CHAPTER 7

PERTURBATIONAL AND VARIATIONAL TECHNIQUES

7-1. Introduction. The differential equation approach of the preced-

ing three chapters leads to an exact solution of the mathematical problem.
However, many problems cannot be treated by this method. We saw in
Sec. 3-11 that electromagnetic field problems can be expressed in integral
equation form. This form is particularly useful for (1) obtaining approxi-
mate solutions and (2) for general expositions of theory. In this chapter,
we shall consider two techniques useful for integral equations arising in
electromagnetic theory.
Perturbational Methods. The word "perturb" means to disturb or to
change slightly. The perturbational methods are useful for calculating
changes in some quantity due to small changes in the problem. Usually
two problems are involved: the "unperturbed" problem, for which the
solution is known, and the" perturbed" problem, which is slightly differ-
ent from the unperturbed one. We have already used perturbational
methods for calculating resonator quality factors and waveguide attenu-
ation constants. Further uses are given in Sees. 7-2 to 7-4.
Variational Methods. The variational methods are useful for deter-
mining characteristic quantities, such as resonant frequencies, imped-
ances, and so on. In contrast to the perturbational procedure, the
variational procedure gives an approximation to the desired quantity
itself, rather than to changes in the quantity. The variational pro-
cedure differs from other approximation methods in that the formula is
" stationary" about the correct solution. This means that the formula
is relatively insensitive to variations in an assumed field about the cor-
rect field. If the desired quantity is real, the variational formula may
be an upper or lower bound to the quantity. Furthermore, if an assumed
field is expressed as a series of functions with undetermined coefficients,
then the coefficients can be adjusted by the Ritz procedure (Sec. 7-6).
In fact, if a complete set of functions is used for the assumed field, the
exact solution can sometimes be obtained, at least in principle.
7-2. Perturbations of Cavity Walls. Figure 7-1a represents a resonant
cavity formed by a conductor covering S and enclosing the loss-free
region T. Figure 7-1b represents a deformation of the original cavity
317
318 TIME-HARMONIC ELECTROMAGNETIC FIELDS

n n

(a) (b)

FIG. 7-1. Perturbation of cavity walls. (a) Original cavity; (b) perturbed cavity.

such that the conductor covers S' = S - ~S and encloses T' = T - ~T.
We wish to determine the change in the resonant frequency due to the
change of the cavity wall.
Let Eo, Hs, Wo represent the field and resonant frequency of the original
cavity, and let E, H, ca represent the corresponding quantities of the
perturbed cavity. In both cases the field equations must be satisfied,
that is,
-v X Eo = jwop.Ho -v X E =jwp.H (7-1)
v X a, = jwoeEo v X H = jweE
We scalarly multiply the last equation by E~ and the conjugate of the
first equation by H. The resulting two equations are
Eri · V X H = jweE · Eri
- H · V X Et = - jwop.Ht • H

Adding these and applying the identity

V • (A X B) = B • V X A - A • V X B
we have
v · (H X Eri) = jweE • E: - jwop.Ht • H
By analogous operations on the second and third of Eqs. (7-1), we obtain
V • (Hri X E) = jwp.H • Hri - jwoeEt • E

These last two equations are now added, and the sum integrated through-
out the volume of the perturbed cavity. The divergence theorem is
applied to the left-hand terms, one of which vanishes, because n X E = 0
on S'. The resulting equation is

1f H
s'
X Et • ds = j(w - wo) JJJ (EE. Et + pH • Ht) dr (7-2)

Finally, since n X Eo = 0 on S, we have

1f H X Et • ds =0
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 319
and the left-hand side of Eq. (7-2) can be written as

1fi H
8'
X Et • ds = 1ft H X Et · ds = -1fi
8 -8 ss
H X Et • ds

The last term is taken as negative, to conform to the convention that

ds points outward. We can now rewrite Eq. (7-2) as

j 1fi H
A8
X Et · ds
= ----------- (7-3)
W - Wo
JJJ(EE. Et + ISH· Ht) d.,.
T'

This is an exact formula for the change in resonant frequency due to an

inward perturbation of the cavity walls. Note that our development
assumes that E and p. are real, that is, we have assumed no losses. Prob-
lem 7-1 gives the general formulation in the lossy case.
The crudest approximation to be made in Eq. (7-3) is that of replacing
E, H by the unperturbed field Eo, Hs, For small perturbations this is
certainly reasonable in the denominator and should be valid in the
numerator if the deformation is shallow and smooth. With this approxi-
mation the integral in the numerator of Eq. (7-3) becomes

1fi H X Et • ds = 1fi ra, X Et) · ds

~8 ~8

= JJJ (EIE - ISIH jwo oI2 oI2) d.,.

AT

The last equality follows from the conservation of complex power [Eq.
(1-62)]. Substituting this into Eq. (7-3), and also substituting Eo, H o
for E, H in the denominator, we have

W - Wo
JJJ(1S1H
AT
o12 - E/E oI2) dr
(7-4)
Wo = j--=j-j--:::---(-p.,-H-o-'2-+-E'-E-o'-2)-d-r

Note that the terms in the numerator are proportional to the electric and
magnetic energies" removed" by the perturbation, while the denominator
is proportional to the total energy stored. Hence, Eq. (7-4) can be
written as
W - "'0 ~Wm - ~W~
~----- (7-5)
Wo W
where ~Wm and ~W~ are time-average electric and magnetic energies
originally contained in AT and W is the total energy stored in the original
320 TIME-HARMONIC ELECTROMAGNETIC FIELDS

cavity. Finally, if dT is of small extent, we can approximate the dW's

by dT times the energy densities at the position of dr. Furthermore,
W can be written as r times a space-average energy density w. Thus,
Eq. (7-5) can be written as

W - Wo ~ (w m -= We) dr = C !::aT
(7-6)
Wo ui-r T

where C depends only on the cavity geometry and the position of the
perturbation.
It is evident from the preceding equations that an inward perturbation
will raise the resonant frequency if it is made at a point of large H (high
tOm), and will lower the resonant frequency if it is made at a point of
large E (high We). The opposite behavior results from an outward pertur-
bation. It is also evident that the greatest changes in resonant frequency
will occur when the perturbation is at a position of maximum E and zero
H, or vice versa.
Numerical calculations using Eqs. (7-4) to (7-6) are easy for the cavi-
ties treated previously, because we calculated W when we determined the
Q's. For the dominant mode of the rectangular cavity of Fig. 2-19, W is
given by Eq. (2-98), or

For dT located at the mid-point of the base (maximum E) we use Eqs.

(2-96) to find dW m = 0, and

t::..W e = ~ IEol2 t::..r

Hence, from Eq. (7-5) we find
W - Wo ~ -2 dr (7-7)
Wo r

If the perturbation occurs at the mid-point of the longer side wall (maxi-
mum H), we have dW e = 0 and
EIEol2
t::..W m = 2(1 + c2/b2) t::..r
Hence, from Eq. (7-5) we find
W - Wo 2 !::ar
---~ (7-8)
1 + (C/b)2-;:-
Note that for a square-base cavity (b = c) the change in resonant fre-
quency due to dT at maximum H is only one-half as great (and in the
opposite direction) as that due to dr at maximum E.
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 321
'rABLE 7-1. THE PARAMETER C OF EQ. (7-6) FOR DEFORMATIONS (a) AT MAXIMUM
E AND (b) AT MAXIMUM H OF THE DOMINANT MODE

Cavity Geometry C

(a) -2
Rectangular
(a ~ b ~ c)
(b) 1 + ~c/W

Short cylinder
(d < 2a) ~(;~J (a) -1.85
(b) 0.5

(a) -0.843
Long cylinder
b 2.86
(d ~ 2a)
( ) 1 +
(1.71a/d)2

(a) -0.361
Spherical
(b) 0.680

(a) -2.02
Hemispherical
(b) 0.680

Table 7-1 gives the value of C in Eqs. (7-6) for cavities of several
geometries for ~T located at (a) maximum E and (b) maximum H. These
values have been obtained using the crude approximations of replacing
E, H by Eo, H o in Eq. (7-3). They are therefore valid only for smooth,
shallow deformations. In general, the frequency shift depends on the
shape of the deformation as well as on the shape of the cavity. The
formulas for deformations of the form of small spheres or small cylinders
can be obtained from the results of the next section by letting E ~ ex>
and p,~ o.
7 -3. Cavity-material Perturbations. Let us now investigate the change
in the resonant frequency of a cavity due to a perturbation of the material
within the cavity. Figure 7-2a represents the original cavity containing
matter E, p: Figure 7-2b represents the same cavity but with the matter
changed to E + dE, P. dp.. +
322 TIME-HARMONIC ELECTROMAGNETIC FIELDS

D D

FIG. 7-2. Perturbation of

matter in a cavity. (a)
Original cavity; (b) per-
turbed cavity.
(a) (b)

Let Eo, H o, Wo represent the field and resonant frequency of the original
cavity, and let E, H, w represent the corresponding quantities of the
perturbed cavity. Within S the field equations apply, that is,
-v X Eo = jwoJ.LH o - V X E = jw(J.L + ~J.L)H (7-9)
V X H o = jwoEE o V X H = jW(E + dE)E
As in the preceding section, we scalarly multiply the last equation by
Eri and the conjugate of the first equation by H, and add the resulting
two equations. This gives
V • (H X Eri) = jW(E + dE)E • Eri - jwoJ.LHri • H
Analogous operation on the second and third of Eqs. (7-9) gives
V • (Hri X E) = jw(J.L + dJ.L)H • Hri - jWoEEri • E
The sum of the preceding two equations is integrated throughout the
cavity, and the divergence theorem is applied to the left-hand terms.
The left-hand terms then vanish, because both n X E = 0 on Sand
n X Eo = 0 on S. The result is

o= III {[wee + .::le) - woe]E • E~ + [W(IL + .::l1L) - wOIL]H • Htl dr

Finally, this can be rearranged as

W - Wo III (.::leE· Et + .::lILH • Ht) dr (7-10)

III (eE. Et + ILH • Ht) dr
This is an exact formula for the change in resonant frequency, due to a
change in E and/or J.L within a cavity. Once again our development has
assumed the loss-free case, that is, E and J.L are real. The general formu-
lation when losses are present is given in Probe 7-5.
In the limit, as dE ~ 0 and ~J.L ~ 0, we can approximate E, H, w by
Eo, Hs, Wo and obtain

w - Wo
::::::
III (.::leIEoI2+ .::lILIHoI2) dr (7-11)
III (elE ol2 + ILIH dr oI2)
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 323
This states that any small increase in E and/or J.L can only decrease the
resonant frequency. Any large change in E and/or p, can be considered
as a succession of many small changes. Hence, any increase in E and/or p,
within a cavity can only decrease the resonant frequency.
We can recognize the various terms of Eq. (7-11) as energy expressions
and rewrite it as

(7-12)

where W is the total energy contained in the original cavity. Now if

the change in E and J.L occupies only a small region dr, we can further
approximate Eq. (7-12) by

(7-13)

where w is the space average of W. The parameters C 1 and C 2 depend

only on the cavity geometry and the position of dr. Note that a small
change in E at a point of zero E or a small change in J.L at a point of zero H
does not change the resonant frequency. If we compare Eq. (7-13) with
Eq. (7-6), it is evident that C = C2 - C 1• For the cases considered .in
Table 7-1, dr is either at a point of zero H, in which case C 2 = 0, or at a
point of zero E, in which case C1 = O. To be explicit, for a material
perturbation at (a) of Table 7-1 we have C 1 = -C and C 2 = 0, while for
a material perturbation at (b) of Table 7-1 we have C 1 = 0 and C 2 = C.
The preceding approximations require that LlE, t:.p., and Llr all be small.
We shall now consider a procedure for removing these restrictions on dE
and djl. This introduces the further complication that the change in
frequency depends on the shape of dr, as well as on its location. The
modification is accomplished by using a quasi-static approximation to the
field internal to dr. This assumes that the field internal to dr is related
to the field external to t:.r in the same manner as for static fields. The
procedure is justifiable, because, in a region small compared to wave-
length, the Helmholtz equation can be approximated by Laplace's
equation.
There are four types of samples for which this quasi-static modification
to the perturbational solution is very simply accomplished. These are
shown in Fig. 7-3 for the dielectric case. For the magnetic case, it is
merely necessary to replace E by Hand E by u, For the thin slab with
E normal to it (Fig. 7-3a), we must have continuity of the normal com-
324 TIME-HARMONIC ELECTROMAGNETIC FIELDS

o_~()
(a) (b) - (c) (d)
FIG. 7-3. Some small dielectric objects for which the quasi-static solutions are simple.

ponent of D, so that
(7-14)

This approximation is valid regardless of the cross-sectional shape of the

cylinder. For the long thin cylinder with E tangential to it (Fig. 7-3b),
we must have continuity of the tangential component of E, so that
(7-15)
Again this approximation is independent of the cross-sectional shape of
the cylinder. For E normal to a long thin circular cylinder (Fig. 7-3c),
we can use the static solution;' which is

(7-16)

Finally, for E normal to a small sphere (Fig. 7-3d), we can use the static
solution," which is

(7-17)

The static solution for a dielectric ellipsoid In a uniform field is also

known but is not very simple in form."
To use the above quasi-static approximations, we approximate E (and
H in the magnetic case) in the numerator of Eq. (7-10) by E in t of the
preceding equations. In the denominator we can still use the approxi-
mations E = Eo and H = Hs, because the contribution from ~T is small
compared to that from the rest of T. Hence, our quasi-static correction
to the perturbational formula is

W - Wo III llEE E~ dT
int •
(7-18)
2 III EIE dr
~

Wo
ol2

1 W. R. Smyth, "Static and Dynamic Electricity," pp. 67-68, McGraw-Hill Book

Company, Inc., New York, 1950.
2 J. A. Stratton, "Electromagnetic Theory," pp. 205-213, McGraw-Hill Book Com..

pany, Inc., New York, 1941.

 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 325

//
/' /I-1~
- -
//

d~a~
(a) (b) (c)

FIG. 7-4. Cavities used to illustrate the perturbational formulas.

for the case dp. = o. (The denominator has been simplified by equating
Wm to We.) The corresponding formula for the frequency shift due to
a magnetic material would be of same form, but with E replaced by H
and E by p. throughout.
Equation (7-18) is, of course, most valuable for problems for which
the exact solution is not known. However, so that we may gain confi-
dence in the results as well as practice in the procedure, let us apply
Eq. (7-18) to problems for which we have the exact solution. These are
illustrated in Fig. 7-4. For a dielectric slab on the base of a rectangular
cavity (Fig. 7-4a), we have E i n t given by Eq. (7-14). The field and energy
expressions for the unperturbed cavity are given in Sec. 2-8. Appli-
cation of Eq. (7-18) then yields

W - Wo 1 Er - 1d
Wo ~ - 2 - Er-- a (7-19)

where d is the slab thickness and a is the cavity height. A comparison

of this with the result of Probe 4-17 for P.l = P.2 = P.O and E2 = EO shows
that our answer is identical to the first term of the expansion for w in
powers of d/a. In fact, if dp. is also nonzero and we treat it to the same
degree of approximation (match tangential H), we again get the correct
first term of the expansion. To illustrate the improvement obtained by
using the quasi-static field, we can compare Eq. (7-19) to the result
obtained from Eq. (7-11), which is
w - Wo 1 d
-- ~ - - (e, - 1)-
Wo 2 a

It is apparent that the above formula is accurate only for Er ~ 1, that is,
when ~E is small.
A nonmagnetic dielectric slab at a side wall of the rectangular cavity
(Fig. 7-4b) has but little effect on the resonant frequency, because E is
zero at the wall. In this case E is tangential to the air-dielectric inter-
face; so Eq. (7-15) should apply. Note that Eqs. (7-18) and (7-11) give
326 TIME-HARMONIC ELECTROMAGNETIC FIELDS

identical approximations in this case. In particular, we obtain.

ta -
WO
WO
~ -
(e, - 1).
a
h
d

0
•
SIn -
2 7rX
a
dx

~ - ~2 (e, - 1) (~y (7-20)

A comparison of this with the answer to Probe 4-18 shows that we again
have the correct first term of the expansion when ~Jl = o.
As a final example, consider the spherical cavity with a concentric
dielectric sphere (Fig. 7-4c). The field of the unperturbed cavity is
defined by

H. = ~Jl(2.744~)Sin8
and the stored energy is given by Eq. (6-35). Applying Eq. (7-18),
using the quasi-static Eq. (7-17), we obtain

w - Wo ~ -0.291 Er
- 1 (2.744 ~)B
Wo Er +2 b
where a is the radius of the small dielectric sphere and b is the radius of
the conductor. This we can compare to the exact solution (Prob. 6-8),
which is the same. The perturbational method used in conjunction with
the quasi-static approximation gives excellent accuracy when properly
used. This shift in resonant frequency caused by the introduction of a
dielectric sample into a resonant cavity can be used to measure the
constitutive parameters of matter.
7-4. Waveguide Perturbations. We shall now consider waveguides
cylindrical in the general sense, that is, all z = constant cross sections
are identical. Figure 7-5a represents a cross section of the unperturbed
waveguide, Fig. 7-5b represents a wall perturbation, and Fig. 7-5c repre-
.sents a material perturbation. All perturbations must, of course, be
independent of z, The guide boundary is taken as perfectly conducting
in all cases.

n n n

c C' c
(a) (b) (c)

FIG. 7-5. Perturbations of cylindrical waveguides. (a) Original cross section;

(b) wall percurbation ; (c) material perturbation.
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 327
At the cutoff frequency a cylindrical waveguide is a two-dimensional
resonator. We should therefore expect formulas similar to those for
perturbations of cavities to apply to waveguides at cutoff. In fact, we
can apply the cavity derivations directly to the region formed by the
cylindrical waveguide bounded by two z = constant planes, changing
only some of the explanations. For example, in deriving Eq. (7-2), the
left-hand side results from the integral

1ft (H X E~ + H~ X E) • ds

taken over the perturbed surface. For a length of a cylindrical wave-

guide at cutoff, the fields are independent of Z; so the surface integrals
over the two z = constant cross sections cancel each other. This leaves
only the surface integral on the left-hand side of Eq. (7-2) taken over the
wall of the waveguide. Following the derivation further, we find that
Eq. (7-3) applies directly for calculating the change in waveguide cutoff
frequency. But both numerator and denominator involve an integration
with respect to z, which reduces to the length of the segment of the cylindri-
cal waveguide. Hence, from Eq. (7-3) we obtain the change in cutoff
frequency ~wc due to an inward perturbation of the waveguide wall as

j J. H X Et • n dl
~wc = ':fliC (7-21)
JJ
s'
(EE • E~ + #LH • Hn ds

where AC is the contour about the volume of the perturbation and S' is
the cross section of the perturbed waveguide (see Fig. 7-5b).
The crude approximation of replacing the perturbed fields E, H by the
unperturbed fields Eo, H o in Eq. (7-21) gives good results for smooth,
shallow perturbations. This leads to

(7-22)

which is analogous to Eq. (7-4). Hence, an inward perturbation of the

waveguide walls at a position of high E will lower the cutoff frequency,
while one at a position of high H will raise the cutoff frequency. For
perturbations not shallow and smooth, we can obtain a better approxi-
mation to Llwc by using a quasi-static approximation for H in the numer-
ator of Eq. (7-21). An example of the perturbation of waveguide walls
is the "ridge waveguide," formed from the rectangular waveguide by
328 TIME-HARMONIC ELECTROMAGNETIC FIELDS

adding ridges along the center of the top and bottom walls.' Such ridges
will lower the cutoff frequency of the dominant mode and will raise the
cutoff frequency of the next higher mode (see Probe 7-12). Hence, a
greater range of single-mode operation can be obtained. The ridges also
decrease the characteristic impedance of the guide; hence, they are used
for impedance matching.
The formulas for material perturbations in cavities can also be special-
ized to the case of material perturbations in waveguides at cutoff. The
reasoning is essentially the same as that used for the wall-perturbation
case. Hence, from Eq. (7-10) we can obtain the exact formula for the
change in cutoff frequency due to a change of matter with the waveguide.
It is
JJ (deE· Et + d#LH • Ht) ds (7-23)
JJ (eE. Et + #LH • Ht) ds
where the integrals are taken over the guide cross section. Note that
an increase in either e or p, can only decrease the cutoff frequency of a
waveguide. If Lle and Llp, are small, we can replace E, H by Eo, H o and
obtain

Llwc ~ Jf + Llp,IHoI2) de
(LlelEol2
(7-24)
w. JJ (elEol2 + #LIB 01 2) ds
This is analogous to Eq. (7-11). If Lle and Llp, are large, but of small
spatial extent, we can improve our approximation by using the quasi-
static method of Sec. 7-3. For example, analogous to Eq. (7-18) we have
in the nonmagnetic case

JJ deE i bt • Et ds
(7-25)
2 JJ elEol2ds
where E i Dt is given by the appropriate one of Eqs. (7-14) to (7-16).
As long as the perturbed guide is homogeneous in e and p" we can
determine the propagation constant at any frequency from the cutoff
frequency according to

(7-26)

1 S. B. Cohn, Properties of Ridge Waveguide, Proc. IRE, vol. 35, no. 8, pp. 783-788,
August, 1947.
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 329
(This is proved in Sec. 8-1.) If the perturbed guide is inhomogeneous,
no such simple relationship exists. In such cases we can obtain pertur-
bational formulas for the change in 'Y. In the loss-free case we can express
the unperturbed fields as
Eo = ~o(x,y)e-i~o.
(7-27)
H o = Ho(x,y)e-i~o.
and the perturbed fields as
E = ~(x,y)e-ilb
(7-28)
H = H(x,y)e-i~:I

The perturbational formulas are then

'" (~t X H) · n dl
'fAC
= - J• ---:--::----'-----------
JI
t:I t:I
tJ - tJO (7-29)
(Et X a + E x at) · u. ds
8

in the case of a wall perturbation, and

JI (~EE. Et + ~Ila • at) ds

~ - fJo = w --=-8=--- _
(7-30)
II (Et x a + E x Ht) · u, ds
8

in the case of a material perturbation. The perturbational formulas in the

lossy case are given in Probs. 7-15 and 7-16.
To illustrate the derivation of the above formulas, consider a material
perturbation. The unperturbed and perturbed fields satisfy Eqs. (7-9)
with w = Wo, for the frequency is kept unchanged. The two equations
following Eqs. (7-9) are still valid, and, with Wo = w, their sum becomes
V • (H X Eri + Hri X E) = jW(~EE • Et + ~ILH • Hri)
Integrating this equation throughout a region and applying the diver-
gence theorem to the left-hand term, we obtain

effi (H X E: + H: X E) • ds = jw III (~EE. Et + ~IlH· Ht) dr (7-31)

This is an identity for any two fields of the
D
same frequency in a region for which E and
IL are changed to E ~E and IL + ~IL. For +
material perturbations in a cylindrical wave- clzI
guide, we express the fields according to Eqs.
(7-27) and (7-28) and apply Eq. (7-31) to FIG. 7-6. DifJerentialslice of a
the differential slice of Fig. 7-6. On the cylinder.
330 TIME-HARMONIC ELECTROMAGNETIC FIELDS

waveguide walls both n X E and n X Eo vanish; so this part of the surface

integral vanishes. Also, since the thickness of the slice is a differential
distance,

ff + ff :z ff
top bottom
= dz
8
= -j({3 - flo) dz ff
S

The right-hand side of Eq. (7-31) can be expressed as the integral over
the cross section times dz; hence Eq. (7-31) reduces to

- j({3 - flo) JJ (:ft X E: + :8:: X E) · u. ds

s
= jw JJ (LlEE. E: + Llp:8: •:8:t> ds
8

Rearrangement of this equation gives Eq. (7-30). In the derivation of

Eq. (7-29), the right-hand side of Eq. (7-31) is zero, and the left-hand
side equated to zero leads to the desired result.
Equations (7-29) and (7-30) as they stand are exact. To use them,
we must make various approximations for E and H, just as we did in
the cavity problems of Sees. 7-3 and 7-4. For example, in the case of
shallow, smooth deformations of waveguide walls, we can approximate
E, It by Eo, Ito in Eq. (7-29). Using the conservation of complex power
[Eq. (1-62)], we arrive at the result

fJ - f30 ~ w _ _~_8
JJ(plb 01 2 - 4~012) ds _
(7-32)
JJ (R: X :8: + :Eo xii:) · u, ds
0
8

(The denominator is twice the time-average power flow in the unper-

turbed guide.) If the perturbation is not shallow and smooth, better
results can be obtained using a quasi-static modification. Similarly, for
small dE and dp. we have the approximation for material perturbations

JJ (LlEIB 01 2+ Llplb01 2) ds
(3 - f30 ~ w 8 . (7-33)
JJ (Et X :8:
8
0 + Eo X :8:t) • ds

For large dE and dp we can obtain better results by using the quasi-static
approximation for the fields within dE and dp.
As an example of the perturbational approach applied to a waveguide
problem, consider a circular waveguide of radius b containing a concen-
tric dielectric rod of radius a. The exact solution to this problem was
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 331
1.0 _ _--.,..---r---r----r-r-----.

0.9
FIG. 7-7. Comparison of
the perturbational solu- .,f 0.8 1-----..---+---+-~~~-+-----1
tion with the exact 80- Qi:-
Iution for the partially 0.7 I---+~~---t--
filled circular waveguide,
E = IOEe, b = O.4Xo• 0.6 J--~-__r--+---+--+-----t

o 0.1 0.2 0.3

alb

considered in Sec. 5-5, and a numerical example is shown in Fig. 5-11.

For the perturbational solution we shall use Eq. (7-30) with dp, = o. In
the numerator we make the quasi-static approximation of Eq. (7-16),
and in the denominator we approximate E, H: by Eo, Ho. The unper-
turbed field of the dominant TEll mode for the circular waveguide is

s, = ~ Jl( 1.841 ~) sin eI> u, = - ~:

E. = 1.~41 J~ (1.841~) cos eI> H. = ~:
where Zo is the characteristic impedance [Eq. (5-32)]. The denominator
of Eq. (7-30) then becomes

;0 ~2'1r del> ~b dp p(E + E.2)

p2 = O.7892~ ~1 _ (:cy
where We is the cutoff frequency. The numerator is easily evaluated as

~:: ~ ~ EO (1.841 ~r
and Eq. (7-30) reduces to
~ -
-k-o-
~o
= VI -
2.146
(W e/W)2
Er -
Er +1
1 (a)2
b
(7-34)

Figure 7-7 compares this solution to the exact solution of Fig. 5-11. Our
approximations give good results for small a/b. At frequencies near the
unperturbed cutoff frequency, the We in Eq. (7-34) may be taken as that
of the perturbed guide.
7-5. Stationary Formulas for Cavities. Suppose we have a resonant
cavity formed by a perfect conductor enclosing a dielectric, possibly in-
homogeneous. The "wave equations" are
v X p,-IV X E - W r 2EE = 0
(7-35)
V X E-IV X H - w r 2J.L H = 0
332 TIME-HARMONIC ELECTROMAGNETIC FIELDS

where W r is the resonant frequency. These reduce to the usual Helmholtz

equations when E and J.L are constants. If the first of Eqs. (7-35) is sealarly
multiplied by E and the resulting equation integrated throughout the
cavity, we obtain

JJJ E · V X }L-IV X E dr
= --JJ-J-E-E-2-dT-- (7-36)
2
r
w

Similarly, multiplying the second of Eqs. (7-35) scalarly by H and inte-

grating throughout the cavity, we obtain

(7-37)

Equations (7-36) and (7-37) are identities, but, even more important,
they are useful for approximating CJJ,. by assuming field distributions in a
cavity. They are particularly well-suited for this latter application
because of their "stationary " character, which we shall now discuss.
We take Eq. (7-36) and substitute for the true field E a trial field

Etrial = E + LlE = E + pe (7-38)

where p is an arbitrary parameter. This procedure gives

w 2 (p) =
JJJ (E + pe) • V X}L-IV X (E + pe) dr
~~-~------------- (7-39)
JJJ E(E + pe) • (E + pe) dr
where we show w2 as a function of p for fixed e. The Maclaurin expan-
sion of w 2 is

w 2 (p ) = w r 2 + p -OW21
op p-O
+ ,p2 02w 21
-2
2. op p-=O
+ .·· (7-40)

Note that the first term is the true resonant frequency, because
= wr 2 • In the variational notation» the above expansion is written
w2 (0 )
as
(7-41)

By definition, each term of Eq. (7-41) equals the corresponding term of

Eq. (7-40). The term Ow 2 is called the first variation of w2, the term 02W2
is called the second variation, and so on. A formula for w 2 is said to be
1 F. B. Hildebrand, "Methods of Applied Mathematics," p.130, Prentice-Hall, Inc.,
Englewood Cliffs, N.J., 1952,
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 333
stationary if the first variation of w2 vanishes. This is equivalent to

(7-42)

The extension to more than one p parameter is straightforward.

We now wish to show that Eq. (7-39) is stationary. The derivative
of the numerator N(p) evaluated at p = 0 is

N'(O) = JJJ (E. V X It-IV X e + e •V X It-IV X E) dT

It is a vector identity that

JJJ E • V X It-IV X e dT = JJJ It-IV X e • V X E dr

+ iffd(It-IV X e) X E] • ds

The last term vanishes, because n X E = 0 on S. A similar identity

states

JJJ It-tv X !; V X E dT = JJJ e· V X It-IV X E dT

-1j) [(It-IV X E) X e] • ds

Using these two identities and the first of Eqs. (7-35), we obtain

N'(O) = 2",,2 JJJ Ee· EdT -1j) [(It-IV X E) X el- ds

The derivative of the denominator D(p) of Eq. (7-39) is, for p = 0,

D'(O) = 2 JJJ Ee· EdT

We then obtain

iJw21 = D(O)N'(O) - N(O)D'(O)

iJp p~O D2(O)
1j) [(It-tv X E) X e] • ds
(7-43)
IJJ EE 2
dT

which has been simplified, using Eq. (7-36). The above equation van-
ishes if n X e = 0 on S, which requires n X E t rial = 0 on S. Hence, Eq.
(7-36) is a stationary formula for the resonant frequency if the tangential
components of the trial E vanish on the cavity walls,
334 TIME-HARMONIC ELECTROMAGNETIC FIELDS

Equation (7-36) can be put into a more symmetrical form by applying

the identity

fJJ E • V X ".-IV X E dr = JJJ ".-IV X E· V X E dr

+ 1ft [(".-IV X E) X E] • ds

The last term vanishes, because n X E = 0 on S. Substituting this

identity into Eq. (7-36), we obtain

(7-44)

This formula proves to be stationary, provided n X Etrial = 0 on S. If

we look carefully at the first variation of Eq. (7-44), it is evident that the
requirement n X Etrial = 0 on S can be relaxed if the term

21ft [(".-IV X E) X E] • ds

is added to the numerator. This gives

JJJ ".-I(V X E)2 dr + 21ft [(".-IV X E) X E] • ds

w
r
2
= JJJ dr EE2 (7-45)

which is stationary, even if n X Etrial ¢ 0 on S. This is an important

modification, because it is not always easy to find a trial field with vanish-
ing tangential components on the cavity walls, especially if the geometry
is complicated. Still further modifications in our formulas are required
if n X E or n X (J.L-1V X E) are discontinuous over some surface within
the cavity. All such modifications can be quite simply effected by the
reaction concept of Sec. 7-7.
A similar procedure shows that Eq. (7-37) is a stationary formula in
terms of H, provided n X (E-1V X H) = 0 on S. The H-field formula
corresponding to Eq. (7-44) is

(7-46)

which turns out to be stationary subject to no boundary conditions on S.

Further modifications to account for discontinuities in n X H or n X
(E-1V X H) over surfaces within the cavity can be made. These modifi-
cations again follow directly from the methods of Sec. 7-7.
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 335

o PI p o PI P
(a) (b)
FIG.7-8. Illustration of w 2 versus p for (a) a stationary formula and (b) a nonstationary
formula.

Let us now briefly consider the advantages of a stationary formula

over a nonstationary one. Figure 7-8 shows pictorally the primary
advantage. Given a class of trial fields of the form of Eq. (7-38), the
parameter w 2 (p ) determined from a stationary formula such as Eq. (7-39)
will have a minimum or maximum at P = 0. 1 This is shown in Fig.
7-8a. The parameter w 2 determined from a nonstationary formula must
have some definite slope at p = 0, as shown in Fig. 7-8b. For a given
error in the assumed field, say LlE = pIe, the corresponding error in the
resonant frequency is Wl 2 - wr 2 • It is evident that for small PI the sta-
tionary formula gives a smaller error in w2 than does the nonstationary
formula. This property is sometimes summarized as follows: " A param-
eter determined by a stationary formula is insensitive to small variations
of the field about the true field." An error of the order of 10 per cent
in the assumed field gives an error of the order of only 1 per cent in the
parameter. In some cases the true field can be shown to yield an abso-
lute minimum or maximum for the parameter. The stationary formula
then gives upper or lower bounds to the parameter. Our formulas for w 2
give upper bounds, as we shall show later.
We might also inquire about the general procedure of establishing
stationary formulas. One characteristic of all such formulas is that the
numerator and denominator contain squares of the trial field. This
insures that amplitude of the trial field will have no effect on the calcu-
lation. Classically, the method of establishing stationary formulas is to
construct formulas of the proper form and then separate the stationary
ones from the nonstationary ones by testing the first variation. In Sec.
7-7 we shall give a general procedure which leads directly to the various
stationary formulas.
1 A complex parameter would have a saddle point at p = o.
336 TIME-HARMONIC ELECTROMAGNETIC FIELDS

N ow let us apply some of our stationary formulas to a problem for

which we have an exact answej, so that we may get an idea of the accu-
racy obtainable. Consider the dominant mode of the circular cavity
(Fig. 5-7), for the case d < 2a. The TM o1o mode is dominant and the
exact resonant frequency is
2.4048
(7-47)
w, - a YEP.

The field is sketched in Fig. 5-8 and is given mathematically by

E. = j i, (2.405~) -H4> = 2.~05 J 1 (2.405~)

Substitution of this true field into any of our stationary formulas must,
of course, give us Eq. (7-47).
Suppose we first try a formula that requires no boundary conditions
[Eq. fi-46)]. Assume as a trial field
H = Uepp v X H = u.2
Equation (7-46) then becomes

w2 =
fo 4pdp -
a
8
Ell fo pvtiP -
a Ep.a2

and our approximation is

2.818
(7-48)
w,. ~ a yEP.

This is 16 per cent too high, which is a relatively poor result. This sug-
gests that our trial field was too crude an approximation. We can
improve our trial field by assuming

H = u, (p - ~:) V X H = u.2 (1 - ~)
which is chosen to satisfy the condition n X E = 0 on S. Equation
(7-46) then yields

foa 4(1 - ~y pdp 180

w2 =
(2 p2)2 = - -2

~
a Ep.31a
EIJ. p - - pdp
o 3a
Q,Dd our approximation is now
2.410
W
r
~ aYEJJ
 PERTURBATIONAL AND VARIATIONAL· TECHNIQUES 337
This is only 0.2 per cent in error. Even though a formula is stationary,
we must use care in choosing trial fields. It is advisable to meet the
physical boundary conditions as closely as possible, for this will help to
obtain a trial field close to the true field. If the same trial field is used in
Eq. (7-37), we again get Eq. (7-49), since n X E = 0 on S.
Now consider a stationary E-field formula, say Eq. (7-44). This
formula requires n X E = 0 on S; hence we choose
1
VXE=uf/>-
a

Substituting this into Eq. (7-44), we obtain

fa e. dp
= l» a 2
6
6'2
fa ( 1 - ~)2 pdp - EILa2
EJ.I. l»

Our approximation is therefore

2.449
W
r
~ a y';p, (7-50)

which is 1.8 per cent too high. If we had chosen a trial E field not
satisfying n X E = 0 on S, we would have had to use Eq. (7-45).
Note that all our approximations are too high. This suggests that the
true resonant frequency is an absolute minimum, which we shall now
show. For example, take Eq. (7-39), and, by means of various identities,
put it into the form

(7-51)

It is known that the eigenfunctions, that is, the fields of the various
modes, form a complete set of orthogonal functions in the cavity space.!
Hence, the error field pe can be expanded in a series

where the Ai are constants and the E, are the various mode fields. Sub-
stituting the above equation into Eq. (7-51), making use of the wave
1 Philip M. Morse and Herman Feshbach, "Methods of Theoretical Physics," part I,

Chap. 6, McGraw-Hill Book Company, Ine., New York, 1953.

338 TIME-HARMONIC ELECTROMAGNETIC FIELDS

equation and the orthogonality relationships, we obtain

W2 _ W 2 =
l (W. 2 - w r 2) A .2
_i_ _---::-::--::-
JJJ EE.2 dr
_
r (7-52)

where the Wi are the resonant frequencies of the ith modes. Since we
have chosen W r as the lowest eigenvalue, Eq. (7-52) is always positive.
Hence, any W calculated from Eq. (7-36) will be an upper bound to the
true resonant frequency. Also, if we choose a trial field orthogonal to
the field of the lowest mode, we have an upper bound to the next higher
resonant frequency, and so on. This, of course, requires that the domi-
nant mode be known exactly, which is seldom the case for complicated
geometries.
Look now at Eq. (7-46). The trial field H = constant vector is a
permissible trial field, since no boundary conditions are required. The
result is W r = 0, which is less than the true resonant frequency [Eq.
(7-47)]. Why do we not have an upper bound in this case? The answer
lies in the fact that we have overlooked the "static mode." A static
magnetic field (wr = 0) can exist in a cavity bounded by a perfect electric
conductor. Fortunately, it is easy to insure that our trial field is orthogo-
nal to all static fields, thereby obtaining an upper bound to the dominant
a-c mode. Any trial field satisfying
V·pH = 0 pH n = 0 on S (7-53)
is orthogonal to all static fields, as we shall now prove. The desired
orthogonality is

JJJ lotH· H.tatic dr = 0

where, in general, Hatatic = - V U. By virtue of the identity

V • (UpH) = pH • V U + UV • pH
the preceding equation becomes

This requirement is met for all U by the conditions of Eq. (7-53). Our
choices for H in the foregoing examples satisfied Eq. (7-53); so we
obtained upper bounds to the dominant TM o1o mode, as desired.
7 -6. The Ritz Procedure. A further advantage of the variational
formulation is that one can choose the best approximation to a stationary
quantity obtainable from a given class of trial fields. This is done by
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 339
including adjustable constants, or variational parameters, in the definition
of the trial field and then choosing those parameters which will give a
minimum or maximum of the stationary quantity. For example, if we
choose
(7-54)
where the Ai are variational parameters, and substitute into the station-
ary formula Eq. (7-36), we obtain
(7-55)
The best approximation to wr 2 will be the minimum value of w 2 , which
can be chosen by requiring

i = 1,2, . . . , n (7-56)

This general method is known as the Ritz procedure.1

The most common way to include variational parameters is to express
the trial field as a linear combination of functions
(7-57)
Since the labor of the calculations increases approximately as the square
of the number of terms in Eq. (7-57), it is desirable to keep n small.
However, it is also necessary that some choice of the Ai will give a
reasonably close approximation to the true field. When a complete set
of functions E, is used, the method may, in principle, lead to an exact
solution. It is also sometimes convenient to choose the E, as an orthog-
onal set.
For an example of the Ritz method, let us again consider the circular
cavity of Fig. 5-7 and trial fields of the form
v X H = u z (2 + 3Ap) (7-58)
where A is a variational parameter. Note that H satisfies no boundary
conditions on S; so we choose Eq. (7-46) as the stationary formula.
Substituting the trial field into Eq. (7-46), we obtain

",2 =
h (2 + 3Ap)2p dp
......=..-
d

EP. fo (p + A p2)2p dp
d

15 8 + 16Aa + 9(Aa)2 (7-59)

= a 2fJJ. 15 + 24Aa + 10(Aa)2

1 The method is also referred to as the U Rayleigh-Ritz procedure."

340 TIME-HARMONIC ELECTROMAGNETIC FIELDS

Note that the approximation of Eq. (7-49) is the special case Aa = -%.
To determine A by the Ritz method, we set
aw 2

aA = 0
and obtain 24 + 55Aa + 28(Aa)2 =0
This can be solved for Aa as

A = -55 ± Vm = { -1.3100 (7-60)

a 56 -0.6543
A substitution of the second of these values into Eq. (7-59) gives
2.4087
w=--- (7-61)
a VEP.
which is smaller than what the first of Eq. (7-60) gives. Hence, Eq.
(7-61) is the desired "best" approximation to the true resonant fre-
quency [Eq. (7-47)]. The solution Aa = -1.31 gives ka = 7.191, which
is an approximation to the next higher eigenvalue 5.520. If the trial
field has two variational parameters, we obtain approximations to the
lowest three eigenvalues, and so on. The Ritz procedure also gives us
an approximation to the true field, but it is difficult to establish the
nature of the approximation.
7-7. The Reaction Concept. 1 A general procedure for establishing
stationary formulas can be obtained, using the concept of reaction as
defined in Sec. 3-8. To reiterate, the reaction of field a on source b is

(a,b) = J(Ea • dJb - Ha • dMb) (7-62)

If all sources can be contained in a finite volume, the reciprocity theorem

[Eq. (3-36)] is
(a,b) = (b,a) (7-63)

The linearity of the field equations is reflected in the identities

(a,b + e) = (a,b) + (a,e) (7-64)

(Aa,b) = A(a,b) = (a,Ab)

where the notation Aa means the a field and source are multiplied by
the number A.
Many of the parameters of interest in electromagnetic engineering are
proportional to reactions. For example, the impedance parameters of a
1 V. H. Rumsey, The Reaction Concept in Electromagnetic Theory, PhY8. Reo., sere
2, vol. 94, no. 6, pp. 1483-1491, June 15, 1954.
 PERTURBATIONAL AND VARIATIONAL TECltNIQl1ES 341
multiport "network" are proportional to reactions, as shown by Eqs.
(3-41). Approximations to the desired reactions can be obtained by
assuming trial fields (or sources) to approximate the true fields (or
sources). It is then argued that the best approximation to a desired
reaction is that obtained by equating reactions between trial fields to
the corresponding reactions between trial and true fields. To be specific,
suppose we want an approximation to the reaction (Ca,Cb). (The symbol
C stands for "correct.") The approximation (a,b) is then best if we
subject it to
(7-65)
because we have imposed all possible constraihts. Equation (7-65) can
be thought of as the statement that all trial sources look the same to
themselves as to the correct sources.
The reaction (a,b) obtained from Eq. (7-65) is also stationary for small
variations of a and b about c; and Cb. This we can prove by letting

and showing that

a(a,b)
apa
IPG-=Pb=O -
- a(a,b)
apb
I -
pG=Pb Z=O -
0 (7-66)

Substituting for a and b into Eqs. (7-65), we have the three relationships
(a,b) = (Ca,Cb) + Pa(ea,cb) + Pb(Ca,eb) + PaPb(ea,eb)
= (Ca,Cb) + Pb(ca,eb)
= (Ca,Cb) + Pa(ea,cb)
Using the last two equations in the first equation, we obtain
(a,b) = (Ca,Cb) - PaPb(ea,eb)
It is now evident that Eqs. (7-66) are satisfied, proving the stationary
character of (a,b).
We have a slightly different case when the reaction concept is used to
determine resonant frequencies of cavities. The true field at resonance
is a source-free field; so the reaction of any field with the true source is
zero. Hence, if we let a = b represent a trial field and associated source,
Eq. (7-65) reduces to
(a,a) = 0 (7-67)
We can think of this as stating that the resonant frequencies are zeros
of the input impedance.
To apply Eq. (7-67), we assume a trial field and determine its sources
from the field equations. For example, an assumed E field can be sup-
342 TIME-HARMONIC ELECTROMAGNETIC FIELDS

(a) (b) (c)

FIG. 7-9. Sources needed to support (a) a trial E field, (b) a trial H field, and (c) both a
trial E field and a trial H field.

ported by the electric currents

J = -jwEE - ~V X (p-lV X E) (7-68)

JW '
However, if the trial field does not satisfy n X E = 0 on S, we need the
additional magnetic surface currents
M. = n X E onS (7-69)
to support the discontinuity in E at S. This is illustrated by Fig. 7-9a.
We now substitute from Eqs. (7-68) and (7-69) into Eq. (7-67) and obtain

o= (a,a) = IfI J. E dr + 1fi M. · C~p V X E) ds

-jw III EE • E dr +~ IIIE. V X (p-tV X E) dr

- ~ 1f (n X E) • (p-tV X E) ds
If n X E = 0 on S, this reduces directly to Eq. (7-36). If n X E ~ 0
on S, the above equation reduces to Eq. (7-45).
If a stationary formula in terms of the H field is desired, we consider
the trial field to be supported by the sources

M = -jwp,H - ~V X (E-1V X H)
JW
(7-70)
M, = n X (!-
JWE
V X H) onS

as represented by Fig. 7-9b. Application of Eq. (7-67) now leads to Eq.

(7-46), or to Eq. (7-37) if M. = o.
Stationary formulas in terms of both E and H are also possible. This
time we consider both electric and magnetic currents, as shown in Fig.
7-9c. They are found from the trial fields according to
J= - jWEE +V X H
M = - jwp,H - V X E (7-71)
M. = n X E on S
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 343
Equation (7-67) then gives

o= fff (E J - HoM) dT -effi HoM. ds

0

ffJ(-jWEE2 + E H + H E + jwp.H2) dT
0 V X 0 V X

-effiE X Hods
which can be rearranged to

W
.fJJ (E
=J
0 V X H +H·V X E) dT -effi E X H · ds
(7-72)
fJJ (p.H2 - EE2) dT

This is sometimes called a "mixed-field" stationary formula. The minus

sign in the denominator might seem strange, but it is easily shown that
E and Hare 90 0 out of phase in the loss-free case (see Sec. 8-4). Hence,
the denominator is twice the stored energy in the cavity.
Finally, if the trial fields have discontinuities in n X E or n X Hover
surfaces within the cavity, we must add the appropriate surface currents
to support the discontinuities. This procedure leads to additional sur-
face integrals in the stationary formulas, as shown in Probs. 7-27 and
7-28.
Earlier we showed that reactions constrained according to Eq. (7-65)
were stationary. But in the above cavity formulas we calculated w by
forcing the reaction to vanish. We shall now prove that the w so deter-
mined is stationary about the true resonant frequency. In the usual
manner, we let the trial field be the true field plus a parameter times an
error field, represented by
a=c+pe
For fixed e the reaction (a,a) is a function of both wand p. Equation
(7-67) constrains (a,a) to vanish; hence, as wand p are varied, we have

The second term of this equation vanishes because (a,a) is stationary

about p = O. The coefficient of the first term is not in general zero; so
~w = 0
Thus, the first variation of ta vanishes, and all formulas for w derived
from Eq. (7-67) are stationary.
The reaction concept also provides us with an alternative way of
viewing the Ritz procedure for improving the trial field or source. We
344 TIME-HARMONIC ELECTROMAGNETIC FIELDS

assume the trial field or source to be a linear combination of functions,

represented by
a = Uu + Vv + · · · (7-73)
where U, V, ... are numbers to be determined. According to the
reaction concept, all trial fields should look the same to themselves as
to the true source; hence we should enforce the conditions
(a,u) = (c,u)
(a,v) = (c,v) (7-74)

Substituting from Eq. (7-73), we obtain the set of equations

U(u,u) + V(v,u) + · .. = (c,u)
U(u,v) + V(v,v) + ··· = (c,v) (7-75)

which can be solved for the parameters U, V, . . .. The solution so

obtained is identical to that obtained by the Ritz procedure.
To illustrate, let us reconsider the example of Sec. 7-6, which was the
Ritz procedure applied to the circular cavity (Fig. 5-7). Our trial field
was Eq. (7-58); so for the same approximation by the reaction concept
we choose
(7-76)
The sources of these fields, according to Eq. (7-70), are
2·
M.u = 2
WE
(7-77)
M." = 3ja
WE

Calculating the various reactions according to Eq. (7-62), we obtain

a2 2 )
(u,u) = 21rda 2 ( jwIJ. 4" + jWE
a2 2 )
(u,v) = (v,u) = 21rda 3 ( j WIJ."5 + jWE (7-78)

(v,v) = 21rda
4
0 ~ + j;E4)
wIJ.

All reactions with the correct source are zero, because the true field is
source-free. Hence, (c,u) = (c,v) = 0 and Eqs. (7-75) reduce to
U(u,u) + V(v,u) = 0
U(u,v) + V(v,v) =0
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 345
These equations can have a jiontrivial solution only if the determinant
of the coefficients of U and V vanishes. Hence,
(u,u)(v,v) - (U,V)2 = 0 (7-79)
is the equation from which w is to be found. The solution of Eq. (7-79),
with the reactions of Eqs. (7-78), yields Eq. (7-61).
'1-8. Stationary Formulas for Waveguides. At cutoff, a waveguide is
a two-dimensional resonator; so we should expect stationary formulas
for the cutoff frequencies to be of the same form as those for the resonant
frequencies of cavities. We must, of course, be careful in applying the
reciprocity theorem, because the sources of our trial fields are not of
finite extent. However, if we take a slice of the waveguide, as was done
in Sec. 7-4, surface integrals over the top and bottom just cancel at
resonance. The height of the slice is common to all terms and therefore
cancels. Starting from Eq. (7-67), we arrive at stationary formulas dif-
fering from our cavity formulas only in that volume integrals are replaced
by surface integrals and surface integrals by line integrals. Hence, the
E-field formula corresponding to Eq. (7-45) is

ff p-l(V X E)2 ds + 2 ¢ [(p-lV X E) X E] • n dl

we
2
= 1/ EE2 de
(7.-80)

where n is the outward-pointing unit vector normal to the waveguide

walls. The H-field formula corresponding to Eq. (7-46) is
f1 E-1(V X H)2 ds

f1pH2 ds
W 2 - -. (7-81)
C -

and the mixed-field formula corresponding to Eq. (7-72) is

.f1(E· V X H + H · V X E) ds - ¢ E X H · n dl
We =J
! /
(pH2 - EE2) d$
(7-82)

N one of the above formulas require boundary conditions on the trial

fields. Corrections for discontinuous trial fields can be made as out-
lined in the preceding section.
As an example, consider the partially filled rectangular waveguide of
Fig. 4-8a. In Sec. 4-6 we obtained a transcendental equation for the
cutoff frequency [Eq. (4-51)]. For a variational solution, let us use Eq.
(7-80) and a trial field
. rX
E = Ut/SIn-
a
346 TIME-HARMONIC ELECTROMAGNETIC FIELDS

which is the empty-guide field. The result is!

We = 1r_ [1 + El - E2 (~ _ ~ sin 21rd)]-~ (7-83)

a VE2J.L2 E2 a 21r a
Note that this is an explicit formula for We, in contrast to the exact equa-
tion, which is transcendental. Table 7-2 compares the above result with
the exact solution for the case El = 2.45EO and E2 = Eo. We should expect
the approximation to become worse as El/E2 becomes larger, since the
field then tends to concentrate more in the dielectric.
TABLE 7-2. RATIO OF WAVEGUIDE WIDTH TO CUTOFF WAVELENGTH FOR THE
RECTANGULA"ll WAVEGUIDE WITH DIELECTRIC SLAB
("Exact" values read from curves by Frank)

d/a a/Xc (exact) a/Xc (approximate)

0 0.500 0.500
0.167 0.485 0.486
0.286 0.450 0.453
0.500 0.375 0.383
0.600 0.350 0.352
1.000 0.319 0.319

A knowledge of the cutoff frequency of a waveguide homogeneous in

E and p, is sufficient to determine the propagation constant at any other
frequency according to Eq. (7-26). If the guide is inhomogeneously
filled, as for example the above-treated rectangular waveguide with
dielectric slab, there is no simple relationship between the cutoff fre-
quency and the propagation constant. We therefore have need of sta-
tionary formulas for propagation constants.
In all of the previous examples, the field equations were given by an
operator which was self-adjoint with respect to the desired integration.t
For inhomogeneously filled waveguides, the field equations lead to an
operator which is not self-adjoint. Hence, an appropriate adjoint opera-
tor must be found and the derivation of the stationary formulas suitably
modified. It turns out that the operator for waves traveling in the -z
direction is the adjoint of the operator for waves traveling in the +z
direction, and the derivation proceeds as follows.
Define +z traveling waves as
E+ = E+(x,y)e-i1b = eE t + uz~z)e-i{Jz
(7-84)
H+ = H+(x,y)e-i{Jz = (Itt + u zf1z)e-i{Jz
1 A. D. Berk, Variational Principles for Elect.romagnet.ic Resonators and Wave-
guides, IRE Trans., vol. AP-4, no. 2, pp. 104-110, April, 1956.
J B. Friedman, "Principles and Techniques of Applied Mathematics," John Wiley

and Sons, Inc., New York, 1956, p. 44.

 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 347
Substituting these into the field equations, we obtain
V X E+ +
jWJJ.H+ = jfju. X ~+
V X H+ - jWER+ = j{3U. X H+
Using analogous definitions for -z traveling waves, we find
V X ~- + jWJJ.H- = -j{3u. X~-
V X H- - jWEE- = -j{3u. X H-
By direct substitution, it can be shown that for any +z traveling wave
solution there exists a -z traveling wave solution given by
E- = ~-(x,y)eif3. = (Rt - uz2z)eif3z
(7-85)
H- = fI.-(x,y)e if3Z = (-He + uzf1z)eif3z
where the ~t, lIt, "2z , and fl. of Eqs. (7-84) and (7-85) are the same
functions.
Now multiply the first of the +z wave equations scalarly by H-, and
the second of the -z wave equations by E+, and add the two resultant
equations. This gives
fI.- · V X ~+ + E+ • V X· a- + jWJJ.H- • H+ - jWEE+ • E;-
= - 2j{3ftt X Ht • U z
which, when integrated over the guide cross section and rearranged, yields

P=
JJ (CAlEE+. E- - CAlI£H+ • H-
~----------:--::----------------
+ jH- • V X E+ + jE+ • V X H-) ds

2 JJ E X H t t • u, ds
(7-86)
This is a mixed-field formula, stationary if n X ~ = 0 on C.
For the E-field formulation, eliminate H from the +z and -z wave
equations, and proceed as in the derivation of Eq. (7-86). The resultant
formula is

/32 JJ 1£-1E 2 de - j2/3 JJ 1£-IE V E. ds

t t •

+ JJ [1£-l(V X t+) · (V X t-) - CAl 2EE+ • t-] ds = 0 (7-87)

stationary if n X E = 0 on C. The H-field formula is given by Eq. (7-87)

with E, JJ., E replaced by JJ., E, H, and it is stationary with no boundary
conditions on H. Equations (7-86) and (7-87) remain stationary in the
lossy case, for which j{3 should be replaced by 'Y = a + j{3.
For an example of the calculation of propagation constants, consider
the centered dielectric slab in a rectangular waveguide, as shown in the
insert of Fig. 7-10. As a trial field, take
t · 1rX
= u, SIn a
348 TIME-HARMONIC ELECTROMAGNETIC FIELDS

1.6
Y~ ~dJ+-
1.4
BEJI
~a~
E = 2.45 EO
x
dfa = 1.0
-0.5
0.3
1.2
0

~
co.
o
0.8

0.4

o 0.2 0.4 0.6 0.8 1.0 1.2

ciAo
FIG. 7-10. Comparison of approximate and exact propagation constants for the
rectangular waveguide with centered dielectric slab, E = 2.45Eoo (After Berk.)

and use Eq. (7-87). The result is!

!k« = [1 + ~ (~ + !
EO a 1r
sin .1rd)
a
- (~)2J~
koa
(7-88)

The exact solution is given in Probe 4-19 and requires the solution of a
transcendental equation. A comparison of a values obtained from Eq.
(7-88) with the exact values for /3lko is shown in Fig. 7-10 for the case
E = 2.45EO.
7-9. Stationary Formulas for Impedance. A formula for impedance
in terms of reaction is given by Eq. (3-41). Such a formula, when con-
strained according to Eq. (7-65), is a stationary formula for impedance.
Figure 7-11 represents a perfectly conducting antenna excited by a
current source. The resultant current on the antenna will distribute
itself so that tangential components of the total electric field vanish on
the conductor. The antenna terminals are close together; so the reaction
of any field with the current source is of the form - VI. If a trial-current
distribution J"a is assumed on the antenna, the formula for input imped-
ance [Eq. (3-41)] is

(7-89)

l13erk, op. cit.

 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 349
where I is the input current. The impedance as calcu-
lated by Eq. (7-89) is stationary about the true current,
as we shall now show. On the antenna surface, the tan-
gential components of the true field Ee are zero except at
the input; hence
(c,a) = - Vel = - I 2Zin = (a,a)

Also, (c,a) = (a,c) by reciprocity; so the constraints of

Eq. (7-65) have been met, and Eq. (7-89) is a stationary
formula.
Equation (7-89) was used to calculate impedance be-
fore its stationary character was noticed.' This method
should not be confused with the induced emf method FIG. 7-11. An
antenna excited
Zin = - 1:1 effi E • J: ds
2
(7-90) by a current
source.
which is based on the conservation of complex power.
Equation (7-90) is not stationary unless both the true current and the
trial current are real. When the trial current is assumed real, we get
the same answer from Eqs. (7-89) and (7-90). Hence, the input imped-
ances for waveguide feeds calculated in Sec. 4-10 are also variational
solutions to the same problems.
If we have two sets of input terminals, as, for example, in the case of
the two linear antennas shown in Fig. 7-12, the variational formula for
mutual impedance is

Zoo = - (a,b) = -1_ ~ Ea . Jib de (7-91)

i.t, Ialb'tr
where I; and L, are the input currents at terminals a and b, respectively.
The demonstration that the constraints of Eq. (7-65) are met is similar
to that for self-impedance. Note that Eq. (7-91) involves the assump-
tion of currents due to both sources, since Ea is the field of J.a. The
extension to N sets of terminals is straightforward.
The calculation of mutual impedance is usually simpler than the calcu-
lation of self-impedance because the source and field points are separated.
Let us therefore take a mutual-impedance problem as our first example.
Consider the parallel linear antennas of length X/2 as shown in the insert
of Fig. 7-12. No appreciable error will be incurred by assuming the cur-
rents as filamentary, as long as the antenna diameters are small compared
to wavelength and compared to antenna separation. Let the z axis lie
1 P. S. Carter, Circuit Relations in Radiating Systems and Applications to Antenna
Problems. Proc, IRE, vol. 20, no. 6, pp. 1004-1041, June, 1932.
350 TIME-HARMONIC ELECTROMAGNETIC FIELDS

80,----..,.-----r------,------------

d
60 T
>"/2 I« [b

40 1
en
E
aC 20
0

-20

-40
FIG. 7-12. Mutual impedance ZGb = RGb + jXGb between parallel X/2linear antennas in
free space.

along antenna a, and assume

27rZ 27rZ
fa. = i, cos T Ib - I b cos- (7-92)
A
Our formula for mutual impedance [Eq. (7-91)] becomes, in this case,

Zab = - -1- jA/4 E/"lb dz

lalb -A/4
By the usual vector-potential method we have
1
Eza. = -.-
JWE
(a
:i2
vZ
2
+k 2)
A.a

where, at antenna b,
1 jA/4 e-ikyd2+(z-z')2
Aza = - la(z') dz'
47r -A/4 vi d 2 + (z - Z')2

Substituting for Eza and [b in our expression for Zoo, we obtain

A/4 dz jA/4 27rZ 27rz'
Zab = - j -A/4 -A/4
dz' cos ~ cos ~ G(z,z')
1\ 1\
(7-93)

where G(z z') = -- -

1 (a 2
+ k vie-ikYdt+(z-~')J
)
2
.
(7-94)
, 47rjWE dZ 2 d + (z - 2 ZI)2
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 351
The integrations of Eq. (7-93) can be expressed in terms of sine integrals
and cosine integrals. The details of the integration can be found in the
literature. I Letting

we obtain for the result

Rob = 4: {2 Ci(kd) - Ci[-\/(kd)2 + 2 + 11"2] - Ci[V(kd)2 + ... 2 - ...2])

X ob = ~T/ {2 Si(kd) - Si[V(kd)2 + 2 + 11"2] - Si[V(kd)2 + '11"2 - 1I"2]l

(7-95)
where Ci(x) and Si(x) are as defined in Probe 2-44. Figure 7-12 shows a
plot of Eqs. (7-95). The mutual impedance between linear antennas of
other lengths and orientations can be fOUQd in the Iiterature.l-"
The evaluation of the self-impedance of a linear antenna is more diffi-
cult because of the singular integrands encountered. Let us use this
problem to illustrate the use of adjustable parameters in the trial cur-
rent. The geometry of the center-fed linear antenna is shown in the
insert of Fig. 7-13. Let the current on the antenna be represented by
two functions, according to Eq. (7-73). Our trial current is then a sur-
face current of the form
(7-96)
where U and V are adjustable parameters. According to the reaction
concept, the trial functions should look the same to the assumed current
as to the true current; hence we enforce the conditions
(a,u) = U(u,u) + V(v,u) = (c,u)
(a,v) = U(u,v) + V(v,v) = (c,v)
where (c,u) and (c,v) can be calculated, as we shall later show. Solving
for U and V, we have in matrix notation

Substituting for U and V into Eq. (7-96) and calculating the self-reaction,
we obtain
(a a) = [(c u) (c v)] [(U,U) (V,U)]-l [(C,U)] (7-98)
, "(u,v) (v,v) (c,v)
1 P. S. Carter, Circuit Relations in Radiating Systems and Applications to Antenna
Problems, Proc, IRE, vol. 20, no. 6, pp. 1004-1041, June, 1932.
J G. Brown and R. King, High Frequency Models in Antenna Investigations, Proc.
IRE, vol. 22, no. 4, pp. 457-480, April, 1934.
352 TIME-HARMONIC ELECTROMAGNETIC FIELDS

5000
I Az
In

4000
~I-

--
T
L =::z=o I Lja
I I
= 22,000

I----~
l~
+I ~2a

t/) 3000
E
.s::::
0

~
.S
2000
Jr'\
I ~Lja = 1800

I} \
1000
) -, ~ ~L/a
I
= 150
I
)
r-, ~
0 - ~V ~ V
(a)
2400
1\1
1600 V' Lta = 22,000
)
7/ r\ /~
t/)

E
800
)~ r - Lfa = 1800
~/
.s::::
0
0 ~~ 1 ~V
l# ~ ~"
r: L/a = 150\\
>< /

- 800
Y"h / l~ .:
vI
/
1// I
-1600
r; V
1/
j
-2400 \I
V
o 2 4 6 8 10 12
kL
(b)

FIG. 7-13. Variational solution for the input impedance of the symmetrical cylindrical
antenna. (After Y. Y. Hu.) (a) Input resistance; (b) input reactance.
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 353
Equations (7-97) and (7-98) also apply to the case of N adjustable con-
stants if the various matrices are extended to N rows and/or columns.
Expanding Eq. (7-98), using the reciprocity condition (u,v) = (v,u), we
obtain
(a a) = (C,U)2(V,V) - 2(c,u)(c,v)(u,v) + (C,V)2(U,U)
, (u,u)(v,v) - (U,V)2
N ow note that n X Ec = 0 on the antenna surface except at the feed; so
(c,x) = - VinI z
for any x, where Yin is the input voltage and L; is the x current at the
input. Using the above two relationships in Eq. (7-89), we obtain
z. = z. 2 I v2(u,u ) - 2IuIv(u ,v) + I u2(v,v)
111 In (U,V)2 - (U,u)(v,v)
which can be rearranged to read
z. = (U,V)2 - (u,u)(v,v)
(7-99)
In I v2(u ,u ) - 2IuIv(u ,v) + I u2(v,v)
where 111, and I; are the values of the u and v trial currents at the input.
Let us now look at the form of the reactions. The currents will be
rotationally symmetric z-directed surface currents on the cylinder p = a,
where a is the antenna radius. These currents can be expressed as

(7-100)

where Ix is the total current and x = u, v. By the potential integral

method we can calculate the field of the current J.z as
1
Ezx = - 8
2·
( k2 iJ2)
+ ';'2 jL/2
dz' ~2r deb'J.xG (7-101)
1r 'JWE oz -L/2 0

e-jk.yp2+a2-2pa cos (cP-q/)+(z-z')J

where G = -,====================================================== (7-102)
V p2 + a 2
- 2pa cos (eb - eb') + (z - Z')2
The various reactions of Eq. (7-99) are then given by

(x,y) = j L/ 2
-L/2
dz
~2r
0
a deb E.z J.u (7-103)

where E.z is given by Eq. (7-101) with p = a. Note the singular nature
of the Green's function [Eq. (7-102)] at p = a.
A precise evaluation of Eq. (7-103) would be difficult; so the following
approximation is usually used. The field of the current is approximated
by the field of a filamentary current of the same magnitude. This is
354 TIME-HARMONIC ELECTROMAGNETIC FIELDS

equivalent to replacing Eq. (7-102) for p = a by

e-j1cyai+(z-z')i
G=--;===========;:::= (7-104)
vi +
a2 (z - Z') 2

For thin antennas, the error introduced by Eq. (7-104) is negligible, as

can be shown by the following argument. The field of the filament of
current is a source-free field in the region external to the linear antenna.
We can therefore assume that this field exists and calculate the equiv-
alent currents on the surface of the antenna. As long as the equivalent
magnetic currents are negligible, as they will be for thin antennas, we can
take the equivalent electric currents for our trial currents. The resultant
current is essentially that of Eq. (7-100). Using the above approxi-
mation for G, we obtain from Eq. (7-103)

(x,y) = -~- fL/2

411"JWE -L/2
dz fL/2
-L/2
dz' I"'(z')II/(z) (k 2
2
+ iJ0z_)
2
G (7-105)

where G is given by Eq. (7-104). Note that, to this approximation, the

self-reaction is equal to the mutual reaction between two identical anten-
nas fed in phase and separated by a distance a. Hence, Eqs. (7-95) with
d replaced by a give the first-order (one trial function) variational solu-
tion for the input impedance of a "X/2 linear antenna. In particular,
note that for very small a = d, Eqs. (7-95) reduce to
(7-106)
as is evident from Fig. 7-12. Resonance (X = 0) occurs for L slightly
less than X/2.
For trial functions in the second-order solution,

t- = Sink(~ -lzl)
(7-107)
I" = 1- COSk(~ -lzl)
have been used in the literature. The evaluation of Eq. (7-105) for
(x,y) = (u,u), (u,v), and (v,v) is long and involved, and formulas in terms
of sine integrals and cosine integrals have been given by Storer! and Hu. 2
Numerical values of the input impedance are given in Fig. 7-13. The
antenna is said to be resonant when X is zero and kL ~ n1l", n odd. It is
said to be antiresonant when X is zero and kL =
n1r, n even. N ote that,

1 J. E. Storer, Variational Solution to the Problem of the Symmetrical Cylindrical

Antenna, Cruft Lab. Rep. TR 101, Cambridge, Mass., 1952.
I Y. Y. Hu, Back-scattering Cross Sections of a Center-loaded Cylindrical Antenna.

IRE Trans., vol AP-6, no. 1, pp. 140-148, January, 1958.

 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 355
in the vicinity of resonance, R is in- E = Ei + E.
sensitive to antenna thickness. It
is in these regions that the analysis tsource
of Sec. 2-10 gives good results. Both
trial currents of Eqs. (7-107) are zero
at the input for kL = 411". Hence, FIG. 7-14. Wave scattering by an ob-
stacle.
the input impedance calculated there-
from cannot be valid in the vicinity
of kL = 411". Perhaps a better choice for the v current would be

I" = ~ -izi
which is finite at z = 0 for all L > o. However, calculations have not
been made for this choice.
7-10. Stationary Formulas for Scattering. Let us first treat the back-
scattering, or radar echo, type of problem by the variational method.
The problem is represented by Fig. 7-14. It consists of a source and one
or more obstacles, and we wish to determine the field scattered back to
the source. For simplicity, the obstacle will be considered a perfect con-
ductor and the source a current element Il. The more general case of
dielectric obstacles is considered in Sec. 7-11.
Let the incident field, that is, the free-space field of the source alone,
be denoted by E'. The total field E with the obstacle present is then the
sum of the incident field Ei plus the scattered field E·. The reaction of
the scattered field on the current element is
(s,i) = IlE ,• = - IV· (7-108)
where V· is the scattered voltage appearing across l. Let the echo be
defined as the ratio of E l • to Il. Then, using reciprocity, we have
E z• (s,i) (i,s)
Echo = It: = (Il)2 = (Il)2
= (I~)21fi E' · J. d8 (7-109)

where J. is the current induced on the perfectly conducting obstacle.

The boundary condition at the obstacle is n X E = 0, or
n X E' = -n X E· on S (7-110)
Hence, Eq. (7-109) can be written as
-1 ~ (c,c)
Echo = (Il)2 'ff E· • J. ds = - (Il)2 (7-111)

where (c,c) stands for the self-reaction of the U correct" currents induced
on the obstacle by the source.
356 TIME-HARMONIC ELECTROMAGNETIC FIELDS

For a stationary formula, we assume a current JO on S and approximate

(c,c) by (a,a), subject to the constraint
(a,a) = (c,a) = -(i,a) (7-112)

The last equality results from Eq. (7-110). To express this constraint in
a form for which (a,a) is insensitive to the amplitude of Ja, ,ve take
(i,a)2
( a,a ) =- ()
a,a

and, replacing (c,c) by (a,a) in Eq. (7-111), we have

-(i a)2 (effi Ei. Ja ds y (7-113)

Echo ~ (Il)2(a,a) =
(Il)2effi s- · Ja ds
where Eo is the field produced by the assumed currents Ja. This is the
variational formulation of the problem. Note the close similarity of the
echo problem to the impedance problem of the preceding section. The
impedance problem is essentially an echo problem for which the source is
at the obstacle. A more general formulation of the echo problem can be
made by replacing Il with an arbitrary source.
The tensor Green's functions of Sec. 3-10 can be used to put Eq. (7-113)
into a more descriptive form. Define [I'(r.r'j] as the tensor of propor-
tionality between a current element dJa at r' and the field dEo that it
produces at f, that is,

Then Eq. (7-113) can be written as

Echo ~
- [izeffi Ei(r) • Ja(r) ds r
{fi ds{fi ds' Ja(r) • [r(r,r')] Ja(r')
-

This equation is in a form characteristic of variational solutions in general.

A commonly calculated parameter is the echo area, defined by Eq.
(3-30). For linearly polarized fields, the echo area is given by

(7-114)

If, in Fig. 7-14, we let Il be z-directed and located on the x axis, and then
let T = X --+ 00, we have, in the vicinity of the obstacle,
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 357
2.5

2.0
~z
~i
lEi L
1.5 Hi
_1 \ L/a
~
= 150

-- -
~

-+t t+2a ~

1.0 /
~ ~I\
J
~
j
fA\~a = 1800
0.5

J 1\
"'--- -
!J ~ ~
j ~
...............
~
.............. ~V
IJ) .......
"--= -l-.....J~
L/a
I
= 22,000
I
o 2 4 6 8 10 12
kL
FIG. 7-15. Broadside echo area A. of a wire. (After Y. Y. Hu.)

Also, by definition, we have echo = Ez//Il; hence from Eq. (7-113)

7]E 0 u. · JBe ib ds) 2
E z· = ---::.....::.=:..--------=--
(1ft
j2XT EB • JB ds 1ft
Therefore, by Eq. (7-114), our stationary formula for echo area is

A = 1r -
(1ft J.Be ds
7] i k% r 2
(7-115)
• X 1ft EB • JB ds
when the incident plane wave is z-polarized and -x traveling.
As an example, consider the scattering of a plane wave by a thin con-
ducting wire, as represented by the insert of Fig. 7-15. The integral in
the denominator of Eq. (7-115) is just the self-reaction of the assumed cur-
rent on the wire. This is the same type of reaction that we encountered
in the linear-antenna problem, approximated by Eq. (7-105). Defining
A as the self-reaction, we have
A = AI: Eo. JO ds ~ -~- fL/2 dz fL/2 dz' [B(Z)[B(Z') (k 2 + ~) G
':tr 41rJWE -L/2 -L/2 dZ2
(7-116)
358 TIME-HARMONIC ELECTROMAGNETIC FIELDS

where G is given by Eq. (7-104). For the current on the wire we should
expect a constant current" forced" by the incident field plus a "natural-
mode" sinusoidal current. At the ends of the wire, the current should
be practically zero; hence we assume for our trial current
L
[a. = cos kz - cos k- (7-117)
2
Equation (7-116) can then be evaluated as

Re (A) = -: [(kL + kL cos kL - 2 sin kL) Si (kL)

+ log 2')'kL - Ci (2kL) - sin" (kL)]
(7-118)
1m (A) = ~1rT'f {(kL + kL cos kL ~2 sin kL) [ Ci (kL) + log 'Y~a ]
+ Si (2kL) - (1 + cos kL) sin -l
where 'Y = 1.781. The integral in the numerator of Eq. (7-115) evalu-
ates to

which defines B.
I L/ 2
-L/2
[a(z) dz = -1 (2k
k
sinL
- - kL cos -kL) = B
2
Hence, the echo area is
2
-
k
(7-119)

Ae = 1~3l :;7'/ r (7-120)

with A and B given by Eqs. (7-118) and (7-119). This solution gives
good accuracy out to about kL = 8. Figure :7-15 shows a plot of Ae/~2
for the second-order solution (two trial functions), as calculated by
Y. Y. Hu.! The results for plane waves incident at an arbitrary angle
are given by Tai." He also shows the effect of choosing different trial
functions.
In two-dimensional problems, the quantity echo width L e corresponds
to the echo area of the three-dimensional problems. The echo width is
defined as the width of incident wave which carries sufficient power to
produce, by cylindrically omnidirectional radiation, the same back-
scattered power density. In equation form, the echo width is

i. = lim (21r p ~~) (7-121)

p-+ 00 S'
1 Y. Y. Hu, Back-scattering Cross Section of a Center-loaded Cylindrical Antenna,
IRE Trans., vol. AP-6, no. 1, pp. 140-148, January, 1958.
I C. T. Tai, Electromagnetic Back-scattering from Cylindrical Wires, J. Appl. Phy.,

vol. 23, no. 8, pp. 909-916, August, 1952.

 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 359
or, for linear polarization, 8..--....----------
L" = ~~ (2r p ~: I I) (7-122) 6
..<
where superscripts 8 and i stand for .......... 4 t - - - T - - - t - - - - t - - - + + - - - - - t
H scattered" and H incident," respec-
~
tively. Going through a develop- 2t---+----+---bl~__+-----4
ment similar to that used for Eq.
(7-115), except that a line source is
used, we obtain a 0.2 0.4 0.6 0.8 1.0
a/A
r (¢J.eikzdZY 2 FIG. 7-16. Echo width L, of a conduct-
i: = 2X 7] ~ (7-123) ing ribbon of width a.
'f E z4J zOdl
if the incident field is z-polarized and -x traveling. Similarly,

(7-124)

if the incident field is y-polarized and -x traveling. From symmetry,

JO in Eq. (7-124) should have no z component. In both Eqs. (7-123)
and (7-124), it is assumed that the scatterers are cylinders generated by
elements parallel to the z axis and the line integrals are in a transverse
(z = constant) plane.
For an example of a two-dimensional problem, consider a z-polarized
plane wave normally incident on a conducting ribbon of width a. This is
illustrated by the insert of Fig. 7-16. Assume that the current induced
on the ribbon is uniform, that is,
(7-125)

Because the current is real, the integral in the denominator of Eq. (7-123)
is

Ja/2
-a/2
E ZoJ Z4 dy = J a/2
-a/2
E ZoJ z0* dy = - p

where P is the complex power per unit length supplied by J.a. But we
have already analyzed the ribbon of uniform current in Sec. 4-12, the
result being
.,,2
P = II2\Z = a 2 "2 Yapert

where Yapert is plotted in Fig. 4-22. The echo width, according to Eq.
360 TIME-HARMONIC ELECTROMAGNETIC FIELDS

Receiver
/

FIG. 7-17. Differential

scattering.
Transmitter

(7-123), is then

£6 ~ ;X Ia2(712i:)2Yapert r = 2; [71Y~pert r (7-126)

A plot of this is shown in Fig. 7-16. For large a we can use Eq. (4-107)
and obtain

(7-127)

which is also the physical optics approximation (see Fig. 3-21).

The more general case of differential scattering, or transmission,' is
represented by Fig. 7-17. The problem consists of a transmitter, which
illuminates the obstacle, and a receiver at which we wish to evaluate the
scattered signal. For simplicity, let us consider both the source and
receiver to be unit electric currents. Then, according to Eq. (3-39), the
voltage across the receiving current due to the transmitting current is
Vr = - (t,r)obstacle present (7-128)

where t and r refer to the source or field of the transmitter and receiver,
respectively. The total signal received is the superposition of the inci-
dent field, due to the transmitter alone, plus the scattered field, due to
the currents c on the obstacle. Hence,
(7-129)
where (t,r) is calculated with the obstacle absent and (c,r) involves the
free-space field of the currents on the obstacle. The transmitter and
receiver currents are assumed to be known (they are current elements in
our simplified case); so V r ' can, in principle, be calculated exactly. Our
problem is to obtain the variational formula for V r "' .
We shall here consider only the simple case of a perfectly conducting
obstacle, the general case being considered in Sec. 7-11. Applying reci-
1 A transmission problem involves the evaluation of the total field at the receiver,
while a scattering problem involves the evaluation of only the scattered field.
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 361
procity, we have, for the scattered voltage at the receiver,
- V,.· = (c,r) = (r,c) = 1f (Ei),. • (J.e)t dB (7-130)

where (J.c)' is the surface current induced on the obstacle by the trans-
mitter and (Ei),. is the field of the receiver current calculated with the
obstacle absent (the incident field). Our boundary conditions on the
various true fields are n X E = 0 at the obstacle boundary; hence
n X (E')" = -n X (Ea),.
(7-131)
n X (Ei)' = -n X (Ea)'
where superscripts i and 8 refer to incident and scattered components,
and t and r refer to transmitter and receiver sources. Hence, by Eqs.
(7-130) and (7-131), we have

V,.· = 1f (E·)" • (J.e)t ds = (C,.,Ct) (7-132)

where (C,.,Ct) stands for the reaction between the field of the "correct"
currents induced on the obstacle by the receiver and the" correct" cur-
rents induced by the transmitter. For our stationary formula, we
approximate (cr,c,) by (a,.,at), where the a's denote assumed currents on
the obstacle, and constrain the latter according to Eq. (7-65), which is
(a,.,a,) = (c,.,a,) = (a,.,ct) (7-133)
In the language of the reaction concept, Eq. (7-133) says that the assumed
currents look the same to each other as to their respective true currents.
By Eqs. (7-131) and reciprocity, Eqs. (7-133) become
(a,.,a,) = (c,.,a,) = -(r,a,)
(7-134)
(a,.,a,) = (ar,Ct) = (ct,ar) = -(t,a,.)
Substituting from Eqs. (7-134) into Eq. (7-132), we have for our vari-
ational formula

[1f (Ei).. • (J.a)l dB ] [1f (Ei)t • (J.a),. dB ]

1f (Ea),. • (J.a)t de
(7-135)

where (Eo)" is the field due to the assumed currents (Jaa)", which approxi-
mate the currents induced by the receiver. Note that Eq. (7-135)
involves the assumption of currents on the obstacle due to sources at
both the transmitter and receiver. Note also that Eq. (7-135) reduces
to the formula for back-scattering [Eq. (7-113)] when the transmitter
and receiver coincide.
362 TIME-HARMONIC ELECTROMAGNETIC FIELDS

7-11. Scattering by Dielectric Obstacles.' The problem of differential

scattering by a dielectric obstacle is represented by Fig. 7-17 if the
obstacle is now considered as a dielectric body. We shall assume it to
be nonmagnetic (~ = ~O), but it may be lossy if E is complex. The
extension to magnetic obstacles is given in Probe 7-42.
When the obstacle is excited by a source, there will be induced in it
polarization currents given by

(7-136)

Superscripts t or r will be added to the various quantities to indicate that

the exciting source is at the transmitter or receiver, respectively. The
treatment of differential scattering of the preceding section made no
assumptions about the nature of the obstacle in the derivation of Eq.
(7-130); hence for unit currents at t and r

(7-137)

where the notation is the same as in the preceding section. Using the
relationship Ei = E - Ea and Eq. (7-136), we can rewrite Eq. (7-137) as

- Vro = JJJ cIUe)r. o-» dr - JJJ (Eo)r. Ue)l dT

= F(cr,ct) - (cr,c,) (7-138)

which defines the functional F. Note that F is symmetrical in c; and

c, and is actually the reaction between Er and (Je), with the obstacle
present.
To obtain a stationary formula for the scattered voltage at the receiver,
we approximate the true currents C by trial currents a and set
(7-139)

subject to constraints of the form of Eq. (7-65) applied to G. Such con-

straints are
G(ar,a,) = G(cr,at) = G(a"ct) (7-140)
and we find
G(cr,a,) = (r,a,) = JJJ (Ei)r. o-» dr
(7-141)
G(ar,c,) = (t,a r) = JJJ (Ei)'. o-» dr

Combining the preceding equations to render V,., insensitive to the ampli-

1 M. H. Cohen, Application of the Reaction Concept to Scattering Problems, IRE
Trans., vol. AP-3, no. 4, pp. 193-199, October, 1955.
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 363
tudes of the trial functions, we have the variational formula
_ Vra = (r,at)(t,a r)
F(a"a,) - (ar,at)
_[III (Ei)r. (JG)I dr] [III (Ei)'. (JG)r dr ]

- IJJ l(-l(Jo)r. (JG)I dr - III (Eo)r • (JO)I dr

(7-142)

For the lossy case, K = jWE + 0'. For a perfectly conducting obstacle,
0'~ 00; hence K- 1 ~ 0 and Eq. (7-142) reduces to Eq. (7-135).
When the transmitter and receiver are represented by the same source,
we have the back-scattering problem. Using the definition of Eq. (7-109)
for echo, when the source is a unit current, we have

(7-143)

The echo area, defined by Eq. (7-114), can be obtained from Eq. (7-143)
by letting the source recede to infinity. The steps parallel those used to
obtain Eq. (7-115). For a z-polarized, -x traveling incident wave, we
obtain
2

(7-144)

In two-dimensional problems, the echo width, defined by Eq. (7-122), is

found to be

L" = ;>. (7-145)

if the incident wave is -x traveling and z-polarized, and

2

(7-146)

if the incident wave is -x traveling and y-polarized. The surface inte-

grals in Eqs. (7-145) and (7-146) are over the cross section of the obstacle
in a z = constant plane.
364 TIME-HARMONIC ELECTROMAGNETIC FIELDS

To illustrate the accuracy that we might expect from the variational

formulas, let us consider a problem for which the exact solution is avail-
able, the circular dielectric cylinder. The incident wave is z-polarized,
and the cylinder is defined by p = a = "X o/ 2, as shown in the insert of
Fig. 7-18. For our first approximation, let us take

(7-147)

where k = w ~ is the wave number of the dielectric. This very crude

assumption yields curve (b) of Fig. 7-18. For a better approximation,
which yields curve (c) of Fig. 7-18, take
Ja = uz(eikz + Ae-ikz ) (7-148)

where A is a variational parameter to be determined either by the Ritz

procedure or by the reaction concept. While Eq. (7-148) is a better
approximation than Eq. (7-147), it is still crude. The integrations
occurring in the various reactions were accomplished by expressing the
exponentials and Hankel functions as Bessel function series, according to
Sec. 5-8. The resulting series converged fairly rapidly.
An alternative procedure for treating dielectric obstacles can be given

0.00012

~ 0.00008
<,
~

o. 00004 r-----t------::IJ"--t---+-----t~~_+_~-..__~_t

o
1.00 1.04 1.08 1.12 1.16 1.20 1.24 1.28
EIEo
FIG. 7-18. Scattering by a dielectric cylinder (a) exact solution, (b) first-order varia-
tional solution, and (c) second-order variational solution. (After Cohen.)
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 365
in terms of equivalent currents over the surface of the obstacle." This
method leads to more than one formula for the desired parameter, and
Rumsey discusses how to choose the best approximation according to
the reaction concept.
7-12. Transmission through Apertures. The problem of transmission
through apertures in an infinitely thin, perfectly conducting plane is
closely related to the problem of scattering by plane obstacles. The pre-
cise interrelationship is shown by the following extension of Babinet's
principle for optics.
Consider the three cases of a given source (a) radiating in free space,
(b) radiating in the presence of an electrically conducting screen, and (c)
radiating in the presence of a magnetically conducting screen, as shown
in Fig. 7-19. The electric and magnetic screens are said to be comple-
mentary if the two screens superimposed cover the entire y = 0 plane
with no overlapping. (The aperture of one is identical to the obstacle
of the other.) Let the fields y > 0 be designated (Ei,Hi) , (Ee,He) , and
(Em,Hm) for the cases (a}, (b), and (c), respectively. Then Babinet's
principle for complementary screens states that
(7-149)
proved aa follows. Let S_ be the screen surface of Fig. 7-19b, and Sa be
the aperture surface of Fig. 7-19b. The total field in each case is the
incident field E' plus the scattered field E- produced by the currents on
the screen. An element of electric current produces no components of
H tangential to any plane containing the element (see Sec. 2-9). The
currents induced on the screen thus produce no tangential H over the
y = 0 plane; hence
n X He = n X Hi over Sa
On the screen itself we have the boundary condition
n X Ee = 0 over S.
For the complementary magnetic screen, following similar reasoning,
we find
n X Em = n X Ei over S.
n X a- = 0 over So.
By the above four equations, the sum Ee + Em, He + Hm satisfies
n X (Ee + Em) = n X E' over S.
n X (He + Hm) = n X Hi over Sa
1 V. H. Rumsey, The Reaction Concept in Electromagnetic Theory, Phys. Reo.,
2 ser., vel. 94, no. 6, pp. 1485-1491, June 15, 1954,
366 TIME-HARMONIC ELECTROMAGNETIC FIELDS

Hence, the e + m field has the same n X E as the incident field over part
of the y = 0 plane and the same n X H over the rest of the y = 0 plane.
These conditions are sufficient to determine E, H in the region y > 0
according to the uniqueness theorem (Sec. 3-3); so Babinet's principle
[Eq. (7-149)] follows.
An alternative statement of Babinet's principle can be given in terms
of the dual problem to Fig. 7-19c, shown in Fig. 7-19d. If the original
source is replaced by its dual (J replaced by K), the magnetic screen
replaced by an electric screen, and the medium replaced by its "recipro-
cal" (17 by 1/11), then E will be numerically equal to -Hm and H numeri-
cally equal to Em (see Table 3-2). If the field of this dual problem is

I
I Electric conductor S.
I
I Ei~ Hi "10
I
I
t Source I
I t Source
I
I s.
I
~n n
I
y=O y;e:O
(a) (b)

Is. Is.
I I
I I
I Em~ Hm~ "10
I
Sa
t Source Magnetic conductor * Dual source
Sa
Electric conductor

I I
IS. Is.
~n ~n
I I
,)'=0 y =0
(c) (d)

Flo. 7-19. Illustration of Babinet's principle.

 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 367

Electric conductor

Magnetic conductor I
\L- 1f 1,. ~I, IT/f
Transmitter Receiver Transmitter Receiver

(a) (b)

FIG. 7-20. The transmitted field Et of (a) is equal to the scattered field E- of (b).

denoted by Ed, Hd, Babinet's principle [Eq. (7-149)] becomes

(7-150)
The problem of Fig. 7-19d is more easily approximated physically than is
the problem of Fig. 7-19c.
The direct application of Babinet's principle to the problems of Fig.
7-20a and b shows that the field transmitted by an aperture in a plane-
conducting screen is equal to the negative of the field scattered by the
complementary obstacle. Hence, stationary formulas forthe signal at
a receiver on the shadow side of a screen are of the same form as the
stationary formulas for the scattered signal at a receiver in the comple-
mentary problem. In Fig. 7-20b, let the sources at the transmitter and
receiver be magnetic currents across the "terminals" I, and L, Then,
dual to Eq. (7-135), we have at the receiver

H- -lr = -
[I1(Hi)r. (M.G)' dS] [I1(Hi)'. (M.G)' ds ] (7-151)
11(HG)r. (M.G)' ds
where M.'J denotes the assumed magnetic current on the obstacle. It
approximates the true magnetic current
M_ = (E+ - E-) X n = 2E- X n (7-152)

where E+ and E- denote E in the regions y > 0 and y < 0, respectively,

and n = Uu. The interrelationships between Fig. 7-20a and b can be
expressed as
H' = -H- E' == -E-
368 TIME-HARMONIC ELECTROMAGNETIC FIELDS

Hence, from Eq. (7-151), we obtain for the aperture problem

where Ea is an assumed field in the aperture and Ha is the magnetic field

calculated from the Ea. The sources of Hi are magnetic current elements
across It and L, and, to apply Eq. (7-153), we must assume an n X E in
the aperture due to (Hi)' alone and due to (Hi)r alone. If It and Ir are
images of each other, as they appear in Fig. 7-20, then the aperture
problem becomes the same as an echo problem, because of the symmetry
of the plane screens about y = o.
Sometimes it is the total power transmitted through the aperture that
is of interest. We define the transmission coefficient T of an aperture
as the ratio of the power transmitted through an aperture to the power
incident on the aperture, that is,

Re JJ E' X H'* • ds
T = apert <Pc (7-154)
Re JJ s- X Hi* • ds = iJ>i
apert

Note that T depends on both the nature of the source and the geometry
of the aperture. Another quantity sometimes defined is the transmission
area, which is the transmission coefficient times the area of the aperture.
We shall explicitly consider uniform plane waves normally incident on
an aperture in a plane screen, as shown in Fig. 7-21a. Let the incident

Electric conductor Complete electric

. conductor

Incident
plane wave
...

n n

y=o y=o
(a) (b)

FIG. 7-21. {a) Transmission through an aperture, and (b) equivalent problem for the
region y > o.
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 369
wave be specified by
Hi = ue-i ku Ei = 71Hi X u" (7-155)
where u is any unit vector orthogonal to UII. In the proof of Babinet's
principle, we noted that in the aperture
n X H' = n X Hi (7-156)
because the currents on the conducting screen produce no tangential
components of H in the y = 0 plane. Equation (7-155) chooses Hi to be
real in the y = 0 plane; so by Eq. (7-156) n X Hi is real in the aperture.
Hence,
CPt = Re JJEI
apert
X HI* ds
0 = Re JJEI
apert
X HI ds 0
(7-157)

Now consider the problem of Fig. 7-21b, which for

Ms = Et X n (7-158)
is equivalent to Fig. 7-21a in the region y > O. Hence, in the equivalent
problem,
<PI = - Re JJ M. 0 H' ds = Re (c,c)
0 (7-159)

where (c,c) is the self-reaction of the correct magnetic currents radiating

in the presence of an electric conductor covering the entire y = 0 plane.
For a variational formulation, we approximate (c,c) by (a,a) and con-
strain (a,a) according to Eqs. (7-65), that is,
(c,c) ~ (a,a) = (c,a) = (a,c)
where all sources radiate in the presence of the conducting plane. We
have (a,c) = (c,a) by reciprocity, and (c,a) can be calculated because we
know n X He = n X Hi. Hence, our stationary formula for (c,c) is

~
(c,a)2
-- = -
(JJHi 0 M." ds Y
(c c)
'(a,a) JJa-
---=-=--=------"-
0 M." ds
(7-160)

where Ha is the field of the assumed current M,a. For the incident field
of Eq. (7-155), we have the power incident on the aperture given by
0\ = 71A (7-161)
where A is the area of the aperture. Hence, combining Eqs. (7-154) to
(7-161), we have

T = .L Re [(JJ non X Ea dSY] (7-162)

'lA JJa- · n X E" ds
370 TIME-HARMONIC ELECTROMAGNETIC FIELDS

2.0 .....--oy---~-..,......~~-..., where Ea is the assumed tangential

electric field in the aperture and Ha is
the magnetic field calculated from Ea
by the methods of Sec. 3-6.
As an example, let us consider the
Variational two-dimensional problem of trans-
0.5 1-----+---+---+---+-----1 mission through a slot, as shown in
the insert of Fig. 7-22. If we assume
Ea in the slot to be real, then
o 0.2 0.4 0.6 0.8 1.0
a/A Ea X Ha* = (Ea X Ha)*
FIG. 7-22. Transmission coefficient for and the denominator of Eq. (7-162)
a slotted conductor, incident wave is
polarized transverse to slot axis. (After
:fltfiles.) / / a- · n X Eo ds =

(// Eo X Ho* • ds) *

In Sec. 4-11 we defined the admittance of an aperture as

Yapert = 1;1 2 II E X H*· ds

and calculated it for a slot for particular assumed E's. Hence, applying
Eq. (7-162) to a unit length of our two-dimensional slot, we have

(7-163)

where a is the width of the slot. When the incident wave is polarized
transverse to the slot, we have the case of Fig. 4-22; hence we take
Ea = 1 (7-164)
in the slot. Now Eq. (7-163) reduces to

T = l-
1Ja
Re (_1_) Y:pe r t
(7-165)

where Yapert = Ga + [B; is shown in Fig. 4-22. From Eqs. (4-106) we

have for small a
1r 2
T ka=:t ka log ka (7-166)

and from Eqs. (4-107) we have for large a

T--+ 1 (7-167)
ka--"ClO
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 371
This last result is the geometrical optics approximation. The variational
solution is compared to the exact solution, which can be obtained by
solving the wave equation in elliptic coordinates 1 (Fig. 7-22). The case
of a plane wave at an arbitrary angle of incidence is considered by Miles."
If the incident wave is polarized parallel to the axis of the slot, we have
the case of Fig. 4-23; so to make use of the analysis of Sec. 4-11 we would
assume
Eo = cos 1rX (7-168)
a

in the slot. Equation (7-163) then reduces to

T = 4a Re (_1_) (7-169)
1r 211 Y:-pert
where Yapert = Go + jBo is shown in Fig. 4-23. From Eqs. (4-115), we
have for small a
T ~
ka-+O
6.85 (~)81\
(7-170)

For large a we should expect the field in the aperture to be uniform.

Hence, we should not expect the trial field of Eq. (7-168) to give good
results for large a, say a > X. Equation (7-169) actually approaches
0.81 for large a, instead of the expected value 1.

PROBLEMS

7-1. Suppose the cavities of Fig. 7-1 contain lossy material characterized by (1', E,
and 11-. Show that the perturbational formula corresponding to Eq. (7-3) is

j effi H X Eo • ds
AS

'" - "'0 = JJJ fEE· Eo - ,.H • HoJ d'T

Note that both wand Wo must be complex. A complex resonance in the low-loss case
can be interpreted according to

where w, is the real resonant frequency and Q is the quality factor (see Sec. 8-14).
7-2. Consider the perturbation of a cavity (say Fig. 7-1a) from one having per-
fectly conducting walls to one having a wall impedance z, defined by
nXE=ZH t
1 Morse and Rubenstein, The Diffraction of Waves by Ribbons and Slits, Phys. Reu.,

vol. 54, no. 11, pp. 895-898, December, 1938.

2 J. W. Miles, On the Diffraction of an Electromagnetic Wave through a Plane

Screen, J. Appl, Phy., vol. 20, no. 8, pp. 760-770, August, 1949.
372 TIME-HARMONIC ELECTROMAGNETIC FIELDS

at the walls. Show that the exact perturbational formula is

-j iff zn- H o d8

'" - "'0 = III (at· Eo - pH • H o) dT

where the subscript 0 denotes unperturbed quantities. Note that Wo is real but w is
complex if Z has a real part.
7-3. Use the results of Probe 7-2 and the approximations
E ~ Eo = IEol H ~ H o = jlHol
to show that

Use the relationships

Z = CR + JOC
and show that the perturbational formula gives

iff XIH oI2d8

ClJr - "'0 ~

2 III plH ol2dT

Note that the formula for Q is identical to the one that we have been using if
CR = Re (-11), where
TJ is the intrinsic impedance of the conducting walls.
7-4. Use the results of Probe 7-3, and show that the fractional change in resonance
due to metal walls is
"'r -
---~
"'0 1
- 2Q

where Wo is the resonant frequency of the cavity with perfectly conducting walls.
7-6. Suppose the cavities of Fig. 7-2 are characterized by a and a + diT in addition
to E, p. and E + ~E, P. + dp.. Show that the perturbational formula corresponding to
Eq. (7-10) is then

III [(~E - j~<T/",)E • Eo - ~pH • Hol dT

III [(E - j<T/w)E· Eo - pH· Hol dr

Both "'0 and '" are complex if a and (T + ilu are not identically zero.
7-6. Use the result of Probe 7-5 for the case (T = 0, and let E ~ Eo = IEol,
H ~ H o = jlHol, w ~ Wr + jwo/2Q, to show that

and that Eq. (7-11) still applies with", changed to "'r.

 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 373
7-7. Suppose that a small sample of lossy dielectric is introduced into a cavity
whose unperturbed resonant frequency is Wo. Show that

where ~ = E' - jE" is the complex permittivity of the sample and Wr is the perturbed
resonant frequency. If the losses of the unperturbed cavity are significant, then
111
Q ~ Q. - Qo
where Q. and Qo are the Q's of the cavity with and without the sample, respectively.
7-8. Consider a rectangular cavity with a small centered dielectric cylinder, as
shown in Fig. 7-23a. Show that the change in the resonant frequency of the domi-
nant mode due to the introduction of the dielectric is

~ ~ 2A (1 - E )
r
Wo be

where A is the cross-sectional area of the cylinder. Use a quasi-static approximation.

Area A

f
1 -~~
1-· ---.V
~---b'---~

(a)
FIG. 7-23. Rectangular cavity with (a) dielectric cylinder and (b) dielectric sphere.

7-9. Consider the rectangular cavity with a small centered dielectric sphere, as
shown in Fig. 7-23b. Show that the change in the resonant frequency of the domi-
nant mode due to the introduction of the dielectric is
'" - "'0 rd 3 Er - 1
--w- ~ - abCEr +2
where d is the diameter of the sphere. Use a quasi-static approximation.
7-10. Consider the circular waveguide of Fig. 5-2. Suppose the wall is slightly
flattened at the point cP = 90°. Show that the change in cutoff frequency for the
z-polarized (E in the center points in the x direction) dominant mode is

AWe ~ -0.418 ~
We rat

where A is the cross-sectional area of the deformation and We = 1.841/a v;;, is the
unperturbed cutoff frequency. For the y-polarized dominant mode,

AWe ~ 1.42~
"'e: raj

Hence, the mode degeneracy has been removed.

374 TIME-HARMONIC ELECTROMAGNETIC FIELDS

7-11. Figure 7-24a shows a small centered dielectric cylinder in a rectangular

waveguide. Show that the change in cutoff frequency of the dominant mode from
that for the empty guide is

where We = rib v;p,. Use a perturbational method and a quasi-static approximation.

I- b
-I I-..-----b---~~
T a 0
Ta

1 -+idf+-
1 -.f d f+-
(a) (b)

FIG. 7-24. Rectangular waveguide with (a) dielectric cylinder and (b) conducting
ridges.

7-12. Consider the rectangular waveguide with small semicircular ridges, as shown
in Fig. 7-24b. Use a perturbational method and a quasi-static approximation to show
that the dominant-mode cutoff frequency differs from the TE o l rectangular guide
cutoff, according to

where We = rib v;p,. Show that the next higher mode (b ~ 2a) cutoff frequency
differs from the TE o2 rectangular guide cutoff, according to

where We = 2r /b v;p,. Hence, the mode separation is increased.

7-13. Consider the rectangular waveguide with the bottom covered by a thin
dielectric slab (Fig. 4-6 with d ~ a). Use a perturbational method and quasi-static
approximation to show that the phase constant is

where fJo = k 2 viI - (fc/f)2 is the empty-guide phase constant. Note that this is
the same as the first term of an expansion of the exact characteristic equation, as
given in Probe 4-14.
7-14. Consider the rectangular waveguide with a centered dielectric cylinder, as
shown in Fig. 7-24a. Use a perturbational method and quasi-static approximation to
show that

where We can be taken as the cutoff frequency of the perturbed guide, given in Prob
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 375
7-11, if '-' is close to '-'c. Show that at the unperturbed TE ol cutoff frequency

fJ = k o ... rd
J
2
Er +1
\J4ab Er - 1

7 -16. Suppose that a waveguide is filled with lossy material, and consider a pertur-
bation of its perfectly conducting walls. Represent the unperturbed fields (subscript
0) and the perturbed fields (no subscript) by

Eo = ~oe-'Yo.
H o == iioe-'Yo.
Note the opposite directions of propagation. Show that the formula corresponding
to Eq. (7-29) is

Show that this reduces to Eq. (7-29) in the loss-free case.

7-16. Consider the perturbation of material in a lossy waveguide from E, P, to (T

E + ~E, 1.& + ~p" + ~(T. Represent the fields as in Probe 7-15, and show that the
(T

formula corresponding to Eq. (7-30) is

=
.II [('"dE - j dO}~ • to - '" dp.H • lI oJde
"Y - "Yo
II
-3 - - - - : : - - - - - - - - - - - - - -
(to X H - t X 110) • U. d«
Show that this reduces to Eq. (7-30) in the loss-free case.
7-17. Use the results of Probe 7-16, and let the unperturbed guide be loss-free.
Denote the propagation constant of the perturbed guide by l' = a + ifJ, and let
E = E; and H = -U:.
Show that the resultant approximation for fJ is Eq. (7-33)
and

II 2 M 1 01 ds
2

a=--":'-';:~------

2 Re II to H: · X u, cis

Note that this is an approximate form of Eq. (2-76).

7 -18. Consider the perturbation of the walls of a waveguide from a perfect con-
ductor to an impedance sheet Z such that
n XE = zU
Represent the unperturbed and perturbed fields as in Probe 7-15, and show that

'Y - 'Yo = II (~O X H: - ~ X Ho) • u, de

7-19. Use the results of Probe 7-18 and let the unperturbed guide be loss-free,
so that "Yo = ;(30. In the perturbed guide, let Z = CR + jX, l' = a + jfJ, E == E:,
376 TIME-HARMONIC ELECTROMAGNETIC FIELDS

H == - 0::, and show that

fJo ~
f
oclbol2 til
----=--------
II ii:·
fJ -
2 Re Eo X u, d8

f CRlb lo2 til

If z = 11, the intrinsic impedance of metal walls, the above formula for a is the
approximation that we have been using to calculate attenuation in metal waveguides.
7-20. Show that

Cl'r 2
JJJ l'-l\V X EI2 dT
= --------
IIJEIEI 2
dT

is a stationary formula for the resonant frequency of a loss-free cavity, provided

n X E = 0 on S, but is not stationary if losses are present.
7-21. Show that Eq. (7-46) is a stationary formula for Wr 2, with no boundary con-
ditions required on H.
7-22. Consider the rectangular cavity (Fig. 2-19) and the stationary formula [Eq.
(7-44)]. Use a trial field
E = uzyz(y - b)(z - e)
and show that Eq. (7-44) gives
VlO ~b2 +c 2
fAr "'"
be Ep,

In the exact solution [Eq. (2-95)], the numerical factor is 1(' instead of Yl0.
7-23. Consider a small deformation of the walls of a cavity, such as represented by
Fig. 7-1. Take the variational formula [Eq. (7-45)], which requires no boundary con-
ditions on E, and take the unperturbed cavity field Eo as a trial field. Show that
Eq. (7-45) reduces to

Show that this formula is essentially the same as Eq. (7-4).

7-24. Figure 5-31b shows a partially filled circular cavity. Use Eq, (7-46) and a
trial field

to show that the dominant mode resonance is

Compare with the results of Probe 5-24.

 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 377
7-25. Consider a waveguide whose cross section is an equilateral triangle of side
length a. Use variational formulas to approximate the lowest cutoff frequency. The
exact solution is
4r
c"c=---
3aw
7-26. Consider the rectangular cavity (Fig. 2-19) and the mixed-field variational
formula [Eq. (7-72)]. Choose a trial field
. 1fy . 1rZ
E =uzslnbsln
c
H = UJiA I
• 1rY
SIn b COS c + u~
1rZ
2
1fy •
cos b SIn C
1rZ

where Al and A 2 are variational parameters. Determine Al and A 2 by the Ritz

method, and show that the resultant formula for c"r is the exact formula [Eq. (2-95)].
Why do we get an exact solution in this case?
7-27. In Fig. 7-25, the surface S represents a perfect electric conductor enclosing a
cavity. A variational solution is desired in terms of a trial field satisfying n X E = 0

FIG. 7-25. Trial fields are discontinuous over 8.

on Sand n X (~-IV X E) continuous at 8, but with n X E discontinuous at 8. Show

that the stationary E-field formula is

",I =
III ; (V X E)t dT +2 II 8
(E 1 - Et) X ; V X E· ds
_
III dJtdT

where subscripts 1 and 2 refer to regions 1 and 2 (Fig. 7-25). Show also that a varia..
tional solution in terms of trial fields satisfying n X E = 0 on Sand n X E continuous
at 8, but with n X (~-lV X E) discontinuous at 8, is given by Eq. (7-44).
7-28. Show that the variational H-field formula for Probe 7-27 is of the same form
as the above E-field formula, given by replacing E by H, E by ~, and ~ by E. Show
that no boundary conditions at S are required in the H-field formula.
7-29. Consider a perturbation of material in a cavity, such as represented by Fig.
7-2. Take the mixed-field variational formula [Eq. (7-72)], and take the unperturbed
cavity field Eo, Ho as a trial field. Show that Eq. (7-72) then reduces to Eq. (7-11).
7-30. Repeat Probe 7-26, using the reaction concept of Sec. 7-7.
7-31. Consider the partially filled rectangular waveguide of Fig. 4-8a. Use the
E-field variational formula [Eq. (7-8)], and the trial field

E = U II sin rZ
a
378 TIME-HARMONIC ELECTROMAGNETIC FIELDS

and show that

E2 (d
"'c ~ -a [1 + ---
ft" El -
E2
-1.SIn 21rd)J-l-i
-a - 21r -
a
Compare some calculated points with the exact solution (Fig. 4-9).
7-32. Use the reaction concept to derive the mixed-field variational formula for
waveguide phase constants

{J=
wJJ (p.f1I+~)d8-j¢~XH.ndl -
JJ~ H· 2 X u.d8

which corresponds to Eq. (7-85) if n X E = 0 on C. No boundary conditions are

required in the above formula.
7-33. Consider the variational formula of Probe 7-32 and a perturbation of wave-
guide walls, as illustrated by Fig. 7-5a and b. Use the unperturbed field Eo, H o as a
trial field, and show that the formula of Probe 7-32 reduces to Eq. (7-32).
7-34. Consider the variational formula of Eq. (7-85) and a perturbation of matter
in a waveguide, represented by Fig. 7-5a and c. Use the unperturbed field Eo, H o as
a trial field, and show that Eq. (7-85) reduces to Eq. (7-33).
7-36. Figure 7-26 shows a coaxial stub to parallel-plate waveguide feed system.
Assume a « X so that a reasonable trial current is a uniform current. Show by the
variational method that the impedance seen by the coax is

Z ~ -~ ka ( 1 - 3.2
- log -~kd)
4 .... 4
where ~ = 1.781.

matched a matched
load load
t

FIG. 7-26. Coax to parallel-plate feed.

7-36. In Probe 7-35, remove the restriction on a and assume a trial current on the
stub
[ :c: cos k(a - a)

Obtain the input impedance seen by the coax by the variational method.
7-37. Repeat Probe 7-36 for the second-order variational solution, assuming trial
currents
[11, = cos k(a - s) [" == 1
Note that only one new reaction is needed in addition to those obtained in Probs.
7-35 and 7-36. Specialize the result to a = X/4.
7-38. Consider the two-dimensional problem of plane-wave scattering by a con-
ducting ribbon, shown in the insert of Fig. 7-16, but with the opposite polarization.
 PERTURBATIONAL AND VARIATIONAL TECHNIQUES 379
"In other words, H' is parallel to the axis of the ribbon. Use the trial current

]. = UI/ cos 1rY

a

and show that the variational solution is

L
•
32a41 1
~---
3X",
1r Yapert
II
where nY a pe r t is given in Fig. 4-23. Show that as ka --+ 00 this answer reduces to
0.66 times the physical optics solution. Why should we expect the above formula
to be inaccurate for large kat
7-39. Consider plane-wave scattering by a wire, represented by Fig. 7-15. At the
first resonance (L ~ X/2), the current is
[G ~ cos kz
and we know that (see Fig. 2-24)
(a,a) ~ 73

The imaginary part of (a,a) is zero because the length is adjusted for resonance.
Using Eq. (7-115), show that at resonance the echo area is
A, ~ 0.86XI
This is relatively insensitive to the diameter of the wire.
7-40. Figure 7-27 represents a resonant length of wire illuminated by a uniform
plane wave at the angle 8, polarized in the r-z plane. Using the approximations of
Probe 7-39, show that the back-scattering area is

Again this is relatively insensitive to the diameter of the wire.

zl r' (to receiver)

1-0 1

ktJ r'(to transmitter)

I
TL
1
FIG. 7-27. Scattering by a resonant wire (L ~ X/2).

7-41. Repeat Probe 7-40 for the case of differential scattering, showing that the
differential echo area is

1 [cos (~cos (J) cos (~cos (J,)]I

A. ~ 0.86X sin (J sin 6'

where A. is defined by Eq. (7-114) with E· evaluated in the 8' direction.

380 TIME-HARMONIC ELECTROMAGNETIC FIELDS

7-42. Consider differential scattering by a magnetic obstacle (Fig. 7-17) and define
"m = jw (p, - p,o)
Show that, instead of Eq. (7-143), we have
2
Echo = «i,a)/l)
F(a,a) - (a,a)

where (i,a) = III (E'· J" - H'· M") dT

F(a,a) = III (K.- 1(J " )2 - Km -1(M")2J dT

(a,a) = III (E"· J" - H"· M") dT

In the above formulas, Ei, Hi is the incident field, JO and Mo are the assumed electric
and magnetic polarization currents on the obstacle, and Eo, Ho is the field from Jo, Ma.
7-43. Figure 7-28a represents a metal antenna cut from a plane conductor and fed
across the slot abo Figure 7-28b represents the aperture formed by the remainder of
the metal plane left after the metal antenna was cut. The aperture antenna, fed

,
,
(a) (b)

FIG. 7-28. (a) A sheet-metal antenna and (b) its complementary aperture antenna.

across cd, is said to be complementary to the metal antenna. Let Zm be the input
impedance of the metal antenna and Y. be the input admittance to the slot antenna,
and show that

Hint: Consider line integrals of E and H from a to band c to d, and use duality.
7-44. Consider a narrow resonant slot of approximate length X/2 in a conducting
screen. Show that the transmission coefficient is

T ~ O.52~
to

where to is the width of the slot. Hint: Use the result of Probe 7-43 and assumptions
similar to those of Prob. 7-39.