AND APPLICATIONS
∗
CARLOS CONCA
†
, RAFAEL ORIVE
‡
, AND MUTHUSAMY VANNINATHAN
§
SIAM J. MATH. ANAL. c 2002 Society for Industrial and Applied Mathematics
Vol. 33, No. 5, pp. 1166–1198
Abstract. The classical problem of homogenization of elliptic operators with periodically os
cillating coeﬃcients is revisited in this paper. As is well known, the homogenization process in a
classical framework is concerned with the study of asymptotic behavior of solutions u
ε
of boundary
value problems associated with such operators when the period ε > 0 of the coeﬃcients is small. In
a previous work by C. Conca and M. Vanninathan [SIAM J. Appl. Math., 57 (1997), pp. 1639–1659],
a new proof of weak convergence as ε → 0 towards the homogenized solution was furnished using
Bloch wave decomposition.
Following the same approach here, we go further and introduce what we call Bloch approximation,
which will provide energy norm approximation for the solution u
ε
. We develop several of its main
features. As a simple application of this new object, we show that it contains both the ﬁrst and
second order correctors. Necessarily, the Bloch approximation will have to capture the oscillations
of the solution in a sharper way. The present approach sheds new light and oﬀers an alternative for
viewing classical results.
Key words. homogenization, Bloch waves, correctors
AMS subject classiﬁcations. 35B27, 35A25, 42C30
PII. S0036141001382200
1. Introduction. In this paper, the classical problem of homogenization of el
liptic operators with periodically oscillating coeﬃcients is revisited. As is well known,
the homogenization process is concerned with the study of the behavior of solutions
u
ε
of boundary value problems associated with such operators when the coeﬃcients
are periodic with small period ε > 0. For an excellent introduction to this subject,
the reader is referred to the book of A. Bensoussan, J.L. Lions, and G. Papanico
laou [5]. In a previous work by C. Conca and M. Vanninathan [11], a new proof
of weak convergence of u
ε
towards the homogenized solution u
∗
was furnished using
Bloch wave decomposition. Following the same approach, we go further and introduce
what we call Bloch approximation of the solution u
ε
. As a simple application of this
new object, we treat the problem of correctors in homogenization. At this point, it
is worthwhile to remark that the homogenized solution u
∗
is merely the weak limit
of solutions u
ε
as ε → 0. The idea behind introducing correctors is to look for terms
(called ﬁrst order correctors) which, when added to the homogenized solution, pro
vide an approximation in the energy norm for all ε suﬃciently small. Second order
correctors yield an error estimate in the energy norm of order O(ε). The main feature
of Bloch approximation is that it contains both the ﬁrst and second order corrector
terms. Another important feature is that it is easily computable in principle.
∗
Received by the editors January 17, 2001; accepted for publication (in revised form) September
18, 2001; published electronically April 8, 2002. This work has been partially supported by FONDAP
through its Programme on MathematicalMechanics.
http://www.siam.org/journals/sima/335/38220.html
†
Departamento de Ingenier´ıa Matem´atica, Facultad de Ciencias F´ısicas y Matem´aticas, and
Centro de Modelamiento Matem´atico, UMR 2071 CNRSUChile, Universidad de Chile, Casilla 170/3
 Correo 3, Santiago, Chile (cconca@dim.uchile.cl).
‡
Departamento de Matem´atica Aplicada, Facultad de Ciencias Matem´aticas, Universidad Com
plutense de Madrid, 28040 Madrid, Espa˜ na (orive@sunma4.mat.ucm.es).
§
IIScTIFR Mathematics Programme, TIFR Center, P.O. Box 1234, Bangalore, 560012, India
(vanni@math.tifrbng.res.in).
1166
BLOCH APPROXIMATION 1167
Historically, a classical way of obtaining such correctors is to work in the physical
space and use multiscale expansion of the solution, which was ﬁrst introduced in the
basic book just cited. As we will see, the method of Bloch waves sheds new light
and oﬀers an alternative for viewing the classical results. This method naturally
leads us to work in the Fourier space, and thus in a framework dual to the one
used in L. Tartar [21], whose method is very general and justiﬁes, in particular, the
ﬁrst term obtained via the multiscale expansion. (There are also other methods
of justiﬁcation based on the analysis in physical space; cf. G. Nguetseng [17] and
G. Allaire [1].) However, it is important to mention that the Bloch wave method
does not presuppose any multiscale structure of the solution; on the contrary, such
a structure of the solution will be a consequence of the present method. Although
correctors are generally not unique, our approach yields a posteriori the same ones as
those obtained in [5].
Bloch waves and their applications are not by any means new. It is a classical tool
which has been in use in solid state physics since the pioneering paper of F. Bloch [6].
However, the basic idea was introduced in the mathematical literature much earlier
by G. Floquet [13]. For later developments, let us cite the works of F. Odeh and
J. Keller [18] and M. Reed and B. Simon [19]. A deep analysis concerning the par
tial regularity of the spectrum of Schr¨ odinger’s equation with periodic potential was
carried out by C. Wilcox [22]. Our point of view regards periodic medium as a per
turbation of homogeneous ones. In this context, the book by T. Kato [15] provides
excellent analysis when the parameter of perturbation is a scalar. We end this rather
incomplete list by citing G. Allaire and C. Conca [2], [3], and P. G´erard et al. [14].
We feel it is also appropriate to cite a recent work by M. Avellaneda, L. Berlyand,
and J.F. Clouet [4], in which the Bloch–Floquet approach is used to provide new ho
mogenization results and handles the boundary layer terms for frequency dependent
problems. To conclude, let us refer the reader to C. Conca [8] for a more complete
general survey on Bloch waves.
Before proceeding further, we mention a word about the notations adopted in
what follows. Unless mentioned explicitly, the usual summation convention with
respect to the repeated indices is understood. The constants appearing in various
estimates independent of ε are generically denoted by c, c
1
, c
2
, etc. Apart from the
usual norms in Sobolev spaces H
1
, H
2
, we will also use the following seminorms:
[v[
H
1
=
_
_
_
N
j=1
_
_
D
j
v
_
_
2
L
2
_
_
_
1
2
, [v[
H
2
=
_
_
_
N
j,k=1
_
_
D
2
j,k
v
_
_
2
L
2
_
_
_
1
2
.
Now let us introduce the problem to be studied in this work. We consider the
operator
A
def
= −
∂
∂y
k
_
a
k
(y)
∂
∂y
_
, y ∈ R
N
, (1.1)
where the coeﬃcients satisfy
_
¸
¸
¸
_
¸
¸
¸
_
a
k
∈ L
∞
#
(Y ), where Y =]0, 2π[
N
, i.e., each a
k
is a
Y periodic bounded measurable function deﬁned on R
N
, and
∃α > 0 such that a
k
(y)η
k
η
≥ α[η[
2
∀η ∈ R
N
, y ∈ Y a.e.,
a
k
= a
k
∀k, = 1, . . . , N.
(1.2)
1168 C. CONCA, R. ORIVE, AND M. VANNINATHAN
For each ε > 0, we consider also the operator A
ε
, where
A
ε
def
= −
∂
∂x
k
_
a
ε
k
(x)
∂
∂x
_
with a
ε
k
(x) = a
k
_
x
ε
_
, x ∈ R
N
. (1.3)
In homogenization theory, it is usual to refer to x and y, the slow and the fast variables,
respectively. They are related by y =
x
ε
. Associated with A
ε
, let us consider the
boundary value problem
A
ε
u
ε
= f in Ω, u
ε
∈ H
1
0
(Ω), (1.4)
which is posed in an arbitrary bounded domain Ω in R
N
and where f is a given
element in L
2
(Ω). It is classical that the above problem admits one and only one
solution.
From the classical work [5], it is known that one can associate to A
ε
a homogenized
operator A
∗
given by
A
∗
def
= −
∂
∂x
k
_
q
k
∂
∂x
_
. (1.5)
The homogenized coeﬃcients q
k
are constants and their deﬁnition is given below.
The solution u
ε
of (1.4) converges weakly in H
1
0
(Ω) to the socalled homogenized
solution u
∗
characterized by
A
∗
u
∗
= f in Ω, u
∗
∈ H
1
0
(Ω). (1.6)
In the present paper, we do not consider the eﬀects of boundaries, postponing
them to a subsequent article [9]. In the case of R
N
, it is natural to replace the
operator A
ε
by (A
ε
+I). In that case, if w
ε
satisﬁes
_
(A
ε
+I)w
ε
= g in R
N
,
w
ε
w
∗
in H
1
(R
N
)weak,
(1.7)
where g is a given function in L
2
(R
N
), then it can be seen that (see Proposition 6.1
below)
w
ε
→ w
∗
in L
2
(R
N
)strong. (1.8)
In view of the above result, there is no concentration of L
2
energy at inﬁnity, and
therefore throughout this paper we will consider a sequence u
ε
and a function f ∈
L
2
(R
N
) satisfying
_
¸
_
¸
_
A
ε
u
ε
= f in R
N
,
u
ε
u
∗
in H
1
(R
N
)weak,
u
ε
→ u
∗
in L
2
(R
N
)strong.
(1.9)
The central issue in the analysis of the ﬁrst order correctors is to obtain functions
u
ε
1
∈ H
1
(R
N
), which can be easily constructed and have the following characteristic
property:
u
ε
−u
∗
−εu
ε
1

H
1
(R
N
)
→ 0 as ε → 0. (1.10)
BLOCH APPROXIMATION 1169
By deﬁnition, second order correctors u
ε
2
∈ H
1
(R
N
) will enjoy the property
u
ε
−u
∗
−εu
ε
1
−ε
2
u
ε
2

H
1
(R
N
)
≤ cε. (1.11)
One of the purposes of this article is to carry out a more general construction than
the classical one for correctors, namely, Bloch approximation θ
ε
, which contains all
the above correctors and justiﬁes the procedure. Apart from this, θ
ε
contains a lot
of information about the periodic medium which will be amply demonstrated in this
paper.
1.1. Survey of the previous results. In the classical book [5] the authors
obtain an asymptotic expansion (with y =
x
ε
) of the form
u
ε
(x) = u
∗
(x) +ε
_
χ
k
(y)
∂u
∗
∂x
k
(x) + ¯ u
1
(x)
_
+ ε
2
_
χ
k
(y)
∂
2
u
∗
∂x
k
x
(x) +χ
(y)
∂¯ u
1
∂x
(x) + ¯ u
2
(x)
_
+ .
(1.12)
Here, χ
k
is the unique solution of the cell problem
_
¸
¸
_
¸
¸
_
Aχ
k
=
∂a
k
∂y
in R
N
,
χ
k
∈ H
1
#
(Y ), M
Y
(χ
k
)
def
=
1
[Y [
_
Y
χ
k
dy = 0.
(1.13)
The function χ
k
is characterized as the unique solution of
_
¸
¸
_
¸
¸
_
Aχ
k
=a
k
+a
km
∂χ
∂y
m
−
∂
∂y
m
(a
mk
χ
)−M
Y
(a
k
)−M
Y
_
a
km
∂χ
∂y
m
_
in R
N
,
χ
k
∈ H
1
#
(Y ), M
Y
(χ
k
) = 0.
(1.14)
Further, ¯ u
1
(x), ¯ u
2
(x), . . . are independent of ε and satisfy equations of the type A
∗
¯ u
j
=
¯ g
j
in R
N
, where, for instance, ¯ g
1
(x) = b
jk
D
3
jk
u
∗
, where b
jk
are constants:
b
jk
= M
Y
_
a
jm
∂χ
k
∂y
m
+a
k
χ
j
_
∀j, k, = 1, . . . , N.
With these notations, the classical formula of the homogenized coeﬃcients is as fol
lows:
q
k
= M
Y
_
a
k
+a
km
∂χ
∂y
m
_
∀k, = 1, . . . , N.
(Another characterization of q
k
is given in Proposition 1.5 below.) Using the above
expansion, the ﬁrst order corrector term is obtained in [5]. More precisely, we have
the following.
Theorem 1.1. We assume that the coeﬃcients a
k
satisfy assumptions (1.2),
f ∈ L
2
(R
N
), and the solution χ
k
∈ W
1,∞
(Y ), k = 1, . . . , N. Then the ﬁrst order
corrector is deﬁned by
u
ε
1
(x) = χ
k
_
x
ε
_
∂u
∗
∂x
k
(x),
1170 C. CONCA, R. ORIVE, AND M. VANNINATHAN
which means that
u
ε
−u
∗
−εu
ε
1

H
1
(R
N
)
→ 0 as ε → 0.
In this paper, we obtain a more general result using a diﬀerent approach intro
duced in [11]. The basic tool of this new approach is Bloch waves ψ associated with
A which we deﬁne now. Let us consider the following spectral problem parameterized
by η ∈ R
N
: ﬁnd λ = λ(η) ∈ R and ψ = ψ(y; η) (not identically zero) such that
_
Aψ(; η) = λ(η)ψ(; η) in R
N
, ψ(; η) is (η; Y )periodic, i.e.,
ψ(y + 2πm; η) = e
2πim·η
ψ(y; η) ∀m ∈ Z
N
, y ∈ R
N
.
(1.15)
Next, we deﬁne φ(y; η) = e
−iy·η
ψ(y; η), and (1.15) can be rewritten in terms of φ as
follows:
A(η)φ = λφ in R
N
, φ is Y periodic. (1.16)
Here, the operator A(η) is deﬁned by
A(η)
def
= −
_
∂
∂y
k
+iη
k
__
a
k
(y)
_
∂
∂y
+iη
__
, (1.17)
which can be rewritten as
A(η) = A+iη
k
C
k
+η
k
η
a
k
(y) (1.18)
with
C
k
φ
def
= −a
kj
(y)
∂φ
∂y
j
−
∂
∂y
j
(a
kj
(y)φ). (1.19)
It is clear from (1.15) that the (η, Y ) periodicity condition is unaltered if we
replace η by (η + q) with q ∈ Z
N
, and η can therefore be conﬁned to the dual
cell η ∈ Y
= [−
1
2
,
1
2
[
N
. It is well known (C. Conca, J. Planchard, and M. Van
ninathan [10]) that for each η ∈ Y
, the above spectral problem admits a discrete
sequence of eigenvalues with the following properties:
_
0 ≤ λ
1
(η) ≤ ≤ λ
m
(η) ≤ → ∞
∀m ≥ 1, λ
m
(η) is a Lipschitz function of η ∈ Y
.
Besides, the corresponding eigenfunctions denoted by ψ
m
(; η) and φ
m
(; η) form or
thonormal bases in the spaces of all L
2
loc
(R
N
)functions which are (η; Y )periodic and
Y periodic, respectively; these spaces are denoted by L
2
#
(η; Y ) and L
2
#
(Y ). It is
worthwhile to remark that these eigenfunctions in fact belong to the spaces H
1
#
(η; Y )
and H
1
#
(Y ), respectively, where
H
1
#
(η; Y ) =
_
ψ ∈ L
2
#
(η; Y )
¸
¸
¸
∂ψ
∂y
k
∈ L
2
#
(η; Y ) ∀k = 1, . . . , N
_
,
H
1
#
(Y ) =
_
φ ∈ L
2
#
(Y )
¸
¸
¸
∂φ
∂y
k
∈ L
2
#
(Y ) ∀k = 1, . . . , N
_
.
BLOCH APPROXIMATION 1171
The functions ψ
m
(; η) and φ
m
(; η) (referred to as Bloch waves) introduced above
enable us to describe the spectral resolution of A (an unbounded selfadjoint operator
in L
2
(R
N
)) in the orthogonal basis ¦e
iy·η
φ
m
(y; η)[m ≥ 1, η ∈ Y
¦. More precisely, we
have the following.
Theorem 1.2. Let g ∈ L
2
(R
N
). The mth Bloch coeﬃcient of g is deﬁned as
follows:
(B
m
g)(η) =
_
R
N
g(y)e
−iy·η
¯
φ
m
(y; η)dy ∀m ≥ 1, η ∈ Y
.
Then the following inverse formula holds:
g(y) =
_
Y
∞
m=1
(B
m
g)(η)e
iy·η
φ
m
(y; η)dη.
Further, we have Parseval’s identity:
_
R
N
[g(y)[
2
dy =
_
Y
∞
m=1
[(B
m
g)(η)[
2
dη.
Finally, for all g in the domain of A, we have
Ag(y) =
_
Y
∞
m=1
λ
m
(η)(B
m
g)(η)e
iy·η
φ
m
(y; η)dη.
To obtain the spectral resolution of A
ε
in an analogous manner, let us introduce
Bloch waves at the εscale:
λ
ε
m
(ξ) = ε
−2
λ
m
(η), φ
ε
m
(x; ξ) = φ
m
(y; η), ψ
ε
m
(x; ξ) = ψ
m
(y; η),
where the variables (x, ξ) and (y, η) are related by y =
x
ε
and η = εξ. Observe that
φ
ε
m
(x; ξ) is εY periodic (in x) and ε
−1
Y
periodic with respect to ξ. In the same
manner, ψ
ε
m
(; ξ) is (εξ; εY )periodic because of the relation ψ
ε
m
(x; ξ) = e
ix·ξ
φ
ε
m
(x; ξ).
Note that the dual cell at εscale is ε
−1
Y
and hence we take ξ to vary in ε
−1
Y
in what
follows. With these notations, we have the following result analogous to Theorem 1.2.
Theorem 1.3. Let g ∈ L
2
(R
N
). The mth Bloch coeﬃcient of g at the εscale is
deﬁned as follows:
(B
ε
m
g)(ξ) =
_
R
N
g(x)e
−ix·ξ
¯
φ
ε
m
(x; ξ)dx ∀m ≥ 1, ξ ∈ ε
−1
Y
.
Then the following inverse formula and Parseval’s identity hold:
g(x) =
_
ε
−1
Y
∞
m=1
(B
ε
m
g)(ξ)e
ix·ξ
φ
ε
m
(x; ξ)dξ,
_
R
N
[g(x)[
2
dx =
_
ε
−1
Y
∞
m=1
[(B
ε
m
g)(ξ)[
2
dξ.
Finally, for all g in the domain of A
ε
, we get
A
ε
g(x) =
_
ε
−1
Y
∞
m=1
λ
ε
m
(ξ)(B
ε
m
g)(ξ)e
ix·ξ
φ
ε
m
(x; ξ)dξ.
1172 C. CONCA, R. ORIVE, AND M. VANNINATHAN
Using the above theorem, the classical homogenization result was deduced in
[11]. Let us recall the main steps. The ﬁrst one consists of considering a sequence
u
ε
∈ H
1
(R
N
) satisfying (1.9). We can express the equation A
ε
u
ε
= f in R
N
in the
equivalent form
λ
ε
m
(ξ)(B
ε
m
u
ε
)(ξ) = (B
ε
m
f)(ξ) ∀m ≥ 1, ξ ∈ ε
−1
Y
. (1.20)
In the homogenization process, one can neglect all the relations for m ≥ 2. More
precisely, it is proved in [11] that the following result holds.
Proposition 1.4. Let
v
ε
(x) =
_
ε
−1
Y
∞
m=2
(B
ε
m
u
ε
)(ξ)e
ix·ξ
φ
ε
m
(x; ξ)dξ. (1.21)
Then v
ε

L
2
(R
N
)
≤ cε.
Thus we can concentrate our attention only on the relation corresponding to the
ﬁrst Bloch wave:
λ
ε
1
(ξ)(B
ε
1
u
ε
)(ξ) = (B
ε
1
f)(ξ) ∀ξ ∈ ε
−1
Y
. (1.22)
The homogenized equation in the Fourier space
q
k
ξ
k
ξ
´ u∗(ξ) =
´
f(ξ) ∀ξ ∈ R
N
(1.23)
is obtained from (1.22) by passing to the limit as ε → 0. Here, the symbol ´ stands
for the classical Fourier transformation
´
f(ξ) =
1
(2π)
N/2
_
R
N
f(x)e
−ix·ξ
dx.
To this end, the following results were established and applied in [11].
Proposition 1.5. We assume that a
k
satisﬁes (1.2). Then there exists δ > 0
such that the ﬁrst eigenvalue λ
1
(η) is an analytic function on B
δ
def
=¦η [ [η[ < δ¦, and
there is a choice of the ﬁrst eigenvector φ
1
(y; η) satisfying
_
η → φ
1
(; η) ∈ H
1
#
(Y ) is analytic on B
δ
,
φ
1
(y; 0) = p
(0)
(= [Y [
−1/2
=
1
(2π)
N/2
).
Moreover, we have the relations
λ
1
(0) = 0, D
k
λ
1
(0) =
∂λ
1
∂η
k
(0) = 0 ∀k = 1, . . . , N,
1
2
D
2
k
λ
1
(0) =
1
2
∂
2
λ
1
∂η
k
∂η
(0) = q
k
∀k, = 1, . . . , N,
and there exist constants c and ¯c such that
c[η[
2
≤λ
1
(η)≤ ¯c[η[
2
∀η ∈ Y
, (1.24)
0 <λ
(N)
2
≤ λ
m
(η) ∀m ≥ 2, η ∈ Y
, (1.25)
where λ
(N)
2
is the second eigenvalue of the spectral problem for A in the cell Y with
Neumann boundary conditions on ∂Y .
BLOCH APPROXIMATION 1173
Apart from the above result of regularity on the Bloch spectrum, we need to
prove that the ﬁrst Bloch transform is an approximation to the Fourier transform.
This result is naturally expected from the fact that φ
ε
1
(x; ξ) → (2π)
−N/2
, as ε → 0,
∀ξ ∈ R
N
.
Proposition 1.6. Let g
ε
and g be in L
2
(R
N
). Then
(i) if g
ε
g weakly in L
2
(R
N
x
), then χ
ε
−1
Y
B
ε
1
g
ε
´ g weakly in L
2
loc
(R
N
ξ
) pro
vided there is a ﬁxed compact set K such that supp(g
ε
) ⊂ K ∀ε;
(ii) if g
ε
→ g in L
2
(R
N
x
), then χ
ε
−1
Y
B
ε
1
g
ε
→ ´ g in L
2
loc
(R
N
ξ
).
These results easily lead us to the following homogenization theorem in R
N
.
Theorem 1.7. We consider a sequence u
ε
satisfying (1.9). Then
a
ε
k
∂u
ε
∂x
q
k
∂u
∗
∂x
in L
2
(R
N
) ∀k = 1, . . . , N.
In particular, u
∗
satisﬁes A
∗
u
∗
= f in R
N
.
Once the homogenization result in R
N
is established, it is an easy matter to
deduce the corresponding result in a bounded domain Ω by localization techniques
using a cutoﬀ function φ ∈ T(Ω) (see [11]).
1.2. Presentation of new results: The Bloch approximation. Let us con
sider the sequence u
ε
satisfying hypotheses (1.9). The Bloch approximation of u
ε
is
deﬁned by the following formula:
θ
ε
(x)
def
=
_
ε
−1
Y
´
u
∗
(ξ)e
ix·ξ
φ
ε
1
(x; ξ)dξ, x ∈ R
N
. (1.26)
First of all, let us remark that this object is not diﬃcult to be computed in principle.
Our goal throughout this paper is to study properties of this function and particularly
its relations with various correctors terms. It is worth noticing that θ
ε
is deﬁned only
in terms of the ﬁrst Bloch mode φ
ε
1
. We will see in section 3 that higher Bloch modes
φ
ε
m
, m ≥ 2, do not contribute at all in the analysis of the correctors of ﬁrst and
second order in the energy norm. (It will be interesting to know whether these higher
order modes play a part in the analysis of correctors in stronger norms H
2
, . . . , etc.
For H
2
estimates, we refer to our work in [12].) Thus we are motivated to introduce
the projection onto the ﬁrst Bloch mode: for all g ∈ L
2
(R
N
), we deﬁne
P
ε
1
g(x) =
_
ε
−1
Y
B
ε
1
g(ξ)e
ix·ξ
φ
ε
1
(x; ξ)dξ, x ∈ R
N
. (1.27)
We note by the item (ii) of Proposition 1.6 that the Fourier transform ´ u∗ is an
approximation of B
ε
1
u
ε
. Therefore, heuristically speaking, the Bloch approximation
θ
ε
is close to P
ε
1
u
ε
and hence to u
ε
. With these notations, we will prove the following
theorem.
Theorem 1.8. Assume that the coeﬃcients a
k
satisfy (1.2). Let u
ε
be the
sequence introduced in (1.9). Then if f ∈ L
2
(R
N
), we have
(u
ε
−θ
ε
) → 0 in H
1
(R
N
). (1.28)
Furthermore, we have the estimate
[u
ε
−θ
ε
[
H
1
(R
N
)
≤ cεf
L
2
(R
N
)
. (1.29)
1174 C. CONCA, R. ORIVE, AND M. VANNINATHAN
It is worth remarking that even though error estimates of the type (1.29) are
sometimes found in the literature, they are usually obtained using the maximum
principle with more regularity hypotheses on a
k
and f. Here, we obtain these natural
estimates under optimal hypotheses.
Thanks to the above result, we are reduced to expanding θ
ε
in terms of ε in order
to be able to compare it with the classical correctors for u
ε
. To fulﬁll this task, it is
clear from the deﬁnition of θ
ε
that it is necessary to obtain asymptotic expansions
of the ﬁrst eigenvalue λ
ε
1
(ξ) and the ﬁrst Bloch mode φ
ε
1
(; ξ). (In addition, for our
purposes below, we need an asymptotic expansion of the ﬁrst Bloch transform B
ε
1
g(ξ)
for which we refer the reader to section 5. These results strengthen earlier results,
particularly those of Proposition 1.6.) We now state results in this direction, and their
proofs will be taken up in the following sections along with other auxiliary results.
First, we introduce some test functions χ
k
, χ
km
, χ
kmn
deﬁned by the following cell
problems (observe that the ﬁrst ones are nothing but the functions already introduced
in (1.14)):
_
¸
¸
_
¸
¸
_
Aχ
k
= (a
k
−q
k
) −
1
2
_
C
k
χ
+C
χ
k
_
in R
N
,
χ
k
∈ H
1
#
(Y ), M
Y
(χ
k
) = 0.
(1.30)
_
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
_
Aχ
km
=
1
3
_
(a
k
−q
k
)χ
m
+ (a
m
−q
m
)χ
k
+ (a
mk
−q
mk
)χ
−C
k
χ
m
−C
χ
mk
− C
m
χ
k
_
in R
N
,
χ
km
∈ H
1
#
(Y ), M
Y
(χ
km
) = 0.
(1.31)
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
Aχ
kmn
=
1
4!
D
4
kmn
λ
1
(0) −
1
4
_
C
n
χ
km
+C
k
χ
mn
+C
χ
mnk
+C
m
χ
nk
_
+
1
3!
_
(a
k
−q
k
)χ
mn
+ (a
m
−q
m
)χ
kn
+ (a
km
−q
km
)χ
n
+ (a
mn
−q
mn
)χ
k
+ (a
n
−q
n
)χ
km
+ (a
kn
−q
kn
)χ
m
_
in R
N
,
χ
kmn
∈ H
1
#
(Y ), M
Y
(χ
kmn
) = 0.
(1.32)
Proposition 1.9. All odd order derivatives of λ
1
at η = 0 vanish, i.e.,
D
β
λ
1
(0) = 0 ∀β ∈ Z
N
+
, [β[ odd.
All even order derivatives of λ
1
at η = 0 can be calculated in a systematic fash
ion. For instance, the fourth order derivatives have the following expressions: for all
k, , m, n = 1, . . . , N
1
4!
D
4
kmn
λ
1
(0) =
1
4
1
[Y [
_
Y
_
C
n
χ
km
+C
k
χ
mn
+C
χ
mnk
+C
m
χ
nk
_
dy
BLOCH APPROXIMATION 1175
−
1
3!
1
[Y [
_
Y
_
(a
k
−q
k
)χ
mn
+ (a
m
−q
m
)χ
nk
+ (a
mn
−q
mn
)χ
k
+ (a
nk
−q
nk
)χ
m
+ (a
km
−q
km
)χ
n
+ (a
n
−q
n
)χ
km
_
dy.
Various derivatives of φ
1
at η = 0 can also be calculated in a systematic fashion.
Proposition 1.10. We have the following expressions:
D
k
φ
1
(y; 0) = ip
(0)
χ
k
(y),
1
2!
D
2
k
φ
1
(y; 0) = − p
(0)
χ
k
(y) −β
(2)
k
p
(0)
,
1
3!
D
3
km
φ
1
(y; 0) = − ip
(0)
χ
km
(y) −
i
3
_
β
(2)
k
χ
m
(y) +β
(2)
m
χ
k
(y) +β
(2)
mk
χ
(y)
_
p
(0)
,
1
4!
D
4
kmn
φ
1
(y; 0) = p
(0)
χ
kmn
(y) −
1
3!
_
β
(2)
k
χ
mn
(y) +β
(2)
m
χ
nk
(y) +β
(2)
mn
χ
k
+ β
(2)
nk
χ
m
(y) +β
(2)
km
χ
n
(y) +β
(2)
n
χ
km
(y)
_
p
(0)
+β
(4)
kmn
p
(0)
with
β
(2)
k
=
1
2!
1
[Y [
_
Y
χ
χ
k
dy,
β
(4)
kmn
=
1
[Y [
_
Y
1
4
_
χ
mn
χ
k
+χ
kmn
χ
+χ
nk
χ
m
+χ
kn
χ
n
_
dy
−
1
[Y [
_
Y
1
6
_
χ
m
χ
kn
+χ
km
χ
n
+χ
k
χ
nm
_
dy
+
1
[Y [
1
2
_
β
(2)
k
β
(2)
mn
+β
(2)
km
β
(2)
n
+β
(2)
kn
β
(2)
m
_
.
We note that all odd order derivatives of φ
1
at η = 0 are purely imaginary and
all even order derivatives are real.
Since φ
1
(; η) is proved to be analytic for [η[ ≤ δ, we can expand it and thus give
rise to an asymptotic expansion of θ
ε
which is as follows:
θ
ε
(x) = u
∗
(x) +εχ
k
_
x
ε
_
∂u
∗
∂x
k
(x) −ε
2
_
χ
k
_
x
ε
_
+β
(2)
k
_
∂
2
u
∗
∂x
k
∂x
(x) + . (1.33)
This can be rigorously proved. Our next result is a sample where we specify the
precise hypotheses needed to justify the above expansion up to three terms.
Theorem 1.11. Assume that the hypotheses of Theorem 1.8 hold.
(i) If u
∗
∈ H
1
(R
N
), then
θ
ε
−u
∗

L
2
(R
N
)
≤ cεu
∗

H
1
(R
N
)
.
(ii) If f ∈ L
2
(R
N
) and χ
k
∈ W
1,∞
#
(Y ), where χ
k
is the solution of (1.13) and
χ
ε
k
(x) = χ
k
_
x
ε
_
, then we have
_
_
_
_
θ
ε
−u
∗
−εχ
ε
k
∂u
∗
∂x
k
_
_
_
_
H
1
(R
N
)
≤ cεf
L
2
(R
N
)
.
1176 C. CONCA, R. ORIVE, AND M. VANNINATHAN
(iii) If f ∈ H
1
(R
N
) and χ
k
, χ
k
∈ W
1,∞
#
(Y ), where χ
k
is the solution of (1.30),
β
(2)
k
are constants deﬁned in Proposition 1.10, and χ
ε
k
(x) = χ
k
_
x
ε
_
, then
_
_
_
_
θ
ε
−u
∗
−εχ
ε
k
∂u
∗
∂x
k
+ε
2
_
χ
ε
k
+β
(2)
k
_
∂
2
u
∗
∂x
k
∂x
_
_
_
_
H
1
(R
N
)
≤ cε
2
f
H
1
(R
N
)
.
It is important to note that these above expansions are of Taylor type owing to
the analyticity of λ
1
(η) and φ
1
(; η). This is the main diﬀerence between this approach
and the classical one found in [5], where the expansion has a multiscale structure.
Concerning the hypotheses on the smoothness of functions χ
k
and χ
k
in state
ments (ii) and (iii), it is worth mentioning from regularity theory of elliptic boundary
value problems that W
1,∞
estimates are hard to come by. This is why they are usually
assumed in homogenization theory. However, several numerical studies with simple
ﬁbers show that these assumptions are valid. Thus they are reasonable hypotheses to
work with as far as certain applications are concerned.
The expansions of λ
1
(η), φ
ε
1
(; η), and B
ε
1
g(ξ) obtained in Propositions 1.5 and 1.9
and Propositions 5.1, 5.2, and 5.3 below have further interesting consequences which
will be developed in a forthcoming paper. For the time being, we will be content
with a few remarks. Since higher order modes can be neglected, the ﬁrst eigenvalue
λ
1
(η) along with the ﬁrst eigenvector φ
1
(; η) represent the periodic medium under
consideration. Their contributions occur somewhat separately without interaction at
the levels of homogenized equation and correctors. More precisely, the ﬁrst eigenvalue
λ
1
(η) contributes at various levels through its derivatives at η = 0. The ﬁrst eigen
vector φ
1
(; η) and its ﬁrst derivatives contribute through the ﬁrst Bloch transform
B
ε
1
g(ξ) and its expansion described in Propositions 5.2 and 5.3.
In the homogenized equation, for instance, we see the product of the second order
derivatives of λ
1
(η) at η = 0 with the 0th order term of B
ε
1
g(ξ), namely, ´ g(ξ). We
see a similar structure in the correctors, too. There seem to be situations where both
interact and contribute jointly in a manner diﬀerent from the above. One example
of such a situation is the study of the propagation of waves in a periodic medium.
It appears that the homogenized medium is not good enough to provide an approx
imation to the propagation for large times because of the appearance of dispersion
eﬀects shown numerically in F. Santosa and W. W. Symes [20]. We feel that this is an
appropriate place to highlight the improvements achieved in this work with respect
to [20]. Apart from the mathematical rigor, the main point is that the third order
derivatives of λ
1
(η) at η = 0 are shown to be zero even in the multidimensional case.
(In fact all odd order derivatives vanish.) Moreover, our arguments are more general
compared with the onedimensional case covered in [20]. This will have consequences
in the propagation of waves in periodic media. We plan to cover these aspects in a
future publication.
We conclude this introduction by saying how the rest of this paper is organized.
Section 2 is devoted to certain fundamental lemmas which are indispensable. As an
immediate application, we prove in section 3 that the higher order Bloch modes are
negligible. Taylor expansions for λ
1
and φ
1
are obtained in section 4 which proves
Propositions 1.9 and 1.10. Section 5 is devoted to the description of the asymptotic
behavior of the ﬁrst Bloch transform B
ε
1
whose deﬁnition is given in Theorem 1.2.
Finally, in section 6, we present the proofs of the main results, namely, Theorems 1.8
and 1.11.
BLOCH APPROXIMATION 1177
2. Fundamental lemmas. In this section, we prove a series of results which
generalize Parseval’s identity stated in Theorem 1.3. These estimates will be useful
later for the analysis of correctors. The following two lemmas are easily seen to be
generalizations of wellknown classical results for −∆.
Lemma 2.1. For all g ∈ H
1
(R
N
), we have
c
1
[g[
2
H
1
(R
N
)
≤
_
ε
−1
Y
∞
m=1
λ
ε
m
(ξ)[B
ε
m
g(ξ)[
2
dξ ≤ c
2
[g[
2
H
1
(R
N
)
,
where c
1
and c
2
are constants independent of ε and g.
Proof. First of all, by uniform ellipticity of A
ε
, we have
α
_
R
N
[∇g[
2
dx ≤
_
R
N
A
ε
g¯ gdx ≤ β
_
R
N
[∇g[
2
dx.
We can rewrite the middle term by applying the Plancherel identity:
_
R
N
g(x)h(x)dx =
_
ε
−1
Y
∞
m=1
B
ε
m
g(ξ)B
ε
m
h(ξ)dξ ∀g, h ∈ L
2
(R
N
). (2.1)
Indeed, using the spectral resolution of A
ε
, we get
_
R
N
A
ε
g¯ gdx =
_
ε
−1
Y
∞
m=1
λ
ε
m
(ξ)[B
ε
m
g(ξ)[
2
dξ.
This completes the proof.
By using the duality between H
1
(R
N
) and H
−1
(R
N
), we deduce the following.
Lemma 2.2. For all g ∈ H
−1
(R
N
), there exist c
1
and c
2
which are independent
of ε and g, such that
c
1
g
2
H
−1
(R
N
)
≤
_
ε
−1
Y
∞
m=1
1
1 +λ
ε
m
(ξ)
[B
ε
m
g(ξ)[
2
dξ ≤ c
2
g
2
H
−1
(R
N
)
.
Proof. It is well known that (A
ε
+ I): H
1
(R
N
) → H
−1
(R
N
) is an isomorphism.
For every g ∈ H
−1
(R
N
) there exists a unique solution u ∈ H
1
(R
N
) of A
ε
u +u = g in
R
N
. We can express the previous equation in the equivalent form
(λ
ε
m
(ξ) + 1)B
ε
m
u(ξ) = B
ε
m
g(ξ) ∀m ≥ 1, ξ ∈ ε
−1
Y
.
Therefore, an application of the Cauchy–Schwarz inequality yields
¸g, v) =
_
ε
−1
Y
∞
m=1
(λ
ε
m
(ξ) + 1)B
ε
m
uB
ε
m
vdξ
≤
_
_
ε
−1
Y
∞
m=1
(λ
ε
m
(ξ) + 1)[B
ε
m
u[
2
dξ
_
1/2
_
_
ε
−1
Y
∞
m=1
(λ
ε
m
(ξ) + 1)[B
ε
m
v[
2
dξ
_
1/2
≤
_
_
ε
−1
Y
∞
m=1
[B
ε
m
g[
2
(λ
ε
m
(ξ) + 1)
dξ
_
1/2
_
_
ε
−1
Y
∞
m=1
(λ
ε
m
(ξ) + 1)[B
ε
m
v[
2
dξ
_
1/2
1178 C. CONCA, R. ORIVE, AND M. VANNINATHAN
for all v ∈ H
1
(R
N
), g ∈ H
−1
(R
N
). Here, ¸, ) denotes the H
1
(R
N
) and H
−1
(R
N
)
duality pairing. By virtue of Lemma 2.1 and Parseval’s identity, the second term in
the righthand side is equivalent to the H
1
norm of v. Thus we deduce the existence
of a constant c
1
such that
c
1
g
2
H
−1
(R
N
)
≤
_
ε
−1
Y
∞
m=1
1
1 +λ
ε
m
(ξ)
[B
ε
m
g(ξ)[
2
dξ,
which is the lower estimate in Lemma 2.2. To prove the upper estimate is enough
to use the continuity of the solution u ∈ H
1
(R
N
) with respect to the righthand side
g ∈ H
−1
(R
N
).
In our next lemma, we consider g
ε
= g
ε
(ξ) a measurable function deﬁned on
ε
−1
Y
, and another function ρ = ρ(y; η) measurable with respect to (y; η) and Y 
periodic in y. We then introduce
G
ε
(x) =
_
ε
−1
Y
g
ε
(ξ)e
ix·ξ
ρ
_
x
ε
; εξ
_
dξ, x ∈ R
N
. (2.2)
The following result estimates its L
2
(R
N
) and H
1
(R
N
) norms.
Lemma 2.3. We assume g
ε
∈ L
2
(ε
−1
Y
) and ρ ∈ L
∞
(Y
; H
1
#
(Y )). Then we
have
G
ε

2
L
2
(R
N
)
=
_
ε
−1
Y
[g
ε
(ξ)[
2
ρ(; εξ)
2
L
2
(Y )
dξ,
[G
ε
[
2
H
1
(R
N
)
=
_
ε
−1
Y
[g
ε
(ξ)[
2
iξρ(; εξ) +ε
−1
∇
y
ρ(; εξ)
2
L
2
(Y )
N
dξ.
Proof. We expand ρ(y; η) as a function of y in the orthonormal basis ¦φ
m
(y; η)¦
∞
m=1
where η is a parameter:
ρ(y; η) =
∞
m=1
a
m
(η)φ
m
(y; η).
Introducing this expression in (2.2), we get
G
ε
(x) =
_
ε
−1
Y
g
ε
(ξ)
∞
m=1
a
m
(εξ)e
ix·ξ
φ
ε
m
(x; ξ)dξ.
Applying Parseval’s identity of Theorem 1.3, we get
G
ε

2
L
2
(R
N
)
=
_
ε
−1
Y
[g
ε
(ξ)[
2
∞
m=1
[a
m
(εξ)[
2
dξ.
This completes the proof of the ﬁrst part of the lemma if we use Parseval’s identity
in L
2
(Y ):
ρ(; η)
2
L
2
(Y )
=
∞
m=1
[a
m
(η)[
2
∀η ∈ Y
. (2.3)
For the second part of the lemma, we formally diﬀerentiate G
ε
(x) with respect
to x. We obtain
∇
x
G
ε
(x) =
_
ε
−1
Y
g
ε
(ξ)e
ix·ξ
_
iξρ
_
x
ε
; εξ
_
+ε
−1
∇
y
ρ
_
x
ε
; εξ
__
dξ.
BLOCH APPROXIMATION 1179
We remark that the above integral is of the same type as the one analyzed in the ﬁrst
part. This completes the proof.
The next lemma presents H
1
estimates on the Bloch modes.
Lemma 2.4. We suppose that the coeﬃcients a
k
satisfy (1.2). Then there exists
a constant c depending on a
k

L
∞
(Y )
such that
_
_
_
∂φ
m
∂y
k
(; η)
_
_
_
L
2
(Y )
≤ c
1
λ
m
(η)
1/2
∀η ∈ Y
, m ≥ 1, k = 1, . . . , N. (2.4)
To prove this, let us introduce the bilinear forms associated with the operators
A(η) and A, respectively.
a(η; φ, ψ) =
_
Y
a
k
(y)
_
∂φ
∂y
+iη
φ
__
∂ψ
∂y
k
+iη
k
ψ
_
dy,
a(φ, ψ) =
_
Y
a
k
(y)
∂φ
∂y
∂ψ
∂y
k
dy.
The basic estimates on them are obtained in [10, p. 190]: There exist constants c, ¯c
which are independent of η ∈ Y
such that for all φ ∈ H
1
#
(Y ),
c (∇φ
2
L
2
(Y )
N
+[η[
2
φ
2
L
2
(Y )
) ≤ a(η; φ, φ) ≤ ¯c (∇φ
2
L
2
(Y )
N
+[η[
2
φ
2
L
2
(Y )
), (2.5)
c ∇φ
2
L
2
(Y )
N
≤ a(φ, φ) ≤ ¯c ∇φ
2
L
2
(Y )
N
. (2.6)
Proof of Lemma 2.4. For simplicity, we denote φ
m
(; η) by φ
m
(η). We recall that
it satisﬁes
a(η; φ
m
(η), ψ) = λ
m
(η)(φ
m
(η), ψ) ∀ψ ∈ H
1
#
(Y ). (2.7)
To deduce (2.4), it is enough to take ψ = φ
m
(η) and use (2.5).
Our next result concerns the estimation of expressions which are inverse to (2.2).
We deﬁne
J
ε
g(ξ) =
_
R
N
g(x)e
−ix·ξ
ρ
_
x
ε
; εξ
_
dx for ξ ∈ ε
−1
Y
, (2.8)
where g = g(x) is a measurable function deﬁned on R
N
and ρ = ρ(y; η) is a measurable
function deﬁned on Y Y
. We assume that ρ is Y periodic in y. The required
hypotheses on these functions will depend on the estimate deduced on J
ε
g. This is
illustrated in the results that follow which are analogous to classical estimates on the
Fourier transform.
Lemma 2.5.
(i) If g ∈ L
2
(R
N
) and ρ ∈ L
∞
(Y
; L
2
#
(Y )), then we have
J
ε
g
L
2
(ε
−1
Y
)
≤ g
L
2
(R
N
)
ρ
L
∞
(Y
;L
2
#
(Y ))
.
(ii) If g ∈ H
1
(R
N
) and ρ ∈ L
∞
(Y
; H
1
#
(Y )), then we have
(1 +[ξ[
2
)
1/2
J
ε
g(ξ)
L
2
(ε
−1
Y
)
≤ c
_
∇g
L
2
(R
N
)
ρ
L
∞
(Y
;L
2
(Y ))
+ε
−1
g
L
2
(R
N
)
∇
y
ρ
L
∞
(Y
;L
2
(Y )
N
)
_
.
1180 C. CONCA, R. ORIVE, AND M. VANNINATHAN
Proof. The idea is to consider the product space L
2
(Y
; L
2
#
(Y )) and expand
ρ(y; η) in two steps. First using the fact that ¦
¯
φ
m
(; η)¦
∞
m=1
is an orthonormal basis
in L
2
#
(Y ), we get
ρ(y; η) =
∞
m=1
a
m
(η)
¯
φ
m
(y; η) ∀y ∈ Y, η ∈ Y
.
Next, for each m, we can expand a
m
(η) in the usual Fourier series:
a
m
(η) =
n∈Z
N
a
mn
e
2πin·η
∀η ∈ Y
.
The corresponding Parseval’s identities are as follows:
ρ(; η)
2
L
2
(Y )
=
m
[a
m
(η)[
2
∀η ∈ Y
,
_
Y
[a
m
(η)[
2
dη =
n∈Z
N
[a
mn
[
2
∀m ∈ N.
Using this expansion, we can rewrite J
ε
g as follows:
J
ε
g(ξ) =
∞
m=1
n∈Z
N
a
mn
e
2πiεn·ξ
_
R
N
g(x)e
−ix·ξ
¯
φ
m
_
x
ε
; εξ
_
dx,
which, according to the deﬁnition of B
ε
m
g(ξ), is equal to
J
ε
g(ξ) =
∞
m=1
n∈Z
N
a
mn
e
2πiεn·ξ
B
ε
m
g(ξ) =
∞
m=1
a
m
(εξ)B
ε
m
g(ξ).
By the Cauchy–Schwarz inequality,
[J
ε
g(ξ)[
2
≤
_
∞
m=1
[a
m
(εξ)[
2
__
∞
m=1
[B
ε
m
g(ξ)[
2
_
= ρ(; εξ)
2
L
2
(Y )
_
∞
m=1
[B
ε
m
g(ξ)[
2
_
≤ ρ
2
L
∞
(Y
;L
2
#
(Y ))
_
∞
m=1
[B
ε
m
g(ξ)[
2
_
.
The proof of (i) is complete if we integrate the above inequality with respect to
ξ ∈ ε
−1
Y
and apply Theorem 1.3. For the proof of (ii), we multiply (2.8) by (−iξ
k
)
and obtain
(−iξ
k
)J
ε
g(ξ) =
_
R
N
g(x)(−iξ
k
)e
−ix·ξ
ρ
_
x
ε
; εξ
_
dx,
which, by integration by parts, can be rewritten as
(−iξ
k
)J
ε
g(ξ) = −
_
R
N
∂g
∂x
k
(x)e
−ix·ξ
ρ
_
x
ε
; εξ
_
dx −ε
−1
_
R
N
g(x)e
−ix·ξ
∂ρ
∂y
k
_
x
ε
; εξ
_
dx.
BLOCH APPROXIMATION 1181
It is now suﬃcient to apply (i) to each of the terms on the righthand side of the
above relation.
Next, we will need some properties of the classical discrete Fourier transform
in our asymptotic description of the ﬁrst Bloch transform. In particular, we are
interested in the relation between discrete and continuous Fourier transforms. To this
end, let us begin by introducing some necessary notations. Let ¦Y
ε
¦
∈Z
N be the mesh
of R
N
generated by the cell εY . More precisely, Y
ε
= x
ε
+ εY where x
ε
= 2πε is
the origin of the cell Y
ε
. We recall the deﬁnition of the discrete Fourier transform of
a function corresponding to this mesh: Let p > N be given. For g ∈ W
1,p
(R
N
) with
compact support we deﬁne
F
ε
g(ξ) =
∈Z
N
g(x
ε
)e
−ix
ε
·ξ
∀ξ ∈ ε
−1
Y
. (2.9)
It is worthwhile to recall that W
1,p
(R
N
) is embedded in C
0
(R
N
) when p > N, and so
g(x
ε
) is well deﬁned.
Lemma 2.6. For g ∈ W
1,p
(R
N
) (p > N) with compact support K, we have
(i) ε
N
(χ
ε
−1
Y
F
ε
g)(ξ) →
1
(2π)
N/2
´ g(ξ) for ξ ∈ R
N
.
(ii) ε
N
F
ε
g
L
2
(ε
−1
Y
)
≤ c[K[
p−2
2p
¦g
L
p
(R
N
)
+ ε∇g
L
p
(R
N
)
N
¦, [K[ denotes the
measure of K.
(iii) ε
N
χ
ε
−1
Y
F
ε
g →
1
(2π)
N/2
´ g in L
2
(R
N
).
Proof. To prove (i), we multiply (2.9) by ε
N
to get
ε
N
F
ε
g(ξ) =
1
(2π)
N
∈Z
N
g(x
ε
)e
−ix
ε
·ξ
[Y
ε
[.
We regard the righthand side of the above equality as a Riemann sum of the integral
1
(2π)
N
_
R
N
g(x)e
−ix·ξ
dx
which converges to it as ε → 0.
To prove (ii), we observe that the righthand side of (2.9) is nothing but the
Fourier series in the variable ξ ∈ ε
−1
Y
. Therefore, by Parseval’s identity, we get
ε
N
_
ε
−1
Y
[F
ε
g(ξ)[
2
dξ =
∈Z
N
[g(x
ε
)[
2
.
We multiply this relation by ε
N
and rewrite it as
ε
2N
_
ε
−1
Y
[F
ε
g(ξ)[
2
dξ =
1
(2π)
N
∈Z
N
[g(x
ε
)[
2
[Y
ε
[. (2.10)
To estimate the righthand side of the above equality, we integrate the inequality
[g(x
ε
)[
2
≤ 2
_
[g(x)[
2
+[g(x) −g(x
ε
)[
2
_
, x ∈ Y
ε
,
over Y
ε
to obtain
[g(x
ε
)[
2
[Y
ε
[ ≤ 2
_
_
Y
ε
[g(x)[
2
dx +
_
Y
ε
[g(x) −g(x
ε
)[
2
dx
_
. (2.11)
1182 C. CONCA, R. ORIVE, AND M. VANNINATHAN
Since p > N, we can use the classical Morrey’s inequality (see Brezis [7, p. 167]) to
deduce
[g(x) −g(x
ε
)[ ≤ cε
1−
N
p
[∇g[
L
p
(Y
ε
)
N
.
Using both the above estimate in (2.11) and the H¨ older inequality and summing over
∈ Z
N
, we complete the proof of (ii).
To prove the statement (iii) we ﬁrst use (i) and (ii) to deduce that
ε
N
χ
ε
−1
Y
(ξ)F
ε
g(ξ)
1
(2π)
N/2
´ g(ξ) in L
2
(R
N
)weak.
Let us now expand
_
_
_
_
ε
N
χ
ε
−1
Y
F
ε
g −
1
(2π)
N/2
´ g
_
_
_
_
2
L
2
(R
N
)
= ε
2N
F
ε
g
2
L
2
(R
N
)
−
2ε
N
(2π)
N/2
(χ
ε
−1
Y
F
ε
g, ´ g)
+
1
(2π)
N
´ g
2
L
2
(R
N
)
.
Now relation (2.10) shows that
ε
2N
F
ε
g
2
L
2
(R
N
)
→
1
(2π)
N
_
R
N
[g[
2
dx =
1
(2π)
N
´ g
2
L
2
(R
N
)
.
Thanks to the above weak convergence, the second term converges to
−
2
(2π)
N/2
1
(2π)
N/2
´ g
2
L
2
(R
N
)
.
This simple computation establishes the strong convergence in L
2
(R
N
).
3. Higher Bloch modes are negligible. In this section, we consider a se
quence of solutions u
ε
of the equation with f ∈ H
−1
(R
N
):
A
ε
u
ε
= f in R
N
, u
ε
∈ H
1
(R
N
). (3.1)
Let us recall that the above equation is equivalent to (1.20) in the Bloch space. In
what follows, we present a systematic method of obtaining estimates on the solution
in Sobolev spaces L
2
and H
1
. In particular, we show that the component of u
ε
in the
higher Bloch modes does not play any role in the analysis of correctors of ﬁrst and
second order provided f is suﬃciently smooth. Thus we consider v
ε
deﬁned in (1.21),
which is nothing but the projection of u
ε
corresponding to all higher Bloch modes.
Estimates on v
ε
derived in this section improve Proposition 1.4.
Proposition 3.1. We have the following estimates for f ∈ L
2
(R
N
):
(i) [v
ε
[
H
1
(R
N
)
≤ cεf
L
2
(R
N
)
,
(ii) v
ε

L
2
(R
N
)
≤ cεf
H
−1
(R
N
)
.
Proof. To show (i), we apply Lemma 2.1 with g = v
ε
and use (1.20). We obtain
∇v
ε

2
L
2
(R
N
)
N
≤ c
_
ε
−1
Y
∞
m=2
1
λ
ε
m
(ξ)
[B
ε
m
f(ξ)[
2
dξ
≤ c sup
m≥2, ξ∈ε
−1
Y
1
λ
ε
m
(ξ)
f
2
L
2
(R
N
)
.
BLOCH APPROXIMATION 1183
Proof of (i) is complete since we have (cf. (1.25))
sup
m≥2, ξ∈ε
−1
Y
1
λ
ε
m
(ξ)
≤
1
λ
(N)
2
ε
2
. (3.2)
For the proof of (ii), we apply Lemma 2.2 with g = f and (1.20). We have
v
ε

2
L
2
(R
N
)
=
_
ε
−1
Y
∞
m=2
[B
ε
m
u
ε
(ξ)[
2
dξ
=
_
ε
−1
Y
∞
m=2
1
λ
ε
m
(ξ)
2
[B
ε
m
f(ξ)[
2
dξ.
Writing
1
λ
ε
m
(ξ)
2
[B
ε
m
f(ξ)[
2
=
1 +λ
ε
m
(ξ)
λ
ε
m
(ξ)
2
[B
ε
m
f(ξ)[
2
1 +λ
ε
m
(ξ)
and using (3.2), we deduce that
1
λ
ε
m
(ξ)
2
[B
ε
m
f(ξ)[
2
≤ cε
2
[B
ε
m
f(ξ)[
2
1 +λ
ε
m
(ξ)
.
The proof is complete if we use Lemma 2.2.
While the above proposition shows that v
ε
can be neglected at the level of the
ﬁrst order correctors (cf. (1.10)), the next result will demonstrate that v
ε
can be
neglected at the level of correctors of ﬁrst and second order. These ﬁner estimates
require naturally higher order regularity of f but not of the coeﬃcients a
k
(y). Let us
state the following proposition, whose proof is similar to the previous one and hence
will not be repeated.
Proposition 3.2. We have the following estimates for f ∈ H
1
(R
N
):
(i) [v
ε
[
H
1
(R
N
)
≤ cε
2
f
H
1
(R
N
)
,
(ii) v
ε

L
2
(R
N
)
≤ cε
2
f
L
2
(R
N
)
.
Assuming a
k
are in W
1,∞
#
(Y ) and further assumptions, we can obtain H
2

estimates on the solution. This is diﬃcult as it involves more subtleties (see [12]).
4. Taylor expansion of the ﬁrst Bloch eigenvalue and eigenvector. The
purpose of this section is to indicate a systematic method to compute derivatives
of the ﬁrst Bloch eigenvalue λ
1
(η) and the ﬁrst Bloch eigenvector φ
1
(; η) at η = 0.
In particular, we will prove Propositions 1.9 and 1.10. Recall that λ
1
(η) and φ
1
(; η)
depend analytically on η in a small neighborhood B
δ
of η = 0. At the cost of reducing
this neighborhood, we claim that the branch η → φ
1
(; η) can be chosen so that the
following conditions are satisﬁed simultaneously:
η ∈ B
δ
→ φ
1
(; η) ∈ H
1
#
(Y ) is analytic, (4.1)
φ
1
(; η)
L
2
(Y )
= 1 ∀η ∈ B
δ
, (4.2)
Im
_
Y
φ
1
(y; η)dy = 0 ∀η ∈ B
δ
. (4.3)
1184 C. CONCA, R. ORIVE, AND M. VANNINATHAN
In what follows, we will see that the above conditions uniquely ﬁx the eigenvector
φ
1
(; η). We remark that the condition (4.2) is classical, whereas the condition (4.3)
is somewhat unusual and can be achieved as indicated below. The idea consists of
multiplying φ
1
(; η) by a complex number (α
1
(η) +iα
2
(η)) where α
1
(η) and α
2
(η) are
real analytic with respect to η and are chosen such that
Im
_
Y
(α
1
(η) +iα
2
(η))φ
1
(y; η)dy = 0.
If we deﬁne
d(η) = (d
1
(η), d
2
(η))
def
=
_
Im
_
Y
φ
1
(y; η)dy, Re
_
Y
φ
1
(y; η)dy
_
,
then the above condition is equivalent to
α
1
(η)d
1
(η) +α
2
(η)d
2
(η) = 0 ∀η ∈ B
δ
.
Obviously, one such choice which is analytic is as follows:
α
1
(η) = −d
2
(η), α
2
(η) = d
1
(η).
Of course, the above procedure has destroyed condition (4.2) (but not condition (4.1)).
However, it can be regained by dividing by [d(η)[. This is possible because d(0) ,= 0
by our choice of φ
1
(; 0) (see Proposition 1.5).
Thanks to our choice of the branch satisfying (4.1)–(4.3), we will now draw some
consequences which will simplify the computations below. In fact, diﬀerentiating (4.2)
with respect to η, we successively get for all k, , m, n = 1, . . . , N
Re¸D
k
φ
1
(; η), φ
1
(; η)) = 0, (4.4)
Re¸D
2
k
φ
1
(; η), φ
1
(; η)) + Re¸D
k
φ
1
(; η), D
φ
1
(; η)) = 0, (4.5)
_
Re¸D
3
km
φ
1
(; η), φ
1
(; η)) + Re¸D
2
k
φ
1
(; η), D
m
φ
1
(; η))
+ Re¸D
2
km
φ
1
(; η), D
φ
1
(; η)) + Re¸D
k
φ
1
(; η), D
2
m
φ
1
(; η)) = 0,
(4.6)
_
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
_
Re¸D
4
kmn
φ
1
(; η), φ
1
(; η)) + Re¸D
3
km
φ
1
(; η), D
n
φ
1
(; η))
+ Re¸D
3
kn
φ
1
(; η), D
m
φ
1
(; η)) + Re¸D
2
k
φ
1
(; η), D
2
mn
φ
1
(; η))
+ Re¸D
3
kmn
φ
1
(; η), D
φ
1
(; η)) + Re¸D
2
km
φ
1
(; η), D
2
n
φ
1
(; η))
+ Re¸D
2
kn
φ
1
(; η), D
2
m
φ
1
(; η)) + Re¸D
k
φ
1
(; η), D
3
mn
φ
1
(; η)) = 0,
(4.7)
where ¸; ) denotes the scalar product in L
2
#
(Y ). On the other hand, diﬀerentiation
of (4.3) yields
Im
_
Y
D
β
φ
1
(y; η)dy = 0 ∀ β ∈ Z
N
+
. (4.8)
From these sets of relations, it follows that
_
Y
D
β
φ
1
(y; 0)dy = 0 ∀ β ∈ Z
N
+
with [β[ odd. (4.9)
BLOCH APPROXIMATION 1185
4.1. First order derivatives. If we diﬀerentiate the eigenvalue equation
(A(η) −λ
1
(η))φ
1
(; η) = 0 once with respect to η
k
, we obtain
D
k
(A(η) −λ
1
(η))φ
1
(; η) + (A(η) −λ
1
(η))D
k
φ
1
(; η) = 0. (4.10)
Taking the scalar product with φ
1
(; η), we get
¸[D
k
(A−λ
1
)]φ
1
, φ
1
) = 0, (4.11)
where we have suppressed the dependence on η for ease of writing. We will continue
with this convention in what follows provided there is no ambiguity. It follows from
(1.18) that
D
k
A(0) = iC
k
∀η ∈ Y
, (4.12)
where the operator C
k
is deﬁned in (1.19).
If we evaluate the relation (4.11) at η = 0 and use the structure of C
k
, we
immediately get that
D
k
λ
1
(0) = 0 ∀k = 1, . . . , N. (4.13)
The next step is to compute the ﬁrst order derivatives of φ
1
at η = 0. To this end,
we go back to (4.10) and use (4.13). We obtain
AD
k
φ
1
(; 0) = −D
k
A(0)φ
1
(; 0) = −iC
k
φ
1
(; 0).
Taking into account (4.9) and the above equation, we can solve for D
k
φ
1
(y; 0) and
obtain
D
k
φ
1
(y; 0) = iφ
1
(y; 0)χ
k
(y) = ip
(0)
χ
k
(y), (4.14)
where, we recall, χ
k
satisﬁes (1.13) and the constant p
(0)
was ﬁxed in Proposition 1.5.
Thus, the ﬁrst order derivative is completely determined and
D
k
φ
1
(y; 0) is purely imaginary. (4.15)
4.2. Second order derivatives. Our starting point is the relation (4.10), which
we diﬀerentiate once with respect to η. We obtain
[D
2
k
(A−λ
1
)]φ
1
+ [D
k
(A−λ
1
)]D
φ
1
+ [D
(A−λ
1
)]D
k
φ
1
+ (A−λ
1
)D
2
k
φ
1
= 0.
(4.16)
Taking the scalar product with φ
1
, we get
¸[D
2
k
(A−λ
1
)]φ
1
, φ
1
) +¸[D
k
(A−λ
1
)]D
φ
1
, φ
1
) +¸[D
(A−λ
1
)]D
k
φ
1
, φ
1
) = 0 (4.17)
for all η ∈ B
δ
. If we use the information obtained in section 4.1 on D
k
λ
1
(0), D
k
φ
1
(; 0),
D
k
A(0), and
D
2
k
A(η) = 2a
k
(y) ∀k, = 1, . . . , N, η ∈ Y
, (4.18)
1186 C. CONCA, R. ORIVE, AND M. VANNINATHAN
we obtain
1
2!
D
2
k
λ
1
(0) =
1
[Y [
_
Y
a
k
(y)dy −
1
2[Y [
_
Y
(C
k
χ
(y) +C
χ
k
(y))dy
=
1
2
(q
k
+q
k
) = q
k
∀k, = 1, . . . , N. (4.19)
As before, the next step is to compute D
2
k
φ
1
(; 0). For this purpose, we go back to
(4.16) and rewrite it with η = 0 as follows:
AD
2
k
φ
1
(; 0) =
_
−2(a
k
−q
k
) +C
k
χ
+C
χ
k
_
φ
1
(; 0).
By comparing the above equation with (1.30) and using the simplicity of the eigenvalue
under consideration, we see that D
2
k
φ
1
(; 0) is of the form
1
2!
D
2
k
φ
1
(y; 0) = −p
(0)
χ
k
(y) −β
(2)
k
p
(0)
for some constant β
(2)
k
. Thanks to (4.5) and (4.8), we can infer that
β
(2)
k
and D
2
k
φ
1
(; 0) are real. (4.20)
Moreover, β
(2)
k
admits the expression given in Proposition 1.10.
4.3. Third order derivatives. From the calculations done so far, it is now
clear how to proceed further to calculate higher order derivatives. Therefore we will
be brief here. Diﬀerentiating (4.16), we get
_
¸
¸
_
¸
¸
_
[D
3
km
(A−λ
1
)]φ
1
+ [D
2
k
(A−λ
1
)]D
m
φ
1
+ [D
2
m
(A−λ
1
)]D
k
φ
1
+ [D
2
km
(A−λ
1
)]D
φ
1
+ [D
k
(A−λ
1
)]D
2
m
φ
1
+ [D
(A−λ
1
)]D
2
km
φ
1
+ [D
m
(A−λ
1
)]D
2
k
φ
1
+ (A−λ
1
)D
3
km
φ
1
= 0.
(4.21)
Taking the scalar product with φ
1
, we get
_
¸
¸
¸
¸
_
¸
¸
¸
¸
_
¸[D
3
km
(A−λ
1
)]φ
1
, φ
1
) +¸[D
2
k
(A−λ
1
)]D
m
φ
1
, φ
1
) +¸[D
2
m
(A−λ
1
)]D
k
φ
1
, φ
1
)
+ ¸[D
2
km
(A−λ
1
)]D
φ
1
, φ
1
) +¸[D
k
(A−λ
1
)]D
2
m
φ
1
, φ
1
)
+ ¸[D
(A−λ
1
)]D
2
km
φ
1
, φ
1
) +¸[D
m
(A−λ
1
)]D
2
k
φ
1
, φ
1
) = 0.
(4.22)
To conclude that D
3
km
λ
1
(0) = 0, it is enough to use the following information in the
above relation:
_
D
k
A is purely imaginary, D
2
k
A is real, D
3
km
A = 0,
φ
1
(0), D
2
k
φ
1
(0) are real, D
k
φ
1
(0) is purely imaginary.
(4.23)
It is evident that the above argument is very general and so can be used to establish
that all odd order derivatives of λ
1
at η = 0 vanish. This proves the ﬁrst part of
Proposition 1.9.
BLOCH APPROXIMATION 1187
To ﬁnd the third order derivatives of φ
1
at η = 0, we realize that (4.21) deﬁnes
a periodic problem for D
3
km
φ
1
(; 0) which can be compared with (1.31). Further, the
relation (4.9) says that its average vanishes. These observations are enough to get the
expression of D
3
km
φ
1
(; 0) given in Proposition 1.10. We conclude by observing the
following important property:
D
3
km
φ
1
(y; 0) is purely imaginary. (4.24)
4.4. Fourth order derivatives. To arrive at the expressions for the fourth
order derivatives of λ
1
and φ
1
at η = 0 given in Propositions 1.9 and 1.10, we follow
the same arguments as in section 4.3.
5. Convergence of the ﬁrst Bloch transform to the Fourier transform.
This section is devoted to the proof of the next proposition which shows the sense in
which the Fourier transform is approximated by the ﬁrst Bloch transform.
Proposition 5.1.
(i) For every g ∈ L
2
(R
N
) with compact support, we have
χ
ε
−1
Y
(ξ)B
ε
1
g(ξ) → ´ g(ξ) in L
∞
loc
(R
N
ξ
).
(ii) If g ∈ L
2
(R
N
), we have
χ
ε
−1
Y
(ξ)B
ε
1
g(ξ) → ´ g(ξ) in L
2
(R
N
ξ
).
This will be a consequence of a more general result. In order to state it, we need
to introduce some new notations. We associate with every function ρ = ρ(y; η) deﬁned
on Y Y
which is Y periodic in y the following function:
¯ ρ
(0)
(η) =
1
[Y [
_
Y
ρ(y; η)e
−iy·η
dy, η ∈ Y
. (5.1)
With this notation, we have the following proposition.
Proposition 5.2. We suppose ρ ∈ L
∞
(Y
; L
2
#
(Y )). Then for all g ∈ W
1,p
(R
N
)
with compact support K and with p > N, we have
χ
ε
−1
Y
(ξ)
_
J
ε
g(ξ) −(2π)
N/2
˜ ρ
(0)
(εξ)´ g(ξ)
_
→ 0 in L
2
(R
N
ξ
), (5.2)
where, we recall, J
ε
g was deﬁned in (2.8).
The proof will be taken up later. Admitting it for the moment, we turn our
attention to the following proof.
Proof of Proposition 5.1. If g ∈ L
2
(R
N
) with compact support K, we have for all
ξ ∈ R
N
[χ
ε
−1
Y
(ξ)B
ε
1
g(ξ) − ´ g(ξ)[ ≤ [χ
ε
−1
Y
(ξ)(B
ε
1
g(ξ) −´ g(ξ))[ +[(χ
ε
−1
Y
(ξ) −1)´ g(ξ)[
≤ c[K[g
L
2
(R
N
)
φ
1
(; εξ) −φ
1
(; 0)
L
2
(Y )
+[(χ
ε
−1
Y
(ξ) −1)´ g(ξ)[.
If [ξ[ is bounded, then by using the fact that the map η → φ
1
(; η) ∈ L
2
#
(Y ) is
Lipschitz near η = 0, we deduce
[[φ
1
(; εξ) −φ
1
(; 0)
L
2
(Y )
≤ cε.
1188 C. CONCA, R. ORIVE, AND M. VANNINATHAN
This completes the proof of (i).
The proof of (ii) is more involved. First, according to Theorem 1.3, we have the
uniform estimate
_
ε
−1
Y
[B
ε
1
g(ξ)[
2
dξ ≤
_
R
N
[g(x)[
2
dx,
and so, by the usual density arguments, it is enough to prove (ii) with g ∈ D(R
N
).
We can now complete the proof using Proposition 5.2. Indeed, with ρ =
¯
φ
1
, we see
that
˜ ρ
(0)
(εξ) → p
(0)
and B
ε
1
g(ξ) = J
ε
g(ξ) ∀ξ ∈ R
N
,
which implies, by Lebesgue’s dominated convergence theorem, that
(2π)
N/2
χ
ε
−1
Y
(ξ)˜ ρ
(0)
(εξ)´ g(ξ) → ´ g(ξ) in L
2
(R
N
ξ
).
Proof of Proposition 5.2. The key point is that the variation of ρ(
x
ε
; εξ) with
respect to x is faster than that of g. To exploit this, we consider the εmesh ¦Y
ε
¦
∈Z
N
generated by the cell εY which was already introduced at the end of section 2. We
decompose
J
ε
g(ξ) =
∈Z
N
_
Y
ε
g(x)e
−ix·ξ
ρ
_
x
ε
; εξ
_
dx (5.3)
=
∈Z
N
g(x
ε
)
_
Y
ε
e
−ix·ξ
ρ
_
x
ε
; εξ
_
dx +r
ε
1
(ξ),
where
r
ε
1
(ξ) =
∈Z
N
_
Y
ε
(g(x) −g(x
ε
))e
−ix·ξ
ρ
_
x
ε
; εξ
_
dx. (5.4)
The ﬁrst term on the righthand side of (5.3) can be, by means of the change of
variables x = x
ε
+εy, transformed into
[Y [ε
N
F
ε
g(ξ)¯ ρ
(0)
(εξ),
where F
ε
g is the discrete Fourier transform of g and ˜ ρ
(0)
is deﬁned in (5.1). Since
we know that χ
ε
−1
Y
(ξ)ε
N
F
ε
g(ξ) →
1
(2π)
N/2
ˆ g(ξ) in L
2
(R
N
) (cf. Lemma 2.6), our
hypothesis on ρ ensures that
_
_
_χ
ε
−1
Y
(ξ)
_
[Y [ε
N
F
ε
g(ξ) −(2π)
N/2
´ g(ξ)
_
˜ ρ
(0)
(εξ)
_
_
_
L
2
(R
N
)
→ 0. (5.5)
Thus, to complete the proof, it is enough to show that
r
ε
1

L
2
(ε
−1
Y
)
≤
c(K)
(1 −
N
p
)
ερ
L
∞
(Y
;L
2
#
(Y ))
∇g
L
p
(R
N
)
. (5.6)
To this end, we rewrite r
ε
1
in a slightly diﬀerent form, namely,
r
ε
1
(ξ) =
_
R
N
¯ g
ε
1
(x)e
−ix·ξ
ρ
_
x
ε
; εξ
_
dx, (5.7)
BLOCH APPROXIMATION 1189
where
¯ g
ε
1
(x) =
∈Z
N
(g(x) −g(x
ε
))χ
Y
ε
(x). (5.8)
We already know how to estimate integrals of the type (5.7) in L
2
(R
N
) (see Lemma
2.5), and so we can deduce (5.6) provided we have the estimate
¯ g
ε
1

L
2
(R
N
)
≤
c(K)
(1 −
N
p
)
ε∇g
L
p
(R
N
)
. (5.9)
Thanks to our hypothesis, we can deduce a stronger estimate, namely,
¯ g
ε
1

L
p
(R
N
)
≤
c
(1 −
N
p
)
ε∇g
L
p
(R
N
)
, (5.10)
where c is a constant independent of K, the support of g. We note that (5.10) is a
simple consequence of Morrey’s estimate (see [7, p. 167]).
Finally, we note that (5.9) can be obtained from (5.10) with c(K) = c[K[
1−
2
p
and
a simple application of the H¨ older inequality.
The proof of Proposition 5.2 shows that the result can be strengthened by as
suming suitable smoothness on g. Our next result is an example in this direction. It
introduces naturally the following quantities:
¯ ρ
(k)
(η) =
1
[Y [
_
Y
ρ(y; η)y
k
e
−iy·η
dy ∀k = 1, . . . , N, η ∈ Y
. (5.11)
Then we have the following corrector result for J
ε
g.
Proposition 5.3. We suppose ρ ∈ L
∞
(Y
; L
2
#
(Y )). Then for all g ∈ W
2,p
(R
N
)
with compact support K and with p > N, we have
χ
ε
−1
Y
(ξ)
_
J
ε
g(ξ) −(2π)
N/2
_
˜ ρ
(0)
(εξ) +iεξ
k
˜ ρ
(k)
(εξ)
¸
´ g(ξ)
_
→ 0 in L
2
(R
N
ξ
).
Proof. We follow the idea of the proof of Proposition 5.2. We decompose J
ε
g(ξ)
as
J
ε
g(ξ) =
∈Z
N
_
Y
ε
¦g(x
ε
) +∇g(x
ε
) (x −x
ε
)¦ e
−ix·ξ
ρ
_
x
ε
; εξ
_
dx +r
ε
2
(ξ), (5.12)
where
r
ε
2
(ξ) =
∈Z
N
_
Y
ε
¦g(x) −g(x
ε
) −∇g(x
ε
) (x −x
ε
)¦ e
−ix·ξ
ρ
_
x
ε
; εξ
_
dx. (5.13)
We can estimate r
ε
2
(ξ) as follows:
r
ε
2

L
2
(ε
−1
Y
)
≤
c(K)
(2 −
N
p
)
ε
2
ρ
L
∞
(Y
;L
2
#
(Y ))
[g[
W
2,p
(R
N
)
. (5.14)
This, in fact, will be a consequence of Lemma 2.5, because we can represent r
ε
2
as
follows:
r
ε
2
(ξ) =
_
R
N
¯ g
ε
2
(x)e
−ix·ξ
ρ
_
x
ε
; εξ
_
dx (5.15)
1190 C. CONCA, R. ORIVE, AND M. VANNINATHAN
with
¯ g
ε
2
(x) =
∈Z
N
(g(x) −g(x
ε
) −∇g(x
ε
) (x −x
ε
)) χ
Y
ε
(x), (5.16)
which admits the following estimates:
¯ g
ε
2

L
2
(R
N
)
≤
c(K)
(2 −
N
p
)
ε
2
[g[
W
2,p
(R
N
)
, (5.17)
¯ g
ε
2

L
p
(R
N
)
≤
c
(2 −
N
p
)
ε
2
[g[
W
2,p
(R
N
)
. (5.18)
As before, (5.17) will be a consequence of (5.18) with c(K) = c[K[
1−
2
p
.
To establish (5.18), what we need is a generalization of Morrey’s inequality for
W
2,p
functions, namely,
[g(x)−g(x
ε
)−∇g(x
ε
) (x−x
ε
)[ ≤
c
(2−
N
p
)
[x−x
ε
[
2−
N
p
[g[
W
2,p
(Y
ε
)
∀x∈Y
ε
. (5.19)
Admitting the above estimate, it is an easy matter to prove (5.18). But the above
estimate is a consequence of Morrey’s inequality for the gradient ∇g ∈ W
1,p
(R
N
) and
the following representation: for all x ∈ Y
ε
,
g(x) −g(x
ε
) −∇g(x
ε
) (x −x
ε
) =
_
1
0
¦∇g((1 −t)x
ε
+tx) −∇g(x
ε
)¦ (x −x
ε
)dt.
This completes the proof of the estimate (5.14) on r
ε
2
. Thus, as expected, r
ε
2
tends to
zero more rapidly. The same cannot be said for the ﬁrst term on the righthand side
of (5.12). Indeed, it is equal to
[Y [
_
ε
N
(F
ε
g)(ξ)˜ ρ
(0)
(εξ) +ε
N+1
_
F
ε
∂g
∂x
k
_
(ξ)˜ ρ
(k)
(εξ)
_
. (5.20)
According to Lemma 2.6, we have the following convergence (apart from (5.5)):
χ
ε
−1
Y
_
[Y [
_
ε
N
_
F
ε
∂g
∂x
k
_
(ξ) −
1
(2π)
N/2
iξ
k
´ g(ξ)
_
˜ ρ
(k)
(εξ)
_
→ 0 in L
2
(R
N
ξ
). (5.21)
This clearly allows us to complete the proof.
6. Proof of the main convergence results. Applying the previously devel
oped techniques and results, we are now in a position to prove the main convergence
results stated in section 1.2 of the introduction (namely, Theorems 1.8 and 1.11 and
the statement (1.8)). We begin by recalling brieﬂy the setup. We take f ∈ L
2
(R
N
)
and consider a sequence u
ε
satisfying (1.9), i.e.,
_
¸
_
¸
_
A
ε
u
ε
= f in R
N
,
u
ε
u
∗
in H
1
(R
N
)weak,
u
ε
→ u
∗
in L
2
(R
N
)strong.
(6.1)
BLOCH APPROXIMATION 1191
6.1. No concentration of energy at inﬁnity. Our hypothesis that u
ε
→ u
∗
in L
2
(R
N
)strong may, at ﬁrst sight, look artiﬁcial. But this in not the case. If Ω
is bounded and smooth, then it is classical that the weak convergence in H
1
(Ω) will
automatically imply the strong convergence in L
2
(Ω). This is not the case in R
N
. To
make comparisons, the correct operator to consider is (A
ε
+ I) instead of A
ε
in R
N
.
In that case, we have the following.
Proposition 6.1. Assume that w
ε
satisﬁes
_
(A
ε
+I)w
ε
= g in R
N
,
w
ε
w
∗
in H
1
(R
N
)weak,
(6.2)
where g is a given function in L
2
(R
N
). Then
w
ε
→ w
∗
in L
2
(R
N
)strong.
Proof. First of all, following the idea of Proposition 3.1, we can neglect higher
Bloch modes of w
ε
and w
∗
. More precisely, we can show
_
ε
−1
Y
∞
m=2
[B
ε
m
w
ε
(ξ)[
2
dξ ≤ cε
4
,
_
ε
−1
Y
∞
m=2
[B
ε
m
w
∗
(ξ)[
2
dξ ≤ cε
2
.
Therefore, it remains to prove
_
ε
−1
Y
[B
ε
1
w
ε
(ξ) −B
ε
1
w
∗
(ξ)[
2
dξ → 0. (6.3)
Equation (6.2) gives the relation
(1 +λ
ε
1
(ξ))B
ε
1
w
ε
(ξ) = B
ε
1
g(ξ), ξ ∈ ε
−1
Y
.
We use it to write
χ
ε
−1
Y
(ξ)(B
ε
1
w
ε
(ξ) −B
ε
1
w
∗
(ξ)) = χ
ε
−1
Y
(ξ)
B
ε
1
g(ξ)
1 +λ
ε
1
(ξ)
−´ w
∗
(ξ) −(χ
ε
−1
Y
(ξ)B
ε
1
w
∗
(ξ) − ´ w
∗
(ξ)).
According to Proposition 5.1, the last term tends to zero in L
2
(R
N
). It suﬃces to
show
χ
ε
−1
Y
(ξ)
B
ε
1
g(ξ)
1 +λ
ε
1
(ξ)
− ´ w
∗
(ξ) → 0 in L
2
(R
N
ξ
). (6.4)
Note that w
∗
satisﬁes the homogenized equation A
∗
w
∗
+ w
∗
= g in R
N
, which is
equivalent to
_
1
2
D
2
k
λ
1
(0)ξ
k
ξ
+ 1
_
´ w
∗
(ξ) = ´ g(ξ), ξ ∈ R
N
.
1192 C. CONCA, R. ORIVE, AND M. VANNINATHAN
So, (6.4) is reduced to
χ
ε
−1
Y
(ξ)
B
ε
1
g(ξ)
1 +λ
ε
1
(ξ)
−
´ g(ξ)
1 +
1
2
D
2
k
λ
1
(0)ξ
k
ξ
→ 0 in L
2
(R
N
ξ
). (6.5)
The above expression can be written in the form
a
ε
+b
ε
c
ε
,
where
a
ε
=
_
1 +
1
2
D
2
k
λ
1
(0)ξ
k
ξ
__
χ
ε
−1
Y
(ξ)B
ε
1
g(ξ) −´ g(ξ)
_
,
b
ε
= −
_
λ
ε
1
(ξ) −
1
2
D
2
k
λ
1
(0)ξ
k
ξ
_
´ g(ξ),
c
ε
= (1 +λ
ε
1
(ξ))
_
1 +
1
2
D
2
k
λ
1
(0)ξ
k
ξ
_
.
Now we have the convergence
a
ε
c
ε
=
χ
ε
−1
Y
(ξ)B
ε
1
g(ξ) −´ g(ξ)
1 +λ
ε
1
(ξ)
→ 0 in L
2
(R
N
ξ
)
because [1 +λ
ε
1
(ξ)] ≥ 1 and by the virtue of Proposition 5.1.
The convergence of
b
ε
c
ε
is not immediate. To show this, we split the energy into
three parts, taking γ > 0 as a ﬁxed constant:
_
ξ≤δε
−1
ξ≤γ
_
b
ε
c
ε
_
2
dξ +
_
ξ≤δε
−1
ξ>γ
_
b
ε
c
ε
_
2
dξ +
_
ξ>δε
−1
_
b
ε
c
ε
_
2
dξ.
In the ﬁrst two parts, we use the estimate
¸
¸
¸
¸
λ
ε
1
(ξ) −
1
2
D
2
k
λ
1
(0)ξ
k
ξ
¸
¸
¸
¸
≤ c[ξ[
3
ε for [ξ[ ≤ δε
−1
, (6.6)
which holds since λ
1
(0) = Dλ
1
(0) = 0 (see Proposition 1.5). In the ﬁrst integral, we
have c
ε
≥ 1 and [b
ε
(ξ)[ ≤ cγ
3
ε[´ g(ξ)[, and consequently it is less than
cε
2
_
R
N
[´ g(ξ)[
2
dξ
and hence converges to zero. In the second integral, we have
c
ε
≥
1
2
λ
ε
1
(ξ)D
2
k
λ
1
(0)ξ
k
ξ
≥ c[ξ[
4
≥ cγ[ξ[
3
since [ξ[ ≥ γ > 0.
With regards to b
ε
, we still have [b
ε
(ξ)[ ≤ c[ξ[
3
ε[´ g(ξ)[, and so the second integral also
converges to zero.
In the third integral, we use the bounds
[b
ε
[ ≤
_
λ
ε
1
(ξ) +
1
2
D
2
k
λ
1
(0)ξ
k
ξ
_
[´ g(ξ)[,
c
ε
≥ λ
ε
1
(ξ) +
1
2
D
2
k
λ
1
(0)ξ
k
ξ
.
BLOCH APPROXIMATION 1193
Thus the third integral is estimated from above by
_
ξ>δε
−1
[´ g(ξ)[
2
dξ.
Obviously, this tends to zero as ε → 0 since g ∈ L
2
(R
N
).
6.2. Corrector result in R
N
. This section is devoted to the proof of Theo
rem 1.8 concerning the Bloch approximation θ
ε
. The proof consists of several steps
which correspond to estimations of the required energy in diﬀerent regions in the
Fourier space (in a neighborhood of the origin [η[ ≤ δ and in its complement [η[ > δ).
Step 1. We decompose u
ε
as follows:
u
ε
= v
ε
+P
ε
1
u
ε
,
where v
ε
and P
ε
1
u
ε
are deﬁned in (1.21) and (1.27), respectively. Thanks to Proposi
tion 3.1, it is enough to prove
P
ε
1
u
ε
−θ
ε

L
2
(R
N
)
→ 0, (6.7)
[P
ε
1
u
ε
−θ
ε
[
H
1
(R
N
)
≤ cεf
L
2
(R
N
)
. (6.8)
Step 2. We estimate the energies in the region [ξ[ > δε
−1
. To this end, we
introduce the quantities
θ
ε,δ
(x) =
_
ξ∈ε
−1
Y
ξ>δε
−1
´ u
∗
(ξ)e
ix·ξ
φ
ε
1
(x; ξ)dξ, (6.9)
P
ε,δ
1
u
ε
(x) =
_
ξ∈ε
−1
Y
ξ>δε
−1
B
ε
1
u
ε
(ξ)e
ix·ξ
φ
ε
1
(x; ξ)dξ. (6.10)
We will obtain the estimates
θ
ε,δ

L
2
(R
N
)
≤ cεf
H
−1
(R
N
)
, (6.11)
[θ
ε,δ
[
H
1
(R
N
)
≤ cεf
L
2
(R
N
)
, (6.12)
P
ε,δ
1
u
ε

L
2
(R
N
)
≤ cεf
H
−1
(R
N
)
, (6.13)
[P
ε,δ
1
u
ε
[
H
1
(R
N
)
≤ cεf
L
2
(R
N
)
. (6.14)
We start with (6.14). Using Lemma 2.3 with ρ = φ
1
and inequalities (2.5), we get
[P
ε,δ
1
u
ε
[
2
H
1
(R
N
)
≤ c
_
ξ∈ε
−1
Y
ξ>δε
−1
[B
ε
1
u
ε
(ξ)[
2
λ
ε
1
(ξ)dξ.
Now (6.14) easily follows if we use (1.22) and (1.24). Next, we prove (6.12). Following
the above procedure, we get
[θ
ε,δ
[
2
H
1
(R
N
)
≤ c
_
ξ∈ε
−1
Y
ξ>δε
−1
[´ u
∗
(ξ)[
2
[ξ[
2
dξ. (6.15)
1194 C. CONCA, R. ORIVE, AND M. VANNINATHAN
If f ∈ L
2
(R
N
), then it is well known that u
∗
∈ H
2
(R
N
) and
_
R
N
[ξ[
4
[´ u
∗
(ξ)[
2
dξ ≤ c
_
R
N
[
´
f(ξ)[
2
dξ. (6.16)
Combining (6.15) and (6.16), we easily get (6.12). We now show (6.11). By Parseval’s
identity, we have
θ
ε,δ

2
L
2
(R
N
)
=
_
ξ∈ε
−1
Y
ξ>δε
−1
[´ u
∗
(ξ)[
2
dξ ≤ c
δ
ε
2
_
ξ∈ε
−1
Y
ξ>δε
−1
[ξ[
−2
[
´
f(ξ)[
2
dξ,
since u
∗
and f are related by the homogenized equation A
∗
u
∗
= f in R
N
. This clearly
implies
θ
ε,δ

2
L
2
(R
N
)
≤ c
δ
ε
2
_
ξ∈ε
−1
Y
ξ>δε
−1
(1 +[ξ[
2
)
−1
[
´
f(ξ)[
2
dξ = c
δ
ε
2
f
2
H
−1
(R
N
)
.
The proof of (6.13) is completely analogous.
Step 3. Now, we consider the energies in [ξ[ ≤ δε
−1
. To this end, let us deﬁne
ω
ε
(x) =
_
ξ≤δε
−1
(B
ε
1
u
ε
(ξ) − ´ u
∗
(ξ))e
ix·ξ
φ
ε
1
(x; ξ)dξ (6.17)
and show that
ω
ε

L
2
(R
N
)
→ 0, (6.18)
[ω
ε
[
H
1
(R
N
)
≤ cεf
L
2
(R
N
)
. (6.19)
To prove (6.18), we decompose the integrand as follows:
B
ε
1
u
ε
− ´ u
∗
= B
ε
1
(u
ε
−u
∗
) + (B
ε
1
u
∗
− ´ u
∗
).
By Parseval’s equality, the ﬁrst term in the L
2
norm is bounded above by u
ε
−
u
∗

L
2
(R
N
)
which, by our hypothesis, converges to zero. That the second term con
verges to zero in L
2
(R
N
) is proved in Proposition 5.1.
Next, we turn are attention to the proof of (6.19). By Lemma 2.1, we have
[ω
ε
[
2
H
1
(R
N
)
≤ c
_
ξ≤δε
−1
λ
ε
1
(ξ)[B
ε
1
u
ε
(ξ) − ´ u
∗
(ξ)[
2
dξ. (6.20)
To estimate the above integral, we write the integrand as
B
ε
1
u
ε
(ξ) − ´ u
∗
(ξ) = λ
ε
1
(ξ)
−1
(B
ε
1
f(ξ) −
´
f(ξ)) +
_
λ
ε
1
(ξ)
−1
−
_
1
2
D
2
k
λ
1
(0)ξ
k
ξ
_
−1
_
´
f(ξ).
Thus we get, using (1.24), that
_
¸
¸
¸
¸
_
¸
¸
¸
¸
_
[ω
ε
[
2
H
1
(R
N
)
≤ c
_
ξ≤δε
−1
[B
ε
1
f(ξ) −
´
f(ξ)[
2
[ξ[
2
dξ
+ c
_
ξ≤δε
−1
λ
ε
1
(ξ)
¸
¸
¸λ
ε
1
(ξ)
−1
−
_
1
2
D
2
k
λ
1
(0)ξ
k
ξ
_
−1
¸
¸
¸
2
[
´
f(ξ)[
2
dξ.
(6.21)
BLOCH APPROXIMATION 1195
To estimate the ﬁrst term on the righthand side of (6.21), we represent the integrand
as
B
ε
1
f(ξ) −
´
f(ξ)
[ξ[
=
_
R
N
f(x)e
−ix·ξ
(φ
ε
1
(x; ξ) −φ
ε
1
(x; 0))
[ξ[
dx.
Applying Lemma 2.5 and using φ
1
(; η) −φ
1
(; 0)
L
2
(Y )
≤ c[η[ for [η[ ≤ δ, we get
_
ξ≤δε
−1
[B
ε
1
f(ξ) −
´
f(ξ)[
2
[ξ[
2
dξ ≤ cε
2
f
2
L
2
(R
N
)
.
The second term on the righthand side of (6.21) can be rewritten, using the homog
enized equation, as
_
ξ≤δε
−1
[λ
ε
1
(ξ) −
1
2
D
2
k
λ
1
(0)ξ
k
ξ
[
2
λ
ε
1
(ξ)
[´ u
∗
(ξ)[
2
dξ.
Using (6.6) and (1.24), we see that the above integral is estimated from above by
cε
2
_
ξ≤δε
−1
[ξ[
4
[´ u
∗
(ξ)[
2
dξ ≤ cε
2
f
2
L
2
(R
N
)
.
This establishes (6.19) and hence the result.
6.3. Asymptotic expansion of the Bloch approximation. In this conclud
ing section, we prove Theorem 1.11.
Proof of (i). We have the following decomposition:
θ
ε
(x) −u
∗
(x) = z
ε
(x) +θ
ε,δ
(x) +u
∗,δ
(x), (6.22)
where
z
ε
(x) =
_
ξ∈ε
−1
Y
ξ≤δε
−1
´ u
∗
(ξ)e
ix·ξ
(φ
ε
1
(x; ξ) −φ
ε
1
(x; 0))dξ, (6.23)
u
∗,δ
(x) =
1
(2π)
N/2
_
ξ>δε
−1
´ u
∗
(ξ)e
ix·ξ
dξ, (6.24)
and θ
ε,δ
is deﬁned in (6.9).
The second term has already been estimated in the L
2
norm (see (6.11)). The
same proof shows that the third term admits a bound
u
∗,δ

L
2
(R
N
)
≤ cεf
H
−1
(R
N
)
≤ cεu
∗

H
1
(R
N
)
. (6.25)
To estimate the ﬁrst term on the righthand side of (6.22), we must proceed
diﬀerently. In fact, it is essential to use Lemma 2.3. We see then that
z
ε

2
L
2
(R
N
)
=
_
ξ∈ε
−1
Y
ξ≤δε
−1
[´ u
∗
(ξ)[
2
φ
ε
1
(; ξ) −φ
ε
1
(; 0)
2
L
2
(Y )
dξ.
Using the Lipschitz continuity of the map η → φ
1
(; η) ∈ L
2
(Y ) for [η[ ≤ δ, we see
that the above integral can be majorized, and we obtain
z
ε

2
L
2
(R
N
)
≤ cε
2
_
ξ≤δε
−1
[´ u
∗
(ξ)[
2
[ξ[
2
dξ ≤ cε
2
[u
∗
[
2
H
1
(R
N
)
. (6.26)
1196 C. CONCA, R. ORIVE, AND M. VANNINATHAN
This ﬁnishes the proof of (i). We note that we cannot, in general, assert that
[u
∗
[
H
1
(R
N
)
≤ cf
H
−1
(R
N
)
as we are working on the entire space R
N
.
Proof of (ii). Because of (i), it suﬃces to prove
¸
¸
¸θ
ε
−u
∗
−εχ
ε
k
∂u
∗
∂x
k
¸
¸
¸
H
1
(R
N
)
≤ cεf
L
2
(R
N
)
. (6.27)
To this end, we use once again the decomposition (6.22) for (θ
ε
−u
∗
) in terms of z
ε
,
θ
ε,δ
, and u
∗,δ
. For θ
ε,δ
, we have the estimate (6.12). For u
∗,δ
, we can easily derive
the estimate
[u
∗,δ
[
2
H
1
(R
N
)
≤ c
_
ξ>δε
−1
[ξ[
2
[´ u
∗
(ξ)[
2
dξ ≤ c
δ
ε
2
f
2
L
2
(R
N
)
. (6.28)
Thus, we are reduced to obtaining the estimate
¸
¸
¸z
ε
−εχ
ε
k
∂u
∗
∂x
k
¸
¸
¸
H
1
(R
N
)
≤ cεf
L
2
(R
N
)
. (6.29)
To this end, we use the representation
∂u
∗
∂x
k
(x) =
1
(2π)
N/2
_
R
N
(iξ
k
)´ u
∗
(ξ)e
ix·ξ
dξ,
and combine it with the representation (6.23) for z
ε
. We get
z
ε
(x) −εχ
ε
k
(x)
∂u
∗
∂x
k
(x) =
_
ξ≤δε
−1
´ u
∗
(ξ)e
ix·ξ
_
φ
ε
1
(x; ξ) −φ
ε
1
(x; 0) −ip
(0)
χ
ε
k
(x)εξ
k
_
dξ
−
_
ξ>δε
−1
ip
(0)
χ
ε
k
(x)εξ
k
´ u
∗
(ξ)e
ix·ξ
dξ. (6.30)
To estimate the ﬁrst term on the righthand side of (6.30), we appeal to Lemma 2.3.
Further, we use
_
_
_φ
1
(; η) −φ
1
(; 0) −ip
(0)
χ
k
()η
k
_
_
_
H
1
(Y )
≤ c[η[
2
for [η[ ≤ δ. (6.31)
The estimate on the second term on the righthand side of (6.30) is more straightfor
ward. We ﬁnally get
¸
¸
¸
¸
z
ε
−εχ
ε
k
∂u
∗
∂x
k
¸
¸
¸
¸
2
H
1
(R
N
)
≤ cε
2
_
R
N
[ξ[
4
[´ u
∗
(ξ)[
2
dξ.
This completes the proof of (6.29) and hence (ii).
Proof of (iii). Consider again the decomposition (6.22). Thanks to (6.9) and
(6.15), we have the estimates
θ
ε,δ

L
2
(R
N
)
≤ cε
2
f
L
2
(R
N
)
, (6.32)
[θ
ε,δ
[
H
1
(R
N
)
≤ cε
2
[f[
H
1
(R
N
)
. (6.33)
BLOCH APPROXIMATION 1197
Similar techniques imply
u
∗,δ

L
2
(R
N
)
≤ cε
2
f
L
2
(R
N
)
, (6.34)
[u
∗,δ
[
H
1
(R
N
)
≤ cε
2
[f[
H
1
(R
N
)
. (6.35)
On the other hand, it is clear from the representation (6.30) that
_
_
_z
ε
−εχ
ε
k
∂u
∗
∂x
k
_
_
_
L
2
(R
N
)
≤ cε
2
f
L
2
(R
N
)
. (6.36)
Thus, it is enough to obtain the estimate
¸
¸
¸
¸
θ
ε
−u
∗
−εχ
ε
k
∂u
∗
∂x
k
+ε
2
(χ
ε
k
+β
(2)
k
)
∂
2
u
∗
∂x
k
∂x
¸
¸
¸
¸
H
1
(R
N
)
≤ cε
2
[f[
H
1
(R
N
)
. (6.37)
Thanks to (6.33) and (6.35), we are reduced to showing that
¸
¸
¸
¸
z
ε
−εχ
ε
k
∂u
∗
∂x
k
+ε
2
(χ
ε
k
+β
(2)
k
)
∂
2
u
∗
∂x
k
∂x
¸
¸
¸
¸
H
1
(R
N
)
≤ cε
2
[f[
H
1
(R
N
)
. (6.38)
We can write
z
ε
(x) −εχ
ε
k
(x)
∂u
∗
∂x
k
(x) + ε
2
(χ
ε
k
(x) +β
(2)
k
)
∂
2
u
∗
∂x
k
∂x
(x)
=
_
ξ≤δε
−1
´ u
∗
(ξ)e
ix·ξ
[φ
ε
1
(x; ξ) −φ
ε
1
(x; 0) −ip
(0)
χ
ε
k
(x)εξ
k
i
+ p
(0)
(χ
ε
k
(x) +β
(2)
k
)ε
2
ξ
k
ξ
]dξ
−
_
ξ>δε
−1
ip
(0)
χ
ε
k
(x)εξ
k
´ u
∗
(ξ)e
ix·ξ
dξ
+
_
ξ>δε
−1
p
(0)
(χ
ε
k
(x) +β
(2)
k
)ε
2
ξ
k
ξ
´ u
∗
(ξ)e
ix·ξ
dξ. (6.39)
The analysis of the righthand side of (6.39) is similar to that of (6.30). The new
information needed is the following:
_
_
_φ
1
(; η) −φ
1
(; 0) −ip
(0)
χ
k
()η
k
+p
(0)
(χ
k
() +β
(2)
k
)η
k
η
_
_
_
H
1
(Y )
≤ c[η[
3
(6.40)
for [η[ ≤ δ, which is a simple consequence of Proposition 1.10. The proof is concluded
via a simple application of Lemma 2.3.
Acknowledgments. This research stems from a couple of visits by R. Orive and
M. Vanninathan to the Center of Mathematical Modelling and to the Department of
Mathematical Engineering of the University of Chile, supported by the FONDAP’s
Programme on MathematicalMechanics. These authors warmly thank these institu
tions.
1198 C. CONCA, R. ORIVE, AND M. VANNINATHAN
REFERENCES
[1] G. Allaire, Homogenization and twoscale convergence, SIAM J. Math. Anal., 23 (1992),
pp. 1482–1518.
[2] G. Allaire and C. Conca, Bloch wave homogenization for a spectral problem in ﬂuidsolid
structures, Arch. Ration. Mech. Anal., 135 (1996), pp. 197–257.
[3] G. Allaire and C. Conca, Boundary layers in the homogenization of a spectral problem in
ﬂuidsolid structures, SIAM J. Math. Anal., 29 (1998), pp. 343–379.
[4] M. Avellaneda, L. Berlyand, and J.F. Clouet, Frequencydependent acoustics of compos
ites with interfaces, SIAM J. Appl. Math., 60 (2000), pp. 2143–2181.
[5] A. Bensoussan, J.L. Lions, and G. Papanicolaou, Asymptotic Analysis for Periodic Struc
tures, NorthHolland, Amsterdam, 1978.
[6] F. Bloch,
¨
Uber die quantenmechanik der electronen in kristallgittern, Z. Phys., 52 (1928),
pp. 555–600.
[7] H. Brezis, Analyse Fonctionnelle, Th´eorie et Applications, Collection Math´ematiques Ap
pliqu´ees pour la Maˆıtrise, Masson, Paris, 1983.
[8] C. Conca, Bloch waves, in Encyclopaedia Mathematics, M. Hazewinkel et al., eds., Kluwer
Academic Publishers, Amsterdam, 1999, pp. 72–74.
[9] C. Conca, R. Orive, and M. Vanninathan, Bloch Approximation in Bounded Domains,
manuscript, 2001.
[10] C. Conca, J. Planchard, and M. Vanninathan, Fluids and Periodic Structures, Rech. Math.
Appl., 38, John Wiley/Masson, New York, Paris, 1995.
[11] C. Conca and M. Vanninathan, Homogenization of periodic structures via Bloch decomposi
tion, SIAM J. Appl. Math., 57 (1997), pp. 1639–1659.
[12] C. Conca and M. Vanninathan, On uniform H
2
estimates in periodic homogenization, Proc.
Roy. Soc. Edinburgh Sect. A, 131 (2001), pp. 499–517.
[13] G. Floquet, Sur les ´equations diﬀ´erentielles lin´eaires ` a coeﬃcients p´eriodiques, Ann. Ecole
Norm. S´er. 2, 12 (1883), pp. 47–89.
[14] P. G´ erard, P.A. Markowich, N.J. Mauser, and F. Poupaud, Homogenization limits and
Wigner transforms, Comm. Pure Appl. Math., 50 (1997), pp. 323–379.
[15] T. Kato, Perturbation Theory for Linear Operators, SpringerVerlag, Berlin, 1966.
[16] F. Murat and L. Tartar, Hconvergence, in Topics in the Mathematical Modeling of Compos
ite Materials, Progr. Nonlinear Diﬀerential Equations Appl. 31, A. Cherkaev and R. Kohn,
eds., Birkh¨auser, Boston, 1997, pp. 21–44.
[17] G. Nguetseng, A general convergence result for a functional related to the theory of homoge
nization, SIAM J. Math. Anal., 20 (1989), pp. 608–623.
[18] F. Odeh and J. Keller, Partial diﬀerential equations with periodic coeﬃcients and Bloch
waves in crystals, J. Math. Phys., 5 (1964), pp. 1499–1504.
[19] M. Reed and B. Simon, Methods of Modern Mathematical Physics, Academic Press, New
York, 1978.
[20] F. Santosa and W. W. Symes, A dispersive eﬀective medium for wave propagation in periodic
composites, SIAM J. Appl. Math., 51 (1991), pp. 984–1005.
[21] L. Tartar, Probl`emes d’homog´en´eisation dans les ´equations aux d´eriv´ees partielles, in Cours
Peccot au Coll`ege de France, 1977.
[22] C. Wilcox, Theory of Bloch waves, J. Anal. Math., 33 (1978), pp. 146–167.
BLOCH APPROXIMATION
1167
Historically, a classical way of obtaining such correctors is to work in the physical space and use multiscale expansion of the solution, which was ﬁrst introduced in the basic book just cited. As we will see, the method of Bloch waves sheds new light and oﬀers an alternative for viewing the classical results. This method naturally leads us to work in the Fourier space, and thus in a framework dual to the one used in L. Tartar [21], whose method is very general and justiﬁes, in particular, the ﬁrst term obtained via the multiscale expansion. (There are also other methods of justiﬁcation based on the analysis in physical space; cf. G. Nguetseng [17] and G. Allaire [1].) However, it is important to mention that the Bloch wave method does not presuppose any multiscale structure of the solution; on the contrary, such a structure of the solution will be a consequence of the present method. Although correctors are generally not unique, our approach yields a posteriori the same ones as those obtained in [5]. Bloch waves and their applications are not by any means new. It is a classical tool which has been in use in solid state physics since the pioneering paper of F. Bloch [6]. However, the basic idea was introduced in the mathematical literature much earlier by G. Floquet [13]. For later developments, let us cite the works of F. Odeh and J. Keller [18] and M. Reed and B. Simon [19]. A deep analysis concerning the partial regularity of the spectrum of Schr¨dinger’s equation with periodic potential was o carried out by C. Wilcox [22]. Our point of view regards periodic medium as a perturbation of homogeneous ones. In this context, the book by T. Kato [15] provides excellent analysis when the parameter of perturbation is a scalar. We end this rather incomplete list by citing G. Allaire and C. Conca [2], [3], and P. G´rard et al. [14]. e We feel it is also appropriate to cite a recent work by M. Avellaneda, L. Berlyand, and J.F. Clouet [4], in which the Bloch–Floquet approach is used to provide new homogenization results and handles the boundary layer terms for frequency dependent problems. To conclude, let us refer the reader to C. Conca [8] for a more complete general survey on Bloch waves. Before proceeding further, we mention a word about the notations adopted in what follows. Unless mentioned explicitly, the usual summation convention with respect to the repeated indices is understood. The constants appearing in various estimates independent of ε are generically denoted by c, c1 , c2 , etc. Apart from the usual norms in Sobolev spaces H 1 , H 2 , we will also use the following seminorms: vH 1 =
N
Dj v
j=1
2 L2
1 2
,
vH 2 =
N
2 Dj,k v
j,k=1
2 L2
1 2
.
Now let us introduce the problem to be studied in this work. We consider the operator (1.1) A =−
def
∂ ∂yk
ak (y)
∂ ∂y
,
y ∈ RN ,
where the coeﬃcients satisfy ∞ N ak ∈ L# (Y ), where Y =]0, 2π[ , i.e., each ak is a Y periodic bounded measurable function deﬁned on RN , and (1.2) ∃α > 0 such that ak (y)ηk η ≥ αη2 ∀η ∈ RN , y ∈ Y a.e., ak = a k ∀k, = 1, . . . , N.
1168
C. CONCA, R. ORIVE, AND M. VANNINATHAN
For each ε > 0, we consider also the operator Aε , where (1.3) Aε = −
def
∂ ∂xk
aε (x) k
∂ ∂x
with
aε (x) = ak k
x , ε
x ∈ RN .
In homogenization theory, it is usual to refer to x and y, the slow and the fast variables, respectively. They are related by y = x . Associated with Aε , let us consider the ε boundary value problem (1.4) Aε uε = f in Ω,
1 uε ∈ H0 (Ω),
which is posed in an arbitrary bounded domain Ω in RN and where f is a given element in L2 (Ω). It is classical that the above problem admits one and only one solution. From the classical work [5], it is known that one can associate to Aε a homogenized operator A∗ given by (1.5) A∗ = −
def
∂ ∂xk
qk
∂ ∂x
.
The homogenized coeﬃcients qk are constants and their deﬁnition is given below. 1 The solution uε of (1.4) converges weakly in H0 (Ω) to the socalled homogenized ∗ solution u characterized by (1.6) A∗ u ∗ = f in Ω,
1 u∗ ∈ H0 (Ω).
In the present paper, we do not consider the eﬀects of boundaries, postponing them to a subsequent article [9]. In the case of RN , it is natural to replace the operator Aε by (Aε + I). In that case, if wε satisﬁes (1.7) (Aε + I)wε = g w
ε
in RN , in H 1 (RN )weak,
w
∗
where g is a given function in L2 (RN ), then it can be seen that (see Proposition 6.1 below) (1.8) wε → w∗ in L2 (RN )strong.
In view of the above result, there is no concentration of L2 energy at inﬁnity, and therefore throughout this paper we will consider a sequence uε and a function f ∈ L2 (RN ) satisfying ε ε in RN , A u =f uε u∗ in H 1 (RN )weak, (1.9) ε ∗ u →u in L2 (RN )strong. The central issue in the analysis of the ﬁrst order correctors is to obtain functions uε ∈ H 1 (RN ), which can be easily constructed and have the following characteristic 1 property: (1.10) uε − u∗ − εuε 1
H 1 (RN )
→0
as
ε → 0.
. . M (χ ) = 0.) Using the above expansion. u2 (x). which contains all the above correctors and justiﬁes the procedure. the classical formula of the homogenized coeﬃcients is as follows: q k = MY ak + akm ∂χ ∂ym ∀k. N. = 1. 1.BLOCH APPROXIMATION 1169 By deﬁnition. . the ﬁrst order corrector term is obtained in [5]. u1 (x). N. Survey of the previous results.12) + ε2 χk (y) Here. g1 (x) = bjk Djk u∗ . Apart from this. and the solution χk ∈ W 1. (Another characterization of qk is given in Proposition 1. θε contains a lot of information about the periodic medium which will be amply demonstrated in this paper.1. . ∂xk x ∂x (1. for instance.11) uε − u∗ − εuε − ε2 uε 1 2 H 1 (RN ) ≤ cε. . namely.∞ (Y ). Y # k k akm ∂χ ∂ym in RN . . f ∈ L2 (RN ). . . where bjk are constants: bjk = MY ajm ∂χk + ak χj ∂ym ∀j.1. Further. k ∂y χ ∈ H 1 (Y ). χk is the unique solution of the cell problem Aχ = ∂ak in RN . Bloch approximation θε . are independent of ε and satisfy equations of the type A∗ uj = 3 gj in RN . . = 1. . we have the following.13) Y χk dy = 0. More precisely. We assume that the coeﬃcients ak satisfy assumptions (1. N .5 below. One of the purposes of this article is to carry out a more general construction than the classical one for correctors. where. Theorem 1. . Then the ﬁrst order corrector is deﬁned by uε (x) = χk 1 x ε ∂u∗ (x). M (χ ) def 1 = Y # k k Y  (1. With these notations. In the classical book [5] the authors obtain an asymptotic expansion (with y = x ) of the form ε uε (x) = u∗ (x) + ε χk (y) ∂u∗ (x) + u1 (x) ∂xk ∂ 2 u∗ ∂u1 (x) + χ (y) (x) + u2 (x) + · · · . . . . second order correctors uε ∈ H 1 (RN ) will enjoy the property 2 (1. The function χk is characterized as the unique solution of Aχ = a +a ∂χ − ∂ (a χ )−M (a )−M k km mk Y k Y k ∂ym ∂ym (1. k.2).14) χ ∈ H 1 (Y ). ∂xk . . k = 1.
η) = λ(η)ψ(·. η) (not identically zero) such that (1. η) ψ(y + 2πm.1170 which means that C. we deﬁne φ(y. the operator A(η) is deﬁned by (1. Y ) 1 and H# (Y ). η) = e in RN . 2 [N . Y )periodic and loc Y periodic. J.15) can be rewritten in terms of φ as follows: (1. these spaces are denoted by L2 (η. where 1 H# (η. η) ∀m ∈ ZN .. . Let us consider the following spectral problem parameterized by η ∈ RN : ﬁnd λ = λ(η) ∈ R and ψ = ψ(y. N # ∂yk ∂φ ∈ L2 (Y ) ∀k = 1. Y ) and L2 (Y ). Besides. The basic tool of this new approach is Bloch waves ψ associated with A which we deﬁne now. Van2 ninathan [10]) that for each η ∈ Y . η) and φm (·.18) with (1. and η can therefore be conﬁned to the dual 1 cell η ∈ Y = [− 1 . we obtain a more general result using a diﬀerent approach introduced in [11]. η). . Planchard.15) Aψ(·. . It is # # 1 worthwhile to remark that these eigenfunctions in fact belong to the spaces H# (η. Conca. N # ∂yk . ORIVE. CONCA. φ is Y periodic. R. η) form orthonormal bases in the spaces of all L2 (RN )functions which are (η. Y ) = ψ ∈ L2 (η. respectively. In this paper. VANNINATHAN uε − u∗ − εuε 1 H 1 (RN ) →0 as ε → 0.16) A(η)φ = λφ in RN .17) A(η) = − def ∂ + iηk ∂yk ak (y) ∂ + iη ∂y . Here. ∂yj ∂yj It is clear from (1. . Y ) ∀k = 1. and (1. and M.19) Ck φ = − akj (y) def A(η) = A + iηk Ck + ηk η ak (y) ∂φ ∂ − (akj (y)φ).15) that the (η. AND M. It is well known (C. Next. . . η) is (η. λm (η) is a Lipschitz function of η ∈ Y . ψ(·. Y )periodic. 2πim·η ψ(y. 1 H# (Y ) = . i. . Y ) # φ ∈ L2 (Y ) # ∂ψ ∈ L2 (η. which can be rewritten as (1. y ∈ RN . .e. Y ) periodicity condition is unaltered if we replace η by (η + q) with q ∈ ZN . . respectively. η) = e−iy·η ψ(y. the above spectral problem admits a discrete sequence of eigenvalues with the following properties: 0 ≤ λ1 (η) ≤ · · · ≤ λm (η) ≤ · · · → ∞ ∀m ≥ 1. the corresponding eigenfunctions denoted by ψm (·.
2. η) (referred to as Bloch waves) introduced above enable us to describe the spectral resolution of A (an unbounded selfadjoint operator in L2 (RN )) in the orthogonal basis {eiy·η φm (y. ψm (·.BLOCH APPROXIMATION 1171 The functions ψm (·. Theorem 1.2. ξ)dξ. The mth Bloch coeﬃcient of g is deﬁned as follows: (Bm g)(η) = RN ¯ g(y)e−iy·η φm (y. Then the following inverse formula holds: ∞ g(y) = Y m=1 (Bm g)(η)eiy·η φm (y. With these notations. for all g in the domain of Aε . η)m ≥ 1. m ∞ ε−1 Y m=1 ε (Bm g)(ξ)2 dξ. Observe that ε φε (x. ξ) = ψm (y. where the variables (x. η). ξ) is εY periodic (in x) and ε−1 Y periodic with respect to ξ. η ∈ Y . we have the following result analogous to Theorem 1. Then the following inverse formula and Parseval’s identity hold: ∞ g(x) = ε−1 Y m=1 ε (Bm g)(ξ)eix·ξ φε (x. η). we get Aε g(x) = ∞ ε−1 Y m=1 ε λε (ξ)(Bm g)(ξ)eix·ξ φε (x. m −1 Note that the dual cell at εscale is ε Y and hence we take ξ to vary in ε−1 Y in what follows. ξ) = φm (y. In the same m ε ε manner. RN g(x)2 dx = Finally. η) and φm (·. m ε ψm (x. η)dη. Theorem 1. ξ). ξ) and (y. ξ) = eix·ξ φε (x. η)dy ∀m ≥ 1. we have ∞ Ag(y) = Y m=1 λm (η)(Bm g)(η)eiy·η φm (y. ξ ∈ ε−1 Y . εY )periodic because of the relation ψm (x. ξ)dξ. To obtain the spectral resolution of Aε in an analogous manner. η)dη. Further. ξ)dx m ∀m ≥ 1. we have Parseval’s identity: g(y) dy = 2 ∞ Y m=1 RN (Bm g)(η)2 dη. Finally. Let g ∈ L2 (RN ). ξ) is (εξ. η ∈ Y }. m φε (x. η) are related by y = x and η = εξ. The mth Bloch coeﬃcient of g at the εscale is deﬁned as follows: ε (Bm g)(ξ) = RN ¯ g(x)e−ix·ξ φε (x. we have the following. More precisely. for all g in the domain of A. Let g ∈ L2 (RN ). m m . let us introduce Bloch waves at the εscale: λε (ξ) = ε−2 λm (η).3.
∀m ≥ 2. def φ1 (y. η) satisfying 1 η → φ1 (·. ξ)dξ. . Thus we can concentrate our attention only on the relation corresponding to the ﬁrst Bloch wave: (1.20) ε ε λε (ξ)(Bm uε )(ξ) = (Bm f )(ξ) ∀m ≥ 1.4. . . one can neglect all the relations for m ≥ 2. Proposition 1. η) ∈ H# (Y ) is analytic on Bδ .25) (N ) cη2 ≤λ1 (η)≤ cη2 0 < λ2 (N ) ∀η ∈ Y .23) qk ξk ξ u∗(ξ) = f (ξ) ∀ξ ∈ RN stands is obtained from (1. = 1. .24) (1. ξ ∈ ε−1 Y .9). We can express the equation Aε uε = f in RN in the equivalent form (1. the classical homogenization result was deduced in [11]. CONCA. The ﬁrst one consists of considering a sequence uε ∈ H 1 (RN ) satisfying (1. More precisely. 1 2 1 ∂ 2 λ1 Dk λ1 (0) = (0) = qk 2 2 ∂ηk ∂η and there exist constants c and c such that (1. we have the relations λ1 (0) = 0. . (2π)N/2 ∀k = 1. To this end. N.5. Here. 0) = p(0) (= Y −1/2 = Moreover. AND M. Dk λ1 (0) = ∂λ1 (0) = 0 ∂ηk 1 ). R. the symbol for the classical Fourier transformation f (ξ) = 1 (2π)N/2 RN f (x)e−ix·ξ dx. VANNINATHAN Using the above theorem. the following results were established and applied in [11]. . and there is a choice of the ﬁrst eigenvector φ1 (y. it is proved in [11] that the following result holds. ≤ λm (η) where λ2 is the second eigenvalue of the spectral problem for A in the cell Y with Neumann boundary conditions on ∂Y .22) ε ε λε (ξ)(B1 uε )(ξ) = (B1 f )(ξ) ∀ξ ∈ ε−1 Y . η ∈ Y . 1 The homogenized equation in the Fourier space (1. We assume that ak satisﬁes (1.1172 C. N. m In the homogenization process. ORIVE.22) by passing to the limit as ε → 0. . Proposition 1.21) v ε (x) = ∞ ε−1 Y m=2 ε (Bm uε )(ξ)eix·ξ φε (x. . Let (1. Let us recall the main steps. Then there exists δ > 0 such that the ﬁrst eigenvalue λ1 (η) is an analytic function on Bδ = {η  η < δ}. ∀k. . m Then v ε L2 (RN ) ≤ cε.2).
then χε−1 Y B1 g ε → g in L2 (RN ). the Bloch approximation ε ε ε θ is close to P1 u and hence to uε . .BLOCH APPROXIMATION 1173 Apart from the above result of regularity on the Bloch spectrum. m ≥ 2.7. We note by the item (ii) of Proposition 1. we deﬁne (1. we need to prove that the ﬁrst Bloch transform is an approximation to the Fourier transform. ξ) → (2π)−N/2 . Let g ε and g be in L2 (RN ). . .29) uε − θε H 1 (RN ) ≤ cε f L2 (RN ) . For H 2 estimates. Let uε be the sequence introduced in (1. heuristically speaking.) Thus we are motivated to introduce the projection onto the ﬁrst Bloch mode: for all g ∈ L2 (RN ). Presentation of new results: The Bloch approximation. . as ε → 0. . We will see in section 3 that higher Bloch modes 1 φε . . First of all. Assume that the coeﬃcients ak satisfy (1. do not contribute at all in the analysis of the correctors of ﬁrst and m second order in the energy norm. u∗ satisﬁes A∗ u∗ = f in RN . Then ε g weakly in L2 (RN ).26) θε (x) = def ε−1 Y u∗ (ξ)eix·ξ φε (x.6. etc.9). Let us consider the sequence uε satisfying hypotheses (1. Then if f ∈ L2 (RN ). x loc ξ These results easily lead us to the following homogenization theorem in RN .6 that the Fourier transform u∗ is an ε approximation of B1 uε .2. We consider a sequence uε satisfying (1. This result is naturally expected from the fact that φε (x. Therefore. 1. . we have (1. Theorem 1. it is an easy matter to deduce the corresponding result in a bounded domain Ω by localization techniques using a cutoﬀ function φ ∈ D(Ω) (see [11]).28) (uε − θε ) → 0 in H 1 (RN ).9). ξ)dξ.2). we will prove the following theorem. Theorem 1. Then aε k ∂uε ∂x qk ∂u∗ ∂x in L2 (RN ) ∀k = 1.8. With these notations. Once the homogenization result in RN is established. . let us remark that this object is not diﬃcult to be computed in principle. Our goal throughout this paper is to study properties of this function and particularly its relations with various correctors terms. . we refer to our work in [12]. then χε−1 Y B1 g ε g weakly in L2 (RN ) pro(i) if g ε x loc ξ vided there is a ﬁxed compact set K such that supp (g ε ) ⊂ K ∀ε. ε (ii) if g ε → g in L2 (RN ). we have the estimate (1. 1 x ∈ RN .27) ε P1 g(x) = ε B1 g(ξ)eix·ξ φε (x. Proposition 1. The Bloch approximation of uε is deﬁned by the following formula: (1. Furthermore.9). N. 1 ∀ξ ∈ RN . It is worth noticing that θε is deﬁned only in terms of the ﬁrst Bloch mode φε . In particular. ξ)dξ. 1 ε−1 Y x ∈ RN . (It will be interesting to know whether these higher order modes play a part in the analysis of correctors in stronger norms H 2 .
we are reduced to expanding θε in terms of ε in order to be able to compare it with the classical correctors for uε . we need an asymptotic expansion of the ﬁrst Bloch transform B1 g(ξ) for which we refer the reader to section 5. Thanks to the above result. ORIVE. CONCA. To fulﬁll this task.14)): Aχ = (ak − qk ) − 1 Ck χ + C χ in RN .31) − C χmk − Cm χk m MY (χk 1 4 ) = 0. they are usually obtained using the maximum principle with more regularity hypotheses on ak and f .30) χ ∈ H 1 (Y ). . Here. the fourth order derivatives have the following expressions: for all k. m m = m −q m )χk + (amk − qmk )χ in RN . These results strengthen earlier results. particularly those of Proposition 1. .6. k # k Aχ k χ k Aχ k χ 1 (ak − qk )χm + (a 3 −Ck χ m 1 ∈ H# (Y ). β odd. .32) 1 (ak − qk )χmn + (a 3! n −q m )χkn + (akm − qkm )χ m + (amn − qmn )χk + (a 1 ∈ H# (Y ). All odd order derivatives of λ1 at η = 0 vanish. χk mn deﬁned by the following cell problems (observe that the ﬁrst ones are nothing but the functions already introduced in (1. AND M. MY (χ ) = 0. For instance. . + All even order derivatives of λ1 at η = 0 can be calculated in a systematic fashion. mn = + 1 4 D 4! k mn λ1 (0) − Cn χk m m + Ck χ mn + C χmnk + Cm χnk n (1. R.9. we introduce some test functions χk . .) We now state results in this direction. N 1 4 D 4! k mn λ1 (0) = 1 1 4 Y  Y Cn χk m + Ck χ mn + C χmnk + Cm χnk dy .e. First. m. for our 1 1 ε purposes below. − q n )χkm + (akn − qkn )χ in RN . ξ)..29) are sometimes found in the literature. Dβ λ1 (0) = 0 ∀β ∈ ZN . k k 2 (1. (In addition. k mn MY (χk mn ) = 0. (1. VANNINATHAN It is worth remarking that even though error estimates of the type (1. n = 1. it is clear from the deﬁnition of θε that it is necessary to obtain asymptotic expansions of the ﬁrst eigenvalue λε (ξ) and the ﬁrst Bloch mode φε (·.1174 C. Proposition 1. i. χk m . we obtain these natural estimates under optimal hypotheses. and their proofs will be taken up in the following sections along with other auxiliary results.
Theorem 1.BLOCH APPROXIMATION 1175 + (amn − qmn )χk − 1 1 3! Y  Y (ak − qk )χmn + (a m −q m )χnk + (ank − qnk )χ m + (akm − qkm )χ n + (a n − q n )χkm dy. then θ ε − u∗ L2 (RN ) ≤ cε u∗ H 1 (RN ) . (i) If u∗ ∈ H 1 (RN ). Assume that the hypotheses of Theorem 1. mn k = Y 1 χ 4 χ + χkmn χ + χn k χm + χk n χn dy χ + χkm χn + χ k χnm dy 1 Y  Y 1 χ 6 m kn 1 1 (2) (2) (2) (2) (2) β β (2) + βkm βn + βkn βm . ∂xk ∂x This can be rigorously proved. we can expand it and thus give rise to an asymptotic expansion of θε which is as follows: (1. Y  2 k mn We note that all odd order derivatives of φ1 at η = 0 are purely imaginary and all even order derivatives are real. 1. Proposition 1. then we have k ε θε − u∗ − εχε k ∂u∗ ∂xk H 1 (RN ) ≤ cε f L2 (RN ) . η) is proved to be analytic for η ≤ δ. (2) mn φ1 (y. 0) (y) − 1 (2) (2) (2) β χ (y) + β m χnk (y) + βmn χk 3! k mn (2) (2) n χkm (y) + βnk χ with βk = βk (4) mn (2) m (y) + βkm χn (y) + β p(0) + βk (4) (0) mn p 1 1 2! Y  1 Y  − + Y χ χk dy.10.8 hold. 0) = ip(0) χk (y).13) and χε (x) = χk x . 1 2 (2) D φ1 (y. We have the following expressions: Dk φ1 (y.11. 2! k 1 3 D 3! k 1 4 D 4! k m φ1 (y. Our next result is a sample where we specify the precise hypotheses needed to justify the above expansion up to three terms. Since φ1 (·. Various derivatives of φ1 at η = 0 can also be calculated in a systematic fashion. 0) = − ip(0) χk = p(0) χk (2) mn m (y) − i 3 βk χm (y) + β (2) (2) m χk (y) + βmk χ (y) p(0) .33) θε (x) = u∗ (x) + εχ k x ε ∂u∗ (x) − ε2 χk ∂xk x (2) + βk ε ∂ 2 u∗ (x) + · · · .∞ (ii) If f ∈ L2 (RN ) and χk ∈ W# (Y ). . 0) = − p(0) χk (y) − βk p(0) . where χk is the solution of (1.
(In fact all odd order derivatives vanish.∞ estimates are hard to come by. Section 5 is devoted to the description of the asymptotic ε behavior of the ﬁrst Bloch transform B1 whose deﬁnition is given in Theorem 1. One example of such a situation is the study of the propagation of waves in a periodic medium. namely. Their contributions occur somewhat separately without interaction at the levels of homogenized equation and correctors.30). we prove in section 3 that the higher order Bloch modes are negligible. φε (·.1.∞ (iii) If f ∈ H 1 (RN ) and χk . where the expansion has a multiscale structure. More precisely. η). . Since higher order modes can be neglected. R. However. the ﬁrst eigenvalue λ1 (η) along with the ﬁrst eigenvector φ1 (·. We plan to cover these aspects in a future publication. Finally. θε − u∗ − εχε k H 1 (RN ) H 1 (RN ) It is important to note that these above expansions are of Taylor type owing to the analyticity of λ1 (η) and φ1 (·. our arguments are more general compared with the onedimensional case covered in [20]. we present the proofs of the main results.3 below have further interesting consequences which will be developed in a forthcoming paper.9 and 1. ε The expansions of λ1 (η). we see the product of the second order ε derivatives of λ1 (η) at η = 0 with the 0th order term of B1 g(ξ).2. too. and B1 g(ξ) obtained in Propositions 1. the main point is that the third order derivatives of λ1 (η) at η = 0 are shown to be zero even in the multidimensional case. βk are constants deﬁned in Proposition 1. χk ∈ W# (Y ). Section 2 is devoted to certain fundamental lemmas which are indispensable.) Moreover. Theorems 1.2 and 5. the ﬁrst eigenvalue λ1 (η) contributes at various levels through its derivatives at η = 0. 5.10. for instance. ORIVE.1176 (2) C. There seem to be situations where both interact and contribute jointly in a manner diﬀerent from the above. η). W. and 5. Taylor expansions for λ1 and φ1 are obtained in section 4 which proves Propositions 1. Apart from the mathematical rigor. AND M. It appears that the homogenized medium is not good enough to provide an approximation to the propagation for large times because of the appearance of dispersion eﬀects shown numerically in F.3.9 1 and Propositions 5. We conclude this introduction by saying how the rest of this paper is organized. CONCA. it is worth mentioning from regularity theory of elliptic boundary value problems that W 1. The ﬁrst eigenvector φ1 (·. in section 6. and χε (x) = χk k ∂ 2 u∗ ∂u∗ (2) + ε2 χε + βk k ∂xk ∂xk ∂x ≤ cε2 f x ε . As an immediate application.2. η) represent the periodic medium under consideration. VANNINATHAN 1. Symes [20].11. Concerning the hypotheses on the smoothness of functions χk and χk in statements (ii) and (iii). Santosa and W. namely. g(ξ).5 and 1.10. We see a similar structure in the correctors. This is the main diﬀerence between this approach and the classical one found in [5]. This is why they are usually assumed in homogenization theory. several numerical studies with simple ﬁbers show that these assumptions are valid. This will have consequences in the propagation of waves in periodic media. we will be content with a few remarks. then . For the time being. We feel that this is an appropriate place to highlight the improvements achieved in this work with respect to [20]. Thus they are reasonable hypotheses to work with as far as certain applications are concerned. where χk is the solution of (1. η) and its ﬁrst derivatives contribute through the ﬁrst Bloch transform ε B1 g(ξ) and its expansion described in Propositions 5.8 and 1. In the homogenized equation.
by uniform ellipticity of Aε . using the spectral resolution of Aε . In this section. h ∈ L2 (RN ). It is well known that (Aε + I): H 1 (RN ) → H −1 (RN ) is an isomorphism. we get Aε g¯dx = g ∞ ε−1 Y m=1 ε λε (ξ)Bm g(ξ)2 dξ.1) RN g(x)h(x)dx = ε−1 Y m=1 ε ε Bm g(ξ)Bm h(ξ)dξ ∀g. we prove a series of results which generalize Parseval’s identity stated in Theorem 1. m RN This completes the proof. First of all. Lemma 2. there exist c1 and c2 which are independent of ε and g. These estimates will be useful later for the analysis of correctors. an application of the Cauchy–Schwarz inequality yields ∞ g. Indeed. H −1 (RN ) Proof.2. For every g ∈ H −1 (RN ) there exists a unique solution u ∈ H 1 (RN ) of Aε u + u = g in RN . For all g ∈ H 1 (RN ). we deduce the following. Lemma 2. we have α ∇g2 dx ≤ Aε g¯dx ≤ β g ∇g2 dx. Fundamental lemmas. RN RN RN We can rewrite the middle term by applying the Plancherel identity: ∞ (2. where c1 and c2 are constants independent of ε and g.BLOCH APPROXIMATION 1177 2. By using the duality between H 1 (RN ) and H −1 (RN ). Proof. such that c1 g 2 H −1 (RN ) ≤ ε−1 Y 1 B ε g(ξ)2 dξ ≤ c2 g 1 + λε (ξ) m m m=1 ∞ 2 .1. For all g ∈ H −1 (RN ). The following two lemmas are easily seen to be generalizations of wellknown classical results for −∆. ξ ∈ ε−1 Y . v = ε−1 Y m=1 ε ε (λε (ξ) + 1)Bm uBm vdξ m 1/2 ε (λε (ξ) + 1)Bm u2 dξ m 1/2 1/2 ε (λε (ξ) + 1)Bm v2 dξ m 1/2 ∞ ∞ ε−1 Y m=1 ∞ ≤ ε−1 Y m=1 ∞ ≤ ε−1 Y ε Bm g2 dξ ε (ξ) + 1) (λm m=1 ε−1 Y m=1 (λε (ξ) m + ε 1)Bm v2 dξ .3. m Therefore. We can express the previous equation in the equivalent form ε ε (λε (ξ) + 1)Bm u(ξ) = Bm g(ξ) ∀m ≥ 1. we have c1 g2 1 N H (R ) ∞ ≤ ε−1 Y ε λε (ξ)Bm g(ξ)2 dξ ≤ c2 g2 m m=1 H 1 (RN ) .
By virtue of Lemma 2. The following result estimates its L2 (RN ) and H 1 (RN ) norms.2. To prove the upper estimate is enough to use the continuity of the solution u ∈ H 1 (RN ) with respect to the righthand side g ∈ H −1 (RN ). ξ)dξ.3) ρ(·. η). 1 Lemma 2. η) and Y periodic in y. η) measurable with respect to (y. εξ) + ε−1 ∇y ρ(·.1 and Parseval’s identity. We then introduce (2. L2 (Y )N Gε 2 H 1 (RN ) ε−1 Y g ε (ξ)2 iξρ(·. we formally diﬀerentiate Gε (x) with respect to x. we get Gε (x) = g ε (ξ) ∞ m=1 ε−1 Y am (εξ)eix·ξ φε (x. Here. Then we have Gε 2 L2 (RN ) = = ε−1 Y g ε (ξ)2 ρ(·. η)}∞ m=1 where η is a parameter: ∞ ρ(y. Thus we deduce the existence of a constant c1 such that c1 g 2 H −1 (RN ) ≤ ε−1 Y 1 B ε g(ξ)2 dξ. η) as a function of y in the orthonormal basis {φm (y. εξ) 2 dξ.1178 C. R. εξ) Proof. m Applying Parseval’s identity of Theorem 1. H# (Y )). CONCA. ·. η) = m=1 am (η)φm (y.2) Gε (x) = ε−1 Y g ε (ξ)eix·ξ ρ x . η) 2 L2 (Y ) ∞ = m=1 am (η)2 ∀η ∈ Y .2). VANNINATHAN for all v ∈ H 1 (RN ). We assume g ε ∈ L2 (ε−1 Y ) and ρ ∈ L∞ (Y . Introducing this expression in (2. 1 + λε (ξ) m m m=1 ∞ which is the lower estimate in Lemma 2. the second term in the righthand side is equivalent to the H 1 norm of v. L2 (Y ) 2 dξ. This completes the proof of the ﬁrst part of the lemma if we use Parseval’s identity in L2 (Y ): (2.3. We obtain x x ∇x Gε (x) = g ε (ξ)eix·ξ iξρ .3. g ∈ H −1 (RN ). AND M. εξ + ε−1 ∇y ρ . εξ dξ. We expand ρ(y. ORIVE. In our next lemma. · denotes the H 1 (RN ) and H −1 (RN ) duality pairing. and another function ρ = ρ(y. εξ dξ. ε ε ε−1 Y . For the second part of the lemma. we consider g ε = g ε (ξ) a measurable function deﬁned on −1 ε Y . we get Gε 2 L2 (RN ) = ε−1 Y g ε (ξ)2 ∞ m=1 am (εξ)2 dξ. ε x ∈ RN .
φ. ∂y ∂yk ∂ψ + iηk ψ dy. k = 1. We deﬁne (2. .5. φm (η). To prove this. a(η. Lemma 2. For simplicity. let us introduce the bilinear forms associated with the operators A(η) and A. .7) 1 a(η. We assume that ρ is Y periodic in y.L2 (Y )) # . N. ψ) ∀ψ ∈ H# (Y ). ψ) = ak (y) ak (y) ∂φ + iη φ ∂y ∂φ ∂ψ dy.2). we denote φm (·. 1 (ii) If g ∈ H 1 (RN ) and ρ ∈ L∞ (Y . ε where g = g(x) is a measurable function deﬁned on RN and ρ = ρ(y.5). (2.4. ∂yk Y Y The basic estimates on them are obtained in [10. Then there exists a constant c depending on ak L∞ (Y ) such that (2. m ≥ 1. Our next result concerns the estimation of expressions which are inverse to (2. ψ) = λm (η)(φm (η). φ) ≤ c ∇φ 2 L2 (Y )N 2 . H# (Y )). L2 (Y )N + η2 φ 2 ). ψ) = a(φ.2). L2 (Y )). p.BLOCH APPROXIMATION 1179 We remark that the above integral is of the same type as the one analyzed in the ﬁrst part.L2 (Y )N ) . η) is a measurable function deﬁned on Y × Y . η) by φm (η). φ) ≤ c ( ∇φ ≤ a(φ. εξ dx for ξ ∈ ε−1 Y . φ. respectively. it is enough to take ψ = φm (η) and use (2. . L2 (Y ) Proof of Lemma 2. The required hypotheses on these functions will depend on the estimate deduced on J ε g. η) ∂yk L2 (Y ) ≤ c1 λm (η)1/2 ∀η ∈ Y .8) J ε g(ξ) = RN g(x)e−ix·ξ ρ x . We suppose that the coeﬃcients ak satisfy (1. (i) If g ∈ L2 (RN ) and ρ ∈ L∞ (Y . We recall that it satisﬁes (2. . The next lemma presents H 1 estimates on the Bloch modes. This completes the proof. c 1 which are independent of η ∈ Y such that for all φ ∈ H# (Y ). To deduce (2. 190]: There exist constants c. then we have (1 + ξ2 )1/2 J ε g(ξ) L2 (ε−1 Y ) ≤c ∇g L2 (RN ) ρ L∞ (Y .5) c ( ∇φ (2. . Lemma 2.L2 (Y )) + ε−1 g L2 (RN ) ∇y ρ L∞ (Y .4) ∂φm (·. This is illustrated in the results that follow which are analogous to classical estimates on the Fourier transform. then we have # J εg L2 (ε−1 Y ) ≤ g L2 (RN ) ρ L∞ (Y .4).6) 2 L2 (Y )N + η2 φ c ∇φ 2 ) L2 (Y ) 2 L2 (Y )N ≤ a(η.4.
AND M.8) by (−iξk ) and obtain (−iξk )J ε g(ξ) = RN g(x)(−iξk )e−ix·ξ ρ x . ε which. m=1 The proof of (i) is complete if we integrate the above inequality with respect to ξ ∈ ε−1 Y and apply Theorem 1.L2 (Y )) # ε Bm g(ξ)2 . By the Cauchy–Schwarz inequality. is equal to J ε g(ξ) = ∞ m=1 n∈ZN ε amn e2πiεn·ξ Bm g(ξ) = ∞ m=1 ε am (εξ)Bm g(ξ). ∀m ∈ N. η) ∀y ∈ Y. εξ) 2 L2 (Y ) ε Bm g(ξ)2 ∞ ≤ ρ 2 L∞ (Y . ORIVE. First using the fact that {φm (·. εξ dx − ε−1 ∂xk ε RN g(x)e−ix·ξ . η) 2 L2 (Y ) = m am (η)2 amn 2 n∈ZN ∀η ∈ Y . R. we can expand am (η) in the usual Fourier series: am (η) = n∈ZN amn e2πin·η ∀η ∈ Y . we get # ∞ ρ(y. η) in two steps. Next. L2 (Y )) and expand # ¯ ρ(y. VANNINATHAN Proof. η) = m=1 ¯ am (η)φm (y. For the proof of (ii).1180 C. by integration by parts. for each m. εξ dx. J ε g(ξ)2 ≤ ∞ m=1 am (εξ)2 ∞ m=1 ∞ m=1 ε Bm g(ξ)2 = ρ(·. ε ∂ρ ∂yk x . Y am (η)2 dη = Using this expansion. η)}∞ is an orthonormal basis m=1 in L2 (Y ). CONCA. η ∈ Y . εξ dx. The idea is to consider the product space L2 (Y . we can rewrite J ε g as follows: J ε g(ξ) = ∞ m=1 n∈ZN amn e2πiεn·ξ RN ¯ g(x)e−ix·ξ φm x . ε ε which. The corresponding Parseval’s identities are as follows: ρ(·. we multiply (2.3. εξ dx. according to the deﬁnition of Bm g(ξ). can be rewritten as (−iξk )J ε g(ξ) = − RN ∂g x (x)e−ix·ξ ρ .
(ii) εN F ε g measure of K.p (RN ) (p > N ) with compact support K. To prove (ii). (iii) εN χε−1 Y F ε g → L2 (ε−1 Y ) ≤ cK p−2 2p { g Lp (RN ) + ε ∇g Lp (RN )N }. More precisely. by Parseval’s identity. K denotes the 1 g (2π)N/2 in L2 (RN ). For g ∈ W 1. we multiply (2. ∈ZN To estimate the righthand side of the above equality. we integrate the inequality g(xε )2 ≤ 2 g(x)2 + g(x) − g(xε )2 . Let {Y ε } ∈ZN be the mesh of RN generated by the cell εY . Yε Yε . we have 1 (i) εN (χε−1 Y F ε g)(ξ) → (2π)N/2 g(ξ) for ξ ∈ RN .9) is nothing but the Fourier series in the variable ξ ∈ ε−1 Y . over Y ε to obtain (2.p (RN ) is embedded in C0 (RN ) when p > N .BLOCH APPROXIMATION 1181 It is now suﬃcient to apply (i) to each of the terms on the righthand side of the above relation.10) ε2N ε−1 Y F ε g(ξ)2 dξ = 1 (2π)N g(xε )2 Y ε .p (RN ) with compact support we deﬁne (2. It is worthwhile to recall that W 1. and so g(xε ) is well deﬁned. 1 (2π)N Proof. For g ∈ W 1. we get εN ε−1 Y F ε g(ξ)2 dξ = ∈ZN g(xε )2 . To this end. Next. we observe that the righthand side of (2. To prove (i). We recall the deﬁnition of the discrete Fourier transform of a function corresponding to this mesh: Let p > N be given. Therefore. Lemma 2. we will need some properties of the classical discrete Fourier transform in our asymptotic description of the ﬁrst Bloch transform.9) by εN to get εN F ε g(ξ) = g(xε )e−ix ∈ZN ε ·ξ Y ε .9) F ε g(ξ) = ∈ZN g(xε )e−ix ε ·ξ ∀ξ ∈ ε−1 Y .11) g(xε )2 Y ε  ≤ 2 g(x)2 dx + g(x) − g(xε )2 dx . let us begin by introducing some necessary notations. Y ε = xε + εY where xε = 2πε is the origin of the cell Y ε . We regard the righthand side of the above equality as a Riemann sum of the integral 1 (2π)N RN g(x)e−ix·ξ dx which converges to it as ε → 0. In particular.6. x ∈ Y ε. we are interested in the relation between discrete and continuous Fourier transforms. We multiply this relation by εN and rewrite it as (2.
we consider a sequence of solutions uε of the equation with f ∈ H −1 (RN ): (3. Higher Bloch modes are negligible.21). We obtain ∇v ε 2 2 N N L (R ) ≤c ≤c 1 B ε f (ξ)2 dξ λε (ξ) m m=2 m 1 f λε (ξ) m 2 . ORIVE. Thus we consider v ε deﬁned in (1. CONCA.11) and the H¨lder inequality and summing over o ∈ ZN . p. the second term converges to − 1 2 g (2π)N/2 (2π)N/2 2 . g) 1 + (2π)N g 2 .1) Aε uε = f in RN . Using both the above estimate in (2. we show that the component of uε in the higher Bloch modes does not play any role in the analysis of correctors of ﬁrst and second order provided f is suﬃciently smooth. To prove the statement (iii) we ﬁrst use (i) and (ii) to deduce that εN χε−1 Y (ξ)F ε g(ξ) Let us now expand εN χε−1 Y F ε g − 1 g (2π)N/2 2 L2 (RN ) N 1 g(ξ) (2π)N/2 in L2 (RN )weak. uε ∈ H 1 (RN ). Estimates on v ε derived in this section improve Proposition 1. L2 (RN ) This simple computation establishes the strong convergence in L2 (RN ).1. 167]) to deduce g(x) − g(xε ) ≤ cε1− p ∇gLp (Y ε )N . we complete the proof of (ii).1182 C. (ii) v ε Proof. We have the following estimates for f ∈ L2 (RN ): (i) v ε H 1 (RN ) ≤ cε f L2 (RN ) . sup ξ∈ε−1 Y . which is nothing but the projection of uε corresponding to all higher Bloch modes. Thanks to the above weak convergence.10) shows that ε2N F ε g 2 L2 (RN ) → 1 (2π)N RN g2 dx = 1 g (2π)N 2 L2 (RN ) . we present a systematic method of obtaining estimates on the solution in Sobolev spaces L2 and H 1 . AND M. = ε2N F ε g 2 L2 (RN ) 2ε − (2π)N/2 (χε−1 Y F ε g. VANNINATHAN Since p > N . In what follows.20). In particular. 3.4. Let us recall that the above equation is equivalent to (1. we can use the classical Morrey’s inequality (see Brezis [7. Proposition 3. L2 (RN ) ∞ L2 (RN ) ≤ cε f H −1 (RN ) . we apply Lemma 2. ε−1 Y m≥2. R.20) in the Bloch space.1 with g = v ε and use (1. In this section. To show (i). L2 (RN ) N Now relation (2.
η) can be chosen so that the following conditions are satisﬁed simultaneously: (4. η)dy = 0 ∀η ∈ Bδ . Let us state the following proposition.3) 1 η ∈ Bδ → φ1 (·. We have the following estimates for f ∈ H 1 (RN ): (i) v ε H 1 (RN ) ≤ cε2 f H 1 (RN ) . For the proof of (ii).2 with g = f and (1.10)). we can obtain H 2 estimates on the solution. This is diﬃcult as it involves more subtleties (see [12]). Taylor expansion of the ﬁrst Bloch eigenvalue and eigenvector.BLOCH APPROXIMATION 1183 Proof of (i) is complete since we have (cf. λε (ξ)2 m m=2 m ∞ ε Bm f (ξ)2 = ε 1 + λε (ξ) Bm f (ξ)2 m ε (ξ)2 1 + λε (ξ) λm m and using (3. Bm f (ξ)2 ≤ cε2 m ε λε (ξ)2 1 + λm (ξ) m The proof is complete if we use Lemma 2. η) at η = 0. φ1 (·.2.9 and 1.2. η) depend analytically on η in a small neighborhood Bδ of η = 0. We have vε 2 2 N L (R ) ∞ = ε−1 Y ε Bm uε (ξ)2 dξ m=2 = Writing 1 λε (ξ)2 m ε−1 Y 1 B ε f (ξ)2 dξ. (ii) v ε L2 (RN ) ≤ cε2 f Assuming ak are in and further assumptions. 4. . These ﬁner estimates require naturally higher order regularity of f but not of the coeﬃcients ak (y). At the cost of reducing this neighborhood. While the above proposition shows that v ε can be neglected at the level of the ﬁrst order correctors (cf.2) (4. Recall that λ1 (η) and φ1 (·.∞ W# (Y ) . In particular. we apply Lemma 2. we deduce that B ε f (ξ)2 1 ε . η) ∈ H# (Y ) L2 (RN ) 1. we will prove Propositions 1. (1. η) Im L2 (Y ) = 1 ∀η ∈ Bδ . The purpose of this section is to indicate a systematic method to compute derivatives of the ﬁrst Bloch eigenvalue λ1 (η) and the ﬁrst Bloch eigenvector φ1 (·.20).1) (4. (1. ξ∈ε−1 Y ε2 . the next result will demonstrate that v ε can be neglected at the level of correctors of ﬁrst and second order. Proposition 3. we claim that the branch η → φ1 (·.10. Y φ1 (y.2) sup 1 λε (ξ) m ≤ 1 (N ) λ2 m≥2.2). whose proof is similar to the previous one and hence will not be repeated.25)) (3. is analytic.
φ1 (·. Thanks to our choice of the branch satisfying (4.5) (4. η) 2 + Re D3 φ1 (·. D φ1 (·.3).8) Im Y Dβ φ1 (y. diﬀerentiating (4. (4. D2m φ1 (·. one such choice which is analytic is as follows: α1 (η) = −d2 (η). d2 (η)) = def Y (α1 (η) + iα2 (η))φ1 (y. + From these sets of relations.4) (4. where ·. η). η). Obviously. 0) (see Proposition 1. η). The idea consists of multiplying φ1 (·. N (4. we successively get for all k. η) + Re Dk φ1 (·. diﬀerentiation # of (4. η)dy. η). . However. On the other hand. Dmn φ1 (·. AND M. η) + Re Dk φ1 (·. η) = 0. η). η). η). η). the above procedure has destroyed condition (4. η). η) + Re Dk φ1 (·. 0)dy = 0 ∀ β ∈ ZN with β odd. η) + Re Dk m φ1 (·. D φ1 (·. η) = 0. 3 Re Dk m φ1 (·. η).2) is classical.1184 C. This is possible because d(0) = 0 by our choice of φ1 (·. η) + Re Dkm φ1 (·. VANNINATHAN In what follows. whereas the condition (4. D2m φ1 (·. Re Y φ1 (y. η). we will see that the above conditions uniquely ﬁx the eigenvector φ1 (·. η) 2 + Re Dkm φ1 (·. η) 2 + Re Dkn φ1 (·. Im Y φ1 (y. n = 1. Dn φ1 (·. In fact. η) by a complex number (α1 (η) + iα2 (η)) where α1 (η) and α2 (η) are real analytic with respect to η and are chosen such that Im If we deﬁne d(η) = (d1 (η). η)dy = 0 ∀ β ∈ ZN . η). η). .2) with respect to η. Of course.5). D3mn φ1 (·. we will now draw some consequences which will simplify the computations below. 2 Re Dk φ1 (·. ORIVE. · denotes the scalar product in L2 (Y ).3) is somewhat unusual and can be achieved as indicated below. .1)–(4. η)dy = 0. η)dy .1)).6) Re Dk φ1 (·. φ1 (·. η) + Re D2 φ1 (·. D φ1 (·. D2n φ1 (·. Dm φ1 (·. φ1 (·. + . η) = 0.7) 4 3 Re Dk mn φ1 (·. Dm φ1 (·.3) yields (4. it can be regained by dividing by d(η). η) 2 + Re Dk φ1 (·. it follows that (4. φ1 (·. m. α2 (η) = d1 (η). CONCA. We remark that the condition (4.9) Y Dβ φ1 (y. . R.2) (but not condition (4. then the above condition is equivalent to α1 (η)d1 (η) + α2 (η)d2 (η) = 0 ∀η ∈ Bδ . η). η). η) = 0. η). . η) k n k 3 2 + Re Dkmn φ1 (·.
.11) at η = 0 and use the structure of Ck . The next step is to compute the ﬁrst order derivatives of φ1 at η = 0.11) [Dk (A − λ1 )]φ1 .BLOCH APPROXIMATION 1185 4.10). where we have suppressed the dependence on η for ease of writing. we go back to (4.19). we get (4. φ1 + [Dk (A − λ1 )]D φ1 . . which we diﬀerentiate once with respect to η. . we can solve for Dk φ1 (y. φ1 = 0. If we diﬀerentiate the eigenvalue equation (A(η) − λ1 (η))φ1 (·. We obtain 2 2 [Dk (A − λ1 )]φ1 + [Dk (A − λ1 )]D φ1 + [D (A − λ1 )]Dk φ1 + (A − λ1 )Dk φ1 = 0.10) Dk (A(η) − λ1 (η))φ1 (·. we obtain (4. we immediately get that (4.17) [Dk (A − λ1 )]φ1 . Dk A(0). = 1.2.5. (4. . . 0) = iφ1 (y. 0). 0) = −iCk φ1 (·.1. If we use the information obtained in section 4. We will continue with this convention in what follows provided there is no ambiguity. . 0) and obtain (4.18) that (4. 0). N. 4.13).1 on Dk λ1 (0). N.13) Dk λ1 (0) = 0 ∀k = 1. η) = 0. 0) = −Dk A(0)φ1 (·. Taking the scalar product with φ1 (·. . Thus. where. and (4. First order derivatives.13) and the constant p(0) was ﬁxed in Proposition 1.15) Dk φ1 (y. χk satisﬁes (1. η) + (A(η) − λ1 (η))Dk φ1 (·.10) and use (4. we get 2 (4.14) Dk φ1 (y. It follows from (1. . To this end. 0)χk (y) = ip(0) χk (y).9) and the above equation. We obtain ADk φ1 (·. Second order derivatives. If we evaluate the relation (4. the ﬁrst order derivative is completely determined and (4. φ1 + [D (A − λ1 )]Dk φ1 . η ∈ Y .18) 2 Dk A(η) = 2ak (y) ∀k. where the operator Ck is deﬁned in (1. Our starting point is the relation (4. we recall. Dk φ1 (·. η) = 0 once with respect to ηk . .12) Dk A(0) = iCk ∀η ∈ Y . Taking into account (4. φ1 = 0 for all η ∈ Bδ .16) Taking the scalar product with φ1 . η). 0) is purely imaginary.
. we get 3 2 [Dk m (A − λ1 )]φ1 + [Dk (A − λ1 )]Dm φ1 + [D2m (A − λ1 )]Dk φ1 2 2 (4.8).16). φ1 + [Dm (A − λ1 )]Dk φ1 . Diﬀerentiating (4. ORIVE. This proves the ﬁrst part of Proposition 1. φ1 2 2 + [D (A − λ1 )]Dkm φ1 .10. By comparing the above equation with (1.9. we see that Dk φ1 (·. 4. Dk m A = 0. φ1 = 0. . . Therefore we will be brief here.1186 we obtain C. . βk (2) (2) βk (2) 2 and Dk φ1 (·. R. φ1 2 + [Dkm (A − λ1 )]D φ1 . From the calculations done so far. CONCA. N. the next step is to compute Dk φ1 (·.3. Dk φ1 (0) are real. 2 As before. For this purpose. we go back to (4. VANNINATHAN 1 1 2 D λ1 (0) = 2! k Y  (4. 0) = −2(ak − qk ) + Ck χ + C χk φ1 (·. φ1 + [D2m (A − λ1 )]Dk φ1 . Thanks to (4. φ1 + [Dk (A − λ1 )]D2m φ1 . Taking the scalar product with φ1 . .20) Moreover. 0) = −p(0) χk (y) − βk p(0) 2! k for some constant βk . 0). AND M. admits the expression given in Proposition 1. (4.19) = Y ak (y)dy − 1 2Y  Y (Ck χ (y) + C χk (y))dy 1 (qk + q k ) = qk 2 ∀k. 2 φ1 (0).22) 3 To conclude that Dk above relation: (4. we can infer that (4. Third order derivatives. 0).23) m λ1 (0) = 0.30) and using the simplicity of the eigenvalue 2 under consideration. it is enough to use the following information in the 2 3 Dk A is real. Dk A is purely imaginary.5) and (4. 0) are real.21) + [Dkm (A − λ1 )]D φ1 + [Dk (A − λ1 )]D2m φ1 + [D (A − λ1 )]Dkm φ1 2 3 + [Dm (A − λ1 )]Dk φ1 + (A − λ1 )Dk m φ1 = 0. it is now clear how to proceed further to calculate higher order derivatives. Dk φ1 (0) is purely imaginary. = 1. φ1 + [Dk (A − λ1 )]Dm φ1 . 0) is of the form 1 2 (2) D φ1 (y. It is evident that the above argument is very general and so can be used to establish that all odd order derivatives of λ1 at η = 0 vanish. we get 3 2 [Dk m (A − λ1 )]φ1 .16) and rewrite it with η = 0 as follows: 2 ADk φ1 (·.
ξ where. 5. ξ This will be a consequence of a more general result.9 and 1. Further. J ε g was deﬁned in (2.31). 0) which can be compared with (1. L2 (Y )). η) ∈ L2 (Y ) is # Lipschitz near η = 0. Fourth order derivatives. we have ε χε−1 Y (ξ)B1 g(ξ) → g(ξ) in L2 (RN ). we have the following proposition. the relation (4.21) deﬁnes 3 a periodic problem for Dk m φ1 (·. η)e−iy·η dy.p (RN ) # with compact support K and with p > N . Proof of Proposition 5. we need to introduce some new notations.1.2) χε−1 Y (ξ) J ε g(ξ) − (2π)N/2 ρ(0) (εξ)g(ξ) → 0 ˜ in L2 (RN ). .3. 0) given in Proposition 1. We associate with every function ρ = ρ(y. we have ε χε−1 Y (ξ)B1 g(ξ) → g(ξ) in L∞ (RN ). These observations are enough to get the 3 expression of Dk m φ1 (·. η) deﬁned on Y × Y which is Y periodic in y the following function: (5. then by using the fact that the map η → φ1 (·.9) says that its average vanishes. To arrive at the expressions for the fourth order derivatives of λ1 and φ1 at η = 0 given in Propositions 1. If g ∈ L2 (RN ) with compact support K.1) ρ(0) (η) = 1 Y  Y ρ(y. 0) L2 (Y ) ≤ cε. 0) is purely imaginary. Convergence of the ﬁrst Bloch transform to the Fourier transform. we have (5.4. 0) L2 (Y ) + (χε−1 Y (ξ) − 1)g(ξ). η∈Y . This section is devoted to the proof of the next proposition which shows the sense in which the Fourier transform is approximated by the ﬁrst Bloch transform.10. The proof will be taken up later. With this notation. we have for all ξ ∈ RN ε ε χε−1 Y (ξ)B1 g(ξ) − g(ξ) ≤ χε−1 Y (ξ)(B1 g(ξ) − g(ξ)) + (χε−1 Y (ξ) − 1)g(ξ) ≤ cK g L2 (RN ) φ1 (·.1. εξ) − φ1 (·.24) 3 Dk m φ1 (y.2. we follow the same arguments as in section 4. We conclude by observing the following important property: (4. Admitting it for the moment. we deduce φ1 (·. If ξ is bounded. εξ) − φ1 (·. We suppose ρ ∈ L∞ (Y .10. Proposition 5. we realize that (4. (i) For every g ∈ L2 (RN ) with compact support. we turn our attention to the following proof. Then for all g ∈ W 1. Proposition 5. 4. In order to state it. loc ξ (ii) If g ∈ L2 (RN ).BLOCH APPROXIMATION 1187 To ﬁnd the third order derivatives of φ1 at η = 0.8). we recall.
Lemma 2. AND M.4) ε r1 (ξ) = ∈ZN Yε (g(x) − g(xε ))e−ix·ξ ρ x . namely. The proof of (ii) is more involved. we consider the εmesh {Y ε } ∈ZN generated by the cell εY which was already introduced at the end of section 2. according to Theorem 1. εξ) with ε respect to x is faster than that of g. First. Thus. by Lebesgue’s dominated convergence theorem. we rewrite r1 in a slightly diﬀerent form. ε To this end. CONCA. with ρ = φ1 . where F ε g is the discrete Fourier transform of g and ρ(0) is deﬁned in (5. (5.2. Since ˜ 1 we know that χε−1 Y (ξ)εN F ε g(ξ) → (2π)N/2 g (ξ) in L2 (RN ) (cf. R. our ˆ hypothesis on ρ ensures that (5.3) J ε g(ξ) = ∈ZN Yε g(x)e−ix·ξ ρ e−ix·ξ ρ x .5) ˜ χε−1 Y (ξ) Y εN F ε g(ξ) − (2π)N/2 g(ξ) ρ(0) (εξ) L2 (RN ) → 0. by means of the change of variables x = xε + εy. ε The ﬁrst term on the righthand side of (5. we have the uniform estimate ε−1 Y ε B1 g(ξ)2 dξ ≤ RN g(x)2 dx. ε . ξ Proof of Proposition 5. ORIVE. Indeed. VANNINATHAN This completes the proof of (i). we see that ρ(0) (εξ) → p(0) ˜ and ε B1 g(ξ) = J ε g(ξ) ∀ξ ∈ RN . εξ dx ε x ε .L2 (Y )) # ∇g Lp (RN ) .7) ε r1 (ξ) = RN ε g1 (x)e−ix·ξ ρ x . that (2π)N/2 χε−1 Y (ξ)˜(0) (εξ)g(ξ) → g(ξ) ρ in L2 (RN ).6).2. by the usual density arguments. ε = ∈ZN g(xε ) Y ε where (5. εξ dx. εξ dx. εξ dx + r1 (ξ). To exploit this.6) ε r1 L2 (ε−1 Y ) ≤ c(K) ε ρ (1 − N ) p L∞ (Y . it is enough to show that (5.3) can be. ¯ We can now complete the proof using Proposition 5. We decompose (5. and so.1). it is enough to prove (ii) with g ∈ D(RN ).3. The key point is that the variation of ρ( x . transformed into Y εN F ε g(ξ)ρ(0) (εξ). to complete the proof. which implies.1188 C.
η)yk e−iy·η dy ∀k = 1. Proposition 5. in fact. . .14) ε r2 L2 (ε−1 Y ) ≤ c(K) 2 ε ρ (2 − N ) p L∞ (Y . we have χε−1 Y (ξ) J ε g(ξ) − (2π)N/2 ρ(0) (εξ) + iεξk ρ(k) (εξ) g(ξ) → 0 ˜ ˜ in L2 (RN ). ε (5. We follow the idea of the proof of Proposition 5. o The proof of Proposition 5. and so we can deduce (5. . namely.9) can be obtained from (5. Thanks to our hypothesis. .BLOCH APPROXIMATION 1189 where (5. It introduces naturally the following quantities: (5. L2 (Y )).10) is a simple consequence of Morrey’s estimate (see [7.15) ε r2 (ξ) = RN ε g2 (x)e−ix·ξ ρ x . We note that (5.L2 (Y )) # gW 2. We suppose ρ ∈ L∞ (Y . 2 Finally. p.6) provided we have the estimate (5. will be a consequence of Lemma 2. ε This. εξ dx.5. Then we have the following corrector result for J ε g.9) ε g1 L2 (RN ) ≤ c(K) ε ∇g (1 − N ) p c ε ∇g (1 − N ) p Lp (RN ) . N. We already know how to estimate integrals of the type (5. η∈Y .10) ε g1 Lp (RN ) ≤ Lp (RN ) . We decompose J ε g(ξ) {g(xε ) + ∇g(xε ) · (x − xε )} e−ix·ξ ρ x ε . ε ε We can estimate r2 (ξ) as follows: (5.8) ε g1 (x) = ∈ZN (g(x) − g(xε ))χY ε (x).2 shows that the result can be strengthened by assuming suitable smoothness on g.7) in L2 (RN ) (see Lemma 2. where c is a constant independent of K. because we can represent r2 as follows: (5.13) r2 (ξ) = ∈ZN Yε {g(x) − g(xε ) − ∇g(xε ) · (x − xε )} e−ix·ξ ρ x . we can deduce a stronger estimate.11) ρ(k) (η) = 1 Y  Y ρ(y. Our next result is an example in this direction. 167]). the support of g.3.p (RN ) . εξ dx + r2 (ξ). Then for all g ∈ W 2.10) with c(K) = cK1− p and a simple application of the H¨lder inequality. (5. ξ as Proof.12) J ε g(ξ) = ∈ZN Y ε where ε (5. we note that (5. εξ dx ε .p (RN ) # with compact support K and with p > N .2.5).
2 of the introduction (namely.p functions. Indeed. R..8 and 1. what we need is a generalization of Morrey’s inequality for W 2. (5. 6. (2 − N ) p 2 ε g2 Lp (RN ) As before.p (RN ) .20) Y  εN (F ε g)(ξ)˜(0) (εξ) + εN +1 F ε ρ ∂g ∂xk (ξ)˜(k) (εξ) .21) χε−1 Y Y  εN Fε ∂g ∂xk (ξ) − 1 iξk g(ξ) ρ(k) (εξ) ˜ (2π)N/2 → 0 in L2 (RN ).5)): (5. which admits the following estimates: (5. it is equal to (5.12).p (RN ) .6.11 and the statement (1. CONCA. we are now in a position to prove the main convergence results stated in section 1. in H 1 (RN )weak. ξ This clearly allows us to complete the proof. We begin by recalling brieﬂy the setup. namely. (5. N (2− p ) Admitting the above estimate. Theorems 1. To establish (5. Proof of the main convergence results.p (RN ) and the following representation: for all x ∈ Y ε .e. Thus. ε ε This completes the proof of the estimate (5. ORIVE. g(x) − g(xε ) − ∇g(xε ) · (x − xε ) = 1 0 {∇g((1 − t)xε + tx) − ∇g(xε )} · (x − xε )dt.1) . (6.17) (5. (2 − N ) p c ε2 gW 2. Applying the previously developed techniques and results.18).14) on r2 . ε ε A u =f uε u∗ ε u → u∗ in RN .9). r2 tends to zero more rapidly. AND M. we have the following convergence (apart from (5.17) will be a consequence of (5.18) ε g2 L2 (RN ) ≤ ≤ c(K) 2 ε gW 2. We take f ∈ L2 (RN ) and consider a sequence uε satisfying (1.1190 with (5. i. as expected.p (Y ε ) ∀x ∈ Y ε . it is an easy matter to prove (5. The same cannot be said for the ﬁrst term on the righthand side of (5. But the above estimate is a consequence of Morrey’s inequality for the gradient ∇g ∈ W 1.8)).19) g(x)−g(xε )−∇g(xε ) · (x−xε ) ≤ N c x−xε 2− p gW 2. ρ According to Lemma 2. in L2 (RN )strong.18).16) C. VANNINATHAN ε g2 (x) = ∈ZN (g(x) − g(xε ) − ∇g(xε ) · (x − xε )) χY ε (x).18) with c(K) = cK1− p .
Then wε → w∗ in L2 (RN )strong. H 1 (RN )weak.1. look artiﬁcial. Our hypothesis that uε → u∗ in L2 (RN )strong may. which is equivalent to 1 2 D λ1 (0)ξk ξ + 1 w∗ (ξ) = g(ξ). Assume that wε satisﬁes (6. First of all. It suﬃces to show (6. No concentration of energy at inﬁnity.3) ε ε B1 wε (ξ) − B1 w∗ (ξ)2 dξ → 0. 1 ξ ∈ ε−1 Y . Proposition 6.1.2) gives the relation ε ε (1 + λε (ξ))B1 wε (ξ) = B1 g(ξ). We use it to write ε ε χε−1 Y (ξ)(B1 wε (ξ) − B1 w∗ (ξ)) = χε−1 Y (ξ) ε B1 g(ξ) 1 + λε (ξ) 1 ε −w∗ (ξ) − (χε−1 Y (ξ)B1 w∗ (ξ) − w∗ (ξ)). In that case. then it is classical that the weak convergence in H 1 (Ω) will automatically imply the strong convergence in L2 (Ω). where g is a given function in L2 (RN ).BLOCH APPROXIMATION 1191 6. . This is not the case in RN . we have the following.1. But this in not the case. following the idea of Proposition 3. ξ Note that w∗ satisﬁes the homogenized equation A∗ w∗ + w∗ = g in RN .1.4) χε−1 Y (ξ) ε B1 g(ξ) − w∗ (ξ) → 0 1 + λε (ξ) 1 in L2 (RN ). ε−1 Y Equation (6. Therefore. Proof. According to Proposition 5. the correct operator to consider is (Aε + I) instead of Aε in RN . If Ω is bounded and smooth. ε Bm wε (ξ)2 dξ ≤ cε4 . it remains to prove (6. More precisely. at ﬁrst sight. we can neglect higher Bloch modes of wε and w∗ . we can show ∞ ε−1 Y m=2 ∞ ε−1 Y m=2 ε Bm w∗ (ξ)2 dξ ≤ cε2 . To make comparisons. the last term tends to zero in L2 (RN ).2) (Aε + I)wε = g wε w∗ in in RN . 2 k ξ ∈ RN .
R. In the second integral. we have cε ≥ 1 and bε (ξ) ≤ cγ 3 εg(ξ). In the ﬁrst two parts. CONCA. 1 2 D λ1 (0)ξk ξ 2 k ε χε−1 Y (ξ)B1 g(ξ) − g(ξ) 1 + λε (ξ) 1 → 0 in L2 (RN ) ξ because [1 + λε (ξ)] ≥ 1 and by the virtue of Proposition 5.5). 1 2 cε ≥ λε (ξ) + Dk λ1 (0)ξk ξ . To show this. . and so the second integral also converges to zero.5) χε−1 Y (ξ) ε g(ξ) B1 g(ξ) − →0 2 1 + λε (ξ) 1 + 1 Dk λ1 (0)ξk ξ 1 2 in L2 (RN ). 1 2 bε = − λε (ξ) − Dk λ1 (0)ξk ξ 1 2 cε = (1 + λε (ξ)) 1 + 1 Now we have the convergence aε = cε g(ξ). AND M. ξ The above expression can be written in the form aε + bε .1. we use the bounds bε  ≤ 1 2 λε (ξ) + Dk λ1 (0)ξk ξ 1 2 g(ξ). ORIVE.1192 C. In the ﬁrst integral. we still have bε (ξ) ≤ cξ3 εg(ξ). we have cε ≥ 1 ε 2 λ (ξ)Dk λ1 (0)ξk ξ ≥ cξ4 ≥ cγξ3 2 1 since ξ ≥ γ > 0. (6. VANNINATHAN So. 1 2 . taking γ > 0 as a ﬁxed constant: ξ≤δε−1 ξ≤γ bε cε 2 dξ + ξ≤δε−1 ξ>γ bε cε 2 dξ + ξ>δε−1 bε cε 2 dξ.6) 1 2 λε (ξ) − Dk λ1 (0)ξk ξ ≤ cξ3 ε 1 2 for ξ ≤ δε−1 .4) is reduced to (6. we split the energy into three parts. which holds since λ1 (0) = Dλ1 (0) = 0 (see Proposition 1. With regards to bε . we use the estimate (6. cε where aε = 1+ 1 2 D λ1 (0)ξk ξ 2 k ε χε−1 Y (ξ)B1 g(ξ) − g(ξ) . and consequently it is less than cε2 RN g(ξ)2 dξ and hence converges to zero. In the third integral. 1 bε The convergence of cε is not immediate.
27).13) (6. ε where v ε and P1 uε are deﬁned in (1. this tends to zero as ε → 0 since g ∈ L2 (RN ). we get ε. 1 ξ∈ε−1 Y ξ>δε−1 We will obtain the estimates (6. Using Lemma 2. Corrector result in RN . To this end.1.δ H 1 (RN ) ≤ cε f ε.BLOCH APPROXIMATION 1193 Thus the third integral is estimated from above by ξ>δε−1 g(ξ)2 dξ.δ P1 uε 2 H 1 (RN ) ≤c ξ∈ε−1 Y ξ>δε−1 ε B1 uε (ξ)2 λε (ξ)dξ. 1 ε B1 uε (ξ)eix·ξ φε (x. Step 1.8 concerning the Bloch approximation θε .9) (6. Thanks to Proposition 3. ξ)dξ.δ P1 uε (x) = ξ∈ε−1 Y ξ>δε−1 u∗ (ξ)eix·ξ φε (x. we introduce the quantities (6. We estimate the energies in the region ξ > δε−1 .δ 2 H 1 (RN ) ≤c ξ∈ε−1 Y ξ>δε−1 u∗ (ξ)2 ξ2 dξ. ≤ cε f H −1 (RN ) ε. ξ)dξ. 6. we get (6. θε.24).10) θε. We start with (6. we prove (6.21) and (1. it is enough to prove (6. We decompose uε as follows: ε u ε = v ε + P1 u ε . ε P1 uε − θε H 1 (RN ) ≤ cε f Step 2. 1 Now (6.2. Following the above procedure.3 with ρ = φ1 and inequalities (2.14).22) and (1.12) (6.12). L2 (RN ) . This section is devoted to the proof of Theorem 1.δ P1 uε L2 (RN ) L2 (RN ) .δ P1 uε H 1 (RN ) ≤ cε f L2 (RN ) . The proof consists of several steps which correspond to estimations of the required energy in diﬀerent regions in the Fourier space (in a neighborhood of the origin η ≤ δ and in its complement η > δ).δ L2 (RN ) ≤ cε f H −1 (RN ) .8) ε P1 u ε − θ ε L2 (RN ) → 0. Next.15) θε. respectively.14) θε. . .11) (6. Obviously.5).δ (x) = ε.7) (6.14) easily follows if we use (1.
13) is completely analogous. we have (6.δ 2 L2 (RN ) = ξ∈ε−1 Y ξ>δε−1 u∗ (ξ)2 dξ ≤ cδ ε2 ξ∈ε−1 Y ξ>δε−1 ξ−2 f (ξ)2 dξ. H −1 (RN ) The proof of (6. we consider the energies in ξ ≤ δε−1 . Next. ξ)dξ 1 → 0. the ﬁrst term in the L2 norm is bounded above by uε − u∗ L2 (RN ) which. By Lemma 2.21) 1 2 + c λε (ξ) λε (ξ)−1 − Dk λ1 (0)ξk ξ 1 1 2 −1 ξ≤δε −1 2 f (ξ)2 dξ.1194 C. Thus we get. VANNINATHAN If f ∈ L2 (RN ). To prove (6. That the second term converges to zero in L2 (RN ) is proved in Proposition 5. RN RN Combining (6.12). CONCA. 1 ξ≤δε−1 To estimate the above integral.19) ωε L2 (RN ) ω ε (x) = ξ≤δε−1 ε (B1 uε (ξ) − u∗ (ξ))eix·ξ φε (x. we write the integrand as ε ε B1 uε (ξ) − u∗ (ξ) = λε (ξ)−1 (B1 f (ξ) − f (ξ)) + λε (ξ)−1 − 1 1 1 2 D λ1 (0)ξk ξ 2 k −1 f (ξ).18). ORIVE.1. we turn are attention to the proof of (6. . let us deﬁne (6.18) (6.15) and (6. we easily get (6.δ 2 L2 (RN ) ≤ cδ ε2 ξ∈ε−1 Y ξ>δε−1 (1 + ξ2 )−1 f (ξ)2 dξ = cδ ε2 f 2 . L2 (RN ) ω ε H 1 (RN ) ≤ cε f . We now show (6. To this end.16) ξ4 u∗ (ξ)2 dξ ≤ c f (ξ)2 dξ.16). by our hypothesis. This clearly implies θε. since u∗ and f are related by the homogenized equation A∗ u∗ = f in RN . AND M. By Parseval’s identity. R.24). By Parseval’s equality. Step 3.1.11). converges to zero. using (1.19). we decompose the integrand as follows: ε ε ε B1 uε − u∗ = B1 (uε − u∗ ) + (B1 u∗ − u∗ ). then it is well known that u∗ ∈ H 2 (RN ) and (6.17) and show that (6. we have θε. that ε B1 f (ξ) − f (ξ)2 ε2 ω  dξ H 1 (RN ) ≤ c ξ2 ξ≤δε−1 (6.20) ω ε 2 1 (RN ) ≤ c H ε λε (ξ)B1 uε (ξ) − u∗ (ξ)2 dξ. Now.
Asymptotic expansion of the Bloch approximation. as 2 λε (ξ) − 1 Dk λ1 (0)ξk ξ 2 ∗ 1 2 u (ξ)2 dξ.24). we represent the integrand as ε B1 f (ξ) − f (ξ) = ξ RN f (x)e−ix·ξ (φε (x. ξ) − φε (x.26) zε 2 L2 (RN ) ≤ cε2 ξ≤δε−1 u∗ (ξ)2 ξ2 dξ ≤ cε2 u∗ 2 H 1 (RN ) . and we obtain (6. Proof of (i). We see then that zε 2 L2 (RN ) = ξ∈ε−1 Y ξ≤δε−1 u∗ (ξ)2 φε (·.21). In this concluding section. In fact. η) − φ1 (·.22). we see that the above integral is estimated from above by cε2 ξ≤δε−1 ξ4 u∗ (ξ)2 dξ ≤ cε2 f 2 . L2 (RN ) ξ≤δε−1 ε B1 f (ξ) − f (ξ)2 dξ ≤ cε2 f ξ2 The second term on the righthand side of (6.11. The same proof shows that the third term admits a bound (6.3. we must proceed diﬀerently. 0))dξ. L2 (Y ) Using the Lipschitz continuity of the map η → φ1 (·.11)). u∗ (ξ)eix·ξ (φε (x.9).21) can be rewritten. .δ (x) = ξ∈ε−1 Y ξ≤δε−1 θε (x) − u∗ (x) = z ε (x) + θε. η) ∈ L2 (Y ) for η ≤ δ. The second term has already been estimated in the L2 norm (see (6. ε (ξ) λ1 ξ≤δε−1 Using (6.19) and hence the result. ξ) − φε (x.BLOCH APPROXIMATION 1195 To estimate the ﬁrst term on the righthand side of (6. 0) 1 1 2 dξ. 1 (2π)N/2 ξ>δε−1 and θε.23) (6. 1 1 u∗ (ξ)eix·ξ dξ. using the homogenized equation. ξ L2 (Y ) Applying Lemma 2. L2 (RN ) This establishes (6.3.δ (x) + u∗. 0) ≤ cη for η ≤ δ. ξ) − φε (·.25) u∗. We have the following decomposition: (6.24) z ε (x) = u∗. it is essential to use Lemma 2. we get 2 .22) where (6. 0)) 1 1 dx. To estimate the ﬁrst term on the righthand side of (6.5 and using φ1 (·.δ (x).6) and (1.δ is deﬁned in (6.δ L2 (RN ) ≤ cε f H −1 (RN ) ≤ cε u∗ H 1 (RN ) . 6. we prove Theorem 1. we see that the above integral can be majorized.
ξ) − φε (x. Further.15).23) for z ε . we can easily derive the estimate (6.22). For θε.12). The estimate on the second term on the righthand side of (6. 0) − ip(0) χk (·)ηk H 1 (Y ) ≤ cη2 for η ≤ δ. We get z ε (x) − εχε (x) k ξ≤δε−1 u∗ (ξ)eix·ξ φε (x.29) z ε − εχε ∂u∗ k ∂xk H 1 (RN ) ≤ cε f L2 (RN ) . VANNINATHAN This ﬁnishes the proof of (i). For u∗.δ 2 H 1 (RN ) ≤c ξ>δε−1 ξ2 u∗ (ξ)2 dξ ≤ cδ ε2 f 2 . Thanks to (6.22) for (θε − u∗ ) in terms of z ε .3. and combine it with the representation (6. it suﬃces to prove (6.29) and hence (ii). CONCA. Consider again the decomposition (6.1196 C.δ . AND M. We note that we cannot. R. We ﬁnally get z ε − εχε ∂u∗ k ∂xk 2 H 1 (RN ) ≤ cε2 RN ξ4 u∗ (ξ)2 dξ. we use (6.30).9) and (6. 0) − ip(0) χε (x)εξk dξ 1 1 k (6.δ L2 (RN ) ≤ cε2 f L2 (RN ) . ORIVE. in general. and u∗. θε. Proof of (iii).δ .28) u∗. This completes the proof of (6.δ . we use the representation ∂u∗ 1 (x) = ∂xk (2π)N/2 ∂u∗ (x) = ∂xk RN (iξk )u∗ (ξ)eix·ξ dξ. we are reduced to obtaining the estimate (6. To this end.27) θε − u∗ − εχε ∂u∗ k ∂xk H 1 (RN ) ≤ cε f L2 (RN ) .30) is more straightforward.32) (6. we have the estimates (6. we appeal to Lemma 2. we use once again the decomposition (6. we have the estimate (6.δ . θε.δ H 1 (RN ) ≤ cε2 f H 1 (RN ) . To this end. Proof of (ii).33) θε. .31) φ1 (·. L2 (RN ) Thus. assert that u∗ H 1 (RN ) ≤ c f H −1 (RN ) as we are working on the entire space RN . Because of (i).30) − ξ>δε−1 ip(0) χε (x)εξk u∗ (ξ)eix·ξ dξ. η) − φ1 (·. k To estimate the ﬁrst term on the righthand side of (6.
Acknowledgments. supported by the FONDAP’s Programme on MathematicalMechanics. Orive and M.3. The proof is concluded via a simple application of Lemma 2. u∗. it is enough to obtain the estimate (6.34) (6.δ H 1 (RN ) ≤ cε2 f H 1 (RN ) . H 1 (RN ) We can write z ε (x) − εχε (x) k ∂u∗ ∂ 2 u∗ (2) (x) + ε2 (χε (x) + βk ) (x) k ∂xk ∂xk ∂x = ξ≤δε−1 u∗ (ξ)eix·ξ [φε (x. 0) − ip(0) χε (x)εξk i 1 1 k + p(0) (χε (x) + βk )ε2 ξk ξ ]dξ k (2) − (6.30) that (6. k (2) The analysis of the righthand side of (6. Thus.δ L2 (RN ) ≤ cε2 f L2 (RN ) . it is clear from the representation (6. This research stems from a couple of visits by R.37) θε − u∗ − εχε ∂u∗ ∂ 2 u∗ (2) + ε2 (χε + βk ) k ∂xk k ∂xk ∂x ≤ cε2 f H 1 (RN ) .36) z ε − εχε ∂u∗ k ∂xk L2 (RN ) ≤ cε2 f L2 (RN ) .40) φ1 (·.39) is similar to that of (6.35). η) − φ1 (·. we are reduced to showing that (6. H 1 (RN ) Thanks to (6. Vanninathan to the Center of Mathematical Modelling and to the Department of Mathematical Engineering of the University of Chile.10. These authors warmly thank these institutions.35) u∗. The new information needed is the following: (6.30).BLOCH APPROXIMATION 1197 Similar techniques imply (6.38) z ε − εχε ∂u∗ ∂ 2 u∗ (2) + ε2 (χε + βk ) k ∂xk k ∂xk ∂x ≤ cε2 f H 1 (RN ) . ξ) − φε (x. . which is a simple consequence of Proposition 1. 0) − ip(0) χk (·)ηk + p(0) (χk (·) + βk )ηk η (2) H 1 (Y ) ≤ cη3 for η ≤ δ.39) + ξ>δε−1 ip(0) χε (x)εξk u∗ (ξ)eix·ξ dξ k ξ>δε−1 p(0) (χε (x) + βk )ε2 ξk ξ u∗ (ξ)eix·ξ dξ. On the other hand.33) and (6.
[15] T. Rech. [10] C. Perturbation Theory for Linear Operators. Vanninathan. Wilcox. Appl. pp. pp. Markowich. Phys. A dispersive eﬀective medium for wave propagation in periodic composites. New York. [13] G. and M. Poupaud.. Analyse Fonctionnelle. Allaire and C. 323–379. 5 (1964). Conca. Santosa and W. Anal. 131 (2001). Phys. [3] G. Conca. 33 (1978). Cherkaev and R. Anal. Conca. eds. Mauser. pp. 1977. Asymptotic Analysis for Periodic Structures. 57 (1997). [16] F. Soc. Tartar. [21] L. Papanicolaou. Tartar. Berlyand. 50 (1997). A. pp. in Topics in the Mathematical Modeling of Composite Materials. Probl`mes d’homog´n´isation dans les ´quations aux d´riv´es partielles. Math. Anal. Proc. Conca and M. Vanninathan. Bensoussan. Comm. R. 31. manuscript. Frequencydependent acoustics of composites with interfaces. pp. SIAM J. SIAM J. 20 (1989). SIAM J. [20] F. Allaire and C. Appl. Bloch. Gerard. P. 1997. J. 52 (1928). Conca and M.L. 984–1005. [8] C.. Boston. Berlin. Math. Ann. SpringerVerlag. Homogenization and twoscale convergence. Sur les ´quations diﬀ´rentielles lin´aires a coeﬃcients p´riodiques. 146–167. ORIVE. 38. Birkh¨user. Paris. Vanninathan. Appl.. AND M. J. Fluids and Periodic Structures. 1999. pp. [5] A. Vanninathan. Kato. A general convergence result for a functional related to the theory of homogenization. [2] G. Anal. Collection Math´matiques Ape e pliqu´es pour la Maˆ e ıtrise. eds. Edinburgh Sect. Kluwer Academic Publishers. Math. Paris. Homogenization of periodic structures via Bloch decomposition. Ration. 1978. 608–623. 1499–1504. 12 (1883).. SIAM J. N. Lions. 2. Bloch Approximation in Bounded Domains. 1482–1518.. 1983. Hazewinkel et al. 343–379.. [11] C. [18] F. Bloch waves. pp. L. e [22] C. [19] M. Avellaneda. Math. Progr. Arch. Reed and B. a [17] G. 1639–1659. 1978. On uniform H 2 estimates in periodic homogenization. 72–74. Amsterdam.. J.. pp. e ´ [14] P. New York. John Wiley/Masson. pp. A. Mech.A. M. 1966. Allaire. pp.1198 C. Th´orie et Applications. pp. Conca. 197–257. Z. Bloch wave homogenization for a spectral problem in ﬂuidsolid structures. Orive. R.. Nonlinear Diﬀerential Equations Appl. 29 (1998). Math. in Encyclopaedia Mathematics. Symes. and F. Kohn. pp. in Cours e e e e e e Peccot au Coll`ge de France. Partial diﬀerential equations with periodic coeﬃcients and Bloch waves in crystals. Appl.. Ecole e e e ` e Norm. and G. VANNINATHAN REFERENCES [1] G. 555–600. Roy. Math.F. Clouet. Homogenization limits and Wigner transforms. Murat and L. Masson. pp. 47–89. Academic Press. Methods of Modern Mathematical Physics. Planchard. Nguetseng. 21–44. Floquet. W. J. Odeh and J. and J... Brezis. [7] H. and M. Amsterdam. 1995. [9] C. Math. [4] M. pp. SIAM J. [12] C. . 135 (1996).J. Keller. Theory of Bloch waves. ¨ [6] F. Simon. 2001.. Math. S´r.. 499–517. 60 (2000).. Pure Appl. Math. 51 (1991). Boundary layers in the homogenization of a spectral problem in ﬂuidsolid structures. Hconvergence. SIAM J. 2143–2181. Math. Conca. CONCA. Uber die quantenmechanik der electronen in kristallgittern. NorthHolland. pp. 23 (1992). Anal.