SPRINGER BRIEFS IN PHYSICS

Kimball A. Milton

Schwinger’s Quantum
Action Principle
From Dirac’s Formulation
Through Feynman’s Path
Integrals, the Schwinger-
Keldysh Method,
Quantum Field Theory, to
Source Theory

123
SpringerBriefs in Physics

Editorial Board
Egor Babaev, University of Massachusetts, Massachusetts, USA
Malcolm Bremer, University of Bristol, Bristol, UK
Xavier Calmet, University of Sussex, Sussex, UK
Francesca Di Lodovico, Queen Mary University of London, London, UK
Maarten Hoogerland, University of Auckland, Auckland, New Zealand
Eric Le Ru, Victoria University of Wellington, Wellington, New Zealand
Hans-Joachim Lewerenz, California Institute of Technology, Pasadena, USA
James Overduin, Towson University, Towson, USA
Vesselin Petkov, Concordia University, Concordia, Canada
Charles H.-T. Wang, University of Aberdeen, Aberdeen, UK
Andrew Whitaker, Queen’s University Belfast, Belfast, UK
More information about this series at http://www.springer.com/series/8902
Kimball A. Milton

Schwinger’s Quantum
Action Principle
From Dirac’s Formulation Through
Feynman’s Path Integrals,
the Schwinger-Keldysh Method,
Quantum Field Theory, to Source Theory

123
Kimball A. Milton
Homer L. Dodge Department of Physics and
Astronomy
The University of Oklahoma
Norman
USA

ISSN 2191-5423 ISSN 2191-5431 (electronic)

SpringerBriefs in Physics
ISBN 978-3-319-20127-6 ISBN 978-3-319-20128-3 (eBook)
DOI 10.1007/978-3-319-20128-3

Library of Congress Control Number: 2015941137

Springer Cham Heidelberg New York Dordrecht London

© The Author(s) 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made.

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media

(www.springer.com)
Abstract

Starting from the earlier notions of stationary action principles, we show how Julian
Schwinger’s Quantum Action Principle descended from Dirac’s formulation, which
independently led Feynman to his path-integral formulation of quantum mechanics.
The connection between the two is brought out, and applications are discussed. The
Keldysh-Schwinger time-cycle method of extracting matrix elements in nonequi-
librium situations is described. The variational formulation of quantum field theory
and the development of source theory constitute the latter part of this work. In this
document, derived from Schwinger’s lectures over four decades, the continuity of
concepts, such as that of Green’s functions, becomes apparent.

v
Contents

1 Historical Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Review of Classical Action Principles. . . . . . . . . . . . . . . . . . . . . . . 3

2.1 Lagrangian Viewpoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Hamiltonian Viewpoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 A Third, Schwingerian, Viewpoint . . . . . . . . . . . . . . . . . . . . . . 7
2.4 Invariance and Conservation Laws . . . . . . . . . . . . . . . . . . . . . . 9
2.5 Nonconservation Laws: The Virial Theorem . . . . . . . . . . . . . . . 13

3 Classical Field Theory—Electrodynamics. . . . . . . . . . . . . . . . . . . . 15

3.1 Action of Particle in Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Electrodynamic Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Momentum and Angular Momentum Conservation. . . . . . . . . . . 21
3.5 Gauge Invariance and the Conservation of Charge . . . . . . . . . . . 24
3.6 Gauge Invariance and Local Conservation Laws . . . . . . . . . . . . 26

4 Quantum Action Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4.1 Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Forced Harmonic Oscillator. . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3 Feynman Path Integral Formulation . . . . . . . . . . . . . . . . . . . . . 44
4.3.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.4 Toward Source Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

5 Time-Cycle or Schwinger-Keldysh Formulation . . . . . . . . . . . . . . . 51

6 Relativistic Theory of Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

6.1 Inference of Particle Properties . . . . . . . . . . . . . . . . . . . . . . . . 70
6.2 The Connection Between Spin and Statistics . . . . . . . . . . . . . . . 72
6.3 Fermi-Dirac Fields of Spin 1=2 . . . . . . . . . . . . . . . . . . . . . . . . 74
6.4 Spin Zero and One. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

vii
viii Contents

6.5 Spin Zero . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

6.6 Spin One . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.7 Electromagnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.8 Introduction of Charge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.9 Quantum Electrodynamics. . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

7 Nonrelativistic Source Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

7.1 Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.2 Bound States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

8 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Chapter 1
Historical Introduction

Variational principles for dynamical systems have a long history. Although precursors
go back at least to Leibnitz (see for example (Euler 1752)) and Euler (1744) the
“principle of least action” was given modern form by de Maupertuis (1744, 1746).
We will not attempt to trace the history here; a brief useful account is given in
Sommerfeld’s lectures (Sommerfeld 1964). The most important names in the history
of the development of dynamical systems, or at least those that will bear most directly
on the following discussion, are those of Lagrange (1788) and Hamilton (1834, 1835).
Here we are concentrating on the work of Julian Schwinger (1918–1994), who
had profound and pervasive influence on 20th century physics, and whose many
students have become leaders in diverse fields.1 For biographical information about
his life and work see (Mehra 2000; Milton 2007). Therefore, we will take up the
story in the modern era. Shortly after Dirac’s work with Fock and Podolsky (Dirac
1932), in which the demonstration of the equivalence between his theory of quantum
electrodynamics, and that of Heisenberg and Pauli (Heisenberg 1929), P. A. M.
Dirac wrote a paper on “The Lagrangian in Quantum Mechanics” (Dirac 1933).
This paper had a profound influence on Richard Feynman’s doctoral dissertation at
Princeton on “The Principles of Least Action in Quantum Mechanics” (Feynman
1942), and on his later work on the formulations of the “Space-Time Approach to
Quantum Electrodynamics” (Feynman 1949). Dirac’s paper further formed the basis
for Schwinger’s development of the quantum action principle, which first appeared
in his final operator field formulation of quantum field theory (Schwinger 1951b),
which we will describe in Chap. 6.
The response of Feynman and Schwinger to Dirac’s inspiring paper was
completely different. Feynman was to give a global “solution” to the problem of

1 For complex reasons, Schwinger’s influence on modern physics is not widely appreciated. His

contributions to our current understanding of nature are underrepresented in textbooks, with some
notable exceptions (Toms 2007).
© The Author(s) 2015 1
K.A. Milton, Schwinger’s Quantum Action Principle,
SpringerBriefs in Physics, DOI 10.1007/978-3-319-20128-3_1
2 1 Historical Introduction

determining the transformation function, the probability amplitude connecting the

state of the system at one time to that at a later time, in terms of a sum over classical
trajectories, the famous path integral. Schwinger, instead, derived (initially postu-
lated) a differential equation for that transformation function in terms of a quantum
action functional. This differential equation possessed Feynman’s path integral as a
formal solution, which remained poorly defined; but Schwinger believed throughout
his life that his approach was “more general, more elegant, more useful, and more
tied to the historical line of development as the quantum transcription of Hamilton’s
action principle” (Schwinger 1973).
Later, in a tribute to Feynman, Schwinger commented further. Dirac, of course,
was the father of transformation theory (Dirac 1927). The transformation function
from a description of the system at time t2 to a description at time t1 is “the product of
all the transformations functions associated with the successive infinitesimal incre-
ments in time.” Dirac said the latter, that is, the transformation function from time t
to time t + dt corresponds to exp[(i/)dt L], where L is the Lagrangian expressed
in terms of the coordinates at the two times. For the transformation
t function between
t2 and t1 “the integrand is exp[(i/)W ]. where W = t21 dt L.” “Now we know, and
Dirac surely knew, that to within a constant factor the ‘correspondence,’ for infin-
itesimal dt, is an equality when we deal with a system of nonrelativistic particles
possessing a coordinate-dependent potential energy V …. Why then, did Dirac not
make a more precise, if less general statement? Because he was interested in a gen-
eral question: What, in quantum mechanics, corresponds to the classical principle of
stationary action?”
“Why, in the decade that followed, didn’t someone pick up the computational
possibilities offered by this integral approach to the time transformation function?
To answer this question bluntly, perhaps no one needed it—until Feynman came
along.” (Schwinger 1989a).
But Schwinger followed the differential route, and starting in early 1950 began
a new, his third, formulation of quantum electrodynamics, based on a variational
approach. This was first published in 1951 (Schwinger 1951b). A bit later he started
developing a new formulation of quantum kinematics, which he called Measurement
Algebra, which got its first public presentation at École de Physique des Houches
in the summer of 1955. There were several short notes in the Proceedings of the
US National Academy published in 1960, explaining both the quantum kinematical
approach and the dynamical action principle (Schwinger 1960a–d), but although
he often promised to write a book on the subject (as he also promised a book on
quantum field theory) nothing came of it. Les Houches lectures, based on notes
taken by Robert Kohler, eventually appeared in 1970 (Schwinger 1970a). Lectures
based on a UCLA course on quantum mechanics by Schwinger were eventually
published under Englert’s editorship (Schwinger 2001). The incompleteness of the
written record may be partly alleviated by the present essay.
We start on a classical footing.
Chapter 2
Review of Classical Action Principles

This section grew out of lectures given by Schwinger at UCLA around 1974,
which were substantially transformed into Chap. 8 of Classical Electrodynamics
(Schwinger 1998). (Remarkably, considering his work on waveguide theory during
World War II, now partially recorded in Ref. (Milton 2006), he never gave lectures
on this subject at Harvard after 1947.)
We start by reviewing and generalizing the Lagrange-Hamilton principle for a
single particle. The action, W12 , is defined as the time integral of the Lagrangian,
L, where the integration extends from an initial configuration or state at time t2 to a
final state at time t1 :  t1
W12 = dt L . (2.1)
t2

The integral refers to any path, any line of time development, from the initial to
the final state, as shown in Fig. 2.1. The actual time evolution of the system is
selected by the principle of stationary action: In response to infinitesimal variations
of the integration path, the action W12 is stationary—does not have a corresponding
infinitesimal change—for variations about the correct path, provided the initial and
final configurations are held fixed,

δW12 = 0. (2.2)

This means that, if we allow infinitesimal changes at the initial and final times,
including alterations of those times, the only contribution to δW12 then comes from
the endpoint variations, or
δW12 = G 1 − G 2 , (2.3)

where G a , a = 1 or 2, is a function, called the generator, depending on dynamical

variables only at time ta . In the following, we will consider three different realizations

© The Author(s) 2015 3

K.A. Milton, Schwinger’s Quantum Action Principle,
SpringerBriefs in Physics, DOI 10.1007/978-3-319-20128-3_2
4 2 Review of Classical Action Principles

Fig. 2.1 A possible path

from initial state to final state • t1

time ↑

• t2
state variables →

of the action principle, where, for simplicity, we will restrict our attention to a single
particle.

2.1 Lagrangian Viewpoint

The nonrelativistic motion of a particle of mass m moving in a potential V (r, t) is

described by the Lagrangian
 2
1 dr
L= m − V (r, t). (2.4)
2 dt

Here, the independent variables are r and t, so that two kinds of variations can be
considered. First, a particular motion is altered infinitesimally, that is, the path is
changed by an amount δr:
r(t) → r(t) + δr(t). (2.5)

Second, the final and initial times can be altered infinitesimally, by δt1 and δt2 ,
respectively. It is more convenient, however, to think of these time displacements as
produced by a continuous variation of the time parameter, δt (t),

t → t + δt (t), (2.6)

so chosen that, at the endpoints,

δt (t1 ) = δt1 , δt (t2 ) = δt2 . (2.7)

The corresponding change in the time differential is

 
dδt
dt → d(t + δt) = 1 + dt, (2.8)
dt
2.1 Lagrangian Viewpoint 5

which implies the transformation of the time derivative,

 
d dδt d
→ 1− . (2.9)
dt dt dt

Because of this redefinition of the time variable, the limits of integration in the action,
 1 1 (dr)2

W12 = m − dt V , (2.10)
2 2 dt

are not changed, the time displacement being produced through δt (t) subject to (2.7).
The resulting variation in the action is now
    2

1 ∂
· ·
dr d dδt 1 dr
δW12 = dt m δr − δr ∇V − m + V − δt V
2 dt dt dt 2 dt ∂t
 1     2 
=
2
dt
d
dt
m
dr
dt
·
δr −
1
2
m
dr
dt
+ V δt
     2 

d2 ∂
·
+ δr −m 2 r − ∇V + δt
dt
d 1
dt 2
m
dr
dt
+V − V
∂t
,

(2.11)

where, in the last form, we have integrated by parts in order to isolate δr and δt.
Because δr and δt are independent variations, the principle of stationary action
implies that the actual motion is governed by

d2
m r = − ∇V, (2.12a)
dt 2
  2 
d 1 dr ∂
m +V = V, (2.12b)
dt 2 dt ∂t

while the total time derivative gives the change at the endpoints,

·
G = p δr − Eδt, (2.12c)

with
 2
dr 1 dr
momentum = p = m , energy = E = m + V. (2.12d)
dt 2 dt

Therefore, we have derived Newton’s second law [the equation of motion in second-
order form], (2.12a), and, for a static potential, ∂V /∂t = 0, the conservation of
energy, (2.12b). The significance of (2.12c) will be discussed later in Sect. 2.4.
6 2 Review of Classical Action Principles

2.2 Hamiltonian Viewpoint

Using the above definition of the momentum, we can rewrite the Lagrangian as

L=p · dr
dt
− H (r, p, t), (2.13)

where we have introduced the Hamiltonian

p2
H= + V (r, t). (2.14)
2m
We are here to regard r, p, and t as independent variables in
 1
W12 =
2
·
[p dr − dt H ]. (2.15)

The change in the action, when r, p, and t are all varied, is

  
1 ∂H ∂H ∂H
· · · ·
d dr dδt
δW12 = δr − δr
dt p + δp − δp − H − δt
2 dt ∂r dt ∂p dt ∂t
 1   
dp ∂ H
· ·
d
= dt (p δr − H δt) + δr − −
2 dt dt ∂r
   
dr ∂ H ∂H
+ δp
dt
·−
∂p
+ δt
dH
dt
−
∂t
. (2.16)

The action principle then implies

dr ∂ H p
= = , (2.17a)
dt ∂p m
dp ∂H
=− = −∇V, (2.17b)
dt ∂r
dH ∂H
= , (2.17c)
dt ∂t

·
G = p δr − H δt. (2.17d)

In contrast with the Lagrangian differential equations of motion, which involve sec-
ond derivatives, these Hamiltonian equations contain only first derivatives; they are
called first-order equations. They describe the same physical system, because when
(2.17a) is substituted into (2.17b), we recover the Lagrangian-Newtonian equation
(2.12a). Furthermore, if we insert (2.17a) into the Hamiltonian (2.14), we identify
H with E. The third equation (2.17c) is then identical with (2.12b). We also note the
equivalence of the two versions of G.
2.2 Hamiltonian Viewpoint 7

But probably the most direct way of seeing that the same physical system is
involved comes by writing the Lagrangian in the Hamiltonian viewpoint as
 2  
m dr 1 dr 2
L= −V − p−m . (2.18)
2 dt 2m dt

The result of varying p in the stationary action principle is to produce

dr
p=m . (2.19)
dt
But, if we accept this as the definition of p, the corresponding term in L disappears
and we explicitly regain the Lagrangian description. We are justified in completely
omitting the last term on the right side of (2.18), despite its dependence on the
variables r and t, because of its quadratic structure. Its explicit contribution to δL is
   
−
1
m
p−m
dr
dt
· d
δp − m δr + m
dt
dr dδt
dt dt
, (2.20)

and the equation supplied by the stationary action principle for p variations, (2.19),
also guarantees that there is no contribution here to the results of r and t variations.

2.3 A Third, Schwingerian, Viewpoint

Here we take r, p, and the velocity, v, as independent variables, so that the Lagrangian
is written in the form
 
L=p · dr
dt
1
− v + mv 2 − V (r, t) ≡ p
2
dr
dt
·
− H (r, p, v, t), (2.21)

where
1
·
H (r, p, v, t) = p v − mv 2 + V (r, t).
2
(2.22)

The variation of the action is now

 1
δW12 =δ
2
·
[p dr − H dt]
 
1 ∂H ∂H ∂H
· · · · ·
dr d
= dt δp +p δr − δr − δp − δv
2 dt dt ∂r ∂p ∂v

∂H dδt
− δt −H
∂t dt
8 2 Review of Classical Action Principles
   
1 dp ∂ H
=
2
dt
d
dt
·
(p δr − H δt) − δr
dt
+
∂r
·
   
dr ∂ H ∂H ∂H
· ·
dH
+ δp − − δv + δt − , (2.23)
dt ∂p ∂v dt ∂t

so that the action principle implies

dp ∂H
=− = −∇V, (2.24a)
dt ∂r
dr ∂ H
= = v, (2.24b)
dt ∂p
∂H
0= − = −p + mv, (2.24c)
∂v
dH ∂H
= , (2.24d)
dt ∂t

·
G = p δr − H δt. (2.24e)

Notice that there is no equation of motion for v since dv/dt does not occur in the
Lagrangian, nor is it multiplied by a time derivative. Consequently, (2.24c) refers to
a single time and is an equation of constraint.
From this third approach, we have the option of returning to either of the other
two viewpoints by imposing an appropriate restriction. Thus, if we write (2.22) as

p2 1
H (r, p, v, t) = + V (r, t) − (p − mv)2 , (2.25)
2m 2m
and we adopt
1
v= p (2.26)
m
as the definition of v, we recover the Hamiltonian description, (2.13) and (2.14).
Alternatively, we can present the Lagrangian (2.21) as
 2    2
L=
m
2
dr
dt
− V + (p − mv) · dr
dt
−v −
m dr
2 dt
−v . (2.27)

Then, if we adopt the following as definitions,

dr
v= , p = mv, (2.28)
dt
the resultant form of L is that of the Lagrangian viewpoint, (2.4). It might seem that
only the definition v = dr/dt, inserted in (2.27), suffices to regain the Lagrangian
description. But then the next to last term in (2.27) would give the following additional
2.3 A Third, Schwingerian, Viewpoint 9

contribution to δL, associated with the variation δr:

(p − mv) · dtd δr. (2.29)

In the next Chapter, where the action formulation of electrodynamics is consid-

ered, we will see the advantage of adopting this third approach, which is characterized
by the introduction of additional variables, similar to v, for which there are no equa-
tions of motion.

2.4 Invariance and Conservation Laws

There is more content to the principle of stationary action than equations of motion.
Suppose one considers a variation such that

δW12 = 0, (2.30)

independently of the choice of initial and final times. We say that the action, which is
left unchanged, is invariant under this alteration of path. Then the stationary action
principle (2.3) asserts that
δW12 = G 1 − G 2 = 0, (2.31)

or, there is a quantity G(t) that has the same value for any choice of time t; it is
conserved in time. A differential statement of that is
d
G(t) = 0. (2.32)
dt
The G functions, which are usually referred to as generators, express the interrelation
between conservation laws and invariances of the system.
Invariance implies conservation, and vice versa. A more precise statement is the
following:
If there is a conservation law, the action is stationary under an infinitesimal transformation
in an appropriate variable.

The converse of this statement is also true.

If the action W is invariant under an infinitesimal transformation (that is, δW = 0), then
there is a corresponding conservation law.

This is the celebrated theorem of Amalie Emmy Noether (Noether 1918).

Here are some examples. Suppose the Hamiltonian of (2.13) does not depend
explicitly on time, or
 1
W12 = ·
[p dr − H (r, p)dt].
2
(2.33)
10 2 Review of Classical Action Principles

Fig. 2.2 δω × r is δω × r
perpendicular to δω and r,
δω
and represents an r
infinitesimal rotation of r
about the δω axis

Then the variation (which as a rigid displacement in time, amounts to a shift in the
time origin)
δt = constant (2.34)

will give δW12 = 0 [see the first line of (2.16), with δr = 0, δp = 0, dδt/dt = 0,
∂ H/∂t = 0]. The conclusion is that G in (2.17d), which here is just

G t = −H δt, (2.35)

is a conserved quantity, or that

dH
= 0. (2.36)
dt
This inference, that the Hamiltonian—the energy—is conserved, if there is no explicit
time dependence in H , is already present in (2.17c). But now a more general principle
is at work.
Next, consider an infinitesimal, rigid rotation, one that maintains the lengths and
scalar products of all vectors. Written explicitly for the position vector r, it is

δr = δω × r, (2.37)

where the constant vector δω gives the direction and magnitude of the rotation (see
Fig. 2.2). Now specialize (2.14) to

p2
H= + V (r ), (2.38)
2m
where r = |r|, a rotationally invariant structure. Then
 1
W12 =
2
·
[p dr − H dt] (2.39)

is also invariant under the rigid rotation, implying the conservation of

2.4 Invariance and Conservation Laws 11

·
G δω = p δr = δω r × p. · (2.40)

This is the conservation of angular momentum,

d
L = r × p, L = 0. (2.41)
dt
Of course, this is also contained within the equation of motion,

d ∂V
L = −r × ∇V = −r×r̂ = 0, (2.42)
dt ∂r
since V depends only on |r|.
Conservation of linear momentum appears analogously when there is invariance
under a rigid translation. For a single particle, (2.17b) tells us immediately that p is
conserved if V is a constant, say zero. Then, indeed, the action
 1 p2

W12 =
2
·
p dr −
2m
dt (2.43)

is invariant under the displacement

δr = δ = constant, (2.44)

and
G δ = p δ · (2.45)

is conserved. But the general principle acts just as easily for, say, a system of two
particles, a and b, with Hamiltonian

pa2 p2
H= + b + V (ra − rb ). (2.46)
2m a 2m b

This Hamiltonian and the associated action

 1
W12 =
2
· ·
[pa dra + pb drb − H dt] (2.47)

are invariant under the rigid translation

δra = δrb = δ, (2.48)

with the implication that

· ·
G δ = pa δra + pb δrb = (pa + pb ) δ · (2.49)
12 2 Review of Classical Action Principles

is conserved. This is the conservation of the total linear momentum,

d
P = pa + pb , P = 0. (2.50)
dt
Something a bit more general appears when we consider a rigid translation that
grows linearly in time:
δra = δrb = δv t, (2.51)

using the example of two particles. This gives each particle the common additional
velocity δv, and therefore must also change their momenta,

δpa = m a δv, δpb = m b δv. (2.52)

The response of the action (2.47) to this variation is

 1
δW12 =
2
· ·
[(pa + pb ) δv dt + δv (m a dra + m b drb ) − (pa + pb ) δv dt]·
 1
=
2
·
d[(m a ra + m b rb ) δv]. (2.53)

The action is not invariant; its variation has end-point contributions. But there is still
·
a conservation law, not of G = P δvt, but of N δv, where ·
N = Pt − (m a ra + m b rb ). (2.54)

Written in terms of the center-of-mass position vector

m a ra + m b rb
R= , M = ma + mb, (2.55)
M
the statement of conservation of

N = Pt − MR, (2.56)

namely
dN dR
0= =P−M , (2.57)
dt dt
is the familiar fact that the center of mass of an isolated system moves at the constant
velocity given by the ratio of the total momentum to the total mass of that system.
2.5 Nonconservation Laws: The Virial Theorem 13

2.5 Nonconservation Laws: The Virial Theorem

The action principle also supplies useful nonconservation laws. Consider, for constant
δλ,
δr = δλr, δp = −δλp, (2.58)

·
which leaves p dr invariant,

· ·
δ(p dr) = (−δλp) dr + p (δλdr) = 0. · (2.59)

But the response of the Hamiltonian

p2
H = T ( p) + V (r), T ( p) = , (2.60)
2m
is given by the noninvariant form

δ H = δλ(−2T + r ∇V ). · (2.61)

Therefore we have, for an arbitrary time interval, for the variation of the action (2.15),
 1  1
· ·
d
δW12 = dt[δλ(2T − r ∇V )] = G 1 − G 2 = dt (p δλr) (2.62)
2 2 dt

or, the theorem

d
dt
·
r p = 2T − r ∇V. · (2.63)

For the particular situation of the Coulomb potential between charges, V =

constant/r , where
·
d
r ∇V = r V = −V, (2.64)
dr
the virial theorem asserts that

·
d
(r p) = 2T + V. (2.65)
dt
We apply this to a bound system produced by a force of attraction. On taking the
time average of (2.65) the time derivative term disappears. That is because, over an
·
arbitrarily long time interval τ = t1 − t2 , the value of r p(t1 ) can differ by only a
·
finite amount from r p(t2 ), and

d
(r p) ≡·
1 t1
dt
d
r p=· ·
r p(t1 ) − r p(t2 ) ·
→ 0, (2.66)
dt τ t2 dt τ
14 2 Review of Classical Action Principles

as τ → ∞. The conclusion, for time averages,

2T = −V , (2.67)

is familiar in elementary discussions of motion in a 1/r potential.

Here is one more example of a nonconservation law: Consider the variations
r
δr =δλ , (2.68a)
r  
δp = − δλ
p rp r
− = δλ
·r × (r × p)
. (2.68b)
r r 3 r3

·
Again p dr is invariant:
   
·
δ(p dr) = −δλ
p rp r
−
· · dr + p · δλ
dr r dr
− δλr 3
·= 0, (2.69)
r r3 r r

and the change of the Hamiltonian (2.60) is now

 
L2
·
r
δ H = δλ − 3 + ∇V . (2.70)
mr r

The resulting theorem, for V = V (r ), is

d r  L2
dt r
·
p =
mr 3
−
dV
dr
, (2.71)

which, when applied to the Coulomb potential, gives the bound-state time average
relation    
L2 1 V
3
=− . (2.72)
m r r

This relation is significant in hydrogen fine-structure calculations (for example, see

(Schwinger 2001)).
Chapter 3
Classical Field Theory—Electrodynamics

This chapter again grew out of Schwinger’s UCLA lectures. These evolved, tortur-
ously, into Chap. 9 of Classical Electrodynamics (Schwinger 1998). Here we use
Gaussian units.

3.1 Action of Particle in Field

It was stated in our review of mechanical action principles in the previous chapter that
the third viewpoint, which employs the variables r, p, and v, was particularly conve-
nient for describing electromagnetic forces on charged particles. With the explicit,
and linear, appearance of v in what plays the role of the potential function when
magnetic fields are present, we begin to see the basis for that remark. Indeed, we
have only to consult (2.21) to find the appropriate Lagrangian:
 
· ·
dr 1 e
L=p − v + mv 2 − eφ + v A, (3.1)
dt 2 c

where φ and A are the scalar and vector potentials, respectively. To recapitulate, the
equations resulting from variations of p, r, and v are, respectively,

dr
= v, (3.2a)
dt  
·
d 1
p = − e∇ φ − v A , (3.2b)
dt c
e
p = mv + A. (3.2c)
c

© The Author(s) 2015 15

K.A. Milton, Schwinger’s Quantum Action Principle,
SpringerBriefs in Physics, DOI 10.1007/978-3-319-20128-3_3
16 3 Classical Field Theory—Electrodynamics

We can now move to either the Lagrangian or the Hamiltonian formulation. For the
first, we simply adopt v = dr/dt as a definition (but see the discussion in Sect. 2.3)
and get
 2
·
1 dr e dr
L= m − eφ + A. (3.3)
2 dt c dt

Alternatively, we use (3.2c) to define

1  e 
v= p− A , (3.4)
m c
and find

L =p · dr
dt
− H, (3.5a)
1  e 2
H= p − A + eφ. (3.5b)
2m c

3.2 Electrodynamic Action

The electromagnetic field is a mechanical system. It contributes its variables to the

action, to the Lagrangian of the whole system of charges and fields. In contrast with
the point charges, the field is distributed in space. Its Lagrangian should therefore
be, not a summation over discrete points, but an integration over all spatial volume
elements, 
L field = (dr) Lfield ; (3.6)

this introduces the Lagrange function, or Lagrangian density, L. The total Lagrangian
must be the sum of the particle part, (3.1), and the field part, (3.6), where the latter
must be chosen so as to give the Maxwell equations, in Gaussian units:

1 ∂
·
4π
∇×B= E+ j, ∇ E = 4πρ, (3.7a)
c ∂t c
1 ∂
−∇ × E =
c ∂t
B, ∇ B = 0. · (3.7b)

The homogeneous equations here are equivalent to the construction of the electro-
magnetic field in term of potentials, or,

1 ∂
A = − E − ∇φ, (3.8a)
c ∂t
B = ∇ × A. (3.8b)
3.2 Electrodynamic Action 17

Thus, we recognize that A(r, t), E(r, t), in analogy with r(t), p(t), obey equations
of motion while φ(r, t), B(r, t), as analogues of v(t), do not. There are enough
clues here to give the structure of Lfield , apart from an overall factor. The anticipated
complete Lagrangian for microscopic electrodynamics is

   
L= pa ·dra
dt
1 ea
− va + m a va2 − ea φ(ra ) + va
2 c
· A(ra )
a
    
1 ∂
· ·
1 1 2
+ (dr) E − A − ∇φ − B ∇ × A + (B − E 2 ) . (3.9)
4π c ∂t 2

The terms that are summed in (3.9) describe the behavior of charged particles
under the influence of the fields, while the terms that are integrated describe the field
behavior. The independent variables are

ra (t), va (t), pa (t), φ(r, t), A(r, t), E(r, t), B(r, t), t. (3.10)

We now look at the response of the Lagrangian to variations in each of these variables
separately, starting with the particle part:
  
· · ·
d dpa va
δra : δL = (δra pa ) + δra − − ∇ a ea φ(ra ) − A(ra ) ,
dt dt c
(3.11a)


·
ea
δva : δL = δva −pa + m a va + A(ra ) , (3.11b)
  c
δL = δpa ·
dra
δpa : − va . (3.11c)
dt

The stationary action principle now implies the equations of motion

 
dpa
dt
= − ea ∇ a φ(ra ) −
va
c
·
A(ra ) , (3.12a)
ea
m a va = pa − A(ra ), (3.12b)
c
dra
va = , (3.12c)
dt
which are the known results, (3.2a)–(3.2c).
The real work now lies in deriving the equations of motion for the fields. In order
to cast all the field-dependent terms into integral form, we introduce charge and
current densities,
18 3 Classical Field Theory—Electrodynamics

ρ(r, t) = ea δ(r − ra (t)), (3.13a)
a

j(r, t) = ea va (t)δ(r − ra (t)), (3.13b)
a

so that

   
ea
· 1
−ea φ(ra ) + va A(ra ) = (dr) −ρφ + j A .
c c
· (3.14)
a

The volume integrals extend over sufficiently large regions to contain all the fields of
interest. Consequently, we can integrate by parts and ignore the surface terms. The
responses of the Lagrangian (3.9) to field variations, and the corresponding equations
of motion deduced from the action principle are

·
1
δφ : δL = (dr) δφ(∇ E − 4πρ), (3.15a)
4π

·
∇ E = 4πρ,

(3.15b)

δA : δL = −
1 d
4π c dt
(dr) δA E ·
  
1 ∂E 4π
+
1
4π
(dr) δA
c ∂t
+
c
·
j−∇×B , (3.15c)

1 ∂ 4π
∇×B= E+ j, (3.15d)
c ∂t c  

1 ∂
·
1
δE : δL = (dr) δE − A − ∇φ − E , (3.15e)
4π c ∂t
1 ∂
E= − A − ∇φ, (3.15f)
c∂t
·
1
δB : δL = (dr) δB (−∇ × A + B), (3.15g)
4π
B = ∇ × A. (3.15h)

We therefore recover Maxwell’s equations, two of which are implicit in the con-
struction of E and B in terms of potentials. By making a time variation of the action
[variations due to the time dependence of the fields vanish by virtue of the stationary
action principle—that is, they are already subsumed in Eqs. (3.15)],
  
d dH
δt : δW = dt (−H δt) + δt , (3.16)
dt dt

we identify the Hamiltonian of the system to be

3.2 Electrodynamic Action 19

  
·
ea 1
H= pa − A(ra ) va − m a va2 + ea φ(ra )
a
c 2
  
+
1
4π
· 1
·
(dr) E ∇φ + B ∇ × A + (E 2 − B 2 ) ,
2
(3.17)

which is a constant of the motion, d H/dt = 0. The generators are inferred from the
total time derivative terms in (3.11a), (3.15c), and (3.16),

δW12 = G 1 − G 2 , (3.18a)

to be

· ·
1
G= δra pa − (dr) E δA − H δt. (3.18b)
a
4π c

3.3 Energy

Notice that the total Lagrangian (3.9) can be presented as


L= pa · dra
dt
−
1
4π c
(dr) E · ∂t∂ A − H, (3.19)
a

where the Hamiltonian is given by (3.17). The narrower, Hamiltonian, description

is reached by eliminating all variables that do not obey equations of motion, and,
correspondingly, do not appear in G. Those “superfluous” variables are the va and
the fields φ and B, which are eliminated by using (3.12b), (3.15b), and (3.15h), the
equations without time derivatives, resulting, first, in the intermediate form

 1  ea 2    2
E + B2

H= pa − A(ra ) + ea φ(ra ) + (dr) − ρφ .
a
2m a c 8π
(3.20)
The first term here is the energy of the particles moving in the field [particle energy—
see (3.5b)], so we might call the second term the field energy. The ambiguity of these
terms (whether the potential energy of particles is attributed to them or to the fields,
or to both) is evident from the existence of a simpler form of the Hamiltonian

1  ea 2  E 2 + B2
H= pa − A(ra ) + (dr) , B = ∇ × A, (3.21)
a
2m a c 8π

where we have used the equivalence of the two terms involving φ, given in (3.14).
This apparently startling result suggests that the scalar potential has disappeared
from the dynamical description. But, in fact, it has not. If we vary the Lagrangian
(3.19), where H is given by (3.21), with respect to E we find
20 3 Classical Field Theory—Electrodynamics
  
1 ∂
·
1
δL = − (dr) δE A + E = 0. (3.22)
4π c ∂t

Do we conclude that 1c ∂t∂ A + E = 0? That would be true if the δE(r, t) were arbitrary.
They are not; E is subject to the restriction—the constraint—(3.15b), which means
that any change in E must obey
∇ δE = 0. · (3.23)

The proper conclusion is that the vector multiplying δE in (3.22) is the gradient of a
scalar function, just as in (3.15f),

1 ∂
A + E = −∇φ, (3.24)
c ∂t
for that leads to 
δL = −
1
4π
·
(dr) (∇ δE)φ = 0, (3.25)

as required.
The fact that the energy is conserved,

dH
= 0, (3.26)
dt
where 
1 E 2 + B2
H= m a va2 + (dr) U, U = , (3.27)
a
2 8π

is a simple sum of particle kinetic energy and integrated field energy density, can be
verified directly by taking the time derivative of (3.20). The time rate of change of
the particle energy is computed directly:
   
d 1 ∂
·
ea
m a va2 + ea φ(ra ) = ea φ(ra ) − va A(ra ) . (3.28)
dt a 2 a
∂t c

We can compute the time derivative of the field energy by using the equation of
energy conservation,  
d
dt
(dr) U = − (dr) j E, ·
(3.29)

to be
     
E 2 + B2 ∂ ∂
·
d
(dr) − ρφ = (dr) −j E − φ ρ − ρ φ
dt 8π ∂t ∂t
3.3 Energy 21
  
∂ 1 ∂
=− (dr) ρ φ − j
∂t c ∂t
A ·
∂ ∂

=− ea
∂t
1
φ(ra ) − va
c ∂t
·
A(ra ) .
a

(3.30)

Here we have used (3.15f), and have noted that

  
∂
·
(dr) j ∇φ − φ ρ = 0
∂t
(3.31)

by charge conservation. Observe that (3.28) and (3.30) are equal in magnitude and
opposite in sign, so that their sum is zero. This proves the statement of energy
conservation (3.26).

3.4 Momentum and Angular Momentum Conservation

The action principle not only provides us with the field equations, particle equations
of motion, and expressions for the energy, but also with the generators (3.18b).
The generators provide a connection between conservation laws and invariances
of the action (recall Sect. 2.4). Here we will further illustrate this connection by
deriving momentum and angular momentum conservation from the invariance of the
action under rigid coordinate translations and rotations, respectively. [In a similar
way we could derive energy conservation, (3.26), from the invariance under time
displacements—see also Sect. 3.6].
Under an infinitesimal rigid coordinate displacement, δ, a given point which is
described by r in the old coordinate system is described by r + δ in thenew one.
(See Fig. 3.1.) The response of the particle term in (3.18b) is simple: δ · a pa ; for
the field part, we require the change, δA, of the vector potential induced by the rigid

Fig. 3.1 Rigid coordinate new old

displacement, where the new
coordinate system is
displaced by a rigid
translation −δ relative to
the old coordinate system r
r+δ
δ
22 3 Classical Field Theory—Electrodynamics

coordinate displacement. The value of a field F at a physical point P is unchanged

under such a displacement, so that if r and r + δ are the coordinates of P in the two
frames, there are corresponding functions F and F such that

F(P) = F(r) = F(r + δ), (3.32)

that is, the new function F of the new coordinate equals the old function F of the
old coordinate. The change in the function F at the same coordinate is given by

F(r) = F(r) + δ F(r), (3.33)

so that
δ F(r) = F(r − δ) − F(r) = −δ ∇ F(r), · (3.34)

for a rigid translation (not a rotation).

As an example, consider the charge density

ρ(r) = ea δ(r − ra ). (3.35)
a

If the positions of all the particles, the ra , are displaced by δ, the charge density
changes to
ρ(r) + δρ(r) = ea δ(r − ra − δ), (3.36)
a

where
·
δ(r − ra − δ) = δ(r − ra ) − δ ∇r δ(r − ra ), (3.37)

and therefore
δρ(r) = −δ ∇ρ(r), · (3.38)

in agreement with (3.34).

So the field part of G in (3.18b) is
 
· ·
1 1
− (dr) E δA = (dr) E i (δ ∇)Ai
4π c 4π c

1
=−
c a
·
ea δ A(ra ) +
1
4π c
·
(dr) (E × B) δ, (3.39)

where the last rearrangement makes use of (3.15b) and (3.15h), and the vector identity

·
δ × (∇ × A) = ∇(δ A) − (δ ∇)A. · (3.40)

Including the particle part from (3.18b) we find the generator corresponding to a
rigid coordinate displacement can be written as
3.4 Momentum and Angular Momentum Conservation 23

·
G = δ P, (3.41)

where
   
ea 1
P= pa − A(ra ) + (dr) E × B ≡ m a va + (dr) G, (3.42)
a
c 4π c a

with G the momentum density. Since the action is invariant under a rigid displace-
ment,
0 = δW = G 1 − G 2 = (P1 − P2 ) δr, · (3.43)

we see that
P1 = P2 , (3.44)

that is, the total momentum, P, is conserved. This, of course, can also be verified by
explicit calculation:
   
d 1 1
(dr) E×B=− (dr) ρE + j × B
dt 4π c c
 1

=− ea E(ra ) + va × B(ra ) , (3.45)
a
c

from which the constancy of P follows.

Similar arguments can be carried out for a rigid rotation for which the change in
the coordinate vector is
δr = δω × r, (3.46)

with δω constant. The corresponding change in a vector function is

A(r + δr) = A(r) + δω × A(r) (3.47)

since a vector transforms in the same way as r, so the new function at the initial
numerical values of the coordinates is

·
A(r) = A(r) − (δr ∇)A(r) + δω × A(r). (3.48)

The change in the vector potential is

·
δA = −(δr ∇)A + δω × A. (3.49)

The generator can now be written in the form

·
G = δω J, (3.50)
24 3 Classical Field Theory—Electrodynamics

where the total angular momentum, J, is found to be

  
1
J= ra × m a va + (dr) r × E×B , (3.51)
a
4π c

which again is a constant of the motion.

3.5 Gauge Invariance and the Conservation of Charge

An electromagnetic system possesses a conservation law, that of electric charge,

which has no place in the usual mechanical framework. It is connected to a further
invariance of the electromagnetic fields—the potentials are not uniquely defined in
that if we let
1 ∂
A → A + ∇λ, φ→φ− λ, (3.52)
c ∂t
the electric and magnetic fields defined by (3.15f) and (3.15h) remain unaltered, for an
arbitrary function λ. This is called gauge invariance; the corresponding substitution
(3.52) is a gauge transformation. [The term has its origin in a now obsolete theory
of Hermann Weyl (1885–1955) (Weyl 1919).]
This invariance of the action must imply a corresponding conservation law. To
determine what is conserved, we compute the change in the Lagrangian, (3.9), explic-
itly. Trivially, the field part of L remains unchanged. In considering the change of the
particle part, we recognize that (3.52) is incomplete; since v is a physical quantity,
p − (e/c)A must be invariant under a gauge transformation, which will only be true
if (3.52) is supplemented by
e
p → p + ∇λ. (3.53)
c
Under the transformation (3.52) and (3.53), the Lagrangian becomes

 ea 
ea ∂

L → L ≡L +
c
∇λ
dra
dt
·− va +
c ∂t
ea
λ + va ∇λ
c
·
a
ea ∂ 
·
dra
=L + λ+ ∇λ
a
c ∂t dt
d
=L + w, (3.54)
dt
where ea
w= λ(ra , t). (3.55)
a
c
3.5 Gauge Invariance and the Conservation of Charge 25

What is the physical consequence of adding a total time derivative to a Lagrangian?

It does not change the equations of motion, so the system is unaltered. Since the
entire change is in the end point behavior,

W 12 = W12 + (w1 − w2 ), (3.56)

the whole effect is a redefinition of the generators, G,

G = G + δw. (3.57)

This alteration reflects the fact that the Lagrangian itself is ambiguous up to a total
time derivative term. [This term may also be seen as arising from the field term in
the generator (3.18b).]
To ascertain the implication of gauge invariance, we rewrite the change in the
Lagrangian given in the first line of (3.54) by use of (3.12c),
  
∂
·
1
L−L= (dr) ρ λ + j ∇λ , (3.58)
c ∂t

and apply this result to an infinitesimal gauge transformation, λ → δλ. The change
in the action is then
 t1   
∂
δW12 = G δλ1 − G δλ2 −
t2
dt (dr) δλ
1
c ∂t
ρ+∇ j , · (3.59)

with the generator being 

1
G δλ = (dr) ρ δλ. (3.60)
c

In view of the arbitrary nature of δλ(r, t), the stationary action principle now demands
that, at every point,
∂
∂t
ρ + ∇ j = 0,· (3.61)

that is, gauge invariance implies local charge conservation. (Of course, this same
result follows from Maxwell’s equations.) Then, the special situation δλ = constant,
where δA = δφ = 0, and W12 is certainly invariant, implies a conservation law, that
of
1
G δλ = δλ Q, (3.62)
c
in which 
Q= (dr) ρ (3.63)

is the conserved total charge.

26 3 Classical Field Theory—Electrodynamics

3.6 Gauge Invariance and Local Conservation Laws

We have just derived the local conservation law of electric charge. Electric charge is
a property carried only by the particles, not by the electromagnetic field. In contrast,
the mechanical properties of energy, linear momentum, and angular momentum are
attributes of both particles and fields. For these we have conservation laws of total
quantities. What about local conservation laws? The usual development of electro-
dynamics refers to local non-conservation laws; they concentrated on the fields and
characterized the charged particles as sources (or sinks) of field mechanical proper-
ties. It is natural to ask for a more even-handed treatment of both charges and fields.
We shall supply it, in the framework of a particular example. The property of gauge
invariance will be both a valuable guide, and an aid to simplifying the calculations.
The time displacement of a complete physical system identifies its total energy.
This suggests that time displacement of a part of the system provides energetic
information about that portion. The ultimate limit of this spatial subdivision, a local
description, should appear in response to an (infinitesimal) time displacement that
varies arbitrarily in space as well as in time, δt (r, t).
Now we need a clue. How do fields, and potentials, respond to such coordinate-
dependent displacements? This is where the freedom of gauge transformations enters:
The change of the vector and scalar potentials, by ∇λ(r, t), −(1/c)(∂/∂t)λ(r, t),
respectively, serves as a model for the potentials themselves. The advantage here is
that the response of the scalar λ(r, t) to the time displacement can be reasonably
taken to be
(λ + δλ)(r, t + δt) = λ(r, t), (3.64a)

or
∂
δλ(r, t) = −δt (r, t) λ(r, t). (3.64b)
∂t
Then we derive
 
∂ 1 ∂
δ(∇λ) = − δt
(∇λ) + − λ c∇δt, (3.65a)
∂t c ∂t
     
1 ∂ 1 ∂ 2 1 ∂ ∂
δ − λ = − δt − λ − − λ δt, (3.65b)
c ∂t c ∂t 2 c ∂t ∂t

which is immediately generalized to

∂
δA = − δt A + φc∇δt, (3.66a)
∂t
∂ ∂
δφ = − δt φ − φ δt, (3.66b)
∂t ∂t
or, equivalently,
3.6 Gauge Invariance and Local Conservation Laws 27

δA = cδtE + ∇(φcδt), (3.67a)

1 ∂
δφ = − (φcδt). (3.67b)
c ∂t
In the latter form we recognize a gauge transformation, produced by the scalar φcδt,
which will not contribute to the changes of field strengths. Accordingly, for that
calculation we have, effectively, δA = cδtE, δφ = 0, leading to

1 ∂ ∂ ∂
δE = − (cδtE) = −δt E − E δt, (3.68a)
c ∂t ∂t ∂t
∂
δB = ∇ × (cδtE) = −δt B − E × ∇cδt; (3.68b)
∂t

the last line employs the field equation ∇ × E = −(1/c)(∂B/∂t).

In the following we adopt a viewpoint in which such homogeneous field equations
are accepted as consequences of the definition of the fields in terms of potentials.
That permits the field Lagrange function (3.9) to be simplified:

1
Lfield = (E 2 − B 2 ). (3.69)
8π
Then we can apply the field variation (3.68b) directly, and get

∂ 1 2∂
δLfield = − δt
∂t
Lfield −
4π
E
∂t
δt −
c
4π
E × B ∇δt ·
∂ ∂
= − (δtLfield ) −
∂t
1
8π
(E 2 + B 2 ) δt −
∂t
c
4π
E × B ∇δt. · (3.70)

Before commenting on these last, not unfamiliar, field structures, we turn to the
charged particles and put them on a somewhat similar footing in terms of a continuous,
rather than a discrete, description.
We therefore present the Lagrangian of the charges in (3.9) in terms of a corre-
sponding Lagrange function,

L charges = (dr) Lcharges , (3.71a)

where
Lcharges = La (3.71b)
a

and
 
·
1 ea
La = δ(r − ra (t)) m a va (t)2 − ea φ(ra , t) + va (t) A(ra , t) ; (3.71c)
2 c
28 3 Classical Field Theory—Electrodynamics

the latter adopts the Lagrangian viewpoint, with va = dra /dt accepted as a definition.
Then, the effect of the time displacement on the variables ra (t), taken as

(ra + δra )(t + δt) = ra (t), (3.72a)

δra (t) = − δt (ra , t)va (t), (3.72b)

implies the velocity variation

 
∂
δva (t) = −δt (ra , t)
d
dt
va (t) − va (t)
∂t
·
δt + va ∇δt ; (3.73)

the last step exhibits both the explicit and the implicit dependences of δt (ra , t) on
t. In computing the variation of φ(ra , t), for example, we combine the potential
variation given in (3.66b) with the effect of δra :

∂ ∂ ∂
·
d
δφ(ra (t), t) = −δt φ − φ δt − δtva ∇ a φ = −δt φ − φ δt, (3.74a)
∂t ∂t dt ∂t
and, similarly,

∂
δA(ra (t), t) = −δt
∂t
· d
A + φc∇δt − δtva ∇ a A = −δt A + φc∇δt. (3.74b)
dt
The total effect of these variations on La is thus
  ∂ 
d
dt
2 ea
δLa = −δt La + δ(r − ra (t)) −m a va − A va + ea φ
c ∂t
·
δt + va ∇δt , ·
(3.75a)
or  
∂
·
d
δLa = − (δtLa ) − δ(r − ra (t))E a δt + va ∇δt , (3.75b)
dt ∂t

where we see the kinetic energy of the charged particle,

1
Ea = m a va2 . (3.76)
2
We have retained the particle symbol d/dt to the last, but now, being firmly back
in the field, space-time viewpoint, it should be written as ∂/∂t, referring to all t
dependence, with r being held fixed. The union of these various contributions to the
variation of the total Lagrange function is

∂ ∂
δLtot = −
∂t ∂t
·
(δtLtot ) − Utot δt − Stot ∇δt, (3.77)

where, from (3.70) and (3.75b),

3.6 Gauge Invariance and Local Conservation Laws 29

1
Utot = (E 2 + B 2 ) + δ(r − ra (t))E a (3.78a)
8π a

and c
Stot = E×B+ δ(r − ra (t))E a va , (3.78b)
4π a

are physically transparent forms for the total energy density and total energy flux
vector.
To focus on what is new in this development, we ignore boundary effects in the
stationary action principle, by setting the otherwise arbitrary δt (r, t) equal to zero
at t1 and t2 . Then, through partial integration, we conclude that
   
t1 ∂
δW12 =
t2
dt (dr) δt
∂t
·
Utot + ∇ Stot = 0, (3.79)

from which follows the local statement of total energy conservation,

∂
∂t
·
Utot + ∇ Stot = 0. (3.80)
Chapter 4
Quantum Action Principle

This Chapter, and the following one, are based on lectures given by the author in
quantum field theory courses at the University of Oklahoma over several years, based
in turn largely on lectures given by Schwinger at Harvard in the late 1960s.
After the above reminder of classical variational principles, we now turn to the
dynamics of quantum mechanics. We begin by considering the transformation func-
tion a  , t + dt|b , t. Here |b , t is a state specified by the values b = {b } of a
complete set of dynamical variables B(t), while |a  , t + dt is a state specified by
values a  = {a  } of a (different) complete set of dynamical variables A(t + dt),
defined at a slightly later time.1 We suppose that A and B do not possess any explicit
time dependence—that is, their definition does not depend upon t. Here

a  , t + dt| = a  , t|U, (4.1)

where the infinitesimal time translation operator is related to the generator of time
translations as follows,
U = 1 + i G = 1 − i dt H. (4.2)

The Hamiltonian H is a function of dynamical variables, which we write generically

as χ (t), and of t explicitly. Thus

a  , t + dt|b , t = a  , t|1 − i dt H (χ (t), t)|b , t. (4.3)

1 Here Schwinger is using his standard notation, designating eigenvalues by primes.

© The Author(s) 2015 31
K.A. Milton, Schwinger’s Quantum Action Principle,
SpringerBriefs in Physics, DOI 10.1007/978-3-319-20128-3_4
32 4 Quantum Action Principle

We next translate states and operators to time zero:

a  , t| = a  |U (t), |b , t = U −1 (t)|b , (4.4a)

−1
χ (t) = U (t)χU (t), (4.4b)

where χ = χ (0), etc. Then,

a  , t + dt|b , t = a  |1 − i dt H (χ , t)|b , (4.5)

or, as a differential equation

δdyn a  , t + dt|b , t = ia  |δdyn [−dt H ]|b 

= ia  , t + dt|δdyn [−dt H (χ (t), t)]|b , t, (4.6)

where δdyn corresponds to changes in initial and final times, δt2 and δt1 , and in the
structure of H , δ H . [By reintroducing dt in the state on the left in the second line,
we make a negligible error of O(dt 2 ).]
However, we can also consider kinematical changes. To understand these, consider
a system defined by coordinates and momenta, {qa (t)}, { pa (t)}, a = 1, . . . , n, which
satisfy the canonical commutation relations,

[qa (t), pb (t)] = iδab , ( = 1) (4.7a)

[qa (t), qb (t)] = [ pa (t), pb (t)] = 0. (4.7b)

A spatial displacement δqa is induced by


n
U = 1 + i Gq , Gq = pa δqa . (4.8)
a=1

In fact (δqa is a number, not an operator),

1
U −1 qa U = qa − [qa , G q ]
i
= qa − δqa , (4.9)

while
1
U −1 pa U = pa − [ pa , G q ] = pa . (4.10)
i
The (dual) symmetry between position and momentum,

q → p, p → −q, (4.11)
4 Quantum Action Principle 33

gives us the form for the generator of a displacement in p:


Gp = − qa δpa . (4.12)
a

A kinematic variation in the states is given by the generators

δkin  | =  | −  | =  |i G, (4.13a)
δkin |  = |  − |  = −i G| , (4.13b)

so, for example, under a δq variation, the transformation function changes by


δq a  , t +dt|b , t = ia  , t +dt| [ pa (t + dt)δqa (t + dt) − pa (t)δqa (t)] |b , t.
a
(4.14)
Now the dynamical variables at different times are related by Hamilton’s equations,

dpa (t) 1
= [ pa (t), H (q(t), p(t), t)]
dt i
∂H
=− (t), (4.15)
∂qa
so
dpa (t) ∂H
pa (t + dt) − pa (t) = dt = −dt (t). (4.16)
dt ∂qa

Similarly, the other Hamilton’s equation

dqa ∂H
= (4.17)
dt ∂ pa

implies that
∂H
qa (t + dt) − qa (t) = dt (t). (4.18)
∂ pa

From this we deduce first the q variation of the transformation function,

δq a  , t + dt|b , t
 ∂H
= ia  , t + dt| pa (t)[δqa (t + dt) − δqa (t)] − dt δqa (t) + O(dt 2 )|b , t
a
∂q a
 

= ia  , t + dt|δq pa (t).[qa (t + dt) − qa (t)] − dt H (q(t), p(t), t) |b , t,
a
(4.19)
34 4 Quantum Action Principle

where the dot denotes symmetric multiplication of the p and q operators.

For p variations we have a similar result:

δ p a  , t + dt|b , t

= − ia  , t + dt| [qa (t + dt)δpa (t + dt) − qa (t)δpa (t)]|b , t
a
 ∂H

= − ia , t + dt| qa (t)[δpa (t + dt) − δpa (t)] + dt (t)δpa (t)|b , t
a
∂ pa
 

= ia  , t + dt|δ p − qa (t).( pa (t + dt) − pa (t)) − dt H (q(t), p(t), t) |b , t.
a
(4.20)

That is, for q variations

 
δq a  , t + dt|b , t = ia  , t + dt|δq dt L q |b , t, (4.21a)

with the quantum Lagrangian


Lq = pa .q̇a − H (q, p, t), (4.21b)
a

while for p variations

 
δ p a  , t + dt|b , t = ia  , t + dt|δ p dt L p |b , t, (4.22a)

with the quantum Lagrangian


Lp = − qa . ṗa − H (q, p, t). (4.22b)
a

We see here two alternative forms of the quantum Lagrangian. Note that the two
forms differ by a total time derivative,

d 
Lq − L p = pa .qa . (4.23)
dt a

We now can unite the kinematic transformations considered here with the dynamic
ones considered earlier, in Eq. (4.6):

δ = δdyn + δkin : δa  , t + dt|b , dt = ia  , t + dt|δ[dt L]|b , t. (4.24)

Suppose, for concreteness, that our states are defined by values of q, so that

δ p a  , t + dt|b t = 0. (4.25)
4 Quantum Action Principle 35

This is consistent, as a result of Hamilton’s equations,

  
∂H
δp Lq = δpa q̇a − = 0. (4.26)
a
∂ pa

In the following we will use L q .

It is immediately clear that we can iterate the infinitesimal version (4.24) of the
quantum action principle by inserting at each time step a complete set of intermediate
states (to simplify the notation, we ignore their quantum numbers):

t1 |t2  = t1 |t1 − dtt1 − dt|t1 − 2dt · · · t2 + 2dt|t2 + dtt2 + dt|t2 , (4.27)

So in this way we deduce the general form of Schwinger’s quantum action principle:
t1
δt1 |t2  = it1 |δ dt L|t2 . (4.28)
t2

This summarizes all the properties of the system.

Suppose the dynamical system is given, that is, the structure of H does not change.
Then
δt1 |t2  = it1 |G 1 − G 2 |t2 , (4.29)

where the generator G a depends on p and q at time ta . Comparing with the action
principle (4.28) we see
t1
δ dt L = G 1 − G 2 , (4.30)
t2

which has exactly the form of the classical action principle (2.3), except that the
Lagrangian L and the generators G are now operators. If no changes occur at the
endpoints, we have the principle of stationary action,


t1 
δ pa .dqa − H dt = 0. (4.31)
t2 a

As in the classical case, let us introduce a time parameter τ , t = t (τ ), such that τ2

and τ1 are fixed. Calling the new time parameter by the original name, the above
variation reads

[δpa .dqa + pa .dδqa − δ H dt − H dδt]
a
 
 
=d pa .δqa − H δt + [δpa .dqa − dpa .δqa ] − δ H dt + d H δt, (4.32)
a a
36 4 Quantum Action Principle

so the action principle says


G= pa .δqa − H δt, (4.33a)
a
dH  dqa dpa

δH = δt + δpa . − δqa . . (4.33b)
dt a
dt dt

We will again assume δpa , δqa are not operators (that is, they are proportional to the
unit operator); then we recover Hamilton’s equations,

∂H dH
= , (4.34a)
∂t dt
∂H dqa
= , (4.34b)
∂ pa dt
∂H dpa
=− . (4.34c)
∂qa dt

(Schwinger also explored the possibility of operator variations, see, for example, his
les Houches lectures (Schwinger 1970a).) We learn from the generators,

G t = −H δt, G q = pa δqa , (4.35)
a

that the change in some function F of the dynamical variable is

dF 1
δF = δt + [F, G], (4.36)
dt i
so we deduce
dF ∂F 1
= + [F, H ], (4.37a)
dt ∂t i
∂F 1
= [F, pa ]. (4.37b)
∂qa i

Note that from this the canonical commutation relations follow,

[qa , pb ] = iδab , [ pa , pb ] = 0, (4.38)

as well as Newton’s law,

1 ∂H
ṗa = − [H, pa ] = − . (4.39)
i ∂qa
4 Quantum Action Principle 37

If we had used L p instead of L q , we would have obtained the same equations of

motion, but in place of G q , we would have obtained

Gp = − qa δpa , (4.40)
a

which implies
∂F 1
= − [F, qa ]. (4.41)
∂ pa i

From this can be deduced the remaining canonical commutator,

[qa , qb ] = 0, (4.42)

as well as the remaining Hamilton equation,

1 ∂H
q̇a = [qa , H ] = . (4.43)
i ∂ pa

It is easy to show that the effect of changing the Lagrangian by a total time derivative
(which is what is done in passing from L q to L p ) is to change the generators.
We now turn to examples.

4.1 Harmonic Oscillator

The harmonic oscillator is defined in terms of creation and annihilation operators,2

y † and y, and the corresponding Hamiltonian H ,

[y, y † ] = 1, (4.44a)
 
1
H =ω y y +
†
. (4.44b)
2

The equations of motion are

dy 1 1
= [y, H ] = ωy, (4.45a)
dt i i
dy † 1 † 1
= [y , H ] = − ωy † . (4.45b)
dt i i

Eigenstates of y and y † exist, as right and left vectors, respectively,

2 We follow Schwinger’s usage of y for the annihilation operator, instead of the more usual a.
38 4 Quantum Action Principle

y|y   = y  |y  , (4.46a)
y |y = y y |,
† † † †
(4.46b)

while y  | and |y †  do not exist.3 These are the famous “coherent states,” to whom the
name Roy Glauber (1963) is invariably attached, although they were discovered by
Erwin Schrödinger (1926), and Glauber’s approach, as he acknowledged, followed
that of his mentor, Schwinger (1953).
The transformation function we seek is therefore

y † , t1 |y  , t2 . (4.48)

If we regard y as a “coordinate,” the corresponding “momentum” is iy † :

1 ∂H ∂H
ẏ = ωy = , i ẏ † = −ωy † = − . (4.49)
i ∂iy † ∂y

The corresponding Lagrangian is therefore4

L = iy † .ẏ − H. (4.51)

Because we use y as our state variable at the initial time, and y † at the final time, we
must exploit our freedom to redefine our generators to write
1
W12 = dt L − iy † (t1 ).y(t1 ). (4.52)
2

Then the variation of the action is

δW12 = − iδ(y1† .y1 ) + G 1 − G 2

= − iδy1† .y1 − iy1† .δy1 + iy1† .δy1 − iy2† .δy2 − H δt1 + H δt2
= − iδy1† .y1 − iy2† .δy2 − H (δt1 − δt2 ). (4.53)

3 If y  |y = y  y  | then we would have an evident contradiction:

1 = y  |[y, y † ]|y   = y  y  |y † |y   − y  |y † |y  y  = 0. (4.47)

4 We might note that in terms of (dimensionless) position and momentum operators

i 1 i d 2
iy † .ẏ = (q − i p).(q̇ +i ṗ) = ( p.q̇ − q. ṗ) + (q + p 2 ), (4.50)
2 2 4 dt
where the first term in the final form is the average of the Legendre transforms in L q and L p .
4.1 Harmonic Oscillator 39

Then the quantum action principle says

δy † , t1 |y  , t2  = iy † , t1 | − iδy1† y1 − iy2† δy2 − ωy1† y1 (δt1 − δt2 )|y  , t2 , (4.54)

since by assumption the variations in the dynamical variables are numerical:

[δy1† , y1 ] = [y2† , δy2 ], (4.55)

and we have dropped the zero-point energy. Now use the equations of motion (4.45a)
and (4.45b) to deduce that

y1 = e−iω(t1 −t2 ) y2 , y2† = e−iω(t1 −t2 ) y1† (4.56)

and hence

δy † , t1 |y  , t2  = y † , t1 |δy † e−iω(t1 −t2 ) y  + y † e−iω(t1 −t2 ) δy 

− iωy † e−iω(t1 −t2 ) (δt1 − δt2 )y  |y  , t2 

= y † , t1 |y  , t2 δ y † e−iω(t1 −t2 ) y  . (4.57)

From this we can deduce that the transformation function has the exponential form

y † , t1 |y  , t2  = exp y † e−iω(t1 −t2 ) y  , (4.58)

which has the correct boundary condition at t1 = t2 ; and in particular, 0|0 = 1.

On the other hand,

y † , t1 |y  , t2  = y † |e−i H (t1 −t2 ) |y  , (4.59)

where both states are expressed at the common time t2 , so, upon inserting a complete
set of energy eigenstates, we obtain (t = t1 − t2 )

y † |Ee−i Et E|y  , (4.60)
E

which we compare to the Taylor expansion of the previous formula,

∞
(y † )n −inωt (y  )n
√ e √ . (4.61)
n=0
n! n!
40 4 Quantum Action Principle

This gives all the eigenvectors and eigenvalues:

E n =nω, n = 0, 1, 2, . . . , (4.62a)
(y † )n
y † |E n  = √ , (4.62b)
n!
(y  )n
E n |y   = √ . (4.62c)
n!

These correspond to the usual construction of the eigenstates from the ground state:

(y † )n
|E n  = √ |0. (4.63)
n!

4.2 Forced Harmonic Oscillator

Now we add a driving term to the Hamiltonian,

H = ωy † y + yK ∗ (t) + y † K (t), (4.64)

where K (t) is an external force (Kraft is force in German). The equation of motion is

dy ∂H
i = † = [y, H ] = ωy + K (t), (4.65)
dt ∂y

while y † satisfies the adjoint equation. In the presence of K (t), we wish to compute
the transformation function y † , t1 |y  , t2  K .
Consider a variation of K . According to the action principle

δ K y † , t1 |y  , t2  K = y † , t1 |iδ K W12 |y  , t2  K
t1
= − iy † , t1 | dt[δ K y † + δ K ∗ y]|y  , t2  K . (4.66)
t2

We can solve this differential equation by noting that the equation of motion (4.65)
can be rewritten as
d  iωt
i e y(t) = eiωt K (t), (4.67)
dt
which is integrated to read
t 
e iωt
y(t) − e iωt2
y(t2 ) = −i dt  eiωt K (t  ), (4.68)
t2
4.2 Forced Harmonic Oscillator 41

or t 
y(t) = e−iω(t−t2 ) y2 − i dt  e−iω(t−t ) K (t  ), (4.69)
t2

and the adjoint5

t1 
y (t) =
†
e−iω(t1 −t) y1† −i dt  e−iω(t −t) K ∗ (t  ). (4.72)
t

Thus our differential equation (4.66) reads

δ K y † , t1 |y  , t2  K
= δ K lny † , t1 |y  , t2  K
y † , t1 |y  , t2  K
t1  t1 

=−i dt δ K (t) y † e−iω(t1 −t) − i dt  e−iω(t −t) K ∗ (t  )
t2 t
t1  t 
∗ −iω(t−t2 )   −iω(t−t  ) 
−i dt δ K (t) e y −i dt e K (t ) . (4.73)
t2 t2

Notice that in the terms bilinear in K and K ∗ , K always occurs earlier than K ∗ .
Therefore, these terms can be combined to read
t1 
− δK dt dt  K ∗ (t)η(t − t  )e−iω(t−t ) K (t  ), (4.74)
t2

where the step function is 

1, t > 0,
η(t) = (4.75)
0, t < 0.

Since we already know the K = 0 value from Eq. (4.58), we may now immediately
integrate our differential equation:

5 The consistency of these two equations follows from

t1

eiωt1 y1 = eiωt2 y2 − i dt  eiωt K (t  ), (4.70)
t2

so that the adjoint of Eq. (4.69) is

 t1  t
 
[y(t)]† = eiωt e−iωt1 y1† − i dt  e−iωt K ∗ (t  ) + i dt  e−iω(t −t) K ∗ (t  )
t2 t2
t
−iω(t  −t)
= eiω(t−t1 ) y1† + i dt  e K ∗ (t  ), (4.71)
t1

which is Eq. (4.72).

42 4 Quantum Action Principle

y † , t1 |y  , t2  K = exp y † e−iω(t1 −t2 ) y 
t1 t1
− iy † dt e−iω(t1 −t) K (t) − i dt e−iω(t−t2 ) K ∗ (t) y 
t2 t2
t1 
 ∗  −iω(t−t  ) 
− dt dt K (t)η(t − t )e K (t ) . (4.76)
t2

The ground state is defined by y  = y † = 0, so

 ∞ 
 ∗  −iω(t−t  ) 
0, t1 |0, t2  = exp −
K
dt dt K (t)η(t − t )e K (t ) , (4.77)
−∞

where we now suppose that the forces turn off at the initial and final times, t2 and t1 ,
respectively.
A check of this result is obtained by computing the probability of the system
remaining in the ground state:
 ∞ 
|0, t1 |0, t2  | = exp
K 2
dt dt  K ∗ (t)e−iω(t−t )
−
−∞

  
× [η(t − t ) + η(t − t)]K (t )
 ∞ 
 ∗ −iω(t−t  ) 
= exp − dt dt K (t)e K (t )
−∞

= exp −|K (ω)|2 , (4.78)

where the Fourier transform of the force is

∞
K (ω) = dt eiωt K (t). (4.79)
−∞

The probability requirement

|0, t1 |0, t2  K |2 ≤ 1 (4.80)

is thus satisfied. We see here a resonance effect: If the oscillator is driven close to
its natural frequency, so K (ω) is large, there is a large probability of finding the
system in an excited state, and therefore of not remaining in the ground state. Let us
calculate this transition amplitude to an excited state. By setting y  = 0 in Eq. (4.76)
we obtain
4.2 Forced Harmonic Oscillator 43
 ∞ 
y † , t1 |0, t2  K = exp −iy † dt e−iω(t1 −t) K (t) 0, t1 |0, t2  K
−∞

= y † , t1 |n, t1 n, t1 |0, t2  K , (4.81)
n

where we have inserted a sum over a complete set of energy eigenstates, which
possess the amplitude [see Eq. (4.62b)]

(y † )n
y † |n = √ . (4.82)
n!

If we expand the first line of Eq. (4.81) in powers of y † , we find

(−i)n
n, t1 |0, t2  K = √ e−inωt1 [K (ω)]n 0, t1 |0, t2  K . (4.83)
n!

The corresponding probability is

|K (ω)|2n −|K (ω)|2

p(n, 0) K = |n, t1 |0, t2  K |2 = e , (4.84)
n!

which is a Poisson distribution6 with mean n̄ = |K (ω)|2 .

Finally, let us define the Green’s function for this problem by

G(t − t  ) = −iη(t − t  )e−iω(t−t ) . (4.86)

It satisfies the differential equation

 
d
i − ω G(t − t  ) = δ(t − t  ), (4.87)
dt

as it must because [see Eq. (4.65)]

 
d
i − ω y(t) = K (t), (4.88)
dt

where y(t) is given by [see Eq. (4.69)]

6A Poisson probability distribution has the form p(n) = λn e−λ /n!. The mean value of n for this
distribution is
∞
 ∞
 ∞

λn e−λ
n̄ = n p(n) = =λ p(n) = λ. (4.85)
(n − 1)!
n=0 n=0 n=0
.
44 4 Quantum Action Principle
∞
y(t) = e−iω(t−t2 ) y2 + dt  G(t − t  )K (t  ). (4.89)
−∞

Similarly, from Eq. (4.72)

∞
y (t) =
†
e−iω(t1 −t) y1† + dt  G(t  − t)K ∗ (t  ). (4.90)
−∞

We can now write the ground-state persistence amplitude (4.77) as

 ∞ 
 ∗  
0, t1 |0, t2  = exp −i
K
dt dt K (t)G(t − t )K (t ) , (4.91)
−∞

and the general amplitude (4.76) as

 ∞ 
y † , t1 |y  , t2  K = exp −i dt dt  K ∗ (t) + iy † δ(t − t1 )
−∞

 
× G(t − t  ) K (t  ) + iy  δ(t  − t2 ) , (4.92)

which demonstrates that knowledge of 0, t1 |0, t2  K for all K determines everything:
 δ(t−t
y † , t1 |y  , t2  K = 0, t1 |0, t2  K (t)+iy 2 )+iy
† δ(t−t
1) . (4.93)

4.3 Feynman Path Integral Formulation

Although much more familiar, the path integral formulation of quantum mechanics
(Feynman 1942, 1949, 1965) is rather vaguely defined. We will here provide a formal
“derivation” based on the Schwinger principle, in the harmonic oscillator context.
Consider a forced oscillator, defined by the Lagrangian (note in this section, H does
not include the source terms)

L = iy † .ẏ − H (y, y † ) − K y † − K ∗ y. (4.94)

As in the preceding section, the action principle says

t1
δ K 0, t1 |0, t2  K = −i0, t1 | dt [δ K y † + δ K ∗ y]|0, t2  K , (4.95)
t2

or for t2 < t < t1 ,

4.3 Feynman Path Integral Formulation 45

δ
i 0, t1 |0, t2  K = 0, t1 |y † (t)|0, t2  K , (4.96a)
δ K (t)
δ
i 0, t1 |0, t2  K = 0, t1 |y(t)|0, t2  K , (4.96b)
δ K ∗ (t)

where we have introduced the concept of the functional derivative. The equation of
motion
∂H ∂H
i ẏ − † − K = 0, −i ẏ † − − K ∗ = 0, (4.97)
∂y ∂y

is thus equivalent to the functional differential equation,

    
δ δ
0 = i K (t), W i ,i − K (t) 0, t1 |0, t2  K , (4.98)
δK ∗ δK

where (the square brackets indicate functional dependence)

t1
W [y, y ] = †
dt [iy † (t).ẏ(t) − H (y(t), y † (t))]. (4.99)
t2

The reason Eq. (4.98) holds is that by definition

δ
K (t  ) = δ(t − t  ), (4.100)
δ K (t)
so
 t1   
iδ
 d iδ iδ iδ
i K (t), dt i . −H ,
t2 δ K (t  ) dt  δ K ∗ (t  ) δ K ∗ (t  ) δ K (t  )
 
d iδ ∂ iδ iδ
=i ∗
− H ∗
, , (4.101)
dt δ K (t) ∂(iδ/δ K (t)) δ K (t) δ K (t)

which corresponds to the first two terms in the equation of motion (4.97), under the
correspondence
δ δ
y↔i , y† ↔ i . (4.102)
δK ∗ δK
Since [[K , W ], W ] = 0, we can write the functional equation (4.98) as
∗ ,iδ/δ K ] ∗ ,iδ/δ K ]
0 = ei W [iδ/δ K K e−i W [iδ/δ K 0, t1 |0, t2  K . (4.103)

The above equation has a solution (up to a constant), because both equations (4.97)
must hold,
∗
0, t1 |0, t2  K = ei W [iδ/δ K ,iδ/δ K ] δ[K ]δ[K ∗ ], (4.104)
46 4 Quantum Action Principle

where δ[K ], δ[K ∗ ] are functional delta functions. The latter have functional Fourier
decompositions (up to a multiplicative constant),

δ[K ] = [dy † ]e−i dt K (t)y † (t)
, (4.105a)

dt K ∗ (t)y(t)
δ[K ∗ ] = [dy]e−i , (4.105b)

where [dy] represents an element of integration over all (numerical-valued) functions

y(t), and so we finally have
∗
0, t1 |0, t2  K ,K
 t1  
∗
= [dy][dy ] exp −i†
dt K (t)y (t) + K (t)y(t) + i W [y, y ]
† †
t2
 t1  
∗
= [dy][dy ] exp i
†
dt iy ẏ − H (y, y ) − K y − K y ,
† † †
(4.106)
t2

where y, y † are now numerical, and the functional integration is over all possible
functions,
 over all possible “paths.” Of course, the classical paths, the ones for which
W − dt (K y † + K ∗ y) is an extremum, receive the greatest weight, at least in the
classical limit, where  → 0.

4.3.1 Example

Consider the harmonic oscillator Hamiltonian, H = ωy † y. Suppose we wish to

calculate, once again, the ground state persistence amplitude, 0, t1 |0, t2  K . It is
perhaps easiest to perform a Fourier transform,
∞ ∞
y(ν) = dt eiνt y(t), y ∗ (−ν) = dt e−iνt y † (t). (4.107)
−∞ −∞

Then
∞ ∞ dν
dt y † (t)y(t) = y(ν)y ∗ (−ν), (4.108a)
−∞ −∞ 2π
∞ ∞
dν
dt iy † (t)ẏ(t) = νy(ν)y ∗ (−ν). (4.108b)
−∞ −∞ 2π
4.3 Feynman Path Integral Formulation 47

Thus Eq. (4.106) becomes

∗
0, t1 |0, t2  K ,K

∗ dν
= [dy][dy ] exp i [y(ν)(ν − ω)y ∗ (−ν)
2π

∗ ∗
− y (−ν)K (ν) − y(ν)K (−ν)]
    
∗ dν K (ν) ∗ K ∗ (−ν)
= [dy][dy ] exp i y(ν) − (ν − ω) y (−ν) −
2π ν−ω ν−ω

dν 1
−i K (ν) K ∗ (−ν)
2π ν−ω
 
∗ dν ∗
= [dy][dy ] exp i y(ν)(ν − ω)y (−ν)
2π
 
dν 1 ∗
× exp −i K (ν) K (−ν)
2π ν−ω
 
dν 1 ∗
= exp −i K (ν) K (−ν) , (4.109)
2π ν−ω

since the functional integral in the third equality, obtained by shifting the integration
variable,

K (ν)
y(ν) − →y(ν), (4.110a)
ν−ω
K ∗ (−ν)
y ∗ (−ν) − →y ∗ (−ν), (4.110b)
ν−ω
∗
is 0, t1 |0, t2  K = K = 0 = 1. How do we interpret the singularity at ν = ω in the
remaining integral? We should have inserted a convergence factor in the original
functional integral:
   
dν dν  
exp i [. . . ] → exp i · · · + iy(ν)y ∗ (−ν) , (4.111)
2π 2π

where  goes to zero through positive values. Thus we have, in effect, ν − ω →

ν − ω + i and so we have for the ground-state persistence amplitude
∗

dt dt  K ∗ (t)G(t−t  )K (t  )
0, t1 |0, t2  K ,K = e−i , (4.112)

which has the form of Eq. (4.91), with

∞ 
 dν e−iν(t−t )
G(t − t ) = , (4.113)
−∞ 2π ν − ω + i
48 4 Quantum Action Principle

which is evaluated by closing the ν contour in the upper half plane if t − t  < 0, and
in the lower half plane when t − t  > 0. Since the pole is in the lower half plane we
get

G(t − t  ) = −iη(t − t  )e−iω(t−t ) , (4.114)

which is exactly what we found in Eq. (4.86).

Now, let us rewrite the path integral (4.106) in terms of coördinates and momenta:


1 ω1
q = √ (y + y ), p =
†
(y − y † ), (4.115a)
2ω 2i
     
ω ip ω ip
y= q+ , y =
†
q− . (4.115b)
2 ω 2 ω

Then the numerical Lagrangian appearing in (4.106) may be rewritten as (see footnote
4 above)

L = iy † ẏ − ωy † y − K y † − K ∗ y
   
ω p ṗ ω2 p2
=i q −i q̇ + i − q + 2
2
2 ω ω 2 ω
     
ω ip ω ∗ ip
− K q− − K q+
2 ω 2 ω
 
ω d p 2 1 d 1
=i q 2 + 2 + pq̇ − ( pq) − ( p 2 + ω2 q 2 )
4 dt ω 2 dt 2
 
2 2
− K q − K p
ω ω
d
= w + L(q, q̇, t), (4.116)
dt
where, if we set q̇ = p, the Lagrangian is

1 2 1 2 2
L(q, q̇, t) = q̇ − ω q + Fq, (4.117)
2 2
if √
K = 0, F = − 2ωK . (4.118)

In the path integral  

 ∂(y, y † ) 
[dy][dy † ] = [dq][dp]  , (4.119)
∂(q, p) 
4.3 Feynman Path Integral Formulation 49

where the Jacobian is

  
   ω2 ω 

 ∂(y, y )  
† 2

 =  = 1, (4.120)
 ∂(q, p)   
 √i − √i 
2ω 2ω

and so from the penultimate line of Eq. (4.116), the path integral (4.106) becomes
 t1 
0, t1 |0, t2  F = [dy][dy † ] exp i dt L(y, y † )
t2
 t1  
1 2 1 2 2
= [dq][dp] exp i dt pq̇ − p − ω q + Fq . (4.121)
t2 2 2

Now we can carry out the p integration, since it is Gaussian:

   
i dt − 21 p 2 + pq̇ i dt − 21 ( p−q̇)2 + 21 q̇ 2
[dp]e = [dp]e
  ∞
dt 21 q̇ 2 1
dpi e− 2 i pi t .
2
= ei (4.122)
i −∞

Here we have discretized time so that p(ti ) = pi , so the final functional integral
√ over
p is just an infinite product of constants, each one of which equals e−iπ/4 2π/Δt.
Thus we arrive at the form originally written down by Feynman [Feynman 1965],
 t1 
0, t1 |0, t2  = F
[dq] exp i dt L(q, q̇, t) , (4.123)
t2

with the Lagrangian given by Eq. (4.117), where an infinite normalization constant
has been absorbed into the measure.

4.4 Toward Source Theory

Let us return to the action principle. Recall from Eq. (4.112)


dt dt  K ∗ (t)G(t−t  )K (t  )
0t1 |0t2  K = e−i . (4.124)

The action principle says

δt1 |t2  = it1 |δ[W = dt L]|t2 . (4.125)
50 4 Quantum Action Principle

In a general sense, the exponent in Eq. (4.124) is an integrated form of the action. In
solving the equation of motion, we found in Eq. (4.89)

y(t) = e−iω(t−t2 ) y(t2 ) + dt  G(t − t  )K (t  ), (4.126)

where the first term is effectively zero here. The net effect is to replace an operator
by a number:
y  (t) = dt  G(t − t  )K (t  ). (4.127)

Then Eq. (4.124) can be written as


dt K ∗ (t)y  (t)
0t1 |0t2  K = e−i . (4.128)

Recall that the action was was the integral of the Lagrangian (4.94), or
 
∂
W = dt y † i y − ωy † y − y † K (t) − yK ∗ (t) , (4.129)
∂t

so we see one term in Eq. (4.128) here, and the equation of motion (4.65) cancels
out the rest! So let’s add something which gives the equation for y  :
 
d 
i dt y † i dt y −ωy † y  −y † K −y  K ∗
0t1 |0t2  = e
K
= ei W . (4.130)

Now insist that W is stationary with respect to variations of y  , y † , and we recover

the equation of motion,  
d
i − ω y  (t) = K (t). (4.131)
dt

This is the starting point for the development of source theory, which will be treated
in Chap. 7.
Chapter 5
Time-Cycle or Schwinger-Keldysh
Formulation

A further utility of the action principle is the time-cycle or Schwinger-Keldysh

formalism, which allows one to calculate matrix elements and consider nonequilib-
rium systems. Schwinger’s original work on this was his famous paper (Schwinger
1961); Keldysh’s paper appeared three years later (Keldysh 1964), and, rather mys-
teriously, cites the Martin-Schwinger equilibrium paper (Martin 1959), but not the
nonequilibrium one (Schwinger 1961). The following was extracted from notes from
Schwinger’s lectures given in 1968 at Harvard, as taken by the author.
Consider the expectation value of some physical property F(t) at a particular time
t1 in a state |b, t2 :

F(t1 )b t2 = b t2 |a  t1 a  |F|a  a  t1 |b t2 , (5.1)
a  a 

which expresses the expectation value in terms of the matrix elements of the operator
F in a complete set of states defined at time t1 , {|a  t1 }. Suppose the operator F has
no explicit time dependence. Then we can use the action principle to write
 t1 
δa  t1 |b t2  = ia  t1 |δ dt L |b t2 , (5.2a)
t2

and so  t1 
δb t2 |a  t1  = −ib t2 |δ dt L |a  t1 , (5.2b)
t2

which can be obtained from the first equation by merely exchanging labels,
 t1  t2
=− . (5.3)
t2 t1

© The Author(s) 2015 51

K.A. Milton, Schwinger’s Quantum Action Principle,
SpringerBriefs in Physics, DOI 10.1007/978-3-319-20128-3_5
52 5 Time-Cycle or Schwinger-Keldysh Formulation

Fig. 5.1 A “time-cycle,” in t1

which a system advances
forward in time from time t2
to time t1 under the influence
of a Lagrangian L + , and then
backward in time from time
L− L+
t1 back to time t2 under the
influence of Lagrangian L −

t2

If we consider 
b t2 |b t2  = b t2 |a  t1 a  t1 |b t2 , (5.4)
a

the above variational equations indeed asserts that

δb t2 |b t2  = 0. (5.5)

We can interpret the above as a cycle in time, going from time t2 to t1 and then
back again, as shown in Fig. 5.1. But, now imagine that the dynamics is different on
the forward and return trips, described by different Lagrangians L + and L − . Then
 t1  t1 
  
δb t2 |b t2  = ib t2 |δ dt L + − dt L − |b t2 . (5.6)
t2 t2

In particular, consider a perturbation of the form,

H = H0 + λ(t)F, (5.7)

where λ(t) is some time-varying parameter. If we have an infinitesimal change, and,

for example, δλ+ = 0, δλ− = 0, then
 t1
  λ+ λ− 
δλ+ b t2 |b t2  = −ib t2 | dt δλ+ F|b t2 . (5.8)
t2

If we choose δλ+ to be an impulse,

δλ+ = δλδ(t − t  ), (5.9)

in this way we obtain the expectation value of F(t  ).

5 Time-Cycle or Schwinger-Keldysh Formulation 53

Fig. 5.2 A time cycle in t1

which the harmonic
oscillator evolves from time
t2 to time t1 under the
influence of a force K + , and
then from t1 back to time t2
K− K+
under a force K −

t2

Let’s illustrate this with a driven harmonic oscillator, as described by Eq. (4.64),
so now
∗
H+ = ω y † y + K + (t)y + K + (t)y † , (5.10a)
∗
H− = ω y y + †
K− (t)y + K − (t)y , †
(5.10b)

which describes the oscillator evolving forward in time from t2 to t1 under the influ-
ence of the force K + , and backward in time from t1 to t2 under the influence of K − ,
as shown in Fig. 5.2. From the variational principle we can learn all about y and y † .
We have already solved this problem by a more laborious method above, in Sect. 4.2.
It suffices to solve this problem with initial and final ground states. If we consider
only a K ∗ variation,

K + ,K −
t1  ∗ ∗

δ K ∗ 0t2 |0t2  = −i0t2 | dt δK + (t)y+ (t) − δK − (t)y− (t) |0t2 . (5.11)
t2

Now we must solve the equations of motion, so since effectively y(t2 ) → 0, we have
from Eq. (4.69),
 t 
y+ (t) = − i dt  e−iω(t−t ) K + (t  ), (5.12a)
t2
 t1  t
 
y− (t) = − i dt  e−iω(t−t ) K + (t  ) − i dt  e−iω(t−t ) K − (t  ). (5.12b)
t2 t1

The last term in the second equation is

 t1 
i dt  e−iω(t−t ) K − (t  )η(t  − t), (5.13)
t2

so naming the advanced and retarded Green’s functions by extending the definition
in Eq. (4.86),
54 5 Time-Cycle or Schwinger-Keldysh Formulation

 η(t  − t)
G a,r (t, t  ) = ie−iω(t−t ) , (5.14)
−η(t − t  )

which satisfy the same differential equation (4.87), we effectively have

 t1
y+ (t) = dt  G r (t − t  )K + (t  ), (5.15a)
t2
 t1  t1

y− (t) = − i dt  e−iω(t−t ) K + (t) + dt  G a (t − t  )K − (t  ), (5.15b)
t2 t2

The solution to the variational equation (5.11) is now

∗ (t)G (t−t  )K (t  )
dt dt  K +
0t2 |0t2  K + ,K − = e−i r +

∗ (t)G (t−t  )K (t  )


∗ (t)e−iω(t−t ) K (t  )
dt dt  K − dt dt  K −
× ei a − e + .
(5.16)

This should reduce to 1 when K + = K − = K , so


− i G r (t − t  ) + i G a (t − t  ) + e−iω(t−t ) = 0, (5.17)

which is, indeed, true.

As an example, consider K − (t) = K (t), K + (t) = K (t + T ), that is, the second
source is displaced forward by a time T . This is sketched in Fig. 5.3. What does this
mean? From a causal analysis, in terms of energy eigenstates, reading from right to
left, 
0t2 |0t2  K − ,K + = 0t2 |nt1  K − =K (t) nt1 |0t2  K + =K (t+T ) . (5.18)
n

The effect on the second transformation function is the same as moving the n, t1 state
to a later time,

nt1 |0t2  K (t+T ) = nt1 + T |0t2  K (t) = e−inωT nt1 |0t2  K (t) , (5.19)

Fig. 5.3 Time cycle in t1

which K − (t) = K (t),
K + (t) = K (t + T ), that is,
the forces are the same on
the two legs, but displaced in
time
K(t)
K(t + T )

t2
5 Time-Cycle or Schwinger-Keldysh Formulation 55

so this says that 

0t2 |0t2  K − K + = e−inωT p(n, 0) K , (5.20)
n

which gives us the probabilities directly. From the formula (5.16) we have, using
Eq. (5.17),


dt dt  K ∗ (t)e−iω(t−t ) [K (t  +T )−K (t  )]
0t2 |0t2  K − K + = e


dt dt  K ∗ (t)e−iω(t−t ) [e−iωT −1]K (t  )]
=e
 
e−iωT −1
= e|γ|
2
, (5.21)

where 
γ= dt eiωt K (t). (5.22)

Thus we immediately obtain Eq. (4.84), or

(|γ|2 )n
p(n, 0) K = e−|γ|
2
. (5.23)
n!
The above Eq. (5.21) can be directly used to find certain average values. For
example,  
2 −iωT −1
e−inωT 0K = e|γ| e . (5.24)

Expand this for small ωT and we find

n0K = |γ|2 . (5.25)

In a bit more systematic way we obtain the dispersion:

2 (e−iωT −1+iωT )
e−i(n−n)ωT  = e|γ| . (5.26)

Expanding this to second order in ωT we get

(n − n)2  = n 2  − n2 ≡ (n)2 = |γ|2 = n, (5.27)

or
n 1
=√ . (5.28)
n n

For large quantum numbers, which corresponds to the classical limit, the fluctuations
become relatively small.
56 5 Time-Cycle or Schwinger-Keldysh Formulation

Now consider a more general variational statement than in Eq. (5.11),


∗ ∗ † †
δ |  K − K + = −i | dt[δK + y+ − δK − y− + δK + y+ − δK − y− |  K ± , (5.29)

we see that since we can change the source functions at will, and make very localized
changes, it makes sense to define the variational derivatives

δ
∗ (t)  |  =  |y+ (t)|  K ± ,
K±
i (5.30a)
δK +
δ †
−i  |  K ± =  |y− (t)|  K ± . (5.30b)
δK − (t)

All expectation values of operator products at any time can be obtained in this way—
in particular, correlation functions. Repeating this operation we get

δ δ δ
(−i) i ∗ (t  ) t2 |t2 
K±
=−i t2 |y+ (t  )|t2  K ±
δK − (t) δK + δK − (t)
†
=t2 |y− (t)y+ (t  )|t2  K ± . (5.31)

The operators are multiplied in the order of the time development. The only place
where K − appears is in the latter part of the time development. See Fig. 5.4.
The distinction between ± disappears if we now set K + = K − :

δ δ
∗ (t  ) 0t2 |0t2  K ± = 0t2 |y † (t)y(t  )|0t2  K . (5.32)
δK − (t) δK + K + =K − =K

As an example, set t = t  = t1 ; then, from Eq. (5.16), this reads for the number
operator N (t) = y † (t)y(t),
 
∗
N (t1 )0K = dt  G r (t1 − t  )K (t  )
dt K (t)G a (t − t1 )
 

=i dt e−iω(t−t1 ) K ∗ (t)(−i) dt  e−iω(t1 −t ) K (t  ) = |γ|2 ,

(5.33)

Fig. 5.4 Variational t1

derivatives pick out operators
at definite times t and t 
−t
t−

t2
5 Time-Cycle or Schwinger-Keldysh Formulation 57

t1

K+
K−
t2
t2

Fig. 5.5 Time cycle with different forces, K + and K − . on the forward and backward moving
segments. Now the initial time of the time cycle, t2 , is different from the final time of the time cycle,
t2 , with τ = t2 − t2 . It is assumed that the time t1 is later than both t2 and t2 , and that the forces are
localized as shown

as before, Eq. (5.25).

We would like to use more general starting and ending states than the ground
state. We can obtain these by use of impulsive forces. It is convenient to deal with
all states at once, as in the generating function for p(n, 0) K considered above. Think
of a time cycle starting at time t2 , advancing forward to time t1 , during which time
the force K + acts, then moving back in time to a time t2 , under the influence of the
force K − —See Fig. 5.5.
Let t2 = t2 + τ . This displacement injects energy information. Consider
 
nt2 |nt2  K ± ≡ trt2 |t2  K ± = e−inωτ nt2 |nt2  K ± , (5.34)
n n

which uses (no force acts between times t2 and t2 )

nt2 | = nt2 |e−inωτ . (5.35)

Analysis of this formula will yield individual transformation functions.

Now we must solve the dynamical equations subject to boundary conditions. Let
us compare trt2 |y+ (t2 )|t2  with trt2 |y− (t2 )|t2 . The first is
 
trt2 |y+ (t2 )|t2  = nt2 |y+ (t2 )|nt2  = nt2 |n  t2 n  |y|n, (5.36a)
n nn 

while the second appears as

 
trt2 |y− (t2 )|t2  = n  t2 |y− (t2 )|n  t2  = n  |y|nnt2 |n  t2 . (5.36b)
n nn 

Here, by introducing a complete set of states at the time of the operator, we have
expressed the formula in terms of the matrix elements of stationary operators.
58 5 Time-Cycle or Schwinger-Keldysh Formulation

Remarkably, we see that the two expressions are equal; in effect, there is a peri-
odicity present here:
y+ (t2 ) = y− (t2 ), (5.37)

as far as traces are concerned. Now, the equations of motion (4.65) for the operators
read  
d
i − ω y(t) = K (t), (5.38)
dt

which has solution (5.12b) with the addition of the initial term, or
 t1
−iω(t−t2 ) 
y− (t) = e y+ (t2 ) − i dt  e−iω(t−t ) K + (t  )
t2
 t1
−iω(t−t  )
+i dt  e K − (t  ). (5.39)
t

In particular,


y− (t2 ) = e−iωτ y+ (t2 ) − i dt  e−iω(t2 +τ −t ) (K + − K − )(t  ). (5.40)

Note that the integrals sweep over the full force history. Let us let t2 = 0 for simplicity,
although we will keep the label. Because of the periodicity condition (5.37) this reads
  
eiωτ − 1 y+ (t2 ) = −i dt eiωt (K + − K − )(t) = −i(γ+ − γ− ), (5.41)

or
1
y+ (t2 ) = (−i)(γ+ − γ− ). (5.42)
eiωτ − 1

What we are interested in is

trt2 |t2  K ±
. (5.43)
trt2 |t2 

The denominator, which refers to the free harmonic oscillator, is immediately eval-
uated as
∞
 1
trt2 |t2  = e−inωτ = . (5.44)
1 − e−iωτ
n=0

(If τ be imaginary, we have thermodynamic utility.) We have then the variational

equation
 
 ∗ 
trt2 |t2  K ± −itrt2 | dt δK + y+ − δK −∗y
− |t2 
K±
δ K ±∗ = . (5.45)
trt2 |t2  trt2 |t2 
5 Time-Cycle or Schwinger-Keldysh Formulation 59

Exactly as before, we get an equation for the logarithm—looking at the previous

calculation leading to Eq. (5.16), we see an additional term, referring to the y+ (t2 )
boundary term in Eq. (5.40). The periodic boundary condition then gives

1
− δ(γ ∗ − γ−
∗
)(γ+ − γ− ). (5.46)
eiωτ − 1 +

Therefore, to convert 0t2 |0t2  K ± in Eq. (5.16) to


trt2 |t2  K ± e−inωτ nt2 |nt2  K ±
=  −inωτ (5.47)
trt2 |t2  e

we must multiply by
1
exp[− |γ+ − γ− |2 ]. (5.48)
eiωτ −1

This holds identically in τ ; in particular, in the limit where τ → −i∞, which

corresponds to absolute zero temperature, we recover 0t2 |0t2  K ± .
We find, generalizing Eq. (5.16)

e−inωτ nt2 |nt2  K ±
 ∗  
n
 −inωτ = e−i dt dt K + (t)G r (t−t )K + (t )
n e

∗ (t)G (t−t  )K (t  )


∗ (t)e−iω(t−t ) K (t  )
dt dt  K − dt dt  K −
× ei a − e +

iωτ −1)−1


∗ −K ∗ )(t)e−iω(t−t ) (K −K )(t  )
dt dt  (K +
× e−(e − + − , (5.49)

which is the exponential of a bilinear structure. This is a generating function for the
amplitudes nt2 |nt2  K ± . But it is useful as it stands.
Put τ = −iβ; then this describes a thermodynamic average over a thermal mixture
at temperature T , where β = 1/kT in terms of Boltzmann’s constant,

e−βnω  | n
n
 −βnω (5.50)
ne

In terms of this replacement,

1 1
→ βω = nβ , (5.51)
eiωτ − 1 e −1

because

ne−inωτ ∂  ∂ 1 1
n −inωτ = ln( e−inωτ ) = ln −iωτ
= iωτ .
n e ∂(−iωτ ) n
∂(−iωτ ) 1 − e e −1
(5.52)
60 5 Time-Cycle or Schwinger-Keldysh Formulation

Now consider a time cycle with displacement T : the system evolves from time t2
to time t1 under the influence of the force K + (t), and backwards in time from t1 to
t2 under the force K − (t):

K − (t) = K (t), K + (t) = K (t + T ). (5.53)

This is again as illustrated in Fig. 5.5, with these replacements. What is the physical
meaning of this? Insert in Eq. (5.49) a complete set of states at time t1 :

nt2 |nt2  K ± = nt2 |n  t1  K − n  t1 |nt2  K + . (5.54)
n

We did this before for the ground state. The effect is the same as moving the starting
and ending times. Appearing here is

n  t1 |nt2  K (t+T ) = n  t1 + T |nt2 + T  K (t) = e−in ωT n  t1 |nt2  K (t) einωT . (5.55)

Therefore,
 
nt2 |nt2  K (t),K (t+T ) = e−i(n −n)ωT p(n  , n) K = e−i(N −n)ωT nK . (5.56)
n

Therefore, as a generalization for finite τ of Eq. (5.26), we have from Eq. (5.49)
 −1
 
−in  ωτ
e e−inωτ e−i(N −n)ωT nK
n n
     
1
= exp e−iωT − 1 |γ|2 − e iωT
− 1 e −iωT
− 1 |γ| 2
, (5.57)
eiωτ − 1

where T gives the final state, and τ the initial state. This used the observation
 
−iωT
dt e iωt
K (t + T ) = e dt eiωt K (t). (5.58)

Expand both sides of Eq. (5.57) in powers of T , and we learn

 e−inωτ
− iωT N − nnK  −in  ωτ
= −iωT |γ|2 , (5.59)
n n  e

or
N − nβK = |γ|2 , (5.60)
5 Time-Cycle or Schwinger-Keldysh Formulation 61

which generalizes the earlier result (5.25). Now apply Eq. (5.59) as a generating
function,
N − nnK = |γ|2 , (5.61)

which reflects the linear nature of the system.

We can rewrite the above generating function more conveniently, by multiplying
by
eiN −nωT = eiωT |γ| ,
2
(5.62)

that is, Eq, (5.57) can be written as

1 
 e−inωτ e−i(N −N )ωT nK
e−inωτ
     
−iωT 1 −iωT
= exp e − 1 + iωT |γ| − iωτ
2
e −1 e iωT
− 1 |γ| .
2
e −1
(5.63)

Now pick off the coefficient of −(ωT )2 /2:

1  1
 e−inωτ (N − N )2 nK = |γ|2 + 2 |γ|2 , (5.64)
e−inωτ eiωτ −1

or
(N − N )2 βK = |γ|2 [1 + 2nβ ]. (5.65)

If, instead, we multiply Eq. (5.64) through by n e−inωτ , we can use this as a
generating function, and learn from Eq. (5.52) that

(N − N )2 nK = |γ|2 (1 + 2n). (5.66)

Note the simplicity of the derivation of this result, which does not involve complicated
functions like Laguerre polynomials.
Chapter 6
Relativistic Theory of Fields

This section is an adaptation of Chap. 5 of lectures given at Stanford by Schwinger

in 1956 (Schwinger 1956).
A state of a physical system is defined in terms of the maximum number of
compatible measurements which can be made upon the system. If the state were
defined on a space-like surface (one in which all points are in space-like relation:
(x)2 − (t)2 > 0) then a measurement at any point is compatible with one made
at any other point, since the disturbances introduced by the measurements cannot
propagate faster than c, and hence cannot interfere. Thus, a state can be specified as an
eigenvector of a complete set of commuting, Hermitian operators a, associated with
a definite space-like surface σ : |a  , σ . There always exists a coordinate system in
which the space-like surface σ is all of three-dimensional space at a given time; in
this particular Lorentz frame the state vector is just: |a  , t. The problem of relativistic
dynamics is to find the transformation function

a1 σ1 |a2 σ2 . (6.1)

As in the non-relativistic case, we assume the existence of an action operator W12

such that
δa1 σ1 |a2 σ2  = ia1 σ1 |δ[W12 ]|a2 σ2 . (6.2)

The contributions to the action operator are now given by


W12 = Wσ +dσ,σ . (6.3)
σ

© The Author(s) 2015 63

K.A. Milton, Schwinger’s Quantum Action Principle,
SpringerBriefs in Physics, DOI 10.1007/978-3-319-20128-3_6
64 6 Relativistic Theory of Fields

σ + dσ
σ

Fig. 6.1 Spacetime volume bounded by two spacelike surfaces, σ and σ + dσ . Points in the blue
and green shaded regions cannot interfere

Since measurements made at points in the space-like shell dσ cannot interfere (e.g.,
in the two shaded areas, shown in Fig. 6.1) their contribution to W12 is additive,

σ1 σ1
W12 = (d x)L(x) = (d x)L[χa (x)], (6.4)
σ2 σ2

where the χa (x) = χa (x, y, z, t) are the dynamical variables of the system, neces-
sarily Hermitian operators; the relativistic requirements automatically introduce the
concept of fields. The relativistic notation used has the form

x 0 = t = −x0 , x k = xk , where k = 1, 2, 3,
∂
(d x) = d x 0 d x1 d x2 d x3 , ∂μ = μ , (6.5)
∂x
where the metric used is ⎛ ⎞
−1 0 0 0
⎜0 1 0 0⎟
gμν =⎜
⎝0
⎟. (6.6)
0 1 0⎠
0 0 0 1

The action principle again states that for a given dynamical system, the variations
arise only from the end-point, that is,

δW12 = G 1 (σ1 ) − G 2 (σ2 ). (6.7)

As before, from this requirement follow the equations of motion and the generators
of infinitesimal transformations which yield the commutation relations of the field
operators. The Lagrangian density L which will yield first order field equations is

1
μ 
L= χ A ∂μ χ − ∂μ χ Aμ χ − H (χ ) , (6.8)
2
where the Aμ are a set of four numerical matrices, and space and time derivatives
appear on a symmetric basis. The symmetrization of the kinematical term relates
to the possibility of adding to L the relativistic analogue of our previous total time
derivative, a four dimensional divergence. If

L = L − ∂μ f μ , (6.9)
6 Relativistic Theory of Fields 65

then
σ1  
W 12 = W12 − (d x)∂μ f = W12 − dσμ f + dσμ f μ ,
μ μ
(6.10)
σ2 σ1 σ2

and
 
G(σ1 ) = G(σ1 ) − dσμ f μ , G(σ2 ) = G(σ2 ) − dσμ f μ . (6.11)
σ1 σ2

As before, the equation of motion is unchanged. Since L is to be Hermitian, we

require
Aμ† = −Aμ , so H† = H (6.12)

The rank of Aμ is that of the number of independent fields. Note that the variation
δa1 σ1 |a2 σ2  is independent of any coordinate system, since L is a Lorentz scalar.
We can now infer some fundamental properties from the requirement of invariance
of L. Consider the coordinate transformation (Poincaré transformation)

x μ = μ ν x ν − μ (6.13)

where
gμν μ λ ν κ = gλκ , μ λ μ κ = δ λ κ . (6.14)

We can divide the transformations into two subsets by considering the effect of (6.13)
on g00 ,

 2  3  2  2   2
− g00 =1=  0 −
0
k 0 0 0 = 1 + k 0 ≥ 1. (6.15)
k=1 k

Since 0 0 = ∂∂ xx 0 it follows that we can never make a continuous change from a

0

positive to a negative sense of time, i.e., generate an improper transformation con-

tinuously. We shall consider only the group of continuous proper Lorentz transfor-
mations.
Under such a coordinate change, the fields χa (x) change to new ones χ a (x)
connected by a real linear transformation:

χ (x) = Lχ (x). (6.16)

66 6 Relativistic Theory of Fields

Using ∂ μ = ν μ ∂ν , and writing the scalar L in the new frame we find

1
μ 
L= χ A ∂ μ χ − ∂ μ χ Aμ χ − H (χ )
2
1  T μ ν  
= χ L A  μ L ∂ν χ − ∂ν χ ν μ L T Aμ L χ − H (Lχ ) . (6.17)
2
Thus
L T Aμ L = μ ν Aν , H (Lχ ) = H (χ ) . (6.18)

If we choose for H the special form

H = χ Bχ = χ L T B Lχ , (6.19)

where B is Hermitian and non-singular, then L T = B L −1 B −1 . From the first of

Eq. (6.18), we then obtain
 
L −1 B −1 Aμ L = μ ν B −1 Aν , (6.20)

showing that the combination B −1 Aμ transforms like a vector; this is implied by the
required invariance of the kinematical term of L.
Consider now a general inhomogeneous infinitesimal Lorentz transformation, of
the form
x μ = x μ − μ + μν x ν , (6.21)

where the  μ and  μ ν are infinitesimals, and the rotational nature of the  μν is
expressed by the relations μν = −νμ . Then L can be written as L = 1 + 2i μν S μν ,
where only the rotational μν terms appear, since the translations  μ do not effect the
gradient operators ∂ν , and no corresponding changes in the χα are required to keep
L invariant. The S μν are (imaginary) operators, acting on the field variable, which
will express the spin character of the fields.
The variation δWα allows us to change the field components at each space-time
point (call these changes δα χα ), and to change the region of integration by displacing
the boundary surfaces σ1 and σ2 . In the previous non-relativistic treatment, instead of
varying the end-point times t1 and t2 ,we used t (τ ). Proceeding similarly here, we can
express the variation of a space-like surface by varying the space-time coordinates
under an infinitesimal Lorentz transformation (δxμ = μ − μν x ν ) so chosen that
on σ1 and σ2 the required displacement is obtained. The change in the action is

σ1  
1
μ 
δW12 = δ0 (d x) χ A ∂μ χ − ∂μ χ Aμ χ − H
2
σ2
6 Relativistic Theory of Fields 67
 
+ dσμ δx μ L − dσμ δ0 x μ L
σ1 σ2
σ1
 
= (d x) δ0 χ Aμ ∂μ χ − ∂μ χ Aμ δ0 χ − δ0 H
σ2
σ1
1

+ (d x) ∂μ χ Aμ δ0 χ − δ0 χ Aμ χ
2
σ2
 
+ dσμ δx μ L − dσμ δx μ L
σ1 σ2
σ1   
μ 1
μ μ

= (d x) δ0 L + dσμ δx L + χ A δ0 χ − δ0 χ A χ
2
σ2 σ1
  
1
μ 
− dσμ δx μ L + χ A δ0 χ − δ0 χ Aμ χ . (6.22)
2
σ2

Applying the action principle, the interior variation δ0 L must vanish, giving the
field equations of motion. What remains is the difference of two generators, G(σ1 ) −
G(σ2 ), where
  
μ 1
μ μ

G(σ ) = dσμ δx L + χ A δ0 χ − δ0 χ A χ . (6.23)
2
σ

To re-write (6.23) we recognize that the total change in the fields is due to both the
δ0 variation of the fields at a given space-time point on σ , and also to the variation
induced by the infinitesimal Lorentz transformation of coordinates as σ is displaced
to σ + dσ ; the latter is obtained from

i
χ (x) = Lχ (x) = χ (x) + μν S μν χ (x), (6.24)
2
But
χ (x) = χ(x) − δx μ ∂μ χ (x), (6.25)

and therefore
i
χ (x) − χ (x) = δx μ ∂μ χ (x) + μν S μν χ (x). (6.26)
2
68 6 Relativistic Theory of Fields

The right hand side of (6.26) is then the variation induced in the χα by the coordinate
transformation; the total variation of the fields is then

i
δχ (x) = δ0 χ (x) + δx μ ∂μ χ (x) + μν S μν χ (x). (6.27)
2
Solving for δ0 χ , and substituting into (6.23), we obtain for the generator
  
1
μ 
G(σ ) = dσμ δx μ L + χ A δχ − δχ Aμ χ
2
 
1

− dσμ χ Aμ ∂ ν χ − ∂ ν χ Aμ χ δxν
2

i

− λν χ Aμ S λν χ − S λν χ Aμ χ . (6.28)
4

Adding a surface term, it is possible to bring G(σ ) into the form:

 
1

G(σ ) = dσμ χ Aμ δχ − δχ Aμ χ + dσμ δxν T μν , (6.29)
2

where T μν is the symmetric stress-tensor operator

1  {μ ν} 
T μν = g μν L − χ A ∂ χ − ∂ {ν χ Aμ} χ
4
i  {ν μ}λ 
− ∂λ χ A S χ − S {μλ Aν} χ , (6.30)
4
and the brackets { } represent symmetrization with respect to μ and ν.
Applying the stationary action principle to the variation δxν , taken as arbitrary,
we note that the invariance of the action operator implies the conservation law:
 
dσμ T μν = dσμ T μν , (6.31)
σ1 σ2

which, in turn, implies the corresponding differential conservation law

∂μ T μν = 0. (6.32)

The generator (6.29) can be split into two parts, one representing changes induced
by the coordinate variation, and the other giving the variation induced by a change
in the field variables,
6 Relativistic Theory of Fields 69

1
μ 
Gχ = χ A δχ − δχ Aμ χ ,
dσμ (6.33a)
2

1
G x = dσμ T μν δxν = ν P ν + μν J μν , (6.33b)
2

where  
 
Pν ≡ dσμ T μν , J μν ≡ dσλ x μ T λν − x ν T λμ ; (6.34)

P ν and J μν are the generators for translations and rotations, respectively, and their
commutation relations are determined by the group of transformations they represent.
Specifically, P ν is recognized as the 4-momentum operator, and J μν as the relativistic
generalization of the angular momentum operator.
The field equations are obtained by the vanishing of δ0 L in (6.22),

∂ H ∂r H
2 A μ ∂μ χ = or − 2 AμT ∂μ χ = , (6.35)
∂χ ∂χ

in terms of left and right derivatives, corresponding to the two equivalent ways of
writing G χ :
 
Gχ = dσμ χ Aμ δχ , or Gχ = − dσμ δχ Aμ χ . (6.36)

If we continued with these two pairs of expressions,we would obtain two forms for
the commutation rules of χ ; their equivalence then leads to the requirement that the
Aμ and χ must decompose:

χ = φ + ψ, Aμ = a μ + s μ , (6.37)

where the a μ are anti-symmetric and real, the s μ are symmetric and imaginary, and
the φ and ψ represent the kinematically independent fields of the Bose-Einstein and
Fermi-Dirac types, respectively. The field equations are then

∂ H ∂r H
2a μ ∂μ φ = = , (6.38a)
∂φ ∂φ
∂ H ∂r H
2s μ ∂μ ψ = =− . (6.38b)
∂ψ ∂ψ

Similarly, G χ can be. divided into its φ and ψ parts, G χ = G φ + G ψ ,

 
μ


Gφ = dσμ φa δφ = dσμ a μ δφ φ,
 


Gψ = dσμ ψs μ δψ = dσ μ s μ δψ ψ. (6.39)
70 6 Relativistic Theory of Fields

Again, we find (where { , } denotes an anticommutator)

 
δψα (x), ψβ (x  ) = 0,
     
δφα (x), φk (x  ) = δφk (x), ψα (x  ) = δψα (x), φk (x  ) = 0. (6.40)

The first of Eq. (6.40) combined with Eq. (6.38b) implies that H must be an even
function of ψ.
The field equations may be written as equations of motion by singling out the
time differentiation,
∂ H
2 A 0 ∂0 χ = − 2 A k ∂k χ . (6.41)
∂χ

If we took A0 to be non-singular, we would be able to solve (6.41) for ∂0 χ . More

generally, we now recognize the existence of the following possible situations:
1. A0 is non-singular. In this case, all of the variables are kinematically independent.
An example of this situation is the Dirac-Majorana spin- 21 field.
2. A0 is singular, but there are enough relations among the variables to determine
all of them. Here, only those variables which possess equations of motion are
kinematically independent, but the non-independent fields are determined from
the independent fields. Examples of this are the spin zero and spin one fields.
3. A0 is singular, and there are not enough relations among the variables to deter-
mine all the fields, as in case 2. The classic example of this is the spin 1, zero
mass, electromagnetic field, where the lack of determination corresponds to the
possibility of introducing a gauge transformation.

6.1 Inference of Particle Properties

We now consider the generators of infinitesimal coordinate (Lorentz) transformations

G k and the commutation relations they imply. From (6.33b) we have

1
G x = ν P ν + μν J μν , (6.42)
2
which, when applied to the space-time coordinates xμ , generates the new x μ ,

x  = x μ − μ + μν x ν , (6.43)

which is Eq. (6.21). Accompanying this transformation, we have the apparent change
in the fields given by (6.26),

i
− δχ (x) = χ(x) − χ (x) = δx μ ∂μ χ (x) + μν S μν χ (x), (6.44a)
2
6.1 Inference of Particle Properties 71

or  
1

− δχ (x) =  μ ∂μ + μν x μ ∂ ν − x ν ∂ μ − i S μν χ (x), (6.44b)
2

using μν = −νμ . Comparing with [χ , G x ] = −iδχ , for arbitrary translations (μ )

and rotations (μν ), we obtain
 
χ , P ν = −i∂ ν χ , (6.45a)
 

χ , J μν = −i x μ ∂ ν − x ν ∂ μ + i S μν χ . (6.45b)

From (6.45b) the identification of S μν with the intrinsic spin characteristics of the
particle is evident. Considering the time component (6.45a) we obtain the standard
commutator equation of motion
  ∂χ
χ , P 0 = −i∂ 0 χ = i∂0 χ = i . (6.46)
∂t
To determine the manner in which the particle interpretation enters, we now represent
the fields by Fourier integrals,

μp
χ (x) = (dp)ei x μ χ ( p), (6.47)

where the χ ( p) are operator functions of the numbers pμ . Substituting into (6.45a)
and equating coefficients, we obtain
 
χ ( p), P ν = p ν χ ( p), (6.48a)

or

P ν χ ( p) = χ ( p) P ν − p ν . (6.48b)

Since the 4 operators P ν all commute, we can have simultaneous eigenstates |P  ,

where P ν | p   = P ν | p  . Applying (6.48a) to these states



P ν χ | p   = χ P ν − p ν |P   = P ν − p ν χ |P  , (6.49)

or writing χ |P   as some new eigenvector |P  − p,


P ν |P  − p = P ν − p ν |P  − p, (6.50)

which shows that the effect of χ on |P   is to produce a state whose momentum

eigenvalue has been changed by − p ν , indicating the capacity of the field to absorb
or emit p ν . Specifically, considering the time component ν = 0, we have two possible
situations, depending on whether p 0 is greater than or less than zero. If p 0 > 0 ,
application of χ on |P   yields a state with lower energy (P 0 − p 0 ), and conversely
72 6 Relativistic Theory of Fields

if p 0 < 0. Since χ is Hermitian, χ can be split, in a Lorentz covariant way, into two
parts, χ = χ (+) + χ (−) , where the superscripts indicate the positive and negative
nature of the p 0 terms which enter in the respective Fourier transforms, and where
(χ (+) )† = χ (−) . Then χ (+) applied to a state of definite energy acts as an energy
annihilation operator, and χ (−) as an energy creation operator.
The χ (+) and χ (−) can then be used to create any physical state from the vacuum
state, where we take the latter as the unique lowest-energy state. This vacuum state
must necessarily correspond to the eigenvalue P k = 0, (k = 1, 2, 3), since if
one of the P k were not zero, a rotation of the coordinate system could yield three
non-zero momentum components, requiring (a super-position of) the corresponding
eigenvectors for its description. But this precludes a description of the vacuum by a
single non-degenerate state; we must therefore require that each P k = 0. Since the
vacuum is to have the lowest energy possible, and we may arbitrarily take this to be
zero, we then characterize the vacuum state as that unique state for which P μ = 0,
where χ (+) |0 = 0.
So far we have considered χ as representing general fields; to introduce particle
properties consider the Fourier transform of χ , and imagine the numbers pμ related
by the relation: 
− pμ p μ = m 2 , p 0 = ± m 2 + p2 . (6.51)

Then χ (+) , for example, remains an operator which annihilates energy, but now is
correlated with a momentum decrease. This is just the usual particle interpretation;
if χ (+) has this character it may be spoken of as a particle annihilation operator, and
conversely for χ (−) .

6.2 The Connection Between Spin and Statistics

We consider the simplest system, corresponding to linear field equations. Taking

H = χ Bχ , this becomes for our two types of fields H = φ B (1) φ + ψ B (2) ψ, where
φ, ψ represent fields of the 1st and 2nd kinds, respectively, meaning bosonic and
fermionic fields. (Note that H must be even in ψ, and therefore, as noted below
Eq. (6.40), no term of the form φψ can occur.) The matrix B (and therefore B (1) and
B (2) ) is Hermitian. Writing

δφ H = δφ B (1) φ + φ B (1) δφ = δφ B (1) φ + B (1)T φ , (6.52)

and since
∂ H ∂r H
= , (6.53)
∂φ ∂φ

for these variations of fields of the first kind, both terms in the above bracket must
be equivalent, implying that B (1) = B (1)T . Since B (1) is also Hermitian, it is real.
6.2 The Connection Between Spin and Statistics 73

For δψ variations, the identical argument shows that B (2) = −B (2)T , i.e., B (2) is
imaginary. It is precisely here that the connection between spin and statistics arises:
We can construct matrices for B −1 a μ only for particles of integral spin, and for
B −1 s μ only when the particle spin is an integer plus one-half. [See Eq. (6.37).]
Writing the equations of motion (6.35) for either field

∂ H
2 A μ ∂μ χ = = 2Bχ , (6.54)
∂χ

and assuming—as the simplest case—that B is non-singular, we obtain

B −1 Aμ ∂μ χ (x) = χ (x). (6.55)

Going to the momentum representation χ ( p) as before, this becomes

i B −1 Aμ pμ χ ( p) = χ ( p). (6.56)

Now consider the matrix (of finite order) i B −1 Aμ pμ ≡ M, which must satisfy its
algebraic minimal characteristic equation M n + a1 M
n−1 + · · · + an =
0, wherethe
an are numbers. From Eq. (6.20) we know that L −1 B −1 Aμ L = μ ν B −1 Aν . If
we insert the proper combinations L −1 L in each term, and use the relation

L −1 B −1 Aμ pμ L = B −1 Aν μ ν pμ = B −1 Aμ p ν , (6.57)

then the result of a Lorentz transformation, giving us back the identical minimal equa-
tion (as it must), shows that the a j are Lorentz invariants, or functions of invariants.
Since the only 4-vectors available are the pμ , we take a j = a j (− pμ p μ ). Further-
more, the minimal equation must be valid independently of the value of the numbers
pμ , i.e., it must be an algebraic identity in pμ ; the coefficients a j (− p μ pμ ) must
then be of the form (− p μ pμ ) j times a numerical factor c j , which is independent of
the pμ . We can identify the two possible cases, corresponding to the degree of the
minimal equation being either even (n = 2k) or odd (n = 2k + 1); in either case the
power to which (B −1 Aμ pμ ) is raised must decrease in steps of two,
 2k
  2k−2
n = 2k : i B −1 Aμ pμ + − pμ p μ c1 i B −1 Aμ pμ + ...


+ − pμ p μ ck = 0,
k
(6.58a)
 2k+1
  2k−1
n = 2k + 1 : i B −1 Aμ pμ + − pμ p μ c1 i B −1 Aμ pμ + ...

 
+ − pμ p μ ck i B −1 Aμ pμ = 0.
k
(6.58b)

Note also that the numbers c j must be real, since

74 6 Relativistic Theory of Fields

†
B † = B, i Aμ = i Aμ , (6.59a)
 † 
and i B −1 Aμ pμ = i Aμ B −1 pμ = B i B −1 Aμ pμ B −1 . (6.59b)

Taking the adjoint of the minimal equation then corresponds to making a similarity
transformation on the matrices and complex conjugating every c j term. Since each
B can be combined with a B −1 term, and since we must still have the same unique
minimal equation, it follows that each c j = c∗j . All of the above is a direct inference
from the requirement of Lorentz invariance.
If we now apply (6.58a) and (6.58b) to the field χ ( p), and use (6.56) we obtain
for either case


2
k 
1 + − pμ p μ c1 + − pμ p μ c2 + · · + − pμ p μ ck χ ( p) = 0. (6.60)

6.3 Fermi-Dirac Fields of Spin 1/2

Consider now only the simplest case of k = 1, which corresponds to a minimal

equation of degree 2. Then Eq. (6.60) becomes

 
1 + − pμ p μ c1 χ = 0, (6.61)

and interpreting c1 as −m −2 , for a non-vanishing χ we have the familiar relation

pμ p μ + m 2 = 0.
We now go to the simplest case of all, that for which n = 2 in Eq. (6.58a); this
then becomes  2 pμ p μ
i B −1 Aμ pμ + = 0. (6.62)
m2
Writing
B = mβ, Aμ = iα μ , and β −1 α μ = γ μ (6.63)

then α μ is Hermitian, and γ μ† = βγ μ β −1 . Eq. (6.62) becomes

μ 2
γ pμ = − pμ p μ = −g μν pμ pν = γ μ γ ν pμ pν , (6.64)

and since only the symmetric combination pμ pν enters here, we have

1  μ ν
γ , γ = −g μν (6.65)
2
as a necessary condition for Eq. (6.62) to be satisfied algebraically by pμ .
Before proceeding further, we remark that the construction of 5 matrices, satisfy-
ing (6.65), each of a definite symmetry (to represent the γ μ and γ 5 = γ 0 γ 1 γ 2 γ 3 ),
6.3 Fermi-Dirac Fields of Spin 1/2 75

can be achieved in only one way: Three of the matrices must be symmetric, and the
remaining two antisymmetric. This statement, which is easily verified, together with
(6.62) shows that we can construct matrices satisfying (6.65) only for Fermi-DIrac
fields, as follows: Since α μ is real, and β imaginary, α μ = βγ μ = −βγ μ∗ , i.e., all
the γ μ are imaginary. From (6.65) we note that
 2  2
γ 0 = −g 00 = +1 γ k = −g kk = −1. (6.66)

Defining γ μ = i μ where the  μ are then real, it follows that

 2  2
0 = −1,  k = +1,
 
 0jn n
0
= −δ j ,  kjn n
k
= δ j . (6.67)
n n

We are looking for matrices of definite symmetry; equating j and , (6.67) can be
satisfied only if n0 j = − 0jn , nk j = + kjn ; i.e., γ 0 is antisymmetric (and imag-
inary, and therefore Hermitian), and the three γ k . are symmetric (and imaginary,
and therefore
 skew-Hermitian).
  γ 5 is then antisymmetric and real. It then follows
that β, γ = β, γ = 0, and since β and γ 0 have the same properties, we may
k 0

identify them: β = γ 0 .
If we now attempt to repeat this for Bose-Einstein fields, then the α μ are anti-
symmetric and imaginary, β is symmetric and real, all the γ μ are the same as for the
Fermi-Dirac case, and again β, γ = 0. But this is a direct violation of the require-
k

ments that there be but three independent symmetric matrices, satisfying (6.66).
Equation (6.65), from which the spin 21 formalism is obtained, therefore refers only
to Fermi-Dirac fields.
The Lagrangian for the Fermi-Dirac spin 21 field is then

1  m
L= ψ, iα μ ∂μ ψ − [ψ, βψ] . (6.68)
2 2
More precisely, this is the Lagrangian for the uncharged Dirac-Majorana field—in
order to represent charge we shall later have to double the number of components
of ψ. For notation, we introduce the use of the dot · to symbolize the proper sym-
metrization brackets to be used for Fermi-Dirac or Bose-Einstein fields. Then

L = ψ· iα μ ∂μ ψ − mψ· βψ. (6.69)

The field equations are then the familiar ones

 
1 μ
α ∂μ + mβ ψ = 0, (6.70)
i
76 6 Relativistic Theory of Fields

where, since ψ is Hermitian, the adjoint of these equations must, and do, yield
identical equations.
To obtain the commutation relations, we use:
 
G ψ = dσμ χ Aμ δχ → dσ0 ψs 0 δψ, (6.71)

where s 0 = iα 0 = i (β = γ 0 ), and form

  i
χ , G χ = δψ(x), (6.72)
2
or    i
dσ0 ψ(x), ψ(x  )s 0 δψ(x  ) = δψ(x), (6.73)
2

where the factor 21 enters because the ψ are “non-canonical,” and [ ] → { } since δψ
anti-commutes with ψ. This then yields

  i  0 −1 (0)
ψ(x), ψ(x  ) = s δ (x − x0 ), (6.74)
2
or
  1
ψα (x), ψβ (x  ) = δαβ δ (0) (x − x  ), (6.75)
2

where x and x  are points on the same space-like surface, and (s 0 )−1 = −i.

6.4 Spin Zero and One


Let us return to the characteristic Eq. (6.58b) of the matrix i B −1 Aμ pμ . We will

−1 μ 
choose k = 1, and examine the odd polynomial in i B A pμ , which gives the
possibility of describing a particle of zero mass. This is necessary to describe the
electromagnetic field.
Define the four vector β μ ≡ im B −1 Aμ . Then we have

μ 3

β pμ + p ν pν β μ pμ = 0,


or β μ β σ β ν + g μν β σ pμ pσ pν = 0. (6.76)

From this equation, which is an identity in pμ , we can make statements only about
the symmetric part of the matrix products. If we completely symmetrize the matrix
factor with respect to μ, σ , and ν, it must vanish. This is the sum of three terms of
the form
β μ β σ β ν + β ν β σ β μ = −g μσ β ν − g νσ β μ , (6.77)
6.4 Spin Zero and One 77

which are the familiar Kemmer-Duffin commutation relations (Petiau 1936; Duffin
1938; Kemmer 1939) which are used to describe a particle of spin zero and one. All
the β matrices are singular. These matrices have 126 independent elements and hence
are reducible to three sub-matrices of dimensionality 10, 5, and l; the sub-matrix of
dimensionality 1 is trivially the null matrix; the one of dimensionality 5 and rank 2
describes a particle of spin 0; the matrix of dimensionality 10 and rank 6 describes
a particle of spin 1.
The Lagrangian can be written, by choosing B = m, as

L = φ· iβ μ ∂μ φ − mφ· φ, (6.78)

which yields the equation of motion,

 
1
β μ ∂μ + m φ = 0. (6.79)
i

Multiplying through by β σ β ν 1i ∂σ 1i ∂ν , and symmetrizing with respect to σ and μ,

using the commutation relations to reduce the triple matrix products and using the
wave equation, we finally find
 
1 1 ν
∂ν ∂ + m φ = 0,
2
(6.80)
i i

which verifies what has been put into the theory as the connection between energy
and momentum. Thus each component of φ satisfies the Klein-Gordon equation.
The generator is 
G φ = −i dσ0 φβ 0 δφ. (6.81)

The commutation relations are developed from

  
 0 
φ(x), dσ0 φ(x )β δφ(x ) = −δφ(x), (6.82)

which yields   
φa (x), β 0 φ(x  )  (0)
= −δab δ (x − x  ), (6.83)
b

where δab is a diagonal matrix having six ones and four zeros along the diagonal,

which refer to the independent and dependent components, respectively. This equa-
tion cannot be solved for the commutation relations because β 0 is a singular matrix.
It eliminates the dependent components
of φ from the  commutation relations.
If we multiply the field equation by 1 − (β 0 )2 and use the commutation rela-
tions, we find   
β k β 0 β 0 ∂k + im 1 − β 0 β 0 φ = 0, (6.84)
78 6 Relativistic Theory of Fields

which is independent of time, and is the equation expressing the dependent compo-
nents of φ in terms of the independent components.
Let us re-examine the spin zero and spin one fields. Integral spin fields may be
described in terms of ordinary tensors. They do not require the special spinor proper-
ties of the half-integral spin fields. We have seen that the spin zero representation of
the Kenner-Duffin algebra has a dimensionality of five. If the only tensor in addition
to the field components which is introduced to form bilinear products in a scalar
Lagrangian is the four-divergence, we must describe the field with tensors differing
in rank by one. We shall construct the spin zero field with a scalar and four-vector
as the necessary 5 components, and the spin one field with a four-vector and an
anti-symmetrical second rank tensor as representing the 10 components.

6.5 Spin Zero

The Lagrangian for a spin zero field is

1
 m 2
L= φ· ∂μ φ μ − φ·μ ∂μ φ − φ − φμ φ μ . (6.85)
2 2
The field equations are determined by varying φ and φ μ ,

∂μ φ μ = mφ, ∂μ φ = mφμ , (6.86)

which imply

1 ν μ
∂ μ ∂μ φ = m∂ μ φμ = m 2 φ, ∂ μ ∂μ φ ν = ∂ ∂ ∂μ φ = m 2 φ ν , (6.87)
m
again yielding the Klein-Gordon equation.
The generator is
 
1
 1
G= dσμ φδφ μ − φ μ δφ → dσ0 φδφ 0 − φ 0 δφ . (6.88)
2 2

Thus only φ and φ 0 are independent variables. This is reflected in the fact that the
Klein-Gordon equation is second order, and hence we must specify both the wave
function and its time derivative.
The equations of motion have to be examined to see if the other components of
the field are determined. We have from the field equations, as equations of motion,

∂0 φ = mφ0 ∂0 φ 0 = mφ − ∂k φ k , (6.89)
6.5 Spin Zero 79

and the following which is not an equation of motion, but does show that the φ k are
determined in terms of the two independent components,

∂k φ = mφk . (6.90)

From the generator we see that we have one set of canonically conjugate variables,
φ and φ0 . This means that the field has only one internal degree of freedom, and
must be a spin zero field. This pair of conjugate variables must obey the canonical
commutation relations (at equal times), as is easily verified by using the generators
for their respective change
     
φ(x), φ(x  ) = φ 0 (x), φ 0 (x  ) = 0, φ 0 (x), φ(x  ) = iδ (0) (x − x  ), (6.91)

where δ (0) (x − x  ) means δ(r − r ) on σ0 . The commutation relations obeyed by

the dependent field components are derived from the field equations and these; for
instance
1 0  1
φ (x), φk (x  ) = − ∂k δ (0) (x − x  ). (6.92)
i m
These commutation relations can be written by inspection in a form not referring to
any particular coordinate system (where now δ (μ) (x − x  ) means δ(r − r ) on σμ ),
on the spacelike surface σμ ,

  1 μ 
φ(x), φ(x  ) = 0 φ (x), φ(x  ) = δ (μ) (x − x  ),
i
1 μ  1  
φ (x), φ ν (x  ) = − ∂ μ δ (ν) (x − x  ) + ∂ ν δ (μ) (x − x  ) . (6.93)
i m

6.6 Spin One

The description of a spin one field requires the use of a ten-component wave function.
As we shall see, not all of these components are independent. We shall use a four-
vector and an anti-symmetrical tensor of rank two. The Lagrangian is chosen to
be  
1
μν μν
 m μ 1 μν
L= φ ν · ∂μ G − G · ∂μ φ ν − φ φμ − G G μν . (6.94)
2 2 2

The factor of 1/2 in the last term is inserted because the unrestricted sum over μ and
ν counts each component twice. The field equations are

∂μ φν − ∂ν φμ = mG μν , ∂μ G μν = mφ ν . (6.95)
80 6 Relativistic Theory of Fields

The generator is 
1
G= dσ0 φk δG 0k − G 0k δφk . (6.96)
2

Thus only six of the ten components can be varied independently, φk and G 0k . Let
us see if they have equations of motion and if the other 4 components are determined
in terms of the independent ones. The equations of motion are

∂0 φk = ∂k φ0 + mG 0k , ∂0 G 0k = mφ k − ∂l G lk , (6.97)

for the independent field components. Also

∂k φ − ∂ φk = mG k , mφ 0 = ∂k G k0 = −∂k G 0k , (6.98)

which determine the four dependent field components. Thus all ten components of
the field are determined.
From the generator we see that we have 3 sets of canonically conjugate field
variables; thus the field has 3 internal degrees of freedom. This corresponds to the
three sub-states of a spin one field.
The commutation relations obeyed by the field components are obviously the
canonically conjugate relations,
   
φ k (x), φ  (x  ) = G 0k (x), G 0 (x  ) = 0,
1  0k 
G (x), φ (x  ) = δk δ (0) (x − x  ). (6.99)
i
The commutation relations obeyed by the dependent components can be realized
from the above, by using their definitions in terms of the independent components.
For instance,

1 0  1 1
φ (x), φ (x  ) = − δk ∂k δ (0) (x − x  ) = − ∂ δ (0) (x − x  ). (6.100)
i m m
These relations can be generalized to refer to an arbitrary coordinate system,

1 μ  1  μ (ν) 
φ (x), φ ν (x  ) = − ∂ δ (x − x  ) + ∂ ν δ (μ) (x − x  ) ,
i m
1  μν  μ
G (x), φλ (x ) = −δλν δ (μ) (x − x  ) − δλ δ (ν) (x − x  ),

i
1  μν  1  μλ  ν (κ) 
G (x), G λκ (x  ) = − g ∂ δ (x − x  ) + ∂ κ δ (ν) (x − x  )
i m
plus antisymmetrical terms . . . . (6.101)
6.7 Electromagnetic Field 81

6.7 Electromagnetic Field

The electromagnetic field is a spin one, massless field. A re-definition of the field
variables in the limit as the mass approaches zero will be made in the spin-one
Lagrangian. Set
√ 1
mG μν ≡ F μν , √ φν ≡ A ν . (6.102)
m

Then the Lagrangian (6.94) becomes

1
 1
L= Aν· ∂μ F μν − F·μν ∂μ Aν + F μν Fμν . (6.103)
2 4
The field equations are

∂μ Aν − ∂ν Aμ = Fμν , ∂μ F μν = 0, (6.104)

and the generator becomes, in the local coordinate system,


1
G= dσ Ak δ F 0k − F 0k δ Ak . (6.105)
2

Thus, as in the case of a spin one non-zero mass field, only six of the field components
can be varied independently. Their equations of motion are

∂0 Ak = ∂k A0 + F0k , ∂0 F 0k = −∂ F k . (6.106)

We must now examine the rest of the field equations to see if the dependent
components are determined,

∂k A − ∂ Ak = Fk , ∂k F 0k = 0. (6.107)

Thus F kl is determined, but because of the vanishing of the photon mass, A0 is not
determined and a further restriction is placed on the electric field, F 0k . This is exactly
the freedom of making a gauge transformation, which is allowed by the vanishing
of the photon mass. That is,
Aμ → Aμ + ∂μ . (6.108)

It is convenient to split the vector part of Aμ into a longitudinal part, which is the
gradient of a scalar, and a transverse part, which is the curl of a vector. Thus we see
that the gauge transformation affects only the longitudinal part of Ak . F 0k is purely
transverse because it has no divergence.
The equations of motion also break up into longitudinal and transverse components,

∂0 ALk = ∂k A0 ∂0 ATk = F0k . (6.109)

82 6 Relativistic Theory of Fields

The first equation is entirely consistent with the possibility of making a gauge trans-
formation which implies

ALk = ∂k , A0 = ∂0 . (6.110)

By setting
Ak = ∂k  + ATk (6.111)

in the generators and integrating by parts, or equivalently adding a surface term to

the Lagrangian, we see that only the transverse components of Ak appear in the
generator, 
1  T 0k
G = dσ0 Ak δ F − F 0k δ ATk . (6.112)
2

Since F 0k and ATk appear as canonical variables, we can impose the canonical
commutation relations
   
ATk (x), AT (x  ) = F 0k (x), F 0l (x  ) = 0, (6.113a)
   T
F 0k (x), AT (x  ) = δk δ (0) (x − x  ) , (6.113b)

where only the transverse component of the term on the right side of (6.113b) can
appear. If we define a term that subtracts off the longitudinal part of the delta function,
the commutation relation can be written as
 
F 0k (x), AT (x  ) = δk δ (0) (x − x  ) − ∂ k ∂ D0 (x − x  ). (6.114)

The divergence of this purely transverse equation must vanish, hence

δk ∂k δ (0) (x − x  ) = −∂ k ∂k ∂ D0 (x − x  ), (6.115)

which can be satisfied if

∇ 2 D0 (x − x  ) = −δ (0) (x − x  ), (6.116)

or
1
D0 (x − x  ) = , (6.117)
4π |x − x |

which shows that the longitudinal part of Ak is intimately connected with the
Coulomb potential.
6.8 Introduction of Charge 83

6.8 Introduction of Charge

It was previously mentioned that the description of charge requires a doubling of the
number of component fields χ .
The simplest case to consider is the case of one internal degree of freedom. Calling
the two basic fields χ(1) and χ(2) , L [Eq. (6.8)] becomes

L = χ(1)· Aμ ∂μ χ(1) + χ(2)· Aμ ∂μ χ(2) − H(χ(1) , χ(2) ). (6.118)

Since both χ(1) and χ(2) have identical space-time properties, the kinematical portion
of L is invariant under rotations and reflections in a new two-dimensional space
composed of the components χ(1) and χ(2) in accordance with the sum-of-squares
notation of (6.118). The rotations in this space, given by

χ (1) = cos λχ(1) + sin λχ(2) , χ (2) = − sin λχ(1) + cos λχ(2) , (6.119)

may be concisely expressed as

χ = eiλq χ , (6.120)

where    
χ(1) 0 −i
χ= q= . (6.121)
χ(2) i 0

The basic improper transformation (reflections) can be taken either as

χ (1) = χ(2) , χ (2) = χ(1) (6.122)

or
χ (1) = χ(2) , χ (2) = −χ(1) . (6.123)

Re-defining Aμ as  
Aμ 0
, (6.124)
0 Aμ

allows us to write (6.118) in matrix notation

L = χ· Aμ ∂μ χ − H(χ ), (6.125)

where we have assumed that H has the required invariance under reflections and
rotations in this space. Use of (6.120) and q T = −q is then sufficient to guarantee
the invariance of L under rotations. The reflection transformations in (6.122) can be
compactly written as
π
χ = Cei 2 q χ χ = Cχ , (6.126)
84 6 Relativistic Theory of Fields

where  
01
C= (6.127)
10

Then the relations C T = C −1 = C show that L is invariant under reflections.

The proper transformation (6.120) can be evolved from a sequence of infinitesimal
transformations of the form

χ = (1 + iδλq) χ = χ − δχ δχ = −iδλqχ , (6.128)

and incorporating this infinitesimal variation into the action principle will yield a
quantity which is conserved. The notation will suggest the interpretation of this
quantity as electrical charge, but until the manner in which it appears in interactions
between the various fields is specified, the nature of this charge is irrelevant.
If we now imagine that δλ of (6.128) is a continuous function of space time, the
finite change in λ which builds up between an initial t (σ2 ) and final t (σ1 ) times
corresponds to different successive choices of χ . The stipulated invariance of L then
implies that δλ L = 0. If δλ were constant, δλ L would vanish trivially, in accordance
with (6.120) and (6.126); we obtain something new since the dependence of δλ on
x μ introduces terms depending on ∂μ δλ:
 
δλ L = −iχ· Aμ qχ ∂μ δλ ≡ j μ ∂μ δλ = ∂μ j μ δλ − δλ∂μ j μ , (6.129)

where

j μ ≡ −iχ· Aμ qχ = −iχ Aμ qχ = χ(2) Aμ χ(1) − χ(1) Aμ χ(2) , (6.130)

The action principle now requires that the coefficient of δλ vanish, yielding the
differential conservation law
∂μ j μ = 0. (6.131)

The term ∂μ [ j μ δλ] gives the generator for changes on the boundary space-like
surfaces σ1 . σ2 . Taking δλ as constant over each surface, which corresponds to
different, but definite, choices of the χ on σ1,2 , we have

G λ = δλ dσμ j μ ≡ Qδλ. (6.132)

Further, if δλ(σ1 ) = δλ(σ2 ) then it is obvious that we get the same description as we
would have obtained with δλ = 0 in both surfaces, i.e.,

[Q(σ2 ) − Q(σ1 )] δλ = 0, and therefore Q(σ1 ) = Q(σ2 ). (6.133)

This result also follows from (6.131); the quantity Q, called the “charge", is therefore
conserved.
6.8 Introduction of Charge 85

Obtaining the commutation relations, from (6.128) and (6.132) we have

1 1
δχ = [χ , G] = [χ , Qδλ] = −iqχ δλ, (6.134)
i i
or
[χ , Q] = qχ , (6.135a)

or    
χ(1) , Q = −iχ(2) , χ(2) , Q = iχ(1) . (6.135b)

The generalization of (6.134) for a finite rotation is

e−iλQ χ eiλQ = eiλq χ , (6.136)

which is analogous to translation induced by the momentum operator Pν ,

1
[χ , Pν ] = ∂ν χ
i
ν
e−iα·P χ (x)eiα·P = eα ∂ν χ (x) = χ (x + a). (6.137)

There must also exist a unitary operator, in analogy with (6.136), which accom-
plishes the improper transformations of Eq. (6.123):

C −1 χ(1) C = χ(1)
C −1 χ C = cχ , or . (6.138)
C −1 χ(2) C = −χ(2)

From (6.132) and (6.130) we then have

C −1 QC = −Q. (6.139)

Thus, C has the interpretation of a charge reflection operator. Since C 2 = +1, its
eigenvalues are C = ±1. From (6.139) we see that [Q, C] = 0 and the two opera-
tors cannot be simultaneously diagonalized, (in the representation of χ(1) , χ(2) , C is
diagonal) except for states of zero charge. If we arbitrarily assign the vacuum state
the eigenvalue C  = +1 , then this state of zero charge, Q  = 0, has both a definite
charge and charge symmetry.
To obtain a state (other than Q  = 0 ) of definite charge, where the operator Q is
diagonalized, we must utilize non-Hermitian operators. Define

χ+ ≡ χ(1) − iχ(2) , χ− = χ(1) + iχ(2) , (6.140)

and then a simple calculation shows that

   
χ+ , Q = χ+ , χ− , Q = −χ− , (6.141)
86 6 Relativistic Theory of Fields

indicating that in this representation Q is diagonal. However, C is no longer diagonal:

C −1 χ± C = χ∓ . (6.142)

From (6.141) we see that if we have a state |Q   representing a definite charge,

then




Q χ+ |Q   = Q  − 1 χ+ |Q  




Q χ− |Q   = Q  + 1 χ− |Q   (6.143)

showing that χ+ creates a state for which the eigenvalue of Q is Q  −1, and conversely
for χ− These, then, represent charge annihilation and creation operators respectively.
If we imagine that every physical state can be created by the action of these operators
on the vacuum |Q  = 0 then the only possible values of Q  are 0, ±1, ±2, …etc.
(Note that charge annihilation represents the destruction of positive charge or the
creation of negative charge, and vice versa.) To connect this with our previous particle
description in Sect. 6.1, where we had χ (+) and χ (−) , we now obviously have the
four possibilities:
(+)
• χ+ destroys a particle and decreases charge by 1,
(−)
• χ− creates a particle and increases charge by 1.

(−) † (+)
These are inverse operations: χ− = χ+ .
(+)
• χ− destroys a particle but increases charge by 1.
(−)
• χ+ creates a particle but decreases charge by 1.

(+) † (−)
These are also inverse operations: χ− = χ+ . This exhaustive description
permits us to describe particles that carry charge.
In terms of the non-Hermitian variables defined by (6.140), the Lagrangian (6.118)
becomes
1 
L= χ−· Aμ ∂μ χ+ + χ+· Aμ ∂μ χ− − H, (6.144)
2
and the generator is

1 
G= dσμ χ− Aμ δχ+ + χ+ Aμ δχ− , (6.145)
2

which shows that χ+ and χ− are a canonically conjugate set of field variables. The
current (6.130) is
i  
jμ = χ+ A μ χ − − χ − A μ χ + . (6.146)
2
For variables of the second kind the commutation rules are
6.8 Introduction of Charge 87
 
χ+ (x), χ− (x  ) A0 = iδ (0) (x − x  ),
 
χ± (x), χ± (x  ) = 0. (6.147)

6.9 Quantum Electrodynamics

As an illustration of the use of non-Hermitian fields, we now consider the ordinary

charged Dirac field. We start with ψ(1) and ψ(2) as the two 4-component Hermitian
fields which form the simplest charged field. Let

ψ(1) − iψ(2) = ψ+ ≡ψ, (6.148a)

ψ(1) + iψ(2) = ψ− ≡ψ . †
(6.148b)

There is an artificial asymmetry here depending on what is defined as ψ and ψ † . The

commutation rules are now
 
ψα (x), ψβ† (x  ) = δαβ δ (0) (x − x  ),
   
ψα (x), ψβ (x  ) = 0 = ψα† (x), ψβ† (x  ) . (6.149)

Since in this case, A0 = i, the Lagrangian becomes:

1 † μ
  
L= ψ· α i ∂μ ψ − i ∂μ ψ † α μ ψ − mψ·† βψ. (6.150)
2 ·

The current is then

j μ = ψ·† α μ ψ. (6.151)

Setting
α μ = βγ μ , ψ † β = ψ, (6.152)

the Lagrangian is

1 
L= ψ · γ μ i∂μ ψ − i∂μ ψ · γ μ ψ − mψ · ψ. (6.153)
2
If we wish to interpret j μ as the electric current, we can write down the Lagrangian
for the electromagnetic field plus the charged Dirac field and specify the coupling
so that the usual field equations arise, in which j μ acts as the source for the vector
potential. The form of the coupling term is of course limited by Lorentz invariance
and the spin 21 and spin 1 algebra. Thus the most general Lagrangian which can be
formed in this case is
88 6 Relativistic Theory of Fields

1 
L= ψ · γ μ i∂μ ψ − i∂μ ψ · γ μ ψ − mψ · ψ
2
  1
+ Aν· ∂μ F μν − F·μν ∂μ Aν + F μν Fμν
4
μ 1 μν


+ e A· jμ + μF· ψ · σμν ψ . (6.154)
2
The symmetrization between Aμ and j μ is necessary because in general, the sources
for the Aμ are partly the jμ and hence these need not commute. The last term, a
Pauli moment, even though it appears covariant, may, in fact, not be covariant due to
its operator properties. These show up only in higher order terms in the perturbation
expansion.
The Lagrangian given by Eq. (6.154) is invariant under the rotation and reflection
of χ(1) and χ(2) , which correspond to the replacements

1
ψ → eiλ(x) ψ, ψ → e−iλ(x) ψ, Aμ → Aμ + ∂μ λ(x), (6.155)
e

and to the interchange of ψ and ψ, respectively. Note that in the second case—that
of reflection—we must also replace Aμ → −Aμ , Fμν → −Fμν , in order to have L
unaltered.
We obtain the equations of motion and generators by variations
   
1
δψ : γμ ∂μ − e Aμ· + m ψ = 0, (6.156a)
i
  
μ
δψ : ψ γ i∂μT − e Aμ· + m = 0, (6.156b)
δ Fμν : ∂μ Aν − ∂ν Aμ = Fμν , (6.156c)
μν μ μ
δ A μ : ∂ν F = eψ · γ ψ ≡ j . (6.156d)

The generators are


1 
G= dσμ ψiγ μ δψ − δψiγ μ ψ − Aν δ F μν − F μν δ Aν , (6.157)
2

and by adding appropriate 4-divergences, this becomes

   
 
G= dσμ ψiγ μ δψ − F μν δ Aν = dσ0 ψiγ 0 δψ − F 0k δ Ak , (6.158)

in a local coordinate system. In this form, the independent variables appear to be

only ψ and the Ak . ψ is definitely an independent variable; its equation of motion
is Eq. (6.156a). To re-examine the electromagnetic field, we write Eqs. (6.156c) and
(6.156d) as
6.9 Quantum Electrodynamics 89

∂0 Ak = ∂k A0 + F0k , Fk = ∂k A − ∂ Ak ,
∂0 F k0
= j − ∂ F ,
k k0
∂k F 0k = j 0 . (6.159)

Since we have the freedom of making a gauge transformation, wherein the longitu-
dinal components of Aμ are then arbitrary, it is advantageous to re-write equations
(6.159) in terms of transverse (T) and longitudinal (L) components:

(1) ∂0 ATk = F0k

T
, (2) ∂0 F k0T = j kT − ∂ F k ,
(3) Fk = Fk
L
= ∂k A − ∂ Ak , (4) ∂0 ALk = ∂k A0 + F0k
L
,
(5) ∂0 F k0L = j kL , (6) ∂k F 0kL = j 0 . (6.160)

L  = 0 since the electric field is no longer divergence-free. The solution of

Here, F0k
item (6) is immediately

F 0kL = −∂ k dσ (0) D(0) (x − x  ) j 0 (x  ), (6.161)

demonstrating that the longitudinal part of the electromagnetic field is not a kine-
matically independent quantity, but depends on the j 0 of all the charged particles
present. Note that item (5) is not an independent statement, since it is the result of
the conservation law for j μ (x). Since ALk = ∂k (x), using item (5) shows that

A0 = ∂0 (x) + dσ  D(0) (x − x  ) j0 (x  ), (6.162)

i.e., A0 is completely arbitrary, corresponding to the freedom of choice of (x).

The generator in Eq. (6.158) can now be written as
  
G= dσ ψiγ 0 δψ − F 0kT δ ATk − F 0kL δ ALk . (6.163)

By the addition of a surface term, the last term in the integrand in (6.163) becomes
equal to = j 0 δ, which does not refer to the kinematically independent part of the
electromagnetic field, and gives the generator for the change in the Dirac field when
an infinitesimal gauge transformation is made—See Eq. (6.132). The electromagnetic
field commutation rules are again as given in equations (6.113a) and (6.113b); the
anti-commutation relation for the Dirac fields works out to be
 
ψ(x), ψ(x  ) = γ 0 δ (0) (x − x  ) = −γ0 δ (0) (x − x  ) → −γμ δ (μ) (x − x  ). (6.164)

Using the non-Hermitian ψ, ψ, the effect of the charge reflection operator C is to

interchange the fields:

C −1 ψC = ψ † , C −1 ψ † C = ψ. (6.165)
90 6 Relativistic Theory of Fields

In order to maintain the invariance of L, we must now also require that this unitary
operator reverse the sign of the Aμ .

C −1 Aμ C = −Aμ . (6.166)

Again, C 2 = +1, and therefore its eigenvalues are C = ±1. From the arguments of
the preceding section, use of the transformation:

C −1 QC = −Q (6.167)

permits a classification, in terms of the quantum number c , for systems of zero net
charge. For example, consider the production of photons by repeated application of
the operator A (actually A(−) , a photon creation operator) to the vacuum state |0,
characterized by c = +1, and assume that no coupling terms are present in L (i.e.,
e = 0). If A is applied n times to |0 then by (6.166)

C (A · · · A) |0 = (−1)n (A · · · A) |0, (6.168)

and we have a state |(A · · · A ) whose eigenvalue of C depends on the (even or odd)
number of photons present: C  = (−1)n = ±1. Whichever +1 or -1 eigenvalue of c
we begin with persists as the coupling is turned on, even though what is now called
the state of n photons is a superposition of many different states.
As an example of this, let us consider the decay of positronium. In the absence of
the coupling term between Aμ and ψ (i.e., e = 0), the system of (e+ + e− ) is stable,
and has a definite eigenvalue c . When the coupling is introduced, the only permitted
final states will be those of the same eigenvalue. From our previous interpretation,
ψ is the operator which creates a particle e− and annihilates
e+ , and conversely
 for
ψ † . To obtain a state with e− at x and e+ at x  , we form ψ † (x)· ψ(x  ) |0, and
recognize that the actual state is a superposition of these with the amplitude wave
function ψ(x, x  ). Using (6.165) we then have
  
C|e+ e−  = C (d x)(d x  )ψ(x, x  ) ψ † (x)· ψ(x  ) |0
  
=− (d x)(d x  )ψ(x, x  ) ψ † (x  )· ψ(x) |0
  
=− (d x)(d x  )ψ(x  , x) ψ † (x)· ψ(x  ) |0. (6.169)

If the wave function ψ(x, x  ) is symmetric, then c = −1; if ψ(x, x  ) is anti-

symmetric, c = +1. As the coupling is now turned on, this no longer remains
the exact state, but the c value remains the same. Thus we have the selection rule
for positronium: from a 1 S state it can decay only into an even number of photons
(c = +1, ψ(x, x  ) is antisymmetric), from a 3 S state it can decay only in to an odd
number of photons (c = −1, ψ(x, x  ) is symmetric).
Chapter 7
Nonrelativistic Source Theory

The following is based on lectures given at Schwinger by Harvard in Spring 1969, as

transcribed by the author. The goal was to construct a general theory of particles, in
a nonrelativistic context. As such, this provides a transition between nonrelativistic
quantum mechanics and source theory, the general development of which was given
later in Schwinger’s three-volume treatise (Schwinger 1970b, 1973, 1989a).
The measurement symbol, or projection operator, which forms the basis for
Schwinger’s approach to quantum mechanics (Schwinger 2001),

|a  , b | = |a  b | (7.1)

represents a idealized process in which the state b is annihilated and a new state a 
is produced. Wouldn’t it be useful to separate these processes? Recall how oscillator
states, as discussed in Chap. 4, were created by forces. We generalize to sources which
can create and destroy particles. A scattering process, which occurs in a more or less
localized scattering region, can be abstracted into a two-stage process, in which first
an incoming particle is absorbed, and then a new particle is created, as sketched
in Fig. 7.1. In the individual processes the source K acts to create or annihilate the
particle.
What does this mean quantum mechanically? We must describe the processes
by probability amplitudes; we want to construct everything from the one particle
production mechanism,
1p+ |0−  K , 0+ |1p−  K , (7.2)

where the first amplitude represents the process in which a single particle state of
momentum p is created, where before the source acts only the vacuum state is present,
while the second represents the process in which a single particle of momentum p is
absorbed after which the vacuum state is present.
Because the processes occur in space and time, the source must be a function
K (r, t), which exhibits a certain degree of localizability. How does the effectiveness

© The Author(s) 2015 91

K.A. Milton, Schwinger’s Quantum Action Principle,
SpringerBriefs in Physics, DOI 10.1007/978-3-319-20128-3_7
92 7 Nonrelativistic Source Theory

Fig. 7.1 A scattering

process in which a particle of p1 p1
momentum p2 scatters into a
state of a particle with
momentum p1 can be
thought of as a combination
of two processes: One in
which the particle is created = +
by a source K in the state p1 K K
and a second in which the
particle of momentum p2 is
absorbed by the source K
p2 p2

of the source vary with different degrees of freedom? The complementary measure is
the corresponding function in momentum space K (p, E), where, nonrelativistically,
E = p 2 /(2m). We expect the relationship between the production amplitude and
the source function to be, at least for a weak source,

(dp)
1p+ |0−  K = (−i)K (p). (7.3)
(2π )3

Here (dp) = dp1 dp2 dp3 . The −i factor is purely conventional for later convenience.
The square root of the momentum-space element is present to properly account for
the density of states in the continuum picture. To compute the annihilation amplitude
0+ |1p− , we can use orthogonality, 1p− |0−  = 0. This must be maintained by the
dynamics. The completeness relation
 
1 = |0+ 0+ | + |1p+ 1p+ | + |1p+ 1p + 1p+ 1p + | + . . . , (7.4)
p p,p

implies

0 = 1p− |0+  K 0+ |0−  K + 1p− |1p +  K 1p + |0−  K + . . . , (7.5)
p

where we shall consider a weak source, so we will drop the higher terms. To lowest
order (in powers of the source)

0+ |0− 0 = 1, 1p− |1p + 0 = δpp , (7.6)

so
0 = 1p− |0+  K + 1p+ |0−  K , (7.7)
7 Nonrelativistic Source Theory 93

or  ∗
0+ |1p−  K = − 1p+ |0−  K . (7.8)

Therefore, the effectiveness of producing or absorbing a particle by a weak source is

 
(dp) (dp)
1p+ |0−  =
K
(−i)K (p), 0+ |1p−  K = (−i)K (p)∗ . (7.9)
(2π )3 (2π )3

These equations, in fact, define what we mean by a source.

Now we need to seek the relation to the space-time description. What happens
when a source is displaced,

K̄ (r, t) = K (r + R, t + T ). (7.10)

Relativity (here Galilean) means that the same effect occurs by displacing the space-
time coordinate system to which the initial and final states are referred, the generator
of such an infinitesimal displacement being

G = P · δ − H δt, (7.11)

which implies that the displacement operator is

U = eiP·R−iH T . (7.12)

Thus the 1 particle states change according to

1p | → 1p |eiP·R−iH T = eip·R−iE T 1p |, (7.13)

while the vacuum state is unchanged,

0+ | → 0+ |. (7.14)

Thus if the one particle amplitude is proportional to K (p),

1p+ |0−  K ∼ K (p), (7.15)

that of the displaced source is

1p+ |0−  K̄ ∼ eip·R−iE T K (p) ∼ K̄ (p). (7.16)

This implies that K (p) is obtained by Fourier transforming K (r, t),


K (p) = K (p, E) = (dr)dte−ip·r+iEt K (r, t). (7.17)
94 7 Nonrelativistic Source Theory

Fig. 7.2 Exchange of a 0+

particle between a source K 2
and a source K 1 , where the
latter is localized later than
the former. Before the first
source acts, the system is in K1
the vacuum state, as it is
after the second source acts.
Time is imagined as plotted
vertically in these “causal”
diagrams K2

0−

In general, one wants to remove the connection between E and p.

What about strong sources? Remember for the oscillator, the most basic object
was the ground-state persistence amplitude 0+ |0−  K . The latter contains the process
of the exchange of particles between temporally separated sources, as illustrated in
Fig. 7.2. Here a single particle is emitted by source K 2 and absorbed later by the
source K 1 . Before and after either source acts, the system is in the vacuum, no-
particle, state. We imagine the total source to be the sum of the two components,

K (x) = K 1 (x) + K 2 (x). (7.18)

The decomposition shown is called a causal arrangement. Because we are so far

considering weak sources, when we insert a complete set of states at an intermediate
time between the action of the two sources, the vacuum persistence amplitude is

0+ |0−  K = 0+ |0−  K 1 0+ |0−  K 2 + 0+ |1p−  K 1 1p+ |0−  K 2 + . . . . (7.19)
p

Now using the one-particle creation and annihilation amplitudes (7.9), together with
the Fourier transform (7.17) we see that the one-particle exchange term here is
   
  (dp) ip·(r−r )−iE(t−t  )
−i (dr)dt (dr )dt K 1∗ (r, t) −i e K 2 (r , t  ). (7.20)
(2π )3

But the source is a unitary whole—results can depend only on the total source K
and not its parts. This is a statement of the uniformity of space and time. This will
introduce terms that will refer to each component source separately; what we don’t
want is a term that involves K 2∗ and K 1 ; therefore, we define the retarded Green’s
function

   (dp) ip·(r−r )−iE(t−t  )
G(r − r , t − t ) = −iη(t − t ) e , (7.21)
(2π )3
7 Nonrelativistic Source Theory 95

Fig. 7.3 Exchange of 0+

noninteracting particles
between spatially
nonoverlapping sources.
Particles emitted by the
lower source of K (α) are
only detected by the upper
source of the same set. It is
K (1) K (2) K (3) ···
arranged that there is no
cross-coupling

0−

in terms of which we infer


0+ |0−  K = 1 − i (dr)dt (dr )dt  K ∗ (r, t)G(r − r , t − t  )K (r , t  ). (7.22)

The Green’s function satisfies the differential equation

1 2 
∂ i∇
i − G(r − r , t − t  ) = δ(t − t  )δ(r − r ). (7.23)
∂t 2m

Now we want to remove the restriction to weak sources. Suppose we have a beam
of noninteracting particles, detected for example by spatially separated sources, as
illustrated in Fig. 7.3. Each particle is produced and detected by a single pair of weak
sources K (α) , α = 1, 2, 3, . . . . There is no interaction between different pairs of
sources. Thus the vacuum persistence amplitude for this arrangement is

  (α)
0+ |0−  K = 1 − i (dr)dt (dr )dt  K ∗ (r, t)G(r − r , t − t  )K (r , t  ) ,
α
(7.24)
because each source is weak. But only K , not K (α) should enter; physics shouldn’t
depend on the channel. So
   
(K ∗ G K )(α) K (α)∗ G K (α) K ∗G K
0+ |0−  K = e−i = e−i α = e−i , (7.25)
α

where the last step depends upon the arrangement that prohibits cross coupling
between the component sources. This looks just like the structure we saw for the
harmonic oscillator (4.91), except now the integrals are over space as well as time:
96 7 Nonrelativistic Source Theory

Fig. 7.4 Exchange of 0+

noninteracting particles
between causally separated
sources, K 2 and K 1
K1

K2

0−


(dr)dt (dr )dt  K ∗ (r,t)G(r−r ,t−t  )K (r ,t  )
0+ |0−  K = e−i . (7.26)

We will apply this to generalized (not weak) emission and absorption processes, but
still assuming that the particles are not interacting.
Figure 7.4 shows the exchange of noninteracting particles between causally sep-
arated sources, K = K 1 + K 2 . The total vacuum persistence amplitude is

K 1∗ G K 2
0+ |0−  K = 0+ |0−  K 1 0+ |0−  K 2 e−i . (7.27)

The fact that there is a causal relation between the two sources means that

  (dp) ip·(r−r )−iE(t−t  )
G(r − r , t − t ) = −i e . (7.28)
(2π )3

Define the discrete specification of the source in momentum space as


(dp)
Kp = K ( p). (7.29)
(2π )3

Then the final exponential term in Eq. (7.27) is

 ∗ (iK
e p (−iK 1 p ) 2p)
. (7.30)

On the other hand, if we insert a complete set of multiparticle states at an intermediate

time, we have 
0+ |0−  K = 0+ |{n} K 1 {n}|0−  K 2 . (7.31)
{n}
7 Nonrelativistic Source Theory 97

Compare this with the expansion of Eq. (7.27):

 ∞ (−iK ∗ )n p (−iK ∗ )n p
1p 2p
0+ |0−  K = 0+ |0−  K 1   0+ |0−  K 2 , (7.32)
p n =0 n p ! n p !
p

where the occupation numbers in a given momentum cell are given by {n} = {n p }.
From this we infer the probability amplitudes for producing and absorbing particles
by a strong source:
(−iK p )n p
{n}|0−  K =  0+ |0−  K , (7.33a)
p n p !
(−iK p∗ )n p
0+ |{n} K =  0+ |0−  K . (7.33b)
p n p !

As a check of this, we verify that the total probability must be unity:

 
p |K p |
2
1= p({n}, 0) K = |0+ |0−  K |2 e . (7.34)
{n}

Independently,
 
|0+ |0−  | = exp − (dr)dt (r )dt  K ∗ (r, t)iG((r − r , t − t  )K (r , t  )
K 2

 
  ∗   ∗  
− (dr)dt (r )dt K (r, t)[iG((r − r, t − t)] K (r , t ) .

(7.35)

But the combinations of Green’s functions appearing here is


(dp) ip·(r−r )−iE(t−t  )
iG((r − r , t − t  ) + [iG((r − r, t  − t)]∗ = e , (7.36)
(2π )3

which is a solution of the homogeneous equation, so indeed


|0+ |0−  K |2 = e− p |K p |
2
, (7.37)

so the probability condition (7.34) is satisfied.

The description so far of the exchange of particles between sources is a sort of
action at a distance picture. We are often concerned with excitations produced by a
source—a more local description. A test source is used to measure effects. So let us
add an additional infinitesimal source,

K (r, t) → K (r, t) + δ K (r, t), (7.38)

98 7 Nonrelativistic Source Theory

which results in the following infinitesimal change in the action,


δW = − (dr)dt[δ K ∗ (r, t)ψ(r, t) + δ K (r, t)ψ ∗ (r, t)], (7.39)

which defines new objects which refer to the pre-existing situation. Here

ψ(r, t) = (dr )dt  G(r − r , t − t  )K (r , t) (7.40)

ψ ∗ (r, t) = (dr )dt  K ∗ (r , t)G(r − r, t  − t)K (r , t  ); (7.41)

the latter is not the complex conjugate of ψ, because

G(r − r , t − t  )∗ = G(r − r, t  − t), (7.42)

G being the retarded Green’s function. The differential equation satisfied by the
Green’s function (7.23) is
 
∂
i − T G(r − r , t − t  ) = δ(r − r )δ(t − t  ), (7.43)
∂t

where T = −∇ 2 /(2m) is the kinetic energy differential operator. Therefore, the field
ψ satisfies  
∂
i − T ψ(r, t) = K (r, t). (7.44)
∂t

This is analogous to the equation satisfied by the harmonic oscillator variable (4.65),
or  
d
i − ω y(t) = K (t). (7.45)
dt

Now because
 
∂
−i − T G(r − r, t  − t) = δ(r − r )δ(t − t  ), (7.46)
∂t

the field ψ ∗ satisfies

 
∂
−i − T ψ ∗ (r, t) = K ∗ (r, t), (7.47)
∂t

which is the complex conjugate equation. The boundary conditions are different in
the two cases: ψ is a retarded solution, while ψ ∗ is an advanced solution.
What does W have to do with action? Let us write the alternative forms
7 Nonrelativistic Source Theory 99
 
W =− (dr)dt K ∗ (r, t)ψ(r, t) = − (dr)dt ψ ∗ (r, t)K (r, t)
  
∗ ∂
= − (dr)dt ψ (r, t) i − T ψ(r, t). (7.48)
∂t

Combining these forms appropriately, we can write

    
∂
W = (dr)dt ψ ∗ i − T ψ − K ∗ ψ − ψ ∗ K . (7.49)
∂t

Think of this last as a functional of K , K ∗ , ψ, and ψ ∗ , so


 
δW = (dr)dt −δ K ∗ ψ − ψ ∗ δ K + δψ,ψ ∗ W. (7.50)

But the definition (7.40) of ψ and ψ ∗ shows that the last variation is zero, which is
a statement of the stationary action principle:
   
∂ ∗
δ ψ,ψ ∗ W = (dr)dt δψ i −T ψ−K
∂t
   
∂ ∗ ∗
+ (dr)dt δψ −i − T ψ − K = 0. (7.51)
∂t

That is, the stationary action principle, that W is unchanged under infinitesimal field
variations, supplies the equations of motion (7.44) and (7.47).
Fundamental to physics is the notion of the uniformity of space and time, that the
laws of physics are independent of the locale. This is reflected in the indistinguisha-
bility of identical particles, which, in turn, is reflected in the probability amplitude
of a source producing a multi-particle distribution, Eq. (7.33a), which says all that
is possible. Experimentally, we know of two kinds of statistics. Here n p is unlim-
ited, so this cannot refer to Fermi-Dirac statistics. In the Bose-Einstein case we have
stimulated emission. Let us see this.
Figure 7.5 shows the interchange of noninteracting particles between sources K 2
and K 1 , but now with a weak source K 0 in between. The total source is composed
of three causally separated pieces,

K = K1 + K0 + K2. (7.52)

Using this causal arrangement, the vacuum persistence amplitude is


K ∗G K
0+ |0−  K = e−i
 
K 1∗ G K 0 + K 0∗ G K 2 ]
= 0+ |0−  K 1 +K 2 0+ |0−  K 0 e−i[ . (7.53)
100 7 Nonrelativistic Source Theory

Fig. 7.5 Effect of a weak 0+

source K 0 acting at an
intermediate time between
strong sources K 2 and K 1
K1

K0

K2

0−

Here, because the disturbance by K 0 is regarded as weak, we approximate

0+ |0−  K 0 ≈ 1, and the exponential term is expanded to first order in K 0 :

0+ |0−  K ≈ 0+ |{n} K 1 {n}|0−  K 2
{n}
⎧ ⎫
⎨  ⎬
∗ ∗
× 1+ [(−iK 1p )(−iK 0p ) + (−iK 0p )(−iK 2p )] . (7.54)
⎩ p
⎭

Compare this with the multi-particle exchange description between the three sources,

0+ |0−  K = 0+ |{n}−  K 1 {n}+ |{n}−  K 0 {n}+ |0−  K 2 . (7.55)
{n}{n}

Now recall the connection between K p and {n}|0−  K , Eq. (7.33b), so we infer

∗

0+ |{n} K 1 (−iK 1p )= n p + 10+ |{n} + 1p  K 1 . (7.56)

Therefore, for a weak source,


{n + 1p }|{n} K = −iK p n p + 1. (7.57)

The corresponding probability of creating 1 more particle with momentum p is

7 Nonrelativistic Source Theory 101

|{n + 1p }|{n} K |2 = |K p |2 (n p + 1). (7.58)

In the last factor, the 1 represents spontaneous emission, and the n p is the enhance-
ment effect of stimulated emission.
The corresponding analysis in the absorption case gives

{n}|{n + 1p } K = −iK p∗ n p + 1, (7.59)

or
√
{n − 1p }|{n} K = −iK p∗ n p , (7.60)

that is, the probability of absorbing one particle is proportional to the incident inten-
sity.
Let us come back to space and time. Adopting a more telegraphic notation, we
can write the vacuum persistence amplitude as
 ∗ 
0+ |0−  K = e−i K G K = 1 − i d1 d1 K ∗ (1)G(1 − 1 )K (1 )

1
− d1 d1 d2 d2 K ∗ (1)K ∗ (2)G(1 − 1 )G(2 − 2 )K (1 )K (2 ) + . . . .
2
(7.61)

Here the numbers represent space-time points, 1 = r1 , t1 , etc. Because the product
of sources is symmetrical, we can replace

1
G(1 − 1 )G(2 − 2 ) → [G(1 − 1 )G(2 − 2 ) + G(1 − 2 )G(2 − 1 )], (7.62)
2
Diagrammatically, the two terms can be represented as in Fig. 7.6. This builds in the
symmetry in the labels. The third term in Eq. (7.61) is

1
− d1 d2 K ∗ (1)K ∗ (2)ψ(1)ψ(2), ψ(1)ψ(2) = ψ(1, 2) = ψ(2, 1), (7.63)
2

Fig. 7.6 Exchange of two particles between spatially and temporally separated sources. Bose-
Einstein symmetry implies that the particles are exchanged between either spatially separated source.
No interaction is to be inferred where the lines cross
102 7 Nonrelativistic Source Theory

which exhibits another characteristic of Bose-Einstein statistics. The field ψ, the

generalization of the wavefunction, must be totally symmetric under interchange of
the particles.

7.1 Interactions

Real particles interact with each other. Thus we should have processes such as
sketched in Fig. 7.7. The particles emitted by the sources are scattered by inter-
actions represented by the black box labeled T. Let us begin with a simpler situation,
scattering from a fixed center, for example, a heavy nucleus. This process can be
represented with only two sources, as shown in Fig. 7.8. This diagram schematically
represents the tracks of particles as seen in a detector. In the absence of interactions

W = − K ∗G0 K , (7.64)

where now we have used the superscript 0 to designate the free particle propagator
or Green’s function previously denoted simply by G. This process in represented
by the dotted line in Fig. 7.8. Now we want to add something to this, the scattering
process, represented generically by T. From a source point of view, both processes
contribute to the vacuum amplitude
 
W = − K ∗ G 0 K − K ∗ G 0 TG 0 K , (7.65)

Fig. 7.7 Two particles

produced by the two earlier
sources scatter by the
processes labeled T and the
scattered particles are
detected by the two later
sources

T
7.1 Interactions 103

Fig. 7.8 Diagram

representing scattering of a
particle off a fixed center.
Here the heavy vertical line
represents the fixed center or
nucleus, the dotted line
represents the exchange of a
particle directly between the
sources, and the solid lines
represent the production of a
particle by the first source, T
its scattering off the fixed
center, by the process
represented by the black box
T, and finally its absorption
by the second source

where the last term means


− K ∗ (1)G 0 (1 − 1 )T(1 , 1 )G 0 (1 − 1 )K (1 )

= − ψ 0∗ (1 )T(1 , 1 )ψ 0 (1 ). (7.66)

The vacuum persistence amplitude is expressed as

0+ |0−  = eiW (7.67)

for the same reason as before. Because we have a well-defined causal situation,

ψ 0 (1) = G 0 (1 − 1 )K (1 ), t1 > t1 , (7.68)

we can write

 (dp) ip·r−iEt 
ψ (r, t) =
0
e (−i)K p ≡ ψp (r, t)(−i)K p . (7.69)
p
(2π ) 3
p

Similarly, 
ψ (1) =
0∗
K ∗ (1 )G 9 (1 − 1), t1 < t1 (7.70)
104 7 Nonrelativistic Source Theory

Fig. 7.9 Multiple-scattering

resolution of the scattering
operator T. Here the diagram
represents repetitions of
elementary scattering
processes each described by
a potential V

V T

implies

 (dp) −ip·r+iEt 
ψ (r, t) =
0∗
e (−i)K p∗ ≡ ψp∗ (r, t)(−i)K p∗ . (7.71)
p
(2π ) 3
p

The term iW is the only term which describes the process being considered. The
vacuum amplitude includes
 ∗ 
0+ |1p  K 1p+ |1p − 1p + |0 K ≈ (−iK p∗ )1p+ |1p − (−iK p ). (7.72)
p p

So, picking out the coefficients of −iK p , −iK p∗ , we infer


1p |1p  = −i ψp (1)T(1, 1 )ψp (1 ). (7.73)

so if we knew T we could compute the scattering amplitude by taking matrix elements

in this way.
Let us analyze the scattering process in more detail, by going into processes which
occur in a definite time. That is, we break up the extended process, represented by the
black box in Fig. 7.8 by viewing it as a repetition of elementary processes represented
by a potential V , as sketched in Fig. 7.9. To describe this, we extend the idea of the
field. The vacuum amplitude is given in terms of

W = − K ∗ [G 0 K + G 0 TG 0 K ], (7.74)
7.1 Interactions 105

where the quantity in square brackets is regarded as the field,


ψ = ψ0 + G 0 Tψ 0 , (7.75)


where ψ 0 = G 0 K and Tψ 0 may be thought of as an effective source. In this way
we get a superposition of effects. Alternatively, we may emphasize the last scattering
act, 
ψ = ψ + G 0 V ψ,
0
(7.76)

where, as above, ψ is due to an infinite number of elementary scattering acts. This

decomposition is self-consistent, because we may write

ψ = ψ0 + G0V ψ
  
=ψ + 0
G Vψ +
0 0
G VG Vψ +
0 0 0
G0V G0V G0V ψ + . . . . (7.77)

So if the expansion makes sense

T = V + V G0V + V G0V G0V + . . . . (7.78)

The differential equation corresponding to Eq. (7.76) is

 
∂
i − T ψ = K + V ψ, (7.79)
∂t
or  
∂
i − T − V ψ = K. (7.80)
∂t

So we see, indeed, that V is the potential energy. Thus, we indeed have a multiple
scattering process.
This also generalizes the concept of the Green’s function. If we write

ψ= GK, (7.81)

where the Green’s function satisfies

 
∂
i − T − V G(1, 1 ) = δ(1, 1 ), (7.82)
∂t

and the action can be written as

106 7 Nonrelativistic Source Theory
    
∂
W =− K ∗ψ − K ψ∗ + ψ ∗ i − T − V ψ, (7.83)
∂t

which in value is equal to 

W = −i K ∗G K . (7.84)

The Green’s functions include the possibility of bound states.

More generally, think of 2-particle scattering, represented in Fig. 7.7. Single-
particle exchange is represented by

W2 = − K ∗G0 K . (7.85)

But here, the particles do their thing, and the scattering is represented by

1
W4 = − (K ∗ G 0 )1 (K ∗ G 0 )2 T(12, 1 2 )(G 0 K )1 (G 0 K )2 . (7.86)
2

Again, the scattering amplitude 1p1 1p1 |1p1 1p2  is given simply in terms of T.
Moreover, we want to analyze the scattering in terms of elementary processes, as
sketched in Fig. 7.10. The terms in the vacuum amplitude of interest are contained
in ei(W2 +W4 ) which when expanded contains the following terms with four sources:

 
1 2 1 ∗ 0 ∗ 0 i
− W2 + iW4 = − K ψ K ψ − (K ∗ G 0 )(K ∗ G 0 )T(G 0 K )(G 0 K )
2 2 2

1
=− K ∗ (1)K ∗ (1 )ψ(1, 2), (7.87)
2

Fig. 7.10 Scattering

process described in terms of
multiple scattering

V T
7.1 Interactions 107

where 
ψ(1, 2) = ψ (1)ψ (2) + i
0 0
G 01 G 02 Tψ 0 ψ 0 . (7.88)

This says that the source detects what comes directly, and what comes from the
scattering process. On the other hand, we can look at this from the last scattering
point of view,

ψ(1, 2) = ψ 0 (1)ψ 0 (2) + i G 01 G 02 V ψ(1 , 2 ), (7.89)

where V is the measure of the single-scattering process. Again one can iterate. Look
at the corresponding differential equation,
  
∂ ∂
i − T1 i − T2 ψ(1, 2) = K (1)K (2) + i(V ψ)(1, 2), (7.90)
∂t1 ∂t2

where a matrix notation is used in the last term. This can be written as
   
∂ ∂
i − T1 i − T2 − iV ψ(1, 2) = K (1)K (2). (7.91)
∂t1 ∂t2

Now make explicit the time localization of the potential:

V (1, 2; 1 , 2 ) = δ(t1 − t2 )δ(t1 − t2 )δ(t1 − t1 )V12 (r1 − r2 , r1 − r2 ). (7.92)

It might also be that the potential is localized in space:

V12 (r1 − r2 , r1 − r2 ) = δ((r1 − r2 ) − (r1 − r2 ))V (r1 − r2 ), (7.93)

but we won’t assume this in the following. Then the Green’s function defined by

ψ(1, 2) = G(1, 2; 1 , 2 )K (1 )K (2), (7.94)

satisfies
   
∂ ∂

i − T1 i − T2 − iV G = δ(1 − 1 )δ(2 − 2 ) sym , (7.95)
∂t1 ∂t2

where the subscript denotes symmetrization. Now because of Eq. (7.89), the integral
equation for the Green’s function is (symmetrization suppressed)

G = G 01 G 02 + iG 0 G 0 V G. (7.96)
108 7 Nonrelativistic Source Theory

The assumption of an instantaneous V means that we can look at the equal-time

Green’s function,

G(r1 , r2 , t; r1 , r2 , t  ), t = t1 = t2 , t  = t1 = t2 . (7.97)

Appearing in Eq. (7.96) is


(dp1 ) (dp2 )
iG 0 (r1 , t; r1 , t  )G 0 (r2 , t; r2 , t  ) = −iη(t − t  )
(2π )3 (2π )3
× exp[i(p1 · (r1 − r1 ) + i(p2 · (r2 − r2 ) − i(E 1 + E 2 )(t − t  )], (7.98)

a Green’s function with energy E 1 + E 2 . This obeys the equation

 
∂
i − T1 − T2 (iG 0 G 0 ) = δ(t − t  )δ(r1 − r1 )δ(r2 − r2 ), (7.99)
∂t

the conventional Green’s function equation. Then

 
∂
i − T1 − T2 iG = δ(t − t  )δ(r1 − r1 )δ(r2 − r2 ) + V12 (iG), (7.100)
∂t

which is what we would expect.

Now we write the vacuum amplitude as
 
1
0+ |0−  K = 1 − i K ∗G0 K − K ∗ K ∗ ψ(12) + . . .
2
 
K ∗ G 0 K − 12 K ∗ K ∗ [ψ(12)−ψ(1)ψ(2)]
= e−i + ..., (7.101)

where in the last three-particle and higher interactions have been omitted. Here we
include in the second term only the effects of interaction, through

χ (1, 2) = ψ(1, 2) − ψ(1)ψ(2). (7.102)

The integral equation (7.96) written in terms of the last interaction can also be
written in term of the first interaction,

G = G 0 G 0 + GiV G 0 G 0 . (7.103)

Symbolically we can solve for G,

1 1
G= G0G0 = G0G0 , (7.104)
1 − G G iV
0 0 1 − iV G 0 G 0
7.1 Interactions 109

which are formally identical. This assumes the instantaneous interaction given by
Eq. (7.92). Let us write

G 0 G 0 = −iG 01+2 , iG = G 1+2 ; (7.105)

then our integral equation reads

G 1+2 = G 01+2 + G 01+2 V12 G 1+2 . (7.106)

which satisfies the differential equation (7.100), or

 
∂
i − T1 − T2 − V12 G 1+2 = δ(t − t  )δ(r1 − r1 )δ(r2 − r2 ). (7.107)
∂t

We might be interested in non-instantaneous initial and final states, even though

we are assuming an instantaneous interaction. Then we could write
  
G = iG 0 (t> − t< )G 12 (t< − t> )G 0 (t> − t< ), (7.108)

where
G 0 (t − t  ) → −iδ(r − r ) as t → t  + 0, (7.109)

hence the factor of i is supplied by comparison with the instantaneous limit, t1 = t2 ,

t1 = t2 .

7.2 Bound States

Sources must be able to create composite structures. We could have started with
composite particles, after all, so this is a aspect of self-consistency. Write the integral
equation for G symmetrically,

G = G 0 G 0 + G 0 G 0 iV [G 0 G 0 + GiV G 0 G 0 ] = G 0 G 0 + G 0 G 0 iV G 0 G 0
+ G 0 G 0 iV GiV G 0 G 0 . (7.110)

We are still working with the instantaneous interaction approximation. So G has the
property of depending on two times. Write again

G = −iG 1+2 , V → V12 . (7.111)

The description of the bound states is contained in G 1+2 . The part of the vacuum
amplitude involving G 1+2 is
110 7 Nonrelativistic Source Theory

i

K ∗ K ∗ G 0 G 0 V12 G 1+2 V12 G 0 G 0 K K
0+ |0−  = e− 2 . (7.112)

This involves all interactions.

Now introduce the coordinates for the two-particle system, where we assume that
the two particles have the same mass,

1
R= (r1 + r2 ), r = r1 − r2 . (7.113)
2
The Green’s function equation (7.107) becomes
 
∂ P2 p2
i − − − V (r) G(r, R, t; r , R , t  ) = δ(t − t)δ(r − r )δ(R − R ),
∂t 2M 2μ
(7.114)
where we see the appearance of the total mass M, the reduced mass μ, the total
momentum P and the relative momentum p. We are interested in the motion of the
center of mass. Let the internal motion be described by an eigenfunction φk governed
by a Schrödinger equation,
 
p2
Ek − − V φk (r) = 0, (7.115)
2μ

so multiplying Eq. (7.114) by φn∗ (r) and integrating over r we obtain

 
∂ P2
i − − En (dr)φ ∗ (r)G(r, R, t; r , R , t  )
∂t 2M
= δ(t − t  )δ(R − R )φ ∗ (r ). (7.116)

The Green’s function can be expanded, therefore, in terms of a single-particle Green’s

functions depending on the state n:

G(r, R, t; r , R , t  ) = φn (r)G n (R − R , t − t  )φn∗ (r ). (7.117)
n

Now from Eq. (7.112) we identify the effective source for an atomic state

1 r
K n (R, t) = √ (dr)φ ∗ (r)V12 (r)G 0 (R + − r1 , t − t1 )
2 2
r
× G 0 (R − − r2 , t − t2 )K (r1 , t1 )K (r2 , t2 )d1 d2. (7.118)
2
We must be very explicit that E n < 0, meaning that the sources are extended: They
must put out less energy than two free particles would have. So G 0 does not refer to
the propagation of a free particle, but rather it propagates an excitation which does
7.2 Bound States 111

not get very far. As a result, the Green’s functions become real in effect, by virtue of
the integration, and

1
K n∗ (R, t) = √ K ∗ K ∗ G 0 G 0 V12 φn (7.119)
2

is actually the complex conjugate of Eq. (7.118). The consistency of the vacuum
persistence amplitude
 
K n∗ (R,t)G n (R−R ,t−t  )K n (R ,t  )
0+ |0−  = e−i n (7.120)

demands that K n∗ really be the complex conjugate of K n ,

Perhaps this remark is clarified by looking at the Fourier transform of the free
Green’s function,
1
. (7.121)
E − T + i

When E < 0 there is no singularity and the Green’s function is real.

The formalism is flexible: it can deal with bound states whether analyzed or not,
which is especially useful in high energy physics.
Can we write down a formalism that expresses the dynamics and supplies the field
equations? The affirmative answer is supplied by writing

0+ |0−  = eiW , (7.122)

with
   
∂ 1
W = ψ i − T ψ − K ∗ ψ − ψ ∗ K − ψ ∗ ψ ∗ V ψψ
∗
∂t 2
1 1
− ψ ∗ ψ ∗ V χ − χ ∗ V ψψ
2  2   
i ∗ ∂ ∂
− χ i −T i − T − iV χ . (7.123)
2 ∂t ∂t

Here we have introduce the 2-particle field χ . What does this imply under field
variations? Varying with respect to ψ ∗ gives
 
∂
i − T ψ = K + ψ ∗ V (ψψ + χ ), (7.124)
∂t

where in the last we see a realistic representation of a single-particle source in the

interaction with the other fields. Varying with respect to χ ∗ yields
   
∂ ∂
−T i − T − iV χ = iV ψψ. (7.125)
∂t ∂t
112 7 Nonrelativistic Source Theory

This says that χ is the part of the two-particle field that has interacted at least once,
thus
χ = GiV ψψ. (7.126)

If we use these equations to evaluate W we recover the previous result (7.112).

We have here two independent field ψ and χ , coupled by interaction. This is the
essence of the many-particle situation in high energy physics. Essentially, we may
regard the − 21 χ ∗ V ψψ as a phenomenological coupling between particles.
This nonrelativistic development of source theory was written down after the rela-
tivistic formulation was developed, which is explicated in great detail in Schwinger’s
three volume treatise (Schwinger 1989a).
Chapter 8
Concluding Remarks

We have traced Schwinger’s development of action formulations from classical sys-

tems of particles and fields, to the description of quantum dynamics through the
Quantum Action Principle. In the latter, we have described quantum mechanical sys-
tems, especially the driven harmonic oscillator. This is ahistorical, since Schwinger
first developed his quantum dynamical principle in the context of quantum electro-
dynamics in the early 1950s, and only nearly a decade later applied it to quantum
mechanics, which is field theory in one dimension—time. At roughly the same time
he was thinking about quantum statistical systems (Martin 1959), and it was natural
to turn to a description of nonequilibrium systems, which was the motivation of the
time-cycle method, although Schwinger put it in a general, although simplified, con-
text. The time cycle method was immediately applied to quantum field theory by his
students, K. T. Mahanthappa and P. M. Bakshi (Mahanthappa 1962; Bakshi 1963).
Then we give a sketch of the application of these methods to quantum field theory,
based on Schwinger’s 1956 Stanford lectures, and to what Schwinger perceived as the
successor to field theory, Source Theory. The latter appeared shortly after he received
his Nobel prize in 1963. In fact the present document, which includes components
of Julian Schwinger’s thinking over nearly four decades, shows that these develop-
ments proceeding organically, and that the action principle and Green’s functions1
played central roles throughout his remarkable career in physics. In fact, it has been
argued (Mehra 2000) that the first “source theory” paper was in fact his most cited
one, written in 1951 (Schwinger 1951a).

1 One of Schwinger’s last publications (Schwinger 1993) described the centrality of Green’s func-
tions to his life work.
© The Author(s) 2015 113
K.A. Milton, Schwinger’s Quantum Action Principle,
SpringerBriefs in Physics, DOI 10.1007/978-3-319-20128-3_8
114 8 Concluding Remarks

Acknowledgments I thank the Laboratoire Kastler Brossel, ENS, UPMC, CNRS, for its hospitality
during the completion of the first part of manuscript. I especially thank Astrid Lambrecht and Serge
Reynaud. The work was supported in part with funding from the Simons Foundation, CNRS,
and the Julian Schwinger Foundation. I thank my many students at the University of Oklahoma,
where much of the material reported here was used as the basis of lectures in electrodynamics,
quantum mechanics, and quantum field theory. I especially thank Herb Fried for permission to use
his transcription of Schwinger’s 1956 lectures as the basis for Chap. 6 here, and Walter Becker for
his LATEX conversion of those notes.
References

P.M. Bakshi, K.T. Mahanthappa, Expectation value formalism in quantum field theory. J. Math.
Phys. 4, 1 (1963)
P.A.M. Dirac, The physical interpretation of the quantum dynamics. Proc. Roy. Soc. Lond. A
113(765), 621–641 (1927)
P.A.M. Dirac, V.A. Fock, B. Podolsky, On quantum electrodynamics. Phys. Zeits. Sowjetunion 2,
468 (1932)
P.A.M. Dirac, The Lagrangian in quantum mechanics. Phys. Zeits. Sowjetunion 3, 64 (1933)
R.J. Duffin, On The characteristic matrices of covariant systems. Phys. Rev. 54, 1114 (1938). doi:10.
1103/PhysRev.54.1114
L. Euler, Methodus Inveniendi Lineas Curvas Maximi Minive Proprietate Gaudentes (Bousquet,
Lausanne and Geneva, 1744)
L. Euler, Investigation of the letter, allegedly written by Leibniz, translated by Wikisource (1752),
http://en.wikisource.org/wiki/Investigation_of_the_letter_of_Leibniz
R.P. Feynman, The principles of least action in quantum mechanics. Ph.D dissertation, Princeton
University, Princeton, NJ. (University Microfilms, Ann Arbor, Publications No. 2948) (1942)
R.P. Feynman, Space-time approach to quantum electrodynamics. Phys. Rev. 76, 769 (1949)
R.P. Feynman, A.R. Hibbs, Quantum Mechanics and Path Integrals (McGraw-Hill, New York,
1965)
R.J. Glauber, Coherent and incoherent states of radiation field. Phys. Rev. 131, 2766–2788 (1963)
W.R. Hamilton, On a general method in dynamics. Part I. Phil. Trans. Roy. Soc. 124, 247–308
(1834)
W.R. Hamilton, On a general method in dynamics. Part II. Phil. Trans. Roy. Soc. 125, 95–144 (1835)
W. Heisenberg, W. Pauli, Zur Quantendynamik der Wellenfelder. Z. Phys. 56, 1–61 (1929)
L.V. Keldysh, Diagram technique for nonequilibrium processes. Zh. Eksp. Teor. Fiz. 47, 1515–1527
(1964) (English translation: Soviet Physics JETP 20: 1018–1026 (1965).)
N. Kemmer, The particle aspect of meson theory. Proc. Roy. Soc. A 173, 91–116 (1939). doi:10.
1098/rspa.1939.0131
J.-L. Lagrange, Méchanique Analytique. p. 226 (1788)
K.T. Mahanthappa, Multiple production of photons in quantum electrodynamics. Phys. Rev. 126,
329 (1962)
P.C. Martin, J. Schwinger, Theory of many-particle systems. Phys. Rev. 115, 1342 (1959)
P.L.M. de Maupertuis, Accord de différentes lois de la nature qui avaient jusqu’ici paru incompati-
bles”. Mém. As. Sc. Paris p. 417 (1744)

© The Author(s) 2015 115

K.A. Milton, Schwinger’s Quantum Action Principle,
SpringerBriefs in Physics, DOI 10.1007/978-3-319-20128-3
116 References

P.L.M. de Maupertuis, Le lois de mouvement et du repos, déduites d’un principe de métaphysique.

Mém. Ac. Berlin, p. 267 (1746)
J. Mehra, K.A. Milton, Climbing the Mountain: The Scientific Biography of Julian Schwinger
(Oxford University Press, Oxford, 2000)
K.A. Milton, J. Schwinger, Electromagnetic Radiation: Variational Methods. Waveguides and
Accelerators (Springer, Berlin, 2006)
K.A. Milton, In Appreciation Julian Schwinger: From nuclear physics and quan-
tum electrodynamics to source theory and beyond. Phys. Persp. 9, 70–114 (2007).
http://arxiv.org/abs/physics/0610054physics/0610054
E. Noether, Invariante Variationsprobleme. Nachr. König. Gesellsch. Wiss. Göttingen. Math-phys.
Klasse, 235–257 (1918)
G. Petiau, University of Paris thesis. Acad. Roy. de Belg., A. Sci. Mem. Collect. 16(2), 1 (1936)
E. Schrödinger, Der stetige Übergang von der Mikro- zur Makromechanik. Naturwissenschaften
14, 664–666 (1926)
J. Schwinger, On gauge invariance and vacuum polarization. Phys. Rev. 82, 664 (1951)
J. Schwinger, The theory of quantized fields. I. Phys. Rev. 82, 914 (1951)
J. Schwinger, Theory of quantized fields. III. Phys. Rev. 91, 728–740 (1953)
J. Schwinger, Differential Equations of Quantum Field Theory, lectures transcribed by R. Blanken-
becler, H. Fried, J. Tiemann, issued by Stanford Research Institute, and transcribed into LaTeX
by W. Becker (1956)
J. Schwinger, The geometry of quantum states. Proc. Natl. Acad. Sci. USA 46, 257 (1960)
J. Schwinger, Unitary operator bases. Proc. Natl. Acad. Sci. USA 46, 570 (1960)
J. Schwinger, Unitary transformations and the action principle. Proc. Natl. Acad. Sci. USA 46, 883
(1960)
J. Schwinger, The special canonical group. Proc. Natl. Acad. Sci. USA 46, 1401 (1960)
J. Schwinger, Brownian motion of a quantum oscillator. J. Math. Phys. 2, 407 (1961)
J. Schwinger, Quantum Kinematics and Dynamics (Benjamin, New York, 1970)
J. Schwinger, Particles, Sources, and Fields (Addison-Wesley, Reading, 1970)
J. Schwinger, A Report on quantum electrodynamics, in J. Mehra, The Physicist’s Conception of
Nature (Reidel, Dordrecht, 1973)
J. Schwinger, Particles, Sources, and Fields, vol. II (Addison-Wesley, Reading, 1973)
J. Schwinger, Particles, Sources, and Fields, vol. I-III. (Addison-Wesley, Redwood City, 1989)
J. Schwinger, A path to quantum electrodynamics. Phys. Today (1989) [Reprinted in Most of the
Good Stuff: Memories of Richard Feynman. (eds. L. M. Brown, J.S. Rigden) (AIP, New York,
1993)]
J. Schwinger, The Greening of quantum field theory: George and I (1993),
http://arxiv.org/abs/hep-ph/9310283arXiv:hep-ph/9310283 (Published in J. Schwinger, The
Physicist, the Teacher, and the Man, ed. Y. J. Ng (World Scientific, Singapore 1996), pp. 13–27
J. Schwinger, L.L. DeRaad Jr, K.A. Milton, W.-y. Tsai, Classical Electrodynamics
(Perseus/Westview, New York, 1998)
J. Schwinger, Quantum Mechanics: Symbolism of Atomic Measurements (Springer, Berlin, 2001)
A. Sommerfeld, Mechanics-Lectures on Theoretical Physics, vol. I (Academic Press, New York,
1964)
D. Toms, The Schwinger Action Principle and Effective Action (Cambridge Monographs on Math-
ematical Physics, 2007)
H. Weyl, Eine neue Erweiterung der Retativitätstheorie. Ann. der Phys. 59, 101–133 (1919)