Cauchy-Dirichlet Problems

For A Class Of Quasilinear Degenerate

Parabolic Equations

By XU Chao-Jiang and NIE Lei
Institute of Mathematics, Wuhan University
430072, Wuhan, P.R.China
ABSTRACT
In this work, we study the quasilinear Cauchy-Dirichlet problem,
t
u +

m
i,j=1
X

i
A
ij
(x, t, u)X
j
u

m
j=1
B
j
(x, t, u)X
j
u+c(x, t, u) = 0, in ]0, T[; u|
t=0
=
, on ; u = , on [0, T[., where X = {X
1
, . . . , X
m
} is a system of real
smooth vector elds which is dened on a open domain M of IR
n
, and satises
the Hormanders condition, M. Assume that is non characteristic for
the system X
1
, . . . , X
m
. Under some hypothesis for the boundary of domain and
the elliptic structure condition for the non-linear coecients A
ij
, B
j
, C, i, j =
1, . . . , m, we have proved that the existence and regularity of solution for above
Cauchy-Dirichlet problems.
Key Words degenerate parabolic equation, vector elds, non-isotropic
Holders space, Cauchy-Dirichlet problems.
Classication 35I, 35H.
1 Introduction
In this work, we study the following quasilinear Cauchy-Dirichlet problem:
Lu
t
u +
m

i,j=1
X

i
A
ij
(x, t, u)X
j
u
+
m

j=1
B
j
(x, t, u)X
j
u + C(x, t, u) = 0, in ]0, T[ (1)
u|
t=0
= , on (2)
u = on [0, T[. (3)

J. Wuahn Univ. 17(1996) 167180

1
where X = {X
1
, . . . , X
m
} is a system of real smooth vector elds dened in an
open domain M IR
n
, n 2, is a bounded open subdomain of M with
smooth. X

j
= X
j
+c
j
is the formula adjoint of X
j
. We assume that the system
of vector elds X = {X
1
, . . . , X
m
} satises the following Hormanders condition
(H):
X
1
, . . . , X
m
together with their commutators X

= [X

1
, . . . [X

s1
, X
s
] . . .]
up to some xed length r span the tangent space at each point of M.
We will study the problem (1)(3) similar to the case of second order parabolic
equations. The role of Laplaceian
x
is substituted by the Hormanders opera-
tors H =

m
j=1
X

j
X
j
+c, where c c
0
> 0. Actually by using the geometry and
the function spaces associated with the system of vector elds X, the operators
H seems to satisfy nearly all properties of Laplacian
x
(see [1, 4, 6, 11]).
We assume that the system of vector elds X and the boundary satises
the following additional conditions (S. E. )
is non characteristic for the system X. And for all 1 j r,
we have X

j
= X
j

T
x
() for all x , and the dimension of X
j
is
constant in a neighborhood of .
Where X
1
is the linear space spanned by the vector elds X
1
, . . . , X
m
with
smooth real coecients in C

(M), X
j
= [X
1
, X
j1
]. And for x , X

1
=
X
1

T
x
(), X

j
= [X

1
, X

j1
]. S
k,
() is the non-isotropic Horlder space asso-
ciated with the system of vector elds X (see [8] and Section 2).
Then, the Hormanders condition implies that X
r
(x) = T
x
M for all x M.
And the condition (S. E. ) implies that the bases of X

1
(vector elds dened
on ) satises the Hormanders condition as well on the manifold at order
r.
For the nonlinear coecients, we assume that there exists constants , > 0
and g(x, t) C
0
([0, T[) such that:
A
i,j
, B
j
, C C

([0, T[ IR), i, j = 1, . . . , m; (4)

|A
ij
(x, t, z)|, |B
j
(x, t, z)|, |C(x, t, z)| g(x, t); (5)
||
2

i,j
A
ij
(x, t, z, )
i

j
||
2
,
(x, t, z, ) [0, T[ IR
n+1
. (6)
We prove in this work the following theorem:
Theorem 1 Assume that the system of vector elds X
1
, . . . , X
m
and satises
the Hormanders condition and (S. E. ). Let the nonlinear coecients of
equation (1) satisfy the conditions (4)-(6). then there exists a solution of problems
(1)(3) in the class C

([0, T[), if , are in the class of C

.
We will give the essential notations and well know results of Hormanders
operators in Section 2. We study linear problems in Section 3, and nonlinear
problems in Section 4.
2
2 Preliminary Lemmas And Notations
We dene now the sub-unit metric on M associated with X as in [4] and [8].
Denition 1 Let C() be a class of absolutely continuous mappings : [0, 1]
M which almost everywhere satisfy the dierential equation

(t) =

|J|r
a
J
(t)X
J
((t)) (7)
with |a
J
(t)| <
|J|
, then we dene
(x, y) = inf{ > 0 | C() with (0) = x, (1) = y}. (8)
Then, is a local metric on M. We introduce now a class of non-isotropic
Holder continuous functions. For 1 > > 0, we dene the function spaces
S
0
([0, T[) = C
0
([0, T[), and S

([0, T[) by
_
_
_
f S
0
([0, T[); sup
x,y;t[0,T[
|f(x, t) f(y, t)|
(x, y)

< +
sup
x;t,[0,T[
|f(x, t) f(x, )|
|t |
/2
< +
_
_
_
. (9)
And for k, l IN, 1 > 0, we dene S
k,l

([0, T[) by
{u S

([0, T[) ;
h
t
X
J
u S

([0, T[),
h k, 2h +|J| l}. (10)
The denition of norm of S
k,l

([0, T[) is classic, and it is also a Banach

space as in [8] (see also [5, 12]).
Using the hypotheses (S. E. ) on , we can also dene the functions spaces
S
k,l

([0, T[) by use the bases of X

1
as in (9) (10) (see [5, 12]). We have also
the following interpolation inequality and compactness results.
Lemma 1 For any k
1
k
2
, l
1
l
2
, , k
1
+l
1
+ < k
2
+l
2
+, and > 0,
there exist C

> 0 such that for any u S

k
2
,l
2

([0, T[), we have

u
S
k
1
,l
1

([0,T[)
u
S
k
2
,l
2

([0,T[)
+ C

u
L

(]0,T[)
. (11)
Then the embedding from S
k
2
,l
2

([0, T[) to S
k
1
,l
1

([0, T[) is a compact map.

3
3 Existence and estimation of solutions
for linear problems
We consider in this section the following linear problems:

t
u + Hu = f(x, t), in ]0, T[, (12)
u|
t=0
= on , (13)
u = on [0, T] . (14)
We will use the method of Hilbert spaces to study existence of weak solution for
above problems. We dene now
M
1
(]0, T[) = {u L
2
(]0, T[);
t
u L
2
(]0, T[),
X
j
u L
2
(]0, T[), 1 j m};
m
1
() = {u L
2
(); X
j
u L
2
(), 1 j m}.
The norm of M
1
(]0, T[), m
1
() is classic, and it is Hilbert spaces. m
1
0
()
denote the closure of D() in m
1
(), and we have D(H) = m
1
0
().
For u, v m
1
0
(), we dene
a(u, v) = (Hu, v)
m
1
()m
1
()
=
_

j=1
X
j
uX
j
vdx + +
_

cuvdx
Then we have
a(u, u) c
0
u
2
m
1
()
; (15)
|a(u, v)| Cu
m
1
()
v
m
1
()
. (16)
Using the abstract existence theorem (see [3]), we obtain the existence of weak
solution for degenerate parabolic problems :
Theorem 2 If f L
2
(]0, T[; m
1
0

()), L
2
(). Then there exists an unique
solution u W(]0, T[; m
1
0
()) = {v L
2
(]0, T[; m
1
0
());
t
v L
2
(]0, T[; m
1
0

())}
of following Cauchy problems:

t
u + Hu(t) = f(t), (17)
u(0) = u
0
. (18)
where m
1
0

() is the adjoint Hilbert spaces of m

1
0
().
Since L
2
(]0, T[) L
2
(]0, T[; m
1
0

()), then for f L

2
(]0, T[),
L
2
(),

M
1
(]0, T[), we have proved the existence of weak solution for
problems (12)14) with =

|
[0,T]
. Here the existence of trace for func-
tion of M
1
(]0, T[) is given in [2, 5, 12]. We have also W(]0, T[; m
1
0
())
4
M
1
(]0, T[). Let u = v

, then v will be a solution of problems (17)(18)
with

f(x, t) = f(x, t) +
t

(x, t) + H

(x, t) L
2
(]0, T[; m

0
()).
For the regularity of this weak solution, if f C

([0, T[), C

(),
C

([0, T[), then u is also in C

([0, T[). But we have to use a more

precise regularity results in the associated Holder spaces S
k,l

([0, T[) to study

nonlinear problems. So we will use the following estimate of heat kernel. Since the
operators
t
+H is hypoelliptic, and the boundary [0, T[ is noncharacteristic,
there exists the heat kernel K
t
(x, y) of Cauchy-Dirichlet problems (12)-(14).
Lemma 2 For any N IN and t > 0 small, we have the following estimate of
heat kernel:
|
k
t
X
J
K
t
(x, y)| C
N,|J|,k
t
k
|J|
2
|B(x; t
1/2
)|
1

_
1 +
(x, y)
2
t
_
N
. (19)
where the dierentiations of X
J
are taken in x or y.
This is just Theorem II of [6]. Using the hypothesis (S. E. ). We can prolong
the functions of S
k,l

([0, T[) and heat kernel K

t
to the exterior of [0, T[, so
we have also the estimate (19) near the boundary of domain. The following is
maximum principle for degenerate parabolic operators (see [12]).
Lemma 3 If f C
0
([0, T[), C
0
(), C
0
([0, T[), and u
S
1,2

(]0, T[) is a solution of problems (12)-(14). Then we have the following a

priori estimate for all t [0, T[,
u
L

(]0,T[)
C{
L

()
+
L

([0,T[)
+ f
L

(]0,T[)
}. (20)
Using the estimate of heat kernel and maximum principle, we have proved the
following precise regularity results (see [12]).
Theorem 3 For k IN, 0 1, let f S
k,2k

([0, T[), S
2k+2,
(),
S
k+1,2k+2

([0, T[), there exists an unique solution of problems (12)-(14) in the

function space S
k+1,2k+2

([0, T[). And the solution veries following estimate:

u
S
k+1,2k+2

([0,T[)
C{
S
2k+2,
()
+
S
k+1,2k+2

([0,T[)
+ f
S
k,2k

([0,T[)
}, (21)
where the constant C is independent of u.
Using the freezing coecients method, we can also study the following linear
problems

t
u +
m

i,j=1
X

i
a
ij
(x, t)X
j
u +
m

j=1
b
j
(x, t)X
j
u
5
+c(x, t)u = f(x, t), in ]0, T[ (22)
u = , on {t = 0} (23)
u = , on [0, T[. (24)
Assume that, for some k IN, > 0, the coecients a
ij
, b
j
, c satises the condi-
tions:
||
2

i,j=1
a
ij
(x, t)
i

j
||
2
,
(x, t, ) [0, T[ IR
n
; (25)
m

i,j=1
a
ij

S
k,2k

+
m

j=1
b
j

S
k,2k

+c
S
k,2k

. (26)
We have the following results.
Corollary 1 Assume that the coecients a
ij
, b
j
, c satises the conditions (25)-
(26), and f S
k,2k

([0, T[), S
2k+2,
(), S
k+1,2k+2

([0, T[). Then

the solutions of problems (22)-(24) is in the class S
k+1,2k+2

([0, T[), and we

have also the estimate
u
S
k+1,2k+2

([0,T[)
C{
S
2k+2,
()
+
S
k+1,2k+2

([0,T[)
+ f
S
k,2k

([0,T[)
}, (27)
where C depends only on , , r.
The proof of this Corollary is given in [12]. We will use those results to study
the existence of smooth solution for quasilinear problems (1)-(3) in Section 4.
4 Existence of solution for quasilinear
problems
We consider now the quasilinear problems (1)-(3). Using the results of Corol-
lary 1 and Leray-Schauders xed point theorem, we have the following existence
theorem.
Theorem 4 Assume that S
2,
(), S
1,2

([0, T[). Under the hypoth-

esis of Theorem 1, if there exist constant B > 0, > 0 such that for all solution
u S
1,2

([0, T[) of following problems (0 1):

t
u +
m

i,j=1
X

i
A
ij
(x, t, u)X
j
u +
m

j=1
B
j
(x, t, u)X
j
u
+C(x, t, u) = 0, in ]0, T[ (28)
u|
t=0
= , on (29)
u = on [0, T[, (30)
6
satises the a priori estimate:
u
S

([0,T[)
B. (31)
Then there exist a solution u S
1,2

([0, T[) for quasilinear problems (1)-(3).

In fact, we dene the operator T : S

([0, T[) S

([0, T[) by u = Tv
is the solution of following linearization problems:

t
u +
m

i,j=1
X

i
A
ij
(x, t, v)X
j
u +
m

j=1
B
j
(x, t, v)X
j
u
+C(x, t, v) = 0, in ]0, T[
u|
t=0
= , on ,
u = on [0, T[.
Since a
j
(x, t) = A
j
(x, t, v(x, t)), b
j
(x, t) = B
j
(x, t, v(x, t)), c(x, t) = C(x, t, v(x, t))
S

([0, T[), using the results of Corollary 1, we have T : S

([0, T[)
S
1,2

([0, T[). Then Lemma 1 give that T : S

([0, T[) S

([0, T[) is a
compact operator, so that T has a xed point in S

([0, T[). Corollary 1 give

the results of Theorem 4.
Now, the proof of Theorem 1 is to construct the a priori estimate (31). We
sckicth this process in two step.
(1) Estimate of max |u|.
Denote by = {{t = 0} }

{[0, T[}, then we have

Lemma 4 Assume that the nonlinear coecients A
ij
, B
j
, C, i, j = 1, . . . , m sat-
ises the hypothesis (4)-(6), and u S
1,2

([0, T[) veries equation (1). Then

there exists constant C = C(sup
[0,T[
|g|, , , ||, sup

|u|), such that

sup
]0,T[
|u| C. (32)
The proof of this Lemma is as following. We have the following Sobolev inequality,
for u M
1
(]0, T[), there exists q ]2,
2n
n2/r
[ such that
u
L
q,l
(]0,T[)
Cu
M
1
(]0,T[)
, (33)
where u
L
q,l
(]0,T[)
=
_
_
T
0
(
_

|u(x, t)|
q
dxdt)
l/q
_
1/q
, and l veries 1/q =
2(2/r)
nl
+
1(2/l)
2
. Using Nash-Moser processor and inequality (33), we can obtain estimate
(32) as in [7] (see also [5]).
(2) Estimate of u
S
.
7
The estimate of u
S
is given by Harnacks inequality for quasilinear degen-
erate parabolic equation (1), there is also two step for this processor, one is in
the interior of domain and on a ball of B(x, ), another is near the boundary
and on a semi-ball B(x, )
+
. The method is standard for degenerate elliptic and
parabolic equation under the hypothesis of Theorem 1, we reference the proof to
the detail paper [5].
End of proof of Theorem 1
Now we have proved that for S
2,
(), S
1,2

([0, T[), there exists

a solution u S
1,2

(]0, T[) for quasilinear problems (1)-(3). Using the linear

regularity results of Theorem 3 with freezing method, we can prove that if
S
2k+2,
(), S
k+1,2k+2

([0, T[), then u S

k+1,2k+2

(]0, T[) for any k

IN. Since S
k+1,2k+2

(]0, T[) C
(2k+2)/r
([0, T[), we have proved Theorem 1.
References
[1] Bony, J. M., Principe du maximum, inegalite de Harnack et uninite du
probl`eme de Cauchy pour les operateurs elliptiques degenerees , Ann. Inst.
Fourier 19 (1969), 227304.
[2] Derrdj, M., Un probl`eme aux limites pour une classe doperateurs du second
ordre hypoelliptiques, Ann. Inst. Fourier 21 (1971), 99148.
[3] Ladyzenskaja O. A. & Solonnikov V. A. & Uralceva N. N., Linear and quasi-
linear equation of parabolic type, Translation of Math. Mono. 23, AMS 1968.
[4] Nagel, A., Stein, E. M., and Wainger, S., Balls and metrics dened by vector
elds I, basic properties, Acta Math., 155 (1985), 103147.
[5] Nie, L., Existence of solutions for a class of quasilinear degenerate parabolic
equation, Master thesis of Wuhan University, April 1993, and paper to ap-
pear.
[6] Sanchez-Calle, A., Fundamental solutions and geometry of the sum of squares
of vector elds, Invent. Math., 78 (1984), 143160.
[7] Xu, C. J., Subelliptic variational problems, Bull. Soc. Math. France, 118
(1990), 147159.
[8] Xu C. J., Regularity for quasilinear second order subelliptic equations,
Comm. Pure Appl. Math., (1992), 7796.
[9] Xu, C. J., Hypoellipticity of nonlliear second order partial dierential equa-
tions, Jour. P. D. E. , 1 (1988), 8595.
8
[10] Xu, C. J., Probl`eme de Dirichlet pour les equations associees `a un syst`eme
de champs de vecteurs, Seminaire E. D. P., Ecole Polytechnique, 1990-1991,
n. 17.
[11] Xu, C. J., Probl`eme de Dirichlet pour les operateurs de Hormander, Prepub-
lication de lEcole Polytechnique, Palaiseau, 1991.
[12] Xu, C. J. & Nie, L., Existence and regularity of solutions for a class of
degenerate parabolic equations, preprint and paper to appear.
9