Nonlinear Analysis 54 (2003) 165 – 186
www.elsevier.com/locate/na
Hardy-type inequalities and Pohozaev-type identities for a class of p-degenerate subelliptic operators and applications Huiqing Zhang∗ , Pengcheng Niu Department of Applied Mathematics, Northwestern Polytechnical University, Xi’an, Shaanxi 710072, People’s Republic of China Received 15 June 2002; accepted 1 September 2002
Abstract The purpose of this paper is to consider a class of p-degenerate subelliptic operators Lp constructed by generalizing Greiner’s vector 3elds. Their fundamental solutions at the origin are established with the aid of the properties of radial functions. A Picone-type identity and a Hardy-type inequality with respect to vector 3elds are proved. Some Pohozaev-type identities and applications to nonlinear equations are given. Finally, a Carleman-type estimate and uniqueness of the operator L2 are discussed. ? 2003 Elsevier Science Ltd. All rights reserved. Keywords: Fundamental solution; Picone identity; Hardy inequality; Pohozaev identity; Nonexistence; Carleman estimate; Uniqueness
1. Introduction In this paper, we consider a class of degenerate subelliptic operators Lp u = divL (|∇L u|p−2 ∇L u);
(1.1)
where p ¿ 1, Xj = @=@xj + 2kyj |z|2k−2 @=@t, Yj = @=@yj − 2kxj |z|2k−2 @=@t, zj = xj + √ n −1yj ∈ C, j = 1; : : : ; n, t ∈ R, ∇L = (X1 ; : : : ; Xn ; Y1 ; : : : ; Yn ), divL (u1 ; : : : ; u2n ) = j=1 (Xj uj + Yj un+j ), k ¿ 1. When p = 2 and k = 1, Lp becomes the sub-Laplacian >H n This paper was supported by the Natural Science Foundation of Shaanxi Province and the National Natural Science Foundation of China. ∗
Corresponding author. E-mail address:
[email protected] (H. Zhang).
0362-546X/03/$ - see front matter ? 2003 Elsevier Science Ltd. All rights reserved. doi:10.1016/S0362-546X(03)00062-2
166
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
on the Heisenberg group H n , see [5]. If p = 2 and k = 2; 3; : : :, Lp is a Greiner operator (see [9]). Operator (1.1) does not possess the translation invariance for k ¿ 1. We refer to [2,11] and the references therein. Since HGormander’s famous foundational paper [12], the study of partial diHerential operators constructed from noncommutative vector 3elds has received great attention. Folland [5] established an explicit fundamental solution for the sub-Laplacian >H n . Garofalo and Lanconelli [7] discussed the unique continuation of SchrGoding-type operators with a suitable potential and proved a Hardy-type inequality for p = 2. A Pohozaev-type identity for the sub-Laplacian was proved in [8] and used to nonexistence of semilinear sub-Laplace equations. Capogna et al. [3] studied the p-degenerate m subelliptic operator j=1 Xj∗ (|Xj u|p−2 Xj u), where Xj are smooth vector 3elds satisfying HGormander’s hypoellipticity condition. We note that the vector 3elds X1 ; : : : ; Xn ; Y1 ; : : : ; Yn in (1.1) do not satisfy HGormander’s condition for k ¿ 1 (see [2,11]). Lp in (1.1) is the Euler-Lagrange equation associated to the functional
p
|∇L u| =
n
(|Xj u|2 + |Yj u|2 )p=2 ;
p¿1
j=1
for a function u satisfying u; ∇L u ∈ Lp . It is the purpose of this paper to present some important properties of Lp , including the fundamental solution with singularity at the origin; Hardy-type inequalities; Pohozaev-type identities; Carleman-type estimates and applications. We now state some preliminaries about the family of vector 3elds X1 ; : : : ; Xn ; Y1 ; : : : ; Yn and the operator Lp in (1.1). Let
2ky1 |z|2k−2
A =
2k−2 2kyn |z| 2k−2 −2kx1 |z| .. . 2k−2 −2kx n |z| .. .
IR2n
2ky1 |z|2k−2 · · · 2kyn |z|2k−2 −2kx1 |z|2k−2 · · · −2kx n |z|2k−2 IR2n
= 2ky|z|2k−2
−2kx|z|2k−2
2ky|z|2k−2
−2kx|z|2k−2 ; 2 4k−2 4k |z|
4k 2 |z|4k−2
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
167
where A is a (2n + 1) × (2n + 1) matrix, IR2n denotes the identity matrix in R2n ,
IR2n B= 2ky1 |z|2k−2 · · · 2kyn |z|2k−2 −2kx1 |z|2k−2 · · · −2kx n |z|2k−2 (2n+1)×2n
=
IR2n 2ky|z|2k−2
−2kx|z|2k−2
:
Then we have Lp u = div(B|∇L u|p−2 ∇L u);
(1.2)
B∇L u = A∇u;
(1.3)
where div and ∇ are taken with respect to the variable (x; y; t) ∈ R2n+1 . A natural family of anisotropic dilations attached to Lp is (z; t) = (z; 2k t);
¿ 0; (z; t) = (x; y; t) ∈ R2n+1 :
(1.4)
One easily checks that Lp ◦ = p ◦ Lp ; therefore Lp is a homogeneous partial diHerential operator of degree p with respect to { } in the sense of [4,13]. It is easy to verify that d (z; t) = Q d z dt; where Q = 2n + 2k
(1.5)
and d z dt denotes the Lebesgue measure on R2n+1 . The generator of the group dilations { }¿0 is the 3eld
n @ @ @ xj + 2kt : + yj X= (1.6) @xj @yj @t j=1
Another smooth vector 3eld on R2n+1 used later is
n @ @ yj : − xj T= @xj @yj
(1.7)
j=1
Consider the function d(z; t) = (|z|4k + t 2 )1=4k ;
(1.8)
168
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
where |z|2 =
n
j=1
Xd = d;
(xj2 + yj2 ). Since d ◦ = d for every ¿ 0, we have
Td = 0:
(1.9)
2n+1
2n+1
| d(z; t) ¡ r}, @Br = {(z; t) ∈ R | d(z; t) = r} and call these We let Br = {(z; t) ∈ R sets, respectively, the generalized ball and the generalized sphere centered at the origin with radius r. Let p = |z|p(2k−1) =dp(2k−1) , ’p = t|z|p(2k−1)−2k =dp(2k−1) . Observing that p is homogeneous of degree zero with respect to (1.4), one has X
p
= 0:
(1.10)
An evident computation yields T’p = 0:
(1.11)
We obtain from (1.8) that
x|z|4k−2 + ty|z|2k−2 1 ∇L d = 4k−1 ; d y|z|4k−2 − tx|z|2k−2 |z|2k−1 |∇L d| = 2k−1 d and |∇L d|p =
p:
(1.12)
Hence we have
x|z|4k−2 + ty|z|2k−2
y|z|4k−2 − tx|z|2k−2 d(p−2)(2k−1)+4k−1 2kt|z|4k−2 x y p(2k−1)−2k 1 |z|p(2k−1) y + t|z| −x = p(2k−1) p(2k−1) d d d 2kt 0
B|∇L d|p−2 ∇L d =
= Noting that div X = Q; we conclude
|z|(2k−1)(p−2)
1 [ p X + ’p T ]: d
div T = 0;
(1.13) (1.14)
1 ( p X + ’p T ) d
1 1 1 1 ’ ’ = + div X + X div T + T p p p p d d d d
Lp d = div
=
p
d
(Q − 1):
(1.15)
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
169
Let S01; p be the completion of C0∞ (R2n+1 ) in the norm
1=p uS 1; p = |∇L u|p + |u|p : R2n+1
The plan of the paper is as follows. In Section 2 we deduce the fundamental solution of Lp according to the properties of the radial functions. Section 3 is devoted to a Picone-type identity associated to the family {Xj ; Yj } and a Hardy-type inequality. In Section 4, we establish the Pohozaev-type identities in our setting. A nonexistence result of the nonlinear degenerate subelliptic equation is given in Section 5. Finally, a Carleman-type estimate and uniqueness of the operator L2 are considered in Section 6 by using the result in Section 4.
2. A fundamental solution Theorem 2.1. The fundamental solution of Lp at the origin is −1 (p−Q)=(p−1) ; %p; k; Q = Cp; k; Q d
1 ¡ p ¡ Q;
(2.1)
where Cp; k; Q satis8es
p−1 |z|p(2k−1) d2k(p−2) Q−p −1 Cp; (Q + 4kp − p − 4k) : = − k; Q 4k (4kp−p+Q)=4k p−1 R2n+1 (1 + d ) Proof. Let d& = (d4k + &4k )1=4k ; We compute ∇ L d& =
1 d&4k−1
|∇L d& |p−2 =
& ¿ 0:
x|z|4k−2 + yt|z|2k−2 y|z|4k−2 − xt|z|2k−2
|z|2k−2 = 4k−1 d&
x|z|2k + yt
y|z|2k − xt
;
|z|(2k−1)(p−2) d2k(p−2)
(2.2)
d&(4k−1)(p−2)
and
B|∇L d& |p−2 ∇L d& =
=
x|z|4k−2 + yt|z|2k−2
|z|(2k−1)(p−2) d2k(p−2) y|z|4k−2 − xt|z|2k−2 (4k−1)(p−1) d& 4k−2 2kt|z| d2k(p−2) d&2k(p−2)+1
[
p& X
+ ’p& T ];
(2.3)
170
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
where p& = |z|p(2k−1) =dp(2k−1) , ’p& = t|z|p(2k−1)−2k =dp(2k−1) . Using the fact that Xd& = & & d4k =d&4k−1 , we have:
p(2k−1) |z|p(2k−1) dp(2k−1) d X p& = X p p(2k−1) = p(2k−1) X d d& dp(2k−1) & p(2k−1) p(2k−1) d |z| dp(2k−1) d4k = p(2k−1) p(2k − 1) p(2k−1) − p(2k − 1) p(2k−1)+1 4k−1 d d& d& d& d4k |z|p(2k−1) dp(2k−1) = p(2k − 1) p(2k−1) p(2k−1) 1 − 4k d d& d& = p(2k − 1) and
X
p&
&4k d4k &
(2.4)
d2k(p−2) d&2k(p−2)+1
= 2k(p − 2)
d2k(p−2) d&2k(p−2)+1
− [2k(p − 2) + 1]
d2kp d&2kp+1
:
(2.5)
Combining these equations, we obtain Lp d& = div(B|∇L d& |p−2 ∇L d& ) 2k(p−2) d = div 2k(p−2)+1 ( p& X + ’p& T ) d& 2k(p−2) d2k(p−2) d = p& 2k(p−2)+1 div X + X d& d&2k(p−2)+1 =Q =
p&
p&
d2k(p−2)
+
d&2k(p−2)+1 d2k(p−2)
d&2k(p−2)+1
d2k(p−2) d&2k(p−2)+1
X
p&
+
p&
p& X
Q + 2k(p − 2) + p(2k − 1)
− [2k(p − 2) + 1]
d4k d4k &
;
d2k(p−2)
d&2k(p−2)+1 &4k d4k & (2.6)
where we have used (1.14) and the fact T’p& = 0. Taking f(d& ) = d&(p−Q)=(p−1) and directly computing show f (d& ) =
p − Q (1−Q)=(p−1) : d p−1 &
We see after using (2.6) and (2.7) that Lp (f(d& )) = divL (|∇L f(d& )|p−2 ∇L f(d& )) = divL (|f (d& )∇L d& |p−2 f (d& )∇L d& )
(2.7)
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
171
p−1 Q−p = divL − d&1−Q |∇L d& |p−2 ∇L d& p−1
Let now
p−1 Q−p p 1−Q =− Lp d& ] [(1 − Q)d−Q & |∇L d& | + d& p−1
p−1 Q−p |z|p(2k−1) d2kp =− (1 − Q)d−Q & p−1 dp(4k−1) & d2k(p−2) + d&1−Q p& 2k(p−2)+1 Q + 2k(p − 2) d&
&4k d4k + p(2k − 1) 4k − (2k(p − 2) + 1) 4k d& d&
p−1 2k(p−2) Q−p &4k −Q d =− Q + 2k(p − 2) + p(2k − 1) p& d& d4k p−1 d&2k(p−2) & d4k d4k − (2k(p − 2) + 1) 4k + (1 − Q) 4k d& d&
p−1 2k(p−2) Q−p &4k −Q d =− (Q + 4kp − p − 4k) 4k p& d& 2k(p−2) p−1 d& d&
p−1 |z|p(2k−1) d2k(p−2) &4k Q−p (Q + 4kp − p − 4k) : (2.8) =− p−1 d4kp−p+Q &
Q−p p−1 Then (2.8) becomes K(z; t) = −
p−1 (Q + 4kp − p − 4k)
|z|p(2k−1) d2k(p−2) : (1 + d4k )(4kp−p+Q)=4k
(2.9)
Lp (d&(p−Q)=(p−1) ) = &−Q K(1=& (z; t)): For any u ∈ C0∞ (R2n+1 ), it follows: (Lp (d(p−Q)=(p−1) ); u) = lim &→0
R2n+1
= lim &−Q &→0
= lim
&→0
R2n+1
R2n+1
= u(0; 0)
Lp (d&(p−Q)=(p−1) )u K(1=& (z; t))u(z; t)
K(z; t)u(&z; &2k t)
R2n+1
The conclusion is deduced from (2.10).
K(z; t):
(2.10)
172
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
Remark 2.1. A direct examination shows that the same formula hold for p ¿ Q. In the critical case p = Q, the fundamental solution is 1 Ck; Q log : d
(2.11)
In fact, let f(d& ) = log 1=d& , clearly f (d& ) = −1=d& and LQ (f(d& )) = divL (|∇L f(d& )|Q−2 ∇L f(d& )) = div(−d&1−Q |∇L d& |Q−2 ∇L d& ) Q 1−Q LQ d& ] = −[(1 − Q)d−Q & |∇L d& | + d&
=−
−Q Q& d&
d2k(Q−2)
(4kQ − 4k) 2k(Q−2)
d&
= −4k(Q − 1)
&4k d4k &
|z|Q(2k−1) d2k(Q−2) &4k d4kQ &
:
Denoting K(z; t) = −4k(Q − 1) and
|z|Q(2k−1) d2k(Q−2) (1 + d4k )Q
Ck; Q =
R2n+1
−1 K(z; t) d z dt
;
then one has at (z; t) = (0; 0),
1 = : LQ Ck; Q log d 3. Hardy-type inequalities In this section, we establish a Picone-type identity for the family {Xj ; Yj }, which is an extension of the identity for the p-Laplacian in the Euclidean space in [1]. Using the identity, we prove a Hardy-type inequality to Lp . Note that for p = 2 and k = 1, it is the result obtained by Garofalo and Lanconelli [7]. Lemma 3.1. For di:erentiable functions v ¿ 0, u ¿ 0 on ) ⊂ R2n+1 , where ) is a bounded or unbounded domain in R2n+1 , it holds L(u; v) = R(u; v) ¿ 0;
(3.1)
where L(u; v) = |∇L u|p + (p − 1)
up up−1 p |∇ v| − p |∇L v|p−2 ∇L u · ∇L v; L vp vp−1
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
p
p−2
R(u; v) = |∇L u| − |∇L v|
∇L
up
173
· ∇L v:
vp−1
Moreover, L(u; v) = 0 a.e. on ) i: ∇L (u=v) = 0 a.e. on ). Proof. It follows from straightforward calculations. It contains Lemma 2.1 in [14]. Lemma 3.2. Suppose that for some ¿ 0, v ∈ C ∞ ()) satis8es − Lp v ¿ gvp−1 and v ¿ 0 in );
(3.2)
where g is some positive, weight function. Then we have for any u in S01; p ()), (3.3) |∇L u|p ¿ g|u|p : )
)
Proof. Thanks to Lemma 3.1, we have 0 6 L(u; v) = R(u; v) )
)
=
)
=
)
6
)
|∇L u|p − |∇L u|p +
|∇L u|p −
∇L
)
up
)
vp−1
)
up vp−1
|∇L v|p−2 ∇L v
Lp v
gup :
Eq. (3.3) now follows. Theorem 3.1. Let u ∈ C0∞ (R2n+1 \{(0; 0)}); 1 ¡ p ¡ Q. Then
p p(2k−1) p |z| |u| Q−p p |∇L u| ¿ ; p d d R2n+1 R2n+1
(3.4)
where d is as in (1.8). Proof. Setting v = f(d) = d(p−Q)=p , one easily sees f (d) =
p − Q −Q=p : d p
Using (1.12), (1.15) and (3.5), we get Lp v = divL (|∇L v|p−2 ∇L v)
p−1 Q(p−1) Q−p − p−2 p = divL − d |∇L d| ∇L d p
(3.5)
174
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
=− + =− + =− =−
Q−p p Q−p p Q−p p Q−p p Q−p p Q−p p
p−1 d
p−1
d
p
p
pd
−
Q(p−1) −1 p
p (Q
− 1)
Q(p − 1) − Q(p−1) −1 p d p
pd
p
Q(p−1) p L
Q(p − 1) − Q(p−1) −1 p |∇L d|p d p
p−1
p−1
−
−
p
Qp−Q+p p
vp−1 : dp
This, combined with Lemma 3.2, yields (3.4). 4. Pohozaev-type identities In what follows we set Zj = Xj ; Zn+j = Yj ; j = 1; : : : ; n, and adopt the summation convention over repeated indices. The following Lemma 4.1 is a generalization of the Proposition 2.2 from [9]. Lemma 4.1. Let ) be a bounded, piecewise C 1 domain and G be a real vector 8eld in R2n+1 . Then, for any di:erentiable function u, |Zu|p G · ˜n − p Gu|Zu|p−2 Zj uZj · ˜n @)
=
)
@)
div G|Zu|p − p
−p
)
)
div Zj Gu|Xu|p−2 Zj u + p
)
[G; Zj ]u|Zu|p−2 Zj u
GuLp u;
(4.1)
where ˜n denotes the outward unit normal to @), |Zu| = (
2n
j=1
|Zj u|2 )1=2 .
Proof. Since
p=2 p=2−1 2n 2n 2n p |Zj u|p = |Zj u|2 G |Zj u|2 G(|Zu|p ) = G 2 j=1
= p|Zu|p−2 GZj uZj u
j=1
j=1
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
175
we deduce from the divergence theorem |Zu|p G · ˜n @)
=
)
=
)
=
)
=
)
=
)
div(G|Zu|p ) div G|Zu|p +
div G|Zu|p + p div G|Zu|p + p
p
)
GZj u|Zu|p−2 Zj u
)
)
[G; Zj ]u|Zu|p−2 Zj u +
[G; Zj ]u|Zu|p−2 Zj u + p
)
Zj Gu|Zu|p−2 Zj u
)
div G|Zu| + p
−p
G(|Zu|p )
Zj (Gu|Zu|p−2 Zj u)
GuZj (|Zu|p−2 Zj u)
)
)
)
div G|Zu|p + p
− =
)
p−2
)
[G; Zj ]u|Zu|
p−2
div Zj Gu|Zu|
Zj u + p
Zj u − p
)
@)
Gu|Zu|p−2 Zj uZj · ˜n
GuLp u:
Then (4.1) is valid. Lemma 4.2. For j = 1; 2; : : : ; 2n; it holds [Zj ; X ] = Zj ;
(4.2)
where X is given by (1.6). Proof. A direct calculation shows for j = 1; : : : ; n;
@ @ @ @ 2k−2 @ xi + 2kt + 2kyj |z| + yi Xj X = @xj @t @xi @yi @t i =
@ @2 @2 @2 @2 + 2kxi yj |z|2k−2 + xi + yi + 2kt @xj @xi @xj @xj @yi @xj @t @xi @t + 2kyi yj |z|2k−2
@2 @2 @ + 4k 2 yj t|z|2k−2 2 + 4k 2 yj |z|2k−2 @yi @t @t @t
176
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
and
XXj =
i
= xi
@ @ xi + yi @xi @yi
@ + 2kt @t
@ @ + 2kyj |z|2k−2 @t @xj
@2 @2 @2 @ + 2kxi yj |z|2k−2 + 4k(k − 1)xi2 yj |z|2k−4 + yi @xj @yi @t @xi @t @xi @xj
+ 2kyj |z|2k−2 + 2kt Hence, we see [Xj ; X ] = =
@ @ @2 + 4k(k − 1)yi2 yj |z|2k−4 + 2kyi yj |z|2k−2 @t @t @yi @t
@2 @2 + 4k 2 yj t|z|2k−2 2 : @t @xj @t
@ @ @ @ + 4k 2 yj |z|2k−2 − 2kyj |z|2k−2 − 4k(k − 1)yj |z|2k−2 @xj @t @t @t @ @ + 2kyj |z|2k−2 = Xj : @t @xj
Similarly, [Yj ; X ] = Yj , j = 1; : : : ; n. Using Lemma 4.1 with G = X and Lemma 4.2 lead immediately to the following Pohozaev-type identity. Theorem 4.1. Let ) be a bounded, piecewise C 1 domain. Then Xu|Zu|p−2 Zj uZj · ˜n − |Zu|p X · ˜n = (p − Q) |Zu|p + p XuLp u: p @)
@)
)
)
(4.3)
Remark 4.1. If p = 2 and k = 1, then (4.3) is the identity in [8]. Also see [10]. Letting G = T in Lemma 4.1 we will have from (1.7) and (1.14) that p−2 Tu|Zu| Zj uZj · ˜n = TuLp u: @)
)
Theorem 4.2. Under the assumptions of Theorem 4.1, one has 2 Xu Xu Tu s |Zu|p ds X · ˜n − p |Zu|p−2 ds X · ˜n − p |Zu|p−2 d ’X · ˜n d d d @Br @Br @Br 2 Xu =(Q + s − p) ds |Zu|p − ps ds |Zu|p−2 d Br Br Xu Tu −p −ps ds ’|Zu|p−2 ds XuLp u; (4.4) d d Br Br where Br = {(z; t) ∈ R2n+1 | d(z; t) 6 r}, @Br = {(z; t) ∈ R2n+1 | d(z; t) = r}, ’2 ; s ¿ 0.
=
2;
’=
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
177
Proof. Set G = ds X in Lemma 4.1. Then [G; Zj ] = ds XZj − Zj (ds X ) = ds XZj − ds Zj X − sds−1 Zj dX = ds [X; Zj ] − sds−1 Zj dX = −ds Zj − sds−1 Zj dX;
(4.5)
div G = div(ds X ) = ds div X + X (ds ) = (Q + s)ds and div Zj = 0; where we have used (4.2) and (1.14), respectively. Substituting these equations in (4.1) yields ds Xu|Zu|p−2 Zj uZj · ˜n |Zu|p ds X · ˜n − p @Br
@Br
= (Q + s − p) −p
Br
Br
ds |Zu|p − ps
Br
ds−1 |Zu|p−2 XuZj uZj d
ds XuLp u:
(4.6)
Noting the fact that ˜n = ∇d=|∇d| on @Br , we infer X · ˜n =
Xd d = |∇d| |∇d|
on @Br :
On the other hand, in view of the formula ZuZd = ∇L u · ∇L d = A∇d · ∇u = we have @Br
1 [ Xu + ’Tu]; d
ds Xu|Zu|p−2 Zj uZj · ˜n
=
@Br
=
@Br
=
@Br
ds Xu|Zu|p−2 Zj uZj d
1 |∇d|
ds−1 Xu|Zu|p−2 [ Xu + ’Tu] ds−2 |Zu|p−2 |Xu|2 X · ˜n +
1 |∇d|
@Br
ds−2 ’|Zu|p−2 XuTuX · ˜n
(4.7)
178
and
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
Br
ds−1 |Zu|p−2 XuZj uZj d =
Br
=
Br
ds−2 |Zu|p−2 Xu[ Xu + ’Tu] ds−2 |Zu|p−2 |Xu|2 +
Br
ds−2 ’|Zu|p−2 XuTu:
(4.8)
Replacing (4.7), (4.8) in (4.6) leads to (4.4). Let G = ds T in Lemma 4.1. We have the following: Theorem 4.3. Under the assumptions of Theorem 4.1, it holds 2 Tu Xu Tu X · ˜n + p p ’ds |Zu|p−2 X · ˜n ds |Zu|p−2 d d d @Br @Br 2 Tu Xu Tu =ps ds |Zu|p−2 ’ds |Zu|p−2 + p ds TuLp u: (4.9) + ps d d d Br Br Br 5. A nonexistence result As a consequence of theorems in Section 4, we give a simple nonexistence result for nonlinear equations. De'nition 5.1. Let ) be a C 1 open domain in R2n+1 , (0; 0) ∈ ). We say that it is strict generalized starshaped with respect to (0; 0) if X · ˜n ¿ 0 at every point of @). Theorem 5.1. Let ) ⊂ R2n+1 be connected, bounded and strict generalized starshaped with respect to (0; 0) ∈ ). Then the problem Lp u + f(u) = 0 in ); (5.1) u=0 on @) L u = 0, if f is locally Lipschitz, has no nonnegative solution u ∈ S01; p ()) ∩ C 1 ()), f(0) = 0 and pQF(u) − (Q − p)uf(u) 6 0; u where F(u) = 0 f(s) ds.
(5.2)
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
179
Proof. If u is a nonnegative solution of (5.1) which satis3es (5.2), we have from the fact ˜n = ∇u=|∇u| on @) (see [8]) that Zj u Xu|Zu|p−2 Zj u |Zu|p X · ˜n: (5.3) Xu|Zu|p−2 Zj uZj · ˜n = = |∇u| @) @) @) Eq. (5.1) and an integration by parts conclude p uf(u) = − uLp u = |∇L u| = |Zu|p )
and
)
)
)
XuLp u = −
Xuf(u) = −
=Q
)
)
)
X
X (F(u)) = −
@)
u
0
=−
)
(5.4)
)
f(t) dt
F(u)X · ˜n +
)
div XF(u)
F(u):
(5.5)
Inserting (5.3)–(5.5) into (4.3) implies p (p − 1) |Zu| X · ˜n = [pQF(u) − (Q − p)uf(u)] ¿ 0; @)
)
which is a contradiction. The claim is proved.
6. A Carleman-type estimate There has been a considerable amount of work in the study of the unique continuation property for second-order elliptic diHerential equations, we refer the reader to the paper [15] and the references therein. The topic for generalized Baouendi–Grushin operator is considered by Garofalo [6]. In this section, we prove a Carleman-type estimate and then the unique continuation of the operator L2 . Theorem 6.1. Let R0 ¡ 1 be 8xed and u ∈ C0∞ (BR0 \{(0; 0)}), for each 2 ¿ 0 and some constants C ¿ 0, satisfying −2
e2d u2 = o(1) −1 2d−2
as d → 0;
e
(Lu)2 ∈ L1 (BR0 )
|Tu| 6 C
|z| 1=2 |u| = C d2
(6.1) (6.2)
and 2k=2(2k−1)
d
|u|:
(6.3)
180
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
where d is as in (1.8), = 2 = |z|2(2k−1) =d2(2k−1) . Then there exist 20 ¿ 1; C0 ¿ 0 depending only on Q such that −2 −1 2+2 2d−2 d e (Lu)2 ; e2d [ d−22−2 u2 + |∇L u|2 ] 6 C0 (2−2 + R20 ) R2n+1
R2n+1
(6.4)
where L = L2 . Theorem 6.2. Suppose that u satis8es |Lu| 6 C1
d
|u| + C2 2
1=2
|∇L u|
d
(6.5)
in BR for some constants C1 ; C2 ¿ 0; u; |∇L u|; Lu ∈ L2 (BR ) and for each 2 ¿ 1 −2
ed uL∞ (Br ) = o(1)
(r → 0)
(6.6)
and −2
ed |∇L u| ∈ L2 (BR ):
(6.7)
Then there exists r0 = r0 (Q; C1 ; C2 ) ¿ 0 such that u ≡ 0 in Br0 . −2
Proof of Theorem 6.1. Let u = f(d)w = e−d w and we have Lu = wL(f(d)) + f(d)Lw + 2∇L w · ∇L (f(d)): From the formula L(f(d)) =
(6.8)
Q−1 f (d) + f (d) d
and the facts that −2
f (d) = 2d−2−1 e−d ; −2
f (d) = 2[(−2 − 1)d−2−2 + 2d−22−2 ]e−d ; we obtain L(f(d)) = e
−d−2
2 d2+2
2 −2−2+Q : d2
(6.9)
On the other hand, it follows ∇L (f(d)) · ∇L w = f (d)∇L d · ∇L w 1 = f (d)A∇d · ∇w = f (d) [ Xw + ’Tw] d = e−d
−2
2 [ Xw + ’Tw]; d2+2
(6.10)
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
181
where ’ = ’2 = t|z|2k−2 =d2(2k−1) . Applying (6.9), (6.10) to (6.8) yields 2 2 22 −d−2 − 2 − 2 + Q w + Lw + 2+2 [ Xw + ’Tw] Lu = e d2+2 d2 d and then d
2+2 2d−2
e
2 −2−2+Q w (Lu) = d d2+2 d2 2 22 + Lw + 2+2 [ Xw + ’Tw] d 2 2 ¿ 42[ Xw + ’Tw] Lw + − 2 − 2 + Q w (6.11) d2+2 d2 2
2+2
2
where we have used the inequality (a + (b + c))2 ¿ 2a(b + c). Now we integrate (6.11) with respect to the measure −1 d z dt and obtain −1 2+2 2d−2 d e (Lu)2 ¿ 42 XwLw R2n+1
R2n+1
+ 422
+ 42 + 422
R2n+1
d2+2
’
R2n+1
R2n+1
2 − 2 − 2 + Q wXw d2
TwLw ’
d2+2
2 − 2 − 2 + Q wTw: d2
Since u ∈ C0∞ (BR0 \{(0; 0)}), letting p = 2 in Theorem 4.1 leads to Q−2 XwLw = |∇L w|2 ¿ 0: 2 R2n+1 R2n+1 By the divergence theorem we 3nd 2 422 − 2 − 2 + Q wXw 2+2 d2 R2n+1 d 2 2 =22 − 2 − 2 + Q Xw2 2+2 d2 R2n+1 d 2 2 = − 22 Q − 2 − 2 + Q w2 2+2 d2 R2n+1 d 2 −222 X − 2 − 2 + Q w2 d2+2 d2 R2n+1
(6.12)
(6.13)
182
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
2 2 2 w2 = − 22 Q w + 22 Q(2 + 2 − Q) 2+2 22+2 R2n+1 d R2n+1 d 2 2 3 w2 w − 22 (2 + 2)(2 + 2 − Q) + 2(22 + 2)2 2+2 22+2 R2n+1 d R2n+1 d
Q 2 2 2 3 w2 : = 42 2 + 1 − w − 22 (2 + 2 − Q) 2+2 22+2 2 R2n+1 d R2n+1 d (6.14) 2
Now we consider the third term on the right-hand side of (6.12). Since L = 6z + 4k 2 |z|4k−2 we have 42
’
R2n+1
@2 @ + 4k|z|2k−2 T; @t 2 @t
TwLw
@w @w t − x 6z w y 2 2 k @x @y R2n+1 (x + y )
2 @w @w t 2 4k−2 @ w − x · 4k y + 2 |z| @t 2 @x @y (x + y2 )k
@w @w t @w @w 2k−2 @ : · 4k|z| y y + 2 − x − x @x @y @x @y @t (x + y2 )k
= 42
Denote that I1 =
@w @w −x 6z w; @x @y R2n+1
2 @w @w t 2 4k−2 @ w − x |z| I2 = 4k y ; 2 2 k @x @y @t 2 R2n+1 (x + y )
@w @w t @w @w 2k−2 @ |z| y : I3 = 4k y −x −x 2 2 k @x @y @t @x @y R2n+1 (x + y ) t (x2 + y2 )k
y
It is clear that I1 = I2 = 0. From the fact −2
−2
−2
−2
Tw = T (ed u) = u(T ed ) + ed Tu = ed Tu and (6.3), one has |Tw|2 |z|−2 6 C
d4
|w|2 :
(6.15)
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
Consequently, we see −2 I3 = 4k |z| d x dy R2n
−∞
= −2k
∞
R2n+1
tTwd(Tw) = 2k
|z|−2 |Tw|2 ¿ − 2kC
R2n+1
d4
−2
R2n
|z|
183
d x dy
∞
−∞
td(Tw)2
w2 :
Returning to (6.15) we have for 2 large enough, ’ 42 w2 : TwLw ¿ − C1 2 22+2 d 2n+1 2n+1 R R
(6.16)
As for the fourth term on the right-hand side of (6.12), it follows 2 ’ − 2 − 2 + Q Tw2 2+2 d2 R2n+1 d 2 ’ =− div T 2+2 − 2 − 2 + Q w2 d2 d R2n+1 2 ’ T − 2 − 2 + Q w2 − d2+2 d2 R2n+1 = 0: Using (6.13), (6.14) and (6.16), we deduce from (6.12) that there exist C0 = C0 (Q) ¿ 0, 20 = 20 (Q; R) ¿ 0 such that when 2 ¿ 20 , −1 2+2 2d−2 w2 (Lu)2 ¿ C0 24 d e 22+2 d 2n+1 2n+1 R R −2 = C0 2 4 e2d u2 ; 22+2 d 2n+1 R that is, R2n+1
d22+2
e
1 u 6 C0 2 4
2d−2 2
R2n+1
−1 2+2 2d−2
d
e
(Lu)2 :
(6.17)
On the other hand, for the smooth function g on R2n+1 , u∇g · A∇u = − u∇(Au∇g) R2n+1
R2n+1
=−
R2n+1
u(∇u · A∇g + uLg)
=− It shows R2n+1
u∇g · A∇u = −
R2n+1
1 2
u∇g · A∇u −
R2n+1
u2 Lg
R2n+1
u2 Lg:
(6.18)
184
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
and then R2n+1
guLu =
R2n+1
gu div(A∇u)
=− =
1 2
R2n+1
R2n+1
u∇g · A∇u − u2 Lg −
R2n+1
R2n+1
g∇u · A∇u
g|∇L u|2
(6.19)
−2
Letting g = e2d , we have from (6.9)
−2 22e2d 22 Lg = +2−Q+2 : d2+2 d2 Due to d ¡ R0 ¡ 1 we obtain |Lg| 6 C2 22
−2
d22+2
e2d :
(6.20)
Using (6.19) and HGolder’s inequality yields
1=2 2d−2 2 2d−2 2 2d−2 2 2 e u + e u e |∇L u| 6 C2 22+2 22+2 R2n+1 d R2n+1 R2n+1 d
1=2 2 −1 22+2 2d−2 d e (Lu) × R2n+1
6 C2
2
R2n+1
d22+2
e
R2 u + 0 2
2d−2 2
−2 1 + e2d u2 2 R2n+1 d22+2
1 1 2 + 2 + R0 6C 24 2 R2n+1
−1 2+2 2d−2
R2n+1
−1 2+2 2d−2
d
e
d
e
(Lu)2 ;
(Lu)2
(6.21)
where we have used (6.17) in the last inequality. From this, together with (6.17), the result follows. Proof of Theorem 6.2. Without restriction, we assume that R ¿ 1 and pick a cut-oH function 7 ∈ C02 (B1 ) with 7 ≡ 1 on B1=2 . Applying Theorem 6.1 to 7u yields −2 e2d [ d−22−2 u2 + |∇L u|2 ] B1=2
6
B1
−2
e2d [ d−22−2 (7u)2 + |∇L (7u)|2 ]
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
6 C0 (2
−2
+
R20 )
−1 2+2 2d−2
d
B1=2
+ C0 (2−2 + R20 )
e
According to (6.5) we have 2 2 2 |Lu| 6 max(C1 ; C2 ) 6 2 max(C12 ; C22 )
(Lu)2
−1 2+2 2d−2
d
B1 \B1=2
d2
1=2
|u| + 2
d4
d
2
|u| +
185
e
(L(7u))2 :
(6.22)
2 |∇L u|
d2
2
:
|∇L u|
Let 20 be large enough such that 2C0 (2−2 + R20 )max(C12 ; C22 ) 6 We have C0 (2
−2
and thus B1=2
e
+
2d−2
R20 )
[ d
1 2
−1 2+2 2d−2
d
B1=2
−22−2 2
e
2
for 2 ¿ 20 : 1 (Lu) 6 2 2
u + |∇L u| ] 6 2C0 (2
−2
−2
B1=2
+
e2d [ d−22−2 u2 + |∇L u|2 ]
R20 )
(6.23) B1 \B1=2
−1 2d−2
e
(L(7u))2 :
Letting 2 → ∞ and noting R0 ¡ 1, we have u ≡ 0 in B1=2 . This accomplishes the proof. Acknowledgements The authors are grateful to the referee for his/her invaluable comments and suggestions on the manuscript. We thank the referee for pointing [15] out to us. References [1] W. Allegretto, Y.X. Huang, A Picone’s identity for the p-Laplacian and applications, Nonlinear Anal. 32 (1998) 819–830. [2] R. Beals, B. Gaveau, P. Greiner, Uniforms hypoelliptic Green’s functions, J. Math. Pures Appl. 77 (1998) 209–248. [3] C. Capogna, D. Danielli, N. Garofalo, Capacity estimates and the local behavior of solutions of nonlinear subelliptic equations, Amer. J. Math. 118 (1997) 1153–1196. [4] Cui shangbin, Some necessary conditions for local solvability of linear partial diHerential operators, J. DiHerential Equations 106 (1993) 1–9. [5] G.B. Folland, A fundamental solution for a subelliptic operator, Bull. Amer. Math. Soc. 79 (1973) 373–376.
186
H. Zhang, P. Niu / Nonlinear Analysis 54 (2003) 165 – 186
[6] N. Garofalo, Unique continuation for a class of elliptic operators which degenerate on a manifold of arbitrary codimension, J. DiHerential Equations 104 (1993) 117–146. [7] N. Garofalo E. Lanconelli, Frequency functions on the Heisenberg group, the uncertainty principle and unique continuation, Ann. Inst. Fourier Grenoble 40 (1990) 313–356. [8] N. Garofalo, E. Lanconelli, Existence and nonexistence results for semilinear equations on the Heisenberg group, Indian Univ. Math. J. 41 (1992) 71–98. [9] N. Garofalo, Z. Shen, Absence of positive eigenvalues for a class of subelliptic operators, Math. Ann. 304 (1996) 701–715. [10] N. Garofalo, D. Vassilev, Regularity near the characteristic set in the nonlinear Dirichlet problem and conformal geometry of sub-Laplacians on Carnot groups, Math. Ann. 318 (2000) 453–516. [11] P.C. Greiner, A fundamental solution for a nonelliptic partial diHerential operator, Canad. J. Math. 31 (1979) 1107–1120. [12] L. HGormander, Hypoelliptic second order diHerential equations, Acta Math. Uppsala 119 (1967) 147–171. [13] Luo Xuebo, Global analyticity for quasi-homogeneous linear partial diHerential equations, Comm. Partial DiHerential Equation 23 (1998) 1171–1179. [14] P. Niu, H. Zhang, Y. Wang, Hardy type and Rellich type inequalities on the Heisenberg group, Proc. Amer. Math. Soc. 129 (2001) 3623–3630. [15] C.D. Sogge, Oscillatory integrals and unique continuation for second order elliptic diHerential equations, J. Amer. Math. Soc. 2 (1989) 491–516.