129

CENTRO INTERNAZIONALE MATWTICO ESTIVO

( O.Ift.E. )
L.B. FELSEN
1)
Department of Electrophysics. Polytechnic Institute of Brooklyn
1 )
2)
OF INTEGRALS
These lectures were presented while the author was on a years'
leave of absence with the Office of Naval Research, London Branch.
2)
This chapter is based on Chapter IV, "KodsI Analysis and Syn-
thesis of Electromagnetic Fields", by L.B.Felsen and N.Marcuvitz,
Research Report R-776-59. 'PIB-705, dcrowave Research Institute,
Polytechnic Institute of Brooklyn, Oct.1959.
131
EVALUATION OF INTEGRALS
by L.B.FELSEN
Department of Electrophysics,
Polytechnic Institute of Brooklyn
1. INTRODUCTORY REMARKS
Radiation and diffraction problems in open regions
(with infinite cross section) are usually solved in of
integral representations for the fields cannot be eva-.
luated in closed form. In many applications, however, the in-
tegrands contain a large parameter, to be called 0- , in t.3I'I:.1S
of which one may obtain an approximation to the integrals.
\Vhile such an evaluation can be treated for rather general
functional dependences of the integrand on Q ,it will suf-
fice wi thin the pr';sent con.text to consider integrals of the
follOwing type:
l
,2
1(0)
= j f(z)
Qo (z) d
e - z , (1)
p
z
where f and q are analytic functions of the complex variable
z along the path of integration P , whose endpOints lie at in-
z
finity, and where the large parameter Q is assumed to be
if.
positive.
If If 0 =\ Q I exp(i arg . Q ) is complex, the phase term is in-
cluded in tho definition of q(z). Alternatively, it :nay h'3
more convenient to obtain an asymptotic evaluation if 1(.0 )
for real values of Q and then oontinue ._ Q analytically into
a range of permitted complex values.
132
- 2 -
L.J3.Felsen
Suppose that Re q(z) has at the p,oint Z Qn P
. s z
a maximum value so that Be q(z) < Be q(z ) on the remainder.
s
of the path. Since Q is very large, it follows that
A = I exp [ Q q ( z I
likewise has a maximum at z and decrea-
. s
ses very rapidly away from zs' It is then suggestive to ap-
proximate I( Q ) by its contribution from the vicinity of z
s
only since the contribution from the remainder of the path
will be exponentially small in comparison. If f(z) is regular
and slowly varying in the vicinity of z , this function may
s
be approximated there by f(z) and taken outside the integrands
s
in (1), thereby] eaving in the integrand only the exponential.
Integration of the latter can be effected approximately by ex-
panding q(z) in a power series about z and retaining only
s
the first few terms; by a basia procedure (see Sec.II),
the integral is compared with a U aanonical " one having similar
properties. This, in rough outline, basis for
an asymptotic approximation of I (Q , fol' large values of Q
details of the evaluation are given in subsequent sections.
In general, A vdll not have the abovedescribed behavior along
the given contour P , but rather along some other path l'
z z
One then attempts to deform. Ii into l' in order to apply' the
z z .
preceding argument. In such a path deformatiQO, prqper account
must be taken of any interfering singularities of f(z) (such
as poles or branoh points) in the oomplex z-plane.
To gain a physioal insight into the disposition of
the pertinent oontours in the oomplex plane, let us deoompose
q and 3 into its real and imaginary parts
q(z) = u(x,y) + iv(x,y), z = x + iy , (2)
133
- 3 -
L.B.Felsen
where u, v, x, yare real. The three dimensional plot of u
or v vs. x and y resembles a mountain relief map with in-
finite peaks, bottomless valleys, and passes, arising from the
well-known fact that neither u nor v can haye absolute maximum
or minimum values in the complex plane. AlthouZft it is possible
to have stationary pOints with
auf aX"= auf Oy = Ov/ Ox = ov/ Oy = 0 , (3)
it follows from the Cauchy Riemann equations au! Ox = av/ ay,
auf ay = - av/ ax, that 02u! ai = - 02u/ ai, 02v/ oi =
= - a
2
v/ ay2. Hence, if the curvature at a simple stationary
point on the surface u(x,y) or v(x,y) is positive along the x
direction, it is negative along the (perpendicular) y direction,
thereby characterizing the stationary points as "saddle points"
(see Fig.l). Equations (3) imply that
dq
-- = 0 at a saddle point z = z
dz S
Fig.l - Relicf of the function u in the
vicinity of a first-order saddle pOint.
(4)
134
- 4 -
L.B.Felsen
Saddle pOints are classified according to the number of deri-
vatives of Q{z) which vanish at z. If the lowest non-
-vanishin,'; derivatives at Zs is q(n)(zs)' the saddle point
is said to be of order (n-l). It is evident from Fig.l that
the flatness of the saddle increase a with the order of the
saddle pOint.
From the preceding remarks oonoerning the behavior
of A = exp ( g u), the most desirable path F 1s the "steepest
z
descent path" (SDF) along which u decreases most rapidly away
from its maximum value. There exists a whole family of auah
curves, for Jifferent of u ,whioh conneots two
max
valley recions, as noted from Fig.l. To obtain as good an ap-
proximation to 1(0) as possible, one selects that path which
is oharaoteri'zed by a minimum value of '\m.x' since both I( Q)
and its asymptotic approxLuation -- and hence their difference,
the error -- \"/i11 be proportional to exp[ Q u ]. The desired
. max
oontour therefore passes through the saddle point z To
s
asoertain the proGress of the SDP through zs' one seeks that
path in the z-plane along whioh du/ds, the rate Qt change ot
u, is a Since
du ox ou oy Ou ou.
- :0: - - + - - = - COSel + - Sln a ..
ds os oy as Ox 6y
where CI is the angle between the path element ds and the
positive x- axis, one obtainsfor the condition of maximum
rate of change
02u au. 6u dv
- = 0 = - - Sln a + -. cos a = -
(6)
oa os Ox 6y ds
135
- 5 -
L.:B.Felsen
wherein the last is a consequence of the Cauchy-Riemann
equations. Hence, u changes most rapidly along a path on which
Imq:v = constant, Le., along a "constant phase" path for
the function exp[ Q q(z)j While the constant phase condition
characterizes both the steepest ascent and the steepest descent
paths, only the latter ie of interest herein.
For a first-order saddle pOint, q" (z ) f. 0, and one
s
may represent
q" (z )
q(z) = q(z ) + I s (z - z )2 +
s 2. s
Upon substituting (7) into exp[ Q q (z)] and defining the angle
r as r = arg(z - zs) + (1/2)arg q" (zs)' one finds that
/exp[ Q q(z)Ji decreases most rapidly when r=!" 12 , in-
creases most rapidly when 0, " , and is constant when
I
f = "/4, :t 3 1t/4 (on the latter contours, the "level 11 is
constant but the phase chan(:es ;;J.ost rapidly). Since only the
lowest order term. in (z - z) has been retained in (7), this
s
simple disposition of the terrain into two valley regions
(" /4 < r <. 3 ,,/4, - ,,/4 > -3"/4) and tV/O mountain
regi ons (- "/4 .( If' <. " 14, 3 ,,14 < <. 5 ,,/4) of equal
angular width, holds only in the immediate vicinity of the
saddle pOint.
whence
For a second order saddle point, q I (z ) = 0= q" (z )
s s
q(3)(z)
q(z) = q(z ) + s
s 3!
3
(z - z) +
s
(8)
136
- 6 -
L.B.Felsen
In this case there are three mountain and three valley regions
emanating from the saddle point, each having an angular width
of 60. Similar considerations apply to saddle point of higher
order.
To effect an approximate evaluation of I(Q) for
large values of Q by the procedure sketched in the biginning
of this section, one seeks to deform the given path Pinto
z
a steepest descent path through the saddle point zs' It may
sometimes be necessary to pass over more than one saddle point
in order to connect the endpoints of P to steepest descent
z
paths; under these circumstances, one includes the contribu-
tion from each point (unless their levels differ appre-
ciably, in which instanct only those giving the largest contri-
bution need be retained). It should be pOinted out that, in
view of the localization of the dominant contribution to I(Q)
to the vicinity of the saddle pOint, it is sufficient to follow
the SDP near Zs only; for example, between points zl and z2
for which He q(ZI,2} ( He q(zs) If on the remainder of the
path Re q(z) He q(Zl 2} , the contribution therefrom is pro-
portional to exp L Q ;(Zl,2)-Q q(zs)] ,i.e., exponentially
small as Q becomes very large. By the same considerations ,
one can effect an approximate evaluation of a finite integral
between the limits zl and z2' provided that the path can
be deformed through a saddle point z and that He q(z} (
s
< Re q(z ) along the entire path.
s
137
- 7 -
L.B. Felsen
2. TRANFORMATION OF THE GIVEN INTEGRAL INTO A CANONICAL FORM
From the discussion in Sec. I it is noted that the
asymptotic evaluation of the integral I( 0) (as C 00 )
by the method of saddle points is intimately connected with
the properties of the analytic function q(z) , in particular
with the nature and disposition of its stationary points. In
this section, it will be assumed that the given integration
path has been deformed into a steepest (or, less stringently,
a rapid) descent path through one or more of the pertinent
saddle pOints of the function q(z) , and attention is given to
the actual evaluation of I(O). In most cases of interest, the
relevant stationary points are isolated, and the asymptotic
evaluation of I (0) is effected by treating separately the
contribution from each. However, there are oJcasions when two
(or more) saddle points are clustered in close proximity and
can no longer be considered in isolation; instead, the combined
configuration laust be taken into account. Moreover, the analytiC
properties of f(z) in the integrand of (1) influence the manner
of evaluation of the integral. If f(z) possesses singularities
near the pertinent saddle pOints, their effect must likewise
be considered.
Since the major contribution arises from the vicinity
of each isolated configuration of stationary points, it is
desirable to transform the integral in (1) into a " canonical "
form wherein the function q(z) is replaced by another function
which characterizes the saddle point arrangement at z in the
s
Simplest manner. Such a function, having the same number and
order of saddle pOints as q(z) (i.e., zeros of ql (z) ), is a
138
- 8 -
L.B.Felsen
polynomial. The transformation will be phrased in terms of a
new variable s and the polynomial denoted by tt: (s;) :
q(z) = r (s)
moreover, it will be convenient to have the point z in the
I s
complex z-plane cOITespond. to s = 0 in the complex s-plane.
Upon changing variables from z to s, one obtains from (1)
where
I( Q) = 1 G(e) .Dr(e) de
p
(lOa)
G(s) = f(z) ddZ , and ddZ = 1:' (lOb)
s s q' z
P is a steepest desoent oontour leading to infinity away from
s = 0 (Le., Re (s) < Re't(O) on p), so that the exponential
term decays very rapidly away from the origin. Hence, if G(s)
is regular and slowly varying near s = 0, one may approximate
I( g) in (lOa) by its asymptotic representation
1(0) { .0 't(s) ds ,
(11)
'If a cluster of saddle pOints exists near s = 0, G( 0)
is generally replaced by a suitable measure of G(s) near
I
s = o.
139
- 9 -
L.B.Felsen
In (11), I(Q ) is expressed in terms of a canonical integral
whioh can be reduced to known funotions for certain Z(s)
(11) represents only a lowest order approximation to I (Q );
higher order terms, smaller by inverse (generally fractional)
powers of Q ,can aleo be obtained (see reference 1).
If it is assumed that f(z) is regular near z ,
a
the regularity Of (1(8) 8 = 0 implies a like behavior
for the mapping derivative (dz!ds). Thus, il(s) must be chosen
so aa to possess at the pOints 8 zeros of the same order as
s
those of q I (z) at z , where the saddle pOints s in the
s s
s-plane correspond to z in the z-plane. From these remarks
s
one deduces for an isolated first-order saddle point at z
8
the transformation
q(z) = 7: (s) = q(z ) - s2 , (12)
s
while for an isolated Mth order saddle point ,
q(z) = "C (s) = q(z ) _ sM+l
s
(13)
If two first-order saddle points at zl and z2 are located so
near one another that they must be treated jointly, one employs3
q(z) = '7:(s) =
s3
a + as - - ,
o 3
(14)
where a and a are constanta and the factor 1/3 in the a
3
o
term has been included for convenience. (The 8
2
-term in the
polynomial ia not required). The two distinct first-order
140
- 10 -
L.B.Felsen
zeros of r' (s), sl,2 =! fa ' correspond to zl,2
q (zl 2) = 'Z' (:!: r; ) , one finds
and from
,
(14a)
For more general saddle point configurations, the polynomial
t(s) takes on a more oomplicated form. However, the above-
-listed examples cover most of the cases arising in connection
with diffraction problems.
The canonical integral in (11) can be evaluated for
each of the special forms in (12)-(14). For (12) and (13), the
integral is expressed in terms of the eamma function, while
...
(14) leads to Airy functions.
So far, it has been assumed that f (z) has no singulari-
ties near the pertinent saddle point(s). If such singularities
do exist, their effect has to be considered explicitly4. For
example, if G(s) has at s = b (b small) a pole of order N,
then one may represent
G(s)
a_
l
+ - + T(s) (15)
(s-b)
where T(s) is regular at s = b and at s = o. The contribution
to I( Q ) from T(s) can be treated as before. Inclusion of the
Conditions on I a I ' which allow the two saddle points to be
treated as isolated, are discussed in Sec. IV
f*
For three equally spaced, colinear saddle pOints, the canoni-
cal integral ('1) is expressibl.e in terms of parabolic cylinder
functions.
141
- 11 -
L.B.Felsen
remaining term.s in (15) requires evalua.tion of a new olass of
oanonioal intsGrals
f
)
-N Or(s)
) = (s-b e ds , u= 1,2
P
(16)
For a first-order saddle. point, with T(s) given by (12),the
integrals in (16) oan be evaluated in terms of the error fun-
ction. If G(s) also has a branch point sinooularity at s= 0
so that the expression on the right-hand side of (15) is multi-
plied by > -1 , then with 1:(s) as in (12), the integrals
(16) can be expressed in terms of Whitaker's confluent hyper-
geometriC function
ll

3. INTEGRANDS WITH ISOLATED SADDLE POINTS
A. Saddle Points
1. First-order Approximation
We now consider in detail the first-order
approximation of the integral in (1), as well as its oomplete
asymptotic expa.nsion, for the case where q(z} has one relevant
fiISt-order saddle point at z and f(z) has no singularities
s
near zs' The pertinent change of variable from the z-plane to
the s-plane is that given in (12) and the steepest descent path
P in the s-plane, along which 1m 1(s) = constant, is clearly
the real s-axis. Thus, the integral in (lOa) is in this case
142
- 12 -
+.0
J G(s)

2
- as
e ds
L.B.Felsen
(17)
with G(s) given by (lOb), and
dz -2s
d; = q' (z)
(17a)
Since G(s) is assumed regular near s = 0, it can be expanded
into a power series
G(s) = G(O) + G' (0) s + Gil (0) + + G(n) (0) + , (18)
2. n!
which oonverges uniformly inside a circle with finite radius
r oertered at s = 0, r being the distance to the nearest
singularity of a(s). Upon L'Hopital's rule to the
indeterminate form for (dz/ds) in (17a) when s= 0, i.e.,
z = z , one evaluates the first coefficient of the expansion
s
as
G(o) - f(z ) - + r--=;-, (18a)
- s ds s=O' ds s=O - -
v/here q" (z ) f. 0 at the first-order saddle point. The choice
s
of sign of (dZ/ds)s=O depends upon the direction of integra-
tion along the steepest descent path through z in the z-plane.
s
(Since ds is positive along the path of integration and, in

partioular, at s = 0, arg(dz/ds) at s = is equal to arg(dz)
at z along the steepest descent path.) If G(s) is approxi-
s .
mated by G(O) only, one obtaim as the first-order asymptotio
143
- 13 -
approximation of I( Q) in (17) :
r:.o
Oq(z ) J
I( Q) tV G(O) e s
_ ....
L.B.Fe1sen
2
- Os
e ds (19 )
Since the integral in (19) is equal to , one obtains
from (19) and (18a)
+ K Qq(zs'
1(0 ) tV _ f(z ) e , 0..,.01'.
Q q" (z) s
s
(20)
2. Complete Asymptotic Expansion
A complete asymptotio expansion for I ( g) in (17)
is obtained upon substituting for G(s) the power series ex-
pansion in (18) and inte6ratin& term by term:+
0>0
Oq(z ) '" (n) ( )
I( 0) N e s L G I 0 In( 0) (21)
n=O n,
+ This manipulation is not rigorously justifiable since the
radius of oonvergenoe r of the power series expansion is
generally finite so that the series representation for G(s)
cannot be employed over the infinite range in s. However,
the error inourred by this (as Q -,cO) is exponen-
tially small, thereby implying the "asymptotio" validity of
the resulting expansion (see ref.l and 2: ; also G.N.Watson,
Proof London Math. SOOt (1918)),
144
- 14 -
L,B,Felsen
The integral I (0) can be evaluated in terms of the gamma
n
function r (z)

I (0 )
n
j
2
n - as 2
= s e ds = , n even
0(1+n)/2
_tIO
(22a)
= 0 , n odd , (22b)
where, in view of the symmetrical integration interval, (22b)
results from the fact that the integrand is an odd function
of s. The values of n=0,1,2,.", are readily in-
ferred from the recursion formula
Alternatively, one may express
0::>
I ( Q) as
n
J
-0 s2
e ds
-()Q
, n even.
(23)
(24a)
(24b)
(24b) gives rise to a recursion relation between In and
1n+2 ' evident from (22a)
d
I 2( Q ) = (--) I (0) , n= 0,2,4... (25)
n+ dO n
145
- 15 -
L.B.Felsen
Thus, the complete asymptotic expansion for I (g ) , as Q 00,
is given by :
O:i(z )
.,0
s
T
G(2n) (0)
nn+l/2)
I ( 0) ,,"'
e
,I
(26)
rD-

(2n)!
n
n=O
Q
or, alternatively,
CIO
Q q(z )
G(2n) (0)
Ii
)
d n "
I(O)",e s
(--) -. (27)
'--
(2n) !
dO Q
n=O
(27) can be written in a convenient operator notation' as
1(0)1\1 e S G (, _.2..) .1l
Qq(z) R rt
e dO Q
(27a)
vlhere the representation of the even function G (x) in t:H'Irrs
e
of a power series about x = 0 is as follows :
0'
"
G I
e _
n=O
G (2n) (0)
e 2n
----x
(2n) !
(27b)
whence the series representation for (27a) is that in (27).
That Eqs. (26) or (27) indeed constitute the asymptotic
expansion of I (Q) as Q -?> oLJ follows from the racogni tion
that the ratio between successive terms of the series, i.e.,
between the (N + l)th and Nth terms, approaches zero as
146
'If
Q C'() , for any N
- 16 -
L.B.Felsen
As a general COL"J!lcnt on as;Yl1I;totic it
should be out th3.t a for;:al asyt1lltotic serias as in
(27) MY for a value of Q ,as N . The
asymptotic expansion IT:.ay nevertheless be t,m.ployed to com}mte
the nUIlerical value of I ( Q) for a .;;i.ven Q since the first
terms in the expansion decrease in magr,i tlloe. In general, the
series should be broken off at the sJ!1allest contributing terll1,
for a given value of Q
One notos that the lowest order approximation to
I( Q) arisingirom the n=O term. in (26) or (27) is .siven
by (20) as required. For the evaluation of the higher order
tems one must lmow the higher crc.or derivatives of G(s) =
= f(z) dz/ds ,3valuatGd at s = O. FOl':r.ally, the dnriv3.tivas
if A fu.nction I ( Q) is said to have 3.1: asymptotic expansion
"..0
'"
I( Q)N ' a f ( Q)
- nn
11=0
as Q -000 ,
if, for any N and for are Q in a ,?-ven intGcral ,
T,T
,-
where I
N
( Q) =) a f ( Q). For d3tails on c:eneral properties
n n
n=O
of asymptotic 0:cpansions see referonces 2.
147
- 17 -
L.B.Felsen
(dnz/ds
n
) can be obtained by successive differentiation of
(lOb) and evaluation of the resulting indeterminate form at
s:= O. An al ternati ve procedure is to expand q (z) in (12) in
a power series about z = z to obtain the dependence of s
s
as a function of (z - z ). To obtain the required behavior
of (z - z) as a of s, this power series must be
s
inverted. (See ref. 10 f'or inversion of series),
A1 though it has been assumed throughout that g
is real, it is evident that I (g) as defined in (22a) can
n
be continued analytically into the range I arg Q I <. ",/2,
since the integral converges as well in this extended range
of g for which He( g s2 ) O. This process of analytic
continuation is frequently more convenient that the inclusion
of the factor exp(iarg g) in q(z), if g is complex (as
mentioned in the of this Chapter). Thus, the asym-
ptotio expansions in (26) or (27) are valid also for complex
g ,in the range I arg Q I < '" /2.
Example
Consider the integral
= J
iO cos (z-a )
e dz , 0 Cl < , ( 28)
z -
p
z
taken over the path P shown in Fig.2. This integral is typical
z
of those which arise in several of the excitation and diffrac-
tion problems studied in this course. Since
1m cos{z- a) = - sin(x-a ) sinh y ,
(29)
148
- 18 -
L.B.Felsen
and Q is assumGd positive, one notes that the exponential
term decays in the regions y) 0, - 'Jt < (x - a ) < 0 , and
y < 0, 0 ({x - a) < 1: 1 \l'hiC:1 a!'e sbovm shaded in Fig.2.
The integrand contains a simple pole at 2; = p, where p is
arbitrary. Upon comparison with (1) ona notes that
y
Fig.2 - Contours of integration in z-plane (z=x+iy)
1
q(z)=i cos(z-a), f(Z)=(z_p)
(30)
The pertinent saddle point Zs in the z-plane is
located at
149
- 19 -
L.B.Felsen
-dd q(z) = - i sin(z- cd =0, i.e. at z = 4. (30a)
z s
The steepest descent path P
z
through the saddle point is de-
fined by
Im q(z) = Im q(z ) = i
s
(30b)
The change of variable to the s-plane is then effected via (12)
bY'
. ( ). 2
1 cos Z - ex = 1 - S , (31a)
or
(3lb)
Although we could proceed immediatelY' to the s-plane via the
transformation (31), we study first the nature of the steepest
descent path P
z
in the z-plane. Along P
z
' s is real, and
the slope of P at z = Z = ex is inferred from (18a) or (Jlb)
z s
to be
(32)
Thus, P makes an angle of (-45
0
) with the x axis at z = 4.
Z
The direction of integration along P in Fig.2 suggests a
z
choice of the indicated direction along P , so that arg(z-z )=
Z s
= -" /4 along P near z = z , and the positive sign is chosen
z a
in Eqs. (32) and (31). The complete steepest descent path can
be plotted readily from Eq. (30b) which requires that
IlIl [i cos(z - )] = i along P
z
' or
150
- 20 -
-1
x - a = cos (scch y)
L.B.Felsen
along P
z
(3)) yiQlds the x-coordinate of any point on P for an as-
z
su,ned value of y. The resulting path is shown in Fig.2 and
has as its asymptotes the lines x = a! /2
The transformation of the P from the z-plane
z
to the s-plane is accomplished via Eq. (3lb), where it is re-
called that the positive sign is chosen. The resulting contour
P is shovr.n in FiG.) and has as its asympotes in the lower
half of the s-plane the lines arg s = - a/2 and arg s =
=1t + ( a/2) Corresponding regions in the z- and s-plane ,
where the exponential tern. decays in rnagni tude, are shown
shaded with slanted lines. Since (ds/dz) = 0 at (z-a) = :t ,
the transformation in (3Ib) Gives rise to first-order branch
point sineuJ.ari ties at s =! sb = {2 c:x::p /4) in the
s-plane. These branch points, and the associated choice of
branch cuts, al'3 also a:lO\m in Fig.). The staepest descent
path P in the s-plane, corresponding to the path P in the
z
z-plane, extends along the rBal a-axis.
Since all paths considered begin and terminate in
a shaded region of the z- or s-plane it is evident that the
clmtours P and P can be deformed ..it infinity into the
z
contours P and P, respectively. Concerning the defomation
z
of the paths in the r8nainder of the z- or s-planes, attention
must be ,gi von to the location of the pole sin5Ulari ty at z = p
in (28). If z= p (or s= sp = i2 ,}1t/4 is situated
in any of the vertically shaded reGions A, B, C in Fig.2 (or
Fig. 3), the residue at the pole must be into account in
151
- 21 -
L.B.Felsen
Fig.) - Contours of integration in s-plane
the contour deformation. Thus,
where
dp) =
Q.()
I
e
iQ
-(1) t( p) + e
iQ
( G(s) e - Qs
2
da, (34)
!
-
-00
+ 1, if P (or lies in regions B or C
- 1, if P (or sp) lies in reeion A
0, if (or sp ) lies outside regions A,B,C.
(34a)
In view of the change of variable in (31), G(s) is given by
)
1 dz
G(s =- -
z-p ds
dz -2is -2i8
- = --- = -;:::::====
ds . sin(z- a) r 2 -
V I-cos (z- Cl)
(34b)
152
- 22 -
L.B.Felsen
The simple form of (dz/ds) permits a direct determination of
the complete series expansion by the binomial theorem
r;
. 2 -1/2
dz 2 e 4 (1 + 1S
2
)
de=
which converges in the interior of a circle of radius \sl= (2
passing through the branch point singularities. Thus, the range
of convergence of the power series expansion for G(s) is
lsI < r2 if i sp I > {2 and is I s\ ( \ sp \ if t s pi ( (2,
i.e., within a circle of finite radius, !lrovided I sp I > 0
The first terms in the asymptotic expansion of Ij Q, p)
in (28) (as 0 are therefore [;iven via (34) and (26) by
Concerning the residue contrimltion from the pole to
Eq.(35), one notes that the magnitude of the exponential term
behaves like exp r
L
- Q I sin( p - a) sinh p. IJ' ,where p and
l' 1, I'
Pi are the real and imaginary parts of p ,respectively. If
p . 0 and p a , the pole contribution is exponentially
1 r
small and can be neglected in comparison the remaining
terms. On the other hand, if p. 70 or if P a , the residue
1 r
contribution may be the dominant one since its magnitude then
remains constant as Q
Although the asymptotic representation in (35) re-
mains valid (as g 00 ) for any p a , the proximity of the
153
- 23 -
L.B.Felsen
pole near the saddle point ( p a) is seen to
the accuracy of any numerical evaluation for large fixed
values of Q In particular, if the pole approaches the
saddle point (( a - P) .. u) , the utility of the representa-
tion in (35) ceases sinoe the error inourred by use of even
the first term becomes large. implies that the radius
of oonvergenoe of the series representation for G(s) shrinks
to zero). One observes in this oonnection that the quantity
nr I a- p I plays a speoial role in assessing whether the
pole is near enough to the saddle point to invalidate (35).
If for Q l, one also has {Q /a-P I 1 , the pole
oan be oonsidered far from the saddle point and (35) applies;
on the other hand, if {Q /a-P I 1 ,with Q 1, the
terms in the asymptotic series are no longer small and the
validity of the expansion is in question. The dependenoe of
the asymptotio expansi on of Il ( Q, a ,p) on the magni tude of
(Q I Cl-P I is studied in detail in Seotion V, whereih is
treated the evaluation of integrals whose integrands contain
a pole near a saddle pOint.
B. Saddle Points of Hi6her Order
If q(z) in (1) has one pertinent M-th order saddle
. . (n) ( ) (14+1) ( ).1
pOlnt at z , l. e., q z = 0, n = 1 ... !'I'l, q z,.. 0 ,
s s s
the appropriate transformation to the s-plane is given by (13):
. 14+1
q(z) = Z(s) = q(z ) - s
s
(36)
requirement Imq(z) = constant, i.e., sM+l positive along
the steepest descent paths, leads to the following possible
contours in the s-plane ,
154
- 24 -
L.B.Felsen
arg , IJ.= 0,1,2 III. (37)
Thus, the (M+1) steepest descent paths originating at the
saddle point z in the z-plane map in the s-plane into the
s .
(1+1) straight lines defined in Eq.(37), which originate at
s = 0 and extend to I s I = pO Since sM+1 is positive alone
any of the paths defined in Eq.(37), we may consider without
loss of generality the case IJ. = o. Let the subscript
o
denote the contribution to I ( Q) in Eq. (1) from a
descent path which maps into the positive real s-axis. Then
the integrals I (Q) in (21) are given in view of (36) by:
n
I (Q)
n
J
( M+l
: n -Qs
= I s e
j
o
ds =
(J8a)
Q -(1+n)/(M+1)
=
M+l 111+1 (38b)
where the integral in {38a} has been evaluated in terms of
the gamma function. The asymptotic nature of the resultin6
expansion in (21) as Q"?>90 follows by the same considera-
tions as before.
Upon expanding q(z} in (36) in a power series
about the pOint z , one obtains directly
s
1
dz I [ -{M+1}! ] M+l
de s = 0 = P (M+ 1) ,
q (z )
s
I(M+1)
P = e ,
155
- 25 -
L.B.Felsen
where p is the appropriate (M+l)-th root of unity whose choice
is determined by the given steepest desoent path (see (37)) ,
and where the principal value of the (M+l)-th root is taken in
the remaining term. From (21) one obtains the following first-
-order approximation to the integral 10 (0 ) ( Il = 0) :
where it is reoalled that G(s) =f(z)(dz/ds) When 14=1, one
recovers from Eq. (40) the previously derived result in (20) ,
save for a factor of 2 which arises since the interval of in-
tegration in (19) extends from s = - tJQ to s = 00
4. INTEGRANDS WI TH TWO ADJACENT FIRST-ORDER SADDLE POINTS
It' q(z) in (1) has two pertinent first-order saddle
pOints which approach each other very closely, the numerical
evalua tion of the integral along a path through one saddle
point is markedly influenced by the presence of the other.
In this case one seeks to approximate I ( g) by a known standard
integral whose integrand also contains two saddle pOints
3
An
appropriate class of standard integrals are the Airy integrals
Ai ( 0') and Bi ( cr )
8
which are defined as follows
Ai( 0 )
1
=-
2d
f e as-.
3
/3 d,
L32
(41a)
156
.. 26 ..
L.B.Fclsen
Bi(a )
1
J
a s-s3/3 d
e s (41b) =-
where Li j' the contours of. integration in the complex s-plane,
are shown in Fig.4. 4Each contour begins and ends at I s I = 0t:I
in a shaded region, Re s3 > 0 , wherein convergence of the in-
tegrals is assured. (The exponent in the integrands of
manifestly has two saddle points located at sl 2 =! ra.
,
Thus, a is a measure of the distance separating the two saddle
points.
I ... S
... -----
-.. -
Fig.4 .. Airy integral paths.
157
- 27 -
L.B.Fe1sen
Let us return to a oonsideration of the integral in
(1), when the derivative q' (z) has two simple neighboring
zeros at zs=z1,2 and f(z) has no singu1a:r:ities near z1,2'
As discussad in Sec. I, we wish to transform the region oon-
taining the two saddle points zl 2 into the vicinity of the
,
origin in the complex s-plane. The required transformation
has been g"venin I Iq.(14), the oonstant a
o
and a
defined in (14a), (Special attention must be given to the
choice of the proper branoh of ( a
3
/
2
) 2/3 , as required for
the eValuation of a from (14a)).
Upon substituting (14) into (11) one obtains the
first-order asymptotic approximation for I ( 0) in Eq. (1) as
o -?> 00 , valid for small values of a
I(O ) rJ g -1/3 G(O) e aoO c( a 0
2
/
3
) ,
wi th a
o
and a defined as in (14a) and
G(O) = S = 0
a(p ) = le
P
t-t
3
/3 dt
p
(42)
(42a)
(42b)
See footnote assooiated with Eq.(ll). For arbitrary a ,
G(O) in (42) must be replaced by (1/2) I G( ra ) + G(- ra
(see ref.l,3). -
158
- 28 -
L.B. Felsen
The inter,ral in (42b) is readily identified in terms of the
Airy integrals in (41),
in Fi004. Since j
L2l
for any specified allowable path P
r + f ,0 one note. that
L3l Ii32
2"i Ai( p)
, if
P =L32
O(p) =
" -iAi(p)]
, if
P = L2l
(43)
,,[B1(P) +iAi(P)]
,if P=L
3l
Eq.(42) is valid for small values of a ,i.e., for
In this case, since f(z) and (dz/ds) are assU!Il.ed to be regu-
lar and slowly varying functions in the vicinity of s = 0, one
may write approximataly
and
I
f(z) I f(zl) ,
's=O
From (14), one finds readily that (see also (18a
ff
,----
dz 1/4 -2 dz 1/4 2
(-) =+ a -, (-) =+ a - ,
ds - ds-
s= ra q' ("1) s=- fa J q" ("2)
(44a)
(44c)
where the choice of sign depends as in (18a) on the direction
of integration along the path. The derivative (dz/ds)s=O must
159
- 29 ...
L.B.Felsen
have a unique value, implying that -q"(Zl) for
Zl?lZ2' i.e., at;::. 0; it thenfollQwB that q"(Z2)=qll(Zl)=0
if zl = z2
If the two first-order saddle points zl,2 coincide,
the point Zl = z2 is a second-order saddle point since both
q'(z ) and q"(z) vanish. Thut;J, in (14), a =0, "2:'(0) ='t"(0)=0,
s s
and the origin in the a-plane i8 likewise a seoond order saddle
pOint. To evaluate the indeterminate form as a 0, write
(44e) a.s
As a ...;7' 0, q" ( zl ) O. Hence, by L' Hopi tal's rule,
whence
2
=
ds 0
s=
-2
s=O
z=z
1
(47) contains an ambiguity regarding the proper choice
(45)
(46)
of the cube root. Its depends on the inte-
gration ion tbe Pl'oblem u,n,der consideration and j s de-
mo.nstrated in an example at the end ef Ws li-
miting form ot (42) for a = 0 is now given by
160
- 30 -
L.B.Fclsen
(48)
where in view of (38) and (42b)
0(0) =
e
-
i
,,/6
3
-1/6 r (!3) ,
if P = L21
(48a)
One verifies readily that (48) agrees with the
saddle point result in (4C) (with M=2) provided that the path
P for 0(0) is taken from s = 0 to s = QO only. In this case,
the value for 0(0) as obtained from (38) is 3-
2
/
3
r (1/3)
The given in terms of this result as
folloVls :
etc.
When 0
2
/
3
in (42) becomes large one may use the
asymptotic representation for the Airy-type function C( O'Q 2/3).
Since Q 04 ,one notes that a Q 2/3 co if f1 is pro-
portional to Q -2/3+ I where > 0 but may be small.
161
- 31 -
L.B.Felsen
As will be seen below, the use of the asymptotio representation
for C( 00
2
/
3
) reduces (42) approximatel1 "1;0 the result ob-
tained when each saddle point is treated separately. Thus, the
double saddle point procedure is required only when 0 =0(0
i.e., very small. To be specific, let us assume that the path
P in Eq. (42) is the same as in Fig.4. Then,from (43) and
the asymptotic repre sentat1 on "(S88 Ref.2o)
1 -(2/3)0
3
/
2
I
Ai ( a) N e , 0 I arg a (j, \ (50)
2.[; i/
4
one obtains
as 0 0
2
/
3
- c>o. Since (51) is valid for small values of a,
i.e., z2 t Eqs.(44) still apply and substitution of (51)
into (42) yields for a small but a 0
2
/
3
-,">4'0 as g -">00'
I (g) (52)
which is identical with the saddle point result in (20). (52)
involves the saddle point at s2 = - /a which is evidently
the pertinent one for the contour L32 in Fig.4 when 0 is
real and a I < ,,/3. To achieve the form in (20), the
quantities in Eqs. (44) have been evaluated at z2 and s2'
162
- 32 -
L.:B.Felaen
Similar results are obtainable when the original integration
path is deformable into any of the other contours in Fig.4.
For the complete asymptotic expansion of I (Q) in
(1), with q(z) given by (14), the reader is referred to re-
ference 3.
Example
Consider the Hankel function ( Q) defined by
the integral 9
.Qq(z)dZ, (.,Q positive), (53)
p
z
q(z)=i rcos z+ sina 1
- -
, sin a
,'f,
=11 '
(53a)
where the path P is the same as in Fig.2. The saddle points
z
of q(z) are located at
= -i z -sin a] = 0 , (54a)
which has pertinent solutions at
(54b)
We assume that (v / Q ) 1 so that 0 <. a '1/2, and seGk a
first-order asymptotio evaluation of the integral in (53) as
D 0Ci) and a -:> ,,/2, i. e., when the order and argument
of the Hankel function are both large and almost equal.
From (53a) and (54b),
163
- 33 -
L.B.Felsen
Q(Z2) =! i [cos 11 - (I -11 ) sin 11]
(55)
Sipoe q (z is imaginarj, a 3/2 as defined in (14a) is imagi-
nary and oh be satisfied by a (0. Let us introduce
(56a)
and choose
=!ra = 1: i R I > 0 (56b)
Then from (14)
2 3/2 (") .
3 = cos 11 - 2 - 11 S1n (1 4 0 ,
(57a)
t (0) = 0 , (57b)
while from (44c)
(570)
Since f(z) =1 in (53) and in view of (57c), one notes that
(44a) and (44b) are satisfied exactly in case, Near
CI =" /2 I one obtains from (57a) and the requirement that
G 0,
(58a)
Thus,
164
- 34 -
L.:B.Felsen
(58b) oould also have been obtained directly from (47) with
zl = z2 = 'II /2, wi th the cube root so chosen that i
l2
is
imaginary (see (56b)). From (14a) and the power series expan-
sion of q(zl) about z2' one finds for q I (z2) = q" (z2) = 0
that i
I2
;-.1(Zl-Z2) [q{J) (z2'] 1/3 Since Q(3)(Z2)=i and
(zl - z2) is real, this rGquirement implies that the principal
value of the cube root in (47) and (48) is multiplied in this
case by exp(i 4'1l 13 ).
To determine the contour of integration in the
s-plane, it suffices to consider the transformation in (14)
as a -) 0 (Le., -"!> 0) in order to establish the loca-
tion of the endpoints of the transformed path. Thus, we examine
r 1 s3
=-1" ' (1 =0) ,
- -'
whioh can be written near z = ",/2 as
i. x 3 s3
-6 (--z)
2 3
Upon taking the cube root of (59a) one obtains
, n=O,l,2,
(59 )
(59a)
(60)
where the last factor expresses the three possible values of
the cube root of unity. The branoh to be chosen is that for
Wllich :!: (ds/dz) evaluatod at z = ,,/2 assumes the value in
(58b). As noted before the proper 'Jhoice is n = 2 so that
165
- 35 -
L.B.Falsen
. -1/3 ( "')
s 1 2 z-2
(60a)
One from (60a) that are: s = 0 when z = ('" /2) + iZi'
zi <. 0, and the are s = -w 12 rlhen z is real and z <. 'II: /2.
It then follows from (59) and a continuity argument that the
entire line He z = ",/2 , !Ill z ,0, I!lP..ps into the :posi ti ve real
a-axis, while the remainder of th:e P terminates in the
z
shaded region in the third g,uadrant of Fig.4. Thus, the con-
tour P
z
is transformed into the path L31 in the a-plane as
shown in Fig.4. From the direction of integration along P
z
in the neiehborhood of z ='1/2 one notas that the ohoice of
the minus sign in (570) and (58b) is appropriate.
The first-order a.symptotic repr)sentation (as Q 4>00)
for H(l)(Q) in (53) ean now be wri tton. directly from
" (42) and from (55) and (57). Since P=L
3l
, one obtains:
where and defined in Eqs.(53a) and (57a),respectively.
When = , use of (4811) and (58b) yields
l/3 - 1/3 -l/6 .
[A1(O) - iB1(O)J = r (J)e-
1n/3
(62)
When a is sufficiently different from n 12 to yield. ('Ill)
la.rge, we may e:n.ploy the asymptotic expressions for the Airy
f
" ti 2c
une ons
166
- 36 -
L.B.FI31sen
U(-a)--.t
1
sin aJ/2 + i)

a
1
/
4
(63a)
Bi(- a) '"
1
(g a 3/2 + ! )
cos ,

1/4
3 4
a
(63b)
valid as a - c>C , I arg a I < "/3 , to obtain the Debye
formula
i gICOS a)sin (1J-i"/4
e (64)
The condition, g 2/3 1 roquired for the validity of (64)
can be phrased in view of (58a),
as
(g _ v) Q 1/3
If (g- v)=O(gl/3), one must employ (61).
5. INTEGRANDS WITH A POLl: SnmULARITY NEAR A FIRST-ORDER
SADDLE POINT
.
A. Simple Pole
(65a)
(65b)
If f(z) in the integrand of (1) has a simple pole
singu].ari ty at z = z near a first-order saddle paint z ,
a s
G(s) in (lOa) will possess correspondingly a simple pole
167
- 37 -
L.B.Fe1sen
singulari ty in the "licini ty of s = 0, at say s = b. Sup:?ose
that G(s) (s - b) --.. a as b; then G(s) can be repre-
sented in the vicinity of s = 0 by (15))
G(s) = + T(s)
s-b
It will be convenient to employ the identity
and to ex:pand
a as ab
2 ,2
s - 0
(66)
(66a)
2
T(s) = T(O) + T' (0) s + Til (0) -T + , (66b)
2.
which is regular as s = b and has a radius of converJence
uninfluenced by the presence of the pole.
Since the saddle point is of order one, the tran-
sformation in (12) applies and the integral in Eq.(l) can De
written as:
I (Q , b)
+..0
Qq(z)( n2
s r ) -.,s
= e I G(s e ds ,
,)
(67)
or, via (66a, b) and (27a) as 4
The function A(a,b) is given in terms of tho inteGral
168
- )8 -
-tOO
A(Q, b) = ab j 2 1 2
s - b
-Do
2
-g s
e ds.
L.B.Felsen
(69)
The integral in (67), with G(s) as in (66), is not defined
when b is real. Viewed as a function of b, a study of the
aDalytia properties of I ( 0, b) as 1m b -? 0 reveals that
the integral is disoontinuous across the real b-axis. Suppose
that b approaohes the real b-axis from the range of posi ti ve
imaginary b,i.e., b'""1'b +i d , with b b real and
r r
& = 0 +. Then the path of integration is indented at
s = b + 1 [ as shown in Fig. 5(a). Similarly, when b -id,
r r
s-plane
II,

(a)
+
1m b = 0
s-plane
(b) 1m b = 0-
Fig. 5 - Contours of integration
the appropriate path is that in Fig.5(b). Upon oonstruoting
the difference I( 0, b + i d) - 1(Q ,b - i to exhibit the
r r
d1soontinui ty in I ( g, b) across the real b-axis, one notes
that the oontributions to the integrals from the straight
portions of the paths in Fig.5 oancel, and there remains a
169
- 39 -
L.B.Felsen
small circular contour enclosing in the positive sense the
pole at s = b Since T(a) in (66) is regular inside this
r
oirole, itg contribu.tion vanishes and one obta.ins from the
residue at s = b
It is easily verified that A( Q, b) satisfies the
differential equation
d 2 rn
(dii+
b
) A(e, b) = -ab
fo solve, substitute
_ Qb
2
A (0 , b) = e B (Q , b)
into (71) whence
dB Qb
,2 fa
- = -ab e
dO
(71)
(72)
(73)
Upon, integrating (73) ove 0 between the limits a and 00,
one obtains
110
B(D , b) = ab r; i .Ob
2
0 -1/2 d 0 (74)
170
- 40 -
L.B.Felsen
2 ft
where it has been assumed for the moment that b < 0 so
that in view of (72), B( 00 , b ) = O. A change of variable in
(74) from Q to or x=:; ib to ,yields
B(Q , b) = 2ah b Q r(+ ib}
( :; i b) L
(75)
where Q(y} is the II error function complement"
Q(y) = (75a)
y
The ambiguity in sign introduced into (75) by the change of
variable is resolved by the previously imposed requirement
B(OO, b) = 0 for b
2
< O. Sinoe the integral in (75a) will
vanish if the lower limit approaches lltfinity along the posi-
tive real axis, we require for b
2
< 0 a choice of sign such
that
(:;: ib) > 0 , (75b)
i. e., the minus sign when b = i I b \ and the plus sign when
b = -i I b I .
The validity of (75) can be extended by analytic
continuation to values other than b
2
< O. Since the expressions
More generally Re b
2
< 0
171
- 41 -
L.B.Felsen
for B in (75) and (72) (with (69)) represent identical
funotions of b
2
when b
2
<.0 t and since (69) clearly remains
valid for all except positive and zero values of b
2
t the
expression in (75) can likewise be continued analytically to
all values of b
2
except b
2
o. This "continuation from pure
imaginary values of b to complex values must be consistent
with condition (75b). The analytic properties of the error
funotion complement imply a choice of sign in (75) according
to the more general condi ti on He (; i b) > 0, 1. e., the r }
sign applies when 1m b o.
From Rqs.(68), (72), and (75), we can now write
down the asymptotic expansion of the integral in Eq. (67) for
Q 1 and for arbitrary values of b:
4
Qq(z )( r- Qb
2
(d r)
I(Q,b)Ne s f12aYl' e - Q(;1blo)+Te - da)
lft
l' 1m o.
(76)
2
The function e -Q b Q(; ib fa) is tabulated for real and
oomplex values of bra. 5 Thus t the asymptotiC expansion
of an integral whose integrand contains a simple pole near a
saddle point has the same form as that for an integrand wi th-
out a pole, except for an additional term involving the
funotion complecent Q.
It is of to verify from (76) the previously
noted expression for the discontinuity in the value of I(Q,b)
when 1m b changes from positive to negative values. As before,
we define
(77)
172
- 42 -
L.B.Felsen
Since Te ( ;-- d d
g
) is continuous for all b, the jump
lI( n,b
l
) -I(D, b
2
)] in the value of I(Q, b) is given by
To treat the sum I Q(i a) + Q(-i a)] ,a real, we choose for
L..
Q{ i a) in (75a) a :path of integration from :!: i a to 0
and then from 0 to 00 along the real axis. Thus,
Q(ia) + Q(-i a)
o
=j
-ia
2 fO
-x
e dx +
i(1
,00
_x2 ji _x
2
e dx + 2 e dx.
o
(79)
The first two terms on the right-hand side of Eq.(79) cancel
while the third is equal to r; . Thus, (78) reduces to the
previous result in (70).
When b is large enough so that (b {Q ) is likewise
large, one may employ an asymptotic expansion for Q(:j:ib {O )
in (76). This expansion is obtained directly from the repre-
sentation in (74) by repeated integrations by patts:
B(Q ,b) = 2ia f; Q(:!: ib ra )1\1- rl + -+ + 0 ( 4
1
2)1. (80)
li L 2b Q b Q J
The asymptotic representation for I(Q,b) in (76)
is then given by
173
- 43 -
L.B.Felsen
Q q(z ) r,; .r
I(O,b)N e G(O), Ibl, Q 1 , (81)
where
G(O) = - + T(O)
(8la)
In this instanoe, the pole is situated" far" from the origin
in the s-plane and the expression in (81) is identical with
that obtained in (20). Just how large I b I {][ has to be be-
fore (80) oan be employed to within a given accuracy can be
assessed by comparing (80) with the exact expression (75)
whose values for a given b are found from numerioal tables.
A detailed oomparison is made in (91) et seq. for the speoial
case arg(! b) = 1(/4
Example
We return now to the evaluation of the integral in
(28) for the case when the pole at z = P is situated near the
saddle point Zs = a. The representation of II (0, 0:, P) in
(34) still applies. However, G(s) in Eq. (34b) should now be
represented as in (66), with the pole contribution exhibited
separately. To determine the behavior of (z - p)-l as a func-
tion of s, first expand s in Eq. (3lb) in a power series
about z = (note: the plus sign applies in (3lb) ) ,
r +ix/4 L . 1 1. I)-a) m
2
J
s = ,2 e L1n(T)+ 2(00S T)(Z-P>- S(sln 2" (z- +... ,
(82)
174
- 44 -
L.B.Felsen
and invert the series 10 to obtain
z - p = . (s-b) + (s-b) . + , (83a)
l
(2 1 2
P - a 2 P-a)
e cos - 2i cos (-
2 . 2
where
b= f2
/4 . p-a
e Sln-;-
(83b)
ThuB,
(P-a)
e cos 2
....L = ....L + (terms finite at s = b), (84)
z-p
(2 s-b
and "a" in (66) is given by
wi th the value of (dz/ds) at s = b obtained from (83a). By (76)
the asymptotic expansion (as Q 0..,) of the integral in (28)
or (34) can now be down directly as
11 (0 , ., p ) - 2IIio
iO
OOB (a -11) E( P) .olll{ !i2!i e _Qb2 Q{;j:ib,if ).!. (1- ili) Ii J
1m b 0, (86)
where t{ p) is defined in (34a) , b is given in (83b) , and
175
- 45 -
L.B.Felsen
T(s)
1 dz 1
=---- (87a)
z - P ds s - b
T(O)
1
(8Th)
=-
-p
It is of interest to note that the expression in (86) is a
continuous function of b although various terms therein are
disoontinuously represented. This is verified upon an inspec-
tion of Fig.5, (34a) and (70). If the pole at s = b crosses
the real axis in Fig.5, the term inside the braces in (86)
experiences a jump as in (70). However, the first term on the
right-hand side of (86) also changes discontinuously under
these conditions and compensates exactly for the first-mentioned
discontinui ty. For values of b such that I b Ira 1, (86)
reduces via (80) to (35).
The special case where p is real is of particular
importance in various diffraction problems. In this instance,
sin l (P -1x)/2 J in (83b) is real and for I p - a I (n ,
+ ib fa = (1- i) = {Q sinl P; aI, 1m b to. (88)
Moreover, 1m b 'l 0 if (p -a) O. Thus, the following term
in (86) can be rewritten as
-0 b
2
- ]
i 2 1'i e Q{; ib fa ) = 2i," sgn{ P-Q:) e Q ,
(89)
where
176
- 46 -
L.B.Felsen
The funotion
9Q (l .. i) {
.1./4 Q = e 111/4 f .-i dx= .1014( f .-i dx J
0 (90a)
oan be expressed via the ohange of variable
x = 1i. (1- i) t
2
(SOb)
in terms of the well-tabulated Fresnel integrals 0 (x) and
S(x) as
e
1
"/4 Q b-i)fl = e
11th
.Ii fC(2{)+i S(2i )]
L -' 2 12 L fi (;"'
where
C(x) = r 00. [(i) t
2
J dt ,
o
S(x) .
(90c)
(90d)
To provide an estimate of how large has to be
before the asymptotio representation for Q L(l-i)i] in (80) oan
be employed, the funotion
(91)
has been plotted. For \ . .-,r 0 , F( ) .., 1 , wh1ie for 1
one has from (80)
177
- 47 -
L.B.Felsen
(9la)
Upon comparing the graphs of (91) and (Sla) as shown in Fig.
(6a), one notes that the first-order asymptotic formula in
(9la) holds with very good aocuracy when 3. In terms
of this estimate one finds from (88) that the "transition
region tt, inside whioh the. simple asymptotic representation
in (35) fails to apply, is given approximately by
I
-1/2 '
P - II 6 Q An analogous estimate can be found for the
case of a double pole singularity, in which case one requires
values for the derivative of F( (see (95) ). The function
7
(92)
whioh ocours in this connection, as well as its asymptotic ap-
proximation
- i
Hi)
,
(92a)
is plotted in Fig. (6b).
B. Multiple Pole
If G(s) "has a pole of order N at s = b; one employs
the representation in (15). To infer the asymptotic expansion
of I(O,b) in (67) in this case, one must investigate integrals
of the form
2
_t'I S
eNds. (93)
178
- 48 -
L.B.Peleen
2 -i2 t 2 r OQ _y2
PLOT OF J = IF I eL)C = - e ' .' e dy
ff' (1-i) j
FOR LARGE
,'\0
I
I
,9
\
45
\
V
,.,
;.
,
/

__ .11
I I'
/
fg.
1\
\
\
I
,8
.7
40'
\: I
V
/
!\ )
_1

AM
I" .. -
p
i
\1 \.
25
/ 1\'
i7
\
,,\
roO
.\\
I


.)
I
"
"-
I
..........
t'-.
r-....
.1. 10'
-
r-
r-
.1
5'
o o
o
.5
1.0 15
2.0
2.5
15 4.0
Pig, 6(a)
179
1.0
.9
..
.$
.1
.1
o
- 49 -
L.B.lelsen
PLOT OF F = I FIe iJ = 1-2 Vi.' J e -1 (2 J 2 + tI /4) f lIOe -l dy
(1-1) J
II
go"
1
.....
I
V

I
V

17
I
/
70"
I
J
I

I
/
I"t
J
\
,
1/\ t
.so"
I!A
1/
11\
'"
if.
/ \
\
1
\
1\
!O.
\,

20'

"""

.....

r-- 10-.
o
1.15 1.0 .u
t
5.0 U . 0
o .5 1.0
11g. 6(b)
180
- 50 ..
L.B.lelsen
For N = 1 (with a -1= 1), the result has .been obtained in (75)
(with (72):
2
A_I (D,b) = ! 2i {i e-Qb Q(:t ib (Q ), 1m b o. (94)
Sinoe one notes fram (93) that
A N(Q ,b) = ...L..![AN1(Q,b)], N=2,3 , (95)
- N -1 db - +
...
all A_N' N 2 , can be inferred fram A_I by suooessive dif-
ferentiation with respect to b (the integral in (93) is uni-
formly convergent for 1m b 0 so that the differentiation
under the sign implied in (95) is permitted). The
evaluation of the derivatives of Q is readily aocomplished
via the formula (see (75a
d 2
dY Q{y) = - e-
y
(96)
so that
etc.
One notes that as b 0 the dependence on Q is
O{ nr) for A_2 and 0(1) for A_
l
Thus, as expected, a
seoond-order pole in the vioinity of the saddle point yields
a larger value for the integral than a first-order pole. If
b is large enough so that I bl ra 1 , one can employ in (97)
the asymptotic representation from (80). Theresul t is found
to be
181
. 51 .
L.B.Felsen
whioh agress with that obtained fram (93) by a direot asymp-
toti0 evaluation.
183
- 52 -
Appendix
L.B.Felsen
Radiation from a Vertical Dipole above a Lossy Plane Earth
As an illustration of the direct application of Eq.
(86) to a problem which has received much attention in this
course, consider the evaluation of the field radiated by a
vertical electric dipole located near a plane interface bet-
ween a lossy and a lossless half-space. Let t
l
, f and t 2 '
r be the dielectric constant and permeability of the loss-
less and lossy regions, respectively, where &1 and fare
positive real while 2 is complex (0 <.arg t2 < 1f /2 , for
an assumed time dependence ). The electromagnetio
fields in both can be inferred from a single scalar
Hertz potentia! function, rr . If the lossless medium (region
I) occupies the half-space z Z 0 and the source (having a
momentum m) is located at ::' = (r', z,), f'= 0, z' <0, where
(f, z) are cylindrical- coordinates, the Hertz potential 1f1
in region I is Given by:
(A-I)
(1-2)
* See chapter by H.Bremmer, Eqs.(7a) and (12). Also, the
chapter II Alternative Green's Function Representations for a
Grounded Dielectric Slab" by L.B.FeIsen, Eqs. (19a) and (24),
wi th Z in Eq. (16) given by Z = ltd tJ fl (to switch to the
exp(-i'->t) time dependence above;J.et j ... -i)
184
.. 53-
L.B.Felsen
where 'i.,2 = - f = 2/1 ' and (Ret
1
> 0
when is real and t.. 5 l.
If r is assumed to be very large, one may replaoe
the Hankel function by its asymptotio approximation
Ho(l) ) 1'\1 (2/11' )1/2 ex:p (i - i 1Y /4), provided that I r/>7l.
The latter condition can be met by distorting the integration
path away from t= O. Upon changing variables to w II 81n-1(Vk1)'
and 1ntr"ducing the spherical coordinates r, 9 via I Z+zll = '
r oos 9, p = sin} , 0< O<.n/2, one obtains
where
(
ikl
{ sin w r sin w) e dw ,
? (A-3)
r
(
. ) sin
2
w - w
Sln w =
sin
2
w + ,cosw
(A-3a)
and we have defined )oCl = + cos w. (j f. - sin
2
w :; rt when w=O).
The inteeration path P in the VI-plane is identioal with P
Z
in the complex z-plane as shown in Fig.2 (except that the path
avoids the origin). Since the letter z denotes a spaoe coor-
dinate here, w has been chosen as the complex integration
variable. To effect an asymptotic evaluation for large values
fit -
of r, the path P is deformed into the steepest descent path
P through the saddle point at w = 9 as in Fig.2.
185
- 54 -

The singularities of the integrand in (A-3) are the
branch pOints at sin wb =! (t , and the poles wp at which
the denominator in (A-3a) vanishes. If It/ 1 and
arg f 1Y /2, appropriate to a highly dissipative medium, any
branoh cut integrals arising from the crossing of a branoh
point during the deformation of is into P can be negleoted
sinoe their oontribution is proportional to
il) icos(wb-9) 11) ;[cos8 {l- + sin f} If]
e = e ,
where the square roots have positive imaginary parts. Hence,
the branch cut integrals are exponentially small. The nearest
pole singularity is located at
-i'iT/4
w "" 1.+_e __
p'" 2 [iiJ
and will not be near the steepest descent path except when
a = V/2, i.e., when the souroe and observation points are
located on the interface. Therefore, the presence of the pole
oan be ignored in the asymptotic evaluation of (A-3) exoept
in the vidni ty of e = 11' /2
To obtain an asymptotio approximation of (A-3) valid
for observation angles in the interval 0 l.. e 1T' /2, it is
oonvenient to write r as
It
r (I) sin w) = -1 + r (I) sin w) ,
II. 2 {f -sin
2
w
r = . , (A-4)
I _sin
2
: Hoos w
186
- 55 -
L.B.Feleen
1\
wherein the numerator of r, unlike th8.t of r , is a slowly
varying funotion of w in the entire interval 0 <: w 11' /2 ,
w real. The oontribution to 1r from the (-1) term in (A-4)
s " It
is read11y shown to be equal to (K/41tfl 1') r) (see
(19) ). Hence,
ilL
J. i1l'/4
rr IV Ie M_e_
s 81ft
l
f e cos(w- 0)
--- F(w) dw ,
1I'i sinO w- w
p p
(A-5)
where
F(w) = {sin w
,.
r sin w)(w"w
p
) (A-5a)
"
Since (w-w ) r and i sin ware slowly varying in the inter-
p
val 0 (. w l' /2, w real, F(w) may be approximated (in a
first-order evaluation) by its value F( 9) at the saddle
point, and removed from the integral. The remaining integral
is identical with that in (28). Its asymptotio approximation
is given in (86), whenoe
where
'11'/4 w -9
b = {2 eJ. sin(..L-),
2
00
Q(y) = j
y
2
-x
e dx ,
(A-6)
(A-6a)
187
- 56-
L.B.'elsen
(2 e -111'/4 e -ill' /4
T(O) ;: +
() -w '2
p 2
(A-6b)
lhile (A-6) i8 Vf:lU,d for 1 . and for all 9 in the in-
terval 0 <: 6 l' /2 , 1 ts present form need be retained only
when i.e., 9 .. 11'/2; for other observation qles,
one may employ the simpler formula. (35). When r /2, the
seoond ter.m inside the bra.ces in (A-6) oan Ie negleoted in
oompariaon with the first term. In particular, one finds for
8=1'/2 (z=z'=o,

1t
1
" : 11 f+i4 fr" -f QHIf ,<-},1, I argI" r,
(A-7)
where f = It I 1s Sommerfeld I s numerioal distanoe. This
resul t agrees with that derived by Prof Bremmer ..If. by a. diffe-
rent method
See chapter by H.Bremmer, Seo. 1.0.
189
- 57 -
L.B.Felsen
REFERENCES
1. L.B.Felsen and N.Marcuvi tz, II Modal Analysis and Synthesis
of Electromagnetic Fields", Ndcrowave Research Institute,
Polytechnic Institute of Report R-776-59, PIB-705,
Oct. 1959.
2. For general information on the asymptotic eValuation of
integrals, see:
a) H.Jeffreys and B.Jeffrays, tt Methods of Mathematical
Physics", Cambridge University Pr3ss, 1946, Chapter 17.
b) Expansion of a Function De-
fined by 8. Definite Integral or a Contour Integral" ,
Admiralty Computing Service, London.(1946)
c) A.Erde1yi, "Asymptotic Expansions" , Dover Publishing
00., 1956, Chapter 2.
d) N. G. deBrui jn, " Asymptotic Methods in Analysis" , Inter-
science Pub. Co., New York, 1958, Chapters 4-6.
e) I.O.Friedrichs, "Speoial Topics in Analysis II , Notes
prepared by K.O.FrieUrichs and H.Kranzer, New York Uni-
versity, 1953-1954, Part B, pp. 1-67.
3. C.Chester, B.Friedman and F.Ursell, "An Extension of the
Method of Steepest Descents", Proc. of the Cambridge Phil.
Soc., Vol. 53, 1957, pp. 599-611.
4. B.L. Van der Waerden, "On the Method of Saddle Points",
Applied Soil Research, B2, pp. 33-45. (For a related method
of evaluating steepest descent integrals in the vicinity of
a pole, see: W.Pauli, Phys. Rev. Vol. 54, 1938; H.Ott,
Annalen d. Physik (Lpz), Vol. 43, 1943; P.C.Clemmow, J.Mech.
App. Math., Vol. 3,1950, and Proc. Roy. Soc. London, Sec. A,
Vol. 205, 1951).
190
- 58-
L.:B.relsen
1
5. P.C.Clemmow and C.M.Munford, "A Table of '2fi e J e dt
g
for Complex Values of f ", Phil. Transaotion of the Royal
Society of London, Vol. At 1952, pp. 189-211.
6. Ott E.Jahnke and F.Emde, "Tabels of Functions" , Dover
Publications, New York, 1945, p. 35-40.
7. F.Horner, "A Table of a Funotion Used in Radio-Propagation
Theory", Proc. of the Institution of El. Engineers, Vol.102,
Part 0, No. 1, 1955, pp. 134-131.
8. J.C.P.Miller, "The Airy Integral II , British Association
Mathematical Tables, Cambridge University Press, 1946.
9. A..SoIlDllerfe1d, "Partial Differential Equations in Physics ",
Academic Press, New York, 1949, p. 89,
10. lor a procedure utilizing Cauohy's theorem, see E.T. Copson,
" Theory of Functions of a Complex Variable \I , Oxford
University Press, London, 1935, Sec. 6.23.
11. F.Oberhettinger, "On a Modification of Watson's Lemma ",
J. Researoh Natl. Bureau of Standarda, July-Sept. 1959.