Scattering of EM waves by many small 6 perfectly conducting or impedance bodies 1 0 2 A.G.Ramm n a KansasStateUniversity,Manhattan,KS66506-2602,USA J 9 [email protected] ] h p - Abstract h at Atheoryofelectromagnetic(EM)wavescatteringbymanysmallparti- m cles of an arbitrary shape is developed. The particles are perfectly con- [ ducting or impedance. For a small impedance particle of an arbitrary shape an explicit analytical formula is derived for the scattering ampli- 1 v tude. Theformulaholdsas a 0, where a isacharacteristicsize ofthe 0 → small particle and the wavelength is arbitrary but fixed. The scattering 6 0 amplitudeforasmallimpedanceparticleisshowntobeproportionalto 2 a2 κ, where κ [0,1) is a parameter which can be chosen by an experi- − 0 ∈ menterashe/shewants. Theboundaryimpedanceofasmallparticleis . 1 assumedtobeoftheformζ ha κ,whereh const,Reh 0. Thescat- − 0 = = ≥ teringamplitudeforasmallperfectlyconductingparticleisproportional 6 1 toa3,itismuchsmallerthanthatforthesmallimpedanceparticle. : v The many-body scattering problem is solved under the physical as- i sumptions a d λ, where d is the minimal distance between neigh- X ≪ ≪ boring particles and λ is the wavelength. The distribution law for the r a smallimpedanceparticlesisN (δ) N(x)dxasa 0.HereN(x) 0is ∼ δ → ≥ anarbitrarycontinuousfunctionthatcanbechosenbytheexperimenter R and N (δ) is the number of particles in an arbitrary sub-domain ∆. It is proved that the EM field in the medium where many small particles, impedanceorperfectlyconducting,aredistributed,hasalimit,asa 0 → andadifferentialequationisderivedforthelimitingfield. Onthis basis the recipe is given for creatingmaterialswith a desired refractioncoefficientbyembeddingmanysmallimpedanceparticlesinto agivenmaterial. 1 Keywords: electromagnetic waves; scattering; impedance bodies; small bod- ies. MSC:78A45;78A25;Z8A40;78M40;35825;35J2;35J57. 1 Introduction Electromagnetic(EM) wave scatteringis a classical area of research. Rayleigh statedin1871.see[17],thatthemainpartofthefield,scatteredbyasmallbody, ka 1,wherek isthewavenumberanda isthecharacteristicsizeofthebody, ≪ is thedipoleradiation,butdidnot giveformulasfor calculatingthisradiation forbodiesofarbitraryshapes. ForsphericalbodiesMie(1908)gaveasolution toEMwavescatteringproblemusingseparationofvariablesinthesphericalco- ordinates. Thismethoddoesnotworkforbodiesofarbitraryshapes. Rayleigh andMieconcludedthatEMfield,scatteredbyasmallbody,isproportionalto O(a3).Weprovethatthefieldscatteredbyasmallimpedancebody(particle)of anarbitraryshapeisproportionaltoa2 κ,whereκ [0,1)isaparameterwhich − ∈ can be chosen by the experimenteras he/she wishes, see formula (1.3) below. Since 2 κ 3, it follows, for a 0, that the scattering amplitude for small − < → impedanceparticleis much larger than the scatteringamplitudefor perfectly conductingordielectricsmallparticle. Thisconclusionmaybeofpracticalim- portance. There is a large literature on low-frequency wave scattering and multiple scattering,see[1],[3],[6],[7],[19]. InthispaperatheoryofEMwavescatteringbyperfectlyconductingandby impedance small bodies of arbitraryshapes is developed. For one-body scat- tering problem explicit formulas for the scattering amplitudes are derived for perfectlyconductingand for impedancesmallbodiesof arbitraryshapes. For many-bodyscatteringproblemthesolutionisgivenasasumofexplicitterms withthecoefficientsthatsolvealinearalgebraicsystem. Ifthesizeofthesmall bodies a 0 and their number M M(a) , a limiting integral equation → = → ∞ is derived for thefield in thelimitingmedium. This equationallows us to ob- tainalocaldifferentialequationforthefieldinthelimitingmediumandtogive explicitanalyticformulasfortherefractioncoefficientofthelimitingmedium. As a result we formulate a recipe for creating materials with a desired re- fraction coefficient by embedding many small impedance particlesin a given material. Themethodsdevelopedinthispaperwereappliedtoacousticproblemsin 2 [11], to heat transfer in themedium where many small bodies are distributed in[13],towavescatteringbymanynano-wiresin[14]. In Section 2 the theory of EM wave scattering is developed for small per- fectlyconductingbodies(particles)ofarbitraryshapes. InSection3thetheoryisdevelopedforEMwavescatteringbyoneimpedance particleofanarbitraryshape. InSection4thetheoryisdevelopedforEMwavescatteringbymanysmall impedanceparticlesofanarbitraryshape. In Section 5 a recipe for creating materialswith a desired refraction coeffi- cient is given is given. The problem of creating materials with a desired mag- neticpermeabilityissolved. Physicalassumptionsinthispapercanbedescribedbytheinequalities: a d λ, (1.1) ≪ ≪ where λ is the wavelength in R3\Ω, Ω is a bounded domain in which many small particles D are distributed, 1 m M M(a), d is the minimal dis- m ≤ ≤ = tancebetweenneighboringparticles. Theboundaryimpedanceisassumedtobe h(x ) m ζ , (1.2) m = aκ wherex D isanarbitrarypointinsideD ,h(x)isanarbitrarycontinuous m m m ∈ functioninΩsuchthatReh 0,κ [0,1)isaparameter. Onecanchooseh and ≥ ∈ κasonewishes. The distributionof the small impedance particles in D is given by the for- mula 1 N (∆): N(x)dx(1 o(1)), a 0, (1.3) = a2 κ ∆ + → − Z where∆ Ωisanarbitraryopenset,N (∆)isthenumberofsmallparticlesin ⊂ theset∆,andN(x) 0isanarbitrarycontinuousfunctioninΩ. ≥ TheexperimentercanchoosethefunctionN(x) 0ashe/shewishes. ≥ Onehas N (∆) 1. (1.4) = xm ∆ X∈ By ω the frequency is denoted, k ω is the wave number, c is the velocity of = c lightintheair. 3 2 Scattering by perfectly conducting particles. 2.1 Scatteringbyoneparticle TheproblemistofindthesolutiontoMaxwell’sequations E iωµH, H iωǫE, inD : R3\D, (2.1) ′ ∇× = ∇× =− = where D is the small body, ka 1, a 0.5diamD, ǫ and µ are dielectric and ≪ = magneticconstantsofthemediumin D′, k ωpǫµ, andtheboundarycondi- = tionis: [N,[E,N]] 0 onS: ∂D. (2.2) = = HereandbelowN : N istheunitnormaltoS pointingintoD ,[E,N] E N s ′ = = × isthevectorproductoftwovectors,E N (E,N)isthescalarproduct, S isthe · = | | surfacearea. TheincidentfieldE is: 0 E E Eeikαx, H ∇× 0, (2.3) 0 · 0 = = iωµ whereα S2 isaunitvector,thedirectionoftheincidentplanewave,anditis ∈ assumedthatE α 0.Thisassumptionimpliesthat · = E 0, H 0. (2.4) 0 0 ∇· = ∇· = ThefieldE tobefoundis: E E v , (2.5) 0 E = + wherethescatteredfieldv satisfiestheradiationcondition E ∂v E r ikv o(1), r : x . (2.6) E ∂r − = =| |→∞ µ ¶ Inequation(2.6)theo(1)isuniformwithrespecttothedirectionβ: x ofthe = r scatteredfieldasr . →∞ ThescatteringamplitudeA(β,α,k)isdefinedasusual: eikr 1 x v A(β,α,k) o , r x , β . (2.7) E = r + r =| |→∞ = r µ ¶ ThemagneticfieldH H v , 0 H = + E v E H ∇× , v ∇× . (2.8) H = iωµ = iωµ 4 Letuslookforthesolutiontothescatteringproblem(2.1)-(2.6)oftheform: eikx t | − | E E g(x,t)J(t)dt, g(x,t) , (2.9) 0 = +∇× = 4π x t ZS | − | where J is a tangential field to S. We assume that S C2, that is, S is twice ∈ continuouslydifferentiable. Equations(2.1)aresatifiedif E E k2E, H ∇× . (2.10) ∇×∇× = = iωµ SinceE satisfiesequations(2.10),theseequationsareequivalentto 0 v v k2v , v ∇× E. (2.11) E E E ∇×∇× = = iωµ Equationforv isequivalenttotheequations: E ( 2 k2)v 0, v 0inD , (2.12) E E ′ ∇ + = ∇· = because v v 2v and v 0.Conversely,equations(2.12) E E E E ∇×∇× =∇∇· −∇ ∇· = areequivalentto(2.10)andto(2.1). Theradiationconditionissatisfiedby v g(x,t)J(t)dt E =∇× ZS foranyvector-function J(t). Theboundarycondition(2.2)yields J J TJ : [N ,[ g(s,t),J(t)]]dt [N ,E ], (2.13) s s s 0 2+ = 2+ ∇ =− ZS wheretheformula J(s) lim[N, g(x,t)J(t)dt] TJ, (2.14) x→s− ∇×ZS = 2 + was used, see [15]. Let us prove that equation (2.13) has a solution and this solutionisuniqueinthespaceC(S)ofcontinuousonS functions. Thisproves thatthescatteringproblemcanbesolvedbyformula(2.9)with J solving(2.13). 5 Theorem2.1. IfD issufficientlysmall,thenequation(2.13)isuniquelysolvable inC(S)anditssolution J istangentialtoS. Proof. Notethatanysolutiontoequation(2.13)isatangentialtoS field.Tosee this,justtakethescalarproductof N withbothsidesofequation(2.13). This s yieldsN J(s) 0.Inotherwords, J isatangentialtoS field. s · = Let uscheckthattheoperatorT iscompact inC(S). Thisfollows fromthe formula ∂g(s,t) TJ g(s,t)N J(t) J(t) dt. (2.15) S s = ∇ · − ∂N ZSµ s ¶ Indeed,if J isatangentialtoS fieldthen N J(s) 0. (2.16) s · = SinceS C2,relation(2.16)implies ∈ 1 N J(t) O( s t ) J(t), g(s,t)N J(t) O J(t). (2.17) s s s | · |= | − | | | |∇ · |≤ s t | | µ| − |¶ Thus, thefirst integral in (2.15) is a weakly singularcompact operator inC(S). Thesecondintegralin(2.15)isalsoaweaklysingularcompactoperatorinC(S) because ∂g(s,t) 1 O , (2.18) ∂N = s t ¯ s ¯ µ| − |¶ ¯ ¯ ifS C2. ¯ ¯ ∈ ¯ ¯ Consequently,equation(2.13)isofFredholmtypeinC(S).Thecorrespond- inghomogeneousequationhasonlythetrivialsolutionifDissufficientlysmall. Thisfollowsfromthefollowingargument. Thehomogeneousversionofequa- tion(2.13)meansthatthefunction v g(x,t)J(t)dt E =∇× ZS solvesequations(2.12),satisfiestheradiationcondition(2.6),and [N,v ] 0 onS. (2.19) E = Thisimpliesthatv 0inD . E ′ = Lemma2.1(seebelow)impliesthatif v 0inD then J 0. Thisconclu- E ′ = = sion and the Fredholm alternativeprove the existence and uniqueness of the solutiontoequation(2.13). ThesmallnessofthebodyD guaranteesthatk2 is notaDirichleteigenvalueoftheLaplacianinD. Theorem2.1isproved. 6 Lemma2.1. Assumethatthefollowingconditionshold: a)v 0inD , E ′ = b) J istangentialtoS, and c)k2isnotaDirichleteigenvalueoftheLaplacianinD. Then J 0. = Proof. Denote A: g(x,t)J(t)dt andusetheformula = S R A Bdx A Bdx N [A,B]ds A Bdx, (2.20) ∇× · = ·∇× − · = ·∇× ZD′ ZD′ ZS ZD′ validforanyB C (D ). If A 0inD ,thenformula(2.20)yields ∈ 0∞ ′ ∇× = ′ A·∇×Bdx=0, ∀B ∈C0∞(D′). (2.21) ZD′ Writethisformulaas dtJ(t) g(x,t)F(x)dx 0, F : B. (2.22) · = =∇× ZS ZD′ Thesetofvector-fieldsF coincidewiththesetofdivergence-freefields F 0 ∇· = inD ,whereF C (D ). ′ ∈ 0∞ ′ Thesetofvector-fields G(t) g(x,t)F(x)dx, F C (D ), = ∀ ∈ 0∞ ′ ZD′ whereitisnotassumedthatthecondition F 0holds,isdenseinthesetL2(S) ∇· = ofvectorfields.Indeed,ifthereexistsanh 0suchthat 6= h(t) g(x,t)F(x)dxdt 0, F C (D ), (2.23) = ∀ ∈ 0∞ ′ ZS ZD′ andw(x): g(x,t)h(t)dt,then = S R w(x)F(x)dx 0, F C (D ). = ∀ ∈ 0∞ ′ ZD′ Thus, w(x) g(x,t)h(t)dt 0 inD . (2.24) ′ = = ZS 7 Consequently, ( 2 k2)w 0 inD, w 0 on S. (2.25) ∇ + = = Sincek2 isnotaDirichleteigenvalueoftheLaplacianinD,equation(2.25)im- plies w 0 in D. Therefore, w 0 in D D . This implies h ∂w ∂w 0. = = ∪ ′ = ∂N − ∂N = Consequently,thesetG(t)isdenseinthesetL2(S)ofvectorfieldso+nS. − Weclaimthatif F 0inD ,whereF C (D ),then G 0onS. ∇· = ′ ∈ 0∞ ′ ∇· = Indeed, g(x,t)F(x)dx g(x,t) F(x)dx g(x,t) F(x)dx 0. (2.26) t x ∇ · =− ∇ · = ∇· = ZD′ ZD′ ZD′ Conversely,if G 0onS,thenequations(2.26)showthat ∇· = g(x,t) F(x)dx 0, t S. ∇· = ∀ ∈ ZD′ Let us use the local coordinate system with the axis x directed along the 3 outernormalN toS,andx (s),x (s)arecoordinatesalongtwoorthogonalaxes s 1 2 tangentialtoS.Letusdenotebye (s)ande (s)theunitvectorsalongtheseaxes 1 2 atapoints S. ∈ Equation(2.22)canbewrittenas J(t) G(t)dt 0 (2.27) · = ZS forallsmoothG(t)suchthat G 0onS,G g(x,t)F(x)dx, F 0. ∇· = = D′ ∇· = Let J(t) J (t)e (t) J (t)e (t) in the local coordinates. For an arbitrary = 1 1 + 2 2 R smallδ 0onecanchooseG (t)andG (t)suchthat 1 2 > J G J G δ, (2.28) 1 1 L2(S) 2 2 L2(S) || − || +|| − || < wheretheover-bardenotesthecomplexconjugate. WithG andG sochosen, 1 2 chooseG suchthat 3 G 0 on S, (2.29) ∇· = whichisclearlypossible.Thenequation(2.27)yields ( J 2 J 2)dt O(δ). (2.30) 1 2 | | +| | = ZS Sinceδ 0isarbitrarysmall,relation(2.30)impliesJ J 0.Therefore, J 0. 1 2 > = = =✷ Lemma2.1isproved. 8 As was stated above, it follows from Lemma 2.1 and from the Fredholm alternativethat equation (2.13) is uniquely solvable for any right-hand side if k2 σ(∆ ), that is, if k2 is not a Dirichlet eigenvalue of the Laplacian in D. If D 6∈ D is sufficientlysmall, which we assumesince a 0, then a fixed numberk2 → cannot be a Dirichlet eigenvalue of the Laplacian in D because the smallest DirichleteigenvalueoftheLaplacianinD isO( 1 ) k2ifa 0. a2 > → Remark2.1. The assumption k2 σ(∆ ) can be discarded if g(x,t) is replaced D 6∈ byg (x,t),theGreenfunctionoftheDirichletHelmholtzoperatorintheexterior ǫ ofaballB : {x: x ǫ},whereǫ 0ischosensothatk2 σ(∆ ). Thischoice ǫ = | |≤ > 6∈ D\Bǫ ofǫ 0isalwayspossible(see[8],p. 29). > Let us denote by V the operator that gives the tangential to S component v oftheuniquesolutionv tothescatteringproblem(2.1)–(2.3),(2.6): Eτ E E E v , v V( [N,E ]). (2.31) 0 E Eτ 0 = + = − If the tangential component v is known, then v is uniquely defined in D . Eτ E ′ Thisisaknownfact,see,forexample,[15].TheoperatorV islinearandbounded inC(S). It mapsC(S) ontoC(S) and v has the same smoothness as the data E [N,E ]. Forexample,ifS Cℓ,thenv Cℓ(D ),whereℓ 0. 0 E ′ ∈ ∈ > Define Q: J(t)dt. (2.32) = ZS Fromformulas(2.7),(2.9)and(2.32)itfollowsthat ik A(β,α,k) [β,Q]. (2.33) = 4π ForbodyD onehas [N,E ]ds E dx E D E c a3, (2.34) 0 0 0 0 D = ∇× =∇× | |=∇× ZS ZD where D isthevolumeofD andc 0isaconstantdependingontheshape D | | > ofD. Forexample,ifD isaballofradiusa,thenc 4π. D = 3 Onehastheformula(see[15],p.8): ∂g(s,t) 1 ds o(1), a 0. (2.35) − ∂N = 2+ → ZS S 9 SinceN J(s) 0andSisC2 smooth,itfollowsthat N J(t) c s t J(t). s s · = − | · |≤ | − || | Therefore 1 I : ds dt g(s,t)N J(t) c ds dt J(t), (2.36) s s = ∇ · ≤ s t | | ¯ZS ZS ¯ ZS ZS | − | ¯ ¯ ¯ ¯ and I O(a¯) J(t)dt. If I wouldsatis¯fytheestimateI o(Q),asa 0,then ≤ S| | = → the theorywould simplify considerably and one would haveQ E D R =−∇× | |= Ec a3. Unfortunately,estimateI o(Q)isnotvalid,andonehastogive D −∇× = anewestimatefortheintegralI : ds dt g(s,t)N J(t). Todothis,inte- 1 = S S ∇s s· grateequation(2.13)overS,useequations(2.15)and(2.35),andget R R Q I c a3 E . (2.37) 1 D 0 + =− ∇× LetuswriteI as 1 I e Γ (t)J (t)dt, (2.38) 1 p pq q = ZS where{e }3 isanorthonormalbasisofR3, p p 1 = ∂g(s,t) Γ (t): N (s)ds, (2.39) pq q = ∂s ZS p and the integral in formula (2.39) is understood as a singular integral. Thus, equation(2.37)takestheform (I Γ)Q c a3 E . (2.40) D 0 + =− ∇× HeretheconstantmatrixΓisdeterminedfromtherelation ΓQ e Γ (t)J (t)dt, (2.41) p pq q = ZS thesummationisunderstoodovertherepeatedindicesp,q,soΓisthematrix which sends a constant vector Q onto the constant vector I defined by the 1 equation(2.38). OnecanprovethattheconstantmatrixΓexistsandcanbedeterminedby equation(2.41),andthematrixI Γisnon-singular. + To prove that a constant matrix Γ exists assume that for every p 1,2,3, = thesetoffunctions{Γ (t)}3 islinearlyindependentinL2(S), Γ2 (t)dt 0 pq q 1 S pq 6= and Q J(t)dt 0. Here=J(t) 3 e J (t). For a fixed p let M be the = S 6= = q 1 q q R p set in L2(S) orthogonalto thelinearspa=n of Γ (t). Then every function J (t) R P pq q 10