See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/255627259

Existence and regularity results for some nonlinear parabolic equations

Article · January 1999

CITATIONS READS

49 417

3 authors, including:

Lucio Boccardo Luigi Orsina

Sapienza University of Rome Sapienza University of Rome
274 PUBLICATIONS 10,729 CITATIONS 122 PUBLICATIONS 3,947 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Lucio Boccardo on 09 June 2016.

The user has requested enhancement of the downloaded file.

 EXISTENCE AND REGULARITY RESULTS
FOR SOME NONLINEAR PARABOLIC EQUATIONS

Lucio BOCCARDO1 Andrea DALL’AGLIO2

Thierry GALLOUËT3 Luigi ORSINA1
Abstract
We prove summability results for the solutions of nonlinear parabolic
problems of the form

u
+ A(u) = f in Ω × (0, T ) ,

with homogeneous Cauchy-Dirichlet boundary conditions, where A is a pseu-

domonotone, coercive, uniformly elliptic operator acting from the space Lp (0, T ; W01,p (Ω))



to its dual Lp (0, T ; W −1,p (Ω)), and the datum f belongs to Lq (0, T ; Lr (Ω)).
As a consequence of these estimates we also obtain an existence theorem in



the case where f does not belong to Lp (0, T ; W −1,p (Ω)).
We also give summability results for data in divergence form.

1 Introduction and statement of the results

Let Ω be a bounded domain in RN , N ≥ 1. For T > 0, let us denote by Q
the cylinder Ω × (0, T ), and by Γ the lateral surface ∂Ω × (0, T ). We will
consider the following nonlinear parabolic Cauchy-Dirichlet problem:


 u
− div (a(x, t, u, ∇u)) = f (x, t) in Q,
u(x, 0) = 0 in Ω, (1.1)


u(x, t) = 0 on Γ.
1
Dipartimento di Matematica, Università di Roma I, Piazzale A. Moro 2, 00185, Roma,
Italia
2
Dipartimento di Matematica, Università di Firenze, Viale Morgagni 67a, 50134,
Firenze, Italia
3
ENS-Lyon, 69364 Lyon Cedex 7, France

1
where −div (a(x, t, u, ∇u)) is a pseudomonotone, coercive, uniformly elliptic



operator acting from Lp (0, T ; W01,p (Ω)) to its dual Lp (0, T ; W −1,p (Ω)) (p is
a real number greater than 1), and f is a measurable function on Q. If



f belongs to Lp (0, T ; W −1,p (Ω)), then it is well known (see [8]) that there
exists at least a solution u of (1.1), with u in Lp (0, T ; W01,p (Ω)).
We will be concerned with some existence and regularity results for the
solutions of (1.1), depending on the summability of the datum f .
We begin recalling some results in the case p = 2 (which includes, for
example, the case of operators which are linear with respect to the gradient):
if f belongs to Lr (0, T ; Lq (Ω)), and if r and q are large enough, that is, if
N 2
+ < 2, r ≥ 1, q ≥ 1,
q r
then Aronson and Serrin (see [1]) proved that any solution u of (1.1) belongs
to L∞ (Q).
On the other hand, if

a(x, t, u, ∇u) = A(x, t) ∇u ,

and if r and q satisfy the opposite inequality and an additional one, more
precisely if  
N 2 N N
2< + ≤ min + 2, + 2 , (1.2)
q r r 2
then in [7], Chapter III, Theorem 9.1, it is proved that any weak solution (if
it exists) belongs to Ls (Q), with s given by
(N + 2)qr
s= , (1.3)
N r − 2q(r − 1)
and this then implies that such a solution belongs to L2 (0, T ; H01 (Ω)).
Observe that if r ≥ 2, then (1.2) becomes
N 2 N
2< + ≤ + 2,
q r r
and in this case the function f belongs to the space L2 (0, T ; H −1 (Ω)), so that
there exists at least a solution of (1.1) (which is indeed unique since a does

2
not depend on u). Conversely, if 1 ≤ r < 2, the hypotheses on r and q can
be rewritten as
N 2 N
2< + ≤ + 2,
q r 2
and in this case the function f is not in general in L2 (0, T ; H −1 (Ω)). However,
one can use the linearity of the operator with respect to the gradient and
the a priori estimates in order to prove the existence of a solution of (1.1) by
means of duality techniques.
The a priori estimates in L2 (0, T ; H01 (Ω)) can be obtained, always in the
linear case, by means of different techniques, such as norm estimates using
the heat kernel, duality properties and interpolation between Lebesgue spaces
([5]).
If a is of the form

a(x, t, u, ∇u) = A(x, t, u) ∇u ,

where A is a Carathéodory matrix-valued function, bounded and uniformly

elliptic, the duality techniques cannot be applied, but the same a priori esti-
mates (with the same bounds on r and q, and the same value of s) continue
to hold. These estimates, together with the linearity of the operator with
respect to the gradient, were used by the authors in order to prove the exis-
tence of at least a solution of (1.1) in the case of data not in L2 (0, T ; H −1 (Ω))
(this case was presented by the first author in [2], where also the results of
the present paper were announced).
In this paper we are going to deal with the general case of a nonlin-
ear pseudomonotone operator. Before stating our results, let us make our
assumptions more precise.
Let p be a real number, p > 1, and let p
be its Hölder conjugate exponent
(i.e., 1/p + 1/p
= 1).
We recall that Ω is a bounded domain in RN , N ≥ 1, Q denotes the
cylinder Ω × (0, T ), and Γ is the lateral surface ∂Ω × (0, T ) of Q, with T > 0.
Let a : Q × R × RN → RN be a Carathéodory function (i.e., a(·, ·, σ, ξ)
is measurable in Q for every (σ, ξ) in R × RN , and a(x, t, ·, ·) is continuous
on R × RN for almost every (x, t) in Q), such that:

a(x, t, σ, ξ) · ξ ≥ α |ξ|p , (1.4)

3
 |a(x, t, σ, ξ)| ≤ β [k(x, t) + |σ|p−1 + |ξ|p−1 ] , (1.5)
[a(x, t, σ, ξ) − a(x, t, σ, η)] · (ξ − η) > 0 , (1.6)
for almost every (x, t) in Q, for every σ in R, for every ξ, η in RN , ξ = η,
where α and β are positive real numbers, and k is a nonnegative function in


Lp (Q).
Let us define the differential operator

A(u) = −div (a(x, t, u, ∇u)) .

Under our assumptions on a, the operator A turns out to be a uniformly

elliptic, coercive and pseudomonotone operator acting from Lp (0, T ; W01,p (Ω))



to its dual Lp (0, T ; W −1,p (Ω)). Moreover (see [8]), for every datum g in



Lp (0, T ; W −1,p (Ω)) there exists at least one solution u in Lp (0, T ; W01,p (Ω))
of the following Cauchy-Dirichlet problem


 u
+ A(u) = g in Q,
u(x, 0) = 0 in Ω, (1.7)


u(x, t) = 0 on Γ.
By a solution u of (1.7) we mean a function in Lp (0, T ; W01,p (Ω)) which
satisfies the equation (1.7) in the sense of distributions. The solution u
belongs to C 0 ([0, T ]; L2 (Ω)) (see for instance [8]), so that the initial datum
has a meaning.
Furthermore, if ψ : R → R is a Lipschitz continuous function such that
ψ(0) = 0, we can choose ϕ = ψ(u) as test function in order to obtain (recall
that u(x, 0) = 0)
 
Ψ(u(T )) dx + a(x, t, u, ∇u) · ∇(ψ(u)) = g, ψ(u) , (1.8)
Ω Q

where  s
Ψ(s) = ψ(ρ) dρ ,
0



and ·, · denotes the duality pairing between the spaces Lp (0, T ; W −1,p (Ω))
and Lp (0, T ; W01,p (Ω)).
Our first result concerns the regularity of solutions of (1.7) in the case of
data which belong to the dual space (see Remark 1.2 below).

4
Theorem 1.1 Let r and q be real numbers such that
N p N
p< + ≤ + p, r ≥ p
, q > 1, (1.9)
q r r
Let f be a function in Lr (0, T ; Lq (Ω)). Then any solution u of (1.7) in the
space Lp (0, T ; W01,p (Ω)) belongs to Ls (Q) where
qr(N + p) + N (p − 2)(qr − q + r)
s= . (1.10)
N r − pq(r − 1)
Moreover, if p > N , one can choose q = 1 in (1.9).

Remark 1.2 Let us make some comments on the bounds on r and q, and
on the value of s. If p < N , it is easy to see that (1.9) implies
Np
q≥ = (p∗ )
,
Np − N + p
where p∗ = N p/(N − p) is the Sobolev exponent of p. This fact yields that



the datum f always belongs to Lp (0, T ; W −1,p (Ω)), so that there exists a
solution for (1.7). Moreover, since p < N , we have that (p∗ )
> 1, so that
the condition q > 1 of (1.9) is always satisfied.
If p = N , then (1.9) becomes
1 1 N
+ > 1, q > 1, r≥ .
q r N −1
If p > N , the condition
N p N
+ ≤ +p
q r r
is always satisfied for r and q greater or equal to than 1, so that the conditions
on r and q become
N p
+ > p, r ≥ p
, q ≥ 1.
q r
If q = 1, then r must satisfy
p
p
≤ r < .
p−N

5
 One can check that the value of s is greater than p(N + 2)/N , which is
the Gagliardo-Nirenberg embedding exponent for functions that belong to
Lp (0, T ; W01,p (Ω)) and to L∞ (0, T ; L2 (Ω)) (see also (2.3), below). Observe
that if p = 2, we obtain the value of s given by (1.3).

Our next result is not only a regularity theorem, but also an existence
result.

Theorem 1.3 Let r and q be real numbers such that

N p N p N p + 2p − N
p< + ≤ 1− + , 1 ≤ r < p
, q ≥ 1 , (1.11)
q r r 2 2
and let f be a function in Lr (0, T ; Lq (Ω)). Then there exists at least one
solution u in Lp (0, T ; W01,p (Ω)) of (1.7); moreover, u belongs to Ls (Q) where
s is as in (1.10).

Remark 1.4 In this case, being r < p
, the datum f does not belong to the



dual space Lp (0, T ; W −1,p (Ω)). If r = 1, then q has to be greater than or
equal to 2.
As in the case of Theorem 1.1, if p < N then the first two conditions of
(1.11) imply q > 1. If p = N , then q = 1 if and only if r = N/(N − 1) = p
,
a value that r cannot attain. Thus, q can be equal to 1 if and only if p > N .
If q = 1, then r must satisfy
N p + 2p − 2N
≤ r < p
.
N p + 2p − 3N
Again, if 1 < p < N the possible values for s are greater than p(N +2)/N .

We will also give a regularity result for solutions of (1.7) in the case of
data in divergence form (see Proposition 2.4, below).
A result concerning local estimates for solutions of nonlinear parabolic
problems was obtained by Porzio in [10], using different techniques.
The following is a picture which summarizes, in the case p < N , the possi-
ble values of r and q given by the bounds in Theorem 1.1 and Theorem 1.3. If
(r, q) lies in region A, then the solution u belongs to Ls (Q)∩Lp (0, T ; W01,p (Ω)),
with s given by (1.10) (see Theorem 1.1); in region B the datum f is not in the

6
dual space, and we have existence of a solution in Ls (Q) ∩ Lp (0, T ; W01,p (Ω)),
with s as in (1.10) (see Theorem 1.3).

The main tool of the proof of both Theorem 1.1 and Theorem 1.3 will be
an a priori estimate on the solutions of (1.7) with data in the dual space. In
the case of Theorem 1.1 this estimate will directly give the result. Under the
hypotheses of the existence Theorem 1.3, however, we will need to proceed
by approximation: more precisely, we will consider data fn (belonging to the
dual space) which converge to the datum f in Lr (0, T ; Lq (Ω)), and take into
account a sequence of solutions un of the corresponding problems. We will
have a uniform estimate on un in Ls (Q) and then in Lp (0, T ; W01,p (Ω)). We
would like to pass to the limit in order to find a solution for our problem
but, since the equation is nonlinear, the a priori estimates are not enough in
order to let n tend to infinity: we need a strong convergence result for the
sequence {∇un }. In order to obtain this result, we will use a theorem proved
by the authors in [3], which will allow us to conclude.
The plan of the paper is as follows: in the next section we will prove the
a priori estimates on the solutions of (1.7) which form the core of the proof
of Theorem 1.1 and Theorem 1.3; we will also state and prove an a priori
estimate for the solutions of (1.7) in the case of data of the form −div (F ).
The third section will be devoted to the proof of the theorems.

7
2 A priori estimates
Before stating and proving our results, let us recall the Gagliardo-Nirenberg
inequality (see [9], Lecture II).

Lemma 2.1 Let v be a function in W01,p (Ω) ∩ Lρ (Ω), with p ≥ 1, ρ ≥ 1.

Then there exists a positive constant C, depending on N , p, ρ, and σ such
that
vLσ (Ω) ≤ C ∇vθ p N
v1−θ
ρ
, (2.1)
L (Ω;R ) L (Ω)

for every θ and σ satisfying

1 1 1 1−θ
0 ≤ θ ≤ 1, 1 ≤ σ < +∞ , =θ − + . (2.2)
σ p N ρ

If p > N , one can choose σ = +∞ (that is, 1/σ = 0).

An immediate consequence of the previous lemma is the following em-

bedding result:  
ρp
|v| ≤ C v ∞
σ N
ρ
|∇v|p , (2.3)
Q L (0,T ;L (Ω)) Q

which holds for every function v in Lp (0, T ; W01,p (Ω)) ∩ L∞ (0, T ; Lρ (Ω)), with
p ≥ 1, ρ ≥ 1 and σ = p(N + ρ)/N (see, for instance, [6], Proposition 3.1).
We now state the a priori estimate we need.

Proposition 2.2 Let r and q be real numbers such that

 
N p N N p N p + 2p − N
p< + ≤ min + p, 1− + , r ≥ 1 , (2.4)
q r r r 2 2
and
q>1 if p ≤ N , q≥1 if p > N . (2.5)



hold. Let f be a function in Lr (0, T ; Lq (Ω)) ∩ Lp (0, T ; W −1,p ). Let u be a
solution in Lp (0, T ; W01,p (Ω)) of (1.7) with datum f , and let s be given by
(1.10). Then the norms of u in Ls (Q) and in Lp (0, T ; W01,p (Ω)) are bounded
by a constant which depends on α, N , p, q, r, Q and on the norm of f in
Lr (0, T ; Lq (Ω)).

8
Remark 2.3 We observe that (2.4) can be written as

N p N p + 2p − N



N p 1− +  if 1 ≤ r < p
,
p< + ≤ r 2 2
q r 
 N
 +p if r ≥ p
,
r
and therefore Proposition 2.2 gives a priori estimates under the assumptions
of both Theorem 1.1 and Theorem 1.3.

Proof of Proposition 2.2. In the following, we will denote by c any con-

stant depending only on α, N , p, q, r and f Lr (0,T ;Lq (Ω)) . The value of c
may vary from line to line.
Let n be a positive integer, and suppose that r > 1 and q > 1.
For γ ≥ 0 (which will be chosen below), and τ ∈ (0, T ), we can take
ϕ = |Tn (u)|γp Tn (u)χ(0,τ ) (t) as test function. This is possible since Tn (u) ∈
L∞ (Q) ∩ Lp (0, T ; W01,p (Ω)), so that ϕ belongs to the same space. Using (1.8),
and setting  s
Ψn (s) = |Tn (t)|γp Tn (t) dt ,
0
we obtain
  τ
Ψn (u(τ )) dx + (γp + 1) a(x, t, u, ∇u) · ∇Tn (u) |Tn (u)|γp
Ω 0 Ω (2.6)
 τ
≤ |f | |Tn (u)| γp+1
.
0 Ω

Observing that
|Tn (s)|γp+2
Ψn (s) ≥ ,
γp + 2
and recalling the ellipticity assumption (1.4) on a, we obtain from (2.6)
  τ
|Tn (u)(τ )| γp+2
dx ≤ (γp + 2) |f | |Tn (u)|γp+1 ,
Ω 0 Ω

and  τ
(γ + 1)p  τ
|∇(|Tn (u)| γ+1
)| ≤
p
|f | |Tn (u)|γp+1 .
0 Ω α (γp + 1) 0 Ω

9
Taking the supremum of both terms for τ in (0, T ) and using Hölder’s in-
equality twice, we obtain
 
Tn (u)γp+2
∞
+ |∇(|Tn (u)|γ+1 )|p ≤ c |f | |Tn (u)|γp+1
L (0,T ;Lγp+2 (Ω)) Q Q
 T
≤c f (t)Lq (Ω) Tn (u)(t)γp+1
(γp+1) q

dt
0 L (Ω)
 T  1
(2.7)
r
(γp+1) r

≤ c f Lr (0,T ;Lq (Ω)) Tn (u)(t)
dt
0 L(γp+1) q (Ω)
 T  1

r
(γp+1) r

≤c Tn (u)(t) (γp+1) q

dt ,
0 L (Ω)

with c depending also on γ. Observe that both r
and q
are real numbers
since r > 1 and q > 1. Setting
w = |Tn (u)|γ+1 ,
formula (2.7) can be rewritten as
  1

γp+2  T (γp+1) r
r

w γ+1
+ ∇wp p ≤c  w(t) γ+1
dt . (2.8)
γp+2 (γp+1) q

L∞ (0,T ;L γ+1 (Ω)) L (Q;RN ) 0 L γ+1 (Ω)

Let us define θ such that

γ+1 1 1 (1 − θ)(γ + 1)
=θ − + . (2.9)
(γp + 1) q
p N γp + 2
We will check below that, under our assumptions on r and q, and with our
choice of γ (see (2.12) below), θ belongs to [0, 1]. Thus, by the Gagliardo-
Nirenberg inequality (2.1), applied with
(γp + 1)q
γp + 2
σ= , ρ= ,
γ+1 γ+1
we have, from (2.8),
γp+2

w γ+1 γp+2 + ∇wp p

L∞ (0,T ;L γ+1 (Ω)) L (Q;RN )
 T  1
(2.10)
γp+1
(1−θ) γp+1
θr
r
≤ c w γ+1
L∞ (0,T ;Lγp+2 (Ω))
∇w(t) γ+1
L (Ω;RN )
p dt .
0

10
If θ < 1, applying the Young inequality with exponents
γp + 2 γp + 2
and ,
(γp + 1)(1 − θ) 1 + θ(γp + 1)
we obtain
γp+2

w γ+1 γp+2 + ∇wp p

L∞ (0,T ;L γ+1 (Ω)) L (Q;RN )
 T 
γp+2 (2.11)
1 γp+2 γp+1
θr
r [1+θ(γp+1)]
≤ w γ+1
γp+2 +c ∇w(t)Lp (Ω;RN ) dt
γ+1
.
2 L∞ (0,T ;L γ+1 (Ω)) 0

Now we choose γ satisfying

γp + 1

θr = p , (2.12)
γ+1
and we will check below that, under our assumptions on r and q, γ is non-
negative. From (2.11) we then obtain
 
γp+2
γp+2 p r
[1+θ(γp+1)]
w γ+1
γp+2 + ∇wp p ≤ c ∇wLp (Q;RN ) . (2.13)
L∞ (0,T ;L γ+1 (Ω)) L (Q;RN )

Since, from (2.12),

θr
(γp + 1) = γp + p ,
we have
γp + 2 γp + 2
=
<1
r
[1 + θ(γp + 1)] r + γp + p
(the latter inequality is equivalent to r
> 2 − p which is true since 2 − p < 1),
and so
γp+2

w γ+1 γp+2 + ∇wp p ≤ c ∇wδ pp , (2.14)

L∞ (0,T ;L γ+1 (Ω)) L (Q;RN ) L (Q;RN )

with δ < 1. If θ = 1, choosing γ as in (2.12), (2.10) becomes (2.14) with

δ = 1/r
< 1. Thus from (2.14) immediately follows
γp+2

w γ+1 γp+2 + ∇wp p ≤ c,

L∞ (0,T ;L γ+1 (Ω)) L (Q;RN )

11
and therefore, by inequality (2.3),

wLσ (Q) ≤ c ,

where
γp+2
N+ γ+1
σ=p .
N
Recalling the definition of w, we thus have proved that

Tn (u)Ls (Q) ≤ c , (2.15)

where c depends on α, N , p, q, r, γ, f Lr (0,T ;Lq (Ω)) but not on n, and

N (γ + 1) + γp + 2 N +p
s=p = (γp + 2) + p − 2. (2.16)
N N
From (2.9) and (2.12) we obtain
N q[(r − 1)(p − 2) + r]
γp + 2 = ,
N r − pq(r − 1)
which implies, by (2.16),
qr(N + p) + N (p − 2)(qr − q + r)
s= ,
N r − pq(r − 1)
which is the value given by (1.10). Passing to the limit as n tends to infinity
in (2.15), we thus obtain an estimate for u in Ls (Q).
We now have to check that γ ≥ 0 and that θ, defined in (2.9), belongs to
[0, 1]. After easy calculations, we obtain that γ ≥ 0 if and only if
N p N p N p − N + 2p
p< + ≤ 1− + ,
q r r 2 2
while the condition θ ≤ 1 holds is satisfied if and only if
N p N
+ ≤ + p.
q r r
The condition θ ≥ 0 is automatically satisfied if γ ≥ 0. Thus, we obtain for
r and q (both strictly greater than 1) the conditions given in the statement.

12
 We now deal with the case r = 1, which corresponds to q ≥ 2. In this
case, we choose ϕ = |Tn (u)|q−2 Tn (u) χ(0,τ ) as test function in (1.7). By means
of the same technique used above, we arrive after straightforward passages
to q−2+p pq
|Tn (u)| p ∈ L∞ (0, T ; L q−2+p (Ω)) ∩ Lp (0, T ; W01,p (Ω)) ,
with a bound on the norms of Tn (u) in these spaces which is uniform with
respect to n. Using inequality (2.3), this implies (again after a passage to
the limit as n tends to infinity) that u is bounded in Ls (Q), with

(N + p)q + N (p − 2)
s= ,
N
which is the value given by (1.10) in the case r = 1.
Finally, we have to treat the case q = 1, which corresponds to the case
p > N . As before, we choose as test function ϕ = |Tn (u)|γp Tn (u) χ(0,τ ) (t),
and we obtain (setting w = |Tn (u)|γ+1 ),
 T  1

γp+2 (γp+1) r
r
w γ+1
γp+2 + ∇wp p ≤c w(t) γ+1 dt .
L∞ (0,T ;L γ+1 (Ω)) L (Q;RN ) 0 L∞ (Ω)

Observe that in this case q
= +∞. Using the Gagliardo-Nirenberg inequality

(2.1) on the right hand side, with ρ = (γp + 2)/(γ + 1) and σ = +∞, we get
(2.10), with θ given by
1
ρ
θ= .
1
N
− p1 + 1
ρ

Going on as in the first part of the proof, one obtains the bound on u in
Ls (Q), with s given by (1.10) with q = 1. The only condition we have to
check is that γ, defined in (2.12), is nonnegative, since this will then imply
that θ ∈ [0, 1]. After straightforward calculation, one can check that γ ≥ 0
if and only if
N p + 2p − 2N p
≤r< ,
N p + 2p − 3N p−N
and these are exactly the bounds on r given by (2.4) for q = 1 (see Remark
2.3).

13
 It remains to prove the estimate in Lp (0, T ; W01,p (Ω)), and in order to do
this, we take T1 (u) as test function in (1.7) (observe that this corresponds to
γ = 0). Throwing away positive terms we obtain, after using (1.4),

α |∇u|p ≤ f L1 (Q) ≤ c f Lr (0,T ;Lq (Ω)) ,
{|u|≤1}

since both r and q are not smaller than 1. On the other hand, if γ is as in
(2.12) (if r > 1) or is such that γp + 2 = q (if r = 1), we have
 
|∇u|p ≤ |∇u|p |u|γp ≤ c ,
{|u|>1} Q

by the first part of the proof. Thus, we obtain

  
|∇u| =
p
|∇u| + p
|∇u|p ≤ c ,
Q {|u|≤1} {|u|>1}

and this gives us the Lp (0, T ; W01,p (Ω)) estimate.

The following result is the analogue of Proposition 2.2 in the case of data
in divergence form.

Proposition 2.4 Let r1 and q1 be real numbers such that

N p N
p−1< + ≤ + p − 1, r1 ≥ p
, q 1 > p
. (2.17)
q1 r 1 r1
Let F : Ω → RN be such that |F | belongs to Lr1 (0, T ; Lq1 (Ω)). Let u be a
solution in Lp (0, T ; W01,p (Ω)) of


 u
+ A(u) = −div (F ) in Q,
u(x, 0) = 0 in Ω, (2.18)


u(x, t) = 0 on Γ.

and let
(N + 2)(p − 1)q1 r1 + N (r1 − q1 )(p − 2)
s1 = . (2.19)
r1 (N + q1 ) − pq1 (r1 − 1)
Then the norm of u in Ls1 (Q) is bounded by a constant which depends on
α, p, N , q1 , r1 , Q and on the norm of |F | in Lr1 (0, T ; Lq1 (Ω)).
Moreover, if p = N we can choose q1 = p
in (2.17).

14
Remark 2.5 In the case of minimal regularity of the datum, that is, in the
case where r1 = q1 = p
, the value given by (2.19) is s1 = p(N + 2)/N , which
is the exponent of the embedding (2.3) for functions in Lp (0, T ; W01,p (Ω)) ∩
L∞ (0, T ; L2 (Ω)).

Proof of Proposition 2.4. The proof follows the same lines as Proposition
2.2.
As before, for γ ≥ 0, and τ ∈ (0, T ), we take ϕ = |Tn (u)|γp Tn (u)χ(0,τ ) (t)
as test function in (2.18). As in the proof of Proposition 2.2, we obtain (after
straightforward passages),
 
Tn (u)γp+2
∞
+ |∇Tn (u)|p |Tn (u)|γp ≤ c |F | |∇Tn (u)| |Tn (u)|γp .
L (0,T ;Lγp+2 (Ω)) Q Q

Using the Young inequality with exponents p
and p in the right hand side,
we have
  


|F | |∇Tn (u)| |Tn (u)|γp ≤ ε |∇Tn (u)|p |Tn (u)|γp + cε |F |p |Tn (u)|γp ,
Q Q Q

so that, choosing ε small enough,

 


Tn (u)γp+2 + |∇Tn (u)| |Tn (u)|
p γp
≤c |F |p |Tn (u)|γp .
L∞ (0,T ;Lγp+2 (Ω)) Q Q

Starting from here, we can repeat exactly the same steps as in the proof of
Proposition 2.2, obtaining the result given in the statement (again, one has
to treat separately the “borderline” cases in which either r1 = p
or q1 = p
).

Remark 2.6 What kind of estimates can be expected if the datum of the
problem is of the form f − div (F ), where f belongs to Lr (0, T ; Lq (Ω)), and
|F | belongs to Lr1 (0, T ; Lq1 (Ω)), with (r, q) and (r1 , q1 ) as in Proposition 2.2
and Proposition 2.4, respectively? What is the regularity of the solutions
with respect to Lebesgue spaces on Q?
If the problem is linear (for example, if a(x, t, s, ξ) = ξ, so that we get
the Laplacian operator), then the solution of the problem with datum f −
div (F ) is the sum of the solutions of the problems with data f and −div F ,

15
respectively, so that, in general, it has the regularity of the less regular of
the two solutions. This fact is also true in the general case.
To see this, let us define s(r, q) the value given by (1.10), and s1 (r1 , q1 ) the
value given by (2.19); denote by E and E1 the sets of pairs (r, q) and (r1 , q1 )
which satisfy the hypotheses of Proposition 2.2 and Proposition 2.4 respec-
tively. If (r, q) ∈ E and (r1 , q1 ) ∈ E1 are such that s(r, q) = s1 (r1 , q1 ) = s,
then clearly we can “combine” the proofs of Proposition 2.2 and Proposition
2.4, choosing the same test function and thus obtaining a regularity result
for u in Ls (Q).
If (for example) s(r, q) < s1 (r1 , q1 ) then we can use the following property
of the set E1 (also valid for E), which is easily seen to hold true: given
(r1 , q1 ) ∈ E1 and given s ∈ [(N + 2)p/N, s1 (r1 , q1 )), there exists (r2 , q2 ) ∈ E1 ,
with r2 ≤ r1 and q2 ≤ q1 , such that s = s1 (r2 , q2 ). In other words, we can use
the classical embeddings between Lebesgue spaces, and consider a function
in Lr1 (0, T ; Lq1 (Ω)) as a function in the bigger space Lr2 (0, T ; Lq2 (Ω)). But
then, as we said before, the regularity of u is Ls(r,q) (Q), that is, u has the
regularity of the less regular of the solutions with data f and −div (F ).
Moreover it is clear that the solutions belong to Lp (0, T ; W01,p (Ω)). We
thus have proved the following result.

Proposition 2.7 Let r and q be real numbers satisfying (2.4), and let r1
and q1 be real numbers satisfying (2.17).



Let f be a function in Lr (0, T ; Lq (Ω)) ∩ Lp (0, T ; W −1,p (Ω)), and let F :
Q → RN be such that |F | belongs to Lr1 (0, T ; Lq1 (Ω)). Let u be a solution
in Lp (0, T ; W01,p (Ω)) of


 u
+ A(u) = f − div (F ) in Q,
u(x, 0) = 0 in Ω,


u(x, t) = 0 on Γ,

and let
s = min{s, s1 } ,
where s is given by (1.10), and s1 is given by (2.19).
Then the norms of u in Ls (Q) and in Lp (0, T ; W01,p (Ω)) are bounded by
a constant which depends on α, p, N , q, r, q1 , r1 , Q, and on the norms of f
in Lr (0, T ; Lq (Ω)) and of |F | in Lr1 (0, T ; Lq1 (Ω)).

16
Remark 2.8 If (r, q) satisfy (1.11) then the a priori estimate of Proposition
2.7 yields an existence result for problem (1.7) with datum f − div (F ); this
result can be obtained proceeding as in the proof of Theorem 1.3 (see next
section).

3 Proof of the results

Proof of Theorem 1.1. The result follows directly from Proposition 2.2,



since in this case Lr (0, T ; Lq (Ω)) is embedded in Lp (0, T ; W −1,p (Ω)) (see
Remark 1.2).
If we are under the hypotheses of Theorem 1.3, then we need to prove the
existence of solutions since the datum f is not in the dual space. As stated
in the Introduction, we will proceed by approximation.
Proof of Theorem 1.3. Let {fn } be a sequence of functions in L∞ (Q) that
converges to f strongly in Lr (0, T ; Lq (Ω)), with r and q as in the statement
of the theorem. Let un be a solution of


 u
n + A(un ) = fn in Q,
un (x, 0) = 0 in Ω, (3.1)


un (x, t) = 0 on Γ.

Observe that such a solution exists (since fn belongs to the dual space)
and belongs to the space Lp (0, T ; W01,p (Ω)). By Proposition 2.2, the se-
quence {un } is bounded both in Ls (Q) (with s as in the statement) and
in Lp (0, T ; W01,p (Ω)). Thus, we can extract a subsequence, still denoted
by {un }, which converges weakly to some function u both in Ls (Q) and
in Lp (0, T ; W01,p (Ω)).
Moreover, since from the equation one obtains that {u
n } is bounded in



L1 (Q) + Lp (0, T ; W −1,p (Ω)), using compactness arguments (see for example
[11]), we obtain that un converges strongly to u in L1 (Q); in particular, it
converges to u almost everywhere in Q.
Furthermore, standard results (see again [11]) yield the compactness of
{un } in C 0 ([0, T ]; W −1,1 (Ω)), so that the initial condition un (0) = 0 passes
to the limit as n tends to ∞.

17
 Since the operator is nonlinear with respect to the gradient, all the con-
vergence results obtained so far are not enough in order to pass to the limit
in the weak formulation of (3.1). To proceed, we need to apply the following
result, which is proved in [3], Theorem 3.3, in a more general setting (see
also [4] for an analogous result).

Theorem 3.1 Let {un } be a sequence of solutions of (3.1) with data fn

which are convergent in the weak-∗ topology of measures; suppose that un
converges to some u weakly in Lσ (0, T ; W01,σ (Ω)) for some σ > 1. Then, up
to subsequences,

∇un → ∇u almost everywhere in Q.

Since our sequences {un } and {fn } satisfy the hypotheses of the above
theorem, we have the almost everywhere convergence of ∇un . This fact
implies that un converges strongly to u in Lρ (0, T ; W01,ρ (Ω)) for every ρ < p,
so that, by (1.5),
ρ
a(x, t, un , ∇un ) → a(x, t, u, ∇u) strongly in L p−1 (Q; RN ),

for every ρ < p. Thus, writing the weak formulation of (3.1), that is
  
− un ϕ
+ a(x, t, un , ∇un ) · ∇ϕ = fn ϕ ,
Q Q Q

for every ϕ ∈ C0∞ (Q), and using all the convergences proved so far, it is easy
to see that u satisfies
  


− uϕ + a(x, t, u, ∇u) · ∇ϕ = f ϕ,
Q Q Q

and so is a weak solution of (1.7).

Acknowledgments. We would like to thank Haı̈m Brezis for the kind letter
containing the results in the linear case we quoted in the Introduction.

18
 References
[1] D.G. Aronson, J. Serrin, Local behavior of solutions of quasilinear
parabolic equations, Arch. Rational Mech. Anal., 25 (1967), 81–122.

[2] L. Boccardo, Existence and regularity results for quasilinear parabolic

equations, in Proceedings of Partial Differential Equations Conference,
Trieste 1995, to appear.

[3] L. Boccardo, A. Dall’Aglio, T. Gallouët, L. Orsina, Nonlinear parabolic

equations with measure data, J. Funct. Anal., 147 (1997), 237–258.

[4] L. Boccardo, F. Murat, Almost everywhere convergence of the gradients

of solutions to elliptic and parabolic equations, Nonlinear Anal., 19
(1992), 581–597.

[5] H. Brezis, Personal communication.

[6] E. DiBenedetto, Degenerate parabolic equations, Springer-Verlag, New

York, 1993.

[7] O. Ladyženskaja, V.A. Solonnikov, N.N. Ural’ceva, Linear and quasilin-

ear equations of parabolic type, Translations of the American Mathe-
matical Society, American Mathematical Society, Providence, 1968.

[8] J.-L. Lions, Quelques méthodes de resolution des problèmes aux limites
non linéaires, Dunod, Gauthier-Villars, Paris, 1969.

[9] L. Nirenberg, On elliptic partial differential equations, Ann. Scuola

Norm. Sup. Pisa, 13 (1959), 116–162.

[10] M.M. Porzio, Local regularity results for some parabolic equations,
preprint.

[11] J. Simon, Compact sets in the space Lp (0, T ; B), Ann. Mat. Pura Appl.,
146 (1987), 65–96.

19

View publication stats