PHYSICAL REVIEW VOLUME 80, NUMBER 3 N O V E M B ER 1, 1950

Non-Linear Interactions between Electromagnetic Fields

ROBERT KARPLUS*^ AND MAURICE NEUMANJ
Institute for Advanced Study, Princeton, New Jersey
(Received June 12, 1950)

The covariant 5-matrix formalism of Dyson has been applied to the calculation of the fourth order
non-linear polarization of the vacuum, which is related to the lowest order non-linear interaction between
electromagnetic fields. The finiteness and the gauge invariance of the interaction are exhibited explicitly by
an expression for the fourth-rank vacuum polarization tensor in momentum space.

I. INTRODUCTION scalar non-local operator. The Fourier transforms of

I T has long been recognized that higher order cor-
rections in quantum electrodynamics include non-
linear interactions between electromagneticfields.1They
arise from the polarizability of the vacuum, from the
possibility that transitions involve pairs only in inter-
mediate states. Since such a correction depends on the
operators of the electromagnetic field alone, it can be
thought of as an addition of fourth degree in the field
strengths to the electromagnetic Lagrangian density
or as a non-diagonal contribution to the scattering
matrix between matter-free states. The scattering of
light by light, which has received some treatment in the
literature,2"5 is an example of a process which can be
described by a specific interaction between photons.
Since the corrections we are discussing are neces-
sarily at least of the order e4, their calculation has
involved considerable complications both because the
treatment of effects involving virtual pairs has been
traditionally accompanied by divergence and gauge-
invariance difficulties and because the expressions en-
countered were lengthy and tedious to manipulate.
With the promise the recent developments in quantum
electrodynamics6"-9 give of eliminating the former and
reducing the latter of these obstacles, it seemed worth
while to re-examine the problem in spite of the smallness
of the effects and the consequent difficulties attending
their experimental detection. We have therefore rear-
ranged the appropriate portion of the fourth order cor-
these are given as integrals over three parameters of
rational functions of the momentum variables. The com-
putation of a cross section requires only the well-known
manipulations of the appropriate element of the scat-
tering matrix in addition to the evaluation of the above-
mentioned integrals for those values of the momenta
which are of interest. The length of the expressions
involved, however, makes the calculation of cross-
sections very tedious except for simple special cases.
A quantity which plays an important role in the cal-
culation is the vacuum polarization tensor G^x^ of
fourth rank,
S/M(*)= ~(a2/12h) f GM,x<r(s, *', *", x'")Av(xf)
XAx{x")Aa{x'")dx'dx"dxf", (1)
where bj^x) is that part of the current induced in the
vacuum which is intrinsically cubic in the potential,
and which cannot be reduced to lower order effects.
This tensor, it will be shown, is finite and divergenceless
with respect to all indices,
^Hv\a\X, X , X , X )U.
dxa
(2)
JfAv\ff \**', .v , wv , v J " v / , CLC

rection to the 5-matrix to display explicitly its finiteness

and gauge-invariance. To this end it has been expressed It will appear further that the tensor depends only on
as a sum of terms each of which is a scalar product of the mass of the pair field and on the nature of the
derivatives of field strengths multiplied by a finite coupling between the pair and vector fields. Hence Eq.
(1) is valid even when A^x) refers to a neutral vector
* Frank B. Jewett Fellow.
t Now at Harvard University, Cambridge, Massachusetts. meson fields, coupled vectorially to the pair field. The
I Now at Brookhaven National Laboratory, Upton, Long effective interaction Lagrangian density, L(x)y and
Island, New York.
1 the contribution to the scattering matrix, 5 (4) , are
For a summary of the literature on this subject the reader is
referred to A. Pais, "Developments in the Theory of the Electron" simply related to the polarization tensor:
(Institute for Advanced Study and Princeton University, 1948),
pp. 21-26.
2 0 . Haipern, Phys. Rev. 44, 855 (1934). 5(4) = [ d*xL(x)
3
E Euler, Ann. d. Physik, 26, 398 (1936). he J
4
W. Heisenberg and H. Euler, Zeits. f. Physik 98, 714 (1936).
5
A. Achieser, Physik Zeits. Sowjetunion 11, 263 (1937).
6
S. Tomonaga, Prog. Theor. Phys. 1, 27 (1946) and subsequent -ia2 1
publications with co-authors. *JHv\a\Xj X j X , X )J\.n\X)Jxv\X )
' J . Schwinger, Phys. Rev. 73, 416 (1948); 74, 1439 (1948). (hCy /
8
R . P. Feynman, Phys. Rev. 76, 749, 769 (1949).
9
F . J. Dyson, Phys. Rev. 75, 1736 (1949). XAx(x")A.(x'")dxdx'dx,'dx?". (!')
380
 NON-LINEAR INTERACTIONS 381

II. THE POLARIZATION TENSOR to zero through positive values. K ^ W ^ is equivalent

The fourth-order term of the electrodynamic 5-matrix
which describes the non-linear effects in which we are
interested is given by the Feynman diagram (Fig. 1)
or by the integral over four four-spaces9
to rMx<rU) in the integrand of Eq. (6) because of the
symmetry of the remaining factors with respect to
simultaneous permutations of ka\ km * and ju, vy .
GMx<r(*}, oi course, is completely symmetric with respect
to these operations. Furthermore,

5(4) = _ (i/64){e/hcY f dtxx&xtfUx^xi

= G M , X , W (- (1 \ - ( 2 \ ~ (3) ? - ( 4 ) ), (9)
XA^(xi)Av(x2)A\(xz)A9(x^)
because the trace of the spinor product of an odd
XTr{ytiSF(oc2Xi)yvSF(xz~X2)y\ number of Dirac matrices vanishes.
At this stage the quantities 5 (4) and G^x^ must be
XSF(xAxz)y<TSF(xi-Xi)}. (3)
defined more precisely because they depend on the
An(x) is the vector potential of a photon and/or a fixed logarithmically divergent tensor rMVxff(K). This is accom-
10 u
electromagnetic field; SF(X) is the function characteristic plished by regularization : -
of the fluctuations in the pair field?9 fxvXcr J- iiv\a *- fivXar (10)
2i r i(yp) K
SF(*)=- lim I d*p e and, correspondingly,
4 2 2
(2TT) -*<>+ J p +K -ie GpvXa=lim^G^x/^ ~ G^xa(M) ],
M> ao
pX=p'T~poXo>
( --rnc/h
(4)
S(4) = -a 2 f J4# f d^d^Wk^dWfe(4)
The transition to momentum space, with
Xexp[f(* (1) +* (2) +* (3) +* (4) )^M(* (1> )

i4M() = (Ac)* fi4M(ft)cited4ft, (5)

X^( ( 2 ) Mx( ( 3 ^,(& ( 4 ) )

yields X G ^ ( ^ ( 1 ) ;fe<2),ft<8\(4)). (11)

To verify the consistency of the theory, the finiteness
S<4> = -a2 fd*x fdWdWVdW'dW* and gauge-invariance of 5 (4) , one may observe that the
polarization tensor G^xa is finite and that it satisfies
the Fourier transform of Eq. (2),
Xexp[*(*> +<2> +*< +k^)x2A(k^)Av(k^)
1 V 1 ) GMX,(* ( 1 ) ,* ( 2 , ,* ( 8 ) ,* ( 4 ) ) = 0,
X ^ x ^ ^ ^ O K W ' ' ^ ' , *w,fe<,#), (6) (12)
where fc%U*(1), *(2), (3)> &(4)) = 0, etc.

= 2W>(* ( 1 > , &(2), (3>, ^ 4 ) ) + ^ , x ( ^ , *<, *<, *<)

+ z w > ( f c < 1 \ *<3), # , (4)) (7)
and
r ^ x . w ( * ( , *<, &<3), * (4) )
iypK iy(pki2)) K
1 d4pTr 7M~ : ~ y v
- / p2 + K2-ie (p~kW)2 + K2-ie
iir2J
iy(p-kW-kM)-K
Xyx~
(p-~k^-k^)2+K2~ie
iy(p-~k^-k^-k^)-K
Xyff . (8) FIG. 1. The Feynman diagram.
(/>-*< >-*< >-# >) + K2-i
2 8 4 2
10
W. Pauli and F. Villars, Rev. Mod. Phys. 21, 434 (1949).
11
Alternatively, the ambiguous integrals can be denned by
It will be understood from now on that integrals which demanding that 5 ( 4 ) be gauge-invariant. Equation (2) would
depend on e are to be evaluated in the limit as e tends then become a defining equation for Guvx<r-
382 R. KARPLUS AND M. NEUMAN

Because GM^x<r is symmetric these four equations are the resulting terms cancel in sets of four. A characteristic
actually equivalent to one another. group is
That GMx<r is indeed finite may be seen more readily
by separating the divergent terms in T^x/*0 and iypK iy{pk(-2)) K
(1/V) fd*pT<>
r{ yv 7x
T * (**>
* nv\<r p2+K2-i* (p-k^)2+K2-h
1 fivXa^ i- nv\<j T L ^ HV\<T J- fxv\a J- fivXcr J } \1^/ iy(p-k^-k^)-K
X y0 [same with K>Ml
where
(p-k^-k^)2+K2-ie
vypK iypK iypK iy(p-\-k(s)) K
- ftvXiT = fd'pTr\
2 7M r^" : -y\ ; 2 7v
iir J I 2 2
(p+k^) +K2-ie
p2+K2ie p2 + K2ie p + K -ie
iy(p+k^+k^)-K
2 2 2j 2 X 7.
P +K ie p rK ie
(p+kW+kW)2 + K2-ie
iypM iypM + [same with KM~] = 0. (17)
-y~rzr^i
p +M2-ie
2 2
p +M -ie 2

These four terms come from 2nMx</'c)(1234),

iypM iypM
XYX~ 2 (14) -2WM)U234), T ^ i - 1 , - 3 , - 2 , -4),
2 2 2
p +M -ie p +M -ie (M)
and - r M x , r ( - l , - 3 , - 2 , - 4 ) in that order. (For
The quantity in brackets in Eq. (13) clearly approaches simplicity, the k{i) have been replaced by i in the
a finite limit as M becomes infinite. The tensor Tn^ arguments of the tensors.) The displacement p^^pn
is easily evaluated to be12 &M (2) ~~ ^ M ( 3 ) m ^ e third and fourth makes them equal
to the first and second, respectively, except for sign.

Tpj,\<r =2 I dAp\ (dMSxr + M(Ax--2i5Mx5ff) III. METHOD OF CALCULATION

iw J I
From Eqs. (7), (8) and (11) it is evident that the
r(V3)(2)2+4/c22 (4/3)( 2 ) 2 +4M 2 2 polarization tensor can be written
X
L [p2+K2-ul* [p2+M2-Uj ] GMx<r(1234)= E
1=2,3,4

1=1,2,4
A^'lm(1234)k^kv^h(l)k/m>
y = l, 3 , 4 w = l, 2 , 3

+ E 5i Im (1234)Mx (l) *. (m)

M4 J=l,2,4
(15) w = l , 2,3

+ E 52^(1234)l,x*/*^> + -
J =1,3, 4
and so does not contribute to G^u because one term m*=l, 2 , 3
vanishes on symmetrization and the other on integra-
tion over the momentum p. +CI(1234)MX+C2(1234)5MX^
One can verify Eq. (12), now, by the use of Eqs. (7), +C,(1234)M,5,x, (18)
(8), (9), and (11), if he notes that
where the ^4, JB, C are invariants which depend on scalar
1 1 products of the four momenta. An expression of this
k^)y^1liy(p^k^-k^-k^)^K2-~Jjyp + K2 form is obtained if the spinor summation and integra-
(16) tion over pp, Eq. (8), are carried out. Because the four
i i
momenta are connected by the conservation equation,
1 1 only three are independent; to maintain a symmetrical
appearance of GMx<r, the three were chosen in a way
12 i i which is dependent on the vector index they carry, as
With the aid of the identities
illustrated by the restrictions on the summations in
(i) fpaf(P2)d*p=fpapepyf(p*)d*p=o, Eq. (18). The problem of obtaining an explicit ex-
pression for G>x<r, and so 5 ( 4 ) , can now be solved in a
(2) fpPfifW*p~l8af}fp*f(p*)d%
straight forward manner by obtaining the coefficients
(3) fpaPfiPyPffW^P^-^afidy^+Bayi^+Ba^fiyl in Eq. (18) from Eqs. (7) to (11). This task is not as
extensive as it seems, because the symmetry of the
polarization tensor means that many invariants can be
 NON-LINEAR INTERACTIONS 383

obtained from others by merely interchanging some of /?^(l)F^(2)F7a(3)Fa(4)

the momenta. Since GM*xr(&(1), k(2\ k(z\ k(i)) is invariant
under simultaneous permutations of the arguments and = ,4 M (l)^,(2),l x (3M,(4)^x, (2) (1234)
indices, the following relations hold among the A's:
gM,xa(2)(1234)~*/Jlj/8>*x(4)*,<1)
yi2111(1234) = ^4441(2341) = ^ 3343 (3412)=^ 2322 (4123)
= .4 4111 (1432)=^ 2122 (2143)=^ 3323 (3214) = ^L4443(4321) +*MC4)fc(1)*x(2)*/8), (21)
= yl3111(1324) = .43313(2413) = yl2422(3142)=yl4442(4231). (19a)
l)
yl2121(1234)=^4141(2341) = ^ 4343 (3412)=^ 2323 (4123) Fafi(l)Ffitt(2)ky^F7i(3)F^W
= ^2112(1243) = ^ 3113 (2314)=^l 3443 (3421)=^ 2442 (4132)
= /l4422(3142) = ^3322(4213) = ^ 3311 (2413)-^ 4411 (3124). (19b)
= -2^ M (lM,(2Mx(3M,(4)g M , x . (3) (1234)
,42123(1234) = ^ 4121 (2341)=^ 4143 (3412)=^1 4323 (4123) g|WX<r(8)(1234)^[(*<J!?^)ft|l<iif,<1>Jkx(1)ft(r(1)
=^ 4341 (1432) = ^2141(2143)=7l2321(3214) = A2343(4321)
= ^3312(1324)=^14311(2431) = ^ 4412 (3142)=^ 4322 (4213)
= ^l4421(1342)=yl3411(2413) = yl3321(3124)=^3422(4231)
= yl2l42(l243) = /l 3112 (2314)-^ 3143 (3421) = ^3422(4132) -(k^k^)k^kp^h{l)K^, (22a)
= yl 3413 (1423)=^l 2113 (2134)=^ 2412 (3241)=^ 2443 (4312). (19c)
.42311(1234) = yl 244l (2341)=^ 334 K3412) = ^2342(4123) Fap(l)Fpa(2)ky^Fy8(3)F^Wl)
= ^l4113(1432)=yl4122(2143) = ^3123(3214)=.44423(4321)
= ^3121(1324) = yl3441(2431)=-.42421(3142) = ^3423(4213) = -2A(l)Av(2)Ax(3)Aa(4;)g^(l234)
= .44112(1342)=^l4313(2413) = yl2312(3124)=^4342(4231)
= ^2411(1243) = /l 2313 (2314)-^ 44 i 3 (3421)=yl 2423 (4132)
= yl3141(1423)=^3122(2134)=5.44142(3241) = ^ 3342 (4321). (19d)
^x<r(4)(1234)^[(*>*^)ft|4<fc<1>*x(2)**(1)
/l2143(1234) = yl 4321 (2341)=^ 34 i 2 (1324). (19e)
/l2341(1234) = .4 4123 (4321)=^ 24 i 3 (1243)=^ 4312 (3124)
= ^3142(4312) = ^ 342 H2431). (19f)
- (k^k^k^k^kx^K^ (22b)

Some of these invariants, of course, are unchanged by
certain interchanges of the momenta. Similar equations
hold for the coefficients B and C, Eq. (18). These rela-
tions, therefore, considerably reduce the labor of finding
ail the coefficients.
One can, moreover, take advantage of the fact that
the polarization tensor satisfies Eq. (12), in other words,
that it is gauge-invariant. It is easy to show that a
gauge-invariant tensor of the form Eq. (18) vanishes
identically if all the coefficients A vanish. Hence,
knowledge of the A's is sufficient to determine G>x<r
completely and it becomes unnecessary to calculate the
coefficients B and C at all.13 They will therefore be
ignored in the subsequent work. The terms which
involve an A will be called "heads" or "leading terms."
One can take further advantage of the gauge-in va-
riance of GpvXa by expressing this quantity as far as
possible as the sum of simpler gauge-invariant tensors.
These can be constructed from a consideration of scalar
products of four field strengths, which will certainly give
rise to gauge-invariant expressions. Thus one can define
the tensors gMvX<r(i) as follows:
^^)F^(4)F^(l)[F 75 (2)Fa 6 (3)-F 75 (3)F 5 e(2)>^ 1 >
= ^M(lM,(2Mx(3M,(4)gM,x.(5)(1234)
g M ,x/ 5) a234)^[(*>*W)(*/ 2 >*/ 8 >*x (1) ^ (1)
-A/ 8) */ftx (2) *a (1) ) + (*(1)*(4))
X(^(3)^(I)^(2)^(2)-^(2)^(3)^(1)^(2))]. (23)
It should be noticed that Eq. (20) involves only heads
in Eq. (19e), Eq. (21) those in Eq. (19f), Eq. (22) those
in Eq. (19a-c), and Eq. (23) those in Eq. (19d). These
few types of tensors, of course, need not be sufficient
to express G^x* completely. To find out to what extent
they and their permutations are represented in G^xa
and what remainder is left after this is done, the coef-
ficients A must be known more precisely. For this
purpose we return to a consideration of T^xo.
IV. EVALUATION OF THE POLARIZATION TENSOR
By the usual methods of carrying out the integration over the
momenta of the virtual particles,8 Eq. (10) can be transformed to

7Vxff(1234)=^/<*rJW
Fa$(l)Ffia(2)Fy,(3)Fiy(4)
f7MCn(^-x(1))-K]7,Cn(^-x(2))-]
= 4^M(l)^,(2)^x(3).4ff(4)^Xff(1)(1234)
Tr
g^(1)(1234)^*|l<2>ft/1>*x(4)*ffw, (20) + ((4)) 2yzy* - (&(1)+&(2)) (kW+kWy2y4
13
The method of defining the ambiguous integrals by requiring
gauge-invariance consists of calculating the coefficients A and
constructing B's and C's that make the tensor gauge-invariant. [same with K>M] y, (24)
This is the case because the A's are unambiguous while the C's
are ambiguous. The method is therefore completely equivalent to i
regularization. J
384 R. KARPLUS AND M. NEUMAN

where
l) J
^i(1234) = - -fdrydyz+ys+ydiy^yz-ys-yt)
yi tk^Ky2Jryz)+ktl^yh
-k^yi+WHyt+yJ+k^yt, X(yi+y 2 +y4-y 3 )i?(1234), (32a)
(25)
X<3> = l)
yi -k^yt+k^Kyt+yi),
-V 4 >y 3 , ^ 1 3ll2 (1234)=-Jj^r(y 1 +y 2 )(y 3 4-y4)(y 2 +y 3 -yi~y4)
and X(y 2 -hy 3 +y 4 -yi)>(1234), (32b)
dr=s
j J J J j dyidyxtysdy*.
yl 1 2141 (1234)=:-5j^Ty 1 (y 2 -fy 3 +y4)(yi+y2+y 3 -y4)
yi>0, Xyi=*1

Since only the leading terms are of interest, the quantities iyp X (yi-fy 2 +y4-y 3 )Z>(1234), (32c)
and K or M in the numerators of the two terms may be dropped;
as was already pointed out (see Eq. (9) and note to Eq. (15)), A^Hm4) = -A1^(m4)^~fdr(yl+y2+y3-yi)
they contribute only quadratically and are then accompanied by
a factor 5Mrj so that they constitute part of a term B or C, Eq. (18). X !(yi4-y 2 )[yiy2+(l-yi)(l-y 2 )]
Then the regularizing term becomes of order M~* and vanishes 4-(y34-y4)Cyi(l-y2)+y 2 (l-yi)]|X>(1234), (33a)
as the auxiliary mass tends to infinity. Hence, as far as head terms
are concerned, the integration over the momentum />M in Eq. (24) Al^Km4)=\fdr(yl-{-y2-y,-y4)
becomes trivial, and
X \(yi+y*)[yiyt+(X-yi)(l-y*)]
TMx<r(1234)~~ fdrD(\2U) 4-(y2+y)[y4(l-yi)+yi(l-y4)]}^(1234), (33b)

rr{rMTX(1)T,TX(2)TX7X(3)T(rTX(4)}. (26) /l 1 24 ^(l234) = - 4 r ( f T f ^ 2 + 3 ' 3 ) ( y 3 + y 4 )

Here
D(1234)-{l-*V+(l/1)C(^1>)s3'0'i+(A()syi3'2 +(>'i+>'2)(yi+y4)]Cyiy3-r-(l-yi)(l-y3)]
+C(>'i+>'4)(y3+y4)+(yi+y2)(y 2 +y 3 )]
+^(3>)2y2y3+(/fef4>)2y3y4-^(1)+^(2))(^(3)+^(4))>'2y4
-(*a>+ifeM))(ife( + ife(*))y 1 y3]|-2 (27) X[yi(l-y 3 )+y 3 (l-yi)]}I>(1234), (34a)
with '=e//c2.
.4 1 2 3 4 K1234)=i/ 1 ^rJ[(l-y4)(l-y3)+y 3 y4]
A further simplification results from the consideration of heads
only when the spin sum in Eq. (26) is carried out, because all but X[(l-yi)(l-y2)+yiy2]
twenty-four terms may be ignored. Thus 4-[(l-y4)y 3 H-(l-y 3 )y 4 ]
Tr {7M7X(1)7*7X(2)7X7X(3)7<rYX(4) | XC(l-y 2 )yi+(l-yi)y 2 ])/?(1234), (34b)
-4[(XM<i)X/2)+X/2>X,<D)(Xx(3)X^4>-fXx(4)X(r(3>)
4-(xM(i)x^3)-hxM(3>x,(1>)(Xx(2>x<r(4)-xx(4>x^2>) ^ 1 2 ^(1234)=^Jdfryiy 3 (l-yi)(l-y3)Z?(1234),(35a)
1 4 4 1 (2) 3 (3) 2
+ (X/ >X/ )-fXM< >X^ >)(Xx X^ )+Xx X^ ))
+ (XM^X,<3>-XM(3>X^))(Xx<4>X(r<i)-Xx^>X(r<4>)
^134i2(1234) = -^frfr(yi+y 2 )(y 2 +y 3 )(y 3 +y 4 )
-f(XM<4)X^)-X/2>Xl,<4))(Xx^)X(r(3)+Xx(3>Xffd))
+ (X<3>X<4> - X.U<4>X<3>) (XX(1%<2> - Xx(2)X<r<1>) ]. (28) X(y 4 +yi)P(1234). (35b)
It must be remembered, of course, that theX(i) have to be expressed
(i) Now, with the help of the symmetrization procedure indicated in
in terms of the three momenta appropriate to the index the X Eq. (7), one can write down the coefficients A in terms of the A\.
carries in accordance with the convention adopted with Eq. (18). From the nature of the A\ one can also deduce certain identities
By substitution of Eq. (25) into Eq. (28) one can then express among the ^4's, and these are indicated in Eqs. (36) to (41).
Tpp\ff in the form
2111 2111 2111 3111
JW(1234)~ 2 2 ^ i*M1234)V l W'WW m \ (29) ^ (1234) = A ! (1234) +A i (1243) +A i (1324)
j=2,3,4 1=1,2,4 =^ 2111 (1243), (36)
i = l,3, 4 w = l , 2 , 3
A 2121(1234) = A !2121(1234) +A i2112(1243) +A i3311(1324)
where the Ai^lm(1234) are the contributions of fMx<r(1234) to the = ^ 2121 (2143), (37)
coefficients A in Eq. (18). Because !PMx<r is less symmetric than
GprXa, as many as fifteen of the A\ must be obtained before the v42i23(1234)=^12ii3(2i34)+^13ii2(23i4).|_^12i4i(2i43)j (SB)
remainder can be generated by symmetry operations. Such a set 2311 2311 3121 2411
^ (1234) = ^i (1234)-fyli (1324)4-^l (1243)
may consist of the following:
= -yl 3121 (1234), (39)
8 r
A (1234) = A i (1234) +A ! (1243) +A !3412(1324),
2143 2143 2143
(40)
-4i2111(1234) =J dTyi(y2+y3+yd(yi+y2yy4)
2341 2341 2413 2413
X(yi+y 2 +y 3 -y 4 )i5(1234), (30a) ^1 (1234) = A ! (1234) +A ! (1243) +A i (4132)
=vl4123(1234). (41)
3111
^i (1234)=^J^r(yi-f-y2)(y 3 -hy4)(y2+y 3 +y4-yi) To apply these relations to G^x*, it is helpful to consider a
X(yi+y 2 +y 3 -y4)/>(1234), (30b) piece G:MvX(r/(1234) of this tensor:
G^x/(1234)=^ 2 1 1 1 (1234)^^^ 1 ^x ( 1 ) ^<r ( 1 )
^ 1 2 1 2 1 (1234)=-^J^ry 1 (y 2 +y 3 +y4)(yi+y 3 +y4-y 2 ) +^2121(1234)^M(2)^a)^x(2^ff(1)4-^2121(1243)^(2)^^1)/fex(1^<r(2)
X(yi+y 2 +y 3 -y 4 )i)(1234), (31a) +^ 2111 (2143)^< 2 >^< 1 ^x (2) ^/ 2) 4-^ 2123 (1234)^( 2 )^( 1 )^x (2) ^ (3)
+^ 2123 (1243)V 2) ^/ 1) ^x (4) ^ ff (2) 4-^ 2123 (2143)^< 2 >^^ 1 >/fex (4) ^ a)
8 r + ^2123(2134)y2)^(l)^x(l)^(T(3)_f_^2311(1234)[^^>^3>^X(1)^(1)
^. 1 2112 (1234)=-- / dTyi(y2+yz+yd(yi+y2-y*~y<d -^(3>^^1^x(2^<r(1)]+^2311(2314)C/feM<2)^<3)^x(1)^<r(2)
X(y 2 +y 3 -yi-y4)Z)(1234), (31b) -^^)^<1>^x(2)^<r(2)]+^2311(3124)[^(2)>fe^3>^x(1^(r(3>
-^/3)V1^x(2)^<r(3)]+^2143(1234)yfeM(2>^<1)^x(4)^(3)
33
^ 1 "(1234)==Jj^r(yi+y 2 )(y 3 +y4)(yi+y 3 +y4-y2) +^ 2341 (1234)C^( 2 >^ 3 >^x (4) ^ ff (1) +V 4) ^ (1) ^x (2) ^ < r (3) ]. (42)
X(yi+y 2 +y 3 ~y4)^(1234), (31c) A look at Eqs. (20)-(28) suggests that GMX(/(1234) be rearranged
 NON-LINEAR INTERACTIONS 385

as follows: indices of the gw. The superscripts on the ^4's define their func-
GMf,x<r,(1234)-^2i43(1234)^X(r(i)(i234) tional form [Eqs. (30) to (41)].
A calculation based on Eq. (46) of the cross section for the
-M^(123% M( , X * (2) (1234)
scattering of light by light is being prepared for publication.
+
(^) C " l 2 1 1 1 ( 1 2 3 4 ) g M , ' X f f ( 3 ) ( 1 2 3 4 ) V. LOW ENERGY APPROXIMATION
+^ "(2134)^ M x<r (3) (2134)+^2i2i(i234)^ x <r (4) (1234)
21

+vl2121U243)gM,ffx(4)(1243)] If the interacting fields vary so slowly in space and

time that the Fourier transforms have appreciable
2311 5
+~^[^ (1234)^Xff< H1234) values only for momenta whose absolute value is much
-^ 2311 (3124)^x^ (6) (3214)]+a(1234)V 2) ^/ 1) ^x (2) ^<r {3)
smaller that the mass of the pair field, the function
+a(1243)V 2 >^/ 1 )^x (4 ^a (2) +a(2143)^^ 2 >^^>^x (4) ^<r (1)
D(1234) in Eq. (27) may be approximated by unity.
The evaluation of the integrals Eqs. (30) to (35) then
X(V 2) ^ (3) ^x (1) ^(r (2) -^ (3) ^ (1 ^x (2) ^<r <2) ), (43) becomes trivial. In particular, the integrals are inde-
where pendent of the momenta. The values of the six basic
ones, Eqs. (36) to (41) are
a(1234) =
Jk^Likmk(4))A2m(UU) A 2111 _ 2121 A 2Ul= A2123=0
4-(^ ^ )^ 2111 (2143) + (^ (1 ^ (4) )^ 2121 (1234)], (44a)
(2 (4)

1 42143 _ A u n
A4321 = 4 / 9 K 4 (47)
6(2314) = -[(<2><4>),42311(2314)
/1 2341 = ^2413==^3412==_14/45/C4I
4-(^ (1) ^ (4) )^ 2311 (1234)+(^ 3 >^^)^ 2311 (3124)]. (44b)
The major part of the polarization tensor has now been ex- Hence
pressed in terms of the tensors gMvX<r(i) derived from field strengths.
The remainder
G>x<r(1234) = (gMxe (1) ( 1 2 3 4 ) + ^ w ( 1 ) ( 1 3 2 4 )
2 a(1234)V2)^'(1)^x(2)^<r(3)+ft(2314)(V2>^3)^x(1)^ff(2) 9/c4
24 perm

14
+6(4312) (V4W4)x(1)&<r(3) - k^k^kx^k/m +g M x (1) (1423))- -(gM,x,(2)(1234)
+6(2341) (V 2 >^< 3 ^x (4) ^ ff (2) -V 2) ^ (4) ^x (2) ^r (3) ) (45)
is still gauge-invariant and must satisfy Eq. (12) with appropriate +g,w (2) (1324)+g Mff ,x (2) (1423)), (48)
B and C terms. It is therefore identically zero. The reason for this
result is the fact that each head in Eq. (45) contains only three
different momenta. The third-rank tensor that results from con- 5(4)=-i-
traction, as in Eq. (12), with the fourth momentum then does not CK2))2 J
180 (hcK
2 2

contain this fourth momentum. The third-rank tensors that result

from the application of Eq. (12) to terms other than heads, terms - U(F,v(x)Fvx(x)FUx)F*(x))~], (49)
that contain a Kronecker delta, will always contain the momentum
vector with which they were multiplied. Hence the heads in Eq. and the effective Lagrangian density
(45) must satisfy Eq. (12) alone, and this means that every term
in Eq. (45) vanishes. As a check on the calculation, a(1234) and a2 1
6(2314) were computed explicitly as functions of null-vector L(x) = [5(^(*)F,,(*))2
momenta; they were indeed found to vanish identically. 180 hoc*
The vacuum polarization tensor can therefore be written
- UF,v(x)FvX(x)FUx)F'(x)J (50)
GVxa(1234)= S {^ 2 ^ 3 (1234)^ X f f ( 1 ) (1234)
24 p e r m L&
This last quantity, when expressed in terms of electric
and magnetic field intensities (D and B) becomes
identical with the result of Euler's calculation:3
+i44 1(12 34)g M> ,x, W (1234) L(x) = - (2a 2 /45^K 4 )[(D 2 -B 2 ) 2 +7(D -B)2]. (50')
It is a pleasure to thank Professor Oppenheimer for
his encouragement and to acknowledge many enlighten-
+ ing discussions with Dr. Res Jost. One of us (M. N.) is
^(^)^2m(1234)g'"X<4)(1234)
indebted to the Institute for Advanced Study for
+ having supported his research by a grant of AEC funds.
5(PW423^1234^^1234^' (46)
The sum over permutations here refers to simultaneous permuta-
tions of the labels of the momentum variables and of the tensor