Homogenization for locally periodic elliptic operators

# Homogenization for non-self-adjoint locally periodic elliptic operators

Nikita N. Senik Saint Petersburg State University, Universitetskaya nab. 7/9, Saint Petersburg 199034, Russia
###### Abstract.

We study the homogenization problem for matrix strongly elliptic operators on of the form . The function  is Lipschitz in the first variable and periodic in the second. We do not require that , so need not be self-adjoint. In this paper, we provide, for small , two terms in the uniform approximation for and a first term in the uniform approximation for . Primary attention is paid to proving sharp-order bounds on the errors of the approximations.

###### Key words and phrases:
homogenization, operator error estimates, locally periodic operators, effective operator, corrector
###### 2010 Mathematics Subject Classification:
Primary 35B27; Secondary 35J15, 35J47
The author was partially funded by Young Russian Mathematics award, Rokhlin grant and RFBR grant 16-01-00087.

## 1. Introduction

Homogenization dates back to the late 1960s, and for more than fifty years it has become a well-established theory. In the simplest case, homogenization deals with asymptotic properties of solutions to differential equations with oscillating coefficients. Given a periodic (with period  in each variable) uniformly bounded and uniformly positive definite function , consider the differential equation

 (1.1) −divA(ε−1x)∇uε−μuε=f,

where , and . The coefficients of the equation are -periodic and hence rapidly oscillate if is small. In homogenization theory one is interested in studying the asymptotic behavior of as  becomes smaller. It is a basic fact that, after passing to a subsequence if necessary, converges to the solution  of the differential equation

 (1.2) −divA0∇u0−μu0=f

with constant . Since, in applications, the elliptic operator on the left side of (1.1) usually describes a physical process in a highly heterogeneous medium, this means that, in certain aspects, the process evolves very similar to that in a homogeneous medium.

It is a basic fact about homogenization theory that converges to in ; we refer the reader to [BLP78], [BP84] or [ZhKO93] for the details. Stated differently, the resolvent of converges in the strong operator topology to the resolvent of . In [BSu01] (see also [BSu03]), Birman and Suslina proved that, in fact, the resolvent converges in norm. Moreover, they found a sharp-order bound on the rate of convergence. Since that time there have been a number of interesting further results in this direction – see [Gri04], [Gri06], [Zh05], [ZhP05], [B08], [KLS12], [Su131], [Su132], [ChC16] and [ZhP16], to name a few.

Here we focus on a more general problem than the periodic one in (1.1). Let with being uniformly bounded functions that are Lipschitz in the first variable and periodic in the second (see Section 3 for a precise definition). Consider the operator  on the complex space  given by

The coefficients now depend not only on the “fast” variable, , but also on the “slow” one, . Assume that, for all in some neighborhood of , the operator  is coercive and furthermore the constants in the coercivity bound are independent of . Then is strongly elliptic for such and there is a sector containing the spectrum of . In this paper, we will obtain approximations for and  (with  outside the sector) in the operator norm and prove that

 (1.3) ∥(Aε−μ)−1−(A0−μ)−1∥L2→L2≤Cε, (1.4) ∥(Aε−μ)−1−(A0−μ)−1−εCεμ∥L2→L2≤Cε2

and

 (1.5) ∥∇(Aε−μ)−1−∇(A0−μ)−1−ε∇Kεμ∥L2→L2≤Cε,

the estimates being sharp with respect to the order (see Theorems 6.1 and 6.2). The effective operator  is of the same form as , but its coefficients depend only on the slow variable. In contrast, the correctors  and  involve rapidly oscillating functions as well. The first of these plays the role of the traditional corrector and differs from the latter in that it involves a smoothing operator. The idea of using a smoothing to regularize the traditional corrector is due to Griso, see [Gri02]. The other corrector has no analogue in classical theory and was first presented in [BSu05] for purely periodic operators. Assume for simplicity that . Then has the form

 Cεμ=(Kεμ−Lμ)−Mεμ+(Kεμ−Lμ)∗

(see Section 5). What is interesting here is that an analog of for periodic operators, while looking similar to this one, does not include the term , see [Se171]. In fact, one cannot remove from if the estimate (1.4) is to remain true, see Remark 5.9 for examples. So this term is a special feature of non-periodic problems.

The results of the present paper extend the author’s work [Se171] on periodic elliptic problems, where we studied non-self-adjoint scalar operators whose coefficients were periodic in some variables and Lipschitz in the others. Put differently, the fast and slow variables were separated in the sense that , where . We proved analogs of the estimates (1.3)–(1.5), yet the correctors were slightly different, see Remark 6.3 below. It should be pointed out that the operators in [Se171] were allowed to involve lower-order terms with quite general coefficients.

Previous results on uniform approximations for locally periodic elliptic operators are due to, on the one hand, Borisov and, on the other hand, Pastukhova and Tikhomirov. In [B08] Borisov established the estimates (1.3) and (1.5) for certain matrix self-adjoint operators with smooth coefficients. In the paper [PT07] of Pastukhova and Tikhomirov, similar results were proved for scalar self-adjoint operators with rough coefficients (although their techniques also apply to non-self-adjoint problems). As far as I know, the estimate (1.4) in the locally periodic settings was not obtained even for the simplest cases.

To prove the estimates, we develop the ideas of [Se171]. In the first step we establish a variant of the resolvent identity that involves the resolvents of the original and the effective operators and a corrector (see Section 7). This combination comes as no surprise, for it is well known that the effective operator and a corrector form a first approximation to the original operator (see, e.g., [BLP78] or [ZhKO93]). When this is done, all the desired estimates will follow at once. However, we cannot use the same technique as in [Se171], so the identity is proved by different means. The point is that the technique depends heavily on the smoothing operator that has been chosen. In the case of periodic operators, the smoothing was based on the Gelfand transform; but it is not as convenient now. To my knowledge, no natural smoothing for operators with locally periodic coefficients is known, so we choose the Steklov smoothing operator, which is the most simple and has proved to be quite useful; see [Zh05] and [ZhP05], where that smoothing first appeared in the context of homogenization, as well as [PT07], [Su131] and [Su132]. We remark that a very similar smoothing had been used earlier in [Gri02] and [Gri04] (see also [Gri06]). Our technique is strongly influenced by all these works.

I believe that the same method can be of use for locally periodic problems on domains with Dirichlet or Neumann boundary conditions as well.

It is also worth noting that, once the estimates (1.3)–(1.5) are verified, a limiting argument will give similar results for operators whose coefficients are Hölder continuous in the first variable, see Remark 6.6. These results, together with the results stated here, have been announced in [Se172].

The plan of the paper is as follows. Section 2 contains basic definitions and notation. In Section 3 we introduce the original operator. We study the effective operator in Section 4 and correctors in Section 5. Section 6 states the main results. Section 7 is the core of the paper, where we first prove the identity and then complete the proofs.

## 2. Notation

The symbol  will stand for the norm on a normed space . If and  are Banach spaces, then is the Banach space of bounded linear operators from to . When , the space  becomes a Banach algebra with the identity map . The norm and the inner product on are denoted by and , respectively. We shall often identify and .

Let be a domain in and a Banach space. The space  consists of those uniformly continuous functions  for which

 ∥u∥C0,1(¯Σ;U)=∥u∥C(¯Σ;U)+[u]C0,1(¯Σ;U)<∞,

where and

 [u]C0,1(¯Σ;U)=supx1,x2∈Σ,x1≠x2∥u(x2)−u(x1)∥U|x2−x1|.

We will use the notation , and  as shorthand for , and  when the context makes clear which and  are meant.

The symbol  stands for the -space of strongly measurable functions on with values in . In case , we write for the norm on and  for the inner product on . We let denote the usual Sobolev space of -valued functions on and , its dual space under the pairing . If  is dense in , then , where is the exponent conjugate to .

Let be the closed cube in with center  and side length , sides being parallel to the axes. Then denotes the completion of in the -norm. Here is the class of -times continuously differentiable functions on whose periodic extension to enjoys the same smoothness. Notice that coincides with the space of all periodic functions in . The spaces  and  are defined in a similar fashion. If , we write for , for , etc. The symbol  will stand for the subspace of functions in with mean value zero. Any  satisfies the Poincaré inequality

 (2.1) ∥u∥2,Q≤(2π)−1∥Du∥2,Q,

as can be seen by using Fourier series. Here and below, .

We will often use the notation  to mean that that there is a constant , depending only on some fixed parameters (these are listed in Theorems 6.1 and 6.2), such that .

## 3. Original operator

Let each  be a function in . Then may be thought of as a bounded mapping  that is Lipschitz in the first variable and periodic in the second. As is well known, for any function  satisfying the Carathéodory condition (i.e., the requirement of continuity with respect to the first variable and measurability with respect to the second) the map defined for and  by

 (3.1) τεu(x,z)=u(x,ε−1x,z),

is measurable (here ). Notice that, if is another function from to , then . We adopt the notation .

Consider the matrix operator  given by

 (3.2) Aε=D∗AεD.

It is easy to see that is bounded, with bound :

 (3.3) ∥Aεu∥−1,2,R≤C♭∥Du∥2,R

for all . Now we impose a condition that will render elliptic. Namely, we assume that is coercive uniformly in , where with , that is, there are and  such that

 (3.4) Re(AεDu,Du)R+CA∥u∥22,R≥cA∥Du∥22,R

for every . It follows that is -sectorial with sector

 S={z∈C:|Imz|≤−1AC♭(Rez+CA)}

independent of . Whenever , the operator  is an isomorphism and hence is invertible; moreover, for any  we have

 (3.5) ∥(Aεμ)−1f∥1,2,R≲∥f∥−1,2,R.

Before proceeding, we make a few remarks about the coercivity condition. It follows from (3.4) (via Lemma 4.1) that satisfies the Legendre–Hadamard condition

 (3.6) Re⟨A(⋅)ξ⊗η,ξ⊗η⟩≥cA|ξ|2|η|2,ξ∈R,η∈Cn,

so is strongly elliptic for all . The Legendre–Hadamard condition does not generally imply (3.4). If we restrict our attention to the real-valued case, then for scalar operators the two statements are equivalent. But this is no longer true for matrix operators, let alone the complex-valued case. A necessary and sufficient algebraic condition on that would guarantee (3.4) is not known.

It is worthwhile to point out that we have to be able to verify the coercivity bound for all in some interval , which may be rather difficult. A sufficient condition not involving  is that the operator  is strongly coercive on and furthermore there is so that for any  and

 (3.7) Re(A(x,⋅)Du,Du)R≥c∥Du∥22,R.

This can be seen by noticing that, by change of variable, the above inequality remains true with in place of . Then a partition of unity argument will do the job, since is uniformly continuous in the first variable.

As an example of satisfying (3.7), let be a matrix first-order differential operator with symbol

 ξ↦b(ξ)=d∑k=1bkξk,

where . Suppose that the symbol has the property that, for some ,

 b(ξ)∗b(ξ)≥α|ξ|2,ξ∈R.

Let be a function in with uniformly positive definite. Now if we take , then application of the Fourier transform will yield

 ≥α∥(Reg)−1/2∥−2C∥Du∥22,R.

Homogenization for self-adjoint operators of this type was studied by Birman and Suslina in the purely periodic setting (see, e.g., [BSu01], [BSu03], [BSu05], [BSu06], [Su131] and [Su132]) and by Borisov in the locally periodic setting (see [B08]).

Observe that the more restrictive Legendre condition, which amounts to the uniform positive definiteness of , does ensure coercivity, but excludes some strongly elliptic operators with important applications – such as certain elasticity operators.

## 4. Effective operator

Given and , we let be the weak solution of

 (4.1)

in . The function  is well defined, since is a continuous linear functional on and the operator  is strongly coercive on , as we shall now see.

###### Lemma 4.1.

For any  and all , we have

 (4.2) Re(A(x,⋅)Du,Du)Q≥cA∥Du∥22,Q.
###### Proof.

Fix with and . We substitute into (3.4) and let tend to 0. Then, because and  converge in to ,

 limε→0Re∫R⟨Aε(x)(Du)ε(x),(Du)ε(x)⟩|φ(x)|2dx≥limε→0cA∫R|(Du)ε(x)|2|φ(x)|2dx.

It is well known that if , then

 limε→0∫Rfε(x)dx=∫R∫Qf(x,y)dxdy

(see, for instance, [A92, Lemmas 5.5 and 5.6]). As a result,

 Re∫R∫Q⟨A(x,y)Du(y),Du(y)⟩|φ(x)|2dxdy≥cA∫R∫Q|Du(y)|2|φ(x)|2dxdy.

Since is an arbitrary function in and since is continuous in the first variable, we conclude that, for any ,

 Re∫Q⟨A(x,y)Du(y),Du(y)⟩dy≥cA∫Q|Du(y)|2dy.\qed

It is clear from Lemma 4.1 and Poincaré’s inequality (2.1) that

 Re(A(x,⋅)Du,Du)Q≳∥u∥21,2,Q

for every . Thus, the definition of makes good sense.

Denote by the map sending to . Evidently, depends linearly on , so is simply an operator of multiplication by a function (still denoted by ). The next lemma shows that has the same regularity in the first variable as .

###### Remark 4.2.

In what follows, we denote differentiation in the first variable by and differentiation in the second variable by . When no confusion can arise, we omit the subscript and write , as we did before.

We have .

###### Proof.

The identity (4.1), together with Lemma 4.1, yields

 cA∥DNξ(x,⋅)∥2,Q≤∥A(x,⋅)∥∞,Q|ξ|,

whence

 (4.3) ∥D2N∥L∞(R;L2(Q))≤c−1A∥A∥C.

Next, by (4.1) again, for any  and

 (A(x2,⋅)(DNξ(x2,⋅)−DNξ(x1,⋅)),Dv)Q =−((A(x2,⋅)−A(x1,⋅))(ξ+DNξ(x1,⋅)),Dv)Q.

Taking  and using Lemma 4.1, we obtain

 cA∥DNξ(x2,⋅)−DNξ(x1,⋅)∥2,Q≤∥A(x2,⋅)−A(x1,⋅)∥∞,Q∥ξ+DNξ(x1,⋅)∥2,Q.

It now follows from (4.3) that

 [D2N]C0,1(¯Rd;L2(Q))≤c−1A(1+c−1A∥A∥C)[A]C0,1.

We have proved that . But then Poincaré’s inequality (2.1) implies that as well. ∎

Let  be given by

 (4.4) A0(x)=∫QA(x,y)(I+D2N(x,y))dy.

Since and  are continuous in the first variable, so is . In fact, we have . Indeed, the estimate

 ∥A0∥C(¯Rd)≤∥A∥C∥I+D2N∥C

is immediate from the definition of , and that

 [A0]C0,1(¯Rd)≤∥A∥C[D2N]C0,1+[A]C0,1∥I+D2N∥C

follows by an easy calculation. Hence, is finite.

Now we define the effective operator  by setting

 (4.5) A0=D∗A0D.

Observe that is bounded and coercive (recall Gårding’s inequality) and thus -sectorial. It can be proved that satisfies an estimate similar to (3.4) with exactly the same constants, however the bound on its norm may be different from (3.3). Nevertheless, the sector for remains the same as for . We briefly sketch the argument; see [Se171, Section 2.3] for a related proof. First consider the two-scale effective system as in [A92] and check that the associated form, which is defined on  by

 u⊕U↦(A(D1u+D2U),D1u+D2U)R×Q,

is -sectorial with sector . We only remark that the coercivity is obtained by substituting (with sufficiently smooth and ) into (3.4) for and letting tend to ; cf. the proof of Lemma 4.1. Then notice that

 (A0u,u)R=(A(D1u+D2U),D1u+D2U)R×Q

provided  (which is definitely in ). The claim is proved.

Thus, we see that the operator  is an isomorphism as long as is outside . In addition, standard regularity theory for strongly elliptic systems (see, e.g., [McL00, Theorem 4.16]) implies that the pre-image of under is all of and for any

 (4.6) ∥(A0μ)−1f∥2,2,R≲∥f∥2,R.

Let us return to our discussion of coercivity at the end of the previous section. As we have seen, (4.2) follows from (3.4), which in turn is a consequence of (3.7). On the other hand, (4.2) does not generally imply (3.7), and there are examples (for , of course) where (4.2) holds, but (3.7) is false, see [BF15]. In such cases, a subsequence of may still converge in the weak operator topology to , but will fail to be strongly elliptic, i.e., will not satisfy the Legendre–Hadamard condition.

## 5. Correctors

Let the operator  be given by

 (5.1) Kμ=ND1(A0μ)−1.

Lemma 4.3, combined with the estimate (4.6), readily implies that is continuous:

 (5.2) ∥Kμf∥1,2,R×Q≲∥f∥2,R.

The very same argument shows that is bounded on  as well:

 (5.3) ∥D1D2Kμf∥2,R×Q≲∥f∥2,R.

Since we do not impose any extra assumptions on the coefficients, the traditional corrector  will not even map into itself. So we must first appropriately regularize the traditional corrector, and a smoothing operator is used for exactly this purpose.

### 5.1. Smoothing

Let be the translation operator

 (5.4) Tεu(x,y,z)=u(x+εz,y),

where and . Certainly, for any satisfying we have . Next, the adjoint of is given by

 (Tε)∗u(x,y)=∫Qu(x−εz,y,z)dz.

Note that is defined on and  as well, by way of identifying these spaces with the corresponding subspaces of . We define the Steklov smoothing operator  to be the restriction of to . In other words,

 (5.5) Sεu(x,y)=∫QTεu(x,y,z)dz.

Here we collect some facts about and .

###### Lemma 5.1.

The restriction of to is an isometry.

###### Proof.

By change of variable,

 ∥τεTεu∥22,R×Q=∫R∫Q|u(x,ε−1x−z)|2dxdz.

But since is periodic in the second variable, this equals . ∎

A related result for is the following.

###### Lemma 5.2.

The restriction of to is bounded, with bound at most .

###### Proof.

This is immediate from Cauchy’s inequality and Lemma 5.1. ∎

It is easy to see that both and  converge in the strong operator topology to the identity operator, yet they do not converge in norm. The uniform convergence will, however, take place if we restrict them to certain Sobolev spaces.

###### Lemma 5.3.

For any  we have

 (5.6) ∥(Tε−I)u∥2,R×Q×Q≲ε∥D1u∥2,R×Q.
###### Proof.

Notice that

 u(x+εz,y)−u(x,y)=εi∫10⟨D1u(x+εtz,y),z⟩dt.

Hence,

 ∥(Tε−I)u(⋅,y,z)∥2,R≤εrQ∥D1u(⋅,y)∥2,R,

where . Integrating out the  and  variables then yields (5.6). ∎

###### Lemma 5.4.

For any  we have

 (5.7) ∥(Sε−I)u∥2,R×Q ≲ε∥D1u∥2,R×Q, (5.8) ∥(Sε−I)u∥2,R×Q ≲ε2∥D1D1u∥2,R×Q.
###### Proof.

The inequality (5.7) comes from (5.6). To prove (5.8), notice that

The first term on the right-hand side has mean value zero for a.e.  and  (because is centered at the origin), so

Integrating over completes the proof. ∎

Now we can prove the following result.

###### Lemma 5.5.

For any  we have

 ∥τεTεu−τεSεu∥2,R×Q≲ε∥D1u∥2,R×Q.
###### Proof.

We write

 τεTεu−τεSεu=τεTε(I−Sε)u+τεSε(Tε−I)u

(here is understood to be defined as , that is, we apply to regarding the new variable resulting from the operator  as a parameter). Then, it follows from Lemmas 5.1 and 5.4 that

 ∥τεTε(I−Sε)u∥2,R×Q≲ε∥D1u∥2,R×Q,

while Lemmas 5.2 and 5.3 imply that

 ∥τεSε(Tε−I)u∥2,R×Q≲ε∥D1u∥2,R×Q.

These observations combine to give the desired estimate. ∎

###### Remark 5.6.

We note that the results of Lemmas 5.15.5 persist if we replace the -norms by the -norms with . This will play a role in what follows.

### 5.2. Correctors

We define the first corrector  by

 (5.9) Kεμ=τεSεKμ.

More explicitly,

 Kεμf(x)=∫QN(x+εz,ε−1x)D(A0μ)−1f(x+εz)dz.

Because of the smoothing , this corrector is bounded with

 (5.10) ∥Kεμf∥2,R ≲∥f∥2,R, (5.11) ∥DKεμf∥2,R ≲ε−1∥f∥2,R.

Indeed, using Lemma 5.2, we see that

 ∥Kεμf∥2,R ≤∥Kμf∥2,R×Q, ∥DKεμf∥2,R ≤∥D1Kμf∥2,R×Q+ε−1∥D2Kμf∥2,R×Q.

The estimates (5.10) and (5.11) then follow from (5.2).

While the -norm of is merely uniformly bounded, the -norm of turns out to be of order .

###### Lemma 5.7.

For any  and  we have

###### Proof.

By definition of and ,

 SεKεμf(x)=∫Q∫QTεKμf(x+εw,ε−1x+z,z)dwdz.

Since  is periodic and has mean value zero, we have

 ∫Q∫QKμf(x+εw,ε−1x+z)dwdz=0,

and hence

 SεKεμf(x)=∫Q∫Q(Tε−I)Kμf(x+εw,ε−1x+z,z)dwdz.

Changing variables and keeping in mind that is periodic in the second variable, we find that

The result is therefore immediate from Lemma 5.3 and the estimate (5.2). ∎

To describe the second corrector, we need some additional notation. Let be the adjoint of . Then we construct the effective operator , the corrector  and the other objects (which will be marked with “” as well) for just as we did for . (It may be noted in passing that is the adjoint of .) Of course, all results for will transfer to . We shall not explicitly formulate these results here, but refer to them by the numbers of the corresponding statements for with “” following the reference (for example, Lemma 5.7+ and the estimate (5.10)+).

Define  by

 (5.12) Lμ=(D1K+μ)∗A(D1(A0μ)−1+D2Kμ)

and  by

 (5.13) Mεμ=ε−1(τεTε(D1((A0μ)+)−1+D2K+μ))∗τε[A,Tε](D1(A0μ)−1+D2Kμ).

A more convenient way of dealing with these operators is to look at their forms. If we set , and , , then

 (Lμf,g)R=(A(D1u0+D2U),D1U+)R×Q

and

 (Mεμf,g)R=ε−1(τε[A,Tε](D1u0+D2U),τεTε(D1u+0+D2U+))R×Q.

Both and  are bounded. Indeed,

 |(Lμf,g)R|≤∥A(D1u0+D2U)∥2,R×Q∥D1U+∥2,R×Q,

and so, according to the estimates (4.6), (5.2) and (5.2)+,

 (5.14) ∥Lμf∥2,R≲∥f∥2,R.

Likewise, observing that