Aspects of Quantum Field Theory

5.
ASPECTS OF QUANTUM FIELD THEORY

T.PADMANABHAN
lnter- University Center for Astronomy (1 Astrophysics
Post Bag No.4, Ganeshkhind
Pune 411 007, India
1. General Introduction
These lectures concentrate on a series of topics in quantum field theory
which, in my opinion, are becoming increasingly relevant to the students of
general relativity and gravity. Since there exists several formal textbooks
in quantum field theory, it is probably worthwhile to explain how these
lectures differ from standard text books and what exactly their scope is.
For most part of the lectures I have chosen topics which are not discussed
in adequate details in standard textbooks. Even while dealing with standard
material I have tried to concentrate on subtleties and questions of technique
which are not usually emphasised. This is done because it is precisely these
aspects of quantwn field theory which become relevant in the study of
quantum fields in curved spacetimes and in other topics in the interface of
gravity and quantum field theory.
The lectures are divided into five parts. Part 2 is a fairly comprehen-
sive and self-contained introduction to the concept and application of path
integrals. No previous acquaintance with the subject is assumed though
familiarity with conventional quantum mechanics (at the level of a first
course) is required. The emphasis is on using path integrals as a technique
for studying non-perturbative aspects of quantum theory. Several impor-
tant concepts in field theory are illustrated in the conceptually simpler
environment of quantum mechanics so as to make the transition in the lat-
ter parts 3 and 4 easier. The discussion of path integrals based on Jacobi
action and its connection with an infinite system of harmonic oscillators
is new and (as far as the author known) has not been emphasised before.
This connection allows one to proceed logically from the path integral for
a relativistic particle to the concept of a quantum field. Part 3 clarifies this
351
352 T.PADMANABHAN
connection and introduces the scalar field to describe a relativistic spinless
particle. Possible generalization to other fields are briefly mentioned.
At this juncture one could take different routes to proceed further. I
have omitted the discussion of conventional topics like canonical quantiza
tion, perturbative study of interactions etc. etc. There are two reasons for
this conscious choice. First of all, many of these concepts are fairly useless
in the context in which gravitational field plays a crucial role. The coo-
ceptual machinery behind this approach - which has been so successful in
quantum electrodynamics and electroweak unification - fails to provide us
with a clear direction to proceed further. Personally, I consider perturbative
renormalizability as an overemphasised trick which needs to be replaced by
sounder and more fundamental concepts if progress has to be achieved in
combining quantum theory with gravity. Since these lectures are primarily
intended for relativists I have not bothered to include historically relevant
topics like perturbation theory. Secondly there exists excellent text books
which discuss perturbation methods in quantum field theory, Feynman rules
etc. and an interested reader will have no difficulty in learning these topics.
Instead of covering these subjects, I have devoted part 4 to a detailed
discussion of the concept of effective action. The basic idea of effective ac-
tion is explained using a quantum mechanical toy model and generalized
to the study of a scalar field with arbitrary self interaction. The renormal-
izability (or otherwise) of a scalar field with self interaction is discussed
in the context of finiteness of effective potential. This also allows the in-
troduction of "running coupling constant" without having to go through
the detailed discussion of Feynman rules etc. I have also included in part 4
an extensive discussion of Schwinger's proper time method applied to the
evaluation of effective Lagrangian and pair creation in external fields. These
are illustrated in the context of creation of scalar particles by an externally
specified electromagnetic field from different points of view.
Part 5 discusses a series of selected examples from the study of quantum
field theory in non-trivial backgrounds. The intimate connection between
the choice of harmonic oscillators to describe the field and the concept of
a field is illustrated by working out the ground state of the scalar field in
(1 + 1) dimension in two different coordinate systems. This is followed by a
discussion of particle creation in several external gravitational fields. Par-
ticle creation in an expanding universe is studied by techniques similar to
those used in part 4; other cases like Schwarzschild metric etc. are discussed
in a unified manner using a semiclassical propagator.
Since the lectures were intended to be of pedagogical nature, I have kept
dtations to original literature to a minimum. The suggestions for further
reading given at the end might help the reader to delve into the original
literature if he so desires.
2. The Path Integral
2.1. INTRODUCTION
353
This part is devoted to a detailed discussion of path integral techniques.
Section 2, 3 and 4 introduce the basic ideas and defines the path integral.
The rigorous definition of path integral by analytic continuation "to Eu-
clidean time is covered in section 5 and various techniques using the path
integral is illustrated in section 7. Section 8 and 9 discuss the system of
harmonic oscillators using path integrals, paving the way for quantum field
theory. The last two sections contain a. discussion of path integrals based
on Jacobi action and its relation with harmonic oscillators.
2.2. AMPLITUDES FROM 'SUM OVER PATHS'
In classical mechanics, the laws governing the motion of a particle in a
potential V(x) can be obtained from the principle of least action. This
principle states that the trajectory followed by a particle in travelling from
(tl,Xl) to (t2' X2) will be the one which makes the 'action'
1
" J." 1 A[x(t)] = dtL(:i;,x) = dt(-m:i;' - V(x))
t) t) 2
(1)
an extremum. This extremum path can- be found as follows: Consider any
two paths x(t) and y(t) connecting the events 'Pl and P2. We can always
write y(t) as x(t) + f(t) where E is an adjustable 'book-keeping' parame-
ter. Since both x(t) and yet) connect the same events it is clear that the
function f(t) vanishes at tl and t2. We can now compute the difference.6.A
between the value of the actions A[x(t)] and A[y(t)] for the two paths. This
difference, will depend on and can be expanded as a Taylor series in E.
The action will be an extremum for a. path x(t) if the difference 6A has no
linear term in f. From (1) it is clear that
AA = A[x(t) + ,/(t)]- A[x(t)]
- -, f dt (m ~ ~ + V'(X) f(t) + 0(,'). (2)
If the linear term has to vanish for arbitrary J(t), then we must have
cf'x '( )
m
dt2
+V x =0 (3)
which determines the extremum path. The solution to this differential equa-
tion connecting the events Pl and P2 gives the classical trajectory of the
354 T.PADMANABHAN
particle. We will denote this classical path by xcU) and the corresponding
value for the action, A(x
c
) by Ac.
The action, in general; assigns a real number to a path x(t) and is called
a <functional' of x(t). The action in (1) is a very special kind of functional
in the sense that it can be expressed as an integral over a iocallagrangian
L(,x). To calculate the action for a particular path x(t), we will first.
evaluate the integrand L(x{t), x(t)) = Ltt) for this path and then integrate
the resulting function over t. The action in (1) is defined only for those
paths x(t) for which this procedure can be carried out in an unambiguous
way; in particular the integrand L(t) should be a single-valued function
of t. This, in turn, implies that the function x(t} should be single-valued.
Paths like the one in Figure 1 are not allowed as valid arguments for the

x(t)

t
action functional.
The classical description of dynamics, outlined above, depends crucially
on the existence of well defined trajectories for motion. To characterize a
path at any instant of time, it is necessary to specify both the position and
velocity of the particle at that instant of time. Since uncertainty principle
forbids such a simultaneous specification of position and momentum the
above description needs to be modified in quantum mechanics ..
A suitable modification can be arrived at by considering the results of
standard two-slit interference experiment with, say, electrons. These ex-
periments suggest that the electrons do not follow a definite trajectory in
travelling from the electron gun to the screen. Instead, we must associate
with each path connecting the electron gun and any particular point all
the screen, a probability amplitude A(path). The net probability ampli-
tude K(2; 1) for the particle to go from the event Pi to the event P2,

ASPECTS OF QUANTUM FIELD THEORY 355
is obtained by adding up the amplitudes for all the pa.ths connecting the
events:
K(2;1) = K(t"x,;tl,Xtl = I: A(path)
(4)
paths
The addition of the amplitudes allows for the quantum mechanical interfer-
ence between the paths. (In the case of the two.slit experiment, for example,
we will add the amplitudes for paths through hole-l and hole-2; this will
lead to the correct interference pattern.). The probability for any process, of
course, is obtained by taking the square of the modulus of the amplitude.
The quantity K(t2,x2;tttXd contains the full dynamical information
about the quantum mechanical system. Given K(t2,x2;tl,xd and the ini-
tial amplitude ~ ( t b X l ) for the particle to be found at Xl, we can compute
the wave function ",,(t,X) at any later time by the usual rules for combining
the amplitudes:
. ",(t, x) = / dx1K(t,x;t"xIl",(t.,xIl
(5)
Therefore the specification of (i) A and (ii) the rule for evaluating the
sum, in (4), will provide a complete quantum mechanical description of the
system.
Since K(t2, X2; tl, Xl) contains the complete dynamical information of a
quantum mechanical system, it is obvious that we will not be able to derive
the rules for its computation from fundamental considerations. We have to
content ourselves by motivating a possible choice for .A and the rule for
computing the sum in (4).
Since we do not expect anyone path to be more probable than the other
it seems reasonable to demand that the probability for any particular path
IA(path)12 must be independent of the path. In other words, A should have
the form
A = expiA(path) (6)
where A is real and we have ignored an overall normalization constant. In
the classical limit, however, we would like the classical path Xc - obtained
by extremising the action - to contribute dominantly to the sum in (4).
This can be achieved most easily by taking A(path) = [A(path)/IiJ. Then
the Kernel becomes
.A[x(t)J
K(t2,x2i
t
ll
X
l) = L exp' h .
all x(t)
(7)
In the limit of h going to zero, the phase of A oscillates rapidly and the
contributions from different paths are mostly cancelled out; the only ones
that survive are those for which A is an extremum. viz. the classical paths.
35G T.PADMA:\ABHAN
This choice in (7) thus provides a natural. explanation for the validity of
principle of least action in the classical theory.
Since the action functional A is well defined only for single-valued func-
tiOIlS x(t), the sum in (7) may also be restricted to such functions. This im-
poses an important constraint on K(t2,x2;t},X}). Consider three instants
of time (tt. t2, ts) such that tl < t2 < ts. Since the paths are single-valued,
the path which connects the event PI to the event P3 must cross the
t = t2 line at some single co-ordinate X2. Hence the usual law of combining
probability amplitudes allows us to write
(8)
which is a constraint on the normalization of K.
The behaviour of the amplitude K(t2,x2;tl,Xl) under time reversal,
tl f+ t2, allows us to put another constraint on the Kernel. Assuming that
the prescription for summing over paths does not single out a direction
of time, and noticing that the action functional changes sign under time
reversal, we get
(9)
Thus time reversal is equivalent to complex conjugation of the amplitude,
which is a general feature of quantum theory. It is easy to show that equa-
tions (8) and (9) together guarantee that the normaliz.ation of wave func-
tions are preserved under time evolution.
2.3. SUM OVER PATHS FOR QUADRATIC ACTIONS
We are nowin a position to work out the form of K(t2,x2;tIoXl) for a
particularly important class of systems. These are the systems described
by an action which is at most quadratic in x and Xj that is, for actions in
the form
A[x(t)] = / (B(t)i;' + C(t)x' + M(t)x)dt (10)
A more general form for a quadratic action
A[x(OI = /[a(t)x' + b(t)x + c(t)xx + d(t)x' + e(t)x + l(t)Jdt (11)
can be reduced to the form in (10) by suitable partial integrations. To
evaluate (7) for such an action we proceed as follows: We write an arbitrary
path x(t) as xc(t) +q(t) where q(t) denotes the deviation from the classical
path. Since all the paths are between two specified end it follows
that q(td = q(t2) = O. We can now write
. .
K(t2,x2;tl,Xl)= L L (12)
a.1I x(t) all q(t)
But since A has only terms which are at most quadratic, A[xc(t) +q(t)] has
the form
A[x,(t) + q(t)] = A[x,(t)] + L[x" q] + Q[q;tj, t2] (13)
in which L is linear in q (and xc) while Q is quadratic in q. (The Q cannot
depend on Xc but it can depend on (tll t2) because these appear as the
limits of t-integration.) But since Xc is an extremum (classical) path, the
linear term must identically vanish; L = O. Therefore
K(t2,x2;t
b
Xl) = L Mexp +q(t)]

-i .-/ i
- exp iiA, L exp iiQ(tj, t2; q(t)) . (14)
all<j,Wl
The expression inside bracket is some function of tt and t2 alone. Calling
this quantity N(tl
l
t2), we can write the form of the Kernel to be
We see that the x-dependence of the Kernel is only through the classical
action Aco We emphasize that the above result is exact for systems with
quadratic actions.
So far we have not specified the exact prescription for calculating the
sum over paths. To determine the value of N we need to specify the rule
for summing over the paths. In the case of quadratic actions, a natural
definition exists allowing us to express N as a determinant of a differential
operator. This is achieved as follows: For actions in the form of (10) the
quantity Q can be expressed as
Q - J." dt[Bq2 + Cl]
"
- [Bqq]:: - t - C)q] dt
-J." [q(<!..B<!.. - C)q] dt = -J." dt q(t)Dq.
tl dt dt tl
(16)
358
In arriving at the last line we have used the fact that q vanishes at the
endpoints and defined an operator
d d
D=(-B--C)
dt dt
(17)
[One may worry that the vanishing of q does not ensure the vanishing
of qq for certain pathsi it turns out that these paths (orm a set of zero
measure in the sum over paths.] In all cases of physical interest tb}s operator
will possess a complete set of orthonormal eigenfunctions {fn(t)} with the
eigenvalues {.\.n}, satisfying the same boundary conditions as q. That is,
there will exist a complete set of functions satisfying the conditions
(18)
We can now expand any q(t) in terms of the functions Un}
(19)
n
obtaining a set of numbers {en}. This set of numbers completely charac-
terize the function q(t). The quantity Q[q(t)] which was a functional of the
path q(t) can now be expressed as a/undion of the numbers {en}'s as above
equation (21):
Q[q(t)] = - j qDqdt = - jCE-Cnfn)b(LCmfm)dt
n . m
- - L CnCm j fnfmAm
dt
= - L AnCnCmOnm = - L A n ~
nm - nm n
Q[c",,>, .. ] (20)
Therefore, it seems reasonable to define the sum over all Junctions q(t) as
equivalent to an integration dver the set of {en} 's. That is, we would like
to make the replacement
(21)
where Mn is a measure which determines the weightage to be given to the
n-th eigenmode. In general, we could have made Mn's explicit functions
of en's. In making the above choice, we have decided that the weightage
given to each path depends only on how 'wiggly' each path is and not on the
explicit shape of the path. We can now write
N =
'" i 1+
00
den i
exp /iQ[q(t)] = II M exp -- L
allq(t) -00 n n n n
[
1
+
00
de. i ,]
- I] -00 Mn exp -1l Anen
(22)
We now face a mathematical difficulty. The integrals over en in the above
expression does not exist because the integrand oscillates for large en if all
the An'S are real. It is necessary to give an extra prescription which will
allow us to interpret these integrals.
We can give meaning to t,his integral if we assume each of the An'S have
an infinitesimal negative imaginary part. That is we replace An by (>\n - if)
and set t: to zero at the end of the calculations. In this case,
1
+00 den i . 1+
00
den i . 2
M
M eXP-Ii{An-tE)c
n
-00 n (-+0 -00 n
-_. __ . ----'--- -_. - --- ----- -----
_ lim -ur m ,
1
(
. Ii ) 1 1 (. Ii)!
(-+0 Mn P .. n - if) Mn An
(23)
- With this definition of the integrals, we get
,
II
.,
N = An 2
n Mn
(24)
The normalization constant Mn was not specified so far. We have to
choose it in such a manner that the resulting Kernel satisfy our conditions
(8) and (9). To do this correctly, we will work out the expression for N in
the special case of a free particle using (8) and (9). Comparing this result
with (24), we can determine Mn.
Since the extremum path for a free particle is a straight line, the classical
action for the free particle will be
Ac = (X2 - xt}2
2 t2 - tt
(25)

glvmg
360 T.PADoIANADHAN
We have set N(t2l tIl = N(t2 - til = N(T) using the time translation
invariance of the system. We shallllow determine N(T) using (8) and (9).
Substituting (26) into (8) we get
Working out the integral, we can write this condition in the form
N(t -t ) = N(t -t )N(t -t ) (21fih)I/2 (t3 - t2)1/2(t2 - tl)I/2. (28)
3 1 3 2 21m (t3 _ tl)1/2
If we now put (t2 - td = a, (t3 - t2) = (3 and
(2
' 1/2
M(z)"= N(z) 1f'::'Z)
the above relation reduces to the form
M(a + (3) = M(a)M(3).
This equation has the general solution
M(z) = exp(kz)
where k is some constant. Therefore N (T) must be of the form
(29)
(30)
(31)
(32)
We now use the condition (9), that K(X2t2;xltl) = K*(X2tljXlt2)' This
will give
exp(k'T) = exp( -kT)
(33)
implying that k is purely imaginary; k = {i/Ti)J.' with reallJ. Therefore, the
Kernel can be written as
(34)
We see that the quantity I-' is only an additive constant to the original
lagrangian. The expression in (34) is what we would ha\'e obtained if we
,
had taken the free particle lagrangian to be (imx2+J.L) rather than
Therefore, we can set J.l = 0 without any loss of generality. The free-particle-
Kernel can be then taken to be
with the normalization constant
[
m ]1/2
N(T) = 2" iliT .
m(X2 -xd
2
2 (t2-ttl

(35)
(36)
Let us now work out the expression for N(T) for a free particle from (24).
-
The differential operator D for a particle of mass m
has the eigenvalues
- mJ'
D ."2 dt2
(37)
(38)
for a complete set of eigenfunctions which vanish at t = 0, T. Substituting
this expression into (24), we get
(39)
Comparing (36) and (39) we see that Mn's should be chosen to satisfy the
condition:
(40)
We shall take this relation to be the 'definition' of Mn in the follOWing
sense: We will. hereafter, write the normalization for an arbitrary quadratic
lagrangian as
N
II (i"Ii)! -L II (i.-Ii)1 en') I -!
=
M An - M >. AnO
n n n n n
-
II en')!
NCree-particle(T} n An
1
m,,2
l
n'
1
C,,';.T)'
,
(41) -
- -
2T'

362 T.PADMANABHAN

This can be computed if An'S are determined from D. All refer-
ence to the weightage Mn has now disappeared.
It can be shown, from the study of the Sturm-Liouville problem for the
operator b that, An ex: n
2
for large n for aU systems. (For a free particle,
An ex: n
2
for all n.) Though we never bothered to display Mn explicitly, it
should be clear from (24) that Mn should scale as Mn IX n-
1
for large n
to ensure convergence of the product. Since the weightage for each path
is M;l 0( n, it follows that the paths which wiggle more are given higher
weightage in the Sum over paths.
As an example of the use of the above formula, let us evaluate the N

for a harmonic oscillator. The operator D will be
which has the eigenvalues
. m
D=-
2
m
A ...:--
2
for eigenfunctions which vanish at t = 0, T. From (41), we get

:
C';;'T) ' 1]
mw'
,
n'
-
2T2 A.
N(T) -
-
I
w
2
T2
-i
(2:nT)' 1]
1-
n
2
:Jr2
Using the relation
(
wT )! (W'T')-!
sinwT = I} 1 - n2:Jr2
we find that
I 1
N(T) = ( m )' ( wT )'
21fi1i.T sinwT
(42)
(43)
(44)
(45)
(46)
The expression for N(T) is conventionally written in a somewhat
ferent form. Noticing that
_1

II =
, 1
= (det D)-' (47)

where the determinant of an operator is defined as the product over its
eigenvalues we can write
N = Dq exp -- qDqdt = N
o
( )-,
f
[
if"] detD,
h . detDo
(48)
where No and Do refer to the free particle expressions and the curly-D
in the integration symbol is a reminder of the particular choice made for
defining the summation over paths ("path integral"). Even though both
determinants may be individually divergent, their ratio will tum out to be
finite in several physically interesting situations. Textbooks often quote this
result as
N = f Dq exp f qDqdt] = (det D)-l
(49)
where it is implicitly assumed that the divergences in the'right hand side are
handled by dividing out the corresponding expression for the free-particle
case. We will adopt this notation when no confusion will arise.
2.4. PATH INTEGRAL WITH EXTERNAL SOURCE
There is an extension of the above result which is repeatedly used in field
theory. Let us consider the pa.th integral for the action
A = - f (xDx + J(t)x)dt
(50)
where J(t) is some externally specified function. This action, of course, is
quadratic and We already know how to evaluate the pa.th integral. But,
in doing that, we will have to expand x(t) as xc{t} + q(t) where Xc is the
classical solution obtained from A. Of teo, it is convenient to treat J(t) as a
separate entity and expand x(t) as xc::(t) + q(t) where :z;c(t) is the classical
solution to the original action - i. e., the ODe with J = O. It is easy to see
that N will now become, instead of (49), the expression
N(t,. tl;J(t)) = ! Dq exp (qDq + J(t)q)dt] (51)
We can evaluate this expression by following exactly the same procedure as
we expand q(t) in terms of the eigenfunctions {In} of the operator
D and evaluate the sum over paths as a product over a set of integrals. The
only difference is that, for the In's in (23),we now get
lim j+= exp [(An - + JnCn]
t"-+O _= n h
I
. 1 ( i1fh ) ! i J;
- 1m exp-.
Mn (An - i,) 4h (An - ,,)
(52)
364
where
Therefore
N=
T.PADMANADHAN
1
"
J. = dtJ(t)I.(t)
"
,
. J2
II
(' . )-'
"'n -IE 2
M.
'", .
ex:p 4h ('\n - if)
(53)
(54)
with the understanding that should be set equal to zero at the end of the
calculations.
In most applications we will require the value of this expression in the
limit of tl -00, t2 -t +oo.-The result CB.D then be recast in a more useful
form as
N = (det D)-\ exp ;" f dt f dt' J(t)G(t, t')J(t') (55)
where the function G(t, f). defined as
G(t, t') = L 1.(t)/.(!')
- n An- U:
(56)
is the Green function for the operator {D - if). It satisfies the differential
equation
"
(D, - ;<)G(t,t')
_ (D" .:...) '" In(t)/n(t')
tlL.J\ .
n "'n-SE
- L In (t)ln(t') = O(t - 1')
(57)
n
OUf final result can be stated as
f Vq exp [- f (qDq + J(t)q)dtj
- (det dt f dt'J(,)G(t,,')J(t') (58)
We will use this result extensively in the later sections.
The if in the definition of the Green function is quite crucial. To see
this, let us evaluate G(t) for a harmonic oscillator. We need to solve the
equation
(D - i<)G(,) = +w' -;< G(t) = ott) (59)
which can be easily done in the Fourier space. Writing
G(t) =
-00 27r

(60)
ASPECTS OF FIELD THEORY
we find that G(v) satisfies the equation
Therefore,
- i, - v'IG(v) = 1.
2
2 1+
00
dv e-
iv
!
G( t) = -- --"-,c,------,,,
m -00 211" 1/
2
- (w
2
- if)
365
(61 )
(62)
The importance of if can be seen at this stage. If it were not present, the
integrand will have poles on the real axis and hence the integral will not
exist. To give meaning to such an integral one has to specify the path of
integration in the complex plane and the results will depend crucially on the
path chosen. The existence of if give an unambiguous value to the integral
by shifting the poles away frQm the real axis. The poles are now located at
/J = w(l- if), infinitesimally away from the real axis. For t > 0, we close
the contour on the lower half plane and obtain the contribution from the
pole at (w - i)j for t < 0 we close the contour on the upper half plane and
the pole at (-w + if) contributes. We thus obtain
G(t) = i + 9( -t)e+""'1 = i e-""l'l. (63)
mw mw
The final result for the harmonic oscillator can be stated as,
N - exp 4
i
" / dtdt' J(t)G(t, t')J(t')
_ (detD)-l/'ex _ i 1 1+
00
IJ(v)I'. (64)
p h2m -00 21r1l2 _w
2
+u:
where we have used the fact that J*(v) = J( -v) for real J(t).
Even though this Green function was obtained as a solution to the har-
monic oscillator operator, it is quite different from the usual Green functions
defined in the classical theory of the oscillator. In classical mechanics, the
equation for a forced harmonic oscillator
m ('" ,
"2 dt2 +w
q(t) = J(t) (65)
is solved. by expressing q(t) as
q(t) = / GR(t - t')J(t')dt'
(66)
where GR(t) is the classical Green function. This object has the following
Fourier representation:
(67)
366
T.PADMANABHAN
Both the poles of the integrand are now in the lower half plane. Therefore,
the integral is zero for t < 0 and is a real quantity for t > 0:
2 .
GR(t) = B(t)smwt.
mw
(68)
This 'retarded Green function' propagates the effect oCthe source J(t) only
'forward in time'. In other words, q at some time t = to is only influenced
by the behaviour of J(t) for t < to. On the other hand, our G(t) propagates
the effects of l(t) both forward and backward in time. Thus these functions
are mathematically quite differentj the position of ie factor is quite crucial.
2.5. THE EUCLIDEAN TIME
We have made one important assumption in arriving at the expressions
quoted above which needs to be highlighted and discussed. We needed to
interpret the integrals which appear in the definition of sum over paths in a
specific manner, by attributing a small imaginary part to all the eigenvalues.
This is equivalent to considering the original operator b as the limit
D '" lim(D - if)
HO
(69)
This procedure, called the 'ie-prescription' is just one of the many ways
of making sense out of an illdefined integral. It is possible to devise other

modifications of the operator D - and corresponding limiting procedures
- to give meaning to the integral. There is no assurance, a. priori, that all
these procedures will lead to the same final result.
One such important alternative procedure, which is extensively used, is
based on analytically continuing the expressions into complex-t plane. Let
us introduce a variable., = it (so that t = -iT") in the action. Under this
substitution, the quantity
exp ~ Q = exp ~ /.:' dt(B(t)q2 + C(t)q2)
(70)
becomes
(
i) ( QE) 1 [T>
exp nQ =exp - h =exP-n}Tl d.,
(71)
where BE(") =: B(t = -i.,) etc. We will assume that the original action is
such that (i) BE(T) and CE(T) are real (ii) BdT) > 0 and (iii) CE(T) < o.
Then the argument of the exponent in (71) is negative definite for all real
ASPECTS OF QUA "TUM FIELD THEORY 367
paths q(T). [This set of paths, of course, is different from the set of paths
obt.ained by substituting t = -iT in the original set of paths; in general:
if q(t) is a real function, qE{T) = q(t = -iT) will not be rea1.] Let us now
consider the sum
(72)
which, by an analysis similar to the one performed before, can be reduced
to the form

(73)
where An are the eigenvalues of the operator DE appearing in the eigenvalue
equation
. [d d 1
DEI. = - ;r;:(BE;r;:) + CE I. = ~ . I .
Expressing An as
(74)
and noticing that BE > 0 and OE < 0, we find that An is positive definite.
Therefore the .integrals in (73) are well-defined, leading to
("Ii)'/2 ( 1 )'/2
NEh,Ttl = II M f' = (detDE)-'/2.

(75)
In arriving at the last expression we have made the implicit assumption
that the determinants are regularized in the usual manner by dividing out
the free particle contribution. We shall now define the original expression
N(t2. tl) as the analytic continuation of NE(T2. 1'1):
(76)
This procedure may be summarized as follows: (i) From the original
expression Q[q(t)) obtain QE[q(T)) by analytically continuing from t to T.
(ii) Check that BE, CE are real and BE > 0 and CE < O. (iii) Evaluate
the sum over paths for QEI by summing over all real q(1'}. (iv) Analyti-
cally continue back to tj this is defined to be the value of the original sum
368
T.PADMANABHAN
over paths. It should be emphasized that this method works only for those
actions for whicIl the condition (ii) above is satisfied. The quantity T is
called the 'Euclidean time' and other variables like AE, G E etc. are called
'Euclidean action', 'Euclidean Green function' etc.
The above idea can be easily extended to path integrals with an external
J(t). Instead of (58), we will now get
J Vq exp
- (det DE)-lexp dt J dt'Jdt)GE(t,t')JE(t') (77)
where G E is the Green function for the operator DE:
DEGE = 6(7)
(78)
This Green function will be well defined even without any additional i
prescriptions.
As an example, let us find G S(T) for a harmonic oscillator. This function
satisfies the equation
.
which can again be solved using the Fourier space. Writing
I
+CQ dv .
G(T) = -2
-00
we find that 0(11) satisfies the equation
Therefore,
T[w' + v'JG(v) = 1.
21+
00
dv e-
iVT
G( r) =;;; -00 2; -v'"+-:-w"
(79)
(80)
(81)
(82)
This integral is perfectly well-defined and can be easily worked out to give
GE(r) = 1 e-WIT!
mw
(83)
We see that the analytic continuation of (77), with the above expression
for G E correctly reproduces (58). Thus for this simple system the Green
function defined by the if prescription and the one defined by analytic
continuation from imaginary time are identical.
ASPECTS OF QUANTe,,1 FIELD THEORY 369
2.6. PATH INTEGRALS FROM Tl,iE SLICING
The procedure described above to define the sum over paths works only
for quadratic actions for which an operator iJ can be defined. We will now
provide a more general definition which will work for any action in the form
(1). Let us divide the time interval T = tf - ti into J equal divisions, each
of size At == J-
1
(t,-ti)' Let tK = with K = O,l,2, ... J, so
that to = tt and tJ = tf. By using (8) repeatedly, we can write
M=J-l
IT .K(tM+l,XM+l;tM,XM)
M=O
(84)
We now consider the limit of the above expression with J oo,.6.t -+ 0
with Jf).t = (tf - til remaining finite. Each of the Kernels appearing in
(84) will then represent the amplitude of propagation between two events
which are infinitesimally separated in time:
- (85)
with Ax = XM+I - XM. We notice that in the limit of l:!:..t---1- 0 the kinetic
energy term will dominate over the potential energy if 'wiggly' paths con-
tribute most to the sum. Therefore, in this limit, we can replace the Kernel
by a free particle Kernel with the potential energy providing an extra phase
factor. That is,
. 1.
lim K(tM + l:!:..t, XM+l, tM,XM) = lim (2 2 exp (86)
.t.t-+o _ .t.t-+O t "
where
.40=
(87)
Substituting (87) into (84) we get the required expression for the Kernel:
-
-
exp
(88)
370
T.PADMANABHAN
This expression provides a definition for the SlIm over paths for all actions
in the form (1). Though formally correct, it is not of much use in the actual
computation of Kernels. The integrals involved in (88), can be evaluated
in a closed, simple form only for quadratic actions for which - it can be
shown that - (88), and (15) give the same result. For non-quadratic actions,
the conventional approach to quantum theory - based on operators and
eigenstates - turns out to be much more convenient. In fact, the physical
meaning of several expressions derived earlier becomes clearer when the
Kernel is connected up with the more conventional description. We shall
now examine these connections.
2.7. KERNELS AND GROUNDSTATE EXPECTATION VALUES
In the conventional approch to quantum mechanics using the Heisenberg
picture the description will be in terms of the position and momentum
operators x and p. Let Ix,t > be the eigenstate of the operator x{t) with
eigenvalue x. The Kernel- which represents the probability amplitude for a
particle to propagate from (tl,Xl) to (t2,X2) - can be expressed, in a more
conventional notation, as the matrix element: .

K(t2, X2; tIl Xl) =< X2, t21tt, Xl >=< x2, 01 exp --H(t2 - tl)lO, Xl > .
. h
(89)
- .
where H is the Hamiltonian for the system. This relation allows one to rep-
resent the Kernel in terms of the energy eigenstates of the system, provided
the hamiltonian is independent of time. We have

- L: exp - En
T
n
(90)
where ,pn(x) =< xlEn > is the n-th energy eigenfunction of the system
under consideration.
In physical applications, we often require the limiting form
W(T;X2,xll - lim lim K(X2t2;xltd
h -+ +00 h -+-0;)
lim K(X2T;X,O)

(91)
This cannot be directly ascertained from (90) because the exponent os-
cillates. However, we can give meaning to this limit if we first transform
(90), to the imaginary time T} = it, and TZ = itz and consider the limit
(7"2 - rd 00. We find that
where the zero-subscript denotes the lowest energy state. (Note that 1/Jo =
1/10). From (92), we see that only the ground state contributes in this infinite
time limit. We may now define the limit in (91), as the analytic continuation
of (92), getting
,
This expression allows one to determine the ground state energy of the
system from the Kernel in a simple manner. We see that
WE(T;O,O) '" (constant)
(94) ..

glvmg
Eo = InWE(T; 0, 0))
(95) A-
The Kernel can also be used to study the effect of external perturbations
on the system. Let us suppose tha.t the system was in the ground state in
the asymptotic past (h -00). At some time t = -T we switch on an
external time dependent disturbance >..(t).. affecting the system. Finally at
t = +T we switch off the perturbation.@ecause of the time-dependence,
we no longer have stationary energy eigenstates for the In fact, the
system is likely to have absorbed energy from the perturbation and would
have ended up at some excited state at t2 = +OOj the probability for it
to be found in the ground state at t2 = +00 will be less than one. This
probability can be computed from the Kernel. Consider the amplitude
-
-
lim lim [j+oo dxdx' < t21 x2[T, x >
h-too tt-t-o;; -00
x
< T,xl- T: x' > < -T,X'[tlXl >]
(96)
Since), = 0 during t2 > t > T and -T > t > tlo matrix elements in these
intervals can be expressed in terms of the energy eigenstates of the original
372

system:
lim < t" x,iT,x >'" ,po(x,).J>o(x) exp - T)
Ii
lim < -T,x'it"xl ,pO(X').J>o(xtlexp - ttl (97)
h-t-oo H
Therefore (setting 1i = 1 for simplicity)
+00
P [.J>o(x,).J>o(xtle-iEo(t,-t,lj! dxdx'(,po(x)e+
iEoT
)
-00
x < T,xlx',-T > (,po (x')e
iEoT
)
+00
-
lim K(t2,x2itlxli A = 0) ! dxdx'{,po(x, T)]*
(t2-h)-+oo
-00
x <T,xix',-T> [,po(x:-T)]
(98)
where 1fo(x, T) represents the ground state wave function at time T etc.
The quantity
1
+00 ,
W = -00 dxdx'[,po(x, T)]' < x,Tix', -T > [.J>o(x, -T)]
(99)
represents the amplitude for the system to remain in the ground state in the
asymptotic future if it started out in the ground state in the asymptotic
past [usually called the "vacuum to vacuum" amplitude]. From (98), we
find that this amplitude is given by the limit:
W Ii
[
' K(t" x,; t,; >-(t))
= m lID
h-+oo h-+-oo K(t2,X2itl,Xl;O)
(100)
This result can be further simplified by noticing that the X2 and Xl depen-
dences cancel out in the ratio in (100) so that we can set X2 = Xl = 0,
getting
W
[
' [' K(t"O;t"O;>-(t))
= un lID
tl-tOO h-t-oo K(t2,Oitt,OjO) .
(101)
Thus the vacuum-vacuum amplitude can be found from the Kernel by a
simple limiting procedure.
The importance of W emerges from the fact that it contains useful
information about the excitation to other levels as well. If Wn denote the
amplitUde for the system to be found in the n-th excited state at t2 = 0:::,
then IW
n
l
2
can be usually approximated as a Poisson distribution:
n
IW
I
' - -"
n
n!
(102)
where 11- = -In IW
o
I
2
Thus, Wo contains full information about the system
if the transition probabilities to different levels are independent of each
other, validating (102). In this case, Jl. denotes the mean level to which the
system is excited: Jl. =< n >. The mean energy absorbed by the system
from the driving source is (EnJl.).
In the above calculation we have not specified the nature of the coupling
between A(t) and the original system. In principle, this could be arbitrary;
in practice, however, we will be able to calculate K[>..(t)]-needed to compute
W- only if the coupling is the coefficients of a quadratic action. The
simplest of such situations correspond to the addition of a. linear coupling
to an external source ),(t) = J(t),
A"'d = / J(t)q(t)dt (103)
to the original quadratic action. We have already discussed this situation
in section (2.3). In this case W is easily computed. We get:,
W = exp :n dt t:: dt'J(t) G(t, t')J(t').
(104)
N; an example of the evaluation of W, let us consider the case of har
monic oscillator. Writing
R = dtdt' J(t) G(t, t')J(t') = -! "J(v1
"
. (105)
-00 m -00 1fV -w +I
we are interested in the probability IWl
2
for which only the imaginary part
of R contributes. Since
1 .1
:c77:. = p- -
X + It::: x
(106)

where P denotes the principle value of the integral, we get
1 1
ImR = - dvIJ(v)I'o(v' - w') = IJ(+w)l'
m -00 tnW
(107)
[We have used the fact that IJ(w)I' = IJ( -w)I' for real J(t).J Therefore,
the probability for a harmonic oscillator to remain in the ground state even
after the action of a perturbation J(t) is
IWI' = I exp = exp
41i 21i
_ exp 1 IIJ(w)I' = exp (_ .f) (108)
(liw)' 2 m
374
T.PAmIA"ABHAN
This result has a very simple interpretation. Note that only the power
!J(wW at the resonance frequency v = w causes the transitions. According
to the discussion following (102), (Eo/hw) denotes the lllean level < n >
to which the system is excited and Eo =< n > 1iw is the mean energy
absorbed by the system. In fact, E is the total amount of energy absorbed
by a classical oscillator when driven by J(t}. This can be most easily seen
by writing the oscillator equation
m(x + w' x) = -J(t) (109)
as
'. z + iwz = -J(t)
where
z(t) =m(x-iwx)
Solving (110) we get
z(t) = _e
iwt
1100 J(t')e+iwt' dt
.
from which it is easy to see that
1 (., ")1
(Energy absorbed) - e. = 2m x + W x t=oo
(110)
(111)
(112)
- ~ {Izl'} = 21 IJ(w)I' (113)
2m t=co m
In fact, for the harmonic oscillator, the probability for transition to the
n-th energy level given by (102), turns out to be exact.
Another case of interest in which the external perturbation can be han-
dled is the "adiabatic caseD. This is the situation in which the perturbatio?
>.(t) varies slowly compared to the intrinsic time scales of the system. In
that case, the ground state evolves in time adiabatically, as
. ,
.po(x, t) = "",(x, 0) exp - ~ 1, Eo(),(t))dt
(114)
where Eo(>.) is the ground state energy of the hamiltonian calculated by
treating >. as some given, time-independent parameter. In this case, it is
easy to see that
. +00
W = exp -i; 100 Eo(),(t))dt
(115)
Lastly, we will establish an important connection between matrix ele-
ments of the form < X2, t2Ix(t)x(t')lXI' tl > and the KerneL Consider the
quantity
K[x"t,;x"t,;J(t)) = / Vx(t)exp [iA[X(t)) + i / J(t)X(t)dt] (116)
,
. ,
where A is some arbitrary action functional. By taking the functional deriva-
tive of this expression with respect to J(t} two times, \ve can write,
Setting J = 0, we get
(118)
The expression on the left hand can be expressed using the Kernel:
f < ")C!,t."
,/ - r: dx 1:UK(i2,t2ix,t)xK(x,tj:dJ)xl K(x',t'jXltd
- 1...:
00
dx 1:
00
dx' < X2-,t2Ix,t > x < > x' <x',t/IXltl >
= < x" t,lx(t)x(t')lx" t, > (119)
Therefore, we get the relation.
< x" t,lx(t)x(t'Jlx"t, >= K(x" t,;x"t,; J)L:o'
(120)
This relation shows that the matrix elements of a string of co-ordinate
operators at different times can be evaluated as a functional derivative of
the Kernel. It is clear that this method can be generalized to give
In this sense, K[J] may be considered the "generating function" for the
matrix elements < X2, t2Ix{t)x{t')x(t") .. . jxlh > .
We will usually use this result in the limit t2 --+ 00, tl --+ -00 with
x, = x, = 0. Writing K[J(t)] = K(O, 00;0, -00; J(t)) we have
, [Ii " Ii " 1
< 0, oolx(t)x(t )10, -00 >= T U(t)'T U(t') .K[J(t)] J=O
(122)
, -
376 T.PADMANABHAN
In this limit, (119), also simplifies because only ground states contribute
in < X2, t2lx, t > and < x', t'lxl, t1 >. We get, by analysis similar to that
which led to (98),
< 0, oolx(t)x(t')IO, -00 >-< 0,0010, -00 >
[:00 dx 1: dx'1f;o(x, t)K(x, tj x', t')x'1/Jo(X', t'} x:..
where .po (x, t) = ""(x) exp( -iEot) etc. Therefore,
< 0, oolx(t)x(t'JlO, -00 >
, I < 0,0010, -':"":'00 >
(123)
= 1: dx 1: dx'[.po(x, t)x]' K(x, I; x', t')[x'''''(x', t'll (124)
which may be interpreted as the amplitude for the system to be in a state
described by the wavefunction
f(x, t) = x.po(x) exp(-iEot) (125)
from time t to tf.
There is one small complication regarding the above formulas which
needs to be taken care of. The evaluation given above assumes that t >
t!. If t < t' the same results can be obtained if we use < x(t')x(t) >
rather than < x(t)x(f') >. It is usual, therefore, to write these results
by replacing < x2t2Ix(t)x(f)[Xltl > by a "time-ordered" product
< x" t,IT(x(t)x(t'lllxl, t, > where we define:
T(x(t)x(t'll = {x(t)x(t') for t ~ t'
x(t')x(t) for t' > t.
(126)
Everything derived so far is valid for arbitrary actions. These results,
however, take particularly simple form for quadratic actions, for which
. 1+
00
K[J] = K[O] exp iii -00 J(t)G(t, t')J(t')dtdt'.
(127)
Using (122), we get
< 0, ooIT(x(t)x(I'llIO, -00 > = .!i,G(t, t') '" g(t, t') (128)
< 0,0010, -00 > 2,
In other words, the Green function can be expressed as the matrix element
of time-ordered product of operators. {The factor (1/2) was due to our
definition of iJ in a somewhat nonstandard manner].
ASPECTS OF QUANTU,I FIELD THEORY 377
For the harmonic oscillator we have already computed G(t, t') earlier.
The expression in the right band side of (124) has a simple interpretation
in this case. We note that the first excited state (n = 1) for a harmonic
oscillator has the wave function
(
2mw) 1/2 .
'p, (x, t) = Ii x"'o(x, t ) e - ~ t (129)
so that f(x, t) in (125) can be written as
(
h )'/2 .
f(x,t)= 2mw eiwt"'l(x)
(130)
Since the Kernel propagates tPl(X, t) unchanged, (124), implies
!itt, t') = ( ~ m ) exp -iw(t - t')
(131)
which, of course, agrees with the expression for (1i.G /2i) if t > t!. Thus, the
Green function contains the information about the first excited state of the
system.
2.8. HARMONIC OSCILLATORS
The systems which we have considered so far have Hamiltonians of the form
(132)
in which p and x appear quite asymmetrically. If we now impose an addi-
tional constraint on the Hamiltonian in (132L that it should be symmetric
under the interchange of position and momentum coordinates, we are forced
to choose V(x) ex: x
2
The constant of proportionality must be positive def-
inite to ensure lower bound for the Hamiltonian. With suitable scaling, we
can write such a Hamiltonian as
(133)
where we have defined the dimensionless variables
p V""" .. 1
P = .jmru) Q = h X; IQ, P] = ft1x,p] = i
(134)
378 T.PADMANABHAN
I ~ l quantum theory, we are interested in the eigenvalues and eigenvectors of
H. The eigenvalues can be easily found. We rewrite the Hamiltonian as
where
1 (Q. 'p).f 1 (Q. 'p.). .f.
a=.../2 +, i a =../2 -s in=a a.
".
(136)
From [Q,F] =;, it follows that
(137)
Let In > be the eigenstates of n with eigenvalue n. It is then easy to
see that atln > and aln >-are also eigenvectors of n with eigenvalues
(n + 1) and (n - 1) respectively. This fact, added to the condition that
< nlnln >=<-nlataJn >= n < nln 0 implies that n = 0,1,2.
[Alternatively, one may argue as follows: The set of complex analytic
functions form a Hilbert space with the norm
f
dzaz - -
< gil >= 2,,; e-"g(z)f(z).
(138)
The commutator {137} can be implemented in this space with the realiza-
tion
(J
a --t ;rz;
(139)
We again see that the operator
(140)
has integral eigenvalues].
From (135), it follows that the hamiltonian has equally spaced energy
eigenvalues
I
En = nw(n + '2); n = 0,1,2
(141)
The eigenstates of the hamiltonian In > are labelled by an integer. We
normalize them in such a way that
< nln >= 1; .In >= ..mIn -1 >; .fln >= -In + lin + I > . (142)
The operators at, a and n are called creatioll, annihilation and number
operators respectively.
The dynamics of the system described by (132) can be studied in either
in Schrodinger picture or in Heisenberg picture. In Schrodinger picture, a
and at have the representations
(143)
The n = 0, ground state o(QL satisfies the condition
(144) "V, ~
with < tPoltPo >= 1. Integrating (144) and normalising we find that 1 0 0 l ~ A' eo,
(145)
which in more familar variables x = (1i,jf"f'IW)1/2Q. becomes:
(
f1lW)I/' (f1lW)
tPo(x) = 7fli exp - 21i x'
(146)
All excited states can be found from the ground state by operating with at
and normalising. Since the ground state has energy ~ 1 i w , the time evolution
of tPo is described by
(
mw) II' (iliw mw)
tPo(t,x) = "Ii exp - 2 t - 21i x'
(147)
In the Heisenberg picture, we have to solve the equations of motion for
the operator d(t)j since
iM = [a,H] = liw[.,ata] = liwa (148)
'!Ie find
(t) (0) -;,." -' -imt
a = a e = ae
(149)
and similarly
(150)
380 T.PADMANADHAN
where we have condensed the notation by writing a(O) = a etc. From these
relations we find that

F(t) - - o(t)) = _ [ae-
iw
' _ ate
i
.']
O(t) _ 1 (ot(t) H(t)) = J... [ae-"" +ate
iw
'] (151)
v'2 v'2
Equivalently.
(152)
and pet) = mx(t). These relations completely solve the problem. Note that

Q and x are Hermitian.
As an example of a calculation with these operators, let us compute
the quantity < Eolx(t')x(t)IEo >, where lEo> is the ground state. Using
(152) and noting that alEo >= 0, we get
< EoJx(t')x(t)JEo >= n e-
iw
('-') = 9
2mw
(153)
which is proportional to the Green function (for t' > t) we have seen several
times before. Repeating the calculation for t < t! and combining the results,
we get
< EoJT(x(t')x(t))JEo >= 2!w [9(t' - t)e-
iwW
-') + 9(t - t')e
iwW
-')]
(154)
where T( ) is the time-ordering operator introduced earlier.
The path integral representation of the same system has already been
covered in the earlier discussion. Since the action corresponding to (132)
is quadratic, the system allows exact solution.
We shall now extend the above analysis to a system of several harmonic
oscillators.
2.9. INFINITE NUMBER HARMONIC OSCILLATORS
Consider a bunch of N harmonic oscillators with the variables
{q,(t), q,(t) ... q,(t), ... qN(t)}. [We cbange the not.tion from x(t) to q(t)
to emphasize the fact that these degrees of freedom have nothing to do with
the (x, y, z) co-ordinates of physical space]. For the sake of convenience, we
will take the mass, m, of each degree of freedom, to be unity. The frequencies
will, in general, be different; let Wk be the frequency of the k-th oscillator.
Since the oscillators do not interact with each other, everything we have
381
developed so far can be tri\jally extended. We will quickly summarize these
generalizations.
The Hamiltonian is now the sum of the individual hamiltonians:
.{!-.12 22
H = L..- 2 (p, + w,q,)
k=1
(155)
Therefore, the energy eigenstates of the system are labelled by a set of N
integers (nl! "2' .. nk,'" nN). This state, denoted by I{ nk} > corresponds
to the situation in which the first oscillator is in the state labelled by nl,'"
the k th oscillator is in the state labelled by nk .. and the N'th oscillator
is in the state nN. The wave' function corresponding to this state will be
N
>l![%'12oooq",ooqN;t) = >l![{q,};t) 0 II >l!n.(q"t)
1;;=1
corresponding to the energy eigenvalue
- N I
E[{n,lI = L Iiw,(n, + -)0
k=1 2
(156) 0
(157)
For example, the lowest energy state will be the one with all no\: IS set to
zero. This "ground-state" has the wave function
N
>l!.[{q.}) = II ",.(q"t) =
k=l
IT ( ~ ) ' / '
1:=1 'lfn
with the ground state energy
N I
Eo = L'i
liw
,
k=l
(159)

Each oscillator has its own creation and annihilation operator 0.1, at. Ex-
cited states can be constructed by operating repeatedly by at. Note that
0.1 changes only the integer corresponding to the kth oscillator:
alln!. n2'" Ink,'" nN >= ..jnk + llnl. n2 ... fik + 1" .. nN > . (160)
The results derived earlier using path integrals for the harmonic oscillator
can also be extended easily for N-dimensional systems. Since the action
.{!-.j 1(02 22
A = L..- dt
2
q, - w,q,)
k=l
(161)
382
T.PADMANABHAN
is a sum of individual actions, the path integral Kernel becomes
(162)
where {qk}F stands (or the final values of all the coordinates
qk : {q[,qf etc. This Kernel represents the probability amplitude
for the system to propagate from an initial state to a final state, such that
first oscillator co-ordinate changes from qf to qi 1 second from to qf, etc.
etc.
We can also generalize the results obtained in the presence of external
sources J(t) to the N-dimensional system. Since there are N-oscillators
we will add N -external somcc functions [J, (t), ],(t) ... J.(t) ... IN(t)] by
adding to the action the term
N
A"'d = L: / dtq.(t)J.(t).
1;;=1
The transition amplitudes computed earlier now generalizes to
with
N
K[O,oo;O-oo;J.(t)] = II (detD.)-I/'
1;;=1

,
exp--W[J.]
2/i
N 1+
00
W[J] - -00 dtdt' J.(t)G.(t, t')J.(t')
N / dv IJ.(v)I'
(163)
(164)
- E (2w) v' - -i<J
N
- / (165)
where Jk(V) is the Fourier transform of Jk(t) and Gk(V) is proportional to
the Fourier transform of the Green function for the k'th oscillator:
1
G.(v) = , (' . r
11 - wk - tt:
(166)
The Green function for each oscillator, of course, can be expressed as
the expectation value of coordinate operators, i.e.,
G.(t, t') =< EoIT(q.(t)qk(t'llIEo > .
(167)
It is more convenient to write this in a somewhat more general form
Gkp(t, t') =< EoIT(q,(t)qp(t'))IEo >= OkpG,(t, t') (168)
With a view to future applications in mind, we want to reexpress the
above formulas in a different manner. Let us first consider the role of the
label'k' in the above analysis. We can think of k as labelling the positive
lattice points in a line - i.e., onedimensional space. In other words, we
consider the interval [0, ooJ in the real line and pick out from this continuum
of points the first N lattice points k = 1,2 .. N. To each of these points,
we associate a harmonic oscillator qk with frequency Wk
This picture suggests an immediate, and useful, extension of our for-
malism. We need not c o n f i n ~ ourselves to the positive side of real axis or
to the lattice points. Let us associate with each real value of a variable
k[-oo < k < 00] an oscillator qk with frequency Wk. We now have an un
countably infinite number oscillators labelled by a continuous index k. [To
be precise, we should now use a notation q(k, t) rather than qk(t); but we
will not bother to do this.]
We can take over all the previous results to this continuum case in a for
mal sense. For example, the ground state wave function of (158) becomes, .
(
Wk)1/4 1
II - exp -iEot - - LWkql
aUk 'lrn. 2n. allk
exp -1+
00
dk {/lw' _ ! In (Wk) }
-00 211" 2 2 I'm
(169)
For this expression to be rigorously meaningful, the integral in the expo--
nent should converge, This will put severe constraints on the nature of the
function WI;;; in fact, for most of the systems which we consider, WI;; will in
crease with k, for large k, The convergence condition will be badly violated
in this case and expressions like (169) need to interpreted very carefully.
We will say more 'about it later.
There are, however, other results which carryover neatly to the contin
uum case. For example, the exponent of (164) becomes
1
+00 dk 1+
00
dv
W = - -l,(v)Gk(V)Jk(V)
-00 21f -00 21r
(170)
This expression suggests another obvious modification. We note that v and
k appear on the same footing in W. The variable v corresponds to t on
Fourier transforming; in a similar manner, we may introduce a Fourier
conjugate variable far k. Let us call this, onedimensianal, real, variable x.

384 T.PADMANABHAN
Writing
J(t; x)
G(t;x)
(171)
We can put W in the form
r- r+=
W = i-co dtdx J-et::J d.t'dx' J(t; x)G(t - t'; x - x/)J(t/;x/)
(172)
Instead of working with the function JJ;;(v) or Jk(t), we can work with the
function J(t,x). All the features of the system are contained in the Green
function Gk{v) or -equivalently - in G(t;x).
The k-space (and its Fourier conjugate x-space) we have introduced is
purely a mathematical construct at this stage. We have merely attached,
to each point in the real line R{1L labelled by k - one harmonic oscillator.
Quite clearly, we can generalize this concept to aD-dimensional k-space,
labelled by a D-dimensional vector k = (k).,k2,kD). To each point, in
this space, we associate an oscillator qk' The k- integrations in previous for-
mulas involve the volume element [dk
1
dkD/(21r)DJ = [dk/(21r)DJ. Cor-
responding to this k-space, we have the Fourier conjugate x- space which
is again D-dimensional.
All the previous expressions can be easily generalized. For example,
(172) will become
r+= r+=
w= J_= dtdx J-= dt'dx'J(t;x)G(t-t';x-x')J(t';x')
(173)
with
. _ 1+
00
dvdk
G(t,x) - (2 )D+l 2 2 +'
-00 7l" V - w
k
J
(174)
and similar expressions for J(tjx) etc.
The original Green function for each oscillator we started with Gk(t, t'),
could be expressed as the expectation value < EoI(qk(t)qk(t'))IEo >. The
quantity G(t,x) can also be expressed in a similar fashion with suitably
defined operators. Note that, for t > t',
G(t-t';x-x')
J
dk J dp G ( 'ikx-ip.x'
- (21r)D (21r)D kp t - t)e
J
dk J dp I 'I ikx-ipx'
- (21r)D (21f)D < Eo q.(t)qp(t) Eo > e .
= J dk J dp E I( () ikx)( (') iPX')'IEo
(21f)D (21f)D < 0 q, t e qp < e . >
(175)
In arriving at the last expression we have used the fact that qk(t)'S are
i.e., qk = qk Let us now define an operator
(176)
Then we can write
G(t - t';x - x') =< EoI(t,x)"l'(t',x')IEo >; t > t'. (177)
We can take care of the condition t > tf by putting a time- ordering oper-
ator. Then we finally get
G(t - t';x - x') =< EoIT((t,x)'(t',x'JJIE
o
> (178)
expressing the Green function in the (x, t) coordinates in terms of the
-
ground state expectation value of the set of operators q,x(t). Note that
-
we have associated with each point X, this operator tPx(t).
Since we have come far away from the simple harmonic oscillator with
one degree of freedom, it is worthwhile to take stock what we have done.
These are the crucial steps of involved:
1. Consider a D-dimensional Cartesian space in which each point is la-
belled by a D-dimensional vector k.
2. We associate with each point in this abstract k dimensional space,
a harmonic oscillator of unit mass and real, positive, frequency WJc.
Let thc(t) be the (Heisenberg picture) operator characterizing the k th
oscillator.
3. We introduce next another D-dimensional space which is Fourier con-
jugate to the k-space. The points in this space are labelled by the
D-dimensional vector x. Physical variables in these two spaces are
related by Fourier transform.
4. The Kernel in the presence of external sources Jk(t) can be written as
.

K[J] = eXP-2ijW
(179)
386 T.PADMANABHAN
with
WIJI
/ 1:
00
dtdt' Jdt)Gdt, t')Jk(t')

/ dxdt / dx'dt' J(t;x)G(t - t';x - x')J(t';x') (180)
where
J(t;x)
G(t;x)
5. The properties of the system is contained in the function G(tj x) which
- in turn - depends only on the frequency of the various oscillators,
i.e., the function w(k) = Wk.
- -
6. To each point in xwspa.ce_associate an operator rh.(t) = ",(tiX) given
by
l( ) / dk _()ik.x
"t;x = t e
The Green function can be then expressed as
/
G(t - t';x - x') =< EoIT(4)(t;xW(t;xllIEo >
(182)
(183)
where lEo > is the state in which all the harmonic oscillators are in
the respective ground state.
The above analysis was done using normal time co-ordinate and hence
contain exponents which oscillate. It is, of course, possible to carry out the
same analysis in the Euclidean sector. It is easy to see that we will then
obtain the expression
with
1
K[J)
2;'
1
+00 1+00
w= -co drdx -00 dr'dx'J(Tjx)G(T-r'jX-x)J(T'i X')
where
-f
dvdk. e-
ill
1"+ik.x / dvdk ei(k.x+vr)
G(T"X) - -
, - (21f)D+l v2+Wk - (21T)D+l v2+wi
(184)
(185)
(186)
This expression has au interesting connection with the Laplace transform
of the free-particle propagator. The Euclidean propagator for a free-particle
in N -dimensions is
(
m )NI' 1 mlxl'
K(xj'T) = 21rnr exp-
n
_ 2r
(187)
where x = X2- - Xl and T = 7"2 - rl. The Laplace transform of this expres-

Slon,
F(x; >'} = /.00 dTe-" K(x; T}
_ can be evaluated easily by wdting K as
!
dk liT . ?
K(X;T} = (21r}N exp(ik.x- 2m"-)
and doing the T integration first. This gives,
! dk . t .. '
F(x; >.}
. e1k.x dTe -T(J..+ 2m )
-
(21r)N 0
-
! dk ik.x 1
(21r)N e (>. + /ik'/2m)
em)! dk e
ikx
-
-
Ii (21r}N (k' + I")
(188)
(189)
(190)
where we have defined - 1'2 = (2m>./h). Let us separate out of the
N-dimensional integral [dk] = [dk,dk, dkN] the N-th integral and write
. (2m) ! dk.dkN ei(k.x+kH:rn)
F(Xl. X, ... XN; >.) = F(x, XN; >.} = Ii (21r)N ( k ~ + k' + I")
(191)
in which x and k stand for (N - 1) dimensional vectors. This expression is
essentially the same as G(T;X) if we identify N = D + Ii XN = Tj kN = v
and choose the frequencies of the harmonic oscillators to be given by
w'(k} = k' + 1"
(192)
We shall later show that p2 is essentially the energy of a free particle.
This gives us the important identification between: (1) one free particle in
N -dimensions with energy p2 'and (2) an infinite number of oscillators in
an (N -1) dimensional space with the frequency relation w
2
(k) = k
2
+ p2.
As we shall see, this result is of fundamental significance.
388
T.PADMANABHAN
2.10. PATH INTEGRALS WITH JACOBI ACTIOK
We shallllow consider a different description of path integral Kernel, which
turns out to be extremely important. Let us begin by observing that the
action in classical mechanics can also be written in the form
A[x(t),p(t)] = {dt[P.X - H(p,x)]
(T p'2
- 10 dt[p.x - 2m - V(x)].
(193)
To facilitate future use, we are considering a theory in aN-dimensional
spacei the position x and momentum p are N-dimensional vectors. If we
vary p and x independently in the above expression we get
T .
5A = I, dt [(X - - (I'> + V'V).ox] + p.oxli (194)
We confine our attention to variations for which ox = 0 at t = 0, T but op
is entirely arbitrary. This gives,
p .- mx;1'> = -V'V (195)
which are the Hamilton's equations. We can now attempt to write the path
integral using (193) in the form
K(X2,Xl;T) = L
allx
dt(p.x-H(p,x))
allp 0
(196)
where the sum is over all functions p(t) but only those x{t) which has the
correct boundary values. The sum over all p can be performed by slicing
the time axis into small intervals, in a manner identical to what was done
earlier. This will lead. to an expression for the Kernel as the sum over all x
of some amplitudej this amplitude turns out to be the usual one with the
lagrangian form of the action.
We will, however, proceed in a somewhat different way in calculating
(196). Instead. of summing over x and p, we will first (i) sum over all x(t)
(satisfying the boundary conditions) for a given value of H(p, x) = E and
then (ii) sum over all values of E. That is, we replace (196) by,
InTdt(P.X-E)
allEIlllx h 0
. T
- L L exp':' In dt {2m(E - V(x))) I/'Ixl - ET
allEallx Ii 0
- L exp (-iET) L exp r
T
dtV2m(E - V(x))';2 (197)
ailE 1i!lx h 10
The 'sum over aU E', is just an integration over the parameter E with
some weightage g(E} [if needed]. Assuming the Hamiltonian is bounded
from below we can normalise the Hamiltonia.n so that E > O. Then we can
write,
K(x2,x,;T) _ /,00 dEg(E)e-;ETG(x2,X,;E)
_ /,00 dEe-;ETB(X2,X,;E) (198)
where
. T
G(X2,X,; E) = L exp k 1 dt"j2m(E - Vx))x
2
(199)
allx 0 ,
Relabelling t by s = (tIT), we get
. I
G(X2, x,; E) = L exp ~ 1 ds
allx 0
dx
2m(E - V)I-I'
ds
(200)
In this form, the paths are parametrised so that x(O) = Xl and x(l} = X2'
This expression offers an interesting interpretation. By inverting the
Fourier transform in (198), we can write.
B = g(E)G(X2,Xl;E) = roo dTekETK(x2,Xl;T) (201)
10
[Strictly speaking, the Fourier integrals go over the range (-00, +00). How-
ever, K(T) vanishes for T < 0 and B(E) vanishes for E < O. We will not
bother to indicate this fact explicitly every time]. Since K(X2,XliT) gives
the amplitude for propagation from Xl to X2 in a time interval T it is
natural to interpret B as the amplitude for propagation between these two
points with energy E. This interpretation is also borne out by the fact
that argument of the exponent in (200) is what is called the Jacobi action
capable of describing paths of particles with definite energy.
Notice that we have also obtained something much more useful: we have
provided an interpretation for the 'sum over paths' appearing in (200)! Since
the Jacobi action has a square root in it, none of our previous definitions of
'sum over paths' will work in this case. We can now take (201) as defining
(200). That is we take
L g(E) exp ~ j ds 2m(E - V)I ~ ~ I'
allx 0
1
- L exp k r d,,,j2m(E - V)lxl'
all x io
390
T.PADMANABHAN
00 T
J dTekET J Vx(t) exp J dt mx' - V] (202)
o 0
In arriving at the second expression we have redefined the measure in "sum
over paths" by absorbing the g(E) factor.
There is one crucial point in the above definition which needs to be
mentioned. The Jacobi action, defined as,
(203)
has the property of being "reparametrisation' invariant" under 'time rela-
belling'. This is clear from the fact that, on performing the transformation
t -+ 'T = T(t) where
t = /,T N(r)dr'
the integrand does not
J dtV2m(E - V)lxl' - J NdrV2m(E - V)N-'lxTI'
(204)
- J dr 2m(E - V)I;j;I" (205)
Therefore two different paths x(t) and x(t(r)) = x'(7) will give the same
value to the action, even though the functional form of x(t) and x'(r) may
be quite different. Obviously, we will get an infinite value for G I if we count
all these paths are different. Therefore, if the sum over paths in (200)
should be well-defined, we must remove this 'reparametrisation invariance'
by choosing a definite parametrisation. This is precisely what is achieved
by the rescaling t -+.,. = tiT in the definition of (200).
The classical equations of motion are invariant even if A is replaced by
-A. If we had used the negative of Jacobi action in (197), then we would
have obtained, in' place of (200) and (201). the relations:
i /,' dx
- I: exp -- ds 2m(E - V)I_1
2
aIlx 1'1, 0 ds
(206)
Let us now consider the Euclidean analogue of this result. With the
lagrangian form of the action, we could tackle the oscillating exponents by
analytically continuing in t through the substitution T = it. This procedure
converts the quantity
(207)
to a negative definite expression (we assume V > 0). But if we change the
parameter s in (200) to 1] = is we will get a positive dennite expression in
the exponent
dx j
2m(E - V)I-I' = <iii
ds
.
dx
2m(-E+ V)ldijl'
(208)
for all E < O. The proper change to make is not is = ij but fJ = -is.
Then, we get, for E < 0,
i j ds 2m(E - V) (.)' = - j d ~ 2m(IEI + V) (.)' (209)
In Euclidean space (202), becomes
= dre- Ii Vxexp-- (-mx2+V)dT
1.
00 Illi j 1 1.' 1
o h 0 2
(210)
Let us evaluate this expression for a free particle in N -dimensions. The
Euclidean free particle Kernel in N -dimension is
K(X;T) = jvxexp-!.[ !.mlxl'dT
Ii 0 2
(
m ) 'l' 1 mlxl'
= 21r"7 exp -Ii 27 jX ~ x2 - xl
Therefore, calling lEI = ~ ,
roo _k ( m )N/' 1 mlxl'
B(x;..\.) = 10 dTe /I. 21rh7 exp -Ii 27
Writing
(211)
(212)
(213)
392 T.PADMANABHAN
and doing the r-integration first, we get
(= "f dk 'k ,,'
B(x;)..) 10 dre-
T
(211")Ne! .X-:r;;;-T
f
~ eik.x [CO dre-;(>.+A::'
7
)
(2.)N Jo
f
dk ik.x 1 h
2
k
2
-
1
- (2.)N
e
Ii A + 2m
(214)
This can be rewritten in the form
'(2m) J dk e'k.x
B(x; A) ~ h (2.)N (p' + Ikl')
(215)
where 1-'2 = (2mA/h
2
). This form is strikingly similar to the Euclidean
Green function we derived earlier for the D-dimensional harmonic oscilla-
tor system. We found that,
J
dvdk eik.x+iVT
GEu(X; T) :- (2.)D+l v' + w'(k)
(216)
The identification between Band GEu follows if we choose the frequencies
of the harmonic oscillators to be
(217)
andN=D+l.Then
f
dvdk. e
ik
.
x
+
iVT
GEu{X,V) = }D 1 2 2
(211" + v +k +,,2
J
dp e
tp
.
x
- (21r)N p2 + ,,' ~ GEu(X); N ~ D + 1
(218)
where the (D + 1) dimensional vectors p and x have the components (k, v)
and (x, T). Comparing (218), and (215), we see that
2m
B(x; A) ~ h GEu(x)
(219)
To make the identification complete, we will choose m = (1/2) and set
h = 1. [This will make ,,' = (2mA/h') ~ A = IEIl. Then we can write
B(x;,,') _ j,= dTe-.'TK(x;T)
- 10= dre-
p2T
/ Vxexp- foT dr [li?]
f
dk e'k.x
- GEu(X) = (21r)N k2 + ,,"
(220)
This is a remarkable result! The B(X;JL2) represent the amplitude for a free
particle in N-dimensional space to propagate from Xl to X2 with energy
/-,2, The GEu{X) is the Green function for a system with infinite number of
harmonic oscillators labelled by a (N - 1) dimensional vector k), with a
particular dispersion relation w
2
(k) = k
2
+ JL2. The above result, therefore,
expresses an equivalence between a free particle in N - dimensions and a
system of infinite number of oscillators in (N -1) dimensions.
)(2.11. RIGOROUS EVALUATION OF THE JACOBI PATH INTEGRAL
In the discussion above, we defined the path integral over the Jacobi action
by connecting it up with the more conventional path integral which we had
already defined rigorously earlier. In doing so, we have bypassed the need
for defining the measure for the Jacobi integral from first principles.
It is, however I interesting to consider the question of defining the Jacobi
path integral from the first principles. This analysis will allow us to explore
an alternative means of defining path integrals.
We will work directly in the Euclidean sector of D-dimensions. Since
we are primarily interested in the issues of principle, regarding the measuz:e
for the path integral it is enough for us to consider the path integral for a
free particle. We have to, therefore, evaluate
i [' (dX)' I
E exp - J. dsv'2mEI d I"
a1lx(t) h 0 s
-
-
(' (dX)'
E exp il" J. <lsI d 1'/'
allx(t) 0 s
(221)
where we have defined ~ ~ = (2mE/n2) as before. We notice that the quan-
tity
I(X2,X1) = J,' dsl (Ts)'I'/2 (222)
is just the length of the curve x(s), connecting x(O) = Xl and x(l) = X2.
We will, therefore, consider the path integral
gE(X2, Xl; 1'0) = E exp-I'ol[x(t)]
allx(t)
in the Euclidean sector, where l is the length of the path.
(223)
This quantity can be defined through the following limiting procedure:
Consider a lattice of points in a D-dimensional lattice with a uniform
lattice spacing of E. We will work out YE in the lattice and will then take
394
the limit of --7 0 with suitable measure M(). To obtain a finite answer,
we also have to treat jlO (which is the Duly parameter in the problem) as
different for different lattice spacings; i.e., I-' = 11{} in a specific manner.
We will reserve the symbol Po with a subscript, for the parameter in the
continuum limit. Thus we have
(224)
In a lattice with spacing of , (223) can be evaluated in a straightforward
manner. Because of the translation invariance of the problem, {IE can only
depend on X2 - Xli so we can set xl = 0 and call X2 = ER where R is a
D-dimensional vector with integral components: R = (nil n2. na'" nD).
Let C(N, R) be the number of paths of length NE connecting the origin to
the lattice point fR. Since aU the paths contribute a term exp-jJ(e)(Ne)
to (223), we get,
00
9E(R;<) = L (225)
N=O
It can be shown from elementary 'combinotrics\ that C(NjR) are deter-
mined by the generating function
pN = [eik1 +eik2 + ... eikD +e-ik1 + ... e:....ikD]N = L:C(NjR)eik.R
R
(226)
Therefore,
00
Le
ik
.
R
9E(R;<) = L
R N=O R
-
00 00 N
L e-p()N FN = L [Fe-
p
(,),]
N=O N=O
- [1- Fe-P(')'r
1
(227)
(228)
We are nOw ready to take the limit of zero lattice spacing. As f --+ 0,
the denominator of the integrand becomes
(230)
so that we will get, for small f,
/ dDp A(,),,-ip.x
QE(X;') '" (2w)D Ipl' + B(,)

(231)
where

A(,) -
,p-2
e
Jl(t)
B(,) ~ [ep() - 2D]
,
(232) -
-
The continuum theory has to be defined in the limit of E -+ 0 with some
measure M(); that is we want to obtain
(233)
The choice of the measure is dictated by the requirement that the right
hand side should be finite in this limit. It is easy to see that, we need to
demand
and
lim [M(E)D-2
e
f
P
(f)] = 1
, ~ o
The first condition implies that, near ~ 0,
()
In2D ,,!, ln2D
IJE:::: + ~
2D
The second condition (235), allows us to determine the measure as
With this choice, we get
1 1
M(,) = 2D ,D-'
. / dDp e-
ip
.
X
11m QE(X; ,)M(,) = (2)D I I' 2
(-+0 11' p - + JLo
(234)
(235)
(236)
(237)
(238)
396 T.PADMANABHAN
which is the continuum Green function we have computed earlier. This
analysis gives a rigorous meaning to the Jacobi path integral.
The measure in (237) has a simple interpretation. Since IJ.. scales with
energy as p. ex: E
1
/
2
, (236) shows that scales as ex: E-
1
j2
j
combining this
with {237} we get M() IX -(D-2) IX E ~ -1. This is precisely the scaling
of density of states g(E) for a D-dimensional system with E ex: Ip12. Thus
the measure merely keeps count of the density of states between E and
E+dE.
3. The Concept of Fields
3.1. INTRODUCTION
The purpose of this short section is to present the material discussed in
section 2 in the context of quantum field theory. In fact, much of the mo-
tivation for the discussion in the last part originates from this connection,
which we will now explore. Section (3.2) establishes the connection between
the path integral for a relativistic particle and the concept of a quantum
field. Some general properties of quantum fields are discussed in sections
(3.3) and (3.4).
3.2. PATH INTEGRAL FOR A RELATIVISTIC PARTICLE
Consider a free particle of mass m moving in accordance with the laws of
special relativity. Its world line xi(s)[with i = 0,1,2,3 and signature + --
-1 satisfies the equation (tPxild-r) = O. This equation can be obtained
from varying several different actions of which the following two are of
primary importance: (i) The quadratic action
Aq"", = ~ J :iix.ds
(I)
and (ii) the square-root action:
Asq = -m f V xiXids (2)
[The multiplicative constant in front is not determined by the classical
equations of motion but is chosen for future convenience]. While the two
actions lead to the same equations of motion, they differ in several other
properties. It is usual to use equation 2 in classical relativity since it gives
the correct expression for the energy and momentum.
Let us now consider how one can obtain a quantum theory for the
relativistic particle. Following the philosophy of path integrals one would
o
like to construct the amplitude for a particle to propagate from the event
yi at s = 0 to the event Xi at s ::::: T:
K(x"r;y',O) = exp; A[x(s)] (3)
alJxi(S)
where A is a suitable action. Several, highly non trivial, problems arise at
this stage:
To begin with, let us consider the possible physical meaning we will
attribute to K(x
i
I Tal. 0). At first sight one might think of K as providing
the probability amplitude for a particle to propagate from ti to xi in a
proper time interval T. However I such an interpretation has some strange
features: To begin with, the proper time lapse T is not a physically relevant
parameter in the problem; wha,t we are really interested in is the probability
amplitude for the particle to propagate from Vi to xi irrespec.titJe of the
elapsed proper time. In other words, we have to integrate over T with a
suitable weightage factor.
The suitable weightage factor is easy to obtain for the action in equa-
tion (1). Since the energy of the particle in the rest frame is m and since
r denotes the time in the rest frame, a world line with a'lapse factor of
T should have a weightage of exp( -imT). Evaluating the quadratic path
integral in equa.tion (1) and incorporating the weightage factor we get

G(x, y) - f dTe-
imr
K(x, ij y, 0)
o
_ -'mT (. m )1/2 (mi )3/' [im(x-v)']
..,. dTe '4' 4 exp 4
- 1rTI 1fT T
o
_ exp im[(x-
y
)' -r] (4)
T2 411" 4r
o
This integral can be expressed in terms of Hankel functions. However, we
have evaluated this integral in a different context in section (2.10). [see
equation ( ... )]. Using the rela.tion
(
m)' im(x-y)' Jd'k (ou.. I
k
,)
exp- = exp T
4"r; 4 7 (2,,)' m
(5)
and performing the T integral, it is easy to show that
roo dr (m)'
i 10 T2 411'" exp im
(6)
398 T.PADMANABHAN
where G(x, y) is as
(7)
The situation is more complicated if we try to use the square root ac
tion. But note that this action is identical to the Jacobi action discussed
in section (2.10) with E = mo, m = (mo/2) and V = O. We saw that such
actions will lead to divergent amplitudes because of reparametrisation in-
variance. In fact, the study of Jacobi action showed that we can choose a
specific parametrisation and interpret the sum over paths by the rule:
looT
L exp - ~ I dl1 rna"; Xi
xi
-= I dr e-
imoT
f Vx exp ~ :0 I :i;
2
ds (8)
path 0 0 0
This is precisely the connection between the two different path integrals.
Thus both action fUDctionals, when correctly interpreted, lead to the same
propagator G. There are, however, several interpretational problems related
to our path integral which we shall now discuss.
3.3. FIELDS AND OSCILLATORS
The propagation amplitude for a free relativistic particle, as we have_seen,
can be defined in terms of the path integral and action functional. To this
extent, relativistic and non-relativistic particles behave in a similar manner.
Unfortunately, this similarity is superficial. Closer inspection shows that
there are serious conceptual differences in the two cases which needs to be
taken into account.
To begin with, note that there is no assurance that Xi is in the future
lightcone of yi or even that xO > yO. There will be non-zero amplitude
for a particle to propagate outside the light cone or even to go 'backward'
in time. The second feature arises from the fact that we are DOW treating
all the coordinates (xO,x
1
,x
2
,x
3
) as the dependent variables and 8 as the
independent variable in evaluating the path integral. There is no reason
for the final value of time (xO) to be greater than the initial value of time
(rl) just as there is no reason for the final value of any other coordinate
to have a definite relation with the initial value. This creates serious prob-
lems of interpretation. One immediate consequence of this fact is that the
propagation amplitude will not obey the 'transitivity rule' in time:
(9)
This implies that one cannot introduce a conserved probability density and
current unlike in non-relativistic mechanics. We cannot study the quantum
theory of a relativistic particle in terms of a wave function ,p{X
O
, x) such
that G(x',x) propagates (XO,x) from Xi to x\
There is a fundamental reason for this difficulty. To see this reason
most clearly, consider the paths which contribute to the amplitude. Even
when xO > yO, the amplitude will pick up contributions from paths like
YABX shown in Figure 2. Consider now an intermediate time t such that
xO > t > yO. From the figure it is clear that the relativistic particle is
located at three different locations 1, 2 and 3 at this instant!. One can,
of course, draw other paths which will cut the t line at arbitrarily large
number of points. In other words, the propagation amplitude takes into
account configurations contcUQing arbitrarily large number of particles at
intermediate times. We can no longer consider the quantum theory of the
relativistic particle as describing the dynamics of a single particle at all
times.
,
x
y
-+----------x
- -
The above fact is of crucial importance. To accommodate this aspect,
one can use two different (but, of course, related) physical pictures. In
the first picture, one interprets the part of the trajectory which is going
"backward in time" ~ i k e the part AB in Figure 2J as an "anti particle" going
"forward in time". Further, we shall assume that particle-antiparticle pairs
can be created and annihilated out of the vacuum repeatedly. In such a
picture, the propagation of particle from y to x in figure 2 will be interpreted
through the following sequence of events: (ij A particle propagates from Y
to Aj (Ii) a particle-antiparticle pair is created out of the vacuum at B; (iii)
the antiparticle created at B annihilates with the (original) particle; (iv)
the particle created at B propagates to X. Notice that we are forced into
400 T.PADMANABHAN
such a description because the path integral takes contribution from the
paths like the one in figure 2.
The second (and more useful) picture for interpreting the situation is as
follows: Notice that the above description stresses the fact that we can no
longer talk about the propagation of a particular particle from y to x. All
we can do is to study the amplitude for a particle to disappear at yi and
a particle to reappear at xi. Let us denote by 10 > a quantum state with
no particles. A quantum state Ixi > with a particle at some location xi can
" .
always be obtained from 10 > by the action of some suitable operator .p(x
1
)j
i.e., we write Ixi >= (xi)IO > . [In fact, this defines the operator (xi)).
The amplitude that a particle at yi will end up at xi is given by
'P =< x'ly' >=< Ol";! (x')";(y')IO >=< Olql(x')qI(y')IO > (10)
[For the sake of simplicity, we have assumed that q, in hermitian; we have

also dropped the caret from q, with the understanding that q, denotes an
operator ]. Since 1> is also given by a path integral expression, we must
have the fundamental relation
< OI.p(x)qI(y)IO >= G(x,y)
(11)
In other words, we need to find an operator q,(x) such that this relation is
satisfied.
But, we have already tackled this problem in Section 2! We showed in
section (2.10) that the propagator G(x,y) can be expressed in terms of
the Green function for an infinite number of oscillators with the dispersion
relation w
2
= k
2
+ m
2
Using the expression for the harmonic oscillator
Green function derived in section (2.9), we can write (for xO > yO)
G(x,y)
e-ik(:r-y}
where the operator q, can be expressed in the form
" f ."" d'k
";(t,x) = q.(t)e' (2,,)'
(12)
(13)
with qk(t) obeying the (Heisenberg picture) equations of motion of a har-
monic oscillator
cP +w2 q" O. w2.=k2+m2
dt
2
k k=,
(14)
and lEo' > representing the ground state of the oscillators. This relation
is the cornerstone of quantum field theory: The amplitude G(x,y) for a
relativistic free particle to propaga.te from 1/ to Xl can be expressed as the
ground state expectation value of a 'fiel9.' 4>(t, x) which can be constructed
from a bunch of oscillators qk(t). Since G(x,y) is the central quantity we

are interested in, we could equally well study the dynamics of (t, x). This,
in turn, is equivalent to studying a system of harmonic oscillators {qk}
which we have already done in Part 2. For the sake of completeness we
shall summarise the key results:

(1) To begin with, it is easy to verify that <p itself satisfies the equation of
motion
(8;iJi + m
2
)(t, x) = 0
which is fully relativistic8J.ly invariant. Similarly, using the relation
%(t) = J (t,x)e-,kxd'x
in the action for the harmonic oscillators
J J
d'k
A = dt (2,,)'
we can find the action for the field (t,x). We find that
(15)
(16)
(17)
(IS)
Where Oi = (8t/>/8:c
i
) etc. Quite clearly, this action is also Lorentz invari-
ant.
(2) Let us next consider the energy eigenstates of the system. Since the
energy eigenstates of the k-th oscillator is labelled by an integer nk, the
general energy eigenstate is described by a set offntegers {nkl' nk
2
,'" etc.}.
This state will have the energy
1 1
E - L"'k(nk + 2") = Lv'k
2
+m2(nk+ 2")
k k
- Eo + Ln
k
v'k
2
+ m
2
(19)
k
where Eo = Lk(Wk/2) is the energy of the ground state 10,0, ... > which
has all nk = 0. Since the energy of one relativistic particle with momentum
k is Wk = .../k
2
+ m
2
, we can interpret I{nk} > as ... the state containing
nk
l
particles with momentum kl and energy Jki + ~ 2 , nk
2
particles with
402
T.PADMANABHAN
momentum k2 and energy + m
2
.. u etc. As an example, consider the
state Ixi >= (t, x)IO > we introduced earlier. Using the results of section
(2.8) and noting that 10 > has to be identified with ground state of the
oscillators lEo >, we find
lx' > = (t,x)lO >= f >
(20)
This state is clearly a superposition of I-particle states with different mo-
mentum k. This is the closeSt to the concept of a "particle located at xi"
which is possible in relativistic field theory.
(3) The ground state ("vacuum") of the quantum field has several pecu-
liarities which need to be stressed. To begin with, note that this state is
an eigenstate of the number operator nk = akak with eigenvalue zero for
all k [This is why we think of ground state as a state with no particles or
a "vacuum state"]. But this has non-trivial dyna.mical features. For
example, it has infinite amount of energy :
1 f d'k 1 .
Eo = L -v'k
2
+ m
2
= -v'k
2
+ m
2
= hm (k')
k 2 2
(21)
No satisfactory interpretation of this feat'ure is available at present, though
several mathematical procedures which will allow one to systematically
subtract out Eo from observable effects can be devised.
The ground state also contains quantum fluctuations in the field tjJ(x, t)
which can lead to nontrivial effects. To see this, we can use the 'coordi-
na.te' representation of the ground state. This will be discussed in the next
section.
The discussion so far can be summarised as follows: The propagation
amplitude for a free relativistic particle can be interpreted either directly
in terms of path integral or in terms of the ground state expectation value
related to the field operator (x). While they lead to the same G(x,y},
these approaches are conc;eptually very different. It turns out that the de-
scription in terms of harmonic oscillators and fields is far more general
than the approach based on path integral for a relativistic particle. In par-
ticular, it is quite easy to discuss particles with nonzero spin in a Lorentz
invariant manner using the field approach. One can easily construct spinor,
vector or tensor fields [all of which can be decomposed into harmonic oscil-
lators] which could represent spin-l/2, spin-l and higher spin particles. The
dynamical equations for these fields can be obtained from suitable action
functionals for the fields themselves.
Interaction between particles can be represented by suitable coupling
between fields in the action functional For example one of the simplest
ways of including an interaction term for a scalar field is to modify m
2
(p2
to a more general function V{IjJ) which we will discuss in the next part.
4. The Technique of Effective Action
4.1. INTRODUCTION
We have developed the technique of path integral in section 2 and the ba-
sic concepts of quantum field theory in section 3. In this section, we shall
combine these two concepts to develop an approximation technique which
will allow one to study the behaviou,r of a quantum field in a given external
background. Such an approximation is particularly useful in studying grav-
itational interactions. Since we do not have a complete theory of quantum
gravity, considerable amount of attention has been focussed in the study of
quantum field theory in a given background gravitational field. The tech-
niques developed in this part' has special relevance to suell studies which
will be. developed in section 5.
In section (4.2) we shall illustrate the basic idea using a quantum me-
chanical example followed by that of a scalar field. The concept of renor-
malization so central in the study of quantum field theory is introduced
in the context of a self interacting scalaX field in sections (4.3) and (4.4).
The last three sections (4.5). (4.6) and (4.7) develop the Euler-Heisenberg
effective action for the electromagnetic field and illustrate the phenomena
of pair creation in an external electric field from different points of view.
4.2. THE CONCEPT OF EFFECTIVE ACTION
Let us consider a theory which describes the interaction between two sys-
tems'having the dynamical variables 0 and q. [This notation is purely for-
mali the symbol 0, for example, could describe a. set of variables, like the
components of a vector field. The detailed nature of these variables is not
of importance at this stage.] The action functional for the system, A[Q,q]
will then depend on both the variables, possibly in a fairly complicated
way. The full theory can be constructed from the exact Kernel
which is quite often impossible to evaluate. It would be, therefore, useful
to have some approximate ways of handling the situation.
404
T.PADMANABHAN
The nature of approximation which one should use in evaluating (1)
depends on the physical situation. Different contexts will require d i f f e r ~
ent approaches and one should not consider anyone technique to be of
univer::;al applicability. The 'effective action' method is one of the many ap-
proximation schemes available for handling (1). This method is of value in
the following context: It may turn out that, we are interested in a physical
situation in which one of the variables, say, Q, behaves nearly classically
while the other variable is fully quantum mechanical. In that case, we can
attack the problem in the following manner:
Let us suppose that we can do the path integral over q in (1) exactly
for some given Q(t). That is, we could evaluate the quantity
(2)
treating Q(t) as some specified function of time. If we could now do
(3)
exactly, we could have completely solved the problem. Since this is not
possible, we will evaluate (3) by invoking the fact that Q is almost classical.
This means that most of the contribution to (3) comes from nearly classical
paths satisfying the condition
JW =0
JQ
(4)
It is often quite easy to evaluate (3) in this approximation and thereby
obtain an approximate solution to our problem. In fact, quite often, we will
be content with obtaining the solutions to (4), and will not even bother to
calculate (3) in this approximation. Equation (4), of course, will contain
some of the effects of the quantum fluctuations of q on Q, and is often called
the 'semiclassical equation'. The quantity W is called the 'effective action'.
It is also convenient to define an 'effective lagrangian' through the relation
W = ! L",dt
(5)
There is, however, one minor complication in the aboye formalism. The
way we have defined our expressions, the quantities K and lV depends on
the boundary conditions (t2, q2. tl. ql). We would prefer to have an effective
action which is completely independent of the qdegree of freedom. The
most natural way of achieving this is to integrate out the effect of q for
all times by considering the limit t2 -? +00, tl --+ -00 in our definition
of the effective action. We will also assume, as is usual, that Q(t) vanishes
asymptotically. From our discussion in section 2 we know that, in this limit,
t.he Kernel essentially represents the amplitude for the system to go from
the ground state in infinite past to the ground state in the infinite future.
Using the results of section (2.7), we can write

F(q" ql; +00, -(0) '" exp ;;W[Q(t)] = N(q"ql) < Eo, +001- 00, Eo >Q,ti
(6)
where < Eo. +001-00, Eo >Q(t) stands the vacuum to vacuum amplitude
in the presence of the source Q(t) and N(q, q2) is a normalization
factor, independent of Q(t). Taking logarithms we get
W[Q(t)] = -in In < Eo,+ool-oo,Eo >Q,ti +(constant) (7)
Since the constant term is independent of Q it will not contribute in the
equation (4). Therefore, for the purposes of our calculation we may take
the the effective action to be defined by the relation
W[Q(t)] = -inln < Eo, +001- 00, Eo >Q,li
(8)
in which all reference to the quantum mode is eliminated. Note that [Eo>
is the ground state of the original systemj that is, the ground state with
Q=O. .
This discussion also highlights an important feature of the effective ac-
tion. We have seen in section 2 that an external perturbation can cause
transitions in a system from ground state to excited state. In other words,
the probability for the system to be in the ground state in the infinite fu-
ture (even though it started in the groundstate iil----rne infinite past) could
be less than unity. This implies that our effective action Wo need not be a
real quantity! If we use this W directly in (4) we have no assurance that
our solution Q will be real. The imaginary part of W contains information
about the rate of transitions induced in the q-system by the presence of
Q(t)j or - in the context of field theory - the rate of production of particles
from the vacuum. The semiclassical equation is of very doubtful validity if
these excitations drain away too much energy from the Q- mode. Thus we
must confine ourselves to the situations in which
ImWReW
(9)
In that case, we can modify the semiclassical equations to read
liRe (Wo)
liQ = 0
(10)
406 T.PADMANABHAN
In most practical situations, (9) will automatically arise because of
another reason. Notice that the success of the entire scheme depends on
our ability to evaluate the first path integral in (2). This task is far from
easy, especially because we need this expression for an arbitrary Q(t}. Quite
often, one evaluates this expression by assuming that the time variation of
Q(t) is slow compared to time scale over which the quantum variable q
fluctuates. [Once (4) is solved we can explicitly verify whether the solution
to (4) which we are interested in satisfies this assumption or not. This
acts as a reasonable test on the reliability of the theory.1 In such a case,
the characteristic frequencies of the q-mode will be much higher than the
frequency at which Q-mode is evolving and hence there will be very little
transfer of energy from Q to q. The real part of W will dominate.
The above discussion allows an alternative picture of the effective action
which is very useful. Let us suppose that Q(t) varies slowly enough for the
adiabatic approximation to be valid. We then know - from our discussion
in section 2 - that the 'vacuum to vacuum' amplitude is given by [see (99)
of section 2).
. +00'
lim lim F(q"ql;t"tl) =W= (const).exp"'-/i ( Eo(Q)dt. (11)
h-+oo h -+-00 J-
oo
This expression allows us to identify the effective lagrangian as the ground
state energy of the q-mode in the presence of Q:
L = -Eo(Q)
(12)
This result, which is valid when the time dependence of Q is completely
ignored in the calculation of Eo, provides an alternative means of compu-
tation of the effective lagrangian if the Q dependence of the ground state
energy can be ascertained.
The transitions to the higher states, indicated by the existence of an
imaginary part to Wo, can also be discussed in terms of the above relation.
The Wo can become complex only if Leff. and hence Eo becotDes complex..
The appearance of an imaginary part to the ground state energy indicates
an exponential decay probability for this state with some half life. This is
preciSely what we expect if transitions to higher states are possible.
The above discussion may suggest that whenever Q varies slowly enough
the real part of W - or, equivalently. the real part of Leff. - will give the
dominant contribution. If that is the case, we should get no imaginary part

to W when the time variation of Q is highly suppressed by treating Q as an
adiabatically varying parameter. This is usually true but one must make
sure that a ground state exists for the range of Q values considered in the
problem. As a simple example, consider the action
(13)
We see that q behaves as a harmonic oscillator with the effective frequency
(14)
It is possible to arrange matters so that Q becomes larger than w in the
course of the evolution even though Q vanished in the asymptotic past and
was increasing arbitrarily slowly. If this happens, no vacuum state will exist
for the q-mode and our calculation will lead to an imaginary part for the
effective action. So, in general, the existence of an imaginary part to the
effective action may either be due to transitions to higher states or due to
the non-existence of the ground state. In the course of our discussion we
will come across examples for both the situations.
We have so far considered the situation in which the action describes two
diHerent physical systems which are interacting. The idea of effective action
CaD also be used in a different context. Consider a system described by a
single dynamical variable, say, Q(t). Let us suppose that we are interested
in a situation in which the behaviour of the system is nearly classical with
small quantum fluctuations around .the classical solution. We would like
to estimate the effects of small quantum fluctuations around the classical
behaviour. For this we can proceed in the following manner:
We expand the action A[Q] as a Taylor series in q = (Q- Q,), where Q,
is the classical solution which we are interested in, retaining up to quadratic
terms in the deviation (Q - Qc). We will then get .
A[Q]
. 1 t'A
- A[Q,+q] = .40+ 2 6Q'
(15)
[The linear term will vanish because Qc is a classical solution.] The action
in (15), CAD be interpreted as describing the coupling between a nearly
classical mode Qc and the quantum fluctuations q. The effect of quantum
fluctuations on Qc can therefore be studied by using an effective action
~ f f [ Q c ] which is obtained by integrating out q in (15). We define
exp *A,.[Qd - J Vqexp *A[Q<+ q]
= exp *A,[Qd. J Vqexp ~ Aodd[Q"q]
(16)
408 T.PAmIANABHAN
where the 'correction' Wcorr[Qc] to the classical action Ac[Qc! is defined by
(17)
This expression is exactly in the form of path integrals we have considered
earlier. The right hand side describes the 'vacuum to vacuum' amplitude
for a system characterised by q, in the presence of Qc' If Qc(t) is varying
adiabatically, then we can write
(18)
where Eo(Qc) is the ground state 'energy of the quantum fluctuations q in
the background of Qc.
If Lc has the form
L, = 1 Q2 _ V(Q)
2
then (18) corrects the potentia! V(Q) to the value
v'" = V(Q) + Eo(Q)
(19)
(20)
In this - quantum mechanical - case, Eo is just the ground state energy of
the harmonic oscillator
1
Eo(Q) =-n
2
82V 1/2
8Q2
so that the corrected "effective" potential is
1 82V 1/2
V",(Q) = V(Q) + 2 8Q2
(21)
(22)
The situation becomes more interestiIig when we apply these ideas to a
field. Consider a scalar field ~ governed by the action
(23) .
where V (cI:.) is a potential describing the self interaction of the scalar field
and D is the dimension of space we are working in. [For usual field theory
D = 3]. Let us suppose that we are interested in studying the effect of
quantum fluctuations around some classical solution ~ = ~ c . This solution
will be taken to be either constant or adiabatically varying so that we can
,
ignore its derivatives. In the spirit of the discussion in the previous section,
we will expand q. around q. c as
(24)
and retain terms upta quadratic order in .p. -The Lagrangian then becomes
where
,,,,' = V"(<I>,) (26)
The correction term Leorr represents a scalar with effective mass
m
2
(4lc). The effective Lagrangian and potential will be, therefore,
L = Lo - E,,(m'); V .. = V + E,,(m')
(27)
where Eo(m
2
) is the ground state energy of a scalar field theory with mass
m.
"'. = J dDk (k' ')!/'
.", 2 (21r)D + m
(28)
Since this expression is badly divergent. for D > 1, we need to consider
methods for making sense out of this expression. We will address ourselves
to this question of 'renormalisation' in the next section. Before that, we
will first cast this expression in a more manageable form.
It is convenient, as usual, to work in the Euclidean sector. In Lorentzian.
space, Loorr = -Eoi but the Leorc in Euclidean sector differ by a sign from
that in Lorentzian space. Sol in the Euclidean sector we need. to calculate
= + ""(m') = ! f dDk (k' + m')!/' = L (29)
corr .<.:I(} 2 (211")D - c
where k is a D-dimensional vector. We begin our manipulations by calcu-
lating
(31))

410 T.PADMANABHAN
We will now use a little trick to eliminate the s-1/2 factor. We introduce a
variable p and rewrite this factor as another integral
1 /+XI dp _!sp2
(211"S)I/2 = _CQ 2'11" e 2
(31)
Then we get
aLe = ! j dDk (JO ds /+00 dp e-!,,{k
2
+pl+m') (32)
8m
2
4 (21l")D 10 -00 211'
We can now combine the k and p integrations into a (D + 1) dimensional
integration over the vector q = (kiP)' Then
- D+1
= ! j d q tx! d _i(q2+m2)
4 (21!")D+1 10 se
_ ! food _!,m2 j d
D
+1q _j"q2
4 10 se (211-)D+1 e
11,(XI ds _1m2,
- - e 2
4 0
(33)
Integrating this expression with respect to m
2
I we get
L 1 (OO ds _1m2"
c = - 210 s(21fs)t(D+l) e 2
(34)
We have omitted an integration constant which is independent of m
2
As it
stands (34) is also divergent at s = OJ however, in this form the divergences
are easy to isolate and handle.
There is another way of deriving (34) which is more straightforward
(though it hides the physical meaning of LeII') and is quite useful. We begin
by noticing that the effect of quantum fluctuations tP which. we are interested
in are contained in the Kernel
. .r:.=exp - j a.iEL
oorr
= jvq,exp - ,frq,Dq,=(detDj-l/2 (35)
". '.
where D is the Euclidean space operator
1 2
V = --(8.8" - m )
2
(36)
in which 8
a
fY'- denotes the (D + 1) dimensional D'Alembertian [containing
D-space and 1 Euclidean time]. We will now write this determinant as
det V = exp[Tr In VI (37)
so that the Kernel becomes
1
= exp -
2
Tr In D
- exp - ~ J diE < xl In Dlx >= exp - J diE L"" (38)
In arriving at the last expression, we have used some basis vectors Ix >
to evaluate the trace. We will now use the integral representation for the
logarithm,
1
=dS
loP = _e-
FIS
" . 0 s
(39)
to get
1 11= ds
- -2<xllnDlx>=-- -<xlexp-sDlx>
2 0 s
1 roo ds
= -2' 10 -;K(x,x; 8)
(40)
where the quantity
K(x,V;s) =< xle-Dlv > (41)
is just the Euclidean path integral Kernel for a quantum mechanical particle
with the hamiltonian D.
This result is of very general validity and should be emphasized. It shows
that if the Euclidean action coupling two systems has the form
(42)
,
where D4I is an operator depending on ~ , then the correction term in
effective lagrangian is given by
1 roo ds
Lcorr = -2' 10 sK(x,xjs)
(43)
where K(X,!li 8) represents the propagation Kernel for some fictitious quan
tum mechanical particle described by the hamiltonian
;. = b ~ (44)
In other words, we have reduced the problem involving a path integral over
fields to a problem involving quantum mechanical Kernel.
In this particular case, the hamiltonian is
1 1
h = D = -( -Baif' + m') = --
2 . 2
(45)

412 T.PADMANABHAN
We can evaluate this Kernel easily. The lagrangian corresponding to this
hamiltonian is
/
_ !((dT), IdXI') 1,
-+ - + - --m
2 cis ds 2
(46)
which represents a free particle in (D+ 1) dimensional space with a constant
background potential (m
2
/2). [Note that m
2
is treated as a constant in the
adiabatic limit] The Kernel K we are after is the one for a free particle:
(
1 )
21l"s
(47)
We thus get the expression for the effective lagrangian to be
( 1 )D+ll/' (1 )
LeIf = -- - .exp __ m
2
s
2 0 S 211"8 2
(48)
which agrees with the previous result. [Note the way in which i-factors
disappeared in the Kernel. In arriving at the last two expressions we have
proceeded as follows: The quantity < x/l exp -iTD1x > withD = -!alllP+
is a proper Schrodinger Kernel with D as hamiltonian and T as time.
Therefore '
(
1 ) !(D+ll .
K(TiX,X) = 21riT 2.
(49)
changing to iT = 8 leads to our expression above}.
4.3. RENORMALISATION OF THE EFFECTIVE LAGRANGIAN
The effective potentials we have derived are badly divergent and cannot be
interpreted as they stand. Physical meaning can be attached to these ex-
pressions only after the divergent terms are isolated and reabsorbed in some
sensible fashion. We will now discuss when and how this can be achieved.
To see what is let us evaluate the effective potential for the
quantum mechanical case, corresponding to D = 0, using the above for-
mula. We find that in D = 0 we have
1
Lcorr = -2' 211" 10 s3/2 exp(- 2
m2s
)
(50)
This integral is divergent at s = O! But since we are studying a quantum
mechanical problem we know that this divergence must be completely spu-
rious; and this is indeed the case. This divergence arises because we used
an integral representation for InD which is mildly divergent at the origin.
That integral reprE".sentation makes sense only when it is used to calculate
the difference between the logarithm of two operators.
This analysis also shows how to avoid the divergence in this case; we
actually only need the difference L{m
2
) - L(m = 0) which is finite. The
easiest way to evaluate this expression is to do the integral from some non-
zero lower limit A and then consider the limit of the expression as A tends
to zero. Performing one integration by parts, we can write
Leorr =
1 ( 1 ) II' { I} m' ( 1 ds "
-- - lim +_ _ e-
jm
$
2 211' A-tO A 1/2 2 211" 0 sl/2
m' ( 1 )1/' (1 )-1/' 1
= L(m =0) + - - -m' .jii = Lo+-m
2 2 2
(51)
where La is an infinite constant. Except for this unimportant - though
.infinite - constant, this is precisely the ground state energy of the harmonic
oscillator we derived earlier.
The situation becomes much more complicated in field theory which
corresponds to the D = 3 case. Our expression for Leff becOmes
L
- 1 ds _1m'. _ 1 I(P) - V.
corr-- -e 2 -- - eft"
81r
2
0 s3 811"2
(52)
with p = = V"(t/I). The divergence near the origin is now quadratic.
We will again evaluate this integral with the lower limit set to some small
value A and then consider the limit of this small cutoff parameter going to
zero. We only need to retain the leading telUlS which will dominate in this
limit. With this aim in our mind let us first calculate [(Pl. By repeated
partial integration, we get:
where
(54)
is the Euler's constant.
414 T.PADMANABHAN
We are supposed to take the limit of A going to zero in this expression.
We see that the last two terms involving p
2
ln p and p2 are finite. The term
in the square bracket diverges as A tends to zero; there are quadratic, linear
and logarithmically divergent terms. Of these, the quadratic term can be
dropped because it is just an infinite constantj this situation is analogous
to the D = 0 quantum mechanical case. But the next two terms A -lp and
p2In A cannot be dropped because they depend on p as well. We must think
up some way of interpreting this.
For an arbitrary V(tjJ), it is impossible to make any sense out of this
expression. This is a major problem in quantum field theory for which no
one has found a solution; it probably suggests that the entire formulation
of field theory is wrong. But no one has succeeded in providing a viable
alternative either.
In such a pathetic situation, we are forced to confine our attention to
those V in which something can be salvaged. This procedure, called
"renormalisation" proceeds as follows:
Let us suppose that V = A2,'" >w) where Ai are a set of con-
stants determining the form of the function. The quantities A -lp and p
2
ln A
will also be some functions of and the constants .\;;. Now suppose we are
lucky enough to have potential for which the quantity (A -1 p+p
2
lnA) has
the same form as our original with a different set of constants Then
we can absorb the divergent terms into these constants and reinterpret the
theory.
Let us see when this will work. It is obvious that non-polynomial form
of V(IjI) will not satisfy our criteria. So let us assume that V(IjI) is an n-th
degree polynomial in with n coefficients .\1 .\2'" >'n.
n
V() = >., + >.,' + ... >'nn = L, >..' (55)
k=l
Then p = (V"/2) will be a polynomial of (n-2)th degree and p'will be
polynomial with 2(n-2)th degree. If this expression is not to have terms
originally not present then we must have
2(n - 2) :s n (56)
Thus if is a polynomial of quartic degree or less we can reinterpret
our theory.
As an example, let us consider the case in which V(IjI) has a quartic and
quadratic terms. That is
p = = + >. ')
2 2 2
(57)
In I, we will drop the first term and rewrite the logarithmic term as
1 2 1 2 1 2 1 2 1 2 (P)
--'"(p - -p InA - -p In p = --p - -p In -
2 2 2 2 2
(58)
where IJ is an arbitrary finite constant introduced to keep the argument of
logarithms dimensionless. Then, we have
1
+ !:.",,) _ + >'",')'lnAp
2A2 2 2 4 2
_ \"')' + >''''').!.
2 4 2 2 2 I'
(59)
We will rearrange this expression dropping infinite constants and absorbing
divergent coefficients into In A, A -1 etc. This will allow us to write
1 =
>."" >", 1 "
--'I' - - + ->.m '" InAp-
SA 32 8
+ >. ",')' In(m' +
8 2 2 21'
- -",' -4>' InAI' +16'';1,(4);1') (60)
SA 8 32
Adding the correction (-1/811:
2
) to the -original V(r/J) we get the effective
potential:
(61)
where
(62)
and
1 (' >. ')' 1 (' >. ,)
'Vtinite = 6411'"2 m + 2 rp In 211 m + 2 .
(63)
Let us analyse this expression. OUf original potential had two constants
m and '\, which were the coefficients of 2 and 4. In Veff(4)) these are re-
placed by two other constants a and b which are functions of m, A, and an
416
T.PADMA;>;ABHAN
arbitrary finite parameter /1. In addition, it also contains divergent terms
involving A. We shall now describe a procedure by which Veff can be inter-
preted.
4.4. 'RUNNING' COUPLING CONSTANTS
We shall, to begin with, pretend corrections to rn
2
and A are 'small', [even
though the corrections are divergent when A -+ 0]. Then, retaining upto
first order accuracy in these corrections, we can replace and >'0 by m
2
and >. in Vfinite' Then we are left with
. ,
V .. = !m'.p' + + 1 (m' + ?:'</>,) In (m' + ), </>') (64)
2 4! 2 21' . 2
Next we notice that the physical values of m and), are actually determined
by comparing the theory with some observation; say, scattering cross sec-
tions measured at some energy E. Thus what is physically meaningful are
the operationally defined parameters \
d'v

(65)

d<P' </=0'
';=0
In the absence of any quantum corrections, we will have these relations
reducing to identities m
2
= m
2
and ). =,).. However, when there is a finite
correction to the classical potential we will get, instead, relations like

(66)
These equations can be inverted to expreSs m
2
and ). in terms of /1. Sub-
stituting these values back into eft', we can express it as
(67)
It may seem that we are still not out of woods because Veft' depends on the
arbitrary scale p.. However, this is really not the case. If I-' is changed to
some value J.t' then it is possible to change mphy and Aphy to some
and such that
(68)
In other words, mphy and Aphy are not constants but change when I-' is
changed.
We shall demonstrate the above results for the case with m
2
= 0, for
which
(69)
ASPECTS OF QUANTuM FIELD THEORY 417
It is clear that V::r(rJ> = 0) = 0 making mph zero. To define>. we cannot
use the previous definition because (d
4
V/dl) diverges at rP = O. However,
we cau define
d'V
A = d'
(70)
where M is an arbitrary scale. This gives
(
5)' 24 AM'
A = A + s;;:A + (16w)' A'ln 4/-1 .
(71)
allowing us to express (4/J./J..) in terms of M
.
AM' 25
In 4/-1 =-6
(72)
Substituting back into (69) we get
A A' ' 25
v.. = 4j' + (16w)'' In M' - 6
(73)
[The constant (25/6), of course, can be reabsorbed into In M'J. In this
expression, we have traded off the Jl dependence for the M -dependence
which is still arbitrary. If we now change M to Mi, then we can retain the
form of \'elf by changing>.. to >..', such that
3,\2 M'
~ ' = A+ 16w,ln M (74)
Under M -t M', >.. -t A' transformations, Veil is invariant: Veff().' 1 M') =
V{A, M) + O{A
3
).
The physical content of the theory is expressed by (74) which shows
how the coupling constant ,\ depends on the energy scale M at which it is
defined. An equivalent form of this relation is
M 8A = 3A'
8M 16w'
(75)
which could h a v ~ also been obtained from (74); note that Jl ex: M2. Equa-
tions like (75) are called 'renormalisation group' equations. The modern
interpretation of field theory draws heavily on this formalism.
4.5. EFFECTIVE ACTION IN ELECTRODYNAMICS
The formalism developed in the previous two sections can be applied to the
study of two extremely important problems: The calculation of effective ac-
tion for gravitational and electromagnetic fields due to their coupling with,
418 T.PADMANABHAN
say, a charged scalar field. We shall study the case of the electromagnetic
field in the remaining sections of this part and take up the gravitational
field in section 5.
The calculation of the effective action for the electromagnetic field al
lows us to determine the quantum correction to the classical Maxwell equa-
tions. Besides, it also illustrates several important conceptual issues in the
formalism and allows us to understand the renormalization procedures in
quantum electrodynamics. We shall provide a straightforward derivation of
the electromagnetic effective lagrangian from the ground state energy in this
section. A more powerful and formal technique based on the Schwinger's
proper time method will be presented in the section (4.6). The divergences
in the effective action and their renormalization is studied in section (4.7).
Consider a system described by the lagrangian density L(Ai. ) where
Ai(X) is a vector potential describing the electromagnetic field and is a
charged (complex) scalar field interacting with the electromagnetic field.
The full quantum theory is described by the Kernel
K = f VA,Vexp [i f LdtdX]
(76)
in which we have set 1i = 1 for convenience. The effective action AeK (and
the effective lagrangian Leff) for electrodyrnimics can be obtained by inte-
grating over the scalar field:
exp(iA",) - exp [i f dtdxL (A,)]
= f Vexp [i f dtdXL(A.,)]. (77)
Thus we need. to evaluate the path integral over in a given background
electromagnetic field.
As usual, this is an impossibly difficult task if Ai(X) is an arbitrary
background field. To make progless we will assume that Ai{X) varies slowly
with x so that we can write
(78)
where Fik are treated as constant. This colresponds to assuming that the
background potential describes a constant electromagnetic field Fik, or -
more precisely - the field tP varies much more rapidly compared to the
background electromagnetic field. Thus we will compute, in the adiabatic
approximation:
exp[iA (F)) = exp [i f dtdXL (F)]
ASPECTS OF QUi\NTU);! FIELD THEORY 419
.
= f Vexp [i f dtdxL [A, = -1/2F,k
xk
,lj (79)
We have seen earlier that, in the adiabatic limit we are considering, Leff is
the negative of the ground state energy of the system. Thus if we compute
the ground state energy Eo(F) of a scalar field ,p in a given background
Fik, then we can determine Lelf{F) = -Eo(F).
This task is particularly easy if the background field satisfies the condi-
tions E.B = a and B2 - E2 > O. In such a ease, the field can be expressed
as purely magnetic in some Lorentz frame. Let B = (O,B,O)j we choose the
gauge such that Ai = (0,0,0, -Bx). The Klein-Gordon equa.tion
can now be separated by taking
(t, x) = /(x) expi(kyy + k.z - wt).
where f(x) satisfies the equatJon
~ ; , + [w
2
- (qBx:'" k.)2lf = (m
2
+ k ~ ) /
This can be rewritten as
where
k.
e=x- i
qB
(80)
(81)
(82)
(83)
(84)
Equation (83) is that of a harmonic oscillator with mass (1/2) and fre-
quency 2(qB}. So, if f(x} has to be bounded for large x, the energy E must
be Qualltis9d:
I
En = 2(qB)(n + 2) = w
2
- (m
2
+ k ~ )
(85)
Therefore the allowed set of frequencies is
(86)
The ground state energy per mode is 2(wn/2) = Wn because the complex
scalar field has twice as many degrees of freedom as a real scalar field. The
total ground state energy is given by the sum over all modes ky and n. The
420 T.PADMANABHAN
weightage factor for the discrete sum over n, in a magnetic field is obtained
by the correspondence:
dk
x
dky --+ L (qB) dk,
211" 211' n 21(" 27r
Hence, the ground state energy is
(87)
(88)
This expression, as usual, is divergent. To separate out a finite part we will
proceed as follows: Consider the quantity
_ (2") 8'E
o
(2") 8'L,ff
I = - qB 8(m')' = qB 8(m')"
which can be evaluated in the following manner:
I =
(89)
The Left" can be determined by integrating the expression twice with respect
to m
2
We get
, ,
qB t:JO dTl e-r,m (00 d'1 e-!pll qB'f}
Leff = (47r)2 10 ;j2 sinhqBf] = 10 (411')2' 11
3
'sinhqBl1
(91)
This - and the subsequent expressions - has a divergence at the lower
limit of integrtation. This divergence can be removed by subtracting the
contribution with E = B = 0; we will ignore this problem right now and
will take it up later. The integration with respect to m
2
also produces a
term like (clm
2
+ C2) with two (divergent) integration constants Cl and C2'
We have not displayed this term herej this divergence is also connected with
the "renormalisation" of Leff and will be discussed later.
If the Leff has to be Lorentz and gauge invariant then it can only depend
on the quantities (E2 - B2) and E.B. We will define two constants a and
b by the relation
ab=E.B
(92)
Then Leff = Leff( a, b). In the case of pure magnetic field we are considering
a = 0 and b = B. Therefore, the Leff can be written in a manifestly invariant
way as:
(93)
Because this form is Lorentz invariant, it must be valid in any frame in
which E2 - B2 < 0 and E.B == O. In all such cases,
1,
00 d ~
L", =
o
e-
FJm2
qrr./B2 - E2
sinh
(94)
The Lef for a pure electric -field ca.n be determined from this expression
if we analytically continue the expression even for B2 < E2. We will find,
for B =0,
. j,= d,., e-'lf1I
2
qf1E
Leff = 3
o (4.-)' . ~ sin q ~ E
(95)
The same result can be obtained by noticing that a and b are invariant
under the transformation E --? iB, B -+ -iE. Therefore, Leff(a, b) must
also be invariant under these transformations: Leff{E,B) = Leff{iB, -iE}.
This allows us to get (95) from (9i).
We will now consider the general case with arbitrary E and B for which
a and b are not simultaneously zero. It is well known that by choosing our
Lorentz frame suitably, we can make E and B parallel, say along the y-
axis. We will describe this field [E = (O,E,O);B = (O,B,O)] in the gauge
Ai = [-Ey, 0, 0, - Bx]. The Klein-Gordon equation becomes
[Cia. - qA.)' - m']4>
-
(
_ a )' a' a' (. a )"
~ 8 t +qEy + ,8x2 + 8y2 - 'a; +qBx -m
4>=0
(96)
Separating the variables by assuming
4>(t,x) =J(x,y)exp-i(wt-k,z)
(97)
422
T.PADMANABHAN
we get
8' 8'
~ + +(w+qEy)'-(k,-qBx)' f=m'f (98)
8x2 8y2
which separates out into x and y modes. Writing
f(x,y) = g(x)Q(y) (99)
where g(x) satisfies the harmonic oscillator equation
rf'9 , 1
dx' - (k, - qBx) 9 = -2qB(n + 2)g
(100)
we get
rf'Q [ 1]
dy' + (w + qEy)'Q = m' + 2qB(n + 2) Q
(101)
Changing to the dimensionless variable
(102)
we obtain
(103)
To proceed further, we use a trick due to Landau. The above expression
shows that, the only dimensionless combination which occurs in the pres-
ence of electric field is T = (qE)-l(m' + qB(2n + 1)). Thus, purely from
dimensional considerations, we expect the ground state energy to have the
form
~
Eo = 2:(2qB)G(T)
(104)
n=O
where G is a function to be determined. Introducing the Laplace transform
F of G, by the relation
(105)
we can write
Summing the geometric series, we obtain
Leff :=; 2(qB)(qE) 10
00
dsF(qEs)e-
sm2
e-
qBs
"1-
_ 2( B)( E) roo ds F(qEs)e-
m
'
q q 10 eQBs _ e-qBs
= (qB)(qE) roo ds e-m"i
1
0
8m q s
423
We now determine F by using the fact that LefI must be invariant under
the transformation E -4 iB,.I! -4 -iE. This means that
L = ( B)( E) /,00 d -m'. F(iqBs)
'" q q se nh(O E )
o -51 1q S
Comparing the two expressions and using the uniqueness of the Laplace
transform with respect to m
2
, we get
F(qEs) F(iqBs)
-
sinhqBs sinh(iqEs)
(107)
Or, equivalently,
F(qEs) sinqEs = F(iqBs) sin(iqBs) (108)
Since each side depends only on either E or B alone, each side must be a
constant independent of E and B. Therefore
F(qEs) sinqEs .,. F(iqBs)sin(iqBs) = constant = A(s) (109)
o 0
gIVing
/,
00 e-m2" A(s)
L,. = (qB)(qE) ds 0 E nh B
o SlDq 881 q S
(110)
The A(s) can be determined by comparing this expression with, say,
(91) in the limit of E --t O. We have .
L,.(E = 0, B)
implying
B /,00 ds _m2$ A(s)
- q -e 0 0 h B
o s sm q s
_ qB roo ds e-m2, 1
Jo (41lVs2 sinhqBs
1
A(s) = (4n)'s
(Ill)
(112)
424 T.PADMANABHAN
Thus we arrive at the final answer
roo ds e-
m
'. ( qEs ) ( qBs )
LefT = 10 (411-)2 s3 sinqEs sinhqBs
(113)
In the situation we are considering E and B are parallel making a
2
- b
2
=
E2 - B2 and ab = E.B. = EB. Therefore E = a and B = h. Thus our
result can be written in a manifestly invariant form as
L ( b) _ roo ds e-
m
'. ( qas ) ( qbS) (114)
elf a, - 10 (4:n)2 83 sinqas sinhqbs
This result will be now valid in any gauge or frames with a and b determined
in terms of (E2 - B2) and (E.B).
The integral, as it stands, 1s ill-defined for two different reasons. (i) The
sine function has poles along the path of integration at qas = mrj n =
1,2, ... (ii) The integral diverges at s = O. The second problem is related
to renormalisation and will be taken up in the next section while the first
problem can be tackled in the following way:
The integral is evaluated by going around each of the poles by a small
semicircle in the upper half plane. This choice of upper hall plane is sug-
gested by the general principle in field theory that m
2
should be treated as
the limit of (m
2
- if). In (104), this is equivalent to treating qE as limit
(qE+if), changing sinqas to sin(qa+i)s. This makes the contour go above
the poles. Equivalently, we can rotate the contour of integration in Leff to
the imaginary axis and express it in the alternative form:
roo <is e-i(m'-i,j. ( qaa ) ( qbs )
Leff = - 10 (41r)2 s3 sinhqas sinqbs
(115)
This expression is sometimes easier to handle; it should be supplemented
by the rule that poles along the real axis should be ignored by going below
the axis.
The occurrence of the poles along the real axis and our if -prescription
has the following important consequence: It shows that Leff has an imagi-
nary part if a is nOll-zero. From (115) we get
(
qaS)(QbS)
sinh qas sin qbs
(116)
and
(
qas)(qbs)
sinh qas sin qbs
(117)
ASPECTS OF QUANT\i,l FIELD THEORY 425
The expression in (llG) can be evaluated by standard contour integration
techniques. However, we can also calculate it from (114) directly; this cal-
culation will explicitly show the origin ofIm (Lell) In (114),
L (E) = (00 ds e-
m
', ( qa ) ( qbS) (118)
eft' )0 (411'")2 8
2
sinqas sinhqbs
the poles at s = Sn = (mr I qa) are to be avoided by going around small
semicircles of radius t: in upper half plane. The n-th pole contributes to this
semicircle the quantity
'fO (fe
i9
idB) --m2s.. qa (QbSn)
e . cos(mr).Eei8 sinhqbs
n
8=rr .
'( 1)"+1 (qa)2 1 m
2
"
- I - "161r
3
n
2
exp - qa n (
qbs" )
sinhqbs
n
(119)
So the total contribution to 1m Leff is:
1m
_ )"+1 1 (qa)2 1 m
2
"
Letr - -1 '2 (2'11-)3 n2 exp - qa n
(
qbsn )
. sinhqbs
n
(120)
It is now clear that (ImLefr) arises because of non-zero Q , i.e., whenever
(i) there is an electric field in the direction of magnetic field or (ii) if E is
perpendicular to B, but E2 > B2. (In this case, we can go to a frame in
which the field is purely electric). For a purely electric field, the imaginary
part is
00 1 (qE)2 (_1)"+1 "m
2
ImLeft" = :; '2 (21r)3 n2 exp - qE n
(121)
Note that this expression is non-analytic in qj perturbation in powers of q
will not produce this result.
4.6, EFFECTIVE LAGRANGIAN FROM PATH INTEGRAL
The above analysis relied heavily on the facts that: (i) the energy levels in
a magnetic field are well known and (ii) the gauge and lorentz invariance
of the theory puts severe restrictions of the form of Left". This method,
therefore, is of only very limited validity. A more formal way of deriving
this result will be to use the proper time representation for Left" discussed
in section (4.2). Since this gives a general formalism for handling arbitrary
time dependence of the electric field, we will discuss this method next.

i -" /
426 T.PADMANABHAN "':;
This method can produce both the Green's function and the effective
lagrangian in a single stroke. The results quoted here will also be relevant
for comparing the quantum theory in an arbitrary, time dependent, elec-
tric field background with quantum theory in an expanding universe. The
central quantity in this description is the Kernel:
(122)
We saw in section (3.2) that the effective lagrangian Leff and the propagator
G(x',x) can be calculated from this Kernel by the relations
1
00 ds
Leff=-i -K(x,x;s)
o S
(123)
and
(124)
In the context of a scalar field interacting with an electromagnetic field, we
can write the kernel in the form
(125)
with the 'Hamiltonian'
h = !(ia _ qA)' _ m' + i,
2 2
(126)
We will first consider an electric field along z-axis, which has an arbitrary
time dependence; i.e., E = E(t) z,B = O. The gauge is chosen such that
A" = (0; 0, 0, A(t))[so that Av = (0; 0, 0, -A(t)); E(t) = -A'(tll. Using the
translational invariance along the spatial coordinates, we can write
I
d3p 0 is[., , ( ))'
<x lexp"2(.a,) -Pl-(P.-qAt -
m
2
+ iElIyo >
d' . a'
I
p _!!.(P2 +m2_i) 1 os [
- (21f")3e 2 J. <x ex
P
"2 -at2 -
(P. - qA(t))'JlyO >
- I +m'-i,))
(127)
where get, t'; s) is the propagator for the one-dimensional quantum mechan-
ical problem with the Hamiltonian
. 1 a' 1
H = - 2 at' - 2 (P. - qA(t))'
(128)
ASPECTS OF QUA);TDI FIELD THEORY 427
Let us now apply this formalism for the case of a uniform electric field
for which the potential is A = - Et. Then
la
2
1(p , 18
2
I",
H=-2at2 - 2 ,+qEt) =-28
p
2 - 2
q
E p
(129)
where p = t + (pz/qE). This is a Harmonic oscillator with mass m = 1 and
imaginary frequency (iqE) ("inverted oscillator"). Since the path integral
Kernel for this problem is well-known we can immediately write down the
coincidence limit for the propagator:
g(t,t;s) =
[
qE . ]'/2 qE2(coshqEs -1) ( P')'
exp -- t+-
{21ti sinh qEs) 2i sinhqEs qE
[
qE ]'/2 . qE ( p,)2
- 2nh E expo . h E (coshqEs -1). t +-E
1flSI q S SID q S q
(130)
Doing the PZIPlI and w we are left with
!.t(2"}
K _ qE (1) e-' m -" _ 1 (qEIs) _i(m'_i,).
- (211")' 2is . sinh(qEsI2) - (2,,)'i3 sinh(qEsI2)"e '
Giving
L
. (IX! ds qE e-
i
j(m
2
-i)
- -, 10 -;-. (211")'.(2is)" sinh(qEs/2)
= fOCI ds qE e-i(m2-it:)$
410 (211")' S2 sinhqEs
(131)
_ (QQ ds qE _i{m2_i}.s (132)
- Jo 41r
2
' s2' sinhqEs
e
.
In this approach it is clear that the imaginary part arises because of
the imaginary frequency (inverted nature) of the harmonic oscillator. This
point is brought out more vividly by the corresponding calculation for the
constant magnetic field. Magnetic fields, in general, give bounded Hamil-
tonians. For example, consider the case with All. = (O;A(z),O,O) giving
B. = (-Mlaz). Then,
K =
(133)
428 T.PADMANABHAN
where the effective Hamiltonian will be now
1 8 1 ,
H = -28z' + 2(Px - qA(z))
(134)
which has a potential bounded from below. Let us apply this equation to a
uniform magnetic field; A = -Bz. Then
where p = Z+ (pxlaB). This is a Harmonic oscillator with mass m = 1 and
real frequency (qB). Therefore
q(z,z;s} =
[
qB ]1/2 qB2(cosqBs-l) ( px)'
exp -- z+-
(27risinqBs) 2i sinqBs qB
[
qB ]'/'
. . exp
21l''lsmqBs
qB ( P)'
i. B (cosqBs-l). z+ B
X
smq s q
(136)
Doing the Px,Py and w integrations, we are left with
K _ qB (1) e-i,(m'-i,) _ 1 (qB/s) -i'(m'-i,) (137)
- 2is . sin(qBs/2} - sin(qBs/2j'e '
Giving
As it stands, the integrand has poles due to the sin(qBs/2) in the denom-
inator. However, notice that the proper definition of harmonic oscillator
path integral involves the prescription: w = li11lt:-+o (w - if). Therefore in
the Kernel the factor sinqBs should be interpreted as the limit of the ex-
pression sinqBs(l - if). So the poles are actually at

Sn = qB (1 + i,)
(139)
We can now transform the integral to one along the imaginary axis. Because
of the exp -is(m
2
/2) factor the contour should be closed in the lower half
plane. Then we get
m'
[00 dy ie--'-Y (qB/2)
10 y2 sinh ( q ~ y ) . (21r')2i
m'
[00 dy e--'-Y I (qB) (140)
10 y2 sinh ( q ~ y ) . (27r)2 2
This expression, which is the same as (111), is real showing that the con
stant magnetic field does not create particles. We shall now discuss the
renormalisation of Leff.
4.7. RENORMALISATION OF THE EFFECTIVE ACTION
We have seen earlier that the real and imaginary parts of the effective
lagrangian lead to different class of phenomenon. Since the Kernel is
K<O<ai - expi[Lo(F) + L .. (F)]
exp i[Lo(F) + Re L,.(F)] exp( -lmL,.)
< 0, +0010, -00 > (141)
we may interpret Re (LetT) as a correction to the original lagrangian for the
electromagnetic field
Lo(F) ~ ~ ( E 2 - B2)
8"
(142)
The (Im( Leff is related to the probability for the system to make tran-
sitions from ground state to the excited state. In this particular case the
excited state will be the one with the quanta of the scalar field present. We
may, therefore, interpret, 21m (Leff) as the probability per unit volume per
unit time for production of scalar particles. In this section, we shall discuss
the effects due to Re(Leff).
The first point to note about Re (Leff) is that it is divergent near s = o.
In fact, Re (Leff) is divergent even when E = B = o. This divergence - in
accordance with the discussion we had before - must be spurious and can
be removed by simply subtracting out the value for E = B = O. Thus we
modify (117) to
q
2
abs
2
=:::c';:cc:-:: - 1
sin qbs sinh qas
(143)
430 T.PADMANABHAN
Since the subtracted term is a constant independent of E and B, the equa-
tions of motion are unaffected. The expression R is still logarithmically
divergent near s = 0, since the quantity in the square brackets behaves
as - b
2
)] near s = O. But notice that this divergent term is
proportional to {a
2
- b
2
} = E2 - B2, -which is the original- uncorrected-
lagrangian. This opens up the possibility-that we can reabsorb the diver-
gence by redefining the field strengths, charges, etc. This can, be done as
follows: Let us first write
L"tai = Lo + L,. = (Lo + L,) + (L,. - L.) (144)
where
1 !.oods ., [1 " ,]
- -cosm s --(qs) (a -b)
(4,,)' 0 S3 6
q' roo ds cosm's.(a' _ b')" _ b') = _ B')
6(411")210 s 811" 811"
(145)
with Z being a formally divergent quantity. With this trick, we can separate
out the and divergent quantities in Ltota! and write
Ld;v = Lo + L, = "!:"'(E' - B') + - B') = -8
1
(1+ Z)(E' - B2)
8" 8" " (146)
and
Lfinite = LetJ - Lc
1 ,
- - (4,,)21
0
-;> cosm 8
x
q2 s
2
ab 1 2 2 2 2)
( b) 'nh( ) -1+ -6
q
8 (a - b
8m qs 81 qsa
(147)
The quantity Lfinite is perfectly welldefined. and finite. [The leading term
coming from the square bracket, near 8 = 0 is proportional to s3 and hence
Lfinite is finite near s = 0.1 So all the divergences are in the first term
(Lo + Lc) = (1 + Z)Lo. We shall now redefine all our field strengths and
charges by the rule
- (1 + Z)l/'E;
(1 + Z)l/'B;
_ (1 + Z)-l/'q
(148)
This is, of course, same as scaling a and b by (1 +Z)1/2 leaving (qphyE
phy
) =
qE invariant. Since only the products qa,qb appear in Lfillite, it can also be
expressed in terms of (qphyEphy). Thus it is possible to redefine the variables
in our theory, thereby absorbing the divergent quantities. The remaining
expression Lfinite is well-defined and possesses a Taylor expansion in q2.
Using this expansion, one can calculate corrections to the electromagnetic
lagrangian in an order-by-order manner.
5. Quantum Theory in Curved Space - -Selected Examples
In this last section, we shalf consider the applications of the formalism de-
rived so' far to SOme importa1.lt research problems. All these problems deal
with the special effects which arise when one studies quantum field theory
in a non-trivial background or with non-trivial boundary conditions. In sec-
tion (5.1), we shall consider a phenomenon called the Casimjr effect which
illustrates the effect of boundary conditions on the concept of vacuum state.
Section (5.2) discusses the by now famous examples of the thermal nature of
the vacuum fluctuations as seen by an accelerated observer. Sections (5.3)
and (5.4) study pair creation in a time dependent background and finally
section (5.5) provides a unified description of the thermal effects seen in
black hole space, deSitter spacetime and Rindler frames. The discussion in
this part is, quite naturally, somewhat more advanced than that in the pre-
vious sections. However, it is still seU contained and only uses techniques
developed so far.
5.1. CASIMIR EFFECT
Consider two plane, parallel, conducting- plates located with a separation L
in between. It is experimentally observed that they attract each other with
a force given by

(1)
It is possible to derive this result in several equivalent ways. For our pur-
pose, it is best to consider it as arising from the fact that the insertion
of the capacitor plates modifies the vacuum functional. The action for the
electromagnetic field can be decomposed into harmonic oscillators by using
some suitable set of mode functions. In the flat space-time with R4 topol-
ogy it is usual to take these mode functions to be plane waves of the form
exp{ikx). This will lead to a vacuum functional of the usual form. Sup-
pose we now introduce two conducting plates in the space-time. We can no
432 T.PADMANABHAN
longer use the mode functions exp(ikx} because they do not vanish on the
location of the conductors. Instead we use some other suitable set of func-
tions (like, say, sin(kx) etc.) ensuring the vanishing of mode functions on
the plates. The vacuum functional now calculated will be quite different. In
other words, the vacuum state depends on the choice of mode functions-
or, equivalently on the choice of the 'harmonic oscillators'. These choices,
in turn, are dictated by the boundary conditions of the problem.
The mathematical description of this phenomenon is quite straightfor-
ward. Since, in the later sections, we will be using the example of a scalar
field in (1 + 1) dimensions, we will briefly illustrate the above phenomenon
in this context. The action for a scalar field in (1 + 1) dimensions will be
(2)
If we expand the field in terms of the plane waves exp( ikx) then the action
reduces to that for a sum of harmonic oscillatorsj
(3)
where
)
f
dK ) iKX
</>(X,T = (21r)q,(T e .
(4)
The ground state wavefunctional can be expressed in terms of the field in
the form
w [</>(X), 0] - N exp ( - ~ f ~ ~ ) IKllq,I')
- N exp [- f dXdYV</>(X)V</>(y)g(X, Y)] (5)
where g is the green's function,
g(X Y) = ~ f dK expiK(X - Y)
, 2 (21r) IKI
(6)
If there are two conductors at x = 0 and at x = L, then we will expand
the field in terms of the mode functions sin(mrx/L) where n is an integer.
The wave functional can again be calculated in terms of the field. It can be
expressed in exactly the form as (5) but with a different Green's function.
We find:
(7)
Since the vacuum functionals are different, so are the fluctuation pat
terns and the vacuum expectation values of operators. In particular, calcu-
lations show that the (T
ik
) calculated from (6) and from (7) are different.
From (6) we get the result
E = LdK
1K1
= dn (mr)
2 210 L
(8)
while from (6) we get
E' = f:
2 fI.=l L
(9)
.
These two quantities differ by a finite amount. If we decide to regularize
the i.n the presence of the plates-:-given by (g)-by merely subtracting
the value in the absence of the plates, then we get the finite remainder:
= - (fr;)
(10)

(To arrive at the above expression we have used the following simple reg-
ularization procedure. In defining the difference between (8) and (9) we
have used the limit
(11)
which can be derived by straightforward calculation). Similar calculation
can be performed. for electromagnetic field in (3 + 1) dimensions. In that
case the corresponding result is:
(12)
This finite remainder can exactly account for the observed Casimir effect.
In other words, it seems very reasonable at least in this context-to com-
pute (1ik) using two different vacuum functionals made of different mode
functions and subtract one from the other to obtain a finite remainder.
It is also important to realize that this finite residual quantity must
contribute to the gravitational attraction. To see this, consider the follow-
ing thought experiment: Two parallel capacitor plates A and B are kept at
a distance L apart. A mass M is attached to each plate through the pul-
ley arrangement. As usual we will assume the plates to be weightless etc.
(These details are given only for the sake of visualizationj our argument is
independent of these details.) The plates were kept in the position by exter-
nal forces for t < 0 and were let of at t = O. Because of the Casimir effect,
434 T.PADMANABHAN
the plates will feel mutual attractive force and will start moving towards
each other. At a short time later, the masses would have acquired some
kinetic energy. This kinetic energy comes from the change in the Casimir
energy between the plates. (The Ca simir energy depends on the plate sep-
aration which changes when the plates come closer.) If (OITikIO}c denotes
the Casimir energy of the vacuum and tft is the stress-energy of the masses,
then we have, from energy conservation:
(13)
However, we know that,
(14)
since the masses are acted upon by external forces.
Now consider a region X far away from our Casimir plates. The grav-
itational field due to our setup is measured at X. If the Casimir energy
(OI'likIO)c does not contribute to gravity, then the only source of gravity in
our arrangement is M. Then Einstein's equations will read as:
1
Gik = Rv. - -Ih,R =
2
(15)
This is blatantly wrong because G
tk
has zero divergence while tik(M) does
not. We are forced to conclude that (OITikIO)c' does contribute to gravity
and that (15) must be modified to
(16)
It therefore follows that, in the absence of M, that is, if we take the limit
of (M -t 0), (OITik 10)c should lead to observable gravitational effects.
The above example is extremely important and should be taken seri-
ously. It shows that the finite parts of the expectation value Tik, computed
by subtracting one divergent quantity from another, must act as a source
of gravity. If it does not, we run into contradictions and even perpetual
motion. In this particular case the two divergent values for (11k) are ob-
tained by using two different sets of mode functions in the flat space-time.
We shall now consider similar effects which arise in the noninertial frames.
5.2. VACUUM FLUCTUATIONS IN THE RlNDLER FRAME
To illustrate the concepts involved, we will use a scalar field in a (1 + 1)
dimensional spacetime and indicate the generalizations to fuur dimensions
whenever needed. Consider the two-dimensional Minkowski space-time with
the line element:
(17)
The complete manifold is covered by the range (-00 < T < +00, -00 <
X < +00). We now introduce two sets of coordinate patches, (x, t) and
(x',n on the regions X > T(" Right ",R) and -X > T ("Left ", L) by
the transformations:
(in R) (18)
x = _g-le
gxl
cosh gt'; T = _g-le
9X
' sinh gt' (in L). (19)
All the coordinates (x, t), (x', tf) vary from (-00, +00). The metric (17) in
terms of (x, t) [or(x\ t')] has the form
(20)
It can be shown that (x, t) corresponds to the proper coordinate system
of a uniformly accelerated observer in the region R, t being the proper
time of his clocks. For an observer, completely confined to R, the surfaces
t = constant provide a set of spacelike hypersurfaces. In the region R, one
can introduce two Killing vector fields:
(21)
and
(22)
The ek leads to translation in the t-coordinate while eM corresponds to
translations in T-coordinate; the latter has components (1, 0) in the inertial
frame.
Consider a scalar field (X, T) = '(x', t'), described by the generally
covariant action
A = ~ J d'xV-gg
ik8
i4>iJk'"
~ r= <fl' r= dX ( 8", ) , _ (8"')'
2 J-
oo
Loo f)T ax
(23)
Tn the region R, the action can be written in terms of x and t in the same
form. (This is because in 2-dimension, Ag
ik
= flik for (20:
AR = - dt dx ~ - - . 11+= 1+= [(8"')' (8"')'
2 -00 _00 at ax
(24)
436 T.PADMANABHAN
We shall quantize this field in the Schrodinger picture by decomposing the
field into harmonic oscillator modes. Consider a field configuration leX)
on the T = 0 hypersurface, expanded as,
(25)
The hypersurface T = 0 is covered completely by (t = 0, x) for X > 0 and
(t' = 0, x') for X < O. In Rand L the field can be decomposed as,
/
+=dk
(X > 0) = (x) = -2 ake;kx
-=
(26)
/
+=dk .,
(X < 0) = .p(x') = -2 bke'kx.
-=
(27)
Using the relations (25), (26) and (27), we can express qk in terms of ak
and bk- After some simple algebra, we get
/
+= . /+= dk
qK = _= .p(X)e-'KxdX = _= 2,; [a.t(k, -K) + bkf(k, +K)]
(28)
where the function f(k, K) is defined by
This function can be expressed in term of the gamma function of imaginary
argument; but we will not need its explicit form.
The vacuum state in the Minkowski frame is described by the functional
q,[.p(X),Oj=q,[(qK)] =
- (30)
Using (28) we can express this wave function in terms of aks and bi:s . A
detailed but straightforward calculation gives,
with

P(k) = cath 9 , Q(k) = esc 9 ; Wk = Ikl
(32)
An observer confined to R will have his observables made out of aks. Let
O(ak) be any such observable. The expectation value of" (at T = t = 0)
in the state 'l' is given by
where
(0) = Inda,/ndb,'P'(a"b,)O'P(a"b,)
, ,
In da,p(a"a,)O(a,) = Tr(pO)
,
a, ) = I n db, b,)'Pa" b,)
,
N exp [- 1+
00
dk Wk (( ak _ aAl coth (W' )
2 -00 211" 4 2T
(33)
+ (a, + aD' tanht';))] (34)
is a thermal density matrix corresponding to the temperature T = (9/211").
-
This is a well known result, showing the thermal nature of the vacuum state
(30) when expressed in terms of the Rindler mode functions. I
The two examples studied so far shows that the concept of vacuum
state in an external background is conceptually non-trivial. Indeed we have
emphasized in section 2 and 3 that the concept of particle (and also the
concept of vacuum state) is inextricably linked with the decomposition of
a quantum field into oscillators. In the Hat spacetimes with J(4 topology
and inertial coordinate system, there exists a very natural decomposition
of quantum fields as harmonic oscillators. This natural decomposition no
longer exists when we use non-inertial coordinate systems or non-standard
boundary conditions. The Casimir effect and Rindler effect discussed in the
last two sections exhibit this feature in a simpler manner. We shall see more
complicated examples of the same as we go along.
5.3. PAIR CREATION IN ELECTRlC FIELD AND EXPANDING UNIVERSE
There is a formal correspondence between pair creation in a time dependent
electric field and pair creation in an expanding Friedmann universe. This
can be seen as follows: Consider, for example, the action for a scalar field
<l>
A = - I [Uia' + m' + !R] <l>
in the Friedmann spacetime with the line element
d,' = a'(t)(dt' - dr')
(35)
(36)
I
438 T.PADUANABHAN
[This action is conformally invariant in the limit of m going to zero. This
fact makes our analysis easy; however, the results are valid even for noo-
conformal coupling]. Writing q:. as (4)/a) and exploiting the conformal flat-
ness of the metric, we can reduce the action to the form
A = - f d'x [8,8'ftat + m'a'(t)]
(37)
To study the pair creation, we can again use the effective lagrangian method.
The Kernel we need' is
K(x,YiS) _ < xle-
i
.tH8;8'+m
2
a
2
(t)-iEJlv >
= f tFp < >
(Z,,)'
= f d'p e-,t(P'-")g(t t s) (38)
(211'")3 1 ,
where g is the propaga.tor for the quantum mechanical Hamiltonian
(39)
Comparing this expression with the conesponding one for the electric field
we can make the identification: m
2
a
2
(t) {::} (p;&' - qA(t))2. Thus there exists
one-to-one correspondence between time dependent electric fields and
expanding Friedmann universes as far as the quantisation of an external
scalar field is concerned.
AB an example, consider the case of constant electric field. The analogue
in cosmology will be a universe with the conformal factor:
a'(t) = + gEt)' >= a'(t + to)'; a = (40)
In the more 'familiar coordinate sytem with
,
ds' = d-? - a'(r)dr' (41)
this corresponds to the expansion law
aCT) = (2aT)1/2 ex: Tt (42)
This corresponds to a radiation-dominated universe. Similar correspon-
dences can be established in other cases which allows one to translate the
results in one physical situation to another.
ASPECTS OF QUANTU).! FIELD THEORY 439
5.4. QUANTUM THEORY IN A MILNE UNIVERSE
We sh<;1U now take up the analogy between gauge and coordinate invariance.
To do this, one requires some region of spacetime ma.nifold which can be
represented conveniently in two different co-ordinate systems: ODe system
in which the metric is static and another in which it depends only on time.
The simplest choice happens to be the upper quarter of the fiat spacetime.
In this 'top-quarter' of the Minkowski spacetime (i.e the region T > IXI
which we will call ('U
'
), the line element can he expressed in two different
ways:
ds
2
= d:['2 _ dX
2
_ dy2 _ dZ
2
_ dp' - g'p'dx' - dY' - dZ'
- e,g'(dt' - dx') _ dY' - dZ' (43)
by the transformation gX = egtsinhgx and gT = egtcoshgxj gp = e
9t

The intermediate form of the transformation shows that the metric belongs
to the class of anisotropically expanding cosmological solutions. Because of
this similarity, we will call this co-ordinate system the I Milne Universe.'
Since the 'static gauge' now is just the inertial co-ordinates, we only have to
work out the quantum theory in the Milne universe. We want to study the
evolution of a quantum field along the hypersurfaces defined by constant-t
(in the Milne coordinates) and compare it with the conventional Minkowski
quantisation. Since the metric in Milne co-ordinates depend only on t, the
Klein-Gordon equation
(44)
can be separated as
4>(x,t) = E {ad.(t)e'k.x + h.c} = E(4)i+) + h.c) (45)
x k
This equa.tion has the two linearly independent solutions which may be
taken to be H ~ ) (p) and Hg)(p) where H,,(z) is the Hankel function. We
write
f(t)
(1)( ) (')
-
c1H
jp
P + C2 Hjp (p)
-
-
Cl (J" + iN,,) + C2(J .... - iN,,)
-
(Cl + c,)Jv(p) + i(Cl - c,)Nv(p)
-
-
b
1
J
v
(p) + ib,Nv(p)
(46)
-
440 T.PADMANABHAN
where 1I = ip and J", and N
v
are the Bessel functions (see eg., Gradsbteyn
and Ryzhik, 1965). We are interested in the limits t -+ oo. From the
properties of the Bessel functions, it is easy to see that
. _{. pV I . p-v I }
,.!;moo J.(t) = (b1 + .b, cot v,,-) 2V r(1 + v) -.b, esc v,,-. 2-v r(1 _ v)
(47)
Since p1I ~ exp(gvt) = exp(ilkxlt) (in U) and we want f to go as
exp( -iwt) for the positive frequency mode, only the second term is admis-
sible. Therefore the positive frequency modes in the infinite past are the
ones obtained by the condition
l>J. = -if>:! ~ S ~ p 1 r = -ib:! ~ ~ h p 1 r = -b:!coth1l"p (48)
smtp7r tS1D(J1f
We should take J.(t) to be
J_
II
= - sinhp1r.J-vo
The value of b:! is fixed by the normalisation condition:
J.(t) = ill, ac v,,-.
b,
iU"i - Ii') = (2,,-)-'.
Straightforward calculation gives,
(49)
(50)
Ib,l' = (2..)' ,,-(sinhp,,-) = (2..)' (!!...) (sinhp1r) (51)
2'11" 29 211" 29
so that
(52)
We know now that, the solution which behaves as e-
iwt
near t -+ -00 in
U, is
I(t)
b,
. h Lv(p)
sm p1l'
(
1 )'/' ("- ) 1/' J_v(p)
- 211" 2g sinhl/2 p7r'
(53)
It is clear from the asymptotic form of the equation for f that we will not
get exp(iwt) in the infinite future. Therefore the positive frequency mode
has to be identified by the WKB analysis, as in the case of electromagnetic
field. This analysis shows that the proper mode is the one which behaves as
, .

exp( -ip) in the infinite future. Since behaves as exp( -ip) near large
p we can set Cl = 0 and take the solution to be
(54)
To normalise this solution, we will again use the condition
W = i(g'g - gg') = (211T' (55)
This gives
(56)
We can now express the positive frequency solution of the infinite past
in terms of the positive and negative frequency solutions of the infinite
future and identify the Bogoliubov coefficients. Using the identities

e
iv7t
H(l)
J
v
-

e"'7t J_
II v
sm 1['1J sm WI!

e -ill'" H(2)
Jv +

e-
i
"'''' J_
II
(57)
-
-
v
sin 1fV 8InV1f
and
(58)
it is easy to show that
f(t) = 2.- ( . 1 )1/2 {e-'P/2g'(t) + eWP/2
g
(t)} (59)
vI2 smh7rp
This corresponds to the Bogoliubov coefficients
a = 1 ( 1 )1/2 e1rp/ 2 .
v'2 sinh "P ,
1 ( 1 ) 1/2
f3 = e-7r1'/2
v'2 sinh1fp
(60)
We see that
2 2 1 (e1l'P - e-
1fP
)
a-(3=- =1
2 sinh 1ffJ
(61)
as it should. The number density of created particles is
(32 1 2e-'
P
1 1 (62)
=-2e
7rP
-e-
7tP
(e
2
'11"P-l)
which corresponds to a thermal spectrum of particles in the longitudinal
momentum with the temperature (g/27r).
442 T.PADMANABHAN
Similar results can be obta.ined in a different context which is well-
known in literature. Investigations in the study of particle production by
expanding Friedmann universes have shown that, in a spatially flat model
with the expansion law a(t) ex: t a thermal spectrum of particles is produced
at late times. Interestingly enough, the analysis is valid in 2-dimensions
as well in which case the spacetime is just the T - X sector of the Milne
universe. Our analysis shows that the other two dimensions merely go for
a ride.
Even though (62) corresponds to a temperature of (gj2tr), the result
is very different from the standard result obtained in Rindier frame in
section (5.2) for two reasons: (i) We are working in the upper and lower
quarters, while the Rindler-' Co-ordinates exist only in the right and left
quarters. This makes the entire situation quite different. (ii) There is no
'particle creation' in the Rindler co-ordinates (r',p, y, z). The Rindler mode
functions behave as exp iwr for all times. The conventional result only
says that these mode functions are connected to the Minkowski modes by
a Bogoliubov transformation with offMdiagonal term which leads to a result
similar to that in (62) . In contrast,. we are now working with a non-static
backgroundj the positive frequency mode in the infinite past does get mixed
up with positive and negative frequency modes of the infinite future.
5.5. SPACETIME MANIFOLD IN SINGULAR GAUGES
We shall next consider the gravitational analogue of the effects discussed
in section (4.7). It turns out that the 'thermal' effects in certain spacetime
provide this analogy. Consider a patch of spacetime, which, in suitable
coordinate system, has the line element,
or
ds' = +B(r)dt' - B-1(r)dr' _ r'(dO' + sin'Odq,')
= +B(r)dt' - B-1(r)dr' - dL' (63)
ds' = B(x)dt' - B-1(x)dx' _ dy' - dz'
= B(x)dt' - B-
1
(x)dx' - dL'

(64)
Co-ordinate systems of the form (63) can be introduced in parts of
Schwarzschild and de Sitter spacetimes while the choice [B(x) = 1 + 2gx]
in (64) represents a uniformly accelerated frame (Rindler frame) in flat
spacetime. We will be concerned basically with the structure of the metric
in the r - t or x - t plane. Since this structure is essentially the same in both
(63) and (64) we shall work throughout with (63); the results are extendible
to (64) in a straight forward manner.
ASPECTS OF Q U A N T U ~ 1 FIELD THEORY 443
The exponent of the Kernel K (x', Xi s) will now contain the integral
ik '2 -12
10
' 10'
A = 0 dT9ikX x = 0 dr [Bt - B x + .... ]
(65)
Quite obviously we will run into problems if B vanishes along the path
of integration. From the nature of our metric it is easy to see that the
surfaces on which B vanishes are null surfaces corresponding to infinite
redshift ('horizons').
To study the effect of horizons, let us proceed in the following manner:
Suppose that at some r = TO(> 0), B(r) vanishes, B'(r) finite and nonzero.
Then near r = TO. we can expand B(r) as,
B(r) - B'(ro)(r - ro) + O[(r - ro)')
- R(ro)(r - ro)
(66)
As long as the points 1 and 2 (between which the transition amplitude is
calculated) are in the same side of the horizon, [i.e., both are at r > TO or
both at r < roJ the integral in the action is well defined and real. But if
the points are located at two sides of the horizon then the hitegral does not
exist due to the divergence of B-l(r) at r = roo
Let us first review briefly the conventional derivation of thermal effects
using path integrals, say, in the context of Schwarzschild black hole. Given
the co-ordinate system of (63), in some region R, we first verify that there
is no physical singularity at the horizon. Having done that, we extend the
geodesics into the past and future and arrive at two further regions of the
manifold not originally covered by the co-ordinate system in (63). Let us
label these regions as F and P. It is now poSsible to show that the probability
for a particle with energy E to be lost from the region R (i.e., probability for
propagation from P to P') in related to the probability for a particle with
energy E to be gained by the region R, (i.e., probability for propagation
from P' to P)by the equation
P(loss) = P(gain) exp -PE
(67)
This is equivalent to assuming that the region R is bathed in radiation at
temperature (3-1.
The above result can be interpreted differently, so as to bring out the
connection with the case of electromagnetic field. This is most easily done
by considering the semiclassical approximation to the path integral propa-
gator, expressed in the saddle point approximation, as:
Q(X" t,; Xl, ttl = Q(2, I) = N expiA(2, I)
(68)
444 T.PADMANABHAN
where A is the action functional satisfying the classical Hamilton-Jacobi
equation. For a particle of mass m, moving in our spacetime the Hamilton-
Jacobi equation will be:
y"(o.A)(o,A) -m' = 0
(69)
The solution to this equation can be represented as
A = -Et + JO + A,(r) (70)
with,
J
T -, 1/2
A,(r) = drB-
1
(r) [E' - B(r)(m' + J'/r')]
(71)
The sign ambiguity of the square-root is related to the "outgoing"
[(oA/or) > 0) or "ingoing" [(oA/ar) < 0) nature of the particle.
Thus, in order to obtain the proba.bility amplitude in (68) for crossing
the horizon, (i.e.
t
when 1 and 2 are on two sides of the horizon), we have to
give some extra. prescription for evaluating the integral. Since the surface
B = 0 is null (just like x = t in the electrodynamic case) we may carry
out the calculation in the Euclidean space or - equivalently - use the if pre-
scription to specify the contour over which the integral has to be performed
around r = ro. The usual if prescription can be easily shown to imply that
we should take the contour for defining-the integral to be an infinitesimal -
semi-circle above the pole at r = ro. Thus, the contour is along the real line
from, say, rl (0 < rl < ro) to (ro - E) and from (ro + E) to, say, r2(r2 > ro).
From (ro - ) to (ro + E) we go along a semicircle of radius to in the upper
complex plane.
Consider an outgoing particle {(8AI8r) > 0] at r = Tl < TO. What is
the amplitude for it to cross the horizon? Clearly, the contribution to A in
the range (ruro - E) and (ro + E, r2) is real. Therefore,
l
To
+
f
A(outgoing) = - "_, (dr/B(r))[E'-B(m'+J'/r'))l/'+(realpart) (72)
[The minus sign corresponds to the initial condition that (BAier) > 0 at
T = Tl < roo For the sake of definiteness we have assumed R in (66) to be
positive, so that B < 0, at r < roo For the cases with R < 0, the answer
has to be modified by a sign change.] Evaluating the integral, in the limit
of (E -+ 0), we get
A(outgoing) - -[E/R(ro))(-i1r) + (realp.rt)
- [i1rE/R(ro)) + (real part) (73)
,
,

,
,
,
,
,

,
,
,
,
Now consider an ingoing particle [(8Aj8r) < OJ at r = r2 > roo The
corresponding action is,
A(ingoing) -1"--(dr/E(r.[E' - E(m' + J' /r')]' + (real part)
TO+f
+ (real part)
+ (real part) (74)
Taking the modulus to obtain the probability, we get,
P(outgoing) = (75)
and
P(ingoing) = N (76)
so that
prout) = exp[-4 .. E/R(ro)].P(in) (77)
This result shows that it is more likely for a particular region to gain parti-
cles than lose them. If one tries to do a consistent quantum field theory in
this region, one has to introduce source terms at the singular boundaries.
Further, the exponential dependence on the energy allows one to give a
interpretation to this result. In a system with temperature {3-1
then the absorption and emission probabilities are related by
P [emission] = exp(-f3E)P [absorption] (78)
Comparing (77) and (78) we identify the temperature of the horizon in
terms of R(ro}. Equation (72) is based on the assumption that R > O. [see
the comment after (72) ]. For R < 0 there will be a change of sign in this
equation. Incorporating both the cases, we can write the general formula
for horizon temperature to be
p-l = (79)
For the Schwarzschild black. hole,
E(r) = (I - 2M/r) '" (1/2M)(r - 2M) + O[(r - 2M)'] (80)
giving R = (2M)-1, and the temperature:
p-l = = (81)
[The following point is worth noting regarding the derivation of the thermal
effect in the case of a Schwarzschild black. hole: The regularisation procedure
which is adopted above is equivalent to replacing M by (M -if:), where M is
446 T.PADMANABHAN
the mass of the black hole. This is identical to the reguiarisation procedure
which would have been adopted in standard field theory if one is dealing
with particles of mass M. Probably this result has no deep significance,
but it certainly appears as an interesting coincidence.]. For the De Sitter
spacetime,
B(r) = (1- H'r') = 2H(H-
1
- r) = -2H(r - H-
1
) (82)
giving
{3-1 = = (83)
Similarly for a metric of the uniformly accelerated frame
B(x) = (1 + 2gx) = 2g[x + (2g)-11 (84)
and
{3-1 = (g/21f)
(85)
The formula can he used for more complicated metrics as well, and gives
the same results as obtained by more detailed methods.
The above analysis is not intended to be a derivation of the thermal
effects; rather, it is an interpretation of results derived by more rigorous
methods. This interpretation, however, has the advantage that it allows
one to obtain the thermal effects by invoking a simple prescription for
handling the integrals across the horizon and emphasizes the role played
by the singular gauge.
The Rindler frame discussed above is ususally considered to be part of
field-nee region, i. e., it represents flat spacetime in a curvilinear
system. Our earlier discussion on the pure gauge potential suggests that
this aspect needs to be looked at closely, especially if Euclidean continua--
tions are used to interpret the theory. In fact, it has been pointed out by
Christensen and Duff that the Euler characteristic of the Euclidean sec-
tor - obtained by analytically continuing in the Rindler time co-ordinate -
is different from that of standard (Euclidean) space. The difference arises
precisely due to the nature of the singularities along the light cone in the
Rindler gauge.

Aspects of Quantum Field Theory - Padmanabhan

Încărcat de

Informații document

Descriere originală:

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Aspects of Quantum Field Theory - Padmanabhan

Încărcat de

Drepturi de autor:

Formate disponibile

5.

S-ar putea să vă placă și