Professional Documents
Culture Documents
PDEs in Riemannian
Geometry
Olivier Druet
Emmanuel Hebey
eric
Robert
Fred
2004
2003064801
Contents
Preface
Chapter 1. Background Material
1.1
1.2
Riemannian Geometry
Basics in Nonlinear Analysis
Palais-Smale Sequences
Strong Solutions of Minimal Energy
Strong Solutions of High Energies
The Case of the Sphere
A Fundamental Estimate: 1
A Fundamental Estimate: 2
Asymptotic Behavior
vii
1
1
7
13
14
17
19
23
25
26
32
45
51
52
54
67
68
72
83
88
143
182
201
209
Bibliography
213
Preface
Let (M, g) be a smooth compact Riemannian manifold of dimension n 3. We
let (h ) be a sequence of smooth functions on M and consider equations like
g u + h u = u2
(E )
(E )
the sum in the right hand side of (I1 ) is a finite sum over i, Bi is a bubble,
obtained by rescaling fundamental positive solutions of the Euclidean equation
u = u2 1 , and the R s are lower order terms which converge strongly to 0
in H12 (M ). This asymptotic description provides a very nice H12 -theory for the
asymptotic behavior of solutions of equations like (E ). Let us assume now that
the h s converge C 0, to h for some 0 < < 1. An important issue in the study
of equations like (E ) is to get a theory in which the above asymptotic description
holds also in the C 0 -space, where pointwise estimates are involved. Developing
such a C 0 -theory is the purpose of these notes.
We wanted these notes to be as self-contained as possible. They should be accessible to graduate students and researchers in other fields. For the sake of clearness,
we decided to present our theory in the specific case of equations like (E ). However, the material we present in these notes is applicable to more general equations.
viii
PREFACE
The Authors
Paris, July 2003
Chapter One
Background Material
We recall in this chapter basic facts concerning Riemannian geometry and nonlinear
analysis on manifolds. For reasons of length, we are obliged to be succinct and partial. Possible references are Chavel [20], do Carmo [22], Gallot-Hulin-Lafontaine
[36], Hebey [43], Jost [50], Kobayashi-Nomizu [53], Sakai [65], and Spivak [72].
As a general remark, we mention that Einsteins summation convention is adopted:
an index occurring twice in a product is to be summed. This also holds for the rest
of this book.
1.1 RIEMANNIAN GEOMETRY
We start with a few notions in differential geometry. Let M be a Hausdorff topological space. We say that M is a topological manifold of dimension n if each point of
M possesses an open neighborhood that is homeomorphic to some open subset of
the Euclidean space Rn . A chart of M is then a couple (, ) where is an open
subset of M , and is a homeomorphism of onto some open subset of Rn . For
y , the coordinates of (y) in Rn are said to be the coordinates of y in(, ).
An atlas of M is a collection of charts (i , i ), i I, such that M = iI i .
Given an atlas (i , i )iI , the transition functions are
j 1
: i (i j ) j (i j )
i
with the obvious convention that we consider j 1
i if and only if i j = .
The atlas is then said to be of class C k if the transition functions are of class C k ,
and it is said to be C k -complete if it is not contained in a (strictly) larger atlas of
class C k . As one can easily check, every atlas of class C k is contained in a unique
C k -complete atlas. For our purpose, we will always assume in what follows that
k = + and that M is connected. One then gets the following definition of a
smooth manifold: A smooth manifold M of dimension n is a connected topological manifold M of dimension n together with a C -complete atlas. Classical
examples of smooth manifolds are the Euclidean space Rn itself, the torus T n , the
unit sphere S n of Rn+1 , and the real projective space Pn (R).
Given two smooth manifolds, M and N , and a smooth map f : M N from
M to N , we say that f is differentiable (or of class C k ) if for any charts (, ) and
)
the map
(,
of M and N such that f () ,
1
f : () (
)
k
is differentiable (or of class C ). In particular, this allows us to define the notion of
diffeomorphism and the notion of diffeomorphic manifolds.
CHAPTER 1
T
(, ) some chart at x, of associated coordinates xi , we define x
x (M )
i x
by, for any f Fx ,
(f ) = Di f 1 (x) .
xi x
As a simple remark, one gets that the x
s form a basis of Tx (M ). Now, one
i x
defines the tangent bundle of M as the disjoint union of the Tx (M )s, x M . If
M is n-dimensional, one can show that T (M ) possesses a natural structure of a
2n-dimensional smooth manifold. Given a chart (, ) of M ,
Tx (M ),
x
BACKGROUND MATERIAL
to the construction of the tangent bundle, one can define the cotangent bundle of
a smooth manifold M as the disjoint union of the Tx (M ) s, x M. In a more
general way, one can define Tpq (M ) as the disjoint union of the Tpq Tx (M ) s,
where Tpq (Tx (M )) is the space of (p, q)-tensors on Tx (M ). Then Tpq (M ) possesses
a natural structure of a smooth manifold of dimension n 1 + np+q1 . A map
T : M Tpq (M ) is then said to be a (p, q)-tensor field on M if for any x
M , T (x) Tpq Tx (M ) . It is said to be of class C k if it is of class C k from
the manifold M to the manifold Tpq (M ). Given two manifolds M and N , a map
f : M N of class C k+1 , and a (p, 0)-tensor field T of class C k on N , one can
define the pullback f T of T by f , that is, the (p, 0)-tensor field of class C k on M
defined for x M and X1 , . . . , Xp Tx (M ), by
f T )(x) X1 , . . . , Xp = T f (x) f (x) X1 , . . . , f (x) Xp .
We now define the notion of a linear connection. Denote by (M ) the space of
differentiable vector fields on M . A linear connection D on M is a map D :
T (M ) (M ) T (M ) which satisfies a certain number of propositions. In local
coordinates, given a chart (, ), this is equivalent to having n3 smooth functions
kij : R, that we refer to as the Christoffel symbols of the connection in
(, ). They characterize the connection in the sense that for X Tx (M ), x ,
and Y (M ),
Y j
j
i
i
+ i (x)Y (x)
DX (Y ) = X (i Y )(x) = X
xi x
xj x
where the X i s and Y i s denote the components of X and Y in the chart (, ),
and for f : M R differentiable at x,
f
= Di f 1 (x) .
xi x
As one can easily check, the kij s are not the components of a (2, 1)-tensor field.
An important remark is that linear connections have natural extensions to differentiable tensor fields. Given a differentiable (p, q)-tensor field, T , a point x in M ,
X Tx (M ), and a chart (, ) of M at x, DX (T ) is the (p, q)-tensor on Tx (M )
defined by DX (T ) = X i (i T )(x), where
j1 ...jq
p
Ti1 ...ip
j ...j
j1 ...jq
i T (x)i11...ipq =
iik (x)T (x)i1 ...ik1 ik+1 ...ip
xi
x
k=1
q
j ...j
jk+1 ...jq
k=1
The covariant derivative commutes with the contraction in the sense that
DX Ckk12 T = Ckk12 DX (T )
where Ckk12 T stands for the contraction of T of order (k1 , k2 ). Given a (p, q)-tensor
field of class C k+1 , T , we let T be the (p + 1, q)-tensor field of class C k whose
components in a chart are given by
j1 ...jq
j1 ...jq
T i ...i
= i1 T i2 ...ip+1 .
1
p+1
CHAPTER 1
2f
xi xj
x
2
= Dij
f 1 (x) .
In the Riemannian context, f is called the Hessian of f and is sometimes denoted by Hess(f ).
The torsion T of a linear connection D can be seen as the smooth (2, 1)-tensor
field on M whose components in any chart are given by the relation Tijk = kij kji .
One says that the connection is torsion-free if T 0. The curvature R of D can
be seen as the smooth (3, 1)-tensor field on M whose components in any chart are
given by the relation
l
=
Rijk
lji
lki
l
+ lj
ki k ji .
xj
xk
l
l
As one can easily check, Rijk
= Rikj
. Moreover, when the connection is torsionfree, one has that
l
l
l
Rijk
+ Rkij
+ Rjki
= 0 and
b
d
d
L() =
dt
g((t))
,
dt t
dt t
a
d
where ( d
dt )t T(t) (M ) is such that ( dt )t f = f (t) for any f : M R
differentiable at (t). If is piecewise C 1 , the length of may be defined as the
sum of the lengths of its C 1 pieces. For x and y in M , let Cxy be the space of
piecewise C 1 curves : [a, b] M such that (a) = x and (b) = y. Then
dg (x, y) = inf L()
Cxy
BACKGROUND MATERIAL
2
xi x
xj x
xm x
where the g ij s are such that gim g mj = ij . Let R be the curvature of the LeviCivita connection as introduced above. One defines
1. the Riemann curvature Rmg of g as the smooth (4, 0)-tensor field on M
,
whose components in a chart are Rijkl = gi Rjkl
2. the Ricci curvature Rcg of g as the smooth (2, 0)-tensor field on M whose
components in a chart are Rij = Rij g , and
3. the scalar curvature Sg of g as the smooth real-valued function on M whose
expression in a chart is Sg = Rij g ij .
As one can check, in any chart,
Rijkl = Rjikl = Rijlk = Rklij ,
and the two Bianchi identities are
Rijkl + Riljk + Riklj = 0 ,
i Rmg jklm + m Rmg jkil + l Rmg jkmi = 0 .
In particular, the Ricci curvature is symmetric, so that in any chart Rij = Rji .
Given a smooth Riemannian manifold (M, g), and its Levi-Civita connection D,
a smooth curve : [a, b] M is said to be a geodesic, if for all t,
d
D d
= 0.
dt
dt t
This means again that in any chart, and for all k,
k
(t) + kij (t) i (t) j (t) = 0 .
CHAPTER 1
chart are such that gij (x) = ij , with the additional property that the Christoffel
symbols kij of the Levi-Civita connection in this chart are such that kij (x) =
0. The coordinates associated with this chart are referred to as geodesic normal
coordinates.
Given a smooth Riemannian n-manifold (M, g), one can define a natural positive
Radon measure on M . In particular,
can be ap
the theory of the Lebesgue integral
plied. For some atlas of M , i , i iI , we shall say that a family j , j , j jJ
is a partition of unity subordinate to i , i iI if the following holds:
(i) (
j )j is a smooth partition of unity subordinate to the covering (i )i ,
(ii) j , j j is an atlas of M , and
(iii) for any j, suppj j .
As one can easily check, for any atlas i , i iI of M , there exists a partition
of unity j , j , j jJ subordinate to i , i iI . One can then define the Riemannian measure as follows:
Given a continuous map f : M R with compact
support, and an atlas i , i iI of M ,
j |g|f 1
f dv(g) =
j dx
M
jJ
j (j )
where j , j , j jJ is a partition of unity subordinate to i , i iI , |g| stands
of g in
for the determinant of the matrix whose elements are the components
j , j , and dx stands for the Lebesgue volume element of Rn .One can
prove
that such a construction does not depend on the choice of the atlas i , i iI and
the partition of unity j , j , j jJ .
The Laplacian acting on functions of a smooth Riemannian manifold (M, g) is
the operator g whose expression in a local chart of associated coordinates xi is
2
u
u
.
kij
g u = g ij
xi xj
xk
For u and v of class C 2 on M , one then has the following formula for integration
by parts:
(uv) dv(g) =
u g v dv(g)
g u vdv(g) =
M
One has that ig > 0 for a compact manifold, but it may be zero for a complete
noncompact manifold. In a similar way, one can define the cut locus Cx of x,
where
Cx is a subset of M , and prove that Cx has measure zero, that ig (x) = dg x, Cx ,
BACKGROUND MATERIAL
1
(M ) =
Sg dv(g) ,
4 M
and when the dimension of M is 4, as shown by Avez [8],
1
1
1 2
2
2
(M ) =
|Wg | + Sg |Eg | dv(g) ,
16 2 M 2
12
where | | stands for the norm associated with g for tensors, and where Wg and Eg
are, respectively, the Weyl tensor of g and the traceless Ricci tensor of g. In a local
chart, the components of Wg are
1
Rik gjl + Rjl gik Ril gjk Rjk gil
n2
Sg
gik gjl gil gjk
+
(n 1)(n 2)
Wijkl = Rijkl
where n stands for the dimension of the manifold. As another striking example of
the relationship that exists between curvature and topology, one can refer to Hamiltons theorem [39]: any three-dimensional, compact, simply connected Riemannian
manifold of positive Ricci curvature must be diffeomorphic to the unit sphere S 3 .
Conversely, by recent results of Lohkamp [59], negative sign assumptions on the
Ricci curvature have no effect on the topology, since any compact manifold possesses a Riemannian metric of negative Ricci curvature. As another example of the
relationship that exists between curvature and topology, we refer to the well-known
sphere theorem of Berger [10], Klingenberg [51, 52], Rauch [61], and Tsukamoto
[75].
1.2 BASICS IN NONLINEAR ANALYSIS
Given a smooth compact n-dimensional Riemannian manifold (M, g), one easily
defines the Sobolev spaces Hkp (M ), following what is done in the more traditional
Euclidean context. For instance, when k = 1 and p > 1, one may define the
Sobolev space H1p (M ) as follows: for u C (M ), we let
uH1p = up + up
CHAPTER 1
where .p is the Lp -norm with respect to the Riemannian measure dvg . We then
define H1p (M ) as the completion of C (M ) with respect to .H1p . A similar
definition holds for Hkp (M ), with
uHkp =
k
i up .
i=0
H1p
2n
n2
be the critical Sobolev exponent. Then, for any q [1, 2 ], H12 (M ) Lq (M ) and
this embedding is continuous, with the property that the embedding is also compact
if q < 2 . The Sobolev inequality corresponding to the continuous embedding
H12 (M ) L2 (M ) is as follows: For any u H12 (M ),
u2 Au2 + Bu2
where A and B are positive constants independent of u, but that may depend on
the manifold. Another very useful inequality is the so-called Poincare inequality.
When dealing with H12 , it reads as the existence of a positive constant A such that,
for any u H12 (M ),
u u2 Au2
where
u=
1
Vg
udvg
M
12
21
2
2
A
|u| dx
|u| dx
.
Rn
Rn
BACKGROUND MATERIAL
u, g dvg +
a(x)udvg =
f (x)dvg .
M
Regularity results for this equation do hold. They are similar to the more traditional
ones expressed in the Euclidean context (regularity is a local notion, so this is not
very surprising). The regularity result we will mostly use is the following: if a is
smooth, and f Hkp (M ) for some k N and p > 1, then a weak solution u to
p
(M ). In particular, it follows from this result and the
the above equation is in Hk+2
Sobolev embedding theorem, that when f is smooth u is also smooth. Needless to
say, the bible for such topics is the exhaustive Gilbarg-Trudinger [37]. A simpler,
but very nice reference is the lecture notes [41] by Han and Lin.
In parallel with regularity, the very useful maximum principles hold for the
Laplacian on Riemannian manifolds. A currently used form is as follows: if a
nonnegative u C 2 (M ) is such that, for any x M ,
g u(x) u(x)f (x, u(x))
for some continuous functions f : M R R, then either u is everywhere
positive, or u is the zero function. This easily follows from the Hopf maximum
principle, as usually stated.
10
CHAPTER 1
In order to end this section, we give a basic example of a possible use of the
above results. We discuss here the existence (and uniqueness) of a solution u to the
Laplace equation
g u = f
on a compact Riemannian manifold (M, g). Although not necessary, we assume
here for convenience that f : M R is smooth. Integrating the Laplace equation,
one sees that a necessary condition for the existence of a solution is that
f dvg = 0 .
M
2
udvg = 0 and
f udvg = 1
H = u H1 (M ) s.t.
M
and
= inf
uH
|u|2 dvg .
|ui |2 dvg = .
lim
i+
By the Poincare inequality we discussed above, there exists A > 0 such that, if
u H12 (M ) is of null average, then
u2 dvg A
|u|2 dvg .
M
It easily follows that the ui s are bounded in H12 (M ). Since H12 (M ) is a reflexive
space, and the embedding H12 (M ) L2 (M ) is compact by the Rellich-Kondrakov
theorem (even when n = 2, noting that H12 H1q for q < 2), there exists a function
u H12 (M ) and a subsequence (ui ) of (ui ), such that
(1) (ui ) converges weakly to u in H12 (M ) and
(2) (ui ) converges to u in L2 (M ).
By (2), u H. By (1), and a basic property of the weak limit (the norm of a weak
limit is less than or equal to the infimum limit of the norms of the sequence), we
get that
|u|2 dvg .
M
Hence,
|u|2 dvg =
11
BACKGROUND MATERIAL
u, g dvg =
dvg +
f dvg .
M
Hence, v u is constant, and this ends the proof of the above claim. As is easily
checked, everything in this proof comes from the compactness of the embedding
H12 L2 . The equations that we discuss below involve critical (noncompact)
Sobolev embeddings.
Chapter Two
The Model Equations
Let (M, g) be a smooth compact Riemannian manifold of dimension n 3. We
let (h ) be a sequence of smooth functions on M , and consider equations like
g u + h u = u2
(E )
14
CHAPTER 2
We briefly discuss in this very basic introductory chapter the existence of PalaisSmale sequences for (E ), and the existence of strong solutions for (E ) of minimal or arbitrarily high energies. More sophisticated examples can be found in
Druet-Hebey [31].
2.1 PALAIS-SMALE SEQUENCES
Let h be a smooth function on M . Following standard terminology, we say that an
operator like Lg = g + h is coercive if there exists C > 0 such that
(Lg u)udvg
(2.1.1)
u2H 2 C
1
H12 (M ),
(Lg u)udvg =
|u|2 + hu2 dvg .
M
Equation (2.1.1) just says that the H12 -norm of a function u is controlled by the
energy of u with respect to Lg . It is easily checked that, if h > 0 everywhere, then
Lg is coercive. As another remark, the existence of a u > 0 solution of (E )
implies that the operator L
g = g + h in the left hand side of equation (E ) is
coercive. From now on, we let Ig be the functional defined on the Sobolev space
H12 (M ) by
1
1
1
|u|2 dvg +
h u2 dvg
|u|2 dvg .
Ig (u) =
2 M
2 M
2 M
Let (u ) be a sequence of functions in H12 (M ). Following standard terminology,
we say that (u ) is a Palais-Smale sequence for equation (E ), or that (u ) is a
Palais-Smale sequence for Ig , if the two following propositions hold:
the sequence (Ig (u )) is bounded and
DIg (u ) 0 strongly in H12 (M ) as + .
(2.1.2)
15
(2.1.3)
Taking inspiration from Brezis-Nirenberg [15], thanks to the mountain pass lemma
of Ambrosetti and Rabinowitz [4], we prove the following proposition in this section.
P ROPOSITION 2.1 Let (M, g) be a smooth compact Riemannian manifold of dimension n 3, and (h ) be a sequence of smooth functions on M . We assume that
(2.1.3) holds. Then there exist Palais-Smale sequences of smooth positive functions
for (E ).
Proof. We let ( ) be a sequence of positive real numbers such that 0 as
+, and let q (1, 2 1). We fix , and let be the functional defined
on H12 (M ) by
1
1
2
|u| dvg +
h u2 dvg
(u) =
2 M
2 M
1
(u+ )2 dvg
(u+ )q+1 dvg
2 M
q+1 M
where u+ = max(u, 0). It is easily seen that the mountain pass lemma of Ambrosetti and Rabinowitz [4], as stated in Brezis and Nirenberg [15], can be applied
to . It follows that there exists c > 0 and a sequence (j ) in H12 (M ) such that
(j ) = c + o(1) and
D (j )(H12 ) = o(1)
(2.1.4)
We claim that by taking for u0 local test functions as in Aubin [6], c can be chosen
such that c < n1 Knn . More precisely, given > 0 small, x0 in M , and > 0,
we let u be the function
1 n2
u (x) = (x) 2 + dg (x0 , x)2
where dg is the distance with respect to g, and is a smooth cutoff function such
that = 1 in Bx0 () and = 0 in M \Bx0 (2), the geodesic balls of center x0 and
radii and 2. Easy computations give that for any q > 1 if n 4, and any q > 3
if n = 3,
u
1
n n2
n n2
2 (q+1) + o
2 (q+1)
=
sup t
u 2
nKnn
t0
where A = A(q, n) is a positive constant depending only on q and n. Assuming
that q > 3 if n = 3, and letting u0 = u with > 0 sufficiently small, this proves
16
CHAPTER 2
1
|j |2 + h 2j dvg
2 M
2
=
(+
)
dv
+
(+ )q+1 dvg + c + o(1)
g
2 M j
q+1 M j
and
|j |2 + h 2j dvg
M
2
q+1
=
(+
)
dv
+
(+
dvg + j H12 o(1) .
g
j
j )
(2.1.5)
(2.1.6)
1
q+1 (2.1.6),
we get that
Writing (2.1.5)
1
1
q
+ (u+
) .
Regularity results and the maximum principle then give that u is smooth and that
either u 0 or u > 0 everywhere. Up to another subsequence, we can assume
that j 22 as j +. If u 0 we then get with (2.1.5) and (2.1.6)
1/2
as j + and that nc = . On the other hand, thanks to
that +
j 2
Hebey and Vaugon [47, 48], there exists B > 0 such that
+ 2
2
2
2
+
j 2 Kn j 2 + Bj 2
so that, passing to the limit as j +, we get that 2/2 Kn2 . Noting that this
is in contradiction with the equations nc = and c < n1 Knn , we have proved
that u > 0 everywhere. Hence, (u ) is a sequence of smooth positive functions,
solutions of the equations
2 1
+ uq .
g u + h u = u
2
u dvg +
uq+1 dvg
Ig (u ) =
2 2
2 M
M
is uniformly bounded in . Similarly, for any H12 (M ),
DIg (u ). =
uq dvg ,
M
17
DIg (u ).
u2 q/(2 1) dvg
(2 1)/2
2
C H12
where C > 0 is independent of . In particular, the conditions in (2.1.2) are
satisfied. This proves the proposition.
2
A more classical construction of Palais-Smale sequences involves bubbles as
defined in the following chapter.
2.2 STRONG SOLUTIONS OF MINIMAL ENERGY
We assume in what follows that, for any , the operator L
g is coercive. We discuss
the existence of strong solutions of minimal energy. Given u H12 (M ), we define
the energy E(u) of u by E(u) = u2 . If u is a strong solution of (E ),
Ig (u) =
1
E(u)2 .
n
(n2)/2
then (E ) possesses a sequence (u ) of smooth positive solutions such that for any
, E(u ) < min .
Proof. The proof of such an assertion is by now very classical. We fix , and let I
be the functional defined on H12 (M ) by
|u|2 dvg +
h u2 dvg .
I(u) =
M
2
|u|2 dvg = 1
H = u H1 (M ) s.t.
M
and
= inf I(u) .
uH
18
CHAPTER 2
2
2
2
2
2
2
2
Kn ui 2 u2 = Kn Kn
|u| dvg +
h u dvg + o(1)
M
By standard regularity results, as developed by Trudinger [74], and using the maximum principle, u is smooth and positive. Noting that u = (n2)/4 u is a solution
2
of (E ), this proves the proposition.
19
Let Sg be the scalar curvature of g. Local arguments as in Aubin [6] give that if
n 4, and if there exists x M such that
h (x) <
n2
Sg (x) ,
4(n 1)
then the inequality on the infimum in Proposition 2.2 is satisfied. This provides
examples of h s for which (E ) possesses a sequence (u ) of smooth positive
solutions such that for any , E(u ) < min . Another framework where strong
solutions of minimal energy arise is when discussing the sharp constant problems.
As already mentioned, we know from Hebey and Vaugon [47, 48] that for any
smooth compact Riemannian manifold (M, g) of dimension n 3, there exists
B > 0 such that for any u H12 (M ),
u22 Kn2 u22 + Bu22 .
Let B0 (g) be the smallest B in this inequality. Taking u = 1, it is easily seen that
2/n
B0 (g) Vg
, where Vg is the volume of M with respect to g. We also have
that, for any u H12 (M ),
u22 Kn2 u22 + B0 (g)u22 .
By the definition of B0 (g), if h : M R is such that h < Kn2 B0 (g), then the
inequality on the infimum in Proposition 2.2 is satisfied. On the other hand, it was
proved by Hebey and Vaugon [49] that there are compact manifolds for which the
above sharp Sobolev inequality does not possess (nonzero) extremal functions. For
such manifolds, if (h ) is a sequence of smooth positive functions satisfying that
h < Kn2 B0 (g) for all , and that h Kn2 B0 (g) in C 0 (M ) as +,
then equation (E ) with respect to g possesses a sequence (u ) of smooth positive
solutions such that for any , E(u ) < min , such that E(u ) min as the
parameter +, and such that u 2 0 as +. In particular, the
u s necessarily blow up as +. We refer to Chapter 3 for the definition of
blowing-up sequences of solutions.
2.3 STRONG SOLUTIONS OF HIGH ENERGIES
We discuss the existence of strong positive solutions of arbitrary high energies in
some specific cases. We claim that the following result holds.
P ROPOSITION 2.3 There are smooth compact Riemannian manifolds with the following property: there exists 0 > 0 such that for any 1 > 0 there exist positive
real numbers K1 (1 ) < K2 (1 ) such that, if (h ) is any sequence of real numbers
in (K1 (1 ), K2 (1 )), then (E ) possesses a sequence (u ) of smooth nonconstant
positive solutions such that for any , 1 E(u ) 1 + min .
Typical examples of such manifolds are the unit sphere in odd dimension and the
product of a circle with any compact Riemannian manifold. We prove the above
proposition in the second case.
20
CHAPTER 2
2i
m
where Kn is the sharp Euclidean constant, and 2 the critical Sobolev exponent.
We let B0 (gt ) be the smallest constant B in the above inequality. Then, for any
u H12 (Mt ),
u22 Kn2 u22 + B0 (gt )u22 .
2/n
M1/m
|u1/m |2 dvg1/m =
M1/m
|u1/m |p dvg1/m
1
m
1
=
m
M1
M1
|u1 |p dvg1 ,
(2)2/n Kn2
u22 + Vg2/n u22
m2/n
where Vg is the volume of M with respect to g. Letting m +, we get that
u22
21
BB0 (g1/m )
B = Kn2 .
2
Noting that min = (Kn2 )(n2)/4 , and that m > 2
min 1 , we let
Vg2/n
< T1 (1 ) < B0 (g1/m )
1/m
be such that, for any T1 (1 ) < B < B0 (g1/m ),
1/2
(n2)/4
2
1 E(2
B
.
min 1 ) + 1
We set T2 (1 ) = B0 (g1/m ), and let (h ) be any sequence of real numbers in
(K1 , K2 ) where K1 = Kn2 T1 and K2 = Kn2 T2 . We write that h = Kn2 ,
so that T1 < < T2 . We set = . Since < Kn2 , we get with the kind
of arguments developed in the preceding section that there exists a minimizer u
for . Moreover, this minimizer can be chosen smooth and positive. Clearly, u
is not constant. If this is not the case,
Vg2/n
= Kn2 ,
1/m
2/n
and this is impossible since Kn2 < 1 and Vg1/m > 1. Independently, we can
choose u such that it is a solution of the equation
2 1
g1/m u + h u = u
(n2)/4
=
and such that u 2 = 1. If we let u
+ h u
=
g1/m u
(n2)/4
and
u 2 =
Then
u , then
2 1
u
. We let u
be the function on M1 such that u
/Gm = u
.
2 1
+ h u
= u
g1 u
(n2)/4
1/2
. Noting that 21 + 2min
1 +
and E(
u ) =
u 2 = m1/2
min , it is easily seen from the above construction that
u ) 1 + min .
1 E(
This proves the proposition.
As a remark, the same proof gives that for any sequence (h ) of smooth functions on M , if for any , K1 (1 ) < h < K2 (1 ), then (E ) on M1 = S 1 M
possesses a sequence (u ) of smooth nonconstant positive solutions such that, for
any , 1 E(u ) 1 + min .
Another possible construction, mixing both the ideas of the preceding section
and of this section, is as follows. Given m 1 integer, we let (h ) be a sequence
1
) M for which equation (E ) on M1/m
of smooth functions on M1/m = S 1 ( m
possesses a sequence (u ) of smooth positive solutions. Let : M1 M1/m
22
CHAPTER 2
and u
= h and
be the canonical projection, and let h
be such that h
u
= u . Then, for any ,
2 1
u
+ h
= u
g1 u
on M1 , and E(
u ) = m1/2 E(u ). Let us now suppose that the dimension n of M
is such that n 3, and that g is conformally flat. Then, g1/m is also conformally
flat on M1/m . By conformally flat we mean that, up to conformal changes of
the metric, we get local isometries with the Euclidean space. A conformal metric
to some Riemannian metric g is expressed as f g where f is a smooth and positive
function. When n 4, g is conformally flat if and only if Wg 0, where Wg is
the Weyl curvature tensor of g. Following Hebey-Vaugon [49], there exists gm , a
conformal metric to g1/m on M1/m , such that the sharp Sobolev inequality
gm )u22
u22 Kn2 u22 + B0 (
with respect to gm does not possess (nonzero) extremal functions, where B0 (
gm ) is
defined as in the proof of Proposition 2.3 (see also the remark at the end of section
2.2). We refer to Hebey-Vaugon [49] for a more general statement. Let (h ) be
gm ), and
a sequence of positive real numbers such that for any , h < Kn2 B0 (
such that h Kn2 B0 (
gm ) as +. By the definition of B0 (
gm ), the kind
of arguments of the preceding section give that, for any , there exists u smooth
and positive on M1/m , and (0, Kn2 ), such that
2 1
gm u + h u = u
(n2)/4
.
(n2)/4
, so that E(
u ) m1/2 min as +,
Moreover, E(
u ) = m1/2
and
u 2 0 as +, where the energy and the L2 -norm in these equations
are given with respect to gm . Summarizing, we have proved that, if (M, g) is a
conformally flat Riemannian manifold of dimension n 3, then, for any m 1
integer, there exists a conformal metric gm to the standard product metric on M1 =
S 1 M , there exists a bounded sequence (h ) of positive real numbers, and there
exists a sequence (u ) of smooth positive functions on M1 , solutions of equations
u ) m1/2 min as +,
(E ) on M1 with respect to gm , and such that E(
and
u 2 0 as +, where the energy and the L2 -norm in these equations
s necessarily blow up as +.
are with respect to gm . In particular, the u
Here again, we refer to Chapter 3 for the definition of blowing-up sequences of
solutions. Independently, note that the same construction can be carried out with
M of order m of
S n , n odd, in place of S 1 M , or with any covering = M
a compact Riemannian manifold of dimension n 4 whose Weyl curvature tensor
vanishes on some open subset of M .
23
u (x ) = maxn u (x) =
xS
Then x = x0 and
4( 1)
.
n(n 2)( + 1)
n2
2
B (x) =
.
d (x ,x)2
2 + hn(n2)
We refer to B (see the following chapters) as the standard bubble with respect to
x and . It is easily seen that there exists C > 1 such that
1
B (x) u (x) CB (x)
(2.4.1)
C
for all x S n and all > 1. Similarly, for any > 0, there exists > 0 such that
1
B (x) u (x) (1 + )B (x)
(2.4.2)
1+
for all x Bx0 ( ) when > 1 is close to 1. Let R be the function given by
u = B + R .
We claim that
R 0 in H12 (S n )
|B |2 dvh 2min and
B2 dvh 2min
Sn
Sn
24
CHAPTER 2
|u |2 dvh 2min
Sn
as 1. Writing that
2
2
|R | dvh =
|u | dvh +
Sn
Sn
|B | dvh 2
Sn
|R |2 dvh = 22min 2
Sn
(u B ) dvh ,
Sn
(u B ) dvh + o(1)
(2.4.3)
Sn
(u B ) dvh =
u2 1 B dvh + o(1) .
Sn
Sn
n
Given > 0, u 0 and B 0 in
S \Bx0 () as 1. Writing that
u2 1 B dvh =
u2 1 B dvh +
u2 1 B dvh
0
Cloc
Bx0 ()
Sn
S n \Bx0 ()
u2 1 B dvh = 2min + o(1) .
Sn
Hence,
Sn
|R |2 dvh 0
Sn
Chapter Three
Blow-up Theory in Sobolev Spaces
An important result of the 1980s describes the asymptotic behavior of Palais-Smale
sequences associated with equations like
u = u2 1 ,
(3.0.1)
where is the Euclidean Laplacian, and u is required to vanish on the boundary
of a smooth bounded open subset of the Euclidean space Rn . This result was
proved by Struwe [73]. Related references are Brezis [11], Brezis-Coron [12, 13],
Lions [58], Sacks-Uhlenbeck [64], Schoen [67], and Wente [76]. Let D12 (Rn ) be
the homogeneous Sobolev space defined as the completion of C0 (Rn ), the space
of smooth functions with compact support in Rn , with respect to the norm
uD12 =
|u|2 dx .
Rn
Nonnegative solutions in D12 (Rn ) of (3.0.1) have been classified by Caffarelli, Gidas, and Spruck [17]. We refer also to Obata [60]. The result we use is as follows:
if u D12 (Rn ), u nonnegative and nontrivial, is a solution of (3.0.1), then u = ua
for some a Rn and some > 0, where
n2
2
.
(3.0.2)
ua (x) =
2
1 + n(n2)
|x a|2
These functions are extremal functions for the sharp Euclidean Sobolev inequality
22
2
2
|u| dx
Kn
|u|2 dx .
(3.0.3)
Rn
Rn
Up to a nonzero multiplicative constant, they are the only nontrivial extremal functions for (3.0.3). Their energy is precisely the minimum energy in the sense that
E(ua ) = ua 2 = min . The free energy Ef , defined for u D12 (Rn ), is given
by
1
1
|u|2 dx
|u|2 dx .
(3.0.4)
Ef (u) =
2 Rn
2 Rn
It is easily checked that if u = ua is a nonnegative nontrivial solution of (3.0.1),
then Ef (u) = n1 Knn .
We present in this chapter the Struwe result [73] for smooth compact Riemannian
manifolds and equations like
(E )
g u + h u = u2 1
as considered in the preceding chapter. For the sake of clarity, we assume that the
h s are uniformly bounded and that they converge L2 to some limiting function.
26
CHAPTER 3
1
1
1
|u|2 dvg +
h u2 dvg
|u|2 dvg ,
Ig (u) =
2 M
2 M
2 M
the u s are characterized by the following:
the sequence (Ig (u )) is bounded and
(3.1.2)
DIg (u ) 0 strongly in H12 (M ) as + .
1
1
1
2
2
|u| dvg +
h u dvg
|u|2 dvg
Ig (u) =
2 M
2 M
2 M
and (E ) be the equation
g u + h u = u2 1
(E )
where h is as above. In the following, given > 0, denotes a smooth cutoff
function in Rn such that = 1 in B0 () and = 0 in Rn \B0 (2). For x M ,
where (M, g) is a smooth compact Riemannian manifold, and < ig /2, where ig
is the injectivity radius, we let ,x be the smooth cutoff function in M given by
,x (y) = exp1
x (y)
where expx is the exponential map at x. Here and in what follows, we regard
expx as defined in Rn . An intrinsic definition is possible if M is parallelizable.
i , i = 1, . . . , N , be open subsets of M such that for any
If not we let i and
i , and such that M = i . The canonical
i, i is parallelizable and i
exponential map gives N maps expx defined in i Rn , and expx is, depending
on the situation, one of these maps. A property of expx that holds for any x M
should then be regarded as a property that holds for any i and any x i .
27
T HEOREM 3.1 Let (M, g) be a smooth compact Riemannian manifold of dimension n 3, (h ) be a sequence of smooth functions on M such that (3.1.1) is
satisfied and (u ) be a Palais-Smale sequence of nonnegative functions for (E ).
j
j
j
), R
> 0, and R
+ as , conThere exist m N, sequences (R
j
verging sequences (x ) in M , a nonnegative solution u0 H12 (M ) of (E ), and
m
j uj + o(1)
(3.1.3)
j=1
where
j n2
j
2
uj (x) = R
uj R
exp1
,
j (x)
x
m n
K + o(1)
n n
(3.1.4)
where o(1) 0 as .
The proof of Theorem 3.1 proceeds in several steps. We follow the proof of
Struwe [73]. Uniqueness of the decomposition (3.1.3) is, for instance, discussed
by Hebey [46]. We use the terminology of a Palais-Smale sequence for (Ig ) instead
of a Palais-Smale sequence for (E ). First we claim that the following result holds:
S TEP 1. Palais-Smale sequences for (Ig ) are bounded in H12 (M ).
Proof of step 1. We let (u ) be a Palais-Smale sequence for (Ig ). Then,
|u |2 dvg +
h u2 dvg
|u |2 dvg
DIg (u ).u =
M
M
M
= o u H12 .
Hence,
Ig (u )
1
=
n
|u |2 dvg + o u H12
|u |2 dvg C + o u H12 .
M
2/2
u2 dvg C + o u H 2
M
where C > 0, like all the constants below, is independent of . Writing that
2
|u |2 dvg +
h u2 dvg = 2Ig (u ) +
|u |2 dvg ,
2 M
M
M
28
CHAPTER 3
|u |2 dvg +
h u2 dvg C + o u H12 .
|u |2 + h u2 dvg + Cu 22 ,
u 2H 2
1
H12 (M ).
Step 1 is proved.
1
1
|u|2 dvg
|u|2 dvg
Ig (u) =
2 M
2 M
so that Ig = Ig when h 0.
S TEP 2. Let (u ) be a Palais-Smale sequence of nonnegative functions for Ig ,
and u0 H12 (M ) a nonnegative function such that u u0 weakly in H12 (M ),
u u0 strongly in L2 (M ), and u u0 almost everywhere as +. Let
u ) is a Palais-Smale sequence for Ig and
u
= u u0 . Then (
Ig (
u ) = Ig (u ) Ig (u0 ) + o(1)
where o(1) 0 as +. Moreover, u0 is a solution of (E ).
Proof of step 2. We first observe that for any C (M ),
2 1
(u ) dvg +
h u dvg
u
dvg
DIg (u ). =
M
(3.1.5)
= o(1)
where o(1) 0 as +, and (u ) is the scalar product with respect to
g of u and . By (3.1.1), h h in Lp (M ) for any p 1 as +.
Writing that
h u dvg =
(h h )u dvg +
h u dvg
(3.1.6)
M
(h h )u dvg u 2 h h n
2
1/2
|| dvg
0
h u0 dvg =
(u0 )2 1 dvg .
u dvg +
M
(3.1.7)
29
. We know by
Hence, u0 is a solution of (E ). Now we compute the energy of u
step 1 that the u s are bounded in H12 (M ). From the Sobolev embedding theorem
and (3.1.6), (3.1.7) with = u , we can thus write that
h u2 dvg =
h (u0 )2 dvg + o(1) .
M
It follows that
Ig (u )
Ig (u0 )
1
+ Ig (
u )
2
dvg + o(1)
(3.1.8)
u +u0 |2 |
u |2 |u0 |2 . Clearly, there exists C > 0, independent
where = |
of , such that
| | dvg C
|
u |2 1 |u0 |dvg + C
|u0 |2 1 |
u |dvg
|
u |2 1 |u0 |dvg = o(1) and
M
Therefore,
M
|u0 |2
|
u |dvg = o(1) .
Summarizing, we are left in the proof of step 2 with the proof that (
u ) is a PalaisSmale sequence for Ig . Let C (M ). From (3.1.6), (3.1.7), and the Sobolev
embedding theorem,
h u dvg =
h u0 dvg + o H12 .
M
It follows that
DIg (u ).
= DIg (
u ).
M
dvg + o H12
(3.1.9)
where
u + u0 |2
= |
(
u + u0 ) |
u |2
u
|u0 |2
2 0
u .
Here again, it is easily checked that there exists C > 0, independent of , such that
| |dvg C
|
u |2 2 |u0 |||dvg + C
|u0 |2 2 |
u |||dvg .
M
| |dvg
M
0 2 2
C |
u |2 2 u0
+ |
u |
u
2 /(2 1)
2 /(2 1)
2 /(2 1)
2
= o(1) .
30
CHAPTER 3
Coming back to (3.1.9), and from the Sobolev embedding theorem, we then get
that
DIg (u ). = DIg (
u ). + o H12 .
This implies that (
u ) is a Palais-Smale sequence for Ig . Step 2 is proved.
1
n
n Kn
2
u ).
u =
|
u | dvg
|
u |2 dvg
DIg (
M
M
= o
u H12 .
Hence,
1
1
u 22 + o(1) =
u 22 + o(1) = + o(1) .
(3.1.10)
n
n
This already implies that 0. Following Hebey-Vaugon [47, 48], there exists
B > 0, independent of , such that
Ig (
u ) =
u 22 Kn2
u 22 + B
u 22 .
Moreover, since the embedding H12 (M ) L2 (M ) is compact, u
0 strongly
in L2 (M ) as +. Letting + in the above sharp Sobolev inequality,
it follows from (3.1.10) that
(n)2/2 Kn2 n .
Ig (u )
The following step is the main ingredient in the proof of Theorem 3.1. We postpone its proof to section 3.2.
S TEP 4. Let (
u ) be a Palais-Smale sequence for Ig such that u
0 weakly in
H12 (M ) but not strongly. Then there exists a sequence (R ) of positive real numbers, R + as , a converging sequence (x ) in M , and a nontrivial
solution u D12 (Rn ) of the Euclidean equation
u = |u|2
(3.1.11)
31
n2
B (x) = R 2 u R exp1
x (x)
2 /2
|u|2 dx =
|u|2 dx Kn2
|u|2 dx
.
Hence,
Rn
Rn
Rn
|u| dx
Rn
Knn
and
1
1
|u|2 dx .
Ef (u) =
2 2
Rn
This proves the above claim. In order to prove the theorem, we let (u ) be a PalaisSmale sequence of nonnegative functions for Ig . According to step 1, (u ) is
bounded in H12 (M ). Up to a subsequence, we may therefore assume that for some
u0 H12 (M ), u u0 weakly in H12 (M ), u u0 strongly in L2 (M ), and
u u0 almost everywhere as +. We may also assume that Ig (u ) c
as +. By step 2, u0 is a nonnegative solution of (3.0.1) and u
= u u 0
is a Palais-Smale sequence for Ig such that
Ig (
u ) = Ig (u ) Ig (u0 ) + o(1) .
If u
0 strongly in H12 (M ), then u = u0 + o(1), and the theorem is proved. If
not, according to the claim at the beginning of this proof, we apply step 4 to get a
new Palais-Smale sequence (
u1 ) of energy
Ig (
u1 ) Ig (
u ) + o(1) .
Here again, either u
1 0 strongly in H12 (M ), in which case the theorem is proved,
1
or u
0 weakly but not strongly in H12 (M ), in which case we again apply
step 4. By induction, at some point, either we do have compactness or the PalaisSmale sequence (
um
) we get with this process has an energy that converges to
2
m
some < . Then, by step 3, u
0 strongly in H1 (M ). Applying step 4
j
in this process, we got some u s, j = 1, . . . , m. Assuming that the uj s are also
nonnegative (see below) this ends the proof of Theorem 3.1.
2
The fact that the uj s we get in the above process are nonnegative is proved in
the next section. We refer to Lemma 3.3.
32
CHAPTER 3
n2
B (x) = R 2 u R exp1
x (x)
|
u |2 dvg = n + o(1)
(3.2.1)
M
Bx (t)
|
u |2 dvg .
Given t0 > 0 small, it follows from (3.2.1) that there exist x0 M and 0 > 0
such that, up to a subsequence,
|
u |2 dvg 0
Bx0 (t0 )
for all . Then, since t (t) is continuous, we get that for any (0, 0 ),
there exists t (0, t0 ) such that (t ) = . Clearly, there also exists x M
such that
(t ) =
Bx (t )
|
u |2 dvg .
33
1
n
x) = R
|
u |2 (x) ,
|
u |2 expx (R
where the norm in the left hand side of this equation is with respect to g, and the
norm in the right hand side is with respect to g . It follows that, if |z| + r < ig R ,
then
2
|
u | dvg =
(3.2.2)
|
u |2 dvg .
1
Bz (r)
expx (R
Bz (r))
When |z| + r < r0 R ,
1
Bz (r) Bexpx
expx R
while
1
(R
z)
1
C0 rR
1
1
expx R
B0 (C0 r) = Bx C0 rR
.
(3.2.3)
(3.2.4)
C0 rt1
0
|
u |2 dvg and
Bz (r)
(3.2.5)
|
u |2 dvg = .
B0 (C0 r)
We let (0, ig ) and C1 > 1 be such that for any x M , and any R 1, if
gx,R (y) = expx g(R1 y), then
1
2
2
|u| dx
|u| dvgx,R C1
|u|2 dx
(3.2.6)
C1 Rn
Rn
Rn
for all u D12 (Rn ) such that suppu B0 (R). Without loss of generality, we also
assume that
1
|u|dx
|u|dvgx,R C1
|u|dx
(3.2.7)
C1 Rn
Rn
Rn
for all u L1 (Rn ) such that suppu B0 (R). We let C0 (Rn ) be a cutoff
function such that
1, = 1 in B0 (1/4), and = 0 in Rn \B0 (3/4). We
10
1
set (x) = R x , where is as above. Then,
|(
u
)|2 dvg = O(1)
Rn
34
CHAPTER 3
weakly in D12 (Rn ). Now we divide the proof of Lemma 3.2 into several steps. As
a first step, we claim that the following holds.
S TEP 1. For r and sufficientl small,
u
u strongly in H12 (B0 (C0 r))
(3.2.8)
as +.
Proof of step 1. We let x0 Rn , and for > 0, we let h be the standard metric
on Bx0 (). By Fatous lemma,
2r
N (
u
)dvh d lim inf
N (
u
)dx C
lim inf
r
Bx0 ()
2
Bx0 (2r)
where Nh (u) = |u| + u , the norm in Nh is with respect to h, and is the Euclidean metric. It follows that there exists [r, 2r] such that, up to a subsequence,
and for all ,
N (
u
)dvh C .
Bx0 ()
)C
where C > 0 is independent of . The embedding
1
0 ()
2
H12 (Bx0 ()) H1/2
(Bx0 ())
as +. Let A be the annulus A = Bx0 (3r)\Bx0 (). Let also D12 (Rn )
u in Bx0 ( + ), and = 0 in Rn \Bx0 (3r ), > 0
be such that = u
small. Then
u
uH 2 (Bx ()) = H 2 (Bx ())
0
0
1/2
1/2
while there exists 0 D12 (A), the closure of C0 (A) in H12 (A), such that
2 (A) .
+ 0 H12 (A) C H1/2
Minimization arguments give that there exists z H12 (A) such that
z = 0 in A ,
z 0 D12 (A) ,
and
z H12 (A) C + 0 H12 (A) .
Hence, z 0 strongly in H12 (A) as +. We let D12 (Rn ) be such
that
= u
u in B x0 () ,
= z in B x0 (3r)\Bx0 () ,
35
and = 0 otherwise. We let r be such that r < min(ig /6, /24), and let be
such that
n2
(x) = R 2 R exp1
x (x)
if dg (x , x) < 6r, and = 0 otherwise. Clearly, 1 exp1
x (x) = 1 if
dg (x , x) < 6r. If in addition |x0 | < 3r, then
DIg (
u ). = DIg (
u
).
=
((
u
) ) dvg
Bx0 (3r)
Bx0 (3r)
|
u
|2
(
u
) dvg
((
u
) ) dvg
Bx0 (3r)
Bx0 ()
Rn
(( + u) ) dvg + o(1)
| |2 dvg + o(1) .
|
u
|2 2 (
u
) dvg =
Bx0 (3r)
| |2 dvg + o(1) ,
Rn
2
| | dvg
| |2 dvg = o(1) .
Rn
Rn
(3.2.9)
| |2 dvg =
|(
u
u)|2 dvg + o(1)
Rn
Bx0 ()
=
so that
Bx0 ()
Rn
|(
u
)| dvg
| | dvg
Bx0 ()
Bx0 ()
|(
u
)|2 dvg + o(1) .
Let N be an integer such that B0 (2) is covered by N balls of radius 1 and centered
in B0 (2). Then there exist N points xi Bx0 (2r), i = 1, . . . , N , such that
Bx0 () Bx0 (2r)
N
i=1
Bxi (r)
36
CHAPTER 3
and we get with (3.2.5) that for x0 and r such that |x0 | + 3r < r0 ,
| |2 dvg N + o(1) .
Rn
(3.2.10)
Independently, from the Sobolev inequality, for C1 as in (3.2.6) and (3.2.7), and
x0 and r such that |x0 | + 3r < , we also have that
2/2
2/2
2/2
2
2
| | dvg
C1
| | dx
Rn
Rn
2/2
C1 Kn2
| |2 dx
Rn
1+(2/2 ) 2
C1
Kn
| |2 dvg .
Rn
| |2 dvg K
| |2 dvg + o(1)
Rn
Rn
where
1+(2 /2)
K = C1
Let > 0 be such that
1+(2 /2) 2
Kn (N )(2 /2)1
C1
so that 0 strongly in
(2 /2)1
Kn2 (N + o(1))
< 1. Then,
| |2 dvg = o(1)
Rn
D12 (Rn )
(3.2.11)
1+(2 /2) 2
Kn (N )(2 /2)1
C1
|
u |2 dvg
=
B0 (C0 r)
=
|(
u
)|2 dvg
B0 (C0 r)
C1
|u|2 dx + o(1) .
B0 (C0 r)
(3.2.12)
37
(3.2.13)
u.
(x) = R 2 (R x) .
1
R0 ). For large, we let be the smooth function on M
Then supp B0 (R
given by = expx . For large,
(
u ) dvg =
((
u
)) dvg
Rn
and
|
u |2
u
dvg =
Rn
|
u
|2
(
u
)dvg .
(u) dx =
|u|2 2 udx .
Rn
In other words, u
Rn
D12 (Rn )
(3.2.14)
(3.2.15)
(3.2.16)
strongly as +. Finally,
u ) Ef (u) + o(1)
Ig (w ) = Ig (
(3.2.17)
38
CHAPTER 3
where o(1) 0 as +
Proof of step 3. We start with the proof of (3.2.15). It suffices to prove that V 0
1
R). For a smooth
weakly in H12 (M ). Given R > 0, we let (R) = Bx (R
function on M and large,
n2
V dvg = R 2
(x)u(R x)(expx (x))dvg
1
B0 (R
R)
(R)
where g = expx g. It follows that for C > 0 such that dvg Cdx,
n+2
|u|dx .
V dvg C R 2
(R)
B0 (R)
Similarly, by Holders inequality,
n+2
2
V dvg C R
|u|dx
M \ (R)
B0 (R )\B0 (R)
1/2
C
|u|2 dx
B0 (R )\B0 (R)
Taking R > 0 sufficiently large and using (3.2.12),
we get that M V dvg 0
as +. Similar arguments give that M (V )dvg 0 as +.
This proves (3.2.15). Now we prove (3.2.16). Here again we let be a smooth
function on M . Then,
DIg (V ). =
(V )dvg
|V |2 2 V dvg .
M
(V )dvg =
(R)
(V )dvg +
Bx ()\ (R)
Bx ()\ (R)
(V )dvg .
(V )dvg = O H12 R
(V )dvg =
(u )dvg .
(R)
B0 (R)
2
|| dvg =
| |2 dvg ,
(R)
B0 (R)
39
we get that
B0 (R)
(u )dvg =
B0 (R)
B0 (R)
(u )dx =
Rn
(u )dx + o H12 .
(u )dx + O H12 R
(V )dvg =
(u )dx+o H12 +O H12 R . (3.2.18)
Rn
2 2
|V |
V dvg =
|u|2 2 u dx
n
M
R
+ o H12 + O H12 R .
Since u is a solution of u = |u|2
(3.2.19)
(3.2.20)
A(). =
dvg =
M
Bx (2)
dvg
and
= |w |2
w |
u |2
u
+ |V |2
V .
2
| 221 dvg
| | 2 1 dvg =
|
(R)
B0 (R)
40
CHAPTER 3
where
= |
u u|2 2 (
u u) |
u |2 2 u
+ |u|2 2 u .
2
(R)
| | 2 1 dvg = o(1) .
Independently,
2
|1 | 2 1 dvg =
)\B0 (R)
B0 (2R
(R)c
|
u
|
C
where
Rn \B0 (R)
|
u
|
2 (2 2)
2 1
2 (2 2)
2 1
2
2
1
|u| 2 1 2
dvg
2
|u| 2 1 dx ,
1
(x) = ,x
expx (R x)
and C > 0 is such that dvg Cdx. Without loss of generality, we may assume
u almost everywhere in Rn . We let
that u
u
|
f = |
2 (2 2)
2 1
and f = |u|
2 (2 2)
2 1
2 1
2 2
Then (f ) is bounded in L
(Rn ) and (f ) converges almost everywhere to f .
From standard integration theory, it follows that
2 1
f f weakly in L 2 2 (Rn )
as +. Hence,
lim
+
Rn \B
0 (R)
|
u
|
2 (2 2)
2 1
|u|
R+ +
R+ +
(R)c
(R)c
2
2 1
dx =
Rn \B
|u|2 dx ,
0 (R)
2
|1 | 2 1 dvg = 0 .
2
|2 | 2 1 dvg = 0 .
Coming back to (3.2.20), and since R > 0 is arbitrary, we get that DIg (w ) 0
strongly as +. This proves (3.2.16). Now we are left with the proof of
(3.2.17). We have that
1
1
Ig (w ) =
|w |2 dvg
|w |2 dvg .
(3.2.21)
2 M
2 M
Concerning the first term in (3.2.21), we write
2
2
|w | dvg =
|w | dvg +
M
Bx (2)
M \Bx (2)
|
u |2 dvg .
41
|w |2 dvg =
Bx (2)
(R)
|w |2 dvg +
(R)c
|w |2 dvg
2
|w |2 dvg =
|(
u u)| dvg
(R)
B0 (R)
(R)
|w |2 dvg = o(1) .
(R)c
Since w = u
V , and (
u ) is bounded in H12 (M ), we have that
|w |2 dvg =
|
u |2 dvg + BR ()
(R)c
(R)c
where
lim lim sup BR () = 0 .
R+ +
Therefore,
|w |2 dvg =
M
|
u |2 dvg
(R)
2
|
u | dvg =
(R)
(3.2.22)
|
u |2 dvg + BR () + o(1)
B0 (R)
|
u |2 dvg
|
u |2 dvg =
|u|2 dx + o(1)
(R)
B0 (R)
=
|u|2 dx + R + o(1)
Rn
where R 0 as R +. Hence,
2
2
|w | dvg =
|
u | dvg
M
Rn
|u|2 dx + BR () + o(1)
|w |2 dvg =
|
u |2 dvg
|u|2 dx + BR () + o(1)
M
Rn
(3.2.23)
(3.2.24)
42
CHAPTER 3
From steps 13, Lemma 3.2 holds for some (0, ig /2) small. Given 1 < 2
in (0, ig /2), it is easily seen that
(2 ,x 1 ,x )B H 2 = o(1) .
1
It follows that Lemma 3.2 holds for any (0, ig /2). This ends the proof of
Lemma 3.2.
2
Another important ingredient we used in the proof of Theorem 3.1 is the following result.
L EMMA 3.3 If (u ) is a Palais-Smale sequence of nonnegative functions for (Ig ),
s
(3.2.25)
ui uN
dvg = o(1) + (R )
u
p
j
N
(R)\j=1 (R )
i=1
j
j
j (R ), (R ) 0 as R +,
where N
(RN
(R) = BxN
), (R ) = By
and the (ui )s and (xi )s are the ordered sequences in i we get in the proof of
Theorem 3.1. We proceed here by inverse induction on s. If s = N 1, then, by
(3.2.13),
2
N
1
ui uN
u
dvg = o(1)
N
(R)
i=1
so that (3.2.25) holds with p = 0. Now we suppose that (3.2.25) holds for some s,
s N 1. If the dg (xs , xN
)s do not converge to 0, then, up to a subsequence,
s
> 0. As a consequence,
(R)
(
R)
=
for
R
N
|us |2 dvg
|us |2 dvg
p
j
N
(R)\j=1 (R )
M \s (R)
|us |2 dvg C
p
j
N
(R)\j=1 (R )
Rn \B
0 (R)
|us |2 dx .
|us |2 dvg = o(1)
p
j
N
(R)\j=1 (R )
43
and thus
2
s1
ui uN
u
dvg = o(1) + (R ) .
p
j
N
(R)\j=1 (R )
i=1
j
R
j
1
1 j
j
exp
(R
)
B
C
Byj
R
(3.2.26)
y
xN
C N
N
N
and
Bxs
R s
C N
1
exp1
(s (R )) Bxs
xN
R C
s
N
.
(3.2.27)
s
i
u dvg = o(1) .
u
s (R)
i=1
Hence, by (3.2.25),
2
|uN
| dvg = o(1) + (R ) ,
p
j
N
(R)\j=1 (R )
s (R)
|uN |2 dx = o(1) + (R ) .
j
s
B0 (R)\p
j=1 B
j (R
y
C
N
) Bx
(R
s
C
(3.2.28)
dg (xs , xN
dg (xs , xN
N
)
)
=
s .
s
N
2
s
i
N
u u dvg = o(1) + (R )
u
p+1 j
N
(R)\j=1 (R )
i=1
44
CHAPTER 3
and
|us |2 dvg
p+1 j
N
(R)\j=1 (R )
|us |2 dvg
M \s (R )
(R ) .
It follows that
2
s1
ui uN
u
dvg = o(1) + (R )
p+1 j
N
(R)\j=1 (R )
i=1
and (3.2.25) holds for s 1. In particular, (3.2.25) is always true. Now we prove
the claim that if the u s in Theorem 3.1 are nonnegative, then u0 and the ui s
of Theorem 3.1 are also nonnegative. As already mentioned, it is clear that u0 is
nonnegative. We let u
N
be given by
N n2
N
2 u
u
N
expxN
(
x)
.
(x) = ( )
2
u
dvg = o(1) + (R ) ,
uN
p
j
N
(R)\j=1 (R )
and it follows
that
B0 (R)\p
j (R C N
j=1 By
N
2
u
uN dx = o(1) + (R )
(3.2.29)
u
N
in L2loc B0 (R)\X
u
consists of the yj s, j = 1, . . . , p. Therefore, we may
as +, where X
N
N
assume that u
u almost everywhere in Rn as +. Let
N n2
2 u0
u
0,N
(N
expxN
(x) = ( )
x) .
Then,
where g (x) =
|u0 |2 dvg =
N
(R)
expxN g (N
x).
u
0,N
u
0,N
2
|
u0,N
| dvg
B0 (R)
It follows that
0 in L2 (B0 (R))
0 almost everywhere in Rn . Summarizing, the
vN (x) = (N
)
n2
2
N
u expxN
(
x)
45
(E )
g u f u
B0 (2)
B0 ()
u, dvg +
hudvg
u2 1 dvg .
M
Then u L (M ). Moreover, for any x in M , any > 0, any p > 0, any q > 2 ,
and any > 0, if u is some nonnegative function of H12 (M ) satisfying the above
equation and
uq dvg ,
Bx (2)
46
CHAPTER 3
then
sup u(y) C
yBx ()
Bx (2)
up dvg
1/p
yBx ()
i=1,...,N
Byi (yi )
Bx ()
(3.3.1)
g u
4/(n2)
u 2
(3.3.2)
and set
)
(E
and
u 2 = 1. Moreover, the s are bounded, and it easily follows from
(3.3.1) and (3.3.2) that there exists c > 0, independent of , such that for any ,
u ) is bounded in H12 (M ), and that (
u ) also
c. It follows that the sequence (
satisfies (3.3.2). Following standard terminology, we say that x is a (geometrical)
blow-up point for (
u ), or for (u ), if for any > 0,
u
2 dvg > 0
lim sup
+
Bx ()
where Bx () is the geodesic ball of center x and radius . We let S be the set
consisting of the geometrical blow-up points for (
u ). Clearly, S = . We claim
that, up to a subsequence, the following holds:
S is a finite set and
0
u 0 in Cloc
(M \S) .
(3.3.3)
47
.
2 u
(3.3.4)
Integrating by parts,
+1
4
2
2
u
g u
dvg =
|(
u
)|2 dvg
( + 1)2 M
M
2( 1)
2
+1
(
)
u
dv
||2 u
+1
g
g
dvg
( + 1)2 M
+1 M
while by Holders inequalities,
2 u
2 +1 dvg
+1
2
(
u )2 dvg
22
Bx ()
u
2 dvg
222
.
Moreover, one gets by the Sobolev inequalities that there exist A, B > 0, independent of and , such that
22
+1
+1
2
2
2
2
(
u ) dvg
A
|(
u )| dvg + B
2 u
+1
dvg . (3.3.5)
M
From (3.3.1), there also exists C > 0, independent of and , such that
+1
4
1
|(
u 2 )|2 dvg
C+
( + 1)2
M
+1
4
2
2
|(
u )| dvg +
h 2 u
+1
dvg
( + 1)2 M
M
so that, for any 1 sufficiently close to 1,
+1
+1
C
4
2
2
|(
u
|(
u 2 )|2 dvg
)|
dv
+
h 2 u
+1
g
dvg .
2 M
( + 1)2 M
M
Integrating (3.3.4) then gives that
+1
|(
u 2 )|2 dvg C1 ()
M
+1
|(
u 2 )|2 dvg + C2 (, ) ,
u 2 = 1,
where, since for some > 0, and
2A
C1 () =
C
and
Bx ()
u
2 dvg
222
4( 1)
(g )
u+1
dvg
( + 1)2 C M
4
2B
+
||2 u
+1
dv
+
2 u
+1
g
dvg .
( + 1)C M
C M
C2 (, ) =
(3.3.6)
48
CHAPTER 3
Clearly, there exists at least one geometrical blow-up point for the sequence (
u ).
Let x0 be such a point. We claim that for all > 0,
222
C
2
.
(3.3.7)
u
dvg
lim sup
2A
+ Bx ()
0
If this is not the case, then there exists C1 (0, 1) such that C1 () C1 for all
. By (3.3.6), and since (
u ) is bounded in L2 (M ), it follows that there exists
C2 > 0 such that for any ,
+1
|(
u 2 )|2 dvg C2 .
(3.3.8)
M
By Holders inequalities,
u
2 dvg
Bx0 (/2)
2 (+1)
2
Bx0 (/2)
n+2
n(+1)
dvg
n(+1)
n2
Bx0 (/2)
n2
n(+1)
dvg
n(+1)
u
2 dvg C3
n(+1)
Bx0 (/2)
Bx0 (/2)
u
n2 dvg
n(+1)
(3.3.9)
u
n2 dvg C4 .
Bx0 (/2)
Fix > 1 close to 1, and let 1 = n( + 1)/(n 2). Then 1 < 1 < 2 . Since
u 1 0 as +. Then, by
u 2 0 as +, we also have that
u 1 0 as +.
Holders inequalities, and since
u 2 is bounded,
This contradicts (3.3.9), so that (3.3.7) is proved. Then, as an immediate consequence of (3.3.7) and the relation
u 2 = 1, we get that, up to a subsequence,
(
u ) has a finite number of geometrical blow-up points. Let S be the set of blow-up
points of this subsequence,
and let x now be a point in M \S. There exists > 0
such that u
0 in L2 Bx () . Coming back to (3.3.6), there exists > 1 such
(+1)/2
that
u
is bounded in H12 (M ). Thanks
to the Sobolev embedding theorem,
it follows that u
is bounded in Lq Bx (/2) for some q > 2 . Noting that
2 1
g u
+ h
u u
49
u2 dvg > 0 .
lim inf
+
j
j (1/R )
x
u2 dvg > 0 .
lim inf
+
Bx
i ()
there exists
This proves that S S. Conversely,
(3.1.3)
gives that for any x S,
2
Chapter Four
Exhaustion and Weak Pointwise Estimates
Let (M, g) be a smooth compact Riemannian manifold of dimension n 3. We
let (h ) be a sequence of smooth functions on M , and consider equations like
g u + h u = u2
(E )
for all . We describe in this chapter an exhaustion method for the blow-up behavior of the u s. This method provides a constructive approach to Theorem 3.1, and
weak (with respect to the material in Chapters 5 and 6) pointwise estimates on the
u s.
We assume in what follows that there exist 0 < < 1 and a smooth (or only
C 0, ) function h on M such that
the operator g + h is coercive and
(4.0.1)
h h in C 0, (M ) as + .
We also assume that there exists > 0 such that E(u ) for all , where
E(u) = u2 is as in Chapter 2.
Assuming (4.0.1) and that E(u ) for all , it is easily seen that the u s are
bounded in H12 (M ). Indeed, we clearly have that h n/2 C for all , where
C > 0 is independent of . Multiplying (E ) by u and integrating over M , we
then get that
u2 dvg 2 .
|u |2 dvg +
h u2 dvg =
M
2
h u dvg
M
2/n
|h |
n/2
u2 dvg
dvg
2/2
.
It follows that there exists C > 0 such that u 2 C for all , and then that the
u s are bounded in H12 (M ) since they are also bounded in L2 (M ) and 2 < 2 .
Up to a subsequence, we may therefore assume that for some u0 H12 (M ),
u u0 weakly in H12 (M )
0
(4.0.2)
52
CHAPTER 4
Another easy remark is that (4.0.1) implies the uniform coercivity condition (3.3.1)
for sufficiently large. Indeed, if g +h is coercive, then there exists > 0 such
that g +(h ) is still coercive. Since h h in C 0, (M ), h h for
sufficiently large. In particular, there exists h such that h h for sufficiently
large, and such that g + h is coercive.
Without loss of generality, we can assume in the following that
lim max u = +
+ M
(4.0.3)
so that blow-up occurs. If this is not the case, up to a subsequence, we easily get
from standard regularity theory that u u0 strongly in C 2 (M ) as +.
The main result of this chapter, Theorem 4.1, is stated in section 4.1. Section 4.2
is concerned with the proof of this result.
4.1 WEAK POINTWISE ESTIMATES
We set up some notation. Given N N , let (xi, ), i = 1, . . . , N , be N converging
sequences of points in M , and (i, ), i = 1, . . . , N , be N sequences of positive
real numbers converging to 0. We set
(4.1.1)
S=
lim xi, , i = 1, . . . , N
+
B
()
,
j
=
i
Si =
j,
j,
xi,
xi,
+ i,
(4.1.2)
for i = 1, . . . , N , where 0 < ig /2, ig is the injectivity radius of (M, g), and
the limits, up to a subsequence, are assumed to exist. As in Chapter 3, we regard
expx as defined in Rn . An intrinsic definition is possible if M is parallelizable.
j , j = 1, . . . , k, be open subsets of M such that for any
If not we let j and
j , and such that M = j . The canonical
j, j is parallelizable and j
exponential map gives k maps expx defined in j Rn , and expx is, depending
on the situation, one of these maps. A property of expx that holds for any x M
should then be regarded as a property that
holds
for any j and any x j . We also
let ui, be the function defined in B0 1
i, , the Euclidean ball of center 0 and
1
radius i, , by
n
2 1
ui, (x) = i,
u expxi, (i, x) .
(4.1.3)
Conversely, if u is a smooth function defined in Rn , we let vi, = vi, (u) be the
function given by
1
1 n
1
2
expxi, (x) if x Bxi, () and
vi, (x) = i, u
i,
(4.1.4)
vi, (x) = 0 otherwise .
53
(4.1.7)
2
weakly in D12 (Rn ), strongly in Cloc
(Rn \Si ), and strongly in C 2 (B0 (0 )), where S
is as in (4.1.1), Si is as in (4.1.2), ui, is as in (4.1.3), and u is as in (4.1.5).
n
N
(x) 2 1 u (x) C, where C > 0 is
(P2) For any x M , and any , R
independent of . Moreover,
n
N
lim
sup
R
(x) 2 1 u (x) u0 (x) = 0
lim
R+ + xM \ (R)
N
where u0 is given by (4.0.2), R
is given by (4.1.7), and (R) is given by
N
(R) = i=1 Bxi, (Ri, ).
(P3) The energy E, given by E(u) = u2 , satisfie
N
2
lim E u u0
vi,
lim E(u )2 u0 22 N 2min
+
i=1
where u0 is given by (4.0.2), vi, = vi, (ui ) is as in (4.1.4), and ui (x) = u(xxi )
where u is as in (4.1.5) and xi is as in (P1).
54
CHAPTER 4
As a remark, because of (P3), we can saturate Theorem 4.1 in order to get some
maximal N by adding sequences (xj, ) and (j, ) such that (P1)(P3) continue to
hold. On the other hand, we clearly have that (4.0.1) implies (3.1.1). We let S be
the set of geometrical blow-up points given by (4.1.1) that we get from Theorem
4.1. We also let S be the set consisting of the limits of the (xik )s we get from
Theorem 3.1. An easy remark is that S and S coincide.
4.2 EXHAUSTION OF BLOW-UP POINTS
We prove Theorem 4.1 in this section. We proceed in several steps. The claims in
what follows are up to a subsequence. First we claim that the following holds.
S TEP 1. There exist a converging sequence (x1, ) of points in M and a sequence
(1, ) of positive real numbers converging to 0 such that, up to a subsequence, the
following propositions hold :
2
(A1) u1, u in Cloc
(Rn ) as +,
u (x1, ) = 1, 2 = max u .
M
From (4.0.3),
lim 1, = 0 .
and for x B0 1
1, , the Euclidean ball of center 0 and
+
n
2 1
u expx1, (1, x) ,
u1, (x) = 1,
g1, (x) = expx1, g (1, x) ,
h1, (x) = h expx1, (1, x) .
where is the Euclidean metric. Note that we also have that g1, is controlled on
both sides by in the sense of bilinear forms. It is easily checked that
u1, (0) = u1, L (B0 (1 )) = 1
(4.2.1)
2 1
g1, u1, + h1, 21, u1, = u1,
(4.2.2)
1,
and that
55
1
(4.2.3)
(4.2.4)
where is the Euclidean Laplacian. We also claim that 1, u1, u1 in D12 (Rn )
as +, where 1, is as in the introduction of this section. In order to prove
these claims, we first note that the 1, u1, s are bounded in D12 (Rn ). Indeed, it
is clear that there exists C > 0, independent of , such that |1, | C1, .
Letting s = 2 /(2 2), so that 2s = n, we then write that
2
2
|1, |2 u21, dvg1, + 2
1,
|u1, |2 dvg1,
Rn
Rn
Rn
1/s
2/2
u21, dvg1,
B0 (/1, )
+ O(1)
O(1) .
This proves that the 1, u1, s are bounded in D12 (Rn ). We may therefore assume
that 1, u1, u1 in D12 (Rn ) as + for some u1 D12 (Rn ). From (4.0.1),
0
the coefficients in equation (4.2.2) are bounded and h1, 21, 0 in Cloc
(Rn ) as
+. By standard elliptic theory, using (4.2.1), this gives (4.2.3) and (4.2.4).
It is clear from (4.2.1) and (4.2.3) that u1 (0) = u1 = 1. From CaffarelliGidas-Spruck [17], we necessarily have that u1 = u where u is as in (4.1.5). In
particular, (A1) holds. Let us now prove that (A2) holds also. We write that
0
2
|u u0 v1, |2 dvg
|u u v1, | dvg =
M \Bx1, (R1, )
|u u0 v1, |2 dvg
Bx1, (R1, )
lim
|u u0 v1, |2 dvg = 0 .
+
Bx1, (R1, )
We write now that for any > 0 there exists C > 0 such that
|u u0 v1, |2 dvg
M \Bx1, (R1, )
(1 + )
+C
M \Bx1, (R1, )
M \Bx1, (R1, )
|u u0 |2 dvg
2
v1,
dvg .
56
CHAPTER 4
Here, C does not depend on R and , and C is defined by the property that for
any x, y R,
|x + y|2 (1 + ) |x|2 + C |y|2 .
lim
R+ +
2
v1,
dvg = 0 .
M \Bx1, (R1, )
|u u0 v1, |2 dvg
lim
+ M
lim
|u u0 |2 dvg
+ M
lim
lim
|u u0 |2 dvg .
R+ +
From (A1),
lim
lim
R+ +
Bx1, (R1, )
|u u0 |2 dvg = lim
Bx1, (R1, )
R+
u2 dx
B0 (R)
= 2min .
|u u0 |2 dvg = lim u 22 u0 22 .
lim
+
Thus (A2) is proved thanks to these relations. This ends the proof of step 1.
dg (xi, ,xj, )
i,
+ as
(B2) for any i {1, . . . , k}, i, ui, u weakly in D12 (Rn ) and strongly in
as +,
2
(Rn )
Cloc
(B3) lim u u0
+
k
i=1
57
k
is given by (4.1.7) and C > 0 is independent
for all x M and all , where R
of x and .
k
k
max R
(x) 2 1 u (x) = R
(y ) 2 1 u (y )
n
xM
(4.2.6)
and we set
1 n
2
.
xk+1, = y and u (xk+1, ) = k+1,
2
in Cloc
(Rn ), where is the Euclidean metric. We also have that gk+1, is on both
sides controled by in the sense of bilinear forms. It is easily checked that
uk+1, (0) = 1
and that
2 1
gk+1, uk+1, + hk+1, 2k+1, uk+1, = uk+1,
n
in B0 1
k+1, . Let x R . For i {1, . . . , k},
58
CHAPTER 4
as +, so that
lim
dg xi, , expxk+1, (k+1, x)
dg (xi, , xk+1, )
=1
2
(Rn )
lim uk+1, = uk+1 in Cloc
and where is the Euclidean Laplacian. We also get that k+1, uk+1, uk+1
weakly in D12 (Rn ) as +. From Caffarelli-Gidas-Spruck
[17], uk+1 = u
1
. Independently, we have
where u is as in (4.1.5). This proves (B2) of Hk+1
1
already seen that (B1) of Hk+1
is true. Thus, it remains to prove that (B3) of
1
Hk+1 is also true. Let R > 0 be given. From the expression for u, there exists
C > 0, independent of , such that for any i {1, . . . , k},
2n
2
vi,
dvg Cni,
dg (xi, , x)
dvg .
Bxk+1, (Rk+1, )
Bxk+1, (Rk+1, )
1
Using (B1) of Hk+1
, it is easily checked that for large enough and for any
x Bxk+1, (Rk+1, ),
2n
dg (xi, , x)
Thus,
2n
2
vi,
dvg
Bxk+1, (Rk+1, )
2n n
2Cdg (xi, , xk+1, )
i, Volg Bxk+1, (Rk+1, )
nk+1,
ni,
C(R)
n
n
dg (xi, , xk+1, ) dg (xi, , xk+1, )
where C(R) is independent of , and for M , Volg () stands for the volume
1
, it follows that
of with respect to g. Using again (B1) of the sequence Hk+1
for any i {1, . . . , k} and any R > 0,
2
lim
vi,
dvg = 0 .
(4.2.7)
+
Bxk+1, (Rk+1, )
1
From (B2) of Hk+1
, we also have that
lim
lim
R+ +
Bxk+1, (Rk+1, )
u2 dvg = 2min .
(4.2.8)
59
2
k
0
vi, dvg
u u
M \Bxk+1, (Rk+1, )
i=1
k
2
0
=
vi, dvg
u u
M
i=1
2
k
0
vi, dvg .
u u
Bx
(Rk+1, )
i=1
k+1,
Since
2
k
0
vi, dvg
u u
Bxk+1, (Rk+1, )
i=1
u2 dvg + 2
Bxk+1, (Rk+1, )
+2
Bxk+1, (Rk+1, )
i=1
2 1 0
u
u dvg
2 1
u
vi, dvg ,
Bxk+1, (Rk+1, )
it follows from (B3) of Hk1 , (4.2.7), and (4.2.8) that
2
k
lim
vi, dvg
lim
u u0
R+ + M \B
(Rk+1, )
x
(4.2.9)
i=1
k+1,
2
u0 22
1)2min .
(k +
lim E(u )
+
1
Using (B2) of Hk+1
, we also have that
|u vk+1, |2 dvg = 0
lim
(4.2.10)
lim
lim
(4.2.11)
Bxk+1, (Rk+1, )
R+ +
Writing that
Bxk+1,
M \Bxk+1, (Rk+1, )
i=1
|u vk+1, |2 dvg
Bxk+1, (Rk+1, )
k
i=1
21
2
k+1
vi, dvg
u u 0
(Rk+1, )
2
vk+1,
dvg = 0 .
Bxk+1, (Rk+1, )
2
vi,
dvg
21
21
+
Bxk+1, (Rk+1, )
0 2
(u ) dvg
21
,
60
CHAPTER 4
k+1
lim
vi, dvg = 0 .
u u0
+ B
(R
)
x
k+1,
(4.2.12)
i=1
k+1,
k+1
0
vi, dvg
u u
M
i=1
2
k+1
vi, dvg
u u0
Bxk+1, (Rk+1, )
i=1
2
k
0
+ (1 + )
vi, dvg
u u
M \Bxk+1, (Rk+1, )
i=1
2
+C
vk+1,
dvg
M \Bxk+1, (Rk+1, )
where C does not depend on R and . Passing to the limit in this expression as
+, then as R + and finally as 0, we get from (4.2.9), (4.2.11),
and (4.2.12), that
2
k+1
vi, dvg lim E(u )2 u0 22 (k+1)2min .
lim
u u0
+ M
+
i=1
1
Hence, (B3) of Hk+1
is true, and this ends the proof of Step 2.
2
Following steps 1 and 2, by induction, we get that there exists k 1, with
2min k 2 u0 22 , that there exist k converging sequences (xi, ) of points in
M and that there exist k sequences (i, ) of positive real numbers converging to
0, such that (B1)(B3) hold, and such that for any x M and any ,
k
R
(x) 2 1 u (x) C
n
(4.2.13)
dg (xi, ,xj, )
min{i, ,j, }
0
(C2) lim u = u0 in Cloc
(M \S) where S is as in (4.1.1),
+
+ for any
61
(C3) there exists 0 > 0 and there exists xi Rn , i = 1, . . . , k, such that for any
2
(Rn \Si ),
i = 1, . . . , k, i, ui, u ( xi ) weakly in D12 (Rn ), strongly in Cloc
2
and strongly in C (B0 (0 )) as +, where ui, is as in (4.1.3), Si is as in
(4.1.2), and u is given by (4.1.5),
k
(C4) for any x M and any , R
(x) 2 1 u (x) C where C > 0 is independent of x and ,
k
(C5) lim u u0 i=1 vi, 22 lim E(u )2 u0 22 k2min
n
(4.2.14)
dg (xi, , xk+1, )
+
i,
(4.2.15)
n2
2
(4.2.16)
2
u (xk+1, ) = k+1,
2
and claim that Hk+1
holds when adding (xk+1, ) and (k+1, ) to the (xi, )s
2
is a consequence
and (i, )s, i = 1, . . . , k. An easy remark is that (C1) of Hk+1
of (4.2.15). We also have that
k
R
(xk+1, ) 0
2
as +. If not, (C2) of Hk implies that
(4.2.17)
|u (xk+1, ) u0 (xk+1, ) | 0
as +, contradicting (4.2.16) since M is compact. Using (4.2.17), we can
rewrite (4.2.16) as
k
(xk+1, )
R
20
k+1,
(4.2.18)
62
CHAPTER 4
(4.2.19)
We also have that gk+1, is controlled on both sides by in the sense of bilinear
forms. It is easily checked that
uk+1, (0) = 1
and that
2 1
gk+1, uk+1, + hk+1, 2k+1, uk+1, = uk+1,
(4.2.20)
in B0 1
k+1, . We let
1
lim
exp1
(x
)
,
x
B
()
,
1
k
Sk+1 =
i,
i,
xk+1,
xk+1,
+ k+1,
1
R
uk+1, (x ) (2R) 2
k
R
(y ) 2
u (y ) (2R) 2
C.
Hence, for any K R \Sk+1 , there exists CK > 0, independent of , such that
n
uk+1, L (K) CK .
(4.2.22)
63
12
1 = uk+1, (0) C
B0 (0 )
u2k+1, dvgk+1,
and where is the Euclidean Laplacian. We also get that k+1, uk+1, uk+1
weakly in D12 (Rn ) as +. Thanks to Caffarelli-Gidas-Spruck [17],
(n2)/2
u (k+1 (x xk+1 ))
2k+1 |xk+1 |2
.
n(n 2)
2
, up to
Since Si with respect to Hk2 is a subset of Si with respect to Hk+1
2
1
changing k+1, into k+1 k+1, and xk+1 into k+1 xk+1 , (C3) of Hk+1 is
proved. By (4.2.17), the limit as + of xk+1, belongs
to2 S,
where S is the
s,
i
=
1,
.
.
.
,
k.
Hence,
(C2)
of
H
set
of
the
limits
of
the
x
i,
k implies (C2) of
2
Hk+1 . Noting that
min
i=1,...,k+1
Hk2
i=1,...,k
2
implies (C4) of Hk+1
. As already mentioned,
we also have that (C4) of
2
(C1) of Hk+1 follows from (4.2.15). We are thus left with the proof that (C5) of
2
Hk+1 holds. Let R > 0. For any > 0, there exists C independent of R and
such that
2
k+1
vi, dvg
u u0
M \Bxk+1, (Rk+1, )
i=1
2
k
0
(1 + )
vi, dvg
u u
M \Bxk+1, (Rk+1, )
i=1
2
+C
vk+1,
dvg .
M \Bxk+1, (Rk+1, )
Noting that
lim
lim
R+ +
M \Bxk+1, (Rk+1, )
2
vk+1,
dvg = 0 ,
64
CHAPTER 4
2
k+1
vi, dvg
u u0
M \Bxk+1, (Rk+1, )
i=1
2
k
0
vi, dvg + R ()
u u
M \Bx
(Rk+1, )
(4.2.23)
i=1
k+1,
where
lim
lim R () = 0 .
R+ +
Letting
2
k+1
vi, dvg
IR () =
u u0
Bxk+1, (Rk+1, )
i=1
2
k
0
vi, dvg ,
u u
Bx
(Rk+1, )
(4.2.24)
i=1
k+1,
k+1
k
vi, dvg
vi, dvg
u u0
u u0
M
M
i=1
i=1
(4.2.25)
+ IR () + R () .
Let
2
2
k+1
k
F = uk+1, u0
vi, ,
vi, uk+1, u0
i=1
i=1
where
n
2 1
vi, (x) = k+1,
vi, expxk+1, (k+1, x)
and
n
2 1
u0 (x) = k+1,
u0 expxk+1, (k+1, x) .
IR () =
Let
B0 (R)
VR =
F dvgk+1, .
1
x R s.t. d (x, Sk+1 )
R
where d is the Euclidean distance. Similar arguments to the ones we used to prove
(4.2.23) give that
F dvgk+1, R () .
B0 (R)VR
Hence
65
IR ()
B0 (R)\VR
F dvgk+1, + R () .
(4.2.26)
Given i {1, . . . , k}, either dg (xi, , xk+1, ) Ck+1, for some C > 0 independent of , or
dg (xi, , xk+1, )
+
k+1,
as +. In the first case, we get by (4.2.15) that
k+1,
+
i,
as +. Hence, given i {1, . . . , k}, either
dg (xi, , xk+1, )
+
k+1,
(4.2.27)
k+1,
+
i,
(4.2.28)
as +, or
2
vi,
dvg 0
Bxk+1, (Rk+1, )
2
vi,
dvgk+1, 0
B0 (R)\VR
(4.2.29)
as +. Let now i {1, . . . , k} be such that (4.2.28) holds and (4.2.27) does
not hold. Let also x B0 (R) \VR . For sufficiently large,
1
k+1, ,
dg xi, , expxk+1, (k+1, x)
2R
and it follows from the expression of vi, that
n2 1
i,
sup vi, CR
k+1,
B0 (R)\VR
for some CR > 0 independent of . In particular, using (4.2.28), we get that for
any R > 0,
sup
B0 (R)\VR
vi, 0
2
, we also have that for any R > 0,
as +. From (C3) of Hk+1
|uk+1, vk+1, |2 dvgk+1, 0
B0 (R)\VR
(4.2.30)
(4.2.31)
66
CHAPTER 4
(4.2.32)
where
lim
lim R () = 0 .
R+ +
From (4.2.24), (4.2.25), (4.2.32), and (C5) of Hk2 , it follows that
2
k+1
lim
vi, dvg lim E(u )2 u0 22 (k + 1)2min
u u0
+ M
+
i=1
2
and (C5) of Hk+1 holds. This ends the proof of step 3.
2
We are now in position to prove Theorem 4.1. As already mentioned, it follows
from steps 1 and 2 that there exists k N,
2 u0 22
1k
,
2min
2
holds and such that
such that HN
lim
lim
sup
R+ + xM \ (R)
n
N
R
(x) 2 1 u (x) u0 (x) = 0
where (R) = N
i=1 Bxi, (Ri, ). This proves Theorem 4.1.
Chapter Five
Asymptotics When the Energy Is of Minimal Type
Let (M, g) be a smooth compact Riemannian manifold of dimension n 3. We
let (h ) be a sequence of smooth functions on M , and consider equations like
g u + h u = u2
(E )
for all . For the readers convenience, we describe in this chapter the C 0 -theory
for the blow-up behavior of the u s when the energy of the u s is of minimal
type. The general situation of arbitrary energies is treated in the next chapter.
As in Chapter 4, we assume in what follows that there exist 0 < < 1 and a
smooth (or only C 0, ) function h on M such that
the operator g + h is coercive and
(5.0.1)
h h in C 0, (M ) as + .
We also assume in this section that the energy of the u s is of minimal type in the
sense that
E(u ) min
(5.0.2)
(n2)/2
Kn
2
2
|u | dvg +
h u dvg =
u2 dvg 2min .
M
2
h u dvg
M
2/n
|h |
n/2
dvg
u2 dvg
2/2
.
It follows that there exists C > 0 such that u 2 C for all , and then
that the u s are bounded in H12 (M ) since they are also bounded in L2 (M ) and
we have 2 < 2 . Up to a subsequence, we may therefore assume that for some
u0 H12 (M ), u u0 in H12 (M ) as +.
Assuming (5.0.1) and (5.0.2), one can prove that, up to a subsequence, either
u0 0, and u u0 strongly in C 2 (M ) as +, or u0 0 and u 0
weakly and not strongly in H12 (M ) as +. This is the subject of Proposition
5.1 of section 5.1. Sharp estimates on the u s are then proved in section 5.2.
68
CHAPTER 5
u2 dvg =
|u |2 dvg +
h u2 dvg
M
M
2
|u | dvg +
hu2 dvg
M
M
C1
|u |2 dvg +
u2 dvg
M
u2 dvg
C2
2/2
where C1 , C2 > 0 are independent of , so that there exists c > 0 such that
u 2 c for all . This proves that, if u0 0, then u 0 weakly and
not strongly in H12 (M ) as +. In the following we assume that u0 0.
A possible proof that the convergence u u0 holds in C 2 (M ) is as follows.
Thanks to regularity theory, it is easily seen that we need to prove only that the
u s are bounded in C 0 (M ). We proceed by contradiction and assume that, up to
4/(n2)
a subsequence, maxM u + as +. We let = u 2
and
1
u
= u 2 u . Then
2 1
g u
+ h u
= u
and
u
2 dvg = 1 .
As already mentioned, there exists c > 0 such that u 2 c. We also have that
Kn2 for all , and still up to a subsequence, we can assume that as
weakly in H12 (M ) as +, where u
is given by
+, and that u
u
(n2)/4 0
u . We let x M and > 0 be such that
u
=
1 n
2
max u
= u
(x ) =
M
Then 0 as +. We use below only the fact that there exists r0 > n/2
such that h
r0 C for all , where C > 0 is independent of , and h is the
69
negative part of h . This easily follows from (5.0.1). Given 0 < < ig , where ig
is the injectivity radius of (M, g), and x Rn such that |x| , we define
n2
expx ( x)
u
(x) = 2 u
where expx is the exponential map at x . Then
2 1
u
g u
+ h
= u
(x) = 2 h exp ( x) and g (x) = (exp ) g( x). For any
where h
x
x
2
|
u | dvg
|
u |2 dvg C and
B0 (R)
M
u
2 dvg
u
2 dvg C
B0 (R)
|( u
)|2 dvg 2
| |2 u
2 dvg + 2
2 |
u |2 dvg
Rn
Rn
Rn
2s
1/s
| | dvg
B0 (3/4 )
Rn
2/2
u
2 dvg
+ O(1) O(1) .
s are bounded in D12 (Rn ), where D12 (Rn ) is the homogeIt follows that the u
neous Sobolev space defined as the completion of C0 (Rn ) with respect to the
norm uD12 = u2 . Up to a subsequence, we can therefore assume that
u in D12 (Rn ) as +, u D12 (Rn ), u 0. In particular, u
u
u
in L2 (Rn ) as +, and thus
2 1 .
u2 lim inf u
+
(5.1.1)
+ |h
r0
Moreover, u
1, and B0 (1) h
dvg C for all , where C > 0 is indepen
dent of , and r0 > n/2 is as above. Hence, thanks to the De GiorgiNashMoser
iterative scheme, as stated in Lemma 3.4,
C
u L2 (B0 (1)) .
1 = max u
B0 (1/2)
It clearly follows that u 0. This proves the above claim. We let C0 (Rn ),
0, and R > 0 such that supp B0 (R). For large,
2 1
dvg (5.1.2)
(
u ) dvg +
dvg
h u
B0 (R)
B0 (R)
B0 (R)
70
CHAPTER 5
B0 (R)
2 rn
|
|h
C
u dvg
h
r0 = o(1)
where C > 0 is independent of , and r0 > n/2 is as above. Hence, passing to the
limit as + in (5.1.2), we get that, for all C0 (Rn ),
1
(u) dx 2
u2 1 dx .
(5.1.3)
K n Rn
Rn
By density, (5.1.3) is true for all D12 (Rn ), 0. Taking = u, it follows
that
1
2
|u| dx 2
u2 dx .
K
n
n
R
n R
Noting that u 0, using the sharp Euclidean Sobolev inequality and (5.1.1), we
can write that
1 22
|u|2 dx
1
1
1
2
Rn
u
dx
2.
2/2
2
Kn2
K
K
n
2
R
n
n
dx
n u
R
2
and that
|
u u
|2
lim
we get that
|
u u
|2 dvg =
|
u|2
|
u|dvg = lim
|
u u
|dvg = 0 ,
u
2 dvg + o(1) .
|
u u
|2 dvg = 1
M
u
2 dvg
u
2 dvg + o(1) .
2
u
dvg =
|
u u
|2 dvg + o(1) .
Bx (R )
Bx (R )
B0 (R)
2
u dx
| u
|2 dvg + o(1)
B0 (R)
71
u
2 dvg + o(1)
Bx (R )
|
u u
|2 dvg + o(1)
Bx (R )
u
2 dvg + o(1) .
1
M
11
u
2 dvg ,
M
and we get the contradiction we were looking for. This ends the proof that the
u
s, and thus the u s, are bounded in C 0 (M ). From regularity theory, and up to
a subsequence, we then get that u u0 strongly in C 2 (M ) as +. This
proves Proposition 5.1.
2
From now on we assume that u 0 weakly and not strongly in H12 (M ) as
+. As above, we assume in addition that (5.0.1) and (5.0.2) hold. Since
(5.0.1) implies the uniform coercivity condition (3.3.1) for large, there exists
4/(n2)
c > 0, independent of , such that u 2 c for all . We let = u 2
1
and u
= u 2 u . Then
2 1
+ h u
= u
g u
and
)
(E
u
2 dvg = 1 .
Kn2
lim inf
(5.1.4)
u
2 dvg = 1
+
Bx0 ()
(5.1.5)
72
CHAPTER 5
shows that, up to a subsequence, there exists R > 0, with the property that R 0
as R +, such that for any R > 0,
1 R
u
2 dvg + o(1)
(5.1.6)
Bx (R )
u
2 dvg
1 R lim inf
+
Bx0 ()
lim sup
+
u
2 dvg 1 .
Bx0 ()
lim
u
2 dvg = 1 .
+
Bx0 ()
(5.2.1)
u (x ) = max u =
M
n2
2
u (x) C
(5.2.2)
and that
lim
lim
sup
R+ + M \Bx (R )
dg (x , x)
n2
2
u (x) = 0 ,
(5.2.3)
where dg is the distance with respect to g. We let u = u10 , where u10 is as in Chapter
3, be the function defined in Rn by
1 n2
|x|2
.
u(x) = 1 +
n(n 2)
Then, u is a positive solution, the only one satisfying u(0) = 1 = maxRn u, of the
equation
u = u2
73
where is the Euclidean Laplacian. Its energy E(u) = u2 is E(u) = min ,
n2
2
(n2)/2
2
n2
x
u
.
=
2
2 + |x|
u(1
x)
n(n2)
n2
n2
2
2
u (x) C
d (x ,x)2
C 2 + dg (x ,x)2
2 + g
n(n2)
n(n2)
where x and are as above. Moreover, for any > 0, there exists > 0 such
that, up to a subsequence, for any and any x Bx0 ( ),
n2
n2
2
2
u (x) (1 + )
.
d (x ,x)2
1 + 2 + dg (x ,x)2
2 + g
n(n2)
n(n2)
2
2 1
+ h u
= u
g u
and
u
2 dvg = 1
)
(E
(5.2.4)
for all . We know (see the proof of Proposition 5.1) that there exists c > 0,
0 weakly in
independent of , such that u 2 c for all . Clearly, u
H12 (M ) as +, and (5.1.4), (5.1.5) hold. An easy claim is that
1
lim = 2 .
(5.2.5)
+
Kn
On the one hand, we know that Kn2 for all . On the other hand, since the
0 strongly
embedding of H12 (M ) in L2 (M ) is compact, we can assume that u
in L2 (M ). Thanks to the sharp Sobolev inequality, as proved by Hebey-Vaugon
[47, 48], there exists B > 0, independent of , such that
u 22 Kn2
u 22 + B
u 22 .
h u
2 dvg Kn2
u 22 + B
u 22 .
u 22 +
M
74
CHAPTER 5
It follows that
u 22
u 22 C
1 Kn2
for all , where C > 0 is independent of . Clearly, there exists c > 0 such that
s converge strongly to
u 2 c. If this is not the case, a subsequence of the u
0 in H12 (M ), contradicting (5.2.4). Passing to the limit as + in the above
equation, this proves (5.2.5).
We define
> 0 by the relations
1 n
2
u
(x ) = max u
M
2/(n2)
n2
2
u
(x) C
and that
lim
lim
sup
R+ + M \Bx (R
)
dg (x , x)
n2
2
u
(x) = 0
(5.2.6)
where dg is the distance with respect to g. It is easily checked that the equations in
Theorem 5.2 reduce to the existence of C > 1, independent of , such that for any
and any x,
n2
n2
2
2
1
u
(x) C
2/n
2/n
d (x ,x)2
d (x ,x)2
C
2 + n g
2 + n g
(5.2.7)
and, for any > 0, to the existence of > 0, independent of , such that for any
and any x Bx0 ( ),
n2
n2
2
2
(x)
2/n
2/n
d (x ,x)2
d (x ,x)2
C
2 + n g4
2 + n g4
(5.2.8)
where C = 1 + . We prove (5.2.7) and (5.2.8) in what follows. The proof
proceeds in several steps. We claim first that there exists C > 0 such that, up to a
subsequence,
n2
2
u
(x) C
2/n
2 + n4 dg (x , x)2
for all and all x M . It is easily checked that the proof of this claim reduces to
the proof that there exists C > 0 such that for any and any x M ,
(x )
u (x) C .
dg (x , x)n2 u
(5.2.9)
75
As a first step we prove that for any > 0 there exists C > 0 such that for any
and any x M ,
n2
2 +
dg (x , x)n2
u
(x) C .
(5.2.10)
In step 2 we prove that we can take = 0 in (5.2.10), so that (5.2.9) is true. Step
3 is concerned with the exact asymptotic profile of the u
s. The proof of (5.2.7)
and (5.2.8), and thus of Theorem 5.2, follows from this exact asymptotic profile.
S TEP 1. We prove that (5.2.10) is true. An easy claim is that it suffices to prove
(5.2.10) for > 0 small. Indeed, let 1 < 2 be positive. It is easily seen that
. On the other hand, if
(5.2.10) with respect to = 1 , 2 is true if dg (x , x)
, then
(5.2.10) is true with respect to 1 , and dg (x , x)
n2 +
2
2
u
(x)
dg (x , x)n22
n2 +1
2
1
dg (x , x)1
dg (x , x)n21
2
u
(x)
so that (5.2.10) with respect to 2 is also true. This proves the above claim. As
already mentioned, (5.0.1) implies the uniform coercivity condition (3.3.1). We
let h C 0, (M ) be such that, up to a subsequence, h h for all , and g + h
0
is coercive. Clearly, there exists 0 > 0 such that for > 0 small, g + h
1 is
coercive. We fix > 0 small, and let G be the Greens function of this operator.
By the maximum principle, G is positive. We let L be the operator
2 2
u.
L u = g u + h u u
= 0 with u
> 0 on M , we get from Berestycki-Nirenberg-Varadhan
Since L u
[9] that the maximum principle holds for L . It is easily seen that in M \{x },
(x , x)
L G1
|G |2 (x , x)
2 2
=
+
h
(x)
h(x)
(x)
+
(1
)
0
G2 (x , x)
G1
(x , x)
so that
(x , x)
L G1
|G |2 (x , x)
2 2
(x)
+
(1
)
0
G2 (x , x)
G1
(x , x)
(5.2.11)
A standard property of the Greens function we use in the following (we refer to
Appendix A for more details) is that there exist C > 0 and > 0 such that, for any
and any x Bx ()\{x },
|G |(x , x)
C
.
G (x , x)
dg (x , x)
(5.2.12)
Let R > 0, to be fixed later on. Combining (5.1.5) and (5.2.11), we get that for
sufficiently large
(x , x) 0
L G1
(x) R
dg (x , x)2 u
76
CHAPTER 5
0 +
.
1
dg (x , x)2
G (x , x)
We choose R > 0 sufficiently large such that C(1 ) R 0. Then
(x , x) 0 in Bx ()\Bx (R
), and we have proved that for suffiL G1
(x
,
x)
0
in
M
\B
). By another standard property
ciently large, L G1
x (R
of the Greens function (we refer here again to Appendix A for more details) there
),
exists C > 0 such that for any , and any x Bx (R
(x , x) C
(1)(n2)
.
G1
(1)(n2) n2
2
If we let C = C 1
(x , x) u
(x)
C G1
2 +
dg (x , x)n2
u
(x) C
G(y , x) (g u
+ h
u ) (x)dvg (x)
u
(y ) =
G(y , x) (g u
+ h u
) (x)dvg (x)
M
2 1
=
G(y , x)
u
(x)dvg (x) .
M
2 1
G(y , x)
u
(x)dvg (x)
M
2 1
=
G(y , x)
u (x)dvg (x) +
Bx ()
M \Bx ()
2 1
G(y , x)
u
(x)dvg (x) .
77
2 1
G(y , x)
u
(x)dvg (x)
M \Bx ()
( n2 )(2 1)
=O
n+2
2 (2 1)
M \Bx ()
M
n+2
(2 1)
=O
2
.
Therefore,
u
(y ) C
Bx ()
n+2
(2 1)
2 1
(5.2.13)
G(y , x)
u
(x)dvg (x) + O
2
2 1
2 1
G(y , x)
u
(x)dvg (x) C
u
dvg .
Bx ()
Bx ()
n+2
2 1
2
u
dvg
Bx (
)
n2
Bx (
)
dvg C
2 .
From (5.2.10), where we choose < 2/(2 1), we then get that
2 1
2 1
G(y , x)
u
(x)dvg (x) C
u
dvg
Bx ()
n+2
2 (2 1)
Bx (
)
+C
Bx ()\Bx (
)
It follows that
dg (x , y )
n2
Bx ()
dg (x , x)(+2n)(2
1)
dvg (x) .
n2
2 1
G(y , x)
u
(x)dvg (x) C
2 .
78
CHAPTER 5
dg (x , y )n2 u
(x )
u (y )
n2
C
2 dg (x , y )n2
1
B
n2
2
+C
2 1
G(y , x)
u
(x)dvg (x)
dg (x , y )n2
2
B
2 1
G(y , x)
u
(x)dvg (x)
1)
n2
+O
2(2
d
(x
,
y
)
g
1
where B = x Bx () s.t. dg (y , x) 12 dg (x , y ) , and B2 is given by the
equation B2 = Bx ()\B1 . By a standard property of the Greens function (we
refer to Appendix A) there exists C > 0 such that G(x, y) Cdg (x, y)2n for all
(x, y) M M , x = y. We once again fix > 0 such that < 2/(2 1). Then,
n2
2 1
dg (x , y )n2 u
(x )
u (y ) C
2
u
dvg
n2
2
+C
1
B
dg (x , y )n2
2 1
dg (y , x)2n u
2
B
n2
2 1
2
dvg C
1
B
for some C > 0 independent of . Independently, still using (5.2.10), we can write
that
n2
2
dg (x , y )n2
C
dg (x , y )
2 1
dg (y , x)2n u
(x)dvg (x)
2
B
2(2 1)
1
dg (x , y )2
2
B
2(2 1)
= o(1)
dg (x , y )
where C > 0 is independent of . Hence,
(x )
u (y ) C
dg (x , y )n2 u
where C > 0 is independent of . Summarizing these cases, we have
proved that,
up to a subsequence, the sequence dg (x , y )n2 u
(x )
u (y ) is bounded. In
particular, (5.2.9) is true.
C
From now on, we let G be the Greens function of the operator g + h , where
h is given by (5.0.1). Let also y0 M , and a sequence (y ) in M such that
y y0 as +.
S TEP 3. We claim that the two following asymptotics hold: if y0 = x0 , then, up
to a subsequence,
n2
2
2/n
n
n2
2
2
dg (x , y )
2 u
(y ) 1
(5.2.14)
4
79
+
dg (x , y )
2 u
(y )
4
(n 2)n1 dg (x0 , y0 )
n2
(5.2.15)
G(x0 , y0 )
G (y , x) (g u
+ h u
) dvg (x)
u
(y ) =
M
2 1
=
G (y , x)
u
(x)dvg (x)
M
2 1
=
G (y , x)
u
(x)dvg (x)
Bx ()
+
From step 2,
2 1
G (y , x)
u
(x)dvg (x) .
M \Bx ()
2 1
G (y , x)
u
(x)dvg (x)
M \Bx ()
n+2
2
=O
M \Bx ()
G (y , x)dvg (x)
n+2
=O
2
.
Therefore,
u
(y ) =
Bx ()
n+2
2 1
.
G (y , x)
u
(x)dvg (x) + O
2
(5.2.16)
(x) =
where h
2 h expx (
x) and g (x) = (expx ) g(
x). The u
s are
0 uniformly. It follows from standard
clearly bounded, and, from (5.0.1), h
80
CHAPTER 5
0,
Cloc
(Rn )
+.
Moreover,
it
is
easily
checked
that
u(0)
=
1
=
maxRn u,
Cloc
that Rn u2 dx = 1, and that
u =
1 2 1
u
Kn2
2/n
n
|x|2
1+
4
n2
2
where n is the volume of the unit n-sphere. Coming back to (5.2.16) we write
that
u (y )
u
(x )
n2
=
2
Bx ()
B0 (
1
)
2
2 1
G (y , x)
u
(x)dvg (x) + O
2 1
2
G y , expx (
x) u
(x)dvg (x) + O
.
2
lim u
(x )
u (y ) = Kn G(y0 , x0 )
u2 1 dx
+
Rn
4
2/n
n2
2
(n 2)n1 G(x0 , y0 )
lim dg (x , y )
n2
u
(x )
u (y ) = lim
dg (x , y )
n2
u
(z )
u
where u
is as above. Clearly, |z | R as +, and since, as above, u
dg (x ,y )
0
n
in Cloc (R ) as +, we get that if y0 = x0 and
R as +,
then
n2
2
2
R
(x )
u (y )
(5.2.17)
dg (x , y )n2 u
2/n
1 + n4 R2
81
n2
=
2 dg (x , y )n2
n2
2
2 1
(x)dvg (x)
G (y , x)
u
dg (x , y )n2
+ as
1
B
dg (x ,y )
2 1
G (y , x)
u
(x)dvg (x)
2
B
2
+O dg (x , y )n2
where B1 = x Bx () s.t. dg (y , x) 12 dg (x , y ) , and B2 is given by the
equation B2 = Bx ()\B1 . Clearly,
n2
2 1
G (y , x)
u
(x)dvg (x)
1
B
2 1
G y , expx (
x) u
(x)dvg (x)
1
1
where =
1
x). For x , the ratio
expx (B ). Let z = expx (
dg (y ,z )
+ as +. Writing that
dg (y , z ) dg (x , z ) dg (x , y ) dg (y , z ) + dg (x , z )
and noting that dg (x , z ) =
|x|, we get that
lim
dg (x , y )
= 1.
dg (y , z )
By a standard property of the Greens function (we refer to Appendix A for more
details) this implies that
1
lim dg (x , y )n2 G y , expx (
x) =
+
(n 2)n1
where n1 is the volume of the unit (n 1)-sphere. Because of the Lebesguedominated convergence theorem and step 2, we then get that
n2
2 1
2 dg (x , y )n2
G (y , x)
u
(x)dvg (x)
lim
+
1
=
(n 2)n1 Kn2
1
B
2 1
dx =
n2
2
2/n
n
where u is as above. On the other hand, from step 2 [see (5.2.9)] we can write that
n2
2 1
2 dg (x , y )n2
G (y , x)
u
(x)dvg (x)
C
C
dg (x , y )
dg (x , y )
Rn
2
2
2
B
1
dg (x , y )2
= o(1)
2
B
82
CHAPTER 5
dg (x ,y )
+ as
(5.2.18)
u
(x)
v (x) =
2/n
2 + n4 dg (x , x)2
and let y be such that v (y ) = minM v . Then, up to a subsequence, it follows
from (5.2.14) and (5.2.15) that v (y ) c for some c > 0 independent of .
Together with step 2, this proves (5.2.7). Independently,
lim dg (x0 , y)n2 G(x0 , y) =
yx0
1
.
(n 2)n1
Hence, given > 0, there exists > 0 such that if y Bx0 ( ), then
1
(n 2)n1 dg (x0 , y)n2 G(x0 , y) 1 + .
1+
If we let y and y be such that
y ) = max v ,
v (y ) = min v and v (
Bx0 ( )
Bx0 ( )
Chapter Six
Asymptotics When the Energy Is Arbitrary
Let (M, g) be a smooth compact Riemannian manifold of dimension n 3. We
let (h ) be a sequence of smooth functions on M , and consider equations like
g u + h u = u2
(E )
for all . We describe in this chapter the C 0 -theory for the blow-up behavior of the
u s when the energy of the u s is arbitrary.
As in Chapters 4 and 5, we assume in what follows that there exist 0 < < 1
and a smooth (or only C 0, ) function h on M such that
the operator g + h is coercive and
(6.0.1)
h h in C 0, (M ) as + .
We also assume that there exists > 0 such that E(u ) for all , where
E(u) = u2 is as in Chapter 2. The existence of u implies the coercivity of
the operator g + h . An estimate like the one in Theorem 6.1 below implies the
coercivity of the operator g + h (see Appendix B). Independently, as in Chapter
4, multiplying (E ) by u and integrating over M , we get that
2
2
|u | dvg +
h u dvg =
u2 dvg 2 .
M
2/n
n/2
2
h
u
dv
|h
|
dv
g
g
M
u2 dvg
2/2
|h |n/2 dvg C
M
for all , where C > 0 is independent of , we get that the u s are bounded in
H12 (M ). Up to a subsequence, we may therefore assume that for some function
u0 H12 (M ),
u u0 weakly in H12 (M )
0
(6.0.2)
84
CHAPTER 6
(6.0.3)
+ M
so that blow-up occurs. If this is not the case, up to a subsequence, we easily get
from standard regularity theory that u u0 strongly in C 2 (M ) as +,
and this provides a complete description of the behavior of the u s. We prove
in this chapter that the following theorem holds. The theorem was announced in
Druet-Hebey-Robert [32]. Applications of the theorem are in Druet [27].
T HEOREM 6.1 Let (M, g) be a smooth compact Riemannian manifold of dimension n 3, (h ) be a sequence of smooth functions on M such that (6.0.1) is satisfied and (u ) be a sequence of positive solutions to (E ) such that E(u )
for some > 0 and all . We assume that (6.0.3) holds. Then there exist N N ,
converging sequences (xi, ) in M , and sequences (i, ) of positive real numbers
converging to 0, i = 1, . . . , N , such that, up to a subsequence,
n2
2
N
1
i,
0
(1 ) u (x) +
C i=1 2 + dg (xi, ,x)2
i,
u (x) (1 + ) u0 (x) + C
n(n2)
N
i=1
i,
2i,
dg (xi, ,x)2
n(n2)
n2
2
n2
2
N
1
i,
0
(1 ) u (x) +
1 + i=1 2 + dg (xi, ,x)2
i,
n(n2)
u (x) (1 + ) u0 (x) + (1 + )
N
i=1
i,
2i, +
dg (xi, ,x)2
n(n2)
n2
2
for all , all x0 S, and all x Bx0 ( ), where S is the set consisting of the
limits of the xi, s as +. Outside the Bx ( )s, x S, the u s converge
C 2, to u0 . The estimate then extends to M in the particular case where u0 0.
We refer to section 6.3 [see, in particular, (6.3.24), (6.3.25), and (6.3.37)] for more
details and a refined statement.
85
Another complement to Theorem 6.1 is that the bubbles in this theorem satisfy
Theorem 3.1. More precisely, we also have that for any x M and any ,
n2
2
N
i,
0
u (x) = u (x) +
+ R (x)
d (xi, ,x)2
2i, + gn(n2)
i=1
where the R s, R H12 (M ) for all , are such that R 0 strongly in H12 (M )
as +. Moreover, if
1
1
|u|2 dvg
|u|2 + h u2 dvg
Ig (u) =
2 M
2 M
and
1
1
Ig (u) =
|u|2 dvg
|u|2 + h u2 dvg
2 M
2 M
are as in Chapter 3, then
N
Ig (u ) = Ig (u0 ) + Knn + r
n
where the r s, r R for all , are such that r 0 as +. We refer to
the end of section 6.3 for the proof of these claims.
Sections 6.16.3 below are devoted to the proof of Theorem 6.1. We set up
some notation in the rest of this introduction. We let the xi, s and i, s be as
in Theorem 4.1, i = 1, . . . , N , and choose N to be maximal. As a remark, the
xi, s in Theorem 6.1 will be slightly distinct from those in Theorem 4.1, but the
i, s are the same. Up to a subsequence, we may assume that for all i, j, k in
{1, . . . , N }, the 1
k, dg (xi, , xj, )s have a limit in [0, +] as +. We let
S = {xi, , i = 1, . . . , N }
(6.0.4)
and rearrange the xi, s in the following way. First, we let x1, be such that 1,
satisfies the condition 1, i, for all i = 1, . . . , N . We let N1 = 1 and
S1 = {xN1 +1, , . . . , xN2 1, }
be the subset of S consisting of the xi, s, i > N1 , which are such that
lim
dg (xN1 , , xi, )
< + .
N1 ,
dg (xN2 , , xi, )
< + .
N2 ,
86
CHAPTER 6
Given i = 1, . . . , p, we let
S0 = {xNi , , i = 1, . . . , p} ,
Si = xNi +1, , . . . , xNi+1 1, ,
S = p S i .
(6.0.5)
i=1
The
Sj s,
As a remark, it might be, in this process of rearranging the xi, s, that Ni+1 =
Ni + 1 or, equivalently, that Si = . In this case, xNi+1 , is chosen such that
Ni+1 , i, for all i = Ni + 1, . . . , N .
Given i = 1, . . . , p, we let
yi, = xNi , and i, = Ni , .
(6.0.6)
(6.0.7)
dg (yi, , yj, )
+
i,
(6.0.8)
dg (yj, , xi, )
R0
.
j,
2
(6.0.9)
(6.0.10)
Up to a subsequence, we may assume that there exists C > 0 such that, for any ,
sup
and that
lim
lim
sup
R+ + M \0 (R)
R (x) 2 1 u (x) C ,
n
M \0 (R0 )
n
R (x) 2 1 u (x) u0 (x) = 0
(6.0.11)
(6.0.12)
where 0 (R) = pj=1 Byj, (Rj, ). We prove (6.0.11) and (6.0.12) in what follows. Let (y ) be a sequence of points in M \0 (R0 ). We distinguish two cases.
In the first case we assume that there exists j {1, . . . , p} such that
dg (yj, , y )
R
j,
as +, where R R0 . By the above construction,
R (y ) = dg (yj, , y )
87
R0
l,
l,
i,
so that
dg (xl, , y )
+
l,
as +. In particular, for any R > 0, y M \ (R) for large, where
(R) is as in Theorem 4.1. By the second claim of proposition (P2) of Theorem
4.1, and since
1
N
R
(y ) R (y ) ,
2
we get that
n
R (y ) 2 1 u (y ) u0 (y ) 0
as +. This proves (6.0.12).
The rest of this chapter is divided as follows. Section 6.1 is devoted to the proof
of an upper estimate on the u s, similar to the one in Theorem 6.1 when we replace
S by S0 . In section 6.2 we prove the upper estimate on the u s with respect to
the full set S . Section 6.3 is devoted to the proof of the estimate from below in
Theorem 6.1, and to the asymptotics of u .
88
CHAPTER 6
p
2n
2
j,
dg (yj, , y)
(6.1.1)
j=1
where the yj, s and j, s are as in (6.0.6). We split the proof of (6.1.1) into
several steps that we refer to as claims. The proof of (6.1.1) is by induction.
First we introduce some notation. We let h0 C 0, (M ) be such that h0 < h
and g + h0 is coercive. We let G : M M \ {(x, x) , x M } R+ , be the
Greens function of the operator g + h0 . Then G is the only symmetric positive
function satisfying in the sense of distributions
g,y G (x, y) + h0 (y)G (x, y) = x
for all x M , where x is the Dirac mass at x. We know (see Appendix A) that
there exist > 0, C1 > 0 and C2 > 0 such that, for any x = y in M ,
dg (x, y) =
C1
|y G(x, y)|
G(x, y)
dg (x, y)
(6.1.2)
and
1
n2
dg (x, y)
G (x, y) C2 .
C2
(6.1.3)
89
p
Gj, (y)
j=1
where
Gj, (y) = G (yj, , y)
and G is the Greens function of g + h0 . Easy computations give that for any
y M \ {y1, , . . . , yp, },
L H (y) =
p
Gj, (y)
A (y)
(6.1.4)
j=1
where
2 2
|Gj, (y)|2
+ (1 )
Gj, (y)
p
Gj, (y )
(h (y ) (1 ) h0 (y ) + o (1))
j=1
0
since we fixed such that h > (1 ) h0 .
Case 2. We assume that dg (y0 , S) 2 , where S is given by Theorem 4.1
and is as in (6.1.2). We let i {1, . . . , p} be such that, up to a subsequence,
R (y ) = dg (yi, , y ). We write that
1
L H (y ) (1 ) Gi, (y )
2 2
u (y )
p
|Gi, (y )|2
2
Gi, (y )
1
Gj, (y )
j=1
L H (y ) (1 )
j=1
90
CHAPTER 6
L H (y ) (1 )
2 2
u (y )
(2n)(1)2
(2n)(1)
C21 pR (y )
C12
2
2 2
1
(1 ) 1 pC2 R (y ) u (y )
C2
R (y )
for sufficiently large. We fix R > R0 , depending only on , such that, for
large,
2
2 2
R (y ) u (y )
(1 ) C12
.
2p
C222
(6.1.5)
u (y) 2j, 2
(6.1.6)
for any j {1, . . . , p} and any y Byj, (R () j, ). This follows from point
(P1) of Theorem 4.1 and the choice of R0 we made in (6.0.9). Independently,
(6.1.3) implies that, for any j {1, . . . , p},
(2n)(1)
H C21 (R () j, )
(6.1.7)
u (y) 2j, 2
( n 1)(12) (2n)(1)
2j,2
j,
(n2)(1)
2C21 R ()
( n2 1)(12)
1,
H (y) .
In particular,
( n2 1)(12)
H (y)
u (y) C()1,
for all j {1, . . . , p} and all y Byj, R () j, . Noting that L u = 0, it
follows from (6.1.5) that
( n2 1)(12)
H u 0
L C()1,
in M \pj=1 Byj, R () j, . From the maximum principle, we then get that there
exists C() > 0, depending only on , such that
( n2 1)(12)
H (y)
u (y) C () 1,
91
in M \
pj=1
The content of our second claim is that the inequality in claim 6.1.1 holds also
with = 0. Claim 6.1.2 can be stated as follows:
C LAIM 6.1.2. Assume that u0 0. There exists C > 0 such that for any > 0
and any y in M \ pj=1 Byj, (R0 j, ),
1
2n
2
u (y) C1,
R (y)
n2
lim sup 1, 2 R (y )
+
u (y ) < + .
(6.1.8)
2 1
u (y )
G (y , y) u (y)
dvg
(6.1.9)
M
2 1
G (y , y) u (y)
dvg
j,
2 1
j, Byj, ()
G (y , y) u (y)
dvg
2 1
j, \Byj, ()
G (y , y) u (y)
dvg .
2 1
G (y , y) u (y)
dvg
j, Byj, ()
j, Byj, ()
2 1
u (y)
(6.1.10)
3
2
for
(6.1.11)
dvg
92
CHAPTER 6
2 1
u (y)
dvg
j, Byj, ()
2 1
j, Byj, (R()1, )
(n+2)(12)
2
u (y)
dvg
+C () 1,
dg (yj, , y)
(n+2)(1)
dvg .
2 1
u (y)
dvg
j, Byj, ()
1
Volg Byj, (R () 1, ) 2 u 22 1
(n+2)(12)
(n+2)(1)
2
+C () 1,
dg (yj, , y)
dvg
n2
2
M \Byj, (R()1, )
(n+2)(12)
2
= O 1, + O 1,
n2
= O 1,2
.
n(n+2)(1)
1,
n2
2 1
.
G (y , y) u (y)
dvg = O 1,2
j, Byj, ()
(6.1.12)
On the other hand, using claim 6.1.1 and (6.1.3), we can write that, for any > 0,
2 1
G (y , y) u (y)
dvg
j, \Byj, ()
( n+2
2 )(12)
C1,
( n+2
2 )(12)
= O 1,
Taking <
2
n+2 ,
j, \Byj, ()
G (y , y) dvg (y)
j, \Byj, ()
G (y , y) u (y)
n
2 1
.
dvg = O 1,
(6.1.13)
Coming back to (6.1.10), we get with (6.1.12) and (6.1.13) that for any j, with
j {1, . . . , p},
n
2 1
2 1
.
G (y , y) u (y)
dvg = O 1,
j,
93
Noting that M =
that (6.1.8) holds.
pj=1 j, ,
2 1
G (y , y) u (y)
dvg
k,
2 1
1
Bk,
G (y , y) u (y)
2
Bk,
dvg
2 1
G (y , y) u (y)
(6.1.14)
dvg
2 1
k, \Byk, ()
G (y , y) u (y)
dvg .
n
2 1
2 1
.
G (y , y) u (y)
dvg = O 1,
(6.1.15)
k, \Byk, ()
G (y , y) C2 dg (y , y)
2n
2n2 C2 R (y )
1
in Bk,
, so that
2 1
2n
G (y , y) u (y)
dvg CR (y )
1
Bk,
1
Bk,
k, Byk, ()
2 1
u (y)
2 1
u (y)
n
2 1
.
dvg = O 1,
dvg .
94
CHAPTER 6
It follows that
n
2 1
2n
2 1
.
G (y , y) u (y)
dvg = O 1,
R (y )
(6.1.16)
1
Bk,
2
, we have that
In Bk,
dg (yk, , y) dg (yk, , y ) dg (y , y)
1
R (y ) R (y )
2
1
R (y ) .
2
Let > 0, to be fixed later on. Since
dg (yk, , y )
+
k,
2
as +, we have that dg (yk, , y) R () k, for all y Bk,
and large.
Noting that for large and i = k,
k, Byi, (R () i, ) = ,
2
we can apply claim 6.1.1. It follows that, for any y Bk,
,
( n2 )(12)
(n2)(1)
dg (yk, , y)
.
u (y) C1,2
Together with (6.1.3) we then get that
2 1
G (y , y) u (y)
dvg
2
Bk,
( n+2 )(12)
(n+2)(1)
C1,2
R (y )
( n+2
2 )(12)
C1,
(n+2)(1)
R (y )
2
Bk,
G (y , y) dvg
By (
2n
R (y )
2
dg (y , y)
dvg .
( n+2 )(12)
2 1
2(n+2)(1)
R (y )
G (y , y) u (y)
dvg C1,2
2
Bk,
n
2 1
2n
C1, R (y )
R (y )
1,
(n+2)2
.
1
Let us assume that 1,
R (y ) + as +. We let > 0 be such that
2
< n+2 . Then,
n
2 1
2n
2 1
.
(6.1.17)
G (y , y) u (y)
dvg = O 1,
R (y )
2
Bk,
95
n2
lim sup 1, R (y )
+
M \ pj=1 Byj,
(R0 j, ), then
u (y ) < + .
2
( n 1)(12)
(2n)(1)
j, (y) = j,2
dg (yj, , y)
.
(6.1.18)
Then we say that (6.1.19)j holds if there exists C > 0, independent of , such that
for any > 0 and any y M \ pi=1 Byi, (R0 i, ),
j1
n 1
n
2n
2n
2
2 1
+ j, Rj, (y)
.
(6.1.19)j
i, dg (yi, , y)
u (y) C
i=1
i, (y) + j,
Rj, (y)
u (y) C ()
i=1
where the
are as in (6.1.18), the Rj, s are as in (6.1.18), the j, s are as
in (6.0.6), and the yj, s are as in (6.0.6).
i, s
1
2
= h (1 ) h0 > 0
(6.1.20)
(6.1.22)
where C1 , C2 , and are given by (6.1.2) and (6.1.3), and where h is given by
(6.1.20). As a starting point
we claim
that there exists R () > R0 such that for
any y M \ pi=1 Byi, R () i, and any k {1, . . . , j 1},
2
2 2
D ()
(6.1.23)
96
CHAPTER 6
2 2
D ()
(6.1.24)
2n
2
i,
dg (yi, , y )
we have that
u (y ) C
j2
(n2)
= i,
(2n)
dg (yi, , y )
i, ,
n
2 1
2n
i, dg (yi, , y )
n
2 1
2n
+ j1, Rj1, (y )
i=1
j2
R
(2n)
i,
n
2 1
2n
(y ) + j1, Rj1, (y )
i=1
Noting that
Rj1, (y ) = min {Rj, (y ) , dg (yj1, , y )} ,
we easily get that
2n
Rj1, (y )
2n
Rj, (y )
2n
+ dg (yj1, , y )
i=1
n
2n
2
Rj, (y )
C (j 1) R(2n) k, (y ) + j1,
Therefore
n 1
min {Rj, (y ) , dg (yk, , y )} 2 u (y )
C (j 1) R
(2n)
dg (yk, , y )
C (j 1) R
1 n
2
+
Rj, (y )
j1,
n
2 1
k,
(y ) +
1 n2
Rj, (y )
j1,
"
.
1 n2
Assuming that
Rj, (y )
+
j1,
as +, and choosing R sufficiently large, it easily follows from the above
equation that (6.1.24) holds. Let us assume now that
Rj, (y ) = O (j1, ) .
(6.1.25)
97
sup
sup
i=1,...,j1 yi,
i, (y)
( n 1)(12)
(2n)(1)
j,2
Rj, (y)
i, = y i, \ pk=1 Byk, (R () k, )
2
2 2
> D () .
(6.1.26)
(6.1.27)
j,
1.
j1,
(6.1.28)
1)(12)
(2n)(1)
Rj, (y )
(6.1.29)
98
CHAPTER 6
n2
4
dg (yk, , y ) 2
u (y )
j1
!
n
1
(n2)
Cdg (yk, , y ) 2
i, (y )
R ()
n
2 1
2n
"
i=1
+j1, Rj, (y )
!
n
1
(n2)
k, (y )
(j 1) R ()
Cdg (yk, , y ) 2
(n2) "
j1,
( n2 1)(12)
(2n)(1)
Rj, (y )
+j1,
Rj, (y )
!
n
(n2)
2 1
(j 1) R ()
k, (y )
Cdg (yk, , y )
(n2) "
j1,
+k, (y )
Rj, (y )
(n2) "
!
j1,
1 n
1 n
(
2
2 )(12)
.
+ R ()
C (j 1) R ()
Rj, (y )
Taking R () sufficiently large, this implies in turn that
Rj, (y ) = O (j1, ) .
(6.1.30)
Rj, (y )
j1,
n
( 2 1)(12) n2 1
= O k,
j1, .
Since k j 1, j1, k, . Thus,
dg (yk, , y ) = O (k, ) .
(6.1.31)
(6.1.32)
and
H (y) =
j1
p
( n 1)(12)
( n 1)(12)
1
1
i,2
Gi, (y)
+ j,2
Gi, (y)
i=1
i=j
99
where Gi, (y) = G (yi, , y). It is clear from point (P1) of Theorem 4.1, (6.0.7),
(6.0.9), (6.1.3), and (6.1.32) that there exists C () > 0 such that
u C () H on Byi, R () i,
(6.1.33)
for all i {1, . . . , p}. We let L be the linear operator given by
2 2
L (u) = g u + h u u
u.
u (y)
+ (1 )
Gi, (y)
p
!
n
( 1)(12)
1
+ j,2
h (y) (1 ) h0 (y)
Gi, (y)
i=j
2 2
u (y)
+ (1 )
(6.1.34)
Gi, (y)
u (y )
D ()
2
h (y ) (1 ) h0 (y ) u (y )
>0
u (y )
D ()
,
2
dg (yk, , y )
2
2 2
D () dg (yk, , y ) u (y )
D () .
2
Hence,
Rj, (y ) dg (yk, , y ) .
(6.1.35)
100
CHAPTER 6
i,2
Gi, (y )
i=1
( n2 1)(12)
+j,
Gl, (y )
p
( n2 1)(12)
j,
(1 )
Gi, (y )
|Gl, (y ) |2
2
Gl, (y )
2 2
u (y )
i=j
C12 ( n2 1)(12)
(2n)(1)2
dg (yk, , y )
C21 k,
j1
( n 1)(12)
(2n)(1)
2 2
2
1
u (y )
C2
i,
dg (yi, , y )
(1 )
i=1
C12 ( n2 1)(12)
(2n)(1)2
dg (y , yl, )
C21 j,
p
n
1
(12)
)
(
(2n)(1)
2 2
C21 j,2
dg (yi, , y )
u (y )
+ (1 )
i=j
C12
2
(1 ) 1
k, (y ) dg (yk, , y )
C2
j1
2 2
C21
i, (y ) u (y )
i=1
!
C12
( n 1)(12)
(2n)(1)2
(1 ) 1
+j,2
Rj, (y )
C2
"
2
2 2
.
C21 (p j + 1) Rj, (y ) u (y )
L H (y ) k, (y ) dg (yk, , y )
( n2 1)(12)
+j,
A,
(2n)(1)2
Rj, (y )
B,
where
A, = (1 )
C12
2
2 2
1
(j 1) dg (yk, , y ) u (y )
1 C2
C2
and
B, = (1 )
C12
2
2 2
1
(p j + 1) Rj, (y ) u (y )
.
1 C2
C2
101
2 2
Rj, (y ) u (y )
2 2
dg (yk, , y ) u (y )
D () .
D ()
,
2
k, , and that dg (yk, , y ) Rj, (y ). Since y
k, , here again
that y
2 2
u (y )
dg (yk, , y )
2
2 2
D () dg (yk, , y ) u (y )
D ()
2
so that dg (yk, , y ) . From (6.1.2), (6.1.3), and (6.1.34),
|Gk, (y ) |2
( n2 1)(12)
1
Gk, (y )
(1 )
L H (y ) k,
2
Gk, (y )
j1
( n 1)(12)
1
2 2
i,2
Gi, (y )
u (y )
i=1
p
( n2 1)(12)
j,
Gi, (y )
2 2
u (y )
i=j
C12
2
(1 ) 1
k, (y ) dg (yk, , y )
C2
j1
2 2
C21
i, (y ) u (y )
i=1
( n 1)(12)
(2n)(1)
2 2
(p j + 1) C21 j,2
Rj, (y )
u (y )
for large. Since y k, , i, (y ) k, (y ) for all i {1, . . . , j 1}.
Thus,
2
L H (y ) k, (y ) dg (yk, , y )
(
(p j + 1) C21 j,
n
2 1
)(12)
A,
(2n)(1)
Rj, (y )
(6.1.36)
2 2
u (y )
where A, is as in case 2. Since Rj, (y ) dg (y , yk, ), and using the definition (6.1.32) of j, ,
( n 1)(12)
(2n)(1)
Rj, (y )
j,2
( n 1)(12)
(2n)(1)
j,2
dg (yk, , y )
(j, max {Aj, ; 1})( 2
1)(12)
(2n)(1)
dg (yk, , y )
102
CHAPTER 6
L H (y ) k, (y ) dg (yk, , y )
C,
where
C, = (1 )
C12
2
2 2
1
(1 + o (1)) dg (yk, , y ) u (y )
.
1 pC2
C2
Since
2
2 2
dg (yk, , y ) u (y )
D () ,
and using the definition (6.1.22) of D (), we have that C, 0 for large.
Hence, L H (y ) 0 for large.
Case 4. We assume that
2 2
u (y )
D ()
2
2 2
dg (yk, , y ) u (y )
> D () .
2 2
Rj, (y ) u (y )
D () .
p
( n 1)(12)
1
2 2
j,2
Gi, (y ) u (y )
.
i=j
i, (y ) u (y )
L H (y ) C2
i=1
C12 ( n2 1)(12)
(2n)(1)2
dg (y , yl, )
1 j,
C2
p
n
( 1)(12)
(2n)(1)
2 2
C21 j,2
dg (yi, , y )
u (y )
+ (1 )
i=j
103
j1
2 2
i, (y ) u (y )
i=1
( n2 1)(12)
+j,
(2n)(1)2
Rj, (y )
B,
where
B, = (1 )
C12
2
2 2
1
(p j + 1) Rj, (y ) u (y )
1 C2
C2
1)(12)
(2n)(1)
Rj, (y )
(6.1.37)
Combining (6.1.37) with (6.1.33) and the fact that L u = 0 in M , the maximum
principle gives the existence of some C () > 0, independent of , such that for any
y M \ pi=1 Byi, (R () i, ), u (y) C () H (y). From (6.1.3) it follows
that there exists C () > 0, independent of , such that
j1
( n2 1)(12)
(2n)(1)
(6.1.38)
i, (y) + j,
Rj, (y )
u (y) C ()
i=1
for all y M \
pi=1
Byi, (R () i, ).
In order to end the proof of claim 6.1.3 (see the definition (6.1.32) of j, ) it
remains to prove that Aj, is bounded from above. We proceed by contradiction
and assume on the contrary that
Aj, +
(6.1.39)
k, (y )
( n 1)(12)
(2n)(1)
j,2
Rj, (y )
104
CHAPTER 6
(6.1.40)
u (y ) C ()
i, (y ) + k, (y ) .
i=1
Since y
k, ,
i,
(y )
k,
u (y ) C () jk, (y ) .
Independently, since y k, ,
n
dg (yk, , y ) 2
u (y ) D ()
n2
4
It follows that
n2
n
D () 4
1
dg (yk, , y ) 2 k, (y )
C () j
( n2 1)(12)
k,
dg (yk, , y )
so that
dg (yk, , y ) = O (k, ) .
(6.1.41)
(6.1.42)
u (y) C ()
i, (y) + j,
Rj, (y )
i=1
105
u (y )
0
i=1 i,
2n
2
(y ) + j,
Rj, (y )
< +
(6.1.43)
(6.1.44)
where
n
2n
2
0i, (y) = i,
dg (yi, , y)
2
small. Given i = 1, . . . , j 1, we let
We fix 0 < < n+2
( n2 1)(12)
(2n)(1)
Rj, (y)
i, = y i, , i, (y) j,
(6.1.45)
i=1
Let k {1, . . . , j 1}, and R () be as in claim 6.1.3. We claim that
By (R () k, )
k,
k,
(6.1.46)
(6.1.47)
(6.1.48)
106
CHAPTER 6
Rj, (z )
n
( 2 1)(12) n2 1
= O j,
k,
(n2)(1)
.
= O k,
dg (yk, , z )
( n2 1)(12)
n
2 1
= O k,
i,
(n2)(1)
.
= O max {k, , i, }
Byk, (R () j, )
j,
(6.1.49)
(6.1.50)
provided that is large. First we prove (6.1.49). Let k {j, . . . , p} and let
(z ) be a sequence of points in Byk, (R () j, ). Assume by contradiction that
. Then, from the definition (6.1.46) of
, we get that there exists i
z
j,
j,
with i {1, . . . , j 1} such that
( n 1)(12)
(2n)(1)
i, (z ) j,2
Rj, (z )
.
Since
Rj, (z ) dg (yk, , z ) = O (j, )
and since i j, we can write, from (6.0.7), that
n 1 (12)
i, ( 2 )
(n2)(1)
(n2)(1)
Rj, (z )
dg (yi, , z )
j,
n
( 1)(12) n2 1
= O i,2
j,
(n2)(1)
.
= O i,
107
(6.1.51)
(6.1.52)
(6.1.53)
j,
j,
j,
dg (yk, , yl, ) k,
O (1) .
k,
j,
In particular, using (6.0.7) and (6.0.8), we get that
dg (yk, , y )
+
j,
108
CHAPTER 6
as +. Hence, dg (yk, , y ) > dg (yl, , y ) for large, and, as an immediate consequence, we get that R (y ) = Rj, (y ). The above claim is proved. By
(6.0.11) we can then write that
n 1
n
n
Rj, (y ) 2
n2 1 2
1
j, u (y ) =
R (y ) 2 u (y )
Rj, (y )
j,
= O (1) ,
and we have proved that, if Rj, (y ) = O (j, ), then (6.1.43) holds. Conversely,
we assume, in addition to (6.1.53), that
Rj, (y )
+
(6.1.54)
j,
as +. Let G be the Greens function of g + h0 , where h0 is as above;
namely, h0 C 0, (M ) is such that g + h0 is coercive and h > h0 . From the
Greens representation formula and equation (E ),
2 1
u (y )
G (y , y) u (y)
dvg
(6.1.55)
M
Bk, = y k, , dg (y , y) dg (yk, , y ) ,
2
1
2
1
2
k,
k,
1
Byk, (R () k, ) Bk,
2 1
G (y , y) u (y)
dvg
k,
2 1
Byk, (R()k, )
G (y , y) u (y)
2 1
1 \B
Bk,
yk, (R()k, )
2
Bk,
dvg
G (y , y) u (y)
2 1
G (y , y) u (y)
(6.1.56)
dvg
dvg .
1
,
From (6.1.3), for any y Bk,
2n
G (y , y) C2 dg (y , y)
Hence,
2n
C2 2n2 dg (yk, , y )
2 1
Byk, (R()k, )
G (y , y) u (y)
2n
Cdg (yk, , y )
2n
Cdg (yk, , y )
dvg
Byk, (R()k, )
2 1
u (y)
dvg
1
Volg Byk, (R () k, ) 2 u 22 1
(6.1.57)
109
2 1
G (y , y) u (y)
dvg = O 0k, (y ) .
(6.1.58)
Byk, (R()k, )
2 1
G (y , y) u (y)
dvg
1 \B
Bk,
yk, (R()k, )
2n
Cdg (yk, , y )
Since <
2
n+2 ,
2 1
1 \B
Bk,
yk, (R()k, )
k, (y)
dvg .
it follows that
2 1
1 \B
Bk,
yk, (R()k, )
G (y , y) u (y)
2n
Cdg (yk, , y )
( n2 +1)(12)
k,
(n+2)(1)
M \Byk, (R()k, )
2n
Cdg (yk, , y )
dg (yk, , y )
dvg
( n2 +1)(12)(n+2)(1)+n
k,
.
2 1
1 \B
Bk,
yk, (R()k, )
dvg
G (y , y) u (y)
dvg = O 0k, (y ) .
(6.1.59)
2
,
Given y Bk,
u (y)
( n2 +1)(12)
(n+2)(1)
Ck,
dg (yk, , y )
.
2 1
G (y , y) u (y)
dvg
2
Bk,
( n2 +1)(12)
(n+2)(1)
Ck,
dg (yk, , y )
( n2 +1)(12)
(n+2)(1)
Ck,
dg (yk, , y )
2n
dvg .
dg (y , y)
By ( 12 dg (yk, ,y ))
2
Bk,
2n
dg (y , y)
dvg
110
CHAPTER 6
Hence,
2
Bk,
2 1
G (y , y) u (y)
( n2 +1)(12)
2(n+2)(1)
dvg Ck,
dg (yk, , y )
C0k,
(y )
k,
dg (yk, , y )
2(n+2)
2
so that by (6.1.53), and since < n+2
,
2 1
G (y , y) u (y)
dvg = o 0k, (y ) .
(6.1.60)
2 1
G (y , y) u (y)
dvg = O 0k, (y )
(6.1.61)
2
Bk,
k,
1
= y k, , dg (y , y) Rj, (y ) ,
2
1
2
Bk, = y k, , dg (y , y) < Rj, (y ) .
2
1
Bk,
By (6.1.49),
.
pi=j Byi, (R () j, )
j,
k,
i,
i=j
k=j
and we can write
that
j,
2 1
G (y , y) u (y)
dvg
2 1
p
i=j Byi, (R()j, )
+
+
1
k=j Bk,
p
k=j
2
Bk,
G (y , y) u (y)
2 1
G (y , y) u (y)
2 1
G (y , y) u (y)
dvg
(6.1.62)
dvg
dvg .
1
Let i {j, . . . , p}. By (6.1.54), dg (yi, , y ) j,
+ as +. It follows
1
that if z Byi, (R () j, ), then dg (z , y ) dg (yi, , y ) 1 as +.
From (6.1.3) we can then write that, for large,
2n
G (y , y) C2 dg (y , y)
2n
C2 2n2 Rj, (y )
111
in
pi=j Byi,
(R () j, ). As a consequence,
2 1
G (y , y) u (y)
dvg
p
i=j Byi, (R()j, )
2n
CRj, (y )
2 1
p
i=j Byi, (R()j, )
2 1
G (y , y) u (y)
p
i=j Byi, (R()j, )
2n
CRj, (y )
u (y)
dvg
dvg
1
u 22 1 Volg pi=j Byi, (R () j, ) 2
2 1
G (y , y) u (y)
dvg
p
i=j Byi, (R()j, )
n
2 1
2n
= O j, Rj, (y )
(6.1.63)
.
1
, from (6.1.3),
In Bk,
2n
G (y , y) C2 dg (y , y)
2n
C2 2n2 Rj, (y )
u (y)
( n +1)(12)
(n+2)(1)
Cj,2
dg (yk, , y)
.
2
,
As a consequence, since < n+2
2 1
G (y , y) u (y)
dvg
1
Bk,
2n
CRj, (y )
2n
CRj, (y )
2n
CRj, (y )
( n +1)(12)
j,2
( n +1)(12)
j,2
1
Bk,
(n+2)(1)
dg (yk, , y)
M \Byk, (R()j, )
dvg
(n+2)(1)
dg (yk, , y)
dvg
( n +1)(12)(n+2)(1)+n
j,2
.
It follows that
n
2 1
2n
2 1
.
G (y , y) u (y)
dvg = O j,
Rj, (y )
1
Bk,
2
In Bk,
, we have that
dg (yk, , y) dg (yk, , y ) dg (y , y)
1
Rj, (y ) Rj, (y )
2
1
Rj, (y )
2
(6.1.64)
112
CHAPTER 6
u (y)
( n +1)(12)
(n+2)(1)
Cj,2
Rj, (y )
.
( n +1)(12)
2 1
(n+2)(1)
G (y , y) u (y)
dvg Cj,2
Rj, (y )
2
Bk,
2
Bk,
G (y , y) dvg
( n +1)(12)
(n+2)(1)
Cj,2
Rj, (y )
2n
dg (y , y)
dvg
2
Bk,
( n +1)(12)
(n+2)(1)
Cj,2
Rj, (y )
2n
dvg
dg (y , y)
By ( 12 Rj, (y ))
( n +1)(12)
2(n+2)(1)
Cj,2
Rj, (y )
2(n+2)
n
j,
2n
2 1
Cj, Rj, (y )
.
Rj, (y )
2
, we then get that
By (6.1.54), and since < n+2
n
2 1
2n
2 1
.
G (y , y) u (y)
dvg = o j,
Rj, (y )
(6.1.65)
n
2 1
2n
2 1
Rj, (y )
G (y , y) u (y)
dvg = O j,
.
(6.1.66)
2
Bk,
j,
p
2n
2
j,
dg (yj, , y)
(6.1.67)
j=1
where the yj, s and j, s are as in (6.0.6). As in subsection 6.1.1, we split the
proof of (6.1.67) into several steps that we refer to as claims. Here again we proceed by induction. As in subsection 6.1.1, we let h0 C 0, (M ) be such that
113
h0 < h and g + h0 is coercive. We let G be the Greens function of the operator g + h0 . Our first claim is the following:
C LAIM 6.1.5. Assume that u0 0, and let 0 < < 21 . If is sufficientl small,
then there exist R() > 0 and C () > 0 such that for any > 0 and any y in
M \ pj=1 Byj, (R () j, ),
n
( 2 1)(12)
(2n)(1)
(2n)
R (y)
+ R (y)
u (y) C () 1,
where R is as in (6.0.10), the j, s are as in (6.0.6), and the yj, s are as in
(6.0.6).
Proof of claim 6.1.5. We fix 0 < < 12 and x0 S, where S is as in Theorem 4.1.
We let
(1 ) C12
(1 ) C12
;
(6.1.68)
D () = min
2(1)
4p
4p
C22
C
2
where C1 and C2 are given by (6.1.2) and (6.1.3). For as in (6.1.2), we let
0 < () < 2 and R () R0 be such that, for sufficiently large, and any
y Bx0 ( ()) \ pj=1 Byj, (R () j, ),
2
2 2
R (y) u (y)
+ h0 (y) h (y) D () and
(6.1.69)
2
2 2
+ (1 ) h0 (y) h (y) D () .
R (y) u (y)
The existence of () and R () easily follows from (6.0.12). Taking () sufficiently small, we may also assume that
dg (x0 , S\ {x0 }) 2 () .
(6.1.70)
1
2
H (y) =
Gj, (y)
and H (y) =
Gj, (y)
j=1
j=1
+ (1 )
|Gj, (y)|2
(y) =
p
(6.1.71)
Gj, (y)
and
L H2
Gj, (y)
2 2
j=1
+ (1 )
|Gj, (y)|2
2
Gj, (y)
(6.1.72)
114
CHAPTER 6
Gk, (y)
C12
C12
dg (y , yk, )
R (y )
Then, by (6.1.69),
C12
L H1 (y ) (1 ) Gk, (y )
2
D () R (y )
R (y )
p
Gj, (y )
j=1
and
L H2 (y ) (1 ) Gk, (y )
C12
D () R (y )
R (y )
p
Gj, (y )
j=1
C2
Gk, (y )
Gj, (y ) C2 dg (yj, , y )
2n
C2 R (y )
Hence,
L H1
(2n)(1)2
C12
pD () C21
(1 ) 1
C2
(y ) R (y )
and
(2n)2
L H2 (y ) R (y )
(1 )
C12
pD () C2
C2
.
(6.1.73)
115
for large. It easily follows from point (P1) of Theorem 4.1, (6.0.7), (6.0.9), and
(6.1.3), that there exists C () > 0, independent of , such that for any j with
j {1, . . . , p},
( n2 1)(12)
1
Gj, (y)
on Byj, (R () j, ) .
u (y) C () 1,
(6.1.74)
Independently, it easily follows from (6.1.70) and point (P1) of Theorem 4.1 that
u (y) C () H2 (y) on Bx0 ( ())
(6.1.75)
2n
2
u (y) u0 (y) C1,
R (y)
sup |z|n2 u (z yj )
j=1,...,p zRn
(6.1.76)
where u is given by (4.1.5). It easily follows from point (P1) of Theorem 4.1 that,
up to a subsequence,
sup
M \p
j=1 Byj, ( )
|u u0 | 0 as +
(6.1.77)
116
CHAPTER 6
Case 2. We assume that there exists k {1, . . . , p} such that for some R R0 ,
dg (yk, , y )
R
k,
as +. By (6.0.9) and point (P1) of Theorem 4.1, we have in that case that
n
1
lim 2 u
+ k,
(y ) = u (y0 yk )
where, up to a subsequence,
y0 = lim
Independently,
n
2
lim k,
1
exp1
yk, (y ) .
k,
2n
2
k,
dg (yk, , y )
= |y0 |2n
so that
lim
u (y )
n
2 1
2n
k, dg (yk, , y )
lim
n
2 1
2n
1, R (y )
1
D0
2
2n
2
R (y )
1,
|u (y ) u0 (y ) |
1, 2 R (y )
n2 1
n
R (y )
1
=
R (y ) 2 |u (y ) u0 (y ) |
1,
n
1
= O R (y ) 2 |u (y ) u0 (y ) |
n2
= o (1) .
It follows that (6.1.78) holds with C > 0.
117
R (y )
+
1,
as +. We let G be the Greens function of g + h . We write with the
Greens representation formula that
2 1
2 1
0
u (y ) u (y ) =
G (y , y) u (y)
u0 (y)
dvg
M
(6.1.79)
+
G (y , y) (h (y) h (y)) u0 (y) dvg .
R (y ) 0 and
We let C2 > 0 (we refer to Appendix A) be such that, for any > 0 and any
x, y M , x = y,
1
2n
2n
dg (x, y)
G (x, y) C2 dg (x, y)
.
(6.1.80)
C2
We clearly have that
and that
M \p
i=1 Byi, ( )
2 1
2 1
dvg = o (1)
G (y , y) u (y)
u0 (y)
where is as in (6.1.77). Coming back to (6.1.79), using (6.1.81) and the above
equation, we get that
2 1
0
u (y ) u (y )
G (y , y) u (y)
dvg + o (1) . (6.1.82)
p
i=1 Byi, ( )
1
= y j, Byj, ( ) s.t. dg (y , y) R (y ) ,
2
1
= y j, Byj, ( ) s.t. dg (y , y) < R (y ) .
2
1
We fix 0 < < n+2
small. Let j {1, . . . , p}. Up to increasing so that
1, = o( ), we can write that
2 1
G (y , y) u (y)
dvg
j, Byj, ( )
2 1
j, Byj, (R()1, )
G (y , y) u (y)
1 \B
Bj,
yj, (R()1, )
2 \B
Bj,
yj, (R()1, )
dvg
2 1
G (y , y) u (y)
2 1
G (y , y) u (y)
(6.1.83)
dvg
dvg
118
CHAPTER 6
1
where R () is given by claim 6.1.5. Since R (y ) 1,
+ as +, we
can write that, for large and any y j, Byj, (R () 1, ),
dg (y , y) dg (y , yj, ) dg (yj, , y)
R (y ) R () 1,
1
R (y ) .
2
Since u 2 , using (6.1.80) and Holders inequality, we have that, for
large,
2 1
G (y , y) u (y)
dvg
j, Byj, (R()1, )
2n
C2 2n2 R (y )
2 1
j, Byj, (R()1, )
u (y)
dvg
1
u 22 1 Volg Byj, (R () 1, ) 2
21
2n 2 1 n1
n n
R () 1,
2n1 C2 R (y )
n
1
n
n
1
2n
n1
2 1 n1 2
2 1
1,
2
C2
R () 2
R (y )
.
n
2n
C2 2n2 R (y )
2 1
j, Byj, (R()1, )
G (y , y) u (y)
n
2 1
dvg
(6.1.84)
2n
D1 1, R (y )
where
D1 = 2n1 C2 2
n1
21
R () 2
(6.1.85)
1
Given y Bj,
, we get from (6.1.80) that
2n
G (y , y) C2 2n2 R (y )
1
and, if y Bj,
\Byj, (R () 1, ), we get from claim 6.1.5 that
2 1
u (y)
22
( n2 +1)(12)
(n+2)(1)
dg (yj, , y)
1,
"
(n+2)
+dg (yj, , y)
.
2 1
C ()
119
2 1
G (y , y) u (y)
dvg
1 \B
Bj,
yj, (R()1, )
22
2 1
C ()
(n+2)
1 \B
Bj,
yj, (R()1, )
2 1
+22 +n4 C ()
2n
C2 R (y )
1
n+2 ,
dg (yj, , y)
dvg .
1 \B
Bj,
yj, (R()1, )
G (y , y) dg (yj, , y)
dvg
( n2 +1)(12)
1,
(n+2)(1)
1 \B
Bj,
yj, (R()1, )
G (y , y) dg (yj, , y)
G (y , y)
n
n1
n1
n
dvg
= O 1(n+2)
= o (1) .
Similarly, it is easily checked that
1 \B
Bj,
yj, (R()1, )
n1
n(n+2)
Byj, ( )
dvg
dg (yj, , y)
(n+2)(1)
dg (yj, , y)
dvg
dvg
2n1
n(n+2)(1)
(R () 1, )
(n + 2) (1 ) n
for large. Therefore,
2 1
G (y , y) u (y)
dvg
1 \B
Bj,
yj, (R()1, )
n
2 1
(6.1.86)
2n
D2 1, R (y )
where
D2 = 22 +n3
+ o (1)
n1
2 1
n(n+2)(1)
C ()
C2 R ()
. (6.1.87)
(n + 2) (1 ) n
2
, we have that
In Bj,
dg (yj, , y) dg (yj, , y ) dg (y, y )
1
R (y ) R (y )
2
1
R (y )
2
so that, applying claim 6.1.5, we get that
!
2 1
2 1
(n+2)
dg (yj, , y)
22 2 C ()
u (y)
"
( n2 +1)(12)
(n+2)(1)
.
+2(n+2)(1) 1,
R (y )
120
CHAPTER 6
Once again, we used the fact that Byi, (R () i, ) j, = for large and any
i = j. Using (6.1.80) we then get that
2 1
G (y , y) u (y)
dvg
2 \B
Bj,
yj, (R()1, )
22
2+(n+2)(1)
2 \B
Bj,
yj, (R()1, )
2 2
+2
22
2 1
C ()
2 1
2 1
C ()
By (
2 1
dg (y , y)
)
G (y , y) n1 dvg
M
= O 1,
n(n+2)
dvg
2(n+2)(1)
R (y )
2n
2
R (y )
= O 1,
n1
n
n1
dg (yj, , y)
( n2 +1)(12)
dvg
dvg
n
C ()
Byj, ( )
G (y , y) dg (yj, , y)
( n2 +1)(12)
(n+2)(1)
1,
R (y )
2n
R (y )
2
(n+2)
2 \B
Bj,
yj, (R()1, )
+22
G (y , y) dvg
C ()
2+(n+2)(1)
C2
( n2 +1)(12)
(n+2)(1)
1,
R (y )
1,
R (y )
+ o (1)
2(n+2)
+ o (1) .
1
+ as +, it follows that
Since R (y ) 1,
2 1
G (y , y) u (y)
dvg
2 \B
Bj,
yj, (R()1, )
n
2 1
2n
= o 1, R (y )
(6.1.88)
+ o (1) .
Substituting (6.1.84), (6.1.86), and (6.1.88) into (6.1.83), we then get that, for any
j {1, . . . , p},
2 1
G (y , y) u (y)
dvg
j, Byj, ( )
(6.1.89)
n
2n
2
(D1 + D2 + o (1)) 1,
R (y )
+ o (1)
where D1 and D2 are given by (6.1.85) and (6.1.87). Since M = pj=1 j, and
Byi, ( ) j, Byj, ( ) j, ,
we get with (6.1.82) and (6.1.89) that
n
2n
2
u (y ) u0 (y ) p (D1 + D2 + o (1)) 1,
R (y )
+ o (1) .
121
(6.1.90)j
i=1
where Rj, is defined by (6.1.18). By claim 6.1.6, (6.1.90)j holds for j = 1. The
subject of claims 6.1.7 and 6.1.8 below is to prove that, for any j {2, . . . , p},
(6.1.90)j1 holds (6.1.90)j holds.
(6.1.91)
(6.1.92)
For k = 1, . . . , j 1, we let
(6.1.93)
k, = y M s.t. k, (y) i, (y) for all i {1, . . . , j 1}
where
( n 1)(12)
(2n)(1)
dg (yi, , y)
.
i, (y) = i,2
We also let
D () = min
(1 ) C12
(1 ) C12
;
2(1)
4p
4p
C22
C
(6.1.94)
(6.1.95)
122
CHAPTER 6
2 2
() u0 (y)
D ()
100
(6.1.97)
when y Bx ( ());
(A3) for any x S, and any y Bx ( ()) \ pi=1 Byi, (R () i, ),
!
"
2
2 2
h (y) + h0 (y) D ()
(6.1.98)
R (y) u (y)
when is sufficiently large.
The proof of (A1) is based on (6.1.90)j1 and (6.0.12). Let (y ) be a sequence
of points in M \ pi=1 Byi, (Ri, ), R > R0 to be fixed later on, such that y
k, for some k {1, . . . , j 1}. In order to prove (A1) we have to prove that,
choosing R sufficiently large,
n
1
u (y ) u0 (y )
min {Rj, (y ) , dg (yk, , y )} 2
1
(6.1.99)
D () 2 2
<
100
for large. From (6.1.90)j1 ,
u (y ) u0 (y )
j2
n 1
n
2n
2n
2
2 1
C
i, dg (yi, , y )
+ j1, Rj1, (y )
+ o (1) .
i=1
Since
Rj1, (y ) = min {Rj, (y ) ; dg (yj1, , y )} ,
it is clear that
2n
Rj1, (y )
2n
Rj, (y )
2n
+ dg (yj1, , y )
2n
2
i,
dg (yi, , y )
(n2)
= i, (y ) i,
(2n)
dg (yi, , y )
it follows that
u (y ) u0 (y )
j1
n 1
n
2n
2n
2
2 1
C
+ j1, Rj, (y )
+ o (1)
i, dg (yi, , y )
i=1
C R(2n)
j1
i=1
n
2 1
2n
i, (y ) + j1, Rj, (y )
+ o (1)
"
!
n
2n
2 1
Rj, (y )
+ o (1) .
C (j 1) R(2n) k, (y ) + j1,
123
Therefore,
u (y ) u0 (y )
n2 1
n
j1,
(2n)
2 1
dg (yk, , y )
k, (y ) +
+ o (1)
C (j 1) R
Rj, (y )
n2 1
n
j1,
+ o (1) .
C (j 1) R1 2 +
Rj, (y )
n
1
+ as +, then (6.1.99) holds with R sufficiently
If Rj, (y ) j1,
large. Without loss of generality, we may thus assume that
Rj, (y ) = O (j1, ) .
(6.1.100)
j1,
j1,
j1,
dg (yl, , yi, ) i,
Rj, (y )
i,
j1,
j1,
dg (yl, , yi, )
O (1) .
i,
By (6.0.8), we then get that
dg (y , yi, )
+
j1,
as +, and it follows from (6.1.100) that, for any i {1, . . . , j 1}, we
have that dg (yi, , y ) > Rj, (y ) when is large. This proves that Rj, (y ) is
such that Rj, (y ) = R (y ) for large. Using (6.0.12), we then get that
n
Rj, (y ) 2
|u (y ) u0 (y ) | R + o (1)
sup
sup
i=1,...,j1 yi,
i, (y)
( n 1)(12)
(2n)(1)
j,2
Rj, (y)
(6.1.101)
where
i, = y i, \ pk=1 Byk, (R () k, ) s.t.
1
n
D () 2 2
0
2 1
.
u (y) u (y)
dg (yi, , y)
100
(6.1.102)
124
CHAPTER 6
+
j1,
We prove (6.1.103) in what follows. We assume that Aj, = . Then, up to a
subsequence, there exist k {1, . . . , j 1} and y k, such that
k, (y ) = (Aj, j, )( 2
1)(12)
(2n)(1)
Rj, (y )
(6.1.104)
n
2 1
2n
+j1, Rj, (y )
+ o (1)
Cdg (yk, , y )
n
2 1
R ()
(2n)
2n
i, (y )
i=1
2
Rj, (y )
+j1,
j1
+ o (1)
Cdg (yk, , y )
n
2 1
( n2 1)(12)
+j1,
(2n)
(j 1) R ()
(2n)(1)
Rj, (y )
Cdg (yk, , y )
+k,
(y )
CR ()
1 n
2
n
2 1
j1,
Rj, (y )
(2n)
(j 1) R ()
j1,
Rj, (y )
k, (y )
(n2)
+ o (1)
k, (y )
(n2)
(j 1) + R ()
+ o (1)
(n2)
j1,
Rj, (y )
(n2)
+ o (1) .
(6.1.105)
125
dg (yk, , y )
n 1 (12)
k, ( 2 )
(n2)(1)
Rj, (y )
j1,
n
( 2 1)(12) n2 1
= O k,
j1, .
(6.1.106)
(6.1.107)
u (y)
C ()
j1
i,
( n2 1)(12)
(y) + j,
(2n)(1)
Rj, (y)
(2n)
+ R (y)
i=1
and
H (y) =
j1
( n 1)(12)
1
i,2
Gi, (y)
i=1
( n 1)(12)
1
+j,2
Gi, (y)
p
i=j
p
i=1
Gi, (y)
126
CHAPTER 6
where Gi, (y) = G (yi, , y), and G is the Greens function of g + h0 . We fix
x0 S, where S is as in Theorem 4.1. We let
0 < () < ,
2
1
() < dg (x0 , S\ {x0 })
2
be such that (6.1.97) and (6.1.98) hold, where is as in (6.1.2), (6.1.3). Since
h > (1 ) h0 , direct computations give that, for any y M \ {y1, , . . . , yp, }
and for large,
j1
( n 1)(12)
1
L H (y)
i,2
Gi, (y)
i=1
|Gi, (y) |2
(1 )
Gi, (y)
2 2
u (y)
( n 1)(12)
1
Gi, (y)
+ j,2
p
i=j
|Gi, (y) |2
(1 )
+
p
Gi, (y)
Gi, (y)
(6.1.108)
2 2
u (y)
2 2
i=1
+ (1 )
|Gi, (y) |2
.
2
Gi, (y)
We claim that, for sufficiently large,
L H 0 in Bx0 ( ()) \ pi=1 Byi, (R () i, ) .
Let (y ) be a sequence of points in Bx0 ( ()) \ pi=1 Byi, (R () i, ). We let
also l {1, . . . , p} be such that R (y ) = dg (yl, , y ). Since () < 2 , it is
clear that R (y ) for large. Thus we can apply (6.1.2) and (6.1.3). We get
that
p
|Gi, (y ) |2
2 2
Gi, (y ) h (y ) h0 (y ) u (y )
+ (1 )
2
Gi, (y )
i=1
1
C12
(2n)
dg (yl, , y )
(1 )
2
C2
dg (yl, , y )
p
"
!
2 2
Gi, (y ) u (y )
h (y ) + h0 (y ) .
i=1
From (6.1.98),
p
Gi, (y )
2 2
u (y )
"
h (y ) + h0 (y )
i=1
2
D () R (y )
p
i=1
Gi, (y )
127
so that, by (6.1.3),
p
Gi, (y )
2 2
u (y )
"
h (y ) + h0 (y )
i=1
2
D () R (y )
D () pC2 R
C2
p
(2n)
dg (y , yi, )
i=1
2+(2n)
(y )
2 2
Gi, (y ) h (y ) h0 (y ) u (y )
i=1
|Gi, (y ) |2
+ (1 )
(6.1.109)
Gi, (y )
3pD () C2 R
2+(2n)
(y )
for large. Now we estimate the other terms in the right-hand side of (6.1.108).
Up to a subsequence, we may assume that y k, for some k {1, . . . , j 1}.
We distinguish four cases.
Case 1. We assume that min {Rj, (y ) ; dg (yk, , y )} as +
for some > 0. Then, by (6.1.96), u (y ) = O (1). From (6.1.3), and since
y k, , this gives in turn that
j1
j1
( n2 1)(12)
2 2
1
u (y )
i,
Gi, (y )
=O
i, (y )
i=1
=O
i=1
k,
(y ) .
u (y )
j1
( n 1)(12)
1
i,2
Gi, (y )
= o (1)
(6.1.110)
i=1
since dg (yk, , y )
that
2 2
u (y )
i=j
n
( 1)(12)
(2n)(1)
= O j,2
Rj, (y )
n
( 1)(12)
.
= O j,2
By (6.1.103) [see also (6.1.107)] j, 0 as +. Hence,
( n 1)(12)
1
j,2
Gi, (y )
= o (1) .
p
2 2
u (y )
i=j
(6.1.111)
128
CHAPTER 6
i=j
(1 )
|Gi, (y ) |2
(6.1.112)
2 2
u (y )
Gi, (y )
( n 1)(12)
2+(2n)(1)
(3p + j 1) D () C21 j,2
Rj, (y )
.
Since y k, ,
1
n
D () 2 2
0
2 1
u (y ) u (y )
dg (yk, , y )
100
so that
2
2 2
dg (yk, , y ) u (y )
D ()
(6.1.113)
2 0
2 2
2 3
+ dg (yk, , y ) u (y )
.
max 1; 2
100
If k Aj (x0 ), Aj (x0 ) as in (6.1.92), then, since () < 12 dg (x0 , S\ {x0 }), we
get that dg (yk, , y ) as + for some > 0. In particular, we get with
(6.1.113) that u (y ) = O (1). As in the first case, we can then prove that
j1
( n 1)(12)
2 2
1
i,2
Gi, (y )
= o (1)
(6.1.114)
u (y )
i=1
and (6.1.114) holds under the assumption that k Aj (x0 ). Let us now assume
that k Aj (x0 ). Then dg (yk, , y ) 2 () for large so that, combining
(6.1.97) and (6.1.113),
2
2 2
D ()
dg (yk, , y ) u (y )
for large. From (6.1.2) and (6.1.3), this implies that
j1
|Gi, (y ) |2
( n2 1)(12)
1
2 2
i,
Gi, (y )
u (y )
(1 )
2
Gi, (y )
i=1
( n2 1)(12)
k,
dg
C21
(2n)(1)
(yk, , y )
2
C21 D () dg (yk, , y )
(1 )
i=1
dg (yk, , y )
j1
( n 1)(12)
(2n)(1)
i,2
dg (yi, , y )
C12
2
(1 ) 1
dg (yk, , y ) k, (y )
C2
C21 D () dg (yk, , y )
C12
j1
i=1
i, (y ) .
129
(1 )
|Gi, (y ) |2
2
Gi, (y )
2 2
u (y )
(6.1.115)
C12
2
dg (yk, , y ) k, (y )
C21
2
D () C21 dg (yk, , y )
j1
i, (y ) .
i=1
(1 )
|Gi, (y ) |2
2
Gi, (y )
2 2
u (y )
(6.1.116)
( n 1)(12)
2
(2n)(1)
(p j + 1) D () C21 j,2
dg (yk, , y ) Rj, (y )
( n 1)(12)
2+(2n)(1)
(p j + 1) D () C21 j,2
dg (yk, , y )
.
By (6.0.7) and (6.1.103), since k j 1 j,
j, k, (1 + o (1)) .
Therefore,
p
( n 1)(12)
2 2
1
u (y )
Gi, (y )
j,2
(6.1.117)
i=j
2
(p j + 1) D () C21 dg (yk, , y )
k, (y ) (1 + o (1)) .
130
CHAPTER 6
Coming back to (6.1.108), we get with (6.1.109), (6.1.116), and (6.1.117) that,
for sufficiently large, L H (y ) 0.
Case 4. We assume that y k, and that min {Rj, (y ) ; dg (yk, , y )} 0
as +. Since y k, ,
n
dg (yk, , y ) 2
u (y ) u0 (y )
D ()
100
212
.
2 2
Rj, (y ) u (y )
D ()
+ o (1) .
100
u (y )
k, ,
j1
( n 1)(12)
1
i,2
Gi, (y )
i=1
2
D () Rj, (y )
C21
j1
i, (y )
i=1
2 1
C2 k,
(j 1) D () Rj, (y )
(y ) .
1)(12)
(2n)(1)
Rj, (y )
( n 1)(12)
(2n)(1)
j,2
Rj, (y )
.
Hence, for large,
2 2
u (y )
j1
( n 1)(12)
1
i,2
Gi, (y )
(6.1.119)
i=1
(
(j 1) D () C21 j,
n
2 1
)(12)
2+(2n)(1)
Rj, (y )
Coming back to (6.1.108), we get with (6.1.109), (6.1.118), and (6.1.119) that,
for sufficiently large, L H (y ) 0.
131
(6.1.120)
Independently, using point (P1) of Theorem 4.1 and (6.1.3), we easily get the existence of some C () > 0 such that
u C () H on Bx0 ( ()) \ pi=1 Byi, (R () i, ) .
(6.1.121)
Since L u = 0, noting that L satisfies the maximum principle (see [9]), it follows from (6.1.120) and (6.1.121) that
u C () H in Bx0 ( ()) \ pi=1 Byi, (R () i, ) .
From (6.1.3), this proves that the estimate of claim 6.1.7 holds when we replace M
by a ()-neighborhood of S. As a straightforward consequence of point (P1) of
Theorem 4.1, we also have that such an estimate holds outside this neighborhood
of S. This ends the proof of claim 6.1.7.
2
The last claim of subsection 6.1.2 is as follows.
C LAIM 6.1.8. Assume that u0 0. If (6.1.90)j1 holds for some j = 2, . . . , p;
then (6.1.90)j holds also.
Proof of claim 6.1.8. The first part of the proof is devoted to proving the following estimate: there exists C > 0 such that for any sequence (y ) of points in
M \ pi=1 Byi, (R0 i, ),
u (y ) u0 (y )
j1
n
2n
0
2 1
C
i, (y ) + j, Rj, (y )
+ o (1)
(6.1.122)
i=1
where
n
2n
2
dg (yi, , y)
0i, (y) = i,
(6.1.123)
i,
i,
i,
j,
where i, is as in (6.1.93), i, is as in (6.1.94), and j, is as in (6.1.107). We
also let
i, .
j, = M \ j1
i=1
(6.1.125)
M \p
i=1 Byi, ( )
|u u0 | 0
(6.1.126)
132
CHAPTER 6
(6.1.127)
By (R () k, )
k,
k,
and
=
Byi, (R () i, )
k,
(6.1.128)
133
Byk, (R () j, )
j,
(6.1.129)
(6.1.130)
Rj, (z )
j,
n
( 2 1)(12) n2 1
= O i,
j,
.
By (6.1.103), j, = O (j1, ), and since i j 1, j1, i, . As a
consequence, dg (yi, , z ) = O (i, ), and we can write that
dg (yi, , yk, ) dg (yi, , z ) + dg (z , yk, )
= O (i, ) + O (j, )
= O (i, ) + O (j1, )
= O (i, ) .
Such an equation is is in contradiction with (6.0.8). This proves (6.1.129). An
easy remark is that (6.1.130) is a direct consequence of (6.1.127).
It follows from claim 6.1.7 and (6.1.128) that, for any k {1, . . . , j 1}, and
\By (R () k, ),
any y
k,
k,
2 1
2 1
(n+2)
C0 k, (y)
+ R (y)
(6.1.131)
u (y)
where C0 > 0 is independent of . Similarly, it follows from claim 6.1.7 and
\ p By (R () j, ),
(6.1.130) that, for any y
i,
j,
i=j
2 1
u (y)
( n2 +1)(12)
C0 j,
(n+2)(1)
Rj, (y)
(n+2)
+ R (y)
(6.1.132)
134
CHAPTER 6
2
u (y ) = u (y0 yk )
lim k,
It follows that
lim
u (y )
= |y0 |n2 u (y0 yk ) .
0k, (y )
(6.1.133)
Letting
D0 = 2 sup
sup |z|n2 u (z yk ) ,
(6.1.134)
k=1,...,p zRn
u (y ) u0 (y )
0
i=1 i,
n
2 1
2n
(y ) + j, Rj, (y )
1
D0 .
2
135
Since j 1 k, it follows from the above equation that dg (yk, , yl, ) = O (k, ),
and this is in contradiction with (6.0.8). Therefore,
R (y ) = Rj, (y ) = O (j, ) .
1
dg (yk, , y ) k,
Since
from (6.0.12) that
Hence,
1 n
u (y ) u0 (y ) = o R (y ) 2 .
u (y ) u0 (y )
1 n
n2
= j, 2 R (y )
u (y ) u0 (y )
n 1
R (y ) 2
=o
j,
1 n
n2
j, 2 Rj, (y )
= o (1) .
In particular, (6.1.122) holds with C > 0.
Case 4. We assume that R (y ) 0 as +, that for any k {1, . . . , p},
dg (yk, , y )
+
k,
as +, and that
Rj, (y )
+
j,
as +. We let G be the Greens function of g +h . Thanks to the Greens
representation formula,
2 1
2 1
dvg
G (y , y) u (y)
u0 (y)
u (y ) u0 (y ) =
M
(6.1.135)
+
G (y , y) (h (y) h (y)) u0 (y) dvg .
M
(6.1.136)
2 1
2 1
dvg = o (1) .
G (y , y) u (y)
u0 (y)
M \p
i=1 Byi, ( )
p
i=1 Byi, ( )
2 1
G (y , y) u (y)
dvg + o (1) .
(6.1.137)
136
CHAPTER 6
(p By
( ))
i=1
k,
i,
2 1
G (y , y) u (y)
2 1
1
Bk,
G (y , y) u (y)
dvg
2 1
2
Bk,
G (y , y) u (y)
dvg
(6.1.138)
dvg .
1
+ as +, for any y Byk, (R () k, ), and
Since dg (yk, , y ) k,
for large,
1
dg (y , y) dg (y , yk, ) .
2
Using (6.1.127), we then get that
1
Byk, (R () k, ) pi=1 Byi, ( ) Bk,
(6.1.139)
G (y , y) C2 2n2 dg (yk, , y )
1
for all y Bk,
. It follows that, for large,
2 1
dvg
G (y , y) u (y)
p
Byk, (R()k, )(i=1 Byi, ( ))
2n
2 1
n2
C2 2
dg (yk, , y )
u (y)
dvg
Byk, (R()k, )
2n
C2 2n2 dg (yk, , y )
1
u 22 1 Volg Byk, (R () k, ) 2 .
2 1
D1 0k,
G (y , y) u (y)
dvg
(6.1.140)
(y )
where
D1 = 2n1 C2 2
n1
21
R () 2
(6.1.141)
137
2 1
G (y , y) u (y)
dvg
1 \B
Bk,
yk, (R()k, )
C0
(n+2)
1 \B
Bk,
yk, (R()k, )
G (y , y) R (y)
dvg
n
2n ( 2 +1)(12)
+C0 C2 2n2 dg (yk, , y )
k,
(n+2)(1)
dg (yk, , y)
dvg .
1 \B
Bk,
yk, (R()k, )
2
n+2 ,
1 \B
Bk,
yk, (R()k, )
dg (yk, , y)
dvg
2n1
n(n+2)(1)
(R () k, )
(n + 2) (1 ) n
when is large. Independently, using Holders inequality and (6.1.80), and since
1
, we easily get that
< n+2
(n+2)
G (y , y) R (y)
dvg
1 \B
Bk,
yk, (R()k, )
G (y , y)
n
n1
n1
n
dvg
n1
n(n+2)
p
i=1 Byi, ( )
R (y)
dvg
= o (1) .
It follows that
2 1
1 \B
Bk,
yk, (R()k, )
D2 0k,
G (y , y) u (y)
dvg
(6.1.142)
(y ) + o (1)
where
D2 = C0 C2 2n1
n1
n(n+2)(1)
R ()
.
(n + 2) (1 ) n
2 1
G (y , y) u (y)
dvg (D1 + D2 ) 0k, (y ) + o (1)
1
Bk,
2
, we have that
for large. Independently, for y in Bk,
dg (yk, , y) dg (yk, , y ) dg (y , y)
1
dg (yk, , y ) dg (yk, , y )
2
1
dg (yk, , y ) .
2
(6.1.143)
(6.1.144)
138
CHAPTER 6
2
It follows, from (6.1.131) and (6.1.139), that for any y Bk,
,
2 1
u (y)
C0
( n2 +1)(12)
k,
dg (yk, , y )
2
(n+2)(1)
(n+2)
+ R (y)
2 1
G (y , y) u (y)
dvg
2
Bk,
C0
(n+2)
2
Bk,
G (y , y) R (y)
( n2 +1)(12)
+C0 k,
dg (yk, , y )
2
dvg
(n+2)(1)
2
Bk,
G (y , y) dvg .
(n+2)
G (y , y) R (y)
dvg = o (1) .
2
Bk,
2n
G (y , y) dvg C2
dg (y , y)
dvg
2
Bk,
2
Bk,
C2
By ( 12 dg (yk, ,y ))
dg (y , y)
2n
dvg
2
= O dg (yk, , y ) .
Hence,
2 1
2
Bk,
G (y , y) u (y)
=O
0k,
(y )
dvg
k,
dg (yk, , y )
2(n+2)
+ o (1) .
1
2
Noting that < n+2
, and that dg (yk, , y ) k,
+ as +, it follows
that
2 1
G (y , y) u (y)
dvg = o 0k, (y ) + o (1) .
(6.1.145)
2
Bk,
Coming back to (6.1.138), we obtain with (6.1.144) and (6.1.145) that, for any
k {1, . . . , j 1},
2 1
dvg
G (y , y) u (y)
p
(
(6.1.146)
i=1 Byi, ( ))
k,
(D1 + D2 + o (1)) 0k, (y ) + o (1)
139
1
1
= y k, pi=1 Byi, ( ) s.t. dg (y , y) Rj, (y ) ,
Bk,
2
1
2
= y k, pi=1 Byi, ( ) s.t. dg (y , y) < Rj, (y ) .
Bk,
2
We write that
2 1
(p By
( ))
i=1
j,
i,
G (y , y) u (y)
k=j
dvg
2 1
k, (
p
i=1 Byi, ( )
G (y , y) u (y)
2 1
p
i=j Byi, (R()j, )
G (y , y) u (y)
dvg
(6.1.147)
dvg .
1
Since Rj, (y ) j,
+ as +, we get that
dg (y , y)
1
Rj, (y )
2
for all y pi=j Byi, (R () j, ) when is large. Then, by (6.1.80), we get that
for large
2 1
G (y , y) u (y)
dvg
p
i=j Byi, (R()j, )
2n
C2 2n2 Rj, (y )
2 1
p
i=j Byi, (R()j, )
2n
C2 2n2 Rj, (y )
u (y)
dvg
1
u 22 1 Volg pi=j Byi, (R () j, ) 2
2n
2
(p j + 1) D1 j,
Rj, (y )
2 1
dvg
G (y , y) u (y)
p
j, (i=1 Byi, ( ))
n
2n
2
(p j + 1) D1 j,
Rj, (y )
p
2 1
+
dvg .
G (y , y) u (y)
p
k=j k, (i=1 Byi, ( ))
(6.1.148)
140
CHAPTER 6
k, (p
i=1 Byi, ( ))
2 1
G (y , y) u (y)
2 1
1
Bk,
G (y , y) u (y)
dvg
2 1
2
Bk,
dvg
G (y , y) u (y)
(6.1.149)
dvg .
1
, we have from (6.1.80) that
In Bk,
2n
G (y , y) C2 dg (y , y)
2n
C2 2n2 Rj, (y )
2 1
1
Bk,
G (y , y) u (y)
C0 C2 2
n2
2n
Rj, (y )
+C0
dvg
( n2 +1)(12)
j,
(n+2)(1)
1
Bk,
(n+2)
1
Bk,
G (y , y) R (y)
dg (yk, , y)
dvg
dvg .
(n+2)
G (y , y) R (y)
dvg = o (1) .
1
Bk,
Since <
2
n+2 ,
(n+2)(1)
dg (yk, , y)
dvg
1
Bk,
(n+2)(1)
M \Byk, (R()j, )
dg (yk, , y)
dvg
2n1
n(n+2)(1)
(R () j, )
.
(n + 2) (1 ) n
n
2 1
2n
2 1
G (y , y) u (y)
dvg D2 j,
Rj, (y )
+ o (1)
1
Bk,
2
, then
where D2 is as in (6.1.143). Independently, if y is in Bk,
1
Rj, (y ) .
2
(6.1.150)
141
2
,
From (6.1.132), it follows that, for y Bk,
(n+2)(1)
n
( 2 +1)(12) Rj, (y )
2 1
(n+2)
C0 j,
+ R (y)
.
u (y)
2
Then,
2 1
2
Bk,
G (y , y) u (y)
dvg
n
( 2 +1)(12)
(n+2)(1)
= O j,
Rj, (y )
2
Bk,
+O
(n+2)
2
Bk,
G (y , y) R (y)
( n2 +1)(12)
= O j,
so that
Rj, (y )
2
Bk,
G (y , y) dvg + o(1) .
G (y , y) dvg C2
By
2 1
2
Bk,
G (y , y) u (y)
n
2 1
2n
= O j, Rj, (y )
Since <
dvg
(n+2)(1)
From (6.1.80),
2
Bk,
G (y , y) dvg
j, (y )
2
dg (y , y)
2n
dvg
dvg
Rj, (y )
j,
(n+2)2
+ o (1) .
1
and Rj, (y ) j,
+ as +, we get that
n
2 1
2n
2 1
+ o (1) .
G (y , y) u (y)
dvg = o j,
Rj, (y )
2
n+2
2
Bk,
2 1
dvg
G (y , y) u (y)
k, (p
i=1 Byi, ( ))
n
2n
2
(D2 + o (1)) j,
Rj, (y )
+ o (1) .
Coming back to (6.1.148) we then get that
2 1
dvg
G (y , y) u (y)
(p By
( ))
i=1
j,
i,
(6.1.151)
n
2n
2 1
(p j + 1) (D1 + D2 ) + o (1) j, Rj, (y )
+ o (1) .
j1
0k, (y )
k=1
n
2n
2
+ (p j + 1) (D1 + D2 + o (1)) j,
Rj, (y )
+ o (1) .
142
CHAPTER 6
i=1
n
Cdg (yk, , y ) 2
+
Rj, (y )
j,
Rj, (y )
j,
(n2)
R ()
j1
i, (y )
i=1
(n2)
( n2 1)(12)
j,
(2n)(1)
Rj, (y )
+ o (1)
Cdg (yk, , y )
n
2 1
k,
(n2)
(n2)
(y ) (j 1) R ()
+ o (1) .
Therefore,
1
1 n (12)
dg (yk, , y ) ( 2 )
D () 2 2
C
100
k,
(n2)
Rj, (y )
(n2)
(j 1) R ()
+
+ o (1) .
j,
(6.1.155)
143
144
CHAPTER 6
i, (x) =
i,
2i, +
dg (xi, ,x)2
n(n2)
n2
2
It follows from section 6.1 [we refer to (6.1.1) and (6.1.67)] that (6.2.1) holds
for all x M \ pj=1 Byj, (R0 j, ). It remains to prove that (6.2.1) holds also in
Byj, (R0 j, ) for all j {1, . . . , p}. We fix j0 {1, . . . , p}, and let
q, = xNj0 +1 1, ,
x
1, = xNj0 +1, , . . . , x
q, = Nj0 +1 1, .
1, = Nj0 +1, , . . . ,
It easily follows from point (P1) of Theorem 4.1 that (6.2.1) holds in the ball
Byj0 , (R0 j0 , ) if there exists C > 0 such that
q
n 1
1 n
2n
2
2
i, dg (
xi, , x)
+ j0 ,
u (x) C
(6.2.2)
i=1
x1, , . . . , x
q, } and all . Without loss of generalfor all x Byj0 , (R0 j0 , ) \ {
ity, we assume in what follows that the set {
xi, }i=1,...,q is not empty. Otherwise,
(6.2.2) is a direct consequence of Theorem 4.1.
We rearrange the x
i, s by a process analogous to that used in the introduction
of this chapter. We let x
1, be such that
1,
i, for all i {1, . . . , q}. We
q2 1 the x
i, s, i = 2, . . . , q, which are such that, up to
denote then by x
2, , . . . , x
a subsequence,
lim
dg (
x1, , x
i, )
< + .
1,
q2 , be such that
q2 ,
i, for all i {q2 , . . . , q}. We
We let q1 = 1, and let x
denote then by x
q2 +1, , . . . , x
q3 1, the x
i, s, q2 + 1 i q, which are such
that, up to a subsequence,
lim
dg (
xq2 , , x
i, )
< + .
q2 ,
and
i, =
qi , .
(6.2.3)
(6.2.4)
145
j,
4
lim sup
+
We have that
SNj0 =
lim
j0 ,
exp1
yj0 ,
(6.2.5)
(
yj, ) j = 1, . . . , r
where SNj0 is as in (4.1.2). We claim now that there exists C > 0 such that
r
n
n
1
1
2n
2
j,
dg (
yj, , y)
+ j0 ,2
u (y) C
(6.2.6)
j=1
where
(y) = min dg (
yj, , y) ,
R
j=1,...,r
where lim
R+ +
Proof of claim 6.2.1. We proceed by contradiction and assume that claim 6.2.1
does not hold. Then there exists a sequence (xN +1, ) in Byj0 , (R0 j0 , ) such
that, for any j {1, . . . , r},
yj, , xN +1, )
dg (
+
j,
(6.2.7)
(6.2.8)
2
u (xN +1, ) = N +1,
(6.2.9)
and prove that Theorem 4.1 continues to hold if we add xN +1, and N +1, to the
xi, s and i, s, i = 1, . . . N . Noting that this contradicts the maximality of N , as
146
CHAPTER 6
assumed in the introduction of this chapter, claim 6.2.1 will be proved. We follow
the lines of step 3 of the proof of Theorem 4.1. We claim that
lim 1
+ j0 ,
exp1
yj0 , (xN +1, ) SNj0
(6.2.10)
where SNj0 is defined by (4.1.2). If (6.2.10) is false, point (P1) of Theorem 4.1
gives that
1 n
|u (xN +1, ) vNj0 , (xN +1, ) | = o j0 ,2 .
(xN +1, ) = O (j , ), we get that
Noting that this contradicts (6.2.8) since R
0
(6.2.10) holds. From (6.2.10),
(xN +1, ) = o (j , ) .
R
(6.2.11)
0
1 n2
Since vNj0 , (xN +1, ) = O j0 , , we get by combining (6.2.8), (6.2.9) and
(6.2.11) that
(xN +1, )
R
20
N +1,
(6.2.12)
N +1, = o (j0 , ) .
(6.2.13)
2
(Rn \SN +1 ), and strongly in C 2 (B0 (0 )),
weakly in D12 (Rn ), strongly in Cloc
where N +1, is as in Chapter 4, SN +1 is as in (4.1.2), uN +1, is as in (4.1.3),
and u is as in (4.1.5).
i=1
2
lim E (u )
+
E(u0 )2 (N + 1) 2min
147
,x
k,
as +. Summarizing, from the above three cases, (D1) is proved. Now we
prove (D2). Given x B0 (1
N +1, ), the Euclidean ball of center 0 and radius
,
we
let
1
N +1,
n
1
uN +1, (x) = N2 +1, u expxN +1, (N +1, x) ,
gN +1, (x) = expxN +1, g (N +1, x) ,
hN +1, (x) = h expxN +1, (N +1, x) .
(6.2.14)
148
CHAPTER 6
Since N +1, 0 as +,
2
lim gN +1, = in Cloc
(Rn ) .
(6.2.15)
We also have that gN +1, is controlled on both sides by in the sense of bilinear
forms. It is easily checked that
uN +1, (0) = 1
(6.2.16)
and that
2 1
gN +1, uN +1, + 2N +1, hN +1, uN +1, = uN
+1,
(6.2.17)
in B0 (1
N +1, ). We let
exp1
xN +1, (xi, )
SN +1 =
lim
, xi, BxN +1, () , 1 i N
+
N +1,
where, up to a subsequence, the limits are assumed to exist. By (6.2.12) [see also
the proof of (D1)]
3
SN +1 B0
0 = .
(6.2.18)
2
Let R > 0 and let (x ) be a sequence of points in B0 (R) such that
1
R
where d is the Euclidean distance. We can write with (6.2.15) that
N +1,
min dg xi, , expxN +1, (N +1, x )
i=1,...,N
2R
d (x , SN +1 )
(6.2.19)
for large. Let y = expxN +1, (N +1, x ). It follows from point (P2) of Theorem 4.1 that
uN +1, (x ) (2R) 2
(2R) 2
min dg (xi, , y ) 2
i=1,...,N
u (y )
for some C > 0 independent of . Thus (uN +1, ) is locally uniformly bounded
in Rn \SN +1 . By standard elliptic theory, and (6.0.1), (6.2.14), (6.2.15), and
(6.2.17), we then get that, up to a subsequence,
lim N +1, uN +1, = uN +1
(6.2.20)
2
(Rn \SN +1 ). We also have that N +1, uN +1, uN +1 weakly in the
in Cloc
space D12 (Rn ) as +. Moreover, uN +1 verifies that
2 1
n
uN +1 = uN
+1 in R
149
1
2 |xN +1 |2 .
n(n 2) N +1
N
+1
vi, dvg
u u0
M \BxN +1, (RN +1, )
i=1
2
N
(1 + )
vi, dvg
u u0
M \BxN +1, (RN +1, )
i=1
2
+C
vN
+1, dvg .
M \BxN +1, (RN +1, )
Noting that
lim
lim
R+ +
2
vN
+1, dvg = 0 ,
we get that
2
N
+1
0
vi, dvg
u u
M \BxN +1, (RN +1, )
i=1
2
N
vi, dvg + R,
u u0
M \Bx
(RN +1, )
i=1
N +1,
where lim
(6.2.21)
lim R, = 0. Hence,
R+ +
2
N
+1
0
vi, dvg
u u
M
i=1
2
N
vi, dvg + IR () + R,
u u0
M
(6.2.22)
i=1
where
2
N
+1
vi, dvg
IR () =
u u0
BxN +1, (RN +1, )
i=1
2
N
vi, dvg .
u u0
Bx
(RN +1, )
N +1,
i=1
(6.2.23)
150
CHAPTER 6
We have that
2
N
+1
vi, dvgN +1,
IR () =
uN +1, u0
B0 (R)
i=1
2
N
i=1
where
n
1
vi, (x) = N2 +1, vi, expxN +1, (N +1, x)
and
n
1
u0 (x) = N2 +1, u0 expxN +1, (N +1, x) .
We let VR be given by
VR =
x Rn s.t. d (x, SN +1 )
Noting that
lim
lim
R+ +
VR
1
R
.
2
vN
+1, dvgN +1, = 0 ,
N
+1
0
vi, dvgN +1,
IR ()
uN +1, u
B0 (R)\VR
i=1
(6.2.24)
2
N
2n
2
vi,
dvg = O ni, dg (xi, , xN +1, )
BxN +1, (RN +1, )
Volg BxN +1, (RN +1, )
2n
= O ni, nN +1, dg (xi, , xN +1, )
n
i,
=o
.
dg (xi, , xN +1, )
151
yj, ),
From (6.2.7) and the construction of (yj, ) and (
i, = O (dg (xi, , xN +1, ))
for all i = 1, . . . , N . Thus,
2
vi,
dvgN +1, 0
B0 (R)\VR
2
vi,
dvgN +1, 0
B0 (R)\VR
as +. Summarizing,
(6.2.27)
as +. Since point (P1) of Theorem 4.1 holds for xN +1, and N +1, ,
|uN +1, vN +1, |2 dvgN +1, 0
(6.2.28)
as +, and
lim
B0 (R)\VR
lim
R+ +
B0 (R)\VR
N
0
vi, dvgN +1, 0
uN +1, vN +1, u
B0 (R)\VR
i=1
N
vi, dvgN +1,
uN +1, u0
B0 (R)\VR
i=1
2 1
0
2
uN
+1, u dvgN +1,
B0 (R)\VR
2
i=1
2min
B0 (R)\VR
+ R,
2 1
uN
i, dvgN +1,
+1, v
(6.2.29)
152
CHAPTER 6
where lim
R+ +
IR () 2min + R,
and using (6.2.22), since R is arbitrary, we get that (D3) is proved. Therefore,
(D1)(D3) hold. As already mentioned, this is in contradiction with the maximality
of N in Theorem 4.1. In particular, claim 6.2.1 is proved.
2
Going on with the proof of (6.2.6), mimicking what was done in section 6.1, we
now prove the following:
0 and C () > 0 such
C LAIM 6.2.2. For any 0 < < 12 , there exist R () R
that
n
( 2 1)(12)
(1 n )(12)
(2n)(1)
(2n)
u (y) C () 1,
+ j0 , 2
R (y)
R (y)
is as in
for all y Byj0 , (R0 j0 , ) \ rj=1 Byj, (R () j, ) and all , where R
claim 6.2.1.
Proof of claim 6.2.2. Let 0 < < 12 . Also let x0 SNj0 , where SNj0 is as in
Theorem 4.1, and x = expyj0 , (j0 , x0 ). We set
D () =
(1 ) C12
4r
C22
(6.2.30)
min {D () ; D (1 )}
(6.2.31)
for all y Bx ( () j0 , ) \
Byj, (R () j, ) and large. In order to
prove (6.2.31), we let (y ) be a sequence of points such that
rj=1
j, )
y Bx (j0 , ) \ rj=1 Byj, (R
for all . Using claim 6.2.1, we can write that
n
n
(y ) 2 1 u (y ) R, + R
(y ) 2 1 vN
R
j0 ,
where lim
(y )
lim R, = 0. Thus,
R+ +
(y ) 2 1 u (y ) R, + (2j , ) 2 1 vN , (y )
R
0
j0
n
.
R, + (2) 2
0 sufficiently large, this proves
Choosing > 0 sufficiently small and R R
(6.2.31). Now we consider the linear operator L given by
2 2
L (u) = g u + h u u
u.
Gj, (y)
and H2 (y) =
Gj, (y)
H1 (y) =
j=1
j=1
153
+ (1 )
|Gj, (y)|2
2
Gj, (y)
and to
L H2 (y) =
r
2 2
Gj, (y)
j=1
+ (1 )
|Gj, (y)|2
Gj, (y)
in M \ {
y1, , . . . , yr, }. Let
y Bx ( () j0 , ) \ rj=1 Byj, (R () j, ) .
By (6.1.2),
|Gj, (y )|2
2
Gj, (y )
C12
dg (
yj, , y )
u (y )
(y )2
min {D () ; D (1 )} R
j=1
+ (1 )
C12
2
dg (
yj, , y )
C12
(2n)(1)2
dg (
yl, , y )
C21
r
r
1
1
(y )2
D (1 ) R
Gj, (y )
O
Gj, (y )
(1 )
j=1
j=1
C12
(2n)(1)2
(1 ) 1
R (y )
C2
(y )(2n)(1)2
rD (1 ) C21 R
(y )(2n)(1)
O R
(y )
2rD (1 ) C21 R
(2n)(1)2
154
CHAPTER 6
(2n)2
rj=1
(6.2.32)
()
1
d x0 , SNj0 \ {x0 } ,
2
u 2j0 ,2 on Bx ( () j0 , )
0 , we also have that
for large. Since R () R
1 n
u 2
j, 2 on Byj, (R () j, )
for all j {1, . . . , r} and large. By (6.1.3) and (6.2.3) we then get the existence
of C () > 0 such that
n
( 2 1)(12) 1
(1 n2 )(12) 2
u C () 1,
H + j0 ,
H
on Bx ( () j0 , ) \ rj=1 Byj, (R () j, ) . From (6.2.32), the above equation, and the fact that L u = 0 in M , we can apply the maximum principle. It
follows that
n
( 2 1)(12) 1
(1 n )(12) 2
H + j0 , 2
H
u C () 1,
in Bx ( () j0 , ) \ rj=1 Byj, (R () j, ). In particular, using (6.1.3), this
proves that the estimate of claim 6.2.2 holds in
Bx ( () j0 , ) \ rj=1 Byj, (R () j, ) .
x0 SNj
Noting that, with a possibly larger C (), point (P1) of Theorem 4.1 gives that this
estimate also holds in
Byj0 , (R0 j0 , ) \ x0 SNj Bx ( () j0 , ) ,
0
C LAIM 6.2.3. There exist C > 0 and a sequence ( ) of positive real numbers
converging to 0 as + such that
n
1 n
2
(y)
+ j0 ,2
u (y) vNj0 , (y) C 1,
R
0 j, and all , where vN , is as
for all y Byj0 , (R0 j0 , ) \ rj=1 Byj, R
j0
2n
155
(y )
R
j0 ,
(y ) = u (y0 yk )
where, up to a subsequence,
1
exp1
yk, (y )
k,
is given by Theorem 4.1. Noting that
dg (
yk, , y )
lim
= |y0 | ,
+
k,
y0 = lim
we obtain that
lim
u (y )
n
2 1
2n
k, dg (
yk, , y )
u (y )
D0
(y )2n
R
n
2 1
1,
where
D0 =
sup
sup |z|n2 u (z yj )
j=1,...,r zRn
(6.2.34)
and yj = xNj0 +qj is as in Theorem 4.1. This gives that (6.2.33) holds with C >
D0 .
d (
y
,y )
1
n2
1, 2 R
|u (y ) vNj0 , (y ) |
(y )
n2 1
(y )
n
R
(y ) 2 1 |u (y ) vN , (y ) |
R
=
j0
1,
n
(y ) 2 1 |u (y ) vN , (y ) |
=O R
j0
n
= o (1) .
156
CHAPTER 6
=
G (y , y) g u vNj0 , (y) dvg
Byj , (2R0 j0 , )
0
+
G (y , y) h (y) (y) u (y) vNj0 , (y) dvg .
Byj , (2R0 j0 , )
0
(6.2.36)
From (6.0.9), (6.2.35), and point (P1) of Theorem 4.1,
1 n
g u vNj0 , (y) = o j0 , 2
in Byj0 , (2R0 j0 , ) \Byj0 , (R0 j0 , ). Thus we can write that
G (y , y) g u vNj0 , (y) dvg
Byj , (2R0 j0 , )\Byj , (R0 j0 , )
0
0
n
1 2
= o j0 ,
G (y , y) dvg .
Byj , (2R0 j0 , )
0
By (6.1.80), we have that
2n
dvg
G (y , y) dvg = O
dg (y , y)
Byj , (2R0 j0 , )
Byj , (2R0 j0 , )
0
0
=O
By (3R0 j0 , )
dg (y , y)
2n
dvg
= O j20 ,
so that
G (y , y) g u vNj0 , (y) dvg
1 n
= o j0 ,2
(6.2.37)
where
= Byj0 , (2R0 j0 , ) \Byj0 , (R0 j0 , ) .
Similarly, it is easily checked that
G (y , y) h (y) (y) u (y) vNj0 , (y) dvg
1 n
= o j0 ,2 .
(6.2.38)
157
1 n
G (y , y) h (y) vNj0 , (y) dvg = o j0 ,2 .
Byj , (R0 j0 , )
0
(6.2.39)
=
G (y , y) (g u (y) + h (y) u (y)) dvg
Byj , (R0 j0 , )
0
1 n
2 1
=
dvg
G (y , y) u (y)
Byj , (R0 j0 , )
0
1 n
2 1
=
dvg
G (y , y) vNj0 , (y)
Byj , (R0 j0 , )
0
1 n
2
+O j0 ,
G (y , y) dvg
Byj , (R0 j0 , )
0
1 n
2 1
=
dvg + o j0 ,2
G (y , y) vNj0 , (y)
Byj , (R0 j0 , )
0
and it follows that
u (y ) vNj0 , (y )
2 1
2 1
=
dvg
vNj0 , (y)
G (y , y) u (y)
Byj , (R0 j0 , )
0
1 n
+ o j0 ,2 .
(6.2.40)
sup
( j0 , )
j20 , |u vNj0 , | 0
(6.2.41)
158
CHAPTER 6
2 1
2 1
dvg
vNj0 , (y)
G (y) u (y)
Byj , (R0 j0 , )\rj=1 Byj, ( j0 , )
0
1 n
2
2n
= o j0 ,
1 n
= o j0 ,2 ,
Byj
(R0 j0 , )
0 ,
dg (y , y)
dvg
1 n
2 1
dvg + o j0 ,2 .
G (y , y) u (y)
rj=1 Byj, ( j0 , )
(6.2.42)
Given j = 1, . . . , r, we define
(y) = dg (
yj, , y) ,
j, = y M s.t. R
1
1
= y j, Byj, ( j0 , ) s.t. dg (y , y) R
(y
)
,
Bj,
2
1
2
= y j, Byj, ( j0 , ) s.t. dg (y , y) < R
(y
)
.
Bj,
2
1
n+2 .
j, Byj, ( j0 , )
G (y , y) u (y)
dvg
2 1
j, Byj, (R()
1, )
G (y , y) u (y)
2 1
1 \B
Bj,
1, )
y
j, (R()
G (y , y) u (y)
dvg
2 1
2 \B
Bj,
1, )
y
j, (R()
G (y , y) u (y)
(6.2.43)
dvg
dvg
(y )
R
1,
+ as +,
dg (y , y) dg (y , yj, ) dg (
yj, , y)
R (y ) R () 1,
1
R
(y )
2
for large and all y j, Byj, (R () 1, ). Using (6.1.80) we then get that
159
for large
2 1
G (y , y) u (y)
j, Byj, (R()
1, )
(y )
C2 2n2 R
2n
dvg
2 1
u
dvg
j, Byj, (R()
1, )
1
u 22 1 Volg Byj, R () 1, 2
21
n
(y )2n 2 1 2 n1 R ()n 1,
C2 2n2 R
.
n
Hence, for large,
2 1
G (y , y) u (y)
dvg
C2 2
n2
(y )
R
2n
j, Byj, (R()
1, )
n
2 1
(6.2.44)
(y )
D1 1, R
2n
where
D1 = 2n1 C2 2
n1
21
n
1
, we can write using (6.1.80) that
Given y Bj,
R () 2
(y )
G (y , y) C2 2n2 R
2n
(6.2.45)
2 1
G (y , y) u (y)
dvg
1 \B
Bj,
1, )
y
j, (R()
22
C ()
+22
1
n+2 ,
( n +1)(12)
j0 ,2
(n+2)
1 \B
Bj,
1, )
y
j, (R()
Since <
2 1
+n4
2 1
C ()
G (y , y) dg (
yj, , y)
n
(y )2n ( 2 +1)(12)
C2 R
1,
(n+2)(1)
1 \B
Bj,
1, )
y
j, (R()
dg (
yj, , y)
(n+2)(1)
1 \B
Bj,
1, )
y
j, (R()
dg (
yj, , y)
dvg .
dvg
2n1
n(n+2)(1)
(R () 1, )
(n + 2) (1 ) n
dvg
160
CHAPTER 6
(n+2)
G (y , y) dg (
yj, , y)
dvg
1 \B
Bj,
1, )
y
j, (R()
= O
n(n2)
n1
By (2R0 j0 , )
dg (y , y)
O
Byj, ( j0 , )
since <
1
n+2 .
n1
n(n+2)
yj, , y)
dvg
dg (
1(n+2)
= O j0 , ( j0 , )
Hence,
1 \B
Bj,
1, )
y
j, (R()
n1
n
dvg
G (y , y) dg (
yj, , y)
(n+2)
2(n+2)
dvg = o j0 ,
since 0 as +. Summarizing,
2 1
G (y , y) u (y)
dvg
1 \B
Bj,
1, )
y
j, (R()
n
2 1
D2 1,
(6.2.46)
n
(y )2n + o 1 2
R
j0 ,
where
D2 = 22 +n3
n1
2 1
n(n+2)(1)
C2 C ()
R ()
. (6.2.47)
(n + 2) (1 ) n
2
In Bj,
, we have that
dg (
yj, , y) dg (
yj, , y ) dg (y , y)
(y )
(y ) 1 R
R
2
1
R
(y ) .
2
Then, applying claim 6.2.2, we obtain as above, using (6.2.3), that
(n+2)(1)
! n +1 (12) 1
(2 )
2 1
2 1
(y )
u (y)
R
1,
22 2 C ()
2
"
n
( +1)(12)
(n+2)
+j0 ,2
dg (
yj, , y)
2
for all y Bj,
\Byj, (R () 1, ). Hence,
2 1
G (y , y) u (y)
dvg
2 \B
Bj,
1, )
y
j, (R()
( n2 +1)(12)
(n+2)(1)
= O 1,
R (y )
+O
( n +1)(12)
j0 ,2
2
Bj,
2
Bj,
G (y , y) dvg
G (y , y) dg (
yj, , y)
(n+2)
dvg
161
2(n+2)
(n+2)
.
G (y , y) dg (
yj, , y)
dvg = o j0 ,
2
Bj,
G (y , y) dvg = O
2
Bj,
By
(y )
R
2
2 1
2 \B
Bj,
1, )
y
j, (R()
G (y , y) u (y)
n
2 1
(y )
= O 1, R
(y )
R
1,
dg (y , y)
dvg
(y )2 .
=O R
Therefore,
Since
2n
2n
1,
(y )
R
dvg
2(n+2)
1 n
+ o j0 ,2 .
1
+ as +, and since < n+2
, we obtain that
2 1
G (y , y) u (y)
dvg
2 \B
Bj,
1, )
y
j, (R()
n
2 1
= o 1,
n
(y )2n + o 1 2 .
R
j0 ,
2 1
dvg
G (y , y) u (y)
j, Byj, ( j0 , )
n
1 n
2n
2 1
(D1 + D2 + o (1)) 1,
+ o j0 ,2
R (y )
(6.2.48)
(6.2.49)
rj=1 Byj, ( j0 , ) = rj=1 j, Byj, ( j0 , ) ,
n
1 n
2n
2 1
+ o j0 ,2 .
R (y )
u (y ) vNj0 , (y ) r (D1 + D2 + o (1)) 1,
This proves that (6.2.33) holds with C > r (D1 + D2 ), where D1 is as in (6.2.45)
and D2 is as in (6.2.47).
Thanks to cases 14, (6.2.33) holds for any sequence (y ) of points in
0 j,
Byj0 , (R0 j0 , ) \ rj=1 Byj, R
if we choose C such that C > D0 + r (D1 + D2 ), where D0 is as in (6.2.34), D1
is as in (6.2.45), and D2 is as in (6.2.47). This ends the proof of claim 6.2.3. 2
From now on, we let j {1, . . . , r}, and let
j, (y) = min dg (
yi, , y) .
R
i=j,...,r
(6.2.50)
162
CHAPTER 6
We say that (6.2.51)j holds if there exist C > 0 and a sequence ( ) of positive
real numbers converging to 0 as + such that
u (y) vNj0 , (y)
j1
n 1
(6.2.51)j
n
n
1
1
2n
2n
2
2
j, (y)
C
i,
dg (
yi, , y)
+ j,
+ j0 ,2
R
i=1
0 i, and all . The objective of
for all y Byj0 , (R0 j0 , ) \ ri=1 Byi, R
claims 6.2.4 and 6.2.5 is to prove that, for any j {2, . . . , r},
(6.2.51)j1 (6.2.51)j .
(6.2.52)
We let j {2, . . . , r} and we assume that (6.2.51)j1 holds. Before stating claim
6.2.4, we prove some preliminary results and set up some notation. We fix such
1
. Given k = 1, . . . , j 1, we let
that 0 < < n+2
k, = y Byj0 , (R0 j0 , ) s.t. k, (y) i, (y) , i = 1, . . . , j 1
(6.2.53)
where
( n 1)(12)
(2n)(1)
i, (y) = i,2
dg (
yi, , y)
.
(6.2.54)
0 such that,
We claim now that there exists R () R
n2 1
j, (y) , dg (
yk, , y)
u (y) vNj0 , (y)
min R
1
min {D () , D (1 )} 2 2
<
100
(6.2.55)
Since
j, (y ) , dg (y , yj1, ) ,
j1, (y ) = min R
R
163
we have that
j, (y )2n + dg (y , yj1, )2n .
j1, (y )2n R
R
Hence,
u (y ) vNj0 , (y )
j1
n 1
n
1 n
2n
2n
2
2 1
2
C
.
i, dg (
yi, , y )
+ j1, Rj, (y )
+ o j0 ,
i=1
2n
2
i,
dg (
yi, , y )
R(2n) i, (y )
for all i {1, . . . , j 1}. Since we have that y k, , we also have that
i, (y ) k, (y ) for all i {1, . . . , j 1}. Thus,
u (y ) vNj0 , (y )
"
!
n
1 n
2n
2 1
,
+ o j0 ,2
C (j 1) R(2n) k, (y ) + j1,
Rj, (y )
yk, , y ) = O (j0 , ), we have that
and since dg (
n2 1
j, (y ) , dg (
yk, , y )
u (y ) vNj0 , (y )
min R
n2
2
!
"
n2
j1,
(2n)
2
C (j 1) R
dg (
yk, , y )
k, (y ) +
+ o (1) .
j, (y )
R
yk, , y ) R
k, , we get that
As a consequence, noting that dg (
n2 1
j, (y ) , dg (
yk, , y )
u (y ) vNj0 , (y )
min R
n2 1
n
j1,
+ o (1) .
C (j 1) R1 2 +
j, (y )
R
1
j, (y ) + as +, it is clear that (6.2.56) holds up to
If j1,
R
j, (y ) = O (
choosing R sufficiently large. Let us assume now that R
j1, ).
We claim that this implies that Rj, (y ) = R (y ) for large. In order to prove
j, (y ) = dg (y , yl, ). For any
this claim, we let l {j, . . . , r} be such that R
i {1, . . . , j 1}, from (6.2.3),
yl, , yi, ) dg (y , yl, )
dg (y , yi, ) dg (
j1,
j1,
j1,
j, (y )
yl, , yi, ) i,
dg (
R
i,
j1,
j1,
yl, , yi, )
dg (
O (1) .
i,
164
CHAPTER 6
as +, and it follows that for any i with i {1, . . . , j 1}, we have that
j, (y ) when is large. In particular, if R
j, (y ) = O (
yi, , y ) > R
j1, ),
dg (
then we have Rj, (y ) = R (y ) for large, and this proves the above claim.
Thanks to claim 6.2.1 we then get that (6.2.56) holds up to choosing R sufficiently
large. In particular, (6.2.55) is proved.
We let now Aj, be defined by
( n2 1)(12)
Aj,
=
where
sup
k, (y)
sup
k=1,...,j1 yk,
( n 1)(12)
(2n)(1)
j,2
Rj, (y)
(6.2.57)
k,
yk, , y) 2
dg (
u (y) vNj0 , (y)
D (1 )
100
212
.
(6.2.58)
k, s
Aj, j,
1.
j1,
(6.2.59)
1)(12)
j, (y )(2n)(1) (1 + o (1)) .
R
(6.2.60)
k, ,
Since y
1
n
D (1 ) 2 2
1
dg (
yk, , y ) 2
u (y ) vNj0 , (y ) .
100
We then get, thanks to (6.2.51)j1 , that
j1
1
n 1
n
D (1 ) 2 2
1
2n
2
Cdg (
yk, , y ) 2
i,
dg (
yi, , y )
100
i=1
n
2n
2 1
+
+ o (1) .
Rj, (y )
j1,
2n
2
i,
dg (
yi, , y )
(2n)
(j 1) R ()
k, (y ) ,
165
Rj, (y )
(n2)
j1,
k, (y ) .
j, (y )
R
Since dg (
yk, , y ) R () k, , we also have that
n
yk, , y ) 2
dg (
k, (y ) R ()(
1 n
2 )(12)
(n2)
1
n
D (1 ) 2 2
j1,
1
CR () 2 (j 1) + R ()
o (1) .
j, (y )
100
R
Taking R () sufficiently large, this gives that
j, (y ) = O (
R
j1, ) .
(6.2.61)
R
j1,
n
( 2 1)(12) n2 1
= O k,
j1, .
Since k j 1, we have by (6.2.3) that j1, k, . Thus,
dg (
yk, , y ) = O (
k, ) .
(6.2.62)
j, (y ) = dg (
yl, , y ). Thanks to
Now we let l {j, . . . , r} be such that R
(6.2.61) and (6.2.62) we can write that
dg (
yk, , yl, ) dg (
yk, , y ) + dg (
yl, , y )
j1, )
= O (
k, ) + O (
= O (
k, ) .
Noting that such an equation is in contradiction with (6.2.4), we get that (6.2.59)
is proved.
We let
j, = j, max {Aj, ; 1}
(6.2.63)
166
CHAPTER 6
i=1
(1 n )(12)
(2n)
+j0 , 2
R (y)
(y) is
for all y Byj0 , (R0 j0 , ) \ ri=1 Byi, (R () i, ) and all , where R
D ()
(6.2.64)
We let
H (y) =
j1
( n 1)(12)
1
i,2
Gi, (y)
i=1
( n 1)(12)
1
+j,2
Gi, (y)
r
(1 n )(12)
+j0 , 2
i=j
r
Gi, (y)
i=1
where
yi, , y) ,
Gi, (y) = G (
and G is the Greens function of g +h0 , where h0 is as in the beginning of section
6.1. We have already seen when proving claim 6.2.2 that
r
(y)(2n)2
L
Gi, (y) 2rD () C2 R
(6.2.65)
i=1
167
j1
( n 1)(12)
1
i,2
Gi, (y)
i=1
(1 )
|Gi, (y) |2
2
Gi, (y)
2 2
u (y)
+ h (y) (1 ) h0 (y)
( n 1)(12)
1
+ j,2
Gi, (y)
r
i=j
(1 )
|Gi, (y) |2
2
Gi, (y)
2 2
u (y)
+ h (y) (1 ) h0 (y)
(1 n )(12)
(2n)2
+ 2rD () C2 j0 , 2
R (y)
(6.2.66)
ri=1
Gi, (y )
C12
dg (
yi, , y )
for all i {1, . . . , r}. Thus, thanks to (6.2.66), we can write that, for large,
L H (y )
j1
( n 1)(12)
1
i,2
Gi, (y )
i=1
C12
(1 )
2
dg (
yi, , y )
2 2
u (y )
+ O (1)
( n 1)(12)
1
+ j,2
Gi, (y )
r
i=j
(6.2.67)
C12
(1 )
2
dg (
yi, , y )
2 2
u (y )
+ O (1)
(1 n )(12)
(2n)2
+ 2rD () C2 j0 , 2
.
R (y )
After passing to a subsequence, we assume that y k, for some 1 k j 1.
We distinguish four cases.
Case 1. We assume that
j, (y ) , dg (
yk, , y )
min R
j0 ,
168
CHAPTER 6
2 2
u (y )
+ O (1)
j2
0 ,
j1
i,
i=1
2
O j0 , k,
( n 1)(12)
1
i,2
Gi, (y )
i=1
=O
j1
(y )
(y )
2 2
u (y )
=o
+ O (1)
1 n
j0 , 2
j1
( n 1)(12)
1
i,2
Gi, (y )
i=1
(6.2.68)
2 2
u (y )
=o
r
n 1 (12)
)
(
1
+ O (1) j,2
Gi, (y )
1 n
j0 , 2
(6.2.69)
i=j
.
Combining (6.2.67) with (6.2.68) and (6.2.69), we then get that, for large,
(1 n )(12)
1 n
(2n)2
L H (y ) 2rD () C2 j0 , 2
+ o j0 , 2 .
R (y )
(y ) 2R0 j , , it follows from this equation that L H (y ) 0
Since R
0
when is large.
j, (y ), and that
yk, , y ) R
Case 2. We assume that y k, , that dg (
j, (y ) = o (j , ). By (6.2.55), we then get that
R
0
j, (y )2 u (y )2
R
D (1 )
for large, and it follows [see, for instance, the proof of (6.2.32)] that
( n 1)(12)
1
Gi, (y )
j,2
r
i=j
C12
(1 )
2
dg (
yi, , y )
2 2
u (y )
+ O (1)
( n 1)(12)
(2n)(1)2
2rD (1 ) C21 j,2
Rj, (y )
(6.2.70)
169
yk, , y ) 2
dg (
u (y ) vNj0 , (y )
D (1 )
100
212
so that
2
2 2
dg (
yk, , y ) u (y )
max 1, 22 3
D (1 )
2
2 2
+ dg (
.
yk, , y ) vNj0 , (y )
100
(6.2.71)
1 n2
d (
y
,y )
If g k,
as
+
for
some
>
0,
then
u
(y
)
=
O
j0 , . It
j0 ,
follows, as in case 1, that
j1
( n 1)(12)
2 2
1
+ O (1)
i,2
Gi, (y )
u (y )
(6.2.72)
i=1
1 n
2
= o j0 ,
.
yk, , y ) = o (j0 , ), we get with (6.2.71) that
If dg (
2
2 2
dg (
yk, , y ) u (y )
D (1 )
j1
( n2 1)(12)
1
Gi, (y )
i,
C12
i=1
(1 )
2 2
u (y )
dg (
yi, , y )
+ O (1)
(1 )
( n2 1)(12) C12
(2n)(1)
k,
dg (
yk, , y )
2
C21
dg (
yk, , y )
2
C21 D (1 ) dg (
yk, , y ) + O (1)
j1
( n 1)(12)
(2n)(1)
dg (
yi, , y )
i,2
i=1
(1 )
C12
2
dg (
yk, , y ) k, (y )
C21
j1
2
yk, , y ) + O (1)
i, (y ) .
C21 D (1 ) dg (
i=1
C12
(1 )
2
dg (
yi, , y )
2 2
u (y )
(6.2.73)
+ O (1)
2
yk, , y )
(4r (j 1) + o (1)) C21 D (1 ) dg (
k, (y ) .
170
CHAPTER 6
Combining (6.2.67) with (6.2.70), (6.2.72), and (6.2.73), we then get, as in case
1, that L H (y ) 0 for large.
j, (y ) and that
Case 3. We assume that y k, , that dg (
yk, , y ) R
dg (
yk, , y ) = o (j0 , ). We then get with (6.2.55) that
2
2 2
yk, , y ) u (y )
dg (
D (1 )
C12
(1 )
2
dg (
yi, , y )
2 2
u (y )
(6.2.74)
+ O (1)
2
yk, , y )
(4r (j 1) + o (1)) C21 D (1 ) dg (
k, (y ) .
j, (y ) dg (
yk, , y ), and since
Since R
2 2
u (y )
D (1 ) dg (
yk, , y )
(r j + 1) D (1 ) dg (
yk, , y )
+ O (1) C21
( n 1)(12)
(2n)(1)
j,2
Rj, (y )
( n 1)(12)
2+(2n)(1)
(r j + 1 + o (1)) D (1 ) C21 j,2
dg (
yk, , y )
.
By (6.2.3), (6.2.59), and (6.2.63),
j, k, (1 + o (1))
since k j 1. It follows that
r
( n 1)(12)
2 2
1
j,2
+ O (1)
Gi, (y )
u (y )
(6.2.75)
i=j
(r j + 1 + o (1)) D (1
) C21 dg
(
yk, , y )
k,
(y ) .
n
2 1
u (y ) vNj0 , (y )
D (1 )
100
212
.
171
j, (y )2 u (y )2
R
D (1 )
C12
(1 )
2
dg (
yi, , y )
2 2
u (y )
+ O (1)
(6.2.76)
j1
j, (y )2 + O (1) C 1
D (1 ) R
i, (y )
2
i=1
(j 1 +
o (1)) C21 D (1
j, (y )2 (y )
) R
k,
j, j,
(
j,
n
2 1
j,
)(12)
(2n)(1)
.
Rj, (y )
This leads to
j1
( n 1)(12)
2 2
1
+ O (1)
i,2
Gi, (y )
u (y )
i=1
(j 1 + o (1)) C21 D (1 )
(6.2.77)
( n 1)(12)
(2n)(1)2
j,2
.
Rj, (y )
Combining (6.2.67) with (6.2.76) and (6.2.77), we then get that L H (y ) 0
for large.
The study of these four cases clearly proves that, for large,
L H 0
(6.2.78)
ri=1
in Bx ( () j0 , ) \
Byi, (R () i, ). Using point
4.1,
(P1) of Theorem
(6.1.3), (6.2.3), (6.2.63), and the fact that () 12 d x0 , SNj0 \ {x0 } , we also
have that there exists C () > 0 such that
u C () H
(6.2.79)
172
CHAPTER 6
on Bx ( () j0 , ) \ ri=1 Byi, (R () i, ) . By [9], L satisfies the maximum principle. Since L u = 0, we then get with (6.2.78), (6.2.79), and (6.1.3)
that the equation of claim 6.2.4 holds in
x0 SNj Bx ( () j0 , ) \ ri=1 Byi, (R () i, ) .
0
Now that claim 6.2.4 is proved, we prove that (6.2.52) holds. This is the subject
of claim 6.2.5. Claim 6.2.5 is as follows:
C LAIM 6.2.5. If (6.2.51)j1 holds for some j = 2, . . . , r, then (6.2.51)j holds
also.
Proof of claim 6.2.5. We claim that there exists C > 0 such
that for any sequence
r
(6.2.80)
i=1
where
n
2n
2
dg (
yi, , y)
0i, (y) = i,
(6.2.81)
1
in the definition (6.2.57)
and j, is as in (6.2.63). We fix such that 0 < < n+2
of Aj, . Before proving (6.2.80), we introduce some notation and discuss some
consequences of claim 6.2.4. Given i = 1, . . . , j 1, we let
( n2 1)(12)
(2n)(1)
(6.2.82)
Rj, (y)
i, = y i, s.t. i, (y) j,
(R0 j , ) \ j1
j,
j0 ,
i=1
i,
(6.2.83)
if k {j, . . . , r} then By (R () j, ) .
k,
(6.2.84)
j,
By claim 6.2.4, (6.2.3), (6.2.63), and (6.2.84) [see also the proof of (6.1.128)] we
\By (R () k, ),
get that for any k {1, . . . , j 1}, and any y
k,
k,
$
#
n
( +1)(12)
2 1
2 1
(y)(n+2)
u (y)
(6.2.85)
C0 k, (y)
+ j0 ,2
R
when is large, and where C0 does not depend on . By claim 6.2.4, (6.2.3),
\r By (R () j, ),
(6.2.63), and (6.2.84), we also have that, for any y
i,
j,
i=j
!
n
( 2 +1)(12)
2 1
(n+2)(1)
C0 j,
u (y)
Rj, (y)
(6.2.86)
"
n
( 2 +1)(12)
(n+2)
+ j0 ,
R (y)
173
sup
( j0 , )
j20 , |u vNj0 , | 0
(6.2.87)
(y )
R
j0 ,
1 n
|u (y ) vNj0 , (y ) | = o j0 ,2 .
1
dg
k,
lim sup
+
u (y )
D0
0k, (y )
where
D0 = sup
sup |z|n2 u (z yk )
i=1,...,r zRn
(6.2.88)
and yk = xNj0 +qk is as in Theorem 4.1. By (6.2.3) and (6.2.63) we then get that
(6.2.80) holds with C > D0 .
1
dg (
yk, , y ) + as + for all k such
Case 3. We assume that k,
j, (y ) = O (j, ). Then, as in case 3 of the proof
that k {1, . . . , r}, and that R
(y ). Since
j, (y ) = R
of claim 6.1.8, we get that for large, R
1
dg (
yk, , y ) +
k,
Hence,
1 n
n2
j, 2 R
u (y ) vNj0 , (y )
j, (y )
n 1
j, (y ) 2
n
R
(y ) 2 1 u (y ) vN , (y )
R
=
j0
j,
n2 1
Rj, (y )
= o
j,
= o (1) .
174
CHAPTER 6
=
G (y , y) g u vNj0 , (y) dvg
Byj , (2R0 j0 , )
0
+
G (y , y) h (y) (y) u (y) vNj0 , (y) dvg .
Byj , (2R0 j0 , )
0
As in case 4 in the proof of claim 6.2.3, this equation reduces to
1 n
u (y ) vNj0 , (y ) = o j0 ,2
2 1
2 1
+
dvg .
vNj0 , (y)
G (y , y) u (y)
Byj , (R0 j0 , )
0
Using (6.1.80) and (6.2.87), we then get that
u (y ) vNj0 , (y )
1 n
2 1
dvg + o j0 ,2
G (y , y) u (y)
ri=1 Byi, ( j0 , )
(6.2.89)
yk, , y ) ,
Bk, = y k, i=1 Byi, ( j0 , ) s.t. dg (y , y) dg (
2
r
1
2
yk, , y ) .
Bk, = y k, i=1 Byi, ( j0 , ) s.t. dg (y , y) < dg (
2
Then,
2 1
k,
ri=1 Byi,
2 1
1
Bk,
G (y , y) u (y)
+
Since
dg (
yk, ,y )
k,
( j0 , ))
G (y , y) u (y)
2
Bk,
dvg
2 1
G (y , y) u (y)
dvg
(6.2.90)
dvg .
(6.2.91)
175
G (y , y) C2 2n2 dg (
yk, , y )
2 1
G (y , y) u (y)
dvg
Byk, (R()
k, )
C2 2
n2
2n
dg (
yk, , y )
2 1
Byk, (R()
k, )
2n
C2 2n2 dg (
yk, , y )
2 1
G (y , y) u (y)
where
D1 = 2n2 C2 2
From (6.2.85),
2 1
u (y)
#
C0
k,
dvg
1
u 22 1 Volg Byk, (R () k, ) 2 .
Byk, (R()
k, )
u (y)
n1
21
2 1
(y)
R () 2
( n +1)(12)
j0 ,2
R
(6.2.93)
(n+2)
(y)
1
for all y Bk,
\Byk, (R () k, ). Noting that
n+2
2
k,
(12)
2n
dg (
yk, , y )
2(n+2)
= 0k, (y ) k,
2 1
G (y , y) u (y)
dvg
1 \B
Bk,
k, )
y
k, (R()
( n +1)(12)
C0 j0 ,2
1 \B
Bk,
k, )
y
k, (R()
(y)(n+2) dvg
G (y , y) R
2(n+2)
(n+2)(1)
dg (
yk, , y)
dvg .
1 \B
Bk,
k, )
y
k, (R()
Since <
1
n+2 ,
1 \B
Bk,
k, )
y
k, (R()
dg (
yk, , y)
(n+2)(1)
dvg
n1 + o (1)
n(n+2)(1)
(R () k, )
.
(n + 2) (1 ) n
Given i = 1, . . . , r, let
Ii, = {j = 1, . . . , r s.t. dg (
yi, , yj, ) 2 j0 , } .
176
CHAPTER 6
(y)(n+2) dvg
G (y , y) R
1 \B
Bk,
k, )
y
k, (R()
Byj
0 ,
n1
n
(R0 j0 , )
G (y , y)
ri=1 Byi, ( j0 , )
n
n1
(y)n(n+2) dvg
R
n1
1(n+2)
= O (j0 , ) O ( j0 , )
.
In particular,
(y)
G (y , y) R
(n+2)
1 \B
Bk,
k, )
y
k, (R()
2(n+2)
dvg = o j0 ,
2 1
G (y , y) u (y)
dvg
1 \B
Bk,
k, )
y
k, (R()
(D2 +
o (1)) 0k,
(y ) + o
1 n
j0 ,2
(6.2.94)
where
n1
n(n+2)(1)
R ()
.
(n + 2) (1 ) n
Combining (6.2.92) and (6.2.94), we then get that
2 1
G (y , y) u (y)
dvg
D2 = C0 C2 2n2
1
Bk,
o (1)) 0k,
(D1 + D2 +
(y ) + o
(6.2.95)
(6.2.96)
1 n
j0 ,2
1
2 dg
2
Independently, it is easily checked that dg (
yk, , y)
(
yk, , y ) in Bk,
. Using (6.2.85) and (6.2.91) we can then write that
( n
2 1
2 1
(n+2)
2 +1)(12)
(n+2)(1)
C0 2
k, (y )
+ j0 ,
u (y)
R (y)
2
. Thus,
for all y Bk,
2 1
G (y , y) u (y)
dvg
2
Bk,
=O
( n +1)(12)
j0 ,2
+O
k,
2 1
(y )
(y)
G (y , y) R
(n+2)
2
Bk,
2
Bk,
G (y , y) dvg
dvg
177
As above,
(y)
G (y , y) R
(n+2)
2
Bk,
2(n+2)
.
dvg = o j0 ,
2n
G (y , y) dvg = O
dg (y , y)
dvg
2
Bk,
2
Bk,
=O
By
dg y
k, ,y
2
yk, , y )
= O dg (
Thus,
2 1
2
Bk,
G (y , y) u (y)
= O 0k, (y )
Since <
1
n+2
) dg (y , y)
2n
dvg
.
dvg
k,
dg (
yk, , y )
2(n+2)
1 n
+ o j0 ,2 .
and
yk, , y )
dg (
+
k,
as +, it follows that
1 n
2 1
G (y , y) u (y)
dvg = o 0k, (y ) + o j0 ,2 .
2
Bk,
(6.2.97)
Combining (6.2.90), (6.2.96), and (6.2.97), we then get that for any k such that
k = 1, . . . , j 1,
2 1
dvg
G (y , y) u (y)
(r By ( j , ))
i=1
k,
0
i,
(6.2.98)
1 n
(D1 + D2 + o (1)) 0k, (y ) + o j0 ,2
where D1 is as in (6.2.93), and D2 is as in (6.2.95). Given k {j, . . . , r}, we now
let
\ r By (R () j, ) s.t. R
j, (y) = dg (
k, = y
yk, , y) ,
j,
i=j
i,
r
1
1
Bk, = y k, i=1 Byi, ( j0 , ) s.t. dg (y , y) Rj, (y ) ,
2
1
2
r
Bk, = y k, i=1 Byi, ( j0 , ) s.t. dg (y , y) < Rj, (y ) .
2
178
CHAPTER 6
We write that
2 1
(r By ( j , ))
j,
i=1
0
i,
2 1
G (y , y) u (y)
1
Bk,
k=j
k=j
G (y , y) u (y)
dvg
dvg
(6.2.99)
2 1
2
Bk,
G (y , y) u (y)
dvg
2 1
G (y , y) u (y)
dvg .
1
j, (y )
Rj, (y ) + as +, we have that dg (y , y) 12 R
Since j,
for all y ri=j Byi, (R () j, ) when is large. Using (6.1.80) we then get that
2 1
G (y , y) u (y)
dvg
ri=j Byi, (R()j, )
C2 2
n2
j, (y )2n
R
j, (y )
C2 2n2 R
2n
2 1
u (y)
dvg
21
u 22 1 Volg ri=j Byi, (R () j, )
2n
2 1
(r j + 1 + o (1)) D1 j,
Rj, (y )
n
2 1
dvg
G (y , y) u (y)
j, (i=1 Byi, ( j0 , ))
r
2 1
G (y , y) u (y)
dvg
1
Bk,
k=j
(6.2.100)
k=j
2 1
2
Bk,
G (y , y) u (y)
dvg
2n
2 1
+ (r j + 1 + o (1)) D1 j,
.
Rj, (y )
n
2n
1
in Bk,
. Using (6.2.86) we also have that
2 1
u (y)
! n +1 (12)
)
(
(n+2)(1)
C0 j,2
dg (
yk, , y)
"
( n +1)(12)
(y)(n+2)
+j0 ,2
R
179
2 1
G (y , y) u (y)
dvg
1
Bk,
C0 C2 2
n2
n
j, (y )2n ( 2 +1)(12)
R
j,
( n +1)(12)
+C0 j0 ,2
As above,
1
Bk,
(n+2)(1)
1
Bk,
dg (
yk, , y)
dvg
(y)(n+2) dvg .
G (y , y) R
(y)(n+2) dvg = o 2(n+2) , and
G (y , y) R
j0 ,
(n+2)(1)
dg (
yk, , y)
dvg
1
Bk,
1
Bk,
n1 + o (1)
n(n+2)(1)
(R () j, )
.
(n + 2) (1 ) n
It follows that
2 1
1
Bk,
G (y , y) u (y)
n
2 1
(D2 + o (1)) j,
dvg
(6.2.101)
n
j, (y )2n + o 1 2
R
j0 ,
2
where D2 is as in (6.2.95). Independently, we have in Bk,
that
1
j, (y) = dg (
yk, , y) R
R
j, (y ) .
2
Using (6.2.86) we then get that
(n+2)(1)
! n +1 (12) R
j, (y )
(2 )
2 1
u (y)
C0 j,
2
"
( n +1)(12)
(y)(n+2)
+j0 ,2
R
2
. Independently, from (6.1.80),
in Bk,
j, (y )2
G (y , y) dvg = O R
2
Bk,
so that
2 1
2
Bk,
G (y , y) u (y)
dvg
( n +1)(12)
(n+2)(1)
= O j,2
Rj, (y )
+O
( n +1)(12)
j0 ,2
2
Bk,
j, (y )2n
= O j, R
G (y , y) dvg
2
Bk,
(y)
G (y , y) R
(n+2)
n
2 1
j, (y )
R
j,
dvg
(n+2)2
n
+ o j1,2 .
0
180
CHAPTER 6
Since <
2
Bk,
1
and j,
Rj, (y ) + as +, it follows that
n
1 n
2 1
2n
2 1
G (y , y) u (y)
dvg = o j,
+ o j0 ,2 .
Rj, (y )
1
n+2
Coming back to (6.2.100), we then get with (6.2.101) and the above equation that
2 1
dvg
G (y , y) u (y)
(r By ( j , ))
j,
i=1
0
i,
n
2n
(6.2.102)
2 1
(r j + 1 + o (1)) (D1 + D2 ) j,
Rj, (y )
1 n
+ o j0 ,2 .
, it follows from (6.2.89), (6.2.98), and
Noting that Byj0 , (R0 j0 , ) = jk=1
k,
(6.2.102) that
u (y ) vNj0 , (y ) (D1 + D2 + o (1))
1 n
0k, (y ) + o j0 ,2
j1
k=1
2n
2 1
+ (r j + 1) (D1 + D2 + o (1)) j,
Rj, (y )
n
(6.2.103)
By (6.2.63), this reduces to proving that Aj, = O (1) where Aj, is as in (6.2.57).
We proceed by contradiction and assume that
Aj, +
(6.2.104)
1)(12)
j, (y )(2n)(1) = k, (y ) (1 + o (1)) .
R
k, ,
k,
D (1 )
100
(y )
212
i,
(6.2.105)
dg (y , yk, ) 2
u (y ) vNj0 , (y ) .
181
i=1
dg (
yk, , y )
j0 ,
Cdg (
yk, , y )
+
n2 1
j, (y )
R
j,
n
2 1
(n2)
R ()
(n2)
j1
i, (y )
i=1
"
( n 1)(12)
(2n)(1)
j,2
+ o (1)
Rj, (y )
!
n
1
(n2)
yk, , y ) 2 k, (y ) (j 1) R ()
Cdg (
(n2)
"
j, (y )
R
+ o (1) .
+
j,
In particular,
1
D (1 ) 2 2
100
1 n (12)
dg (
yk, , y ) ( 2 )
C
K + o (1)
k,
where
K =
d (
y
j, (y )
R
j,
(6.2.106)
(n2)
+ O (1) .
,y )
=
dg (
yk, , y )
Rj, (y )
k,
n
( 2 1)(12) n2 1
.
k,
= O j,
182
CHAPTER 6
and this is in contradiction with (6.2.4). In particular, the above claim is proved and
dg (
yk, ,y )
j, (y ) = o (j, ),
+ as +. Then, thanks to (6.2.106), R
k,
and coming back to (6.2.105) we get that
n 1 (12)
k, ( 2 )
(n2)(1)
j, (y )(n2)(1)
dg (
yk, , y )
=
R
j,
n
( 2 1)(12) n2 1
= o k,
j,
n
( 2 1)(12) n2 1
j1,
= o k,
(n2)(1)
.
= o k,
Since
dg (
yk, , y )
+
k,
as +, this is the contradiction we were looking for. In particular, (6.2.104)
is absurd, and (6.2.103) is proved. As already mentioned, this ends the proof of
claim 6.2.5.
2
By claim 6.2.3, (6.2.51)j is true for j = 1. By induction, thanks to claim 6.2.5,
we then get that (6.1.51)j is true for all j = 1, . . . , r. In particular, (6.2.51)r is
true. Noting that (6.2.51)r implies (6.2.6), we have proved that (6.2.6) holds. As
already mentioned, by finite induction, repeating the above arguments, we in fact
proved that (6.2.2) is true. Then, (6.2.1) is also true, and the upper estimate on the
u s of Theorem 6.1 holds.
6.3 ASYMPTOTIC BEHAVIOR
We prove the estimate from below in Theorem 6.1. For that purpose, we let (x ) be
a sequence of points in M and compute an exact asymptotic expansion of u (x ).
We split this section into two subsections. The first subsection is concerned with
the case u0 0. The second subsection is concerned with the case u0 0.
First we introduce some notations. After passing to a subsequence, we can assume that (x ) converges and we let
x = lim x .
(6.3.1)
x
k = lim xk, .
(6.3.2)
(6.3.3)
183
n2 1
k,
k, (x) =
,
d (
xk, ,x)2
2
k, + gn(n2)
(6.3.4)
if x = y
1
S (x, y) =
(6.3.7)
n2
G (x, y)
if x = y
(n 2) n1 dg (x, y)
where (x, y) M M , and G is the Greens function of g + h . From the
property (P3) of Appendix A, S is indeed continuous on M M .
We assume in the following that the xi, s are ordered such that
1, 2, N, .
We define the sets i, , i = 1, . . . , N , in the following way:
1, = x M s.t. 1, (x) i, (x) , i = 1, . . . , N and
(6.3.8)
j, = x M \ j1
i=1 i, s.t. j, (x) i, (x) , i = 1, . . . , N
when j = 2, . . . , N . It is easily seen that M = N
i=1 i, and that the i, s are
disjoint. From (6.2.1) and (6.3.6), for any k {1, . . . , N } and any x k, ,
u (x) u0 (x) (1 + ) + Ck, (x)
(6.3.9)
(6.3.10)
184
CHAPTER 6
2 1
u (x ) =
G (x , x) u (x)
dvg
M
k=1
k,
G (x , x) u (x)
2 1
(6.3.11)
dvg
2 1
G (x , x) u (x)
dvg = IR, + IIR, + IIIR,
(6.3.12)
k,
where
IR, =
IIR, =
IIIR, =
k, Bx
k, (Rk, )
1 \B
Bk,
x
k, (Rk, )
2 \B
Bk,
x
k, (Rk, )
G (x , x) u (x)
2 1
G (x , x) u (x)
2 1
G (x , x) u (x)
2 1
dvg ,
dvg ,
dvg .
Ck, 2 1 +
4n (n 2)
2k,
n
+1
(n+2)
2
Ck,
dg (
xk, , x )
In particular,
n
2 +1
(n+2)
IIIR, = O k, dg (
xk, , x )
2
Bk,
G (x , x) dvg
185
From (6.1.80),
G (x , x) dvg = O
2
Bk,
2n
2
Bk,
dg (x , x)
dvg
=O
Bx
2n
dg x
k, ,x
2
xk, , x )
= O dg (
) dg (x , x)
dvg
.
Since dg (x , x
k, ) 1
k, + as +, we then get that
n
2n
2 1
.
dg (
xk, , x )
IIIR, = o k,
Coming back to the expression for k, (x ), it easily follows that, for any R > 0,
IIIR, = o k, (x ) .
(6.3.13)
1
Now we estimate IIR, . From (6.1.80) and the definition of Bk,
,
2n
G (x , x) C2 2n2 dg (
xk, , x )
1
for all x Bk,
. Together with (6.3.10) this gives that
2n
xk, , x )
IIR, Cdg (
1 \B
Bk,
x
k, (Rk, )
2 1
k,
dvg
n
2 1
2 1
k,
dvg = R, k,
M \Bx
k, (Rk, )
(6.3.14)
2 1
u
dvg .
(6.3.15)
k, Bx
k, (Rk, )
1
+
k,
n (n 2)
2i,
n (n 2)
186
CHAPTER 6
so that
2
dg (
xi, , z ) n (n 2) 2i,
k,
i,
1+
R2
n (n 2)
$
1 .
2 1
u
dvg =
k, Bx
k, (Rk, )
(6.3.16)
2 1
u
dvg
Bx
k, (Rk, )
2 1
u
dvg
Bx
k, (Rk, )\k,
2 1
2 1
u
dvg =
u
dvg
k, Bx
k, (Rk, )
Bx
k, (Rk, )
+O
1
Volg Bxk, (Rk, ) \k, 2
n
2 1
2 1
.
u
dvg + o k,
Bx
k, (Rk, )
1 n
2 1
lim k,2
u
dvg =
u2 1 dx
lim
R+ +
Bx
k, (Rk, )
Rn
n
n1
(n (n 2)) 2
u2 1 dx =
n
n
R
this leads to
Bx
k, (Rk, )
2 1
u
dvg =
n1
n
n
2 1
(n (n 2)) 2 + R, k,
187
xk, , x ) 1
Since dg (
k, + as +, we also have that
lim
k, (x )
n
2 1
2n
k, dg (
xk, , x )
= (n (n 2)) 2
Hence,
xk , x ) + R, ) k, (x ) .
IR, = (S (
(6.3.17)
2 1
G (x , x) u (x)
dvg = S (
xk , x ) + o (1) k, (x ) . (6.3.18)
k,
It follows that
2
\Bxk, (Rk, ) =
Bk,
lim IIIR, = 0 .
R+ +
1
Now we estimate IIR, . For any x Bk,
\Bxk, (Rk, ), we have that
xk, , x) dg (x , x
k, )
dg (x , x) dg (
xk, , x) (Rk + o (1)) k,
dg (
(R Rk + o (1)) k, .
1
\Bxk, (Rk, ),
Thus, for large, for R large, and for any x Bk,
1
dg (
xk, , x) .
2
Then we can write, using (6.1.80) and (6.3.10), that
2n
2 1
dg (
xk, , x)
k, (x)
dvg
IIR, C
dg (x , x)
1 \B
Bk,
x
k, (Rk, )
(6.3.19)
188
CHAPTER 6
n
2n
2 1
2 1
lim k,
dg (
xk, , x)
k, (x)
dvg = 0
lim
R+ +
M \Bx
k, (Rk, )
so that
1 n
IIR, = R, k,2
(6.3.20)
radius 1
k, , we also let
n
2 1
u expxk, (k, x) ,
uk, (x) = k,
2
IR, = k,
where
Vk, (R)
2
G (x)uk, (x)
dvgk,
G (x) = G expxk, (k, z ) , expxk, (k, x)
and
Vk, (R) =
1
k,
exp1
k, (Rk, ) .
xk, k, Bx
1 + o (1)
1 n
2n
2 1
k,2
1Vk, (R) dgk, (z , x)
uk, (x)
dvgk,
IR, =
(n 2) n1
Rn
189
where 1Vk, (R) is the characteristic function of Vk, (R). It is easily seen that
2
(Rn ) as +, where is the Euclidean metric. Using (6.3.3)
gk, in Cloc
and (6.3.16), we have that
1Vk, (R) 1Bxk (R) a.e.
as +. From point (P1) of Theorem 4.1, we have that
uk, u (. xk ) a.e.
as +. Finally, up to a subsequence,
dgk, (z , .)
2n
|z .|2n a.e.
as +, where z = lim+ z . By (6.3.10), 1Vk, (R) uk, is uniformly bounded in Rn . Using the Lebesgue-dominated convergence theorem we
then easily get that
1
1 n
2 1
IR, = k,2
|z x|2n u (x xk )
dx + R, .
(n 2) n1 Rn
Since u = u2
, we have that
2 1
|z x|2n u (x xk )
dx = u (z xk ) .
1
(n 2) n1
Rn
2
k,
k, (x ) u (z xk )
(6.3.22)
Combining (6.3.12) with (6.3.19), (6.3.20), and (6.3.22), we then get that
2 1
G (x , x) u (x)
dvg = (1 + o (1)) k, (x ) .
(6.3.23)
k,
2 1
G (x , x) u (x)
dvg = S (
xk , x ) + o (1) k, (x )
k,
N
xk , x ) + o (1) k, (x ) .
(S (
(6.3.24)
k=1
We already know from the preceding section and (6.3.6) that there exists C > 1
such that, for any x M ,
u (x) C
N
k=1
k, (x) .
190
CHAPTER 6
Let C > 1 be sufficiently large such that S (x, y) 2/C for all x, y M . Since
S is positive and continuous, such a C exists. Then, using (6.3.24), we get that,
for large,
u (x)
N
1
k,
C
k=1
and this proves Theorem 6.1. As a remark, the xi, s of Theorem 6.1 are not those
of Theorem 4.1, but the x
i, s defined in (6.3.3). Summarizing, Theorem 6.1 is
proved in the special case u0 0.
From standard elliptic theory, it easily follows from (6.3.24) that, up to a subsequence,
n 1
N
i, 2
1 n
lim 1, 2 u = Cn
G (
xi , .)
(6.3.25)
lim
+
+ 1,
i=1
2
(M \S), where
in Cloc
n
Cn = (n (n 2)) 2
(n 2) n1 ,
2 1
2 1
0
u (x ) u (x ) =
dvg
G (x , x) u (x)
u0 (x)
M
+
G (x , x) (h (x) h (x)) u0 (x) dvg .
M
so that
u (x ) u0 (x ) =
2 1
2 1
dvg + o (1) .
G (x , x) u (x)
u0 (x)
M \N
i, ( )
i=1 Bx
|u u0 | 0
2 1
2 1
G (x , x) u (x)
u0 (x)
dvg = o (1)
M \N
i, ( )
i=1 Bx
191
and that
N
i, ( )
i=1 Bx
G (x , x) u0 (x)
2 1
dvg = o (1) .
Hence,
u (x ) u0 (x )
N
(6.3.27)
2 1
=
dvg + o (1) .
G (x , x) u (x)
N
i, ( ))
k=1 k, (i=1 Bx
We fix k {1, . . . , N }, and let
N
1
1
Bk, = x k, i=1 Bxi, ( ) s.t. dg (x , x) dg (
xk, , x ) ,
2
N
1
2
xk, , x ) .
Bk, = x k, i=1 Bxi, ( ) s.t. dg (x , x) < dg (
2
1
2
1
2
Bk,
and Bk,
Bk,
= . Up to replacing by
Clearly, k, = Bk,
= max , ( max i, )1/2 ,
i=1,...,N
2 1
dvg
G (x , x) u (x)
k, (N
i, ( ))
i=1 Bx
(6.3.28)
IR, =
k, Bx
k, (Rk, )
G (x , x) u (x)
IIR, =
1 \B
Bk,
x
k, (Rk, )
G (x , x) u (x)
2 1
G (x , x) u (x)
2 1
IIIR, =
2 \B
Bk,
x
k, (Rk, )
2 1
dvg ,
dvg ,
dvg .
Now we estimate these three integrals. Here again we distinguish two cases.
Case 1. We assume that, up to a subsequence,
k, )
dg (x , x
+
k,
as +. First we estimate IIIR, . From (6.3.26) and the definition of
2
, noting that dg (
xk, , x) dg (
xk, , x ) dg (x , x), there exists C > 0
Bk,
2
independent of such that for any x Bk,
,
n
2 1 2
n
(
x
,
x
)
d
1
1
g
k,
2 1
C k, 2 1 +
+ 1
u (x)
4n (n 2)
2k,
n
(n+2)
2 +1
dg (
xk, , x )
+1 .
C k,
192
CHAPTER 6
It follows that
n
2 +1
(n+2)
xk, , x )
IIIR, = O k, dg (
+O
G (x , x) dvg
By (6.1.80),
2
Bk,
2
Bk,
2
Bk,
2n
2
Bk,
G (x , x) dvg
G (x , x) dvg = O
dg (x , x)
dvg
=O
Bx
2n
dg x
k, ,x
xk, , x )
= O dg (
) dg (x , x)
dvg
.
2
N
Since 0 and Bk,
i=1 Bxi, ( ), it also follows from (6.1.80) that
G (x , x) dvg = o (1) .
2
Bk,
Since dg (x , x
k, ) 1
k, + as +, this gives that
n
2n
2 1
dg (
xk, , x )
+ o (1) .
IIIR, = o k,
Coming back to the definition of k, , it easily follows that
IIIR, = o k, (x ) + o (1)
(6.3.29)
1
,
for all R > 0. Now we estimate IIR, . From (6.1.80) and the definition of Bk,
2n
G (x , x) C2 2n2 dg (
xk, , x )
1
for all x Bk,
. We can then write with (6.3.26) that
2n
IIR, Cdg (
xk, , x )
1 \B
Bk,
x
k, (Rk, )
+C
1 \B
Bk,
x
k, (Rk, )
2 1
k,
dvg
G (x , x) dvg
n
2 1
2 1
k,
dvg = R, k,
M \Bx
k, (Rk, )
G (x , x) dvg = o (1) .
1
Bk,
193
(6.3.30)
Finally, we estimate IR, . It is easily checked that the arguments used to prove
(6.3.17) can be used here. Following the proof of (6.3.17) we then get that
xk , x ) + R, k, (x ) .
(6.3.31)
IR, = S (
Combining (6.3.28) and (6.3.29)(6.3.31), it follows that
2 1
dvg
G (x , x) u (x)
k, (N
i, ( ))
i=1 Bx
= S (
xk , x ) + o (1) k, (x ) + o (1) .
(6.3.32)
It follows that
2
\Bxk, (Rk, ) =
Bk,
lim IIIR, = 0 .
R+ +
1
Now we estimate IIR, . For any x Bk,
\Bxk, (Rk, ), we have that
xk, , x) dg (x , x
k, )
dg (x , x) dg (
xk, , x) (Rk + o (1)) k,
dg (
(R Rk + o (1)) k, .
1
\Bxk, (Rk, ),
Thus, for large, for R large, and for any x Bk,
1
dg (
xk, , x) .
2
Then we can write, using (6.1.80) and (6.3.26), that
2n
2 1
dg (
xk, , x)
k, (x)
dvg
IIR, C
dg (x , x)
1 \B
Bk,
x
k, (Rk, )
+C
1
Bk,
G (x , x) dvg
(6.3.33)
194
CHAPTER 6
n
2n
2 1
2 1
lim k,
dg (
xk, , x)
k, (x)
dvg = 0 .
lim
R+ +
M \Bx
k, (Rk, )
Moreover, since dg (
xk, , x ) 1
k, Rk as +, we also have that
1 n2
2
R
1 n
k
+ o (1) .
k, (x ) = k,2
1+
n (n 2)
1
Independently, since 0 and Bk,
N
i, ( ), it also follows from
i=1 Bx
(6.1.80) that
G (x , x) dvg = o (1) .
1
Bk,
IIR, = R, k, (x ) + 1 .
(6.3.34)
Finally, we estimate IR, . It is easily checked that the arguments used to prove
(6.3.22) can be used here. Following the proof of (6.3.22) we then get that
IR, = 1 + R, k, (x ) .
(6.3.35)
Combining (6.3.28) with (6.3.33)(6.3.35), it follows that
2 1
dvg
G (x , x) u (x)
k, (N
i, ( ))
i=1 Bx
(6.3.36)
= (1 + o (1)) k, (x ) + o (1) .
This ends case 2.
From cases 1 and 2, and since S (x, x) = 1,
2 1
dvg
G (x , x) u (x)
k, (N
i, ( ))
i=1 Bx
= (S (
xk , x ) + o (1)) k, (x ) + o (1)
for all k {1, . . . , N }. By (6.3.27), we then get that
u (x ) = u0 (x ) (1 + o (1))
+
N
xk , x ) + o (1) k, (x ) .
(S (
(6.3.37)
k=1
We already know form the preceding section and (6.3.6) that there exists C > 1
such that, for any x M ,
u (x) u0 (x) (1 + o (1)) + C
N
k=1
k, (x) .
195
Let C > 1 be sufficiently large such that S (x, y) 2/C for all x, y M . Since
S is positive and continuous, such a C exists. Then, thanks to (6.3.37), we get
that, for large,
N
1
k, (x)
u (x) u (x) (1 + o (1)) +
C
0
k=1
and this proves Theorem 6.1. As a remark, the xi, s of Theorem 6.1 are not those
of Theorem 4.1, but the x
i, s defined in (6.3.3). Summarizing, Theorem 6.1 is
proved in the special case u0 0. Together with subsection 6.3.1, Theorem 6.1 is
proved.
The remark following Theorem 6.1 easily follows from (6.3.37) and the fact that
S (x, x) = 1 for all x. We prove now that the bubbles of Theorem 6.1 satisfy
Theorem 3.1. More precisely, we prove that, up to a subsequence,
u = u0 +
N
i, + o (1)
(6.3.38)
i=1
1
2
and
Ig (u) =
1
2
1
|u|2 + h u2 dvg
2
1
|u|2 + h u2 dvg
2
(6.3.39)
|u|2 dvg
|u|2 dvg .
Since
2 1
g u + h u = u
in M , we have that
Ig
1
(u ) =
n
u2 dvg .
Since
g u0 + h u0 = (u0 )2
in M , we also have that
1
Ig u0 =
n
(u0 )2 dvg .
u2 dvg =
(u0 )2 dvg + N Knn .
lim
+
(6.3.40)
196
CHAPTER 6
lim
u2 dvg
+
= lim
u (x) +
2
N
i, (x) S (
xi , x)
(6.3.41)
dvg .
i=1
Since S is continuous on M M ,
u0 (x) +
N
2
i, (x) S (
xi , x)
i=1
(u0 )2 dvg +
N
(u0 )2
+O
M
+O
i, (x) S (
xi , x)
dvg
2 1
i,
dvg
(6.3.42)
i,
i=1
N
2
i=1
N
dvg
dvg
i=1
i, dvg 0 and
M
2 1
i,
dvg 0
u0 (x) +
N
2
i, (x) S (
xi , x)
i=1
(u0 )2 dvg +
N
dvg
(6.3.43)
2
i, (x) S (
xi , x)
dvg + o (1) .
i=1
2
i, (x) S (
xi , x)
dvg
i=1
i=1
i, (x)
2
2
S (
xi , x)
+O
i=j
dvg
2 1
i, j,
dvg .
(6.3.44)
197
2
2
i, (x) S (
xi , x) dvg
M
2
2
=
i, (x) S (
xi , x) dvg
Bx
i, (Ri, )
M \Bx
i, (Ri, )
i, (x)
2
2
S (
xi , x)
= (1 + o (1))
Bx
i, (Ri, )
M \Bx
i, (Ri, )
i, (x)
i, (x)
2
2
dvg
dvg
2
S (
xi , x)
dvg .
2
lim
i, (x) dvg = Knn
lim
R+ +
and that
Bx
i, (Ri, )
lim
lim
R+ +
Thus,
M \Bx
i, (Ri, )
lim
i, (x)
2
i, (x)
2
S (
xi , x)
2
dvg = 0 .
dvg = Knn
(6.3.45)
2 1
2 1
i, j,
dvg
+ j, i,
M
2 1
2 1
i, j,
dvg
+ j, i,
Bx
j, (Rj, )
+
M \Bx
j, (Rj, )
2
j,
dvg
Bx
j, (Rj, )
Bx
j, (Rj, )
M \Bx
j, (Rj, )
2
i,
dvg
+
M
221
2
i,
dvg
2 1
2 1
+ j, i,
i, j,
dvg
Bx
j, (Rj, )
221
Bx
j, (Rj, )
2
j,
dvg
2 1
2
2
i,
dvg
221
21
2
j,
dvg
2
i,
dvg
21
M \Bx
j, (Rj, )
2
j,
dvg
21
.
21
198
CHAPTER 6
As above,
M \Bx
j, (Rj, )
where lim
2
j,
dvg = R,
R+ +
get that
2 1
2 1
i, j,
+ j, i,
dvg = O
Bx
j, (Rj, )
21
2
+ R,
i,
dvg
+O
If
dg (xi, ,xj, )
i,
221
2
.
i,
dvg
Bx
j, (Rj, )
+ as +, we have that
Bx
j, (Rj, )
2n n
2
i,
dvg = O ni, dg (xi, , xj, )
j,
= o (1) .
If dg (xi, , xj, ) = O (i, ), we then have by (4.1.6) that j, = o (i, ) and thus
that
2
n
i,
dvg = O n
i, j,
Bx
j, (Rj, )
= o (1) .
Therefore,
2 1
i, j,
dvg 0
(6.3.46)
N
i, .
i=1
2
|R |2 dvg 0 and
R
dvg 0
M
(6.3.47)
2
dvg 0
(6.3.48)
R
M
199
|R |2 dvg
M
|u |2 dvg +
i=j
i=1
|i, |2 dvg
u u0 dvg
(i, j, ) dvg 2
i=1
|u0 |2 dvg +
(6.3.49)
(u i, ) dvg + 2
u0 i, dvg .
i=1
2
|u | dvg
(6.3.50)
|u0 |2 dvg ,
u u0 dvg
M
M
0
u i, dvg 0 ,
M
|i, |2 dvg Knn
(6.3.51)
(i, j, ) dvg 0
(6.3.52)
(u i, ) dvg =
(g u )i, dvg
M
2 1
i, dvg + o(1) .
(u i, ) dvg =
u
M
2 1
u
i, dvg
M
2 1
=
u
i, dvg +
Bx
i, (Ri, )
Since
M \Bx
i, (Ri, )
lim
lim
R+ +
M \Bx
i, (Ri, )
2 1
i, dvg .
u
i, (x)
2
dvg = 0 ,
200
CHAPTER 6
it follows that
2 1
u
i, dvg
where
lim
Bx
i, (Ri, )
2 1
u
i, dvg + R,
R+ +
2 1
u
i, dvg = Knn + R,
Bx
i, (Ri, )
(u i, ) dvg Knn
(6.3.53)
|R |2 dvg 0
(6.3.54)
Appendix A
The Greens Function on Compact Manifolds
We discuss in this appendix existence and properties of Greens functions for operators like g + h. We let (M, g) be a smooth compact Riemannian manifold of
dimension n 3, and let h C 0, (M ), 0 < < 1. We assume that g + h is
coercive, and let > 0 be such that for any u H12 (M ),
2
2
(A1)
|u| + hu dvg
|u|2 + u2 dvg .
M
Given K > 0 and > 0, we let H(K, ) be the set of the functions h in C 0, (M )
which are such that (A1) is satisfied and
|h(x)| K for all x ,
|h(y) h(x)| Kdg (x, y) for all x, y .
(A2)
202
APPENDIX A
x (y)
(n 2)n1 dg (x, y)n2
where x (y) = (x, y), and n1 is the volume of the unit (n 1)-sphere. Following Aubin [7], it is easily seen that there exists C > 0 such that for any x, y,
x = y,
C
,
dg (x, y)n2
C
|g,y H(x, y)|
dg (x, y)n2
|H(x, y)|
(A4)
(A5)
Z(x, y) =
X(x, z)Y (z, y)dvg (z) .
M
From a result by Giraud [38] (see also Aubin [7]) Z : M M \DM R is continuous and there exists C(, , M, g) > 0, depending only on , , and (M, g),
such that
|Z(x, y)| CX CY C(, , M, g)dg (x, y)+n if + < n ,
|Z(x, y)| CX CY C(, , M, g) (1 + | ln dg (x, y)|) if + = n , and (A6)
|Z(x, y)| CX CY C(, , M, g) if + > n .
Moreover, when + > n, Z is continuous on M M . We let 1 be the function
from M M \DM to R given by
1 (x, y) = g,y H(x, y) h(y)H(x, y)
(A7)
(A8)
We also let
203
n
+1
2
where E 2 is the greatest integer not exceeding n2 . It is easily seen from (A4)
and (A6) that for h H(K, ), any i 1, . . . , k + 1, and any x = y,
k=E
n
|i (x, y)|
C
if 2i < n ,
dg (x, y)n2i
(A9)
where C > 0 depends only on (M, g) and K. We also have that k and k+1 are
continuous on M M . Given x M , we let ux be given by
g ux + hux = k+1,x
(A10)
where k+1,x (y) = k+1 (x, y). Since g + h is coercive, ux exists. We then let
G : M M \DM R be given by
k
G(x, y) = H(x, y) +
i (x, z)H(z, y)dvg (z) + u(x, y)
(A11)
i=1
u
(x) =
G(x, y) (g u
(y) + h(y)
u(y)) dvg (y) .
M
In particular, (A3) is satisfied. Now we claim that there exists C > 0, depending
only on (M, g) and K, such that for any x, y, y M ,
|k+1 (x, y ) k+1 (x, y)| Cdg (y, y ) .
(A12)
In order to prove this claim, we proceed as follows. It is easily checked from (A7)
and (A9) that
|k+1 (x, y ) k+1 (x, y)|
Cdg (y, y ) + C
|f (z, y ) f (z, y)| dvg (z)
M
1
1
+C
dg (z, y)n2 dg (z, y )n2 dvg (z)
M
(A13)
where C > 0 depends only on (M, g) and K. We let > 0 to be chosen later on,
and assume that dg (y, y ) < /2. We can write that
1
1
dvg (z)
dg (z, y)n2
n2
dg (z, y )
M
1
1
dvg (z)
=
(A14)
n2
n2
dg (z, y )
By ( ) dg (z, y)
1
1
dvg (z) .
+
n2
dg (z, y )n2
M \By ( ) dg (z, y)
204
APPENDIX A
1
1
dvg (z) Cdg (y, y )
n2
dg (z, y )n2
M \By ( ) dg (z, y)
(A15)
where C( ) > 0 depends only on (M, g), , and . Similarly, we can write that
n3
1
1
(n 2) (dg (z, y) + dg (z, y ))
dg (y, y )
dg (z, y)n2
dg (z, y )n2
dg (z, y)n2 dg (z, y )n2
so that
1
1
dvg (z)
n2
n2
dg (z, y )
By ( ) dg (z, y)
(A16)
By ( )
We choose > 0 sufficiently small such that for any x M , and any s, t Rn , if
|s| and |t| , then
1
|t s| dg (expx (s), expx (t)) C( )|t s|
C( )
where C( ) > 1 does not depend on x, s, and t. We let z = expy (t) and let
y = expy (t0 ), |t0 | < /2. Then easy computations give that
n3
(|t| + |t t0 |)
dt C( )
(A17)
I(y, y ) C( )
n2 |t t |n2
|t|
0
B0 ( )
where C( ) > 0 depends only on (M, g) and . Combining (A14)(A17), we
proved that there exists C > 0, depending only on (M, g), such that for any points
y, y M , if dg (y, y ) < /2, then
1
1
dvg (z) Cdg (y, y ) .
(A18)
dg (z, y)n2
n2
d
(z,
y
)
g
M
Clearly, (A18) extends to all y, y M , so that, from (A13) and (A18),
|k+1 (x, y ) k+1 (x, y)|
Cdg (y, y ) + C
|f (z, y ) f (z, y)| dvg (z)
(A19)
for all y, y M , where C > 0 depends only on (M, g) and K. We can write (see,
for instance, Aubin [7]) that
f (x, y) =
Q(x, y)
dg (x, y)n1
205
and Qx (y) = Q(x, y). In particular, there exists C > 0, depending only on (M, g),
such that for any x, y, y ,
|Q(x, y)| Cdg (x, y) and |Q(x, y ) Q(x, y)| Cdg (y, y ) .
It follows that
1
1
dvg (z) .
+C
dg (z, y)
n1
)n1
d
(z,
y)
d
(z,
y
g
g
M
Writing that
n2
1
1
(n 1) (dg (z, y) + dg (z, y ))
dg (y, y )
dg (z, y)n1
dg (z, y )n1
dg (z, y)n1 dg (z, y )n1
we get, as when proving (A18), that for any y, y M ,
1
1
dvg (z) Cdg (y, y )
dg (z, y)
dg (z, y)n1
dg (z, y )n1
(A20)
where C > 0 depends only on (M, g). Combining (A19) and (A20), it follows that
there exists C > 0, depending only on (M, g) and K, such that for any x, y, y in
M,
|k+1 (x, y ) k+1 (x, y)| Cdg (y, y ) .
This proves (A12). Now that we have (A12), we return to the definition (A10) of
ux . Clearly, ux C 2, (M ). Thanks to standard elliptic estimates as in GilbargTrudinger [37], see for instance Theorems 6.2 and 8.17, we have that
ux C 2, C ux C 0 + k+1,x C 0,
and
ux C 0 C ux 2 + k+1,x C 0
and we get, using (A9), (A12), and the above equations, that for any x M ,
ux 2
ux C 2, C
(A21)
where C > 0 depends only on (M, g), K, and . Coming back to the definition
(A11) of G, and combining (A4), (A6), (A9), and (A21), we get that there exists
C > 0, depending only on (M, g), K, and , such that for any x, y M ,
dg (x, y)n2 G(x, y) C .
This proves (P2). Noting that
g (ux ux ) + h(ux ux ) = k+1,x k+1,x
we get, thanks to standard elliptic estimates as above, that
ux ux C 0 Ck+1,x k+1,x C 0
(A22)
206
APPENDIX A
(A23)
2,
(M \{x}). Given x, y M such that
in M \{x}, we also have that Gx is in Cloc
0 < dg (x, y) < 1, we get with (A4), (A6), (A9), and (A21), that
(A24)
It follows that there exists > 0 such that for any x, y M , if 0 < dg (x, y) < ,
just said, G
x has its support in M \Bx (). Moreover, Gx is Lipschitz so that
2
Gx H1 (M ). Using (A23) we can then write
0=
G
x (g Gx + hGx ) dvg
M
2
2
=
|G
dvg .
x | + h(Gx )
M
It follows that G(x, y) = G(y, x) for all x = y, and (P1) is proved. We are thus
left with the proof of (P3). The first equation in (P3) is an easy consequence of
(A24). It remains to prove the second equation in (P3). We fix x, y M such that
0 < dg (x, y) < , and let y Rn be such that y = expx (
y ). Given X S n1 ,
n
the unit sphere of R , and t R, we set
y + tX) .
yt = expx (
Then,
d
(
y , X)
H(x, yt )t=0 =
dt
n1 dg (x, y)n1 |
y|
(A25)
where (., .) is the Euclidean scalar product, and we also have that
d
C
H(z, yt )
dt
t=0
dg (z, y)n1
207
(A26)
where C > 0 depends only on (M, g). For any r > 0, and any i = 1, . . . , k, from
(A26),
d
i (x, z)H(z, yt )dvg (z)
dt
t=0
M \By (r)
(A27)
d
H(z, yt )
dvg (z) .
=
i (x, z)
dt
t=0
M \By (r)
Independently, for r > 0 small, r < dg (x, y)/2, and t > 0 small, we can write that
H(z, yt ) H(z, y)
dvg (z)
i (x, z)
By (r)
t
dg (x, y)n2 By (r)
t
n3
(dg (z, yt ) + dg (z, y))
C
dvg (z)
H(z, yt ) H(z, y)
dvg (z) C(r)
i (x, z)
(A28)
By (r)
t
where C(r) 0 as r 0. Combining (A27) and (A28) we then get that for any
i = 1, . . . , k,
d
i (x, z)H(z, yt )dvg (z)
dt
t=0
M
(A29)
d
H(z, yt )
=
i (x, z)
dvg (z) .
dt
t=0
M
Using (A25) and (A29), coming back to the definition (A11) of G, we can write
that
d
G(x, yt )t=0
dt
(
y , X)
=
(A30)
n1 dg (x, y)n1 |
y|
k
d
d
+
H(z, yt )t=0 dvg (z) + ux (yt )t=0 .
i (x, z)
dt
dt
i=1 M
By (A21),
d
ux (yt ) C
dt
t=0
(A31)
208
APPENDIX A
where C > 0 depends only on (M, g), K, and . Independently, as an easy consequence of (A6), (A9), and (A26), for any i = 1, . . . , k,
C
d
H(z, yt ) t=0 dvg (z)
(A32)
i (x, z)
dt
d
(x,
y)n2
g
M
where C > 0 depends only on (M, g) and K. In geodesic normal coordinates at x,
we refer for instance to Lemma 8 in Hebey-Vaugon [47], |g ij (y) ij | Cdg (x, y)
where C > 0 can be chosen such that it depends only on (M, g). Combining
(A30)(A32), we then get that
n1 dg (x, y)n1 |y G(x, y)| 1 Cdg (x, y)
where C > 0 depends only on (M, g), K, and . This ends the proof of (P3).
Appendix B
Coercivity Is a Necessary Condition
We prove that the operator g + h has to be coercive if the equation in Theorem
6.1 is true, and thus that the coercivity of the operator g + h in Theorem 6.1 is
a necessary condition. We let (M, g) be a smooth compact Riemannian manifold
of dimension n 3, and for h C 0, (M ), (0, 1), and u H12 (M ), we let
|u|2 + hu2 dvg .
Ih (u) =
M
Then we let h = inf u Ih (u) where the infimum is with respect to the us in
H12 (M ) such that M u2 dvg = 1. The operator g + h is said to be coercive
[see (2.1.1)] if there exists C > 0 such that Ih (u) Cu2H 2 for all u H12 (M ).
1
Easy claims are as follows:
C LAIM 1. The operator g + h is coercive if and only if h > 0.
uh C 2, (M ), uh > 0 everywhere, such
C LAIM 2. Whatever h is, there
exists
2
that g uh + huh = h uh and M uh dvg = 1.
C LAIM 3. If there exists u C 2, (M ) such that u > 0 and g u + hu > 0
everywhere, then g + h is coercive.
We prove claim 1 by noting that, if h > 0, then
|u|2 dvg +
hu2 dvg
M
M
|u| dvg +
hu2 dvg + (1 )h
u2 dvg
M
M
M
|u|2 dvg +
u2 dvg
M
0<
uh (g u + hu) dvg
M
= h
uuh dvg
M
210
APPENDIX B
1
C
0,
(B1)
(M ) as +. We assume that
N
Bi u (1 + )u0 + C
i=1
Bi
(B2)
i=1
n2
2
Bi (x) =
N
i,
2i,
dg (xi, ,x)2
n(n2)
the xi, s are converging sequences of points in M , the i, s are positive real
numbers such that i, 0 as +, C is a positive constant independent of
and x, N N , and u0 H12 (M ) is a nonnegative function such that u u0
in H12 (M ). Then we claim that, necessarily, the operator g + h is coercive.
We prove the claim in what follows. We let = h and u = uh . By the
maximum principle and regularity theory, u0 C 2, (M ),
g u0 + h u0 = (u0 )2
2 1
(g u + h u ) u dvg =
u
u dvg .
M
(B3)
Since = 0,
(g u + h u ) u dvg
M
=
u (g u + h u ) u dvg +
(h h ) u u dvg
M
M
=
(h h ) u u dvg
M
(g u + h u ) u dvg = o
M
u dvg
(B4)
211
u dvg C
M
Bi dvg
i=1
n
2 1
Bi dvg + O i,
Bxi, ()
n
2 1
Bi dvg Ci,
Bxi, ()
+1
2
Ci,
rn1
2i, + r2
Bi dvg
i,
dr
n2
2
rn1
n2
(1 + r2 ) 2
n
i,
2 +1
Ci,
rdr
C+
0
dr
so that
Bxi, ()
n
2 1
.
Bi dvg = O i,
n
2 1
.
(g u + h u ) u dvg = o 1,
(B5)
Still, using (B2), and since u > 0 everywhere, we can write that
2 1
2 1
u
u dvg C
u
dvg
Bx1, (1, )
Bx1, (1, )
B1
2 1
dvg .
Noting that
Bx1, (1, )
1 2 1
dvg C
1,
0
n
2 1
C1,
it follows that
1,
2
1, + r2
rn1 dr
rn1
(1 + r2 )
n
2 1
2
u
u dvg C1,
n+2
2
n+2
2
dr
(B6)
for some C > 0 independent of . Inserting (B5) and (B6) into (B3), we get the
desired contradiction. This proves the above claim.
Bibliography
214
[13]
BIBLIOGRAPHY
[14] Brezis, H., and Lieb, E., A relation between pointwise convergence of functions and convergence of functionals, Proc. Amer. Math. Soc., 88, 486490,
1983.
[15] Brezis, H., and Nirenberg, L., Positive solutions of nonlinear elliptic equations involving critical Sobolev exponents, Comm. Pure Appl. Math., 36,
437477, 1983.
[16] Brezis, H., and Peletier, L. A., Asymptotics for elliptic equations involving
critical Sobolev exponents, in Partial Differential Equations and the Calculus of Variations, eds. F. Colombini, A. Marino, L. Modica, and S. Spagnalo,
Birkhauser, Basel, 1989.
[17] Caffarelli, L. A., Gidas, B., and Spruck, J., Asymptotic symmetry and local behavior of semilinear elliptic equations with critical Sobolev growth,
Comm. Pure Appl. Math., 42, 271297, 1989.
[18] Chang, A., and Yang, P., A perturbation result in prescribing scalar curvature
on S n , Duke Math. J., 64, 2769, 1991.
[19] Chang, A., Gursky, M., and Yang, P., The scalar curvature equation on 2and 3-spheres, Calc. Var. Partial Differential Equations, 1, 205229, 1993.
[20] Chavel, I., Riemannian geometryA modern introduction, Cambridge Tracts
in Mathematics, 108, Cambridge University Press, Cambridge, 1993.
[21] Chern, S. S., A simple intrinsic proof of the Gauss-Bonnet formula for closed
Riemannian manifolds, Ann. Math., 45, 747752, 1944.
[22] do Carmo, M. P., Riemannian geometry, Mathematics: Theory and Applications, Birkhauser, Boston, 1992.
[23] Druet, O., The best constants problem in Sobolev inequalities, Math. Ann.,
314, 327346, 1999.
[24]
, Optimal Sobolev inequalities and extremal functions. The threedimensional case, Indiana Univ. Math. J., 51, 6988, 2002.
[25]
[26]
, Sharp local isoperimetric inequalities involving the scalar curvature, Proc. Amer. Math. Soc., 130, 23512361, 2002.
[27]
BIBLIOGRAPHY
215
[28]
2003.
[29] Druet, O., and Hebey, E., The AB program in geometric analysis. Sharp
Sobolev inequalities and related problems, Memoirs of the American Mathematical Society, MEMO/160/761, 2002.
[30]
[31]
[32] Druet, O., Hebey, E., and Robert, F., A C 0 -theory for the blow-up of second order elliptic equations of critical Sobolev growth, Electronic Research
Announcements of the AMS, 9, 1925, 2003.
[33] Druet, O., and Robert, F., Asymptotic profile for the sub-extremals of the
sharp Sobolev inequality on the sphere, Comm. P.D.E., 25, 743778, 2001.
[34] Ekeland, I., and Ghoussoub, N., Selected new aspects of the calculus of
variations in the large, Bull. Amer. Math. Soc., 39, 207265, 2002.
[35] Fenchel, W., On total curvatures of Riemannian manifolds: I, J. London
Math. Soc., 15, 1522, 1940.
[36] Gallot, S., Hulin, D., and Lafontaine, J., Riemannian geometry, second edition, Universitext, Springer, Berlin, 1993.
[37] Gilbarg, G., and Trudinger, N. S., Elliptic partial differential equations
of second order, second edition, Grundlehren der Mathematischen Wissenschaften, 224, Springer, Berlin, 1983.
[38] Giraud, G., Sur le probl`eme de Dirichlet generalise, Ann. Sc. Ecole Norm.
Sup., 46, 131245, 1929.
[39] Hamilton, R. S., Three-manifolds with positive Ricci curvature, J. Differential Geom., 17, 255306, 1982.
[40] Han, Z. C., Asymptotic approach to singular solutions for nonlinear elliptic
equations involving critical Sobolev exponent, Ann. Inst. H. Poincare. Anal.
Non Lineaire, 8, 159174, 1991.
[41] Han, Q., and Lin, F., Elliptic partial differential equations, CIMS Lecture
Notes, Courant Institute of Mathematical Sciences, Vol. 1, 1997. Second
edition published by the American Mathematical Society, 2000.
[42] Hebey, E., Courbure scalaire et geometrie conforme, J. Geom. Phys., 10,
345380, 1993.
216
BIBLIOGRAPHY
[43]
, Introduction a` lanalyse non lineaire sur les varietes, Diderot Editeur, Fondations, Paris, 1997.
[44]
, Nonlinear analysis on manifolds: Sobolev spaces and inequalities, CIMS Lecture Notes, Courant Institute of Mathematical Sciences, Vol.
5, 1999. Second edition published by the American Mathematical Society,
2000.
[45]
, Nonlinear elliptic equations of critical Sobolev growth from a dynamical viewpoint, Contemporary Mathematics, Proceedings of the conference in honor of H. Brezis and F. Browder, to appear.
[46]
, Variational methods and elliptic equations in Riemannian geometry, Preprint, Notes from lectures given at ICTP, 2003.
[47] Hebey, E., and Vaugon, M., The best constant problem in the Sobolev embedding theorem for complete Riemannian manifolds, Duke Math. J., 79,
235279, 1995.
[48]
[49]
2001.
, Uber
Riemannsche Mannigfaltigkeiten mit positiver Krummung,
Comment. Math. Helv., 35, 4754, 1961.
[57] Li, Y. Y., and Zhu, M., Yamabe type equations on three dimensional Riemannian manifolds, Communications in Contemporary Math., 1, 150, 1999.
[58] Lions, P.-L., The concentration-compactness principle in the calculus of
variations. The limit case I, II, Rev. Mat. Iberoamericana, no. 1, 145-201;
no. 2, 45121, 1985.
BIBLIOGRAPHY
217
[59] Lohkamp, J., Metrics of negative Ricci curvature, Ann. Math., 140, 655683,
1994.
[60] Obata, M., The conjectures on conformal transformations of Riemannian
manifolds, J. Differential Geom., 6, 247258, 1971/72.
[61] Rauch, H. E., A contribution to differential geometry in the large, Ann.
Math., 54, 3855, 1951.
[62] Robert, F., Asymptotic behavior of a nonlinear elliptic equation with critical
Sobolev exponent The radial case, Adv. Differential Equations, 6, 821846,
2001.
[63]
[64] Sacks, P., and Uhlenbeck, K., On the existence of minimal immersions of
2-spheres, Ann. Math., 113, 124, 1981.
[65] Sakai, T., Riemannian geometry, Translations of Mathematical Monographs,
149, American Mathematical Society, Providence, R.I., 1996.
[66] Schoen, R., Conformal deformation of a Riemannian metric to constant
scalar curvature, J. Differential Geom., 20, 479495, 1984.
[67]
, Variational theory for the total scalar curvature functional for Riemannian metrics and related topics, in Topics in calculus of variations (Montecatini Terme, 1987). Lecture Notes in Mathematics, 1365, Springer, Berlin,
1989, 120154.
[68]
[69]
, A report on some recent progress on nonlinear problems in geometry, Surveys in Differential Geometry, Cambridge, Mass., 1990, 201241.
[70]
[71] Schoen, R., and Zhang, D., Prescribed scalar curvature on the n-sphere,
Calc. Var. Partial Differential Equations, 4, 125, 1996.
[72] Spivak, M., A comprehensive introduction to differential geometry, 5 vols.,
Publish or Perish, Houston, Tex., 1970.
[73] Struwe, M., A global compactness result for elliptic boundary problems involving limiting nonlinearities, Math. Z., 187, 511517, 1984.
218
BIBLIOGRAPHY
[74] Trudinger, N. S., Remarks concerning the conformal deformation of Riemannian structures on compact manifolds, Ann. Scuola Norm. Sup. Pisa, 22,
265274, 1968.
[75] Tsukamoto, Y., On Riemannian manifolds with positive curvature, Mem.
Fac. Sci. Kyushu Univ. Ser. A, 15, 9096, 1961/1962.
[76] Wente, H. C., Large solutions to the volume constrained Plateau problem,
Arch. Rational Mech. Anal., 75, 5977, 1980.