Low dimension anomalies and solvability in higher dimensions for some perturbed Pohozaev equation G. Mancini∗ Dipartimen...
7 downloads
324 Views
170KB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Low dimension anomalies and solvability in higher dimensions for some perturbed Pohozaev equation G. Mancini∗ Dipartimento di Matematica, Universit´a di Roma Tre Abstract In this talk, we will present a new solvability condition in higher dimension for a Brezis Nirenberg type equation. This result, which we believe to be sharp in some sense, has been obtained in collaboration with Adimurthi
1
Introduction
Very surprising ”low dimension phenomena” have been observed by Brezis and Nirenberg in their pioneering work [BN] concerning perturbations of the ”Pohozaev equation”
(P )
N +2
−∆u = u N −2 + p(u) u>0 in Ω u=0 in ∂Ω
in Ω
Here, Ω ⊂ 3, (P ) has a solution ∀Ω (ii)if N = 3, (P ) is not solvable in general (in fact (P ) has a solution in a ball Br iff r is large enough) As for (i) it follows from the very general sufficient condition they found (assuming p ≥ 0, but this is non really relevant)
(∗) lim →0
4−N 2
Z
|x|≤ √1
P ((
N −2 1 ) 2 ) = +∞ + |x|2
Rs N where P (s) = 0 p(t)dt. Of course, p(s) = s N −2 does not satisfy (∗) if N = 3. This example shows two facts : 0∗
Supported by M.U.R.S.T.
1
- anomalies may occurr in dimension 3 - condition (∗) is sharp in dimension 3. In view of this example, two related questions naturally arise: - is condition (∗) sharp also in higher dimensions ? - do anomalies occur also in higher dimensions ? We also refer to [PS], [BG] for related problems and interpretation of ”low dimension phenomena”. In a recent paper [AY], Adimurthi and Yadava gave an adfirmative answer to the question above in dimensions 4, 5, 6: they exibited a class of compactely supported perturbations PN such that (k) if N = 3, 4 and p ∈ PN , there are anomalies (i.e. there are solutions on Br iff r is large) R (kk) if N = 5, 6, and p ∈ PN and, in addition, P ( |x|N1 −2 ) = 0, then there are anomalies. (kkk) if N ≥ 7, and p ∈ PN , then (P ) is solvable on any ball. Of course, in cases (k) − (kk) Rcondition (∗) is violated: in case (k), because obviously P ( |x|N1 −2 ) < ∞, while, in case (kk), because R R N −2 1 2 ) = 0() (see Lemma 4.1 below). P ( |x|N1 −2 ) = 0 implies |x|≤ √1 P (( +|x| 2)
In the attempt to extend the existence result in (kkk) to more general perturbations and domains, Adimurthi and myself [AM] realized that condition (∗) is not sharp in dimension N ≥ 7. To be more precise, let us first give a closer look to condition (∗) , limiting ourselves, for sake of simplicity, to perturbations p in the class P := C0∞ ((0, ∞)). If we denote, for a given p ∈ P
Ik = Ik (p) :=
Z
[
1 dk P ( N −2 )](|x|2 )dx k dt t 2
it is easy to see that (j) If Ik = 0 ∀k , orI0 = I1 ... = Ik−1 and Ik < 0 for some k, (∗) fails in any dimension N (jj) If I0 = I1 = ... = Ik−1 = 0, and Ik > 0 for some k = 0, 1... then (∗) holds iff N > 2k + 4.In fact, 4−N R N −2 1 2 )= if I0 = I1 = ... = Ik−1 = 0 then lim→0 2 |x|≤ √1 P (( +|x| 2) k N −2 4+2k−N R d 1 2 ) lim→0 2 P (( +|x| 2) 0, and this explains the ”surprising” existence result (kkk)). 2
In particular: R I0 = 0 and 0 > I1 = − N 2−2 p( |x|N1 −2 ) |x|1N ⇒ (∗) fails ∀N . On the other hand, we will prove that, if
(∗∗) I0 = 0 and
Z
1 N
|∇H|2 −
Z
p(
1 1 ) >0 |x|N −2 |x|N
then(P ) has a solution on any Ω if N ≥ 7. Here, H is the finite energy solution of − ∆u = p(
1 ) in
(1.1)
So, we see that, in case I0 = 0, a much weaker condition than (∗) insures, in dimension N ≥ 7, the solvability of (P ) on any domain, i.e. (∗) is not sharp in dimension N ≥ 7. Of course, all this does not answer the question whether the anomaly described above occurs in dimension N ≥ 7. Even if we do believe that such anomalies occur in any dimension ( i.e. ”anomaly” is not a low dimension phenomena!), we content ourselves with the weaker conjecture N
S 2 if (P ) has solutions with energy smaller than N (S := best Sobolev con2N 1 N N −2 stant in the embedding H → L (< )) on any small ball Br , then
Z
1 |∇H| − N 2
Z
p(
1 1 ) ≥0 |x|N −2 |x|N
In other words, we suspect that a weak inequality in (∗∗) is necessary for ”low energy” solutions to exist on any small ball.
2
The main result and the variational approch
For sake of simplicity we will assume p ∈ C0∞ ((0, ∞)). Let H be the unique solution in D1,2 of − ∆H = p(
1 |x|N −2
) in
R Theorem 2.1 If P ( |x|N1 −2 ) > 0 and N ≥ 5 , (P ) has a solution. R R R If P ( |x|N1 −2 ) = 0 and |∇H|2 > N1 p( |x|N1 −2 ) |x|1N , then - if N ≥ 7, (P ) has a solution ∀Ω - if N = 3, 4, 5, 6, (P ) has a solution if Ω contains a large ball. 3
By standard arguments, a non zero critical point of 1 ˜ E(u) := 2
Z
N −2 |∇u| − 2N 2
Z
+
(u )
2N N −2
−
Z
P (u+ ) u ∈ H01 (Ω)
˜ enjoies the mountain pass is a solution of (P ). The energy functional E geometry, and, again by standard arguments, a mountain pass level is critical N
S 2 (S := best Sobolev constant). provided it is strictly smaller than N In fact it is enough to prove that, for some V ∈ H01 (Br ) , Br ⊂ Ω, it results N
S 2 max E(tV ) < (2.1) t≥0 N R R R 2N −2 where E(u) = 21 R|∇u|2 − N2N |u| N −2 − P (u) , u ∈ H01 (Br ) and P is s the even extension of 0 p(t)dt, s ≥ 0. This is because, after extending V equal to zero outside Br , it results ˜ E(t|V |) ≤ E(tV ).
3
Choise of the test function and basic estimates
To get (2.1), we will choose V = V , for suitably small, where V ∈ H01 (Br ) satisfies N +2
− ∆V = UN −2 + p(U ) on Br ⊂ Ω
(3.1)
N −2
Here U (x) = − 2 U ( x ) p N −2 cN 2 , cN = N (N − 2) U (x) = ( 1+|x| 2) N +2
Recall that −∆U = UN −2 in
2
|∇U | =
Z
2N
UN −2 = S
N 2
ˆ N −2 and In what follows, we will always assume < 4δ r2 , where δ = c−1 N δ ˆ ˆ δ < 1 is such that p(s) ≡ 0 for s ≤ δ. We will denote w := V − U , so that −∆w = p(U ) in Br and w ≡ −b N −2 N 2 . on ∂Br where b = ( 2c+r 2) N −2
We will also write sometimes c˜N instead of cN 2 4
Lemma 3.1 Let V be given by (3.1).Then
E(V ) ≤
N 2
S N
N −2 Z N +2 cN 2 U N −2 ]( )N −2 − r 2 N Z < Z 1 + o(1) − |∇w |2 − P (U ) 2 Br Br
+ +[o(1) +
(3.2)
where o(1) are small in uniformly with respect to r ∈ (1, ∞) Proof. Multiplying (3.1) by V and integrating by parts, and using the inequality 2N
2N
2N
|V | N −2 − UN −2 ≥
N −2 2N w , N −2 U
we get, by straightforward computations:
N +2 N R S 2 E(V ) ≤ N − 12 [ |x|≤r UN −2 w − p(U )U + p(U )w ]− R R − Br P (U ) − Br ×[0,1] (1 − t)p0 (U + tw )w2 N +2
Now, from −∆U = UN −2 , −∆w = p(U ), and Green formulas, we get Z
N +2 N −2
U Br Z
|x|≤r
Z
R
−
1 2
Z Br
UN −2 ≤
S N Z |∇w |2 −
E(V ) ≤
(3.3)
Br
N +2
and hence, since
N +2
p(U )(U − w UN −2 ) , Br Z p(U )w = |∇w |2 − w p(U ) w =
N 2
+
|x|≤r
N −2 2
R
N +2
c˜N N −2 ( ) 2 r Z P (U ) +
U N −2 : Z
N +2
U N −2 + b
Z
p(U ) −
Br
(1 − t)p0 (U + tB )w2
(3.4)
Br ×[0,1]
To complete the proof of the Lemma, we just use (3.4) and the estimates gathered in the following Lemma 3.2 R N R (i) p(U ) = 2 [
Proof 1 To see (i), we just make the change of variable y = − 2 x : R R N −2 N −2 N cN N 2 ) = 2 2 )dy p(( 2c+|x| p(( +|y| 2) 2) Br
because
N −2 2
cN p(( +|y| 2)
≤ r2 , |y| ≥
r √
cN ⇒ ( +|y| 2)
N −2 2
≤ ( crN2 )
N −2 2
≤ (δcN )
N −2 2
= δˆ ⇒
)=0
R N −2 N N 2 ) = 2 (o(1) + Hence p(( 2c+|x| p( |y|c˜NN−2 )) where o(1) is uniform 2)
Z (
2N
|w | N −2 )
N −2 N
Br
≤ [(
1 S
Z
1
|∇w |2 ) 2 + b (volBr )
N −2 2N
]2 ≤
Br
1 ≤ 2( S
Z
N −2 −2 N −2 |∇w |2 + ωNN cN ( ) ) N r
(3.5)
R where ωN := B1 dx. To see (iii), we can use the representation formula (recall that p(U ) ≡ 0 outside B√ ⊂ Br ) : δ
c−2 w (x) + b = N ωN
Z
1 1 √ p(U (y))( |x − y|N −2 − |x − y|N −2 )dy |y|≤ δ
(3.6)
where x is any R ppoint on ∂Br . p Now , |x| ≥ 2 δ ⇒ |x − y| ≥ δ ⇒ |y|≤√ |x − y|2−N ≤ δ R p N −2 N ( δ ) 2 ( δ ) 2 ωN = ωN δ , while |x| ≤ 2 δ implies |y|≤√ |x − y|2−N δ
≤
R
|y|≤3
√ |z|2−N = 9 ωN . 2 δ δ
N −2
So, for any x we obtain from (3.6) |w (x)| ≤ ( crN2 ) 2 + dN (dN just 1 depending on N and p) that is k w k∞ = 0( 2 ) uniformly for r ≥ 1. Finally, by H˘ older inequality and (ii), we have (t − 1)|p0 (U + tw )w2 | R 2N N −2 R 1 R N 2 ≤ ( |w | N −2 ) N 0 ( Br |p0 (U + tw )| 2 ) N R R R N 2 1 ≤ 0(1)[ |∇w |2 + ( r )N −2 ] 0 ( Br |p0 (U + tw )| 2 ) N dt |
R
Br ×[0,1]
R1 R N 2 By dominated convergence, 0 ( Br |p0 (U + tw )| 2 ) N →→0 0 for any given r, because k w kL∞ (Br ) →→0 0 for every r by elliptic theory. Furthermore, by p (iii), if r ≥ 1 , p0 (U + tw )(x)) ≡ 0 for |x| ≥ δ and < N , for some N 6
R N 2 depending only on N and not on r, and hence we get ( Br |p0 (U + tw )| 2 ) N ≤ 4 δ
4
2
k p0 k∞ ωNN . This complets the proof of (iiii).
Higher order estimates and the proof of the Theorem
We first estimate the last two terms in (3.2). Lemma 4.1 N +2 R R N +2 2 2 (c 2 |∇w | = |∇H|2 + o(1)) N B
(i) (ii)
Furthermore, all estimates are uniform in r ∈ (1, ∞). Proof √ (i) Let w ˜ (x) := −1 w ( x) in B √r , so that − ∆w ˜ = p((
N −2 cN ) 2 ) in B √r . 2 + |x|
Now, if H ∈ D1,2 (
N −2 cN ) 2 ) in
clearly w ˜ − H ≡ const on B √r , and hence R N +2 R N +2 R |∇w |2 = 2 B r |∇w ˜ |2 = 2 B Br √
r √
|∇H |2 =
N +2 2
R ( |∇H0 |2 +
o(1)), where
− ∆H0 = p(
R
c˜N ) in
and o(1) is uniform with respect to r in [1, ∞). Then (i) follows from N +2 R |∇H0 |2 = cN 2 |∇H|2 . 1 (ii) After the change of variable y = − 2 x we have 7
Z
N
P (U ) = 2 +1 (
Br
1
Z
P ((
B √r
N −2 cN ) 2 ) 2 + |x|
where the second integral is in fact taken over B √1 , because ≤ δr2 . First δ
lim
→0
Z
P ((
B √r
N −2 cN ) 2 )= 2 + |x|
Z
P(
c˜N ). |x|N −2
(4.1)
If this integral vanishes, lim→0
1
R
= lim→0
cN P (( +|x| 2)
|y|≤ √1
R
δ
|y|≤ √1
= − N 2−2 c˜N
R
δ
N −2 2
)=
cN c˜N 2−N 2 p(( +|y|2 )
p( |y|c˜NN−2 ) |y|1N
N −2 2
N
1 2 = )( +|y| 2) R = − N 2−2 c˜N
where the limit is obviously uniform in r ≥ 1. Finally, we prove (iii). We have R N −2 N −2 N R cN cN 2 ) = 2 p(( lim→0 − 2 Br U p(U ) = lim→0 |y|≤ √r ( +|y| 2) +|y|2 ) R c˜N c˜N =
N +2
U p(U ) = 2 [o(1) − c˜N + |x|N −2 p0 ( |x|c˜NN−2 )]dx] Br
N −2 ˜N 2 c
R
1 |x|N
[p( |x|c˜NN−2 )+
Proof of the theorem We have to show how to deriveR(2.1) from Lemma 3.1 and 4.1. We will limit ourselves to the ”vanishing case” P ( |x|N1 −2 ) = 0. Inserting the estimates (i) - (ii) of Lemma 4.1 in (3.4), we get N
S2 + (aN + o(1))( )N −2 − N r N +2 Z Z cN 2 N +2 1 1 1 2 2 [o(1) + |∇H| − − p( )] 2 N |x|N |x|N −2 E(V ) ≤
8
(4.2)
where o(1) are uniform in r ≥ R non depend on r. R 1 and aN does Since, by assumption, I1 := |∇H|2 − N1 |x|1N p( |x|N1 −2 ) > 0, we see from (4.2) that for smaller than some (r)
E(V ) ≤
S N
N 2
− N −2 (
I1 6−N cN 2 − N −2 ) 2 r
and in fact (r) can be taken independent on r if r ≥ 1. Now, (2.1), and hence the result, follows from the
Claim Let E(t V ) = maxt>0 E(tV ) , < (r). Then (t − 1)2 = o(1)(
N +2 2
+ ( )N −2 ) r
where o(1) is uniform with respect to r ∈ [1, ∞).
Assuming the claim, we have 1
d2 E((tt + (1 − t))V ) = dt2 0 N +2 = 0(1)(1 − t )2 = o(1)( 2 + ( )N −2 ) r
0 ≤ E(t V ) − E(V ) = −
Z
t
uniformly in r ≥ 1, and hence, from (4.3): N
S2 I1 6−N 1 + cN max E(tV ) ≤ − N −2 ( 2 − N −2 ) t>0 N 4 r for ≤ (r) ∀r, or ≤ 0 := inf r≥1 (r) > 0, if r ≥ 1. So N
S2 if N ≥ 7 then, ∀r > 0 : max E(tV ) < for < ˜(r) t 2 N S2 if 3 ≤ N ≤ 6 then max E(tV ) < for ≤ 0 and r ≥ r() t>0 N Proof of the Claim 9
(4.3)
Let us first observe that Z
|∇V |2 = S
Z
|V | N −2 = S
N 2
+ 0(1)(
N +2 2
1 + ( )N −2 ) 2 r
(4.4)
1 + ( )N −2 ) 2 r R R R R In fact, |∇V |2 = |∇U |2 + |∇w |2 + 2 ∇V ∇w = 2N
N 2
+ 0(1)(
N +2 2
(4.5)
R N +2 0(1)( r )N −2 + 0( 2 ) + 2 B√ p(U )V δ R 2N N −2 N +2 R and | p(U )V | ≤ LN 4 ( |V | N −2 ) 2N , for some pure constant LN . Also,
R
2N
|V | N −2 =
N +2 + N2N −2 N −2
R
R
2N
UN −2 +
2N N −2
R
4
|U + tw | N −2 (U + tw )w 4
Br ×[0,1]
(1 − t)|U + tw | N −2 w2 = S
N 2
+ 0(1)(
N +2 2
1
+ ( r )N −2 ) 2
by H˘ older inequality, Lemma 3.2-ii and Lemma 4.1-i. Also, all estimates in (4.4) - (4.5) are uniform in r ≥ 1. Now, from (4.4)-(4.5) and
0=
Z Z Z 4 2N d 1 p(t V )V ] E(tV )|t=t = t [ |∇V |2 − tN −2 |V | N −2 − t dt
we get 4
|1 − tN −2 | = 0(1)( R because | p(t V )V | ≤ LN |V |
N +2 2
N
2N N −2
1 + ( )N −2 ) 2 r
( 2 )
N +2 2N
This proves the claim, because |t − 1| ≤
4
N −2 N −2 2 |t
− 1|.
References . [AM] Adimurthi and Mancini, A sharp condition in higher dimensions for a Brezis Nirenberg type equation, in preparation
10
[AY] Adimurthi and Yadava, Critical Sobolev exponent problem in a ball with nonlinear perturbation changing sign, Advances in Differential equation 2,(1997), 161-182. [BG] Bernis F. and H. -Ch. Grunau, Critical exponents and multiple critical dimensions for polyharmonic operators, J.Differ Equations 117, (1995), 469486. [BN] Brezis H. and L. Nirenberg, Positive solutions of nonlinear elliptic equations involving critical exponents, Comm.Pure. Appl. math.36 (1983), 437-477. [PS] Pucci P. and J. Serrin, Critical exponents and critical dimensions for polyharmonic operators, J. Math. Pures at Appl. 69 (1990), 55-83.
11