PH YS I CAL REVI EW VOLUME 101, NUM 8 ER 2 JANUARY 15, 1956

Vacuum Polarization in a Strong Coulomb Field*f

EYVIND H. WICHMANN$ AND NORMAN M. KROLL
Co1icmbiu University, New York, New York
(Received July 8, 1955)

A study is carried out of the vacuum polarization in a strong Coulomb 6eld. Radiative corrections are
neglected. A perturbation calculation is avoided by making use of the explicit solutions of the Dirac equation
in a Coulomb 6eld. The Laplace transform of the polarization charge density times r~ is found and used as
a basis for further study. It is proved to be an analytic function of the strength of the inducing charge.
It is verified that the erst-order term in a power series expansion in the strength of the inducing charge just
corresponds to the Uehling potential. The third-order term is studied in some detail. The leading term in
the polarization potential close to the inducing charge and the space integral of the induced potential
divided by r are found to all orders in the strength of the inducing charge. Ambiguities are handled by a
method corresponding to regularization.
Some experimental applications are considered. The corrections to the Uehling term in these cases are
found to be small.

E. INTRODUCTION the inducing charge is a point charge, the 6rst-order

ECENT measurements of energy level diGerences potential will be referred to as the Uehling potential. It
in mu-mesonic atoms have raised the question as is well known that it gives rise to a measurable con-
to whether quantum electrodynamical corrections to tribution to the Lamb shift in hydrogen. The effect of
these level separations are of observable magnitude. '—' the Uehling potential on the x-ray 6ne structure
It is expected that the main quantum electrodynamical
eGect on the levels in mu-mesonic atoms would be the also been considered. "'
separation and on the levels in m, u-mesonic atoms has
' ' It may here be remarked
that the Uehling potential falls o6 exponentially at
eGect of vacuum polarization, arising from the coupling
of the electron-positron Geld to the Coulomb field of large distances from the inducing charge, and behaves
the nucleus. ' as (Inr) jr at small distances.
Likewise quantum electrodynamical corrections to The Uehling potential is the leading term in a per-
the x-ray fine structure separations in heavy elements turbation expansion in which o,Z is treated as a small
may be of observable magnitude. In this latter case, expansion parameter. (n is the fine structure constant,
vacuum polarization can be expected to be important, and Z is the magnitude of the inducing charge, in units
although it is not the only quantum electrodynamical of the elementary charge. ) The use of the first-order
eGect expected to play a role. ' ' term only when considering the effects of vacuum
The phenomenon of vacuum polarization in an polarization on hydrogen levels can thus be expected
external field, to first order in a power series expansion to be a very good approximation. This may not be the

previously. "
in the strength of the inducing field, has been discussed
Furthermore, in the case of a constant
external field, it has been discussed to all orders in the
case when one considers mu-mesonic atoms or x-ray
fine structure in heavy elements, since nZ is then of
order unity. It is thus of interest to consider higher
strength of the inducing field. "" For the case in which order e8ects.
7Vith this in mind, we have undertaken a study of the
*Work partially supported jointly by the Signal Corps, the
Once of Naval Research, and the Air Research and Development vacuum polarization in a strong Coulomb field. Ke
Command. avoid a perturbation expansion by making use of the
t Work partially performed at Brookhaven National Laboratory explicit solutions of the Dirac equation in a Coulomb
under the auspices of the U. S. Atomic Energy Commission.
f Submitted in partial fulfillment of the requirements for the field. Radiative corrections are neglected.
degree of Doctor of Philosophy, in the Faculty of Pure Science, %e first consider the general expression for the
Columbia University, New York, New York.
' L. Fitch and J. Rainwater, Phys. Rev. 92, 789 (1953). induced charge density and we show how the sum over
' V.
L. Cooper and E. M. Henley, Phys. Rev. 92, 801 (1953). states representing the charge density may be broken
' H. N.
C. Corben, Phys. Rev. 94, 789 (1954). up into partial sums referring to diferent angular
4 D. L. Hill and K. W. Ford, Phys. Rev.
94, 1917 (1954). momenta. %e next show how these partial sums may be
' Koslow, Fitch, and Rainwater, Phys. Rev. 95, 291 (1954).
'A. B. Mickelwait and H. C. Corben, Phys. Rev. 96, 1145 expressed in terms of a contour integral of the Green's
(1954). functions of the radial Dirac equations. The Green's
7A. L. Schawlow and C. H. Townes, "The EGect on X-Ray
Fine Structure of Deviations from a Coulomb Field near the functions are explicitly constructed, and their relevant
Nucleus" (to be published). We are indebted to the authors for properties discussed. Some further discussion is given
the opportunity to read their manuscript. in Appendix I.
R. F. Christy and J. M. Keller, Phys. Rev. 61, 147 (1942).
' R. Serber, Phys. Rev. 48, 49 (1935). The Laplace transform of the polarization charge
is E. A. UehIing, Phys. Rev. 48, 55 (1935).
density times r' is found and discussed. The expression
"V. Weisskopf, Kgl. Danske Videnskab Selskab, Mat. -fys. is regulated and renormalized, and shown (in Appendix
Medd. 14, 6 (1936).
's J. Schwinger, Phys. Rev. 82, 664 (1951). II), to be an analytic function of nZ inside the circle
8
 E. H. Mf I CH MANN AN 0 N. M. KROLI.
sentation based on the solutions to the Dirac equation
in the Coulomb field.
The sum (2) is divergent as it stands. In the course
of our study we shall try to give a more proper defi-
nition and to separate the physically meaningful
FIG. i. Feynman diagram for quantities from meaningless in6nities.
the polarization potential using a
Coulomb field interaction repre- We study the Dirac equation in the Coulomb field,
sentation. and consider simultaneous eigenstates to E, and the J,
I

l
I
I
Hamiltonian. "E
has all nonzero integers as eigenvalues,
I
I
and I, has the eigenvalues;
I
I
m= —1k/+a' —)kf+p ''' (kf —-'. (3)
I
I
I We are then led to radial equations which we write in
the form
~crZ~ =1. The first-order term is extracted and shown (X),+s)Lw(x)) =0, (4)
to correspond to the Uehling term. The third-order where
term is discussed in som, e detail, and likewise the be- d
—+1.
havior of the polarization potential close to the origin )
g dx g
is discussed. Except for the Uehling term, the charge
renormalization turns out to be finite up to o,Z= 1.
Higher-than-first-order vacuum polarization effects
on the energy levels of mu-mesonic atoms are con-
sidered and found to be small. Ql S= K] S j'N2X )
Ke also study vacuum polarization sects on the
x-ray 6ne structure separation, arising from the first- y = ctZ = e'Z/4sr pkc, p

and third-order terms.

The contribution to the Lamb shift in hydrogen from
x=r moc
third-order vacuum polarization is found to be neg- s= E/mpc'
ligible.
In Appendix III, the asymptotic behavior of the (E is here the energy).
polarization potential is discussed briefiy from the The complete solutions can be written in terms of
standpoint of the Euler-Heisenberg Hamiltonian. the radial eigenfunctions, and spherical harmonics as
In Appendix IV, some summation formulas are i k k+m ——,'&
derived. pi(km; r) = —— wi(x) F'lp;l;" —&(8; p),
x [k[ —12k
II. EXPRESSION OF THE POLARIZATION CHARGE
DENSITY AS A SUM OF CONTRIBUTIONS FROM k—m—-' &
DIFFERENT ANGULAR MOMENTUM STATES
Pp(km; r) =
2
——wi(x) 2
I'i p-:i-i"+*'8" &)
We are interested in the vacuum expectation value S 2k l
(~)
of the current operator; k 1 'k —m+-',
j„=-',e(vac py„f py„p vac), — Pp(km; r) =— -wp(x) '(~' &)
~ ~
(1) 2k+1 .
where — e is the charge of the electron, in the presence
k+m+-',
1 &

of a Coulomb field arising from a point charge of mag- &4(km; r) =-wp(x) I"l~+.l-. "+'(~' V)
nitude eZ located at the origin r=0. x 2k+1
Only the timelike component of (1) is different from
zero, and we may thus write the induced charge density; The labels k and s are suppressed in zo~ and mg. Using
addition theorems for spherical harmonics, we And
p(r) = 2e P(„&trace(P(r)P(r)),
—i2e P& i trace(P(r)P*(r)), (2) trace(P(km; r)P*(km; r))
where f(r) is the solution to the time-independent
Dirac equation in a Coulomb field, and (+) indicates J 4
= P Py, *(km;r)y, (km;r)
a sum over all positive energy states (= electron states), re~7 rs~l
and (— ) indicates a sum over all negative energy states
(= positron states), as de6ned by the Coulomb field. —
Q w„(x)w.*(x), (j= ~
k ~
—')
—, (6)
This charge density gives rise to an electrostatic 4m
potential, Vt (r), which may be represented by the n L. I. Schiff, Quantum 3lechanics (McGraw-Hill Book Com-
Feynman diagram in Fig. 1, using an interaction repre- pany, Inc. , New York, 1949), first edition.
 QACUU1VI POLARIZATION IN STRONG COULOM 8 F I ELD

so that
R - plone

with
L ( (R:R')
e2[u( 1 p~, cq
~2(r) =- / ~+ R'.
2 42r r20l2) a a a aa a a aa a\a & aa a\&%aaa aaa aa%~.

(.~(R:R')
XLP(+){wl(x)w1 (x)+w2(x)w2 (*)}
—P, , {w (x)w *(x)+w (x)w *(x)}), (8)
. -iR
where fw(x)j are solutions to the radial equation (4) Fro. 3. Special contours used in the discussion of the contour
for a given k, and given E, and with appropriate nor- integral representation of the sum over the energy states.
malization. (+) and (— ) indicate respectively sum-
mations over positive and negative energy eigenstates The Green's function is an analytic function of z,
to the radial equation. After the summation over m the except possibly at the location of the eigenvalues. The
spherical symmetry of the induced charge density is set of eigenvalues consists of a point set on the real
apparent. z-axis, between 0 and 1, and with z=i as a point of
We shall now focus our attention on the radial equa- accumulation, and of the half-lines
tion (4) and for convenience we shall suppress the
index k from the solutions in the next paragraph, as we z~ 1 and z ~ i.
have done so far.
Let us introduce cuts in the z-plane along the half-
III. EXPRESSION OF THE SUMMATION OVER THE lines defined by (12). It will be found that the Green's
RADIAL EIGENSTATES AS A CONTOUR INTEGRAL function has simple poles at the discrete eigenvalues
OF THE RADIAL GREEN'S FUNCTION and branch points at z=i and z= 1, and that it is —
Let 0(p (i. The boundary conditions,
otherwise a single-valued analytic function in the cut
plane.
(a) (w) finite at x=o, Let us consider the contours E(R) and P(R) in the
cut z-plane. (See Fig. 2.) E(R) is a simple curve, starting
(b) [w$ bounded at infinity, at the point +R+Oi, ending at the point +R — Oi,
define the eigenvaluesand eigenfunctions of the radial which encircles all the discrete eigenvalues precisely
equation (4). I,et z be any complex number, 220t an once. I'(R) is a simple curve, starting at the point
eigenvalue to (4). We may then construct a Green's
—R —Oi, ending at the point R+Oi— , which encircles
function, E, to the radial equation in the form of a none of the discrete eigenvalues.
bilinear sum; After the introductory remarks on the Green's
function as defined by (10), it is apparent that the sum
wp(xi 2)w„(x2i 6) over states in (8) that we are interested in can be
E„.
(xi, x2, z) =g
(e)
q

z
(10) represented by:

where the sum is over all eigenvalues e. This Green's

P {wi(x)wi*(x)+w2(x)w2*(x)}
function has the property (provided [wj is properly —P {wi(x)wi*(x)+w2(x)w2*(x)}
normalized); (—)B

(n~i+ s) (E (x,x2, s)) = I'1

(xI x2). — 1
dz trace(E(x, x; s))
2m'i ~ p (gl)

Z - plone ds trace(E(x, x; s)) (13)

27K u P(R)

in the limit E. , 8~00. This limit will exist only after

-RJ
aaaaaaaaaa ~ aaag ~
IR
aaaaaaoaaaaaaaaaaP
some regularization process has been performed.
+I In view of the analyticity of the Green's function, we
(
P(R) may deform the contours E(R') and E(R) . In particular,
Fxo. 2. Contours used in connection with the contour integral we consider the contours Ci(R), C2(R), C2(R), C4(R),
representation of the sum over the energy states. E(R), Li(R; R') and L2 (R; R') shown in Fig. 3.
 E. H. WI CH MAN N AN 0 N. iVI. KROLL
Thus The utility of (13) is a consequence of the fact that
there exists an alternative expression for the Green's
dz E= —, dz E— dz E—, dz E function in terms of the regular and irregular solutions
~ z(a) C1(gr, ) ~ C'4(a) to (4) for z not an eigenvalue, so that the summation
over states in (10) can be avoided. Thus: For z not an
ds E — &ts E, (14) eigenvalue, we construct two linearly independent
L, 1(a; a') "I.~(R; a') solutions fw&'&(x; z)$ and Lw"&(x; z)j to (4), by re-
quiring that Lw&i&j be finite at the origin and Lw&'& J
bounded at in6nity. Let
ds E=,~ '
dz E+ dz E.
1(a) ~ gg(a) E(z)=ws&'&(x; z)wi&'&(x; z) —wi&'&(x; z)ws&'&(x; z); (15)

This is a convenient set of contours, since it will turn then

out that the physically meaningful contribution will (8/Bx)E (z) =—0, (16)
appear as an integral along the contour I(R), while the I.et
ambiguities are connected with the other contours.
and E(z) 00, if z is not an eigenvalue.

Upon regularization of the expression for the charge +1 if x)0

density, and passing to the limit E~~ and E'~, 8(x)=~+-,', if x=0
only the integral along the imaginary s-axis will con-
tribute.
.'0 if x&0.

8(xp —xi) f wl (xl j z)wl (x2 j z) j wl (xi j z)W2 (x2) z) )

E (s) Ewp (xi s)wl (x2 j z) j w2 (xl j z)W2 (x2 z) ~
y

e(xi —x,) (wi& (xi, z)wi"'(xsj s); wi "(xi,. S)ws"'(xs, z))

(1't)
E(s) Ewe (xi s)wi (xs j z) j ws (xi j z)ws (xs j z) )

The diagonal elements of E are continuous functions functions:

of x~ and x~, the nondiagonal elements have a finite
step-discontinuity at x&=x2. It is apparent that I'(p+e) I'(q) t"
—Z
~(p;q; t)= (21)
(n*&+z). (E(xi, xp, z))=8(xi —xp), (18)
=p F (q+&s) I'(p) e!

so that we have indeed found another expression for the

Green's function. This construction is very similar in G(p;q;t)=—t' 'I'(p+1 q)1'(q)
F(P+1 q; 2 —q; t), —
I (2 q)I (p)
principle to the construction of a Green's function for
an ordinary second order linear diQ'erential equation. " &(p q; t) =
—~(p; q; t) —G(p; q; t),
(22)
(23)
The solutions fw&'&j and ['w&'&j can be expressed in
terms of the functions defined below. Ke de6ne single where
valued functions in the cut plane: arg(tt ')= (1—q) argt.
(s+1)'* by (z+1)', p 1;
——
For some of the properties of these functions, see Ap-
(s — 1)'* by (z — 1)i, =p ——i; pendix I. We note here the integral representations

(s' —1)'*= (z+1)'*(s —1)'. I" (p;q t)

1
Thus I'(q)
dxe*'xi' —'(1 —x) p-" ',
Im{ (z' —1)i} )0. 1(pF(q- p) ~.
~
t
(24)

valid when
s= (k' —y') '*,
Re{p})0, Re{q—p})0;
a= s —iyz/(z' —1)1,
t&= 2s+1,
(20) P(p; q; t)
X= 0 —iP/(s —1) '. r(p+1 —q)
e' ~
dxe ~'x™-l(x 1)~' (25)
We will use the following conQuent hypergeometric 1(p)P(1-q) ~.
where —ss. &argt &-', vr, valid when Re{ 0, and q not p})
Physt'k (Veriag Ju1ius Springer, Ber1in, 1931), Part L a positive integer. Forfurtherdiscussion& see Appendix I.
 VACUUM POLARIZATION IN STRONG COULOMB FIELD
Having made these definitions, we are able to deter- %e discuss in Appendix I how various propagation
]
mine $w&'&] and Lw&" in terms of the confluent hyper-
geometric functions, by examining the behavior of F
functions may be de6ned in terms of contour integrals
of the Green's function, and how these can be used to
and 6
at the origin and at in6nity, The variables x1 and solve the time-dependent radial equation.
x2 are real and non-negative, and by (19) we may
IV. THE LAPLACE TRANSFORM OF THE TRACE OF
write THE GREEN'S FUNCTlON WHEN xI=x2=x
—
-'22r &arg{ i X(Z' 1— —
)l} &-2, 2r, (26)
Using the results of the preceding section, we find:
and thus get
wi&'& )'$2x(z2 —1)&]' expgix(z2
(x; z) =i(z 1— —1)2] Db (x; z) = trace(Ãb (x,x; z)+E b (x,x; 2) )
X{XF(a;b; 2— ix(z' 1—)b) r (a)r (1 —b) p x(z'
2i— 1)—
b]"
= —2z
+aF(a+1 b; 2i— x(z' 1)—
&)} r (b)r (a+1 —b) (z' —1)b
w2&'&(x; z) = (z+1)bt 2x(z2 —1):]'exptix(z2 —1)'] X exp {2ix(z'—1)&}
X {XF(a; b; —2ix(z' —1) '*)
ZV
aF(a+— 1; b; 2ix—(z' 1)—
l)}, X t
F(; b; — ("—1)')
—1)&L2x(z2 —1)~]' exp(ix(z' —1)&) (27) 1(. —1)b.
wl& &(x z) =i(z
1)b— XH(a; b; 2ix—
(z' 1)')—
X{X H(a; b; 2ix—
(z' )
—2ix(z' —1)&)}
w2&2&
(x; z) = (z+1)~L2x(z2
+aH(a+1 b

1) l]' explbx(z' —1)&]

+
1+a b—
F(a+1; b; 2ix—
(z' 1)'—
)

X {liH(a; b; 2ix—
(z' 1)'*)—
X H(a+1; b; 2ix—
(z' 1)1)—
aH(a+1; — z' 1)')—
b; 2ix(— }.
Thus, +azkF(a; b; z' 1)')—
2ix(—
I'(a+1 —b)1'(b) (i b)
K(z) =4'(z2 —1)' exp (. (28) XH(a+1; b; —2ix(z' —1)i)
r(2 —b)r{a) )
~

E 2
+F(a+1; b; 2ix(z—' 1)')—
The last result is best obtained by letting x become very
small in the solutions (27), and using (16). XH(a; b; 2ix(z' —
1) &)] .—(29)
Using (27) and (28) and the definition (17), we may
construct the Green's functions explicitly in terms of
the confluent hypergeometric functions. By examining F- (p; — )= "D (; )d* (30)
the Green's function we may establish that (see Ap-
pendix I for some details):
This Laplace transform exists everywhere in the cut
{a) The Green's function has simple poles at the
discrete eigenvalues. These, of course, lie on the real plane when s is not at a pole of E.
z-axis between s=0 and a= 1. YVe denote:

(b) With the exception of the poles, the Green's

function is a single-valued analytic function of s in the
I = ip/2(z2 —1)l, (31)
cut s-plane. g=L1 —~]yt1+ ~i]LIy~(~ —1)], (32)
(c) The Green's function has branch points at z=1
and at s= — 1. pz L1 —~]LId~i]
g= ln (33)
(d) With xi (or x2) fixed and finite, and z in the cut (z' —1)b t1+t(2b —1)]
plane, not at a pole, the Green's function considered as
a function of x2 (or xl) is finite at the origin, and bounded Using the integral representations (24) and (25), we
at in6nity. get
—2z p1
Db(X; Z) = — —1)b]2a dg
L 2iX(Z' dlt exp{ 2ix(z' —
1)'((—q—
)}
(z' —1) b "0 41. (z' —1)& I" (a)r(b —a)
~0

Xp —l(q —1)a—ly/b-a —l(1 —$)b-a —l+ ~.(~-1).~~.-'(1-8~.-'

r (a+1)r (b —a —1)
(a+1 —b) 8
+z (a—l 1
( q)
b——l~ b —a—2
a
(~ 1) a+ ga(1 ])b-a—2~b —a—l(~ 1)a—l (34)
r(a)I'(b —a) I' (a+1)I' (b —a —1)
 E. H. %'I CH MANN AND N. M. KROLL
whence
122
l
iv r (b) r (b)
E,(p; —is) = — dp dn ]a—1(1 () D a — ba
l~— -1(— 1)a—
~ I+
z2 $ dp (s' —1)' I'(a)r(b —a) r (a+ 1)r (b —a —1)
I aI'(b)
y Pa(1 () ba ba -2(~—
-2~— 1)a pa(1 p) ~ 1)a
b a— b a —
2~— 1(— —I
r(a+1)r(b —a —1)
(a+1 —b) r (b)
V-'(1-&)"-'~" -'(~-1). ~

L +~-&]-'
r(a)r(b —a)
2«ba 1(-~ —1 )a-I
-
«I
oa

t
i~
z2 (s' —1) [u+2«]'- [u+2« —1]a [u+2«]
&
'[u+2« — 1]'+'
— — —
a~b a 1(~ 1)a I
——2(I«
(a+1 b)~b 1)
[u+2«)~ I
[u+2« 1]—
+' [u+2«]b '[u+2« 1]—
iv [1—t]'
dt
(s' 1)b -[1+ut]~ [1+t(u —1))' [1+ut]~' '[1+« (u —1)]'+'.
a[1—t] (a+1 —b) [1—t]

with «=1/2«. Thus

Eb(p; is) =—
z' —
1

1 ~p
tI
dtQ' exp( ig}——
iy
'
1

(s' 1) -[I t]L1+ut] [1+t (u —1)]

b
|
[1+ut]~ -'[1+t(u —1)]'+' [I+ut]~'[I+«(u —1)]'

Q (a+1 —b)
z (35)
[1—t][1+t(u —1)] [1+ut]
V. DIVERGENCE DIFFICULTIES AND iy
REGULARIZATION OF THE
LAPLACE TRANSFORM s' —1 J II (s' —I)&
According to our earlier considerations, we should
now perform the contour integrations of (35) in ac- X
cordance with (13) and pass to the limit, and finally [1—t][1+ut] [1+t(u —1)]. (s' —1)b
sum the result over k as indicated by (8). We should
then get an expression for the Laplace transform of the
X
induced charge density times r . By inspection of (35), [1—«][1+ut] [I+ t(u —1)]
however, it is seen that the result would diverge. In
2
addition, delicate considerations would arise as to
what order should be followed in the integrations and +sk 1 —ig+-'g2 —7—lnQ
summations. In our opinion, the theory does not give
2k [1—t)[1+t(u —1)]
any answer to such questions. iVz'
To deal with this situation, we will proceed as follows: [1 2g]
—t]L1+t( —1)]
We first sum Eb over k in accordance with (8). We then [1+.t] ("-1)' -L1
carry out the integration over t (which will give an
infinite term in first order in y). We finally carry out the + . (37)
contour integrations over z and at the same time we [1+ut].
will remove ambiguities by a regularization process.
For convenience, we shall-erst remove what will be The expression (37) arises from the first three terms
shown to be the Uehling term from our expression. We in a power series expansion of (35) in y. Let
write

is) =Eb'(p; is)+E—

b"(p; —is), — (36)
W" (p; is) =p k—
Eb" (p; —is). (38)
Eb(p; tt«' 1
 VACUUM POLARIZATION IN STRONG COULOMB FIELD
Let 0«&~ and 0(8(~. A(P; —is; e;b)
z'y 8 zs is'
W'(p; —iz; e; b) = dt Q kQ" ln(1+u)
s —1 4~ (s' —1) & ris 3u'(s' —1) &
oonss 3 (»'- 1)"'

x1— —+
1- 1 i -
1 1 ~1
+—+
3y
x [1—ig) (z' —
[1+t (u —1)]. u u' 1)&. 2 2u' Ee 2)

+sk 1 —ig —'g' ——lnQ —,

2

2k
1

[1—t)[1+ t(u —1)]

1

4u
p1
E 2eus
1

4u'
— 4ui
1y
I +O(e)+O(b). (43)

The first sum in (42) is independent of e and b. Inte-

1 its' grated around the contours and passing to the limit R,
[1 ig]
[1+ut) (s' —1)& R' —+~, the only contribution comes from the integral
along the imaginary s-axis. Let s= iy in this integration.
1 Finding the inverse Laplace transform, we thus get for
X +- (39)
[1+ut] [1—t][1+t(u —1)] 1
the induced charge density and the potential, to first
order in y,
It may be shown (see Appendix II) that for u finite, ~" y'L3+2y']
the order of summation over k and integration over t (y i
yl"."'(r) =
may be reversed in (38). The expression (38) is thus 47resr E3z. ) ~o [y +1
consistent with our program, and it is also gratifying
to see that the ambiguity concerning the order of in-
tegration and summation exists only for EI,'. X exp —218per (y'+1) l, (44)
k
The expression (39) is a polynomial of second degree
in y. Let us write this explicitly as ' p" y'[3+2y')
—is; e; b) Vp"'(r) =- e t' p q (2mscp
W'(p; ; b) = W&'&(p;
is; e— «r (3z& ( 5 ) ~s [y'+1)&
+~W& &(p; -'.;.; b)+~'W&'&(p; -", ; b). (40)
Consider the coefficients of the even powers. When X exp —2mocr (y'+1) ' (45)
integrated over the contour I(It), the integrals will
vanish. The only possible contribution actually comes In the expression for the charge density the infinite
from the contours Li(R:R') and Ls(R:R'), and so is point charge at the origin, required to eGect charge
dependent on the way we pass to the limit in the contour renormalization, has been omitted. The potential (44)
integration. Ke mill set these terms equal to zero. It is is precisely the Uehling potential.
also clear that the induced charge must change sign The contribution from the term 2 in (43) we set
when the inducing charge does, and therefore even equal to zero. Its actual value depends on the way we
powers of y cannot occur in our result. Evaluating perform the limiting process for the contour. Ke see
g &'), we get that it is of the form
1

W"'(pl «l ei b) co+ cs (1/p)',

1 1 3 or, 6nding the inverse Laplace transform, this corre-
sponds to a charge density
(z' —1)& 2u' e 2 4u 2(1+u)
cs'(1/r)'b (r)+ cs'(1/r).
—s' 1 t'1
1
—2) 1) —
——
1
1
s'
+—ln(1+u) Such terms in the charge density vanish upon regulari-
zation, which is easily seen from a study of the dimen-
2us Ee 4u
sions of the coeKcients. "
+0(.)+o(b), (41) We now study W" (p; iz) and — write
W&'&(p; —is; e; b) W"'(P; iz) = W" (P—; —is) —ysW&si (P; is), (46—
)
"W. Pauli and F. Villars, Rev.
Modern Phys. 21, 434 (1949).
co' has the dimensionof a charge and is therefore independent of
1+u. 3 (s' —1) "' 2 (s' —1)& the pair 6eld mass, while c2', having the dimension of charge
divided by an area must be proportional to the square of the

+a(p; —is; ', b),

mass. In the notation of Pauli and Villars, the conditions
(42) J'p(x)dx=0, J'p(x)x'dx=0, remove these terms.
 E. H. Vf I CHMANN AND N. M. KROLL
where
+-
p" dx
'

—ln(1 —x') In 1+x

ysW
"&
(p; —is) — l x
1
j. 1 1+u 1+u
dt Q kQ' --+u In —4u' lnu
—1 J, (z' —1)'.[1+ut)[1—
—

zs s=t ij $—I 1 —Q

2 (1 —u) ln'(1+u)
(s' —1)'u'(1 —u')

ks
I1 —ij[1+i(u —1)3 [1+ui)
—(11 u) lP (2,u') —u'(1 —u')+ —u'
3

X —g +——g lnQ+
——g— I
2 ks (s' —1)» 2 In(1 —us) —u In +3us(1 —u ) Inu
I —I
X——
[1 &][1+&(u—1)l [1+uij- (1-u)»I (2»)
(z' —1)u(1 —u')
)& ——lnQ —
1
-g' . (47)
—(1+3u)k(2, —u) —2u(1 —u') —~'(1+u)
6
This expression arises from the third power of y in a
power series expansion in y of E" in (36). I
I et us define the function &p(2; x), when xI &1, by —In —u') —2u(1 —u')
I+u In(1
lnu . (50)
1—
I

the power series:

The expression W&s& is a continuous function of u for

&P(2; x) =Q —, positive N. It further satisfies the relation
n=j ~2

and let I (x) denote the Riemann zeta function. We I

IT'"&(p; —is) «& Iu/(s' —1)'I
I

carry ottt the integration over t in (47) and get

for some ct if 0&u &1. We thus see that W&s& inte-
W&" (p; —"& (p;
is) = W iz) i (z' —
—
', 1—)— grated over the contours Cj, C2, C3, C4, 1.~, 1.2 vanishes
in the limit E, R'—+~. We shall call W&s& the "regulated
4s' —15s'+6 2 lnu "
5'(3), and in general denote regulated functions by a
(49)
3(s' —1)' 6(z' —1) (s' —1)' bar. On the other hand, the bracketed expression in
(49) gives a contribution to the contour integral

is)— depending on the manner in which we go to the limit.

where
The result would be of the form
2i (s' 1)»W&'—
(p; &

S4 4
kt+ks lnp.
—Ins(1+u)+2&P(2 u') ln(1 —
(s' —1)'u' 3
u) Both these terms can be removed by regularization. "
The remaining terms, W"'(p; — is), can be handled
2 7r2 dx in exactly the same way. In this case only the integral
+ u' [ln(1 —u')+u'~+2 ~l ln'(1 x' along the imaginary axis of this function contributes
3 3 0 g to the contour integral for the charge density, and the
contribution is finite for all p. To renormalize the charge
1 n
—2 ln(1 —u') ln +2u' Inu it is necessary to remove a term constant in p; the
1 —I remainder, denoted by W"'(p; — is), then vanishes at
p=0. We have further proved that the integral of W"'
S2 1+u is an analytic function of y in the region yI &1 (see
+ (z' —1)u' —
&P(2,u) ln
1 —u Appendix II) .
I

"This corresponds to a charge density of the form: k&'b(r)/(r')

1+u
——ln —2u +u lns(1+u)+-,'u»I (2,us)
+k, '(1/r'). This has, oi course, no very precise meaning, since in
order that the Laplace transform "exist" it is necessary that k1'
6 1 —I be inhnite.
 VACUUM POLARIZATION IN STRONG COULOM 8 F I EL D

VI. FURTHER STUDY OF THE RENORMALIZED converging when. 0(p(2, or

LAPLACE TRANSFORM
e y'(mpc) '
By the results of the preceding section, and of Ap-
pendix II, a power series expansion in y can be carried 4«& ) les

out. The radius of convergence is IVI =1, and only odd

powers of y will occur. Let us denote + (0 170044) (p/2)' —(0.274793) (p/2)'
—(0.118519)(p/2) 4
lnp
q(p) = due-& p(r)xs, (51)
+ (0.303301)(p/2)4+O(p') }. (59')

q(p) P +2n+lq(2n+1) (p) (52) The term proportional to p'lnp is not unexpected.
n=p
It gives rise to the leading term in the third-order
q'"(p) =q(p) —yq&'&(p) —7'q"&(p), charge and potential at large distances:
(53)
q'(p) = q(p) —yq"'(p). e (2 )(Aq'
We thus have 42repr (2252r) Empcr)
t" (60)
"q"'(p) =-",
4~2(a) J
e
I
(mpc)s
p"'(r)-
e (40'( Aq'

(p)= —a) 4+&

e
I
(mpc)'
I
("
t dye'"(p (55)
42rr &2252r] Empcr)

(as r-+oo). The same term, can be derived from the

Euler-Heisenberg Lagrangian. (See Appendix III. ")
For large p, we get
The induced charge gives rise to an electrostatic poten-
tial Vr (r), according to
(p) = —
e (mpC)'
——
2r 7 2

V, (r) =
1

rap
—
~„
p~
p(u)u'du+ —
&p
i
~ ~
p(u)udu. (56)
limq'2&
42rs&
I

We now return to W"'(p; y). After an integration by

A, J
I

6
— —|'(3)
9 3
. (61)

parts with respect to t, we may write it in the form

We also de6ne
&"'(p y)
f V„(r)e 1'xdh, r= (A/mpc)x,
U(p) = (57) f' [1—t][1+ut]+[1+t(u—1)]
dt
and get (using Poisson's equation) [1+ys]» [1—t][1+ut][1+t(u—1)]
p

(A)21—1 kQ*cosg —Q"I k ——g' —

~(p) = I I —,„i q(p') d p'. (58) k 1
-'2ysinQ
2 )
I

hmpc) ep p' jp

From this we may find the Laplace transform of positive 2&y' p' dt
integral powers of x times the polarization potential by
[kQ' cosg
a differentiation with respect to p.
[1+ys]& "p [1+ut] &-1
We expand (50) in powers of p, and carry out the (1+y')'
integration over y. Using (54), we thus have —Qs(k —kgs —ys lnQ)] ——
-', [ksQ' sing
-',

7'q"'(p) =~',
e
(mpc)
4~2( tt )
I I I
(p—) (2r)
II
&2) &4)
— —
52r' 13
I

72
+-24 —(&'g —-'&v —-'»'g l Q —lv'g)Q"] +~"'(y) (62)

(py' 2 ~2 19
+ E2) + + gpss(2rq X'"(y) is determined so that, identically in y,
3 6 15 (2) E4)
lim W'"(p; y) =0. (63)
312rs 1- pq
4
16
x — -+- +I — — lnp t
I '1 The discussion in Appendix III implies p"'(r) falls off faster
360 2 2) 135 &
than (1/rr) so that (60) indeed gives the leading term in the
charge density at large distances. Since the Uehling term will be
8x 3347' seen to. dominate at small distances, one notes that the induced
+ +O(p') 59) charge densities at large and small distances are always of op-
6 is 2835 posite sign.
 E. H. %I CHMANN AND N. M. KROLL
We find (see Appendix II) The f's are terms added to remove the (diverging)
terms to erst and third order in y.
ks For small p, we may expand
~"'(y) =
L1+y']"=' "+ Y[1+y'] '
+,
t 2k'
8 e (Bzpc)
q"—
'(p)
7$ 2p 00 00
-ap (,=o) 4or'L h )
ko(1+y') [1+y']'*
ZZ -
i ~~ ig
X &' —
5
|(2)—-f'(3)+ —
1
f(4) +—— f(5)
(e+s)k k .64 8 64 32 2
(nPs)'+ y'&'[ -
95 55 e (moc)'
+0 (y ) '=
)

t (4) t (6)
~2ky2 16 16 l
I
4or' E f) )
(64)
(rs+ k) '(1+y') X(y'(0.015191)+y'(0.007127)+O(y')). (67)
We may also find the limit of the integral of (62) with For y t t
~ 1, we msy write
respect to y, when p goes to infinity. The result is (see '
8 e (mpcy
Appendix II) q"'(p)—
e ' -~p — b=o) 4~o& e )
(mocy
lim q"'(p) = Xy'[0.015191]F)(y'). (68)
4~p( a )
The function Ii j is shown on the graph in Fig. 4.
t'ai v 1v'
X —2&k tan VII. POLARIZATION POTENTIAL AT SMALL
(s) k 6k' DISTANCES. CORRECTIONS TO ENERGY
LEVELS IN MU-MESONIC ATOMS
eke eke
+4K 2
~=»=) (++s)'+y' (n+k)'
We return to the question of the polarization charge
at the origin. We combine (61) and (65) and write

—k tan- l+
yk
&Q'=4~
(5) »m Lv'q") (p)+q'"(p)]
(~yk)
t
)
~+s& (~+k) t
Eeoc)
l
~" (69)

kv3 Expanding for small y, we get

3 (e+ k)' 2 (0+k)'
~Q'=- y' -f (3)+
3
— —+y' ——
9 6.
f(4)+-f'(2)
20 4
The fact that this limit is finite corresponds to the
presence of a point charge at the origin, just as in the
case of q(')(p). ——
27 2
(f(3) —f(4))+-(| (3) —l (5))
q"'(p) can be expanded in powers of p, for small p, 5 3
and expressions analogous to (65) obtained for each
-34
coeKcient. We have carried out the calculation for the
lowest terms only, obtaining
+v' l (5)+ —
t(7) t
-1g+ —
—lf(6)+12t (5)
21 3 E 24)
i
8
Bp
q'"(p)—
4s'( („=p)
e (nzpcq'

))i j I (v~)
5
+-f'(2)f(4)+-l'(3) +oh')
2
(7o)
or
k[rP+2es] 8Q' ——e(y'(0. 020940)+y (0.007121)Fp(y )} (71)
X PP (rl+s) + y + (tt+s)[(i+s) +y ] The function Fo has been rather roughly evaluated
numerically and is represented graphically in Fig. 4.
2k(e+s)
(e+ s)+ [(e+s)'+y']l
f ~( —gof„o(&)
) For small y, we have [from (70)]
~oh') =1+&'(0 51g3)+O(V') (72)
I-i; y'-
+P —k s —k+ —— k ks 8Q' represents the part of the higher-than-first-order
polarization charge which is located at the origin. Since
2 2k 2(k+s ) (k+s ) the total polarization charge (to higher than first order)
—fo") —V'fa") (66) vanishes; — l)Q' is the part of the higher-than-first-order
polarization charge located outside the origin. Thus, for
 VACUUM POLAR I ZAT I ON IN STRONG COULOM 8 F I ELD
Thus, for uranium, Z= 92:
AE„'/E„1.
6—
X 10-4.
For low Z, the approximation (75), while not so well
justified, is almost certainly an upper limit. The con-
tribution is then particularly small due to the smallness
of y.
In this estimate, we have ignored the effect of the
finite nuclear size. Nevertheless, we are confident that
(75) gives a good picture of the order of magnitude
involved. Our conclusion is thus that with present day
experimental accuracy, level displacements in mesonic
atoms, due to vacuum polarization in higher order than
0.5 I.O the first, are not detectable.
y2 The effect of the Uehling potential (44) on level dis-
Fro. 4. The functions Fp(~') and fl&(p') defined in the text in
connection with the polarization potential near the inducing
charge and the space integral of the polarization potential divided
elsewhere. '"
placements in mesonic atoms has been considered
' In this case, the eGect must be con-
sidered susceptible to measurements. In some recent
by r.
measurements' on x-rays from mesonic atoms, for the
r small compared to the mean radius of the outside purpose of obtaining a value for the mass of the mu-
meson, there is indeed some indication that the eGect
charge distribution, we get for the higher-than-first-
order potential of vacuum polarization has to be considered in the inter-
pretation of the results for consistency with other mass
Vp'(r) = Vr (r) 7Vr ~'i(r) —8Q—'/(4prepr). (73) determinations. In the case of the first-order term, it is
A mean radius, r~, for the charge distribution outside also easy to extend the result to the case of a nucleus of
the origin may be defined by finite size. '

goo VIII. EFFECT OF VACUUM POLARIZATION ON X-RAY

r p' 4pr —
r'—
p'(r)dr —8Q' FINE STRUCTURE. THIRD-ORDER CONTRIBUTION
TO THE LAMB SHIFT IN HYDROGEN

f'A )4 We consider the effect of vacuum polarization on the

=»m — —V'"(p)+ —
f} 8
C'"(p) 4 x-ray fine structure separation in heavy elements, in
particular the 2pi —
~

-Bp c}p Empc) 2p; separation. Our interest in this

question derives from attempts that have been made to
1+ps (0 2692)Pi (p~)- infer something about the nuclear size from an analysis
=(0.8579) ~
. (74)
~

(mpc) 1+y'(0.3401)Fp(p ) of measured separations.

Schawlow and Townes '~ have pointed out the
The mean radius is thus about equal to the Compton existence of a systematic deviation of the experimental
wavelength of the electron. It is also interesting to note separation from the theoretical prediction, which varies
that the mean radius does not change much in the rapidly with Z. They demonstrate that such an effect
range: could be attributed to short range departures from the
{) w~ w] Coulomb interaction. An attempt to attribute the
eAect entirely to the finite size of the nucleus leads to a
We use (73) to estimate the displacement of energy
nuclear radius considerably larger than that obtained
levels in mesonic atoms, due to higher than first order
from other experiments. It is clear that quantum elec-
terms in the vacuum polarization.
trodynamical modifications of the Coulomb interaction
We use Schrodinger wave functions for a pure
will also contribute to the effect, and indeed if one now
Coulomb 6eld to describe the meson, and assume that
regards the nuclear radius as known from other experi-
Z is sufficiently high, and the principal quantum
ments, one may make use of this eGect as a means of
number sufficiently small, so that the meson is well
observing quantum electrodynamical eGects in heavy
inside the mean radius (74). We thus find the leading
elements.
term in the displacement, AE„',simply as the expecta-
One such eGect, although by no means the only one,
tion value of the potential Vi'(r). Let E„denote the
is the eBect of vacuum polarization. Estimates of the
unperturbed energy. Then
contribution from the Uehling potential to the 2pi — 2p*,
~E.'=(~~ V '(r) ~~), separation have been made previously, taking the
DE —2 5Q'/Ze=
'/E„— 2n(y'(0. 020940) leading term only in an expansion of the expectation

+y4(0. 007121)Fp(y')) (75) 's A. L. Schawiow and C. H. Townes, Science 115, 284 (1952).
 E. H. WI CH MANN AN D N. M. KROLL

.50 important one. It could in a rough way be depicted as

a diminuition of the parameter Z occurring in the wave
.20 functions for the 2P electrons. It would thus decrease
the parameter C. Such an estimate would, however,
necessarily be rough, since it is the very different be-
Sc&)
P havior of the 2P~ and 2P; wave functions near the
.I0 origin that gives rise to the shift due to vacuum
.09 polarization, and only a calculation using "screened"
.08 wave functions could establish how screening affects
.07
this different behavior. The second eQ'ect is given by the
.05 difference in radiative corrections to the interactions
between, respectively, the 2P~ and 2P~ electrons and
.04
the rest of the electrons in the atoms. This can be
.03 expected to be small.
Let us next consider the eQect of the third-order
polarization potential on the fine structure separation.
.02 From the Laplace transform (57) it is easy to obtain
the expectation value of V~&" (r) over nonrelativistic
wave functions in a Coulomb field, as an expansion in
powers of p. It is considerably more difEcult to find the
.Ol expectation values over Dirac wave functions because
'FP 75 80 85 90
2 of the occurrence of fractional powers of r in the ex-
FrG. 5. The contribution of the Uehling potential to the pression for the square of the wave function.
2Pg—2pg separation in heavy elements in units of the Rydberg.
As an orientation we shall first consider the con-
tribution to the fine structure separation for small y.
values in 7.8 This approximation turns out to be This may be found as follows: We expand the square of
inadequate when Z is large, and we have therefore the Dirac wave function in powers of p and take only
carried out a more precise computation, as follows: the leading term. We thus get (after an integration over
Using relativistic Coulomb wave functions for the angles)
2P; and 2P; states, we have calculated the level shift
due to the Uehling potential by evaluating the expec-
tation values in the appropriate states. We express the J~ x~IP(2P~) I~= (1/24)y(yx)4+O(y6)
result in the form of correction factors to the shifts one (78)
would obtain by taking the leading term only in an
expansion in y. Thus dw x'IP(2p)) I'= (1/24)y(yx)'
—9mpc'o, 'Z'
—~(2P. 7~ I
"'I2P:-)= -(2P:), X (v*)'+ (9/4)7'$+O(v').
I

1120m
Using (57), (58), and (59'), we thus get
—27spc Q Z "'
—e(2P~ el'p"'I2P~) = -(2P:) ~(2P) y'l'i 2pl) e(2P& v'l'i "' 2P&)
—
I I I I
I

ii20x (3/32m)ny'mac'(0. 01417)+O(y'), (79)

The contribution to the 6ne structure separation thus giving the ratio of the third- and first-order shifts:
becomes
"'= (n'Z'— Bi ~'&/8p&'&
~—(0.212) y'+O(y'). (80)
EBi &". 5i 80m-)C,
/—
(77) tIn obtaining this result, the constant C has been set
C= L9C(2P~) —2C(2P1) 3/7,
equal to 1 in (77).g
where E„ is the Rydberg constant. C(2P~), C(2P;), C, We have tried to improve the estimate given by (80)
and 8~&') are given in Table I, and 8~&') has been repre- by a method which cannot really be strictly justified.
sented graphically in Fig. 5 for some values of Z.
The large values of C shows that the lowest order
TABLE I. Values of the parameters de6ned in Eqs.
(76) aiid (77).
approximation is not justified. Note in particular the
rapid variation of b~&'& with Z. c(&py) cg&~) g„(3)/g~0)
We have here neglected the presence of the other 95 3.82 0.65 4.73 0.287 —0.06
electrons in the atom. They give rise to two corrections 90 3.20 0.65 3.92 0.172 —0.06
which may properly be considered in this connection. 85 2.74 0.65 3.33 0.104 —0,06
78 2.24 0.66 2.69 0.0500 —0.06
The first correction is the effect of screening on the 70 1.85 0.67 2.19 0.0213 -0.05
wave functions for the 2P electrons, and is the more
 UACUUM POLARIZATION IN STRONG COULOMB FIELD
We carried out the following computation: Using (57), when —~x & argt ~-', x, where
(58), and (59') we 6nd the averages of the polarization o. = sign (argt),
potential V(3& over xe ~, x'e &, x'e ~ and x'e & . We
then use these averages to find, by interpolation, such r (22+q)r (22+ p)
averages as x'e & where g represents the fractional g(p;q t)-Z t ".
exponents occuring in the square of the Dirac wave r{q)r(p)N!
function. In this way, we arrive at the values of 8„")
The integral representations (24) and (25) are easily
given in Table I. In view of the fact that the numbers
established: (24) by direct evaluation of the integral,
in the last column are fairly small, it did not seem justi-
and (25) by noting that the integral satisfies the dif-
fied at this time to carry out a more elaborate evalua-
ferential equation (1), and that it has the same leading
tion. We estimate that the errors in the last column of
term in the asymptotic expansion as II, as given by (4).
Table I may be as high as 50/o.
Let us now study E'(x), x2, z). If we write (for the
One may thus assume that the Uehling term gives the
case x1 (x2):
main contribution to the shift arising from vacuum
1
polarization. The shift has the same rapid Z dependence
as the deviation found by Schawlow and Townes, "and
E(xt, x2., z) = g Q A~„F(a+22,b; —2ix(z' —1)&)
nm m=O
is of a similar order of magnitude. It has, however the XII(a+2)2 (z'
2ix—
b ),
1)&— (5)
wrong sign, which may be taken as an indication that we may write
other quantum electrodynamical eGects play an im-
portant role. +(x1 x2 + (x1 x2 «)+It (x1 x2 z)
z) (6)
The contribution from V&&" to the energy level dis- I i
placement in hydrogen is easily computed, using (57), K'(xt, xs, z) = P Q A„„F(a+I; (z' 1)'—
b; 2ix— )
n=o m=o
(58), and (59). The result is, for the 2s state, 308 cycles/ —2ix(z' —1)&),
sec, and therefore entirely negligible. XF(a+2)2; b; (7)
1 1
APPENDIX I E"(xtxs, z) = P P A „F(a+n;b; —2ix(z' —1)&)
F(p; q; t), G(p; q; t), and H(p; q; t) are special solu- X G(a+2)2; b; —2ix(z' —1)&). (8)
tions to the differential equation Let
[20«) (x; z)]— = z)] —[2()(2) (x; z)].
+(q —t) —
[2()(') (x (9)
t —
d2

dt'
p 1!'(t) =o. We shall study the behavior of as we pass the cut E"
dt in the z-plane. Let us denote: (where f(z) is a function
6 and H are defined so that they satisfy the same well- of z)
known recursion relations as Ii. p)1 or —1, p(
Using Pochhammer's contour I', we may 6nd an " f+(p) = lim f(z), f (p) = lim f(z), (10)
integral representation for F, valid for all p and q,
except when p or p — q is an integer: (p' —1)+'= —(p' —1)-',
—r (q) exp( —i2rq) 1=a++a, —
b (12)
F(p t)= = &- = a+ = a- I,
4I'(p)r(q —p) sin(2rp) sin[2r(q —
q
p) j I &+ I I I I I I
(13)
(p-1)+'/(p-1)-'= —p/I pl (14)
(1 —z)2-~-'e"
dzz
—
(2)
P F+(a; b; 2ix(p' — 1)')—
By deforming this contour, we get the asymptotic ex- =exp[2ix(p' 1)'* (a+1; bi —»x(p' —1) )~
jF-— (15)
pansions:
r(q) (1) — ( 1)a .2()1 (1)
F(p' q' t) = (+t) "g(p' p q+1' t)e'""' Ipl l~+a+
r(q —p)
(1) — ( 1)s
(3) Ipl ~+a+
(16)
— p l~l' r(1 —a+)r(a+)
&(p;q;t)=
r(p+1
r (1 —q) g(p; p —q+1; t), (4)— (2) —( 1) ~
pl l+a+
I
r(1 —a-)r(a-)
~+( )

' K. Y. Whittaker and G. N. Watson, 2 Coze'se of Modern p l~l' I'(1 —a+)I'(a+)

A)salysts (Cambridge University Press, Cambridge, 1950), fourth (2) =( 1)-~ g) ~(ft)
edition, p. 256, I p lt+ai
I
r (1—a )I (a )
 E. H. WI CH MANN AND N. M. KROLL
() X I'(()+)I'(1 —a+) co. Using the radial equation inside the integrals (21)
E-(p) = — — E+(P) (17) and (22), and letting r — +0, the result (23) follows
a+~+I'(~-) I'(1 ~-) readily, provided that the interchanges of order of dif-
Thus: ferentiation and integration are allowed.
E "(x,,xg, p) =E+"(x)',x, ; p). All these remarks are purely formal; to be more
By inspection, we see that E" has no poles, and that precise means essentially to prove an expansion theorem
E" is regular at s=1 and s= — 1. Thus; E"(x),x2, s) for the radial eigenfunctions.
is a single-valued analytic function in the whole complex We thus see that the transformations (20) play the
plane. It follows that E"
when integrated along E(R) role of time development transformations ("propagation
or I'(R) gives zero. The charge density could thus have functions"), and various types of such functions can
been dined only in terms of E'. However, an examina- be constructed by selecting suitable contours.
tion of the asymptotic behavior of E' using (3) shows The procedure here outlined makes possible the con-
that then the method of deforming the contours so struction of radial propagation functions for each k.
that all physically signi6cant contributions would come From the matrix elements of the radial propagation
from the integral along the imaginary axis would fail. functions and products of spherical harmonics, we may
Concerning E' we see that possible poles are located construct the corresponding propagation functions of
at the zeros of E(s), i.e., at the points where a is a non- the full Dirac equation in the case of a Coulomb field,
positive integer. A closer examination shows that the in the form of a sum over k. If we let the strength of the
poles are precisely at the bound-state eigenvalues, and Coulomb 6eld go to zero, the sum over k can be ex-
the correctly normalized bound-state eigenfunctions plicitly carried out, and we obtain the usual propagation
may be obtained from the residues of E' at these poles. functions for a free Dirac field.
The correctly normalized continuum state eigenfunc- APPENDIX II
tions may likewise be obtained by considering the sum
of the integrals of E' over the contours I. (R); R ')
L~(R; R'), letting R go to R'.
and. We prove here that

I.et us now consider the time-dependent radial equa- (P' y)

tion:
J,
(&.~+~~/~~) [(t (»' ~)) =0 (19) is an analytic function of p and y in the region
I.etC be some contour in the cut s-plane, which does Re[p)&0; ]y[ &1.
not go through a pole of E (x)x~, s). Let Writing
x= (1-&)/[1+&(u-1)), (1)
(Ge(x), x, ; 7)) = ds(E(x, ,xg, s))e-"*. (20) then, after performing an integration by parts, we may
2)l Z write W" in the form
Consider in particular the contours I'(R) and E(R) ~ ~" V;y)=~."(p;y)+~." V;y),
These contours de6ne transformations G~(g) and G~~g)
such that if [f(x)) is some "physically well behaved" W "'(p; y) = dx P T (') (p; y; x), (2)
function, then: Jp k=1
1
Lf(* )) ( )=J dy(G ( )( y', ))Lf(y)), (21) ll'"'(P'y)= t dxZT2(")(f y x)
0

[f(x; 7))p(s) J
0
dy(G&(z)(x, y; r))[f(y)), (22) T("'(P y *)
. =(1+y')',
1

1
are solutions of the time-dependent radial equation. kQ' cosg
This result is easily verified formally by substituting p' 1+x(u —1) x[1+u —x) .
(21) or (22) into (20), and inverting the order of inte-
gration and diGerentiation. kQ" —) kQ'g' —) y~Q" lnQ) (3)
For certain functions f
the transformed functions T2(") (p; y; x)
approach a limit as E goes to inanity. In this case,
2pg Q
[f(x; o))s+[f(x; o))~=[f(x)) (23) [kQ' cosg
We can see this in a somewhat unprecise way as follows: [1+u —x)[1+x(u —1))
p'
Let f(x) be expanded in terms of the radial eigen-
functions, (both discrete and continuum eigenfunc-
—Q" (k —kg' —'y' lnQ)) ——
—',
P
[kQ's
—, sing
f
tions), and let be such that there is a eo such that no Vg
eigenfunction occurs in the expansion whose eigenvalue —(k'g —k'gs+-,'ky'g
-', lnQ —y'g)Q")
-', . (4)
is larger in absolute value than cp. Let E be larger than
 VACUUM POLARIZATION IN STRONG COULOMB FIELD
It is the behavior of the T~~' at x=0 and x= 1 which '&hen,
prevents the analyticity properties of the integral over
8"" from being almost self evident. We therefore split A2(p; y; y) = Q dx T2~"', (10)
a=l Jo
the x-integration into three intervals
and
0&x&(,. (&@&1—6; —5&x&1,
with
1
IA (p;y;y)l =E. "1+y' l1+Nl'
0&&&-, '; 0&b&-', .
We consider the region for some E& independent of p, y, y, and e. Also,

«(p} )0; (6)

dy. A (2P;y; y)
&E2~"
IPI
The idea of the proof is to show that the integral over 1+ IPI
the central interval has the desired analyticity proper-
for some E3 independent of p, y, and e.
ties, and the other two integrals tend to zero with e and
b, uniformly in the region de6ned by (6).
(b) Let e &x&1— k Then
We have &1—b,
Re (1 —y')'*& )p) 0 I QI
(13)
P B~(p; y; y) =2 Ch T2(~),
for some 0&p, &2, depending on yo only.
k=1 4,
Q=xL1+x(N —1)3/I 1+I—xJ, and B2(p; y; &) is an analytic function of p and y in
the region
xL1+~—*j
g= —ln
vy I l~l &vo, «(P})0.
P L1+x(u —1) j Furthermore,
Ke consider 8"2'". I.et
y
IB~(p y'») &Ea
A2(P;y y)=„ 0
dx Q Tm'"', I

1+y' I1+ml'
for some El independent of p, y, and y. Then,

~ dy
B2(p;y;y)
~o
C2(p; y; y) = dx Q T2&".
is an analytic function of p and y in the region

(a) Let 0 &x & e. Then there is a Eo, independent of Ivl&yo, «(P})0,
p, y, y, e, xsuchthat: and
IPI
IQ('- ')'cosgl &x E,, dyB&(p; y; y) &E4
IQ~'-
1+I pl
'&
singl &xi'Eo,
for some E4 independent of p and y.
IQgl =* E., (c) Let 1 —8&x&1. Then,
&x~E„ —
l»QI &3(1—
IQg'I
x) &-' lgl &3(1 x) &-',
IQg I
=* E., IkQ o g— — —-'y'1 Ql &E,(1 —*)*k,
Q (k -'kg
IQ lnQI &x~E„ ksQ' sing —Q" (k'g —~k'g'+-', ky'g lnQ —p'g
-',
I I

IQg lnQI &x~EO. & E|;(1—x) 'x~k'

Furthermore,

IQI =-:, IQ"'-"'I &x"Eo(4)" ', for some Eq independent of p, y, y, and 5. Also,

yb
dxT2"& &E6
I
1+~I' C1+y2j
(16)
1 Ill yb
ds ~ —c~ I~2(p' y' v) &E~
~0 1+x(N —
I

1) I1+el I1+~l' ll+y g

 E. H. NICHMANN AND X. M. KROI. L
fol soine Ks independent of p, p, 3) and y. Then) —ksP"it" [ln'rt)+-', kg'ri" [in+
-',

Ipl —-,'[in&)[lnrt) p"rts I . (21)

Cs(p; y; y)dy ~Kryo (17)
1+ IPI
for some Kr independent of p, y, b. We then expand [1— f'rt) ' in a power series, and inte-
(d) Thus grate the resulting double series term by term.
The expression (66) can be derived by similar
dvlvs'"(P; y) methods. It may be noted that a certain care has to be
~o exercised in all these integrations, especially in the
is an analytic function of p and y in the region: regions
0. t~] y~ 00 .
Furthermore APPENDIX III
OQ oo
We consider the Euler-Heisenberg Lagrangian den-
&~s'"(P y)= dx P Ts"&=+
k=l k~ls)p
dxTsis&,
s,ty2o, m.

Ipl L[x) =-', epE'+ —

+O(Es).
dye "'(P; y) ~Ks 3t50x'mp4cv
"o 1+I~I
for some Es, depending on pp only. E is the electric Geld strength, and the magnetic Geld
The proof for lV&"' is entirely similar, only the has been put equal to zero. We assume that the 6.eld can
estimate now becomes be derived from a spherically symmetric potential, and
thus get the 6eld equations from the variational prin-
ciple:
dylvi'"(ip' y) ~ (19)
1+ IPI E= —~v(r), =0,
3 rsdrLtV(r))
for some E9 depending on yo only.
OI
Using similar methods, we may prove that the order
of integration over t and summation over k may be d dV (dv) (dv)
reversed in the expression for 8"&'&, and the analytic — r'ep +
e4A
+Ol I
& dr )
I
E dr )
I

properties of W&s& may be studied using (47) in the dr dr 90s'mp4c7

text.
Concerning the derivation of the expression (64) for Assuming the leading term to be a Coulomb potential,
)I, "'(y), we see that the contribution from 5's"' vanishes we immediately get the asymptotic form of the poten-
because of (12), (14) and (17). To find the contribution tial:
I
from W'i'" we may set = 0 in 2'i &s' in (3) and expand
V(r) =
2Zsns( ft q'
I+ol
(1)
— .
the denominators in powers of x. Integrating the double 4xeor 225' (mpcr) &r ) I

series term by term we get (64).

'

In deriving (65), we first note that Wi"' gives no This agrees with (60).
contribution because of (19). The constant comes from
APPENDIX IV
integrating X'"(y), and from 1Vs"'. To find the con-
tribution from 8
2'" we replace t and y by the new vari-
(a) We may define a function it (n; x), when I xl (1
ables (which may be justified): by the power series:
[1—t)[1+It)
rt= (20) it + —.
(I; x) =]gag
1+It [1+t(st—1)) QtL

We let ~pe and get We shall study only lt (2; x), which occurs in the ex-
pression for W&s&. We have the integral representation:
dyes"'(P; y) I' —
dt
f(2 i x) = — ~
in[1 —tx),
~l pl ~p t
8) ' k-c
00

dg I

J,
drt[1 2 —kP'rt' cos[y lnrt)
when lxl (1.
Introducing an appropriate cut in the
x-plane, we may continue lt, using the integral repre-
1
—— ksP'st* sin[y inst)+O'P"rts[lnst) —kerf' ss W. Heisenberg and H. Euler, Z. Physik 98, /14 (1936).
7 "H. Euler, Ann. Physik 26, 398 (1936).
 PA CUU M POLAR I ZATION I N STRONG COU LOM 8 F IELD

sentation. In particular we get the relations: Using these results, we may easily constru'ct the

P(2; 1 —x) =P(2; 1) —Llnx)Lln(1 —x)) —P(2; x),

P(2 —x)= —P(2; 1) —' ln'x —P~ —, 2; —— ~,

x)
table:

Z-~ ~=~
00
Z,
00
m(m+e)'
1
=l-(3),

~(2; -1)= —l~(2; 1), =-ll (4) -!r'(2),

~=~ ~=& m(m+n)'
P(2; -,') =-', P(2; 1) ——,' ln'(2),
e0 00 1
O(2; 1) =~(2). =!1'(2)-lt. (4),
o
n 1m=1 m2( m+ e)
(b) In deriving (66) and (67), we have to evaluate
certain double series. We shall tabulate the sums we 00 00 1
have evaluated, and indicate the method. Let 1 and p) =2K(5) —f(2)l(~),
q&1. Then:
=i m(m+e)' ,
00 00
00 00 1 1
+ m&(m+e)& =31 (2)t (3) (11—
/2)l (5),
-~ m&(m+e) &
~-& ~-& m'(m+e)'

=2
00 00

Z
=
—+
+1 ns~e~ ns~e~
=~(P)t(~)-1(p+q)
00

~=& ~=&
60

mo(m+e)o
= (9/2)l (5) —2l (2)f(3),

Let r ) 1 (r an integer). Then 00 ce 1

=-.&(6) ——.-V(3),
~=~ ~=~ m(m+e)'
~-r m=x me(m+e)" 00 00
= t'(3) (4/3) f—
(6),
'dx n=i ~-i m (m+e)
= ( —1)'L(r —1) g '
r
—Dn 'x)Eln'(1 x))
~p X 00 00

Z 2 =-:V(3)—-'0(6),
= ( —1)~'P(r —1)!) ' lim
n-1 m=1 ~3 ~
y -,' Ogyr —1
00 00 1

'pl
Z Z =f(2)f(4) 'l( )3—+ll'(6)
a=~ m=& m4(m+e)
X ~lim dxx" '(1 —x)'
a. ~,
a -~ r (y)r (1+.) n=l m=1
=l(p-1)-l8); »3,
m &
= ( —1)~'f(r —1)!) ' lim lim
Q
o:t, o ay~i .-e r (1y y+,)
=-:Ef(p —1) —~(p)); P&3
a=i m=i (m+e)~'
X
-=~-(e+~)' (e+~+y)'- The last two relations are trivial if one chosen (m+e)
1 1 as a new variable of summation.

.-i (e+ ~+y e+ei

To evaluate these sums numerically, we have used
the tables of Davies for the Riemann zeta function for
"
= (+1)l.(2+ ) Z~(2+-I) l. ( E)- integral arguments.
~ H. T. navies, Tables of Higher Mathematical Functions, Vol.
II, p. 244.