Interior Integral Estimates on Weak Solutions of Nonlinear Parabolic Systems
J. Naumannand M. Wolff
Abstract
This paper concerns various types of CACCIOPPOLI and POINCARE inequalities on weak solutions uof nonlinear parabolic systems. The main result of the paper is the local integrability of the spatial gradientD uto an exponent p>2.
Contents
1. Introduction
2. Regularization and localization 3. CACCIOPPOLI inequalities 4. POINCARE inequalities
5. Extended CACCIOPPOLI inequality 6. Local higher integrability ofD u References
1 Introduction
Let
IR
n (n
2) be a bounded domain, and let 0< T <
+1 be xed. SetQ
= (0T
).We consider the following system of PDE's:
@u
i@t
;D a
i(xtuDu
) =f
i;D g
i 1) inQ
(i
= 1:::N
) (1.1)where:
u
= fu
1:::u
Ng (N
1)Du
= fD u
ig (matrix of rst spatial derivatives ofu
)D
=@
@x
= 1:::n
f
=ff
1:::f
Ng andg
=fg
ig are given functions inQ
. Leta
=fa
ig be a matrix. Denek
a
k= Xn=1
N
X
i=1(
a
i)21=2:
1)Throughout a repeated Greek resp. Latin index implies summation over 1 ::: nresp. 1 ::: N.
Throughout the whole paper, we impose on the functions
a
i in (1.1) the following conditions:8
>
<
>
:
a
i is a Caratheodory function onQ
IR
NIR
nN (= 1:::n i
= 1:::N
)(1.2)
8
>
<
>
:
k
a
(xtu
)ka
0(1 +ju
j(n+2)=n+kk)8(
xtu
)2Q
IR
NIR
nN (a
0= const) (1.3)8
>
<
>
:
a
i(xtu
)i 0kk28(
xtu
)2Q
IR
NIR
nN (0= const>
0):
(1.4)Let
W
p1( ) =fv
2L
p( ) :D v
2L
p( ) (= 1:::n
)g denote the usual Sobolev space over . Next, deneW
211(Q
) =nv
2L
2(Q
) :D v @v @t
2L
2(Q
) ( = 1:::n
)o andW
210(Q
) = fv
2L
2(Q
) :D v
2L
2(Q
) (= 1:::n
)gV
210(Q
) = nv
2W
210(Q
) : esssup(0T)
Z
v
2(xt
)dx <
+1o:
The following imbedding theorem is well-known (cf. e.g. 5]):8
>
>
>
>
>
>
>
>
>
>
<
>
>
>
>
>
>
>
>
>
>
:
Let ! IR
n be a bounded domain with Lipschitz boundary@!: Let
@!
be relatively open. Then:k
v
kL2(n+2)=n(! (0T))c
0 esssup(0T)
Z
!
v
2(xt
)dx
+ZT0 Z
!
j
Dv
j2dx
dt
1=2 for allv
2V
210(!
(0T
))v
= 0 a.e. on (0T
) (c
0= const>
0):
(1.5)By
L
p(Q IR
N),W
211(Q IR
N) etc. we denote the space of vector valued functionsv
=fv
1:::v
Ng the components of which belong toL
p(Q
) resp.W
211(Q
) etc.We introduce the notion of weak solution of (1.1) regardless of whether or not this solution is subject to any boundary or initial condition.
Let (1.2) and (1.3) be satised. Without any further reference, throughout the whole paper, we assume
g
2L
2(Q IR
nN):
DEFINITION 1.1 Assume
1
f
2L
2(n+2)=(n+4)(Q IR
N) resp.2
f
2L
1(Q IR
N).The function
u
2V
210(Q IR
N) is called a weak solution of (1.1) ifZ
u
i(xt
)v
i(xt
)dx
;Zt0 Z
u
i@v
i@s
dx
ds
+Zt0 Z
a
i(xsuDu
)D v
idx
ds
==Zt
0 Z
(
f
iv
i+g
iD v
i)dx
ds
(1.6)for a.a.
t
2(0T
) and all test functions1
v
2W
211(Q IR
N), supp(v
) (0T
] resp.2
v
2W
211(Q IR
N)\L
1(Q IR
N), supp(v
) (0T
].The aim of the present paper is to prove various interior integral estimates on any weak solution
u
of (1.1). Following 5], we rst regularizeu
and localize then (1.6) with respect tot
. Then CACCIOPPOLI and POINCARE inequalities are readily obtained (Sections 3 and 4).Next, in Section 5 we prove an extended version of the preceeding CACCIOPPOLI inequality which is appearently new in the theory of parabolic systems. Our main result is the interior higher integrability of
Du
(i.e.Du
2L
plo cfor ap >
2) which is presented in Section 6. It is based on the parabolic analogue of the well-known higher integrability by reverse Holder inequality due to GEHRING-GIAQUINTA-MODICA.This method has been used in 3], 6] to prove the higher integrability of the spatial gradient of bounded weak solutions to parabolic systems with quadratic growth nonlinearities. Under more restrictive assumptions on the coecients
a
i, similar results have been obtained in 1], 2]by an entirely dierent technique.
The higher integrability of the spatial gradient of weak solution to a nonlinear parabolic system is of interest in itself. On the other hand, it is also a basic tool in the proof of partial regularity of weak solutions of nonlinear parabolic systems.
2 Regularization and localization
Let
f
2L
1(Q
). Given anyt
0 2(0T
) andk > T
;1t
0 we introduce the Steklov mean off
:f
k(xt
) =k
t+ 1
k
Z
t
f
(xs
)ds
for a.a. (xt
)2 (0t
0):
We note some properties of
f
k which will be used in what follows. Lett
02(0T
).PROPOSITION 2.1 Let
f
2L
p(Q
) (1p <
+1) andg
2L
p0(Q
). Thent0+Z 1k 0
Z
f
(xt
) Ztt;k1
g
(xs
)ds
dx
dt
= t0
Z
0 Z
tZ+1k
t
f
(xs
)ds
g
(xt
)dx
dt
for allk > T
;1t
0.PROPOSITION 2.2 1. Let
f
2L
p(Q
) (1p
+1). Thent0
Z
0 Z
j
f
kjpdx
dt
Zt00 Z
j
f
jpdx
dt
(1p <
+1) esssup(0t0)j
f
kjesssup(0t0)j
f
j for allk > T
;1t
0.2. Let
f
2L
p(Q
) (1p <
+1). Then (i)f
k !f
inL
p( (0t
0)) ask
!+1(ii)
f
k(t
)!f
(t
) inL
p( ) ask
!+1 for a.a.t
2(0t
0).3. Let
f
2L
1(Q
),g
2L
1(Q
). Then there exists a subsequencefk
jg such thatt0
Z
0 Z
f
kjg
kjdx
dt
;!Zt00 Z
fg
dx
dt as j
!1:
The following result shows the eect of regularization with respect to
t
of the Steklov mean.PROPOSITION 2.3 Let
f
2W
210(Q
). Thenf
k 2W
211( (0t
0)) and there holdsD f
k(xt
) = (D f
)k(xt
) (= 1:::n
)@f
k@t
(xt
) =k
hf xt
+ 1k
;
f
(xt
)i for a.a. (xt
)2 (0t
0) and allk > T
;1t
0.We are now going to localize (1.6) with respect to
t
. To this end, lett
0 2(0T
) be arbitrary.We consider the Steklov mean with integers
k > T
;1t
0.THEOREM 2.1 Assume
1
f
2L
2(n+2)=(n+4)(Q IR
N) resp.2
f
2L
1(Q IR
N).Let
u
2V
210(Q IR
N) be a weak solution of (1.1). ThenZ
@u
ik@t '
idx
+Z
(
a
i)kD '
idx
=Z
((
f
i)k'
i+ (g
i )kD '
i)dx
(2.1)for a.a.
t
2(0t
0), all integersk > T
;1t
0 and all test functions 1'
2W
21(IR
N), supp('
) resp.2
'
2W
21(IR
N)\L
1(IR
N), supp('
) .Proof. Fix an integer
m > n
2. Let j (j
= 12:::
) be open sets such that:j j j+1
::: @
j smooth 1j=1 j =
:
The Sobolev imbedding theorem implies
W
2m( j)C
( j) (j
= 12:::
). DeneW
2m( j) =n'
2W
2m( j) :'
=@'
@
=:::
=@
m;1'
@
m;1 = 0 a.e. on@
jo
:
Let
'
2W
2m( jIR
N) be arbitrary. We extend'
by zero onto n j and denote this function in again by'
. Let 2C
(IR
) have its support in (0t
0). Then the functionv
(xt
) =k'
(x
) Ztt;1k
(s
)ds
(xt
)2Q k
integer> T
;1t
0 is admissible in (1.6).Let
t
1 2t
0+ 1kT
. Clearly,
v
(t
1) = 0 andt1
Z
0 Z
u
i(xt
)@v
i@t
(xt
)dx
dt
= ;k
Zt00 Z
h
u
ixt
+ 1k
;
u
i(xt
)i'
i(x
)(t
)dx
dt
= ;Zt0
0 Z
@u
ik@t
(xt
)'
i(x
)(t
)dx
dt
(cf. Prop. 2.3). Now (1.6) witht
1 2t
0+ 1kT
gives
t0
Z
0 Z
@u
ik@t '
idx
dt
+ t0
Z
0 Z
(
a
i)kD '
idx
dt
= t0
Z
0 Z
(
f
i)k'
i+ (g
i)kD '
i]dx
dt:
Thus, by a standard argument,
Z
j
@u
ik@t
(xt
)'
i(x
)dx
+Z
(
a
i)k(xt
)D '
i(x
)dx
==Z
j
(
f
i)k(xt
)'
i(x
) + (g
i)k(xt
)D '
i(x
)]dx
(2.2)for all
t
2(0t
0)nE
jk where measE
jk = 0 (notice that by virtue of the separability ofW
2m( j) the setE
jk does not depend on'
). DeneE
=j1S=1 S
k>(T;t0);1
E
jk. Then measE
= 0, and (2.2) is true for a.a.t
2(0t
0), forj
= 12:::
and all integersk > T
;1t
0.Let
'
satisfy 1 resp. 2 above. Fixj
such that supp('
) j. Let'
be the standard mollication of'
. Then'
2W
2m( jIR
N) for all 0< <
dist(supp('
)@
j) and'
!'
inW
21( jIR
N) as ! 0. If, in addition,'
2L
1(IR
N) then there exists a subsequence off
'
g(not relabelled) such that'
!'
a.e. in j. Hence, inserting'
in (2.2) and letting tend !0 gives the claim.3 CACCIOPPOLI inequalities
Dene
B
r =B
r(x
0) =fx
2IR
n:jx
;x
0j< r
gQ
r =Q
r(x
0t
0) =B
r(x
0)(t
0;r
2t
0):
Let (
x
0t
0)2Q
. Fix any 0< r <
12pt
0 such thatB
2r . Let 2C
1(IR
n)2)and 2C
1(IR
) be cut-o functions as follows:8
>
<
>
:
1 onB
r 0 inIR
nnB
2r 01 jD
jc
0r
inIR
n8
>
<
>
:
1 on (t
0;r
2+1) 0 on (;1t
0;4r
2) 0 1 00c
0r
2 onIR
n (c
0= const>
0 independent ofr
).Following 3] we dene for any
v
2L
1(Q
2r)v
~2r(t
) = ZB2r
2dx
;1 ZB2r
v
(yt
)2(y
)dy
for a.a.t
2(t
0;4r
2t
0):
Let
v
2W
211(Q
2r). Then the functiont
7! ~v
2r(t
) possesses a weak derivative d~v
2rd
t
2L
2(t
0 ; 4r
2t
0) and there holdsd~
v
2rd
t
(t
) = gdv
dt
2r(
t
) for a.a.t
2(t
0;4r
2t
0):
2)To emphasize the dependence of onB2r, below we shall write2r in place of,r etc.
Let (1.2) { (1.4) be satised. In what follows, we consider the two cases3)
8
>
<
>
:
f
2L
2(n+2)=(n+4)(Q IR
N)u
2V
210(Q IR
N) is a weak solution of (1.1) (3.1)or
8
>
<
>
:
f
2L
1(Q IR
N)u
2V
210(Q IR
N)\L
1(Q IR
N) is a weak solution of (1.1):
(3.2)We begin by proving the following
THEOREM 3.1 Assume (3.1) or (3.2). Then, for every
" >
0, 12Z
B2r
j
u
(xt
);j22(x
)dx
2(t
) + (0;"
) Ztt0;4r2
Z
B2r
k
Du
k222dx
ds
c
1 1 + 1"
1
r
2Z
Q2r
j
u
;j2dx
ds
+c
1 ZQ2r(1 +j
u
j2(n+2)=n22)dx
ds
(3.3)+ Z
Q2r
j
f
jju
;j22dx
ds
+c
1 1 + 1"
Z
Q2r
k
g
k2dx
ds
for a.a.t
2(t
0;4r
2t
0) and all 2IR
N12
Z
B2r
j
u
(xt
);u
~2r(t
)j22(x
)dx
2(t
) + (0;"
) Ztt0;4r2
Z
B2r
k
Du
k222dx
ds
c
2 1 + 1"
1
r
2Z
Q2r
j
u
;u
~2rj2dx
ds
+c
2 ZQ2r (1 +j
u
j2(n+2)=n22)dx
ds
(3.4)+ Z
Q2r
j
f
jju
;u
~2rj22dx
ds
+c
2 1 + 1"
Z
Q2r
k
g
k2dx
ds
for a.a.
t
2(t
0;4r
2t
0) where the constantsc
1,c
2 depend neither onr
nor on"
. Proof. Letk
be any integer> T
;1t
0. The function'
= (u
k(t
);)22(t
)t
2(t
0;4r
2t
0) is admissible in (2.1). We obtain12
Z
B2r
j
u
k(t
);j22dx
2(t
) + Ztt0;4r2
Z
B2r(
a
i)k(D u
ik)22dx
ds
==;2 Zt
t0;4r2
Z
B2r(
a
i)k(u
ik;i)(D
)2dx
ds
3)Recall thatg2L2(QIRnN) throughout.
+ Zt
t0;4r2
Z
B2r
j
u
k ;j220dx
ds
+ Ztt0;4r2
Z
B2r(
f
i)k(u
ik;i)22dx
ds
+ Ztt0;4r2
Z
B2r(
g
i)k(D u
ik)2+ 2(u
ik;i)(D
)]2dx
ds
for allt
2(t
0;4r
2t
0). Using Prop. 2.2 we nd by letting tendk
!+112
Z
B2r
j
u
(t
);j22dx
2(t
) +0 Ztt0;4r2
Z
B2r
k
Du
k222dx
ds
2
c
0r
t
Z
t0;4r2
Z
B2r(1 +j
u
j(n+2)=n+kDu
k)ju
;j2dx
ds
+c
0r
2t
Z
t0;4r2
Z
B2r
j
u
;j2dx
ds
+ Ztt0;4r2
Z
B2r
j
f
jju
;j22dx
ds
+ Ztt0;4r2
Z
B2r
k
g
k kDu
k22+ 2c
0r
ju
;jdx
ds:
Then (3.3) is readily seen by employing Young's inequality.
To prove (3.4), we rst note that
Z
B2r
u
ik(xt
);(gu
ik)2r(t
)]2(x
)dx
= 0 (i
= 1:::N
) for allt
2(t
0;4r
2t
0). HenceZ
B2r
@u
ik@t
(t
)(u
ik(t
);(gu
ik)2r(t
))2dx
= 12 ZB2r
dtd j
u
k(t
);(gu
k)2r(t
)j22dx
and thereforet
Z
t0;4r2
Z
B2r
@u
ik@s
(u
ik;(gu
ik)2r)22dx
ds
== 12BZ
2r
j
u
k(t
);(gu
k)2r(t
)j22dx
2(t
); Zt0;4r2
Z
B2r
j
u
k;(gu
k)2rj220dx
ds
for allt
2(t
0;4r
2t
0).Now we insert
'
= (u
k(t
);(gu
k)2r(t
))22(t
)t
2(t
0;4r
2t
0) into (2.1) and nd12
Z
B2r
j
u
k(xt
);(gu
k)2r(t
)j22(x
)dx
2(t
) + Ztt0;4r2
Z
B2r(
a
i)k(D u
ik)22dx
ds
==;2 Zt
t0;4r2
Z
B2r(
a
i)k(u
ik;(gu
ik)2r)(D
)2dx
ds
+ Zt0;4r2
Z
B2r
j
u
k ;(gu
k)2rj220dx
ds
+ Zt0;4r2
Z
B2r(
f
i)k(u
ik ;(gu
ik)2r)22dx
ds
+ Ztt0;4r2
Z
B2r(
g
i)k(D u
ik)2+ 2(u
ik;(gu
ik)2r)(D
)]2dx
ds
for all
t
2 (t
0;4r
2t
0). Lettingk
! +1 and using an analogous reasoning as above we nd (3.4).From Theorem 3.1 we derive
COROLLARY 3.1 (CACCIOPPOLI inequalities) Assume (3.1). Then esssup
(t0;r2t0)
Z
Br
j
u
(xt
);j2dx
+ZQr
k
Du
k2dx
dt
c
28
>
<
>
:
r
12Z
Q2r
j
u
;j2dx
dt
+ ZQ2r(1 +j
u
j2(n+2)=n+kg
k2)dx
dt
(3.5)+ Z
Q2r
j
f
j2(n+2)=(n+4)dx
dt
(n+4)=(n+2)9
>
=
>
for all 2
IR
N, and esssup(t0;r2t0)
Z
Br
j
u
(xt
);u
~r(t
)j2dx
+ZQr
k
Du
k2dx
dt
c
28
>
<
>
:
r
12Z
Q2r
j
u
;u
~2rj2dx
dt
+ ZQ2r(1 +j
u
j2(n+2)=n+kg
k2)dx
dt
(3.6)+ Z
Q2r
j
f
j2(n+2)=(n+4)dx
dt
(n+4)=(n+2)9
>
=
>
(
c
2= const>
0 independent ofr
).Proof. Fix
"
= 02 . Then (3.3) implies 12
Z
B2r
j
u
(xt
);j22(x
)dx
2(t
) + 0 2t
Z
t0;4r2
Z
B2r
k
Du
k222dx
ds
c
8
>
<
>
:
r
12Z
Q2r
j
u
;j2dx
ds
+ ZQ2r(1 +j
u
j2(n+2)=n+kg
k2)dx
ds
(3.7)+Z
Q2r
j
f
jju
;j22dx
ds
9
>
=
>
for a.a.
t
2(t
0;4r
2t
0). Hence, the essential supremum of the functiont
7! ZB2r
j
u
(xt
);j22(x
)dx
2(t
)over the interval (
t
0;4r
2t
0), as well as the integralZ
Q2r
k
Du
k222dx
ds
are bounded by the right-hand side of (3.7). Thus esssup
(t0;4r2t0)
Z
B2r
j
u
(xt
);j22(x
)dx
2(t
) + ZQ2r
k
Du
k222dx
ds
c
8
>
<
>
:
r
12Z
Q2r
j
u
;j2dx
ds
+ ZQ2r(1 +j
u
j2(n+2)=n+kg
k2)dx
ds
(3.8)+Z
Q2r
j
f
jju
;j22dx
ds
9
>
=
>
:
Next, we combine the imbedding (1.5) and Holder's and Young's inequality to obtain
Z
Q2r
j
f
jju
;j22dx
ds
8
>
<
>
:
esssup
(t0;4r2t0)
Z
B2r
j
u
(xt
);j22(x
)dx
2(t
) + ZQ2r
k
D
((u
;)22)k2dx
dt
9
>
=
>
+
c
Z
Q2r
j
f
j2(n+2)=(n+4)dx
dt
(n+4)=(n+2)for all
>
0. Thus, by choosing appropriately small, from (3.8) we obtain (3.5).By an analogous argument, from (3.4) it follows that esssup
(t0;4r2t0)
Z
B2r
j
u
(xt
);u
~2r(t
)j22(x
)dx
2(t
) + ZQ2r
k
Du
k222dx
dt
c
8
>
<
>
:
r
12Z
Q2r
j
u
;u
~2rj2dx
ds
+ ZQ2r(1 +j
u
j2(n+2)=n22+kg
k2)dx
ds
(3.9)+ Z
Q2r
j
f
j2(n+2)=(n+4)dx
ds
(n+4)=(n+2)9
>
=
>
:
We estimate the rst term on the left of (3.9) from below. To this end, let
r denote the cut-o function with respect toB
r (analogously as2r with respect toB
2r cf. footnote 2). We haveZ
Br
r2d