An Introduction to Variational Methods in Electromagnetic Scattering

Author(s): Ralph D. Kodis

Source: Journal of the Society for Industrial and Applied Mathematics, Vol. 2, No. 2 (Jun.,
1954), pp. 89-112
Published by: Society for Industrial and Applied Mathematics
Stable URL: http://www.jstor.org/stable/2098757
Accessed: 16-09-2016 22:09 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

Society for Industrial and Applied Mathematics is collaborating with JSTOR to digitize, preserve and
extend access to Journal of the Society for Industrial and Applied Mathematics

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
J. SOC. INDUST. APPL. MATH.
Vol. 2, No. 2, June, 1954
Printed in U.S.A.

AN INTRODUCTION TO VARIATIONAL METHODS IN

ELECTROMAGNETIC SCATTERING*

RALPH D. KODIS

1. Introduction. The theory of electromagnetic scattering may be

divided in a natural way into what have been called interior and exterior
problems, and these in turn may be considered either in the harmonic
time domain or from the viewpoint of transient behavior. This paper is
limited to those aspects of the theory which deal with harmonic fields ex-
terior to perfectly conducting scatterers, although in some circumstances
the extension to other configurations in both space and time is not dif-
ficult. Stated more precisely, the question is one of finding solutions to
the time-independent Maxwell equations for the electric and magnetic
field vectors subject to appropriate boundary conditions on the surface of
the scatterer and at infinity.
For the electromagnetic field, this solution can be carried out explicitly
only in a very few cases because of both the limited number of coordinate
systems in which the partial differential equation can be separated and
the vector nature of the fields. However, by methods analogous to those
which are well-known in the corresponding scalar theory a completely
general formal solution is possible in terms of an integral involving the
boundary values of one or the other of the field vectors. The problem is
thus transformed from one of determining the fields everywhere through
a partial differential equation to one of solving an integral equation for the
required boundary values. Unfortunately the problem is not noticeably
easier when stated in these terms.
There is, nevertheless, a considerable advantage in the integral equation
formulation since it enables one to construct stationary expressions for a
number of quantities of interest in diffraction theory. In this form the
first-order variation of, for example, the far-zone scattered amplitude is
zero with respect to similar variations of the surface distribution. Conse-
quently, good approximate results can be obtained whenever experience or
ingenuity suggests a good trial distribution. In some circumstances, when
the unknown distribution can be expressed as a series expansion with un-
determined coefficients, the variational principle provides a systematic
way of making an optimum determination of successive coefficients to what-
ever order of accuracy is deemed practical.
The theory of electromagnetic scattering has already been presented
from the variational viewpoint in connection with apertures in plane
screens [1]. To avoid repetition, the review which follows restates the

* Received by the editors April 19, 1954.

89

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 90 RALPH D. KODIS

problem in terms of the scattering by conducting obstacles. The discus-

sion begins with the construction of the tensor Green's function for free
space, the interpretation of its physical significance, and the development
of some of its properties. With this function and the vector analogue of
Green's second identity it becomes possible to find compact integral
formulae for the scattered fields. These integrals may be viewed as precise
statements of an electromagnetic Huyghens' principle which leads directly
to corresponding Kirchhoff approximations. Some simple examples are
then considered.
In the final sections of this article we consider a variational principle for
the scattered amplitude far from a conducting obstacle. It is shown in the
usual way that this amplitude obeys a reciprocity relation and that it is
connected to the total scattering cross section of the obstacle. The general
ideas that have been developed are then applied to a specific example.

2. The tensor Green's function. In scalar diffraction theory it is con-

venient to express the solution of the wave equation in terms of a scalar
Green's function which represents the contribution of a unit source at the
point P' to the scalar wave function at another point P. The correspond-
ing procedure in a vector theory is complicated by the fact that the direc-
tion of the vector wave function is not simply related to that of its sources.
Since all three components of the source vector contribute to each com-
ponent of the electromagnetic field, nine functional relationships between
the points P and P' are required. These relations are the components of
the tensor (dyadic) Green's function.
To find the required tensor representation we start with Maxwell's
equations in an isotropic homogeneous medium carrying a volume current
density J(r) which may have both impressed and induced components. In
M.K.S. units these equations are

V X E = icogH
V X H = -icoWE + J,

where the field vectors are functions only of position (harmonic time de-
pendence corresponding to Re [e-it] is implied here and in what follows).
It has been shown in standard treatises on electromagnetic theory [2]
that a formal solution of the field equations can be obtained by means of a
pair of auxiliary functions called vector and scalar potentials. The vector
potential, A, satisfies the inhomogeneous vector wave equation

( 72 + k2)A = - j (k2 =2C

where the Laplacian must be interpreted in general according to the
identity

V2A V 7-A- V X V XA.

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 91

If the vector potential can be found, the magnetic field follows from the
definition

4H = V XA,

and the electric field is then obtained from the second of Maxwell's equa-
tions.
An integral representation for the vector potential is possible in rec-
tangular coordinates because in this system the vector differential operator
reduces to the scalar Laplacian of each component. Thus the ith rectangu-
lar component of the vector wave equation is

(1) (72 + k2)Aj(r)= -,=Ji(r),

and the Helmholtz integral for this scalar equation is

(2) A i(r) = A f G(r, r')Ji(r') dV'.

The volume of integration includes all sources, and

G(rr') = exp(ik r - r' f)

4irjr- r'I
is the free space scalar Green's function. Since this function is a solution
of the inhomogeneous equation,

(3) ( 72 + k2)G(r, r')- -b(r -r')

the fact that (2) is a solution of (1) can be verified by direct substitution.
It is now a straightforward matter to construct the corresponding
Helmholtz integral for the electromagnetic field vectors. The electric
vector, for example, is given by

-icEE =-J +- V X V X A.

Writing out the x-component in detail, we have

-ioEEx(r) = -Jx(r) + ax fG(r, r')J, (r') dV' + axz fG(r, r')Jz(r') dV'

- 32 f G(r, r')Jx(r') dV' - a f G(r, r')Jx(r') dV'.

The form of this result is simplified by making use of the scalar wave
equation for G. This tells us that

(3e + &2)G(r, r') = -zG(r, r') - k2G(r, r') -(r - r').

Since the differential operators apply only to the unprimed coordinates,

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 92 RALPH D. KODIS

the order of integration and differentiation may be interchanged. Then

with

I(r - r')Jx(r') dV' =Jr),

it follows that

E (r) = i f G(r, r')J,(r') dV' + a, [(8xG)J, + (ayG)Jo

+ (, G)JJ] dV'.
In equivalent vector notation

x E(r) = iWI tfG(r, r')(x-J(r')) dV' + V) VG(r, r') -J(r')) dV'1

where x is a unit vector in the x-direction. Similar expressions apply for

the y and z components with the substitution of appropriate unit vectors.
Consequently the entire electric field can be written compactly in the
Helmholtz form

(4) E(r) = icoA If F(r, r')-J(r') dV',

where

(5) F(r, r' _7 (+ V )G(r, r')

and

I = xx + Y3Y + ZZ.

The nine components, Frjj of the tensor Green's function thus defined
give the contribution of the j-component of unit current density at r'
to the i-component of the field at r, and together they constitute the vector
counterpart of the free-space scalar Green's function.
Further properties of the tensor (dyadic) Green's function follow from
the vector wave equation for E which is obtained by eliminating the
H-vector from the field equations. The result of this manipulation is

(6) V X V X E-k2E = icowJ.

By substituting (4) in (6) we find that a sufficien

to be a solution of the inhomogeneous vector wave equation is for the
tensor Green's function to satisfy

(7) V X V X IC(r, r') - k2r(r, r') = 16(r - r').

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 93

It can further be shown that (5) does in fact satisfy (7) provided that
G(r, r') satisfies the inhomogeneous scalar wave equation (3).
With (7) and the vector analogue of Green's second identity it is now a
straightforward matter to obtain the vector counterpart to the symmetry
in r and r' of the scalar Green's function, The required identity applies to
two vector functions of position, A and B, which are well-behaved in a
volume V and on the surface S which bounds it. For such functions it is
easily shown that [3]

f [A V X v X B-B V X V X A] dV

(8)
= f -[B X V X A- A X > X B] dS,

where ni is the outward normal from V. Let us use two arbitrary

vectors b' and b" to make the identifications

A = ir(r, r')-b', B = r(r, r")-b".

Then the surface integral on the right of (8) becomes

I n.[r(rr").b" X VX 1r(r,r')sb' - r(r,r') b' X V X LI(rr")-b"IdS

in which the points designated by r' and r" are at finite, fixed dis
from the origin, whereas the volume V is taken to be large enough
on all points of S, I r I >> I r' 1, I r I >> I r" 1, and I r - oo.
Under these conditions the surface integral can be simplified somewhat
by making use of the explicit form of r and its curl. Since the operator
"curl grad" is identically zero,

F(r, r') *b' = ? V V) G(r, r') *b'

V X r(r, r') * b' = V X (G(r, r')b') = -b' X VG(r, r').

Now for very large r (= r 1) the scalar Green's function exhibits the
asymptotic forms
ikjIr-r' ikr
e e -i(A.'
G(r, r') = -r ' _rre
4 (- )r' 4rr
Vikr
e (-ik(r') ( )
4irr
ik r
AA2e -ik(r.r')
17VVG(r, r) '--'-fk2 e
4wr

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 94 RALPH D. KODIS

It follows that the tenso

behavior*
ikr

F(r, r').*b' - (I - rr) *b' eik(rr)

(A~ ~~ i r ik (Pr so).

V X r(r, r').b' r (r X b')ik

With these expressions the first term

2 ikr

r(r, r").b" X V X r(r, r').b' - ik e e ikr(rff+r )

(4 w2 )r2

- frrb" X r X b'),

where

(I - )b" X (r X b') = b" X (r X b') -(b11r X (r X b')

= (b'b") r- r ( r r.
It is evident that for large r the above combination is unaffected by the
interchange of primed and double-primed coordinates so that the two terms
of the surface integral are equal, and I = 0. As a consequence,

f [F(rr)b'.7 X V X r(r,r").b"

- r(r, r").b". V X V X r(r, r').b'] dV = 0

Replacing the V X V X r by its equivalent from the vector wave equa-
tion (7), we find that

/r [(r, r').b'.I b"6(r - r") - r(r, r").b".I b'b(r - r')] dV = 0.

So

b" r(r", r') .b' = b'. *r(r', r").*b".

Since b' and b" are arbitrary unit vectors, the last relation states that
the components of the dyad are equal to the corresponding components of
its transpose when the position variables are interchanged. This statement
constitutes the symmetry property of the tensor Green's function and
may be written

(11) F(r', r") = [F(r", r')]T.

* It should be noted in passing that this form, which r
of a unit source, has no component in the direction of the radius vector.

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 95

3. Scattering formulae for a conducting obstacle. In the problem we pro-

pose to examine, a plane harmonic electromagnetic wave is incident upon
a finite, perfectly conducting obstacle in free space. The incident wave
induces on the obstacle a surface current density which radiates a sec-
ondary or scattered field. In the region exterior to the obstacle this field
satisfies the source-free Maxwell equations and hence the homogeneous
vector wave equation, subject to appropriate boundary conditions. Spe-
cifically, we require for the electric vector a solution of the second-order
equation

V X V X E - k2E = 0,
where

E(r) = Einc(r) + EPc(r).

This solution must describe the field at all points in space exterior to the
volume V of the obstacle. In particular, the tangential component of the
total field at the surface of the perfectly conducting obstacle must vanish,
and as we shall see, the scattered field must obey certain conditions far
from the scatterer.
In order to find the field that meets these requirements let us construct
two vector functions of position*

A = E(r'), B = (r', r) .b

and apply Green's second vector identity (8) to them. The volume of
integration is V', exterior to V; it is bounded by S and by the surface
of a large sphere SO centered on the obstacle. With integration variables
corresponding to the primed coordinates, Green's identity gives

[E(r')*v' X V' X 17(r', r)b - b r(r', r)*b*V' X V' X E(r')] dV'

= | ti' [r(r', r).b X (V' X E(r')) -E(r') X (7' X r(r', r).b)] dS'.
+So~

When the differential equation satisfied by each vector is taken into

account, the volume integral reduces to

(12) f E(r')*I.bS(r - r') dV' = E(r)*b (r in V').

The integral over S can also be simplified immediately by interchanging

* The position vector of a point is directed from an arbitrary but fixed origin
in the vicinity of the obstacle to the point in question. The corresponding unit
vector denotes only a direction in space and so may be translated at will.

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 96 RALPH D. KODIS

vector and scalar products in the second term and imposing the boundary
condition on the tangential component of E. In this way

A.E(r') X (7' X r(r', r).b)

- (7' X r(r', r).b).(i' X E(r')) = 0 (r' on S).

The first term, on the other hand, contributes to the scattered field. With
the help of Maxwell's equation for curl E it takes the form

(13) fii'.[F(r', r) . b X icoptH(r')] dS' = iwm f Kr,,(r') .(r', r).b dS',

where

KA(r') = H(r') X i', (r' on S),

and the subscript indicates that excitation is in the direction r". This
integral, it will be noticed, has the same form as that found in the previous
section, K being the surface current density induced on S.
In order to deal with the integral over So, it is convenient to decompose
the electric field into its two parts, Einc and ESc, and to consider separately
the two resulting integrals,

Imb =ifn n [F(r', r)*b X (V' X Einc(r'))

0

-Einc(r') X ( V ' X 1r(r', r) * b)] dS'

18C = / n'[Fr(r', r)*b X (V' X ESC (r'))

-ESC (r') X (V'1 X r(rl, r) -b)] dS'.

The first of these is evidently the surface integral encountered when

Green's vector identity is applied to Einc(r') and r(r', r) b in an unoc-
cupied region of volume V + V'. The corresponding volume integral is
readily evaluated by the procedure that led to (11). Thus

(14) 1inc = V [Einc(r').7 ' X ' X rF(r', r).b

- r(r', r).b.V' X V' X Einc(r')] dVT = Einc(r).b.

The significance of ISC can best be appreciated by allowing the radius of

So to become large enough so that the asymptotic form of the Green's
function can be used. In the appropriate spherical coordinate system

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 97

(R, 0, p) the substitution of (10) gives

ISC / kR'.[(I - R'R').b X (V' X ESC(R')) + ikEsc(R

ikR'

erR e-ikR'r)R,2 sin 6' do' d+p'.

By straightforward manipulations involving double vector products, this

integral is readily reduced to

I 4- f R'fikEsc(R) -R' X V' X ESC(R')] .bt eikR eik(R'r) sin ' do' d+,

where b = b - (b.R')R' is the component of b transverse to R'. In this

form the integrand is recognized as the vector analogue of the Sommerfeld
radiation condition, and arguments similar to those invoked in the scalar
theory [4] require that

(15) limR,- [R X V X ESC(R) - ikEBc(R)] R = 0.

When (12)-(15) are put together, it is found that

(16) E(r) = Einc(r) + iw,.t Kvr') *(r', r) dS',

where the common factor b has been dropped since it is an arbitrary

constant vector. The corresponding solution for the magnetic field is
obtained by making use of the field equations; it is

(17) H(r) = Hinc(r) + V X f K-(r').F(r', r) dS'.

Alternatively, the symmetry condition (11) makes it possible to write

(18) H(r) = Hinc(r) + fK,,(r').V' X r(r', r) dS'.

These integral formulae express the scattered field exterior to a region

bounded by a surface S in terms of the boundary values of H on S, pro-
vided nosH = ni X E = 0. They are clearly analogous to the mathematical
form of Huyghens' principle in optics, and so suggest the possibility of
making calculations which correspond to the scalar Kirchhoff approxima-
tion. The vector formulation has the advantage of starting with integrals
that satisfy the field equations.

4. Kirchhoff scattering approximation. In the theory of optics, Kirchhoff

approximations are usually associated with the diffraction of light through

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 98 RALPH D. KODIS

holes in plane screens. Integrals of the type (16) for the scalar amplitude
1+ are evaluated by assuming that both i, and &4i/&n vanish on the shadow
side of the screen and have the unperturbed incident values in the aperture.
The analogous procedure for electromagnetic diffraction by apertures in
perfectly conducting screens consists in taking the surface current density
on the back of the screen to be zero, while assuming that on the illuminated
side nf X H is everywhere the same as it would be with an unperforated
screen. For thick obstacles a further extension of the idea is necessary.
The current density on the shadow side of the obstacle is taken to be zero
as before. On the illuminated side, however, the density at every point is
assigned the value it would have if it were induced on a perfectly conduct-
ing plane of infinite extent tangent to the obstacle at the point in question.
In other words, the obstacle surface is considered to be locally plane over
distances of the order of a few wavelengths, beyond which interactions
between current elements are assumed negligible. A current distribution
constructed in this way resembles the correct distribution very closely
except near the shadow boundary.
This kind of Kirchhoff approximation has a simple analytic form if the
incident wave is plane. For a wave of unit amplitude whose electric vector
is polarized in the direction e", we can write

Ein(r) = eik(?fer)

The field equations then show that

HAc(r) = ( V X EAjY=(r))h =-h eik(2er)

where e", h", and i" form a cartesian set of unit vectors at the point r.
When this wave is reflected from a plane conducting surface, the induced
current density is

KA,,(r= HA& (r') X A = -2A X HA,(r,)

= -k (n A fi)(ikr" *r')
CO/u cop

The high frequency scattered field, as given by formula (16), is then

(19) E,(r) = -2ik A (f X h") *(r', r)e k(rf*r') dS'

where S_ designates the illuminated portion of the obstacle surface (Fig. 1).
For most shapes the complete evaluation of (19) presents considerable
difficulty. However, some interesting information can be obtained by re-
stricting attention to the distant field. At very large distances, for which

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 99

r j >> I r' 1, the asymptotic form of the Green's f

the identity

(A X hf").(I- = -r X [r X (n X fi")],
leads to
ikr

(20) Esef(r) - 2ik er X i X X ^h"l)eik(r -r)r dS' (r oo).

The corresponding magnetic field follows from the fact that

KA,,(r'). *7' X F(r', r) = Kr,,(r') X V'G(r', r),

which reduces (18) to

2 ikr (f_
(21) Hscf(r) -2i _ corW
e r X |A( X Ri") e i rr)t dS' (r oo).
47r s_

These "geometrical optics" formulae for the electromagnetic field display

the properties that are characteristic of a radiating field far from its sources.

GEOMETR CAL

+ (~~S'

FIG. 1

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 100 RALPH D. KODIS

The field vectors, for exam

and the ratio of their ma
space (k2 = co2Eu). Moreover, the complex Poynting vector, E X H*, is
real, r-directed and has the correct 1 r2 dependence to ensure finite scat-
tered power. It should also be noticed that to this degree of approximation
the field amplitudes can be calculated for one direction in space without
specifying the shape of the obstacle. That is, if the direction of observation
r is the same as the direction of incidence r", the integral to be evaluated is

I = A" X f (f X h") dS' = f [(i"rh")i- (")h"] dS'.

Now i" and h" are orthogonal, while -r f) dS' is the projection of an
element of the area S_ on a plane perpendicular to the direction r" (Fig. 1
So the integral is simply I = Sph", where S, is the projected cross-sec-
tional area of the obstacle. Then
k2 ikr
H8/1r -2i k S S. h

(22) ikwr
ik i
E"'(r) - -2ik e S e"

This result provides us with two general facts about scattering at high
frequencies. In the first place there is no rotation of polarization between
scattered and incident waves along the direction of incidence. Secondly, the
total scattering cross section has a particularly simple value. For if the r
independent part of the distant electric field in (22) is called the amplitude
for forward scattering, we can write

ik ifi
A 27rSe

It will be shown at a later stage that the total scattering cross section is

= 47r
ImIMe A-A,

so that in the optical limit, a conducting obstacle of ar

is scattering an incident plane wave has a total cross section that is twice
its projected area,

0 = 2SP.

5. A variational principle for the electromagnetic scattering cross sec-

tion. Formulations like that of the previous section help to place optical
diffraction theory on a firmer foundation than it has hitherto enjoyed.

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 101

When the consequences are worked out, however, it is found that only a
limited amount of additional information can be obtained. Not only are
the Kirchhoff corrections to geometrical optics generally unsatisfactory,
but at the relatively low frequencies that are used in microwave optics
these approximations produce results which are highly inaccurate-even
misleading. In order to be able to work out useful approximations in the
frequency range between the static and optical limits a different approach
to the problem is required. One such approach is through the variational
principle of Levine and Schwinger.
To construct a variational principle for the vector wave equation we
need only the integral formula (16) for the field:

(16) E(r) = e eik("r) + i Kf (r) * I(r', r) dS'.

By making use of the asymptotic form of the Green's function

ikr

r(r, r') - (I - r) r e k(r r) (r -- co)

and the identity

Kr,,(r') (I-rr) = -r X r X Kr,,(r'),

the scattered field in the far zone is found to be

ikr

Es c (r) - r rJ4Le rXrX KA,, (r') ei(

47r r d

That part of ESC which is independent of r is called the scattered amplitude

and may be written

(23) A(i, i) = f Krl#(r')e-k(rr') ds'

47r

where it will be remembered r" is the direction of incidence and r is the

direction of observation. The evaluation of (23) depends on knowledge of
the current distribution on the obstacle which is determined by the re-
quirement that A X E = 0 on S. When this boundary condition is im-
posed, formula (16) gives the integral equation for K

(24) (A X &f)eik rr) = iw A Kr,(r') F r(r', r) X A dS' (r on S).

With the integral equation it is possible to construct a stationary ex-

pression for the vector scattered amplitude. We note that if the left mem-
ber of (24) is multiplied by a suitable current distribution and integrated,

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 102 RALPH D. KODIS

a form very much like (23) will result. With this in mind the integral equa-
tion may be vector premultiplied by the distribution Kp(r) with the result*

(rr)X (ft X (IF = -i f Kr(r)(F(r r) X nf) X K-(r) dS'.

Since the vectors K and nf are everywhere orthogonal, the double vector
products reduce to

K X (n X e") = (K.e") n and (r X ni) X K = (Knr)i.

On integrating over the obstacle surface we find that

e" *fr Kr(r)e dS = -ic. ff K (r')IF(r', r).Kr(r) dS dS'

(25)

= e * f KA#(r)eik(r) dS.

The last term follows from the fact that the interchange of r" and r leaves
the double integral unaffected and must therefore be a property of the left
hand side of (25) as well. Comparison with (23) establishes the reciprocity
condition between the scattered vector amplitudes which result when the
directions of observation and of incidence are interchanged, namely,

(26) e".A(ir, -e') = e'A(f", -r)

The stationary form of these far zone amplitudes can be constructed

from a proper combination of the integrals of (25). This results in the
homogeneous form

e7 Ar r!

-e f | (r)e dS - e" * KA (r)eik(rffr) dS

(27)
-ico, K| , K(r') - IF r) -KA(r) dS dS'
47r ^ ,, ^ ^ tt ~ ~ r

-. e *A(A,) A).

The functional (27) is readily shown to be an extremum with respect to

* It should be emphasized that in all subsequent integrals the unit vector

r (whether in subscript or exponential) denotes an arbitrary but fixed direction of
incidence; the vector r on the other hand designates a variable point on the surface
of integration.

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 103

small independent variations of the distributions Kr and K-, . For example,

the variation with respect to K? is

A= - dS SK?(r)* [&~eik e r) + ico f K?(r')IF(r', r) dS'].

Since MKA is a vector tangential to S the stationary property is proved sub-

ject to the necessary and sufficient condition that the tangential com-
ponent of the bracketted quantity is zero. This condition is satisfied, of
course, when K?", is the solution of the integral equation (24).
Before trying to use (27) for calculations it is advantageous to render
the stationary expression scale invariant. This may be accomplished by
making scale transformations in the current distributions. Thus, if we
write aK-,, for Kis" and bKW for KA , we find that

(28) - e*A(", -A = al"- bI - iwlabJ,

where I, I" and J are identified with the appropriate integrals in (27).
The variations of A with respect to independent variations in the scale
factors must vanish, so

a -. and b=

Substitution of these extremizing values for the scale factors in (28) gives

-4ire* A(i, A-A) -

- (e f Kr(r)eix) dS) (e" f Kr(r)eik(rr) ds)

(29)- ____________________
Jf KAN(r') * r(r', r) .KA(r) dS dS'

- 47ret A Ar,-r".

This form of the scattered amplitude is useful in practical calculations since

it is both stationary and independent of the scale of the current distribu-
tions.
The computation of polar scattering diagrams from (29) is a difficult
numerical task at best, although it can be carried out in a few special
cases. A simpler quantity of considerable interest is the total scattering
cross section of the obstacle, which is defined as the total scattered power
per unit incident power. That is
Total scattered power
Power incident per unit area

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 104 RALPH D. KODIS

One obvious way to calculate the scattered power is to integrate the

normal component of the Poynting vector Ssc over the surface So of a very
large sphere which encloses the obstacle. It is evident that such a pro-
cedure would encounter formidable mathematical difficulties since it in-
volves the integration of I A 12, which is itself a complicated integral.
Fortunately, however, the total scattered power can be related to the
scattered amplitude in a much simpler way. Since there are no sources or
sinks in the volume V bounded by S and So, the divergence of Ssc is zero
everywhere in the region. So one can write

L (VSsc) dV = A(nSes) dS + I (n SsC) dS = 0.

The last integral is the scattered power, and we find that

psc = 12 Re f (n Ssc) dS

= Ref no (Esc X H *) dS,

where n' is now directed out of the scatterer rather than out of the volume V.
In order to see how this integral is related to the amplitude A, the
scattered fields are expressed as the difference between the total and the
incident fields. The normal component of the complex Poynting vector
then becomes

n i(Esc X Hsc) = nV[(E X H*) - (E X Hinc*)

+ (Einc X Hinc*) - (Einc X H*)].
By rearranging the vector products and imposing the boundary condition
that nL X E = 0 on S, the first two terms are seen to vanish, and

psc = 2 Re [f n (Einc X Hinc*) dS - f E'nc. ( X H*) dS].

Now the incident power flow through a closed surface is zero, so the first
term gives no contribution. Remembering that for a plane wave incident
in the direction r" the current induced on S is K-y' = H X ni, we can trans-
form the only remaining integral to obtain

Psc = I Ref e.K* (r)eik(rf r) dS

2 r
=2Re -. e" -A* r" r"

= - Im e".A(", r") .
coyL

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 105

The scattering cross section follows from the fact that the power incident
per unit area in a plane normal to the direction of propagation of a plane
wave is kl/2w,, so that in the direction F"

(30) (r" T Im eA" A(r^" r")

Substituting from (29), we find that

(e31 . = O(, (r) e ik rr ) dS)( A. JK,, (r)e-ik (rr) dS)

(31) far" -k I
toK~r (r') * r (r', r) * Krt

6. Application of the variational principle. A number of interesting and

instructive applications are given in the literature for variational principles
similar to (31). The example that has been chosen to conclude this review
indicates how the basic ideas of the previous sections can be extended to
more complex configurations and has the additional advantage of yielding
non-trivial results for very simple trial functions.
Let us consider, then, a pair of finite obstacles with perfectly conducting
surfaces Si and S2. The application of Green's vector identity to the
electric vector and the free space dyadic Green's function in the region
exterior to the obstacles gives, as in section 3, eq. (16),

E(r) = Ein(r)

(32) + iwnk [f K-, (r') r (rl, r) dS' + f K,(r')or (r'2,

The angular dependence of the far zone field is again found by making
use of the asymptotic form of the free space tensor Green's function
ikr

r(r, r ' ( Irr)

47r
- e ( (r soo

FIG. 2

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 106 RALPH D. KODIS

Then if the scattered amplitude is defined as

(33) A(r", r) = "-n [L KrN(ri)eik(rr{) dSl + 2 KA,,(r')eik(rr2) dsj,

the far zone scattered field is
ikr

EAr,,r - rX r X A(r", r) (r -- > oo )

r

The two current distributions, K(r1) and K(r2) are to be determined

from the pair of integral equations which result when the boundary con-
ditions are imposed. These conditions are

A X E(r1) = n X E(r2) = 0.

The integral equations are therefore

(ni X ,)eik(r~r1) = -iywln X [f Kv (r')*(rl, ri) dS'

() S K,,(r'2).r(r'2, ri) dSj2

(34) 2
(^ X tt~
L 1

KA, t (r') r (r' r2) d

To construct the scattered amplitude from these integral equations, each

equation is premultiplied vectorially with the current distribution induced
on the surface in question when it is illuminated by a plane wave propa-
gating in the direction rA. Integrating over the appropriate surface and
adding the results, we find that

A" [f K e(r eik(rr1) dS1 + (Kr(r2)eik(r1er2) dS2]

= -iwi [ff KA,,(rl) * r(rl, ri) * KA (rj) dS, dS

+ tt Kv (rl) r(r'1, r2) .KA(r2) dS2 dSl

(35) S
+2S1
ff KA,,(r') r

+ ffs Kp (r>)'r(r', r2) KA(r2) dS2 dSj

4r2 2 r2
=. r e"A*
-A( A~,-r) =
.r. r e*A~rr
, .

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 107

The stationary expression for the scattered amplitude can now be con-
structed by an appropriate combination of these forms. First we introduce
the shorthand notation

= b". f K (rj)eik(rr) dSj ; I= . (rj)e dSj;

(36)
Jjm = ff Kr (r') * r (r4, rm) * Kr(rm) dS; dSm.
SiSm

Then,

_47r A A
- . -eA(r ) =-(Il + I2 + I'1' + I2')
(37)
-iWM(J11 + J12 + J21 + J22) = - r-

That (37) is the desired stationary form may be verified by carrying out
independent variations of the current distributions.
Finally, it is possible to obtain a scale-independent formula by making
the scale transformations

K-(r1) - aKA(r1) Kr(r2) - cK-(r2)

KA,,r l WArx (ri) Krt-(r2) dlKr (r2) .

Since the amplitude is an extremum, the coefficients are determined by

the four conditions

47r aA 11
-2-= bJl + dJ21 +. = 0
W28 A2 a iWA

2 2--bJ12O+ dJ22 + .2 =

2 aJ11 + CJ12 + .=

47r&aA
---2 = aJ21 + CJ22 + 2 = 0.
CW A2 Td iWIA
Only a pair of these is required to obtain the final result. Thus

1 I2'J12 - I1J22 1 I'1'J21 -I2J

iWA J11 J22 - J12 J21 iW Jll J22 - J12
and since

-. e r- = -ali - C2

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 108 RALPH D. KODIS

the scale-independent, stationary form is

(38) -Ad e A(ir" - i) = 1IiJ22 - 1I2'J12 - 12'1'J21 + 12121J11

J11 J22 - J12 J21

The scattering cross section again is obtained from the general formula

( r" = k: Ime eA~r r"

It is of interest to note that while (38) is considerably more cumbersome
than the corresponding result for a single obstacle, this is the price that
must be paid for using the simple free-space Green's function. On the
other hand, the formal result for the scattered amplitude from any number
of obstacles can always be reduced to (29) at the expense of using a more
involved r. Thus if there are m obstacles, the application of Green's
theorem gives a sum of m integrals of the form

kcon J K-,, (r') r (r', r) -b dSj.

By choosing a tensor Green's function that vanishes on those obstacle

surfaces for which j 5 m, the sum is reduced to a single integral over the
mth obstacle, and the scattered amplitude carn be written in the simple
form of (29)

- 47r6,*A~ -r ) = __

Here I and I" are integrals over the mth obstacle identical in form
for one obstacle, but

J = ffm Kv (r') rm(r) K- (rm) dSm dSM.

Although the formalism is as simple as it is in the case of a single obstacle,

the difficult task now is to find the Green's function rm which not only
satisfies the radiation condition but also vanishes on all scattering surfaces
save one.
As an example of the above theory let us take the case of a pair of
parallel conducting cylinders of infinite extent in the z-direction. The
origin of coordinates is taken at the point midway between cylinder
centers (Fig. 3). If the incident wave is polarized in the z-direction (e" =
and propagates in the x-direction (A = x), none of the field quantitites is
a function of z. It follows from Maxwell's equations that the magnetic
vector can have only transverse (to z) components so that the currents
induced on the cylindrical surfaces must be longitudinally directed and

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 109

Ate ( aO 2 -d_)
d

_~~~~ -- __ / ~ ~~/

FIG. 3

constant in z. On substituting K = I K I I in (36), we find that the non-

trivial longitudinal integrals all have the form

ff ZF r(r,, rm)Z dz dz' = L - d d G(r' , rm) dz dz'

By making use of the transform of the scalar Green's function, [51,

1 fff+ ei[kz-x)+ky(y-y')+k,(z-z)]
G(r, r') = JJJ-G0 k2 k2- k2 k2

it can be shown that the result of the integration is the Green's function
for two dimensions,

G(Q, V') = -

where H ") is the Hankel function of the f

vector in the polar coordinates of the xy-plane
With these specializations, the integrals of

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 110 RA LPH D. KODIS

tude in the direction of incidence are (per unit length of cylinder)

2 r 2 r

Ii = aj 1 K-(aj)e doj j = aj = I(aj)e ik hi doj;

* 2 r

im= ff K(a')HW"(k a; - am )K-i

where

aj cos ej = x rj = x* ?yd+ aj + z).

In order to proceed further with these integrals it is necessary to select

distribution functions for the current densities, and as a first approxima-
tion they may be taken to be constants. It should be noted that these
constants need not be the same since a, 5 a2, and in fact they turn out
be rather complicated functions of both radii. However, since (38) is scale
independent, this is of no importance for our calculations and we may
consider that all the densities are equal to unity. Then
27r

= al f eika1coso1 d4l = 2raJo(kai) =

27r

12 = a2 eka2COS2 d2 = 2ra2Jo(ka2) = I2

ia 2 2 o
(39) Jil = f H ("(k a, - a' ) do, do =ir a, Jo(kal)H"'(kai)
. 2 2 r

J22 = 2 f H(1"(k I a2 - a j) d/2 do' = izr2a2

*a a ,27r
J12 = aa2 H(1)(k I a,-a' I do1 d' = 21
For j = k the double integrals are evaluated with the help of the addition
theorem for Hankel functions,

H(l)(k0]- M~~~~~0
& I) = ZJm(kp)H(l)(kp')e`m(O'), (I p),
and the orthogonality relation for exponentials.
The evaluation of the final integral, J12, is a little more difficult since
it involves the distance between points on the circumferences of different
cylinders. An appropriate addition theorem can be found by referring to
Fig. 3. The usual addition theorem gives

a, - a2 I) = EJm(kb)H(l)(kd)eim.

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 VARIATIONAL METHODS IN ELECTROMAGNETIC SCATTERING 111

Summing the angles of the triangle efg we find that

+ = J - a -2 ,

where t02 iS the angle between a2 and the negative y-axis. In terms of these
angles
00

Ho (k I a - a2 I) = Ee-imae'mrJm(kb)H(l)(kd)Cima2
-oo

But

e eim7Jm(kb) = ei Jm(kb)
oo

- EJ,(ka,)J._m(ka2)e"8(02-01
-oo

o ) (k a-a1 - a2 I) = JZZ (ka)J,(m(ka2)H)(kd)e-intYlei(n-m) 2.

so that

-oo -oo

Since tj = Oj + ir/2, the angle integrals

n = m = 0 and are zero otherwise. Therefore

(40) J12 = J21 = i7r2aaa2Jo(kai)Jo(ka2)H0'1)(kd).

When the results (39) and (40) are put into formula (38), it is found that

47r
= - Im 2.A(xI)

(41) Jo(kaD)H1)'(ka2) + Jo(ka2)H1)'(kai)

4 - -2Jo(ka_)Jo(ka2)H__)(kd)
=- IM i -
k Im - H1) (ka1
The above scattering cross section can be simplified further on physical
grounds. For a pair of cylinders a current distribution which is independent
of angle is a good approximation at long wavelengths (small radii) and
large separations, for which kd >> 1 and ka1, ka2 << 1. Under these con-
ditions Hol) (kd) -> 0 and Hol)(kal), H1) (ka2) -> - oo, and the denom-
inator of (41) can be expanded in a power series of the small quantity
Jo(kai)Jo(ka2)[H0(1 (kd)]2/Ho') (kai)H ') (ka2). Retaining terms up to

0 { [Hj(1) (kd)]2/UH (1) (ka)H (J1) (ka2) 12

we find that

4 4m Jo(ka ) Jo(ka2) Jo(kai)Jo(ka2)H0 )(kd)

k LH~H ((kai) + Ho' (ka2) H2N)(kaH jH(1(ka2)

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms
 112 RALPH D. KODIS

=4 r >?(kaj)
7_J2(kaj) +N2(kai) Jo
-2( ? i
-2 Im 2iJ(a)
I Jo (
Ho1) (ka,)H(1
To this degree of approximation th
vidual low-frequency cross section
action term which applies to small
enough so that the distribution of
from that of an isolated obstacle. F

I Jo/No I << 1

so that the correction can be written

?c -2 Im iJo(kaD)Jo(ka2)HO
- NO(kai) NO(ka2)

2 Jo (kai)Jo(ka2)JO(kd)
No(kai)No(ka2)

The correction is thus seen to disappear like the reciprocal of a logarithm

when either radius is vanishingly small, while for fixed radii and large
separations it has the asymptotic behavior (kd)-I cos (kd -/4).

This paper was written at the suggestion of Dr. Harold Levine, and I
am happy to take this opportunity to thank him for his encouragement
and helpful discussions.

REFERENCES

1. H. LEVINE and J. SCHWINGER, On the Theory of Electromagnetic

by an Aperture in an Infinite Plane Conducting Screen, Comm
Pure and Applied Mathematics, Vol. 3, No. 4, 1950.
2. J. A. STRATTON, Electromagnetic Theory, McGraw-Hill (1941), page 23.
3. Ibid. page 464.
4. Ibid. page 486.
5. H. LEVINE and J. SCHWINGER, On the Theory of Diffraction by an Apertur
Infinite Plane Screen, Phys. Rev., Vol. 74 (1948) pp. 958-974.

HARVARD UNIVERSITY

This content downloaded from 128.123.44.23 on Fri, 16 Sep 2016 22:09:29 UTC
All use subject to http://about.jstor.org/terms