This action might not be possible to undo. Are you sure you want to continue?
75
FIELD THEORY
A Path Integral Approa,
Second Edition
ASHOK DAS
FIELD THEORY
A Path Integral Approach
Second Edition
World Scientific Lecture Notes in Physics
Published Vol. 54: Introduction to the Theory of Spin Glasses and Neural Networks V Dotsenko Vol. 55: Lectures in Particle Physics D Green Vol. 56: Chaos and Gauge Field Theory T S Biro, et al. Vol. 57: Foundations of Quantum Chromodynamics (2nd ed.): An Introduction to Perturbative Methods in Gauge Theories TMuta Vol. 58: Concepts in Solids, Lectures on the Theory of Solids P. W. Anderson and H. Bunke Vol. 59: Lattice Gauge Theories: An Introduction (2nd ed.) H J Ftothe Vol. 60: Massive Neutrinos in Physics and Astrophysics (2nd ed.) R N Mohapatra and P B Pal Vol. 61: Modern Differential Geometry for Physicists (2nd ed.) C J Isham Vol. 62: ITEP Lectures on Particle Physics and Field Theory (In 2 Volumes) M A Shifman Vol. 64: Fluctuations and Localization in Mesoscopic Electron Systems M Janssen Vol. 65: Universal Fluctuations: The Phenomenology of Hadronic Matter R Sorer and M Ploszajczak Vol. 66: Microcanonical Thermodynamics: Phase Transitions in "Small" Systems DHE Gross Vol. 67: Quantum Scaling in ManyBody Systems M A Continentino Vol. 69: Deparametrization and Path Integral Quantization of Cosmological Models C Simeone Vol. 70: Noise Sustained Patterns: Fluctuations & Nonlinearities Markus Loecher Vol. 71: The QCD Vacuum, Hadrons and Superdense Matter (2nd ed.) Edward V Shuryak Vol. 72: Massive Neutrinos in Physics and Astrophysics (3rd ed.) R Mohapatra and P B Pal Vol. 73: The Elementary Process of Bremsstrahlung W Nakel and E Haug Vol. 74: Lattice Gauge Theories: An Introduction (3rd ed.) HJRothe
World Scientific Lecture Notes in Physics  Vol. 75
FIELD THEORY
A Path Integral Approach
Second Edition
X V ^ i l U J x . JL/x\k3
University of Rochester, USA
i  p World Scientific
NEW JERSEY > LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TAIPEI • CHENNAI
Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library CataloguinginPublication Data A catalogue record for this book is available from the British Library.
FIELD THEORY: A PATH INTEGRAL APPROACH Second Edition Copyright © 2006 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN 9812568476 ISBN 9812568484 (pbk)
Printed in Singapore by B & JO Enterprise
To Lakshmi and Gouri
This course was aimed at second year graduate students who had already taken a one year course on nonrelativistic quantum mechanics but had not necessarily specialized into any area of physics and these lecture notes grew out of this course which I taught. it has become quite clear that the methods and techniques of field theory are widely applicable in many areas of physics. Even in the published form. In fact. In writing these lecture notes. do not bring out this feature of field theory. It is with this goal in mind. namely.Preface to the First Edition Traditionally. I have endeavored to keep as much of the detailed derivations of various results as I could — the idea being that a reader can then concentrate on the logical development of concepts without worrying about the technical details. that the Department of Physics and Astronomy at the University of Rochester introduced a new one semester course on field theory in Fall 1991. the lecture notes are identical to what was covered in the class. Most of the concepts were developed within the context of quantum mechanics — which the students were expected to be familiar with — and subsequently these concepts were applied to various branches of physics. field theory had its main thrust of development in high energy physics. I have added some references at the end of vii . Over the years. however. to make graduate students aware of the applicability of the field theoretic methods to various areas. which is how conventional field theory courses are taught. The canonical quantization methods. the conventional field theory courses are taught with a heavy emphasis on high energy physics. A path integral description of field theory is the appropriate setting for this. Consequently.
the many interesting topics of gauge theories are also not covered in these lectures. I would like to thank our chairman. There are many who were responsible for these lecture notes. There is so much literature that is available in this subject that it would have been impossible to include all of them. would have taken for granted. in computergenerating all the figures in the book. Judy Mack. The support of many colleagues was also vital for the completion of these lecture notes. Finally. for asking me to teach and design a syllabus for this course. They are only intended to be suggestive. Zhu Yang for straightening out many little details which were essential in presenting the material in a coherent and consistent way. It simply would have been impossible to do justice to these topics within a one semester course. The references are not meant to be complete and I apologize to many whose works I have not cited in the references. has done a superb job as far as the appearance of the book is concerned and I sincerely thank her. I am grateful to my students Paulo Bedaque and WenJui Huang as well as to Dr. Ashok Das. I am grateful to Ammani for being there. Rochester. . I would also like to thank Michael Begel for helping out in numerous ways. in particular. as always. otherwise.viii Field Theory: A Path Integral Approach every chapter. The students deserve the most credit for keeping all the derivations complete and raising many issues which I. Since this was developed as a course for general students. Paul Slattery.
Ashok Das. While these topics were not covered in the original course on path integrals. There are many people who have helped me to complete this edition of the book and I would like to thank. Boersma for all their assistance.Preface to the Second Edition This second edition of the book is an expanded version which contains a chapter on path integral quantization of gauge theories as well as a chapter on anomalies. In addition. chapter 6 (Supersymmetry) has been expanded to include a section on supersymmetric singular potentials. in particular. they are part of my lectures in other courses that I have taught at the University of Rochester and have been incorporated into this new edition at the request of colleagues from all over the world. Rochester. Arsen Melikyan. . Judy Mack. Dave Munson and J.
.
5 Particles and Fields Metric and Other Notations Functionals Review of Quantum Mechanics References vii ix 1 1 2 3 7 10 11 11 13 20 22 25 30 31 31 33 36 45 2.2 1. Introduction 1.3 1.4 Path Integral for the Harmonic Oscillator Method of Fourier Transform Matrix Method The Classical Action xi .4 2. Harmonic Oscillator 3.1 1.2 3.6 Basis States Operator Ordering The Classical Limit Equivalence with the Schrodinger Equation Free Particle References 3.1 2.5 2.1 3. Path Integrals and Quantum Mechanics 2.Contents Preface to the First Edition Preface to the Second Edition 1.3 3.2 2.3 2.4 1.
Generating Functional 4.2 Alternate Regularization References 6.4 5.5 5.2 7.5 4.1 4.5.6 Euclidean Rotation Time Ordered Correlation Functions Correlation Functions in Definite States Vacuum Functional Anharmonic Oscillator References 5.5 Supersymmetric Oscillator Supersymmetric Quantum Mechanics Shape Invariance Example Supersymmetry and Singular Potentials 6.5 WKB Approximation Saddle Point Method SemiClassical Methods in Path Integrals Double Well Potential References .xii Field Theory: A Path Integral Approach 3. Path Integrals for Fermions 5.1 7.3 4.6 7.3 5.3 6.4 6. Supersymmetry 6.1 5.2 6.5 References 51 53 53 59 61 64 71 73 75 75 78 83 86 91 95 97 97 102 105 110 Ill 115 117 118 121 121 127 130 134 142 4. SemiClassical Methods 7.4 7.6 Fermionic Oscillator Grassmann Variables Generating Functional Feynman Propagator The Fermion Determinant References 6.3 7.2 5.4 4.1 6.5.2 4.1 Regularized Superpotential 6.
Path Integral for the Double Well 8.3 9.5 8.3 Complex Scalar Field 11.2 NonAbelian Gauge Theory 12.1 10.1 9.6 Goldstone Theorem 11. Invariances and Their Consequences 11.1 Maxwell Theory 12.Contents xiii 8.4 Ward Identities 11.5 Spontaneous Symmetry Breaking 11. Effective Action 10.2 10.4 10.4 8.1 8.7 References 12.6 Instantons Zero Modes The Instanton Integral Evaluating the Determinant MultiInstanton Contributions References 143 143 150 154 158 163 166 167 167 170 181 184 186 187 187 193 200 203 208 209 209 212 215 218 222 226 235 236 239 239 246 255 9.2 9.3 Path Integral for Gauge Theories .3 8.1 Symmetries of the Action 11.1 Example 11. Gauge Theories 12.2 Noether's Theorem 11.5 Systems with Many Degrees of Freedom Relativistic Scalar Field Theory Feynman Rules Connected Diagrams References 10.3 10.2.4 9. Path Integral for Relativistic Theories 9.2 8.5 The Classical Field Effective Action Loop Expansion Effective Potential at One Loop References 11.
4 15.3 15. Ising Model 15.5 15.8 Index One Dimensional Ising Model The Partition Function Two Dimensional Ising Model Duality High and Low Temperature Expansions Quantum Mechanical Model Duality in the Quantum System References .4 14. Systems at Finite Temperature 14.1 Anomalous Ward Identity 13. Anomalies 13.2 15.4 12.7 15.5 12.2 Schwinger Model 13.2 14.3 References 14.1 15.3 14.1 14.xiv Field Theory: A Path Integral Approach 12.6 BRST Invariance Ward Identities References 266 274 278 279 279 289 307 309 309 314 318 324 326 327 327 332 337 339 343 349 356 358 359 13.6 15.5 Statistical Mechanics Critical Exponents Harmonic Oscillator Fermionic Oscillator References 15.
1 Particles and Fields Classically. a theory describing the displacements of the onedimensional string would constitute a 1+1 dimensional field theory whereas the more familiar Maxwell's equations (in four dimensions) can be regarded as a 3+1 dimensional field theory. t) or equivalently by the potentials (0(x. namely. a theory described by fields is usually known as a D + 1 dimensional field theory where D represents the number of spatial dimensions on which the field variables depend.Chapter 1 Introduction 1. Therefore. t). there is the motion of a particle or a rigid body (with a finite number of degrees of freedom) which can be described by a finite number of coordinates. fields depend continuously on some space variables as well. while the coordinates of a particle depend only on time. t)). And then. the displacement field. we can think of such a theory as a 0+1 dimensional field theory. it is clear that a theory describing the motion of a particle can be regarded as a special case. First. Familiar examples of classical fields are the electromagnetic fields described by E(x. Such systems are described by fields. In this language. Similarly. t) and B(x. namely. Thus. there are physical systems where the number of degrees of freedom is nondenumerably (noncountably) infinite. then. the motion of a onedimensional string is also described by a field 4>(x.t). A(x. there are two kinds of dynamical systems that we encounter. 1 . For example.
) • (1. For the relativistic case. we will use the BjorkenDrell convention. .4) The gradients are similarly obtained from Eqs. 2 . (1.2) Here we have assumed the speed of light to be unity (c = 1). l . .2 Field Theory: A Path Integral Approach 1. Namely.2) to be a =aMf' v )' (L6) <17) <"=£=(*•*)• so that the D'Alembertian takes the form • = d^ = rrd»dv = ~V2.x)..8) .MVX'/ = (t. therefore. the contravariant coordinates are assumed to be x» = (t. The covariant metric.5) (1. . . (1. follows to have a diagonal form with the signatures Vnu = ( + . (1.1) and (1...1) while the covariant coordinates have the form X/j. 3 . we will discuss both nonrelativistic as well as relativists theories.2 Metric and Other Notations In these lectures.x2 . .) • The invariant length is given by x2 = x % = rTxpXv = ri^x" = t2 . . /i = 0 . = T]. (1.x ) . TT = ( + . .3) The inverse or the contravariant metric clearly also has the same form.. namely. (1.
Introduction
3
1.3
Functionals
It is evident that in dealing with dynamical systems, we are dealing with functions of continuous variables. In fact, most of the times, we are really dealing with functions of functions which are otherwise known as functionals. If we are considering the motion of a particle in a potential in one dimension, then the Lagrangian is given by L(x, x) =  m i 2  V(x), (1.9)
where x(t) and x(t) denote the coordinate and the velocity of the particle and the simplest functional we can think of is the action functional defined as S[x] = [
f
dtL(x,x).
(1.10)
Note that unlike a function whose value depends on a particular point in the coordinate space, the value of the action depends on the entire trajectory along which the integration is carried out. For different paths connecting the initial and the final points, the value of the action functional will be different. Thus, a functional has the generic form
F\f] = JdxF(f(x)),
where, for example, we may have F(f(x)) = (f(x))n.
(1.11)
(1.12)
Sometimes, one loosely also says that F(f(x)) is a functional. The notion of a derivative can be extended to the case of functionals in a natural way through the notion of generalized functions. Thus, one defines the functional derivative or the Gateaux derivative from the linear functional
F'\v]=^F{f
+ ev]
=
/
d l
^ "
W

(U3)
4
Field Theory: A Path Integral Approach
Equivalently, from the working point of view, this simply corresponds to defining Sf(y)
eo l €
' '
It now follows from Eq. (1.14) that Sf{x) Sf(y) 8(xy). (1.15)
The functional derivative satisfies all the properties of a derivative, namely, it is linear and associative,
5 mm ZUiJ 5f(Xy llJi •(FM+FM=m+ 5f{x) 6f(x)' 5
6f{xy
<F1[f]F2[f]) = 5^4F2[f} + F1[f}S^4.
xWJ ,uu
Sf{x)
.WJ
xWJ
(1.16)
5f{x)
It also satisfies the chain rule of differentiation. Furthermore, we now see that given a functional F[f], we can Taylor expand it in the form F[f] = fdx P0(x) + fdXldx2 + where P0{x) = F(f(x))\nx)=0 Pi(xi,x2) P2(xi,x2,x3) = 8F(f(xi)) Sf(x2) 1 2! , (1.18)
f(x)=0
Pi(xux2)
f(x2) , (1.17)
dx1dx2dx3P2(xi,x2,x3)
f(x2)f(x3)\
=
PFifixt)) 5f(x2)8f(x3) /(x)=0
and so on. As simple examples, let us calculate a few particular functional derivatives.
Introduction
5
(i) Let
F[f] = Jdy F(f(y)) = J dy (f(y))n,
where n denotes a positive integer. Then, SF(f(y)) Sf(x)
= Hm
(1.19)
F(f(y) + e +
+ n
e5(yx))F(f(y)) e5(yx))n(f(y))n
e™
l]m(f(y)
= Hm(/(y))n
<^y))n~l^y
 *) + °( e2 )  (f(y»n
(1.20)
= n(/(y))n_1%a;). Therefore, we obtain SF\f] _ f 5F(f(y))
V
6f(x)
J
8f(x)
=
Jdyn(f(y)r15(yx)
(1.21)
= n(f(x))n~1.
(ii) Let us next consider the onedimensional action in Eq. (1.10) S[x] = [ with L(x(t),x(t)) = \m{x{t)f V(x(t)) (1.23)
f
dt' L(x(t'),x(t')),
(1.22)
= T{x(t))V(x(t)). In a straightforward manner, we obtain SV(x(t')) Sx(t)
=
e^o
y ( s ( f ) + e£ft'  t)) e
V{x(t'))
= ^ ' ( x ^ ' ) ) * ^  *),
(124)
6
Field Theory: A Path Integral Approach
where we have defined
Similarly, ST(x(t')) 5x(t)
= Um
T(x(t>) + e^O
e£r5(t>t))T(x(t>)) € (1.25)
= mx(t')—5(t't). It is clear now that SL(x(t'),x(t')) 8x(t) _ ~ 5{T{x(t'))V{x{t'))) Sx(t) V'{x(t'))5{t'  i).
= mx(t') ^6(1/ t)
(1.26)
Consequently, in this case, we obtain for ti <t <tf 5S[x] 8x(t)
=
/•*/ Jti
5L(x(t'),x(t')) 6x(t) {mx{t')^/8(t'
dt
= ffdt'
Jti
t)
V'(x{t'))5{t'  t))
= mx(t)
 V'(x(t))
+
ddL(x(t),x(t)) dt dx{t)
dL(x(t),x(t)) dx(t) '
{
' '
The right hand side is, of course, reminiscent of the EulerLagrange equation. In fact, we note that 6x(t) dt 8x(t) 8x{t) '
K
' '
gives the EulerLagrange equation as a functional extremum of the action. This is nothing other than the principle of least action expressed in a compact notation in the language of functionals.
Introduction
7
1.4
Review of Quantum Mechanics
In this section, we will describe very briefly the essential features of quantum mechanics assuming that the readers are familiar with the subject. The conventional approach to quantum mechanics starts with the Hamiltonian formulation of classical mechanics and promotes observables to noncommuting operators. The dynamics, in this case, is given by the timedependent Schrodinger equation **!*£» _
H h W
>,
(13)
where H denotes the Hamiltonian operator of the system. Equivalently, in the one dimensional case, the wave function of a particle satisfies
h2 d2 \ 2 + V(x))^(x,t), 2m dx where we have identified # E , * ) = <*#(*)>.
(1.30)
(131)
with \x) denoting the coordinate basis states. This, then, defines the time evolution of the system. The main purpose behind solving the Schrodinger equation lies in determining the time evolution operator which generates the time translation of the system. Namely, the time evolution operator transforms the quantum mechanical state at an earlier time ti to a future time t\ as M*l)) = tf(*l,*2M*2)>. (132)
Clearly, for a time independent Hamiltonian, we see from Eq. (1.29) (the Schrodinger equation) that for t\ > t2, U(h,t2) = e^tlt2)H. (1.33)
8
Field Theory: A Path Integral Approach
More explicitly, we can write U(h,t2) = B(h  t2)e^^H . (1.34)
It is obvious that the time evolution operator is nothing other than the Green's function for the time dependent Schrodinger equation and satisfies (ih^ H\ U(h,t2) = ih5(h  *2) • (135)
Determining this operator is equivalent to finding its matrix elements in a given basis. Thus, for example, in the coordinate basis defined by X\x)=x\x), (1.36) we can write {xi\U(t1,t2)\x2) = U(ti,x1;t2,x2). (137)
If we know the function U(ti,xi\t2,x2) completely, then the time evolution of the wave function can be written as ip(xi,h) = dx2U{ti,xi;t2,x2)ip(x2,t2). (1.38)
It is interesting to note that the dependence on the intermediate times drops out in the above equation as can be easily checked. Our discussion has been within the framework of the Schrodinger picture so far where the quantum states \i/j(t)) carry time dependence while the operators are time independent. On the other hand, in the Heisenberg picture, where the quantum states are time independent, using Eq. (1.32) we can identify
IV>H = m = o))s = m = o))
(1.39)
Introduction
9
In this picture, the operators carry all the time dependence. For example, the coordinate operator in the Heisenberg picture is related to the coordinate operator in the Schrodinger picture through the relation XH{t) = e*tHXe*tH. The eigenstates of this operator satisfying XH{t)\x,t)H = x\x,t)H, (1.41) (1.40)
are then easily seen to be related to the coordinate basis in the Schrodinger picture through \x,t)H = e*tH\x). It is clear now that for t\ > t2 we can write H(xi,t1\x2,t2)H = ( m l e  i ^ e t * 2 ^ ) = (xl\e^t^H\x2) (1.42)
= {xi\U{h,t2)\x2) = U(t1,x1;t2,x2). (143)
This shows that the matrix elements of the time evolution operator are nothing other than the time ordered transition amplitudes between the coordinate basis states in the Heisenberg picture. Finally, there is the interaction picture where both the quantum states as well as the operators carry partial time dependence. Without going into any technical detail, let us simply note here that the interaction picture is quite useful in the study of nontrivially interacting theories. In any case, the goal of the study of quantum mechanics in any of these pictures is to construct the matrix elements of the time evolution operator which as we have seen can be identified with transition amplitudes between the coordinate basis states in the Heisenberg picture.
10
Field Theory: A Path Integral Approach
1.5
References
Das, A., "Lectures on Quantum Mechanics", Hindustan Book Agency. Dirac, P. A. M., "Principles of Quantum Mechanics", Oxford Univ. Press. Schiff, L. I., "Quantum Mechanics", McGrawHill Publishing.
Chapter 2
P a t h Integrals and Quantum Mechanics
2.1
Basis States
Before going into the derivation of the path integral representation for U(tf,xf,ti,Xi) or the transition amplitude, let us recapitulate some of the basic formulae of quantum mechanics. Consider, for simplicity, a one dimensional quantum mechanical system. The eigenstates of the coordinate operator, as we have seen in Eq. (1.36), satisfy X\x)=x\x). (2.1) These eigenstates define an orthonormal basis. Namely, they satisfy (x\x') = 5(x — x'), dx\x)(x\ = l. (2.2)
Similarly, the eigenstates of the momentum operator satisfying P\p)=p\p), (2.3)
also define an orthonormal basis. Namely, the momentum eigenstates satisfy (p\p') = 8(pp>),
Jdp\p){p\ = l.
ll
(2.4)
12
Field Theory: A Path Integral Approach
The inner product of the coordinate and the momentum basis states gives the matrix elements of the transformation operator between the two basis. In fact, one can readily determine that (p\x) = ^ e " ^ = (xW . (2.5)
These are the defining relations for Fourier transforms. Namely, using the completeness relations of the basis states, the Fourier transform of functions can be defined as
f(x) = {x\f) = Jdp{x\p){p\f)
1
[
V^ J '
f(k) = Vhf(p)
dk el"xf{k),
(2.6)
f dxe~y;pxf(x)
}=Jdxe^f(x).
(2.7)
These simply take a function from a given space to its conjugate space or the dual space. Here k =  can be thought of as the wave number in the case of a quantum mechanical particle. (Some other authors may define Fourier transform with alternate normalizations. Here, the definition is symmetrical.) As we have seen in Eq. (1.42), the Heisenberg states are related to the Schrodinger states in a simple way. For the coordinate basis states, for example, we will have \x,t)H = e%m\x).
It follows now that the coordinate basis states in the Heisenberg
Path Integrals and Quantum
Mechanics
13
picture satisfy
H(x,t\x',t)H
=
^tHAtH\J (x\eKthe*tH\x')
= (x\x') = 5(x  x1), and dx \x,t}H
H(x,t\
(2.8)
= = enm
' dxehtH\x)( x\e
h \tH
fdx\x){ x\e~*tH ,
= 1.
(2.9)
It is worth noting here that the orthonormality as well as the completeness relations hold for the Heisenberg states only at equal times. 2.2 Operator Ordering
In the Hamiltonian formalism, the transition from classical mechanics to quantum mechanics is achieved by promoting observables to operators which are not necessarily commuting. Consequently, the Hamiltonian of the classical system is supposed to go over to the quantum operator H(x,p)^H(xop,Pop). (2.10)
This, however, does not specify what should be done when products of x and p (which are noncommuting as operators) are involved. For example, classically we know that xp — px. Therefore, the order of these terms does not matter in the classical Hamiltonian. Quantum mechanically, however, the order of the operators is quite crucial and a priori it is not clear what such a term
14
Field Theory: A Path Integral Approach
ought to correspond to in the quantum theory. This is the operator ordering problem and, unfortunately, there is no well defined principle which specifies the order of operators in the passage from classical to quantum mechanics. There are, however, a few prescriptions which one uses conventionally. In normal ordering, one orders the products of x's and p's such that the momenta stand to the left of the coordinates. Thus,
N.O.
xp —> px ,
N.O.
px —• px , 2 N.O. 2 x p —• px ,
N.O.
o
/r. 1 1 \
xpx —> px , (2.11) and so on. However, the prescription that is much more widely used and is much more satisfactory from various other points of view is the Weyl ordering. Here one symmetrizes the product of operators in all possible combinations with equal weight. Thus, xp —^ ~{xp + px), w.o. I , , . px —• (xp + px),
2 W.O. I / 9 , , 2\
x p —> [x p + xpx + px ) , xpx —>' (x2p + xpx + px2), (212)
and so on. For normal ordering, it is easy to see that for any quantum Hamiltonian obtained from the classical Hamiltonian H(x,p) (x'\HN°\x) = f dp(x'\p)(p\HN°\x)
= J' ^.e&*^H{x,p).
(2.13)
3) and (2. we obtain using the BakerCampbellHausdorff formula Q\ 2 )ePPope( 2 I = gl 2 ) (APPop~\ 2 ~ > — e(ax0p+/3p0p) _ (2.op and pop is a constant.) To understand Weyl ordering. .op+/3pop) would.Path Integrals and Quantum Mechanics 15 Here we have used the completeness relations of the momentum basis states given in Eq.^ anr(xnopP™)W° (2. generate all such powers and by analyzing the matrix elements of this exponential operator. on the other hand. (2. (The matrix element of the quantum Hamiltonian is a classical function for which the ordering is irrelevant.1).14) The expansion of the exponential operator e(aa. we will learn about the matrix elements of Weyl ordered Hamiltonians.4) as well as the defining relations in Eqs. we can easily show that (*xop + (3pop)N = ]T n+m=N . it can now be easily shown that dp (x \e( 2 >epp°p\p){p\eK 2 >\x) (2. Prom the fact that the commutator of a. generates the Weyl ordering of products of the form a^opPop naturally if we treat xop and pop as noncommuting operators. of course.15) Using this relation. let us note that the expansion of (axop + j3pop)N.5). (2. In fact.16) / pefc** \C^+0v). (2.
1). let * = tjjr • (2. (2. tn = U + ne.16 Field Theory: A Path Integral Approach Once again. (2.tN2)H (2. (2.4) as well as the defining relations in Eqs. the matrix elements of the Weyl ordered Hamiltonian leads to what is known as the midpoint prescription and this is what we will use in all of our discussions.h\Xi. (2. Let us divide the time interval between the initial and the final time into N equal segments of infinitesimal length e.9)).xf. we obtain U(tf.20) .tf\xNi. we discretize the time interval and in the end. We are now ready to calculate the transition amplitude.ti)H. We can now label the > intermediate times as.ti)H .. we will have <z'#w°(Wop).Xf]ti.Xi) = H(xf. Let us recall that in the Heisenberg picture.2.3) and (2. we are interested in taking the continuum limit e — 0 > and N — oo such that Eq. X H{XNl.5). (2. say.ti. for tf > U. we have used here the completeness properties given in Eq..tNi)H • • H {xi.tf\xi.tNl\xN2..tf\Xi. n = l. we have U (tf.(iVl).17) As we see. It follows from this that for a Weyl ordered quantum Hamiltonian.18) holds true.19) Introducing complete sets of coordinate basis states for every intermediate time point (see Eq.p) • (2. for simplicity.i8) In other words.Xi) = H(Xf. (2. Namely.ti)H = lim N^oo dx1dxNlH{xf.) = f ^ ei«°*H {^f..
we have clearly assumed an inherent time ordering from left to right. It is worth emphasizing here that the number of intermediate coordinate integrations differs from the number of momentum integrations and has profound consequences in the study of the symmetry properties of the transition amplitudes.tn—i)H = \xn\e ft eft \xn—\) = {age^'^lxni) = = (xn\e~^eH\xni) f d Pn JpnjXnXn^^H^^^^) ^ ^ ^ Here we have used the midpoint prescription of Eq.23) This is the crudest form of Feynman's path integral and is defined in the phase space of the system.»0«fi(=^*. (2. we obtain U(tf. (2.xf. we can write > . (2.22) In writing this. Note that in the continuum limit.Path Integrals and Quantum Mechanics 17 In writing this. for e — 0. Substituting this form of the inner product into the transition amplitude.17) corresponding to Weyl ordering. we note that any intermediate inner product in Eq.ti.20) has the form H\%nitn\xni. Furthermore. XN = Xf. there are only (N— 1) intermediate points of integration. namely.)) _ (2. Let us also note here that while there are iV inner products in the above expression. we have identified XQ = Xi.Xi)= dpi dpN / dxi • • • dajjvi 2Kh'" 2nh e>0 J lim N—>oo xeiE^K.
p)) (2. it is proportional to the action in the mixed variables.xf. To obtain the more familiar form of the path integral involving the Lagrangian in the configuration space. can be done readily.25) n such a case. (2. therefore.26) The momentum integrals are Gaussian and.Xi) = hm / dzi • • • dxNi„ fc • N—>oo y/xn+xn. (2. Namely.p) = ^ + V(x). We note that U A„ icfPn Pn<. ^ ^ / = hm . Namely.22) as N I™ ^ Yl [P^(xn ~ xni) eH ( N^oo " n=1 ^n' 1 . (2. we have from Eq.H( .ti. let us specialize to the class of Hamiltonians which are quadratic in the momentum variables. p n ) ) I %n ~r Xn—\ N / v / / Xn \ Xn— 1 \ TT i.22) U(tf.x„xn_1) Pn 6 t I 2m £ / 2TT^ An a n i£ ( „1 e2^h{Pn 2m Pn(xnxnl) / 2irh P i .18 Field Theory: A Path Integral Approach the phase factor of Eq.Pr H(x. dpN 2KK xe Vv *)). let us choose H(x.24) =[ dt L. (2.e £ [Pn [ *£ U dtip± J.
27) \2mheJ Substituting this back into the transition amplitude in Eq. (2.xi) = lim ( ^ ) iey^AT 2 I m fxnxnl\ y(Xn+xnl e / dxi • • • dxNieh n _ 1 V2^ ' ^ = A f Vx etti<l*m±2V^ (2.10).26). .tl.xf.28) = AJvxe*sW. where A is a constant independent of the dynamics of the system and S[x] is the action for the system given in Eq.Path Integrals and Quantum m(xnxn_i)\ji Mechanics ( m(xnxn_i) ' 19 " / 1 2nE e (2'Kmh\2 ')(• gif 1 "" 1 ".1 = (—Y N_ (2. (1. This is Feynman's path integral for the transition amplitude in quantum mechanics. we obtain U(tf.
xi) lim A .28). let us try to understand the meaning of the path integral measure Vx. connecting the two points. HTAr 1 P e^O JV—xx> " J . Quantum mechanically. Feynman's path integral simply says that the transition amplitude between an initial and a final state is the sum over all paths.ti. We know from the study of quantum mechanics that if a process can take place in several distinct ways. the end points are held fixed and only the intermediate coordinates are integrated over the entire space. what we see is that all the paths contribute to the transition amplitude. . Thus. therefore.20 Field Theory: A Path Integral Approach To understand the meaning of this. For Hamiltonians which are not quadratic in the momenta. quite expected. the expression holds in general.29) y(xn+xnl\ / rl'ri . = A f Vxei5^ mfxnxnl\ 2 (2. then the transition amplitude is the sum of the individual amplitudes corresponding to every possible way the process can take place. it is represented as U(tf. of course. Therefore. The sum over the paths is. However. It is also worth pointing out here that even though we derived the path integral representation for the transition amplitude for a special class of Hamiltonians. gives rise to a trajectory between the initial and the final points. of the weight factor eft ^J. it is the weight factor e^s^x' that is quite crucial and unexpected. we know that it is the classical action that determines the classical dynamics. however. one should simply be careful in defining the path integral measure Vx. . (2.3 The Classical Limit As we have seen in Eq. Classically. Any spatial configuration of the intermediate points. integrating over all such configurations (that is precisely what the integrations over the intermediate points are supposed to do) is equivalent to summing over all the paths connecting the initial and the final points.xf. In this integration. 2. the transition amplitude can be written as a sum over paths and for the case of a one dimensional Hamiltonian which is quadratic in the momentum.
in the path integral. we will ignore this technical point. Even though one can be more quantitative in the discussion of the behavior of the transition amplitude. therefore. concentrate only on paths connecting the initial and the final points that differ from one another only slightly. we only look at continuous paths which are differentiable. it is only the classical path that is singled out in the classical limit. all such contributions will average out to zero. We note that for paths where Xn ^ the first term in the exponential would be quite large. Thus. Let us. namely. But for simplicity of argument. there would be a nearby xn differing only slightly which would have a cancelling effect. let us try to be qualitative in the following. For simplicity. We note here that the weight factor in the > . for every such xn. (A more careful analysis shows that the paths which contribute nontrivially are the continuous paths which are not necessarily differentiable.Path Integrals and Quantum Mechanics 21 where AN = (^k)2 • Xn—X. particularly since e is infinitesimally small. when h — 0. Therefore. the weight factor can easily be positive or negative. In other words. such paths will lead to a very large phase and consequently.) The question that we would like to understand is how among all the paths which can contribute to the transition amplitude.
But. Namely. eft *x\ is a phase multiplied by a large quantity when h — 0.4 Equivalence with the Schrodinger Equation At this point one may wonder about the Schrodinger equation in the path integral formalism. clearly. there will be a nearby path. Mathematically. only the trajectories close to the ones satisfying iM=0' < 3) 20  would contribute significantly to the transition amplitude in the classical limit. it is clear that the • dominant contribution to the path integral would arise from paths near the one which extremizes the phase factor. of course. therefore. we know that these are precisely the trajectories which a classical particle would follow. For every such path. but rather because there are paths infinitesimally close to it which add coherently. All such paths. One can. namely. the action will not change. In other words. where the action would differ by a small amount. we are considering a path. Suppose. Therefore. 2.22 Field Theory: A Path Integral Approach path integral. which is quite far away from the classical trajectory. say # 3 . make various estimates as to how far away a path can be from the classical trajectory before its contribution becomes unimportant. however. all such paths will add up coherently and give the dominant contribution as h — 0. the action is stationary. but since it is multiplied by a large constant would produce a large phase. the classical trajectories. Consequently. Once again. it is not clear how we can recover the time dependent Schrodinger equation (see Eq. namely. say # 2 . infinitesimally close. But let us not go into these details here. It is in this way that the » classical trajectory is singled out in the classical limit. Then. because h is small. if we choose a path infinitesimally close to the classical path. from the principle of least action. the phase along this trajectory will be quite large.30)) from the . will average out to zero in the sum. (1. Near the classical trajectory. we can see this more intuitively in the following way. not because it contributes the most.
e) = ( ~ ) h V ^H{x\ 0).29). Therefore.32) Therefore. (2. so that we can write rP(x. Let us recall that the Schrodinger equation is a differential equation.32).xf. e) = ( ^  .34) J°° dV e S ^ .Xi) v \2Kihe) ' We also know from Eq. /m^Kl^'vpP)). we merely have to examine the infinitesimal form of the transition amplitude or the path integral.38) that the transition amplitude is the propagator which gives the propagation of the wave function in the following way. if 7 is large.x')il>(x'.36) .^ ^ + l ) ] ^ . we obtain 1>(x.x. (M1) oo / dx' U(e. The dominant contribution will.35) It is obvious that because e is infinitesimal.33) (2.31) into Eq. to derive the Schrodinger equation. + v. 0). come from the region of integration 0 < H < ( ^ ) \ (2. oo (2. Consequently.x . we obtain for infinitesimal e U(tf = e.0). substituting the form of the transition amplitude namely. Eq. (2. (2. (2. therefore. (2.ti = 0.0. then the first 7 term in the exponent would lead to rapid oscillations and all such contributions will average out to zero. Prom the explicit form of the transition amplitude in Eq.) *  ° ° Me&l**'^f Let us next change variables to r] = x' .Path Integrals and Quantum Mechanics 23 path integral representation of the transition amplitude. (1. it determines infinitesimal changes in the wave function.
24 Field Theory: A Path Integral Approach where the change in the first exponent is of the order of unity. . 9 inL„2 ihe f2mhe\2 I dr? 7 ezfie'7 = — 7 Joo m \ m J Note that these integrals contain oscillatory integrands and the simplest way of evaluating them is through a regularization.< ( z . 0) + rtfix. Thus. OO /"OO / . Therefore. we obtain x U(x. \ V = (™**)\ (2. e 2 ) The individual integrations can be easily done and the results are f°° . 1 (.0) V(x)1>(x.37) . we can keep terms consistently up to order e. 0 ) + O ( r 7 3 . For example. we can Taylor expand the integrand and since we are interested in the infinitesimal behavior.2 + # ' ( * . iEL„2 f2mhe\* / dr? e^ne7! = Joo V ™ / oo drirje^2 = 0.39) and so on.0) (2. 0) +  .38) / •oo /•°° . 0) + 0{rf) = (^)T*i oo im 2 a Ihe 'I il>{x.oo dr? e ^ 7 ' = lim / = lim <5>0+ OO <5^0+ J_00 dry e^zsi" 1 . . 0) + \tl/'{x. (2.
we obtain . Since we do not have a feeling for such quantities.jV (x) i. contains the Schrodinger equation and is equivalent to it. ( .^ ^ + y(a. (1. the Lagrangian has the form L = mi2 . (2.0) = .37). (2. For a free particle in one dimension.40) In the limit e — 0.30)) The path integral representation.( — J ^"(x.0) + O(62).Path Integrals and Quantum Mechanics 25 Substituting these back into Eq.5 Free Particle We recognize that the path integral is a functional integral.£)^(x. 0) + ^ V (x. Namely.))v(x. The free particle is probably the simplest of quantum mechanical systems. therefore. (2. (x.0) + O( e 2 ) =tf>(x. 0) + O (e2) or^(x. therefore. the integrand which is the phase factor is a functional of the trajectory between the initial and the final points.41) . we obtain the time dependent • Schrodinger equation (Eq. 0) . let us evaluate some of these integrals associated with simple systems. 2.
Xi)= hm ( _ ) iV>oo —>oo X ^ _ J /"dyi • • dj/jvi e*E^=i(i/nvni)a .45) .ti.42) Defining / m \ ^ = U i d \2heJ we have *«. (2.U. We note that [dyx e*[(wi»)2+(i«yi)a] = [dyi eiPfoi^T^+^ww)2] ™\*ettovo)\ (2.xf. the simplest method probably is to work out a few lower order ones and derive a pattern. we obtain U(tf.xf.28) or (2.29). (2.i e J h ^n=1 2 ^ £ ) lim ( ) 2 / dxi • • e>o J TV—>oo v 2nine J •dxNie^'^n=l(Xn~Xn i)2 (2. from our definition of the transition amplitude in Eq.Xi) = lim I ——) e^o V 2mne I JV—>oo / dxi • • • d ^ .44) This is a Gaussian integral which can be evaluated in many different ways.26 Field Theory: A Path Integral Approach Therefore. However. (243) U(tf.
xf. the transition amplitude can be explicitly evaluated.) \f fm^IV _\fl m f^ l e^o e e2HNc N—*oo = hm ( ) e^hNeixf x i) e^o AT—>oo \2mhNeJ I i a m x x ( f. It has the right behavior in the sense that. then we will have / dyidy 2 e i [^ 1 ~ yo ^ 2+ ^ 2 ~ ?/1 ^ + ^ 3 '" ?/2 ^ 2 ] = (*I\* fdy2eih(y2y0)2+(y3y2?] dy2eifte^) 2 +t(«)2 )'/ 2 / V 3 i ' (wr .xf.47) Thus. we see as tf —> U. (2.Xi) e+o \2irihe) e \ m \ N (xN—xo) JV—>oo h \2mne/ .46) A pattern is now obvious and using this we can write U(tf. we see that for a free particle.Path Integrals and Quantum Mechanics 27 If we had two intermediate integrations. 2 \ 2 e f(»/3!/o) (2.ti. U{tf.48) .ti.Xi)>8(xfXi).i) m 2irih(tf — ti) (*/*0 (2.
(1.50) dt \^mx\ = ^mv2(tf . Second. in this case.28 Field Theory: A Path Integral Approach which is nothing other than the orthonormality relation for the states in the Heisenberg picture given in Eq.Xi = V(tf ti) or. v=Xf^. we obtain (2. Let us note here that since S[x] = J dt mi. even the simplest of equations has only dispersive solutions. therefore. Furthermore this is exactly what one would obtain by solving the Schrodinger equation. we have S[xd] = f f (2. we can write Xf . That is. since v is a constant. ti the EulerLagrange equations give (see Eq. simply as .U). all the potentially dangerous singular terms involving e have disappeared. (2. (2.28)) This gives as solutions Xcl(t) = v = constant. (2 53)  We recognize.8). Substituting this back into Eq. Thus.51) On the other hand. (2. for the classical trajectory. 2 . It expresses the well known fact that even a well localized wave packet spreads with time.51). that we can also write the quantum transition amplitude.52) *M=Mt^)'<''«=^.
J2^ ( V \ / (im\ 2 fxf —„ \ 2 Xi .— fXf ~Xi\2 Tf] 2 \tfu) ) .„„.8U xh ti2 d2U . wr^!*f> ( 5) 27  which is equivalent to saying that the transition amplitude obtained from Feynman's path integral.Path Integrals and Quantum Mechanics 29 This is a particular characteristic of some quantum systems which can be exactly solved. indeed. 2m {. the transition amplitude can be written in the form U(tf. it follows that . (2.ti.„„. for these systems. (1. .BU\ Therefore.55) where A is a constant. (2. Finally. solves the Schrodinger equation for a free particle (compare with Eq..xf.35)).47) that dU _ U ~dff ~ ~2{tf U) im /xf — Xi\ ~~2h\tfUj TT dU _ im fxf — %i\ dxf h \tf — t d2U _ im U dxi h tf — ti . Namely. \ h J \ tf — t in h U 2{tfU)+ 4.xi) = Ae^x^. let us note from the explicit form of the transition amplitude in Eq.
6 References Das. Feynman. R.30 Field Theory: A Path Integral Approach 2. "Techniques and Applications of Path Integration" .. Hibbs. McGrawHill Publishing. "Quantum Theory of Many Variable Systems and Fields". "Quantum Mechanics and Path Integrals". "Lectures on Quantum Mechanics". R. and A. L. World Scientific Publishing.. John Wiley Publishing. Schulman. B.. Sakita. A. S. . Hindustan Book Agency. P.
we can think of the time dependent external source J(t) as an electric field if the oscillator is supposed to carry an electric charge.2) Here.1 Path Integral for the Harmonic Oscillator As a second example of the path integrals.1) with the action given by / ' dt L. we know that if the external source were time independent. Zi Z (3.m i 2 — mu>2x2 + Jx. > Furthermore. let us consider the oscillator interacting with an external source described by the Lagrangian L = .Chapter 3 Harmonic Oscillator 3. The well known results for the free harmonic oscillator can be obtained from this system in the limit J(t) — 0. let us consider the one dimensional harmonic oscillator which we know can be solved exactly. (3.3) 31 .1) as 1 J2 2 L = rax2 + Jx mu x 25 Z 1 2 = = mx \ muj2 ( x J N 2 J2 rruoz j) + 2muj2 2 1 I m s 2 _ _ m w2—2 + _ J _ )> ^ r2 mu 2mui (3. In fact. then the problem can also be solved exactly simply because in this case we can write the Lagrangian of Eq. for example. (3.
mxc\ + moj xc\ — J = 0 . of considerable interest because we can obtain various known special cases in different limits.1) is. namely. we have 5S[x] 5x(t) 0. as we have seen in Eq.5) that the action is an extremum for the classical trajectory. if we define x(t) = xcl(t) + r)(t). (3. (3.32 Field Theory: A Path Integral Approach where we have defined J x= x— (3.28). let us note that the action is at most quadratic in the dynamical variables x(t).xf. in such a case. The system described by Eq. U. X—Xc\ (3.6) To evaluate this functional integral. we can Taylor expand the action about the classical path as S[x] = S[xcl + v] = S[xcl] + fdt rj(t) SS[x] X—Xc\ + J dtidta T7(ti)T7(t2) lJ 6x(ti)5x(t2) (3. and the general form of the transition amplitude. Xi) = A Vx e%s[x]. the classical equilibrium position of the oscillator is shifted by a constant amount. (3. the system behaves like a spring suspended freely under the effect of gravity. Therefore.2) gives the classical trajectory and takes the form SS[x] 5x(t) 0 (3. The EulerLagrange equation for the action in Eq. (3. is given by U(tf. Therefore.9) .7) then.5) or.4) mtu' In other words.8) X—Xc\ We note from Eq. (3. therefore. (2.
10) X—Xc\ If we evaluate the functional derivatives in Eq.m w V ) . we note that the integrand in the exponent of the functional integral does not depend on time explicitly. Consequently.2 Method of Fourier Transform First of all. e^ 4 ' <^v2m^r. (3. (3.2). we will evaluate this integral in three different ways so as to develop a feeling for the path integrals.Harmonic Oscillator 33 Consequently.12). (3. namely. (3. 3. Since the harmonic oscillator is a fundamental system in any branch of physics. we can rewrite the transition amplitude in this case as = A e t 5 ^ ' ] f Vr. (3. Since the end points of the trajectories are fixed. (3. it measures the deviation of a trajectory from the classical trajectory. we can also rewrite the action as S[x] = S[xcl] + 1 ffdt (mf]2 . we can redefine the variable of integration as t>tU. (3.10) for the action in Eq.2) _ (3_13) This is an integral where the exponent is quadratic in the variables and such an integral can be done in several ways. the fluctuations satisfy the boundary conditions r){U)=v(tf) = 0.11) 2 hi The variable T](t) represents the quantum fluctuations around the classical path.14) . we can also write Eq. Therefore.8) as S[x] = S[xcl] + ^ fdhdt2 S2S[x] vitiMh) 5x(t\)8x(t2) (3.12) It is clear that summing over all the paths is equivalent to summing over all possible fluctuations subject to the constraint in Eq.
(3. the value of the fluctuation at any point on the trajectory can be represented as a Fourier series of the form r/(t) = ^ a n s i n ^ V n = 1.15) where we have identified the time interval with T = tfU. Substituting this back. (3. Similarly. (3.1.16) The variable rj(t) satisfies the boundary conditions (see Eq.18) We note here that since we have chosen to divide the trajectory into N intervals. we also obtain / dtr]2(t) = J2 dt a a ™™ sin ( ^Y ) Sln ( ~Tr ) n . since there are (N — 1) intermediate time points.18).17) Consequently. namely. we can write the transition amplitude as U(tf. . . .34 Field Theory: A Path Integral Approach in which case. (3. (3. 2 .12)) 77(0)=77(T) = 0.Xi) = A e i 5 ^ /' Vr\ e^^dtH2~m^2) . (3.N . there can only be (N — 1) independent coefficients an in the Fourier expansion in Eq. .xf.U.19) where we have used the orthonormality properties of the cosine functions. we find that Ja dt" = E / 0 dta°™(T) = E(T)2°»n ("FJ cos {r) cos ("FJ < 3 .
ti. n=l .2^2 / /. (3.nni.Xi) = lim A'et™ 6^0 N—>oo J /dax.4mh\2 //n7r\2 9 2\ i "2 Substituting this form of the individual integrals into the expression for the transition amplitude in Eq. we obtain Nl / \ / m\ 2 \ ~\ U(tf. In fact. (3. is a product of a set of decoupled integrals each of which has the form of a Gaussian integral which can be easily evaluated. we note that integrating over all possible configurations of r)(t) or all possible quantum fluctuations is equivalent to integrating over all possible values of the coefficients of expansion an. in this case.da^e^^C^^Wa) = lim A'e*sl*d N—KX) / dai • • • dajyi e~^ E "=1 ^ } ~" N .21) Here we note that any possible factor arising from the Jacobian in the change of variables from 7 to the coefficients an has been lumped 7 into A' whose form we will determine shortly. . Thus we can write the transition amplitude also as U(tf. {uTy\ sinuT .f ^ \ n=l j / . the individual integrals have the values (see Eq.21).xi)= lim A"e*s^ N±™ JV*oo J] ( l . (2. ATl TT / .Harmonic Oscillator 35 Furthermore.xf.39)) imT (fmr\2 .xf.23) If we now use the identity. We note here that the transition amplitude. (3.ti.
26) lim A"=(^Y N—>oo . the harmonic oscillator reduces to a free particle for which we have already evaluated the transition amplitude.N. and comparing with Eq. Let us parameterize time on a trajectory as tn = U + ne.36 Field Theory: A Path Integral Approach we obtain U(tf..*)=(sjHi_)\i«W...M.25).xf. . (3. (3. (2.3 Matrix Method If the evaluation of the path integral by the method of Fourier transforms appears less satisfactory.27) Therefore. In fact. recalling from Eq..(t. we obtain (3...Xi)= lim A"eis^ £—•0 N—>oo \ (^P\ UJl J ' . n = 0. then let us evaluate the integral in the conventional manner by discretizing the time interval. • 3.U. (3. we determine the complete form of the transition amplitude for the harmonic oscillator to be It is quite straightforward to see that this expression reduces to the transition amplitude for the free particle in the limit of u> — 0.l.25) We can determine the constant A" by simply noting that when u = 0..54) that ft.
we can write the transition amplitude in Eq.31) corresponding to the boundary conditions in Eq.15) in the explicit form U{tf. let us define the values of the fluctuations at these points as V(tn) = Vn • (329) Then.33.dr?jv . . in this case.V) f N_ AT—>oo a (2^) ^ y eB_1 2e c. The transition amplitude.12).U. (3. we are supposed to identify Vo = r)N = 0 . (3. . namely.Xi) = A et5^.1 V 2 . let us rescale the variables as r>n>(^Vn. (3.] j V i q e ^/t/dt(m^mu. To simplify the integral. (3.xf.Harmonic Oscillator 37 Correspondingly._ ( 3 3 0 ) x eM \ ^ > In this expression.] d7?i e ".32) U(tf.Xf.ti. will take the form (3.Xi) N—>oo x «(<— >'M^)') .
37) V '• where we have defined x=2 1 / eW e2tu2 (3. f m\ m \VNIJ then.1 ••• / ( 2 1 0 0 2 1 0 1 1 2 •••\ •• •• 2 2 eu 1 0 (3. (3.34) and the (N — 1) x (N — 1) matrix B has the form ( B = 2 1 0 0 •••\ 1 21 0 ••• 01 2 . we can also write the transition amplitude in terms of matrices as (3. (3. namely.38) y = [i + . therefore. we can write it as y 0 0 .34) AT—>oo x I dqe^Bi.38 Field Theory: A Path Integral Approach If we think of the ?7n's (there are (N — 1) of them) as forming a column matrix.35) Here r}T represents the transpose of the column matrix in Eq.^ y x y 0 •• 0 y x y •• (3.36) / V : (x B \: This is a symmetric matrix and.
we determine the form of the transition amplitude for the harmonic oscillator to be U(tf.35).40) / / d C eJC BDC / dCi • • • dCiviiE^bnC e T n(r" 71=1 N Nl = (z7r)^f i (det J B)~5 . (3.xf. In other words.ti. (bx 0 0 . defining we obtain &neiriTBl1 i i J (3.41) Here we have used the familiar fact that the Jacobian for a change of variable by a unitary matrix is unity. (3.39) \\ Therefore.2 e \S\xa\ . e>o \27rzrtedet £>/ lim f m J^M (3. therefore. (3.^ 0 b2 0 ••• BD o o h ••• = UBU]. therefore.42) . Using this result in Eq. can be diagonahzed by a unitary matrix (more precisely by an orthogonal matrix) which we denote by U.Xi) = N—KX) S l s * J Ur ( rn \ 2 v ( m \f {2he\^ <*>•<***)"* \S[xc lim — 6^0 N+00 \2mheJ ( d e t m .Harmonic Oscillator 39 The matrix B is clearly Hermitian (both x and y are real) and.
(3. We note then. of course.37) and (3.40 Field Theory: A Path Integral Approach It is clear from this analysis that the transition amplitude can be denned only if the matrix B does not have any vanishing eigenvalue..46) In the continuum limit. (3. e^0 N+oo (3.. that if we denote the determinant of the n x n submatrices of B as In. We note here that the main quantity to calculate in order to evaluate the transition amplitude is lim e det B .38). In+\ — 2In + Inl = + ^)2/„1 + ^n1 H eJ2. In+1 — ^In + Inl g = W 1(T 17\ n 2 / . we can identify t = ne as a continuous variable as e — 0. that > e»0 AT^oo lim e det B = lim e/jvi = <X*f .45) We are. we can think of this as a continuous function of t. In other words.2. Substituting the form of x and y. where we restrict /_i = 0.43) W = 2(i^)4(i or. interested in the continuum limit. then it is easy to check that they satisfy the recursion relation In+1=xIny2In1. r + Inl .l.44) This recursion relation can be checked trivially for low orders of the matrix determinants.. /o = 1. .U) = 4>{T).2 w 7T~ \ In ^—IJI1 or. £•0 Let us note from the special structure of B in Eqs. In order to do so. let us define a function (/>(tn ~ U) = <p(ne) = e/„ ..47) . we obtain n = 0. (3. H — Jnl (3. (3.
(3. (3. for the harmonic oscillator.53) v y \2mhBxnuTJ This is.Harmonic Oscillator 41 We also note from Eq. smut . Let us recall from . (3. _„.51) It now follows from this that lim e d e t 5 = lim eijvi = 4>(T) = ^ " ^ N—»oo N*oo • (3. t. x) lim ( ^ V ^Q \2TrihedetBJ 1 ess[*d] =( r JT^SM (3. we conclude that in the limit e —• 0. in the continuum limit. 0(0) = lim e/ 0 = lim e = 0.^r) = H (2" ¥ ~ 0 =l • (3'49) Furthermore. (3.46) that. (3. <f>(t) = .52) Consequently.ti._.xf. Let us next describe an alternate way to determine detB which is quite useful in studying some specific problems. e—>0 e—>0 (3. *°> = H'.45).49)) is clearly .28) using the method of Fourier transforms. (3. what we had already derived in Eq.. from the recursion relation for the J n 's in Eq. /.42) to be U(tf.48) and similarly. of course.48) and (3. the function 4>{t) satisfies the > differential equation ^ = ^(t). (350) We recognize this to be the harmonic oscillator equation and the solution subject to the initial conditions (Eqs.xi)= v /. we obtain the transition amplitude in Eq. .
43) that the determinant of the nxn matrices. i .58) .1 r We note here that we can write these recursion relations also in the simple matrix form as In+l In x y* 1 0 (3. In fact.57) with S~x = A+A_ (3. the 2 x 2 matrix can be trivially diagonalized by a similarity transformation.55) in a straightforward manner. (3. (3.54) Inl. we obtain INi IN2 x y 1 0 .2X /7l Io (N2)factors o \ (JV2) x —y* V 1 0 ' l x (3.42 Field Theory: A Path Integral Approach Eq. for n = N — 2.56) Furthermore. In. Iterating this (N — 2) times.55) We can determine the eigenvalues of the fundamental 2 x 2 matrix in Eq. Prom det we obtain A± x — A y2 1 A 0. x ± \Jx2 — Ay1 (3. satisfy the recursion relations In+x = xlny I. if we define <S = cA+ GL\_ c d (3.
A _ = (x24j/2)^ = (4e 2 u. We can easily check now that 11=0. (3.43) and (3. (3. (3. k = l. (3. 2 )5 =2iecu. (3.56) and (3.57) and (3.44). Let us next note from Eqs. we obtain <**»> (&)*(T*?.63) which is consistent with our earlier observations in Eqs.Harmonic Oscillator 43 where c and d are arbitrary parameters.MD. .60) /tf^T^HA^A?).«"» We recall from Eq.38) that A + .H(Ao+A°>Using this in Eq. we obtain from Eq. (3.56) that x = A+ + A_ . (3. (3.61) Using this as well as the forms for S and <S_1 in Eqs. then it is easy to check that (r".55). (3.58).62) = \+ + \_ = x.
(3.((1 + zew)^ .64) Consequently. (3. TV—>oo . sinwT lim e det B = . namely. what we had obtained earlier in Eq.e " ^ ) = ^ ^ . substituting these relations into Eq.62). (3.44 Field Theory: A Path Integral Approach xN = fx + ^/x2Ay2\ (i^)+W N = (l + ieuj + 0 (e2))N viV ~ (1 + iev) N — \jx2 — Ay2 i 11 \ N 2(1££•)2iew\ 2 2\\N = (1 . /• • ~(lieuj)N \iV . we obtain lim e/jvi = lim e e^O AT—>oo 1 e^O N—«x> A+ — A_ ( A £ .A^) lim e J . (3.65) Here T is the time interval between the initial and the final times and this is. of course.52).(e*r . TV—>oo ieu)N) 1 // N—HX3 ^ io.TVv / i^r ' = J .(1 e>o 2zeo.ietu + O (e2)) . that T.
U.66) Once again we see from Eq. A complete determination of the transition amplitude in this case.68) with A and B arbitrary constants to be determined from the boundary conditions. (3. we obtain the transition amplitude for the harmonic oscillator to be U(tf. where the homogeneous solution is of the form xn(t) = Aeiujt + Be'™*. In other words.Harmonic Oscillator 45 In any case.Xi) = ( 3. therefore.66) that the transition amplitude has a generic form similar to the one found for the case of the free particle. (3. (3.5)) mxc\ + mco2xc\ — J = 0. This can be done simply in the following way.67) is defined by the equation . we use the method of Green's function. We recall that the EulerLagrange equation for the present system is given by (see Eq. (3. it is proportional to eHs>XclL>. To determine the inhomogeneous solution. obviously. (367) The solution. Here the Green's function for Eq.69) (3.28) or (3. the classical trajectory is a solution of the equation (£+^)*W = ^ . consists of a homogeneous and an inhomogeneous part and can be written as xci(t) = xR(t) + xx(t).xf.4 The Classical Action ™ T )§e^c]. Namely. (3. would require us to evaluate the classical action for the system.
46 Field Theory: A Path Integral Approach It is clear that if we know the Green's function G(t — t'). the Green's functions that are of fundamental interest are the retarded and the advanced Green's functions.6) that the Fourier transform in time is defined with an opposite phase.i') — • (3. (3. in classical mechanics.70) we obtain (Note from Eq. Therefore. where G(k) is the Fourier transform of G(t — t'). (2. then the inhomogeneous solution can be written as Xl(t) = . But a Green's function that . we must specify a contour in the complex /cplane in order to evaluate the integral. Defining G(tt>) = J^=eik^G(k).) (J^+J^G(t (') = «((*') ™>wJk2+^G{k) = h Consequently. the Green's function takes the form G(t1?)= 1 [jLeikWG(k) r = rJ i^j Ak e ifc(tf) (3J4> A quick inspection shows that the integrand has poles at k — ±UJ. Normally. (3. and substituting these into Eq.71) Ju m The Green's function can be easily determined by transforming to the Fourier space./ ' tit' G(t .72) 6(tt') = J^eik<t*).
LO2 ' .LO2 + ie (3.75) as the Fourier transform of the function which satisfies the differential equation lim+ ( —2 + J1 . (3. in this language. we can think of the Green's function in Eq. for completeness.75) 2 1 1 1 = lim 5^o+ v/^r k + to — i5 k — to + i5 ' where we have defined 5 e 2LO In other words.73)) 1 GF(k) = lim 0+ v ^ 1 k . (3.t') (3. Im k Im k LO LO I x \ LO > X + iS *~ Re k LO x — iS Rek Equivalently.ie J GF(t . correspond respectively to choosing the Fourier transforms as 1 1 GK>A{k) = lim 7 e^o+ v ^ (k ± ie)2 . that the retarded and the advanced Green's functions.76) We note here.t') = 6(t . it corresponds to defining (see Eq.Harmonic Oscillator 47 is of fundamental significance in quantum theories is the Feynman Green's function and corresponds to choosing a contour as shown below.
Thus. W i t h such a choice of contour for the Feynman Green's function. enclosing the contour in the upper half plane. the Feynman Green's function has the form iuj(tt') cMtt') (3. id + ie X X id + ie X * U — ie x u — ie Re k Re A .5^0+ 27 J "T 1 eiw(tt>) — 2iri 2TT 2U ^. we find eik(tt') G()(t0= lim (k + id — iS)(k — u + iS) .79) . enclosing the contour in the lower half plane for t — t' > 0.78) GF(tt') = 9(tt') 2iid + 9(t'ty 2iui (3.48 Field Theory: A Path Integral Approach with the respective contours Im k Im A .f dfc 1 Ju(tt') 2zo.77) 2iid On the other hand. we obtain eik(tt') G(+\tt')= v ' 5+0+ 2?r J lim — f dk 2 (k + to — iS)(k — id + i<5) e iw(tt') = s<. for £ — £' < 0."> 2w J _ eiaj{tt') (3.
Harmonic
Oscillator
49
Using this Green's function in Eq. (3.71), we can now obtain the inhomogeneous solution as
(t)= r M GF{tt'yJti ' rt iw{tt')
m
rtf piui{tt') N
/"nrM
+L
At
'^r «\
(3.80)
J
Thus, substituting Eqs. (3.69) and (3.80) into Eq. (3.68) we can write the classical trajectory as xd(t) =xn(t) + xj(t)
= Aeiu}t + Be~iu>t
(3.81) Imposing the boundary conditions
AeM* + Beivti
_ _ J _ J f di/eMuf)j^
2imto Jti if.
?
Xd(tf)
= Xf
= Ae^f
+ Be~iu,tf
1
2imuj
j t Ju
[ ' dt'e^f^Jit'),
/•*/
(3.82)
we see that we can solve for A and B in terms of the initial and the
50
Field Theory: A Path Integral Approach
final coordinates of the trajectory as A = 1 2i sin LOT
eiutf
(*/<
ft ezs
ivU _
iuitf
^—iuitt
H muj B 1 2z sin
/ Jt (Xieiujtf
rtj rtf
dt'smoo(t'~ti)J(t')
UJT
x/e*"**)
iujti piu>ti e
+
/ mu Jt
dt'smu(tft')J(t')
Substituting these relations into Eq. (3.81), we determine the classical trajectory to be xc\(t) = Xf sin u(t — ti)+Xi sinwT ' ^ r , * , / f dt'J(t')(elujT sin u>(tf — t)
+
1
cosco(t  t')  costu(tf
+U t 
t'))
— ( f dt'jit'y^^
muJ \Jti
+ ftf dt'juy ,iuj(tt')
Jt
(3.83)
We can now derive the classical action from Eqs. (3.2) and (3.1) in a straightforward manner to be mui SW\] = 7T u;T KXi + 2 sin T
X
/ ) C°SUT
~
Xf
2XiX
i\
+ sintuT i
mwsin
I dtJ(t) s i n w ( t / t)+ /*.
I dtJ(t) smco(t  t.
r*
/
J ti
f
sinu(tf  t) sinu)(t' U)J{t'). (3.84)
Jti
dt I dt'j(t)
Harmonic
Oscillator
51
This, therefore, completes the derivation of the transition amplitude for the harmonic oscillator interacting with a time dependent external source. 3.5 References
Feynman, R. P. and A. R. Hibbs, "Quantum Mechanics and Path Integrals", McGrawHill Publishing. Kleinert, H., "Path Integrals", World Scientific Publishing. Schulman, L. S., "Techniques and Applications of Path Integration" , John Wiley Publishing.
Chapter 4
Generating Functional
4.1
Euclidean Rotation
We have seen in Eq. (2.39) that the standard Gaussian integral (where the exponent is quadratic), namely,
Joo
\
a
J
generalizes in the case of a n x n matrix as (see Eq. (3.41))
/_> — {&•
provided A is a Hermitian matrix. In fact, we will now see explicitly that this result holds true even when we replace the matrix A by a Hermitian operator. In other words, we will see that we can write fvrj jlutoiWOViW = N [detO(t)]5 , (4.2)
where 0(t) is a Hermitian operator and iV a normalization constant whose explicit form is irrelevant. To establish the identification, let us go back to the harmonic oscillator and note that the quantity of fundamental importance in
53
54
Field Theory: A Path Integral Approach
this case was the integral (see Eq. (3.13))
I
ftUt{\me\rr^W)
= fvVe^^dtT>^+uj2Mt) with the boundary conditions V(ti) = r,(tf) = 0.
,
(4.3)
(4.4)
The value of this integral was determined earlier (see Eq. (3.25)) to be ... fsincoT\ A V uT
_i2
where T = tf — ti represents the total time interval. We evaluated this integral earlier by carefully discretizing the time interval and calculating the determinant of a matrix (see Eqs. (3.30) and (3.35)) whose matrix elements were nothing other than the discrete form of the matrix elements of the operator in the exponent in Eq. (4.3). This would already justify our claim. But let us, in fact, calculate the determinant of the operator in the exponent of Eq. (4.3) explicitly and compare with the result obtained earlier. The first problem that we face in evaluating the functional integral is that the exponential in the integrand is oscillatory and, therefore, we have to define the integral in some manner. One can, of course, use the same trick as we employed in defining ordinary oscillatory Gaussian integrals (see Eq. (2.39)). Namely, let us define
Pr?e
if/t/dtw2V) [Vri
e%tf!/dtvM&+" 2
/
lim
ieMt)
(45)
Generating
Functional
55
This provides proper damping to the integrand and, as we will see, leads to the Feynman Green's functions for the theory. In fact, as we have already seen in Eq. (3.76), the inverse of the operator in the exponent in Eq. (4.5) gives the Feynman Green's function which plays the role of the causal propagator in the quantum theory. It is in this sense that one says that the path integral naturally incorporates the causal boundary condition. There is an alternate but equivalent way of defining the path integral which is quite pleasing and which gives some sense of rigor to all the manipulations involving the path integral. Very simply, it corresponds to analytically continuing all the integrals to imaginary times in the complex iplane. More explicitly, we let t' = ir, real.
(4.6)
Imt
Ret
With this analytic continuation, the integral in Eq. (4.3) becomes
I
/
f a 3%Sti dtvm&+» )r,{t)
•
^CfdrV(r)(^+^)r,(T)
= N' fvV
e/^
d
^()(^+2)"W .
(4.7)
^ 2 + ^ 2 J ^ n = An^n. (4.22. (4. Therefore.8).7) or .10) subject to the boundary conditions in Eq. (4. (The analytically continued operator has a positive definite spectrum.12) \Tfn Thus.) We can now evaluate this integral and at the end of our calculations. (4. we are supposed to analytically continue back to real time by letting r > T' = it. (4. t real.2) we see that the quantity which we are interested in is det(—g^+a. (4.7) subject to the boundary conditions V(ri) = v(rf) = 0.7) is now a well defined quantity since the integrand is exponentially damped. we see that the determinant of the operator in Eq. this determinant has to be evaluated in the space of functions which satisfy the boundary conditions liji) = V(T/) = 0 .10) are easily obtained to be with n a positive integer. (4. 2 ). The normalized eigenfunctions of Eq. we are basically interested in solving the eigenvalue equation f .56 Field Theory: A Path Integral Approach Here we have scaled the variables in the last step and N' represents the Jacobian for the change of variables.9) From Eq. we have to evaluate the integral in Eq.8) The right hand side of Eq. (4. (4. The corresponding eigenvalues are K= (^— niT \ ) +u. (4. Furthermore. Furthermore.
of course. Analytically continuing this back to real time.28) through a careful evaluation. (3.10) has the form det = B Sinh "(T/ . (3. we conclude that p^w/*?^22) / ^rH^r™ Later. (4.24) and B is a constant representing the first product whose value can be absorbed into the normalization of the path integral measure.13) where we have used a relation similar to the one given in Eq. Therefore. we have identified T = tf — ti with the total time interval. we obtain d2 .Generating Functional 57 (4. Here. This is. . 22 det I ——^ + UJ\ I > det I (& + u2 —^ = A^ffU)=Asmu/T_ oj(tf . as before.? . we will generalize this result to field theories or systems with an infinite number of degrees of freedom. related to the value of the path integral which we had obtained earlier in Eq.U) UJT V .
under the analytic continuation.75)) that in the complex energy plane. Let us note that an analytic continuation is meaningful only if no singularity is crossed in the process.17) . only if we let k° ^k'° = iK. that an analytic continuation from Re k to Im k° is meaningful only if the rotation is anticlockwise. (Although. (3. corresponds to a rotation to Euclidean space. where we leave the dimensionality of spacetime arbitrary. the singularities occur at Irak0 co + i5 X X N u) — i5 It is clear. in higher dimensional field theories it is quite meaningful. (4. z2 = ( i 2 . We know from our study of the Green's function in the last chapter (see Eq.58 Field Theory: A Path Integral Approach Let us next discuss in some detail the analytic continuation to the imaginary time.> ( . namely. Consider a Minkowski space with coordinates x» = (t. Then.x).i T . x ) .( r 2 + x 2 ) . The analytic continuation. in one dimension. K real. (4.16) Therefore. x» = ( t . we note that r is nothing other than a Euclidean time.x 2 ) ^ . it does not make sense to talk about a Euclidean space. therefore. x ) .) The sense of the rotation is completely fixed by the singularity structure of the theory. consequently.
We note that each inner product in the integrand represents a transition amplitude and. can be written as a path integral. then we can insert complete sets of coordinate basis states and write HiXf^flXftit^XH^lxut^H = dx1dx2H(xf.t2\xi. T real. We have obtained the transition amplitude in the form of a path integral as H(xf.41). we can write (for t\ > t2) H{xf.ti)HH{xi. (4.t2\xi.tf\XH(ti)\xi.t2)HH(x2. (4. Here we have used the relation in Eq. therefore.ti)H x H(xi. Combining the products.ti)H.20) ldxidx2xix2H(xf.ti)H = NJvxx{tl)x(t2)eTis^.ti)H. tf>ti>t2>U.ti\x2.19) Let us next consider a product of operators of the form XH(h)XH(t2).ti)H = (4. and evaluate the matrix element H(xf.Generating Functional 59 Since we can represent it follows now that in the complex iplane.tf\xi.21) .2 T i m e O r d e r e d Correlation Functions Let us recapitulate quickly what we have done so far.tf\xi. (1.tf\XH(ti)XH(t2)\Xi. the consistent rotation will be t>1? = ir. (4.tf\XH(tl)XH{t2)\xi. Since t\ > t2.t2)HH(x2.18) 4.ti)H = N fvx eLn SW .ti\XH(t2)\x2.
ti)H = N fvxOtixih)) • • • On(x(tn))e^s^ .t2\XH(ti)\x1. Thus.60 Field Theory: A Path Integral Approach Here we have used the identification x1 = x(t1).22) Similarly. we can write H(Xf. therefore.ti)H = N fvx x(ti)x{t2) d M.23) that the path integral naturally gives the time ordered correlation functions as the moments H{xf. x2 = x(t2). (4.tf\XH{t2)XH(t{)\Xi. (4.tf\XH{t2)\x2.23) In the last step.21) and (4. we note that for t2 > *i. we have used the fact that factors in the integrand such as x{t\) and xfa) are classical quantities and.t2)xH(h)xH(t2) + e(t2 . There are no operators any more.26) Furthermore.ti)x„(t 2 )Mti) • (4.ti)H = / dxidx 2 H{xf. . (4.h)H H{xi.ti\xi.25) In fact. we see from Eqs.tf\T{XH{h)XH{t2))\xi.U)H = N jvx x(ii)x(t 2 ) e* S^ .tf\T(Oi(XH(t1))On(XH(tn)))\Xi. it is obvious now that the time ordered product of any set of operators leads to correlation functions in the path integral formalism as H{Xf. the beauty of the path integrals lies in the fact that all the factors on the right hand side are cnumbers (classical quantities).24) where the time ordering can be explicitly represented as r(x H (i!)x H (i 2 )) = o(ti .ti)H S x H{x2. (4. (4. their product is commutative.
(4. we would like to know the probability amplitude for a system in a state \ipi)H at time ti to make a transition to a state \ipf)H at time tf.ti) fvxe*sW.tf)ipi(xi.27) = N f dxfdxnl>*f(xf.t). (4. Let us note that by definition.tf)il>i{xi. we are often interested in calculating expectation values.26)). we see that the time ordered correlation functions between such physical states can also be written as H <V/r(Oi (*„(*!)) • • • on(xH(tn)))\ii>i)H = N / dxfdxi i/>*f(xf. Namely. Namely. Here we have used the usual definition of the wavefunction. this transition amplitude is given by = / dxfdxiH(i(. In physical applications.t\ip)H = ip(x.ti) x Jvx Oi(x(ti)) • • • On{x(tn))eisW .28) In dealing with physical systems. we have calculated the transition amplitude between two coordinate states. H{x. Following our discussion earlier (see Eq.tf)i/ji(xi. This is what the Smatrix elements are supposed to give. This is simply obtained by noting that ff(^r(Oi(xH(ti)) • • • on{xH itn)))\^i)H = N x Jvx dxfdxii}*(xf.tf\xi.ti) Oi(x(ti)) • • • On{x{tn))e*sW .29) . we are often interested in transitions between physical states.tf)H H{xf.ti)H H&iM^iJH (4. however. (4.3 Correlation Functions in Definite States So far.f\xf.Generating Functional 61 4.
the subscript H signifying the description in the Heisenberg picture. (4.U) Clearly. if we define a modified action of the form S[x.31) S[x. Let us next note that we can generate the various correlation functions in a simple way in the path integral formalism by adding appropriate external sources.tf)Mxi.0] = S[x]. where S[x] defines the dynamics of the system. (4. for convenience. then. J] = S[x] + / then.62 Field Theory: A Path Integral Approach Since the states need not necessarily be normalized. Let us further define {il>i\A)j = N fdxfdxiiP*(xf. In most field theoretic questions one is primarily interested in calculating the expectation values of time ordered products in the ground state and. clearly. dtx(t)J(t). (A\A)J=O JVxe*s^ . one tends to be sloppy about this factor in such cases.32) = N J' dxfdxitfixfJfMxitU) jVxeisW (433) = {A\A) • . consequently. Prom now on let us suppress. we obtain the expectation value to be (TiO^Xviti^'OniXM))) _ nmnoiix^h)) = • • • on(xH(tn)))\^)H H(A\A)H (4.30) fdxfdxi^ixf^f^iixi^JVxOiixjh)) Jdxfdxi^ixfitf^iixi^fVxe^W ••• On(x(tn))e^s^ Note that the normalization constant N has cancelled out in the ratio and it is for this reason that we do not often worry about the explicit form of the normalization constant. Thus.
35) WilxfaMi). (4. therefore. (4. tf)Mxi.Generating Functional 63 It is clear now from Eqs.tf)il>i(xi.32) that (tf >h> SJ(h) = N J dxfdxt # ( * / . that S(ipi\i>i)j SJ(tx) = N U) J Vx { 6 ti) §^ e i s ^ (4.tf)il. (4.36) (4.34) fvx^x(h)e^x'Jl j=o dxfdxiip*(xf. In general. = N fdxfdxitfixf^fWiixuU) It follows.U) / T>xx(ti)e* S[x] (4.29). we have for 6J(h)SJ(t2 = N I J=0 dxfdxi$(xf. where we have used the relation in Eq.31) and (4. Similarly.i(xi.ti) X DX ] \hj SJ(tl) 5J(t2)eh JVx j=o 2 N Jdxfdxitf(xf.37) J=0 .ti) ^ ) xit^x&y^W tyir(*(*iW2))&>.tf)ipi(xi. it is quite straightforward to show that ^ M i l 6J(t1)5J(tn) h (iPi\T(X(t1)X(tn))\7pi).
39) It is clear from our earlier discussion in Eq. (4. That is.tf\T^ifdtX^\xl.ti).38) j=o It is for this reason that (^i\i>i)j is also known as the generating functional for the time ordered correlation functions. let us calculate the amplitude for the system to make a transition from the coordinate state in the infinite past labeled by the .26) that we can also think of this quantity as the matrix element of the operator (xf./(*„) (4.40) = (xf.64 Field Theory: A Path Integral Approach Consequently. (xf. 4.tf\xi.ti)j =N = N jvx fvxe^x^ es s W + i In d* WW) . Let us next take the limit t j —> —CO.4 Vacuum Functional An object of great interest in quantum theories is the vacuum to vacuum transition amplitude in the presence of an external source. (4. we can write (r(x(ti)A:(tn))) _ (rl>i\T(X(t1)X(tn))\iJ>i) _ (ih)n (lPMi)j PtyiWiU <JJ(ti) •••<*.tf\xuU)j =N Jvxe&W (4. The simplest way to obtain this is to go back to the transition amplitude in the coordinate space. tf —> CO .
we can > write lim (xf. We will consider this limit by assuming that the external source is nonzero within a large but finite interval of time.40) lim (xf. (4. We. H\n) = En\n). choose a derivation parallel to that of a relativistic quantum field theory and assume that the ground state energy is zero.44) (We wish to point out here that in a relativistic field theory.tf\T (e* i:dt V JX ) / tj—> —OO if—*oo T—>O0£i —>_ oo \xhU).Generating Functional 65 coordinate x\ to the coordinate state in the infinite future labeled by Xf in the presence of an external source which switches on adiabatically.ti)j = lim lim {xf.) Although for simplicity of discussion we have assumed the energy eigenstates to be discrete. tf—>oo (4.tf\xi. That is. (4. therefore. In such a case. the ground state energy does not vanish in general and in such a case. it is not essential for our arguments.43) Let us further assume that the ground state energy of our Hamiltonian is normalized to zero so that H\0) = 0. (4. En > 0 .41) and we will take the limit r — oo at the end. tf\xi. In quantum mechanics. Lorentz invariance requires which leads to a vanishing ground state energy. for \t\>T. U)j = N fvxen tj—>00 tf—>00 J S«. we can write from Eq. let us assume that J if) = 0. however. Introducing complete sets of energy .x)+Jx) _ (4 42) Alternatively.dt ^{x. the asymptotic limits are not well defined and the derivation becomes involved.
Thus. we can write (0TfCi/co««^o>= lim N ' V J ' ' u^oo tfKX { X / ^ U \ J (xf\0){0\Xi) • (447) V ' The left hand side is independent of the end points and. in this asymptotic limit. In the limit ti — —oo and > tf — oo .ti)j If —>00 = Jdmjx/IOXOIT (e*^dUX^ 0)(0x<) = (x/lOXOlxiXOlT (e^°°oo d * JX )  0 ) . the right hand side must also be independent of the end points.\xi.U) — lim lim 'S^(xf\e~xHtf\ri) x (nT ( e ^ .42) and (4.tf\xi. we obtain ^ limjz/.tf\n)(n\T tfHX> ™>m fe^rd*JX) \m){m.45) where we have used Eqs.r ^ ^ l m X m l e S ^ I ^ ) = lim lim y ^ e .ti)j if—>oo = lim lim y^(xf. (4.46) Consequently.tf\xi. (4. the exponentials oscillate out to zero except for the ground • state. Furthermore. the right hand side has the structure of a path integral .^ " * / " ^ ^ / ^ In) t/>oo ™>m x (nT (e* !T d * J * ) m) (mxi).44). therefore. we obtain lim t j — > —OO (xf.66 Field Theory: A Path Integral Approach eigenstates into the transition amplitude. One can alternately see this also by analytically continuing to the imaginary time axis (Euclidean space in the case of field theories). (1.
48) dt(L(x. (4. therefore. one can construct the Smatrix of the theory and. Let us define Z[J] = eiw^ (4.28)). In quantum mechanics.50) Namely. j=o fvxe*8^ (4. We have already seen in the case of the free particle as well as the harmonic oscillator that the path integral for the transition amplitude is proportional to the exponential of the classical action (see Eqs.49) (4. Let us note that by definition SW[J] = sj(h) J=0 (ih) 1 SZ[J] ~Z\J\ 6 J fa) \j=o W*i)> (4. (2. (4.52) .38) that (ih)n Z[J] SnZ[J] 5J{ti)SJ{tri (T{X(tx)X{tn))). Z[J] is correspondingly known as the vacuum functional or the generating functional for vacuum Green's functions. therefore.51) or.47) also as (0T (e^°°oo d t J X ) o) = (00)j = N [vxe&W without the end point constraints and with oo . we are often interested in various statistical deviations from the mean values.54) and (3. it follows from Eq.Generating Functional 67 and we can write Eq. Z[J] generates time ordered correlation functions or the Green's functions in the vacuum.x) / oo + Jx). Let us note that if we define Z[J] = (0\0)j = N then. It is for this reason that W[J] is also called an effective action. If one knows all the vacuum Green's functions. (4. In quantum field theory. these correlation functions or the vacuum Green's functions play a central role. This can be obtained in the path integral formalism in the following way. solve the theory. W[J] = ih]nZ[J\.
(T(X(t3)X(tl)))(X(t2)) + 2(X(t1))(X(t2))(X(t3)) = (T ((X(h) (TiX^XfaVHXfa)) (T(X(t2)X(t3))){X(tl)) (4. (4.(X(t2)))(X(t3) (X(t3))))). W[J] can still be shown to generate various . obtain (ihf (ih) 83W[J] 5J{h)SJ(t2)dJ(ts) j=o 63Z[J] 1 Z[J] 6J{h)SJ(t2)5J(t3) 5Z[J] 6 J fa) 52Z[J] Z2[J] SJ(h)5J{t2) 1 62Z[J] Z2[J] SJ(t2)6J(t3) 8Z[J] 6J{t3) 1 52Z[J] Z2[J] 5J(t3)6J{h) SZ[J] 5J(h) + 2 6Z[J] SZ[J] SZ[J] \ Z3 SJih) 5J(t2) 6J(t3) J  J=Q = (TiXMXWXfo))) . we note that (ih) S2W[J] SJ(h)SJ(t2) (ih) j=o = S2Z[J] 1 Z[J] 5J(h)5J(t2)  SZ[J] 5Z[J] \ Z [J] 6J(h) SJ(t2) J 2 j=o (T(X(tl)X(t2))) (T((X(tl) (X(tl))(X(t2)) (X(t2))))).(X(h)))(X(t2) We recognize this to be the second order deviation from the mean and we note that we can similarly. .(X(tl)))(X(t2) We can go on and the expressions start to take a more complicated form starting with the fourth functional derivative of the effective action VF[J]. Next.53) . However.54) .68 Field Theory: A Path Integral Approach where (• • •) stands for the vacuum expectation value from now on.
we have (X(h)) = (ih) 1 SZ[J] Z[J] 5J(h j=o (4.53) that (T(X(tl)X(t2))) = (ih)2 1 52Z[J] Z[J] dJ(h)8J(t2) d2W[J] 6J(h)dJ(t2) J=0 = (ih) (4. in this case. In this case. we have Z[J] = N where °° / oo fvxe*s[x'J] /l dt I mx \^ 2 1 — mu x ^ 2 2 (4. Therefore. In quantum field theory. we obtain from Eqs. we can write Z[J] = N fvxeif«>dt^m*2zmu2x2+Jx) = lim W I Vxe^ e^0+ J 5 (4.55) + Jx Obviously. (4.57) J=0 Let us also note that because the action for the harmonic oscillator is quadratic in the variables.Generating Functional 69 statistical deviations and their moments. Let us next go back to the example of the harmonic oscillator which we have studied in some detail.50) and (4.W M t ) ) .H ^ . for the harmonic oscillator. W[J] is known as the generating functional for the connected vacuum Green's functions.58) °°At M W ^ .^ W .56) NjVxx(ti)eisfc NfVxetsW This vanishes because the integrand in the numerator is odd.
62) = (ihf ~ ) GF(h .if) = 5(t .57)) <T(X(tx)X(t2))> = {ihf Z[J] 52Z[J] SJfatfJfa) j=o (4. (4.t2) = GF(h am) m . for the harmonic oscillator.61) Consequently. In other words. we can define 1 1"°° x(t) = x(t) + — / dif GF(t . the twopoint time ordered vacuum correlation function. namely. we have (see Eq. This is a general feature of all quantum mechanical theories. . gives the Feynman Green's function.60) We now obtain in a straightforward manner S2Z[J] <5J(ti)<JJ(t2) j=o /im GF(t!i2)Z[0] (4.t2).59) and the generating functional will then take the form Z[J]= lim JV x e 2hm / /Mef^ d f i W(p+ u 2 ^) ffZedtdt'JWGHtt'W) JJ oo = lim N e>Q+ det (TPT +uj2 dt'2 it) X e 27k J/~o= d i d t ' JWF{tt>)J(t>) = Z{0] e»^tt°°dtdt' J(t)GF(tt')J(t') (4. that the twopoint connected vacuum Green's function is nothing other than the Feynman propagator of the theory. (3.76)) lim+ f —j + J1 .70 Field Theory: A Path Integral Approach Let us recall that (see Eq.if) Using this.if) J {If). ^ J — CO (4. in the present case.ie J GF(t .
A > 0. In such a case.64) = N [vxey™°°dt^m±25m"2x2$xi+Jx\ If we write / oo \ where >Q[x.) The Gaussian (recall the free particle and the harmonic oscillator) is the simplest of the path integrals which can be exactly evaluated. the quantum mechanical system corresponding to the Lagrangian L = . there are only a few path integrals that can be exactly evaluated. (4. the manifestation of this lies in the fact that the path integral corresponding to the Lagrangian for this anharmonic oscillator cannot be evaluated exactly and has to be calculated perturbatively in the following way.mu2x2 + Jx j . However. J}= dtx4.66) dt( mx2 . use perturbation theory and calculate corrections to the unperturbed system in a power series in A.xA. oo ^ (4. (Fortunately. similarly. In other words. we also know that if we perturb the harmonic oscillator even slightly by an additional potential.65) (4.muj2x2 . the problem cannot be analytically solved. of course.m i 2 . Let us introduce an external source and write the vacuum functional for this theory as Z[J] = N fvxe^x^ (4. there is a one to one correspondence between the quantum mechanical problems that can be analytically solved and the path integrals that can be exactly evaluated. .63) is impossible to solve exactly even when A<C 1.Generating Functional 71 4. say a quartic potential.5 Anharmonic Oscillator Just as there are a handful of quantum mechanical problems which can be solved analytically. we. In the Feynman path integral approach.
This is perturbation theory in the framework of Feynman's path integral.t t / " c o d * ( . then we note that SS0\x. In other words. J] .^ 4 t ) ) 4 ) z0[J]. we obtain (4.68) Now.60) that it has the form Z0[J] = Z0[0] efflfe//"««'««" JCOG^^*")^'). we can Taylor expand the first exponential and we will be able to obtain the vacuum functional as a power series in A. . fVxe^SoM (4.J]. Consequently. . .69) = where ZQ[J] is the vacuum functional for the harmonic oscillator interacting with an external source. i. (4.70) z[j] = z0[o] X e2ki \e&Ir°°dt{ih*m)4' H^t' At" J{t')GF{t>t")J(t") _ (471) It is clear that if A is small. we can use this identification to (eS/00**4) (e£l°°dt{ih*m)4) em*. (4. We have already seen in Eq. for weak coupling. all the vacuum Green's functions can also be calculated perturbatively. write Z[J] = N jvx x(t).72 Field Theory: A Path Integral Approach is the action for the harmonic oscillator in the presence of a source. operationally we can identify 6 6J(t) when this acts on S0[x.e.J\ = N fvx e* s »M] = (e&nod*(%^>4)iV ^ e .. Substituting this back then.
. Huang.Generating Functional 73 4. "Secret Symmetry". ibid 91. Schwinger. . Rev.. World Scientific Publishing. 713 (1953). Leptons and Gauge Fields". Erice Lectures (1973). "Quarks.6 References Coleman. 914 (1951). K. J. Phys. S. 82..
.
(5. Let us discuss one that is the most intuitive. The operators describing bosons. we are assuming that h = 1.Chapter 5 P a t h Integrals for Fermions 5. there are two kinds of particles in nature. Let us recall that the bosonic harmonic oscillator in one dimension with a natural frequency u has a Hamiltonian which.1) Here. for simplicity. The symmetric structure of the Hamiltonian. a^ = 1 . (52) B with all others vanishing. obey commutation relations whereas the fermionic operators (i. namely. They are described by quantum mechanical operators with very different properties. operators describing fermions) satisfy anticommutation relations. bosons and fermions. The creation and the annihilation operators are supposed to satisfy the commutation relations aB.. the states must have a symmetric form. As a preparation for such systems. namely. 75 . takes the form HB = 2^ ( a s « s + aBaBj .1 Fermionic Oscillator As we know.e. written in terms of creation and annihilation operators. the fermionic oscillator. There are many ways to introduce the fermionic oscillator. in this case. is a reflection of the fact that we are dealing with Bose particles and. consequently. for example. let us study a prototype example.
let us note that if we identify the operators aF and aF with the annihilation and creation operators for such a system.76 Field Theory: A Path Integral Approach Fermionic systems. Let us assume. therefore. let us try a Hamiltonian for a fermionic oscillator with frequency u> of the form HF = — (aFaF .3) If aF and aF were to satisfy commutation relations like the Bose oscillator. 2 / x \2 dpi dp = dpdp + dpdp = 1— dpi. let [aF. (5. then we can define a number operator as usual as NF = aFaF. Namely. if we had &F. (5. we can rewrite the fermionic Hamiltonian in Eq. the anticommutators are by definition symmetric.4) with all others vanishing. on the other hand. ctF (5.aFaFj . the particles must obey FermiDirac statistics. then using this. (5. To see this.5) In other words. therefore. dp (5.7) . Therefore. have an inherent antisymmetry.6) is that in such a system.)2 = 2(4) 2 = o. that the fermionic operators aF and aF instead satisfy anticommutation relations. dp (4) 2 + (4.aF}+ = aF + aF = 2aF = 0. in such a case. there would be no dynamics associated with the Hamiltonian. (5. namely.6) In contrast to the commutators.3) to be HF = — (aFaF — (aFaF + 1 j j — tjQ ~~2' (5. An immediate consequence of the anticommutation relations in Eq. dp.
3) as HF — — (aFaF . aF and NF can now be calculated in a straightforward manner.12) .6) we note that NF — aFaFapaF. = at (5. that we can at the most have one fermion in a given quantum state. (5. dp aF = a. we have NF\nF) = nF\nF). Given this. dF = aF dp. the commutation relations between aF.10) Consequently.1. HF\0)=U(NF^J\0) = ~\0).aFaF) J = UJ aFa = w JV. This shows that the anticommutation relations are the natural choice for a fermionic system.dFdp &p.Path Integrals for Fermions 77 From the anticommutation relations in Eq.NF] = aF. [aF. namely. The ground state with no quantum is denoted by 0) and satisfies NF\0)=0.NF dp. (5. NF (NFl) = 0. (5.8) Therefore.11) with nF = 0. let us rewrite the Hamiltonian for the fermionic oscillator in Eq. the eigenvalues of the number operator can only be zero or one.9) Furthermore. (5.(1 . if we assume an eigenstate of NF to be denoted by \nF).ata dp. This is the reflection of the Pauli principle or the Fermi statistics. = aF (1 — aFaF J aF = aFaF = NF or. (5. (5.
implies that for any given i. (5. n .. .16) This.14) It is clear from the anticommutation relations in Eq. . df = 0. then they satisfy 0i0j +6361 = 0. 40> = 1>. (5. (5. we cannot directly write down a Lagrangian for the fermionic oscillator with the usual notions of coordinates and momenta. HF\1)=U>(NF^)\1) = ^\1). if Oi.j = 1. .. the state with one quantum is denoted by 1) and satisfies NF\1) = \1). (5.78 Field Theory: A Path Integral Approach Similarly. they have very uncommon properties and let us note only some of these properties which we will need for our discussions.6) that 41) = OUF0)=0. . (517) .13) The ground state is annihilated by aF and we have aF0)=0. in particular. For example. 5.15) Therefore.2.. the Hilbert space.. 2 . i. is two dimensional and we note here that the ground state energy has the opposite sign from the ground state energy of a bosonic oscillator. in this case. i not summed. we need the notion of anticommuting classical variables.i = l . Obviously. defines a set of Grassmann variables (classical). (5.2 G r a s s m a n n Variables Since fermions have no classical analogue.n. As one can readily imagine. Such variables have been well studied in mathematics and go under the name of Grassmann variables.
Furthermore. then it has the simple Taylor expansion f{6) = a + b9. the Grassmann variables are nilpotent.' =0  (522) In other words.«A . the conventional commutation relation between derivatives and coordinates now takes the form 9 f) wfi+d>wr5ij (5 23)  The notion of integration can also be generalized to Grassmann variables.(S) * "' (I?) . Denoting by D the operation of differentiation with respect . dOi d9j + d6j dOi ' *  ' These derivatives. are nilpotent just like the variables themselves. w»> .(s0 = ' ~ " • ' whereas a left derivative would give h s. the derivatives.te s h (1 59 k «w . Namely. the derivatives have to be defined carefully in the sense that the direction in which the derivatives operate must be specified. behave quite like the exterior derivatives in differential geometry. a right derivative for Grassmann variables would give w. for example.w • (520) Thus.18) Since 6iS are anticommuting. We note in particular that for a fixed i c^2 an. in this case. the derivatives with respect to these variables also anticommute. Thus. we will use left derivatives. This has the immediate consequence that if f(9) is a function of only one Grassmann variable."> (ft) . in fact. the sense of the derivative is crucial and in all our discussions.Path Integrals for Fermions 79 In other words. Let us note that like the Grassmann variables. (5.
24) Namely. then. we note that these must satisfy the relations ID = 0. (5.22)). must give zero upon differentiation. for a function of a single Grassmann variable. (5. (5. DI = 0. (5.28) /*/«•)T^i/.6 9 = 1.80 Field Theory: A Path Integral Approach to one Grassmann variable and by / the operation of integration.27) This is an essential difference between ordinary variables and Grassmann variables and has far reaching consequences. in this case. An immediate consequence of the definition of the integral in Eq. Thus. the integral of a total derivative must vanish if we ignore surface terms and furthermore. it satisfies the above properties and hence for Grassmann variables integration can be naturally identified with differentiation.26) is that if we redefine the variable of integration as 9' = a 9. Note that since differentiation with respect to a Grassmannn variable is nilpotent (see Eq. Namely. an integral. we obtain a^O. (5.25) Jd0f(0) = yyi. (5.26) This immediately leads to the fundamental result that for Grassmann variables 16.'(i) <» " . we have I = D. we have (5. being independent of the variable.
Furthermore. fd66(0)f(0)= [d00f(0)= = Jd6ea=^ fdO0(a + b6) = a = f(0). This result can be easily generalized to integrations involving many Grassmann variables and it can be shown that if d[ = ai5Bj .30) with det dij ^ 0 and repeated indices being summed.27).32) That this satisfies all the properties of the delta function can be easily seen by noting that ! 69 5(6) = f d6 8=l.34) Here we have used the nilpotency of the Grassmann variables. then n „ n / Hd0i f(9i) = (det aij) / J ] dO'J ( a ^ .31) We can also define a delta function in the space of Grassmann variables as 6(0) =0. (5. then. (5. consistent with the rule for change of variables for the Grassmann variables. (5. Namely. we note that the Jacobian in the case of redefinition of Grassmann variables is the inverse of what one would naively expect for ordinary variables. which follows from Eq. . (5. In addition.33) (5. if f(0) = a + b9.Path Integrals for Fermions 81 But this is precisely the opposite of what happens for ordinary variables. (5. if 9(0)= off.
82 Field Theory: A Path Integral Approach then.37) id where we are assuming that Ci and c* are independent Grassmann variables.. then y*dCe** = JdC(l + K0) = — (1 + i(0) = i6 = i6(6). and 9? = e: + c*jM^. we obtain 5(g{O)) = a0 = a6(O) = ^6(e). we obtain using Eqs. then. 0£) and analyze the integral 1 = IIIde* J d0 i eWW'+'SWM . namely. (5. Let us consider two sets of independent Grassmann variables.• •. (5..31).38) (5.62. (5. Note that if we make the change of variables (we are assuming that M~l exists) 0. the convention for summation over repeated indices is always assumed. An integral representation for the delta function can be obtained simply by noting that if £ is also a Grassmann variable.* * * ' ) = det My / 1 ] <W Mj eW*** V<*$ ..36) Let us next evaluate the basic Gaussian integral for Grassmann variables.39) = f I I d0i d°i e(fl? W ^ .35) where g(6) is assumed to be Grassmann odd.. (5.. (61.= MijOj + a or.38) and (5.39) that 1 6i = Mrj\e'jcj). (5. Furthermore. 6n) and (O^O^. (5.
let us consider the Lagrangian L = ^[^h)^[M]. (541) where tp and tp are two independent Grassmann variables. Indeed. This leads to an essential difference between quantum mechanical bosonic and fermionic theories. (5.40) Here N is a constant and we note that the Gaussian integral in the case of Grassmann variables has the same form as the integral for ordinary variables except for the positive power of the determinant.Path Integrals for Fermions 83 = det Mtj f J ] d#* < } W J CTW+^WCM^) ij = det My / n d0*' d*J. continue with the first form of the Lagrangian. (5.42) We will. This is a first order Lagrangian and one can define canonical conjugate momenta associated with the Grassmann variables ip and tp as usual . we can now ask whether it is possible to write a Lagrangian for the fermionic oscillator. 5.3 Generating Functional With all this background on Grassmann variables. Eq.^]. however. (5. e «:'»JK^^ = N det My ec»*M^lcj. namely.41). Quite often one eliminates a total derivative to write an equivalent Lagrangian also as L= #^$.
) With the convention of left derivatives.^ ) . (5.45) With this identification of ip and ip with the annihilation and the creation operators respectively. We will. the hermiticity properties for these variables now follow. that this simple Lagrangian will yield the Hamiltonian of Eq. (5. (5.3) for the fermionic oscillator if we identify ip = aF. the proper definition of the Hamiltonian (which is only a function of coordinates and momenta and which also leads to the correct dynamical equations of motion) is H = tpUf + ^ % ' — .84 Field Theory: A Path Integral Approach (Such a system is known as a constrained system and one should use the formalism of Dirac quantization to quantize such a system. Tp = ai. we note that •^ = ip.48) . the number operator defined as NF = aFaF = ipip .46) We also note that with this convention.L %— % / — • — \ to — + = . $ = ip • (5. not go into these technical details which are not quite relevant for our present discussion. however. Namely.47) is Hermitian since Art = (^)t = ^ t = ^ . (5.g ^ . f [^'^1 (5. therefore.2 yw ~ ^ ) =  [ « ] .44) It is clear.2 ^ ~ 2 \ ~ ^V "2 ^ ' ^ + = g ( ^ .
even when we are dealing with the reality questions of ordinary Grassmann variables (not operators). (rixT = x V (5.49) In other words.Path Integrals for Fermions 85 Since Grassmann variables have no classical analogue.. The complete action for the oscillator.51) where we have denoted the sources for ip and •tj) by fj and rj respectively.g ( ^ . Namely. (5. = .^ ) . we follow the above prescription in defining complex conjugation.^t>M = L. we can write the vacuum functional for the fermionic oscillator as Z[rj. (552) with S[^M = jdtL = jto(^(U h) . This is the only way a consistent transition is possible from a classical to a quantum Lagrangian involving fermions. let us note that the Lagrangian for the fermionic oscillator given in Eq. i>] = g ( ^ .fj = N I V^ViP e l s W>. in this case. then. even classically.50) With this. With this prescription. fj] = (0\0)r. has the form oo / oo dt ( # + H).41) is Hermitian (real). we continue to treat Grassmann variables like operators.V^] . (5. we define for any pair of Grassmann variables r\ and x.f [<M) • (553) . (5.dip) ~ 2$.
namely. since e is now a Grassmann variable. (Incidentally. (557) As we have mentioned earlier. the limit e — 0 is redundant since e2 = 0 and • the highest power of e in the expansion of the functional is linear.14).) Thus. we will assume the hermiticity conditions for the sources similar to the ones given in Eq. we essentially .F Mt))] e"1.41) for the fermionic oscillator and note that when fermions are involved.lim F { m + e5{t ~ * /)} ~ F { m ) (5 55) However. we will always work with left derivatives even when dealing with functionals involving fermionic variables.86 Field Theory: A Path Integral Approach Once again. (5. we note here that for polynomial functionals. 5. (5. The definition of the functional derivative is still the same as given in Eq. there are right and left functional derivatives with respect to fermionic variables. (5.52) is Hermitian. the position of e _ 1 in the expression defines the direction of the derivative.4 Feynman Propagator Let us next go back to the Lagrangian in Eq. (5. Secondly. Just as an ordinary derivative with respect to a Grassmann variable is directional.56) whereas a right functional derivative would be defined as SF S^f = J?0 iF W ) + ^(t .0) . (5. namely.54) in order that the complete action in Eq. rf — r?. similarly. one can think of e _ 1 simply as J^.46). (1. rf =V. SF i m ) . a left functional derivative corresponds to defining SJ 1 ^ = Jim e" 1 [F (m + eS(t t))F (^(t))] .
(5 e = ev 3 = ^ r^. This is another reflection of the fact that the Grassmann variables inherently behave like operators.58) 0 °* .l We note. * # = •/ ?> ip [Z = (tpipiptp) = [ip. 1 03=1 (5.tp] .V o . that 4 = ip — ip \ o \{i> itp ^ \ = i (ipip + iptp J = i (Tpip — 'ipip 1 . (5.59) . then. = {rkn + fjip) = m . namely.Path Integrals for Fermions 87 have a matrix structure. Let us define the following two component matrices. Here 03 denotes the Pauli matrix.
88 Field Theory: A Path Integral Approach where we have used the anticommuting properties of the Grassmann variables.60) We note here that. (5. we could have written the source term also as d**e. therefore. 27T J (5. (5. We know that we can evaluate the generating functional if we know the Green's function associated with the operator in the quadratic term in the exponent (see Eq.61) = N JVV JS™*'^***")*+©*). (5. study the equation (ia3^. we can write the vacuum functional as (h=l) Z[S] =N fv^eiS (5. Let us.63) .59).t') = 5(t . It is now straightforward to show that the action for the fermionic oscillator in Eq.t').52) can be written as = [ dt(^(i<T3^u)* + ey\. (5. using Eq. 6(t .62) This is clearly a matrix equation and it can be solved easily in the momentum space. J —( oo In other words..uA G(t . (5. O and 6 are not really independent and.40)). Thus.fdk e"^**'). therefore.t') = !. we define G(t .0 = J J L G(k) e~ik^ .
(3.75)) GF(t ~ t>) = lim J . . the Green's function in Eq. (5. c****) v .OJ + le (5.76)) lim f iaz— .62). (5.t').t') = 5(t . (3.68) . kA — u* (5.t') = lim !.u + ie) GF(t . . e»0+ 27T J er3£. will lead to the propagator (see Eq. Substituting these expressions back into Eq. in this case.f dk 2 °*2k + U.64) Consequently./ dfc — — ^ e . we obtain 1 2TT (a3k .(**'). (5.66) This can also be written in the alternate form ifc GF(t . G(k) = a3k + LU V2TT k2 U>2' (5.w)G(ife) 1 1 2ir o3£.w 1 l 2TT or.65) The singularity structure of the integrand is obvious and the Feynman prescription. e>0+ \ dt J This defines the Feynman propagator in the present case.Path Integrals for Fermions 89 where G(k) is a matrix in the Fourier transformed space.62) has the form oit^i^mc2TT ifc(tt') y dk ^ ± ^ ett(tf). (5.67) and satisfies the equation (see Eq. ^ o + 2?r J lim k UJ + ir] g = J _ [dk 3fc + " ett(tf).
75) arises from the change in the order of the fermionic variables. in this case.0(t2 .' £ . (5. (5.70) This is because the integrand in Eq.t 2 )*(*i)*(*2) . (4.69) 8Z[Q] = 0. which anticommute. Going back to the vacuum functional.) Incidentally. /dt*(«73^w+.57) for h — 1. we note that since the exponent is quadratic in the variables .e)# (5.72) 69(h) e=e=o Thus. (5. we see that if we write Z[Q] = e^[ @ ] then.74) (Compare this with Eq.t i ) * ( t 2 ) * ( t i ) • (5. (5. we can obtain from Eq.71) Similarly. the 1point function can be obtained to be SZ[&] = lim N J V9 (19(h)) €2 69(h) e=e=o e .> .70) is odd under * .0 + = 0.+i e )*+e*) _ ( 5 69) Therefore.90 Field Theory: A Path Integral Approach Going back to the vacuum functional. (5. (5. in the case of fermionic variables. in the second term. is denned as T (y(h)y(t2)) = 0(h . * >• . we note that we can write Z[Q] = lim+ N Ivy e </di(§*(ia 3 &u.* . (5. time ordering.75) The relative negative sign between the two terms in Eq.73) H) se(h)s@(t2) 0 = @ = o = 62w[e] ( l) S2Z[@] Z[0] dS(h)S@(t2) e=e=o I {T(y(h)V(t2))).
Therefore. .Path Integrals for Fermions 91 (namely. 5. we have. Then. (5. (3. it is a Gaussian integral). (4.77) Therefore. (5. it can be explicitly evaluated using Eq.78) This again shows (see Eq. In this section.62)) that the time ordered twopoint correlation function in the vacuum gives the Feynman Green's function.2).5 T h e Fermion D e t e r m i n a n t The fermion action following from Eq. As we have argued earlier. j^iGF(ht2)z[0} (5. we will evaluate the generating functional for the fermions in the absence of any sources. let us take the dynamical Lagrangian of Eq. the generating functional can be easily evaluated.42) is quadratic in the dynamical variables just like the action of the bosonic oscillator in Eq. using Eq. (5. this is a general feature of all quantum theories. e=e=o (5. (5.40) to be Z\e]= lim N e0+ J fVmeiS^(. For simplicity. (5. It is obvious now that s2z[e] se(ti)se(t2 = iGF(t1t2)Z[0].42).74) e=e=o = (*)* = iGF(t1t2).41) or Eq.™zii"+^+Q*) = JV e !// d *i d *2©(*i) G H*i*2) e (k) = Z[0] e"5 / / d t l d t 2 ©(*i)GF(ii*2)e(t 2 ) } (576) where Z[0] represents the value of the functional in the absence of sources.
.17) and (2.21) along with the earlier observation that the variable tp represents the momentum conjugate to ip.81) Here we have used the midpoint prescription of Weyl ordering as discussed in Eqs. we should discretize the time interval as in Eq. (2. (2. .=l(*( f^)U>i>n{ J2"1)) . (5. (5.79) Here we have used the anticommuting properties of the Grassmann variables to rewrite the commutator of the fermionic variables in a simpler form.2. n= 1. The constant N.18). defining the intermediate time points to be tn = ne.79) also as Z[0]=N f Vi>V^ J So dt <iW<»H>). where the infinitesimal interval is defined to be T we can write the path integral to be Z[0] = lim N / dipi • • • d^jvidV'i • • • d0jvi e^O J N—>oo .. (5. to be Z[0] =N f VipVtp e iS[tp.. We can once again define tf —U = T. as the time interval and translate the time coordinate to write the generating functional of Eq.ip] = N f Vi>V^ e* • # dt (*W«W). in the absence of sources. (5. .N1. representing the normalization of the path integral measure is arbitrary at this point and would be appropriately chosen later.80) To evaluate the path integral. Thus.92 Field Theory: A Path Integral Approach we can write the generating functional.
84) e^0 J .81) can be written out in detail as E (1 + — )^„Vn + [ 1 + ~Y J V'iV^AT n=l JV1 ^ ' . 1 2~ ) ^N4>NI Thus. Yl ( l ~ ~2~) Mnl n2 ieuj\ ~ ( 1 "^] V^O (5.81) also as Z[0] = lim N f #d^ e^B^+jr^TJHi+^NM iV—>oo ? (5 . (5. defining (iV — 1) component matrices </> = ^ \ipNiJ lecu J o / o \ J ieu> 0 (5.Path Integrals for Fermions 93 The exponent in Eq. (5. / 16LJ \  .  .„ leu.82) zeu.83) we can write the path integral of Eq.
85) \.89) . has the form Z[0] = TV e ^ e<e = N e ? e "il>Ntli0ipNil)N) iuT (ei»T$f1.88) In the continuum limit of e — 0 and N — oo such that Ne — T. e—>0 jV—>oo The matrix B has a very simple structure and we can easily evaluate the determinant as well as the appropriate element of the inverse matrix which have the following forms.km)1 JV2 „"•' JV2 ^JV1. with i .84) can now be easily evaluated using Eq.( 1 + i f i ) ^ ^ ) .f) (5.. (5.M " / i . therefore.i$f4. i / 1 +zeoA ^J (5. l .1 .I l ) xNl ~ f1 + yujNl (5.86) V= . A r . (5.l B ^ .2  The path integral in Eq.87) = lim N d e t B e ( J . the > > path integral.40) and the result is Z[0}= Urn N JV—>oo detBe(JTB~1J(1+^N4>N) (5. detB = x JV _ 1 Vl XEUJ JV1 = 1+ v h .94 Field Theory: A Path Integral Approach where we have denned a (N — 1) x (N — 1) matrix B as (x 0 0 0 \ v x 0 0 ••• y x 0 0 B = 0 y x 0 (5.
6 References e(^ T ^VvV>/). Cambridge Univ. F. Press. "Supermanifolds". $N = '4>f (5. DeWitt..91) Berezin. Academic Press. . for simplicity and for future use.90) We choose.Path Integrals for Fermions 95 Here we have identified $o = A. the normalization of the path integral measure to be N = 1 so that the free fermion path integral takes the form Z[0] = e^ 5. "The Method of Second Quantization".. (5. B.
.
(See Eqs.aF}+ = 0 = aF.av = 1.3) with all others vanishing. aB 1. (5.2) and (5.1 Supersymmetric Oscillator We have seen in Chapters 3 and 5 that a bosonic oscillator in one dimension with a natural frequency u is described by the Hamiltonian HB = ~2 ( a ^ f l s + aBa] B) = w (asaB + 2 ) ' t61) while a fermionic oscillator with a natural frequency u is described by the Hamiltonian HF = — (aFaF — aFaF) = w ( aFaF — . (6.Chapter 6 S u p er sy m m e t r y 6.) The creation and the annihilation operators for the bosonic oscillator satisfy a s . (5.3). For the fermionic oscillator. (6.6)) [aF. on the other hand.2) Here we are assuming that h — 1. CiF (6. the creation and the annihilation operators satisfy the anticommutation relations (see Eq.4) .) . 97 flj?.
5) that the constant term in the Hamiltonian for this system has cancelled out. . Let us next consider a system consisting of a bosonic and a fermionic oscillator with the same natural frequency u>. nB = 0. (6. NF\nF) — nF\nF). let us define \nB.= u [aBaB +aFaF) .11) that the eigenvalues for the fermion number operator are 0 or 1 consistent with the Pauli . The Hamiltonian for this system follows from Eqs.2) to be H = HB + HF = — (aBaB + aBa*B + aFaF — aFaF) ( t 1 t 1 = w aBaB + . nF) = \nB) ® \nF).9) (6. This is known as the supersymmetric oscillator. (6. Consequently.+ aFaF . then. If we define the number operators for the bosonic and the fermionic oscillators as NB = aBaB.1. (5. (6.l.8) Here we are using our earlier result in Eq. we can write the Hamiltonian for the system also as H = LU(NB + NF).7) that the energy eigenstates of the system will be the eigenstates of the number operators NB and /VF. where NB\nB) = nB\nB). 2 . (6. . (6.1) and (6. nF — 0. (6.6) NF = aFaF .98 Field Theory: A Path Integral Approach Let us note here (as we have pointed out earlier in chapter 5) that the ground state energy for the bosonic oscillator is ^ whereas that for the fermionic oscillator is — ^. . (6.5) We note from Eq. .7) It is clear from Eq.
H]= aBaF. all other energy eigenstates of the system are doubly degenerate. the ground state is assumed to satisfy aB\0) = 0 = a F 0). (6.0 = 0. .10) that the ground state energy of the supersymmetric oscillator vanishes.1. (6.0) have the same energy for any value of n # . We also note from Eq.8) and (6.13) We can show using the commutation relations in Eqs. We also observe from Eq.12) The vanishing of the ground state energy is a general feature of supersymmetric theories and as we will see shortly it is a consequence of the supersymmetry of the system. Prom Eqs. + aFaF) dp ~r Q>B @>Fi 0>F Q'F = to \ a . Let us next consider the following two fermionic operators in the theory.10) with nB = 0 . £•0. Namely. 1 . .u(aBaB a'j^a.nF).Supersymmetry 99 principle while the eigenvalues for the bosonic number operator can take any positive semidefinite integer value. (6.4) that (The bosonic operators commute with the fermionic ones. is again a consequence of the supersymmetry of the system. The degeneracy in the energy value for a bosonic and a fermionic state. (6.) [Q. Q aFaB (6. (6. (6. and nF = 0.11) Incidentally. 2 . (6.7). (6.nF) = EnBynF\nB. the states \nB. ( ^ — QiB(LF . namely.nF) = uj(nB + nF)\nB. 1) and \nB +1.9) we note that the energy eigenvalues for the supersymmetric oscillator are given by H\nB.10) that except for the ground state.3) and (6. as we will see.
aB = aBaB + aFaF = to H.15) (6. we note from Eq.15) and (6.16) that (0\H\0) = u (0\QQ + QQ\0) = 0. (6. define conserved quantities of this system (charges) and would correspond to the generators of symmetries in the theory. (6.17) Namely. and similarly. Q and H define an algebra which involves both commutators and anticommutators. the ground state energy in a supersymmetric theory vanishes. Q and Q.16) that the operators Q. Furthermore.14) The operators. then it follows from Eq. (6. Such an algebra is known as a graded Lie algebra and defines the infinitesimal form of the supersymmetric transformations. (6. We also note that [Q. namely. (6. An immediate consequence of the supersymmetry algebra is that if the ground state is invariant under supersymmetry transformations. therefore. (6. (6.18) (6. [Q.13) that Q is really the . H] = aFaB.16) Thus. (6. if (see Eq.12)) Q  0 ) = o = Q0>.100 Field Theory: A Path Integral Approach to i—aBaF + aBaF\ = 0.u>(aBaB + aFaF) 0.Q}+ = d^dp j CLpClQ = aB CI pi dF aB a'f aB. we see from Eqs.14).
NF ClftQjp) (XpCip = ai — (XQCIF &F) &F aF (6. [Q. we have _. it follows from Eq. .nF)= \ *fn~B~~+l\nB + l. It now follows that for nF \nB. Consequently. (6.Supersymmetry 101 Hermitian conjugate of Q..2.nF . its expectation value in any state must be positive semidefinite.NB] = Q.. [Q.I) \ 0 ifnF^0.. Let us next analyze the effect of Q and Q on the energy eigenstates of the system. We note from the commutation rules of the theory that [Q. aBo. (6. aB aF (1B<XF — LJ . Q\nB.. (6. if nF = 0 . therefore.16) that in a supersymmetric theory H is really a positive semidennite operator and.21) where we recognize nF = 0.NF] = Q. we can think of Q as raising the bosonic number nB while lowering the fermionic number nF by one unit whereas Q does the opposite. . (6.1 and nB = 0.NE aBaF.20) In other words.1.19) — \cl j and similarly [Q.22) .B J — aB.nF) 0>.
. We also note here that fix) can be any chosen monomial of x at this point.nF)) = Q(H\nB.np{Q\nB The supersymmetric oscillator is the simplest example of supersymmetric theories.nF) = ^ =  n B . if nB = 0 or nF = 1.e. it now follows that such paired states will be degenerate in energy.24) that when f(x)=ux. From our discussion in the last section.15)). 6.nF)) H(Q\nB.nF(Q\nB. the operators Q and Q take a bosonic state (with nF = 0) to a fermionic state (with nF = 1) or vice versa. since Q and Q commute with the Hamiltonian of the system (see Eqs (6. (6. Namely H(Q\nB. for consistency with earlier discussions we have set m = 1 for the bosonic part of the Lagrangian. It is clear from Eq.nF +1) 0 if nB ^ 0 or nF ^ 1.14) and (6. therefore. Namely. quantum mechanical theory. nF)).25) the Lagrangian of Eq. i. This is the manifestation of supersymmetry on the states in the Hilbert space of the Hamiltonian.2 Supersymmetric Quantum Mechanics Let us next study a general supersymmetric. The concept of supersymmetry and graded Lie algebras generalizes to other cases as well and there exist many useful realizations of these algebras in the context of field theories. (6.102 Field Theory: A Path Integral Approach Q\nB.nF)) = EnB. . (6.nF)) = Q(H\nB.23) = EnB. we note that acting on any state other than the ground state.nF)).l. (6. the bosonic and the fermionic states are paired.24) reduces to that of a supersymmetric oscillator discussed in the last section..fww • (6 24)  Here. let us consider a Lagrangian of the form L ±2 =\ ~ Vf{x))2+^ . Furthermore. we note that supersymmetry necessarily involves both bosons and fermions and.
Supersymmetry 103 In general. (6. that while the Lagrangian in Eq. the action for the Lagrangian can be seen to remain unchanged. are reminiscent of the supersymmetry transformations which we discussed earlier. In other words.27) where e and e are infinitesimal Grassmann parameters. (6. n = 0.. let us consider monomials only of the form. let us next look at the generating functional for the supersymmetric quantum mechanical theory in . (See chapter 11 for a detailed discussion of symmetries.2. (Instantons are discussed in chapter 8..) Consequently.26) 5^ = 0.27) are generated respectively by the two supersymmetric charges Q and Q in the theory. the transformations in Eqs. (6. and —j=xej=f(x)e.) These symmetry transformations mix up the bosonic and the fermionic variables of the theory and.26) and (6. without going into detail. in the case of even monomials the presence of instantons breaks supersymmetry. i _ 1 5&l> = j=xej= f(x)e. In fact. We also note here. one can explicitly show that the two sets of transformations in Eqs. we note that under the infinitesimal transformations 5ex = —= xpe.1.27) define symmetries of the system. f(x)~x2n+1. therefore. (6.24) is supersymmetric for any monomial f(x). 8e*l> = 5£ip = 0. (6.28) With these preparations..26) and (6.. (6.
29).24) that since the Lagrangian is quadratic in the fermionic variables.29) As we have seen in the last section. (6. Correspondingly.31). In particular.24) (h = 1) Z = N f V^V^Vx iS[x ] e ^. . (5. (6.104 Field Theory: A Path Integral Approach Eq. (6. (6.31) Let us next note that if we define a new bosonic variable through the relation p = ixf{x). the spectrum of a supersymmetric theory has many interesting features.40) and (5. we can write jv^Vrj. (6.32) then the Jacobian for this change of variables in Eq.31) will be given by i l J = det ( » ! .30) Substituting Eq.f'(x)\ . (See Eqs. we obtain Z = N fvipV^Vx = N jvxJi^kUW)2) = N fvxdet (i± . (6.87)). (6.33) This is precisely the inverse of the determinant in Eq. e '/<tttf(*&/'(*)W = N'det (i± . (6.f'(x)] iS[x e ^'^ [v$Thl>eiSdtMi&f'WM e *J*<i*M</(*» a >. the generating functional for such a theory is also quite interesting. Thus. (6. (6.30) into Eq. the functional integral for these variables can be done easily using our results in chapter 5./ ' ( * ) ) (6. let us note from Eq.
33) and (6. The solubility of such systems now appears to be related to a special symmetry associated with these systems known as shape invariance.3 Shape Invariance As we have noted earlier.31). Let us consider a one dimensional quantum mechanical system . (6. (6. 6. Here we have used the fact that for f(x) of the form in Eq. we see that the generating functional for a supersymmetric theory can be redefined to have the form of a free bosonic generating functional.34) into the generating functional in Eq.Supersymmetry 105 Furthermore. (6.2i J°° dx .) Substituting Eqs.35) In other words.28). there are only a handful of quantum mechanical systems which can be solved analytically. (6.32)) and generalizes to field theories in higher dimensions as well. the last integral vanishes. we note that f dtp2 = f dt(ixf(x))2 + (f(x))2) f(x) (6. This symmetry is also quite useful in the evaluation of the path integrals for such systems.(f(x))2) = Jdt(x2(f(x))2) . (For even monomials. This is known as the Nicolai map (namely.34) = J'dt(x22ixf(x) = _ j di (x2 . this does not vanish giving the contribution due to the instantons which breaks supersymmetry. we find Z = N fvpeydtP\ (6. Eq. (6. on the other hand.32).
and we identify U_(x) = ^(W2(x) .36) If we assume the ground state of the system to have vanishing energy.+ (W2(x) 2 2 i + W'(x)). or. then we can write the Hamiltonian in Eq. (6.+ U+(x) = V.42) A^O.38) (6.41) . namely. then it follows that QH_\i>) = \{Q\iP)) or. Q = j=(p + iW(x)). if H_\$) = QQ\il>) = \\rl>). (6.40) It now follows that if \ip) is any eigenstate of the Hamiltonian H_ other than the ground state.W'(x)). (6. QQ{Qm ff+(QV)) = A(QV» = A(Q^)). (6.106 Field Theory: A Path Integral Approach described by the Hamiltonian P2 ff_ = y + C/_(x). we can construct a second Hermitian Hamiltonian of the form H+ = QQ = P. (6.37) where Q = j=(piW(x)).36) also in the factorized form H_ = QQ.39) We note that Q and Q are Hermitian conjugates of each other and that given these two operators. (6. (6.
In such a case.45) Since we have assumed that the ground state energy of H_ vanishes and since we know that H+(a0) and H_(a0) are almost isospectral.43) P2 H_(a0) = Q{a0)Q(a0) = — + U(x. (6. P2 H+(a0) = Q{a0)Q(a0) = — + U+(x. it follows now from Eq. are almost isospectral in the sense that they share the same energy spectrum except for the ground state energy of H_. (6.44) then we say that the potential is shape invariant. (6. (6. (6.45) that the energy value for the first excited state of H_(a0) must be E. (6. we can . The potential.aQ) v2 = ff_(a1) + i2(a1) = Q(a 1 )g(a 1 ) + i2(a I ). depends on some parameters such as the coupling constants. H_ and H+.43) with R^di) a constant and the parameters a0 and al satisfying a known functional relationship ai = /(ao).a0) = U_{x.46) It is also easy to see now that for a shape invariant potential. we can write using Eq. If the potential of the theory is such that we can write U+{x. of course.Supersymmetry 107 Namely. = R(ai). we note that the two Hamiltonians.a1) + R(a1).a0).
(6. Furthermore. ) .47) we can write down the relation Q(as)Q(as) = Q(as+1)Q(as+1) or.52) = eitH(s+1)Q(as) . + R(as+1) (6. (6.50) This defines a recursion relation between the sequence of Hamiltonians. defining the time evolution operator for a particular Hamiltonian H^ in the sequence to be n(s) = eitHM ^ ( g51) we note that Eq. .47) will be almost isospectral and from this. fc=i (6. (6. (6. (6. . k=\ (6.50) gives Q(as)U^ = Q(as)eitH(s) = U(s+^Q(as).108 Field Theory: A Path Integral Approach construct a sequence of Hamiltonians such as tf(1) = H+(a0) = H_(ai) + R{ai)._1) + '£iR(ak) fc=i s = H_(as) + Y/R(ak). we can determine the energy levels of H_(a0) to be n £n = ^i?(afe). . Q(as)H^ = H^s+^Q(as). . ( / ( a 0 ) ) .49) Given the sequence of Hamiltonians in Eq.47) Here we have identified as = / s ( a 0 ) = / ( / . sl HW=H+(a.48) All the Hamiltonians in Eq. for t > 0. with a little bit of analysis.
we can solve for the time evolution operator of the original system.52) and (6. we obtain du(s) dt _ = (s iHin(s)„itH(°) / s \ = i I Q(ass)y{as)) + .J2R{ak. by taking the time derivative of Eq. This will determine the path integral for the system. (6.51).53) The relations in Eqs.  ^ + i£ i2(afc) J U& = iQ(as)U^+1^Q(a3).y.))e.Supersymmetry 109 Similarly. (6..*tH(s) y{a )Q(as . \ s fc=i / or.aa)\u^(x.54) It is clear from this discussion that for a shape invariant potential if one of the Hamiltonians in the sequence coincides with a system which we can solve exactly. (6.53) define recursion relations for the time evolution operator and have the coordinate representation of the form ^ + W(x.54).t) ~ + W(y. (6.y.as)^U^+1Hx. .t). then using the recursion relations in Eq. (6.
^ 2W This determines the transition amplitude for the original system.47)) uW(x. . is H<® =H_(a0) = ^sech2x Zt + .t) = fL= e ^ ^ .a0)) (6.a0) = i (W2(x.55) Prom Eqs. (2.a0) 1  W'(x. „2 i2(Oi) = 0. the free particle Hamiltonian for which we know the transition amplitude to be (see Eq. (6. we find U.a0)) = . sechy (iktanhx)(ik + t&nhy) 2 1 + fc e(ifc(a .58) and the next Hamiltonian in the sequence is given by (see Eq. i (6.— sech x.{W2(x.a0) + W(x.a0) = l.39) and (6. we can identify a0 = 1.40)._ y) _tt (fc a +1)) 2TT J. (6. it is easy to see that U(°'(x.y. (6.2 _ 1 ~ 2" ~ In this case. fll = a0 .57) The Hamiltonian for the system.a0) = a 0 tanhx. (6. therefore. of course.4 Example Let us consider a quantum mechanical system with W(x.54). (6.1 = 0.56) v.t) dk = secha. Zi (6.(x. in this case.y. 2 U+(x.110 Field Theory: A Path Integral Approach 6.47)) HW = H+(a0) = H_(ai) + R(ai) = Y + l' ( 6 . \J2mt (6.60) into Eq. a0 — 1.59 ) This is.^ .60) Substituting Eq.
rather it is a consistency condition which arises from treating the singular potential in a carefully regularized manner. we expect the wave function to vanish at the origin leading to the conclusion that.5 Supersymmetry and Singular Potentials Singular potentials within the context of quantum mechanics are interesting because they remind us of the necessity of regularization even in such simple systems. to bring out these features. To understand the supersymmetric "half" oscillator or the supersymmetric oscillator on the half line. Namely. Let us consider a particle moving in the potential { h{co2x2 u). such a working rule should not be blindly imposed which can lead to erroneous conclusions. In going beyond simple quantum mechanical systems. It must be emphasized that such a condition should not be thought of as a boundary condition.61) for x < 0. This will also bring out clearly how choosing a regularization consistent with the symmetries of the system under study is important. The spectrum of this potential is quite clear intuitively. a careful analysis of singular systems using a regularization have led to the working rule that the quantum mechanical wave function must vanish at points where the potential becomes singular. the supersymmetric "half" oscillator. for x > 0. it is useful to recapitulate briefly the results of the "half" oscillator. however. of all the solutions of . We know from studies in relativistic quantum field theories that a regularization must always be chosen consistent with the symmetries of the theory under study in order to be able to extract meaningful results. Singular potentials within the context of supersymmetric quantum mechanical systems provide an excellent example of this and we will discuss this in this section. Other quantum mechanical systems with a complex singularity structure can be found in the literature.Supersymmetry 111 6. because of the infinite barrier. An improper choice of regularization can lead to incorrect conclusions about the theory. oo. (6. We will study a simple model. In simple quantum mechanical systems.
vanishes for large . let us consider the particle moving in the regularized potential \{u?x2 co).63) Since the system no longer has reflection symmetry. the solutions. we have the asymptotically damped solution. let us note that singular potentials are best studied in a regularized manner because this is the only way that appropriate boundary conditions can be determined correctly. The solutions of the Schrodinger equation. cannot be classified into even and odd solutions. Z (6. on the "half" line there is no notion of even and odd) would survive in this case.112 Field Theory: A Path Integral Approach the oscillator on the full line. The Schrodinger equation 1 d2 \ •2^2+V(x)j4>(x) = ei>(x). Since \c\ — oo at the end. x). y.62) for x < 0. with the understanding that the limit \c\ — oo is to be taken at the > end. Therefore. First. are known as the parabolic cylinder functions and the asymptotically damped physical solution is given by ^l)(x) = BU((Id + l). (6. can now be solved in the two regions. Rather. of course. While this is quite obvious. the normalizable (physical) solution would correspond to one which vanishes asymptotically. ^{x) = Ae^2t^x.V2^x). let us analyze the problem systematically for later purpose. for x > 0 V{x) = { 2 (6.64) The parabolic cylinder function. for x < 0. in the region x > 0. in the region x > 0. > for any finite energy solution. only the odd solutions (of course. U(a.
(remember that the zero point energy is already subtracted out in (6. V^x) = Bn e .Supersymmetry 113 values of x.x)x^°—f P .67 ) In other words. One can define a superpotential { —LUX.61) or (6. (6. for x > 0. Thus.65) It is now straightforward to match the solutions in Eqs. then. (6.68) The corresponding physical wave functions are nontrivial only on the half line x > 0 and have the form Mx) = Bn U((2n +  ) . we see that the correct boundary condition naturally arises from regularizing the singular potential and studying the problem systematically.x) 2i(2a+l)r(3 + a)' U'(a.66) It is clear. this can be satisfied only if > ~h + \ lcb"°° ""> ^ = 0'1'2 ( 6 . We now turn to the analysis of the supersymmetric oscillator on the half line. when the regularization is removed. For small values of x. (6. it satisfies U(a. . (6. that as \c\ — oo.63. 6. only the odd Hermite polynomials survive leading to the fact that the wave function vanishes at x = 0. (6.69) Namely. for x < 0 .^ + i) (6.62)) en = co(2n + 1).70) oo.> 2 H2n+l{^x).64) and their first derivatives across the boundary at x = 0 and their ratio gives i VJ^Te _ i nt) 2 v ^ r ( . namely. the energy levels that survive are the odd ones.
Furthermore. ^. for x < 0. V{x) = { c2 j. Since. tl>+. the state with e = 0 no longer belongs to the Hilbert space (since it corresponds to an even Hermite polynomial solution). in particular. H2n+i(Vux). we obtain e+. l .n(x) = Btne~^x2 1 2 H2n+i(Vux). the supercharge is an odd operator and hence connects even and odd Hermite polynomials. However. in this case. .72) for x < 0. as before.73) Here n = 0 . lead to the pair of potentials { V+(x) = k(to2x2Tu). that in such a case.71) for x < 0. iorx>0. First. .n(x) = B+iU e~2"x ip. It is straightforward and without going into details. (6.72). with the understanding that \c±\ —• oo at the end. This boundary condition arises from a systematic study involving a regularized potential. we can study it. (6. . (6. Second.114 Field Theory: A Path Integral Approach which would. . it would appear that the superpartner states do not belong to the physical Hilbert space (Namely. the energy levels for the supersymmetric pair of Hamiltonians are no longer degenerate. forx>0. This leads to the conventional conclusion that supersymmetry is broken in such a case and let us note. namely. in this case. 2 .n = 2w(n + 1). by regularizing the singular potentials as \{UJ2X2 c2 u).n = w(2n + 1). » The earlier analysis can now be repeated for the pair of potentials in Eq. . this involves singular potentials. e_. forx>0. (6. There are several things to note from this analysis. only the odd Hermite polynomials survive as physical solutions since the wave function has to vanish at the origin. that. oo. let us simply note the results. 2 2 \{OJ X +u>). naively.
The regularized superpotential now leads to the pair of regularized . therefore. The answer is quite obvious. (6. the regularization would have broken supersymmetry through instanton effects as we have mentioned earlier). it is. it is the superpotential which gives the pair of supersymmetric potentials through Riccati type relations. the breaking of supersymmetry that results when the regularization is removed cannot be trusted as a dynamical effect. 6. let us look at the regularized superpotential W(x) = LOX6(X) + c0(x). It is natural. for any value of the regularizing parameters.Supersymmetry 115 the boundary condition selects out only odd Hermite polynomials as belonging to the physical Hilbert space.72) do not define a supersymmetric system and hence the regularization itself breaks supersymmetry. Rather. to regularize the superpotential which would automatically lead to a pair of regularized potentials which would be supersymmetric for any value of the regularization parameter. the form of the superpotential in Eq. namely.70) selects out c > 0 (otherwise.74) Here c is the regularization parameter and we are supposed to take \c\ — oo at the end.). hence the boundary condition) used. In other words. it is not the potential that is fundamental. namely. c± (even if \c+\ = c_).5. the boundary condition) breaks supersymmetry. the pair of potentials in Eq. then.1 Regularized Superpotential Another way to understand this is to note that for a supersymmetric system. meaningful to ask if supersymmetry is broken when the regularization parameter is removed at the end. Namely. The question that needs to be addressed is whether it is a dynamical property of the system or an artifact of the regularization (and. (6. There is absolutely no doubt that supersymmetry is broken in this case. (6. with such a regularization. Note that the existence of a normalizable ground > state. Consequently. that supersymmetry is broken mainly because the regularization (and. therefore. With this in mind. such a regularization will respect supersymmetry and.
^_.77) It is now clear that. (6..77) give respectively • e + i n = 2ion.75) V_{x) = \ [{LO2X2 + u)9{x) + c29{x) + c6(x)] .79) This is indeed quite interesting for it shows that the spectrum of H+ contains the ground state with vanishing energy.76) and (6. Furthermore. n e . we obtain the two conditions (c 2 _ 2e+)i/2 _c _ + C 2^r(g r(£) + i) (6. Carefully matching the wave function and the discontinuity of the first derivative across x = 0 for each of the wavefunctions and taking their ratio. have the forms 1 2 ll>+. However. (6. Consequently..n = 2w(n + 1).. as c — oo. in this case.^ + i) (6.2.cd(x)] .72)) lies only in the presence of the 5(x) terms in the potentials. which are supersymmetric for any c > 0.78) The corresponding wave functions.116 Field Theory: A Path Integral Approach supersymmetric potentials V+(x) = \ [{u2x2 . Let us note that the difference here from the earlier case where the potentials were directly regularized (see Eq. all the other states of H+ and H_ are degenerate in energy corresponding to even and odd Hermite polynomials as one would expect .n(x) = B + . ra = 0.1. e_.76) I ( C 2_2 £ _)V2 2 I r(^ + i) v ^ r ( .u)0{x) + c29(x) . n (x) = 5 _ . the matching conditions are now different because of the delta function terms. (6. the earlier solutions in the regions x > 0 and x < 0 continue to hold. (6.^ 2 H2n+i(V^x). n e " ^ H2n(Vux) . (6. .
u)9(x) + ( A V .80) Here A is the regularization parameter and we are to take the limit A — oo at the end.XxO(x). Namely.X)9(x)] . This regularized superpotential would now lead to the pair of supersymmetric potentials of the form V+(x) = \ [{u\2 .2 Alternate Regularization Of course. let us choose a regularized superpotential of the form W{x) = uxO(x) . In fact. equivalently. 6. This should be contrasted with the general belief that supersymmetry is broken in this system (which is a consequence of using boundary conditions or. then. the regularization merely introduces a supersymmetric pair of oscillators for x < 0 whose frequency is to be taken to infinity at the end. Once again. Consequently. it is quite clear that if the supersymmetric "half" oscillator is defined carefully by regularizing the superpotential. it is worth investigating whether our conclusions would continue to hold with an alternate regularization of the superpotential which would not introduce such singular terms to the potentials. we note that. consequently. .Supersymmetry 117 from superpartner states. (6. we should worry at this point as to how regularization independent our conclusion really is.81) V(x) = ^ [ ( w V + u)0(x) + ( A V + X)e(x)} There are no singular delta potential terms with this regularization. our results appear to follow from the matching conditions in the presence of singular delta potential terms and. although both signs » of A appear to be allowed. With this in mind. (6. of regularizing the potentials in a manner which violates supersymmetry).5. supersymmetry is manifest in the limit of removing the regularization. existence of a normalizable ground state would select A > 0. .
78) and (6. A.n(x) = £+. 123 (1997). Phys. Eqs. this shows that this conclusion is independent of the regularization used as long as the regularization preserves supersymmetry which can be achieved by properly regularizing the superpotential. Furthermore.118 Field Theory: A Path Integral Approach Since there is a harmonic oscillator potential for both x > 0 and x < 0. Now matching the wave function and its first derivative at x = 0 for each of the Hamiltonians and taking the ratio. (6. Nucl. of course.6 References Das. The corresponding wave functions are given by ip+. ^_.82) and (6.. Huang.85) These are. B505.n = 2co(n + 1). Phys. They are the parabolic cylinder functions which we have mentioned earlier. and S. and W..n e ^ 2 ^ 2 H2n(Vux). (6. the solutions are straightforward.. we obtain 1 r( 2A) _ ! r ( It) (682) i n^ + l) _ i r ( . n = 0. (6. n e " > 2 H2n+1(V^x). A. the same energy levels and wave functions as obtained in Eqs.^ + i) It is clear now that.2. as A — oo. n (x) = 5 _ . . Das. Pernice.79) respectively showing again that supersymmetry is manifest. 3241 (1990).84) e_. 6. (6..83) give respec> tively e+iW = 2um. Rev. D41.1. J. This analysis can be carried out in a straight forward manner to more complicated superpotentials and the conclusions hold without any change.
Nucl.Supersymmetry 119 Das. Pernice. A. 561. Nucl. . 419 (1980). Witten. Phys. 253 (1982). E. H. B170.. Phys. L. Nucl. Gendenshtein. Phys. Nicolai. and S. 357 (1999). 38. JETP Lett. 356 (1983). B202...
.
the stationary states of the system will satisfy the timeindependent Schrodinger equation given by {JL&+VM)«*>=E*<*). However. It is. These are inherently nonperturbative phenomena.Chapter 7 SemiClassical Methods 7. The basic idea behind this is quite simple. Then. For example. 121 ( ") Here E is a constant representing the energy of the state. even though we may be able to obtain the energy levels and the eigenstates for the motion of a particle in a potential well by using perturbation theory. we will never learn about barrier penetration from such an analysis. we use perturbation theory and perturbation theory brings out many interesting properties of the system. by definition perturbation theory cannot provide information about nonperturbative aspects of the theory. of course. useful to develop an approximation scheme which brings out some of these nonperturbative characteristics.1 W K B Approximation As we know. In such a case. Let us assume that we have a particle moving in a complicated potential V(x). Similarly. WKB is such an approximation scheme. the Born approximations used in scattering theory give more accurate estimates of the scattering amplitudes as we go to higher orders of perturbation. therefore. but we cannot obtain information on the bound states of the system from this analysis. most quantum systems cannot be solved analytically. We know .
but changes slowly.2) with p(x) = ^2m(E . is complex.3) It is clear.V(x)). With all these information.1) can still be written as plane waves of the form of Eq. if V{x) = V = constant. (7.1) can be represented as a phase where the phase. ip{x) = A e ± ^ x .4) where N is a normalization constant and furthermore.2) When the potential changes with the coordinate. then. (7. namely. let us write the general solution of the timeindependent Schrodinger equation to have the form r/>(x) = N efcW. the phase clearly will be a function of h. (7. therefore. in general. Since the Schrodinger operator depends on h. Namely.122 Field Theory: A Path Integral Approach that if the potential were a constant. we conclude that the general solution of Eq. the solutions of Eq.5) If we substitute this wave function back into the Schrodinger equation . (7. noting that we can write (for nonnegative A(x)) A(x) = elnA^ . that we can try a general solution to the timeindependent Schrodinger equation of the form </>(£) = NA(x) eiB{x) . (7. (7. the solutions of Eq. (7. then it is easy to convince ourselves that within a region where the potential does not change appreciably. (7.1) will be plane waves (for E > V). where p = ^2m(E V).
( . (7. therefore.8) For this to be true. So far. we obtain d2ip{x) 2m +Rr(EV(x)Mx) dx2 or. we obtain 1 h2 [(<t>'0(x) + H'^x) i + • • • ) 2 + 2m(E . represent quantum corrections to the classical phase. we obtain ((f)'0(x))2 + 2m(EV(x)) =0 or. the coefficients of the individual terms in Eq. (7. Let us next assume a power series expansion for <f>(x) of the form 4>(x) = (p0(x) + Hi{x) + tffoix) + ••• .2cP'0(x)(f>'1(x)) + O(h0) = 0. everything has been exact.( ^ ( x ) ) 2 ( h2 ~ + 2m(£y(x))) + i « ( x ) .SemiClassical Methods 123 (Eq.1)). then. Other terms in > the series. Equating the coefficient of the J^ term to zero. l ( .6) + ^"{x) .V(x)^j e**<*> = 0 (7.6). 4>'0{x) = ±p(x) .^ ( 0 ' W ) 2 + y\x) or. ( . (7.^ ( * ) ) 2 =0 + ^(E + ^(E .V[x)) = p2(x) or.V(x))] +(</£(*)+ ^'/(z) + )=o h or.8) must be zero. (7. Substituting the power series back into Eq.V(x))"j = 0. (7.7) It is clear. that (f>o(x) will correspond to the classical phase since that is what will survive in the limit h — 0. (cf>'0(x))2 = 2m(E .
8) to zero. we obtain #S(x) . (7.9) Here p(x) is the momentum of the particle at the point x defined by Eq. (7.**'**') _ ( 7 n ) The timedependent stationary wave function. (7. (7.t) = = e~iEttp(x) Ne*Et+*I*odx'pW = JVesJotd*'(P*£) = Neislx<d. in this case. Furthermore.2tf.12) This is exactly what we would have expected in the classical limit.(x)ti(x) = 0 or.13) . M%) = ± I dx'p(x'). (7.124 Field Theory: A Path Integral Approach .9). This. would be given by i/)(x. Let us next include the first order correction to the phase.\np{x). Setting the coefficient of the ^ term in Eq.10) If we keep only the leading order term in the expansion of 4>(x). then the wave function would have the form 4>(x) = Nei*°(*) = NeUZ. consistency of the subsequent relations will pick out only the positive sign. therefore.3). (7. corresponding to motion with energy E. determines the classical phase to be M*)= f Jxo dx'p(x'). let us note that even though both the signs of the solution are allowed in Eq. M*) = \ In 4/Q{x) = l. (7.
is inversely proportional to the momentum or the velocity. the classical momentum vanishes and. clearly. breaks down for small p(x) and in particular when p(x)=0. This approximation. the probability density. From the form of the WKB wave function in Eq. Namely. we can write the wave function to be tl){x) ~ = NeUMx)+f^i(x)) Neji(S x dx'p{x')+%hLp(x)) (714) = N ey x dX'P(X')_ This is known as the WKB approximation for the quantum mechanical wave function of the system. we would expect qualitatively that the system is more likely to be found at points where its velocity is smaller. we note that ^•(xMaOoc^. what we would expect from classical considerations alone.14).16) In other words. (7. keeping up to the first order correction to the classical phase. in these regions the WKB approximation breaks down and we must examine the Schrodinger equation and its solutions more carefully. We also note here that this selects out the positive root for <po(x) in Eq. the WKB approximation gives us a quantum wave function which retains some of the classical properties. (7. Thus.9). in this case.SemiClassical Methods 125 The constant of integration in Eq. (7. at the classical turning points. (7. Thus. classically. of course.15) Namely. . It is for this reason that the WKB approximation is often also called the semiclassical approximation. This is.13) can be absorbed into the normalization constant iV of the wave function. (7. consequently.
.19) or. N = N* = Therefore. p{x) / •I Xn Recalling that the classical period of oscillation is given by T =2 we obtain 1 rXb dx fXb _da^ _ „ Jxn V X ( ) dx ^^ Jxn p(x) ' fXb f* (7.18) ' 2m 2m (7. Since the wave function damps outside the well. we can write the normalization condition to be approximately CO />£(. / co da. dx ip*(x)ip(x) (7. Then. the normalization constant for the WKB wave function can be determined approximately in the following way.126 Field Theory: A Path Integral Approach Xb X • Let us consider a particle moving in a potential of the form shown above.17) wi'T^i. we can write the normalized WKB wave function to be ™°\l$r)'ir""} " TTV(x) e±f*dx'p(x>) (7. ip* (x)ij){x) ~ / Jx„.20) where ui — %£• denotes the classical angular frequency of motion.
2 Saddle Point Method Let us consider an integral of the form oo dxe*fW. Furthermore. let us assume that the function f(x) has an extremum at x = XQ which is a maximum. /"(*)U. (7. < 7 . this is the most dominant contribution to the integral.x0f).SemiClassical Methods 127 7.<oWe can now expand the function around this extremum as f(x) = f(x0) + \{x .e[^(/(^o)+i(xxo) 2 /"(^)+0((x:Co)3))] 2 / oo = es/(*o) /•oo / J—oo /•oo dxe(±(xxo) \f"(x0)\+0((xx0)3)) = eifM ~ e»/(lo) / J—oo dy ^e(h2\f"^)\+o(^y3)) 2n ^ l/"(*o)l ±/(*o) ea / ( x °). . we are assuming that f'(x)\ =0.23) Substituting this back into the integral in Eq.xQ)2 f"(x0) + 0((x .24) It is easy to see that the terms we neglected are higher order in a and.21). therefore. (7.22 ) (7.21) / oo where a is a very small constant. we obtain oo da. 2lUX l/"(*o) (7. In other words.
then the value of the integral will approximately equal the sum of the contributions around each of the extrema.128 Field Theory: A Path Integral Approach Irc^ Note from Eq.26) . then. x n /<(. let us analyze the Gamma function for large values of the argument. this method of evaluating the integral is also referred to as the method of steepest descent. The Gamma function has the integral representation given by /•OO T(n + 1) = n ! = / Jo /•OO dxxne~x dxe<x+nlnx) = / Jo /•OO = / dxen(lnx~nK (7.in our previous discussion in Eq. f(x) = In a. (7.23) that if we consider the function in the complex xplane. then along the imaginary axis it has a minimum at the extremal point. It is worth emphasizing here that if the function in the exponent has several extrema. the extremal point is really a saddle point in the complex xplane. (7.25) Jo Let us next assume that n is very large. the saddle point method.I . Therefore. This method is quite useful in obtaining approximate values of very complicated integrals. would correspond to .21). In the present case.) = I . therefore. As an example. This. Hence the name. (7. Let us also note that the direction of our integration has been along the direction of steepest descent (along the real axis) and.
therefore. we can write T(n + 1) ~ e n / ( s °) / Jo = e n/(xo) dx n{ e ^(xXo)) / dxeto^"^)2 oo Jo / oo = v&en(hn1) or. n! y/tom ( . this is also written in the form Inn! ~ n(lnn — 1).^ = ( l n n . Furthermore.29) Therefore.30) (Since XQ = n is large and positive. (7.27) /(x 0 ) = I n x 0 .< 0 . we have extended the integration to the negative axis in the intermediate step because the contribution from this region is negligible.! 2.28) /"(x 0 ) = .31) .i . XQ — n is a maximum of the function and. Sometimes. Thus. (7. (7. x (7.) " . for large n .SemiClassical Methods 129 Requiring the first derivative to vanish gives XQ = n.) This is known as Stirling's approximation and holds when n is large. for large n . we have = . in this case.l ) . is the only extremum. / » and. 2 n (7. for large n.
it provides an extremum of the exponent in the path integral. Namely. (2. (7. Let us recall that the classical trajectory satisfies the equation (see Eq. Thus. Namely. the action is a minimum for the classical trajectory. (1. Therefore. The transition amplitude associated with this action. is given by (see Eq. Let us consider a general action S[x] which is not necessarily quadratic. we are assuming that the actual calculations are always done in the Euclidean space and then the results are rotated back to Minkowski space. The path integral cannot always be evaluated exactly for such systems and the method of steepest descent gives rise to a very useful approximation in such a case.U) = N f Vx eS s N .tf\xi. we know that.21). this integral is similar to the one in Eq.32) Since h is a small parameter. we can expand the action around the classical trajectory. let us define x(t) = xcl(t) + ri(t). we can rotate to the Euclidean space as we have discussed earlier and the integrand becomes well behaved. but is not exactly in the same form. But any realistic theory will involve interactions which are inherently nonlinear. in such a case. Furthermore. However. (7.3 SemiClassical Methods in Path Integrals The saddle point method or the method of steepest descent can be applied to path integrals as well.28) or (3.28)) {xf.130 Field Theory: A Path Integral Approach 7. this integral is oscillatory. as we have seen.33) . (7.9)) 6S[x] 5x(t) = 0. We will continue to use the real time description keeping in mind the fact that in all our discussions. Note that so far we have only evaluated path integrals which involve quadratic actions.
2). (7. the path integral has to be evaluated more carefully using the method of collective coordinates as we will discuss in the next chapter. (Recall from Eq. we have S[x] = S[xcl + rj\ (7. When this happens. (7.35) ) where we have used the result of Eq.35) is surprisingly similar in form to the WKB wave function in Eq.) In fact. (7. (1. the phases are identical (see Eq. (7.35) that the transition amplitude is a Schrodinger wave function for a delta function source. (We are also keeping the h term explicitly to bring out the quantum nature of the calculations.34) Substituting this back into the transition amplitude in Eq.20).SemiClassical Methods 131 Then.ti) ^ (5 g[x c ^ d e t (H <5xci(ti)(5xci(t2) 2 3 *SM. (7. this happens when there is some symmetry or its spontaneous breakdown occurring in the theory. We note here that the form of the transition amplitude in the saddle point approximation in Eq.tf\Xi. we obtain (xf.32).12)). (4.) Clearly. the saddle point method breaks down if dct *2gM _Q feci(ii)feci(t2) Normally. .
we note that S2S[xcl] 6xci(ti)5xci(t2) Therefore. the left . we are interested in evaluating determinants of the form G( m S + w ( x ) in the space of functions where rj(^) = 0 = r)(—^). We have already evaluated such determinants earlier in connection with the harmonic oscillator (see Eq. Let us recall here some general results which hold for determinants of operators containing bounded potentials. relate the two multiplying factors as well. We will describe below. We note that for A coinciding with an eigenvalue of one of the operators. we will choose ti = —\ and tf = ^ which will also be useful later. With a little bit of analysis. in fact.V(xU . (4. we see that (mc^ + V "(Xd)) 5 h ( ~ h ) • ( 7  3 7 ) HlsdM^)Hl(m^+v"M))det ( 3) 78  Thus. how we can. Let us note that a general action has the form S[x] = f f dt [\rnx2 . if I ( m oS det(\(m& det(\(m^ + W{X) ) ^ =A = ^' 0 an< (7 39) " = with the initial value conditions ip\y(~ 1") then ^ ^iV(—"2) *> + W1(x))\) + W2(x))\) _^() ^)(f) (7.10)).132 Field Theory: A Path Integral Approach It is the multiplying factors that we readily do not see to be comparable. In this case. (7. without going into too much detail.40) where W\{x) and W2(x) are two bounded potentials.36) For simplicity. it is possible to show that.
44) . let us use this result to write the transition amplitude in Eq. (7. ij)vl(t) the equation (it is an eigenstate with zero eigenvalue) satisfies ( m S + V "(X^) ^vkt) = 0 • (7. (7. But so will the right hand side for the same value of A because in such a case ip^ would correspond to an energy eigenfunction satisfying the boundary conditions at the end points. (7.28)) m ^ + V'(xcl) = 0. they must be equal. (1.39)).43) We note here that the classical equations (EulerLagrange equations) following from our action have the form (see Eq. But.SemiClassical Methods 133 hand side will have a zero or a pole. (7.42) (xf. for the moment.41) That is. Since both the left and the right hand side of the above equation are entire functions of A with identical zeroes and poles.35) in the form eis^'l ydet(i(mg + V(xd))) iV e*s[xd] . We will define a particular normalization later. this ratio is independent of the particular form of the potential W(x) and can be used to define the normalization constant in the path integral.ti) = Let us note here that by definition (see Eq. It follows from this result that det(±(m^ + Wl(x))) d e t ( I ( m g + ^2(x))) = = (7.tf\xi.
4 Double Well Potential As an application of the WKB method. (7. . (7. 2 2 y/p(xf) (7.a2)2) . (7. (7. let us consider a particle moving in onedimension in an anharmonic potential of the form V(x) = 9~(x2a2)2.48) where g and a are constants. (7.134 Field Theory: A Path Integral Approach It follows from this that d / ni d2£ci di V dt2 + V'{xcl)\ = 0 „ dxd d2 fdxd\ or. (7. we recognize that we can write the transition amplitude in Eq.20) is now complete and we recognize that the method of steepest descent merely gives the WKB approximation. (m^2 + v"M)^r = ° ( 7 .47) The correspondence with the WKB wave function in Eq. the action has the form S[x] = f dt (^mx2 . Consequently. . 7.% = £=etsM. we readily identify that 1>$(t)cx^<xp(xd).46) Consequently.( x 2 . when plotted. has the shape of a double well.49) This potential.42) also in the form (xf^\Xi.45 ) Comparing with Eq.43).
there will be no tunneling from one well to the other.50) The height of the potential at the origin is given by (7. Namely. Furthermore.SemiClassical Methods 135 This is a very interesting potential and shows up in all branches of physics in different forms. The motion of the particle is easy to analyze in this case. (7. then it stays there forever. We note from Eq. x <> — x ). Each well has quantized levels of energy and if the particle is in one well. then tpo(x) will describe the ground state wave function of the left well with the same . from the symmetry of the problem at hand (namely. Note that it is an even potential with a local maximum at the origin. The two minima of the potential are symmetrically located at x = ±a. we conclude that both the wells in the present case will have degenerate energy levels. (7.51) Let us also define here for later use V"(x = ±a) = g2a2 = rmo2 .51) that for infinitely large coupling. Thus.52) where we can identify to with the natural angular frequency of harmonic oscillations near the minima. (7. if 4>o{x) denotes the ground state wave function of the well in the positive xaxis with energy EQ. the potential separates into two symmetrical wells with an infinite barrier.
Consequently. not be degenerate in energy any longer because of tunneling and we wish to calculate the splitting in the energy levels due to tunneling using the WKB approximation. ^ ^ + f ? ( £ o . let us assume. If ipo(x) denotes the wave function in the well I (that is. for simplicity. the well on the right). Thus. it is obvious that V'IOE) would represent the ground state of the system (after taking tunneling into account). (7.53) will also be degenerate in energy. (7. Let us write down the timeindependent Schrodinger equations that various wave functions satisfy. The symmetry of the system (Hamiltonian) still dictates that the eigenstates of the Hamiltonian can only be even and odd linear combinations as described above.56) Furthermore. They will. (7. that the wave functions are all real. in particular. we can . When the coupling constant g is finite. the combinations A(x) = j=(ip0(x) + ipo(x)). (7. then the potential barrier will be finite. any linear combination of the two wave functions and.136 Field Theory: A Path Integral Approach energy. Mx) = ^M>0*0 .55) ^ l + fiE2V(X))Mx) = 0. or the well on the left. In fact.V(x))Mx) = 0.Mx)). however. then. Consequently. we can easily see that its value will be vanishingly small in the well II. Let us note that ipi (x) is a symmetric wave function whereas tp2 (x) is antisymmetric.54) f ^ 2 + ^(ElV(x))Mx) = 0. the particle initially confined to one well can tunnel into the other well and the states of the two wells will mix.
53). . .^ .l ^ W . a product such as ipo(x)ipo(—x) will be negligible everywhere.54) by il>i(x) and Eq.d2ip0(x) . then it would have vanishingly small value in well I. (7.E ^ . so that ^OO (7.(MWoW = ^iWo(0)^o(OM(0) or. . (7.58) 1 /*0O 1 / dx^(x)Mx) ~m I da. E o .57) Let us note similarly that if 4>o(~x) denotes the wave function in the well II.Vo(0)Vi(0)) . that ^ ( ^ ( x j . (7.60) Integrating this equation and using Eq. Vo(z) .55) by I)}Q[X) and subtract one from the other.d2V>i(x) 2m. We see from Eq.^ (Vi(O)Vd(O) .j= • (7.SemiClassical Methods 137 normalize the wave function as /•oo / Jo dx^2(a. .59) Let us next multiply Eq. .)~l.2 /•OO Jm(Eo +W (F . . This gives .59). Consequently. . .In MxWi(x))\~ = . (7. ~(E0 ^i(ar) d 2 ^.E.61) ._.2 7p0(x) Hi{x) da. we obtain I?< or. mv2 (7. (7.)~ Ex) / da.(x) da. . therefore. . . (7.
We recall from Eq. ^i(O) = V2 ^o(O).El = . V(0)^E0. (7.65) = ftu(o)_^eit/o 7TU(0) = ^b(*)i (7.EQ).64) .138 Field Theory: A Path Integral Approach Let us note that since V»i(aO = 75M>(aO + i/>o(x)). mv2 m (7. (7. we obtain „ h2 mv(0). m fi ..62) It is here that we would like to use the WKB approximation.66) ^eI/oadbWI.^ = ^ o ( 0 ) V o ( 0 ) = . Substituting this back into Eq. Here.61).V o ( 0 ) ^ ( 0 ) . (7.20) that we can write the WKB wave function as W 7TU(0) ^(O)c^M^o(O).63) (7. . we note that v(0) = \l^(V(0) Putting these back. 7T .„. we obtain E0 . 9 0 (7. Vi(0) = 0.
of course. (7. note that because of the reflection symmetry in the problem.70) This would. we know that the quantum mechanical ground state energy is nonzero in general. x — a = ±1 " muj or. however. consequently.68) reflects the effects of tunneling. (7. are given by .71) . we can write e . (7. in this case. In fact. TV (7. then we can identify the ground state energy of the system with E0 = ~ .68) which gives the splitting between the two degenerate levels in this approximation. 1^12! = —eiK**W*)\. This. if we approximate the potential near each of the minima by a harmonic oscillator potential. assumes that the particle under consideration has vanishing energy.SemiClassical Methods 139 This is the splitting in the energy level of the true ground state from the case of the infinite well.af = EQ = — or.f tfdx \p(x)\ = ei/ad* p(x) ^ (769) and this gives the coefficient for tunneling from the minimum at x = —a to the one at x = a. x\ = a± \l . then. •K = —eiSoa**W*)\. In fact.67) (7. The damping exponential in Eq. In reality. We can similarly show that E2Eo and. imply that the turning points for motion in both the wells.mw\x .
Correspondingly.140 Field Theory: A Path Integral Approach and. (7. for a more accurate estimation of the splitting in the energy levels.68) by _ 2 fa" e ft Jo _2 e ft Jo°d*b(z)l dx \p(x) (7.3)) .mu\x + a)2 = EQ = — or. would correspond to tunneling from —a + A / — to a — v / ^ j .72) The tunneling from one well to the other in this case. we should replace the exponential in Eq.73) Furthermore. therefore. recalling that (see Eq. (7. similarly. . x\i = — a ± muj h mto (7. x + a = ± or.
77) . (7. (7.\ h 2 2\+0(K V mu>a / Here we have defined 2mtoa2 So = — g — • (7.SemiClassical Methods 141 we can evaluate the exponent in a straightforward manner as I v 7nbJ \ Jo dx \p(x)\ ra~\J^ Jo ^ .x2) ( 2a \ 4a?h \* 2 2 2 mco(a — x ) J : i x ^ ^ _ _ ^ _ ) (774) This integral can be trivially done and has the value I y 7TIQJ / JO dx \p(x)\ mtoa2 x 3 h h. we obtain the splitting in the energy levels to be E2E1~ TV 4e \ V e~ns° n 3 i n _ = — Vmftwza e'fi* 0 .76) Substituting Eqs.76) into Eq. /— (7.73) and (7. h h. / o + 9lnV 2~oln \ 2 2 V mioa* 2 \ 2 I .68).^ muj(a 2 .
Lifshitz. Pergamon Press. Migdal. 7.. and L. "Quantum Theory of Many Variable Systems and Fields". we will calculate this energy splitting using the path integrals and compare the two results. "Nonrelativistic Quantum Mechanics". World Scientific Publishing. "Approximation Methods in Quantum Mechanics". . M. B.. Erice Lectures. 1977.142 Field Theory: A Path Integral Approach In the next chapter. A. "The Uses of Instantons". Landau. Krainov. Benjamin Publishing. and V. L. Sakita.5 References Coleman. B. S. D .
(2. ) = (xf\eiHT\xi) = N jvxe&W. (7. (8. (8. (8. .1 Instantons Let us next try to do the path integral for the double well potential.42) {xf\e~*HT\xi) = N [vxexSsW . by rotating to imaginary time t —> —it.Chapter 8 P a t h Integral for the Double Well 8.4) .1) where for a double well potential (see Eq.  N . We recall from Eq.28) that the transition amplitude is defined as ^ / .2) As we have seen earlier in chapter 4. (8.49)) the action is given by S[x] = f2 dt (Knx2 V{x) = f^dt Qmi2^(z2a2)2).3) where the Euclidean action has the form T SE[x] = f^dt (^mx2 143 + V(x)j . the best way to evaluate the path integral is to go to the Euclidean space. (1. we obtain using Eq. Thus.
V(x).7) —a a x + Two solutions to the Euclidean classical equation of motion in Eq. Namely.8) satisfy the classical equation with E — 0. the particle stays at rest on top of one of the hills in such a case. x(t) = ±a. (8. we can evaluate the path integral by the saddle point method. The Euclidean energy associated with such a motion is given by E = . (8. The classical equation which is obtained from the extremum of the action in Eq. In other words. (8.5) with minimum energy are obvious. this would correspond to the case where the particle .4) has the form 8SE[X] 6x(t) or. (8. therefore. mxV'(x) with = 0. (8.144 Field Theory: A Path Integral Approach In the semiclassical method.6) The Euclidean equations. (8.5) V(x) = ^(x2a2f. Quantum mechanically.m i 2 . correspond to a particle moving in an inverted potential otherwise also known as a double humped potential.
U) 1 .9) where tc is a constant and we have identified as in Eq. (7. (8.12) q2 = V'(xcl).X T a Lu2 U! 4 Ta{X«a2)) 2 = ^ ^ci^ci .52) mw2 = V"(±a) = g2a2. (8. m . ^ ( 4 . in the large T limit.Path Integral for the Double Well 145 executes small oscillations at the bottom of either of the wells in the Minkowski space and these small oscillations can be approximated by a harmonic oscillator motion.5) which play a dominant role in evaluating the path integral.° ) = ^ X c I ( x c i . OUJ 9uj(t (8.10) xc\ = ±—seen 2 aw — tc) 2 ~ ~~2 l ^)^^') =^ ^ .a 2 ) (8. However. we see that . when T —> oo) in which we are ultimately interested in. there are nontrivial solutions to the Euclidean equation of motion in Eq. Then.^ 2a =F. (8. (namely. Let xci(t) = ± a t a n h ^ ~ c' .
it follows that mxcl . . these also correspond to minimum energy solutions like the trivial motion.14) a a J Zi f tc —a t — A —a v(8. The value of the action corresponding to such a classical motion can be easily calculated.16) tc These solutions. We also note that. correspond to the particle starting out on one of the hill tops at t —»• — oo and then moving over to the other hill top at t —> oo. therefore. the Euclidean energy defined in Eq. (8.5). xc\{t—> oo) = Ta. In other words. (8. (8. Let us also note (see Eq.13) and we conclude that the solutions in Eq. (8.7) has the value E = ^mx2clV(xcl) = 0. for such motions.11)) that for such solutions 1 1 2 mx\ = . for these solutions. (8.V'{xc\) = 0 . (8.9) represent nontrivial solutions of the Euclidean equation of motion in Eq.m x — V(xcl) = V(xcl).146 Field Theory: A Path Integral Approach Consequently. xc\{t ^ oo) = ±a .15) & ill Therefore. (8.
3 277TCJ 3 " m a = ~W where we have used Eq. (8.11). They are known respectively as the instanton and the antiinstanton solutions (classical solutions in Euclidean time). A. finite action solutions in the Euclidean space and have the graphical form as shown above.Path Integral for the Double Well 147 SE[XC\] = So = j di Umx2cl + r± V(xd)j dt mx\ — ml dx c i xc\ Cdxa (T£(^a2)) mcu (\ o 9 ±a =pa TTIW 4a 3 ~2alT (8. therefore.18) .17) 2 .uj{ttc 'a2u2 m  —•—seen ma — sech ^ UJ . (8.11) we find t h a t LE = 2 m i d + V(xc\) = mx2cl . 2. .76) in connection with the W K B calculation of the splitting of the energy levels for the double well. The solutions in Eq. We note t h a t this action has the same value as the So defined in Eq. (7. If we look at the Lagrangian for such a solution.9) are. (8.uj{t — tc) (8. then using Eq.
35) of the saddle point method in connection with path integrals. before going into this. the Lagrangian is fairly localized around t = tc with a size of about At. we conclude that (O. we can also have multiinstanton solutions in such a theory. i . This is a direct reflection of the time translation invariance in the theory. However. c l ( t 1 ) ( 5 a .a ) 0 . Just as we can have a one instanton or one antiinstanton solution. stands for One Instanton) (a\e nHT\ . N Vxe~r*SE[x] Ne~*SE[Xcl] Vr)e lh JJ " A 'V 1 ' < 5 a .19) It is in this sense that one says that instantons are localized solutions in time with a size of about . (7.I. The constant tc which signifies the time when the solution reaches the valley of the Euclidean potential is really arbitrary. c l (12 ) ^ t ) . let us calculate the contribution to the transition amplitude coming from the one instanton or antiinstanton trajectory. From our earlier discussion in Eq. 1 _ _ LU y/m ga (8..148 Field Theory: A Path Integral Approach In other words.
as we have already seen in Eq. (8.m ^ + F"(xcl)))oi Here we have denned x(t) = xcl(t) + r)(t). (8.O .9) or (8. / d e t ( l ( . we can define the normalized zero eigenvalue solution of Eq.21) and.11) that dxc\ au> ^f = T sech . (8.9). using Eq. T^oo. (8. ^T^' (8 23)  satisfies the zero eigenvalue equation (see Eq. in a bit more detail.ci)))oc^o(). (8.17).24) as As we had seen earlier in Eq. xd(ttc) = a t a n h ^ ~ c'. (8. r.20) can be obtained from this solution simply as d e t Q (  m ^ + V "(a.20) (8. (8. (8. We know from Eq.22) Let us next analyze the determinant in Eq. (8. (7. for the one instanton case.uj(t — tc) .24) In fact.41). it is clear that rSo ^° (I) ^ ° (8 27)  . the determinant in Eq. for the one instanton case. (7.26) But from the form of the solution in Eq.Path Integral for the Double Well 149 = e*So.20). (8. (8.m ^ + V f c ^ .45) rotated to imaginary time) ( .14).
in the present case ipo(t) happens to be an exact eigenstate of the operator (—m^z + V"{xc\)) with zero eigenvalue. then the Gaussian does not change.29) defines a symmetry of the quadratic action. the term that we need to reexamine is / VVe^ndtldt^(tl)^£&^v(t2).30) . in this case. In other words.150 Field Theory: A Path Integral Approach In other words.28) Note that. Another way to visualize the trouble is to note that if we were to expand the fluctuations around the classical trajectory in a complete basis of the eigenstates of the operator T—rrvr—7TS. n>0 Vrj(t) = J ] dcn n>0 (8. The reason for this is obvious. let us recall that the determinant in Eq.20) arose from integrating out the Gaussian fluctuations. since ipo(t) represents a zero mode of the operator Sx . the transformation in Eq.t •}$* . then we can write 5xci(ti)dxci(t2)' V(t) = Yl cnipn(t). if we make a change of the integration variable as 6r.2 Zero Modes As we have argued before. in this case.) > This is what one means in saying that there is a zero mode in this theory. (829) where e is a constant parameter. (8. 8. Therefore. the determinant identically vanishes. (This means that if>o(±^) = 0 for T — oo.(t) = e^o(t). (8. a zero mode is present in the theory whenever there is a symmetry operative in the system. namely. To see this.t \. (8.
(8. in this case. we have to evaluate the integral more carefully. x cl  = — . The most general classical . we obtain Y\ dcn e n>0 c Ih 2ft 2Jn>0 ^" n = dc ^ / ° / n dcn e~^ En> ° Anc"' ^ . quite analogous to the instanton calculation. (8.3 i ) ^ n>0 where An denotes the eigenvalues corresponding to the eigenstates tpn. consequently. To understand further the origin of the problem.5 2 ( x 2 ) 2 . namely. in fact.Path Integral for the Double Well 151 Substituting this expansion into Eq. (8. there is no Gaussian damping for the dco integration. <7X x ' or. in this case. Let = // J J—oo oo oo da.32) / / Here a is a small parameter and we have defined /(x) = i x 2 .33) This example is. x 2 = x\ + x\ . let us examine a simple two dimensional integral. corresponds to a local minimum.34) It is easy to see that the other solution.28). the origin. (8.i da?2 e«^x) dx1dx2ek«(12*292(*2)2). In such a case. lead to the maximum f=x(lVx 2 )=0. The classical equations. Here we note that the zero mode drops out of the exponent and.
(8. we would obtain the value of the integral in Eq. (8. namely. (8. (8. is a consequence of the rotational invariance of the function / ( x ) in Eq.33). In the present case. if we use the saddle point method naively.32) to be I ~ e ^ > / dr?Q e2° * ^ . therefore. Expanding around the classical solution. we note that dx "\dxc\ = 8^c>2In the matrix form.38) \ (**a\ \^cos0sin# sin Thus. we can write <92/(*cl) dxadx^ n( cos2e cos 6 sin 9 (8.36) ^ ^ V p + 0{rf). 9. the presence of the angular parameter.35) Z2. (8. therefore. we have / ( x ) = / ( x c l + 77) ~ /(x c l ) + L a a = 1. be written as Zl. (8. This is very much like the arbitrary parameter tc which arises in the case of the instantons.40) .34) can.37) From the form of / ( x ) in Eq.33).cl = 25 where 9 is an arbitrary constant angular parameter.cl = 2*7 1 cos sin (8.2.152 Field Theory: A Path Integral Approach solution following from Eq. choosing xa = xatCi + r]a. (8.
::)• «»•<» which we recognize as an infinitesimal rotation. (8. writing out the exponent completely. the quadratic exponent in Eq.44) drhdrh /"OO e~~^ rfl2 J J — OO d?7ie" dff2 J—00 J—00 . (8.45) The analogy with the instanton case is now complete. Namely. we have r^i e i/(*c. (8. —2.) J d7 ?1 d7 ?2 e^ r ' lCOse+??2Sin '') 2 .43) Furthermore. In fact. (8.39) that the eigenstate with zero eigenvalue has the form xo=f Si iV (8.Path Integral for the Double Well m 153 Let us note that the matrix }„& Eq. (8.39) has two eigen values. There is no damping for the dr/2 integration. redefining the variables as rfi = 771 cos 9 + r\2 sin 9 . In this simple example.40) does not change. A = 0. we note that we can write the integral in Eq.41) \ — cos 9 J Consequently. it is easy to check from Eq. .40) does not exist. That is the origin of the divergence and it is a consequence of rotational invariance in this case. under a transformation of the variables of integration of the form '(s)=—'(:r»)=*4 (:. As a result.43) also as j ~ e^/(xci) / / = ea/( x ci)/ /»00 (8. the Gaussian integral in Eq. (8. the solution to the problem is obvious. This is very much like the instanton calculation that we did. In fact. rJ2 — —ill sin 9 + r)2 cos 9 . (8.
tc) + rj(t . x(t + tc) = xcl(t) + rj(t)=xd(t) + ^2cntpn(t). Let us recall that expanding around the instanton trajectory yields x(t) = xd(t . So. Let us discuss very briefly how this is done. in the present case.tc) or. we will like to replace the dco integration by an integration over the position of the center of the instanton.33) is rotationally invariant.154 Field Theory: A Path Integral Approach since the function / ( x ) in Eq. . is a consequence of time translation invariance.47) (Since the trajectory is independent of the center of the instanton trajectory. we have already seen in Eq. The angular integral. namely. the fluctuations must balance out the tc dependence. This method generalizes readily to other systems with more degrees of freedom and is known as the method of collective coordinates.) Multiplying Eq. (8. (8. can be trivially performed giving a finite result after which the saddle point approximation can be applied to the radial integral which will have no zero mode. The divergence. (8.47) with ipo(t) and integrating over time. following our earlier discussion. it is appropriate to use circular (polar) coordinates. This is what we will try to use in order to evaluate the instanton integral. in this case.46 ) which is divergent. 8.31) that / = / dco / II dCn e~ ^ S n > ° X " C" ' •* •* n>0 ( 8 . the position of the center of the instanton can be arbitrary. n>0 (8.3 The Instanton Integral In the case of the instanton.
Path Integral for the Double Well 155 we obtain / 2 Tdtx(t + tc)ip0(t) I T dt \xcl(t) + ^cn^n{t) \ I Mt) 2 " n>0 I.) This simple analysis shows that co = co(tc) • (849) Therefore. 5x(t + tc) = SV = Sc0 Mt) = 5c0 (^) 2 ^ ^ .z * d Imj Xci{t) ~dT +C ° Hi)'**® = co. (It is worth emphasizing here that we are only interested in large T limits when all these results hold. In this case. + c0 (8. (8. let 8V = 6c0Mt).48) The first term vanishes because it has the same value at both the limits. To obtain the Jacobian of this transformation to the leading order.47) arising from a change in the coefficient of the zero mode. we can easily change the cointegration to an integration over tc. we also note that this is precisely the change in the path that we would have obtained to .25). However. (8. let us consider an infinitesimal change in the path in Eq. Namely. (8. (850) where we assume that <5co is infinitesimal.51) where we have used Eq.
in this case. .52).48) that since co(*c)= / 2 dtx(t + tc)i)0{t) dx(t + tc) dco _ 2 f2 T die ~ 7Z dt—^—Mt) = / > ^ * < < > 2 . from Eq.156 Field Theory: A Path Integral Approach leading order had we translated the center of the instanton as tc^tc Namely.54) dtc \ m ' A more direct way to arrive at this result is to note from Eq. we note that to leading order. we can determine the Jacobian of the transformation from the integration variable Co to tc to be i dc 0 (i c ) .x(t + tc) _5t dx(t + tc) ^ ^ dt . (8. /S 0 \ i 2 + 5tc = tc + 5co (±\ 2 . (8. 5x(t + tc) = x(t + tc + 8tc) . (8.T dt /_ x 2 \^r+^Cndr\Mt) \ n>0 ^ ' J 2 n>0 2 .52) /SoY^ \ m J dx(t + tc) dt dxcl(t) Thus. (8.
Thus. (8. (8. the integrand is a total derivative of •0Q(£) which vanishes at both the limits. (8.25). therefore.Path Integral for the Double Well 157 where we have used Eq.56) into Eq.57) can be done trivially. we will leave it as it is for later purposes. (8. that to leading order (since ipo is normalized to unity). (8.57) / y/detf(^m^ + V"(xd))) Here det' stands for the value of the determinant of the operator without the zero mode.54) or (8. Let us also note here that even though the di c integral in Eq. We substitute Eq. dc0(tc) _fs0y+o{h) vr dtc (856) V m Namely. It is clear. Thus.20) and (8.c«r) ^(t2)^) / = g y ndtc f Hdcne^n>0^ci ^ ' ~~~2 n>0 I _T S0\ m /_ 3 T di c (8. we are ready to do the determinant calculation now.57) we obtain the form of the transition amplitude in the presence of an instanton to be . we are using here the fact that the higher moments of a Gaussian of the kind that we are dealing with in Eq. The n = 0 term drops out in the second term because for n = 0. (8. from Eqs.46) to obtain 52S[xcA] L Pt7e^//d*id«2^i)fac.46) are higher orders in h.
')) Clearly.62). (8. let us consider the scattering problem for the Schrodinger equation or. It is easy to see again that both sides of Eq. To evaluate this.59) by dividing it out.158 Field Theory: A Path Integral Approach 8. then we would have evaluated det' since we already know the value of the determinant for the harmonic oscillator. then we obtain _dAW dE ls dek'(Km& =° + V"(xcl))) ' ( j det(£(mg+**. therefore must be equal. (8. if we can evaluate the left hand side of Eq.58). (8. (8. (8. (Mi) .4 Evaluating the Determinant To evaluate det' in Eq.61) If we eliminate the zero mode in Eq. A(E = oo) = l.60) since there is a zero eigenvalue for the determinant in the numerator and further. (8. (m^ + V"(xa)\i> = hEi>. We note that A(£7 = 0) = 0. let us define the quantity where the determinant in the denominator corresponds to that of a free harmonic oscillator which we have already evaluated.59) have the same analytic structure and.
E) > B_ (E) eikt + A. shows that the two coefficients B+ (E) and S_ (E) are identical. In particular.68) . namely. we can write lim f+(t.E). f(t.65) The Jost functions are two linearly independent solutions of the Schrodinger equation in Eq. lim t—»±oo (8.Path Integral for the Double Well 159 If we define the asymptotic solutions (Jost functions) as lim t—>oo f+(t.E)^eikt.E)^elkt.(t.E)) = 2ik B+ (E) = 2ik £ _ (E). This. (8. E) > A+(E) eikt + B+(E) £—•—oo e~ikt. ikt lim t^—oo (8. (8. t—>oo The linear independence of the Jost functions can be easily seen by calculating the Wronskian which has the value W(f+(t. B+(E) = B{E) = A(E). they can also be shown to be equal to A(E'). m V'Cxci) »• muj2.67) where the equality in the last step results from evaluating the Wronskian at the two different time limits t —• ±oo.66) lim /_ (t.63) and consequently. (8. With a bit more analysis.f. (8.64) where we identify .(E) e~ikt. in fact. any general solution can be written as a linear combination of the two.2 hE n kA = uT.
(8. we conclude t h a t A+(E = 0) = 1.63) and (8.72) with t h e asymptotic form of the Jost functions in Eqs.65). (8. (8. (8.i Mt) t—>oo A (8J0) Thus. (8.160 Field Theory: A Path Integral Approach Let us also note from Eq. . B+(E = 0) = 0. The asymptotic equations which the Jost functions satisfy (see Eqs. we note that we can identify (k(E — 0) = — ico) lim f±(t. (8.66) and (8. a result which we already know. E = 0)= £—»oo > e"1*1 .71) Thus.73) . E') = 0. B_(E = 0) = 0.25) t h a t t h e zero m o d e has t h e asymptotic form h m ip0(t) = lim — t—>±oo t—>±oo \ m J 2au (—} where we have defined 2 — sech 2 2 e^1 v 2 ' = KeTu}t. comparing Eqs. (8.69) K = ^ (S) 2 • lim .65)) are md2f+d^ md2f~Qt2E'] E) A _ ( £ = 0) = 1.64) a n d (8. from Eqs. (8. (8.mu.(hE> ~ ™2)f(t. (8.71) a n d (8. E) = 0 .69).64).*)f+(t. (8. Consequently.(HE .72) (8.74) .66). we obtain A(E = 0) = B+(E = 0) = 0.
we obtain > t=oo ^Q^W(f+(t.76) On the other hand. (8.E)) at or ' = m Ef+(t.77) .f(t.E>) p.Path Integral for the Double Well 161 Multiplying the first of these equations by /_(*.E) m (E~E>)f+(t.E)) £=—oo h mK2 (8. (8.66).f(t.EfWlf_it.E) d2 W(f+(t. (8. from the asymptotic form of the Jost functions in Eq.E)f(t.0).E') dt2 or. ) with T — oo.t(B_(E)eikt .E)) dEdt mK2 E=0 = —/ + (i.0). If+frE)—^ h r\ f(t.o) m h ^S).(ewt ik eikt .o)/_(t.0)f{t.0)J.ik + A(E)eikt) A{E)e~ikt) t=oo .we^V**) .75) Integrating this equation between (—f. m or.E))\^co = (e^iikB_(E)eikt + ujeu.(t. E) and subtracting one from the other.Ut.E)f_(t. we obtain m{Mt. W(f+(t.£') and the second by /+(£.^u(^ 2 dt h(EE')f+(t. we see that W(f+(t.0).
E)) dB_(E) dE dA{E) dE 2LU = 2u E=0 (8. then. (8.m ^ + V"(xcl))) d e t ( i ( . Comparing Eqs. (8.79) or. (8.68) as well as the results of Eq.80) The one instanton contribution in Eq.76) and (8.0)J.78) £=0 Here we have used the identification in Eq. we obtain 2LO dA(E) dE dA(E) dE E=0 h mK2 h 2mu>K2 ' (8. E=0 We.72). determine the ratio in Eq.162 Field Theory: A Path Integral Approach from which we determine ^o^W(f+(t. therefore. (8.l.78). (8.m ^ + mu )) 2 = QA(E) dE E=0 h 2mcoK2 (8.58) can now be explicitly determined and takes the form (a\e \HT\ a}o.(t. N d2 det(I(m^+^)) + V»(xd))) det(£(m& + a/>))\ detmm$ 5o^2 m Z 2 dtr 2 .62) to be d e t ' (  ( .
calculate the transition amplitude in the presence of an antiinstanton and it can be shown to be identical to the result obtained in Eq. . separates into a product of two factorsone that of a simple harmonic oscillator arising from the trivial solution of the Euclidean equation of motion and the second giving the true contribution due to an instanton.81) 2 where we have used Eq. (8. The instanton density is small for weak coupling and in such a case these multiinstanton solutions will contribute to the transition amplitude as well and their contribution can be evaluated under an approximation commonly known as the dilute gas approximation. (8. a string of widely separated instantons and antiinstantons also satisfies the Euclidean classical equation given in Eq. A typical example of a multiinstanton solution has the following form. (3. in this case. r.82) The transition amplitude. similarly.28) (rotated to Euclidean space). We have also defined a new quantity.80) as well as the value of the path integral for the harmonic oscillator given in Eq. We can.5). (8. (8.70) is given by r = y[??f K efr = 2yJ^Jae~^.81). whose value using Eq.5 MultiInstanton Contributions As we had discussed earlier. (8.Path Integral for the Double Well I 163 T (8. 8.
_^r(rT)n n! 8.. a n.i . we see from Eq. only a total of odd number of instantons and antiinstantons can contribute to transition amplitudes of the form (a\e*HT\ a).85) . (a\e~*HT\ . the integral over the centers of the instantons gives T ~2 /**1 ftnl rjpn / d*iy d* a y_ f dt » = iz (8 84)  Furthermore."instanton" solution will contribute an amount (see Eqs. (8.84)) (Z) / m a n 3 _<r nTn (mu\\ e 2 r» = —) V 7m / n! \ 7rn / e 2 ^ ^ . Thus. (8.86) Similarly.tn (8.t2..81) and (8.83) In such a case.85 We have to recognize here that only an even number of instantons and antiinstantons can contribute to the transition amplitude of the form (a\eTiHT\a). their contributions to the transition amplitude will simply be multiplicative.a).instanton solution with centers satisfying "2 < t n < * n . (8. (8." < t i < J". {a\e~^HT\a).. &tti.164 Field Theory: A Path Integral Approach h h ' ' £3 h ' ' t5 Let us consider a n.87) Adding all such "instanton" contributions. since the instantons and the antiinstantons are assumed to be noninteracting.
we will obtain (a\e ft  .a  .a). we can show that for odd number of instanton contributions.Path Integral for the Double Well 165 that we will have (a\e*HT\a) E = \2n /rnuJ\\ _ujz(rT) 2 e 2 Y^Ph) ~ 9«! \ixhJ 2nT~ — — fmu\\ (^) e 2 cosh(rT) = _u/r (erT + e~~rT) 6 " 2 2 + e(+r)T^ = I ( ™ ) 3 ^ . then. (8.^ . .* " T  _ ) ( _  _ G) + ( _ a  e .g 8 ) Similarly. . we obtain E± = h(^±r). If we identify the two low lying states of the Hamiltonian as ±) with energy eigenvalues E± respectively.(  r)T _ ( g.88). then we note that by inserting a complete set of energy states we will obtain for large T (a\e~^HT\ a) a) (g90) ~ ( _ a  e .) ( .a) + ekE+T(a\+){+\ Comparing this with Eq.^ T  + ) ( +  _ = e .T ( . (8.91) .gg) = I ( ™ ) ^ ( e .( f ~r)T _ e (t+r)T^ .a) = ( — ) — V 7TAJ. / e 2 smh(rT) (g.
J. the splitting between the two energy levels is obtained to be AE = E+ .. Sakita... Shifman. Oxford Univ. (7.= 2hr /2m = 2hx 2\ —u V n = W2mhLU%a e~*So ..6 References Coleman.E.166 Field Theory: A Path Integral Approach Therefore. "The Uses of Instantons".92) This splitting of energy levels calculated in the path integral formalism can then be compared with the result obtained through the WKB approximation in Eq. ZinnJustin. Phys. 1977. 8. et al. B.77). "Quantum Theory of Many Variable Systems and Fields". Usp. "Quantum Field Theory and Critical Phenomena". Sov. (8. . Press. S. Erice Lectures. World Scientific Publishing. 195 (1982). M. 25.
1) The action generically has the form S[xa] = f f dtL(xa. Let us consider a system with ndegrees of freedom characterized by the coordinates xa(t). However.1 Systems with Many Degrees of Freedom Thus far. 2.Chapter 9 P a t h Integral for Relativistic Theories 9. (9. a = 1. through the couplings S[xa. These coordinates.xa). then the transition amplitude in Eq. (9. n.28) can be easily shown to generalize to {xf. can denote the coordinates of nparticles in one dimension or the coordinates of a single particle in ndimensions.2) and we are supposed to integrate over all paths starting at xf at t = ti and ending at xj at t = tf. if it describes the dynamics of the system). • • • . in this case.ti) = N lvxae*s^. If S[x] denotes the appropriate action for the system (namely.3) . for example. We can also introduce appropriate sources. (2. (9. we have only discussed one particle systems. the method of path integrals generalizes readily to systems with many particles or systems with many degrees of freedom. Ja] = S[xa] + f 167 f dt Ja(t)xa(t).tf\xi.
if <p(x. this allows to derive the various transition amplitudes or matrix elements a simple manner. (4.5) where in the infinite time interval limit.4) us in to as As we have seen earlier in Eqs. in all these discussions.tf\xi. (9.8) (Incidentally. the action in Eq. the vacuum vacuum transition amplitude in the limit of infinite time interval (see section 4. The path integrals can also be extended to continuum field theories once we recognize that these theories describe physical systems with an infinite number of degrees of freedom. (4. (9. as before.4) Z[J] = <00>J = N fvxa eis[*a'ja]. we are going to assume that the relation between the Lagrangian and the Hamiltonian of the system is the canonical one which would lead to path integrals of the form .5) has no end point restriction in the sense that the initial and the final coordinates of the paths can be chosen arbitrarily. We can also define.ti}J = Nfvxae^xa'jaK (9. Thus.7) dt dx J(x. then the vacuum to vacuum transition amplitude in the presence of an external source can be written as Z[J] = (00) J = N j V<\>e^>J].50). (9. t).38) and (4. J] = S[(/>] + ff J J—oo (9.40)) {xf.xa) + Ja(t)xa(t)) . t)(j>{x. (9. t) is the basic variable of a 1 + 1 dimensional field theory.6) oo and the integration over the paths in Eq. (9.168 Field Theory: A Path Integral Approach to define the transition amplitude in the presence of these sources as (see Eq.3) has the form oo / dt (L(xa. where S[<p.
18) and (2. let us assume that §<*<§. This is why it is the vacuum functional which is the quantity of fundamental significance in these studies.20)).both nonrelativistic and relativistic. If this is not the case.Path Integral for Relativistic Theories 169 in Eq.7). Returning now to the question of the functional integration. (2.r + me. namely. in addition.11) .22). we defined the path integral by dividing up the time interval into infinitesimal steps (see Eqs.10) I I I I I I I I I I I I I I I I I I I 1I —k k 2 2 If we now label the position as well as the value of the field variable at any intermediate point on the trajectory as L xm = . (2. Depending on the particular form of the action. let us recall that in the 0 + 1 dimensional case. the path integral in the phase space as obtained in Eq. (9. it is the time ordered Green's functions in the vacuum which play the most important role in a field theory because the scattering matrix or the Smatrix can be obtained from them. 4>{xm) = <t>m. Here. Let us further divide the length interval into N equal steps of length e such that Ne = L. (9. then one should take as the starting point. Thus. (99) with the understanding that we will take the limit L — oo at the > end.. we will be dealing with different kinds of field theories .) Before going into the discussion about the functional integration in the present case. (9. The second point to note is that we have left the specific form of S[<p] arbitrary. 0<m<N. we have to divide up the space interval into infinitesimal steps as well.5) or (9. it is worth emphasizing what we have discussed earlier.
just as in the case of quantum mechanics. However.JV. which we have extensively discussed. then the Green's functions can still be defined uniquely since they are defined as ratios for which the divergent constants simply drop out. the path integral. (2.5) or (9.170 Field Theory: A Path Integral Approach then.15) . (9. For field theories in higher dimensions where the basic variables are cj)(x. is defined by taking a hypercube divided into a lattice of infinitesimal spacing and then identifying the functional integral with a product of ordinary integrals of the field values at each of the lattice sites. where S[(f). N.29). let us choose a relativistic scalar field theory in 3 + 1 dimensions described by the Lagrangian density £(</>. does not exist in the sense that the integrations defined in Eq. in Eq.A( . (9. t)(f>(x.12) are divergent in the continuum limit. in the present case. (9. (9. if we absorb the divergence into the normalization constant.14).13) and the integrations are over the entire spacetime manifold in higher dimension with n denoting the dimensionality of the spacetime manifold.tf 2 .O. for example) lV<j>= lim J L—>oo lim E .t). we can define the vacuum generating functional exactly in an analogous manner.2 Relativistic Scalar Field Theory With all these preliminaries. however.OO /TTd0TO.14) (9.7 ! .^ V . The functional integral. Namely.7).—<f> ^ v . J (9. let us take a specific form of the action in Eq. (9. d^) = d^d^ . t).12) Unlike the case of quantum mechanics. J] = S[<f>} + ! dnx J(x. 9. we can define (see Eq. in such a case. Namely. we have Z[J] = (00) J = N / ' z t y e W .
17) take the form S ^ =W + m20 + ^ _ J{x) = „_ (919) Here we have chosen to represent the spacetime variables in a compact notation of x for simplicity. The EulerLagrange equations following from the action in Eq. (4.19) is also known as the </>4theory.x') = 5A{x . t). namely.15) by integrating over the space variables as L= fd^xC^d^). (9. This is a selfinteracting theory which can describe spin zero particles with mass 77i. 4).20) . and S[<p.17) (9. this theory described by the action in Eq. 8^). t)<j>(x.17) or by the dynamical equations in Eq. (9. (9.) so that S[4>]= fd4x C{4>. In the absence of interaction. (9.17) is quadratic in the field variables and hence the generating functional can be evaluated in much the same way as in the case of quantum mechanical systems (see chapters 2. J] = S[<i>] + f d4x J(x. It is worth noting here that the Lagrangian can be obtained from the Lagrangian density in Eq. when A = 0. However. (9. 3 and in particular. The equation satisfied by the Green's function is {d^ + m2) G(x . (9.63)) to the anharmonic oscillator which we discussed earlier except that it is a relativistic field theory invariant under global Poincare transformations. Commonly. (9.Path Integral for Relativistic Theories 171 with A > 0 (This condition merely corresponds to the fact that we would like the potential to be bounded from below so that the quantum theory will have a meaningful ground state. let us first define the Feynman Green's function associated with this theory.16) This theory is quite similar (see Eq. the action in Eq.18) (9.x').
x1) = — ^ f d4k eik<xx'^. GF(x — x') = lim / — .21) and substituting these back into the differential equation. (9.172 Field Theory: A Path Integral Approach Defining the Fourier transforms as ik(xx') <54(x .76)) lim (dad» + m2e»0+ ie) GF(x . (1.23) We can also think of the Feynman Green's function as satisfying the differential equation (see Eq.75) as eiH**'). if A = 0. Eq. (9. (9.14).24) Going back to the generating functional in Eq.25) .x') = 64{x . The Feynman Green's function or the propagator is.20). then defined following Eq. (9.4) and k2 represents the invariant length square of the conjugate four vector k^.4 2 — V ' e^o+J (2TT l k — m2 + ie d k l (9. we obtain 1 (2TT)2 V (k2+m2)G(k) V ' OT G ' = 1 (2TT)4 ' W . (3.3) and (1. we note that for the present case. then we can define ZQ[J] =N I' V<t> eR So[x ' J1 = iV rV4>eyd4x^W+m2)*J(l>). (3.( 2 ^ A ^ ' (9 22)  Here we are using the scalar product for the four vectors with the metrics introduced in Eqs. (9.x').
for the path integral is normally chosen such that Z[0] = 1. we obtain lim+ l d4x ^ 4>(d^ + m2. with GF defined in Eq. (9. Alternately. t). the normalization constant. quadratic part of the action.i e ) 0 . Substituting Eq.26). we can also define Z0[J]= lim N / ' p 0 e .J ^ ) _ e >+ J —0 (9_26) If we now redefine the variable of integration to be 4>(x) = 4>(x) + f d V GF(x . we obtain (Note that the .27) x (d„d" + m 2 . N.i / d 4 x ( ^ ( 9 M ^ + m 2 .28) where we have used Eq. (9.J{x)4>{x) ~ f d4x'J(x)GF(xx')J(x') (9.28) back into the generating functional in Eq. </>(x. J] representing the free.24). The field.x')J{x'). (9.23). then.Path Integral for Relativistic Theories 173 with SQ[X. t) > 0 . (9.25) should be properly evaluated by rotating to Euclidean space as discussed in section 4.ie) U(x) + / d V G F ( x = lim / d4x d>(x)(dltdti + m2 x")J{x")\ ie)ct>(x) . (9. Let us note here once again that the integral in Eq. x—>oo Furthermore. is assumed to satisfy the asymptotic condition lim 0(x.ie)4> = lim+ / " d 4 x i U(x)+ f d4x'GF(xx')J(x')\ (9.1.
As in the case of the harmonic oscillator. we note that when A = 0.29) Here we have used a generalization of the result in Eq. (9. (9. we obtain once again the result that the Feynman propagator is nothing other than the time ordered two point correlation function in the vacuum (see Eq. 2 n Z0[J] GF(xy))Z0[J] S2Z0[J] SJ(x)SJ(y) j=o Z0[J] j=o = ihGF(x — y). <ow*)o> = Zo[J] 6J{x) J=0 J=0 — u.62)). (4. .27) is trivial.2) for a field theory.30) Namely.174 Field Theory: A Path Integral Approach Jacobian for the change of variable in Eq.eyd*x(fr(xKd^^2ieMx)J(xMx)) *dVj(z)GF(x*')J(z') X N [ V4>e~TiJ 'd*x ^)(9^+m2ie)4>(x) = 7V[det(d/i<9'J + m 2 )]~^ X e^K If d4xd*x' J(x)GF(xx>)J(x>) = ZQ[0} e^K If d4xd4x ' J X G ( ) F(XX')J(X') _ (9. (4.) ZQ[J}= lim N = lim e^//d 4 [v<j.
ei/d^(l9^a^2#^iT^4+^) = N / V<j>(eT%\fd*x<fiHx)\ JS0[<j>. Therefore. We note that we can write (as we had also noted earlier in Eq. (4. A power series expansion in A for the generating functional in Eq. we have Z[J} = iXh3 f . J}.68) in the case of the anharmonic oscillator) <j>{x) 5J(x) (9. (9.J] ) = (V^/d4*(~^)4) A f^e^IM T mld4x(ih S 14 5J(x) 1 = (« )Z0[J}.69) for the anharmonic oscillator. However. we can evaluate it perturbatively at least when the coupling is weak. quadratic action SQ[(J>.32) Once again. the ^theory does not also have a closed form expression for the generating functional. we can rewrite the generating functional of Eq.31) when acting on the free. i\h6 f ax 64 8J (x) 4 + i 2! • 4! J (/ 5 J 4 (x) d y SJHv) + • Z0[J] . (4.Path Integral for Relativistic Theories 175 Just as the path integral for the anharmonic oscillator cannot be evaluated in a closed form. we note that this is very analogous to the result obtained in Eq. Thus. (9. (9.32) follows by Taylor expanding the exponential involving the interaction terms.13). also as Z[J]=N /"p«/. in the present case.
29) and (9. only the even order Green's functions will be nontrivial in this theory and let us calculate only the 2point and the 4point functions up to order A. (9.36) { ^rj JjHx)) i\h3 f . we conclude that the vacuum expectation value of the time ordered product of an odd number of fields must vanish in this theory. (9.A 84 dx 4! . (4. Let us recall from Eq. By definition.37) .32) (namely. <or(^(xi)0(x 2n+ i))o) _ (tfi) 2 "* 1 5^Z[J] Z[J] 5J(Xl)6J(x2n+1) J=0 ' l • } Consequently./ " ~ SJ4(x)/ (9.50) that. by definition. 52Z[J] 5J{x1)6J(x2) j = o Keeping terms up to order A. In other words. from the fact that it is invariant under J <> — J).^ J J d4xid4x2 64 Id4ysm) J(xt)GF(xix2)J(x2) + (9.176 Field Theory: A Path Integral Approach = Zo[0] i\h3 r d4x s4 i_ + i\h3\2 5J4(x) 2! ' 4! J dV 4 '<JJ (x) ( / X e . from the symmetry of Z[J] in Eqs. let us derive some of the Green's functions to low orders in the coupling constant for the 0 4 theory.33) that {ihf {0\T(</>(Xl)(f>(x2))\0) = Z[J] Z[J] ~ Z0[0] l (9.33) To obtain a feeling for how the actual calculations are carried out. we note from Eq. <or(0(xiMx 2 ) (ih)n Z[J] <f>(Xn))\0) 5nZ[J] 6J{Xl)5J(x2)6J(xn) J=0" (934) Furthermore.
ix 2 )J(x2) > \ / ~/T 2r G F (0)G F (0) 6i + ^GF X I / dAXiGF{x — X4)J(X4) 1 ( / d 4 x 3 G F (x .39) +^ x f / dAx5GF(xx5)J(x5) x e~2^ //cl4a. i d 4 a . F (a:ia:2)J(x2) _ With some algebraic manipulations.i)GF(xix2)J(x2) .) U ^ W d4x .x3)J(x3))( d4x4GF(x x4)J(x4)j (9.F( i.2J(3.^ r ff d 4 iid 4 » a J( 3 . 2J(a.2J(xi)G.x 3 )J(x 3 ) jf / d 4 z 4 G F (x .x 4 )J(x 4 ) )( / d 4 a./ __3 4r aX 2 ^ f ^ r . let us note that S2 / / d 4 xid 4 :r2 e 2R 2 SJ (x) 5 8J{x) \ fj x e as J(xi)GF(x1x2)J(x2) h ( / d4x3GF(xx3)J(x3) d4xld4x2J(x1)GF{xlx2)J(x2) •He.Path Integral for Relativistic Theories 177 To evaluate this.i)G .38) X e ^i// d 4 3 .2)J( 2) 5J (x / 4 / d 5J {x) 7 <Jj2(a.id4a. this.(0) pf d4x3GF(x . leads to the result <54 4 4 :c :E :r d x 4 e 2ft //d x 1 d a.x 6 )J(x 6 ) (9. then.i)G P ( a . 6 G F (a.
37). (9.xi)GF(x .40) X e~^h N d4xid4x2J{xi)GF(xiX2)J(x2) It now follows that Z[0] = Z0[0] ( l + ^ G F ( 0 ) G F ( 0 ) y d 4 *) . the divergence can be absorbed into the normalization constant.42) . From Eq. (9.x2) (9.41) This is clearly divergent and as we have argued earlier.x5)J(x5)\(  x6)J(x6) (9.40) we can also calculate to linear order in A d2Z[J] SJ(xi)6J(x2) Zo[0] j=o + ^GF(0)GF(0)Jd X I I rGF(Xl ~X2] + ^ G F ( O ) J d4x GF(x . (9.GF(x — X±)J(xt) / d4x5GF(x / d4x6GF(x x .178 Field Theory: A Path Integral Approach Putting this back into Eq. we obtain the generating functional to linear power in A to be Z[J] = Zo[0] 1 + (0)G F (0) ^GFK (fd4x3GF(x fdAx + \GF (0) fd4x  x3)J(x3) x I / d4X4GF(x — X4)J(x±) I — / d 4 x I / d4x3GF(x  x3)J(x3) X I / d4X4.
40) that . we have only kept the leading order term coming from the expansion of the denominator since we are interested in the 2point function up to order A. (9. Next. we note from Eq. of course. It is worth pointing it out here that GF(0) is a divergent quantity as we can readily check from the form of the propagator.Path Integral for Relativistic Theories 179 Therefore. to linear order in A.43) Here in the second term. is a first order quantum correction.xi)GF(x + ^GF(0) ~/i2 . indeed.^ G F ( x i . the quantum corrections in a field theory lead to divergences which are then taken care of by what is commonly known as the process of renormalization.23). (9.x2) fd4xGF(xXl)GF(xx2) . let us calculate the 4point function up to order A.x 2 ) + ^G F (0) \h2 f —GF(0) — ihGF(xi — x2) / d4xGF(x — x\)GF{x — x2). To leading order. on the other hand. we find that the first order correction to the propagator in this theory is divergent. the Feynman propagator for the free theory defined in Eq. (9. Thus. We note that the first term is. The second term. namely. This is. a general feature of quantum field theories. we obtain (0T(^(ari)^(x 2 ))0> {ihf Z[J] 52Z[J] 5J(xi)5J{x2) J=0 h2 *fGF(0)GF(0)fd*x) GF(xi x2] Z0[0](l + x Z0[0} 1 + ^ G F (0)<M0)/ d4x J d 4 x G F (x .
34) and keeping terms only up to order A.x2) > — — / d4xGF(x — x\)GF{x — X2)GF{x — x3)GF(x — £4) (9.x3)GF(x — X2)GF{x — X4) + GF(xi .180 Field Theory: A Path Integral Approach 64Z[J] SJ(xi)5J(x2)5J(xs)5J(x4. (9.x3yj .) j=o l = Z0[0] .1 (l + x (GF{xi .xi)G F (x .a.x2)GF{x3 .x±) ^GF(0)GF(0)Jd4x) + GF{x\ .^ G F ( 0 ) / d4x{GF(xi .3)GF(a.xA) + GF(xi .x 2 )G F (x .X4)GF(x2 .x3) + GF(x2 — xs)GF(x — xi)G F (x — X4) + GF(x2 . .X4)GF(o.Xi)GF{x — X2)GF(x .x 3 )G F (a.a.44) Substituting this back into the definition of the 4point function in Eq.2 . we obtain <0T(<Kzi)<Kz2)#c3)<Kz4))0> {ih)4 Z[J] 84Z[J] SJ(xi)6J(x2)SJ(x3)6J(x4) j=o . .£4) + G F (xi .4)GF(x — xi)G F (x — £3) + GF(x3 .
45) 9.xi)GF(x .x4)GF(x2 —GF(0) / d 4 xG F (a.x2) > .x2.xz) . The Feynman rules do precisely this.Xi)GF(x . (9.x2)GF(x .x2)GF{x .x2)GF(x + GF(x2 .Path Integral for Relativistic Theories 181 = h2[GF{xi x2)GF(x3  xA) x3)) + GF(XI x3)GF(x2 .Xi)GF(x .i .xi)GF(x + GF(x3 . ^ ^ X 3 = V(xi.xz)GF{x .xz)GF{x .x4) —iXhs / d4xGF(x — x\)GF{x — x2)GF(x — x^)GF(x — X4).x3) .a?4) .3 Feynman Rules These lowest order calculations are enough to convince any one interested in the subject that a systematic procedure needs to be developed to keep track of the perturbative expansion.x4) + GF(xi . Let us represent these diagrammatically as xi X\ X2 x2=i^GF(xi £4 x2).x3.x4) .x3)GF(x + GF(x\ .x4) + GF(xi . Let us note that the basic elements in our <?!>4theory are the Feynman propagator for the free theory and the interaction.x\)GF(x + GF(x2 .x4)GF(x .
47) There is one final rule. For the present case. With these rules.yi)GF(y2 . (9. (9. in this case.2/4) = (ift)3 / dV!ftd 4 Jted 4 Iftd 4 i/4Gp(si .2 / 3 ) % . we must integrate over the intermediate points where a vertex connects with the propagators. (9. Let us further use the rule that in evaluating such graphs.2/i)<% .2/4) = Xh2GF(0) J d4?/ GF(X! . (The symmetry . then the true value of the diagram is obtained by dividing with this symmetry factor.2/3. .47) is invariant under a rotation by 180°. we can obtain the value for the following simple diagram to be X\ 2/1 2/2 X2 = / d4yidAy2d4y3d4y4ihGF(xi x ihGF(y3 y4)V(yi. The symmetry factor.3/4) ( .yi)ihGF(y2  x2) y2.182 Field Theory: A Path Integral Approach _i_ 64S[<p] h 5(j){x\)8(j){x2)8(t>{xz)6(j>{xi) iX f = — — / d4xS(x — x\)5{x — x2)6(x — xz)5(x — X4). we can construct various nontrivial graphs by joining the vertex to the propagators. is 2 1 = 2.x2). then. that if the internal part of a diagram has a symmetry. the internal bubble in the diagram for Eq. It is clear that given these basic elements.x2) x GF(y3 .46) The interaction vertex is understood to be the part of the graph without the external lines or the propagators.2 / 2 ) % ." ^ % . Namely.y)GF(y .
Thus. (9. we can write <or(<K*i)<Kz2))o> <OT(<MX1)0(X2)<XZ3)<MX4))O) Xi X3 X2 XA Xi + X2 Xz XA X\ + X2 X4 X3 X3 + X\ \J XA X2 V XA X2 \J Xz X\ X2 + X\ X3 + X\ Xi + X2 \J XA XI \7 X9.43) and (9.Path Integral for Relativistic Theories 183 factor for a Feynman diagram is the most difficult to determine by naive inspection and should be obtained through a careful evaluationwhen necessary. XA Xz + x2 Xi + Xz \J Xo XA X\ XA + x2 X ^ xz (9. we now immediately recognize from Eqs.49) . for example.48) which we recognize to be the first order (linear in A) correction to the propagator in Eq. (9. (9. going back to the Wick expansion of field theory. (9. we obtain the value of this diagram to be Xl^—x2 = >^GF(0)Jd4xGF(xlx)GF(xx2).45) that up to order A.43). Each such diagram that can be constructed from the basic elements in Eq.) Dividing by this factor.46) is known as a Feynman diagram of the theory and corresponds to a basic term in the perturbative expansion.
The Feynman diagrams.<o0(xi)o) <o0(x2)o> = <OT(0(a. (4.51) J=0 Similarly.50) generates connected Green's functions. (9. we obtain 82W[J] ih 5J(xi)5J(x2) = (ih) j=o 1 S2Z[J] _Z[J] 8J(xi)5J(x2) 5Z[J] 5Z[J] Z [J] 5J{xi) 5J(x2)_ J=0 2 = (or(0(xi)^2))o) . J=0 (9. W[J] = ih\nZ[J\.i)<Kz2))O)c. (9.52) . as can be seen from the diagrams in Eq. It is clear from the above simple example that the generating functional Z[J] generates Green's functions which contain disconnected diagrams as well. are known as disconnected Feynman diagrams. the logarithm of Z[J] generates Green's functions which contain only the connected diagrams (otherwise known as the connected Green's functions) and these give rise to the physical scattering matrix elements. (9.50). As we have discussed earlier (see the discussion following Eq. clearly consist of two classes of diagrams. for the two point function. We note from Eqs. Namely.4 Connected Diagrams The Feynman diagrams. one where each part of the diagram is connected to the rest of the diagram and another where parts of the diagram are disconnected.51)). (9.184 Field Theory: A Path Integral Approach 9.49).35) that 6W[J] 5J(xi) ih 5Z[J] Z[J) 6J(Xl) = (0^i)0).34) and (9. (9. which consist of parts that are not connected to one another.
up to order A. as we have discussed earlier in Eq.52) that (0T(^(xi)^(x 2 ))0) = (Or(<^(x1)0(x2))O)c (9. for the 4point function. Consequently.55) Similarly.)O> = 0 .W ( * i )  0 > <OT(0(x 2 Mx 3 ))O) (O0(x 2 )O)(Or(0(x3)0(xi))O)(O^(x3)O>(OT(^(xi)^(x 2 ))O) + 2(O^(x1)O)(O0(x2)O)(O0(x3)O) = (0\T(<t>(xl)cf>(x2)4>(x3)\0)c.Path Integral for Relativistic Theories 185 In a similar manner.53) and so on. (9. of course. it can be shown that {ihf^2 63W[J\ SJ(xi)5J(x2)5J(x3) j=o = < 0 r o ( s i M x 2 M x 3 ) )  0 ) . Note that for the ^ 4 theory. with some algebra. We can. check explicitly that the connected Green's functions involve only connected Feynman diagrams as follows.35) <O0(a. (9. (9.54) Xl X2 +x1^—x2 • (9. we have <OT(^(xi)0(x 2 )^(x 3 )0(x4))O)c = (OT(0(x1)0(x2)^(x3)(/>(x4))O) <Or(^(xi)0(x 2 ))O)<OT(^(x3)^(x 4 ))O> <or(^(xi)^(x3))o)<or(^(x2)^(x4))o) (0r(^(x 1 )^(x4))0)(0r(^(x 2 )^(x 3 ))0> Xl X4 = x2Xx3 <»•*» . we note from Eq.
Proc. J. USA 37. it contains diagrams that are reducible to two connected diagrams upon cutting an internal line. X\ X2 + X\ 0 . Schwinger. Nuovo Cimento 34. we can write down the diagrammatic expansion of the connected 2point Green's function up to order A2 in this theory simply as Xl^Y/JX2 ®~. . K.. 1790 (1964). we see that W[J] generates connected Green's functions.186 Field Theory: A Path Integral Approach Thus. 452 (1951). 9. JonaLasinio. Thus.5 References Huang. "Quarks.. the third graph is reducible upon cutting the internal propagator. 455 (1951). J.. X2 + X\ 00 X2 + Xl (^J X2 + X\ x2(9. Note that while W[J] generates connected diagrams. in the 2point function represented above. We will take up the study of the 1PI (one particle irreducible) diagrams in the next chapter. Natl. Sci. Given this.57) The organization of the perturbation series now becomes quite straightforward. It is clear that the IP irreducible diagrams are in some sense more fundamental since we can construct all the connected diagrams from them. Such diagrams are called IP (one particle) reducible. Leptons and Gauge Fields". World Scientific Publishing. ibid 37.
3) where we have identified e~fi with the generator of spacetime translations. 187 .4) or. that it satisfies ei p a .\0) = 0) (10. (10. Assuming that the vacuum state in our Hilbert space is unique and that it is Poincare invariant.1 The Classical Field As we saw in the last chapter. namely. (9. let us note that we can write (O0(x)O) = (0\eipx<p(0)e^px\0). We note that the one point function in the presence of an external source is given by For the ^ 4 theory.1) where W[J] generates connected Green's functions. the generating functional for Green's functions in a scalar field theory is given by Z[J] = eiwW = A fvte&W T .35) that this vacuum expectation value of the field operator vanishes in the absence of external sources (namely.Chapter 10 Effective Action 10. In general. when J = 0). we have seen in Eq. (10. however. PM0) = 0.
J)e^^ = 0.5) Thus. (10. To understand the meaning of the classical field as well as the reason for its name. from the symmetry arguments alone. As we will see later. Namely. in general. Let us denote this by and note that it is indeed a functional of the external source.8) .6) The value of the one point function is quite important in the study of symmetries.4) that (0\<p(x)\0) = <O0(O)O> = constant.3) and (10. independent of spacetime coordinates. this constant coincides with zero. let us analyze the generating functional in Eq. Since Z[J] is independent of (f>(x). be a constant. we conclude that the one point function can. (10. For the 0 4 theory. The field (f>c(x) which is only a classical variable is known as the classical field. SZ[J] = N jv4>%5S[(t). (10. under such a change. the vacuum expectation value of the field operator is a functional of the source and need not be zero. In the presence of an external source. (10. an arbitrary. a nonvanishing value of this quantity signals the spontaneous break down of some symmetry in the theory. namely. infinitesimal change in <j){x) in the integrand on the right hand side will leave the generating functional invariant. in some detail.1) Z[J] = N fvteiW'^. (10.188 Field Theory: A Path Integral Approach we obtain from Eqs. we have <0^(ar)0> = 0 . however.
(If it does change.12) Using Eq.11) where the quantity F{(f>{x)) depends on the specific dynamics of the system and for the (^theory. (10. (10.13) (10. it now follows that we must have N /p^^JMeis[M = o. we note from.11) in Eq. that it has the particular form WW) = . (10. Such a term plays an important role in the study of anomalies.   = id^ + m2)t(x) + ^(x). then. (10. Eq.^ ^ = F(<p(x))J(x)=0.) Since the relation in Eq.1) does not change under a redefinition of the field variable.8) must be true for any arbitrary variation 6<p(x) of the field variable. then we will have an additional term coming from the change of the measure. we obtain J o(f>{x) = N fvcj) (F(cf)(x)) . (9.J(x)) e ^ ' J ' = 0. Eq.14) . that the EulerLagrange equations of a theory hold only as an expectation value equation (Ehrenfest's theorem) or more explicitly.15).Effective Action 189 Here we are assuming that the functional integration measure in Eq. Let us recall that ' ms8J(x) = M fv<l)(l)(x)e^'J\ M (10.17)) .9). (9. (10. for example. < 0 'w 0>J =°S (iai0) The EulerLagrange equation (the classical equation) has the generic form (see.9) This merely expresses what we already know from our study of quantum mechanics.8) or (9. Namely. (10. (10.
(10.12). It is quite different from the classical EulerLagrange equation in Eq.J(x) = (8^ + m2)4>{x) + ^ 3 ( x ) . let us consider specifically the </>4> theory. the EulerLagrange equation takes the form F(4>(x)) .J(x) = 0. In this case.J(x)] eiwW = o .11).15) „ /5W[J] . F ( </>C(:E) — ih 5 8J(x) This is.15) as well as Eq. (10..16) which is the familiar classical EulerLagrange equation in Eq.190 Field Theory: A Path Integral Approach which allows us to use the identification (x) — —ih> 5J(x) ' using which.16) in the h — 0 limit. . 5 or. as we have noted earlier in Eq. (10. (10. e SJ(x) . (10.13) also as F[ ih ih •±W[J] 6 d SJ(x) 6 5J{x) F ih J{x) J(x) Z[J] = 0 e*wW = 0 or. or. the full dynamical equation of the theory at the quantum level. (10. we can write Eq. To get a better feeling for the quantum equation in Eq.15) reduces to the form F(<t>c(x))J{x) = Q.ih V 5J(x) 5J(x) or. (10.11). But let us note that in the limit h — 0. F { „ WL J . of course. the • complete equation in Eq.J{x) = 0 J(x) = 0. (10. (10. It is for this reason that 4>c(x) is called the classical field.
These equations are also known as the SchwingerDyson equations and the BetheSalpeter equations are a special case of these equations.18) We can think of Eq. (d^ ~ J(x) \H2 52<j>c(x) ^rMx)ij{x)~^sj(X)6j(x)=0In terms of VF[J]. the expressions become ill defined and have to be regularized (defined) in some manner.18). The equation involves second and third functional derivatives at the same spacetime point which are not at all well defined. We should note here that in quantum field theory. By taking higher functional derivatives. S . the dynamical equations satisfied by various Green's functions of the theory. whenever there are products of field operators at the same spacetime point.18) as the master equation governing the full dynamics of the quantum theory. 3! 6J3(x) (10. this can also be written as (m7) <^>«H(™) W i\h5W[J}S2W[J] ~ 2! 5J(x) 6J2(x) _ Xh2 63W[J] = 0. One needs to develop a systematic regularization procedure to handle these difficulties. The manifestation of this problem is quite clear in Eq. (0^ + m2) ^ c ( x )  itifij^r) + 3\{Mx)ih6J&) + m2)<f>c(x) + ^4(x) iXH SMx) ~J{X) = 0 or. the quantum equation takes the form or. (10.Effective Action 191 Consequently. The discussion of these topics lies outside the scope of these lectures. we can determine from this. (10.
x')J(x') + A fd4x'GF(x + . First. (10. the iterative solution for the classical field can be written .x{) H fd4x1d4x2d4x3d4x4GF(x .17) (d^ + m2)4>c(x) + ^ c ( z ) .19) which is the classical EulerLagrange equation. all such > ill defined terms vanish and we have from Eq.^ ^4{x"))j .x')<l>l{x'). that in the limit h — 0. (10.x')J(x') .x"){J{x") = . (10.fd4x'GF(x .x') x (^Jd4x"GF{x> .24) (Green's function) as follows.21) We can diagrammatically represent this if we introduce a vertex describing the interaction of the field with the external source as x ^ _ % SS[<t>. we note that the solution for <f>c(x) can be written as an integral equation Mx) = J#x'GF{xx'){J{x') ±/c(x')) = . on the other hand. let us note that we can solve this equation iteratively through the use of the propagator defined in Eqs. The iterative solution has the form </>c(x) ~ . (9. (10.J\ h 6</)(x) = 0=0 "• l J(x).fd4x'GF{x .22) In this case.x2)GF(xi .x3)GF(xi (10.J d V G f ( x .20) which can be solved iteratively. Furthermore.x')J{x') + ^ f d V GF(x .192 Field Theory: A Path Integral Approach Let us note.23) and (9.Jix) = 0.x 4 ) J(x2) J(xz)J{xi) x GF(xi .
the classical field is a functional of the source J(x). (1023) In other words. However. It is. as we have seen in Eq. This relation indicates that the variables J(x) and 4>c(x) are in some sense conjugate variables.2 Effective Action The classical field. therefore. is defined as SW[J] 6J(x) 4>c(x). x ~ ^ 1 .24) .* + . The combinatoric factors cancel out when taking functional derivatives of (j)c with respect to sources and we obtain the npoint tree amplitudes. let us define a new functional through a Legendre transformation as r[0 c ] = W[J] f d4x J(x)(j>c{x). we can also invert the defining relation for the classical field and solve for the source J(x) as a functional of $c(x) at least perturbatively. (10. As we have argued earlier. 10.7).Effective Action 193 diagrammatically as 4>c{x) = % x + 3! x + ^ . (10. the classical field in the limit h —> 0 generates all the tree diagrams or the Born diagrams. In fact. also known as the Born functional or the generating functional for tree diagrams.
we have used the chain rule for functional derivatives as well as the definition of the classical field. • In the framework of the effective action. To understand the meaning of this new functional. (10.4 sj(y) I ^ = J V SM^J sM^)My)~J{x) V dy x 5J(y) 5Mx) J SMx) SJ(y) J[X) In this derivation. (10. we see from Eq. defines the source as a functional of the classical field and has the same structure as the classical EulerLagrange equation for a system in the presence of an external source.12) that it is given by It is for this reason that r[(j>c] is also known as the effective action functional. then. then we can write 5 f . Note that as we have discussed earlier in Eq.27) This is an extremum equation and is much easier to analyze to determine whether a symmetry is spontaneously broken. let us note that if we treat <j>c(x) as our independent variable. 4 Hc{y) d*y SJ{x) " / J " 6J{x) & SMy) . (10. when J — 0. (10. (10. r[0 c ]. indeed.11) and (10. that ar[</>c] sw[J] /• .25). = 0.25) that the value of this constant is determined from the equation sr[<t>c} S(pc(x) c (x)=const.5). Let us recall from Eqs. • <t>c(x) — constant.194 Field Theory: A Path Integral Approach It is clear. We note that Eq.
31) We recall from Eqs. recalling t h a t when J = 0. (10.30) 8nT[4>c .) ^(2)r(2) 1.25). Thus. (10. we have from Eq.1 + E . is t h e inverse of the propagator at every order of the perturbation theory.33) T^ 1 = G. then.(xi) • • • 8(j)c(xn) p(n) _ we can write Eq. (10.29) Introducing the compact notation W^ = 5nW[J] 5J(x\) • • • 8J(xr (10.30) and (9. writing r (2) = if} + E = G. (10. (10. <j)c(x) = <pc = constant.31) W& j=o or.29) also in the compact form (We can view this as an operatorial equation where the appropriate coordinate dependences will arise by taking matrix elements in the coordinate basis.52) t h a t the full propagator of the theory is denned to be W< 2 > J=0 Furthermore.Effective Action Using this in Eq.34) . (10. (9. (10. we obtain 195 (sry>c 6J(y) 4 \54>c(x) = S\xy) or. T^  . GT^ = 1.32) In other words. < • / d z / S2W[J] 52T[cpc} 8J{y)8J{z) 5<f>c(z)5ct>c(x) 84(xy). (10.
(10.29) d42 . Given the relation in Eq. we have from Eq. 52W[J] L J <52LyLJ r[^c / = S4(x .36) we have the diagrammatic relation for the propagator as + xcyoy +•••• (io37) It is clear from the above relation that £ is nothing other than the IP irreducible (IPI) 2point vertex function.33) GiG'1 + £) = 1 or. (10.T S2W[J] 63F[(PC] ' 62W[J] J 8J{y)8J{z) 5<l>c{z)8<l>c{x)8<l>c[a) 8J(a)SJ(u) " . (10.35) Introducing the diagrammatic representation ^ =~ 0 .196 Field Theory: A Path Integral Approach where £ denotes the quantum corrections in r( 2 ) . G = 1 G^ + E 1 1 G —1 F /"(— 1 lrF v 1 1 t_r F /^r— 1 CJF /~t— 1 v 1 (jrF /"(—1 v l /"(IJTJ = GF GFZGF + G F E G F £ G F + • • • .v) 6J(y)8J(z) 5M*)5Mx) t3mj] we can differentiate this with respect to 5fi \ to obtain d4„ s*m} K / _ SJ{u>)SJ(y)6J(z) 8<f>c{z)8<f>c(x) /H^H4. It is also known as the proper self energy diagram. (10. .
we can write this equation also as W(3) = W(2)W(2)W(2)F{3) _ {WA0) Furthermore.39) or (10. introducing the diagrammatic representations.y. —0—y =ihwW{x.41) we note that Eq.39) In compact notation.Effective Action 197 Recalling that we can also rewrite this equation as 63W[J] SJ(x)SJ{y)6J{z) /dVdVdV~^[J1 5J{x)5J{x') 53T[^c pww S2W[J] 5J(y)6J(y') SJ(z)SJ(z') (10.y) j=o (ih)2W^(x.z) (10. (10.y.z) > j=o y / z y =rW(x.40) can also be diagrammatically represented as (10.42) > .
it is IPI).43) where (we are assuming (j>c = 0 in the above expansion) 1 vEli ' ' " > •En) is the proper (IPI) npoint vertex function of the theory. A similar calculation for the 4point connected Green's function leads to the diagrammatic relation X V y W w / \ . it can also be expanded alternately in powers of the derivative or momentum like the classical . + permutations Thus. It is for this reason that T[<pc] is known as the IPI generating functional. we see that we can expand the effective action functional as 00 f 1 71=1 (10. Since r[^ c ] is an effective action.198 Field Theory: A Path Integral Approach This shows that r( 3 )L gives the proper 3point vertex function (in other words. Let us note here that the IPI vertex functions with suitable external wave functions lead to the scattering matrix of the theory.
this condition becomes equivalent to dF eff (0 c ) = 0. then the classical field takes a constant value (/>c(x) — 0C.45) where the terms neglected are higher order in the derivatives.27) <5r[0c S(f)c(x) <pc(x)=(pc = 0.Effective Action 199 action. (10. The quantity f d4x which represents the spacetime volume is also conventionally written as d 4 x = (27r)4(T(0).47) Let us also note that the constant value of 0C = (0) when the sources are turned off is obtained from the extremum equation in Eq. (10. (10.48) In terms of the effective potential. leading to r[0 c ] = . we expect to be able to write r[0 c ] also in the form (10. in this limit.45) vanish. In this limit all the derivative terms in the expansion in Eq.F(0)) . (10.^ 0 2 . Thus. (10. Let us recall that when the sources are set equal to zero.46) In other words.^ 4 = Jd4x Q ^ 0 ^ 0 . (10.49) 50 c . / • (10. since the tree level action has the form 5[0] =  d 4 x Q ^ 0 ^ 0 .j d4x Ves(0c) = Kff (0c) J d4x.44) r[0c] = Jd4x ivMMx)) + lAfaWWMxWMx) + •••). the effective action simply picks out the effective potential including quantum corrections to all orders.
this equation is much easier to analyze than a functional equation. quite important.50) The study of the effective potential is. for most practical calculations.) For the (f)4theory.200 Field Theory: A Path Integral Approach which is a familiar extremization condition from the study of classical mechanics. only ordinary partial derivatives are involved in the above extremum condition. It is particularly useful in analyzing when the quantum corrections can change the qualitative tree level or classical behavior of a theory. However. therefore. (This simply involves taking Fourier transforms. we should integrate over the inter . We also note that the renormalized values of the masses and the coupling constants (including all quantum corrections) can be obtained (in this theory) from the effective potential simply as 84>l d4VeE = 4>c={<t>) m R = 4>c = <4>) \R (10. it is quite useful to work in the momentum space. Note here that since </>c is an ordinary variable (namely. In this sense. the momentum space Feynman rules take the form P ihGF(p) = lim 2 e>o p — m 2 + ie' p2Xp = 'J 54 (Pi +P2+P3+P4) • (1051) In evaluating a Feynman diagram. for example. 10. The Feynman rules can be readily generalized to the momentum space given the rules in the coordinate space. it does not depend on spacetime coordinates).3 Loop Expansion We have already described the Feynman rules in the coordinate space.
Let us next recognize from the form of the exponent in the path integral that the quantity which determines the dynamics of the system is \cfa drf) = I Q ^ W . we have not explicitly included the ie term although it should always be kept in mind in evaluating the integral. In writing the propagator. On the other hand. let us evaluate the 1PI 2point vertex function at order A. suppose we are considering a proper vertex diagram (1PI diagram) . Thus. the consequence of this is that each vertex has a factor of ^ associated with it.51). let us simply note that the calculations indeed take a simpler form in the momentum space. we obtain 1 / i\\ f d4k r4/ . We will discuss the actual evaluation of the integrals later. k 1= Pi SL ih P2 According to our rules.52) simply corresponds to the symmetry factor of the diagram which we discussed earlier. the momenta of the internal propagators. namely. for example.53) The Planck's constant which measures the quantum nature of an amplitude comes as a multiplicative factor in the exponent. comes multiplied with a factor of h.Effective Action 201 mediate momenta. (10.^ 0 2 .^ .46) and (10. . For the present. = 5^»»/^i^b (m52) Note that the factor of  in front of the integral in Eq. Thus. (9. the propagator which is the inverse of the operator in the quadratic part of the Lagrangian. As we have seen in Eqs. (10..
Therefore. therefore. Of course. (10. However. This follows mainly from the fact that the expansion parameter. Since there are / internal lines. unaffected by any such separation. Since there are V vertices. there must. The loop expansion provides a valid perturbative expansion simply because h is a small quantity.55) But the number of independent momentum integrations precisely measures the number of loops in a diagram and from the above relation we note that the number of loops associated with a diagram is related to the power of h associated with a diagram. the ^functions will effectively reduce the number of momentum integration by V — 1. then the total number of h factors associated with such a diagram is given by P = IV. therefore. This is particularly useful if the theory exhibits spontaneous . there are no external propagators or legs. let us note that in a proper vertex diagram. Let us also calculate the number of independent momentum integrations associated with such a diagram. Consequently. there will be as many momentum conserving <5functions. Hence. the number of independent internal momentum integrations will be given by L = I(Vl) = IV + l = P+l.202 Field Theory: A Path Integral Approach with V vertices and / internal lines or propagators. This expansion is quite useful and is very different from expanding in powers of the coupling constant. such a diagram will behave like ~ hp. be / momentum integrations. The loop expansion is. (10. all the momenta associated with the internal lines must be integrated.54) In other words. h. the number of loops exceeds the power of h by one. it is insensitive to how we divide the Lagrangian into a free part and an interaction part. Second. we will need to have an overall momentum conserving ^function for the amplitude. expanding an amplitude in powers of h is also equivalent to an expansion in the number of loops. First. In fact. Therefore. Each such ^function will reduce the number of momentum integration by one. multiplies the entire Lagrangian. not all such momenta will be independent since at each vertex there are momentum conserving ^functions.
7) and (10. As we have seen earlier in Eq. In this case. as we have argued. 4>c(x). 5(j>c{x) T] 5<t>c{x) _ 62W[J] SJ(y) SJ(x)SJ(y) G(xy. However.25)) 5T[4>c] _ J(x). the loop expansion is unaffected by such a shift. 10. the classical dynamical equations are given by ~Hi ^ ={d +m2)(p{x)+ 3{x) J{x) ^ = • Furthermore.4 Effective Potential at One Loop Let us next calculate the effective potential for the (p4.57) and remember that we are interested only in one loop effects. (10. in this case. as we will see later. satisfies the equation If we now use the relations (see Eqs.theory at one loop. (10. then .17) that the classical field. Shifting the fields around such a value complicates perturbation in powers of the coupling constants.26). (10. (10. we note from Eq.Effective Action 203 symmetry breakdown in which case. the vacuum expectation value <j)c = (4>) becomes dependent on the coupling constants of the theory.<j>c).
(10. ^^(r[</>c]S[<j>c]) = %—Mx)G(0.45)) Si[&] = Jdix{Vl{cf)c{x)) we obtain 5S\ [cf)c 6<f>c(x) dVx (10.63) + ) . (10.(j>c) + o(h2).<t>c)+O{h2). v ™' ' 2 (10. Eq. we can consistently write SSxfa] 5<j)c{x) iX <f>c(x)G(0.cj)c).59) 5(j)c(x) (hSM) + 0(h2) = l—4>c(x)G(o. if we expand the effective action as T[(f>c] = S[<f>c] + hSifc] then.204 Field Theory: A Path Integral Approach keeping terms u p to linear power in h. (10. then Eq. (10.62) Thus.^ + ra»)W. we obtain from Eq. consistency requires t h a t .)^(„) + fM^ + 0(*»> or. if we restrict to <f>c(x) = 4>c = constant.60) Therefore.64) Let us note that although the Green's function G{x — y. to linear order in h.61) From the structure of the action (see Eq. (10.58) Now. (10. (10. <pc) c a n itself have a power series expansion in h. 2 (10. (10.56) H .58) gives v + 0(h2).61) takes the form ^ ) = ^ G ( 0 ^ ) .
(10. Furthermore.y. V. (9. this relation gives d4z •\x)54>c(z) (G(zy. 4>c) = 54(x . we are assuming that V\{4>c = 0) = 0.c J H t 2 _ J _ ^ .66) The Green's function can now be trivially determined and as we have seen before in Eq. Jd4z((d^ or.y.32) and have defined me2ff = m 2 + ^ .29) that <52r[^c] 5*w[j) dZ 5<t>c{x)84>c{z)8J{z)5J{y) / ^ § ^ V) ' to the lowest order of the Green's function that we are interested in.y).z))G(z .y) .«) Here. (10.r 2 or.) =  [' * .<j>c)) = d4(xy) or.15) as well as Eq.23) has the form Substituting this back into Eq. (9. <t>c) = 54(x . c 2 . noting from Eq. If that is not true.y. (10. we obtain dVxfa) iA iX 1_ f d44k dk 4 2 J (2vr) k . (10.65) or. (d^ + m2 + ^l)5\x . The one .y) + m2 + ^<t>2c)G(x . ( « V + mls)G{x Here we have used the form of the action from Eq. (10.Effective Action 205 we only use the lowest order expression for the Green's function in the above equation. <f>c) = 54(x .O0.64). then we have to add a constant term to the above expression.«.
k2 2 2 ^c + m2 .k i f d4k = 2JW id4kE 1WJ ln ( %<% + rr? 2k& \ *<% + m 2 2 2 m2 — k2 { m k • ^ Now. la70 Since the integrand does not depend on the angular variables. . + 4 j '< > (10.. (10. rotating to Euclidean space (see section 4. 3 „ . {U2 + m2 + k%\ = 2 i w * f e ' " ( • ro. the angular integration can be done trivially and has the value [d3tt so that we have = 2ir2.206 Field Theory: A Path Integral Approach loop correction to the potential can now be determined by going over to Euclidean space and doing the integral.1). 2) (2TT)4 1 f d3n . we obtain .71) _. (um) ._L^d44l„(^±). First.68). we obtain ™y&r« J J^rl i f d4k . {%(f>2 + rn2 + k2E\ m2 + /c. we note that if we interchange the orders of integration in Eq. d4k k2m2 ±<j>l f era p i J d4k r*° _d(±4>l) ^4>l I + m2 .
the integrals are divergent and. / ^ 1 l n _ . ji. to write the expression in a meaningful manner.(10. meff f i m eff *\ V l w.2 (m 2 ff . diverges in the limit A — oo which is the physical • . as it stands.Hx + m2)) "A4 327T2 2 lnA2±)+rr&A2 ^lnA2 + ^(lnme2ffi Yh2l)^2^2^^ 32TT 2 . Note that the one loop potential.._ 2)^ 2 m2 A2 etfln \iA m + ln 2 2 1 /i2 .75) + o ( m +7T^c) ln M* Here we have introduced an arbitrary mass scale. . therefore. we have to cut off the integral at some high momentum scale to obtain 1 TOc) = ^ 2 / fA2 AXX ( m ( X + meff) . — ^Ei (1073) we note that the effective potential at one loop takes the form 1 /00 Vl{4>c) = 32^J dXX ( l n ( x + m'ff) ~ HX + m 2 )) • (1074) Clearly. 2 ^^ 2 / n 22 A 2 2 V A m 2 ^ ( 2 ^ + 0 I n .Effective Action 207 Defining x *E.2 +^(ln^J^ln^ 327T2 2 4 A 77^A 2 2^c A 2 .
208
Field Theory: A Path Integral Approach
limit for the true value. This brings out one of the essential features of quantum field theory. Namely, pointlike interactions necessarily induce divergences simply because the Heisenberg uncertainty principle, in this case, allows for an infinite uncertainty in the momentum being exchanged. This necessitates a systematic procedure for eliminating divergences in such theories. This is known as the renormalization theory which we will not go into. Let us simply note here that up to one loop, then, we can write the effective potential of the </>4theory to be v & =V + V1. (10.76) 10.5 References
Goldstone, J., A. Salam and S. Weinberg, Phys. Rev. 127, 965 (1962). JonaLasinio, J., Nuovo Cimento 34, 1790 (1964). Nambu, Y., Phys. Lett. 26B, 626 (1966). ZinnJustin, J., "Quantum Field Theory and Critical Phenomena", Oxford Univ. Press.
Chapter 11
Invariances and Their Consequences
11.1
Symmetries of the Action
Let us continue with the ^theory and note that, in this case, we have
S[<j>] = J d4x £(</>, d^),
where the Lagrangian density has the form
(11.1)
£(<f>, d^) = \d^d^
 ^
2
 \<f •
(H.2)
We can write the action in Eq. (11.1) also in terms of the Lagrangian in the form S[<f>] = f dtL, (11.3) where L= [ d3 x £(</>, 0^). (11.4)
Given this theory, where the basic variables are the fields 4>{x), we can define the momentum conjugate to the field variables in a straightforward manner as U(x) = ^L = V ' d</>(x) fa). (11.5)
This is the analogue of the relation between the momentum and velocity in classical mechanics, namely, p = x (for m = 1). In quantum field theory, in operator language, this then is the starting point for
209
210
Field Theory: A Path Integral Approach
quantization. However, in the path integral formalism, we treat all variables classically. Therefore, let us analyze various concepts in the classical language. First, let us note that given the Lagrangian in Eq. (11.3), we can obtain the Hamiltonian through a Legendre transformation as H = f d3xU(x)<j)(x)  L . In the present case, we can write this out in detail as H= f d3x (u(x)cj){x)  ^<j)2(x) + ^V<j>(x) • V<(>(x) (11.6)
=jd3x Qn 2 (x) + \v<t> • v<p(x) + ^ V ( x ) + ±4>\x)) .
(11.7) Sometimes, this is also written as H = Jd3x^2(x) +lV4>{x)• V*(x) + ^ V ( s ) + ^ 4 ( z ) ) .
(11.8) Given a Lagrangian density, which depends only on 4>(x) and dfj,<p(x), the EulerLagrange equation is obtained to be (This is simply the generalization of Eq. (1.28) to the case of a field theory.)
^dd^ix)
dcf){x)
which gives the dynamics of the system. Given the dynamical equations, we can ask how unique is the Lagrangian density for the system. The answer, not surprisingly, turns out to be that the Lagrangian density is unique only up to total derivatives. Namely, both £(<f>,d^), and Ci^d^ + d^K^dx^), (11.10)
Invariances and Their Consequences
211
give the same EulerLagrange equation. We can, of course, check this directly. But a more intuitive way to understand this is to note that with the usual assumptions about the asymptotic fall off of the field variables, we have
SK = JdAxdllK»(<t>,dxct)) = 0.
(11.11)
In other words, a total divergence in the Lagrangian density does not contribute to the action. Consequently, the variation of SK cannot contribute to the dynamical equations. (We note here that even when the asymptotic fall off of the fields is not fast enough, this statement remains true.) We can, of course, check this for specific examples explicitly. Thus, choosing CK = d^K" = d M ( < W ) = d^d^ the EulerLagrange equation gives + <t>d^ct>, (11.12)
(11.13) With this analysis, therefore, it is clear that a given system of dynamical equations will remain invariant under a set of infinitesimal transformations of the field variables of the form <l> ^ <f> + S</>, (11.14)
if and only if the corresponding Lagrangian density changes, at the most, by a total divergence under the same transformations. Namely, if C^C + d^K", (11.15)
under a field transformation, then it defines an invariance of the dynamical equations. Note that in the special case when K^ = 0, then the Lagrangian density itself is invariant under the set of field transformations in Eq. (11.14) and, therefore, also defines a symmetry of the system. However, this is a very special case. In
212
Field Theory: A Path Integral Approach
general, if under
4> —• (j> + S(f>,
S[<i>]>S[<t> + 6<l>] = S[<l>],
(11.16)
then we say that the field transformations define an invariance or a symmetry of the system, namely, the dynamical equations. Continuous transformations, by definition, depend on a parameter of transformation continuously. This parameter can be a spacetime independent parameter or it can depend on the coordinates of the field variables. In the first case, the transformations would change the field variables by the same amount at every spacetime point. On the other hand, the change in the field variables, in the second case, will be different at different spacetime points depending on the value of the parameter. Accordingly, these transformations are called global and local transformations respectively. The basic symmetries in gauge theories are local symmetries. 11.2 Noether's Theorem
Noether's theorem, very simply, says that for every continuous global symmetry of a system, there exists a current density which is conserved. More specifically, it says that for a system described by a Lagrangian density C(<f>,d,j,<f)), if the infinitesimal global transformations (f>{x)  > <f>{x) + 5e<p(x), (11.17) where e is the constant parameter of transformation, define a symmetry of the system, in the sense that under these transformations C^C + d^K^dx^S^), then, dd^(j){x) defines a current density which is conserved. (11.19) (11.18)
Invariances and Their Consequences
213
To see that ji? is indeed conserved, let us note that
dC • 5€4>{x) + , / f . . ^ ( 5 ^ ^ ) )  0 ^ " . (11.20) 5(/>(ic) dd^x)
Here, we have used the EulerLagrange equation in Eq. (11.9) as well as the fact that d^8e4>{x) — <5e<9M<?!>(x). We note next that the first two terms in Eq. (11.20) simply give the change in the Lagrangian density under the transformations. Therefore, we can also write using Eq. (11.18) dtf? = 6e£  d^K" = 0. (11.21)
This shows that the current density given in Eq. (11.19) is indeed conserved. The current density defined in Eq. (11.19) depends on the parameter of transformation as well. A more fundamental quantity is the current without the parameter of the transformation and let us denote this symbolically as j? = ef. (11.22)
We have to remember that this is only a symbolic relation simply because the parameter e may, itself, have a tensorial structure in which case the current without the parameter will have a more complicated tensor structure as we will see shortly. As in classical electrodynamics, we know that given a conserved current density, we can define a charge which is a constant of motion as Q= fd3xj°(x,t). (11.23)
The fact that this charge is a constant, independent of time, can be
214
Field Theory: A Path Integral Approach
seen simply as follows.
=
Jd3xd0j°(x,t) + Vj(X,t)) . (11.24)
= Jd3x{d0j°(x,t)
Here, we have added a total divergence which vanishes under our assumptions on the asymptotic behavior of the field variables. Thus, we have ^ = Jd3xd„f = 0, (11.25) which follows from Eqs. (11.21) and (11.22), namely, the conservation of the current density. This shows that the charge is a constant of motion. Another way to understand this result is to note that this implies classically that the Poisson bracket of Q with H vanishes. Quantum mechanically, the commutator of the two operators must vanish. [Q,H] = 0. (11.26)
But this is precisely a symmetry condition in quantum mechanics. Namely, we know from our studies in quantum mechanics that a transformation is a symmetry if the generator of infinitesimal symmetry transformations commutes with the Hamiltonian. Conversely, any operator which commutes with the Hamiltonian is the generator of a symmetry transformation which leaves the system invariant. Thus, we recognize Q to be the generator of the infinitesimal symmetry transformations in the present case. This simply means that the infinitesimal change in any variable can be obtained from 5e</> = i[eQ,<f>]. (11.27)
(Classically, we should use appropriate Poisson bracket relations.) It is now clear that the vanishing of the commutator between Q and
the operator implementing finite symmetry transformations can be written in terms of the generator of infinitesimal transformations as U(a) = e~iaQ . Q. then. U(a)\0) = e~iaQ\0) = 0).27) that (0\5£<f>(x)\0) = i(O[eQ. the conserved charge. (11. does not annihilate the vacuum and that the vacuum expectation value of the change in some operator in the theory becomes nonzero. a true symmetry is supposed to leave the ground state or the vacuum invariant.0(x)]O) = 0 .33) . A field variable. 11. under such a transformation. As we will see later.31). (11. namely. we note from Eq.• x" + e" or. in such a case. (11.31) (11. Let us continue to use the 0 4 theory for this discussion.30) that Q0)=0. (11. (11.28) where a is the parameter of finite transformation. for a true symmetry. it follows from Eq.1 Example As an example of Noether's theorem. Therefore. the conserved charge annihilates the vacuum. furthermore. Let us define the infinitesimal translations x" .32) where we have used Eq. Equivalently. (11.Invariances and Their Consequences 215 H simply corresponds to the Hamiltonian being invariant under the symmetry transformationswhich we expect.2.30) In other words. let us study global spacetime translations as a symmetry of quantum field theories. (11. In quantum field theory. (11. is supposed to change as <t>{x) > U(a)<f){x)U1 (a) = e~iaQ<j){x)eiaQ . 5ex» = e^ . if there is a spontaneous breakdown of a symmetry.29) And.
37) As a result.36) (11. (11. (11. 5tC = C{x + e) . (11. we can. (11. is to note that the Lagrangian density is effectively a function of x. S€<j>{x) = <f>(x + e) . Thus.C(x) = e^d^Cix) = d^K* . C = C(x). we see from Eq.34) Given this.2) that for this theory dC dd^{x) d»(f>(x). (11.19).19). however. where eM is the constant parameter of transformation.35) On the other hand. we readily identify K* = e»C{x) = e»£(4>. of course. we see from Eqs.2) £ = \d„cf>(x)d^(x) .e"C = e" (d^(x)dl/(P(x) .36) and (11. Sed^ix) = d^S^ix)) = e^d^x). in a straightforward manner.^V(z) ±4>\x).5££) . (11. Therefore. (11. in this case.<f>(x) = e^d^x). In such a case. A much simpler way to evaluate this. (11. that the Noether current defined in Eq. (11. obtain the infinitesimal change in the Lagrangian density in Eq. 3^). follows to be = d»<t>{x) {evdv<t>{x)) .34).rfvC) . namely.216 Field Theory: A Path Integral Approach as the global transformations.38) = ev (df"cf>(x)du4>{x) .
of course. (11. let us note that the fundamental conserved quantity (in this case T^v) is not necessarily a vector. (11. let us write them out explicitly.rfvL. (11. has a vectorial character and has the form P ^ = fd3fd3xT0^ = xT°». (11. rplXV _ rpVfl.42) (there are.22)) j?(x) = euT^.42) that P°= fd3xT00 = J d3x ((4>(x))2 .23) and (11. _ (1141) This is known as the stress tensor of the theory. (11.42) To understand the meaning of the charges in Eq.40) that the conserved quantity. (11. the conserved current density and the current without the parameter of transformation has the form (see Eq. (11.40) that the conserved charge. in fact. Second. We note from Eqs. in this case. in this case. is a symmetric second rank tensor. First. we note from Eq.39) where we see from Eq. Let us also note from Eqs. Its tensorial character depends completely on the parameter of transformation. (11.38) that T"" = d»4>{x)dv4>{x) . namely.Invariances and Their Consequences 217 This is.40) There are several comments in order here.C) = Jd3x((<j>(x))2±<j>2(x) + ±Vct>Vt + ^<t>\x) + ^(x) . (11.40) and (11. four of them).
as we have mentioned before. 0*) = d^*d^ . in this case. Such a theory. can describe spin zero mesons which are charge neutral. (11. The real Lagrangian density describing quartic interactions can be generalized from Eq.44) One way to study such a theory is to expand the complex field in terms of two real fields as. 4>*(x)^<l>(x). P°) as the Hamiltonian of the system obtained in Eq. (1146) .218 Field Theory: A Path Integral Approach = j d 3 x Q ^ 2 ( x ) + \ V 0 • V<f> + —^(x) + ^4>\x) = H. <fi{x). say <j>(x) = j=(Mx) + i<h(x))(1145) However./ d3£</>(x)V<Ha:) • We recognize the first quantity (namely. as the basic variable. P{ = fd3xT0i= /"d3x^> (11. 11. Thus. (11. Namely.2) and written as C(d>. (11. Let us next consider a scalar field theory where the basic field variable is complex. P = .8) and from relativistic invariance we conclude that P must represent the total momentum of the system. we recover the familiar result that the spacetime translations are generated by the energymomentum operators of the theory. let us continue with the complex field.43) or.^(<t>*4>? .m 2 0 V .3 Complex Scalar Field So far we have discussed the (^4theory where the basic field variable is real.
(11. (11. constant (global) parameter of transformation or equivalently. We can treat 4> and <j>* as independent dynamical variables.49) where a is a real. an infinitesimal transformation of the form (e is infinitesimal) de(f>(x) = ie<j>(x). (11.51) (11.47) or. we note that under such a transformation </>*</> . the two dynamical equations in Eqs. (11.• e}a<\?eia<\> = <t>*<j>. We should have expected this since having a complex field doubles the number of degrees of freedom. Let us next note that if we make a phase transformation of the form (j)*{x)>eia<t>*{x). (11.Invariances and Their Consequences 219 with A > 0. then. the two EulerLagrange equations following from Eq.46) are given by B dC dC = or.48) correspond to two coupled scalar field equations. 8£<f>*(x)=ie<l>*(x).48) Thus. and dC dC (11. (d^ + m2W + ^(<£W = 0.50) . (djXd» + m2)<t>+ ^ « W = 0.47) and (11. Correspondingly.
54) = ied^d^cj) . (11. (11. in the sense that. (11. For such an invariance.50) 4>*4> remains unchanged. under the transformation in Eq.50).49) or (11.ied^d^cf) = 0. from the form of the infinitesimal transformations in Eq. we note that K» = 0.m2cf>*^ . therefore.) In this case. (11.54) results because the parameter of transformation is assumed to be independent of spacetime coordinates. under the infinitesimal transformations of Eq. we see that the constant phase transformations define a symmetry of the theory. (11. 6e£ = 0. we obtain = 8^5^)3^+ d^*{d^{5^)) (11. Similarly.^(</>V) 2 »• C. (11.55) Such a symmetry is called an internal symmetry since the transformations do not change the spacetime points. (11.50).57) Therefore. C = d^4>*d^(j) .49) d^d^cp + ^ ( e ^ * ) ^ . (11.ie</>*</> = 0 .53) and (11. (11. we note that 5e(<l>*<f>) = (6£<f>*)<j> + <j>*{8t<i>) = it(\>*4> . the conserved current constructed through the Noether . (It is important to recognize that the invariance in Eqs.52) Namely. Equivalently.56) (11.^ ) = d^d^cf).220 Field Theory: A Path Integral Approach Equivalently.53) Alternatively. we note that under the transformation of Eq.
where we have defined f = icf)* & $ = i {4>*d^ct> . Q would represent the electric charge operator. Q0) = 0 . therefore. (11.19)) = &t<l>(ie<t>*)+ &*<!>* (iet) = ie(0*d/>(d'y)0) = e(i<l>*&l<l>) = ef.Invariances and Their Consequences 221 procedure has the form (see Eq. (11. Therefore. in this case. if the phase transformations in Eq.60) In quantum field theory. for the present case. (11. In other words.58) The conserved current. (11. (11. we must have.61) . This theory. As we have mentioned earlier. can be written as Q = Jd3xj° = f d3xi(<f>*j)<j)*<f)) = i [d3x(<j)*4>j>*</>).59) (11. has a vectorial character very much like the electromagnetic current density.49) or (11. it can be identified with the electromagnetic current associated with this system. j M . in such a case. The conserved charge. can describe charged spin zero mesons.( # > * ) (f)) . then the charge operator in Eq.60) must annihilate the vacuum.50) define a true symmetry of the system. (11.
let us consider the generating functional for the complex scalar field. 6£S[<t>. J*] is not unless we simultaneously change J and J* also. if the redefinition corresponds to the infinitesimal phase transformations defined in Eq. J. First of all. J*} = eiW^J'J^ =N f V<f>V</>* eiS\<M\J.64) Since the generating functional does not depend on the field variables (that is. But more importantly. they lead to conserved quantities and conserved quantum numbers. (11.50). making a field redefinition in the integrand of the path integral should not change the generating functional. (11.<f. 0*] + S£(f = d4x (J*(f> + J(f>*) ¥ fd4x(J*5ed> + JSed>*) — —16 f d4x (J*<f> . In fact.r].222 Field Theory: A Path Integral Approach 11.63) with S[(p. (/)*] representing the dynamical action for the system. (11.50).J. we note that J* is the source for the field 4> whereas J corresponds to the source for <j>*. In particular. the complete action S[(f).J. the fields are all integrated out). between the transition amplitudes. therefore. Z[J.J*] = S[0>*] + J dAx(T<t> + J<t>*). <f>*] is invariant under the global phase transformations of Eq. J*] = 5eS[4>. Here. (11. 4>*. Note also that even though the action S[4>. as an example. </>*. Thus.4 Ward Identities Symmetries are quite important in the study of physical theories for various reasons. then . (1L62) Let us note here that we have now set h = 1 for simplicity and that we have defined S[</>. they give rise to relations between various Green's functions and. let us note that infmitesimally.J4>*).
**' J . the symmetry relations .65) becomes 5Z '5J(x) This must hold for any arbitrary value of the parameter e and. we can obtain relations between various connected Green's functions as a result of the symmetry in the problem. By taking higher functional derivatives of Eq. we conclude that or. it does not contribute. (11. 5Z J J ] }.65) where we have used Eq.64).66) 5Z[J. In general. J *].J*] . (11.J</>*)\ e tf[MV. (11. In this case.Invariances and Their Consequences 223 we will have SeZ[J. Eq. Jd4x 5W SW ( r i x ) ^ ^ . (11. one should also worry about the change coming from the Jacobian under a field redefinition. therefore. In the present case.68) {X '6J*(x) 'SJ(x) This is the master equation for defining symmetry relations. (11.68). J* SJ*(x) iN f V$D<\>* (f>(x) e i S t*.J{x)i^ ) = 0.* 6J{x) =iN J f V<f>V<t>* <j>\x) eiS^*'J^ . Using this then. (11. Let us recall that by definition. J*]=0 = N = N tvm* [l)(j)V4>*i5eSeiStt'r'J>J*] (e fd4x ( J > .
iSeW[J. it is easy to see that the complete action in Eq. 5eS[<j>. such relations are extremely useful and go under the name of Ward Identities of the theory (also known as SlavnovTaylor identities particularly in the case of gauge symmetries).J*} = 0. Se<fi* = ie<f>*. (11. of course.70) or. Therefore. It is interesting to note that we could also have obtained the Ward identities from a combined set of transformations of the form 5e4> = —ie(fi. or.J*}eiW^J^=0 SeW[J. J*] = N fv(pV(t>* or.68).224 Field Theory: A Path Integral Approach are quite simple (simply because the symmetry transformations in Eq. .J{x)^~) =0 or ' Id4x{r^smJ^&° (1L71) This is. from Z[J. 6eJ* = ieJ* . (11.63) is invariant. J*} = 0 (i6£S)eiS^*'J'J*} ) (11. 5£J = ieJ. Namely. iej^x (j*{x)^ . but in the case of more complicated symmetries such as gauge symmetries. the same relation as in Eq.50) are simple).J. (11.69) In such a case. J*\ = e w [ J ' J *l = N fvcl>T><t>* eiSlWAJ'] we obtain S€Z[J. (11.4>*.
72) From the transformation properties of the fields (j>(x) and 4>*{x) in Eq. we obtain (This corresponds to asking. (11. From the transformation properties of the classical fields in Eq. by how much would 4>c (x) change if we change 4>{x) according to Eq. # ] = Jd4x(J*6£<l>c(x) + J(x)5ecj>*c(x)) J{x)</>l(x)) = ie f d4x(J*(x)(f)c{x) = 0. (11. (11. # ] = W[J. (11.74) . (10.75) . we can immediately determine the transformation properties of the vacuum expectation values in Eq.73). Namely.) 5€<j>c{x) = (O<5e0(x)O)J'J* = ie(0<Kz)0) J ' J * = ie<k(x). tfe#(x) = <0 W : z )  0 ) J < J * = ie(0\cf>*(x)\0)J'J* = ie<fc(x).73) where we have assumed the invariance of the ground state under such a transformation. we can now work out the Ward Identities for the 1PI vertex functions. (11. (11. we will have a complex classical field defined by (see Eq.50). (11.J d 4 x ( J > c ( x ) + J ( x ) # ( x ) ) . from which it follows that W c (11.72). we note that we have r [ 0 c . In the present case.Invariances and Their Consequences 225 Let us note that in the case of a complex scalar field.7)) 5W 7 r* 0c*(z) = ^ = W*(z)O)J'J*.50).J*} .
following Eq.2) except for the sign in the mass term which is opposite.226 Field Theory: A Path Integral Approach Here in the last step. It is clear that this Lagrangian density is also invariant under the global phase transformations in Eq. However. (11. (11.68).50) since each of the terms is. (11. On the other hand. V^.73). (11. (11.79) . we obtain / d 4 l G^* ( I ) «sii* : ( I ) ) . namely.59). (11.m V > + ^ ( ^ ) 2 . Therefore.49) or (11.4>*) = . we have used the relation in Eq. by taking higher order functional derivatives.a (11 77)  This is the master equation from which we can derive relations between various IPI vertex functions.75). we can set this to zero and noting that the parameter e is arbitrary. we obtain . (11. there exists a conserved charge which is the same as given in Eq. £ = 9 / ^ +mV^(^) 2 .<£ I** («sb* (I) " mfilx)) •(1L76) Therefore. A>0. 11.5 Spontaneous Symmetry Breaking Let us next consider the complex scalar field theory defined by the following Lagrangian density. the phase transformations define a symmetry of this theory as well and according to Noether's theorem. if we look at the potential of this theory. (11. using Eq. as a consequence of the symmetry in the theory.78) This is the same Lagrangian density as in Eq.
85) . 6C6C = However.Invariances and Their Consequences 227 then. To better understand what is involved here.84) Note that since for constant field configurations the derivative terms vanish. let us rewrite the complex field in terms of two real scalar fields. the extrema of the potential occur at 8V 86* 8V m2 + 6* 'V 2 A = ' 9 A * = 0. = m' (11.80) There are two solutions of these extremum conditions which are easily obtained to be 6C == ^ = 0. this also defines the true minimum of energy or the true ground state of this theory.81) . the extremum at 6C = 6* = 0 is really a local maximum of the potential energy whereas the true minimum occurs at 2m2 (11. (11. M or. (11. let us write V2 (cr + ip) . it is quite easy to see that d2V 86*86 whereas 82V 86*86 4>*4>=^ 2m2 (11.82) \?* 4>*4>=^f —m (11. d6 [ m +2' = 0. Namely.83) Consequently. we note that for constant field configurations.
90) . let us plot the potential in Eq. to be at 2m Cc = (11. the minimum of the potential can be chosen to be at 4= or. For simplicity.228 Field Theory: A Path Integral Approach where we assume that a and p are real (Hermitian) scalar fields. in fact choose the minimum . (11. 9 9 4m2 + (%) = ^y(11. in this case. we see that one of the fields develops a vacuum expectation value. the minimum of the potential occurs at <%& = ^ or. namely.P) = ^(a2 + p'2) + ±(v2 + p2)2. let us choose pc = 0. therefore. (11. a2 + p2 = — . V(a. In terms of these variables. then. p c = (0p(x)0)=0. (11. Then. there is an infinite number of degenerate minima lying on a circle in the a — p plane.89) To understand further what is involved. (11.79) as a function of a and p for constant values of the fields. ac = (0\a(x)\0) = ^ = .88) In this case. A>0. ac = 4m 2 A 2m 7x' Pc = 0.87) Let us.86) It is clear that.
45).92) In other words. (11. the global phase transformations correspond to a rotation in the a .'2m <T • Thus. (7.Invariances and Their Consequences 229 _ . but the minima are infinitely degenerate. 5a + iSp = e(p — ia). is very much like the instanton potential in Eq. such a potential is also known as the Mexican hat potential.50) 8cf) = —iecf). Popularly. Let us also note that since 4>= j=(a + ip). (11. 5p = —ea. we conclude that under the global phase transformations. that /=(&r + iSp) = *e j= (o. the real scalar fields transform as 5a — ep. the potential. we can deduce from the transformation rule in Eq.+ ip) or. in the present case.p plane.91) Prom this. . (11.
(11. (11.e. (11. we must have Q0)^0. a] — ep. therefore.p] = ea.89) <0<7(:r)0>=<7 c = ^=. we . 5p=ie[Q.26) that the symmetry of the Hamiltonian implies that Assuming that the vacuum state has zero energy (i. we say that the symmetry of the Hamiltonian (or the theory) is spontaneously broken.6 ^ = 1m or.94) It is clear.94) is consistent. In such a case. we conclude using Eq. Since Q does not annihilate the vacuum of the present theory. (11. that in the present case. let Q\0) = \x)(1196) We know from Eq. (11. H\0) = 0).92) that 1m <0M0> = e<0M0> = . (11.93) Therefore. ie(0\[Q. since with our choice in Eq. (11.27) that the infinitesimal change in any operator can be expressed as a commutator with the charge associated with the transformation as da = —it [Q.p]\0) = e^=. (11.95) in order that the relation in Eq.230 Field Theory: A Path Integral Approach Let us also note from our earlier discussion in Eq.
98) We have already seen that the Hamiltonian commutes with Q expressing the fact that it is independent of time. (11. H\X)=0. we can write [P^Q}=0.96) and (11. it follows that the momentum operator also commutes with Q.97) In other words. In fact. The most intuitive way is to note that PM generates spacetime translations whereas Q generates a phase transformation in the internal Hilbert space. We can. therefore. think of this as another vacuum. This can be easily seen from Eq.Invariances and Their Consequences 231 then obtain using Eq. (11.60) to be hermitian.H]\0) = 0 or. (QHHQ)\0)=0 or.96) would appear to be degenerate with the vacuum in energy. (11. (Q is seen from Eq. therefore. (11. (11. HQ\0) = 0 or. their order should not matter which is equivalent to saying that the generators must commute.t)Q\0) = fd3x(0\eipxj°(0)eipxQ\0).99) There are many ways of obtaining this result besides the argument given above. (11. A consequence of their commutativity is that we have e~iPxQ = Qe~iPx . in general. the state x) defined in Eq.100) .23) as follows. The problem with this interpretation is that this state is not normalizable. Since Q does not depend on spatial coordinates.) (x\x) = (0\QQ\0) = {0\Jd3xf(x. (11.96) [Q. (11. Both these transformations are independent of each other and.
Let us note. it does not act unitarily on the Hilbert space.<t>(x)}. This is another manifestation of spontaneous symmetry breaking.232 Field Theory: A Path Integral Approach Using this in Eq. (11. let us note that even classically.101) where we have used the property of the ground state. however.98). we obtain <XX) = Jd3x(0\eiPxj°(0)QeiPx\0) = Jd3x(0\f(0)Q\0) = (0\j°(0)Q\0) fd3x^oo. commutators such as [Q. (11. if the potential has a non . PM0)=0. To analyze further the consequences of spontaneous symmetry breaking. it is straightforward to show that the charge Q does not exist when there is spontaneous breakdown of the symmetry. the state \x) is not normalizable and hence cannot be thought of as another vacuum. (11. then. This analysis also shows that the finite transformation operator U(a) = eiaQ. that even though Q may not exist. In fact. are well defined in such a theory and as a result expressions such as U(a)(t){x)Ul{a) = eiaQ4>(x)eiaQ . are also well defined. namely.103) does not act unitarily on the Hilbert space. (11.102) In other words. Another way of saying this is to note that while the operator U(a) defines unitary transformations for the field variables.
Thus.^ ( c 2 + p2)~(ff2 + ^)2. then a stable perturbation would require us to expand the theory about the stable minimum. let us expand .^ ( 0 * 0 ) 2 1_ 2 / 2m\ _ / 2m \ 1 _ m2 / / 2m\2 2\ A // 2m V A / 9 o 4m 4m 2 \ 16 V VA A ~z 2 3 rn 2 = I a M a a ^ + ^ P a ^ + c72f m n /m2 m2\ /2m m' TT 2m 3 2 + (w_^_mv^2 + p2)__^2+p2)2 . (11.104) Then.Invariances and Their Consequences 233 trivial minimum. (11.78) would become £ = d^*d»(t> + m? </>*<!> . p^p. 2m a » {a) + a = = + a. (11. .105) . the Lagrangian density of the theory in Eq.
In fact. The minimum of the potential occurs along a valley and the Goldstone mode simply reflects the motion along the valley of the potential. there necessarily arise massless fields (particles). The Goldstone particles were.234 Field Theory: A Path Integral Approach Thus. Namely. The closest that comes to being massless is the pimeson. the change in the Goldstone field under the symmetry transformation acquires a nonzero vacuum expectation value. namely./ = . This is a general feature of spontaneous symmetry breaking. certain correlation lengths will become infinite. In particle physics. we conclude that when Goldstone modes are present. 277? <0M0> = . (11. not received well by the particle physics community. we recognize that the two point function in such a theory will have an infinite correlation. Therefore. However. two particles at infinite separation will still feel the presence of each other. one does not know of elementary spin zero particles which are massless. in such theories.e . we see the interesting fact that while the field a remains massive with the right sign for the mass term. In the present case. of course. namely. whenever a continuous symmetry is spontaneously broken in a manifestly Lorentz invariant theory. therefore. A massless field or a particle. in the presence of gauge fields like the photon field. The most familiar massless field is the photon field and we know that as a consequence of the photon being massless. In terms of the potential. it is easier to understand the Goldstone mode intuitively. the Goldstone modes get absorbed into the longitudinal modes of the gauge bosons effectively making them massive.106) This is also a general feature of theories with spontaneously broken symmetries. has associated with it an infinite characteristic length (Compton length). . Namely. the field p indeed has become massless. the Coulomb force has an infinite range. we note that p corresponds to the Goldstone field and let us recall our earlier result. This is known as the Higgs mechanism and is widely used in the physical models of fundamental interactions. These are known as the Goldstone fields or Goldstone modes (particles).
Jk=0 (11. let us assume that we have a theory of nscalar fields described by the Lagrangian density £ = £(&. corresponds to having nontrivial <^ic's when the sources are turned off.108) that the classical fields would transform under the symmetry transformations as Se<l>ic(x) = Tij(e)(t)jc(x).109) Furthermore.n. even if for one of the values of i... (11. (11. To see a general proof of this theorem. then there must exist massless particles (Goldstone particles) in the theory. the classical fields are defined to be SW T 4>ic(x) = 7TTT = <0<fc(aO0)J* • (ll. (11. the Goldstone theorem states that if there is spontaneous breakdown of a continuous symmetry. Let us note from Eqs. A. e. (11. 3 ^ ) .112) . we can define the generating functional with appropriate sources as Z[Ji] = e' w [ J i l =N fvfa eiS^Ji]. ( \\ 6 W (Pic = 9K{X)\J=0 = SJi(x) ^0.2.73) and (11. In this case. of course. i = l. A.107) Furthermore. may itself have an index.108) where we assume summation over repeated indices and where the global parameter of transformation. (11.Invariances and Their Consequences 235 11.111) then. let us assume that the global transformations 6e<f>i = Tij(e)<l>j.107).6 Goldstone Theorem In a manifestly Lorentz invariant quantum theory with a positive metric for the Hilbert space. Namely.HO) oJi(x) The case of spontaneous symmetry breaking. define a symmetry of the Lagrangian density in Eq. we will have spontaneous breakdown of the symmetry. (11.
consistency of Eq. 1 (p A i = 0 ) ^ = 0. there will be spontaneous breaking of the symmetry) only if det(G. will have at least one nonzero value if the symmetry is spontaneously broken. (11. Tjk{e)4>kc = 0 (11. {G~'l{pti = Q))ijTik[e)<t>ka = Q. This proves the Goldstone theorem.y)). 11. . (10. Given the above relation.. there must exist massless particles in the theory./(Vltfi( MM <11114) =/d4"*^wT(^w' When we switch off the sources.7 References Goldstone. This system of equations will have a nontrivial solution (namely. this defines an extremum equation whose solutions.25) that the 1PI vertex functional satisfies the defining relation When the source is turned off.114) will lead to (in this case <piC(x) = </>jC = constant) . we can also obtain seMx)= t^y. 154 (1961). Nuovo Cimento. </>jc.116) In other words.4 / d y or . J. (11. Jd y 4 *2r c(x)54>jC{y) Tjk{e)(f>kc = 0 (G\x . 19..236 Field Theory: A Path Integral Approach We also know from Eq.115) or.
. Itzykson. and J. JonaLasinio. 253 (1957). L. L. Hill. Zuber.B. S. Ed. McGrawHill Publishing. Rev. "Quantum Field Theory".. 345 (1961). R. Nambu. E. Y.Invariances and Their Consequences 237 Guralnik. Mod.. 23.. et al. Phys. Cool and R. Rev. and G. JohnWiley Publishing. C. in "Advances in Particle Physics". G. E. 122. Marshak. Phys.
.
along with beautiful structures. we will start with Maxwell's theory where the difficulties can be simply handled and then we will go into a detailed discussion of nonAbelian gauge theories within the context of the path integral formalism.1) . Gauge theories are defined to be theories with a local symmetry and the symmetry is based on some relevant symmetry group. 239 (12. these difficulties can be handled with ease. have important consequences and in the case of a local symmetry. (Similar difficulties also manifest in the canonical quantization of such theories. For theories with simple local symmetries such as Maxwell's theory.1 Maxwell Theory Let us recall that the Lagrangian density for Maxwell's theory is given by C = ~Fia/F^. but problems become quite severe when the relevant local symmetries are more complex such as a nonAbelian symmetry. we will study how such theories are described in the path integral formalism. However. in this chapter. local invariances also bring difficulties. As we have seen in the last chapter symmetries. in general. The path integral description in this case needs to be analyzed carefully.) In this chapter.Chapter 12 Gauge Theories Gauge theories are very fundamental in our present understanding of physical forces and. the consequences are even more powerful. 12.
3) defines a symmetry (invariance) of the Maxwell theory.U(x). (12.3) the field strength tensor remains invariant /w OJJ. of course. under a local change of the vector potential by a gradient.dvA» = Fvli. = d^Av . For completeness. (12. (12.3) can be written in terms of local U(l) phase transformations as = U1(x)Ali(x)U{x) + iU1(x)df. restricts the structure of the theory (Lagrangian density) and leads to difficulties that can be seen as follows.5) where U(x) = e~ia(^ e U(l). Maxwell's theory is invariant under gauge transformations which are local transformations. The gauge invariance. we note here that the gauge transformation of Eq.AV — Oi/A^ d^ (A„ + dva{x)) . the gauge transformation of Eq.duA^ = F^ . (12. the physically observable electric and magnetic fields are not sensitive to a redefinition of the vector potential by a gauge transformations. A^x) ^ A^x) + d^x).2) As we know.4) Namely.240 Field Theory: A Path Integral Approach where the field strength tensor describing the electric and the magnetic fields is defined in terms of the vector potential A^ as Fp. (12. These are simple local gauge transformations belonging to the Abelian U(l) group and correspondingly. Let us note that the canonical momenta conjugate to the field variables A^ can be calculated from the Lagrangian density and . Explicitly.dv (A^ + d^a{x)) d^Av . Maxwell's theory is known as an Abelian gauge theory. Since Maxwell's equations can be described in terms of the field strength tensors. (12.
j or. we can choose a gauge (a particular form of the vector potential defined by the gauge choice) to work with. The canonical . a general feature of gauge theories. the true dynamics of the theory is. namely. would lead to V2A0 = . the local invariance leads to constraints.A = 0.6) n° = .8) On the other hand. contained in the transverse physical degrees of freedom and the longitudinal degrees of freedom can be expressed in terms of these. if sources (charges and currents) are present.F 0 0 = o. the canonical quantization of such theories is nontrivial.10) 1 Ao = . di (9*A0 . However.9) = 0.3). (12. the gauge V . then the equations of motion 3 ^ lead to (for v = 0) d^Ff* = diFi0 = 0 or. (12. A° = Ao = 0. If we choose the gauge V • A = 0 (Coulomb gauge). 0 (12.a0A*) = 0 or.7) This is. we note that since Maxwell's theory as well as physical quantities such as the field strength tensor are invariant under the gauge transformation in Eq. (12. (12. (12. therefore.Gauge Theories 241 take the forms IP = ~ which implies the constraint <9AM = F°". in fact.11) In either case.— 2 Jo (12. As a result.
d»du) (12.13) With a suitable normalization P^v can be thought of as a projection operator.12) p^pux = (n^n . . namely. Here we note that we can write the Lagrangian density for the Maxwell theory also as L = FtwFlw where P^ = r ^ D .Ddu) = 0 = dvP"u . = A^P^Av + total derivatives . (P^ = ^P^ is the normalized projection operator. the Feynman propagator cannot be defined. Let us emphasize here that the final result for the calculation of any amplitude in the canonical formalism remains manifestly Lorentz invariant.d^<9AD .d»du . This implies that if we were to apply the path integral formalism naively. As a result.d»dx\ = OP»x. we note that d^ v = dM (v^a . therefore.14) = (dvD . the inverse of P^ does not exist and the Green's function and. (12. We can ask what would happen if we were to treat the theory in the path integral formalism as opposed to the canonical formalism. this is the transverse projection operator. it projects the components of any vector transverse (perpendicular) to the gradient operator <9M.d^d") (sxa .dudx) = r / ^ D 2 .) In fact. neither the generating functional will exist nor can we carry out perturbation theory. (12.d»dxU + ^ d A D = • (VAD . Therefore. but we lose manifest Lorentz invariance in the process. However.242 Field Theory: A Path Integral Approach quantization can now be carried out. there is no manifest Lorentz invariance in the intermediate steps.
we could modify the theory to make it nonsingular. let us look at the equation of motion following from the action in the present case 8d. without any further input. However.Av dAu . Therefore. since we realize that the difficulties in quantization arise because of the singular nature of the Lagrangian density. Namely.) C = . But to understand the issue better.\ F ^ F ^ . in this process. for example.16) Here we have generalized Maxwell's theory to include a conserved current. the Lagrangian density (This formulation of the theory is due to Fermi and this gauge is known as the FeynmanFermi gauge.\ {d»A»f . (12. But clearly this would appear to be different from Maxwell's theory.15) where j ^ represents a conserved current (the sources in Maxwell's theory) V = 0. We can. we can solve for the constraints and quantize only the true dynamical degrees of freedom.A» . In such a case. of course. we give up manifest Lorentz invariance since we single out the transverse degrees of freedom. we see that the naive canonical quantization has unpleasant features in the case of Maxwell's theory since the fields are constrained and the momentum corresponding to AQ vanishes. This term breaks gauge invariance and consequently leads to a nonsingular theory. But more than that we have also added a term —  {d^A^) to Maxwell's Lagrangian density. Let us consider. take an alternative approach.j. In a physical gauge such as the Coulomb gauge. simply because the Green's function does not exist. at this point there is no justification for adding this new term to the Lagrangian density. (12. the Cauchy initial value problem cannot be uniquely solved. the system is singular and contains constraints among the field variables. As a consequence.Gauge Theories 243 We note here that whenever the determinant of the matrix of highest derivatives in the Lagrangian density vanishes.
+ &/{dA)j'. we have 0„ V " + • (d • A) = dvf or. Without the second term on the left hand side in Eq. and .j = 0. U{dA) = 0. V " + d {8 • A) = f .17). the presence of this additional term in the Lagrangian density would not change the physics of Maxwell's theory.244 Field Theory: A Path Integral Approach or. (12. (12.* = 0 . v =0 (12.17) or.19) would determine x = 0 at all times and we get back our familiar Maxwell's theory. at t = 0. (12.18) f An alternative way to see this is to note that if we take the divergence of the equations of motion in (12. UAu = f or.17). .dvA») + du{dA)= or. UdA = d. (12. x is in reality a free field and. classically we can think at t = 0. Thus. (12.21) ot then Eq. Thus we see that although the presence of the term —\x1 in the Lagrangian density where X = dA.20) seems to modify the theory. dllF'a. therefore.19) where we have used the antisymmetry of the field strength tensor as well as the conservation of j ^ . we realize that if we restrict classically X = 0. Furthermore. it takes the form ^ (&iAv . If we now write out the left hand side. (12. this is just Maxwell's equations in the presence of conserved sources.
it can be decomposed into positive and negative frequency parts uniquely in a relativistically invariant manner and this decomposition is preserved under time evolution. The state with zero norm. too. .24) where A^ (+) (x) is the positive frequency part of the Maxwell's field and contains only the destruction operator. Gupta and Bleuler weakened the supplementary condition on the physical states to have the form V^«0z:)phys)=O.24) selects out three kind of photon states as being physical. One can think of imposing such a condition on the vector space of states to select out the physical Hilbert space. this. (12. Therefore. (12.22) is hard to impose on the theory as an operator relation. Furthermore. V^phys)=0. two transverse photon states with a positive norm and a linear combination of the timelike and the longitudinal photon states with zero norm.19)) ndftA'l(x) = 0. but it also requires that those photons cannot be emitted either.15) with the supplementary condition dA = 0. (12. the field variables. (12. the GuptaBleuler condition (12. the resulting vector space of states has the problem of a negative metric. A^.25) dlj.22) In the quantum theory. (12. turns out to be too stringent a condition.) We remark here that since (see Eq. we note that the theory has four degrees of freedom resulting from the four components of A^ and since the components can be timelike or spacelike.23) However. This not only demands that certain kinds of photons are not present in the physical state. namely. are operators and Eq. (This is commonly known as the GuptaBleuler quantization. (12.Ati(x) is like a free scalar field. On the other hand.Gauge Theories 245 of Maxwell's theory as described by the modified Lagrangian density of Eq. however. (12.
Thus. there is no problem with a probabilistic interpretation). we see that the physical subspace of the theory selected by the supplementary condition contains states with positive semidefinite norm (negative norm states are eliminated by the supplementary condition or the physical state condition and. namely. decouples from the theory and the physical Hilbert space effectively consist of the physical transverse photon states. there is an easy solution to the problem associated with the local invariance of the theory. on the physical quantum states since (Vlc^lVHO.26) where \ip) represents a physical state. (12. therefore. We can think of the GuptaBleuler supplementary condition as imposing the Lorentz condition dnA»{x) = 0. therefore. In this simple gauge theory. we have the true physical subspace of the theory where the norm of states is positive definite. it can be described in the path integral formalism in the standard manner. Since the modified theory in Eq. if we further mod out the states by the zero norm states. (12.2 NonAbelian Gauge Theory As we have mentioned earlier. As we will see in the next section.246 Field Theory: A Path Integral Approach however. 12. (12.15) (which is equivalent to Maxwell's theory) is nonsingular. Vphys = ^ . Since the zero norm states are orthogonal to all the states including themselves. is orthogonal to every other state (including itself) and. nonAbelian gauge theories are based on nontrivial symmetry groups where the generators of the group .27) where VQ represents the set of states with zero norm. the difficulties become more severe in a nonAbelian gauge theory. consequently.
a particular representation that is very important as well as useful in the study of gauge theories is given by (Ta)bc = ifabc. (12.a = 1.28) and the Jacobi identity for the algebra is given by rpa rpb rpC + [Tc. Tb] + ifbcp [Tp. If we assume that Ta. then the generators satisfy the Lie algebra of the form (we take the generators to be Hermitian) rpa rpb = ifabcTc.Tb + rpb rpC T a = 0. Let us take brief digression into the structure of SU(n). (12.Gauge Theories 247 satisfy a noncommutative algebra. However.Ta]. n 2 — l(dimS17(n)) define the generators of the group.29) This imposes a restriction on the structure constants of the form ifabp [Tp.2. ••• .30) o r We can find the generators in various representations of the group much like in the case of angular momentum. rpa rpb Acq / rparpb •rpb rpa = = (ifacp) (Ta)cp{Tb)pg(Tb)cp(T«)pq (ifbpq) ~ (ifbcp) {ifapg) .32) Furthermore. (12. Ta] = 0 o r tabp tpcqrpq _j_ reap fpbqrpq __ fbcp ipaqrpq tabpspcq __ scapspbq __ rbcp tpaq _ Q _ Q (12. Tc] + ifcap \TP. we can easily check that this representation satisfies the Lie algebra.31) This is consistent with the hermiticity requirement for the generators (r°t) 6c = «Ta)cby = (»/«*)* = ifacb = ifabc = (Ta)bc. (12.
(no sum on c). (12.36) v J [Ka if a = b. so that the identification (^adj))6c = . To prove this let us note that we can always diagonalize the tensor Tr (TaTb^ such that (this is a symmetric real matrix) / \ f 0 if a ^ b. For any representation of a simple Lie group we can write Tr TaTb = C25ab. Furthermore. (1234) indeed forms a representation of the Lie algebra known as the adjoint representation. using the commutation relations we have habc= ^ fifabpTpTc\ = ifabp Tr = ifabpKp6pc (TpTc) = iKcfabc.35) We note that C2 is a normalization constant which determines the values of the structure constants. (12.38) . (12.37) = Tr (raTbTc) is completely antisymmetric in all its indices. Tr(raT6) = ^ (12.r 6 c (1233) .248 Field Theory: __ facp fbpq _ ecap fpbq A Path Integral i ebcp fapq ibcp fpaq Approach = fabPfPci = ifabp (ifpcq) (by Jacobi identity) = ifabp(Tp)cq. Let us next note that the quantity habc= Tr rpa rpb r . It depends on the representation but not on the indices a and b.Tr (rbTaTc) .
42) (12. (no sum on b).) In other words. (12.Tc]Tb) (TpTb) = ifacp Tr = ifacPKp5pb = iKbfacb = ~iKbfabc.39) However. Tr TbTa = KbSab . Similarly. (12.45) then T(R) is known as the index of the representation R. (12.46) .Gauge Theories 249 On the other hand. since habc is completely antisymmetric hacb = _habc _ (12.41). we see that we can write Tr (raTb) = C26ab . (12.41) By cyclicity of the trace.) We note that if we write TrTaTb = T{R)6ab. (12.40) Comparing the two results we see that Kb = Kc = K. (12. we note that /!«*= Tr ([Ta.44) where the constant C2 depends only on the representation. (It is chosen to be 2 for the fundamental representation to which the fermions belong in SU(n). (12. we have (TaTa)mn = C(R)8mn. (Alternatively. note that Tr TaTb = KaSab.43) which proves Eq. the two must equal and hence we have Ka = Kb.
250 Field Theory: A Path Integral Approach where C(R) is known as the Casimir of the representation R. since U and A^ are now matrices. The two are clearly related as T(R) dim G = C(R) dim R. 9{x)] = A^+D^ix).52) (12. 6(x)] . The gauge transformation in Eq.5) can now be generalized to A„{x) . where we have defined D„e(x) = dffiix) . To construct a gauge theory with SU(n) symmetry. in the present case. (12. these can be expanded in terms of the basic generators of the Lie algebra as Afl(x) = TaAl(x). (12. U{x) = e~iaW = ei<*a(x)T" } (1250) + iU\x)dllU(x).47) where dim G.49) belongs to the nonAbelian group SU(n). they do not commute in general and the actual gauge transformation in Eq. can be written as matrices.3). we note that under a gauge transformation (12. . (12. Furthermore.51) also known as the covariant derivative. (12.U~1(x)Afl(x)U(x) where. (12. In fact. (12.49) is much more complicated than (12.49) takes the form (here the parameter of transformation 9{x) is assumed to be infinitesimal) A„(x) > ^ ( x ) + < V ( x M [A^x). Furthermore.49). To construct the Lagrangian density for the gauge field in a gauge invariant manner. Eq. we note that the gauge potentials must belong to some representation of the symmetry group and.48) where summation over repeated indices is understood. therefore.i [A^x). for the simple case of an infinitesimal gauge transformation. dim R denote respectively the dimensionalities of the group G and the representation R. (12.
(dfJJ1) (OyU)) .i [ ( £ / " % £ / + i t / " 1 (d^U)). (UlAyU + iU\dvU))} = iU~l [Af. [UlA„U + iU1 (duU)} dv [U^AfJJ + iU'1 {dfJJ)] = i (dfJJ1) (dvU) . in the present case f^y = d^Ay — dyAfj. Ay] . namely. Let us also note that under a gauge transformation ~~* [Afli AV\ = —l [Af^Ay — AyAfj. .i {dyU1) {dfJJ) + (d^1) AVU + U~lAv (d^U) . F^^U^Ff.55) then under a gauge transformation.UlAf.53) Namely. (12. (12. (12.57) . Ay] U + i {{dyU1) {dfJJ) .\ ^ F ^ = \^yF^a .yU.2)) would transform as JfJ..Gauge Theories 251 the tensor representing the Abelian field strength (see (12. is not invariant under a gauge transformation. we see that. F^y will transform covariantly.( (dfJJ1) AyU + U~lAy {dfJJ) .U^Af. unlike QED.(dyU1) AfJJ ..) .Ay . if we define the field strength tensor as F^ = df.ig [Af. in the present case. (dyU)) .56) It is now easy to construct the Lagrangian density for the kinetic part of the gauge field as (quadratic in derivatives) £gauge = .dyAf.{dvU~l) AfJJ . (12.ll — Of! Ay — UyAf! f df. (12. it is clear that. (dvU) +U1(dflA1/d„Afl)U.54) Thus.
(12. to be invariant under the gauge transformation (12. T a 's in the fundamental representation). (we have set the coupling constant to unity for simplicity) F^ = d^Al . the field strength tensor takes the form.56) and leads to = i (ifcab) ecFbu . (12.49). transforms according to the adjoint representation of the group. A^.59) = (^ai(rfadj))ac^c). The infinitesimal transformation of the field strength tensor can be obtained from Eq. the infinitesimal form of the transformation for the gauge fields in Eq.60) which shows that the gauge field. so that we can write 6A° = (D&®oY .51) take the form = (d^ea + fabcAbftec'\ = (d^a ~ fbacAbfl0c') = {dll6ai(ifhac)Abfic) (12.dvA% + fabcAlAcv = F„l. (12. In components. This Lagrangian density is easily seen. (12. from the cyclicity of trace.58) In components.252 Field Theory: A Path Integral Approach where a particular normalization for the trace is assumed (namely.
Physically we understand this in the following way. the field strength tensor F^ as well as the gauge field A^ transform according to the adjoint representation of the group.65) dA°(x) . here the gauge field has self interaction. (12. Wa{x) = = F°i"a(x). Since gauge fields couple to any particle carrying charge of the symmetry group. (12. (It does not matter what representationt the matter fields belong to. (12. (DvF'w)a dA« = 0. The EulerLagrange equations following from Eq. (12. the gauge field carries the charge of the nonAbelian symmetry group (they have a nontrivial symmetry index) in contrast to the photon field which is charge neutral.) We note here that unlike the photon field.63) For /i = 0.62) where the covariant derivative is defined to be in the adjoint representation of the group and F% = dpAl ~ dvAl + rhcA\Acv . the gauge field must transform in the adjoint representation. Let us next briefly examine the difficulties that arise in trying to canonically quantize the theory along the lines of Maxwell theory. it can be explicitly checked the coefficient matrix of highest derivatives is the transverse projection operator just like in the Maxwell theory. In the present case. In fact. we see that the EulerLagrange equations lead to a constraint of the form DiF0ia = 0.64) which is a reflection of the gauge invariance of the theory.57) take the form v ddvA« or. in the case of nonAbelian symmetry they must possess self interactions. (12. Let us define the momenta canonically conjugate to the field variables A".Gauge Theories 253 Thus.
. (12. let us look at the theory C = F^F^ a . We can try to quantize a nonAbelian gauge theory covariantly. (12. much like in the Maxwell theory. there are serious differences from Maxwell's theory. the theory can be defined in terms of only physical transverse gauge fields in this case. For example.67) where we can think of Ef(x) as the nonAbelian electric field strength. makes the theory nonsingular. by modifying the theory. we have U0a(x) = F0Oa = 0. (12.254 Field Theory: A Path Integral Approach Noting that the field strength F£v is antisymmetric in the indices fj. However.70) and hence classically we can impose the condition dA = 0. We note that the momentum conjugate to AQ does not exist much like in the Maxwell theory. A%(x) = 0. very much along the lines of the Abelian theory. although it is much more complicated. consequently. for example.69) The additional term in this theory clearly breaks gauge invariance and. in the present case. u. This implies that AQ is like a cnumber quantity which commutes with every other operator in the theory. Barring technical issues such as the Gribov ambiguity (related to the existence of large gauge transformations). Therefore.66) and Uia(x) = F0ia = Ef(x). (12. namely. Thus we can choose a physical gauge condition to suitably set it equal to zero. Thus.71) . namely. (12. let us note that in the Abelian theory. D(8A) = 0.\ (O^A^)2 . canonical quantization leads to a lack of manifest Lorentz invariance. (12.68) The analysis is parallel to what we have discussed in the case of Maxwell theory.
(12. we note that (dtlA. in contrast to the Abelian theory. even classically the equations of motion are given by V " a + fabcA^F^ c + dv (d^a) = 0. we need to analyze the question of modifying the theory in a more systematic and detailed manner. consequently.Gauge Theories 255 which then translates to the GuptaBleuler condition on the physical states dA ( + ) phys) = 0 . (12. let us go back to the Maxwell theory described the Lagrangian density £( J ) = ^F^F^ + J^A^ .73) Contracting with du. Furthermore. (12. since it is not invariant under time evolution. since (dflAfJ'a) is not a free field.A^F^ C W 0.) Correspondingly the analog of the GuptaBleuler condition for nonAbelian gauge theories does not appear to exist.74) Thus. nor can we think of a supplementary condition such as ^ A ^ ( + )  p h y s ) = 0. (12.76) . 12. it cannot be uniquely decomposed into a positive and a negative frequency part (in a time invariant manner). We would see next how we can derive intuition on this important question from the path integral quantization. we obtain • ( d M ^ a ) = fahcdu (. Therefore.M) does not behave like a free field and. (Namely. (12.3 Path Integral for Gauge Theories To understand better how gauge theories can be handled in the path integral formalism. however.75) in a physically meaningful manner. the physical subspace would keep changing with time which is not desirable.72) In the case of the nonAbelian theory. the additional term has truly modified the theory.
(12. (12. (The operator possesses zero modes and.^)]. y)Av{y).80) .O H ) ~ * eH J M >°^ J V ) .0'"'^)+(^.• 4 a ) = UA^U1 + ill"1 (d„U). Clearly therefore.78) (Jfi. the determinant of O^ vanishes.77) Here we have introduced a compact notation to describe integrations as described above. The functional integral is a Gaussian integral which we have worked out in earlier chapters and leads to Z [jy = eiM/lJ"l = N (det ( .79) where U(a) = e~iaW .y). The Lagrangian density for Maxwell's theory is invariant under the gauge transformation A„ .d^d") S\x . the inverse of the matrix cannot be defined either. as we have seen before. The longitudinal vectors k^ (or d^F) are its eigenvectors with zero eigenvalue. The generating functional in the path integral formalism is given by Z [JM] = e * ^ l = N J V A ^ ^ ei[(^.) Going over to the Euclidean space does not help either. The source of the difficulty is not hard to see. the operator O^ is a projection operator for transverse photons. as we can see even in the limit of vanishing sources. the generating functional does not exist. J^(x)Afl(x).AfM) = Jd'x (A^AV) = J d"xd% A^x)0^{x However. In fact. (12. This implies that the generating functional does not exist. consequently. (12. = N fvAp where iV is a normalization constant and O^ix y) = (DrT .256 Field Theory: A Path Integral Approach where J^ represents a conserved current (source).
82) defines the hypersurface which intersects the orbits once. We recognize that we should not integrate over all gauge field configurations because they are not really distinct. all the 4 that are obtained by making a gauge transformations with all possible a(x) are said to lie on an "orbit" in the group space. (12.Gauge Theories 257 s For a fixed AM. namely. F(4"))=0. has a unique solution for a(x).81) (In the nonAbelian case.84) is known as the gauge condition. fixing and the condition (12. The way this is carried out is by choosing a hypersurface which intersects each orbit only once. gauge independent and do not depend on the choice of the hypersur . (12. of course. then even if A^ does not satisfy the condition. (12.83) This procedure is known as gauge F(A^) = 0.. Physical quantities are. i. Therefore. The action S. is proportional to the "volume" of the orbits denoted by /"jjda(x). this should be replaced by the group invariant Haar measure J\x dU(x). even in the absence of any sources. is constant (invariant) on such orbits. The method for extracting this factor out of the path integral is due to Faddeev and Popov and relies on the method of gauge fixing.) This is an infinite factor (which is the reason why the Gaussian functional integral does not exist) and must be extracted out before doing any calculations. on the other hand. Rather we should integrate over each orbit only once. we can find a gauge transformed Ap which does. the generating functional.e. if F(A„) = 0.
Therefore. (12. A3 = 0 is the axial gauge. AQ = 0 is the temporal gauge. in general. (Only physical quantities need to be gauge independent. let us do the following trick due to Faddeev and Popov.88) .) Note that the quantity App [A^] is gauge invariant which can be seen from the fact that AFP [A. (12. F (Ap.258 Field Theory: A Path Integral Approach face (gauge).) To extract out the infinite gauge volume factor.85) and so on. We can already see the need for gauge fixing from the fact that because the action is gauge invariant so is the generating functional (if sources are transformed appropriately in the nonAbelian case). for example. On the other hand. we know from ordinary perturbation theory that the Green's functions are. it would lead only to gauge invariant Green's functions. [Ap] .86) (The integration measure should be dU(x) which is essential in the case of nonAbelian theories. Then > App 4 a. (12.] fl[da(x) S (F ( 4 a ) ( x ) ) ) = 1 . V • A = 0 is the Coulomb gauge.] = fl[da(x) S (F (4 a ) (s))) • (12.87) Let us make a gauge transformation A^ — A^ . Let us define A F P [A.) — d^A11 = 0 is the Lorentz/Landau gauge. Thus one has to fix a gauge without which even the Cauchy initial value problem cannot be solved. Thus. gauge dependent although the ^matrix (the scattering matrix) elements are gauge independent.) ] = [l[da{x) 6 (F (A£+*\XJ)) ^ X = f Y[da(x) S {F (A^))) •* x = A ^ .
we still have to determine what App [A^] .)) eiSiJ)^ (12.JVAJAFP [A. That is f da(x) = f d (a(x) + a'(x)) . This.]jl[da(x)s(F ( A ^ ) ) = N [j II d a ^) j VA ^p [AJ 5 (F (A.)) eisU)^ (12.91) Furthermore.93) = N J VA^App [A^] 5 (F (A„)) elS{J) ^ .90) Remembering that App [A^] is gauge invariant we can now insert this identity factor into the generating functional to write je«s(J)^l. namely. let us make an inverse gauge transformation A^A^\ under which the generating functional takes the form Z[J] = N J VA^App [A^] j H da(x) 6 (F (A. we should have the Haar measure which is gauge invariant.92) Z[J] = N. gives the correct functional form for the generating functional. where the gauge volume has been factored and absorbed into the normalization constant N of the path integral. therefore. I d{UU') = f dU.Gauge Theories 259 This follows from the fact that the measure in the group space is invariant under a gauge transformation. (12. (12.89) In the nonAbelian case. (12. However.
we determine SF A F P [A„] = det  (12. We see that the FaddeevPopov determinant can be calculated simply by restricting to infinitesimal gauge transformations.95) (4 a) ) ^ ^ / a=0 .96) = 0. (12. (12. be thought of as the Jacobian that goes with a particular gauge choice.) We can further generalize our derivation by noting that a general equation of the hypersurface has the form (Physical results are not sensitive to the choice of the hypersurface.97) The FaddeevPopov determinant can. On the other hand. To do this let us note that •* x =det (ini) • (1294) We note that since App [A^] is gauge invariant we can make an inverse gauge transformation to make F [A^] = 0 in the above derivation. therefore. (Here we are completely ignoring the problem of Gribov ambiguity associated with large gauge transformations.260 Field Theory: A Path Integral Approach is.99) .) F(A(t(x)) = f(x). (12.f{x)) = 1. (12.98) Here f{x) is independent of A^. Thus. for gauge fields which satisfy the condition F(Ali) we have a(x) = 0. Then we can insert the identity A F P [AJ Jl[da(x) S (F (A^(x)) .
] AsiJ)kJ^A^y = NJvA^AFP[A^}ei(s(J)+s^).Gauge Theories 261 into the functional integral. We note that physical quantities are independent of f(x). where SGF = fd4x CGF = ^J^xiFiA^x)))2 and £ is known as the gauge fixing parameter. note that since SF A F p [Ap] = det  we can write this as . (12.f(x)) 2 xefii^(m) eis(^ =NJVAIXAFP [A.101) (4* ±—^ j a=0 . (12. (12. The FaddeevPopov determinant is unchanged by this modification because f(x) does not depend on A^x).100) Following 't Hooft. Thus the generating functional in this case is given by Z[J} = N J 2M„AFP [AJ < (F (AM(x)) .f(x)) 5 lS(J) e ^ . Thus. Hence we can multiply the generating functional by a weight factor and integrate over all f(x).102) fSF(A^(x))\ \ / a=0 [veDceW£)°=oc) . we can now do the following. the generating functional becomes Z[J] = N J VA^AFP [AJ J VfS (F (A^x)) . We furthermore.
). i.104) Thus although these behave as scalar objects under Lorentz transformations.262 Field Theory: A Path Integral Approach SF ifd4xd*yc(x)l (»gw) Lfa) M C W = I VcDc e V / «=o where Sghost = ' J dxdy c{x) I ^ Here we have introduced two independent fictitious fields c(x) and c(x) to write the determinant in the form of an action.. c. but opposite statistics.. [c(x). We note here that this is possible only if the ghost fields c(x) and c(x) anticommute (Ghosts have the same Lorentz structure as the parameters of transformation. (12. These fields are known as FaddeevPopov ghosts and as is obvious from their anticommutation relations. c] = S^ [A.106) .105) = JtfxdglA^cc]. [c(x).e. (12. Thus the generating functional now takes the form Z [J] = e w[J] where S$ [A„. they obey anticommutation rules. (12.c(y)}+ = 0.c(y)] + = 0.] + SGF + S ^ =N f VA^VcDc j & ^ d . graphs involving these fictitious particles in closed loops must have an additional (—1) factor just like the fermions.c(y)]+ = 0. [c(x).
(12.112) . (12. we note that F ( 4 a ) ) = d^Alila\x) so that SF = dM (A" + d"a(x)) . (12.Gauge Theories 263 Let us now look at a simple gauge fixing condition. (12.109) (4a)(*)) 6a(y) a=0 U5\xy).110) In this case.107) Clearly the gauge fixing Lagrangian density has the form CGF = {F{A»{x))f (12. (12.108) = ^(M"(*)) 2  It is clear that this provides the longitudinal components of the quadratic terms in fields and hence breaks gauge invariance. To obtain the corresponding ghost Lagrangian density for this gauge choice. Thus our effective Lagrangian density for this choice of gauge condition becomes C*s = \F»vF^ .j d 4 zd 4 y c{x) (nS4(x ~ y)) c(y) = I d4x d^c(x)d»c(x) = f d 4 x £ g h o s t . the covariant condition F(Ali)=dllA"(x) = f(x). for example. the ghost action is obtained to be f A A %ost = Jdxdy c(x) 6F(A?(x)) 5a»y) a=0 c(y) = .1 (d^)2 + dlxc{x)d^c{x) + J^A^ .111) where we have neglected total divergence terms.
As a result. we cannot neglect them even if they are noninteracting if we are doing calculations at finite temperature.264 Field Theory: A Path Integral Approach We note here that the ghost fields are noninteracting in the case of the Abelian theory and. the ghost fields automatically couple to the geometry also. Hence omitting the ghost Lagrangian density in such a case would lead to nonsensical results. In nonAbelian gauge theories. (The ghost degrees of freedom subtract because they have the unphysical statistics. This naive counting works pretty well as we will see later. adding a gauge fixing Lagrangian density changes the theory. however. In particular. Such a term is called a gauge fixing term and any term which makes the coefficient matrix of the quadratic terms nonsingular and maintains various global symmetries of the theory is allowed for this purpose. we may omit them and then for £ = 1 we recognize that our effective Lagrangian density is nothing other than Maxwell's theory in the FeynmanFermi gauge. if we take Cmv as describing the dynamics of the theory. we cannot define propagators and the entire philosophy of doing perturbative calculations with Feynman diagrams breaks down. . Hence one can think of the effective Lagrangian density as having two helicity states (4 — 2 x 1 = 2) effectively. On the other hand. has only one helicity component. then. Let us recall from our study in the Abelian theory that gauge invariance puts a very strong constraint on the structure of the Lagrangian density for the gauge field. On the other hand each of the ghost fields.) Let us now apply these ideas of path integral quantization to the case of nonAbelian gauge theories. It is also true that when Maxwell's theory is coupled to gravitational fields. therefore.Smatrix. being a scalar. therefore. the ghost fields are interacting and have to be present. the An field has four helicity components. noninvertible. we add to the gauge invariant Lagrangian density a term which breaks gauge invariance and thereby allows us to define the propagator for the gauge field. Furthermore. One way of looking at the ghost fields is as if they are there to subtract out the unphysical field degrees of freedom. since these fields are really necessary for the unitarity of the . For example. In order to circumvent this difficulty. the coefficient matrix of the quadratic terms in the Lagrangian density is singular and.
(12. we can write the ghost action corresponding to this gauge choice as c°(y). In the case of the nonAbelian gauge theory.^ ( v n 2 . (12. We note that for the covariant gauge choice that we are assuming.dyA^ix) + fhcA\Acv The standard covariant gauge fixing. Following the prescription of Faddeev and Popov. as we have seen. (12. consists of adding to the invariant Lagrangian density a term of the form £GF = . (12.115) . (12. (12.117) Here £ represents an arbitrary constant parameter known as the gauge fixing parameter. (12113) is invariant under the infinitesimal gauge transformation A^x) + A^a(x) = A«(x) + D„6a[x).114) where 9a{x) is the infinitesiihal parameter of transformation and D»aa{x) = d^ix) F. .118) Here we have left the dimensionality of spacetime arbitrary since our discussions apply in any dimension. we can write Fa[A^\x)] = d"A^a = d» (Aa{x) + D^aa(x)) . and to compensate for that we have to add a Lagrangian density for the ghost fields following the prescription of Faddeev and Popov.119) Q [AT f A A ~a< JFa[A^(x)} Sghost = dx £ g h o s t = .Gauge Theories 265 in general.^ V ^ a .H6) which would correspond to a gauge fixing condition of the form Fa[A] = d^a(x) = fa{x)./ dxdy 5 ° ( x ) — L b v b . the Lagrangian density for the gauge fields Anv = . in a nonAbelian gauge theory.v{x) = d^Alix) + fahcAl{x)ec{x).
without going into too much technical details. for this choice of gauge fixing. the ghost Lagrangian density. 12.120) where we have dropped a total derivative term in the last line of the above equation and have identified D°bx = 6abd^x + rbAl(x).122) which for the covariant gauge that we are considering is given by £TOT = \F%F^a . (12. the total Lagrangian density for a nonAbelian gauge theory has the form £TOT = A n v + £ G F + £ghost > (12.123) The path integral can now be defined with this £ T O T and will be well behaved. (12./ dy e°(x) (dZD^Six . In a deeper sense. the gauge fixing and the ghost Lagrangian densities modify the original theory in a compensating manner which allows us to define the Feynman rules of the theory and carry out any perturbative calculation.121) With all these modifications.1 {d^An2 + d»ca {Dpcf . follows to be £ghost = . merely correspond to a multiplicative factor of unity which does not change the physical content of the theory.y)) c(y)b = d»ca{x) {Dllc{x))a . (12.266 Field Theory: A Path Integral Approach Consequently. . in the path integral formulation.4 B R S T Invariance As we have mentioned earlier. we can see that these additional terms in the Lagrangian density have no physical content in the following way. the gauge fixing and the ghost Lagrangian densities. However.
(12. Similarly.A^) where LU is the constant anticommuting parameter of the global transformations. . with gauge fixing and ghost terms. therefore. (12. It is easy to check that the total Lagrangian density is invariant under the global transformations 8Aatl = 5ca = u(Dlic)a. remembers the gauge invariance of the original theory. in some sense. 2 fabccbcc.j (df. The invariance of the Lagrangian density can be seen by first noting that 6(Dlic?) = Dli6<? + fabc6Ablld: = \Dp ]fabccbcc J = fabcScbcc — —— ( fabc (fabccbcc) = /<** + ufabc (p^cb) ftofftfj cc = 0.Gauge Theories 267 The total Lagrangian density has been gauge fixed and. develops a global fermionic symmetry which. we obtain S (d^a) = u d^D^ca = 0.124) 5ca = . (12. fbPQ __ tabp fbqc __ cabq ibcp\ cPc1cc = 0.125) Here we have used the Jacobi identity for the symmetry algebra. This shows that d26i<f>a = 0. the total Lagrangian density.ca independent of the parameters of the transformations where 612 correspond to transformations with parameters .126) when the ghost equations of motion are used.127) for 4>a = A'jx. However.ca. (12. does not have the gauge invariance of the original theory.
In this derivation.129) = &i(^(dvAva)DIMA. namely. is slightly unpleasant in the sense that the nilpotency of the antighost field transformation holds only onshell. however. This is known as the BRST transformation for a gauge theory and arises in the presence the gauge fixing and the ghost Lagrangian densities. This is a reflection of the fact that the theory is lacking some auxiliary fields and once the correct auxiliary fields are incorporated. (12. We note that the nilpotency of the transformations holds offshell only for the fields A^. therefore. while for ca it is true only onshell (when the ghost equations of motion are used). The invariance of the Lagrangian density can now be easily checked. the symmetry algebra will close offshell.ca. we note that the transformation for A" is really a gauge transformation with the parameter aa = u>ca and.^ {dvAva) {VSAfi = ~ [dvAva) VD^ + (d»5ca) Z V a .128) Consequently. We can write the gauge fixing Lagrangian density by introducing an auxiliary field which will also be quite useful for our later discussions. The present formulation of the BRST symmetry. we need to worry only about the gauge fixing and the ghost Lagrangian densities which lead to S (CGF + £ g h o s t ) = .268 UJI}2 Field Theory: A Path Integral Approach respectively. we have used the fact that 5 (DMca) = 0 which we have seen earlier. First. This shows that the total Lagrangian density is invariant under the global transformations with an anticommuting constant parameter. <5Anv = 0. 5" {dvAva) Z V a (12. Let us rewrite £ G F = d»FaAl + ^FaFa . the invariant Lagrangian density is trivially invariant under these transformations. (12.130) . so that the action is invariant.
It is clear from the form of £ Q F that the equation of motion for the auxiliary field takes the form £Fa = &>A°. 5ca = oo Fa. allows us to take such gauge choices as the Landau gauge which corresponds to simply taking the limit £ = 0. 2 fabccbcc.ca.134) for <pa = A^. The total Lagrangian density can now be written as £ T O T = Anv + £GF + £ghost = ^F*vF»va + d^FaA° + ^FaFa + d»caD^ca .135) . does not transform at all which leads to SCTOT = 5 (CGF + £ghost) = d»Fa5A« + d»5ca {D^cf = ujd^Fa (D^cf = 0. the BRST transformations take the form 6Al = 5ca = u(Dltc)a. Fa represents the missing auxiliary field that we had alluded to earlier. we recover the original gauge fixing Lagrangian density (if we ignore a total divergence term). We note that £.Gauge Theories 269 where Fa is an auxiliary field. Fa.Lud^Fa (£>Mc)a (12. (12. nv is invariant under the BRST transformation as we had argued earlier and the auxiliary field.133) and it is straightforward to check that these transformations are nilpotent offshell.132) In this case. a (12. . ca. 5Fa = 0. (12. (12. <52<5i0a = O. Fa. namely.131) and when we eliminate F using this equation. Among other things £ Q F as written above. Therefore.
These are known as the antiBRST transformations. However. bosonic infinitesimal parameter and the generator of the symmetry merely counts the ghost number of the fields. 5Fa = iJfabcFbcp. There is also a second set of transformations involving the antighost fields of the form Sca = Lu(Fa6ca = .2 fabccb~A fabc^cc. (The fact that these . here we see that the total Lagrangian density is invariant under the BRST transformations (as opposed to changing by a total divergence). known as BRST transformations. These transformations. 8? = 9&. replace the original gauge invariance of the theory. we will ignore this symmetry for the rest of our discussions. (12. which define a residual global symmetry of the full theory. (12.270 Field Theory: A Path Integral Approach Unlike the formulation without auxiliary fields. . It is also worth noting here that these symmetries arise naturally in a superspace formulation of gauge theories. Here 6 represents a constant. since these do not lead to any new information beyond what the BRST invariance provides. and play a fundamental role in the study of nonAbelian gauge theories. We note here that the BRST and the antiBRST transformations are not quite symmetric in the ghost and the antighost fields which is a reflection of the asymmetric nature of these fields in the ghost Lagrangian density. In addition to these two anticommuting symmetries. in some sense.136) which leave the total Lagrangian density invariant.137) with all other fields remaining inert. the total Lagrangian density is also invariant under the infinitesimal bosonic global symmetry transformations 5ca = 9ca.
for example.Gauge Theories 271 transformations are like scale transformations and not like phase transformations has to do with the particular hermiticity properties that the ghost and the antighost fields must satisfy for consistent quantization of the theory. the generators of the BRST symmetry. needs to be properly selected for a discussion of physical questions. The corresponding operator in a nonAbelian theory. are conserved and hence can be used to define a physical Hilbert space which would remain invariant under the time evolution of the system. Furthermore. we can identify the physical space of the gauge theory as satisfying QBRSTJphys) = 0 . as is clear by now. therefore. contains many more states than the physical states alone. Qc. (12. In QED. The physical Hilbert space. On the other hand. QBRST. in reality it is an infinite number of conditions since it has to hold for every value of the coordinates. the physical space must be selected in such a way that it remains invariant under the time evolution of the system. the GuptaBleuler condition works because d^A^ satisfies the free KleinGordon equation in the covariant gauge and hence the physical space so selected remains invariant under time evolution.139) Note that even in the case of QED. does not satisfy a free equation as we have seen and hence is not suitable for identifying the physical subspace.) The vector space of the full theory.138) where the superscript. Qcphys)=0. on the other hand. and the ghost scaling symmetry. (QBRST a n d Qc are the charges constructed from the Noether current for the respective transformations whose explicit forms can be obtained from the Noether procedure. we have seen that the states in the physical space are selected as satisfying the GuptaBleuler condition ^4+)(x)phys)=0. We recognize that even though this looks like one condition. In the covariant gauge in QED. stands for the positive frequency part of the field. (+). (12.) Thus. these would appear to correspond to only two conditions and not an infinite number of conditions as we .
) = F^'SAl + b!?D»ca + (0"c°) Sca + FaD^ca . It is. (12.143) We recognize that the condition Q c phys) = 0. If we use the field decomposition for the fields and normal order the BRST charge (so that the annihilation operators are to the right). / ' 0v f dx (<% (F0ic) + Ft) (12. in particular.. let us note that the Noether current associated with the BRST transformations has the form 4RST(O. (12.144) . this has the explicit form Qm^ = i f 6k{c^\k)F^\k)F^\k)c^\k)) . the BRST charge can be obtained to be QBRST = J dx (F dvc + Fd°c) = f dx (F0idiC + Fc)= dxFc. therefore. not clear a priori if these conditions are sufficient to reproduce the GuptaBleuler conditions in the case of QED.fabc (3"e°) cbcA . we can obtain the current without the parameter of transformations j g R S T = F^aDuca + FaD^ca . To see that these conditions indeed lead to the GuptaBleuler conditions in QED.141) and.272 Field Theory: A Path Integral Approach have seen is the case with the GuptaBleuler condition.i fabc (d^c") cbcc . for the Abelian theory where fabc = 0.142) where we have used Maxwell's equations in the intermediate steps. (12. From this.140) = u (F^aDvca where we have used the fact that the auxiliary field does not transform under the BRST transformations. (12.
in principle. (12.0) = \A„). £ G F + £ghost = 6 (d»c?A° QBRST. ) ) phys) = 0. This is precisely the GuptaBleuler condition in momentum space and this derivation shows how a single condition can give rise to an infinite number of conditions (in this case. This allows.c°F°) (WAI . (12.^1?FA . phys) = \A„) ® 0.Gauge Theories 273 implies that the physical states must have zero ghost number. (12. namely. (12. . be written as a BRST variation (without the parameter of transformation).148) where we have used the equation of motion for the auxiliary field.F()(fe)cW(fc)) x \Afj) <> \c. Thus. for every momentum mode k).146) Namely. then this implies that c{+\k)\c.c) g = 0.149) (12.145) we note that if the physical states have to further satisfy the condition QBRSxIphys) = ijdk (c(\h)F(+\k) . the physical states should have no ghost particles. states containing equal numbers of ghost and antighost particles.150) .c) = 0 = fW(ifc)S^). Denoting the physical states of the theory as phys> = A /x )(8)c. we feel confident that the physical state conditions are the right ones even for a nonAbelian theory.c).147) (12. in fact. We note that these extra terms in the Lagrangian density can. and must satisfy F^(k)\phys) = ( d ^ + ) ( A . We are now ready to show that the gauge fixing and the ghost Lagrangian densities lead to no physical consequences.
(0 f dx FaFa\0)J = %fdx {\5{<fFa) 0) J (12.5 Ward Identities The BRST invariance of the full theory leads to many relations between various scattering amplitudes of the theory. 12.274 Field Theory: A Path Integral Approach Here we have used the fact that the BRST charge is the generator of the BRST transformations so that the transformations for fermionic operators are generated through the anticommutator of the operators with the generator. We can also show that all the physical matrix elements are independent of the choice of the gauge fixing parameter £ in the following manner (the BRST variation denoted is without the parameter of transformation) ^(00)^^1 = %.151) phys') This shows that the terms added to the original Lagrangian density have no contribution to the physical matrix elements of the theory. ( W A £ + g°F°) + = 0. It now follows from the physical condition that (phys ( £ G F + £ g h o s t ) phys') = ~(phys QBRST.152) = %.J dx (0 [ Q B R S T . where we have used the fact that the vacuum belongs to the physical Hilbert space of the theory and as such is annihilated by the BRST charge. (12. c a n + 0) J = 0. These are known as the Ward identities or the SlavnovTaylor identities of the theory and are quite essential in establishing the renormalizability of gauge .
These identities are best described within the context of path integrals which we will do next. we have not only introduced sources for all the field variables in the theory.c V ) .Gauge Theories 275 theories. we have assumed the convention of a left derivative for the anticommuting fields. but. we can write the generating functional for the theory as Z[J] = eiW^ = N JVA e^ dxC s  . in the presence of sources. in addition. The fields A^ are known as the classical fields . The need for this will become clear shortly. _ T?(c)a _ sw 6Ja: _tSW_ ~ = (cax (c)a = ' 5r)a ((\fabC^C)) = ^a (12155) Here. (12.154) The vacuum expectation values of operators.153) + K»a (D„c)a + Ka (~]:fahccbA Here. (12. Let us consider an effective Lagrangian density which consists of £ T O T as well as source terms as follows £ eff = £TOT + J " M £ + JaFa + i (rfca . can now be written as {AD = Afa= (pa\ SJHa> °W ^ . Denoting all the fields and the sources generically by A and J respectively. we have also added sources for the composite variations under the BRST transformation.
156) On the other hand. the effective Lagrangian density is no longer invariant under the BRST transformations. through a Legendre . with a redefinition of the fields under a BRST transformation inside the path integral. It is here that the usefulness of the sources for the composite operators becomes evident. the generating functional should be invariant. Most often. to the generating functional for the proper vertices. These can be obtained by passing from the generating functional for the connected Green's functions. This immediately leads to 5Z[J] = 0 = N fvA = iujdx (i f dx 6£eS) e^ dxC s  ( j ^ ( ( D M c ) 0 ) + ifja{9rbccbcc) + iVa(Fa}) .158) This is the Master equation from which one can derive all the identities relating the connected Green's functions of the theory. 5J (x)J a (12.8car]a) = ui 1 J^a (D^c)a + i ( fabcrfcbcc + Farf (12. we obtain the change in Ceg (remember that the parameter is anticommuting) to be 8CeS = J»a5Aa + JaSFa + i (rf5ca . the generating functional is defined by integrating over all possible field configurations. (12.157) The measure can be easily checked to be invariant under such a fermionic transformation and we note that the above relation can also be written as / SW 5W 5W \ V 8K» {x) a ' y '8K (x) a ' v . When the external sources are held fixed. recalling that £ T O T is BRST invariant and that the BRST transformations are nilpotent. we are interested in the proper (1PI) vertices of the theory.276 Field Theory: A Path Integral Approach and in what follows. W^J]. T[A]. Therefore. we will ignore the superscript (c) for notational simplicity. however. In fact.
let us note that we can write the Master identity in the momentum space .161) This is the Master equation from which we can derive all the relations. we see that we can rewrite the Master equation in terms of the generating functional of the proper vertices as / Ar I F r) y) I = 0 J \5Aa^(x) 5K»a(x) 8ca(x) 5Ka(x) Sca(x)J (12. iff. K}fdx (j»aAl + JaFa + i {rfca .Gauge Theories 277 transformation. K] = W[J. for example. (12. we have T[A. 5ca = 6T a ~Sc^ = iV . This. Thus. is essential in proving the renormalizability of gauge theories. resulting from the BRST invariance of the theory. Defining a Legendre transformation involving (only) the field variables of the theory (The field variables are really the classical fields and we are dropping the superscript (c) for simplicity. it is clear that ST — —jr 6A° 6T = 6Fa 8T Ja. 5T 5K5T 5Ka 5W 5K^ SW 8Ka (12. between various (1PI) proper vertices.160) Using these definitions. in turn.). From the definition of the generating functional for the proper vertices.159) where K stands generically for the sources for the composite variations.c V ) ) .
the counter terms (quantum corrections) should satisfy such a relation. 29 (1967). Mohapatra. D. World Scientific Publishing (1997).. 42. C . is very fundamental in the study of gauge theories as far as renormalizability and gauge independence are concerned.+ SA (k) 5T ST ST 6T SKa(k) F°«^)=°Taking derivative of this with respect to field variables to zero gives <52r 62T 8Fh(p)5A(lx{p) 5cc{p)5K»a{p) spb. Proc. Gupta. in this way.278 Field Theory: A Path Integral Approach as / < • * ( . Acta 23. shows that this relates the mixed two point function involving FA^ with the two point function for the ghost fields and. Nuclear Physics B139.162 > Lcci \ and setting all 82V = 0.6 References Becchi. N. Phys. S. 567 (1950). 378 (1971). Phys. then. Bleuler. Physical Review D4. K. (London) A63. 12. N. . V. 681 (1950). A. and V.163) A simple analysis."a. Faddeev. L. Gribov.. The BRST invariance. Das.. 127 (1975).. < 12 . Soc. 1 (1978). consequently... Rouet and R. N. A. 5cc(p)Scb(p) (12. "Finite Temperature Field Theory". Physics Letters 25B. Helv. Communications in Mathematical Physics. Popov. N. Stora.8K»a{k) + 5c?(k) " . R.
then the fermions can be thought of as quarks and the theory would describe quantum 279 . 13.Chapter 13 Anomalies We have seen earlier that a continuous global symmetry in a quantum field theory leads to a current which is conserved and the Ward identities of the theory reflect this conservation law. the classical (tree level) conservation of the current is violated by quantum corrections (loop effects). needs to be taken care of before one can make sense of such a theory. On the other hand. If we choose the gauge group to be SU(3). if the regularization used to define divergent amplitudes does not respect the symmetry that leads to the conservation law. Anomalies are quite important from a physical point of view. Sometimes. for example. In this chapter. an anomaly in a gauge theory (namely. Global anomalies such as the chiral anomaly have direct physical consequences. In this case. the divergence of the current density no longer vanishes and one says that there is an anomaly in the conservation law. however. This can happen. In turn. therefore. in the conservation of the gauge current) can render the theory inconsistent and. this modifies the tree level Ward identities of the theory to new ones that are known as anomalous Ward identities of the theory (which reflect the modified conservation law). we will study some issues associated with such phenomena.1 Anomalous Ward Identity Let us consider a non Abelian gauge theory with massless fermions.
1) where a = 1. (13. Thus for example for the chiral symmetry. of course.2. i = 1. The invariances of the theory. which corresponds to an Abelian group of transformations (t/(l)). .N.5) where j» = ftl5l^\ (13. is also invariant under the global transformation 5ft = i\j5ft. lead to conserved currents. and the covariant derivative for the fermions is defined as D^ft = d^ft .4) 5ft = i\fty5 . we have d^ = 0. • • • .2. (13. (13. (13. Here A is an infinitesimal spacetime independent parameter and 75 denotes the (pseudoscalar) Dirac matrix (denned to be Hermitian). ••• . the theory is described by the Lagrangian density C = \F«vF^>a + ift^D^ft. This global symmetry. 5ft = i6a(x) (Tai/jY .3) where 9a is an infinitesimal parameter of the gauge transformation. For a general gauge group of SU(n).iA% (Ta)ij ft .6) The Ward identities are nothing but generalizations of the fact that the currents are conserved.2) Note that this theory in addition to having the nonAbelian local gauge invariance 5Al = Dliea[x). (13.280 Field Theory: A Path Integral Approach chromodynamics. is commonly known as the chiral symmetry of the theory. 8ft = i6a(x) (tpTaY .n 2 — 1.
Anomalies 281 We note here that.mfiiA .6) would not be conserved. the chiral current in (13. the relation (13.10) Let us now make a local chiral transformation of the form (see (13. Rather.7) then the global chiral transformations in Eq. we discover that the axial Ward Identities are violated at the loop level. (13. More specifically. when we calculate various matrix elements regularized in a gauge invariant way. we can treat the A^ field as an external . the modification to the tree level identities arise only at the one loop level. (13. (13. it will satisfy the relation dtfg = liming . i.4)) Sip1 = i\{x)^i)\ (13. There are various ways to understand this violation. if there is a mass term in the Lagrangian density of the form £ m = mtpip .8) For massive fermions.4) will not be a symmetry of the theory and as a result. However.9) and here we have kept the mass term just for generality so that STOT = J d4x (\F^uF^a + i ^ D ^ . (13. Furthermore. for this analysis. Let us first begin with an analysis in the path integral formalism where we can write the generating functional as ZTOT = fv^V^VA^ eiSTOT ..8) would have related various matrix elements and would have led to the axial (chiral) Ward Identities.e.11) Sft = iA(x)^75 • The generating functional should be invariant under such a field redefinition since we are integrating over all field configurations. (13. if the quarks were massive.
<t>n{x)ct>i{y) = 5\xy). then we have / d 4 X ^n(x)(f)m(x) = 5nm . let us note that in this derivation we have assumed that the integration measure is insensitive to changes in the field redefinitions. let us assume that we can solve for the eigenstates of the gauge invariant Dirac operator.f y ^ S y V . we take so that under the redefinition (13.11) SZ = 0 = fvipVip = Iv^Vii /d 4 x{i5 /i A(x)^757 M V + 2A(x)m^7sV' i } eiS ( . if these sates define an orthonormal complete basis (the Dirac operator is a Hermitian operator).2 z m ^ 7 5 ^ ) eiS • (13.d ^ T & V V . Namely.15) Here </>n's represent the eigenstates of the Dirac operator and An's are the corresponding eigenvalues.13) Therefore. To understand the question of the measure. (8^ ( . this implies that ( fd4x or. (13. i7liDtl<j>n = \n<i>n. Y.282 Field Theory: A Path Integral Approach field (since it does not transform under the chiral transformations) and look at the only fermionic part of the action. We have assume that these are discrete states just for simplicity. (13. Therefore. Furthermore.14) i fd4x\(x) This is the naive (tree level) Ward Identity and the question is why does it fail at the loop level. To understand this.16) .2 z m ^ 7 5 ^ ) \(x)) = 0 + 2im^S^> = 0• (13. we should analyze the behavior of the functional measure under such a field redefinition.
17) where we are suppressing the index i for simplicity. We can also expand the new variables ip'(x). For example.20) ip(x) > tp'{x) = ip(x) + iA^(x)7 5 . In terms of bra and ket states we can write tp{x) = (x\il>) = n ^2(x\n){n\ip) (13.• tp'(x) = ip(x) + i\{x)j5?p(x). n (13.19) Let us now make an infinitesimal chiral transformation (see (13. n (13.21) n.11)) ip(x) .ip'(x) in the same eigenbasis. we can write (x). (13. (13. This can. n n and so on.Anomalies 283 We can expand the fermion fields in the basis of these eigenfunctions as ij)(x) = n ^2an(f)n(x). be thought of as a transformation from a wave function basis to the occupation number basis or any other discrete quantum number basis.22) .18) = ^2 (l)ri{x)an = ^2 an<t>n(x). n (13. $(x) = J2<l>Ux)bn.m where the matrix cnm is given by + i [d4x 4(x)A(x)75^ m (a. for example. The integration measure is now clearly understood to be V${x)Vip(x) = Y[dbndan.).
e.x')<f>m(x').284 Field Theory: A Path Integral Approach This can be seen easily in the following way 1/)'(x) = (x\rJ>') = n Tri(x\n)(nW) = ^2(x\n)(n\c\xl>). If we use this fact then we obtain Cnm= dAxdAx'(f)\l(x)c{x)8{x . We note that the chiral transformation (13.)0rn(a.x') is a local function.) = J d4x <f>l{x) (1 + i\(x)<y5) <f>m{x) = 8nm + i d4x X(x)(()l(x)f5(()m(x).24) where Cnm = (n\c\m) = / d xd = fdAxd4x' x'(n\x)(x\c\x')(x'\m) (13. </>)l(a. i. Thus n = ^2(x\n)(n\c\m)(m\ip) n. n = ^a'n4>n(x) (13.)c(a.26) = I d4a. tp'(x) = c(x)ip{x) = (1 + iA(s)7 5 ) ip{x).23) where c denotes the infinitesimal chiral transformation. (13.28) .m =^cnmam4>n{x). (13. (13..27) The above analysis also shows that under the chiral transformation dan > da'n = (det cnm)~l dan .11) is a local transformation so that c(x.25) <l>ll(x)c(x.x')(j>m{x') (13.
where d e t c n m = det ( 5nm + i I d4x X(x)4>\l(x)^(f)rn{2 (13.)750m(rc) J J d 4 z \{x)4>t(x)K(f>n(x) J • = expliJ2f (1331) Since n runs over infinitely many values. Similarly. (13. it is clear that under the chiral transformation the fermionic integration measure changes and contributes a nontrivial amount given by VipVij) »• (det cnm)~2 VipVi/j.30) = exp (Tr f i / d 4 x A(x)(/>]l(a.) Yl n }}m m / d*x A(x)4(x)7 5 ^ n (aj)e(&) = J M2—•ooz—' E n / / ^ A(x)4(x)75e(^) ^ ( x ) (13. Therefore. this sum is ill defined. under the chiral transformation (13. we regularize the expression in the exponent as J2 / n J = d4 * A(x)^(s)7s^„(a.29) Therefore. rather than the determinant appears in this case simply because a n 's are fermionic in character (see chapter 5).11) d6 n f db'n = (det Cnm)'1 dbn.Anomalies 285 and we recognize that the inverse of the determinant.32) .
^2Uk)4>{(k') n = (27r)464(k . namely.k'). v (2TT) ' 4 V x TrA(a. that the regu .» r . Eq.. 7 1 ) ^ ^ = DliD»>f'f\[Dli.32) takes the form Y2 / d 4 x A(x)0+7s0 n (x) = M^ooJ [d4x lim (2TT) 4 ^kA^ki^)^\k~k') . we note that using the properties of the Dirac matrices. namely.Dv] (13. where we are using the matrix representation for the gauge fields. (13.34) To simplify this expression. (13.) e ^ ^ e " ^ ) eik'x d4x J~TX(X) (27r) 4 TV eikxy5e \~**) eik'x .35) 1 = DliD*+%'f>fFlu. therefore. A^ = A*T°. (13.33) then.286 Field Theory: A Path Integral Approach If we use the completeness relation for the basis functions in momentum space. F^ = F^Ta with the generators Ta in the representation of the fermions. we can write = . It is clear.o2 [ 7 ^ .
(13.^ ( (2TT) ' M 2^oo7 7 5 e^(^ D M + ^1'F)e^ fit/ kukV W^ = lim M2—>oo VF v) ' e /*»^A W (^)l*( 1 »yY'lV x TrF„„Fve «• = . we have determined detc nTO = exp I z ^ / d4x X(x)^ll(x)'y54>n(x) J .36) We note here that in the intermediate steps we have used the fact that the "Trace" is defined over the Dirac indices as well as the gauge indices and that in our discussions we have set the coupling constant for the gauge theory to unity.M ^ M e " " * ' • (13.37) = exp ( 3 ^ J d 4 x X(x)e^xPFlluFXp\ .34) can be written as ^2 / d4x A(a.5 ^ 2 y d4i M ^ . Therefore.Anomalies 287 larized exponent in (13.)^(x)750n(x) n ^ = lim / d 4 x 7 ^ I 4A V ^ ) T r e .
(13. For example.42) The anomalies have observable effect.^e^^F^F^) This corresponds to the operator identity =0. (13.40) d^ = 2im^ + ^e^F^Fxp > (1341) = .38) j ^ f ^ x As a result.39) d^2im^)ys. (13. Furthermore.. if we have a theory where there is both .2 i m ^ + J^2F^U where the dual field strength tensor is defined as F^ = e^XpFxP .288 Field Theory: A Path Integral Approach In other words. the functional integration measure changes under the chiral redefinition of the fields and the change in the fermionic measure under an infinitesimal chiral transformation is given by = ((det cnmy2 = T>I/JVIJJ \}V$V\I> X(x)e^xPFlll/(x)FXp(x) (13. which leads to the relation (J d*x X(x) (d^ + 2 i m ^ . we note that the change of the generating functional under an infinitesimal chiral transformation is correctly given by SZ = 0 = J Vm [i J d4x X(x) { j^e^F^Fx. the correct pion life time can be calculated only if the chiral anomaly is taken into account.
13. This theory is invariant under local (7(1) transformations as well as chiral U(l) transformations (since fermions are massless). In natural units.43) where e denotes the electric charge of the fermions. they have precisely this kind of coupling.ieA^) xj. Let us see how this arises. . Secondly. But first let us note some of the special features of two dimensional space time. This is a model that can be explicitly solved and the structure of the resulting effective gauge theory is directly deduced from the anomaly in the theory. First. the model cannot be renormalized unless the anomaly is cancelled. The Lagrangian density describing the interactions of massless charged fermions with an Abelian gauge field is given by C = \F^VF^ + # y (d„ . (13. We would examine the model in two different approaches. This can be seen as follows. . note that in two dimensions the photon does not have any true degrees of freedom since there cannot be any transverse polarization. the action is dimensionless [S}= f d2x £ = 0. the electromagnetic coupling in two dimensional QED is dimensional unlike in four dimensional QED where it is a dimensionless constant. In string theories that are currently fashionable similar anomaly free considerations fix the gauge group to be uniquely SO(32) or E(8) x E(8). Therefore. In grand unified theories where we treat both quarks and leptons as massless.Anomalies 289 vectorial and axial vectorial coupling of the fermions (corresponding to local symmetries) . one cannot make a consistent grand unified theory unless one can cancel out the chiral anomaly (associated with the local invariances) and this leads to the study of groups and representations which are anomaly free.2 Schwinger Model The Schwinger model describes quantum electrodynamics (QED) in 1 + 1 dimension.
(7T = . (13. this determines [e] = 1.44) Furthermore. So this is a completely finite theory. by power counting alone it is easily seen that this theory has no infinities.. In turn. and M = 0] = ^. Note also that in two dimensions if we choose our Dirac matrices to satisfy 70t=7°.7 \ and 75 = 7°7 1 so that t it (7°)2 = 1 .1\2 „l\t .( 7 ° ) 2 ( 7 1 ) 2 = 1. This determines the canonical dimension of fields to be [4J=0. 752 = 7°7 1 7 0 7 1 = .290 Field Theory: A Path Integral Approach which implies that [4 = 2.1 ="!• 0+ 10 0 1 75 = 7 7 =7 7 =77 =75. (7 1 ) /„. (1345) then a possible representation for these matrices can be given by M: :)• .
e 1 0 .^ A p e " " < v . " ) .51) 2 or. Similarly.Anomalies 291 Furthermore. note that in addition to the second rank metric tensor which is diagonal and has the form rT = ( + . (13.49) 1 ~"^e lulu = \^v\l^lu\ = Afv^u. expj5 = axp a"" = e""75 . It is completely antisymmetric and is chosen as e01 = 1 = . we have eoi = .52) .48) (13. where we have defined (in analogy with four dimensions) < V = g h^T"] • This leads to eAp75 = . or.50) (13. we can write 1 75 = ~~z^^vl 7 = (13.47) This is known as the LeviCivita tensor of two dimensions. for the covariant LeviCivita tensor. (13.1 = . (1346) in two dimensions we have another second rank tensor given by e^ = e"" • (13.d o • Using this tensor.
x1) = (T {^{x)i){x')))A . 757M = ^vlu • (13.55) we note that this satisfies the differential equation i V (dp . Furthermore.53) As a result of these identities. (13.292 Field Theory: A Path Integral Approach It is clear now that we can write 7 M 7 y = \ {7^. we note that in two dimensions.56) The free fermion two point function can be related to the two point function for the massless scalar field as iSf>{xx') = i#£$)(xx'). j ^ = e^ju. lv} + \<f.x') = 52(xx'). or. . • Conversely. Denoting by iS*{x . lv\ = rT + ^ 7 5 • y = e^757. (13. (13. and are related as f = e""J5„.ieAp) Sj(x .54) Let us now calculate the fermion two point function in the presence of an external electromagnetic field. the vector and the axial vector currents defined as f = e^V.
57) To solve for the fermion Green's function we note that the most general solution can be written as S$(x . .61) 7 5 ^ ^ . and satisfies the equation • Ai 0) (x ..64) (13.ieAJ = 0 or.x ' ) . (13. ) ) s£ 0 ) (x . # ( x ) = e7MAM or. .Anomalies 293 where iA^ (x . where SF (x — x') is the free Green's function satisfying i0S$\xx') = £i(xx').62) Let us note here that the whole point of this exercise is to show that in two dimensions one can always find a matrix function <p(x) such that fAp = fd^ix).59) i^ (d^iftx)) .x') = e i W s ) * ( x . ( x ) + erf A^ = 0 or.x ' ) .7 ^ < / .x') = <52 (x . (13.58) (13. (13. = e ( ^ + e ^ s ) d„Av = e a ^ +  so that • Tr 4>{x) = led^A* .63) That this is true can also be seen from the fact that in two dimension we can always decompose a vector as A„ = d^a + e^r.60) (13. D</)(x) = e 7 ' V ^ A .x') = (T ^(x^x'))). and • Tr 75<Kx) = ee^F^ . and (p(x) is a 2 x 2 matrix function satisfying (13. (13.
= 0 (x°) iA<+l(x°.f" 2n J0 / ± "2(2TT) L 2 _ c . o > .+ sinA. 1 * f°° / dk ' i\k\x°+ikx . with <t>{x) = a{x) + j5r](x). (13.x). x) = ifdp (± = — / 27rJ0 = —/ j f ° f e** cos **) —ifca. we have A$\x°. and for x° > 0.65) To obtain the complete Green's functions let us note that the time ordered scalar Green's function can be written in terms of positive and negative frequency functions as A^\x) where •A±.cosA. f—i717°cosfca. — 7XA. we have iS^(x°.x) = ±.sinfca.x) .0 —(—z7°A.6 (x°) iA^(x°. k x° > 0.i k x ° (pikx k * e l e +e i —ikx\ ) e*kx° c o g y.294 Field Theory: A Path Integral Approach so that = 'fd^tr + 7^75^7? = 7 ^ ( a + 7577) = 7 ^ .»y0 = ± Therefore.) } k y ! dA.x) ( .x _ 2xJ0 ~eikx°coskx.a.
. we determine that iSA(x .I F v . = 7*75 1 71 2n i"fc.69) This leads explicitly to (j/*(x))= Tl lim .Anomalies 295 = 1 —/ 27T Jo dk (i75 cos kx + sin kx) e lltx 7 7=.ij5 sinfcx)e lKx _ 7 75 / ^ eikx°ii5kx 27T Jo For x° = 0.^ Tr f (d^1 l ei Ax) (13.x1) = e W ( s W ( * ' ) ) _ .70) Neglecting the uninteresting infinite part we obtain 0'"(z)> = .x 2Trix (13. (13.66) iSJ^x) 27T Jo dfc e . (13. f°° _t o = —— / dA.x') The expectation value of the current is defined in a gauge invariant way as f{x) = rl lim V0—T'^ ei>{x')1^{x)eie $* dxXAx . this leads to (13.0—T^ eTr^(iP(x)Tp(x')eieS*'dxXAx Una d .. T. eTr ( 7^ l + i(x — x')—(f>(x) [l + ie(x' x)A1(x)]) 7 —^r 2m(x lim Tr x—*x' _2TTI(X — x') 2TT 2K (13.68) 2iri(x .i75fea. (cos fcz .71) .67) Thus for x — x'0.
5i Tr <f>(x) .d0 Tr ( 75 0(x)) = 2e^ 1 .73) we can write (j°(x)) = ^ Tr lb4>{x) = ^(d0Tr<^)2eJ4°) ~(d°Trt(x)2eA°). d0 Tr 0(s) + <9i Tr (75<K*)) = 2e. 7 5 0(x) = Tre7°4 or.e7V^i (Tr </>(*)) .^ [^di . and Tr7V ^ Tr 0(x) + ^ 7 5 Tr 7 ^ ( 3 ) = Tr e 7 x 4 (13.e^di Tr ( 7 5 ^ ) ) . since or.72) 7 V 7 5 # i Tr . note that (aO = ^ T r (<f>(x)) + so that we can write (f{x)) = ^Tv Q y V ^ i T r </>(*) 7 5 0(x) p5Tr(l5<f>(x)) (13.74) or. .40. using these relations as well as the definition in (13.296 Field Theory: A Path Integral Approach Furthermore. Therefore.73) Furthermore.lerf1 Ar] (13. @ ^ Tr </>(*) + ll5 Tr 75</)(x) we obtain Tr7V ^ Tr 0(x) + i 7 5 Tr = e4.
Going over to Euclidean space.77) P2 (13.2eA») . the chiral current is no longer conserved.~ {2e&iU1 ff'Au . (13. we can write •^TOT = jvA^D^e o(TOT) where 4 T ° T) = J d2xE (jF^F^ + i^D^A .61) Tr (f>(x) = 2 e D ~ 1 d M ^ .79) Namely.Anomalies 297 and (j1(x)) = = — ~(d1Ti4>(x)2eA1) (d1 Tr <j>{x) . 7T V (13. it is also clear from this relation that d^ = d^ju = ^F^ ? 0.2eA») = — {rT ~ d^D'1^) Av . therefore.80) The integration over the fermion fields can be easily done (the action .78) This is a gauge invariant current which is conserved as is readily seen from its form.75) Covariantly. (13. (13. However. we can write (f(x)) = —{dflTr <f>(x) . Let us next study the generating functional for the Schwinger model.2eAl) .76) Putting in the solution from (13. (13. we obtain (ji>(x)) = .
/ d M ? F / « ^ * ) HiHe4)eId2^E(\F^F^) (13.(jv{x)) = 0 2 or. r. x)lu = 0 or. dSefi _ dSefi _ "dd^A..e.85) in which case the EulerLagrange equation becomes dv. (13. let us look at the Euler Lagrange equation satisfied by the photon field.T r (i$ + e ^ ) " 1 e 7i . (13. dAv " U or.. d^F^ + e Tr SJ{x.84) This can be readily seen by choosing a covariant gauge such as the Landau gauge 8 ^ = 0. (13. d^F^ .86) which is the equation for a massive gauge field. 7T e 2 (13.mlhA„ = 0 (nEm%h)Av = 0.FnV .duDldx^j Ax = 0. UEAV . or.) = 0 or.m2vhAv = 0 or.83) This now describes a massive photon with the mass of the photon given by <h = . . (^ = JDA^e3^.298 Field Theory: A Path Integral Approach is quadratic in the variables) and the result is Z = fvA^det = I'D Ay. dpF^ .(Sux .( . (13. dpF^ .81) SeS = J d2xE (^F^FfJ) .In (i? + eAT). where {i$ + eA\) e .82) To understand the meaning of this better.
(13. (13.Anomalies 299 The effective action for the photon field in this case becomes Seff = AF^F„V + .. % ( V .')Al. Another way of seeing the same result is to look at the generating functional for the fermions (in Mankowski space) Z= where S= It follows now that oAfl(x) fvipV4> eiS .c ^ D . Integrating this.88) .1 ^ ) Av . we obtain Z= = e—^ I e^Id'xA^rrd^ar^A^ tfxA^ifdKnid^A. = / VipVip J ieip^ip(x)eiS = • 9 i(f(x))Z is* {rT TV d»Uxdv)AvZ. we can write ZTOT = (vApVm / e^MiW)^ m2 I I VA^e ijd2x[ \FlluF»v+2±(r)fl'/d»a1d.87) which again shows that the photon in the effective theory has become massive. fd2xiTp(@ie4)ijj. Therefore. VA^ elb*s .
Note that the nonlocal effective Lagrangian density 1 Til 9 £ eff = —^F^ + ^A„ ( T T . the novel feature in this case is that there is no scalar field in the theory. On the other hand once we accept this form of the effective action given.ieA^) ip (13. jFfiuFpv + i ^ (d^ . (13.^B(x)B(x). massive scalar field. the complete Schwinger model including all quantum corrections is equivalent to a free.mvhe^A^Bix). This is often referred to as the bosonization of the model.93) . In this sense it is called a dynamical symmetry breaking and the symmetry that is broken in this case is the chiral symmetry.300 Field Theory: A Path Integral Approach The phenomenon of the gauge field becoming massive through spontaneous breaking of a local symmetry is quite well understood.89) can be written in a local manner by introducing a scalar field B(x) as jCeS = ~\F^F^ + ~dllB(x)d"B(x) . (13. Let us now analyze the Schwinger model from a different point of view. (13.90) Integrating out the scalar field B(x) (on which the action depends quadratically) we obtain the earlier form of the effective action. is given by / VA^VipVipeSE c(TOT) .91) That is. we can integrate out the A^ field completely since the Lagrangian density depends on the gauge fields at most quadratically. (13.^ D " 1 ^ ) Av . The Euclidean generating functional. as we have seen.92) where the Euclidean action has the form 4 TOT) = / d 2 * . However. The resulting Lagrangian density in this case is given by Asff = dliB{x)&iB{x) . The fermions have completely disappeared from the theory.
(x) = e * e 7 5 Q ( a V(z).99) Therefore. ieA ii) tp ip (13. This is quite different from what we have seen in the earlier analysis. the fermion would decouple and the theory would appear to consist of free photons and free massless fermions.ier/sdrf) ip'.94 ) (13. (13.97) (dp .96) In other words. . Let us note next that if we make a local chiral transformation of the form ^. (13. ${x) = ft{x)eie^a^ .iee^rj) = #7^(^^750^77)^.98) then the fermion Lagrangian density becomes = # V {&„ + ie^5d^a(x) .95) (13. Therefore. the Landau gauge condition 0 ^ = 0 ==> a = 0. in this gauge we can write A^ = e^d^rj and the fermion Lagrangian density takes the form £ = #7^ (^ = i^f.Anomalies 301 Since the theory has a local gauge invariance let us choose a covariant gauge condition such as the Landau gauge condition V^ = °Let us recall that in two dimensions we can write A^ = d^a + e^duTj. The real story lies in the change of the functional measure. if we choose a(x) = rj(x). ( 13 .
(13. Let us define the eigenvalue equation itf<f>k(x) = «7M (dp + ieibd^ne) = XkMx) • .104) We can also expand the functions tp and ip as ip(x) = ^ a m 0 T O ( a . $'(x) = 4){x) .102) .103) If the <?Vs form a complete set.101) Furthermore. (13.105) .j. The fermion Lagrangian density in this case is given by C = itp^ (d. To do this.iee^ 7 5 . / • J2Mx)d>l(y) = S2(xy). (13. let us assume that we have already made n infinitesimal transformations. let us assume that the finite transformation with a(x) is obtained by making N successive infinitesimal transformations with parameter e(x) such that lim e—>0. iV—>oo Ne(x) = a{x). then as discussed earlier.302 Field Theory: A Path Integral Approach In this two dimensional case the change in the measure has to be calculated very carefully.100) The infinitesimal chiral transformations are given by ip(x) = ip'{x) + iey5e(x)tp'(x). + ie75<9/J(ne(x)) . m i>{x) = YJbm<t>l{x). they satisfy d2xE 4>\{x)(f)m(x) = 5km. ip(x) = ip'(x) + iee(x)ip'(x)j5 which can be inverted to give ip'(x) = tp(x) — ie^etp .ie^d^rf) ip. k (13. ) .iej5r]) 4>k(x) (13. (13.
(13. as before.109) Similarly.ie J d2xE e{x)4>\n^A = exp lie^2 d2x E e(x)(f)m(x)f5(l)m I . calculate the determinant.108) (13.111) Let us.110) such that XtyXty = (det cmtf Thp'Vr/. m (13.Anomalies 303 so that VTpViP = Y[dbmdam.x).ie d2xE e(x)(/)mj5<f>i j J = exp I TV In \5mi . (13.'. J J dam = det cmi f j da'm .112) .106) If we make an infinitesimal transformation. m (13. m m d2xEe(x)(j)m(x)'y5<pe(x). m m.l (13. l[dbm m = detcmel[db'm.107) where cmt = <W ie Therefore. we can determine that ip(x) » ip'{x) = ^2 a'm<t>m{x) = ^2 cmeae<t>m(. (13. therefore. det cmi = det f 5mi .
icy5dur]) = d^d^ + e2dll{ne .vdIJI.r?)<9M(ne .) e ~ ^ d2xE e ( x ) ^ ( x ) 7 5 in A2 /ft / e^^m{x) xeik'xl5e^eikx4>m(k). Using the completeness relation (13. m (13. (13.(ne .ie^nd^e + ie^d^rj) x (du + i&y5nd„e .rj).115) Let us note the two dimensional identity = liiTtv (df.)75</)m(a. (13.304 Field Theory: A Path Integral Approach The exponent can be calculated in a regularized manner as before ie^2 / d2x B e(aO$„(aO750m(aO d2xE e(x)^]n(a.114) the regularized exponent in the above relation takes the form lim ie [d2xE M2HX) J = £ r % e ( z ) Tr e~ik'x^ (27T) 2 e^ eik'x .113) J24>m(k)ftm(k') = (2ir)252(kk'). .rj) 2ieeIJ.116) .rj)dv + ie^d^d^ne .
119) x=Y[(l n=0 + ae + nbe2) . (13.118) As we make N infinitesimal transformations.Anomalies 305 Using this we obtain = fc / ^ <x) d ^ ^ .rj) J n=0 \ ^ J J Using the result that if N / 2 f \ V^Vxj/. (13.77) ] V^'V^1 = (1 + — f d2xE e{x)d^{ne(x) .117) Thus. we see that under this single infinitesimal transformation. the measure changes as N Vi>Vi> = Y[ ( 1 + — / d2xE e(x)df.rj) J V$V$ .dfM(ne . (13. the measure changes as VipVip = (det cmi)2 Vi>'Vip' = exp ( — f d2xE e{x)dlld^{ne{x) .120) .*l) • (13.
if we choose a(x) = rj(x). there is a change in the functional measure leading to the .122) This gives the change in the measure for a finite chiral transformation as V^ViJ. N In a. Therefore. — ^ n=0 N In (l + oe + nbe2) = ] P (ae + nbe2) n=0 = aA^e + iV(iV + l)e 2 .121) +f< (13.123) Furthermore. (13.124) = exp ( — = exp f . in the limit e — 0. then we obtain Vi>V*p = exp ( — f d2xE ri{x)d^ \\r]{x) .306 Field Theory: A Path Integral Approach then. iVe = a. we obtain > > lnx = a a + ^ a 2 . iV — oo. = exp (?fd2xE a(x)dfid(i (\&{x) .  Thus if we make a finite chiral transformation to decouple the fermions.r](x)\ J Ztyi'Zty/ = exp (h \^XE d^(x)d^xn V$W / d2xE eIJ. =>a: = aa e (13. (13.vdur}(x)eIJt\d\r)(x) j Vrp'Vip' / " d 2 ^ AM(x)AM(s) J Th?Vi(/.Vj) V^Vtj/.
K.125) 1 e2  / • (13. D 21. Nuovo Cimento 60 A.3 References Adler. Rev. This method can be extended to a very general class of two dimensional Abelian models which reduce to different known models in different limits. Zeit. D23. 128. 707 (1995). A. we see that the solubility of the model is closely related to the Jacobian of the chiral transformation that decouples the fermions. Rev. Jackiw. J. Das. 177. 558 (1981). Phys. Roskies. S. Phys. Phys. J. in turn. Rev. is connected with the anomaly of the system. Phys. Lett. and F.. Bell. Rev. C 67. 2426 (1969). 1195 (1979). Phys. S. This. D 33. Mathur. Rev. 2848 (1980).Anomalies 307 generating functional IvApTty'Vtf where SeH = I <$ XE 2 SeS (13. Fujikawa. Schwinger. R. and M. Das. and R.. 2425 (1962). Hott. Phys. 47 (1969). In this method. Schaposnik. and V. Phys.. 489 (1986). 42. A.126) This shows again that the photon becomes massive in this theory and the fermions decouple completely from the spectrum. 13. Rev. .
.
the value of any observable quantity averaged over the entire ensemble will take the form (A)=A = J2Pn(n\A\n) = J > n A n . Each system in this ensemble can. Thus.1) where we are assuming that the energy eigenstates are normalized and that An = (n\A\n).Chapter 14 Systems at Finite Temperature 14. Let us consider not one quantum mechanical system. completely statistical in the sense that pn can be identified with the number of physical systems in the state n) divided by the total number of systems in the ensemble. (14. we can define pn to represent the probability of finding a system in the ensemble to be in an energy eigenstate \n). Namely. of course. Such a situation is quite physical as we know from our studies in statistical mechanics.2) denotes the expectation value of the operator in the quantum me309 . it can be an ensemble of oscillators or any other physical system. n n (14. For a given ensemble. be in any eigenstate of energy. but a whole collection of identical quantum systemsan ensemble. Thus. for example. Let us further assume for simplicity that the physical system under consideration has discrete eigenvalues of energy. of course. This is.1 Statistical Mechanics Let us review very briefly various concepts from statistical mechanics. we may have an ensemble of physical systems in thermal equilibrium with a heat bath.
. k is the Boltzmann constant and T the temperature of the system. Namely. in this case. Z(P) = Txe~m where we have defined . we can write pn = — e'^r . However.3) It is in general very difficult to determine the probability distribution for an ensemble. is given by the MaxwellBoltzmann distribution. Z = y ^ e n kT = V \ n  e n \n) (14. an ensemble interacting with a large heat bath. (14. and if we allow sufficient time to achieve thermal equilibrium. namely. in this case.4) Here En is the energy of the nth quantum state. Thus.5) or. The normalization factor Z can be determined from the relations for the probabilities in Eq. we have the average in a quantum state (expectation value) and second. then we know that the probability distribution. 1 > Pn > 0. if we are dealing with a thermodynamic ensemble.310 Field Theory: A Path Integral Approach chanical state \n). pn has to satisfy certain conditions. (14.3) as J2P" = 1 1 V^ Sn 1 or. Namely. we have the averaging with respect to the probability distribution of systems in the ensemble. there are two kinds of averaging involved here.2^ e n *T =1 l3H or. ]>>n = l. » = h (14 6) ' . (14. Being a probability. First. n for all n.
For a statistical ensemble. The amount of order or the lack of it. it is easy to see that the thermodynamic average of any quantity defined in Eq. (14.8) dp lnZ(/3). the average energy associated with the system follows from Eq.1) will be given by Z((3) 1 •fiEn(n\A\n) n Z(P) J2 {n\e0HA\n) n 1 T r ( e " PHA) ~ Z((3) Tr(e'f*HA) ~ TVe^ ' <14T> In particular. is denned through the entropy as S = ^2pnlnpn = (hip).7) to be (H)pU Tr(e^ff) ^e_m 1 ( d Z Z(P) V d ^ \ d(3 J 9(3 (14. for an ensemble.Systems at Finite Temperature 311 Z(/3) is known as the partition function of the system and plays the most fundamental role in deriving the thermodynamic properties of the system.9) . (14. (14.
it is an ordered ensemble. (14. n PnEn + In Z((3)J2Pn n = 0U + lnZ((3) = (3^lnZ(P) + lnZ(0) = Pw\3>*m] #9 n (14 n)  Here we have used Eq. for a fixed m . and the entropy. (14. the larger the number of states the physical system can be in. (14.312 Field Theory: A Path Integral Approach By definition. ™ Z{(3) Therefore.10) For such an ensemble.4).8) and (14. therefore. all the individual systems are in the same energy state and. it is clear that the entropy is always positive semidefinite since 0 < pn < 1. U.^ n Pn^Pn = ^ P r l (Rln^)) n = (5 Y.3) in the intermediate steps. Furthermore. On the other hand. Given the internal energy. as we have seen in Eq. S. we can calculate the entropy of the ensemble to be S =. the free energy . the more disordered the ensemble becomes and the entropy increases. its value is zero for a pure ensemble for which Pn = $nm. For a thermodynamic ensemble.
we know that below the Curie temperature. such as the magnets. For T > Tc. Namely. The amount of residual magnetization decreases as the temperature of the system approaches the Curie temperature and vanishes at Tc. the hysteresis effect or the effect of spontaneous magnetization provides an example of a phase transition. the system exhibits no . One of the major interests in the study of statistical mechanics is the question of phase transitions in such systems. Even in solids.12) that the partition function takes a particularly simple form when expressed in terms of the free energy.12) In terms of the free energy. if a magnetic material is subjected to an external magnetic field. Phase transitions are all too familiar to us from our studies of the different phases of water.Systems at Finite Temperature 313 for an ensemble can be written as = ±lnZ(/9).14) We have gone over some of these concepts in some detail in order to bring out the essential similarities with the concepts of path integral that we have been discussing so far. T c . then the material develops a residual magnetization even when the external field is switched off. we can define the other thermodynamical quantities as r) 3 BF1 s <G H^IZ{f3) = epFW . (14. (14. ta (4 1) 1 3  It is also interesting to note from Eq. we can write (14. Namely.
the Ising model assumes nearest neighbor interaction for the spins which are supposed to be pointing only along the . This model goes under the name of planar Ising model or the Ising model in two dimensions. ri2). — 1 for spin down. The behavior of physical systems near the critical point is of great significance. let us recapitulate how one uses statistical mechanics to study critical phenomena. 14. a critical temperature separating the different phases of a magnetic material. The temperature T = Tc is. then. Accordingly. we assume { 1 for spin up. stands for either of the two unit vectors on the lattice. (14.16) where we have assumed a simplified coupling for the problem. before we discuss this. This should be quite familiar from our studies of atomic systems where we know that elementary particles with a nontrivial spin possess magnetic dipole moments Let us consider a square lattice with equal spacing in both x and y directions. let us discuss the critical exponents in the context of a specific model which explains the properties of magnetization quite well. In fact. The crucial feature of this model is that it ascribes the magnetic properties of a material to its spin content.314 Field Theory: A Path Integral Approach spontaneous magnetization. Let us also assume that at each lattice site labeled by n = (ni. They can also be studied with equal ease using the concepts of path integrals. the Hamiltonian for the Ising model is taken to be H = J^2S(n)S(n n. In simple language. However. Here fj.(i + p). This can be studied from the point of view of statistical mechanics quite well. (14. therefore.2 Critical Exponents To fix ideas clearly. let us assume that the spins interact as locally as is possible.15) Furthermore. there is a spin S(n) which can either point up or down.
ft + fl) + B^2S{n). for the present case.14). (14.Systems at Finite Temperature 315 zaxis.18) has to be carried out. (14. Let us note now from Eqs. then the Hamiltonian does not change. n (14. takes the form Z(f3. (14. if iV denotes the total number of lattice points. here. The true partition function. It is clear that if its value is positive. It is worth pointing out here that the Hamiltonian for the Ising model has a discrete symmetry in the sense that if we flip all the spins of the system.17) The partition function defined in Eqs. such a coupling is known as a ferromagnetic coupling. measures the strength of the spinspin interaction.18) The summation. Let us next subject this spin system to a constant external magnetic field B. of course.S ) = rftefiH = £ config eW . (i4.B) = e . then there are 2 ^ possible spin configurations over which the summation in Eq.' i * = *('»X»«"') or.19) Using the translation invariance of the theory. In this case. is over all possible spin configurations of the system. The constant. Consequently. Let us note that at every lattice site. we can write (S(n))p = (S(0))p.W . the spin can take two possible values.17) and (14. then the coupling is known as antiferromagnetic. in this case. Conversely. Accordingly.5) and (14. J.18) that. H = lift ( E S M e . we have i = . (14.20) ." " ) = £>(>>))«>. has to be calculated in the thermodynamic limit when N —> oo. then a minimum of the energy will be obtained when all the spins are pointing along the same directioneither up or down. (W. if J is negative. the Hamiltonian becomes H = jJ2S(n)S(n n.
from Eq..19) and (14.22) This is. if {S(0))/3\B=o = 0.. (14. (14. (1424) then.m n / B_Q = jf (NJ2(S^S(°))0N2(s(°m V . It is large at those temperatures where the correlation between the spins is large. the magnetic susceptibility is completely determined by the spinspin correlation function. B) = jj(£.25) . . the mean magnetization per site can be obtained to be M(P. of course. = 1 1  . n S(n))0 = (S(0))f.e. in this case.21) Thus. (14. . namely. we have x = (3 J2 (S(n)S(0))f. we see that the magnetic susceptibility is related to the fluctuations in the spin. (14. Thus. a function of both the temperature and the applied magnetic field and its value can be calculated once we know the free energy or the partition function. The magnetic susceptibility is proportional to the rate of change of magnetization with the applied field and is defined to be _ _ dM B=o i N d2F *>B* B=0 P % £ (S(n)S(m))f.316 Field Theory: A Path Integral Approach so that we obtain ^ = N(S(0))f.21). Namely. no spontaneous magnetization. (14. Note that if the system has no net magnetization.(£ S{n))} \n. i.20). (1423) / B=Q n where we have used Eqs.
However. as the temperature of the system is lowered to the critical temperature.27) can be calculated once we know the partition function. namely. the system will show no spontaneous magnetization simply because the thermal motion will dominate. (14.29) . Let us emphasize here that the parameter (3 in the exponent is not ^ which was defined earlier but represents a critical exponent. the system may develop long range correlations and the behavior of the correlation length near the critical temperature is parameterized by another critical exponent of the form ^(r)~(rrc)".Systems at Finite Temperature 317 If. The critical temperature and the behavior of spontaneous magnetization near the critical temperature. how the magnetization vanishes as the temperature approaches the critical temperature M\B=0 ~ (T . (The notation is unfortunate. it is clear that the thermal motion will not allow any appreciable correlation between the spins. but this is the convention. we note that the discrete symmetry of the system is spontaneously broken. (14. for some temperature (3 > 0C. the system shows spontaneous magnetization or residual magnetization. In this case.r c )~T . (14. £(T).) Furthermore. the spontaneous magnetization defines an order parameter in the sense that its value separates the two different phases.Tcf . (14.26) then. at any temperature by analyzing the magnetic susceptibility. The spontaneous magnetization vanishes as we approach the critical temperature and for f3 < (3C. For very high temperatures.28) The magnetic susceptibility may similarly become large at this point and its behavior near the critical temperature is parameterized by yet another critical exponent as X(T) ~ (T . we find in our spin system that (S(0))p\B=0^0. We can similarly calculate the correlation length between the spins.
31) to be En=(n Z(0) = Tref)H ePEn n e 2 J>"^ \n=0 1 _PiO f e 2 1 . the energy levels are given by + l)uj.31) 2. (14. other thermodynamic quantities in the system such as the specific heat defined as C = .l. n = 0.5) and (14.3 °) may also display a singular behavior at the critical point and all these can be calculated once we know the partition function.. (14.2. where we have set h = 1. 14. We know that for an oscillator with a natural frequency to. the partition function can be derived using Eqs. we can calculate the thermodynamic properties of the system.3 Harmonic Oscillator The calculation of the partition function for the one dimensional quantum harmonic oscillator is quite straightforward. For this system then.&j*> T d2F ( 14 .318 Field Theory: A Path Integral Approach Similarly.eP" 1 e2 e~~2 (14.32) 2smh^ Since we know the partition function. .
(3. (1. we note that the trace can be taken in any basis. the integrand really is the transition amplitude between the same coordinate state in the Euclidean time with (3 — ^ (T is the temperature here) playing the role of the Euclidean time .35) We now recognize the integrand in Eq. if we choose the coordinate basis in the Schrodinger picture. (3.34) We note here that we have set h = 1.84) with J = 0) Six*] = r 0 ^ £* b i l l UJL M + x)) cos coT . In other words. (14. (14. (14.T\xh0) = (xf\eiHT\xi) =N fvxeiSW W^wCs^)'^"whereas we have noted earlier (see Eq. (14 33)  (14. From the definition of the partition function. Z{0) = TrefiH.35) merely as the transition amplitude (see Eq. In particular.66)) (xf. J = 0 in the above equations and T here denotes the time interval between the initial and the final points of the trajectory.43)) for the harmonic oscillator with the identification T = 0.Systems at Finite Temperature 319 Let us next see how we can calculate the partition function for the harmonic oscillator through the path integral method.2xiXf] .36) Xf = Xi = x . then we can write Z(J3) = f dx {x\epH\x). Let us recall that we have already calculated the transition amplitude for the harmonic oscillator which has the form (see Eq.
then.38) 1 1 / 8LO F = .35) i dx mio 2m{—isinhfiu) mui V 2TT sinh (3u muj 2 f dxe (iiSffc(^i)* 2 ) Z7T s m h pu ) J muj ir 2irsmh(3cvJ ^mwtanh^ 2sinh/3a. (14.320 Field Theory: A Path Integral Approach interval.InZ = .37) This is.32). (14.39) .ln2 + l n s i n h ^ p p \ I (14. we obtain from Eq. the partition function which we had found by a direct calculation in Eq.. Using this. Let us note next that since m we obtain = oPF = 1 2 sinh 0U (14. of course.tanh^ 4 sinh2 &) 2 sinh %* ' (14.
for very high temperatures or small /3.42) This is.41) Namely.Ddimensional quantum statistical .13) that UJ cosh ®Y = 2 sinh &• go) /3m o. we get < ^ = t/~ l + ^ .e^ +^ T . of course. we note from Eq. (H)p = U=^ 1.) This analysis of the derivation of the partition function for the harmonic oscillator from the path integral is quite instructive in the sense that it shows that a (D + l)dimensional Euclidean quantum field theory can be related to a . (14. in such a case. (We expect the system to behave in a classical manner at very high temperature. (14. (14. it tells us that for low temperatures or large (3. e 2 + e n /3a> 2 /3J*J ^ e 2 —e 2 w 1 + eP" 2" 1 .Systems at Finite Temperature 321 Consequently. the oscillators remain in the ground state. the expression for the equipartition of energy. Among other things.e#" 2 or.^ f c T .40) This is exactly what we would have obtained from Planck's law (remember that K = 1). (14. On the other hand. we have (H)p = U ~  .
Let .36). (14. arises from the trace in the definition of the partition function.322 Field Theory: A Path Integral Approach system since the Euclidean time interval can be consistently identified with P = ^ as in Eq. Furthermore. we have Z(J3) = N f VxjjViP eSE ^^] with the boundary conditions V(0) = </>(/?). in such a case. as is clear. are assumed to satisfy a periodic boundary condition </>(t + /3) = <f>(t). This way of describing a quantum statistical system in equilibrium through a Euclidean path integral is known as the Matsubara formalism or the imaginary time formalism (since we rotate to imaginary time). can be written exactly in the same manner but with antiperiodic boundary conditions. A careful analysis for the fermions shows that the partition function.43) Here we are assuming integration over the end points which is equivalent to taking the trace. (14. In fact. = N f ^(O)=0O8) (14. for fermions. ^(0) = $((3).46) These boundary conditions can be shown to be related to the question of quantum statistics associated with the different systems. Namely. (14. in this case.44) This. we note that the field variables. the relation between the two for a bosonic field theory can be simply obtained as Z(J3) = N f Vcf)e~SE V<Pef°dtSd3xCE.
therefore. t. The identification of the two above. in this case should be treated as a space variable and not as a time. the generating functional takes the form Z = N fvxeyodt^mi2+v{x^ . This connection can be simply extended to a field theory where the Euclidean action would act as the Hamiltonian for the corresponding classical statistical system.mx2 .48) Here we have put back Planck's constant for reasons which will be clear shortly. the generating functional for such a system will have the form Z = N I Vx eK s N = N I Vx e ^ dt L . (14. (14. relates the quantum fluctuations in a quantum mechanical system with the thermal fluctuations in a corresponding classical statistical system.47) Then. If we rotate to Euclidean time. Let us also recall that the Planck's constant measures quantum fluctuations in a quantum mechanical system whereas temperature measures thermal fluctuations in a statistical system.49) This has precisely the form of a classical partition function if we identify H= f dt [\mx2 + V(x)j (14. These go under the name of real time formalisms. (14.V(x).Systems at Finite Temperature 323 us note here without going into details that there exist other formalisms which allow for the presence of both time and temperature in the theory simultaneously. (14. Let us consider a quantum mechanical system described by the Lagrangian L = ]. . The Matsubara formalism also suggests that a (L>+l)dimensional bosonic Euclidean quantum field theory can be related to a (D + 1)dimensional classical statistical system in the following way.50) as governing the dynamics of the system and h = kT = ^.51) Here we should note that the variable.
(14. for the calculation of the partition function. from Eqs. (14. (14.36) and (14.4 Fermionic Oscillator Let us next calculate the partition function for a fermionic oscillator with a natural frequency u. (5.324 Field Theory: A Path Integral Approach 14.^ i ' $f = $i • (14. namely. It.1 that the Hilbert space for the fermionic oscillator is quite simple. (14. In fact. both using the path integrals and the standard methods. (5. then.53) The evaluation of the partition function for the fermionic oscillator follows from the form of the transition amplitude derived in Eq.5) that for the fermionic oscillator we have using Eq. we have to impose antiperiodic boundary conditions (see Eq.13). follows from the definition of the partition function in Eq.12) and (5.46)). We note from our discussion in section 5. we require </>/ = .54) We can now calculate the partition function for the system with the identifications in Eqs. we note that it is like a two level system with energy eigenvalues Once again.52) Z(0) = Tref}H = e^0 + em = e2 + e 2 = e^(l + e^)=2cosh^.54) as well as the result in .91). we have set h = 1 for simplicity here. (14. Following our discussion in the last section. we note that in the case of fermions.
We note now from the definition in Eq.27).12) that the free energy for the fermionic oscillator is given by F(J3) = .91) as = er /3ul I dtjatyi ( l .53) for the partition function of the system. (14. (14. Here we have used the nilpotency properties of Grassmann variables (see Eq.^ In Z{(3) = .17)) as well as the integration rules given in Eqs. This is exactly the same result which we had obtained earlier in Eq. (5. (14.(1 + e " ^ ) ^ ) (l + e " ^ ) (14.Systems at Finite Temperature 325 Eq.57) It now follows that for low temperatures or large j3 (/3 = ^ ) . we have {H)P = U~^. (5. (14. <tf>/J = tf = _ ! + _ £ _ .55) = e 2 = 2 cosh — .26) and (5.56) The average energy for the ensemble can now be calculated from Eq. (5. (14.13) to be (H)^ = U=~(PF(f3)) LO s i n h ^ ~~2 cosh^ 0U1 0U to e 2 — e L 2 2 e 2 + e or. (14.i fin 2 + In cosh ^ \ .58) .
51. Kogut. 14. World Scientific Publishing. T. A. the system likes to remain in the ground state for low temperatures whereas for high temperatures or small (3.. Mod. North Holland Publishing. (14. Rev. J. H. John Wiley Publishing. Prog. "Finite Temperature Field Theory".59) In this case.. Phys. "Statistical Mechanics". Umezawa.5 References Das.326 Field Theory: A Path Integral Approach Namely.. Theor.. Phys. et al.. Matsubara. K.. 351 (1955).. we obtain &. we see that the average energy of the system goes to zero inversely with the temperature which amounts to saying that the system tries to populate equally the two available energy states. 659 (1979). 14. "Thermofield Dynamics and Condensed States". Huang.„**£—£. .
(15. which is a constant. Sj=±l e _ / m =E Sj=±l e(PJZ?=x°i'i+iPBi:?=1si).Si (lfU) Here we have assumed that the total number of lattice sites is N and that the spin system is being subjected to an external magnetic field. B. with the example of the one dimensional Ising model. 327 . namely. which we have developed in the last chapter. let us assume periodic boundary condition on the lattice (cyclicity condition).3) and ask whether there exists a quantum mechanical system whose Euclidean generating functional will give rise to the partition function for the one dimensional Ising model. Si+N = Si. Z(0)= Y. The Hamiltonian for spins interacting through nearest neighbors on a one dimensional lattice (chain) is given by JV N H = jj2sisi+i+BY.1 One Dimensional Ising Model Let us pursue the ideas of statistical mechanics. by definition. The classical partition function for this system is. (15>2) To study this system.C h a p t e r 15 Ising Model 15.
(15.5) We can now calculate the Euclidean transition amplitude for the quantum system described by Eq. (15. which is defined to be <Sfine. Furthermore. (15..T ^s i n ).10) .3.e7£73)si).7) and introducing a complete set of eigenstates of £73 at every intermediate point. (15.4) between two eigenstates of £73.9) Note that if e is small.e ^s7Vi)(s2e" e H ? si).328 Field Theory: A Path Integral Approach Let us consider the quantum mechanical system described by the Hamiltonian Hq = a ai + 7 cr3 . s—±1.8) and write (s i+ ie~ eH «si) =* (s i + 1 (l + ea£7X .JV. (15.. Let s) denote the two component eigenstates of (73 such that &3\s) — s\s).8) where the intermediate sums are for the values i = 2. then we can expand the individual exponents in Eq. we obtain \Sfinp Pin/ = ] T (sMe" e ^5jv)(sive. we have also identified Sin = Sl.6) Dividing the time interval into N steps of infinitesimal length e such that (for large N) Ne = T.4) where o\ and £73 are the two Pauli matrices and a and 7 are two arbitrary constant parameters at this point. (15. (15. (15.. S6n = SM(15..
/1 x 2 forn>l.13) Using the relations in Eq. we also have the following identities.13). ^ = .( S j + Sj+i) ) . forn>l. (Sisi+i)j 1 \ n = f(sjsi+i)j x m . 1 \2n (si + Sj+i)) = ( . + s i + i ) J + e a l . (15. (15.si+1 = (si + Si+i).( s i . (15.S i + i ) 1 .( S J .( S J + SJ+I). (sj + s i + i ) j l(sisi+i)) /1 =0. x2 (si+i\ai\si) (si+1\a3\si) 1 = ( .m>l. I \ 2n+l (si + Sj+i)J forn>0. we obtain {si+1\eeH"\Si) 2(si l. A2 l + si+1). (15. 1 ^2 (Si+l\Si) = ( . we can obtain (for constant parameters 5 and A) . then.11) which can be explicitly checked. s» = ± 1 . forn.Ising Model 329 Using the relations.ej(si A2 (\.12) Prom the fact that for any i.( s j + Sj+i) ) .
( s i . n=\ ^ n! A ( . \ 2 fysi + Si+iU + Q ( s < . <5n / l .1) ( (si + Si+i) +smh5((si + si+1)j (15.s i + i ) j = 1 . t ( 2n ) ! V2 = 1 + (e .330 Field Theory: A Path Integral Approach A((sisi+1)) +5(sj+si+1.l ) ((siSi+i)j + (cosh5.s i + i ) j +<5(si + s i+ i) .*+i)J + Z ^t(2n)! V2 si T ^ T T ( n( A /l \* J±i.ai+i) J An / l v An/1.V 2n n=l 00 rx=l 00 ^rU ( *. (15.1) ( y * .14) Let us now use the algebraic relation 2 /.15) . + (Si+Si+l) ^[ U \ 7 £2n+l v /]_ \ 2n+l + E(^TT)!(2 ^ n=0 ~ ' S2n v ( + Sl+l) N2n /. §2n+l 1 + £^_^( ai + fli+1) ^ s2n fK + s i+l) l + ( e A .
with the identification e = ea. we can write (8i+1\eeH<\8i) = e (A(i(«*+o) a +*K*+* + o) _ (15.e 7 . _ es sinh<5= Equivalently. (15.17) e5 = l . si+i) (15. e8 = 1. (15.18) (1519) Consequently.16) We note that this has precisely the same form as the transition amplitude between two neighboring sites in Eq.e H «si) "y Si = ± l e (AEiLi(("i«i+i)) +*Eiii(«i+«i+i)) = J2 Si = ± l e (AE£i£(i**+i)+*E£i«0 . we can write Eq. using this identification. (15.14) also as = cosh 5 ( (si + Sj+i) j + e A ( {s^ + smhs((Si + si+i)j . = —€7.Ising Model 331 Then. we obtain T[e THq ~ = E 3i=±l = (si\e~tH"\sN) (sNle^lsNx) 2 • • • (s 2 e.12) provided we make the identification e<5 __ e5 cosh 5 = e = ea.
332 Field Theory: A Path Integral Approach e— \ Si = ±l e (T^i=is's*+i+.22) where Z{j3) represents the partition function for the one dimensional ising model. let us rewrite the exponent in Eq. Note that e(/3JsiSi+1~/3B±(si+si+i)) = e (2/3Jxi(l+lsisi+i)/?Bxi(si+si+1)) (20j(\{sisi+1))20Bl(si+si+1)) pj e^ e aPJ cosh (3B ( {Si + Sj+i) ] + e 2/3 J (Si  Si+li •smh/3B(si + si+i) (15.23) .20) provided we identify A = 2/3 J. (15. 15.5^i=i^J e 2 JVA e 2 »i=±l E /wr (15. S = 0B .21) With these identifications. Once again. (15. then we see that we can write TteTH«=e^Z(P). this shows that the quantum fluctuations in a quantum theory can be related to the thermal fluctuations of a classical statistical system. (15.2) in the partition function in a way that is easy to use.2 The Partition Function To evaluate the partition function for the one dimensional Ising model explicitly.
26) Prom this analysis. It is now clear that if we define a matrix operator 2/3J K = efiJ cosh p5 + e" CTi .27) where Ai and A2 are the two eigenvalues of the matrix K in Eq.sinh /?B a3 . (15. (15.24) then. det^ &_pj e/3(J+s)_AJ =0 or.26).2Ae^J cosh/3fi + e2f3J .Ising Model 333 where we have used the identities in Eq. The eigenvalues of the matrix K can be easily obtained from det{K .25) Note that K is a 2 x 2 matrix and has the explicit form 1 /e^J(cosh/35sinh05) e~pJ J e'PJ smhpB) eP (cosh(3B + e/3(JB) &pj \ e/3J eP{J+B) J • (15. A2 . = TrKN (si\K\sN)(sN\K\sNi){s2\K\8l) = \? + A2V. (15. the matrix element of this operator between the eigenstates \si) and SJ+I) of the a3 operator will be obtained using Eq. Si = ±l E e f/lT^N ao^N ( / 3 J E j = l SiSi+l/3#Li=l (si+si+l)\ 2 ^ ) = Y. A 2 2Ae / 3 J cosh/3S + 2sinh2/3J = 0.e~2f>J = 0 or. (15.13). (15. (15.11) to be (si+1\K\si) = e(f}j8iai+1f>B^8i+8i+l)) . (15.28) .XI) = 0 or. it is clear that we can derive the partition function for the one dimensional Ising model explicitly as follows.
we can approximately write Z(j3) = TrK N = Af + A^ Vf JV e0J cosh # B + (e 2/3J sinh2 /35 + e'2pj) 5 . Let us note from Eq.32) .31) that we can write In Z{P) = Nln \epJ cosh pB + (e 2/3J sinh2 0B + e" 2 / ? J )^ = iV PJ + In jcosh pB + (sinh2 pB + e~ 4/3J ) 5 } .e20J + e " 2 ^ ) = eP3 cosh pB ± (e2/3J sinh2 /?5 + e~2fiJ If we identify the two eigenvalues as A: = e?J cosh PB + (e2?J sinh2 0B + e~20J) A2 = e?J cosh PB . (15.334 Field Theory: A Path Integral Approach This is a quadratic equation whose solutions are easily obtained to be A = epj cosh pB ± (e2(3J cosh2 /3B2 = e?Jcosh/35 sinh 2(3J ± (e 2 / 3 J (l + sinh 2 pB) . We can now derive various quantities of thermodynamic interest.31) This method of evaluating the partition function is known as the matrix method and we recognize K as the transfer matrix for the system (see also section 3. we note that since Ai > A2.30) then. (15.29) (15.(e2(3J sinh2 /?£ + e" 2 / 3 J ) (15.3). for large N. (15.
it cannot describe the properties of a magnet.Ising Model 335 Therefore. (14. The magnetic susceptibility for such a system can also be easily calculated (see Eq. (14.33) (sinh 0B + e~^J) * 2 It is interesting to note that when the external magnetic field is switched off. being ordered. These. Namely.23)) and takes the form X ~~ dM ~dB = /3e 2 / 3 J . 1 3F dF 1 I 3d // 1 . (15.34) This shows that for \2{3J\ < 1. (15. the susceptibility obeys Curie's law. In this one dimensional system. there is no spontaneous magnetization and consequently. therefore. the average magnetization per site defined in Eq.22) can now be derived from Eq. we would have expected the configurations where all the spins are "up" or "down" to correspond to minimum energy states. the magnetization vanishes.4 as well as chapter 8) . Let us recall that for the doublewell potential (see section 7. xZ^fcjf (1535) The absence of spontaneous magnetization in the present system may appear puzzling because naively. we would have expected spontaneous magnetization for the system. The lack of magnetization can actually be understood through the instanton calculation which we discussed earlier. in this case.32) to be . NdB~ 1 Nf3 dB NdB\ P dlnZ(P) N (pSmhpB + 1 2/3sinh/3Boosh/3B 1 N y (Sinh2/3B+e4/")3 P cosh 0 5 + (sinh2 /3B + e~^J) * sinh/3B (15..
x+ (15. (15. is a mixture of these two states (the symmetric state) such that (x) true = 0 . in the present case. (15. (7.38) However.37) In this case.53).336 Field Theory: A Path Integral Approach a the naive ground states would give (x) = ±a. In the one dimensional Ising spin system. we showed explicitly that the tunneling or the presence of instanton states contributes significantly leading to the mixing of the states and restoring the symmetry. we can correspondingly think of the following two configurations as denoting the two ground states for which the magnetization is nonzero or (M)p^O. as we have seen earlier in Eq. . TTTIIITTT —two kinks or one instantonantiinstanton. there are other spin configurations such as TTTTIUU .o n e kink or one instanton.36) The true ground state.
obtained in the limit when the lattice spacing goes to zero. let us use periodic boundary conditions along both the axes so that s i = s i\. (15.3 Two Dimensional Ising Model Let us next consider a two dimensional array of spins on a square lattice interacting through nearest neighbors. Even though these configurations have higher energy.N J > i i continuum dx(—(ds(x)) + /ys(x)j + constant.i2 = S h+N. It is worth recalling that in a thermodynamic ensemble. Viewed in this way.41) . of course. it is the free energy which plays the dominant role. (15.i2+N > (15.i2 = S ii. Consequently. 15. spacetime lattices are often used to define a regularized quantum field theory. ( M )  u e = 0. Once again. we can think of the one dimensional Ising model as corresponding to a one dimensional free scalar field theory interacting with a constant external source in the continuum limit. Namely.^ ) 2 + B Y^ « .Ising Model 337 and so on which contribute significantly.40) where a and 7 are two constants. In fact. As we have seen. path integrals are defined by discretizing spacetime variables. they also are more disordered. The continuum theory is.39) This qualitative discussion can actually be made more precise through the use of the path integrals. Namely. they will have a higher entropy and as a result can have a lower free energy.J^(si+i . let us note that H = J^2 i S S J «+I + B^Si i = . The consequence resulting from the contributions of these spin configurations is that the true ensemble average of magnetization vanishes.
^.i2=l (sii.17)) N H = J^SiSj (ij) = J ^ ii. The total number of points on the lattice is then obtained to be n = N2 .i2.43) (15. The spins are assumed to take only the values ± 1 .i2+l) (1544) The symbol (ij) is introduced as a short hand for sites which are nearest neighbors.338 Field Theory: A Path Integral Approach where we are using the notation that i = ( u . (15.44) as being taken over all the links of the lattice.i2+sii.42) The Hamiltonian describing the interaction of the spins is given by (see also Eq.) The partition function for the system described by the Hamilto . ^ ) denotes a point on the two dimensional lattice and we are assuming that N denotes the total number of lattice sites along any axis. (15. \h. (Remember that a link connects two nearest neighbors on a lattice. (14. That is. Si = Silti2 = ± 1 .i2Sii + l.i2Sh. foralHi. We can also think of the sum in Eq.
2).45) where we have defined K = 0J. (15.Ising Model 339 nian in Eq.4 Duality Let us note that since Si = ±1.45) to obtain eKSiSj = coshre+ SiSj sinh re = cosh K (1 + SiSj tanh re) . (15. (15. let us discuss some of the symmetries associated with this system. appears to be only slightly more complicated than that for the one dimensional case in Eq. this partition function is much more difficult to evaluate in closed form. (15.44) can now be defined to be Z(/3) = £ 3i=±l e '* Y" e (" £<«>*«*) Si=±l J2 * i = ± l (ij) *[[eKSiS\ (15. However. as we will see. This partition function.47) we can expand the exponent in the partition function in Eq. (15. 15. as it stands. Before going into the actual evaluation of this partition function.46) Note that we are discussing the simpler case when the spin system is not interacting with an external magnetic field.48) . (15.
50) S J = ± I (ij) Let us next note that the product on the right hand side in Eq.340 Field Theory: A Path Integral Approach Therefore. j . (15. Namely. for nearest neighbors j . J f cosh K (1 + SiSj tanh/t) Si=±l (ij) SiS tann K — (cosh n)2n 2J Si = ±l I I (* + •• (ij) 1 J ) = (coshK) 2n ^2 ]~[y^(gigjtanh/c)' S j = ± l (ij) (=0 1 = (coshK) 2n J2 n E ^ ^ ^ ^ i ) ' si=±l (ij) 1=0 (1549) We see that we can simplify this expression by assigning a number h — hj = Iji = (0.52 ) . we can also write z(P) = E neKsisj Si = ±l (ij) = y .1) to each link between the sites i and j and rewriting Z(P) = (coshK)2n'^2(timhK)h+h+h ^ YlistSj)1* . (15. for nearest neighbors.51) Si=±l (ij) Sj=±l i Si=±l i where we have defined. «i = E j z «• ( 15 .50) can simply be understood as the product of the spins at each lattice site with an exponent corresponding to the sum of the link numbers for links meeting at that site. E II(W 0 = E Ilte) 5 ^ = E nw*14. (15.
mod 2.h and I4 denote the link numbers for four links meeting at a common site. Let us also define a dual variableCTJat each site of the dual lattice and . mod 2 . In other words. if h. (15.Ising Model 341 The sum over the nearest neighbors can now be done to give E Si=±l n(* s .h.55) must satisfy E 3 hj = 0.56) for any four links joining at a site.54 ) Putting everything back in Eq. then h + h + h + k = 0...53) vanishes when n« is odd.50). each plaquette of the dual lattice encloses a given site of the original lattice and intersects the four links originating from that site.55) The constraint here is that the l^s in Eq. (15. (15. (15. (15. k 1 1 2 1 4*" K Thus. It is constructed by placing a lattice site at the center of each plaquette of the original lattice. (15.)' w =n (ij) E ^) n i =n ( x +C^D • (i553) i i Sj=±l It is clear that the expression in Eq. V. we obtain Z(J3) = (2 cosh2 «) n E ( t a n h K)h+l2+= Z(K) . on the other hand. it has the value E Hi***)1* = T • « i = ± l (ij) ( 15 . For even rij.57) Let us next consider the dual lattice associated with our original lattice.
(15.T2) = e " ^ 1 .3. we obtain Z(K) = (2 cosh2 n)n J^ e{2™*+K* £<«> "^ (15. (15.60) ' (2cosh2^ ^ (e 2 «T' l ' .^ ) .0304 . *4 = i(l<74<ri).342 Field Theory: A Path Integral Approach assume that it can take values ± 1 .cr2(J3 . Furthermore. Denoting by (1. '3 = 2 I 1 °"304). we note that for every link that is intersected by a dual link.2.58) We see that each of the Zfc's have the value 0 or 1 as required. mod 2. Going back to the expression for the partition function in Eq.(Tia2 .61) (15.59) In other words. where we have denned tanhK = e .55).2 * \ Substituting this back into Eq. (15. the constraint equation in Eq.58) that (tanh/c)' 1 = (tanhK)^ 1 ' T1. (15.4) the sites of the dual lattice which enclose the point k of the original lattice. (15.57) can be naturally solved through the dual lattice variables. we also have h + h + h + k = z (4 . we can define h = (l02cr3).55). (15. we note using Eq.0401) = 0. 1.
we have seen in Eq. Therefore. (15. e2Kc . if there exists a single phase transition in this model (which was known from general arguments due to Peirels). the partition function cannot be evaluated exactly. (15. In such a case.Ising Model 343 This relation in Eq. Consequently.51) that the temperature can be related to the Planck's constant which in some sense measures the quantum coupling. it must occur at a unique point where or.46))). /?c = ^ l n ( 7 2 + l) • 15.5 High and Low Temperature Expansions Quite often in statistical mechanics. we would like to study the system at very high temperatures as well as at very low temperatures to see if any meaningful conclusion regarding the system can be obtained.62) is quite interesting in that the relation tanh K = e _2re . sinh2rec = 1 or. rec = Jpc=In (y/2 + l ) (15.63) defines a transformation between strong and weak couplings (or high and low temperatures (see Eq. (14. sinh 2rec = 1 or. And we find that the corresponding partition functions are related as well.e~2n° = 2 or. which can also be written as sinh 2resinh 2K* = 1.64) or. high and low . In the language of field theory. (15. e2Kc = y/2 + 1 or.
We have Z(K)= ] T e("E<y> * * ' ) . so is tanh K and the right hand side can be expanded in a power series in tannic. then a bond will connect the sites. (15. then we will not draw a bond connecting the two lattice sites whereas if Ik = 1. Since K is small for high temperatures. Let us go back to the partition function for the 2d Ising model. let us postulate the rule that if Ik = 0. the constraint on the link numbers simply says that there must be an even number of bonds originating from a given lattice site. To do that. Z^'s. mod 2 . we note that the first term on the right hand side of the expansion will correspond to the case where there are no bonds on the lattice. let us note that the link numbers. for any four links meeting at a lattice site. as in Eq.55) that we can write where the link numbers are assumed to satisfy h + h + h + k = 0. then K is small.344 Field Theory: A Path Integral Approach temperature expansions are also known as strong coupling and weak coupling expansions (or approximations). With this rule then. can only take values 0 or 1.46). (15. Consequently. (15.65) Si=±l where. . We have seen in Eq. we have defined H kT If the temperature is high enough. Accordingly.
m:: ^ —> (tanhre) It is not hard to see that such a diagram can be drawn in 2n different ways and hence this term will come with a multiplicity of 2n. The plaquette can be drawn in ndifferent ways on the lattice (recall the periodic boundary condition) and hence this term will come with a multiplicity of n. At the next order the diagrams that will contribute are rrn: — • (tanhre) 5 . the first nontrivial term in the series will correspond to the product of the weight factor tanh re over a single plaquette.Ising Model 345 The next term in the series will be of the form * £ J • • • * * • • — (tanhre)4 In other words. The next term in the series will represent the product of the weight factor tanh re over a plaquette involving two lattice lengths.
dividing the partition function by e2nK (which is the value of the partition function when all the spins are pointing along one direction). Namely. the low temperature expansion will be a measure of how many spins flip as T becomes nonzero but small. the low temperature expansion would merely measure the deviation from such an ordered configuration. Therefore. let us note that when T is small.45) Z K ( ) _ V . J K = fcT' is large.346 Field Theory: A Path Integral Approach (tanh/c)^ (tanlm) 8 .67) To obtain the low temperature expansion. then we will expect all the spins to be frozen along one axis. Thus. The combinatorics can be worked out in a straightforward manner for these graphs so that the high temperature expansionof the partition function will have the form Z(K) (2 cosh2 K)" = 1 + n(tanh«) + 2n(tanh/t) + n(n + 9) (t&nh K)8 + (15. If T — 0.0 («£ w >(i+*«i)) (15. we have from Eq. say up.68) O2TIK =£ =±i . (15.
the two possibilities can be represented as • x x • • —> e _ 1 2 K .Ising Model 347 To develop the right hand side diagrammatically. the first term on the right hand side will correspond to a diagram of the form 1. this leads to two possibilities. The next term in the series will correspond to two flipped spins. Furthermore. Namely. The next term in the series will correspond to the case where one of the spins on the lattice has flipped and will represent a diagram of the form x • • • In other words. let us draw a cross on the lattice to represent a flipped spin. Thus. • x • x • —• Pi6« . Interestingly enough. the term would have a weight e~8K. the flipped spins can be nearest neighbors or they need not be. a single flipped spin will interact with a nearest neighbor with weight e~2K and since there are four nearest neighbors for any site. Diagrammatically. the flipped spin can occur at any lattice site and hence this term will come with a multiplicity of n.
• • • • X X • • x x • • e Thus. for low temperatures. ones where there are three or four flipped spins which are nearest neighbors also contribute the same amount. x x x • • • • • • • x • • x x • —> e" 1 6 K .70) ^ l = i + n e 8«* + 2ne 12K ' + \n{n + 9) e " 16K * + • • • . Furthermore. The multiplicity of the second diagram. re = re* . the interaction between the two nearest neighbor spins which are flipped does not contribute to the partition function. • • • • • • —> e~ 16K . will be \n{n — 5). namely. there is another class of diagrams.348 Field Theory: A Path Integral Approach In other words. obviously. that is not the only kind of diagram which contributes an amount e _16re . (15. In fact.69) It is clear now that if we denote. then.71) . (15. we can write the low temperature expansion also as Z (15. the number of ways a pair of flipped spins can occur as nearest neighbors is In. However. in the first configuration. we can consistently derive a low temperature expansion of the partition function which has the form ^ = 1 + ne8K + 2ne" 12K + \n{n + 9)e" 16K + • • • .
In the present case. (15. (15.73) 2 n 2nK (2 cosh k) e * This is. comparing Eqs.75) . the interaction energy between two adjacent rows. we can write the interaction energy between the spins on a row.2K *. (15.J ^ Si(m)si(m + 1). can be written as N H{m. we see explicitly that the duality mapping (transformation) really takes us from the high temperature expansion to the low temperature expansion and vice versa.6 Quantum Mechanical Model Z Before finding the correspondence of the two dimensional Ising model with a quantum mechanical model. m Then.Ising Model 349 Thus. say m and (m + 1). m + 1) = . m. let us derive the transfer matrix for the system.67) and (15.71) we see that under the mapping tanhK = e. Let us label the sites on a given row by 1 < i < N and the rows by 1 < m < N.62). of course. Let us begin by writing the Hamiltonian for the system in a way that is better suited for our manipulations. the duality relation that we have derived earlier in Eq. (15.74) Similarly. as N H(m) = J^2si(m)si+i(m).72) we have W = ^ > . i=l (15. 15. (15.
76) m=l i=l If we desire..S2.m+l)) N — TT e /3Jsi(m)si(m+l) e /3Jsj(m)s i + i(m) _ i=l (15.SN). let us look at eP(H(rn)+H(rn.s'N\si.78) Let us next note that on every row there are T sites and if we V introduce a two component eigenvector of 03 at every site. (15.s'2..44)) the total energy of the system as N H=Y1 (# M + # (m>m + !)) N N = .. we can also add an external magnetic field at this point.S2... (15. concentrate only on a single factor of this exponent. However. we can write (see Eq. (15..sN) (15. then we can define a 2N dimensional vector space on every row through a direct product as \s(m)) = \s\) <> \s2) 0 • • • \SN) g = \SI. for simplicity. let us ignore it for simplicity.. Namely..80) = $sls'1Ss2s'2SsNs'N .350 Field Theory: A Path Integral Approach Given this.79) We can define an inner product on such states as (s(m + l)s(m)) = (s'1.. The partition function will involve the exponent e~m = e (/ ? E^ = i(ffM+^Krn+i))) _ (15....77) Let us.Y1 J ] C {si(m)si+i(m) + Si(m)si(m + 1)).
g 8 8 a3(i) = l ® l ® . (15. W i t h these preliminaries.9 + aaisinh9\si) (\ZQA\ ci!sinh#\ a s i n h # a cosh 9 J ' where a is a constant. we note t h a t a t e r m such as e/3Jsi(m)si(m+l) can also b e written as a 2 x 2 matrix of the form S /3J e0J ^  1 %J \ J )• (1585) Therefore. we note t h a t we can identify (s'i\aee^\si) = e0Js^.. we can write {s'j\ae ai \si) = (s^acosh9 acosh. t h e completeness relation will take t h e form £ \s(m))(s(m)\ = l. therefore. (15.81) where 1 denotes t h e 2 ^ x 2N identity matrix.84) and (15.) <Ti(i) = 1 <> 1 <> • • • <> <ri ® 1 • • • ® 1. comparing the two relations in Eqs. (15.. On the other hand.85). (15.Ising Model 351 Similarly. as a 2 x 2 matrix.83) We also note t h a t t h e &\ in t h e i t h place acts on t h e vectors \si) and. all t h e entries in t h e above expression correspond t o the trivial 2 x 2 identity matrix except at t h e i t h entry.82) Namely. (There will b e iV of each.< g > c r 3 ( g ) 1 • • • (8) 1 . namely. Let us also record here t h e product formula for matrices defined through direct products. (15. (A®B){C®D) = AC®BD.86) . let us now introduce t h e following 2 ^ x 2 ^ matrices.
90) (15 91) (15. asinhe = e~PJ.88) (15.352 Field Theory: A Path Integral Approach provided the following relations are true.87) Consequently.92) . a cosh 6 = e^ 7 . with tanh 0 = e~WJ .0\s{m)) — {s(m + l)\s(m)) (cosh# + Si + iSjsinh#) = (s(m + l )  s ( m ) ) e ^ + l S i ) = (s{m + l)\s{m))ePJSiSi+\ = e^J(Ei»i(m)«i(m+i)) _ (15. Equivalently. it is clear that if we define a 2 ^ x 2 ^ matrix as = (2sinh2/?J)^ e f c ^ i W ) .89) (15.2 / 3 J . (15. we can write (s(m + l)\Ki\s(m)) Let us also note that (s(m + l )  e ( ^ 3 ( m ) a 3 ( i ) )  s ( m ) ) = (s(m + l)cosh# + oz{i + l)a^{i) smh. then. we can make the above identification provided tanh# = e. namely. a = (2sinh2/3J)5.
to . Si=±l = TrKN.95) which defines the transfer matrix for the system. in field theory..97) = Y. therefore.. In field theory language.93) (15. we are looking for a quantum Hamiltonian whose Euclidean transition amplitude will yield the partition function.96) = e/3(H(m)+H(m.m+l)) The partition function can now be written as Z((3) = £ Si = ± l e"^ (s(l)\K\s(N))(s(N)\K\s(Nl)). is the starting point for the Onsager solution of the two dimensional Ising model. Namely.(s(2)\K\s(l)) (15. Furthermore. (s(m + l)\K\s(m)) = (s(m + l ^ i ^ ^ m ) ) = (s(m + l) J FsTis(m))e / 3 J S£i^+i = eP J lli'=isi(. even if we are dealing with a theory on the lattice. let us identify one of the axes. This. say the vertical one.i+V°*W). defining K2 = e?J<£?=i°*(. Thus. Thus.Ising Model 353 provided we make the identification 6 = /3J. we would prefer the time variable to be continuous. we note that we can write K = KXK2 = ( 2 s i n h 2 / 5 J ) ^ e 0 S £ i ' T l ( i ) x e ^E l 1 1 3(i+i)a 3 W ) (15.m)si(m+'i)e0J12iLi Si(m)si+i(m) _ (15.94) (15.
m + 1) = —J' ^2 sisi • (15. be different. we have actually destroyed the isotropy of the system and. Let us note that we are only changing the spacing among > the rows. allowing for different couplings along the two axes. in principle. consequently. The spacing along a row. is unchanged.98 ) where N J^SiSi+i. a very small quantity. then the correlation lengths become quite large in this limit and in such a limit the lattice structure becomes quite irrelevant. But. of course. let us recall that if there is a critical point in the theory. Let us also note here that by making the lattice asymmetric. this may appear bothersome. of course. At first sight.354 Field Theory: A Path Integral Approach correspond to time and we choose the separation between the rows to be e.99) . • • • • • • The continuum time limit. we can write JV N H = ^ m=l N ^2 (J'si(m)Si(m i=l + 1)  Jsi(m)si+i(m)) = Y^ (H(m) + H(m>m + !))» m=\ ( 15 . the couplings along different axes can. N H(m) = H(m. will be obtained by choosing e — 0. Thus.
102). we immediately conclude that H i = ~ X > 1 W + Xa^ i + ^(O)> (15.104) where A is a constant parameter and as before.100) Sj=±l m = l In the quantum field theory language. we recognize that the two can be identified if {s(m + l)\eeH"\s(m)) = e^it^s.100) and (15.102) S j = ± l m=l Thus. we can identify tanhe~e = e ~ w ' . (15.Ising Model 355 The partition function. we can write the Euclidean time interval as TEUCI.m+l)) N = ^ JJeC^Si^+i+^'^i^O. = Ne.95) and (15. will have the form ZIP) = £ *~m Si=±l = £e Si=±l /3E^=i(H(m)+H(m. (15.101) and assume that there exists a quantum Hamiltonian Hq such that we can write Z(J3) = Tre~ TBuc '.103) From our discussion of the transfer matrix in Eqs.) _ (15. (15. (15.96).ff « = Tre _JVeH « = J2 N m\eeH<\s(N))(s(2)\e*H<\s(l)) = £ Yl(s(m + l)\eeH*\s{m)). in this case. comparing Eqs.105) . (15. (15.+i+A*.
the corresponding coupling between the rows becomes stronger. This is renormalization group behavior of the couplings in its crudest form.J = / • (15.106) • • • • • • • • • original lattice N • • • • • dual lattice 1 2 3 Let us next consider the dual lattice corresponding to this one dimensional lattice and define the dual operators on the dual lattice as M i W = 0 3 ( « + 1 ) 0 .108) .107) It is easy to see that /x?(z) = (a3(i + l W i ) ) 2 = J.356 Field Theory: A Path Integral Approach This relation is quite interesting in the sense that it brings out a relationship between the coupling strength as a function of the lattice spacing. i (15.3 (*). In particular. M ! « = (n<r.7 Duality in the Quantum System We have been able to relate the 2d Ising model to a one dimensional quantum mechanical system with a Hamiltonian N H i = ~ I ] M * ) + x^{i i=l + l)<T3(i)) • (15. 15. we note that as we make the spacing between the rows smaller.
110) On the other hand. al{i) = I = al(i). then.111) Thus.(ri(j)]= ° = [ff3(*). \lii(i). JJ<7i(j) il [/*l(*)»A»3(*)]+ = = 03 (* + 1) YlW3(i).Ising Model 357 These results can be shown to follow from the basic commutation relations of the Pauli matrices. Wi(i). H3(i + 1)/X3(«) = cri(i + 1). let us note that by definition. k i ( i ) . Furthermore.<ri(i)]+(riU) = 0. ^3(j + l)"3(j)] = 0. Using these. namely.109) . cr3(i + l)(J3(i). we can also derive that for i ^ j .A*3(j)] = (15.HiU)] = ka(i + 1)^3(0.ff3(j)l. (15.fe=i n^wii^w i=i i 3 = 0. //i(i) and //3(i) also have the same algebraic properties as the Pauli matrices on the original lattice. we note that we can write N i=l AT (15. a 3 ( i ) ] + = 0.2 ( M 3 ( « + l)A«3(i) + Vi(*)) i=l .112) = . [A»3(*). if * ^ 3. (15. Using these.
"Statistical Mechanics". R.. 52. Rev. Rev. Savit. then the above duality relation implies that this must happen at A = 1.. 453 (1980). J. (15.64). John Wiley Publishing. Phys. K.114) For some finite value of A. Mod.. Mod. (15.358 Field Theory: A Path Integral Approach N = \J2 (fn{i) + AVa(* + 1)M3(*)) = \Hq(\1). 659 (1979). that the energy eigenvalues of this system must also satisfy the relation E(\) = \E(\1). Kogut. Phys. if there is a phase transition such that the correlation lengths become infinite or that some energy eigenvalue becomes zero (zero mode). Namely. This shows. 51. (15. it maps the strong coupling properties of the system to its weak coupling properties. Let us also note here that since we have a quantum mechanical description of the 2d Ising model...113) This is the self duality relation for this system. . 15. in particular.8 References Huang. This is precisely how we had determined the critical temperature for the 2d Ising model in Eq. we can also develop a perturbation theory in the standard manner.
124 Classical statistical system. 6. 134 Dual lattice. 323 Classical trajectory. 243 Fermionic oscillator. 356 Effective action. 67 Coulomb gauge. 28. 311 Euclidean action. 327 Euclidean rotation. 15 Basis states. 8. 184 Continuous transformation. 49. 279 Classical field. 266 Chiral anomaly. 184 Connected Green's function. 32. 53 Euclidean space. 71 Anomalous Ward identity. 335 Doublewell potential. 130 Complex scalar field. 40 Coordinate basis. 241 Critical exponent. 199 Entropy. 160 BakerCampbellHausdorff formula. 143 EulerLagrange equation. 45 FaddeevPopov determinant. 143 Double well potential. 339 Duality in quantum systems. 76 FermiFeynman gauge. 76 Antiinstanton. 123. 240 Anharmonic oscillator. 322 Asymptotic equation. 11 BetheSalpeter equation. 260 FermiDirac statistics. 193 Effective potential. 67. 341 Duality. 279 Anticommutation relation. 143 Euclidean equation. 218 Connected diagram. 324 . 75.Index Abelian gauge theory. 59. 321 Euclidean generating functional. 61. 187. 9 Correlation function. 191 Born diagram. 45. 300 BRST invariance. 147 Antiperiodic boundary condition. 193 Bosonization. 187. 212 359 Continuum limit. 314 Double well. 279 Anomaly. 32. 21 Classical phase. 75. 188 Classical path. 58. 144 Euclidean field theory.
12 Imaginary time. 324. 2 Midpoint prescription. 311. 186 Operator ordering. 235 Grassmann variable. 332 Pauli principle. 159 Left derivative. 235 Goldstone mode. 13. 47. 77 Periodic boundary condition. 174. 143 Interaction picture. 111 Harmonic oscillator. 111. 209 Ising model. 45. 327 Regularization. 337 Isospectral. 9 Invariance.360 Field Theory: A Path Integral Approach Ising model two dimensional. 19 Feynman propagator. 239 Gaussian integral. 309 Fourier series. 17. 86 legendre transformation. 14 One loop effective potential. 113 Regularized superpotential. 25. 147 Instantons. 105 Noether's theorem. 318 Heisenberg picture. 313 Proper self energy. 8 Heisenberg states. 3 Functional integral. 212 Goldstone particle. 55. 196 Quantum Quantum Quantum Quantum Quantum chromodynamics. 200 Many degrees of freedom. 53. 46 Free energy. 143 Instanton. 179 Feynman rules. 25 Gauge choice. 31. 178 Global transformation. 312. 107 Jost function. 254 GuptaBleuler quantization. 115 . 3 Functional derivative. 72 Perturbative expansion. 321 Feynman diagram. 78 Green's function. 171 Feynman path integral. 167 Maxwell theory. 200 Field. 69. 212 Loop expansion. 181 Phase transformation. 181. 58. 219 Phase transition. 16 Multiinstanton. 349 statistical system. 179 fluctuation. 246 Normal ordering. 203 One particle irreducible. 183 Feynman Green's function. 91. 163 Nicolai map. 36 Functional. 34 Fourier transform. 33 mechanical model. 241 Gauge theories. 176 Gribov ambiguity. 82 Generating functional. 70. 280 correction. 79. 14 Partition function. 212 NonAbelian gauge theory. 234 Goldstone theorem. 239 Metric. 245 Half oscillator. 327 Ising model one dimensional. 86. 89. 325 Free particle. 12. 64. 210 Local transformation. 1 Finite temperature. 322 Perturbation theory. 148.
176 Time ordering. 170 Schrodinger equation. 149 Zero mode. 7. 125. 193 Vacuum functional. 148. 126.Index Relativistic field theory. 134. 209 361 Temperature expansion. 127. 113 Supersymmetric Oscillator. 232 Steepest descent. 169 Vacuum generating functional. 130. 85. 148. 12 Schwinger model. 134 Stirling's approximation. 346 Time evolution operator. 7 Time ordered product. 102 Supersymmetry. 334 Transition amplitude. 16. 138 Zero eigenvalue. 125. 134 WKB wave function. 130. 274 Weyl ordering. 144 Shape invariance. 121. 36. 88. 346 Temperature expansion low. 51. 131 Tree diagram. 97 Supplementary condition. 335 Spontaneous symmetry breaking. 121. 86 Saddle point method. 67. 343 Temperature expansion high. 245 Symmetries. 170 Right derivative. 129 Stress tensor. 222 Ward identity. 14 WKB approximation. 105 Singular potentials. 154 Transfer matrix. 111 SlavnovTaylor identities. 130. 316. 8 Schrodinger states. 215 Spontaneous magnetization. 121 Schrodinger picture. 289 SchwingerDyson equation. 152 Scalar field theory. 150. 226. 41. 9. 79. 158. 170 Ward identities. 131. 97 Supersymmetric quantum mechanics. 64. 125 Semiclassical methods. 224 Spacetime translation. 358 . 60. 128. 217 Superpotential. 90 Time translation invariance. 191 Semiclassical approximation. 32.
the subject matter is primarily developed within the context of quantum mechanics before going into specialized areas. With its utility in a variety of fields in physics. Two extra chapters cover path integral quantization of gauge theories and anomalies. this second edition is an important expansion of the popular first edition. and a new section extends the supersymmetry chapter.2 o o 6 www. Adding new material keenly requested by readers. 6145 he ISBN 9812568476 v n « > D I PUBLISHlNi brld Scientific 9 789812 568472 i .com .w FIELD THEORY A Path Integral Approach Second Edition This unique book describes quantum field theory completely within the context of path integrals.worldscientific. where singular potentials in supersymmetric systems are described.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.