Lecture Notes on the Mathematics of ACOUSTICSLecture Notes on the Mathematics of ACOUSTICS Edited by M. C.M. WRIGHT (University of Southampton, UK) Imperial College Press ee 3 7 7 7 T T pees enePublished by Inmpedal College Pres 57 Shelon Suet Covent Garden Landon WC2E SHE Distbuted by ‘Word Seiente Publishing Co, Pe. Ld ‘Toh Tuck Link, Singapore 596224 USA ees 27 Warren Stree, Suite 401-402, Hackensack, NOTED! UR office: 57 Shion Stes, Covent Garde, London WC2H SHE Brits Library Cataloguing i-Poblication Data ‘Aentalogue record for thi book i avallable om he Beish Libary [LECTURE NOTES ON THE MATHEMATICS OF ACOUSTICS ‘Copyright © 2005 by InpeialCollege Press All ight reserved Ths book, o parts thereof, may not be reproduced in any form or by ay means, tlecronicor mechanical, lading photocopying, recording or any information storage and retrieval {tem now known ovo be ivented, without writer person fom the Publisher. For photocopying of maseil inthis volume, please pay a copying fe trough the Copyright (Clearance Canter, Inc, 222 Rosewood Drive, Danvers, MA 01923, USA. ta this case permission to hotacopy isnt roguied from the publisher, ISBN 1-86004.496-5, Pred Singapore by Wid Slot Pairs (S) PBL. ae ‘To CarolynPreface ‘The chapters in this volume are based on lectures given at a one week summer School on mathematis for acoustics, funded by the UK Engineering and Physical Sciences Research Council (EPSRC) and held at the University of Southampton from the 14th to the 18th of July, 2003. The idea of such a summer school was conceived by Keith Attenborough in response to a call from EPSRC for propos- als for funding to increase the level of mathematical competency for engineer- ing researchers in the UK, issued by Engineering Programme Manager Elizabeth Hilton, and administered by Stephen Elsby. The SMART (Support Mathematics for Acoustics Research Training) summer school was the frst such summer school tobe funded and run; another has since taken place and a call has been issued for ‘more to follow. Although it was arranged at fairly short notice the summer school proved a ‘reat success. Over forty young acoustics researchers and, sbout twenty leetur- ers made their way to Southampton for an intensive week of mathematies, along with staff and research students of Southampton's School of Mathematies who provided additional tutorial support, The students lived together, ate together and spent most of the day in lectures, tutorials, seminars and one-to-one surgeries where they could raise subjects of interest with any of the staff they chose, They also had the chance to exchange ideas with one another, an opportunity that [hope ‘will benefit those who go on to form part ofthe UK’s future acoustics research ca- pability. The response was such that it is planned to run such a summer school biannuslly with this book as a textbook, though this does not, of course, preclude the addition of other topics in future, We do not claim thatthe topics discussed here provide anything like complete coverage of the field of mathematical acous- tics, merely that forms a useful body of material ata level that may be of use to students ofthe subject, Chapters 8 and 11 are based on seminars, which were given at the end ofa ‘Lecture Notes onthe Mathematics of Acourtes each day and the morning ofthe final day, These aimed to give an overview of particular area without fitting into the series of lectures which had a theme foreach day, similar though not identical tothe parts ofthis book. Additional seminars on subjects such as nonlinear acoustics, quantum chaos in acoustics and atmospheric acoustics could not be included for one reason or another, The remaining chapters are based on lectures apart from Chapters 5 and 7, which didn't find a place in the final timetable. Dring the editing process the material in this book has been somewhat ex- panded from what was presented during the summer school It was decided atthe start of our preparations forthe school that total consistency was not only imprac~ tical but actually undesirable, since several notations are in common use in the literature and it would be unfair to gloss over this inevitable feature of acoust- cal research. That excuse is ess valid for this book; nevertheless a few instances of diversity remain where the alternative would be too restrictive. Principally, in some chapters f denotes an integral transform of F whereas in others itis given by F. On the other hand exponential time factors are given as exp(—iut)through- ut. The following people should be thanked for their help with the production of this book: Mirko Schaedlich, Chris Howls, Chris Morfey and Jeremy Astley at Southampton; Dr K. K. Phua and Ying Oi Chiow at Imperial College Press and finally the dedicatee who, in the time since our wedding a month after the sum- ‘mer school, has never complained about the time I'v spent poring through ISTEX: ‘manuals or hunched over a laptop wrestling with this manuscript. Matthew Wright Contents Preface vii PartI Mathematical Methods 1. Vector Cateulus 3 JW. Bliont 1.1 Motivation wee LLL Velocity potential 11.2 Plane waves 1.13 Spherical waves 12 Scalars and Vectors oe ee 1.21 Rectangular Cartesian coordinates 122 — Geometric vectors . . . 123 Vector algebra 12.4 Thescalar product 1.25 Thevector product. 1.2.6 The triple scalar product 1.2.7. ‘The triple vector product... 12.8 Thestandard basis : 129 The position vector . 1.3 Vector and Scalar Functions podeead 1.3.1 Vector-valued functions ofa real variable 13.2 Vector and sealar fields 133. Differential operators 134 Thechain rules... ,Lecture Notes on the Mathemates of Acouaies 1.35 The del operator . 1.36 Vector identities . 13.7 Taylor's theorem . 14 Carves and Surfaces in 3-D Space 141 Cures ‘ 142 Surfaces 15 Curvilinear Coordinate Systems 15.1 Cartesian coordinates. 152 Cunilinear coordinates : 153° Orthogonal curvilinear coordinates 154 Cylindrical polar coordinates: (R, 4, 2) 155 Spherical polar coordinates: (r,8, 4) 1.6 Integrals BUG oho : 16.1 Line integrals... , 162 Surface integrals 1.63 Multiple integrals 17 Integral Theorems 5 17.1. Green's lemma 172 Gause's divergence theorem 173. Stokes's theorem 174 Green's theorems 1.75 Fundamental solutions... 1.76 Tre Green's function. 1.8 Suffix Notation 18.1 The Kronecker delta 1.82 The ltemating tensor . 183 Cartesian coordinates... 2... 184 Differential operators. Problems . 2. Functions of a Complex Variable 1.W Blioet 21 Complex Numbers... 6. 2a Algebraic operations. 2 Complex conjugate and modulus . . ‘The Areand diagram Polar coordinates. : 5 “The complex exponential form... 3 13 4 4 4 15 16 16 16 1 19 19 20 20 20 au 23 2B 23 24 25 28 26 2 21 28 29 29 30 33 33 33 34 35 36 Contents 21.7 Complex polynomials |... 2.1.8 Square and nth roots 2.2 Functions of a Complex Variable 22.1 Mappings . . . 222 The complex exponential function 22.3 Multiple-valued functions . 22.5 Analytic functions . 23. Contour Integration 23.1 Properties ...... 23.2 Closed contours... . 233 Cauchy's theorem 234 — Cauchy's integral theorem . 24 Power Series Expansions 24.1 The Taylor series. 242 The Laurent se 25 — Residue Theory 25.1 — Singularities . oe 252 Cauchy's residue theorem . , 253 Evaluation of real integrals. 2.6 Analytic Continuation Problems. . 3. Integral Transforms JW. Elion 3.1 The Laplace Transform . ‘The Heaviside step function Properties, : Differential properties . Convolution... 6... 5 ‘The convolution theorem". ‘The inverse transform 32 TheInversion Theorem... . 33. Fourier Series 33.1 Harmonics . 224 The complex logarithmic function . 45 26.1 Chains of analytic continuation . . . . 3.2.1 Evaluation of the contour integral 31 31 38 39 39 40 40 41 2 43 44 45 46 46 46 48 9 50 2 54 35 37 st 58 59 © 6 6 GI 6 6 6Lecture Noten te Mathematics of Acoustics a 332 Odd and even functions o 333. Comergence . : o 334 Differentiation and integration @ 335. The complex exponential form. 6 34 The Fourier Transform. . aos vee 10 3.4.1 The Fourier transform pair... 0 342 Fourier'stheorem . 343. General comments. 3.44 The duality property... 345 Further properties . . 346 Differential properties : 347 Parsevals theorem... eee eee 348 Theconvolution theorem... . 5 349 Thedeltafunction...... 15 3410 Furtherestensions . seve 1 35 TheHeat Conduction Equation 6.0... ses... 7B 35.1 Fourier ansform . 35.2 Green's function 565 353 Laplace transform . cece B 36 The Wave Equation seen 361 Timeharmonic motion... 2... ses BL 3.62 Fourier transform . . viene 363 Time-domain Green's function . 87 Problems... 88 4. Asymptotic Expansion of Integrals 91 RH.Self 4.1 Introduction. . 2 42. Two Elementary Methods « 93 42.1 Term by term integration... 93 422 Integration by paris... 2... % 423° Watson's lemma . ee 43. Method of Laplace 99 44 Method of Stationary Phase . » 100 44.1 Outline proof 102 45 Method of Steepest Descents - 103 FurtherReading 0.20... 005 see 104 Problems. 105 nents Part II Wave Motion 5. ‘The Wiener-Hopf Technique M.C.M, Wright 5.1 “Introduction... 52. Rigid Screen Diffraction 4 52.1 , Generalised Fourier transforms 522 Branchouts .. .. 523° One-sided transforms... , 5.24 Transformed boundary conditions. 5:3 ‘The General Wiener-Hopf Equation Analytic continuation . . Behaviour at infinity in the s-plane , 53.4 ‘Behaviour at the edge . , 54 A Systematic Decomposition Procedure. 54.1 Application to the sereen problem $5 Summary... 2.6.2.0. 6. Waveguides ‘M, Melver & C. M. Linton 6.1 Basic Theory... . 6.11 Introduction 6.1.2 Boundary conditions. 6.13 Time harmonic motion. 6.1.4 Plane waves . 6.15 Acoustic waveguides . 6.1.6 Cut-off frequencies . 6.1.7 Cylindrical guides . 62. Eigenfunction Expansions , 62.1 Arolling piston . . 7 63 An Abrupt Change in Boundary Condition 63.1 Matched eigenfunction expansions 64 Complex Variable Methods . 64.1 Cauchy's residue theorem , 642 Abrupt change in boundary condition: soltion using residue calculus theory . . + U2 Basie) us Application tothe diffraction problem ++ 126 + 126 - 129 == 130 - 132 - 132 109 109 ++ 109 ul m2 us 6 NT U8 ++ 120 12 123 125 125 125 127 127 134 135 . 137 . 137 . 138Lecture Notes on the Mathematics of Acustice i Conn 5 643 The Wiener-Hopf technique. 40 : Part I Aeroacoustics Further Reading wees . 143 > uw Problems... 218 9, Generalised Funetions in Aeroacousties 1" Peake 7. Wavelet Decomposition as aca i 8.1 ngodsetion « 19 9.1.1 Introductory remarks and basic concept. “19 7.1 Introduction 145 te ree eee + 19 12 Wiavefeds , : as 9113 Sealingand general argument |. sss 181 721 Thyee-dimensional Fourier wansform . 5... 146 9.14 Differentiation : ‘ie2 73. TwoDimensional Representation vo... esc sees ro 9.15 Extension to higher dimensions... “184 74 Extension to Three Dimensions «ss. a6 92 Applcaionin Aeroaoustos «es 0,2. sees see 184 7S AngularRepresentaion 60... s ves c sve e es 150 92.1 Abasicequaion. 0... 184 76 Solution to Original Problem 153 922 Shock noise « 187 Problems . . 154 9.3 Fourier Transforms «188 9.3.1 Definitions: and basics . 189 8, Acoustics of Rigid-Porous Matra 137 932 Some examples 189 . Atenborugh & 0. Unnova Rte Reading i 81 Inroducton . 2137 ; fy eee coe rai 10, Monopotes, Dipoes, nd Quadrupotes 195 83 Impedance 1159 ©. Chapman 84 ‘Analysis for a Single Slit bene wee, 161 10.1 Introduction 195 41 Misty ein aaitiope 622.211 102 Te Tie Dinonal Wave auton 2022226 ot verre fs ina slit like pore i } 10.2.1 Arbitrary forcing. eee eee 196 843 Viscous and thermal effects in cylindrical pore.» 164 | eee oe es 8.4.4 — Low and high frequency approximations... .. . . 165 | 1023 Examples ee 198 85 Analysis fora Balk Absorier’ . seta se ese ss 166 10.3. ‘The Three-Dimensional Convected Wave Equation . 201 85:1 lowzersivtyofastitabeotoer . ss |. 166 1 1031 Avbieey facing om 85.2. Surfeceimpedance ofa sit absorberlayer |... | 167 | isa) ase om ey ee a ner 10.4 The Two-Dimensional Wave Equation 203 854 The phenomenologiea! model and structure factor || 169 | 104.1 Arbitrary forcing non 85:5 The generalised model of Johason and Allard...» 170 1042. Singlesequency forcing |. 2208 85.6 Acoustical properties of stacked spheres. - 1m 105 The Two-Dimensional Cnvected Wave Equation |... 203 8.6 Comparisons with Data and Nonlinear Aspects. hia 1081 Aspley rag ao 8.6.1 Comparisons with data for three materials. a7 105.2 Single-frequency forcing. - "os 8.6.2 Nonlinear theory and comparisons with data, 174 106 Conclusion... ... . 204 81 Conclusions vee vee ee eves 2175 i Problems. 204 ee ee ee eer ee ee erry —Lecture Notes onthe Mathematics of Acoustics i 11. Corrugated Pipe Flow 207 i LW Biot 1.1 The Problem eee TA Theexperimental set-up... eo 2 Experimental results 208 3. Proposed mechanism eee oil 11.14 Other experiments seeee ee UD 112. Pipe Flow 212 1121 Theadmittance . 23 11.2.2 Planar pipe resonance . 21s 11.3 Cummings's Corrugated Model . . 217 114 Further Thoughts 20 Part IV Signal Processing 12, Digital Filters 225 PI Duncan 12.1 Mathematical Overview... . : eee 22 12.2 Fourier‘Transform . : peer za 123 Impulse Response and Frequency Response... 208 124 Convolution Principle . . Bouduoooner) 125 Dirac Delta Functions and Sifting Property... ss... 250 125.1 The Fourier wansform of adelta function»... 231 125.2 The inverse Fourier transform of a delta function in frequency... 2... peeeeee trai 126 Laplace Transform Analysis... . eee ee eae 126.1 The Zetransform. oe =. 233 1262 Thedlscrete Fourier tansform «1s... 234 127 Digital Filters... . ceenee 235 127.1 Introduction... 2. . : ae 1272 FIR filters... .. on 235 12.73 Three point moving average FIR filter. 236 12.74 High-pass FIR filter . A eee 231) 12.75 Infinite impulse response (IR) fers oo 239 12.7.6 Bilinear IR filters (bilins)... . . 240 12.7.7 Bilinear high-pass filter example... oes 2A 12.7.8 Biquadratic UR filter design (biquads) ..... . . . 242 Contents 12.79 Variable Q noteh filter. . 127.10 Notch-pass filter Seca 12.7.11 The bilinear transform... . 126) Summaly eee eee oe Further Reading . 13, Measurement of Linear Time-Invariant Systems TJ, Cox & P Darlington 13.1 Introduction... . 132 Estimating Stasties Using Fourier Methods 132.1 Autocovariance function ‘i 13.2.2 Auto power spectral density function». 13.2.3. Wiener-Khintchine theorem ...... . 13.24 Cross power spectral density function 13.25 Estimating CPSDs cee 13.2.6 Transfer function measurement in noise 13.2.7 The ordinary coherence function 13.2.8 Response of LI systems to random input 13.3 Maximum Length Sequences... . 13.3.1 Transfer function measurement using maximum length sequences 133.2 Measure once 13.3.3 No truncation errors 1334 Crestfactor 134 Practical... , 1341 FFTs 1342 Lis 13.43 Transfer function measurement using noise . 13.44 Transfer function measurement using MLS . . Further Reading»... 2.2... : 14, Numerical Optimisation T.J. Cox & P Darlington 14.1 Introduction... 142 Genetic Algorithms . . 143 Adaptive Filtering 143.1 Basios . . , = 257 = 242. = 243 = 243 244 247 247 248, = 248 = 249 254 256 257 + + 258 = 258 259) = 259 = 260 = 261 265sill Lecture Notes on the Mathemates of Acourtcs 1432 Discretestime formulation of the noise cancelling problem ........ eee 143.3 Adaptive methods for noise cancellation... - 277 143.4 Stochastic gradient search methods... . - 278 Index 283 PARTI Mathematical MethodsChapter 1 Vector Calculus J. W. Eliott Department of Mathematics University of Hull 1.1 Motivation In the absence of any mean flow, viscous dissipation and heat transfer, the propa- gation of sound in a stationary medium of uniform mean density pg and pressure ois governed by the linearised Buler equations of motion ou 0%) PG = VP +E w Hore ¢p is the uniform speed of sound, p = p(x,¢) isthe acoustic pressure and u = u(r, ) is the fluid velocity at a point P with position vector r = OP at a time f. In addition we have considered the presence of both an unsteady body force f, which vanishes inthe undisturbed state, and volume sources 9 ‘We can eliminate the velocity u as follows ‘Thus the governing equation forthe pressure pis, (GE -v') rs = P60 =e —v-r o Bon 30 Lecture Notes onthe Mathematics of Acoustics ‘This is the inhomogeneous wave equation describing the production of sound by the volume source q and body force f. LLL Velocity potential In the absence of body forces, where £ = 0, the flow is irrotational since oe 8 Since the fui is disturbed from an initial state of rest, we can set wxaavx(%)--tvxwnss Po so w= Veir,2), ©) where @ isthe velocity potential, The linearised equations of motion ate then BT AVs = pon ¥ (r+ 5) wget 6 ‘where the momentum equation integrates to yield the linearised Bernoulli equa- tion, Furthermore, it follows that a Boe Outside the source region, wheie g = 0, the Huctuations w and pall propagate as sound waves governed by the homogeneous wave equation, namely GB -*)ees ‘The velocity u associated with the sound is called the acoustic particle velocity, ?) 468) = als a @) 141.2 Plane waves {A plane wave is one in which uaulot, p= P(e, ® Hore a measures distance in the direction ofthe unit vector i, which is (say) to the right. In the absence of body forces the linearised equations of motion are Lap ou Baton. ROA 0) Tess sen that bth pand stay the casa wav uation 1 # 18 & (gie-)ano-0 (Gap aa)we=o aD Vector Calcul 5 ‘These have a general solution p(x, 4) = F(x —egt) +(e + cot), u(z,t i Foes Fle cot) ~ Gl + cat)), a2) where F and G are arbitrary, known as d’Alembert’s solution. ‘This solution represents the linear superposition of two arbitrary disturbances of invariant form both moving witha speed c, one to the right and one tothe left. In particular for a wave simply travelling to the right P= pocu=F(o-at), and Z= P= poee, a3) where Z is the acoustic spedance and = 1/Z is the acoustic admittance. L121 Planar harmonic wave More generally a planat harmonic wave of angular frequency «is ofthe form (ryt) = acos(hy + kay + Rgz ~ wt) + bsin(ky2 + hay + haz — wt) Recos(by2 + kay + hoz ~ wt +6) as ‘aking tane = b/o, with r = (2,y,2) the postion vetor and e = (bx, the wavenumber vector, then we can write a Plryt) = Reos (k-r—wt +e) Re {Roer-utt} a a {Alter} ) where A = Re'* isthe complex amplitude, The points of constant phase, where $(e,t) = e-r — wt = constant define a wave front, and such points satisfy dp =k-dr—wdt = 0, (16) ‘Thus at any given time ¢ the wave-front is given by dg k+ dr = 0, so the wavenumber vector is perpendicular to any wave-ffont (crest or trough). Tf we rove with given crest (or rough) then we must have ap dr caries drow ao ep an where vg is the phase-velocity. ‘Thus the wave propagates in the direction of the ‘wavenumber vector at a wave-speed WW” Vea and the wavelength, the distance between two adjacent crests, is A= 2n/|k. 0 = Ive! as)‘ Lecture Noes onthe Mothematics of Acoustes 113. Spherical waves [A spherical wave is one in which uaubie, p= plrt), (9) ‘where r measures distance inthe radial direction f. Inthe absence of body forces the linearised equations of motion are 18 os ou op Baten (=o, mg =-2 20) It is easly seen that p and u satisfy the (spherical) wave equation (22-H)eo-0 an 18 [200] Oy, 2a 1 a5 | rear) 2) ‘Thus we have a general solution prt) Lire cat) + G(r + cnt} 3) where F and G correspond to waves radiating outward to infinity from the origin, and inward to the origin from infinity respectively 113.1 Causality ‘Usually when sound is generated in open space we insist thatthe sound must not anticipate its cause, the so-called causality condition. Thus G(r + cot) = 0 since it mast have existed for an infinitely long time in the past. This condition is indistinguishable from Sommerfeld's radiation condition, namely tim + (B +02) Ht (ay 1.1.3.2 Harmonic spherical waves A harmonie outward propagating spherical wave of frequency w is of the form A treo) alerw), sy p=Relp'(rt)], where p'(r,t) Vector Cateulas 7 where! ‘ofeg is and A isa complex amplitude. Now if u = Re[u!(r, )] then ae Jemma so i) grtoteor/a ) ; @ which we can write as ? iegrt(e~r/ea) 1 fo? ¢ ais (2) res a8) sat (2) vot-H 28) where @ = tan“*(cp/wr) = x/2 — tan~!(wr/ea) is the phase angle. Thus, in contrast to the planar ease, the velocity and pressure are notin phase. Indeed a¢ +0, 6-9/2 and p! ~ —iwporu’. We also see the radiation impedance ie Poco z S poco. 9) ‘1+ (co fury Ast — 00, 4 — 0 the pressure and velocity are very nearly in phase, with the spherical wave acting like a plane wave with p' ~ pocow! 1.2 Scalars and Vectors ‘Any entity that can be represented by a single real number, A € R, is called @ scalar. A vector, a € R®, is usually defined to be a quantity that has both a ‘magnitude and a direction. 12.1 Rectangular Cartesian coordinates In 3-D Buclidean R°-space we draw, through a fixed point O, the origin, three fixed, mutually perpendicular, lines Ox, Oy and Oz called the a-axis, y-axis and ‘axis respectively, and collectively known as rectangular Cartesian axes Ozyz, as shown in Fig. 11. Place the thumb, index finger and middle finger of the right hand, at right angles in the most natural way. If the index finger points along Oz and the middle finger point along Oy then, for a right-handed set, the aumb points along Oz, This statement exhibits eyelic symmetry in zy, z since it remains true when we replace 2 by yy by 2, and z by x ‘The positon of a general point P may be specified by drawing a line from O to P, of longthr = OP. Ira, fy denule the angles OP makes with Oz, Oy Oz,A Lecture Nolet onthe Mathematics of Acoustics RPspace Fig Li Rectangular Cansian aes, then we define the Cartesian coordinates of P to be the numbers zaresa, y=rcosf, cos. G0) By considering the right-ongled triangles OPL, OPM, OPIN where L, M,N are the feet ofthe perpendicular from P to the 2, y and z-axes, we see that p=slenghofOL, y=slenghofOM, z= +tlength of ON. ‘We refer to P as the point P: (2, y, 2). By Pythagoras avPryee. en 12.2 Geometric vectors [A geometrical representation of a vector isthe directed line segment a = OA, ‘which has both a magnitude (he length OA) and a direction (indicated by an row in Fig. 12) _ Let the vectors a = OA and b = OB represent adjacent sides of a parallelo- gram, This representation is not unique, since wealso have a = BC. The triangle {or parallelogram) law states that ¢ = a + bis represented by the diagonal OC. ‘A.unit yeetor is a vector of unit magnitude, often denoted by a circumex. ‘We denote the modulus, or magnitude, ofa vector a by the scalar a= [also = a/lal isa unit vector lying in the direction of a, The 2ar0 vector O, represented by the point O, isthe only vector without an associated direction. Vector Callus 9 RYspace Fig. 2 A vector inthe plane 12.3 Yeetor algebra Let and fédenote unit vectors in the directions of the 2-axs, y-axis and z-axis respectively, then for any vector a € R®, we ean uniquely write asattaxjtask or (a1, 42,43), (32) (0,1,0), & = (0,0,1). Then (21, a2, 43) isthe scalar lal = Vaj +03 +03. 6s (bss 2563) and a scalar A € RR then Where a1, a2,a3 € R, For examplet = (1,0,0),) = the modulus or magnitude of a vector a Given vectors = (01442503), b= Atb= (a +b,a2+baa3+bs), a= (Aa, Aap,Abs). 34) 124° The scalar product The sala product of to vetors a = (as,02,03), B= (by basbs) isthe sealar a b= aby + a2bp + aby = [al |b cosd os) where if a = OA and b = OB then @ isthe angle between OA and OB. Conse- quently aang t diel = eo ‘Two non-zero vectors a and b are orthogonal to each other fa -b ~y = Io!10 Lesaure Hotes on ine wtathenuancs of Acoustiés 1.25 The vector product ‘The vector product of two vectors a = (a1, 42,03) and b = (by, 63, s) then tak a 2 @3 by be by axb laljb|sing, 67) where the unit vector A is perpendicular to both a = Od and b = OB. Here 8, the angle between OA and OB, is such that when looking along ON, the sense of 8 increasing is clockwise. Two non-zero vectors a and b are parallel to each other if ax b = 0, Also ifa = OA, b = OB represent two adjacent sides of a parallelogram, then [a x b| = [al [b| sin @ represents the area of the parallelogram, 12.6 The triple scalar product For three vectors a = (21,42, 03), b = (bi, bas) and © = (¢,, 09,03) the tiple scalar product, sometimes denoted by [a,b,c], is defined to be the scalar 2: 0205) by by by 62 23 a:(bx¢)=b-(¢xa)=e-(axb) 8) Ifa = OA, b = OB and ¢ = OG represent adjacent edges ofa parallelepiped, then Ja: (b x c) | represents the volume of the parallelepiped. Consequently three non-zero vectors a, b and c are co-planar if and only ifa- (b x c) 1.2.7 The triple vector product The triple vector product of three vectors a = (a3, 2,43), b = (bi, b2, bg) and © = (61,02, ¢3) is the vector defined by (axb) xe= (a-c)b~(b-e)a 9) From which it follows that (axb)- (ex d) = (a-e)(b-4)—(a-d)(b-c). (40) tor Calculus n 1.28 The standard basis ‘The set of vectors {8,5}, also denoted by {61,2 83} and {fff}, is known as the standard basis, Is a right-handed, orthonormal basis set satisfying Piejjok ay (42) «3 12.9 The position vector ‘The position vector ofa general point P with Cartesian coordinates (x,y,z) rla- tive to an origin O is the directed line segment r = OB, which we write as ftyltzk, or r= (29,2) 4a) ‘The modulus r = OP Ir) in agreement with Eq, (31) is given by wae eet (45) ‘An alternative notation to the above is to write x. te GP as the position vector of the point P: (23, 22,23), and correspondingly write 18, +2282 +2965, or (t1,22,09) (46) 13 Vector and Sealar Funetions 1.3.1 Vector-valued functions of a real variable AA vector-valued fuiction F': Rt + IR of a real variable ¢ assigns to each scalar #1 = [e,8] C Ra unique vector ¥ € R, In Cartesian coordinates we write FHF()=u(i+uG)j+wi)k forasi 0 preserves the sense of direction. . ‘The unit tangent veetor ‘to the curve C is given by fora st sb, m ectr Catt 13 ‘where s measures the ae length along the curve C. As ds/dt > 0, 6 = s(t) has ‘unique iaverse t = ¢(s), yielding the intrinsic equation of C, r=r(s) =2(s)i+ y(sj + 2(s)k, O 0, then the condition of an outward propagating wave Ge“, is equivalent to the Sommerfeld condition + (-m2) +0, asr = |x| 00, Exercise 1.2 (D Show that one corollary ofthe divergence theorem is that [fhe [| Bes where 5 is a finite region Z of 3-D space enclosed by the surface S, and in which n measures distances in the direction of the outward-drawn unit normal, A, to 8, (ii) By considering the scalar field @ of Ex. 1.1, and a sphere S: |r — yl = ro, show that ae dA = (1- Aro) e*. [[aper- 0 Hence confirm the above corollary for this G by direct mumerical evaluation fer the case where isthe rion enclosed by sphers r= yl =. band jr—yl = ewhere 0 < ¢ Flg.23. Complex mappings.© Lecture Note on the Mathemate of Acoutice By allowing = to vary, we see that / maps the region D, called the domain (of definition) of f, into a region R, called the range off. 22.2: The complex exponential function ‘We define the exponential function by fl) ee = oF [cosy + ising] an so that |e*| =o and arg(e*) = y. We can then define the hyperbolic functions ty ae ete coshs= (ete), sinhz= 2 and the trigonometric functions by ). (43) 22.3 Multiple-valued functions ‘A multiple-valued function f assigns more than one value to a point z € D. To void confusion we construct a branch of a multiple-valued function j, which is 4 single-valued function F that in some domain D takes only one of the values of f(z) for = € D. Often all points z on a given line, a branch cut, are removed from the z-plane in order to define the single-valued branch. Forexampleif 2 = re! where ~x < 8 < x, then the square-root function has two possible values, namely 2 ye!9/2. Choosing only the postive value, taking Fell? forr > 0, — ina {FOr eroacrctee aw o for z= 0, then this (single-valued) function Fp is well-defined on the entire complex plane except along the ray @ = 1, the negative real axis, ‘Any point that is common to all branch cuts of f is called a branch point. In ‘general z = 2g is a branch point of f then for some rq > 0 S (z0-+ roel) #7 (20+ roelo*?")) @s) 223.1 The nth rot fnction For integer n > ( consider the n-valued function fla) = 0 = rine, 6) “3 mig Be ry ‘ Functions of¢ Complex Varcble 41 where z = re”, Here = = O is a branch point since f [rly] = primi = cPeIn FOr) sir). ee) ‘To make w = 24/ single-valued we eut the z-plane, between the branch point 2 = 0 along some ray 8 = ay and forbid z to cross this baer, effecdvely restiting @ = argz to the range a —2r <0 O-nO—r 0, n 1, at the point 2 = 2, Let ybe a smal ctclar anticlockwise contour, with cents and f= : 06) radius § < 1, connected to C by the siaigh line L: A —» B. By considering the closed contour’ = C+ ~~, we have from Cauchy’ theorem with residue 2. Alteatively, from Bq, (0), the residue is given by ee fedz= ($+ f—$-[) teaz=o, Aan tay (oF r0e] = tm $ [E54] = fp =] f (f+f-£- [so 20 3 3 since f(z) is analytic at all points inside the contour P. Consequently = Inga on ‘Again at 2 = —2 the residue is given by toa freee aon a In the vicinity of z = 2, the function f(2) has a Laurent series = Goer t tpeaytetaGadt, 25 Residue Theory (102) 1 is the residue of the pole at z = 2. Consequently fiteyas 2 on fle~ayPde= Qrir, (103) since from Example 2.2 we have the result 28.1 Singularities From Liouville’ theorem, the only fonction analytic everywhere, including the f flay point at infinity, is constant. However entire functions, such as polynomials and ic Cxponentils ate analyti at al pont of theft comple pane. “A singula point is sad tobe isolated iff i analytic at every point of some deleted neighbourhood, 0 < [2~ 70] < of zp, Wecan always obtain a Laurent eo Series expansion about an isolated singular point, Non-isolated singularities in- fe-arar= 04) clude branch points, for which the branch cut is an exemple of a singular curve, I otherwise. tnd ester pon, Forexauple 0 isa cluster point of (2) = eot(n/2) us Ope ; general ase we can take, to be the small cirular antcookw tothe sequence of isolated singular points at z = L/n for n= 4,22... contour surrounding the singularity» = =, = 1,2,...,n. Weean similarly7 “Lectre Nott onthe Mathemadles of Acoustics show that o Example 2.5 If C is an anticlockwise simple closed curve large enough to en- close all the singularities of the integrand, evaluate (106) Solution The singularities ofthe integrand are simple poles at B45r=2(242)=0, 90 aon The residue at z = 0s given by n= lyf@) = lp 2S 4 (108) The residue at z = 421i given by 3-61 y= lim (2—20)f(2) = tim 2—2 42}. 3—8E yey gy 2 = ding (+ — 21) 4(2) = Jima, Gay = HO - 8) (108) The residue at s = ~21s given by . . 846 sage rae lm, (242) §(2) = lim, ay = Taneay = HEHE aig) ‘Thus the sun ofthe resides is given by Son =~} = 6-65) - Ho+ 6) = an) Consequently Parl tf a (112) 28.3 Evaluation of real integrals Suppose we wish to evaluate the rea! integral I eC Gamer para a3) Funcilons of ¢ Complex Variable si We first consider the complex contour integral ae ioe f aanerray oe where C = L+-T'is a simple closed contour taken in the anticlockwise sense, shown in Fig. 2.4. Here C consists ofa straight line segment L along the a-axis, = ¢ [-R,R, and a semicircular are C with centre 2 = 0 and of large radius R>1 v seplane we ag =R a R ib Fig.24 Contour integration, Although the integrand has four poles at z = kia and at z = -Lib, we see that the contour encloses only the wo poles at z = i, 2 = Ib, The residue atz leis piven by 1 ni lim ~ al) 2) = lim Sayre We) The residue at z = ibis given by = lim (= — 81 tim n= in -W/@)= ln oraere mel OO ‘The sum of the residues 1 Bee? — Faxat ey 17)2 Lectre Notes on the Mathematics of Acoustics ‘Thus ff0)8=95 sarery) * aeED Now we see that since C= L-+T we have fears [rears [ separ. On the sogment L we have z = 2, dz = de where ~R 1 since the nth term does not approach zero as 1+ 00. We can also consider the function #1, (128) ‘which is defined and analytic at all points except z = 1, Moreover fale) = f(z) forall je] <1, a9) since Bg (127) isthe Taylor series representation, about z = 0, of Eg (128) inside the unit ezcle, Infact we can state that f(z) isthe unigue analytic continvation of filo the domain | > 1, = #1, This follows from the following theorem. ‘Theorem 25 A function that is analytic in a domain Dis uniquely determined by its values in some interior domain D! CD, or along an arc P interior to D. If 2p isa point on a curve T € D and zy is any other point in D, then we can construct a curve C from zy to zy, with intermediate points zp, k= 1,2, .+.yn—1 such that |2u—24-a| < d, Where dis tho smallest distance from C to the boundary of D. Since f(z) is analytic in every region [2~ z4| < d we can construct a Taylor series using its value at z = zy 6 obtain its values at z = 2,1. By this repeated overlapping of Taylor series we can uniquely determine fat any point in D. Example 2.6 Consider the function f, defined by a= [° which is defined only for D,; Rez > 0. However we can also write at, (130) fle) | 7 2 when Rez> 0, «anu Lecture Notes on he Mathemaiesof Acoust Let fo(2) be defined by the geometic series nani (EY, etic. a3) & “Then within the circle Dy: [z +i] < 1 the series converges to falz) when |z +i] <1. (133) Trey Evidently fa(2) = f(z) foreach 2 € Dy 1 Dz and so fr isthe analytic continu- ation of fi into Da. Furthermore the function Feat, 2x0, 034 js the analytic continuation of both f and fa into the entire z-plane for z # 0. 2.6.1 Chains of analytic continuation Suppose we have a functions fa, fe and fs analytic in regions Dy, Dz and Ds respectively, Suppose further that Li) =fo(2) im DMD, and f(z) = fol) in Da Da, (35) so that fz isthe analytic continuation off into Dz and fs isthe analyte conin- uation of fp into Ds. Then we cannor conclude that falz)= f(z) forallz € Di MDs (136) because Ds, Da, Ds might enclose a branch point ofa multi-valued function. Example 2.7 Consider the branch of 24/2 flz)=VFe"®, > 0, O0, 1/2<8< 2m. (138) ‘An analytic continuation of fo into the first quadrant is fle)= vie, roo, Beoc - ass) Note that fa(2) # fa(z). Infact fa(2) Functions of « Complex Vrlable 3s Fig. 25. A possible aangement ofthe regione D1, Dz and Ds, However suppose we analytically continue a function along two distinct ‘curves O; and Cp, to a given point, so that C; — Cp is a closed contour. If there are no singular points in the region D enclosed by C ~ Ca, then the result of ‘each analytic continuation is identical and we obtain a single-valued function for allz€ D, Problems Exercise 2.1 By using Cauchy’s residue theorem, evaluate I ahell dz, fo: etm for intéger n > 0. Exercise22 By using Cauchy's residue theorem, evaluate zo 242 @ n=§ Gow © n=$ ea. ics |alag 2 +1) OR Fe en Het Ty Exercise 2.3 Suppose that the rational function f(2) has a simple pole at 2 such that FE where h(z0) = Oand g(20) #0,56 Lecare Notes onthe Mathematics of Acoustics where g and A are polynomial functions. By writing A(z) = (z—z0)R(z) where Rlzo) #0, show that the residue of f at 2 = 20 is given by (20) (Go) Exercise 24 By using contour integral methods, evaluate the Cauchy Principal Value Res( f(z) : 20) = np dz = Jim ee An) Ht Note that this may difer in value from ‘ «at in fi a L. eae ee Ee aca eee where «is some finite number. You may use the magnitude inequality, namely that |, neva s [iret and |z—a| > [2| — [al Chapter 3 Integral Transforms 1.W. Elliot. Department of Mathematics University of Hull 34 The Laplace ‘Transform ‘The Laplace Transform ofa function f(t), defined forall ¢ > 0, is given by L{F()} = [reeves ® provided the integral converges. The Laplace transform converts f(t) into a new function of the complex transform variable s = o + iw. ‘We say that f(t) is of exponential order fort > 7, where T > 0, if there exist real constants o!, M > Oand 7 > 0, such that |f(t)| < Me® forall t > T. Fora given fo is not uniue bt o> op, where op (he greatest lower bound fore’) iscalled the abscissa of convergence. ‘Theorem 3.1 If fis piecewise-continuous and of exponential order on t > T > 0, with abscissa of convergence og 2 0, then its Laplace transform, defined by Eq (1) exists but converges only for Ro(s) > av. Proc, ‘The poof of existence rae echnical bt from the magnitude in- (ote equal if = + hen e= cone weot=|[" soetae s [veo * ee Se GO5 Lectre Notes on the Mathematics of Acoustics Ife scum that fs absolutely integrable on 0,7}, 50 that 1 7 1) = [ toierrtars f \reniae 90 Q ‘Thus for some functions, whether it's because oftheir rapid growth as ¢ +00, for example f(t) = e, or because of theiintegrability, for example f(t) = 1/42, their Laplace transform does not converge for any value of Example 3.1 For integer n > 0, we have for Re(s) > 0 ere [ eetaa [rear = @) Bxample 32. For complex, we have for Re(s) > Re(v) elena [etre [heen © ‘The Laplace transform is linear operator, with L{af(t) + Balt)} = aL {f()} + AL {o(t)}, © for constants and A. It follows that, for Re(s) > 0, we have L{eos(64)} = £ ( [ett + ey} =} [ee + al ae @ Lfsin(@t)} = £ if [ew = aa peeeel Similarly for Re(s) > |k| we have £fcosh(kt)} = =" woe Lfsinhen} = 341 The Heaviside step function ‘The Heaviside unit step function, H(), is the discontinuous fonction _ ft tort> 0, cen = {2 tet > 07 wo={y mice Bodet-mon{? B12 0 Inert Toons ° With this notation the Laplace transform Ba (1) may be writen as cu@h= fea f° most at an We say that f(s) is the two-sided Laplace transform of H(t) f(t). ‘We say that f(t) is eausal if f(t) = 0 for all ¢ < 0. The multiplying of f(¢) by H(t) converts f into the causal function H(:) f(t), which, unless F(0) = 0, is Giscontinuous at ¢ = 0, It follows that for a > 0 coate—a)= [Pe * ot [Pere nemcay=te a2) 31.2 Properties B21 The sale ride ctstee) = [regener = 3) 31.22 The Ist shifting theorem £{ ste} = [ * F(t)e Oe at = f{e—a). a 31.23 The nd shifing theorem Fora > O we st that Lf F(t~ a) H(t} = as) 312A Muliplying by > cresoya [PP ertipen* tero}= [erecta Ro eeyk C (te at) (16)o Lectre Notes on the Mathemate of Acoustics Example 3.3 From Eqs. (14) and (16) it follows that L£{ain(24)e~*} = oo an L£{teosh(8) = ate ag) 7 2 6 £ (Peony) = & aac ee 3.1.3 Differential properties We have, upon using integration by parts, ciroy= [° = [(0e"]P +s f ” spent dt = sf(s) 40), 20) cto) = [Shona = [roe wef roe =sl{FO}- 0 =F) -sf0)- 7. aD ‘Thus, by induction, we se that {22} fo-Se =hf(0) @ Also for a function of more than one variable {ena} = [Slower 2 (J rremes) 31.4 Convolution Given two piecewise-continuous functions f(t) and g(t), defined for all & the convolution of f and g is defined by sata |” tnae-nar= J” se-naear= ato elf. 0 Integral Transforms a ‘The notation (f + g)(t) = f(t) «g(t indicates that convolution converts the two funetions f and g into a third new function of t. In addition to being commutative, convolution is linear and associative, since for constants a and 8 Jit) +[ag(t) + AACE) = aF(t) + 9(8) + BFA, 25) £ (8) # fale) + A@)] = [FE # (0) «AC. (26) In particular, if both f and g are causal, so that f(r) = g(t) = Oforr < 0 and f(t=r) = oft 7) =O forr > t then Ba, (24) reduces to £() * ot) y= ff F(r)a(t- 1) ner ff F(E—r)9lr) ar = a(t) * fe). GN Henceforth, when discussing conyolution in connection with Laplace transforms, wwe shall assume the functions involved are causal 2.5 The comolution theorem Theorem 3.2 [fF f(t) and g(t) are causal and defined fort > 0, then ef [ soe ner} = 2170) +000) = £1709) 2 tO} = Hoste (28) For example, for any causal function f(), the convolution theorem Eq, (28) yields the intogral property e{ [roe} =e Ho) = cco = 1. 3.1.6 The inverse transform We take the symbol £71 { fo} to denote the causal function f(t) whose Laplace transform is f(s). That is, provided f is continuous £0} =f, 50 efi} Othe Ia shifting theorem gives i AY gat 4 pn [st Or 7 4) =£7 {()} =e {4} eee {% +} a2) From known transforms we have, fort > 0, the result te + [Cy cos(-yt) + Cz sin(t)] e*, 3) 3.1.6.2 Convolution 1 we ake the inverse transfom ofthe convolution theorem we obtain Ee OLL(O} = £7 {FE} = MO *9). 4) For example suppose 5) 46) = Gayer5 then upon inversion, we have for ¢ > 0 Ue) = 109 Ae = Her) [P= MP] a, 8) 31.63. The 2nd shifting theorem TE we take the inverse transform ofthe 2nd shifting theorem we bave FO) = £7 {G(e)er*} = oft a) H(E~ 0) en Bxample34 Solve the pari dferential equation ou | Ou St petusst ine>0,t>0, Rt ge tent >0t> 38) subject to the conditions u = Oat 2 =O and t= 0, Solution If we take the Laplace transform of the PDE with respect to t we obtain a linear 1st order ODE for (r,s), namely Ba (e+ em sce) +ule,0), 50 evans 2 [eeeng] = 3 a9) Iniegral Transforms a where we have used an integrating factor. Using the fact that @ = = 0, we have the particular solution - 4 enlettie 7 8te,9) = [eter 1 1 tern] = seen {*- gayle »d} (40) Resolving the rational terms into partial fractions gives (2,8) <2 [3 aoe wl s* GD = (1-ees) [a-3+ chy tad ay ‘Taking the inverse transform gives us u(e.t)=2 (tLe) [7 -e*Ft-2)Ht-a)), (42) where ~24(Q4t}et (43) and J) ft 2) He -2) = (t-24 B+ ier!) ~e°* [fe 2) 24. @Ht— ayers] sa(t-1 +e) - [t-2- ¢- 2-2-4] 4) Thus u(z, t) = u(t,a) fort > z, since wien {50> 1d bos oe fort <2, i He 14e*)~ [2-24 (24s)e*] fort>s. 32. The Inversion Theorem ‘Theorem 3.3. The inversion theorem states that if fo=ciro)= [jee = [Ls0xee*a, a6o Lectre Notes onthe Mathematicr of Acusice where the integral converges for Res) > oo, then ify > op i fyghab [oO fyettas = HOLM (Ho) = 5m [howe ee SHEE, 4 opening on whether fi continuous ator not. In partly fort = Othe integral converges to 37 (0+), 32.1 Evaluation of the contour integral Generally #(s) is an analytic function of s except at isolated poles at s = sk, and at Branch points at s = ay all of which be in Re(s) < ap, In general wwe tke branch cuts to emanate from s = a, and lie along the stsigh lines eels — 25 “Typically we considera closed contour Ca, which consists ofthe Bromwich contour s = 7+ ir, —R Sr < R, whichis closed by a semicircle Sp of radius FR,on the right fort < O, and on te left for > 0. Then, by Cauchy's theorom dif torent ( [t+ since j(s)e" is analytic inside Op. For t > 0 we have f foetas (Rohe) Fleas, 49) where Rs sufcintly large to enclose all the pole, and the loop integrals weap around each branch cut. Thus, by the residue theorem mf, f° a= Dn fort > 0, (50) where rp is the residue of the kth pole, It is standard practice then to show that the integrand is exponentially small on Sa, implying that \ haters totes, 48) Hisjettds +0 as R00. 61) sn Here use is made of the fact that on Sig, for R >> 1 wehave s = sothat 7tRel ~ Rel record _, fevfitles® +0 int <0, cos0 > 0, ~ len Atleordl 49 int > 0,cos@ <0. ( (o mcg eset “s Tethen follows that 1 HR » 1 fost) F(t) H(t) = lim (sje ds = So ry = ds, Om a Ai | ig Homtes= are Fle) as (63) Example35 Solve the wave equation Pu 1 Fu Ramage A> Oa 6 for constant ep > 0, subject to the boundary conditions Fon nd initial conditions a, and u-0 atz—oo, forallt>0, (55) u=du/at=0 at forall r, (656) Solution Taking a Laplace transform of the differential equation we have Pa S on for (zx, s), subject to the boundary conditions a=fls), 2 0, azo, (58) ‘Thus we have a general solution Ax, ) = A(s)et/ + B(s)e™*/ 6) Clearly A = 0, to satisfy conditions for x large, and B = f(s) so that, ife,s) = fler*!, 50 us,t) = [ 7 dlsyet* ds, (60) which, from Eq, (47), we ean interpret as u(a,t) = H(t - 2/co) f(t - 2/ea)- oy Example 3.6 Solve the partl differential equation au ou Rte (@) for constant on > D,subject to the boundary conditions u=2Up (const) onr=a, and wis finite atr = forall > 0, (63)Axquope s2qng Su, of poanouodss xmduoo ayy sg solos jouSu0 oy uot Lams Bujaq aouasanuo YM, “suoppuos s rqyoinic saystes 3) f 31 KYuo (2) f 1 soBs0Atoo sexes patnuoiay -hp oy hosronuoz “souos youd u0 amp ven tsof(9)f3o keane ay 01 soBran “woo sopas poweoruyotp‘uttoy Aq as (3) f Jo soyas Jato oxy spo4Bequ om JT uoqpeusonn pun uormmenfia yee + Ase gnturunp su soop yorys Jo apraueur sip ‘uousuiouoqd sqqyg ai se usoty Ynunuoosi zo tod a 2 oousepuyiooysiono Burkuedusooge we Go ‘muoosip o med ap seu somo sf aUodsoALOD (,_u)C az %q pub “o atp ‘snonanuossp st (af BOE." '9 = 3 15 sORUTNOD St yf Hg "38 404 (,1)0 30 %4 poe Ho gyn moss ouaBsaquoD mn snonutuossp st 3 3190 snonuuod st f 101 fo soysayen uodn Suqpuadsp ‘Ws + NE ‘OF, 00 — jy $0 7042 septs adoayy S.1o4anag ays (92.62 Aq u2N}8 24D “9 puo ¥o (Pup ‘suompuon s sayiuig Suklsnos uonsunf qipoued vst (fff pre waxoDKL, (98) } = [(eema)ups 29 + (aemu)s00 Yo] TAZ + OO ‘Qynuanuoosip oyuy Jo siufod pue vunqurw pue wusnrem pareost Jo Joquinu ony v sey 9) ‘porsed ouo Aue u's! suoRypHOD saoqmpEN Asses (©) pos s| wonoung oxpotsad papuinog ¥y “(3)/ uonouny au 01 sadsoAuod f 30 sotI0s Jounog oup yey sozuerend ‘29H ou stg pu Yo squayoys00D Jo aOURISTXO OL, sourBioweg eee (50 (emu)uasg” Km (8 “(gmu)soo aL oe OOF = (a) og ago su08 sting ony“ sypoued ppo we pur '(-)F = (af yom w Q)y 6 aro (9)6 wowouny nowny sfpouod aKa wy suowounf uoas pu eyo Te e suofunaypstosy (sh (0/4) =~ eA = ? Mp ups ty— = Mg "Hp so0 ty = Mo sprork (LL) “ba ta (68) “ba Jo wosteduioc “pquomey (pu sapres ap Jo aseyd pue apmyydus stp axe *p pus 9 < “yr azay (a) “(4p + gemi)s00 “y+ oy = (I) oy orp ut (11) “ba sou98 soe om SHE STE UBD OAL mommy Pee , cru 30y = FP (em) (rnu).00 cc) 2mm (imuryoo odustmy oft stmujas a a 3} momangrayae Lf oxusmsy c/s =p (no (gunn oa) wuss i} selon Graon | ‘Ayourwu‘suonouny smawouod en ain soy suonepsy AqeUoSoMpI0 o1p yA zayyeBOY as POEM TY Lo omujsco | 1p (arm) * af i ap (gnu) > a { ‘neu 0 Jo 9pour 967] ‘Sunesfonu pur “us roBoy owos soy “(yuu 30 (ymu0o soy 6 (La) "ba Busters kq smoyoy (31) ba a1 Aremyg sp Jo aotous amp azeyHA L , 1 ae ‘ baw ao worms [Ea wameooe [¢ @ ‘omen ‘eqns spo 9 woAS are 9 pus Yo suoyoyoo JENOg aA AHL w [Gruae%4 + (u}e00 0] “C 4-90 = (DF ojsuedxo s9p06 somo 29 sey p/ug = m Kouanbayy aejnSuw pue \z, poyrod Jo“(2)f uonouns o1pouod y sopog piney CC aumoay Jo younyroy 4) wo 0 207] »ou i (gz) as Saag me = (PUT = G20 uso aupsar O05 wo «(gaa (E) ae ya = Hem) + (Moog Suonbosven wt = (xu 2800 = (eyo uo wD “eppoan® (oa) a 6 pepo OO sneeau (sou) 8 Tem Gas = whan Tareas aos = (uui)ysco nuts (00/04) 4909 84 oyosul Go aaua)uus Og ~ (Go/Pe)qus Hy 0 st Ys = 8 yw ompisas orp sau, CIs T exp esas om He = s yw sojod yp sapisuos oy, “Og st q = ev onpysasoxp Smt, 17 dur) wopesBe jo momo ay, 16 I (e)oa subs teh (yay suofowon sory Ww) Wordge={lno+ nee +1} z jaye **} owound op soy ponstonu oxy g'= 9 sayy (3) ue = (0)6 ary osqe pur 9 = s ye sojad ojdans sey puesBauy oy, toa Cec =tsp=8 os Gfayass tay f =O my, "20 + 45 ajifou sone fg way worngundc ust, ‘00, Y se [[euIS Ajpenusuodya st puesSoqur axp ‘Q > (s)oy pue T <& yr ~ [| 205 “Te Bajo enous sy super sm J fg apes wf ot ovo pes a SS ay arbi = eau eons ago Supytoo Wp Into psoas) i spp [anaes | 8 8Of oc cofosqs tJ 8 rag (fees +t Dacre! 1 (a) Uonnjos any aM “(1») waxoonn woyssoxur aun Aq a0u0H] (39), spjett = 48 uompuos ay “0 saxnbar g wonuy 2 Assnes oF uw) [(S)s8@e+ (3) or] sedate opin resoua8 & ney om snus /2 Ads st jax sejnonwd y 69) = ta = (tae}7 swomy suopypuoa Azepunog axp 01 309fqns “(sm 0p (9) ‘key 2m uoHenbe jenuarayp ou Jo uo;suEN sop[de] e SuNyRI, WORNIOS os) meso <1 gan suonspuco jensur pue s2uone3y fo soousney 9 wo 210M HE 80 Lecure Notes on the Mathai of Acne ‘we can write the Foutier series expansion Eq (77) inthe form J(0) = feo +3 Alon ~ tba} + Be + iba)emI™] 8) where 9 = fo, en = of and wn on = H(dy — iby) = rf F(8) [cos(nwt) — isin(nt)) dé 1 er apf Moerma. @) Recalling the harmonic series Eq, (83) we see that n= H(Qn— ibn) = Amel? 80 fen] = An = VaR +0R. (90) 3.3.5.1 Parseval’s theorem ‘Theorem 3.5 Ifcq and da are the complex Fourier coefficients of {(t) and g(t) then DO orden = D> endl. eo» ar FL Hone Corollary 3.1 koh +2 >> (02 +82), 2) where we ell the lft hand side the root mean square (RMS) value ofthe periodic function. 34 The Fourier Transform 34.1 The Fourier transform pair The Fourier transform, of funtion f(x), defined for ~co 0 find the Fourier transform ofthe causal function Fla) =Hlz)z"e"™, — (@> 0). 8)n Lectre Noes on the Mathematics of Acoustics Solution Clealy f satisfies Eq, (05). From Bq. (97) and Example 3.1 we have gues Serre eS nl Flatayeterss) = [steroids = 61") = 9) [Brample 36 iad he Fourier transform ofthe non-ausl funtion f(z) = nail a>d. (100) Solution Again satisfies Ea. (95), We can interpret Flew} = [eee as a contour integral along the real-axis in the complex z = z + iy-plane, Thus ie [7 eaeeuraras, on Fferet} a erete oter ay ate weet [NM cacanreras, a eottyte Here we have used Cauchy's theorem to lower the contour into y < 0 because the imegrandis analytic (everywhere) and decays exponentially fast fr |2| > 1. ‘Thus (103) where w = 2-+ik/2a, v= YGu. Given the well-known result ve=ryy= [Pwirerae= J" ea, 09 where (2) is the gamma function, we have the transform \ Beeps Zorn 08) mcg anon 344. The duality property For the Fourier transform pai Ea. (93), (4) symmetry yields the property FUCA, 9 F{Fla)}=2n/(-H). (106) Example 3.9 Find the Fourier transform of, @ fe) Clile+0)—Hle-a)], ste) = “Psin(as) Solution (@) We have Fle) [cetae = FE (ka). 07) (b) From the symmetry property Bq, (106) we must then have {2 sin(az) } = 2nC [H(-k +0) — H(—k — a)] = 2n0 [E(k + a) — Hk a) 408) 34.8 Further properties 345. We see that The Ist shift property F{stayer} = [7 geese = fea). 09 345.2. The 2nd shift property We see that F(pler—1)) = [” flor —He-Hae ; [i senerteer =itb/a (—k) and F {f(a — @)} More simply F {(—1)}" Lecture Notes one Mathematics of Aeaisticr 34453 Multiplication by 2” ‘We see that F{a"4(e)} In * an p(ayertas & ([° nowmes) any ample 30 Obtain the Furie transform of f(e) = CfA(2) — H(e ~ 20)] a1 Solution Using the Ist ift property Bq, (109), we have = Hla ay) = 2. FU} =F (Clie +0) he sinha). (113) 36 Differential properties Fie) = £ emeds ik ie Fl@e Repeating this argument n times, gives {e) = [@e™ ede = (ik) F(R). (114) KY" F(R), as) 34.7 Parseval’'s theorem ‘Theorem 3.7 Given that F{f(z)) = #(l) and F {o(z)} = a(k) we can show that @ —— [ teygner= 3 [” Fwa-ner, a9 0) fC rew-nae= xf” Twoawman ary Corollary 32 By Parseval’s theorem the constant p= [ ve@t a= 3 : * FU) FR ak = * L7G? ak, (8) ert Trango since for f real, F(—K) = F*(k). 34.8 The convolution theorem ‘The convolution of two piecewise-continuous functions f(x) and g(a), defined for all is once again defined by Eq, (24), that is eave) = [7 selotemrar= [7 fe=naledar = (oe). C19 ‘Theorem 38 The convolution theorem states that Fureeaen=#{[” serate-nar} =F UO} F (ole) = F000) C20), and correspondingly F (s(e)ote)) = & | feate-nar 349° The delta function ‘The Dirac delta function is a generalised function having the properties ‘eo! ifs=o, - aa (0) ife0, 33) subject to the initial conditions Se), and Buf Solution Taking a Fourier transform (in 2) of the equation yields the problem att=0, forall, (134) eee wa (435) for G(k,t), subject tothe initial = ) att=0, (136) 9 u Integret Transforms n ‘The general solution ofthe transformed equation is given by ‘Uk, t) = A(k) cos(kt) + B(k)sin(ke), 37) where A and B are arbitrary, To satisfy the initial conditions we require A(k) = J {h) and B(E) = 0. Thus we have the transform solution, “G(k t) = F(R) cos( kt) = $ F(A) ‘Thus by the inversion theorem (94), we have the solution wena [704 nee Ly; aga eet apy 2 e-8 iz [7 ims ak [FO ax} (138) Which i Alember's solution othe wave equation, namely [e+ +fle-a). (140) ) fel + oe] (138) Ie 4 oH) oll a ua, 34.10 Further extensions So far we have taken the transform variable k to be real. A complex k= a+ ir, where c= Re(k), 7 = Im(k), allows us to consider functions which are not absolutely integrable inthe usual sense, ‘Theorem 3.9 Suppose that f is real and, for real constants Ay. and ri, satisfies [f@)| 0, aan uO asttoo, and w= f(z) att=0. (148) 38.1 Fourier transform ‘Taking a Fourier transform (in 2) ofthe differential equation yields (149) subjet to the inital condition {s@)} =F), a From Example 3,9 the transform solution satisfying the initial condition is given by \ wie} % Ce) (450) L ant whe) Foyer = 7 gtene { ‘Integral Transforms @ ‘Thus by the convolution theorem (120), we have Sal © payee ta, 152) u(x,t) 352. Green's function If f(z) = 6(x — a), then the solution is given by u = G(x — a, t), where. ay. 2 ’ Ai focttant) 1 tata sy fe }e5 A 53) Where G is called the Green's function. The initial condition, namely G(x ~ 4,0) = (x ~ a) confirms Eq. (123). Further the general solution Eq, (152) is siven by le ~a,t) = vini)= [see ee os 252s anon fore #a, form =a, 2 Fo st 1a) =e a) = {° . 55) ‘oo! subject to i ~» Os x + :boo. Indeed any funetion G satisfying LG(e — a) = ~6(2—a) where Lis some linear differential operator, is referred to a8 a Green’s funetion. Solving in x ~ a > O and « ~ a < 0 separately, yields Ax, s) = Aa (sera (156) ‘where we take s!/? to have a branch cut along the negative real axis, so s = re, = <6 0 we close to the left as shown in Fig. 3.2 to obtain the integral 1 aod acids = ah, afr tht Leth where on Sp, 6 = Rell, $ <0 < —m <0 < ~Fandon Os = eo, ar <0 0 because of the poles on the real axis at fa Fig. 3.3. The contour of negation for Example 3.14 = sx, These poles can be avoided by indenting the path of integration into the complex h-plane to pass either above or below the singularity, Four different solutions canbe obtained, Bt only one exhiis outgoing weve Behniour If k= ky + ik; then yen i FED 0 as|k|soo, if (nz fre>% am hy SO fora y we can close the contour by a large semicircle in the upper half- plane, where the integrand decays exponentially. Only the pole at & = ris en- closed by the contour, and hence i x Be t=—9) 179) Bai x 5 079) For a < y we can close the contour by a large semicircle in the lower half-plane, where the integrand decays exponentially. Only the pole at & = —r is enclosed by the contour, and hence lon where the negative sign is due to the contour orientation. Therefore iRes(e) forz>y) 4 geeoal extend {TRO gmeciy= Ree aan Example 3.15 Obtain the -D frequency domain Greens function that sass -(¥? +5?) Gale.) = dle -y)- (1s) Solution From Eq, (173) we have alles ? akdkadks. (83) ‘Taking spherical coordinates (k, 0,6), defined by ky =ksindeosé, fr =sindsing, hy = eos, (184) with @ measured from the source-observer direction so that Galrysu) Kk: (e-y)=Klr—yleos8, and dky dig dky = A sin@dk dd dg. (185) Hence we have (186) Integrat Transforms s Integrating with respect to @ and ¢ yields yu) x byee [M2 _#_ finest Cale) = ali [aes [Soa 3 ~ oral wa eas ee K__ uie-y1 oral ae dk «as7) Following an identical approach to that adopted for the previous example, we ‘void the poles at k = en by indenting the path of integration to pass below the pole at & = x and above the pole at k= —n. Here, though, shiv, cage Since |r —y| > 0 we can close the contour by a large semicitele in the upper half-plane, where the integrand decays exponentially, Only the pole at k = ris enclosed by the contour, and hence A hlovlag = = ile-y Le ayer ak = aeiRes(x) = ani x fetin-rl, (180) ‘Therefore Galt,y,) = wo (nic-2!) “2 Example 3.16 Obtain the 3-D Green's function that satisfies eM-¥l, (190) = (V8 +8) Galey) = dle —y) = (2 — a)ély ~ ge) 6le — 2 inz>0, (isn) where here y = (2, Ys, 2) and z5 > 0. In addition we must satisfy the Sommer- fold radiation condition of no incoming waves, and atthe solid interface %e ate 192 Solution —() The simplest approsch isto recall that Galny) = 28 ay) = EE + Hey), (093) where (W+2)H@y)=0 in 230, cis) 20°38 8 a 36 Lectre Mites on he Mathemotisof Acoustics 0, 199) subject to 6@)/0x = 0on «= 0, and outgoing wave behaviour Solving for in > ay and 2 < 25 separately then requires imposing the conditions st (azt=0, iz ey, (200) ‘The first condition ensures continuity and the second comes from an integration of the equation for G from 2, — € to 2, + ¢ in the limit as ¢ — O+. A solution satisfying the condition of outgoing waves is given by a [Aaeesl ee a. {fe fors> ku «con Aa{ett + Ueto MY ford 0T(n + 1) = nl, the factorial function, Indeed upon integration by paris we see for Re(z) > 0 retys [rere terete es [> Exercise 32 Solve the following model for sound propagation near to outdoor round surfaces by describing the field due to point source, located atthe point = (0,0,2,), in a semi-infinite fuid in 2 > 0, with a semi-infinite porous ‘medium in 2 <0, Tn the upper medium the equation for the acoustic pressure, D ) (+52) G¢) where 0, Integral Trenaforms » ‘where « = w/co is the acoustic wavenumber, and here the time-dependent factor 7 is understood but suppressed throughout, Inthe lower medium, the govern- ing equation forthe acoustic pressure is (VHK)G=0 inz 00). ‘Many problems which arise in acoustics can be posed in such a way. “Asymptotic methods seek to find the dominant behaviour of the solution t6 problems for regions where a variable or parameter tends to some limit (fnite or infinite), We begin by considering order symbols and what is meant by an asymptotic expansion. Given two real valued functions f() and g(z) defined in some neighbourhood ofr = 29 we say that Jim OS = e F(z) =o(g(2)) asta, if ‘This is pronounced as “fis litle oof gas = tends to 9. We also define a relation “7 is big O of gas tends too’, writen in symbols as F(z) = O(9(2)) and which means that there exists a constant, M, such that |f(2)| < MIg(2)| for 2 ‘lose’ to 20 (i. [2 — 20] near 0). ‘We also define the important symbol ~. For two functions f() and g(z) we write f(z) ~ 9(z) as © 0 if ims ze | (@)/9(2)| = 1. [Note that zo may be zero, finite or infinite. ase 20, 3) + f(@) = o(1) as 2 co means that f(z) ++ 0.a8 2 ~+ c0. # {{2) = O(1) as 2 — oo means that f(2) is bounded. If fla) = bz? — 72 +9 then as 2 —+ 00 we have: f(z) = 0(2"), S(@)~52%, fle) = ofa"), © “The symbols 0, o and ~ allow us to compare functions in tenms of ratios rather than in terms of differences. If we use 52? to approximate the function Sc! — Tz +9 (as.2 — oc), then we have captured the ‘dominant’ behaviour. Notice thatthe absolute error involved tends to infinity, while the percentage error tends to zero, Tis i typical of asymptotic approximations. ‘A sequence {gq (z)} i an asymptotic sequence as ray if forall m Gnt1(2) = 0(Gn(2)) a8: > so. © Likewise, a series "5 angq(z) is an asymptotic expansion of the function Sle) as2 + agit S 3 |p0)-35 enontel]=olan(st). zz.) Asymprotc Expansion of negate 93 If Dar Gngn(z) is an asymptotic expansion of f(z) then x $(e)~ > angela) ® In the remainder of this chapter we will consider methods for obtaining asymp- ‘otic expansions to integrals. Generally we will be content with finding only the first, or leading order, term in such series, 42 Two Elementary Methods ‘wo simple ehnigistatcan cts be used ime expand cn bused involve expanding he integrand or patf the iapran) inser or te rpeated use of meron y a, We ust thes method by example 421. Term by term integration ‘Consider the integral Te -[ Sa, ase p00 o ‘The integrand is small everywhere apart from lose 00. Since ae peed, fertct, a0) ‘we might try and formally expand the integrand: 1a)» [° Screen Eo) pi [ * ayrerenstat Ss Syn nt ~ LOW a ) Notice that this is an example of an asymptotic series that diverges, The result makes sense, however, because for x —+ oo the dominant contribution clearly ‘comes from the region close tot = 0, The result can also be obtained by repet use of integration by partso Lecture Noes on the Motenates of Acoustics 422° Integration by parts Ta) = £ os 2.28.2 00 50 write: vi [* 8 a [ oat (13) Consider the integral a acest (2) Clearly I(x) T(z) and use integration by parts [fe a Since as) wwehave as) an 4.2.3. Watson's lemma ‘Watson's lemma is an important too! in the analysis of Laplace type integrals: 1)= [~ ‘whore a solution is sought in the limit x + 00. Consider a graph of exp(~t) as shown in Fig. 4.1. As increases the ite tegrand will be close to zero everywhere except near t = 0, and ghus the main contribution to the integral comes from the segion close to ¢ = 0. Fle at, as “AsyupoieExpension of iegrals 95 Fig. 4.1 Graphs ofe~* against for varying Example 4.1 Consider the integral Ie) tf Solution ‘The exact solution is *sintdt. (9) +e @) aia gtate en "We can luo oan an approximate solution to the int J solaion tothe integral dey by nti that, near ¢ = 0, sin t ~ ¢. If we approximate the integral by . T(2)~ [ te-** dt, 2) a integration by parts recovers the correct leading order behaviour I(x) ~ ‘This type of argument can be made more rigorous through Watson's lemma,96 Lecture Notes onthe Mathematics of Acoustics Lemma 4.1 Let 3) then if (i) the integral [2° e-*|f(#)| At exists for some o> O-and if (ii) (0) has the asymptotic expansion f(t) ~ Ty ant"? as t —+ 0 and OA anh (go, 4) where the Gamma function is defined by T(A) = [2° e~* nl for integer n. at and P(n-+ 1) = A special case of this result is when f(t) can bé expanded as a power series about t= Din which case: 12) = f e*poaen > 2O 25) Proof. For z — oo the dominant contribution tothe integral is from the region close to t = 0, This motivates us to split the integral in two by writing: 12) = Petras Pemnnat=10)+() 26) ‘where 7 is small positive number. Now: ita) =| [ er*tpear| c [er fenha wacal=| [es rnat s [ emipioiae -f[ eeolem*t f(a 2) and this gives ate) ger@r-om f™ “1 F(|dt se“ (28) for some number M > 0, Hence we have proved that [/2(z)| i exponentially small and negligible compared to any power of 2. e Asymptotic Expansion of Integral 7 Now consider the first integral and use the asymptotic approximation for f(t): w MO = Dart! tev, (60 ey =o), 29) Then nye [em (Somer are zy By By making the substitution u = ce we obtain: ey dt, G0) GBD ma for some B since ¢y = oft’ = puyeet du f wey S Ef eran 2) By writing Lf “f oe 6) we ean now express (2) as wo [femel and where we note thatthe second integral is asymptotically small beceuse [[eeralsfem Perera), 05% ‘acne Neon he Matemas of Aco ‘Similerly we find that By is a sum of an integral of order 2” and one which is asymptotically small, Hence we have shown that: Tle) = hs) +(2) ote [omen du + ent +007), fo 66) T(a) ~ 0 ant *T (rn): en ot Fs Example 4.2. Consider again the example “er*taintat 28) a5 1 00, Solution Since sint = tt9/31419/S!+... wehave ay = (—1)"*1/(2n—1)}, Yn = 2n and Watson's lemma gives: T(a) ~ ayP(Ay)2™™ + aT (Aa)2™™* +. =P (2)2? - Bret + ; @) hich agres wih he approximation found above. Example 3 Consider he behaviour of “eg pat 40) as200. Solution By introducing the new variable w= 19 we obtain % asu(t + ul/8) 1a)= Pon Saw «) and Watson's lemma then gives I(x) ~ 0(3)=~*/*-+ §2(3)a¥* Atympite Expansion of Integrals ° 43° Method of Laplace In the last example a simple substitution was used to reduce the integral to one ‘where Watson’s lemma was applicable. Sometimes substitutions are hard to find or do not work and in these cases Laplace's method is useful. ‘Thus we look at integrals of the form (42) and seck the asymptotic behaviour as + 00. JF A(t) has a maximum then we may expect the major contribution tothe integral to come from a subregion surrounding this maximum. The essence of “Laplace's method is to replace ht) by the fist few terms of its Taylor series about this maximum. Suppose then that h(t) obtain its maximum value at =e. There axe three cases to consider, Ife = a ore = bthen we can have hc) #0 and h() is approximated by Ale) + (t- e)h'(c). (43) Ha ¢ band Me) = 0,4"() 0 then aproxiate Mt) by 19+ GE ane, a ar more generally by Hey+ THe, 9) when h(c) is the fist’non-vanishing derivative. In each case we obtain an integral to which Watson's lemma can be applied after a suitable substitution, Consider, for example, the second of these cases, We make the approximation I(2) ~ [ * p(eyentoehe-ata' (ay at, ween pejrtore orgy, 46) Now extend the range of integration to the whole ofthe real line, (since we have #"(@) < Othis change lends only to exponentially small terms), split the range of integration at = c, and make the substitution BRE 0)? 7)100 Lecture Notes onthe Mathematics of Acoustics togive 1) ~ 29 (VANS [Sau 2) and apply Wetson's emia to obtain ; Ua) we (VBR. 49) 44 Method of Stationary Phase ‘The method of stationary phase is of use when considering generalised Fourier integrals with complex integrands of the form 4 IQ)= [ Fl) dt, ase 00. 6) Integrals such as these are govemed by a very general result known as the Riemann-Lebesgue lemma. This states that I(z) + 0 as & + 00 so long as L/(O)| is integrable, 4(t) is continuously differentiable on (a, 5], and ¢(t) is not constant on any part of [a]. Assuming these conditions are satisfied, the leading order behaviour of (2) can be found by integrating by pars: °F) 4 1e)= | ay@eae -[ £0) es) fe By J. Je Both terms are O(1/z) as x —+ 00, but a second integration by parts quickly es- tablishes that the second term is actually O(1/22). Assuming the boundary terms do not vanish, they give the leading order asymptotic behaviour of the intégral. "We wl not prove the Riemann-Lebesgue lemma but the result can be under stood as arising from a destructive interference, The integrand is an oscillatory function, an the greater 2th more rapidly it slates, with contributions from neighbouring egion nearly cancelling. The east effective cancelation occurs a the end pont ofthe interval If the conditions on ¢(¢) are relaxed then integration by parts may not work because other subregions may exist where the cancellation i nefcent and these will also contribute (or dominate) the lading order asymptotic behaviour ofthe (e600) at 61) Asympotc Expansion of Itegrals 101 integral. Consider the integral tea) = [a4 irye™ at, 2) ‘The graph of the rel part of the inte; Te pach afihe grand about the origin is shown in Fig. 4.2 for i od Fig.42 Graphot Rofo!) agunst fork = 60, vas oa the. Lees of the integrand k¢? has a minimum at t = 0, the real part significantly to the asymptotic behaviour of the integral. : a If thereis only one point c for which a < ¢ < band #/(c) | = Othen so long as #(€) # O the method of stationary phase states that Pensclong He) ~ fle) a)” efizoortis) ae asz—oo, (53) withthe + or ~ sign acconding as $"(e) > Dor $"(c) <0,10a Lecture ote on he Mathomate of Acousicr 441 Outline proof Ree pen ote wie of bere! ine na mae poration LO~ LO, oe) ~ de) + He- 76") 64) giving 11a) ~ (900 [” exp {Hatt— 99") a 5) “This integral can be evaluated exactly using the substitution t-aa2 {Sa}? 66 with the being chosen depending on the sign of 4"(¢), to give f exp {Fia(t— o)?9"(e)} at =2 {eal £ ott Ge : (ia}"en en Example 4.4 i= /2) ap [remains = 2? f ens ap lo atenent [onan b va ae! e * fas melee int, (8) ‘The result proved here is easily generalised to cases where #')(c) is the first non- vanishing derivative. Anympttic Expansion of gras 13 45. Method of Steepest Descents This is a far more general method and encompasses both types of integral we have considered above. It is a technique for finding the asymptotic behaviour of integrals ofthe form Ia) = £ Fle) de (59) ‘Where f(z) and é(2) ae complex analytic functions, Cis a contour in the complex Plane, and we are again interested inthe limit as (real) x — co, If we decompose (2) as 4(z) = u(2) + iu(z) where w and 9 are the (teal valued) real and imaginary parts we may suspect that the major contribution to {he integral comes from points close tothe maximum value of u along the contour and seek an approximation by applying Laplace's method. However, this might well lead to an overestimate since the imaginary part, v(z), may vary rapidly along the contour, giving rise to rapid oscillations of the integrand and significant cancellation. ‘The method of stespest descents gets around this difficulty by exploiting the analyticity ofthe integrand. In such cases the value of the integral depends only ‘upon the end points and not on the contour joining them. Ifthe otiginal contour C’ 4s deformed to a contour C” where the phase u(z) varies as little as possible, then ‘an application of Laplace's method will give a far beter asymptotic approxima tion. ‘Because (2) is analytic, it satisfies the Cauchy-Riemann equations: Ou_ dv du__ bv Or By By az and these imply that Vu ‘Vv = 0. Hence the directional derivative of v(e) along a contour with tangent vector parallel to Vu must be zero, such a contour is & contour of constant phase. It is also the contour slong which u(2) vaties ‘ost rapidly, and since Vje"#| = WeF™() itis the contour along which the ‘magnitude ofthe integrand varies the most rapidly. In other words, itis a contour of steepest descent. Once the contour C has been deformed to the contour O which has constant phase, the asymptotic behaviour ofthe integral is determined by the value of the integrand close to the maximum of u(z) along C’. This maximum may accur at one ofthe end points but it may occur ata local interior poim. Because (2) is ‘assumed analytic, maxima of v(z) must correspond to saddle points of gz), At saddle points, curves of steepest descent can intersect and care must be taken to ensure that a curve of steepest ascent isnot chosen, (60)106 Lecture Naat onthe Mahemaris of Acoustics Example 4.5. ‘The Bessel function Jo(2) can be represented as teane{ 2 [Pertecon a, and we seek the asymptotic behaviour as x —+ co, @) Solution Here 4(2) = icosh(z) and g"(2) = isinh(z) = Oat z = 0. Since cosh(z + iy) = —sinh(z) sin(y) + icosh(z) cos(y) (62) the imaginary partis equal to 1 at the origin, and along the contour of constant phase we must have cosh(a) eas(y) = @) ‘We can now apply steepest descents extending the limits of integration to in- fnty and weting no {2 f exoticeahs)ech, where C'is any contour joining —o0 — i /2 and +00 +in/? and passing through the saddle point at the origin. The gradient, dy/ds, of the contour of steepest descent is equal to 1 at the origin and we can therefore approximate z = (1+ i)t where tis real, Likewise, cosh z ~ 1 + it? and Jo(2) = CO) so(e)~Re{ 22! [” expte-=2)at} ) “This negra scaly evaluated to give Jo(2) ~ Vifracos(z— n/a), , as 2+ 00. 6) Further Reading Leppington, F. G. (1992) Asymptotic Evaluation of Integrals, chapter 4 of ‘Modern Methods in Analytical Acoustics by Crighton, D. G. et al. Springer Verlag ‘Cautics, G. F, Krook, M, and Pearson, C. B. (1966) Functions of a Complex Variable. McGraw-Hill. og cena eg a ‘Asya Espanson of tegras 105 Problems Exercise 4.1 Show that as 2 —» 0+ (yeaa nes aa Exercise 4.2 By using term by term integration, show that the elliptic integral 2 Km (1-meos*@)-? ae cean be approximated for m <1 by (am, 9 K¢m) = (14 4 Se (=F (ie Be Exercise 4.3 Under the,change of variable u Hte)= [O° vamterwnot to [Ve hag 8 iter change of variable,» = wand apply Watson's oma to show at becomes cay, Exercise 4.4. Use the method of steepest descent show that HQY(p) ~ yf Delonte ass —o0, omy sg ao aay omyPARTI Wave MotionChapter 5 The Wiener-Hopf Technique ‘M,C. M, Wright Institute of Sound and Vibration Research University of Southampton 5.1 Introdueti ‘The Wiener-Hopf technique is, for acousticians, a way to determine wavefields in situations with separable. geometry but mixed boundary conditions, Here ‘mixed boundary conditions’ means different conditions on different regions of the boundary (as opposed to what are otherwise known as Robin boundary con- ditions, a ‘mixture’ of Neumann and Dirichlet conditions.) The technique works by applying our knowledge of the behaviour of complex. functions to complex Fourier transforms of the wavefield. We shall illustrate the technique by applying ito the problem of diffraction of plane waves by a semi-infinite sereen, Similar treatments of the same problem are given by Noble (1988) and Leppington (1992). 52. Rigid Sereen Diffraction Consider, then, a rigid soreen located at 2 <0, y = 0 extending to infinity in the positive z-direction, all quantities being independent of 2. Plane waves with wavenumber i are incident at an angle to the screen, as show in Fig, 5.1. The total velocity potential field is (ust) = Relbe“™]. wo ‘We can write the total field as the sum of the incident and scattered fields + flew, ®0 Lect Notes onthe Mathematics of Acousés wf | Incident plane waves | wavenumber k Screen (w+ BY Fig. 5.1 Geomety ofthe iid seren iftaction groom, where fis the scattered field that we wish to find and & = w/c ‘Since the total feld satisfies the two-dimensional Helmholtz equation ee ) a Sate) atKe (aera tens musth sated eld for which we mst now determine he boundary conditions. On the sereen we havé the usual vanishing normal velocity so the soattered y-velocity field is equal and opposite to the incident y-velocity field oF a hich e rest of y = 0, whic This is not enough, we need a boundary condition forthe resto we can get by recalling that the problem of finding the field scattdyed by a rigid body is equivalent o that of finding the eld radiated by "a vibrating body of the same size and shape whose normal velocity isthe negative of what is associated ‘with the incident wave" Pierce, 1981). Since the screen has no thickness, but its normals have opposite signs on opposite sides, the scattered feld must have also have opposite signs on opposite sides, in other words f(x,y) is an odd function ingy-and continuous for = > 0, hence (2,0) @ iksin@e™**°™°, §— fors 0. 6 The WienerHopf Technique m In order to ensure that our waves are physically realistic we let k ns k + ie, where € > 0 for the time being but we can let it vanish once we are satisfied with our solution, We expect the scattered field to account forthe ordinary reflection from the sereen away from its edge and the cancellation ofthe incident field on the other side, and itcan be seen that equal and opposite plane wave behaviour on each side is consistent with this, It is perhaps also worth remarking thatthe problem doesn't ‘have an intrinsic length scale other than the wavelength ofthe incident wave so there is no need to rescale variables Fig.52 The incident, casted and ttl potential els, ‘Figure 5.2 shows snapshots of the incident, scattered and total wavetields that We are trying to find, calculated using the results ofthis chapter. Ordinarily, of Course, we won't know the wavefield in advance, bit it may be useful to refer to these pictures in what follows, 5.21 Generalised Fourier transforms ‘We can see that the geometry is separable, so we can hope to treat the and dependencies separately and this suggests that Fourier transforming a spatial variable will be beneficial. Since we are allowing complex wavenumbers we must also allow the transform variable to be complex, giving F(sy)= f f(y)e™ dx, © where $ = 1 + isa. Assuming for the time being that this transform is valid the transformed wave equation takes the form (3 et x) F(sy) =0, m1m Lectre Notes on she Mahemates of Acousies ® (B-7) Fon iz denote the spatial com- ing defined = (s* — &)"?, Whereas we used & to dente the spa towel of inc hari fn efetaFute tesen Wik pe to time) we use F to denote the Fourier transform with respect to space (inthe 2 direction) off. 52.2 Branch euts ince is defined asa factional power we have to choose which rot to take for exh cps abe a's weve we move any fom» = 3h whee y= 0 without ambiguity. Remember that k is complex so k lic in the (+,+) and =, ~) quadrants ofthe complex plane. : ‘al from §2.2.3 (p. 120) thet tis impossible to choose roots in such a way that 7 is continuous overalls, Both rool, and —, are the origin of branch cuts. A single, finite branch cut joining the two roots would be possible. But for reasons which will become apparent we will prefer to make a separate branch cut from each zero of 1 infin its ovm half-plane In his way the stip we < sa <« will ot contain any branch cuts, and since it doesn't contin any other singularities either we can ascertain that -y so defined is analytic (j.c. single~ Valued and cifereniable) inthis stip. It is perhaps worth remarking that the Satement that a complex funtion isdifferentiable i considerably moce restrictive than itis whea applied to a sealae funtion, 523. One-sided transforms With 7 thus defined we can solve the ODE obtained from Bq. (8) by separating variables as Alsen", for y > 0, ® ~Als}e, fory <0, ron={ because Re(y) > O between the two branch cus, Note that Hany 0s F(8,) = (3,04) # F(¢,0-), which should not be surprising since’ is discontinuous across the screen. The form of A(s) remains to be determined from the boundary conditions, : Ikis at this point thatthe difficulty raised by the mixed boundary conditions becomes evident, and we have to (a) find a way to transform the two regions separately and (b) solve the resulting equations. Step (a) is a matter of defining oe Corey The Wiener-Hop Tecdgue 4s so-called generalised Fourier transforms, step () is the Wiener-Hopf technique (or more strictly Jones's method for solving Wiener-Hopf problems.) Itis simple enough to define two half-trensforms thus Fone [_setdes [~ jeae, Fa) +F(60), 0) but the time has now come to consider the implications of letting s be complex. Each integrand contains a factor exp(—s3z) so for each integral there will be values of sa for which it will grow exponentially. This ean only be accommodated as long as f shtinks at a greater rate 8s > co, implying a maximum value of 2 for which F_ can exist and a minimum s» for F,.. Specifically, as — 00 the behaviour of f will be dominated by cylindrical waves (due to the line sources on the screen, which satisfy the equivalent radiation problem) with an exponential decay factor ofc, implying that sg > ~e for F. As 2 + —co there will similarly be cylindrical radiation but there wil also be cancelling/efleting incident plane ‘waves, whose exponential decay in the ~2-direction willbe €cos requiring s» < £0080. As long as the relevant condition is satisfied each transform exists and is furthermore analytic since f is continuous in the z-direction. Crucially there is an overlapping strip in which both F_ and F, are analytic. Any equation containing both quantities must have 52 in this strip to be valid, Since we're doing one-sided transforms with a complex transform variable (and even calling its) it might be asked why we don't just use Laplace transforms 4s indeed many authors do. There is no strong reason not to, other than that the references quoted here use Fourier transforms and, perhaps, going from left and right hand spatial domains to upper and lower transform domains might help to ‘keep the two domains distinct in our minds, . 5.24 Transformed boundary conditions ‘We can now write our boundary conditions in terms of transformed quantities 50 thatthe condition off the screen gives (9,0) 6, ay (continuous across y = 0) whereas the condition on the screen becomes a state- ‘ent about the transform’s y-derivative (denoted by a dash) (12) (aso continuous deross y = 0 because 8f/8y, the seatered vertical particle ve- Jocity in the el, reaches zaro on both sides of the sreen). Since we know thet,1 Lecrure Noes on the Matenaies of Acoustics” undary conditions and hence Fis odd with respect to y we can work with the boundary eather eer af We shal ere wt Fy for F(s,0+) et. heeafe, noting thatthe ‘+? in Prefers tothe half ofthe a- domain from which it was transformed, not the side ofthe screen on which we're letting y vanish, If his symmetry argument were not available we would have to ‘work with new functions defined to be the sum and difference of F, on either side of the screen. AS itis, Eq (9) gives us FtheA (3) FL4 FL =-7A, a4) with which we can eliminate A and substitute in our transformed boundary con- ditions to get sin Eapauee as = keos" y PAR Both sides will be analytic as long as s lies in the strip between ~e and € cos. Figure 5.3 summarises the situation in the complex plane. “4 Sunetions analytic above this ne SE Branch eut Fig.53.Argand diag showing domains of scaly for Ea (1). The Wener-Hopf Technique us 53. The General Wiener-Hopf Equation Equation (15) s in the form ofthe general Wiener-Hopf equation K(s)Us(s)+ Vs) =P(), a a, or with V. a0 that oy Whereas the domains from which Uy and V- were transformed do nt oven, ‘he domains ofanalticity of th transforms do (or at east should fr the Wiens Hopf technique tobe applicable.) The procedures ten to eamange thi equation so that one side ill ‘sand the oter all ‘minus. Tis canbe done ts tllons (assoming, of course, htt can be done at all) factrise Kat Jes and divide through by F- pve FO Obviously must be non-zero for histo work. Then decompose Rs a sum, R= R_ + Ry and rearrange to get an we KU, Ry = R= aa) ‘where the left hand side is analytic for sp > ox and the right hand side is analytic for 2 <8. 53.1 Analytic continuation Recall now the earlier comment that analyticity is a more severe restriction for 4 complex function than itis for a scalar one, This is borne out in a surprising result, uterly different to that of real functions, that if a complex funetion can be differentiated once it can be differentiated any number of times. One of the Consequences of this is a theorem that can be stated colloquially as saying that an analytic function can't have a flat spot without being flat everywhere. That is ‘o say, fa function js known to be analytic over a domain and is also known to ‘be zero everywhere in a subdomain then it must be zero everywhere in the whole domain of analyticity (and hence likewise for any other constant value). This is because if the function is zero over a subdomain then the coefficients of its Taylor series expansion fn that subdomain must all be zero, And if this is true in the subdomain it must also he ie in the domain containing i, throughout which thecure Notes on the MathemaicofAcoucice us Lace N ce functions analy, sinc each analyte urtion only bas on unique Taylor series expansion abot parca point *onsequenee of this was given in §26 (p52), namely tat evo func one analyte over domains that veiap, and ae el in tht ve region, then thee exists unique function that analyte over oth domains, ele the vets continuo’ fone funtion inte ee’ domatn. Tie fallow ie treaty fom apljng the ‘0 Ma spo’ rule tthe difeeac bevean the tne fotos inthe ovetap rion, Tiss exact th ce forte two sides of 1 (U8), cach sie forms an analyte continuation of the ter, s thre mist Ba ofan funtion B() analytic ove the entre comple plane, defined by Vals) =E(), foralls. (19) Ky(9Us(6) ~ Rolo) = RAG) K-@) ‘We shall now do this for our diffraction problem, 53.2 Application to the diffraction problem ‘We have the following correspondence between the general Wiener-Hopf equa tion and the quantities in the half-plane diffraction problem: PM, anne, B+ ccos8. ‘We can factorise the kernel by inspection to give (6 RY = (oR (s+ AY, 20) where each factor has one of the two branch cuts in -y so that our version of Eq, (11) is, Ia Fi kein? G-WIPE+ CTR” Gomer =O ‘The right hand side of this equation is our R(s), which we need to decompose into the sum of a ‘++ anda “—" function, 1 hus a pole at s = eus@ aud a branch cut from ~K to 00 in the lower half plane. By inspection the pole due tothe factor The Wener-tHopf Tchigue oo 1/(s ~ kcos6) can be removed by multiplying by a function with a zero at that point as follows: snint ee -ksine [77 Z = Roos | GEA ~ EF Reo E keos8 | (s +k)? (E+ kos 6) | SSE malyie above Kos [zoalyic teow] + ~Esind (Fees 8)7* (6 Reos8)" —_—_ so that the two terms on the right hand side are R,(s) and R_(s) respectively, Us- ing these we can finaly write our Wiener-Hopf equation in the form of Eq, (18) (22) ~k)ip y ksind _ (RIOR. + Cras hema = chang [a A ae ‘head ‘eae = aa ~ GEE), @ £0 that since the two sides are analytic and equal inthe overlapping sip they axe equal everywhere that iter side is analy, in ober words aves the et complex plane. Product fctorstions are more stightorvard, partly because the roots of the product of two functions is just the union ofthe roos of each fcton The form of K(s) influences the difficulty of both decompositions because R(s) = PO) K-(s) 533 Behaviour ati nity inthe s-plane ‘The function (3) is still defined in terms of the unknown functions FY. and Fy, ‘but it turns out that we can deduce all we need to know about its form from its asymptotic behaviour as s ~+ oo in the half plane for which the right hand side of cach equation is analytic, Starting withthe frst equation, and hence considering the upper half plane where sg > —ecos8, the second term obviously tends toie enna Note on he Mathematics of Acoushes sero as $ — 00, but we also need to know the asymptotic behaviour of F...Even though we are using asymptotic behaviour we will still get a final result that is exact. “The so-called Abelian theorem relates the larg s asymptotic behaviour of the +4 tansform of a funtion to that function's small = asymptote, For a function (2) that satistes|g(2)| < Ae so that a4) = [ee ar, @ is analytic for s2 > avthen as |s| — oo in the upper half plane (writen [s| ~ co) then the integral will be dominated by the values of f when 2 is small so that lim exy= [" (29, 4) eae, 5) bina andi. g(2) —+ 2 as © — 0+ for some A > —1 then this gives eae age g jm, x(0 [fe 7 ff Gy 26) Note that =1 < \ < O corresponds to some sort of ‘blowing-up* at x = O+. Ifit is known that this doesn’t happen ie. is bounded at 0+ then A must be positive and G., must diminish atleast as fast as O(s™*) as |s| goes to infinity inthe upper half plane, ‘Similarly if G_ is analytic in a lower half plane and g(x) tends to (—2)/ with p> -Lase— O- then 6.0) = Gear en ‘These results also hold when |s| > oo in the relevant half plane of analyticity with ep remaining finite, as long as o(z) is infinitely differentiable, lim lelseat 534 Behaviour atthe edge ‘The Abelian theorem tells us that we can deduce the behatviour of Fs) atinfinty in the upper half plane if we know the behaviour of f(2,0)8f/By as x —~ 0+, and of F(s) at infinity in the lower half plane if we know the behaviour of /(,0) as z+ 0 with y = 0. We can infer this from the requirement thatthe kinetic nergy ofthe field should be finite since we don’t have any energy sources, If we write @ = |]. which varies with distance from the edge r according tog ~ r® The Wener-Hopf Technique 19 as — O+ then the kinetic energy in a region © surrounding 2 proportional to [[stesean [[pprrecar= [foe an and for this quantity to remain finite we need 2a +1 > -1 soa > -1, We therefore know that 8f/6y = O(a) asa —+ O+- and this implies, by the Abelian theorem, that F = O(s~!~*) as |3| + 007. Similar arguments require that f ’be bounded as < ~+ O= 50 that F_ = O(s"#) as |s| + 00 |. We can therefore conclude that Z(s) — 0 as |s| ~» oo in any direction. Once again we can exploit the analyticty of (s), this time by observing that it mast have a Taylor series = Owill be B(s)=entersbers? te, 9) Within a ciole of radius r about the origin. Since B(s) is analytic it must be bounded everywhere, and it can be shown that if [(s)| < A then these coeffi- cients are bounded by len] s Ar“, 0) ‘And this must stil be rue when we let r + 09, which leads us to conclude that all the em apart from e, must vanish. This is Liouville’ theorem, which tells us that en analytic function tha s bounded at infinity in all directions can only be a constant. A generalisation of Liouville’ theorem tht is often useful says that if E(s) < Als!" everywhere then £(s) must be a polynomial of degree IV o less. In our ease since we know that H(s) 0 as |s] — oo we must conclude that 9 = 0 a8 wel, leaving H(s) = 0 everywhere. We therefore have from Ba, (23) ksing EF Fees 7 (6 keas tyes B Gn) ‘and a corresponding equation for F, but if we just solve for F_ we can deter- rine f completely because A(s) = F_(s). This expression must then be inverse Foutier transformed with ¢ — 0. The result can be writen stan) = gt fowronorng e242) wo ($5!)),126 Lecture Notes one Mathematics of Acoustics where x = 1 cos © and y = rsin @ (@ defined to have a discontinuity where the screen lies), and where is a Fresnél function defined by Fa)= f[ ae @3) ‘The more usual definition ofa Fresne! function is (Abramowitz & Stegun, 1972) Fe) [ at, 6s) ‘The two are related by F(a) +F(@) 05) 544A Systematic Decomposition Procedure ‘The factorisation and sum decomposition in the half-plane diffraction problem were obtained in a distinctly ad hoc fashion, a mote systematic approach is desir- able, In fact if we can find such an approach to the sum decomposition we will also have found one for the factorisation since using it to find Ink =n +k, 66) where K is the kernel gives K. and K that satisfy K = KCK, atleast as long as n.K is analytic inthe stip in which we require both ‘+" and ‘— functions to be analytic So we wish to find R_ and Ry, such that R = R_+-Ry fora given R, analytic in the sip a < s2 < 8. Recall that Cauchy's integral formula gives us 1 f2Ox SS en RO) af Ona” ‘where the integra is around a closed path in the plane of ( = ¢: +i¢a, that encloses ¢ = 3 for whichever value of s we are evaluating it. The integral around a closed path is the sum of integrals over any set of path segments that make up the closed path so if we can find a way to cu the path into two segments so thatthe integral ‘over one is a+ function and the other isa ‘~" function we will have achieved our goal. Since (¢) is analytic in a < G < i the integfand won't have any singularities other than ¢ = s inthis strip so we are free to deform the path as we please as long as it stays within the stip, But since we want the result to work for 6 equal 1 any and all of the values of $ within the strip we would like w make it a5 large as possible within the strip. ole nena elec lle eal tamales ‘The Wener-Hopf Technique ma at | Fig. 5A Interation path inte ¢-plane for sum decomposition Start by fixing satan arbitrary position in the strip inthe ¢-plane and draw a rectangular path enclosing it, which crosses the Ca axis at ax and By. At this stage we have got four integrals that add up to R(s) rather than the two we want, but in order to include all s inthe strip we are going to want to expand the rectangle infinitely atthe sides. If R behaves suitably there then the contributions from inte- arating along the sides will vanish and we will be left with two terms as required. For this to be the case requires (C) = O(¢74) as |¢| —+ oo in the stip for some positive 6, since then RO = FE) - ou, 8) and this is a condition forthe sum decomposition method to work. Ifitis satisfied then we have 1 fit REO) A R= aco Broo ‘which is one sum decomposition but fort tobe the one we want we must examine the analtiity of each integral. AS long as lies within our (infinitely wide) rectangle then the inegrand will be analyte along each ofthe paths so the integrals will both be analytic. If s lies outside the rectangle then the integrals will still bea ee oe ee ee ee Tern tices ite athena of Aceon analytic functions, but not the same ones, since they will now add up to zero instead of R(s) by the residue theorem. We can therefore think ofthe frst integral as defining function that is analytic for s2 > a and the second one defining a function that is analytic for s2 0. Thisin Lecture Notes on the Mathematics of Acoustic éan allow the calulation of potential flow in situations where no suitable confor- tral mapping can be found, Vortcity can also be calculated in some situations ‘And although the previous derivation was fora steady state problem i is slays possible to inlude the time factor throughout and solve transient problems inthe (ther conguaonscansil betes long asthe goomety ofthe problem Teas toa separable solution ofthe wave equation, for example the radiation froma circular pipe, or diftection by parallel plates can be obtained by the Wiener-Hopf procedure. Its also necessary thatthe transform chosen allows the formulation Of the Wiener-Hopf equation. ‘The complex Fourier transform is not the only choice, others that may be necessary are the Mellin transform or occasionally the Kontorovich-Lebedey transform, details of which are given in Noble (1988) ‘Some problems, such as difriction by a plate with fnite thickness or finite extent, or radiation from a pipe with a finite flange, cannot be solved exacly by this procedure but are amenable to approximate methods based on it. In some cases, however exact results can be obtained by otber methods, such as matched asymptotic expansion. Difficulties als arise in problems involving matix-valued functions since the fact that matrix multiplication isnot, in general, commutative can prevent the factorization of the kernal. Recent work by Abrahsms (2000) has used Padé approximants, which effectively replace branch cuts with arrays of poles, to good effect. ‘The Wienes-Hopf technique can be applied to Fourier transforms of time as well as space. An obvious example of a semi-infinite interval is time divided into the past and the future, Tn this way itis fundamental to much of the signal processing literature concerning optimal prediction References Abrahams, I. D. (2000) The application of Padé approximants to Wiener-Hopf factorization, IMA Journal of Applied Mathematics 65, pp. 257-281. Abramowitz, M. and Stegun, I. (1972) Handbook of Mathematical Functions. Dover. Leppington, B.G. (1992) Wiener-Hopf Technique, chapter 5 of Modern Methods in Analytical Acoustics by Crighton, D. G. eta. Springer Verlag, Noble, B, (1988) Methods Based on the Wiener-Hopf Technique for the Solution of Pertal Differential Equations. American Mathematical Society Pierce, A. D. (1989) Acoustics: An Introduction to its Physical Principles and ‘pplication. Acosta Sei of Arai, Woodbury, NY, 9.425 Chapter 6 Waveguides 'M, Melver & C. M. Linton Department of Mathematical Sciences Loughborough University 6.1 Basie Theory 6.11 Introduction Consider @ small disturbance of a uniform, stationary, acoustic medium, The total pressure pr is pr=potr, [pl (4) For the case of rigid walls we have 08/0 = 0 and this implies that ¥"(0) = r'@ : Tf we assume that A <0 and write A= —7? we get yr—ly (25) hich has general solution Y= Ae 4 Bew™ (26) ‘The boundary conditions then show that A = B = 0,50 this is no good. Hence assume that A = 47 > 0, We get Y"47¥ en ‘which has general solution ¥ = Coos qy + Dainyy. 28) ‘The condition ¥"(0) = O shows that D =O and ¥"(d) =O leads to Cysingd = 9) We can'thave C = 0,sosin-yd = 0, Hence we musthave qdanm nah Qe. 0) tee, alien ciamrs cinema) cilia) aaah elit Wevepulier 19 Finally we need to consider the casey and the boundary conditions show that just what Eq, (28) gives when + = 0. ‘The solution in the cross-guide variable must therefore be ofthe form ‘Then ¥” = Osothat Y = B4+Fy 0. Hence ¥ is a constant, which is YW) =Cneos tain a= mH/, Wz BD ‘The equation for X is now X" 4 08x on = VEO om ‘Assuming tat an #0, itis helpful to write the solutions in the form of complex exponentials X a etlens, (33) We have shown that Cue cosa isa solution to our problem for any n = 0, 1,2,.. Since the equations we ate solving are linear and homogeneous, any linear combination of these solution is also a solution. Hence we waite the general solution to our problem as B= D> (Gye + Dyo™"*) cos yay. 64) J 6.1.6 Cutoff frequencies ‘The nature of the solutions we have just derived depends erucially on whether = VF = 75 is real or imaginary, Consider the solution (Cyel9>* + Dye) cos yy, Suppose first that k < 7%q. Then dq = IV9—F? and the Cn term decays exponentially as —> oo but grows exponentially as 2 —+ -00, The Da term has the reverse behaviour. In any physical application we must set the constants that ‘multiply solutions that grow at infinity to zero, ‘Suppose next that k = 4. Then a, = 0 and the solutions look like (Cx + Dp2) cos tay. ‘The resulting solutions with Dj = 0 are called standing waves. Suppose finally that k > “j. Then a is real (and positive). The Cy, tein represents a wave traveling in the direction of increasing x whereas the Dy, term represents a wave travelling in the dizection of decreasing 2.130 Lectne Notes onthe Mathematles of Acouses Now k = w/cp sow = kep. We define the quantities in = co these are the cutoff frequencies for the waveguide. Note that they depend on both the guide ‘geometry and the boundary conditions onthe guide walls. For Yq~1 << ‘fy f= 1,2,-.» there are n possible wave modes in the 64.7 Cylindrical guides In oylindrcal polar coordinates (7,8, 2), the Helmholtz equation is PR 100 105 FO, ee Soe tet hb en. 35) oe trae tae tage tO =O CD Consider a circular cylindrical guide of radius a, If we have pressure-release walls then onr=a. (6) Separate variables, Let (7,8,2) = ROO)X(2), en then wrox+ Lr'Ox + ZRO"x + ROX" + PROX (38) from which RoR oF ee eoeee _ 39) manta say, 69) Next say (40) ‘The equation for © is "+20 = ay and in order that the solution is single valued we require © to be periodic with Period Or. Hence we must have '= m, where mis an integer and then © = Acosmé + Bsinmé. (42) ‘The equation for Ris then PRY 4 RE (Pr? “This is a linear second-order ODE and hence it has two linearly-independent solutions. They are called Bessel functions and are represented by the symbols m)R =O. . 43) i \ (43) ; i Waveguides i Jm(77) and Yn(om). Properties ofthese functions can be looked up in standard reference books and the functions J and Yq are plotted in Fig. 6.2. One of the ‘ost important facts about them is that Yin (77) is singular at = O whereas J, isnot. Hence ifr = 0 is within the physical domain (as its in oUF Gaze, only the ‘Tm'functions can be present in the solution. " Fig,62 Besse functions Jp (solid ins) and Yo (dashed lin) Ifthe guide wall has zero impedance we would have R(a) = 0 and hence In() : 44) ‘The Bessel function Jy is an oscillatory function with a set of zeros (much like the sine or cosine function), These zeros ate labelled jp, ,,n = 1,2, and can be easily computed. We must then have " 1= Yan = bn [0 45) Finally, the a-varaton is ofthe form etomns, (49) where Onin = an ‘The cut-off frequencies are now min = Tent = (48)iat Lecture Note onthe Mathematics of Acoutice is tour plots of their cross section, (One way to visualise the modes is to draw contour plots ie. curves of constant &(r, 6,0). AS an example, the mode with r, @-dependence Talia: 7) 00839, where j,. is the first zero of Ja is illustrated in Fig, 63. Fig. 63. Curves of constant Jal 1) cos 3. The ight areas comespond to posive values and the dk reas to negative values, 62. Bigenfunction Expansions 62.1 Arolling piston We can represent arg piston, making small oscillations about its midpoint, at one end ofa vaveguide, by the boundary condition leno = ((y~ $a) coswt,0) 9) | and then we have the boundary-value problem shown in Fig. 6.4. ‘As we hive seen the general solution, satisfying the boundary conditions on the guide walls, canbe written = » Da Li (CacleF + Dace") cosrays, | 60) Waveguides a ved (2, y,t) = Re[@(e,y)e™*) (4E)P=0, keu/eo ig. 64 Aboundary-alu problem fo aoling piston ina waveguide ec ee ane eae een For k

yp the Cr, terms represent waves travelling to the right and the D, terms are waves travelling to the left. Hence, since we can only have outgoing waves 2s @ 00, Dy = O forall n. Thus B= YP Oye!** cos yt (32) ‘The coefficients C,, must be determined from the boundary condition on the pis- ton, which is Siac cosy =v — 4d ce) ‘We can think of this as a Fourier cosine series for the 2d-periodic extension of the function y—d/2. Multiply by cosrmx/d, m = 0,1,2,...and use the fact that ; oe [ce Men" y= where & = 1s ¢m = 2, m > 1, and dn isthe Kronecker delta (Se §1.8.1,p.27.) This yields 64) a 4d) 60s EY a Flamm = fy ~ fe) ece ha Balt ccems). 65) Hence ifm is even, Cy, is zero. On the other hand, if m is odd we have Oy = ee ee ee eee ee ee ee oe learis ae Lazare Hotes on te Mathenatce of Acoarier 4id Jugs? Thos =) —— ) Note that Ginga = WR (2g — 1)? fd 7) and so if k < 1/d there is no propagation down the guide. For (4-5 0 (tegion M) can be written as linear combinations ofthe terms exp(tky2}n(y), NB 0, and exp(-ten2) Uq(y),n 2 1, respectively, where By =O NA, fe = R=, tin Valu) = el? cos dau, Valu) = 24? sin uaa, An = ni, (59) (ns dn, (60) Woveulder bs and & = 1, ¢, = 2,n > 1, Both {a(p)}2 and {Wn (y)}@29 form complete orthonormal sets on (0,1). In particular, J Yoldrntv ay bon = [Bal %m ode @ For simpliity we assume thatthe wavenumber is in the range 0 < ke< 7/2 and then ky’ = —ik is purely imaginary whereas ky, tn, > 1 ave real, Without loss of generality fy, n> 1; ae all ken tobe positive, The fact that his imaginary but tha all tervals of and eal shows that one possible wave mode cin exist region | whereas no such modes se posible inregion IH. Thus we can define a scatring problem by imposing the edation conditions BW (EP 4 Reb yo(y) = el 4 RE — asa soo, (62) a0 as e—+00. (63) Hore R, the reflection coefficient is an unknown complex constant but eneruy considerations show that || = 1 and so we can write R= exp(ié). Then b~ cos(ke ~ ut) + cos(ke + wt — 8) asé—+—oo. (64) 631 Maiched eigenfunction expansions In region I we expand 6 as & yt a (ote seb aly) + Yo FB etry) 5) == where Uj, n > 0 are complex constants to be determined and Uj = ko(R — 1). Similarly, i 2 > 0, we write ee = emp), 66) Physical considerations require that the pressure and velocity are continuous across = = 0. Continuity of & and 80/8 ensures this and yields 2¥o(0) + 2 Fal) = — , O0 1, ey yt Frente ep so from Eq, (73) Srp Gediton (4 : ea Jy iw or; from the definitions of pgp and we -yialt te @ A Balm which can be written as a) where ul aru =14-0 = a n21. (85) Wart BeR yaa, ne 65) ‘The behaviour ofthe velocity near the origin can be used to show that uL,ul =0(n7¥) asm + 00, (86) and hence Vq rust satisfy Va = O(n~8/9) as n+ 00, ‘We can solve Eq, (84) a follows, Consider the numbers jim tf Ha, mor, en > pte Ba doy 2 = em where Cy is a sequence of contours (to be determined) on which z > oo as N+ 00 and f(z) is ameromorphie function that has the following properties: | (P1) f(z) has simple poles at z = ky. > 0, and at ko, 1 (22) (2) has simple zeros at z= ey. > 1, Waveguides 1 (P3) (2) = oft) as = + 00 0n Cy a8 N+ oo. ‘The last of these conditions ensures that Im = 0, m > 1, and then Cauchy's residue theorem gives So Res(f tn). Res(F ho) _ Sh Pei tyme MER XE we compar hs with Eg, (4 we that he oan tg, (8 is aen by Va = Res(f : kn) (89) provided Res(f : ho) = 1. Can a suitable f be found? Let fle) (90) Where A is a constant. The infinite product converges uniformly on any compact set excluding z= km, m > Oand z = ky since Th 2 = Ff com) & Itcan be shown that if we choose Cry to be a circle centred on the origin with radius (n ~ 4), then as 2 +00 on the curves Cy as NV +00 we have ry l= z/tm ins 12h and so, from Eg. (90), f(2) = O(28/*), which cetaily satisfies condition (3). ‘The function f(z) defined by Eq. (90) clearly satisfies conditions (Pl) and (P2) ‘and we can make sure that Res(f : —ka) = 1 by choosing oe") 2) A= ~2ky I See. (93) ethan hove pe fT (1+ Ko/Fim)(1 — ko/im) ee and the condition Vz) = O(n-9/2) can be checked directly. The numerator ‘within the products the complex conjugate f the denominator since hy is purelyM0 Lecture Note onthe Mathemaules of Acoues imaginary and soit immediately follows that [R| = 1 as required. If we write R= ~ exp(2ix) then ae Sf (e) -=(wE)] wm . & (wo btw ) SS 643. The Wiener-Hopf technique In the problem with the abrupt change in boundary condition we can extract the incident wave field by setting Bau) = Hay) +e. on) ‘Taking Fourier transforms as in the previous chapter leads to the Wiener-Hopf equation 4F.(6,0) = 28 Fi (s,0), (98) anh where y= (2-H? @) ‘and the prime indicates differentiation with respect to y. To ensure the appropri- ‘ate convergence ofthe Fourier transforms we assume that k has a small postive imaginary part ¢; the final result i obtained by letting ¢ — 0 Equation (98) can be rearranged to give wrk _ Fi(s,0) SE + F008) = - KEE, «100 where K(s) = yeothy. aol) ‘The next step is to write K(9) = K()K-(0), (402) where Kis a plus function and JC. is a minus function, and each is analytic and ‘non-zero in its respective half-plane. This can be done by noting that Ks) has an infinite number of simple poles and zeros (bul no branch points) aud expanding it 1s an infinite product. The poles and zeros in the respective half-planes can then — Wosegides i be easily separated, (It urns out that K_(—s) = K,(s).) Suppose that this has ‘been done. Then we can reanrange Eq, (100) to give i (srk, te) FO) eC ac +g) * ye -*) ~Ky() ED. wae (103) ‘Using the appropriate conditions on the solution near the origin we can show that both sides of Eq. (103) must be identically zero and hence F(s,0) and F-(s,0) can be determined. Knowledge of either of these function is suficient to calculate F(s,y) and then Fourier inversion yields tk costly = 1) FO J Ky @)rsiahy ‘The contour of integration is from —oo to co inthe strip of analyticity, On leting € ~ O this reduces to the real axis, but we must indent the contour above any singularities on (—00,0) and below any singularities on (0,00). For < 0 we can close the contour in the upper-half plane and we will pick up modes corresponding tothe zeros of sinh. On the other hand, for x > O we would have to close the contour in the lower-half plane, Since slew) eas, (08) Lathe ag Ko (405) we could rewrite the solution as Flay) = [LBW Vic (seas 406) 7K. J 7Feoshy and 50 we would pick up modes from the zeros of cosh 7. ‘We will explicitly solve the « < case, for which we will have to find K's). ‘The only poles of the integrand in Eq, (104) that lead to waves in the far field are those with s real and positive (the contour is indented above poles on the ‘imaginary axis) and if shas positive imaginary part then exp(—isz) decays expo- nentially as 2 ~» ~oo. Thus the reflection coeficont is determined by the zeros of ysinb-y. Note that sinhz=2]](+27/5), An =n, (107) coch's = TT 1 27/8), tn — (m3), (108)sik =a? J] 0-428) = (6? 2) JL + -2P2). 09) = hand ase k, (iio) Hence, from Cauchy's residue theorem kere othe AR RE Tee SEO aD F(@,y) ~ ani Ra -[KQ)? ay Next, ‘yeoshy Shy Tle +P /un) [LG +9?/a8) “G8 G8) K(s) os (e+ iea)(s i) TAG iby) oie)” Hence K,.(s), which is analytic and non-zero in the upper-half plane, is given by Ty An(s + itn) Ky oe) am (= TN erik) with K_(6) just given by changing the pls signs to minus signs. In this way we ensure that Ky(s) = K_(=8). The flection coeficient i then A Ty alk + tka)? aI (82 + 2) (hi ihn)? ~ RARE ig)? 7 LA PF RYE ea)” Ty (k= itn)(k + ikn) Ty (ik/ tn + 1)(—ik/ kn + 1) Ty cea ikaygeinen) "LL epee + Cie je F)” KO)" qs) aa & which isthe same as Eq. (94). nn YT ee, a Traveguldce 143 Further Reading For those interested in mathematical techniques relevant to acoustic waveguides, a few references are provided below. Some, in particular the books of Noble and Mittra & Lee, are ata fairly advanced level. The more basic techniques, such as separation of variables, can be found in standard texts on advanced engineering ‘mathematics Argence, E. and Kahan, T. (1967) Theory of Waveguides and Cavity Resonators Blackie & Son, Billingham, J. and King, A. C. (2000) Wave Motion, Cambridge University Press, Collin, R. B, (1991) Field Theory of Guided Waves, IEEE Press. Coulson, C. A. and Jeffrey, A. (1977) Waves: A Mathematical Approach to the Common Types of Wave Motion. Longman, Courant, R. and Hilbert, D. (1953 & 1962) Methods of Mathematical Physics, vols. 1 & II. Interscience Publishers, ones, D. S. (1986) Acoustic and Electromagnetic Waves. Clarendon Press. Leppington, F. G. (1992) Wiener-Hopf Technique, chapter 5 of Modern Methods in Anaiytical Acoustics by Crighton, D. G. etal, Springer Verlag, ‘Mittra, R. and Leo, S. W. (1971) Analytical Techniques in the Theory of Guided Waves, Macmillan Morse, P.M. and Ingard, K. U, (1986) Theoretical Acoustics. Princeton University Press. Noble, B. (1988) Methods Based on the Wiener-Hopf Technique for the Solution of Partial Differential Equations, American Mathematical Society, Rossing, T. D. and Fletcher, N. H. (1995) Principles of Vibration and Sowrid. Springer Verlag. Problems Exercise 6.1 Calculate the modes for a rigid rectangular guide with eross- section ~a Oand vice versa, We can make this explicit by writing separate equations forthe two spaces or, more compactly, by wating -a/f. = a ake dh ay ii: ims aie’) lem) tna) mitt. ih150 Lechre Notes on We Mathemater acoustic [As ky and iy vary the Value off will be either purely real or purely imag- inary. A postive imaginary value will attenuate the amplitude of a plane wave propagating in a direction that lies in the x = 0 plane by a factor that depends exponentially on the distance from that plane. Waves such as this are referred 10 as ‘evanescent waves’, More properly the wavelield is sid to evanesce in the 2 dieotion and propagate in the and y directions. When this happens the apparent ‘wavenumber of the propagating component will be (&2 ++ K3)*/2, so as KZ + hz increases beyond k the wavelength ofthe propagating wave i reduced. If ky were negative imaginary then the amplitude would grow exponentially with z, which ‘would be physically unrelisto ‘here are four different paths inthe complex plane that satisfy ke = (k? — i —K3)0/4 the alternatives being obtained by reflection in the eal andlor imagi- nary axes, We choose the correct version (shown in Fig. 7.1) on physical grounds, arguing that when ral the wavenumber should be positive for the waves to propa te avay from the source and that when imaginary it should be positive to ensure exponential decay (or evanescence). We could also reason that in a lossy medium ‘k-would have a small postive imaginary parte so thatthe correct path would be one given by fey = lim[(h + ie)? — BE — Rg), a) which follows the path T, shown in Fig, 7.1. To see that this eliminates'two of the four paths consider the locus ofthe quantity under the square root when ¢ is finite, and recall that taking the square root of a complex number halves is complex angle ‘The integrand of Eq. (17) is singular at k = 0 so this point on the path of integration would have tobe avoided or omitted somehow if his expression were to be evaluated numerically. It turns out that the development that follows avoids this problem, 7S. Angular Representation With Bq. (17) we have decomposed a spherical wave into an integral of plane waves as desired, In order to make use ofthis result, however, it would be useful to express the plane wave that corresponds tothe integrand in terms ofthe angle itmakes with some plane, since in our original problem the reflection coefficient ‘was specified to be angle dependent. In the same way that we converted (2, spatial dependence to (r,) depen- dence, and converted (ke, ky) wavenumber dependence to (k,,9) dependence in the plane, we will now convert from (key y, ) dependence to (k, 9,7) depen- dence where 7 isthe azimuthal angle as before and wis the polar angle, giving OE t i Wovefe Decompotion 151 ' Imag. Tae Real Fig. 7.1 Patho megration or he. the transformation hy = ksinoos, (9) ky = ksinwsiny, 20) iz = kos. @) ‘These values can be substituted directly into our integral (although in fact we ‘won't bother to do so just yet), and we can find dk. dy as before by laa) ft Gevacs “page| dyay keoswsin-y sin yp cos-y, [cos psin yl dvdy ake dh, addy = kk, sinydydy, (22) slving us an expression of the form Ok ik [lector anyores @) R On Sy for which we must determine the paths of integration. Since the spherical wave we are trying to represent hes no particular azimuthal dependence the range for yean be from 0 to 2r. The polar angle , however, implies dependence on k, which, ‘wehave seen, cai become imaginary, We must choose a path in the complex plane for that corresponds tothe path for ky as determined by Eq, (21) that we have12, Lecture Note on the Mathematics of houses previously chosen, To see how this can be done writ} = a+ end recall that [k= cosy = cos cosh 6 ~isincsinh 8, oy which follows directly from Euler's identity and the definitions of sinh and cosh. Using this expression we can treat the two straight sections ofthe path separately: For. 1>k/k>0 set For O-+k,/k-si00 set osass, 0-8 -+ 00, siving the path for p shown in Fig, 7.2. Imaginary values of y will correspond Imag, Real Te Fig. 72 Pat ofiegration 0. to waves that are purely evanescent in the = direction as did imaginary values of kz. We ought to check that the corresponding values of ft, and ky remain purely real along the y-path, which we can do by recalling that sin(a ++ iB) sin e cosh 8 + icosasinh 8. If either a = ¥ andlor 6 = 0 then sin(a + i8) will bbe purely real as required. “We now have the expression: GR AR ff atesio conveying sin y4lslcos¥) ak qHlesaveorsysindsia ysl) sin gaya, 25) ‘which decomposes a spherical wave into an integral of plane waves, specified by ‘cirmuthal und polar uugle of propagation, and whose integral is non-singular for all finite points on the path Ty. Wave Decomposition 158, 7.6 Solution to Original Problem (29,2) Fig. 73. Problem geometry, Consider @ monopole source located atthe origin witha reflecting interface parallel to the 2 = 0 plane but a distance d below it, The boundary has the property that a plane wave incident at angle tothe normal i reflected so that is ‘complex amplitude is modified by a factor V (y) (which may be complex to allow fora change of phase atthe boundary) A plane wave component of the reflected field, observed at the point (2, y, 2) will have travelled a distance (2, y,2d-+ z). ‘The spatial component ofthe total wavefield a that point wil then be ha Sy tpaleye) 26) Plz, y2) where, using Bq, (25) pan hf petseertinssaionrensy(y dada TIT yO en ‘This expression can be simplified by writing x = roos¢ and y = rsing and ss ee a a154 “Lecture Nott onthe Mathematics of Acoustice using a convenient identity to evaluate the inner integral: 9 in(acoutsytin) slay gy =f gikr(cond coo ytsin pain) sing ¢ ay= [fe ey l = [erinventr Day = In Jo(krsiny), (28) where Jo is the zeroth order Bessel function. We can now write the reflected wavefield as Pea = ik | Io(kersin p)e~iM24+=) con vy (y) sin gh dab. (29) ‘An expression of this form is known as a Sommerfeld integral and can contain more general transfer functions than the reflection coefficient we have used here, ‘The integrand no longer represents a plane wave field but it does retain explicit dependence on y, which allows us to apply the refletion factor, Approximation techniques such as the method of steepest descent (see §4.5, p, 103) can be used to evaluate this integral; alternatively numerical integration is possible, References Baxandall, P. and Liebeck, H. (1987) Vector Calculus. Oxford University Press. Brekhovskikh, L. M. and Godin, 0. A. (1992) Acoustics of Layered Media I: Point Sources and Bounded Beams. Springer Verlag Lighthll,J.(1978) Waves in Fluids. Cambridge University Press, Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. Vol. I 1433 McGraw-Hill, Sommerfeld, A. (1949) Partial Diferenial Equations in Physics, Academic Press. j Weyl, H, (1919) Ausbeitung elekiromdgnetscher Wellen tber cinem ebenen Leiter, Annas of Physies 60, pp. 481-500, Problems Exereise7.1._ Write computer programs to very as many of the equations in this chapter as you can by numerical integration. Wherever possible use symmetry to reduce the dimensionality and size ofthe region over which you have to solve, Exercise 7.2 Verify the thrae-cimensional Fourier transform given in Ea. (6). Wbsefeld Deconpostion 155 Exercise 7.3. Hankel functions are given by Hf!” = Jo+iYo’and HO Jo=L¥. Draw the path of integration T,. necessary forte following iden- ty to hold HL) (hr sin eM oY sin dy. Hint: Which functions are odd? Which functions are even? Exercise 7.4 The Hankel transform pair can be written as Ihe = Ze J dalhedoerar and de) =a [ B(bir Bebe a Show that if €(2,) is azimuthally uniform then £ is also the two-diménsional Fourier transform of € when k2 = B24 2.Chapter 8 Acoustics of Rigid—Porous Materials K. Attenborough & 0. Umnova Departinent of Engineering University of Hull 8.1 Introduction Porous matetals ae used widely for sound absorption, They may be fibrovs, cellular or granular. Fibrous matecils may be inthe form of mats, boards of preformed elements manufactured of gas, mineral or organic fibers (natural or tmanmade) and include felts and felted textiles, Cellular materials include polymer foatns of varying degrees of igi. Increasing uses being made of porous mete als for example aluminium foams, se sound absorbers. In comparison with other tmiteils, glass woo, fr example, which gives high absorption over wide fe- Guency range, meal foams are not very good sound absorbers, Howeves, the high Weight-specite stifiness, good crash-energy absorption ably, and fire rsistance of porous metals make them suitable fr sound absorption panels inthe srrat and automotive industries. Granular materials canbe regarded as an alternative to fibrous and foam ab- sorbers in many indoor and outdoor applications. Sound absorbing granular mate- Tals combine good mechanical stengt and very low manufictring costs. Gran- tar materials may be consolidated though use of some form of inde on he par ticles as in woodchip panels, porous concrete and pervious road surfaces. Gran | larmaterials may be unconsolidated (loose). There are many naturally-occuring | granular materials including sands, gravel, soils and snow. The acoustical proper- om: ties of such materials are an important factor in outdoor sound propagation. ‘Six example? of porous materials used for sound sorption are shown in Fig. 8.1, Since the particles are nearly spherical and identical the lead shot serves st ms) ai iain cl i | = = -— = = oe oe oe== 188 “Lectre Notes on ie Mationaes of Acoustes 1s a reference material when testing theories for the acoustical properties of rgi framed porous materials. This contribution investigates some of these theories, In theory rigid-porous materials may be considered to be dissipative fluids since the wave motion is confined to the fluid in the pores. The amplitude of the sound ‘wave in a porous material is attenuated and the phase is changed as a result of viscous drag and thermal exchange in the pores. Ideally these effects should be treated simultaneously but here they are treated separately. This approach has been shown by the authors and others to be justified (see bibliography) and the later comparisons between data and predictions demonstrate its validity. Fig, 8.1 Seven examples of porous materia 8.2 Wave Equations ‘The vector wave equation in an isotropic elastic sotid with equilibrium density po is given by au poe = +H) VV w+ nV +X, ® ‘where w is the displacement vector, Xi the body force, isthe rigidity modulus and the balk modulus is given by K = A+ 2u/3. In avid : = 0, X = O; principal stresses pressure, and = A: =p, where p is the hydrostatic e u=V4, KVV-Vé=po55V0. ) ” ro STAconticn of Rgid-Poows Motels 159 Using V - V¢ = V26, this leads to ay. Pb V6 = ar ® ‘A plane harmonio wave solution of this equation is A (2, 1) = $5 costw(t — 2/c0) + 9), @ Pow’ with 9 = (K/p0)*/ representing the phase speed ofthe wave and representing the phase, This solution can be written 6) with k= w/ep being the wavenumber and a = Ae™#, Note the time dependence convention being used here, ‘Using this expression for ¢, pressure and velocity are found to be, ee (at) = —porzee = po" = agit), © Fb 8 ites-un) 9 = Brae ~ pass” ° 83. Impedance ‘The ratio of pressure to velocity in air isthe characteristic impedance p/u = P0%0 = Ze. Consider forward and backward-travelling plane harmonic waves at two in- finite vertical planes in air separated by distance d (Fig. 82). There are both Fig, 2 Forward and backward pane waves in iz160 Lectre Notes on the Mathematics of Acoutes positve-going (incident) and negative-going (reflected) waves. Ifthe amplitude of the positive-going wave is a and that of the negative-going wave isa’ then the pressure velocity and impedance (i. the ratio of pressure to particle velocity) at any point are given by plat) = gel 4. fel(-te-w0, © (ee) = Agta (est) = 2 gite-un_ 2 (a t) u(t) @) &z,t) 0) ALA, aohs(Fs)~e) 4 afel-Het due) aF) = 2 RRO = ale Faw)” acttH(P) 4 alert) Geta) — ale TAD” ap Hence 2(P 1) = Ze eet) @~ AAT zS eo APa, al t(F)—wt) 4 gll(=Ae(P2) ut) Ps) = Le aTT-O — alah ATO” a3) Using Eqs. (12) and (13), itis possible to write 1Z(P,) cot(hd) + Ze AP) = 20706) 5 iz cot( kd)” an Here Ze and k denote the. charzeteristic impedance and propagation constent in the fui, Now consider sound incident from fluid 2 (usually sir) on a layer of material (fluid 1) in front of a rigid wall (Fig. 8.3) Since the wall is rigid, Z(P:) + c0, so from Eqs. (12) & (13) (Ps) = iZecot{kd) = Zecoth(-ikd). (3) attics?) MSc ica) | at) ac), ise? Acoustics of Rigid-Porow Materiels 16 Fie 63. Plane waves incident on a har-becked aye, Note that Z, and k inthis equation refer to the characteristic impedance and prop- agation constant in fluid 1. With the assumption that Z(Ps) = Z(Pa), Eq. (15) gives the surface impedance of a hard-backed layer (Sid 1) of thickness d. The (pressure) reflection coefficient is given by a’/a, The absorption coeffi- cient ofthe layer follows from 1 — |a! af? ‘As shown later, « consequence of dissipation due to viscous and thermal ef- fects in the pores is that both propagation constant and characteristic impedance ina porous medium are complex: b= ky tks, a6) Ze= By + ifh an ‘Also the surface impedance of a porous layer and the plane wave pressure reflec- tion are complex quantities. 84 Analysis for a Single Slit BAL Viscosity effects in aslitslike pore ‘According to Newton's law of viscosity, the viscous stress ina fiuid normal tothe {y-direction (Fig. 84.1) is given by ousted, as) la) =, he uid. ‘where 7 isthe ectificient of dynamic viscosity i ‘The change in viscous stress across an imaginary fluid element between 2a ctre Nene on Ne Methenater Acoust * Fig. 84 Change in viscous sess across an imaginary fui element and =; + Acs (see Fig. 84.1) is given by aus(ai) , Sealer + Aa) AR en Ct as) ‘As the width ofthe fluid element becomes infinitesimally small, then the viscous force per unit volume, ie. force per unit length if unit area in the 2; direction is being considered, is given by (20) ‘Making use of Eq, (20) in the equation of motion for a sound wave travelling in a parallel-walled slit of width 2a (Fig. 8.5), a, Buy res Oak (21) Applying the ‘no-slip’ boundary conditions i.e. vy = 0 at 21 = ta, the solution of Bg, (21) is 2) where ky = (—iwpo/n)*/? is the wavenumber of the viscous wave in a fitid. —aSs eS Acoues of Riid-Porous Materials 163 x3 | v(2;) Fig. 85. Velocity profile across a parlal-alied ait du to dynam viscous effects, ‘The averake of the patcle velocity over the width of the sit is given by el - es) oe Qv=) | de: Hore Aisa dimensionless parameter given by = Equation (23) may be rewritten inthe form 2a apo dz ane eh 3) (p00? /n)¥/? = alupo/n)/?. ~iwplw) 4) where Pl) = po/F(2), (25a) roy=1- od as and p(s) represents a ‘complex density’ of air in the pores. For some developments introduced later, a convenient alternative form for Bq, (24) is, Po spas ~ wwpaey 2H OVYOVAD Bey — ee Ey 84.2 Thermal effects in a slit-like pore ‘As the sound wave propagates in the slit it causes compressions and rarefactions in which the temperature respectively rises above ambient and falls Lelow am164 etre Note on the Mathematics of Acousties bient. If the solid walls of the slit are assumed to be heat conducting there is heat exchange between the walls and the sound wave. After combining the heat conduction equation and the equation of state, i is possible to obtain an equation for the temperature change 7 (‘acoustical temperature’) associated with a plane harmonic sound wave in a slit-like pore: Or tee meee an where 7/= ¢p/ev, the ratio of specific heats (constant pressurelconstant volume) and v! = n/pey, x being the thermal conductivity ‘Equation (27) for ris rather similar to Eq, (21) for vs. Given the boundary conditions r = Oat 2; = -ba, and, after averaging over the cross-section ofthe slit, the result forthe (complex) bul modulus of air in the sit, has a similar form tog, (24) for complex density i. Po, K= Tq Fra) 28) where Pr = 177Cv /isis the Prandtl number and Ps is the adiabatic bulk modulus of air. 84.3 Viscous and thermal effects in a cylindrical pore ‘The equivalent to Eg, (20) fora cylindrical pore may be written ‘After similar analysis to that for a sound wave in. slit-ike pore, Eq, (256) for a cylindrical pore of radius r may be written 20) where Ay = (upor?/n)!? = r(woo/n)!?, and J; arid Jo are eylindrical Bessel functions of order one and z2r0 respectively. . | ‘Corresponding results have been obtained for triangular pores andectanguler | pores. The results for four pore shapes are summarised in Table 8.1. ~ | ie Ee — om oe a a ee ee) Acoust of Rigid-Porows Moelle 165 ‘Tle 81 Dimensionless parameters and complex density functions fr fur pore shapes Raa i FO ste wit 2) wens 1= taah(aV=/OV=D exndr aus 2) atu? 1 = (fav SiC 00a) quilaer ianl ideo) (&V5/4)(u/v¥/2—— 1 ~ cot(aV=H/(AV=D +1727 eetangle (sides 2,28) Beco(u/y? ATH paolim + Hn + HF ovata Ka ns p+ et HOA SS ee ee ee 84.4 Low and high frequency approximations ‘The low frequency or small 4 approximations of Eq, (26), the equation of particle ‘motion in a slit, and the cylindrical pore equivalent are 6, 31, — Ph Sho + Hs, on and a 4 8 See a @2) “The first terms on the right hand sides of Eqs. (31) and (32) indicate an increase in the effective density of the fluid in the pores that depends on the pore shape. ‘The second terms onthe right hand side of Eqs, (31) and (32) represent the effects of viscosity in the pores. For steady flow through a single pore, the ratio of the applied pressure gradient to the average particle velocity is the flow resistivity Selting w = 0 gives the ‘d.c (zero velocity) flow resistivity Gym defined as = (8p/02s)/0s. $0 dyn = 3/2? in a slit pore of width 2a and dpa; = 8/1? in a cylindrical pore of radius. ‘The low frequency or small approximations of Eq. (28), the equation forthe complex bulk modulus ofthe fluid ina sit, andthe cylindrical pore equivalent are KeP jeder, 6) and KEP h + poste] 64) Since the second termns on the right hand sides of these equations involve the square of A, they are quite small for small A. For example in a slit of 20 xm width ‘and at 100 Hiz, } ='0,0084, So the thermal exchange in pores (of any shape) tends to be isothermal at low frequencies.