Вы находитесь на странице: 1из 16

Applications of critical point theory

to variational problems on Rn

Antonio AMBROSETTI

Abstract
This paper is a expanded versionof the talk delivered at Equadiff 99. We will review
some recent advances in critical point theory and some of their applications to elliptic
problems on Rn .
The abstract set up deals with the existence of critical points of C 2 functionals of the
type
f (u) = f0 (u) + G(u), uE
where E is a Hilbert space and f0 possesses a finite dimensional, possibly non compact
manifold Z of critical points.
The abstract results provide a unified frame for a broad variety of variational problems.
Here we will focus on the problem of finding positive solutions of some elliptic equations on
Rn , including the prescribed scalar curvature problem.

1 Introduction: Some Motivation


Many problems arising in Nonlinear Analysis are perturbative in nature. We will present here-
after some examples that will be discussed in some detail in the sequel. They furnish a motivation
for the abstract setting that will be outlined in Section 2 below.

1.1 Bifurcation from the essential spectrum


Let us consider the problem of finding solutions of
(
00 + = h(x)||p1 , x R,
(B)
lim|x| (x) = 0,

which bifurcate from = 0, the infimum of the essential spectrum of on H 1 (Rn ). Here p is
any number greater than 1. Suppose that there exists ` > 0 such that h ` L1 (R) and make
the change of variable
u(x) = 2/(1p) (x/), = 2 ,
Eq. (B) becomes

(1) u00 + u = `|u|p1 u + (h(x/) `)|u|p1 u, u H 1 (R).

We can consider (1) as a perturbation of the problem

(2) u00 + u = `|u|p1 u, u H 1 (R).


SISSA, via Beirut 2-4, Trieste 34014 Italy. Supported by M.U.R.S.T. under the national project Variational

methods and nonlinear differential equations.

1
Let denote the unique positive radially symmetric solution of (2). Since such an equation is
autonomous, it possesses a continuum of solution, namely all the function ( + ), for all R.
If, for small, (1) has a solution u , then (B) has a family of nontrivial solutions , = 2
which branch off the trivial solution at = 0 (the norm such that 0 as 0 depends
on p).

1.2 Entire solutions of elliptic semilinear equations


As a second example, we consider the question of finding positive solutions of elliptic problems
on Rn such as
(
u + u = |u|p1 u + Du (x, u), x Rn ,
(E)
lim|x| u(x) = 0,

where > 0, n 3 and 1 < p < 2 1 = (n + 2)/(n 2).


When = 0 the unperturbed problem has again an n dimensional family of solutions like
in the preceding case. The problem is to find a possible Rn such that (E) has solutions u
which converge to v( + ) as 0. Problems (B) and (E) will be discussed in Section 3.

1.3 The scalar curvature problem


In the specific case that p = (n + 2)/(n 2), the critical Sobolev exponent, an equation like
(E) with = 0 arises in Differential Geometry in the classical problem of prescribing the scalar
curvature of a manifold.
Let (M, g0 ) be a compact n-dimensional riemannian manifold and let R0 denote its scalar
curvature. One seeks for a metric g conformally equivalent to g0 such that the scalar curvature
of (M, g) is a prescribed function R. We will consider hereafter the case that n 3 (when n = 2
the approach is slightly different). Letting g = u4/(n2) g0 , u > 0, one is led to find positive
solutions of (up to a positive unimportant constant)

(3) g0 u + R0 u = Ru(n+2)/(n2) , u H 1 (M ),

where g0 denotes the Laplace-Beltrami operator.


We will address the problem of prescribing the scalar curvature on the n sphere. More
precisely, we take (M, g0 ) to be the standard n sphere (S n , g0 ) and focus our attention on the
case in which the prescribed curvature is close to a positive constant, say R = 1 + K. Using
stereographic co-ordinates on S n , equation (3) becomes

(SC) u = (1 + K(x)) u(n+2)/(n2) , u D1,2 (Rn ).

The new feature of (SC) is that the unperturbed equation u = u(n+2)/(n2) involving the
critical Sobolev exponent (n + 2)/(n 2) is not merely translationally invariant, but possesses
an n + 1 family of solutions. Some recent results concerning (SC) will be outlined in Section 4
where we also discuss (i) when K is invariant under the action of various groups of symmetries.
Equation (SC) is a particular case of

(E) u = u(n+2)/(n2) + Du (x, u), u D1,2 (Rn ).

Some results dealing with (E) taken from [6, 7] are autlined in Section 5.
Although the preceding problems are usually faced by different tools, it can be convenient
to find a functional setting which permits to handle all of them. Actually, equations (B), (E),
(SC) and (E) have some common features:

2
they are variational, namely their solutions are the critical points of a functional f defined
on a Hilbert space E;
the unperturbed problems obtained putting = 0 inherit a symmetry (in the examples
this is due to the fact that they are autonomous problems);
the perturbations break the symmetry.

Recently we have introduced in [2, 3] a perturbative method in critical point theory that provides
a unified abstract frame for all the preceding problems (and for other ones as well). This abstract
set up is reviewed in Section 2.
The results discussed in this paper are taken from various papers, see [2, 3, 6, 7, 8], where
we refer for other results, complete proofs and more details. The rest of the paper contains the
following 4 sections:

Section 2. The abstract setting


2.1. A perturbation results in critical point theory
2.2. Further perturbation results
2.3. Bifurcation from the critical manifold
Section 3: Applications to subcritical elliptic problems
3.1. Solutions of (E)
3.2. Bifurcation of positive solutiosn of (B)
Section 4: The scalar curvature problem on S n
4.1. The general case
4.2. Curvatures close to a constant
4.3. The equivariant case
Section 5: Elliptic equations with critical exponent

2 The Abstract Setting


Motivated by the above examples, we discuss in this section a general abstract set up that can
be used to handle the preceding problems.

2.1 A Perturbation Results in Critical Points Theory


Let us consider a Hilbert space E and a family of functionals f C 2 (E, R) of the form

(4) f (u) = f0 (u) + G(u).

We assume that the unperturbed functional f0 satisfies the following assumptions:

(A1) f0 possesses a finite dimensional manifold Z of critical points; Z will be called critical
manifold;
(A2) D2 f0 (z) is a Fredholm map with index 0, for all z Z;
(A3) Ker(D2 f0 (z)) = Tz Z, for all z Z (Tz Z denotes the tangent space to Z at z).

3
Roughly, condition (A1) is due to the invariance of f0 under a symmetry; furthermore, f0 (z)
0 implies that D2 f0 (z)[] = 0, for all Tz Z, namely Tz Z Ker(D2 f0 (z)). Hence condition
(A3) amounts to requiring that if v is any solution of the linear equation D2 f0 (z)[v] = 0 then
v Tz Z. A critical manifold Z satisfying (A 2, 3) will be called nondegenerate.
Let = G|Z .

Theorem 2.1 [2] Let f C 2 (E, R) be of the form (4), where f0 satisfies (A 1,2,3) and suppose
that there exists a critical point z Z of such that one of the following conditions hold:
(i) z is nondegenerate;
(ii) z is a strict local minimum or maximum;
(iii) z is isolated and the local topological degree of 0 at z, degloc (0 , 0) is different from zero.
Then for || small enough, the functional f has a critical point u such that u z as 0.

The proof relies in 2 main steps. The first one is a sort of Lyapunov-Schmidt reduction (non-
linear and variational in nature): for all z Z one finds w = w(, z) (Tz Z) solving for ||
small the auxiliary equation

(5) f (z + w) Tz Z.

Letting Z = {z + w(, z)}, it turns out that Z is locally diffeomorphic to Z and the critical
points of f constrained on Z give rise to critical points of f . Since w(, z) 0 as 0, there
results
f (z + w(, z)) = b + (z) + o(), ( 0).
Finally, the conditions (i), (ii) or (iii) imply that f has on Z a critical point of the form
u = z + w(, z).

Remark 2.2 (a) The assumption that z is isolated can be removed. Of course, in such a case
we cannot localize the limit of u as 0.
(b) If z is a non-degenerate critical point of then u does for f and there results
(
m(f , u ) = m(f0 , z0 ) + m(, z) if > 0
m(f , u ) = m(f0 , z0 ) + m(, z) if < 0

Here m(, q) denotes the Morse index of q as a critical point of the functional , namely the
maximal dimension of the subspace where D2 (q) is negative defined.

Remark 2.3 The case in which the critical manifold Z is compact has been handled e.g. in [21,
22, 36, 37, 40, 5] and usually applied to the existence of periodic solutions of forced Hamiltonian
systems, see e.g. [25, 5]. In such a case f has always a maximum and a minimum on Z , unless
it is identically constant. Unlike in the forementioned papers, the new feature of Theorem 2.1
is that Z is possibly not compact.

2.2 Further perturbation results


In some cases, like when one deals with (B), the functional f has the more general form

(6) f (u) = f0 (u) + G(, u).

In such a case, in addition to (A 1,2,3) one makes the following hypotheses:

4
(A4) G is of class C 2 on R E;
b : Z 7 R such that G(, z) (z)
(A5) > 0 and b as 0, and u G(, z) = o(/2 ).

If (A 4,5) hold one can still carry out the preceding arguments yielding

f (z + w(, z)) = b + (z)


b + o( ), ( 0).

Then we obtain:

Theorem 2.4 [3] Let f0 C 2 (E, R) satisfy (A 1,2,3) and G satisfy (A 4,5). If
b has a strict
local maximum or minimum, then the same conclusion of Theorem 2.1 holds true.

For the proof we refer to [3, Sections 2, 3].

2.3 Bifurcation from the critical manifold


Theorem 2.1 is a bifurcation result. Actually, let us write our problem in the form F (, u) = 0,
where F (, u) := f (u). The set of trivial solutions of F = 0 is given by {0} Z. We can say
that z Z is a bifurcation point (for F = 0) if there exist sequences n 6= 0, and un E, such
that
n 0, un z , F (n , un ) = 0.
One can show

Theorem 2.5 ([5, Section 4],[4, Section 7]) A necessary condition for z to be a bifurcation
point for F = 0 is that (z ) = 0.

Theorem 2.1 gives a sufficient condition for bifurcation.

Remark 2.6 Under the condition of Theorem 2.1 we can prove that there is a connecetd set of
non trivial solutions branching off z . Indeed, one can also prove a kind of global bifurcation
result. But up to now we did not find any interesting application. We take this occasion to
mention that some bifurcation results of a recent paper of ours Branching points for a class of
variational operators, J. dAnalyse Math. 76 (1998), can be deduced by some previous work by
N.Dancer, Proc. London Math. Soc. 23 (1971) and 26 (1973), and by J.Ize, Mem. AMS vol.
174 (1976). We thank prof. N. Dancer who brought these references to our knowledge.

The preceding abstract results can be used to handle a great variety of nonlinear problems.
For example, one can use Theorem 2.1 to find homoclinic orbits of Duffing like systems as well
as to find semiclassical stationary states of nonlinear Schrodinger equations with a potential. In
the former case the auxiliary function is nothing but the so called Poincare function, namely
the primitive of the Melnikov function. It is worth mentioning that the preceding perturbation
arguments can be further carried out in order to prove the existence of a chaotic behaviour. For
this kind of problems and related questions, we refer to the recent papers [14, 15, 16]. As for
the Schrodinger equation, we refer to [4] and references therein.

3 Applications to Subcritical Elliptic Problems


In this section we will discuss the applications of Theorem 2.1 to equations like (E) and (B).

5
3.1 Solutions of (E)
We consider the problem

(E) u + u = |u|p1 u + Du (x, u), u H 1 (Rn ),

where > 0, n 3 and 1 < p < 2 1 := (n + 2)/(n 2) (subcritical case).


Solutions of (E) will be found as the critical points u H 1 (Rn ) of
Z
1 1
(7) f (u) = kuk2 |u|p+1 G(u)
2 p + 1 Rn

where
Z
1
kuk2 |u|2 + u2 dx

=
2 Rn
Z
G(u) = (x, u) dx.
Rn

Consider the unperturbed problem

(8) u + u = |u|p1 u, u H 1 (Rn ),

and let be its unique radial positive solution. In this case the critical manifold introduced in
Section 2 is given by
Z = {( + ) : Rn }.
It is well known that Z is non degenerate in the sense specified before that conditions (A 2, 3)
hold true. As for one finds
Z
() = (x, (x + )) dx.
Rn
n
Suppose R is, say, a maximum or a minimum of . Then, according to the abstract result,
(E) has a solution that bifurcates from ( + ).
It is worth pointing out that the f given by (7) does not verify the Palais-Smale, (PS) for
short, condition. This lack of (PS) is reflected here in the non-compactness of Z. The specific
feature of our approach is that, unlike the Concentration-Compactness principle introduced by
P.L. Lions [30], the (PS) condition is recovered whenever has suitable critical points. In general,
for perturbation problems like (E), our approach permits to work in a greater generality.
Some concrete examples will highlight the kind of results one can find.

Example 1. If (x, u) = h(x)u with h L2 (Rn ), one has that


Z
() = h(x )(x) dx.
Rn

Since there results lim|| () = 0 then has a maximum or a minimum provided there
exists such that ()
6= 0. If this is the case, an application of Theorem 2.1 yields a solution of

(9) u + u = |u|p1 u + h(x), > 0, u H 1 (Rn ).

Two more solutions of (9) can be found: one near the non degenerate critical manifold Z 0 =
{( + )} and a second one near u = 0 by means of the standard Implicit Function Theorem.

6
Example 2. If we take (x, u) = a(x)|u|p+1 + h(x)u, with a L1 (Rn ) and h L2 (Rn ), one
finds that Z  
1 p+1
() = a(x )|(x)| h(x )(x) dx.
Rn p + 1
We will assume that
(10) 1 > 2 > 0 such that a |x|1 , h |x|2 as |x| .
Theorem 3.1 [32] Let a L1 (Rn ), h L2 (Rn ) satisfy (10) and suppose that there exists Rn
> 0. Then
such that ()
u + u = ((1 a(x))up + h(x), > 0, u H 1 (Rn )
has at least four positive solutions, provided > 0 is sufficiently small.
Here has 2 minima and one maximum, that give rise to 3 solutions bifurcating from Z. A fourth
solution is again found near u = 0. Let us point out that one can allow a polinomial growth
of a, h. This would not be possible by a stright application of the Concentration-Compactness
principle. Theorem 3.1 improves [1], in the perturbative case. For other results in this direction,
we refer to [32].

3.2 Bifurcation of positive solutions of (B)


As a last application, we discuss the existence of solutions of (B) that bifurcate from the essential
spectrum. As we have seen in Subsection 1.1, after a change of variable, we are led to study
equation

u00 + u = `|u|p1 u + (h(x/) `)|u|p1 u, u H 1 (R),


which can be faced by means of Theorem 2.4. Actually, here the perturbation term has the more
general form Z
1
G(, u) = (h(x/) `)|u|p+1 (x) dx.
p+1 R
One shows that Theorem 2.4 applies provided, e.g.
Z
(h(x) `)dx 6= 0,
R

b : R R given by
with = 1 and
Z
1
()
b = (h(x) `)dx p+1 ().
p+1 R

Since, obviously,
b has a maximum at = 0, it follows that (B) possesses a family of nontrivial
solution (x) for = 2 0. If 1 < p < 5 these solution branch off from the trivial solution
= 0. For more details and other results dealing with the bifurcation from the essential
spectrum for equations like (B), see [31, 39, 3].
Other applications of the abstract results to elliptic equations can be found e.g. in [10, 11,
12, 24].

4 The Scalar Curvature Problem on S n


In this section we deal with the Scalar Curvature problem on the standard n-dimensional sphere
Sn.

7
4.1 The general case
According to the discussion made in Section 1, prescribing the scalar curvature R on S n is
equivalent, in stereographic coordinates to look for positive solutions of

(11) u = R(x)u(n+2)/(n2) , u D1,2 (Rn ).

Here and in the sequel we use the same symbol R to denote a function on S n and R 1 , where
: S n Rn denotes the stereographic projection. The context will make it clear which one is
the function we are dealing with.
The problem has first been raised by Nirenberg [35]. Necessary conditions for (11) to have
a solution has been found e.g. in [28], see also [27]. Roughly, existence results have been proved
under two sets of assumptions. A group of results deals with the case in which R is symmetric.
The first one is due to Moser.

Theorem 4.1 [34] If R is even on S 2 , then (11) has a solution if and only if maxS 2 R > 0.

When n = 3 the existence of a solution has been proved by escobar - Schoen in [23]. For
future reference, let us also mention another result by Chen - Li that deals with rotationally
symmetric curvatures.

Theorem 4.2 [20] If R C 2 is rotationally symmetric on S n and nondegenerate (i.e. R00 6= 0


whenever R0 = 0), then (11) has a solution if and only if maxS n R > 0 and R0 changes sign on
the set R > 0.

For other results dealing with the case that R is invariant under the action of a group, possibly
with fixed points, we refer e.g. to [19, 26].
In another group of papers the symmetry assumption is replaced by conditions involving the
critical points Q of R and the sign of R(Q). For example, the following result has been proved
by Bahri - Coron:

Theorem 4.3 [13] Let n = 3 and suppose that R satisfies:

(1) R C 2 has finitely many non-degenerate critical points xi with Morse index mi ;
(2) R(xi ) 6= 0
mi
P
(3) R(xi )<0 (1) 6= 1.

Then there exists on the standard sphere S 3 a conformal metric with scalar curvature equal to
R.

Dealing with dimensions n 4 the situation becomes more complicated. In general, one has to
require a flatness condition at the critical points Q of R: the derivatives D R(Q) = 0 for all
n 2. In particular, when n > 3, D2 R(Q) = 0 at the critical points of R, which therefore
are degenerate. An assumption of this sort has been first introduced by Escobar - Schoen in
[23] to handle the Scalar Curvature problem on S n when R is invariant under the action of a
free group action.
Similarly, Yanyan Li has extended the Bahri - Coron result to any dimension n 4
assuming that there exists , n 2 < < n such that, in local coordinates
n
X
(12) R(P ) = R(Q) + aj |Pj Qj | + o(|P Q| ), P Q.
1

where P = (P1 , . . . , Pn ) and Q = (Q1 , . . . , Qn ).

8
P
Theorem 4.4 [29] Let (12) hold and suppose that aj 6= 0 and aj 6= 0. Then (11) has a
solution provided X
(1)i(Q) 6= (1)n .
R(Q)<0,
P a <0
j

where i(Q) = ]{aj < 0}.

Remark 4.5 It is worth pointing out that in [29] it is also shown that if we do not assume a
flatness condition, the hypotheses of Theorem 4.3 (with 1 substituted by (1)n ) do not suffice
any more for the existence of a solution when n > 3, and the Scalar Curvature Problem can
have no solution at all.

4.2 Curvatures close to a constant


Here we outline some recent result from [6]. One goal will be to show that the preceding flatness
condition is no more necessary when R is close to a constant. Precisely we take R = 1 + K so
that (11) becomes the equation (SC) already introduced in Section 1,

(SC) u = (1 + K(x))u(n+2)/(n2) , u D1,2 (Rn )

We will merely give an idea of the arguments needed to prove the following extension of Theorem
4.3

Theorem 4.6 [6] Let (1 2) of Theorem 4.3 hold and suppose that
X
(13) (1)mi 6= (1)n .
K(xi )<0

Then, for || small, (SC) has a solution.

Let us sketch the proof that is carried out in three steps.

Step 1. Applying the abstract setting.

Consider the unperturbed problem

(14) u + u(n+2)/(n2) = 0, u D1,2 (Rn )

As anticipated before, (14) is not merely translationally invariant, but it also inherits an addi-
tional symmetry. Actually if z0 (x) = (1 + |x|2 )(n2)/2 denotes, up to a positive constant, the
unique radial positive solution of (14), then the critical manifold Z is now given by

Z = {z, : > 0, Rn }

where  (n2)/2
(2n)/2 x
z, (x) := z0 ( )= .
2 + |x |2
It is possible to prove that Z is still nondegenerate. Moreover, has the following form
Z Z
1 2 1
(, ) = K(x)z, (x)dx = K(y + )z02 (y)dy.
2 Rn 2 Rn

9
The new feature is that in the present case Z is open and has a boundary, namely Z = {(0, ) :
Rn }. This creates some problem. However, a control of the behaviour of as 0 allows
us to use again the abstract results to obtain solutions of (14).
First of all, can be extended by continuity to = 0 and by eveness to all of R Rn . Actually,
there exist c0 > 0 such that
(
(0, ) = lim0 (, ) = c0 K(),
(15)
D (0, ) = 0.

We will use the same notation to denote this extension. As a consequence possesses critical
points of the type Xi = (0, xi ) for every critical point xi of K. The key observation is that
the Morse index Mi of Xi is related to the Morse index mi of xi in dependence to the sign of
K(xi ). Precisely, if K(xi ) 6= 0 then Xi is non-degenerate. Furthermore one has:

mi if K(xi ) > 0,
(16) Mi =
mi + 1 if K(xi ) < 0.

To deduce (16) it suffices to evaluate the second derivatives of at points of the type (0, ).
Actually one has that D2 (0, ) = 0. Furthermore, since there results
Z

2 2
D (0, ) = Dij K()xi xj z02 (x)dx,

then we infer

(17) D2 (0, ) = c1 K(), (c1 > 0),

and hence

D2 K()

0
D2 (0, ) =
0 c1 K()
This shows that (16) holds. Now, let us write the Morse equalities for K and .

Step 2. Morse equality for K.

Without loss of generality, we can choose the north pole PN of the stereographic projection in
such a way that K(pN ) = minS n K. Then, passing to stereographic co-ordinates, there exists
R > 0 such that K(x) x < 0 for |x| = R. Thus the Morse relationship for K becomes
X
(1)n = (1)mi .

Since K(xi ) 6= 0, the right hand side can be split:


X X
(18) (1)n = (1)mi + (1)mi .
K(xi )>0 K(xi )<0

Step 3.Morse equality for .

First of all, one shows that (X) X < 0 as |X| = R  1, where X = (0, ). Next, suppose
that Xi = (0, i ) are the only critical points of . Then is a Morse function on Rn+1 and there
results

10
X X X
(1)n+1 = (1)Mi = (1)Mi + (1)Mi .
K(xi )>0 K(xi )<0

Using (16) we infer


X X
(1)n+1 = (1)mi (1)mi .
K(xi )>0 K(xi )<0

This, together with (18) implies


X
(1)mi = (1)n ,
K(xi )<0

in contrast with the assumption (4.6). Sharpening the preceding arguments, one can prove that
with > 0 at which Theorem 2.1 applies. This proves Theorem
has a critical point X = (, )
4.6.
A slight modification of the preceding arguments leads to prove:

Theorem 4.7 [6] Let K C 1 (Rn ) have a finite number of critical points i and suppose there
are i [2, n[ and i -homogeneous functions Qi = Qi ,i such that

(19) K(x) = K(i ) + Qi (x i ) + o(|x i |i , x i .



Moreover, we assume that Ai := Qi (y)z02 (y)dy 6= 0 for all i and that
R

X
degloc (K, i ) 6= (1)n .
Ai <0

Then, for || small,(SC) has a solution.

Remark 4.8 Concerning curvatures close to a positive constant, we mention the papers [18, 29].
In the former it is assumed that (19) holds for some < n, if n is even and < n 1, if n
is odd. Furthermore a technical condition on an auxiliary function is required. Other technical
hypothesis involving Qi are also made in [29], see Theorem 6.1 and Corollary 6.1 in [29, Part I].
The proofs employ arguments different than ours.

We end this subsection mentioning the recent paper [33] where a Morse theory for scalar
curvatures near to a positive constant has been discussed.

4.3 The equivariant case


Here we briefly report some results of a recent paper [8] dealing with (SC) in the case that K is
invariant under the action of a group of symmetries.
We suppose that there is a compact group O(n) acting on S n and that K is -invariant.
Here K denotes the function on S n . Let us remark that, in stereographic coordinates, the
symmetries to be considered would be different. For example, if K(P ) = K(P ), P S n , then
for the corresponding K on Rn there results K(x) = K(x/|x|2 ).
The idea is to use the abstract setting in connection with the observation that one can find a
e related to , acting on (0, +) Rn under which and f (z + w(, z)) are invariant.
group ,
Referring to [8] for details, we limit ourselves to indicate the relationship between the fixed
points of and those of e in some specific cases:

1) if Fix() = then Fix()


e = {(1, 0)};

11
2) if Fix() = {P , P )} then Fix()
e = {(, 0) : > 0};

3) if Fix() = S nk then Fix()


e = (0, +) Vnk where Vnk = { Rn : 1 = . . . = k = 0}.

In item 2) above, we have taken the north pole of the stereographic projection to be one of the
fixed points, say PN = P . Similarly, in item 3) we have chosen PN to belong to the fixed point
set S nk . Obviously, it suffices to look for critical points of restricted to Fix()
e and this
allows us to obtain rather general results. For example, if Fix() = then (1, 0) is a critical
point of as well as of f on Z and it immediately follows that (SC) has a solution. Dealing
with 2) and 3) we consider constrained on Fix(). e In the former, a solution exists provided

we impose some conditions on K at P and P . For example one has:
Theorem 4.9 Suppose that Fix() = {P , P )}. Then for || small (SC) has a solution
provided K C 2 (S n , R) satisfies one of the following two conditions:

(a) K(P ) = K(P );

(b) K(P ) < K(P ), resp. K(P ) > K(P ), and K(P ) < 0, resp. K(P ) > 0.
Here and in the sequel, we always assume that K C 2 even though in some cases, as in (a)
above, less regularity suffices.
The specific case that K has a symmetry with respect to the axis passing through the north
and south poles (namely when K on Rn is radial) has been studied in [6].
Unlike the preceding Theorem, where the assumptions are made on the function K defined
on S n , dealing with the case 3) it is convenient to work directly with K : Rn R. Since
e = (0, +) Vnk , we can look for critical points of constrained on such subspace. It
Fix()
turns out that we can repeat the argument used in the proof of Theorem 4.6 yielding:

Theorem 4.10 Let Fix() = S nk and suppose that K C 2 (Rn ) satisfies


(K) K has a finite number of critical points i such that i Vnk . Moreover these i are
nondegenerate for K constrained to Vnk and K(i ) 6= 0.
Then for || small (SC) has a solution provided
X
(20) (1)mi =6 (1)nk mi = m(K|Vnk , i ).
K(i )<0

Remark 4.11 In the limit cases k = 0 and k = n, Theorem 4.10 is nothing but Theorem 4.6,
and Theorem 4.9 respectively.

The following Theorem is related to a result of [19].

Theorem 4.12 Suppose that K C 2 (Rn ) achieves the maximum on Vnk , k > 0, at a point
0 such that K(0 ) > 0. Then (SC) has a solution provided || is small.
2
Actually, (0, 0 ) = maxVnk (0, ). Moreover, K(0 ) > 0 implies D (0, 0 ) > 0, see
(17). Since, as remarked before, (X) X < 0 as |X| = || + || , (, ) R Vnk ,
then achieves its global maximum on [0, ) Vnk at some point (, ) with > 0. Using
Theorem 2.1 the result follows.
The assumptions made in Theorem 4.12 agree with those of Theorem 4.2. Let us also point
out that, obviously, we must take k > 0, otherwise Vnk = Rn and K(0 ) 0 at the maximum
0 of K.

12
Remark 4.13 In the above Theorems of this subsection we find symmetric solutions. In general,
this is no more true for curvatures which are not close to a constant. For example, it is shown
in Theorem 3 of [20] that there exists a family of nondegenerate radially symmetric functions
R satisfying the assumptions of Theorem 4.2 but such that (SC) has no radially symmetric
solutions at all. Unlike our R, this family R is obtained by perturbation of a strictly monotone
function R1 .

For other results dealing with equivariant equation with critical Sobolev exponent, see also [17].

5 Elliptic equations with critical exponent


Equation (SC) is a particular case of (E) in the case that p = (n + 2)/(n 2), the critical Sobolev
exponent, and = 0:

(E) u = u(n+2)/(n2) + Du (x, u), u D1,2 (Rn ).


1
Here we will look for positive solutions of (E). The specific case in which (x, u) = q+1 h(x)uq+1
with 1 q < (n + 2)/(n 2), resp. 0 < q < 1, has been faced in [6, Sections 5 and 6] and [7],
respectively. For example, when q = 1 one finds:

Theorem 5.1 [6] The problem

(21) u = h(x)u + u(n+2)/(n2) u D1,2 (Rn ),

has, for || small, a positive solution provided that supp[h] is compact and either
(h1 ) n > 4, h C 2 (Rn ), and h(x) 6 0,
or
h L1 (Rn )
R
(h2 ) n 3, and h(x)dx 6= 0.

Roughly, one applies the abstract setting with


Z Z
1 2 1
(, ) = h(x)z, (x)dx = 2 h(y+ )z02 (y)dy.
2 Rn 2 Rn

It turns out that in the present case can be extended to = 0 by setting (0, ) = 0 and there
still results (, ) 0 as + || +.
Moreover, if (h1 ) holds and is such that h( ) 6= 0, then
Z Z
2 2
2 (, 0) = h(y+ )z0 (y)dy h( ) z02 (y)dy as 0.
Rn Rn

Similarly, if (h2 ) holds, there results


Z Z
2n2 (, 0) = h(x)z02 (x/)dx z02 (0) h(x)dx as +.
Rn Rn

In both cases 6 0 and achieves the minimum or the maximum at some (, ) Rn+1 with
> 0. An application of Theorem 2.1 yields the result.

13
Remark 5.2 (a) The assumption that supp[h] is compact is used only to prove that the solutions
bifurcating from Z are positive. On the other side, we do not assume that h 0. Indeed, if (h1 )
holds and h changes sign, the above arguments imply that achieves both the minium and the
maximum on (0, ) Rn and hence (21) possesses at least two positive solutions.
(b) In general, the preceding solutions cannot be found by means of the Mountain-Pass
Theorem. For example, if > 0 and has a minimum, then, according to Remark 2.2-(b), the
corresponding solution of (21) has Morse index equal to n + 2.

When 1 < q < (n + 2)/(n 2) the preceding result can be improved eliminating the assumption
that supp[h] is compact.

Theorem 5.3 [6] Let 1 < q < (n +2)/(n 2) and suppose that h L1 (Rn ) L (Rn ), h(x) 6 0.
Then, for || small, (E) has a positive solution.

In the case 0 < q < 1 the functional f is no more C 2 and the abstract setting needs to be
modified. However, one can prove an existence result similar to the first statement of Theorem
5.1 above:

Theorem 5.4 [7] Let n 3 and suppose that h C(Rn , R) has compact support and is positive
somewhere. Then, for || small, (E) with 0 < q < 1 has a positive solution.

Remark 5.5 (a) The solution found in the preceding Theorem is close to the critical manifold
Z. In addition to this solution, when 0 < q < 1 equation (E) has another solution u b with
b 0 as 0, that can be found by minimizing the energy functional f , see [7]. For the
u
existence of this u
b we do not need to require that h has compact support. But, unlike in
Theorem 5.4, to show that u b is positive we have to assume that > 0 (small) and h 0.
(b) If h L1 L , h 0 and > 0 is small, a second positive solution of (E) has been
found in [7] by the Mountain Pass Theorem. The same procedure can be used in Theorem
5.1. In both cases the support of h can possibly be not compact. The lack of (PS) is bypassed
by using the Concentration Compactness method. Using this tool, we have been not able to
prove the existence of a positive solution when h is not 0, or when < 0. From this point of
view, Theorems 5.1 and 5.4 are other examples in which our perturbation approach allows us
to recover the (PS) condition in a case in which the Concentration Compactness method seems
not to work.
(c) The equation (SC) makes sense also when K is not the stereographic projection of a
function defined on S n . Existence of solutions for this or, more in general, for a problem like

u = h(x)uq + (1 + K(x))u(n+2)/(n2) u D1,2 (Rn ),

are discussed in [6].

References
[1] S. Adachi - K. Tanaka, Four positive solutions for the equation u + u = a(x)up + f (x)
in Rn , Preprint (1998).
[2] A. Ambrosetti - M. Badiale, Homoclinics: Poincare-Melnikov type results via a variational
approach, Ann. Inst. H. Poincare Analyse Non Lineaire 15 (1998) 233-252. Preliminary note
on C. R. Acad. Sci. Paris 323, Serie I (1996), 753-758.
[3] A. Ambrosetti - M. Badiale, Variational Perturbative methods and bifurcation of bound
states from the essential spectrum, Proc. Royal Soc. Edinburgh, 128 A (1998), 1131-1161.

14
[4] A. Ambrosetti - M. Badiale - S. Cingolani, Semiclassical states of nonlinear Schrodinger
equations, Arch. Rat. Mech. Anal. 140 (1997), 285-300.
[5] A. Ambrosetti - V. Coti Zelati - I. Ekeland, Symmetry breaking in Hamiltonian systems,
J. Diff. Equat. 67 (1987) 165-184.
(N +2)
[6] A. Ambrosetti - J. Garcia Azorero - I. Peral, Perturbation of u + u (N 2) = 0, the scalar
curvature problem in RN and related topics, J. Funct. Anal. 165 (1999), 117-149.
[7] A. Ambrosetti - J. Garcia Azorero - I. Peral, Elliptic variational problems in Rn with critical
growth, Preprint S.I.S.S.A. - Trieste (1999).
[8] A. Ambrosetti - A. Malchiodi, On the symmetric scalar curvature problem, to appear on J.
Diff. Equat. (1999).
[9] Th. Aubin, Some Nonlinear Problems in Differential Geometry, Springer, 1998.
[10] M. Badiale - A. Duci, Concentrated solutions for a non variational semilinear elliptic equa-
tion, Preprint S.N.S. Pisa (1999).
[11] M. Badiale - J. Garcia Azorero - I. Peral, Perturbation results for an anisotropic Schrodinger
equation via a variational method, to appear on Non. Diff. Eq. Appl.
[12] M. Badiale - B. Pellacci - S. Villegas, Elliptic problems in Rn with jumping nonlinearities:
perturbation results, to appear on Diff. Int. Equat.
[13] A. Bahri - J.M. Coron, The scalar curvature problem on the standard three dimensional
sphere, J. Funct. Anal. 95 (1991) 106-172.
[14] M. Berti - P. Bolle, Homoclinics and chaotic behaviour for perturbed second order systems,
Ann. Mat. Pura Appl. 176 (1999), 323-378.
[15] M. Berti - P. Bolle, Variational construction of homoclinics and chaos in presence of a
saddle-saddle equilibrium, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 27 (1998), 331-377.
[16] M. Berti - C. Carminati, Chaotic dynamics for perturbations of infinte dimensional hamil-
tonian systems, Preprint S.I.S.S.A. (1999).
[17] G. Bianchi - J. Chabrowski - A. Szulkin, On symmetric solutions of an elliptic equation with
a nonlinearity involving critical Sobolev exponent, Nonlin. Anal. TMA 25 (1995),41-59.
[18] S.A. Chang - P. Yang, A perturbation result in prescribing scalar curvature on S n , Duke
Math. J. 64-1 (1991) 27-69.
[19] W. Chen, Scalar Curvatures on S n , Math. Annalen 283 (1989), 353-365.
[20] W. Chen - C. Li, A necessary and sufficient condition for the Nirenberg problem, Comm.
Pure Appl. Math. 48 (1995), 657-667.
[21] D. Chillingworth, Bifurcation from an orbit of symmetry, in Singularities and Dynamical
Systems, S.N. Pnevmatikos Ed., pp. 285-294, North holland, 1985.
[22] E.N. Dancer, An implicit function theorem with symmetries and its applications to nonlinear
eigenvalue problems, Bull. Austr. Math. Soc. 21 (1980), 81-91.
[23] J. Escobar - R. Schoen, Conformal metrics with prescribed scalar curvature, Invent. Math.
86 (1986) 243-254.

15
[24] J. Garcia Azorero - E. Montefusco - I. Peral, Bifurcation for the p-laplacian in RN , to
appear on Advances in Diff. Eq.
[25] J.K. Hale - P.Z. Taboas, Interaction of damping and forcing in a second order equation,
Nonlin. Anal. 2 (1978), 77-84.
[26] Hebey E.: Changements de metriques conformes sur la sphere. Le probleme de Nirenberg,
Bull. Sci. Math. 114 (1990), 215-242.

[27] J. Kazdan, Prescribing the curvature of a Riemannian manifold, Reg. Conf. Series in Math.
57, A.M.S., 1984.
[28] J. Kazdan - F. Warner, Curvature functions for compact two manifolds, Annals of Math.
99 (1974), 14-47.
[29] Yanyan Li: Prescribing scalar curvature on S n and related topics, Part I, J. Diff. Equat.
120 (1995) 319-410; and Part II, Existence and compactness, Comm. Pure App. Math. 49
(1996), 437-477.
[30] P.L. Lions, The concentration-compactness principle in the calculus of variations. The limit
case, part 1; and Part 2, Rev. Matematica Iberoamericana 1-1 (1985), 541-597, and 1-2
(1985), 45-121.
[31] R.J. Magnus, On perturbations of translationally invariant differential equation, Proc. Roy.
Soc. Edinburgh, 110 (1988), 1-25.
[32] A. Malchiodi, Multiple positive solutions of some elliptic equations in Rn , to appear on J.
Nonlin. Anal. TMA.
[33] A. Malchiodi, The scalar curvature problem on S n : an approach via Morse theory, preprint
S.I.S.S.A. - Trieste (July 1999).
[34] J. Moser, On a nonlinear problem in Differential Geomerty, in Dynamical Systems, M.
Peixoto Ed., Academic Press, New York, 1973.
[35] L. Nirenberg, Monge-Ampere equations and some associated problems in geometry, Pro-
ceedings of the International Congress of Mathematicians, Vancouver (1974) 275-279.
[36] L. Recke - D. Peterhof, Abstract forced symmetry breaking and forced frequency locking of
modulated Waves, J. Diff. Equat. 144-2 (1998) 233-262.
[37] M. Reeken, Stability of critical points under small perturbation, Part I and II, Manus.
Math. 7 (1972), 387-411, and 8 (1973), 69-92.
[38] R. Schoen, On the number of constant scalar curvature metrics in a conformal class, in
Differential Geometry: a Symposium in honor of M. do Carmo, Wiley (1991), 311-320.
[39] C. Stuart, Bifurcation of homoclinic orbits and bifurcation from the essential spectrum,
SIAM J. Math. Anal. 20 (1989), 1145-71.
[40] A. Vanderbauwhede, Note on symmetry and bifurcation near families of solutions, J. Diff.
Equat. 47 (1983) 99-106.

16

Вам также может понравиться