On the derivation of boundary integral equations for scattering by an infinite one-dimensional rough surface J. A. DeSanto and P. A. Martina) Department of Mathematical and Computer Sciences, Colorado School of Mines, Golden, Colorado 80401-1887 ~Received 24 June 1996; revised 10 December 1996; accepted 29 January 1997! A crucial ingredient in the formulation of boundary-value problems for acoustic scattering of time-harmonic waves is the radiation condition. This is well understood when the scatterer is a bounded obstacle. For plane-wave scattering by an infinite, rough, impenetrable surface S, the physics of the problem suggests that all scattered waves must travel away from ~or along! the surface. This condition is used, together with Green’s theorem and the free-space Green’s function, to derive boundary integral equations over S. This requires careful consideration of certain integrals over a large semicircle of radius r; it is known that these integrals vanish as r→` if the scattered field satisfies the Sommerfeld radiation condition, but that is not the case here—reflected plane waves must be present. The integral equations obtained are Helmholtz integral equations; they must be modified for grazing incident waves. As such integral equations are often claimed to be exact, and are often used to generate benchmark numerical solutions, it seems worthwhile to establish their validity or otherwise. © 1997 Acoustical Society of America. @S0001-4966~97!01406-9# PACS numbers: 43.20.Fn, 43.30.Hw @ANN# INTRODUCTION Consider the scattering of a plane time-harmonic acoustic wave by a bounded obstacle. To fix ideas here, we consider a two-dimensional obstacle, with a smooth, sound-hard surface S. Mathematically, this is an exterior Neu mann problem for the Helmholtz equation. In order to have a wellposed problem, we impose the Sommerfeld radiation condition, Ar S D ]u 2iku →0 ]r ~1! as r→`, uniformly in all directions. Here, u is the scattered field, r is a plane polar coordinate, k is the wave number, and we have assumed a time dependence of e 2i v t . Physically, the radiation condition ensures that the scattered waves propagate outwards, away from the obstacle. Mathematically, the radiation condition also yields uniqueness and existence for the boundary-value problem. A familiar method for solving the above problem is to derive a boundary integral equation for the boundary values of u on S. In the derivation, Green’s theorem is applied to u and a fundamental solution G, in the region bounded internally by S and externally by C r , a large circle of radius r. It turns out that the radiation condition implies that the integral I ~ u;C r ! [ ES Cr u D ]G ]u 2G ds→0 ]r ]r as r→`, ~2! and so only boundary integrals over S remain. Thus, the radiation condition is a crucial ingredient for two results: a well-posed boundary-value problem; and the vanishing of a a! Permanent address: Department of Mathematics, University of Manchester, Manchester M13 9PL, United Kingdom. 67 J. Acoust. Soc. Am. 102 (1), July 1997 standard integral over a large circle. See Colton and Kress ~1983! for more information. For a sound-hard surface S, the procedure described above leads to the following boundary integral equation: u~ p !2 E u~ q ! S ]G ~ p,q ! ds q 5 ]nq E ]] S u inc G ~ p,q ! ds q , nq ~3! pPS, here, u inc is the given incident wave. One can also derive an equation for the boundary values of the total field u tot5u inc1u; this boundary integral equation is u tot~ p ! 2 E S u tot~ q ! ]G ~ p,q ! ds q 52u inc~ p ! , ]nq pPS. ~4! We shall refer to ~3! and ~4! as standard Helmholtz integral equations. Similar equations can be derived for sound-soft surfaces ~exterior Dirichlet problem, u tot50 on S). Now, it is known that the waves scattered by a bounded two-dimensional obstacle have the form u ~ r, u ! 5 e ikr Ar f ~ u ! 1O ~ r 23/2! ~5! as r→`, where (r, u ) are plane polar coordinates, and f is called the far-field pattern. Thus, apart from being outgoing (e ikr ), the waves decay with distance from the obstacle (1/Ar). Indeed, the radiation condition implies that u5O(r 21/2) as r→`. As G also satisfies the radiation condition, we find that the integrand in ~2! is u S D S D ]G ]u 2ikG 2G 2iku 5o ~ r 21 ! ]r ]r 0001-4966/97/102(1)/67/11/$10.00 as r→`, © 1997 Acoustical Society of America 67 whence I(u;C )→0 as r→`, where C is any piece of C r . This shows that the result ~2! is due, essentially, to the decay of u ~and G), not to any cancellation effects induced by the integration. The description given above changes completely when the obstacle is unbounded. For example, suppose that S is an infinite, flat plane. Then, an incident plane wave will be scattered ~reflected! specularly as a single propagating plane wave. More generally, suppose that S is an infinite corrugated surface; then, an incident plane wave will be scattered into a spectrum of plane waves. The specification of a ‘‘radiation condition’’ for such problems continues to attract attention ~see, for example, Ramm, 1986!; clearly, the Sommerfeld radiation condition is not appropriate, as it is not satisfied by a propagating plane wave. Nevertheless, it is customary to proceed, assuming that the scattered field can be represented in terms of plane waves, at least at some distance from S. Typically, this requires the discarding of an integral such as ~2!, but with the large circle C r replaced by a large semicircle H r . Can this step be justified? This paper began as an attempt to do this. Another possible approach is to assume that u can be written as a surface distribution of sources or dipoles; see Sec. VI E. One might also invoke the limiting absorption principle, in which the wave number k is replaced by k1i«, where « is small and positive; the corresponding u is required to decay as r→`. However, this is delicate ~compared to scattering by a bounded obstacle! as the limits «→0 and z→` for exp $iz(k1i«)% do not commute. The motivation behind the present work is the pervasive view that solving a boundary integral equation gives a rigorous, exact way ~apart from numerical errors! of solving problems involving the scattering of plane waves by infinite rough surfaces. Indeed, one can find many books and papers setting out this view. ~References to the literature will be given later.! However, very little attention has been given to the derivation of the boundary integral equations themselves, most writers being content to start by writing down a standard Helmholtz integral equation, ~3! or ~4!. We will show that ~3! is valid for plane-wave scattering by an infinite, onedimensional, rough surface. We will also show that ~4! is valid, except for grazing incident waves ~in which case the right-hand side should be replaced by u inc). The paper is organized as follows. Section I is devoted to formulating the problem, with some discussion on radiation conditions and some background on angular-spectrum representations and integral representations ~using G). Estimation of integrals over the semicircle H r is carried out in Secs. II, III, and IV. Thus, the method of stationary phase and an expansion method are used in Secs. II and III, respectively, but only for a single plane wave. Results for I(u;H r ) are obtained in Sec. IV, and are then used in Sec. V to derive various boundary integral equations of Helmholtz type. Extensive discussion of the results is given in Sec. VI. For example, it is shown that the standard Helmholtz integral equations are valid for a finite patch of roughness and for finite incident beams. 68 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 I. FORMULATION Consider the scattering of a plane wave by an infinite rough surface, S. In this paper, we assume that the surface is one-dimensional, so that it can be described by z5s ~ x ! , 2`,x,` with 2h,s(x)<0 and some constant h>0. The acoustic medium occupies z.s and, for definiteness, we assume that S is a smooth, sound-hard surface. Thus we can write the total field as u tot5u inc1u, where u is the scattered field and u inc~ r, u ! 5e 2ikr cos~ u 1 u i ! , u u i u < 21 p , ~6! is the incident plane wave; u i is the angle of incidence ~it is the angle between the direction of propagation and the negative z axis!, and (r, u ) are plane polar coordinates: x5r sin u and z5r cos u. All the fields u tot , u inc , and u satisfy the Helmholtz equation ~ ¹ 2 1k 2 ! u50, ~7! for z.s. The boundary condition is ] u tot 50 ]n ~8! on S, where ] / ] n denotes normal differentiation out of the acoustic medium. A. Reflection by a flat surface If S is flat (s50), the problem is trivial. Nevertheless, this problem can still teach us something. It is well known that the scattered field is u ~ r, u ! 5e ikr cos~ u 2 u i ! for u u i u , 21 p . ~9! When u u i u 5 21p ~‘‘grazing incidence’’!, we have u[0: The incident wave satisfies the boundary condition on S. Thus, for u u i u , 21p , u tot52 e ikx sin u i cos~ kz cos u i ! and 2 e ikx sin u i cos~ kzcos u i ! 1A 1 e ikx 1A 2 e 2ikx both ‘‘solve’’ the problem, where A 1 and A 2 are arbitrary constants. Of course, we disallow this second solution, unless A 1 5A 2 50: but why? The answer is because of the radiation condition ~which we have yet to specify!. Physically, we want to exclude all ‘‘incoming’’ plane waves, apart from the incident wave. We will be more precise in Sec. I B. B. Angular-spectrum representations For any S, the scattered field above the corrugations, z.0, may be written using an angular-spectrum representation, J. A. DeSanto and P. A. Martin: Rough surface integral equations 68 u ~ x,z ! 5 5 E E ` F ~ m ! e ik ~ m x1mz ! 2` p /2 dm m~ m ! ~10! A ~ a ! e ikrcos~ u 2 a ! d a 2 p /2 ~11! 1 evanescent terms. Here, F( m ) is the spectral amplitude, A( a )5F(sin a! and m~ m !5 H A12 m 2 , i Am 2 21, u m u ,1, 1 H6 r 5 $ ~ r, u ! :0<6 u < 2 p % u m u .1. The integrals are superpositions of plane waves; they are propagating, homogeneous plane waves when u m u ,1 and they are evanescent, inhomogeneous plane waves when u m u .1 In ~11!, we see the propagating plane waves explicitly: They propagate at an angle a to the positive z axis, with an ~unknown! complex amplitude, A( a ); the ‘‘evanescent terms’’ decay exponentially with z. For more information on angular-spectrum representations, see Clemmow ~1966! and DeSanto and Martin ~1996!. In general, the spectral amplitude must be considered as a generalized function, and not as a continuous or analytic function. This simple observation is motivated by known results for particular surfaces. Thus, for a flat surface we have u u i u , 21 p , F ~ m ! 5 d ~ m 2sin u i ! cos u i , where d is the Dirac delta function, whereas for a periodic surface F is a discrete sum of delta functions. So, we split the scattered field into three parts as ~12! u5u pr1u ev1u con , where ( n51 being quarter-circles, we require that in the region x>0, z.0, we use 0< a n < 21p , so that all plane waves propagate out through H 1 r . Similarly, in the region x<0, z.0, we use 2 21p < a n <0, so that all plane waves propagate out through H2 r . This form of the radiation condition will be used to derive boundary integral equations. C. Boundary integral equations One way to determine the scattered field is to derive a boundary integral equation over the rough surface S. The appropriate fundamental solution is G ~ P,Q ! 5G ~ y,x! 5 2 21 iH ~01 ! ~ k u x2yu ! , where x and y are the position vectors of Q and P, respecis a Hankel tively, with respect to the origin O, and H (1) n function. Apply Green’s theorem to u and G in the region D r with boundary ] D r 5H r øS r øT r , where H r is a large semicircle of radius r and center O, S r 5 $ ~ x,z ! :z5s ~ x ! ,2r,x,r % A n v~ r, u ; a n ! , M u ev~ r, u ! 5 ( B m w ~ r, u ; m m ! , E C ~ m ! e ikr ~ m sin u 1m cos u ! m51 u con~ r, u ! 5 ` 2` v~ r, u ; a ! 5e ikr cos~ u 2 a ! , ~13! dm , m~ m ! with u a u < 21 p , ~14! ~15! 2u ~ P ! 5 Am 2 21 The first term in ~12! is a sum of propagating plane waves; the coefficients A n and the angles a n are unknown in general. The second term in ~12! is a sum of evanescent waves; the coefficients B m and m m are unknown in general. The third term in ~12! is a continuous spectrum of plane waves; the unknown function C is continuous. Properties and consequences of the general representation ~12! were investigated by DeSanto and Martin ~1996!. Let us now return to the radiation condition. Having chosen an origin O, arbitrarily, we consider a large semicircle H r , with radius r and center at O. We then require J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 u~ q ! ]Dr J ]G ]u G ~ P,q ! ds q , ~ P,q ! 2 ]nq ]nq 2u ~ P ! 5 EH u~ q ! Sr J ]G ] u inc G ~ P,q ! ds q ~ P,q ! 1 ]nq ]nq ~18! 1I ~ u;H r ! 1I ~ u;T r ! , with u m u .1. ~16! 69 E H where PPD r , qP] D r and ] / ] n q denotes normal differentiation at q. Use of the boundary condition ~8! yields and w ~ r, u ; m ! 5e ikr m sin u e 2kr cos u ~17! is a truncated rough surface, and T r consists of two line segments joining the ends of H r and S r . Then, as both u and G satisfy the Helmholtz equation ~7! in D r @apart from the singularity in G( P,Q) at P5Q], we obtain N u pr~ r, u ! 5 that all propagating plane-wave components v (r, u ; a n ) in u propagate outwards through H r , away from O. This is almost built into the decomposition ~12!: we have to be careful with grazing waves ( u a n u 5 21p ). For example, if a n 5 21p , v 5e ikx ; this wave leaves the semicircle at u 5 21p but enters at u 52 21p . A simple way to impose our radiation condition without excluding grazing waves is to split the half-space z.0 and the semicircle H r into two parts. Thus with where I ~ u;S ! 5 EH S u~ q ! J ]G ]u G ~ P,q ! ds q ~ P,q ! 2 ]nq ]nq and normal differentiation is taken in a direction away from the origin @so that ] / ] n5 ] / ] r on H r , consistent with ~2!#. The scattered field u and its derivative ] u/ ] x are bounded in the neighborhood of S. This assumption together with simple bounds and the large-argument asymptotic behavior of Hankel functions show that I(u;T r )5O(r 21/2) as r→`, whence I ~ u;T r ! →0 as r→`. J. A. DeSanto and P. A. Martin: Rough surface integral equations ~19! 69 Before estimating I(u;H r ) for large r, using ~12!, we consider a single propagating plane-wave component in ~12!. Thus, we shall evaluate I( v ;H r ) as r→`, where v is defined by ~15!. Indeed, we shall evaluate the limit using two different methods: the method of stationary phase ~Sec. II! and an expansion method ~Sec. III!. We shall discuss the evaluation of I(u;H r ) itself for large r in Sec. IV. Boundary integral equations will then be derived from ~18! in Sec. V. We use the method of stationary phase to estimate I( v ;H r ). There are two cases, depending on the value of a , which can be smoothed together using a uniform approximation. A. The method of stationary phase: z a z < 21p We are interested in large values of r5 u xu for fixed values of y and k. We have 5 exp$ ik ~ r2y•x̂! % Akr Akr e e 1O ~ l 21 ! as l→`, where F5l sin a2kr sin w. Hence, from ~21!, we obtain as kr→`, for u a u , 21 p . ~23! B. The method of stationary phase: z a z 5 21p Suppose that a 5 21p . In this case, F( u )5sin u is stationary at u 56 21p , which are end points of the range of integration. We have g ~ 21 p ! 50 and g ~ 2 21 p ! 52 e ik r sin w . I ~ v ;H r ! 5e ik r sin w 1O„~ kr ! 21/2… ~20! Hence, for large r, B ]G ]v 2G .ik @ 12 cos~ u 2 a !# ]r ]r Akr 3e as l→`, ~24! as kr→`, for a 5 21 p . ~25! When a 52 21p , we obtain the same result except that w is replaced by 2 w . In this case, the relevant stationary-phase point is at u 5 21p . When r 50, we can give an independent check of the result ~24!. In this case, we have L ~ l ! 5 Al ikr ~ 11 cos~ u 2 a !! 2ik r cos~ u 2 w ! E p /2 ~ 12sinu ! e il sin u d u 2 p /2 e 5 p Al $ J 0 ~ l ! 2iJ 1 ~ l ! % , and then I ~ v ;H r ! .iBe ikr L ~ kr ! , ~21! where L ~ l ! 5 Al E p /2 g~ u !e ilF ~ u ! du, A 2 p /2 F ~ u ! 5 cos~ u 2 a ! . For large l[kr, the dominant contribution to L(l) comes from those points c in the range of integration at which the phase F is stationary: F 8 (c)50. As F 8 ( u )5 sin(a2u) and u a u , 21p , the only stationary-phase point is at u 5 a . Then ~Bleistein and Handelsman, 1986, p. 220! L ~ l ! ;Bg ~ a ! e ilF ~ a ! as l→`, ~22! where H S 2 1 1 cos l2 m p 2 p pl 2 4 D as l→`. ~26! C. Uniform asymptotics and A where J m is a Bessel function. The result follows from the well-known asymptotic approximation, J m~ l ! ; g ~ u ! 5 @ 12 cos~ u 2 a !# e 2ik r cos~ u 2 w ! 70 @~ 12sin a ! e iF 1 ~ 11sin a ! e 2iF # whence ikr 2ik r cos~ u 2 w ! B52 12 i A~ 2/p ! e 2i p /4. B5 Alcos a L ~ l ! ; A2 p e i p /4e 2il e ik r sin w as r→`, where x̂5x/r, y5( r sin w,r cos w) and v i It follows that B B L~ l !5 I ~ v ;H r ! 5O„~ kr ! 21/2… II. THE METHOD OF STATIONARY PHASE G ~ P,q ! . But g( a )50, and so L(l)5o(1) as l→`. In fact, an integration by parts gives J 2p 1 exp i p sgn F 9 ~ a ! 5 A2 p e 2i p /4. u F 9~ a !u 4 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 We have seen that the results for u a u , 21p and u a u 5 21p are different, that is, the asymptotic estimate of I( v ;H r ) is not uniform in the parameter a . However, we can obtain a uniform approximation ~see Appendix A!; for example, if a is near 21p , we find that I ~ v ;H r ! . cos ~ 21 d ! e ik r sin w erfc~ m ! , ~27! where erfc~ m ! 5 E Ap 2 ` 2 e 2x dx ~28! m is the complementary error function, m 5 A2le 2i p /4 sin 21 d and d 5 21 p 2 a . J. A. DeSanto and P. A. Martin: Rough surface integral equations 70 Note that if a 5 21p ( d 50), we recover ~25!. On the other hand, if a , 21p ( d .0), we recover ~23!, since erfc( m ); p 21/2m 21 exp(2m2) as m →`. 22i G ~ u !5 p (j e i ~ 2 j11 ! u 4 5 2 j11 p ` ( j50 sin~ 2 j11 ! u , 2 j11 ~33! this is a familiar Fourier series: III. AN EXPANSION METHOD G ~ u !5 The integral I( v ;H r ) is over a semicircle of radius r, between u 52 21p and u 51 21p . We can evaluate this integral explicitly, using appropriate expansions of v and G. Thus ` v~ r, u ; a ! 5e ikr cos~ u 2 a ! 5 ( m52` i m J m ~ kr ! e 2im ~ u 2 a ! , ~29! henceforth, we suppress the limits when the summation is over all integers. Similarly, 1 G ~ P,q ! 5 2 i 2 (n H n~ kr ! J n~ k r ! e in ~ u 2 w ! ~30! for r. r , where H n [H (1) n . Hence I ~ v ;H r ! 5 H 1, 0,u,p, 21, 2p,u,0, 0, and is defined by periodicity for other values of u . Hence, for large kr, F n ;1 for 0< u a u , 21p but F n 5o(1) for u a u 5 21p . Thus we obtain the same ~nonuniform! results as derived in Secs. II A and II B. The drawbacks with this method are that it does not yield results that are uniform in a for a near 6 21p , and it is very complicated to use for three-dimensional problems. IV. ASYMPTOTIC BEHAVIOR OF I ( u ; H r ) When a plane wave is reflected by a rough surface S, we can use the angular-spectrum representation ~12! for the reflected field above the corrugations. Thus we have I ~ u;H r ! 5I ~ u pr ;H r ! 1I ~ u ev ;H r ! 1I ~ u con ;H r ! . (m (n i m J n~ k r ! W mn~ kr ! A mn e i~ m a 2n w !, For I(u ev ;H r ), with u ev defined by ~13!, we have where I ~ w;H r ! .iB Akre ikr 1 2 8 ~ w ! H n~ w !% W mn ~ w ! 5 2 i p w $ J m ~ w ! H n8 ~ w ! 2J m E p /2 g ~ u ! e ikrF ~ u ! d u , 2 p /2 where w is defined by ~16!, and 1 A mn 5 p 5 E H p /2 e i ~ n2m ! u F ~ u ! 5 m sin u 1i Am 2 21cos u du and 2 p /2 1 g ~ u ! 5 @ 12F ~ u !# e 2ik r cos~ u 2 w ! , if m5n, 2 ~ 21 ! j ~ 2 j11 ! p 0 as u m u .1, integration by parts shows that if m5n12 j11, I ~ w;H r ! 5O„~ kr ! 21/2… otherwise, I ~ v ;H r ! 5 I ~ u ev ;H r ! →0 (n i n J n~ k r ! $ W nn~ kr ! 2F n~ kr, a ! % e in~ a 2 w !, 22i p 1 (j 2 j11 W n12 j11,n e i~ 2 j11 !a . ~31! We want to estimate I( v ;H r ) for large r. We can evaluate the first term in the braces exactly: W nn is essentially a Wronskian, given by W nn 51. For F n , we have W mn ~ w ! ;exp$ i ~ m2n ! p /2% as w→`, for fixed m and n. ~32! We proceed formally, and use this approximation in ~31!. ~This procedure can be justified; see Appendix B.! The result is F n ~ kr, a ! ;G ~ a 1 21 p ! as kr→`, independently of n, where J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 as r→`. For u con , we have u con~ r, u ! ; where F n ~ kr, a ! 5 as kr→`. Hence, from ~13!, here, j is an arbitrary integer. It follows that 71 ~34! u50,6 p , A 2 p i ~ kr2 p /4! e C ~ sin u ! kr as kr→`. This result makes essential use of the continuity of C( m ) ~see Clemmow, 1966, Sec. 3.2!. Thus, u con satisfies the Sommerfeld radiation condition ~1!, whence I ~ u con ;H r ! →0 as r→`. @A direct derivation of this result, based on ~14! and the method of stationary phase, is given in Appendix C.# Finally, consider I(u pr ;H r ). If u a n u , 21p , the results of the previous sections are immediately applicable, and show that I(u pr ;H r )→0 as r→`. Next, consider grazing waves, u a n u 5 21p , and write v 6 5 v~ r, u ;6 21 p ! 5e 6ikx . We have 2 I ~ u pr ;H r ! 5I ~ u pr ;H 1 r ! 1I ~ u pr ;H r ! , J. A. DeSanto and P. A. Martin: Rough surface integral equations 71 where H 6 r are quarter-circles defined by ~17!; the radiation condition ~all plane waves must propagate outwards through H r , away from O) implies that we limit our attention to 7 I( v 6 ;H 6 r ), as v 6 propagates inwards through H r . So, from Sec. II A, we have il I ~ v 1 ;H 1 r ! .iBe Al E p /2 ~ 12sin u ! 0 3e 2ik r cos~ u 2 w ! e il sin u d u for large l[kr. There is one point of stationary phase ~cf. Sec. II B! at u 5 21p , but the integrand vanishes there whence I( v 1 ;H 1 r )→0 as r→`. A similar argument succeeds for I( v 2 ;H 2 r ). In summary, we find that our radiation condition ensures that I ~ u;H r ! →0 ~35! as r→`. This is ~formally! the standard Helmholtz integral equation for the boundary values of u on S. It is common to not work with ~37! but with an integral equation for the total field, u tot . To obtain such an equation, we start by defining a region D̃ r with boundary ] D̃ r . Given r, let H̃ r denote the semicircle in z<0, with radius r and center O; then, D̃ r is the bounded region in z,s enclosed by H̃ r and the rough surface. The boundary ] D̃ r consists of a piece of S r , namely S t with t(r),r, and a piece of H̃ r , namely H̃ r \ T̃ r where T̃ r consists of two circular arcs joining the ends of H̃ r and the ends of S t . Now, apply Green’s theorem to u inc and G in D̃ r . As both fields satisfy ~7! in D̃ r , the result is 05 EH St u inc~ q ! J ]G ] u inc G ~ P,q ! ds q ~ P,q ! 2 ]nq ]nq ~38! 2I ~ u inc ;H̃ r ! 1I ~ u inc ; T̃ r ! V. BOUNDARY INTEGRAL EQUATIONS In Sec. I, we used Green’s theorem to obtain the integral representation 2u ~ P ! 5 EH u~ q ! Sr J ]G ] u inc G ~ P,q ! ds q ~ P,q ! 1 ]nq ]nq I ~ u inc ;H̃ r ! →0 when PPD r , the region bounded by the semicircle H r , the truncated rough surface S r , and the two line segments T r . Note that the left-hand side of this equation does not depend on r, so that the right-hand side of the equation must have a limit as r→`. Taking this limit, using ~19! and ~35!, we obtain EH u~ q ! S J ]G ] u inc G ~ P,q ! ds q , ~ P,q ! 1 ]nq ]nq 5 lim r→` E as u x u →`, where U(x) is only required to be bounded. We remark that Beckmann and Spizzichino ~1963, p. 180! and Ogilvy ~1991, p. 75! discard I(u;H r ) by assuming erroneously that u50 on H r . Letting P→ pPS in ~36! gives S u~ q ! ]G ~ p,q ! ds q 5 ]nq ~39! Letting r→` in ~38!, and adding the result to ~36!, we obtain E S u tot~ q ! ]G ~ P,q ! ds q 2U i ~ P ! , ]nq PPD ` . E J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 ~40! Then, letting P→p P S gives E S u tot~ q ! ]G ~ p,q ! ds q 52u inc~ p ! 2U i ~ p ! , ]nq ~41! Now, the standard Helmholtz integral equation for the total field is w~ p !2 E w~ q ! S ]G ~ p,q ! ds q 52u inc~ p ! , ]nq pPS. ~42! Thus, for nongrazing incident waves (U i 50), we see that u tot does satisfy the standard Helmholtz integral equation. However, for grazing incident waves, u tot does not satisfy the Helmholtz integral equation ~42!, but 2u tot does. VI. DISCUSSION A. Previous work: Helmholtz integral equations ] u inc G ~ p,q ! ds q , S ]nq pPS. 72 as r→` ~ u u i u < 21 p ! . pPS. U ~ x ! e ik u x u u x u 21/2 E I ~ u inc ;H̃ r ! →Ui ~ P ! , Sr which is the standard definition of a principal-value integral at infinity. In fact, the integral over S exists as an ordinary improper integral; to see this, we note that the integrand behaves like u~ p !2 We combine these formulas and write u tot~ p ! 2 where D ` is the unbounded region z.s and S I ~ u inc ;H̃ r ! 5u inc~ P !~ u u i u 5 21 p ! . 2u ~ P ! 5 ~36! PPD ` , E as r→` ~ u u i u , 21 p ! . For grazing incidence, we have 1I ~ u;H r ! 1I ~ u;T r ! 2u ~ P ! 5 when PPD r , taking into account the direction of the normal vector on S. As before, simple bounds show that I(u inc ; T̃ r )→0 as r→`. If u u i u , 21p , u inc @given by ~6!# is a plane wave propagating outwards through H̃ r , whence ~37! The idea that a boundary integral equation may be used to solve the problem of plane-wave scattering by an infinite rough surface is familiar. It is discussed in books on such J. A. DeSanto and P. A. Martin: Rough surface integral equations 72 problems; see for example Maystre and Dainty ~1991!, Ogilvy ~1991, Secs. 4.1.1 and 6.3! and Voronovich ~1994, Sec. 3.1!. In particular, the standard Helmholtz integral equation ~4! is equation ~6.53! in Ogilvy’s book and equation ~3.1.37! in Voronovich’s book. Many recent authors refer to the paper by Holford ~1981! on the scattering of a plane wave by a periodic soundsoft surface. He obtains the integral representation @his equation ~20!# u tot~ P ! 5u inc~ P ! 1 1 2 E S u tot~ q ! ]G ~ P,q ! ds q , ]nq ~43! PPD ` , as in Sec. I C, by applying Green’s theorem to u tot and G in the semicircular region D r @using our notation and the boundary condition ~8!#. He claims that ‘‘the term u inc( P) is the contribution from the large semicircle’’ H r as r→`. He does not prove this statement and, moreover, it is not true for grazing incident waves. To see this, we note that applying Green’s theorem to u tot and G in D r gives 2u tot~ P ! 5 E Sr u tot~ q ! ]G ~ P,q ! ds q 1I ~ u tot ;H r ! ]nq that we can use the large-argument approximation for H (1) 1 (kR). Thus, assuming for simplicity that U5U i 50, ~40! gives 2u ~ x,z ! 5 I ~ u tot ;H r ! 5I ~ u;H r ! 1I ~ u inc ;H r ! 5I ~ u;H r ! 1I ~ u inc ;C r ! 2I ~ u inc ;H̃ r ! , where C r 5H r øH̃ r is a large circle. But, for an incident plane wave, I ~ u inc ;C r ! 52u inc , E S u tot~ q ! E ;iB Ak S H ~11 ! ~ kR ! u tot~ q ! R n~ q ! • ~ q2y! ds q e ikR n~ q ! • ~ q2y! ds q R 3/2 ~44! as kz→`, where B is defined by ~20! and n(q) is the unit normal vector at q pointing out of D ` . If u tot(q) has a compact support ~so that it vanishes for u qu .L, say!, or if S is finite ~bounded scatterer!, we can make a second approximation: R5 $ r 2 22y•q1q 2 % 1/25r2ŷ•q1O ~ q 2 /r ! as r→`, ~45! where r5 u yu , ŷ5y/r, and q5 u qu . ~The notation used here differs from that used in Secs. I C and II A.! To the same order, we can also replace (q2y) in ~44! by (2y). Hence, we find that u is given by ~5!, where the far-field pattern is f ~ u !5 2 1I ~ u tot ;T r ! . Now, I(u tot ;T r )→0 as r→` and ik 2 iB Ak 2 E S u tot~ q ! n~ q ! •ŷ exp$ 2ikŷ•q% ds q . However, one cannot justify the use of the approximation ~45! for plane-wave incidence and unbounded surfaces ~see the discussion by Ogilvy, 1991, p. 78!. This is immediately clear, because there must be a reflected plane wave, whereas ~45! leads to a cylindrical wave. For explicit confirmation, consider the reflection of a plane wave by a flat surface, so that u tot(q)52 exp$ikj sin ui%; the integral in ~44! can then be estimated using the method of stationary phase @and yields the correct u, given by ~9!#, whereas the integral for f ( u ) diverges. exactly, whence I ~ u tot ;H r ! →2u inc2Ui as r→`, where Ui is defined by ~39!. Thus ~43! is correct whenever Ui [0. Holford himself refers to earlier papers by Urosovskii ~1960!, who in turn refers to Lysanov ~1956!. For more recent work, we can cite Thorsos ~1988!, Bishop and Smith ~1992! and McSharry et al. ~1995!. All these papers start from the Helmholtz integral equation for u tot , ~4!, or the analogous equation for a sound-soft surface. Moreover, all but one of these papers are concerned with plane-wave incidence, the exception being the paper by Thorsos ~1988!. He considers a ‘tapered’ plane wave; we will discuss beams of finite extent in Sec. VI D. B. Far-field asymptotics Care is needed when approximating the scattered field at large distances from an infinite surface. To illustrate this, consider the integral representation ~40!. Let P[(x,z) PD ` and q[( j ,s( j ))PS have position vectors y and q, respectively. In the far field, kR@1, where R5 u y2qu , so 73 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 C. A finite patch of roughness Suppose that the infinite surface S is flat, apart from a finite patch of roughness, S patch , confined to u x u ,L, say. A plane-wave incident on such a patch will generate a specular plane wave and a cylindrical wave. Thus, for nongrazing incidence, the standard Helmholtz integral equation for the total field, ~4!, is valid. To see that the decomposition itself is valid, write u tot5u flat1u cyl , where u flat5e 2ikrcos~ u 1 u i ! 1e ikr cos~ u 2 u i ! ~ u u i u , 21 p ! is the total field for reflection by an infinite flat sound-hard surface. Thus ] u cyl ] u flat 52 ]n ]n on S patch. ~46! There are now three cases to consider, namely, ridges, grooves, and a combination thereof. J. A. DeSanto and P. A. Martin: Rough surface integral equations 73 1. Ridges Suppose that S patch consists of a finite number of ridges ~‘‘bosses’’!, so that s(x)>0. Then, the scattering problem is equivalent to the scattering of two plane waves, u flat , by a finite bounded obstacle ~a ‘‘double-body’’! with boundary S patchøS 8patch , where S 8patch is the reflection of S patch in the line z50. Thus u cyl is a cylindrical wave, satisfying the Sommerfeld radiation condition ~1!. The use of bosses to model rough surfaces is well known ~Ogilvy, 1991, Sec. 6.1!. The use of images to treat scatterers near flat impenetrable boundaries is also well known; for a recent application, see Chao et al. ~1996!. Indeed, if we introduce the exact Green’s function for the halfplane z>0, 5 2 21 i $ H ~01 ! ~ k A~ x2 j ! 2 1 ~ z2 z ! 2 ! 1H ~01 ! ~ k A~ x2 j ! 2 1 ~ z1 z ! 2 ! % , we find that E H S ridge 1 u cyl~ q ! ]GE ~ P,q ! ]nq J E ] u flat E G ~ P,q ! ds q , ]nq S ridge From the discussion above, we see that if S patch consists of a finite number of ridges and grooves, then u cyl can be represented using u cyl~ P ! 5 E S1 n ~ q ! G E ~ P,q ! ds q , u tot~ p ! 52u inc~ p ! 1 PPD ` , where S ridge is the union of all the ridge surfaces (S patch\ S ridge is part of z50). Letting P→pPS ridge yields a boundary integral equation for u cyl(p). Alternatively, we can write u cyl~ P ! 5 3. Ridges and grooves where S 1 5S ridgeøS mouth , P[(x,z) and z.max$s(x),0% . The determination of n on S 1 is complicated, although S 1 is a finite surface. If we use the Helmholtz integral equation for u tot , which we know is legitimate, we have to solve an integral equation over an infinite surface. However, this can be reduced to an integral equation over S patch as follows. Since ] G(p,q)/ ] n q 50 when both p and q are on the flat part of S, S flat5S \ S patch , ~4! gives G E ~ P,Q ! [G E ~ x,z; j , z ! 2u cyl~ P ! 5 with grooves ~compared to ridges! has given rise to more sophisticated methods for solving such problems, involving more complicated integral representations; see Willers ~1987! and Asvestas and Kleinman ~1994!. Applications of G E to rough-surface scattering were made by Berman and Perkins ~1985! and by Shaw and Dougan ~1995!. n 1 ~ q ! G E ~ P,q ! ds q , PPD ` , ~47! where the boundary condition ~46! implies that the source density n 1 solves a Fredholm integral equation of the second kind over S ridge . E u tot~ p ! 2 E S patch E S mouth n 2 ~ q ! G E ~ P,q ! ds q , P[ ~ x,z ! and z.0. ~48! To find the source density n 2 , we can apply Green’s theorem inside each groove to u cyl and G; we have the boundary condition ~46! on the surface of each groove, and we have ~transmission! conditions enforcing the continuity of u cyl and ] u cyl / ] z across the mouth of each groove. Note that we cannot use ~48! inside the grooves because of the image singularities in G E . This extra complication 74 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 pPS flat . u tot~ q ! K ~ p,q ! ds q 52 f ~ p ! , pPS patch , where K ~ p,q ! 5 ]G ~ p,q ! 1 ]nq 2. Grooves u cyl~ P ! 5 ]G ~ p,q ! ds q , ]nq This means that u tot on the ~infinite! flat part of S is known in terms of u tot on the rough part of S. Hence, we can write ~4! for pPS patch as f ~ p ! 5u inc~ p ! 1 Suppose that S patch consists of a finite number of grooves, so that s(x)<0. Then, ] u cyl / ] z is known, in principle, for all x on z50: It is zero except across the mouth of each groove. As there is a finite number of grooves, it follows that u cyl is a cylindrical wave; it has an angularspectrum representation with a continuous spectral amplitude. Let S mouth be the union of all the groove mouths; it is part of z50. We can write S patch u tot~ q ! E S flat E ]G ]G ~ p,l ! ~ q,l ! ds l , ] n ] nq S flat l u inc~ q ! ]G ~ p,q ! ds q . ]nq D. Finite beams So far we have taken the incident field to be a plane wave. However, for many applications, the incident field is a finite beam. To construct such a beam, we start by considering a single line-source at Q, u inc~ P ! 5G ~ P,Q ! . As u inc satisfies the Sommerfeld radiation condition, an energy argument ~DeSanto and Martin, 1996! shows that the scattered field cannot include any reflected plane waves. Thus, the standard Helmholtz integral equations, ~3! and ~4!, are valid; see DeSanto and Brown ~1986, Sec. 4.1!. Next, we distribute the line-sources over a finite curve Q ~or a finite region!, to give u inc~ P ! 5 En Q inc~ Q ! G ~ P,Q ! ds Q , where n inc is prescribed and can be adjusted to make u inc beamlike. ~If Q is far from S, the asymptotic approximations J. A. DeSanto and P. A. Martin: Rough surface integral equations 74 described in Sec. VI B can be used.! It follows that the standard Helmholtz integral equations remain valid. Another way to generate a beam is to use an angularspectrum representation, u inc~ r, u ! 5 E p /2 2 p /2 A inc~ a ! e ikr cos~ u 1 a ! d a , ~49! where A inc is a prescribed continuous function, and is typically taken as a Gaussian ~Saillard and Maystre, 1990!. The standard Helmholtz integral equations are valid for incident fields of this type. Thorsos ~1988! uses a ‘‘tapered’’ plane wave. This incident field does not satisfy the Helmholtz equation, and so the derivation of the Helmholtz integral equation for u tot fails. @The Helmholtz integral equation for u, ~3!, is valid without this qualification.# Nevertheless, as Thorsos points out, the tapered plane wave is an approximation to an actual wave field, constructed using ~49!; see Thorsos ~1988, Sec. I B!. E. Other integral equations An alternative way of solving scattering problems, touched on above, is to assume that the scattered field can be written as u~ P !5 Eg S ~ q ! G ~ P,q ! ds q , PPD, where the source density g is unknown. For sound-hard surfaces, the boundary condition ~8! yields an integral equation for g , g~ p !2 Eg ~q! S ]G ] u inc , ~ p,q ! ds q 5 ]np ]np pPS. whereas the integral equation for the boundary values of the total field is valid for nongrazing incident waves ~a simple modification is required for grazing incident waves!. These results underpin the use of these integral equations for numerical computations. The standard Helmholtz integral equations are valid if the roughness is confined to a finite portion of an otherwise flat but infinite surface. They are also valid for incident beams of finite width. Similar results may be obtained for sound-soft surfaces. Extension to electromagnetic and elastodynamic problems, and to penetrable interfaces, should be straightforward. Finally, extension of these ideas to two-dimensional rough surfaces can also be made, although the analysis is more difficult and the results are different. Some of these aspects are currently under investigation. ACKNOWLEDGMENTS PAM acknowledges receipt of a Fulbright Scholarship Grant. He also thanks the Department of Mathematical and Computer Sciences, Colorado School of Mines, for its kind hospitality. Both authors are grateful to A.G. Voronovich for his perceptive remarks on an earlier version of the paper, which encouraged us to clarify the role of grazing waves. APPENDIX A: UNIFORM ASYMPTOTICS We derive a uniform approximation for I( v ;H r ), using a method discussed by Bleistein and Handelsman ~1986, Sec 9.4!. We start by focussing on the non-uniformity at a 5 21 p ; the nonuniformity at a 52 21p can be treated similarly. Write L5L 1 1L 2 where For sound-soft surfaces, the corresponding integral equation is Eg S ~ q ! G ~ p,q ! ds q 5 2u inc~ p ! , pPS; this has been solved numerically by Lentz ~1974!, Rodrı́guez et al. ~1992!, and others. Chandler-Wilde and Ross ~1995, 1996! use a doublelayer potential for sound-soft surfaces, u~ P !5 Eg S ~q! ]G1 ~ P,q ! ds q , ]nq VII. CONCLUSIONS We have seen that the use of standard Helmholtz integral equations for the scattering of a plane wave by an infinite, sound-hard, one-dimensional, rough surface is justified in most circumstances. In particular, the equation for the boundary values of the scattered field is always valid, J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 E 0 g ~ u ! e ilF ~ u ! d u 2 p /2 and L 2 5L2L 1 . We have L 2 5o(1) as l→`, uniformly in a , for a bounded away from 2 21p . For L 1 there is a point of stationary phase at u 5 a 2 p ~outside the range of integration! which approaches the end point at u 52 21p as a → 12p . Let us make a preliminary change of variables, mapping the end point to the origin: Put u 5x2 21p and a 5 21p 2 d giving PPD, where G 1 satisfies an impedance condition on a line z52h 0 (h 0 .h). All of the formulations mentioned in this section are indirect, in that they assume that u( P) can be represented in a specified form. Thus the radiation condition is implicit in the representation. 75 L 1 ~ l ! 5 Al L 1 ~ l ! 5 Al E p /2 h ~ x ! e il f ~ x; d ! dx 0 with h ~ x ! 5 @ 11cos~ x1 d !# e ik r sin ~ w 2x ! and f ~ x; d ! 5 2cos~ x1 d ! . Thus L 1 has a stationary-phase point at x52 d , which approaches the end-point x50 as d →0. The prototype special function with this property is the complementary error function, defined by ~28!. In order to relate this function to L 1 , we change the integration variable from x to t, using f ~ x; d ! 2 f ~ 0; d ! 5 21 t 2 1 g t, J. A. DeSanto and P. A. Martin: Rough surface integral equations 75 requiring that x52 d is mapped to t52 g gives g 52 3 sin 21d . Hence, L 1 ~ l ! 5 Ale il f ~ 0; d ! E T h̃ ~ t ! e il ~ 1/2 t 21gt ! dt, 0 where h̃ ~ t ! 5h ~ x ~ t !! APPENDIX C: ASYMPTOTIC BEHAVIOR OF I ( U con ; H r ) dx t1 g dx and 5 . dt dt f 8 ~ x; d ! For the dominant contribution, we can set the upper limit T5` and replace h̃ (t) by h̃ (0); this gives L 1 ~ l ! . Ale il f ~ 0; d ! gh~ 0 ! f 8 ~ 0; d ! E ` e il @~ 1/2 ! t 21gt # dt. p i p /4 2 1/2il g 2 e e erfc~ m ! 2l with m 5 g Disregarding the evanescent terms in ~14!, we can write u con as the integral in ~11!, whence I ~ u con ;H r ! 5 D~ l !5 A 5 L 1 ~ l ! . A2 p cos 21 d e i p /4e 2il e ik r sin w erfc~ m ! In order to estimate F n (kr, a ) for large kr, we substituted the asymptotic approximation ~32! into ~31!. However, this requires some justification, as ~32! presupposes that m and n are fixed. @A hint that nonuniform behavior might be expected comes from ~29!: The left-hand side is a plane wave whereas every term on the right-hand side is O((kr) 21/2) as kr→`.# We start by writing F n ~ w, a ! 5w $ H 8n ~ w ! F n ~ w, a ! 2H n ~ w !~ ] / ] w ! F n ~ w, a ! % , where ~B1! Next, we use a standard integral representation for Bessel functions, J m~ w ! 5 1 2p E p e i ~ m u 2w sin u ! d u , 2p whence F n ~ w, a ! 5 i 4 E p 2p G ~ u 1 a ! e i ~ n u 2w sin u ! d u ~B2! where G ( u ) is defined by ~33!. Finally, we use the method of stationary phase to estimate ~B2! for large w; there are stationary-phase points at u 56 21p , whence ~22! gives F n ~ w, a ! 5 A~ 21 p /w ! sin ~ w2 21 n p 2 41 p ! 1O ~ w 21 ! as w→`, for 0< u a u , 21p , whereas F n ~ w,6 21 p ! 5O ~ w 21 ! 76 as w→`. J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 A ~ a ! I ~ v ;H r ! d a .iBe il AlD ~ l ! E p /2 A~ a ! 2 p /2 E p /2 2 p /2 @ 12cos~ u 2 a !# E p /2 A~ a ! 2 p /2 E p /22 a ~ 12cos c ! 2 p /22 a As A is a continuous function of a , we can change the order of integration to give APPENDIX B: ASYMPTOTIC BEHAVIOR OF F n 1 2 p /2 3e 2ik r cos~ c 1 a 2 w ! e il cos c d c d a . whence the final result ~27! follows. (j 2 j11 J n12 j11~ w ! e i~ 2 j11 !a . p /2 3e 2ik r cos~ u 2 w ! e il cos~ u 2 a ! d u d a 1 2i p /4 le . 2 Substituting back, we obtain F n ~ w, a ! 5 E as l[kr→`, where we have used ~21!, and 0 Standard manipulations show that the integral can be expressed as A These results are exactly the same as we would have obtained if we had replaced the Bessel function in ~B1! by the leading term in its large-argument asymptotic expansion, namely ~26!. D~ l !5 E p ~ 12cos c ! Q ~ c ! e il cos c d c , ~C1! 0 where Q~ c !5 E p /22 c 2 p /2 $ A ~ a ! e 2ik r cos~ c 1 a 2 w ! 1A ~ 2 a ! e 2ik r cos~ c 1 a 1 w ! % d a . Now we can estimate D(l) for large l using the method of stationary phase. The stationary-phase points are c 50 and c 5 p . At c 50, 12cos c50, whereas Q( p )50. Hence, the integrand in ~C1! vanishes at both stationaryphase points, whence D(l)5O(l 21 ) as l→`. Thus, we deduce that I(u con ;H r )→0 as r→`. Asvestas, J. S., and Kleinman, R. E. ~1994!. ‘‘Electromagnetic scattering by indented screens,’’ IEEE Trans. AP-42, 22–30. Beckmann, P., and Spizzichino, A. ~1963!. The Scattering of Electromagnetic Waves from Rough Surfaces ~Pergamon, Oxford!. Berman, D. H., and Perkins, J. S. ~1985!. ‘‘The Kirchhoff approximation and first-order perturbation theory for rough surface scattering,’’ J. Acoust. Soc. Am. 78, 1045–1051. Bishop, G. C., and Smith, J. ~1992!. ‘‘A scattering model for nondifferentiable periodic surface roughness,’’ J. Acoust. Soc. Am. 91, 744–770. Bleistein, N., and Handelsman, R. A. ~1986!. Asymptotic Expansions of Integrals ~Dover, New York!. Chandler-Wilde, S. N., and Ross, C. R. ~1995!. ‘‘Scattering by onedimensional rough surfaces,’’ in The 3rd International Conference on Mathematical and Numerical Aspects of Wave Propagation, edited by G. Cohen, E. Bécache, P. Joly, and J. E. Roberts ~SIAM, Philadelphia!, pp. 208–215. Chandler-Wilde, S. N., and Ross, C. R. ~1996!. ‘‘Scattering by rough surfaces: the Dirichlet problem for the Helmholtz equation in a non-locally perturbed half-plane,’’ Math. Methods Appl. Sci. 19, 959–976. Chao, J. C., Rizzo, F. J., El-Shafiey, I., Liu, Y. J., Udpa, L., and Martin, P. A. ~1996!. ‘‘A general formulation for light scattering by a dielectric body near a perfectly conducting surface,’’ J. Opt. Soc. Am. A 13, 338–344. J. A. DeSanto and P. A. Martin: Rough surface integral equations 76 Clemmow, P. C. ~1966!. The Plane Wave Spectrum Representation of Electromagnetic Fields ~Pergamon, Oxford!. Colton, D., and Kress, R. ~1983!. Integral Equation Methods in Scattering Theory ~Wiley, New York!. DeSanto, J. A., and Brown, G. S. ~1986!. ‘‘Analytical techniques for multiple scattering from rough surfaces,’’ in Progress in Optics XXIII, edited by E. Wolf ~Elsevier, New York!, pp. 1–62. DeSanto, J. A., and Martin, P. A. ~1996!. ‘‘On angular-spectrum representations for scattering by infinite rough surface,’’ Wave Motion 24, 421– 433. Holford, R. L. ~1981!. ‘‘Scattering of sound waves at a periodic, pressurerelease surface: An exact solution,’’ J. Acoust. Soc. Am. 70, 1116–1128. Lentz, R. R. ~1974!. ‘‘A numerical study of electromagnetic scattering from ocean-like surfaces,’’ Radio Sci. 9, 1139–1146. Lysanov, Iu. P. ~1956!. ‘‘One approximate solution for the problem of the scattering of acoustic waves by an uneven surface,’’ Sov. Phys. Acoust. 2, 190–197. Maystre, D., and Dainty, J. C. ~Eds.! ~1991!. Modern Analysis of Scattering Phenomena ~Adam Hilger, Bristol!. McSharry, P. E., Moroney, D. T., and Cullen, P. J. ~1995!. ‘‘Wave scattering by a two-dimensional pressure-release surface based on a perturbation of the Green’s function,’’ J. Acoust. Soc. Am. 98, 1699–1716. 77 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Ogilvy, J. A. ~1991!. Theory of Wave Scattering from Random Rough Surfaces ~Adam Hilger, Bristol!. Ramm, A. G. ~1986!. Scattering by Obstacles ~Reidel, Dordrecht!. Rodrı́guez, E., Kim, Y., and Durden, S. L. ~1992!. ‘‘A numerical assessment of rough surface scattering theories: Horizontal polarization,’’ Radio Sci. 27, 497–513. Saillard, M., and Maystre, D. ~1990!. ‘‘Scattering from metallic and dielectric rough surfaces,’’ J. Opt. Soc. Am. A 7, 982–990. Shaw, W. T., and Dougan, A. J. ~1995!. ‘‘Half-space Green’s functions and applications to scattering of electromagnetic waves from ocean-like surfaces,’’ Waves Random Media 5, 341–359. Thorsos, E. I. ~1988!. ‘‘The validity of the Kirchhoff approximation for rough surface scattering using a Gaussian roughness spectrum,’’ J. Acoust. Soc. Am. 83, 78–92. Urosovskii, I. A. ~1960!. ‘‘Sound scattering by a sinusoidally uneven surface characterized by normal acoustic conductivity,’’ Sov. Phys. Acoust. 5, 362–369. Voronovich, A. G. ~1994!. Wave Scattering from Rough Surfaces ~SpringerVerlag, Berlin!. Willers, A. ~1987!. ‘‘The Helmholtz equation in disturbed half-spaces,’’ Math. Methods Appl. Sci. 9, 312–323. J. A. DeSanto and P. A. Martin: Rough surface integral equations 77