introduction to koopman-von neumann mechanics an …
TRANSCRIPT
INTRODUCTION TO KOOPMAN-VON NEUMANN MECHANICS
AN HONORS THESIS
SUBMITTED ON THE FIFTH DAY OF DECEMBER, 2021
TO THE DEPARTMENT OF PHYSICS AND ENGINEERING PHYSICS
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
OF THE HONORS PROGRAM
OF NEWCOMB-TULANE COLLEGE
TULANE UNIVERSITY
FOR THE DEGREE OF
BACHELOR OF SCIENCES
WITH HONORS IN PHYSICS
BY
Daniel W. Piasecki
APPROVED:
Denys I. Bondar, Ph.D.
Director of Thesis
Ryan T. Glasser, Ph.D.
Second Reader
Frank J. Tipler, Ph.D
Third Reader
arX
iv:2
112.
0561
9v2
[qu
ant-
ph]
25
Dec
202
1
Abstract
This work is to consolidate current literature on Koopman-von Neumann
(KvN) Mechanics into a simple and easy to understand text. KvN Mechanics is
a branch of Classical Mechanics that has been recast into the mathematical lan-
guage of Quantum Mechanics. KvN Mechanics utilizes a Hilbert phase space
with operators to calculate the expectation values of observables of interest (ex-
pectation values such as position, momentum, etc.) It is an important tool in
statistical physics and guide to illuminate the mysterious relationship between
Quantum and Classical Physics. A lot of important applications of KvN Me-
chanics have been developed in the last decades.
ii
Acknowledgements
Thanksgiving is approaching at the time I’m writing this, so there are some people
who will get some messages of gratitude from me a few days early.
First, I would like to thank the Tulane Summer Research Program and the Tulane
Honors Program. This project would not have been possible without their support.
Many thanks to the readers of this thesis: Dr. Ryan Glasser, Dr. Frank Tipler, and
especially my advisor for this project, Dr. Denys Bondar. I am grateful for his advice
on this paper and guidance for how I can take what I’ve learned from it and apply
it to the rest of my career, both in terms of new knowledge about the subject matter
and skills I’ve been practicing while working on this. His openness and directness
in advising me made a big difference, and I’ve learned a lot from working on this
project. Also, 80 pages is a lot to read, so I am grateful for the time all three of these
professors have dedicated to reading my work.
******************************************************************
I am also very appreciative to my friends and family for their encouragement,
prayers, and patience with me. This thesis was a huge commitment for me, and they
were respectful of my time when I was busy and curious and uplifting during my few
moments of downtime. One of them also helped look over some of it and did some
proofreading, which I also am thankful for. Thanks goes especially to my sweetheart
Clover Robichaud for her continual advice, patience, and support while the project
has been ongoing. Also, thanks goes to my good friend Kushal Singh for helping me
prepare for the Thesis Defense itself.
Finally, I am grateful to all other physicists and aspiring physicists who read this;
if you are one of these people, you are exploring a little known topic that can make
a big difference, and I appreciate that you are likely branching out from the standard
iii
topics you are already taught in class. I am hopeful that my work contributed to
their knowledge of Koopman–von Neumann Mechanics and that this will help them
expand the field of Physics. This is currently a little known branch of Physics, and the
goal of this project was to make this information more accessible to others unfamiliar
with the topic, both in Physics and otherwise. If anyone could take what I did and
have that help him or her in contributing to Physics, then I will know that I have made
a difference.
iv
Contents
1 Introduction 1
2 General Review of Quantum Mechanics 2
2.1 Fundamentals of Hilbert Space . . . . . . . . . . . . . . . . . . . . 3
2.2 Operators and Commutators . . . . . . . . . . . . . . . . . . . . . 7
2.3 Expectation Values and the Eigenvalue Problem of Quantum Mechanics 10
2.4 Review of Dirac Delta Functional . . . . . . . . . . . . . . . . . . 13
2.5 Summary of Quantum Mechanics and its Postulates . . . . . . . . . 14
2.6 The Density Operator . . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Introduction to KvN Mechanics 19
4 Path Integral Formulation of Quantum and Classical Mechanics 34
4.1 Feynman Path Integral of Quantum Mechanics . . . . . . . . . . . . 34
4.2 Koopman-von Neumann Classical Path Integral . . . . . . . . . . . 40
5 Classical and Quantum Comparison 46
5.1 The Phase in Koopman-von Neumann and Quantum Mechanics . . 46
5.2 Uncertainty Principle in Koopman-von Neumann and Quantum Me-
chanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.3 Measurement in Koopman-von Neumann and Quantum Mechanics . 51
5.4 The Double Slit Experiment . . . . . . . . . . . . . . . . . . . . . 61
6 Miscellaneous KvN Topics of Interest 74
6.1 Operational Dynamic Modeling . . . . . . . . . . . . . . . . . . . 74
6.2 Wigner Quasiprobability Distribution . . . . . . . . . . . . . . . . 77
6.3 Time Dependent Harmonic Oscillator . . . . . . . . . . . . . . . . 81
6.4 Aharonov–Bohm Effect . . . . . . . . . . . . . . . . . . . . . . . . 89
7 Other Research 100
v
References 104
Appendices 107
A Useful Mathematics for Hilbert Space Physics 107
B The Principle of Causality and Stone’s Theorem 110
C Exponated Operators and the Propogator 111
vi
List of Figures
1 This is the set up of the double slit experiment. Middle wall is at
y = yM and rightmost wall is at y = yR. Adapted from Mauro 2002 114
2 a. Left: Graph of KvN probability distribution. b. Right: Graph of
QM probability distribution. Adapted from Mauro 2002. . . . . . . 114
3 a. Left: Example graph of a single variable gaussian curve. b. Right:
Example graph of a double variable gaussian curve. Adapted from
Wolfram Alpha 2021. . . . . . . . . . . . . . . . . . . . . . . . . . 114
4 Aharanov-Bohm Experiment set up. Reproduced from Aharonov and
Bohm 1959. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
vii
1 Introduction
Quantum Mechanics (QM) is known for its unique mathematical structure unlike that
traditionally found in other branches of Physics. In QM, one has operators acting on
the wavefunction in complex Hilbert Space in order to compute what the expectation
values of observables seen in the laboratory are (Shankar 1988). Classical Mechanics
(CM), on the other hand, is based on continuous, real valued functions of spatial and
time dependent variables. One merely has to plug in the values of the variables at
a particular point, and one knows the value of the function of interest. QM makes
probabilistic predictions of the quantum particle in question, whereas CM makes
deterministic predictions of all classical behavior.
The large differences in mathematical form has historically made it difficult to
compare QM to CM (Mauro 2003). There have been many attempts to make QM
appear more mathematically similar to CM (Mauro 2003). For instance, the Born
and Madelung approach to QM has a modified Hamilton-Jacobi equation as its basis
(Mauro 2003). But there have been few attempts to make CM more similar in form
to QM. This changed in the 1930’s, however, when a form of statistical CM was
devised by Bernard Koopman and John von Neumann that had the same operatorial
mathematical structure as QM (Koopman 1931; von Neumann 1932). Koopman-von
Neumann Mechanics (KvNM), as it later became known, was subsequently used by
mathematicians to formulate ergodic theory, while physicists mostly forgot about the
existence of this field. Currently, there is a large renewal of interest by the physics
community in this relatively obscure field of study.
In this Thesis I intend to articulate many significant contributions of KvNM to
physics in a straightforward and easy to understand fashion. I am writing this text
in such a way that even an undergraduate with limited training in Physics could un-
derstand. I will point out significant topics of research in KvNM and how they are
changing physics as we know it. By making this topic more accessible to a variety of
readers, I hope that many will find it of interest and continue to build upon this novel
and fascinating avenue of study.
1
KvNM is statistical CM cast in the language of complex wavefunctions in Hilbert
Space. Although wavefunctions in Hilbert Space are thought to be tools of QM, they
can be used to exactly model the statistical properties of purely classical systems.
Classical wavefunctions have been used to study a wide range of physical phenom-
ena. For example, they have been used in studying chaotic systems in quantum and
classical theory (Wilkie and Brumer 1997), been used to develop a model of stochas-
ticity from determinism (McCaul and Bondar 2021), been used to study the inter-
play of connected quantum and classical systems (Bondar, Gay-Balmaz, and Tronci
2019), been used to study free fall and Fischer Information (Sen, Dhasmana, and
Silagadze 2020), been used to reproduce equations governing spin 1/2 relativistic
particles (Cabrera et al. 2019), among many other areas of research.
Of great interest, a field like KvNM might be able to decipher the true meaning
of QM, which has been hotly debated for over a century.1
2 General Review of Quantum Mechanics
What follows is a brief review of Quantum Theory. QM depends on operations on
the vector |ψ〉 in Hilbert Space. The wavefunction |ψ〉 is a complete description of
the state of the system in QM (Shankar 1988). It contains probabilistic information
of the system’s behavior.
It has a number of convenient properties, such as the fact that it makes switching
between one set of bases and another is a rather straightforward task. Hilbert Space
useful for all sorts of mathematical procedures. It is, however, not the typical tool of
CM.1Despite a century of progress, QM has no consensus interpretation of its equations.
‘Shut up and calculate’ rings as true today as it did a century ago. Feynman gives us a glimmer ofhope for the situation, however, as it always takes several generations to understand a great, mysterioustruth: “We have always had a great deal of difficulty understanding the world view that quantummechanics represents. At least I do, because I’m an old enough man that I haven’t got to the point thatthis stuff is obvious to me... You know how it always is, every new idea, it takes a generation or twountil it becomes obvious that there’s no real problem.” (Feynman 1982) Perhaps we’re getting closerto an understanding and perhaps a budding field like KvNM can hold clues to the answer.
2
2.1 Fundamentals of Hilbert Space
To understand QM, we must first be comfortable with the concept of a Hilbert Space.
A Hilbert Space is an infinite dimensional vector space with a well defined inner
product. One can conceive of the Hilbert Space as a coordinate system with an infinite
number of perpendicular axes as a crude model. For QM, the Hilbert Space we use
not only contains real numbers, but has also been extended to the complex numbers.
Although there are Hilbert Spaces that only contain real numbers, we will not need
to concern ourselves with those when working with QM or KvNM.
To see why Hilbert Spaces are so useful in QM, imagine a generic function f(x)
which is continuous on its domain (Shankar 1988). Since any continuous function
can be represented by an infinite number of points, we can represent any continuous
function in Hilbert Space with state vectors. Each point along a generic function
f(x) can be represented by a vector in Hilbert Space. Because there is an infinite
number of points along the function f(x) (even in a finite segment), we should not
be surprised it will take an infinite number of vectors to represent f(x).
This collection of vectors can be represented in an orthogonal basis (i.e., each
vector is perpendicular to any other vector in the collection or set, as there are an
infinite number of perpendicular axes in Hilbert Space.) A vector can be represented
with the following bra-ket notation (original to Dirac):
~v =
v1
v2
v3
=⇒ |v〉 =
v1
v2
v3
, 〈v| =(v∗1 v∗2 v∗3.
)
where |v〉 is the ket vector and 〈v| is the bra vector (Shankar 1988). The asterisk
(*) represents the complex conjugate. Vectors like ~v are ambiguous, because it is
not often clear if to write them as columns (as depicted above) or rows of numbers.
However, the Dirac bra-ket notation overcomes the ambiguity of typical vector nota-
tion in physics, by having that the ket is always equivalent to a column and the bra is
equivalent to a row of values. Hilbert Space utilizes bra-ket notation in its infinite di-
3
mensional mathematics. A traditional Hilbert Space ket might have more than three
entries therefore; it might be represented with an infinite number of entries.
With this in mind, the continuous function f(x) would be represented in Hilbert
Space as:
f(x) =⇒ |f(x)〉 =
f(x1)
f(x2)
f(x3)
...
f(xn)
...
. (1)
You can think about this representation of f(x) as the value of each point along f(x),
i.e. f(xi), having a unique independent basis vector in Hilbert Space multiplying it.
So according to the rules of Linear Algebra you may expand |f(x)〉, for example:
|f(x)〉 =
1
0
0
...
0
...
f(x1) +
0
1
0
...
0
...
f(x2) +
0
0
1
...
0
...
f(x3) + ...+
0
0
0
...
1
...
f(xn) + ...
In the above expansion, notice that the column vectors form a unique set of linearly
independent basis vectors.
The way to switch between bras and kets in a complex Hilbert Space is to take
the transpose of the vector and complex conjugate of each of the values, like so:
4
|v〉 =
v1
v2
v3
...
⇐⇒ 〈v| =
(v∗1 v∗2 v∗3 . . .
)
The asterisks represents the complex conjugate. This operation will be represented
by †, so that |v〉† = 〈v| and vice versa. Just like any ordinary vector, bra-ket vectors
can be added component wise and multiplied together:
v1
v2
v3
...
+
w1
w2
w3
...
= |v〉+ |w〉 =
v1 + w1
v2 + w2
v3 + w3
...
= |v + w〉 , (2)
(w∗1 w∗2 w∗3 . . .
)
v1
v2
v3
...
= v1w
∗1 + v2w
∗2 + v3w
∗3 + . . . = 〈w|v〉 ,
where 〈w|v〉 is a generalized “dot product” or inner product for the vector space. Two
vectors are said to be orthogonal if their inner product is 0. So in order for two vectors
|v〉 and |w〉 in Hilbert Space to be orthogonal, we must have 〈w|v〉 = 0. Also, like
for any common vector, you can rescale a vector by a complex constant c to stretch
or shrink its magnitude: |v′〉 = c |v〉.
The vectors additionally obey the following properties:
(〈w|v〉)† = 〈v|w〉 ,
|c1v + c2w〉† = (c1 |v〉+ c2 |w〉)† = (c1 |v〉)† + (c2 |w〉)† = c∗1 〈v|+ c∗2 〈w| .
5
Since continuous functions like f(x) can be represented in Hilbert Space, we can
recast continuous functions of QM into the Hilbert Space formalism. We will shortly
see why this is useful to do.
The wavefunction ψ is the backbone of the quantum mechanical description of
reality. The wavefunction ψ is an equation of state, i.e., it provides us with a descrip-
tion of the state of the system (Shankar 1988). 2The wavefunction got its name from
the fact that its form resembles that of a wave equation. In configuration space, the
wavefunction is often written as:
ψ(q1, q2, ..., qn, t) = R(q1, q2, ..., qn, t)eiS(q1,q2,...,qn,t)/~, (3)
where q1, q2, ..., qn are positions in configuration space and t is time. S represents the
Action and R is the amplitude of the wave equation. In light of the above discussion
and since ψ is a continuous function, we can represent ψ as |ψ(t)〉 in complex Hilbert
Space. Schrodinger’s equation, for example, becomes:
i~∂ψ
∂t= − ~2
2m
∂2ψ
∂q2+ V (q)ψ =⇒ i~
∂
∂t|ψ(t)〉 = − ~2
2m
∂2
∂q2|ψ(t)〉+ V (q) |ψ(t)〉 .
(4)
The benefits of using Hilbert Space instead of just utilizing ψ is because Hilbert
Space provides a rich mathematical, abstract structure where various manipulations
are straightforward and easy to do (Shankar 1988; Bondar et al. 2012). We can very
easily, for instance, change our basis of interest from one to another. We can ‘de-
construct’ functions and put them in various representations. This is not a flexibility
given by simply using ψ. As we will see, changing representations is a fundamental
ingredient to QM, so the Hilbert Space form is absolutely necessary.2One way you can think of an ‘equation of state’ is, for instance, the ideal gas law, PV = NkT .
This equation will tell you what the state of the system, i.e., what the pressure P , volume V , numberof gas particles N , and temperature T , for an ideal gas is.
6
2.2 Operators and Commutators
An operator is a mathematical object that “acts on” a function, vector, or matrix to
produce a new function (the derivative), a new vector, or a new matrix. For example,
the derivative ddt
acting on a function f(t) is an operator, as is a matrix M acting on
a vector |v〉. The operator produces a new mathematical entity, such as the df(t)dt
from
f(t) or a new vector |v′〉 from the original vector (|v′〉 = M |v〉). All operators will
be represented with triangular “hats”, for example, A.
Two important types of operators are Hermitian operators and anti-Hermitian op-
erators, which obey the following rules respectively:
A† = A,
A† = −A,
where the † symbol is the transpose with complex conjugate as before. The operators
we most use in QM are Hermitian operators. Hermitian operators are incredibly
useful to utilize, because if we take the † of an expression with a Hermitian operator,
the form of the Hermitian operator is left unchanged. For example, for two vectors
|v〉 and |w〉 and Hermitian operator H:
(〈w| H |v〉)† = 〈v| H† |w〉 = 〈v| H |w〉 .
An observable in QM is a property of a system you would like to measure or observe.
Examples of observables in QM would be the position of a particle, its momentum,
its spin, etc. Observables in QM are given by Hermitian operators A acting on the
state vector |ψ(t)〉, i.e., A |ψ(t)〉. The state vector |ψ(t)〉 is a vector in Hilbert Space
that contains complete dynamical information about the system in question.
Hermitian operators relate directly to properties you measure inside the labora-
tory. For any generic operator A, there exist special types of ket vectors which when
acted on by an operator A they equal themselves rescaled by a constant. In other
7
words, they obey the mathematical form of
A |ψ(t)〉 = A |ψ(t)〉 , (5)
where ‘A’ is always a real number (Shankar 1988).
Note that this special relationship is not true for just any vector |ψ(t)〉. This is only
true for specific vectors which we will call eigenkets. The rescaling real number ‘A’
is known as the eigenvalue, and the mathematical problem to identify for an operator
A the correct eigenkets and eigenvalues is called the eigenvalue problem(Shankar
1988)3. The eigenvalue problem is central to QM because it directly relates to the
observables measured in the laboratory, as will be developed below.
Perhaps one of the most important properties of a Hermitian operator is that its
eigenvalues must necessarily always be real (Shankar 1988). Other types of opera-
tors might give, for example, imaginary eigenvalues. Hermitian operators will only
give you real numbers, which are the types of values we would measure inside the
laboratory.
Two very important operators are the ones for the observables momentum and
position. In coordinate representation (as other representations are possible), the mo-
mentum and coordinate operators are given respectively as follows:
p =~i
∂
∂q, (6)
q = q. (7)
These operators are important, because as we shall see, they are related to the average
momentum and position of interest. 4
3The function version of the eigenvalue problem is Aψ = Aψ, where ψ is called an eigenfunction.To recover the Hilbert Space version, remember the colloquial transformation ψ =⇒ |ψ(t)〉. In theliterature, terms like eigenvectors and eigenfunctions are used interchangeably, as they are technicallythe same thing, just one written in an abstract Hilbert Space and one written without.
4For example, in standard function form one can easily show that the pψ = pψ where p is themomentum expectation value measurable in the laboratory and p is the above momentum operator.Earlier, we said that a common form of ψ in configuration space is ψ = ReiS/~, where R is the
8
Commutators describe relations between operators. The principle that two num-
bers A and B obey A · B = B · A is known as the principle of commutativity.
Operators do not need to, however, obey the principle of commutativity. We might
have, for example, that AB 6= BA for two operators A and B. This might be most
explicitly clear when we think about the example of matrix multiplication. The order
matrices are multiplied in might give you different end products. We say therefore
that matrix multiplication is noncommutative.
In order to describe this property for operators, we use the idea of a commutator.
For two operators A and B, the commutator is defined as:
[A, B] ≡ AB − BA.
If the commutator equals zero, the two operators commute:
[A, B] = AB − BA = 0 =⇒ AB = BA.
However, if the commutator is nonzero, then obviously the operators do not commute.
The position and momentum operators are an example of noncommuting opera-
tors. It can be demonstrated that:
[q, p] = qp− pq = q~i
∂
∂q− ~i
∂
∂qq.
[q, p]ψ = q~i
∂ψ
∂q− ~i
∂
∂q(qψ) = q
~i
∂ψ
∂q− ~i
∂q
∂qψ − q~
i
∂ψ
∂q= −~
iψ.
Ergo, the commutator relation for q and p is
amplitude of the wavefunction and S is the Action. Let’s assume a simple, one dimensional casewhere R = R(t) and S = S(q, t). Then we would have:
pψ =~i
∂
∂qReiS/~ =
~iReiS/~
i
~∂S
∂q=∂S
∂qψ.
From CM, we know that p = ∂S∂q , so the shoe fits. Please note that this exercise is a bit of an over-
simplification. Generally, R = R(t) is not considered a legitimate choice because it can lead tononnormalizable wavefunctions (see Tipler 2010 for further discussion.) For the sake of a simpledemonstration of how operators acting on wavefunctions lead to eigenvalues measured in the labora-tory, I chose this oversimplified approach to highlight connections between ideas .
9
[q, p]ψ = −~iψ = i~ψ =⇒ [q, p] = i~. (8)
This is the famous Heisenberg Canonical Commutator (Shankar 1988). In section
5.2 we will derive the Heisenberg Uncertainty Principle from the Canonical Com-
mutator. The following are some important commutator relationships given below
with derivation for brevity:
[A, B] = −[B, A]. (9)
[cA, B] = [A, cB] = c[A, B]. (10)
[AB, C] = A[B, C] + [A, C]B. (11)
[A, BC] = B[A, C] + [A, B]C. (12)
[A, B + C] = [A, B] + [A, C]. (13)
2.3 Expectation Values and the Eigenvalue Problem of Quantum
Mechanics
What exactly are the eigenvalues that we get out of the eigenvalue problem? I have
stated before that they are linked to measured observables, but in what way exactly?
Observables can be anything you would like to measure, such as the momentum or
position of a quantum particle.
Eigenvalues are going to be the numbers you measure directly inside the labora-
tory. If you conduct an experiment on a quantum system, the eigenvalues will be the
exact numbers you record. Remember that for a Hermitian operator, its eigenvalues
10
will always be real numbers. This is critical, because any number you measure in an
experiment will be real instead of, say, complex.
Eigenvalues are connected to finding the expectation values of an experiment.
Imagine you conduct an experiment on a quantum system and you measure the po-
sition or momentum. Instead of predicting individual momenta or position, QM can
predict the weighted mean value that you should expect. If you took many measure-
ments on identical systems, and took the weighted average of all those measurements,
the value you get out of this is the expectation value.
I will demonstrate how expectation values arise in QM in the following. I will
use both the Hilbert Space formulism and the function ψ that some might be more
comfortable with. Both derivations will end up with identical conclusions.
First, start with the eigenvalue expression Aψ = Aψ and multiply both sides by
the complex conjugate, like so:
ψ∗Aψ = ψ∗Aψ = Aψ∗ψ
In Hilbert Space, the same can be achieved by multiplying the eigenvalue expression
A |ψ(t)〉 = A |ψ(t)〉 with a bra vector from the left, so that the operator is ‘sand-
wiched’:
〈ψ(t)| A |ψ(t)〉 = 〈ψ(t)|A |ψ(t)〉
For the typical function form, we are going to multiply both sides by the ‘operator’´∞−∞ dx, in order to get:
ˆ ∞−∞
ψ∗Aψdx =
ˆ ∞−∞
Aψ∗ψdx (14)
We now introduce a fundamental postulate of QM, namely, the Born Rule. The
Born Rule states that ρ = ψ∗ψ where ρ represents the probability density for contin-
uous ψ. From probability theory, we know that if a probability density is integrated
11
over all possible states, it must be equal to 1 (the renormalization condition):
ˆ ∞−∞
ψ∗ψdx = 1 (15)
In other words, the probability of something happening is certain (= 1) if all possible
outcomes occur. The equivalent renormalization condition in Hilbert Space is that
〈ψ(t)|ψ(t)〉 = 1.5
Equation (14) is defined to be the expectation value of the observable given by
operator A, as can be seen from the similarity of both:
〈f(x)〉 =
ˆ ∞−∞
f(x)ρ dx⇐⇒ˆ ∞−∞
ψ∗Aψdx =
ˆ ∞−∞
Aψ∗ψdx,
where you can see that the eigenvalueA takes on the role of f(x) in the above expres-
sion and ψ∗ψ is the probability density. Please keep in mind that there could be many
different values you can measure for observable A, so it makes sense it is equivalent
to the function f(x). Therefore the expectation value of observable A is given by
〈A〉 =
ˆ ∞−∞
ψ∗Aψdx = 〈ψ(t)| A |ψ(t)〉 . (16)
The expectation values measured in the laboratory happen to be the exact ones pre-
dicted by the above equation. The expectation value expression was strictly derived
from the eigenvalue problem and the Born Rule.5Both renormalization conditions can be shown to be equivalent. From Appendix A, we have the
important completeness theorem, which says:
I =
ˆdx |x〉 〈x|
Since I is just the identity operator (operator where IA = A or I |ψ〉 = |ψ〉, just like multiplyingsomething by the number ‘1’), we can do the following operations:
〈ψ(t)|ψ(t)〉 = 〈ψ(t)| I |ψ(t)〉 =ˆ〈ψ(t)|x〉 〈x|ψ(t)〉 dx
Recall from before that we have said that 〈x|ψ(t)〉 = ψ(x, t), so we get:
〈ψ(t)|ψ(t)〉 =ˆψ∗ψdx
12
2.4 Review of Dirac Delta Functional
The Dirac delta functional is important to QM for the renormalization of all the basis
vectors. For any Hilbert Space, one can find an orthogonal set of bases to represent it.
But we can do one better - one can always find an orthonormal basis. An orthonormal
basis is a basis where not only are all the basis vectors orthogonal, but they are also
normalized to unit length 1:
|A〉normalized =|A〉√〈A|A〉
(17)
Any two generic vectors |A〉 and |A′〉 in Hilbert Space are orthonormal if and only if
they satisfy
〈A|A′〉 = δAA′ =
1 if |A〉 = |A′〉
0 if |A〉 6= |A′〉
where δAA′ is the Kronecker delta. The Kronecker delta functional is however only
appropriate when dealing with strictly discrete systems. Many problems in QM, how-
ever, can be either discrete or continuous. We therefore need a functional that can
encompass both discrete and continuous systems to describe an orthonormal basis.
This is done by using the Dirac delta functional6:
. 〈A|A′〉 = δ(A− A′) (18)
The Dirac delta functional is a generalization of the Kronecker delta (Shankar 1988).
The Dirac delta functional obeys the following properties:
δ(A− A′) = 0 if A 6= A′.
ˆ b
a
δ(A− A′) dA′ = 1 if a ≤ A ≤ b.
6Please note that the A here is a generic placeholder variable with arbitrary units. It could bea placeholder for position q, momentum p, or some other measurable quantity. So you could, forexample, have 〈q|q′〉 = δ(q − q′) for position and 〈p|p′〉 = δ(p− p′) for momentum, etc.
13
The functional is zero everywhere but at A′ (Shankar 1988).
The Dirac delta functional should not be treated as a regular function. It has its
own unique properties and form, distinct from how functions behave. For example,
the area under a single point when A = A′ should be zero (the area under a point is
zero). But for the Dirac delta functional, it is precisely 1. This and other properties
of this functional highlight the fact that it should not by treated like the functions
you have seen up to this point. It is a functional, not a function. It is, however,
fundamental to the renormalization of vectors in Hilbert Space.
Other important properties of the functional we will utilize in the text of this
article include the following:
ˆδ(A− A′)f(A′)dA′ = f(A). (19)
δ(A− A′) = δ(A′ − A). (20)
d
dAδ(A− A′) = − d
dA′δ(A− A′). (21)
Some useful relations for QM:
dn
dAnδ(A− A′) =
ˆdω
2π(−iω)ne−iω(A−A′). (22)
ˆdn
dAnδ(A− A′)f(A′) dA′ = (−1)n
dnf(A)
dAn. (23)
δ(αA) =δ(A)
|α|, (24)
where α is a constant.
2.5 Summary of Quantum Mechanics and its Postulates
The postulates of QM can be reviewed as follows (Shankar 1988; Bondar et al. 2012):
14
1. The state of a quantum system (particle or collection of particles) is represented
by a vector |ψ〉 in complex Hilbert Space.
2. For any observable A, there is an associated Hermitian operator A with eigen-
value problem A |A〉 = A |A〉 where |A〉 is the specific eigenket and A is the
eigenvalue which is measured in the laboratory. Examples of common observ-
ables are position of a particle q and momentum p. The variables position q
and momentum p are represented by Hermitian operators q and p which obey
the following properties: 〈q| q |q′〉 = qδ(q−q′) and 〈q| p |q′〉 = −i~ ddqδ(q−q′).
3. Born Rule: For a continuous distribution, the probability density for a particle
is ρ = ψ∗ψ. Equivalently in Dirac Notation, it can be written that the proba-
bility density of observing A is ρ(A) ∝ 〈ψ|A〉 〈A|ψ〉 = ‖ 〈A|ψ〉 ‖2 where “∝”
is the proportionality sign. More precisely we could say that the Born Rule
dictates that
ρ(A) =〈ψ|A〉 〈A|ψ〉〈ψ|ψ〉
As long as the above equation meets the renormalization condition 〈ψ|ψ〉 = 1
(see equation (15) and associated footnote), the Born Rule in its final form is
often written as
ρ(A) = 〈ψ|A〉 〈A|ψ〉
For a discrete system, the Born Rule is written as:
P (A) = 〈ψ|A〉 〈A|ψ〉 ,
assuming renormalization, where P is the probability. For both the continu-
ous and discrete cases, the state of the system collapses from the initial state
|ψ(t)〉 to the final state |A〉 whenever a measurement of the quantum system is
undertaken. Under the Copenhagen Interpretation of QM, the collapse of the
wavefunction |ψ〉 to the specific eigenket |A〉 is instantaneous and irreversible.
If A is observed in experiment, the wavefunction collapses to the associated
15
eigenket |A〉.
4. The state space of a composite system is the tensor product of the subsystem’s
state spaces
There are a couple other important things to say about the above overview of QM.
Firstly, although not included as a postulate, the expectation value relationship (16)
plays a critical role inside of QM. They are of fundamental importance as we reviewed
in section 2.3.
Not discussed thoroughly is the centrality of the Schrodinger equation to QM.
The state of the quantum system |ψ〉 obeys the Schrodinger equation:
i~∂
∂t|ψ〉 =
[− ~2
2m
∂2
∂q2+ V (q)
]|ψ〉 = H |ψ〉 (25)
where H is the quantum Hamiltonian that bears great similarity to the classical Hamil-
tonian7, except that the momentum and position are replaced by the momentum and
position operators of QM:
H(q, p) =p2
2m+ V (q) =⇒ H(q, p) =
p2
2m+ V (q) = − ~2
2m
∂2
∂q2+ V (q)
Of important note, the Hamiltonian operator which acts on the wavefunction in equa-
tion (25) is Hermitian (Shankar 1988). Since it is Hermitian, its eigenvalues will al-
ways be real instead of complex or imaginary, as required for numbers measured in
the laboratory.
There will be a slight adjustment of notation from the previous couple sections.
Although it was convenient to previously write A |ψ〉 = A |ψ〉 as the eigenvalue
problem, it is more precise to write the eigenvalue problem as A |A〉 = A |A〉. As
described before, for an operator A there are specific eigenvalues and eigenkets that7That is, a classical Hamiltonian where the potential energy has no velocity or time dependence,
which takes on the form H(q, p) = p2
2m + V (q). Note that p is momentum and V (q) is a one dimen-sional potential energy.
16
satisfy the above equation. Although |ψ〉might be used to describe the specific eigen-
ket in the context of the eigenvalue problem, |ψ〉 is often the generic state vector of
the quantum system, so this might lead to some confusion. To highlight the fact that
only specific eigenkets |A〉 satisfy this relationship, A |A〉 = A |A〉will be used from
now on.
Two very important eigenvalue problems are q |q〉 = q |q〉 for position q and
p |p〉 = p |p〉 for momentum p. The expectation values of the position and momentum
measured in the laboratory will be the eigenvalues q and p. For any observable A,
there is a relationship A |A〉 = A |A〉 for which combined with the Born Rule axiom,
you can deduce the expectation value of A (as described in section 2.3).
One of the initially confusing things about the Dirac notation is that the symbols
inside the bras and kets are just labels. So you might see vector bras and kets with
labels such as |1st〉, |2nd〉, |I〉, |II〉, |ω = 2.5〉, etc. All of these have valid nota-
tion, as the labels simply describe the associated vector, and that is all. Usually, the
eigenvalue is used to label the associated eigenket. Hence the eigenvalue ‘A’ being
associated with eigenket |A〉 in the eigenvalue problem. It is why in equation (2), we
could make a statement such as |v〉 + |w〉 = |v + w〉, because both v and w are just
labels.
Finally, the collapse of the wavefunction describes one of the ongoing mysteries of
QM. It can be seen that the Schrodinger equation is fully deterministic from equation
(25), but the collapse of the wavefunction is (as far as we know) indeterministic and
random. There is no known formula to describe how the wavefunction collapses.
More will be said about wavefunction collapse in section 5.3 of this article, in the
context of QM and CM.
2.6 The Density Operator
Before we delve straight into KvNM, it is critical to briefly look at one more tool
that will be of use to us in this work. Instead of representing the state of a system as
a wavefunction vector in Hilbert Space |ψ〉, we can instead use something called a
17
density matrix for certain systems, represented by ρ(t) (not to be confused with the
previously used ρ for probability density.) The density matrix is defined by
ρ(t) =∑j
pj |ψj(t)〉 〈ψj(t)| , (26)
for the discrete case. The density matrix is used when you are not certain of the
configuration of your system at some point in time. You can represent it as statistical
ensemble of possible wavefunction configurations |ψj(t)〉, each with probability of
occurrence pj . Note with probability, we must have
∑j
pj = 1.
We will see an example of the density matrix used in action later in section 5.3.
Density matrices are good for describing the above described so-called “mixed states”
of a system (with the “pure states” being the individual wavefunctions |ψj〉). If j = 1
in (26) the density matrix represents a pure state, but if it is greater than 1 it represents
a mixed state.
The probability of measuring A, or P (A), is given by the density matrix formu-
lation as follows:
P (A) = Tr[ρ(t) |a〉 〈a|], (27)
where Tr[...] refers to the trace of the matrix. You can represent the trace by
Tr[...] =∑n
〈n| ... |n〉 , (28)
where |n〉 are arbitrary basis of your choice. This is the density matrix version of the
Born Rule. It is easy to see that this is fully equivalent to the Born Rule, for instance,
when we plug (26) with j = 1 into (27):
P (A) = 〈a|ψ1〉 〈ψ1|a〉 〈a|a〉 = | 〈A|ψ1(t)〉 |2, (29)
18
since 〈a|a〉 = 1 assuming orthonormality. The expectation value of operator A is
given by
〈A〉 = Tr[Aρ(t)
], (30)
where Aρ(t) form a matrix and Tr[...] refers to the trace of the matrix. You can reduce
(30) to (16) just by applying the trace formula (28), easily demonstrated for the pure
state j = 1:
Tr[Aρ(t)
]=∑n
〈n| A |ψ1(t)〉 〈ψ1(t)|n〉 =∑n
〈ψ1(t)|n〉 〈n| A |ψ1(t)〉 .
Applying resolution of identity (Theorem II of Appendix A) we get the standard ex-
pectation value formula:
Tr[Aρ(t)
]= 〈ψ1(t)| A |ψ1(t)〉 = 〈A〉.
Other useful properties of the density matrix:
1. Tr[ρ(t)] = 1, which is the normalization condition.
2. Tr[ρ(t)2] = 1 for pure state.
3. Tr[ρ(t)2] < 1 for mixed state.
4. ρ(t) = ρ(t)†, in other words, ρ(t) is Hermitian.
(Based on MIT OpenCourseWare 2006.) We will utilize the density operator in future
sections.
3 Introduction to KvN Mechanics
Koopman-von Neumann Mechanics is unique because it is a form of classical statis-
tical mechanics that truly takes on a lot of the mathematical structure of QM (Koop-19
man 1931; von Neumann 1932; Bondar et al. 2012). It was invented in the 1930’s by
Bernard Koopman and John von Neumann. Subsequently forgotten for some time by
most physicists, its resurgence raises intriguing questions about the nature of QM to
CM. Just like with QM, we have a set of operators that act on vectors in Hilbert Space
in order to retrieve eigenvalues of observables (Koopman 1931; von Neumann 1932;
Bondar et al. 2012). In order to have a proper Hilbert Space, we first must define the
inner product of that Hilbert Space (Koopman 1931):
〈ψ|φ〉 =
ˆdA 〈ψ|A〉 〈A|φ〉 =
ˆdA ψ∗(A)φ(A).
CM always has distinct trajectories that particles travel along. In QM, we have the
Heisenberg Uncertainty Principle which states you cannot have both a well defined
position of a particle and traveling momentum (or velocity) at the same time. Un-
der the Copenhagen Interpretation, this is not due to some principle of measurement,
rather, this is a fundamental property of the fabric of reality (Shankar 1988). Reality
forbids quantum particles from having a precise momentum and position simultane-
ously. In QM, trajectories could be viewed as being “fuzzy” instead of well defined
and localized like in CM. You can no longer draw nice curves to describe how a
particle travels in space under QM.
Because of this fact, both position and momentum would have a common set of
eigenstates in KvN Hilbert Space (Bondar et al. 2013; McCaul and Bondar 2021).
The state vectors for KvN would therefore have the following properties (Koopman
1931):
|A〉 = |q, p〉 = |q〉 ⊗ |p〉 ,
q |q, p〉 = q |q, p〉 ,
p |q, p〉 = p |q, p〉 ,
20
〈q, p|q′, p′〉 = δ(q − q′)δ(p− p′),
ˆdp dq |q, p〉 〈q, p| = I. (31)
The observable A’s eigenket |A〉 is the KvN ket |q, p〉 because of the common set of
eigenstates between q and p. Because it contains information about q and p, this is
providing information in phase space instead of the previously discussed configura-
tion space (Koopman 1931). Previously for configuration space we had 〈q|ψ(t)〉 =
ψ(q, t),8 whereas here we will have 〈q, p|ψ(t)〉 = ψ(q, p, t) in phase space.
In QM, we have the Schrodinger equation (25) to dictate to us the time evolution of
the quantum particle described by |ψ〉. To estimate the probability of an observable,
we utilize ρ(A, t) = ψ∗(A, t)ψ(A, t) = | 〈A|ψ(t)〉 |2 (Born Rule). In CM, we have the
following equation known as the Louiville Equation to describe classical trajectories:
∂ρ
∂t+∑(∂H
∂pj
∂ρ
∂qj− ∂H
∂qj
∂ρ
∂pj
)= 0 (32)
The Louiville Equation is important because it tells you the probability density ρ of
finding a particle in observable point ω = (q, p) in phase space. In other words, it
defines the probabilistic time evolution of an ensemble of particles. We can define
the Louiville operator to be (Mauro 2003):
L =∑(
− i∂H∂pj
∂
∂qj+ i
∂H
∂qj
∂
∂pj
)(33)
By doing so, we can rewrite equation (32) in operational terms:
i∂ρ
∂t= Lρ (34)
Notice each term in equation (33) has been multiplied by the imaginary unit i. Just8Which can be generalized into 〈q1, q2, q3, ..., qn|ψ(t)〉 = ψ(q1, q2, q3, ..., qn, t) for configuration
space.
21
like in QM, we postulate a Born Rule for this probability density. If ρ(A, t) = ψ∗ψ
(where A is the observable), then we can easily show that 9
i∂ψ
∂t= Lψ (35)
This equation, of course, bears a striking resemblance to equation (25), the Schrodinger
equation in its operatorial form. The classical wavefunction ψ here is interpreted as a
probability density amplitude, like under most interpretations of QM (Shankar 1988).
Just like with QM, KvN depends on the generic state vector |ψ(t)〉 containing statis-
tical information about the system. Colloquially, just as before, ψ =⇒ |ψ(t)〉.
An important property of KvN Mechanics is that the Louiville operator L is Her-
mitian, just like the Hamiltonian operator of the Schrodinger equation. The norm
of 〈ψ(t)|ψ(t)〉 =´dωψ∗(ω)ψ(ω) is therefore conserved, further strengthening the
Born’s Rule postulated for KvN Mechanics (Mauro 2003).
The Axioms of KvN Mechanics can be summarized as follows (based off Bondar
et al. 2012):
1. The state of a system is represented by a vector |ψ〉 in a complex Hilbert Space.
2. For any observable A, there is an associated Hermitian operator A with eigen-
value problem A |A〉 = A |A〉 where |A〉 is the specific eigenket and A is the
eigenvalue measured in the laboratory.
3. Born Rule: The probability of measuring A is given by P (A) = | 〈A|ψ(t)〉 |2.
The wavefunction |ψ〉 describing the system instantaneously collapses into the9Demonstrated as follows. If ρ = ψ∗ψ, then for a simple case with index j = 1 we plug it into
equation (34):
i∂ψ∗ψ
∂t+ i
∂H
∂p
∂ψ∗ψ
∂q− i∂H
∂q
∂ψ∗ψ
∂p= 0
i∂ψ∗
∂tψ + i
∂ψ
∂tψ∗ + i
∂H
∂p
∂ψ∗
∂qψ + i
∂H
∂p
∂ψ
∂qψ∗ − i∂H
∂q
∂ψ∗
∂pψ − i∂H
∂q
∂ψ
∂pψ∗ = 0
[i∂ψ∗
∂t+ i
∂H
∂p
∂ψ∗
∂q− i∂H
∂q
∂ψ∗
∂p
]ψ +
[i∂ψ
∂t+ i
∂H
∂p
∂ψ
∂q− i∂H
∂q
∂ψ
∂p
]ψ∗ = 0
Since generally neither ψ nor ψ∗ are zero, this means the expressions in square brackets must equalzero. The bracketed expressions are, of course, equation (35) and its complex conjugate.
22
eigenket |A〉 associated with the observed eigenvalue A.
4. The state space of a composite system is the tensor product of the subsystems’
state space
Notice, these are exactly the same postulates of QM in section 2.5, just with anything
“quantum” removed (Bondar et al. 2012). The formula for the expectation value of
the system is still the same, since Axioms 2 and 3 remain unchanged, and therefore
the derivation in section 2.3 is the same. The very important expectation value rule
(16) therefore still applies in KvNM.
Given these four ‘universal’ postulates, one can derive both QM and KvNM from
the Ehrenfest Theorems. The Ehrenfest Theorems are as follows:
d
dt〈q〉 =
〈p〉m, (36)
d
dt〈p〉 = 〈−V ′(q)〉, (37)
and are usually considered strictly within the context of QM.
Under QM, the fabric of reality forbids the mutual existence of a well defined
position and momentum. Under CM, we can know the position and momentum to an
arbitrary precision. This difference is encoded with the commutator relationship. In
section 5.3, we will show that the Canonical Commutator of QM (8) directly produces
the Heisenberg Uncertainty Principle. We can also write a commutator relationship
for CM under the axioms of KvNM. We can say:
[q, p] = 0 (38)
I shall call this the “Classical Commutator”. This commutator relationship would
produce no uncertainty relationship between the classical position q and momentum
p. Ergo, position and momentum can be arbitrarily known.
For QM, one can show that the Canonical Commutator combined with the Ehren-
fest Theorem produces the traditional Schrodinger equation with quantum Hamilto-
23
nian. We will demonstrate this as follows. According to the common set of Axioms,
we can say that the expectation value of observable A is simply 〈ψ(t)| A |ψ(t)〉 (16).
Therefore we can easily say that
〈q〉 = 〈ψ(t)| q |ψ(t)〉
〈p〉 = 〈ψ(t)| p |ψ(t)〉
Then from the Ehrenfest Theorems one can carry out the following operations:
d
dt〈ψ(t)| q |ψ(t)〉 =
1
m〈ψ(t)| p |ψ(t)〉 ,
d
dt〈ψ(t)| p |ψ(t)〉 = 〈ψ(t)| − V ′(q) |ψ(t)〉 .
We are using the Schrodinger picture of QM, so we assume time dependence
solely resides in |ψ(t)〉 and any operators are independend of time. 10 Using the
chain rule of calculus, we write:
⟨dψ(t)
dt
∣∣∣∣ q |ψ(t)〉+ 〈ψ(t)| q∣∣∣∣dψ(t)
dt
⟩=
1
m〈ψ(t)| p |ψ(t)〉 ,
⟨dψ(t)
dt
∣∣∣∣ p |ψ(t)〉+ 〈ψ(t)| p∣∣∣∣dψ(t)
dt
⟩= 〈ψ(t)| − V ′(q) |ψ(t)〉 . (39)
Stone’s Theorem (see Appendix B) dictates that there exists a unique Schrodinger-like
equation i ∂∂t|ψ〉 = G |ψ〉 where G is the Hermitian generator of motion in operator
form. We deduce there must be a G that governs the evolution of CM, just as the
Hamiltonian operator does for QM. The only thing is to figure out what this generator
G is.
For QM, we can derive what the generator of motion is by using the commutator
relationships. We can take Stone’s Theorem and then derive:10This is the opposite of the Heisenberg picture, which assumes no time dependence in the wave-
function vector ket and puts all the time dependence inside the operators!
24
∣∣∣∣dψ(t)
dt
⟩=H
i~|ψ(t)〉 ⇐⇒
⟨dψ(t)
dt
∣∣∣∣ = −〈ψ(t)| Hi~
Which means:
−〈ψ(t)| Hi~q |ψ(t)〉+ 〈ψ(t)| q H
i~|ψ(t)〉 =
1
m〈ψ(t)| p |ψ(t)〉
− 〈ψ(t)| Hi~p |ψ(t)〉+ 〈ψ(t)| p H
i~|ψ(t)〉 = 〈ψ(t)| − V ′(q) |ψ(t)〉
You can expand the vector ‘sandwich’ and rewrite it in terms of a commutator to get:
〈ψ(t)| − Hq + qH |ψ(t)〉 = −〈ψ(t)| [H, q] |ψ(t)〉 =i~m〈ψ(t)| p |ψ(t)〉
〈ψ(t)| − Hp+ pH |ψ(t)〉 = −〈ψ(t)| [H, p] |ψ(t)〉 = 〈ψ(t)| − i~V ′(q) |ψ(t)〉
Since the Ehrenfest Theorems are true regardless of the particular state of the system
|ψ(t)〉 (i.e., they are true for any state you might find), we can remove the wavefunc-
tion vector from both sides to get:
−[H, q] =i~mp
[H, p] = i~V ′(q) (40)
We will make the assumption that H = H(q, p) and utilize Theorem VI from Ap-
pendix A. By this Theorem, we have q = A1, p = A2, and H(q, p) = f(A1, A2).
Therefore:
[H, q] = [H(q, p), q] =n=2∑k=1
[Ak, q]∂H(A1, A2)
∂Ak= [q, q]
∂H(q, p)
∂q+ [p, q]
∂H(q, p)
∂p
25
[H, p] = [H(q, p), p] =n=2∑k=1
[Ak, p]∂H(A1, A2)
∂Ak= [q, p]
∂H(q, p)
∂q+ [p, p]
∂H(q, p)
∂p
A commutator with itself will always equal zero:
[q, q] = qq − qq = 0
[p, p] = pp− pp = 0
Ergo, the above expressions based in Theorem VI reduce to:
[H, q] = [p, q]∂H(q, p)
∂p
[H, p] = [q, p]∂H(q, p)
∂q
Plugging these expressions into (40) gives us:
−[p, q]∂H(q, p)
∂p=i~mp
[q, p]∂H(q, p)
∂q= i~V ′(q) (41)
From the Canonical Commutator we know that [q, p] = i~.11 This step is absolutely
critical. It is in fact the step that is going to distinguish QM from classical KvNM
- the usage of the Canonical Commutator. We are going to plug in the Canonical
Commutator to the above expressions, so that we will get:
∂H(q, p)
∂p=
p
m
11We will also use the fact that [q, p] = −[p, q]. It’s not hard to prove, just takes a couple lines ofalgebra:
[q, p] = qp− pq = −(−qp+ pq) = −(pq − qp) = −[p, q]
26
∂H(q, p)
∂q=dV (q)
dq
Theorem VII of Appendix A will be useful, as we can now translate this expression
with operators into more familiar analytical functions in order to do operations with
derivations. We will have the following analytical functions as a result of Theorem
VII:
∂H(q, p)
∂p=
p
m
∂H(q, p)
∂q=dV (q)
dq
Some basic calculus reveals:
ˆ∂H(q, p)
∂pdp = H(q, p) =
ˆp
mdp =
p2
2m+ C(q)
whereC(q) is some function solely in terms of q so that it disappears when the partial
with respect to p is taken. Taking the partial derivative with respect to q of the above
expression reveals:
∂H(q, p)
∂q=dC(q)
dq=dV (q)
dq
Ergo, C(q) = V (q) so that the final solution isH(q, p) = p2
2m+V (q), which happens
to be equivalent to the quantum Hamiltonian. Therefore, plugging this solution of H
into the above version of Stone’s Theorem we reclaim the Schrodinger equation(25):
i~∂
∂t|ψ〉 =
[ p2
2m+ V (q)
]|ψ〉
The same series of steps can be carried out to derive KvNM. The only differ-
ence in the derivation is that we will assume the Classical Commutator instead of
the Canonical Commutator. Once again we start with the Ehrenfest Theorems and
follow a similar series of steps. Starting from (39), we will utilize Stone’s Theorem
27
as i ∂∂t|ψ〉 = K |ψ〉 to get:
−〈ψ(t)| Kiq |ψ(t)〉+ 〈ψ(t)| q K
i|ψ(t)〉 =
1
m〈ψ(t)| p |ψ(t)〉
− 〈ψ(t)| Kip |ψ(t)〉+ 〈ψ(t)| p K
i|ψ(t)〉 = 〈ψ(t)| − V ′(q) |ψ(t)〉
Following a similar series of steps, we can derive that:
−[K, q] =i
mp
[K, p] = iV ′(q) (42)
Now we run into a problem. If we assume K = K(q, p) as before, then we will derive
once again (41). However, now we will utilize the Classical Commutator instead of
the Canonical Commutator, since we want well defined classical trajectories. By
imposing the Classical Commutator [q, p] = 0, we run into a contradiction, namely:
−[p, q]∂K(q, p)
∂p= (0)
∂K(q, p)
∂p= 0 =
i
mp =⇒ p = 0
[q, p]∂K(q, p)
∂q= (0)
∂K(q, p)
∂q= 0 = iV ′(q) =⇒ dV
dq= 0
Obviously, this cannot be true for all systems, because many classical systems def-
initely have a nonzero momentum. This contradiction between theory and observa-
tion suggests that we must approach this from a different angle. One such angle is to
propose two new operators θ and λ (Bondar et al. 2012). These operators obey the
eigenvalue problems, θ |θ〉 = θ |θ〉 and λ |λ〉 = λ |λ〉, by postulate 2 of the ‘universal
axioms’, but are not tied to any direct physically observable features of reality (as
noted in Ramos-Prieto et al. 2018).12 They serve as an analogue to Lagrange Multi-
pliers (McCaul and Bondar 2021), serving to constrain the realm of possible behavior
to a very specific subset.12Even though these operators are not directly physically observable, their existence has been seen
through indirect means. These operators are the famous Bopp Operators of the phase-space formula-tion of QM (Bondar et al. 2013). We will explore the Wigner Distribution in section 6.
28
For KvNM, we will postulate the following behavior for these operators:
[q, θ] = [p, λ] = i
[qc, pc] = [q, λ] = [p, θ] = [θ, λ] = 0 (43)
where i is the imaginary unit. Equations (43) make up what is known as the Koopman-
von Neumann Algebra. If we do this, then we can assume K = K(q, p, θ, λ). By
Theorem VI of Appendix A again, we will get:
[K(q, p, θ, λ), q] =n=4∑k=1
[Ak, q]∂K(A1, A2, A3, A4)
∂Ak= [q, q]
∂K(q, p, θ, λ)
∂q+[p, q]
∂K(q, p, θ, λ)
∂p
+[θ, q]∂K(q, p, θ, λ)
∂θ+ [λ, q]
∂K(q, p, θ, λ)
∂λ
[K(q, p, θ, λ), p] =n=4∑k=1
[Ak, p]∂K(A1, A2, A3, A4)
∂Ak= [q, p]
∂K(q, p, θ, λ)
∂q+[p, p]
∂K(q, p, θ, λ)
∂p
+[θ, p]∂K(q, p, θ, λ)
∂θ+ [λ, p]
∂K(q, p, θ, λ)
∂λ
Three terms from each expression will disappear because of the KvN Algebra above
and the Classical Commutator. The equations of motion will end up being:
[θ, q]∂K(q, p, θ, λ)
∂θ= − i
mp
[λ, p]∂K(q, p, θ, λ)
∂λ= iV ′(q) (44)
Once we substitute in [q, θ] = [p, λ] = i we will simply have:
∂K(q, p, θ, λ)
∂θ=
p
m
29
∂K(q, p, θ, λ)
∂λ= −V ′(q)
Utilizing Theorem VII again, we get the differential equations:
∂K(q, p, θ, λ)
∂θ=
p
m
∂K(q, p, θ, λ)
∂λ= −V ′(q)
Which we proceed to solve:
ˆ∂K(q, p, θ, λ)
∂θdθ = K(q, p, θ, λ) =
ˆp
mdθ =
pθ
m+ C(q, p, λ)
∂
∂λK(q, p, θ, λ) =
∂
∂λ(pθ
m+ C(q, p, λ)) =
∂C(q, p, λ)
∂λ= −V (q)
dqˆ∂C(q, p, λ)
∂λdλ = C(q, p, λ) = −
ˆV (q)
dqdλ = −V (q)
dqλ+ C(q, p)
We can therefore derive
K = K(q, p, θ, λ) =pθ
m− V ′(q)λ+ C(q, p) (45)
which is the important Koopman Generator of KvNM. The Koopman Generator will
become very important when we discuss the Path Integral Formulation of QM and
CM (section 4).
We can show that the Koopman Generator gives us the Louiville Equations again
and produces the KvN classical wavefunction. To show this is quite straightforward.
Since the Born Rule for KvNM postulates that ρ = | 〈q, p|ψ(t)〉 |2, let us try to use the
Koopman Generator to find an expression that governs the evolution of the classical
probability density.
First, refer to (31). From these basic properties, one can easily show the following:
〈q, p| q |ψ(t)〉 = [〈q, p| q] |ψ(t)〉 = 〈q, p| q |ψ(t)〉 = q 〈q, p|ψ(t)〉 (46)
30
〈q, p| p |ψ(t)〉 = [〈q, p| p] |ψ(t)〉 = 〈q, p| p |ψ(t)〉 = p 〈q, p|ψ(t)〉 (47)
Using Theorem IV from Appendix A, we can demonstrate that the following are true.
Since [q, θ] = i, we can set κ = 1, |A〉 = |q, p〉, and B = θ for the parameters in
Theorem IV. We can say that the following is true then:
〈q, p| θ |ψ(t)〉 = −i ∂∂q〈q, p|ψ(t)〉 (48)
Notice that although A = (q, p), we do not write a partial derivative with respect to
p, and that is because [p, θ] = 0 per the Koopman Algebra.
The same can be done with [p, λ] = i. Using Theorem IV with parameters κ = 1,
|A〉 = |q, p〉, and B = λ, we get:
〈q, p| λ |ψ(t)〉 = −i ∂∂p〈q, p|ψ(t)〉 (49)
Again, there is no partial with respect to q here, since [q, λ] = 0 per the Koopman
Algebra.
Plugging the Koopman Generator into Stone’s Theorem gives us:
i∂
∂t|ψ(t)〉 = K |ψ(t)〉 =
[ pθm− V ′(q)λ+ C(q, p)
]|ψ(t)〉
And now we will ‘sandwich’ it from the left hand side with the vector bra 〈q, p| and
work out the algebra:
〈q, p| i ∂∂t|ψ(t)〉 = 〈q, p|
[ pθm− V ′(q)λ+ C(q, p)
]|ψ(t)〉
i∂
∂t〈q, p|ψ(t)〉 = 〈q, p| pθ
m|ψ(t)〉 − 〈q, p|V ′(q)λ |ψ(t)〉+ 〈q, p|C(q, p) |ψ(t)〉
31
Utilizing Theorems I of Appendix A13and equations (46),(47),(48) and (49), we can
derive:
i∂
∂t〈q, p|ψ(t)〉 =
p
m〈q, p| θ |ψ(t)〉 − V ′(q) 〈q, p| λ |ψ(t)〉+ C(q, p) 〈q, p|ψ(t)〉
i∂
∂t〈q, p|ψ(t)〉 = −i p
m
∂
∂q〈q, p|ψ(t)〉+ iV ′(q)
∂
∂p〈q, p|ψ(t)〉+ C(q, p) 〈q, p|ψ(t)〉
(50)
We can derive the time evolution equation for the probability density by utilizing
equation (50) above. This will be done in this fashion:
i∂
∂t| 〈q, p|ψ(t)〉 |2 = i
∂
∂t[〈ψ(t)|q, p〉 〈q, p|ψ(t)〉] =
= i∂
∂t[〈ψ(t)|q, p〉] 〈q, p|ψ(t)〉+ i 〈ψ(t)|q, p〉 ∂
∂t[〈q, p|ψ(t)〉]
Into this equation plug in (50) and its complex conjugate:
i∂
∂t| 〈q, p|ψ(t)〉 |2 =
[−i pm
∂
∂q〈ψ(t)|q, p〉+iV ′(q) ∂
∂p〈ψ(t)|q, p〉−C(q, p) 〈ψ(t)|q, p〉
]〈q, p|ψ(t)〉
+ 〈ψ(t)|q, p〉[− i p
m
∂
∂q〈q, p|ψ(t)〉+ iV ′(q)
∂
∂p〈q, p|ψ(t)〉+C(q, p) 〈q, p|ψ(t)〉
]=
= −i pm〈q, p|ψ(t)〉 ∂
∂q〈ψ(t)|q, p〉+ iV ′(q) 〈q, p|ψ(t)〉 ∂
∂p〈ψ(t)|q, p〉+
−〈q, p|ψ(t)〉C(q, p) 〈ψ(t)|q, p〉 − i 〈ψ(t)|q, p〉 pm
∂
∂q〈q, p|ψ(t)〉
+i 〈ψ(t)|q, p〉V ′(q) ∂∂p〈q, p|ψ(t)〉+ 〈ψ(t)|q, p〉C(q, p) 〈q, p|ψ(t)〉 =
13This Theorem tells us that V ′(q) |q, p〉 = V ′(q) |q, p〉 and C(q, p) |q, p〉 = C(q, p) |q, p〉
32
= −i pm〈q, p|ψ(t)〉 ∂
∂q〈ψ(t)|q, p〉 − i 〈ψ(t)|q, p〉 p
m
∂
∂q〈q, p|ψ(t)〉
+iV ′(q) 〈q, p|ψ(t)〉 ∂∂p〈ψ(t)|q, p〉+ i 〈ψ(t)|q, p〉V ′(q) ∂
∂p〈q, p|ψ(t)〉
Recognizing the involvement of chain rule applied to the wavefunction:
i∂
∂t| 〈q, p|ψ(t)〉 |2 = −i p
m
∂
∂q| 〈q, p|ψ(t)〉 |2 + iV ′(q)
∂
∂p| 〈q, p|ψ(t)〉 |2
Since | 〈q, p|ψ(t)〉 |2 = ρ, the probability density, we can see right away that this
equation is fully equivalent to the Louiville equation (32)14, and that 〈q, p|ψ(t)〉 =
ψ(q, p, t) is the previously identified KvN classical wavefunction in phase space.
In summary, KvNM is a classical theory of physics that utilizes complex Hilbert
Spaces, just like in QM. Like QM, it is probabilistic in nature, so it can only make
probabilistic predictions of classical systems. Like in QM, you have a wavefunction
that completely describes the state of the system (35). This wavefunction is in phase
space, and KvNM is a phase space theory. It operates on the same axioms as QM,
but is a fully classical theory. It is not an extension of classical theory, but fully
equivalent to it, recast in a non-traditional form. The mathematical conveniences of
QM exist in KvNM. For example, one reason QM has its odd Hilbert Space form is
to conveniently change between vector bases. One can do this in KvNM with equal
ease. Just like in QM, the expectation value of observable A is given by 〈A〉 =
〈ψ(t)| A |ψ(t)〉.
The backbone of KvNM lies in the standard Hilbert Space behaviors (31), the
Koopman Generator (45), and the Koopman Algebra (43), which lets one derive the
Louiville Equation governing classical probabilities via usage of the KvNM classical
wavefunction. One can derive KvNM from the Ehrenfest Theorems by assuming the14Recall from Classical Mechanics that
F = −∂H∂q
= −V ′(q)
dq
dt=∂H
∂p=
p
m
33
Classical Commutator. The main difference between Classical and Quantum Physics
appears to lie in the choice of the commutator for position q and momentum p (Bondar
et al. 2012).
In the pages that follow, we will demonstrate the many applications of KvNM.
Even though technically it is strictly equivalent to any classical theory, it allows for
a wide variety of applications and uses. KvNM is a very important and convenient
tool for statistical physics.
4 Path Integral Formulation of Quantum and Classi-
cal Mechanics
4.1 Feynman Path Integral of Quantum Mechanics
Feynman Path Integral is another formulation of QM that is completely equivalent to
the Schrodinger equation. It was invented by the famous physicist Richard Feynman
in 1948 and has applications many usages in QM, some which you will see in this
Thesis. Whereas the Schrodinger equation is more analogous to Hamiltonian Dy-
namics (for instance, see section 5.1 for the similarity to Hamilton-Jacobi Theory),
the Path Integral is more analogous to Lagrangian Dynamics (Shankar 1988).
Under Lagrangian Dynamics in Classical Theory, you select two points, a starting
point and an endpoint point. In the allowed time interval, the classical particle would
take the path where the function S, called the Action, would be minimized. From
CM, the term T − V is called the Lagrangian (where T is kinetic energy and V is
potential energy) and the Action is defined as the time integral of the Lagrangian:
S =
ˆ t
0
dt(T − V ) (51)
This function has to be minimized to define the classical path. The minimization rule
for classical particles - called the Least Action Principle or Hamilton’s Principle - is
often written as:
34
δ
ˆ t
0
dt(T − V ) = 0
Things are not so simple in Quantum Theory.
We begin with a simple derivation of the Feynman Path Integral to understand
the differences between QM and Lagrangian Theory. We start with the following
expression, which is simply a rewrite of the Schrodinger equation:
|ψ(t)〉 = e−i~ tH |ψ(0)〉 (52)
In Appendix C, one can see how this expression is completely equivalent to the
Schrodinger equation. One might also notice that the operator H is exponated, but
this is not unusual in QM (more information about exponated operators and what
they mean is found in the same Appendix). The term exp− i
~tH
is called the uni-
tary time-evolution operator and defines how the quantum particle will evolve in time
from its starting condition |ψ(0)〉 to its final condition |ψ(t)〉 at time t (see Appendix
B for more on unitary operators).
To begin, we insert a position q bra on both sides of (52):
〈q|ψ(t)〉 = ψ(q, t) = 〈q| e−i~ tH |ψ(0)〉 (53)
And then insert resolution of identity (Appendix A Theorem II) for initial position
q′:
〈q|ψ(t)〉 = ψ(q, t) =
ˆdq′ 〈q| e−
i~ tH |q′〉 〈q′|ψ(0)〉 =
ˆdq′ 〈q| e−
i~ tH |q′〉ψ(q′, 0)
(54)
The term 〈q| e− i~ tH |q′〉 is called the kernell of the Path Integral. The kernell rep-
resents the probability amplitude while traveling from the starting point to the final
point. Think of the |q′〉 = |q0〉 as the starting point in the Lagrangian formulism and
|q〉 = |qf〉 as the final prespecified point.
35
Let ∆t be a very small change in time, where ∆t = tj+1 − tj . We will break the
total allowed time t into N equal increments, so that we have ∆t = t/N . We can
therefore write:
〈q| e−i~ tH |q′〉 = 〈q| e−
i~ (∆t1+∆t2+...+∆tN )H |q′〉 = 〈q| e−
i~∆t1H− i
~∆t2H+...− i~∆tN H |q′〉
〈q| e−i~ tH |q′〉 = 〈q| e−
i~∆t1He−
i~∆t2H · · · e−
i~∆tN H |q′〉
The subscripts on each of the ∆t are there just to make clear that we are breaking the
total time t into a large number of N increments. Since the ∆t here are very small,
we will drop the subscripts from here on out. Just keep in mind there are now a total
N exponential terms inside the above expression.
Using Theorem II of Appendix A we can insert between each exponential the
resolution of identity:
〈qf | e−i~∆tHe−
i~∆tH · · · e−
i~∆tH |q0〉 = 〈qf | e−
i~∆tHIe−
i~∆tHI · · · Ie−
i~∆tH |q0〉 =
=
ˆ ˆ· · ·ˆdqN−1dqN−2 · · · dq1 〈qf | e−
i~∆tH |qN−1〉 〈qN−1| e−
i~∆tH |qN−2〉×
× 〈qN−2| · · · |q1〉 〈q1| e−i~∆tH |q0〉 (55)
We will take one of the “bra-ket” sandwiches above and do some further opera-
tions on it. We can represent each “sandwich” as 〈qj+1| e−i~∆tH |qj〉 and plug in the
value of the Hamiltonian operator H:
〈qj+1| e−i~∆tH |qj〉 = 〈qj+1| e−
i~∆t[ p
2
2m+V (q)] |qj〉 = 〈qj+1| e−
i~∆t p
2
2m e−i~∆tV (q) |qj〉
From Theorem I of Appendix A we know that:
36
e−i~∆tV (q) |qj〉 = e−
i~∆tV (qj) |qj〉
Next we apply the completeness theorem with momentum p:
〈qj+1| e−i~∆t p
2
2m Ie−i~∆tV (q) |qj〉 =
ˆdp 〈qj+1| e−
i~∆t p
2
2m |p〉 〈p| e−i~∆tV (qj) |qj〉
Again applying Theorem I we know:
e−i~∆t p
2
2m |p〉 = e−i~∆t p
2
2m |p〉
so that the whole thing becomes
〈qj+1| e−i~∆t p
2
2m e−i~∆tV (q) |qj〉 =
ˆdp 〈qj+1|p〉 〈p|qj〉 e−
i~∆t[ p
2
2m+V (qj)]
Since QM obeys the Canonical Commutator [q, p] = i~, we know what 〈qj+1|p〉
and 〈p|qj〉 are from Theorem V of Appendix A where we set κ = ~. Plugging these
results into the above equation we get:
〈qj+1| e−i~∆t p
2
2m e−i~∆tV (q) |qj〉 =
ˆdp e−
i~∆t[ p
2
2m+V (qj)]
1
2π~ei~pqj+1e−
i~pqj =
=
ˆdp e−
i~∆t[ p
2
2m+V (qj)]
1
2π~ei~p(qj+1−qj)
This is known as a Gaussian Integral. From Theorem VIII of Appendix A, we know
how to compute Gaussian Integrals. We simply set a = i~
∆t2m
, b = i~(qj+1 − qj), and
c = − i~∆tV (qj) in this Theorem and conduct the following steps:
1
2π~
ˆdp e−
i~
∆t2m
p2+ i~ (qj+1−qj)p− i
~∆tV (qj) =1
2π~
√π
aeb2
4a+c =
37
=1
2π~
√π
( i~∆t2m
)exp
[( i~(qj+1 − qj))2
4( i~∆t2m
)− i
~∆tV (qj)
]
∴ 〈qj+1| e−i~∆tH |qj〉 =
1
2π~
√2πm~i∆t
exp
[im(qj+1 − qj)2
2~∆t− i
~∆tV (qj)
](56)
Plugging equation (56) into (55) we get the following for the total time t:
〈qf | e−i~ tH |q0〉 =
( 1
2π~
)N(2πm~i∆t
)N2
ˆ ˆ· · ·ˆdqN−1dqN−2 · · · dq1×
× exp
[N−1∑j=0
(im(qj+1 − qj)2
2~∆t− i
~∆tV (qj)
)]
This hideous looking expression can be notationally simplified a bit by treating all
the integrals with respect to qj as a product of integrals
ˆ ˆ· · ·ˆdqN−1dqN−2 · · · dq1 =
N−1∏j=0
ˆdqj
where Π just means product from j = 0, 1, ..., N − 1, similarly like Σ means sum
from j = 0, 1, ..., N − 1. Compactly written:
〈qf | e−i~ tH |q0〉 =
( 1
2π~
)N(2πm~i∆t
)N2
N−1∏j=0
ˆdqj exp
[N−1∑j=0
( i~m(qj+1 − qj)2
2∆t− i
~∆tV (qj)
)](57)
Factoring out a factor of i∆t~ from the exponent:
〈qf | e−i~ tH |q0〉 =
( 1
2π~
)N(2πm~i∆t
)N2
N−1∏j=0
ˆdqj exp
[N−1∑j=0
(i∆t~
(m(qj+1 − qj)2
2∆t2− V (qj)
))](58)
Now, let us take the limit as ∆t → 0. Since ∆t = t/N , this is equivalent to taking
N →∞.
lim∆t→0
[qj+1 − qj∆t
]2
=(dqdt
)2
(59)
38
This is simply the definition of a derivative. And:
limN→∞
N−1∑j=0
=
ˆ t
0
(60)
We end up with:
〈qf | e−i~ tH |q0〉 =
( 1
2π~
)N(2πm~i∆t
)N2
N−1∏j=0
ˆdqj exp
[i
~
ˆ t
0
dt(m
2
(dqdt
)2
− V (q))]
(61)
Since q is the position, dqdt
is obviously the velocity. A quick glance will reveal that
the exponated term m2
(dqdt
)2 is simply the Kinetic Energy T . Ergo, if you compare the
exponent to (51), the last term becomes eiS/~, which should be familiar. It is exactly
the same term that appears in equation (3), the wavefunction represented in radial
form.
We can write some of the messy, ugly terms into a nice format. We say that
limN→∞
(−i2πm∆t
)N2
N−1∏j=0
ˆdqj =
ˆDq(t)
And therefore the entire thing can be written simply as:
〈qf | e−i~ tH |q0〉 =
ˆDq(t) exp
[i
~
ˆ t
0
dt(m
2
(dqdt
)2
− V (q))]
〈qf | e−i~ tH |q0〉 =
ˆDq(t) eiS/~ (62)
Equation (62) represents the final Kernel of the Feynman Path Integral. This inte-
gral is fascinating because it says that for a fixed starting point |q0〉 and ending point
|qf〉, the quantum particle takes all possible paths inbetween within the allotted time
interval. This is like saying that if you are throwing darts at a dart board, the flying
dart takes all possible paths from your hand to the bullseye. This shocking outcome
highlights another strange property of quantum particles. This goes hand in hand
with the earlier comments about quantum particles having “fuzzy” trajectories.
39
4.2 Koopman-von Neumann Classical Path Integral
Just like a Path Integral can be formulated for Quantum Theory, one can also formu-
late a path integral for classical theory. The path integral, however, must enforce New-
ton’s Laws and particles must travel along well defined paths instead of the “fuzzy”
trajectories of QM. This classical path integral is possible thanks to the KvN for-
mulism (Mauro 2003; McCaul and Bondar 2021).
First, in order to develop the KvN Path Integral, let us develop some initial use-
ful tools. These will be ingredients we will use in a future step in making the Path
Integral. Using Theorem V of Appendix A, you can compute what expressions like
〈q, λ|q′, p〉 are. LetA = p, B = λ, and κ = 1. Utilizing the Koopman-von Neumann
Algebra (43), you can get:
〈q, λ|q′, p〉 = 〈q| ⊗ 〈λ| |q′〉 ⊗ |p〉 = 〈q|q′〉 〈λ|p〉 =1√2πδ(q − q′)e−ipλ (63)
Note that there is no need to evaluate 〈q|p〉 and 〈λ|q′〉 since [q, p] = 0 and [λ, q] = 0.
From (18), recall 〈q|q′〉 = δ(q− q′). You can follow a similar procedure to calculate
the following two expressions:
〈θ, p|q, p′〉 = 〈θ| ⊗ 〈p| |q〉 ⊗ |p′〉 = 〈θ|q〉 〈p|p′〉 =1√2πδ(p− p′)e−iθq (64)
〈θ, p|q, λ〉 = 〈θ| ⊗ 〈p| |q〉 ⊗ |λ〉 = 〈θ|q〉 〈p|λ〉 =1
2πeipλ−iθq (65)
The three expressions above are critical ingredients which we will utilize soon.
Now we can begin with the proper formulation of the Path Integral. Similar to
(52), we will utilize
|ψ(t)〉 = e−itK |ψ(0)〉 (66)
40
with the Koopman Generator K (45) instead of the quantum Hamiltonian H .15 Since
KvNM deals with phase space, we will attack this expression with a q, p bra on the
left to get the following:
〈q, p|ψ(t)〉 = ψ(q, p, t) = 〈q, p| e−itK |ψ(0)〉 (67)
And like before, we will introduce the resolution of identity, this time with respect to
q′, p′:
〈q, p|ψ(t)〉 =
ˆdq′ dp′ 〈q, p| e−itK |q′, p′〉 〈q′, p′|ψ(0)〉 =
=
ˆdq′ dp′ 〈q, p| e−
i~ tK |q′, p′〉ψ(q′, p′, 0) (68)
The KvN Kernel is 〈q, p| e− i~ tK |q′, p′〉 with |q′, p′〉 = |q0, p0〉 being the initial state
of the system in phase space and |q, p〉 = |qf , pf〉 being the final state of the system.
Again, let ∆t be a small change in time. As before in the quantum case, it is a very
small increment where we will have ∆t = t/N . We will get:
〈q, p| e−itK |q′, p′〉 = 〈q, p| e−i∆t1Ke−i∆t2K · · · e−i∆tN K |q′, p′〉 (69)
The subscripts on each of the ∆t are there just to make clear that we are breaking the
total time t into a large number of N increments. Since the ∆t here are very small,
we will drop the subscripts from here on out. Just keep in mind there are now a total
N exponential terms inside the above expression.
Using Theorem II we can insert between each exponential a completeness rela-
tionship, as before:
〈qf , pf | e−i∆tKe−i∆tK · · · e−i∆tK |q0, p0〉 =
=
ˆ· · ·ˆdqN−1dpN−1 · · · dq1dp1 〈qf , pf | e−i∆tK |qN−1, pN−1〉×
15More information about exponated operators in Appendix C.
41
× 〈qN−1, pN−1| e−i∆tK |qN−2, pN−2〉 〈qN−2, pN−2| · · · |q1, p1〉 〈q1, p1| e−i∆tK |q0, p0〉
(70)
We will take one of the “bra-ket” sandwiches above and do some further operations
on it. We can represent each “sandwich” as 〈qj+1, pj+1| e−i∆tK |qj, pj〉 and plug in
the value of the Koopman Generator K:
〈qj+1, pj+1| e−i∆tK |qj, pj〉 = 〈qj+1, pj+1| exp
(−i∆t
[ pθm− V ′(q)λ
])|qj, pj〉 =
= 〈qj+1, pj+1| e−i∆tpθm ei∆tV
′(q)λ |qj, pj〉
Please note, in keeping with McCaul and Bondar 2021, we have omitted the constant
of integration C(q, p) from (45) in the following calculations. It will not affect the
dynamics of the experiment for classical systems (Bondar et al. 2012; McCaul and
Bondar 2021). Next we apply the completeness theorem twice with θ, p and q, λ:
〈qj+1, pj+1| e−i∆tK |qj, pj〉 = 〈qj+1, pj+1| e−i∆tpθm ei∆tV
′(q)λ) |qj, pj〉 =
= 〈qj+1, pj+1| e−i∆tpθm IIei∆tV
′(q)λ |qj, pj〉 =
=
˘dθ dp dq dλ 〈qj+1, pj+1| e−i∆t
pθm |θ, p〉 〈θ, p|q, λ〉 〈q, λ| ei∆tV ′(q)λ |qj, pj〉
(71)
Note that by Theorem I, we have that
exp
(−i∆t pθ
m
)|θ, p〉 = exp
(−i∆tpθ
m
)|θ, p〉
And also that
〈q, λ| exp(i∆tV ′(q)λ
)= 〈q, λ| exp(i∆tV ′(q)λ)
42
Therefore, plugging into the expression (71) we get:
˘dθ dp dq dλ 〈qj+1, pj+1|θ, p〉 〈θ, p|q, λ〉 〈q, λ|qj, pj〉 e−i∆t
pθm ei∆tV
′(q)λ
For the expressions 〈qj+1, pj+1|θ, p〉, 〈θ, p|q, λ〉, and 〈q, λ|qj, pj〉, we substitute in our
ingredients we derived earlier (63), (64), and (65):
˘dθ dp dq dλ
1√2πδ(p−p′)eiθqj+1
1
2πeipj+1λ−iθqj 1√
2πδ(q−q′)e−ipjλe−i∆t
pθm ei∆tV
′(q)λ
Remember that according to (19), the integrals of Dirac delta functionals will reduce
to the following, neater expression:
=
¨dθ dλ
1√2πeiθqj+1
1
2πeipj+1λ−iθqj 1√
2πe−ipjλe−i∆t
pθm ei∆tV
′(q)λ
And simplifying:
=1
(2π)2
¨dθ dλ eiθqj+1−iθqj−i∆t pθm eipj+1λ−ipjλ+i∆tV ′(q)λ
Factoring out the observables θ and λ from each exponent we get:
1
(2π)2
¨dθ dλ exp
(iθ∆t
[qj+1 − qj∆t
− p
m
])exp
(iλ∆t
[pj+1 − pj∆t
+ V ′(q)])
Inserting this important expression into equation (70) we will get that:
〈qf , pf | e−itK |q0, p0〉 =
=
ˆ···ˆdqN−1dpN−1···dq1dp1
[1
(2π)2
¨dθN−1 dλN−1 exp
(iθN−1∆t
[qj+1 − qj∆t
−pjm
])×
× exp(iλN−1∆t
[pj+1 − pj∆t
+V ′(qj)])]···[
1
(2π)2
¨dθ1 dλ1 exp
(iθ1∆t
[qj+1 − qj∆t
−pjm
])×
× exp(iλ1∆t
[pj+1 − pj∆t
+ V ′(qj)])]
43
We can utilize some of the simplifying notation we used for Feynman’s Path Integral:
〈qf , pf | e−itK |q0, p0〉 =1
(2π)2N
˘ N−1∏j=1
dqj dpj dθj dλj×
× exp(i∆t
N−1∑j=0
[θj
(qj+1 − qj∆t
− pjm
)])exp
(i∆t
N−1∑j=0
[λj
(pj+1 − pj∆t
+V ′(qj))])(72)
Just as before, let ∆t→ 0 (equivalent to N →∞) to get:
lim∆t→0
qj+1 − qj∆t
=dq
dt
lim∆t→0
pj+1 − pj∆t
=dp
dt
limN→∞
N−1∑j=0
=
ˆ t
0
One more piece of notation. Just as before, let us define that for a variable f , that we
will have:
Df = limN→∞
N∏n
dfn√2π
(73)
And so the classical analog of the path integral can be written:
〈qf , pf | e−itK |q0, p0〉 =
˘DqDpDθDλ exp
(i
ˆ t
0
dt[λ(t)
(dqdt− p
m
)+θ(t)
(dpdt
+V ′(q))])
(74)
What is the relevance of this? This encodes the equations of CM (McCaul and Bondar
2021). The path integral is a kind of redundancy that can be shown to enforce the laws
of CM. We can show that it leads to a Dirac delta functional that enforces Newton’s
Laws.
Recall that the Dirac delta functional δ(A − A′) is nonzero when A = A′ but
44
zero if A 6= A′. Let us use equation (72) above, as it is written in a slightly easier
to mathematically manipulate form (following the footsteps of McCaul and Bondar
2021). Let us look at the portion of the integral with respect to θj:
ˆdθj exp
(i∆t[θj
(qj+1 − qj∆t
−pjm
)])=
ˆdθj exp
(i∆t
m
[θj
(mqj+1 − qj
∆t−pj
)])
Next, we will utilize equation (22) for our convenience. We will take the 0th
derivative of the Dirac delta functional in the formula (which is the same as taking
no derivative at all):
d0
dA0δ(A− A′) = δ(A− A′) =
ˆdω
2π(−iω)0e−iω(A−A′) =
ˆdω
2πe−iω(A−A′)
If you set θ = ω/2π, you can plug into the integral with respect to θj to get:
ˆdθj exp
(i∆t
m
[θj
(mqj+1 − qj
∆t−pj
)])=
ˆdω
2πexp
(−i∆t
m
[2πω
(pj−m
qj+1 − qj∆t
)])=
=
ˆdω
2πexp
(− iω
[2π
∆t
m
(pj −m
qj+1 − qj∆t
)])By comparing it to the Dirac delta expression above, we see right away that (A−A′) =[2π∆t
m(pj −m qj+1−qj
∆t)], and so:
ˆdθj exp
(i∆t[θj
(qj+1 − qj∆t
− pjm
)])= δ( ∆t
2πm(pj −m
qj+1 − qj∆t
))
=
= 2πm
∆tδ(pj −m
qj+1 − qj∆t
)(75)
utilizing equation (24) for the last step. This reinforces the dynamical equation pj −
mqj+1−qj
∆t= 0, which is simply the discretized version of the momentum relation,
45
p = mv. If this delta functional is plugged back into equation (74), then you can
reduce the total number of variables integrated over from four to just q and λ (McCaul
and Bondar 2021). The two variable version of the Classical Path Integral ensures
CM holds (Mauro 2003; McCaul and Bondar 2021).
The above integration over θ can also be conducted over the variable λ instead to
get
ˆdλj exp
(i∆t[λj
(pj+1 − pj∆t
+V ′(qj))])
=
ˆdω
2πexp
(−iω
[∆t
2π
(−pj+1 − pj
∆t−V ′(qj)
)])=
=2π
∆tδ(pj+1 − pj
∆t+ V ′(qj)
)(76)
This reinforces the dynamical equation pj+1−pj∆t
+ V ′(qj) = 0, which is simply the
discrete version of Newton’s 2nd Law of Motion, dpdt
= −V ′(q) = F .
All the above equations are elaborate formulations of the laws of CM. These are
different ways of enforcing the laws of CM hold true. We will go over some practical
usages of the Classical Path Integral in the pages that follow.
5 Classical and Quantum Comparison
5.1 The Phase in Koopman-von Neumann and Quantum Mechan-
ics
In QM, the interaction of the quantum phase S with the wavefunction amplitude R
gives rise to quantum mechanical behavior (Mauro 2003). Recall that one way of ex-
pressing the wavefunction ψ(t) is (3). If we plug this expression into the Schrodinger
equation (4) we get two equations:
∂S
∂t+
(∇S)2
2m− ~2
2mR∇2R + V (q) = 0 (77)
46
which is the classical Hamilton-Jacobi theory with quantum potential16, and
∂R2
∂t+∇ · (R2∇S
m) = 0 (78)
which is the continuity equation for density R2. 17 From the above expression, you
can see that the S and R are coupled together and cannot be separated from each
other in QM.
This is not the case in CM. From the KvNM equation (35), you can see if you plug
in ψ(q, p, t) = R(q, p, t)eiG(q,p,t) you will get complete decoupling of the amplitude
R from the phaseG. All one has to do is plug the phase space wavefunction ψ(q, p, t)
into (35), like so:
i∂ψ
∂t= Lψ = i
∂
∂tR(q, p, t)eiG(q,p,t) =
[− i p
m
∂
∂q+ i
dV
dq
∂
∂p
]R(q, p, t)eiG(q,p,t)
This process begets:
i∂R
∂t= LR (79)
i∂G
∂t= LG (80)
The amplitude and the phase are completely seperate. Quantum mechanics has there-
fore been described as the theory that tells how the phase interacts with the amplitude
(Mauro 2003).
If the phase and the amplitude evolve separately in KvNM, one may ask what is16As an interesting aside, one can trivially show that the quantum potential disappears if one as-
sumes the Classical Commutator. For instance, equation (77) can be rewritten since we know that(i~)2 = −~2:
∂S
∂t+
(∇S)2
2m+
[q, p]2
2mR∇2R+ V (q) = 0
If you assume the Canonical Commutator, you retrieve the Schrodinger equation as is. If you assumethe Classical Commutator, you retrieve the completely classical Hamilton-Jacobi equation. This resultis trivial, but still interesting to consider from the perspective of unity of ideas.
17Keep in mind that if ρ = ψ∗ψ, then ρ = ψ∗ψ = Re−iS/~ReiS/~ = R2.
47
the practical use of the KvNM formulism? Why the use of complex wavefunctions
if the exponated phase is completely independent of the amplitude? While it is true
that in the momentum position representation |q, p〉 of KvNM the formulism can be
redundant, one can always use the convenient properties of Hilbert Spaces to change
to other representations, which gives us new insights into CM (Bondar et al. 2012;
Bondar et al. 2013; McCaul and Bondar 2021, etc.) We will see examples of this
change in basis utilized in section 6. In order to have the most general form of CM, it
is important to write the classical wavefunction in this complex form (Mauro 2003).
5.2 Uncertainty Principle in Koopman-von Neumann and Quan-
tum Mechanics
One of the hallmarks of QM is the existence of an uncertainty principle between the
position and momentum. The uncertainty principle tells you that the more precisely
you measure where a wavefunction is localized, the less precisely you know the mo-
mentum distribution. This contrasts with CM, where you can measure the values of
both the momentum and position to an arbitrary precision at the same time. This
trade off is another re-articulation of the strange property of “fuzzy” trajectories in
QM and well-defined trajectories in CM.
One can naturally derive the Heisenberg Uncertainty Principle from the Canoni-
cal Commutator. Through the same series of steps, one can also demonstrate the lack
of an uncertainty principle between position and momentum for the KvN classical
case. The procedure is quite straightforward.
Firstly, from statistics we know that the standard deviation of some measured
quantity A is given by:
σA =√〈A2〉 − 〈A〉2 (81)
And this will be a crucial ingredient in deriving the Heisenberg Uncertainty Principle.
48
From equation (16), we know that
〈A〉 = 〈ψ(t)| A |ψ(t)〉
〈A2〉 = 〈ψ(t)| A2 |ψ(t)〉
Treat 〈A〉 as a constant. Then we can see the following important relationship:
(A− 〈A〉)2 = A2 − 2〈A〉A+ 〈A〉2,
〈ψ(t)| [A2−2A〈A〉+〈A〉2] |ψ(t)〉 = 〈ψ(t)| A2 |ψ(t)〉−〈ψ(t)| 2〈A〉A |ψ(t)〉+〈ψ(t)| 〈A〉2 |ψ(t)〉 =
= 〈ψ(t)| A2 |ψ(t)〉 − 2〈A〉 〈ψ(t)| A |ψ(t)〉+ 〈A〉2 〈ψ(t)|ψ(t)〉 =
= 〈ψ(t)| A2 |ψ(t)〉 − 2〈A〉2 + 〈A〉2 = 〈A2〉 − 〈A〉2.
utilizing (16) and 〈ψ(t)|ψ(t)〉 = 1.
We have all the ingredients we need to derive the Heisenberg Uncertainty Prin-
ciple. First, let us compute the variance σ2 for both the position and momentum.
Definitionally following (81) and our above result, they would be
σ2q = 〈q2〉 − 〈q〉2 = 〈ψ(t)| (q − 〈q〉)2 |ψ(t)〉
σ2p = 〈p2〉 − 〈p〉2 = 〈ψ(t)| (p− 〈p〉)2 |ψ(t)〉
We can define a new vector to simplify our calculation. We can define |Q〉 = (q −
〈q〉) |ψ(t)〉 and |P 〉 = (p− 〈p〉) |ψ(t)〉 in order to write the variance of position and
momentum simply as
σ2q = 〈Q|Q〉
σ2p = 〈P |P 〉
49
Then we apply the Cauchy-Schwarz Inequality (Theorem III of Appendix A) which
tells us that:
σ2pσ
2q = 〈P |P 〉 〈Q|Q〉 ≥ | 〈P |Q〉 |2 (82)
We know that for any complex number z = a+ bi, that
|z|2 = a2 + b2 ≥ b2
where we know from complex analysis that b = z−z∗2i
. Ergo,
|z|2 = | 〈P |Q〉 |2 ≥(z − z∗
2i
)2
=(〈P |Q〉 − 〈Q|P 〉
2i
)2
,
where
〈P |Q〉 = 〈ψ(t)| (p−〈p〉)(q−〈q〉) |ψ(t)〉 = 〈ψ(t)| [pq−q〈p〉−〈q〉p+〈q〉〈p〉] |ψ(t)〉 =
= 〈ψ(t)| pq |ψ(t)〉 − 〈ψ(t)| q〈p〉 |ψ(t)〉 − 〈ψ(t)| 〈q〉p |ψ(t)〉+ 〈ψ(t)| 〈q〉〈p〉] |ψ(t)〉 ,
and
〈Q|P 〉 = 〈ψ(t)| qp |ψ(t)〉−〈ψ(t)| q〈p〉 |ψ(t)〉−〈ψ(t)| 〈q〉p |ψ(t)〉+〈ψ(t)| 〈q〉〈p〉] |ψ(t)〉 .
With the calculations above, we plug into (82) and derive:
| 〈P |Q〉 |2 ≥(〈P |Q〉 − 〈Q|P 〉
2i
)2
=(〈ψ(t)| pq |ψ(t)〉 − 〈ψ(t)| qp |ψ(t)〉
2i
)2
,
| 〈P |Q〉 |2 ≥(〈ψ(t)| [p, q] |ψ(t)〉
2i
)2
. (83)
50
From this, it is simple to see that:
σpσq ≥| 〈ψ(t)| [p, q] |ψ(t)〉 |
2(84)
If we plug in the Canonical Commutator for [p, q] we derive the Heisenberg Uncer-
tainty Principle,
σpσq ≥ ~/2 (85)
If we plug in the Classical Commutator for [p, q] = 0 there is no uncertainty rela-
tion between the position and momentum, hence there is no theoretical restriction to
knowledge of precision of both at the same time. Depending on your choice of the
Commutator, you derive either Classical or Quantum Physics (Bondar et al. 2012).
This does not, however, mean there are no uncertainty relations for KvNM. Be-
cause the Koopman Algebra (43) tells us that [q, θ] = i and [p, λ] = i, we can develop
uncertainty relationships for q-θ and p-λ by plugging the commutators into expres-
sions analogous to (84):
σpσλ ≥| 〈ψ(t)| [p, λ] |ψ(t)〉 |
2=
1
2
σqσθ ≥| 〈ψ(t)| [q, θ] |ψ(t)〉 |
2=
1
2
These would be the ‘classical’ uncertainty relationships. Since neither θ nor λ are
directly observable (Ramos-Prieto et al. 2018), these uncertainty relations will not
impact our observations of classical systems. For KvNM, many “quantum” phenom-
ena are tucked into the not directly observable operators θ and λ.
5.3 Measurement in Koopman-von Neumann and Quantum Me-
chanics
QM is known to have the strange property that it makes probabilistic predictions
instead of deterministic ones. One strange feature related to this fact that tucked
51
into the previously mentioned postulates of QM you have a phenomenon termed the
collapse of the wavefunction. Any wavefunction can be expanded across a basis using
the completeness theorem (Theorem II of Appendix A):
|ψ〉 = I |ψ〉 ,
where I is the identity operator (analogous to the number ‘1’.) Inserting the discrete18
completeness relation we get:
|ψ〉 =∑j
|j〉 〈j|ψ〉 = |A〉 〈A|ψ〉+ |B〉 〈B|ψ〉+ |C〉 〈C|ψ〉+ · · ·, (86)
where the collection of all |j〉 vectors (j = A,B,C, ...) are your basis vectors ac-
cording to the principles of Linear Algebra and 〈j|ψ〉 are scalar values or functions
(recalling that 〈j|ψ〉 = ψ(j)). This is simply expanding the state vector |ψ〉 along a
particular set of basis vectors |j〉. A, B, C, etc., are different eigenvalues associated
with each ‘eigenbasis’, for example, A |A〉 = A |A〉.
If you measure B then the initial wavefunction |ψ〉 instantaneously collapses the
entire quantum system to state |B〉. How the state of the system |ψ〉 instantaneously
changes to |B〉 when B is measured is still a modern day quantum mystery. As
described by Postulate 3 in section 2, the probability for observing B is P (B) =
| 〈B|ψ〉 |2 for discrete systems and the probability density is ρ(B) = | 〈B|ψ〉 |2 in
continuous systems.
The strange phenomenon of collapse is preserved in KvNM. You have a KvN
classical wavefunction that again can be expanded over a particular basis of your
choice
|ψ〉 = |A〉 〈A|ψ〉+ |B〉 〈B|ψ〉+ |C〉 〈C|ψ〉+ · · ·,18Note that in (86) I used the discrete case instead of the continuous completeness identity. This is
simply due to the ease of highlighting certain concepts. The continuous completeness identity woulddo the same thing as the discrete case, but might obscure for some readers results we will discuss.Remember that the integral is just a continuous summation of infinitely small things.
52
and that when a particular eigenvalue is observed, the system collapses to its asso-
ciated eigenbasis. The same Born Rule applies. You have the same postulate for
collapse from the set of ‘universal axioms’ .
The common behavior of the wavefunction collapsing during measurement in
QM and KvNM is mystifying, and perhaps one day further light can be shed on it.
Maybe it is a common feature of probabilistic models built in Hilbert Space, or per-
haps there will be some other explanation developed. As far as we know, the collapse
is instant, irreversible, and no causal process exists to explain it.
There is an important difference though in the collapse of the wavefunction dur-
ing measurement for the KvNM versus QM cases. The wavefunction will not col-
lapse for non-selective measurements in the KvN case but will in the QM case (Gozzi
and Mauro 2002; Mauro 2003). Non-selective measurements are measurements con-
ducted on the system where the results are not ‘read out loud’, meaning the system
would be hypothetically collapsed but we do not ‘record’, remember, or consider
which eigenbasis it collapses to. We remain completely ignorant of whether the sys-
tem state |ψ〉 hypothetically collapses into |A〉, |B〉, |C〉, or some other state. This
difference in collapse for KvN and QM wavefunctions will be demonstrated (follow-
ing the work of Gozzi and Mauro 2002; Mauro 2003).
First, let us consider the QM case. Consider a Hamiltonian H which obeys the
following eigenvalue relationships:
H |+〉 = ~ω |+〉 , (87)
H |−〉 = −~ω |−〉 , (88)
where |+〉 and |−〉 are the eigenvectors and ~ω and −~ω are the energy eigenvalues
of the operator H , respectively. 19 Let us consider an observable with operator Ω,
which has the eigenvectors |a〉 and |b〉 defined by:19This should bring to mind, for instance, that the energy of a photon is given by E = ~ω, where ω
is the angular frequency.
53
|a〉 =1√2
[|+〉+ |−〉
], (89)
|b〉 =1√2
[|+〉 − |−〉
]. (90)
For the sake of our analysis, we can choose the initial state of the system to be the
following superposition of states (Mauro 2003):
|ψ(t = 0)〉 =1
2|+〉+
√3
4|−〉 (91)
We will utilize the Schrodinger equation in the form of (52) (see also Appendix C)
order to evolve the above wavefunction, |ψ(0)〉. We want to evolve the system to
t = 2τ and see what the probability of measuring a, that is P (a), will be (Mauro
2003).
|ψ(t = 2τ)〉 = e−i~ H2τ |ψ(0)〉 = exp
[− i~H2τ
][1
2|+〉+
√3
4|−〉
]. (92)
We will utilize the very useful Theorem I from Appendix A alongside (87) and (88).
This will give us:
|ψ(t = 2τ)〉 =1
2exp
[− i~H2τ
]|+〉+
√3
4exp
[− i~H2τ
]|−〉 =
=1
2exp
[− i~
2τ(~ω)
]|+〉+
√3
4exp
[− i~
2τ(−~ω)
]|−〉 =
=1
2exp[−2iτω] |+〉+
√3
4exp[2iτω] |−〉 . (93)
To find the probability of observing a at t = 2τ we will use the Born Rule:
P (a) = 〈ψ(2τ)|a〉 〈a|ψ(2τ)〉 = | 〈a|ψ(2τ)〉 |2,
where we can evaluate:54
〈a|ψ(2τ)〉 =1
2exp[−2iτω] 〈a|+〉+
√3
4exp[2iτω] 〈a|−〉 =
=1
2exp[−2iτω]
1√2
[〈+|+ 〈−|
]|+〉+
√3
4exp[2iτω]
1√2
[〈+|+ 〈−|
]|−〉 .
Using the orthonormality condition, we know that 〈+|−〉 = 0 while 〈+|+〉 = 〈−|−〉 =
1, so that the above expression reduces to
〈a|ψ(2τ)〉 =1
2√
2exp[−2iτω] +
√3
8exp[2iτω].
Therefore, taking (〈a|ψ(2τ)〉)† = 〈ψ(2τ)|a〉 and plugging into the Born Rule:
P (a) = 〈ψ(2τ)|a〉 〈a|ψ(2τ)〉 =( 1
2√
2exp[2iτω] +
√3
8exp[−2iτω]
)×
×( 1
2√
2exp[−2iτω] +
√3
8exp[2iτω]
)=
=1
2√
2exp[2iτω]
1
2√
2exp[−2iτω] +
1
2√
2exp[2iτω]
√3
8exp[2iτω]+
+
√3
8exp[−2iτω]
1
2√
2exp[−2iτω] +
√3
8exp[−2iτω]
√3
8exp[2iτω] =
=1
8+
√2
4
√3
8exp[4iωτ ] +
√2
4
√3
8exp[−4iωτ ] +
3
8. (94)
There are one important trigonometric identity we can make use of. It is:
cos(θ) = (1/2)(eiθ + e−iθ) (95)
Taking the identity (95) with (94), we can calculate that
55
P (a) =1
2
(1 +
√3
4cos(4ωτ)
). (96)
The above usage of the wavefunction is for “pure states” (term defined in section 2.6).
Now, instead of letting the quantum system evolve continuous from t = 0 to
t = 2τ , let us introduce a non-selective measurement at t = τ . How will this affect
the probability of measuring a at t = 2τ? We will start with the same |ψ(0)〉. We
will, however, use the density operator formulism to describe the system (see section
2.6) instead of strictly using only the ‘pure’ wavefunction. This is because we have
here what is called a “mixed state”. When we take the measurement at t = τ , the
original system given by (91) collapses, but we do not know whether it collapses into
|a〉 or |b〉, as we do not ‘read out’ or record what the collapse leads to. We can use
(26) to write down:
ρ(τ) = pa(τ) |a〉 〈a|+ pb(τ) |b〉 〈b| , (97)
where pa(τ) and pb(τ) are the probabilistic weights to the ‘statistical mixture’ of
possible states |a〉 and |b〉 of the system. The factor pa(τ) refers to the probability
that we observe a at t = τ (and hence the system collapsed to |a〉) and the term pb(τ)
refers to the probability that we observe b at t = τ (and hence the system collapsed to
|b〉). Since we are ignorant of which way the wavefunction collapses, the probabilities
became weights of likelihood in (97).
We calculate what the factors pa(τ) and pb(τ) are in a similar fashion to calculat-
ing (96):
pa(τ) = 〈ψ(τ)|a〉 〈a|ψ(τ)〉 =(√2
4exp[−iωτ ] +
√3
8exp[iωτ ]
)×
×(√2
4exp[iωτ ] +
√3
8exp[−iωτ ]
)=
=1
2
(1 +
√3
4
[1
2exp[−2iωτ ] +
1
2exp[2iωτ ]
])=
1
2
(1 +
√3
4cos(2ωτ)
)(98)
56
pb(τ) = 〈ψ(τ)|b〉 〈b|ψ(τ)〉 =(√2
4exp[−iωτ ]−
√3
8exp[iωτ ]
)×
×(√2
4exp[iωτ ]−
√3
8exp[−iωτ ]
)=
=1
2
(1−
√3
4
[1
2exp[−2iωτ ] +
1
2exp[2iωτ ]
])=
1
2
(1−
√3
4cos(2ωτ)
)(99)
We will use (52) again as before to evolve the system then from (97) at t = τ to
t = 2τ , in the following fashion:
|ψa(2τ)〉 = e−i~ H(2τ−τ) |a〉 = exp
[− i~Hτ
]1√2
[|+〉+ |−〉
], (100)
|ψb(2τ)〉 = e−i~ H(2τ−τ) |b〉 = exp
[− i~Hτ
]1√2
[|+〉 − |−〉
], (101)
utilizing the expressions (89) and (90) above.
As before, utilizing the eigenvalue expressions for H (87) and (88), and Theorem
I we get:
|ψa(2τ)〉 =1√2
exp
[− i~τ(~ω)
]|+〉+
1√2
exp
[− i~τ(−~ω)
]|−〉 , (102)
|ψb(2τ)〉 =1√2
exp
[− i~τ(~ω)
]|+〉 − 1√
2exp
[− i~τ(−~ω)
]|−〉 . (103)
The density operator therefore at t = 2τ can be represented as the following using
the above expressions:
ρ(2τ) = pa(τ) |ψa(2τ)〉 〈ψa(2τ)|+ pb(τ) |ψb(2τ)〉 〈ψb(2τ)| (104)
And we can use this to evaluate the probability of measuring a by using (27):
57
P (a) = Tr[ρ(2τ) |a〉 〈a|] = 〈a| ρ(2τ) |a〉 〈a|a〉+ 〈b| ρ(2τ) |a〉 〈a|b〉 .
Remember that 〈a|b〉 = 0 and 〈a|a〉 = 1, so:
P (a) = 〈a| ρ(2τ) |a〉 .
Plugging in (104) we get:
P (a) = pa(τ) 〈a|ψa(2τ)〉 〈ψa(2τ)|a〉+ pb(τ) 〈a|ψb(2τ)〉 〈ψb(2τ)|a〉 . (105)
And now we can substitute the expressions (102), (103), (98), and (99) to get after
some algebra our final expression:
P (a) =1
2
(1 +
√3
4cos2(2ωτ)
). (106)
Obviously, comparing (96) with (106) shows that they are two different equations.
This shows us that in the quantum case making a non-selective measurement will
change the outcome of a later measurement. A non-selective measurement will mod-
ify the probability of outcome at a later point in time t.
So far, what we looked at is an abstract example to demonstrate non-selective
measurements in QM. Let us look at a more concrete example in QM, and then figure
out what the implications are for KvNM (following Mauro 2003). Starting off, let us
measure the probability of a quantum particle’s position from t = 0 to t = τ .
This will be given by (26) except we are no longer dealing with a discrete case.
Since position is continuous, we replace the summation sign with an integral (as an
integral is simply continuous summation of infinitely small things). We start with a
pure state with wavefunction |ψ(0)〉 at t = 0. The probability density ρ (not to be
confused with the density matrix) of observing q at time t = τ will be given by
58
ρ(q, τ) = Tr[ρ(τ) |q〉 〈q|] = |ψ(q, τ)|2, (107)
where ρ(τ) = |ψ(τ)〉 〈ψ(τ)| (pure state density matrix).
Let us say that we make a non-selective measurement immediately after t = 0,
which we will represent as time t = 0+. The pure state density matrix gets replaced
with a mixed density matrix. We utilize a continuous version of (26) to write the
density matrix
ρ(0+) =
ˆdq0|ψ(q0, 0)|2 |q0〉 〈q0| .
To represent a state of the system immediately before t = 0, we can write time t = 0−
and say that:
ρ(0−) = |ψ(0−)〉 〈ψ(0−)|
The system goes from ρ(0−) to ρ(0+) as a non-selective measurement is made at
t = 0.
In QM, the measurement right before and the measurement right after will ob-
serve the same thing (Mauro 2003). You can convince yourself of this by computing
the following:
ρ(q, 0−) = Tr[ρ(0−) |q〉 〈q|] = ψ∗(q, 0)ψ(q, 0)
ρ(q, 0+) =Tr[ρ(0+) |q〉 〈q|]
Tr[ρ(0+)]= ψ∗(q, 0)ψ(q, 0)
Remember that the denominator, Tr[ρ(0+)], just represents the normalization condi-
tion (described in section 2.6). We can evolve ψ∗(q, 0)ψ(q, 0) to see that in the long
term the non-selective measurement does affect the quantum system, just like in the
above worked out example (Mauro 2003). The probability density will be the same
instantaneously right before and right after measurement, but will alter in the long
run in QM.
59
Under KvNM, however, we can observe some differences arise. If we start with
a pure state again
ρ(0−) = |ψ(0−)〉 〈ψ(0−)| ,
and conduct no measurements, we will have the KvN wavefunction evolve to
ρ(q, p, τ) = Tr[ρ(τ) |q, p〉 〈q, p|] = |ψ(q, p, τ)|2, (108)
similarly to the quantum case (107). Also like the quantum case, in KvNM measuring
the probability density right before and after will not alter the distribution:
Tr[ρ(0−) |q, p〉 〈q, p|]Tr[ρ(0−)]
=Tr[ρ(0+) |q, p〉 〈q, p|]
Tr[ρ(0+)]
However, a non-selective measurement can be shown to leave the KvN wavefunction
unaltered, whereas a quantum wavefunction will be changed due to the non-selective
measurement .
ρ(0+) =
ˆdq0dp0|ψ(q0, p0, 0)|2 |q0, p0〉 〈q0, p0| .
The above probability density, taken right after the measurement, can be shown to
naturally evolve into (108), as if no measurement has taken place (Mauro 2003).
There is one final similarity worth mentioning between QM and KvNM. Under
QM, when a measurement is taken for position q, it affects the outcome of the distri-
bution of p. Perhaps, this is not surprising, considering the Heisenberg Uncertainty
Principle. Under KvNM, a measurement taken for (q, p) will alter the statistical dis-
tribution of the variable λ (Mauro 2003). This should also not be surprising because,
as we discussed in section 5.2, there exists a classical uncertainty relationship be-
tween p and λ. The effects of QM are still present in KvNM therefore, if we replace
q with (q, p) and p with λ.
60
5.4 The Double Slit Experiment
The Double Slit Experiment is at the center of Quantum Theory. QM has photons,
electrons, neutrons, and other quanta going through two tiny slits and forming an
interference pattern on the other side. Even if we send one particle at a time through
the double slit, the particle build up on the other side will reflect an interference
pattern. Even if the particle does not seem to interfere with any other particle, it still
is part of an interference pattern. This shows that the interference patterns are due to
something more fundamental in nature. 20
The wavefunction mathematically describes this strange behavior of elementary
particles, even though the exact interpretation of the wavefunction is not known.
Feynman believed that the strange behavior must lie in the complex nature of the
wavefunction (Mauro 2003). KvNM also contains complex wavefunctions in Hilbert
Space, so we might logically ask the obvious question: if KvNM has complex wave-
functions, will it successfully preserve our knowledge of CM, without interference
patterns arising? Here we will test the outcome of complex wavefunctions in classical
phase space. This analysis is based on the work of (Mauro 2003).
The double slit experiment is set up in Figure 1 above. The y axis is aligned
parallel to the direction of travel. The x axis will be the coordinate parallel to the
back wall where the interference is supposed to occur for the quantum case. In the
middle, at y = yM you have the wall with the two slits. At y = yF you have the wall
where interference occurs in the quantum case.
Under KvNM, we should expect no interference build up at the backwall at y =
yR. Under QM (and in reality), we want there to be an interference build up at the
backwall. We will start with analyzing the quantum case, and how it works. We want
a gaussian wave emerge from y = 0 and travel right until it hits the double slit wall.
Some of the wavefunction will travel through the two slits and eventually interfere at
y = yR if the system behaves quantum.20Interestingly, if you cover up on of the two slits, you will recover particle travelling in straight
trajectories without any interference pattern, just as in CM. This makes the relationship between CMand QM even more bizarre.
61
A gaussian function has the following form:
f(u) =1√
2πσ2e−(u−µ)2/2σ2 (109)
where µ is the average value µ = 〈u〉 and σ is the standard deviation, given by
σ =√〈u2〉 − 〈u〉2.
Recalling the form of a quantum wavefunction (3), we can make it a gaussian
wavefunction by setting R equal to (109) with a generic renormalization constant
N .21 Let µ = 0 in order to center the wavefunction of the double slit experiment.
The equation will have the form:
ψ(x, t = 0) = N exp
[− x2
2σ2x
]exp
[i
~S
]Note that for this wavefunction, S = 0, however. This is obvious when you consider
the definition of the Action (51) for a free particle. Since we decided that the time at
the leftmost wall is t = 0, the Action at the leftmost wall would be:
S =
ˆ 0
0
dtT = 0
Therefore, a better way to write the previous equation at t = 0 is:
ψ(x, t = 0) = N exp
[− x2
2σ2x
](110)
Following the Path Integral Formulism (54), in order to know the know the wavefunc-
tion ψ(x, t) at some later time t we will need to utilize the Kernel (57). The Kernel
〈x| e− i~ tH |x0〉 will evolve the system to the time of interest.
〈x|ψ(t)〉 =
ˆdx0 〈x| e−
i~ tH |x0〉ψ(x0, 0)
We will utilize a very special trick in order to evaluate the above expression. Since21In order to keep things nice and streamlined, the constant in front of (109) will simply be absorbed
into the Renormalization Constant.
62
we are dealing with a free particle, there are no potentials, or V (x, y) = 0 (hence no
forces or accelerations). Because of this, we do not have to use the full, intimidating
form of the Feynman Kernel (57). In this semiclassical treatment there is no acceler-
ation of the quantum particle in the y direction, and so the Kernel can be written in a
much simpler form.
Instead of taking on the form (57), it will instead have the form (56). This is
because we can treat the final position of the wavefunction as qj+1 and the initial
position as qj if there is no acceleration. The term qj+1−qj simply becomes one large
step instead of a miniscule step as before. This is analogous to how in kinematics,
when there is no acceleration, the formula for instantaneous velocity is v = ∆x/∆t,
even though in every other case the formula gives you average velocity (as it is a slope
formula). You can also think of this simplification as taking (57) and setting N = 1
(one large step due to no acceleration). Please note that this is not an approximation,
but this will give us an exact expression.
Our Kernel K (not to be confused with the Koopman Generator) is therefore:
K =
√m
2πi~(t− t0)exp
[im(x(t)− x0)2
2~(t− t0)
](111)
And then we can write the following Path Integral:
〈x|ψ(t)〉 =
ˆ ∞−∞
dx0
√m
2πi~(t− t0)exp
[im(x(t)− x0)2
2~(t− t0)
]N exp
[− x2
0
2σ2x
]
We want to evaluate the path integral up until the point it reached the middle wall with
the double slits, at time tM . Remembering that we have set the initial time t0 equal
to 0 and treating t and x(t) as constants, we can evaluate the integral. First, we will
pull out all constants and absorb them into N , the renormalization constant (recall
the renormalization condition described in section 2.3). Then we do the following
algebraic manipulations, in preparation to evaluating the integral itself:
63
exp
[im(x(t)− x0)2
2~t
]exp
[− x2
0
2σ2x
]= exp
[im(x(t)2 − 2x(t)x0 + x0
2)
2~t− x2
0
2σ2x
]=
= exp
[( im2~t− 1
2σ2x
)x2
0 −imx(t)
~tx0 +
imx(t)2
2~t
]We can then use Theorem VIII of Appendix A to evaluate the integral. Comparing
the two, we can recognize the following coefficients of the Gaussian integral:
a = −( im
2~t− 1
2σ2x
)b = −imx(t)
~t
c =imx(t)2
2~t
And so by the Theorem, the integral turns out to be:
N
ˆ ∞−∞
dx0 exp
[( im2~t− 1
2σ2x
)x2
0 −imx(t)
~tx0 +
imx(t)2
2~t
]=
= N
√π
−(im2~t −
12σ2x
) exp
( imx(t)~t )2
−4(im2~t −
12σ2x
) exp
[imx(t)2
2~t
]=
= N
√2π~tσ2
x
−imσ2x + ~t
exp
[m2x(t)2σ2
x
2~t(−imσ2x + ~t)
]exp
[imx(t)2
2~t
]
∴ 〈x|ψ(t)〉 = N
√2π~tσ2
x
−imσ2x + ~t
exp
[−1
2
mx(t)2
mσ2x + i~t
]The above wavefunction 〈x|ψ(t)〉 = ψ(x, t) is true for anytime 0 < t ≤ tM , where
tM is the time to reach the middle barrier with the two slits. This is because the above
wavefunction is that of a free particle, and once it encounters a barrier (or potential),
the particle is obviously no longer free.
Next, since the expression ψ(x, tM) gives you the quantum wavefunction at the
double slits, we have to take into consideration the portion of the wavefunction that
64
goes through the double slits. To describe the wavefunction going through the double
slit we will utilize step functions. The step functions will closely model the fact that
the wavefunction that hits the middle wall and stops will not be important in creating
any effects on the other side of the wall. Only the wavefunction that goes through the
two slits will be important in creating the diffraction pattern on the other side (Shankar
1988; Mauro 2003). The step function, also called the θ-Heaveside function, is given
by:
H(x) =
1 x > 0
0 x ≤ 0(112)
Each slit has a width of 2δ. The first slit has width ∆1 = (xA − δ, xA + δ) and the
second slit has spans ∆2 = (−xA − δ,−xA + δ) (following the notation of Mauro
2003).
Then, the wavefunction right after the two slits can be written as
ψ(x, tM+dt) = Nψ(x, tM)[C1(x)+C2(x)] = N exp
[−1
2
mx2
mσ2x + i~t
][C1(x)+C2(x)]
(113)
where the following functions can be defined:
C1(x) = H(x− xA + δ)−H(x− xA − δ) (114)
C2(x) = H(x+ xA + δ)−H(x+ xA − δ) (115)
This is done in order to filter out the effects of the wavefunction that actually goes
through the slits. We do not care for the portion of the wavefunction that does not go
through the double slit. They will produce no effects. The above equations make cer-
tain that only the wavefunction that goes through slits ∆1 and ∆2 will be studied. The
N in (113) is a renormalization constant, so that´dxψ∗ψ = 1 (the renormalization
condition described in section 2.3).
Using our Kernel for the free particle again, we will evaluate the wavefunction
65
using the path integral for the region after the two slits. We again utilize (111). Notice
that the initial wavefunction ψ(x0, 0) in that equation starts at time zero, so we will
need to adjust the time at the double slit wall so that it is set to zero. We can do
this by substituting the variable t in the Kernel (111) with t− tM . This can be done
because time has the unique property that any point can be arbitrarily set to t = 0,
and by doing this swap we effectively set the time at the middle wall equal to zero
(starts at tM and evolves from there). So the integral we need to evaluate will have
the following form:
〈x|ψ(t)〉 = N
ˆdxM exp
[im(x(t)− xM)2
2~(t− tM)
]ψ(xM , tM) =
= N
ˆdxM exp
[im(x(t)− xM)2
2~(t− tM)
]exp
[−1
2
mx2M
mσ2x + i~tM
][C1(xM) + C2(xM)]
(116)
where all constants, as is the custom, have been absorbed into N .
This expression above cannot be evaluated explicitly. However, because of the
properties of the step functions (Mauro 2003), it can be rewritten as
〈x|ψ(t)〉 = N(ˆ xA+δ
xA−δdxM exp
[im(x(t)− xM)2
2~(t− tM)− 1
2
mx2M
mσ2x + i~tM
]+
+
ˆ −xA+δ
−xA−δdxM exp
[im(x(t)− xM)2
2~(t− tM)− 1
2
mx2M
mσ2x + i~tM
])(117)
You can consider each of the two integrals to represent an independent wavefunction,
so that you can rewrite the whole expression as
〈x|ψ(t)〉 = N [ψ1(x) + ψ2(x)]
You can think of ψ1 as being the wavefunction coming out of ∆1 and ψ2 as the wave-
function associated with ∆2.
The equation (117) shows us that the phases of the two wavefunctions ψ1 and ψ2
are out of phase. You can see this because the integration is over two different inter-
66
vals, so the two phases will turn out different. This is significant, because wavefunc-
tions out of phase is characteristic of interference. When two wavefunctions with
different phases overlap, they form the characteristic interference pattern (Shankar
1988; Mauro 2003).
If we want the probability of finding a particle at the rightmost wall (the inter-
ference wall), we need to know the wavefunction 〈x|ψ(t)〉 at the rightmost wall. If
it takes time tR to reach this wall, that wavefunction can be simply represented as
〈x|ψ(tR)〉 = ψ(x, tR) (equivalent to taking (117) and plugging in t = tR). Then,
according to the postulates of section 2.5, we just have to apply the Born Rule to find
the probability density as a function of x:
ρ(x, t = tR) = ψ(x, tR)∗ψ(x, tR) = 〈ψ(tR)|x〉 〈x|ψ(tR)〉 =
= ψ1(x, tR)∗ψ1(x, tR)+ψ2(x, tR)∗ψ2(x, tR)+ψ1(x, tR)∗ψ2(x, tR)+ψ2(x, tR)∗ψ1(x, tR)
The last two terms are not identically zero (Mauro 2003). If we plot ρ vs x, we will
get the typical interference pattern we expected, as seen in Figure 2b.
This, of course, comes as no surprise. We are still dealing with quantum theory.
Following the same steps, though, will we get the same results with KvNM? We will
follow a similar series of steps to evaluate classical KvN theory.
For the KvNM case, we will use a double gaussian function because we are mak-
ing considerations with both the position and momentum (KvNM is a phase space
theory). A double gaussian (a gaussian in x and y coordinates; see Figure 3b) takes
on the mathematical form:
f(u, v) =1
√πσuσv
e−(u−µu)2/2σu2
e−(v−µv)2/2σv2 (118)
We are interested in the interference pattern that lies along the x direction. Taking
the KvN wavefunction ψ(x, px, t) = R(x, px, t)eiG(x,px,t) (like in section 5.1) and
plugging in (118) for R at t = 0 we will have
67
ψ(x, px, t = 0) =1
√πσxσp
exp
[− x2
2σx2− p2
x
2σp2+ iG(x, px, t = 0)
](119)
Please note here that σp refers to the momentum in the x direction. Following the Path
Integral Formulism, in order to know the know the wavefunction ψ at some later time
t we will need to utilize the Kernel. For this situation, the travelling classical particle
is free of any potential V (x, y), so we will need the free particle wavefunction. As
V (x, y) = 0, there will be no derivatives of the potential. We can take the results (75)
and (76) to get a straightforward and concise Kernel, which will make the calculation
we are about to do simple.
We can take (75) and rewrite it in the following way (dropping the constant 2π,
as it can be normalized away, and replacing q with the desired variable x):
δ(∆t
m(pj −m
xj+1 − xj∆t
))
= δ(pj∆t
m− xj+1 + xj
)In the continuum limit, we can just simply write this as:
= δ(p0t
m+ x0 − x
)
where x0 is the initial position of the particle, p0x is the initial momentum, and m is
the mass as before. This comes straight from elementary kinematics, as since there
is no potential, the derivative of the potential must be 0, so there is no acceleration:
Fx = −∂V∂x
= max ⇒ ax = 0
x(t) = x0 + v0xt+1
2axt
2 ⇒ x(t) = x0 + v0xt = x0 +p0xt
m
Using the momentum relation px = mvx on the last step. In other words, the Dirac
delta functional is enforcing classical position evolution, x(t).
68
The same can be done with (76) to get:
1
∆tδ(pj+1 − pj
∆t+ V ′(qj)
)= δ(pj+1 − pj
)
In the continuum limit remembering the direction is in x:
= δ(px − p0x
)This makes sense, considering that there are no forces present (because potential
is zero), and hence the momentum should be constant. This Dirac delta functional
enforces classical momentum conservation.
The Kernel K can therefore be succinctly represented as
K = δ(px − p0x
)δ(x− p0xt
m− x0
)(120)
And plugging this Kernel into the original equation (68) shows us that indeed it pro-
vides us with the correct result that we desire, forcing classical evolution of the KvN
wavefunction:
〈x, px|ψ(t)〉 =
ˆdx0 dp0xδ
(px − p0x
)δ(x− p0xt
m− x0
)ψ(x0, p0x, 0) =
= ψ(x− pxt
m, px, 0) = ψ(x, px, t),
utilizing the property (19) . This statement checks out, as we know that 〈x, p|ψ(t)〉 =
ψ(x, p, t).
Returning to (119), we will evaluate the same integral utilizing the Kernel. By
equation (68), the evolution of the wavefunction from its original state when t = 0 is
given by
〈x, px|ψ(t)〉 =1
√πσxσp
ˆdx0 dp0xδ
(px − p0x
)δ(x− p0xt
m− x0
)×
69
× exp
[− x2
0
2σx2− p2
0x
2σp2+ iG(x0, p0x, 0)
]=
=1
√πσxσp
exp
[− 1
2σ2x
(x− pxt
m
)2
− p2x
2σp2+ iG
(x− pxt
m, px
)](121)
Note that the above is strictly in terms of the position and momentum in the x direc-
tion. The above equation is true at least until it hits the middle wall with the slits,
because up until that point the wavefunction is that of a free particle. Using geomet-
ric considerations, we can figure out the wavefunction at the double slit wall. Since
there is no acceleration in the direction of travel (as −∂V/∂y = 0), we know from
kinematics:
vy =p0y
m=
∆y
∆t=yf − y0
tf − t0
Under the assumption that y(0) = 0 we can write:
yMt
=p0y
m⇒ tM =
yMm
p0y
, (122)
where tM is the time to the double slit wall. We can substitute this value into (121)
to know what is the classical wavefunction at that point in time.
Then, the wavefunction right after the two slits can be written as
ψ(x, px, tM + dt) = Nψ(x, px, tM)[C1(x) + C2(x)] (123)
Utilizing, as before, the functions (114) and (115) we had defined in the quantum
picture above. This is done in order to filter out the effects of the wavefunction that
actually goes through the slits. We do not care for the portion of the wavefunction
that does not go through the double slit. They will produce no effects. The above
equations make certain that only the wavefunction that goes through slits ∆1 and ∆2
will be studied. The N in (123) is a renormalization constant, similar to before, so
that´dx dpxψ
∗ψ = 1 (the renormalization condition for the KvN wavefunction.)
Using our Kernel for the free particle again, we will evaluate the wavefunction
70
using the path integral for the region after the two slits. We again utilize (68). Notice
that the initial wavefunction ψ(q0, p0, 0) in that equation starts at time zero, so we
will need to adjust the time at the double slit wall so that it is set to zero. We can do
this by substituting the variable t in the Kernel (120) with t− tM . This can be done
because time has the unique property that any point can be arbitrarily set to t = 0, and
by doing this swap we effectively set the time at the middle wall equal to zero (starts
at tM and evolves from there). So the integral we need to evaluate has the following
form:
〈x, px|ψ(t− tM)〉 =
ˆdxM dpMxδ
(px − pMx
)δ(x− pMx(t− tM)
m− xM
)×
×Nψ(xM , pMx, tM)[C1(xM) + C2(xM)]
Plugging everything in:
〈x, px|ψ(t− tM)〉 = N
ˆdx0 dp0xδ
(px − p0x
)δ(x− p0x(t− tM)
m− x0
)×
× exp
[− 1
2σ2x
(x0 −
p0xtMm
)2
− p20x
2σp2+ iG
(x0 −
p0xtMm
, p0x
)][C1(x0) + C2(x0)]
And then using (19) to evaluate:
〈x, px|ψ(t− tM)〉 = Nexp[− 1
2σ2x
(x− px(t− tM)
m− pxtM
m
)2
− p2x
2σp2+
+iG(x−px(t− tM)
m−pxtM
m, px
)][C1
(x−px(t− tM)
m
)+C2
(x−px(t− tM)
m
)]=
= N exp
[− 1
2σ2x
(x− pxt
m
)2
− p2x
2σp2+ iG
(x− pxt
m, px
)]×
×[C1
(x− px(t− tM)
m
)+ C2
(x− px(t− tM)
m
)](124)
71
Similar to (122), under geometric considerations one can show that the time to the
back wall is tR = myR/p0y, and so we can substitute this and (122) into the above
expression to get the expression for the wavefunction at the back wall (wall where we
should see the interference pattern emerge in QM).
We want to know the probability distribution for finding a particle at coordinate
x at this back wall (Mauro 2003). We have no need of the momentum information,
as the interference pattern is a shape that appears in space. So what we can do is get
the probability distribution in terms of x is the following:
P (x) =
ˆ ∞−∞
dpx|ψ(x, px, tR−tM)|2 =
ˆ ∞−∞
dpx 〈ψ(tR − tM)|x, px〉 〈x, px|ψ(tR − tM)〉
Plugging in (124) for 〈x, px|ψ(tR − tM)〉 we get:
P (x) = N
ˆ ∞−∞
dpx exp
[− 1
2σ2x
(x− pxtR
m
)2
− p2x
2σp2
]2
exp
[−iG
(x− pxtR
m, px
)]×
× exp
[iG(x− pxtR
m, px
)][C1
(x− px(tR − tM)
m
)+ C2
(x− px(tR − tM)
m
)]2
=
= N
ˆ ∞−∞
dpx exp
[− 1
2σ2x
(x− pxtR
m
)2
− p2x
2σp2
]2
×
×[C1
(x− px(tR − tM)
m
)+ C2
(x− px(tR − tM)
m
)]2
We will use a very special property of the θ-Heavyside step function at this point. We
know that the following important property is true (Mauro 2003):
(C1 + C2)2 = C1 + C2
Therefore:
P (x) = N
ˆ ∞−∞
dpx exp
[− 1
2σ2x
(x− pxtR
m
)2
− p2x
2σp2
]2
×
72
×[C1
(x− px(tR − tM)
m
)+ C2
(x− px(tR − tM)
m
)]This last equation is very important, because it shows that no interference pattern
will emerge from the classical KvN wavefunction. If you look at the above equation
closely, you can see that the probability distribution of x is a superposition of two
probability distributions:
P (x) = N
ˆ ∞−∞
dpx exp
[− 1
2σ2x
(x− pxtR
m
)2
− p2x
2σp2
]2
C1
(x− px(tR − tM)
m
)+
+N
ˆ ∞−∞
dpx exp
[− 1
2σ2x
(x− pxtR
m
)2
− p2x
2σp2
]2
C2
(x− px(tR − tM)
m
)The first integral in the above expression is the probability at x if only the first slit
∆1 was opened and the second slit ∆2 was closed, and the second integral in the
above expression is the probability at x if only the second slit ∆2was opened and
the first slit ∆1 was closed. This can be seen from how C1 filters out any portion
of the wavefunction that does not go through ∆1 and C2 filters out any portion of
the wavefunction that does not go through ∆2. This means that the total probability
distribution is the sum of the probability distribution of each slit, which is exactly
what you would expect for particle instead of wave behavior (no interference effects).
Notice also that the phase of the wavefunction has completely disappeared from
the above equation. You will see this is important, because for interference effects
we will want the phase to play an active role. Waves must be out of phase in order
to produce interference effects, not have the phase cancel each other out like in the
above equations.
If you graph the above function, you will see that indeed no interference fringes
were produced (see Figure 2a). Koopman wavefunctions therefore successfully pre-
dict the behavior of classical particles. Despite their complex nature in a Hilbert
space, they do no produce any interference effects. All the interference effects com-
pletely disappear. It shows how wavefunctions can describe classical physics.
73
6 Miscellaneous KvN Topics of Interest
6.1 Operational Dynamic Modeling
The history of physics is full of failed models, models that did not stand the test of
time. Physics is often carried out in a trial and error fashion, a process that can be
repetitive, messy, and slow. This raises the question if discovery and model building
in physics can be optimized, for the sake of speeding up the progress of discovery.
One such optimization model is called Operation Dynamical Modeling (ODM)
and utilizes KvNM as an essential ingredient (Bondar et al. 2012). The two main
pillars of modern physics is QM and CM. 22 Since we have seen that both QM and
CM must obey the Ehrenfest Theorems, any particular model must conform to either
QM or CM to the best of our knowledge, and if it falls outside either pillar, the model is
quite likely false. If repeated measurements influence the outcome of the experiment,
one knows that there must be a noncommutative algebra involved (i.e., there must
be a nonzero commutator involved, like in QM.) If repeated measurements do not
generally alter the outcome of the experiment in the fashion of QM, you can assume
a commutator of 0 between all variables involved.
Since the Ehrenfest Theorems deal with dynamical weighted averages (expecta-
tion values), this is a statistical test of different models. If a model does not obey
the Ehrenfest Theorems, it is quite likely to be false, and progress can be saved by
either fixing the model so that it is inline with the fundamental Ehrenfest Theorems
or scrapping the model.
Not only does ODM allow quantum and classical physics, it also explains the
nexus of the two. Semi-classical systems are incorporated into ODM, as they must
also obey the Ehrenfest Theorems. By construction, one could create a commutator
that encompasses both the classical and quantum worlds, such as [q, p] = i~κ where
κ is a constant that lies between 0 (classical world) and 1 (purely quantum world).
By taking the limit of κ goes to 0, one should be able to recover either CM.22Note, General Relativity and Special Relativity are here considered to be part of CM.
74
This melding, in fact, results in the derivation of the semi-classical Wigner quasiprob-
ability distribution. This function is known to accurately describe the interface be-
tween the quantum world and the classical world. As the Schrodinger equation is
generally written in terms of configuration space, the Wigner quasiprobability dis-
tribution is important because it recasts QM into a phase space theory. The Wigner
function has a few odd properties, including the fact that it can take on negative prob-
abilities, but thanks to ODM some light has been shed to the reasons why this might
be the case (as we will discuss later.)
We will begin with Stone’s Theorem again like we did in section 3, i.e. i ∂∂t|ψ〉 =
H |ψ〉, and try to create an operator HQC which will encompass both the classical and
quantum worlds using our newly constructed commutator [q, p] = i~κ, 0 ≤ κ ≤ 1.
We must have a unified quantum-classical algebra which gives us the Koopman-
von Neumann Algebra (43) when κ = 0 and a quantum algebra for κ = 1. We
desire:
limκ→0
qQC = qc
limκ→0
pQC = pc
limκ→0
HQC = ~K
In order to reflect the KvN Algebra, Bondar et al. 2013 proposed the following unified
Classical-Quantum Algebra:
[qQC , θQC ] = [pQC , λQC ] = i (125)
All other commutators will equal 0. For the unified dynamics, we first begin with the
Ehrenfest Theorems (equations (36) and (37) for reference) as before:
d
dt〈q〉 =
d
dt〈ψ(t)| qQC |ψ(t)〉 =
1
m〈p〉 =
1
m〈ψ(t)| pQC |ψ(t)〉
d
dt〈p〉 =
d
dt〈ψ(t)| pQC |ψ(t)〉 = 〈−V ′(q)〉 = 〈ψ(t)| − V ′(qQC) |ψ(t)〉
75
Following the same series of steps as in section 3, we will obtain the differential
equations:
κm∂H
∂pQC+m
~∂H
∂θQC= pQC
κ∂H
∂qQC− 1
~∂H
∂λQC= V ′(qQC)
The general solution H to the two above differential equation is simply:
H(qQC , pQC , θQC , λQC) =1
κ
[ p2QC
2m+ V (qQC)
]+ F (pQC − ~κθQC , qQC + ~κλQC)
where the function F is a function undetectable to experiment because one can show
that for any observable A, you have [F , A] = 0 (Bondar et al. 2012).
The quantum operators can be constructed from classical ones, so Bondar et al.
2012 propose that
qQC = qc − ~κλc2
pQC = pc + ~κθc2
θQC = θc
λQC = λc
It is clear that the above expressions have a nice, smooth classical limit as κ → 0.
All the variables operators become classical operators in that limit.
To make the limit smooth as you go from the quantum Hamiltonian H to ~K as
κ goes from 1 to 0, Bondar et al. 2012 dervied:
HQC =1
κ[p2QC
2m+ V (qQC)]− 1
2mκ(pQC − ~κθQC)2 − 1
κV (qQC + ~κλQC) (126)
Equivalently, you can write by plugging in the quantum-classical operators:
76
HQC =~mpcθc +
1
κV (qc −
~κ2λc)−
1
κV (qc +
~κ2λc)
The above expression can be shown to lead to the phase space formulation of QM, the
Wigner quasiprobability distribution. This derivation will be left for the next section.
The Wigner distribution is significant because it is known to accurately model the
‘nexus’ between quantum and classical worlds (Bondar et al. 2012; Bondar et al.
2013). It also has a nice classical limit (Bondar et al. 2012; Bondar et al. 2013).
Independently deriving this distribution indicates that the Ehrenfest Theorems apply
at all times, including in the regime of quantum/classical boundary. ODM, therefore,
is a very successful melding of the two worlds.
ODM suggests that any physical theory will follow KvNM, QM, or the above
melding of the two. KvNM is fundamentally important because any novel classical
theory proposed must be consistent with it. We will see certain applications of ODM
and its successes in the pages that follow. Without KvNM, there would be no unifying
paradigm based in the Ehrenfest Theorems. ODM is a methodology that will pave
for more efficient discovery in Physics.
6.2 Wigner Quasiprobability Distribution
The Wigner Quasiprobability Distribution connects our understanding of the quan-
tum wavefunction to phase space. It is an important area of study developed by Eu-
gen Wigner in 1932 and independently formulated by another eminent mathematical
physicist, Jose Enrique Moyal. The Wigner Distribution has a variety of applications,
from quantum optics to quantum computing (Bondar et al. 2013). It has one signifi-
cantly odd feature though. This distribution allows for negative probabilities, which
is axiomatically impossible by mainstream probability theories (Bondar et al. 2013).
ODM sheds some light on this odd feature. The Wigner Distribution can be derived
from the unified dynamics of ODM, and we will proceed to do so below.
It is very simple to derive the Wigner quasiprobability distribution from ODM’s
77
unified dynamics. Again, we hark back to Stone’s Theorem, in order to try to figure
out the generator of motion:
i~d
dt|ψ(t)κ〉 = HQC |ψ(t)κ〉
We will sandwich it from the left with |qc, λc〉 to get the λc, qc representation. Refer-
ring back to the rules for operators acting on vector kets, we can see that:
〈q, λ| qc |ψ(t)〉 = [〈q, λ| qc] |ψ(t)〉 = q 〈q, λ|ψ(t)〉 (127)
〈q, λ| λc |ψ(t)〉 = [〈q, λ| λc] |ψ(t)〉 = λ 〈q, λ|ψ(t)〉 (128)
where q and λ are the eigenvalues. From Theorem IV of Appendix A, we can say
thatB = p,A = λ, κ = −1 (since [λ, p] = −i) 23 and keep in mind that the Classical
Commutator is [qc, pc] = 0. Therefore:
〈q, λ| pc |ψ(t)〉 = −i(−1)∂
∂λ〈q, λ|ψ(t)〉 = i
∂
∂λ〈q, λ|ψ(t)〉 (129)
There is no partial derivative with respect to q as the Classical Commutator between
q and p is 0. Theorem IV can also be applied in the situation where B = θ in order
to derive:
〈q, λ| θc |ψ(t)〉 = −i ∂∂q〈q, λ|ψ(t)〉 (130)
Now we create the actual ‘sandwich’ from Stone’s Theorem:
〈q, λ| i~ ddt|ψ(t)κ〉 = 〈q, λ| [ ~
mpcθc +
1
κV (qc −
~κ2λc)−
1
κV (qc +
~κ2λc)] |ψ(t)κ〉
i~d
dt〈q, λ|ψ(t)κ〉 = 〈q, λ| ~
mpcθc |ψ(t)κ〉+ 〈q, λ| 1
κV (qc −
~κ2λc) |ψ(t)κ〉+
23This is because [λ, p] = −[p, λ] and as indicated earlier, [p, λ] = i.
78
−〈q, λ| 1κV (qc +
~κ2λc) |ψ(t)κ〉
Utilizing the previous results for Theorem I (see footnote 13), we can see:
i~d
dt〈q, λ|ψ(t)κ〉 =
~m〈q, λ| pcθc |ψ(t)κ〉+
1
κV (q−~κ
2λ) 〈q, λ|ψ(t)κ〉−
1
κV (q+
~κ2λ) 〈q, λ|ψ(t)κ〉
Finally utilizing the results from Theorem IV above:
i~d
dt〈q, λ|ψ(t)κ〉 = −i2 ~
m
∂2
∂λ∂q〈q, λ|ψ(t)κ〉+
1
κV (q − ~κ
2λ) 〈q, λ|ψ(t)κ〉+
−1
κV (q +
~κ2λ) 〈q, λ|ψ(t)κ〉
Let us move all the terms to the left hand side of the equation, so that the left hand
side equals 0 pm the right hand side. Introducing two new variables, u = q− ~κλ/2
and v = q + ~κλ/2, we will rewrite the whole equation so it has a prettier form. To
eliminate the double partial derivative, we can take the following steps:
u− v = q − ~κλ2− q − ~κλ
2= −~κλ
u+ v = q − ~κλ2
+ q +~κλ
2= 2q
Therefore the partial derivative can be rewritten:
∂2
∂λ∂q=
∂2
∂(u−v−~κ)∂(u+v2
)= −2~κ
∂2
∂(u− v)∂(u+ v)= −2~κ
[ ∂2
∂u2− ∂2
∂v2
]
Recasting the entire equation in terms of u and v:
[i~κ
∂
∂t− (~κ)2
2m
( ∂2
∂u2− ∂2
∂v2
)− V (u) + V (v)
]〈q, λ|ψ(t)κ〉 = 0 (131)
Here, 〈q, λ|ψ(t)κ〉 is equal to the density matrix ρκ(u, v, t) up to a constant of pro-
79
portionality C, i.e., ρκ(u, v, t) = C 〈q, λ|ψ(t)κ〉.
We can transition to q, p representation to derive the Wigner distribution from the
above using Theorem II with A = (q, λ):
〈q, p|ψ(t)κ〉 = 〈q, p| I |ψ(t)κ〉 =
ˆdλ 〈q, p|q, λ〉 〈q, λ|ψ(t)κ〉
Note that we do not integrate over q due to the commutator relationships [q, q] =
[qc, pc] = 0, whereas [pc, λc] = i. Since [q, q] = [qc, pc] = 0 it makes no sense
to integrate over q, as might be initially implied by looking at the expressions akin
to (31). Since the other commutator is nonzero, we will retain integration over λ,
however. We know that ρκ(u, v, t) = C 〈q, λ|ψ(t)κ〉, so we will plug that in.
The question now is to figure out what 〈q, p|q, λ〉 is. We will utilize Theorem IV
and the commutator relationships in order to deduce this. Since [q, q] = [qc, pc] = 0,
solving this expression is equivalent to solving 〈p|λ〉, as the other terms commute
(i.e., the commutator of their operators is equal to 0) and are therefore irrelevant to
this problem. By Theorem IV, we set A = p, B = λ, and κ = 1. We know that this
expression is therefore
〈q, p|q, λ〉 = 〈p|λ〉 =1√2πeipλ
Putting it all together:
〈q, p|ψ(t)κ〉 =
ˆdλ
C√2πρκ(q −
~κλ2, q +
~κλ2, t)eipλ (132)
The above expression shows us that the wavefunction 〈q, p|ψ(t)κ〉 is equal to the
Wigner quasi-probability distribution up to a constant of proportionality C. The
Wigner distribution is very accurate when it comes to describing the boundary of
Quantum and Classical (Bondar et al. 2012; Bondar et al. 2013), so this is no trivial
derivation.
What ODM reveals is that the Wigner “distribution” is therefore a type of wave-
function for a quantum particle at a point in classical phase space. The strange prop-
80
erty of having negative probabilities can by explained in the ODM framework as
being due to the fact that the Wigner “distribution” is not a distribution at all! It is a
wavefunction. Wavefunctions, unlike probability distributions, can take on negative
values (Bondar et al. 2013). The KvN wavefunction can have negative or imaginary
weights to it, whereas traditional Louiville theory gives only positive, real weights to
the same trajectories (Bondar et al. 2013).
It is worth to highlight one more interesting result of the above analysis. Although
not directly physically observable, the operators θ and λ can be indirectly observed
through the Wigner wavefunction. They are a critical ingredient as the so-called
‘Bopp Operators’.
6.3 Time Dependent Harmonic Oscillator
The Harominc Oscillator is one of the most thoroughly studied models in all of
physics. Its application arises in many situations. We are especially interested in
the Time Dependent Harmonic Oscillator, where factors like the mass or frequency
might change with time. Because of the common mathematical basis for QM and
KvNM, it should come to no surprise that common sets of tools can be used to in-
vestigate both. The Harmonic Oscillator is no exception. The operatorial tools to
investigate quantum oscillators can be used to explore classical (KvN) oscillators
and vise versa.
For a classical or quantum Time Dependent Harmonic Oscillator we need a quadratic
potential of the following form:
V (q, t) = k(t)q2
2
The above potential is modelled off of the classical spring potential. An ideal block-
mass and spring system can undergo harmonic oscillation. The spring potential is
given by V = (1/2)kq2 with spring constant k and displacement q. Here, we add an
extra layer of complexity by making the spring constant time dependent. This is our
81
model potential for all Time Dependent Harmonic Oscillation, whether quantum or
classical.
In the quantum case, we would plug in the above expression for the potential term
into the Schrodinger equation to solve for the Quantum Oscillator. In the classical
case, we plug in the derivative of the above expression into the Koopman Generator
(45):
∂
∂qV (q, t) =
∂
∂qk(t)
q2
2= k(t)q = V ′(q)
i∂
∂t|ψ(t)〉 =
[ pθm− V ′(q)λ
]|ψ(t)〉 =
[ pθm− k(t)λq
]|ψ(t)〉 .
Similar to what we did before (50) in section 3, we can attack both sides with the bra
vector 〈q, p| to get:
〈q, p| i ∂∂t|ψ(t)〉 = 〈q, p|
[ pθm− k(t)λq
]|ψ(t)〉 ,
i∂
∂t〈q, p|ψ(t)〉 = 〈q, p| pθ
m|ψ(t)〉 − 〈q, p| k(t)λq |ψ(t)〉 ,
i∂
∂tψ(q, p, t) =
[ pθm− k(t)λq
]ψ(q, p, t) (133)
where we colloquially say that q = q, p = p, λ = −i ∂∂p
, and θ = −i ∂∂q
, just as we
derived in section 3. Notice as for the Path Integral (section 4.2), we choose to set
the constant C to zero in the Koopman Generator for simplicity’s sake.
One way to solve for the equations of motion of a Harmonic Oscillator is to use
Ermakov-Lewis invariants (Lewis Jr. 1967). If you have a Hamiltonian of the form
H =1
2[p2 + Ω2(t)q2],
where Ω(t) is an arbitrary function with time dependence, you can use the invariant
to write a transformation where the time dependent term Ω(t) completely disappears.
82
This makes solving the problem much simpler. Like it sounds, the Ermakov-Lewis
invariant I is an operator that does not change (stays constant in time). The invariant
I obeys the following rule for the quantum system (Lewis Jr. 1967; Lewis Jr. and
Riesenfeld 1968):
dI
dt=∂I
∂t+
1
i~[I , H] = 0. (134)
H is the generator of motion, usually the quantum Hamiltonian, but in the classical
case it will be the Koopman Generator. Based on the Koopman Algebra (43), Ramos-
Prieto et al. 2018 deduced the following expression for I:
I = α0q2 + α1θ
2 + α2p2 + α3λ
2 + α4qp+ α5θλ
where all the αj are time dependent. This was a scientific guess basically, as there is
no simple way to derive this fact. Plugging the above expression into (134), we get:
∂α0
∂tq2 +
∂α1
∂tθ2 +
∂α2
∂tp2 +
∂α3
∂tλ2 +
∂α4
∂tqp+
∂α5
∂tθλ+
1
i~[I , K] = 0 (135)
Evaluating [I , K], we first can do (13):
[I , K] =[I ,pθ
m− k(t)λq
]=[I ,pθ
m
]−[I , k(t)λq
](136)
Taking the first term and applying (11) and (13):
[I ,pθ
m
]= −
[ pθm, I]
= −[ pθm, α0q
2 + α1θ2 + α2p
2 + α3λ2 + α4qp+ α5θλ
]=
= −[ pθm, α0q
2]−[ pθm, α1θ
2]−[ pθm, α2p
2]−[ pθm, α3λ
2]−[ pθm, α4qp
]−[ pθm, α5θλ
]=
= − p
m
[θ, α0q
2]−[ pm, α0q
2]θ− p
m
[θ, α1θ
2]−[ pm, α1θ
2]θ− p
m
[θ, α2p
2]−[ pm, α2p
2]θ+
83
− p
m
[θ, α3λ
2]−[ pm, α3λ
2]θ− p
m
[θ, α4qp
]−[ pm, α4qp
]θ− p
m
[θ, α5θλ
]−[ pm, α5θλ
]θ
(137)
Utilizing our nifty identities (10) and (12) and Koopman Algebra again the following
can be said about the above commutators:
[θ, α0q
2]
= α0q[θ, q]
+[θ, α0q
]q = −2iα0q
[θ, α1θ
2]
= α1θ[θ, θ]
+[θ, θ]α1θ = 0
[θ, α2p
2]
= α2p[θ, p]
+[θ, p]α2p = 0
[θ, α3λ
2]
= α3λ[θ, λ]
+[θ, λ]α3λ = 0
[θ, α4qp
]= α4q
[θ, p]
+[θ, q]α4p = −iα4p
[θ, α5θλ
]= α5θ
[θ, λ]
+[θ, θ]α5λ = 0
[ pm, α0q
2]
= α0q[p/m, q
]+[p/m, q
]α0q = 0
[ pm, α1θ
2]
= α1θ[ pm, θ]
+[ pm, θ]α1θ = 0
[ pm, α2p
2]
=α2
mp[p, p]
+[p, p]α2
mp = 0
[ pm, α3λ
2]
=α3
mλ[p, λ]
+[p, λ]α3
mλ =
2iα3
mλ
84
[ pm, α4qp
]=α4
mq[p, p]
+[p, q]α4
mp = 0
[ pm, α5θλ
]=α5
mθ[p, λ]
+[p, θ]α5
mλ =
iα5
mθ
Ergo, the above expression (137) reduces to:
[I ,pθ
m
]= 2iα0
pq
m+ iα4
p2
m− 2iα3
λθ
m− iα5
θ2
m(138)
Now, we do the same for [I , k(t)λq]:
[I , k(t)λq
]= −
[k(t)λq, I
]= −
[k(t)λq, α0q
2+α1θ2+α2p
2+α3λ2+α4qp+α5θλ
]=
= −[k(t)λq, α0q
2]−[k(t)λq, α1θ
2]−[k(t)λq, α2p
2]+
−[k(t)λq, α3λ
2]−[k(t)λq, α4qp
]−[k(t)λq, α5θλ
]=
= −k(t)λ[q, α0q
2]−[k(t)λ, α0q
2]q − k(t)λ
[q, α1θ
2]+
−[k(t)λ, α1θ
2]q − k(t)λ
[q, α2p
2]−[k(t)λ, α2p
2]q+
−k(t)λ[q, α3λ
2]−[k(t)λ, α3λ
2]q − k(t)λ
[q, α4qp
]+
−[k(t)λ, α4qp
]q − k(t)λ
[q, α5θλ
]−[k(t)λ, α5θλ
]q
We can perform a similar series of steps to achieve:
[q, α0q
2]
= 0
[k(t)λ, α0q
2]
= 0
85
[q, α1θ
2]
= α1θ[q, θ]
+[q, θ]α1θ = 2iα1θ
[k(t)λ, α1θ
2]
= 0
[q, α2p
2]
= 0
[k(t)λ, α2p
2]
= k(t)α2p[λ, p]
+[λ, p]k(t)α2p = −2ik(t)α2p
[q, α3λ
2]
= 0
[k(t)λ, α3λ
2]
= 0
[q, α4qp
]= 0
[k(t)λ, α4qp
]= k(t)α4q
[λ, p]
+[λ, q]k(t)α4p = −ik(t)α4q
[q, α5θλ
]= α5θ
[q, λ]
+[q, θ]α5λ = iα5λ
[k(t)λ, α5θλ
]= 0
Therefore, putting it all together:
[I , k(t)λq
]= −2iα1k(t)λθ + 2ik(t)α2pq + ik(t)α4q
2 − ik(t)α5λ2 (139)
86
Utilizing equations (136), (138), (139) with (135), we can compute the invariant
to be:
∂α0
∂tq2+
∂α1
∂tθ2+
∂α2
∂tp2+
∂α3
∂tλ2+
∂α4
∂tqp+
∂α5
∂tθλ+
1
i~
(2iα0
pq
m+iα4
p2
m−2iα3
λθ
m−iα5
θ2
m+
+ 2iα1k(t)λθ − 2ik(t)α2pq − ik(t)α4q2 + ik(t)α5λ
2)
= 0 (140)
Combining like terms:
(∂α0
∂t− k(t)α4
~
)q2 +
(∂α1
∂t− α5
~m
)θ2 +
(∂α2
∂t+α4
~m
)p2 +
(∂α3
∂t+k(t)α5
~
)λ2+
+(∂α4
∂t+
2α0
~m− 2k(t)α2
~
)qp+
(∂α5
∂t− 2α3
~m+
2k(t)α1
~
)θλ = 0 (141)
It must equal zero, but each of the operators we know is nonzero, so the only way
for this equation to be true is if all the coefficients above equal zero. Setting all the
coefficients equal to zero and solving the system of equations will lead you to the
dynamics of the (here classical) Harmonic Oscillator.
For instance, if we solve for α1 we will get the equation:
∂2α1
∂t2+ 2k(t)α1 =
1
2α1
(∂α1
∂t
)2
+C
α1
And if we define an arbitrary function called ρ(t) so that α1 = ρ2(t)2
, we will have
d2ρ
dt2+ k(t)ρ = C/ρ3
which is an equation we call the Ermakov equation (Lewis Jr. 1967; Lewis Jr. and
Riesenfeld 1968). Using the Ermakov equation, all other coefficients can be rewritten
so that the invariant has the following form:
I =1
2
[ qρ
+ (dρ
dtq − ρp)2 +
λ2
ρ2+ (
dρ
dtλ+ ρθ)2
].
87
Notice that the above classical Ermakov-Lewis invariant cannot be directly observed.
This is because operators like θ and λ are not directly tied to physically observable
quantities (Ramos-Prieto et al. 2018). Nevertheless, it serves as an important tool in
solving Harmonic Oscillation problems.
We can define following the unitary transformation
T = exp
[iln(ρ)
2(qp+ pq)
]exp
[− i
2ρ
dρ
dtq2
],
which for the quantum case transforms the time dependent problem described above
to a time independent problem (Lewis Jr. 1967; Lewis Jr. and Riesenfeld 1968) by
the following:
1
2[p2 + q2] = T IT †.
In the classical case, we can define the following two unitary transformations (Ramos-
Prieto et al. 2018):
T1 = exp
[i
ρ
dρ
dtqλ
],
T2 = exp
[iln(ρ)
2(qθ + θq)
]exp
[−i ln(ρ)
2(pλ+ λp)
].
We can define the new wavefunction Ψ(q, p, t) = T2T1ψ(q, p, t) and we can rewrite
our equation of interest (133):
i∂
∂tΨ(q, p, t) =
1
ρ2
[ pθm− λq
]Ψ(q, p, t). (142)
where the values of the operators are exactly the same as in (133). The original
wavefunction equation got transformed into a wavefunction equation without the time
dependent potential term in front of the position operator. This can be much more
easily solved to deduce the statistical motion of the classical particles.
Note that this method was originally used in the operatorial form for the Schrodinger
equation (Lewis Jr. 1967). We are able to successfully implement this procedure for
88
KvNM as well. If you graph the phase space diagrams for a selected k(t), you can
see that it is indeed the graph of an oscillator (Ramos-Prieto et al. 2018). KvNM
is successful at bridging methods that involve quantum-like operators with classical
physics.
6.4 Aharonov–Bohm Effect
The Aharonov-Bohm Effect is a surprising phenomena that emerges from the quan-
tum wavefunction interacting with the magnetic potential. In classical electromag-
netic theory, one can take the electric ~E and magnetic ~B fields and decompose them
into various derivatives of electric and magnetic potentials:
~E = −~∇φ,
~B = ~∇× ~A,
where φ is the electric scalar potential and ~A is the magnetic vector potential. Both
of these potentials have the unique property that they make ~E and ~B gauge invariant,
in other words, that a ‘shift’ χ will not produce different magnetic or electric fields.
In other words, the new potentials ~A′ and φ′ defined by
~A′ = ~A+ ~∇χ,
and
φ′ = φ− ∂χ
∂t,
will produce the same magnetic and electric fields as ~A and φ. These potentials have
historically been seen as useful mathematical constructs with little physical bearing
on reality in themselves. The Aharonov-Bohm effect soon put this assumption to the
test.
The Aharonov-Bohm effect is a a quantum interference, much like that in Young’s
Double Slit Experiment, produced by the magnetic potential changing the phases of
89
the quantum wavefunctions. When different wavefunctions with difference phases
meet, they interfere with themselves. The surprising thing is that the interference
pattern does not depend on the magnetic field ~B but directly on the magnetic potential
~A. In other words, if the magnetic field ~B = ~∇ × ~A is zero but the potential is
nonzero, an effect will still be observed. This suggests that the magnetic potential
~A is more fundamental than the magnetic field ~B, which have always been seen as
being on equal footing in classical theory.
Aharonov and Bohm described the theory behind the phenomena in 1959, but
the idea was not experimentally verified for the magnetic potential until after their
deaths. The experiment used a solenoid, which is the model we will use here for the
magnetic potential. Recall that a solenoid is a cylindrical device made of winding
wire which produces a magnetic field inside the solenoid and zero magnetic field
outside the solenoid.
The generic setup of the experiment for either type of potential is depicted in Fig-
ure 4. A coherent beam of electrons would be split into two branches, travel through
an electric or magnetic potential where ~E = ~0 or ~B = ~0, and the two branches would
then be recombined. Each branch’s wavefunction took a slightly different path, so
when the two branches are recombined an interference pattern should emerge be-
cause of a phase difference acumulated along each path. This is a semiclassical way
to visualize the experiment (recall that in QM proper, the particle actually takes all
possible paths between source and sink in the allotted time.)
To understand mathematically how the phase shift happens, first recall the polar
form of the wavefunction (3) and what the Action represents (51). We can write that:
ψ = R exp
[i
~
ˆ t
0
T − V dt]. (143)
The above wavefunction can be written for a region with no potential energy (free
particle case) as:
ψ0 = R exp
[i
~
ˆ t
0
Tdt
].
90
And so (143) can be rewritten in the generic case as:
ψ = ψ0 exp
[− i~
ˆ t
0
V dt
].
For a wavefunction going through an electric potential, from CM we know we can
write that the potential energy is V = eφ, where e is the charge of the particle’s
wavefunction and φ is the scalar potential. And so the generic wavefunction can be
written as the following for the case with electric potential:
ψ = ψ0 exp
[−ie
~
ˆ t
0
φdt
]. (144)
The above formulas can be used to describe what should happen in the experiment in
the presence of the electric potential.
Refer to Figure 4. Starting off, the beam of coherent electrons has wavefunction
ψ0 in the absence of an external potential. It is split into two, and both branches travel
through a region of potential φ, so each wavefunction gains the form of (144). The
branching could be represented as ψ01 and ψ0
2 upon reaching the area with electric
potential at time t1. After some evolution they leave the potential at time t2, and they
are then recombined. The first branch travelled along potential φ1 and the second
branch travelled along φ2, in this semiclassical picture. You can write the following
superposition for the final wavefunction:
ψf = ψ1 + ψ2 = ψ01 exp
[− i~C1
]+ ψ0
2 exp
[− i~C2
],
where we define the phases
C1 = e
ˆ t2
t1
φ1dt,
and
C2 = e
ˆ t2
t1
φ2dt.
91
Since C1 and C2 are different, we have gained a phase shift of (C1 −C2)/~ 24 There
is absolutely no classical force ~F = e ~E exerted on the electrons as there is no elec-
tric field, but a measurable effect should still be produced on the electron’s behavior
(Aharonov and Bohm 1959).
A similar analysis can be done with the magnetic vector potential. Following
Aharonov and Bohm 1959, we can generalize (144) to include the magnetic potential.
In relativity, you can write the electromagnetic potential as the four vector Aµ =
(φ/c ~A) = (φ/c Ax Ay Az) where as before, φ is the electric potential and ~A is
the magnetic potential. The position four vector can be represented is xµ = (ct ~x) =
(ct x y z). Rewriting the position four vector as the differential dxµ, we can generalize
the potential integral to be:
e
~
˛Aµdxµ =
e
~
˛ (φdt−
~A
c· d~x). (145)
The path of integration has to be closed in spacetime (Aharonov and Bohm 1959).
Let us assume that the variable t is a constant in order to understand how interference
could arise strictly from the magnetic potential. The above integral suggests that we
will expect the following phase shift to hold true:
(C1 − C2)/~ = − e
c~
˛~A · d~x
We can set up a similar situation as to before.
Starting off, the beam of electrons has wavefunction ψ0 in the absence of an ex-
ternal magnetic potential. It is split into two, and both branches travel through a
region of potential ~A. Each wavefunction gains the form of (144) except we re-
place exp[−(ie/~)
´φdt]
with exp[−(ie/~c)
´~A · d~x
]as described above. In other
words:24This comes straight from the Euler formula. Recall that:
ei(θ+C) = cos(θ + C) + isin(θ + C)
. Here, the C is the phase. If the phase of two waves are different, we should expect interference.
92
ψ = ψ0 exp
[− ie~c
ˆ~A · d~x
]. (146)
The branching could be represented as ψ01 and ψ0
2 upon reaching the area with mag-
netic potential at time t1. After some evolution they leave the potential at time t2, and
they are then recombined. You can write the following superposition:
ψf = ψ1 + ψ2 = ψ01 exp
[− i~C1
]+ ψ0
2 exp
[− i~C2
],
where we define
C1 =e
c
ˆ ~x(t2)
~x(t1)
~A · d~x′,
and
C2 = −ec
ˆ ~x(t2)
~x(t1)
~A · d~x′,
where we assume equal path length for each branch. If the path length is indeed kept
the same, then you will have phase difference (C1 − C2)/~ = (e/c~)´~A · d~x in the
end (Aharonov and Bohm 1959). Note that this is true even if the magnetic force
~F = q~v × ~B is zero because ~B = ~0. This is a purely quantum effect.
The electric Aharonov-Bohm effect described above has been theorized, but no
experimental data exists for it as of yet. We will focus on the magnetic potential there-
fore. No further space will be dedicated to the electric potential, but all considerations
from here on out will be framed in the context of the magnetic potential.
One may ask the obvious question, in the descriptions above, if all possible paths
are taken in the Path Integral, would not some paths go through the magnetic field,
explaining the outcome of the experiment? Since the Feynman Path Integral defines
a set starting point and endpoint in a set time period, it is actually possible to show
that you can set up a barrier (described in Aharonov and Bohm 1959) so that no
wavefunction directly interacts with the inside of the solenoid in the magnetic case.
93
The changes we see, therefore, must be due to the nonzero magnetic potential instead
of the magnetic field itself.
KvNM, having the mathematical form of quantum theory, might be one of the
best tools to show that the Aharonov-Bohm effect does not exist in CM (Mauro 2003).
This is a purely quantum phenomenon. Interference is not something expected in the
domain of CM, but is expected under QM. This can be traced to the behavior of the
generator of motion, the quantum Hamiltonian, having the following form under QM
in the presence of the magnetic potential for a free particle:
H =1
2m
(p− e
c~A)2
In the presence of the magnetic potential, the free particle Hamiltonian has the p
operator replaced with p − (e/c) ~A. This is described as a minimal coupling rule
(Gozzi and Mauro 2002). If KvNM describes CM precisely, we should expect that
the generator of motion, the Koopman Generator (45), will not be affected by the
magnetic potential even if we treat it mathematically in the exact same ways as we
treat the quantum Hamiltonian. In the following pages, we will put this to the test.
First, imagine two cylinders, one within the other (following Gozzi and Mauro
2002; Mauro 2003). Let them be infinite in height. The smaller cylinder on the
inside will represent the solenoid which houses a magnetic field. The outside of the
solenoid (area inside the larger cylinder but outside of the smaller cylinder) will have
no magnetic field, but there will be a magnetic potential. Since we are dealing with
cylinders, it would be useful to rewrite the generators of motion in terms of cylindrical
coordinates. The conversion of cartesian to cylindrical coordinates is given by:
x = r cos(θ),
y = r sin(θ),
and
z = z, (147)
94
where r is the distance from the z axis and θ is the angle in the xy plane.
With this notation, the quantum Hamiltonian for a free particle can be written:
H = − ~2
2m~∇2 = − ~2
2m
( ∂2
∂x2+∂2
∂y2+∂2
∂z2
)= − ~2
2m
( ∂2
∂r2+
1
r
∂
∂r+
1
r2
∂2
∂θ2+∂2
∂z2
).
This is now in cylindrical coordinates. The eigenvalue problem we want to solve is:
Hψ(r, θ, z) = Eψ(r, θ, z).
where E are the energy eigenvalues. The above expression is often termed the time
independent Schrodinger equation, as it uses the quantum Hamiltonian to extract en-
ergy values with no explicit time derivative.
Since the operators ∂/∂θ and ∂/∂z commute with the above form of H , we can
choose solutions of the form (Gozzi and Mauro 2002):
ψ(r, θ, z) =1
2πexp
[ip0zz
~
]exp[inθ]R(r), (148)
where p0z is a constant and n is an integer. Plugging in this wavefunction into the
time independent Schrodinger equation will give you solutions in the form of Bessel
Functions:
d2R(r)
dr2+
1
r
dR(r)
dr+(2mE
~2− p0
z2
~2− n2
r2
)R(r) = 0.
Bessel Functions are a class of functions that govern cylindrical harmonics.
If you turn on the magnetic potential, we will now have to solve
HAψ(r, θ, z)A = EAψ(r, θ, z)A.
In order to achieve this, we have the following minimal coupling rule in cylindrical
95
coordinates:
pθ = −i~ ∂∂θ⇒ pθ −
e
c
Φ
2π= −i~ ∂
∂θ− e
c
Φ
2π(149)
The quantum Hamiltonian therefore becomes:
HA = − ~2
2m
( ∂2
∂r2+
1
r
∂
∂r+
1
r2
( ∂∂θ− ie
chΦ)2
+∂2
∂z2
),
where Φ is the flux (constant) through an area perpendicular to the axis of the smaller
cylinder. This area will be larger than the smaller cylinder’s cross-section. Notice
that pθ is not present in the wavefunction, so the form of (148) will remain unaltered.
The energy values in the presence of the magnetic potential, as you will see, will be
different.
As before, substituting (148) into the magnetic potential version of the time inde-
pendent Schrodinger equation will give you the following Bessel function
d2R(r)
dr2+
1
r
dR(r)
dr+(2mE
~2− p0
z2
~2− (n− α)2
r2
)R(r) = 0,
where α = eΦBch
. If you compare the two Bessel functions, you can immediately see
that the presence of the magnetic potential shifts the function’s behavior. For the
quantum case, the behavior of the system is impacted by the presence of the vector
potential.
Now we will do the same procedure for L as the generator of motion (33). For
the free particle in 3 dimensions, we know that
L = −ipxm
∂
∂x− ipy
m
∂
∂y− ipz
m
∂
∂z,
based off of (33). Just as in the quantum case, we want this to be in cylindrical coor-
dinates. We will therefore need to convert the cartesian momenta px, py, and pz into
pr, pθ, and pz. The latter are the canonical momenta, which can be calculated directly
from the Lagrangian L.25 As before, any cartesian coordinates can be converted to25Do not confuse the Lagrangian L with the operator L. These are two different mathematical
entities.
96
cylindrical coordinates via (147), which we will make use of.
To start off, we write down the Lagrangian of motion. Recall from section 4.1,
the Lagrangian is just the kinetic energy minus the potential energy, L = T − V .
Since the potential is zero, the Lagrangian is just equal to the total kinetic energy, so
that
L = T =1
2m(x2 + y2 + z2), (150)
where we use the standard dot notation, so that any variable γ has a time derivative
γ = dγ/dt. We can take time derivatives of (147) to get:
x = r cos(θ)⇒ x = r cos(θ)− r sin(θ)θ,
y = r sin(θ)⇒ y = r sin(θ) + r cos(θ)θ,
with the z coordinate unaffected. Plugging these terms into (150), we get our La-
grangian in cylindrical coordinates:
L =1
2mr2 +
1
2mr2θ2 +
1
2mz2
Recalling from Lagrangian Dynamics that the canonical momentum associated with
coordinate σ is given by
pσ =∂L
∂qσ
Therefore applying the above formula for coordinates r, θ, and z, we have:
pr =∂L
∂r= mr
pθ =∂L
∂θ= mr2θ
pz =∂L
∂z= mz
97
If we apply the same rule to the cartesian Lagrangian (150) we get:
px =∂L
∂x= mx,
py =∂L
∂y= my,
with the z coordinate momentum unaltered. Using all the above information, we
can rewrite the cartesian momenta in terms of the cylindrical momenta, using the
momenta expressions above:
px = mx = m[r cos(θ)− r sin(θ)θ] = pr cos(θ)− 1
rpθ sin(θ),
py = m[r sin(θ) + r cos(θ)θ] = pr sin(θ) +1
rpθ cos(θ).
We therefore can rewrite the generator of motion L using the above information in
terms of cylindrical coordinates:
L = −ipxm
∂
∂x−ipy
m
∂
∂y−ipzm
∂
∂z= −ipr
m
∂
∂r−i pθmr2
∂
∂θ−ipzm
∂
∂z−i p
2θ
mr3
∂
∂pr. (151)
For the case with no magnetic potential, we will be solving the analogue to the time
independent Schrodinger equation,
Lψ = Eψ,
where ψ is the KvN classical wavefunction and E is the eigenvalue of the above
expression. Note that E is certainly not the same as the Energy like in the Schrodinger
equation. We observe that L commutes with−i ∂∂θ
,−i ∂∂z
, pθ, and pz. We can therefore
deduce that the wavefunction will have the form
ψ =1
2πR(r, pr)δ(pθ − p0
θ)δ(pz − p0z) exp[inθ] exp
[iλ0zz]. (152)
98
If we plug this wavefunction into (151), we will get (Gozzi and Mauro 2002):
(− i
mpr∂
∂r+p0θn
mr2− ip0
θ2
mr3
∂
∂pr+λ0zp
0z
m− E
)R(r, pr) = 0. (153)
We have a similar expression for the magnetic potential case:
LAψA = EAψA
Just as in the quantum case, we will utilize the minimal coupling rule (149). The
only difference is that the representation of the operator pθ is no longer going to be
a derivative with respect to θ, and instead we will have pθ = pθ for the KvNM case
(Gozzi and Mauro 2002). Making the substitution
pθ ⇒ pθ −e
c
Φ
2π, (154)
we will get the following two expressions:
L = −iprm
∂
∂r− i 1
mr2
(pθ −
e
c
Φ
2π
) ∂∂θ− ipz
m
∂
∂z− i 1
mr3
(pθ −
e
c
Φ
2π
)2 ∂
∂pr, (155)
ψ =1
2πR(r, pr)δ(pθ − p0
θ −eΦ
2πc)δ(pz − p0
z) exp[inθ] exp[iλ0zz]. (156)
Plugging (156) into (155) we will get:
(− i
mpr∂
∂r+p0θn
mr2− ip0
θ2
mr3
∂
∂pr+λ0zp
0z
m− EA
)R(r, pr) = 0 (157)
If you compare (157) (case with magnetic potential) to (153) (case without magnetic
potential) you will notice they happen to be the same equation. This shows us that
the same solutions will exist for classical mechanics regardless of the presence or
absence of the gauge potential. As demonstrated, the gauge potential will however
99
affect quantum systems (Gozzi and Mauro 2002; Mauro 2003).
On a philosophical level, the above analysis raises the question what do the gauge
potentials physically mean? Historically, it was simply seen as a mathematical tool
or convenience in the study of Electromagnetism. The electromagnetic field and its
potentials were fully equivalent descriptions of reality, but the potentials were only
seen as a mathematical construct with useful properties. This might be a historic
example of physicists not taking the mathematics of their theory seriously enough.
For example, Albert Einstein famously invented the theories of special relativity by
acknowledging that the speed of light c is the same in all noninertial reference frames.
This is a fact tacitly present in Maxwell’s equations, but physicists at the time did
not treat the mathematics seriously enough and thought the equations must only be
true relative to a hypothetical ‘fluid’ filling all of space and time, the aether, an idea
Einstein showed to be wrong. Aharonov and Bohm 1959 say in their paper “a further
interpretation of the potentials is needed in the quantum mechanics”, so perhaps this
is an area for further scientific investigation and discovery in the future.
7 Other Research
Wavefunctions in a complex Hilbert phase space can exactly model CM in the form of
statistical physics. Both QM and CM can be articulated in a common mathematical
language utilizing Hermitian operators to recover expectation values of observables.
In the pages above we covered a host of different applications of the KvN wave-
function inside classical physics. We hardly exhausted the possible applications of
KvNM, however. There are many areas of research we have not even touched upon.
Some recent examples of the widespread applicability of the KvNM formulism:
1. Why the usage of finite-dimensional Hilbert Spaces in QM and KvNM is often
ill-advised
A recent trends in QM research has been to try utilize QM with finite dimen-
sions instead of the regular infinite Hilbert space. This is done to aid in clas-
100
sical computer modeling of quantum systems and an interest in the possibility
of rooting QM in a finite instead of an infinite space. For instance, approxima-
tions of the quantum Hamiltonian tend to be finite matrices (Bondar, Cabrera,
and Rabitz 2013). Bondar, Cabrera, and Rabitz 2013 demonstrate that this will
unavoidably violate the Ehrenfest Theorems. This is a serious shortcoming of
finite dimensional treatments of QM, but there are a small number of ways that
the issues might be circumvented (Bondar, Cabrera, and Rabitz 2013).
2. KvNM has been used to study aspects of String Theory and the Gravitational
Principle of Equivalence
The KvN Path Integral is utilized in Carta and Mauro 2002 to study the prop-
erties of Classical Yang-Mills Theories (a subset of String Theories.) The Path
Integral provides geometrical tools like exterior derivatives, forms, etc. The
Classical Path Integral might shed light on the geometrical aspects of these
string theories (Carta and Mauro 2002). Sen, Dhasmana, and Silagadze 2020
investigates the validity of Einstein’s principle of mass equivalence for both
classical KvNM and quantum systems. Fisher information is discussed and
how it relates to insights from KvNM and QM. It is demonstrated that the weak
principle of equivalence holds in QM just as it does in KvNM (Sen, Dhasmana,
and Silagadze 2020).
3. ODM can be used to describe relativistic particles
The ODM paradigm is upheld in relativistic settings. The following two Ehrenfest-
like theorems were postulated (Cabrera et al. 2019) based on relativistic prin-
ciples
d
dt〈xk〉 = 〈cγ0γk〉.
d
dt〈pk〉 = 〈ce∂kAνγ0γν〉.
If the quantum commutator is chosen, you can follow a series of steps similar
101
to those in section 3 to derive the famous Dirac equation, which governs rel-
ativistic quantum particles (Cabrera et al. 2019). If the Koopman Algebra is
chosen and you constrain the system from making antiparticles, the same se-
ries of steps as deriving the KvNM will give you the classical Spohn equation
(Cabrera et al. 2019). This is quite a surprising result - ODM can study spin
1/2 relativistic particles.
4. ODM has been used to study Time’s Arrow and stochastic behavior from de-
terminism
A lot of processes are reversible in physics, but our intuitions of the world are
that it is irreversible and things tend towards decay. How does irreversibility
arise out of the reversible? McCaul and Bondar 2021 utilizes the KvN Path
Integral to formulate a functional model for how irreversibility arises out of
the reversible laws we have.
5. Reformulation of Electromagnetism
Rajagopal and Ghose 2016 provides a reformulation of electromagnetic theory
that draws inspiration from the KvN theory.
6. KvN wavefunctions were utilized to understand the interaction of classical with
quantum systems
The problem of modeling quantum and classical system interaction is an old
one. Utilizing the KvN classical wavefunction, however, had led to new models
of how the interaction occurs. Not only does it describe the classical system’s
effect on the quantum system, but uniquely it models how the quantum system
will impact the classical system in return (Bondar, Gay-Balmaz, and Tronci
2019). It also overcomes difficulty in some other models. For instance, the
quantum density matrix will always be positive-definite in this model (Bondar,
Gay-Balmaz, and Tronci 2019).
Among other areas of inquiry. Doubtlessly, many more areas of research that will
benefit from KvNM and a paradigm like ODM.102
It is important to have many “tools” in science when investigating novel phenom-
ena. Feynman famously said that the reason he had solved problems others could
not, is that his toolbox had a different set of tools than everyone else. KvNM is yet
an another novel tool for you, the reader, to have in your toolkit. KvNM has and will
continue to shed light on new and exciting phenomena in science.
103
References
Aharonov, Yakir and David Bohm (1959). “Significance of Electromagnetic Poten-
tials in the Quantum Theory”. In: Physical Review 115, p. 485. doi: https :
//doi:10.1103/PhysRev.115.485.
Bondar, Denys I., Renan Cabrera, and Herschel A. Rabitz (2013). “Conceptual incon-
sistencies in finite-dimensional quantum and classical mechanics.” In: Physical
Review A: Atomic, Molecular, and Optical Physics 88.1, p. 012116. doi: https:
//doi.org/10.1103/PhysRevA.88.012116.
Bondar, Denys I., Francois Gay-Balmaz, and Cesare Tronci (2019). “Koopman wave-
functions and classical-quantum correlation dynamics”. In: Proceedings of the
Royal Society A 475.2229, p. 20180879. doi: https://doi.org/10.1098/
rspa.2018.0879.
Bondar, Denys I. et al. (2012). “Operational Dynamic Modeling Transcending Quan-
tum and Classical Mechanics.” In: Physical Review Letters 109.19, p. 190403.
doi: https://doi.org/10.1103/PhysRevLett.109.190403.
Bondar, Denys I. et al. (2013). “Wigner phase space distribution as a wave function.”
In: Physical Review A: Atomic, Molecular, and Optical Physics 88.5, p. 052108.
doi: https://doi.org/10.1103/PhysRevA.88.052108.
Cabrera, Renan et al. (2019). “Operational Dynamical Modeling of spin 1/2 rela-
tivistic particles: the Dirac equation and its classical limit.” In: The European
Physical Journal Special Topics 227, 2195–2207. doi: https://doi.org/10.
1140/epjst/e2018-800075-7.
Carta, P. and D. Mauro (2002). “Functional Approach to Classical Yang-Mills The-
ories.” In: Nuclear Physics B: Proceedings Supplements 104.1-3, pp. 247–250.
doi: https://doi.org/10.1016/S0920-5632(01)01625-5.
Feynman, Richard (1982). “Simulating Physics with Computers”. In: International
Journal of Theoretical Physics 21, pp. 467–488.
104
Gozzi, E. and D. Mauro (2002). “Minimal Coupling in Koopman–von Neumann The-
ory”. In: Annals of Physics 296.2, pp. 152–186. doi: https://doi.org/10.
1006/aphy.2001.6206.
Koopman, Bernard O. (1931). “Hamiltonian Systems and Transformation in Hilbert
Space”. In: Proceedings of the National Academy of Sciences 17.5, pp. 315–318.
doi: https://doi.org/10.1073/pnas.17.5.315.
Lewis Jr., H. R. (1967). “Classical and Quantum Systems with Time-Dependent Harmonic-
Oscillator-Type Hamiltonians”. In: Physical Review Letters. 18.13, p. 510. doi:
https://doi.org/10.1063/1.1664991.
Lewis Jr., H. R. and W. B. Riesenfeld (1968). “An Exact Quantum Theory of the
Time-Dependent Harmonic Oscillator and of a Charged Particle in a Time-Dependent
Electromagnetic Field”. In: Journal of Mathematical Physics 10, p. 1458. doi:
https://doi.org/10.1063/1.1664991.
Mauro, D. (2002). “On Koopman-von Neumann Waves”. In: International Journal
of Modern Physics A 17.09, pp. 1301–1325. doi: https://doi.org/10.1142/
S0217751X02009680.
— (2003). “Topics in Koopman-von Neumann Theory”. In: Ph.D. Thesis. doi: https:
//arxiv.org/abs/quant-ph/0301172.
McCaul, Gerard and Denys I. Bondar (2021). “How to win friends and influence
functionals: deducing stochasticity from deterministic dynamics.” In: The Euro-
pean Physical Journal Special Topics 230, 733–754. doi: https://doi.org/
10.1140/epjs/s11734-021-00068-2.
MIT OpenCourseWare (2006). THE DENSITY MATRIX. url: https://ocw.mit.
edu/courses/chemistry/5-74-introductory-quantum-mechanics-ii-
spring-2009/lecture-notes/MIT5_74s09_lec12.pdf.
Rajagopal, A.K. and P. Ghose (2016). “Hilbert space theory of classical electrody-
namics”. In: Pramana 86, p. 1161. doi: https://doi.org/10.1007.
105
Ramos-Prieto, Iran et al. (2018). “KvN mechanics approach to the time-dependent
frequency harmonic oscillator”. In: Scientific Reports 8.1, p. 8401. doi: https:
//doi.org/10.1038/s41598-018-26759-w.
Sen, Abhijit, Shailesh Dhasmana, and Zurab K. Silagadze (2020). “Free fall in KvN
mechanics and Einstein’s principle of equivalence.” In: Annals of Physics 422,
p. 168302. doi: https://doi.org/10.1016/j.aop.2020.168302.
Shankar, Ramamurti (1988). Principles of Quantum Mechanics. New York and Lon-
don: Plenum Press.
Tipler, Frank J. (2010). “Hamilton-Jacobi Many-Worlds Theory and the Heisenberg
Uncertainty Principle”. In: ArXiv. doi: https://arxiv.org/abs/1007.4566.
von Neumann, John (1932). “Zur operatorenmethode in der Klassi sehen Mechanik”.
In: Ann. Math 33, p. 587.
Wilkie, J. and P. Brumer (1997). “Quantum-classical correspondence via Liouville
dynamics. II. Correspondence for chaotic Hamiltonian systems.” In: Physical Re-
view A 55.43. doi: https://doi.org/10.1103/PhysRevA.55.43.
Wolfram Alpha (2021). url: https://mathworld.wolfram.com/GaussianFunction.
html.
106
Appendices
A Useful Mathematics for Hilbert Space Physics
Note: A lot of the following useful mathematics was found in the Appendix of Bon-
dar et al. 2012.
Theorem I: If you have a function of operator A, i.e. f(A), then the following
eigenvalue expressions are true:
f(A) |A〉 = f(A) |A〉
〈A| f(A) = 〈A| f(A)
This should purposefully resemble the prior eigenvalue problem you are familiar
with, A |A〉 = A |A〉. This theorem is true for multivariable operators, for instance,
A(q, p) |q, p〉 = A(q, p) |q, p〉
.
Theorem II (Resolution of Identity): The following theorem says that
I =
ˆdA |A〉 〈A|
where I is the identity operator, functionally equivalent to multiplying an object by
1. Since it is just equal to 1, you can introduce the Resolution of Identity in a variety
of calculations. As always A is an arbitrary placeholder variable, and you can make
it any variable of interest.
Theorem III (Cauchy-Schwartz Inequality): It simply states that the following is al-
107
ways true for complex vectors in Hilbert Space:
〈C|C〉 〈D|D〉 ≥ | 〈C|D〉 |2
where C and D are generic placeholder variables.
Theorem IV : If the commutator relationship between two generic operators A and
B is [A, B] = iκ, then the following statements hold true:
〈A| B |ψ〉 = −iκ ∂
∂A〈A|ψ〉
〈B| A |ψ〉 = iκ∂
∂B〈B|ψ〉
Recall that 〈B|ψ〉 = ψ(B) and 〈A|ψ〉 = ψ(A).
Theorem V : If the commutator relationship between two generic operators A and
B is [A, B] = iκ, then the following statements hold true:
〈A|B〉 =1√2πκ
eiAB/κ (158)
This is not difficult to demonstrate. First, take Theorem IV and realize that |ψ〉 is a
general vector that can be substituted for a more specific vector. This is a common
practice in QM. Let us say that |ψ〉 = |B〉. Then by Theorem IV we have:
〈A| B |B〉 = B 〈A|B〉 = −iκ ∂
∂A〈A|B〉
You can solve this common differential equation for the function 〈A|B〉 to get:
〈A|B〉 = CeiAB/κ
where C is a constant of integration. C can be found in the following way: Start with
equation (18)
108
〈A|A′〉 = δ(A− A′)
and use Theorem II with A = B 26 to get:
〈A|A′〉 = δ(A− A′) = 〈A| I |A′〉 =
ˆdB 〈A|B〉 〈B|A′〉
Since we know what 〈A|B〉 is, we plug it in to get:
〈A|A′〉 = δ(A− A′) =
ˆdB[CeiAB/κ][Ce−iA
′B/κ] = C2
ˆdBei(A−A
′)B/κ
From equation (22) of section 2, we can easily see that
ˆdBei(A−A
′)B/κ = 2πκδ(A− A′)
Ergo,
δ(A− A′) = C22πκδ(A− A′)
C22πκ = 1
C2 =1
2πκ
C =1√2πκ
We now know C and have successfully demonstrated equation (158).
Theorem VI: As strange as it sounds, the mathematics of Hilbert Space defines deriva-
tives with respect to operators. If you have [Ak, B] = Constant, then a function of
operators f(A1, A2, ..., An) will obey the following rule:
[f(A1, A2, ..., An), B] =n∑k=1
[Ak, B]∂f(A1, A2, ..., An)
∂Ak
26In other words, sinceA is just a generic placeholder for Theorem II, we will use I =´dB |B〉 〈B|
109
Notice the partial derivative with respect to an operator. The next theorem shows how
to change this odd operator derivative to regular derivatives that we know and love.
Theorem VII The following is the relationship between derivatives with respect to
operators and the regular derivatives we are most generally familiar with:
∂f(A1, A2, ..., An)
∂Ak=
1
(2π)n
ˆ n∏l=1
dεl dµl exp[in∑q=1
µq(Aq − εq)]∂f(ε1, ε2, ..., εn)
∂εk
where on the left you can see the derivative with respect to an operator and on the
right you can see the derivative with respect to a scalar variable.
Theorem VIII: An integral of a Gaussian function has the following form:
ˆ ∞−∞
e−ax2+bx+cdx =
√π
aeb2
4a+c
where a, b, and c can be real or imaginary constants.
B The Principle of Causality and Stone’s Theorem
A unitary operator U(t2, t1) always has the following properties:
1. Group product: U(t3, t2)U(t2, t1) = U(t3, t1)
2. Reversibility: U(t2, t1)† = U(t2, t1)−1 = U(t1, t2)
3. Identity: U(t1, t1) = 1
The job of the unitary operator U(t2, t1) is to bring you from t1 to t2. The group
product encapsulates the concept of cause and effect, as you move from cause to
effect through time. Causality is central to the philosophy of science, and we should
not be surprised that it pops up inside the mathematics of physics.
The three properties of the unitary operator above imply Stone’s Theorem:
110
If the three properties hold true, then the following relationship is also true:
i∂
∂tU(t2, t1) = GU(t2, t1)
where G is a Hermitian operator. If we assume that the wavefunction ket evolves via
the unitary operator
|ψ(t2)〉 = U(t2, t1) |ψ(t1)〉 ,
then we recover the Schrodinger-like equation we spoke of in earlier sections:
i∂
∂tU(t2, t1) |ψ(t1)〉 = GU(t2, t1) |ψ(t1)〉 ⇒ i
∂
∂t|ψ(t2)〉 = G |ψ(t2)〉
This is the reason why Stone’s Theorem becomes an important ingredient in section
3.
C Exponated Operators and the Propogator
Although it might seem odd at first, exponated operators are allowed within Quantum
Mechanics. The Maclaurin series for the f(x) = ex is given by:
ex =∞∑n=0
xn
n!= 1 +
1
1!x+
1
2!x2 +
1
3!x3 +
1
4!x4 + · · ·
Let us say that you have a generic operator A, like a square matrix. As long as An is
defined for any real number n, then you can say:
eA =∞∑n=0
An
n!= 1 +
1
1!A+
1
2!A2 +
1
3!A3 +
1
4!A4 + · · ·
The Maclaurin series defines exponated operators as an infinite sum.
If we have an expression such as eAt where A is time-independent, we can also
derive a derivative rule for it:
111
eAt =∞∑n=0
(At)n
n!= 1 +
1
1!At+
1
2!(At)2 +
1
3!(At)3 +
1
4!(At)4 + · · ·
d
dteAt =
d
dt1 +
d
dt
1
1!At+
d
dt
1
2!(At)2 +
d
dt
1
3!(At)3 + · · · =
=1
1!A+
1
2!A2(2t)+
1
3!A3(3t2)+
1
4!A4(4t3)+··· = A
[1+
1
2!A(2t)+
1
3!A2(3t2)+
1
4!A3(4t3)+···
]=
= A[1+
1
2 · 1A(2t)+
1
3 · 2 · 1A2(3t2)+···
]= A
[1+
1
1!At+
1
2!(At)2+
1
3!(At)3+···
]= AeAt
∴d
dteAt = AeAt (159)
The above relation is true even if a real or complex constant is placed in front of the
exponated term At. You can redo the exercise to convince yourself of this. We will
use this important fact in order to demonstrate the important relationship (52) is fully
equivalent to the Schrodinger equation:
|ψ(t)〉 = e−i~ tH |ψ(0)〉
i~∂
∂t|ψ(t)〉 = i~
∂
∂te−
i~ tH |ψ(0)〉 = i~
−i~He−
i~ tH |ψ(0)〉 = H |ψ(t)〉
∴ i~∂
∂t|ψ(t)〉 = H |ψ(t)〉
A similar derivation can be done for (66), to demonstrate that it is fully equivalent to
Stone’s Theorem (Appendix B).
112
113
Figure 1: This is the set up of the double slit experiment. Middle wall is at y = yMand rightmost wall is at y = yR. Adapted from Mauro 2002
Figure 2: a. Left: Graph of KvN probability distribution. b. Right: Graph of QMprobability distribution. Adapted from Mauro 2002.
Figure 3: a. Left: Example graph of a single variable gaussian curve. b. Right:Example graph of a double variable gaussian curve. Adapted from Wolfram Alpha2021.
114
Figure 4: Aharanov-Bohm Experiment set up. Reproduced from Aharonov andBohm 1959.
115