Khayyam J. Math.

6 (2020), no 1, 57–72
DOI: 10.22034/kjm.2019.97170

STRONGLY QUASILINEAR PARABOLIC SYSTEMS

IN DIVERGENCE FORM WITH WEAK MONOTONICITY

∗
ELHOUSSINE AZROUL1 AND FARAH BALAADICH1

Communicated by P.I. Naumkin

Abstract. The existence of solutions to the strongly quasilinear parabolic

system
∂u
− div σ(x, t, u, Du) + g(x, t, u, Du) = f,
∂t
0
is proved, where the source term f is assumed to belong to Lp (0, T ;
0
W −1,p (Ω; Rm )). Further, we prove the existence of a weak solution by means
of the Young measures under mild monotonicity assumptions on σ.

1. Introduction
Let Ω be a bounded open subset of Rn and let Q be the cylinder Ω × (0, T )
with some given T > 0. By ∂Q we denote the boundary of Q and Mm×n the real
vector space of m × n matrices equipped with the inner product A : B = Ai,j Bi,j
(with conventional summation). Consider first the quasilinear parabolic initial-
boundary value system
∂u
− div σ(x, t, u, Du) = f in Q,
∂t
u(x, t) = 0 on ∂Q, (1.1)
u(x, 0) = u0 (x) in Ω,
0 0
where u : Q → Rm . In (1.1) the right hand side f belongs to Lp (0, T ; W −1,p (Ω; Rm ))
for some p ∈ (1, ∞). In [13], Young introduced Young measure as a powerful tool

Date: Received: 05 December 2018; Revised: 6 October 2019; Accepted: 19 October 2019.
∗
Corresponding author.
2010 Mathematics Subject Classification. Primary 35K55; Secondary 35D30.
Key words and phrases. Quasilinear parabolic systems, weak monotonicity, weak solution,
Young measures.
57
58 E. AZROUL, F. BALAADICH

to describe the weak limit of sequences. N. Hungerbühler [8] obtained the exis-
tence of a weak solution for (1.1) by using the concept of Young measures. The
author assumed weak monotonicity assumptions on σ.
If A(u) = −div σ(x, t, u, Du), u : Q → R and A is a classical operator of
the Leray-Lions type with respect to the Sobolev space Lp (0, T ; W01,p (Ω; Rm ))
for some 1 < p < ∞, then the existence of solutions for (1.1) was proved in
[3, 10, 11, 12]. The authors required the strict monotonicity or monotonicity in
the variables (u, F ) ∈ Ω × Rn . Nevertheless, we will not use the previous type of
monotonicity.
In this paper, we will be using the Young measures and Galerkin method to
prove the existence result for the following strongly quasilinear parabolic system
∂u
− div σ(x, t, u, Du) + g(x, t, u, Du) = f in Q, (1.2)
∂t
u(x, t) = 0 on ∂Q, (1.3)
u(x, 0) = u0 (x) in Ω. (1.4)
The problem (1.2)-(1.4) can be seen as a more general form of (1.1), where g :
Q × Rm × Mm×n → Rm . Similar problems to (1.2)-(1.4) were studied, we refer
the reader [1, 4, 6].
This article is organized as follows: in Section 2, we present our assumptions
and main result. Section 3 is a brief review of Young measures. Section 4 deals
with the Galerkin approximations and necessary a priori estimates. Section 5
concerns the identification of weak limits by means of Young measures, while
Section 6 is devoted to the proof of the main result.

2. Assumptions and main results

Let Ω be a bounded open subset of Rn and set Q = Ω × (0, T ) for T >
0. Throughout this paper, we denote Qτ = Ω × (0, τ ) for every τ ∈ [0, T ].
Consider the problem (1.2)–(1.4), where σ : Q × Rm × Mm×n → Mm×n and
g : Q × Rm × Mm×n → Rm satisfy the following assumptions:
(H0) σ and g are Carathéodory functions (i.e., measurable w.r.t (x, t) ∈ Q and
continuous w.r.t other variables).
0
(H1) There exist c1 ≥ 0, β > 0, d1 ∈ Lp (Q) and d2 ∈ L1 (Q) such that
|σ(x, t, u, A)| ≤ d1 (x, t) + c1 |u|p−1 + |A|p−1 ,

σ(x, t, u, A) : A + g(x, t, u, A).A ≥ −d2 (x, t) + β|A|p .

(H2) σ satisfies one of the following conditions:
(a) For all (x, t) ∈ Q, A 7→ σ(x, t, u, A) is a C 1 -function and is monotone,
that is, for all (x, t) ∈ Q, u ∈ Rm and A, B ∈ Mm×n , we have


σ(x, t, u, A) − σ(x, t, u, B) : (A − B) ≥ 0.
(b) There exists a function W : Q × Rm × Mm×n → R such that σ(x, t, u, A) =
∂W
∂A
(x, t, u, A) and A → W (x, t, u, A) is convex and C 1 for all (x, t) ∈ Q
and u ∈ Rm .
 STRONGLY QUASILINEAR PARABOLIC SYSTEMS... 59

(c) σ is strictly monotone, that is, σ is monotone and

σ(x, t, u, A) − σ(x, t, u, B) : (A − B) = 0 ⇒ A = B.
(d)
Z Z


σ(x, t, u, λ) − σ(x, t, u, λ) : (λ − λ)dν(λ)dxdt > 0,
Q Mm×n

where λ = hν(x,t) , idi and ν = {ν(x,t) }(x,t)∈Q is any family of Young mea-
sures generated by a bounded sequence in Lp (Q) and not a Dirac measure
for a.e. (x, t) ∈ Q.
(H3) g satisfies one of the following conditions:
0
(i) There exist c2 ≥ 0 and d2 ∈ Lp (Q) such that
|g(x, t, u, A)| ≤ d2 (x, t) + c2 |u|p−1 + |A|p−1 .

(ii) The function g is independent of the fourth variable, or, for a.e. (x, t) ∈ Q
and all u ∈ Rm , the mapping A → g(x, t, u, A) is linear.
Remark 2.1. Assumptions (H1) and (H3)(i) state standard growth and coercivity
conditions. The assumption (H1)(b) allows to take a potential W (x, t, u, A) which
is only convex but not strictly convex in A ∈ Mm×n and to consider (1.2) with
σ(x, t, u, A) = ∂W
∂A
(x, t, u, A). Note that if W is assumed to be strictly convex,
then σ becomes strict monotone. Thus, the standard method may apply. Finally,
(H2)(d) states the notion of strict p-quasimonotone in terms of gradient Young
measures.
We shall prove the following existence theorem.
Theorem 2.2. Suppose that the conditions (H0)–(H1) are satisfied. Let u0 ∈
0 0
L2 (Ω; Rm ) and f ∈ Lp (0, T ; W −1,p (Ω; Rm )) be given. Then
(1) if σ satisfies one of the conditions (H2)(a) or (b), then for every g satis-
fying (H3)(ii), the system (1.2)–(1.4) has a weak solution.
(2) if σ satisfies one of the conditions (H2)(c) or (d), then for each g satisfying
(H3)(i), the system (1.2)–(1.4) has a weak solution.
Remark 2.3. A simple model of our problem is as follows:
∂u
− div (|Du|p−2 Du) + |u|p−2 u = f in Q,
∂t
u(x, t) = 0 on ∂Q,
u(x, 0) = u0 (x) in Ω.
For the potential W , one can take W := p1 |A|p .

3. A review of Young measures

In the following, C0 (Rm ) denotes the closure of the space of continuous functions
on Rm with compact support with respect to the k.k∞ -norm. Its dual space can
60 E. AZROUL, F. BALAADICH

be identified with M(Rm ), the space of signed Radon measures with finite mass.
The related duality pairing is given by
Z
hν, ϕi = ϕ(λ)dν(λ).
Rm
R
Note that, as id(λ) = λ then hν, idi = Rm λdν(λ).
Definition 3.1. Assume that the sequence {wj }j≥1 is bounded in L∞ (Ω; Rm ).
Then there exist a subsequence {wk }k and a Borel probability measure νx on Rm
for a.e. x ∈ Ω, such that for almost each ϕ ∈ C(Rm ) we have
ϕ(wk ) *∗ ϕ weakly in L∞ (Ω),
where Z
ϕ(x) = ϕ(λ)dνx (λ).
Rm
We call ν = {νx }x∈Ω the family of Young measures associated with the subse-
quences {wk }k .
The fundamental theorem on Young measures may be stated in the following
lemma.
Lemma 3.2 ([5]). Let Ω ⊂ Rn be Lebesgue measurable (not necessarily bounded)
and let wj : Ω → Rm , j = 1, 2, . . . be a sequence of Lebesgue measurable functions.
Then there exist a subsequence wk and a family {νx }x∈Ω of nonnegative Radon
measures on Rm , such that
R
(i) kνx kM(Rm ) := Rm dνx ≤ 1 for almost x ∈ Ω.
(ii) ϕ(wk ) *∗ ϕ weakly in L∞ (Ω) for all ϕ ∈ C0 (Rm ), where ϕ(x) = hνx , ϕi.
(iii) If
lim sup {x ∈ Ω ∩ BR (0) : |wk (x)| ≥ L} = 0 (3.1)
L→∞ k

for all R > 0, then kνx k = 1 for a.e. x ∈ Ω, and for all measurable
Ω0 ⊂ Ω, there holds ϕ(wk ) * ϕ = hνx , ϕi weakly in L1 (Ω0 ) for a continuous
function ϕ provided the sequence ϕ(wk ) is weakly precompact in L1 (Ω0 ).
The following lemmas are considered as the applications of the fundamental
theorem on Young measures (Lemma 3.2), which will be needed in what follows.
Lemma 3.3 ([7]). If |Ω| < ∞ and νx is the Young measure generated by the
(whole) sequence wj , then there holds
wj → w in measure ⇔ νx = δw(x) for a.e. x ∈ Ω.
Lemma 3.4 ([7]). Let F : Ω×Rm ×Mm×n → R be a Carathéodory function and let
uk : Ω → Rm be a sequence of measurable functions such that uk → u in measure
and such that Duk generates the Young measure νx , with kνx kM(Mm×n ) = 1 for
almost every x ∈ Ω. Then
Z Z Z
lim inf F (x, uk (x), Duk (x))dx ≥ F (x, u, λ)dνx (λ)dx,
k→∞ Ω0 Ω0 Mm×n
−
provided that the negative part F (x, uk (x), Duk (x)) is equi-integrable.
 STRONGLY QUASILINEAR PARABOLIC SYSTEMS... 61

Remark 3.5. (1) Lemma 3.4 is a Fatou-type inequality.

(2) Under condition (3.1), it was proved [2] that for any measurable Ω0 ⊂ Ω,
ϕ(., uk ) * hνx , ϕ(x, .)i in L1 (Ω0 ),
for every Carathéodory function ϕ : Ω0 ×Rm → R such that {ϕ(., uk )} is se-
quentially weakly relative compact in L1 (Ω0 ). Further, the author showed
that if uk generates the Young measure νx , then for ϕ ∈ L1 (Ω; C0 (Rm ))
we have
Z Z Z
lim ϕ(x, uk (x))dx = ϕ(x, λ)dνx (λ)dx.
k→∞ Ω Ω Rm

4. Galerkin approximation
We choose an L2 (Ω; Rm )-orthonormal base {wi }i≥1 such that
C 1 (Ω;Rm )
{wi }i≥1 ⊂ C0∞ (Ω; Rm ), C0∞ (Ω; Rm ) ⊂ ∪ Vk ,
k≥1

where Vk = span{w1 , . . . , wk }. Define the following approach for approximating

solutions of (1.2)–(1.4):
k
X
uk (x, t) = αki (t)wi (x), (4.1)
i=1

where αki : [0, T ] → R are measurable bounded functions. Assume that uk ∈

Lp (0, T ; W01,p (Ω; Rm )). Thus uk satisfies the boundary condition (1.3) by con-
struction. For the initial condition (1.4), one can choose the initial coefficients
αki (0) := (u0 , wi )L2 , with (., .) denotes the inner product of L2 , such that
k
X
uk (., 0) = αki (0)wi (.) → u0 in L2 (Ω)
i=1

as k → ∞. To complete the construction of uk , it remains to determine the

coefficients αki (t). For this, let k ∈ N be fixed (for the moment), 0 < τ < T
and I = [0, τ ]. Furthermore, we choose r > 0 large enough, such that the set
Br (0) := B(0, r) ⊂ Rk contains the vectors (α1k (0), . . . , αkk (0)). Consider the
function
Θ : I × Br (0) → Rk
 Z k
X k
X
(t, α1 , . . . , αk ) 7→ hf (t), wj i − σ(x, t, α i wi , αi Dwi ) : Dwj dx
Ω i=1 i=1
Z k
X k
X 
− g(x, t, αi wi , αi Dwi ).wj dx ,
Ω j=1,...,k
i=1 i=1
0
where h., .i denotes the dual pairing of W −1,p (Ω) and W01,p (Ω). The operator
Θ is a Carathéodory function by the condition (H0). Next, we will estimate
62 E. AZROUL, F. BALAADICH

Θj . By using the conditions (H1) and (H3)(i), one gets together with the Hölder
inequality
Z k
X k
X
σ(x, t, αi w i , αi Dwi ) : Dwj dx
Ω i=1 i=1
Z k k  10  Z  p1
p0 (4.2)
X X p p
≤ σ(x, t, αi wi , αi Dwi ) dx |Dwj | dx
Ω i=1 i=1 Ω
Z
≤c d1 (x, t)dx + c,
Ω

and
Z k
X k
X Z
g(x, t, αi wi , αi Dwi ).wj dx ≤ c d2 (x, t)dx + c, (4.3)
Ω i=1 i=1 Ω

where c depends on k and r but not on t.

Note that (4.2) and (4.3) are obtained by the following arguments: firstly, we
have W0s,2 (Ω) ⊂ W01,p (Ω) for s ≥ 1+n( 12 − p1 ), secondly Dwj ∈ W s−1,2 (Ω) ⊂ L∞ (Ω)
for wj ∈ W s,2 (Ω). For the first term in the definition of Θ, we have
hf (t), wj i ≤ kf (t)k−1,p0 kwj k1,p .
As a consequence, the j th term of Θ can be estimated as follows:
|Θj (t, α1 , . . . , αk )| ≤ c(r, k)b(t) (4.4)
uniformly on I × Br (0), where c(r, k) is a constant, which depends on r and k,
and where b(t) ∈ L1 (I) does not depend on r and k. Thus, the Carathéodory
existence result on ordinary differential equations (cf. Kamke [9]) applied to the
system
 0
αj (t) = Θj (t, α1 (t), . . . , αk (t)),
(4.5)
αj (0) = αkj (0),
(for j ∈ {1, . . . , k}) ensures the existence of a distributional, continuous solution
αj (depending on k) of (4.5) on a time interval [0, τ 0 ), where τ 0 > 0, a priori, may
depend on k. Furthermore, the corresponding integral equation
Z t
αj (t) = αj (0) + Θj (t, α1 (s), . . . , αk (s))ds
0

holds on [0, τ 0 ). Hence

k
X
uk (x, t) = αki (t)wi (x)
i=1
is the desired solution to the system of ordinary differential equations
Z Z
∂uk

, wj L2 + σ(x, t, uk , Duk ) : Dwj dx + g(x, t, uk , Duk ).wj dx
∂t Ω Ω (4.6)
= hf (t), wj i,
 STRONGLY QUASILINEAR PARABOLIC SYSTEMS... 63

with the initial condition uk (., 0) = ki=1 αki (0)wi (.) → u0 in L2 (Ω) as k → ∞.
P
Now, we will extend the local solution defined on [0, τ 0 ) to a global one. For this,
we multiply each side of (4.6) by αkj (t) and we sum. This gives for an arbitrary
time τ ∈ [0, T )
Z Z
∂uk

uk dxdt + σ(x, t, uk , Duk ) : Duk + g(x, t, uk , Duk ).uk dxdt
Qτ ∂t Qτ
Z τ
= hf (t), uk idt,
0
which is denoted as I1 + I2 = I3 . By integrating and (H1), we have
1 1
I1 = kuk (., τ )k2L2 (Ω) − kuk (., 0)k2L2 (Ω) ,
2 2
and Z Z
I2 ≥ − d2 (x, t)dxdt + β |Duk |p dxdt.
Qτ Qτ
By Hölder’s inequality
|I3 | ≤ kf kLp0 (0,T ;W −1,p0 (Ω)) kuk kLp (0,T ;W 1,p (Ω)) .
0

From the estimations on I ,  = 1, 2, 3, we deduce

αki (τ ) i=1,...,k |2Rk = kuk (., τ )k2L2 (Ω) ≤ c,

where c is a constant independent of τ (and of k).

Consider now

M := t ∈ [0, T ) : there exists a weak solution of (4.5) on [0, t) .
We have M is nonempty, because it contains a local solution. Moreover, thanks
to [8], we then have M is an open set which is also closed. Thus M = [0, T ).
From the estimations on I ,  = 1, 2, 3, we conclude that the sequence (uk )k is
bounded in Lp (0, T ; W01,p (Ω; Rm ))∩L∞ (0, T ; L2 (Ω; Rm )). Therefore, by extracting
a suitable subsequence (still denoted by (uk )k ), we may assume
uk * u in Lp (0, T ; W01,p (Ω; Rm )), (4.7)
uk *∗ u in L∞ (0, T ; L2 (Ω; Rm )). (4.8)
p 1,p m ∞ 2 m
The function u ∈ L (0, T ; W0 (Ω; R )) ∩ L (0, T ; L (Ω; R )) is a candidate to
be a weak solution for the problem (1.2)–(1.4). Using the growth condition in
(H1)
 and (H3), together with  (4.7), we can extract a suitable subsequence of
− div σ(x, t, uk , Duk ) and g(x, t, uk , Duk ) such that
0 0
− div σ(x, t, uk , Duk ) * χ in Lp (0, T ; W −1,p (Ω; Rm )) (4.9)
and
0 0
g(x, t, uk , Duk ) * ξ in Lp (0, T ; W −1,p (Ω; Rm )), (4.10)
p 0 −1,p0 m
where χ, ξ ∈ L (0, T ; W (Ω; R )).
Since (uk )k is bounded in L∞ (0, T ; L2 (Ω; Rm )), there exists a subsequence,
which is again denoted by (uk )k , such that
uk (., T ) * z in L2 (Ω; Rm ) as k → ∞.
64 E. AZROUL, F. BALAADICH

We will prove that z = u(., T ) and u(., 0) = u0 (.). For simplicity, we denote
u(., T ) as u(T ) and u(., 0) as u(0). For every φ ∈ C ∞ ([0, T ]) and v ∈ Vj , j ≤ k,
we have
Z Z Z
∂uk
vφdxdt + σ(x, t, uk , Duk ) : Dvφdxdt + g(x, t, uk , Duk ).vφdxdt
Q ∂t Q Q
Z
= f.vφdxdt.
Q

After integrating, one gets

Z Z Z Z
uk (T )φ(T )vdx − uk (0)φ(0)vdx = f.vφdxdt − σ(x, t, uk , Duk ) : Dvφdxdt
Ω Ω Q Q
Z Z
− g(x, t, uk , Duk ).vφdxdt + uk vφ0 dxdt.
Q Q

We pass to the limit as k → ∞ in the previous equality

Z Z
zφ(T )vdx − u0 φ(0)vdx
Ω
ZΩ Z Z Z
= f.vφdxdt − χ.vφdxdt − ξ.vφdxdt + uvφ0 dxdt.
Q Q Q Q

Let φ(0) = φ(T ) = 0. Then

Z Z Z Z
− χ.vφdxdt − ξ.vφdxdt + f.vφdxdt = − uvφ0 dxdt
Q Q Q Q
Z
= φvu0 dxdt.
Q

Consequently, we obtain
Z Z Z Z
0
zφ(T )vdx − u0 φ(0)vdx = φvu dxdt + uvφ0 dxdt
Ω Ω Q Q
Z
T
= uφvdx 0
ZΩ Z
= u(T )φ(T )vdx − u(0)φ(0)vdx.
Ω Ω

If we take φ(T ) = 0 and φ(0) = 1, then we have u(0) = u0 ; if φ(T ) = 1 and

φ(0) = 0, then u(T ) = z.
The principal difficulty will be to identify χ with −div σ(x, t, u, Du) and ξ with
g(x, t, u, Du).

5. Identification of weak limits by means of Young measures

The Young measure is a device that comes to overcome the difficulty that
may arises when weak convergence does not behave as one desire with respect to
nonlinear functionals and operators. The following lemma describes limit points
of gradient sequences of approximating solutions.
 STRONGLY QUASILINEAR PARABOLIC SYSTEMS... 65

Lemma 5.1. If (Duk )k is bounded in Lp (0, T ; Lp (Ω)), then (Duk ) can generates
R satisfy kν(x,t) k = 1,1 and there
the Young measure ν(x,t) which is a subsequence of
(Duk ) weakly convergent to Mm×n λdν(x,t) (λ) in L (0, T ; L (Ω; Rm )).
1

Proof. To prove the first part of Lemma 5.1, it is sufficient to show that (Duk )
satisfies equation the (3.1). Since (Duk ) is bounded, it follows that there exists
c ≥ 0 such that
Z Z
p
c≥ |Duk | dxdt ≥ |Duk |p dxdt
Q {(x,t): |Duk (x,t)|≥L}

≥ Lp {(x, t) : |Duk (x, t)| ≥ L} .

Thus
c
sup {(x, t) : |Duk (x, t)| ≥ L} ≤ → 0, as L → ∞.
k∈N Lp
According to Lemma 3.2(iii), kν(x,t) k = 1.
For the remaining part, the reflexivity of Lp (0, T ; Lp (Ω)) implies the existence
of a subsequence (still denoted by (Duk )) weakly convergent in Lp (0, T ; Lp (Ω)),
thus weakly convergent in L1 (0, T ; L1 (Ω)). By Lemma 3.2(iii) and by taking ϕ
as the identity mapping id, it result that
Z
Duk * hν(x,t) , idi = λdν(x,t) (λ) weakly in L1 (0, T ; L1 (Ω)).
Mm×n


Lemma 5.2. For almost every (x, t) ∈ Q, ν(x,t) satisfies the following identifica-
tion
hν(x,t) , idi = Du(x, t).

Proof. Since uk * u in Lp (0, T ; W01,p (Ω; Rm )) and uk → u in Lp (0, T ; Lp (Ω)), we

have
Duk * Du in Lp (0, T ; Lp (Ω)).
Moreover, Duk * Du in L1 (0, T ; L1 (Ω)) (up to a subsequence). By virtue of
Lemma 5.1, we can infer that
Du(x, t) = hν(x,t) , idi for a.e. (x, t) ∈ Q.

The following lemma, namely div-curl inequality, is the key ingredient to pass
to the limit in the approximating equations and to prove that the weak limit u
of the Galerkin approximations uk is indeed a solution of (1.2)–(1.4).
Lemma 5.3. The Young measure ν(x,t) generated by the gradients Duk of the
Galerkin approximations uk has the following property:
Z Z


σ(x, t, u, λ) − σ(x, t, u, Du) : (λ − Du)dν(x,t) (λ)dxdt ≤ 0.
Q Mm×n
66 E. AZROUL, F. BALAADICH

Proof. Let us consider the sequence

Jk := σ(x, t, uk , Duk ) − σ(x, t, u, Du) : (Duk − Du)
= σ(x, t, uk , Duk ) : (Duk − Du) − σ(x, t, u, Du) : (Duk − Du)
=: Jk,1 + Jk,2 .
We have by the growth condition (H1) that
Z Z
p0 0
|d1 (x, t)|p + |u|p + |Du|p dxdt,


|σ(x, t, u, Du)| dxdt ≤ c
Q Q
0 0
and since u ∈ Lp (0, T ; W01,p (Ω)) we obtain σ ∈ Lp (0, T ; W −1,p (Ω)). By virtue of
Lemma 5.1, it follows that
Z


Jk,2 * σ(x, t, u, Du) : λdν(x,t) (λ) − Du ,
Mm×n
which gives by Lemma 5.2 that Jk,2 → 0 as k → ∞.
Since (uk ) is bounded, then uk * u in Lp (0, T ; W01,p (Ω; Rm )) and in measure on
Q. It follows from the equi-integrability of σ(x, t, uk , Duk ) and Lemma 3.4, that
Z Z
J := lim inf Jk dxdt = lim inf Jk,1 dxdt
k→∞ Q k→∞ Q
Z Z (5.1)
≥ σ(x, t, u, λ) : (λ − Du)dν(x,t) (λ)dxdt.
Q Mm×n

To get the result, it is sufficient to prove that J ≤ 0. On the one hand, we have
Z Z T
lim inf − σ(x, t, uk , Duk ) : Dudxdt = − hχ, uidt
k→∞ Q 0
Z T Z (5.2)
1 2 1 2
= ku(., T )kL2 − ku(., 0)kL2 − hf, uidt + ξ.udxdt,
2 2 0 Q

where we have used the following energy equality related to χ and ξ:

Z s Z s Z s
1 2 1
ku(., s)kL2 + hχ, uidt + hξ, uidt = hf, uidt + ku(., 0)k2L2
2 0 0 0 2
for all s ∈ [0, T ]. On the other hand, by the Galerkin equations
Z
σ(x, t, uk ,Duk ) : Duk dxdt
Q
Z T Z Z
∂uk
= hf, uk idt − uk dt − g(x, t, uk , Duk ).uk dxdt.
0 Q ∂t Q

We pass to the limit inf in the last equation and using the fact that uk (., 0) →
u0 (x) = u(x, 0) and uk (., T ) * u(., T ) in L2 (Ω; Rm ), we get
Z
lim inf σ(x, t, uk , Duk ) : Duk dxdt
k→∞ Q
Z T Z (5.3)
1 1
≤ hf, uidt − ku(., T )k2L2 + ku(., 0)k2L2 − ξ.udxdt.
0 2 2 Q
 STRONGLY QUASILINEAR PARABOLIC SYSTEMS... 67

Due to (5.2) and (5.3)

Z
J = lim inf σ(x, t, uk , Duk ) : (Duk − Du)dxdt ≤ 0.
k→∞ Q

According to Lemma 5.2

Z Z
σ(x, t, u, Du) : (λ − Du)dν(x,t) (λ)dxdt = 0.
Q Mm×n

This together with (5.1) imply the needed result. 

Lemma 5.4. Suppose that ν(x,t) satisfies the inequality of Lemma 5.3. Then for
a.e. (x, t) ∈ Q we have


σ(x, t, u, λ) − σ(x, t, u, Du) : (λ − Du) = 0 on supp ν(x,t) .
Proof. We have by Lemma 5.3, that
Z Z


σ(x, t, u, λ) − σ(x, t, u, Du) : (λ − Du)dν(x,t) (λ)dxdt ≤ 0.
Q Mm×n

The above integral is nonnegative, and this is according to the monotonicity

assumption of σ. Hence, it must vanish almost everywhere with respect to the
product measure dν(x,t) (λ) ⊗ dx ⊗ dt. Consequently, for almost all (x, t) ∈ Q


σ(x, t, u, λ) − σ(x, t, u, Du) : (λ − Du) = 0 on supp ν(x,t) .


6. Proof of the main result

In this section, we give the proof of Theorem 2.2 based on the two cases listed in.
We start with the case (2) where we have supposed that σ satisfies the condition
(c) or (d).
Note that, in these cases, we will prove that we may extract a subsequence
with the property
Duk → Du in measure on Q. (6.1)
Case (c): By strict monotonicity, it follows from Lemma 5.4 that supp ν(x,t) =
{Du(x, t)}, thus ν(x,t) = δDu(x,t) for a.e. (x, t) ∈ Q.

Case (d): Suppose that ν(x,t) is not a Dirac measure on a set (x, t) ∈ Q0 ⊂
Q of positive Lebesgue measure |Q0 | > 0. Since kν(x,t) k = 1 and Du(x, t) =
hν(x,t) , idi = λ, it follows from the strict p-quasimonotone that
Z Z


0< σ(x, t, u, λ) − σ(x, t, u, λ) : (λ − λ)dν(x,t) (λ)dxdt
Q Mm×n
Z Z
= σ(x, t, u, λ) : (λ − λ)dν(x,t) (λ)dxdt.
Q Mm×n

Hence
Z Z Z Z
σ(x, t, u, λ) : λdν(x,t) (λ)dxdt > σ(x, t, u, λ) : Du dν(x,t) (λ)dxdt.
Q Mm×n Q Mm×n
68 E. AZROUL, F. BALAADICH

From Lemma 5.3 and the above inequality, we get

Z Z Z Z
σ(x, t, u, λ) : Du dν(x,t) (λ)dxdt ≥ σ(x, t, u, λ) : λdν(x,t) (λ)dxdt
Q Mm×n Q Mm×n
Z Z
> σ(x, t, u, λ) : Du dν(x,t) (λ)dxdt
Q Mm×n

which is a contradiction. Hence ν(x,t) is a Dirac measure. Assume that ν(x,t) =

δh(x,t) . Then
Z Z
h(x, t) = λdδh(x,t) (λ) = λdν(x,t) (λ) = Du(x, t).
Mm×n Mm×n

Thus ν(x,t) = δDu(x,t) .

To complete the proof of this part, we argue as follows: we have ν(x,t) =
δDu(x,t) for a.e. (x, t) ∈ Q. Then by Lemma 3.3 Duk → Du in measure on
Q as k → ∞, and thus σ(x, t, uk , Duk ) → σ(x, t, u, Du) and g(x, t, uk , Duk ) →
g(x, t, u, Du) almost everywhere on Q (up to extraction of a further subsequence).
Since by (H1) and (H3)(i) the sequences σ(x, t, uk , Duk ) and g(x, t, uk , Duk ) are
bounded. It follows that σ(x, t, uk , Duk ) → σ(x, t, u, Du) and g(x, t, uk , Duk ) →
g(x, t, u, Du) in Lβ (Q), for all β ∈ [1, p0 ) by the Vitali convergence theorem. It
then follows that
− div σ(x, t, uk , Duk ) * χ = −div σ(x, t, u, Du) (6.2)
and
g(x, t, uk , Duk ) * ξ = g(x, t, u, Du). (6.3)
These properties are sufficient to pass to the limit in the Galerkin equations and
to conclude the proof of the part (2) of Theorem 2.2.
For the remaining part (i.e., the first part) of Theorem 2.2, we note that the
property (6.1) does not hold (in general), but we will obtain σ(x, t, uk , Duk ) *
0
σ(x, t, u, Du) and g(x, t, uk , Duk ) * g(x, t, u, Du) in Lp (Q). To do this, we
need the convergence in measure of the sequence uk . Since (uk )k is bounded
in Lp (0, T ; W01,p (Ω; Rm )), we have then uk * u in Lp (0, T ; W01,p (Ω; Rm )) and in
measure on Q as k → ∞.

Case (a): We prove that for a.e. (x, t) ∈ Q and every µ ∈ Mm×n the following
equation holds on supp ν(x,t)


σ(x, t, u, λ) : µ = σ(x, t, u, Du) : µ + ∇σ(x, t, u, Du) : (λ − Du), (6.4)
where ∇ denotes the derivative with respect to the third variable of σ. Due to
the monotonicity of σ, we have for all τ ∈ R


0 ≤ σ(x, t, u, λ) − σ(x, t, u, Du + τ µ) : (λ − Du − τ λ)
= σ(x, t, u, λ) : (λ − Du) − σ(x, t, u, λ) : τ µ − σ(x, t, u, Du + τ µ) : (λ − Du − τ µ)
= σ(x, t, u, Du) : (λ − Du) − σ(x, t, u, λ) : τ µ − σ(x, t, u, Du + τ µ) : (λ − Du − τ µ),
by Lemma 5.4. Hence
−σ(x, t, u, λ) : τ µ ≥ −σ(x, t, u, Du) : (λ−Du)+σ(x, t, u, Du+τ µ) : (λ−Du−τ µ).
 STRONGLY QUASILINEAR PARABOLIC SYSTEMS... 69

Using the fact that

σ(x, t, u, Du + τ µ) = σ(x, t, u, Du) + ∇σ(x, t, u, Du)τ µ + o(τ )
to deduce


−σ(x, t, u, λ) : τ µ ≥ τ ∇σ(x, t, u, Du)µ : (λ − Du) − σ(x, t, u, Du) : µ + o(τ ).

Since the sign of τ is arbitrary in R, the above inequality implies (6.4). On the
other hand, the equiintegrability of σ(x, t, uk , Duk ) implies that its weak L1 -limit
σ is given by
Z
σ= σ(x, t, u, λ)dν(x,t) (λ)
supp ν(x,t)
Z
(6.4)

= σ(x, t, u, Du) + ∇σ(x, t, u, Du) : (Du − λ) dν(x,t) (λ)
supp ν(x,t)

= σ(x, t, u, Du),
R
where we have used kν(x,t) k = 1 and supp ν(x,t) (Du − λ)dν(x,t) (λ) = 0.
Evidently,
0
σ(x, t, uk , Duk ) * σ(x, t, u, Du) in Lp (Q).
Case (b): We start by proving that for almost all (x, t) ∈ Q, supp ν(x,t) ⊂
K(x,t) , where
K(x,t) := λ ∈ Mm×n : W (x, t, u, λ) = W (x, t, u, Du) + σ(x, t, u, Du) : (λ − Du) .


If λ ∈ supp ν(x,t) , then by Lemma 5.4

(1 − τ ) : σ(x, t, u, λ) − σ(x, t, u, Du) : (λ − Du) = 0 ∀τ ∈ [0, 1]. (6.5)
The monotonicity of σ together with (6.5) imply


0 ≤ (1 − τ ) : σ(x, t, u, Du + τ (λ − Du)) − σ(x, t, u, λ) : (Du − λ)

(6.6)
= (1 − τ ) : σ(x, t, u, Du + τ (λ − Du)) − σ(x, t, u, Du) : (Du − λ).
Again by the monotonicity of σ and τ ∈ [0, 1], it follows that the right hand side
of (6.6) is nonpositive, because


σ(x, t, u, Du + τ (λ − Du)) − σ(x, t, u, Du) : τ (λ − Du) ≥ 0,
which implies for all τ ∈ [0, 1]


σ(x, t, u, Du + τ (λ − Du)) − σ(x, t, u, Du) : (1 − τ )(λ − Du) ≥ 0.
Thus, for all τ ∈ [0, 1]


σ(x, t, u, Du + τ (λ − Du)) − σ(x, t, u, Du) : (λ − Du) = 0,
whenever λ ∈ supp ν(x,t) . From the hypothesis of the potential W we get
Z 1


W (x, t, u, λ) = W (x, t, u, Du) + σ x, t, u, Du + τ (λ − Du) : (λ − Du)dτ
0
= W (x, t, u, Du) + σ(x, t, u, Du) : (λ − Du).
70 E. AZROUL, F. BALAADICH

We conclude that λ ∈ K(x,t) , i.e., supp ν(x,t) ⊂ K(x,t) . Due to the convexity of W ,
we have for all λ ∈ Mm×n
W (x, t, u, λ) ≥ W (x, t, u, Du) + σ(x, t, u, Du) : (λ − Du).
For all λ ∈ K(x,t) , put
F (λ) = W (x, t, u, λ) and G(λ) = W (x, t, u, Du) + σ(x, t, u, Du) : (λ − Du).
Since λ 7→ F (λ) is continuous and differentiable, it follows for µ ∈ Mm×n and
τ ∈R
F (λ + τ µ) − F (λ) G(λ + τ µ) − G(λ)
≥ if τ > 0,
τ τ
F (λ + τ µ) − F (λ) G(λ + τ µ) − G(λ)
≤ if τ < 0.
τ τ
Consequently, DF = DG, i.e.,
σ(x, t, u, λ) = σ(x, t, u, Du) ∀ λ ∈ K(x,t) ⊃ supp ν(x,t) .
Hence
Z Z
σ= σ(x, t, u, λ)dν(x,t) (λ) = σ(x, t, u, λ)dν(x,t) (λ)
Mm×n supp ν(x,t)
Z
(6.7)
= σ(x, t, u, Du)dν(x,t) (λ)
supp ν(x,t)

= σ(x, t, u, Du).
This shows that σ(x, t, uk , Duk ) * σ(x, t, u, Du) in L1 (Q), and we will show the
strong convergence. Consider the Carathéodory function
h(x, t, s, λ) = σ(x, t, s, λ) − σ(x, t) , s ∈ Rm , λ ∈ Mm×n .
0
We have σ(x, t, uk , Duk ) is weakly convergent in Lp (Q), hence equi-integrable.
This implies the equi-integrability of hk (x, t) := h(x, t, uk , Duk ) and
hk * h in L1 (Q),
where
Z
h(x, t) = h(x, t, s, λ)dδu(x,t) (s) ⊗ dν(x,t) (λ)
Rm ×Mm×n
Z
= σ(x, t, u, λ) − σ(x, t) dν(x,t) (λ)
Mm×n
Z
= σ(x, t, u, λ) − σ(x, t) dν(x,t) (λ) = 0,
supp ν(x,t)

by (6.7). Since hk ≥ 0, it follows that

hk → 0 in L1 (Q).
0
Using the fact that hk is bounded in Lp (Q) together with the Vitali convergence
0
theorem, we conclude that σ(x, t, uk , Duk ) * σ(x, t, u, Du) in Lp (Q).
From cases (a) and (b), we have
0
σ(x, t, uk , Duk ) * σ(x, t, u, Du) in Lp (Q).
 STRONGLY QUASILINEAR PARABOLIC SYSTEMS... 71
0
It remains then to prove that g(x, t, uk , Duk ) * g(x, tu, Du) in Lp (Q). If g does
not depend on the third variable, then by the convergence in measure of uk to
u and the continuity of g, we get the needed result. On the other hand, if g is
linear in A ∈ Mm×n , then
Z
g(x, t, uk , Duk ) * g(x, t, u, λ)dν(x,t) (λ)
Mm×n
Z
= g(x, t, u, .) ◦ λdν(x,t) (λ)
Mm×n
= g(x, t, u, .) ◦ Du = g(x, t, u, Du),
R
where we have used Du(x, t) = Mm×n λdν(x,t) (λ).
In conclusion, we can now pass to the limit in the Galerkin equations. Note
that the energy equality
Z T Z T Z T
1 2 1
ku(., T )kL2 (Ω) + hχ, uidt + hξ, uidt = hf, uidt + ku(., 0)k2L2 (Ω)
2 0 0 0 2
holds true with χ replaced by −div σ(x, t, u, Du) and ξ by g(x, t, u, Du).

References
1. L. Aharouch, E. Azroul and M. Rhoudaf, Existence result for variational degenerated par-
abolic problems via pseudo-monotonicity, in: Proceedings of the 2005 Oujda International
Conference on Nonlinear Analysis, pp. 9–20, Electron. J. Differ. Equ. Conf. 14, Southwest
Texas State Univ. San Marcos, TX, 2006.
2. J.M. Ball, A version of the fundamental theorem for Young measures, in: PDEs and con-
tinuum models of phase transitions (Nice, 1988), pp. 207–215, Lecture Notes in Phys. 344,
Springer, Berlin, 1989.
3. H. Brézis and F.E. Browder, Strongly nonlinear parabolic initial-boundary value problems,
Proc. Nat. Acad. Sci. U.S.A. 76 (1979), no. 1, 38–40.
4. J. Chabrowski and K.W. Zhang, Quasi-monotonicity and perturbated systems with critical
growth, Indiana Univ. Math. J. 41 (1992), no. 2, 483–504.
5. G. Dolzmann, N. Hungerühler and S. Muller, Nonlinear elliptic systems with measure-valued
right hand side, Math. Z. 226 (1997) 545–574.
6. P. Dreyfuss and N. Hungerbühler, Navier-Stokes systems with quasimonotone viscosity ten-
sor, Int. J. Differ. Equ. Appl. 9 (2004), no. 1, 59–79.
7. N. Hungerbühler, Young measures and nonlinear PDEs, Birmingham, 1999.
8. N. Hungerbühler, Quasi-linear parabolic systems in divergence form with weak monotonic-
ity, Duke Math. J. 107 (2001), no. 3, 497–519.
9. E. Kamke, Das Lebesgue-Stieltjes-Integral, Teubner, Leipzig, 1960.
10. R. Landes, On Galerkin’s method in the existence theory of quasilinear elliptic equations,
J. Funct. Anal. 39 (1980) 123–148.
11. R. Landes, On weak solutions of quasilinear parabolic equations, Nonlinear Anal. 9 (1985)
887–904.
12. R. Landes and V. Mustonen, On parabolic initial-boundary value problems with critical
growth for the gradient, Ann. Henri Poincaré, 11 (1994) 135–158.
13. L.C. Young, Generalized curves and the existence of an attained absolute minimum in the
calculus of variations, Comptes Rendus de la Société des Sciences et des Lettres de Varsovie,
30 (1937) 212–234.
72 E. AZROUL, F. BALAADICH

1
Department of Mathematics, Faculty of Sciences Dhar El Mehraz, B.P. 1796
Atlas, Fez-Morocco.
E-mail address: [email protected]; [email protected]