An Introduction to Homogenization

and G-convergence
Anneliese Defranceschi

This paper contains the notes of ve lectures concerning an introduction to Homogenization and G-
convergence, delivered on September 6-8, 1993 as a part of the \School on Homogenization" at the ICTP,
Trieste. The main topics treated are the following ones:
I. Homogenization of second order linear elliptic operators
II. Homogenization of monotone operators
III. G-convergence; H-convergence.
0. Introduction
Composite materials (bred, stratied, porous,: : :) play an important role in many branches of Mechan-
ics, Physics, Chemistry and Engineering. Typically, in such materials, the physical parameters (such as
conductivity, elasticity coecients, : : :) are discontinuous and oscillate between the dierent values charac-
terizing each of the components. When these components are intimately mixed, these parameters oscillate
very rapidly and the microscopic structure becomes complicated.
On the other hand we may think to get a good approximation of the macroscopic behaviour of such
a heterogeneous material by letting the parameter "h , which describes the neness of the microscopic
structure, tend to zero in the equations describing phenomena such as heat conduction and elasticity. It
is the purpose of homogenization theory to describe these limit processes, when "h tends to zero.
More precisely, homogenization deals with the asymptotic analysis of Partial Dierential Equations
of Physics in heterogeneous materials with a periodic structure, when the characteristic length "h of the
period tends to zero.
A good model for the study of the physical behaviour of a heterogeneous body with a ne periodic
structure, e.g. in electrostatics, magnetostatics, or stationary heat diusion is given by
 ;div(a( x )Duh ) = f on
;
(0:1) "h
uh j@
= 0 on @
;
here
is a bounded open subset of Rn which will be considered as a piece of the heterogeneous material
and "h is the period of the structure in all directions. The function uh can be interpreted as the electric
potential, magnetic potential, or the temperature, respectively. The coecients a( "x ) = (aij ( "x )) are
h h
"h -periodic functions and describe the physical properties of the dierent materials constituting the body
(they are the dielectric coecients, the magnetic permeability and the thermic conductivity coecients,
respectively). The function f is a given source term.
When the period of the structure is very small, a direct numerical approximation of the solution
to (0.1) may be very heavy, or even impossible. Then homogenization provides an alternative way of
2 A. Defranceschi

approximating such solutions uh by means of a function u which solves the problem corresponding to a
\homogenized" material
 ;div(bDu) = f on
;
(0:2)
uj@
= 0 on @
;
where b is a constant matrix (for a homogeneous material the physical properties does not depend on x).
The \homogenized" matrix b may be interpreted as the physical parameters of a homogeneous body, whose
behaviour is equivalent, from a \macroscopic" point of view, to the behaviour of the material with the
given periodic microstructure (described by (0.1)) (the coecients bij are called also \eective" coecients
or \eective" parameters since they describe the macroscopic properties of the medium).
In these lectures we shall consider the asymptotic analysis of the solutions to
 ;div(ah(x; Duh)) = f on
;
uh j@
= 0 on @

in the following main cases:

(i) ah (x; ) = a( "x ) , with a(x) periodic matrix;
(ii) ah (x; ) = a( "x ; ), with a(
; ) periodic;
h

(iii) ah (x; ) = ah (x) , without periodicity assumptions on ah :

The main references for the homogenization theory of periodic structures are the books by Bensoussan-
Lions-Papanicolaou [9], Sanchez Palencia [40], Lions [31], Bakhvalov-Panasenko [6], and Oleinik-Shamaev-
Yosian [37]. Other general references for the theory of the homogenization of partial dierential equations
are Babuska [4], Bensoussan [7], and Bergman-Lions-Papanicolaou-Murat-Tartar-Sanchez Palencia [10].

Notation
Let n 2 N be xed. Given m 2 N , the elements of Rm are usually considered as column vectors,
(
;
) denotes the scalar product on Rm , and j
j will be the usual euclidean norm. Let M nn be denote
the set of all n  n realPmatrices. Given M = (Mij ) 2 M nn and P2 Rn , M is the vector of Rn with
components (M)i = nj=1 Mij j ; i 2 f1; : : :; ng and (M; ) = ni;j =1 Mij j i for every ,  2 Rn .
We shall identify M nn with Rn2 .
We use the symbol jAj to the denote the Lebesgue measure of the set A  Rn . The notation a.e.
stands for almost everywhere with respect to the Lebesgue measure.
For every open subset A R2 Rn and f 2 L1(A) we denote by MA (f) the average of f (with respect
to A) dened as MA (f) = jA1 j A f(x) dx : If no confusion can occur, we shall simply write M(f).
For what concerns Lp and Sobolev spaces we refer to the Appendix.

I. Homogenization of second order linear elliptic operators

 An Introduction to Homogenization and G-convergence 3

1. Setting of the problem

Let z1 ; : : :; zn be n linearly independent vectors of Rn , and let P be the parallelogram with sides
z1 ; : : :; zn , i.e.,
P = ft1z1 + : : : + tn zn : 0 < ti < 1 for i = 1; : : :; n g :
We say that a function ' : Rn ! Rm is P -periodic if '(x) = '(x + zi ) for every x 2 Rn and for every
i = 1; : : :; n. In this case we say that P is a periodicity cell of the function '. For the sake of simplicity
(without loss of generality) we shall assume from now on that the periodicity cell P is the unit cube
Y = ]0; 1[n . Hence, if e1 ; : : :; en denotes the canonical basis of Rn then ' : Rn ! Rm is Y -periodic if
'(x) = '(x + ei ) for every x 2 Rn and for every i = 1; : : :; n.
Let us consider the function a : Rn ! M nn , with a(x) = (aij (x)) for x 2 Rn , satisfying the
following properties:
(1.1) aij is Y -periodic on Rn for every i; j = 1; : : :; n;
(1.2) aij 2 L1 (Rn) for every i; j = 1; : : :; n; P
(1.3) there exists a constant  > 0 such that (a(x); ) = ni;j =1 aij (x)j i  j j2 for a.e. x 2 Rn and
for every  2 Rn .
We then dene ah : Rn ! M nn by
(1:4) ah (x) = a( "x ) ;
h
where ("h ) is a sequence of positive real numbers converging to 0. Note that the functions ahij are
"h Y -periodic on Rn .
Let
be a bounded open subset of Rn (we shall consider it as a piece of a heterogeneous material).
For a xed "h > 0, let us consider the Dirichlet boundary value problem on

 ;div(ah(x)Duh ) = f on
;
(1:5)
uh j@
= 0 ; on @
;
where f is a given smooth function on
. Assume f 2 H ;1;2(
). The variational (weak) formulation of
(1.5) becomes then: nd uh 2 H01;2(
) such that
(R h 1 ;2
(1:6)
(a (x)Duh ; Dv) dx = hf; vi for every v 2 H0 (
)
uh 2 H01;2(
)
(note that this presentation does not require regularity assumptions for the functions aij . Moreover, we
get a priori estimates on uh which are independent of "h and are not based on the regularity of the
coecients).
Let us note that by the Lax-Milgram lemma (see Lemma A.3.1) we have existence and uniqueness
of a solution to (1.6). Indeed, let Zus dene the bilinear form ah1 : H01;2(
)  H01;2(
) ! R by
ah1 (u; v) = (ah (x)Du; Dv) dx for every u, v 2 H01;2(
) :

We observe that the boundedness assumption (1.2) and Holder's inequality yield immediately
jah1 (u; v)j  ckukH01 2 (
)kvkH01 2 (
) for every u, v 2 H01;2(
)
; ;

(take into account Remark A.1.15). Moreover, the ellipticity condition (1.3) ensures that
ah1 (u; u)  kuk2H01 2(
) for every u 2 H01;2(
) :
;

Hence, ah1 denes a bilinear continuous and coercive form on H01;2(

) and the existence and uniqueness
of a solution to (1.6) is guaranteed.
4 A. Defranceschi

Remark 1.1. Instead of the Dirichlet boundary conditions in (1.5) one can consider also more general
boundary conditions; for example, Neumann boundary conditions or mixed boundary conditions. However,
to x the ideas about homogenization, we will consider for the moment Dirichlet boundary conditions.
Let us come back to the Dirichlet boundary value problem
 ;div(ah(x)Duh ) = f on
;
(1:7)
uh 2 H01;2(
) :
We can associate to ah the (second order elliptic) operator Ah : H01;2(
) ! H ;1;2(
) dened by
Ah u = ;div(ah Du) ;
and (1.7) can be written also in the form
 Ah uh = f on
;
(1:8)
uh 2 H01;2(
) :
Now, let us consider the sequence (uh ) of solutions to (1.8) corresponding to the sequence ("h ). Let us
note that our assumptions on ah guarantee that
kuh kH01 2 (
)  c ;
;

where c is a constant independent of h (for more details see Section 2 and Section 4). Therefore, there
exist a subsequence (u(h) ) of (uh ) and a function u0 2 H01;2(
) such that
u(h) * u0 weakly in H01;2(
) :
At this point it is natural to ask whether u0 solves a boundary value problem of the type (1.8), i.e.,
 ;div(b(x)Du0) = f on
;
u0 2 H01;2(
) :
The aim of the next sections is to answer at this question. We shall construct a second order elliptic
operator B such that (uh ) converges to u0 (in an appropriate topology), where u0 is the solution to
 Bu0 = f on
;
(1:9)
u0 2 H01;2(
) ;
with Bu = ;div(b(x)Du) : The operator B is the so called homogenized operator of the family (Ah ) and
b(x) the homogenized coecients.
As pointed out in the introduction, this convergence analysis is related to the problem of nding the
physical properties of a homogeneous material, whose overall response is close to that of the heterogeneous
periodic material (whose physical properties are described by (1.4)), when the size "h of the periodicity
cell tends to 0.
The problem of passing to the limit in (1.7), when "h approaches to 0, is rather delicate (as we will
see soon) and requires the introduction of new techniques. The main diculty lies in the passage to the
limit in (ah (x)Duh ), which is the product of only weakly convergent sequences.

Before attacking the study of the general case, let us consider a simple particular case.
 An Introduction to Homogenization and G-convergence 5

2. An example in dimension 1
Let
= ]x0 ; x1[  R and f 2 L2 (
). Let ("h ) be a sequence of positive real numbers converging
to 0 and let ah (x) = a( "x ), where a : R ! R is a measurable Y -periodic function satisfying
h

(2:1) 0 <   a(x)   < +1 a.e. on R :

We consider the Dirichlet boundary value problems
 ; d (ah (x) du (x)) = f in
;
h

(2:2) dx dx
uh (x0 ) = uh (x1 ) = 0
(for every "h > 0, (2.2) is, for example, the stationary heat equation in a 1-dimensional "h Y -periodic
medium). The weak formulation is then
(R R
h duh dv

a dx dx dx =
fv dx for every v 2 H01;2(
) ;
(2:3)
uh 2 H01;2(
) :
As seen in the previous section, for every xed "h , there exists a unique solution uh 2 H01;2(
) to problem
(2.3). By taking v = uh in (2.3) and using Holder's inequality we get
Z duh
ah ( dx )2dx  kf kL2 (
) kuh kL2 (
) :

Using (2.1) and the Poincare inequality (see Theorem A.1.14 and Remark A.1.15) we obtain
(2:4) kuh kH01 2 (
)  c kf kL2 (
) ;
;

where c is a positive constant depending only on

. Hence the sequence (uh ) is uniformly bounded in
H01;2(
). Therefore (see Theorem A.1.3 and Theorem A.1.11) there exist u0 2 H01;2(
) and a subsequence,
still denoted by (uh ), such that
(2:5) uh * u0 weakly in H01;2(
) :
Moreover, the periodicity assumption on a (see Theorem A.1.18) yields
 1 Z
(2:6) a * M(a) = jY j a(y) dy in L1 (
) weak* (hence weakly in L2 (
))
h
Y
(note that jY j = 1). From (2.5), (2.6), and (2.2) it is tempting to believe that in the limit one has
 ; d (M(a) du ) = f
0 in
;
(2:7) dx dx
u0(x0 ) = u0 (x1) = 0 :
But this is false in general, being  h (x)  ah (x) dudx (x) the product of two sequences converging both
h

weakly (see Remark A.1.19). To obtain the correct answer let us proceed as follows: note that by the
boundedness of ah in L1 (
) and (2.4) we have that  h is uniformly bounded in L2 (
), and by (2.2)
satises
h
(2:8) ; ddx = f in
:
6 A. Defranceschi

Hence,  h is uniformly bounded in H 1;2(

). Since the injection H 1;2(
) ,! L2 (
) is compact (see
Theorem A.1.12), it follows that one can assume (at least passing to a subsequence) that

(2:9)  h !  0 strongly in L2 (
)

so that
1 h 1 0
ah  * M( a ) weakly in L (
):
(2:10) 2

(Note that 0 < 1  a1  1 < +1 . Moreover the periodicity assumption on a implies by Theorem
A.1.18 that ( a1 ) converges in L1 (
) weak* to M( a1 ) and 1  M( a1 )  1 .) But
h

1 h duh
ah  = dx ;
so that (2.5) and (2.10) imply
du0 = M( 1 )  0 :
dx a
By passing to the limit in the sense of distributions in ; ddx = f we have ; ddx = f , so that u0 is the
h 0

solution to the Dirichlet boundary value problem

(; d 1 du0
(2:11) dx ( M( a1 ) dx ) = f in
;
u0 (x0) = u0(x1 ) = 0 :

The homogenized operator B associated to Ah is given by

2
(2:12) B = ; M1( 1 ) dx
d :
2
a

Note that in this case the homogenized operator is related to the harmonic mean (and not to the arithmetic
mean) of a (compare (2.7) with (2.12)). Finally, by the uniqueness of the solution to (2.11) it follows that
the whole sequence (uh ) converges weakly in H01;2(
) to u0 , without extracting a subsequence (use the
Urysohn property).

Remark 2.1. Let us note that

M(a)  M1( 1 )
a
with
R pastrict inequality in general. This fact follows immediately by the Holder inequality applied to
p1 dy (recall that 0 <   a(x)   < +1 for a.e. x 2 R ).
Y a
 An Introduction to Homogenization and G-convergence 7

3. Asymptotic expansions using multiple scales

For every h 2 N , let ah : Rn ! M nn be the function dened by (1.4). In order to study the
asymptotic behaviour of the solutions uh to
 Ahuh  ;div(ah (x)Duh ) = f on
;
(3:1)
uh j@
= 0 on @
;
where f 2 L (
) , an ecient technique consists in applying asymptotic expansions using multiple scales
2
(i.e., \slow" and \fast" variables). More precisely, the heuristic device is to suppose that uh has a two-scale
expansion of the form
(3:2) uh (x) = u0(x; "x ) + "h u1(x; "x ) + "2h u2 (x; "x ) + : : : ;
h h h
where the functions ui (x; y) are Y -periodic in y for every x 2
. This means that we postulate the
existence of smooth functions ui(x; y) dened on
 Rn , Y -periodic in y and independent of "h such
that the right hand side of (3.2) is an asymptotic expansion of uh (as well as its derivatives).
Let us note that the two variables x and "x take into account the two scales of the homogenization
phenomenon; the x variable is the macroscopic variable, whereas the "x variable takes into account the
h

h
\microscopic" geometry.
The method we are going to develop turns out to be very useful to obtain the right answers in
the study of the limit behaviour of the solutions to problem (3.1) (but also for more general cases). The
proof of the correctness of the formulas obtained by this method can sometimes be made directly, but in
general other tools will be needed (for example, the use of particular test functions and a compensated
compactness lemma).
Remark 3.1. It is technically complicated to keep track of the boundary conditions when seeking uh in
the form (3.2) and this is actually the source of serious technical diculties in justifying the method. The
method will nevertheless give the \right answer" because it will turn out that, in this sort of problems,
the boundary conditions are somewhat irrelevant.
The idea of the method is (simply) to insert (3.2) in equation (3.1) and to identify powers of "h . In
order to present these computations in a simple form, given a smooth function (x; y) of two variables,
dene the function h (x) of one variable by
h (x) = (x; "x )
h
and note that @h (x) = ; @ + 1 @
;x; x
:
@xi @xi "h @yi "h
Then, one can write
(3:3)
 
Ah h = (";h 2 A0 + ";h 1 A1 + "0h A2) (x; "x ) ;
h
where
X
A0 = ; @y@ ;aij (y) @y@

n

i;j =1 i j
X n @ ;a (y) @
; Xn @ ; @

A1 = ; ij a ij (y)
i;j =1 @xi @yj i;j =1 @yi @xj
X
A2 = ; @x@ ;aij (y) @x@
:
n

i;j =1 i j
8 A. Defranceschi

By using (3.2) and (3.3), the equation (3.1) becomes, under the assumptions that a and the ui(x; y) are
smooth,
(3:4) (A0 u0)(x; "x ) = 0 on
;
h

(3:5) (A0 u1 + A1u0 )(x; "x ) = 0 on

;
h

(3:6) (A0 u2 + A1 u1 + A2 u0)(x; "x ) = f(x) on

;
h
of course one can (formally) proceed:
(A0 u3 + A1 u2 + A2 u1 + A3u0 )(x; "x ) = 0 on
etc.
h
Let us see that the homogenized operator can be constructed from (3.4), (3.5) and (3.6), which will be
done in the sequel.
Because of the Y -periodicity of ui (x;
), around any point x the function z 7! ui(z; "z ) behaves
like z 7! ui (x; "z ) : Hence we shall determine ui by means of the following problems where x is now a
h

xed parameter:
 (A0u0)(x;
) = 0 on Y ;
(3:7)
u0(x;
) Y -periodic (i.e., u0 (x;
) has the same values on the opposite faces of Y );
 (A0u1)(x;
) = ;(A1u0)(x;
) on Y ;
(3:8)
u1(x;
) Y -periodic;
 (A0u2)(x;
) = f(x) ; (A1u1 + A2u0)(x;
) on Y ;
(3:9)
u2(x;
) Y -periodic.
Let us consider problems (3.7)-(3.9) in the framework of weak solutions.
Let us start by proving an existence result for a boundary value problem on the unit cube. Let
Hper (Y ) denote the subset of H 1;2(Y ) of functions u which have the same trace on the opposite faces of
1 ; 2
Y . Moreover, we denote by H]1;2(Y ) the subset of H 1;2(
) of all the functions u with mean value zero
which have the same trace on the opposite faces of Y .

1;2 (Y )) . Then

Lemma 3.2. Let F 2 (Hper
(R (a(y)D'; D ) dy = hF; i for every 2 Hper
1;2 (Y ) ;
(3:10) Y
' 2 H]1;2(Y )
admits a unique solution if and only if
(3:11) hF; 1i = 0 :
 An Introduction to Homogenization and G-convergence 9

R
Proof. Condition (3.11) is clearly necessary since Y (a(y)D'; D ) dy = 0 if is constant. Note that
1;2
H] (Y ) is a closed subset of H 1;2(Y ),
and therefore a Hilbert space. Moreover, by the Poincare-
Wirtinger inequality (see Theorem A.1.14) kDvkL2 (Y ;R ) denes a norm on H]1;2(Y ) equivalent to the
n

norm kvkH 1 2 (Y ) . Let us consider a1 : H]1;2(Y )  H]1;2(Y ) ! R dened by

;

Z
a1 ('; ) = (a(y)D'; D ) dy :
Y
Clearly, a1 is a bilinear form. Moreover, by the boundedness assumption of a it follows that
ja1('; )j  ckD'kL2 (Y ;R ) kD kL2 (Y ;R
n n )

for every ', 2 H]1;2(Y ) : Therefore, a1 is continuous. Moreover, the ellipticity condition satised by
a implies immediately that a1 is coercive. Therefore, by the Lax-Milgram lemma there exists a unique
function ' 2 H]1;2(Y ) satisfying
Z
(a(y)D'; D ) dy = hF; i for every 2 H]1;2(Y ) :
Y
Since hF; 1i = 0 it turns out that ' satises also
Z
(a(y)D'; D ) dy = hF; i for every 2 Hper
1;2(Y ) :
Y
Hence, (3.10) admits a unique solution in H]1;2(Y ).
Let us apply this lemma to the solution of (3.7), (3.8) and (3.9).
Step 1: Study of (3.7).
Let us look for a solution to the problem (3.7), i.e.,
 (A0u0)(x;
) = 0 on Y ;
u0 (x;
) Y -periodic
(note that the periodicity condition plays the role of boundary conditions). By using the Green formula
and by taking into account the periodicity assumptions, one proves easily that problem (3.7) is equivalent
to the following one: nd u0(x;
) 2 Hper
1;2 (Y ) such that
Z
(3:12) (a(y)Du0 ; D ) dy = 0 for every 2 Hper
1;2 (Y ) :
Y
Let us recall that x plays the role of a parameter (hence Du0 is the gradient with respect to y). By
Lemma 3.2 we can conclude that u0(x;
) 2 Hper
1;2(Y ) is determined by (3.12) up to a constant. By taking
= u0(x;
) in (3.12) and by using the ellipticity condition satised by a, it follows immediately that
u0(x;
)=costant, i.e.,
(3:13) u0 (x; y) = u0 (x) :
10 A. Defranceschi

Step 2: Study of (3.8).

Using (3.13), problem (3.8) reduces to
( P ;
0 (x) on Y ;
(A0 u1)(x;
) = ni;j =1 @y@ aij (
) @u
@x
(3:14) i j

u1(x;
) Y -periodic.
This is still a problem in y, where x is a parameter. Due to the separation of variables on the right hand
side of (3.14), we are able to represent u1 in a simple form. Let us note that by Green's formula and the
periodicity assumptions the weak formulation of (3.14) becomes: nd u1 (x;
) 2 Hper
1;2 (Y ) such that
Z n @u Z
X 0
(3:15) (a(y)Du1 ; D )dy = ; @xj aij Di dy for every 2 Hper (Y ) :
1;2
Y i;j =1 Y
Let us consider for k = 1; : : :; n the problem
(R (a(y)Dwk (y); D (y)) dy = ; R (a(y)e ; D (y)) dy for every 2 H]1;2(Y ) ;
Y Y k
(3:16) 1; 2
w 2 H] (Y ) :
k
R
Note that the function F k : H]1;2(Y ) ! R dened by 7! F k ( ) = ; Y (a(y)ek ; D ) dy is a linear and
continuous function on H]1;2(Y ). By the Lax-Milgram lemma for every k = 1; : : :; n there exists a unique
solution to (3.16). Then the general solution to (3.15) becomes
Xn
(3:17) u1 (x; y) = wk (y) @u@x0 (x) + ue1 (x) ;
k=1 k
where ue1 is an additive constant (function of the parameter x).
Step 3: Study of (3.9).
We now consider (3.9) where we think of u2 as the unknown, x being a parameter. Let us consider the
1;2(Y ) ! R dened by
function F : Hper
Z n @ Z
X @u1
hF; i = f(x)@xi (y) dy + @xi aij (y) @yj dy
Y i;j =1 Y

n Z
X Z
1 @ dy + X @
n
; aij (y) @u
@xj @yi a ij (y) @u0 dy :
i;j =1 Y i;j =1 @xi Y @xj
We note that problem (3.9) is equivalent (the proof is analogous to the previous ones) to nd u2 (x;
) 2
1;2 (Y ) such that
Hper Z
(a(y)Du2 ; D ) dy = hF; i for every 2 Hper 1;2 (Y ) :
Y
By virtue of Lemma 3.2, u2 exists if and only if
(3:18) hF; 1i = 0 :
Condition (3.18) is the homogenized equation we are looking for. Indeed, by taking = 1 and the
expression of u1 into account, (3.18) becomes
Xn 2 u0
(3:19) ; bik @x@ @x = f on
;
i;k=1 i k
 An Introduction to Homogenization and G-convergence 11

where
Z X
n k
(3:20) bik = jY1 j (aik (y) + aij (y) @w@x(y) ) dy
Y j =1 j
(recall that jY j = 1 and will be therefore successively omitted). The equation (3.19) is the homogenized
equation and the coecients bij are the homogenized coecients. We will prove (see Proposition 4.2) that
the homogenized matrix is symmetric if a has this property and satises an ellipticity condition like a.
Finally, to obtain a well posed problem for u0 , we only need a boundary condition for u0 . From (3.1) and
(3.2) we obtain u0j@
(x) = 0 on @
. Note that this relation is formal, but it will be rigorously proved
below.

Remark 3.3. Let us note that the preceeding considerations can be summarized as follows: if we postulate
an expansion of the form (3.2), the rst term u0 is determined as a solution to the equation (3.19) with
the boundary condition u0 j@
(x) = 0 on @
. The formal rule (which will be justied below) to compute
the homogenized coecients is as follows:
i) solve (3.16) on the unit cell Y , for k = 1; : : :; n;
ii) bik is given by (3.20).
We shall prove in Theorem 4.1 that (uh ) converges (in an suitable topology) to the function u0 given
above.

Remark 3.4. Let us conclude with some remarks on the homogenized operator.
(i) In the one-dimensional case one has wk = w solution to
; dyd ;a(y) dw
da(y)
dy = dy ;
hence, a(y) dw
R c
dy = ;a(y) + c. The condition on w to be periodic implies that Y (;1 + a(y) )dy = 0,
i.e.,
;1 + cM( a1 ) = 0 dw = ;1 + c :
dy a(y)
Then, the homogenized coecient has, according to (3.20), the form
Z
b = (a(y) ; a(y) + c) dy = c = M1( 1 ) ;
Y a
and we nd (2.12).
(ii) Let us note that the homogenized coecients have the form
X n k
bik = M(aik ) + M(aij @w @yj ) :
j =1
As we have already seen in the one dimensional case
X
n X
n
ahij Dj uh 6* M(aij )Dj u0
j =1 j =1
and M(aij @w
@y ) appears as a \corrector".
k

j
12 A. Defranceschi

4. Homogenization (symmetric case)

By S] we denote the set of all functions a : Rn ! M nn such that a(x) = (aij (x)), i; j = 1; : : :; n
is Y -periodic and satises the following properties:
(4.1) aij 2 L1 (Rn) for every i; j = 1; : : :; n;
(4.2) aij = aji on Rn for every i; j = 1; : : :; n; P
(4.3) there exists a constant  > 0 such that (a(x); ) = ni;j =1 aij (x)j i  j j2 for a.e. x 2 Rn and
for every  2 Rn .
Given a 2 S] , we consider the following Dirichlet boundary value problems on the bounded open
subset
of Rn :
( ;div(a( x )Du ) = f on
;
" h h
(4:4) h

uh 2 H01;2(
) ;
where (fh ) is a sequence of functions converging strongly in H ;1;2(
) to f , and ("h ) is a sequence of
positive real numbers converging to 0.
In this section we shall prove the convergence, as ("h ) tends to 0, of the solutions uh to (4.4) to
the solution u0 of the following homogenized problem
 ;div(b Du0) = f on
;
(4:5)
u0 2 H01;2(
) :
The constant matrix b = (bij ) is dened by
Z X
n k
(4:6) bik = (aik (y) + aij (y) @w@y(y) ) dy ;
Y j =1 j

where wk is the unique solution to the local problem

(R (a(y)(e + Dwk (y)); Dv(y)) dy = 0 for every v 2 H 1;2(Y ) ;
Y k ]
(4:7)
wk 2 H]1;2(Y ) :
More precisely, we shall present here Tartar's proof (known as the energy method) of the following con-
vergence theorem of De Giorgi and Spagnolo (see [44], [23]). (Note that some homogenization results for
(4.4) are proven also in [4] and [5].)
Theorem 4.1. Let a 2 S] and let ("h ) be a sequence of positive real numbers converging to 0 . Assume
that (fh ) converges strongly in H ;1;2(
) to f . Let uh and u0 be the solutions to (4.4) and (4.5),
respectively. Then,

(4:8) uh * u 0 weakly in H01;2(

) :

(4:9) a( "x )Duh * bDu0 weakly in L2(

; Rn) :
h
 An Introduction to Homogenization and G-convergence 13

Proof. Recall that ah (x) = a( "x ) 2 M nn for every x 2 Rn . The weak formulation of the Dirichlet
h
boundary value problems (4.4) becomes then
(R

(a (x)Duh ; Dv) dx = hfh ; vi
h for every v 2 H01;2(
) ;
(4:10)
uh 2 H01;2(
) :
By taking v = uh in (4.10), and by taking (4.3) into account we have
Z Z
 jDuh j2dx  (ah (x)Duh ; Duh ) dx  kfh kH ;1 2 (
) kuh kH01 2 (
)  ckuh kH01 2 (
) ;
; ; ;

where c is a constant independent of h. By Remark A.1.15 this implies that

(4:11) kuh kH01 2 (
)  C ;
;

where C is a constant independent of h. Consider now the vector in Rn dened as

(4:12)  h (x) = ah (x)Duh (x) on
;
P
i.e., ih (x) = nj=1 ahij (x) @u@x(x) for every i = 1; : : :; n. Since (4.1) and (4.11) hold, we get immediately
h
j

(4:13) k h kL2 (
;R )  C 0 ;
n

where C 0 is a constant independent of h. Therefore, there exist u 2 H01;2(

),   2 L2(
; Rn) and two
subsequences, still denoted by (uh ), ( h ) such that
(4:14) uh * u weakly in H01;2(
) ;

(4:15)  h *   weakly in L2(

; Rn) :
Now, by writing (4.10) in the form
Z
( h ; Dv) dx = hfh ; vi for every v 2 H01;2(
);

we can pass to the limit for any xed v 2 H01;2(

) and we get
Z
(4:16) (  ; Dv) dx = hf; vi for every v 2 H01;2(
)

(note that here the weak convergence in H ;1;2(

) of (fh ) to f would suce). Let us suppose that
(4:17)   (x) = b Du (x) for a.e. x 2
:
Then, (4.16) shows that u 2 H01;2(
) satises the weak formulation of the problem (4.5). By the
uniqueness of the solution to problem (4.5), we may conclude that u = u0 (in Proposition 4.2 we prove
that b satises the same ellipticity conditions as a; hence, the solution of (4.5) is unique). Therefore we
have only to prove (4.17). This will be done by means of the so called \energy method" (developed by
14 A. Defranceschi

L. Tartar) which is based on the introduction of test functions of a special suitable form (let us underline
that they have to be enough to identify the limit problem. As pointed out in other occasions, the main
diculty lies in the passage to the limit in products of only weakly convergent sequences).
Let us consider the local problem (4.7) and let wk 2 H]1;2(Y ) be the solution to (4.7). Let us still
denote by wk its Y -periodic extension to Rn . By Lemma A.1.16 it turns out that wk 2 Hloc 1;2(Rn ). Let
us dene for every k = 1; : : :; n the sequence of functions
(4:18) whk (x) = xk + "h wk ( "x ) = (ek ; x) + "h wk ( "x ) for a.e. x 2 Rn
h h
(note that this function is in fact the sum of the rst terms of the expansion u0(x) + "h u1(x; "x ) with
u0(x) = (ek ; x) and ue1(x) = 0). The periodicity property of this function yields easily that
h

 wk ! xk strongly in L2 (
) ; (as h ! 1)
(4:19) h
Dwhk * ek weakly in L2 (
; Rn) ; (as h ! 1).
Moreover, by Lemma A.1.17 (with g(y) = a(y)(ek + Dwk (y))), the functions whk satisfy the equations
(4:20) ;div(ah (x)Dwhk (x)) = 0 in D0 (Rn) :
Then, by multiplying (4.20) by any function v 2 H01;2(
) we have
Z
(4:21) (ah (x)Dwhk ; Dv) dx = 0 :

In order to avoid diculties with the boundary condition, let us take a function ' 2 C01 (
) and let us
write (4.10) with v = 'whk 2 H01;2(
) and (4.21) with v = 'uh 2 H01;2(
). We have then
Z Z
(ah (x)Duh ; (D')whk ) dx + (ah (x)Duh ; (Dwhk )') dx = hfh ; 'whk i
(4:22) Z
Z

(ah (x)Dwhk ; (D')uh ) dx + (ah (x)Dwhk ; (Duh )') dx = 0

Since aij = aji for every i; j 2 f1; : : :; ng ; we have that

Z Z
(ah (x)Dwhk ; (Duh )') dx = (ah (x)Duh ; (Dwhk )') dx :

Therefore, by subtracting the second equation in (4.22) from the rst one, we get
Z Z
(4:23) (ah (x)Duh ; (D')whk ) dx ; (ah (x)Dwhk ; (D')uh ) dx = hfh ; 'whki

for every ' 2 C01(

). Now we are in the position to pass to the limit in (4.23) as h ! 1 , since each term
is the scalar product in L2 (
; Rn) of an element which converges weakly and another which converges
strongly in L2 (
; Rn). Indeed
 h (x) = ah (x)Duh *   weakly in L2 (
; Rn)
(by (4.15)), and
(D')whk ! (D')xk strongly in L2 (
; Rn)
 An Introduction to Homogenization and G-convergence 15

(note that (D') is xed). Moreover,

X
n
h (x) = X;aij (
)(jk + @w (
))
; x

k n k
(ah Dwhk )i (x) = aij ( "x ) @w
@x @y "
j =1 h j j =1 j h

for every i 2 f1; : : :; ng . Hence,

Z X
n k
(4:24) (ah Dwhk )i * (aik (y) + aij (y) @w@y(y) ) dy = (4:6) = bik weakly in L2(
) :
Y j =1 j

Finally, (4.14) and Rellich's theorem imply that

(D')uh ! (D')u strongly in L2 (
; Rn) :
Since by (4.1) the sequence (whk ) converges to xk weakly in H01;2(
) and by assumption (fh ) converges
to f strongly in H ;1;2(
), we can nally assert that
Z ;X
n X
n

(4:25) i (Di ')xk ; bik (Di ')u dx = hf; 'xk i :

i=1 i=1

Moreover, by (4.16) with v = 'xk , the previous equation becomes

Z X
n Z X
n
(i xk ; bik u )(Di ') dx = i Di ('xk ) dx for every ' 2 C01(
) ;

i=1
i=1

and we get for every k = 1; : : :; n

Z X
n
(k ; bik Di u )' dx = 0 for every ' 2 C01 (
) :

i=1

Since the last equation holds for every ' 2 C01 (

), we get that
X
n
k = bik Di u a.e. on

i=1

for every k = 1; : : :; n. By the simmetry of the matrix b, which is shown in the next proposition, the proof
of (4.17) is accomplished. Since the homogenized operator is uniquely dened and we have uniqueness of
the solution to (4.5) we may conclude that the convergences
uh * u0 weakly in H01;2(
)
ah Duh * bDu0 weakly in L2 (
; Rn)
hold for the whole sequence, and not only for the above extracted subsequence.
16 A. Defranceschi

Proposition 4.2. Let a : Rn ! M nn be a function in S] . Let b be the constant matrix dened by
(4.6). Then b is still symmetric and satises the same ellipticity condition as a , i.e.,
(i) bik = bki for every i; k = 1; : : :n ;
Xn
(ii) (b; ) = bik k i  j j2 for every  2 Rn .
i;k=1
Proof. Let us prove (i). Fix k and s in f1; : : :; ng and let v = ws in (4.7). We obtain
Z
(a(y)(ek + Dwk (y)); Dws (y)) dy = 0 :
Y
R
By adding to both sides the quantity Y (a(y)(ek + Dwk (y)); es ) dy we get
Z X
n k Z
(ask (y) + asj (y) @w@y(y) ) dy = (a(y)(ek + Dwk (y)); (es + Dws (y))) dy ;
Y j =1 j Y

i.e., Z
bsk = (a(y)(ek + Dwk (y)); (es + Dws (y))) dy :
Y
Since a(x) is symmetric on Rn , the proof of (i) is accomplished.
Let us show (ii). Given  2 Rn , let us dene the sequence of functions
X
n
vh (x) = k whk (x) for a.e. x 2 Rn ;
k=1
where whk (x), k = 1; : : :; n are the functions dened by (4.18). Note that by (4.19) and (4.24) we have
8 vh ! Pnk=1 kxk = (; x) strongly in L2(
) ;
>
>
<
(4:26) > Dvh *  weakly in L2 (
; Rn) ;
>
: h 2
(a Dvh )i * (b)i weakly in L (
), for every i = 1; : : :n :
Moreover by (4.20) we obtain ;div(ah Dvh ) = 0 in D0 (Rn ) : Let us show that
Z Z
(4:27) (ah (x)Dvh ; Dvh )' dx ! (b; )' dx for every ' 2 C01 (
) :

Note that Z Z
(ah (x)Dvh ; Dvh )' dx = ; (ah (x)Dvh ; D')vh dx ; hdiv(ah Dvh ); 'vh i

Z

= ; (ah Dvh ; D')vh dx :

By virtue of (4.26) the last integral converges to

Z Z
; (b; D')(; x) dx = (b; )' dx ;

which proves (4.27) (note that this result can be obtained also immediately by the compensated compact-
ness lemma (see Lemma A.2.1 with gh = ah Dvh and uh = vh ).
 An Introduction to Homogenization and G-convergence 17

Let us note that the ellipticity condition of a implies that

Z Z
(ah (x)Dvh ; Dvh )' dx  jDvh j2 ' dx

for every ' 2 C01 (

), '  0. Now, by passing to the limit as h ! 1 , (4.27) and the weak lower
semicontinuity of the norm in L2 (
; Rn) imply that
Z Z
(4:28) (b; )' dx   j j2' dx

for every ' 2 C01 (

), '  0, which implies immediately (ii).
Let us state now some results which follow with minor modications from the previous homogeniza-
tion result.
Let a0 : Rn ! R be a Y -periodic function belonging to L1 (Rn). Moreover assume that there
exists a constant  > 0 such that a0   a.e. on Rn . Then we have:
Corollary 4.3. Let a 2 S] , a0 be as dened above, and let ("h ) be a sequence of positive real numbers
converging to 0 . Assume that (fh ) converges to f strongly in H ;1;2(
) . For every h 2 N , let uh be the
solution to the Dirichlet boundary value problem
( ;div(a( x )Du ) + a ( x )u = f on
;
" h 0 " h h
(4:29) h h

uh 2 H01;2(
) :
Then
weakly in H01;2(
) ;
uh * u0
where u0 is the unique solution to the homogenized problem
 ;div(bDu0) + M(a0)u0 = f on
;
(4:30)
u0 2 H01;2(
) :
The constant matrix b is dened by (4.6) and M(a0 ) is the mean value of a0 on Y .
Proof. It follows easily from Theorem 4.1 and the fact that
Z Z
a0( "x )uh v dx ! M(a0)u0 v dx

h

for every v 2 H01;2(

) (note that a0( "x ) * M(a0) in L1 (
) weak*, and hence we have a0 ( "x )uh *
M(a0 )u0 weakly in L2 (
)).
h h

Corollary 4.4. Assume @

be Lipschitz. Let a 2 S] , a0 be as dened above, and let ("h ) be a sequence
of positive real numbers converging to 0 . Assume that (fh ) converges to f weakly in L2(
) . For every
h 2 N , let uh be the solution to the Neumann boundary value problem
( ;div(a( x )Du ) + a ( x )u = f on
;
" h 0 " h h
(4:31) h

(a( "xh )Duh ; ) = 0 on

h

@
;
18 A. Defranceschi

where  denotes the unit outer normal to the boundary @

. Then
uh * u0 weakly in H 1;2(
) ;
where u0 is the unique solution to the problem
 ;div(bDu0) + M(a0)u0 = f on
;
(4:32)
(bDu0 ; ) = 0 on @
:
The constant matrix b is dened by (4.6) and M(a0 ) is the mean value of a0 on Y .
Proof. Let us use as above the notation ah (x) = a( "x ) and set ah0 (x) = a0 ( "x ). By weak solution of
(4.31) we mean a function uh 2 H 1;2(
) satisfying
h h

Z;
Z
(4:33) h h
(a (x)Duh ; Dv) + a0 (x)uh v dx = fh v dx for every v 2 H 1;2(
) :

Proceeding as in Theorem 4.1 and taking into account that a0 (x)   > 0 for a.e. x 2 Rn , we get
that the sequence (uh ) is uniformly bounded in H 1;2(
) and therefore converges (up to a subsequence)
weakly in H 1;2(
) and strongly in L2 (
) to a function u . Moreover, by the periodicity property of a0
we have that (ah0 ) converges to M(a0 ) in L1 (
) weak*. We then obtain instead of the equation (4.16)
the following relation
Z;
Z
(4:34) 
( ; Dv) + M(a0)u v dx = fv dx for every v 2 H 1;2(
)

(note that   is the weak limit in L2 (

; Rn) of the sequence ( h ) dened as ah (x)Duh (x)). This shows
that the functions   and u satisfy a certain equation and an associated boundary condition as in the
classical Neumann boundary value problem. Therefore, to conclude the proof it remains to show that
  (x) = bDu (x) for a.e. x 2
.
But the proof of this relation is of course the same as in Theorem 4.1, since it is a local property independent
of the boundary conditions.
Remark 4.5. The example studied in Section 2 shows that in general (Duh ) does not converge strongly
in L2 (
; Rn) to Du0 . Indeed, assume for a moment that the solutions uh and u0 to (2.3) and (2.11)
respectively, satisfy
(4:35) duh ! du0 strongly in L2(
) :
dx dx
Then,
(4:36) ah (x) du h
dx * M(a) dx
du0 weakly in L2(
) ;
and one would be able to pass to the limit directly in the equation
Z duh dv Z
ah dx = fv dx

dx dx

and obtain Z Z
M(a) du 0 dv dx = fv dx
dx dx for every v 2 H 1;2(
) :

But this is not the limit equation (see also Remark 2.1) and we get a contradiction.
Therefore, (4.8.) cannot be improved without adding extra terms (of the \corrector" type). In [9]
(Chapter 1, Section 5) one can nd the proof of the following corrector result.
 An Introduction to Homogenization and G-convergence 19

Theorem 4.6. Let us assume that the hypotheses of Theorem 4.1 hold true. Moreover, assume that
i) fh ; f 2 L2 (
) ;
ii) wk dened by (4.7) belongs to W 1;1(Y ) for every k = 1; : : :; n :
Then
Duh = Du0 + P h Du0 + rh with rh ! 0 strongly in L2 (
; Rn) ;
where the matrix P h (x) = (Pikh (x)) is dened by Pikh (x) = @w x
@x ( " ) .
k

i h

Remark 4.7. Note that from a numerical point of view correctors are important since the weak H 1;2 -
convergence is not completely satisfactory. Correctors give a \good" approximation of Duh , since it is an
approximation in the strong topology of L2 (
; Rn) (the term P h Du0 \corrects" rapid oscillations of the
gradient of (uh ; u0 )).
Furthermore, the corrector result turns out to be a basic tool in the study of the asymptotic be-
haviour of the bounded solutions uh to quasilinear equations of the form
;div(a( "x )Duh ) + uh = H h (x; uh ; Duh ) ;
h
where a 2 S] , > 0 and the Hamiltonians H h = H h (x; s; ) are measurable in x, continuous in the pair
[s; ] and have, for example, quadratic growth in  (see [8], where also the case a non-symmetric has been
considered).

II. Homogenization of monotone operators

5. Homogenization and correctors for monotone operators
Let us deal now with the homogenization of a sequence of nonlinear monotone operators Ah :
H01;2(
) ! H ;1;2(
) of the form
Ah u = ;div(a( "x ; Du)) ;
h
where a(x;
) is Y -periodic and satises suitable assumptions of uniform strict monotonicity and uniform
Lipschitz-continuity. The results presented here are contained in [47] (see also [2] and [3]).
By N] we denote the set of all functions a : Rn  Rn ! Rn such that for every  2 Rn , a(
; )
is Lebesgue measurable and Y -periodic and satises the following properties: there exist two constants
0 <    < +1 such that
(5.1) (strict monotonicity) (a(x; 1) ; a(x; 2); 1 ; 2 )  j1 ; 2 j2
(5.2) (Lipschitz-continuity) ja(x; 1) ; a(x; 2)j   j1 ; 2 j
for a.e. x 2 Rn and for every 1 , 2 2 Rn . Moreover
(5.3) a(x; 0) = 0 for a.e. x 2 Rn :
20 A. Defranceschi

Remark 5.1. Note that a(x; ) = a(x), where a : Rn ! M nn is Y -periodic and satises (4.1) and (4.3)
(without any symmetry assumption) belongs to N] . Therefore we shall deduce from a homogenization re-
sult proven for N] a homogenization result for a sequence of operators of the form Ah u = ;div(a( "x )Du),
h

where a is not necessarily symmetric.

Given a 2 N] , for every "h > 0 and fh 2 H ;1;2(
) let us consider the following Dirichlet boundary
value problem on the bounded open subset
of Rn :
( ;div(a( x ; Du )) = f on
;
" h h
(5:4) 1 ; 2
h

uh 2 H0 (
) :

Remark 5.2. By a classical result in existence theory for boundary value problems dened by monotone
operators (see Theorem A.3.2) for every fh 2 H ;1;2(
) and for every "h > 0 there exists a unique solution
uh 2 H01;2(
) to (5.4). Indeed, let us consider the operator Ah : H01;2(
) ! H ;1;2(
) dened by
Ah u = ;div(a( "x ; Du)) :
h
By (5.1) we have that
hAhZu1 ; Ah u2; u1 ; u2i = Z
= (a( "x ; Du1 ) ; a( "x ; Du2 ); Du1 ; Du2 ) dx   jDu1 ; Du2 j2dx

h h

for everyu1 , u2 2 H01;2(

), which guarantees that Ah is a strictly monotone and coercive map on H01;2(
)
(take into account (5.3)). Moreover, by (5.2) we get
kAh u1 ; Ah u2kH ;1 2 (
)   ku1 ; u2kH01 2 (
)
; ;

for every u1 , u2 2 H01;2(

) which proves the continuity of Ah . Therefore, by Theorem A.3.2 we have
existence and uniqueness of a solution to (5.4)
In this section we shall prove the following homogenization result:
Theorem 5.3. Let a 2 N] and let ("h ) be a sequence of positive real numbers converging to 0 . Assume
that (fh ) converges strongly in H ;1;2(
) to f . Let (uh ) be the solutions to (5.4). Then,
uh * u0 weakly in H01;2(
) ;
a( "x ; Duh ) * b(Du0 ) weakly in L2 (
; Rn) ;
h
where u0 is the unique solution to the homogenized problem
 ;div(b(Du0)) = f on
;
(5:5)
u0 2 H01;2(
) :
The operator b : Rn ! Rn is dened for every  2 Rn by
Z
(5:6) b() = a(y;  + Dw (y)) dy ;
Y
where w is the unique solution to the local problem
(R (a(y;  + Dw (y)); Dv(y)) dy = 0 for every v 2 H]1;2(Y ) ;
(5:7) Y
w 2 H]1;2(Y ) :
 An Introduction to Homogenization and G-convergence 21

Remark 5.4. Proceeding analogously as in Remark 5.2 one can prove the existence and uniqueness of a
solution to the local problem (5.7). It can be shown directly by using the denition of b and the properties
satised by a, that b : Rn ! Rn is monotone and continuous on Rn (hence, by Theorem A.3.2 maximal
monotone). Furthermore, it will be seen in the sequel that the operator b satises strict monotonicity
properties like a (this implies in particular the uniqueness of the solution to (5.5)).
Let us show that b is monotone. Given 1 , 2 2 Rn , by the denition of b there exist w 2 H]1;2(Y ) i

i = 1; 2 satisfying
Z
(5:8) (a(y; i + Dw (y)); Dv) dy = 0
i
for every v 2 H]1;2(Y )
Y
and Z
b(i ) = a(y; i + Dw (y)) dy :
i

Y
Therefore, by taking (5.8) and (5.1) into account, we get
Z Z
(b(1 ) ; b(2 ); 1 ; 2) = ( a(y; 1 + Dw1 (y)) dy ; a(y; 2 + Dw2 (y)) dy; 1 ; 2 )
Z Y Y
= (a(y; 1 + Dw (y)) ; a(y; 2 + Dw2 (y)); 1 ; 2 ) dy
1

Z Y
= (a(y; 1 + Dw1 (y)) ; a(y; 2 + Dw2 (y)); (1 ; Dw1 (y)) ; (2 + Dw2 (y))) dy
Y
 0;
this proves that b is monotone.
Let us prove that for every 1 , 2 2 Rn we have
2
(5:9) jb(1 ) ; b(2 )j   j1 ; 2 j :
Let w 2 H]1;2(Y ) i = 1; 2 satisfying
i

Z
(5:10) (a(y; i + Dw (y)); Dv) dy = 0
i
for every v 2 H]1;2(Y )
Y
and Z
b(i ) = a(y; i + Dw (y)) dy :
i

Y
Then, by taking (5.2), (5.1) and (5.10) into account, we get
Z Z
jb(1 );b(2 )j2 = j a(y; 1 + Dw1 (y)) dy ; a(y; 2 + Dw2 (y)) dyj2
;Z
Y Y
1


 ja(y; 1 + Dw (y)) ; a(y; 2 + Dw (y))j dy 2 2
YZ
  j(1 + Dw (y)) ; (2 + Dw (y))j dy
2
; 1 2

;Z
Y
 2


j(1 + Dw1 (y)) ; (2 + Dw2 (y))j2 dy
2 Z
Y
  ; (a(y; 1 + Dw1 (y)) ; a(y; 2 + Dw2 (y)); (1 + Dw1 (y)) ; (2 + Dw2 (y)) dy

Y
2 2
  (b(1 ) ; b(2 ); 1 ; 2 )   jb(1 ) ; b(2)jj1 ; 2 j ;
22 A. Defranceschi

and (5.9) follows.

Proof of Theorem 5.3. By Remark 5.2, for every h 2 N , there exists a unique solution uh to the problem
(R

(a( "h ; Duh ); Dv) dx = hfh ; vi
x for every v 2 H01;2(
) ;
(5:11)
uh 2 H01;2(
) :
By taking v = uh in (5.11) and by means of the assumptions (5.1) and (5.3) (take into account also that
(fh ) is uniformly bounded in H ;1;2(
)), we get immediately
(5:12) kuh kH01 2 (
)  c ;
;

where c is a constant independent of h. Let us dene

 h = a( "x ; Duh ) :
h
By (5.2), (5.3) and (5.12) we obtain that
(5:13) k h kL2 (
;R )  C ;
n

where C is a constant independent of h. Therefore, there exist u 2 H01;2(

) and   2 L2 (
; Rn) and
two subsequences, still denoted by (uh ) and ( h ), such that
uh * u weakly in H01;2(
) ;
(5:14)
h *  weakly in L2 (
; Rn) :
By passing to the limit in (5.11) we get (in the sense of distributions)
;div  = f on

(note that here the weak convergence in H ;1;2(

) of (fh ) to f would suce). If we show that
  = b(Du ) a.e. on
;
then by the uniqueness of the solution to problem (5.5) we have to conclude that u = u0 . Arguing as in
the proof of Theorem 4.1 we obtain then that the convergences
uh * u  weakly in H01;2(
) ;
h *  weakly in L2 (
; Rn)
hold for the whole sequence, and not only for the above extracted subsequence. Therefore, the proof of
Theorem 5.3 is accomplished if we show that   = b(Du ) a.e. on
:
In order to prove that   = b(Du ) a.e. on
we dene a sequence of suitable functions wh 2
H 1;2(
), "h Y -periodic, in the following way. Given  2 Rn , let us consider a solution w 2 H]1;2(Y ) to
problem (5.7). Let us still denote by w its Y -periodic extension to Rn . It can be proved (see Lemma
A.1.16) that w 2 Hloc1;2(Rn ) and
Z
(a(x;  + Dw (x)); Dv(x)) dx = 0
R n
 An Introduction to Homogenization and G-convergence 23

for every v 2 C01(Rn ) (see Lemma A.1.17). Let us dene

(5:15) wh (x) = (; x) + "h w ( "x ) for a.e. x 2 Rn :
h
The periodicity properties of this function and of a yield easily that
8 w * (; x) weakly in H 1;2(
) ;
>
> h
>
< 
(5:16) > Dwh *  weakly in L2 (
; Rn) ;
>
: a( "x ; Dwh (x)) = a(
;  + Dw (
)); "x
* b() weakly in L2(
; Rn) :
h h

By the monotonicity of a we have

Z x
(a( " ; Duh (x)) ; a( "x ; Dwh (x)); Duh (x) ; Dwh (x))'(x) dx  0

h h
1
for every ' 2 C0 (
), '  0. By passing to the limit as h tends to 1 , the compensated compactness
lemma A.2.1 implies that Z
(  (x) ; b(); Du (x) ; )'(x) dx  0

for every ' 2 C01(

), '
 0 (note that ;div(a( "x ; Duh )) = fh , and (fh ) converges to f strongly
in H (
); moreover ;div(a( "x ; Dwh )) = 0 for every h 2 N ; and (5.16) hold). Therefore, for every
; 1; 2
h

 2 Rn we have
h

(5:17) (  (x) ; b(); Du (x) ; )  0 for a.e. x 2

:
In particular, if we denote by (m ) a countable dense subset of Rn , (5.17) yields that
(5:18) (  (x) ; b(m ); Du (x) ; m )  0 for a.e. x 2
; for every m 2 N :
This implies by the continuity of b (see Remark 5.4) that
(  (x) ; b(); Du (x) ; )  0 for a.e. x 2
and for every  2 Rn :
By taking the maximal monotonicity of b into account the last inequality guarantees that   (x) =
b(Du (x)) for a.e. x 2
; which was our goal.
Proposition 5.5. The operator b : Rn ! Rn dened by (5.6) satises the following property:
(5:19) (b(1 ) ; b(2); 1 ; 2)  j1 ; 2 j2
for every 1 , 2 2 Rn .
Proof. Let i 2 Rn , i = 1; 2. For every i = 1; 2 let us consider the sequence of functions wh 2 H 1;2(
)
i

such that
8 w * ( ; x) weakly in H 1;2(
) ;
>
> h
i
i
< 
(5:20) > Dwh * i weakly in L2 (
; Rn) ;
i

>
: x  ;x

a( " ; Dwh (x)) = a(
; i + Dw (
)) "
h
i  i
h
* b(i ) weakly in L2 (
; Rn) :
24 A. Defranceschi

By the monotonicity of a it follows that

Z x Z
(a( " ; Dwh1 ) ; a( "x ; Dwh2 ); Dwh1 ; Dwh2 )'(x) dx   jDwh1 ; Dwh2 j2'(x) dx

h h

for every ' 2 C01 (

), '  0. By passing to the limit as h tends to 1 , the compensated compactness
lemma (used on the left hand side) and the weak lower semicontinuity of the norm in L2 (
; Rn) (applied
on the right hand side) ensure that
Z Z
(b(1 ) ; b(2 ); D1 ; D2 )'(x) dx   j1 ; 2 j2'(x) dx

for every ' 2 C01 (

), '  0, which implies (5.19).
Finally, let us mention that a corrector result for the case a 2 N] has been proven in [22]. It can
be stated as follows:
Theorem 5.6. Assume that the hypotheses of Theorem 5.3 hold true. Let uh be the solutions to the
equations (5.4) and let u0 be the solution to problem (5.5). Then
Duh = ph (
; Mh Du0) + rh with rh ! 0 strongly in L2 (
; Rn) :

Here, for every "h > 0, the function ph : Rn  Rn ! Rn is dened by ph (x; ) =  + Dw ( "x ),
where w is the unique solution to the local problem (5.7). Moreover, for every ' 2 L2 (
; Rn) the function
h

Mh ' : Rn ! Rn is dened by
X Z
(Mh ')(x) = 1Y (x) jY1i j '(y) dy ;
i

i2I h
h
h Y i
h

where Yhi = "h (i + Y ) (for i 2 Zn ), Ih = fi 2 Zn : Yhi 

g and 1A is the characteristic function of a set
A  Rn .

Remark 5.7. This corrector result permits to study the limit behaviour of the bounded solutions uh to
quasi-linear equations of the form
;div(a( "x ; Duh )) + uh = H( "x ; uh ; Duh ) ;
h h
where a 2 N] , > 0, and H = H(x; s; ) is Y -periodic in x, continuous in the pair [s; ] and grows at
most like j j2 (for more details see [20]).
Conclusive remarks
Let us conclude this chapter with the statement of some further results on homogenization of
nonlinear monotone operators in divergence form.
The case 1 < p < +1 has been studied under analogous hypotheses of uniform strict monotonicity
and equicontinuity for a by Fusco and Moscariello in [27] and [28]. Given two positive constants  and
, they prove an homogenization result for
Ah u = ;div(a( "x ; u; Du)) ;
h
 An Introduction to Homogenization and G-convergence 25

where a(x; s; ) veries the following structure conditions:

a) a(
; s; ) is Y -periodic and Lebesgue measurable on Rn ;
b) for a.e. x 2 Rn , for every s, s1 , s2 2 R , and 1 , 2 2 Rn
if p  2
i) (a(x; s; 1) ; a(x; s; 2); 1 ; 2)  j1 ; 2jp
ii) ja(x; s1; 1) ; a(x; s2; 2 )j  (1 + js1j + js2j + j1j + j2 j)p;2(js1 ; s2 j + j1 ; 2 j)
if 1 < p  2
j) (a(x; s; 1) ; a(x; s; 2); 1 ; 2 )  j1 ; 2j2 (j1j + j2 j)p;2
jj) ja(x; s1; 1 ) ; a(x; s2 ; 2)j  (js1 ; s2 j + j1 ; 2 j)p;1
c) a(x; 0; 0) 2 Lq (
; Rn) if p > n, or a(x; 0; 0) 2 Lp0 (
; Rn) with p0 > p;n 1 if p  n.
The main result is the following:
Theorem 5.8. Let a : Rn  R  Rn ! Rn satisfying a), b) and c). Assume that f 2 Lp0 (
) with
p0 > np . Let ("h ) be a sequence of positive real numbers converging to 0 . Let uh be the solutions to the
Dirichlet boundary value problems
( ;div(a( x ; u ; Du )) = f on
;
" h h
h
uh 2 W01;p (
) :
Then
uh * u0 weakly in W01;p (
) ;
a( "x ; uh; Duh ) * b(u0 ; Du0) weakly in Lq (
; Rn) ;
h
where u0 is the unique solution to the problem
 ;div(b(u0; Du0)) = f on
;
u0 2 W01;p(
) :
The homogenized operator b : R  Rn ! Rn is dened by
Z
b(s; ) = a(y; s;  + Dw (y)) dy ;
Y
where w is the unique solution to
( R (a(y; s;  + Dw (y)); Dv(y)) dy = 0 for every v 2 W]1;p (Y )
Y
w 2 W]1;p (Y ) :

Finally, in [19] the regularity conditions on a (required until this point) has been weakend and also
the general case where a is a possibly multivalued map has been considered. To state the main result let
us introduce some notation and denition.
For every open subset U in Rn we denote by L(U) the  -eld of all Lebesgue measurable subsets
of U , and by B(Rn ) the  -eld of all Borel subsets of Rn . Let 1 < p < +1 , and let us x two constants
m1  0, m2  0, and two constants c1 > 0, c2 > 0.
26 A. Defranceschi

Denition 5.9. By M(Rn) we denote that class of all (possibly) multivalued functions a : Rn ! Rn
which satisfy the following conditions:
i) a is maximal monotone;
ii) the estimates
jjq  m1 + c1(; )
j jp  m2 + c2(; )
hold for every  2 Rn and  2 a().
For every open subset U of Rn , by MU (Rn ) we denote the class of all multivalued functions
a : U  Rn ! Rn with closed values which satisfy the following conditions:
iii) for a.e. y 2 U , a(y;
) 2 M(Rn );
iv) a is measurable with respect to L(U)
B(Rn ) and B(Rn ), i.e.,
a;1 (C) = f[y; ] 2 U  Rn : a(y; ) \ C 6= g 2 L(U)
B(Rn )
for every closed set C  Rn .
Now we can state the homogenization result:

Theorem 5.10. Let a 2 MR (Rn) be such that a(
; ) is Y -periodic for every  2 Rn . Let ("h ) be a
n

sequence of positive real numbers converging to 0 . Let uh be the solutions and gh be the momenta to the
Dirichlet boundary value problems
8 g (x) 2 a( x ; Du (x))
>
<h "
h
h for a.e. x 2
;
> ;divg = f on
;
: uh 2 Wh 01;p(
) :
Then, up to a subsequence,
uh * u weakly in W01;p (
) ;
gh * g weakly in Lq (
; Rn) ;
where u is a solution and g is a momentum of the homogenized problem
8 g(x) 2 b(Du(x)) for a.e. x 2
;
>
<
> ;divg = f on
;
: u 2 W01;p(
) :
For every  2 Rn , the set b() is dened by
Z
b() = f 2 Rn : 9w 2 W]1;p(Y )9k 2 Lq (Y ; Rn) satisfying (5.21) and = k(y)dyg ;
Y
and
8 w 2 W 1;p(Y ); k 2 Lq (Y ; Rn) ;
>
< ]
(5:21) 2 a(y;  + Dw (y)) for a.e. y 2 Y ;
: k(y)
> R (k(y); Dv(y)) dy = 0 for every v 2 W 1;p (Y ) :
Y ]
 An Introduction to Homogenization and G-convergence 27

Note: The main examples of maps of the class MR (Rn ) have the form
n

(5:22) a(x; ) = @ (x; ) ;

where @ denotes the subdierential with respect to  and : Rn  Rn ! [0; +1[ is measurable in
(x; ), convex in , and satises the inequalities
c1 j jp  (x; )  c2(1 + j jp)
for suitable constants 0 < c1  c2 . In this case the operator ;div(a( "x ; Du)) is the subdierential of the
h

functional
Z x
(5:23) h (u) = ( ; Du) dx :

"h
Note that the homogenization of a family of variational integrals of the form (5.23) has been studied by
Marcellini in [32] and by Carbone and Sbordone in [17] using the techniques of ;-convergence introduced
by De Giorgi.
Let us point out that if is not assumed to be dierentiable the map a can be multivalued. More-
over, the \multivalued approach" nds also a motivation in the fact that, under the general assumptions
on a 2 MR (Rn ), the additional hypothesis on a to be single-valued is not enough to ensure the same
n

property for the homogenized operator b (see [19], Section 4).

III. G-convergence; H-convergence

6. Setting of the problem. G-convergence for second order linear (uniformly) elliptic oper-
ators. The symmetric case
Let
be a bounded open subset of Rn . Let  and  be constants such that 0 <    < +1 .
Let us denote by M(; ) the set of all functions a :
! M nn satisfying the following properties:
(6.1) aij 2 L1 (
) for i; j = 1; : : :; n, and ja(x) j   j j for a.e. x 2
and for every  2 Rn ;
(6.2) (a(x); )  j j2 for a.e. x 2
and for every  2 Rn :
Let us consider a sequence (ah ) in M(; ) and let f 2 H ;1;2(
) (for the sake of simplicity,
without loss of generality, we consider from now on a right hand side term independent of h). Then, for
every xed h, there exists a unique solution uh to the Dirichlet boundary value problem
 ;div(ah (x)Duh) = f on
;
uh 2 H01;2(
) ;
and
kuh kH01 2 (
)  kf kH ;1 2 (
) :
; ;

Hence, there exists a subsequence (u(h) ) of (uh ) such that

u(h) * u0 weakly in H01;2(
) :
As in the periodic case, the problem is then the following: what can we say about u0 ? Does u0 satisfy an
equation of the same type as uh ?
28 A. Defranceschi

Remark 6.1. If (ah ) 2 M(; ), and

ah ! a0 strongly in L1 (
; Rn2 )
we can pass to the limit in ah Duh and we have
ah Duh * a0 Du0 weakly in L2(
; Rn)
and hence u0 is the solution (unique since a0 2 M(; )) to
 ;div(a0Du0) = f in
;
u0 2 H01;2(
) :
Let us note that the previous result is not true if we do not have the strong convergence of the sequence
(ah ).
Indeed, let
=]x0; x1[ R and f 2 L2 (
). Let us consider the sequence ah 2 M(; ) = fah 2
1
L (
) :   ah (x)   for a.e. x 2
g dened by ah (x) = g(hx), where g : R ! R is a 1-periodic
function dened on ]0; 1[ by (
 if 0 < x < 21
g(x) =
 if 12  x < 1 :
Then we get (up to subsequences)
1 * 1 1 1 1
in L1 (
) weak*
a h a0 = 2 (  +  ) ;
while
ah * b0 = 12 ( + ) in L1 (
) weak*
and the sequence of solutions uh to
 ; d (ah (x) du (x)) = f in
;
h
dx dx
uh (x0 ) = uh (x1 ) = 0
converge in the weak topology of H01;2(
) to the solution of the Dirichlet boundary value problem
(
; dxd (a0 (x) dudx0 (x) ) = f in
;
u0(x0) = u0(x1 ) = 0 :
Let us point out that only in dimension n = 1 the weak* limit of ( a1 ) caracterizes the coecients of the
h

matrix we are looking for. This is not longer true in dimension n > 1 as shown by an example in [33].
In order to answer to the above questions and other related questions for a more general class of
problems we follow the approach which uses the theory of G-convergence. A rst notion of G-convergence
for second order linear elliptic operators was introduced by De Giorgi and S. Spagnolo in [23], [42], [43],[44]
as the convergence, in a suitable topology, of the Green's operator associated to the Dirichlet boundary
value problems, in the case that ah 2 M(; ) and ah (x) is symmetric. Let us recall it brie y here.
 An Introduction to Homogenization and G-convergence 29

Denition 6.2. For every h 2 N let ah 2 M(; ), ahij (x) = ahji(x) for a.e. x 2
and for every
i; j = 1; : : :; n and let a0 2 M(; ), a0ij (x) = a0ji (x) for a.e. x 2
and for every i; j = 1; : : :; n. We
then say that (ah ) G-converges to a0 if for every f 2 H ;1;2(
) the solutions uh of the equations
 ;div(ah (x)Duh) = f on
;
(6:3)
uh 2 H01;2(
)
satisfy
uh * u0 weakly in H01;2(
) ;
where u0 is the solution to
 ;div(a0(x)Du0) = f on
;
(6:4)
u0 2 H01;2(
) :

The main result (which motivates the denition) is the sequential compactness of the class of sym-
metric functions belonging to M(; ) with respect to the G-convergence.

Theorem 6.3. Given a sequence (ah )  M(; ) , ah (x) symmetric, then there exist a subsequence
(a(h) ) of (ah ) and a0 2 M(; ) , a0(x) symmetric such that (a(h) ) G-converges to a0 :

Remark 6.4. The above result can be expressed as follows: given a sequence (ah )  M(; ), ah (x)
symmetric, there exist a matrix a0 2 M(; ), a0 (x) symmetric (called the G-limit) and an increasing
sequence of integers ((h)), such that for every f 2 H ;1;2(
) the sequence (u(h) ) of the solutions to
(6.3) corresponding to (a(h) ) converges weakly in H 1;2(
) and strongly in L2 (
) to the solution u0 to
(6.4).

The original proof of Spagnolo is rather technical and uses results of the semigroup theory for
linear operators and of the G-convergence of parabolic equations. Many dierent proofs have been given
subsequently (see, for example [46], [41]).
We would like to notice that in [44] also the following localization property is proven.

Theorem 6.5. Assume that (ah ) , (bh ) , a0 and b0 belong to M(; ) and are symmetric. If (ah ) G-
converges to a0 , (bh ) G-converges to b0 , and ah (x) = bh (x) for a.e. x in an open subset
0 of
, then
a0 (x) = b0(x) for a.e. x 2
0 .

7. H-convergence for second order linear (uniformly) elliptic operators. The non-symmetric
case
The notion of G-convergence has been extended to the non-symmetric case by Murat and Tartar
under the name of H-convergence (see [33], [47] and [48]). Let us recall the denition (see [33]). Let 0
and  0 be constants satisfying 0 < 0   0 < +1 .
30 A. Defranceschi

Denition 7.1. Let ah 2 M(; ) and let a0 2 M(0 ;  0). We then say that (ah ) H-converges to a0 if
for every f 2 H ;1;2(
) the solutions uh to the equations
 ;div(ah (x)Duh) = f on
;
(7:1)
uh 2 H01;2(
)
satisfy
(
(7:2) uh * u0 weakly in H01;2(
) ;
ah Duh * a0Du0 weakly in L2 (
; Rn) ;
where u0 is the solution to
 ;div(a0(x)Du0) = f on

(7:3)
u0 2 H01;2(
) :

Remark 7.2. Let us note that in the non-symmetric case (see also the nonlinear cases) a denition of
H-convergence as in the symmetric case would not determine uniquely the H-limit as the following example
shows.
Assume n = 3, and let ' 2 C01 (
). Let us dene a(x) = I , where I is the identity matrix and let
0 0 ;D3 '(x) D2 '(x)
1
b(x) = I + @ D3 '(x) 0 ;D1 '(x) A :
;D2 '(x) D1 '(x) 0
It is easy to see that a and b belong to M(; ) with  = 1 and  = (1 + max

jD'j). Note that

b(x) =  + D' ^  , where ^ denotes the external product in R and
n

Z
((D' ^ Du); Dv) dx = 0 for every u; v 2 H 1;2(
) :

It follows that
Z Z
(a(x)Du; Dv) dx = (b(x)Du; Dv) dx for every u; v 2 H 1;2(
) :

This implies that the operator Au = ;div(a(x)Du) coincides with the operator Bu = ;div(b(x)Du) in
spite of the fact that a(x) 6= b(x).

Let us show now that the condition (7.2) in the above denition determines uniquely the H-limit
a0 .
 An Introduction to Homogenization and G-convergence 31

Proposition 7.3. Let (ah ) be a sequence of functions of the class M(; ) and let a0 2 M(0;  0 ) and
b0 2 M(00;  00 ) such that (ah ) H-converges to a0 and (ah ) H-converges to b0 . Then, a0 = b0 a.e. on

.
Proof. Let ! 
and let ' 2 C01 (
) with ' = 1 on ! : For every  2 Rn let us dene f  =
;div(a0 D((; x)')) : Let us consider for h = 0; 1; : : : the solutions uh to the equations
(
;div(ah Duh ) = f  on
;
uh 2 H01;2(
) :

By the coercivity of a0 it turns out that

u0 = (; x)' on

;

and being a0 by assumption an H-limit of (ah ) we have

(
uh * u0 weakly in H01;2(
) ;
ah Duh * a0 Du0 weakly in L2 (
; Rn) :

Analogously for b0 we have

(
uh * u0 weakly in H01;2(
) ;
ah Duh * b0Du0 weakly in L2 (
; Rn) :

By the uniqueness of the weak limit in L2 (

; Rn) we may conclude that a0 Du0 = b0 Du0 a. e. on
.
Since Du0 =  on !, we get a0 = b0 a.e. on !. Thus, a0 = b0 a.e. on
:

The main result obtained by Tartar and Murat (see [33]) is the sequential compactness of the class
M(; ) with respect to the H-convergence.

Theorem 7.4. Given a sequence (ah )  M(; ) , then there exist a subsequence (a(h) ) of (ah ) and
0 
a 2 M(;  ) such that (a(h) ) H-converges to a0 :
2

Note. The above result shows that the class M(; ) is \stable" with respect to the H-convergence as
far as coerciveness is concerned, but unstable with regard to the norm of the matrices (compare with the
compactness result for the symmetric case).

The rest of this section is devoted to the study of some properties of the H-convergence and the
proof of Theorem 7.4.
The next lemma, together with the compensated compactness lemma (see Lemma A.2.1), will be
crucial in the sequel. Given M 2 M nn , we denote by M T the transpose matrix of M .
32 A. Defranceschi

Lemma 7.5. Let ah 2 M(; ) . Let (uh ) and (vh ) be two sequences in H 1;2(
) such that the following
conditions are satised:
8
>
< uhh * hu0 weakly in H 1;2(
)
(7:4) >  = a Du *  0 weakly in L2 (
; Rn)
: ;div(ah Duhh ) ! ;div0 strongly in H ;1;2(
)
8
>
< vhh * vh0 T weakly in0 H 1;2(
)
(7:5) >  = (a ) Dv *  weakly in L2(
; Rn)
: ;div((ah )T Dvhh ) ! ;div0 strongly in H ;1;2(
) :
Then
(7:6) ( 0; Dv0 ) = (Du0 ; 0) a.e. on
:

Proof. Let us write

( h ; Dvh ) = (ah Duh ; Dvh ) = (Duh ; (ah )T Dvh ) = (Duh ; h ) :
Hence Z Z
( h ; Dvh )' dx = (Duh ; h )' dx

for every ' 2 C01 (

). By the compensated compactness lemma we may conclude that
Z Z
( 0 ; Dv0 )' dx = (Du0 ; 0 )' dx

for every ' 2 C01 (

), and (7.6) follows immediately.
Proposition 7.6. Let (ah ) be a sequence in M(; ) which H-converges to a0 2 M(0;  0 ) . Then, the
sequence (ah )T H-converges to (a0)T .
Proof. Let g 2 H ;1;2(
). We have to prove that the solutions vh to
 ;div((ah )T Dvh ) = g on
;
vh 2 H01;2(
)
satisfy (
vh * v0 weakly in H01;2(
)
(ah )T Dvh * (a0 )T Dv0 weakly in L2 (
; Rn) ;
where v0 is the solution to  ;div((a0)T Dv0 ) = g on
;
v0 2 H01;2(
) :
Let us note that the sequence (vh ) is uniformly bounded in H01;2(
); furthermore, ((ah )T Dvh ) is uniformly
bounded in L2 (
; Rn). Hence, there exist a subsequence (h) of h and two functions v 2 H01;2(
) and
 2 L2 (
; Rn) such that
(
v(h) * v weakly in H01;2(
) ;
(a(h) )T Dv(h) *  weakly in L2 (
; Rn) :
 An Introduction to Homogenization and G-convergence 33

Clearly, ;div = g on
. On the other hand, given f 2 H ;1;2(
) and uh the solutions to
 ;div(ah Duh) = f on
;
uh 2 H01;2(
) ;
we have by assumption that
(
uh * u0 weakly in H01;2(
) ;
ah Duh * a0Du0 weakly in L2 (
; Rn) ;
where u0 is the solution to  ;div(a0Du0) = f on
;
u0 2 H01;2(
) :
By Lemma 7.5 we get
(7:7) (a0 Du0 ; Dv) = (Du0 ; ) a.e. on
:
Since f can be chosen arbitrarily in H ;1;2(
), arguing as in the proof of Proposition 7.3, we can take on
! 
, Du0 = , where  2 Rn is arbitrary. Then (7.7) becomes
(a0 ; Dv) = (; ) a.e. on ! :
Since this is true for every  2 Rn we can conclude that  = (a0 )T Dv on
. The equality ;div = g
implies then v = v0 ,  = (a0 )T Dv0 . By the uniqueness of the limits, we can conclude that the whole
sequences (vh ) and (ah Dvh ) converge to v0 and a0 Dv0 , respectively. This concludes the proof.
Proof of Theorem 7.4. The proof of Theorem 7.4 is divided in several steps. The proof of Step 1 is
given in the Appendix.
Step 1:
Proposition 7.7. Let F be a separable Banach space and let G be a re exive Banach space. Let L(F ; G)
be the set of all linear and continuous operators from F into G . Assume that for every h 2 N
(i) T h 2 L(F ; G) ;
(ii) kT h kL(F ;G)  c c > 0:
Then there exist a subsequence (T (h) ) of (T h ) and an operator T 0 2 L(F ; G) such that for every f 2 F
(7:8) T (h) f * T 0f weakly in G :

Proposition 7.8. Let V be a re exive and separable Banach space. Let  and  be two positive constants
and let (T h ) be a sequence of operators such that for every h 2 N
(i) T h 2 L(V ; V  ) ;
(ii) kT h kL(V ;V  )   ;
(iii) for every v 2 V; hT h v; viV  ;V  kvk2V :
34 A. Defranceschi

Then there exist a subsequence (T (h) ) of (T h ) and an operator T 0 2 L(V ; V  ) such that
8 T 0 2 L(V ; V )
>
<
(7:9) 0 kL(V ;V  )  
: kforT every
2
> 
v 2 V; hT 0 v; viV  ;V  kvk2V :
Moreover, for every f 2 V  we have
(7:10) (T (h) );1 f * (T 0 );1 f weakly in V :
Step 2. We construct the test functions which will be used in Lemma 7.5.
Let
0 be a bounded open subset of Rn such that

0 . Let us consider the sequence (bh )
in M(; ;
0) (note that M(; ;
0) denotes the set M(; ), where
has been replaced by
0 ) such
that
(7:11) bh = (ah )T on

(for example take bh =  I on

0 n
).
Let us consider the sequence of operators (Bh )  L(H01;2(
0); H ;1;2(
0 )) dened for h 2 N by
Bh u = ;div(bh Du) :
By Proposition 7.8 (it is easy to verify that Bh satises the hypotheses (ii) and (iii) of Proposition 7.8)
there exist a subsequence (B(h) ) of (Bh ) and an operator B0 2 L(H01;2(
0 ); H ;1;2(
0 )) such that for
every g 2 H ;1;2(
0)
(7:12) (B(h) );1 g * (B0 );1g weakly in H01;2(
0 ) :
Given ' 2 C01 (
0 ) such that ' = 1 on
, we denote by gi the function in H ;1;2(
0 ) dened by
(7:13) gi = B0 ((ei ; x)') :
For every i 2 f1; : : :; ng , let us denote by v(h);i the solutions to
(
B(h) v(h);i = gi on
0 ;
v(h);i 2 H01;2(
0 ) :
This denition together with (7.12) and (7.11) implies that for every i 2 f1; 2; : : :; ng we have
(
;div((a(h) )T Dv(h);i ) = gi on
;
v(h);i 2 H 1;2(
) :
Furthermore, by (7.12)
v(h);i * (ei ;
) weakly in H 1;2(
) :
By passing to a subsequence of (h), let us denote it by (h), we have for every i 2 f1; 2; : : :; ng
(a (h) )T Dv (h) ;i * i weakly in L2 (
; Rn) :
 An Introduction to Homogenization and G-convergence 35

Note that for every i 2 f1; 2; : : :; ng , the sequence (v (h) ;i) satises (7.5).
Let us dene a0 2 L2 (
; Rn2 ) by
(a0 (x))i;j = (i(x))j for a.e. x 2
, for every i; j 2 f1; 2; : : :; ng :
In the remaining steps we shall prove that (a (h) ) H-converges to a0 .
Step 3. For the sake of simplicity we shall in the sequel simply write h instead of (h). For every h 2 N ,
let us denote by Ah the operator in L(H01;2(
); H ;1;2(
)) dened by
Ah u = ;div(ah Du) :
It turns out that Ah is an isomorphism. Moreover, let us consider the operator T h 2 L(H ;1;2(
); L2 (
; Rn))
dened by
T h f = ah D((Ah );1 f) :
We have
kT h f kL2 (
;R )   k(Ah );1 f kH01 2 (
)   kf kH ;1 2 (
)
n ; ;

for every f 2 H ;1;2(

). By applying Proposition 7.8 to the operator Ah and Proposition 7.7 to the
operator T h we deduce that there exist a subsequence (h) of h (recall that h stands here for the
subsequence (h); however, no confusion can occur) and two operators A0 2 L(H01;2(
); H ;1;2(
)) and
T 0 2 L(H ;1;2(
); L2 (
; Rn)) such that for every f 2 H ;1;2(
) we have
(A(h) );1 f * (A0 );1 f weakly in H01;2(
)
T (h) f * T 0 f weakly in L2 (
; Rn) :
For f 2 H ;1;2(
), we set
(Ah );1 f = uh ; (A0);1 f = u0 ;
(here h stands for (h)). We have then
8 u * u weakly in H 1;2(
)
>
< ((hh)) 0 0
> a Du * T 0 f =  weakly in L2 (
; Rn)
: ;div(a(h(h)Du
)
) = f on
:
(h)
We note know that the sequence (u(h) ) satises the hypothesis (7.4) of Lemma 7.5. Moreover, by taking
into account the sequence (v (h) ) constructed in the previous step and Lemma 7.5, we obtain for every
i 2 f1; 2; : : :; ng
(; D(ei ; x)) = (Du0 ; i) a.e. on
:
By the denition of a0 this is nothing but
T 0f =  = a0Du0 :
Step 4. We prove that a0 belongs to M(;  ).
2

By denition a0 2 L2 (
; Rn2 ). Hence, for every u0 2 H01;2(
) we have a0 Du0 2 L2 (
; Rn) . By
the compensated compactness lemma we get that
(7:14) (a(h) Du(h) ; Du(h) ) ! (a0 Du0 ; Du0) in D0 (
) :
36 A. Defranceschi

By the ellipticity assumption of ah we have

Z Z
(7:15) (a(h) Du(h) ; Du(h) )' dx   jDu(h) j2' dx

for every ' 2 C01(

), '  0. Then, by (7.14), (7.15) and the weak lower semicontinuity of the norm in
L2 (
; Rn) we get
Z Z
(7:16) (a0 Du0 ; Du0)' dx   jDu0 j2' dx

for every ' 2 C01 (

), '  0. Note that (7.16) holds for every u0 2 H01;2(
) (since f ranges all over
H ;1;2(
) and A0 is an isomorphism). By taking u0 = (; x) , where 2 C01 (
) and = 1 in a
neighbourhood of the support of ' and  2 Rn arbitrary, from (7.16) we deduce
Z Z
(a0(x); )' dx   jj2' dx

for every ' 2 C01 (

), '  0. Hence
(a0 (x); )  jj2
for every  2 Rn and for a.e. x 2
.
Let us prove now that ja0(x)j   jj for a.e. x 2
and for every  2 Rn .
2

By the assumptions on ah it follows that for every h 2 N the following inequality holds
((ah );1 (x); )  2 jj2 for a.e. x 2
, for every  2 Rn :

This yields Z Z
(Duh ; ah Duh )'2 dx  2 jah Duh j2'2 dx

1
for every ' 2 C0 (
) and for every h 2 N ; hence, in particular it holds for every (h). By passing to the
limit (taking into account the compensated compactness lemma and the weak lower semicontinuity of the
norm in L2 (
; Rn)) we obtain
Z  Z
(Du0 ; a0Du0 )'2 dx   2 ja0 Du0j2 '2 dx :

Proceeding as above we get for every  2 Rn and for every ' 2 C01(
)
Z Z
(; a0)'2 dx  2 ja0j2'2 dx :

From this inequality we can deduce

 0
 2 ka 'kL2 (
;R )  ka 'kL2 (
;R ) k'kL2 (
;R ) :
0n n n

Finally,
2
ka0'kL2 (
;R )   k'kL2(
) jj ;
n
 An Introduction to Homogenization and G-convergence 37

for every ' 2 C01 (

) and for every  2 Rn . By the converse of Holder's inequality (see [26], Proposition
6.14) we obtain that a0  2 L1 (
; Rn) and
2
ka0kL1 (
;R )   jj
n

for every  2 Rn .
Step 5. In the previous step we have shown that a0 belongs to M(;  ). The limit u0 of the sequence
2

(u(h) ) is dened in a unique manner (independent of the subsequence (h) extract from the sequence
(h)) by
 ;div(a0Du0) = f in
;
u0 2 H01;2(
) :
Moreover, by the uniqueness of the limits, we have that the whole sequences (u (h) ) and (a (h) u (h) ) (and
not the subsequences determined by (h)) converge. We may conclude that a (h) H-converges to a0 ; the
proof of Theorem 7.4 is so accomplished.

Remark 7.9. Let us conclude this section by noting that a corrector result for the class M(; ) has
been proved in [33]. Moreover, some properties of the H-convergence for quasi-linear elliptic operators
were studied by L. Boccardo, Th. Gallouet and F. Murat in [12], [13] and [14].

8. Some further remarks on G-convergence and H-convergence

The rst results in the nonlinear case are due to L. Tartar, who studied (in [47]) the properties of
the H-convergence for monotone problems of the type
 ;div(ah(x; Duh)) = f on
;
uh 2 H01;2(
) ;
assuming that the maps ah are uniformly strictly monotone and uniformly Lipschitz-continuous on Rn
(note that the vector-valued case is considered in [45] whereas more general classes of uniformly equicon-
tinuous strictly monotone operators on W 1;p(
), with p  2, are considered by Raitum in [39]).

By N (; ) we denote the set of all functions a :

 Rn ! Rn such that for every  2 Rn , a(
; )
is Lebesgue measurable and satises the following properties:
(8.1) (strict monotonicity) (a(x; 1) ; (x; 2); 1 ; 2 )  j1 ; 2 j2
(8.2) (Lipschitz-continuity) ja(x; 1) ; a(x; 2)j   j1 ; 2 j
for a.e. x 2 Rn and for every 1 , 2 2 Rn . Moreover
(8.3) a(x; 0) = 0 for a.e. x 2 Rn :

Let 0 and  0 be constants satisfying 0 < 0   0 < +1 .

38 A. Defranceschi

Denition 8.1. Let ah 2 N (; ) and let a0 2 N (0 ;  0 ). We say that (ah) H-converges to a0 if for
every f 2 H ;1;2(
) the solutions uh to the equations
 ;div(ah(x; Duh)) = f on
;
(8:4)
uh 2 H01;2(
)
satisfy
(
(8:5) uh * u0 weakly in H01;2(
) ;
ah (
; Duh ) * a0(
; Du0 ) weakly in L2 (
; Rn) ;
where u0 is the solution to  ;div(a0(x; Du0)) = f on
;
u0 2 H01;2(
) :
The followingtheorem, due to Tartar (see [47] and [50]), justies the denition (8.1) of H-convergence;
its proof is reproduced in [24].
Theorem 8.2. Given a sequence (ah )  N (; ) , there exist a subsequence (a(h) ) of (ah ) and a0 2
N (; 2 ) such that (a(h) ) H-converges to a0 :

Remark 8.3. Let us mention that a corrector result for the class N (; ) has been proved by Murat in
[36].
Moreover, in [11] a convergence result for the strongly non linear equations
 ;div(ah (x; Duh)) + ch(x)g(uh ) = f on
;
uh 2 H01;2(
) ;
where ah 2 N (; ), has been proved.
A compactness result (in the sense of H-convergence) for equations of the type
 ;div(ah (x; uh; Duh)) = f on
;
uh 2 H01;2(
) ;
with ah (x; s; ) 2 N (; ) for every s 2 R is shown in [24].
Finally, a general notion of G-convergence for a sequence of maximal monotone (possibly multival-
ued) operators of the form Ah u = ;div(ah (x; Du)) has been introduced in [18]. Let us point out that,
in order to include the case (5.22), the authors consider the class M
(Rn ) (see Denition 5.9) and do
not assume the maps a to be continuous or strictly monotone. The main results of the paper are the
local character of the G-convergence and the sequential compactness of M
(Rn ) with respect to the
G-convergence.
 An Introduction to Homogenization and G-convergence 39

Appendix
A.1. Lp and Sobolev Spaces
We give here only the denitions and main results that we used in the previous chapters. Most of the
theorems are standard and their proofs as well as a deeper analysis are available in several textbooks on
Functional Analysis.
We start with the abstract denition of the notion of weak convergence (for more details on it we
refer to [25], [51] or to [16]) and then apply it to Lp and Sobolev spaces.
A.1.1. Weak convergence
Let us start with the denition.
Denition A.1.1. Let X be a real Banach space, X  its dual and h
;
i the canonical pairing over
X  X .
i) We say that the sequence (xh ) in X converges weakly to x 2 X and we denote
xh * x in X
if hx ; xh i ! hx; xi for every x 2 X  .
ii) We say that the sequence (xh ) in X  converges weak * to x 2 X  and we denote
 
xh * x in X 
if hxh ; xi ! hx ; xi for every x 2 X .
Then the following results hold.
Theorem A.1.2. Let X be a Banach space. Let (xh ) and (xh) be two sequence in X and in X  ,
respectively.
i) Let xh * x , then there exists a constant K > 0 such that kxh k  K ;
furthermore kxk  lim inf kx k :
h!1 h
 
ii) Let xh * x , then there exists a constant K > 0 such that kxh kX   K ;
furthermore kxkX   lim inf kx k  :
h!1 h X
iii) If xh ! x (strongly), then xh * x (weakly).
 x (weak *).
iv) If xh ! x (strongly in X  ), then xh *
v) If xh * x (weakly) and xh ! x (strongly in X  ), then hxh ; xhi ! hx ; xi .
Theorem A.1.3. Let X be a re exive Banach space. Let (xh ) be a sequence in X and K be a positive
constant such that kxh k  K . Then there exist x 2 X and a subsequence (x(h) ) of (xh ) such that
x(h) * x in X .
Theorem A.1.4. Let X be a separable Banach space. Let (xh) be a sequence in X  and K be a positive
constant such that kxh kX   K . Then there exist x 2 X  and a subsequence (x(h) ) of (xh ) such that
x(h) * x in X  .
40 A. Defranceschi

A.1.2. Lp spaces
We apply the above results to the Lp spaces which are dened as follows (for more details see [1],
[16], [52]).
Denition A.1.5. Let
be an open subset of Rn .
i) Let 1  p < +1 . We denote by Lp (
; Rn) the set of all measurable functions f :
! Rn such
that
kf kL (
;R ) 
p n
;Z jf(x)jpdx
1=p < +1 :

It can be shown that k
kL (

;R ) is a norm.
p n

ii) Let p = +1 . A measurable function f :

! Rn is said to be in L1 (
; Rn) if
kf kL1 (
;R )  inf f : jf(x)j   a.e. in
g < +1 :
n

One proves that k
kL1 (

;R ) denes a norm.
n

iii) Lploc (
; Rn) denotes the linear space of measurable functions u such that u 2 Lp (
0 ; Rn) for every

0 
(note that uh ! u in Lploc (
; Rn) if uh ! u in Lp (
0 ; Rn) for every
0 
).
Note: When dealing with scalar functions dened on
, we drop the target space Rn in the notation, and
write just Lp (
) or Lploc (
).
Remark A.1.6.
a) Let 1  p  +1 . We denote by q the conjugate exponent of p, i.e., 1=p + 1=q = 1, where it is
understood that if p = 1 then q = +1 and reciprocally.
b) Let 1  p < +1 . Then the dual space of Lp (
; Rn) is Lq (
; Rn). We point out also that the
dual space of L1 (
; Rn) contains strictly L1 (
; Rn).
c) The notion of weak convergence in Lp (
; Rn) becomes then as follows: If 1  p < +1 , then
fh * f weakly in Lp (
; Rn) if
Z Z
(fh (x); g(x)) dx ! (f(x); g(x)) dx

for every g 2 Lq (
; Rn). For the case p = +1 , fh * f in L1 (
; Rn) weak* if
Z Z
(fh (x); g(x)) dx ! (f(x); g(x)) dx

for every g 2 L1 (
; Rn).
Theorem A.1.7. For every 1  p  +1 , Lp (
; Rn) is a Banach space. It is separable if 1  p < +1
and re exive if 1 < p < +1R. Moreover, L2 (
; Rn) turns out to be a Hilbert space with the scalar product
dened by (f; g)L2 (
;R ) =
(f(x); g(x)) dx .
n

A.1.3. Sobolev spaces

We mention here some important results on Sobolev spaces that we have used in the previous
chapters.
Let us give rst the denition of Sobolev spaces.
 An Introduction to Homogenization and G-convergence 41

Denition A.1.8. Let

be an open subset of Rn and 1  p  +1 . The Sobolev space W 1;p (
) is
dened by
W 1;p (
) = fu 2 Lp (
) : Du 2 Lp (
; Rn)g ;
@u ; @u ; : : :; @u ) denotes the rst order distributional derivative
where Du = (D1 u; D2u; : : :; Dnu) = ( @x 1 @x2 @x
n
of the function u.
On W 1;p (
) we dene the norm
kukW 1 (
) = ;kukpL (
) + kDukpL (
;R )
1=p
;p p p n

Denition A.1.9. Let 1  p < +1 . W01;p (

) denotes the closure of C01(
) in W 1;p (
). W ;1;q (
)
with 1=p + 1=q = 1 indicates the dual space of W01;p(
).
Remark A.1.10. If p = 2, the notations H 1;2(
) or H 1 (
) are very common for W 1;2(
). Moreover,
H01;2(
) or H01 (
) stand for W01;2(
). ThePspaces H 1;2(
) and H01;2(
) are naturally endowed with the
scalar product (u; v)H 1 2 (
) = (u; v)L2 (
) + ni=1 (Di u; Div)L2 (
) which induces the norm kukH 1 2 (
) :
; ;

Theorem A.1.11. The space W 1;p(

) is a Banach space for 1  p  +1 . W 1;p(
) is separable if
1  p < +1 and re exive if 1 < p < +1 .
Moreover, the space W01;p(
) endowed with the norm induced by W 1;p (
) is a separable Banach
space; it is re exive if 1 < p < +1 .
The spaces H 1;2(
) and H01;2(
) are separable Hilbert spaces.
We now quote the Sobolev and Rellich-Kondrachov imbedding theorems.
Theorem A.1.12. Let
be a bounded open subset of Rn with Lipschitz boundary.
i) If 1  p < n , then
W 1;p(
)  Lq (
) for every 1  q  np=(n ; p)
and the imbedding is compact for every 1  q < np=(n ; p) .
ii) If p = n , then
W 1;p (
)  Lq (
) for every 1  q < +1
and the imbedding is compact.
iii) If p > n , then
W 1;p (
)  C(
)
and the imbedding is compact.
Remark A.1.13.
a) The regularity of the boundary @
in the theorem can be weakened (see, for example, [1]). Note
that if the space W 1;p(
) is replaced by W01;p(
), then no regularity of the boundary is required.
b) The compact imbedding can be read in the following way. Let
uh * u weakly in W 1;p(
):
Case I : If 1  p < n, then uh ! u strongly in Lq (
); 1  q < np=(n ; p);
Case II : If p = n, then uh ! u strongly in Lq (
); 1  q < +1 ;
Case III : If p > n, then uh ! u strongly in L1 (
) :
Let us state two important inequalities.
42 A. Defranceschi

Theorem 1.14.
i) (Poincare inequality) Let
be a bounded open set and let 1  p < +1 . Then there exists a
constant K > 0 such that
kukL (
)  K kDukL (
;R )
p p n

for every u 2 W01;p(

) .
ii) (Poincare-Wirtinger inequality) Let
be a bounded open convex set and let 1  p < +1 . Then
there exists a constant K > 0 such that
ku ; M
(u)kL (
)  K kDukL (
;R
p p n )

for every u 2 W 1;p(

) .

Remark A.1.15. From the previous theorem it follows that kDukL (

;R ) denes a norm on W01;p(
),
p n

denoted by kukW01 (
) , which is equivalent to the norm kukW 1 (
) .
;p ;p

A.1.4. Extension and convergence lemmas for periodic functions

Let us start with the extension properties of periodic functions (see [45] Annexe 2). Let Y = ]0; 1[n
be the unit cube in Rn and let 1 < p < +1 . By W]1;p (Y ) we denote the subset of W 1;p (Y ) of all the
functions u with mean value zero which have the same trace on the opposite faces of Y . In the case p = 2
we use the notation H]1;2(Y ).
Lemma A.1.16. Let f 2 W]1;p(Y ) . Then f can be extended by periodicity to an element of Wloc
1;p (Rn) .

Lemma A.1.17. Let g 2 Lq (Y ; Rn) such that

R (g; Dv) = 0 for every v 2 W]1;p (Y ) . Then g can
Y
be extended by periodicity to an element of Lqloc (Rn; Rn) , still denoted by g such that ;div g = 0 in
D0 (Rn) .
Let us conclude this section with a result for the weak convergence on Lp spaces which has been
used frequently in the previous chapters. For a proof of it we refer to [45] Annexe 2, [21] Chapter 2,
Theorem 1.5.
Theorem A.1.18. Let f 2 Lp(Y ) . Then f can be extended by periodicity to a function (still denoted by
f ) belonging to Lploc (Rn ) . Moreover, if ("h ) is a sequence of positive real numbers converging to 0 and
fh (x) = f( "x ) , then Z
fh * M(f) = jY1 j f(y) dy weakly in Lploc (Rn )
h

Y
if 1  p < +1 , and
 M(f) in L1 (Rn ) weak*
fh *
if p = +1 .
It is clear that the above results still hold for Y not necessarily the unit cube in Rn but a paral-
lelogram of the type described in Section 1.
 An Introduction to Homogenization and G-convergence 43

Remark A.1.19. Let us point out some features of the weak convergence. To this aim, let us consider
Y = ]0; 2[ and f(x) = sin x. Let ("h ) be a sequence of positive numbers converging to 0. By Theorem
A.1.18 we have that fh (x) = f( "x ) converges to 0 in L1 (Y ) weak* (hence weakly in L2 (Y )). In
h

particular, Z 2
fh (x) dx ! 2 1 Z 2 sin y dy = 0 ;
0 0
i.e., the mean values of fh converges to 0. On the other hand, we have that (fh ) does not converge a.e.
on Y . Furthermore,
Z 2 Z
(A:1:1) kfh ; 0k2L2(Y ) = sin2( "x )dx ! ( 1 sin2 y dy)2 =  6= 0 ;
0 h 0
which shows that we do not have convergence of (fh ) to f in the strong topology of L2 (Y ).
This example shows also another mathematical diculty one meets by handling with weak conver-
gent sequences. More precisely, if two sequences and their product converge in the weak topology, the
limit of the product ist not equal, in general, to the product of the limits. Indeed, (A.1.1) proves that
fh2 = fh  fh does not converge weakly in L2 (Y ) to 0.
A.2. A Compensated Compactness Lemma
noindent The next lemma, which has been used frequently in the previous chapters, helps to over-
come the diculties present by passing to the limit in products of only weakly convergent sequences.
Lemma A.2.1. Let 1 < p < +1 . Let (uh ) be a sequence converging to u weakly in W 1;p (
) , and let
(gh ) be a sequence in Lq (
; Rn) converging weakly to g in Lq (
; Rn) . Moreover assume that (;div gh )
converges to ;div g strongly in W ;1;q (
) . Then
Z Z
(gh ; Duh )' dx ! (g; Du)' dx

1
for every ' 2 C0 (
) .
Proof. The lemma is a simple case of compensated compactness (see ([34], [35], [49]). It can be proved by
observing that Z Z
(gh ; Duh )' dx =< ;div gh ; uh ' > ; uh (gh ; D') dx

for every ' 2 C01 (

).
Note that (gh ; Duh ) is the product of two sequences which converge only in the weak topology, and
that by passing to the limit we get the product of the limits. This fact is known as the phenomenon of
\compensated compactness".
A.3. Abstract existence theorems
A.3.1. Lax-Milgram Lemma
Let H be a Hilbert space. A bilinear form a on H is called continuous (or bounded) if there exists a
positive constant K such that
ja(u; v)j  K kxkkvk for every u, v 2 H
and coercive if there exists a positive constant  such that
a(u; u)  kuk2 for every u 2 H :
A particular example of continuous, coercive bilinear form is the scalar product of H itself.
44 A. Defranceschi

Lemma A.3.1. Let a be a continuous, coercive bilinear form on a Hilbert space H . Then for every
bounded linear functional f in H  , there exists a unique element u 2 H such that
a(u; v) = hf; vi for every v 2 H :

For a proof of this classical lemma we refer to [16], [29].

A.3.2. Maximal monotone operators
Let X be a Banach space and X  its dual space. Let A be a single-valued operator from D(A) to
X  , where D(A) is a linear subspace of X and is called the domain of A. The range R(A) of A is the
set of all points f of X  such that there exists x 2 D(A) with Ax = f . Then
a) A is said to be monotone if
hAx1 ; Ax2; x1 ; x2i  0 for every x1 , x2 2 D(A) :
b) A is said to be strictly monotone if for every x1 , x2 2 D(A)
hAx1 ; Ax2; x1 ; x2i = 0 implies x1 = x2
c) A is said to be maximal monotone if for every pair [x; y] 2 X  X  such that
hy ; A; x ;  i  0 for every  2 D(A)
it follows that y = Ax.
d) A is said to be hemicontinuous if
lim A(x + ty) = Ax weakly in X 
t!0
for any x 2 D(A) and y 2 X such that x + ty 2 D(A) for 0  t  1.
Theorem A.3.2. Let X be a Banach space and let A : X ! X  be everywhere dened (i.e., D(A) = X ),
monotone and hemicontinuous. Then A is maximal monotone. In addition, if X is re exive and A is
coercive, i.e.,
lim hAx; xi = +1 ;
kxk!1 kxk
then R(A) = X  .
Proof. If X is a Hilbert space the proof of the previous theorem can be found in [15]. For the general
case see [38] Chapter III, Corollary 2.3 and Theorem 2.10, or [30] Chapter 2, Theorem 2.1.
A.4. Proof of Proposition 7.7 and of Proposition 7.8
Proof of Proposition 7.7. Since F is separable, there exists a countable dense subset X of F . By the
assumptions on T h and G and by using a diagonalization argument there exists a subsequence (T (h) ) of
(T h ) such that for every x 2 X , (T (h) x) converges weakly to a limit in G. Let us denote this limit by
T 0 x.
Now, given f 2 F and g 2 G , by approximating f by x 2 X one proves easily that the sequence
(hT (h) f; g iG;G ) is a Cauchy sequence in R . Let us denote by hT 0f; g i its limit. The linearity of T 0 is
immediate; by taking into account the weak lower semicontinuity of the norm and assumption (ii) we get
kT 0 f kG = k hlim
!1
T (h) f kG  lim
h!1
inf kT (h) f kG  ckf kG :
Hence T 0 2 L(F; G). The proof of Proposition 7.7 is then accomplished.
 An Introduction to Homogenization and G-convergence 45

Proof of Proposition 7.8. Since (i) holds, we can dene the bilinear form ah : V  V ! R by
ah (u; v) = hT h u; vi
for every u, v 2 V . By the hypotheses (ii) and (iii) it follows immediately that ah is continuous and
coercive. Hence, by the Lax-Milgram lemma for every f 2 V  there exists a unique function u 2 V such
that
ah (u; v) = hf; vi for every v 2 V :
It turns out that the operators T h are invertible and
k(T h );1 f kV = kukV  1 kf kV 
for every f 2 V  ; thus k(T h );1 kL(V ;V  )  1 : By Proposition 7.7 there exist a subsequence (h) of h
and an operator S 2 L(V  ; V ) such that for every f 2 V 
(T (h) );1 f * Sf weakly in V :
We get
h(T (h) );1 f; f iV;V  = h(T (h) );1 f; T (h) (T (h) );1f iV;V  
 k(T (h) );1f k2V  2 kf k2V  :
Hence, for every f 2 V 
hSf; f iV;V   2 kf k2V  :
This proves that S is coercive. This fact together with the property that S 2 L(V  ; V ) ensures that S is
invertible. Let us denote by T 0 2 L(V ; V  ) its inverse. Note that for every v 2 V we have
 0 2
 2 kT vkV   hST v; T viV;V  hv; T viV;V  kvkV kT vkV :
0 0  0  0 

We conclude that 2
kT 0kL(V ;V  )   :
On the other hand, we have for every f 2 V 
k(T (h) );1f k2V  hT (h) (T (h) );1 f; (T (h) );1 f iV  ;V = hf; (T (h) );1 f iV  ;V ;
by taking the weak lower semicontinuity of the norm in V into account we obtain for every f 2 V 
kSf k2V  hf; Sf iV  ;V :
By taking in particular f = T 0v we conclude that
kvk2V  hT 0 v; viV  ;V
for every v 2 V , which concludes the proof of Proposition 7.8.
Acknowledgments. The author thanks the Department of Mathematics of the University of Trento for
the hospitality during the preparation of these notes.
46 A. Defranceschi

References
[1] ADAMS R.: Sobolev Spaces. Academic Press, New York, 1975.
[2] ATTOUCH H.: Introduction a l'homogeneisation d'inequations variationnelles. Rend. Sem. Mat.
Univers. Politecn. Torino 40 2 (1981), 1-23.
[3] ATTOUCH H.: Variational Convergence for Functions and Operators. Pitman, London, 1984.
[4] BABUSKA I.: Solution of interface problems by homogenization I, II, III. Tech. Note, Univ. of
Maryland (1974-75).
[5] BABUSKA I.: Homogenization and its application. Mathematical and computational problems.
Proc. Symp. Numerical Sol. Partial Di. Eq., III, Maryland (1975), Academic Press, (1976), 89-116.
[6] BAKHVALOV N.S., PANASENKO G.P.: Averaged Processes in Periodic Media. Nauka, Moscow,
1984.
[7] BENSOUSSAN A.: Homogenization theory. Atti del S.A.F.A. III \A Survey on the Theoretical and
Numerical Trends in Nonlinear Analysis" (Bari, 1978), 1-46 Conferenze del Seminario di Matem-
atica dell'Universita di Bari 158, Laterza, Bari, 1979.
[8] BENSOUSSAN A., BOCCARDO L., MURAT F.: H-convergence for quasilinear elliptic equations
with quadratic growth. Preprint Univ. Pierre et Marie Curie, Lab. D'Analyse Numerique (1990).
[9] BENSOUSSAN A., LIONS J.L., PAPANICOLAOU G.: Asymptotic Analysis for Periodic Structures.
North Holland, Amsterdam, 1978.
[10] BERGMAN D.J., LIONS J.L., PAPANICOLAOU G., MURAT F., TARTAR L., SANCHEZ-PALENCIA
E.: Les Methodes de L'Homogeneisation: Theorie et Applications en Physique. Ecole d'Ete d'Analyse
Numerique C.E.A.-E.D.F.-INRIA (Breau-sans-Nappe, 1983), Collection de la Direction des Etudes
et Reserches d'Electricite de France, 57, Eyrolles, Paris, 1985.
[11] BOCCARDO L.: Homogeneisation pour une classe d'equations fortement non lineaires. C.R. Acad.
Sci. Paris 306 (1988), 253-256.
[12] BOCCARDO L., GALLOUET T.: Homogenization with jumping nonlinearities. Ann. Mat. Pura
Appl. 138 (1984), 211-221.
[13] BOCCARDO L., MURAT F.: Homogeneisation de problemes quasi-lineaires. Studio di problemi-
limite della analisi funzionale (Bressanone, 1981), 13-51, Pitagora ed. Bologna, 1982.
[14] BOCCARDO L., MURAT F.: Remarques sur l'homogeneisation de certaines problemes quasilineaires.
Portugal. Math. 41 (1982), 535-562.
[15] BREZIS H.: Operateurs Maximaux Monotones et Semi-groupes de Contractions dans les Espaces
de Hilbert. North Holland, Amsterdam, 1973.
[16] BREZIS H.: Analyse Fonctionnelle. Masson, Paris, 1983.
[17] CARBONE L., SBORDONE C.: Some properties of ;-limits of integral functionals. Ann. Mat. Pura
Appl. 122 (1979), 1-60.
[18] CHIADO PIAT V., DAL MASO G., DEFRANCESCHI A.: G-convergence of monotone operators.
Ann. Inst. H. Poincare. Anal. Non Lineaire 7 3 (1990), 123-160.
[19] CHIADO PIAT V., DEFRANCESCHI A.: Homogenization of monotone operators. Nonlinear Anal.
14 (1990), 717-732.
[20] CHIADO PIAT V., DEFRANCESCHI A.: Homogenization of quasi-linear equations with natural
growth terms. Manuscripta Math. 68 (1990), 229-247.
[21] DACOROGNA B.: Direct Methods in the Calculus of Variations. Springer Verlag, Berlin, 1989.
[22] DAL MASO G., DEFRANCESCHI A.: Correctors for the homogenization of monotone operators.
Dierential and Integral Equations 3 (1990), 1151-1166.
[23] DE GIORGI E., SPAGNOLO S.: Sulla convergenza degli integrali dell'energia per operatori ellittici
del secondo ordine. Boll. Un. Mat. Ital. (4) 8 (1973), 391-411.
 An Introduction to Homogenization and G-convergence 47

[24] DEL VECCHIO T.: On the homogenization of a class of pseudomonotone operators in divergence
form. Boll. Un. Mat. Ital. 7 (1991), 369-388.
[25] DUNFORD N, SCHWARTZ J.T.: Linear Operators. Wiley, New York, 1957.
[26] FOLLAND G.B.: Real Analysis. Modern Techniques and their Applications. Wiley, New York, 1984.
[27] FUSCO N., MOSCARIELLO G.: On the homogenization of quasilinear divergence structure oper-
ators. Ann. Mat. Pura Appl. 146 (1987), 1-13.
[28] FUSCO N., MOSCARIELLO G.: Further results on the homogenization of quasilinear divergence
structure operators. Preprint Univ. Napoli (1986).
[29] GILBARG D., TRUDINGER N.S.: Elliptic Partial Dierential Equations of Second Order. Springer
Verlag, Berlin, 1977.
[30] LIONS J.: Quelques methodes de resolution des problemes aux limites non lineaires. Dunod Gauthier-
Villars, Paris (1969).
[31] LIONS J.: Some Methods in the Mathematical Analysis of Systems and their Control. Science Press,
Beijing, China. Gordon and Breach, New York, 1981.
[32] MARCELLINI P.: Periodic solutions and homogenization of non linear variational problems. Ann.
Mat. Pura Appl. (4) 117 (1978), 139-152.
[33] MURAT F.: H-convergence. Seminaire d'Analyse Fonctionelle et Numerique de l'Universite d'Alger,
1977.
[34] MURAT F.: Compacite par compensation. Ann. Sc. Norm. Sup. Pisa Cl. Sci. (4) 5 (1978), 489-507.
[35] MURAT F.: Compacite par compensation II. Proc. Int. Meeting on "Recent Methods in Nonlinear
Analysis", Rome 1978, ed. E. De Giorgi, E. Magenes, U. Mosco, 245-256, Pitagora ed. Bologna,
1979.
[36] MURAT F.: Correctors for monotone problems in homogenization. Manuscript, 1990.
[37] OLEINIK O.A., SHAMAEV A.S., YOSIFIAN G.A.: Mathematical Problems of the Theory of
Strongly Nonhomogeneous Elastic Media. Nauka, Moscow, 1990.
[38] PASCALI D., SBURLAN S.: Nonlinear Mappings of Monotone Type. Editura Academiei, Bucuresti,
1978.
[39] RAITUM U.E.: On the G-convergence of quasilinear elliptic operators with unbounded coecients.
Soviet Math. Dokl. 24 (1981), 472-476.
[40] SANCHEZ-PALENCIA E.: Non-Homogeneous Media and Vibration Theory. Lecture Notes in Physics
127, Springer-Verlag, Berlin, 1980.
[41] SBORDONE C.: La ;-convergenza e la G-convergenza. Rend. Sem. Mat. Univers. Politecn. Torino
40 2 (1982), 25-51.
[42] SPAGNOLO S.: Sul limite delle soluzioni di problemi di Cauchy relativi all'equazione del calore.
Ann. Sc. Norm. Sup. Pisa Cl. Sci. (3) 21 (1967), 657-699.
[43] SPAGNOLO S.: Sulla convergenza delle soluzioni di equazioni paraboliche ed ellittiche. Ann. Sc.
Norm. Sup. Pisa Cl. Sci. (3) 22 (1968), 571-597.
[44] SPAGNOLO S.: Convergence in energy for elliptic operators. Proc. Third Symp. Numer. Solut.
Partial Di. Equat. (College Park, 1975), 469-498, Academic Press, San Diego, 1976.
[45] SUQUET P.: Plasticite et Homogeneisation. Thesis Univ. Paris VI, 1982.
[46] TARTAR L.: Convergence d'operateurs dierentiels. Analisi convessa, Roma, 1977.
[47] TARTAR L.: Cours Peccot au College de France. Paris, 1977.
[48] TARTAR L.: Quelques remarques sur l'homogeneisation. Proc. of the Japan-France seminar 1976
"Functional analysis and numerical analysis", 469-482, Japan Society for the Promotion of Science,
1978.
48 A. Defranceschi

[49] TARTAR L.: Compensated compactness and applications to partial dierential equations. Nonlinear
analysis and mechanics. Heriot-Watt Symposium vol. IV. Research Notes in Mathematics 39, 136-
211, Pitman, London, 1979.
[50] TARTAR L.: Homogeneisation d'operateurs monotones. Manuscript, 1981.
[51] YOSIDA K.: Functional Analysis. Springer Verlag, Berlin, 1966.
[52] W.P. ZIEMER: Weakly Dierentiable Functions. Springer-Verlag, Berlin, 1989.
A wider list of references is contained in the book by G. Dal Maso: An Introduction to ;-convergence
(Birkhauser, Boston, 1993).

Anneliese Defranceschi
Dipartimento di Matematica
Universita di Parma
Via M. D'Azeglio, 85/A
I-43100 PARMA (Italy)