RICHARD A. BRANDT Center for Theoretical Physics, Department of Physics and Astronomy, University of Maryland, College Park, Maryland 20740 A formulation of quantum electrodynamics based on finite local field equations is employed in order to prove and discuss the gauge invariance of the theory in a meaningful and rigorous way. The Dirac and Maxwell equations have the usual forms except that the current operators f(x) and iM(x) are explicitly expressed as finite local limits of sums of nonlocal field products and suitable subtraction terms. The electric current, for example, involves the terms A(x) and : A+):. The field equations are used to derive renormalized Dyson-Schwinger-type integral equations for the renormalized proper part functions JY,II@“, Au, and X”@ (the four-photon vertex function), etc.Application of the boundary conditions E(b = m) = Z’(fi = m) = II(O) = n’(O) = U”(O) = LI($ = m, 0) = X(0,0,0,0) = 0 is shown to completely specify the current operators. It is shown that the theory is gauge invariant in the sense that the divergence conditions k,W”(k) = koX**(k,...) = 0, etc. and the generalized Ward identities k,,N(p, k) = eZ(p - k) - e-J?(p), etc. are all satisfied in each order or perturbation theory. This is shown to be equivalent to the invariance of the field equations under gauge transformations of the second kind. I. INTRODUCTION
The purpose of this paper1 is to use a formulation of quantum electrodynamics based on finite local field equations2 in order to rigorously prove the gauge invariance of the theory and to show that this gauge invariance is equivalent to the invariance of the field equations under local gauge transformations. The field equations will have the usual forms8
l$P”(x)
= j”(x),
* Supported in part by the U. S. Air Force Office of Scientific Research under Grant AFOSR 68-1453. 1 This paper is taken from Ref. (I). Other parts of (I) will be contained in another paper (2). We shall refer to Ref. (2) as B throughout this work. 2 Such equations were derived in perturbation theory in (21). An independent derivation was known to W. Zimmermann at an earlier date. For recent investigations and applications, see references (3)-(6), where earlier references can be found. We shall refer to (3) as A in this paper. s We use the notations and conventions of Bjorken and Drell (7); e.g., c = ti = 1, k - x = kpx,, = kOxO- k . x. Wewritei = y ‘p = yfipp, 0 = a* = a02 - v*, (a/ax,) F(x) = Lw(x) = F’fi(x). Three-vectors will be denoted by bold-face letters.
122
GAUGE
INVARIANCE
123
except that the current operators f(x) and j”(x) will be finite limits of nonlocal expressions. Thus (1.3) where the operator q”(x; 0 has singularities at f = 0 which compensate those of the local product q(x) yU$(x), and similarly for p(x; 5). Gauge invariance in classical electrodynamics means an invariance under the transformations #(x) + e-iea(z)#(x), i&x) -+ eiea(r)#(x), (1.4) AL(x) + Au(x) + 4L44 for arbitrary smooth functions a(x). Here A, is the classical vector potential and + a Schrijdinger wavefunction (matter field). In the usual formulation of quantum electrodynamics, a canonical proof of gauge invariance in this sense has only formal significance because the field equation for the quantized field A, involves the meaningless product $(x) rub(x) of quantum fields at, the same point. In practice the requirement of gauge invariance has been replaced by the requirement that various generalized divergence conditions are satisfied. The simplest of these are the identity kd7,,,(k) = 0 (1.5) and the generalized Ward identity ku/l,(p,
k) = eZ((p - k) - eZ(p).
(1.6)
By the gauge invariance of electrodynamics we shall mean that all these conditions are satisfied. In this paper, on the other hand, one has well-defined field equations for A, and I,L and hence can return to the original form of gauge invariance and discuss its consequences for renormalized electrodynamics. In particular, the requirement that the field equations (1.1) and (1.2) be invariant under (1.4) implies nontrivial restrictions on the subtractions qu(x; .$) and p(x; t) used in definingj@(x) andf(x), especially as the product #(x) y,$(x + .$) is not gauge invariant for nonzero f. We shall show, in fact, that the invariance of (1.1) and (1.2) under (1.4) is equivalent to the requirement that all of the generalized divergence conditions be satisfied. The analysis of the requirement of gauge invariance on the subtraction functions q”(x; 5) andp(x; 8) is given in Section VI, Subsections E to H. A full understanding of this problem requires a detailed knowledge of the general, behavior of qU and p.
124
BRANDT
To determine this, in previous sections the renormalized field equations are related to integral equations for Green’s functions and the subtraction functions are determined from conventional renormalization conditions. A preliminary study of Subsections VI.E-H will, however, show what is involved. In order to see what the problem is, let us consider the usual formulation of quantum electrodynamics based on the “unrenormalized” field equations
Calculations with these equations lead to numerous divergences which can, however, by careful use of covariance properties, be isolated into infinite but unobservable mass, charge, and field operator renormalization. For example, for the second order polarization tensor one obtains
where M is a large cutoff mass and U$“(k) is finite for M + co. Now (1.5) requires that n@“(O) = 0 and other requirements of renormalization theory demand that the first two derivatives of the renormalized nuy(k) vanish at k = 0. Thus one must renormalize (1.9) by subtraction at k = 0. The resultant expression is finite and satisfies n““(O) = 0 but is not guaranteed to satisfy (1.5). In this paper we shall in fact prove that the divergence conditions are satisfied in every order of renormalized perturbation theory. Previous proofs are all of a formal nature and do not guarantee that the appropriate identities will hold in perturbation theory. The local field equations (l.l), (1.2) form an ideal setting for a meaningful discussion of gauge invariance. All calculations proceed directly and mathematically from (1.1) and (1.2) without renormalization or regularization. Thus if (1.1) and (1.2) are gauge invariant, the divergence conditions should be satisfied. Note that the expression lim,,, X,&(X)yU#(x + 0 is not invariant under (1.6). It behaves like [-3 for f - 0 and so, after (1.4) is applied, terms like 511Q(x) contribute. They must be cancclled by the q” term in (1.3). In contrast, in the usual formalism based on the divergent equations (1.7), (1.8) and the ad hoc renormalization rules, gauge invariance cannot even be discussed in a meaningful way. Equations (1.7) and (1.8) are not really gauge invariant and, in any case, the need for renormalizations means that they do not involve the correct current operators for the theory. In this approach, gauge invariance is usually maintained by simply discarding non-gauge-invariant quantities or by using the special Pauli-Villars regularization. These procedures cannot, however,
GAUGE
125
INVARIANCE
be formulated in a local way (or even in a simple way) and so they make gauge invariance appear to be rather artificial. With gauge-invariant finite local field equations, on the other hand, gauge invariance is implemented from the beginning in a natural and mathematically precise manner. In Section II we describe how to calculate renormalized Green’s functions in quantum electrodynamics. The Feynman rules and elegant Bogoliubov-ParasiukHepp renormalization prescriptions are outlined. We show how the subtraction points are specified by the requirement that the parameters have their physical values and indicate the restrictions imposed by gauge invariance. Our field equations and the general forms of our current operators are exhibited in Section III. In Section IV we define numerous Green’s functions and proper part functions and then use the field equations and current definitions to derive integral equations relating them. (Most of the details are omitted. They are either identical to those in A or are given in B.) By imposing the conventional boundary conditions on these equations, we completely specify the current operators. (We show in B that our integral equations are all finite and self-consistent by deriving them from renormalization theory. It follows that the iterative solutions of our equations are equivalent to conventional perturbation theory and hence that quantum electrodynamics can be based on finite field equations.) In Section V we show that our theory, and hence conventional theory, satisfies the requirements of gauge invariance. In particular, we derive many generalized Ward identities and other divergence conditions. The arguments (given in the Appendix) are quite involved and indicate the subtleties of gauge invariance. We find in Section VI, moreover, that, given the general forms for the current operators, the gauge invariance of the theory is equivalent to the gauge invariance of the field equations-including the gauge invariance of the electric current operator. II. GENERAL
PROPERTIES
OF QUANTUM
ELECTRODYNAMICS
Our purpose in this section is to summarize the means by which calculations in quantum electrodynamics are performed. This theory describes the Lorentzinvariant interaction of electrons (spinor particles of mass m and charge -e), positrons (spinor particles of mass m and charge +e), and photons (massless and chargeless vector particles) according to the (formal) interaction Hamiltonian density 4(x> = e&4 r?W 4dx). Gw * When necessary to avoid infrared difficulties, we can assume that the photon has a small but finite mass.
126
BRANDT
We shall first describe the general calculational procedure and then describe the particular procedure which must be invoked in order that the results describe the particles encountered in the laboratory. This particular procedure is determined by the imposition of the requirements dictated by a specification of the operational definitions of the parameters M, e, and by gauge invariance and charge-conjugation invariance. A. RENORMALIZATIONTHEORY We shall be concerned exclusively with the Green’s functions5
The S-matrix elements can be obtained from these by means of the usual reduction formula (8). A suitable formal starting point for our discussion is the Gell-Mann-Low (9) perturbation expansion, the various terms in which can be represented by Feynman diagrams in the usual way. An arbitrary &h-order Feynman diagram G(V, ,..., V, , 2) connecting the vertices V, ,..., V, with the lines 2 = {fI . .. 8S> corresponds in momentum space to an unrenormalized function6 (2.3)
Here pc is the momentum corresponding to the line 8 and a, represents its discrete spin or polarization indices. Thus if /is an electron line we have
and if /is a photon line we have (2.5) 5 Throughout
this work we denote vacuum expectation values of time-ordered products simply