Regularity for nonlinear equations involving square Hörmander operators

Regularity for nonlinear equations involving square Hörmander operators

Nonlinear Analysis, Theory, Methods Pergamon REGULARITY & Applications, Vol. 23, No. 1, pp. 49-73, 1994 Copyright 0 1994 Elsevier Science Ltd Pr...

1MB Sizes 1 Downloads 72 Views

Nonlinear

Analysis,

Theory,

Methods

Pergamon

REGULARITY

& Applications, Vol. 23, No. 1, pp. 49-73, 1994 Copyright 0 1994 Elsevier Science Ltd Printed in Great Britain. All rights reserved 0362-546X/94 $7.00+ .Ml

FOR NONLINEAR EQUATIONS INVOLVING HijRMANDER OPERATORS

SQUARE

Uco GIANAZZA Dipartimento di Matematica, Universita di Pavia, via Abbiategrasso 209, I-27100 Pavia, Italy (Received 12 July 1992; received for publication

16 June 1993)

Square HBrmander operators, homogeneous spaces, extraintegrability.

Key words and phrases:

1. INTRODUCTION IN 1973 Gehring

[l] showed that an Lq(&‘) function g is actually Lp integrable in a cube Q for some p > q provided a certain inequality relating to the maximal functions holds. In a successive paper Giaquinta and Modica [2] observed that the lemma is true only if g = 0 on a” - Q and proved a correct local version which is the main tool in their proof of an Lp estimate for a nonlinear elliptic system. In this way they gave an extension of a similar result that Meyers had obtained for an elliptic equation in divergence form [3]. At the same time a big interest arose in the development of a theory of singular integrals on homogeneous spaces; therefore, some useful results, such as covering lemmas and, above all, the Calderon-Zygmund lemma, were proved in the new setting (see [4]). Recently the structure of homogeneous space has been of great importance in studying Dirichlet forms as a tool in the description of the variational behaviour of possibly highly nonhomogeneous and nonisotropic bodies (see [5]). In this work we consider the problem - ;

Xi*x((u) + b(x, 24,X;(u)) = 0

in L!

(1.1)

i=l

u=o

on

where Xi satisfies a uniform Hormander condition k span the tangent space), Xi* is the formal adjoint Precisely we have W,

u, Xi(U)) = a,(x)u

asz

(1.2)

(Xi and its commutators up to a fixed order and b(x, s, p) is a first order nonlinear term. + “0% u, Xi(U))

with @o(X)> CY> 0 and f is a Caratheodory function which has at most a quadratic growth in Xi(u). Our aim is to study the regularity properties of weak solutions for equation (1.1). Under a smoothness assumption on L2, we will prove in theorem 5.1 that a weak solution u exists, which actually belongs to H,‘(a) n L”(a). To do that we adapt to our context an idea of Boccardo-Murat-Puel (see [6]), which is essentially based on the weak maximum principle. This result opens the way to finer estimates relative to our nonlinear equation. 49

50

U GIANAZZA

As a matter of fact, it is known that an intrinsic distance d(x, y) can be associated to our Xi in a natural way and this, together with Lebesgue measure, gives to @li” the structure of homogeneous space. Therefore, in theorem 4.1 we prove a homogeneous version of the Giaquinta-Modica lemma (see [2]) that will be a major tool in the proof of theorem 7.1 where we show that an L!’ estimate is still valid, much along the lines of Meyers’ result (see [3]). Finally in theorem 7.2 we demonstrate that our solutions are also locally Holder continuous. 2. FUNCTIONAL

Let us consider C”(al”) functions now define the vector fields

FRAMEWORK

a;,, where i = 1, . . . , m (m fixed) and I = 1, . . ., n. We can

Xi = ~ ai, ~ I=1 I and suppose that they satisfy a uniform (for further details see [7]). By making our vector fields (see [7-91) we denote

i=

l,...,m

Hormander condition as discussed in the Introduction use of the intrinsic distance which can be associated to

B(r, x) = (y: d(x, y) < r). Its main feature is that this ball has different dimensions along the directions defined by the Xi and their commutators. In [7] a Poincare inequality has been proved for intrinsic balls. Precisely we see that given a relatively compact open subset A of a”, there exists a constant C, > 0 and an integer k 2 1 such that for every x E CR”and every r > 0 with B(r, x) C A the following inequality holds

is



124 - iiJ2dx 5 C,r2 i where u E C”(A) let us consider

IxiCu)12

dX

(2.1)

.w, x)

B(r/k,x)

and ii is the average

of u on @r/k,

x). Given an open set Q such that Q E 61n 0

E(u, v; Q) = : i=l

Xi(U)Xi(V) dx +

uvdx !

.i Cl

R

which naturally gives rise to the norm [E(u, U; Q)]“2. We can then extend the classical Sobolev spaces to our setting. Therefore, by H’(SZ; E) and Hd(0; E) we denote the closure of C”(Q) and C,“(a) for the norm E. An easy extension of (2.1) to these new spaces is then proved. Let us now consider L = -t

We can associate

the following

bilinear

i=l

x;xi.

form on H’(Q;

(2.2) E) x Hd(sZ; E)

xi (“)xi(v) dX

(2.3)

Square

Hiirmander

operators

51

and the question arises about the existence of the Green function for the Dirichlet problem relative to our operator in CR”. Such a problem has a positive answer and the following estimates have been proved r2

r2 (2.4)

A, J&r, x)1 s GX(y) 5 *l JB(r, x)1 2-s lxj,

9 *..9 x,G”W

5

As ,&,

where G’(Y) is the Green function relative to L in 611”with singularity r I f?, R > 0 suitable and ]B(r, x)1 is the volume of the ball. As in the usual Dirichlet problem, G”(Y) is such that a@n(u, GX) = U(X)

(2.5)

x),

at x and Y E aB(r, x),

v U E C,“(@“).

Since a(u, v; B(r, x)) is coercive on Hd(B(r, x); E) it has been possible to define the Green function relative to L in B(r, x) with singularity at x (see [lo]). The above written estimates (2.4) and (2.5) are easily extended to the new setting. In particular we have

(2.6) for y E dB(qr, x), q E (0, qO], q0 < 1 suitable, A3 and A, constants which do not depend on x. It is also useful to consider the so-called regularized Green function relative to the ball B(r, x) (G,“(y)) as defined in [IO]. We refer to that work for the properties of (or to a”), GpX;BB(r,xj convergence

of G,“;.,,,,(G,“(y))

to G&,,,(GX(~)) 3. HOMOGENEOUS

Let us consider a set X and a function it satisfies the following properties: (i) &,Y) > 0 es x f Y; (ii) P(X, Y) = P(Y, x); (iii) there exists a positive

constant

as P + 0. SPACES

p: X x X --t 6l’. We say that p is a pseudodistance

K such that for every x,y, z E X

P(X, Y) 5 K]P(Y, z) + P(Z, We define

the pseudoball

if

41.

B(r, x) of ray r and centre x as the set B(r,x)

= lY:Y EX,P(X,Y)

< rl.

To simplify the notations we will use “distance” and “ball” instead of “pseudodistance” and “pseudoball”. Let us suppose that the set X considered above is a topological space and that the associated distance is such that: (i) the balls B(r, x) form a base of open neighbourhoods of x; (ii) there exists an N E N such that for every x E X and for every r > 0 there are no more than N points xi belonging to B(r, x) such that p(Xi, Xj) > r/2 (homogeneity property). We say that X is a homogeneous space and the two numbers N and K occurring in the definitions are its constants. In the following the letter C will stand for a generical constant, different in various contexts, depending only on N and K.

U. GIANAZZA

52

It is interesting to observe that the homogeneity property is automatically verified if a positive Borelian measure ,D defined on a o-algebra of subsets of X which contains the balls B(r, x) satisfies the following inequality 0 < p(B(r, x)) < C/@(r/2,

x)) < +=J

(3.1) for every x E X and r > 0. Therefore, we shall say that a topological space X is a homogeneous space if it is endowed with a distance p and a measure ,u satisfying those conditions. We now want to state some lemmas whose proofs have already been given in [4]. They are practically the homogeneous version of the analogous ones given in [ 1l] for the Euclidean case of 6li”. LEMMA 3.1. Let E 5 X be a bounded subset of a homogeneous

space X which is covered by the family (B@(x), x)). From this family we can then select a sequence of disjoint balls B(r(x,), Xi) so that the family (B(kr(Xi), Xi)) covers E, the constant k depending only on Nand K. Moreover, we have CP(E) 5 C P(B(‘?xJ, Xi)). Actually the last part concerning the inequality about the measure is new, but can be proved very easily thanks to (3.1) and the numerable additivity of Borelian measures. LEMMA 3.2. Let 0 c X be a bounded open set. There exists a sequence of balls B(p,, xi) such

that:

(i) Q = Ui &pi,

X;);

(ii) a point x E a cannot belong to more than M balls &pi, Xi); (iii) the balls B(hpi, Xi) intersect the complement of Q, CQ. The two constants M and h depend only on X. Proof. The proof is the same as in [4]. It is important to observe that in the above-mentioned work we have r(x) = & where K is the constant of the homogeneous

dist(x, CQ) space and k is defined in lemma 3.1

1 pi = kri = - dist(x,, CSJ) 2K h = 3K * hpi = 3 dist(xi, CQ) but there is no need to put exactly hp, = t dist(xi, CQ). It is enough that the balls intersect CQ; therefore, any choice of h that brings about this result, will do. Definition 3.1. We define the local maximal function function f as 1

@cx)= suPR
s BR

li;r,(x) of an L’(X, dp) nonnegative

f(t) Wt).

Square Hiirmander operators It

is obvious

53

that a.e.

f(x> 5 n;i,cx>

PROPOSITION3.1. Let us consider f E L’(X, dp), f 2 0. We have p((x E x: &-f(X) > CrJ)5 5o1 x f(t) d/At). I Proof.

The proof is as in [4]. There is no formal change due to the upper limit we now set

for R. PROPOSITION 3.2. Let f be a bounded support function, such that f 2 0. Let us further suppose that f E L’(X, dp) and that

There exists a sequence of balls B(pi, Xi) such that: (i) f(X) I cd a.e. on X - Ui B(pi, Xi) = X - S2; (3 l/IB(Pi, XJI SB(,,,,~ f(t)+(t) (iii) Ci IB(pj, xi)1 5 C/a jxf(t)

5 CW W);

(iv) a point x E X cannot belong to more than M balls B(p,, Xi). Proof. Again we work as in [4]. The fact that we consider the local maximum function instead of the global one requires that hp;
Vi.

This is always possible (see the remark in the proof of lemma 3.2) even if any set X can require a different choice of R, and (or) of h. 4. L!’ INTEGRABILITY

ON HOMOGENEOUS

SPACES

We can now come to theorem 4.1. For the sake of completeness, we will first state a global version of it which is of some interest but cannot be applied to our context. PROPOSITION 4.1. Let us consider a homogeneous space X and a function g with bounded support such that g: X + [0, +oo] and g E L*(X, dp) with q E (1, +a). Suppose that p(X) < +co, b E (1, +a~) and that

M&X) 5 NM, WI4

a.e. in X

(4.1)

where Mf(x) is the global maximal function for f. Then g is Lp integrable in X and satisfies (4.2)

for p E [q, q + c), where c is a positive constant which depends only on q, b and the constants of the homogeneous space.

U. GIANAZZA

54

Remark 4.1. An inequality such as (4.1) is commonly known as reverse Holder. It is interesting to observe that the global nature of this proposition is deeply linked to the use it makes of global maximal functions. Under the point of view of the applications, we are led to consider H{(X) functions, where the boundary value gives no problem, but this can be the biggest limitation. That is why we have to give a local version of the above result and that can be done only by putting an upper limit on R in the maximal function, as done in the preceding paragraph. With the new idea and a finer but rather technical subdivision argument, we can overcome the main difficulties and prove theorem 4.1. Proof. See [12]. Let us finally state the main theorem of this section. space X and two nonnegative functions g and f such that g E Lq(x), 4 > 1 andf E L’(X), r > q. Fix a ball B, = B(R*, x0) in Xand suppose that for every x E B, and R < i dist(x, SS,) the following estimate holds

THEOREM 4.1.Let us consider a homogeneous

(4.3) with b > 1. There exists a constant 8, = &(q, r, K, N) such that if 6 < &,, then g E L$,,(B,) for P E h, q +

~1and (,&,,

{.,gpdp)l’p 5 C/(\B(2;,x),

i,,,gqdp)L’9

+ (,,(,A,.,

.i,,,fp4’p]

where C and E are positive constants depending only on b, 8, q, r, K, N. Proof.

Let us consider the ball B, such that B, = B(3R, x,)

and let us define c, = B(R, x0) C, = (x E X: 2rekR < dist(x, dB,) 5 22-kR) It is clear that B,(3R,

X0) =

u ck. kz0

k = 1,2...

C4e4)

Square

HGrmander

operators

55

Let us now set (Yk

=

(KILL)

K, a proper

g(x)

(g’)B, =

G(x)= (g”);;

constant

j$ i B, gq dp

on ck

f(x)

F(x)= (gQ);( S(x)

=

F<“>

on ck

Olk E(h, t) = (x Fix t E [ 1, +a)

E

B,: h > t).

and choose s = K2 t such that for convenience

Since

we can apply lemma 3.2 on each Gk. If the constant h,p,

and the upper bound

k+5

= k

+

4

diNxi, c&d

R,, in the definition &,k

we obtain

a sequence

that is

k+4R

= -

k+32k

-

in lemma 3.2 is such that

k+5i? 5

k+4

of the local maximal

of balls Bh (pi, Xi) such that:

(i) BL (Pi 3Xi> C ck ; (ii) G 5 (Y,$ a.e. in Gk - UjBi(pi,xi); (iii) l/IBL[ IBj, Gq(t) d&t) I K&/$)‘. If we divide by CY$we obtain

hk which appears

2

function

is given as

U. GIANAZZA

56

Then

IE(s,s)

- uj,kBjkI

= 0 and, hence,

where we take into account

the doubling

property

and we put

K4 = K,C. Let us now recall that !& is such that MGQ(X) > sqcYq k Therefore,

for each x E fik there exists a ball @(r,

vx

E

Qk.

x) such that (4.6)

with r < ROk = (k + 4)/(k + 3)R/2k. Denote by Bi the ball which has the same centre

as i$ and F = $1. Of course

B{ c c, u c, u c,

(4.7)

k+2 Bi

c

u

(4.7’)

cj

i=k-1

and also (4.8) i=O k+4 Lg

c

u

(4.8’)

cj.

i=k-1

Thanks

to (4.7) we have @ = (Bi n ck) u @j, n ck+r) u (@ n &+2).

Therefore,

If we divide

(4.6) by CY~we have

I

(4.9)

51

Square HGrmander operators

Let us now consider

our inequality

+

.i elBIL 4/3R

and take two balls & and B’,. To simplify (4.10) becomes

Dividing

qdp

+

ak

I BnCk

i BnCk+i

z+

+

sQd,z + 2 =+I

If we also take into account

If we reason

Taking

as before

into account

5q

-t(B*l q-1

I

let us simply

write B and B*. Then

side

[BnCk+23p]

Rq&+4[BnCk+2Sq‘b]

+,[BS’q4-;-

s BnCk+l

Bnc,,

LI

(4.10)

dpu.

the notations

by (gq)B, 01: we get on the left-hand

,A -li

gq

the right-hand

we finally

the definition

s

gdp

side, we have

get

of s, we have

+ (B*(‘-

llfl’B*~Qlp)i’q

($i’q([B*SqdP>“q

+

lB*1’-“‘.

B*

It is also easy to see that in this new setting

[B*sdp ([,*CFqdlr)Lb

we have as in [2]

5 i’ris,rlnB*Sdp

5 [jEC5,CloB*5qd,u

5 2t\B*)

+ flB*l + tqiB*~]l’*jB*I’-l’q

t’-’ S CFq dp

+

.E(‘S,t)nB*

lB*ll-l’q (;y’q( I’..6’ dp]‘q

I

2tlB*I

+

;tl-q

6’ dp

j w2.r)

nB*

U.

58

G~ANAZZA

if 0 < b. Therefore, n

5 -t(P q-1

cr

6 dp + t1-q

e + _t’-4 b

5q dp I E(S,f)fIB’

EG,t) oB*

.i EG,1) nfs*

Sq b.

(4.11)

Let us now define Dk = u@. j Since Dk is bounded

we can apply the usual covering

lemma

and obtain (4.12)

where the Bici) are pairwise

disjoint.

Combining

ID,1 5 K, c I@fi’\ I KS? i

(4.11) and (4.12) we get

‘i4 i=k-I I>

x

t-l i

Sdfi

+ t-q


! E(5.t) nci

If k = 0 the sum has obviously to be corrected observe that Dk 2 a,. Therefore, 1%

c

k

Moreover, ; Thanks

5

when we sum over k, the integrals

ID,\

5

;&(q-

1) t-’ i

c k

to the estimates

!

t-l i

Sdp

0 I ws*s)

E(S,t)

.

(4.8). Let us now

at most six times each. Then

!Yqd,u + %f-q

+ t-q i E(3.t)

- 1)

+ trq

Tqdp
.i E(S,t)

1

ID,\.

over Ci are counted

sd,u
Sq&

to (4.5) we have

x

with

+ Bf-q i m,r)nc,

according

‘;K,(q

or

Fqdfi

+

;t-q .i,,,,,

‘”

dpj

Square

Hbrmander

operators

59

On the other hand $j4dp =

&

66’-’ s E(S,1)-E(S,s)

E(s,t)-HS,s)

sq-1

5

6 diu

i E(S,~)-E(S,s)

K 7 btq-’

< -

6 !

dp

E(S,r)

with

Therefore,

if K

(& + K7)b>

=

o

1 -Kg0

8

that is, if K,d<

1

then sqdp

I KS tq-’

.iE(S,t) From

(

here on we can reason 5. EXISTENCE

c E(S,t)

as in [2] and we come to the result.

OF WEAK

SOLUTIONS

QUADRATIC

FOR

Let us now come back to our nonlinear - ~ ~Xj(U)

A QUASILINEAR

GROWTH

IN THE

equation.

+ a,(X)u

+ f(X,

We consider U,

EQUATION

WITH

X, TERM

Xi(U)) = 0

the following in D

i=l

on

u=o

and we suppose

%(4

E J?fi)

we take a Caratheodory if(x,.sp)I

function

5 Co + b(bl)l#

with C,, > 0 and b: CR++ CR+monotone

(5.1) a.e. in Sz.

go(x) z cyo > 0 For f(x. s,p)

an

that

(5.2)

which satisfies a.e. in Q

increasing.

vsE63,vpEw

(5.3)

U. GI.~NAzzA

60

We see that f is bounded relative to the variable x and has at most a quadratic growth in the Xi term. We will prove the existence of at least a solution u E Hd(fiz; E) n L”(Q). We are, therefore, dealing with a weak solution in a sense we will explain in the following. The hypotheses are almost minimal with regard to the nonlinear term, whereas we must assume Sz = B(R, x0) as we rely upon a maximum principle which is valid only on balls: this is a typical situation when working with subelliptic operators as we are doing now. A particular simple case which satisfies the hypotheses given above is - f

x,*X,(u)

+ a&u

+ g(u) i

i=l

lx;(u)12

= h

i=l

with h E L”, g: @ --t CFtcontinuous and a,(x) such as required in (5.1). First of all we require a preliminary lemma, a sort of weak maximum principle equations. LEMMA 5.1.Let u E Hi(Q

E) be a weak solution -i~lx*xitu)

with a = B(R, x0), R suitable,

for nonlinear

of +

aOu

=

-f

f E L”(Q) and a,(x) > o. > 0 a.e. in Q. Then

where M = We now state our theorem

Ilf IL”(n) *

that is just an adapted

version

of the analogous

result given in [6].

THEOREM 5.1.Let Q be a ball B(R, x0) and let us consider a Caratheodory function f(x, S, p) defined in Q x 6l x 61”. Actually this means that it verifies the following hypotheses: (i) v (s, p) E & x @I” the function x + f(x, s, p) is measurable; (ii) v a.e. x E Q the function (s, p) + f(x, S, p) is continuous. We further suppose that f(x, s, p) satisfies (5.3). Finally we consider a coefficient a, defined in 0 such that (5.1) and (5.2) are valid. Under the previous hypotheses we claim that there exists a solution u such that u E Hd(Q; E) n L”(cI) - ~ ~Xi(U)

+ aO(X)u + f(X, 24,Xi(U)) = 0

(5.4) in D’(Q).

(5.5)

i=l

Remark 5.1. It is interesting to observe that as u E Hd n L”, f belongs to L’ thanks hypothesis made above and each term of our equation is therefore a distribution. Remark function

5.2. Generally, when f is a Caratheodory that assigns f(Z, u(2), Xi(U(X))) to Z E Q.

function,

to the

by f(x, u, Xi(U)) we mean

the

61

Square Hiirmander operators

Proof of lemma. As u E H&2; E), given a positive constant C, the function v = (u - C)’ is admissible as a test function. Then Xi(U)Xi((U - C)‘) dX +

iR

i cl

Thanks to the coercivity of the operator,

.r R f,

I&(@ - C>‘)12~+

i

a,(x)u(u

- C)’ dx = -

f(u - C)’ dx.

at least if R is not too big, we have

Qol(U - a+12~

5

1(-f

- a,C)(u

- C)’ dx.

R

R

If we choose C = M/a,

In

we have on the right-hand side

s (-f

- M)(u

- C)’ dx I 0.

n

Therefore,

each of the integrals on the left has to be zero and we conclude (U-c)+=o*U
Proof of theorem. We will proceed in the following proof by steps. of our problem. We consider the function f, defined by

Step 1. Approximation

f fc = 1 + Elf/

(5.6)

where E belongs to a sequence that converges to zero but must be considered fixed for the moment. As If,] < (fl, our new function is still a Caratheodory one and verifies growth condition (5.3). Moreover, we have (5.7) If we consider the approximated

problem U, E H&2; E)

- F Xxi(&) i=l

+ aou, + f,(x, 4,

xi(Uc))

(5.8) =

0

in LO’

(5.8’)

we can reason as in [6] and applying the Schauder fixed point theorem, we can be sure that (5.8), (5.8’) admits at least a solution u,. What is most important, as f, belongs to L”(O) we can use any Hd(!22;E) function as a test function in the approximated problem. Moreover, we can apply lemma 5.1 because of (5.7) and we obtain that any solution of (5.8), (5.8’) belongs to L”(Q) and satisfies the following inequality (5.9)

62

U.

GIANAZZA

Remark 5.3. If we had a maximum

principle valid on a more general class of domain result with weaker regularity assumptions on aQ.

could prove an existence

Step 2. Estimate in L”(Q). We now want to refine estimate does not depend

on E any more.

(5.9) and give a better version we want to show that

Precisely,

Q, we

that

(5.10) As a matter

of fact we will prove that

co

u, < -

a.e. in Q.

(5.11)

010

The symmetric relation U, > -Co/~, is given in the same way and together marized in (5.10). To simplify the notations we define 2, = 2.4,- -

they are sum-

.

010

This function

solves

xi*X.(

- f

I

-fEk,

zE) + aOzE =

XA4J) - a0

i=l

Furthermore,

s

in YB’.

(5.12)

010

if we define (5.13)

where b is the function

that appears

in (5.3), we have

Wl4) 5 c, as b is nondecreasing.

It easily follows

that

-h(Uc,Xi(uc))- a02 5 CO+ b(lu,l) i lXi(Q12 - a02 i=l

s c.5 f Ixi(utd12 i=l = c&E, lXkE)12 Let us now consider

the test function v = exp(t,lz:12)z:

where: (i) z,’ = maxtO, zEl; (ii) t, = C,2/2 and let us define e, =

exp(t,lz:l’).

a.e. in a.

(5.14)

Square

Hiirmander

63

operators

We observe that Z: E L”(a) thanks to the estimate proved at the end of step 1 and Z: E Hd(Q; E); in fact z,’ is zero on &2 as Z, = -Co/o,, a.e. on %2. Moreover, as Xi(U) = e,Xi(Z,‘) u also belongs

+ 2tEe,jZ,‘12Xi($)

to Hd(G; E). We can then use it in equations

Xi(Z,)Xi(Z,+)dX + 2t,

e, i i .n

at least if R is not too big, we can estimate

I ReeiC, lxi(Z:)12dX+ 2tc e,lzi12 ! ,Q

Hfc

69 uc

5-

9

.n

xi Cu.&)+

aO

Ixi(Z:)l k + Qo 46 I2 i .a

f i=l

co 1

e,z,C dx.

by (5.14) we have 7

I eE i

.I fl

&,‘I” f

C,

,,fl

If we apply the Young

jl

dX +

Ixi(Z:)12

C$J

i=l

,I fi

4 5-

n

IXi(Z:)I'dx + 2, I

i=l

, a e, i

1

eelz,f12dx

R

2 IXi(Z,>12e,Z: d-X.

(5.15)

i=l

inequality

we further

lXAz312 do + %

i .a

get

&,+I2 f

&,+I2 b

+ a0 i ,fi



1 ~ R e, j, 2 1

Ixi(Z,‘>12dx

i=l

Ixi(Z,‘)12 dx + i

e,ClzE+12 5

IR

IXi(Z,+)J2dX.

i=l

to the choice for t,, 2 t,

we

=

+

have 1” 2 !n

In eE J,

the

QO

0

Thanks

eEuoz,z,+ dx

n

n

,?I

On the other hand,

s

i=l

Thanks to the coercivity of our operator, left-hand term and we obtain

, I

(5.12) and we have

eh,‘12ii xi(Z.E)xi(Z:) dX +

i .R

i=l

also

Ixik:)12 d.Y+ i

cl

As e, 2 1 it must be Z: = 0. Therefore,

e,lz,+I’ E Ixi(zZ>l”~ i= 1 U, I Co/~,

.

+

eE(zE+12

010 s

cl

dx

5

0.

64

U.

GIAN~~ZA

Step 3. Estimate in Hd(CJ; E). If we define

thanks

to the estimate

proved

in the previous

step, we have

b(lu,l) as b is a growing

function.

Let us consider

5 C,

the function +.

u, = exp(tu,2h4, If we define

E, = exp(tuz) then we have + 2t,uz exp{tuz)Xi(u,)

X~(U,) = exp(tu~jXi(u,)

= EEXi(UE) + 2t,u,2E,Xi(u,).

Since u, E H&S; E) fl L”(Q) we can conclude that u, E Hd(Cl; E) and can be used as a test function in our approximated equation (5.8). We have m a0E, uf dx u.ZEe F IXi(U.&12 dx + Ee iC, Ixi(UJ* b + 2t i=l I cl ia .i a

Thanks

to the estimate

on a, we have

m

EciClIXi(u.s)I’ do + 2t

.i cl

=--

R

s

E.zu.Z f Ixi(uc)12~+ QO i=l

E&X

n

fe(~3UC Xi (ucJ)Ec UCdo-

(5.16)

3

i n On the other hand -

fe

txY

u.5 9 xi

(“.A)Ec

ue

dX

cl

i=l

f i=l

IXi(U,)(*dX

(5.17)

Square

HBrmander

65

operators

where IQz(is the measure of Q. If we rewrite (5.16) taking into account estimate (5.17), we have 1 E, : 2 s D i=l

IXi(u,)?~

Ku,2 ,! IXi(u,)l*~

+ f n

+ ~0

i=l

E,U,2 dX n

If we recall that E, 2 1 we have (5.18) and, therefore,

U, E #(a;

E) fl L”(0).

Step 4. Strong convergence in Ho@; E). As the functions u, are bounded in H,‘(Q; E), we can extract a subsequence, which we still denote by uE, such that UE ‘U

weakly in H&Q E)

% + 24

strongly in L*(Q)

u, + u

a.e. in 52.

What we want to show is that actually U, + u strongly in Hd(Q; E). We define ii, = u, - u. This function belongs to Hd(Q; E) n L”(L2) and is a solution for the equation - ~ xi”Xi(ii,) + a(jE, = -f,(X, U,,Xi(U,)) + E xi*Xi(U) - UOU. i=l

i=l

In a similar way as before we define UE= exp(tii,2Jii, where t = 2C,2 and as usual we put E, = exp(t@). It is easy to see that v, E H,‘(Q; E) and can, therefore,

in

E, i

IX&)l*ti

+ 2t

5

Eciiz i .i n

i=l

i -

R

i=l

be used as a test function. We have

IXi(U,)1*dx + (~0 E,tiif ti ia

Et iil xi(~tJK(U)do If,(x, u.s> K(Ue))IEegc ia 2t

n

a, uE, ii, dx

E, ~ Xi(ii,)Xi(U)ii,2 dx i=l

n

(5.19)

66

GIANAZZA

U

If we recall that Ixi(u.5>Iz

=

U + &)I2 = IX;(U) + X;(U,)l2 12JXi(U)12

Ixi(

+ 2lX;(ii,)l2

we get from (5.19) n R

E, 2 ~X;(Z&))~dx + 2t E,U,2 ~ (X;(ii,)l’dx i=l i=l i .Q 1 5

.i[cl

CO + 2C,

f

lX;(~)1~ E,Izi,( dx + ;

E,i$ dX !a

.i a

1

i=l

+ ~yg

E, t

bM%)i2~

i=l

n + i

E,(2C,)2ii,z i ,e

i

IX;(ii,)12dx

-

E, f !

i=l

n

Xi(%)X; (u) dx

i=l

m -

2t

E, C X;(U,)X;(U)~,2 dx R

Taking

into account

a, uE, ii, dx.

(5.20)

n

i=l

the definition

of t we can rewrite

(5.20) in this way

(5.21) As Ei, + 0 strongly in L2(sZ) and also X;(U,) --t 0 at least weakly in L2(n), it is easy to see that all the terms in the right-hand side converge to zero and the same must happen to the other side. As E, 2 1 we have, therefore, that ii, + 0 strongly in H,‘(Qz; E). Step 5. Passage to the limit. As u, converges extracting a new subsequence) Xi(&)

--f

to u strongly

Xi(U)

in HJ(Oz; E) we have (eventually

a.e. in Q.

As in [6] we claim that: (i) f,(X, u,, xi(Uc)) * .I%, U, xi(u)) a.e. in Q (ii) f8(U,, Xi(U,)) + f(U, Xi(u)) strongly in L’(a). Therefore, going to the limit in each term of equation (5.8), we have that the limit function belongs to Hd(Q; E) fl L”(Q) and satisfies equation (5.5). Remark 5.4. It is interesting that we have proved by providing a function which satisfies

the existence

u

of a Hd(Q; E) fl L”(Q) solution

(5.22) In [6] it is proved

that all limited

solutions

satisfy

(5.22).

SquareH(irmanderoperators

67

6. A CACCIOPPOLI INEQUALITY FOR A NONLINEAR PROBLEM We can now prove a Caccioppoli inequality for our problem: it will basically be for the regularity estimates of the next paragraph. Just like the previous paragraph, we have &2= &R, x0). As we work with H,‘(fiz; E) n L”(a) solutions, we can use exponential cut-off functions, so that we can take into account a quadratic growth in the Xi terms. Otherwise a milder condition should be considered.

PROPOSITION6.1. Consider problem (5.4), (5.5) and suppose that (5.1)-(5.3) u E HJ(nz; E) fl L”(a) be a solution. We have

I’ [

+

bi2 G’&,) dx + 1

B(qr,xo)

I ,B(%,),

hold.

Let

su~~(qr,x& - k/*

Iu - kl*dx + C2ra

B(r,xo)-B(w,xo)

(6.1)

where 0 < q 5 i& with Go E (0, l), 0 < r < R, k E 03, C, and C2 are constants which depend on 4, Il&-~~~9k and the quantities that appear in the nonlinear term estimate.

Proof. Firstly we observe that to prove (6.1) it is enough to prove the same inequality with G$2r,x0j replaced by GXo. We use as a test function

u = (u - k) e”“-kt2cp2((G~)-1)Gj z E B(qr, x0) where G,” is the regularized Green function relative to a” and the point Z, t will be fixed in the following and ~1E C,“(B) with

2 @(t) s (C, _ 1y

A and Co suitable. We have

E Xj(u)Xj((u

- k) et’u-k’2cp2GJ dx +

,i Ri=l

+

n

f(x, u, Xj(u))(u

a,u(u - k) er’u-k’2p2G~ dx 0

- k) et~U-k~Zp2G~ dx = 0.

68

U. GIANAZZA

If we put Et = e tiu-k’2 for short as we did in the previous

T lXi(u)12Et~2G;

cl

(U - k)E, i

+ n

+

z

(U - k)‘E,

dx + 2t

fl i=l

we get

(Xi(U)12~‘GpZdX

i=l

Xi(U)Xi(p*)G,ZdX

a, u(u - k)E,p2G; d_x

+

i=l

a

(U - k)Ef f n

paragraph,

Xi(U)Xi(G~)~*dX

i=l

f(xv u, Xi(U>)(U- ~)E~P*G,Z k

=R

and also

Et f IXi(u)~*~*G;dx + 2t .i n

i=l

12)~ IXi(U)12~2G,ZdX i=l

s

Et ~ Xi((U - k)2~2)Xi(G,“)~

n

n

s

(U - k)E,~2G~ dx + ~

-aok

1

i=l

E~(u -

0

--

F

E,(u -

(u - /c)~E,~~~G,Zdx + ;

+ 010

I

i .n

k)* ~

E, ~ Xi((U - k)*)Xi((P2)GpZ dw -

i! .i n

i=l

S[ + ~ Co

Cl

n

i

a

term, putting

f(xv ~9 Xi(U))(U - k)Etp2G;

dx-

+

D zif,

Et E IXi(U)12p2GpZti. i=l

for short and applying

I

i

n

Cl = b(ll~&~,)

IU - klE~~*G,Zdx

IXi(U)12

i=l

C,lu - kjE,p*G;dx

I

dx

n

Recalling the estimate on the nonlinear the Young inequality we have

5

Xi(~")Xi(G,Z)

i=l

cl

IXi
Square

t = CF/2 and apply the Young

If we choose 1 i sa

Hiirmander

E, ~ IXi(U)12~2GpZdX + ~ i=l +

(u

010

1”

k)2E,cp2G;dx

-

(U - k)E,lo’G,Zdx

1

+ ;

+ ~

E,(u - k)2 ~ Xi(lo’)Xi(G~)dx in

(u - k)2E,p2G;

\u - k~E,q~~G;dx

+ Co .i n E,p2G,Z dx

n

1

E,(u - k)’ ~ Xi(~2)Xi(G,Z)dx

2 ia +

i=l

dx + C, k2

0 f-

IX,(~)(~cp~GpZd_x

E1 ~ Xi((U - k)‘~‘)Xi(GpZ)dX i=l

E, ~ Xi((U - k)2)Xi(~2)G~dX i=l

I ?

we find

i=l

iR

2

once more,

E,(u - k12 f

sD

I -aok

--

inequality

.r0

D

.r

operators

Iu

co

- ~

Et E Xi((U - k)2)Xi(p2)G,” dx

i=l

i=l

n

k[Etq2G;dx.

-

fl

Therefore, 1 2 1n

Ed ~ IXi(U)12~2GpZdx + ~ i=I 3 + 4ao 1

(U - k)‘Etp2G;dx

2

i 0

+

T

+ ;

Et i i ,a

Xi(p’)Xi(G,Z)dx

-

Xi((U - k)2~2)Xi(G,“)dw

i=l

- i

Et ~ Xi((U - k)2)Xi(~2)GpZ~

i=l

(u

lXi(U)12v2GpZdx

i=l

i cl

E~(u - k)2 f

I-

E~(u - k)2 t n

sn

k)2Etp2G;dx

+ C2k2

i=l

Etp2G;dx

+ C,

.rn

in

E,cp’G; dx. in

Finally,

sn

Et E [Xi(U)(2p2Gidx i=l + c:

Et f In

5

+

R c4

IXi(U)I’(U - k)2p2G,ZdX +

Et i .i n

i=l

Xi(Cp’)Xi(G,Z) ti

i=l

Et (P’G; dx cl

d.x

sn

E~(u - k)2 f

I

(u - k)2Etp2G;

+ (~0

-

Et i sn

i=l

Xi((U - k)2p2)Xi(GpZ) ti

i=l

Xi((U - k)2)Xi(p2)GidX

U. GIANAZZA

70

having

put c,

If we take into account

and taking

= 2(C#

that u E L”(Q),

into account

-t C,).

putting

CS = inf,(u

- k\

C, = mint1

+ CfCi,

01~)

that 1 5 Et 5 C,

where C, stands

G

for its upper

bound,

we conclude

) i( itlIXi(U>12+ N2‘

p2G; dx +

(u -

R

r~

Xi((U - k)2~2)Xi(G,") ~

, I2i=l

s I c,

(U - k)’ u cl

+ c8

~ Xi(~2)Xi(G~) i=l

m c Xi((u

dx -

- @2)Xi((02)G,” dx

! Cli=l

1

p2G; dx. Ia

If we recall that (U - /Q2 > ;

- !$

we find

c9

f i=l

i( n

I&(u)12

I c,

+

u’)v’C~

(U - k)’ f irtJfi

CIX

+

in

Xi((U - k)2~2)Xi(G,“) dx

Fl

Xi(p2)Xi(GpZ) dx -

i=l

f

Xi((U - k)*)Xi(~“)G,” dx

0 i=l

1

I\ +

cp”G; dx

Cl0

n

where C9 = C6/4 and Cl0 takes into account ’

c9

the estimate

given above.

Then

m

.i( a

I

iCl

2

IxiCu)I

+

u*

P’G;

CLY +

,Bc;,

zj,

B(p,~)

(u

-

W2

CLX

)

c,

n

(U

s

-

‘:’ + kj21p’It Ixi(G,Z)I’(~33 i=l

&)k+

Clo[Qr2G;~.

71

Square Hiirmander operators

Passing

to the limit as p --t O+ we obtain

c9

n

jt,

+

Ixi(u)12

u2

>

i(

I c,

.in

p2Gz dx + (u - /c)~(z)

$$

; Ixi(Gz)/2 i=l

(u - k)‘Iv’I

Choosing A and C, such that: (i) (x: GZ 1 l/A) > B(2qr, z) (ii) (x: Gz 2 l/C,,A) c B((l - q)r,z) (possibly for small r) and taking into account derivatives, we obtain

s(

+ &++

the estimates

j, lxitu)12+ u2 GZ dx

B(v, z)

>

C,,sflGz~.

on the Green

function

and its

+ (u - /c)~(z)

>

s &j

~/ B(r,z)-B(qr,z)(’ - k)2 dx + c12ra

with 01 suitable (we use the duplication property of measures of intrinsic only on the dimension of the space and the Hormander constant k). Finally, taking the supremum for z E B(qr, x0) we get

balls and LYdepends

m C

i=l

Ixi(U)I’

+

s

u2

>

G”‘~-x

(24-

+

SUPB(qr,x,,)(U

-

k12

Jq2dx + C12P.

B(r,q+BWvW

7. REGULARITY

FOR THE

WEAK

We can now come to the estimate

SOLUTION

for the solution

OF A NONLINEAR

to our nonlinear

EQUATION

problem

in Q = B(R, x0).

THEOREM

7.1. Let u be a Hd(Q; E) II L”(Q) solution of (5.5) and suppose that (5.1)-(5.3) are satisfied. Then there exists an exponent p > 2 such that u E yt;p(sZ; E) and the following estimate holds for B, c BZr c Sz 1

(f (

p’2

lU12+ ~ IXi(U)12 )

B,

i=l

&)l”

If we define

we also have the following

theorem.

s

KI[(,~Br(lulz + j,

lxi(u)12) &y”]

+

K2*

U.

12

THEOREM 7.2. Under

the same hypothesis V(r) I

GIANAZZA

as above, r

0

Cl

-

RCI

with r 5 (q/2)Ro < R, 5 (q/2)R,

y

u is locally

V(R,) + CzR,

q as in the previous

Holder

continuous

and

6

paragraph

and y, 6 E (0, 1).

Proof of theorem 7.1. It is easy to see that the metric topology induced by the intrinsic distance defined in Section 1 is equivalent to the natural Euclidean topology of Gl”. Moreover, the Lebesgue measure satisfies the doubling condition for our intrinsic balls. We can then see that the space a” wth the intrinsic distance acquires the structure of homogeneous space, as discussed in Section 2. Let us now come back to our Caccioppoli inequality

s( BG7r,xo)

i

f i=l

k(#

+

Gg2r,xoj

u2

b

+

su~s(~‘.x,,)(u

- Q2

>

,B&,l,

(u -

k)‘dx + C2ra.

~(~,~o)-~(qr,xO)

If we neglect the second term on the left-hand Green function, we have

side and take into account

the estimate

for the

n

i B(V,%)

(

i~llxioi2+~2)~r~~xii,i,,smi,r,,(~~k)2dX+C~’B(~~~‘r~.

If we recall how ra was calculated

we find

Now putting q = a, k the average of u on B(qr, x,), using the Sobolev inequality q* such that l/q* + l/s = 1 where s is the Sobolev exponent, we have

If C, = 0 we could directly apply theorem 4.1 and we have the result. the case, we can reason as in [2] and again we obtain our inequality.

Otherwise,

and assuming

as is usually

Proof of theorem 7.2. We use as a test function u = (u _ 0) er(l~-“‘*G;rp where ii is the average of u on B(qr, x0), p < r, Gj = G&f,,zj,p is the regularized Green function of B(2r, z), rp is the potential of B(2r, z) with respect to B(tr, z) and I is as in proposition 6.1.

Square

Hormander

operators

73

We can reason as in theorem 1 of [lo], to which we refer. The initial difference due to the presence of the exponential term can be adjusted as in the proof of proposition 6.1, thanks to the fact that u E H~(Qz; E) n L”(S2). REFERENCES of the partial derivatives of a quasiconformal 1. GEHRINC F. W., The Lp integrability 265-271 (1973). 2. GIAQUINTA M. & MODICA G., Regularity results for some classes of higher order

mapping, nonlinear

Acta Math. 130, elliptic

systems,

J.

Math. 311, 145-169 (1979). 3. MEYERS N. G., An Lp estimate for the gradient of solutions of second order elliptic divergence equations, Ann. Scu. norm. sup. III XVII, 189-205 (1963). 4. COIFMAN R. R. & WEISS G., Analyse harmonique non-commutative sur certaines espaces homogenes, Lecture Notes in Mathematics, Vol. 242. Springer, Berlin (1971). principle for Dirichlet forms on discontinuous media (preprint). 5. BIROLI M. & Mosco U., A Saint-Venant 6. BOCCARDO L., MURAT F. & PUEL J. P., Existence de solutions faibles pour des equations elliptique quasi-lineares a croissance quadratique, Nonlinear Partial Differential Equations and their Applications, College de France Seminar, (Edited by H. BREZIS and J. L. LIONS), Vol. IV, Research Notes in Mathematics. Pitman, London (1983). I. JERISON D. & SANCHEZ-CALLE A., Subelliptic second order differential operators, in Lecture Notes in Mathematics, Vol. 1277, pp. 46-77. Springer, Berlin (1987). 8. NAGEL A., STEIN E. M. & WAINGER S., Balls and metrics defined by vector fields: basic properties, ActaMath. 137,

247-320 (1986). 9. SANCHEZ-CALLE A., Fundamental

solutions

and geometry

of the sum of squares

of vector fields, Invent. Math. 78,

143-160 (1984). 10. BIROLI M., Local

properties

of solutions

to equations

involving

Potential Theory, Nagoya, August (1990). 11. STEIN E. M., Singular Integrals and Differentiability Princeton (1970). 12. GIANAZZA U., The Lp integrability

on homogeneous

square

Hormander

Properties of Functions.

spaces,

operators, Princeton

Proc. Int. Conf. University

Rend. Ist. Lomb. A 126, 82-92 (1992).

Press,