Under consideration for publication inJ. Fluid Mech. 1 Capillary-gravity wave transport over 0 spatially random drift 0 0 2 By GUILLAUME BAL∗ and TOM CHOU† n a ∗ Department of Mathematics, University of Chicago, Chicago, IL 60637 J †Department of Mathematics, Stanford University,Stanford, CA 94305 1 (Received 2 February 2008) ] n y We derive transport equations for the propagation of water wave action in the pres- d ence ofa static,spatiallyrandomsurface drift.Using the Wigner distributionW(x,k,t) - to represent the envelope of the wave amplitude at position x contained in waves with u l wavevectork,wedescribesurfacewavetransportoverstaticflowsconsistingoftwolength f . scales; one varying smoothly on the wavelength scale, the other varying on a scale com- s parabletothewavelength.Thespatiallyrapidlyvaryingbutweaksurfaceflowsaugment c i the characteristic equations with scattering terms that are explicit functions of the cor- s y relations of the random surface currents. These scattering terms depend parametrically h on the magnitudes and directions of the smoothly varying drift and are shown to give p risetoaDopplercoupledscatteringmechanism.TheDopplerinteractioninthepresence [ of slowly varying drift modifies the scattering processes and provides a mechanism for 1 coupling long wavelengths with short wavelengths.Conservation of wave action (CWA), v typically derived for slowly varying drift, is extended to systems with rapidly varying 1 flow. At yet larger propagation distances, we derive from the transport equations, an 0 equation for wave energy diffusion. The associated diffusion constant is also expressed 0 1 in terms of the surface flow correlations. Our results provide a formal set of equations 0 to analyse transport of surface wave action, intensity, energy, and wave scattering as a 0 function of the slowly varying drifts and the correlationfunctions of the random, highly 0 oscillatory surface flows. / s c i s y h 1. Introduction p v: Water wave dynamics are altered by interactions with spatially varying surface flows. i The surface flows modify the free surface boundary conditions that determine the dis- X persion for propagating water waves. The effect of smoothly varying (compared to the r wavelength) currents have been analysed using ray theory (Peregrine (1976), Jonsson a (1990)) and the principle of conservation of wave action (CWA) (cf. Longuet-Higgins & Stewart (1961), Mei (1979), White (1999), Whitham (1974) and references within). These studies have largely focussed on long ocean gravity waves propagating over even larger scale spatially varying drifts. Water waves can also scatter from regions of un- derlying vorticity regions smaller than the wavelength Fabrikant & Raevsky (1994) and Cerda & Lund (1993). Boundary conditions that vary on capillary length scales, as well as wave interactions with structures comparable to or smaller than the wavelength can also lead to wave scattering (Chou, Lucas & Stone (1995), Gou, Messiter & Schultz (1993)), attenuation (Chou & Nelson (1994), Lee et al. (1993)), and Bragg reflections (Chou (1998), Naciri & Mei (1988)). Nonetheless, water wave propagation over random 2 Bal & Chou static underlying currents that vary on both large and small length scales, and their interactions, have received relatively less attention. Inthispaper,wewillonlyconsiderstaticirrotationalcurrents,butderivethetransport equations for surface waves in the presence of underlying flows that vary on both long and short (on the order of the wavelength) length scales. Rather than computing wave scattering fromspecific static flow configurations(Gerber (1993),Trulson& Mei (1993), Fabrikant & Raevsky (1994)), we take a statistical approach by considering ensemble averagesoverrealisations of the static randomness.Different statistical approaches have beenappliedtowavepropagationoverarandomdepth(Elter&Molyneux(1972)),third soundlocalizationinsuperfluidHeliumfilms(Kleinert(1990)),andwavediffusioninthe presenceofturbulentflows(Howe(1973),Rayevskiy(1983),Fannjiang&Ryzhik(1999)). In the next section we derive the linearised capillary-gravity wave equations to low- est order in the irrotational surface flow. The fluid mechanical boundary conditions are reduced to two partial differential equations that couple the surface height to velocity potential at the free surface. We treat only the “high frequency” limit (Ryzhik, Papani- colaou,&Keller(1996))wherewavelengthsaremuchsmallerthanwavepropagationdis- tancesunderconsideration.InSection3,weintroducetheWignerdistributionW(x,k,t) which represents the wave energy density and allows us to treat surface currents that vary simultaneously on two separatedlength scales. The dynamicalequations developed in section 2 are then written in terms of an evolution equation for W. Upon expanding W in powers of wavelength/propagationdistance, we obtain transport equations. In Section 4, we present our main mathematical result, equation (4.1), an equation describing the transport of surface wave action. Appendix A gives details of some of the derivation. The transport equation includes advection by the slowly varying drift, plus scattering terms that are functions of the correlations of the rapidly varying drift, representingwaterwavescattering.Uponsimultaneouslytreatingbothsmoothlyvarying andrapidlyvaryingflowsusingatwo-scaleexpansion,wefindthatscatteringfromrapidly varying flows depends parametrically on the smoothly varying flows. In the Results and Discussion, we discuss the regimes of validity, consider specific forms for the correlation functions, and detail the conditions for doppler coupling. CWA is extended to include rapidlyvaryingdriftprovidedthatthecorrelationsofthedriftsatisfycertainconstraints. We also physically motivate the reason for considering two scales for the underlying drift. In the limit of still larger propagation distances, after multiple wave scattering, wave propagation leaves the transport regime and becomes diffusive (Sheng (1995)). A diffusion equation for water wave energy is also given, with an outline of its derivation given in Appendix B. 2. Surface wave equations AssumeanunderlyingflowV(x,z) (U (x,z),U (x,z),U (x,z)) (U(x,z),U (x,z)), 1 2 z z ≡ ≡ where the 1,2 components denote the two-dimensional in-plane directions. This static flow may be generated by external, time independent sources such as wind or inter- nal flows beneath the water surface. The surface deformation due to V(x,z) is denoted η¯(x) where x (x,y) is the two-dimensional in-plane position vector. An additional ≡ variation in height due to the velocity v(x,z) associated with surface waves is denoted η(x,t). When all flows are irrotational, we can define their associated velocity poten- tials V(x,z) ( +ˆz∂ )Φ(x,z) andv(x,z,t) ( +ˆz∂ )ϕ(x,z,t). Incompressibility x z x z ≡ ∇ ≡ ∇ requires ∆ϕ(x,z,t)+∂2ϕ(x,z,t)=∆Φ(x,z)+∂2Φ(x,z)=0, (2.1) z z Water wave transport 3 X +y - x +−εy |y|~1/k x x 2 x 1 O(1) O(L) Figure 1. The relevant scales in water wave transport. Initially, the system size, observation point, and length scale of the slowly varying drift is O(L), with surface wave wavelength and scaleoftherandomsurfacecurrentofO(1).Uponrescaling,thesystemsizebecomesO(1),while the wavelength and random flow variations are O(ε). where ∆ = 2 is the two-dimensional Laplacian. The kinematic condition (Whitham ∇x (1974)) applied at z =η¯(x)+η(x,t) ζ(x,t) is ≡ ∂ η(x,t)+U(x,ζ) ζ(x,t)=U (x,z =ζ)+∂ ϕ(x,z =ζ,t). (2.2) t x z z ·∇ Upon expanding (2.2) to linear order in η and ϕ about the static free surface, the right hand side becomes U (x,ζ)+∂ ϕ(x,ζ,t)=U (x,η¯)+η(x,t)∂ U (x,η¯)+∂ ϕ(x,η¯,t)+O(η2). (2.3) z z z z z z Atthestaticsurfaceη¯,U(x,η¯) η¯(x)=U (x,η¯).Nowassumethattheunderlyingflow x z ·∇ isweakenoughsuchthatU (x,z 0)andη¯arebothsmall.Arigidsurfaceapproximation z ≈ is appropriate for small Froude numbers U2/c2 η¯2 U (x,0)/U(x,0) 1 (c φ ∼ |∇x | ∼ z | | ≪ φ is the surface wave phase velocity) when the free surface boundary conditions can be approximately evaluated at z = 0 (Fabrikant & Raevsky (1994)). Although we have assumed U (x,z 0) = ∂ Φ(x,z 0) 0 and a vanishing static surface deformation z z ≈ ≈ ≈ η¯(x) 0, U(x,0)= ∂ U (x,0)=0. x z z ≈ ∇ · − 6 Combining the above approximationswith the dynamic boundary conditions (derived frombalance of normalsurface stressesat z =0 (Whitham (1974))),we have the pairof coupled equations ∂ η(x,t)+ (U(x,z =0)η(x,t))= lim ∂ ϕ(x,z,t) t x z ∇ · z→0− (2.4) lim [ρ∂ ϕ(x,z,t)+ρU(x,z) ϕ(x,z,t)]=σ∆η(x,t) ρgη(x,t) t x z→0− ·∇ − where σ and g are the air-water surface tension and gravitational acceleration, respec- tively.Althoughitisstraightforwardtoexpandtohigherordersinη¯(x)andη(x,t),orto include underlying vorticity, we will limit our study to equations (2.4) in order to make the development of the transport equations more transparent. The typicalsystemsize,ordistance of wavepropagationshownin Fig.1 is of O(L) with 4 Bal & Chou L 1. Wavelengths however, are of O(1). To implement our high frequency (Ryzhik, ≫ Papanicolaou,& Keller (1996))asymptotic analyses,we rescale the system such that all distances are measured in units of L ε−1. We eventually take the limit ε 0 as an ≡ → approximationforsmall,finiteε.Surface velocities,potentials,andheightdisplacements are now functions of the new variables x x/ε,z z/ε and t t/ε. We shall further → → → nondimensionalise all distances in terms of the capillary length ℓ = σ/gρ. Time, c velocitypotentials,andvelocitiesaredimensionalisedinunits of ℓ /g, pgℓ3,and√gℓ c c c respectively, e.g. for water, U =1 corresponds to a surface driftpvelocitypof 16.3cm/s. ∼ Since U (x,z 0) 0, we define the flow at the surface by z ≈ ≈ U(x,z =0) U(x)+√εδU(x/ε). (2.5) ≡ In these rescaled coordinates, U(x) denotes surface flows varying on length scales of O(1) much greaterthan a typicalwavelength,while δU(x/ε) varies over lengths of O(ε) comparable to a typical wavelength. The amplitude of the slowly varying flow U(x) is O(ε0),whilethatoftherapidlyvaryingflowδU(x/ε),isassumedtobeofO(√ε).Amore detailed discussion of the physical motivation for considering the √ε scaling is deferred to the Results and Discussion. After rescaling, the boundary conditions (2.4) evaluated at z =0 become ∂ η(x,t)+ U(x)+√εδU(x/ε) η(x) = lim ∂ ϕ(x,0) t x z ∇ · z→0− (cid:2)(cid:0) (cid:1) (cid:3) ∂ ϕ(x,t)+U(x) ϕ(x,t)+√εδU(x/ε) ϕ(x,t)=ε∆η(x,t) ε−1η(x,t). t x x ·∇ ·∇ − (2.6) Althoughdriftthatvariesslowlyalongonewavelengthcanbetreatedwithcharacteristics and WKB theory, random flows varying on the wavelength scale require a statistical approach. Without loss of generality, we choose δU to have zero mean and an isotropic two-point correlation function δU (x)δU (x′) R (x x′ ), where (i,j)=(1,2) and i j ij h i ≡ | − | ... denotes an ensemble average over realisations of δU(x). h i We now define the spatial Fourier decompositions for the dynamical wave variables coshq(h+z) ϕ(x, h6z 6ζ,t)= ϕ(q,t)e−iq·x , η(x,t)= η(q,t)e−iq·x, − Z coshqh Z q q (2.7) the static surface flows x U(x)= U(q)e−iq·x, δU = δU(q)e−iq·x/ε, (2.8) Zq (cid:16)ε(cid:17) Zq and the correlations R (x)= R (q)e−iq·x, (2.9) ij ij Z q where q = (q ,q ) is an in-plane two dimensional wavevector, q q = q2+q2, and 1 2 ≡ | | 1 2 (2π)−2 dq dq . The Fourier integrals for η exclude q = 0 due to thpe incompress- q ≡ 1 2 Ribility constrRaint η(x,t)=0, while the q=0 mode for ϕ gives an irrelevant constant x shifttothevelocitRypotential.Notethatϕin(2.7)manifestlysatisfies(2.1).Substituting Water wave transport 5 (2.8) into the boundary conditions (2.4), we obtain, ∂ η(k,t) i η(k q)U(q) k i√ε η(k q/ε)δU(q) k=ϕ(k,t)ktanhεkh t − Z − · − Z − · q q ∂ ϕ(k,t) i U(q) (k q)ϕ(k q) i√ε δU(q) (k q/ε)ϕ(k q/ε) t − Z · − − − Z · − − q q = (εk2+ε−1)η(k). − (2.10) where the δU(q) are correlated according to δU (p)δU (q) =R (p)δ(p+q). (2.11) i j ij h i | | SincethecorrelationR (x)issymmetricini j,anddependsonlyuponthemagnitude ij ↔ x, R (p q) is real. ij | | | − | In the case where δU = 0 and U(x) U is strictly uniform, equations (2.10) can 0 ≡ be simplified by assuming a e−iωt dependence for all dynamical variables. Uniform drift yields the familiar capillary-gravitywave dispersion relation ω(k)= (k3+k)tanhkh+U k Ω(k)+U k. (2.12) 0 0 · ≡ · p However,forwhatfollows,wewishtoderivetransportequationsforsurfacewaves(action, energy,intensity)inthepresenceofaspatiallyvaryingdriftcontainingtwolengthscales: U=U(x)+√εδU(x/ε). 3. The Wigner distribution and asymptotic analyses Theintensityofthedynamicalwavevariablescanberepresentedbytheproductoftwo Greenfunctionsevaluatedatpositionsx εy/2.Thedifferenceintheirevaluationpoints, ± εy, resolves the waves of wavevector k 2π/(εy). Elter & Molyneux (1972) used this | | ∼ representationtostudyshallowwaterwavepropagationoverarandombottom.However, for the arbitrary depth surface wave problem, where the Green function is not simple, andwheretwolengthscalesaretreated,itisconvenienttousetheFourierrepresentation of the Wigner distribution (Wigner (1932), G´erard et al. (1997), Ryzhik, Papanicolaou, & Keller (1996)). Define ψ =(ψ ,ψ ) (η(x),ϕ(x,z =0)) and the Wigner distribution: 1 2 ≡ εy εy W (x,k,t) (2π)−2 eik·yψ x ,t ψ∗ x+ ,t dy (3.1) ij ≡ Z i(cid:16) − 2 (cid:17) j (cid:16) 2 (cid:17) where x is a central field point from which we consider two neighbouring points x εy, ± 2 and their intervening wavefield. Fourier transformingthe x variable using the definition (2.7) we find, p k p k W (p,k,t)=(2πε)−2ψ ,t ψ∗ ,t . (3.2) ij i(cid:18)2 − ε (cid:19) j (cid:18)−2 − ε (cid:19) The total wave energy, comprising gravitational, kinetic, and surface tension contribu- 6 Bal & Chou tions is 1 1 0 = η(x)2+ η(x)2 + dz U(x,z)+ˆzU (x,z)+v(x,z)2 x z E 2Z |∇ | | | 2Z Z | | x(cid:2) (cid:3) x −h 1 0 dz U(x,z)+ˆzU (x,z)2 (3.3) z −2Z Z | | x −h 1 = (k2+1)η(k)2+ktanhkhϕ(k,z =0)2. 2Z | | | | k The energyabovehas beenexpandedto anorderin η(x,t) andϕ(x,z,t)consistentwith the approximationsusedto derive(2.4).In arrivingatthe lastequalityin (3.3),we have integratedbyparts,usedthe Fourierdecompositions(2.7)andimposedanimpenetrable bottomconditionatz = h.Thewaveenergydensitycarriedbywavevectorkis(G´erard − et al. (1997)) 1 (k,t)= Tr[A(k)W(k,t)], (3.4) E 2 whereA (k)=k2+1,A (k)=ktanhkh,A =A =0.Thus,theWignerdistribution 11 22 12 21 epitomises the local surface wave energy density. In the presence of slowly varying drift, we identify W(x,k,t) as the local Wigner distribution at position x representing waves of wavevector k. The time evolution of its Fourier transformW(p,k,t), canbe derivedby considering time evolutionof the vector field ψ implied by the boundary conditions (2.4): ψ˙ (k,t)+iL (k)ψ (k,t)= i U(q) (k qδ )ψ (k q,t) j jℓ ℓ j2 j Z · − − q (3.5) +i√ε δU(q) (k qδ /ε)ψ (k q/ε,t), q · − j2 j − R where the operator L(k) is defined by 0 ik tanhεkh | | | | L(k)= . (3.6) i(εk2+ε−1) 0  −  We have redefined the physical wavenumber to be k/ε so that k O(1). Upon using ∼ (3.5) and the definition (3.2), (see Appendix A) k p k p W˙ (p,k,t) =iW (p,k,t)L† + iL W (p,k,t) ij iℓ ℓj(cid:18)ε 2(cid:19)− iℓ(cid:18)ε − 2(cid:19) ℓj k p +i U(q) + qδ W (p q,k+εq/2,t) i2 ij Z ·(cid:18)−ε 2 − (cid:19) − q k p i U(q) +qδ W (p q,k εq/2,t) (3.7) j2 ij − Z ·(cid:18)−ε − 2 (cid:19) − − q k p q +i√ε δU(q) + δ W (p q/ε,k+q/2,t) i2 ij Z ·(cid:18)−ε 2 − ε (cid:19) − q k p q i√ε δU(q) + δ W (p q/ε,k q/2,t), j2 ij − Z ·(cid:18)−ε − 2 ε (cid:19) − − q where only the index ℓ= 1,2 has been summed over. If we now assume that W(x,k,t) can be expanded in functions that vary independently at the two relevant length scales, Water wave transport 7 functions of the field p (dual to x) can be replaced by functions of a slow variation in p and a fast oscillation ξ/ε; p p+ξ/ε. → This amountstothe Fourierequivalentofatwo-scaleexpansioninwhichxisreplaced by x and y = x/ε (Ryzhik, Papanicolaou, & Keller (1996)). The two new independent wavevectors p and ξ are both of O(1). Expanding the Wigner distribution in powers of √ε and using p p+ξ/ε, → W(p,k,t) W (p,ξ,k,t)+√εW (p,ξ,k,t)+εW (p,ξ,k,t)+O(ε3/2), (3.8) 0 1/2 1 → we expand each quantity appearing in (3.7) in powers of √ε and equate like powers. Uponexpandingtheoff-diagonaloperatorL( k/ε+p/2)=ε−1L (k)+L (k,p)+O(ε), 0 1 − where 0 iktanhkh 0 ip kf(k) · L0(k)= , L1(k,p)   (3.9) ≡ i(k2+1) 0 ip k 0  −   ·  and hk+sinhkhcoshkh f(k) . (3.10) ≡− 2kcosh2kh 3.1. Order ε−1 terms The terms of O(ε−1) in (3.7) are ξ W (p,ξ,k,t)L†(k ) L (k )W (p,ξ,k,t)=0, k k (3.11) 0 0 + − 0 − 0 ± ≡ ± 2 Tosolve(3.11),weusetheeigenvaluesandnormalisedeigenvectorsforL anditscomplex 0 adjoint L†, 0 iτ α(k)/2 iτ τΩ(k) iγ, b = p ; τΩ(k)+iγ, c = 2α(k) , (3.12) − τ 1 τ p  2α(k)   α(k)/2  p p Ω(k) where α(k) , τ = 1, and iγ 0 is a small imaginary term. A W (p,ξ,k,t) ≡ k2+1 ± → 0 that manifestly satisfies (3.12) can be constructed by expanding in the basis of 2 2 × matrices composed from the eigenvectors: W0(p,ξ,k,t) = aττ′(p,k,t)bτ(k−)b†τ′(k+). (3.13) τ,Xτ′=± Right[left] multiplying (3.11) (using (3.13)) by the eigenvectors of the adjoint prob- lem, c (k ) c†(k ) , we find that a = a = 0, and a (x,k,t) a (x,k,t) = τ − τ + +− −+ −− ≡ − a++(x, k,t)(cid:2) a+(x(cid:3), k,t). Furthermore, a+,a− = 0 only if ξ = 0. Thus W0 has the − ≡ − 6 form W (p,ξ,k,t)=W (p,k,t)δ(ξ). (3.14) 0 0 FromthedefinitionofW ,weseethatthe(1,1)componentofW isthelocalenvelop 0 0 oftheensembleaveragedwaveintensity η(x,k,t)2 a (x,k,t)α(k).Similarly,fromthe + | | ≃ energy (Eq. (3.4)), we see immediately that the local ensemble averagedenergy density (x,k,t) =A (k)α(k) a(x,k,t) +A (k) a(x,k,t) 11 22 hE i h i h i (3.15) =Ω(k) a(x,k,t) . h i 8 Bal & Chou Therefore,sincethestartingdynamicalequationsarelinear,wecanidentify a(x,k,t) as h i the ensemble averagedlocalwave action associatedwith wavesof wavevectork (Henyey et al. (1988)). The wave action a(x,k,t) , rather than the energy density (x,k,t) h i hE i is the conserved quantity (Longuet-Higgins & Stewart (1961), Mei (1979), Whitham (1974)). The physical origin of γ arises from causality, but can also be explicitly derived from considerations of an infinitesimally small viscous dissipation (Chou, Lucas & Stone (1995)). Although we have assumed γ 0, for our model to be valid, the viscosity → need only be small enough such that surface waves are not attenuated before they have a chance to multiply scatter and enter the transport or diffusion regimes under consid- eration. This constraintcan be quantified by noting that in the frequency domain, wave dissipation is givenby γ =2νk2 (Landau (1985)) where ν is the kinematic viscosity and c (k) Ω(k) (3.16) g k ≡|∇ | isthegroupvelocity.Thecorrespondingdecaylengthk−1 c (k)/(νk2)mustbegreater d ∼ g than the relevant wave propagationdistance. Therefore, we require ε2c (k/ε) g (1,ε−1) (3.17) νk2 ≫ for(transport,diffusion)theoriestobevalid.Theinequality(3.17)givesanupperbound for the viscosity νk2 (εc (k/ε),ε2c (k/ε)) (3.18) g g ≪ which is most easily satisfied in the shallow water wave regime for transport. Otherwise we must at least require ν <o(√ε). The upper bounds for ν (and hence γ) given above provide one criterion for the validity of transport theory. 3.2. Order ε−1/2 terms Collecting terms in (3.7) of order ε−1/2, we obtain W (p,ξ,k,t)L†(k ) L (k )W (p,ξ,k,t)+ U(q) ξW (p q,ξ q,k,t) 1/2 0 + − 0 − 1/2 Z · 1/2 − − q δU(q) k W (p,ξ q,k+q/2,t)+ δU(q) k W (p,ξ q,k q/2,t) − 0 + 0 −Z · − Z · − − q q δU(q) q[W (p,ξ q,k+q/2,t)S+SW (p,ξ q,k q/2,t)]=0 0 0 −Z · − − − q (3.19) 0 0 where S= . (cid:20) 0 1 (cid:21) Similarly decomposing W in the basis matrices composed of b (k )b† (k ) (as in 1/2 τ − τ′ + 3.13), substituting W (p,0,k,t)δ(ξ) from (3.13 into (3.19), and inverse Fourier trans- 0 forming in the slow variable p, we obtain W1/2(x,k,ξ,t)= δτU′Ω(ξ()k·Γ)τ,τ′τ(Ωx,(kξ,k),+t)bUτ((xk)−)ξb+†τ′(2ki+γ), (3.20) τ,Xτ′=± + − − · Water wave transport 9 where Γτ,τ′(x,ξ,k,t)≡k−aτ′(x,k+,t)c†τ(k−)bτ′(k+)−k+aτ(x,k−,t)b†τ(k−)cτ′(k+) ξ +2 aµ(x,k+,t)c†τ(k−)bµ(k+)+aµ(x,k−,t)b†µ(k−)cτ′(k+) . µX=±(cid:2) (cid:3) (3.21) 3.3. Order ε0 terms The terms of order ε0 in (3.7) read W˙ (p,k,t)=iW (p,k,t)L†( p) iL (p)W (p,k,t) i k U(q)q W (p q,ξ,k,t) 0 0 1 − − 1 0 − Z · ·∇k 0 − q +i U(q) pW (p q,ξ,k,t) i U(q) q[SW (p q,ξ,k,t)+W (p q,ξ,k,t)S] 0 0 0 Z · − − Z · − − q q +i δU(q) k W (p,ξ q,k q/2,t) i δU(q) k W (p,ξ q,k+q/2,t) + 1/2 − 1/2 Z · − − − Z · − q q δU(q) q SW (p,ξ q,k+q/2,t)+W (p,ξ q,k q/2,t)S 1/2 1/2 −Z · − − − q (cid:2) (cid:3) +iW L† iL W + U(q) ξW (p q,ξ,k,t). 1 0− 0 1 Z · 1 − q (3.22) To obtain an equation for the statistical ensemble average a (x,k,t) , we multiply + h i (3.22)byc† (k)ontheleftandbyc (k)ontherightandsubstituteW fromequation + + 1/2 (3.20). We obtain a closed equation for a(x,k,t) a (x,k,t) (we henceforth suppress + ≡h i the ... notationfora(x,k,t)and (x,k,t))bytruncatingtermscontainingW .Clearly, 1 h i E from (3.12), c†(k)(iW L† iL W )c (k) = 0. Furthermore, we assume ξW (p + 1 0 − 0 1 + h 1 − q,ξ,k,t) 0 which followsfromergodicity of dynamicalsystems,andhas been used in i≈ the propagation of waves in random media (see Ryzhik, Papanicolaou, & Keller (1996), Bal et al. (1999)). The transport equations resulting from this truncation are rigorously justified in the scalar case (Spohn (1977), Erdo¨s & Yau (1998)). 4. The surface wave transport equation The main mathematical result of this paper, an evolution equation for the ensemble averagedwave action a(x,k,t) follows from equation (3.22) above (cf. Appendix A) and reads, a˙(x,k,t)+ ω(x,k) a(x,k,t) ω(x,k) a(x,k,t) k x x k ∇ ·∇ −∇ ·∇ (4.1) = Σ(k)a(x,k,t)+ σ(q,k)a(x,q,t), − Z q where ω(x,k)= (k3+k)tanhkh+U(x) k Ω(k)+U(x) k. (4.2) · ≡ · p The left hand side in (4.1) corresponds to wave action propagation in the absence of random fluctuations. It is equivalent to the equations obtained by the ray theory, or a WKB expansion (see section 5.1). The two terms on the right hand side of (4.1) representrefraction,or“scattering”ofwaveactionoutofandintowaveswithwavevector 10 Bal & Chou krespectively.Inderiving(4.1)wehaveinverseFouriertransformedbacktotheslowfield pointvariablex,andusedtherelation(α(k) f(k)α−1(k))k Ω(k).Toobtain(4.1), k − ≡∇ we assumed R (q)q = R (q)q = 0, which would always be valid for divergence-free ij i ij j flowsintwo dimensions.Althoughthe perturbationδUis notdivergence-freeingeneral, δU(x,z =0) = ∂ δU (x,0) = 0, using symmetry considerations, we will show in z z ∇· − 6 section 5.2 that R (q)q =R (q)q =0. ij i ij j Explicitly, the scattering rates are Σ(k) 2π q R (q k)k b†(k)c (q)b†(q)c (k)δ(τω(x,τq) ω(x,k)) ≡ Z i ij − j + τ τ + − q τX=± σ(q,k) 2π τq R (τq k)k b†(τq)c (k)2δ(τω(x,q) ω(x,k)) ≡ i ij − j| τ + | − τX=± (4.3) where (τα(k)+α(q))(τα(q)+α(k)) b† (k)c (q)b†(q)c (k) = + τ τ + 4α(k)α(q) (4.4) (τα(q)+α(k))2 |b†τ(k)cτ′(q)|2 = 4α(k)α(q) . Physically, Σ(k) is a decay rate arising from scattering of action out of wavevector k. The kernel σ(q,k) represents scattering of action from wavevector q into action with wavevectork. Note that the slowly varying drift U(x) enters parametrically in the scat- teringviaω(x,k)intheδ functionsupports.Theargumentsω(x,k)intheδ functions − − mean that we can consider the transport of waves of each fixed frequency ω ω(x,k) 0 ≡ independently. The typical distance travelled by a wave before it is significantly redirected is defined by the mean free path c (k) g ℓ = O(1). (4.5) mfp Σ(k) ∼ Themeanfreepathdescribedherecarriesadifferentinterpretationfromthatconsidered inweaklynonlinear,ormultiplescatteringtheories(Zakharov,L’vov&Falkovich(1992)) where one treats a low density of scatterers. Rather than strong, rare scatterings over every distance ℓ O(1), we have considered constant, but weak interaction with mfp ∼ an extended, random flow field. Although here, each scattering is O(ε) and weak, over a distance of O(1), approximately ε−1 interactions arise, ultimately producing ℓ mfp ∼ O(1). 5. Results and Discussion We have derived transport equations for water wave propagation interacting with static, random surface flows containing two explicit length scales. We have further as- sumed that the amplitude of δU scales as εβ with β = 1/2: The random flows are correspondinglyweakenedas the high frequency limit is taken.Since scattering strength is proportional to the power spectrum of the random flows and is quadratic in δU, the mean free path can be estimated heuristically by ℓ c (k)/Σ(k)ε1−2β. For β >1/2, mfp g ∼ thescatteringistooweakandthemeanfreepathdiverges.Inthislimit,wavesarenearly freely propagatingandcan be described by the slowly varyingflows alone, or WKB the- ory.Ifβ <1/2,ℓ 0andthescatteringbecomessofrequentthatoverapropagation mfp → distance of O(1), the large number of scatterings lead to diffusive (cf. Section 5.4) be-