JOSEPH F. BIRD and RICHARD A.

FARRELL

ELECTROMAGNETIC SCATTERING THEORY

Various theoretical investigations of electromagnetic scattering are reviewed that illustrate the in-
terplay between basic researches and application needs. Topics covered include stochastic variational
techniques for vector wave scattering by random systems of general electromagnetic properties, such
as rough surfaces or chaff clouds; the development of simple but effective trial functions for varia-
tional usage; and the treatment of electromagnetic induction and color vision as vector field scatter-
ing problems.

INTRODUCTION ous media and discrete bodies, all possessing a vari-

ety of material properties. Such "scattering," in a gen-
The purpose of this article is to survey selected eral sense, includes wave reflection, refraction, diffrac-
aspects of the basic research program in electromag- tion, or absorption phenomena, as well as more slow-
netic scattering theory that is conducted by the Theo- ly varying inductive interactions. In general, the elec-
retical Problems Group of APL's Milton S. Eisen- tromagnetic field varying in space and time can be
hower Research Center. In the process, we shall indi- Fourier analyzed into elemental waves over a spectrum
cate the interrelations between this research program of frequencies. At frequencies below those at which
and mission requirements in other departments of quantum effects dominate, the classical electromag-
APL. In general, the interconnections come about netic field theory usually is adequate to describe scat-
through collaboration with another department in one tering phenomena. This classical regime can extend
of two ways. A technical problem may be uncovered over 15 decades of frequency in hertz-from the opti-
whose solution requires the development of fundamen- cal through microwave and radio bands down through
tal sciences, e.g., the needs for a broadband method quasistationary induction to the zero-frequency static
of analysis of radar scatter by the sea surface and for limit.
investigation of hydromagnetic signals in the ocean. The Group has pursued a long-standing basic re-
Or a problem may arise whose solution involves sci- search program in the theory of electromagnetic wave
entific techniques developed in other applications, e.g., scattering. Since the electromagnetic field is a vector,
the analyses of the magnetic suspension in satellite dis- vector-wave theory is used, but special cases of scalar
turbance compensation systems (DISCOS) and of ex- waves also relevant to acoustics have been considered.
periments in color vision. Naturally, much effort has focused on the electromag-
The propagation, scattering, and absorption of elec- netic wave scattering because of its greater difficulty
tromagnetic and other waves provide potential tools and the importance of optical, radar, and induction
for probing various media, such as the ocean's sur- technology to APL missions. Classical methods of vec-
face, particulate matter in the ocean, chaff (and oth- tor field theory are used, with emphasis on the power-
er obscurants), aerosols, bubbles, and military targets. ful variational technique. 1
These phenomena have been important to the tasks Classical field theories are typically expressed in
of APL since its inception, for the specification, de- terms of partial differential equations with associated
sign, and use of such systems as radar, sonar, al- boundary conditions. Alternatively, the method of
timeters, and radio communication links. Also, from Green's functions can be employed to obtain an in-
time to time technologic problems in other areas arise tegral representation for the field. In the latter, a scat-
at APL or in other Johns Hopkins Divisions, e.g., cor- tered field is expressed in terms of integrals that depend
neal light scattering, color vision analysis, and elec- on the unknown field on and within the scatterers. An
tromagnetic induction effects, which in our experience efficient method of solution is to use physical intui-
can be attacked fruitfully by vector-field wave scat- tion to obtain trial approximations for the fields at the
tering techniques. scatterers and then calculate the scattered field by eval-
Surveying a representative sample of such researches uating the integrals. Of course, any errors in the cho-
entails a rather lengthy article, and its theoretical char- sen trial fields will generally result in errors of the same
acter involves considerable mathematics. However, the relative order in the calculated scattered field.
most rigorous section (Eqs. 6 through 22) may be The Schwinger variational principle 2 is an extreme-
skimmed or skipped without impaired understanding ly important contribution in that it recasts the integral
of subsequent sections. representation into an invariant form, that is, an ex-
pression in which first-order errors in the trial fields
OVERVIEW lead only to second-order errors in the calculated scat-
Many macroscopic phenomena are manifestations tered field. Thus, this variational principle enables one
of electromagnetic field interactions with inhomogene- to improve existing approximations. It also facilitates

58 Johns Hopkin s APL Technical Digest, Volume 7, Number J (1986)

------------------------------------------------------------------------------ ----

the design of new trial approximations that are effec- (a)

tive and yet simple. However, the absence of first- Vertical scale
order errors in the variational formulation is obtained
at the cost of replacing a single integral by a quotient
that involves the product of two such integrals divid- -,' ,-- .... ...... ,~--,
.,
Horizontal scale
..... -
ed by a third (more complicated) integral. Thus, al-
though the method has been applied to a variety of (b)
problems,I-8 only limited application 9, l0 was made
relevant to stochastic scattering problems, i.e., random
ensembles of scatterers such as the ocean's surface or
rain clouds. 11 The apparent need to evaluate statisti-
cal averages of the quotient of integrals would certainly
discourage such application.
The development of variational techniques as a cal- (c)
culational tool for stochastic scattering problems has 5r-~--.---.--.--_r--~~~~--~ __
been a central goal of the Group's program. A break- c:
o
through was achieved when we obtained an alterna- 't; 4
~
tive invariant stochastic formulation. 12 Specifically,
when the integrals appearing in the above-mentioned ~
(,)
3
quotient are first averaged and then the quotient is tak- "0

en, one has an invariant expression for the averaged .~ 2

scattered field. This form is inherently simpler to evalu-
ate than the average of the quotient, yet is variation-
ally equivalent to it. As will be discussed, the research
effort so far has culminated in a general stochastic 7 8 9 10
variational principle (SVP) for electromagnetic scat- ka
tering by arbitrary random distributions of scatterers Figure 1-(a) Schematic of rough-surface element. (b) Ray-
leigh model: Circular cylindrical boss (radius a) on a perfect-
with any linear electric and magnetic properties. ly conducting plane with a plane-wave (vector k i ) incident at
These wave-scattering studies, initiated in support the angle 0i and scattering into directions Os (defining a scat-
of the submarine security program by R. W. Hart, tered wave vector k s ). (c) Relative scattering cross sections
then chairman of the Research Center, led first to a as functions of the size/wavelength parameter (ka) for the Ray-
leigh model above for 0i = Os = 0 and horizontal polariza-
scalar form of the SVP. 12 The motive for this re- tion. Curves denote results for perturbation (P) and Kirchhoff
search was to facilitate analysis of radar returns from (K) approximations, for the variational improvement (V) of the
sea-surface "scars" left by submarines. Because the latter, and for the exact solution (E). Note that whereas P and
scars contained roughness scales of the order of radar K depart from Eat ka ~ 1 and ka !5 3, respectively, the varia-
wavelength (as well as larger and smaller scales), it was tional curve V follows E from large ka nicely through the gap
desirable to improve on standard "two-scale" the-
(ka - 3 to 1) down to ka = O. Even more accurate broadband
results have been achieved with novel, yet simple trial func-
ory.13 That analysis uses a long-wavelength (pertur- tions, as illustrated in Fig. 8.
bation) approximation for small vertical roughnesses
and a short wavelength (Kirchhoff) approximation for
large horizontal scales (Fig. la). This leaves untreat- sional tensor) formulation is desirable because it al-
ed the roughness features comparable to wavelength lows one to account for the anisotropy of the material
in both height and breadth. Thus there was a need to properties through which an electromagnetic field vec-
develop a fundamental theory that remained valid tor in one direction can produce polarization, mag-
through the transition region. This central omission ' netization, or current vectors in other directions. The
has received prominent attention only very recent- dyadic formalism also facilitates the Green's function
ly,14-17 though still with limitations (e.g., neglect of solution of the governing vector equations. Test ap-
multiple scattering). In contrast, early tests of the sca- plication has shown that the SVP can account for
lar SVP showed that it can account largely for multi- polarization, interference, and multiple-scattering
ple scattering well into the transition regime. 18-21 effects-even with simple trial functions that do not
Subsequent work has been directed toward develop- yield these effects in standard noninvariant calcula-
ing and testing the SVP as a calculational tool. This tions. 25 ,26 The tests used trial approximations valid at
has been accomplished largely in collaboration with long wavelengths and showed that the variational tech-
the Fleet Systems Department. 19-29 It formed the ba- nique extended their range of validity to include rough-
sis for a Ph.D. thesis and a post-doctoral research proj- ness scales up to the order of the wavelength.
ect of staff members now in that department. The Recent effort has focused on trial-function selection
work has produced a dyadic SVP that is applicable to and design to achieve accurate but tractable broadband
vector wave scatterers of arbitrary inhomogeneous and scattering calculation. 27-3o The initial work again con-
anisotropic permittivity, conductivity, and permeabil- cerned long wavelengths. 27 Then, as a next step to-
ity, possessing arbitrary random characteristics. The ward variational improvement of "two-scale" theory,
dyadic (i.e., vectorvector, or second-rank three-dimen- variational treatment of a short wavelength (Kirchhoff)

John s Hopkin s APL Technical Digest, Volume 7, Number 1 (1986) 59

Bird, Farrell - Electromagnetic Scattering Theory

trial field was carried out for a test surface. As illus- age to corneal cells. These studies of electromagnetic
trated in Fig. la, a rough surface is described by ver- interactions with biological tissue were surveyed recent-
tical deviation as a function of horizontal dimensions, ly in the Johns Hopkins APL Technical Digest, 31 and
which includes height, slopes, curvatures, and higher extensions of the infrared studies are described in the
variations. A tractable test model for analysis is Ray- article by McCally et al. elsewhere in this issue. In an-
leigh's classic surface-scattering model, a circular other biomedical effort with the Johns Hopkins Med-
hemicylindrical boss on a perfectly conducting plane, ical Institutions, early APL vision research eventuated
as depicted in Fig. lb. For that model, all the vertical- in a general vector field theory of vision, considered
horizontal variation scales are represented just by the as the "scattering" process first defined and investigat-
boss radius, a. Thus a unique size/wavelength ratio ed by Maxwell: electromagnetic field absorptions at
parameterizes the model, viz., ka, where k = 27r/A the retina transmuting into a brightness-color spatio-
is the wavenumber of the incident radiation of wave- temporal sensation field. 32 ,33 Concomitant electro-
length A. More importantly, the model has a tracta- physiological studies with APL's Biomedical Programs
ble exact solution to which we may compare different Office suggested a novel electroretinographic analy-
approximate solutions. This is important because it is sis technique. 34
the only ironclad way to assess the accuracy of approx- Finally, various problems in electromagnetic induc-
imation methods. Such a comparison is displayed in tion that have concerned different departments of APL
Fig. Ic for the initial case we treated-backscatter of have been attacked via vector-field methods with con-
normally incident, horizontally polarized waves. One siderable success. Thus, analyses of hydro magnetism
sees from this graph (with the exact solution, curve induced by ocean motion across the geomagnetic field,
E, as benchmark) that the long-wave perturbation re- initiated for the SSBN security program, yielded many
sult (curve P) is satisfactory for small ka but fails badly useful numerical estimates, proved a new eddy the-
for ka ~ I, while the short-wave Kirchhoff result orem, identified the phenomenon of sonomagnetic
(curve K) may be acceptable for large ka but fails badly pseudowaves (hydro magnetic fields propagated via
at ka $ 2. In contrast, the variational result (curve acoustic radiation), and determined their scattering by
V) shows good agreement at all wavelengths, i.e., not the air-sea surface. 35 -38 For the Space Department, lev-
only from high frequencies into the transition region, itational force-torque in the magnetic suspension of
as hoped, but all the way through to the low-frequency the disturbance compensation system (DISCOS), now
(Rayleigh) limit. 28 (Other examples of this all- deployed in NOVA satellites, was analyzed to aid in
frequency variational improvement at different angles malfunction diagnosis and design optimization for
and polarization are shown in Figs. 6 and 7.) DISCOS. The power of vector field analysis is exem-
Although detailed analysis for this model over all plified by the fact that the analysis produced both sig-
polarizations and angles showed that the all-frequency nificant practical design simplification and new con-
improvement was not fully satisfactory, 30 these tributions to magnetic levitation theory. 39-41
studies encouraged subsequent attempts to design new
trial functions for accurate broadband scattering cal- STOCHASTIC VARIATIONAL PRINCIPLES
culations. 28 ,29 For perfect conductors, we modified General features of variational principles that make
the long-wavelength trial fields so that they became them invaluable as calculation tools are reviewed brief-
capable of satisfying the boundary conditions. At least ly in this section. The generic form of the Schwinger-
for simple test models, this approach has yielded ex- type variational principle is discussed, without mak-
tremely accurate all-frequency variational results. 29 ing explicit the integrals that compose it; these are giv-
(An example is displayed in Fig. 8.) Presently, in col- en in the next section. After pointing out the difficulty
laboration with the Space Department, similar meth- of applying the Schwinger form to stochastic scat-
ods are being tried on models for ocean surface radar terers, the APL-developed stochastic version of this
scatter. Future studies will be directed at surfaces with principle is presented and discussed.
arbitrary electromagnetic properties, where the fields Many physical problems can be expressed in terms
penetrate the surface. of a set of field functions, ir, that obey certain field
Most of this article will be devoted to elaborating equations. Often, primary interest is not so much in
on the SVP theory and its applications. However, we accurate calculations of ir, but in some related quan-
also present brief descriptions of some of the other tity, F(e.g., scattering cross section), that can be writ-
electromagnetic scattering research that the Group has ten as a functional of the field functions, F(ir) , in
done in collaboration with colleagues from other APL various ways. A particular functional is said to be a
departments or from other divisions of Johns Hop- variational principle if it is invariant for ir approach-
kins. In long-standing biomedical research collabora- =
ing the exact solution irE, i.e., if aF(ir) F(irE + air)
tion with the Johns Hopkins Medical Institutions, a - F(irE) = 0 to first-order in small variations, air.
major eye corneal structure program has investigated When F is formulated such that the invariance (Eul-
both visible light scattering and infrared absorption er) equations reproduce the original field equations,
in the cornea. An intricate interplay between experi- aF = 0 represents a compact statement equivalent to
ment and theory led to the development of light scat- the original problem. Further, since the form of F is
tering as a tool to probe the ultrastructure of the chosen such that F(irE) represents a desired result in
cornea and yielded an explanation of infrared dam- the problem, the variational principle offers a power-

60 fohn s Hopkin s APL Technical Digest, Volume 7, Number 1 (1986)

 Bird, Farrell - Electromagnetic Scattering Theory

ful method of approximately calculating that result tegrand contains the fields and the Green's function
from reasonably accurate approximate values of the [G(r,r')] appropriate to the scattering problem.
field if. The variational method may be used either The form of Eq. 2 shows that variational invariance
to improve an existing approximation or to fashion is obtained at the extra cost of evaluating the integrals
efficient new trial approximations. In either case, by N2 and D, once the noninvariant solution N} is
virtue of the variational invariance, small errors in the known. Because ;j; is the solution to the adjoint prob-
trial functions for the field if lead to much (quadrati- lem in which ks +-+ -k;, it is convenient to use trial
cally) smaller errors in the calculated quantity, F. In- fields for the reciprocal problem, ;j;1, that are ob-
deed, judicious selection of trial fields if such that they tained from the trial field for the original problem,
can mimic exact behavior where important, viz., in F, 1/;1, also by the transformation ks +-+ -k;. In that case,
no matter how poorly they may behave elsewhere, will not only is it true that N2 (;j;) = N} (1/;), but also that
be seen to yield efficient and accurate approximations. N2(;j;/) = Nt (1/;/), so that the only additional cost of
An attractive form of variational principle is the invariance is to evaluate D. Several specific examples
Schwinger type, which is a ratio of functionals that of invariant formulations of scalar problems are giv-
is independent of the overall amplitude of the trial en in Ref. 1 (pp. 1128-1134), and a few are presented
function. For example, consider scalar wave problems in the next section.
consisting of a plane wave of amplitude A, frequency Vector wave (electromagnetic) scattering has a vari-
w, and wave vector k;, viz., A exp[i(k; -r - wt)], in- ational principle of the same generic form as Eq. 2,
teracting with a scatterer to produce a superposition with 1/; now the electromagnetic field represented by E:
of harmonic waves, say 1/;(r) exp(-iwt). The field dis-
tribution near the scatterer may be quite intricate, but
interest often centers on the far field (r ~ scatterer
size), where the scattered waves reduce to a spherical where T is any component of the vector scatter am-
wave whose amplitude varies with scattering direction plitude. The adjoint field E and the integrals Nt , N 2,
denoted by unit vector ks • Thus, the field 1/;(r) be- and D are identified in the next section. Here we only
comes asymptotically the sum of incident plane and note that the integrals are more complex than in the
scattered spherical fields, scalar case because their integrands involve dyadic
operations on the vector E,E field. Like the scalar wave
1/;(r) ::::: A exp(ik; -r) + T(ks ,k;) - A exp(ikr)/r , (1) principle (Eq. 2), the vector wave principle (Eq. 3) was
initially derived for deterministic scatterers in which
where k denotes wave vector magnitude Ik; I, and ks the scatterers have no element of randomness. Of
== k ks • The key ingredient in Eq. 1 is T, the relative course, there are important applications (such as sea
amplitude of the scattered wave, whose determination surface or chaff cloud scattering) in which the scat-
represents a complete solution for the far-field scat- tering system is necessarily treated as a random ensem-
tering. For example, the differential cross section is ble of scatterers (of sea waves or chaff particles), i.e.,
given by 1112. as a stochastic scatterer.
The Schwinger variational principle for scatter am- Stochastic scattering presented a seemingly insuper-
plitude T is obtained by considering the adjoint field able difficulty for the Schwinger-type variational prin-
distribution, ;j;(r), which describes the reciprocal prob- ciples (Eqs. 2 and 3). When a scatterer is characterized
lem of a plane wave with wave vector -ks scattering by randomness in its geometrical or material proper-
into the direction -k; (i.e., transforming ks +-+ -k; in ties, the quantity of interest is a statistical moment of
the original problem). As will be illustrated by the T or of the differential cross section, 1112. Even the
general vector derivation given in the next section, one first moment of Eqs. 2 and 3 involves averaging a quo-
way to derive Schwinger's expression is to use ;j; to tient of complicated integrals,
eliminate the incident amplitude A and thereby obtain
the homogeneous, symmetric variational principle for (4)
the scatter amplitude,
which is generally intractable and deterred application.
But this impasse is broken by the work at APL report-
ed in Ref. 12, which demonstrates that, for arbitrary
This functional TV of the field functions 1/;,;j; possess- scatterer statistics, the integrals Nt, N 2, and D in Eq.
es the valuable invariance properties discussed above 2 can be individually averaged and then recombined to
(identify if as 1/;,;j; and F as TV), provided the func- form an invariant ratio for the averaged amplitude.
tionals N} , N 2 , and D in Eq. 2 are appropriately de- Subsequent work at APL has extended this result to
fined. This will be seen in detail for the general vector the vector case, Eq. 3, and to all higher moments.
case in the next section, where N} , N 2 , and D are ex- 22,23,29 Thus, in lieu of Eq. 4 a general stochastic vari-
hibited to be integrals over the field distribution at the ational principle for mean amplitude is
scatterer. The integral N} (1/;) is proportional to a
known noninvariant expression for T, while N2 (;j;) is (5)
the analogous expression in the adjoint problem. The
denominator D(1/;,;j;) is a double integral whose in- in the sense that (T) v is exact when exact fields are

Johns Hopkins APL Technical Digest, Volume 7, Number 1 (1986) 61

Bird, Farrell - Electromagnetic Scattering Theory
used to evaluate it and that o(T) v = 0 to first order trary. For another example, in the case of a lossy di-
(cf. above). Note that this is not a claim that Eq. 5 electric body with complex index of refraction, m,
is equal to Eq. 4 (except for exact fields), but simply
variationally equivalent to it in that first-order terms
vanish; the nonvanishing higher order terms in Eq. 5
N J (,p) 1 dV exp(-ik, ·r) u(r) ,p(r) •

differ from those in Eq. 4. Analogous variational ex-

pressions hold for higher statistical moments, e.g.,
D(,p,if;) 1 dV u(r) ,p(r) (7)
(l1l 2n ) V = (1N1 12n)(1N2 12n)/(47r)2n(/D12n).
These SVPs rest on the sole assumption that the in-
cident amplitude, A, is not stochastic, whence their
X 1
(if;(r) - dV'G(r,r') u(r') if;(r')] ,

proof follows by manipulation 12,22 of the general re- where V is the scatterer volume and u(r) == k 2 (m 2 -
lations T = NI/47rA and D = N2 e A, which obtain 1), assuming m = 1 (i.e., u = 0) outside V, and, again,
in the derivation of deterministic Schwinger principles N2 (~) = NI (t/;). The scalar principle discussed in Ref.
(e.g., Eqs. 12 and 14 below). The SVP, Eq. 5, is in- 1 is sufficient for the geometries of simple test prob-
herently more tractable than Eq. 4, since evaluating lems, but practical applications will require the more
a quotient of averages is less difficult than evaluating general vector principle about to be discussed.
averages of quotients. A special case of Eq. 5 was given Scattering from an object or surface with general
passing notice (but with neither proof nor subsequent electric and magnetic properties (Fig. 2) is described
generalization or application) in an early rough-surface by the vector time-harmonic wave equation for the
scattering theory.9 Sample results of our SVP appli- electric field,
cations will be given shortly, but first we will outline
V x V x E - k 2E = BeE , (8)
the derivation and some important aspects of the
general vector variational principle.
where k is the wavenumber of an incoming plane wave
Ej = Aej exp(ik j er) of amplitude A, propagating in
VARIATIONAL EXPRESSIONS FOR
the direction k j with linear polarization ej, and the
GENERAL SCATTERING PROBLEMS dyadic operator 0 characterizes the geometric and ma-
The application of variational principles to scatter- terial properties of the scatterer. For notational sim-
ing problems requires explicit expressions for the in- plification, arguments of functions will be omitted
tegrals N 1 , N 2 , and D. This section presents these when no confusion can arise. The explicit form of
expressions for several cases and reviews the deriva- o e E for general inhomogeneous and anisotropic scat-
tion of the vector variational principle. These mathe- terers with tensor permeability II, permittivity E, and
matical considerations are the foundation for appli- conductivity (1, is
cations and for the analyses of test cases described in
the next section. The comparisons there between the BeE = DeE + V x [pe[V X E]] , (9)
exact and variational results for these test cases pro-
vide a measure of the potential utility of variational where, with I the unit dyadic,
methods. That section and all subsequent ones have
been written in such manner that they can be read with- U == k2[r + (47rilw)u - I] , (9a)
out an appreciation for the mathematical foundations
presented in this section.
es component of the scattered wave
Scalar wave scattering is applicable to acoustics and
I ncident wave
to electromagnetics with special geometries. As dis-
cussed in Ref. 1, scattering from a perfectly conduct-
ing cylinder leads to

N J (,p) = 1 dS ,p(r) x !D ± (n) X exp(-ik, .r);

D(,p,if;) = 11dS dS' ,p(r) if;(r') (6)

and reciprocity as indicated earlier yields N2 (~)

NI (t/;). In Eq. 6, S denotes scattering surface, 5) ±(n)
denotes normal derivate a/an operating to the right
( + ) or left (-) for vertical or horizontal polarization, €(r},cr(r),ji(r)
respectively, and G(r,r') is the Helmholtz Green's
Figure 2-Vector wave scattering from volume Vo, bound-
function exp(iklr' -rl)/47rlr' -rl. For the acoustic prob- ed by surface So, of material with inhomogeneous anisotrop-
lem, hard (soft) scatterers correspond to vertical (hor- ic p~rmittivity, conductivity, and permeability tensors (E(r),O(r),
izontal) polarization, and scatterer geometry is arbi- and ji(r)).

62 Johns Hopkins APL Technical Digest, Volume 7, Number 1 (1986)

 Bird, Farrell - Electromagnetic Scattering Theory
.....
(9b) ond field, E, and an associated operator, 0, a;:e in-
troduced, Eq. 10 is multiplied by the product (OeE),
The permittivity and permeability have been normal- and the result is integrated over the volume of the scat-
ized by those of the homogeneous background medi- terer to obtain
um in which the scatterers are assumed embedded, and
the superscript -1 indicates matrix inversion. (14)
A careful application of Green's theorem yields the
integral equation 42,43 Here the integrals N2 and Dare

E(r) = Ej(r) + i Vo
dV' Go (r,r') • [O(r') • E(r')] ,
(10)
N2 = r
JVo
dV (geE) e [e i exp(ik i er)], (15)

where Vo is the volume of the scatterer and the in- D

finite space Green's dyadic is given by43
(16)
...... 2 exp Uk Ir - r' I )
P.V. (I + VVlk) - - - --
47r lr - r ' l
1 ......
- - Lo(r - r') (11) Using Eq. 14 to replace II A in Eq. 12 leads finally to
k2 '

(17)
where the symbol P. V. implies a principal value when
the term in brackets in Eq. lOis integrated and L is
the depolarization tensor 44 appropriate to the exclud- the form advertised earlier in Eqs. 2 and 3. Compari-
ed volume used to define the principal value. The in- son of Eqs. 7 above with Eqs. 13, 15, and 16 shows
finite-space Green's dyadic reduces to the usual free the similar form, but more complex construction, of
space Green' s dyadic 42 if the field point r lies outside the NI , N 2 , and D here.
the source region Vo, since o(r - r') = 0, and the prin- The requirement that Eq. 17 be stationary with re-
cipal value designation is unnecessary in that event. The spect to variations about the exact fields E and E
form of the infinite-space Green's dyadic can be de- results in separate integral equations (the Euler equa-
rived by using the free-space Green's dyadic and tions) for E and E. (These equations assume that the
Green's vector theorem to derive an integral equation inverse, 0-1 , exists. 23) The variational requirement
for the field within the source region. The singular na- suggests the choice g ot,
= where the adjoint is ot
ture of V V (1 I Ir - r ' I) requires that the point r = r '
be excluded from the region to which Green's theorem
defined by 1dV E e [ot
eE] == 1dV [OeE] e E. The
Euler equation for E is Eq. 10. The equation for E
is applied. Removal of an infinitesimal region about is similar to Eq. 10 except that it describes a plane wave
r = r' yields the principal value integral, and the resul- e
of linear polarization s and amplitude A = D / NI
tant integral over the surface enclosing this excluded incident along -ks being scattered by an object char-
volume yields the Lo(r - r') term. 43,45
The component of the scattering amplitude pola- acterized by 0+. With these identifications, the field
e
rized along a direction s is obtained from the asymp- E is usually called the adjoint field. This reciprocity
totic form of Eq. 10 on letting r - 0 0 in Eq. 11, and relationship between E and E suggests that a similar
one finds relationship should be imposed on the trial approxi-
mations for these fields, which are used in perform-
1 ing variational calculations via Eq. 17.
-N (12)
47rA I The dyadic Green's function formalism was con-
venient for deriving the above vector variational ex-
pressions 22,23 and for discussing general properties of
the variational principle. However, evaluation of the
integral D defined in Eq. 16 requires the infinite-space
where as before k s = k k s with k s rsI lrs I, and we Green's dyadic Go within the source region where the
e e
have defined s such that s eks = 0 in the far field depolarization tensor term, Lo(r - r'), of Eq. 11 is non-
limit. The form of the scattering amplitude given in zero. Straightforward evaluation of this depolariza-
Eq. 12 is noninvariant, in the sense that first-order er- tion tensor and the corresponding principal value
rors made in approximating the field E in Eq. 13 lead integral proved to be difficult. To remedy this, we used
to first-order errors in the approximation for T. some results from Yaghjian 43 to reexpress D in terms
An invariant form for T is obtained by using proce- of the scalar Helmholtz Green's function G introduced
dures analogous to those employed by Levine and in Eq. 6 and the text following. The double volume
Schwinger 46 for the scalar case. In particular, a sec- integral in D then becomes 24

John s Hopkins A PL Technical Digest, Volume 7, N umber 1 (1986) 63

Bird, Farrell - Electromagnetic Scattering Theory
as polarization effects for vector waves. But the test
models that are chosen are also simple enough to ad-
mit exact solutions, at least for some cases. This is im-
portant, since only by knowing exact results can one
truly judge the accuracy of approximations. Thus, the
models considered are constructed from infinite cir-
cular cylinders, which play the role of test model here
much like the hydrogen atom in quantum mechanics.
- k12 (OeE)
- + SVo ] ' (18)
We first summarize some results for scalar wave scat-
tering from a classic rough-surface model 47 and then
where SVo is a surface integral over the boundary, outline both the approach and the results for vector
So, of tIie scatterer involving G and the normal wave scattering from a random dielectric-body mod-
derivative aGlan. In particular, el. These examples involve small scatterer-size/wave-
length ratios; larger size-parameters are considered in
the next section.
r
JSo
dS' ( OeE) x (v'G XIlk2)
Scalar Test Problem
Our first application of the SVP involved calculat-
ing the averaged differential cross section, (111 2), for
the scattering of a scalar plane wave by a model rough
where superscript Tr denotes matrix transpose. Each surface where homogeneous (Dirichlet) boundary con-
integral of the right side of Eq. 18 contains integrable ditions are obeyed, i.e., where the wave function van-
singularities of the form IIR and IIR 2- e , € > O. ishes at the surface. The surface consists of a large
These should, therefore, be simpler to evaluate than number, N, of parallel, nonoverlapping hemicylindri-
the principal value integral and the depolarization ten- cal bosses of equal radius randomly distributed on a
sor required by the nonintegrable II R3 singularity of plane, and the wave is incident normal to the hemicyl-
Go. Thus, Eq. 18 offers a practical method for evalu- inders' axes (Fig. 3). The equivalent electromagnetic
ating the required variational expression (Eq. 16). problem is the scattering of a horizontally polarized
Vector variational expressions were also obtained wave by an embossed, perfectly conducting plane. We
for perfectly conducting objects. 8 ,22 The integrals in computed the first -order perturbational approximation
this case have the form to (1112) and compared it to the stochastic variation-
al improvement of the perturbation approxi-
mation. 18,20,21
NI
~ So
exp(-iks er) (es eK) dS , (20) In the low-frequency (Rayleigh) limit, both the vari-
ational and perturbational approximations were found

~
to be of the form
N2 CRee;) exp(ik i er) dS , (21)
So
(23)

D=~ So
dS K 0 ~
So
dS' GoK' , (22) to first order in v, the "packing density" or area frac-
tion of the plane occupied by the hemicylinders. In
where the surface integrals are over the scatterer sur- both cases the constant C in Eq. 23 is N times the cross
face So, K and K are the original and Adjoint surface section for an isolated hemicylinder. However, the pa-
currents, respectively, and in this case G is the reduced rameter Ci is equal to 2 for the perturbational approx-
form of Eq. 11 for r ¢ r'. This free-space dyadic imation and 4.08 for the variational result.
Green's function can be shown to be appropriate from
the limits taken in Green's theorem at the surface of
a perfect conductor. As a result, Din Eq. 22 is evalu-
ated by first performing the r' integration with r above
So and then letting r approach So to perform the r in-
tegration. I ,22 For two-dimensional systems, Eqs. 20
through 22 may be reduced to the scalar wave Eqs.
6 above.
STOCHASTIC VARIATIONAL TEST
APPLICATIONS
To assess the accuracy and efficacy of the SVP, we
have tested it on idealized random models that are
complex enough to exhibit cooperative phenomena Figure 3-Classic rough-surface scattering model: a stochas-
such as interference and multiple scattering, as well tic array of the Rayleigh surface elements of Fig. 1b.

64 Johns Hopkins APL Technical Digest, Volume 7, Number 1 (1986)

 Bird, Farrell - Electromagnetic Scattering Theory

In order to investigate the discrepancy beween these

two results, we considered a special case of this sur-
face with only two hemicylinders present. 19,20 We ob-
tained the exact solution, as well as the first-order
perturbational approximation and its variational im-
provement. Again we found the result Eq. 23, where
now the parameter a equals 1.89 for the exact solu-
tion, 2.04 for the variational approximation, and 1 for
the perturbational approximation. The variational re-
sult is thus the more accurate of the two approxima-
tions. A careful examination of these solutions for two
hemicylinders revealed that the variational approxima-
tion accounts for multiple scattering, but the first-
order perturbation approximation, which it clearly im-
proves upon, does not. 19
Vector Test Case with Inhomogeneous
Boundary Conditions
To assess the more general vector SVP, we applied
it to a random model of sufficient complexity to ex-
hibit polarization effects, as well as interference and
multiple scattering, but again, simple enough to ad-
mit an exact solution. Our previous experience led us
to consider a random assembly that consists of an en-
semble of systems, each of which is a pair of infinite-
ly long, parallel dielectric cylinders of radius a and
index of refraction m (Fig. 4). The cylinder separa-
tion varies randomly from ensemble member to en- Figure 4-Scattering configuration for two parallel dielectric
cylinders with randomly variable separation.
semble member, except that the cylinders are restricted
to be nonoverlapping and to have a maximum sepa-
ration L. (Of course, L must be much smaller than the
distance to a field point in order that the asymptotic fields E and E inside the scatterer. The analysis sim-
field and therefore the scattering amplitude is well de- plifies in the case of Rayleigh dielectric cylinders, for
fined.) The cylinders are illuminated by a plane elec- which ka ~ 1 and mka ~ 1. Then the fields inside
tromagnetic wave of wavenumber k propagating in the the cylinders for the original and adjoint problems are
direction kj with linear polarization ej, and we exam- reasonably approximated by the incoming plane
ine the component of the scattered field propagating waves, i.e., the Born approximation, E(r) ::::: ej
in the direction ks with polarization es. exp(ik j er) and E(r) ::::: es exp( -iks er), is appropriate.
The scattering configuration in Fig. 4 is the N = We compared 25,26 the exact, Born, and variational so-
2 case of the generic problem of scattering by N ran- lutions to first order in the packing density of the
domly separated Rayleigh cylinders. Eventual interest cylinders, JI = 4alL, for the plane waves incident nor-
resides in the limit of large N and small packing den- mal to the axes of the cylinders.
sity, JI = N2al L. For this reason, the solutions are ex- Waves with transverse electric (TE) and transverse
pressed in powers of packing density, and only linear magnetic (TM) polarization relative to the cylinders'
terms are retained. The N = 2 system has the virtue axes decouple and thus can be analyzed separately. The
of admitting an exact solution to test the vector sto- transverse magnetic wave is relatively simple. To first
chastic variational results. order in the packing density JI, the exact, Born, and
The exact field for each cylinder is expressed in stan- variational solutions for averaged cross section all
dard fashion 19,47,48 as a series of cylindrically outgo- agree:
ing waves centered on the axis of that cylinder. An
addition theorem 49 for Bessel functions is then used
to translate the waves centered on the axis of one cyl-
inder to a coordinate system centered on the other cyl- where ao is the single-cylinder Rayleigh transverse
inder. The boundary conditions on the surface of each magnetic cross section, and the term proportional to
cylinder can then be easily satisfied and the expansion JI is due to interference between the waves scattered
coefficients in the series determined. This procedure by different cylinders. Because the Born result is cor-
gives a series expansion of the scattering amplitude, rect to this order in JI, it follows that multiple scatter-
and in the small (i.e., Rayleigh) cylinder limit only the ing affects the transverse magnetic wave only in terms
leading term contributes. of higher order in v.
The variational approximation to the scattering am- The transverse electric wave is considerably more
plitude is obtained by choosing trial functions for the complicated. The Born approximation has the famil-

fohns Hopkins APL Technical Digest, Volume 7, Number 1 (1986) 65

Bird, Farrell - Electromagnetic Scattering Theory

iar cosine-squared behavior in scattering angle, Os used in noninvariant formulations. In any event, we
(measured from the forward direction), have seen that they were susceptible to vast improve-
ment via the SVP. Since our primary concern was to
test the SVP solutions against known exact solutions,
we gave no initial attention to other size-ranges or oth-
The Born approximation ignores the geometric polar- er trial functions. But, of course, realistic random scat-
izability of the cylinder in the transverse electric terers are characterized by ranges of particle dimen-
wave. 44 Thus, Eq. 25 disagrees with the single-scat- sions or roughness scales that may include or exceed
tering result, the wavelengths of interest. Hence, we have devoted
considerable subsequent study to the judicious selec-
tion or modification of trial functions, and to varia-
tional improvement of the Kirchhoff short-wave ap-
which is obtained by taking this geometric polariza- proximation appropriate to large size-parameters.
bility into account. A comparison with the exact so- These studies were mostly restricted to deterministic
lution to order v, systems, since the primary interest was in the spectral
behavior of the variational integrals (Nl , N 2 , and D)
that are the same in the deterministic Eqs. 2 and 3 as
in the SVP Eq. 5.
Our initial study 27 of trial functions treated a near-
shows that for m ~ 1 the multiple scattering effects ly transparent (m ~ 1) dielectric sphere, of size-
(through this order in JI) are given by parameter ka • (m - 1), via Eqs. 2 and 7, again with
2 the known exact solution available for comparison.
af,s = - [2/(m 2 + 1)]2 ( m - 1) For the simple approximation in which the incident
2
m + 1 plane wave is used as the trial function, the variation-
al total cross section was found to be accurate to 10
x 2ao v cos Os cos (Os - 2(3) , (28) percent up to ka(m - 1) ~ 0.8. Modifying the plane
wavenumber to that inside the sphere, mk, increased
this limit to ka(m - 1) ~ 1.2. Introducing an adjust-
to relative order (m 2 - 1)/24(m2 + 1), where {3 is the able wavenumber, exk, and determining ex variation-
angle the plane of the cylinder axes makes with the for- ally, raised the limit to ka(m - 1) ~ 1.6. Ultimately,
ward direction (Fig. 4). The variational improvement a nonplane (spherical lens) wave function was found
of the Born approximation has the form the most effective simple trial function tested, raising
the limit to ka(m - 1) ~ 4 (see Fig. 5). Indeed, rea-
sonably accurate near-forward scatter amplitudes were
obtained for ka(m - 1) as large as 10, which for m =
where the multiple scattering contribution (last term) 1.2 corresponds to a sphere radius of approximately
is found to be identical to the exact contribution (Eq. 8A.
28) to the same relative error. The higher order terms
of relative size (m 2 - 1)/24(m2 + 1) in both the ex-
act and variational multiple scattering contributions 4~----~----~------~----'-----~
are discussed and compared in Refs. 25 and 26. The
multiple scattering term arises entirely from the sur-
face integral term in Eq. 18. 3
In summary, comparison of the exact (Eq. 27) and
variational (Eq. 29) forms shows that the SVP correct-
ly accounts for the geometric polarizability of the
transverse electric wave (even using the Born trial field,
which does not); it shows further that, to lowest or-
der in the small parameter m 2 - 1, the SVP
reproduces the exact multiple scattering contribution.

VARIATIONAL TRIAL-FUNCTION
SELECTION
The tests so far described were limited to Rayleigh 2 3 4 5
scatterers, i.e., ones of small size compared to wave- ka(m - 1)
length. For such size-parameters, one generally expects Figure 5- Total scattering cross section (Utod for a dielec-
the plane-wave trial fields we adopted to be reason- tric sphere of radius a and refraction index m = 1.2 as a func-
tion of size parameter (ka(m - 1)). The variational result using
able first approximations, despite rather gross flaws the spherical lens trial function (colored curve) agrees with
(e.g., in Eqs. 23 and 25) and even total failures (cf. the exact solution (black curve) within 10 percent up to the
the Rayleigh transverse magnetic cylinder 30 ) when vertical bar.

66 Johns Hopkins APL Technical Digest, Volume 7, Number 1 (1986)

 Bird, Farrell - Electromagnetic Scattering Theory
We next investigated the effects of trial function
selection on the accuracy achievable with the surface- 3~---r----~---'----~----~--~

scattering test model described above (Fig. 3). For

plane-wave trial functions, substantially accurate vari-
ational amplitudes had been found for ka :5 1 and,
not surprisingly, complete breakdown at ka > 1, 2
where plane waves are a poor starting approxima-
tion. 18-21 In order to test the SVP at higher frequen-
cies, we have subsequently investigated the physical
optics (Kirchhoff) approximation as a trial function
in the variational integrals of Eq. 3. The Kirchhoff trial
function is the surface field determined by adopting
the tangent-plane approximation for boundary con-
ditions at the scatterer surface and the geometric op-
tics approximation for shadowing. Using this trial
function, we found variational improvement not only TM:
in the expected regime (ka > 1), but also through the 0i = 50°
important "resonance" region (ka - 1) and all the way Os = 40°
down to the Rayleigh limit (ka - 0). To investigate sys-
tematically this all-frequency behavior, we analyzed in
detail the classic single-boss model of Rayleigh (Fig. 1b)
over all size-parameters, polarizations, and bistatic con-
o
figurations. 30 This model involves much the same in-
tegrals, and ka behavior, as the multi-boss random
array; but again, its exact solution is available for strict
evaluation of results. Sample comparisons of Kirchhoff,
variational, and exact calculations are shown in Figs. -1r/4
lc, 6, and 7 via the curves labeled K, V, and E, respec-
tively. One sees for both transverse magnetic and trans-
verse electric polarization and in both magnitude and
phase that the variational (V) curves improve substan- -1r/2~ __~____~__~____~____L -_ _-J
tially over the Kirchhoff (K) in the latter's domain (ka o 2 4 6 8 10 12
~ 1), and then nicely follow the exact (E) curves ka
through the transition (ka - 1) where the Kirchhoff Figure 6- Typical bistatic scattering for the Rayleigh model
curves have departed significantly from the exact. (Fig. 1b) with horizontal polarization (transverse magnetic,
The findings over the entire ka spectrum from these E II axis). The scatter cross section relative to boss width (a/2a)
model calculations 30 are: (a) general variational im- and phase of scatter amplitude are shown as functions of
size-parameter (ka) for exact, Kirchhoff, and variational (E,K,V)
provement of the Kirchhoff approximation at high fre- solutions.
quencies (ka ~ 1); (b) better variational extension into
the intermediate-frequency region (ka - 1) than expect-
ed; and (c) substantial low-frequency improvement.
Regarding the latter, we fmd in brief that, whereas Ray- ational usage. Instead, it is attractive to exploit the
leigh limits of the Kirchhoff approximation are totally freedom inherent in the variational principle in order
wrong in wavelength and angle dependence (or, by a to design trial functions that are both simple and ef-
factor 2 in one case), the variational technique always fective. For simple scattering models, the liberated ap-
improves them by correcting the wavelength dependence proach has yielded extremely accurate all-wavelength
(and the factor 2). And, while not correcting faulty an- variational results. 28,29
gular distributions, the variational treatment is found In particular, we have investigated "boundary-
to improve integrated (total) cross sections. Born" approximations, i.e., plane-wave trial functions
The variational Rayleigh limits are evidently suffi- modified by a parametric function adjustable to suit
cient improvement that they enable the excellent short- the boundary conditions. One treats the function or
wave results to extend satisfactorily through the reso- its derivatives (depending on polarization) on the
nance regime. Nevertheless, the low-frequency accura- boundary as variational parameters and determines
cy is not uniformly adequate to consider the variation- them by the stationarity property of the variational
al-Kirchhoff approximation a reliable all-frequency principle. As an example, consider transverse electric
method. Furthermore, the Kirchhoff trial function can plane-wave scattering from the classic Rayleigh mod-
be awkward for variational calculation, as our forth- el (Fig. Ib). We used a simple modification of the in-
coming treatment of variational Kirchhoff divergences cident plane wave (t/;inc) in which the field near the
for transverse electric polarization will attest. In oth- scattering surface (So) is approximated as
er words, the canonical Kirchhoff approximation is
neither fully adequate nor suitably tractable for vari-

Johns Hopkins APL Technical Digest, Volume 7, Number 1 (1986) 67

Bird, Farrell - Electromagnetic Scattering Theory

3~--~----~---.--~-r----.----' 1.8

c 1.6
0
.';:;
u 1.4
2 Sl

e 1.2
<J)

u
Cl
c 1.0
..::
~
.....
co 0.8
~
"0
(1)
0.6
.~
ro
o E 0.4
0
z
0.2
1(/4 0.0
0.01 0.1 10 100
ka

o Figure 8-Variational results using the boundary-Born trial

function (colored curve) compared to the exact solution (black
curve) for the Rayleigh model (Fig. 1b) with vertical polariza-
tion (transverse electric) and OJ 50 Os =
-20 Note 0
, = 0
•

(1) logarithmic size scale (ka).

~ -1(/ 4
a...

scalar waves incident normally to the y axis to describe

-1(/2 scattering in different polarizations independently.
Thus, in horizontal polarization, E = jE(x,z), we ap-
proximate the electric field near the surface by the
boundary-Born form,
-31(/4 L -_ _---L_ _ _ _..L..-_ _---L_ __ _-'--_ _ _ _L.....-_ _- - '

o 2 4 6 8 10 12
E = E
inc
- E
inc
Iz= r • I(x,z) , (31)
ka
Figure 7-Same as Fig. 6, except with vertical polarization and the boundary condition E = 0 on Z t(x) is
(tranverse electric, B II axis).
guaranteed provided that fIx,t(x)] == 1. A simple
general form that satisfies this proviso homologously
Here l/; represents the magnetic field (H II axis), so the to Eq. 30 is I(x,z) = g[z - t(x)] with g(O) = 1. One
boundary condition is al/;Ian = 0 on So. This is iden- finds that in the variational integrals (Eq. 2) the para-
tically satisfied by Eq. 30, provided only that metric function, I, and its derivatives appear only via
al(P)lan lsQ = 1. We treatl(P) on So (i.e.,/(a» itself allazlr , which for the simple form indicated reduces
as a variatIOnal parameter and determine it by the sta- to g' (0). In that case, there results a completely para-
tionarity condition, a I1l 2 /al(a) = O. meter-independent variational scattering amplitude,
Figure 8 shows an example of the variational results which we have evaluated for a sinusoidal surface
for (normalized) cross section as a function of size- (again: known exact solution). This variational result
parameter ka. One sees excellent agreement with the has been shown analytically to be in precise agreement
exact solution for ka ranging over four orders of mag- with the exact solution in a variety of limiting cases-
nitude around unity. Similarly very accurate broad- including both low-frequency and high-frequency
band results were found for transverse magnetic waves, limits-even where the trial approximation gives quite
and also for isolated cylinders with either polarization, wrong noninvariant results. Numerical studies will
by using boundary-Born trial functions analogous to evaluate the solution for intermediate regimes of scat-
Eq. 30 but appropriate to the pertinent boundary con- tering parameters. If required, the variational results
ditions. 28 Thus, choosing simple trial functions that may be fine-tuned by noting that with a periodic sur-
are capable of satisfying the boundary conditions leads face, t(x) = t(x + A), the tuning function, allazlr ,
to excellent results for perfectly conducting cylindri- is likewise periodic and representable as a Fourier se-
cal scatterers, for radius to wavelength ratios, kaI27r, ries, the first term being just the constant g'(O). Of
ranging from very small to very large. course, if all terms of the series were retained, one ex-
Presently, trial function design for simple sea-sur- pects the exact solution. However, it is important to
face radar scattering models is under investigation in note that this is not the same as expanding the surface
collaboration with APL's Space Department. For ex- field in a complete set (e.g., Papas' early variational
ample, considering a one-dimensional, perfectly con- work 3 ) because the latter trial function takes no ac-
ducting corrugated surface, say z = t(x), we again use count of the boundary condition. Here it is built into

68 John s Hopkin s APL Technical Digest, Volume 7, Number 1 (1986)

 Bird, Farrell - Electromagnetic Scattering Theory
the trial functions such as Eqs. 30 and 31 from the internal field that satisfies the divergenceless vector
start. Helmholtz equation, with k 2 the inductive value (Eq.
Another problem presently being investigated is that 32),
of a beam with a Gaussian irradiance profile incident
on a perfectly conducting, infinitely long cylinder. The v x V x B - k 2 B = 0, div B = 0 , (33a)
case of normal incidence has been analyzed in detail
for a boundary-Born Gaussian trial field. The first sur- and an external scattered field (vl/;) given by Laplace's
prising result was that the adjoint problem is not ob- equation, i.e.,
tained merely by interchanging k; +-+ -ks in the original
Gaussian problem, but rather is the solution of a plane- B = Bo + Vl/;, Ill/; = 0, (33b)
wave scattering problem with the field incident from
-ks and detected in the direction -k;. Since we know subject only to the boundary condition
the boundary-Born trial field for a plane-wave scat-
tering problem, this unexpected result does not pres- B continuous at interfaces. (33c)
ent a difficulty. Numerical results for transverse mag-
netic polarization were obtained for a wide variety of Equations 33 completely define the field, which then
system parameters, and the agreement between the ex- gives the levitational force-torque system by standard
act and variational cross sections was as good as that formulas, as well as the correct eddy current distribu-
found earlier in the plane wave problem. tion via j = curl BI p,.
The solution of Eqs. 33 was calculated for a great
variety of shell constructions and shell-filament orien-
ELECTROMAGNETIC tations for infinitely long, stationary configura-
INDUCTION AND VISION tions. 39 Subsequently, both finite-length and kinet-
We now leave the subject of variational analysis of ic-dynamic effects were analyzed. The former 40 yield-
wave scattering to summarize some related low-fre- ed good agreement with stationary laboratory mea-
quency electromagnetic and visual analyses. At low surements. The latter 41 determined the kinetic magne-
frequencies or high conductivities, such that displace- tic torque affecting the spin-orbital dynamics of the
ment currents are negligible, pre-Maxwellian methods suspension over a wide range of parameters and con-
are commonly adopted (e.g., current-current interac- ditions. Along with aiding in malfunction diagnosis
tion at a distance). This may seem simpler, but actu- and design optimization, 50 it is interesting that al-
ally can be clumsy, ad hoc, and lead to incomplete ready infinite-length theory dissolved the apparent
solutions; Maxwell field theory instead facilitates pre- need for a difficult fabrication of end-caps for finite-
cise and complete solution of induction problems. 35-41 length nonhomogeneous shells. 39 This may seem par-
Consider an isotropic linear medium, wherein the adoxical but only shows once more the value of exact
electromagnetic Fourier components exp[i(ker - wI)] analysis of idealized models in order to understand
satisfy the vector wave equation with dispersion rela- practical systems.
tion k 2 = €P,W 2 + ip,aw, where €, p" a are permittivi- For a different example, consider the hydromagnetic
ty, permeability, conductivity, all assumed scalars here field, b, induced by interaction of a given magnetic
(cf. Eqs. 9). For the optical-to-radar waves considered field, B o , with a flow field, u, in a weakly conduct-
so far, and outside of good conductors, the wavenum- ing fluid. An important instance is oceanic hydromag-
ber, k, is nearly real and waves readily propagate. But netism, where the geomagnetic field interacts with the
in the induction regime (w ~ al €), great variety of seawater motions to induce manifold
oceano-magnetic effects. 35 The induced b is too weak
k =::: ..J(ip,aw) == (1 + 1)10 , (32) to act back on u to produce true hydromagnetic radi-
ations (e.g., Alfven waves) and, as Eq. 32 showed,
where 0 is skin depth, so that 1m k =::: Re k. This means electromagnetic waves do not propagate. However, an
that "waves" are so strongly damped that they sim- interesting phenomenon that we have called "nseu-
ply diffuse rather than propagate. Thus, instead of in- doradiation" does arise. When the inducing flo~ u is
coming plane-waves from infinity being scattered into a mechanical radiation (e.g., surface or internal gravity
far-field outgoing waves, here one has to deal with waves, or sound), it can carry along with it and there-
proximate interactions. by propagate the hydro magnetic field. The acoustic
For example, in the DISCOS magnetic suspension case entailing "sonomagnetic" pseudowaves and their
system, which consists of a metallic cylindrical shell scattering by the air-sea interface (see Fig. 9, top) has
encircling an AC filament, one can treat the levitation lreen analyzed in detail. 36,38
as repulsion between the AC current and the eddy cur- Sonomagnetic pseudoradiation is described by the
rents it induces in the shell. However, a model deduced inhomogeneous divergenceless vector Helmholtz equa-
from current flow at equilibrium gives a misleading tion, again with k 2 of Eq. 32,
physical picture that appears to entail a difficult fabri-
cation problem for the device. 39 The alternative vec- V x V x b - k 2b = -p,a curl (Bo x V cf»,
tor field treatment considers the magnetic field of the
current filament (Bo) scattering via the shell into an div b = 0 . (34a)

Johns Hopkins APL Technical Digest, Volume 7, Number 1 (1986) 69

Bird, Farrell - Electromagnetic Scattering Theory
9. The time-dependent case (k 2 in Eq. 34a reverting
z
to -p.aalat) was subsequently analyzed for strong pulse
External (air) B1
sources in an infinite medium, revealing a sonomag-
netic shock accompanied by precursor and relaxation
- - - - ... - - -........ - - waves. 38 Of particular interest are the vector charac-
Interface (z = O):'__-':~I --- ex
--'

.....--'!:
-- """'-i - ,. _ BII -_-_ teristics of the sonomagnetic signal, as opposed to the
scalar sound pulse, since they could aid in locating an
unknown source.
In closing, it seems felicitous to remark our appli-
cation of vector-field-analytic methods to a quite dif-
ferent facet of electromagnetic theory: the mathema-
tical description of visual sensation. This has exercised
physicists from the founder of electromagnetic theory,
Maxwell, through the early vector mathematization (of
5~--~---'---------'--------1 both space and color) by Helmholtz and Grassman,
to the Riemannian geometrization (following Ein-
stein's general relativity) of color theory by Schrl>din-
ger. Thus, there is consensus that vision requires a vec-
4 tor theory, despite a period of confusion over just what
are the vector components-retinal photon absorp-
c
0
tions, or cortical brightness and color sensations. 32
Ii>
+-'
+-'
'0, Only recently, however, has the mathematical descrip-
co ~
~ 3 ~
tion evolved to incorporate spatiotemporal variations,
Cl...
Q)
thus by definition forming a vector field theory, which
0
Cl
!co we have accordingly analyzed as we did the wave-
-I Q)
z scattering theory above via a Green's function
description. 33
2 I bll < 10-4 'Y Maxwell devoted a great deal-by some measures,
a third-of his efforts to the analysis of vision. His
mature insights that "All vision is color vision ... ,"
and" ... essentially a mental science," 51 seemed lost
1L-____- L__-L__________ ________
~ ~

in subsequent controversies that raged over Helmholtz-

o 0.5 1.5
ean "three-color" and Heringean "four- (or more) col-
R (kilometers) or" theories. However, controversy dissolved in mod-
Figure 9-Sonomagnetic pseudowave scattering by the ern times upon acceptance of the zone concept, viz.,
ocean surface (upper sketch) for an undersea acoustic source that Helmholtz red-green-blue (RGB) color compo-
(d ~ ocean depth) in geomagnetic field of horizontal and ver- nents represent the retinal photoabsorption zone, while
tical components, B. and B.l , respectively. The graph shows
the source power, P, versus range, R, domain (shaded) in
Hering's opponent colors (red versus green, blue ver-
which the sonomagnetic amplitude Ib.l I radiated at low al- sus yellow) represent the mental sensation zone. Again
titudes from a 1 hertz vertical acoustic dipole (at any d, see to quote Maxwell: " ... there is one word on which
Ref. 36) in a polar ocean exceeds a magnetometer sensitivi- we must fix attention. That word is Sensation." 51
ty of 10-4 ),.
The presently accepted model for visual sensation is
shown in Fig. 10.
Herein the acoustic field is represented by the poten- Recognizing the vector-field character of space-time
tial flow, U == V C/>, where c/> satisfies the scalar Helm- varying color-brightness vision, V(z ') as delineated in
holtz equation with acoustic wavenumber ka' the Fig. 10 caption, we formulated a general theory
in which V(z ') is a nonlinear vector functional repre-
senting all the many physiological operations of the
visual system that transmute the quantum absorptions,
in which S is the sound-source distribution and Cs is Q(z), by cone or rod cells at retinal coordinates z ==
sound speed_ The sonomagnetic field equations (Eqs. t,x,Y into the visual sensations. 32 For the large class
34), subject to the conditions b continuous and c/> = of vision experiments in which linearization is valid,
o at the air-sea boundary, were solved 36 in quadra- we expressed the theory by means of an analog to the
tures that are analogous to the Sommerfeld integrals dyadic Green's function for wave scattering encoun-
of antenna theory but are more complicated in that tered above (cf. Eq. 10), i.e.,33
two wavenumbers enter: the electromagnetic k (Eq. 32)
and acoustic ka (Eq. 34b). Analysis showed that the
sonomagnetic field, b, not only propagates with the
V(z') = Vo(z') + 1 dz F(z' ,z) • <1Q(z) (35)

sound but also is transmitted into the (nearly) sound- for small-, fast-, or fine-scale fluctuations .£1Q ==
free air. Therein it is radiated to considerable distances Q - Qo about an ambient state Vo ,Qo. Notke that we
for high-power infrasonic sources, S, as seen in Fig. have denoted the Green's dyadic here by F, since in

70 Johns Hopkins APL Technical Digest, Volume 7, Number 1 (1986)

 Bird, Farrell - Electromagnetic Scattering Theory
A The core of the research program remains the de-
velopment of variational techniques for wave scatter-
ing calculations, especially for stochastic scatterers. A
V A (Z) - - - - - - - - - - - - - - - - number of outstanding problems remain to be solved
to facilitate application of the technique. For exam-
ple, formulation of a surface-integral variational prin-
ciple for penetrable scatterers would ease calculations
for lossy dielectric scatterers. Some progress toward
this end has been made for high frequencies, and the
principle will be tested on a stochastic grating model.
Similarly, extension of our variational trial-function
design process to the case of imperfect conductors is
T
required. Also, calculations for infinite cylinders need
to be extended to finite-length, undulatory, and rough-
ened cylindrical or spheroidal scatterers. Complex ma-
croscopic bodies may then be modeled by combining
contributions from a set of such elementary constitu-
o ent scatterers.
Figure 10-Vector field representation of visual sensation
M in three Cartesian dimensions of achromatic (A), tritanopic REFERENCES
(T), and deuteranopic (0) opponent-color sensation. Bright- I P . M. Morse and H . Feshbach, Methods of Theoretical Physics, McGraw-
ness is measured by lVI, and hue and saturation by ratios Hill, New York (1953).
Vr/V D and VA/ lVI, respectively. All are functions of senso- 2H . Levine and 1. Schwinger, "On the Theory of Diffraction by an Aper-
ry space-time coordinates (z ;: t' ,x' ,V ' ). (For the color-defi- ture in an Infinite Plane Screen, I and II," Phys. Rev. 74,958-974 (1948)
cient, including one of the authors: the T axis is red-green, and 75, 1423-1431 (1949).
the 0 is blue-yellow.) 3c. H. Papas, "Diffraction by a Cylindrical Obstacle, " J. Appl. Phys. 21,
318-325 (1950).
4H. Levine and 1. Schwinger, " On the Theory of Electromagnetic Wave
Diffraction by an Aperture in an Infinite Plane Conducting Screen," in
vision the letter G is preempted for the color green and Theory of Electromagnetic Waves-A Symposium, Wiley Interscience,
New York (1951).
unavailable for George Green. The dyadic properties 5 C . - T . Tai, "Electromagnetic Back-Scattering from Cylindrical Waves,"
in Eq. 35 refer not to ordinary three-space, however, J. Appl. Phys. 23, 909-916 (1952).
but to the retinal (RGB) and cortical (ATD, cf. Fig. 6R. D. Kodis, "An Introduction to Variational Methods in Electromagnetic
Scattering," J. Soc. Indust. Appl. Math. 2, 89-112 (1954).
10) spaces. This theory has elucidated brightness-con- 7R. D. Kodis, " Variational Principles in High-Frequency Scattering," Proc.
trast studies and 'standard color models, has provided Cambridge Phi/os. Soc. 54, 512-529 (1958).
color vision generalizations of classic space-time bright- 8L. Cairo and T. Kahan, Variational Techniques in Electromagnetism, Gor-
don and Breach, New York (1965) .
ness laws, has analyzed wavelength-pulse and color- 9W . S. Ament, " Toward a Theory of Reflection by a Rough Surface, "
flicker experiments, and has yielded understanding of Proc. IRE 41, 142-146 (1953) .
10 C. 1. Palermo and L. H. Bauer, "Bistatic Scattering Cross Section of Chaff
heterochromatic luminance additivity in flicker and Dipoles with Application to Communication," Proc. IEEE 53, 1119-1121
border observations. 32,33 (1965) .
II A. Ishimaru, Wave Propagation and Scattering in Random Media, Aca-
demic Press, New York (1978) .
CONCLUSION 12 R. W . Hart and R. A. Farrell, "A Variational Principle for Scattering from
The theoretical researches we have surveyed concern Rough Surfaces," IEEE Trans. Antennas Propag. AP-25, 708-710 (1977).
13 For a recent review, see E. Lilneburg, "Wave Scattering by Random Sur-
topics that at first can seem too disparate to belong faces and Application to Natural Surfaces," in Wave Propagation and Re-
under one title. Yet, the unity of physics relates the mote Sensing, Proc. URSI Commission F 1983 Symp., European Space
Agency-ESTEC, Noordwijk , The Netherlands, pp. xliii-I (1983).
different topics closely. One way to appreciate this is 14 A . K. Fung and M. F. Chen, Scatteringfrom a Perfectly-Conducting Ran-
to compare the basic field equations (Eqs. 8, 33, and dom Surface-Extinction Method, Univ. of Kansas, Remote Sensing Lab-
34). They are all variations on the vector Helmholtz oratory TR 592-3 (Dec 1983).
15 E . Bahar and D. E. Barrick, "Scattering Cross Sections for Composite
equation that governs general vector field behavior (see Surfaces that Cannot Be Treated as Perturbed Physical Optics Problems,"
Ref. 1, Eq. 13.1.1 and text following). The differences Radio Sci. 18, 129-137 (1983).
16E. Bahar, C. L. Rufenach, D. E . Barrick, and M. A . Fitzwater, "Scatter-
reside in boundary conditions (e.g., Eqs. 1 versus 33b), ing Cross Section Modulation for Arbitrarily Oriented Composite Rough
or in subsidiary conditions (e.g., Eq. 34b), or in the Surfaces: Full Wave Approach," Radio Sci. 18, 675-690 (1983).
media (e.g., Eqs. 9 versus 32). Vision theory lacks such 17E. Bahar and M. A. Fitzwater, "Scattering Cross Sections for Composite
Rough Surfaces Using the Unified Full Wave Approach, " IEEE Trans.
dynamical field equations (although visual kinemat- Antennas Propag. AP-32, 730-734 (1984) .
ics hints of analogs via the existence of perceptual 18E. P . Gray, R. W . Hart, and R. A. Farrell, "A New Variational Approach
Lorentz transformations 52), but its vector field nature to Scattering by Random Media or Rough Surfaces," in Proc. Open Sym-
posium, URSI Commission F, pp. 111-115 (1977).
(Fig. 10) and Green's function representation (Eq. 35) 19 1. A. Krill and R. A. Farrell, "Comparisons Between' Variational , Per-
are well established. Of course the differences in topics turbational, and Exact Solutions for Scattering from a Random Rough
Surface Model," J. Opt. Soc. Am. 68, 768-774 (1978).
lead to dissimilar treatments and consequences. But 20E. P . Gray, R. W. Hart, and R. A. Farrell, " An Application of a Varia-
as emphasized through this review, the mathematical tional Principle for Scattering by Random Rough Surfaces," Radio Sci.
similarities of the different vector field analyses allow 13, 333-343 (1978).
21 E . P . Gray, R. W . Hart, and R. A . Farrell, A Variational Approximation
them to share common language and lend mutual for the Scattering of Scalar Waves by Stochastic Surfaces, IHU/ APL
support. TG 1322 (1979) .

Johns Hopkins APL Technical Digest, Volume 7, Number 1 (1986) 71

Bird, Farrell - Electromagnetic Scattering Theory

22J. A. Krill and R. H. Andreo, "Vector Stochastic Variational Principles 38 J. F. Bird, "Sonomagnetic Pulses from Underwater Explosions and Im-
for Electromagnetic Wave Scattering," IEEE Trans. Antennas Propag. plosions," l. Acoust. Soc. Am. 67, 491-495 (1980).
AP-28, 770-776 (1980). 39 J. F. Bird, "Theory of Magnetic Levitation for Biaxial Systems," l. Appl.
23R. H. Andreo and J. A. Krill, "Vector Stochastic Variational Expressions Phys. 52, 578-588 (1981) .
for Scatterers with Dielectric, Conductive, and Magnetic Properties," l. 40 J. F. Bird, "Levitational End-Effects in a Cylindrical Magnetic Suspen-
Opt. Soc. Am. 71, 978-982 (1981). sion," l. Appl. Phys. 52, 6032-6040 (1981).
24J . A. Krill, R. H. Andreo, and R. A. Farrell, "A Computational Alter- 41 J. F. Bird, "Kinetic Torque and Dynamic Behavior in a Magnetic Levita-
native for Variational Expressions that Involve Dyadic Green Functions," tion Device," l. Appl. Phys. 53, 1326-1333 (1982).
IEEE Trans. Antennas Propag. AP-30, 1003-1005 (1982). 42c. - T. Tai, Dyadic Green's Functions in Electromagnetic Theory, Intext,
25 J. A. Krill, R. H. Andreo, and R. A. Farrell, "Variational Calculations Scranton (1971).
of Electromagnetic Scattering from Two Randomly Separated Rayleigh 43 A. D. Yaghjian, "Electric Dyadic Green's Functions in the Source Re-
Dielectric Cylinders," l. Opt. Soc. Am. 73, 408-410 (1983). gion," Proc. IEEE 68, 248-263 (1980).
26 J. A. Krill, R. H. Andreo, and R. A. Farrell, Calculation Procedures for 44H. C. van de Hulst, Light Scattering by Small Particles, Dover, New York
Variational, Born, and Exact Solutionsfor Electromagnetic Scattering from (1981).
Two Randomly Separated Dielectric Rayleigh Cylinders, JHU I APL 45J . Van Bladel, "Some Remarks on Green's Dyadic for Infinite Space,"
TG 1344 (1983). IRE Trans. Antennas Propag. AP-9, 563-566 (1961).
27M. R. Feinstein and R. A. Farrell, "Trial Functions in Variational Ap- 46H . Levine and J . Schwinger, "On the Theory of Electromagnetic Wave
proximations to Long Wavelength Scattering," l. Opt. Soc. Am. 72, Diffraction by an Aperture in an Infinite Plane Conducting Screen,"
223-231 (1982). Comm. Pure Appl. Math. 3, 355-391 (1950).
28 J. A. Krill, J . F. Bird, and R. A. Farrell, "Trial Functions in Variational 47 V. Twersky, "Multiple Scattering of Radiation by an Arbitrary Planar
Calculations," in Proc. 1982 CSL Scientific Con! on Obscuration and Configuration of Parallel Cylinders and by Two Parallel Cylinders," l.
Aerosol Research, R. H. Kohl & Assoc., pp. 201-209, (1983). Appl. Phys. 23, 407-414 (1952).
29 J. A. Krill and R. A. Farrell, "The Development and Testing of a Stochas- 48G. Olaofe, "Scattering by Two Cylinders," Radio Sci. 5, 1351-1360 (1970).
tic Variational Principle for Electromagnetic Scattering," in Wave Prop- 49G. N. Watson, A Treatise on the Theory of Bessel Functions, 2nd ed.,
agation and Remote Sensing, Proc. URSI Commission F 1983 Symp., Cambridge University Press, New York (1966).
European Space Agency-ESTEC, Noordwijk, The Netherlands, pp. 50V. L. Pisacane et aI., TIP-III DISCOS Study Group Report, JHU/APL
299-307 (1983). SDO-5592 (1980).
30 J. F. Bird, "Analysis of All-Frequency Variational Behavior of the Kirch- 51 J. C. Maxwell, "On Colour Vision," Scientific Papers, Vol. 11,267-279,
hoff Approximation for a Classic Surface-Scattering Model," l. Opt. Soc. Dover reprint (1890).
Am. A 2, 945-953 (1985). 52T. Caelli, W. Hoffman, and H. Lindman, "Subjective Lorentz Transfor-
31 R. A. Farrell, C. B. Bargeron, W. R. Green, and R. L. McCally, "Col- mations and the Perception of Motion," l. Opt. Soc. Am. 68,402-411
laborative Biomedical Research on Corneal Structure," Johns Hopkins (1978).
APL Tech. Dig. 4, 65-79 (1983).
32 R. W. Massof and J. F. Bird, "A General Zone Theory of Color and
Brightness Vision I. Basic Formulation," l. Opt. Soc. Am. 68,1465-1471 ACKNOWLEDGMENT -This work has been supported in part by the
(1978). Army, the Navy, the Strategic Defense Initiative Office, the National Aero-
33 J. F. Bird and R. W. Massof, "A General Zone Theory of Color and nautics and Space Administration, and the National Institutes of Health. We
Brightness Vision II. The Space-Time Field," l. Opt. Soc. Am. 68, are pleased to acknowledge a number of members of the Laboratory's staff
1471-1481 (1978). who made important contributions to the analyses that are reviewed in this
34 J. F. Bird, R. W. Flower, and G. H. Mowbray, "Analysis of the Retina article. They include: Robert W. Hart, Assistant Director for Research and
via High Frequency Electroretinography," Biophys. l. 29, 379-396 (1980). Exploratory Development; Jerry A. Krill and Matthew R. Feinstein, presently
35 J. F. Bird and H. Ko, Ocean Magnetics I. Fundamental Survey and Esti- in the Fleet Systems Department; David E. Freund of the Research Center;
mates of Induction Phenomena, JHU I APL TG 1315A (1977). Ernest P. Gray of the Space Department and members of the DISCOS Study
36 J. F. Bird, "Hydromagnetism Induced by Submerged Acoustic Sources: Group in that Department; Harvey W. Ko of the Submarine Technology
Sonomagnetic Pseudoradiation," l. Acoust. Soc. Am. 62, 1291-1296 Department; and Robert H. Andreo, who is no longer at APL. It is also a
(1977). pleasure to acknowledge collaboration on the color vision theory with Robert
37 J. F. Bird, "Hydromagnetic Perturbations Due to Localized Flows: An W. Massof of the Wilmer Institute of The Johns Hopkins University School
Eddy Theorem," Phys. Fluids 22, 585-586 (1979). of Medicine.

THE AUTHORS

JOSEPH F. BIRD (left) is a physicist in the Theoretical Problems

Group. He joined the Milton S. Eisenhower Research Center in 1958
and was appointed to the Principal Professional Staff in 1962. He
was born in Scranton, Pa., in 1930 and earned an A.B. degree in
physics from the University of Scranton in 1951 and a Ph.D. de-
gree in theoretical physics from Cornell University in 1958. Dr. Bird's
research results at APL have been published as journal articles in
various fields: combustion instability in solid-fuel rockets, star for-
mation theory and cosmogony, psychophysics and electrophysiology
of vision, oceano-hydromagnetics theory, neural noise modeling,
the theory of chromatic visual sensations, electromagnetic levitation
calculations, and wave scattering theory. The variety of these re-
searches largely reflects the diversity of his collaborations with other
APL departments and the Johns Hopkins Medical School.

RICHARD A. FARRELL (right) is a physicist and the supervisor

of the Theoretical Problems Group in the Milton S. Eisenhower Re- mg techniques used in his corneal research and the Laboratory's
search Center. Born in Providence, R.I., he obtained a B.S. degree interest in alterations of the sea surface led to his involvement in
from Providence College in 1960, an M.S. from the University of fundamental investigations of scattering by random systems. Dr.
Massachusetts in 1962, and a Ph.D. from The Catholic University Farrell is the principal investigator on contracts and grants from
of America in 1965. Dr. Farrell's research interests include relating the Army and the National Eye Institute and is an investigator on
the cornea's structure to its function, especially relative to its light- the Laboratory's Space Sciences Consortium in support of the Stra-
scattering properties; developing theoretical methods for calculating tegic Defense Initiative. He is a member of various professional or-
wave scattering in random media; and analytic treatments of the ganizations, including the American Physical Society, the Optical
statistical mechanics of phase transitions. APL's collaborative ef- Society of America, the Association for Research in Vision and
forts with the Johns Hopkins Medical School stimulated his interest Ophthalmology, the International Society for Eye Research, and
in biomedical problems when he joined APL in 1965. The scatter- the New York Academy of Sciences.

72 fohns Hopkins APL Technical Digest, Volume 7, Number 1 (1986)