E.

WOLF, PROGRESS IN OPTICS XXXVI

0 1996 ELSEVIER SCIENCE B.V
ALL RIGHTS RESERVED

PHOTON WAVE FUNCTION

BY

IWO BIALYNICKI-BIRULA

Center for Theoretical Physics, Polish Academy of Sciences,

Al. Lotnikdw 32. 02-668 Warsaw, Poland
AND
Rochester Theory Center for Optical Science and Engineering,
University of Rochester, Rochester, NY 14627, USA

245
 CONTENTS

PAGE

4 1. INTRODUCTION . . . . . . . . . . . . . . . . . . . 248

§ 2. WAVE EQUATION FOR PHOTONS . . . . . . . . . . . 254

§ 3. PHOTON WAVE FUNCTION IN COORDlNATE

REPRESENTATION . . . . . . . . . . . . . . . . . . 259

§ 4. PHOTON WAVE FUNCTION IN MOMENTUM

REPRESENTATION . . . . . . . . . . . . . . . . . . 263

§ 5. PROBABILISTIC INTERPRETATION . . . . . . . . . . 267

§ 6. EIGENVALUE PROBLEMS FOR THE PHOTON WAVE

FUNCTION . . . . . . . . . . . . . . . . . . . . . 273

9 7. RELATIVISTIC INVARIANCE OF PHOTON WAVE

MECHANICS . . . . . . . . . . . . . . . . . . . . . 278

§ 8. LOCALIZABILITY OF PHOTONS . . . . . . . . . . . 279

§ 9. PHASE-SPACE DESCRIPTION OF A PHOTON . . . . . . 281

4 10. HYDRODYNAMIC FORMULATION . . . . . . . . . . 283

tj 11. PHOTON WAVE FUNCTION IN NON-CARTESIAN

COORDINATE SYSTEMS AND IN CURVED SPACE . . . . 285

4 12. PHOTON WAVE FUNCTION AS A SPINUR FIELD . . . . 286

9 13. PHOTON WAVE FUNCTIONS AND MODE EXPANSION OF

THE ELECTROMAGNETIC FIELD . . . . . . . . . . . 288

246
$ 1 4. SUMMARY . . . . . . . . . . . . . . . . . . . . . 290
ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . 291
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . 292

241
 0 1. Introduction
The photon wave function is a controversial concept. The controversies stem
from the fact that photon wave functions cannot have all the properties of the
Schrodinger wave functions of nonrelativistic wave mechanics. Insistence on
those properties which, owing to the peculiarities of photon dynamics, cannot
be rendered, led some physicists to the extreme opinion that the photon wave
function does not exist. I reject such a fundamentalist point of view in favor of
a more pragmatic approach. In my view, the photon wave function exists as long
as it can be defined precisely and made useful. Many authors whose papers are
quoted in this review share the same opinion and had no reservations about using
the name ‘photon wave function’ when referring to a complex vector-function
of space coordinates r and time t that adequately describes the quantum state of
a single photon.
The notion of the photon wave function is certainly not new, but strangely
enough it has never been explored systematically and fully. Some textbooks on
quantum mechanics start the introduction to quantum theory with a discussion
of photon polarization measurements (cf., for example, Dirac [1958], Baym
[1969], Lipkin [1973], Cohen-Tannoudji, Diu and Laloe [1977]), but in all
these expositions a complete photon wave function never takes on a specific
mathematical form. Even Dirac, who wrote: “The essential point is the
association of each of the translational states of the photon with one of the wave
functions of ordinary wave optics”, never expresses this association in an explicit
form. In this context he also uses the now famous phrase: “Each photon interferes
only with itself”, which implies the existence of photon wave functions whose
superposition leads to interference phenomena.
In the textbook analysis of polarization, only simple prototype two-component
wave functions are used to describe various polarization states of the photon,
and with their help the preparation and the measurement of polarization is
analyzed. However, it is not explained why a wave function should not be used
to describe also the “translational states of the photon” mentioned by Dirac.
After such a heuristic introduction to quantum theory, the authors go on to the
study of massive particles, and if they ever return to quantum theory of photons,
it is always within the formalism of second quantization with creation and

248
v,g 11 INTRODUCTION 249

annihilation operators. In some textbooks (cf., for example, Bohm [ 19541, Power
[1964]), one may even find statements which negate completely the possibility
of introducing a wave function for the photon.
A study of the photon wave function should be preceded by an explanation of
what a photon is, and why a description of the photon in terms of a wave function
must exist. According to modern quantum field theory, photons, together with
all other particles (and also quasiparticles, phonons, excitons, plasmons, etc.),
are the quantum excitations of a field. In the case of photons, these are the
excitations of the electromagnetic field. The lowest field excitation of a gioen
type corresponds to one photon and higher field excitations involve more than
one photon. This concept of a photon (called the “modern photon” in a tutorial
review by Kidd, Ardini and Anton [1989]) enables one to use the photon wave
function to describe not only quantum states of an excitation of the free field, but
also of the electromagnetic field interacting with a medium. Conceptually, the
difference between free space and a medium is not essential since the physical
vacuum is like a polarizable medium. It is filled with all the virtual pairs -
zero point excitations of charged quantum fields. Therefore, even in free space,
photons can be also viewed as the excitations of the vacuum made mostly of
virtual electron-positron pairs (Bialynicki-Birula [ 19631, Bjorken [ 19631).
Even though, in principle, all particles can be treated as field excitations,
photons are much different from massive particles. They are also different from
massless neutrinos since the photon number does not obey a conservation law.
There are problems with the photon localization, and as a result the position
operator for the photon is ill-defined, but the similarities between photons and
other quantum particles are so ample that the introduction of the photon wave
function seems to be fully justified and even necessary in order to achieve a
complete unification of our description of all particles.

I , I , COORDINATE VS. MOMENTUM REPRESENTATION

In nonrelativistic quantum mechanics, the term coordinate representation is used

to denote the representation in which the wave function +(r) is defined as a
projection of the state vector I+) on the eigenstates Ir) of the components i,$,
and i of the position operator i ,

The wave function in coordinate representation, therefore, becomes automatically

a function of the eigenvalues of the position operator i . The position operators
250 PHOTON WAVE FUNCTION [v, 9 1

act on the wave function simply through a multiplication. In quantum mechanics

of photons, this approach does not work due to difficulties with the definition
of the photon position operator (cf. 5 8). One may still, however, introduce
functions of the coordinate vector r to describe quantum states of the photon.
By adopting this less stringent point of view that does not tie the wave function
in coordinate representation with eq. (l.l), one avoids the consequences of
the nonexistence of the photon position operator i . In principle, any function
of r that adequately describes photon states may be called a photon wave
function in coordinate representation, and it is a matter of taste and convenience
regarding which one to use. It should be pointed out that in a relativistic quantum
theory, even for particles with nonvanishing rest mass, the position operator
and the localization associated with it do not live up to our nonrelativistic
expectations. The differences in localization of photons and, say electrons, are
more quantitative than qualitative since they amount to the “spilling of the wave
function” beyond the localization region governed by a power law versus an
exponential decay.
The photon wave function in momentum representation has not stirred any
controversy since the photon momentum operatorp is well defined. Its existence,
as the generator of translations, follows directly from the general theory of
representations of the Poincare group developed by Wigner [1939]. It has
always been taken for granted by all physicists working in relativistic quantum
electrodynamics that the notion of the photon wave function in momentum
representation is well founded. Such wave functions describing initial and final
states of photons appear in all formulas for transition amplitudes in the S-matrix
theory of scattering phenomena (cf., for example, Schweber [ 19611, Akhiezer
and Berestetskii [ 19651, Bialynicki-Birula and Bialynicka-Birula [ 19751, Cohen-
Tannoudji, Dupont-Roc and Grynberg [ 19891). Thus, one may safely assert that
the photon wave function in momentum representation is a well-defined and fully
established object.

1.2. PHASE REPRESENTATION

The photon wave functions discussed in this review are distinct from the one-
mode wave functions which have been introduced in the past (London [1927],
Bialynicki-Birula and Bialynicka-Birula [ 19761, Pegg and Barnett [ 19881) to
describe multi-photon states. These functions depend on the phase cp of the field
and were called the wave functions in the phase representation by Bialynicki-
Birula and Bialynicka-Birula [ 19761. The wave functions Y( cp) characterize
quantum states of a selected mode of the quantized electromagnetic field and,
v, 9: 11 INTRODUCTION 25 1

in general, they describe a superposition of states with different numbers of

photons. All spatial characteristics of these states are contained in the mode
function that defines the selected mode of the electromagnetic field. One-mode
wave functions Y ( q )describe properties of multi-photon states of the quantized
electromagnetic field with all photons being in the same quantum-mechanical
state. In contrast, the photon wave function in the coordinate representation can
be identified with the mode function itself (cf. 5 13). It describes a state of a
single photon and not a state of the quantized field.

1.3. LANDAU-PEIERLS WAVE FUNCTION

The concept of the photon wave function in coordinate representation was

introduced for the first time by Landau and Peierls [ 19301. The same function has
been independently rediscovered more recently by Cook [ 1982a,b], and Inagaki
[ 19941. The Landau-Peierls proposal has not been met with great enthusiasm,
since their wave function is a highly nonlocal object.
The nonlocality of the Landau-Peierls wave function is introduced by
operating on the local electromagnetic field with the integral operator (-A)-’l4:

This integral operator corresponds to a division by Jl‘cl of the Fourier transform,

and it changes the dimension of the wave function from LP2, characteristic
of the electromagnetic field, to L-3/2. Therefore the modulus squared of the
Landau-Peierls wave function has the right dimensionality to be interpreted as
a probability density to find a photon. In particular, these wave functions can be
normalized to one with the standard definition of the norm since the integral
of the modulus squared of the wave function is dimensionless. However, as
has been already noted by Pauli [1933], despite its right dimensionality, the
nonlocal wave function has serious drawbacks. First, it does not transform under
Lorentz transformations as a tensor field or any other geometric object. Second,
a nonlocal wave function taken at a point in one coordinate system depends on
the values of this wave function in all of space in another coordinate system.
Third, the probability density defined with the use of a nonlocal wave function
does not correspond to the probability of interaction of localized charges with
the electromagnetic field. The vanishing of the wave function at a definite point,
in Pauli’s words (Pauli [1933]), has ‘<nodirect physical significance” because the
electromagnetic field does act on charges at the points where the probability of
252 PHOTON WAVE FUNCTION [Y § I

finding a photon vanishes. The Landau-Peierls wave functions can not be used
as primary objects in the presence of a medium since one is unable to impose
proper boundary conditions on such nonlocal objects. These functions can be
introduced, if one wishes to do so, as secondary objects related to the local wave
function by a nonlocal transformation (9 5). Scalar products and expectation
values look simpler when they are expressed in terms of the Landau-Peierls
wave hnctions, but that is, perhaps, their only advantage.

1.4. RIEMANN-SILBERSTEIN WAVE FUNCTION

The mathematical object that fully deserves the name of the photon wave
function can be traced back to a complex version of Maxwell's equations that
was already known at the turn of the century. The earliest reference is the
second volume of the lecture notes on the differential equations of mathematical
physics by Riemann, which were edited and published by Weber [1901]. Various
applications of this form of Maxwell's equations were given by Silberstein
[1907a,b, 19141 and Bateman [1915] in the framework of classical physics.
The complex form of Maxwell's equations is obtained by multiplying the first
pair of these equations,

a,D(r, t ) = V x H(r, t), V.D(r,t ) = 0, (1.3)

by the imaginary unit and then by subtracting from it the second pair

d,B(r,t ) = -V x E(r, t), V . B ( r ,t ) = 0. (1.4)

In the SI units that are used here, the vectors D and B have different dimensions
and prior to subtraction one must equalize the dimensions of both terms. The
resulting equations in empty space are:

idtF(r,t) = cV x F(r,t),
V . F ( r ,t ) = 0,

where

and c = I/-. Around the year 1930, Majorana (unpublished notes quoted
by Mignani, Recami and Baldo [1974]) arrived at the same complex vector
Y § 11 INTRODUCTION 253

exploring the analogy between the Dirac equation and the Maxwell equations.
Kramers [1938] made extensive use of this vector in his treatment of quantum
radiation theory. This vector is also a natural object to use in the quaternionic
formulation of Maxwell's theory (Silberstein [ 19141). With the advent of spinor
calculus that superseded the quaternionic calculus, the transformation properties
of the Riemanr-Silberstein vector have become even more transparent. When
Maxwell's equations were cast into the spinor form (Laporte and Uhlenbeck
[ 193 11, Oppenheimer [193 l]), this vector turned into a symmetric second-rank
spinor. The use of the Riemann-Silberstein vector as the wave function of
the photon has been advocated by Oppenheimer [1931], Moli&re [1949], Good
[ 19571, Bialynicki-Birula [ 19941, Sipe [ 19951, and Bialynicki-Birula [1996a].
It was already noticed by Silberstein [ 1907al that the important dynamical
quantities associated with the electromagnetic field - the energy density and
the Poynting vector - can be represented as bilinear expressions built from the
complex vector F . Using modern terminology, one would say that the formulas
for the energy E , momentum P, angular momentum M , and the moment of
energy N of the electromagnetic field look like quantum-mechanical expectation
values; i.e.,

E =
1 d3rF*.F, (1.8)

P = -
21c 's d3rF*xF, (1.9)

M = -
21c 'S d3rr x (F*xF), (1.10)

(1.11)

evaluated in a state described by the wave function F. All these quantities

are invariant under the multiplication of F by a phase factor exp(ia). Such
a multiplication results in the so called duality rotation (Misner and Wheeler
[1957]) of the field vectors
D' - cosaD sinaB
(1.12)
& & & '
B' cosaB
-
+-s i n a D (1.13)
& fi A .
The coupling of the electromagnetic field with charges fixes the phase a but for
a free photon one has the same complete freedom in choosing the overall phase
of the photon wave function as in standard wave mechanics of massive particles.
254 PHOTON WAVE FUNCTION [v, 5 2

The Riemann-Silberstein vector also has many other properties which one
would associate with a one-photon wave function, except for a somewhat
modified probabilistic interpretation. Insistence on exactly the same form of the
expressions for transition probabilities as in nonrelativistic wave mechanics leads
back to the Landau-Peierls wave function with its highly nonlocal transformation
properties.

5 2. Wave Equation for Photons

The wave equation for the photon is taken to be the complexified form [eq. (1.5)]
of Maxwell’s equations. In order to justify this choice, one may show (cf. 5 4)
that the Fourier decomposition of the solutions of this wave equation leads to the
same photon wave functions in momentum representation that can be introduced
without any reference to a wave equation directly from the general theory of
representations of the Poincart group (Bargmann and Wigner [ 19481, Lomont
and Moses [1962]). There is also a heuristic argument indicating that eq. (1.5)
is the right choice. Namely, the wave equation (1.5) can be written in the same
form as the Weyl equation for the neutrino wave function. As a matter of fact,
all wave equations for massless particles with arbitrary spin can be cast into the
same form (cf. 5 12).

2.1. WAVE EQUATION FOR PHOTONS IN FREE SPACE

In order to see a correspondence between Maxwell’s equations and quantum-

mechanical wave equations, one may follow Oppenheimer [ 193 13 and Molikre
[1949] and rewrite eq. (1.5) with the use of the spin-1 matrices s,,sy,sz well
known from quantum mechanics (see, for example, Schiff [ 19681). The matrices
which will be used here are in a different representation from the one usually
used in quantum mechanics, since they act on the Cartesian vector components
of the wave function and not on the components labeled by the eigenvalues of
s,. That is the reason why the matrix s, is not diagonal:

s,=
00 0
[oo-i],
Oi 0
sy= [ OOi
0 001,
-i 0 0
s,= [ 0 -i 0
0i 0 0 1 . (2.1)

Equation (1.5) can be written in terms of these spin matrices if the following
conversion rule from vector notation to matrix notation is applied:
Y § 21 WAVE EQUATION FOR PHOTONS 255

The resulting equation,

is of a Schrodinger type but with a different Hamiltonian. The divergence

condition (1.6) can also be expressed in terms of spin matrices, either as

or equivalently (Pryce [ 19481) as

The form (2.3) of the Maxwell equations compares directly with the Weyl
equation for neutrinos (Weyl [ 19291):

Equation (2.6) differs from eq. (2.3) only in having the Pauli matrices,
appropriate for spin-; particles, instead of the spin-1 matrices which are
appropriate for photons. Of course, one may cancel the factors of h appearing
on both sides of eqs. (2.3) and (2.6), but their presence makes the connection
with quantum mechanics more transparent.
Some authors (Oppenheimer [193 11, Ohmura [1956], Moses [ 19591) intro-
duced a different, although equivalent, form of the photon wave equation for a
four-component wave function. The inclusion of the fourth component enables
one to incorporate the divergence condition in a natural way. This approach is
related directly to the spinorial representation of the photon wave function, and
will be discussed in fj12.
In quantum mechanics the stationary solutions of the wave equation play a
distinguished role. They are the building blocks from which all solutions can be
constructed. Stationary solutions of the wave equation are obtained by separating
the time variable and solving the resulting eigenvalue problem. The eigenvalue
equation resulting from the photon wave equation (2.3) is:

fl
c(s.?V)F(r) = hwF(r). (2.7)

Assuming that the photon energy hw is positive, one reads from eq. (2.7) that
the projection of the spin on the direction of momentum (helicity) is positive.
256 PHOTON WAVE FUNCTION [v, § 2

It can be easily checked that one can reverse this sign by changing i into -i
in the definition (1.7) of the Riemann-Silberstein vector. Thus, the choice of
sign in the definition of this vector is equivalent to choosing positive or negative
helicity, corresponding to left-handed or right-handed circular polarization. In
order to account for both helicity states of the photons, one must consider both
vectors; one may call them F, and F-. This doubling of the vectors F had already
been considered by Silberstein [I9141 in the context of the classical Maxwell
equations.
Of course, if one is interested only in translating the Maxwell equations into
a complex form, one can restrict oneself to either F+or F-. In both cases, one
obtains a one-to-one correspondence between the real field vectors D and B
and their complex combination F+.However, to have a bona fide photon wave
function one must be able to superpose different helicity states without changing
the sign of the energy (frequency). This can be done only when both helicities
are described by different components of the same waue function as in the theory
of spin-i particles. One can see even more clearly the need to use both complex
combinations when one deals with the propagation of photons in a medium.

2.2. WAVE EQUATION FOR PHOTONS IN A MEDIUM

In free space, the two vectors F* satisfy two separate wave equations:

iatF+(r, t ) = &cV x F + ( r , t). (2.8)

In a homogeneous medium, using the values of E and p for the medium in the
definition of the vectors F + ,

one also obtains two separate wave equations. The new vectors F* are linear
combinations of the free-space vectors F;,

F+ = !.2 [(E+ E) Fy + (E E) Fo] , (2.10)

[(&E)F:+(E+E)Fo].
-

F - = 2l (2.1 1)

Thus, the positive and negative helicity states in a medium are certain linear
superpositions of such states in free space. The necessity to form linear
v, 0 21 WAVE EQUATION FOR PHOTONS 257

superpositions of both helicity states shows up even more forcefully in an

inhomogeneous medium, because then it is not possible to split the wave
equations into two independent sets. For a linear, time-independent, isotropic
medium, characterized by space-dependent permittivity and permeability, one
obtains the following coupled set of wave equations:

idtF+(r,t ) = u(r)(V x F+(r,t ) (2.12)

1 1
- -F+(r, t ) x Vo(r)- -F-(r, t ) x Vh(r)),
24r) 2h(r)

where F*(r,t) are built with the values of ~ ( r and ) p ( r ) in the medium,
u(r) = l / d m ) is the value of the speed of light in the medium, and
h(r) = d m
is the “resistance of the medium” (the sole justification for the
use of this name is the right dimensionality of Ohm). The divergence condition
(1.6) in an inhomogeneous medium takes on the form

1 1
V . F + ( r t, ) = -F+(r, t).Vu(r)+ -F-(r, t).Vh(r), (2.14)
2uW 2h(r)
1 1
V . F - ( r ,t ) = ---F-(r, t).Vo(r)+ -F+(r, t).Vh(r). (2.15)
24r) 2hW
The quantum-mechanical forms of eqs. (2.12) and (2.13) are:

- ih- (s.Vh(r))F-(r, t), (2.16)

2h(r)

+ ih- 4 - 1 (s.Vh(r))F+(r,t). (2.17)

2h(r)
In view of the coupling in the evolution equations (2.16) and (2.17) between
the vectors F+ and F-, one must combine them to form one wave function, 3,
with six components:

3= .I;[ (2.18)
258 PHOTON WAVE FUNCTION [Y fi 2

The wave equation for this function can be written in a compact form,

(2.19)

where the spin matrices si operate separately on upper and lower components,

(2.20)

and the three Pauli matrices pi act on 3 as follows:

(2.21)

The divergence conditions (2.14) and (2.15) can also be written in this compact
form:

(2.22)

One would not be able to write a linear wave equation for an inhomogeneous
medium in terms of just one three-dimensional complex vector, without doubling
the number of components. Note that the speed of light u may vary without
causing necessarily the mixing of helicities. This happens, for example, in the
gravitational field (cf. 9 11). It is only the space-dependent resistance h(r) that
causes mixing.
It is worth stressing that the study of the propagation of photons in an
inhomogeneous medium separates clearly local wave functions from nonlocal
ones. In free space there are many wave functions satisfying the same set
of equations. For example, differentiations of the wave functions or integral
operations of the type (1.2) do not change the form of these equations. The
essential difference between various wave functions shows up forcefully in the
study of the wave equation in an inhomogeneous medium, or in curved space
(9 11). All previous studies, except Bialynicki-Birula [1994], were restricted to
propagation in free space, and this very important point was missed completely.
The photon wave equations in an inhomogeneous medium are not very simple
but that is due, perhaps, to a phenomenological character of macroscopic
electrodynamics. The propagation of a photon in a medium is a succession
Y I 31 COORDINATE REPRESENTATION 259

of absorptions and subsequent emissions of the photon by the charges which

form the medium. The number of photons of a given helicity is, in general, not
conserved in these processes, and that accounts for all the complications. The
photon wave equations in an inhomogeneous medium describe in actual fact a
propagation of some collective excitations of the whole system and not just of
free photons.

2.3. ANALOGY WITH THE DIRAC EQUATION

The analogy with the relativistic electron theory, mentioned in 4 1, becomes the
closest when the photon wave equation is compared with the Dirac equation
written in the chiral representation of the Dirac matrices. In this representation
the bispinor is made of two relativistic spinors

(2.23)

and the Dirac equation may be viewed as two Weyl equations coupled by the
mass term:

(2.24)

(2.25)

These equations are analogous to eqs. (2.16) and (2.17) for the photon wave
function. For photons, the role of the mass term is played by the inhomogeneity
of the medium.

0 3. Photon Wave Function in Coordinate Representation

Despite a formal similarity between the wave equations for the photon [eqs.
(2.16) and (2.17)] and for the electron [eqs. (2.24) and (2.25)], there is an
important difference. Photons, unlike electrons, do not have antiparticles and
this fact influences the form of solutions of the wave equation and their
interpretation.
260 PHOTON WAVE FUNCTION tv, P 3
3.1. PHOTONS HAVE NO ANTIPARTICLES

Elementary, plane-wave solutions of relativistic wave equations in free space are

of two types: they have positive or negative frequency,

exp(-iot + k.r), or exp(iot - k-r). (3.1)

According to relativistic quantum mechanics, solutions with positive frequency

correspond to particles, and solutions with negative frequency correspond to
antiparticles. Thus, positive and negative frequency parts of the same solution
of the wave equation describe two different physical entities: particle and
antiparticle.
Photons do not have antiparticles, or to put it differently, antiphotons are
identical with photons. Hence, the information carried by the negative frequency
solutions must be the same as the information already contained in the positive
frequency solution. Therefore, one may disregard completely the negative
frequency part as redundant. An alternative method is to keep also the negative
frequency part, but to impose an additional condition on the solutions of the
wave equation. This condition states that the operation of particle-antiparticle
conjugation,

leaves the function 3 invariant:

q r , t ) = F(r,t). (3.3)

This condition is compatible with the evolution equation (2.19), and it eliminates
the unwanted degrees of freedom (cf. Bialynicki-Birula [ 19941). Equation (3.3)
is satisfied automatically if the wave function is constructed according to the
definition (2.18). This follows from the fact that F+ and F _ are complex
conjugate to each other. The information carried by the six-component function
7 satisfying the condition (3.3) is contained in its positive energy part, and is
the same as that camed by the initial Riemann-Silberstein vector F. It follows
from eq. (3.3) that the negative frequency part f l - 1 can always be obtained by
complex conjugation and by an interchange of the upper and lower components
of the positive frequency part 3(+),
v, D 31 COORDINATE REPRESENTATION 26 1

In this review, I shall use the symbol Y to denote the properly normalized,
positive energy (positive frequency) part of the function F:

t).
Y ( r ,t ) = 3(+)(r, (3.5)

This is the true photon wave function. Proper normalization of the photon wave
function is essential for its probabilistic interpretation and is discussed in 5 5.
Note that the function Y carries the same amount of information as the original
Riemanr-Silberstein vector since Y can be constructed from F by splitting this
vector into positive and negative frequency parts and then using the first part as
the upper components of Y and the complex conjugate of the second part as
the lower components:

The positive frequency part of the solutions of wave equations is also a well-
known concept in classical electromagnetic theory, where it is called the analytic
signal (Born and Wolf [ 19801, Mandel and Wolf [ 19951).

3.2. TRANSFORMATION PROPERTIES OF THE PHOTON WAVE FUNCTION IN

COORDINATE REPRESENTATION

In free space, the components of the electromagnetic field form a tensor, and
that allows one to establish the transformation properties of 3.Transformation
properties of the photon wave function Y are the same as those of F.Under
rotations, the upper and the lower half of Y transform as three-dimensional
vector fields. Under Lorentz transformations, the upper and the lower part also
transform independently and the corresponding rules can be inferred directly
from classical electrodynamics (cf., for example, Jackson [ 19751). Under the
Lorentz transformation characterized by the velocity u, the vectors F% change
as follows:

uxF y2 u(u.F)
~i=y(~~i-)--- (3.7)
C y+l c2 ’

where y is the standard relativistic factor y = d w .One may check that

this transformation preserves the square of these vectors,
262 PHOTON WAVE FUNCTION [v, I 3

This is understood easily if one observes that F i is a combination of the well-

known scalar invariant S = (EOE’- B2/h)/2 and the pseudoscalar invariant
P = m E . B of the electromagnetic field,
F: = S f iP. (3.9)

Thus, rotations and Lorentz transformations act on vectors Fk as elements of

the orthogonal group in three dimensions (Kramers [I9381 p. 429) leading to the
following transformation properties of the wave function:

(3.10)

where C is a three-dimensional, complex orthogonal matrix and C* is its

complex conjugate. The unification of rotations and Lorentz boosts into one
complex orthogonal transformation is even more transparent for infinitesimal
transformations,

Y’ = Y + ((i6y + p&).s) Y, (3.1 1)

where 6 y is the vector of an infinitesimal rotation. Under the space reflection

r + -r, the upper and lower part of Y do not transform independently but
are interchanged because D and B transform as a vector and a pseudovector,
respectively:

Y ’ ( - r , t ) = pl Y ( r ,t). (3.12)

All these transformation properties can also be simply stated in terms of

second-rank spinor fields (cf. 0 12).

3.3. PHOTON HAMILTONIAN

The operator appearing on the right hand side of the evolution equation (2.19)
for the photon wave function is the Hamiltonian operator for the photon:

(3.13)

In free space, this expression reduces to:

k
fio = cp3 ( sy). (3.14)
v, 5 41 MOMENTUM REPRESENTATION 263

The formulas (3.13) and (3.14) define a Hermitian operator with continuous
spectrum extending from -00 to 00. Hermiticity is defined here with respect
to the standard (mathematical) scalar product,

(3.15)

Wave functions of physical photons are built from positive-energy solutions

of the eigenvalue problem for the Hamiltonian,

ii P ( r ) = E P ( r ) . (3.16)

There is a simple relation between positive-energy solutions and negative-energy

solutions. One may obtain all solutions for the negative energies just by an
A h

interchange of upper and lower components, since p l H p l = -H. Such a simple

symmetry of solutions is a result of photons being identical with antiphotons
and it is not found, in general, for particles whose antiparticles are physically
distinct. For example, the solutions of the wave functions describing electrons in
the Coulomb potential of the proton are quite different from the wave function of
positrons moving in the same potential. In the first case the potential is attractive
(bound states), while in the second case it is repulsive (only scattering states).
Explicit solutions of the energy eigenvalue problem for photons are obtained
easily in free space, but in the presence of a medium this can be done only in
special cases. In this respect, wave mechanics of photons are not much different
from wave mechanics of massive particles, where explicit solutions can also be
found only for special potentials.

0 4. Photon Wave Function in Momentum Representation

The most thorough textbook treatment of quantum mechanics of photons has

been given by Akhiezer and Berestetskii [1965], who devoted a long chapter
to this problem. Their discussion is limited to momentum representation except
for a brief subsection under a characteristic title: “Impossibility of introducing a
photon wave function in the coordinate representation”. This impossibility will
be addressed in 9; 8.
Wave mechanics of photons in momentum representation can be derived
directly from relativistic quantum kinematics and group representation theory,
but here the analysis will be based on the Fourier representation of the photon
wave function in coordinate representation.
264 PHOTON WAVE FUNCTION [v, 5 4

In this review, I shall use traditionally the wave vector, k, instead of the photon
momentum vector, p = hk, as the argument of the wave function in momentum
space. The explicit introduction of Planck's constant is necessary only when the
proper normalization of the wave function is needed.

4.1. PHOTON WAVE FUNCTION AS A FOURIER INTEGRAL

The standard procedure for solving the wave equations (1.5) or (2.3) is based on
Fourier transformation. Since every solution of eqs. (1 .S) and (1.6) is a solution
of the d'Alembert equation,

the vectors F* can be represented as superpositions of plane waves

F,(r, t ) = &/m d3k

[f,(k) e-iwrtik'r+fl(k) ,
eiwr-ik'r]

where w = clkl and the factor v& has been introduced for future convenience.
The remaining integral has the dimension of l/length2, so that the Fourier
coefiicientsf* have the dimension of IAength. It has already been taken into
account in eqs. (4.2) and (4.3) that the vectors F, and F- are complex conjugate
of each other. In order to fulfill Maxwell's equations, the two complex vectors
f+(k) and f_(k) must satisfy the set of linear, algebraic equations which result
from eqs. ( 1 . 5 ) and (1.6); respectively,
ik x f * ( k ) = &lklF*(k), (4.4)
k.f*(k) = 0. (4.5)
Actually, the second equation is superfluous since it follows from the first.
Solutions of these equations are determined up to a complex factor. Denoting
by e(k) a normalized solution of the first equation taken with a plus sign
ik x e(k) = Ikle(k), (4.6)
e*(k).e(k)= 1, (4.7)
one can express the vectors f*(k) in the form:
f+(k) = e ( k ) f ( k ,11, f-(k) = e * ( k ) f ( k , - l ) . (4.8)
The two complex functions f (k, A), where A = k1, describe the independent
degrees of freedom of the free electromagnetic field. The vector e(k) can be
v, 9: 41 MOMENTUM REPRESENTATION 265

decomposed into two real vectors li(k) which form, together with the unit vector
n(k) = k/lkl, an orthonormal set:

e ( k ) = (ll(k)+ i l Z ( k ) ) / h , li(k).l,(k)= 6,,, (4.9)

n(k).li(k)= 0, ll(k) x l2(k) = n(k). (4.10)

The only freedom left in the definition of e(k) is its phase. A multiplication by
a phase factor amounts to a rotation of the vectors li(k) around the vector n(k).
The same freedom characterizes the coefficient functions f ( k ,A). This phase
may, in general, depend on k and it plays an important role in the study of the
photon wave function in momentum representation. The final form of the Fourier
representation for vectors F* is:

F+(r,t ) = f i / m e d3k
(k) [f ( k , I)e?t+ik'r +f *(k,-1) eiw'-ik.r], (4.1 1)

4.2. INTERPRETATION OF FOURIER COEFFICIENTS

In free space, the energy, momentum, angular momentum, and moment of energy
of the classical electromagnetic field are given by the expressions (1,8)-( 1.1 1).
With the help of eqs. (4.1 1) of (4.12) they can be expressed in terms of the
coefficient functions f ( k ,A) (cf., for example, Bialynicki-Birula and Bialynicka-
Birula [ 19751):

(4.13)

(4.14)

(4.16)

where
266 PHOTON WAVE FUNCTION [v, § 4

and

a ( k ) = I , ( k ) . ak12(k)- 12(k). akl,(k). (4.18)

The operation Dk is a natural covariant derivative on the light cone (Starusz-

kiewicz [ 19731, Bialynicki-Birula and Bialynicka-Birula [ 1975, 19871). Note
that this operation depends, through the vector a(k), on the phase convention
for the polarization vector e ( k ) . It is interesting to note that the vector a(k)
has properties simular to the electromagnetic vector potential. It changes by a
gradient under a change of the phase, but its curl is defined uniquely. Indeed, it
follows from the definition (4.18) of a(k) and from the orthonormality conditions
(4.10) that the vector a ( k ) obeys the equation:

a l U l - q U l = --EIlknk/k2. (4.19)

This equation determines the Berry phase (Bialynicki-Birula and Bialynicka-

Birula [ 19871) in the propagation of photons.
The formulas (4.13) and (4.14) indicate thatf(k, f l ) describe field amplitudes
with energy hw and momentum hk. The formula (4.15) shows that f ( k , 1)
andf(k, -1) describe field amplitudes with positive and negative helicity, since
their contribution to the component of angular momentum in the direction of
momentum is equal to *th, respectively.
The functions f ( k , A) actually have a dual interpretation. In classical theory
they yield full information about the electromagnetic field. In wave mechanics
of the photon, these functions are the components of the photon wave function
in momentum representation. In order to distinguish between these two cases,
I shall denote the two components of the photon wave function in momentum
representation by a new symbol, @(k,A). Wave function must be normalized and
the proper normalization of the photon wave function @ ( k A) , is discussed in (j 5.
The expansion of the photon wave function into plane waves has the following
form:

(4.20)

where

e(k, 1) = e(k), e(k,-1) = e*(k). (4.21)

The integral (4.20) defines a certain (continuous) superposition of the wave

functions @(k,A)with different values of the wave vector. If the photon wave
V, I51 PROBABlLlSTlC INTERPRETATION 261

function in momentum representation is accepted as a legitimate concept, then

the superpositions of such functions must also be accepted. Those who are not
sure about the meaning of a continuous superposition, in the form of an integral,
may restrict the electromagnetic field to a box and replace the integral by a
discrete sum.

4.3. TRANSFORMATION PROPERTIES OF THE PHOTON WAVE FUNCTION IN

MOMENTUM REPRESENTATION

Transformation rules for the wave function in momentum representation may be

derived from the transformation properties of the Riemann-Silberstein vector.
They are the same in the classical theory of the electromagnetic field and in the
quantum theory of photons. Under space and time translations, functions @ ( k A)
,
are multiplied by the phase factors

@'(k,A) = exp(-iwto + ik.ro) @(k,A), (4.22)

where (Q,to) is the four-vector of translation. From expressions (4.13) and (4.14)
one may deduce that under rotations and Lorentz transformations, the functions
@(k,A) are also multiplied by phase factors,

@'(k',A) = exp(-iAO(k, A)) @ ( k A),

, (4.23)

where the phase function O ( k , A ) depends on the PoincarC transformation A.

This transformation property can be derived easily from the transformation law
of the energy-momentum four-vector for the electromagnetic field. The right
hand side in the formulas (4.13) and (4.14) has three factors: the integration
volume d3W(2n)31kl,the four-vector ( k , w), and the moduli squared of @(k,A).
The integration volume is an invariant (cf., for example, Weinberg [ 19951 p. 67),
and therefore I@(k,A)I2must also be invariant. An explicit form of the phase
O(k,A) corresponding to a given Poincare transformation A can be given (cf.,
for example, Amrein [1969]), but it is not very illuminating.

8 5. Probabilistic Interpretation

Probabilistic interpretation of wave mechanics requires, first of all, a definition

of the scalar product (Yl I Yz) that is to be used in the calculation of transition
probabilities. The modulus squared of the scalar product of two normalized wave
268 PHOTON WAVE FUNCTION [v, § 5

functions I ( Y
lI Y2)l2 determines the probability of finding a photon in the state
Yl when the photon is in the state Y2. The probability, of course, must be a
pure number and - as a true observable - must be invariant under all Poincare
transformations. The most obvious definition of the scalar product (3.15) cannot
be used, because it is neither PoincarC invariant nor dimensionally correct. There
is essentially only one candidate for the correct scalar product. Its heuristic
derivation is the easiest in momentum representation.

5.1. SCALAR PRODUCT

According to quantum mechanics, each photon with momentum hk carries

energy hw. Thus, the total number of photons N present in the electromagnetic
field is obtained by dividing the integrand in the formula (4.13) by hw,

A photon wave function describes just one photon. The normalized wave function
must, therefore, satisfy the condition N = 1. Normalized photon wave functions
in momentum representation satisfy the normalization condition:

The form of the scalar product can be deduced from the expression for the
norm and it reads:

This scalar product can be also expressed in terms of the photon wave functions
in coordinate representation by inverting the Fourier transformation in eq. (4.20):

J'
1
#(k,A) = -e*(k, A). d3rexp(-ik.r) Y(r, t ) , (5.4)
6
where the scalar product with e*(k,A) is evaluated separately for upper and lower
components, and for each A only one of them does not vanish. Upon substituting
v, P 51 PROBABILISTIC INTERPRETATION 269

this expression into eq. (5.3), interchanging the order of integrations, and using
the following properties of the vectors e(k,A),

e*(k,A).e(k,A') = 6A,Af, (5.5)

C e:(k, A)e,(k,A) = 6 , - n,(k)n,(k), (5.6)
A
one obtains the following expression for the scalar product in coordinate
representation

The norm associated with this scalar product is:

N = IIY[12= - d3r
2n2hc 'ss d3r' Yt(r)---
1
Ir - r'I2
Y(r').

The scalar product (5.7) and the associated norm (5.8) for photon wave
functions have been arrived at by numerous authors starting from various
premises. Gross [ 19641 has proven that this scalar product and this norm are
invariant not only under Poincark transformations but also under conformal
transformations. Zeldovich [1965] derived the formula (5.8) for the number of
photons in terms of the electromagnetic field vectors. Recently, the norm (5.8)
has been found very useful in the formulation of wavelet electrodynamics (Kaiser
[1992]). The same expression (5.3) for the scalar product can also be derived
by considering quantum-mechanical expectation values. That approach has been
used by Good [ 19571 and is presented below.

5.2. EXPECTATION VALUES OF PHYSICAL QUANTITIES

In wave mechanics of photons, the normalized photon wave function #(k,A)

replaces the classical field amplitudes f ( k , A). The classical expressions for the
energy, momentum, angular momentum, and moment of energy become the
formulas for the quantum-mechanical expectation values:

(5.9)
(5.10)

These equations compared with the formulas (4.13)-(4.16) enable one to identify
, - . A h h

the operators H , P , J , and K in momentum representation as:

fi = hw, (5.1 1)
270 PHOTON WAVE FUNCTION [v, D 5

= hk, (5.12)

(5.13)

h
K = hoTDk. (5.14)
1

- A h

The operators H , P,J , and K are Hermitian with respect to the scalar product
given by eq. (5.3).
The formulas (5.1 1)-(5.14) are fully consistent with the interpretation of
#(k, A) as the probability amplitude in momentum representation. The probability
density to find the photon with the momentum hk and the helicity A is:

(5.15)

In order to express the quantum-mechanical expectation values (5.1 1)+5.14)

in coordinate representation, one must identify the proper form of the scalar
product for the photon wave function Y. This identification was made by Good
[ 19571 who compared the classical formula for the energy of the electromagnetic
field with the quantum-mechanical expression involving the Hamiltonian

and came to the conclusion that the scalar product for the photon wave function
has to be modified as follows:

1
1 Y2)= / d 3 r Y/z:Yz.
(Y, (5.17)
H
It is assumed here that the wave functions are built from positive energy states
only, and that assumption guarantees the positive definiteness of the norm,

1
11 = /d3r Yt Y, (5.18)

associated with that scalar product. This form of the scalar product leads to the
following expectation values of the energy, momentum, angular momentum, and
moment of energy operators:

(E) = J’d3r Ytfi-’ (s.:V) Y, (5.19)

Y 9: 51 PROBABILISTIC INTERPRETATION 27 1

h
(P) = /d3r Y t f i - l1y V Y , (5.20)

( M ) = Jd3r ~ + i l(-r 'x :V + d s ) Y, (5.21)

(5.22)

Thus, the operators H,P,J , and k in coordinate representation have the form
h e -

il = c ( yA ) ,
(5.23)
- h
P = TV, (5.24)
1
A
(5.25)
h

J =rxTV+hs,
1
it = Gr. (5.26)

All these operators preserve the divergence condition (1.6), and are Hermitian
with respect to the scalar product (5.17). It is also reassuring to note that
the quantum-mechanical operators of momentum and angular momentum in
coordinate representation have the same form as in standard quantum mechanics.
This can be taken as another indication that Y ( r ,t ) is a legitimate and useful
object.
One may prove directly (without using the Fourier expansions) with the help
of the following identities:

and with the use of eq. (2.4), that the expectation values (5.19H5.21) reduce to
the classical expressions ( 1 .S)-( 1 . 1 1) when the wave function is replaced by the
classical electromagnetic field.
The scalar product (5.17) in coordinate representation has been obtained from
the scalar product (5.3) in momentum representation. However, the scalar product
that contains the division by the Hamiltonian can be derived on more general
grounds, and its definition does not depend on the choice of representation. It
has been shown (Segal [1963], Ashtekar and Magnon [1975]) that such a scalar
product is a general feature of geometric quantization in field theory.
Even though the number of photons is given by a double integral, so that there
is no local expression for the photon probability density in coordinate space,
212 PHOTON WAVE FUNCTION [v, D 5

the expression for the energy has the form of a single integral over IY(r)I2.
Therefore, one may introduce a tentative notion of the “average photon energy
in a region of space” and try to associate a probabilistic interpretation of the
photon wave function with this quantity (Bialynicki-Birula [ 19941, Sipe [ 19951).
More precisely, the quantity p s ( Q ) ,

(5.28)

may be interpreted as the probability of finding the energy of the photon localized
in the region Q. In other words, p ~ ( L 2is) the fraction of the average total energy
of the photon associated with the region Q. The probability density pE(r, t ) of
finding the energy of the photon at the point r ,

(5.29)

is properly normalized to one and it also satisfies the continuity equation

(5.30)

with the normalized average energy flux

(5.3 1)

as the probability current. The direct connection between the wave function
Y(r, t ) and the average energy density justifies the name “energy wave function”
used by Mandel and Wolf [1995]. It is understandable that the localization of
photons is associated with their energy because photons do not carry other
attributes like charge, fermion number, or rest mass. It is worth noting that
for gravitons not only the probability but even the energy cannot be localized
(cf., for example, Weinberg and Witten [ 19801). The probabilistic interpretation
of the energy wave function Y is still subject to all the limitations arising
from the lack of the photon position operator, as discussed in 4 8. In particular,
there are no projection operators whose expectation values would give the
probabilities pE(Q).
The transition amplitudes, the operators representing important physical
quantities, and the expectation values can be expressed with equal ease in
momentum representation and in coordinate representation. For photons moving
in empty space, both representations are completely equivalent and give the same
v, 5 61 EICENVALUE PROBLEMS 273

results. The only relevant issue is whether a particular superposition of wave

functions in momentum representation is useful for the description of quantum
states of the photon. The distinguished and unique feature of the superposition
given by the Fourier integrals (4.20) is that they represent local fields. They have
local transformation properties (3.10) and they satisfy local boundary conditions.
Therefore, for photons moving in an inhomogeneous or bounded medium, it is
the coordinate representation that is preferred because only in this representation
one may easily take into account the properties of the medium (cf. 5 6).

5.3. CONNECTION WITH LANDAU-PEIERLS WAVE FUNCTION

One may easily convert the scalar product (5.7) into a standard form containing
a single integration with the use of the following identity:

(5.32)

This enables one to convert the double integral (5.7) into a single integral:

(YI1%) = /d3r @!(r>W r ) . (5.33)

The new functions @ are the Landau-Peierls wave functions. They are related
to the photon wave functions Y through the formula:

(5.34)

The form of the scalar product for the Landau-Peierls wave functions is simple,
but one must pay for this simplicity with the nonlocality of the wave functions.
There is also a simple mathematical argument that shows shortcomings of the
Landau-Peierls wave function. While for every integrable wave function Y the
transformation (5.34) defines the Landau-Peierls wave function @, the inverse
transformation is singular since it contains a nonintegrable kernel Ir - r’17/2
(Amrein [ 19691, Mandel and Wolf [1995]). This leads to a paradox that for many
“reasonable” functions @ (for example, for every function that becomes zero
abruptly at the boundary), the energy density is infinite. Thus, it is much more
natural to treat Y as the primary and @ as the derived object.

$ 6. Eigenvalue Problems for the Photon Wave Function

In wave mechanics of photons, as in wave mechanics of massive particles, one

may study eigenvalues and eigenfunctions of various interesting observables. The
274 PHOTON WAVE FUNCTION [v, 5 6

most important observables, of course, are the momentum, angular momentum,

energy, and moment of energy - the generators of the Poincark group. The
eigenhnctions of these observables will be given in coordinate representation
to underscore the validity and usefulness of the photon wave function in this
representation.

6.1. EIGENVALUE PROBLEMS FOR MOMENTUM AND ANGULAR MOMENTUM

The eigenvalue problems for the components of the photon momentum operator
have the standard quantum-mechanical form,

and their solutions depend on r through the exponential functions exp(ik-r).

The eigenvalue problem for the photon angular momentum also has the
standard quantum-mechanical form. It contains, as usual, the eigenvalue problem
for the z-component of the total angular momentum,

and the eigenvalue problem for the square of the total angular momentum

The solutions of eqs. (6.2) and (6.3) are well-known vector spherical harmonics
(cf., for example, Messiah [ 19611). The direct connection between the quantum-
mechanical eigenvalue problems and multipole expansion in classical electro-
magnetism was explored systematically for the first time by Molikre [1949].

6.2. EIGENVALUE PROBLEM FOR THE MOMENT OF ENERGY

The solution of the eigenvalue problem for the moment of energy shows the
versatility of the calculational methods based on the coordinate representation
and sheds some light on the problem of the localizability of the photon that
is discussed in $8. Of course, the same result can be obtained by Fourier
transforming the solution of the eigenvalue problem obtained in momentum
representation.
The three components of the moment of energy, like the components
of angular momentum, do not commute among themselves. Therefore, the
V, P 61 EIGENVALUE PROBLEMS 275

eigenvalue problem can be posed only for one component at a time. Choosing, for
definiteness, the z-component, one obtains the following eigenvalue equation:

-ip3(s.V)z Y =K Y. (6.4)

The solution of this equation becomes unique when one chooses two additional
eigenvalue equations to be solved concurrently. For example, eq. (6.4) can be
solved together with the eigenvalue problems for the x and y components of
momentum, since the three operators Fx,Fy, and kzcommute. The solutions for
the upper and lower components of the wave function differ only in the sign of
K and one can solve them independently. When the wave function in the form
Y = exp(ikxx + ikyy)(qx(z), qy(z), qZ(z))is substituted into eq. (6.4), one obtains
a set of ordinary differential equations:

where the prime denotes the differentiation with respect to z. These equations
are solved by the following substitution (k: = k,' + k,):

which results in a Bessel-type equation for qZ,

z2q: +zq: +( ~ k:z2)qZ

K - = 0. (6.10)

The physically acceptable solution of this equation is given by the Macdonald

function of the imaginary index,

(6.11)

The other solution grows exponentially when z -+ 00 and must be rejected. The
physical solution falls off exponentially for large Iz( and represents a photon
state that is localized as much as possible in the z-direction. The remaining two
components of the eigenfunction are obtained from eqs. (6.8) and (6.9). The
216 PHOTON WAVE FUNCTION [v, 5 6

photon wave functions which describe eigenstates of gzare not normalizable,

because the spectrum of the eigenvalues is continuous: K can be any real number.

6.3. PHOTON PROPAGATION ALONG AN OPTICAL FIBER AS A QUANTUM-

MECHANICAL BOUND STATE PROBLEM

The eigenvalue problem for the photon energy operator in the absence of
a medium is solved by Fourier transformation as described in $4. In the
presence of a medium, one can search for eigenstates of the photon Hamiltonian
closely following the path traveled in nonrelativistic wave mechanics of massive
particles. This procedure usually involves selecting a set of operators commuting
with the Hamiltonian and then solving the appropriate set of eigenvalue
equations. The photon propagation along an infinite cylindrical optical fiber (cf.,
for example, Bialynicki-Birula [ 19941) is a good illustration of this approach.
In order to take care of the boundary conditions at the surface of the fiber, one
must work in the coordinate representation.
Consider an infinite, cylindrical optical fiber of diameter a characterized by a
dielectric permittivity E . The symmetry of the problem suggests the inclusion in
the set of commuting operators, in addition to the Hamiltonian, the projections
of the momentum operator and the total angular momentum on the direction
of the fiber axis. In cylindrical coordinates the eigenvalue equations for the
z-components of momentum and angular momentum and the Hamiltonian have
the form:

-idz Y = k,Y, (6.12)

( - id, + (s.ez>) Y = MY, (6.13)

(6.14)

where ep, e,, and e, are the unit vectors along the coordinate lines and u
equals to c outside the fiber. Due to the symmetry of the problem, there is no
coupling between the upper and lower components of Y and the solution of these
eigenvalue equations can be sought in the form of a three-dimensional vector,

In order to separate the variables and obtain a set of ordinary differential

equations, one needs the following differential and algebraic relations:

d,ep = e,, d,e, = -ep, (6.16)

v, § 61 EIGENVALUE PROBLEMS 211

(s.ep)e,, = ie,, (s.e,)ep = ie,, (s.e,)e, = ie,. (6.17)

All unlisted terms of the type (6.16) and (6.17) vanish. The dependence on 9
and z of all three components I),, I),, and I), of the photon wave function can
be separated out on the basis of eqs. (6.12) and (6.13),

The three p-dependent components of the wave function satisfy the equations

1 M . 0
-8,Pfq- = ;fi. (6.21)
P P
These equations lead to a Bessel equation forf,,

(6.22)

where k: = w2/u2- k,'. The remaining two functions f can be determined in

terms off,:

(6.23)

(6.24)

The photon wave function obeys the Bessel equation inside the fiber with one
value of k l , and with a different value of k l in the surrounding free space.
The behavior of the solution of eq. (6.22) depends on whether kl is real or
imaginary. A general solution of this equation is either (for real k l ) a linear
combination of Bessel functions of the first kind J M ( ~and ) the second kind
YM(p),or (for imaginary k l ) a linear combination of modified Bessel functions
I M ( p ) and K M ( ~ )In. full analogy with the problem of a potential well in
quantum mechanics, one can search for bound states in the transverse direction
by matching a regular oscillatory solution inside (i.e. the J M ( p ) function)
with an exponentially damped solution outside the fiber (i.e. the KM(p ) function).
The matching conditions, well known from classical electromagnetic theory, are
278 PHOTON WAVE FUNCTION tv, 5 7

the continuity conditions for the E, and H, field components at the surface of the
fiber, when p = a. Bound states occur because the speed of light is greater in the
vacuum than inside the fiber. Therefore, it may happen that k l is real inside and
imaginary outside the fiber. Since there are two matching conditions and only
one ratio of the amplitudes inside and outside the fiber, both conditions can be
satisfied only for a set of discrete eigenvalues of the photon energy hw. It is worth
noting that in order to have an imaginary k l , one must have a nonvanishing k,.
Thus, a photon may be bound in the plane perpendicular to the fiber, but it is
always moving freely along the fiber, as in the quantum-mechanical description
of a charged particle moving in a homogeneous magnetic field. This analysis
gives an interpretation of electromagnetic evanescent waves as quantum bound
states. Of course, true bound states of photons, which are described by a photon
wave function decaying exponentially in all directions, are not possible.

0 7. Relativistic Invariance of Photon Wave Mechanics

In a relativistically invariant quantum theory, the Poincare transformations

are represented by unitary operators. The ten Hermitian generators of these
transformations must satisfy the commutation relations characteristic of the
Poincare group. The ten generators of the PoincarC group are identified with
the operators H , P,J , and 2.They generate infinitesimal time translation, space
translations, rotations, and boosts (special Lorentz transformations), respectively.
The structure of the Poincare group leads to the following commutation relations
obeyed by these generators (cf., for example, Bargmann and Wigner [ 19481,
Bialynicki-Birula and Bialynicka-Birula [ 19751, Itzykson and Zuber [ 19803,
Weinberg [I9951 p. 61):

[-K;,H-1 = i@;,

All the remaining commutators vanish. One may check by a direct calculation
that the operators H , P, J? and K , given in momentum representation by
A -
V, 5 81 LOCALIZABILITY OF PHOTONS 279

the formulas (5.11)<5.14) and in coordinate representation by the formulas

(5.23)<5.26), obey the commutation relation for the generators of the Poincarh
group. In the proof of the commutation relations in momentum representation,
one needs the condition (4.19). Since all generators of the Poincart group are
represented by operators which are Hermitian with respect to the scalar product
(5.3) or (5.7), the Poincart transformations are represented by unitary operators.
Therefore, the scalar product is invariant under these transformations and all
transition probabilities are the same for all observers connected by Poincart
transformations. Thus, in both coordinate and momentum representations, wave
mechanics of photons is a fully relativistic theory.

0 8. Localizability of Photons

The problem of localization of relativistic systems was first posed and solved by
Newton and Wigner [ 19491, and later refined by Wightman [ 19621. According
to this analysis it is possible to define position operators and localized states
for massive particles and for massless particles of spin 0, but not for massless
particles with spin. Thus, the position operator in the sense of Newton and
Wigner does not exist for photons (cf. also a recent tutorial review on that
subject by Rosewarne and Sarkar [1992]). As a simple heuristic explanation of
why a position operator for the photon does not exist, one may observe (Pryce
[ 19481) that the multiplication by r cannot be applied to the photon wave function
because it destroys the divergence condition (1.6).
A weaker definition of localization that is applicable, even when the position
operator does not exist, was proposed by Jauch and Piron [1967] and a
very detailed analysis of this problem has been given by Amrein [1969].
The Jauch-Piron localizability allows for noncompatibility of “photon position
measurements” in overlapping regions. The main weakness of such an abstract
analysis is that an operational definition of the photon position measurement
for photons has not been incorporated into it. The existence of position
measurements for photons is just taken for granted, regardless of the feasibility
of their physical realizations. When a realistic model of the photon detector
is brought in, it is the wave function Y rather than @ that appears as the
correct probability amplitude for photodetection (Mandel and Wolf [ 19951).
Thus, in practical applications the energy wave functions Y always seem to play
a dominant role.
It must, however, be stressed that even for massive particles, the localization
is not perfect, because it is not relativistically invariant. Two observers who
280 PHOTON WAVE FUNCTION [V, 9; 8

are in relative motion would not quite agree as to the localization region of a
relativistic particle. This follows from the fact that the Newton-Wigner wave
function yjNW is related to the relativistic wave function y j that transforms locally
under Poincare transformations by a nonlocal transformation (cf. Haag [ 19931):

yjNW(r)= /d’r’K(r - r’) @(r’),

where the kernel K can be represented in terms of the Macdonald function,

In the limit, when m -+ 0, K(r) + ~ / ( 2 n r ) - ~ /and ’ eq. (8.1) becomes the

relation (1.2) between the local wave function of the photon and the Landau-
Peierls wave function. Thus, the difference between the localizability of massive
particles and photons is not that great. In both cases, localization cannot be
defined in a relativistic manner. However, for massive particles departures from
strict localization are only exponentially small due to the fast decay of the
Macdonald function in eq. (8.1). In the nonrelativistic limit, when c + m, the
exponential tails become infinitely sharp and the localization is restored.
Difficulties with the position operator for relativistic particles have a profound
origin connected with the structure of the Poincare group. In nonrelativistic
physics the position operator is the generator (up to a factor of mass) of Galilean
transformations (cf., for example, Gottfried [1966], Weinberg [ 19951 p. 62). In
a relativistic theory, the Galilean transformations are replaced by the Lorentz
transformations and the position operator (multiplied by the mass) is replaced
by the boost generator K . The main difference between Galilean and Lorentz
transformation affecting the discussion of localizability is that boost generators
do not commute. Therefore, one may only hope to localize relativistic particles
in one direction at a time. The possibility to localize photons in one direction
has been discussed in general terms as the “front” description by Acharya and
Sudarshan [1960]. The eigenfunctions of the boost operator K, given in tj 6 may
serve as an explicit realization of the front description for the photon.
The considerations of photon localizability, while important for the under-
standing of some fundamental issues, do not much influence the practical
applications of the photon wave function. All that should really matter there
is that the wave function be defined precisely and that its interpretation be not
extended beyond the limits of applicability.
v, P 91 PHASE-SPACE DESCRIPTION OF A PHOTON 28 1

0 9. Phase-Space Description of a Photon

Distribution functions in phase space are a very convenient tool in the description
of statistical properties and the study of the classical limit of wave mechanics.
A direct analog of the Wigner function (Wigner [1932]) introduced in wave
mechanics may also be introduced for photons with the help of the photon
wave function. This is done by Fourier transforming the product of the wave
function and its complex conjugate. Fourier transforms of the electromagnetic
fields similar to the Wigner function have been introduced in optics, first by
Walther [ 19681 in the two-dimensional context of radiative transfer theory and
then by Wolf [1976] and by Sudarshan [1979, 198la,b] in the three-dimensional
case. In these papers, phase-space distribution functions were defined only for the
stationary states of the electromagnetic field, and they were treated as functions
of the frequency. The time-dependent distribution functions can be defined
(Bialynicki-Birula [ 19941) with the use of the time-dependent wave function. The
only formal difference between the standard definition of the Wigner function in
nonrelativistic wave mechanics 'of massive particles and the case of photons is
the presence of vector indices. Thus, the photon distribution function in phase
space is not a single scalar function but rather a 6 x 6 Hermitian matrix defined
as follows:

Similar multi-component distribution functions also arise for a Dirac particle,

and one can use some of the techniques developed by Bialynicki-Birula, Gomicki
and Rafelski [1991] to deal with such functions.
Every 6 x 6 Hermitian matrix can be written in the following block form:

where all 3 x 3 matrices are Hermitian. This decomposition can also be

expressed in terms of the p matrices,

The vector indices i and j refer to the components within the upper and
lower parts of the wave function and the matrices p act on these parts as
282 PHOTON WAVE FUNCTION [Y 9 9

a whole. The most general photon distribution function, as seen from this
analysis, is quite complicated. In general, when the medium induces mixing
of the two polarization states, all components of the distribution function are
needed. However, when photons propagate in free space, only a subset of these
components is sufficient to account for the dynamical properties of photon
beams. The simplest case is that of a given helicity. A more interesting case
is that of an unpolarized photon beam: a mixture of both helicities with equal
weights. This state must described by the distribution function because a mixed
state cannot be treated by pure Maxwell's theory. In all these cases, phase-space
dynamics can be described by a 3x3 Hermitian matrix; i.e., by just one scalar
function and one vector function. To this end, one may introduce the following
reduced distribution function:

Wq(r,k, t ) =
J d3s e-ik'sq+(r+ is,t ) %*(r- is,t ) , (9.4)

where vi are the upper or the lower components of the original wave function.
The Hermitian matrix W; can be decomposed into a real symmetric tensor and
a real vector according to the formula

(9.5)

The tensor corresponds to the symmetric part of W$ and the vector corresponds
to the antisymmetric part. The factor of c has been separated out in the second
term since the vector u is related to the momentum density, while the trace of
wq is related to the energy density (Bialynicki-Birula [ 19941).
The equations satisfied by the components of the photon distribution function
in free space can be obtained from Maxwell's equations (1.5) and (1.6) for the
vector F ,

(k + ?i V ) Wq = 0 = (k 5 V I j Wq.
- 1

This leads to the following set of coupled evolution equations for the real
components wii and ui:
v, § 101 HYDRODYNAMIC FORMULATION 283

and to the subsidiary conditions

(9.10)
(9.1 1)

The k-dependent terms in the evolution equations describe a uniform rotation of

the vector ui and of the tensor wij around the wave vector k so that these terms
can be eliminated by “going to a rotating coordinate frame”.
With the help of the subsidiary conditions (9.10,9.11) one can eliminate the
remaining components and obtain from the evolution equations (9.8,9.9) the
equations for w = C wii and u,

(9.12)
(9.13)

These evolution equations do form a simple, self-contained set. However, as is

always the case with the phase-space distribution functions in wave mechanics,
not all solutions of these equations are admissible. Only those distribution
functions are allowed which can be represented in the form (9.4) at the initial
time (with u and w satisfying the subsidiary conditions (9.10) and (9.11)).

0 10. Hydrodynamic Formulation

It was shown by Madelung [ 19261 that the Schrodinger wave equation can be cast
into a hydrodynamic form. In this form, the complex wave function is replaced
by real variables: the probability density p and the velocity u of the probability
flow. The wave equation is replaced by the hydrodynamic evolution equations for
the variables p and u . In order to reduce the number of functions from four to the
original two, one must impose auxiliary conditions - the quantization condition -
on the velocity field. Later, other wave equations in quantum mechanics (Pauli,
Dirac, Weyl) were also presented in the hydrodynamic form. The wave equation
for the photon wave function is not an exception in this respect. It can also be
written (Bialynicki-Birula [ 1996b1) as a set of equations for real hydrodynamic-
like variables. Since the Riemam-Silberstein vector F carries all the information
284 PHOTON WAVE FUNCTION [Y § 10

about the photon wave function, one may use F to define these variables. They
comprise the energy density p and the velocity of the energy flow v,
C
p(r, t ) = F*(r,t ) . F(r, t), p(r, t ) u(r, t ) = -F*(r, t ) x F(r, t), (10.1)
2i
the components tij of the following tensor,

tij(r,t ) = F;(r, t ) q ( r ,t ) +F;C(r,t)Fi(r,t), (10.2)

and another vector u,

C
p(r, t ) u(r,t ) = 7(F*(r,t)VF(r,t ) - ( V F * ( r ,t))F(r,t)). (10.3)
21
Owing to the existence of the following identities satisfied by the hydrodynamic
variables,

only one component of tv is arbitrary, but the hydrodynamic equations look more
symmetric when all the components are treated on equal footing. The number of
algebraically independent hydrodynamic variables is reduced from eight to six
(the number of degrees of freedom described by F ) by the following quantization
condition
1
Jd,.?. [vx ii- s&(,k(UiVujx vuk +uiV$ x vtkl-2tilvtjl x v u k ) ] = 2nn, (10.5)

where n is a natural number. This condition must hold for every choice of the
integration surface and it states, in essence, that the phase of the wave function
is defined uniquely (up to an overall constant phase).
The evolution equations for the hydrodynamic variables are

(10.6)
(10.7)

(10.8)
 285

(10.9)

express the divergence

(10.10)

(10.11)

Hydrodynamic description of the photon dynamics is not simple, but its

existence again underscores the unification of the quantum theory of the photon
with the rest of quantum mechanics. Everything that can be done with other
particles can also be done for photons.

8 11. Photon Wave Function in Non-Cartesian

Coordinate Systems and in Curved Space

It was shown by Skrotskii [I9571 and Plebanski [ 19601 (for a pedagogical review,
see Schleich and Scully [ 19841) that the propagation of the electromagnetic field
in arbitrary coordinate systems, including the case of curved space-time, may be
described by Maxwell’s equations, with all the information about the space-time
geometry contained in the relations connecting the field vectors E , B and D, H .
This discovery can be enhanced further by the observation (Bialynicki-Birula
[ 19941) that in contradistinction to the case of an inhomogeneous medium, in
the gravitational field the two photon helicities do not mix. This follows from
the fact that for arbitrary metric g,, the constitutive relations can be written as
a single equation connecting two complex vectors: the vector F(r, t ) defined by
eq. (1.7) and a new vector C(r,t ) defined as

(11.1)

In curved space (or in curvilinear coordinates), the constitutive relations for two
complex vectors F(r, t ) and G(r,t ) have the form
F’. = - - (If i g U + i g o k & ; k i ) ~ j ,
(11.2)
goo
286 PHOTON WAVE FUNCTION [v, 5 1 2

(11.3)

where g,, is the metric tensor, gp'' is its inverse, and g is the determinant of
g,,. The Maxwell equations expressed in terms of vectors C and F in curved
space-time are the same as in flat space:

i&F(r, t ) = V x C(r, t ) , (11.4)

V.F(r,t) = 0. (11.5)

In these equations, all derivatives are ordinary (not covariant) derivatives as in

flat space. The whole difference is in the form of the constitutive relations (1 1.2)
and (1 1.3). These relations contain all information about the gravitational field
or the curvilinear coordinate system. Since the relations between C and F are
linear, one may again write two separate wave equations for the two helicity states
as in flat space. By combining these two equations, one obtains the following
wave equation for the six-component photon wave function F(r, t ) :

where

(11.7)

These equations contain only the matrix p3 that does not mix the helicity states.
The true photon wave function Y(r,t) may be introduced only in the time-
independent case, when the separation of 3 ( r , t ) into positive and negative
energy parts is well defined.

5 12. Photon Wave Function as a Spinor Field

Soon after the formulation of spinor calculus by van der Waerden in the context
of the Dirac equation, it was discovered by Laporte and Uhlenbeck [1931]
that the Maxwell equations can also be cast into a spinorial form. The spinor
representation of the electromagnetic field and the Riemann-Silberstein vector
are closely connected. The components of the vector F are related to the
components of a second-rank symmetric spinor qj,,,~,

q500 = -F, + iFy, (12.1)

Y 5 121 WAVE FUNCTION AS A SPINOR FIELD 287

(12.2)
(12.3)

and the components of the complex conjugate vector F' are related to the
components of a second-rank symmetric primed spinor @'B',

(12.4)
(12.5)
(12.6)

The property that even in curved space both helicities propagate without mixing
is, in the spinorial formalism, a simple consequence of the fact that both spinors
$AB and @,' satisfy separate wave equations (Laporte and Uhlenbeck [193 11,
Penrose and Rindler [ 19841)

(12.7)
(12.8)

where the matrices and ~ ' A J Bare built from the unit matrix and the Pauli
matrices,

(12.9)
(12.10)

Under a Lorentz transformation the second-rank spinor changes according to the

formula

where SAC is a 2 x 2 matrix of the fundamental (spinor) representation of the

Lorentz group. Thus, from the point of view of the representation theory of
the Lorentz and PoincarC groups (Streater and Wightman [ 19781, Weinberg
[ 19951 p. 23 l), the photon wave functions for a given helicity are just the three-
component fields which transform as irreducible representations (1,O) or (0,l)
of the proper Lorentz group (without reflections). In order to accommodate
reflections, one must combine both representations and introduce the six-
dimensional objects F or Y.
Equations (12.7) [and similarly eqs. (12.8)] represent a set of four equations
obeyed by four components ($11,$12,$21,$22) of the second order spinor. All
288 PHOTON WAVE FUNCTION [Y0 13
four equations can be written in the form of the Dirac equation (Ohmura [ 19561,
Moses [ 19591):
h
ih&4(r, t ) = a . TV@(r,t ) , ( 12.1 2)
1

where the matrices a are

0010 00-i 0 1 0 0 0
0001 0 0 0 -i 01 0 0

0100 O i 0 0 0 0 0 -1
One may check that in this formulation the divergence condition takes on the
following simple algebraic form: @12 = 4 2 1 .
The Maxwell equations expressed in spinor notation and the Weyl equation
provide just the simplest examples from a hierarchy of wave equations for
massless fields described by symmetric spinors ~ B , B ~ . . .orB ~@B:B;.'.B:. All these
equations have the form (Penrose and Rindler [ 19841)
(12.14)
(12.15)
This universality of massless wave equations for all spins gives an additional
argument for treating the Riemann-Silberstein vector as the photon wave
function.

9 13. Photon Wave Functions and

Mode Expansion of the Electromagnetic Field

The concept of the photon wave function is also useful in the process of
quantization of the electromagnetic field. One may simply apply the procedure
of second quantization to the photon wave function in the same manner as one
does for other field operators. In order to see this analogy, one may recall that
the field operator $(r) for, say the electron field, is built from a complete set of
wave functions for the electrons t,hi(r) and from a complete set of wave functions
for positrons t,h;(r) according to the following rule (cf., for example, Schweber
[ 19611, Bialynicki-Birula and Bialynicka-Birula [ 19751, Weinberg [ 19951):
(13.1)

where 2, and inare the annihilation operators for electrons and positrons,
respectively. The second part of the field operator is related to the first by the
Y 5 131 WAVE FUNCTIONS AND MODE EXPANSION 289

operation of charge conjugation performed on the wave functions and on the

operators. The analog of charge conjugation for photon wave functions is given
by eq. (3.2). Following this procedure, one may construct the field operator of
the electromagnetic field in the form

(13.2)

where the identity of particles and antiparticles for photons has been taken
into account by using only one set of creation and annihilation operators. The
field operator ( 13.2) is non-Hermitian, but it satisfies the particle-antiparticle
conjugation condition (3.2). Therefore, it has only six Hermitian components.
These Hermitian operators are identified as the field operators 6 ( r ,t ) and g(r, t ) ,
and they are obtained from 3 through the formula
A

(13.3)

As a direct consequence of the formula (13.2), one may identify the photon
wave functions in the second-quantized theory with the matrix elements of the
electromagnetic field operators or 6 and 6 taken between one-particle states
and the vacuum:

In the simplest case of the free field, when the complete set of photon wave
functions may be labeled by the wave vector k and helicity A, the formula (13.2)
takes on the form

In the presence of a medium, the expansion (1 3.2) of the electromagnetic field

operator requires the knowledge of a complete set of wave functions Ynwhich
satisfy the photon wave equation in the medium. These functions are usually
called the mode functions of the electromagnetic field (cf., for example, Louise11
[1973] p. 240 and Mandel and Wolf [I9951 p. 905) but the term photon wave
functions is perhaps more appropriate (Moses [ 19731, Bialynicki-Birula and
Bialynicka-Birula [ 19751). The advantage of using the terminology of wave
functions is that it brings in all the associations with wave mechanics and makes
the classification of the modes more transparent. In particular, one may use
290 PHOTON WAVE FUNCTION [ Y P 14

the quantum-mechanical notion of eigenfunctions and eigenvalues to classify

the functions used in the mode expansion (Moses [1973]) and also borrow
from quantum mechanics the methods of proving their completeness (Bialynicki-
Birula and Brojan [ 19721).
This discussion shows that the photon wave function is not restricted to the
wave mechanics of photons. The same wave functions also appear as the mode
functions in the expansion of the electromagnetic field operators.

0 14. Summary

The aim of this review was to collect and explain all basic properties of a certain
well-defined mathematical object - a six-component function of space-time
variables - that describes the quantum state of the photon. Whether or not one
decides to call this object the photon wave function in coordinate representation
is a matter of opinion, since some properties known from wave mechanics of
massive particles are missing. The most essential property that does not hold
for the photon wave function is that the argument r of the wave function cannot
be directly associated with the position operator of the photon. The position
operator for the photon simply does not exist. However, one should remember
that for massive particles also, the true position operator exists only in the
nonrelativistic approximation. The concept of localization associated with the
Newton-Wigner position operator is not relativistically invariant. Since photons
cannot be described in a nonrelativistic manner, there is no approximate position
operator.
The strongest argument that can be made for the photon wave function in
coordinate representation is based on the most fundamental property of quantum
states - on the principle of superposition. According to the superposition
principle, wave functions form a linear space. By adding wave functions, one
again obtains a legitimate wave function. Once this principle is accepted, the
existence of photon wave functions in coordinate representation follows from
the existence of the photon wave functions in momentum representation, and
these functions are genuine by all standards; their existence follows simply
from relativistic quantum kinematics (or more precisely, from the representation
theory of the Poincark group). The Fourier integral (4.20) represents a special
combination of momentum space wave functions with different momenta and as
a matter of principle, such linear combinations are certainly allowed. One may
only argue which superpositions to take as more natural or useful, but rejecting
totally the very concept of the photon wave function in coordinate representation
is tantamount to rejecting altogether the superposition principle.
VI ACKNOWLEDGMENTS 29 I

There is not much advantage in using the photon wave function in coordinate
representation to perform calculations for photons moving in free space. The
relation of this wave function to momentum wave function is so straightforward
that one may as well stick to momentum representation. It is only in the presence
of a medium, especially in an inhomogeneous medium, that the photon wave
function in coordinate representation becomes useful and even essential. Only
in the coordinate representation may one hope to solve the eigenvalue problems
and to take into account the boundary conditions.
The introduction of the wave function for the photon has many significant
benefits. The photon wave function enables one to formulate a consistent wave
mechanics of photons that could be often used as a convenient tool in the
quantum description of electromagnetic fields, independent of the formalism of
second quantization. In other words, in constructing quantum theories of photons
one may proceed, as in quantum theory of all other particles, through two stages.
At the first stage, one introduces wave functions and a wave equation obeyed by
these wave functions. At the second stage, one upgrades the wave functions to the
level of field operators in order to deal more effectively with the states involving
many indistinguishable particles and to allow for processes in which the number
of particles is not conserved. Many methods which have proven very useful in
the study of particles described by the Schrodinger wave functions can also be
implemented for photons, leading to some new insights. These methods include
relationships between symmetries and operators, the definitions of various sets of
modes for the electromagnetic field and their completeness relations, eigenvalue
problems for various observables ( Q 6), phase-space representation (9 9), and
the hydrodynamic formulation ( Q 10). Finally, there are important logical and
pedagogical advantages coming from the use of the photon wave function. The
quantum-mechanical description of all particles, including photons, becomes
uniform.

Acknowledgments

Four lectures based on this review were presented while I was a Senior Visiting
Fellow at the Rochester Theory Center for Optical Science and Engineering in
February 1996. I would also like to acknowledge discussions with Z. Bialynicka-
Birula, M. Czachor, J.H. Eberly, A. Orlowski, W. Schleich, M.O. Scully, and
E. Wolf.
292 PHOTON WAVE FUNCTION

References

Acharya, R., and E.C.G. Sudarshan, 1960, “Front” description in relativistic quantum mechanics,
J. Math. Phys. 1, 532.
Akhiezer, A.I., and VB. Berestetskii, 1965, Quantum Electrodynamics (Interscience, New York)
ch. I .
Amrein, W.O., 1969, Localizability for particles of zero mass, Helv. Phys. Acta 42, 149.
Ashtekar, A,, and A. Magnon, 1975, Quantum fields in curved space-times, Proc. R. SOC.(London)
A 346, 375.
Bargmann, V, and E.P. Wigner, 1948, Group theoretical discussion of relativistic wave equations,
Proc. Natl. Acad. Sci. USA 34, 211. Reprinted: 1966, in: Symmetry Groups in Nuclear and
Particle Physics, ed. EJ. Dyson (Benjamin, New York) p. 205.
Bateman, H., 1915, The Mathematical Analysis of Electrical and Optical Wave Motion on the Basis
of Maxwell’s Equations (Cambridge). Reprinted: 1955 (Dover, New York).
Baym, G., 1969, Lectures on Quantum Mechanics (Benjamin, Reading, MA) ch. I .
Bialynicki-Birula, I., 1963, Quantum electrodynamics without electromagnetic field, Phys. Rev. 130,
465.
Bialynicki-Birula, I., 1994, On the wave function of the photon, Acta Phys. Polon. A 86, 97.
Bialynicki-Birula, I., 1996a, The photon wave function, in: Coherence and Quantum Optics VII, eds
J.H. Eberly, L. Mandel, and E. Wolf (Plenum Press, New York).
Bialynicki-Birula, I., 1996b, Hydrodynamics of relativistic ptobability flows, in: Nonlinear Dynamics,
Chaotic and Complex Systems, eds E. Infeld, R. Zelazny and A. Galkowski (Cambridge University,
Cambridge).
Bialynicki-Birula, I., and Z. Bialynicka-Birula, 1975, Quantum Electrodynamics (Pergamon Press,
Oxford).
Bialynicki-Birula, I., and Z. Bialynicka-Birula, 1976, Quantum electrodynamics of intense photon
beams. New approximation method, Phys. Rev. A 14, 1101.
Bialynicki-Birula, I., and Z. Bialynicka-Birula, 1987, Berry’s phase in the relativistic theory of
spinning particles, Phys. Rev. D 35, 2383.
Bialynicki-Birula, I., and J. Brojan, 1972, Completeness of evanescent waves, Phys. Rev. D5, 485.
Bialynicki-Birula, I., P. Gornicki and J. Rafelski, 1991, Phase-space structure of the Dirac vacuum,
Phys. Rev. D 44, 1825.
Bjorken, J., 1963, A dynamical origin of the electromagnetic field, Ann. Phys. (New York) 24, 174.
Bohm, D., 1954, Quantum Theory (Constable, London) p. 91.
Born, M., and E. Wolf, 1980, Principles of Optics (Pergamon, Oxford) p. 494.
Cohen-Tannoudji, C., B. Diu and F. Laloe, 1977, Quantum Mechanics, Vol. I (Wiley, New York).
Cohen-Tannoudji, C., J. Dupont-Roc and G. Grynberg, 1989, Photons and Atoms, Introduction to
Quantum Electrodynamics (Wiley, New York).
Cook, R.J., 1982% Photon dynamics, Phys. Rev. A 25, 2164.
Cook, R.J., 1982bi Lorentz covariance of photon dynamics, Phys. Rev. A 26, 2754.
Dirac, P.A.M., 1958, The Principles of Quantum Theory, 4th Ed. (Clarendon Press, Oxford) p. 9.
Good Jr, R.H., 1957, Particle aspect of the electromagnetic field equations, Phys. Rev. 105, 1914.
Gottfried, K., 1966, Quantum Mechanics (Benjamin, New York) p. 252.
Gross, L., 1964, Norm invariance of mass-zero equations under the conformal group, J. Math. Phys.
5, 687.
Haag, R., 1993, Local Quantum Physics (Springer, Berlin) p. 33.
Inagaki, T., 1994, Quantum-mechanical approach to a free photon, Phys. Rev. A 49, 2839.
Itzykson, C., and J.-B. Zuber, 1980, Quantum Field Theory (McGraw-Hill, New York) p. 50.
Jackson, J.D., 1975, Classical Electrodynamics (Wiley, New York) p. 552.
VI REFERENCES 293

Jauch, J.M., and C. Piron, 1967, Generalized localizability, Helv. Phys. Acta 40, 559.
Kaiser, G., 1992, Wavelet electrodynamics, Phys. Lett. A 168, 28.
Kidd, R., J. Ardini and A. Anton, 1989, Evolution of the modem photon, Am. J. Phys. 57, 27.
Kramers, H.A., 1938, Quantentheorie des Elektrons und der Strahlung, in: Hand- und Jahrbuch
der Chemischen Physik (Eucken-Wolf, Leipzig) [English translation: 1957, Quantum Mechanics
(North-Holland, Amsterdam)].
Landau, L.D., and R. Peierls, 1930, Quantenelektrodynamik im Konfigurationsraum, 2. Phys. 62,
188.
Laporte, O., and G.E. Uhlenbeck, 1931, Application of spinor analysis to the Maxwell and Dirac
equations, Phys. Rev. 37, 1380.
Lipkin, H.J., 1973, Quantum Mechanics (North-Holland, Amsterdam) ch. I .
Lomont, J.S., and H.E. Moses, 1962, Simple realizations of the infinitesimal generators of the proper
orthochronous inhomogeneous Lorentz group for zero mass, J. Math. Phys. 3, 405
London, F., 1927, Winkelvariable und kanonische Transfornationen in der Undulationsmechanik,
Z. Phys. 40, 193.
Louisell, W.H., 1973, Quantum Statistical Properties of Radiation (Wiley, New York).
Madelung, E., 1926, Quantentheorie in Hydrodynamischer Form, 2. Phys. 40, 322.
Mandel, L., and E. Wolf, 1995, Optical coherence and quantum optics (Cambridge University Press,
Cambridge).
Messiah, A,, 1961, Quantum Mechanics, Vol. I1 (North-Holland, Amsterdam) p. 1034.
Mignani, R., E. Recami and M. Baldo, 1974, About a Dirac-like equation for the photon according
to Ettore Majorana, Lett. Nuovo Cimento 11, 568.
Misner, C.W., and J.A. Wheeler, 1957, Classical physics as geometry: Gravitation, electromagnetism,
unquantized charge, and mass as properties of curved empty space, Ann. Phys. 2, 525.
Moliire, G., 1949, Laufende elektromagnetische Multipolwellen und eine neue Methode der Feld-
Quantisierung, Ann. Phys. 6, 146.
Moses, H.E., 1959, Solution of Maxwell’s equations in terms of a spinor notation: the direct and
inverse problem, Phys. Rev. 113, 1670.
Moses, H.E., 1973, Photon wave functions and the exact electromagnetic matrix elements for
hydrogenic atoms, Phys. Rev. A 8, 1710.
Newton, T.D., and E.P. Wigner, 1949, Localized states for elementary systems, Rev. Mod. Phys. 21,
400.
Ohmura, T., 1956, A new formulation on the electromagnetic field, Prog. Theor. Phys. 16, 684.
Oppenheimer, J.R., 1931, Note on light quanta and the electromagnetic field, Phys. Rev. 38, 725.
Pauli, W., 1933, Prinzipien der Quantentheorie, in: Handbuch der Physik, Vol. 24 (Springer, Berlin)
[English translation: 1980, General Principles of Quantum Mechanics (Springer, Berlin)] p. 189.
Pegg, D.T., and S.M. Bamett, 1988, Unitary phase operator in quantum mechanics, Europhys. Lett.
6, 483.
Penrose, R., and W. Rindler, 1984, Spinors and Space-Time, Vol. I (Cambridge University Press,
Cambridge) ch. 5.
Plebanski, J., 1960, Electromagnetic waves in gravitational fields, Phys. Rev. 118, 1396.
Power, E.A., 1964, Introductory Quantum Electrodynamics (Longmans, London) p. 61.
Pryce, M.H.L., 1948, The mass-centre in the restricted theory of relativity and its connection with
the quantum theory of elementary particles, Proc. R. SOC.(London) A 195, 62.
Rosewame, D., and S. Sarkar, 1992, Rigorous theory of photon localizability, Quantum Opt. 4,405.
Schiff, L.I., 1968, Quantum Mechanics (McGraw-Hill, New York) p. 203.
Schleich, W., and M.O. Scully, 1984, General relativity and modem optics in: New Trends in Atomic
Physics, Les Houches, Session XXXVIII, eds G. Grynberg and R. Stora (Elsevier, Amsterdam)
p. 995.
294 PHOTON WAVE FUNCTION [V

Schweber, S.S., 1961, An Introduction to Relativistic Quantum Field Theory (Row, Peterson, and
Co., Evanston, IL).
Segal, I.E., 1963, Mathematical Problems of Relativistic Physics (American Mathematical Society,
Providence, RI).
Silberstein, L., 1907a, Elektromagnetische Grundgleichungen in bivectorieller Behandlung, Ann.
Phys. 22, 579.
Silberstein, L., 1907b, Nachtrag zur Abhandlung uber “Elektromagnetische Grundgleichungen in
bivectorieller Behandlung”, Ann. Phys. 24, 783.
Silberstein, L., 1914, The Theory of Relativity (MacMillan, London). 2nd enlarged Ed.: 1924.
Sipe, J.F., 1995, Photon wave functions, Phys. Rev. A 52, 1875.
Skrotskii, G.B., 1957, The influence of gravitation on the propagation of light, Sov. Phys. Dokl. 2,
226.
Staruszkiewicz, A,, 1973, On affine properties of the light cone and their application in quantum
electrodynamics, Acta Phys. Polon. B 4, 57.
Streater, R.F., and A S . Wightman, 1978, PCT, Spin and Statistics, and All That (Benjamin,
Reading, MA).
Sudarshan, E.C.G., 1979, Pencils of rays in wave optics, Phys. Lett. A 73, 269.
Sudarshan, E.C.G., 1981a, Quantum electrodynamics and light rays, Physica A 96, 315.
Sudarshan, E.C.G., 1981b, Quantum theory of radiative transfer, Phys. Rev. A 23, 2803.
Walther, A,, 1968, Radiometry and coherence, J. Opt. SOC.Am. 68, 1256.
Weber, H., 1901, Die partiellen Differential-Gleichungen der mathematischen Physik nach Riemann’s
Vorlesungen (Friedrich Vieweg und Sohn, Braunschweig) p. 348.
Weinberg, S., 1995, The quantum theory of fields, Vol. I (Cambridge University, Cambridge) p. 61.
Weinberg, S., and E. Witten, 1980, Limits on massless particles, Phys. Lett. B 96, 59.
Weyl, H., 1929, Elektron und Gravitation, Z. Phys. 56, 330.
Wightman, A.S., 1962, On the localizability of quantum mechanical systems, Rev. Mod. Phys. 34,
845.
Wigner, E.P., 1932, On the quantum correction for thermodynamic equilibrium, Phys. Rev. 40, 749.
Wigner, E.P., 1939, On unitary representations of the inhomogeneous Lorentz group, Ann. Math. 40,
39. Reprinted: 1966, Symmetry Groups in Nuclear and Particle Physics, ed. F.J. Dyson (Benjamin,
New York) p. 120.
Wolf, E., 1976, New theory of radiative energy transfer in free electromagnetic field, Phys. Rev. D
13, 869.
Zeldovich, Ya.B., 1965, Number of quanta as an invariant of the classical electromagnetic field,
Dokl. Acad. Sci. USSR, 163, 1359. In Russian.