Path Integrals in Quantum Mechanics

and Low-Dimensional QFT

Bachelor of Science Degree in Physics
Division of Theoretical Physics

Author: Johanna Adbo

Supervisor: Pietro Longhi
Subject Reader: Lisa Freyhult

July 21, 2023

 Abstract
The focus of this thesis is to introduce the path integral and some of its ap-
plications. One interpretation of quantum mechanics is that a microscopic
system which moves from an initial- to a final state moves through each
possible intermediate state. The path integral uses the principle of least
action to sum over all such intermediate states to find the evolution of a
quantum mechanical system. We compare the path integral approach to
that of the Schrödinger equation and show that the two give an equivalent
description of quantum mechanics.

To demonstrate the usefulness of the path integral, we introduce low-dimen-

sional quantum field theory (QFT). In particular, we discuss Feynman dia-
grams. The idea behind Feynman diagrams is to sum over all possible weak
interactions between fields to evaluate the properties of a system through
the path integral. We also carry out a computation of a low energy effec-
tive action in a 0-dimensional model. The result of the computation shows
that there is free energy also in a vacuum. Finally, we briefly generalize
some of the previous discussion to 1-dimensional QFT. To give an example
of a practical application, we give a qualitative discussion of how the path
integral can be applied to statistical mechanics to predict the behaviour of
superfluids.

Sammanfattning

Målet med den här rapporten är att introducera konceptet vägintegral och
några av dess applikationer. En tolkning av kvantmekanik är att ett mikro-
skopiskt system som går från ett initialt- till ett slutgiltigt tillstånd kommer
att passera genom alla möjliga mellanliggande tillstånd. Vägintegralen an-
vänder sig av principen om minsta verkan för att summera över alla sådana
mellanliggande tillstånd för att hitta utvecklingen hos ett system. Vi kom-
mer att jämföra vägintegralen med Schrödingers ekvation och visa att de
två ger en ekvivalent beskrivning av kvantmekaniken.

För att demonstrera vägintegralens användbarhet introducerar vi lågdimen-

sionell kvantfältteori. Vi diskuterar speciellt Feynmandiagram. Idén bakom
Feynmandiagram är att summera över alla möjliga svaga interaktioner mel-
lan fält för att utvärdera fysikaliska egenskaper hos system med hjälp av
vägintegraler. Vi kommer också att utvärdera en effektiv verkan i 0-dimen-
sionell kvantfältteori. Resultatet visar att det finns fri energi även i ett
vakuum. Slutligen generaliserar vi delar av vår tidigare diskussion till 1-
dimensionell kvantfältteori. Som ett exempel på praktiska applikationer
för vi en kvalitativ diskussion kring hur vägintegraler kan användas inom
statistisk mekanik för att förutsäga egenskaper hos superfluider.

i
Contents
1 Introduction 1

2 Important Concepts in Quantum Mechanics 3

2.1 The Lagrangian in Quantum Mechanics . . . . . . . . . . . . . . . . . . . . 3
2.2 Wave-Particle Duality and a Brief History of Quantum Theory . . . . . . . 7
2.3 The Double-Slit Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 The Uncertainty Principle in Experiments . . . . . . . . . . . . . . . . . . 9
2.5 The Kernel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 The Path Integral in Quantum Mechanics 12

3.1 The Path Integral in the Classical Limit . . . . . . . . . . . . . . . . . . . 12
3.2 Evaluation of the Path Integral for a Free Particle . . . . . . . . . . . . . . 13
3.3 Momentum Representation of the Path Integral . . . . . . . . . . . . . . . 16
3.4 The Path Integral in Relation to the Schrödinger Equation . . . . . . . . . 21

4 Important Concepts in QFT 23

4.1 Manifolds and Riemannian Metric in Different Dimensions . . . . . . . . . 23
4.2 The Partition Function and Correlation Functions . . . . . . . . . . . . . . 25
4.3 Grassmann Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5 The Path Integral in 0-Dimensional QFT 27

5.1 Feynman Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.2 Supersymmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3 Deformation Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.4 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.5 Landau-Ginzburg Supersymmetry . . . . . . . . . . . . . . . . . . . . . . . 40

6 The Path Integral in 1-Dimensional QFT 43

6.1 Feynman Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.2 Supersymmetry and Supercharges . . . . . . . . . . . . . . . . . . . . . . . 44
6.3 The Partition Function in Statistical Mechanics and Thermal QFT . . . . 46

7 Conclusions 50

A The Central Equations in Lagrangian and Hamiltonian Mechanics 54

B Cauchy’s Theorem over Infinite Domains 55

B.1 Complex-Valued Gaussian Integral . . . . . . . . . . . . . . . . . . . . . . 55
B.2 Complex-Valued Gaussian Integral with a Quadratic Term . . . . . . . . . 57
B.3 Complex-Valued Gaussian Integral with a Linear Term in the Exponent . . 57
B.4 Integral of a Complex-Valued Exponential Function . . . . . . . . . . . . . 58
B.5 The Principal Part of the Integral of a Complex-Valued Exponential Function 59

C An Introduction to Green’s Functions 60

ii
 1 Introduction

1 Introduction
Quantum mechanics was originally formulated using Hamiltonian mechanics. This was
the natural choice over Lagrangian mechanics since Hamiltonian mechanics is described
by a set of canonical coordinates and momenta in phase space. (For an introduction to
analytical mechanics, see Goldstein et al. [14]). In quantum mechanics, we can use Her-
mitian operators in Hilbert space, denoted by a hat, which satisfy similar commutation
relations as those for the classical canonical variables. To go from the classical theory
to the quantum theory, we make the exchange [25]

[Â, B̂]
{A, B} → .
iℏ
Here, {} denotes a Poisson bracket. If A and B are functions of some coordinates qi and
conjugate momenta pi , the Poisson bracket has the property {A, B} = ∂q ∂A ∂B
i ∂pi
∂B ∂A
− ∂qi ∂pi
.
The index i implies a sum over all variables. In the quantum mechanical case, [Â, B̂] =
ÂB̂ − B̂  is a commutator, and  and B̂ are constructed from operators. We may also
formulate a Hamiltonian from such operators, and this Hamiltonian generates the time
evolution of our system. This relatively simple transition from the classical theory to
the quantum theory motivates the use of Hamiltonian formalism in quantum mechanics.

When Dirac proposed what was to become the path integral in 1933 [6], he did it with
the intention of implementing Lagrangian formalism in quantum mechanics. Why, if
Hamiltonian formalism works so well, would we switch to Lagrangian formalism? Partly
because the latter can be considered more fundamental. Lagrangian mechanics is based
on the action principle, and there is no equivalent action principle involving the coordi-
nates and conjugate momenta in Hamiltonian mechanics. The Lagrangian is constructed
from the kinetic energy and the potential energy of a system. The same cannot always
be said about the Hamiltonian. Second, the Lagrangian formalism is more suitable for
special relativity. This is because the action is a scalar quantity, and therefore invari-
ant under coordinate transformations, e.g. Lorentz transformations. The Hamiltonian
formalism is not compatible with special relativity: it involves a fixed time which is
taken to be the same in all reference frames. Under a Lorentz transformation, the time
parameter changes, and Hamiltonian mechanics breaks down. Lagrangian mechanics
instead treats time and space on equal footing. This is useful when we want to combine
quantum mechanics and special relativity, for example in quantum field theory, or QFT.

The complete formulation of the path integral was realized by Feynman in 1942 [10].
Feynman also showed that the path integral is equivalent to the Schrödinger equation in
evaluating the time-evolution of a quantum mechanical system. We will prove the same
in this thesis. The idea of the path integral is to sum, or integrate, over all possible
paths for a system as it evolves from an initial state to a final state. This integral gives
us the probability amplitude for the evolution of the system. Each path is weighed by a
phase factor where the phase is given by the classical action. This approach to quantum

1
 1 Introduction

mechanics is attractive not only for the discussion about Lagrangian mechanics above,
but also from a theoretical standpoint. One way to interpret quantum mechanics is to
imagine that a quantum mechanical system takes not one path from a state to another,
but every possible intermediate path simultaneously.

The path integral is however rarely used in non-relativistic quantum mechanics. But
it has many other applications, notably in QFT. QFT treats fields, such as the elec-
tromagnetic field, and combines quantum mechanics, classical field theory and special
relativity. It is a theory which revolves around counting all possible ways that a process
can occur, and the path integral is well suited for this. In this thesis, we introduce path
integrals in QFT through the partition function, which is a central equation in QFT
containing much information about the physical properties of a system. However, it is
often difficult to evaluate these path integrals. To simplify computation, one can use
several tools, e.g. perturbation theory, which involves Feynman diagrams. The idea
behind Feynman diagrams is to sum over all possible weak interactions between fields to
evaluate terms in perturbation expansions. Another important aspect of QFT is sym-
metries. In general, finding symmetries of the action puts constraints on the dynamics
of a system and helps us evaluate its physical properties. The specific symmetry which
will be discussed in this thesis is supersymmetry, i.e. a symmetry between bosonic- and
fermionic fields. For QFT:s possessing this symmetry, computations often simplify.

Today, the path integral is one of the cornerstones in many field theories. It has applica-
tions in e.g. statistical mechanics and string theory [18] [3]. One specific application is
brought up in this thesis: the predictive power of the path integral for the behaviour of
superfluids. A superfluid is a fluid which adopts exotic behaviours at low temperatures.
For example, at a specific temperature, its heat capacity grows exponentially and it no
longer experiences friction.

In conclusion, it is well worth the time to gain some understanding of the path integral
and its role in modern physics. This thesis is a literature study of path integrals in
quantum mechanics and QFT. It aims to explain the following questions at the level of
a bachelor student.

• Why was the path integral introduced in quantum mechanics and how is it defined?

• What are the fundamental properties of the path integral in quantum mechanics?

• How can one prove equivalence between the path integral and the Schrödinger
equation in quantum mechanics?

• How is the path integral used in QFT, specifically in the formulation of the parti-
tion function?

• How do Feynman diagrams and supersymmetry facilitate evaluation of path inte-

grals?

2
 2 Important Concepts in Quantum Mechanics

• How can path integrals be used in statistical mechanics to predict the behaviour
of superfluids?
As an illustration of the above points, we will perform a computation using path in-
tegrals. More specifically, we study an effective action in a 0-dimensional QFT and
evaluate some properties of the system using both classical mechanics and QFT. We
then compare the results.

We also seek a deeper understanding of quantum theories throughout, from the point of
view of this new formalism.

2 Important Concepts in Quantum Mechanics

We begin with some background theory. In section 2.1, we discuss the Lagrangian
formalism and how one can apply it to quantum mechanics. The discussion is based
mainly on Dirac’s original article on the subject from 1933 [6]. It also involves some
input from Feynman’s PhD thesis from 1942 [10] and Brown’s summary of Feynman’s
PhD thesis [1]. This provides some of the mathematical foundation for the path integral.

In sections 2.2-2.5, we discuss wave-particle duality and the strange nature of quantum
mechanical systems, which gives a more intuitive understanding of the path integral.
The discussion is based on Feynman and Hibbs [11, ch. 1-5].

2.1 The Lagrangian in Quantum Mechanics

In the introduction, we discussed reasons for wanting to introduce Lagrangian formalism
in quantum mechanics. However, there are issues with translating Lagrangian mechanics
directly to quantum mechanics. (We give a brief summary of the central equations in
Lagrangian- and Hamiltonian mechanics in Appendix A.) In Lagrangian mechanics, the
time evolution of a system is obtained from the Euler-Lagrange equations. These equa-
tions involve partial derivatives of the Lagrangian L = L(q, q̇, t) with respect to some
generalized coordinates q, q̇. In quantum mechanics, these variables are replaced with
operators, and it is not obvious how to define derivatives of operators; the derivative is
itself an operator, and operators do not act on other operators.

Thus we cannot directly apply the Euler-Lagrange equations in quantum mechanics.

Instead we must introduce the Lagrangian by another method. Dirac suggested we find
a generating function which gives a time-transformation of the variables of our system, to
find its evolution. To achieve this, we begin with classical mechanics and a more general
transformation: that from some independent variables (q, p) to some other independent
variables (Q, P ), such that

Q = Q(q, p, t),
P = P (q, p, t). (2.1)

3
 2 Important Concepts in Quantum Mechanics

For simplicity, we examine the case with only one coordinate and one conjugate mo-
menta. The variables in equation (2.1) must fulfill

∂F ∂L
P =− , P = ,
∂Q ∂ Q̇
(2.2) (2.3)
∂F ∂L
p= . p= .
∂q ∂ q̇

The equations (2.2) is the general formalism of canonical transformations given some
generating function F (q, Q, t). The equations (2.3) are the definitions of conjugate mo-
menta, see for example Goldstein et al. [14]. To find a transformation in time, we
impose (Q, P ) = (qi , pi ), in which (qi , pi ) is the coordinate and conjugate momentum at
the initial time ti , and (q, p) = (qf , pf ), where (qf , pf ) is the coordinate and conjugate
momentum at a later time tf . We also introduce qk as the coordinate at some time in
the interval [ti , tf ]. We note that equations (2.2) are fulfilled if we let

ˆ tf
F =S= L(q, q̇, t)dt. (2.4)
ti

A demonstration of this is given below. It may be skipped by the reader.

We demonstrate that equation (2.4) fulfills the desired relations (2.2) by considering the
expression

ˆ tf
!
∂S ∂L ∂q ∂L ∂ q̇
= + dt. (2.5)
∂qk ti ∂q ∂qk ∂ q̇ ∂qk

We use integration by parts on the second term in the integrand

ˆ tf
" #t ˆ tf
∂L ∂ q̇ ∂L ∂q f ∂L ∂q
dt = {EL-equations, Appendix A} = − dt
ti ∂ q̇ ∂qk ∂ q̇ ∂qk ti ti ∂q ∂qk
ˆ tf
∂qf ∂qi ∂L ∂q
= {Equation (2.3)} = pf − pi − dt. (2.6)
∂qk ∂qk ti ∂q ∂qk

We use equation (2.6) to rewrite equation (2.5) and obtain

∂S ∂qf ∂qi
= pf − pi .
∂qk ∂qk ∂qk
By setting qk = qi , qf we obtain

4
 2 Important Concepts in Quantum Mechanics

∂S
pi = − ,
∂qi
∂S
pf = .
∂qf
which agrees with equations (2.2).

The above discussion about transformations has involved only classical mechanics. We
return to quantum mechanics, where a canonical transformation between two coordinate
representations is given by some transformation function ⟨q|Q⟩. We can write

⟨q|q̂|Q⟩ = q ⟨q|Q⟩ , (2.7)

∂
⟨q|p̂|Q⟩ = −iℏ ⟨q|Q⟩ , (2.8)
∂q
⟨q|Q̂|Q⟩ = Q ⟨q|Q⟩ , (2.9)
∂
⟨q|P̂ |Q⟩ = iℏ ⟨q|Q⟩ . (2.10)
∂Q

q̂, Q̂ simply pick out the position eigenvalue and p̂ acts as a derivative. For a derivation
of this, see Sakurai and Napolitano [25, p. 52]. Equation (2.10) follows naturally from
the property demonstrated in equation (2.8), since ⟨q|P̂ |Q⟩ = ⟨Q|P̂ |q⟩ = −iℏ ∂Q ∂
⟨Q|q⟩ =
iℏ ∂Q ⟨q|Q⟩. This holds since p̂ is a hermitian operator.
∂

Equations (2.7)-(2.10) have great implications. If we consider some function ĝ = ĝ(q̂, Q̂, t),
then equation (2.7) and (2.9) imply

⟨q|ĝ(q̂, Q̂, t)|Q⟩ = g(q, Q, t) ⟨q|Q⟩ . (2.11)

This is easily seen for every function which can be rewritten as a power series. Again,
q̂, Q̂ are operators and q, Q are variables in equation (2.11).
i
We make an ansatz for our transformation function, ⟨q|Q⟩ = e ℏ U , where U is a function
of the form U = U (q, Q, t). We get, from equation (2.8) and (2.10)

∂
⟨q|p̂|Q⟩ = U (q, Q) ⟨q|Q⟩ ,
∂q
∂
⟨q|P̂ |Q⟩ = − U (q, Q) ⟨q|Q⟩ . (2.12)
∂Q
Comparison between (2.12) and (2.11) gives

5
 2 Important Concepts in Quantum Mechanics

∂ Û ∂ Û
p̂ = , P̂ = − . (2.13)
∂ q̂ ∂ Q̂
Equation (2.13) implies that Û is the quantum analogue of our generating function F
in equation (2.2). By equation (2.4), we also have F = S, so that our transformation
function is given by

i
⟨qf |qi ⟩ = e ℏ U , (2.14)
where U is the quantum analogue of the classical action. U (qi , qf , t) is a function of the
coordinate at the initial time and the coordinate at the final time. It is reasonable to
postulate that equation (2.14) holds only for infinitesimal time intervals [t, t + ϵ], since in
this case, both the velocity and the position of the system can be expressed in terms of
the initial- and final coordinates. This is because we can set the velocity to be constant
and the position to be an intermediate point.

Dirac also introduced a classical analogue to ⟨qf |qi ⟩ called A(tf , ti ), where

i
 
A(tf , ti ) = exp Ldt ,
ℏ
where L is the classical Lagrangian. Also A(tf , ti ) is valid only for an infinitesimal
transformation. To achieve a finite transformation in time, we must apply many such
consecutive infinitesimal transformations. In the classical case, we get

A(tT ) = A(ttm )A(tm tm−1 )A(tm−1 tm−2 )A(tm−2 tm−3 )...A(t1 T ).

The corresponding finite transformation in quantum mechanics can be constructed by
considering some finite transformation from t to T and introducing a complete basis
at every intermediate timestep. According to the mathematical formalism of quantum
mechanics, the resulting expression is given by
ˆ ˆ ˆ
⟨qt |qT ⟩ = ... ⟨qt |qm ⟩ dqm ⟨qm |qm−1 ⟩ dqm−1 ⟨qm−1 |qm−2 ⟩ dqm−2 ... ⟨q2 |q1 ⟩ dq1 ⟨q1 |qT ⟩ .
(2.15)
Equation (2.15) implies that the path taken by a quantum mechanical system is not
deterministic. At each intermediate step between the initial and final time, we must
integrate over all possible positions for our system.

The above discussion by Dirac comes very close to Feynman’s final formulation of the
path integral, although Dirac’s transformation function is missing a normalizing factor.
Also, while Dirac argued that U in equation (2.14) is the quantum analogue of the
classical action, Feynman showed in his PhD thesis that it is actually equivalent to the
classical action. We prove the same in section 3.4.

6
 2 Important Concepts in Quantum Mechanics

2.2 Wave-Particle Duality and a Brief History of Quantum

Theory
Particles can be described as both localized particles and as waves: which description is
the most accurate depends on the system we choose to investigate and also on the way in
which we measure it. Historically, the concept of wave-particle duality has been widely
discussed, and the debate was first settled for photons. The wave nature of light was
well explained by Maxwell’s equations, first formulated in 1865 [20] and the theory of
quantized light was proposed by Planck as he attempted to solve the problem of black-
body radiation in 1901 [23]. The combined theory of wave-particle duality for photons
was proposed by Einstein as he discovered the photoelectric effect in 1905 [8]. Later,
de Broglie proposed in his PhD thesis in 1924 that the theory of wave-particle duality
should be extended to all particles [21]. Later experiments confirmed this for electrons
[5] and other particles. The complete theory of wave-particle duality is part of today’s
quantum mechanics, developed in the 1920s [15].

Later in this report, we discuss quantum field theory, QFT. The foundations of QFT was
laid out as early as 1927 by Paul Dirac as he attempted to quantize the electromagnetic
field [7][19]. Quantum fields showed early promise as they could describe both antimatter
and annihilation- and creation of particles. QFT also combined special relativity and
quantum mechanics, which was another reason for its development [29]. The formulation
of QFT continued throughout the 1900s and is the subject of much current research.
Feynman made many important contributions, among others his Feynman diagrams
which appeared in print for the first time in 1949 [12]. They will be discussed in section
5.1.

2.3 The Double-Slit Experiment

A famous demonstration of wave-particle duality is the double-slit experiment, first per-
formed with photons in 1801 by Thomas Young [30][28]. However, Young argued at the
time that the result of his experiment was proof of the wave nature of light, rather than
of wave-particle duality.

The experiment is conducted as follows. Particles are emitted from a source and allowed
to pass through two narrow slits in a wall, and then strike a detector. Each particle is
detected at some location in the detector, thus the particles behave as localized parti-
cles when we measure their spatial position. The original experiment in 1801 was not
sophisticated enough to detect individual particles, which explains Young’s original con-
clusion. If we leave our detector on for some time, we are left with a pattern made up of
many detected particles. If we allow the particles to pass through only one of the slits,
we observe a smooth distribution in the detector, with a maximum in front of the open
slit. If we allow the particles to pass through both slits, we observe bands of alternating
high and low intensity. This behaviour is observed even if we allow just one particle
at a time to pass though the experiment. Since the intensity from both slits in our
experiment cannot be explained as the superposition of intensities from slit 1 and slit

7
 2 Important Concepts in Quantum Mechanics

2, each particle cannot have passed through just one slit or the other. Rather, it seems
that the particles must be described as a non-localized waves, able to pass through both
slits simultaneously. This behaviour is illustrated in figure (1).

Only the upper slit open

Both slits open

Only the lower slit open

Figure 1: Illustration of the double-slit experiment, where the slits are on the left
hand side and the detector is on the right hand side, with white dots representing
detected particles. The figures were generated in Python.

We denote the amplitude of a particle wave just before it strikes some point in the
detector Φ. We think of the slits as two particle sources, where the amplitude for arriving
in the detector after passing though the upper slit is denoted by Φ1 and similarly Φ2 for
the lower slit. In general, to obtain the intensity of a wave, we square its amplitude. If our
two sources are independent, we obtain the intensity in the detector as I = |Φ1 |2 + |Φ2 |2 .
If the two sources are not independent, but are instead interfering, we must add the
amplitudes before taking the square as in

I = |Φ1 + Φ2 |2 . (2.16)

This is indeed the equation which describes the observed pattern in figure (1), with
both slits open. Our particles are then best described as interfering waves as they pass
through the experiment and as localized particles when they strike the detector. This is
a clear demonstration of wave-particle duality.

We can of course determine which of the two slits a particle passes through on its way
to the detector. We could for example place a light source after the slits. A photon from
the light source then interacts with each particle just after it passes through one of the
slits, and alerts us of its position. However, we disturb the momentum of the particle

8
 2 Important Concepts in Quantum Mechanics

in the process, and smear out the interference pattern to where it is no longer visible.
The resulting pattern in the detector is the superposition of two smooth distributions:
the one which results from keeping only slit 1 open, and the one from keeping only slit
2 open. It seems as though our measuring which slit the particles pass through results
in the particles passing through only one slit at a time.

2.4 The Uncertainty Principle in Experiments

The discussion in the previous section demonstrates a general rule for all experiments.
It is well expressed in Feynman and Hibbs [11, p.9]:

"Any determination of the alternative taken by a process capable

of following more than one alternative destroys the interference
between alternatives."

This is another version of Heisenberg’s uncertainty principle. In the case of the double-
slit experiment, it is easy to see that the statement above is compatible with the original
statement by Heisenberg, ∆x∆p ≥ ℏ2 . When we try to determine the position of a
particle with a light source, we necessarily introduce an uncertainty in its momentum.
This means that we alter the trajectory of the particle, and we disturb the pattern in
the detector.

2.5 The Kernel

The amplitudes in equation (2.16) can be written as Φi (b, a), where the index i again
denotes the amplitude for a specific path through the experiment. The argument (b, a)
specifies the endpoint and the starting point for the path. In other words, b = (xb , tb )
denotes that the path ends at the point xb at the fixed time tb . The initial point is
similarly given by a = (xa , ta ). The total amplitude Φ(b, a) = Φ1 (b, a) + Φ2 (b, a) is the
sum of the amplitudes for alternative paths through the experiment. In our case, it
consists of the paths 1 and 2, as illustrated in figure (2). The amplitude Φ(b, a) is also
associated with an intensity through equation (2.16). Properly normalized, the intensity
gives the probability for any one particle to go from point xa to point xb in the time
interval tb − ta . We therefore refer to Φ(b, a) as a probability amplitude, assuming it is
properly normalized.

We would now like to make a more general statement. We imagine putting more walls in
between the source and the detector in the double-slit experiment. We also make some
extra slits in the walls. To find the total amplitude to go from the detector at point a to
the point b, we must sum over the amplitudes from all possible paths between the two
points. The scenario is illustrated in figure (3), where examples of 3 possible paths are
drawn. We call this total amplitude the kernel

K(b, a) = Φi (b, a), (2.17)

X

9
 2 Important Concepts in Quantum Mechanics

Particle source

Detector
a

Figure 2: Two possible particle paths through the double-slit experiment.

Particle source

Detector
a

Figure 3: Three possible particle paths through a multiple-slit experiment.

If we keep putting up more walls and making more slits, eventually we have to sum over
an infinite amount of paths through the space between a and b. The kernel in equation
(2.17) is generalized to an integral

ˆ b
i
K(b, a) = C Dx(t)e ℏ S[b,a] . (2.18)
a

where C is some normalizing factor, and the notation Dx(t) tells us that we must in-
tegrate over every possible path from (xa , ta ) to (xb , tb ). The chosen path will in turn
affect the value of S[a, b], which is path-dependent. Equation (2.18) is Feynman’s path
integral. It is similar to Dirac’s equation (2.15), except it involves the classical action,
and it has been normalized to give the probability amplitude for the time evolution of
our system.

Let us discuss some properties of the kernel in equation (2.18). We assume there is
some intermediate time tc , with ta < tc < tb , when the system is at some point xc . We
want to know how to express the amplitude for several events occurring in succession.
Probabilities for successive events multiply, and it follows that the associated probability
amplitudes also multiply. We can write

10
 2 Important Concepts in Quantum Mechanics

ˆ ∞
K(b, a) = K(b, c)K(c, a)dxc . (2.19)
−∞

Note that we must integrate over all possible positions xc since the system may go
through any possible point at time tc .

xc tc

Time
• xb tb
ta < tb < tc

xa ta

Figure 4: Evolution of a quantum mechanical system through time, from an infinite

set of points xa at time ta to a final point (xb , tb ). The purple cone indicates that all
states at the earlier time ta influences the probability to arrive at (xb , tb ). However, no
points at the later time tc influences the probability to arrive at (xb , tb ), by causality.

Also, we want the kernel to be 0 for ta > tb . This is a statement about causality: events
prior to tb should affect the system at tb , but later events should not. We put this
restriction on the kernel in general

K(b, a) = 0 if ta > tb . (2.20)

One more thing: how is the kernel K(b, a) related to the wave function of a system,
ψ(x, t)? The absolute value squared of the wave function is the probability for the
system to be in some state, much like the absolute value squared of the kernel. The
difference between the two is that a kernel also carries information about some previous
point in time; it is the amplitude that a system is in some state at tb given that it was in
some previous state at ta . The wave function carries only information about the system
at one specific time. We give some intuition for the relation between the two by the
following relation

ˆ ∞
ψ(xb , tb ) = K(b, a)ψ(xa , ta )dxa . (2.21)
−∞

11
 3 The Path Integral in Quantum Mechanics

On the right hand side, we have the probability amplitude ψ(xa , ta ) for the system to be
at the initial point (xa , ta ). We integrate this initial state against the amplitude K(a, b)
to go from point (xa , ta ) to (xb , tb ) and obtain the amplitude ψ(xb , tb ) for the system to
arrive at the final point (xb , tb ).

We can visualize both equation (2.21) and the restriction (2.20) as in figure (4). We have
a quantum mechanical system which evolves in time. If we want to find the amplitude
for the system to be at (xb , tb ), we consider all possible initial points (xa , ta ) and their
associated amplitude ψ(xa , ta ). We integrate our initial states against the kernel K(b, a),
here visualized by the pink cone. By doing so, we obtain the amplitude for the system to
be at (xb , tb ). No states at a later time tc can contribute to this amplitude by causality,
i.e. K(b, c) = 0. The blue planes are supposed to extend to infinity.

3 The Path Integral in Quantum Mechanics

This section is concerned with the validity of the path integral. In section 3.1 we show
that the path integral behaves as expected in the classical limit, and in section 3.2-3.3
we evaluate the kernel of a free particle in position- and momentum basis. This gives
us a concrete example of how to evaluate a path integral. It also shows that where we
interpret the results, they correspond to our physical reality. Finally, in section 3.5, we
show that the path integral formulation is equivalent to the Schrödinger formulation of
quantum mechanics.

The discussion in section 3 is based on Feynman and Hibbs [11, ch. 2-4].

3.1 The Path Integral in the Classical Limit

It is important to emphasize that some paths give a negligible contribution to the path
integral in equation (2.18). Let’s consider a macroscopic system. This action in the
argument of the exponential function is of the order of magnitude of a classical system.
The action is divided by ℏ, which is of the order of magnitude of a microscopic system.
i
Therefore, each path is weighed by a rapidly oscillating function e ℏ S . When we change
the path slightly, the action changes only slightly. However, the phase of the oscillating
function changes drastically, since the action is again divided by ℏ. Therefore, every small
step along a path gives alternating positive and negative contributions to the amplitude.
In every such point, a neighboring path gives an equal contribution of opposite sign.
Contributions from neighboring paths cancel everywhere but for the path where the action
stays constant when we alter the path slightly. This corresponds to the extremum of the
action, i.e. the classical path. As a result, in the classical limit, equation (2.18) reduces
to some kernel

i
K(b, a) ∝ e ℏ Sc ,

12
 3 The Path Integral in Quantum Mechanics

where Sc is the action for the classical path.

The above reasoning applies also in the classical limit when we take ℏ → 0. In conclu-
sion, the path integral reduces to the classical path in the classical limit, as expected.

In the case of microscopic systems, the difference in the action as we change our path
slightly is of the order of magnitude of microscopic systems, i.e. of the order of magni-
tude of ℏ. Therefore, many paths contribute to equation (2.18). However, the largest
contributions are still given by those paths which minimizes the action. That is, even for
quantum mechanical systems, a longer path contributes a smaller probability amplitude,
since this requires more kinetic energy. A path which passes though a large positive po-
tential also contributes a smaller probability amplitude. For example, the probability is
negligible that a particle moves through a wall in the double-slit experiment. Further
support for this reasoning is given in section 4.2 when we discuss Wick-rotations.

3.2 Evaluation of the Path Integral for a Free Particle

Most of the time, path integrals are difficult to evaluate, and we must use some trick. In
this section, we introduce a general method for evaluating the path integral of a particle
in some potential V . The action is given by

ˆ tb
1
 
S[b, a] = mẋ2 − V (x, t) dt. (3.1)
ta 2

We use the property introduced in equation (2.19) and divide our time interval into not
2 steps, but infinitely many steps. For the sake of computation, we denote the number
of time intervals N , where each interval is of length ϵ. We will later take N → ∞,
and consequently ϵ → 0. For small time intervals, the action can be approximated by
Si [xi , xi−1 , ϵ] = ϵLi xi −xϵ i−1 , xi +x2 i−1 . Here the velocity is approximated by xi −xϵ i−1 and
the position is approximated by xi +x2 i−1 . We write

ˆ ˆ
1 1 1
K(b, a) = lim N ... e ℏ ϵL1 ... e ℏ ϵLN dx1 ... dxN −1 . (3.2)
ϵ→0 A

That is, we approximate the paths between ti and ti + ϵ as straight lines and at each
time ti we integrate over all possible positions for the particle. The approximations
introduced by these finite timesteps will not affect the solution when we eventually let
ϵ → 0. It is also implicit that x0 = xa and xN = xb . A is a normalization factor
associated with each time step and our first task is to find an explicit expression for it.
We start from equation (2.21) and write

ˆ ˆ
∞
1 ∞
i
ψ(x2 , t + ϵ) = K(2, 1)ψ(x1 , t)dx1 = e ℏ ϵL2 ψ(x1 , t)dx1 , (3.3)
−∞ A −∞

13
 3 The Path Integral in Quantum Mechanics

where ϵ is a small time interval. We use equation (3.1) to expand upon equation (3.3)

ˆ
1 ∞
i m(x2 − x1 )2 x2 + x1
" #
i

ψ(x2 , t + ϵ) = exp − ϵV ,t ψ(x1 , t)dx1 . (3.4)
A −∞ ℏ 2ϵ ℏ 2

The first term in the exponent involves a factor 1ϵ . This term will in most cases result
in a quickly oscillating function which forces the integral to zero. The only instance
where this is not the case is when |x2 − x1 | << 1. We therefore make the substitution
x1 − x2 = η and note that x2 is fixed, i.e. dx1 = d(η + x2 ) = dη. We rewrite equation
(3.4) with our new variable η

ˆ
1 ∞
2x + η
!
imη 2 i
  
ψ(x, t + ϵ) = exp exp − ϵV ,t ψ(x + η, t) dη, (3.5)
A −∞ 2ℏϵ ℏ 2

where we have also made the substitution x2 → x. Those paths where η << 1 give the
only notable contributions to the integral. This is intuitive from a physical standpoint,
since we consider a very small timestep and it is unlikely that the particle travels far.
We may expand equation (3.5) as a power series in η and ϵ

ˆ
1 ∞
!
∂ψ imη 2 i

ψ(x, t) + ϵ = exp 1 − ϵV (x, t)
∂t A −∞ 2ℏϵ ℏ
" #
∂ψ(x, t) η 2 ∂ 2 ψ(x, t)
ψ(x, t) + η + dη, (3.6)
∂x 2 ∂x2

where we have kept terms up to order ϵ and η 2 . This is done since the integrand in
2
equation (3.5) involves an oscillating function with a phase mη
2ℏϵ
. The main contribu-
tion
q to the integral comes from a phase up to 2π radians, i.e. when η goes from 0 to
2π 2ℏϵ m
. Thus, those terms which make a significant contribution to the integral must
fulfill η ∼ ϵ.
2

The entirety of equation (3.6) will be used later, in section 3.4. In our search for A,
we consider only the term of order 0 in ϵ and η. We do this because it simplifies the
expression, and it is allowed since the equality in (3.6) must hold for any value of ϵ, also
for ϵ → 0, and consequently η → 0. We obtain

ˆ s
1 ∞
1 2πiℏϵ
" #
imη 2
ψ(x, t) = ψ(x, t) exp dη = {Appendix B.1} = ψ(x, t) . (3.7)
A −∞ 2ℏϵ A m
 1
From equation (3.7), we conclude A = 2πiℏϵ
m
2
.

Our final expression for the path integral of a particle in an arbitrary potential is given
by this normalizing factor together with equation (3.2)

14
 3 The Path Integral in Quantum Mechanics

ˆ ˆ
m(xi−1 − xi )2 xi−1 + xi
N N
" !#
m i
 
2
K(b, a) = lim exp
Y
... − ϵV ,t x1 ... xN −1 .
ϵ→0 2πiℏϵ ℏ 2ϵ 2
i=1
(3.8)
We see now the purpose of splitting our time interval into steps. As long as our potential
consists of at most quadratic terms in x, we are left with a product of Gaussian integrals
as in equation (3.8), and they are often easy to solve.

We cannot proceed further unless we specify the potential. Depending on our choice,
the path integral may become very difficult to evaluate. We choose V = 0 for simplicity.
For a free particle, then, we get from equation (3.8)

 
ˆ ˆ "ˆ
im[(x0 − x1 )2 + (x1 − x2 )2 ] im(x2 − x3 )2
N ! # !
m
 
2
K(b, a) = lim exp dx1 exp
 
...  dx2 
 ...
ϵ→0 2πiℏϵ 
 2ℏϵ 2ℏϵ 
| {z }
I1
| {z }
I2
(3.9)
We denote k = m
2ℏϵ
and evaluate I1

ˆ ∞  
I1 = exp(ik(x20 + x22 )) exp ik(2x21 − 2x1 (x0 + x2 ) dx1 = {Appendix B.3}
−∞
!s
1 π
 1
ik 2
= exp (x0 − x2 )2 . (3.10)
2 2 k
We use this result to calculate I2 from equation (3.9) in a similar way. We obtain

!s
ik 1 π
 
I2 = exp (x0 − x3 )2 . (3.11)
3 3 k
We notice a possible pattern in equation (3.10) and (3.11)

!n
1 2πℏϵ
s !
2
im
In = f (n) = exp (x0 − xn+1 )2 . (3.12)
n+1 m 2(n + 1)ℏϵ
To establish this relation, we must show In+1 = f (n + 1). We know from equation (3.9)
that In+1 is given by

ˆ
im
 
In+1 = In exp (xn − xn+1 )2 dxn . (3.13)
2ℏϵ

15
 3 The Path Integral in Quantum Mechanics

We compute (3.13) explicitly in the same way that we did I1 , I2 , and this gives exactly
In+1 = f (n + 1). Since we proved in equation (3.10) and (3.11) that the formula holds
for n = 1 and n = 2, it must hold also for all n > 2. Finally, we can solve equation (3.9)
with the help of equation (3.12)

1 s
1
N
m m im
   
2 2
K(b, a) = lim IN −1 = lim exp (x0 − xN )2 . (3.14)
ϵ→0 2πiℏϵ ϵ→0 2πiℏϵ N 2ℏN ϵ

We set (x0 , t0 ) = (xa , ta ) and (xN , tN ) = (xb , tb ). We recall also that N = tN −t0
ϵ
= tb −ta
ϵ
.
Using these relations, (3.14) can be written as

!1
im(xb − xa )2
!
m 2
K(b, a) = exp . (3.15)
2πiℏ(tb − ta ) 2ℏ(tb − ta )

Equation (3.15) is the final expression for the kernel of a free particle. It will be studied
further in the next section.

3.3 Momentum Representation of the Path Integral

We know that a state in quantum mechanics can be expressed in both position basis
ψ(x, t) and momentum basis Φ(p, t). We can do the same for kernels. A momentum
representation K̃p (p2 , t2 ; p1 , t1 ) gives the amplitude for a system to have a certain mo-
mentum p2 at some time t2 , given that it had some momentum p1 at an earlier time t1 .
The kernel obeys

ˆ
Φ(p2 , t2 ) = K̃p (p2 , t2 ; p1 , t1 )Φ(p1 , t1 )dp1 , (3.16)

in analogy with (2.21).

However, we want to take it one step further. Since one of our goals is to give a
description of quantum mechanics which is compatible with special relativity, there is a
point in also transforming the variable of time to energy, and obtain Kp (p2 , E2 ; p1 , E1 ).
We can then describe our kernel using either the variables (x, t) or the variables (p, E),
similar to how we group position with time and momentum with energy in special
relativity. We have

ˆ
1
ΦE (p2 , E2 ) = Kp (p2 , E2 ; p1 , E1 )ΦE (p1 , E1 )dp1 , (3.17)
2πℏ

where the factor 2πℏ

1
is inserted by convention. To find an expression which allows us
to move from one representation of the kernel to another, we use the known Fourier
transforms from position basis to momentum basis and back

16
 3 The Path Integral in Quantum Mechanics

ˆ ˆ
1 ipx ipx
ψ(x, t) = e ℏ Φ(p, t)dp, Φ(p, t) = e− ℏ ψ(x, t)dx. (3.18)
2πℏ
The equations in (3.18) along with equations (2.21) and (3.16) give us

¨
ip2 x2 ip1 x1
K̃p (p2 , t2 ; p1 , t1 ) = e− ℏ K(x2 , t2 ; x1 , t1 )e ℏ dx2 dx1 . (3.19)

How would we construct a similar transformation from time to energy? From the
Schrödinger equation, we have ⟨t|Ĥ|E⟩ = ih ∂t
∂
⟨t|E⟩. This expression can be used to
itE
write ⟨t|Ĥ|E⟩ = E ⟨t|E⟩ = iℏ ∂t ⟨t|E⟩ =⇒ ⟨t|E⟩ = Ae− ℏ , where we set A = 2πℏ
∂ 1
.

We find our desired Fourier transforms as

ˆ ˆ
⟨t|Φ⟩ = ⟨t|E⟩ ⟨E|Φ⟩ dE, ⟨E|Φ⟩ = ⟨E|t⟩ ⟨t|Φ⟩ dt,
ˆ ˆ
1 − itE itE
Φ(p, t) = e ℏ ΦE (p, E)dE, ΦE (p, E) = e ℏ Φ(p, t)dt. (3.20)
2πℏ
Equations (3.20) together with (3.16), (3.17) and (3.19) give

˘
ip2 x2 iE2 t2 ip1 x1 iE1 t1
Kp (p2 , E2 ; p1 , E1 ) = e− ℏ e ℏ K(x2 , t2 ; x1 , t1 )e ℏ e− ℏ dt2 dx2 dt1 dx1 .

(3.21)
This is the final expression for the transform between our two basis representations of
the kernel. Now, we would like to obtain the kernel for the free particle in the (p, E)-
basis. This will allow us to better interpret equation (3.15).

We begin by substituting equation (3.15) for K(x2 , t2 ; x1 , t1 ) in equation (3.19)

¨ !1
im(x2 − x1 )2
!
ip2 x2 m 2
ip1 x1
   
K̃p (p2 , t2 ; p1 , t1 ) = exp − exp exp dx1 dx2 .
ℏ 2πiℏ(t2 − t1 ) 2ℏ(t2 − t1 ) ℏ

We use the computation in Appendix B.3 to simplify the above expression and arrive at

!ˆ
ip2 i
 
K̃p (p2 , t2 ; p1 , t1 ) = exp − 1 (t2 − t1 ) exp − (p2 − p1 )x2 dx2
2ℏm ℏ
!
ip2
= 2πℏδ(p2 − p1 )exp − 1 (t2 − t1 ) . (3.22)
2ℏm

17
 3 The Path Integral in Quantum Mechanics

The integral in equation (3.22) evaluates to the delta function, which seems reasonable
i
considering that {e ℏ xpn } is an orthogonal basis. The factor 2πℏ is not so obvious, but
we show that it is necessary below. This part may be skipped by the reader.

We look at Fourier transforms of some arbitrary functions G, F

ˆ ˆ
1 ipx ipx
F (x) = e ℏ G(p)dp, G(p) = e− ℏ F (x)dx.
2πℏ
We use these equations to write

ˆ ˆ ˆ ˆ
1 1
! !
−ipx ip′ x
− ℏi (p−p′ )x
G(p) = e ℏ e ℏ G(p )dp
′ ′
dx = G(p ) ′
e dx dp′ .
2πℏ 2πℏ

Since G and p, p′ are arbitrary, for the equality to hold, the equation in parenthesis must
fulfill

ˆ
i ′
e ℏ (p−p )x dx = 2πℏ δ(p′ − p), (3.23)

which motivates equation (3.22).

We substitute equation (3.22) for K̃p in equation (3.21)

¨ !
iE2 t2 ip2 iE1 t1
   
Kp (p2 , E2 ; p1 , E1 ) = 2πℏ δ(p2 − p1 ) exp exp − 1 (t2 − t1 ) exp − dt2 dt1 .
ℏ 2ℏm ℏ
(3.24)

To evaluate this, we make a change of variables t2 = t1 + τ . We recall that for t1 > t2 ,

the kernel must equal zero. Therefore, t1 runs from −∞ to ∞ and τ runs from 0 to ∞.
We rewrite equation (3.24) using this change of variables

ˆ ∞ ˆ ∞ p2
i i 1 )τ
Kp (p2 , E2 ; p1 , E1 ) = 2πℏδ(p2 − p1 ) e ℏ
(E2 −E1 )t1
dt1 e ℏ
(E2 − 2m
dτ
−∞ 0
 
i
= lim(2πℏ)2 δ(p2 − p1 )δ(E2 − E1 )  p21
. (3.25)
ϵ→0 1
E
ℏ 2
− 2m
+ iϵ

The factor 2πℏδ(E2 − E1 ) comes from the integral over t1 , in analogy with equation
(3.23). The integral over τ is more difficult to evaluate since it doesn’t converge. We
p21
 
must make a slight rotation into the complex plane ω → ω + iϵ, where ω = ℏ E2 − 2m .
1

18
 3 The Path Integral in Quantum Mechanics

ϵ is some positive real-valued variable which we will later send to zero to return to our
original problem. This method allows us to evaluate the integral

ˆ ˆ
∞ ∞
1 i
e iωτ
dτ = {ω → ω + iϵ} = lim e(iω−ϵ)τ dτ = lim − = lim . (3.26)
0
ϵ→0
0
ϵ→0 iω − ϵ ϵ→0 ω + iϵ
However, we cannot immediately take the limit ϵ → 0, since it would introduce a factor
1
ω
in equation (3.25) and the kernel would diverge at ω = 0. This is unfavourable, since
the kernel is used as a propagator and we may want to integrate over it with respect to
all values of p and E, and therefore all values of ω. We could imagine setting ϵ = 0 and
taking the principle part of any integral which includes ω = 0. However, this doesn’t
work for a different reason: when we take the inverse of the transform in (3.25), we want
to get back to the time-representation of our kernel. Such an inverse transform involves
a factor

ˆ ∞
i
e−iwτ dω. (3.27)
−∞ ω + iϵ

We said earlier that by causality, our kernel should evaluate to zero when t1 > t2 .
Equation (3.27) evaluates to zero when τ = t2 − t1 < 0 only if ϵ is finite and positive (see
Appendix B.4). Because of this, we cannot immediately set ϵ = 0, but we can rewrite
equation (3.26) as

i i(ω − iϵ) iω ϵ
= 2 = 2 + 2 . (3.28)
ω + iϵ ω +ϵ 2 ω +ϵ 2 ω + ϵ2
As ϵ → 0, we take the first term to be ωi . Whenever we integrate over this term, it is
implied that we take the principal part of the integral. As ϵ → 0, the second term goes
to zero everywhere except at ω = 0. Thus we write the second term cδ(ω) for some
constant c. We find this constant by integration

ˆ ˆ ˆ
∞ ∞ ∞
1
∞
ϵ ω ω
   
c= cδ(ω)dω = dω = =u = du = arctan = π.
−∞ −∞ ω +ϵ
2 2 ϵ −∞ u +1
2 ϵ −∞

We conclude that the second term in equation (3.28) can be written πδ(ω). The whole
expression becomes

i i
 
lim = P.P. + πδ(ω). (3.29)
ϵ→0 ω + iϵ ω
However, we are not satisfied with this expression until we know that it preserves causal-
ity. In other words, the equation integrated against e−iωτ must evaluate to zero when
τ < 0. It turns our that this is the case since the second term integrated against e−iωτ

19
 3 The Path Integral in Quantum Mechanics

evaluates to π and the first term integrated against e−iωτ when τ < 0 evaluates to −π ,
see Appendix B.5.

We rewrite our kernel using equation (3.25) and (3.29)

(2π)2 ℏ3 iδ(p2 − p1 )δ(E2 − E1 )

Kp (p2 , E2 ; p1 , E1 ) = lim p21
ϵ→0
E1 − 2m
+ iϵ
   !
i p2
= 4π 2 ℏ3 δ(p2 − p1 )δ(E2 − E1 ) P.P.   + πδ E1 − 1  .
E1 −
p21 2m
2m
(3.30)

How do we make sense of equation (3.30)? First of all, we have two delta functions in
the expression. We remember that we are considering a free particle in this case, and it
is reasonable that the energy and the momentum of the particle remain unchanged as it
moves through space with no force acting on it.

We can also obtain a better understanding of the term in brackets. We look back at the
original expression for the kernel of a free particle in the (p, t)-basis, equation (3.22). It
p2
is an oscillating function in τ with frequency 2m1 , defined only for τ = t2 − t1 > 0. If we
plot the real or imaginary part of K̃p as a function of τ it should look something like in
figure (5).

Amplitude

K̃p

Figure 5: The kernel K̃p of a free particle plotted against the time difference between
the final- and initial state. The kernel is zero for τ < 0 and for τ ≥ 0 it oscillates with
constant frequency.

To go from the (p, t)-representation of the kernel to the (p, E)-representation, we per-
formed a Fourier transform of the kernel in equation (3.22) with respect to time. The
frequencies included in the Fourier transform were the energies of our system. Because

20
 3 The Path Integral in Quantum Mechanics

the kernel has a jump at τ = 0, we must include all frequencies to recreate the curve.
As the time difference τ increases, one frequency/energy begins to dominate. This fre-
p2
quency corresponds to 2m , see equation (3.22). The above is intuitive also from a physical
standpoint: when we treat energy as its own variable, independent from momentum, the
p2
energy doesn’t necessarily correspond to the familiar expression 2m . However, we can
measure the energy more accurately as time passes (from the Heisenberg uncertainty
p2
principle), and it will in time approach E = 2m .
!
This behaviour is implicit in equation (3.30) where the term P.P. i
p2
accounts for
1
E1 − 2m
p21
 
the behaviour at τ = 0 and the term δ E1 − 2m
accounts for the behaviour at times
τ > 0.

3.4 The Path Integral in Relation to the Schrödinger Equation

We would like to prove that our path integral is equivalent to the Schrödinger equation
for an infinitesimal time interval. We can do this for some arbitrary Lagrangian which
is quadratic in ẋ. One might think this is a weakness in our proof, but most physical
systems fit under this desciption. Also, we can read in Brown and Feynman [1, p.58]
about considering only Lagrangians which are quadratic in ẋ:
"This is not a limitation, however, as it includes all the cases for which
the Schroedinger equation has been experimentally verified."
We return to equation (3.6), which was derived from a Lagrangian of the required form.
We evaluate each term on the right hand side which is at most of first order in ϵ or
second order in η. The lowest order term has already been evaluated, in equation (3.7)

ˆ
1 ∞
" #
imη 2
ψ(x, t) exp dη = ψ(x, t), (3.31)
A −∞ 2ℏϵ
 1
where we set A = 2πiℏϵ
m
2
. The term of first order in η evaluates to

ˆ
1 ∂ψ(x, t) ∞
" #
imη 2
exp ηdη = 0, (3.32)
A ∂x −∞ 2ℏϵ
since the integrand is an odd function. The two remaining terms evaluate to

ˆ
1 ∂ 2 ψ(x, t) ∞
" #
imη 2 2 iℏϵ ∂ 2 ψ(x, t)
exp η dη = {Appendix B.2} = . (3.33)
2A ∂x2 −∞ 2ℏϵ 2m ∂x2

ˆ ∞
" #
iϵ imη 2 i
− V (x, t)ψ(x, t) exp dη = {Appendix B.1} = − ϵV (x, t)ψ. (3.34)
Aℏ −∞ 2ℏϵ ℏ

21
 3 The Path Integral in Quantum Mechanics

We rewrite equation (3.6) using equation (3.31)-(3.34)

ψ iℏϵ ∂ 2 ψ i
ψ+ϵ =ψ+ − ϵV (x, t)ψ,
∂t 2m ∂x 2 ℏ
2 2
∂ψ ℏ ∂ ψ
iℏ =− + V (x, t)ψ. (3.35)
∂t 2m ∂x2

Equation (3.35) is the Schrödinger equation and we have proven equivalence between
the path integral and the Schrödinger equation for an infinitesimal time interval. This
proof was first given by Feynman in 1942 [10] [1]. Since the Schrödinger equation was
a well-established description of quantum mechanics at the time, it established also the
path integral as a valid description of quantum mechanics. Further, Feynman argued as
Dirac did in his article [6], see section 2.1. That is, to obtain a finite transformation, we
integrate over each possible path between the initial- and final state of our system, and
apply an infinitesimal path integral at each point. The resulting probability amplitude
is given by the path integral in equation (2.18).

We would like to discuss one more aspect of the Schrödinger equation with respect to
the path integral. We remember that the kernel is a special case of the wave function.
Therefore, it should satisfy the Schrödinger equation

∂K(b, a) ℏ2 ∂ 2 K(b, a)
iℏ =− + V (x, t)K(b, a). (3.36)
∂t 2m ∂x2
However, this equation is not complete: the kernel is nonzero only for ta < tb , from
equation (2.20). At ta = tb , we have a discontinuity in the kernel, and equation (3.36)
fails to describe this behaviour. In order to examine what happens to K(a, b) as ta = tb ,
we use equation (2.21) and take the limit as tb → ta

ˆ ∞
lim ψ(xb , tb ) = ψ(xb , ta ) = lim K(b, a)ψ(xa , ta )dxa .
tb →ta tb →ta
−∞

Thus we get lim K(b, a) = δ(xa − xb ).

tb →ta
We can write equation (3.36) as

∂K(b, a) iℏ ∂ 2 K(b, a) i
= − V (x, t)K(b, a) + δ(xa − xb )δ(ta − tb ). (3.37)
∂t 2m ∂x 2 h
If we integrate the above function over an infinitesimal time interval around ta , all terms
in the right hand side go to zero except for δ(xa − xb ), and we obtain the desired ex-
pression for the kernel at ta = tb . If we also impose K(b, a) = 0 when ta < tb , equation
(3.37) uniquely defines the kernel. From Appendix C, we see that this equation tells us
that the kernel is a Green’s function for the Schrödinger equation.

22
 4 Important Concepts in QFT

Green’s functions have the property that they propagate the effect of some "source" via
integration. The kernel acts as a Green’s function in that it propagates the effect of
some initial state to some final state, under a given potential. Green’s functions also
obey the reciprocity relation G(x2 , t2 ; x1 , t1 ) = G(x1 , t2 ; x2 , t1 ), see again Appendix C.
This means that a source at the point x1 has some influence on a state at point x2 and
this influence is the same if we interchange the position of the source and the response.

4 Important Concepts in QFT

The path integral rarely simplifies computations in non-relativistic quantum mechanics.
We saw an example of this in the complicated computation of the free particle kernel in
section 3.2. It would be nice to see some applications where the path integral is central
and useful, and we therefore turn to QFT. In this section, we introduce some concepts
and follow mainly Hori et al. [17, ch. 8-9]. We do not attempt mathematical rigor, but
rather an intuitive understanding.

4.1 Manifolds and Riemannian Metric in Different Dimensions

The first important concept is a manifold, which is a topological space. A topological
space, in turn, is a mathematical space where we can define convergence, continuity
and connectedness. A manifold is a topological space that locally resembles Euclidian
space, Rn , which is the space that we are used to from classical mechanics. We note in
particular that Rn involves a flat spacetime, and that a circle is a simple example of a
manifold: if we zoom in enough, it looks like a straight line, or R1 . The outline of a
triangle is not a manifold: no matter how much we zoom in on the neighborhood of a
corner, it will never resemble a straight line.

When we formulate a QFT, the starting point is our choice of manifold. Most of the time,
we choose to work with a Riemannian manifold with a smooth metric on it. A metric is
something which allows us to define angles and distances. A Riemannian manifold is just
a space with a Riemannian metric, which gives a very intuitive definition of a distance:
for a differentiable path we define a distance at every point in space as the length of an
infinitesimal tangent vector at that point. This distance in spacetime can be written ds,
with ds2 = dx2 + dt2 in Euclidean signature, or ds2 = dx2 − dt2 in Minkowski signature.
Note that we work in natural units, i.e. c = 1. To obtain the length of some path, we
walk an infinitesimal step along the tangent vector at each point of the path, as in

ˆ ˆ
L[γ] = ds = |γ̇(t)|dt,

where L[γ] denotes the length of γ, and γ is a differentiable path. The distance between
two points in our manifold is defined as the minimum of L[γ] if γ runs between the points
[27]. The Riemannian metric is applicable not just to 3 dimensions, so it generalizes our

23
 4 Important Concepts in QFT

familiar notion of distance.

Throughout this report, we have used Minkowski signature, since Minkowski space is
the common choice for quantum mechanics. Now we want to move from Minkowski
signature to Euclidean signature, since we will be considering QFT:s in Euclidean space,
or Euclidean QFT:s. We can do so with a Wick rotation, i.e. a rotation into imaginary
time t → it. This rotation clearly transforms ds2 in the desired way. A Wick rotation
also results in a change in our path integral

ˆ ˆ
i
e ℏ
S
Dx → e−S Dx, (4.1)

where we have switched to natural units for convenience.

It might be easier to evaluate path integrals in Euclidean space. In fact, the Wick ro-
tation is very similar to our previous tricks for evaluating integrals in this report, see
Appendix B. In the appendix, we use the Cauchy integral theorem to change the interval
of integration from the real line to some contour in the complex plane. This allows us
to go from an oscillating function to an exponentially decreasing one.

We stop briefly to comment on the nature of path integrals. It is obvious that the
Euclidean signature path integral in equation (4.1) receives contributions mainly from
those paths associated with a smaller value of the action. If the physics of our system
are to remain the same under a change of coordinates, the same must hold for the path
integral in Minkowski space. This further motivates our discussion in section 3.1.

Now to the dimension of our manifold. We deal mostly with 0-dimensional QFT in
this report. Our manifold is then just a point, and there is no evolution in time and
no spatial directions to move along. Also, we consider scalars rather than functions.
0-dimensional QFT:s are difficult to visualize and it is much more common to work in
higher dimensions. However, many concepts can be easily introduced in 0 dimensions
and then generalized to higher dimensions. That is what we aim to do here. Our action
in 0 dimensions is given by

S(ϕi ),

where ϕi are scalars.

In 1-dimensional QFT, we consider one degree of freedom, often taken to be time.

Another word for 1-dimensional QFT is quantum mechanics, where our manifolds cor-
respond to the worldlines of particles. The action becomes

ˆ
S[L(ϕi (t), t)] = L(ϕi (t), t)dt,

24
 4 Important Concepts in QFT

where ϕi are functions of time.

In higher dimensional QFT:s, we integrate over fields. For example in 4-dimensional

QFT, the action can be written

˘
S[L(ϕi (x, y, z, t), t)] = L(ϕi (x, y, z, t), t)dxdydzdt,

where ϕi are fields. At each given time they take on a value for every point in 3-
dimensional space.

There exist many fields, where the most familiar are bosonic fields and fermionic fields.
Excitations of bosonic fields are bosonic particles and excitations of fermionic fields are
fermionic particles. There is some intuition to this, since an excitation of a field means
that energy is added to the field. Energy can only exist in quanta when we’re dealing
with quantum theories. So, we can never add an infinitesimal amount of energy to a
quantum field, only a finite amount: a particle, which moves around as the field’s energy
distribution changes with time.

4.2 The Partition Function and Correlation Functions

We introduce the partition function

ˆ
Z[α, ϵ] := Dϕe−S[ϕ,α,ϵ] , (4.2)

which is just a path integral in Euclidean space. In the above expression, we have in-
cluded only one field ϕ. The notation in the exponent which includes variables α, ϵ will
be useful in the following sections. α indicates the presence of a mass term, which can be
thought of as a kinetic energy term. ϵ indicates the presence of an additional term, e.g.
an interaction between fields, or a source. The partition function can be normalized as
Z[α,ϵ]
Z[α,0]
, where Z[α, 0] is the partition function for the corresponding free theory, where we
include only the mass term. This notation will become more intuitive in the following
sections.

Equation (4.2) is very similar to the partition function in statistical mechanics, although
it looks different at first glance. We discuss this relationship more in section 6.3.

The partition function often encodes all information about a system, as it is the generator
of correlation functions. The correlation function of some function f (ψ) is defined as

ˆ
⟨f (ϕ)⟩ := Dϕf (ϕ)e−S[ϕ,α] . (4.3)

25
 4 Important Concepts in QFT

Equation (4.3), if normalized, is an expectation value averaged over all fields in a theory
(in our case, only one field). The factor e−S /Z[α, 0] is a probability density.

Often it is useful to express the correlation function as a derivative of the partition

function. To achieve this, we add a term to the action S → S − af (ψ), take the
derivative of the partition function w.r.t a and set it equal to zero

ˆ
∂Z[α, a]
Z[α, a] = dϕe−S[ϕ,α]+af (ϕ) =⇒ ⟨f (ϕ)⟩ = . (4.4)
∂a a=0

4.3 Grassmann Variables

We construct wave functions in quantum mechanics to obey the Pauli exclusion principle,
i.e. two fermions cannot coexist in the same state. Similarly in QFT, two fermionic
fields cannot coexist in the same state. Say we would like to express a path integral
in a theory involving fermionic fields. Such an integral involves fermionic fields in the
argument of an exponential function. The exponential function can in turn be expanded
in a power series, where the n:th term involves n − 1 factors of the same fermionic field.
Thus all terms except the first two are ill-defined. In order to solve this problem and
express a path integral involving fermionic fields, we introduce a new mathematical tool:
Grassmann variables. They are denoted θn and are also called fermionic variables/fields.
They exist alongside bosonic variables/fields, ψm , which behave like ordinary variables.
Grassmann variables, on the other hand, are anti-commuting

θa θb = −θb θa , (4.5)

and obey strange integration rules

ˆ
dθ1 dθ2 ... dθn = 0, (4.6)

ˆ
θ1 θ2 ... θn dθ1 dθ2 ... dθn = 1, (4.7)

where the ordering of the variables in equation (4.7) is important. The value of the
integral picks up a factor (-1) for each transposition of two neighboring fields θi , θi+1 .
Removing one or more of the fields θa from equation (4.7) means again that the integral
evaluates to zero.

We note that equation (4.5) implies (θa )2 = 0. Equation (4.5) also implies that a pair
of Grassmann variables obey the same commutator relationship as a bosonic variable

θc (θa θb ) = −θa θc θb = (θa θb )θc .

26
 5 The Path Integral in 0-Dimensional QFT

The above formalism solves our issue with path integrals involving fermionic fields, since
a power series in Grassmann variables will terminate. To demonstrate this, we consider
a 0-dimensional QFT with only two fermionic fields and one bosonic field, with the
action S[ψ, θ1 , θ2 ] = f (ψ) − g(ψ)θ1 θ2 . f and g are arbitrary functions. In general, we
expect the action to behave like an ordinary variable, i.e. we don’t want it to act like
a fermionic variable. Therefore, each term in an action function must include an even
number of fermionic fields. By this reasoning, the above expression for S[ψ, θ1 , θ2 ] is
the most general action for our chosen set of fields. Any term involving only one of the
fermionic fields would be fermionic itself, and any term of higher order than two in the
fermionic fields would equal zero, since (θa )2 = 0.

The path integral resulting from this action involves a term ef (ψ)θ1 θ2 , which can be
expanded in a Taylor series like any ordinary function, since the exponent is bosonic.
The resulting expansion terminates, and can be written

∞
(f (ψ)θ1 θ2 )n
ef (ψ)θ1 θ2 = = 1 + f (ψ)θ1 θ2 .
X

n=0 n!

5 The Path Integral in 0-Dimensional QFT

In this section, we explore some basic applications of the partition function in 0-dimensional
QFT; in particular, how to simplify calculations of partition functions and correlation
functions using Feynman diagrams and supersymmetry. In 0 dimensions, we do not con-
sider each possible path through spacetime, since there is no spacetime to move through.
Rather, the partition function in equation (4.2) reduces to a Lebesgue integral over the
real line. The discussion in section 5 follows Hori et al. [17, ch. 9].

5.1 Feynman Diagrams

We consider a free theory, since it is easy to work with. This means that our action
involves only terms which are quadratic in the fields. A quadratic term is called a
mass term, analogous to a kinetic energy term in higher dimensional QFT - although
in 0 dimensions, we have no actual kinetic energy since we have no time evolution. A
term which is more than quadratic in the fields corresponds to an interaction between
fields. We may introduce weak interactions in our free theory as perturbations, or small
higher-order terms. We consider an example

α 2
S(ψ, α, ϵ) = ψ + iϵψ 3 , (5.1)
2
The perturbation consists of the second term, where ϵ is a small, bosonic variable. We
also note

27
 5 The Path Integral in 0-Dimensional QFT

ˆ s
∞
−α ψ2 2π
Z(α, 0) = dψe 2 = , (5.2)
−∞ α

so that we can use this factor to normalize our partition function later on.

We write out the partition function associated with the action in equation (5.1) and
choose to normalize it. Since ϵ is small, we also make an expansion in ϵ

ˆ ˆ
1 −α ψ 2 −iϵψ 3 1 α 2
∞
(−iϵψ 3 )m
Z(α, ϵ) = = dψe− 2 ψ (5.3)
X
dψe2
Z(α, 0) Z(α, 0) m=0 m!
∞
(−iϵ)m
The above expression can be written Z(α, ϵ) = 1
⟨(ψ 3 )m ⟩ϵ=0 , where the
P
Z(α,0) m!
m=0
subscript signifies that these correlation functions are evaluated with ϵ set to zero. To
compute these correlation functions, we use the method introduced in equation (4.4),
with a modified partition function

ˆ
1 α 2 +jψ
Z(α, j) = dψe− 2 ψ . (5.4)
Z(α, 0)

We see that n consecutive derivatives of Z(α, j) with respect to j gives us the correlation
function ⟨ψ n ⟩ϵ=0 , given that we set j = 0 after performing the derivatives. j is here called
a source. Sometimes our theories contain actual sources, but here it is only a temporary
tool which we eventually set to zero. We solve equation (5.4) explicitly, and use our
expression for the normalization factor in equation (5.2)

ˆ
1 α 2 +jψ j2
Z(α, j) = dψe− 2 ψ = e 2α . (5.5)
Z(α, 0)

The expression (5.5) can in turn be used to calculate the correlation functions ⟨ψ n ⟩ϵ=0

  n
∂ n Z(α, j) c 1 2 if n is even,
⟨ψ n ⟩ϵ=0 = = α (5.6)
∂ nj j=0 0 otherwise,

where c is some constant. This result follows directly from equation (5.5), since the
first derivative of Z(α, j) brings down a factor αj from the exponent. Each consecutive
derivative acts either on the exponential function, bringing down an additional factor
of αj , or it acts on one of the j:s in front of the exponential function. Since we end
our calculation by setting j = 0, the only terms which survive are the ones where an
equal amount of derivatives have acted on the exponential as on the factors in front, so
that there are no factors of j left. How do we actually calculate c in equation (5.6)?
We show below how to do it for n = 6, i.e. the term with m = 2 in the expansion in

28
 5 The Path Integral in 0-Dimensional QFT

equation (5.3). The term with m = 1 is proportional to ⟨ψ 3 ⟩ϵ=0 , which is zero according
to equation (5.6), so there is no need to evaluate it.

What the computation comes down to is to find all possible ways of combining the
derivatives in equation (5.6) into pairs. One way to do it is to label each ψ by some
number, i.e. we want to calculate ⟨ψ1 ψ2 ψ3 ψ4 ψ5 ψ6 ⟩ϵ=0 . We let each field ψi correspond
to a source ji . This numbering will make more physical sense in section 6.1 when we
discuss 1-dimensional QFT. Here, it is only a mathematical tool.

We start easy by calculating the correlation function for the first two fields

∂ 2 Z(α, ji ) 1
! " #
∂ j1 ji2 δj1 ,j2 ji2 j1 j2 ji2
⟨ψ1 ψ2 ⟩ϵ=0 = = e 2α = e + 2 e 2α
2α = . (5.7)
∂j2 ∂j1 j=0
∂j2 α j=0
α α j=0
α

We can visualize each factor of ψi in the correlation function as a vertex with one
"leg" emanating from it (corresponding to a derivative). We pair up these legs, where
each such pairing produces a propagator. Each propagator corresponds to a δ in the
final expression for our correlation function, see equation (5.7), and each propagator
contributes a factor α1 to its value. Finally, the number of possible ways of constructing
these diagrams introduces a numerical factor. In the case of equation (5.7), we have two
instances of ψi , so the pairing up of legs (or derivatives) can be done in only one way, as
in figure (6). There is also only one propagator. This results in a factor α1 as the only
contribution to the correlation function, in agreement with equation (5.7).

ψ1 • • ψ2

Figure 6: A Feynman diagram involving only one propagator.

For simplicity, we denote δja ,jb = δab from here on. If we carry out the computation in
equation (5.7) again, but with all 6 fields, we obtain

"
δ12 δ34 δ56 ji2 δ12 δ35 δ46 ji2 δ12 δ45 δ36 ji2 δ14 δ25 δ36 ji2 δ23 δ45 δ16 ji2
⟨ψ1 ψ2 ψ3 ψ4 ψ5 ψ6 ⟩ϵ=0 = e 2α + e 2α + e 2α + e 2α + e 2α
α3 α3 α3 α3 α3

+ 10 identical terms containing all different permutations of δa,b

#
+ terms involving one or more factors of ji
j=0
15
= . (5.8)
α

29
 5 The Path Integral in 0-Dimensional QFT

Instead of performing this calculation, we can visualize each factor of ψ 3 in ⟨(ψ 3 )n ⟩ as

a vertex with 3 legs emanating from it (each one corresponding to a derivative). The
expression in equation (5.8) corresponds to two factors of ψ 3 . The different possible
Feynman diagrams in this case are given in figure (7). The legs (1,2,3) can be connected
to the legs (4,5,6) to make two different shapes, A and B. To make the shape in A, we
take one of the blue legs, say 1, and connect it to one of the pink legs. There are three
ways to do this. Then, we pair up leg 2 with one of the remaining pink legs. In total, we
can construct diagram A in (3 · 2 · 1 = 3!) different ways. To construct diagram B, we
must choose which one of the 3 blue legs connects to one of the pink legs. There are then
3 different pink legs to connect this blue leg to. Finally, there is only one choice of how
to connect the remaining 4 legs. The total amount of combinations that form diagram
B is (32 ). In conclusion, diagram A and B contribute a total amount α1 (3! + 32 ) = 15 α
to
the correlation function, which is consistent with our explicit calculation (5.8).

1 2 4 5

• + •

3 6
=

A: • •

or

B: • •

Figure 7: Feynman diagrams in a vacuum state.

Since we have no sources in our original problem, equation (5.1), we are effectively in a
vacuum. If we read our Feynman diagrams in figure (7) from left to right and let this
be the direction of time, we see that they are consistent with the vacuum interpretation:
our initial- and final state consists of nothing. Fields are simply created from the vac-
uum and then annihilate.

We conclude from equations (5.3), (5.2) and (5.8) that our normalized partition function
is given by

30
 5 The Path Integral in 0-Dimensional QFT

(−iϵ)2 1
 3
Z(α, ϵ) = 1 + 15 + higher order corrections.
2 α
This was a very simple example where the use of Feynman diagrams was not neces-
sary for computation; we were able to find the terms in our perturbation expansion
directly. This is often difficult. Instead, a deeper analysis of a given perturbed action
gives us all possible Feynman diagrams, and we can then reconstruct the perturbation
expansion from our Feynman diagrams. From here on, we focus only on the latter step,
i.e. how to translate between Feynman diagrams and terms in a perturbation expansion.

Let us briefly analyze the action in equation (5.1). The mass term α2 ψ 2 is quadratic in
the field. The power series which would result from expanding such a term is represented
by diagrams with two legs per vertex. This produces Feynman diagrams like the one in
figure (8).

• • • • •

Figure 8: Feynman diagrams involving only one propagator per vertex.

This is clearly a field which propagates without interacting with other fields, so the mass
term is analogous to a kinetic term, as we discussed in the beginning of this section.
The term iϵψ 3 is instead a perturbation term, which represent self-interactions in the
field ψ, shown visually in figure (7).

From the above discussion, we may draw some general conclusions, so that in the future
we can more easily interpret Feynman diagrams as terms in a perturbation expansion.
To make such an analysis, we must also examine the action which describes our system.
For example, a mass term of the form α2 ψ 2 in the action implies

• Each factor α1 in the perturbation expansion corresponds to a propagator of ψ,

represented by a line.

The term iϵψ 3 implies

• Each factor ϵ in the perturbation expansion corresponds to a vertex. Each factor

of ψ corresponds to a leg emanating from the vertex.

If we have a term jψ in our action, this corresponds to a system with sources, where the
field ψ interacts once with each source. By the general logic of this section, we conclude

• Each factor j in the perturbation expansion corresponds to an external vertex (a

vertex connected to only one propagator of ψ), with a circle at the vertex indicating
a source.

31
 5 The Path Integral in 0-Dimensional QFT

Numerical factors which arise from all the possible ways of constructing a Feynman dia-
gram are called symmetry factors. We saw an example of such a factor in equation (5.8).

Computation of the Dynamics of a System with and without

Quantum Corrections

We demonstrate some concepts from the previous sections by analyzing a system using
both classical methods and QFT. By comparing these approaches, we deduce the effect
of quantum corrections.

We consider the action

a 1
S(ψ, ζ) = ψ 2 + ψ 2 ζ + jψ, (5.9)
2 2
where a is a constant, j is a source, ψ is a dynamic field and ζ is a small, non-dynamic
field. Dynamic fields are those which describe the evolution of a system. There is no
term in the action which is quadratic in ζ, so this field cannot propagate; only perturb
our system.

We use the action in (5.9) to find the classical equations of motion of ψ and the classical
free energy. We then use the path integral to evaluate the QFT partition function and
free energy. We use this free energy to find the one-particle irreducible (or 1PI) quantum
effective action, which is the quantum analogue of the classical action. We then compute
the equations of motion of ψ again, this time including quantum mechanical corrections.

1. Calculating the classical equation of motion of ψ.

In the classical case, a system takes the path which minimizes the action, i.e. the
equations of motion are found by evaluating δS = 0. In our 0-dimensional case, we have
no time variable to sum over, so the variation of the action simply gives

∂S j
δS = δψ = (aψ + ψζ + j)δψ = 0 =⇒ ψcl (j, ζ) = − .
∂ψ a+ζ
2. Calculating the classical free energy Fcl .

The classical free energy is given by a Legendre transform of the action, Fcl (j, ζ) =
[jψ − S(ψ, ζ)]ψ=ψcl . In our case

a 1 j2
 
Fcl (j, ζ) = − ψ 2 − ψ 2 ζ =− . (5.10)
2 2 ψ=ψcl 2(a + ζ)

32
 5 The Path Integral in 0-Dimensional QFT

3. Expanding Fcl and interpreting the terms as tree diagrams.

We expand our expression (5.10) in powers of ζ, using the power expansion 1

1+x
=
1 − x + x2 − x3 + ...

!
j2 ζ ζ2 ζ3
Fcl = − 1 − + 2 − 3 + ... . (5.11)
2a a a a

We draw the corresponding Feynman diagrams in figure (9). The term ψ 2 ζ represents
an interaction between two different fields. According to the rules in section 5.1, we
represent the interaction as tri-valent vertices. Solid lines represent propagators of ψ,
corresponding to a factor a1 in the perturbation terms. A dashed line represents a per-
turbation from the non-dynamical field ζ, corresponding to a factor ζ. Since ζ cannot
propagate, there are no internal dashed lines. Finally, the circles represent sources and
can only exist as external vertices connected by solid lines since they interact only once
with ψ. These circles correspond to a factor j.

Fcl = + • + • • + • • • + ...

= − j2 + j 2ζ − j 2ζ 2 + j 2ζ 3 − ...
2a 2a2 2a3 2a4

Figure 9: Tree diagrams corresponding to each term in the power expansion of Fcl .

4. Computing the partition function and the free energy.

Now we move to QFT. The partition function in Euclidean signature is given by

ˆ ˆ
∞
a 1 ∞
a 1
      
Z[j, ζ] = exp − ψ 2 + ψ 2 ζ + jψ dψ = exp − + ζ ψ 2 − jψ dψ
−∞ 2 2 −∞ 2 2
2π
s !
j2
= exp . (5.12)
a+ζ 2(a + ζ)
The free energy is given by F (j, ζ) = −logZ(j, ζ). We use this expression and equation
(5.12) to obtain

1 2π
!
j2
F (j, ζ) = − log − . (5.13)
2 a+ζ 2(a + ζ)
We conclude from a comparison between (5.10) and (5.13) that the free energy obtained
from the QFT partition function includes one extra term compared to the classical free

33
 5 The Path Integral in 0-Dimensional QFT

energy.

5. Expanding the free energy in a power series and interpreting the

terms as Feynman diagrams.
2 3
We use equation (5.13) along with the power expansion log(1 + x) = x − x2 + x3 − ... to
obtain

1 2π 1
! !
ζ ζ2 ζ3 ζ4 j2 ζ ζ2 ζ3
 
F (j, ζ) = − log + − + − + ... − 1 − 2 + 3 − 4 + ... .
2 a 2 a 2a2 3a3 4a4 2 a a a
| {z } | {z }
A B
(5.14)

The first term in equation (5.14) is just a constant. The expansion which we call B
is identical to the expression for Fcl in equation (5.11) and is represented by the same
Feynman diagrams as in figure (9). Expansion A is represented as Feynman diagrams
in figure (10).

• • • • • •
F = ... + + + + + ...
• • • •

= ... + ζ − ζ2 + ζ3 − ζ4 + ...
2a 4a2 6a3 8a4

Figure 10: Feynman diagrams corresponding to terms in the power expansion of F .

There are no sources present in these diagrams: only loops with propagators of ψ and
interactions between ψ and ζ. Following from the discussion in section 5.1, these are
vacuum Feynman diagrams. Our conclusion is that the free energy in the quantum the-
ory gets an extra contribution from vacuum fluctuations.

6. Obtaining the 1PI quantum effective action and the equation of mo-
tion of ψ.

Classically, the free energy is given by the Legendre transform of the action, Fcl =
[jψ − S]ψ=ψcl . In QFT, we instead write F = [jψ − Γ]ψ=ψc where Γ(ψ, ζ) is the 1PI
effective action. Γ is the generator of 1PI irreducible Feynman diagrams, in the same
way that the partition function is the generator of correlation functions, as demonstrated
by equation (5.6). 1PI irreducible Feynman diagrams are those which cannot be sepa-
rated into two independent diagrams by disconnecting any one internal line. Γ can also
be minimized to give the equations of motion ψc . Equations of motion are implicitly
classical, since they assume a deterministic time evolution. However, the equations of

34
 5 The Path Integral in 0-Dimensional QFT

motion obtained from Γ are semi-classical, since they include quantum corrections. Such
corrections arise from higher order Feynman diagrams, i.e. loop diagrams, which are by
construction 1PI irreducible.

In this way, Γ is the quantum analogue of the classical action, and ψc is obtained by
considering the variation

∂(jψc ) ∂F
δΓ = 0 = δj + δj
∂j ∂j
∂
=⇒ ψc = F. (5.15)
∂j

We use equation (5.15) and our expression for the free energy (5.13) to obtain

j
ψc (j, ζ) = − . (5.16)
a+ζ

ψc is thus identical to our expression for ψcl (5.10). This is because our expression for
the QFT free energy was identical to the expression for the classical free energy, apart
from one extra term without sources. From equation (5.15), we see that only those terms
which contain sources determine the equations of motion. Along the same line, if we set
j = 0 in equation (5.16), we find that ψc is static.

Our final task is to find the 1PI quantum effective action for our system. We use equation
(5.16) and (5.13) to obtain

1 2π 1 2π 1
! !
j2 a
Γ(ψc , ζ) = − log + = − log + ψc2 + ψc2 ζ. (5.17)
2 a+ζ 2(a + ζ) 2 a+ζ 2 2

Γ in equation (5.17) is similar to

 the classical action in equation (5.3). However, Γ(ψ, ζ)
contains the extra term − 2 log a+ζ
1 2π
, which accounts for quantum corrections. Also, Γ
includes no sources. This is a result of our formalism: we are here interested in the
behaviour of ψ rather than in the sources, and we obtained the above expression free
from sources by performing a suitable Legendre transform.

5.2 Supersymmetry
In section 5.1, we used perturbation theory to evaluate the partition function. In this
section, we examine QFT:s which are constructed to possess supersymmetry. This qual-
ity allows us to evaluate the partition function and other quantities exactly.

To demonstrate supersymmetry, we consider a system of two fermionic fields and one

bosonic field with the action from section 4.3

35
 5 The Path Integral in 0-Dimensional QFT

S(ψ, θ1 , θ2 ) = f (ψ) − θ1 θ2 g(ψ),

This time, we specify the action further, by putting f (ψ) = 12 h′ (ψ)2 and g(ψ) = h′′ (ψ),
where h is a function called the superpotential. The resulting action S[ψ, θ1 , θ2 ] =
h (ψ)2 − θ1 θ2 h′′ (ψ) is invariant under some transformations which exchange bosonic-
1 ′
2
and fermionic fields. Such transformations are called supersymmetry transformations.
In our case, the following is a supersymmetry transformation

δψ = ϵ1 θ1 + ϵ2 θ2 ,
δθ1 = ϵ2 h′ , (5.18)
δθ2 = −ϵ1 h′ .

ϵi are fermionic variables, and they are small, to generate an infinitesimal transformation.
The corresponding transformation of the action is given by

∂S ∂S ∂S
δS = δψ + δθ1 + δθ2
∂ψ ∂θ1 ∂θ2
= (ϵ1 θ1 + ϵ2 θ2 )(h h − θ1 θ2 h′′′ ) + (ϵ2 h′ )(−θ2 h′′ ) + (−ϵ1 h′ )(θ1 h′′ ) = 0,
′ ′′
(5.19)

where the ordering of the terms is important and we have taken into consideration the
rule θa2 = 0 and the commutation relations for fermionic fields. We have also used the
convention that derivatives be taken as ∂θ∂ 2 (θ1 θ2 ) = −θ1 . From equation (5.19), we see
that the action is invariant under the transformation (5.18). One can easily show that
also the measure is invariant. In conclusion, (5.18) generates a supersymmetry and the
partition function is invariant under the transformation.

We demonstrate some useful consequences of supersymmetry in the following sections.

5.3 Deformation Invariance

We begin with an important consequence of supersymmetry: deformation invariance.
In the context of this thesis, it means that our superpotential h(ψ) from section 5.2 can
be altered in ways that leave the partition function unchanged. This simplifies many
calculations.

To prove deformation invariance, we first introduce another property of symmetric

QFT:s, namely that correlation functions evaluate to zero for variations of fields, if the
action is invariant under the variation. In our case, this means that the correlation func-
tion ⟨δγ⟩ = 0 for any function γ(ψ, θ1 , θ2 ), where δ is a supersymmetry transformation.
We prove this as follows

36
 5 The Path Integral in 0-Dimensional QFT

ˆ ˆ
⟨δγ⟩ = δγe−S
dψdθ1 dθ2 = δ(γe−S )dψdθ1 dθ2
ˆ
∂(γe−S ) ∂(γe−S ) ∂(γe−S )
!
= δψ + δθ1 + δθ2 dψdθ1 dθ2 = 0. (5.20)
∂ψ ∂θ1 ∂θ2

The second equality follows from the property δS = 0 =⇒ δ(γe−S ) = δγe−S +γδ(e−S ) =
δγe−S . The final equality in equation (5.20) is a consequence of several things. First,
each term in γ contains at most one factor of each fermionic field. Therefore, the deriva-
tive of γ w.r.t. θ1 or θ2 will eliminiate one of the fermionic fields from each term. By the
rules of Grassman integration, equation (4.6), the integral of the resulting terms goes
−S )
to zero. We are then left with only the last term ∂(γe ∂ψ
δψ, where δψ can be moved
outside the integral and the remaining expression is easily rewritten as a total derivative
in ψ. The integral reduces to the evaluation of (γe−S ) at ±∞. We expect that the
action goes to infinity far away, so also these boundary terms go to zero. The result in
equation (5.20) thus holds as long as the supersymmetry transformation does not change
the behaviour of S at infinity, since this could alter the behaviour of the boundary terms.

We would like to use the property in equation (5.20) to show that ⟨δρ S⟩ = 0, where δρ is
the specific transformation given by a change in the superpotential h → h + σρ, where
ρ = ρ(ψ), and σ is an infinitesimal bosonic parameter. Our end goal is to show that this
transformation of the superpotential does not affect the partition function. Indeed, we
see that ⟨δρ S⟩ = 0 implies invariance of the partition function by the following

ˆ ˆ ˆ
∂ ∂ −S
⟨δρ S⟩ = δρ Se −S
dψdθ1 dθ2 = (S) δρ h e−S dψdθ1 dθ2 = − (e )δρ h dψdθ1 dθ2
∂h ∂h

ˆ !
= − δρ e−S dψdθ1 dθ2 .

Thus ⟨δρ S⟩ = 0 =⇒ δρ Z = 0. We remind ourselves of our expression for S,

S(ψ, θ1 , θ2 ) = 21 h′ (ψ)2 − θ1 θ2 h′′ (ψ). To prove ⟨δρ S⟩ = 0, we begin by writing out S
under the transformation δρ

1
δρ S = S(h) − S(h + σρ) = (σ 2 ρ′2 + 2σh′ ρ′ ) − σρ′′ θ1 θ2 .
2
We assume the term of second order in σ is negligible and write

δρ S = σh′ ρ′ − σρ′′ θ1 θ2 . (5.21)

We consider a suitable function g(ψ, θ1 ) = ρ′ (ψ)θ1 . The variation δg under our super-
symmetry is identical to δρ S. We show this by utilizing the transformation in (5.18)
along with the restriction ϵ1 = ϵ2 = ϵ

37
 5 The Path Integral in 0-Dimensional QFT

δg = δψρ′′ θ1 + δθ1 ρ′ = σρ′ h′ − σρ′′ θ1 θ2 . (5.22)

By equation (5.20), ⟨δg⟩ = 0. Also, a comparison between equation (5.22) and (5.21)
gives ⟨δρ S⟩ = ⟨δg⟩ = 0. We have proven that the partition function is unchanged under
the infinitesimal transformation h → h + σρ. This implies invariance also for h → h + ρ,
since a finite transformation is achieved from many consecutive infinitesimal transforma-
tions. As a reminder, ρ is a bosonic function whose only restriction is that the behaviour
of the action cannot be dominated by ρ at infinity.

We will see a consequence of this deformation invariance in the next section.

5.4 Localization
The idea behind localization is that in some supersymmetric QFT:s, the partition func-
tion receives contributions only from a limited amount of paths (or in 0 dimensions, a
limited amount of points). Contributions from the remaining paths cancel against each
other due to boson-fermion cancellations. As we saw in section 5.1, calculations in QFT
often come down to approximations using perturbation theory. However, when a QFT
allows for localization, we may use this technique to evaluate many quantities exactly.

To show that localization applies in our case, we return again to deformation invariance.
Say our superpotential h is a polynomial. Then the transformation h → h + ρ does not
change the behavior of S at infinity as long as ρ is a polynomial of at most the same
degree as h. We could for example set ρ = βh(ψ), where β is a real constant. This
corresponds to rescaling our superpotential by some constant, h → λh.

If we allow λ to grow, the first term in our action S(ψ, θ1 , θ2 ) = 21 λ2 h′ (ψ)2 − θ1 θ2 λh′′ (ψ)
quickly becomes the dominant term, i.e. we can effectively write S(ψ, θ1 , θ2 ) = 21 λ2 h′ (ψ)2
as λ → ∞. As a consequence, our partition function goes to zero everywhere but at the
points where h′ (ψ) = 0. This happens to correspond to the points where our fermionic
fields are unchanged by the supersymmetry transformation, see (5.18). The statement
holds true for every QFT with supersymmetry, as stated in Hori et al. [17, p. 158]

"This is the localization principle: The path-integral is localized at loci where

the R.H.S of the fermionic transformation under supersymmetry is zero."

Since the partition function is independent of our choice of λ, it will receive contributions
only from the points ψ0 where h′ (ψ0 ) = 0, even if we go back to our original problem
with λ = 1. We do so, and try to find an expression for Z. We consider the case where
h is a polynomial of degree n with n − 1 distinct critical points. Around each of the
points ψ0 , we can expand h and keep only the first two non-zero terms

a0
h(ψ) = h(ψ0 ) + (ψ − ψ0 )2 ,
2

38
 5 The Path Integral in 0-Dimensional QFT

where a0 = h′′ . Our normalized partition function is then of the form

ˆ ˆ
1 X − 21 a20 (ψ−ψ0 )2 +a0 θ1 θ2 1 X 1 2 2
Z= e dψdθ1 dθ2 = √ e− 2 a0 (ψ−ψ0 ) (1 + a0 θ1 θ2 )dψdθ1 dθ2
Z0 ψ0 2π ψ0

ˆ
1 X − 21 a20 (ψ−ψ0 )2 a0 X h′′ (ψ0 )
=√ dψ = = (5.23)
X
a0 e .
2π ψ0 ψ0 |h (ψ0 )|
|a0 | ′′
ψ0

Z0 is the partition function for the corresponding free theory. We see that when we
consider localization, the evaluation of our partition function reduces to counting the
number of critical points of the superpotential. Each root contributes a term ±1, where
the sign depends on the slope of h′ (ψ0 ).

Amplitude Amplitude

h′ (ψ) h′ (ψ)

ψ ψ

Figure 11: Some polynomial h′ (ψ), where h(ψ) is of degree n = 8.

The function has been shifted by the addition of a constant.

Amplitude
Amplitude

h′ (ψ)
h′ (ψ)

ψ
ψ

Figure 12: Some polynomial h′ (ψ), where h(ψ) is of degree n = 7.

The function has been shifted by the addition of a constant.

39
 5 The Path Integral in 0-Dimensional QFT

We can simplify the expression in equation (5.23) even further. To do this, we consider
some deformation of our superpotential again. We set ρ = bψ, and thus h′ → h′ + b,
where b is some real constant. If we let b grow large enough, the number of critical points
h′ (ψ0 ) reduces to one for an even degree n of h, or zero for an odd degree n of h, see
figure (11). In the latter case, our partition function reduces to zero. In the first case,
our partition function reduces to ±1, where the sign is determined by h′′ (ψ) at our only
remaining critical point. This sign is in turn decided by the sign in front of the leading
term in h. We conclude that the evaluation of the partition function is sensitive only
to the nature of the superpotential. If the superpotential is a polynomial, it is sensitive
only to its degree, and the nature of its extremum points.

In other words, the evaluation of the partition function may simplify significantly when
we construct a supersymmetric QFT. As a consequence, the computation of many cor-
relation functions also simplify. We will see an example of this in the next section.

5.5 Landau-Ginzburg Supersymmetry

In this section, we again treat supersymmetry, with the addition that our fields are
complex. We include twice as many fields, (ψ, θ1 , θ2 ) → (ψ, ψ, θ1 , θ2 , θ1 , θ2 ), where the
overline denotes a complex conjugate.

We choose a new action

S(ψ, ψ, θ1 , θ2 , θ1 , θ2 ) = |W ′ |2 − W ′′ θ1 θ2 − W ′′ θ1 θ2 ,

where W = W (ψ) is a complex differentiable function and our new superpotential. We

consider two supersymmetry transformations

δψ = ϵ1 θ1 + ϵ2 θ2 δψ = 0,
δθ1 = ϵ2 W ′ δθ1 = 0, (5.24)
δθ2 = −ϵ1 W ′ δθ2 = 0.

and

δψ = 0 δψ = ϵ1 θ1 + ϵ2 θ2 ,
δθ1 = 0 δθ1 = ϵ2 W ′ , (5.25)
δθ2 = 0 δθ2 = −ϵ1 W , ′

where δ denotes a transformation which transforms only the fields and not to the com-
plex conjugated fields. δ denotes a separate transformation which transforms only
the complex conjugated fields. ϵi are again small, fermionic variables. If we also set
2
ϵ1 = ϵ2 , ϵ1 = ϵ2 , we see from the transformations in (5.24) and (5.25) that δ 2 = δ = 0.

40
 5 The Path Integral in 0-Dimensional QFT

We see this by noting that each term in (δθi )2 and (δψ)2 contains a factor of ϵ2 = 0.
The same logic applies to the transformations of the complex conjugated fields.

We have already proven that localization can be applied to supersymmetric QFT:s such
as this one. In analogy with the supersymmetric QFT in section 5.4, the partition
function simplifies also in this case to include only those points where the superpotential
has a critical point, W ′ (ψ0 ) = 0. However, the expression for the partition function
doesn’t necessarily look the same as it did in equation (5.23). To evaluate the partition
function in this QFT, we again expand the superpotential around its critical points and
write W (ψ) = W (ψ0 ) + α2 (ψ − ψ0 )2 , with α = W ′′ . We also set ψ = x + iy, where x and
y are real, bosonic variables. Our partition function reduces to

ˆ
1 X h  i
Z= exp − |α(ψ − ψ0 )|2 − αθ1 θ2 − αθ1 θ2 dψdψdθ1 dθ2 dθ1 dθ2
2π ψ0
ˆ
1 X 2  
= |α| exp −|α|2 |ψ − ψ0 |2 dψdψ
2π ψ0
   
∂ψ ∂ψ
= dψdψ = abs det ∂x
∂ψ
∂y
∂ψ
 dxdy = 2dxdy 
∂x ∂y
ˆ ˆ
1 X 2 |α|2 x2 2 y2 1 X 2 2π
= |α| 2 e|α| dy = = 1. (5.26)
X
e dx |α|
2π ψ0 2π ψ0 |α|2 ψ0
´
To´obtain the expression on the last line, we use the property exp (−|α|2 |ψ − ψ0 |2 ) dψdψ
= exp (−|α|2 |ψ|2 ) dψdψ, which holds since a change of variables ψ ↔ ψ − ψ0 leaves the
integral unchanged. We see from equation (5.26) that our complex-valued supersymme-
try has reduced the partition function to counting the number of critical points of W .
As opposed to equation (5.23) in section 5.4, here we don’t have to consider the nature
of the critical points.

We also want to investigate how to evaluate correlation functions using localization. We

note first that a function f (ψ, ψ, θ1 , θ2 , θ1 , θ2 ) which is invariant under at least one of the
two supersymmetry transformations in (5.24) and (5.25) is in general easier to evaluate
than a function which is not invariant. We explain why this is. Consider f and assume it
is invariant under the δ-transformation. Assume also that δρ denotes the transformation
W → λW . By writing out ⟨f ⟩ we see that −δρ ⟨f ⟩ = ⟨f δρ S⟩ = ⟨f δh⟩ = {δf = 0} =
⟨δ(f h)⟩ = 0. Here, h = h(ψ, ψ, θ1 , θ2 , θ1 , θ2 ) is some function which fulfills δh = δρ S,
similar to the function g we found in section 5.3 to prove deformation invariance. Since
δρ ⟨f ⟩ = 0, we may apply localization also to this correlation function and evaluate it
only in those points ψ0 where W ′ = 0.

An example of a function which is invariant is any function of only ψ or ψ, i.e. some

function β(ψ) or ω(ψ). β and ω are automatically invariant under one of our super-
symmetry transformations, since δψ = 0, δψ = 0, see (5.24) and (5.25). It turns out
that functions of the type β(ψ) or ω(ψ) are the only functions of the bosonic fields

41
 5 The Path Integral in 0-Dimensional QFT

which are invariant. To see this, we note that the function ψψ is not invariant, since
δ(ψψ) = δψψ + ψδψ ̸= 0, and the same goes for δ(ψψ). By the same reasoning, no
function ψ n ψ m ; n, m ̸= 0 is invariant.

We evaluate the correlation function of β(ψ) similarly to how we evaluated the expression
for Z in equation (5.26)
ˆ
1 X h  i
⟨β(ψ)⟩ = β(ψ0 )exp − |α(ψ − ψ0 )|2 − αθ1 θ2 − αθ1 θ2 dψdψdθ1 dθ2 dθ1 dθ2 = β(ψ0 ).
X
2π ψ0 ψ0

Again, supersymmetry has simplified an otherwise complicated computation. We read

in David Skinner’s lecture notes on QFT [26]

"In the absence of experimental evidence for a supersymmetric extension of the

Standard Model, the close connections between supersymmetric QFTs and deep
mathematics and the fact that supersymmetry helps tame otherwise intractable
path integrals now provide the main reason for studying supersymmetry."

It is interesting to note that supersymmetry has many applications beyond trying to

extend the Standard Model, although the search for supersymmetric particles is still
very much ongoing.

Finally, we mention the importance of functions like β(ψ). From now on, we refer to
them not as functions, but as fields. Fields such as β which are invariant under the
δ-transformation are called chiral fields. Chiral fields have certain properties. Most
importantly, we may evaluate their correlation functions exactly. Also, the product of
two chiral fields, ζ and ν, is also a chiral field
δ(ζν) = δ(ζ)ν + ζδ(ν) = 0.
The same logic tells us that the addition of two chiral fields is also a chiral field. Under
2
our constraint ϵ1 = ϵ2 = ϵ, we saw that δ = 0, so we may also construct trivially chiral
fields of the form µ = δκ, where κ is some arbitrary field. It will always hold that δµ = 0
and ⟨µ⟩ = 0.

In general, a QFT is well described by the various correlation functions involving its
fields. We could try to calculate the correlation function of any arbitrary field in our
theory, but this often proves difficult, and we must turn to perturbation theory. It may
therefore be useful to construct a QFT whose main building blocks are chiral fields.
Trivially chiral fields like the ones mentioned previously, µ = δκ, are then those fields
which play no physically significant role in the theory, since they do not affect the
correlation function if added to a chiral field. Often, two chiral fields differ only by
the addition of a trivially chiral field, and we can in principle treat those two fields as
equivalent. This treatment helps us focus on the essential properties of a theory and
only consider relevant degrees of freedom.

42
 6 The Path Integral in 1-Dimensional QFT

6 The Path Integral in 1-Dimensional QFT

In a sense we’re now returning to the beginning of this report, since 1-dimensional QFT
is another word for quantum mechanics. One difference from the 0-dimensional case is
that we have boundaries to our manifold, and we always assume some boundary condi-
tions.

In sections 6.1 and 6.2 we show very briefly how one may generalize Feynman diagrams
and supersymmetry from 0 dimensions to 1 dimension. Section 6.2 also introduces
supercharges, which are conserved quantitites under supersymmetry. The discussion
mostly follows Hori et al. [17, ch. 10]. Section 6.3 discusses the use of path integrals in
statistical mechanics and thermal QFT, with applications to superfluids. The section is
based on Feynman and Hibbs [11, ch. 10] and several articles where path integrals are
used to model Helium-4, [2] [4] [24].

6.1 Feynman Diagrams

We will not perform any explicit calculations related to Feynman diagrams - we will
simply give an idea of how to generalize some concepts from section 5.1 to 1 dimension.

In 1-dimensional QFT, we have a time variable, which means that our system can evolve
and our equations may involve time derivatives. We consider an action similar to that
in equation (5.1)

ˆ 
α 2

S[ψ, ϵ] = ψ̇ + iϵψ 3 + jψ dt.
2
Now, the first term is an actual kinetic term and the second is a potential term. We
have also included a source term. We choose to write out the corresponding partition
function and expand it in a power series in ϵ

ˆ ´
Dx(t)e− ( 2 ψ̇(t) +iϵψ(t) +j(t)ψ(t))dt
α 2 3
Z=
ˆ ˆ ˆ ˆ
´ (−iϵ)2
!
= Dx(t)e − [ α2 ψ̇(t)2 +j(t)ψ(t)]dt 1 − iϵ ψ(t1 ) dt1 +
3
ψ(t1 ) ψ(t2 ) dt1 dt2 + ...
3 3
2
ˆ ´
Dx(t)e− ( 2 ψ̇(t) +j(t)ψ(t))dt
α 2
=
ˆ ˆ ˆ "ˆ
∞
(−iϵ)n ´
#
− (α 2 +j(t)ψ(t)
)
+ ψ̇(t)
ψ(t1 ) ψ(t2 ) ... ψ(tn ) .
3 3 3
X
... dt1 dt2 ...dtn Dx(t)e 2

n=1 n!

The last expression in brackets may be rewritten as correlation functions, although we

must here employ functional derivatives. For example

43
 6 The Path Integral in 1-Dimensional QFT

ˆ ´ α
δZ δ
e− ( 2 ψ̇(t) +j(t)ψ(t))dt
2
⟨ψ(t1 )⟩ϵ=0 = = Dx(t)
δj(t1 ) δj(t1 )
ˆ ˆ !
´ α
δj(t)
ψ(t)dt e− ( 2 ψ̇(t) +j(t)ψ(t))dt
2
= Dx(t)
δj(t1 )
ˆ ´ α
= Dx(t)ψ(t1 )e− ( 2 ψ̇(t) +j(t)ψ(t))dt ,
2

in analogy to equation (5.6). The integrals involved in correlation functions are clearly
more difficult to evaluate in 1 dimension than in 0 dimensions. Constructing Feynman
diagrams to facilitate the calculations works in much the same way that it did in section
5.1. However, we see that the numbering of the fields is more intuitive in 1 dimension:
here, a vertex represents an interaction at a specific point in time. When we move to
still higher dimensions, vertices exist at different points in spacetime.

6.2 Supersymmetry and Supercharges

We consider a supersymmetric 1-dimensional QFT, or rather, supersymmetric quantum
mechanics. We choose for our fields one bosonic field x, a complex fermionic field θ and
its complex conjugate θ. We consider the action

ˆ 
1 2 1 ′ i ˙ − h′′ (x)θθ dt,

S[x, θ, θ] = ẋ − (h (x))2 + (θθ̇ − θθ)
2 2 2
and a supersymmetry transformation

δx = ϵθ − ϵθ,
δθ = ϵ(iẋ + h′ (x)), (6.1)
δθ = ϵ(−iẋ + h′ (x)),

where ϵ is a complex fermionic variable given by ϵ = ϵ1 + iϵ2 , with ϵ1 , ϵ2 ∈ R. Also, ϵ

has no time-dependence. We see that the transformation in (6.1) simplifies to that in
(5.18) if we allow the terms involving time derivatives to go to zero. This is equivalent to
going from a 1-dimensional manifold to a 0-dimensional manifold. For this to generate
a symmetry, the action must be invariant under the transformation. This is equivalent
to saying that the Lagrangian can change by at most a total time derivative under the
transformation. By evaluating δS, we can show

ˆ !
+ δ θ˙ ˙ dt
∂L ∂L ∂L ∂L ∂L ∂L
δS = δx + δ ẋ + δθ + δ θ̇ + δθ
∂x ∂ ẋ ∂θ ∂ θ̇ ∂θ ∂θ
ˆ ˆ
ϵ d   ϵ d
= ẋθ + ih′ θ dt + (−ẋθ + ih′ θ) dt = 0. (6.2)
2 dt 2 dt

44
 6 The Path Integral in 1-Dimensional QFT

Since ϵ, ϵ do not depend on time, we move them outside of the integrals, and we are left
with two integrals involving a total time derivative. The whole expression goes to zero.

Furthermore, Noether’s theorem tells us that for every differentiable symmetry of the
action, there exists a corresponding conservation law [22] [14, ch.13]. Supersymmetry in
particular leads to conserved supercharges. We show now that our superymmetric QFT
involves two distinct supercharges Q and Q.

First of all, we introduce a method of finding such conserved quantities. If we first

consider some transformation involving a constant term ϵ, as in (6.1), the action is
invariant under this transformation. If we instead let ϵ be a function of time, this is not
necessarily the case. We instead arrive at some expression

ˆ
δS = ϵ̇κdt, (6.3)

where κ is a conserved quantity. The reason why this works is because we have to include
extra terms in our expression for δL when we allow ϵ to depend on time. These terms
involve a factor of ϵ̇, and we arrive at an expression like the one in equation (6.3). We
may then use integration by parts to rewrite equation (6.3), resulting in an integrand
on the form ϵκ̇. From the principle of least action, this expression must equal zero for
all ϵ. Thus, κ̇ = 0.

To find the quantity which is conserved in our case, we  set ϵ = ϵ(t) in equation (6.1).
We have already concluded that δL = 2 dt ẋθ + ih θ dt + 2ϵ dtd (−ẋθ + ih′ θ) when ϵ is
ϵ d ′

constant in time, see equation (6.2). Now that ϵ is time-dependent, we can no longer
move ϵ and ϵ out of the integral, and these terms remain. We must also include additional
˙ By considering all the above,
terms in δL, namely those including ϵ̇ in δ ẋ, δ θ̇ and δ θ.
we obtain a new expression for δS after some algebra

ˆ  
δS = −iϵ̇(θh′ (x) + iθẋ) − iϵ̇(θh′ (x) − iθẋ) dt. (6.4)

A comparison between equation (6.3) and (6.4) reveals that our conserved supercharges
are given by

Q = θ(iẋ + h′ (x)),
Q = θ(−iẋ + h′ (x)),

where Q is the complex conjugate of Q.

45
 6 The Path Integral in 1-Dimensional QFT

6.3 The Partition Function in Statistical Mechanics and Ther-

mal QFT
Partition functions in QFT are closely related to those in statistical mechanics. This
means that the mathematical structure of statistical mechanics and QFT is quite similar
and the theory behind QFT is often useful when we analyze many-particle systems. In
this section, we give an example of how one can apply QFT to statistical mechanics.
We examine specifically the canonical partition function from statistical mechanics and
its analogue in thermal QFT. The canonical partition function describes a system at
fixed temperature, volume and number of particles which can exchange heat with its
environment. The analogous field theory is QFT at a fixed temperature, thermal QFT.
At the end of this section, we give a non-rigorous discussion about how thermal QFT
may be used to predict the behaviour of superfluids.

In statistical mechanics our canonical partition function is given by a sum over the
probabilities for the system to be in a given energy state. In a quantum mechanical
system with quantized energy levels, we write this as the trace over the Boltzmann
function

Z = tr(e−βH ).

We rewrite the above using the general Hamiltonian for a particle in a potential

ˆ
p2
Z= dx ⟨x|e−β( 2m +V (x)) |x⟩ . (6.5)

It is implied that p and x are operators in the above expression. We would like to
rewrite equation (6.5) into something similar to a path integral. In order to do so, we
p2 p2
must first rewrite the exponential function as in e−β( 2m +V (x)) → e−β 2m e−βV (x) . However,
x and p do not commute. The Baker-Campbell-Hausdorff formula (BCH formula) tells
1
us that in general ef (p) eg(x) = ef (p)+g(x)+ 2 [f (p),g(x)]+... , for some operators f and g. We
can easily prove the BCH formula by expanding the logarithm of eA eB , where A and B
are operators

!n
∞
(−1)n−1 A B ∞
(−1)n−1 X ∞ X ∞
Am B k
log(e e ) =
A B
(e e − 1) =
n
−1
X X

n=1 n n=1 n m=0 k=0 m!k!

1 1
= (A + B + AB + ...) − (A2 + B 2 + AB + BA + ...) +... = A + B + [A, B] + ...
| {z } 2 2
n=1
| {z }
n=2

Thus, if f (p) and g(x) are infinitesimal, we can approximate ef (p)+g(x) = ef (p) eg(x) .

46
 6 The Path Integral in 1-Dimensional QFT

We do what we have done previously in this thesis, and split an interval into smaller
pieces. We take β in equation (6.5) to be the length of some interval and split it into
p2 p2
N pieces, each of length ϵ. This allows us to write e−β( 2m +V (x)) = (e−ϵ 2m e−ϵV (x) )N as
N → ∞ and ϵ → 0. We use this expression to rewrite equation (6.5), and we insert the
identity operator in position- and momentum basis at each step along our interval β

ˆ
p2
Z = lim dx ⟨x|(e−ϵ 2m e−ϵV (x) )N |x⟩
N →∞
ˆ ˆ ˆ ˆ ˆ
dp1 dpN p2 p2
= lim dx dx1 ... dxN ⟨x|e−ϵ 2m |p1 ⟩ ⟨p1 | e−ϵV (x) |x1 ⟩ ⟨x1 | e−ϵ 2m |p2 ⟩
N →∞ 2π 2π

p2
⟨p2 | e−ϵV (x) |x2 ⟩ ... ⟨xN −1 |e−ϵ 2m |pN ⟩ ⟨pN | e−ϵV (x) |xN ⟩ ⟨xN |x⟩

N ˆ ˆ
dpi −ϵ( p2i +V (xi ))
= lim
Y
dxi e 2m ⟨xi−1 |pi ⟩ ⟨pi |xi ⟩
N →∞
i=1 2π x0 =xN
N ˆ ˆ p2
dpi i +V
= lim −ϵ( 2m (xi )+ipi (xi−1 −xi ))
(6.6)
Y
dxi e ,
N →∞
i=1 2π x0 =xN
´ ´
where the constraint x0 = xN follows from the integration dx ⟨xn |x⟩ = dxδ(x − xN ),
and a change in notation x → x0 .

We integrate over each pi in expression (6.6) to obtain

N ˆ 
xi −xi−1
2
N ˆ
m −ϵ m m
r r
+ϵV (xi )
Z = lim = lim dxi e−ϵL(xi ,ẋi )
Y 2 ϵ
Y
dxi e
N →∞
i=1 2πϵ x0 =xN
N →∞
i=1 2πϵ x0 =xN

ˆ x(β)
= Dxe−S[x,ẋ] . (6.7)
x(0) x(0)=x(β)

Here, we interpret ϵ as an infinitesimal piece of the "time interval" β, and we denote

ẋ = xi −xϵ i−1 . This is only an analogy since we observe the similarity between the path
integral in equation (3.2) and equation (6.7), although the latter is Wick-rotated and
expressed in natural units. Note that the interval over which the path integral runs is
not a true time interval here, so we are not considering a true motion between states.

Equation (6.7) is the partition function in thermal QFT. The expression agrees with our
general formula for partition functions (4.2), but here we have additional constraints.
Namely, the intial- and final state for the path integral must coincide and our interval
is given by β, which in turn depends on the temperature T , as β = kT 1
. Observe also

47
 6 The Path Integral in 1-Dimensional QFT

that we must integrate over all initial (final) states in equation (6.7). This differs from
path integrals in previous sections, where we have assumed a fixed initial state. In sta-
tistical mechanics, we are interested in systems that exist in one of many states, each
with a given probability. Therefore we must consider every possible initial position for
a particle in this case.

We imagine a many-particle system: a liquid. The initial (final) state for the system
then consists of a particle distribution. One possible path for such a system is illustrated
in figure (13), part A. Each blue dot represents an initial position for a particle, to
which it returns at the end of the interval. Now let’s assume our liquid consists of
indistinguishable bosons. For such particles, the final state in equation (6.7) looks the
same if we interchange two or more bosons. In figure (13), part B, we illustrate one such
path, which includes a permutation of final positions. In conclusion, both figure A and
B illustrate allowed paths for the system, and both must be considered as we evaluate
equation (6.7).

•
•
•
•
A: • •
•
• •

•
•
•
•
B: • •
•
• •

Figure 13: Two possible paths for particles in a liquid, from an initial state to an
identical final state, where the dots represent initial/final positions. All particles are
indistinguishable bosons.

Sidenote: if we consider fermions instead of bosons, some permutations of final positions

give positive contributions to the partition function, and other permutations give nega-
tive contributions. For simplicity, we stick to the treatment of bosons in this example.
Note also that all integer spin particles are bosons. Our discussion therefore applies to
helium-4, which is a boson.

Normally, the paths which contribute most to the partition function are those which
minimize the action, i.e. the shortest possible paths. When we evaluate equation (6.7),
we mainly consider paths that move a very short distance away from the initial (final)

48
 6 The Path Integral in 1-Dimensional QFT

state. Paths such as those in figure (13), part B, force the particles to travel a longer
distance. In other words, paths which involve permutations generally contribute less
to the partition function. We can see just how suppressed such contributions are by
considering equation (6.7), which for a system of particles is proportional to

ˆ ˆ β
! ˆ !
mX mX
ẋ2i du ≈ ρ(xi )exp − |xi − P (xi )|2 dxi .
X
Z∝ Dx(u)exp −
2 i 0 P 2β i
(6.8)

Here, P (xi ) denotes the new final position of xi after a permutation. In the last step,
we approximate the paths from initial to final position as straight lines. We also as-
sume some type of lattice model for the liquid, which enables us to introduce a particle
density ρ(xi ). The sum over P represents a sum over all possible permutations of final
positions, and the sum over i is a sum over all particles in our liquid. This is a very
crude approximation, but it works for our qualitative discussion.

From equation (6.8), we see that a path which includes a permutation of just two parti-
−mkT d2i
cles contributes to the partition function with a term proportional to C = exp ℏ2
,
where the factor ℏ2 has been introduced by dimensional analysis. T is the temperature
of the liquid. di is the distance travelled from initial to final state for particle i. This
distance is at least of the order of magnitude of the average distance between particles.
A path involving permutations of N particles includes a factor C N , and its contribution
is much suppressed. However, there are also N ! possible permutations involving the N
particles, so the contribution is not necessarily negligible. For low enough temperatures
and small enough spacing between particles, C may be large enough that contributions
from paths involving permutations compares to those from paths without permutations.
A deeper analysis of equation (6.7) for Helium-4 reveals that for a specific temperature,
permutations involving an arbitrary amount of particles suddenly contribute notice-
ably to the partition function [2]. This behaviour arises very suddenly, and completely
changes the behaviour of the liquid, since most information about the system is encoded
in the partition function. Among other things, the liquid starts to flow without fric-
tion and its heat capacity grows exponentially. The liquid becomes a superfluid, and
this is due to quantum mechanical effects at low temperatures, as we have discussed here.

To give an example of how these permuted paths might affect the properties of a liquid,
we investigate the heat capacity. Heat capacity is defined in terms of the partition
function as

∂ 2 (kT lnZ) 1 ∂Z 2 1 ∂Z 1 ∂Z 2 1 ∂ Z
2
Cv = T = kT − kT + kT . (6.9)
∂T 2 Z ∂T Z ∂T Z ∂T Z ∂T 2
If we perform the derivatives in expression (6.9) on equation (6.8), the resulting expres-
2
sion for Cv contains two terms which are proportional to ⟨d2i ⟩ and ⟨d4i − ⟨d2i ⟩ ⟩ respec-
tively, again with di = |xi − P (xi )|. We see that the first term is the average squared

49
 7 Conclusions

distance travelled for each particle, and the second is the average of the difference be-
tween the quadratic distance and the average squared distance. When the partition
function receives a majority of its contributions from paths without permutations, all
distances di are small and of the same magnitude. However, at some low temperature we
suddenly include particle paths of all different lengths in our computation. The above
terms increase drastically, along with the heat capacity.

We excluded the potential term from the partition function in equation (6.8), and we
made several other approximations throughout this section. A deeper analysis is re-
quired to accurately predict the quantitative behaviour of superfluids. This was done
by Feynman in 1953 [9] [4], as he took into consideration the strong interaction between
atoms by considering only paths which do not cross over other atoms. Such paths are
achieved by allowing obstructing atoms to move a small distance out of the way. Since
this requires additional kinetic energy, it introduces an extra factor in the partition func-
tion, i.e. a multiplicative factor in the integrand of equation (6.8). Feynman also argues
that for a chosen set of particles, only permutations between closest neighbors contribute
noticeably to the partition function. This can be shown mathematically, from analyzing
the multiplicative factor just mentioned. In conclusion, the problem of modeling super-
fluids reduces to considering a lattice model of a liquid and a partition function similar
to that in equation (6.8). We then evaluate the contributions from different paths for
the particles in the system. Permutations are taken into consideration by constructing
polygons between lattice sites. A numerical simulation based on the above reasoning
was performed by Elser and Ceperley in 1987 [24]. They employed Monte Carlo methods
to construct polygons in lattices models of Helium-4 (which enters a superfluid phase
around 2 Kelvin). Many more have conducted similar studies, and a summary of similar
articles was published by Ceperley in 1995 [2]. Resulting predictions for e.g. energy,
specific heat and superfluid pressure show good agreement with experiment.

7 Conclusions
We introduced the subject of path integrals with a discussion about Hamiltonian vs.
Lagrangian mechanics, where the more fundamental nature of Lagrangian mechanics
provided one of the motivations for introducing the path integral in quantum mechan-
ics. Another motivation came from the need of a formulation of quantum mechanics
compatible with special
´ relativity. We gave the mathematical definition of the path in-
i
tegral, K(b, a) = Dx(t)e ℏ S , and also discussed interference between alternative paths
in quantum mechanical systems for a more intuitive interpretation. We gave a summary
of the properties of the path integral. Our discussion showed that the path integral
behaves as expected in the classical limit, that the path integral preserves causality
and that it treats consecutive events in a consistent manner. A common method for
evaluating path integrals was introduced, i.e. splitting time intervals into infinitesimal
pieces. With this method, we obtained an expression for the kernel of a free particle
and for a particle in an arbitrary potential. The latter result led to Feynman’s proof
of equivalence between the path integral and the Schrödinger equation. We argue that

50
 7 Conclusions

this proof validates the path integral formulation of quantum mechanics.

In later sections, we introduced 0-dimensional QFT. We discussed partition functions

and correlation functions in particular, and Feynman diagrams and supersymmetry as
tools to evaluate these functions. We also solved a simple problem in 0-dimensional QFT:
for a given action, we found the free energy and the equations of motion by classical
means, and then by taking quantum mechanical corrections into account. Our results
showed that these corrections were in fact vacuum fluctuations.

We moved to 1-dimensional QFT and found that our theories apply also in this case,
although computations become more complex. We also showed that the canonical par-
tition function in statistical mechanics can be rewritten to give the partition function
in thermal QFT. The result led to a practical application of the path integral in many-
particle systems as we attempted to quantitatively describe the behaviour of superfluids.
By studying articles on the subject, we saw that many successful simulations have been
performed using path integrals to predict the behaviour of Helium-4 in the transition
from fluid to superfluid.

Modern applications of the path integral far exceed those discussed in this report. What
started out as a mathematical curiosity by Dirac has developed into a useful tool in
many fields. One might ask whether theoretical physics is worth the effort when many
new theories lack obvious applications. The efforts by Dirac and Feynman, and many
others, show that there may be unexpected applications down the road. An interesting
continuation of this thesis would be to investigate QFT in higher dimensions and how
path integrals are applied in e.g. string theory, to see if path integrals could be a piece
of the puzzle in finding a theory of quantum gravity.

51
References
[1] Brown (Editor), Lauri M. Feynman’s Thesis — A New Approach to Quantum
Theory. (Northwestern University, USA, 2005).
[2] Ceperley, David M. Path integrals in the theory of condensed helium. Reviews
of Modern Physics 67, 279 (1995).
[3] Chaichian, Masud and Demichev, Andrei. Path Integrals in Physics: Vol-
ume II Quantum Field Theory, Statistical Physics and other Modern Applications.
(CRC Press, 2018).
[4] Cross, Michael. Statistical Physics - Applications of Path Integrals to Super-
fluidity. Caltech: Physics 127c. (2006).
[5] Davisson, C. and Germer, L. H. Diffraction of Electrons by a Crystal of Nickel.
Phys. Rev. 30, 705–740 (1927).
[6] Dirac, P.A.M. The Lagrangian in Quantum Mechanics. Physikalische Zeitschrift
der Sowjetunion, Band 3, Heft 1. 312–320 (1933).
[7] Dirac, P.A.M. The quantum theory of the emission and absorption of radiation.
Proceedings of the Royal Society of London. Series A, Containing Papers of a
Mathematical and Physical Character 114, 243–265 (1927).
[8] Einstein, Albert. Über einen die Erzeugung und Verwandlung des Lichtes be-
treffenden heuristischen Gesichtspunkt. Annuals of Physics 322, 132–148 (1905).
[9] Feynman, R. P. The λ-Transition in Liquid Helium. Physical review 90, 1116–
1117 (1953).
[10] Feynman, Richard. The principle of least action in quantum mechanics. the
Faculty of Princeton University. (1942).
[11] Feynman, Richard P. and Hibbs, Albert R. Quantum Mechanics and Path
Integrals. (Dover Publications, Inc. 2010).
[12] Feynman, Richard Phillips. Space-time approach to quantum electrodynam-
ics. Physical Review 76, 769 (1949).
[13] Gamelin, Theodore W. Complex Analysis. (Springer Science - Business Media
New York, originally published by Springer-Verlag New York, Inc. in 2001, 2001).
[14] Goldstein, Herbert, Poole, Charles, and Safko, John. Classical Mechan-
ics. (Addison Wesley, 2000).
[15] Greiner, Walter. Quantum mechanics: an introduction. (Springer Science &
Business Media, 2000).
[16] Haberman, Richard. Applied partial differential equations with Fourier series
and boundary value problems. (Pearson Higher Ed, 2013).
[17] Hori, Kentaro et al. Mirror Symmetry. (the American Mathematical Society,
Providence, RI, for the Clay Mathematics Institute, Cambridge, MA, 2003).

52
[18] Ishibashi, Nobuyuki, Okuyama, Kazumi, and Satoh, Yuji. Path integral
approach to string theory on AdS3. Nuclear Physics B 588, 149–177 (2000).
[19] Mandl, Franz. Quantum field theory. 2. ed. (Wiley, 2010).
[20] Maxwell, James Clerk. A dynamical theory of the electromagnetic field. Philo-
sophical transactions of the Royal Society of London. 459–512 (1865).
[21] Nobel Prize Outreach. Louis de Broglie – Facts. https://www.nobelprize.
org/prizes/physics/1929/broglie/facts/. Accessed 2023-05-05.
[22] Noether, E. Invariante Variationsprobleme. Nachrichten von der Gesellschaft
der Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse 1918, 235–
257 (1918).
[23] Planck, Max. On the Law of the Energy Distribution in the Normal Spectrum.
Annals of Physics 4, 553, Originally published in Verh. Dtsch. Phys. Ges. Berlin,
1900, 2, 202 and 237, Translated from German by Kuyanov Yu.V. (1901).
[24] Pollock, E. L. and Ceperley, D. M. Path-integral computation of superfluid
densities. Phys. Rev. B 36, 8343–8352 (1987).
[25] Sakurai, J.J. and Napolitano, Jim. Modern Quantum Mechanics, 2nd edition.
(Pearson Education, Inc., publishing as Addison-Wesley, 1301 Sansome Street, San
Fransisco, CA 94111, Copyright 1994, 2011).
[26] Skinner, David. Quantum field theory ii. Lecture notes, Part III of the Mathe-
matical Tripos, University of Cambridge. (2018).
[27] Sommer, Stefan, Fletcher, Tom, and Pennec, Xavier. Introduction to dif-
ferential and Riemannian geometry. 3–37 (2020).
[28] Tretkoff, Ernie. This Month in Physics History - May 1801: Thomas Young
and the Nature of Light. APS News 17, (2008).
[29] Weinberg, Steven. The search for unity: Notes for a history of quantum field
theory. Daedalus. 17–35 (1977).
[30] Young, Thomas. The Bakerian Lecture. On the theory of light and colours.
Philosophical transactions of the Royal Society of London. 12–48 (1802).

53
A The Central Equations in Lagrangian and Hamil-
tonian Mechanics
We give a brief recap of the central equations in Lagrangian- and Hamiltonian mechanics.
First: F unctionals are functions which acts on functions. One example is the action

ˆ b
S[L(qi , q̇i , t)] = L(qi , q̇i , t)dt, where L = T − V. (A.1)
a

The principle of least action applied to equation (A.1) leads to the Euler-Lagrange
equations

!
∂L ∂ ∂L
− = 0. (A.2)
∂qi ∂t ∂ q̇i

For a derivation, see Goldstein et al. [14, ch. 2].

Hamiltonian mechanics is closely related to Lagrangian mechanics. The Hamiltonian is

defined as

H(qi , pi , t) = pi q̇i − L(qi , q̇i , t), (A.3)

where the indices in the first term on the RHS implies a sum over all variables. The
Hamiltonian often evaluates to the total energy of a system, but not always. We may
solve for L in equation (A.3) and insert this expression in equation (A.2). If we consider
qi and pi to be our new variables, we obtain

!
∂ ∂ ∂ ∂H
(pi q̇i − H(qi , pi , t)) − (pi q̇i − H(qi , pi , t)) = 0 → q̇i = ,
∂pi ∂t ∂ ṗi ∂pi
!
∂ ∂ ∂ ∂H
(pi q̇i − H(qi , pi , t)) − (pi q̇i − H(qi , pi , t)) = 0 → ṗi = − .
∂qi ∂t ∂ q̇i ∂qi

The resulting equations are called Hamilton’s equations

∂H
q̇i = ,
∂qi
∂H
ṗi = − .
∂qi

54
B Cauchy’s Theorem over Infinite Domains
We state Cauchy’s theorem here as expressed in Gamelin [13, p.110].

Cauchy’s Theorem. Let D be a bounded domain with piecewise

smooth boundary. If f(z) is an analytic function on D that extends
smoothly to ∂D, then

ˆ
f (z)dz = 0. (B.1)
∂D

For a proof, see the same book by Gamelin. In the following sections, we perform some
computations using Cauchy’s theorem, to be used in the main text. Throughout, we use
the notation z = x + iy.

B.1 Complex-Valued Gaussian Integral

Assume we have an integral of the form

ˆ ∞
2
eiax dx, (B.2)
−∞

where x is a real variable and a is a real constant.

We choose a closed contour in the complex plane as in figure (14).

Im

•(p,p)

γ3 γ2

γ1 Re

γ4

(-p,-p)•

Figure 14: Closed directed curve in the complex plane.

The closed contour is made up of curves γi . We let Ii denote the integral of eiaz w.r.t.
2

z over the correspoding curve γi . By Cauchy’s integral theorem (B.1) we have

55
 I1 + I2 + I3 + I4 = 0 =⇒ I1 = −I2 − I3 − I4 , (B.3)

I1 is given by

ˆ p
2
I1 = eiax dx, (B.4)
−p

so that I1 with p → ∞ corresponds to the integral which we aim to evaluate, see (B.2).
We express the integrals I2 and I4 as

ˆ p
2
I2 = {z = p + iy} = eia(p+iy) idy.
0

ˆ 0 ˆ p
ia(−p+iy)2 2
I4 = {z = −p + iy} = e idy = eia(p+iy) idy = I2 . (B.5)
−p 0

We find an upper bound to this integral

ˆ ˆ
1 − e−2ap
p p 2
ia(p+iy)2
|I2 | = e dy ≤ −2apy
e dy = . (B.6)
0 0 2ap

From equation (B.6) and (B.5), we conclude lim |I2 | = lim |I4 | = 0.
p→∞ p→∞

The integral I3 evaluates to

ˆ −p
2
I3 = {z = x + ix} = (1 + i) e−2ax dx. (B.7)
p

We let p → ∞ and evaluate equation (B.7) as a Gaussian integral

ˆ −∞ √ iπ 1 r π
s
π iπ
r
−2ax2
lim I3 = (1 + i) e dx = −(1 + i) = − 2(e 2 ) 2 =− . (B.8)
p→∞
∞ 2a 2a a

We combine our results from (B.3), (B.4) and (B.8) to get

ˆ ∞
s
iax2 iπ
e dx = − p→∞
lim I3 = . (B.9)
−∞ a

56
B.2 Complex-Valued Gaussian Integral with a Quadratic Term
Assume we have an integral of the form

ˆ ∞
2
x2 eiax dx.
−∞

This can be written as

ˆ ˆ s 
p p
1 √
!
iax2 d iax2 d iπ 
lim x2 e dx = −i e dx = {Equation (B.9)} = −i  = 3 −iπ.
p→∞
−p da −p da a 2a 2

Our final result is

ˆ ∞
2 1 √
x2 eiax dx = 3 −iπ.
−∞ 2a 2

B.3 Complex-Valued Gaussian Integral with a Linear Term in

the Exponent
Assume we have an integral of the form

ˆ ∞
2 +bx)
ei(ax dx.
−∞

We solve this by completing the square in the exponent

ˆ √ 2 !ˆ ∞ s
∞ 2
√ 1
!
i b
ax+ 2√ b
−i 4a b ib2 iy 2 iπ ib2
e a
dx = {y = ax + √ } = √ exp − e dy = exp − ,
−∞ 2 a a 4a −∞ a 4a

where the last equality follows from equation (B.9). Our final result is

ˆ ∞
s !
i(ax2 +bx) iπ ib2
e dx = exp − .
−∞ a 4a

57
B.4 Integral of a Complex-Valued Exponential Function
Assume we have an integral of the form

ˆ ∞
i
I= e−ixτ dx, (B.10)
−∞ x + iϵ

where x is a real variable and τ and ϵ are real constants. We choose a closed contour as
in figure (15).

Im

R γ2

Re
γ1
•(0,−ε)

Figure 15: Closed directed contour in the complex plane with a pole at (0, −ϵ).

We let Ii denote the integral of equation z+iϵ

i
e−izτ w.r.t. z over the corresponding curve
γi . Since the integrand consists of an exponential function multiplied by the inverse of a
polynomial, it is analytic everywhere except at z = −iϵ. Thus, the integrand is analytic
inside our closed contour, as long as ϵ > 0. By Cauchy’s theorem (B.1), we have

I1 + I2 = 0 → I1 = −I2 .

We note also that the integral I1 as R → ∞ is our original problem (B.10).

We evaluate I2 as the radius R → ∞

ˆ
i
lim I2 = lim e−iτ x eτ y dz. (B.11)
R→∞ R→∞ γ2 x + iy + iϵ

As R → ∞, either |x| or y or both go to infinity along the curve γ2 . We impose

τ < 0, which forces the integral in equation (B.11) to zero when y → ∞. Similarly, the
oscillating function e−iτ x forces the integral to zero when |x| → ∞, since this oscillating
function is integrated against a bounded function. Our result is then I1 = −I2 = 0 and
the solution to our problem

58
 ˆ ∞
i
I= e−iwτ dw = 0, if τ < 0.
−∞ w + iϵ

B.5 The Principal Part of the Integral of a Complex-Valued

Exponential Function
Assume we have an integral of the form
ˆ ∞ ˆ −ϵ ˆ ∞
e−iτ x e−iτ x e−iτ x
!
I = P.P. dx = lim dx + dx ,
−∞ x ϵ→0
−∞ x ϵ x
where τ is a constant, with τ < 0 and x is a real variable. We choose a closed contour
as in figure (16). We let Ii denote the integral of the function e−iτ z /z with respect to z
over the corresponding curve γi .

Im

R γ4

γ2
Re
γ1 -ϵ ϵ γ3

Figure 16: Closed directed curve in the complex plane.

Using Cauchy’s theorem (B.1), we can write

I = I1 + I3 = −I2 − I4 , (B.12)
where it is implied that we take R → ∞ and |ϵ| → 0 as we evaluate the integrals I1 − I4 .

By the same argument as in section B.4, we find that I4 = 0 as R → ∞. The integral

I2 is evaluated as
ˆ
e−izτ
lim I2 = lim dz = {e−izτ = 1, to the lowest order in Taylor series} =
|ϵ|→0 |ϵ|→0 γ2 z
ˆ ˆ 0
1
= lim dz = {z = |ϵ|e } = i
iθ
dθ = −iπ. (B.13)
|ϵ|→0 γ2 z π

59
From equation (B.12) and (B.13) we obtain

ˆ ∞
e−izτ
P.P. dz = iπ, if τ < 0.
−∞ z

C An Introduction to Green’s Functions

The concept of Green’s functions is used in the main text when we discuss the Schrödinger
equation. Green’s functions are used to solve inhomogeneous linear differential equa-
tions, such as

L[u(x, t)] = Q(x, t), (C.1)

where Q is a source, and L is an operator, L = ik1 ∂t ∂

+ k2 ∇2 , with k1 , k2 ∈ R. We
also impose some inhomogenous boundary conditions u(xa , t) = α, u(xb , t) = β and the
initial condition u(x, 0) = g(x). The function u(x, t) is then the response to the source
Q, under the given boundary- and initial conditions.

The Green’s function G(x, t; x0 , t0 ) is defined by

L[G(x, t; x0 , t0 )] = δ(x − x0 )δ(t − t0 ), (C.2)

with corresponding homogenous boundary conditions as those for u at xa , xb . G(x, t; x0 , t0 )

is then the response at (x, t) to a concentrated source at (x0 , t0 ). By causality, this re-
sponse should be zero if t < t0 , i.e. we impose G(x, t; x0 , t0 ) = 0 if t < t0 . The Green’s
function defined in equation (C.2) can then be used to solve the problem in equation
(C.1). The explicit solution is

ˆ t˚ ˚
u(x, t) = G(x, t; x0 , t0 )Q(x0 , t0 )d x0 dt0 − ik1
3
G(x, t; x0 , 0)g(x)d3 x0
0
ˆ t‹
+ k2 u(x0 , t0 )∇x0 G(x, t; x0 , t0 ) · n̂dS0 dt0 .
0

(C.3)

A derivation of this will follow below. First, we focus on the expression in equation
(C.3) and see that the Green’s function propagates the effect of each nonhomogenous
part of our problem to give the response u(x, t). The first term gives the influence from
the source Q. The second term gives the influence from the initial condition. The third

60
term gives the influence from the boundary conditions. This is a general property of
Green’s functions: they propagate the effect of some function via integration.

There is one more property of Green’s functions which we should discuss: the reciprocity
theorem

G(x1 , t1 ; x0 , t0 ) = G(x0 , t1 ; x1 , t0 ), (C.4)

This formula tells us that the location of the source and the location of the response can
be interchanged, and the influence from the source remains the same. Below follows a
derivation of equation (C.3) and (C.4).

First, we introduce an adjoint operator L∗ = −ik1 ∂t ∂

+ k2 ∇2 . We also consider a Green’s
function where we let the time of the source vary instead of the time of the response,
G(x, t1 ; x1 , t). Note that the influence on some function at time t from a source at
time t0 depends only on the elapsed time, so we can consider the time interval [t, t1 ]
as well as [−t1 , −t]. Using this fact, we rewrite our source-varying Green’s function
G(x, t1 ; x1 , t) = G(x, −t; x1 , −t1 ). This is identical to our original Green’s function,
equation (C.2), but with the opposite sign for the time variable. We see that our source-
varying Green’s function fulfills L∗ [G(x, t1 ; x1 , t)] = δ(x − x1 )δ(t − t1 ).

We also consider two arbitrary functions u(x, t), v(x, t) and employ a method often used
to solve similar problems. We write

ˆ tf ˚ ˆ tf ˚ " ! #
∂v ∂u
[uL∗ (v) − vL(u)]d3 xdt = −ik1 u +v − k2 (v∇2 u − u∇2 v) d3 xdt.
ti ti ∂t ∂t
(C.5)
We rewrite the first term in the integrand on the RHS using integration by parts

ˆ tf ˚ ! ˚ tf ˆ tf ˚ !
∂v ∂u 3 ∂u ∂u 3
ik1 u +v d xdt = ik1 uv d x+
3
ik1 −v +v d xdt
ti ∂t ∂t ti ti ∂t ∂t

˚ tf
= ik1 uv d3 x. (C.6)
ti

We see that the adjoint operator L∗ was introduced since L is not self-adjoint. If we
had used L in place of L∗ in equation (C.6), we would have been left with an additional
integral on the right hand side.

We also rewrite the second term in the integrand on the RHS of equation (C.5) using
Green’s Second Identity, which is derived from the divergence theorem as

61
‹ ˚
(u∇v − v∇u) · n̂dS = ∇ · (u∇v − v∇u)d3 x
˚ ˚ ˚
= (∇u∇v − ∇v∇u)d x +
3
(u∇ v − v∇ u)d x =
2 2 3
(u∇2 v − v∇2 u)d3 x.
(C.7)
Here, n̂ is the normal unit vector to the boundary S. We use equation (C.7) and (C.6)
to rewrite (C.5) as

ˆ tf ˚ ˚ tf ˆ tf ‹
[uL (v) − vL(u)]d xdt = −ik1
∗ 3
uv 3
d x − k2 (v∇u − u∇v) · n̂dSdt.
ti ti ti
(C.8)
This will be an important result moving forward.

To finish our proof, we set u = u(x, t), v = G(x, t0 ; x0 , t) and tf = t0+ , ti = 0 in equation
(C.8). t0+ here means that we integrate an infinitesimal slice of time beyond t0 . Note
also that v is a source-varying Green’s function. We obtain

ˆ t0+ ˚
(u(x, t)δ(x − x0 )δ(t − t0 ) − G(x, t0 ; x0 , t)Q(x, t))d3 xdt
0
˚ t0+
= −ik1 u(x, t)G(x, t0 ; x0 , t) d3 x
ˆ t0+ ‹ 0

− k2 (G(x, t0 ; x0 , t)∇u(x, t) − u(x, t)∇G(x, t0 ; x0 , t)) · n̂dSdt. (C.9)

0

The first
´ t term
‚ in the integrand on the LHS in equation (C.9) evaluates to u(x0 , t0 ). The
term 0 0+ (G(x, t0 ; x0 , t)∇u(x,
˝ t) · n̂)dSdt evaluates to zero since G equals zero at the
boundary. Also, the term ik1 u(x, t0+ )G(x, t0 ; x0 , t0+ )d x evaluates to zero by causal-
3

ity, since t0+ > t0 .

We rewrite equation (C.9) according to the above discussion. One can also show that
we may now set t0+ = t0 . Finally, a change of variables (x0 , t0 ) ↔ (x, t), along with the
reciprocity relation in equation (C.4), allows us to obtain the expression in (C.3).

We prove the reciprocity formula (C.4) in a similar way. We again consider equation
(C.8) and set u = G(x, t; x0 , t0 ), v = G(x, t1 ; x1 , t) and tf = ∞, ti = −∞. We obtain

ˆ ∞ ˚
(G(x, t; x0 , t0 )δ(x − x1 )δ(t − t1 ) − G(x, t1 ; x1 , t)δ(x − x0 )δ(t − t0 ))d3 xdt
−∞
˚ ∞
= ik1 G(x, t; x0 , t0 )G(x, t1 ; x1 , t) d3 x + integral of G over boundary.
t=−∞

62
The final term where we evaluate G at the boundary equals zero. The first term on
the right hand side also equals zero from causality. Thus, by evaluating the left hand
side, we obtain G(x1 , t1 ; x0 , t0 ) − G(x0 , t1 ; x1 , t0 ) = 0. This concludes the derivation of
equation (C.4).

For similar proofs and a more in-depth discussion, see Haberman [16, ch.11].

63