNk (2.59)
k
Each quantum has ”mechanical” momentum pk = ℏk. So the closed string can be consid-
ered as a collection of traveling waves ”quasi”particles with energy momentum relation

ϵk ≡ ℏωk = ℏv|k| = v|pk | ⇒ ϵ(p) = v|p| (2.60)

It is useful to pay attention that this result can be viewed (obtained by a shortcut) as a
consequence of the three fundamental relations - two basic quantum mechanical relations
- the Plank-Einstein ϵ = ℏω and the de Broglie p = ℏk and the string dispersion relation
ω = v|k|. In a similar way we will find below that the quanta of the EM field will be
particles (photons) with energy-momentum relation ϵ = ℏω = ℏck = cp i.e. of massless
relativistic particles. The quanta of the Schrödinger field will have ϵ = ℏω = ℏ2 k 2 /2m =
p2 /2m, i.e. the energy-momentum relation of non relativistic particles.
2 The presence of ±ℏk terms in this expression helps to cancel the 1/2 ”zero point motion” term present

in the expression for the energy.

Version of April 11, 2023 82

Transformation to creation and annihilation operators

In practice it is very convenient to introduce creation and annihilation operators in the
standard way
   
Q̂k = ℏ/2ωk âk + â†k , P̂k = i ℏωk /2 â†k − âk
p p
   
âk = 1/2ℏωk iP̂k + ωk Q̂k , â†k = 1/2ℏωk −iP̂k + ωk Q̂k (2.61)
p p

[âk , â†k′ ] = δkk′ , [âk , âk′ ] = 0 = [â†k , â†k′ ]

Using these operators we can write the Hamiltonian

ℏωk â†k âk

X
Ĥr = E0 + (2.62)
k

and its eigenstates

Y Y (↠)Nk
k
|{Nk } >= |Nk >= |0 > (2.63)
k k
(N k !)1/2

Great advantage of using â and ↠operators rather than P̂ and Q̂ is the simplicity of the
”action” of these operators on the ”number states”, i.e the states with a fixed numbers
of quasi particles in each normal mode. Schematically
√ √
â|n >= n|n − 1 > , ↠|n >= n + 1|n + 1 >

In detailed notation
p Y
âk |{Nk } >= Nk |Nk − 1 > |{Nk′ } >
k′ ̸=k

â†k |{Nk } >=

p Y
Nk + 1|Nk + 1 > |{Nk′ } >
k′ ̸=k

One says that the operators â†k and âk create and destroy quasi particles of energy ℏω
and momentum ℏk.
It is useful to express the field operators in terms of a and a† . Using (2.53) we obtain
r r
ℏ h ikx † −ikx
i ℏ h i
âk − â†−k eikx
X X
ϕop (x) = −i âk e − âk e = −i
2ωk L 2ωk L
k k
r r (2.64)
X ℏωk h † −ikx
i X ℏωk h †
i
ikx ikx
πop (x) = − âk e + âk e =− âk + â−k e
2L 2L
k k

which are sums of terms which either create a quantum with momentum ℏk or annihilate
one with the opposite momentum −ℏk.
Version of April 11, 2023 83

2.3 Quantization of the EM Field

The quantization of the electromagnetic field follows the same route as with the simple
string above. The classical equations of the field are the Maxwell equations. We will now
cast them into Hamilton form and identify the Hamiltonian and the canonical coordi-
nates and momenta of the field. We will then replace them by operators with canonical
commutation relations acting on the appropriate wave functionals.

2.3.1 Hamilton form of the Maxwell equations

The Maxwell equations have the familiar form3
ρ(r)
∇ · E(r, t) = , ∇ · B(r, t) = 0 (2.65)
ϵ0
∂B(r, t) 1 ∂E(r, t) j(r, t)
∇ × E(r, t) = − , ∇ × B(r, t) = + 2
∂t c2 ∂t c ϵ0
Here c is the light velocity and ϵ0 is a constant ϵ0 = 8.85 · 10−12 F m−1 called vacuum
permittivity which is related to our choice of the SI measurement unit system.
The Maxwell equations describe the EM field configuration for a given distribution
of the electric current j(r, t) and density ρ(r, t) of electric charges. Assuming that we
are dealing with a system of N charges and denoting by ra (t) , va (t), a = 1, ..., N their
positions and velocities we have
N
X N
X
ρ(r, t) = qa δ(r − ra (t)) , j(r, t) = qa va δ(r − ra (t)) (2.66)
a=1 a=1

These expressions must be supplemented by the mechanical equations of motion for the
charges as they move in the given E(r, t) and B(r, t). These equations are just the Newton
equations for the charges
dva dra
ma = qa E(ra , t) + qa (va × B(ra , t)) , va = , a =, ..., N (2.67)
dt dt
The coupled equations (2.65), (2.66) and (2.67) provide the complete system which deter-
mines how the positions of the charges and their motion determine the EM field and how
this field determines the motion of the charges. Our first goal will be to cast this system
in the Hamiltonian form thereby determining its canonical variables and the Hamiltonian.

Vector potential. The A0 = 0 gauge

We start by noting that the first pair of Maxwell equations (2.65) does not involve time
derivatives. They are in a sense constraints on the possible functional dependence of E(r)
and B(r). Both constraints are easy to resolve. The condition ∇ · B = 0 means that there
are no magnetic charges in nature and that B can be represented as a curl of an arbitrary
vector function
B=∇×A (2.68)
3 In this Chapter we use the SI system of units.
Version of April 11, 2023 84

which is conventionally called the vector potential. In the Chapter where we considered
the motion in an external EM field we have seen that the quantum mechanical formulation
was impossible without an explicit use of this function. Also presently we will find that
the quantization of EM field can not avoid using A.
The second pair of the Maxwell equations consists of dynamical equations. Inserting
Eq. (2.68) in the first of these equations we obtain

∂A ∂A
∇ × (E + )=0 ⇒ E=− − ∇A0 (2.69)
∂t ∂t
Here A0 is (in non relativistic parlance) the ”scalar potential” which together with A
completely determine the fields E and B. The potentials A and A0 are not uniquely
defined. We can choose instead different function A′0 (r, t) and A′ (r, t) related to A0 and
A by the gauge transformation
∂χ
A′0 = A0 − , A′ = A + ∇χ (2.70)
∂t
with arbitrary function χ(r, t). We shall use this freedom and take A0 to be identically
equal to zero and write
∂A
= −E (2.71)
∂t
This choice is called ”working in the A0 = 0 gauge”. Importantly this choice does not
exhaust the full gauge freedom. We can still add to A a gradient of a time independent
function χ(r) without changing our A0 = 0 assumption.
Inserting (2.68) in the last of the Maxwell equations (2.65) we obtain

∂E j
= c2 ∇ × ∇ × A − (2.72)
∂t ϵ0
Using (2.71) this equation becomes

∂2A j
2
= −c2 ∇ × ∇ × A + (2.73)
∂t ϵ0

Regarding the 2nd time derivative on the left as acceleration of A(r, t) one can view the
above equations as coupled Newton equations for the degrees of freedom A(r). In this view
at every point in space there are three such degrees of freedom (field coordinates) which
can symbolically be represented as Ai,r . The 3 dimensional vector index is i = 1, ..., 3 and
r is running over all points in the 3 dimensional space in a way similar to x running over
points of the x axis in the example of a string. The coupling between different Ai,r is via
complicated combination of second order vector derivative ∇ × ∇× connecting different
vector components of A(r) in neighboring points.
The last term in (2.72) is the ”force” acting on the field coordinates on the part of
the matter. Ignoring this force for a moment (i.e. considering the EM field in an empty
space region) we can view the coordinates A(r) as representing coupled oscillators. This
is because the above equation without the last term is linear. Although complicated from
the vector analysis point of view the derivatives combination ∇×∇× is a linear operation.
Version of April 11, 2023 85

The Hamiltonian
Continuing with the ”mechanical” interpretation of the EM field dynamics we notice that
the pair of the 1st order equations (2.71) and (2.72) without the last term can be regarded
together as Hamilton equations with the following field Hamiltonian
ϵ0
Z
d3 r E2 (r) + c2 (∇ × A(r))2
 
Hf = (2.74)
2
and canonical variables Ai (r) as coordinates and −ϵ0 Ei (r) as momenta. We will verify
this in a moment but first we note that perhaps the simplest way to guess the expression
of the Hamiltonian is to notice that on account of Eq. (2.71) the first term in it has the
form of the kinetic energy. One can determine how it changes with time by forming a
scalar product of the left hand side of (2.72) with E. Multiplying the first term on the
right hand side (remember we still are ignoring the current term) with the equal quantity
−∂A/∂t we can integrate both sides over r. After simple manipulations4 one can show
that the change in time of the kinetic energy is equal to minus the change in time of the
second term in Hf which has the meaning of the potential energy. This of course verifies
that Hf is conserved, dHf /dt = 0.
Returning to the Hamiltonian (2.74) we form its first variation
Z
d3 r ϵ0 E · δE + ϵ0 c2 (∇ × A) · (∇ × δA)
 
δHf =
Z
d3 ϵ0 E · δE + ϵ0 c2 ϵijk ∂j Ak ϵilm ∂l δAm
 
=
Z
d3 r ϵ0 E · δE − ϵ0 c2 δAm ϵilm ϵijk ∂l ∂j Ak
 
=
Z
d3 r ϵ0 E · δE + ϵ0 c2 δA · ∇ × ∇ × A)
 
=

where we performed integration by parts in the second term. From this it follows that
δHf δHf
= −E , = ϵ0 c2 ∇ × ∇ × A (2.75)
δ(−ϵ0 E) δA

showing that the Hamilton equations with this Hamiltonian and canonical variables A(r)
and −ϵ0 E(r)

∂A δHf ∂(−ϵ0 E) δHf

= = −E , =− = −ϵ0 c2 ∇ × ∇ × A (2.76)
∂t δ(−ϵ0 E) ∂t δA
4 Hereare the details
Z Z Z
∂E ∂A
d3 r E · = −c2 d3 r ·∇×∇×A = −c2 d3 r∂t Ai ϵijk ∂j ϵklm ∂l Am =
∂t ∂t
Z Z  
∂A
= (integrate by parts) = c2 d3 rϵijk ∂j ∂t Ai ϵklm ∂l Am = −c2 d3 r ∇ × · (∇ × A) .
∂t
c2 d
Z Z
d 1
Rewrite as d3 r E2 = − d3 r(∇ × A)2
dt 2 2 dt
Version of April 11, 2023 86

indeed coincide with Eqs. (2.71) and (2.72) without the current term.
Let us now show how to account for the current term and the dynamical Newton
equations (2.67) for the charges. Here we are guided by our knowledge of the Hamiltonian
of charges moving a given EM field (cf. chapter Motion in External EM Field). We simply
add it to Hf above and obtain
N
ϵ0 1
Z
 X 2
d3 r E2 (r) + c2 (∇ × A(r))2 +

H= [pa − qa A(ra )] (2.77)
2 a=1
2m a

Using exactly the same calculation as in the chapter Motion in External EM Field we can
show that the Hamilton equations
dra ∂H dpa ∂H
= , =− (2.78)
dt ∂pa dt ∂ra

are equivalent to the Newton equations (2.67). Let us now consider the first variation of
the last term in (2.77) with respect to A(r). We obtain
N N
qa qa
X X Z
− [pa − qa A(ra )] · δA(ra ) = − [pa − qa A(ra )] · d3 rδ(r − ra )δA(r)
a=1
ma a=1
m a

N
qa
Z X Z
= − d3 r [pa − qa A(ra )] δ(r − ra ) · δA(r) = − d3 r j(r) · δA(r)
a=1
ma

where j(r) is the current as defined in Eq. (2.67) with

1
va = [p − qa A(ra )]
ma a
With this result the second equation of (2.76) with Hf replaced by the full H (2.77) now
reads
∂(−ϵ0 E) δH
=− = −ϵ0 c2 ∇ × ∇ × A + j (2.79)
∂t δA
reproducing the full equation (2.72). We also note that since the second term in the full H
does not depend on E the first equation in (2.76) remains unchanged when Hf is replaced
in it by H.

2.3.2 Canonical quantization

Having established the canonical structure of the theory we can now quantize it. Attentive
reader should have noticed that we have not yet accounted for the first equation in the
set (2.65) expressing the Gauss law. We also seem to be missing from the Hamiltonian
(2.77) the regular Coulomb interaction energy between the charges {qa }. We will address
these issues shortly but meanwhile let us proceed with the quantization.
Moving from classical to quantum description we recognise that the coordinate set of
our system consists of the vector potential A(r) (i.e. 3 vector components in each point of
the position space, i.e. 3 × ∞3 variables ) and 3N vectors {ra } of the particles’ positions.
Version of April 11, 2023 87

Accordingly we introduce the wave functional of the field A(r) which also depends (is a
function of) the N particles’ positions and the

Ψ = Ψ[A(r), r1 , ..., rN , t] (2.80)

This should be viewed as a correspondence between all field configurations and set of N
particles’ positions {ra } and (in general) complex probability amplitudes which in general
change with time.
The physical operators are constructed from the corresponding classical quantities by
the canonical substitution
iℏ δ
A(r) → Aop (r) = A(r) , E(r) → Eop (r) = (2.81)
ϵ0 δA(r)
ra → r̂a = ra , pa → p̂a = −iℏ∇a

where we accounted for the fact that it is the combination −ϵ0 E(r) which is canonical to
A(r) not just E(r). Using the equality

δAj (r′ )
= δij δ(r − r′ ).
δAi (r)

it follows that the field operators obey the commutation relations

h i iℏ h i h i
Êi (r), Âj (r′ ) = δij δ(r − r′ ) , Âi (r), Âj (r′ ) = Êi (r), Êj (r′ ) = 0 (2.82)
ϵ0
The time dependence of the wave functional/function Ψ[A(r), r1 , ..., rN , t] is governed
by the Schrödinger equation
∂Ψ(t)
iℏ = Hop Ψ(t) (2.83)
∂t
in which the Hamiltonian operator Hop is obtained by replacing in the classical expression
(2.77) the fields A(r) , E(r) and the particle variables {ra }, {pa } by the corresponding
operators (2.81),
N
ϵ0 1
Z
 X 2
d3 r E2op (r) + c2 (∇ × Aop (r))2 +

Hop = [−iℏ∇a − qa Aop (r̂a )] . (2.84)
2 a=1
2m a

The above Hamiltonian does not depend on time as we are dealing with a closed EM
field + matter system. The energy is therefore conserved and quantum mechanically we
can reduce in the standard way the solution of the above time dependent Schrödinger
equation to solving the static equation

Hop Ψ = EΨ (2.85)

This is a complicated equation for the coupled field-matter system. No exact solution is
possible. We will discuss approximate solutions below. But before that we have to clarify
several formal but very important issues which will allow us to somewhat simplify the
problem.
Version of April 11, 2023 88

2.3.3 Gauge invariance

The Gauss law
We will now show that the Hamiltonian formulation presented above neatly accounts for
both the Gauss law and the Coulomb interaction between the charges. The key to this is
to observe that the Hamiltonian is invariant under the gauge transformation
A(r, t) → A(r, t) + ∇χ(r) , E(r, t) → E(r, t) (2.86)
pa (t) → pa (t) + qa ∇χ(ra (t)) , ra (t) → ra (t)
with an arbitrary time independent function χ(r). This is the residual gauge transforma-
tion we have briefly mentioned after Eq.(2.72).
The symmetry of H under (2.86) is the result of the way the vector potential A(r)
enters it, i.e. only via the combinations ∇ × A and [pa − qa A(ra )]. It is a local symmetry
meaning that it is characterised by parameters χ(r) which depend on r. Schematically
there are ∞3 parameters corresponding to the ”number” of points in the 3D space of
vectors r. As we will show below the generators of this symmetry are
gop (r) = −ϵ0 ∇ · Eop (r) + ρop (r) (2.87)
Their dependence on r means that there are ∞3 generators corresponding to ∞3 param-
eters χ(r) in (2.86).
Classically expressions corresponding to the generators of symmetries of the Hamilto-
nian are conserved by the Hamilton equations. Momentum and angular momentum are of
course the classic examples of such conservations. Accordingly let us show that equations
(2.71), (2.72) and (2.67) conserve the above expression for the generator gop (r, t) when
it is taken as classical and when E and ρ in it are allowed to evolve according to these
equations. We have
∂ ∂E ∂ρ ∂ρ
[−ϵ0 ∇ · E(r) + ρ(r)] = −ϵ0 ∇ · + =∇·j+ =0 (2.88)
∂t ∂t ∂t ∂t
where we used Eq.(2.72) and the continuity equation for the charges.5
The vanishing of ∂g(r)/∂t means that local quantities −ϵ0 ∇ · E(r) + ρ(r) form a
constant, time independent function of r. It is natural to denote this function by ρ0 (r)
−ϵ0 ∇ · E(r) + ρ(r) = ρ0 (r)
and interpret it as a density of fixed static electric charges. We notice that these charges
appear in addition to the dynamical charges qa described by the equations (2.67). Under
normal circumstances there are no such extra static charges. In fact there presence would
violate such symmetries as translational, rotational, Lorenz. So one should assume that
ρ0 = 0. Using this in the above relation we recover the Gauss law.
5 Continuity equation is a general relation between ρ(r, t) and j(r, t) given by (2.66)
N N
∂ρ(r, t) ∂ X X dra
= qa δ(r − ra (t)) = qa ∇ra δ(r − ra (t)) · =
∂t ∂t a=1 a=1
dt
= (using ∇ra δ(r − ra ) = −∇r δ(r − ra ) ) = −∇r · j(r, t)
Version of April 11, 2023 89

Quantum mechanics of the gauge transformation

Let us work out the quantum mechanics of the gauge transformation (2.86) . What we
want to show is that it is generated by the ∞3 operators gop (r), Eq. (2.87), i.e. that the
following relations hold

Aop (r′ ) Aop (r′ ) + ∇χ(r′ )

   
−(i/ℏ) d3 rχ(r)gop (r) (i/ℏ) d3 rχ(r)gop (r)
R R
e e = (2.89)
p̂a p̂a + qa ∇χ(ra )

As usual it is enough to consider the infinitesimal χ(r) for which the left hand side reduces
to
Aop (r′ ) −iℏ∇r δ(r − r′ )
    
i i
Z Z
3 3
− d rχ(r) gop (r), =− d rχ(r) (2.90)
ℏ p̂a ℏ −iℏqa ∇r δ(r − ra )

Here we omitted the identity term and used

[gop (r), Aop (r′ )] = [−ϵ0 ∇ · Eop (r), Aop (r′ )] = −iℏ∇r δ(r − r′ ) (2.91)

and
[gop (r), p̂a ] = [ρop (r), p̂a ] = iℏ∇ra δ(r − ra ) = −iℏ∇r δ(r − ra ) (2.92)
Now we do the integration by parts and use the delta function

−iℏ∇r δ(r − r′ ) δ(r − r′ ) ∇χ(r′ )

  Z    
i
Z
− d3 rχ(r) = d3 r∇χ(r) =
ℏ −iℏqa ∇r δ(r − ra ) qa δ(r − ra ) qa ∇χ(ra )
(2.93)
obtaining exactly what is needed to get the χ(r) dependent term in the right hand side
of Eq.(2.89).
Thus ∞3 operators gop (r) are indeed the generators of the gauge transformation. Since
the Hamiltonian operator Hop is invariant under this transformation one must have that
Hop commutes with gop (r)
[Hop , gop (r)] = 0 (2.94)
As in simpler quantum mechanical systems this means that Hop and gop (r) have common
eigenfunctions. We write symbolically

Hop Ψ = EΨ , gop (r)Ψ = ρ0 (r)Ψ (2.95)

where we denoted by ρ0 (r) the eigenvalues of the ∞3 operators qop (r). As in the classical
case the meaning of ρ0 (r) is the density of static (”background”) electric charges. They
are ”background” because they are not a part of the dynamics. Just sit there as a
part of initial conditions. Their presence would violate basic symmetries (translational,
rotational, Lorenz) so the physics dictates that one must select only the eigenfunctions
which belong in the ”sector” of the system Hilbert space for which

gop (r)Ψ = 0 (2.96)

In other words - the gauge invariant sector.

Version of April 11, 2023 90

Separating the longitudinal components of the fields

In the Hamiltonian formulation the electromagnetic field has 3 × ∞3 degrees of freedom
which in our formulation are described by the coordinates Ai (r) and the corresponding
momenta −ϵ0 Ei (r) , i = 1, 2, 3. Using the local gauge symmetry (2.86) one can eliminate
one third of these degrees of freedom. For this reason let us represent the functions A(r)
and E(r) as sum of the so called transverse and longitudinal components

A(r) = AT (r) + AL (r) , E(r) = ET (r) + EL (r) (2.97)

where AT , AL , ET and EL satisfy

∇ · AT (r) = ∇ · ET (r) = 0 , ∇ × AL (r) = ∇ × EL (r) = 0 (2.98)

Such a representation is possible for any vector field. This can be shown (and the origin
of the names longitudinal and transverse understood) using Fourier expansions. Let us
take for example A(r) and expand

Ak eik·r
X
A(r) = (2.99)
k
For convenience in order to have discrete values of k we consider the fields in a large but
finite volume. The precise boundary conditions are not important for this discussion.
The Fourier amplitudes Ak are vectors. Their directions in principle bear no relation
to the direction of the corresponding wave vectors k. We can however represent each of
them as a sum of two vectors which are parallel and perpendicular to ”their” k
(L) (T ) (L) (T )
A=A +A , with A ×k=0 , A ·k=0
k k k k
Using this we can write the Fourier expansion as a sum

A eik·r + A eik·r
X (L) X (T )
A(r) = (2.100)
k k
k k
Using
∇ · (aeik·r ) = ik · aeik·r , ∇ × (aeik·r ) = ik × aeik·r
we see that the two terms in the Fourier expansion of A(r) are respectively AL (r) and
AT (r) as appear in (2.97).
We note also that longitudinal components of the vector fields can be written as a
gradient of a scalar function. Therefore we can write

A(r) = AT (r) + ∇ξ(r) , E(r) = ET (r) − ∇ϕ(r) (2.101)

where two scalar functions ξ(r) and -ϕ(r) fully determine the longitudinal components
AL (r) and EL (r) respectively. As will become clear in the next section ϕ(r) is the scalar
electric potential so familiar from the Coulomb and other electrostatic problems.
Version of April 11, 2023 91

Recovering the Coulomb interaction. Resulting Hamiltonian

Inserting the above expressions for A(r) and E(r) in the Hamiltonian (2.77) we obtain
N
ϵ0 1
Z
 X 2
d3 r E2T (r) + (∇ϕ)2 + c2 (∇ × AT (r))2 +

H= [pa − qa AT (ra )] (2.102)
2 a=1
2m a

We have transformed pa to pa + qa ∇ξ(ra ). The mixed term containing ET · ∇ϕ does not

appear since it vanishes as can be seen after integrating it by parts
Z Z
d3 r ET · ∇ϕ = − d3 r (∇ · ET )ϕ = 0

As a last step in transforming H we note that the Gauss law allows to express ϕ(r) in
terms of ρ(r)

ρ(r) 1 1
Z
2
∇ · E(r) = −∇ ϕ(r) = ⇒ ϕ(r) = d3 r′ ρ(r′ ) (2.103)
ϵ0 4πϵ0 |r − r′ |

We see that in this formulation the familiar scalar potential appears in the longitudinal
component of the electric field EL (r) and is completely determined by the density of the
charge. This allows to express the term in H containing (∇ϕ)2 as

ϵ0 ϵ0 1 1 ρ(r)ρ(r′ )
Z Z Z Z
d3 r(∇ϕ)2 = − d3 rϕ∇2 ϕ = d3 rϕ(r)ρ(r) = d3 rd3 r′
2 2 2 8πϵ0 |r − r′ |

This is just the Coulomb interaction between the charges in ρ(r) and can be written using
the expression for ρ(r) given in (2.67) as
N N
ϵ0 1 X qa qb
Z X
d3 r(∇ϕ)2 = + ϵa (2.104)
2 8πϵ0 |ra − rb | a=1 self interaction
a̸=b

where ϵaself interaction are constants which express the Coulomb self-energy of each particle.
They diverge for point particles. We will not deal with this in details but assuming that
particles have small but finite sizes (cutoffs) we will simply disregard this constant term.
To conclude, the Hamiltonian has the form
N
X 1 2
H = Hr + [pa − qa AT (ra )] + VCoul (2.105)
a=1
2m a

where we defined the radiation and the Coulomb interaction parts of the Hamiltonian
ϵ0
Z
d3 r E2T (r) + c2 (∇ × AT (r))2
 
Hr = (2.106)
2
N
1 X qa qb
VCoul =
8πϵ0 |ra − rb |
a̸=b
Version of April 11, 2023 92

It is often convenient to write this Hamiltonian as a sum of three parts

H = Hr + Hmatter + Uradiation−matter interaction (2.107)

where Hr is given by (3.25) and

N
X p2a
Hmatter = + VCoul (2.108)
a=1
2ma
N
X qa
Uradiation−matter interaction = − pa · AT (ra ) + (2.109)
a=1
ma

N
X qa2
+ AT (ra ) · AT (ra )
a=1
2ma

Note that when switching to operators there will be no operator ordering ambiguity in
the term p̂a · AT (ra ) since the difference, i.e. the commutator

p̂a · AT (ra ) − AT (ra ) · p̂a = −iℏ∇a · AT (ra ) = 0

An aside - separating transverse and longitudinal Maxwell equations

Let us now examine how the Hamilton (Maxwell) equations (2.71) and (2.72) look in
terms of the transverse fields AT (r), ET (r). One can see that each equation separates
into two relating separately the transverse and longitudinal components
∂AT ∂AL
= −ET , = −EL (2.110)
∂t ∂t
∂ET jT ∂EL j
= c2 ∇ × ∇ × AT − , =− L
∂t ϵ0 ∂t ϵ0
The last equation is equivalent to the continuity equation for the current
∂ρ(r, t) ∇ · jL (r, t) ∇ · j(r, t)
=− =− (2.111)
∂t ϵ0 ϵ0
as can be see by taking divergence of both parts and using the Gauss law for ∇ · EL .
The equation ∂AL /∂t = −EL , shows how the longitudinal component of A which does
not enter the Hamiltonian (i.e. is the ”cyclic” coordinate) develops in time for a given
EL which in turn is determined by the Gauss law via the charge density. One can see
this as analogous to say the motion of the angular coordinates in a spherically symmetric
problem as determined by the (conserved) angular momentum.

2.4 Photons
2.4.1 Field oscillators
In the present and following sections we will disregard the radiation-matter interaction
and will concentrate of the radiation part described by Hr . Since this Hamiltonian is
Version of April 11, 2023 93

quadratic we will continue as in the case of a string. We will impose periodic boundary
conditions and will expand ET (r) and AT (r) in terms of traveling waves. As we will
see Hr will become a sum of decoupled oscillators so the traveling waves are the normal
modes of the radiation.

Expansion in traveling waves

Following the string example, cf., Eq.(2.44) and the footnote6 below we expand the field
canonical coordinates and momenta A(r) and −ϵ0 E(r), cf. Eq. (2.76) in a large volume
Ω

 
1 X 1
A(r) = √ qk sin(k · r) + pk cos(k · r) (2.112)
Ωϵ0 ck
k
1 X

E(r) = −√ pk sin(k · r) − ck qk cos(k · r)
Ωϵ0
k
with vector expansion coefficients qk and pk . To make the above expansions more sym-
metric with respects to the appearance of ϵ0 we changed our canonical variables to
√ √
A(r) → A(r)/ ϵ0 , −ϵ0 E(r) → − ϵ0 E(r)
The periodic boundary conditions lead to discrete values of the wave vectors
 
2π
k = (nx , ny , nx ) 1/3 , ni = 0, ±1, ±2 ... (2.113)
Ω
where we assumed the volume to be a cube, i.e. have the same length, width and height
each equal to Ω1/3 .
The more conventional form of the expansion Eq. (2.112( found in the literature (cf.,
cf. Landau and Lifshitz, Classical Field Theory, Sec.52 or Ref. [7] ) is written in terms of
canonically transformed variables
1
qk → − pk ; pk → ωk qk (2.114)
ωk
which for the transverse components of the fields results in
 
1 X 1
AT (r) = √ Qk cos(k · r) − P sin(k · r) (2.115)
Ωϵ0 ωk k
k
1 X

ET (r) = − √ Pk cos(k · r) + ωk Qk sin(k · r)
Ωϵ0
k
6 For easy comparison we reproduce this expansion here
r  
1 X Pk (t)
ϕ(x, t) = Qk (t) sin kx + cos kx
L k v|k|
r
1 X 2πν
π(x, t) = [Pk (t) sin kx − v|k|Qk (t) cos kx] , k = , ν = ±1, ±2, ...
L k L
Version of April 11, 2023 94

The transversality of AT (r) and ET (r) means that the vectors Qk and Pk are always
orthogonal to the corresponding k,

k · Qk = 0 , k · Pk = 0.
It is convenient to use a pair of fixed unit polarization vectors λkα , α = 1, 2 with

λk1 · λk2 = 0 , λkα · k = 0 , α = 1, 2 (2.116)

so we can write X X
Qk = Qkα λkα , Pk = Pk α λ k α (2.117)
α=1,2 α=1,2

We now insert expansions (2.115) and (2.117) into the Maxwell equations for the
transverse component, i.e. into the 1st and 3rd equation of the set (2.110) in the absence
of the current (recall we are discussing pure radiation). We obtain in the straightforward
manner separate linear equations for Qkα and Pkα

Q̇kα = Pkα , Ṗkα = −ωk2 Qkα with ωk = ck (2.118)

One clearly sees that these are Hamilton equations of harmonic oscillators labeled by kα
each with the Hamiltonian
1 2 
Hkα = Pkα + ωk2 Q2kα (2.119)
2
and Qkα and Pkα being the generalized coordinates and momenta. One could also obtain
this by inserting expansions (2.115) and (2.117) into Hr to find
X 1 X 2 
Hr = Hk α = Pkα + ωk2 Q2kα with ωk = ck (2.120)
2
kα kα
It is seen that indeed we represent Hr as a sum of decoupled oscillators with frequencies
given by the well know dispersion relation of the EM waves. There are two oscillators
with different polarizations for each k. Since ωk depends on the magnitude of k all the
oscillators with |k| = k have the same frequency.

Field wave functions and eigenstates. Photons appear

We now turn to the quantum mechanics of the radiation. It is the easiest to do this in the
decoupled eigenmodes of the field as encoded in the Hamiltonian (2.120). Instead of clas-
sical time dependent variables Qkα (t) and Pkα (t) we consider wave function Ψ({Qkα }, t)
which contains all the quantum information. This is ”extracted” by using operators for
every physical quantity which are build of two sets of basic operators of the ”coordinate”
and ”momentum”.
∂
Qkα → Q̂kα = Qkα ; Pkα → P̂kα = −iℏ (2.121)
∂Qkα
     
Pkα , Pk′ α′ = Qkα , Qk′ α′ = 0 , Qkα , Pk′ α′ = iℏδkk′ δαα′
Version of April 11, 2023 95

All in the usual way as in quantum mechanics of mechanical systems.

The time dependence of Ψ({Qkα }, t) is governed by the Schrödinger equation

∂Ψ({Qkα }, t)
iℏ = Ĥr Ψ({Qkα }, t) (2.122)
∂t
with the Hamiltonian operator Ĥr obtained from Eq. (2.120) by replacing in it the coor-
dinates and momenta with the corresponding operators
1 X 2 
Ĥr = P̂kα + ωk2 Q̂2kα (2.123)
2
kα
The most important solutions of the Schrödinger equation are the stationary states which
are the eigenstates of Ĥr
Ĥr Ψ({Qkα }) = EΨ({Qkα }) (2.124)
Since Ĥr is a sum of independent terms each representing an oscillator the eigenvalues of
Ĥr are sums of eigenenergies of independent oscillators
 
X 1 X
E{N } = ℏωk Nkα + ≡E+ ℏωk Nkα
kα 2
kα kα
X ℏωk (2.125)
E0 = , Nkα = 0, 1, 2, 3, ...
2
kα
The corresponding eigenfunctions are products
Y
Ψ{N } ({Qkα }) = ψN (Qkα ) , Nkα = 0, 1, 2, 3, ... (2.126)
kα kα
kα
where ψN (Qkα ) are the standard eigenfunctions of a harmonic oscillator, cf., Eq.(2.33),
kα
with unit mass and frequency ωk = ck.
As in our discussion of the string quantization (and actually in the quantization of
any linear dynamical system) we find that the EM field can be viewed as a collection of
energy quanta
ϵkα = ℏωk
in its normal modes. In the following sections we will show that these quanta have all
the characteristics of particles. They carry momentum, angular momentum and spin and
have energy-momentum relation of massless particles moving with the speed of light, cf.,
Eq.(2.141). These quantum particles are photons.
Focusing on the details we note that the EM modes are 3D vector waves. In our
developments we have chosen them as plane waves with wave vectors k and polarisations
λkα . But let us note that the eigenfrequencies of these modes ωk = ck and as a result
the energies ϵkα of the quanta depend only on the magnitude of k, i.e. on the wavelength
and not on the direction of the vector k and the polarisation of the modes.
As with the ordinary matter particles with e.g. ϵk = ℏ2 k 2 /2m this means that there
is a continuum degeneracy of the modes and therefore of the quantum mechanical states
Version of April 11, 2023 96

of the (free) photons. This degeneracy results in a freedom to change the basis states
with a given energy ϵ from the vector plane wave (like we did above) to e.g. spherical
(vector spherical!) or cylindrical (vector cylindrical!) etc waves. The quantum numbers
kα will then be replaced by appropriately changed ones like k, l, m replacing kx , ky , kz in
the scalar waves. A recent reference to the vector spherical waves is e.g. Ref.[4].

The wave function of the vacuum. The Casimir effect

The ground state of the EM field is the vacuum of the theory in the absence of matter
and other quantum fields. Its wave function is the product of Gaussians familiar from our
discussion of the guitar string, Eq. (2.34),
 
Y Y  ωk 1/4 X
Ψ{N =0} ({Qkα }) = ψ0 (Qkα ) = exp − ωk Q2kα /2ℏ (2.127)
kα πℏ
kα kα kα
It provides perhaps the simplest example of quantum vacuum fluctuations of field degrees
of freedom in a quantum field theory.
Can one observe these fluctuations? In a ground breaking paper, Ref.[1], Casimir
addressed this issue. He suggested that such fluctuations induce ”attraction between two
perfectly conducting plates”. On such plates the parallel to the plates components of the
electric field must vanish so that the field normal modes for which this doesn’t happen
will be excluded from the field degrees of freedom and consequently from the vacuum
fluctuations. This is schematically illustrated in Fig.2.4. The density of the normal
modes frequencies between the plates will be smaller than in the free space outside.
We have considered a one dimensional version of this effect in the context of the
quantum guitar string, Sec.2.1.4. We have shown there that it leads to an attractive force
between the (analogue of) the plates with the more narrow spacing than that of the other
part of the string. The same happens in the realistic 3D case with quantized EM field.
Casimir calculations predicted that an attractive force per unit area (pressure) at plate
separation a is given by
ℏπ 2 c
P =−
240a4
Note that the inverse quartic dependence on the distance is most unusual in physics. It
is also worth mentioning that for certain special combinations of the plates materials, the
Casimir force can be repulsive. The results obtained by Casimir were later extended to
various geometries of the plates and his predictions were confirmed experimentally, cf.,
Ref.[6].
Version of April 11, 2023 97

Figure 2.4: Schematic depiction of the normal modes of the field vibrations in the presence
of two plates which enforce vanishing of the field at the plates positions (Fig.1 from
Ref.[5]). This leads to the difference in the frequency densities of the field normal modes
between the plates as compared to the outside free space. This difference depends on the
distance between the plates and leads to the Casimir effect of plate attraction

Photon creation and annihilation operators. Field operators

In practice it is convenient to introduce creation and annihilation operators of photons in
the standard way
   
Q̂kα = ℏ/2ωk ↠+ âkα , P̂kα = i ℏωk /2 ↠− âkα
p p
kα kα
   
†
p p
âkα = 1/2ℏωk iP̂kα + ωk Q̂kα , â = 1/2ℏωk −iP̂kα + ωk Q̂kα (2.128)
kα
[âkα , ↠′ ′ ] = δkk′ δαα′ , [âkα , âk′ α′ ] = 0 = [↠, ↠′ ′ ]
kα kα k α
Using these operators we can write the Hamiltonian

ℏωk ↠âkα

X
Ĥr = E0 + (2.129)
kα
kα
and its eigenstates
N
Y (↠) kα
kα
Y
|{Nkα } >= |Nkα >= |0 > (2.130)
(Nkα !)1/2
kα kα

Great advantage of using â and ↠operators rather than P̂ and Q̂ in dealing with photons
is the simplicity of the ”action” of these operators on the ”number states”, i.e the states
with a fixed photon numbers in each normal mode. Schematically
√ √
â|n >= n|n − 1 > , ↠|n >= n + 1|n + 1 >
Version of April 11, 2023 98

In ”full glory”
q Y
âkα |{Nkα } >= Nkα |Nkα − 1 > |{Nk′ α′ } >
′
k ̸=k ,α′ ̸=α

q
a† |{Nkα } >=
Y
Nkα + 1|Nkα + 1 > |{Nk′ α′ } >
kα ′
k ̸=k ,α′ ̸=α

†
One says that the operators â and âkα create and destroy(annihilate) photons.
kα
It is useful to express the operators of the fields (2.115) as well as B̂(r) in terms of
the creation and annihilation operators

X 1/2 h i
ℏ
ÂT (r) = âkα λkα eik·r + ↠λkα e−ik·r =
2ϵ0 ωk Ω kα
kα
X 1/2
ℏ
= λkα eik·r (âkα + ↠)
2ϵ0 ωk Ω −k α
kα
(2.131)
X  ℏωk 1/2 h i
ÊT (r) = i âkα λkα eik·r − ↠λkα e−ik·r
2ϵ0 Ω kα
kα
X  ℏ 1/2 h i
B̂(r) = i âkα (k × λkα )eik·r − ↠(k × λkα )e−ik·r
2ϵ0 ωk Ω kα
kα
Once the longitudinal components of the fields Aop (r) and Eop (r) were separated the
remaining transverse parts ÂT (r) and ÊT (r) do not obey the canonical commutations
(2.82). Rather the delta function there gets replaced by the so called ”transverse” delta
function cf., Ref.[7], Ch.İII.A.1.

2.4.2 Photon momentum

In this subsection we will consider the operator of the momentum of the EM field Pfield .
Using it we will be able to show that photons are not just ”portions” of energy of the EM
field but that they also carry a ”corresponding” portion of its momentum. Moreover the
relation between the energies and momenta of these portions are as of massless particles
traveling with the speed of light.

Generators of translations in the matter-field system

Quantum mechanically it is probably the easiest to guess the expression of the momentum
by recalling that it is the generator of infinitesimal translations. For the interacting system
of particles (matter) and EM field described by the Hamiltonian (2.84) the operation of
Version of April 11, 2023 99

infinitesimal translation is the transformation

ˆ ˆ
e(i/ℏ)a·P Aop (r)e(−i/ℏ)a·P = Aop (r + a) ≈ Aop (r) + (a · ∇)Aop (r)
ˆ ˆ
e(i/ℏ)a·P Eop (r)e(−i/ℏ)a·P = Eop (r + a) ≈ Eop (r) + (a · ∇)Eop (r) (2.132)
ˆ ˆ
e(i/ℏ)a·P r̂a e(−i/ℏ)a·P = r̂a + a, a = 1, ..., N
Therefore the(vector) momentum operator P̂ = {P̂x , P̂y , P̂z } of the system should be such
that for each of its component P̂k , the commutators hold
[P̂k , Âj (r)] = −iℏ∂k Âj (r) , [P̂k , Êj (r)] = −iℏ∂k Ej (r) (2.133)
∂rj,a
[P̂k , rj,a ] = −iℏ = −iℏδkj (2.134)
∂rk,a
It is actually very easy to guess what such P̂ should be
N 3 Z
X ϵ0 X n o
P̂ = P̂matter + P̂field = p̂a + d3 r Êj (r)∇Âj (r) + h.c. (2.135)
a=1
2 j=1

where p̂a = −iℏ∇a , a = 1, ...N and the ”h.c.” abbreviation stands for ”hermitian
conjugate”.
Indeed the first term P̂matter has the required commutator with r̂a while commuting
with Eop (r) and Aop (r) and the second term P̂field commutes with r̂a and satisfies
3 Z
ϵ0 X n o
[P̂field,k , Âj (r)] = d3 r′ [Ên (r′ )∂k′ Ân (r′ ), Âj (r)] + ... =
2 n=1
3 Z
X iℏδnj
= −ϵ0 d3 r ′ δ(r′ − r))∂k′ Ân (r′ ) = −iℏ∂k Âj (r)
n=1
ϵ 0

3 Z
ϵ0 X n o
[P̂field,k , Êj (r)] = d3 r′ [Ên (r′ )∂k′ Ân (r′ ), Êj (r)] + ... =
2 n=1
3 Z
X −iℏδnj
= −ϵ0 d3 r′ Ên (r′ ))∂k′ δ(r′ − r)) = −iℏ∂k Êj (r)
n=1
ϵ 0

where in the last line we used the integration by parts.

Let us recall the classical expression for the conserved momentum in the presence of
the EM field. On the basis of the Maxwell and Newton equations Eqs.(2.65 - 2.67) one
finds that
N
X Z
P= ma va + ϵ0 d3 r E(r) × B(r) (2.136)
a=1
is the conserved total momentum of the field-matter system
dP
=0 (2.137)
dt
cf. p. 61, in Ref.[7] or a less formal text - Ref.[8]. In Appendix 2.5.3 below we show the
equivalence of the expressions (2.135) and (2.136).
Version of April 11, 2023 100

Momentum of the EM radiation

In the absence of charged particles we can use ∇ · Ê = 0 and replace Êj by the transversal
ÊT j in the field part of the momentum in (2.135). The same can be done with Âj in
it. Indeed the longitudinal part of  can be written as a gradient AL = ∇ξ. ˆ Thus it
contributes (after the replacement Êj → ÊT j ) the term
3 Z 3 Z
ϵ0 X n o ϵ X
0
n
ˆ + h.c.
o
d3 r ÊT j (r)∇ÂLj (r) + h.c. = d3 r ÊT j (r)∇∂j ξ(r)
2 j=1 2 j=1

in P̂field . This term is however zero as can be seen by integrating by parts in the right
hand side and using ∂j ÊT j = 0.
Thus we can write the momentum of the ”pure” radiation as
3 h
ϵ0
Z X i
P̂r = d3 r ÊT,j (r)∇ÂT,j (r) + h.c. (2.138)
2 j=1

Inserting expressions for the operators AT (r) and ET (r) one obtains

ℏk ↠âkα
X
P̂r = (2.139)
kα
kα

where we used kα ℏk = 0 7 . As it should P̂r commutes with the Hamiltonian Ĥr . Its
P
eigenvalues are X
P{N } = ℏk Nkα , Nkα = 0, 1, 2, 3, ... (2.140)
kα
kα
We can see that every state with Nkα quanta has momentum ℏkNkα so that every
energy quantum with ϵk = ℏωk carries momentum pk = ℏk. Using the dispersion relation
ωk = c|k| of the (classal) light waves (EM normal modes) we find the energy-momentum
relation of light quanta
ϵk = c|pk | (2.141)
i.e. of the massless particle moving with the light velocity.

2.4.3 Common states of light

Number states
These are just the eigenstates |{Nkα }⟩ of the Ĥr , cf., Eq. (2.129). Although most natural
from the formal point of view they are highly nonclassical and in fact are extremely hard
to produce ”on demand”8 . Number states are states of well defined energy but not of the
7 There is a subtle point here - this sum diverges and must be regularized by, say, assuming a cutoff at

some large kc .
8 E.g. M. Oxborrow and A.G. Sinclair, Contemp. Phys. 46, 173 (2005).
Version of April 11, 2023 101

states of well defined EM field. As an example consider a single mode of the electric fields,
i.e. just one term with a given k, α in the expression for ET in (2.131) and calculate
 1/2 h i
ℏωk
⟨Nkα |Ê(r)|Nkα ⟩ = iλkα ⟨Nkα | âkα eik·r − â+ e−ik·r |Nkα ⟩ = 0 (2.142)
2ϵ0 Ω kα

and
q
∆E = ⟨Nkα |Ê(r) · Ê(r)|Nkα ⟩ =
 1/2  1/2 (2.143)
ℏωk ℏωk
= ⟨Nkα |âkα â+ + â+ âkα |Nkα ⟩ = (2Nkα + 1)1/2
2ϵ0 Ω kα kα 2ϵ0 Ω

so that the everage value of Ê is zero while the fluctuations grow with the number of
photons.

Quantum mechanics behind the classical EM field. Coherent states of light

The correct description of the world is quantum mechanical while the classical physics
is just an approximation. So it is natural to ask what is the quantum mechanical state
behind the classical EM field? Since the quantum mechanical operators of electric and
magnetic components Ê(r) and Â(r) of the field are non commuting there is no state in
which they both have definite values.
Under these restrictions the appropriate quantum state |Ψ(t)⟩ behind the classical EM
field must be such that the averages, i.e. the expectation values of the field operators

E(r, t) ≡ ⟨Ψ(t)|Ê(r)|Ψ(t)⟩ , A(r, t) ≡ ⟨Ψ(t)|Â(r)|Ψ(t)⟩

will be developing in time as solutions of the classical Maxwell equation and be ”classically
large” i.e. much larger than the quantum uncertainties i.e. the standard deviations of
these fields from the averages..
It is not hard to find the state with the above properties for a free EM field. Since
such a field can be represented as a collection of independents modes it is useful to start
by considering a simple case of just a single mode with a given wave number k and
polarisation λ. Concentrating on the electric field we have the operator
λ h i
Ê(r) = √ k ck Q̂k cos(k · r) − P̂k sin(k · r) (2.144)
Ωϵ0

The quantum mechanics of Ê(r) and its non non commutativity with Â(r) is ”encoded”
in the canonical non commuting pair of the operators P̂k , Q̂k . Their dynamics (for a free
field) is simple - just that of harmonic oscillator, cf., Eq. (2.123).
So the task is to find a quantum state of harmonic oscillator, i.e. solutions |ψ(t)⟩ of
the Schrödinger equation

∂|ψ(t)⟩ 1 2
p̂ + ω 2 q̂ 2


iℏ = ĥ|ψ(t)⟩ with ĥ = (2.145)
∂t 2
Version of April 11, 2023 102

for which the averages

q(t) = ⟨ψ(t)|q̂|ψ(t)⟩ , p(t) = ⟨ψ(t)|p̂|ψ(t)⟩

obey the classical equations of the harmonic oscillator

q̇ = p , ṗ = −ω 2 q (2.146)

and have smallest possible quantum uncertainties.

Such a state was first discussed by Schrödinger already in 1926 and has a name -
coherent state. Its common formal definition is that it is an eigenstate of the annihilation
operator
√
 
∂
â|α⟩ = α|α⟩ ⇒ ℏ + ω q ψα (q) = 2ℏωαψα (q) (2.147)
∂q
with eigenstates |α⟩ labeled by the eigenvalues α and where we used the coordinate rep-
resentation of â √
â = (ip̂ + ω q̂) / 2ℏω
Note that since â is non hermitian α’s are in general complex valued9 . Also note that for
α = 0 the coherent state is just a ground state of the harmonic oscillator
2
ψ0 (q) = Ae−ωq /2ℏ
, A = (ω/πℏ)1/4 (2.148)

The properties of the coherent state are discussed in the Appendix of this Chapter.
It is shown there that coherent state is a wave packet the dynamics of which is such that
the averages
q0 ≡ ⟨α|q̂|α⟩ , p0 ≡ ⟨α|p̂|α⟩ (2.149)
move along the corresponding classical trajectories with uncertainties obeying the mini-
mum uncertainties relation
∆q∆p = ℏ/2 (2.150)
It is useful to schematically present this picture in the classical phase space as is explained
in Fig. 2.5.
9 There are many unfamiliar features of |α⟩ states as a result of this. Like non orthogonality at different

α’s or over completeness. This will be partly covered in the Appendix

Version of April 11, 2023 103

Figure 2.5: Schematic representation of a coherent state and its motion as a smeared
distribution (like e.g. Wigner distribution) in a classical phase space, cf. Eq.(2.163). We
use here ℏ = 1 units. For comparison also a number state ψn (q) = ⟨q|n⟩ centered at the
phase space origin (⟨n|q̂|n⟩ = ⟨n|p̂|n⟩ = 0) is shown schematically.

Let us briefly consider how the EM field ”looks like” in a coherent state. Consider a
single mode (2.144) written in terms of the photon and assume it is in a coherent state
|αk (t)⟩. Then
 1/2 h i
ℏωk
⟨αk (t)|Ê(r)|αk (t)⟩ = iλk αk (t)eik·r − αk
∗
(t)e−ik·r =
2ϵ0 Ω
 1/2 (2.151)
2ℏωk
= λk |αk (0)| sin(ωk t − k · r − ϕk )
ϵ0 Ω
where we used the results (2.168, 2.169) from the Appendix. This expression for the
average Ê(r) has the form of a classical field. Its amplitude is controlled by |αk (0)|, cf.,
the radius of the classical trajectory in Fig. 2.5.
Calculating
ℏωk 
1 + 4|αk (0)|2 sin2 (ωk t − k · r − ϕk )

⟨αk (t)|Ê(r) · Ê(r)|αk (t)⟩ = (2.152)
2ϵ0 Ω
we obtain for the quantum fluctuations
q
ℏωk
∆E ≡ ⟨αk (t)|Ê(r) · Ê(r)|αk (t)⟩ − ⟨αk (t)|Ê(r)|αk (t)⟩2 = (2.153)
2ϵ0 Ω
which is independent of the magnitude of the average, cf., again Fig. 2.5. So for the
electric field ≫ than the quantum scale of the fluctuations ∆E the field can be viewed as
classical.
Version of April 11, 2023 104

Thermal light
Thermal radiation is radiation in thermal equilibrium, which means (as is usual in quan-
tum statistical physics) that this radiation is described not by a wave function (or rather
wave functional) but by the density matrix. This density matrix is diagonal in the eigenen-
ergy basis X
ρ= w({Nkα })|{Nkα }⟩⟨{Nkα }| (2.154)
{N
kα }
with the probabilities given by the Boltzmann factor
E({Nkα }) E({Nkα })
   
1 X
w({Nkα }) = exp − , Z(T ) = exp −
Z(T ) T T
{N
kα }
which of course is equivalent to saying that the radiation power follows the Plank law.
Just to remind - by using X
E({Nkα }) = ℏωk Nkα
kα
separating exponentials in w({Nkα }) and Z(T ) into products with different kα and sum-
ming over Nkα for each kα in Z(T ) one obtains
Y
w(Nkα ) with w(Nkα ) = 1 − e−βℏωk exp (−βℏωk Nkα )


w({Nkα }) = (2.155)
kα
and β = (kB T )−1 . The average energy per mode is
X
ℏωk
⟨E⟩kα = ℏωk Nkα w(Nkα ) = ℏωk ⟨N ⟩kα = βℏω
e k −1
N
kα
and the Plank spectral energy density
XZ d3 k 2k 2 dkdγ 8πhν 3 hν
Z
ℏωk
dn = ⟨E⟩kα 3
= βℏωk − 1 3
= 3 βhν − 1
dν
α γ∈4π (2π) e γ∈4π (2π) c e

with hν = ℏω.
Such a spectrum is an idealization of a radiation spectrum emitted by matter sources
which by themselves are in a thermal equilibrium and moreover the radiation which they
emit ”has enough time” inside the matter to reach equilibrium with it. The major factors
”distorting” such spectra are layers of matter (like sun and earth atmospheres) between
the equilibrated matter-radiation system and the observer. If such layers have different
temperature and are too thin the light will ”have no time” to re-equilibrate as it passes
through them. The layers will just absorb some of the passing radiation at particular wave
lengths depending on their chemical composition. This will produce the corresponding
”absorption lines” in the radiation spectrum. Hot excited atoms, molecules, etc, inside the
layers will also emit and add non equilibrated light at particular wavelengths producing
the ”emission lines”. Example of the observed solar radiation spectrum, cf., Fig. 2.6
provides a good illustration of these features.
Version of April 11, 2023 105

Figure 2.6: Above - the Plank black body spectrum and its modifications in real world.
Below - discrete absorption lines on the background of the continuum solar light spectrum

2.4.4 Photon angular momentum and spin

Using the Maxwell and Newton equations (2.65), (2.67) together with the charge cur-
rent and density (2.66) one can show (cf., Ch 1A in Ref.[7]) that the conserved angular
momentum of the matter-field system is
N
X Z
J= ra × ma va + ϵ0 d3 r r × [E(r) × B(r)] (2.156)
a=1

Comparing with the expression (2.136) for the matter-field momentum both terms have
intuitively clear meaning.
It is important to note that as is usual with the definition of angular momentum the
expression (2.156) refers to a specific point - the origin of the chosen coordinate system -
with respect to which J is calculated. This of course can be easily changed by replacing
ra → ra − r0 and r → r − r0 with an arbitrary vector r0 in both terms of J respectively.
This change leads to a straightforward generalisation of the classical mechanics relation
for such transformations of angular momenta

J ⇒ J′ = J − r0 × P
Version of April 11, 2023 106

Generators of rotations in the matter-field system

Consider infinitesimal rotations of the coordinate system
r → r′ = r + δr = r + δα × r
where as usual the magnitude of the vector δα is equal the rotation angle and it is directed
along the axis of rotation (right hand rule). We want to determine how the wave functional
Ψ[A(r), r1 , ..., rN ] changes under this transformation.
Let us start by recalling that a scalar field change obeys the intuitive rule
ϕ(r) → ϕ′ (r′ ) = ϕ(r)
saying that the values of the rotated field ϕ′ at rotated points r′ are the same as non
rotated field ϕ in original points r. Using r = r′ − δr and dropping the prime in r′ on
both sides can write
ϕ′ (r) = ϕ(r − δr) ≈ ϕ(r) − δr · ∇ϕ(r) =
= ϕ(r) − (δα × r) · ∇ϕ(r) = [1 − δα · (r × ∇)]ϕ(r) =
i
= [1 − δα · l̂]ϕ(r) with l̂ = −iℏ[r × ∇] (2.157)
ℏ
For a vector field one also has to rotate the field itself
A(r) → A′ (r′ ) = [1 + δα×]A(r′ − δα × r)
which gives (after dropping the prime on r′ )
A′ (r) ≈ A(r) + δA(r) = A(r) + δα × A(r) − [(δα × r) · ∇]A(r) =
= A(r) + δα × A(r) − [δα · (r × ∇)]A(r) (2.158)
As in the scalar field case the last term corresponds to the ”orbital” rotation with l̂ =
−iℏr × ∇ while in the Appendix below we show that the additional second term is (not
surprisingly) a rotation of the components of the vector A with spin one matrices.
Let us now examine what happens to a wave functional Ψ[A(r)] when its argument is
transformed as in (2.158). To simplify things we leave out the particle coordinates {ra }
aince the part of the rotation generator for them is obvious. Have
δΨ[A(r)]
Z
Ψ[A(r)] → Ψ[A(r) + δA(r)] ≈ Ψ[A(r)] + d3 r δA(r) · =
δA(r)
 
δ
Z
= 1 + d3 r δA(r) · Ψ[A(r)] =
δA(r)
 
i
Z
3
= 1 − ϵ0 d r δA(r) · Eop (r) Ψ[A(r)]
ℏ
where we used the expression for Eop (r) as defined in (2.81). Using the explicit form of
δA from (2.158) we can write for the integral in the second term
Z Z
ϵ0 d3 r δA(r) · Eop (r) = ϵ0 d3 r {δα × A(r) − [δα · (r × ∇)]A(r)} · Eop (r)
Z ( )
X
= ϵ0 δα · d3 r A × Eop − [(r × ∇)Ai (r)]Êi
i
Version of April 11, 2023 107

From this we can read off the generator of rotations for the field part. It can be written
as a sum of two parts - spin and orbital

Ĵf ield = L̂f ield + Ŝf ield (2.159)

with
Z X
L̂f ield = ϵ0 d3 r Êi (r × ∇)Âi
i
Z
Ŝf ield = ϵ0 d3 r [Eop × Aop ] (2.160)

where we indicated that A in this expression should be regarded as operator (although it

is diagonal, Aop = A, in the representation of Ψ[A(r)].
Note that in the expressions for L̂f ield and Ŝf ield we were free to commute Êi compo-
nents to the left. Indeed in L̂f ield the commutator of (r × ∇)Âi and Êi is proportional to
the derivative of the delta function δ(r − r′ )
h i iℏ
(r × ∇)Âi (r), Êi (r′ ) =− (r × ∇)δ(r − r′ ) r=r′ = 0
r=r ′ ϵ0

which vanishes at r = r′ . In Ŝf ield only different i.e. commuting components Âi and Êj
with i ̸= j enter in their vector product.

2.4.5 Photon parity and photon statistics

The vector potential A(r) is a polar vector - it changes it sign under parity transformation

A(r) → −A(−r) (2.161)

This property of A(r) is the basis of the statement that the photon, i.e. the quanta of the
vibrations of A(r) have negative parity. We note that at this stage this is a fairly cryptic
statement which becomes clear when photon emission by matter system is studied (cf.,
later in the course).
Photons are bosons! This too is a somewhat cryptic statement at this stage. It will
become clear when dealing with the second quantitation formalism of the Schrödinger
field in relation to quantum many body systems. Here we only remark that one can have
any number of photons in the same state, i.e. in the same mode characterised by k, α
quantum numbers.
One can calculate the commutator of the operators of the electric and magnetic fields
ÊT (r) and B̂(r) = ∇ × ÂT (r) and find that they do not commute. This have all the usual
quantum mechanical consequences. In fact in tutorials and home works we/you will deal
with issues related to questions like ”what is the electric/magnetic field of a photon?”
Version of April 11, 2023 108

2.5 Appendix
2.5.1 Details of the standing to traveling waves transformation
What does the transformation Eq. (2.43) achieve
Let us start by noting that the Hamiltonian for a given kν has the same form in the new
variables
1 X 2 1
Hν = (P + ων2 Q2i,ν ) = [(Pk2 + ωk2 Q2k ) + (P−k
2
+ ωk2 Q2−k )]
2 i=1,2 i,ν 2

and due to their canonicity (cf., below) the dynamical equations for Q±k , P±k are the
same as for Qi,ν , Pi,ν , i = 1, 2 so their solutions have the same form as in Eq. (2.42). Now
both terms in this solution give traveling waves when inserted in the expansion (2.44).
Indeed have for the first terms in these solutions when inserted in the expansion for ϕ(x)
r r
1X 1X
Qk (0) [sin kx cos ωt − cos kx sin ωt] = Qk 0) sin(kx − ωt)
L L
k k

and for the second terms

r r
1 X Pk (0) 1 X Pk (0)
[sin kx sin ωt + cos kx cos ωt] = cos(kx − ωt)
L ων L ω
k k

These traveling waves are ”running” in the positive or negative x-direction depending on
the sign of k.

Verifying canonicity
Our transformation from the standing waves expansion (2.40) to the traveling waves (2.44)
amounted to transforming from Qi,ν , Pi,ν phase space variables to Q±k , P±k , Eq. (2.43).
Let us now check the canonicity of this transformation.
Let us recall that in a mechanical system described by a set of generalised coordinates
and momenta {q, p} the transformation to a canonically conjugate set {Q, P } must satisfy
X X
pi dqi = Pk dQk + dF
i k

where dF denote a complete differential. In our case the set {q, p} is Qi,ν , Pi,ν and we are
Version of April 11, 2023 109

transforming to Q±k , P±k . We obtain

X 1
Pi,ν dQi,ν = [(Pk − P−k )d(Qk − Q−k ) − (Qk + Q−k )d(Pk + P−k )] =
i
2
1
= [Pk dQk + P−k dQ−k − Pk dQ−k − P−k dQk −
2
− Qk dPk − Q−k dP−k − Qk dP−k − Q−k dPk ] =
1
= Pk dQk + P−k dQ−k − (Pk dQk + P−k dQ−k + Qk dPk + Q−k dP−k ) −
2
1
− (Pk dQ−k + P−k dQk + Qk dP−k + Q−k dPk ) =
2
1
= Pk dQk + P−k dQ−k − d(Pk Qk + P−k Q−k − Pk Q−k − P−k Qk )
2
It is instructive also to verify the canonicity of the general transformation (2.44).
In this case the set {q, p} is {ϕ(x), π(x)}, the sum over i is integral over x and we are
transforming to Qk , Pk . So we have
L
∂ϕ(x, t)
Z
dxπ(x, t) =
0 ∂t
1X L
Z
= dx [sin k ′ x Pk′ (t) − v|k ′ | cos k ′ x Qk′ (t)] ×
L ′ 0
kk
 
1
× sin kx Q̇k (t) + cos kx Ṗk (t) =
v|k|
X1h i
= Pk (t)Q̇k (t) − Qk (t)Ṗk (t) =
2
k
X 1X d
= Pk (t)Q̇k (t) − [Qk (t)Pk (t)]
2 dt
k k

which shows the canonicity of Pk and Qk .

2.5.2 Details of the coherent states

Useful averages. Minimum uncertainty
It easy to find an explicit solution of the equation (2.147). But before doing that it is
useful first to calculate the following averages
p p p
q0 ≡ ⟨α|q̂|α⟩ = ℏ/(2ω)⟨α|(â + â+ )|α⟩ = ℏ/(2ω)(α + α∗ ) = 2ℏ/ω Re α
p p √ (2.162)
p0 ≡ ⟨α|p̂|α⟩ = i ℏω/2⟨α|(â+ − â)|α⟩ = i ℏω/2(α∗ − α) = 2ℏω Im α

which give p p
α= ω/2ℏ q0 + i 1/(2ℏω) p0 (2.163)
Version of April 11, 2023 110

Also have
⟨α|q̂ 2 |α⟩ = (ℏ/2ω)⟨α|â2 + ââ+ + â+ â + (â+ )2 |α⟩ =
= (ℏ/2ω)⟨α|â2 + 2â+ â + 1 + (â+ )2 |α⟩ = (2.164)
∗ 2 2
= (ℏ/2 ω)[(α + α ) + 1) = ⟨α|q̂|α⟩ + ℏ/(2ω)

and
⟨α|p2 |α⟩ = −(ℏω/2)⟨α|â+2 − â+ â − ââ+ + â2 |α⟩ =
(2.165)
= −(ℏω/2)[(α − α∗ )2 − 1] = ⟨α|p̂|α⟩2 + ℏω/2

which shows that the coordinate and momentum uncertainties in this state are indepen-
dent of α
p p
∆q ≡ ⟨α|q̂ 2 |α⟩ − ⟨α|q̂|α⟩2 = ℏ/(2ω)
p p (2.166)
∆p ≡ ⟨α|p̂2 |α⟩ − ⟨α|p̂|α⟩2 = ℏω/2

which in turns means that for large (classical) values of q0 and p0 , i.e. for large |α|, cf.,
Eq. (2.163), the quantum uncertainties are negligible. The actual values of ∆q and ∆p
show that |α⟩ is a minimum uncertainty state, Eq. (2.150).

Dynamics of coherent states

Let is now consider the dynamics of a coherent state, i.e. find

|α(t)⟩ ≡ e−iĥt/ℏ |α⟩ with ĥ = ℏω(â+ â + 1/2)

For this we use the Heisenberg representation â(t) = eiĥt/ℏ â e−iĥt/ℏ of â and the corre-
sponding Heisenberg equation which is easily solved
∂â(t)
iℏ = −ĥâ(t) + â(t)ĥ = ℏω[−â+ (t)â(t)â(t) + â(t)â+ (t)â(t)] =
∂t (2.167)
= ℏωâ(t) ⇒ â(t) = âe−iωt

This gives

â|α(t)⟩ = âe−iĥt/ℏ |α⟩ = e−iĥt/ℏ â(t)|α⟩ = e−iĥt/ℏ âe−iωt |α⟩ = αe−iωt |α(t)⟩ (2.168)

which shows that |α(t)⟩ remains a coherent state with

α(t) = αe−iωt (2.169)

In terms of the corresponding q0 (t) and p0 (t)

p p
q0 (t) = 2ℏ/ω Re α(t) = 2ℏ/ω[ Re α cos ωt + Im α sin ωt] = q0 cos ωt + (p0 /ω) sin ωt
√ √
p0 (t) = 2ℏω Im α(t) = 2ℏω[ Im α cos ωt − Re α sin ωt] = p0 cos ωt − ωq0 sin ωt

which coincide with the solution of the classical equations (2.146).

Version of April 11, 2023 111

Explicit expressions. Ground state of a shifted harmonic oscillator

Using the explicit expression (2.148) for the coherent state at α = 0 it is easy to find
solutions of Eq. (2.147) for a general α by using the decomposition (2.163) in (2.147)
    
∂ ∂
ℏ + ω q ψα (q) = [ωq0 + i p0 ]ψα (q) ⇒ ℏ − ip0 + ω(q − q0 ) ψα (q) = 0
∂q ∂q

and noticing that this equation is similar to the one with α = 0, Eq. (2.148), but with a
shift q → q − q0 and a p0 dependent phase

ψα (q) = A exp{−[ω(q − q0 )2 + ip0 ]/ℏ} , A = (ω/πℏ)1/4 (2.170)

It clearly can be regarded as a ground state of a shifted harmonic oscillator, i.e. of

1
(p̂ − p0 )2 + ω 2 (q̂ − qo )2

ĥ = (2.171)
2
This observation is important for a qualitative discussion of the laser light.
Let us note that the coherent state can also be written as an expansion in a complete
set of number states, i.e. the harmonic oscillator eigenstates |n⟩
∞
X
|α⟩ = cn |n⟩
n=0

Acting with â we obtain

∞ ∞
X X √ X
â|α⟩ = cn â|n⟩ = cn n|n − 1⟩ = α cn |n⟩ =
n=0 n=0 n=0
∞
X √
=α ck−1 |k − 1⟩ ⇒ ncn = αcn−1
k=1
∞
αn X αn
⇒ cn = √ c0 ⇒ |α⟩ = c0 √ |n⟩
n! n=0 n!

Find c0 from normalization

∞
X |α|2 2 2
1 = ⟨α|α⟩ = |c0 |2 = |c0 |2 e|α| ⇒ c0 = e−|α| /2

n=0
n!

so
∞
−|α|2 /2
X αn
|α⟩ = e √ |n⟩ (2.172)
n=0 n!
It is also easy to calculate the overlap
2
/2 −|β||2 /2 α∗ β 2
⟨α|β⟩ = e−|α| e e ⇒ |⟨α|β⟩|2 = e−|α−β|
Version of April 11, 2023 112

showing non orthogonality of different |α⟩ states. The set |α⟩ is over-complete but satisfies
a useful resolution of unity relation
∞
d2 α
Z X
|α⟩⟨α| = |n⟩⟨n| = 1̂ with d2 α = d Re α d Im α
π n=0

which is easy to prove by using the expansion (2.172) and changing to polar coordinates
α = reiϕ , d2 α = rdrdϕ in the integral.

2.5.3 More on the EM field momentum

Relation to the classical expressions for the matter-field momentum
Classical expression (2.136) can be written
Z Z
Pf = ϵ0 d r E(r) × B(r) = ϵ0 d3 r E(r) × ∇ × A(r)
3
(2.173)

It is related to the integral of the Poynting vector, cf., the reference to the Feynman
lectures given above for the physics discussion of this result.
Let us write this expression in components (using the Levi-Civita tensor and the
summation convention)

(E × ∇ × A)i = ϵijk Ej ϵklm ∂l Am = ϵkij ϵklm Ej ∂l Am = (2.174)

= (δil δjm − δim δlj )Ej ∂l Am = Ej ∂i Aj − Ej ∂j Ai

so that
Z Z Z
ϵ0 d3 r (E × ∇ × A)i = ϵ0 d3 r (Ej ∂i Aj − Ej ∂j Ai ) = ϵ0 d3 r (Ej ∂i Aj + ∂j Ej Ai )
(2.175)
where we integrated by parts in the last equality. Using the Gauss law ∂j Ej = ρ/ϵ0 this
gives
Z Z Z N
X
(Pf )i = ϵ0 d3 r Ej (r)∂i Aj (r) + d3 rρ(r)Ai (r)) = ϵ0 d3 r Ej (r)∂i Aj (r) + qa Ai (ra )
a=1
(2.176)
PN
where we used ρ(r) = a=1 qa δ(r − ra ) to integrate.
Using this in the expression for the total momentum (2.136) (and restoring for better
clarity the summation symbol for the repeated index j) we obtain
N
X Z 3
X
P= pa + ϵ0 d3 r Ej (r)∇Aj (r) (2.177)
a=1 j=1

with
pa = ma va + qa A(ra ) (2.178)
This coincides with the expression (2.135).
Version of April 11, 2023 113

Field momentum in terms of the transverse components

Both terms in the expression (2.136) are separately gauge invariant. However the two
terms in the transformed expression (2.177) are not. Only their sum is. We can repair
this if we repeat the calculation (2.175) but first replacing A by AT in the starting left
hand side. This will lead to the same expression as (2.177) but with AT replacing A in it
N
X Z 3
X
3
P= (ma va + qa AT (ra )) + ϵ0 d r Ej (r)∇AT,j (r)
a=1 j=1

Now both terms are gauge invariant. We can moreover in the second term replace E by
ET . Indeed writing
Ej = ET,j + EL,j = ET,j − ∂j ϕ
and using
 
Z 3
X Z X3
d3 r ∂j ϕ(r)∇AT,j (r) = − d3 rϕ(r)∇  ∂j AT,j (r) = 0
j=1 j=1

we express
N
X Z 3
X
P= (ma va + qa AT (ra )) + ϵ0 d3 r ET,j (r)∇AT,j (r) (2.179)
a=1 j=1

In the absence of the charged matter (i.e. when all qa ’s are zero) the field part of this
momentum becomes the momentum of the free radiation as we have already derived in
(2.138).

2.5.4 More on the EM field angular momentum

Relation to the classical expression
See Ref.[7], Complement BI .

Spin 1 part of rotations of a vector field

We can write the 2nd term in Eq. (2.158) as
i i
[δα × A]j = ϵjkl δαk Al = − δαk skjl Al = − [δα · s]jl Al
ℏ ℏ
where the matrices
skjl = iℏϵjkl
are spin 1 matrices written in cartesian components basis x1 = x, x2 = y, x3 = z rather
than in the more familiar spherical components basis (xm , m = ±1, 0 )
1 1
x+1 = − √ (x + iy) , x−1 = √ (x − iy) , x0 = z
2 2
Version of April 11, 2023 114

i.e. xm ∼ rY1m (θ, ϕ).

One can easily verify that the commutators indeed have the correct form

[si , sj ] = iℏϵijn sn (2.180)

For this must prove that

[si , sj ]kl = −ℏ2 [ϵkim ϵmjl − ϵkjm ϵmil ]

is equal to
iℏϵijn snkl = iℏϵijn iℏϵknl = −ℏ2 ϵijn ϵknl
Have

ϵkim ϵmjl − ϵkjm ϵmil = (δkj δil − δkl δij ) − (δki δjl − δkl δji ) = δkj δil − δki δjl

which indeed is equal to

ϵijn ϵknl = ϵijn ϵlkn = δil δjk − δik δjl

Bibliography

[1] H. B. G. Casimir, Proc. K. Ned. Akad. Wet. 51, 793 (1948)

[2] T. H. Boyer, American Journal of Physics 71, 990 (2003)

[3] Action-angle Variables. University of Rochester.
[Link]
[4] G. Kristensson. Spherical Vector Waves. Lund, 2014
[Link]

[5] K. Kingsbury, The Casimir Effect, Version 3, A Comprehensive Exercise, 2009.

[Link]
[6] A. Stange, D. K. Campbell, and D.J. Bishop, Science and technology of the Casimir
effect. Physics Today 74, 1, 42 (2021);

[7] C. Cohen-Tannoudji, J. Dupont-Roc, G. Grynberg, Photons and Atoms: Introduc-

tion to Quantum Electrodynamics, 2004 Wiley.
[8] [Link]
.

115
Chapter 3

Photon-Matter Interactions

This Chapter is the continuation of the Chapter ”Quantized EM Field”. We will use the
quantum description of the EM field discussed there to provide several simple examples
of how photons are emitted and absorbed by quantum matter systems.

3.1 Interaction Hamiltonian

3.1.1 Separating the interaction terms
As was shown in the Chapter ”Quantized EM Field” the Hamiltonian operator of the EM
field interacting with (non relativistic) matter is
N
1 ϵ0
X Z h i
Ĥ = [p̂a − qa ÂT (ra )]2 + VCoul + ÊT (r)2 + c2 (∇ × ÂT (r))2 d3 r (3.1)
a=1
2ma 2

with
X 1/2 h i
ℏ
ÂT (r) = âkα λkα eik·r + ↠λkα e−ik·r (3.2)
2ϵ0 ωk Ω kα
kα
X  ℏωk 1/2 h i
ÊT (r) = i âkα λkα eik·r − ↠λkα e−ik·r (3.3)
2ϵ0 Ω kα
kα
and
N
1 X qa qb
VCoul =
8πϵ0 |ra − rb |
a̸=b

This expression can be written as

Ĥ = Ĥmatter + Ĥr + Ĥmatter−radiation interaction (3.4)

116
Version of April 11, 2023 117

with
N
X p̂2a
Ĥmatter = + VCoul (r1 , ..., rn )
a=1
2ma
ϵ0
Z h i
Ĥr = ÊT (r)2 + c2 (∇ × ÂT (r))2 d3 r (3.5)
2
Ĥmatter−radiation interaction = ĤI1 + ĤI2
and
N
X qa h i
ĤI1 = − p̂a · ÂT (ra ) + ÂT (ra ) · p̂a (3.6)
a=1
2ma
N
X qa2
ĤI2 = [ÂT (ra )]2 (3.7)
a=1
2m a

The expressions for ĤI1 and ĤI2 depend on the coordinates and momenta of the particles
and on the ”coordinates” ÂT (r) of the field. It is worth noting that the transversality of
AT means that pa and ÂT (ra ) commute
3 h
X i
p̂a,i , ÂT,i (ra ) = −iℏ∇a · ÂT (ra ) = 0
i=1

so that the interaction ĤI1 can be written as one term

N
X qa
ĤI1 = − ÂT (ra ) · p̂a (3.8)
a=1
ma

3.1.2 Adding spin and external fields

When matter particles have spins one must add spin degrees of freedom sa to the particle
coordinates ra . As a rule spinning particles have non zero magnetic moment µa 1 which
is parallel to the spin and follows its dynamics. The proportionality relation between the
corresponding operators is conventionally written
qa
µ̂a = ga sˆa (3.9)
2ma
where ga is the so called Lande factor or g-factor (see e.g. the appropriate section in the
Chapter ”Motion in External Electromagnetic Field”).
Particle magnetic moments interact with the magnetic field so one must add a new
term to the interaction Hamiltonian Ĥmatter−radiation interaction ,
N
X
ĤI3 = − µ̂a · B̂(ra ) (3.10)
a=1
1 This is obvious for charged particles but in fact also neutral particles with spin, e.g. molecules,

atoms, neutrons, etc, may have non zero µ due to the ”spinning” charges inside the overall neutral
system. Charged quarks in a neutron is an obvious example.
Version of April 11, 2023 118

with the operator of the magnetic field (cf., the Chapter ”Quantized EM Field”)

X  ℏ 1/2 h i
B̂(r) = i âkα (k × λkα )eik·r − ↠(k × λkα )e−ik·r (3.11)
2ϵ0 ωk Ω kα
kα
We have up to now considered a closed matter-EM field system. One often encounters
a situation in which in addition there are external fields acting on the matter particles.
Examples are Coulomb potential of a heavy nucleus acting on atomic electrons or external
magnetic field acting on electrons in Landau levels. Such external fields are to a good
approximation classical with prescribed space and time dependence. In their presence
the Hamiltonian (3.1) should be modified by adding external classical vector potential,
external scalar potential and external magnetic field. The full Hamiltonian will then have
the form 2

N i2
X 1 h
Ĥ = p̂a − qa Aexternal (ra , t) − qa ÂT (ra + VCoul +
a=1
2ma
N
X N
X
+ U external (ra , t) − µ̂a · Bexternal (ra , t) − (3.12)
a=1 a=1
N
ϵ0
X Z h i
− µ̂a · B̂(ra ) + ÊT (r)2 + c2 ∇ × ÂT (r) d3 r
a=1
2
N
1 X qa qb
VCoul =
8πϵ0 |ra − rb |
a̸=b

where we have also added the spin degrees of freedom interacting with external magnetic
fields via the particles magnetic moments.

3.1.3 Disentangling radiation from the matter degrees of freedom

The objects like ÂT (ra ) and B̂(ra ) in the expressions (3.6), (3.7) and (3.10) are operator
valued functions (fields) of operators (particle coordinates). It is easy and convenient to
disentangle this complicated dependence using the identities
Z Z
2 2
ÂT (ra ) = δ(r − ra )ÂT (r) d r , ÂT (ra ) = δ(r − ra )ÂT (r) d3 r
3

and Z
B̂(ra ) = δ(r − ra )B̂(r) d3 r

2 Note that external fields influence the radiation only via matter. There is no direct effect on the

dynamics of the radiation. This is a consequence of the linearity of the Maxwell equations.
Version of April 11, 2023 119

Using these one can write the interactions (3.6), (3.7) and (3.10) as
N
qa
Z X
ĤI1 = − d3 r [p̂ δ(r − ra ) + δ(r − ra )p̂a ] · ÂT (r) (3.13)
a=1
2ma a
N
qa2
Z X
ĤI2 = d3 r δ(r − ra )[ÂT (r)]2 (3.14)
a=1
2m a
Z N
X
3
ĤI3 = − d r µ̂a δ(r − ra ) · B̂(r) (3.15)
a=1

The 1st and the 3rd of these expressions have a simple form
Z
ĤI1 = − ĵ(r) · ÂT (r) d3 r (3.16)

and Z
ĤI3 = − m̂(r) · B̂(r) d3 r (3.17)

with current operator

N
1 X qa
ĵ(r) = [p̂ δ(r − ra )) + δ(r − ra )p̂a ]
2 a=1 ma a

and magnetization operator

N
X
m̂(r) = µ̂a δ(r − ra )
a=1

The second term ĤI2 simplifies when all the charges and masses of the particles are equal
q1 = q2 = ... = qN = q, m1 = m2 = ... = mN = m. Then
q
Z
ĤI2 = ρ̂(r)[ÂT (r)]2 d3 r (3.18)
m
with charge density operator
N
X
ρ̂(r) = qδ(r − ra )
a=1

3.1.4 Resulting insights

Matter creates, annihilates, scatters photons
Qualitative insights into the nature of the interaction terms is gained if the expressions
for the fields ÂT (r) and B̂(r) in terms of the photon creation and annihilation operators
Version of April 11, 2023 120

written in the form3

X 1/2
ℏ
ÂT (r) = λkα eik·r (âkα + ↠) (3.19)
2ϵ0 ωk Ω −kα
kα
X  ℏ 1/2  
B̂(r) = i (k × λkα )eik·r âkα + ↠(3.20)
2ϵ0 ωk Ω −kα
kα

are inserted in Eqs. (3.13 - 3.15). The interaction term ĤI1 takes the form

X 1/2
ℏ
ĤI1 = − ( ĵ−k · λkα )(âkα + ↠) (3.21)
2ϵ0 ωk Ω −k α
kα
with
N
qa h
Z i
ĵ(r)e−ik·r d3 r = p̂a e−ik·ra + e−ik·ra p̂a
X
ĵk = (3.22)
a=1
2ma

It is seen that to 1st order 4 this interaction acts by creating or annihilating single photons
with (not surprising but worth noting) opposite signs of the momentum ℏk. It is also
important to note that (as will become clearer later and especially in the chapter on
Second Quantization) the expression

λkα eik·r

can often be regarded as a photon wave function having definite momentum p = ℏk and
polarization λα .
Inserting the expression for B̂(r) into the interaction ĤI3 , Eq.(3.15), one obtains

X  ℏ 1/2
ĤI3 =− i [m̂−k · (k × λkα )] (âkα + ↠) (3.23)
2ϵ0 ωk Ω −k α
kα
with
Z N
m̂(r)e−ik·r d3 r = µ̂a e−ik·ra
X
m̂k = (3.24)
a=1

One observes that this interaction term also creates or annihilates one photon in 1st order.
The difference with ĤI1 is that in the former case the photon creation or annihilation was
”accompanied” with the ”action” on the matter variables of the corresponding component
3 To simplify expressions we assume here and in the following that the polarization vectors for k and

−k modes are chosen to be the same λkα = λ−kα .

4 By ”to 1st order” here and in the following we mean that the interaction acts one time on a wave

function. Note that in solving the Schrödinger equation the Hamiltonian acts ”infinitely many times” so
to speak. This can be seen by viewing the time evolution
ψ(t) = exp[−(i/ℏ)Ĥ(t − t0 )]ψ(t0 ) = [1 + (−i/ℏ)Ĥ(t − t0 ) + (−i/ℏ)2 Ĥ 2 (t − t0 )2 + ...]ψ(t0 )
Version of April 11, 2023 121

ĵ−k of the current operator projected on the photon polarization λkα . In ĤI3 this action
is replaced with m̂−k component of the magnetization density m̂(r) projected on k×λkα .
Turning now to the ĤI2 , Eq. (3.14), we note that the presence of the square [ÂT (r)]2
means that the creation and annihilation operators will appear in this expression in the
products
âkα âk′ α′ , ↠↠′ ′ , ↠âk′ α′ , âkα ↠′ ′
kα k α kα kα
which shows that these interaction terms in 1sr order either create or destroy two photons
or simultaneously create and destroy a photon with different momentum and polarization.

Matter ”shifts”, ”mixes” the radiation oscillators

Let us recall that ÂT (r) is written in terms of the running plane waves as
 
1 X 1
AT (r) = √ Qk cos(k · r) − P sin(k · r)
Ωϵ0 ωk k
k
where λkα ’s are fixed polarization vectors orthogonal to k. Recalling also that Ĥr is the
sum of the normal modes oscillators
1 X 2 
Ĥr = P̂kα + ωk2 Q̂2kα (3.25)
2
kα
we find that in terms of P̂k,α ’s and Q̂kα the Hamiltonian is written
1 X 2  X 
Ĥ = Ĥmatter + P̂kα + ωk2 Q̂2kα + Ŝkα Q̂kα + Ĉkα P̂kα + ĤI2 + ĤI3 (3.26)
2
kα kα
with
1
Z
Ŝkα = − √ λkα · ĵ(r) cos(k · r) d3 r (3.27)
Ωϵ0
1
Z
Ĉkα = √ λkα · ĵ(r) sin(k · r)d3 r
Ωϵ0 ω 2
Schematically one can say that via the ĤI1 interaction the matter causes shifts of the
oscillators of the radiation normal modes. The shift is in both the coordinates Qkα and
momenta Pkα . For fixed classical Skα and Ckα each oscillator gets shifted
 2 2 
1 2 2 2
 1  (0) 2

(0) (0)
P̂kα + ω Q̂kα → P̂kα − P + ωk Q̂kα − Q +E (3.28)
2 2 kα kα kα
(0) (0) (0)
with P ,Q and E determined by Skα and Ckα in an obvious way. Of course in
kα kα kα
a real situation Skα and Ckα are dynamical and quantized.
Let us also note that the interaction term ĤI3 may schematically be viewed in a similar
way as we outlined above for ĤI1 since it is linear in B̂(r) = ∇ × ÂT (r) and therefore in
Qkα and Pkα variables.
The interaction term ĤI2 on the other hand is quadratic in ÂT (r). Its dependence on
the field normal modes variables is therefore quadratic depending on products Qkα Qk′ α′ ,
Pkα Pk′ α′ and Qkα Pk′ α′ mixing the normal modes kα’s already in the 1st order.
Version of April 11, 2023 122

Generation of coherent states. Schematic model of a laser

Let us recall the properties of the coherent states which were discussed in the Section 6.1.3.
of the Quantized EM Field chapter. It was shown there that such states can be viewed as
ground states of a shifted harmonic oscillator. Turning to the expression (3.28) we notice
that if just one photon mode kα is selected and the current which ”feeds” this mode is
external, constant in time and classical then the lowest eigenstate of the corresponding
Hamiltonian will be a coherent state.
Such a Hamiltonian can actually be used as a simplest schematic model to begin under-
standing the quantum mechanics of the light emitted by a laser. Selecting a single mode is
modeling (in the simplest way) of the laser resonator. The classical external current is (a
very much simplified description of ) the source of excitations of the electric charges which
de-excite by emitting photons into the resonator mode. This shifted harmonic oscillator
model obviously is extremely schematic and misses many important laser features and
details. It nevertheless correctly indicates that a simple reasonable approximation to the
state of light which (one mode) laser emits is a coherent state.

3.2 Emission and Absorption of Photons

In this section we discuss the details of quantum mechanical description of photon emis-
sion and absorption. We will do this treating the radiation-matter interaction using the
perturbation theory and will limit ourselves to the leading 1st order terms. As should be
clear from our discussion above the relevant terms for such 1st order processes are ĤI1
and ĤI3 . We will begin by considering only the effect of ĤI1 i.e. photon emission and
absorption resulting from the change of the state of the electric current of the matter sys-
tem. Classically this would correspond to emission of radiation by an alternating current
(like e.g. in a simple antenna). The treatment of the photon emission by changing the
spin states of matter, i.e. the effect of the ĤI3 interaction term will fit naturally in the
discussion of these processes in relation to the changes of the current magnetic moment,
cf., Section 3.2.6.
Following this introduction we will begin by considering the Hamiltonian

Ĥ = Ĥ0 + ĤI1 (3.29)

where the unperturbed part is

ℏωk ↠âkα

X
Ĥ0 = Ĥmatter + (3.30)
kα
kα
P
and where we dropped the constant vacuum energy term Evacuum = (1/2) kα ℏωk .
Version of April 11, 2023 123

3.2.1 Paradigm of spontaneous emission of radiation - discrete

matter level coupled to a photon continuum
Unperturbed energies
We assume that we know how to solve the matter Hamiltonian i.e. that we know its
eigenstates and the corresponding eigenvalues

Ĥmatter |n⟩ = En |n⟩ (3.31)

We therefore know the eigenstates of the unperturbed Ĥ0 , Eq. (3.30) ,

X
|n⟩|{Nk,α }⟩ with eigenenergies E(n, {Nkα }) = En + Nkα ℏωk (3.32)
k,α

We assume that (as is typical for atomic, molecular or nuclear systems) the low lying
matter eigenenergies in (3.32) form discrete system of levels following by higher lying
continuum states (like e.g. simplest hydrogen atom at rest5 ). Let us consider the sector
of unperturbed levels with zero photons

En + 0 photons

and compare to the corresponding levels in a one photon sector

En + 1 photon = En + ℏωk

It is important to note that

ℏωk = ℏck
form a continuum of levels because of essentially continuum values of k (for large quanti-
sation volume).
Plotting these energies, cf. Fig. 3.1, one can see discrete levels of the matter without
photons ”embedded” in the continuum of matter + one or more photon levels. The
simplest is e.g. the first excited matter level with no photons

|n = 1⟩|{0kα }⟩ with E(1, {0kα }) = E1 + 0 photons

vs the ground state E0 plus one photon

|n = 0⟩|1kα , {0k′ α′ }⟩ with E(0, 1kα , {0k′ α′ }) = E0 + ℏωk

continuum of levels.
5 We ignore at the moment the center of mass motion of the emitting. Its effects will be discussed

below
Version of April 11, 2023 124

...

...
...

...

...
<latexit sha1_base64="kQkDxPnVv0Mgc/tsfqEuJYOLy9w=">AAAB7XicbVBNS8NAEJ3Ur1q/qh69BIvgqSRS1GPRi8cK9gPaUDabTbt2sxt2J0Ip/Q9ePCji1f/jzX/jts1BWx8MPN6bYWZemApu0PO+ncLa+sbmVnG7tLO7t39QPjxqGZVpyppUCaU7ITFMcMmayFGwTqoZSULB2uHodua3n5g2XMkHHKcsSMhA8phTglZq9USk0PTLFa/qzeGuEj8nFcjR6Je/epGiWcIkUkGM6fpeisGEaORUsGmplxmWEjoiA9a1VJKEmWAyv3bqnlklcmOlbUl05+rviQlJjBknoe1MCA7NsjcT//O6GcbXwYTLNEMm6WJRnAkXlTt73Y24ZhTF2BJCNbe3unRINKFoAyrZEPzll1dJ66LqX1Zr97VK/SaPowgncArn4MMV1OEOGtAECo/wDK/w5ijnxXl3PhatBSefOYY/cD5/AL8Nj0E=</latexit>

<latexit sha1_base64="kQkDxPnVv0Mgc/tsfqEuJYOLy9w=">AAAB7XicbVBNS8NAEJ3Ur1q/qh69BIvgqSRS1GPRi8cK9gPaUDabTbt2sxt2J0Ip/Q9ePCji1f/jzX/jts1BWx8MPN6bYWZemApu0PO+ncLa+sbmVnG7tLO7t39QPjxqGZVpyppUCaU7ITFMcMmayFGwTqoZSULB2uHodua3n5g2XMkHHKcsSMhA8phTglZq9USk0PTLFa/qzeGuEj8nFcjR6Je/epGiWcIkUkGM6fpeisGEaORUsGmplxmWEjoiA9a1VJKEmWAyv3bqnlklcmOlbUl05+rviQlJjBknoe1MCA7NsjcT//O6GcbXwYTLNEMm6WJRnAkXlTt73Y24ZhTF2BJCNbe3unRINKFoAyrZEPzll1dJ66LqX1Zr97VK/SaPowgncArn4MMV1OEOGtAECo/wDK/w5ijnxXl3PhatBSefOYY/cD5/AL8Nj0E=</latexit>
<latexit sha1_base64="kQkDxPnVv0Mgc/tsfqEuJYOLy9w=">AAAB7XicbVBNS8NAEJ3Ur1q/qh69BIvgqSRS1GPRi8cK9gPaUDabTbt2sxt2J0Ip/Q9ePCji1f/jzX/jts1BWx8MPN6bYWZemApu0PO+ncLa+sbmVnG7tLO7t39QPjxqGZVpyppUCaU7ITFMcMmayFGwTqoZSULB2uHodua3n5g2XMkHHKcsSMhA8phTglZq9USk0PTLFa/qzeGuEj8nFcjR6Je/epGiWcIkUkGM6fpeisGEaORUsGmplxmWEjoiA9a1VJKEmWAyv3bqnlklcmOlbUl05+rviQlJjBknoe1MCA7NsjcT//O6GcbXwYTLNEMm6WJRnAkXlTt73Y24ZhTF2BJCNbe3unRINKFoAyrZEPzll1dJ66LqX1Zr97VK/SaPowgncArn4MMV1OEOGtAECo/wDK/w5ijnxXl3PhatBSefOYY/cD5/AL8Nj0E=</latexit>

<latexit sha1_base64="kQkDxPnVv0Mgc/tsfqEuJYOLy9w=">AAAB7XicbVBNS8NAEJ3Ur1q/qh69BIvgqSRS1GPRi8cK9gPaUDabTbt2sxt2J0Ip/Q9ePCji1f/jzX/jts1BWx8MPN6bYWZemApu0PO+ncLa+sbmVnG7tLO7t39QPjxqGZVpyppUCaU7ITFMcMmayFGwTqoZSULB2uHodua3n5g2XMkHHKcsSMhA8phTglZq9USk0PTLFa/qzeGuEj8nFcjR6Je/epGiWcIkUkGM6fpeisGEaORUsGmplxmWEjoiA9a1VJKEmWAyv3bqnlklcmOlbUl05+rviQlJjBknoe1MCA7NsjcT//O6GcbXwYTLNEMm6WJRnAkXlTt73Y24ZhTF2BJCNbe3unRINKFoAyrZEPzll1dJ66LqX1Zr97VK/SaPowgncArn4MMV1OEOGtAECo/wDK/w5ijnxXl3PhatBSefOYY/cD5/AL8Nj0E=</latexit>

<latexit sha1_base64="kQkDxPnVv0Mgc/tsfqEuJYOLy9w=">AAAB7XicbVBNS8NAEJ3Ur1q/qh69BIvgqSRS1GPRi8cK9gPaUDabTbt2sxt2J0Ip/Q9ePCji1f/jzX/jts1BWx8MPN6bYWZemApu0PO+ncLa+sbmVnG7tLO7t39QPjxqGZVpyppUCaU7ITFMcMmayFGwTqoZSULB2uHodua3n5g2XMkHHKcsSMhA8phTglZq9USk0PTLFa/qzeGuEj8nFcjR6Je/epGiWcIkUkGM6fpeisGEaORUsGmplxmWEjoiA9a1VJKEmWAyv3bqnlklcmOlbUl05+rviQlJjBknoe1MCA7NsjcT//O6GcbXwYTLNEMm6WJRnAkXlTt73Y24ZhTF2BJCNbe3unRINKFoAyrZEPzll1dJ66LqX1Zr97VK/SaPowgncArn4MMV1OEOGtAECo/wDK/w5ijnxXl3PhatBSefOYY/cD5/AL8Nj0E=</latexit>
E
<latexit sha1_base64="sxhrtqlJFrvCaQX4o+BHfbB9UbA=">AAAB6HicbVDLSgNBEOyNrxhfUY9eBoPgKexKUI9BETwmYB6QLGF20puMmZ1dZmaFEPIFXjwo4tVP8ubfOEn2oIkFDUVVN91dQSK4Nq777eTW1jc2t/LbhZ3dvf2D4uFRU8epYthgsYhVO6AaBZfYMNwIbCcKaRQIbAWj25nfekKleSwfzDhBP6IDyUPOqLFS/a5XLLlldw6ySryMlCBDrVf86vZjlkYoDRNU647nJsafUGU4EzgtdFONCWUjOsCOpZJGqP3J/NApObNKn4SxsiUNmau/JyY00nocBbYzomaol72Z+J/XSU147U+4TFKDki0WhakgJiazr0mfK2RGjC2hTHF7K2FDqigzNpuCDcFbfnmVNC/K3mW5Uq+UqjdZHHk4gVM4Bw+uoAr3UIMGMEB4hld4cx6dF+fd+Vi05pxs5hj+wPn8AZvbjNI=</latexit>

"2 + 1ph
<latexit sha1_base64="TNhW/9L1Lz4+mnSmSSmHDLaCUP0=">AAACBXicbVDJSgNBEO2JW4xb1KMeGoMgCGEmBPUY9OIxglkgM4SeTiVp0rPQXRMMQy5e/BUvHhTx6j9482/sLAdNfFDw+r0quur5sRQabfvbyqysrq1vZDdzW9s7u3v5/YO6jhLFocYjGammzzRIEUINBUpoxgpY4Eto+IObid8YgtIiCu9xFIMXsF4ouoIzNFI7f+wOmYJYC2leJXpOXYQHVEHqxP1xO1+wi/YUdJk4c1Igc1Tb+S+3E/EkgBC5ZFq3HDtGL2UKBZcwzrmJhpjxAetBy9CQBaC9dHrFmJ4apUO7kTIVIp2qvydSFmg9CnzTGTDs60VvIv7ntRLsXnmpCOMEIeSzj7qJpBjRSSS0IxRwlCNDGFfC7Ep5nynG0QSXMyE4iycvk3qp6FwUy3flQuV6HkeWHJETckYcckkq5JZUSY1w8kieySt5s56sF+vd+pi1Zqz5zCH5A+vzBzgpmGg=</latexit>

...

... ...
<latexit sha1_base64="kQkDxPnVv0Mgc/tsfqEuJYOLy9w=">AAAB7XicbVBNS8NAEJ3Ur1q/qh69BIvgqSRS1GPRi8cK9gPaUDabTbt2sxt2J0Ip/Q9ePCji1f/jzX/jts1BWx8MPN6bYWZemApu0PO+ncLa+sbmVnG7tLO7t39QPjxqGZVpyppUCaU7ITFMcMmayFGwTqoZSULB2uHodua3n5g2XMkHHKcsSMhA8phTglZq9USk0PTLFa/qzeGuEj8nFcjR6Je/epGiWcIkUkGM6fpeisGEaORUsGmplxmWEjoiA9a1VJKEmWAyv3bqnlklcmOlbUl05+rviQlJjBknoe1MCA7NsjcT//O6GcbXwYTLNEMm6WJRnAkXlTt73Y24ZhTF2BJCNbe3unRINKFoAyrZEPzll1dJ66LqX1Zr97VK/SaPowgncArn4MMV1OEOGtAECo/wDK/w5ijnxXl3PhatBSefOYY/cD5/AL8Nj0E=</latexit>

"2
<latexit sha1_base64="//t5RYYkC+Rgr3jWePziRNIVhPE=">AAAB9HicbVDLTgJBEOzFF+IL9ehlIjHxRHYJUY9ELx4xkUcCGzI79MKE2dl1ZpaEEL7DiweN8erHePNvHGAPClbSSaWqO91dQSK4Nq777eQ2Nre2d/K7hb39g8Oj4vFJU8epYthgsYhVO6AaBZfYMNwIbCcKaRQIbAWju7nfGqPSPJaPZpKgH9GB5CFn1FjJ746pwkRzEctepVcsuWV3AbJOvIyUIEO9V/zq9mOWRigNE1Trjucmxp9SZTgTOCt0U40JZSM6wI6lkkao/eni6Bm5sEqfhLGyJQ1ZqL8npjTSehIFtjOiZqhXvbn4n9dJTXjjT7lMUoOSLReFqSAmJvMESJ8rZEZMLKFMcXsrYUOqKDM2p4INwVt9eZ00K2Xvqlx9qJZqt1kceTiDc7gED66hBvdQhwYweIJneIU3Z+y8OO/Ox7I152Qzp/AHzucP7B+SNQ==</latexit>

<latexit sha1_base64="kQkDxPnVv0Mgc/tsfqEuJYOLy9w=">AAAB7XicbVBNS8NAEJ3Ur1q/qh69BIvgqSRS1GPRi8cK9gPaUDabTbt2sxt2J0Ip/Q9ePCji1f/jzX/jts1BWx8MPN6bYWZemApu0PO+ncLa+sbmVnG7tLO7t39QPjxqGZVpyppUCaU7ITFMcMmayFGwTqoZSULB2uHodua3n5g2XMkHHKcsSMhA8phTglZq9USk0PTLFa/qzeGuEj8nFcjR6Je/epGiWcIkUkGM6fpeisGEaORUsGmplxmWEjoiA9a1VJKEmWAyv3bqnlklcmOlbUl05+rviQlJjBknoe1MCA7NsjcT//O6GcbXwYTLNEMm6WJRnAkXlTt73Y24ZhTF2BJCNbe3unRINKFoAyrZEPzll1dJ66LqX1Zr97VK/SaPowgncArn4MMV1OEOGtAECo/wDK/w5ijnxXl3PhatBSefOYY/cD5/AL8Nj0E=</latexit>
"1 + 1ph
<latexit sha1_base64="vyxIAp+cjur8i6ZWuyZQrHTYdlg=">AAACBXicbVDLSgNBEJz1GeNr1aMeBoMgCGFXgnoMevEYwTwgG8LspJMMmX0w0xsMSy5e/BUvHhTx6j9482+cJHvQxIKGmqpuprv8WAqNjvNtLS2vrK6t5zbym1vbO7v23n5NR4niUOWRjFTDZxqkCKGKAiU0YgUs8CXU/cHNxK8PQWkRhfc4iqEVsF4ouoIzNFLbPvKGTEGshTQvl55RD+EBVZC6cX/ctgtO0ZmCLhI3IwWSodK2v7xOxJMAQuSSad10nRhbKVMouIRx3ks0xIwPWA+ahoYsAN1Kp1eM6YlROrQbKVMh0qn6eyJlgdajwDedAcO+nvcm4n9eM8HuVSsVYZwghHz2UTeRFCM6iYR2hAKOcmQI40qYXSnvM8U4muDyJgR3/uRFUjsvuhfF0l2pUL7O4siRQ3JMTolLLkmZ3JIKqRJOHskzeSVv1pP1Yr1bH7PWJSubOSB/YH3+ADaWmGc=</latexit>

"1 <latexit sha1_base64="ncJvIIziYBeq+ZKhyu3Th13DerI=">AAAB9HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE1GPRi8cK9gPaUDbbSbt0s4m7m0IJ/R1ePCji1R/jzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU1HGqGDZYLGLVDqhGwSU2DDcC24lCGgUCW8Hobua3xqg0j+WjmSToR3QgecgZNVbyu2OqMNFcxLLn9coVt+rOQVaJl5MK5Kj3yl/dfszSCKVhgmrd8dzE+BlVhjOB01I31ZhQNqID7FgqaYTaz+ZHT8mZVfokjJUtachc/T2R0UjrSRTYzoiaoV72ZuJ/Xic14Y2fcZmkBiVbLApTQUxMZgmQPlfIjJhYQpni9lbChlRRZmxOJRuCt/zyKmleVL2r6uXDZaV2m8dRhBM4hXPw4BpqcA91aACDJ3iGV3hzxs6L8+58LFoLTj5zDH/gfP4A6puSNA==</latexit>

<latexit sha1_base64="kQkDxPnVv0Mgc/tsfqEuJYOLy9w=">AAAB7XicbVBNS8NAEJ3Ur1q/qh69BIvgqSRS1GPRi8cK9gPaUDabTbt2sxt2J0Ip/Q9ePCji1f/jzX/jts1BWx8MPN6bYWZemApu0PO+ncLa+sbmVnG7tLO7t39QPjxqGZVpyppUCaU7ITFMcMmayFGwTqoZSULB2uHodua3n5g2XMkHHKcsSMhA8phTglZq9USk0PTLFa/qzeGuEj8nFcjR6Je/epGiWcIkUkGM6fpeisGEaORUsGmplxmWEjoiA9a1VJKEmWAyv3bqnlklcmOlbUl05+rviQlJjBknoe1MCA7NsjcT//O6GcbXwYTLNEMm6WJRnAkXlTt73Y24ZhTF2BJCNbe3unRINKFoAyrZEPzll1dJ66LqX1Zr97VK/SaPowgncArn4MMV1OEOGtAECo/wDK/w5ijnxXl3PhatBSefOYY/cD5/AL8Nj0E=</latexit>
"0 + 2ph
<latexit sha1_base64="wnB+LmHvWxkc8YAbwGiE0tLyCtk=">AAACBXicbVDJSgNBEO2JW4xb1KMeGoMgCGEmBPUY9OIxglkgM4SeTiVp0rPQXRMMQy5e/BUvHhTx6j9482/sLAdNfFDw+r0quur5sRQabfvbyqysrq1vZDdzW9s7u3v5/YO6jhLFocYjGammzzRIEUINBUpoxgpY4Eto+IObid8YgtIiCu9xFIMXsF4ouoIzNFI7f+wOmYJYC2leNj2nLsIDqiAtxf1xO1+wi/YUdJk4c1Igc1Tb+S+3E/EkgBC5ZFq3HDtGL2UKBZcwzrmJhpjxAetBy9CQBaC9dHrFmJ4apUO7kTIVIp2qvydSFmg9CnzTGTDs60VvIv7ntRLsXnmpCOMEIeSzj7qJpBjRSSS0IxRwlCNDGFfC7Ep5nynG0QSXMyE4iycvk3qp6FwUy3flQuV6HkeWHJETckYcckkq5JZUSY1w8kieySt5s56sF+vd+pi1Zqz5zCH5A+vzBzaKmGc=</latexit>

"0 + 1ph
<latexit sha1_base64="at/tHNm0fXzDP0RSwCKl1xhXRTc=">AAACBXicbVDLSgNBEJz1GeNr1aMeBoMgCGFXgnoMevEYwTwgG8LspJMMmX0w0xsMSy5e/BUvHhTx6j9482+cJHvQxIKGmqpuprv8WAqNjvNtLS2vrK6t5zbym1vbO7v23n5NR4niUOWRjFTDZxqkCKGKAiU0YgUs8CXU/cHNxK8PQWkRhfc4iqEVsF4ouoIzNFLbPvKGTEGshTQvh55RD+EBVZC6cX/ctgtO0ZmCLhI3IwWSodK2v7xOxJMAQuSSad10nRhbKVMouIRx3ks0xIwPWA+ahoYsAN1Kp1eM6YlROrQbKVMh0qn6eyJlgdajwDedAcO+nvcm4n9eM8HuVSsVYZwghHz2UTeRFCM6iYR2hAKOcmQI40qYXSnvM8U4muDyJgR3/uRFUjsvuhfF0l2pUL7O4siRQ3JMTolLLkmZ3JIKqRJOHskzeSVv1pP1Yr1bH7PWJSubOSB/YH3+ADUDmGY=</latexit>

"0
<latexit sha1_base64="qxLaPEvriLT/pWrboVClmbXHudU=">AAAB9HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE1GPRi8cK9gPaUDbbSbt0s4m7m0IJ/R1ePCji1R/jzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU1HGqGDZYLGLVDqhGwSU2DDcC24lCGgUCW8Hobua3xqg0j+WjmSToR3QgecgZNVbyu2OqMNFcxLLn9soVt+rOQVaJl5MK5Kj3yl/dfszSCKVhgmrd8dzE+BlVhjOB01I31ZhQNqID7FgqaYTaz+ZHT8mZVfokjJUtachc/T2R0UjrSRTYzoiaoV72ZuJ/Xic14Y2fcZmkBiVbLApTQUxMZgmQPlfIjJhYQpni9lbChlRRZmxOJRuCt/zyKmleVL2r6uXDZaV2m8dRhBM4hXPw4BpqcA91aACDJ3iGV3hzxs6L8+58LFoLTj5zDH/gfP4A6ReSMw==</latexit>

"n + 1ph "n + 2ph

<latexit sha1_base64="ek1eOppEYEFX4BxM7OocCfta4yk=">AAACBXicbVDLSgNBEJyNrxhfUY96GAyCIITdENRj0IvHCOYBSQizk04yZHZ2mekNhiUXL/6KFw+KePUfvPk3Th4HTSxoqKnqZrrLj6Qw6LrfTmpldW19I72Z2dre2d3L7h9UTRhrDhUeylDXfWZACgUVFCihHmlggS+h5g9uJn5tCNqIUN3jKIJWwHpKdAVnaKV29rg5ZBoiI6R9KXpOmwgPqIOkEPXH7WzOzbtT0GXizUmOzFFuZ7+anZDHASjkkhnT8NwIWwnTKLiEcaYZG4gYH7AeNCxVLADTSqZXjOmpVTq0G2pbCulU/T2RsMCYUeDbzoBh3yx6E/E/rxFj96qVCBXFCIrPPurGkmJIJ5HQjtDAUY4sYVwLuyvlfaYZRxtcxobgLZ68TKqFvHeRL94Vc6XreRxpckROyBnxyCUpkVtSJhXCySN5Jq/kzXlyXpx352PWmnLmM4fkD5zPH5gkmKU=</latexit>

<latexit sha1_base64="lXyRoY9GChuXmmMAiuK94//gk9c=">AAACBXicbVA9SwNBEN3zM8avqKUWi0EQhHAnQS2DNpYRzAckIextJsmSvb1jdy4YjjQ2/hUbC0Vs/Q92/hs3yRWa+GDg7Xsz7MzzIykMuu63s7S8srq2ntnIbm5t7+zm9varJow1hwoPZajrPjMghYIKCpRQjzSwwJdQ8wc3E782BG1EqO5xFEErYD0luoIztFI7d9QcMg2REdK+FD2jTYQH1EHiRf1xO5d3C+4UdJF4KcmTFOV27qvZCXkcgEIumTENz42wlTCNgksYZ5uxgYjxAetBw1LFAjCtZHrFmJ5YpUO7obalkE7V3xMJC4wZBb7tDBj2zbw3Ef/zGjF2r1qJUFGMoPjso24sKYZ0EgntCA0c5cgSxrWwu1LeZ5pxtMFlbQje/MmLpHpe8C4KxbtivnSdxpEhh+SYnBKPXJISuSVlUiGcPJJn8krenCfnxXl3PmatS046c0D+wPn8AZadmKQ=</latexit>

"n + 0ph
<latexit sha1_base64="8GusF8dUG3AYmHHU7ZjW8LCHJUs=">AAACBXicbVA9SwNBEN3zM8avqKUWi0EQhHAnQS2DNpYRzAckIextJsmSvb1jdy4YjjQ2/hUbC0Vs/Q92/hs3yRWa+GDg7Xsz7MzzIykMuu63s7S8srq2ntnIbm5t7+zm9varJow1hwoPZajrPjMghYIKCpRQjzSwwJdQ8wc3E782BG1EqO5xFEErYD0luoIztFI7d9QcMg2REdK+FD2jTYQH1EHiRv1xO5d3C+4UdJF4KcmTFOV27qvZCXkcgEIumTENz42wlTCNgksYZ5uxgYjxAetBw1LFAjCtZHrFmJ5YpUO7obalkE7V3xMJC4wZBb7tDBj2zbw3Ef/zGjF2r1qJUFGMoPjso24sKYZ0EgntCA0c5cgSxrWwu1LeZ5pxtMFlbQje/MmLpHpe8C4KxbtivnSdxpEhh+SYnBKPXJISuSVlUiGcPJJn8krenCfnxXl3PmatS046c0D+wPn8AZUWmKM=</latexit>

Figure 3.1: The presence of the continuum of the photonic levels i.e. of photons in the
continuum of the EM field modes (represented schematically as colored bands in the
figure) means that the discrete excited levels of matter are embedded in this continuum.
As explained in this section the coupling of the matter to the EM field means in turn that
the discrete matter levels get ”smeared” over the nearby continuum of photonic levels.
The result is that their energy position gets shifted and they acquire a width becoming
somewhat analogous to classical resonances.

Coupling to the continuum - time domain. Exponential decay

We now turn to the discussion of what will the perturbation ĤI1 which has matrix ele-
ments connecting such levels cause. We will do this in the framework of a simple model
- a single discrete state coupled to a continuum of states. This is known as Weisskopf-
Wigner model. We present here the main results for this model. Details are found in the
Appendix of this chapter, as well as in Ch.I-C3 and Complement CI of Ref.[1].
We will use simplified notations. Consider a quantum state with energy E0 and wave
function ψ0 imbedded into a broad continuum of levels with energies Eν and wavefunc-
tions ψν . In the notation of the previous section E0 stands for E(1, {0kα }) while Eν for
E(0; 1kα , {0k′ α′ }) with the corresponding wave functions.
Let V be the interaction between the levels with matrix elements
∗ ∗
V0ν = Vν0 , Vνµ = Vµν

We want to consider how the system develops in time if it was initially (say at t = 0 )
prepared in the discrete state ψ0 . Formally we need to solve the Schrödinger equation of
this system
∂Ψ(t)
iℏ = (Ĥ0 + V̂ )Ψ(t) (3.33)
∂t
Version of April 11, 2023 125

with the initial condition

Ψ(t = 0) = ψ0 (3.34)
Let us write Ψ(t) as an expansion in the basis of the unperturbed states {ψ0 , ψν }
Z
Ψ(t) = c0 (t)ψ0 e−iE0 t/ℏ + cν (t)ψν e−iEν t/ℏ dν (3.35)

where for convenience we ”pull out” the factors e−iE0 t/ℏ and e−iEν t/ℏ from the (yet un-
determined) time dependent coefficients c0 (t) and cν (t). We note that the coefficient
c0 (t)e−iE0 t/ℏ determines the time dependence of the ”persistence amplitude” of the initial
state ψ0
⟨ψ0 |Ψ(t)⟩ = c0 (t)e−iE0 t/ℏ (3.36)
while the amplitudes cν (t)e−iEν t/ℏ provide the time dependence of the spreading of the
initial discrete state over the continuum states.
Inserting the expansion (3.35) into the Schrödinger equation, using

Ĥ0 ψ0 = E0 ψ0 , Ĥ0 ψν = Eν ψν

and projecting on ψ0 and ψµ we obtain coupled equations for the coefficients

dc0
Z
iℏ = V0µ cµ e−iωµ0 t dµ
dt
dcµ
Z
iℏ = Vµ0 c0 e−iω0µ t + Vµν cµ e−iωνµ t dν (3.37)
dt
with the notation
ωνµ = (Eν − Eµ )/ℏ
and initial conditions
c0 (0) = 1 , cν (0) = 0 (3.38)
The crucial step/approximation in the Weisskopf-Wigner approach is to neglect the cou-
pling between the continuum levels, i.e. to set

Vµν = 0 (3.39)

in the equations (3.37). This approximation allows to integrate the second equation (recall
that c0 (0) = 0)
1 t
Z
′
cµ (t) = Vµ0 e−iω0µ t c0 (t′ ) dt′ (3.40)
iℏ 0
Inserting this into the first equation we obtain a single integro-differential equation for
c0 (t)
Z t
dc0
= K(t − t′ )c0 (t′ )dt′ (3.41)
dt 0
where we introduced notation for the kernel K(t − t′ )
1
Z
K(t) = − 2 |V0µ |2 eiω0µ t dµ (3.42)
ℏ
Version of April 11, 2023 126

Equations of this type are called equations with memory (for obvious reason). The mem-
ory time is finite if the kernel K(t) has finite ”range” T , i.e. vanishes for t much larger
than some finite time interval T .
Let us make an important observation here - K(t) is proportional to the time correla-
tion of V̂ (t) in the initial state ψ0
Z
dµ⟨ψ0 |V̂ |ψµ ⟩⟨ψµ |V̂ |ψ0 ⟩ei(E0 −Eµ )t/ℏ = ⟨ψ0 |V̂ (t)V̂ (0)|ψ0 ⟩ (3.43)

where
V̂ (t) = eiĤ0 t/ℏ V̂ e−iĤ0 t/ℏ
is the interaction V̂ in the so called interaction representation. In order to understand
what this means for the spontaneous photon emission let us recall what are the unper-
turbed energies and the corresponding wave functions in that problem, cf,. Eq. (3.32)
and the following discussion. Let us also recall the explicit form of the interactions,
Z Z
3
ĤI1 = − ĵ(r) · ÂT (r) d r , ĤI3 = − m̂(r) · B̂(r) d3 r (3.44)

cf., Eqs. (3.16,3.17). Using this in the correlator ⟨ψ0 |V̂ (t)V̂ (0)|ψ0 ⟩ we observe that in
this case it is a product of the matter part involving correlators of the current ĵ(r) or
magnetization m̂(r) in the initial matter state and the correlations

⟨vacuum|ÂT,a (r, t)ÂT,b (r, 0)|vacuum⟩ and ⟨vacuum|B̂a (r, t)B̂b (r, 0)|vacuum⟩

of the components of the EM field in the vacuum. These correlators measure the vacuum
fluctuations of the field which drive the matter (say an atom) in an excited state to
spontaneously emit a photon and decay to a lower state.
Returning Eq. (3.41) we note that it can be formally solved by Laplace transform.
To invert the transform however one must use approximations. In Appendix we discuss
a different method of solving Eq. (3.41) using
R the Maslov approximation. To state the
results it is useful to rewrite the integral dµ over the continuum states ψµ in Eq. (3.42)
by splitting it into the integral over the states with a fixed energy Eµ = E following by
the integral over E. This can be done using
Z Z Z
dµ... = dE dµ δ(E − Eµ )... (3.45)

The kernel K(t) is then

1
Z
K(t) = − dE|V0µ |2 ei(E0 −E)t/ℏ with
ℏ2 Eµ =E
Z (3.46)
2
|V0µ |2 = dµ δ(E0 − Eµ )|V0µ |
Eµ =E

Using this we show in the Appendix that using the Maslov approximation approach
to solve Eq. (3.41) one finds that the time dependence of the ”persistence amplitude”
Version of April 11, 2023 127

Eq. (3.36) of the initial state ψ0 in the long time limit (cf., Eq. (3.145)) is given by the
exponential
⟨ψ0 |Ψ(t)⟩ ≡ c0 (t)e−iE0 t/ℏ = b0 e−Γt/2 e−i(E0 +∆E)t/ℏ (3.47)
where b0 is a constant which depends on the short times behavior of c0 (t) and

2π 2π
Z
Γ = |V0µ |2 = dµ δ(E0 − Eµ )|V0µ |2
ℏ Eµ =E0 ℏ
(3.48)
1
Z
∆E = P dE |V0µ |2
Eµ =E E0 − E

Here P denotes the ”principle value” of the integral, cf.,

Z b "Z Z b #
−ϵ
f (x) f (x)
P dx ≡ lim dx + dx , for a < 0 , b > 0
a x ϵ→0 a ϵ x

We see that the ”survival probability” of the initial state asymptotically decays exponen-
tially with Γ controlling the decay rate

w0 (t) ≡ |⟨ψ0 |Ψ(t)⟩|2 = |b0 |2 e−Γt (3.49)

The inverse ratio 1/Γ is often called the lifetime of the level.
We observe that Γ is a sum Z
Γ = dµΓ0→µ

of partial Γ0 → µ ’s given by

2π
Γ0 → µ = |V0µ |2 δ(E − Eµ )
ℏ
which are just the golden rule probabilities per unit time of transitions into particular
continuous state ψµ .
The irreversible dynamics of a discrete state decaying into a continuum may serve
as a simple example of how irreversibility appears in a formally reversible theoretical
framework. In this respect it is instructive to follow a chain of considerations which
starts by replacing the continuum of levels by just one level, then a few, then many but
still discrete and finally by the continuum. It should be clear that in the few levels case
there will be finite times that the system will ”visit” back the initial level. These ”return
times” are growing with the number of levels and turning to infinite (i.e. to a decay) in
the continuum case.

Coupling to the continuum - energy domain. Line shape, shift and width
The quantity ∆E in Eq. (3.47) is the energy shift of the unperturbed energy E0 caused by
the coupling via V0µ of ψ0 to the continuum of ψµ ’s. To understand this statement better
we would like to present now the ”stationary” version of the above discussion, i.e. to
determine how the discrete state ψ0 of the unperturbed Hamiltonian Ĥ0 gets ”smeared”,
Version of April 11, 2023 128

i.e. becomes distributed over the exact states of the problem with the coupling to the
continuum states.
We note that the solution (3.35) which we found to the time dependent Schrödinger
equation (3.33) can be formally expanded in terms of the eigenfunctions Ψχ of the ”full”
Hamiltonian (Ĥ0 + V̂ )Ψχ = Eχ Ψχ
Z
Ψ(t) = dχAχ Ψχ e−iΩχ t , Ωχ = Eχ /ℏ (3.50)

with the expansion coefficients Aχ determined by the initial condition (3.34)6

Aχ = ⟨Ψχ |ψ0 ⟩ (3.51)

The amplitude (3.47) can then be written as

Z Z
⟨ψ0 |Ψ(t)⟩ = dχ|⟨ψ0 |Ψχ ⟩|2 e−iΩχ t = dE|⟨ψ0 |Ψχ ⟩|2 e−iEt/ℏ (3.52)
Eχ =E

R
where for the integral dχ we used the identity Eq. (3.45) with the notation similar to
Eq. (3.46) for |⟨ψ0 |Ψχ ⟩|2 . We obtain
Eχ =E

∞
1
Z
|⟨ψ0 |Ψχ ⟩|2 = dt⟨ψ0 |Ψ(t)⟩eiEt/ℏ (3.53)
Eχ =E 2π −∞

The left hand side is what we are interested in - the distribution of probabilities of the
unperturbed discrete state ψ0 among the exact stationary states of the problem.
To evaluate the integral in the r.h.s. we need to extend the solution (3.47) to negative
times t < 0. This is simply done by noting that all the elements of the solution going from
cµ (t) to c0 (t) keep their formal expressions. The only difference is found in the discussion
of the long time limit t0 → ∞ in Eq. (3.146) of the Appendix which must be replaced by
t0 → −∞. To calculate such limit we will again use the shift in the energy integration
contour but this time we will need to do this into the positive Im E half plane. It is easy
to see that this will lead to the same result as for positive t but with the sign change of
Γ. The integral above therefore consists of two parts
Z 0 Z ∞ 
1 −i(E0 +∆E+iℏΓ/2−E)t/ℏ −i(E0 +∆E−iℏΓ2−E)t/ℏ
dte + dte
2π −∞ 0

which are easily evaluated with the result

1 ℏΓ/2
|⟨ψ0 |Ψχ ⟩|2 = (3.54)
Eχ =E π (E − E0 − ∆E)2 + (ℏΓ/2)2

This shows that the unperturbed discrete state with a fixed energy E0 gets ”smeared”
over the exact states in the energy range ℏΓ shifted by ∆E relative to E0 . The function in
the r.h.s. of the above equality is a Lorentzian (also known as Breit-Wigner distribution).
6 We assume that the continuum eigenfunctions Ψ are normalized to the delta function ⟨Ψ |Ψ ′ ⟩ =
χ χ χ
δ(χ − χ′ )
Version of April 11, 2023 129

One often says that the discrete state with a sharp position in energy ”acquires” a line
shape with a width and a shift.
One can also describe the result Eq. (3.54) as a discrete state turning into a resonance.
This due to the analogy with what happens to a classical harmonic oscillator with an
oscillation frequency ω0 under an influence of the dissipative force −γv. The oscillator
motion (for unit mass)
q
q(t) = q0 e−γt/2 sin(ωt + ϕ0 ) , ω = ω02 − γ 2 /4

is damped oscillations with a shifted frequency and the amplitude exponentially decaying
with time.

3.2.2 Photon emission rate

Following the discussion in the previous section we will now use the Fermi golden rule
2π
Γi→f = |⟨f |V |i⟩|2 δ(En − E0 − ℏωk ) (3.55)
ℏ
to calculate the rate (probability per unit time) of spontaneous photon emission in which
the state of matter changes from higher to lower energy state. Denoting these matter
states as |n⟩ and |0⟩ we have in this case

|i⟩ = |n⟩|{0kα }⟩ |f ⟩ = |0⟩|1kα , {0k′ α′ }⟩ (3.56)

where k′ α′ denote all photon states except kα. We have already inserted the correspond-
ing initial and final energies En and E0 + ℏωk in the δ function above.
Using ĤI1 of Eq. (3.21) with these initial and final states we calculate

⟨f |ĤI1 |i⟩ = ⟨0| ⟨1kα , {0k′ α′ }|ĤI1 |{0kα }⟩ |n⟩ =

X  ℏ 1/2
= − ′′ Ω ⟨0|ĵ−k′′ · λk′′ α′′ |n⟩⟨1kα , {0k′ α′ }|âk′′ α′′ + ↠′′ ′′ |{0kα }⟩
2ϵ ω
0 k −k α
k′′ α′′
We have a sum of products of the matter and the radiation matrix elements. The latter
are trivial to calculate

⟨1kα , {0k′ α′ }|âk′′ α′′ + ↠′′ ′′ |{0kα }⟩ = δ−k′′ , k δα′′ α (3.57)

−k α

which means that only one term is not zero in the sum over the modes.
So we obtain  1/2
ℏ
⟨f |ĤI1 |i⟩ = − ⟨0|ĵk · λkα |n⟩ (3.58)
2ϵ0 ωk Ω
The appearance of the inverse quantization volume 1/Ω in the square of this expression is
easy to understand. We are calculating the probability rate to find the emitted photon in
a given k momentum state. For a macroscopically large Ω this probability is very small
∼ 1/Ω but the values of k are very dense. In fact their density is ∼ Ω which will cancel
the 1/Ω in the probability rate. We will now consider an example showing this.
Version of April 11, 2023 130

What a typical detector measures

Let us consider a practical situation in which the emitted photons are detected by a de-
tector placed sufficiently far from the emitting system and measuring all photons emitted
in a small sold angle dγ around k.

Figure 3.2: Schematic geometry of the detection of emitted photons. The detector opening
spans dγ solid angle centered at the direction θ, ϕ along which the emission rate is detected.

To calculate what the detector measures we note that the probability per unit time
to measure a photon with a given polarization α and momentum ℏk in a small ”volume”
∆3 k around a given k is given by
X Ωd3 k
Γi→k′ α ≈ Γi→kα × number of k′ s in ∆3 k → Γi→kα


dwkα =
(2π)3
k′ in ∆3 k

where we denoted schematically by Γi→k′ α the rate of the photon emission into k′ α state
and assumed that ∆3 k is small enough to have this rate changing little in the above sum.
We have also conventionally switched to the differential d3 k in our notations and used the
expression Ωd3 k/(2π)3 for the number of k’s in d3 k.
To continue with what we assumed this detector measures we should adjust the above
expression to account for all the k’s in the solid angle dγ. For this we express d3 k =
k 2 dkdγ, keep dγ fixed and integrate over dk. Using the explicit expression for Γi→k′ α
with matrix element (3.58) and changing to k = ω/c we have that the probability or
more practically the relative number of photons per unit time measured by the detector
in repeated experiments is given by

Ωω 2 dω
 
2π
Z
ℏ
dNkα = dγ |⟨0|ĵk · λkα |n⟩|2 δ(En − E0 − ℏω) (3.59)
ℏ 2ϵ0 ωΩ (2πc)3
Version of April 11, 2023 131

We note that Ω cancels out. Using the δ function to do the integral we find
dNkα ω
= |⟨0|ĵk · λkα |n⟩|2 (3.60)
dγ 8π 2 c3 ϵ0 ℏ
where we must remember that ω and the magnitude of k are fixed by the energy conser-
vation
ℏω = ck = En − E0 (3.61)
The above expression for the emission rate is the main result of this section. It shows
all one needs in order to find the deexcitation rate with photon emitted in the small angle
in the direction k with the polarization vector λkα . One should be able to calculate the
matrix element
⟨0|ĵk |n⟩
of the k-th Fourier component of the matter current, then project it on the polarization
λkα , square the result and multiply by the coefficient in front of (3.60).
Let us indicate that working in spherical coordinates

k = k(sin θ cos ϕ, sin θ sin ϕ, cos θ) (3.62)

a convenient choice of linear polarization vectors for the photon emission problem is

λ1 = (cos θ cos ϕ, cos θ sin ϕ, − sin θ) , λ2 = (− sin ϕ, cos ϕ, 0)

(3.63)
k · λ1,2 = λ1 · λ2 = 0

This choice corresponds to λ1 lying in the k, ez plane, i.e. parallel to eθ while λ2 is

perpendicular to it, i.e. parallel to eϕ .
If the detector does not distinguish between the photon polarizations (as is often the
case) one must sum
dNk X dN
kα
= (3.64)
dγ α=1,2
dγ

The classical limit

In the following sections we will discuss various properties and simplifications of the
current matrix element in (3.60). Before that let us compare this expression with the
corresponding classical result. For this let us it by the photon energy ℏω. In this way we
will find the power emitted by the system
dPkα ω2
= |⟨0|ĵk |n⟩| · λkα |2 (3.65)
dγ 8π 2 c3 ϵ0
Remarkably there is no explicit ℏ dependence in this expression and the quantum me-
chanics manifests itself in the presence of the matrix element of the current.
Comparing this expression with the classical result (cf., Ref.[3], p.279) one finds that
the expressions are formally identical7 provided one identifies the matrix element of the
7 One should remember the extra 1/4πϵ factor when passing from CGS to SI of the square of electric
0
charge
Version of April 11, 2023 132

current operator ĵ(r) in quantum mechanical expression with the Fourier component with
the frequency (3.61) of the classical current j(r, t).
This correspondence fits the semiclassical rule (cf., Sec.48 in Ref. [4]) that the matrix
elements fmn in the classical limit approach the components fm−n of the Fourier expansion
of the classical function f (t). This rule was originally guessed by Heisenberg in his matrix
quantum mechanics approach.

Momentum conservation and recoil energy

Let us consider the common case that the initial and final states of the photon emitting
matter system are momentum eigenstates with total momentum Pi and Pf respectively.
Isolated atoms, molecules, nuclei will be in such states. The initial and the final states in
such systems will then be

|i⟩ = |n, Pi ⟩|{0kα }⟩ |f ⟩ = |0, Pf ⟩|1kα , {0k′ α′ }⟩

with the transition matrix element (3.58)

 
2 ℏ
|⟨f |V |i⟩| = |⟨0, Pf |ĵk |n, Pi ⟩ · λkα |2 (3.66)
2ϵ0 ωk Ω

The operator ĵk has the property that when acting on a matter state having a given
total momentum P it transforms this Pstate into a state with P − ℏk. To show this let
N
us use the momentum operator P̂ = a=1 p̂a and calculate the action of its components
P̂m on the state which ĵk generates acting on |n, Pi ⟩
  h i
(i)
P̂m ĵk |n, Pi ⟩ = P̂m , ĵk |n, Pi ⟩ + Pm ĵk |n, Pi ⟩ (3.67)

(i)
where to avoid confusion we denoted by Pm the m-th component of the Pi vector and
(i)
used P̂m |n, Pi ⟩ = Pm |n, Pi ⟩ in the second term on the r.h.s. Let us now calculate the
commutator in the first term using the explicit expression (3.22) for ĵk and the following
relation for the components p̂a of P̂
h i h i
p̂a , e−ik·rb = −δab iℏ∇a e−ik·rb = −δab ℏke−ik·rb → P̂m , ĵk = −ℏkm ĵk

This gives for Eq. (3.67)

   
P̂m ĵk |n, Pi ⟩ = (Pi − ℏk)m ĵk |n, Pi ⟩ (3.68)

showing that indeed the state ĵk |n, Pi ⟩ had a definite value of the momentum P = Pi −ℏk.
Since states with different momenta are orthogonal this property means the transitions
matrix elements ⟨0, Pf |ĵk |n, Pi ⟩ is non vanishing only for

Pi = Pf + ℏk

i.e. the emitted photons conserve the total momentum.

Version of April 11, 2023 133

The above discussion concerned the change of the momentum of matter systems emit-
ting photons. But this recoil momentum ∆P ≡ Pf − P1 implies that there is also a
corresponding recoil energy. This energy should in principle be included in the energy
conservation relation Eq.(3.61). However one can show that for typical photon momenta
the recoil energy can to a very good approximation be neglected and the matter system
assumed to remain at rest in its c.m. frame.
Indeed for the photon energy ϵphoton = ℏω the momentum transferred to the recoiling
matter is precoil = ℏk = ℏω/c. Thus the matter recoil kinetic energy ϵrecoil = p2recoil /2M =
(ℏω)2 /2M c2 where we assumed that the matter is non relativistic and work in a reference
frame in which it was initially at rest.
The ratio of the recoil energy to the photon energy is therefore
ϵrecoil ℏω
∼
ϵphoton M c2

which for typical emitting matter systems (molecules, atoms, nuclei) is

1eV ÷ 10 M eV
≪1
(1 ÷ 100) GeV

so that the recoil energy is indeed negligible for such system.

Let us note that the dimensionless recoil velocity is given by the same expression
v p ℏω
= =
c Mc M c2
For e.g. hydrogen atom this gives
vrecoil 10eV
∼ 9 → vrecoil ∼ 10−8 c = 3m/s
c 10 eV

3.2.3 Long wavelength approximation

Consider two typical photon emitting quantum systems - atoms and nuclei and examine
the relation between their sizes and the wavelengths of emitted photons. The latter are
related to the photon energies as
2π 2πc 2πℏc 6.28 × 197 eV · nm 1200 eV · nm
λ= = = ≈ ≈
k ck ℏω ℏω ℏω
The typical atomic sizes are ∼ 0.1 ÷ 0.2 nm while typical emission energies of atomic
photons are ∼ 1 ÷ 103 eV . This means that the emitted photon wavelengths are

λ ∼ (1.2 ÷ 1200) nm ≫ 0.1 ÷ 0.2 nm atomic sizes

Similar result holds for nuclei for which the sizes are 5 ÷ 10f m while typical emission
energies are (1 ÷ 10) M eV . So

1200 M eV · f m
λ≈ ∼ 120 ÷ 1200 f m ≫ 5 ÷ 10 f m nuclear sizes
ℏω
Version of April 11, 2023 134

Similar estimates hold for solid state emission systems (there the typical size is the crystal
unit cell, etc) and small molecules.
These estimates have important consequence for the evaluation and magnitude of the
current matrix element in the emission rate expression Eq. (3.60). Writing it out explicitly
Z
⟨0|ĵk |n⟩ = d3 r e−ik·r ⟨0|ĵ(r)|n⟩ (3.69)

we see that the range of the integration where the integrand is not vanishing is determined
by the matrix element of the current. So this range must be |r| ≤ a where a ∼ size of
the emitting system. As we have seen above for the majority of the matter systems of
interest this range will be a ≪ λ - the wave lengths of the emitted photons. This gives
the condition ka ≪ 1 under which one can expand the exponent in the above integral
Z Z
⟨0|ĵk |n⟩ = d r(1 − ik · r + ...)⟨0|ĵ(r)|n⟩ = ⟨0|ĵ0 |n⟩ − i d3 r (k · r)⟨0|ĵ(r)|n⟩ + ... (3.70)
3

and keep only the lowest non-vanishing term.

This is the basis of the important element of the photon emission (and as we will see
below photon absorption) treatment – the Long Wavelength Approximation (LWA).

3.2.4 Electric dipole emission

Let us discuss the photon emission rate which one should expect retaining only the lowest
term in the LWA expansion (3.70). We use
N N
qa qa
Z X X
3
ĵ0 = d r [p̂a δ(r − ra ) + δ(r − ra )p̂a ] = p̂a (3.71)
a=1
2m a a=1
ma

To evaluate matrix elements of this operator between matter eigenenergy states it is

convenient to use the commutation relation
N
X p̂2b p̂
[ra , Ĥmatter ] = [ra , ] = iℏ a
2mb mb
b=1

Therefore
N N
X qa 1 X
ĵ0 = p̂ = [d̂, Ĥmatter ] with d̂ = qa ra (3.72)
a=1
ma a iℏ a=1

where d̂ is the operator of the dipole moment of the matter system.

For the matrix element in the first term of (3.70) we therefore have
1 En − E0
⟨0|ĵ0 |n⟩ = ⟨0|[d̂, Ĥmatter ]|n⟩ = ⟨0|d̂|n⟩ (3.73)
iℏ iℏ
where we used that |0⟩ and |n⟩ are eigenstates of Ĥmatter . Thus to lowest order in ka

⟨0|ĵk |n⟩ ≈ −iω⟨0|d̂|n⟩

Version of April 11, 2023 135

Transitions described by these matrix elements are called electric dipole transitions. Using
this in the expression (3.60) for the photon emission rate we obtain
dNkα ω3
= |⟨0|d̂|n⟩ · λkα |2 (3.74)
dγ 8π c3 ϵ0 ℏ
2

Radiation described by this formula is called electric dipole radiation.

The above expression depends on three factors - the vector of the matrix elements of
the dipole operator between the matter eigenstates,
don ≡ ⟨0|d̂|n⟩ ,
the energy difference ℏω between these states and the polarization λkα of the emitted
photon.
As we will see below the fact that the dipole operator d̂ is a vector allows to make
many general statements concerning the resulting vector don of matrix elements. For the
situation in which the initial and final states |n⟩ and |0⟩ are eigenstates of the angular
momentum of the matter system it will be possible to determine when don is non vanishing
and to derive general relations between the components of don , i.e. to find its direction.
˙
The direction k of the photon emission enters the dipole emission rate Eq.(3.74) via its
dependence on the polarization vectors λkα which are perpendicular to k. It is obvious
that the angular distribution of the emitted photons is symmetric around the direction of
the vector don . Moreover since λkα are perpendicular to k, the emission rate is zero along
the line of the direction of don . For an arbitrary direction of k it is convenient to work
with polarization vectors λk1 and λk2 which are respectively parallel and perpendicular
to the plane defined by don and k. Choosing the coordinate system with don along its
z-axis and denoting by θ and ϕ the spherical angles of k it is easy to see that such a choice
corresponds to Eq. (3.63). Then
|don · λk1 |2 = |don |2 sin2 θ (3.75)
We plot the resulting pattern in Fig. 3.3. Clearly the emission rate with the polarization
λk2 is identically zero for all the directions of such emission
|don · λk2 |2 = 0

Figure 3.3: Angular distribution of the electric dipole radiation vs the spherical angles
θ and ϕ of the emitted photon wave vector k with the dipole matrix element vector don
chosen to lie along the z-axis.
Version of April 11, 2023 136

Multiplying Eq. (3.74) by ℏω gives the power emitted in the dipole radiation

dPkα ω4
= |⟨0|d̂|n⟩ · λkα |2 (3.76)
dγ 8π 2 c3 ϵ0

As in our discussion following Eq. (3.65) we note that there is no explicit ℏ dependence
in this expression and that the quantum mechanics manifests itself ”only” in the matrix
element of the dipole operator. Once again comparing with the classical expression, cf.8
one sees that this matrix element in the classical limit becomes the Fourier component of
the classical dipole moment d(t) with frequency (3.61).
Here is a pictorial representation of the classical electric dipole radiation.

[Link]

3.2.5 Angular momentum and parity selection rules

In discussing photon emission by individual molecules, atoms, nuclei and subnuclear par-
ticles one deals with rotationally invariant matter Hamiltonians with eigenstates which
are also eigenstates of the total angular momentum (including the spin)
N
X N
X
Ĵ = L̂ + Ŝ = ra × p̂a + ŝa
a=1 a=1

2
or more precisely of its square Ĵ and one of its projections, commonly chosen as Jˆz . The
sum here is over the components of the molecule, atom, etc, which is under consideration.
So in these (very common) cases the dipole matrix elements to be considered are

⟨0|d̂|n⟩ → ⟨ν2 J2 M2 |d̂|ν1 J1 M1 ⟩ (3.77)

where we indicated explicitly the angular momentum quantum numbers and denoted by
ν all the remaining ones needed to completely specify the states of the matter system.
For example levels of a particle with spin 1/2 moving in a spherically symmetric potential
and with spin-orbit coupling have 4 quantum numbers

nr , l , j , m

so ν will stand in this case for nr , l numbers.

Dipole moment is an ℓ = 1 object. Spherical components of vectors

When the dipole operator d̂ in the matrix element Eq. (3.77) acts on the initial state
|ν1 J1 M1 ⟩ it creates a state which doesn’t have the same angular momentum and is in
general expected to be a superposition of states with definite J’s and M ’s. The vectorial
character of d̂ allows to determine the range of possibles values of these quantum num-
bers and to a certain extent also the coefficients in the resulting linear combination. To
8 [Link]
Version of April 11, 2023 137

demonstrates this it is useful to transform the vector d̂ from cartesian to the so called
spherical components.
The general expressions for such components of any vector v are by definition
vx + ivy vx − ivy ∗
vµ=1 = − √ , vµ=0 = vz , vµ=−1 = √ = −vµ=1 (3.78)
2 2
The scalar product of vectors in spherical components is expressed as
X X
a·b= (−1)µ aµ b−µ = aµ b∗µ (3.79)
µ=−1,0,1 µ=−1,0,1

The usefulness of forming the spherical components’ combinations can be seen especially
clear in terms of the spherical coordinates9
vx = v sin θv cos ϕv , vy = v sin θv sin ϕv , vz = v cos θv (3.80)
so that
r
1 4π
v0 = v cos θv , v±1 = ∓ √ v sin θv e±iϕv → v Y1µ (Ωv ) µ = 1, 0, −1
vµ =
2 3
(3.81)
emphasizing that three components of a vector behave under rotations as Y1µ . In the
group theoretical terminology one says that vectors transform as j = 1 representation of
the group O(3) of rotations.
This of course holds true also for the vector d̂ of the electric dipole moment. Let us
now explore the consequences of this insight.

Dipole angular momentum selection rules - hydrogen atom first

Let us start with a simplest case of electric dipole transitions in a hydrogen atom. With
its single electron the dipole operator and its spherical components in this simple system
are just r
4π
d̂ = er → dˆµ = e r Y1µ (θ, ϕ)
3
We can ignore the spin and consider dipole transitions between the orbital eigenstates of
the hydrogen atom
|n, l, m⟩ → |n′ , l′ , m′ ⟩
with the coordinate representation of these states having the familiar form
⟨r|n, l, m⟩ = Rnl (r)Ylm (θ, ϕ)
In this representation electric dipole operator acting on the initial state |n, l, m⟩ results
in a state dˆµ |n, l, m⟩ which in the coordinate representation is
r
ˆ 4π
⟨r|dµ |n, l, m⟩ = erRnl (r)Y1µ (θ, ϕ)Ylm (θ, ϕ) (3.82)
3
9 Note that the subscript v in the angles here indicates that they are not necessarily the same as of the

real space coordinate vector {x, y, z}.

Version of April 11, 2023 138

Let us use the intuition from the quantum angular momentum algebra and view the
product of the two spherical harmonics Y1µ Ylm as an eigenfunction of the (quantum) sum
of two angular momenta ℓ1 = 1 and ℓ2 = l. As we know the resulting angular momentum
ℓ has possible values given by

l − 1 , l , l + 1 with the projection µ + m

Continuing with this understanding we expect that the state dˆµ |n, l, m⟩ is a linear com-
bination of states with the above values of ℓ and its projection.
Forming the dipole matrix element ⟨n′ , l′ , m′ |dˆµ |n, l, m⟩ means that the final state
|n , l , m′ ⟩ is projected on this linear combination. The resulting overlap should be zero
′ ′

unless the final angular momentum l′ m′ is equal to one of the above values, i.e. satisfy
the familiar triangular rule of adding angular momenta

|l − 1| ≤ l′ ≤ l + 1 , m′ = m + µ (3.83)

Formally these considerations are supported and extended by using the known expansion
of the product of two spherical harmonics Y11 m1 (θ, ϕ)Yl2 m2 (θ, ϕ) viewed as a function of
the angles θ, ϕ in terms of the complete set {YLM (θ, ϕ)}
∞
X L
X
Yl1 m1 (θ, ϕ)Yl2 m2 (θ, ϕ) = GM m 1 m2
L l1 l2 YLM (θ, ϕ) (3.84)
L=0 M =−L

where GM m 1 m2
L l 1 l2 are the so called Gaunt coefficients which are proportional to the respec-
tive Clebsh-Gordan (CG) coefficients, cf. Ref.[5], p.57
M m 1 m2
GL l1 l2 = a(l1 , l2 , L)⟨LM |l1 m1 , l2 m2 ⟩ (3.85)

Here the proportionality factor a(l1 , l2 , L) doesn’t depend on the projections m1 , m2 , M .

The CG coefficient is zero unless

|l1 − l2 | ≤ L ≤ l1 + l2 and M = m1 + m2

which constraints the sum over L in the expansion (3.84) and removes the sum over M.
When applied to our case, Eq. (3.82), with the product Y1µ Ylm one recovers what we
have guessed using qualitative arguments, i.e. the rules (3.83). These are called electric
dipole angular momentum selection rules. In words they state that only transition with
at most one unit change in the angular momentum are allowed, i.e ∆l = 0, ±1. Below
we will complete the discussion of these rules by examining also the consequences of the
parity conservation.
Let us further observe that the dependence of the Gaunt coefficients on the angular
momentum projection quantum numbers M, m1 , m2 enter only via the CG coefficient.
To see what this means for the electric dipole transitions let us sketch schematically the
calculation of the dipole matrix element ⟨n′ , l′ , m′ |dˆµ |n, l, m⟩. We will need to calculate
Z Z
⟨n′ , l′ , m′ |dˆµ |n, l, m⟩ = (radial part) (angular part) (3.86)
Version of April 11, 2023 139

where
4π ∞ 2
Z r Z
(radial part) = e r drRn∗ ′ l′ (r)rRnl (r)
3 0
Z Z (3.87)
(angular part) = Yl∗′ m′ (θ, ϕ)Y1µ (θ, ϕ)Ylm (θ, ϕ) sin θdθdϕ

Expansion (3.84) shows that the angular integral equals the appropriate Gaunt coefficient
′
Glm′ 1µm
l . Using Eq. (3.85) we see that

⟨n′ , l′ , m′ |dˆµ |n, l, m⟩ = ⟨lm, 1µ|l′ m′ ⟩⟨n′ l′ ||d||nl⟩

ˆ (3.88)

where we introduced the common notation ⟨n′ l′ ||d||nl⟩ ˆ called reduced matrix element for
the part of the full matrix element which is independent of m, m′ and µ. In the present
case it is the product of the radial part in (3.87) and the factor a(l, 1, l′ ) in the relation
(3.85).
Expression (3.88) is a particular case of a more general relation known as the Wigner-
Eckart theorem which will be discussed in the next Section. It shows that the dipole
matrix element dependence on m, , m′ and µ is entirely determined by known (tabulated)
CG coefficients, cf. Ref.[5].
From this it follows that if for given nl and n′ l′ quantum numbers one needs to find
all the matrix elements ⟨n′ , l′ , m′ |dˆµ |n, l, m⟩ it is be enough to determine just one of them,
say, with m = m′ , µ = 0. Using its value one can calculate the reduced matrix element
ˆ
⟨n′ l′ ||d||nl⟩ and then all the (2l + 1) × 3 via the relation Eq. (3.88) with appropriate CG
coefficients.
We also note that for given initial and final states the selection rules Eq. (3.83) show
that only one spherical component of the vector10 ⟨n′ , l′ , m′ |dˆµ |n, l, m⟩ is non zero, that
with µ = m′ − m. Let us recall that in the present context this vector is what was denoted
⟨0|d̂|n⟩ in the expression (3.74) for the electric dipole emission rate. We then conclude
that the scalar product ⟨0|d̂|n⟩ · λ in that expression has correspondingly only one term
⟨0|dˆµ |n⟩λ∗µ with that µ and the angular distribution is given by the angular dependence
of |λµ |2 .
Let us consider as an example the case of transitions between states with equal m = m′
for which the only non zero matrix element is ⟨n′ , l′ , m|dˆ0 |n, l, m⟩. This is ⟨0|dz |n⟩ in the
notation of Eq. (3.74) and correspondingly the angular distribution of the emitted photons
is given by λ2z which for the choice (3.63) of λ1 is given by Eq. (3.75) and zero for λ2 .
More examples and details will be considered in tutorials and homework.

Dipole parity selection rule - hydrogen atom first

Let us now examine limitations which parity conservation imposes on the possible final
states of electric dipole transitions from a given initial state. We start by noticing that
under the parity transformation r → −r, i.e. under mirror reflection
x, y, z → −x, −y, −z (3.89)
10 We use the term ”vector” for complex valued matrix elements of the dipole operator dˆ for the brevity

of presentation. It is the relative size of its three components that will be of our interest
Version of April 11, 2023 140

of the coordinate system the electric dipole operator changes sign d̂ → −d̂. Let us make
this coordinate change in the integral Eq. (3.86). In spherical coordinates this change is

r, θ, ϕ → r, π − θ, ϕ + π

so that the radial part doesn’t change while the spherical harmonics transform as11

Ylm (θ, ϕ) → Ylm (π − θ, ϕ + π) = (−1)l Ylm (θ, ϕ) (3.90)

The result is that the entire integral on the r.h.s. of Eq. (3.86) is equal to itself multiplied
′
by −(−1)l (−1)l . This of course means that it is zero and together with it the matrix
element ⟨n′ , l′ , m′ |dˆµ |n, l, m⟩ is zero unless
′
(−1)l (−1)l = −1 (3.91)

i.e. l′ and l are of opposite parity (i.e. odd vs even or even vs odd). This is called parity
selection rule. Taken together with the angular momentum we find that electric dipole
selection rules can be formulated as

l′ = l ± 1 (3.92)

In Fig.3.4 we show examples of electric dipole transitions

Figure 3.4: Radiative transitions in hydrogen. Only dipole transitions between adjacent
angular momentum columns are allowed, as per combined angular momentum and parity
selection rule ∆l = ±1
11 To see this start with the easy Y ∼ sinl θeilϕ and then use Y
ll lm ∼ L̂− Ylm+1 together with L̂− being
even under r → −r to show that all Ylm transform as Yll .
Version of April 11, 2023 141

Angular momentum selection rules - general view. The Wigner-Eckart theo-

rem
In this section we will extend and formalize our discussion of the angular momentum
selection rules from the simplest case of radiative transitions in a hydrogen atom to a
general case of any physical system (e.g. multi-electron atoms, nuclei, molecules) the
Hamiltonian of which is invariant under rotations. We will show that the main relation,
Eq.(3.88) holds for such systems with all its consequences.
The general structure of the eigenstates in systems with rotationally invariant Hamil-
tonian is |νJM ⟩, cf., Eq. (3.77), with ν denoting all the quantum numbers needed to
specify this state apart of the angular momentum J and its projection M . What this
structure means is that under O(3) rotations these states transform as
J
ˆ
iαn·J
X
J ′
Û (αn)|νJM ⟩ ≡ e |νJM ⟩ = DM M ′ (αn)|νJM ⟩
M ′ =−J (3.93)
ˆ
J ′ iαn·J
DM M ′ (αn) = ⟨νJM |e |νJM ⟩

i.e. the multiplets of states with different J’s do not mix. Here we denoted by α the angle
of rotation and by the unit vector n the direction of the rotation axis.
In our discussions of the hydrogen atom case we have seen that the vector character
of the dipole operator, i.e. its behavior under rotations played a very important part. We
will now generalize this discussion. Let us recall that under any unitary transformation
which transforms wavefunctions as |ψ⟩ → U |ψ⟩ the operators transform as U fˆU −1 . This
is trivially seen by considering how the states obtained by acting with fˆ transform

fˆ|ψ⟩ → U fˆ|ψ⟩ = U fˆU −1 U |ψ⟩

which demonstrates that indeed U fˆU −1 acting on transformed wavefunctions U |ψ⟩ pro-
duces the correctly transformed result.
Following this understanding one defines spherical tensor operators T̂jµ as a set of
2j + 1 operators which transform among themselves under O(3) rotations
X j
Û −1 (αn)T̂jµ Û (αn) = Dµµ′ (αn)T̂jµ′
µ′

Obviously the electric dipole operator dˆµ is an example of the spherical tensor T̂jµ with
rank j = 1. In the following section we will encounter examples of electric and magnetic
multipole operators which will correspond to spherical tensors T̂jµ with higher rank j.
One also encounters similar expansions of physical operators in terms of spherical tensor
operators T̂jµ in other fields of physics, e.g. in the context of atomic and nuclear shell
models.
To understand the properties of the spherical tensor operators let us examine how the
state which is obtained when T̂jµ acts on |νJM ⟩ behaves under rotations
XX j
U T̂jµ |νJM ⟩ = U T̂jµ U −1 U |νJM ⟩ = J
Dµµ′ DM ′
M ′ T̂jµ′ |νJM ⟩
µ′ M′
Version of April 11, 2023 142

The product of the two D matrices appearing here is identical to what would be obtained
when rotating the direct product of states with angular momentum j, µ and J, M . This
suggests that T̂jµ acting on |νJM ⟩ generates a state having total angular momentum
equal (quantum mechanically) to the sum of j, µ and J, M . This would mean that in the
matrix
⟨ν ′ J ′ M ′ |T̂jµ |νJM ⟩ (3.94)
only matrix elements satisfying the quantum mechanical rules of summing the angular
momenta
|J − j| ≤ J ′ ≤ J1 + j , M ′ = M + m (3.95)
can be non zero.
These intuitive expectations find rigorous proof in the classic Wigner-Eckart theorem.
It generalizes the equality Eq. (3.88) to matrix elements (3.94), i.e. to the most general
spherical tensor operators and eigenstates of any physical system with spherical symmetry

⟨ν ′ J ′ M ′ |T̂jµ |νJM ⟩ = ⟨J ′ M ′ |jµ, JM ⟩⟨ν ′ J ′ ||T̂j ||νJ⟩ (3.96)

Here ⟨J ′ M ′ |jm, JM ⟩ are the Clebsh-Gordan coefficients and the notation ⟨ν ′ J ′ ||T̂j ||νJ⟩
called reduced matrix elements stands for the parts of the full matrix elements which are
independent of the projections M, M ′ and m. This dependence is fully incorporated in the
CG coefficients which also carry the information about the angular momentum selection
rules, Eq. (3.95).
As in the hydrogen atom case the reduced matrix elements represent the orientation
independent context of the original matrix elements, Eq.(3.94). To find them it is enough
to calculate ⟨ν ′ J ′ M ′ |T̂jm |νJM ⟩ for one particular set of values of M, m, M ′ = M + m and
divide the result by the corresponding CG coefficient. For fixed νJ and ν ′ J ′ this amounts
to just one calculation to determine all the (2J + 1) × (2j + 1) matrix elements in the left
hand side of the relation (3.96) via the (known, tabulated) CG coefficients.
Finally let us note that the formal proof of Eq. (3.96) can be found in many references,
e.g. p. 252 in Ref. [7].

An aside - review of the parity symmetry

In our discussion of the parity selection rules in hydrogen atom they looked like a special
case depending on the behavior of the spherical harmonics Ylm (θ, ϕ) under the trans-
formation of the angles, Eq. (3.90). We now wish to generalized these considerations to
photon radiation in more complicated systems.
Parity transformation is an inversion transformation of a coordinate system in which
all of its axes change signs, e.g. Eq. (3.89). Let us note that in two dimensions this
transformation can be accomplished by a π rotation of the axis. This is not so in three
dimensions where the coordinate system changes from right-handed to left-handed. This
is the reason the parity transformation probes additional features in three dimensional
physical systems.
Let us note that technically the coordinate inversion can be achieved by a reflection in
any plane, followed by a π rotation about an axis normal to this plane. We also note that
under coordinate inversion vectors are expected to change signs, cf., Fig.3.5. However as
Version of April 11, 2023 143

we will see below there exist a category of vectors which do not change signs under parity
transformation. Such vectors are called axial vectors or pseudo-vectors to distinguish from
the real vectors also called polar vectors.

Figure 3.5: Parity transformation - the same physics (e.g. the same particle position,
momentum, etc) is seen in the inverted coordinate axes system with r → −r, p → −p etc

Is the nature invariant with respect to the parity transformation? Historically this was
a very important question and the brief answer is that physical systems interacting via
gravity, electromagnetic and strong interactions are invariant but the weak interactions
violate this. It is beyond the scope of these lectures to go into the details of this statement,
cf., Ref. [8]. Rather let us remain in the framework of what we study and examine this
˙
issue starting with the Hamiltonian given by Eq.(3.1). We observe that this Hamiltonian
remains invariant if we change
ra → −ra , , pa → −pa A(r) → −A(−r) , E(r) → −E(−r) (3.97)
which is obviously the parity transformation. The extension to the remaining part ĤI3 ,
˙
Eq.(3.10) of the (non relativistic) matter-EM field Hamiltonian is discussed in the Ap-
pendix 3.3.2 where it is shown that magnetic field B(r) and particles’ angular momenta
la and spins sa are axial vectors, i.e. they do not change under the coordinate inversion.
Let us now consider what does the invariance of the Hamiltonian under the parity
transformation imply. It will be sufficient for our goals to limit the discussion to the
matter part of the Hamiltonian Ĥmatter in Eq. (3.5). We introduce the parity operator
by defining its action on the wavefunctions of the matter particles
P̂ ψ(r1 , σ1 ; r2 , σ2 ; ...; rN , σN ) = ψ(−r1 , σ1 ; −r2 , σ2 ; ...; −rN , σN ) (3.98)
or formally
⟨r1 , σ1 ; ...; rN , σN |P̂ |ψ⟩ = ⟨−r1 , σ1 ; ...; −rN , σN |ψ⟩
Here σ’s denote the particle spin variables (e.g. for spin 1/2 they are σ = ±1/2) and we
used the axial vector nature of the spins.
Clearly
P̂ 2 ≡ P̂ P̂ = 1
which means that
P̂ = P̂ −1 (3.99)
Version of April 11, 2023 144

As usual with symmetries the invariance of the matter Hamiltonian under the parity
transformation means that to transform the result of Ĥ acting on any |ψ⟩ will produce
the same result as of Ĥ acting on the transformed |ψ⟩
 
P̂ Ĥmatter |ψ⟩ = Ĥmatter P̂ |ψ⟩ (3.100)

Formally this means

P̂ Ĥmatter = Ĥmatter P̂ → [Ĥmatter , P̂ ] = 0 (3.101)

or using (3.99)
P̂ Ĥmatter P̂ = Ĥmatter
Eigenstates of such Hamiltonian are or can be chosen to be eigenstates of P̂ . Indeed,
acting with P̂

Ĥmatter |n⟩ = En |n⟩ → P̂ Ĥmatter |n⟩ = En P̂ |n⟩ → Ĥmatter P̂ |n⟩ = En P̂ |n⟩

we see that if |n⟩ is an eigenstate of Ĥmatter so is P̂ |n⟩ with the same eigenenergy En .
This implies one of the two possibilities - either P̂ |n⟩ is proportional to |n⟩

P̂ |n⟩ = const|n⟩

or it is a different state. In the former case we find

P̂ 2 |n⟩ = constP̂ |n⟩ = const2 |n⟩

and since P̂ 2 = 1 have const2 = 1 → const = ±1.

When P̂ |n⟩ is a different state from |n⟩ we have degeneracy and can form linear
combinations of these states
1
|n⟩± ≡ (1 ± P̂ )|n⟩
2
which are eigenstates of P̂
1 1 1
P̂ |n⟩± = P̂ (1 ± P̂ )|n⟩ = (P̂ ± P̂ 2 )|n⟩ = ± (1 ± P̂ )|n⟩ = ±|n⟩± (3.102)
2 2 2
exactly as in the non degenerate case.

Parity selection rule - general view

We now want to learn what limitations the parity symmetry imposes on the dipole matrix
elements in the expression (3.74). Following the discussion in the previous section we can
assume that the states |0|⟩ and |n⟩ have well defined parity which we denote respectively
by Pf and Pi . Then using

P̂ 2 = 1 and P̂ d̂|n⟩ = −d̂P̂ |n⟩

we can write
  
⟨0|d̂|n⟩ = ⟨0|P̂ 2 d̂|n⟩ = ⟨0|P̂ P̂ d̂|n⟩ = −(−1)Pf (−1)Pi ⟨0|d̂|n⟩ (3.103)
Version of April 11, 2023 145

This means that must have

(−1)Pf (−1)Pi = −1 (3.104)
in order to have non zero dipole matrix element ⟨0|d̂|n⟩.
The above relation for the parities of the initial and final states of the transition is
called dipole parity selection rule. Together with the dipole angular momentum selection
rule they impose fairly stringent limitations on the allowed pairs of matter states which
can be ”connected” by non zero radiative dipole transitions.
To conclude this section we note that formal manipulations in (3.103) actually take a
very simple form if we write explicitly the dipole matrix elements as integrals
"N #
X Z X
⟨0|d̂|n⟩ = ψ0∗ (r1 σ1 , ..., rn σn ) qa ra ψn (r1 σ1 , ..., rn σn ) d3 r1 ....d3 rn
σ1 ,...,σN a=1

where σ’s denote the spin variables. Changing the integration variables ra = −r′a and
using
ψn (−r′1 σ1 , ..., −r′n σn ) = (−1)Pn ψn (r′1 σ1 , ..., r′n σn )
reproduces the formal arguments of Eq. (3.103).
In the following sections the above discussion will help to derive the parity selection
rules for higher terms in the long wavelength expansion Eq. (3.70).

3.2.6 ”Forbidden” (higher multipole) transitions

When the dipole matrix element between a pair of states |ni ⟩ and |nf ⟩ vanishes because
of the selection rules the radiative transitions between such states are traditionally called
forbidden. But of course there is a possibility that the transitions still occur via higher
order terms in the long-wavelength expansion (3.70) of ⟨0|ĵk · λkα |n⟩.
In this section we examine the next order term after the dipole in this expansion. This
term is Z
−i⟨0| d3 r (k · r)(ĵ(r) · λkα )|n⟩ (3.105)

It is useful to transform the integrand (omitting the subscript of λ and using the summa-
tion convention)
1
(k · r)(ĵ · λ) = kl rl ĵs λs = kl λs [(rl ĵs + rs ĵl ) + (rl ĵs − rs ĵl )] (3.106)
2
We shall consider the two parts of this expression separately.

Electric quadrupole transitions

We start by considering the symmetric term in (3.106). This term contributes

i
Z
− kl λs d3 r⟨0|rl ĵs (r) + rs ĵl (r)|n⟩ (3.107)
2
Version of April 11, 2023 146

in the transition matrix element (3.105). We will transform this expression using the
continuity equation for the operators ρ̂ and ĵ

∂ ρ̂(r, t) ∂ ĵm
= −∇ · ĵ(r, t) ≡ − (with summation over repeated indices)
∂t ∂rm
Let us consider

∂ ρ̂ ∂ ĵm
Z Z Z Z
d3 r rs rl = − d3 r rs rl = − d3 r[δms rl + rs δml ]jm = d3 r[rl js + rs jl ]
∂t ∂rm
where we used the continuity equation followed by integration by parts. This resulting
relation allows to express the matrix element of the symmetric term (3.107) as

i ∂ ρ̂
Z
− kl λs d3 r rs rl ⟨0| |n⟩
2 ∂t
Using the Heisenberg equation for ρ̂ we can write
∂ ρ̂ 1 En − E0
⟨0| |n⟩ = ⟨0|[ρ, Hmatter ]|n⟩ = ⟨0|ρ|n⟩ = −iω⟨0|ρ|n⟩
∂t iℏ iℏ
With this the symmetric term becomes
ω
Z
− kl λs d3 r rl rs ⟨0|ρ̂|n⟩ (3.108)
2
Using k · λ ≡ kl λl = 0 this can be written
ω 1 ω
Z
− kl λs d3 r (rl rs − δls r2 )⟨0|ρ̂|n⟩ = − kl λs ⟨0|Q̂ls |n⟩ (3.109)
2 3 6

where Q̂ls are components of the operator of the electric quadrupole tensor of the radiation
emitting matter
Z N
X
3 2
Q̂ls = d r (3rl rs − δls r )ρ̂(r) = qa (3ra,l ra,s − δls ra2 ) (3.110)
a=1

The emitted photon parameters enter via the factor (ω/6)kl λs in Eq.(3.109) while the
matter is represented by the electric quadrupole moment operator. Radiative transitions
arising through this term are called electric quadrupole transitions.

Electric quadrupole moment is an ℓ = 2 object. Selection rules

Electric quadrupole moment is a symmetric traceless tensor. This means that it has 5
independent components which transform between themselves under rotations. This is
similar to the 5 components of the second order spherical harmonic Y2µ or in a more
formal language to the 5 components of the ℓ = 2 representation (multiplet) of the group
of rotations.
Version of April 11, 2023 147

To see the relation it is useful to step back to Eq. (3.108), write the product rl rs in
spherical components
4π 2
rm rm′ = r Y1,m (Ω)Y1,m′ (Ω) with m, m′ = −1, 0, 1
3
and use the relations Eqs. (3.84,3.85) for l1 = l2 = 1
X
Y1,m (Ω)Y1,m′ (Ω) = a(1, 1, l)⟨lµ|1, m; 1, m′ ⟩Ylµ (Ω) , µ = m + m′
l=0,1,2

where in this case

3
a(1, 1, l) = p ⟨l0|1, 0; 1, 0⟩
4π(2l + 1)
Here the allowed values of l = 0, 1, 2 in the sum correspond to adding two units of angular
momenta and are formally dictated by the CG coefficients. The l = 1 term in the sum
vanishes since ⟨10|1, 0; 1, 0⟩ = 0 reflecting the vanishing of the antisymmetric (vector)
product of two identical vectors Y1m , cf, Ref. [5]. In the l = 0 term (the scalar product)
the corresponding CG coefficient ⟨00|1, 0; 1, 0⟩ ∼ δm,−m′ and since rm rm′ enter Eq. (3.108)
when written in spherical components as
X
∗ ∗
km λm′ rm rm′
m,m′ =−1,0,1

it vanishes (as with such terms earlier) due to orthogonality k · λ = 0.

We are thus left with only l = 2 term which shows that in spherical components the
cartesian tensor of the quadrupole moment becomes (a linear combination of) the five
components of the spherical representation of this tensor12
r r N
4π 4π X
Z
2 3
Q̂2µ = r Y2µ (θ, ϕ)ρ̂(r) d r = qa ra2 Y2µ (θa , ϕa ) (3.111)
5 5 a=1
p
Summarizing the above and using ⟨20|1, 0; 1, 0⟩ = 2/3 we have for Eq. (3.108)
ω
Z
− kl λs d3 r rl rs ⟨0|ρ̂|n⟩ =
2
ω X (3.112)
= −√ Φµ (Ωk , λ)⟨0|Q̂2µ |n⟩
6 µ

where we denoted
X
Φµ (Ωk , λ) = ⟨2µ|1, m; 1, m′ ⟩km
∗ ∗
λm′ =
m,m′ =−1,0,1
r (3.113)
4π X
′ ∗
= ⟨2µ|1, m; 1, m ⟩ k Y1µ (Ωk )λ∗m′
3
m,m′ =−1,0,1

12 Explicit
expressions relating the
√ spherical components Q2µ to √
the Cartesian Qls , Eq. (3.110) are
Q20 = −(1/2)Qzz , Q2,±1 = ±(1/ 6)(Qxz ± iQyz ) , Q2,±2 = −(1/2 6)(Qxx − Qyy + 2iQxy )
Version of April 11, 2023 148

The above expression is useful for finding the selection rules for electric quadrupole tran-
sitions in physical systems with eigenstates having definite angular momentum values

⟨0|Q̂2µ |n⟩ → ⟨ν ′ J ′ M ′ |Q̂2µ |νJM ⟩ (3.114)

Using the Wigner-Eckart theorem, Eq. (3.96), for the operator of the electric quadrupole
moment one has

⟨ν ′ J ′ M ′ |Q̂2µ |νJM ⟩ = ⟨J ′ M ′ |2µ, JM ⟩⟨ν ′ J ′ ||Q̂2 ||νJ⟩

which shows that the angular momentum selection rules for transitions with this operator,
i.e. for electric quadrupole transitions are

J ′ = |J − 2|, ...., J + 2 , M′ = M + µ (3.115)

Applying the parity transformation ra → −ra , a = 1, ..., N to Q̂ml , Eq. (3.110) we see
that it does not change. Thus the parity selection rule for electric quadrupole transitions
is
Pf = Pi (3.116)
This rule which is ”opposite” to the dipole selection rule, Eq.(3.104), is the main reason
why the electric quadrupole and magnetic dipole transitions explained below are the
leading mechanisms for the transition, which are forbidden by the dipole selection rules.
Since related to higher order terms in the long wavelength expansion, Eq. (3.70) such
transitions have order of magnitude smaller transition rates in the small ka parameter
than the allowed dipole transitions.
This is of course for the levels which satisfy the angular momentum selection rules,
Eq. (3.115). Transitions between levels with larger angular momentum differences are
controlled by higher terms in the LWA expansion, which are correspondingly weaker, cf.,
our discussion below and Ref. [3]. In this respect an interesting situation arises when a
matter system has a low lying high angular momentum state. If all the levels below such
state have low angular momenta this state will have a long radiative lifetime. Such states
are called isomeric and are metastable if probabilities of non radiative transitions (e.g.
via collisions in gases or phonon emission in solids) are small too.

Angular distribution of electric quadrupole radiation

Let us now briefly discuss the angular distribution of photons emitted in electric quadrupole
transitions in the very common case when the relation (3.114) is valid. Since in this case
µ is fixed, µ = M ′ − M only one term will remain in the sum in Eq. (3.112). The angular
distribution is obviously given by the corresponding function Φµ (Ωk , λ) with the depen-
dence on the angles of the polarization vectors as given e.g. by the relations Eq. (3.63).
The spherical components of the latter are

cos θe±iϕ
(λ1 )µ=±1 = ∓ √ , (λ1 )µ=0 = − sin θ
2
±iϕ
ie
(λ2 )µ=±1 =− √ , (λ2 )µ=0 = 0
2
Version of April 11, 2023 149

As an example let us consider transitions with M = M ′ , µ = 0. Using

3m2 − 1
⟨2µ|1, m; 1, −m⟩ = (−1)1−m √
6
one finds that the sum in (??) is −3 sin θ cos θ for the λ1 photon polarization while it van-
ishes for λ2 . Accordingly the corresponding angular distributions in electric quadrupole
transitions are
dNk1 dNk2
∼ sin2 θ cos2 θ , =0 (3.117)
dγ dγ
cf., Fig. 3.6.

Figure 3.6: Angular distribution of electric quadrupole radiation with λ1 polarization,

cf., Eq.(3.117) plotted in a similar way as in Fig. 3.3. Note that the independence of
the azimuthal angle ϕ means that the 3D version of this figure is obtained by rotating it
around the z-axis.

Magnetic dipole transitions

The antisymmetric part of (3.106) is conveniently expressed via vector products
1 1 1
kl λm (rl ĵm − rm ĵl ) = kl λm ϵlmn (r × ĵ)n = (k × λ) · (r × ĵ) (3.118)
2 2 2
which will contribute in Eq. (3.105) as

1
Z
−i(k × λkα ) · ⟨0| d3 r(r × ĵ(r))|n⟩ (3.119)
2

The emitted photon parameters enter via the factor (k × λkα ) while the matter is repre-
sented by the magnetic dipole moment of the current generated by the orbital motion of
the matter constituents,
N N
1 1 X qa qa
Z X
d3 r(r × ĵ(r)) = [−p̂a × ra + ra × p̂a ] = (ra × p̂a ).
2 2 a=1 2ma a=1
2ma
Version of April 11, 2023 150

At this stage it is important to recall the interaction term HI3 , Eq. (3.23) which we have
not treated so far. This term also depends on the magnetic moments of the matter
constituents. However not the ones generated by the orbital motion but rather by their
intrinsic motion, i.e. their spins.
The contribution of ĤI3 to the transition matrix element is straightforward to derive
just following the same steps which led us to the expression (3.58) for ⟨f |ĤI1 |i⟩ with the
result  1/2
ℏ
⟨f |ĤI3 |i⟩ = −i (k × λkα ) · ⟨0|m̂−k |n⟩ (3.120)
2ϵ0 ωk Ω
where
Z N
m̂(r)e−ik·r d3 r = µ̂a e−ik·ra
X
m̂k =
a=1

In the long wavelength limit

Z N
X
⟨0|m̂−k |n⟩ ≈ ⟨0|m̂0 |n⟩ = ⟨0| d3 r m̂(r) |n⟩ = ⟨0| µ̂a |n⟩
a=1

Taking this term into account the expression (3.119) becomes

−i(k × λkα ) · ⟨0|M̂|n⟩ ,

with
   X N   (3.121)
1 qa
Z 
3
M̂ = d r r × ĵ(r) + m̂(r) = (ra × p̂a ) + µ̂a
2 a=1
2ma

Radiative transitions arising through this term are called magnetic dipole transitions.
Electrons in atoms have equal charge to mass ratio e/m so the orbital part of this
expression reduces to
N
e X e
(ra × p̂a ) = L̂ (3.122)
2m a=1 2m
while the spin part
N N
X e X e
µ̂a = g ŝa = g Ŝ (3.123)
a=1
2m a=1 2m
where L and S are respectively the total orbital angular momentum and the spin of the
emitting matter system. In the spin part expression, Eq. (3.123) we used the gyromagnetic
ratio g(e/2m) to relate intrinsic magnetic moments to the spins. The part e/2m denotes
the classical value while g - known as the g-factor - is the dimensionless quantity to
account for deviations from the classical g = 1 value. Dirac relativistic equation for spin
1/2 particles predicts the value g = 2 while field theoretical corrections change it slightly
to g = 2(1 + α/2π + · · · ) ≃ 2.002319... where α = 1/137.
The expression (3.121) for electrons is therefore
ie
− (k × λkα ) · ⟨0|L̂ + g Ŝ|n⟩ (3.124)
2m
Version of April 11, 2023 151

i.e. it is proportional to the matrix element of the combination L̂ + g Ŝ ≈ L̂ + 2Ŝ of the

electronic angular momentum and spin.

Selection rules and angular distribution of magnetic dipole transitions

Let us now focus as in the electric dipole and quadrupole transitions on emitting systems
with eigenstates having defined angular momentum values

⟨0|M̂|n⟩ → ⟨ν ′ , J ′ , M ′ |M̂µ |ν, J, M ⟩ , µ = −1, 0, 1 (3.125)

where we also introduced the spherical components of the vector M̂. The angular mo-
mentum selection rules are obviously

J ′ = |J − 1|, ..., J + 1 , M′ = M + µ (3.126)

as in the electric dipole case.

However the parity selection rule is different. Indeed the magnetic dipole is a pseudo
vector as it does not change under the parity transformation r → −r , p → −p. Therefore
the parity will not change in the transitions i.e. the magnetic dipole parity selection rule
is
Pf = Pi
Let us also note the following. One can rewrite the operator L̂ + g Ŝ in Eq. (3.124) as
L̂ + Ŝ + (g − 1)Ŝ = Ĵ + (g − 1)Ŝ where Ĵ is the total angular momentum. In emitting
systems with eigenstates as in Eq. (3.125) the operator Ĵ can not cause transitions so that
the magnetic dipole emission must go via ”spin-flips”, i.e. (in conventional language)
via the change of the spin projection Sz . That in turn means that spin must not be a
conserved quantity in the eigenstates of Eq. (3.125). Which implies that there must be a
spin-orbit interaction in the matter Hamiltonian of the emitting system. Thus magnetic
dipole emission is the measure of such an interaction.
Finally let us address the angular distribution of photons emitted in magnetic dipole
transitions. This is determined by the angular dependence of the components of the vector
(k×λkα ) in Eq. (3.121) which are weighted by the vector of the matrix elements ⟨0|M̂|n⟩.
Noting that our favorite choice Eq. (3.63) of polarizations form right handed system of
unit vectors with the direction of k we deduce that the vectors (k × λkα ) with α = 1 and
α = 2 are respectively proportional to λk2 and λk1 . Therefore the angular distribution
of the expression (3.121) is identical to the electric dipole one with ⟨0|d̂|n⟩ replaced by
⟨0|M̂|n⟩ and appropriate adjustment of the polarization vectors.

General multipole expansion

What we have seen so far in our discussions of the electric dipole and quadrupole and
magnetic dipole emissions is essentially a transformation of the terms in the Taylor ex-
pansion (3.70) to the expansion in terms of ”angular” multipoles. The reason the latter
is more appropriate is that the small parameter of the long wavelength expansion ka ≪ 1
concerns the ”radial size” a of the system, |r| ≤ a with obviously no limitation on the an-
gles. Perhaps the simplest familiar example of this is a ”move” from the Taylor expansion
Version of April 11, 2023 152

of the Coulomb potential

1 ρ(r′ ) 3 ′
Z
ϕ(r) = d r
4πϵ0 r ′ ≤a |r − r′ |

’”outside” of a charge distribution, r > a to the multiple expansion

1
Z
Taylor expansion − ϕ(r) = ρ(r′ )(1 + r · r′ /r2 + ...)d3 r′ .
4πϵ0 r
∞ l E (3.127)
1 X X Mlm
Multipole expansion − ϕ(r) = Ylm (Ωr )
4πϵ0 Rl+1
l=0 m=−l

with
4π
Z
E
Mlm = rl ρ(r)Ylm
∗
(Ωr )d3 r − multipole electric moments
2l + 1
In deriving the latter expression one uses the known expansion of 1/|r − r′ | into a sum
∗
of products Ylm (Ωr )Ylm (Ωr′ ) which allows to factorize the outside r > a and the inside
′
r ≤ a regions. The result is the Coulomb potential represented as a sum of multipole
E
potentials which the electric multipole moments Mlm generate.
Returning to our problem we want to find a similar multipole expansion for the ex-
pressions Eqs. (3.58) and (3.120). Let’s concentrate on the former and consider
Z
ĵk · λ = d3 r ĵ(r) · λ e−ik·r (3.128)

Rather than expanding exp(−ik · r) in Taylor series as we have done in Eq. (3.70) we shall
use
X∞ X l
exp(−ik · r) = 4π (−i)l gl (kr)Ylm (Ωk )Ylm
∗
(Ωr )
l=0 m=−l
p
where the spherical Bessel functions are gl (kr) = π/2krJl+1/2 (kr). With this expansion
of the exponent
∞ X
X l 1
X Z
ĵk · λ = 4π (−i)l Ylm (Ωk )λq d3 r gl (kr)ĵq∗ (r)Ylm
∗
(Ωr ) (3.129)
l=0 m=−l q=−1

where we used the spherical components of the vectors λ and ĵ.

The resulting expression (3.129) for ĵk · λ has similar features with the multiple ex-
pansion Eq. (3.127) of the Coulomb potential. It is a sum of terms each factorized in
the product of components depending on the photon variables k and λ and the matter
variables j(r). One still has the matter component depending on k via gl (kr) but this will
decouple in the long wavelength approximation (LWA) kr ≪ 1 for which gl (kr) ∼ k l rl
when the matrix element ⟨0|ĵk |n⟩ · λ is considered.
The remaining problem in the expansion Eq. (3.129) is that photon and matter com-
ponents in each term do not have definite multipolarities. This is most obvious in the
photon related parts which transform under rotations as a product of Ylm (Ωk ) and λq ,
Version of April 11, 2023 153

i.e. as a sum of representations l − 1, l and l + 1. The technical reason for this is trivially
∗
obvious - the photon related Ylm (Ωk ) is coupled to the matter related Ylm (Ωr ) and λq
to ĵq∗ (r). What one needs is to ”recouple” the products into the photon and the matter
groups. This can be done using the Clebsch-Gordan completeness relation, cf., Ref. [3],
p. 338, X
⟨lm, 1q|l1LM ⟩⟨l1LM |lm′ 1q ′ ⟩ = δmm′ δqq′
LM

Inserting it into Eq. (3.129) one obtains

X
ĵk · λ = 4π (−i)l ΦLM,l (Ωk , λ)M̂LM,l (3.130)
L,l,M

where with gl (kr) ≈ (kr)l /(2l + 1)!! in the long wavelength limit have
X
ΦLM,l (Ωk , λ) = ⟨lm, 1q|l1LM ⟩k l Ylm (Ωk )λq , l = L, L ± 1 , (3.131)
m,q

and
1 XZ
M̂LM,l = d3 r rl ⟨l1LM |lm′ 1q ′ ⟩ĵq∗′ (r)Ylm
∗
′ (Ωr ) , l = L, L ± 1 , (3.132)
(2l + 1)!! ′ ′
m ,q

The values of l = L, L ± 1 in both expression correspond of course to the vector addition

of a unit angular momentum of the vectors λ and j to the l of Ylm (Ωk ) and Ylm (Ωr )
respectively.
The general expansion (3.130) has the structure we were looking for. Both ΦLM,l (Ωk , λ)
and M̂LM,l transform as the M components of the L-th representation of rotations. While
it is obvious for the photon components a bit more work is needed to show this for the
integral representing M̂LM,l . This is left as an exercise.
It is useful to consider a few simple cases.
Version of April 11, 2023 154

Monopole emission

Starting with the L = 0 term it is easy to show that it vanishes. Indeed for L = 0 have
M = 0 → m = −q and only l = 1 as a possible value. So k l Ylm (Ωk ) → kY1m (Ωk ) ∼ km
and X
Φ00,1 ∼ (−1)m km λ−m = k · λ = 0
m

So - no monopole photon emission. One can intuitively relate this to the fact that pho-
tons have spin 1 - one can’t emit a photon without changing the emission system angular
momentum by at least one unit.

Dipole emission
The terms with L = 1 have l = 0 and l = 1. For l = 0 have
Z
Φ1M,l=0 ∼ λM , M̂1M,l=0 ∼ d3 rĵM ∗
(r)

recovering the electric dipole case, cf. Eq.(3.73).

For the l = 1 value one has
X
Φ1M,l=1 = k ⟨1 M − q, 1q|111M ⟩Y1m (Ωk )λM −q ∼ (k × λ)M
q
1X
Z Z
3
M̂1M,l=1 = d r r⟨111M |1 M − q1q⟩ĵq∗ (r)Y1M
∗
−q (Ωr ) ∼ d3 (r × ĵ(r))M
3 q

i.e. the magnetic dipole emission.

Higher multipoles
Discussions of the higher values of L cf., Ref.[3], p.376, confirms this pattern - the
l = L terms correspond to magnetic multipoles while the l = L ± 1 terms are electric
radiation terms. So the sum over L in Eq.(3.130) is the sum over different multipoles
of the matter ”vibrations” (quantum mechanical transition matrix elements) causing the
photon emission.

Angular distribution, selection rules of the general multipole terms

The expression (3.131) for ΦLM,l (Ωk , λ) can be interpreted as the probability amplitude of
the photon emitted by the l, L matter multipole into the solid angle Ωk with polarization
λ. It reflects the expectation that the total angular momentum of a photon is a sum of
its orbital angular momentum (encoded in Y1m (Ωk )) and its unit spin (associated with
the polarization vector).
Angular momentum selection rules for the terms in the expansion Eq. (3.130) follow
by applying the Wigner-Eckart theorem to matrix elements of the multipole moments
operators between matter eigenstates with defined angular momentum values

⟨ν ′ , j ′ , m′ |M̂LM,l |ν, j, m⟩ = ⟨j ′ m′ |LM, jm⟩⟨ν ′ , j ′ ||M̂L,l ||ν, j⟩ (3.133)

Version of April 11, 2023 155

From this we have the angular momentum selection rules

|L − j| ≤ j ′ ≤ L + j , m′ = m + M (3.134)

Perhaps not surprisingly they do not involve the l index which distinguishes between
electric and magnetic multipoles. This index is important however in the parity selection
rules. Perhaps the fastest way to see this is to observe that ΦLM,l (Ωk , λ) changes under
the parity r → −r transformation as (−1)l+1 where l comes from the orbital Ylm (Ωk )
while the extra minus from the polar vector of the polarization. So

(−1)Pf = (−1)l+1 (−1)Pi

As a final remark we note that the above arguments based on the parity properties of
the photons amplitudes could be made more formal and rigorous by examining how the
matrix elements Eq. (3.133) behave under the parity transformation, cf., Ref. [3], p. 379.

3.2.7 Induced photon emission

In our discussion above of the photon emission by an excited state of quantum matter
(atom, solid, nucleus, molecule) we have assumed that prior to the emission (i.e. in the
initial state) there were no photons present in the radiation mode kα into which the
matter system emits the photon, cf. Eq. (3.56). Such an emission is called spontaneous.
Let us now consider what happens if the initial state already contained N photons
before the emission, i.e. have

|i⟩ = |n⟩|Nkα , {0k′ α′ }⟩ |f ⟩ = |0⟩|(Nkα + 1, {0k′ α′ }⟩

With this change the calculation of the field matrix element in Eq. (3.57) becomes
q
⟨Nkα + 1, {0k′ α′ }|âk′′ α′′ + ↠′′ ′′ |Nkα , {0kα }⟩ = δ−k′′ , k δα′′ α Nkα + 1
−k α

because of the basic matrix element of the harmonic oscillator creation operator
√
⟨N + 1|↠|N ⟩ = N + 1

This produces the following result in the absolute values square of the interaction
 
2 ℏ
|⟨f |ĤI1 |i⟩| = |⟨0|ĵk · λkα |n⟩|2 (Nkα + 1) (3.135)
2ϵ0 ωk Ω

which means that the emission rate Γn→0,kα is Nkα + 1 times larger than in the sponta-
neous emission case. So just the initial presence of Nkα photons in the radiation modes
into which the emission occurs leads to this increase of the emission rate. This effect is
called induced or stimulated emission. It is often interpreted as a quantum mechanical
effect of ”bosons like to stick together”, i.e. to be in the same state and is the key to the
idea of lasers.
Very schematically this idea can be outlined as following. Assume a large number
of identical ”emitters” (e.g. atoms, molecules, etc) which can be ”continuously” excited
Version of April 11, 2023 156

to a certain energy level and then de-excite to low lying levels via photon emission. As
we learned earlier the angular distribution and polarization of the emitted photons will
depend on the angular momentum projections M and M ′ of the initial and final states
but if only the initial energy is specified the M values will be random and so will be the
emitted photons directions and polarizations. This is as long as only the spontaneous
emission is considered.
If some particular photon modes kα contain a (large) number of (pre emitted) photons
then high probability (∼ Nkα ) induced emission, i.e. ”lasing” will occur into these
particular modes. Schematically the needed accumulation of photons in controlled modes
is achieved e.g. by placing the emitters in a resonator. This selects resonating modes in
which photons ”bounce back and forth” before escaping.
All this is very sketchy of course. More detailed explanations are found in appropriate
quantum optics literature.
Let us note that historically it is common to write the expression for the emission rate
as a sum of the term containing the Nkα and the term containing 1 from the sum Nkα + 1

Γ = Γinduced + Γspontaneous

This the expression for the spontaneous photon emission rate (3.60) is changed to
induced sponteneous
dNkα dNkα dNkα
 
= + (3.136)
dγ dγ dγ

with induced sponteneous

dNkα dNkα
 
= Nk α (3.137)
dγ dγ

3.2.8 Photon absorption

Consider now the process of the photon absorption. We have

|i⟩ = |0⟩|Nkα , ...⟩ , |f ⟩ = |n⟩|Nkα − 1, ...⟩ (3.138)

The matrix element of ĤI1 between these states gives

 
ℏ
|⟨f |ĤI1 |i⟩|2 = |⟨n|ĵ−k · λkα |0⟩|2 Nkα (3.139)
2ϵ0 ωk Ω

Since
⟨n|ĵ−k · λkα |0⟩ = ⟨0|ĵk · λkα |n⟩∗
we find equality relation
Γabsorption
0→n = Γinduced
n→0
emission
(3.140)
for absorption and induced emission rates of photons with the same k and λ. This
relation is crucial for laser physics. Indeed it shows that having Nkα incident photons
(per unit time) of energy ℏωk a photon has an equal probability of being absorbed by
a ground-state atom or being duplicated (amplified!) via an induced emission by an
Version of April 11, 2023 157

excited-state atom. To favor emission over absorption, there need to be more excited-state
atoms than ground-state atoms. This of course doesn’t happen in thermally equilibrated
systems. A non equilibrium situation must be created by adding energy via a process
known as “pumping” in order to raise enough atoms to the upper level. The result called
“population inversion” leads to light amplification. Pumping may be electrical, optical or
chemical.

3.3 Appendix
3.3.1 Discrete level coupled to continuum
Here we present details of a simple non perturbative approach to deal with the Weisskopf-
Wigner model as defined in Section (3.2.1). A more general treatment of this problem is
reviewed in e.g. Ref.[2].

Neglecting coupling between continuum levels

As was described following Eq. (3.37) the crucial step/approximation in the Weisskopf-
Wigner approach is to neglect the coupling between the continuum levels, i.e. to set
Vµν = 0. This means that the Hamiltonian matrix in the basis {ψ0 , ψν } has the ”bordered”
form  
E0 V01 ... V0ν ...
 V10 E1 ... 0 .... 
 
H=  ... ... ... ... ...  
 Vν0 0 ... Eν ... 
... ... ... ... ...
Here we tacitly assumed discrete values of the ν index. Such matrices are easy to diag-
onalize especially when simplifying assumptions about Eν ’s and V0ν are made. This is
described in e.g. Ref. [1].

Markov approximation
Examining the integral expression Eq. (3.46) for the kernel K(t) which for convenience we
rewrite here
1
Z
K(t) = − 2 dE|V0µ |2 ei(E0 −E)t/ℏ
ℏ Eµ =E

we observe that the integrand is a product of in general a smooth function of E and an

exponential which oscillates in E with the period ∼ ℏ/t. Denoting by ∆ the scale over
which |V0µ |2 changes it is clear that the integral will tend to zero for long times
Eµ =E

t ≫ ℏ/∆. Under this condition the kernel K(t) has the ”range”
ℏ
T ∼
∆
Let us change the variable t′ in the integral in (3.41) to τ = t − t′
Z t Z t
dc0 ′ ′ ′
= K(t − t )c0 (t )dt = K(τ )c0 (t − τ )dτ (3.141)
dt 0 0
Version of April 11, 2023 158

For a given t only the values of c0 (t − τ ) within ”memory times” τ ≤ T of K(τ ) contribute
in the integral. To simplify further we next assume that c0 (t) changes little over the time
T . We will address below the meaning of this assumption. When it is valid we can
approximate under the integral
c0 (t − τ ) ≈ c0 (t)
and write
t
dc0
Z
= c0 (t) K(τ )dτ (3.142)
dt 0

This approximation is called the Markov approximation - dynamics of c(t), i.e. how it
changes at the time t depends only on its value at the time t and not on earlier times
t′ < t, i.e., it has no memory of the past.
The integral on the right hand side of the above equation is a known function of t so
the equation can be integrated but let us first make one more simplification. We will be
interested in the long time behaviour of c0 (t) for t ≫ T . Since by assumption K(τ ) is
small for τ ≫ T we can approximate
Z t Z ∞
K(τ )dτ ≈ K(τ )dτ
0 0

Let us introduce the following notation for the real and imaginary part of this integral
Z ∞ Z ∞
∆E Γ
Im K(τ )dτ = − , Re K(τ )dτ = (3.143)
0 ℏ 0 2

With this we have for the time dependence of the ”persistence amplitude” of the initial
state ψ0
⟨ψ0 |Ψ(t)⟩ = c0 (t)e−iE0 t/ℏ |t≫T = c(0)e−i(E0 +∆E)t/ℏ e−Γt/2 (3.144)
We will see below that Γ is positive so this amplitude decays exponentially with the decay
rate Γ. Its phase acquires energy shift ∆E.
We will discuss the explicit form of ∆E and Γ in the next subsection. Here we note
that the time scale over which c0 (t) changes is ∼ 1/Γ or ∼ ℏ/∆E. This our assumption
of c0 (t) changing slowly in the interval T = ℏ/∆ means that must have

ℏΓ ≪ ∆ , ∆E ≪ ∆ (3.145)

Decay rate (width) and the energy shift of a decaying state

We now provide explicit expressions for Γ and ∆E. Consider the integral
Z t0
ei(E0 −E)t0 /ℏ − 1
Z
ℏ lim K(τ )dτ = i lim dE |V0µ |2 (3.146)
t0 →∞ 0 t0 →∞ Eµ =E E0 − E

where we used Eq. (3.46) for K(t). To calculate the t0 → ∞ limit we will use the following
device (cf., Ref. [4], Sec. 43). Let us shift the integration contour over E slightly into the
lower imaginary half plane (with Im E < 0), cf. dashed line in Fig.3.7
Version of April 11, 2023 159

Figure 3.7: The resulting integration contour in the iϵ prescription. It was obtained first
shifting the integration contour along the real axis in Eq.(3.146) to the lower ImE < 0
half-plane, then letting t0 → ∞ for which the 1st term in the integrand vanishes and then
bringing the contour back to the real axis with a small semicircle around the singularity
point at E = E0 . Note that choosing the semicircular shape is a matter of convenience
allowing to obtain easily the conventional result (3.149) as explained in the text.

This can be done without changing the value of the integral since the integrand has
no singularities on the real axis13 . The integral above can then be separated into a sum
of two
ei(E0 −E)t0 /ℏ 1
Z Z
dE |V0µ | 2 and − dE |V0µ |2
Eµ =E E0 − E Eµ =E E0 − E
This was not possible when the integration was over the real E axis because each term
separately is singular at E = E0 .
In the limit t0 → ∞ the first integral tends to zero (due to the presence of the eImE t0
factor in its integrand) and we are left with the second integral. There we can bring the
integration over E back to the real axis taking care that it doesn’t cross the pole at E = E0 ,
cf., Fig.3.7 Z ∞
1
Z
ℏ K(τ )dτ = i dE |V0µ |2 (3.147)
0 contour in Fig.3.7
Eµ =E E − E0
It is convenient and conventional to view the resulting integration contour in the above
integral in a following way. Let us first add a small positive imaginary quantity iϵ to E0 ,
then shift the contour to the real axis (it goes under the pole so there is no problem) and
at the end consider the limit of ϵ → 0 deforming the contour to prevent the pole crossing
it.
13 To be precise one should write the definite energy integral with its limits EEmax dE... and assume
R
min
that the contour end points (which are fixed and can’t be moved to the complex plane) give negligible
contribution.
Version of April 11, 2023 160

This procedure is often called the iϵ prescription and using it we write the expression
(3.147) as Z ∞
1
Z
ℏ K(τ )dτ = i lim dE |V0µ |2 (3.148)
0 ϵ→0 E − E0 − iϵ

The integral here can be transformed using the formula

b b
f (x) f (x)
Z Z
lim dx = P dx ∓ iπf (0) (3.149)
ϵ→0 a x ± iϵ a x

valid for a < 0 and b > 0. Here P denotes the principle value of the integral
Z b "Z Z b #
−ϵ
f (x) f (x)
P dx = lim dx + dx (3.150)
a x ϵ→0 a ϵ x

The two terms in (3.149) correspond to the integral along the real axis with the excluded
interval −ϵ < x < ϵ and the integral along the semicircle of radius ϵ around the singularity
at x = 0, cf., Fig.3.7. The value of the second term is just half of the ∓2πif (0) from
the application of the Cauchy’s residue theorem to the full circle (or just calculating the
integral using polar coordinates Re x = r cos ϕ, Im x = r sin ϕ in the complex plane).
Using this for the integral in (3.148) we arrive at the expressions (3.48) for the energy
shift ∆E and the width Γ.

3.3.2 The ĤI3 part of the Hamiltonian and the parity transfor-
mation
The invariance under parity transformation of the part ĤI3 , Eq.(3.10) of the (non rel-
ativistic) matter-EM field Hamiltonian follows since both B and the particle spins ŝa
do not change their signs under the parity transformation. They are axial (or pseudo)
vectors. For the magnetic field this is already seen in the Lorenz force

F = q(E + v × B)

Since the force F = mdv/dt it must be a polar vector so must be E. In the second term
since v is polar B must be axial. This is also seen in the Maxwell equations

∇ × E = −∂B/∂t , c2 ∇ × B = ∂E/∂t + j/ϵ0

as well as in the relation B = ∇ × A.

Spin vectors s are axial as they are part of the total angular momentum j = l + s with
the orbital part l = r×p which obviously doesn’t change under the parity transformation.
It is not difficult to gain intuition about this peculiar property of l. Indeed the orbital
angular momentum reflects/measures the magnitude and the direction of the ”rotational”
component with respect to the origin r = 0 in a (chosen) coordinate system of a particle
motion at a position r moving with the momentum p. Changing the particle position
r → −r and the momentum p → −p leave the direction and the magnitude of the
rotational component of the motion the same.
Bibliography

[1] Atom-Photon Interactions: Basic Processes and Applications, Claude Cohen-

Tannoudji, Jacques Dupont-Roc, Gilbert Grynberg

[2] K.J.F Gaemers and T.D Visser, Deviations from Exponential Decay in Quantum Me-
chanics, Physica, A153 (1988) 234-251
[3] G. Baym, Lectures on Quantum Mechanics, Addison-Wesley, 1993
[4] L.D Landau and E.M. Lifshitz, Quantum Mechanics, Pergamon Press, 1965

[5] D. Brink and G. R. Satchler, Angular Momentum, Oxford University Press, 2nd Edi-
tion. 1968
[6] J. D. Bozek, et al, J. Phys. B: At. Mol. Opt. Phys. 39 (2006) 4871?4882
[7] J.J. Sakurai, Jim Napolitano, Modern quantum mechanics, 2nd ed. Addison Wesley,
1994
[8] D. Griffiths, Introduction to Elementary Particles, 2nd Revised Ed, Wiley, 2008.

161
Chapter 4

Quantization of the
Schrödinger Field - The Second
Quantization

4.1 Introduction
Let us consider the Schrödinger equation for a free particle

∂ψ(r,t) ℏ2 2
iℏ =− ∇ ψ(r,t) (4.1)
∂t 2m
and regard it as equation for a classical field ψ(r, t) just like we regarded the Maxwell
equations for the electromagnetic field. To remind - the Maxwell field after the quan-
tization describes free quanta - photons - which behave like quantum particles. Their
energy-momentum relation ϵ = c|p| is determined by the classical dispersion relation
ω = c|k| of the free EM waves supplemented with the basic QM particle-wave relations
ϵ = ℏω and p = ℏk.
For the free Schrödinger field the dispersion relation is read off the equation (4.1) as

ℏ2 |k|2
ℏω =
2m
which suggests that the quantization of this field will lead to the description of free quanta
with the energy-momentum relation

p2
ϵ=
2m
i.e. that of non relativistic particles. This procedure is called second quantization for
obvious reasons.
We will start by confirming this picture and then extending it to describe particles
moving in an external potential and also interacting between themselves. In the process of

162
Version of April 11, 2023 163

doing this we will discover that the formalism describes identical particles obeying boson
statistics. We will then understand how to extend the formalism to describe particle
obeying fermion statistics.

4.2 Free Schrödinger field. Quantization

The Schrödinger field ψ(r, t) is a scalar field and in that it is simpler than the vector
EM field. It is however complex valued unlike the real valued EM field. The last feature
means that actually the equation (4.1) should be considered as a pair of equations for real
and imaginary parts or equivalently for ψ(r) and its complex conjugate ψ ∗ (r)
∂ψ(r,t) ℏ2 2 ∂ψ ∗ (r,t) ℏ2 2 ∗
iℏ =− ∇ ψ(r,t) ; −iℏ =− ∇ ψ (r,t) . (4.2)
∂t 2m ∂t 2m
Following the motivation outlined in the Introduction we consider the quantization of
this field. We follow the standard quantization procedure and start by identifying the
hamiltonian structure and the canonical conjugate pairs of the Schrödinger field.
We note that the pair of equations (4.2) can be regarded as the Hamiltonian pair of
equations with the Hamiltonian
ℏ2
Z
H= d3 r|∇ψ(r)|2 (4.3)
2m
Indeed, the variation of this expression gives
ℏ2
Z
δH = d3 r [∇ψ ∗ (r) · ∇δψ(r) + ∇δψ ∗ (r) · ∇ψ(r)] =
2m
ℏ2
Z
d3 r [∇2 ψ ∗ (r)]δψ(r) + [∇2 ψ(r)] δψ ∗ (r)

= − (4.4)
2m
Now regarding δψ(r) and δψ ∗ (r) as independent we read off that
δH ℏ2 2 ∗ δH ℏ2 2
=− ∇ ψ (r) ; ∗
=− ∇ ψ(r) (4.5)
δψ(r) 2m δψ (r) 2m
This shows that the pair (4.2) is indeed the Hamiltonian pair provided one considers
ψ(r) and iℏψ ∗ (r) (i.e. their values at every space point r) as respectively canonically
conjugate coordinates and momenta,
∂ψ(r, t) δH ∂(iℏψ ∗ (r, t)) δH
= , =− (4.6)
∂t δ(iℏψ ∗ (r, t)) ∂t δψ(r, t)

4.2.1 Separating the real and imaginary parts

The complex valuedness of ψ and iℏψ ∗ presents a slight problem in applying the standard
rules of the canonical quantization procedure. A possible way to avoid this problem is to
transforms to the real and imaginary parts

ψ = Reψ + iImψ ; ψ ∗ = Reψ − iImψ (4.7)

Version of April 11, 2023 164

We will proceed with this for a little while and use it to learn how to quantize using the
original ψ and iℏψ ∗ . As we will see it will be a more convenient (and conventional) option.
One must take care that the transformation (4.7) is canonical to make sure that the
transformed variables also form a canonical pair. This is achieved by
1 1
ψ(r) = √ [ϕ(r) + iπ(r)] ; ψ ∗ (r) = √ [ϕ(r) − iπ(r)] (4.8)
2ℏ 2ℏ

with real ϕ and π. To verify that ϕ and π are canonical let us use the known property1
that time independent canonical transformation from a set pk , qk to Pk , Qk obeys
X X
pk dqk = Pk dQk + dF
k k

where dF is a total differential. In our case the sum over k is the integral over r so that
1
Z Z
iℏψ ∗ (r)dψ(r)dr = iℏ [ϕ(r)dϕ(r) + iϕ(r)dπ(r) − iπ(r)dϕ(r)+
2ℏ
i 2
Z Z
ϕ (r) + π 2 (r) + 2iϕ(r)π(r) dr

+ π(r)dπ(r)] dr = π(r)dϕ(r)dr + d
4
showing that indeed ϕ and π are canonical i.e. difference of the symplectic forms in the
old and the new canonical variables is a complete differential.
Inserting (4.8) into the equations (4.2) we obtain

∂ϕ(r,t) ℏ 2 ∂π(r,t) ℏ 2
=− ∇ π(r,t) ; = ∇ ϕ(r,t) . (4.9)
∂t 2m ∂t 2m
The Hamiltonian becomes
ℏ2
Z Z
3 2 ℏ
d3 r (∇π(r))2 + (∇ϕ(r))2
 
H= d r|∇ψ(r)| = (4.10)
2m 4m
Its variation
Z
ℏ
δH = d3 r [∇π(r) · ∇δπ(r) + ∇ϕ(r) · ∇δϕ(r)] =
2m
Z
ℏ
d3 r ∇2 π(r)δπ(r) + ∇2 ϕ(r)δϕ(r)
 
= − (4.11)
2m
so that
δH ℏ 2 δH ℏ 2
=− ∇ ϕ(r) , =− ∇ π(r) (4.12)
δϕ(r) 2m δπ(r) 2m
Thus we see that Eqs. (4.9) indeed are in the Hamiltonian form

∂ϕ(r,t) δH ∂π(r,t) δH
= , =− . (4.13)
∂t δπ(r, t) ∂t δϕ(r, t)
1 cf, L. Landau and E. Lifshitz, Mechanics, Sec.45, Elsevier Ltd. 1976
Version of April 11, 2023 165

with ϕ(r) as coordinates and π(r) as momenta. These are real and we can quantize the
theory in the usual way by introducing wave functionals Ψ[ϕ(r)] and operators which act
on them
δ
ϕ̂(r)Ψ[ϕ(r)] = ϕ(r)Ψ[ϕ(r)] ; π̂(r)Ψ[ϕ(r)] = −iℏ Ψ[ϕ(r)] . (4.14)
δϕ(r)
with the Schrödinger equation
∂
Ψ[ϕ(r), t] = Hop Ψ[ϕ(r), t]
iℏ (4.15)
∂t
where the Hamiltonian operator is given by the expression (4.10) with ϕ(r) and π(r)
replaced by the corresponding operators
Z h i
ℏ
Hop = d3 r (∇π̂(r))2 + (∇ϕ̂(r))2 (4.16)
4m
We note that the commutation relations for the operators ϕ̂(r) and π̂(r) are
[ϕ̂(r), ϕ̂(r′ )] = [π̂(r), π̂(r′ )] = 0
δϕ(r′ )
[π̂(r), ϕ̂(r′ )] = π̂(r) ϕ̂(r′ ) − ϕ̂(r′ ) π̂(r) = −iℏ = −iℏδ(r − r′ ) (4.17)
δϕ(r)

4.2.2 Back to the complex valued field

As was already stated it is more convenient to work with complex valued field. Using
ˆ and π̂(r))
(4.8) we introduce non hermitian combinations of the operators ϕ(r)
1 1
ψ̂(r) = √ [ϕ̂(r) + iπ̂(r)] ; ψ̂ + (r) = √ [ϕ̂(r) − iπ̂(r)] (4.18)
2ℏ 2ℏ
We note here a clear analogy with the familiar operators â and â+
1 1
â = √ [x̂ + ip̂] , â+ = √ [x̂ − ip̂] ,
2ℏ 2ℏ
the non hermitian combinations of coordinate and momentum operators for a single degree
of freedom.
In our case we have such combinations (4.18) for every r, i.e. for ∞3 coordinate-
momentum pairs.
From the commutation relations (4.17) we have
[ψ̂(r), ψ̂ + (r′ )] = δ(r − r′ ) , [ψ̂(r), ψ̂(r′ )] = 0 = [ψ̂ + (r), ψ̂ + (r′ )] (4.19)
Looking back at (4.6) we observe that had we postulated the usual commutation relations
for the operators corresponding to the complex field canonical coordinates and momenta
ψ(r) and iℏψ(r)
[ψ̂(r), iℏψ̂ + (r′ )] = iℏδ(r − r′ ) (4.20)
we would have arrived at the same result Eq. (4.19).
As we will soon see the commutation relations (4.19) will be essentially all (well almost
all) we will need to know about the operators ψ̂(r) and ψ̂ + (r) in order to understand how
they act on any wave function.
Version of April 11, 2023 166

4.2.3 The Hamiltonian of the free Schrödinger field

The Hamiltonian operator is directly obtained from Eq. (4.3) by replacing ψ(r) and ψ ∗ (r)
with the operators ψ̂(r) and ψ̂ + (r). One usually finds two expressions in the literature
ℏ2
Z
Hop = d3 r∇ψ̂ + (r)∇ψ̂(r) (4.21)
2m
or  2 2
ℏ ∇ ℏ2
Z Z
3 +
Hop = d rψ̂ (r) − ψ̂(r) = − d3 rψ̂ + (r)∇2 ψ̂(r) (4.22)
2m 2m
The difference is obviously just a ”surface term” at large distances. This is an opera-
tor valued term so disregarding this difference means the requirement that all the wave
functions of the field will produce zero when acted upon by such ”surface” operators.
Note also the order of the operators chosen in the above expression for Hop with ψ̂ + (r)
acting after ψ̂(r). As will become clear below this order of operators (called ”normal
ordering”) assures that the vacuum of the theory has zero energy.

4.2.4 The eigenstates. Field quanta are free non relativistic par-
ticles
Having established the form of the Hamiltonian operator of the theory we should proceed
to solve the Schrödinger equation of the theory
∂
iℏ |Ψ(t)⟩ = Hop |Ψ(t)⟩ (4.23)
∂t
Note - to solve the Schrödinger equation for the quantum Schrödinger field!

The normal modes

Since Hop is time independent we can solve the above equation in a standard way by first
finding the eigenfunctions of the Schrödinger field Hamiltonian i.e. solutions of
Hop Ψ = EΨ (4.24)
To this end we will go to the normal modes of the field. As we know from the systems with
finite number of degrees of freedom quadratic Hamiltonians become sums of independent
terms when the original degrees of freedom are transformed to the normal modes.
To find the normal modes let us recall that they are special solutions of the classical
equations in which all the degrees of freedom of the physical system oscillate with the
same frequency. Classical equations in the present case are just the field equations (4.2).
Their solutions with a given frequency ψ(r, t) = u(r) exp(−iωt) satisfy
ℏ2 2
∇ u(r) + ℏω u(r) = 0 (4.25)
2m
and can be chosen as plane waves
1 ℏ2 k 2
uk (r) = √ eik·r with ℏω = (4.26)
Ω 2m
Version of April 11, 2023 167

where we assumed the usual periodic boundary condition in a very large volume Ω
2π
k= (nx , ny .nz ) with nx , ny , nz = 0, ±1, ±2, ...
Ω1/3
We now expand the field operators using these normal modes
1 X
âk eik·r
X
ψ̂(r) = âk uk (r) = √
Ω
k k
+
X
+ ∗ 1 X + −ik·r
ψ̂ (r) = â uk (r) = √ â e (4.27)
k Ω k
k k

The coefficients âk and â+ in the above expansions of the field operators are obviously
k
operators. This expansion must be viewed as a transformation from a canonical set of
2 × ∞3 operators ψ̂(r) , ψ̂ + (r) to another canonical set of 2 × ∞3 operators âk , â+ .
k
Using orthonormality of uk ’s it is easy to invert (4.27)

1 1
Z Z
−ik·r
âk = √ 3
d r ψ̂(r)e +
, â = √ d3 r ψ̂ + (r)eik·r
Ω k Ω
Using the commutations (4.19) one can then find the commutation relations between âk ’s
and â+′ ’s,
k
[âk , â+′ ] = δkk′ , [âk , âk′ ] = 0 = [â+ , â+′ ] (4.28)
k k k
which of course express the harmonic oscillator character of the normal modes for each k
and their independence for different k’s.

Diagonalizing the field Hamiltonian

Inserting the expansions (4.27) into the Hamiltonian Eq. (4.22) we obtain a sum of inde-
pendent (commuting) oscillators
X
Hop = ϵk â+ âk (4.29)
k
k
where we have denoted the energies of the oscillator quanta

ℏ2 k2
ϵk = (4.30)
2m
Based on this it is trivial to find the eigenfunctions and eigenenergies of each term. Clearly
the eigenstates of this Hop are products of the familiar harmonic oscillator-like states (cf.,
Appendix, Eq. (4.159))
n
Y (â+ ) k
pk
Y
|{nk }⟩ = |nk ⟩ = |0⟩ (4.31)
nk !
k k
Version of April 11, 2023 168

with eigenvalues which are

X
E{n } = ϵk n k with each nk = 0, 1, 2, . . . (4.32)
k
k
So the eigenenergies of the free Schrödinger field are sums over the modes uk (r) of integer
numbers nk of quanta with energies ϵk . To understand the physics of these quanta it is
useful to ask/determine what are their momenta. For this one must find the corresponding
operator. We deal with this in the next section.
We note that the ground state corresponds to all nk = 0, i.e. it is the vacuum state
|0⟩. Its energy is equal to zero which was assured by the normal ordered form of Hop ,
Eq. (4.22), which we have adopted. Let us also note that in this formulation the only
properties we will ever need of the vacuum state are that it gives zero when acted upon
with anyone of the operators âk and that it is normalized

âk |0⟩ = 0 , ⟨0|0⟩ = 1 (4.33)

Let us also note that the most general states of the theory are linear combinations of
the eigenstates (4.31) X
|Ψ⟩ = C{n } |{nk }⟩ (4.34)
k
{n }
k
They may appear e.g. as solutions of the time dependent Schrödinger equation of the
field
∂|Ψ(t)⟩
iℏ = Hop |Ψ(t)⟩ (4.35)
∂t
with coefficients depending on time via the usual

C{n (t) = C{n (0) exp(−iE{n } t/ℏ)

k} k} k
We note that the number of particles in the above expressions for C{n is given by
k}
X
N= nk
k
It is important to note that nowhere in the formalism there appears a requirement that N
is fixed, i.e. has the same value in the e.g. expression for the general wave function |Ψ⟩.
The formalism in principle allows to have states with coherent combinations of different
particle numbers. We will address this issue in the last section.

Degeneracy of the normal modes. Spherical waves

The normal modes Eq. (4.26) are clearly infinitely degenerate having the same frequency
ω for all k with the same k = |k|. This of course follows from the degeneracy of the
solutions of the (free Schrödinger ) equation (4.25). This degeneracy means that other
sets can be chosen for a given k. The familiar spherical or cylindrical waves rather than
the plane waves would supply examples of such sets.
Version of April 11, 2023 169

Let us consider the spherical waves set of solutions

uklm (r) = Rkl (r)Ylm (θ, ϕ)

with l, m the angular momentum and its projection (for a free particle) and Rkl (r) and
Ylm (θ, ϕ) the radial and angular parts2 . We can expand the field operators using such
normal modes3
X
ψ̂(r) = âklm uklm (r, θ, ϕ)
klm
X
+
ψ̂ (r) = â+ ∗
klm uklm (r, θ, ϕ) (4.36)
klm

with the operators

Z Z
âklm = 3
d rψ̂(r) u∗klm (r, θ, ϕ) , â+
klm = d3 rψ̂ + (r) uklm (r, θ, ϕ)

and (as can be easily checked) the commutation relations equivalent to Eq. (4.28) with k
and k′ indices replaced by klm and k ′ l′ m′ .
Inserting the expansions (4.36) into the Hamiltonian Eq. (4.22) we obtain
X
Hop = ϵklm â+
klm âklm (4.37)
klm

As with the plane waves it is a sum of independent (commuting) oscillators with quanta
energies depending only on k
ℏ2 k 2
ϵklm =
2m
i.e. equal to the energy of the plane wave quanta - reflecting the degeneracy of the normal
modes.

4.2.5 Momentum and angular momentum

Field momentum
We now discuss the total (mechanical) momentum of the Schrödinger field. To find it
expression we could go back to the classical fields and use the Noether theorem. We
prefer to find it by considering the generator of the translations r → r + a with a constant
vector a. The field operators change as ψ̂(r) → ψ̂(r + a) and ψ̂ + (r) → ψ̂ + (r + a). So we
are looking for the operator Pop with which
   
ψ̂(r) ψ̂(r + a)
e−ia·Pop /ℏ eia·Pop /ℏ = (4.38)
ψ̂ + (r) ψ̂ + (r + a)
2 Recall the solutions of the stationary Schrödinger equation for a free particle in spherical coordinates,

cf., Sakurai, Modern Quantum Mechanics, Sec.3.7, Addison-Wesley, 1994.

3 For convenience we assume that k values are made discrete by imposing boundary condition in a

large spherical box

Version of April 11, 2023 170

For infinitesimal a this is

   
ψ̂(r) (1 + a · ∇)ψ̂(r)
(1 − ia · Pop /ℏ) (1 + ia · Pop /ℏ) =
ψ̂ + (r) (1 + a · ∇)ψ̂ + (r)
which means that must have the commutator
    
ψ̂(r) iℏ∇ψ̂(r)
Pop , =
ψ̂ + (r) iℏ∇ψ̂ + (r)
This is achieved with the expression
Z
Pop = d3 r′ ψ̂ + (r′ )(−iℏ∇r′ )ψ̂(r′ ) (4.39)

Indeed
ψ̂ + (r′ )(−iℏ∇r′ )ψ̂(r′ ) , ψ̂(r)
   Z  
ψ̂(r) 3 ′
Pop , = d r =
ψ̂ + (r) ψ̂ + (r′ )(−iℏ∇r′ )ψ̂(r′ ) , ψ̂ + (r)
δ(r − r′ )iℏ∇r′ ψ̂(r′ )
Z    
3 ′ iℏ∇r ψ̂(r)
= d r =
ψ̂ + (r′ )(−iℏ∇r′ )δ(r − r′ ) iℏ∇r ψ̂ + (r)
where in the second line of the last equality we used integration by parts.

Field quanta are free nonrelativistic particles

The momentum Pop commutes with the Hamiltonian Hop , Eq. (4.22)
[Hop , Pop ] = 0 (4.40)
Verifying this explicitly with Hop and Pop written in terms of the field operators ψ̂(r)
and ψ̂ + (r) is a good exercise which is left to the reader. Physically this is the result of
the invariance of Hop under the translation.
Let us write Pop in terms of the normal modes operators âk ’s and â+′ ’s. Using the
k
expansions (4.27) in (4.39) we obtain
X
Pop = ℏk â+′ âk′ (4.41)
k
k
This expression compared to Eq. (4.29) trivially shows that indeed Pop commutes with
Hop . It has the same eigenfunctions (4.31) and its eigenvalues are
X
P{n } = ℏk n k (4.42)
k
k
This shows that each field quantum with energy ϵk carry momentum pk = ℏk. The
energy momentum relation ϵ(p) follows from the explicit dependence of ϵk = ℏ2 k2 /2m
on k
|p |2
ϵk (p) = k (4.43)
2m
which is the familiar energy-momentum relation of non relativistic particles. This indi-
cates that quanta of the free Schrödinger field behave like such particles.
Version of April 11, 2023 171

Field angular momentum

In analogy with the field momentum one can find the expression for the operator of the
field angular momentum by considering infinitesimal rotations r → r + δϕ n × r with δϕ -
angle of rotation and n - unit vector along the rotation axis (with the usual ”right hand
rule” convention). As with the momentum we should look for the operator Lop for which
   
ψ̂(r) ψ̂(r + δϕn × r)
e−iδϕn·Lop /ℏ eiδϕn·Lop /ℏ = (4.44)
ψ̂ + (r) ψ̂ + (r + δϕn × r)

For infinitesimal δϕ it is straightforward to conclude that Lop must satisfy

      
ψ̂(r) iℏ(n × r) · ∇ψ̂(r) iℏn · (r × ∇)ψ̂(r)
n · Lop , = =
ψ̂ + (r) iℏ(n × r) · ∇ψ̂ + (r) iℏn · (r × ∇)ψ̂ + (r)

This is achieved with the expression

Z
Lop = d3 r′ ψ̂ + (r′ )[r′ × (−iℏ∇r′ )]ψ̂(r′ ) (4.45)

Indeed

ψ̂ + (r′ )[r′ × (−iℏ∇r′ )]ψ̂(r′ ) , ψ̂(r)

   Z  
ψ̂(r) 3 ′
Lop , = d r =
ψ̂ + (r) ψ̂ + (r′ )[r′ × (−iℏ∇r′ )]ψ̂(r′ ) , ψ̂ + (r)
δ(r − r′ )iℏ(r′ × ∇r′ )ψ̂(r′ )
Z    
iℏ(r × ∇r )ψ̂(r)
= d3 r′ = (4.46)
ψ̂ + (r′ )[−iℏ(r′ × ∇r′ )]δ(r − r′ ) iℏ(r × ∇r )ψ̂ + (r)

where in the second line of the last equality we used integration by parts.
The angular momentum operator commutes with the free field Hamiltonian, Eq. (4.22)

[Hop , Lop ] = 0 (4.47)

The reader is advised to carry out this calculation the result of which essentially follows
from the commutativity of the ”first quantized” h0 = −ℏ2 ∇2 /2m and l = r × (−iℏ∇)
which enter the expressions of these operators. Physically of course it reflects the invari-
ance of the free field Hop under rotations. Another useful calculation for the reader to
work out is to verify the validity of the standard commutation relations for the components
of Lop X
[Lop,i , Lop,j ] = iℏ ϵijn Lop,n (4.48)
n

Here again the corresponding commutators of li = [r × (−iℏ∇)]i ’s which enter the expres-
sions of Lop,i ’s are the ”cause” of this result.
Following the experience of transforming the field momentum operator Pop to the
plane wave basis it is instructive to consider transforming the field operators in the field
angular momentum operator Lop , Eq. (4.45) to the spherical wave normal modes basis
uklm (r) as given in Eq. (4.36). In contrast to Pop the non commutativity of different
Version of April 11, 2023 172

components of Lop leads to different forms of the expressions for different Lop,j ’s. The
simplest is for Lop,z X
Lop,z = ℏm â+
klm âklm
klm

The expressions for Lop,x and Lop,y will contain non diagonal m → m ± 1 terms. We leave
for the reader to work this out explicitly.

4.3 Adding external potential

So the conclusions at this stage are that the quantized free Schrödinger field describes a
collection of quanta which behave like free moving non interacting non relativistic quantum
particles. We also note that these particles are identical (see longer discussion of this
aspect in the following sections). It is therefore natural to ask how to include interactions
of the particles and how to account for their statistics?

4.3.1 The Hamiltonian

We begin by considering the Schrödinger field in the presence of an external potential.
The field equation is the familiar
ℏ2 2
 
∂ψ(r, t)
iℏ = − ∇ + U (r) ψ(r, t) ≡ hψ(r, t) (4.49)
∂t 2m
with h defined as
ℏ2 2
h=− ∇ + U (r) (4.50)
2m
Following what we did in the case of the free field, cf., Eq.(4.2) we consider this
equation and its complex conjugate as the pair of Hamilton equations with ψ(r) and
iℏψ ∗ (r) as canonical variables and the following Hamiltonian
Z  2 
ℏ
H = d3 r |∇ψ(r)|2 + U (r)|ψ(r)|2 (4.51)
2m
Indeed from
ℏ2
Z 
3
δH = d r (∇ψ ∗ (r)∇δψ(r) + ∇δψ ∗ (r)∇ψ(r)) +
2m
+U (r) (ψ ∗ (r)δψ(r) + δψ ∗ (r)ψ(r))]

we find that Hamilton equations for iℏψ ∗ (r) and ψ(r)

ℏ2 2
 
∂ψ(r) δH 1
= = − ∇ ψ(r) + U (r)ψ(r)
∂t δ[iℏψ ∗ (r)] iℏ 2m
∂[iℏψ ∗ (r)] ℏ2 2 ∗
 
δH
=− = − − ∇ ψ (r) + U (r)ψ ∗ (r)
∂t δψ(r) 2m
reproduce correctly the field equation (4.49) and its complex conjugate.
Version of April 11, 2023 173

On this basis we will quantize this field following the by now familiar pattern

ψ(r) → ψ̂(r) , ψ ∗ (r) → ψ̂ + (r) (4.52)

with commutation relations (4.19) and the Hamiltonian operator

Z  2 
ℏ
Hop = d3 r ∇ψ̂ + (r) · ∇ψ̂(r) + U (r)ψ̂ + (r)ψ̂(r)
2m

or in an equivalent form (cf., the remark after Eq. (4.22))

ℏ2 2
Z   Z
Hop = d3 r ψ̂ + (r) − ∇ + U (r) ψ̂(r) ≡ d3 r ψ̂ + (r)hψ̂(r) (4.53)
2m

with h defined above in Eq. (4.50). As in the case of the free field the general goal of
the theory is to solve the Schrödinger equation (4.35) but with the Hamiltonian operator
given by (4.53). As always the general method of doing this is to find the eigenfunctions
of this operator.
Before this let us note that the Heisenberg equation for the field operators calculated
with the Hamiltonian Hop , Eq. (4.53) coincides in form (as they should) with the wave
equation (4.49) which we have quantized

∂ ψ̂(r, t) h i
iℏ = ψ̂(r, t), Hop = hψ̂(r, t)
∂t
with h defined in Eq. (4.50). The calculation of the commutator in this equation can be
efficiently done by commuting the operator ψ̂(r, t) through the elements of the expression
R 3 ′ + ′
d r ψ̂ (r , t)hψ̂(r′ , t) for Hop . Since the only non zero commutator of ψ̂(r, t) is with
+ ′
ψ̂ (r , t) we get
h i Z
ψ̂(r, t), Hop = d3 r′ δ(r − r′ )hψ̂(r′ , t) = hψ̂(r, t)

The Heisenberg equation for ψ̂ + (r, t) coincides in form with complex conjugate of Eq. (4.49).

4.3.2 The eigenstates. Field quanta are particles in the external

potential
To find the eigenfunctions of the above Hamiltonian

Hop Ψ = EΨ

we use the experience with the free field and look for the basis ui (r) to expand the
field operators ψ̂(r) in which Hop will become a sum of decoupled commuting terms like
Eq. (4.29) for the free field Hamiltonian. Before doing this let us briefly consider the
general aspects of changing basis.
Version of April 11, 2023 174

Changing basis
The transformation from ψ̂(r) and ψ̂ + (r) to âk and â+ can be viewed as a particular
k
example of a more general operator transformation
X X
ψ̂(r) = âi ui (r) , ψ̂ + (r) = â+ ∗
i ui (r) (4.54)
i i

with {ui (r)} - any complete orthonormal basis, i.e. set of functions which obey
Z
d3 r u∗i (r)uj (r) = δij orthonormality (4.55)
X
ui (r)u∗i (r′ ) = δ(r − r′ ) completeness
i

Inverting the transformation

Z Z
âi = d3 r ψ̂(r)u∗i (r) , â+
i = d3 rψ̂ + (r)ui (r) (4.56)

and using the commutation relations (4.19) for the field operators ψ̂(r) and ψ̂ + (r) and
orthogonality of the basis set {ui (r)} one finds that the commutations of the âi , â+
i set
remain canonical
[âi , â+ + +
j ] = δij , [âi , âj ] = 0 = [âi , âj ] (4.57)
Let us note a useful view of the expansion (4.54) as transforming ”vectors” of operators
from one basis to another. E.g. a vector ψ̂r (i.e the set {ψ̂r }with r regarded as an index)
in the operator valued Hilbert space of functions of r gets transformed to the vector {âi }
in this space with the use of the transformation matrix {ur,i } (with r in ui (r) regarded as
index). The orthogonality and completeness relations (4.55) of the set {ui (r)} are just the
expressions of the unitarity of the matrix {ur,i }. In Appendix we review the properties
of the operators âj and â+
i for a general basis set {ui (r)} and the quantum states which
they generate.
Using the expansions Eq. (4.54) in the expression Eq. (4.53) for Hop , we obtain
X
Hop = hij â+
i âj (4.58)
ij

where Z
hij ≡ d3 r u∗i (r)huj (r)

are matrix elements of h in the basis ui (r).

The normal modes

We now choose ui (r)’s to be solutions of

hui (r) = ϵi ui (r) (4.59)

Version of April 11, 2023 175

These solutions are obviously the normal modes of the field described by the linear equa-
tion (4.49). Indeed in a trivial way the field configurations ψ(r, t) = ui (r) exp(−iϵi t/ℏ)
solve the (classical) field equation (4.49), i.e. in each of these configurations all the field
degrees of freedom (indexed by r) oscillate with the same frequency ϵi /ℏ.
We note that in the non interacting limit U (r) = 0 the operator h reduces to

ℏ2 2
h0 = − ∇
2m
and ui (r)’s become the plane waves uk (r), Eq.(4.26).
It is important to observe that h0 and h appear as operators acting on functions of
r. As such they are very different from the operator Hop which acts on the states of the
field ψ̂, like e.g. the states Eq. (4.34). As was already noted the field operators are on
the one hand operators in the space of the states of the field (and in this role r is just an
index labelling these operators) and on the other hand they are functions of r on which
the operator h acts. Perhaps a helpful analogy is the quantized EM field in which the
components of Eop (r) and Bop (r) are both operators and functions of r. In the present
context for reasons which will become clear in the sections below operators like h0 and h
will often be called single particle operators and the bases of functions like uk (r) or ui (r)
- single particle bases.
In the basis of the eigenstates of h we have

hij = ϵi δij

so that as in the free field case Hop is a sum of independent (commuting) oscillators
X
Hop = ϵi â+
i âi (4.60)
i

corresponding to the ”vibrations” of amplitudes of the normal modes Eq. (4.59). The
eigenfunctions of Hop are products of eigenstates |ni ⟩ of these field oscillators, i.e eigen-
states of the operators
n̂i = â+
i âi (4.61)
cf., Appendix, Eq. (4.159) while the eigenenergies are the corresponding sums
Y Y (â+ )ni X
|Ψ{ni } ⟩ ≡ |n1 , n2 , . . . , ni , . . . ⟩ = |ni ⟩ = √i |0⟩ , E{ni } = ϵi ni (4.62)
i i
ni ! i

To conclude, the quantization of the Schrödinger field in the presence of an external

potential, Eq.(4.49) describes collections of independent quanta of the normal modes
given by the solutions of the equation (4.59). Since this equation is just a Schrödinger
equation for a single particle in the potential U (r) we therefore obtained a description of
systems of such particles in this potential occupying its eigenstates ui (r).

The particle number operator. U (1) symmetry

We note that the operators n̂i , Eq. (4.61) ”count” the number of particles ni in each single
particle state ui (r). We had similar operators n̂k in the free field case, cf., Eq. (4.29). It
Version of April 11, 2023 176

is useful and important to introduce the total number of particles operator

X
Nop = n̂i (4.63)
i

which ”measures” the sum of all ni ’s

X
Nop |Ψ{ni } ⟩ = Nop |n1 , n2 , . . . , ni , . . . ⟩ = N |n1 , n2 , . . . , ni , . . . ⟩ with N = ni (4.64)
i

This operator has the same form in any complete orthonormal basis
X X Z
+ +
Nop = âi âi = â âk = d3 r ψ̂ + (r)ψ̂(r) (4.65)
k
i k
as can be verified by inserting the expansions (4.54) with different sets ui (r) in the last
integral.
The result that the eigenfunctions (4.62) of the Hamiltonian (4.60) are also eigenfunc-
tions of the number operator (4.63) is linked to the fact that Nop commutes with the
Hamiltonian
[Hop , Nop ] = 0 (4.66)
so that the particle number is a conserved quantum number in this theory, not a fixed
quantity prescribed from ”outside”.
Let us note that Hop commutes with the individual mode number operators n̂i ,
Eq. (4.61). This however is only for the eigenmodes of the field, i.e. for the single particle
states Eq. (4.59). The conservation of Nop is a much more general property independent
of the basis, cf. Eq. (4.65). It is intuitively related to the manner in which the operators
âj and â+
i enter the general Hamiltonian Eq. (4.58) and can be traced to the way the field
Hamiltonian Eq. (4.53) contains the field operators ψ̂ + and ψ̂. Formally this is reflected
in the invariance of the expression (4.53) under a global (coordinate independent) phase
transformation
ψ̂(r) → eiα ψ̂(r) , ψ̂ + (r) → e−iα ψ̂ + (r) (4.67)
In a more general context such a transformation is called a global U (1) gauge transfor-
mation and the operator Nop is its generator. This means that

e−iαNop ψ̂(r)eiαNop = eiα ψ̂(r) , e−iαNop ψ̂ + (r)eiαNop = e−iα ψ̂ + (r) (4.68)

As usual to prove this it is sufficient to consider an infinitesimal α. It is enough to do this

for ψ̂(r) since the relation for ψ̂ + (r) is just the hermitian conjugate. We have

(1 − iαNop )ψ̂(r)(1 + iαNop ) = (1 + iα)ψ̂(r) → [Nop , ψ̂(r)] = −ψ̂(r) (4.69)

Simple calculation supplies the proof

Z Z
[Nop , ψ̂(r)] = d3 r′ [ψ̂ + (r′ )ψ̂(r′ ), ψ̂(r)] = − d3 r′ δ(r′ − r)ψ̂(r′ ) = −ψ̂(r)
Version of April 11, 2023 177

Using Eq. (4.68), the invariance of Hop under (4.67) and denoting

Uop (α) ≡ e−iαNop

one has
Z 
+
Uop (α)Hop Uop (α) = Uop (α) d3 r ψ̂ + (r)hψ̂(r) Uop
+
(α) =
Z
= d3 r Uop (α)ψ̂ + (r)Uop+ +
(α) h Uop (α)ψ̂(r)Uop (α) = Hop

For infinitesimal α
+
Uop (α)Hop Uop (α) → (1 − iαNop )Hop (1 + iαNop ) = (Hop − iα[Nop , Hop ]) (4.70)

and to have it equal to Hop must have Eq. (4.66).

Going back to the eigenfunctions and eigenvalues of Hop we note that the general
solution of the Schrödinger equation (4.35) with this Hop is a familiar linear combination
X
|Ψ(t)⟩ = C{ni } |Ψ{ni } ⟩e−E{ni } t/ℏ (4.71)
{ni }

with (as always) the coefficients C{ni } determined by the initial condition for |Ψ(t)⟩ at
t = 0. And we note that the formalism in principle
P allows to have states with coherent
combinations of different particle numbers N = i ni . The choice to have a fixed N , i.e.
to have it the same for all components in the above solution is in the freedom of setting
the appropriate initial condition supported (conserved in time) by the commutativity of
Hop with Nop .

4.3.3 Working with the field operators

The last equality in the expressions (4.65) for Nop in terms of the field operators represents
Nop as a sum (integral) over particle number operators dn̂(r) = ψ̂ + (r)ψ̂(r)d3 r in the
infinitesimal volume d3 r situated at r. This suggest that

ρ̂(r) = ψ̂ + (r)ψ̂(r) (4.72)

is the particle density operator. This also explains what is the physical meaning of the
field operators ψ̂ + (r). Indeed let us consider a state

|r′ ⟩ ≡ const ψ̂ + (r′ )|0⟩ (4.73)

where we introduced a multiplicative constant for normalization, see below. Let us act
on this state with the operator ρ̂(r)

ρ̂(r)|r′ ⟩ = const ψ̂ + (r)ψ̂(r)ψ̂ + (r′ )|0⟩ = const δ(r − r′ )ψ̂ + (r′ )|0⟩ = δ(r − r′ )|r′ ⟩ (4.74)

where we commuted ψ̂(r) with ψ̂ + (r′ )to its right and then used ψ̂(r)|0⟩ = 0. The result
shows that ψ̂ + (r) acting on the vacuum state creates a particle at the position r. More
precisely it creates delta like particle density at this position.
Version of April 11, 2023 178

What happens if several ψ̂ + ’s act on the vacuum? E.g. consider the state
|r1 , . . . , rN ⟩ = constN ψ̂ + (r1 ) . . . ψ̂ + (rN )|0⟩ (4.75)
Let us act on this state with ρ̂(r). As in the one particle case we find the result by first
commuting ψ̂(r) through ψ̂ + (ra )’s to its right all the way to the vacuum. This calculation
will appear in several places below so we show it in details
N
Y N
Y
+ +
ψ̂(r) ψ̂ (ra )|0⟩ = [δ(r − r1 ) + ψ̂ (r1 )ψ̂(r)] ψ̂ + (ra )|0⟩ =
a=1 a̸=1
N
Y N
Y
= δ(r − r1 ) ψ̂ + (ra ) + ψ̂ + (r1 )[δ(r − r2 ) + ψ̂ + (r2 )ψ̂(r)] ψ̂ + (ra )]|0⟩ =
a̸=1 a=3
N
Y N
Y
= δ(r − r1 ) ψ̂ + (ra ) + δ(r − r2 ) ψ̂ + (ra ) + · · · +
a̸=1 a̸=2
N
Y −1
+ ψ̂ + (ra )[δ(r − rN ) + ψ̂ + (rN )ψ̂(r)]|0⟩ =
a=1
 
XN N
Y
= δ(r − rb ) ψ̂ + (ra ) |0⟩ (4.76)
b=1 a̸=b

Acting on this with ψ̂ + (r) and using the delta function in each term to replace r → rb in
it we get
N
X
ρ̂(r)|r1 , . . . , rN ⟩ = [ δ(r − ra )]|r1 , . . . , rN ⟩
a=1

i.e. have N particles (delta like particle densities) at the positions ra , a = 1, . . . , N . In the
same manner one can show that ψ̂ + (r) creates a particle at r when it acts on any general
state (discussed below). We also note that the result (4.76) shows that ψ̂(r) destroys
(annihilates) a particle if its coordinates coincide with r.
It is important to notice that ψ̂ + and ψ̂ create and annihilate particles only when they
act to the right. Acting to the left they produce an opposite result - they correspond-
ingly annihilate and create particles. For example the state ⟨r1 , . . . , rN | is the hermitian
conjugate of |r1 , . . . , rN ⟩ so

⟨r1 , . . . , rN | = [constN ψ̂ + (r1 ) . . . ψ̂ + (rN )|0⟩]+ = ⟨0|ψ̂(rN ) . . . ψ̂(r1 )(constN )∗ (4.77)

since [ψ̂ + ]+ = ψ̂. Thus the state ⟨r1 , . . . , rN | is the result of acting with N ψ̂’s to the left
on the vacuum ⟨0|.
What is the norm of |r1 , . . . , rN ⟩? Take as an example one particle state Eq. (4.73)
and calculate
⟨r′ |r⟩ = |const|2 ⟨0|ψ̂(r′ )ψ̂ + (r)|0⟩ = |const|2 δ(r − r′ )
The result shows that such a state is non normalizable. This should not be surprising
as one has a continuum of states labeled by r. Just as with more familiar momentum
Version of April 11, 2023 179

states labeled by p. Also the momentum states are non normalizable. The common
regularization is to make p discrete by introducing very large but finite volume, i.e. to
introduce an infrared cutoff. In the same way one can make r discrete by introducing a
lattice of discrete r’s. If this is not done - then one can normalize as convenient.
As we will see in the next section the most common use of the states |r⟩ or their N
particle generalization |r1 , . . . , rN ⟩, Eq. (4.75), makes it convenient to choose the normal-
ization of these states as
1
constN = √
N!

4.4 Wave functions. Operators. Comparison with the

first quantized description
The quantization of the Schrödinger field is (for obvious reasons) called the second quanti-
zation. For the field governed by Eq.(4.49) this seems to result in an alternative description
of quantum non interacting particles in the external potential U (r).
Here we want to understand if this description is indeed complete and how it is related
to the standard quantum mechanical description of say N particles with the wave function
Φ(r1 , r2 , . . . , rN , t) obeying the N particle Schrödinger equation
N 
ℏ2 2

∂Φ(r1 , ..., rN , t) X
iℏ = − ∇ + U (ra ) Φ(ra , ..., rN , t) (4.78)
∂t a=1
2m a

4.4.1 Wave functions in the second quantization

Coordinate representation. Second vs first quantization
The states |r1 , . . . , rN ⟩ introduced in the previous section, cf., Eq. (4.75), form a very
convenient basis to represent a general N particles wave function in the second quantiza-
tion
Z YN
|Φ⟩ = d3 ra Φ(r1 , r2 , . . . , rN )|r1 , . . . , rN ⟩ (4.79)
a=1
The interpretation of this expression is quite clear - we have a linear combination of N
particles in different coordinate positions r1 , . . . , rN weighted each with the probability
amplitude Φ(r1 , r2 , . . . , rN ). These amplitudes form the wave function |Φ⟩ in the coordi-
nate representation and clearly are equivalent to this wave function in the first quantiza-
tion formalism. We will see this equivalence even more explicitly in the discussions below
of how physical operators of particle observables act on |Φ⟩.
As discussed in the Appendix in order to have both |Φ⟩ and Φ(r1 , . . . , rN ) normalized
to unity, i.e. to have
Z
d3 r1 . . . d3 rN |Φ(r1 , r2 , . . . , rN )|2 = 1 and ⟨Φ|Φ⟩ = 1 (4.80)
√
one must choose the normalization constN = 1/ N ! in the definition (4.75) of the states
|r1 , . . . , rN ⟩ as they appear in the relation (4.79) between |Φ⟩ and Φ(r1 , r2 , . . . , rN ). We
Version of April 11, 2023 180

thus have
N
1
Z Y
|Φ⟩ = √ d3 ra Φ(r1 , r2 , . . . , rN )ψ̂ + (r1 )ψ̂ + (r2 ) . . . ψ̂ + (rN )|0⟩ (4.81)
N! a=1

Permutation symmetry
The commutativity properties of the field operators ψ̂ + (r) imply that the coordinate prob-
ability amplitudes Φ(r1 , r2 , . . . , rN ) in Eq. (4.81) can not be arbitrary. These functions
must be symmetric under all possible permutations of the particles’ coordinates.
Let us demonstrate this for the simplest case of two particles
1
Z
|Φ⟩ ≡ √ d3 r1 d3 r2 Φ(r1 , r2 )ψ̂ + (r1 )ψ̂ + (r2 )|0⟩ (4.82)
2
Functions of two variables can belong to one of the two symmetry representations - sym-
metric or antisymmetric,

ΦS (r1 , r2 ) = ΦS (r2 , r1 ) and ΦA (r1 , r2 ) = −ΦA (r2 , r1 )

and in general have

1 1
Φ(r1 , r2 ) = [Φ(r1 , r2 ) + Φ(r2 , r1 )] + [Φ(r1 , r2 ) − Φ(r2 , r1 )] ≡ ΦS (r1 , r2 ) + ΦA (r1 , r2 )
2 2
It is straightforward to show that |Φ⟩A obtained with ΦA (r1 , r2 ) in Eq. (4.82) vanishes
identically. Have
1
Z
|Φ⟩A = √ d3 r1 d3 r2 ΦA (r1 , r2 )ψ̂ + (r1 )ψ̂ + (r2 )|0⟩ =
2
1
Z
= −√ d3 r1 d3 r2 ΦA (r2 , r1 )ψ̂ + (r1 )ψ̂ + (r2 )|0⟩ =
2
1
Z
= −√ d3 r1 d3 r2 ΦA (r2 , r1 )ψ̂ + (r2 )ψ̂ + (r1 )|0⟩ =
2
1
Z
= −√ d3 r1 d3 r2 ΦA (r1 , r2 )ψ̂ + (r1 )ψ̂ + (r2 )|0⟩ = −|Φ⟩A
2

where in the 3rd line we commuted ψ̂ + (r1 )ψ̂ + (r2 ) = ψ̂ + (r2 )ψ̂ + (r1 ) and in the 4th line have
interchanged the integration variables r1 ↔ r2 . So we have proved that |Φ⟩A = −|Φ⟩A
which means that |Φ⟩A = 0.
The same proof obviously holds for any pair of coordinates in a general wave function
Φ(r1 , . . . , rN ). Thus only Φ(r1 , . . . , rN )’s which are symmetric with respect to permu-
tation of any two particles produce non zero result in Eq. (4.81). This means that this
is true also for Φ(r1 , . . . , rN )’s which are symmetric under permutations of any number
of particles. Indeed (as is simple to understand4 and can be proved by induction) any
4 Cf, Messiah, Quantum Mechanics (Dover Books in Physics), Ch. XIV. Denote for example by

(1532476) a permutation 1 → 5 → 3 → 2 → 4 → 7 → 6 → 1. It can clearly be written as an or-

dered product (15)(53)(32)(24)(47)(76) of transpositions (with right to left order)
Version of April 11, 2023 181

such permutation can be decomposed into a product of permutations of two particles

(transpositions).
The above symmetry under permutations of the wave functions is one of the most
important features of the second quantization formalism. Together with the symmetry of
the physical observables as represented by the operators as discussed below this property
means that the quanta of the theory are bosons, i.e. identical particles obeying Bose
statistics. We will provide more details to this discussion in Section 4.4.3.

Occupation number representation

Expanding the field operators in Eq. (4.81) in an arbitrary complete and orthonormal
single particle basis, cf., Eq. (4.54), we obtain
X
|Φ⟩ = Ci1 ,...,iN â+ +
i1 ...âiN |0⟩ (4.83)
i1 ,...,iN

with
N
1
Z Y
Ci1 ,...,iN = √ d3 ra Φ(r1 , r2 , . . . , rN )u∗i1 (r1 ) . . . u∗iN (rN )
N! a=1

The coefficients Ci1 ,...,iN represent the function |Φ⟩ in the basis of products of the single
particle states ui (r). As we discussed in the previous Section the functions Φ(r1 , r2 , . . . , rN )
are symmetric with respect to permutations of the particle coordinates ra ’s. One can use
this to replace the products of ui (r)’s in the above expression for Ci1 ,...,iN ’s by symmetrized
products X
u∗i1 (r1 ) . . . u∗iN (rN ) → const u∗i1 (rp1 ) . . . u∗iN (rpN )
P
with P denoting the permutations of particle coordinates r1 , ..., rN → rp1 , ..., rpN , the
normalization constant p
const = n1 !...nN !/N !
and appropriate adjustment of the expansion constants Ci1 ,...,iN . It is a useful exercise
to work this out starting with the simple N = 2 case, writing u∗i1 (r1 )u∗i2 (r2 ) as a sum
of symmetric and antisymmetric products, with the antisymmetric part vanishing in the
integral of its product and the symmetric Φ(r1 , r2 ).
It is useful and conventional to write the expansion (4.83) using the notation of
Eq. (4.62) with the occupation numbers ni of the single particle states. In this repre-
sentation the state |Φ⟩ will be written as
X
|Φ⟩ = Cn1 ,...,ni ,... |n1 , n1 , ..., ni , ...⟩ (4.84)
P
n1 ,...,ni ,...;with i ni =N

with appropriate adjustment of the coefficients Cn1 ,...,ni ,... . Such representation of the N
particles wave functions is called occupation number representation. It emphasises
the fact that we are dealing with identical quanta (particles) so that all one needs is
their numbers ni in each single particle state. Note that in this representation one must
3
”supply” infinite
P (actually ∞ ) set of (positive) integers ni . But since they are subject to
the constraint i = N only ≤ N of them are not zero.
Version of April 11, 2023 182

4.4.2 Operators in the second quantization

In this section we want to establish how the operators of the physical observables act on
wave functions in the second quantization formalism. In this way we will also understand
much better the connection with the first quantization.

The one body Hamiltonian

We will start with the discussion of the action on |Φ⟩ by the Hamiltonian (4.53). Let us
write it as a sum of two terms - kinetic and potential

Hop = Kop + Uop (4.85)

ℏ2 2
Z   Z
3 +
Kop = d r ψ̂ (r) − ∇ ψ̂(r) , Uop = d3 rψ̂ + (r)U (r)ψ̂(r)
2m

and let us consider first the action on |Φ⟩ of the potential part
N N
1
Z Y Z Y
3
Uop |Φ⟩ = √ d ra Φ(r1 , r2 , . . . , rN ) d3 rψ̂ + (r)U (r)ψ̂(r) ψ̂ + (ra )|0⟩ (4.86)
N! a=1 a=1

Using the result Eq. (4.76), multiplying it by U (r) and ψ̂ + (r) and doing the d3 r integral
with the help of the δ-functions we get
N
Z "N # N
Y X Y
d3 rψ̂ + (r)U (r)ψ̂(r) ψ̂ + (ra )|0⟩ = U (rb ) ψ̂ + (ra )|0⟩ (4.87)
a=1 b=1 a=1

so that
N N
1
Z Y Y
Uop |Φ⟩ = √ d3 ra Φ′ (r1 , r2 , . . . , rN ) ψ̂ + (ra )|0⟩ (4.88)
N! a=1 a=1

with
N
" #
X
Φ′ (r1 , r2 , . . . , rN ) = U (ra ) Φ(r1 , r2 , . . . , rN ) (4.89)
a=1

We see that the action of thePsecond quantized operator Uop on |Φ⟩ is equivalent to the
action of the first quantized a U (ra ) on Φ(r1 , . . . , rN ) i.e. on the first quantized partner
of |Φ⟩.
To calculate the action of Kop on |Φ⟩ is a bit more involved but straightforward. The
details are given in the Appendix 4.7.3 with the result
N N
1
Z Y Y
Kop |Φ⟩ = √ d3 ra Φ′ (r1 , r2 , . . . , rN ) ψ̂ + (ra )|0⟩ (4.90)
N! a=1 a

with
N 
" #
X ℏ2 2
Φ′ (r1 , r2 , . . . , rN ) = − ∇ Φ(r1 , r2 , . . . , rN ) (4.91)
2m rb
b=1
Version of April 11, 2023 183

As in the Uop case we see that the action of Kop on |Φ⟩ is equivalent to the action of the
first quantized kinetic energy operator
N 
ℏ2 2
X 
− ∇
a=1
2m rb

on the first quantized partner Φ(r1 , r2 , . . . , rN ) of |Φ⟩

Combining these results we find that
N
"N # N
1
Z Y X Y
3
Hop |Φ⟩ = (Kop + Uop )|Φ⟩ = √ d ra ha Φ(r1 , r2 , . . . , rN ) ψ̂ + (ra )|0⟩
N ! a=1 a=1 a
(4.92)
with the single particle hamiltonian h given by Eq. (4.50)

Other one body operators

The operators Kop , Uop and Hop discussed above are all of the type which in the first
quantization formulation have the form
N
X
(1)
F̂ = fa(1) (4.93)
a=1

(1)
with each fa being a function of ra and pa = −iℏ∇a . Such operators act on wave
functions of N particles but at one particle at a time. They are called one-body operators
and the subscript which we attached to F̂ (1) and f (1) serves to make this distinction.
On the basis of our above discussion of the operators Kop , Uop and Hop we can make
a general statement that in the second quantization one body operators have the form
Z
Fop = d3 rψ̂ + (r)f (1) ψ̂(r)
(1)
(4.94)

where f (1) in the last expression is one (any) of the operators in the sum (4.93) and it is
acting on ψ̂(r) as a function of r. E.g. angular momentum
N
X N
X
L= la ≡ ra × (−iℏ∇ra ) (4.95)
a=1 a=1

becomes Z
Lop = d3 rψ̂ + (r)[r × (−iℏ∇r )]ψ̂(r) (4.96)

in the second quantization formalism.

It is important to observe that the particle number N which appears in the operators
in the 1st quantization Eq. (4.93) is a part of their definition while the corresponding
operators in the second quantization do not contain any information about N . It is
the wave functions on which these operators act, like |Ψ⟩ in the previous section which
Version of April 11, 2023 184

(1)
depend on N . The second quantization Fop ’s ”are ready to act” on |Ψ⟩ with any value of
N including a linear combination with different N ’s (see the section below on the general
Fock space). At the same time these particular type of operators do not change N since
they contain an equal number of creation and annihilation operators - one of each type.
But nothing intrinsically in the formalism prevents having operators which change N . In
fact the elementary ones ψ̂(r) and ψ̂ + (r) do just that.
(1)
Formally the conserving N property of the operators Fop is expressed by their com-
R 3 +
mutativity with the particle number operator Nop = d rψ̂ (r)ψ̂(r) ,
(1)
[Fop , Nop ] = 0 (4.97)

which as in the case of Hop , Eq. (4.66) follows from the invariance of Eq. (4.94) with
respect to the global U (1) transformation Eq. (4.67).
From our derivations in the previous section it should also be clear in details the
”mechanics” of how the one body second quantized operators act on functions like |Ψ⟩.
Pictorially one can say that first the destruction operator ψ̂(r) acts on |Ψ⟩ ”seeking out”
all the particles at their position ra , a = 1, . . . , N and ”annihilating” them one at a time.
The result depends on the coordinates of the particle positions. The operator fˆ then
acts on these coordinates and then the operator ψ̂ + (r) puts the particles back (”creates”
them) where they originally were. All this gets weighted with the probability amplitude
Φ(r1 , r2 , . . . , rN ) and integrated over all possible ra ’s.

Particle interactions. Two body operators

Let us now understand how do we write in the second quantization the operators which
represent interactions between particles. The most common such operators are potential
energy which is a sum of all pairwise interactions (e.g. Coulomb interaction). Their form
in the first quantization is
N
1 X
V = V (ra − rb ) (4.98)
2
a,b=1,a̸=b

As we see this is a sum of operators with each acting on two particles at a time. Such
operators are called two body operators.
Based on the experience of the previous sections it is not difficult to guess that the
following corresponding expression holds in the second quantization
1
Z
Vop = d3 rd3 r′ ψ̂ + (r)ψ̂ + (r′ )V (r − r′ )ψ̂(r′ )ψ̂(r) (4.99)
2
To verify this guess let us do what we did with one body operators - let us act with this
expression on the general N particle wave function in Eq. (4.81).
N
1
Z Y
Vop |Φ⟩ = √ d3 ra Φ(r1 , r2 , . . . , rN ) × (4.100)
N ! a=1
N
1
Z Y
× d3 rd3 r′ ψ̂ + (r)ψ̂ + (r′ )V (r − r′ )ψ̂(r′ )ψ̂(r) ψ̂ + (ra )|0⟩
2 a=1
Version of April 11, 2023 185

To evaluate this we use the relation Eq. (4.76), act on it with ψ̂(r′ ) and obtain
 
N
Y XN N
Y
ψ̂(r′ )ψ̂(r) ψ̂ + (ra )|0⟩ =  δ(r′ − rc )δ(r − rb ) ψ̂ + (ra ) |0⟩
a=1 b̸=c a̸=b,c

Using this we find

N
1
Z Y
d3 rd3 r′ ψ̂ + (r)ψ̂ + (r′ )V (r − r′ )ψ̂(r′ )ψ̂(r) ψ̂ + (ra )|0⟩ =
2 a=1
 
N N
1
Z X Y
d3 rd3 r′ ψ̂ + (r)ψ̂ + (r′ )V (r − r′ )  δ(r′ − rc )δ(r − rb ) ψ̂ + (ra ) |0⟩ =
2
b,c=1,b̸=c a̸=b,c
N N
1 X Y
= V (rb − rc ) ψ̂ + (ra )|0⟩
2 a=1
b,c=1,b̸=c

and therefore
 
N N N
1 1
Z Y X Y
Vop |Φ⟩ = √ d3 ra  V (rb − rc ) Φ(r1 , r2 , . . . , rN ) ψ̂ + (ra )|0⟩
N! a=1
2 a=1
b,c=1,b̸=c
(4.101)
So indeed the action of Vop on |Φ⟩ is equivalent to/results in the action of the first
quantized V , Eq. (4.98) on Φ(r1 , r2 , . . . , rN )
The intuitive understanding of the expression (4.99) is similar to what we saw in the
one body operators case - the operators ψ̂(r′ )ψ̂(r) ”search” to annihilate two particles (as
r and r′ are integrated over) and thereby ”reveal” their position. The function V (r − r′ )
weighs the result while the operators ψ̂ + (r)ψ̂ + (r) put the particles back. All this is
integrated over all possible positions r and r′ .
The general form of the two body operator in the first quantization is
N
1 X (2)
F̂ (2) = fab (4.102)
2
a,b=1,a̸=b

with the second quantized counterpart

1
Z
(2)
Fop = d3 rd3 r′ ψ̂ + (r)ψ̂ + (r′ )f (2) ψ̂(r′ )ψ̂(r) (4.103)
2
with f (2) in general being a function of r, r′ and p̂ = −iℏ∇r , p̂′ = −iℏ∇r′ .
It is important to observe that also here as with the one body operators the second
quantized operators do not contain any information about the number N of the particles
which is encoded in the wave functions on which these operators act. Also here the
(2)
operators Fop do not change the value of N and commute with the particle number
operator
(2)
[Fop , Nop ] = 0 (4.104)
(2)
which is a consequence of the U (1) unitary symmetry Eq. (4.67) which the operators Fop
posses.
Version of April 11, 2023 186

Changing the single particle basis

The one and two body operators discussed above were expressed in terms of the basic field
operators ψ̂(r) and ψ̂ + (r). It is easy and instructive to express them using the expansion
(4.54) of these operators in a general single particle basis {ui (r)}. Inserting (4.54) into
the expressions (4.94) and (4.103) we obtain
X
(1)
Fop = ⟨i|f (1) |j⟩â+
i âj (4.105)
ij
X
(2)
Fop = ⟨ij|f (2) |kl⟩â+ +
i âj âl âk (4.106)
ijkl

where we used the notation for the matrix elements of elementary one and two body
operators
Z
(1)
⟨i|f |j⟩ = d3 ru∗i (r)f (1) uj (r) (4.107)
Z
⟨ij|f (2) |kl⟩ = d3 rd3 r′ u∗i (r)u∗j (r′ )f (2) uk (r)ul (r′ ) (4.108)

(1) (2)
We draw attention to the ”logic” of how the operators Fop and Fop in a general single
particle basis act on a wave function in this basis as written in the occupation number
(1)
representation of Eq. (4.84). In the one body Fop one starts with the operator âj an-
nihilating a particle in a (single particle basis) state uj reducing the corresponding nj
occupation to nj − 1. This is ”weighted” with a corresponding amplitude to find this nj
as encoded in the coefficients Cn1 ,...,nj ,... of the occupation number representation. The
following action of the operator â+
i creates (puts back) a particle in the state ui and the
result gets multiplied by the transition matrix element ⟨i|f (1) |j⟩. At the end one sums
(2)
over all such transitions. The two body Fop operates in a similar fashion but with two
particles annihilation and creation and the sum over all two particle transitions.

4.4.3 Second quantization via commutators describes identical

bosons
The following important features of the above formalism must be observed at this stage.
The first quantization operators which are counterparts of the operators in the second
quantization are always symmetric sums over all the particles or their pairs etc in the
wave functions on which they act.
The symmetry of these sums follows since all their terms are identical in acting on
different particles. They have the same functional dependence on the coordinates and
momenta with the same parameters - masses, charges, etc., e.g. same kinetic energy,
same external potential, same inter-particle interactions, etc. This is seen in the formal
correspondence Eq. (4.93) → Eq. (4.94) and Eq. (4.102) → Eq. (4.103) and in the explicit
examples in Eqs. (4.89, 4.91, 4.95, 4.98). There is no possibility to have second quantized
(1)
operators representing observables distinguishing a particular particle, say f5 or sets
Version of April 11, 2023 187

(1) (1)
of particular particles, e.g f7 + f15 . This is a general feature of quantum systems of
identical particles.
Let us also recall that as we discovered in Section 4.4.1 the first quantization wave
functions Φ(r1 , ..., rN ) which are counterparts of the second quantization |Φ⟩ are sym-
metric under the permutation of all the particles coordinates. This confirms the particle
being identical and moreover obeying the spin-statistics theorem requirements for systems
of bosons.
Let us remind that the spin-statistics theorem, proved by Pauli, states (in its first part)
that the wave functions of a system of identical integer-spin particles must be symmetric
under the exchange of the coordinates of any two particles. Such particles are bosons
obeying the Bose-Einstein statistics.
In our case of the identical particles without spin the symmetry requirement dictated
by the Pauli theorem is an additional rule which is imposed in the first quantization
formalism on selecting the wave function solutions of the Schrödinger equation (4.78).
As we have seen it is automatically fulfilled in the second quantization wave functions
Eqs. (4.79,4.81).
The second part of the spin-statistics theorem concerns wave functions of system of
identical half-integer spin particles. The theorem states that they must be anti-symmetric
under any pair of particle exchange. Such particles are fermions obeyong the Fermi-Dirac
statistics. We will discuss in the next Chapter how the second quantization allows for a
simple and straightforward modification to be extended to the descriptions of fermions.
Concerning the proof of the spin-statistics theorem - as Feynman states in his Lectures
on Physics: ”...An explanation has been worked out by Pauli from complicated arguments
of QFT and relativity...but we haven’t found a way of reproducing his arguments on an
elementary level...”5 .

4.4.4 Self interacting Schrödinger field.

Summing up. Interacting Hamiltonian
To summarize we learned how to translate the wave functions and physical operators
into the second quantization formalism. The N particle wave function of N bosons
Φ(r1 , ..., rN ) should be symmetric and becomes the amplitude of the |r1 , ..., rN ⟩ state
in the expression (4.79) or the more explicit (4.81).
Dealing with operators one should first determine to which type they belong - one
body, two body, etc. Examples of one body operators are momentum, density, current,
etc
N
X N
X N
X
P = pa = (−iℏ∇ra ) , ρ(r) = δ(r − ra )
a=1 a=1 a=1
N
1 X
j(r) = [δ(r − ra )(−iℏ∇ra ) + (−iℏ∇ra )δ(r − ra )]
2m a=1
5 cf.,I. Duck and E. Sudarshan, Towards an understanding of the spin-statistics theorem, Am. J.
Phys., 66 (4) 1998
Version of April 11, 2023 188

Note that identical particles imply that any such operator is a sum of identical operators
acting on each particle. So one takes one member of the sum and uses it in the expres-
sion (4.94) to find the corresponding 2nd quantized operator. If one prefers a general
basis {ui } rather than the coordinate representation of the field operators one needs to
calculate the matrix elements between all possible pairs of ui ’s and use them in the ex-
pression (4.105). The number N of particles appears explicitly in the operators of the
first quantized formalism but not in the 2nd quantization.
There are not too many examples of two body operators. Beside the two body inter-
action (4.98) there are various correlators like density-density or current-current, etc
N
X
ρop (r)ρop (r′ ) = δ(r − ra )δ(r′ − rb ) , etc
a,b=1

As in the one body case one must take one term in such a double sum and either use it in
the expression (4.103) with field operators or calculate all its two particle matrix elements
in a chosen basis of the single particle states ui ’s. One should then form an expression
(4.106) using these matrix elements.
As a rule it is extremely rare to find 3 body operator but it is straightforwardly clear
how to extend what we have learned to such cases.
Let us follow the above rules to write the full 2nd quantization Hamiltonian of a many
body interacting system. Consider its (most common) expression in the 1st quantization
N N
" #
X p̂2a 1 X
Ĥ = + U (ra ) + V (ra − rb ) , p̂a = −iℏ∇a (4.109)
a=1
2m 2
a,b=1,a̸=b

with externally fixed number N of the particles. Assuming that the particles are bosons
their particles statistics must be imposed ”by hand” allowing only symmetric wave func-
tions.
The 2nd quantized version of the above Hamiltonian is
1
Z Z
3 +
Hop = d rψ̂ (r) h ψ̂(r) + d3 rd3 r′ ψ̂ + (r)ψ̂ + (r′ )V (r − r′ )ψ̂(r′ )ψ̂(r) (4.110)
2
with
ℏ2 2
h=− ∇ + U (r)
2m
In a general single particle basis this Hamiltonian is written
X 1X
Hop = ⟨i| h |j⟩â+
i âj + ⟨ij|V |kl⟩â+ +
i âj âl âk (4.111)
ij
2
ijkl

If one knows the solutions of the one body part, i.e. knows the eigenfunctions ui (r) and
the eigenenergies ϵi of h, Eq. (4.59) one can ”incorporate” this knowledge in the above
expression for Hop . Using the set {ui } as the basis one has ⟨i| h |j⟩ = ϵi δij and
X 1X
Hop = ϵi â+
i âi + ⟨ij|V |kl⟩â+ +
i âj âl âk (4.112)
i
2
ijkl
Version of April 11, 2023 189

In the Mean Field Approximations chapter of this course we shall discuss and give exam-
ples of even more optimal ways to choose the single particle basis which incorporate on
the average the effect of the interaction term in many body systems.

Heisenberg equations. Classical limits - field vs particles

Let us consider the Heisenberg equations for the field operators ψ̂(r) and ψ̂ + (r) and the
general interacting Hamiltonian (4.110). We have
    
∂ ψ̂(r, t) ψ̂(r, t)
iℏ = , Hop
∂t ψ̂ + (r, t) ψ̂ + (r, t)

Straightforward calculations produce Hermitian conjugate equations

ℏ2 2
 
∂ ψ̂(r, t)
Z
iℏ = − ∇ + U (r) ψ̂(r, t) + V (r − r′ )ψ̂ + (r′ , t)ψ̂(r′ , t) d3 r′ ψ̂(r, t)
∂t 2m
(4.113)
∂ ψ̂ + (r, t) 2
  Z
ℏ
−iℏ = − ∇2 + U (r) ψ̂ + (r, t) + V (r − r′ )ψ̂ + (r′ , t)ψ̂(r′ , t) d3 r′ ψ̂ + (r, t)
∂t 2m

These equations find many uses in the theory of many-particle systems. The Green’s
functions method provides a good example6 .
Here we want to point out a simple but conceptually important aspect - their classical
limit. Like in other quantum systems this limit is intuitively obtained by replacing coor-
dinate and momentum operators by the corresponding classical functions of time turning
Heisenberg equations into classical Hamilton equations. In the above equations (4.113)
this means replacing ψ̂(r, t) and ψ̂ + (r, t) by the c-number (classical, commuting) functions
ψ(r, t) and ψ ∗ (r, t). In the non interacting limit V = 0 the resulting wave equations bring
us back to where we started, cf., the Schrödinger equation (4.49). The classical limit of
the fully interacting case leads to a non linear Schrödinger equation with cubic non linear
term controlled by the interaction.
Let us add two more remarks.
a) Our intuitive ”derivation” of the classical limit of Eq. (4.113) requires formal jus-
tification which will be discussed in the Chapter ”Mean Field Approximations for Many
Body Problems”. On the intuitive level the classical limit of the quantum field corresponds
to physical processes in which very large number of quanta (particles) are ”condensed” in
the same wave mode, i.e the same single particle state.
b) The classical limit referred to above is different from the common classical limit
for the N particle Hamiltonian Eq. (4.109). The latter is given by replacing the operators
for the particle coordinates and momenta by the classical variables in the correspond-
ing Heisenberg equations. It is easy to show that this results in the classical Hamilton
equations
dra ∂H dpa ∂H
= , =− (4.114)
dt ∂pa dt ∂ra
6 cf., Quantum Theory of Many-Particle Systems, A. Fetter and J. Walecka, Dover, 2003
Version of April 11, 2023 190

We therefore have two classical limits - the ”field” classical limit for the fields ψ(r, t) and
ψ ∗ (r, t) vs the more familiar ”particle” classical limit for the particle coordinates ra ’s and
momenta pa ’s. The latter classical limit is the limit of ℏ → 0 while the former is achieved
for the large number N0 ≫ 1 of condensed quanta (i.e. the boson particles) of the theory.

4.5 Fermions – another alternative of the second quan-

tization
As we have learned so far the quantization of the Schrödinger field leads to a very efficient
and elegant description of many particle bosonic systems in all their aspects. A natural
question is if this treatment can be extended to systems of fermions.

4.5.1 Quantization via anticommutators

A clear hint towards a positive answer can be found in our discussions in Section 4.4.1.
There we saw that the symmetry of the bosonic wave functions was assured by the most
basic property of the field operators ψ + (ra )’s creating the particles - their commutativ-
ity. As we will now show there is a consistent way of quantizing the Schrödinger field by
postulating anticommutativity of the basic operators. This single change of the quanti-
zation postulate will lead to a description of many fermion systems similar to the second
quantized formalism for many bosons.
Dealing with fermions one must introduce spin variable together with position coor-
dinates in order to describe the particles of the theory. Accordingly we start with the
classical field which is described by functions
ψσ (r) and ψσ∗ (r) (4.115)
with the spin projection index σ = ±1/2 (we assume spin 1/2 fermions as by far the most
common). It is often useful to write/view these functions in the explicit spinor form as
!
∗
ψ1/2 (r)
 
ψ1/2 (r)
and ∗
ψ−1/2 (r) ψ−1/2 (r)

We quantize this field by introducing two sets of operators

ψ̂σ (r) and ψ̂σ+ (r)
with r and σ = ±1/2 labelling each set. We need to define the space of states on which
these 2 × ∞3 operators act and the results of their action. We have seen with the bosonic
field ψ(r) treated above that to achieve this it was sufficient to define an abstract vacuum
state |0⟩ and the commutation relations between the field operators. Following this we
could define the basis of the space of states on which the operators act and calculate any
matrix element for any given operator.
Following this experience we start by defining the vacuum state |0⟩ with the properties
a) ⟨0|0⟩ = 1 (4.116)
b) ψ̂σ (r)|0⟩ = 0 for all values of r and σ
Version of April 11, 2023 191

This we supplement with imposing (posulating) the anticommutation relations as follows

ψ̂σ (r)ψ̂σ+′ (r′ ) + ψ̂σ+′ (r′ )ψ̂σ (r) ≡ {ψ̂σ (r), ψ̂σ+′ (r′ )} = δσσ′ δ(r − r′ )
ψ̂σ (r)ψ̂σ′ (r′ ) + ψ̂σ′ (r′ )ψ̂σ (r) ≡ {ψ̂σ (r), ψ̂σ′ (r′ )} = 0 (4.117)
ψ̂σ+ (r)ψ̂σ+′ (r′ ) + ψ̂σ+′ (r′ )ψ̂σ+ (r) = {ψ̂σ+ (r) ψ̂σ+′ (r′ )} =0
where the curly brackets { , } define anticommutators.
As we will demonstrate below these two definitions are sufficient to define a quantum
mechanical fermion field with any dynamics. We note that while the definition of the
vacuum is the same as in the bosonic case the anticommutation relations define a new
quantization ”paradigm” which is different from the familiar canonical quantization via
the commutators.

4.5.2 Fermions in external potential

In order to understand the consequences of the new quantization scheme defined above
we start by considering a simple example - particles in an external potential.

The field equations and the Hamiltonian

The dynamical equation for the field (4.115) in an external potential is a generalization
of the Eq. (4.49) to include the spin
∂ψσ (r, t) X
iℏ = hσσ′ ψσ′ (r, t) (4.118)
∂t ′ σ

As an example we consider the following hσσ′

ℏ2
 
hσσ′ = δσσ′ − ∇2 + Uσσ′ (r) (4.119)
2m
We assumed a spin dependent external potential, like for instance the interaction of the
spin with an inhomogeneous magnetic field (e.g. in the Stern-Gerlach experiment)
Uσσ′ (r) = −γB(r) · sσσ′
with a constant γ and vector s of spin 1/2 matrices.
Using our experience with the spinless field and appropriately generalizing it we con-
sider the equation (4.118) and its complex conjugate as the pair of Hamilton equations
with ψσ (r) and iℏψσ∗ (r) as canonical variables and the following classical Hamiltonian
function
ℏ2
XZ  
H= d3 r δσσ′ |∇ψσ (r)|2 + Uσσ′ (r)ψσ∗ (r)ψσ′ (r) (4.120)
′
2m
σσ
Indeed from
ℏ2
XZ 
3
δH = d r δσσ′ [∇ψσ∗ (r)∇δψσ′ (r) + ∇δψσ∗ (r)∇ψσ′ (r)] +
2m
σσ ′
+Uσσ′ (r) [ψσ∗ (r)δψσ′ (r) + δψσ∗ (r)ψσ′ (r)]}
Version of April 11, 2023 192

we find
ℏ2 2
 
∂ψσ (r) δH 1 X
= = −δσσ′ ∇ ψσ′ (r) + Uσσ′ (r)ψσ′ (r)
∂t δ[iℏψσ∗ (r)] iℏ ′ 2m
σσ

∂[iℏψσ∗ (r)] X ℏ2 2 ∗

δH
=− =− −δσσ′ ∇ ψσ′ (r) + Uσσ′ (r)ψσ∗ ′ (r)
∂t δψσ (r) ′
2m
σσ

which reproduce correctly the field equation (4.118) and its complex conjugate.
On this basis we quantize this spinor field by replacing it with the field operators

ψσ (r) → ψ̂σ (r) , ψσ∗ (r) → ψ̂σ+ (r) (4.121)

with the anticommutation relations (4.117) and the Hamiltonian operator

ℏ2
XZ  
Hop = d3 r δσσ′ ∇ψ̂σ+ (r) · ∇ψ̂σ′ (r) + Uσσ′ (r)ψ̂σ+ (r)ψ̂σ′ (r) (4.122)
′
2m
σσ

or in an equivalent form (cf., the remark after Eq. (4.22))

ℏ2 2
XZ   XZ
Hop = d3 r ψ̂σ+ (r) −δσσ′ ∇ + Uσσ′ (r) ψ̂σ′ (r) = d3 r ψ̂σ+′ (r)hσ′ σ ψ̂σ (r)
2m
σσ ′ σσ ′
(4.123)

Transforming to the normal modes

Let us now solve the quantum mechanical problem defined by the Hamiltonian (4.123).
This is not hard since it is quadratic. We need to find its normal modes. Following a very
similar route as in dealing with (4.53) we consider a single particle equation
X
hσσ′ ui (r, σ) = ϵi ui (r, σ) (4.124)
σ′

The set {ui (r, σ)} is complete and orthonormal in the space of functions of r, σ
XZ X
d3 u∗i (r, σ)uj (r, σ) = δij , ui (r, σ)u∗i (r′ , σ ′ ) = δσσ′ δ(r − r′ ) (4.125)
σ i

We expand the field operators using this set

X X
ψ̂σ (r) = âi ui (r, σ) , ψ̂σ+ (r) = â+ ∗
i ui (r, σ) (4.126)
i i

The operators âi and â+

i can be expressed as

X Z XZ
âi = d3 r ψ̂σ (r)u∗i (r, σ) , â+
i = d3 r ψ̂σ+ (r)ui (r, σ) (4.127)
σ σ
Version of April 11, 2023 193

Using the anticommutators (4.117) and the completeness of the set {ui (r, σ)} it is easy
to see that âi ’s and â+
i ’s satisfy anticommutation relations too

{âi , â+ + +
j } = δij , {âi , âj } = 0 = {âi , âj } (4.128)

Inserting the expansions (4.126) in the Hamiltonian (4.123) we obtain

X
Hop = ϵi â+
i âi (4.129)
i

exactly as in the bosonic case but with the operators obeying the anticommutation rela-
tions.

The eigenstates. Working with anticommiting â’s and â+ ’s

The Hamiltonian (4.129) is a sum of commuting parts. Indeed as is easy to verify that

[n̂i , nˆj ] = 0 (4.130)

where we denoted
n̂i = â+
i âi (4.131)
We need to find the eigenfunctions of n̂i ’s. We follow the same construction as in the
bosonic case, cf., Sec.4.7.1. We note that from Eqs.(4.127) it follows that the vacuum
state |0⟩ defined in (4.116) is annihilated by all âi ’s

âi |0⟩ = 0 for all i′ s (4.132)

Since it is also annihilated by all n̂i ’s it is clearly an eigenstate of the Hamiltonian (4.129)
with zero energy eigenvalue.
We now define one particle states

|1i ⟩ ≡ â+
i |0⟩

for any i. We note the following properties of such states

⟨1i |1i ⟩ = ⟨0|âi â+ +

i |0⟩ = ⟨0|1 − âi âi |0⟩ = 1
⟨0|1i ⟩ = ⟨0|â+
i |0⟩ = 0 , â+ + 2
i |1i ⟩ = (âi ) |0⟩ = 0 (4.133)

In the 1st equality we used the anticomutation relation {âi , â+

i } = 1 and

⟨1i | ≡ [â+ + + +
i |0⟩] = ⟨0|[âi ] = ⟨0|âi

In the 2nd equality we used

⟨0|â+ +
i = [âi |0⟩] = 0

In the 3rd we used the anticommutator

{â+ + + 2
i , âi } = 2[âi ] = 0
Version of April 11, 2023 194

Remarkably this last relation is the expression of the Pauli exclusion principle that two
(or more) identical fermions cannot occupy the same quantum state - in this case the
state ui .
The most relevant for us property of the states |1i ⟩ is that they are eigenstates of n̂i
with eigenvalue ni = 1

n̂i |1i ⟩ = â+ + + + +

i âi âi |0⟩ = âi [1 − âi âi ]|0⟩ = âi |0⟩ = |1i ⟩ (4.134)

The last relation in Eq. (4.133) means that there are only two eigenstates of each n̂i - |0⟩
and |1i ⟩ with respective eigenvalues ni = 0 and ni = 1.
It follows then that the eigenfunctions of the Hamiltonian (4.129) are the products of
all possible eigenstates of n̂i
Y Y
|Ψ{ni } ⟩ = |n1 , n2 , . . . , ni , . . . ⟩ = |ni ⟩ = [â+ ni
i ] |0⟩ with ni = 0 or 1
i i

and with the corresponding eigenenergies

X X
E{ni } = ϵi ni with ni = 0 or 1 ; N = ni
i i

The restriction of the occupations ni to 0 or 1 is of course another expression of the Pauli

principle and is a direct result of the anti-commutation relations which we assumed in the
process of the quantization.
As is the bosonic case the total number of particles N is an eigenvalue of the total
particle number operator X
Nop = n̂i (4.135)
i

which commutes with the Hamiltonian Hop , Eq.(4.129). We will expand on this below in
Section 4.5.2.
So to summarize - the solution of this problem amounts to solving the single particle
Schrödinger equation (4.124) and then populating (filling in) the resulting single particle
states ui with N particles according to the Pauil principe. This solution is of course
identical to what we would obtain in the 1st quantization formalism for N fermions with
the difference that there N was a fixed, given parameter of the problem while it is a
quantum number and can take any value in the 2nd quantization formalism.

Spin independent potential

Let us discuss an important limiting case of the single particle hamiltonian in Eq.(4.118)
which is spin-independent, i.e. diagonal in spin indices

h2 2
hσσ′ = δσσ′ h with h = − ∇ + U (r)
2m
and correspondingly
XZ
Hop = d3 r ψ̂σ+ (r)hψ̂σ (r)
σ
Version of April 11, 2023 195

The normal modes are then products of space and spin parts

ui (r, σ) = uk (r)χs (σ)

with uk (r) solving

huk (r) = ϵk uk (r)
and χs (σ), s = ±1/2 being just two orthogonal space independent spinors, e.g.
   
1 0
χ1/2 = , χ−1/2 =
0 1

The single particle energies ϵk are now spin degenerate and the expansion in normal modes
has the form
X X
ψ̂σ (r) = âks uk (r)χs (σ) , ψ̂σ+ (r) = â+ ∗ ∗
ks uk (r)χs (σ)
ks ks

with the commutation relations

{âks , â+ + +
k′ s′ } = δkk δss , {âks , âk s } = 0 = {âks , âk′ s′ }
′ ′ ′ ′

The Hamiltonian is expressed as

X X
Hop = ϵk â+
ks âks = ϵk n̂ks
ks ks

The number operators n̂ks commute and their eigenfunctions are easily found as before to
be |0⟩ and |1ks ⟩ ≡ â+ks |0⟩ with corresponding eigenvalies nks = 0 and 1. The eigenfunctions
of Hop are then
Y Y
|Ψ{nks } ⟩ = |n1 , n2 , . . . , nks , . . . ⟩ = |nks ⟩ = [â+
ks ]
nks
|0⟩ with nks = 0 or 1
ks ks

and with the corresponding eigenenergies

X X
E{nks } = ϵk nks with nks = 0 or 1 ; N = nks
ks ks

The particle number operator

As in the bosonic case it is useful to express the total particle number operator Nop in
terms of the field operators. Using (4.127) in the expression (4.135) we obtain
XZ
Nop = d3 rψ̂σ+ (r)ψ̂σ (r)
σ

As in the bosonic case this operator is the generator of the global U (1) gauge transfor-
mation, the analogue of Eq. (4.67 for the fermion field

ψ̂σ (r) → eiα ψ̂σ (r) , ψ̂σ+ (r) → e−iα ψ̂σ+ (r) (4.136)
Version of April 11, 2023 196

It is indeed easy to check that ”despite” the anticommutation relations for the fermion
field operators the relation Eq. (4.69) holds for each spin component

[Nop , ψ̂σ (r)] = −ψ̂σ (r)

and therefore so is the corresponding generalization of Eq. (4.68)

e−iαNop ψ̂σ (r)eiαNop = eiα ψ̂σ (r) , e−iαNop ψ̂σ+ (r)eiαNop = e−iα ψ̂σ+ (r)

Since the Hamiltonian Eq. (4.123) is invariant under this transformation it commutes with
Nop .

Working with the fermion field operators

The expression for the particle number operator shows that

ρ̂σ (r) = ψ̂σ+ (r)ψ̂σ (r)

is the density operator of particles with the spin projection σ. Let us consider a state

|r, σ⟩ ≡ ψ̂σ+ (r)|0⟩ (4.137)

and let us act on it with the operator ρ̂σ′ (r′ ). Using the anticommutation relations (4.117)
to commute ψ̂σ′ (r′ ) towards |0⟩ and using Eq. (4.116) we find

ρ̂σ′ (r′ )|r, σ⟩ = ψ̂σ+′ (r′ )ψ̂σ′ (r′ )ψ̂σ+ (r)|0⟩ = δσσ′ δ(r − r′ )ψ̂σ+′ (r′ )|0⟩ = δσσ′ δ(r − r′ )|r, σ⟩
(4.138)
which shows that ψ̂σ+ (r) creates a particle at the position r with spin projection σ. More
precisely it creates delta like particle density of particles with spin projection σ at this
position.
Continuing as we did in the boson case let us consider the state

|r1 σ1 , . . . , rN σN ⟩ = constN ψ̂σ+1 (r1 ) . . . ψ̂σ+N (rN )|0⟩ (4.139)

where we introduced a multiplicative constant for normalization, see below. Acting on

this state with the operator ψ̂σ (r), commuting it towards |0⟩ and using Eq. (4.116) we get
N
X N
Y
ψ̂σ (r)|r1 σ1 , . . . , rN σN ⟩ = constN (−1)Pa δσσa δ(r − ra ) ψ̂σ+b (rb )|0⟩ (4.140)
a=1 b̸=a

where Pa is the parity of the number of permutations one needs to make in order to move
ψ̂σ (r) to the right of ψ̂σ+a (ra ). The result (4.140) means that ψ̂σ (r) destroys (annihilates)
one particle if its coordinates coincide with r and its spin projection with σ. In doing
this it also changes the sign of the resulting part of the wave function if the permutation
number Pa is odd. In this way it’s action is sensitive to the order of the destroyed particle
in the wave function.
Version of April 11, 2023 197

We can use the above result to act on the state (4.139) with the operator ρ̂σ (r). We
obtain in the same manner as in the boson case (cf., Eq.(4.76))
N
X
ρ̂σ (r)|r1 , . . . , rN ⟩ = [ δσσa δ(r − ra )]|r1 , . . . , rN ⟩
a=1

showing that this state describes N particles (delta like particle densities) with spin pro-
jections σa at the positions ra , a = 1, . . . , N .

4.5.3 Relation to the first quantization

In this Section we follow a similar development as in the boson case but with the addi-
tional spin index in the field operators and anti-commutation instead of the commutation
relations.

The wave functions

Consider the fermionic version of the N particles wave function in the second quantization
N
1 X Z Y
|Φ⟩ = √ d3 ra Φ(r1 σ1 , r2 σ2 , . . . , rN σN )ψ̂σ+1 (r1 ) . . . ψ̂σ+N (rN )|0⟩ (4.141)
N ! σ1 ,...σN a=1
The interpretation of this expression is quite clear - we have a linear combination of
N particles in positions r1 , . . . , rN with spin projections σ1 , ..., σN weighted each by the
probability amplitude Φ(r1 σ1 , r2 σ2 , . . . , rN σN ). The anticommutation of ψ̂σ+ (r)’s assures
that this amplitude is antisymmetric with respect to the exchange of any pair of (r, σ)’s7 .
This amplitude is clearly the first quantization partner of the wave function |Φ⟩
As in the bosonic case the normalization of |Φ⟩ assures that it is normalized, i.e.
⟨Φ|Φ⟩ = 1 provided the amplitude Φ(r1 σ1 , . . . , rN σN ) is
X Z Y N
d3 ra |Φ(r1 σ1 , r2 σ2 , . . . , rN σN )|2 = 1
σ1 ,...,σN a=1

In the arbitrary single particle basis ui (r, σ) the above wave function looks exactly as in
the boson case X
|Φ⟩ = Ci1 ,...,iN â+ +
i1 ...âiN |0⟩ (4.142)
i1 ,...,iN

with the ”only” difference that the operators â+i ’s are anticommuting.
As in the bosonic case it is useful and practical to work with the wave functions in the
occupation number representation, cf., Eq (4.84),
X
|Φ⟩ = Cn1 ,...,ni ,... |n1 , n1 , ..., ni , ...⟩ (4.143)
P
n1 ,...,ni ,...;with ni =0 or 1 , i ni =N

with the ”only” difference that the fermionic occupations ni ’s are restricted to be zero or
one.
7 As was already discussed in the bosonic case one can prove that any permutation of N objects can

be achieved by a an ordered ”product” (sequence) of pairwise transpositions.

Version of April 11, 2023 198

The operators
As in the bosonic case the operators in the 1st quantized formulation of fermions are
classified as one-body, two-body, etc.
One body operators
To remind - these operators act on wave functions of identical particles one particle at a
time and have a general form given by the expression (4.93). The difference in the present
(1)
fermion case is that each fa operator in addition to being a function of ra , p̂a = −iℏ∇a
(1)
may also depend on the spin matrices sa . This means that in general fa ’s are 2 × 2
spinor matrices with matrix elements depending on ra and p̂a , cf., the example of hσσ′
in Eq.(4.118).
In a very similar way as in the bosonic case one can show (cf., Appendix 4.7.4) that
in the second quantization one body operators have the form
XZ (1)
(1)
Fop = d3 rψ̂σ+′ (r)fσ′ σ ψ̂σ (r) (4.144)
σσ ′

(1)
where fσσ′ is one (any) of the operators in the sum (4.93) generalized to include the spin
dependence. It is acting on ψ̂σ (r) as a spinor function of r. The expression (4.123) for
the Hamiltonian in an external potential provides a good example of such an operator.
Two body operators
The two body operators for identical particles with spins in the 1st quantization have
(2)
the same form (4.102) as in the bosonic case but with the elementary operators fab in
general depending in addition to ra , rb , p̂a and p̂b also on the spin matrices sa , sb . An
example is given by the so called spin exchange term in a (phenomenological) two particle
interaction
N
1 X
[V (ra − rb ) + W (ra − rb ) (ŝa · ŝb )]
2
a,b=1;a̸=b

(2)
For simplicity we will consider only spin independent fab . One can show that such two
body operators in the fermionic 2nd quantization have a form similar to the bosonic
expression (4.103) with the addition of the spin indices in the field operators

1X
Z
(2)
Fop = d3 rd3 r′ ψ̂σ+ (r)ψ̂σ+′ (r′ )f (2) ψ̂σ′ (r′ )ψ̂σ (r)
2 ′
σσ

with f (2) being a function of r, r′ and p̂ = −iℏ∇r , p̂′ = −iℏ∇r′ 8 . Note the relative order
of the field operators. Since they anticommute it is important to keep it.
General single particle basis
8 The (2)
general spin dependent two body Fop will have the pairwise f (2) ’s depending in addition on the
(2)
spin operators ŝ,sˆ′ of the particles’ pairs. This means they will be four index matrices fσσ′ ,σ”σ′′′ and the
(2)
expression for Fop will be
Z
(2) 1 X (2)
Fop = d3 rd3 r′ ψ̂σ
+ +
(r)ψ̂σ ′ ′
′ (r )fσσ ′ ,σ ′′ σ ′′′ ψ̂σ ′′′ (r )ψ̂σ ′′ (r)
2
σσ ′ σ ′′ σ ′′′
 Version of April 11, 2023 199

(1) (2)
To obtain the expression for Fop and Fop in a general basis ui (r, σ) one just has to
expand the field operators in their expressions in this basis, cf., Eq.(4.126). The result
has identical form to the bosonic expressions (4.105) and (4.106) but the matrix elements
have spin summations in addition to space coordinates integrals
XZ (1)
(1)
⟨i|f |j⟩ = d3 ru∗i (r, σ ′ )fσ′ σ uj (r, σ)
σσ ′
XZ
⟨ij|f (2) |kl⟩ = d3 rd3 r′ u∗i (r, σ)u∗j (r′ , σ ′ )f (2) uk (r, σ)ul (r′ , σ ′ ) (4.145)
σσ ′

where for the two-body operator we write only for the simple (but very common) case of
the spin independent f (2) .

4.5.4 Interacting fermions

Hamiltonian
The most common Hamiltonian of interacting fermions has the form
XZ
Hop = d3 rψ̂σ+′ (r) hσ′ σ ψ̂σ (r) + (4.146)
σσ ′
1X
Z
+ d3 rd3 r′ ψ̂σ+ (r)ψ̂σ+′ (r′ )V (r − r′ )ψ̂σ′ (r′ )ψ̂σ (r)
2 ′
σσ

with
ℏ2 2
hσ′ σ = −δσ′ σ
∇ + Uσ′ σ (r)
2m
and a spin independent two body interaction. In a general single particle basis this
Hamiltonian is X 1X
Hop = ⟨i| h |j⟩â+
i âj + ⟨ij|V |kl⟩â+ +
i âj âl âk (4.147)
ij
2
ijkl

As in the boson case if the solutions of the non interacting part are known, i.e. if one knows
the eigenfunctions of the single particle Hamiltonian h, cf. Eq. (4.124) (e.g. Coulomb
wave functions in atoms) one can use the operators â+ i , âj in this basis. The matrix
⟨i| h |j⟩ is then diagonal making the first term in Hop trivial
X 1X
Hop = ϵi â+
i âi + ⟨ij|V |kl⟩â+ +
i âj âl âk (4.148)
i
2
ijkl

and helping to ”focus attention” on the particle interactions.

with the corresponding generalization of the expression (4.145)
X Z (2)
⟨ij|f (2) |kl⟩ = d3 rd3 r′ u∗i (r, σ)u∗j (r′ , σ ′ )fσσ′ ,σ′′ σ′′′ uk (r, σ ′′ )ul (r′ , σ ′′′ )
σσ ′ σ ′′ σ ′′′
Version of April 11, 2023 200

Heisenberg equations. No classical limit

Despite anticommutation relations of the fermion field operators ψ̂σ (r) and ψ̂σ+ (r) the
Heisenberg equations for these operators
    
∂ ψ̂σ (r, t) ψ̂σ (r, t)
iℏ = , Hop
∂t ψ̂σ+ (r, t) ψ̂σ+ (r, t)

for the general interacting Hamiltonian (4.146) have the same formal appearance as for
bosons apart of the presence of the spin indices. It is a useful exercise for the reader to
work this out explicitly. The equation for ψ̂σ (r) is

∂ ψ̂σ (r, t) X
Z X
iℏ = hσσ′ ψ̂σ′ (r, t) + V (r − r′ ) ψ̂σ+′ (r′ , t)ψ̂σ′ (r′ , t)d3 r′ ψ̂σ (r, t) (4.149)
∂t ′
σ σ′

and the Hermitian conjugate of this equation for ψ̂σ+ (r, t). We note that unlike the boson
case these equations do not have classical limit. This for the obvious reason that Pauli
principle and formally the anti commutation relations of the field operators prevent having
more than one fermion in any given field mode9 .

Mean field approximation

Let us assume for simplicity the spin independent U (r)) and write the Heisenberg equation
(4.149) in the following form

ℏ2 2
 
∂ ψ̂σ (r, t)
Z
′ ′ 3 ′
iℏ = − ∇ + U (r) + V (r − r )ρ̂(r , t)d r ψ̂σ (r, t) (4.150)
∂t 2m

with X
ρ̂(r, t) = ψ̂σ+ (r, t)ψ̂σ (r, t)
σ

The potential U (r) in these equations is formally modified by the last term which is
a convolution of the two body interaction V (r − r′ ) and the operator of the particle
density ρ̂(r′ , t) . For a classical particle density function ρ(r, t) this term would have a
natural meaning of the potential which the particles of the system induce10 . In quantum
mechanic context one can qualitatively think of ρ̂(r, t) as a random variable the probability
amplitude distribution of which is determined by the wave function |Φ⟩ of the many
fermion system under consideration.
9 The so called anticommuting c-numbers (Grassman variables) are often related to the classical limit

of fermionic second quantized operators. In a very crude way they are obtained by setting to zero all the
anticommutators in Eq. (4.117),
+ ′ ′ + + ′
{ψ̂σ (r), ψ̂σ ′ (r )} = {ψ̂σ (r), ψ̂σ ′ (r )} = {ψ̂σ (r) ψ̂σ ′ (r )} = 0

This is in (again a crude) analogy with the classical limit of the bosonic case in which all the canonical
commutators vanish, cf., Berezin, F. A., “The Method of Second Quantization,” Academic Press, 1965.
The Grassman variables are most often used in constructing functional integrals for femionic systems,
cf., Negele, J. W., and Orland, H., “Quantum Many-Particle Systems,” Perseus Books Group, 1998,
pp.25-37
Version of April 11, 2023 201

Given |Φ⟩ one can write

ρ̂(r) = ⟨Φ|ρ̂(r)|Φ⟩ + δ ρ̂(r)

separating the average and the fluctuations of ρ̂(r, t). It is natural to ask if neglecting the
fluctuations would be a good approximation. This would certainly greatly simplify the
problem. It would also be in line with similar approximations known in other fields under
the name ”mean field approximation”10 . In the many-fermion systems such mean field
approximations were first introduced in atomic physics by Hartree and then supplemented
by Fock to result in the Hatree-Fock method. We will address these developments in a
separate chapter. Mean mean approximation and its extensions play a very important
role in theoretical treatment of such many fermion systems as atoms, nuclei and solids.

4.6 The Fock space.

In the first quantization formalism we encountered the notion of the Hilbert space. For
N particles this was the space of all functions of N variables

Φ(x1 , x2 , ...., xN ) with

xa = ra , a = 1, ..., N, symmetrized for spinless bosons ,
xa = ra , σa , a = 1, ..., N, antisymmetrized for fermions

The operators acting on such functions didn’t change the particle number N . The situa-
tion is different in the second quantization formulation. Here already the most elementary
operators ψ̂(r), ψ̂ + (r), âi , â+
i , etc., change the particle number and the most general wave
function should be a linear combination of functions like ΦN with different N ’s and in-
cluding the vacuum
X (1) X (2)
|Φ⟩ = C (0) |0⟩ + Ci â+i |0⟩ + Cij â+ +
i âj |0⟩ + .... +
i ij
(2)
X
+ Ci1 i2 ...iN â+ + +
i1 âi2 ...âiN |0⟩ + ...... (4.151)
i1 i2 ..

The Hilbert space of all such functions is called the Fock space and is a direct sum
M M M
(vacuum) (1 particle Hilbert space) (2 particle Hilbert space) ... (4.152)
M M
... (N particle Hilbert space) ...
10 There is an important aspect which must be addressed first. This is related to the fact that ρ̂(r, t)

and therefore its average includes all the particles in the system while the mean field potential acting on
any given particle Z
V (r − r′ )ρ(r′ , t)d3 r′

must exclude this particular particle. This problem is elegantly solved in the Hartree-Fock method
described in the Mean Field Approximations chapter
Version of April 11, 2023 202

4.7 Appendix
4.7.1 Bosons - reviewing the properties of â’ s and â+ ’s
The vacuum state
Let us defined a special state denoted |0⟩. We shall call this state a vacuum state. The
only properties we will ever need of this state are that it gives zero when acted upon with
anyone of the operators âi and that it is normalised

âi |0⟩ = 0 i = 1, 2, ... (4.153)

⟨0|0⟩ = 1

Single mode
We start by considering the pair âi , â+
i of operators with a fixed index i. We will call
them operators of a single mode ui (r). We then define the state (following an analogy
with the oscillator ladder operators)

|1i ⟩ ≡ â+
i |0⟩ (4.154)

As is easy to see this state is normalised. Indeed using the commutation relations and
the properties of |0⟩ find

⟨1i |1i ⟩ = ⟨0|âi â+ +

i |0⟩ = ⟨0|1 + âi âi |0⟩ = ⟨0|0⟩ = 1

Also have orthogonality

⟨1i |0⟩ = ⟨0|âi |0⟩ = 0
In the same way we define
1 +
|2i ⟩ = const â+
i |1i ⟩ = √ âi |1i ⟩ (4.155)
2
√
The normalization constant is found as const = 1/ 2 by calculating the norm

⟨2i |2i ⟩ = |const|2 ⟨1i |âi â+ 2 + +

i |1i ⟩ = |const| ⟨1i |âi âi âi |0⟩ =
= |const|2 ⟨1i |(1 + â+ + 2 +
i âi )âi |0⟩ = |const| [⟨1i |âi |0⟩ +
+ ⟨1i |â+ + 2
i (1 + âi âi )|0⟩] = 2|const| ⟨1i |11 ⟩ = 2|const|
2

We have orthogonality
1 1 1 1
⟨2i |1i ⟩ = √ ⟨1i |âi |1i ⟩ = √ ⟨1i |âi â+ +
i |0i ⟩ = √ ⟨1i |1 + âi âi |0i ⟩ = √ ⟨1i |0i ⟩ = 0
2 2 2 2
and even more trivially
1
⟨2i |0i ⟩ = √ ⟨1i |âi |0⟩ = 0
2
Version of April 11, 2023 203

By iterating we define
1 1 1
|ni ⟩ = √ â+ |ni − 1⟩ = p (â+ 2
i ) |ni − 2⟩ = · · · = √ (â+ )ni |0⟩ (4.156)
ni i ni (ni − 1) ni ! i

One can prove that the resulting states |ni ⟩ form orthogonal set

⟨mi |ni ⟩ = 0 for mi ̸= ni

Indeed writing
1
⟨mi |ni ⟩ = √ ⟨0|(âi )mi (â+ ni
i ) |0⟩
mi ni
and commuting each âi ’s to the right all the way to |0⟩ one proves this to vanish for
mi ̸= ni .
We also have
+ 1
√ 1
â+
i |ni ⟩ = âi √ (â+ ni
i ) |0⟩ = ni + 1 p (â+
i )
ni +1
|0⟩
ni ! (ni + 1)!
1 1 √ 1
âi |ni ⟩ = âi √ (â+ )ni |0⟩ = ni √ (â+ )ni −1 |0⟩ = ni p (â+
i )
ni −1
|0⟩
ni ! i ni ! i (ni − 1)!

where the factor ni in the second equality of the second line results from commuting âi
through n operators in (â+ i )
ni
to get it acting on |0⟩. The above calculation shows that
√ √
â+
i |ni ⟩ = ni + 1|ni + 1⟩ , âi |ni ⟩ = ni |ni − 1⟩ (4.157)

Note also that by hermitian conjugation

√ √
⟨ni |âi = ni + 1⟨ni + 1| , ⟨ni |â+
i = ni ⟨ni − 1| (4.158)

The last two sets of equalities define the action of the operators âi and â+ i on any state
”belonging” to the mode ui (r). Indeed for any such state |ξi ⟩ we can determine P the result
of acting on it with âi or â+ i by writing it as a linear combination |ξi ⟩ = ni cni |ni ⟩ of
the basis states |ni ⟩.
Let us now consider the operator n̂i = â+ i âi . The basis states |ni ⟩, Eq. (4.156) are its
eigenstates
√
n̂i |ni ⟩ = â+
i âi |ni ⟩ = ni â+i |ni − 1⟩ = ni |ni ⟩ (4.159)
This operator is the i-th mode number operator.

Many modes
We now generalize the above single mode construction to all modes of the complete set
ui (r). This is easily done mostly because pairs of âi and â+
i commute for different i’s.
The general multimode analogue of the states |ni ⟩ is
Y Y 1
|{ni }⟩ ≡ |n1 , n2 , . . . , nk , . . . ⟩ = |ni ⟩ = √ (â+ )ni |0⟩ (4.160)
i i
ni ! i
Version of April 11, 2023 204

The operators âi and â+ i act on these states as

√
â+
i |n1 , . . . , ni , . . . ⟩ = ni + 1|n1 , . . . , ni + 1, . . . ⟩
√
âi |n1 , . . . , ni , . . . ⟩ = ni |n1 , . . . , ni − 1, . . . ⟩ (4.161)

and have number operators for all modes

n̂i |n1 , . . . , ni , . . . ⟩ = â+

i âi |n1 , . . . , ni , . . . ⟩ = ni |n1 , . . . , ni , . . . ⟩ (4.162)

It is useful and important to introduce the total particle number operator

X X
Nop = n̂i = â+
i âi (4.163)
i i

which ”measures” the sum of all ni ’s

!
X
Nop |n1 , n2 , . . . , ni , . . . ⟩ = ni |n1 , n2 , . . . , ni , . . . ⟩ (4.164)
i

4.7.2 Bosons - wave function normalization

Let us consider the norm of the wave function Eq. (4.79)

⟨Φ|Φ⟩ = (4.165)
N
Z Z Y
= d3 ra d3 rb′ Φ∗ (r′1 , r′2 , . . . , r′N )Φ(r1 , r2 , . . . , rN )⟨r′1 , . . . , r′N |r1 , . . . , rN ⟩
a,b=1

We need to evaluate the overlap ⟨r′1 , . . . , r′N |r1 , . . . , rN ⟩. In a straightforward way by

commuting ψ̂(r′a )’s to the right all the way to the vacuum state |0⟩ we obtain

⟨r′1 , . . . , r′N |r1 , . . . , rN ⟩ = |constN |2 ⟨0|ψ̂(r′N ), ...ψ̂(r′1 )ψ̂ + (r1 ) . . . ψ̂ + (rN )|0⟩ =
N
X N
Y
2
= |constN | ⟨0|ψ̂(r′N ), ...ψ̂(r′2 ) δ(r′1 − ra ) ψ̂ + (rb )|0⟩ =
a=1 b̸=a
N
X N
X N
Y
= |constN |2 ⟨0|ψ̂(r′N ), ...ψ̂(r′3 ) δ(r′1 − ra ) δ(r′2 − rb ) ψ̂ + (rc )|0⟩ =
a=1 b=1,b̸=a c=1,c̸=a,b
N
X N
X N
X
= ... = |constN |2 δ(r′1 − ra ) δ(r′2 − rb ) δ(r′3 − rd )..⟨0|0⟩ =
a=1 b=1,b̸=a d=1,d̸=a,b,c

X N
Y
= |constN |2 δ(r′a − rP a ) (4.166)
P a=1

where P stands for permutations of the particle indices a = 1, 2, .., N . There are N !
permutations of N indices and therefore N ! terms in the last sum.
Version of April 11, 2023 205

Using this result in Eq. (4.165) one can use the delta functions to reduce the norm
⟨Φ|Φ⟩ to a sum of integrals
N
XZ Y
⟨Φ|Φ⟩ = |constN |2 d3 ra Φ∗ (rP 1 , rP 2 , . . . , rP N )Φ(r1 , r2 , . . . , rN )
P a=1

Since Φ(r1 , r2 , . . . , rN ) is symmetric with respect to the permutations of its arguments

the above N ! integrals are identical
N
Z Y
2
⟨Φ|Φ⟩ = N !|constN | d3 ra |Φ(r1 , r2 , . . . , rN )|2
a=1

which leads to the consistent normalization conditions to unity of both first and second
quantization wave functions Eq. (4.80) for the choice of the const as
1
constN = √
N!

4.7.3 Bosons - calculating Kop |Φ⟩

As with Uop |Ψ⟩ we start by considering
N
ℏ2 2
Z   Y
d3 r ψ̂ + (r) − ∇r ψ̂(r) ψ̂ + (ra )|0⟩
2m a=1

Using in this expression the result (4.76) and

∇2r δ(r − rb ) = −∇r ∇rb δ(r − rb ) = ∇2rb δ(r − rb )
we get it in the form
 
Z N  2
 N
X ℏ Y
d3 r ψ̂ + (r)  − ∇2 δ(r − rb ) ψ̂ + (ra ) |0⟩
2m rb
b=1 a̸=b

Therefore
N N
ℏ2 2
 
1
Z Y Z Y
3 3 +
Kop |Φ⟩ = √ d ra Φ(r1 , r2 , . . . , rN ) d r ψ̂ (r) − ∇r ψ̂(r) ψ̂ + (ra )|0⟩ =
N! a=1
2m a=1
 
N N  2
 N
1
Z Y Z X ℏ Y
=√ d3 ra Φ(r1 , r2 , . . . , rN ) d3 r ψ̂ + (r)  − ∇2 δ(r − rb ) ψ̂ + (ra ) |0⟩ =
N! a=1
2m rb
b=1 a̸=b
 
N N  2
 N
1
Z Z Y X ℏ Y
=√ d3 r ψ̂ + (r) d3 ra  − ∇2rb Φ(r1 , r2 , . . . , rN )δ(r − rb ) ψ̂ + (ra ) |0⟩
N! a=1
2m
b=1 a̸=b

in the last line we changed the order of integration and then did integration by parts
(twice) to free the delta functions and transfer ∇2rb to act on Φ(r1 , . . . , rN ).
Changing the order of integrations back again and using the delta functions we obtain
Version of April 11, 2023 206

4.7.4 Fermions - calculating Fop |Φ⟩

(1)
Deriving the action of the operator Fop , Eq. (4.144) on the many fermion wave function
(1) (1)
Eq. (4.141) let us start by applying the part ψ̂σ+′ (r)fσ′ σ of Fop to the expression (4.140),
with the result
N
(1)
X Y Y
constN fσ′ σ (r)δσσa δ(r − ra ) ψ̂σ+b (rb )ψ̂σ+′ (r) ψ̂σ+b (rb )|0⟩ (4.167)
a=1 b<a b>a
(1)
Here we for simplicity assumed that fσ′ σ is a function of r only so that we could bring
ψ̂σ+′ (r) ”through it” and commute to where ψ̂σ+a (ra ) was. This commuting generated
additional the factor (−1)Pa giving overall unity when combined with the same factor in
(1)
Eq. (4.140). We note that for fσ′ σ depending on −iℏ∇r one should use the intermediate
integration by parts in analogy with what we did in the kinetic energy case with bosons,
cf., Appendix 4.7.3.
To finish the calculation let us sum the result (4.167) over σ and σ ′ and integrate over
r. Using δσσa to perform the sum over σ and δ(r − ra ) to do the integral we obtain
N N X
(1)
Y X Y Y
(1)
Fop ψ̂σ+a (ra )|0⟩ = fσ′ σa (ra ) ψ̂σ+b (rb )ψ̂σ+′ (ra ) ψ̂σ+b (rb )|0⟩
a=1 a=1 σ ′ b<a b>a

This gives
N N
(1) 1 X Z Y Y
Fop |Φ⟩ =√ d3 ra Φ(r1 σ1 , r2 σ2 , . . . , rN σN )Fop (1)
ψ̂σ+a (ra )|0⟩ =
N ! σ1 ,...σN a=1 a=1
N
1 X Z Y
=√ d3 rb Φ(r1 σ1 , r2 σ2 , . . . , ra σa , . . . , rN σN ) ×
N ! σ1 ,...σN b=1
N X
(1)
X Y Y
× fσ′ σa (ra ) ψ̂σ+b (rb )ψ̂σ+′ (ra ) ψ̂σ+b (rb )|0⟩ =
a=1 σ′ b<a b>a
N N X
1 X Z Y X (1)
=√ d3 rb fσa σ′ (ra )Φ(r1 σ1 , r2 σ2 , . . . , ra σ ′ , . . . , rN σN ) ×
N ! σ1 ,...σN b=1 a=1 σ ′
Y Y
× ψ̂σ+b (rb )ψ̂σ+a (ra ) ψ̂σ+b (rb )|0⟩
b<a b>a

where after the last equality sign we have used the presence of sums over both σa and σ ′
and interchanged notation of their summation variables σa ↔ σ ′ . This finally gives
N N
(1) 1 X Z Y Y
Fop |Φ⟩ = √ d3 ra Φ′ (r1 σ1 , r2 σ2 , . . . , rN σN ) ψ̂σ+a (ra )|0⟩
N ! σ1 ,...σN a=1 a=1

with
N X
" #
(1)
X
′
Φ (r1 σ1 , r2 σ2 , . . . , rN σN ) = fσa σ′ (ra ) Φ(r1 σ1 , r2 σ2 , . . . , ra σ ′ , . . . , rN σN )
a=1 σ ′