1,257
Views
5
CrossRef citations to date
0
Altmetric
Articles

Stochastic homogenisation of high-contrast media

, &
Pages 91-117 | Received 07 Dec 2017, Accepted 27 Jun 2018, Published online: 06 Aug 2018

ABSTRACT

Using a suitable stochastic version of the compactness argument of [Zhikov VV. On an extension of the method of two-scale convergence and its applications. Sb Math. 2000;191(7–8):973–1014], we develop a probabilistic framework for the analysis of heterogeneous media with high contrast. We show that an appropriately defined multiscale limit of the field in the original medium satisfies a system of equations corresponding to the coupled ‘macroscopic’ and ‘microscopic’ components of the field, giving rise to an analogue of the ‘Zhikov function’, which represents the effective dispersion of the medium. We demonstrate that, under some lenient conditions within the new framework, the spectra of the original problems converge to the spectrum of their homogenisation limit.

COMMUNICATED BY:

MATHEMATICS SUBJECT CLASSIFICATION (2010):

1. Introduction

Asymptotic analysis of differential equations with rapidly oscillating coefficients has featured prominently among the interests of the applied analysis community during the last half a century. The problem of understanding and quantifying the overall behaviour of heterogeneous media has emerged as a natural step within the general progress of material science, wave propagation and mathematical physics. In this period several frameworks have been developed for the analysis of families of differential operators, functionals and random processes describing multiscale media, all of which have benefitted from the invariably deep insight and mathematical elegance of the work of V. V. Zhikov. In the present paper, we touch upon two subjects in which his contributions have inspired generations of followers: the stochastic approach to homogenisation, in particular through his collaboration with S. M. Kozlov during the 1980s, and the analysis of differential operators describing periodic composites with high contrast, which started with his fundamental contribution [Citation1].

Our present interest in the context of stochastic homogenisation of high-contrast composites stems from the relationships that have recently been indicated between media with negative material properties (‘metamaterials’), and more generally time-dispersive media, and ‘degenerate’ families of differential operators, where e.g. loss of uniform ellipticity of the symbol is known to lead to non-classical dispersion relations in the limit of vanishing ratio ϵ of the microscopic (l) and macroscopic (L) lengths: ε=l/L0. The work [Citation1] has provided an example, in the periodic context, of what one should expect in the limit as ε0 in terms of the two-scale structure of the solution as well as the spectrum of the related differential operator, in the case when the metamaterial is modelled by disjoint ‘soft’ inclusions with low, order O(ε2) values of the material parameters (say, elastic constants in the context of linearised elasticity), embedded in a connected ‘stiff’ material with material constants of order O(1). In mathematical terms, the coefficients of the corresponding differential expression alternate between values of different orders in ϵ, where the contrast increases as ϵ gets smaller.

In the present article, we introduce a stochastic framework for the analysis of homogenisation problems with soft inclusions and explore the question on what version of the results of [Citation1] can be achieved in this new framework. In particular, we are interested in the equations that describe the stochastic two-scale limit, in an appropriate sense, of the sequence of solutions to the probabilistic version of a Dirichlet problem in a bounded domain of Rn. Furthermore, we show that the spectra of such problems converge, in the Hausdorff sense, to the spectrum of the limit problem, which we analyse in a setting that models distributions of soft inclusions whose shapes are taken from a certain finite set and whose sizes vary over an interval. To our knowledge, the present manuscript is the first work containing an analysis of random heterogeneous media with high contrast that results in a ‘complete’ Hausdorff-type convergence statement for the spectra of the corresponding differential operators. Various aspects of multiscale analysis of high-contrast media in the stochastic context have been looked at in a handful of papers, e.g. [Citation2–4].

While in the periodic context norm-resolvent convergence results been obtained for high-contrast media, see [Citation5,Citation6], the stochastic case remains open to developments of a similar nature. It is anticipated that the operator-theoretic approach to problems of the kind we discuss in the present article will provide a general description of the types of spectral behaviour that can occur in the real-world applications where it is difficult to enforce periodicity of the microstructure. On the other hand, as we show in the present work, new wave phenomena should be expected in the stochastic setting (e.g. a non-trivial continuous component of the spectral measure of the homogenised operator for a bounded-domain problem), which makes the related future developments even more exciting.

Next, we outline the structure of the paper. In Section 2 we recall the notion of stochastic two-scale convergence, which we use, in Sections 3 and 4, to pass to the limit, as ε0, in a family of homogenisation problems with random soft inclusions. In Section 3 we give a formulation of the high-contrast problem we study and provide some auxiliary statements. In Section 4 we describe the limit problem and prove the strong resolvent convergence of the ϵ-dependent family to the limit system of equations. In Section 5 we provide a link between the spectra of the Laplacian operator on realisations of the inclusions and of the corresponding stochastic Laplacian. In Section 6 we prove that sequences of normalised eigenfunctions of the ϵ-dependent problems are compact in the sense of strong stochastic two-scale convergence. Finally, in Section 7 we discuss two examples of the general stochastic setting and describe the structure of the corresponding limit spectrum.

In conclusion of this section, we introduce some notation used throughout the paper. For a Banach space X and its dual X, we denote by X,X the corresponding duality. For a Hilbert space H the inner product of a,bH is also denoted by a,bH and, if H=Rn, by ab. For a set O we denote by χO its characteristic function, which takes value one on the set O and zero on the complement to O in the appropriate ambient space. For DRn we denote by D¯ its closure and by |D| its Lebesgue measure. Further, we use the notation Br(0) for the ball in Rn of radius r with the centre at the origin; Y denotes the cube [0,1)n with torus topology, where the opposite faces are identified; and N0l:={0,,l}. For an operator A on some Hilbert space, we denote by SpA its spectrum. Finally, for a Lipschitz open set DRn, we denote by ΔD the (positive) Laplace operator with the Dirichlet boundary condition on D. For xRn, we denote by [x] the element of Zn which satisfies [x]x<[x]+(1,,1). For k=1,,n, by ek we denote the kth coordinate vectors.

2. Stochastic two-scale convergence

2.1. Probability framework

Let (Ω,F,P) be a complete probability space. We assume that F is countably generated, which implies that the spaces Lp(Ω), p[1,), are separable. For a function uL1(Ω), we will sometimes write u for Ωu.

Definition 2.1

A family (Tx)xRn of measurable bijective mappings Tx:ΩΩ on a probability space (Ω,F,P) is called a dynamical system on Ω with respect to P if:

  1. TxTy=Tx+y  x,yRn;

  2. P(TxF)=P(F)  xRn, FF;

  3. T:Rn×ΩΩ,  (x,ω)Tx(ω) is measurable (for the standard σ-algebra on the product space, where on Rn we take the Lebesgue σ-algebra).

We next define the notion of ergodicity for dynamical systems introduced above.

Definition 2.2

A dynamical system is called ergodic if one of the following equivalent conditions is fulfilled:

  1. f measurable, f(ω)=f(Txω)  xRn,a.e.ωΩf(ω)isconstant Pa.e.ωΩ.

  2. P((TxBB)(TxBB))=0  xRnP(B){0,1}.

Henceforth we assume that the dynamical system Tx is ergodic.

Remark 2.1

Note that for the condition (b) the implication P(B){0,1} has to hold, if the symmetric difference between TxB and B is a null set. It can be shown (see, e.g. [Citation7]) that ergodicity is also equivalent to an a priori weaker implication TxB=BxRnP(B){0,1}.

For fLp(Ω), we write f(x,ω):=f(Txω), defining the realisation fLlocp(Rn,Lp(Ω)). There is a natural unitary action on L2(Ω) associated with Tx: (1) U(x)f=fTx,fL2(Ω).(1) It can be shown that the conditions of Definition 2.1 imply that this is a strongly continuous group (see [Citation8]). It is often necessary that the set of full measure be invariant in the sense that together with the point ω it contains the whole ‘trajectory’ {Txω,xRm}. This requirement can always be met on the basis of the following simple lemma (see [Citation8, Lemma 7.1]).

Lemma 2.1

Let Ω0 be a set of full measure in Ω. Then there exists a set of full measure Ω1 such that Ω1Ω0, and for a given ωΩ1 we have TxωΩ0 for almost all xRm.

For each j=1,2,n, we define the infinitesimal generator Dj of the unitary group {U(x)}xRn by the formula (2) Djf(ω)=limxj0f(Txjω)f(ω)xj,fL2(Ω),(2) where the limit is taken in L2(Ω). Notice that iDj, j=1,,n, are commuting, self-adjoint, closed and densely defined linear operators on the separable Hilbert space L2(Ω). The domain Dj(Ω) of such an operator is given by the set of L2(Ω)-functions for which the limit (Equation2) exists. We consider the set (3) W1,2(Ω):=j=1nDj(Ω)(3) and similarly Wk,2(Ω):={fL2(Ω):D1α1DnαnfL2(Ω),α1++αn=k},W,2(Ω):=kNWk,2(Ω). It is shown by the standard semigroup property that W,2(Ω) is dense in L2(Ω). We also define the space C(Ω):={fW,2(Ω):(α1,,αn)N0n D1α1DnαnfL(Ω)}. By the smoothening procedure discussed in [Citation8, p.232] (see also the text before Lemma 3.1 below), it is shown that C(Ω) is dense in Lp(Ω) for all p[1,) as well as in Wk,2(Ω) for all k. Furthermore, it is shown that W1,2(Ω) is separable. Notice that, due to the infinitesimal generator being closed, Djf can be equivalently defined as the function that satisfies the property (4) ΩDjfg=ΩfDjggC(Ω).(4) If fW1,2(Ω), we may also define Djf(x,ω):=Djf(Txω) for all xRn. It can be shown that the following identity holds (see [Citation9]): (5) W1,2(Ω)={fWloc1,2(Rn,L2(Ω)):f(x+y,ω)=f(x,Tyω)  x,y, a.e.ω}={fC1(Rn,L2(Ω)):f(x+y,ω)=f(x,Tyω)  x,y, a.e.ω}.(5) Moreover, for a.e. ωΩ the function Dif(,ω) is the distributional derivative of f(,ω): a proof of this fact can be found in [Citation9, Lemma A.7].

Following [Citation10], we denote by #,k,2 the seminorm on C(Ω) given by u#,k,22=αNn, α=kDαuL2(Ω)2. By Wk,2(Ω) we denote the completion of C(Ω) with respect to the seminorm #,k,2. The gradient operator ω:=(D1,,Dn) and the operator divω:=ω are extended uniquely by continuity to mappings from W1,2(Ω) to L2(Ω,Rn) and from W1,2(Ω,Rn) to L2(Ω), respectively. Finally, by a density argument, it is easily seen that W1,2(Ω) is also the completion of W1,2(Ω) with respect to #,1,2.

2.2. Definition and basic properties

The key property of ergodic systems is the following theorem, due to Birkhoff (for a more general approach, see [Citation11]).

Theorem 2.1

‘Ergodic Theorem’

Let (Ω,F,P) be a probability space with an ergodic dynamical system (Tx)xRn on Ω. Let fL1(Ω), and let BRn be a bounded open set. Then for P-a.e. ωΩ one has (6) limε0Bf(Tx/εω)dx=|B|Ωf(ω)dP(ω).(6) Furthermore, for all fLp(Ω),1p, and a.e. ωΩ, the function f(x,ω)=f(Txω) satisfies f(,ω)Llocp(Rn). For p< one has f(/ε,ω)=f(T/εω)ΩfdP weakly in Llocp(Rn) as ε0.

The elements ω such that (Equation6) holds for every fL1(Ω) and bounded open BRn are refereed to as typical elements, while the corresponding sets (Txω)xRN are called typical trajectories. Note that the separability of L1(Ω) implies that almost every ωΩ is typical, and in what follows we only work with such ω.

For vector spaces V1,V2, we denote by by V1V2 their usual tensor product. We define the following notion of stochastic two-scale convergence, which is a slight variation of the definition given in [Citation12]. In [13], the authors average over the probability space and do not use the Birkhoff Ergodic Theorem. As a consequence, they do not obtain convergence almost everywhere but only in mean, which results in a weaker notion of stochastic two-scale convergence than the one introduced in [12]. In the context of calculus of variations, the first results are obtained in [14,15]. The authors of these papers do not use stochastic two-scale convergence at all, as this was introduced later on, but rely on a formula for non-periodic homogenisation for a.e. ω∈Ω as well as on the ergodic theorem. We shall stay in the Hilbert setting (p=2), as it suffices for our analysis. Finally, we denote by S a bounded open LIpschitz set in Rn.

Definition 2.3

Let (Txω)xRn be a typical trajectory and (uε) a bounded sequence in L2(S). We say that (uε) weakly stochastically two-scale converges to uL2(S×Ω) and write uε2u, if (7) limε0Suε(x)g(x,Tx/εω)dx=ΩSu(x,ω)g(x,ω)dxdP(ω)  gC0(S)C(Ω).(7)

If additionally uεL2(S)uL2(S×Ω), we say that (uε) strongly stochastically two-scale converges to u and write uε2u.

Remark 2.2

The convergence of (uε) is defined along a fixed typical trajectory and a priori the limit depends on this trajectory. In applications, such as the analysis of the PDE family in Section 4, it often turns out that the limit does not depend on the trajectory chosen. For this reason, and to simplify notation, in what follows we often do not indicate this dependence explicitly.

Note also that, by density, the set of admissible test functions g in (Equation7) can be extended to L2(S)L2(Ω).

In the next proposition, we collect the properties of stochastic two-scale convergence that we use in the present work.

Proposition 2.1

The following properties of stochastic two-scale convergence hold.

  1. Let (uε) be a bounded sequence in L2(S). Then there exists a subsequence (not relabelled) and uL2(S×Ω) such that uε2u.

  2. If uε2u then uL2(S×Ω)lim infε0uεL2(S).

  3. If (uε)L2(S) is a bounded sequence with uεu in L2(S) for some uL2(S), then uε2u.

  4. If (vε)L(S) is uniformly bounded and vεv strongly in L1(S) for some vL(S), and (uε) is bounded in L2(S) with uε2u for some uL2(S×Ω), then vεuε2vu.

  5. Let (uε) be a bounded sequence in W1,2(S). Then on a subsequence (not relabelled) uεu0 in W1,2(S), and there exists u1L2(S,W1,2(Ω)) such that uε2u0+ωu1(,ω).

  6. Let (uε) be a bounded sequence in L2(S) such that εuε is bounded in L2(S,Rn). Then there exists uL2(S,W1,2(Ω)) such that on a subsequence (8) uε2u,εuε2ωu(,ω).(8)

Proof.

In view of analogies with the periodic case, we just give a sketch of the proof. A proof of (a) can be found in [Citation12, Lemma 5.1]. For the proof of (b), we take an arbitrary gC0(S)C(Ω) and calculate lim infε0S|uε(x)g(x,Tx/εω)|2dx=lim infε0|uε(x)|2dx2Suε(x)g(x,Tx/εω)dx+S|g(x,Tx/εω)|2dx=lim infε0|uε(x)|2dx2S×Ωu(x,ω)g(x,ω)dxdP(ω)+S×Ω|g(x,ω)|2dxdP(ω). We obtain the claim by approximating uL2(S×Ω) with functions gC0(S)C(Ω). The proof of (c), (d) is straightforward. The proof of (e) goes in the same way as in the periodic case, by the duality argument.

First, one proves that if fL2(Ω,Rn) is such that Ωfg=0g{gC(Ω,Rn):divωg=0}, then there exists ψW1,2(Ω) such that f=ωψ. One then proceeds in the same way as in the periodic case (see [Citation16]). In order to show the claim (f), take the subsequence such that uε2u, where uL2(S×Ω) and εuε2z, where zL2(S×Ω,Rn). We choose the test functions of the form ϕε(x)=a(x)b(Tx/εω), where aC0(S) and bC(Ω), and using integration by parts we conclude limε0Sεuε(x)ϕε(x)=SΩu(x,ω)a(x)ωb(ω)dxdP(ω)=SΩz(x)a(x)b(ω)dxdP(ω), from which the claim follows by a density argument, in view of the property (Equation4).

3. Problem formulation and auxiliary statements

Let SRn be a bounded open Lipschitz set. We take OΩ such that 0<P(O)<1 and for each ωΩ consider its ‘realisation’ Oω={xRn:TxωO}. We assume that the following conditions are satisfied.

Assumption 3.1

For a.e. ωΩ one has (9) Oω:=k=1Oωk,(9) where:

  1. Oωk, kN, are open connected sets with Lipschitz boundary;

  2. One has Oω¯=k=1Oωk¯;

  3. There exist c1,c2>0 such that c1diamOωkc2  kN;

  4. There exists a sequence of disjoint bounded domains Bωk such that Oωk¯Bωk, kN, and Cω>0 such that for all kN the following extension property holds: for all uW1,2(BωkOωk¯) there exists u~W1,2(Bωk) satisfying u~=uonBωkOωk¯,Bωk|u~|2CωBωkOωk|u|2,Δu~=0onOωk.

It is easily seen that Assumption 3.1 holds for the examples given in Section 7.1. Denote by Λ the set of typical elements ωΩ satisfying the conditions listed in Assumption 3.1, and for all ωΛ, ε>0 define S0ε(ω) as the union of all components εOωk that are subsets of S and stay sufficiently far from its boundary, in the sense that there exists C=C(ω)>0 such that (10) S0ε(ω):=kKωεεOωk,Kωε:={kN:εOωkS, dist(εOωk,S)>Cε}.(10) We denote the complement of the set S0ε(ω) by S1ε(ω):=SS0ε(ω)¯ and the corresponding set indicator functions by χ0ε(ω) and χ1ε(ω).

For each ωΛ, we consider the following Dirichlet problem in S: for λ<0 and fεL2(S), find uεW01,2(S) such that (11) SAε(,ω)uεvλSuεv=SfεvvW01,2(S),(11) where Aε(,ω)=χ1ε(ω)A1+ε2χ0ε(ω)I,ωΛ, with a symmetric and positive-definite matrix A1.

For all ωΛ we also define the Dirichlet operator Aε(ω) in L2(S) corresponding to the differential expression divAε(,ω)u, e.g. by considering the bilinear form SAε(,ω)uv,u,vW01,2(S).

It is well known that the spectrum of Aε(ω) is discrete. The following subspace of W1,2(Ω) will play an essential role in our analysis: W01,2(O)={vW1,2(Ω):v(Txω)=0onRnOω  ωΛ}. Notice that as a consequence of Ergodic Theorem (Theorem 2.1) one has W01,2(O)={vW1,2(Ω):χOv=v}, i.e. W01,2(O) consists of W1,2-functions that vanish on Ω∖O. Henceforth we assume that ωΛ without mentioning it explicitly.

The next two lemmas use a standard smoothening (or ‘mollification’) procedure, which we now describe. We take gL2(Ω) and (cf. [Citation8, p.232]) choose a nonnegative even function ρC0(Rn) with Rnρ=1, suppρB1(0) and write ρδ(x)=δnρ(x/δ) for all δ>0. For each δ>0, we define the mollification Rδ[g] of g by Rδ[g](ω)=Rnρδ(y)g(Tyω)dy,=Rnρδ(y)g(Tyω)dy,ωΩ. Notice that (12) Rδ[g](Txω)=Rnρδ(y)g(Txyω)dy=Rnρδ(xy)g(Tyω)dy=Rnρδ(yx)g(Tyω)dy,(12) from which we infer that DjRδ[g](ω)=Rnjρδ(y)g(Tyω)dy,ωΩ, j=1,2,,n. Arguing by induction, we show that Rδ[g]W,2(Ω), and if gL(Ω) then Rδ[g]C(Ω). Before we state and prove the lemmas, we introduce additional notation. We define the space C0(O):={vC(Ω):v=0onΩ∖O}, as well as the sets Dωk,m:=xOωk:dist(x,Oωk)>1m,k,mN. Also, for all mN we define the set Bm:={ωΩ:0Dωk,mfor somekN}Ω. By using the density of Qn in Rn it can be seen that for all mN the set Bm is measurable. Notice that for each fixed ω,k,m, where m is large enough, there exist constants C1,C2>0 such that (13) C1m|Oωk||ΩωkDωk,m|C2m|Oωk|.(13) In the next lemma we assume that a relaxed version of the right inequality in (Equation13) holds uniformly in ω.

Lemma 3.1

Suppose that for a.e. ωΩ there exists a sequence of positive values Cm converging to zero, such that |OωkDωk,m|Cm|Oωk|kN. Then the set C0(O) is dense in L2(O).

Proof.

Using Ergodic Theorem and the assumption of the lemma, it can be shown that P(OBm)0 as m. To prove the density, it suffices to approximate g:=χBmf, where fL(O) by a function from C0(O), for which we use the above mollification procedure. Notice that for δ>0 small enough, one has Rδ[g]C0(O). It remains to check Rδ[g]g as δ0, but this follows from the strong continuity of the group U(x), see (Equation1): gRδ[g]L2(Ω)Rnρδ(y)U(y)ggL2(Ω)dy0, as required.

Notice that, by the standard Poincaré inequality, for each Dωk,m there exists C>0 such that (14) OωkDωk,mu2dxC|OωkDωk,m|2OωkDωk,m|u|2dxuW01,2(Oωk).(14) In the following lemma we impose this condition uniformly.

Lemma 3.2

Assume that for a.e. ωΩ there exists a constant C>0 such that (15) |OωkDωk,m|Cm|Oωk|k,m,(15) and that (Equation14) is satisfied for all k and large enough m. Then the set C0(O) is dense in W01,2(O).

Proof.

We take fW01,2(O) and define fM:=χ|f|Mf+χ|f|MM. Notice that as a consequence of (Equation5), fMW01,2(O) and DjfM=χ|f|<MDjf. It is easily seen that fMf in W1,2(Ω) as M. Thus we can assume, without loss of generality, that fL(Ω)W01,2(O). We define hm=R1/2m[χBmf]. It can be seen from the proof of Lemma 3.1 that hmf in L2(O) as m. Notice that for a.e. ωB2m/3 we have, for i=1,,n, Djhm(ω)=R1/2m[Dif](ω)=Rnρ1/2m(y)Djf(Tyω)dy, and therefore DjhmDjfL2(B2m/3)R1/2m[Dif]DjfL2(Ω)0.

Notice also that for a.e. ωΩ there exist C1,C2>0 such that for all k,mN, where m is sufficiently large, we have jhm(,ω)L2(OωkDωk,2m/3)2C1m2fL2(OωkDωk,m/2)2C2fW1,2(OωkDωk,m/2)2, where we have used (Equation12), (Equation14), (Equation15) and Young's inequality. Using the Ergodic Theorem we conclude that there exists C>0 such that DjhmL2(OB2m/3)2CfW1,2(OBm/2)2, from which the claim follows.

4. Limit equations and two-scale resolvent convergence

We define the quadratic form A1homξξ:=infϕW1,2(Ω)Ω∖OA1(ξ+ωϕ)(ξ+ωϕ),ξRn, and denote by DW1,2(Ω∖O) the completion of DW1,2(Ω):={ωϕ:ϕW1,2(Ω)} with respect to the seminorm ωϕL2(Ω∖O), ϕW1,2(Ω). The proof of the following lemma is straightforward.

Lemma 4.1

For each ξRn there exists pξDW1,2(Ω∖O) such that A1homξξ=Ω∖OA1(ξ+pξ)(ξ+pξ), or, equivalently, (16) Ω∖OA1(ξ+pξ)ωϕ=0ϕC(Ω).(16) In particular, one has A1homA1.

Remark 4.1

It follows from the observations in [Citation12, p.265–266] that if the following extension property is satisfied for a.e. ωΩ: for all uC0(B1(0)) there exists εω0>0 and a sequence of functions (u~ε) such that u~ε=uinB1(0)kNεOωkforεεω0,B2(0)|u~ε|CωB1(0)kNεOωk|u|2, where Cω is a constant independent of u and ϵ, then the matrix A1hom is positive definite.

Notice that under Assumption 3.1, the extension property in Remark 4.1 is satisfied. We define the space H:=L2(S)+{uL2(S×Ω):u|S×(ΩO)=0}, which is clearly a direct sum, naturally embedded in L2(S×Ω). Before stating the next theorem we prove a simple lemma that implies that gives norm bounds for each component of H by the norm in L2(S×Ω).

Lemma 4.2

Let f0L2(S) and f1L2(S×Ω) such that f10 on Ω∖O. Then there exists a constant C>0 such that f0L2(S)+f1L2(S×Ω)Cf0+f1L2(S×Ω), where we use the natural embedding L2(S)L2(S×Ω).

Proof.

By Cauchy-Schwartz inequality we have 2S×Ω|f0(x)||f1(x,ω)|dxdP(ω)=2S|f0(x)|O|f1(x,)|dPdxf0L2(S)2+Of1L2(S)2f0L2(S)2+P(O)f1L2(S×Ω)2, and hence (17) f0+f1L2(S×Ω)2f0L2(S×Ω)22S×Ω|f0||f1|dxdP+f1L2(S×Ω)2(1P(O))f1L2(S×Ω)2.(17) It remains to bound f0L2(S×Ω) by f0+f1L2(S×Ω), which is done by the triangle inequality: f0L2(S)22(f0+f1L2(S)2+f1L2(S×Ω)2)2(2P(O))1P(O)f0+f1L2(S×Ω)2.

By P:L2(S×Ω)H we denote the orthogonal projection. For fL2(S×Ω) we have Pf(x,ω)=Ω∖Of(x,)dP+χO(ω)f(x,ω)Ω∖Of(x,)dP.

Theorem 4.1

Under Assumption 3.1, let λ<0 and suppose that (fε) is a bounded sequence in L2(S) such that fε2fL2(S×Ω). For each ε>0, consider the solution uε to (Equation11). Then for a.e. ωΩ one has uε2u0+u1(,ω), where u0W01,2(S),u1L2(S,W01,2(O)) satisfy (18) SA1homu0ϕ0λS(u0+u1Ω)ϕ0=SfΩϕ0ϕ0W01,2(S),(18) (19) Oωu1(x,)ωϕ1λO(u0(x)+u1(x,))ϕ1=Of(x,)ϕ1ϕ1W01,2(O).(19)

Remark 4.2

The system (Equation18) and (Equation19) is understood in the weak sense: (20) SA1homu0ϕ0+S×Ωωu1ωϕ1λS×Ω(u0+u1)(ϕ0+ϕ1)=S×Ωf(ϕ0+ϕ1)ϕ0W01,2(S), ϕ1L2(S,W01,2(O)).(20) Noting that W01,2(O) is a closed subspace of W1,2(Ω) and bearing in mind Lemma 4.2, it follows by the Lax-Milgram lemma that for all fL2(S×Ω), xS the problem (Equation20) has a unique solution in W01,2(O). Its solutions for the right-hand sides fL2(S×Ω) and Pf coincide. The solution of the Equation (Equation19) has the form (21) u1(x,ω)=u(x,ω)+u0(x)w(ω),(21) where uL2(S,W01,2(O)) is the solution of (Equation19) obtained by setting u0=0 and wW01,2(O) is the solution of (Equation19) obtained by setting u0=1 and f=0. Substituting (Equation21) into (Equation18), we obtain an equation on u0.

Proof.

The proof follows a standard argument. First, by (Equation11), there exists a constant C>0 such that (22) uεL2(S1ε)+εuεL2(S0ε)+uεL2(S)C.(22) For each ε>0 we extend uε|S1ε, using Assumption 3.1, to a sequence u~ε, which is bounded in W1,2(S). From Proposition 2.1 we infer that there exist u0W1,2(S), u1L2(S,W1,2(Ω)), u2L2(S,W1,2(Ω)) such that on a subsequence we have (23) u~εu0strongly inL2(S),u~ε2u0+ωu2,uεu~ε2u1,ε(uεu~ε)2ωu1.(23) To obtain the Equation (Equation18), we take test functions of the form ϕ0(x)+εa(x)ϕ(Tx/εω) in (Equation11), where ϕ0W01,2(S), ϕW1,2(Ω) and aC01(S). In the limit as ε0 we obtain (24) SΩ∖OA1(u0+ωu2)(ϕ0+aωϕ1)dPdxλS(u0+u1Ω)ϕ0=SfΩϕ0.(24) Setting ϕ0=0, it follows that Ω∖OA1(u0(x)+ωu2)ωϕ1dP=0a.e. xS, and the characterisation (Equation16) yields ωu2(x,ω)=pu0(x)(ω) a.e. xS, ωΩ. Taking arbitrary ϕ0W01,2(S) in (Equation24), we obtain the ‘macroscopic’ part (Equation18) of the limit problem. The ‘microscopic’ part (Equation19) is obtained by taking test functions of the form a(x)ϕ(Tx/εω) in (Equation11), where aC01(S), ϕW01,2(O). The convergence of the whole sequence can be deduced by uniqueness of the solution of the system (Equation18)–(Equation19).

Remark 4.3

The following observation was made in [Citation1] in the periodic setting. The formulation (Equation20) can be interpreted from the operator-theoretic point of view. Namely, we define a positive-definite operator A on a dense linear subset of V=W01,2(S)+L2(S,W01,2(O)) (which is a dense subset of H under the condition of Lemma 3.1), as follows. One takes λ<0 and defines the domain dom(A) as the set of solutions of (Equation20) obtained for varying fH. To see that dom(A) is dense in H, take the solutions u0+u1,w0+w1V for f,gH, respectively. Setting ϕ0=u0, ϕ1=u1 as the test function in the equation for (w0,w1) and ϕ0=w0, ϕ1=w1 as the test function in the equation for (u0,u1) yields S×Ωf(w0+w1)=S×Ωg(u0+u1). Thus, if gu0+u1 then necessarily w0+w1=0, which implies g=0. The operator A:dom(A)H defined by A(u0+u1)=f+λ(u0+u1) is a bounded linear mapping between Hilbert spaces, where the norm on dom(A) is given by u0+u1dom(A)2=A(u0+u1)H2+u0+u1V2.

We shall need the following statement for the convergence of spectra of the operators associated with (Equation11). It is proved in the same way as the previous theorem, and we omit the proof.

Proposition 4.1

Under Assumption 3.1, let λ<0 and suppose that (fε) is a bounded sequence in L2(S0ε) such that χ0εfε2fL2(S×Ω). For each ε>0, let zεW01,2(S0ε) be the solution of (25) ε2S0εzεvλS0εzεv=S0εfεvvW01,2(S0ε).(25) Then for a.e. ωΩ one has zε2z(,ω), where zL2(S,W01,2(O)) is the solution of the problem (26) Oωz(x,)ωvλOz(x,)v=Of(x,)vvW01,2(O).(26)

Remark 4.4

Theorem 4.1 and Proposition 4.1 are still valid if, instead of a fixed λ<0, we take a sequence (λε)R such that λελR and lim infε0dist(λε,SpAε)>0, for Theorem 4.1, i.e. lim infε0dist(λε,SpTε)>0 for Proposition 4.1, where Tε:=ε2ΔS0ε. Notice that SpTε splits into the spectra of scaled Laplace operators on each inclusion contained in S0ε: (27) SpTε=kKωεSp(ΔOωk),(27) where Kωε is defined in (Equation10). Notice that there exists C>0 such that for all λR the solution uε of (Equation11) satisfies (28) uεL2(S1ε)+εuεL2(S0ε)+uεL2(S)C(dist(λ,SpAε)1+λ+1)fεL2(S),(28) and similarly the solution of (Equation25) satisfies (29) εzεL2(S0ε)+zεL2(S0ε)C(dist(λ,SpTε)1+λ+1)fεL2(S).(29)

In what follows we denote by Δω the operator generated by the bilinear form (30) Oωuωv,u,vW01,2(O).(30) As a consequence of Proposition 4.1 and Remark 4.4, we have the following statement.

Corollary 4.1

Assume that Assumption 3.1 holds. Then Sp(Δω)kNSp(ΔOωk)¯a.e. ωΩ.

Proof.

Take λkNSp(ΔOωk)¯ and fL2(O), and define fε(x,ω):=χ0εf(Tx/εω)2f. As a consequence of Remark 4.4, the sequence of solutions of (Equation25) converges weakly two-scale to the solution of (Equation26), which is a resolvent equation. Moreover, (Equation27) and (Equation29) imply the existence of C>0 such that u1L2(S×Ω)lim infε0zεL2(S0ε)CfL2(O), and therefore λSp(Δω).

5. Spectral completeness for inclusions

Next we prove that kNSp(ΔOωk)¯Sp(Δω)a.e. ωΩ. We shall use the assumptions of Lemma 3.2 as well as assume that for each λ0>0 there exists Mλ0>0 such that for a.e. ωΩ the following implication holds: (31) Δu=λu,uW01,2(Oωk),for some kN,λλ0uL(Oωk)Mλ0uL2(Oωk).(31) Notice that, by regularity theory, the above condition is satisfied for a fixed ωΩ and kN, whenever the boundary Oωk is sufficiently regular. In what follows we use a sequence {ϕ~k}kNC0([0,c2+1]n) that is dense in W01,2([0,c2+1]n), where the constant c2 is defined in Assumption 3.1.

We will now define a sequence of random variables that is invariant for all ωO whose realisation is such that the shape that contains the origin is the same. For q=(q1,,qn)Qn define the set Oq:={ωO: there exists k0Nsuch that qOωk0}.

Lemma 5.1

For every qQn,OqΩ is measurable.

Proof.

Notice that (32) ωOqThere exists a polygonal line that connects0andqandconsists of a finite set of straight segments with rational endpointssuch that for alllQnon this line one hasTlωO.(32) Since for each fixed qQn there is a countable set of lines satisfying the property (Equation32), the set Oq is measurable.

We define the random variables Fj(ω):=inf{qj: ωOq},ωΩ, j=1,,n. Notice that Fj=+ whenever ωO, and also, due to the assumption, c2Fj0 for a.e. ωO. We denote by F the random vector (33) F:=(F1,,Fn)+(12,,12).(33) For a.e. ωO, mN we define the set Fωm:=xRn:there exists k0Nsuch that xOωk0and dist(x,Oωk0)>1m. Furthermore, we introduce the set Uω[0,c2+1]n, which is a translation of the set Oωk0 containing the origin: Uω:={x[0,c2+1]n:xFOωk0for k0Nsuch that 0Oωk0}. Finally, we define a characteristic function of the translation of the set Fωm and a measurable function of ω taking values in W01,2([0,c2+1]n): (34) χm(x,ω):=χDωm(xF),ϕk,m(x,ω):=ρ1/2m(χm(x,ω)ϕ~k(x)).(34) Notice that for a.e. ωO one has suppϕk,m(,ω)Uω.

Lemma 5.2

For every k,mN, the function ωϕk,m(,ω) taking values in W01,2([0,c2+1]n) is measurable with respect to the Borel σ-algebra on W01,2([0,c2+1]n).

Proof.

Firstly notice that (35) ωχm(,ω)ϕ~k(),(35) is a measurable mapping taking values in the set L2([0,c2+1]n), with Borel σ-algebra. To check this notice that for each qQn the set Lq:={ωΩ:qFωm}, is measurable: the related proof is similar to that of Lemma 5.1. Further, for ψC0(Rn) the norm ψχmϕ~kL2(Rn) is written as a limit of Riemann sums, and each Riemann sum can be written in terms of a finite number of χLq and values of function ϕ~k(). Thus ωψχmϕ~kL2(Rn) is measurable. Since the topology in L2(Rn) is generated by the balls of the form B(ψ,r), where ψC0(Rn) and rQ we have that the mapping given by (Equation35) is measurable. The claim follows by using the fact that the convolution is a continuous (and thus measurable) operator from L2 to W1,2.

Notice that by construction {ϕk,m(,ω)}k,mN C0(Uω) is a dense subset of W01,2(Uω) for a.e. ωΩ (see also the proof of Lemma 3.2). For 0ab we introduce the following subset of O: (36) Ea,b:={ωO:ΔOωk0 has an eigenvalue in[a,b]for k0Nsuch that 0Oωk0}.(36) For 0ab and a.e. ωEa,b we also define Sa,b,ωW01,2(Uω) as follows: Sa,b,ω:={ψW01,2(Uω):ψis an eigenfunction of ΔUωwhose eigenvalue is in [a,b]}. Finally, for every rR and k,mN we define the random variable (37) Xrk,m(ω):=Δϕk,m(,ω)rϕk,m(,ω)W1,2(Uω)ϕk,m(,ω)L2(Uω)ifϕk,m(,ω)0,+otherwise.(37)

Lemma 5.3

For every rR and k,mN, the function Xrk,m is measurable.

Proof.

We use Lemma 5.2 and the fact that Δ is a continuous map from W1,2 to W1,2 and W1,2(Uω) is a measurable function, since ψ(,ω)W1,2(Uω):=supk,mNW1,2(Uω)ψ(,ω),ϕk,m(,ω)W01,2(Uω)ϕk,m(,ω)W1,2(Uω):ϕk,m(,ω)0.

Lemma 5.4

For 0ab, the set Ea,b is measurable.

Proof.

The claim follows by observing that Ea,b=ωO:infk,mN,rQ[a,b]Xrk,m(ω)=0.

Now we are going to define a measurable mapping from O to the subspace Sa,b,ω. We set it to be an L2-projection onto Sa,b,ω of a specially chosen function of x and ω. We need the following measurability lemma.

Lemma 5.5

Assume that ωϕ(,ω) is a measurable function taking values in L2(Uω) for a.e. ωEa,b. Then the L2-distance ωdistL2(Uω)(ϕ(,ω),Sa,b,ω), ωEa,b, is a measurable map.

Proof.

The claim follows from the formula distL2(Uω)(ϕ(,ω),Sa,b,ω)=lim supninfk,mN1nϕk,m(,ω)ϕ(,ω)L2(Uω):Xrk,m(ω)<1nfor somerQ[a,b].

For 0ab and ωEa,b we define a measurable map ωϕa,b(,ω) as follows: ϕa,b(,ω)=ϕk0(ω),m0(ω)(,ω), where (38) k0(ω):=minkN{k:distL2(Uω)(ϕk,m(,ω),Sa,b,ω)ϕk,m(,ω)L2(Uω),12ϕk,m(,ω)L2(Uω)1for somemN},(38) and m0(ω) is the minimal value of m in (Equation38) Notice that in this way for a.e. ωEa,b the L2-projection of ϕa,b(,ω) on Sa,b,ω is not zero. We also define the random variable R:Ω[0,+) in the following way: R(ω):=dist(0,Oωk0)if0Oωk0for some k0N,0otherwise. By invoking the measurability of Oq, qQn, see Lemma 5.1, it is easily seen that R is indeed measurable. Next, for 0ab, l>0 we define the random variable ψa,b,l:ΩR by ψa,b,l(ω):=lim supnB(D,min{l,R(ω)})ϕk1(ω,n),m1(ω,n)(,ω)ifR(ω)>0,ωEa,b,0otherwise, where, for all nN, k1(ω,n):=minkNk:Xrk,m<1nfor somerQ[a,b],ϕk,m(,ω)ϕa,b(,ω)L2(Uω)<distL2(Uω)(ϕa,b(,ω),Sα,b,ω)+1nfor some mN, m1(ω) is the corresponding minimal valueFootnote1 of m, and B(D,min{l,R(ω)}) is the ball with the centre at D and radius min{l,R(ω)}, see (5.1). We also define ψa,b:=lim supl1N, l0ψa,b,l. Notice that in this way ψa,b is the value at the origin (taking into account for ωO the relative position of the origin with respect to the shape) of the (unique) L2-projection of ϕa,b onto Sa,b,ω. As a consequence of (Equation31), we have |ψa,b|Mb. Notice that by construction ψa,b0 if P(Ea,b)>0. We are ready for the proof of main statement.

Theorem 5.1

Under Assumption 3.1, the assumption of Lemma 3.2 and (Equation31), one has kNSp(ΔOωk)¯Sp(Δω)a.e. ωΩ.

Proof.

We take l0. There are two possibilities:

  1. There exists ε>0 such that Elε,l+ε has zero probability. In this case we denote ε0(l):=supε>0{ε:P(Elε,l+ε)=0}.

  2. For all ε>0 the set Elε,l+ε has positive probability.

In the case (a), by the continuity of probability, we conclude that P(Elε0(l),l+ε0(l)0)=0, where (cf. (Equation36)) Ea,b0:={ωO:ΔOωk0 has an eigenvalue in(a,b)for k0Nsuch that 0Oωk0}. By Lemma 2.1 and Corollary 4.1 we infer that (lε0(l),l+ε0(l))CkNSp(ΔOωk)¯CSp(Δω)a.e. ωΩ. In particular, we conclude that lSp(Δω).

In the case (b) we construct a Weyl sequence showing that lSp(Δω). To this end, we define ψn:=ψl1/n,l+1/nL2(O)1ψl1/n,l+1/n,nN. Then, by the above construction and using Ergodic Theorem, one has ΔωψnlψnL2(O)1n,nN. It follows from the above that Sp(Δω) consists of exactly those lR that satisfy the property (b). The set Sp(Δω) is closed, hence its complement is a countable union of open disjoint intervals. Every element of such an interval (d1,d2) satisfies the property (a) with l=(d1+d2)/2, ε0(l)=(d2d1)/2, and therefore P(Ed1,d20)=0. Using Lemma 2.1, we obtain (d1,d2)CkNSp(ΔOωk)¯a.e. ωΩ. The claim follows since there is only countable number of such intervals.

6. Convergence of spectrum

In our analysis we keep in mind the examples set in Section 7, for which it is shown that Sp(Δω)Sp(A). In the present section we assume that this holds, as well as the conclusion of Theorem 5.1, i.e. (39) Sp(Δω)=kNSp(ΔOωk)¯.(39) We are interested in approximating the spectra SpAε(ω) of the operators Aε(ω) (see Section 3) by the spectrum SpA of the limit operator. We claim that SpAε(ω)SpA for a.e. ωΩ, where the convergence is understood in the Hausdorff sense:

  1. For all λSpA there are λεSpAε(ω) such that λελ.

  2. If λεSpAε(ω) and λελ, then λSpA.

We prove this claim by adapting the argument of [Citation17]. First, we introduce the notion of strong resolvent convergence.

Definition 6.1

Let Aε(ω) and A be the operators acting on L2(S) and on HL2(S×Ω), respectively. We say that Aε(ω) strongly two-scale resolvent converge to A and write Aε2A if fε2f,fL2(S×Ω)(Aε(ω)+I)1fε2(A+I)1ffor a.e. ωΩ.

It can be shown that the property (a) is satisfied if we have strong two-scale resolvent convergence (see the proof of [Citation17, Proposition 2.2]). Theorem 4.1 shows that the following implication holds: fε2f,fL2(S×Ω)(Aε(ω)+I)1fε2(A+I)1f. It can be shown that this is equivalent to strong two-scale resolvent convergence (see [Citation17, Proposition 2.8]) and thus the property (a) is satisfied.

In order to prove (b), we start from the eigenvalue problem for the operator Aε(ω) (it has a compact resolvent and its spectrum is discrete), i.e. (40) Aε(ω)uε=sεuε,S(uε)2=1.(40) If we have that sεs and uε2u, then we would also have Au=su. However, the problem would be if u=0, because then sSpA. The next lemma tells us if sSp(Δω) then necessarily the sequence of eigenvalues is compact with respect to the strong two-scale convergence and thus s belongs to the point spectrum of the operator A, since then necessarily u0.

Theorem 6.1

Suppose that (Equation39) holds and that for each ε>0, (sε,uε) satisfy (Equation40). If sεsSp(Δω), then for a.e. ωΩ the sequence (uε) is compact in the sense of strong two-scale convergence.

Proof.

uεW01,2(S) satisfies S1εA1uεv+ε2S0εuεv=sεSuεvvW01,2(S). We use Assumption 3.1 and for each ϵ extend uε|S0ε, denoting the extensions by u~ε. Notice that there exists C>0 such that (41) u~εW1,2(S)C.(41) The difference zε:=uεu~ε satisfies: (42) zεW01,2(S0ε),ε2S0εzεvsεS0εzεv=sεS0εu~εvvW01,2(S0ε).(42) From the estimate (Equation41) we see that (u~ε) is weakly compact in W01,2(S) and thus there exists u~W01,2(S) such that u~εu~, which immediately implies sεχ0εu~ε2su~χO(ω). Furthermore, as a consequence of (Equation27), (Equation29) and (Equation39), the following estimate holds for some C>0: εzεL2(S0ε)+zεL2(S0ε)C. Therefore, from Proposition 4.1 and Remark 4.4 we conclude that zε2zL2(S,W01,2(O)), where the limit z satisfies (43) Oωz(x,)ωvsOz(x,)v=sOu~(x)vvW01,2(O).(43) We also consider the problem (44) mεW01,2(S0ε),ε2S0εmεvsεS0εmεv=sεS0εzεvvW01,2(S0ε).(44) In the same way as before we conclude that for some C>0: εmεL2(S0ε)+mεL2(S0ε)C. Analogously, we conclude that mε2mL2(S,W01,2(O)) which satisfies (45) Oωm(x,)ωvsOm(x,)v=sOz(x,)vvW01,2(O).(45) By testing (Equation42) with mε and (Equation44) with zε we conclude limε0S0ε(zε)2=limε0S0εu~εmε=S×Ωu~m. Finally, by testing (Equation43) with m(x,) and (Equation45) with z(x,) and integrating over S we conclude S×Ωu~m=S×Ωz2, which completes the proof.

7. Spectrum of the limit operator: examples

This section is devoted to the description of the spectrum of the limit operator. Since it crucially depends on the intrinsic properties of the microscopic part of the operator and the properties of the probability space, it does not seem feasible (at least at the current stage of research in this area) to provide a characterisation of the spectrum in a general setting. We shall consider several interesting, from the point of view of applications, examples of probability spaces and configurations of soft inclusions. The general example of a finite number of shapes of randomly varying size is described in Section 7.1. Then we consider the case of a single shape of fixed size in Section 7.2, and the case of a single shape of randomly varying size in Section 7.3, for which we provide the full description of the spectrum of the limit operator with the proofs. The characterisation of the spectrum in the general case of Section 7.1 is analogous to the case of a single shape considered in Section 7.3.

7.1. The setting of finite number of shapes of varying size

Let (ω~j)jZn be a sequence of independent and identically distributed random vectors taking values in N0l×[r1,r2], where 0<r1r21 and (Ω~,F~,P~) is an appropriate probability space. We also assume that we have a finite number of shapes YkY:=[0,1)n, kN0l, that represent the inclusions, where the first and the second components of ω~j=(kj,rj) model the shape and the size, respectively. We also set Y0=. On Ω~ there is a natural shift T~z(ω~j)=(ω~jz), which is ergodic. We next state the discrete analogue of Lemma 2.1.

Lemma 7.1

Assume that Ω~0Ω~ is a set of full measure. Then there exists a subset Ω~1Ω~0 of full measure such that for each ω~Ω~1,zZn we have T~zω~Ω~0.

We treat Y as a probability space with Lebesgue measure dy and the standard algebra L of Lebesgue measurable sets, and define Ω=Ω~×Y,F=F~×L,P=P~×dy. On Ω we define a dynamical system Tx(ω~,y)=(T~[x+y]ω~,x+y[x+y]). We consider the set O={(ω~,y):ω~0N0l×[r1,r2],yr0Yk0}Ω. It is easily seen that O is measurable. For a fixed ω=(ω~,y), the realisation Oω consists of the inclusions rjYkj+jy,jZn. Next, we describe the generators Dj, j=1,2,,n, in the present example. Taking fW1,2(Ω) and using the above lemma, note that there exists a subset of full measure Ω~1Ω~ such that for all ω~Ω~1 and zZn we have f(T~zω~,)W1,2(Y). It is clear that for xY+zy one has f(x,ω):=f(Txω)=f(T~zω~,x(zy)). Using this fact and the statement following (Equation5), we infer that W1,2(Ω)={fL2(Ω~×Y):for a.e. ω~Ω~, f(ω~,)W1,2(Y),f(T~z+ekω~,)|{yk=0}=f(T~zω~,)|{yk=1} zZn,k{1,,n}}. and (46) (Djf)(ω~,y)=yjf(ω~,y),j=1,2,,n.(46)

7.2. Simple example

In this section we set l=0,r1=r2=1, so that N0l×[r1,r2]={0,1}, and, by a standard procedure, see e.g. [Citation18], identify the elements of the probability space Ω~ with sequences ω~=(ω~z)zZn whose components ω~z take values in the two-element set {0,1}. Let Y1 be an open subset of Y whose closure is contained in Y (‘soft inclusion’). The value 0 or 1 of ω~z,zZn, corresponds to the absence or the presence of the inclusion in the ‘shifted cell’ Y+z, respectively. We also set O={ω=(ω~,y):ω~0=1,yY1}Ω. Then, for a given ω=(ω~,y)Ω, the realisation Oω={x:Tx(ω~,y)O} is the union of the sets (‘inclusions’) Y1+zy over all zZn such that ω~z=1. For this example the space W01,2(O) consists of all functions of the form (47) v(ω)=v(ω~,y)=vω~(y),(ω~,y)O, where vω~W01,2(Y1),0otherwise.(47) It is also important to understand how one applies the stochastic gradient. For a function v(ω)W01,2(O)L2(Ω) we have (see (Equation46)) (48) ωv=vω~(y),(ω~,y)O, where vω~W01,2(Y1),0otherwise.(48) Consider formally the spectral problem for the limit operator: (49) SA1homu0ϕ0=λS(u0+u1)ϕ0ϕ0W01,2(S),(49) (50) Oωu1(x,)ωϕ1=λO(u0(x)+u1(x,))ϕ1ϕ1W01,2(O).(50) We write the solution to the ‘microscopic’ Equation (Equation50) in the form u1(x,ω)=λu0(x)v(ω), where (recall Remark 4.2) (51) vW01,2(O),Δωv=λv+1.(51) In other words, v is given by (Equation47) with vω~(y) satisfying (52) Δvω~(y)=λvω~(y)+1,yY1,(52) whenever ω~ such that ω~0=1 and vω~=0 otherwise.

We label the eigenvalues of the operator in (Equation52) in the increasing order, where we repeat multiple eigenvalues, so that νj,jN, and νj,jN, are, respectively, the eigenvalues whose eigenfunctions ϕj have non-zero integral over Y1 and the eigenvalues whose eigenfunctions ϕj have zero integral over Y1. Following [Citation1], we write the solution to (Equation52) via the spectral decomposition (53) vω~=j=1(νjλ)1Y1ϕjϕj,(53) and thereby (54) vΩ=P({ω~:ω~0=1})Y1vω~dy=P({ω~:ω~0=1})j=1(νjλ)1Y1ϕj2.(54) Substituting the obtained representation for u1 into the ‘macroscopic’ equation (Equation49) yields (55) divA1homu0=β(λ)u0,u0W01,2(S),(55) where (56) β(λ):=λ(1+λvΩ)=λ+λ2P({ω~:ω~0=1})j=1(νjλ)1Y1ϕj2(56) is a stochastic version of the ‘Zhikov function’ β in [Citation1]. Assume for the moment that S=Rn. Then the intervals where β(λ)0 are the ‘spectral bands’ of A, and additionally a Bloch-type spectrum is given by {νj:jN}. The set {λ:β(λ)<0}{νj:jN} corresponds to the gaps in the spectrum of A.

In the setting of this paper, namely, for a bounded domain SRn, instead of each spectral band β(λ)0 lying to the left of νj we have a ‘band’ of discrete spectrum: a countable set of eigenvalues (57) {λj,k:νj1<λj,k<νj,β(λj,k)=μk},(57) with the accumulation point at the right end νj of each band, where μk are the eigenvalues of the operator divA1hom defined by the form (58) SA1homuv,u,vW01,2(S).(58) The Bloch-type spectrum of A consists of eigenvalues νj of infinite multiplicity with eigenfunctions of the form f(x)vj(ω) with fL2(S) and (59) vj(ω)=vj(ω~,y)=ϕj(y),(ω~,y)O,0otherwise.(59) Summarising, the spectrum of A is given by (60) σ(A)=j{νj,νj}{λj,k:j,kN}.(60)

7.3. More advanced example

Here we allow the inclusions to randomly change size, so that l=0, 0<r1<r2<1. By analogy with the previous section, we assume that Ω~ consists of sequences ω~=(ω~z)zZn such that ω~z{0}[r1,r2], zZn. We also assume that the restriction to to [r1,r2] of the probability measure on Ω~ is absolutely continuous with respect to Lebesgue measure. As before, consider Y1Y, and denote by Y1,r:=r(Y1yc)+yc, where yc is the centre of Y, the ‘scaled inclusion’, requiring that Y1,r2¯Y, in order for the extension property in Assumption 3.1 to hold. The values 0 or r[r1,r2] of ω~z correspond to the absence of an inclusion or the presence of the inclusion Y1,r in the cell Y+z, respectively. Furthermore, define O:={ω=(ω~,y):yY1,ω~0}Ω. Then a realisation Oω={x:Tx(ω~,y)O} is the union of the inclusions Y1,ω~z+zy for all zZn, where in the case ω~z=0 we set Y1,ω~z=. The space W01,2(O) consists of functions of the form (61) v(ω)=v(ω~,y)=vω~(y),(ω~,y)O, where vω~W01,2(Y1,ω~0),0,otherwise.(61) Consider the spectral problem for u1, namely Oωu1(x,)ωϕ=λO(u0(x)+u1(x,))ϕϕW01,2(Ω), and separate the variables, as in Section 7.2: u1(x,ω)=λu0(x)v(ω), where the function v satisfies (62) Oωvωϕ=O(1+λv)ϕϕW01,2(Ω).(62) The stochastic gradient is given by ωv=vω~(y),(ω~,y)O, where vω~W01,2(Y1,ω~0),0,otherwise, and therefore the problem (Equation62) is equivalent to (63) {ω~0[r1,r2]}Y1,ω~0yvyϕdydP(ω~)={ω~0[r1,r2]}Y1,ω~0(1+λv)ϕdydP(ω~).(63) For each r[r1,r2], the eigenvalues νj,r,νj,r and (orthonormal) eigenfunctions ϕj,r,ϕj,r of the operator Δy acting in W01,2(Y1,r) are obtained by scaling the eigenvalues and eigenfunctions of Δy acting in W01,2(Y1), in particular, νj,r=r2νj, νj,r=r2νj. Therefore, the formula (Equation53) with νj, ϕj replaced by νj,r, ϕj,r gives the solution to (64) Δyvr=1+λvr,vrW01,2(Y1,r).(64) If 0<r1r2 and the set {νj,r:jN,r[r1,r2]} has gaps, then for λR{νj,r:jN,r[r1,r2]} the solution to (Equation63) is given by (Equation61), where the functions vω~(y) solve (Equation64) with r=ω~0. Substituting it into the spectral problem for (Equation49) yields the problem (Equation55) with the Zhikov-type function β given by (cf. (Equation56)) (65) β(λ):=λ(1+λvΩ)=λ+λ2{ω~0[r1,r2]}j=1(νj,ω~0λ)1Y1,ω~0ϕj,ω~02dP(ω~).(65) The integral in (Equation65) is well defined for λR{νj,r:jN,r[r1,r2]}, and the description of the spectrum on the intervals where β(λ)>0 follows Section 7.2.

Theorem 7.1

Under the assumptions of the current subsection, the spectrum of A is given by σ(A)=jN,r[r1,r2]{νj,r,νj,r}{λj,k:j,kN} where for each k, the values λj,k are solutions to β(λj,k)=μk, see (Equation57). The point spectrum of the operator A is given by {λj,k:j,kN}.

It is clear that if the set jN,r[r1,r2]{νj,r,νj,r} has gaps, then σ(A) also has gaps. We are going to prove the theorem in several steps formulated in the following lemmas. We begin by studying the spectrum of the ‘microscopic’ part of the limit operator.

Lemma 7.2

The spectrum of the operator Δω (see (Equation30)) is given by σ(Δω)=jN,r[r1,r2]{νj,r,νj,r} and does not contain eigenvalues of Δω.

Proof.

Let λ=νj0,r0 for some j0N,r0[r1,r2], and assume that vW01,2(O) is an eigenfunction corresponding to λ, i.e. Δωv=λv. (For λ=νj0,r0 argument is similar.) Then v is of the form (Equation61), where Δvω~=λvω~ in Y1,ω~0, whenever ω~0[r1,r2]. But λ is only an eigenvalue of the operator Δ acting in W01,2(Y1,ω~0) if ω~0=r0, hence (66) v(ω)=v(ω~,y)=ϕj0,r0(y),ω~0=r0,yY1,r0,0otherwise.(66) It remains to observe that {ω~0=r0,yY1,r0} is a set of measure zero in Ω and hence vL2(Ø)=0. The second claim of the lemma follows.

Now we show that λσ(Δω) by constructing a Weyl sequence. Without loss of generality we can assume that r0(r1,r2). For small enough δ>0 we choose an L2-function wδ=wδ(r) such that suppwδ(r0δ,r0+δ) and wδL2(r0δ,r0+δ)=1, e.g. we can choose wδ to be equal to a constant proportional to δ1/2 on (r0δ,r0+δ). Consider the sequence vδ(ω)=vδ(ω~,y)=wδ(ω~0)ϕj0,ω~0(y),(ω~,y)O,0otherwise. We have vδW01,2(O),vδL2(Ω)=1 and Δωvδ(ω,y)=wδ(ω~0)Δϕj0,ω~0(y)=νj0,ω~0wδ(ω~0)ϕj0,ω~0(y), hence ΔωvδλvδL2(O)2={ω~0[r1,r2]}Y1,ω~0((νj0,ω~0νj0,r0)wδ(ω~0)ϕj0,ω~0(y))2dydP~(ω~)0,δ0. It follows that vδ is a Weyl sequence for λ=νj0,r0.

It remains to prove that λ is in the resolvent set whenever λjN,r[r1,r2]{νj,r,νj,r}. Assume the contrary and let fL2(O), then the resolvent equation Δωvλv=f has a unique solution given by (Equation61) with vω~ solving Δvω~(y)λvω~(y)=f(ω~,y),  yY1,ω~0. Moreover, since d:=distλ, jN,r[r1,r2]{νj,r,νj,r}>0, we have vω~L2(Y1,ω~)d1f(ω~,)L2(Y1,ω~0), and it follows immediately that vL2(O)d1fL2(O), which concludes the proof.

Next, we focus on the spectrum of A.

Lemma 7.3

The inclusion σ(Δω)σ(A) holds.

Proof.

The proof of the inclusion {νj,r:jN,r[r1,r2]}σ(A) repeats the related part of the proof of the Lemma 7.2. Namely, for λ=νj0,r0, j0N, r0(r1,r2), we define a Weyl sequence uδ:=u0δ+u1δV, where u0δ0 and u1δ is given by u1δ(x,ω):=f(x)vδ(ω), with an arbitrary fixed fL2(S) and vδ defined as in Lemma 7.2. In order to show that {νj,r:jN,r[r1,r2]}σ(A), suppose that λ=νj0,r0 for some j0N and r0(r1,r2). Assume, to the contrary, that there exists a bounded resolvent (Aλ)1, i.e. the system  (Equation18)–(Equation19) has a unique solution for all fL2(S×Ω). For f=f(x)L2(S) the second equation reads {ω~0[r1,r2]}Y1,ω~0(Δyu1λu1)ϕ1dydP(ω~)=(f+λu0){ω~0[r1,r2]}Y1,ω~0ϕ1dydP~(ω~). Then u1 must necessary be of the form u1=(f+λu0)v, where v is of the form (Equation61) and Δyvω~=λvω~+1 in Y1,ω~0, i.e. (67) vω~=j=1(νj,ω~0λ)1Y1,ω~0ϕj,ω~0ϕj,ω~0.(67) which clearly blows up as ω~0r0. We show that the corresponding v is not an element of L2(O), leading to a contradiction. Indeed, using the identity νj0,ω~0νj0,r0=ω~02νj0r02νj0=r02ω~02(r0ω~0)(r0+ω~0)νj0, one has {ω~0[r1,r2]}Y1,ω~0|v|2dydP(ω~)={ω~0[r1,r2]}j=1(νj,ω~0νj0,r0)2Y1,ω~0ϕj,ω~02dP(ω~){ω~0[r1,r2]}(νj,ω~0νj0,r0)2Y1,ω~0ϕj,ω~02dP(ω~)C{ω~0[r1,r2]}|ω~0r0|2dP(ω~), where the last integral diverges.

Lemma 7.4

If β(λ)=μk for some kN then λ is an eigenvalue of A.

Proof.

Let β(λ)=μk, and denote by ψkW01,2(S) be the corresponding eigenfunction of divA1hom. Since λ{νj,r:j,rN}, the problem vW01,2(O),Δωv=λv+1, has a solution given by (Equation61), (Equation67). Therefore u=ψk+λψkv is the eigenfunction of A corresponding to λ.

Lemma 7.5

A point λ belongs to the resolvent set of A if λσ(Δω) and β(λ)σ(divA1hom), i.e. β(λ)<0 or β(λ)0 and β(λ)μk,kN.

Proof.

We claim that the problem (Equation18) and (Equation19) has bounded resolvent. Indeed, suppose that fL2(S×Ω) and write (Equation19) in the form Δωu1λu1=λu0+f. Since λ is not in the spectrum of Δω, the latter has a bounded resolvent at λ and u1=λu0v+g, where v=(Δωλ)11 is as in (Equation61) and g=g(x,ω)=(Δωλ)1f(x,ω), xS. In particular, vL2(O)dist(λ,σ(Δω))11L2(O),g(x,)L2(O)dist(λ,σ(Δω))1f(x,)L2(Ω). Substituting the expression for u1 in (Equation18) we obtain divA1homu0β(λ)u0=f+λg. For β(λ)σ(divA1hom) the operator divA1homβ(λ) is invertible and u0L2(S)dist(β(λ),σ(divA1hom))1f+λgL2(S)dist(β(λ),σ(divA1hom))1(fL2(S×Ω)+|λ|gL2(S×O)), from which the claim follows.

Proposition 7.1

The set σ(Δω){λj,k:j,kN} does not contain eigenvalues of the operator A.

Proof.

Assume that λ=νj0,r0, for some j0N and r0[r1,r2], is an eigenvalue of A, i.e. there exists u=u0+u1V such that (68) divA1homu0=λ(u0+u1O),Δωu1(x,)=λ(u0(x)+u1(x,)).(68) Suppose that u0(x)0 for some xS, then u1(x,)=λu0(x)v(), where v solves (69) Δωv=λv+1.(69) Arguing as for the second inclusion of Lemma 7.3, we see that (Equation69) has no L2-solution for the given λ. It follows that u0=0 and therefore u1(x,) is an eigenfunction of Δω, which cannot be true by Lemma 7.2.

Now we assume that λ=νj0,r0. Arguing as above, for u0(x)0 we have u1(x,)=λu0(x)v(), where v solves (Equation69). The solution exists and is given by (Equation61), (Equation67). Substituting u1 into (Equation68) we see that u0 must satisfy divA1homu0=β(λ)u0, which cannot be true since β(λ)σ(divA1hom).

Finally, if u0=0, then we argue as above for the case λ=νj0,r0, again arriving at a contradiction.

This completes the proof of Theorem 7.1.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

KC is grateful for the support of the Engineering and Physical Sciences Research Council [grant number EP/L018802/2] ‘Mathematical foundations of metamaterials: homogenisation, dissipation and operator theory’. IV has been supported by the Croatian Science Foundation – Hrvatska Zaklada za Znanost [grant number 9477 (MAMPITCoStruFl)].

Notes

1. The function ϕk,m(,ω) is an ‘approximate eigenfunction’ for ΔUω, see (Equation37).

References

  • Zhikov VV. On an extension of the method of two-scale convergence and its applications. (Russian) Mat Sb. 2000;191(7):31–72. Translation in Sb Math. 2000;191(7–8):973–1014.
  • Bellieud M. Homogenization of stratified elastic composites with high contrast. SIAM J Math Anal. 2017;49(4):2615–2665. doi: 10.1137/15M1012050
  • Bouchitté G, Bourel C, Manca L. Resonant effects in random dielectric structures. ESAIM Contr Optim Calculus Variations. 2015;21:217–246. doi: 10.1051/cocv/2014026
  • Bourgeat A, Mikelić A, Piatnitski A. On the double porosity model of a single phase flow in random media. Asymptotic Anal. 2003;34:311–332.
  • Cherednichenko K, Cooper S. Resolvent estimates for high-contrast homogenisation problems. Arch Ration Mech Anal. 2016;219(3):1061–1086. doi: 10.1007/s00205-015-0916-4
  • Cherednichenko K, Cherdantsev M, Cooper S. Extreme localisation of eigenfunctions to one-dimensional high-contrast periodic problems with a defect. 2017. arXiv:1702.03538.
  • Cornfeld IP, Fomin SV, Sinai YG. Ergodic Theory. New York: Springer-Verlag; 1982. (Grundlehren der Mathematischen Wissenschaften; 245). Translated from the Russian by A. B. Sossinskii.
  • Zhikov VV, Kozlov SM, Oleĭnik OA. Homogenization of differential operators and integral functionals. Berlin: Springer-Verlag; 1994.
  • Duerinckx M, Gloria A. Stochastic homogenization of nonconvex unbounded integral functionals with convex growth. Arch Ration Mech Anal. 2016;221(3):1511–1584. doi: 10.1007/s00205-016-0992-0
  • Sango M, Woukeng J-L. Stochastic two-scale convergence of an integral functional. Asymptot Anal. 2011;73(1–2):97–123.
  • Ackoglu MA, Krengel U. Ergodic theorems for superadditive processes. J Reine Angew Math. 1981;323:53–67.
  • Zhikov VV, Pyatnitskiĭ AL. Homogenization of random singular structures and random measures. Izv Ross Akad Nauk Ser Mat. 2006;70(1):23–74. doi: 10.4213/im554
  • Bourgeat A, Mikelić A, Wright S. Stochastic two-scale convergence in the mean and applications. J Reine Angew Math. 1994;456:19–51.
  • Dal Maso G, Modica L. Nonlinear stochastic homogenization. Ann Mat Pura Appl. 1986;144(4):347–389. doi: 10.1007/BF01760826
  • Dal Maso G, Modica L. Nonlinear stochastic homogenization and ergodic theory. J Reine Angew Math. 1986;368:28–42.
  • Allaire G. Homogenization and two-scale convergence. SIAM J Math Anal. 1992;23(6):1482–1518. doi: 10.1137/0523084
  • Zhikov VV. Gaps in the spectrum of some elliptic operators in divergent form with periodic coefficients. (Russian) Algebra i Analiz. 2004;16(5):34–58. Translation in St Petersburg Math J. 2005;16(5):773–790.
  • Shiryaev AN. Probability. New York: Springer-Verlag; 1996.