€¦ · acknowledgements my supervisor ezequiel rodrigues barbosa and my co-supervisor marcos...

111
Universidade Federal de Minas Gerais Institute of Exact Sciences - ICEx Department of Mathematics PhD Thesis About a Class of Optimal Sobolev Vector Inequalities of Second Order Aldo Peres Campos e Lopes Advisor : Prof. Ezequiel Rodrigues Barbosa Co-Advisor: Prof. Marcos Montenegro Belo Horizonte - April 9, 2014

Upload: others

Post on 04-Aug-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

Universidade Federal de Minas Gerais

Institute of Exact Sciences - ICEx

Department of Mathematics

PhD Thesis

About a Class of Optimal Sobolev Vector Inequalitiesof Second Order

Aldo Peres Campos e Lopes

Advisor : Prof. Ezequiel Rodrigues Barbosa

Co-Advisor: Prof. Marcos Montenegro

Belo Horizonte - April 9, 2014

Page 2: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

To my parents, Alfreu and Eunice.

Page 3: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

Acknowledgements

My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro.

I appreciate the encouragement and support, and by the beautiful theme proposed. The

discussions always so enlightening and instructive. I am grateful to my supervisor for his

patience and kindness to address several questions that have arisen, by having their time

even on Saturdays and other non-school days.

Members of the bank examiner who kindly agreed to participate in the completion of

this work, professors Ezequiel Rodrigues Barbosa, Marcos Montenegro, Emerson Jurandir

and Joao Marcos Bezerra do O.

To my parents, Alfreu and Eunice, by ample incentive to the “domain of knowledge”.

Even before entering the University, the great desire of my father was watching me with

the title of Doctor, “Dr. Aldo”,as always told me.

To my many colleagues in the UFMG who helped me in various ways.

To Professor Francisco Dutenhefner by the orientation in the MSc and for under-

standing and flexibilitye. To Susana C. Fornari for her patience and dedication during

my undergraduate research and encouraging academic research.

To the professor Sylvie Marie who without knowing it helped me get into the math

career. I do not forget a phrase she said.: “Many people choose to do engineering because

they like math. Why not choose mathematics since they like math?”

To many professors of UFMG math department who helped answering questions and

teaching courses I attended as a student. In this regard, I am grateful to professors like:

Gastao, Mario Jorge, Rogerio Mol, Marcelo T. Cunha, among several other.

To UFMG math department that helped me, as much as possible, in several events

that I participated . It is certainly well deserved the concept 6 by CAPES..

UNIFEI colleagues who helped in the reduction of my working hours for a few semesters.

To other colleagues, friends and professors who should be included here. However, do

not do a list with names so I could forget to mention someone.

Again, to all, my sincere thanks.And above all, God, Jehovah, the Almighty, Great

Scientist and Designer Creator of the universe and the wonderful human mind.

Page 4: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

Science cannot solve the ultimate mystery of nature. And that is because, in the last

analysis, we ourselves are part of nature and therefore part of the mystery that we are

trying to solve.

Max Planck

Page 5: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

Abstract

We approach potential elliptic systems involving Paneitz-Branson operators and crit-

ical nonlinearities. First, we present conditions for the existence of regular solutions of

potential systems in Riemannian Geometry, a decomposition in diagonal bubbles to ap-

plications of Palais-Smale and theoretical applications of this decomposition. Then, we

Euclidean space, we present another decomposition in bubbles and apply the decomposi-

tion in bubbles o a result of compactness. Finally, we apply all those results in extremal

applications for optimal Sobolev inequalities on compact manifolds.

Page 6: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

6

Contents

General Introduction 7

0.1 Historical Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

0.2 Proposal and Relevance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

0.3 Organization and Ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1 Preliminary Mathematical Material 17

1.1 Curvatures in a Riemannian manifold . . . . . . . . . . . . . . . . . . . . . 17

1.2 The Musical Isomorphism and Divergence of Tensors . . . . . . . . . . . . 21

1.3 Homogeneous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

1.4 Sobolev Spaces of Vector Valued Maps of Second Order . . . . . . . . . . . 26

1.5 Coercivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

1.6 The Scalar AB Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

1.6.1 Partial Answers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

1.7 AB Vector Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

1.7.1 Partial Answers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

2 Elliptic Systems of Fourth Order 45

2.1 Existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.2 Regularidade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.3 Bubbles Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

2.4 Pointwise Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

2.5 Concentracao L2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

2.6 Compacidade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

3 Sharp Sobolev Vetorial Inequality of Second Order 99

3.1 Extremal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

4 Final Considerations 103

4.1 Final Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

Page 7: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

7

General Introduction

0.1 Historical Overview

In 1983, Paneitz [27] introduced the fourth order operator P 4g : C4(M) → C0(M),

defined by

P 4g u := ∆2

gu− divg

((2

3Rgg − 2 Ricg

)(∇u)#

),

for all u ∈ C4(M), where (M, g) is a Riemannian manifold of dimension n = 4, Ricg is the

Ricci tensor Ricci, Rg is the scalar curvature, divg is the divergente and ∆g is the Laplace-

Beltrami operator with respect to the metric g. This operator P 4g has some properties of

conformal invariance. Accurately, if g = e2ϕg is conformal to the metric g, ϕ ∈ C∞(M),

then

P 4g = e−4ϕP 4

g .

Associated with this operator, we have the notion of Q-curvatura, a curvature which also

has conformal properties. For this case n = 4, a Q-curvature is given by

Q4g =

1

6

(∆gRg − 3|Ricg |2g +R2

g

).

Beyond this conformal invariance, the operator P 4g appears in the following relation be-

tween the curvatures Q4g and Q4

g:

P 4g ϕ+Q4

g = Q4ge

4ϕ .

It is noteworthy that the Q-curvature in the dimension n = 4, and for locally conformally

flat manifold, is inside the integral in the Gauss-Bonnet formula for Euler characteris-

tic, and thus has a very important role in the study of topology and geometry of the

Riemannian manifold of dimension 4. We have the following integral identity

4π2χ(M) =

∫M

(Qg +

1

8|Weylg |2

)dvg , (1)

where χ(M) is the Euler characteristic of the manifold M and Weylg denotes the Weyl

tensor with respect to the metric g. As the |Weylg |2dvg is an punctual conformal invariant,

we obtain that the integral of Q-curvatura∫Qg dvg is a conformal invariant. For more

details on this, see the articles by Chang [8] and Chang-Yang [7].

The generalization for the case n ≥ 5 was made by Branson [6] in 1987. Let (M, g) be a

Riemannian manifold of dimension n ≥ 5. We define the operator P ng : C4(M)→ C0(M)

by

Page 8: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

8

P ng u := ∆2

gu− divg

((anRgg + bnRicg) (∇u)#

)+n− 4

2Qngu ,

where

an =(n− 2)2 + 4

2(n− 1)(n− 2), bn = − 4

n− 2,

and

Qng =

1

2(n− 1)∆gRg +

n3 − 4n2 + 16n− 16

8(n− 1)2(n− 2)2Rg −

2

(n− 2)2|Ricg |2g

is the Q-curvatura for dimension n ≥ 5. We denote P ng also by Pg. This operator also has

conformal invariance properties. That is, considering u ∈ C∞(M), u > 0, and the metric

g = u4

n−4 g wich is conformal to the metric g, we have

P ng ϕ = u−

n+4n−4P n

g (uϕ), (2)

for all ϕ ∈ C∞(M). In particular, by taking ϕ ≡ 1, we get the following elliptic semilinear

diferencial equation satisfied by the conformal factor u:

P ng u =

n− 4

2Qngu

n+4n−4 , u > 0 .

It is interesting to compare this relationship between the Paneitz-Branson operator

and the Q-curvatura of the metric g with the scalar curvature Rg. Let’s see what we

mean. Initially, we have following

Lgu = Rgun+2n−2 ,

where Lg = 4n−1n−2

∆g +Rg; g = u4

n−2 g and n ≥ 3. We have also a equivalent identity in the

case n = 2. The operator Lg is a conformal operator. An important result in conformal

differential geometry is the resolution of the Yamabe problem , where the operator Lg

plays a key role. That is, we can always find a metric g, in the conformal class of g, such

that the scalar curvature Rg is constant, considering that the manifold is smooth, closed

and dimension n ≥ 2.

Because of similarities of the properties, in the conformal differential geometry, be-

tween the operators Lg e Pg, it is natural to ask whether the Q-curvatura has the same

property, that is, how it would be the Yamabe problem for Q-curvature. Partial results

have been established in response to this question for manifolds of dimensions ≥ 5 (see

[15], [22], [23], [28]).

But these results are limited by the lack of the maximum principle for differential

operators of higher orders. Because of this problem, we need positivity in the conformal

factor u which is determined by Paneitz-Branson equation and is not clear way to guar-

antee that this condition is satisfied more generally. This problem can be solved using

Page 9: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

9

the conformal covariance property of the Paneitz-Branson operator, that is, by (2). This

property tells us that uv is a solution for the Paneitz-Branson equation in the geometry of

the metric g if u is a solution in the geometry of g. This condition plays an important role

in the type of functions which can be a local minimum of a functional which is naturally

associated with the Yamabe problem for the Q-curvature.

There is another problem, an analytical problem, when we consider the Yamabe prob-

lem for the Q-curvature. As in the case of the classical Yamabe problem, we have problems

when we try to use variational methods to find solutions of Paneitz-Branson equation be-

cause of the exponent of the nonlinearity, n+4n−4

, where n is the dimension of the manifold,

which is the critical Sobolev exponent W 2,2(M) less one, because the immersion W 2,2(M)

in L2nn−4 is not compact.

One idea contained in [10], by Hebey-Djadli-Ledoux, is that the Sobolev inequalities

of second order can be used to deal with the problem of concentration in approximating

sequences for the solution, since the infimum of the functional associated with the problem,

the Paneitz-Branson functional, is smaller than a critical value. In this method, it is not

clear when the infimum is positive and the Yamabe constant is less than or equal to the

Yamabe constant of the sphere.

A partial solution to this problem was made by F. Robert and P. Esposito in 2002,

see [15]. It has been shown that if n ≥ 8 and the manifold is locally conformally flat, then

there exists a minimizer for the Paneitz-Branson functional.

The effect of this result is that the part of existence of the Yamabe problem for the

Q-curvature is brought to a point analogous to that Aubin led the Yamabe problem for

the scalar curvature. But, is not yet clear when the Green function of the Paneitz Branson

operator is positive, in case the operator be coercive. This excludes try to use the methods

of Schoen to complete the problem.

It was shown by D. Raske (see [29]) that there exists a metric in the conformal class

of the arbitrary metric in a smooth closed Riemannian manifold, of dimension n ≥ 5,

such that the Q-curvature of the metric is constant. Existence of solutions is obtained

through the combination of variational methods, Sobolev inequalities of second order and

the blow-up theory of W 2,2(M). Below is the result.

We define the Paneitz Branson constant as

λg(M) := infw∈C∞+ (M)

∫MwPgw dvg

‖w‖22nn−4

,

where Pg is the Paneitz-Bransonoperator. Let λ(Sn) be the Paneitz-Branson constant of

the unit sphere with the canonical metric. We have the following result (see [29]):

Theorem 1 (David Raske, 2011). Let (M, g) be a Riemannian manifold of dimension

n ≥ 5. Suppose that at least one of the following conditions is valid:

i. The Paneitz-Branson constant is less than λ(Sn);

Page 10: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

10

ii. The Yamabe constant of g is greater than or equal to the inverse of the Yamabe

constant of the n-sphere;

iii. n ≥ 8 and g is not locally conformally flat.

then exists a smooth minimizer, positive of Paneitz-Branson functional and there exists a

metric h in the conformal class of g such that Qh = λ, where λ is the Paneitz-Branson

constant of g.

thus, the study of operators like the Paneitz-Branson is very important in the analysis of

geometrical problems as the problem of the prescribed Q-curvature, where the Yamabe

problem for the Q-curvature is a particular case.

Consider the system of equations

−∆2gui + divg

(Ai(∇ui)#

)+

k∑j=1

Aij(x)uj = u2#−1i , (3)

where 2# = 2nn−4

, U = (u1, . . . , uk), A = (Aij) is a continuous map of M to M sk(R) such

that A(x) is positive definit for all x ∈ M , M sk(R) is the space of real symmetric matrices

k × k, and the Ai are symmetric tensor fields of type (2,0). In this context, we consider

ui > 0 for all i. This system can be seen as a natural generalization of the equations

involving operators of type Paneitz Branson. Thus, considering the scalar case, that is

k = 1, and the tensor Ai = f · g, where f is a smooth function, the system (3) can be

rewritten as

−∆2gu+ bα∆gu+ cαu = u2#−1 . (4)

Assume that the constants bα and cα are converging sequences of real positive numbers,

satisfying cα ≤ b2α4

. From 2000 to 2004, many authors studied the case (4) above, for

example F. Robert, E. Hebey, Z. Djadli and M. Ledoux in [10, 15, 22] and [24]. Hebey-

Robert-Wen in [24] discussed the compactness of solutions of (4), precisely when bα and

cα converge respectively to b0 and c0 and the solutions uα converge weakly in H2,2(M).

They found conditions such that the limit uα is nontrivial.

The following theorems have been proved by Hebey-Robert-Wen in 2004 (see [24]).

Let

Ag =(n− 2)2 + 4

2(n− 1)(n− 2)Rg g −

4

n− 2Ricg (5)

be field of (2, 0)-tensor. We denote by λi(Ag)x, 1, ..., k, the g-eigenvalues of Ag(x) and

define λ1 as infimum of the i and x of the λi(Ag)x and λ2 as the supremum of the i and

x of the λi(Ag)x. We denote por Sc the critical set defined by

Sc = λ ∈ R : λ1 ≤ λ ≤ λ2 . (6)

Page 11: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

11

We have the following results:

Theorem 2 (Hebey-Robert-Wen, 2004). Let (M, g) be a compact manifold locally confor-

mally flat of dimension n and (bα)α, (cα)α converging sequences of positive real numbers

with positive limits b∞ e c∞ such that cα ≤ b2α4

for all α. We consider equations of type

∆2gu+ bα∆gu+ cαu = u2#−1 , (7)

and we assume that b∞ /∈ Sc, where b∞ is the limit of bα and Sc is the critical set given

by (6). Then the family (7) is pseudo-compact when n ≥ 6 and compact when n ≥ 9.

We say that a family of equations are solutions of (7) is pseudo-compact if, for any

sequences (uα) in H2,2(M) of positive solutions that converges weakly in H2,2(M), the

weak limit u0 of uα is nonzero.

The following theorem is a complement of the above theorem when the dimension is

n = 6, 7 or 8 and b∞ is is below the lower limit λ1 de Sc.

Theorem 3 (Hebey-Robert-Wen, 2004). Let (M, g) be a compact manifold locally con-

formally flat of dimension n = 6, 7 or 8 and (bα)α, (cα)α converging sequences of real

positive numbers with positive limit b∞ and c∞ such that cα ≤ b2α4

for all α. We consider

equations like

∆2gu+ bα∆gu+ cαu = u2#−1 ,

and we assume that b∞ < λ1 = min Sc, where b∞ is the limit of bα and Sc is the critical

set given by (6). Then the family (7) is compact.

Pseudo-compactness has a traditional interest because it seeks nontrivial solutions of

limit equation we obtain from (7) making α→ +∞.

On the other hand, we say that the family of equations (7) is compact if any sequence

(uα)α in H2,2(M) of positive solutions of (7) is limited in C4,θ(M), 0 < θ < 1 and then

converges, if necessary take a subsequence, on C4(M) for some function u0.

Compactness is a concept clearly stronger than pseudo-compactness.

Pseudo-compactness for elliptic equations of second order the type of Yamabe has

been widely studied. Compactness for Yamabe equations of second order was studied by

Schoen from 1988 to 1991 (see [32, 33, 34, 35]).

As application of the results of compactness, we can study also the existence of ex-

tremal in sharp Sobolev scalar inequalities of second order.

From the continuous immersion H2,2(M) → L2#(M), then there are constants A,B >

0 such that:

‖u‖22# ≤ A‖∆gu‖2

2 +B‖u‖2H1,2(M) , ∀ u ∈ H2,2(M). (8)

We are interested in the sharp constants A,B from the above inequality. More pre-

cisely, the best constant A is defined by

Page 12: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

12

A0 = inf

A: exists B such that the above inequality (8) is valid ∀ u ∈ H2,2(M). (9)

A natural question that arises is: the infimum A0 is achieved? The answer is yes, the

infimum A0 is achieved in (8). In 2000, Djadli-Hebey-Ledoux [10] proved the result with

the restriction that the metric g is conformally flat. Later, in 2003, Emmanuel Hebey

proved the result for any Riemannian manifold in [25] (see also [26]).

Follow from the Sobolev theorem that the constant A0 is well defined. This constant

was calculated by Lieb [25], Lions [26], Edmunds-Fortunato-Jannelli [14] and Swason [37].

Precisely, we have:

1

A0

=n(n2 − 4)(n− 4)w

4nn

16,

where wn is the volume of the sphere Sn in Rn+1.

The infimum in (9) is achieved and there exists a constant B > 0 such that:

(∫M

|u|2#dvg) 2

2#

≤ A0

∫M

(∆gu)2 dvg +B

∫M

(|∇u|2g + u2

)dvg , (10)

for all u ∈ H2,2(M).

The second best Sobolev constant associated with (8) is defined by:

B0 = inf B ∈ R; (10) is valid . (11)

The second Sobolev Riemannian inequality states that, for any u ∈ H2,2(M), we have

(∫M

|u|2# dvg) 2

2#

≤ A0

∫M

(∆gu)2 dvg +B0

∫M

(|∇gu|2 + |u|2

)dvg . (12)

A nonzero function u0 ∈ H2,2(M) is said extremal for the inequality (12) if

(∫M

|u0|2#

dvg

) 2

2#

= A0

∫M

(∆gu0)2 dvg +B0

∫M

(|∇gu0|2 + |u0|2

)dvg .

You can find some comments about the second best constant in [10] (Djadli, Hebey,

Ledoux, 2000). But for the extremal of the sharp inequalities there are studies only for

the case of first order. To this reasoning, see [11] also [5, 4].

All has been said above is being considered for the case k = 1. The main objective

of this thesis is to extend some of these results for the case k ≥ 2 and apply them to the

study of existence of extremal in sharp Sobolev vector inequalities of second order. We’ll

talk more about this in the next section.

Page 13: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

13

0.2 Proposal and Relevance

One of the main goals of this thesis is extend the results on compactness of solutions of

equations involving Paneitz-Branson type operators for systems and apply these results

to obtain results of existence of extremal for a class of sharp Sobolev inequalities of

second order. This is an important question, both the mathematical point of view, for

involve a larger structure and his understanding, as from the point of view of analytical

applications, by allowing the study of several elliptic PDE systems of second order on

Riemannian manifold. Let me be a bit more clear..

Let (M, g) be a compact Riemannian manifold, of dimension n ≥ 5, whose volume

element is dvg. We consider here maps U ∈ H2,2k (M) that are solutions the following

model equation:

−∆2gU + divg

(A(∇U)#

)+∇UG(x, U) = ∇F (U) . (13)

That is,

−∆2gui + divg

(Ai(∇ui)#

)+ ∂iG(x, U) = ∂iF (U) , (14)

for each i = 1, ..., k, where 2# = 2nn−4

is the Sobolev critical exponent for the immersions of

H2,2(M) in Lp(M) spaces. Note that the operator Pg := −∆2g+divg

(Ai(∇·)#

)+∂iG(x, ·)

is Paneitz-Branson type. Here arise some questions, such as:

• There is a nonzero solution U for (13)?

• If there is a nonzero solution U , which regularity we can get??

• The set of the solutions of (13) is compact in some topology?

We will answer these questions in the following chapters for a class of homogeneous

functions F and G.

Answering the above questions, we continue with the study of existence of extremal

in sharp Sobolev inequalities of second order in the vector case. This study, for the

case of second order, is also motivated by the study of the best constants for first order

inequalities in the vector case which was done by E. Barbosa e M. Montenegro (see [5]).

E. Barbosa e M. Montenegro studied inequalities like:

(∫M

F (U) dvg

) pp∗

≤ A0(p, F,G, g)

∫M

|∇gU |p dvg + B0(p, F,G, g)

∫M

G(x, U) dvg , (15)

where A0 and B0 are the best constants, 1 ≤ p < n, U = (u1, · · · , uk), F : Rk → R is a

positive continuous function, and p∗-homogeneous and G : M × Rk → R is a continuous

function, positive and p-homogeneous in the second variable. The vector theory of best

Page 14: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

14

constants developed (in [5]) considers a number of questions involving the constants A0

and B0. Some follow directly from scalar theory, others are more complex, for example,

the behavior of B0 for all the parameters involved and problem of existence and C0

compactness of the extremal. A map U is extremal when it achieves equality in (15).

When F and G are C1 maps a extremal is weak solution of the system:

−A0(p, F,G, g)∆p,gui +1

pB0(p, F,G, g)

∂G(x, U)

∂ti=

1

p∗∂F (U)

∂ti, i = 1, · · · , k ,

where ∆p,gu = divg(|∇gu|p−2∇gu) denotes the p-Laplacian operator associated with the

metric g.

Now consider Uα ∈ H1,p(M,Rk) weak solutions of the system:

−∆p,gU +1

p∇UGα(x,U) =

1

p∗∇Fα(U) em M ,

where Fα : Rk → R are positive C1 functions and p∗-homogeneousand the Gα : M×Rk →R are C1 functions and p-homogeneous in the second variable. Consider sequences of

limited solutions (Uα)α whose limite is U ≡ 0, taking a subsequence if it is necessary, we

have the following bubbles decomposition:

Uα =l∑

j=1

Bj,α +Rα ,

for all α > 0, where (Bj,α)α, j = 1, · · · , l, are k-bubbles and (Rα)α ⊂ H1,p(M,Rk) is

such that Rα → 0 in H1,p(M,Rk) when α → +∞. With this result we can add more

properties to the sequences of limited solutions such that Uα 0 in H1,p(M,Rk). We have

also pointwise estimates or C0 for (Uα)α (see [9]). The blow-up points of the sequences

(Uα)α have much of the information of the sequences, this is the Lp concentration property.

Concentration properties were studied by Druet, Hebey and Robert (see [17] and [12])

and extensions of these works were made by G. Souza [9].

In this work, we follow similar ideas of E. Barbosa and M. Montenegro in [5]. We

consider the sharp Sobolev inequalities of second order:

(∫M

F (U)dvg

) 2

2#

≤ A0

∫M

(∆gU)2 dvg +

+ B0

∫M

((A((∇gU)#, (∇gU)#

)+G(x, U)

)dvg , (16)

where F is a 2#-homogeneous map and G is 2-homogeneous on the second variable, and

also the systems of second order are associated to this inequality:

Page 15: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

15

−∆2gU + divg

(A(∇U)#

)+

1

2∇UG(x, U) =

1

2#∇F (U) . (17)

We consider sequences of limited solutions (Uα)α whose limite is U ≡ 0. If necessary take

a subsequence, we obtain the following bubbles decomposition :

Uα =l∑

j=1

Bj,α +Rα ,

for all α > 0, where (Bj,α)α, j = 1, · · · , l, k-bubbles and (Rα)α ⊂ H2,2(M,Rk) is such

that Rα → 0 in H2,2(M,Rk) when α → +∞. Added to this result, we have pointwise

estimates or C0 for (Uα)α. The blow-up points or concentration of the sequences (Uα)α

have a lot of information about the sequences and generating L2 concentration properties.

We apply all results for the study of existence of extremal for the inequalities (16).

0.3 Organization and Ideas

This work has three chapters. In Chapter 1, we show some definitions and some basic

results that will be used in the others chapters. In special, we show the Sobolev spaces

of second order and we make fazemos um panorama detalhado da scalar theory de best

constants on sharp Sobolev inequalities of second ordem e destacamos alguns problemas

em aberto. Morover, we describe some important problems of the vector theory of best

constants and we state our main contributions. We include in this chapter, some basic

results about Euclidean and Riemannian Sobolev vector inequalities. Precisely, we show

that the best constant associated to the inequality

(∫RnF (U) dx

) 2

2#

≤ A∫Rn

(∆U)2 dx

is given by

A0(F, n) = M2

2#

F A0(n)

where MF = maxSk−12

and Sk−12 =

t ∈ Rk;

∑ki=1 |ti|2 = 1

. Morover, characterized the

associated extremal as the type of U0 = u0t0, where u0 is a extremal function associated

to the Euclidean Lp-Sobolev scalar inequality and t0 is a maximum point of F in Sk−12 .

Chapter 2 has the largest number of contributions made. Turn our attentionto the

following system:

−∆2gU + divg

(A(∇U)#

)+∇UGα(x, U) = ∇Fα(U) (18)

In this chapter, we study the blow-up decomposition of the solutions (Uα)α of (18).

We put conditions for existence of minimal energy solutions. Then we study the behavior

Page 16: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

16

of sequences (Uα)α, given by the solutions of systems (18), as function of the behavior of

Fα and Gα.

With the hypothesis that the (Uα)α is limited, we obtain that these sequences is a

Palais-Smale sequences of the functional J associated to (18). That is, the sequences Uα

is such that:

(J(Uα))α is limited and DJ(Uα)→ 0 in(H2,2k (M)

)′.

From the limitation of Uα we get the existence of a weak limit U0 in H2,2k (M). Working

with the case where U0 ≡ 0, the hypotheses about Fα and Gα let us decompose the

functional J in terms of others functional Lik such that the components of Uα form Palais-

Smale sequences for the functional Lik. This decomposition in functional Lik allow us to

use the results obtained by F. Robert (see [30]) to obtain bubbles decomposition for the

components of Uα, which concludes this part.

Using bubbles decomposition we have the pointwise estimates for a sequences of solu-

tions of (18) that have 0 as a weak limit.

Then, we have the L2 concentration, where we extended for the vector case some

results. In some of these, we assume that G(x, U) =∑k

i,j=1Aij(x)ui(x)uj(x). We use some

estimates and the Bochner-Linerowicz-Weitzenbook formula for the main L2 concentration

result.

In the main part of chapter 2, we proof the compactness. For this, we use several

previous results, as bubbles decomposition and L2 concentration.We also use an important

and useful tool that is the Pohozaev type identity. Our contributions at this point extend

some results ound in studies of E. Hebey, F. Robert, Z. Djadli, M. Ledoux and V. Felli,

made in the scalar case (fourth order equations), for the vector case (systems of fourth

order).

In the chapter 3, we study the existence of etremal maps. The main ideia here is the

following. Suppose that the second order vector Sobolev sharp inequality has no extremal

map. We consider then a numerical sequence (α) such that 0 < α < B0 and α → B0.

Associated to this sequence, we construct functions Fα and Gα and also solutions Uα of

systems of type studied on the previous chapters. The sequence (Uα) converges for U

such that will be a extremal map.

Page 17: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

17

Chapter

1Preliminary Mathematical Material

In this chapter, introduce some basic notations and definitions we will use throughout

the remainder of this paper. We remember some basic facts of Riemannian geometry.

Then, there is the Sobolev space theory in Riemannian manifolds. In this section and

in the rest of this paper, we assume that the manifold is compact. We show the norms

that we use, and some properties of these Sobolev spaces.

Some definitions are more usual. But present below the definitions which will be useful

in the following work.

Over the years, about forty years, much attention has been given to the sharp Sobolev

Riemannian inequalities. There is a vast literature with a wide theory of the best constants

that is connected with areas such as analysis, geometry and topology. These inequalities

play an important role in geometric analysis, especially in the study of the existence

and multiplicity of solutions to the Yamabe problem (see [2], [20], [34]), Riemannian

isoperimetric inequalities (see [13]), among other applications. Most of these results shows

the influence of geometry and topology.

Some efforts were made in the study of shapr Sobolev Riemannian inequalities in some

decades. Part of the obtained results is kown as the AB program. Sharp inequalities and

first order Sobolev best constants in the scalar case were studied by Aubin, Druet, Hebey,

Vaugon, among others (see [3], [13], [19], [21]). But sharp inequalities and best Sobolev

constants in the vector case were studied by Hebey, E. R. Barbosa and M. S. Montenegro

(see [17], [18] and [5], [4]). For sharp Sobolev inequalities of second order in the scalar case,

the paper made by Djadli-Hebey-Ledoux show a introduction to study the first constant.

In this chapter, we show the AB program for scalar sharp Sobolev inequalities, and

after in the vector case, of second order. We show here some contributions, that is, we

answer some of the questions of AB program of the inequalities of second order.

1.1 Curvatures in a Riemannian manifold

Let (M, g) be a compact Riemannian manifold of dimension n. Let x ∈ M and X, Y, Z ∈TxM . We define:

Page 18: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

18

R(X, Y )(x)Z = ∇X(x)(∇Y (x)Z)−∇Y (x)(∇X(x)Z)−∇[X,Y ](x)Z ,

where X, Y , Z are vector fields on M such that X(x) = X, Y (x) = Y, Z(x) = Z. This

definition is independant of the choice of the extensions X, Y , Z. Given x ∈ M and

X, Y, Z ∈ TxM and η ∈ (TxM)∗, we define the curvature tensor as follows:

R(x)(X, Y, Z, η) = η(R(Y, Z)(x) .X) = 〈R(X, Y )Z,W 〉 .

The function R is a (3,1)-tensor. The coordinates of R in a chart are given by:

R(x)lijk =

(∂Γlki∂xj

)x

(∂Γlji∂xk

)x

+ Γljα(x)Γαki(x)− Γlkα(x)Γαji(x) , (1.1)

where Γkij are the Christoffel symbols.

The Riemman tensor is a (4,0)-tensor and the coordinates in a chart are Rijkl :=

giαRαjkl.

The curvature operator R in x ∈ M , R : Λ2x → Λ2

x is defined by the relation

〈R(X ∧ Y ),W ∧ Z〉 = 〈R(X, Y )Z,W 〉 ,

where Λ2x is the space generated by the 2-forms X ∧ Y and

(X ∧ Y ) (Z,W ) = 〈Y, Z〉〈X,W 〉 − 〈X,Z〉〈Y,W 〉 .

Note that, if Xini=1 is a orthonormal basis of TpM , then Xi ∧ Yii<j is a orthonormal

basis of Λ2x. Thus, the symmetric bilinear function R is well defined.

Now let σ ⊂ TpM be a bidimensional space of TpM and X, Y a basis for σ. The

sectional curvature of σ in x is given by:

K(σ) = K (X, Y ) =R(X, Y, Y,X)

‖X ∧ Y ‖2,

where ‖X ∧ Y ‖2 = ‖X‖2‖Y ‖2 − 〈X, Y 〉2.

The Ricci tensor which we denote by Ricg or Ric is a symmetric (2,0)-tensor defined

as the following bilinear function:

Ric : TM × TM → R ,

that associates to each pair of fields (X, Y ) the trace of the function Z 7→ R(X,Z)Y .

Then we have:

Ric (X, Y ) =n∑i=1

〈R(Xi, X)Y,Xi〉 ,

where Xini=1 is a orthonormal basis of TpM .

Page 19: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

19

The Ricci curvature in the direction of X ∈ TM , with |X| = 1, is defined as

Ric (X) = Ric (X,X) .

If X ∈ TM is unitary and X(p) = v, p ∈ M and v ∈ TpM , then we write Ricp(v)

instead of Ric(X,X). Let e1, · · · , en be a orthonormal basis with v = ei for some i.

Then:

Ricp(v) =∑j 6=i

〈R(v, ei)v, ei〉

=∑j 6=i

K(v, ei)

The scalar curvature we denote by Rg or Scalg is the trace of the Ricci tensor Rg :=

gijRij, where Rij are the components of Ricg in a chart.

The Riemann curvature Rmg is defined by

Rmg(x)(X, Y, Z, T ) = g(x) (X,R(Z, T )Y ) .

The components of Rmg in a chart are given by Rijkl = giαRαjkl where Rα

jkl are the

components of the curvature R as described above in (1.1).

Now let (M, g) be a Riemannian manifold of dimension n ≥ 3. The Weyl curvature

of g, denoted by Weylg, is a field of tensor of class C∞ four times covariant on M . Such

curvature is defined by:

Weylg = Rmg−1

n− 2Ricgg +

Rg

2(n− 1)(n− 2)g g ,

where is the Kulkarmi-Nomizu product, define below. For all x ∈ M we have:

Weylg(x) = Rmg(x)− 1

n− 2Ricg(x) g(x) +

Rg(x)

2(n− 1)(n− 2)g(x) g(x) .

The components Wijkl of Weylg are given by the relation:

Wijkl = Rijkl −1

n− 2(Rikgjl +Rjlgik −Rilgjk −Rjkgil)

+Rg

(n− 1)(n− 2)(gikgjl − gilgjk)

Let E be a vector space and h, k two tensores twice covariant and symmetric on E,

that is, for all X and Y in E:

Page 20: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

20

h(X, Y ) = h(Y,X) and k(X, Y ) = k(Y,X) .

We define the Kulkarni-Nomizu product of h and k, we denote by h k, as the four

times covariant tensor in E, defined for all X, Y, Z, T ∈ E by

h k(X, Y, Z, T ) = h(X,Z)k(Y, T ) + h(Y, T )k(X,Z)− h(X,T )k(Y, Z)− h(Y, Z)k(X,T ) .

We see that the Kulkarmi-Nomizu product is symmetric in the sense that, for all h

and k, we have h k = k h. We also see that the product is distributive with respect

to addition, in that, for all h, k1 and k2 we have h (k1 + k2) = h k1 + h k2.

A Riemannian manifold is locally conformally flat if, in each point of M there exists

a neighborhood conformally equivalent to the Rm. That is, if x ∈ M , there exists a

neighborhood U of x and a function u class C∞, that is, u : U → R, such that a metric

(local) g = e2ug is flat in U .

If a manifold (M, g) has dimension n ≥ 4, then (M, g) is conformally flat if and only

if, Weylg ≡ 0.

Page 21: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

21

1.2 The Musical Isomorphism and Divergence of Tensors

Consider x ∈M . # is the musical isomorphism between TxM and (TxM)∗, defined as:

# : TxM −→ (TxM)∗

X 7−→

(TxM) −→ R

Y 7−→ 〈X, Y 〉g(x) ,

This isomorphism is the identification of a Euclidean space with its dual. We denote:

X# the image of X via # and η# the image of η ∈ (TxM)∗ via the inverse of #. This

definition extends naturally vector field (that is, (0,1)-tensors) and to (1,0)-tensors. If X

is vector field and η is a (1, 0)−tensor, the coordinates of their images in a chart are:

Xi := (X#)i = gijXj

ηi := (η#)i = gijηj ,

that is an expression that is independent of the chart. Clearly, we have (X#)# = X e

(η#)# = η.

In what follows, in this work, A is defined as:

A((X)#, (X)#

)=

k∑i=1

Ai((Xi)

#, (Xi)#), (1.2)

where Ai, i = 1, ..., k, are positive and symmetric tensores, that is, A is a sum of continuous

symmetric (2, 0)−tensor. We have X = (X1, . . . , Xk) is such that each Xi, i = 1, . . . , k

is a (1,0)-tensor. In what follows, for simplicity, we say that A is a smooth symmetric

(2,0)-tensor when we refer the sum (1.2) above. Here Al(Xl)# is a (1, 0)-tensor whose

local coordinates are

(A(Xl)

#)i

= Aij((Xl)

#)j

= Aijgik(Xl)k

As the manifold M is compact and A is continuous, there exists a constant C > 0

such that: ∣∣∣∣∫M

A((X)#, (X)#

)dvg

∣∣∣∣ ≤ C

∫M

|X|2g dvg ,

for all X, where

Page 22: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

22

∫M

A((X)#, (X)#

)dvg =

k∑i=1

∫M

Ai((Xi)

#, (Xi)#)dvg .

We have that the manifoldM is compact andA is a symmetric and positive (2, 0)−tensor,

then there are positive constants c and C, such that:

c|X|2g ≤ Ai((X)#, (X)#

)≤ C|X|2g , (1.3)

for X a (1,0)-tensor.

Let X be a smooth vector field in M . The divergence of X is a smooth function in M

given by:

divg X : M → R

p 7→ (divg X)(p) = tr v 7→ (∇vX)(p) ,

where v ∈ TpM and tr is the trace of the linear operator. To define on charts, consider

X a (1,0)-tensor in M . The divergence is defined as

divg(X) = gij(∇X)ij = gij(∂iXj − ΓkijXk

).

This expression is independent of the chart.

Throughout this work we use sometimes the following useful theorem.

Theorem 4 (Divergence theorem). Let (M, g) be a compact Riemannian manifold without

boundary. Let η be a smooth (1,0)-tensor. Then we have:∫M

divg(η) dvg = 0 .

in particular, given u, v ∈ C∞(M), we have∫M

u∆gv dvg =

∫M

〈∇u,∇v〉g dvg∫M

(∆gu) v dvg .

where 〈·, ·〉 is the scalar product associated with g for 1-form.

Page 23: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

23

1.3 Homogeneous Functions

Consider G : M × Rk → R a continuous function and 2-homogeneous in the second

variable class C1 in the second variable. For example:

G(x, t) =k∑

i,j=1

Aij(x)titj ,

where t = (t1, . . . , tk) ∈ Rk and A = (Aij)M → M sk(R) is continuous, such that (Aij(x))

is positive defined, for all x ∈ M , and M sk(R) is the space of real symmetric matrices

k × k.

Let F : Rk → R be a positive function class C1 and 2#-homogeneous1, where 2# = 2nn−4

.

For example:

F (t) =k∑i=1

|ti|2#

,

where t = (t1, . . . , tk). In this case,

∂iF (t) = |ti|2#−2ti .

Now let F : Rk → R be a continuous function and q-homogeneous. Consider the

unitary sphere in the norm p, that is:

∂Bp [0, 1] =t ∈ Rk; |t|p = 1

,

where |t|p = (|t1|p + · · ·+ |tk|p)1p being t = (t1, . . . , tk). As the set ∂Bp [0, 1] is compact,

then there are constants mF,p > 0 e MF,p > 0 such that:

mF,p ≤ F (t) ≤MF,p ∀ t ∈ ∂Bp [0, 1] .

Now consider any t ∈ Rk \ 0. Then, by the q-homogeneity of F , we have

F

(t

|t|p

)=

1

|t|qpF (t) ,

for all t ∈ Rk \ 0. Thus,

mF,p|t|qp ≤ F (t) ≤MF,p|t|qp ∀ t ∈ Rk . (1.4)

As the space Rk has finite dimension, then all norms are equivalent. Therefore, there

exists a constant c > 0 such that

1F (λt) = λ2#

F (t), ∀ t ∈ Rk and λ > 0

Page 24: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

24

|t|p ≤ c|t|q ∀ t ∈ Rk .

then

F (t) ≤M ′F,p|t|qq ∀ t ∈ Rk ,

where M ′F,p = cqMF,p is a positive constant.

In this work, in most situations, we use that F is a 2#-homogeneous function. There-

fore, this case will simplify the notation where F is a 2#-homogeneous function and the

norm is Euclidean: mF,2 = mF e MF,2 = MF . Then

mF |t|2#

2 ≤ F (t) ≤MF |t|2#

2 ∀ t ∈ Rk .

Similarly, if G : M × Rk → R is a q-homogeneous function in the second variable, we

have:

mG,p|t|qp ≤ G(x, t) ≤MG,p|t|qp ∀ t ∈ Rk .

In this work, we use, unless the contrary that G is a 2-homogeneous function on

the second variable. Therefore, this case will simplify the notation where G is a 2-

homogeneous function on the second variable and the norm is Euclidean: mG,2 = mG and

MG,2 = MG. Thereby, we have:

mG|t|22 ≤ G(x, t) ≤MG|t|22 ∀ t ∈ Rk . (1.5)

Let F : Rk → R be a C1 function q-homogeneous. In this work, we use the following

Euler identity:

k∑i=1

∂iF (t)ti = q F (t) .

Part of the vector theory of best constants does not follow directly from scalar theory.

One of the differences is in relation to the nature of vector functions satisfying the condi-

tions of homogeneity. In the case (k ≥ 2), there are examples of homogeneous functions

they are just continuous. Note the following example. Consider

F (t) = |t|2#µ e G(x, t) = β(x)|t|2µ ,

where | · |µ is a µ-norm defined by |t|µ =(∑k

i=1 |ti|µ) 1µ

for 1 ≤ µ < ∞ e |t|∞ =

max |t|i; i = 1, . . . , k.The absence of regularity of F and G creates an obstacle for the study of several

questions, since the arguments are based on Euler equations satisfied by critical points of

functional. That is, the approach in what follows would be restricted without assuming

Page 25: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

25

regularity of F and G.

Page 26: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

26

1.4 Sobolev Spaces of Vector Valued Maps of Second Order

We consider (M, g) a compact Riemannian manifold and dvg = dv(g) the Riemannian

measure associated with the metric g. Given a function u : M → R class C∞(M) and k

a integer, we denote ∇ku the k-nth covariant derivative of u and |∇ku| the norm of ∇ku,

defined by

|∇ku| = gi1j1 · · · gikjk(∇ku

)i1···ik

(∇ku

)j1···jk

,

where(∇ku

)i1···ik

denotes the components of ∇u in a chart. However this definition does

not depend on the choice of the chart.

If (M, g) is a compact Riemannian manifold, the Sobolev space H2,2(M) is, by defini-

tion, the completion of C∞(M) em L2(M) by the norm

‖u‖′H2,2(M) =2∑j=0

(∫M

|∇ju|2g dvg) 1

2

,

where dvg = dv(g) is Riemannian measure associated with the metric g.

The space H2,2(M) is a Hilbert space, with norm:

‖u‖2H2,2(M) =

∫M

|u|2 dvg +

∫M

|∇u|2 dvg +

∫M

(∆gu)2 dvg

Note that the norms ‖·‖′H2,2(M) and ‖·‖H2,2(M) are equivalent. Indeed, by the Bochner-

Lichnerowitz-Weitzenbock formula, we have that∫M

(∆gu)2 dvg =

∫M

|∇2u|2g dvg +

∫M

Ricg

((∇u)#, (∇u)#

)dvg

for all u ∈ H2,2(M). then, we have

‖∇2u‖22 + ‖∇u‖2

2 + ‖u‖22

= ‖∆gu‖22 −

∫M

Ricg

((∇u)#, (∇u)#

)dvg + ‖∇u‖2

2 + ‖u‖22

≤ |∆gu|22 + C‖∇u‖22 + ‖u‖2

2

for all u ∈ H2,2(M). But, hence we have that there exists a positive constant C such that

‖ · ‖′H2,2(M) ≤ C‖ · ‖H2,2(M). On the other hand, note that, for all function u ∈ C∞(M):

(∆gu)2 ≤ |∇2u|2

thus, as the two norms of H2,2(M) are equivalent.

Page 27: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

27

The associate scalar product is defined by:

〈u, v〉 =

∫M

〈u, v〉g dvg +

∫M

〈∇u,∇v〉g dvg +

∫M

〈∇2u,∇2v〉g dvg

=2∑j=0

∫M

〈∇ju,∇jv〉g dvg .

As M is compact, the space H2,2(M) does not depend on a metric Riemannian. If M

is a compact manifold with two metrics g and g, then exist a real number C > 1 such

that, in any point of M :

1

Cg ≤ g ≤ Cg .

These two inequalities we reagrd as inequalities between bilinear forms.

A linear operator T : E → F between two Banach spaces (generally, we use E =

H2,2(M) and F = Rk, with k ∈ N) is compact if, for any sequences (un)n∈N ∈ E

uniformly bounded in the norm of E, then exists u ∈ F and a subsequence (unk) such

that:

limn→∞

T (unk) = u ,

strongly in F .

The space H2,2(M) is reflexive. Thus, all bounded sequence in H2,2(M) have weakly

convergent subsequences. And, if T : H2,2(M) → Rk is compact, then T takes limited

sequences in H2,2(M) in sequences that have convergent subsequences in Rk, for all k ∈ N.

The space H2,2(M) it is also a separable space. Therefore, every bouded sequence in

(H2,2(M))∗

has convergent subsequences (in the weak∗ topology).

Consider (Tn)n∈N ∈ (H2,2(M))∗

and T ∈ H2,2(M). We say that (Tn) converges weakly

for T if

limn→∞

Tn(u) = T (u) for all u ∈(H2,2(M)

)∗,

or

Tn T em(H2,2(M)

)∗,

where n→∞. (H2,2(M))∗

is the space of continuous linear forms of H2,2(M).

If M is a Riemannian manifold of dimension n ≥ 5, then the immersion

H2,2(M) → Lq(M) ,

is compact for q ∈(

1, 2nn−4

]. The immersion

Page 28: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

28

H2,2(M) L2nn−4 (M) ,

is continuous, but not compact. That is, there exists a constant C > 0 such that

‖u‖L

2nn−4 (M)

≤ C‖u‖H2,2(M) .

We denote by H2,2k (M) = H2,2

(M,Rk

)the vector Sobolev space H2,2(M) × · · · ×

H2,2(M), that is

H2,2k (M) =

U = (u1, · · · , uk);ui ∈ H2,2(M) for all i = 1, · · · , k

.

This space has the norm:

||U||H2,2k (M) =

(∫M

(∆gU)2 dvg +

∫M

(|∇gU|2 + |U|2

)dvg

) 12

,

where U = (u1, . . . , uk) and,

∫M

|U|22dvg =k∑i=1

∫M

|ui|2 dvg,∫M

|∇gU|22dvg =k∑i=1

∫M

|∇gui|2dvg

∫M

(∆gU)2 dvg =k∑i=1

∫M

(∆gui)2 dvg .

Also:

|U|p = (|u1|p + · · ·+ |uk|p)1p .

To simplify, we use that:

|U| = |U|2 .

The vector-valued Sobolev space H2,2k (M) has different properties due to the properties

of the space H2,2(M). The space H2,2k (M) is a Hilbert space and has the following scalar

product:

〈U ,V〉H2,2k (M) =

k∑i=1

〈ui, vi〉 ,

where 〈·, ·〉 is the usual scalar product in H2,2(M) and U = (u1, · · · , uk) and V =

(v1, · · · , vk).The space H2,2

k (M) is separable and reflexive. Thus, the unit ball of H2,2k (M) is weakly

compact. In other words, for any sequences (Un)n∈N ∈ H2,2k (M) such that

Page 29: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

29

‖Un‖H2,2k (M) ≤ C for all n ∈ N ,

there exists a subsequence (Uni)n∈N ∈ H2,2k (M) and exists U ∈ H2,2

k (M) such that

Uni U , (1.6)

weakly in H2,2k (M) when n→∞.

Let us return to the same bounded sequences (Un)n∈N in H2,2k (M). Let T : H2,2

k (M)→F be a compact operator and F a Banach space. Then the sequences (T (Un))n∈N converge

to T (U) (note the weak convergence of (1.6)). Furthermore, if Tn → T in(H2,2k (M)

)∗,

then

Tn(Un)→ T (U) .

We define the spaces Lqk(M) = Lq(M,Rk), for each q with 1 ≤ q < ∞, as the space

Lq(M) × · · · × Lq(M). That is,

Lqk(M) = U = (u1, · · · , uk); ui ∈ Lq(M), i = 1, · · · , k ,

with norm

‖U‖Lqk(M) =

(k∑i=1

‖ui‖q

) 1q

,

where U = (u1, . . . , uk) and ‖ · ‖p is the norm of Lp(M), defined by

‖u‖p =

(∫M

|u|p dvg) 1

p

.

To simplify, when there is no ambiguity,

‖U‖p

to indicate the norm of U = (u1, . . . , uk) in Lpk(M)

The Sobolev immersion H2,2k (M) → Lqk(M) is compact for 1 ≤ q < 2n

n−4and continuous

if q = 2nn−4

. So we say that 2∗ = 2nn−4

is the critical exponent with respect to the immersion

H2,2k (M) → Lqk(M).

Page 30: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

30

1.5 Coercivity

Consider U = (u1, ..., uk) ∈ H2,2k (M). We define the functional Φ : H2,2

k (M) → R and

Ψ = ΨG : H2,2k (M)→ R, respectively, by:

Φ(U) =

∫M

F (U) dvg (1.7)

and

Ψ(U) = ΨG(U) =

∫M

(∆gU)2 dvg +

∫M

A((∇U)#, (∇U)#

)dvg +

∫M

G(x, U) dvg(∫M

F (U) dvg

) 2

2#

,

where A is as defined at the beginning (a sum of smooth (2,0)-tensors). We also define

Ψ : H2,2k (M)→ R by

Ψ(U) =

∫M

(∆gU)2 dvg +

∫M

A((∇U)#, (∇U)#

)dvg +

∫M

G(x, U) dvg .

Thus,

Ψ(U) = ΨG(U) =Ψ(U)(∫

M

F (U) dvg

) 2

2#

. (1.8)

We denote by L2k(M) the Sobolev space of L2(M)×L2(M)× · · · ×L2(M) with norm:

‖U‖2L2k(M) =

∫M

|U |2 dvg =k∑i=1

∫M

|ui|2 dvg ,

where U = (u1, · · · , uk). Similarly, we define the space L2#

k (M).

Using the Sobolev immersions H2,2k (M) → L2

k(M) and H2,2k (M) → L2#

k (M) the

functional ΨG and Φ are well defined.

Definition 1 (coercivity). We say that Ψ is coercive if exists α > 0 such that:∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg ≥ α

∫M

|U |2 dvg ,

for all U ∈ H2,2k (M), where A and G : M × Rk → R are as defined before.

We have the following proposition that show equivalent definitions of coercivity.

Proposition 5. The definitions below are equivalent:

(i) Ψ is coercive.

Page 31: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

31

(ii) Exists α > 0 such that:∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg ≥ α‖U‖2

2#

≥ αM− 2

2#

F

(∫M

F (U) dvg

) 2

2#

,

for all U ∈ H2,2k (M).

(iii) Exists α > 0 such that:∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg ≥ α‖U‖2

H2,2k (M)

,

for all U ∈ H2,2k (M).

Proof. (iii)⇒ (ii)

Follows from the immersion:

H2,2k (M) → L2#

k (M) .

(ii)⇒ (i)

We obtain using the immersion:

L2#

k (M) → L2k(M) .

(i)⇒ (iii)

With (i) we have:∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg ≥ α‖U‖2

L2k(M) ,

for some α > 0. Consider 0 < ε < 1, such that ε ≤ αα+k

, where k > 0 is a constant

presented below. We have:

L(U) :=

∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg , (1.9)

is equal to

L(U) = εL(U) + (1− ε)L(U) .

Hence,

L(U) ≥ ε

∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg + (1− ε)α

∫M

|U |2 dvg .

Page 32: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

32

Using the homogeneity of G and that A is bounded, we have:

L(U) ≥ ε

∫M

((∆gU)2 + c|∇U |2 +mG|U |2

)dvg + (1− ε)α

∫M

|U |2 dvg .

Consider ε0 = min ε, εc, εmG, that is, ε0 = rε, where r = 1, c, or mG. Then

L(U) ≥ ε0

∫M

((∆gU)2 + |∇U |2 + |U |2

)dvg + (1− ε)α

∫M

|U |2 dvg ,

take (1− ε)α ≥ ε0. Hence:∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg ≥ ε0‖U‖2

H2,2k (M)

.

Page 33: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

33

1.6 The Scalar AB Program

In Rn, there exists a constant A > 0 such that

‖u‖2

L2# (Rn)≤ A

∫Rn

(∆ξu)2 dx , (1.10)

for all u ∈ C∞c (Rn) (the set of smooth functions with compact support in Rn) and where

ξ is the Euclidean metric of Rn.

Let A0 = A0(n) be the sharp constant in the Sobolev inequality (1.10). That is, A0

is the smallest constant satisfying (1.10). We define:

1

A20

= infu∈D2

2(Rn)\0

∫Rn

(∆ξu)2 dx(∫Rn|u|2# dx

) 2

2#

, (1.11)

where D22(Rn) is is the completion of the C∞c (Rn) com a norma ‖u‖D2

2(Rn) := ‖∆ξu‖2.

Follows from the Sobolev immersion theorem that the constant A0 > 0 is well defined.

This constant was calculated by Lieb [25], Lions [26], Edmunds-Fortunato-Jannelli [14]

and Swason [37]. We have:

1

A20

=n(n2 − 4)(n− 4)w

4nn

16,

where wn is the volume of the canonical unit sphere Sn in Rn+1.

Moreover, the extremal of the sharp inequality, that is, functions in D22(Rn) that

achieve the infimum in (1.11) are known and they are of the form:

uλ,µ,x0(x) = µ

λ2 + |x− x0|2

)n−42

,

where µ 6= 0, λ > 0 and x0 ∈ Rn are arbitrary.

The following Sobolev immersion is continuous, but not compact. If (M, g) is a Rie-

mannian manifold of dimension n ≥ 5 then H2,2(M) → L2#(M) continuously, where

2# = 2nn−4

. That is, exists A > 0 such that:

‖u‖L2# (M)

≤ A‖u‖H2,2(M) .

Further, from the continuous immersion H2,2(M) → L2#(M), then there are constants

A,B > 0 such that:

‖u‖22# ≤ A‖∆gu‖2

2 +B‖u‖2H1,2(M) , ∀ u ∈ H2,2(M) , (1.12)

where

Page 34: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

34

‖u‖2H1,2(M) = ‖∇u‖2

2 + ‖u‖22 .

We are interested in the sharp constant A,B from the inequality above. More precisely,

we are interested in taking A minimized.

Easily see that A ≥ A20. Moreover, it holds true, for all ε > 0 exists Bε > 0 such that,

for any u ∈ H2,2(M) we have:

(∫M

|u|2# dvg) 2

2#

≤ (A20 + ε)

(∫M

(∆gu)2 dvg

)+Bε

∫M

(|∇gu|2 + |u|2

)dvg . (1.13)

In particular, we define

Kn = infA ∈ R; exists B such that the above inequality (1.12) is valid ∀ u ∈ H2,2(M)

,

(1.14)

then Kn = A20 for all manifold (M, g). A question that naturally comes is the following:

• The infimum Kn is achieved?

Or, equivalently, we can take ε = 0 in (1.13)?A positive response was given in 2000

by Djadli-Hebey-Ledoux [10] with the restriction that the metric g is conformally flat.

Then in 2003, Emmanuel Hebey proved the result for any Riemannian manifold [16].

Specifically:

Theorem 6 (E. Hebey, 2003). Let (M, g) be a compact manifold of dimension n ≥ 5.

Then exists B > 0, which depends on the manifold and the metric g, such that, for all

u ∈ H2,2(M)

(∫M

|u|2# dvg) 2

2#

≤ A20

∫M

(∆gu)2 dvg +B

∫M

(|∇gu|2 + |u|2

)dvg .

In particular, the infimum is achieved in (1.12).

Let (M, g) be a compact Riemannian manifold of dimension n ≥ 5. We denote

H2,2(M) as the completion of the space C∞(M) in the norm:

‖u‖2H2,2(M) =

∫M

(∆gu)2 dvg +

∫M

(|∇gu|2 + |u|2

)dvg .

As 2# = 2nn−4

, consider two positive functions f ∈ C∞(M) and h ∈ C0(M), exist two

positive constantes A, B such that, for any u ∈ H2,2(M):

(∫M

|u|2# dvg) 2

2#

≤ A

∫M

(∆gu)2 dvg +B

∫M

(f(x)|∇gu|2 + h(x)|u|2

)dvg . (1.15)

Page 35: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

35

The first best Sobolev constant associated to the (1.15) is defined by:

A0(f, h, g) = inf A ∈ R; exists B ∈ R such that (1.15) is valid .

The first sharp Riemannian Sobolev inequality asserts that, for any u ∈ H2,2(M),

(∫M

|u|2# dvg) 2

2#

≤ A0(f, h, g)

∫M

(∆gu)2 dvg +B

∫M

(f(x)|∇gu|2 + h(x)|u|2

)dvg ,

(1.16)

for some constant B ∈ R.

The second sharp Sobolev constant associated to (1.15) is defined by:

B0(f, h, g) = inf B ∈ R; (1.16) is valid .

The second Sobolev Riemannian sharp inequality sates that, for any u ∈ H2,2(M) we

have:

(∫M

|u|2# dvg) 2

2#

≤ A0(f, h, g)

∫M

(∆gu)2 dvg +

+ B0(f, h, g)

∫M

(f(x)|∇gu|2 + h(x)|u|2

)dvg . (1.17)

A nonzero function u0 ∈ H2,2(M) is called extremal for the inequality (1.17) if

(∫M

|u|2# dvg) 2

2#

= A0(f, h, g)

∫M

(∆gu)2 dvg+B0(f, h, g)

∫M

(f(x)|∇gu|2 + h(x)|u|2

)dvg .

The AB scalar program consists of several questions of interest involving the sharp

constant A0(f, h, g) e B0(f, h, g), and the inequalities otimas (1.17) and (1.16). In what

follows, we will divide this program into parts: A program A and B program.

The A program consists of some problems involving A0(f, h, g) and the inequality

(1.16).

• question 1A: What is the exact value, or estimates of A0(f, h, g)?

• question 2A: The inequality (1.16) is valid?

• question 3A: The validity of (1.16) implies some geometric obstruction?

• question 4A: A0(f, h, g) depends continuously of f and h in some topology?

• question 5A: A0(f, h, g) depends continuously on the metric g in some topology?

• question 6A: What is the role of geometry on these questions?

Page 36: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

36

The B program consists of some questions involving B0(f, h, g) and the inequality

(1.17):

• question 1B: What is the exact value, or estimates of B0(f, h, g)?

• question 2B: B0(f, h, g) depends continuously on f and h in some topology?

• question 3B: B0(f, h, g) depends continuously on the metric g in some topology?

• question 4B: The inequality (1.17) has extremal function?

• question 5B: The set of the extremal functions E(f, h, g) of L2#-norm is compact

in the C0 topology?

• question 6B: What is the role of geometry on these questions?

1.6.1 Partial Answers

As shown at the beginning of the previous section, we have answers only to the questions

1A and 2A.

There are no reponses for the questions of the scalar case in the B program. That is,

no author has worked on these questions.

We will answer some of the questions of the vector case, which in turn will provide

ansswers to similar questions of the scalar program.

Page 37: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

37

1.7 AB Vector Program

Let n ≥ 5 and k ≥ 1 be integer. We denote by D2,2k (Rn) the Euclidean Sobolev vector

space D2,2(Rn)× · · · × D2,2(Rn) with norm

‖∆U‖D2,2k (Rn) =

(∫Rn

(∆U)2 dx

) 12

,

where

U = (u1, . . . , uk)

and

∫Rn

(∆U)2 dx =k∑i=1

∫Rn

(∆ui)2 dx .

Let F : Rk → R be a positive continuous function and 2#-homogeneous. In this case,

follows directly from (1.10) the existence of a constant A > 0 such that

(∫RnF (U) dx

) 2

2#

≤ A∫Rn

(∆U)2 dx , (1.18)

for all U ∈ D2,2k (Rn).

The sharp Euclidean Sobolev constant A associated with the inequality (1.18) is:

A0(F, n) = inf A ∈ R; (1.18) is valid .

The sharp Euclidean Sobolev vector inequality states that:

(∫RnF (U) dx

) 2

2#

≤ A0(F, n)

∫Rn

(∆U)2 dx , (1.19)

for all U ∈ D2,2k (Rn).

A nonzero map U0 ∈ D2,2k (Rn) is called extremal of (1.19), if

(∫RnF (U) dx

) 2

2#

= A0(F, n)

∫Rn

(∆U)2 dx . (1.20)

two basic questions related to (1.19) are:

(a) What is the exact value of A0(F, n)?

(b) (1.19) has extremal map?

We have reponses for these two questions in the following result.

Page 38: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

38

Proposition 7. We have that

A0(F, n) = M2

2#

F A0(n) ,

where MF = maxSk−12

F and Sk−12 =

t ∈ Rk;

∑ki=1 |ti|2 = 1

and A0(n) is given by (1.11).

Furthermore, U0 ∈ D2,2k (Rn) is a extremal map of (1.19) if and only if, U0 = t0u0 for some

t0 ∈ Sk−12 such that MF = F (t0) and some extremal function u0 ∈ D2,2(Rn) of (1.10).

Proof. From the 2#− homogeneity of F :

F (t) ≤ MF

(k∑i=1

|ti|2) 2#

2

, ∀ t ∈ Rk .

Thus, using the Minkowski inequality and by (1.10), we have for any U ∈ D2,2k (Rn):

(∫M

F (U) dx

) 2

2#

≤ M2

2#

F

∫M

(k∑i=1

|ui|2) 2#

2

dvg

2

2#

≤ M2

2#

F

k∑i=1

(∫M

|ui|2#

dvg

) 2

2#

≤ M2

2#

F A0(n)

∫Rn

k∑i=1

(∆ui)2 dx = M

2

2#

F A0(n)

∫Rn

(∆U)2 dx . (1.21)

Hence, we obtain that:

A0(F, n) ≤M2

2#

F A0(n) .

now, by choosing U0 = t0u0 where t0 ∈ Sk−12 is such that MF = F (t0) and u0 ∈

D2,2(Rn) is a extremal function of (1.10), we have2

(∫RnF (U0) dx

) 2

2#

= M2

2#

F

(∫Rn|u0|2

#

dx

) 2

2#

= M2

2#

F A0(n)

∫Rn

(∆u0)2 dx

= M2

2#

F A0(n)

∫Rn

(k∑i=1

∆(ti0u0

))2

dx

= M2

2#

F A0(n)

∫Rn

(∆U0)2 dx . (1.22)

A0(F, n) ≥M2

2#

F A0(n) .

2Note that, in this case:

(∆U0)2

=

k∑i=1

(∆(ti0u0

))2= (∆u0)

2

Page 39: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

39

Therefore,

A0(F, n) = M2

2#

F A0(n) .

We conclude also that the k-maps U0 = t0u0, as constructed above, are extremals. We

affirm that all extremals (1.19) has this form. In fact, let U ∈ D2,2k (Rn) be a extremal

of (1.19). In this case, U satisfies (1.21) with equality instead of the three inequalities.

But note that the second equality corresponds to the Minkowski inequality . This implies

that exists t ∈ Rk in |t|2 = 1 and u ∈ D2,2(Rn) such that U = tu. And, concluding, by

the first equality it follows that F (t) = MF and, from the third equality, u is extremal

function of (1.10).

Let F : Rk → R be a positive continuous and 2#-homogeneous function, and G :

M × Rk → R a positive continuous and 2-homogeneous function on the second variable.

Follows from the continuous immersion H2,2(M) → L2#(M) that exist two positive

constants A and B such that

(∫M

F (U)dvg

) 2

2#

≤ A∫M

(∆gU)2 dvg +

+ B∫M

(A((∇gU)#, (∇gU)#

)+G(x, U)

)dvg . (1.23)

The first Sobolev best constant associated to (1.23) is defined as:

A0 = A0(A,F,G, g) = inf A ∈ R; exists B ∈ R such that (1.23) is valid .

The first sharp Riemannian Sobolev vector inequality states that, for any U ∈ H2,2k (M),

we have:

(∫M

F (U)dvg

) 2

2#

≤ A0

∫M

(∆gU)2 dvg +

+ B∫M

(A((∇gU)#, (∇gU)#

)+G(x, U)

)dvg , (1.24)

for some constante B ∈ R.

The second Sobolev best constant associated to (1.24) is defined as:

B0 = B0(A,F,G, g) = inf B ∈ R; (1.24) is valid .

The second sharp Sobolev Riemannian vector inequality states that, for any U ∈H2,2k (M), vale

Page 40: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

40

(∫M

F (U)dvg

) 2

2#

≤ A0

∫M

(∆gU)2 dvg +

+ B0

∫M

((A((∇gU)#, (∇gU)#

)+G(x, U)

)dvg . (1.25)

This inequality is sharp in relation to the first and the second sharp Sobolev constant, in

that where neither can be reduced.

A natural question arises: It is possible to achieve equality in (1.25)? A nonzero

k-function U0 ∈ H2,2k (M) is called extremal of (1.25), if

(∫M

F (U0)dvg

) 2

2#

= A0

∫M

(∆gU0)2 dvg +

+ B0

∫M

((A((∇gU0)#, (∇gU)#

)+G(x, U0)

)dvg .

The AB vector program consists of several questions of interest involving the best

constants A0(A,F,G, g) and B0(A,F,G, g) and the sharp inequalities (1.24) AND (1.25).

This program is separated into two parts: the A program and the B program.

The A program is composed of the following questions involving A0(A,F,G, g) and

(1.24):

• question 1A: What is the exact value or estimates of A0(A,F,G, g)?

• question 2A: The inequality (1.24) is valid?

• question 3A: The validity of (1.24) implies in some geometric obstruction?

• question 4A: A0(A,F,G, g) depends continuously of F and G in some topology?

• question 5A: A0(A,F,G, g) depends continuously of g and A in some topology?

• question 6A: What is the role of geometry in these questions?

The B program consists of the following questions involvingB0(A,F,G, g) and (1.25):

• question 1B: What is the exact value or estimates of B0(A,F,G, g)?

• question 2B: B0(A,F,G, g) depends continuously of F and G in some topology?

• question 3B: B0(A,F,G, g) depends continuously on the metric g and A in some

topology?

• question 4B: The inequality (1.25) has extremal?

Page 41: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

41

• question 5B: The set of the extremal E(A,F,G, g) normalized by∫MF (U)dvg = 1

is compact in the C0-topology?

• question 6B: What is the role of geometry in these questions?

1.7.1 Partial Answers

Some questions of the A program have been answered only by some authors for the scalar

case, that is, no previous studies exist for the vector case. One goal of this work is to give

a contribution to these questions in the vectorial case. The B program does not have any

previous result, even for the scalar case. We will show below answers to some questions

of the B vector program, which will give us, as already mentioned, responses to program

B in the scalar case.

We begin answering the question 1A of the AB program

Proposition 8. We have

A0(A,F,G, g) = M2

2#

F A0 ,

where MF = maxSk−12

F and Sk−12 =

t ∈ Rk;

∑ki=1 |ti|2 = 1

. In particular the inequality

otima (1.24) is valid.

Proof. We have, from the 2#− homogeneity of F :

F (t) ≤ MF

(k∑i=1

|ti|2) 2#

2

, ∀ t ∈ Rk .

Thus, using the Minkowski inequality, we have:

(∫M

F (U) dx

) 2

2#

≤ M2

2#

F

∫M

(k∑i=1

|ui|2) 2#

2

dvg

2

2#

≤ M2

2#

F

k∑i=1

(∫M

|ui|2#

dvg

) 2

2#

.

That is,

(∫M

F (U) dx

) 2

2#

≤M2

2#

F

(∫M

|U |2# dvg) 2

2#

. (1.26)

On the other hand, by the definition of A0, we have

k∑i=1

(∫M

|ui|2#

dvg

) 2

2#

≤ A0

∫M

k∑i=1

(∆gui)2 dvg +B

∫M

k∑i=1

(|∇ui|2 + u2

i

)dvg .

That is,

Page 42: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

42

(∫M

|U |2# dvg) 2

2#

≤ A0

∫M

(∆gU)2 dvg +B

∫M

(|∇U |2 + |U |2

)dvg . (1.27)

By the hypotheses of G (remember that G : M×Rk → R is a positive continuous function

and 2-homogeneous on the second variable), there exists a constant m > 0 such that:

G(x, t) ≥ mk∑i=1

|ti|2, ∀ x ∈ M, t ∈ Rk ,

where m = minM×Sk−12

G. Hence, by joining (1.26) and (1.3) in the inequality (1.27), we

obtain for all U ∈ H2,2k (M):

(∫M

F (U) dvg

) 2

2#

≤M2

2#

F A0

∫M

(∆gU)2dvg+BM

2

2#

F

max c, m

∫M

(A((∇U)#, (∇U)#

)+G(x, U)

)dvg ,

this implies que

A0(A,F,G, g) ≤ M2

2#

F A0 . (1.28)

Now consider constant A and B such that

(∫M

F (U) dvg

) 2

2#

≤ A∫M

(∆gU)2 dvg + B∫M

((A((∇gU)#, (∇gU)#

)+G(x, U)

)dvg ,

for all U ∈ H2,2k (M). We use (1.3) and let us now consider the constant A and B =

min BC,B, such that

(∫M

F (U) dvg

) 2

2#

≤ A0

∫M

(∆gU)2 dvg + B∫M

(|∇U |+G(x, U)) dvg .

We choose U = ut0 =(ut10, ut

20, ..., ut

k0

), where t0 =

(t10, . . . , t

k0

)∈ Sk−1

2 is such that

F (t0) = MF .using also that G is 2-homogeneous, we obtain:

Page 43: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

43

(∆U)2 =k∑i=1

(∆uti0

)2= (∆u)2|t0|2 = (∆u)2

|∇U |2 =k∑i=1

∣∣∇uti0∣∣2 =k∑i=1

|ti|2|∇u|2 = |t0|2|∇u|2 = |∇u|2

|U |2 =k∑i=1

(uti0)2

= u2|t0|2 = u2

G(x, U) = G(x, t0u) = |u|2G(x, t0) ≤ c|u|2 .

where u ∈ H2,2(M). We found a constant B1 > 0 such that:

(∫M

|u|2# dvg) 2

2#

≤ A0M− 2

2#

F

∫M

(∆gu)2 dvg +B1

∫M

(|∇u|2 + u2

)dvg .

thus, we obtain that

A0M− 2

2#

F ≥ A0 . (1.29)

Hence, from (1.29) and (1.28), we obtain:

A0(A,F,G, g) = M2

2#

F A0 .

If F (U) =∑k

i=1 |ui|2#

, then A0 = Kn, because MF = 1. That is, in this case, the first

best constant A0 only depends on n.

Note that the first best constant from the vector theory does not depend on the

geometry and neither G, and depends on n and F with respect to C0loc(M) topology.

The first best constant A0 depends continuously on F . Let (Fα)α be a family of

positive continuous functions, Fα : Rk → R and 2#−homogeneous for each α which

converges toF : Rk → R on the C0 topology. Thus we have Aα0 = M2

2#

FαKn is the first

best constant associated to Fα:

(∫M

Fα(U) dvg

) 2

2#

≤ Aα0∫M

(∆gU)2 dvg +B∫M

((A((∇gU)#, (∇gU)#

)+G(x, U)

)dvg .

But note that we have the following convergence: Aα0 → A0. That is, A0 depends

continuously on F .

What was done above answer some of the questions involved in A vector program.

Let us now focus on B vector program. More precisely, we answer some questions of the

B vector program in chapter 3.

Page 44: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

44

In chapter 2 that follows, we developed a theory that will be used to demonstrate the

existence of extremal in chapter 3.

Page 45: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

45

Chapter

2Elliptic Systems of Fourth Order

This part that follows return attention to the system of fourth order:

−∆2gU + divg

(A(∇U)#

)+

1

2∇UG(x, U) =

1

2#∇F (U) , (2.1)

where U = (u1, . . . , uk). That is,

−∆2gui + divg

(Ai(∇ui)#

)+

1

2∂iG(x, U) =

1

2#∂iF (U) ,

for i = 1, . . . , k. For most of the results in this section, we consider F : Rk → R a

C1 positive function and 2#-homogeneous and G : M × Rk → R a C1 positive function

and 2-homogeneous on the second variable. But, some results (as the L2 concentration

and compactness) we will take the following particular function: G(x, t) =∑k

ij=1Aijtitj,

where t = (t1, . . . , tk) ∈ Rk, Ai = big (bi ∈ R and g is the metric of the manifold) and

Aij : M → R are functions such that a matrix (Aij(x)) is positive defined and symmetric,

for all x ∈M . Thus, o system (2.1) is being:

−∆2gui + bi∆gui +

1

2

k∑j=1

Aijuj =1

2#|ui|2

#−2ui . (2.2)

In the scalar version of (2.1), a particular case is the following:

∆2gu+ bα∆gu+ cαu = u2#−1 (2.3)

where bα and cα are positive continuous functions. This case has been studied by many

authors (see for exemple [10], [15] and [24]). They analyze the solutions and the energy

functional associated. One of the questions in the work of Hebey-Wen is if bα and cα

converges to b0 and c0 respectively, that is, we have

∆2gu+ b0∆gu+ c0u = u2#−1 (2.4)

and if the solutions uα converges weakly in H2,2(M), then which conditions the weak limit

is not trivial? In this section we put the same question to solutions of (2.2) and we estab-

lished a result in the same direction of Hebey-Wen, that is the theorem of compactness

Page 46: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

46

that is the end of this section.

One of the main differences between (2.1) and (2.3) is that the second equation can

be written, in most cases, as the product of elliptic operators of second order. This helps

to use standards techniques of second order as the Moser iteration from analysis. But

in the case of (2.1) we will not have a product in general. Thus, we are not able to use

directly the techniques of second order. Thus we used a technique of Moser iteration for

operators of fourth order done by Sandeep (see [31]).

Consider the operator Pg := ∆2g − divg

(A(∇·)#

)+ a (where a ∈ C∞(M) and A

is a smooth symmetric (2,0)-tensor). This operator Pg does not satisfy the principle of

punctual comparison, even when it is coercive. This is an important difference to operators

of second order.

A question of importance is the following: the operator Pg satisfies the maximum

principle? A resposta is positive if Pg is the product of two elliptic operators, each one

coercive. Thus, Pg = (∆g + a) (∆g + a′), where a and a′ ∈ C∞(M). Also, if 4b0 ≤ c20 in

(2.4), we obtain the decomposition of Pg in elliptic coercive operators and therefore we

have also the maximum principle (see [30]).

One of the highlight of this section is the bubbles decomposition theorem, since using

this theorem, we were able to prove the main results and get our contributions. We

demonstrate that theorem proving initially that Palais-Smale sequence converges weakly

and we obtain the weak limit of the sequence. We obtain a new Palais-Smale sequence

for a new functional. Thence we use the scalar case to complete the theorem..

Our contribution here gives extension to some work already done. (for example E.

Hebey, F. Robert, Z. Djadli, M. Ledoux, V. Felli, and others) where has been studied

(2.2) in the escalar case.

This section will begin with the existence of solutions, then we will through the regu-

larity, punctual estimates and L2 concentration, ending with compactness.

2.1 Existence

Let q ∈[2, 2#

]. We define:

µq = infIq(U);U ∈ H2,2

k (M) \ 0 e Φ(U) = 1,

where Φ(U) =

∫M

F (U) dvg (see (1.7)) and Iq : H2,2k (M) \ 0 → R is the following

functional:

Iq(U) =

∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg(∫

M

F (U) dvg

) 2q

, (2.5)

Page 47: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

47

defined for all U ∈ H2,2k (M) \ 0, where F : Rk → R is a q-homogeneous function. The

functional Iq is well defined, because Sobolev embeddingH2,2k (M) → Lqk(M) is continuous.

To get solutions of (2.1) in the critical case, that is, when q = 2#, we have to use

another method. This presented later. In the result which follows, we focus on the

subcritical case, that is, when q < 2#.

Proposition 9. Let (M, g) be a compact Riemannian manifold of dimension n ≥ 5. Let A

be a symmetric (2, 0)−tensor; F : Rk → R is a C1 function positive and q−homogeneous

and G : M × Rk → R is a continuous function, of class C1 and 2-homogeneous on the

second variable. Assume that q ∈[2, 2#

). Then, µq is finite and is achieved. That is,

µq ∈ R and there is U ∈ H2,2k (M) \ 0 such that Iq(U) = µq.

Proof. Initially, we prove that µq > −∞. Let U ∈ H2,2k (M) \ 0. As A is smooth, then

there exists C > 0 such that∣∣∣∣∫M

A((∇U)#, (∇U)#

)dvg

∣∣∣∣ ≤ C

∫M

|∇U |2g dvg , (2.6)

for all U ∈ H2,2k (M). We note that there is a constant C ′ > 0 such that∣∣∣∣∫

M

A((∇U)#, (∇U)#

)dvg

∣∣∣∣ ≤ 1

2

∫M

(∆gU)2 dvg + C ′‖U‖22 , (2.7)

for all U = (u1, . . . , uk) ∈ H2,2k (M), where ‖U‖2

2 =∑k

i=1

∫M|ui|2 dvg.

To prove the above inequality, we prove the two following lemmas.

Lemma 10. Let (M, g) be a compact Riemannian manifold. Then, for each ε > 0, exists

C(ε) > 0 such that

‖∇u‖2 ≤ ε‖∆gu‖2 + C(ε)‖u‖2, ∀ u ∈ H2,2(M) .

Proof. Suppose, by contradiction, that there is ε > 0 such that, for each α ∈ N∗, exists

uα ∈ H2,2(M) such that

‖∇uα‖2 > ε‖∆guα‖2 + α‖uα‖2, e ‖∇uα‖2 = 1 .

Thence,

1 > ε‖∆guα‖2 e 1 > α‖uα‖2 .

That is,

‖∆guα‖2 <1

εe ‖uα‖2 <

1

α.

Thus,

Page 48: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

48

‖∆guα‖2 + ‖∇uα‖2 + ‖uα‖2 <1

ε+ 1 +

1

α.

Therefore, we have that exists a constant C > 0 such that

‖uα‖H2,2(M) ≤ C .

Furthermore, we have ‖uα‖2 → 0. As the embedding H2,2(M) → H1,2(M) is compact,

up to a subsequence, we have uα u (u ∈ H2,2(M)) and uα → u in H1,2(M). Therefore,

‖∇u‖2 = 1 and ‖u‖2 = 0, that is a contradiction.

We will use this lemma 10 above to prove the lemma which follow. To prove (2.7),

just we check the escalar case. The vector case follows directly from the lemma below.

Lemma 11. Let A be a (2, 0)−tensor. There exists a constant C ′ > 0 such that∣∣∣∣∫M

A((∇u)#, (∇u)#

)dvg

∣∣∣∣ ≤ 1

2

∫M

(∆gu)2 dvg + C ′‖u‖22 .

Proof. Let C > 0 be a constant on the inequality (2.6). We take ε = 1√4C

in the lemma

10 above:

‖∇u‖2 ≤1√4C‖∆gu‖2 + C(ε)‖u‖2 .

Thence,

‖∇u‖22 ≤

(‖∆gu‖2√

4C+ C(ε)‖u‖2

)2

=‖∆gu‖2

2

4C+

2C(ε)√4C‖∆gu‖2‖u‖2 + C2(ε)‖u‖2

2

≤ ‖∆gu‖22

4C+‖∆gu‖2

2

4C+

4C2(ε)‖u‖22

2+ C2(ε)‖u‖2

2 .

Thus,

C

∫M

|∇u|2 dvg ≤1

2

∫M

(∆gu)2 dvg + C ′‖u‖22 .

Therefore, by (2.6) we obtain the result of the lemma.

Let us now turn to the proof of the proposition. As F is q-homogeneous and positive,

we have that there exists a constant MF,2 = M ′F > 0 such that:

F (U) ≤M ′F |U |

q2 ≤ M ′

F |U |qq , para toda U ∈ Rk , (2.8)

Page 49: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

49

where |U |22 = |u1|2 + · · · + |uk|2 and U = (u1, . . . , uk) (see section 1.3, p. 23). And, from

the homogeneity of G, we obtain that there is a constant mG > 0 such that:

mG|U |22 ≤ G(x, U) , para toda U ∈ Rk . (2.9)

We will use now the above inequalities (2.9), (2.8), (2.7) and the Holder inequality (also

used the idea at the beginning of proof of proposition 8). Note that, for each i = 1, . . . , k

we have

∫M

|ui|2 dvg ≤(∫

M

|ui|q dvg) 2

q

vol(M)1− 2q

and as ‖ui‖2q ≤ ‖U‖2

q, we have:

∫M

|U | dvg ≤ k

(∫M

|U |qq dvg) 2

q

vol(M)1− 2q

Then we have:

Iq(U) =

∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg(∫

M

F (U) dvg

) 2q

1

2

∫M

(∆gU)2 dvg − (C ′ −mG)‖U‖22(∫

M

F (U) dvg

) 2q

∫M

(∆gU)2 dvg

2

(∫M

F (U) dvg

) 2q

− (C ′ −mG) ‖U‖22

V olg(M)2q−1(M ′

F

) 2q ‖U‖2

q

=

∫M

(∆gU)2 dvg

2

(∫M

F (U) dvg

) 2q

− k (C ′ −mG)

V olg(M)2q−1(M ′

F

) 2q

, (2.10)

for all U ∈ H2,2k (M) \ 0, where ‖U‖2

2 =∑k

i=1 ‖ui‖22. Note that we are assuming that

(C ′ −mG) > 0. Otherwise, if (C ′ −mG) < 0 we would have:

Iq(U) ≥

∫M

(∆gU)2 dvg

2 (M ′F )

2q ‖U‖2

q

This prove, for both cases, that µq > −∞ and then µq ∈ R. Now, let (Uα)α∈N ∈

Page 50: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

50

H2,2k (M) \ 0 be a minimizing sequence for Iq, that is:

limα→∞

Iq(Uα) = µq . (2.11)

Without loss of generality, we can assume that∫M

F (Uα) dvg = 1 , (2.12)

for all α ∈ N. From (2.10) and (2.11), we have that theres exists a constant C0 > 0 such

that: ∫M

(∆gUα)2 dvg ≤ C0 ,

for all α ∈ N. From (1.4) and (2.12), we obtain:

‖Uα‖22 =

∫M

|Uα|22 ≤ C1, para uma constant C1 > 0 .

And by lemma 10, using the two limitations above, we have:

‖∇U‖22 ≤ C2 ,

where C2 > 0. Thence, exists a constant C > 0, such that:

‖Uα‖H2,2k (M) ≤ C ,

for all α ∈ N. As the unit ball of H2,2k (M) is weakly compact, we have that there exists

U ∈ H2,2k (M) that is the weak limit of (Uα)α∈N, up to the extraction of a subsequence.

That is, exists a subsequence (Uα′) ⊂ (Uα) such that

Uα′ U ,

weak convergence in(H2,2k (M)

)∗, when α → ∞. Without loss of generality, we assume

that the convergence is for the initial sequence (Uα)α∈N. As the embedding H2,2k (M) →

H1,2k (M) is compact, we have

limα→∞

Uα = U em H1,2k (M) .

For 2 ≤ q < 2#, the embedding H2,2k (M) → Lqk(M) is compact, we have also

limα→∞

Uα = U em Lqk(M) .

We also have Uα → U qtp. Thus∫MF (U) dvg = 1 and by (1.4) we conclude that U 6= 0,

where U = (u1, . . . , uk). Consider Θα = Uα − U ∈ H2,2k (M) for α ∈ N. We have:

Page 51: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

51

∫M

((∆gUα)2 + A

((∇Uα)#, (∇Uα)#

)+G(x, Uα)

)dvg

=

∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg +

∫M

(∆gΘα)2 + o(1) ,

where limα→∞ o(1) = 0, because Θα 0 in(H2,2k (M)

)∗and Uα → U in H1,2

k (M) when

α→∞. Assume that:

µq =

∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg +

∫M

(∆gΘα)2 + o(1) , (2.13)

when α→∞. As U 6= 0, we have Iq(U) ≥ µq. But we also have:

µq ≤∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg +

∫M

(∆gΘα)2 . (2.14)

that is µq ≤ Iq(U). Joining (2.13) and (2.14), we have:

µq =

∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg = Iq(U) ,

and

limα→∞

∫M

(∆gΘα)2 dvg = 0 .

In particular, the infimum µq is achieved in U ∈ H2,2k (M).

Now, consider the set Λ defined by:

Λ =U ∈ H2,2

k (M); Φ(U) = 1,

where Φ(U) =∫MF (U) dvg (see (1.7)). Consider also the number λ given by:

λ = infU∈Λ

Ψ(U) ,

where Ψ(U) = I2#(U) (see (1.8) and (2.5)).

We see easily that if (Uα)α is a minimizing sequence for λ, then (Uα)α is a limited

sequence in H2,2k (M). In fact, such statement arises from the fact that (Uα)α ⊂ Λ and

from the existence of a constant mF > 0 such that mF |U |2#

2 ≤ F (U).

In the theorem which follows, we see that the system (2.1) has a solution of minimum

energy.

Page 52: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

52

Theorem 12. Let (M, g) be a compact Riemannian manifold of dimension n ≥ 5. Let

F : Rk → R be a positive function of class C1 and 2#−homogeneous and G : M×Rk → Rbe a continuous function of class C1 and 2-homogeneous on the second variable. Let also

A be a smooth symmetric (2, 0)−tensor and assume that Ψ is coercive. If λ < 1A0

, then

(2.1) has weak solution U0 ∈ Λ such that

Ψ(U0) = λ .

Proof. As Ψ is coercive, we have that exists α > 0 such that

Ψ(U) ≥ α‖U‖22# ≥ M

− 2

2#

F,2#α

(∫M

F (U) dvg

) 2

2#

,

(see the proposition 5). We claim that the infimum λ is positive, that is, λ > 0. From

the above inequality we obtain:

Ψ(U) =

∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg(∫

M

F (U) dvg

) 2

2#

≥ α

M2

2#

F

,

for all U ∈ H2,2k (M). Thus λ ≥ αM

− 2

2#

F > 0. This proves the assertion.

Let (Uα)α ⊂ Λ be a minimizing sequence for λ. That is,

limα→∞

Ψ (Uα) = λ .

The sequence is limited, existe U0 ∈ H2,2k (M) such that, up to a subsequence, we have:

Uα U0 weakly in H2,2k (M)

Uα → U0 strongly in H1,2k (M) and in L2

k(M) .

We consider Θα = Uα − U0 ∈ H2,2k (M) for all α ∈ N. As in the proof of the previous

proposition (proposition 9), we have:

λ = Ψ(U0) +

∫M

(∆gΘα)2 dvg + o(1) , (2.15)

where limα→∞ o(1) = 0. Without loss of generality, we can assume that:

limα→∞

Θα(x) = 0 q.t.p x ∈ M .

By the Brezis-Lieb generalized lemma (see [1]), we have:

Page 53: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

53

limα→∞

(∫M

F (U0 + Θα) dvg −∫M

F (U0) dvg −∫M

F (Θα) dvg

)= 0 . (2.16)

With the inequality (1.24) (also from (1.3) and (1.5)) and the convergence of Θα in

H2,2k (M), we have:

(∫M

F (Θα) dvg

) 2

2#

≤ A0

∫M

(∆gΘα)2 dvg +B

∫M

(|∇Θα|2 +G(x,Θα)

)dvg

≤ A0

∫M

(∆gΘα)2 dvg + o(1) . (2.17)

By the definition, we have Ψ is 2-homogeneous, that is, Ψ(αU) = α2Ψ(U), for α ∈ R.

Thus:

λ ≤ Ψ

U0(∫M

F (U0) dvg

) 1

2#

=1(∫

M

F (U0) dvg

) 2

2#

Ψ(U0)

Therefore,

Ψ(U0) ≥ λ

(∫M

F (U0) dvg

) 2

2#

. (2.18)

Using in (2.15) the inequalities (2.16), (2.17) and (2.18), we obtain:

λ ≥ λ

(∫M

F (U0) dvg

) 2

2#

+ (A0)−1

(∫M

F (Θα) dvg

) 2

2#

+ o(1)

≥ λ

(∫M

F (U0) dvg

) 2

2#

+ (A0)−1

(1−

∫M

F (U0) dvg

) 2

2#

+ o(1) .

When α→ +∞ in this last inequality we obtain:

λA0

(1−

(∫M

F (U0) dvg

) 2

2#

)≥(

1−∫M

F (U0) dvg

) 2

2#

.

Note that 22#

= 1− 4n< 1 and

F (Uα) F (U0) em L1(M) ⇒∫M

F (U0) dvg ≤ lim inf

∫M

F (Uα) dvg = 1 .

We then conclude then:

Page 54: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

54

∫M

F (U0) dvg ≤ 1 . (2.19)

As 1−Xp ≤ (1−X)p for all X ∈ [0, 1] and for all 0 ≤ p ≤ 1, we obtain1:

(λA0 − 1)

(1−

(∫M

F (U0) dvg

) 2

2#

)≥ 0 .

From the hypothesis λ < 1A0

, we obtain:∫M

F (U0) dvg ≥ 1 .

On the other hand, from weak convergence (see (2.19)), we obtain that:∫M

F (U0) dvg ≤ 1 .

Therefore, ∫M

F (U0) dvg = 1 .

Thus, U0 ∈ Λ and U0 6= 0 (by (1.4), because F is 2#-homogeneous). Thus as in

the proof of the previous proposition, we also have limα→∞Θα = 0 in H2,2k (M) and

limα→∞∫M

(∆Θα)2 dvg = 0. Thus, from (2.15), we have

λ = Ψ(U0) .

Therefore, U0 ∈ H2,2k (M) \ 0 is a minimizer for Ψ and U0 is solution for the problem

(2.1), that is,

Ψ′(U0)V = λΦ′(U0)V for all V ∈ H2,2k (M) ,

Lagrange multipliers theorem.

1If (X − 1) ≥ 0, then 1 = (1−X +X)p ≤ (1−X)p +Xp, because (a+ b)s ≤ as + bs if a, b > 0 ands ∈ [0, 1] .

Page 55: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

55

2.2 Regularidade

The Trudinger strategy for second order operators, as in the case of Yamabe type op-

erators, does not apply satisfactorily to the case of fourth order operators. But we can

note that Sandeep ([31]) was able to develop the De Giorgi-Nash-Moser method for fourth

order equations, which is a technique that is very close to the technique of Trudinger.

Some techniques used to fourth order operators were developed by Van der Vorst and,

and in the Riemannian context, by Esposito-Robert (see [15], [38] and also [10]).

An important result that we obtain with coercivity is the following.

Proposition 13. Let (M, g) be a compact Riemannian manifold and let A be a sum of

smooth symmetric and positive (2, 0)−tensores. Assume that S ∈ Hr,pk (M) and G :

M × Rk → R of class C1 is a positive 2-homogeneous function on the second variable, of

class C1. Then existe U ∈ H4+r,pk (M) such that

−∆2gU + divg

(A(∇U)#

)+

1

2∇UG(x, U) = S(x) . (2.20)

Morover, we have

‖U‖H4+r,pk (M) ≤ C‖S‖Hr,p

k (M) .

for all U ∈ H2,2k (M) where C = C(M, g, k) > 0. And if G(x, t) =

∑ki,j=1 Aij(x)titj, where

(Aij(x)) is positive as bilinear form and symmetric, then we have unicity in (2.20).

Proof. Let A be a sum of smooth and symmetric (2, 0)−tensores and G a C1 and 2-

homogeneous function on the second variable. The functional Ψ is coercive, that is, exists

λ > 0 such that:

Ψ(U) =

∫M

((∆gU)2 + A

((∇U)#, (∇U)#

)+G(x, U)

)dvg ≥ λ

∫M

|U |22 dvg .

In this proof, we consider the operator

PgU = −∆2gU + divg

(A(∇U)#

)+

1

2∇UG(x, U) ,

or, considering the coordinates functions of U = (u1, . . . , uk), we have (we use the same

notation for the operator Pg):

Pgui = −∆2gui + divg

(Ai(∇ui)#

)+

1

2∂iG(x, U) ,

for i = 1, . . . , k.

Affirmation 1. Let p > 1. We affirm that there is C > 0 such that

‖U‖Lpk(M) ≤ C‖PgU‖Lpk(M) ∀ U ∈ H4,pk (M) .

Page 56: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

56

We prove this affirmation by contradiction. We assume that, for all α ∈ N∗, exists

Uα = (u1α, . . . , u

kα) ∈ H4,p

k (M) such that

‖Uα‖p = 1 e ‖PgUα‖p ≤1

α. (2.21)

We denote Sα(x) = (S1α, . . . , S

kα). Let Siα = Pgu

iα. Thus,

−∆2gu

iα + divg

(Ai(∇uiα)#

)+

1

2∂iG(x, Uα) = Siα(x) ,

for i = 1, . . . , k. Applying the Lp theory in

−∆2gui + divg

(Ai(∇ui)#

)= Siα(x)− ∂iG(x, Uα)

e we obtain

‖uiα‖H4,pk (M) ≤ C

(‖Siα − ∂iG(x, Uα)‖p + ‖uiα‖p

)≤ C

(‖Siα‖p + ‖∂iG(x, Uα)‖p + ‖uiα‖p

)≤ C0

(‖Siα‖p + ‖uiα‖p + ‖uiα‖p

)≤ C0

(1

α+ 2

)≤ C ′ ,

where C ′ > 0 does not depend on α. Thus, the sequence (uiα) is limited in H4,p(M).

Then, up to a subsequence, we have:

uiα ui em H4,p(M) ,

for i = 1, . . . , k. And, up to another subsequence,

uiα → ui em H2,p(M) ,

for i = 1, . . . , k, because the embedding H4,pk (M) → H2,p

k (M) is compact. For any

ϕ ∈ C∞(M), we have

∫M

(∆gu

iα∆gϕ+ Ai

((∇uiα)#, (∇ϕ)#

)+

1

2∂iG(x, Uα)ϕ

)dvg =

∫M

Siα(x)ϕ dvg ,

for i = 1, . . . , k. Or, writing in another way:∫M

ϕPguiα dvg =

∫M

Siα(x)ϕ dvg .

Now, doing α → ∞ and by using (2.21) together with the Holder inequality, we have

Page 57: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

57

Pgui = 0 in the weak sense, for i = 1, . . . , k, or PgU = 0. Form the Schauder theory, we

have ui ∈ C4(M). And, form the coercivity, we have:

0 =

∫M

UPgU dvg ≥ λ

∫M

|U |2 dvg ,

that is, U ≡ 0. That is a contradiction, because:

‖U‖p = limα→+∞

‖Uα‖p = 1 .

So, we prove the affirmation.

Affirmation 2. Let α ∈ (0, 1). We affirm that for any S ∈ C0,αk , G and A as in

the affirmation 1, then there exists U ∈ H4,2k (M) such that

PgU = S .

Now we prove the affirmation. Let us consider the functional:

I(U) =1

2

∫M

UPgU dvg −∫M

S · U dvg ,

for all U ∈ H2,2k (M), where U(x) = (u1(x), . . . , uk(x)), S(x) = (S1(x), . . . , Sk(x)) e S ·U =

s1u1 + · · ·+ uksk. From the coercivity and Holder inequality, we have:

I(U) ≥ λ‖U‖22 − ‖S‖2‖U‖2 ≥ −

‖S‖22

4λ. (2.22)

Then, µ = infI(U) ;U ∈ H2,2

k (M)> −∞ is defined. Let (Uα)α∈N ∈ H2,2

k (M) be a

minimizing sequence for µ, that is,

limα→∞

I(Uα) = µ . (2.23)

With the first inequality of (2.22), we have ‖Uα‖2 < α for all α ∈ N. From (2.23) and

the coercivity we have

‖Uα‖H2,2k (M) = O(1) ,

when α → +∞. As the unit ball of Hm,pk (M) is weakly compact and the sequence

(Uα) is limited in H2,2k (M), then exists a subsequence (Uα′) ∈ H2,2

k (M) and there exists

U ∈ H2,2k (M) such that

Uα′ U fracamente em H2,2k (M) .

The embedding H2,2k (M) → H1,2

k (M) is compact, then up to another subsequence, we

have

Uα → U fortemente em H1,2k (M) .

Page 58: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

58

where, without loss of generality, we return to index of the original sequence. Then we

have:

I(Uα) = I(U) +1

2

∫M

(∆g(Uα − U))2 dvg + o(1) = µ+ o(1) ,

when α→ +∞. As µ is the infimum, we have µ ≤ I(U) and

limα→∞

∫M

(∆g(Uα′ − U))2 dvg = 0 .

Therefore, µ = I(U). We have then I ′(U) = 0, because I ∈ C1(H2,2k (M),R). That is,

PgU = S in the weak sense. From

−∆2gU + divg

(A(∇U)#

)= S(x)− 1

2∇UG(x, U) .

Follows from the Schauder theory that U ∈ H4,2k (M).

Affirmation 3. We affirm that, for any S ∈ Lpk(M) and G(x, t) =∑k

i,j=1Aij(x)titj,

where (Aij(x)) is positive as bilinear form and symmetric for all x ∈ M , exists a unique

U ∈ H4,pk (M) such that

−∆2gU + divg

(A(∇U)#

)= S − 1

2∇UG(x, U) .

Let (Sm)m∈N ∈ C∞k (M) be a sequence such that:

limm→∞

Sm = S

strongly in Lpk(M). For each m ∈ N, let Um ∈ C4k(M) be such that (see affirmation 2):

−∆2gUm + divg

(A(∇Um)#

)= Sm .

As Ψ is coercive and from Lp theory, we have for any m, n ∈ N:

‖Um − Un‖H4,pk (M) ≤ C ′

(‖Sm − Sn‖Lpk(M) + ‖Um − Un‖Lpk(M)

)≤ k‖Sm − Sn‖p

.

We have then that (Um) is a Cauchy sequence in H4,pk (M) and then exists U ∈ H4,p

k (M)

such that limm→∞ Um = U in H4,pk (M). We have then:

−∆2gU + divg

(A(∇U)#

)= S − 1

2∇UG(x, U) .

Now assume that V ∈ H4,pk (M) satisfies:

Page 59: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

59

−∆2gV + divg

(A(∇V )#

)= S − 1

2∇VG(x, V ) .

Subtracting the last two equalities, we obtain that

−∆2g (U − V ) + divg

(A(∇(U − V ))#

)+

1

2∇G(x, U − V ) = 0 ,

remembering that G is as defined in the statement of Affirmation 3. Follows from coer-

civity that U ≡ V :

0 ≥ Ψ(U − V ) ≥ λ

∫M

|U − V |2 dvg ,

and this proves the affirmation.

The part of existence is proven in affirmation 3 above. The estimate a priori is a

consequence of affirmation 1 and Lp theory:

‖U‖H4+r,pk (M) ≤ C

(‖S‖Hr,p

k (M) + ‖U‖Lpk(M)

)≤ C ′‖U‖Hr,p

k (M)

For the particular cases of the functions F and G, we can improve the regularity of

the solution. See the proposition below. Thus, the question of regularity is resolved.

Proposition 14. Let (M, g) be a compact Riemannian manifold of dimension n ≥ 5. Let

A be a sum of symmetric and positive (2, 0)−tensores . Let F : Rk → R be a positive C1

function and 2#-homogeneous and G : M × Rk → R a 2-homogeneous function on the

second variable of class C1, given by F (t) =∑k

i=1 |ti|2#

and G(x, t) =∑k

i=1Aij(x)titj. In

wath (Aij(x)) is symmetric and positive defined as bilinear form, for all x ∈ M . Assume

that U = (u1, . . . , uk) ∈ H2,2k (M) is a weak solution of:

−∆2gui + divg Ai((∇ui)#) +

1

2

k∑j=1

Aijuj =1

2#|ui|2

#−2ui (2.24)

i = 1, . . . , k. Then U ∈ C4k(M) and U is a solution of (2.24) in the usual sense.

Proof. Let p ≥ 1, R > 0 (defined after) and V = (v1, · · · , vk) ∈ Lpk(M). From Holder

inequality, we have

∂iF (U)χ|U |≥R vi ∈ Lr(M) where

1

r=

1

p+

4

n,

where denote here |U | = |U |2# and χ is the characterictic function. In fact, we obtain:

Page 60: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

60

‖∂iF (U)χ|U |≥R vi‖r =

(∫M

|∂iF (U)1|U |≥R vi|r dvg

) 1r

((∫M

|∂iF (U)χ|U |≥R|n4

) 4rn(∫

M

|vi|p dvg) r

p

) 1r

= ‖∂iF (U)χ|U |≥R‖n4‖vi‖p .

Follows from the regularity theory (see the previous result) that there is a unique W =

(w1, · · · , wk) ∈ H4,rk (M) such that:

Pgwi = ∂iF (U)χ|U |≥Rv

i , (2.25)

for i = 1, . . . , k. Exists C = C(p, r, n) > 0 such that

‖wi‖H4,r(M) ≤ C‖∂iF (U)χ|U |≥Rvi‖Lr(M) . (2.26)

We know that H4,r(M) → Lq(M) continuously, where 1q

= 1r− 4

n= 1

p. Then wi ∈ Lpk(M)

and exists C = C((M, g), p, r, n) > 0 such that

‖wi‖p ≤ C‖∇ F (U)χ|U |≥R‖n4‖vi‖p . (2.27)

We define the o operator Tp,R : Lp(M) → Lp(M) such that, for each v ∈ Lp(M),

Tp,R(v) = w where w is as above. From we have alread done above (2.27), Tp,r is a

continuous map. But also, note that this map is a linear map. Let vi1 and vi2 ∈ Lp(M) be

such that:

Pgwi1 = ∂iF (U)χ|U |≥Rv

i1

Pgwi2 = ∂iF (U)χ|U |≥Rv

i2

Then,

Pg(wi1) + Pg(w

i2) = Pg(w

i1 + wi2) = ∂iF (U)χ|U |≥R(vi1 + vi2)

On the other hand, assume that wi ∈ Lp(M) is the unique function such that:

Pg(wi) = ∂iF (U)χ|U |≥R(vi1 + vi2)

Thus, form the uniqueness of the solution of (2.25), we have wi = wi1 + wi2. Therefore,

the operator Tp,R is linear.

Page 61: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

61

‖Tp,R‖Lp→Lp ≤ C(p, r, n)

(∫|U |≥R

|U |2#) 4

n

.

Thus, as U ∈ L2#

k (M) exists R0 = R((M, g), p, r, n) > 0 such that

‖Tp,r‖Lp→Lp ≤1

2.

Then we have IdLp − Tp,r : Lpk(M) → Lpk(M) is linear and continuous with the inverse

linear and continuous.

As ∇F (U)χ|U |≤R ∈ L∞k (M) we have, by the regularity (see the previous result), that

for all p ≥ 2# exists U = (u1, · · · , uk) ∈ H4,pk (M) such that:

−∆2gui + divg

(Ai(∇ui)#

)+ ∂iG(x, U) = ∂iF (U)χ|U |≤R ui ,

for i = 1, ..., k. Let U = (IdLp − Tp,R)−1 (U) ∈ Lpk(M). We have

−∆2gui + divg

(Ai(∇ui)#

)+ ∂iG(x, U) = ∂iF (U)χ|U |≥Rui + ∂F (U)χ|U |≤R ui

and

−∆2g (ui − ui) + divg

(Ai(ui − ui)#

)+ ∂iG(x, U − U) = ∂iF (U)χ|U |≥Ru

i ,

for i = 1, . . . , k. Thence we have U − U = T2#,R(U) and

(Id

L2# − T2#,R

)(U) = U = (IdLp − Tp,R) (U) =

(Id

L2# − T2#,R

)(U) ,

since p ≥ 2# and U, U ∈ L2#

k (M).

As the operator(Id

L2# − T2#,R

)has inverse, we have U = U ∈ Lpk(M) for all p ≥ 2#.

From the regularity theory (Lp theory), we have U ∈ H4,pk (M) for all p ≥ 2#. From

Sobolev embedding Hr,pk (M) → C0,α

k (M) for all α ∈ (0, 1) such that α < r − np

we have

∇F (U) ∈ C0,αk (M) and for the regularity theory (Schauder theory), we have U ∈ C4

k(M).

Page 62: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

62

2.3 Bubbles Decomposition

Consider Uα =(u1α, ..., u

)solutions of the system (2.1). Note that (2.1) has a energy

functional given by:

J(U) =

∫M

((∆gU)2 + A

((∇U)#, (∇U)#

))dvg +

+

∫M

G(x, U) dvg −∫M

F (U) dvg . (2.28)

or,

J(U) =

∫M

UPg(U) dvg −∫M

F (U) dvg ,

where

Pg(U) = −∆2gU + divg

(A(∇U)#

)+

1

2∇UG(x, U) .

Remember in which follows the definition of Palais Smale sequence.

Definition 2. Let (Uα)α ∈ H2,2k (M). The sequence (Uα)α is a Palais-Smale (or PS)

sequence if:

• J(Uα) is limited;

• limα→∞ J′(Uα) = 0 em

(H2,2k (M)

)∗.

Let (xα)α∈N be a converging sequence of points in M and let (µα)α∈N ∈ R be such that

µα > 0 for all α and limα→+∞ µα = 0. Let δ ∈(

0, ig(M)

2

), where ig(M) is the injectivity

radius of (M, g). Assume that ηδ, xα ∈ C∞(M) defined by ηδ, xα = ηδ exp−1xα , at which we

consider expxα : B2δ(0) → B2δ(xα) a exponencial map in x defined in B2δ(0) (Euclidean

ball of Rn) and that ηδ ∈ C∞(Rn), ηδ ≡ 1 in Bδ(0), ηδ ≡ 0 in Rn \ B2δ(0). We define in

which follows a family of functions, called scalar bubbles or 1-bubbles:

Bα(x) = βnηδ, xα(x)

(µα

µ2α + dg(x, xα)2

)n−42

,

for all x ∈ M . In this case we say that thr points xα are the centers and the numbers µα

are the weights of (Bα)α. The constant βn is βn = (n(n− 4)(n2 − 4))n−48 .

A k−bubbles is a sequence dof maps Bj,α = (B1j,α, · · · , Bk

j,α) such that one of the

coordinates is a scalar bubbles and the others are null..

Note that we have:

Bα(x) = ηδ, xα(x)µ−n−4

2α u

(exp−1

xα (x)

µα

), (2.29)

Page 63: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

63

for all x ∈ M e

u(x) = βn

(1

1 + |x|2

)n−42

, (2.30)

for all x ∈ Rn. Note that u ∈ D22(Rn) is a extremal for:

1

A0

= infu∈D2

2(Rn)\0

∫Rn

(∆ξu)2 dx(∫Rn|u|2# dx

) 2

2#

, (2.31)

where D22(Rn)is the completamento de C∞c (Rn) com a norma ‖u‖D2

2(Rn) = ‖∆ξu‖2 .

Note that the function u in (2.30) satisfies the following equation:

∆2ξ u = u2#−1 em Rn .

The extremal for the sharp Euclidean inequality, that is, functions in D22(Rn) that

achieve the infimum in (2.31), are in the following form:

uλ, µ, x0(x) = µ

λ2 + |x− x0|2

)n−42

for all x ∈ Rn ,

where µ 6= 0, λ > 0 and x0 ∈ Rn are arbitrary.

For clarity, we are considering u ≥ 0, where the bubbles decomposition with 1-bubbles

defined as in (2.29), where u ∈ D22(Rn) ∩ C∞(Rn) is a solution of ∆2

ξu = |u|2#−2u in Rn.

The lack of strong convergence of Palais-Smale sequences for J can be described by

the 1-bubbles. The following theorem shows how fundamental they are for description of

Palais-Smale sequences. A description of Palais-Smale sequences for critical functional is

done by Struwe (see [36]) which was provided Palais-Smale sequences for critical functional

associated with an elliptical operator of second order in a subset of Rn. The initial idea

for the scalar case came from studies of Hebey and Robert which is the extension of the

associated functional to the fourth order operator, in a Riemannian manifold.

Theorem 15. [Bubbles Decomposition] Let (M, g) be a compact Riemannian manifold

of dimension n ≥ 5, Fα, F : Rk → R positive functions C1, and 2#−homogeneous and

Gα, G : M × Rk → R are continuous functions 2-homogeneous on the second variable,

and Gα is C1 such that

Fα → F em C1loc(Rk)

Gα → G em C0loc(M × Rk) .

Let Uα ∈ H2,2k (M) be a weak solution of (2.1). If the sequence (Uα)α∈N is limited in

Page 64: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

64

H2,2k (M), then exists U0 ∈ H2,2

k (M) the weak limit of Uα and, up to a subsequence, we

have:

Uα = U0 +l∑

j=1

Bj,α +Rα (2.32)

for all α > 0, where (Bj,α)α∈N , j = 1, ..., l, k-bubbles and (Rα)α∈N ⊂ H2,2k (M) is such

that

limα→∞

Rα = 0 em H2,2k (M)

Proof. Without loss of generality, we suppose that the limit U0 ∈ H2,2k (M) is trivial, that

is, U0 ≡ 0.

From limitation of Uα, up to a subsequence, exists U0 ∈ H2,2k (M) such that:

Uα U0 em H2,2k (M) .

Note that, from the convergence above, we have:∫M

∆gUα∆gΘ dvg +

∫M

A((∇Uα)#, (∇Θ)#) = o(1), ∀ Θ ∈ H2,2k (M) .

By the dominated convergence, we have:

k∑i=1

∫M

∂iFα(Uα)Θi dvg =k∑i=1

∫M

∂iF (U)Θi dvg + o(1) ,

and ∫M

Gα(x, Uα) dvg = o(1) . (2.33)

As U0 ≡ 0, we have uiα 0 in H2,2k (M).

Thus, by using the energy functional J defined in (2.28):

J(Uα) =

∫M

((∆gUα)2 + A

((∇Uα)#, (∇Uα)#

))dvg −

∫M

F (Uα) dvg + o(1)

= Lk(Uα) + o(1) ,

where

Lk(Uα) =

∫M

((∆gUα)2 + A

((∇Uα)#, (∇Uα)#

))dvg −

∫M

F (Uα) dvg .

On the other hand, we have:

Page 65: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

65

DJ(Uα)Θ =

∫M

(∆gUα∆gΘ + A

((∇Uα)#, (∇Θ)#

))dvg

+k∑i=1

∫M

∂iGα(x, Uα)Θi dvg −k∑i=1

∫M

∂iFα(Uα)Θi dvg .

We also have that

DLk(Uα)Θ = DJ(Uα)Θ−k∑i=1

∫M

∂iGα(x, Uα)Θi dvg

or,

DLk(Uα)Θ =

∫M

(∆gUα∆gΘ) + A((∇Uα)#, (∇Θ)#

)dvg −

k∑i=1

∫M

∂iF (Uα)Θi dvg .

Thus, by using (2.33), we have:

DJ(Uα)Θ =

∫M

(∆gUα∆gΘ + A

((∇Uα)#, (∇Θ)#

))dvg −

−k∑i=1

∫M

∂iFα(Uα)Θi dvg + o(1)

= DLk(Uα)Θ + o(1) .

Therefore, if (Uα)α∈N is a Palais-Smale sequence for the functional J , then (Uα)α∈N is also

a Palais-Smale sequence for Lk. As

k∑i=1

∫M

∂iF (Uα)Θi dvg = o(1) ,

and, ∫M

|uiα|2#−2uiαΘi dvg = o(1) ,

we obtain that, if (Uα)α∈N is a Palais-Smale sequence for J , then that sequence is also

Palais-Smale for Lk. We have then that (uiα)α∈N also is a Palais-Smale sequence for Lik,

with i = 1, ..., k, where

Page 66: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

66

Lik(uiα) =

∫M

((∆gu

)2+ Ai

((∇uiα)#, (∇uiα)#

))dvg −

1

2#

∫M

|uiα|2#

dvg .

Then:

DLik(uiα)Θi =

∫M

(∆gu

iα∆gΘ

i + Ai((∇uiα)#, (∇Θi)#

))dvg −

∫M

|uiα|2#−2uiαΘi dvg .

So, for each i exists a ki and a scalar bubbles(Bij,α

)α, where j = 1, ..., ki, such that, up

to a subsequence

uiα = u10 +

ki∑j=1

Bij,α +Ri

α ,

and,

Liα(uiα) =

ki∑i=1

E(uiα) + o(1) ,

where uiα ∈ D22(Rn) is a non-trivial solution of

−∆2ξ u = |u|2#−2u em Rn ,(

Bij,α

is a scalar bubbles and

E(uiα) =1

2

∫Rn

(∆gu

)2dx− 1

2#

∫Rn|uiα|2

#

dx .

We finished putting l =∑k

i=1 ki, because, for each coordinate function of the sequence

(Uα)α ∈ H2,2k (M) we have a bubbles decomposition. Thus, if Uα = (u1

α, . . . , ukα), we have:

Uα =l∑

j=1

Bj,α +Rα

where Bj,α =(B1j,α, · · · , B

kij,α

)is a k-bubbles and Rα =

(R1α, · · · , Rk

α

)is such that

Rα → 0 em H2,2k (M)

2.4 Pointwise Estimates

The bubbles decomposition enables us to add properties to the sequences of limited solu-

tions in H2,2k (M). With this result, we add the pointwise estimates for (Uα)α.

Page 67: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

67

Theorem 16 (Pointwise Estimates). Let (M, g), Gα, Fα be as in the bubbles decomposition

theorem (theorem 15). Let Uα be a limited sequence of solutions of

−∆2gU + divg

(A(∇U)#

)+∇UGα(x, U) = ∇Fα(U) em M , (2.34)

converging to 0 in H2,2k (M). Considering the bubbles decomposition in the theorem 15,

then exists, up to a subsequence, a constant C > 0, independent of α such that

(mini,j

dg(xij,α, x

))n−42

√√√√ k∑i=1

(uiα)2 ≤ C ,

for all α and for all x, where xij,α are the centers of the bubbles Bj,α. And in particular

the |Uα| are uniformly bounded in any compact subset of M \ xj,0lj=1 and uiα → 0 in

C0loc(M \ xj,0

lj=1) where xj,0 is the limit of xij,α.

Proof. We define

Φα(x) = dg(xij,α, x

)e Ψα(x) = Φα(x)

n−42

(k∑i=1

(uiα(x)

)2

) 12

,

where the xij,α are the centers of the bubbles Bj,α.

We will do the proof by contradiction. Initially, consider the sequence (yα)αormed by

the points of maximum of Ψα and such that:

Ψα(yα) = maxM

Ψα(x) e limα→+∞

Ψα(yα) = +∞ .

Up to a subsequence we can assume that:

|ui0α (yα)| ≥ |uiα(yα)| ,

for some i0 = 1, ..., k and for all i. Let

µα = |ui0α (yα)|−2

n−4 ,

then µα → 0 when α→ +∞. Then, by the definition of yα

limα→+∞

dg(xij,α, yα

)µα

= +∞ . (2.35)

In fact, to demonstrate (2.35),develop the expression of Ψα:

Page 68: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

68

Ψα(yα) =

(k∑i=1

|uiα(yα)|2) 1

2

Φα (yα)n−42

≤(k|ui0α (yα)|2

) 12 dg

(xij,α, yα

)n−42

=√k|ui0α (yα)|dg

(xij,α, yα

)n−42

=√k

(dg(xij,α, yα

)µα

)n−42

.

As Ψα(yα)→∞ when α→ +∞, we obtain (2.35).

Let 0 < δ < ig(M), where ig is the radius of injectivity of (M, g). For i = 1, ..., k on

the Euclidean ball B0 (δµ−1α ) of center 0 and radius δµ−1

α , we define the function:

wiα(x) = µn−42

α uiα(expyα(µαx)

), (2.36)

where expyα is the exponential map in yα. Given R > 0 and x ∈ B0(R) Euclidean ball

centered on 0 and radius R. By (2.35) and (2.36) we have:

|wiα(x)| ≤ µn−42

α

(k∑j=1

|ujα(expyα(µαx)

)|2) 1

2

≤ µn−42

α

Ψα

(expyα(µαx)

)Φα

(expyα(µαx)

)n−42

.

Thus,

|wiα(x)| ≤ µn−42

α

Ψα

(expyα(µαx)

)Φα

(expyα(µαx)

)n−42

, (2.37)

for all i and all α big enough. For all i, j and x on the Euclidean ball B0(R) of center 0

and radius R > 0, we obtain the inequalities:

Page 69: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

69

dg(xij,α, expyα(µαx)

)≥ dg

(xij,α, yα

)− dg

(yα, expyα(µαx)

)≥ dg

(xij,α, yα

)−Rµα

=

(dg(xij,α, yα

)Φα(yα)

− RµαΦα(yα)

)Φα(yα)

≥(

1− RµαΦα(yα)

)Φα(yα) .

From the above inequality, by (2.37) and the definitions of yα and Ψα, we obtain:

|wiα(x)| ≤ µn−42

α

Ψα

(expyα(µαx)

)Φα

(expyα(µαx)

)n−42

≤ µn−42

αΨα(yα)

Φα

(expyα(µαx)

)n−42

≤ µn−42

α k12 |ui0α (yα)| Φα(yα)

n−42

Φα

(expyα(µαx)

)n−42

≤ k12

(1− Rµα

Φα(yα)

)−n−42

.

Therefore,

|wiα(x)| ≤ k12

(1− Rµα

Φα(yα)

)−n−42

, (2.38)

for all x ∈ B0(R) and for any i = 1, . . . , k when α is big enough. In particular, from

(2.35) and (2.38), up to a subsequence, we obtain the uniforme limitation of wiα in any

compact subset of Rn for all i.

Let Wα =(w1α, ..., w

). The Wα are solutions of

−∆2gα Wα + µ2

α divg

(A(∇Wα)#

)+

1

2µ4α∇gα,UG(x,Wα) =

1

2#∇gα F (Wα) , (2.39)

where

G(x, U) = G(expyα(µαx), U

)e gα =

(exp∗yα g

)(µαx) ,

gα is the pull-back de g. Let ξ be a Euclidean metric. For each compact set K ⊂ Rn, as

µα → 0, follows that gα → ξ and, C2(K) when α→ +∞.

Page 70: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

70

Then, from the elliptic theory, follows by (2.38) that the wiα are uniformly bounded

in C2,θloc (Rn), 0 < θ < 1, for all i. In particular, up to a subsequence, we can assume that

wiα → wi in C2loc(Rn) when α → +∞. Follows that the wi are limited in Rn by (2.37)

and they are such that |wi0(0)| = 1 by construction. Furthermore we consider the wi

belonging to the D22(Rn) space and wi ∈ L2#(Rn). Let W = (w1, ..., wk) 6= 0. For all i

and R > 0, we have: ∫Byα (Rµα)

|uiα|2#

dvg =

∫B0(R)

|wiα|2#

dvgα .

Follows from dominated convergence that, for all R > 0:∫Byα (Rµα)

|uiα|2#

dvg =

∫Rn|wi|2# dx+ εR(α) , (2.40)

where εR(α) is such that:

limR→+∞

limα→+∞

εR(α) = 0 . (2.41)

By bubbles decomposition, we have

∫Byα (Rµα)

|uiα|2#

dvg =

∫Byα (Rµα)

∣∣∣∣∣ui0 +

mi∑i=1

Bij,α +Rα

∣∣∣∣∣2#

dvg

≤ 22(2#−1)

mi∑i=1

∫Byα (Rµα)

|Bij,α|2

#

dvg + o(1) .

Therefore,

∫Byα (Rµα)

|uiα|2#

dvg ≤ c

mi∑i=1

∫Byα(Rµα)

|Bij,α|2

#

dvg + o(1) , (2.42)

where o(1)→ 0 when α→ +∞ and c > 0 is independent of α and R.

Recall that the k-bubbles are vector functions where one of the coordinates functions

is a scalar bubbles and the others are null. We have, by (2.35) , that

limα→+∞

∫Byα(Rµα)

|Bij,α|2

#

dvg = 0 ,

for all R > 0 and i = 1, ..., k. From (2.40) and (2.42) we have∫Rn|wi|2# dx = εR(α) .

Using (2.41) and making α→ +∞ and R→ +∞, we have:

Page 71: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

71

∫Rn|wi|2# dx = 0 ,

for all i = 1, . . . , k. This leads to a contradiction, because W 6= 0.

Page 72: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

72

2.5 Concentracao L2

In what follows, we consider (Uα)α a sequence of solutions of

∆2gu

iα − divg(Ai(∇uiα)#) +

1

2∂iG(x, Uα) =

1

2#∂iF (Uα) . (2.43)

where G : M × Rk → R is a C1 positive function and 2-homogeneous and F : Rk → R is

a positive C1 function and 2#-homogeneous. For some of the results that follow, due to

technical details, we use particular cases of (2.43), we will comment later.

The points of blow up, or concentration of the sequence. (Uα)α has much of the

information of the sequence. This property is called L2 concentration.

Remark. We consider S the set of geometric blow-up points defined as

S =

lim

α→+∞xij,α; i = 1, . . . , l

. (2.44)

where l is as in the bubbles decomposition theorem.

Before discussing the L2 concentration, we prove some inequalities. Let A be a sum of

(2,0)-tensores. Consider also U = (u1, ..., uk) a k-map in H2,2k (M). We say that U satisfies

−∆2gui + divg(Ai(∇ui)#) + ∂iG(x, U) ≤ ∂iF (U) (2.45)

in the sense of distributions for all i = 1, ..., k and Φ = (ϕ1, ..., ϕk) in H2,2k (M)

∫M

(∆gui,∆gϕi) dvg +

∫M

Ai((∇ui)#, (∇ϕi)#

)dvg +

∫M

∂iG(x, U)ϕi ≤∫M

∂iF (U)ϕi dvg

where (∆gui,∆gϕi) is the punctual scalar product ∇ui and ∇ϕi.For the lemma below, we consider the coercivity of the operator ψ (see section 1.5).

Lemma 17. Consider Uα =(u1α, ..., u

)solution of (2.43), and let G(x, U) =

∑ki,j=1Aij(x)ui(x)uj(x),

where (Aij(x)) is positive as bilinear form and symmetric and let ψ be coercive. Exists,

then, C > 0 such that, up to a subsequence,∫M

|Uα| dvg ≤ C

∫M

|Uα|2#−1 dvg ,

for all α, where |Uα| =∑k

i=1 |uiα| and |Uα|2#−1 =

∑ki=1 |uiα|2

#−1.

Proof. Let f iα = sign(uiα) be a function given by:

f iα = χuiα>0 − χuiα<0 , (2.46)

where χA is the characteristic function of A. Then,

f iαuiα = |uiα| ,

Page 73: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

73

for all α and for all i. Note that we have |f iα| ≤ 1 for all α and for all i.

By coercivity (see section 1.5, p. 30), we have:

ψ(Uα) ≥ C‖Uα‖2H2,2k (M)

, (2.47)

where,

ψ(U) =

∫M

(∆gU)2 dvg +

∫M

A((∇U)#, (∇U)#

)dvg +

∫M

G(x, U) dvg . (2.48)

By (2.47), there exists a solution U ′α for the minimization problem consisting in finding

a minimum for ψ(U) under the constraint∫M

(fα, U) dvg = 1 where (fα, U) =∑k

i=1 fiαui

and U = (u1, ..., uk), that is, the ui are components of U and fα = (f 1α, . . . , f

kα). If λα

is the minimum of ψ(U) where U ∈ H2,2k (M) satisfies the constraint

∫M

(fα, U) dvg = 1,

follows by (2.47), that λα > 0. Let Uα = λ−1α U ′α. Then, Uα is solution of the system:

−∆2gu

iα + divg

(Ai(∇uiα)#

)+

k∑j=1

Aijujα = f iα , (2.49)

for all i and for all α, where the uiα are components of Uα and f iα are as in (2.46).

By multiplying (2.49) by uiα, integrating over M and summing on i, we obtain, together

with (2.47), the square of the norm H2,2k (M) of Uα is uniformly controlled by the L1 norm

of the |Uα|. That is, we have

‖Uα‖22 ≤ ‖Uα‖2

H2,2k (M)

≤ ‖Uα‖1

In particular, the uiα are uniformly bounded in L2. By the standard elliptic theory the uiαare in the Sobolev spaces Hq

2(M) for all q. Thence, the uiα are continuous.

Note that, if G(x, U) =∑k

i,j=1Aijui(x)uj(x), then ∂iG(x, U) =∑k

j=1Aijuj(x).

By the above discussion and by the elliptic theory, we have that exists a constant

C0 > 0 such that |uiα| ≤ C0 in M for all α and for all i. And, as F is 2#-homogeneous,

∂iF is 2# − 1 homogeneous. By (2.49) and (2.1):

Page 74: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

74

k∑i=1

∫M

|uiα| dvg =k∑i=1

∫M

uiαfiα dvg

=k∑i=1

∫M

(−∆2

guiα + divg

(Ai(∇uiα)#

)+

k∑j=1

Aijujα(x)

)uiα dvg

=k∑i=1

∫M

(−∆2

guiα + divg

(Ai(∇uiα)#

)+

k∑j=1

Aijujα(x)

)uiα dvg

≤k∑i=1

∫M

∂iF (Uα)|uiα| dvg

≤ C0

k∑i=1

∫M

∂iF (Uα) dvg

≤ C

k∑i=1

∫M

|uiα|2#−1 dvg

for all α. Thus, ∫M

|Uα| dvg ≤ C

∫M

|Uα|2#−1 dvg ,

for all α, where C > 0 does not depend on α. This finalizes the proof.

Let us consider now Uα ∈ H2,2k (M) weak solution of (2.1). Suppose that the sequence

(Uα)α is limited in H2,2k (M).

Before we prove the L2 concentration, we need the following lemma. In dimensions

greater or equal to 9 this lemma is a consequence of Holder inequality and the fact that

‖Uα‖2 → 0. In dimension 8 we will use the bubbles decomposition of the theorem 15.

Lemma 18. Let n ≥ 8, then

∫M

|Uα|2#−1 = o(1)

(∫M

|Uα|2) 1

2

,

where o(1)→ 0 when α→∞.

In what follows, we consider Bδ as the union of the balls Bxi(δ), xi ∈ S, set defined

in (2.44), where i = 1, . . . , l and l is as in the bubbles decomposition theorem.

Proof. We divide the proof into two parts.. First we prove the result when n = 8. After,

we demonstrate the case n > 8.

Let n = 8 and consider:

Page 75: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

75

∫M

|Uα|2#−1 dvg√∫

M

|Uα|2 dvg

=k∑i=1

∫M

|uiα|2#−1 dvg√∫

M

|Uα|2 dvg

,

≤k∑i=1

∫M

|uiα|2#−1 dvg√∫

M

|uiα|2 dvg

. (2.50)

Returning to the bubbles decomposition of the Uα in H2,2k (M), given by the theorem

15. Let xij,α and µij,α be the the centers and weights of the 1-bubbles (Bij,α)α involved in

the decomposition of each k-bubbles (Bα)α given by (2.32). Let R > 0 and l be as in the

bubbles decomposition theorem. We define Ωi,α(R) as the union (i = 1 to i = l) of the

geodesic balls centered in xij,α and radius Rµij,α:

Ωi,α(R) = ∪lj=1Bxij,α(Rµij,α)

We fix i = 1, ..., k. The 2# = 4 when n = 8, then we obtain, by Holder inequality:

∫M

(uiα)2#−1

dvg ≤∫

Ωi,α(R)

(uiα)2#−1

dvg +

√∫M\Ωi,α(R)

(uiα)2# dvg

√∫M

(uiα)2 dvg .

Thus,

∫M

(uiα)2#−1 dvg(∫M

(uiα)2 dvg

) 12

∫Ωi,α(R)

(uiα)2#−1

dvg√∫M

(uiα)2 dvg

+

√∫M\Ωi,α(R)

(uiα)2# dvg , (2.51)

Let ϕ ∈ C∞0 (Rn), where C∞0 (Rn) is the set of smooth functions with compact support

in Rn. Let us consider ϕij,α a function defined by the equation:

ϕij,α(x) =(µij,α)−n−4

2 ϕ((µij,α)−1 expxij,α(x)) (2.52)

By the bubbles decomposition in H2,2k (M), we obtain by a direct calculation, for any

R > 0:

Page 76: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

76

(i)

∫M\Ωij,α(R)

(Bij,α

)2#dvg = εR(α) (2.53)

(ii)

∫Ωij,α(R)

(Bij,α

)2#−1ϕij,α dvg =

∫B0(R)

(u)2#−1 ϕdx+ o(1) (2.54)

(iii)

∫Ωij,α(R)

(Bij,α

)2(ϕij,α)2#−2 dvg =

∫B0(R)

(u)2 ϕ2#−2 dvx + o(1) ,

where u is as in (2.30), Ωij,α(R) = Bxij,α

(Rµij,α), o(1)→ 0 when α→ +∞ and where

limR→+∞

limα→+∞

εR(α) = 0 , (2.55)

From (i), we obtain: ∫M\Ωi,α(R)

(uiα)2#

dvg = εR(α) (2.56)

In wath Ωi,α is as previously defined and εR(α) is such that (2.55) is valid. From now on,

let ϕ in (2.52) be such that ϕ = 1 in the B0(R). Thus,

∫Ωi,α(R)

(uiα)2#−1

dvg ≤l∑

j=1

(µij,α)n−4

2

∫Ωij,α(R)

(uiα)2#−1

ϕij,α dvg

By the bubbles decomposition and (ii):

∫Ωij,α(R)

(uiα)2#−1

ϕij,α dvg ≤ C

∫Ωij,α

(Bij,α

)2#−1ϕij,α dvg + o(1)

≤ C

∫B0(R)

u2#−1 dx+ o(1)

where o(1)→ 0 when α→ +∞ and C > 0 does not depend on α or R. In particular, we

have:

∫Ωi,α(R)

(uiα)2#−1

dvg ≤(C

∫B0(R)

u2#−1 dx+ o(1)

) l∑j=1

(µij,α)n−4

2 (2.57)

where o(1)→ 0 when α→ +∞. Independently we also have:

∫M

(uiα)2 dvg ≥∫

Ωij,α(R)

(uiα)2 dvg

≥(µij,α)n−4

∫Ωij,α(R)

(uiα)2(ϕij,α

)2#−2dvg

Page 77: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

77

Here, 2# − 2 = 2, because n = 8. Together with the bubbles decomposition, we have

∫Ωij,α(R)

(uiα)2(ϕij,α

)2#−2dvg =

∫Ωij,α(R)

(l∑

m=1

Bim,α

)2 (ϕij,α

)2#−2dvg + o(1)

≥∫

Ωij,α(R)

(Bij,α

)2 (ϕij,α

)2#−2dvg ,

By (iii) we obtain that,∫Ωij,α(R)

(uiα)2(ϕij,α

)2#−2dvg ≥

∫B0(R)

u2 dx+ o(1) .

Thence, for any j, we have:∫M

(uiα)2 dvg ≥(µij,α)n−4

(∫B0(R)

u2 dx+ o(1)

).

And we can conclude that:∫M

(uiα)2dvg ≥

(maxj=1,...,l

µij,α

)n−4(∫B0(R)

u2 dx+ o(1)

), (2.58)

where o(1)→ 0 when α→ +∞. We denote:

R(α) =

∫M

|uiα|2#−1 dvg∫

M

|uiα| dvg.

Then, from (2.51), (2.50), (2.56), (2.57), (2.58), we obtain:

lim supα→+∞

R(α) ≤ εR + C

∫B0(R)

u2#−1 dx√∫B0(R)

u2 dx

. (2.59)

where εR → 0 when R→ +∞ and C > 0 does not depend on R. Note that we have:

limR→+∞

∫B0(R)

u2#−1 dx =

∫Rnu2#−1 dx < +∞ .

On the other hand, when n = 8,

limR→+∞

∫B0(R)

u2#−1 dx = +∞ .

By equation (2.59), we obtain then that R(α) → 0 when α → +∞. Thus we obtain the

result for n = 8.

Page 78: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

78

For convenience, let Uα = ‖Uα‖−12#Uα, so that

∫M

|Uα|2#

dvg = 1.

Now we prove the result for n > 8. We will do the proof in two steps: n ≥ 12 and

9 ≤ n < 12.

Let n ≥ 12. Note that 2# − 1 = n+4n−4

= 1 + 8n−4

. Then 1 < 2# − 1 ≤ 2. Follows from

Holder inequality:

∫M

|Uα|2#−1 dvg ≤ C

(∫M

|Uα|2 dvg) 2#−1

2

, (2.60)

where C > 0 is independent of α. Thus,∫M

|Uα|2#−1 dvg(∫

M

|Uα|2 dvg) 1

2

≤ C

(∫M

|Uα|2 dvg) 2#−2

2

= o(1) . (2.61)

If 9 ≤ n < 12, then 2 < 2# − 1 < 2#. By the Holder inequality we obtain:

∫M

|Uα|2#−1 dvg ≤

(∫M

|Uα|2 dvg)n−4

8(∫

M

|Uα|2#

dvg

) 12−n8

(2.62)

≤(∫

M

|Uα|2 dvg)n−4

8

,

because ‖Uα‖2# = 1. We have then:∫M

|Uα|2#−1 dvg(∫

M

|Uα|2 dvg) 1

2

≤(∫

M

|Uα|2 dvg)n−8

8

= o(1) . (2.63)

As we have Uα → 0 in L2(M) when α→ +∞, by (2.61) and (2.63) follows the result,

that is, o(1)→ 0, when α→ +∞.

In the result which follows we consider a particular case of (2.43). We consider G :

M × Rk → R a positive C1 function 2-homogeneous, given by:

G(x, t) =k∑ij

Aijtitj

where (Aij) is symmetric and positive as bilinear form. We also consider that Ai = big,

where g is the metric and bi ∈ R (Ai is a (2,0)-tensor as in the section 1.2). Thus, in the

lemma which follows, we consider the following system:

Page 79: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

79

−∆2gu

i + bi∆gui +

k∑j=1

Aijuj = ∂iF (U) (2.64)

for i = 1, . . . , k and F : Rk → R is a positive C1 function and 2#-homogeneous.

By the lemma which follows we got a useful estimate that we use for the L2 concen-

tration.

Lemma 19. Let Uα and U0 as in the bubbles decomposition (theorem 15). Assume that,

for each i = 1, . . . , k, Ai = big, where g is the metric and bi ∈ R. For all δ > 0 exists

C > 0 such that, up to a subsequence,

maxM\Bδ

|Uα| ≤ C

∫M

(1 + |Uα|2

#−2)|Uα| dvg ,

for all α, where Bδ = Bx0(δ) is the ball of center x0 and radius δ. |Uα| =∑k

i=1 |uiα| e

|Uα|2#−2 =

∑ki=1 |uiα|2

#−2 and x0 is the limit of the centers of the 1-bubbles of which the

k-bubbles are formed.

Proof. Let B = Bx(r) be such that Bx(2r) ⊂ M \ x0 and let (Uα)α, Uα = (u1α, . . . , u

kα),

solution of:

−∆2gu

iα + bi∆gu

iα +

k∑j=1

Aijuj = ∂iF (Uα) . (2.65)

for i = 1, . . . , k. Thus, by theorem 16 and (2.64), we obtain:

| −∆2gu

iα + bi∆gu

iα| ≤ C|Uα| .

We also obtain that, considering a ≤ b2i4

:

| −∆2gu

iα + bi∆gu

iα + auiα| ≤ C ′|Uα| , (2.66)

for all α and for all i and C does not depend on α or i. Let Uα be solution of:

−∆2gu

iα + bi∆gu

iα + auiα =

∣∣−∆2gu

iα + bi∆gu

iα + auiα

∣∣ , (2.67)

for all α and all i. As

−∆2g(u

iα ± uiα) + bi∆g(u

iα ± uiα) + a(uiα ± uiα) ≥ 0 ,

follows from maximum principle (see [30]) that uiα ≥ |uiα| on M and for all α and all i. In

particular, each uiα is not negative. Note that we have:

−∆2g|Uα|+ b0∆g|Uα|+ a|Uα| ≤ C|Uα| ,

Page 80: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

80

on the ball B for all α and b0 = kmini bi. From above inequality, Uα = (u1α, . . . , u

kα) and

|Uα| =∑k

i=1 uiα. The constant C > 0 is independent of α and each uiα is not negative.

Follows from bubbles decomposition and (2.66), (2.67) that the (Uα)α are uniformly

bounded in L∞(B). We can apply the De Giorgi-Nash-Moser for the functions Uα. In

particular we have,

maxBx( r4)

|Uα| ≤ C

∫Bx( r2)

|Uα| dvg . (2.68)

where C > 0 does not depend on α. Since B is basically any ball in M \ x0, by (2.68)

we have:

maxM\Bδ

|Uα| ≤∫M

|Uα| dvg .

By (2.65), we have:

| −∆2gu

iα + bi∆gu

iα + auiα| ≤ C

(1 + |Uα|2

#−2)|Uα| . (2.69)

And by (2.66) and (2.67) we obtain that:∫M

|Uα| dvg ≤ C

∫M

|Uα| dvg .

Thus,

maxM\Bδ

|Uα| ≤ maxM\Bδ

|Uα| ≤∫M

|Uα| dvg ≤∫M

|Uα| dvg .

Therefore, we have from (2.69):

maxM\Bδ

|Uα| ≤ C

∫M

(1 + |Uα|2

#−2)|Uα| dvg .

Note that, from the above lemma, we obtain that:

maxM\Bδ

|Uα| ≤ C‖Uα‖2 .

Consider Uα = ‖Uα‖−12 Uα and B(δ) = x ∈ M ; dist(x, S) < δ. Thus we have

‖Uα‖2 = 1.

In the proof of L2 concentration which follows we will use the lemma (18). We assume

also Uα is solution of (2.64).

Theorem 20. Let Uα be solution of (2.43), where Aij(x) is positive as bilinear form and

symmetric. Then

Uα → 0 ,

Page 81: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

81

in H2,2k (M \Bδ) for all δ > 0.

Proof. Initially show that ‖Uα‖L2k(M\Bδ) → 0.

Suppose (Uα) is solution of (2.43). By the lemma 17, we obtain:∫M

|Uα| dvg ≤ C

∫M

|Uα|2#−1 dvg . (2.70)

From the above inequality, together with the above lemmas, we have:

∫M\Bδ

(Uα)2 dvg ≤(maxM\Bδ |Uα|

) ∫M\Bδ

|Uα| dvg ≤

∫M\B δ

2

|Uα|2 dvg

12 ∫

M

|Uα|2#−1 dvg .

Thence, by the lemma 18, we have ‖Uα‖L2k(M\Bδ) = o(1).

From the proof of De Giorgi-Nash-Moser (see [31]), we have∫M\Bδ

|∇uiα|2 dvg ≤ C

∫M\B δ

2

(uiα)2 dvg ,

for some constant C independent of α, but it depends on δ. Thence∫M\Bδ

|∇Uα|2 dvg ≤ C

∫M\B δ

2

(Uα)2 dvg −→ 0 ,

by the previous step.

To prove the last part, let φ be a cut-off function such that 0 ≤ φ ≤ 1, φ = 0 on the

ball B δ2

and φ = 1 in M \Bδ. By multiplying (2.64) by φ2uiα and integrating over M , we

have:

∫M

∆guiα

(∆gφ

2uiα)dvg +

∫M

Ai(∇uiα,∇(φ2uiα)

)dvg +

1

2

∫M

∂iG(x, Uα)φ2uiα dvg =

=1

2#

∫M

∂iF (Uα)φ2uiα dvg , (2.71)

for i = 1, ..., k, or

∫M

∆gUα(∆gφ

2Uα)dvg +

∫M

A(∇gUα,∇g(φ

2Uα))dvg +

∫M

G(x, U)φ2 dvg =

=

∫M

F (x, Uα)φ2 dvg .

By a direct calculation, the first term can be written as:

Page 82: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

82

∫M

∆gUα(∆gφ

2Uα)dvg =

∫M

(∆g(φUα))2 dvg +O

(‖Uα‖

H1,2k

(M\B δ

2

)),

where

‖Uα‖2H1,2k (Ω)

=

∫Ω

(|Uα|2 + |∇Uα|2

)dvg .

The other remaining terms in (2.71) can be estimated by O

(‖Uα‖2

H1,2k

(M\B δ

2

))

. Thus,

rewrite (2.71) as:

∫M

(∆g(φUα))2 dvg = O

(‖Uα‖2

H1,2k

(M\B δ

2

)),

and by the Bochner-Lichnerowicz-Weitzenbook formula

∫M

|∇2(φuiα)|2 dvg =

∫M

(∆g(φu

iα))2dvg −

∫M

Ricg(∇(φuiα),∇(φuiα)

)dvg

≤∫M

|∆g(φuiα)2|2 dvg + k

∫M

|∇(ϕuiα)|2 dvg

= O

(‖uiα‖

H1,2

(M\B δ

2

)).

Thus,

∫M\Bδ

|∇2Uα|2 dvg = O

(‖Uα‖

H1,2

(M\B δ

2

)).

Therefore,

∫M\Bδ

|∇2Uα|2 dvg ≤ C

∫M\B δ

2

(|Uα|2 + |∇Uα|2

)dvg

,

for each i = 1, . . . , k. Thus,

∫M\Bδ

|∇2Uα|2 dvg ≤ C

∫M\B δ

2

(|Uα|2 + |∇Uα|2

)dvg

,

and this last inequality converges to zero hanks to the previous step. This ends the proof

of the lemma.

We have in which follows a global estimate.

Lemma 21. Uα satisfies ‖Uα‖2 = o(1)‖∇Uα‖2, where o(1)→ 0 when α→∞.

Page 83: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

83

Proof. By the Holder inequality we have:∫Bδ

|Uα|2 dvg ≤ vol (Bδ)2∗−22∗ ‖Uα‖2

2∗ ,

where 2∗ = 2nn−2

and V ol(Bδ) is the volume of Bδ. Now, by Sobolev embedding H1,2(M) →L2∗(M), we have

‖Uα‖22∗ ≤ A

(‖∇Uα‖2

2 + ‖Uα‖22

),

where A > 0 is independent of α. By separating the integral:∫M

|Uα|2 dvg =

∫Bδ

|Uα|2 dvg +

∫M\Bδ

|Uα|2 dvg ,

and by using as two above inequalities, we obtain:∫M

|Uα|2 dvg ≤ C1

∫M\Bδ

|Uα|2 dvg + C2V ol(Bδ)2∗−22∗

∫M

|∇Uα|2 dvg ,

for all δ > 0 small enough where C1 and C2 are positive constants independent of α and

δ. As ∫M\Bδ

|Uα|2 dvg −→ 0 ,

by the proof of the previous result, we have:

1 ≤ C2V ol(Bδ)2∗−22∗ lim inf

α→∞

((∫M

|Uα|2 dvg)−1 ∫

M

|∇Uα| dvg

),

for δ small enough, whence follows the result.

As an immediate consequence of the above two results (theorem 20 and lemma 21),

we have the following.

Lemma 22. Let Uα = ‖∇Uα‖−12 Uα, then

Uα −→ 0 em H2,2k (M \Bδ) ,

for all δ > 0.

Proof. Note that, by the theorem 20 we have:

‖Uα‖H2,2k (M\Bδ) = ‖Uα‖−2

2

(∫M\Bδ

((∆Uα)2 + |∇Uα|2 + |Uα|2

)dvg

)= o(1)

And by the lemma 21:

Page 84: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

84

‖Uα‖2

‖∇Uα‖2

= o(1)

Therefore,

‖Uα‖2H2,2k (M\Bδ)

= ‖∇Uα‖−22

‖Uα‖−22

‖Uα‖−22

(∫M

((∆Uα)2 + |∇Uα|2 + |Uα|2

)dvg

)= o(1)

Page 85: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

85

2.6 Compacidade

In the previous section we studied existence of solutions for the following system

∆2gui − divg(Ai(∇ui)#) + ∂iG(x, U) = ∂iF (U) , (2.72)

where i = 1, . . . , k and U = (u1, . . . , uk). We also studied the bubbles decomposition for

Uα solution of (2.72). F : Rk → R is a positive C1 function and 2#−homogeneous and

G : M × Rk → R is a positive C1 function 2-homogeneous on the second variable. Ai is

a smooth (2,0)-tensor symmetric.

Throughout this section we use the case where Gα(x, U) =∑k

ij=1Aαij(x)ui(x)uj(x)

and Aiα = big for each i = 1, . . . , k. (A(α))α, α ∈ N is a sequence of smooth maps,

A(α) : M → M sk(R), where A(α) = (Aαij). Remember that M s

k(R) is the vector space of

the real symmetric matrices of order k × k. Consider the following system

∆2gu

i − biα∆gui +

k∑j=1

Aαij(x)uj(x) = ∂iF (U) , (2.73)

for i = 1, . . . , k. In what follows we consider that Uα = (u1α, · · · , ukα) is solution of (2.73)

and (Uα)α is a limited sequence in H2,2k (M). We assume that A(α) satisfies that there is

a C1 map, A : M → M sk(R), A = (Aij) such that

Aαij 7→ Aij em C1(M) (2.74)

when α → +∞, for all i and j. We also consider that Aiα = biαg (in (2.72)) converges to

Ai = big. The limit system, by combining (2.73) and (2.74) is:

∆2gu

i − bi∆gui +

k∑j=1

Aij(x)uj(x) = ∂iF (U) ,

We will use in this section some of the results already been demonstrated before as

the bubbles decomposition and the L2 concentration of the previous section.

Let η be a cut-off function in Rn with η = 1 on the ball B0(δ) and η = 0 out of ball

B0(2δ), where B0(r) is the Euclidean ball with center 0 and radius r. We consider ηuiα as

a function defined in Rn and with support in B0(2δ). Then, in the below lemma, we will

use a Pohozaev type identity:∫Rnxk∂k(ηu

iα)∆2(ηuiα) dx+

n− 4

2

∫Rn

(∆(ηuiα))2

dx = 0 . (2.75)

where xk is the kth coordinate of x ∈ Rk. Note that, from the lemmas 21 and 22, for

j = 0, 1, 2, we have:∫B0(2δ)\B0(δ)

|∇jUα|2 dx = o(εα) =

∫B0(2δ)

|Uα|2 dx ,

Page 86: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

86

where ε−1α o(εα)→ 0 when α→∞ and

εα =

∫M

|∇Uα|2 dvg .

Using these estimates above in (2.75), we obtain:

Lemma 23. Let Uα = (u1α, · · · , ukα) be a limited sequence in H2,2

k (M). Then we have the

following estimate:∫Rnη2(∆2uiα

)xk∂ku

iα dx+

n− 4

2

∫Rnη2uiα∆2uiα dx = o(εα) . (2.76)

where o(εα)εα→ 0 when α→ 0.

Proof. We start with the second term of (2.75). In order to simplify, uiα ≡ u. by expanding

the term (∆(ηu))2 easily we obtain that

∫Rn

(∆ (ηu))2 dx =

∫Rnη2 (∆u)2 dx+

∫Rnu2 (∆η)2 dx

− 4

∫Rn〈∇η,∇u〉 (∆η) u dx− 4

∫Rn〈∇η,∇u〉 (∆u) η dx

+ 2

∫Rnη (∆η)u (∆u) dx+ 4

∫Rn〈∇η,∇u〉2 dx .

where for the two functions ϕ and ψ, 〈∇ϕ,∇ψ〉 is the scalar product of ∇ϕ and ∇ψ.

Integrating by parts we have:

∫Rnη2 (∆u)2 dx =

∫Rnη2u∆2u dx−

∫Rn

(∆η2

)u (∆u) dx

+ 4

∫Rnη 〈∇η,∇u〉 (∆u) η dx .

By Holder inequality, for p = 1, 2, we have:

∫Rn

(∆ηp)2 u2 dx ≤(∫

|∆ηp|n2 dx

) 4n(∫

u2# dx

) (n−4)n

,

where Aδ = B0(2δ) \ B0(δ). Note that |∆pη| ≤ C where C > 0 and u ∈ L2#(Rn). Thus

we have, ∫Rn

(∆ηp)2 u2 dx = o(εα) .

Then we have:

Page 87: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

87

∫Rn|∆η2|u|∆u| dx ≤

(∫Rn

(∆η2

)2u2 dx

) 12(∫

Rn(∆u)2 dx

) 12

∫Rnη|∆η|u|∆u| dx ≤

(∫Rn

(∆η)2 u2 dx

) 12(∫

Rn(∆u)2 dx

) 12

,

and as ∆u ∈ L2(Rn), we obtain then que:∫Rn

(∆η2)u(∆u) dx = o(εα) e

∫Rnη(∆η)u(∆u) dx = o(εα)

where o(εα) is as defined above. Due to Holder inequality, we write:

∫Rn〈∇η,∇u〉2 dx ≤

∫Rn|∇η|2|∇u|2 dx ≤

(∫Aδ

|∇η|n dx) 2

n(∫

|∇u|2∗ dx) (n−2)

n

where 2∗ = 2nn−2

. Note that |∇η| ≤ C and |∇u| ∈ L2∗(Rn). Then we obtain:∫Rn〈∇η,∇u〉2 dx = o(εα) .

We can write that,

∣∣∣∣∫Rn〈∇η,∇u〉 u∆η dx

∣∣∣∣ ≤ ∫Rn| 〈∇η,∇u〉 |u|∆η| dx

√∫Rn〈∇η,∇u〉2 dx

√∫Rn

(∆η)2 u2 dx ,

we obtain that ∫Rn〈∇η,∇u〉 u∆η dx = o(εα) .

Thus, form the above estimates, we obtain that:∫Rn

(∆ (ηu))2 dx =

∫Rnη2u∆2u dx+ o(εα) . (2.77)

Now we calculate the first term of (2.75). Note that we have:

∆2(ηu) = η∆2u+ ∆η∆u− 2 〈∇η,∇∆u〉+

+u∆2η + ∆u∆η − 2 〈∇u,∇∆η〉 − 2∆ 〈∇η,∇u〉 .

Thus,

Page 88: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

88

∫Rn

∆2(ηu)xk∂k(ηu) dx

=

∫Rnη2(∆2u)xk∂ku dx+

∫Rnηu(∆2η)xk∂ku dx

+ 2

∫Rn

(∆η)(∆u)ηxk∂ku dx− 2

∫Rnη∆ 〈∇η,∇u〉xk∂ku dx

− 2

∫Rnη 〈∇u,∇∆η〉xk∂ku dx− 2

∫Rnη 〈∇η,∇∆u〉xk∂ku dx

+

∫Rnηu(∆2u)xk∂kη dx+

∫Rnu2∆2ηxk∂kη dx

+

∫Rnu∆η∆uxk∂kη dx− 2

∫Rnu(∆(〈∇η,∇u〉))xk∂kη dx

− 2

∫Rnu 〈∇η,∇∆u〉xk∂kη dx− 2

∫Rnu 〈∇u,∇∆η〉xk∂kη dx . (2.78)

Note that |∆2η| ≤ C and |x| ≤ 2δ in Aδ = B0(2δ) \ B0(δ). By Holder inequality we

obtain:

∫Aδ

u|∇u| dx ≤

√∫Aδ

|∇u|2 dx

√∫Aδ

u2 dx ,

and also

∫Aδ

|∇u|2 dx ≤ |Aδ|2n

(∫Aδ

|∇u|2∗ dx) 2

2∗

∫Aδ

u2 dx ≤ |Aδ|4n

(∫Aδ

|u|2# dx) 2

2#

.

Then from the inequality

∣∣∣∣∫Rnuη(∆2η)xk∂ku dx

∣∣∣∣ ≤ C

∫Aδ

u|∇u| dx ,

as |Aδ| ≤ C, u ∈ L2#(Rn) e |∇u| ∈ L2∗(Rn), we obtain that

∫Rnuη(∆2η)xk∂ku dx = o(εα) .

Similarly, we have

Page 89: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

89

∣∣∣∣∫Rnη(∆u)(∆η)xk∂ku dx

∣∣∣∣ ≤ C

∫Aδ

|∇u||∆u| dx

≤ C

√∫Aδ

(∆u)2 dx

√∫Aδ

|∇u|2 dx

Thence ∫Rnη(∆u)(∆η)xk∂ku dx = o(εα) (2.79)

Note that ∣∣∣∣∫Rnη 〈∇u,∇∆η〉xk∂ku dx

∣∣∣∣ ≤ C

∫Aδ

|∇u|2 dx .

Thus we obtain that ∫Rnη 〈∇u,∇∆η〉xk∂ku dx = o(εα) .

Note that ∣∣∣∣∫Rnu2(∆2η)xk∂kη dx

∣∣∣∣ ≤ C

∫Aδ

u2 dx .

Thus we obtain ∫Rnu2(∆2η)xk∂kη dx = o(εα) .

Similarly, we write that

∣∣∣∣∫Rnu(∆η)(∆u)xk∂kη dx

∣∣∣∣ ≤ C

∫Aδ

u (∆u)

≤(∫

(∆u)2, dx

) 12(∫

u2 dx

) 12

.

As previously, we have ∫Rnu(∆η)(∆u)xk∂kη dx = o(εα) . (2.80)

Noting that ∣∣∣∣∫Rnu 〈∇u,∇∆η〉xk∂kη dx

∣∣∣∣ ≤ C

∫Aδ

u|∇u| dx ,

we also obtain that

Page 90: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

90

∫Rnu 〈∇u,∇∆η〉xk∂kη dx = o(εα) .

Note that from the identity

∫M

〈∇u,∇v〉 dvg = −∫M

u∆v dvg ,

by doing u = u1u2, we obtain that

∫M

(u1 〈∇u2,∇v〉+ u2 〈∇u1,∇v〉 dvg) = −∫M

u1u2∆v dvg .

Independently, integrating by parts (by doing v = η, u1 =(xk∂ku

)η and u2 = ∆u in the

above equality and then we developed the term ∇(ηxk∂ku)), we have

∫Rnη 〈∇η,∇∆u〉 xk∂ku dx =

=

∫Rnη (∆η) (∆u) xk∂ku dx−

∫Rn

(∆u)⟨∇η,∇

(η xk∂ku

)⟩dx

=

∫Rnη (∆η) (∆u) xk∂ku dx−

∫Rn|∇η|2 (∆u) xk∂ku dx

−∫Rnη (∆u) 〈∇η,∇u〉 dx−

∫Rnη (∆u) 〈x,∇η〉∇2u dx .

But note that

∣∣∣∣∫Rn|∇η|2 (∆u) xk∂ku dx

∣∣∣∣ ≤ C

∫Aδ

|∇u||∆u| dx∣∣∣∣∫Rnη (∆u) 〈∇η,∇u〉 dx

∣∣∣∣ ≤ C

∫Aδ

|∇u||∆u| .

Then, by (2.79) we obtain that

∫Rnη 〈∇η,∇∆u〉 xk∂ku dx = o(εα)−

∫Rnη (∆u) 〈x,∇η〉∇2u dx .

Noting that |∆u| ≤√n|∇2u|, we have

∣∣∣∣∫Rnη (∆u) 〈x,∇η〉∇2u dx

∣∣∣∣ ≤ C

∫Aδ

|∇2u|2 dx .

Using the Bochner-Lichnerowicz-Weitzenbock formula, we have

Page 91: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

91

∫Rnη|∇2u|2 dx =

∫Rnη|∆u|2 dx−

∫Rnη Ric 〈∇u,∇u〉 dx

≤∫Rnη|∆u|2 dx+ k

∫Rnη|∇u|2 dx .

Thence we obtain that |∇2u|2 ∈ L2(Rn) and∫Aδ

|∇2u|2 dx = o(εα) .

Thus we obtain that ∫Rnη 〈∇η,∇∆u〉 xk ∂ku dx = o(εα) . (2.81)

Similarly,

∫Rnη (∆ 〈∇η,∇u〉) xk ∂ku dx =

∫Rn〈∇∆η,∇u〉 η xk ∂ku dx

+

∫Rnη 〈∇η,∇∆u〉 xk ∂ku dx− 2

∫Rn

⟨∇2u,∇2η

⟩η xk ∂ku dx .

Noting that ∫Rn| 〈∇∆η,∇u〉 η xk ∂ku dx| ≤ C

∫Aδ

|∇u|2 dx ,

and that

∣∣∣∣∫Rn

⟨∇2u,∇2η

⟩η xk ∂ku dx

∣∣∣∣ ≤ C

∫Aδ

|∇u||∇2u| dx

≤ C

√∫Aδ

|∇2u|2 dx

√∫Aδ

|∇u|2 dx .

By using (2.81) we obtain that∫Rnη (∆ 〈∇η,∇u〉) xk ∂ku dx = o(εα) .

Doing similar calculations, we obtain that

∫Rn

(∆ 〈∇η,∇u〉)uxk∂kη dx =

∫Rn〈∇∆η,∇u〉uxk∂kη dx

+

∫Rn〈∇η,∇∆u〉uxk∂kη dx− 2

∫Rn

⟨∇2η,∇2u

⟩uxk∂kη dx .

Page 92: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

92

But note that we have∣∣∣∣∫Rn〈∇∆η,∇u〉uxk∂kη dx

∣∣∣∣ ≤ C

∫Aδ

u|∇u| dx = o(εα) ,

and also

∣∣∣∣∫Rn

⟨∇2η,∇2u

⟩uxk∂kη dx

∣∣∣∣ ≤ C

∫Aδ

u|∇2u| dx

≤ C

√∫Aδ

|∇2u|2 dx

√∫Aδ

u2 dx = o(εα) .

Integrating by parts,

∫Rn〈∇η,∇∆u〉uxk∂kη dx

=

∫Rn

(∆η) (∆u)uxk∂kη dx−∫Rn

(∆u)⟨∇η,∇

(uxk∂kη

)⟩dx

=

∫Rn

(∆η) (∆u)uxk∂kη dx−∫Rn

(∆u) 〈∇η,∇u〉 xk∂kη dx

−∫Rnu (∆u) |∇η|2 dx−

∫Rnu (∆u) ∇2η 〈x,∇η〉 dx . (2.82)

Note that we have,∣∣∣∣∫Rn

(∆u) 〈∇η,∇u〉 xk∂kη dx∣∣∣∣ ≤ C

∫Aδ

|∇u||∆u| dx = o(εα) , (2.83)

and

∣∣∣∣∫Rnu (∆u) |∇η|2 dx

∣∣∣∣+

∣∣∣∣∫Rnu (∆u)∇2η 〈x,∇η〉 dx

∣∣∣∣ ≤ C

∫Aδ

u|∇u| dx . (2.84)

By using the above estimates (2.83) and (2.84) in the equality (2.82) and by using (2.80),

we obtain that ∫Rn

(∆ 〈∇η,∇u〉)uxk∂kη dx = o(εα) ,

and ∫Rn〈∇η,∇∆u〉uxk∂kη dx = o(εα) .

Finally, note that we have,

Page 93: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

93

∫Rnη u(∆2u

)xk∂kη dx =

∫Rn

(∆u) ∆(uη xk∂kη

)dx

=

∫Rnη(xk∂kη

)(∆u)2 dx+

∫Rnu (∆u) ∆

(η xk∂kη

)dx

− 2

∫Rn

⟨∇(η xk∂kη

),∇u

⟩(∆u) dx .

Note that we have

|∆(ηxk∂kη

)| ≤ C e |∇

(ηxk∂kη

)| ≤ C .

Thence we have

∣∣∣∣∫Rnηu(∆2u

)xk∂kη dx

∣∣∣∣ ≤ C1

∫Aδ

(∆u)2 dx

+ C2

∫Aδ

u|∆u| dx+ C3

∫Aδ

|∇u||∆u| dx .

Thus we obtain that ∫Rnηu(∆2u

)xk∂kη dx = o(εα) .

Using these above estimates in (2.78), we obtain that∫Rn

∆2 (ηu)xk∂k (ηu) dx =

∫Rnη2(∆2u

)xk∂ku dx+ o(εα) . (2.85)

Therefore, by (2.77) and (2.85) the Pohozaeh identity (2.75) it becomes∫Rnη2(∆2u

)xk∂ku dx+

n− 4

2

∫Rnη2u∆2u dx = o(εα) . (2.86)

The above estimate (2.86) we use at the following theorem.

Theorem 24. Let (M, g) be a compact Riemannian manifold locally conformally flat of

dimension n ≥ 8. Assume that the tensor Aiα = biαg converges, for all i = 1, ..., k, to the

smooth symmetric tensor Ai = big.

Consider the system in (2.73) where (A(α))α is the sequence of smooth maps A(α) :

M → M sk(R), satisfying (2.74). Let Uα be solution of (2.73) that convereges weakly to U0

in H2,2k (M). Then U0 in non-trivial if big−Ag is positive or negativr definite, for some i.

Proof. We want to show, under the conditions of theorem, that the weak limit U0 is

non-trivial. Let us assume that U0 = 0 and this leads to a contradiction.

Page 94: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

94

The Uα forms a Palais-Smale sequence for the following energy functional (see the

proof the bubbles decomposition theorem, 15):

J(U) =1

2

∫M

((∆gU)2 +G(x, U)

)dvg −

∫M

F (U) dvg .

that is defined in H2,2k (M). As already seen, up to a subsequence, the Uα have a bubbles

decomposition.

Let x ∈ S, where S is the set of the blow-up points. The (M, g) is locally conformally

flat, then we can choose δ > 0 such that g is conformal to the flat metric g = φ−4

n−4 g,

where φ is smooth and positive. Even more, we can choose δ small enough such that

S∩B(x, 4δ) = x. Note that in the Euclidean metric g = ξ, we have |∇u|2g = φ4

n−4 |∇u|2g.Let Uα = φUα, that is, uiα = φuiα, for i = 1, . . . , k. We use the conformal property of

the Paneitz-Branson operator PBg

PBg(ϕu) = ϕ2#−1PBg(u) ,

for any smooth function u and g = φ−4

n−4 g and n ≥ 5. The geometric Paneitz-Branson

operator PBg is defined by

PBgu = ∆2gu− divg(Ag du) +

(n− 4

2

)Qgu ,

where Ag is the following smooth symmetric (2,0)-tensor

Ag =(n− 2)2 + 4

2(n− 1)(n− 2)Rgg −

4

n− 2Ricg ,

Rg and Ricg denotes the scalar and Ricci curvatures respectively and Qg is the Q-curvature

Qg =1

2(n− 1)∆gRg +

n3 − 4n2 + 16n− 16

8(n− 1)2(n− 2)2R2g −

2

n− 2|Ricg |2 .

Then, by a direct calculation we have

∆2uiα − φ8

n−4 divg

((biαg − Ag)duiα

)+ 2φ

12−nn−4 (biαg − Ag)(∇uiα,∇φ) +

+ hiαuiα + φ

8n−4

k∑j=2

Aαij(x)uiα(x) = ∂iF (Uα) , (2.87)

where

hiα = −(n− 4

2

8n−4Qn

g − φn+4n−4 divg

((biαg − Ag

)dφ−1

).

Let η be a cut-off function in Rn with η = 1 on the ball B0(δ) and η = 0 out of the

ball B0(2δ), where B0(r) is the Euclidean ball with center 0 and radius r. We consider

Page 95: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

95

ηuiα as a function defined in Rn and with support in B0(2δ). We will use the Pohozaev

type identity:

∫Rnx · ∇(ηuiα)∆2(ηuiα) dx+

n− 4

2

∫Rn

(∆(ηuiα)

)2dx = 0 . (2.88)

We prove the theorem by calculating the terms in the above identity. We calculate

the parcels in terms of εα, where:

εα =

∫M

|∇Uα|2 dvg . (2.89)

Note that, from lemmas 21 and 22, for j = 0, 1, 2, we have:

∫B0(2δ)\B0(δ)

|∇jUα|2 dx = o(εα) =

∫B0(2δ)

|Uα|2 dx , (2.90)

where ε−1α o(εα)→ 0 when α→∞. Using this estimate in (2.88) (as we did previously at

the beginning of section, in the lemma 23), we have:

∫Rnη2(x · ∇uiα

)uiα∆2uiα dx+

n− 4

2

∫Rnη2uiα∆2uiα dx = o(εα) . (2.91)

Now, we multiply the equation (2.87) by η2uiα, we integrate over Rn and sum over i =

1, . . . , k:

k∑i=1

∫Rnη2uiα∆2uiα dx−

k∑i=1

∫Rnφ

8n−4η2uiα divg

((Aiα − Ag

)duiα)dx

+ 2k∑i=1

∫Rnφ

12−nn−4 η2uiα

(Aiα − Ag

)(∇uiα,∇φ) dx+

k∑i=1

∫Rnhiα(ηuiα)2

dx+

+k∑

i,j=1

∫Rnη2φ

8n−4Aαij

(uiα)2

= 2#

∫Rnη2F (Uα) dx . (2.92)

Writing φ8

n−4η2 divg ((Aiα − Ag) duiα) as aij∂ijuα+ bk∂kuα, where aij and bk are smooth

functions with support in B0(2δ). Integrating by parts we have:

∫Rn

φ8

n−4 η2uiα divg

((biαg − Ag

)duiα)dx

= −∫Rnφ

8n−4η2

(biαg − Ag

) (∇uiα,∇uiα

)dx+ o(εα) . (2.93)

We also have, by Holder:

Page 96: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

96

∣∣∣∣∫Rnhiα(ηuiα)2dx

∣∣∣∣ ≤ C

∫B0(2δ)

uiα|∇uiα| dx

≤ C

2

∫B0(2δ)

(|∇uiα|2 + uiα

)dx

Using (2.90):

∫Rnhiα(ηuiα)2dx = o(εα) =

∫Rnφ

12−nn−4 η2uiα

(biαg − Ag

) (∇uiα,∇φ

)dx

=

∫Rnφ

8n−4η2

k∑j=1

Aαij(x)(uiα)2dx . (2.94)

Replacing (2.93) and (2.94) in (2.92), we obtain:

k∑i=1

∫Rnη2uiα∆2uiα dx +

k∑i=1

∫Rnφ

8n−4η2

(biαg − Ag

) (∇uiα,∇uiα

)dx

= 2#

∫Rnη2F (Uα) dx+ o(εα) . (2.95)

In view of the first term in (2.91), we multiply (2.87) by η2 (∇uiα · x), we integrate in

Rn and we sum on i = 1, . . . , k:

k∑i=1

∫Rnη2(x · ∇uiα

)∆2uiα dx−

k∑i=1

∫Rnφ

8n−4η2

(x · ∇uiα

)divg

((biαg − Ag

)duiα)dx

+ 2k∑i=1

∫Rnφ

12−nn−4 η2

(∇uiα · x

) (Aiα − Ag

) (∇uiα,∇φ

)dx+

k∑i=1

∫Rnhiαη

2(x · ∇uiα

)uiα dx+

+k∑

i,j=1

∫Rnφ

8n−4η2

(x · ∇uiα

)Aαij(x)uiα dx =

k∑i=1

∫Rnη2(x · ∇uiα

)∂iF (Uα) dx . (2.96)

Note that, integrating by parts and by using (2.90), we have:∫Rnη2(x · ∇uiα

)(uiα)2#−1 dx = − n

2#

∫Rnη2(uiα)2# dx+ o(εα)

Thus, ∫Rnη2(x · ∇uiα

)∂iF (Uα) dx = −nk

2#

∫Rnη2(uiα)2#

dx+ o(εα) . (2.97)

Again by (2.90) we have:

Page 97: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

97

∫Rnhiαη

2(x · ∇uiα

)uiαdx = o(εα) .

Also,

∫Rnφ

8n−4η2

(x · ∇uiα

) k∑j=1

Aαij(x)uiαdx = o(εα) .

And, ∫Rnφ

12−nn−4 η2

(x · ∇uiα

) (biαg − Ag

) (∇uiα,∇φ

)dx = δO(εα) ,

when |x| ≤ 2δ in the support of the integral and |ε−1α O(εα)| ≤ C, independent of α and

δ. Proceeding as in the case (2.93) and by using the fact that aij = aji:

∫Rn

φ8

n−4 η2((biαg − Ag

)duiα)dx

=n− 2

2

∫Rnφ

8n−4η2

(biαg − Ag

) (∇uiα,∇uiα

)dx+ δO(εα) . (2.98)

Replacing (2.97) and (2.98) in (2.96), we obtain:

k∑i=1

∫Rn

η2(x · ∇uiα

)∆2uiα dx−

n− 2

2

k∑i=1

∫Rnφ

8n−4 η2

(biαg − Ag

) (∇uiα,∇uiα

)dx

= −nk∫Rnη2F (Uα) dx+ o(εα) + δO(εα) . (2.99)

Replacing (2.99) and (2.95) in (2.91), we obtain:

k∑i=1

∫Rnφ

8n−4η2

(biαg − Ag

) (∇uiα,∇uiα

)dνg = o(εα) + δO(εα) . (2.100)

Returning to the manifold, we consider η defined in M . We have:∫M

φ8

n−4η2 (bαg − Ag) (∇Uα,∇Uα) dνg = o(εα) + δO(εα) .

As big−Ag has sign for some i, biαg−Ag has sign for α big enough. Thus there exists

t > 0 such that: ∫Bx(tx)

|∇Uα|2 dνg = o(εα) + δO(εα) , (2.101)

for δ > 0 small enough and for α big enough. Now we sum (2.101) over all x ∈ S and, by

using the lemma 22, we obtain:

Page 98: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

98

εα = o(εα) + δO(εα) .

By dividing by εα and taking the limit when α → ∞, we obtain 1 ≤ Cδ, where C is

independent of δ. This is a contradiction when δ is small enough.

Page 99: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

99

Chapter

3Sharp Sobolev Vetorial Inequality of

Second Order

3.1 Extremal

The second vector inequality L2− Riemannian Sobolev states that, for any U ∈ H2,2k (M):

(∫M

F (U)dvg

) 2

2#

≤ A0

∫M

(∆gU)2 dvg +

+ B0

∫M

((A((∇gU)#, (∇gU)#

)+G(x, U)

)dvg . (3.1)

is sharp concerning the first and the second Sobolev best constant, in the sense where

neither can be reduced.

As the end of the previous chapter, we assume that in this chapterG : M ×Rk → R is

a function given by∑k

i,j=1Aijuiuj, where (Aij) is positive as linear symmetric form. We

also assume that Ai = big, bi ∈ R, for i = 1, . . . , k.

Let E(A,F,G, g) be the set of the extremal maps normalized by∫MF (U) dvg = 1

associated with (3.1).

We present the following results about compactness of extremal maps for a fixed

metric.

Theorem 25 (Existence and Compactness of Extremal Maps). Let (M, g) be a compact

Riemannian manifold and locally conformally flat. Suppose that n ≥ 8 and, for some

i = 1, . . . , k,

big − Ag > 0 ou big − Ag < 0 ,

where Ag = (n−2)2+42(n−1)(n−2)

Rg g − 4n−2

Ricg. Then, the inequality (3.1) has extremal. In addi-

tion, the set E(A,F,G, g) is compact in the C0 topology.

Proof. Remember that we are assuming that G(x, U) =∑k

i,j=1 Aijuiuj so we can use the

result of the compactness and the (2, 0)-tensor Ai is given by big, bi ∈ R. However, note

Page 100: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

100

that the demonstration that follows is valid for A being a sum of (2, 0)-tensor Ai and

G : M × Rk → R a C1 positive function and 2-homogeneous. Consider U ∈ H2,2k (M),

then:

(∫M

F (U) dvg

) 2

2#

≤ Ao∫M

(∆gU)2 dvg + B0

∫M

(A((∇U)#, (∇U)#

)+G(x, U)

)dvg .

Let (α) be sequences such that 0 < α < B0 and α→ B0.

Now consider the following functional defined in Λ:

Jα(U) =

∫M

(∆gU)2 dvg + αA−10

∫M

(A((∇U)#, (∇U)#

)+G(x, U)

)dvg ,

where

Λ =

U ∈ H2,2

k (M) :

∫M

F (U) dvg = 1

.

Consider

λα = infU∈Λ

Jα(U) .

We have that λα <1A0

. Otherwise:

Jα(U) ≥ λα ≥1

A0

, (3.2)

where U ∈ H2,2k (M). If F (U) 6= 1, just normalize it, taking 1

(∫M F (U) dvg)

12#U . By (3.2) we

have:

A0

∫M

(∆gU)2 dvg + α

∫M

(A((∇U)#, (∇U)#

)+G(x, U)

)dvg ≥

(∫M

F (U) dvg

) 2

2#

,

what is one contradiction, because 0 < α < B0. Thus, we have

λα <1

A0

.

We define

Aα = αA−10 Aiα e Gα = αA−1

0 G .

Thus

Aα → BoA−10 Ai e Gα → BoA−1

0 G .

Page 101: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

101

Note that, from hypotheses of the theorem, Aiα = biαg converges to Ai = big. The

condition λα < A−10 implies that exists a minimizer Uα ∈ Λα for λ. Moreover, Uα satisfies

the system

∆2uiα + αA−10 divg

(Aiα(∇uiα)#

)+

1

2αA−1

0 ∂iGα(x, Uα) =λα2#∂iF (Uα) .

Then we have that Uα U0 in H2,2k (M), where U0 = (u1

0, . . . , uk0). Therefore

∆2ui0 +1

2B0A−1

0 divg

(Ai(ui0)#

)+ B0A−1

0 ∂iGα(x, U0) =λ

2#∂iF (U0) . (3.3)

Note that λα → λ = A−10 . Using Uα ∈ Λα, we obtain

limα→+∞

∫M

F (Uα) dvg = 1 .

In the inequality

(∫M

F (Uα) dvg

) 2

2#

≤ Ao∫M

(∆gUα)2 dvg+B0

∫M

(A((∇Uα)#, (∇Uα)#

)+G(x, Uα)

)dvg .

we take the limit when α→ +∞. Thus

lim infα→+∞

∫M

(∆gUα)2 dvg ≥ A−10 .

then the affirmation follows noting

lim supα→+∞

∫M

(∆gUα)2 dvg ≤ lim supα→+∞

λα ≤ A−10 .

Now multiply the equation (3.3) above by ui0 and we sum from i = 1 to i = k and

then we integrate. By theorem 24 note that U0 is a nontrivial extremal and the equality

is valid in (3.1) for U0

(∫M

F (U0) dvg

) 2

2#

= A0

∫M

(∆gU0)2 dvg+B0

∫M

(A((∇U0)#, (∇U0)

)#+G(x, U0)

)dvg .

By the above, note that the set E(A,F,G, g) of the extremal with norm equal 1 is

compact.

As a direct consequence of the previous theorem, we have the following results.

Corollary 26. Let (M, g) be a compact Riemannian manifold and locally conformally

flat. Suppose that n ≥ 8 and is true one of the statements for some i = 1, . . . , k,

i. bi >(n−2)2+4

2(n−1)(n−2)Rg and Ricg > 0 in (2.73).

Page 102: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

102

ii. bi <(n−2)2+4

2(n−1)(n−2)Rg and Ricg < 0 in (2.73).

then (3.1) has extremal.

Another consequence of the above theorem is the following corollary.

Corollary 27. Let (M, g) be a Einstein compact Riemannian manifold and locally con-

formally flat and, for some i, bi <(n−2)2+4

2(n−1)(n−2). Then (3.1) has extremal.

Page 103: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

103

Chapter

4Final Considerations

We will estimate the second best constant from vector theory B0(A,F,G, g) depending

on the corresponding constant in the scalar theory B0(g). In what follows, the potential

functions F and G are taken only positive continuous and homogeneous.

In these final considerations we are assuming here that there exists a Riemannian

manifold (M, g) does not have extremal for the sharp Sobolev scalar inequality.

(∫M

|u|2# dvg) 2

2#

≤ A0(g)

∫M

(∆gu)2 dvg + B0(g)

∫M

(|∇u|2g + u2

)dvg (4.1)

where A0(g) is a constante A0 defined in (9) and B0 is defined in (11). We are assum-

ingthen that there exists a nonzero function u0 ∈ H2,2(M) such that

(∫M

|u0|2#

dvg

) 2

2#

= A0(g)

∫M

(∆gu0)2 dvg + B0(g)

∫M

(|∇u0|2g + u2

)dvg

We start this section with a proposition about the estimates for the second best con-

stant in the vector theory B0(A,F,G, g) in depending on the corresponding constant in

the scalar theory B0(g). Our examples and counterexamples will be motivated by these

estimates.

Proposition 28. Let (M, g) be a compact Riemannian manifold of dimension n ≥ 5. For

each t0 ∈ Sk−12 such that F (t0) = MF , we have

M2

2#

F B0(g)

min maxx∈M G(x, t0), CA≤ B0(A,F,G, g) ≤ M

2

2#

F B0(g)

max cA,mG

where mG = minM×Sk−12

G and

cA|∇gu|2 ≤ Ai((∇gu)#, (∇gu)#

)≤ CA|∇gu|2 . (4.2)

In particular, if exists t0 ∈ Sk−12 such that F (t0) = MF e MG = maxx∈M G(x, t0) and

the condition below (4.4) is satisfied, then

Page 104: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

104

B0(A,F,G, g) =M

2

2#

F B0(g)

mG

and, furthermore, if

(∫M

|u|2# dvg) 2

2#

≤ A0(f, h, g)

∫M

(∆gu)2 dvg +B0(f, h, g)

∫M

f(x)|∇gu|2 + h(x)|u|2 dvg(4.3)

has extremal, then (1.25) has extremal (with B0(f, h, g) = B0(1, 1, g) = B0(g)).

The condition (4.4) is the following:

MG = maxx∈M

G(x, t0) ≤ CA e cA ≤ mG (4.4)

Proof. Consider U ∈ H2,2k (M), we have

(∫M

F (U) dvg

) 2

2#

≤ M2

2#

F A0

∫M

(∆gU)2 dvg +

+ B0(A,F,G, g)

∫M

(A((∇gU)#, (∇gU)#

)+G(x, U)

)dvg

thus, taking U = ut0, with u ∈ H2,2(M), we obtain:

(∫M

|u|2# dvg) 2

2#

≤ A0

∫M

(∆gu)2 dvg +M− 2

2#

F B0(A,F,G, g) maxx∈M

G(x, t0)

∫M

|u|2 dvg

+ B0(A,F,G, g)M− 2

2#

F CA

∫M

|∇gu|2 dvg

≤ A0

∫M

(∆gu)2 dvg +

+ B0(A,F,G, g)M− 2

2#

F min

maxx∈M

G(x, t0), CA

∫M

(|∇gu|2 + |u|2

)dvg

By the definition of B0(g), then we have:

B0(A,F,G, g) ≥ M2

2#

F B0(g)

min maxx∈M G(x, t0), CA

On the other hand, we have from the proof of the proposition 8:

Page 105: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

105

(∫M

F (U) dvg

) 2

2#

≤ M2

2#

F A0

∫M

(∆gU)2 dvg +B0(g)M

2

2#

F

mG

∫M

G(x, U) dvg

+B0(g)M

2

2#

F

CA

∫M

A((∇gU)#, (∇gU)#

)dvg

≤ M2

2#

F A0

∫M

(∆gU)2 dvg +

+B0(g)M

2

2#

F

max mG, CA

∫M

(A((∇gU)#, (∇gU)#

)+G(x, U)

)dvg

for all U ∈ H2,2k (M). Then, by the definition of B0(A,F,G, g), we have:

B0 ≤M

2

2#

F B0(g)

max cA,mG

Example 1. Let (M, g) be a compact Riemannian manifold of dimension n ≥ 8

such that

B0 =n2 − 2n− 4

2n(n− 1)A0 Rg(x)

and (4.3) has extremal.

Let G : M × Rk → R, G(x, t) =∑k

i,j=1 Aij(x)|ti||tj| where Aij are nonnegative

continuous functions such that Ai0i0 > 0 does not depend on x and Aii ≥ Ai0i0 for some

i0. We have:

Ai0i0|t|2 ≤k∑i=1

Aii(x)|ti|2 ≤∑i,j

Aij(x)|ti||tj|

thus, mG = Ai0i0 .

Let F : Rk → R be a positive continuous function and 2#-homogeneous such that

F (ei0) = MF where ei0 is the i0-nth element of canonical basis de Rk. Then, by the

proposition 28:

B0(A,F,G, g) =M

2

2#

F B0(g)

Ai0i0

Let u0 ∈ H2,2(M) be a extremal function of (4.3). Then U = u0ei0 is a extremal map

of (1.25). Note that the regularity of F was not required.

Example 2. Let (M, g) be a compact Riemannian manifold of dimension n ≥ 8

such that

Page 106: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

106

B0(g) =n2 − 2n− 4

2n(n− 1)A0 Rg(x)

and (4.3) has no extremal.

Consider G : M × Rk → R, G(x, t) =∑k

i,j=1 Aij(x)|ti||tj| where Aij are nonnegative

continuous functions such that Ai0i0 > 0 does not depend on x and Aii ≥ Ai0i0 for some

i0. We have:

Ai0i0 |t|2 ≤k∑i=1

Aii(x)|ti|2 ≤∑i,j

Aij(x)|ti||tj|

thus, mG = Ai0i0 .

Let F : Rk → R be a positive continuous function and 2#-homogeneous such that

F (ei0) = MF where ei0 is the i0-nth element da canonical basis of Rk. Then, by the

proposition 28:

B0(A,F,G, g) =M

2

2#

F B0(g)

Ai0i0

Assume, by contradiction, there exists a extremal map of U0 in (1.25). The (2, 0)-

tensor Ai satisfies eqrefcCA. Then

B0(A,F,G, g)

∫M

(A((∇gU0)#, (∇gU0)#

)+

k∑i,j=1

Ai,j|uio||uj0|

)dvg =

=

(∫M

F (U0) dvg

) 2

2#

−M2

2#

F A0

∫M

(∆gU0)2 dvg

≤ M2

2#

F

k∑i=1

(∫M

|uio|2 dvg) 2

2#

−M2

2#

F A0

∫M

(∆gU0)2 dvg

≤ M2

2#

F A0(g)k∑i=1

∫M

(∆gu

io

)dvg +M

2

2#

F B0(g)k∑i=1

∫M

(|∇gu

i0|2 + |ui0|2

)2dvg

− M2

2#

F A0

∫M

(∆gU0) dvg

≤ M2

2#

F B0(g)k∑i=1

∫M

(|∇gu

i0|2 + |ui0|2

)2dvg

≤ M2

2#

F B0(g)

Ai0i0

∫M

k∑i,j=1

Ai,j|ui0||uj0| dvg +

M2

2#

F B0(g)

cA

k∑i=1

∫M

k∑i=1

Ai((∇gu

i0)#, (∇gu

i0)#)dvg

≤ M2

2#

F B0(g)

max Ai0i0 , cA

∫M

(k∑

i,j=1

Aij|ui0||uj0|+

k∑i=1

Ai((∇gu

i0)#, (∇gu

i0)#))

dvg

This implies

Page 107: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

107

k∑i=1

(∫M

|ui0|2#

dvg

) 2

2#

= A0(g)k∑i=1

∫M

(∆gu

i0

)2dvg +

+ B0(g)

∫M

(k∑i=1

|∇guio|2 +

k∑i=1

|uio|2)dvg (4.5)

Independently, follows from the scalar sharp Sobolev inequality (4.1), that

(∫M

|ui0|2 dvg) 2

2#

≤ A0(g)

∫M

(∆gu

i0

)2dvg +B0(g)

∫M

(|∇gu

i0|2 + |ui0|2

)dvg (4.6)

for each i = 1, ..., k. then, by (4.6) and (4.5), there exists j ∈ 1, ..., k such that uj0 6= 0

and,

(∫M

|uj0|2 dvg) 2

2#

= A0(g)

∫M

(∆gu

j0

)2dvg +B0(g)

∫M

(|∇gu

j0|2 + |uj0|2

)dvg

this contradicts the initial hypothesis that (4.4) has no extremal.

4.1 Final Comments

Several studies have been devoted to the study of sharp Sobolev inequalities over the

last 30 years. Ithis is mainly due to its connection with some geometric and analytic

problems. Actually, these inequalities are related, for example, with the Yamabe problem,

isoperimetric inequalities and some properties of the Ricci flow. A scalar theory of best

constants associated to the classical Sobolev inequality was then developed in parallel

to the study of some geometric problems. Important results on the validity of sharp

Sobolev inequalities, existence or non-existence of extremal maps, characterization and

compactness of extremal maps were obtained in the last decades. Initially, the focus was

to equations of order 2 mostly because of the Yamabe problem. But in recent years several

authors have studied fourth order equations and Paneitz Branson type operators.

Every scalar theory of best constants, including our results of compactness of extremal

maps, naturally arises in the vector context.Part of this thesis was then dedicated to the

best constants vector theory. We observed that some known facts of the scalar theoryare

easily extended to the vector case. On the other hand, other facts were more complex and

required the development of some ideas that are available only in scalar context. We also

provide sufficient conditions for the existence of extremal maps. A result of compactness

of extremal maps were also established.

Although the best constants vector theory developed here, now extends much of what’s

Page 108: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

108

in the scalar theory, some questions arise in this new context.. For example, we provide a

sufficient condition for the existence of extremal maps, with F and G of C1 class. From

the examples and counter-examples presented in the previous section, we expect some

results are valid for F and G only continuous. We also believe in more general results of

non-existence of extremal aplications that depend only on geometry and not the functions

F and G, as shown in Example 2 in the previous section. For this, the introduction of

the notion of critical maps provide a path in this direction (similarly to the scalar case of

second order). Another interesting question is the following:

Given functions F and G, we can guarantee that (3.1) has extremal map if and only

if, exists t0 ∈ Sk−12 , where F (t0) = MF , such that exists extremal?

The example 2 of the previous section shows that in fact it occurs in a particular case.

However, it is not clear that this occurs in general.

Finally, we list other issues that arise from the results of this thesis.

(a) Exists a Riemannian manifold such that does not exists extremal map for (1.17)?

(b) For any functions F and G, there exists a metric h conforme the given metric g such

that (3.1) has extremal map?

(c) How to guarantee the existence of extremal for any Riemannian manifold?

Page 109: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

109

Bibliography

[1] M. AMSTER, P.; CRISTINA. Existence of solutions for elliptic systems with critical

Sobolev exponent. J. Differential Equations 49, p. 13, 2002.

[2] T. AUBIN. Equations differentielles non lineaires et probleme de Yamabe concernant

la courbure scalaire. J. Math. Pure Appl. 55, p. 269-296, 1976.

[3] T. AUBIN. Problemes isoperimetriques et espaces de Sobolev. J. Differential Geom.

11 (4), p. 573-598, 1976.

[4] M. BARBOSA, E.; MONTENEGRO. Extremal maps in best constants vector theory

- Part II: Extended Lp-theory. Preprint.

[5] M. BARBOSA, E.; MONTENEGRO. Extremal maps in best constants vector theory

- Part I: Duality and compactness. Journal of Functional Analysis, 2012.

[6] T. P. BRANSON. Group representations arising from Lorentz conformal geometry.

J. Funct. Anal., 74, 199-291, 1987.

[7] P. C. CHANG, S. Y. A.; YANG. On fourth order curvature invariant. Comp. Math.

237, Spectral Problems in Geometry and Arithmetic, Ed: T. Branson, AMS, p. 9-28,

1999.

[8] S. Y. A. CHANG. On Paneitz operator - a fourth order differential operator in

conformal geometry, Harmonic Analysis and Partial Differential Equations, Essays

in honor of Alberto P. Calderon. Eds: M. Christ, C. Kening and Sadorsky, Chicgo

Lectures in Mathematics, p. 127-150, 1999.

[9] G. F. DE SOUZA. Teoria C0 Vetorial em Geometria Riemanniana e Decomposicao

em Bubbles para Aplicacoes Palais-Smale. Tese (Doutorado em Matematica), Depar-

tamento de Matematica - Universidade Federal de Minas Gerais, 2010.

[10] E. ; LEDOUX M. DJADLI, Z. ; HEBEY. Paneitz-type operators and applications.

Duke. Math. J. 104, 129-169, 2000.

[11] O. DJADLI, Z. ; DRUET. Extremal functions for optimal Sobolev inequalities on

compact manifolds. Calc. Var.,12, 59-84, 2001.

[12] E.;ROBERT F. DRUET, O.; HEBEY. Blow-up Theory for Elliptic PDEs in Rie-

mannian Geometry. Mathematical Notes,Princeton University, vol. 45, 2004.

Page 110: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

110

[13] O. DRUET. Isoperimetric inequalities on compact manifolds. Geometriae Dedicata,

217-236, 2002.

[14] E. EDMUNDS, D. E.; FORTUNATO ; JANNELLI. Critical exponents, critical di-

mensions and the biharmonic operator. Arch. Rational Mech. Anal., 112), 269-289,

1990.

[15] F. ESPOSITO, P.; ROBERT. Mountain pass critical points for Paneitz-Branson

operatos. Calc. Var. Partial Differential Equations, 15, 493-517, 2002.

[16] E. HEBEY. Sharp Sobolev inequalities of second order. J. Geom. Anal., 13, 145-162,

2003.

[17] E. HEBEY. Critical Elliptic Systems in Potential Form. Adv. Differential Equa-

tions,11, no. 5,p. 511-600, 2006.

[18] E. HEBEY. Sharp Sobolev inequalities for vector valued maps. Math. Z. 253, no. 4,

681-708, 2006.

[19] E.; M. VAUGON HEBEY. Meilleures constantes dans le theoreme d?inclusion de

Sobolev et multiplicite pour les problemes de Nirenberg et Yamabe. Indiana Univ.

Math. J. 41, no. 2, 377-407, 1992.

[20] E.; M. VAUGON HEBEY. The best constant problem in the Sobolev embeding theorem

for complete Riemannian manifolds. Duke Math. J., 79, 235-279, 1995.

[21] E.; M. VAUGON HEBEY. Meilleures constantes dans le theoreme d inclusion de

Sobolev. Ann. Inst. H. Poincare. 13, 57-93, 1996.

[22] F. HEBEY, E.;ROBERT. Coercivity and Struwe compactness for Paneitz-type oper-

ator with constant coefficients. Calc. Var. Partial Differential Equations, 13, 491-517,

2001.

[23] F. HEBEY, E.;ROBERT. Compactness and global estimates for the geometric equa-

tion in higher dimensions. E.R.A./A.M.S, 10, 134-141., 2004.

[24] F. Robert; WEN Y. HEBEY, E. Hebey ; ROBERT. Compactness and global esti-

mates for a fourth order equation os critical Sobolev growth arising from conformal

geometry. Preprint, 2004.

[25] E. H. LIEB. Sharp constants in the Hardy-Littlewood-Sobolev and related inequalities.

Ann. of Math, 118, 349-374, 1983.

[26] P. L. LIONS. The concetration-compactness principle in the calculus of variations.

The limit case I, II. Rev. Mat. Iberoamericana, 1, 206-231, 1998.

Page 111: €¦ · Acknowledgements My supervisor Ezequiel Rodrigues Barbosa and my co-supervisor Marcos Montenegro. I appreciate the encouragement and support, and by the beautiful theme proposed

111

[27] S. PANEITZ. A quadratic conformally covariant differential operator for arbitrarey

Riemannian manifolds. Preprint, 1983.

[28] Raske QING, J.; RASKE. On positive solutions to semi-linear conformally invariant

equations on locally conformally flat manifolds. Int. Math. Res. Not., Art. ID 94172,

20 pp., 2006.

[29] D. RASKE. The Yamabe problem for the Q-curvature. ArXiV, 2011.

[30] F. ROBERT. Fourth Order Equations With Critical Growth In Riemannian Geom-

etry. Personal Notes, 2009.

[31] K. SANDEEP. A Compactness type for Paneitz-Branson Operators with Critical

Nonlinearity. Differential and Integral Equations, 2005.

[32] R. SCHOEN. Lectures Notes from courses at Stanford. written by D. Pollack,

Preprint, 1988.

[33] R. SCHOEN. Varitional theory for the total scalar curvature functional for Rieman-

nian metrics and related topics, Topics in Calculus of Variations (Montecatini Terme,

1987). Lecture Notes in Math., vol. 1365, Springer-Verlag, Berlin, pp. 120-154, 1989.

[34] R. SCHOEN. On the number of constant scalar curvature metrics in a conformal

class, Differential Geometry: A symposium in honor of Manfredo do Carmo, Proc.

Int. Conf. (Rio de Janeiro, 1988). Pitman Monogr. Surveys Pure Appl. Math.,vol

52, Longman Scientific and Thecnical, Harlow, pp. 311-320, 1991.

[35] R. SCHOEN. A report on some recent progress on nonlinear problems in geometry.

Surveys in Differential Geometry, (Cambridge, Mass, 1990), Suppl. J. Diff. Geom.,

vol 1, Lehigh University, Pensylvania, pp. 201-241, 1991.

[36] M. STRUWE. A Global Compactness Result for Elliptic Boundary Value Problems

Involving Limiting Nonlinearities. Math. Z., 187, 511-517, 1984.

[37] C. A. SWASON. The best Sobolev constant. Appl. Anal., 47, no4,227-239, 1992.

[38] R.C.A.M. Van der Vorst. Best constant for the embedding of the space H2 ∩H10 (Ω)

into L2NN−4 (Ω). Differential Integral Equations, 6 , 259-276, 1993.