An Improved Uniform Approximation for Diffraction Integrals

14
An Improved Uniform Approximation for Diffraction Integrals Author(s): Ian Thompson Source: Proceedings: Mathematical, Physical and Engineering Sciences, Vol. 462, No. 2069 (May 8, 2006), pp. 1341-1353 Published by: The Royal Society Stable URL: http://www.jstor.org/stable/20208946 . Accessed: 22/06/2014 00:26 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. . The Royal Society is collaborating with JSTOR to digitize, preserve and extend access to Proceedings: Mathematical, Physical and Engineering Sciences. http://www.jstor.org This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AM All use subject to JSTOR Terms and Conditions

Transcript of An Improved Uniform Approximation for Diffraction Integrals

Page 1: An Improved Uniform Approximation for Diffraction Integrals

An Improved Uniform Approximation for Diffraction IntegralsAuthor(s): Ian ThompsonSource: Proceedings: Mathematical, Physical and Engineering Sciences, Vol. 462, No. 2069 (May8, 2006), pp. 1341-1353Published by: The Royal SocietyStable URL: http://www.jstor.org/stable/20208946 .

Accessed: 22/06/2014 00:26

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

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

The Royal Society is collaborating with JSTOR to digitize, preserve and extend access to Proceedings:Mathematical, Physical and Engineering Sciences.

http://www.jstor.org

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 2: An Improved Uniform Approximation for Diffraction Integrals

PROCEEDINGS -OF- A Proc. R. Soc. A (2006) 462, 1341-1353

THE ROYAL MX doi:10.1098/rspa.2005.1623 SOCIETY A?\. Published online 17 January 2006

An improved uniform approximation for diffraction integrals

By Ian Thompson*

Department of Mathematical Sciences, Loughborough University,

Loughborough, Leicestershire LEU 3TU, UK

A saddle point formula for integrals whose integrands possess a multi-valued exponent, and one or more poles is derived. The result includes a number of error (or Fresnel) functions, which is equal to the product of the number of poles and the number of sheets

possessed by the exponent's Riemann surface. Such formulae have previously been

suggested on the basis of known exact results and transformations, both of which are

specific to a particular exponent. In general, the multi-valuedness can be taken into

account by a straightforward modification to a standard procedure, which would

ordinarily yield only a single error function for each pole. In the context of diffraction

theory, this type of approximation remains valid in the vicinity of an optical boundary, even in some cases where a wavefield is incident in a direction almost or exactly parallel to a sharp obstacle (grazing incidence).

Keywords: integral asymptotics; saddle point; pole; uniform approximation; error function; diffraction

1. Introduction

The purpose of this article is the derivation of an improved saddle point formula for integrals whose integrands possess poles and a multi-valued exponent. For ease of presentation, and physical interpretation, we consider a class of such

integrals, members of which occur frequently in diffraction theory, for example in

the half plane problem illustrated in figure 1. Thus, let

0(r, 0; 0) = ?kr cos^ + -L

f JW-e-^Mda, (1.1) 2iriJcoc-oi0

where the location of the pole a = a0 is given by

a0 = ?kcos &, (1.2)

with k>0 and 0g[O,7t). The exponent % is defined as

X(a, 6) =

Y(a)|sin 0\ + \a cos 0, (1.3) with

y{a) =

{a2-k2)1'2, (1.4)

*[email protected]

Received 16 February 2005

Accepted 25 November 2005 1341 ? 2006 The Royal Society

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 3: An Improved Uniform Approximation for Diffraction Integrals

/. Thompson

? incident wave

-optical boundary

?? barrier

r *''

Figure 1. A wave incident at angle 0 upon a barrier occupying the half line 0 = 0. The arrows show

the direction of phase velocity, hence the optical boundaries are located at 6 = 0 and 6 = 2ir?0,

while the shadow and reflection regions occupy O<6<0 and 2ir ?

0<6<2ir, respectively.

so that 0 is a solution to the Helmholtz equation in polar coordinates (r, 6), i.e.

(V2 + k2)(j) = 0.

Integral solutions to other partial differential equations, which have different

exponents, can be treated in a similar manner. The contour C consists of the real

line, traversed from left to right, with indentations above a= ? k and a =

a0 and below a =

k, see figure 2, below, and the branch of the exponent is chosen so that

t(0) = ?ik. This corresponds to the situation in which 0 is derived from a time

dependent wavefield, with harmonic factor e~l?Jt. The integral in equation (1.1) generally represents the outgoing diffracted field generated when the plane wave

term strikes one or more sharp obstacles. We are concerned with the evaluation, in

the far field r^$> 1, of the contribution from the first order saddle point a = as, where

as = ?kcos 6, (1.5)

and, in particular the effect of the pole a = a0 upon this procedure. The function /is

taken to be free from essential singularities; this is generally the case for integrals arising in diffraction problems. In addition, for the purposes of the derivation we

shall assume that it has no poles. We generalize our result to deal with multiple

poles, and those outside the interval ( ?

k, k) at the end of ?2. If/should possess branch points other than a = + k, then these may contribute to the solution in some cases. We therefore choose to present our discussion in terms of the method of

steepest descents, rather than that of stationary phase, since if the path of

integration has to be diverted around a branch cut, one can often show that the

resulting contribution is exponentially small. In addition, certain standard

integrals appear automatically when the calculation is carried out using this

technique.

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 4: An Improved Uniform Approximation for Diffraction Integrals

An improved approximation for diffraction integrals 1343

(a) (b) integration path saddle point pole

3(a)

"$

3(i2)

*l

jm

Figure 2. (a) The original contour of integration C, and a typical steepest descent path S for a

diffraction integral. At the saddle point c? = o?s, the angle of intersection with the real line is 37r/4.

A residue must be included in cases where as<a0. (b) The corresponding situation in the r plane,

following indentation to the right of the point a = as.

One way to view the method of steepest descents, which is particularly effective in problems involving poles, is a contour deformation, followed by a

complex mapping, which transforms equation (1.1) into a real line integral to

which Watson's lemma may be applied (for large r), provided that the saddle is

not close to the pole. This results in a non-uniform approximation, whose

greatest merit is simplicity; it forms the basis of many results in ray theory

(Keller 1962), however, it becomes singular as the saddle approaches the pole.

Alternatively, following the transformation, the pole can be removed by the

addition and subtraction of a simple integral which can be evaluated exactly. The result is a uniform approximation, consisting of a scaled complex error

function (Abramowitz & Stegun 1965; henceforth, we shall just write error

function for brevity), and an integral which is approximated by Watson's lemma.

The former is not an approximation; it is a component of the exact value of 0, and satisfies the governing partial differential equation. It accounts for the rapid, but continuous, activation and deactivation of the residue term as the variation

in 6 causes the saddle to cross the pole a0. In place of the error function, many authors use the complex Fresnel function (Noble 1988), however the formulae

that follow are slightly more compact when given in terms of the former. The

relationship between the two is given in ?2. The non-uniform approximation can

be retrieved by replacing the error function with the first term in its asymptotic

expansion for large argument. The uniform approximation has been applied to

the solutions of diffraction problems involving Helmholtz equations by several

authors, including Rawlins (1975) and Abrahams & Wickham (1990). For a

general discussion of these methods, the reader is referred to Bleistein &;

Handelsman (1986) and Felsen & Marcuwitz (1994). As we shall see, even the

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 5: An Improved Uniform Approximation for Diffraction Integrals

1344 /. Thompson

uniform approximation breaks down in certain limits. The main result of this

article is the fact that the multi-valued nature of the exponent function x in

equation (1.1) causes the steepest descent mapping to generate as many poles as

the Riemann surface possesses sheets (two in this case). We derive an improved uniform asymptotic formula for (1.1) which consists of two error functions and, as usual, an integral to be approximated by Watson's lemma. In order to give a

physical interpretation to these remarks, we now consider the integral (1.1) in

the context of the diffraction problem shown in figure 1, in which a wave is

incident at angle 0 upon a barrier occupying the half line 0 = 0. This may

represent diffraction of acoustic waves by a half plane (Noble 1988), or of bending waves by a crack in a thin plate (Norris & Wang 1994) for example. The

resulting scattered field can be derived via the Wiener-Hopf technique and

expressed as an integral of the form (1.1) (additional terms, representing evanescent modes are also present in the latter case). Typically, the function /in

(1.1) takes different forms on either side of the half line 6 = tt, though the field is of course continuous here. The merger of the saddle and pole occurs at the optical boundaries of shadow and reflection shown in figure 1. This ought to be expected, as these divide the regions in which the incident and reflected fields are present from those where they are not. Typically, two separate uniform approximations

must be derived, one that includes an error function to account for the activation

and deactivation of the incident field, and one that does likewise for the reflected

wave. The former is applied for 6<tt, and the latter for d>ir. (In a problem with

a more complicated geometry, it may be necessary to consider several different

regions.) However, in the limits 0??0 and 0?>7r, which are known as 'grazing

incidence', the optical boundaries are close together, and in their vicinity, both

uniform approximations break down. It is reasonable to suppose that, in fact, both error functions are present in the solution for all 6. However, to directly establish this, we would need to assume differentiability to all orders, which is

certainly not apparent across 0 = tt given the integral form of 0 (equation (1.1)). Various alternative approaches exist for integrals satisfying the Helmholtz

equation. One such method is to transform the integral via the so-called 'complex

angle substitution'

a = ?kcos ?. (1.6)

This results in a steepest descents problem in which all functions are single valued,

though the exact evaluation of contour integrals in the ? plane is not

straightforward (Noble 1988). Another is to deduce the improved approximation

by comparison with the exact evaluations of the Sommerfeld half plane integrals, as in Borovikov Sz Kinber (1994). Finally, were we to make the assumption that the

field is indeed differentiable to all orders across 0 = 7T, then a ray method could be

employed, as in Lewis Sz Boersma (1969). For integrals satisfying other differential

equations, which have a different exponent function x, however, there may be

neither exact results in literature, nor a convenient mapping such as (1.6). The

present method can be applied in these cases, even when the exponent function

takes a complicated form, as it does in problems involving wave propagation and

diffraction in anisotropic media, see for example Norris & Achenbach (1983), Felsen & Marcuwitz (1994) and Thompson & Abrahams (2005). In addition, it

explicitly shows the precise means by which multiple error functions arise from the

asymptotics of diffraction integrals possessing only one pole.

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 6: An Improved Uniform Approximation for Diffraction Integrals

An improved approximation for diffraction integrals 1345

2. Derivation of the improved approximation

Return to equation (1.1), consider 6^(6,tc), and deform the contour C onto the

path of steepest descent S. This interval for 6 is sufficient to illustrate the

method; the approximation in the limiting cases 6=0 and 6 = ir will be

determined from our final results by continuity. Also, we must have

/(?o) = 1, (2.1)

here, since the residue from a = a:o eliminates the incident plane wave term in the

shadow region 6<0. The expression for 0 now becomes

J_ f

J(0d_e-rX(a,6)da + [l-Hf0-e\Wkr cos(@H?)

/^ 2m)sa-a0

where H(-) represents Heaviside's unit function. In the case where OLs =

a0 (i.e. 6 =

0), S is indented to the right, leaving the pole beneath the contour and hence

we take H(0) = 0. Given that

X(as,6)=-ik, (2.3)

it is not difficult to show that the steepest descent path is defined by the equation

a2 + k2cos26 + 2a(k + iu)cos 6 + 2iuk - u2 = 0, (2.4)

where u>0, and the different roots represent the two branches emanating from

the saddle point. The arcs at infinity yield no contributions, and we can use

equation (2.4) to show that the descent path always crosses the real line precisely

twice, at the saddle and also at the point a= ? ksec 6. The branch points a=+k,

therefore cause no difficulty. It should be emphasized that equation (2.2) is

continuous across the half line 6 = 0-, the effect of the contour crossing the pole

accounts for the presence of the Heaviside function. A typical steepest descent

path in the a plane is shown in figure 2. The angle at which the contour S crosses

the interval ( ?

k, k) turns out to be of some importance. To determine it, we first

note that

X"(a,e)=-k2Sme/j3(a), (2.5)

where the prime symbol denotes differentiation with respect to a; this convention

is maintained throughout. It is not difficult to see that this expression is always

positive imaginary at the saddle point, in fact

If we now write

a ?

as =

eelr?,

then in the limit a->as, we have

2 i(2r7+(7T/2))

note that the left-hand side of this expression is clearly positive real for a?5.

Given that C, and therefore S, moves from upper to lower half plane as ( ?

k, k) is

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 7: An Improved Uniform Approximation for Diffraction Integrals

1346 /. Thompson

crossed, it follows that the tangent to the descent path at the saddle point intersects the real line at a clockwise angle 37r/4 (see figure 2). Let us now

introduce the steepest descent mapping

i2 = x(M)-xK,0), (2.8)

with , x dt f(a)

g(t) = ^U_. (2.9) *v

yda a-a0 v J

In the t plane, the (first order) saddle is fixed at the origin, and the image of the

steepest descent path is clearly the real line, since the exponential in the integral is now simply e-7** . The location of poles in the t plane, the direction in which the

path of integration is to be traversed and the manner in which it should be

indented if necessary must all be carefully considered. The function g(t) cannot

usually be determined for general t, however its value (or perhaps residue) at

certain critical points, such as the origin, can be found relatively easily. The

location of poles in the t plane can be deduced from the following observations.

First, we note that the integrand in equation (1.1) exists not on a single complex

plane, but on a Riemann surface, which possesses at least two sheets, due to the

branch points in the exponent. If the function /also possesses branch points (as is

generally the case), then there are often multiple sheets. On each sheet of this

Riemann surface, there is a simple pole located at the point a = a0. Since dt/da is

finite, it follows from equation (2.9) that the function g(t) has poles at

corresponding points in the t plane. Those poles which lie on sheets with

y(0) = ?\k are mapped by equation (2.8) to the point t=to, which is defined up

to a factor +1 by

i =iJfc[l-cos(?-0)], (2.10) whereas those on sheets with y(0)=ik are taken to the point t=t\, with

t\ =iife[l-cos(0 + 0)]. (2.11) The next step is to determine the direction in which the real line is to be

traversed in the t plane, and any appropriate indentations that may be required. To achieve this, we first note that, except on indentations, i2ER+ for a S.

Second, since we must take a square root to obtain t, we avoid the point ^ = 0.

Thus, in the a plane, S is indented to the right of the saddle point in a semi

circular arc with radius e. This is consistent with the indentation used previously in the special case 0 =

0; it now applies for all 0. The radius e must be sufficiently small that the orientation of S with respect to the pole a =

a0 is not changed. On

the arc, we may apply equation (2.7), with 7] decreasing from an initial value of

37t/4 to ? (7r/4) (see figure 2a), to show that arg(?2) varies from 27T to 0 as the

indentation is traversed. It now follows that the image of S in the t2 plane traverses the lower side of IR+ from right to left, encircles the origin, and finally traverses the upper side from left to right, as shown in figure 2. If we now take

the square root t= l^e1 arg^

^2, we find that the real line is traversed from left to

right in the t plane, and that the pole t=ti lies on the line arg(?) =

7r/4, i.e. it is

always above the path of integration, hence

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 8: An Improved Uniform Approximation for Diffraction Integrals

An improved approximation for diffraction integrals 1347

As usual, the surd symbol yf7 is used only when taking the positive root of a

positive, real quantity. The situation with regard to to is slightly more

complicated, since for 6 = 0, we have ?o

= 0, i.e. as 6 varies, this point may

'jump' from one sheet of the square root's Riemann surface to the other.

Consequently, we cannot appeal to any complex analysis argument to determine

its position, however this is unnecessary?continuity of the resulting expression in 6 will suffice. We now let the indentation radius e?>0 in all cases except 6 = 0.

Clearly, to lies on one side of the contour for 0>6 and on the opposite side

otherwise; only this behaviour can maintain continuity given the presence of the

Heaviside function in equation (2.2). The real line has an indentation above

the origin in the critical case 6 = 0, therefore we may conclude that to lies on the

upper side of the contour for 0>6, and on use of equation (2.10) we obtain

to =

V2kei{ir/A)sm 0-6

Writing

9(t) = h(t) (t-k)(t-txy

(2.12)

and separating into partial fractions, we find that 0 is now reduced to the form

0 =

0! +02 +03> (2.13)

where

<t>i = Kh) Afar

?

~-rt'

%-ti 27ri irt-tQ dt + [l-H(0-6)]

Kh) h-h\

J.kr cos(&?6) (2.14)

02 =

and

Kk) ) eikr r e^

l 27T? J-oo t ?

ti dt,

h =

tQ ?

tx

eikr ,o

2^? J e 2m

h(t)-h(tQ) /i(?)-/i(?x) t-to t-tx

e^di.

(2.15)

(2.16)

In the first instance, the path of integration T passes below the pole in all cases, and a residue has been included where appropriate. Since there is now no

possibility of a pole crossing an integration path, the critical value h(to) follows

immediately; we must have

Kh) =

<o ?*i?

in order to maintain continuity in 6. This result can also be obtained by using

equation (2.12) in (2.9) and rearranging to yield

h(t) (dt/da)(a-a0) t-U t-to

= /(?);

the limit t-*to, which implies a??ao may now be taken without difficulty. The

procedure described above cannot easily be used to determine h(ti), since in this case the function /must be evaluated on a sheet which does not contain S. If/has

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 9: An Improved Uniform Approximation for Diffraction Integrals

1348 /. Thompson

branch points (as is usually the case), then it is not always obvious which sheet should be chosen. Thus, at this stage, we simply write

h(tx) =?c(?o ?*i),

where c is a constant to be determined. In fact, we shall see later that c is the reflection coefficient for the barrier in question. Finally, we determine the value of /i(0); thus we differentiate equation (2.8) twice with respect to a, to obtain

X"(M) =

2(i') + 2tt". (2.17)

Next, on substitution of equation (2.12) into (2.9) and evaluation at the point ?=0, which corresponds to a =

as we find that

h(0) = /fo)V2< >?(t/4)

x/?FKJ? (2.18)

where we have used the identity as ?

a0 =

i?o?i. The argument is determined from

equation (2.17) by noting the direction of variation of a and t as the paths of

integration pass through the saddle in the respective complex planes. As t passes from left to right through the origin a travels down a path whose tangent intersects the real line at a clockwise angle 37r/4 (see figure 2 a), and hence

arg[f'(as, 0)] =

(7r/4). It now remains to simply collect together all of our results, evaluate the standard integrals (2.14) and (2.15) and take the leading order behaviour of (2.16) via Watson's lemma. It should be emphasized that this is the

only approximation applied at any stage of the entire procedure. After some

manipulation, we find that

eikr ^e'^sin

e-e'

(2.19)

02 c?w '2kr? 2

and

03 ei(?r/4)eiAr

2V2irkr

X

0 + 0 0-0 csc-csc

?-?

sin Of (?k cos 0) ? c sin ?-sin + 0[(kr)-3/2}.

(2.20)

(2.21)

2 2

Here, we have made use of the standard integral (Abramowitz & Stegun 1965)

1 f?? e~z* (*)=- -dz' [3(*&)>?]> 17T J^o Z

? Zq

in fact this is valid for all Zq provided that the path of integration is adjusted so as

to pass below the pole. The notation w(-) represents the scaled complex error

function, i.e. _ 2

w(z) = e z

erfc(?iz);

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 10: An Improved Uniform Approximation for Diffraction Integrals

An improved approximation for diffraction integrals 1349

this is related to the aforementioned complex Fresnel function F( ) via

F(z)=?^^w(?^z). As noted above, the terms 0X and 02 are not approximations; they are exact solutions to the Helmholtz equation with the special property that they may 'contain' a plane wave. To see this, we need the two results (Abramowitz & Stegun

1965) w(z) =2e~z2-w(-z) (2.22)

and

w(z)-1-7=, |z|-*oo, -^<arg(*)<-^. (2.23) zy/?'

Thus, 0i includes the incident field for 6 > 0, however the plane wave included in 02 cannot be activated for 6 (0, it). This is as we should expect; there are no such terms

other than the incident field in this region. Indeed, the only other plane wave in the entire problem is the reflected field, which is present for 6>2ir?0, precisely the

region in which the plane wave in 02 is activated. In fact, from equation (2.22), we can

write

j> = celbr

cos(<9+6>)

Afar

W -V2^ei(7r/4) sin 0 + 6'

and immediately we recognize the plane wave term as the reflected field. The constant c is thereby determined as the reflection coefficient for the barrier occupying 6=0.

More rigourously, c can be obtained by repeating the entire procedure for 6 G (it, 2tt) and imposing continuity as 0?>it. The two simpler approximations can be retrieved

by using equations (2.22) and (2.23) in (2.13). In particular, the non-uniform

approximation, with residues omitted, is given by the first term in equation (2.21), i.e.

ei(7T/4)eifcr @ + e @__e ~ ?

csc ?-?

csc-sin 6f (?k cos 6). (2.24)

The effect of the additional terms is to regularize this expression at the two poles 6 = 0 and 6=2n?0, and to include the plane waves where appropriate. If we now consider the integral (1.1) and the approximation given by (2.13) with (2.19)-(2.21) as functions of the parameter 0, an immediate generalization to deal with integrals possessing poles outside the interval (

? 1,1) becomes available. Thus, for a simple

pole located at a = ap, we write ap= ?A:cos 0P, where 0< 9t(@p)

< it. Addition of the correction terms 0+ and 0?., where

Akr

4>? =

A??lw ?V2k?ei^ sin^?i ?(t/4)

y/l-nhr csc 0p?6

(2.25)

to (2.24) now yields the improved formula. The coefficients A+ and A- can be determined by comparison of residues?the resulting approximation must be regular at 6 =

0p and 6 = 2tt?0p. In both cases, we can determine whether to take the upper

or lower sign by ensuring that the first term in (2.22) is activated in the correct region

(cf. equation (2.23)). Thus, we have reached the result suggested by Borovikov & Kinber (1994), via a method which does not depend upon comparisons to other diffraction integrals whose values are known a priori, and can therefore be applied in

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 11: An Improved Uniform Approximation for Diffraction Integrals

1350 /. Thompson

cases, where the governing equation is not that of Helmholtz. The procedure of

correcting the diffraction coefficient can simply be repeated to account for integrals whose integrands possess multiple poles. For 0p^ [0, it], equation (2.22) yields plane waves as above, whereas if

9?(0p) =?0, we obtain evanescent modes (these may

propagate without loss in some special direction, such as along a barrier, and can

therefore play an important role in the far field behaviour). Before examining a more complicated problem, it is useful to briefly

investigate the properties of the approximation given by equations (2.13) and

(2.19)-(2.21) by applying it to the classical problem of diffraction by a half plane. The geometry is identical to that shown in figure 1, and the boundary condition

requires that 90/90 = 0 for 0 = 0 and 0 = 2ir. The solution to this problem in the

form (1.1) is (Noble 1988)

(a + kfl?

where the fractional power in the denominator is defined so that (a+ k)1'2 =

aA* + k for a> ?

k. The reflection coefficient c is easily shown to be unity, and on

recalling that equation (2.13) is derived for 0 (O,tt), the third term now vanishes identically. We are left with

ihr

0 W /2^ei(7r/4) sin 0-0

+ w f2kr?{nl^ sin 0 + 0'

(2.27)

which clearly does not break down in any limits?it is the exact solution to the

problem in question, originally obtained by Sommerfeld (1896) (for an English reference, see Sommerfeld (1954)). Thus, in this case nothing is actually gained;

we have simply retrieved an old and widely known result. It is worth noting,

however, that had we taken h(ti) =

0, we would obtain a formula in place of

equation (2.27) which is singular in the limit 0 + 0?>0. Another point of interest

is that the improved formula continues to give the correct result for all 0 if we

multiply (2.26) by csc 0/2 and then set 0 = 0, thereby merging the pole with the branch point of the function /

3. An example from thin plate theory

Finally, we consider the situation in which </> represents the transverse

displacement of an infinite thin plate, with a rigid strip along the half line

0 = 0. Then </> satisfies the partial differential equation (Timoshenko 1940)

(V2 + k2)(V2-k2)<j) = 0,

with boundary conditions 0= (d(/)/d0) = O for 0 = 0 and 0 = 2k. The solution to

this problem, originally derived by Norris & Wang (1994), is

0 Jkr cos(0?&) +

2tt?

Yo ?o (Ae~

t ? K+ L(e^T|sin

y|sin 0\_rvo~rMs^n 0\ \

e^A|8in?|)sgn(sin?)

ye

iar cos 6

a-a0 -da, (3.1)

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 12: An Improved Uniform Approximation for Diffraction Integrals

An improved approximation for diffraction integrals 1351

where we have introduced the conventions that a function shown without

argument depends upon a alone, and a subscript '0' denotes evaluation at the

point a = a0. It should be emphasized that our notation is rather different from

that used by Norris &; Wang (1994). Here, y is defined by equation (1.4), and the second exponent by

X(a) =

(a2 + k2)1/2, (3.2)

with X(a) = Va2 + k2 for a G IR. The contour C is the same as before; it passes

between the branch points a=?ik. A superscript ' +

' ('

? ') denotes a function

that is analytic on and above (below) C; these are obtained via an analytic

product factorization (Noble 1988) so that K+ K~ = K, etc. The function K(a) is

given by

y(a)-X(a)

and K+ can be expressed in terms of finite, non-singular contour integrals

(Norris Sz Wang 1994). Thompson Sz Abrahams (2005) apply a procedure that

requires only one such integral to a similar problem. The remaining factorizations can be performed by inspection, thus

y?(a)=eT^/A)(a?kf2,

and

X?(a)=eTl^\a?ik)1^2',

where the branches are chosen so that y^(0) =

Vke~l(<7r^\ and ̂ (0) = Vk. It is

important, for the purposes of our subsequent analysis, to briefly discuss the

possibility of evanescent modes in equation (3.1). First, consider terms with an

exponent involving X. A straightforward calculation shows that these have a

saddle point located at a= ? ikcos?, and that the associated steepest descent

path is defined by a =

?i(k + u)cos 0 + sin 6yu2 + 2u,

where u>0. Clearly, this is the real line if cos 6 = 0; it lies in the upper or lower

half plane for cos 6<0 and cos 6>0, respectively. In the latter case, a diversion

around a= ? k is necessary, note that the terms in question have no branch point

at a = k. All contributions are exponentially small, except in cases, where 6 ? 0

and 6~2it, i.e. in the vicinity of the rigid strip. Next, consider the terms with an

exponent involving y. These have a branch point at a= ? ik, but not at a = ik.

We can apply the method of steepest descents as before; equation (2.4) shows

that the deformed integration path crosses the imaginary axis at the point a = ?ik cos 0|csc 6\. Therefore, it must be diverted if cos 6> |sin 6\, or else it will

pass below a= ? ik. Since y is pure imaginary for a =

(l + u)ik, u>0, the largest contribution due to the diversion comes from the branch point itself, and it

decays exponentially, at least as rapidly as e_?T/ . To summarize, there are

evanescent modes present for all 6, though these do not contribute significantly to the far field in general. There are also modes that propagate along the faces of

the rigid strip 6=0 and 6 = 2k\ these are evanescent in other directions. If we now

treat equation (3.1) as an integral of the class (1.1) (i.e. disregard terms with

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 13: An Improved Uniform Approximation for Diffraction Integrals

1352 /. Thompson

exponent A), with

/(a) = i^T^ [Y0sgn(sin 0) -y?^Vrl, (3.3) we should expect a good approximation, except possibly in the immediate

vicinity of the rigid strip. The reflection coefficient may be obtained by taking the

residue in (1.1) at a = a0, on the lit side of the strip; we obtain

c=^>. (3.4)

Introducing a subscript 's' to refer to evaluation at the saddle point, we find that, for 0e(O, it),

sin Of (a.) = i K0~K+y^yt [To+Ys~ - t?] ; (3.5)

it is generally simplest to perform manipulations in terms of y0, As, etc., where

possible. Equations (2.13), (3.4) and (3.5) define an approximation that is valid for all 0; the same result occurs if we repeat the entire procedure considering instead 0^(tt, 2tt). For 0 = 0 and 0 =

2k, equation (3.5) vanishes identically, due

to the factor y^. Interestingly, this means that, if 0 is sufficiently large to permit the use of (2.23), then (2.13) accurately predicts the displacement at the strip

(zero). In addition, if 0 = 0 (head-on incidence), then c= ?1, and equation (2.13) satisfies the boundary condition exactly. Thus, the omitted modes, all of which

are evanescent except possibly along the axis of the strip, make a significant contribution to the displacement here for small, non-zero 0 only. Finally, it is of some interest to assess the accuracy of equation (2.13) for 0 =

tt, since the uniform

approximation is known to break down in the vicinity of this half line as 0 ? 7T.

Thus, return to (3.1), set 0 = ir and deform the path of integration to pass along the faces of the branch cuts in the upper half plane. Ignoring exponentially small

contributions, we find that

0~01+02> (3-6) wherein

i*" Jk 7R (a-ao)

and K JL>v- ffc+ioo piar

j

02 = e

P+1?? piar dry

0i =

_1^^2L [rt + _ ?+]

e da

Jk JR {o?-Oq)

v- rk+ioo Jar j

Tri )k yR a-a0

Here, the first integral has no pole at the point a = a0, and the subscript 'R' has

been introduced to denote the fact that y~ is to be evaluated on the right face of

the branch cut. Following an application of Watson's lemma in equation (3.7), the substitution t =

?iy^(a)y/r reduces both integrals to standard forms, which

evaluate to give

^ei(7r/4)cos^

i ?\kr cos & i ikr J jy -\ ?,, (p ~e + e < KqAqW

i(7r/4) i^ "| + ~==sec-K0-AO[K+(k)A+(k)-K0+X?] + 0(r^2) L

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions

Page 14: An Improved Uniform Approximation for Diffraction Integrals

An improved approximation for diffraction integrals 1353

Setting 6 = tt in equation (2.13), and making use of the identities (2.2) and c?l =

2X0Kq, we can derive precisely the same expression for 0, showing that the

improved approximation is valid here for all angles of incidence.

4. Concluding remarks

We have demonstrated the derivation of an improved saddle point formula for

integrals possessing poles and a multi-valued exponent. For ease of presentation, the procedure is performed upon an integral which satisfies the Helmholtz

equation; it can easily be adapted to other cases. The resulting formula includes a

number of error functions, which is equal to the number of poles multiplied by the number of sheets possessed by the exponent's Riemann surface. The method

is easy to apply, regardless of the number and location of the poles, and the range of validity often extends beyond the interval for which it was derived.

References

Abrahams, I. D. & Wickham, G. R. 1990 The scattering of sound by two semi-infinite, parallel

staggered plates II: evaluation of the velocity potential for an incident plane wave and an

incident duct mode. Proc. R. Soc. A 427, 139-171.

Abramowitz, M. & Stegun, I. 1965 Handbook of mathematical functions. New York: Dover.

Bleistein, N. & Handelsman, R. A. 1986 Asymptotic expansions of integrals. New York: Dover.

Borovikov, V. A. & Kinber, B. Ye. 1994 Geometrical theory of diffraction. London: Institute of Electrical Engineers.

Felsen, L. B. & Marcuwitz, N. 1994 Radiation and scattering of waves. New York: Institute of

Electrical and Electronics Engineers.

Keller, J. B. 1962 Geometrical theory of diffraction. J. Opt. Soc. Am. 52, 116-130.

Lewis, R. M. & Boersma, J. 1969 Uniform asymptotic theory of edge diffraction. J. Math. Phys. 10,

2291-2305. (doi:10.1063/l. 1664835) Noble, B. N. 1988 Methods based on the Wiener-Hopf technique. New York: Chelsea.

Norris, A. N. & Achenbach, J. D. 1983 Elastic wave diffraction by a semi-infinite crack in a

transversely isotropic material. Q. J. Mech. Appl. Math. 37, 565-580.

Norris, A. N. & Wang, Z. 1994 Bending-wave diffraction from strips and cracks on thin plates. Q.

J. Mech. Appl. Math. 47, 607-627.

Rawlins, A. D. 1975 The solution of a mixed boundary value problem in the theory of diffraction by a semi-infinite plane. Proc. R. Soc. A 346, 469-484.

Sommerfeld, A. J. W. 1896 Mathematische th?orie der diffraction. Math. Ann. 47, 317-374. (doi:10.

1007/BF01447273) Sommerfeld, A. J. W. 1954 Optics. New York: Academic Press.

Thompson, I. & Abrahams, I. D. 2005 Diffraction of flexural waves by cracks in orthotropic thin

elastic plates. I: formal solution. Proc. R. Soc. A 461, 3413-3436. (doi:10.1098/rspa.2004.1418)

Timoshenko, S. 1940 Theory of plates and shells. New York: McGraw-Hill.

Proc. R. Soc. A (2006)

This content downloaded from 188.72.126.55 on Sun, 22 Jun 2014 00:26:38 AMAll use subject to JSTOR Terms and Conditions