Upload
alfredo-perez
View
31
Download
4
Embed Size (px)
DESCRIPTION
Matemática
Citation preview
Eigenvalues on Riemannian Manifolds
Publicaes Matemticas
Eigenvalues on Riemannian Manifolds
Changyu Xia UnB
29o Colquio Brasileiro de Matemtica
Copyright 2013 by Changyu Xia
Impresso no Brasil / Printed in Brazil
Capa: Noni Geiger / Srgio R. Vaz
29o Colquio Brasileiro de Matemtica
Anlise em Fractais Milton Jara Asymptotic Models for Surface and Internal Waves - Jean-Claude Saut Bilhares: Aspectos Fsicos e Matemticos - Alberto Saa e Renato de S
Teles Controle timo: Uma Introduo na Forma de Problemas e Solues -
Alex L. de Castro Eigenvalues on Riemannian Manifolds - Changyu Xia
Equaes Algbricas e a Teoria de Galois - Rodrigo Gondim, Maria Eulalia de Moraes Melo e Francesco Russo
Ergodic Optimization, Zero Temperature Limits and the Max-Plus Algebra - Alexandre Baraviera, Renaud Leplaideur e Artur Lopes
Expansive Measures - Carlos A. Morales e Vctor F. Sirvent Funes de Operador e o Estudo do Espectro - Augusto Armando de
Castro Jnior Introduo Geometria Finsler - Umberto L. Hryniewicz e Pedro A. S.
Salomo Introduo aos Mtodos de Crivos em Teoria dos Nmeros - Jlio
Andrade Otimizao de Mdias sobre Grafos Orientados - Eduardo Garibaldi e
Joo Tiago Assuno Gomes ISBN: 978-85-244-0354-5
Distribuio: IMPA Estrada Dona Castorina, 110 22460-320 Rio de Janeiro, RJ E-mail: [email protected] http://www.impa.br
eigenvaluecm2013/9/2page 3
i
i
i
i
i
i
i
i
Contents
1 Eigenvalue problems on Riemannian manifolds 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Some estimates for the first eigenvalue of the Laplacian 5
2 Isoperimetric inequalities for eigenvalues 142.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 142.2 The Faber-Krahn Inequality . . . . . . . . . . . . . . . 162.3 The Szego-Weinberger Inequality . . . . . . . . . . . . 172.4 The Ashbaugh-Benguria Theorem . . . . . . . . . . . 192.5 The Hersch Theorem . . . . . . . . . . . . . . . . . . . 23
3 Universal Inequalities for Eigenvalues 293.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 293.2 Eigenvalues of the Clamped Plate Problem . . . . 343.3 Eigenvalues of the Polyharmonic Operator . . . 463.4 Eigenvalues of the Buckling Problem . . . . . . . . . . 57
4 Polya Conjecture and Related Results 734.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 734.2 The Krogers Theorem . . . . . . . . . . . . . . . . . . 774.3 A generalized Polya conjecture by Cheng-Yang . . 814.4 Another generalized Polya conjecture . . . . . . . . . . 85
5 The Steklov eigenvalue problems 945.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 945.2 Estimates for the Steklov eigenvalues . . . . . . . . . . 95
3
eigenvaluecm2013/9/2page 4
i
i
i
i
i
i
i
i
4 CONTENTS
Bibliography 103
eigenvaluecm2013/9/2page 1
i
i
i
i
i
i
i
i
Chapter 1
Eigenvalue problems on
Riemannian manifolds
1.1 Introduction
Let (M, g) be an n-dimensional Reimannian manifold with boundary(possibly empty). The most important operator on M is the Lapla-cian . In local coordinate system {xi}ni=1, the Laplacian is givenby
=1G
n
i,j=1
xi
(Ggij
xj
)
where (gij) is the inverse matrix (gij)1, gij = g(
xi
, xj ) are the
coefficients of the Riemannian metric in the local coordinates, andG = det(gij). In local coordinates, the Riemannian measure dv on(M, g) is given by
dv =Gdx1...dxn.
Let C(M) and set
||||21 =
M
||2 +
M
||2.
1
eigenvaluecm2013/9/2page 2
i
i
i
i
i
i
i
i
2 [CAP. 1: EIGENVALUE PROBLEMS ON RIEMANNIAN MANIFOLDS
Here and in the future, the integrations on M are always taken withrespect to the Riemannian measure on M . Let us denote by H21 (M)
ando
H21 (M) the completion of C(M) and C0 (M) with respect to
|| ||. The theory of Sobolev spaces tells us that H21 (M) =o
H21 (M)when M is complete. Our purpose is to study some eigenvalue prob-lems associated to the Laplacian operator on a compact manifold M .When M = , we consider the closed eigenvalue problem:
u+ u = 0. (1.1)
When M 6= , we are interested in the following eigenvalue prob-lems.
The Dirichlet problem:{
u = u in M,u|M = 0. (1.2)
The Neumannn problem:{
u = u in M,u
M
= 0,(1.3)
where is the unit outward normal to M .
The clamped plate problem:{
2u = u in M,u|M = u
M
= 0,(1.4)
The buckling problem:{
2u = u in M,u|M = u
M
= 0,(1.5)
The eigenvalue problem of poly-harmonic operator:{
()lu = u in M,u|M = u
M
= = l1ul1
M
= 0, l 2. (1.6)
eigenvaluecm2013/9/2page 3
i
i
i
i
i
i
i
i
[SEC. 1.1: INTRODUCTION 3
The buckling problem of arbitrary order:{
()lu = u in M,u|M = u
M
= = l1ul1
M
= 0, l 2. (1.7)
The Steklov problem of second order:{
u = 0 in M,u = u on M.
(1.8)
The Steklov problem of fourth order:{
2u = 0 in M,u = u u = 0 on M.
(1.9)
Let us denote by 1 the first non-zero eigenvalue of the above prob-lems. We can arrange the eigenvalues of these problems as follows:
0 < 1 2 +.
For many reasons in Mathematics and Physics, it is important toobtain nice estimates for the s. We will concentrate our attentionon this problem. Let us list some basic facts in this direction.
Theorem 1.1 (Weyls asymptotic formula, [97]). In each of theeigenvalue problems (1.1), (1.2), (1.3), let N() be the number ofeigenvalues, counted with multiplicity, . Then
N() n|M |n/2/(2)n (1.10)
as , where n is the volume of the unit ball in Rn and |M | isthe volume of M . In particular,
n/2k {(2)n/n}k/|M | (1.11)
as +.
There are similar asymptotic formulas for the other eigenvalueproblems above (Cf. [1], [79], [80]).
Define a space H as follows:
eigenvaluecm2013/9/2page 4
i
i
i
i
i
i
i
i
4 [CAP. 1: EIGENVALUE PROBLEMS ON RIEMANNIAN MANIFOLDS
For the closed eigenvalue problem (1.1),
H =
{f H21 (M)
M
f = 0
}. (1.12)
For the Dirichlet eigenvalue problem (1.2),
H =o
H21 (M). (1.13)
For the Neumann eigenvalue problem (1.3),
H =
{f H21 (M)
M
f = 0
}. (1.14)
A fundamental fool in the theory of eigenvalues is the
Mini-Max principle. We can find a countable orthonormal ba-sis {fi}, fi C(M) for the problems (1.1), (1.2) and (1.3) suchthat
1 = inf{
M|f |2
Mf2
| f H},
i = inf{
M|f |2
Mf2
| f H,
Mffj = 0, j = 1, , i 1
}.
(1.15)
In particular, we have the
Poincare inequality:
M
|f |2 1
M
f2, f H. (1.16)
For other eigenvalue problems above, similar mini-max principlesalso hold.
Theorem 1.2 (The Co-Area formula, [17]). Let M be a compactRiemannian manifold with boundary, f H1(M). Then for anynon-negative function g on M ,
M
g =
(
{f=}
g
|f |
)d (1.17)
eigenvaluecm2013/9/2page 5
i
i
i
i
i
i
i
i
[SEC. 1.2: SOME ESTIMATES FOR THE FIRST EIGENVALUE OF THE LAPLACIAN 5
1.2 Some estimates for the first eigenvalue
of the Laplacian
In this section, we will prove some estimates for the first eigenvalueof the Laplacian.
Theorem 1.3 ([73]). Let M be an n-dimensional complete Rie-mannian manifold with Ricci curvature RicM n 1. Then thefirst non-zero eigenvalue of the closed eigenvalue problem (1.1) of Msatisfies 1(M) n.
The proof of Theorem 1.3 can be carried out by substituting afirst eigenfunction into the Bochner formula and integrating on Mthe resulted equality (see the proof of theorem 1.6 below).
An important classical result about eigenvalue is the following
Theorem 1.4 (Chengs Comparison Theorem, [18]). Let M bean n-dimensional complete Riemannian manifold with Ricci curva-ture satisfying RicM (n 1)c and let BR(p) be an open geodesicball of radius R around a point p in M , where R < /
c, when
c > 0. Then the first eigenvalue of the Dirichlet problem (1.2) ofBR(p) satisfies
1(BR(p)) 1(BR(c)), (1.18)
with equality holding if and only if BR(p) is isometric to BR(c), whereBR(c) is a geodesic ball of radius R in a complete simply connectedRiemannian manifold of constant curvature c and of dimension n.
An immediate application of Chengs eigenvalue comparison the-orem is a rigidity theorem for compact manifolds of positive Riccicurvature.
Theorem 1.5 (The Maximal Diameter Theorem, [18]). Let Mbe an n-dimensional complete Riemannian manifold with Ricci cur-vature RicM n 1. If the diameter of M satisfies d(M) , thenM is isometric to an n-dimensional unit sphere.
Proof. Take two points p, q M so that d(p, q) ; thenB/2(p)B/2(q) = . Let f and g be the first eigenfunctions corre-
eigenvaluecm2013/9/2page 6
i
i
i
i
i
i
i
i
6 [CAP. 1: EIGENVALUE PROBLEMS ON RIEMANNIAN MANIFOLDS
sponding to the first Dirichlet eigenvalues of B/2(p) and B/2(q), re-spectively. We extend f and g on the wholeM by setting f |M\B/2(p) =g|M\B/2(q) = 0 and take two non-zero constants a and b such that
M
(af + bg) = 0
Observe that the first Dirichlet eigenvalue of an n-dimensional unithemisphere is n. The mini-max principle and Chengs comparisontheorem then imply that
n 1(M)
M|(af + bg)|2
M(af + bg)2
=a2
B/2(p)|f |2 + b2
B/2(q)
|g|2
a2
B/2(p)f2 + b2
B/2(q)
g2
=a21(B/2(p))
B/2(p)
f2 + b21(B/2(q))
B/2(q)g2
a2
B/2(p)f2 + b2
B/2(q)
g2
na2
B/2(p)
f2 + nb2
B/2(q)g2
a2
B/2(p)f2 + b2
B/2(q)
g2= n.
We conclude from the equality case of the mini-max principle andChengs comparison theorem that each of B/2(p) and B/2(q) isisometric to the n-dimensional unit hemisphere and
M = B/2(p) B/2(q)
Consequently, M is isometric to a unit n-sphere.
The maximal diameter theorem can be also used to prove theObata theorem below.
Theorem 1.6 ([76]). Let M be an n-dimensional complete Rie-mannian manifold with Ricci curvature RicM n 1. If the firstnon-zero eigenvalue of the closed eigenvalue problem (1.1) of M is n,then M is isometric to a unit n-sphere.
eigenvaluecm2013/9/2page 7
i
i
i
i
i
i
i
i
[SEC. 1.2: SOME ESTIMATES FOR THE FIRST EIGENVALUE OF THE LAPLACIAN 7
Proof. Let f be a first eigenfunction corresponding to the firsteigenvalue n of M . From the Bochner formula, we get
1
2|f |2 = |2f |2 + f,(f) + Ric(f,f) (1.19)
(f)2
n n|f |2 + (n 1)|f |2 = nf2 |f |2.
Integrating on M and noticing
M(nf2 |f |2) = 0, we conclude
that the inequalities in 1.19 should take equality sign. Thus, we have
1
2(|f |2 + f2) = 1
2|f |2 + 1
2f2
= nf2 |f |2 + ff + |f |2 = 0
and so |f |2 + f2 is a constant. Without lose of generality, we canassume that |f |2 + f2 = 1 and so
|f |1 f2
= 1.
Let p and q be points of M such that f(p) = f(q) = 1 and take aunit speed minimizing geodesic : [0, l] M from p to q. Integratingthe above equation along , we obtain
l =
ds =
|f |1 f2
ds 1
1
dt1 t2
= .
It then follows from the maximal diameter theorem that M is iso-metric to an unit n-sphere.
Remark 1.1. Let Mn be a compact Riemannian manifold withRicci curvature RicM n 1 and nonempty boundary. If the meancurvature of M is nonnegative, then the first Dirichlet eigenvalueof M satisfies 1 n with equality holding if and only if Mn isisometric to an n-dimensional unit hemisphere [82]. Similarly, if theboundary ofM is convex, then the first non-zero Neumann eigenvalueof M must satisfy 1 n with equality holding if and only if Mn isisometric to an n-dimensional unit hemisphere [34, 100].
eigenvaluecm2013/9/2page 8
i
i
i
i
i
i
i
i
8 [CAP. 1: EIGENVALUE PROBLEMS ON RIEMANNIAN MANIFOLDS
We now prove another rigidity theorem using the techniques ofeigenvalues.
Theorem 1.7 ([101]). Let M be an n-dimensional complete Rie-mannian manifold with Ricci curvature RicM n 1 and let N be aclosed minimal hypersurface which divides M into two disjoint opendomains 1 and 2. If there exists a point p M such that d(p,N),the distance from p to N , is no less than /2, then the pair (M,N) isisometric to the pair (Sn(1),Sn1(1), being Sn(1) the unit n-sphere.
Proof. Assume without lose of generality that p 1. We knowfrom d(p,N) /2 that B/2(p) 1. It then follows from the do-main monotonicity [17] that the first Dirichlet eigenvalues of B/2(p)and 1 satisfy
1(B/2(p)
) 1(1). (1.20)
On the other hand, Chengs comparison theorem tells us that
1(B/2(p)
) n (1.21)
and Reillys estimate implies that 1(1) n. Thus, the inequalitiesin (1.20) and (1.21) should be equalities. Consequently, B/2(p) = 1is isometric to an n-dimensional unit hemisphere and so N = 1 =S
n1(1) is totally geodesic. It then follows from a result of [39] that2 is also isometric to an n-dimensional unit hemisphere.
Let 1 be the least nontrivial eigenvalue of an n-dimensional com-pact manifold M and let be the corresponding eigenfunction. Bymultiplying with a constant it is possible to assume that
a 1 = infM; a+ 1 = sup
M
where 0 a() < 1 is the median of .Suppose that Mn is a compact manifold without boundary of
nonnegative Ricci curvature and of diameter d. Li-Yau [72] showedthat the first nontrivial eigenvalue satisfies
1 2
(1 + a)d2
eigenvaluecm2013/9/2page 9
i
i
i
i
i
i
i
i
[SEC. 1.2: SOME ESTIMATES FOR THE FIRST EIGENVALUE OF THE LAPLACIAN 9
and conjectured that
1 2
d2. (1.22)
Li-Yaus conjecture was proved by Zhong and Yang in [103]. Let usprovide a proof of (1.22) given by Li in [71].
Lemma 1.1. The function
z(u) =2
(arcsin(u) + u
1 u2
) u
defined on [-1, 1] satisfies
uz + z(1 u2) + u = 0; (1.23)
z2 2zz + z 0; (1.24)
2z uz + 1 0; (1.25)
and
1 u2 2|z|. (1.26)
Proof. Differentiating yields
z =4
1 u2 1, z = 4u
1 u2.
Thus (1.23) is satisfied.To see (1.24), we note that
z2 2zz + z = 4
1 u2{
4
(1 u2 + u arcsinu
) (1 + u2)
}.
Since the right hand side is an even function, it suffices to check that
4
(1 u2 + u arcsinu
) (1 + u2) 0
eigenvaluecmb2013/9/2page 10
i
i
i
i
i
i
i
i
10 [CAP. 1: EIGENVALUE PROBLEMS ON RIEMANNIAN MANIFOLDS
on [0, 1]. It is easy to see that
d
du
{4
(1 u2 + u arcsinu
) (1 + u2)
}=
4
arcsinu 2u
which is nonpositive on [0, 1]. Hence
4
(1 u2 + u arcsinu
) (1 + u2)
[
4
(1 u2 + u arcsinu
) (1 + u2)
]u=1
= 0.
Inequality (1.25) follows easily because
2z uz + 1 = 4
arcsinu+ 1 u 0.
To see (1.26), let us consider the cases 1 u 0 and 0 u 1separately. It is clearly that the inequality is valid at -1, 0 and 1.Setting
f(u) = 1 u2 4
(arcsinu+ u
1 u2
)+ 2u;
then
f = 2u 4
(2
1 u2) + 2,
f = 2 + 8u
1 u2,
and
f =8
(1 u2)3/2 .
When 1 u 0, f 0. Hence f(u) min{f(1), f(0)} = 0. Forthe case 0 u 1, f 0. Thus
f max{f (0), f (1)} = max{
2 8, 0
}= 0.
eigenvaluecm2013/9/2page 11
i
i
i
i
i
i
i
i
[SEC. 1.2: SOME ESTIMATES FOR THE FIRST EIGENVALUE OF THE LAPLACIAN 11
Therefore f(u) f(1) which proves (1.26).
Lemma 1.2. Suppose M is a compact manifold without boundaryof nonnegative Ricci curvature. Assume that a nontrivial eigenfunc-tion corresponding to the eigenvalue is normalized so that for0 a < 1, a+ 1 = supM and a 1 = infM . If u = a, then
||2 (1 u2) + 2az(u) (1.27)
where
z(u) =2
(arcsinu+ u
1 u2
) u. (1.28)
Proof. We need only to prove an estimate similar to (1.27) foru = ( a) where 0 < < 1. The lemma will follow by letting 0. By the definition of u; we have
u = (u+ a)
with u . We may assume a > 0. Consider the function
Q = |u|2 c(1 u2) 2az(u),
We can choose c large enough so that supM Q = 0. The lemmafollows if c ; for a sequence of 1, hence we may assume thatc > .
Let the maximizing point of Q be x0. We claim that |u(x0)| > 0since otherwise u(x0) = 0 and
0 = Q(x0) = c(1 u2)(x0) 2az(x0) (c a)(1 2)
by (1.26), which is a contradiction.Differentiating in the ei direction gives
1
2Qi = ujuji + cuui azui. (1.29)
We can assume at x0 that u(x0) = |u(x0)| and using Qi = 0, wehave
ujiuji u211 = (cu az)2. (1.30)
eigenvaluecm2013/9/2page 12
i
i
i
i
i
i
i
i
12 [CAP. 1: EIGENVALUE PROBLEMS ON RIEMANNIAN MANIFOLDS
Differentiating again, using the commutation formula, Q(x0) = 0,(1.26), (1.29), and (1.30), we get
0 12Q(x0) (1.31)
= ujiuji + uj(u)j + Ric(u,u) + (c az) + (cu az)u (cu az)2 + (c az)(c(1 u2) + 2az)
(cu az)(u+ a)= ac((1 u2)z + uz + u) + a22(2zz + z2 + z)
+a(c )(uz + 2z + 1) + (c )(c a).
However by (1.23), (1.24), and (1.25), we conclude that
0 ac(1 )u a22(1 )z + (c )(c a) (1.32)
ac(1 ) a22(1 )(
4
1)
+ (c )(c a)
(c+ )(1 ) + (c )2.
This implies that
c {
2 + (1 ) +
(1 )(9 )2
}.
Taking 0 one gets the desired estimate.
Theorem 1.8. ([103]) Suppose M is a compact manifold withoutboundary whose Ricci curvature is nonnegative. Let a u 0 be themedian of a normalized first eigenfunction with a+ 1 = supM anda 1 = infM ; and let d be the diameter. Then the first non-zeroeigenvalue of M satisfies
d21 2 +6
(2 1)4a2 2(1 + 0.02a2). (1.33)
Proof. Let u = a and let be the shortest geodesic from theminimizing point of u to the maximizing point with length at most
eigenvaluecm2013/9/2page 13
i
i
i
i
i
i
i
i
[SEC. 1.2: SOME ESTIMATES FOR THE FIRST EIGENVALUE OF THE LAPLACIAN 13
d. Integrating the gradient estimate (1.27) along this segment withrespect to arc-length and using oddness, we have
d1/2
ds
|u|ds1 u2 + 2az
1
0
{1
1 u2 + 2az+
11 u2 2az
}du
1
0
11 u2
{2 +
3a2z2
1 u2}du
+ 3a2( 1
0
z1 u2
)2
= +3a2
2
(2 1)4.
Remark 1.2. It has been shown by Hang-Wang [40] that if theequality holds in (1.33) then M is isometric to a circle.
Remark 1.3. Let Mn be a compact manifold with smoothboundary and nonnegative Ricci curvature. Suppose that the sec-ond fundamental form of M is nonnegative. Then the first nontrivialeigenvalue of the Laplacian with Neumann boundary conditions alsosatisfies the inequality (1.27). The proof runs the same as Lemma1.1 except that the possibility of the maximum of the test functionQ at the boundary must be handled. In fact, the boundary convex-ity assumption implies that the maximum of Q cannot occur on theboundary.
eigenvaluecm2013/9/2page 14
i
i
i
i
i
i
i
i
Chapter 2
Isoperimetric
inequalities for
eigenvalues
2.1 Introduction
In this chapter, we will prove some isoperimetric inequalities for eigen-values on manifolds which have always been important problems ingeometric analysis. Owing to the limitation on the materials, we onlyselect some of the results in the area. For more interesting results,we refer to [3] , [8], [17] and the references therein. The isoperimetricinequalities to be proved are : the Faber-Krahn inequality for the firsteigenvalue of the Dirichlet eigenvalue; the Szego-Weinberger inequal-ity for the first nontrivial Neumann eigenvalue; the Hersch theoremfor the first closed eigenvalue on a compact Riemannian surface ofgenus zero; the Ashbaugh-Benguria theorem; etc. For the conve-nience of later use, we recall now the notion of spherically symmetricrearrangement. Suppose that f is a bounded measurable functionon the bounded measurable set Rn. Consider the distributionfunction f (t) defined by
f (t) = |{x ||f(x)| > t}| (2.1)
14
eigenvaluecmb2013/9/2page 15
i
i
i
i
i
i
i
i
[SEC. 2.1: INTRODUCTION 15
where | | denotes Lebesgue measure. The distribution function canbe viewed as a function from [0,) to [0, ||] and is nonincreasing.The decreasing rearrangement f of f , is the inverse of f and isdefined by
f(s) = inf{t 0|f (t) < s}. (2.2)
It is a nonincreasing function on [0, ||]. For a bounded measurableset Rn, its spherical rearrangement is defined as the ballcentered at the origin having the same measure as . The spherically(symmetric) decreasing rearrangement f : R is defined by
f(x) = f(Cn|x|n) for x (2.3)
where Cn = n/2/
(n2 + 1
)is the volume of the unt ball in Rn. An
important fact we will use is that
f2 =
||
0
(f(s))2ds =
(f)2. (2.4)
It is known that for any function f in the Sobolev space H10 (),f H10 () and
|f|2
|f |2. (2.5)
For two nonnegative measurable functions f and g on we have
fg
fg. (2.6)
Let us recall the notion of spherically (symmetric) increasing rear-rangement, which we denote by a lower . The definition is almostidentical to that of spherically decreasing rearrangement, except thatg should be radially increasing (in the weak sense) on
. In thiscase, we have
fg
fg. (2.7)
eigenvaluecmb2013/9/2page 16
i
i
i
i
i
i
i
i
16 [CAP. 2: ISOPERIMETRIC INEQUALITIES FOR EIGENVALUES
2.2 The Faber-Krahn Inequality
In this section, we will prove the Faber-Krahn inequality which is oneof the oldest isoperimetric inequalities for an eigenvalue.
Theorem 2.1 (Faber-Krahn [36],[61]). For a bounded domain Rn, the first Dirichlet eigenvalue satisfies
1() 1() (2.8)
with equality if and only if = .
Proof. Let u1 be a first Dirichlet eigenfunction for . We havefrom (2.4), (2.5) and the mini-max principle that
1() =
|u1|2u21
(2.9)
=
|u1|2
(u1)
2
|u1|2
(u1)2
1().
For the characterization of the case of equality, we refer to [56].
The Faber-Krahn inequality is valid for more general manifolds.Let M be an n-dimensional complete Riemannian manifold and for afixed R, let M be the complete simply connected n-dimensionalspace form of constant sectional curvature . To each bounded do-main in M , associate the geodesic ball D in M satisfying
|| = |D|. (2.10)
If > 0 then only consider those for which || < |M|.
Theorem 2.2. If, for all such in M , equality (2.10) impliesthe isoperimetric inequality
|| |D|, (2.11)
eigenvaluecm2013/9/2page 17
i
i
i
i
i
i
i
i
[SEC. 2.3: THE SZEGO-WEINBERGER INEQUALITY 17
with equality in (2.11) if and only if is isometric to D, then we alsohave, for every bounded domain in M , that equality (2.10) impliesthe inequality for the first Dirichlet eigenvalue
1() 1(D), (2.12)
with equality holding if and only if is isometric to D.
For a proof of Theorem 2.2, we refer to [17].
2.3 The Szego-Weinberger Inequality
In this section, we prove the Szego-Weinberger inequality which is acounterpart to the first non-zero Neumann eigenvalue of the Faber-Krahn inequality.
Theorem 2.3 ([96]). Let be a bounded domain in Rn. Thenthe first non-zero Neumann eigenvalue of satisfies
1() 1() (2.13)
with equality holding if and only if = .
Proof. Let R be the radius of and let g be the solution of theequation
{g + n1r g
n1r2 g + 1()g = 0g(0) = 0, g(R) = 0
(2.14)
By a topological argument, we can take as trial functions Pi, suchthat
Pi = 0 for i = 1, , n, with
Pi(x) = h(r)xir,
where the xis are Cartesian coordinates, x = (x1, , xn) Rn, r =|x|, and
h(r) =
{g(r) for 0 r Rg(R) for r R.
eigenvaluecm2013/9/2page 18
i
i
i
i
i
i
i
i
18 [CAP. 2: ISOPERIMETRIC INEQUALITIES FOR EIGENVALUES
Observe that by an appropriate choice of sign, g(r) is increasing on[0,R] and hence that h is everywhere nondecreasing for r 0. Bysubstituting our trial functions Pi into the mini-max inequality for1, we find
1()
P 2i
|Pi|2.
Summing this in i for 1 i n, we arrive at
1()
ni=1 |Pi|2
ni=1 P
2i
(2.15)
=
(h(r)2 + n1r2 h(r)
2)
h(r)2
=
A(r)
h(r)2
where
A(r) = h(r)2 +n 1r2
h(r)2. (2.16)
A(r) is easily seen to be decreasing for 0 r R by differentiatingand using the differential equation (2.14). One finds
A(r) = 2(1()hh + (n 1)(rh h)2/r3) < 0, 0 < r < R.(2.17)
Also, A(r) = (n 1)g(R)2/r2 for r R shows that A is decreasingfor r > R. Since A is continuous for all r 0, it is also decreasingthere. Observe that
A(r)
A(r) (2.18)
since the volumes integrated over are the same in both cases, while inpassing from the left to right hand sides we are exchanging integratingover \ for integrating over \ which are sets of equal volume.Since A is (strictly) decreasing this clearly increases the value of theintegral unless = , when equality obtains. Similarly we find that
h(r)2
h(r)2 (2.19)
eigenvaluecm2013/9/2page 19
i
i
i
i
i
i
i
i
[SEC. 2.4: THE ASHBAUGH-BENGURIA THEOREM 19
since h is nondecreasing. Thus we arrive at
1() A(r)
h(r)2
= 1(), (2.20)
since each Pi is precisely a Neumann eigenfunction of with eigen-value 1(BR) for the domain BR =
. This completes the proofof the Szego-Weinberger inequality, including the characterization ofthe case of equality.
2.4 The Ashbaugh-Benguria Theorem
In this section we consider the sharp upper bound for 2/1 for theDirichlet eigenvalue problem proved by Ashbaugh-Benguria. In 1955and 1956, Payne, Polya and Weinberger [77], [78], proved that
21
3 for R2
and conjectured that
21
21
disk
=j21,1j20,1
with equality if and only if is a disk and where jp,k denotes thekth positive zero of the Bessel function Jp(t). For general dimensionn 2, the analogous statements are
21
1 + 4n
for Rn,
and the PPW conjecture
21
21
nball
=j2n/2,1
j2n/21,1, (2.21)
with equality if and only if is an n-ball. This important conjecturewas proved by Ashbaugh-Benguria (see [5], [6], [7]).
eigenvaluecm2013/9/2page 20
i
i
i
i
i
i
i
i
20 [CAP. 2: ISOPERIMETRIC INEQUALITIES FOR EIGENVALUES
We proceed now with the proof of (2.21). Let us start from thevariational principle for 2
2() = minH10 (),06=u1
||22
, (2.22)
which, by integration by parts, leads to
2() 1() (2.23)
|P |2u21P 2u21
, Pu1 H10 (),
Pu21 = 0, P 6= 0.
To get the isoperimetric result out of (2.23), one must make veryspecial choices of the function P , in particular, choices for which(2.23) is an equality if is a ball. Thus we shall use n trial functionsP = Pi, such that
Piu
21 = 0 for i = 1, , n where
Pi = g(r)xir
(2.24)
and
g(r) =
{f(r) = right radial function on BR for 0 r R,f(R) for r R. (2.25)
The right R in this case turns out to be the unique R such that1(BR) = 1(). Substituting Pi into (2.23) and summing on i, wefind
2() 1() B(r)u21
f(r)2u21
(2.26)
where
B(r) = f (r)2 +n 1r2
f(r)2. (2.27)
Now the equation (2.26) does not depend on the P is and so we arein a position to define the function f . The idea is to take f as aproperly quotient of Bessel functions so that the equality occur if is a ball in Rn. This motivates the choice of :
f(r) = w(r), (2.28)
eigenvaluecm2013/9/2page 21
i
i
i
i
i
i
i
i
[SEC. 2.4: THE ASHBAUGH-BENGURIA THEOREM 21
where
w(x) =
{jn/2(x)
jn/21(x), if 0 x < 1,
w(1) limx1 w(x), if x 1,(2.29)
with = jn/21,1, = jn/2,1 and =1/.
Lemma 2.1. The equality occurs in (2.26) when is a ball with1 as the first Dirichlet eigenvalue and f is given by (2.28).
Proof. If S1 is a closed ball of Rn of appropriate radius centered
in the origin in which the problem{
z = z in S1,z|S1 = 0.
(2.30)
has 1 as the first eigenvalue, then
S1 = {x Rn; |x| /1 = 1/}.
The second eigenvalue of the problem (2.30) is 2 =2
21. The firsteigenfunction of S1 is
z(x) = c|x|1n/2jn/21(1|x|),
and the eigenfunctions corresponding to 2 are:
fi(x) = c|x|1n/2jn/2(2|x|)
xi|x| , i = 1, , n,
where c is a non-zero constant.Let
Q(r) =
jn/2
(2r
)
jn/21(
1r), if 0 r < 1/,
limr1/jn/2
(2r
)
jn/21(
1r), if r 1/,
(2.31)
Observe that Q(r) = w(r) = g(r) and let
Qi(x) = Q(|x|)xi|x| = g(r)
xi|x| ,
eigenvaluecm2013/9/2page 22
i
i
i
i
i
i
i
i
22 [CAP. 2: ISOPERIMETRIC INEQUALITIES FOR EIGENVALUES
then
S1
Qiz2 = 0, i = 1, , n
and Qiz are eigenfunctions of 2. Thus, we have
2 =
S1
|(Qiz)|2S1
(Qiz)2, i = 1, , n.
Summing over i and simplifying, we get
2 1 =
S1
((f (r))2 + (n 1) f
2(r)r2
)z2
S1f2(r)z2
. (2.32)
This completes the proof of Lemma 3.1.Substituting (2.28) into (2.26), we get
2 1 1B(r)u21
w2(r)u21
(2.33)
where
B(x) = (w(x))2 + (n 1)w2(x)
x2. (2.34)
From the definition of w and the properties of Bessel functions one canprove that w(t) is nondecreasing and B(t) is non-increasing. There-fore, we have
B(r)u21
B(r)(u1)
2
B(r)(u1)
2 (2.35)
and
w(r)2u21
w(r)(u
1)
2
w(r)(u1)
2 (2.36)
In order to continue the proof, we need a result of Chiti: If c is chosenso that
u21 =
u21 =
S1
z2, (2.37)
eigenvaluecm2013/9/2page 23
i
i
i
i
i
i
i
i
[SEC. 2.5: THE HERSCH THEOREM 23
then
f(r)(u1)
2
S1
f(r)z2, (2.38)
if f is increasing, and the reverse inequality if f is decreasing. Itfollows from (2.38) and the monotonicity properties of B and w that
B(r)(u1)
2
S1
B(r)z2 (2.39)
and
w(r)(u1)
2
S1
w(r)2z2. (2.40)
Combining (2.33), (2.35), (2.36), (2.39), (2.40), and using the defini-tion of z, we finally get
2 1 1
S1B(r)z2
2
S1w2(r)z2
=12
(2 2). (2.41)
From here the inequality
21
j2n/2,1
j2n/21,1(2.42)
follows immediately. Also, it is clear from the proof that the equalityholding in (2.42) if and only if is a ball.
2.5 The Hersch Theorem
In 1974, Hersch proved an isoperimetric inequality for the first non-trivial eigenvalue on the 2-dimensional sphere S2.
Theorem 2.5 ([48]). For any metric on S2, the first non-trivialeigenvalue satisfies
1 8
A(S2). (2.43)
eigenvaluecm2013/9/2page 24
i
i
i
i
i
i
i
i
24 [CAP. 2: ISOPERIMETRIC INEQUALITIES FOR EIGENVALUES
Proof. For any metric ds2 on S2, we can construct a conformal map : (S2, ds2) (S2, ds20), here ds20 denotes the standard metric on S2.From the mini-max principle, we have
1 = infS2
fdv=0
S2|f |2dv
S2f2dv
, (2.44)
where dv is the area element with respect to ds2. Take the coordi-nate functions xi(i = 1, 2, 3) on (S2, ds20); then x
i , i = 1, 2, 3, arefunctions on (S2, ds2).
Observe that is a conformal map and that in the case of surfaces,the Dirichlet integral of a function is a conformal invariant. Thus wehave
S2
|(xi )|2dv =
S2
|xi|2dv =
S2
xixi = 2
S2
(xi)2 =8
3.
Since
Area(S2) =
S2
dv =
3
i=1
S2
(xi )2dv,
there exists at least one i such that
S2
(xi )2dv Area(S2)
3.
Also, we can choose satisfying
S2xi dv = 0 [85]. Thus
1
S2|(xi )|2dv
S2(xi )2dv
8
A(S2). (2.45)
For the discussion of equality case, we refer to [48].
Remark 2.1. S2 is a Riemann surface of genus g = 0. ForRiemannian surface g of genus g > 0, Yang-Yau obtained a similarresult.
Theorem 2.6 ([99]). For any metric on g, the first eigenvaluesatisfies
1 8(1 + g)
|g|(2.46)
eigenvaluecm2013/9/2page 25
i
i
i
i
i
i
i
i
[SEC. 2.5: THE HERSCH THEOREM 25
Remark 2.2. Herschs theorem cant be generalized directly tothe case of higher dimensions [86]. That is, one cant expect that
1Vol(M)2/n C,
with a constant depending only on n. It must depend also on othergeometric invariants of M .
Here is an interesting application of Herschs theorem.
Theorem 2.7 ([19]). Suppose that M is homeomorphic to S2 and1, 2, 3 are three first eigenfunctions such that their square sum isa constant. Then M is actually isometric to a sphere with constantsectional curvature.
Proof. The assumption of Theorem 2.7 says that
{i + 1(M)i = 0, i = 1, 2, 3,3
i=1 2i = c, c is a constant.
Thus,
0 =
(3
i=1
2i
)= 2
3
i=1
|i|2 + 23
i=1
ii
= 2
3
i=1
|i|2 21(M)3
i=1
2i
which gives
3
i=1
|i|2 = c1(M). (2.47)
Taking the Laplacian of both sides of (2.47) and using the Bochner
eigenvaluecm2013/9/2page 26
i
i
i
i
i
i
i
i
26 [CAP. 2: ISOPERIMETRIC INEQUALITIES FOR EIGENVALUES
formula, we get
0 =1
2
3
i=1
|i|2 (2.48)
=3
i=1
|2i|2 +3
i=1
i (i) +3
i=1
Ric(i,i)
=
3
i=1
|2i|2 1(M)23
i=1
2i +K
3
i=1
|i|2
3
i=1
|i|22
c1(M)2 +Kc1(M)
= 1(M)2/2 +Kc1(M),
where K is the sectional curvature of M . Thus we have
1(M) 2K. (2.49)
Integrating (2.49) and using the Gauss-Bonnet formula, we have
1(M) area(M) 8. (2.50)
Combining (2.50) and Herschs theorem we know that M is a 2-sphere.
We have an isoperimetric upper bound for the first eigenvalue ofthe Laplacian of a closed (compact without boundary) hypersurfaceembedded in Rn.
Theorem 2.8 ([92]). Let M be a connected closed hypersurfaceembedded in Rn(n 3). Let be the region bounded by M . Denoteby V and A the volume of and the area of M , respectively. Thenthe first non-zero eigenvalue 1 of the Laplacian acting on functionson M satisfies
1 (n 1)AnV
(nV
)1/n. (2.51)
with equality holding if and only if M is an (n 1)-sphere.
eigenvaluecm2013/9/2page 27
i
i
i
i
i
i
i
i
[SEC. 2.5: THE HERSCH THEOREM 27
Proof. Let us denote by x1, , xn, the coordinate functions onR
n. By choosing the coordinates origin properly, we can assume that
M
xi = 0, i = 1, , n.
Since M cannot be contained in any hyperplane, each xi is not a con-stant function, i = 1, , n. It follows from the Poincare inequalitythat for each fixed i {1, , n}
1
M
x2i
M
|xi|2,
with equality holding if and only if xi = 1xi.Summing over i from 1 to n, we get
1
M
n
i=1
x2i
M
n
i=1
|xi|2 =
M
(n 1) = (n 1)A, (2.52)
with equality if and only if
xi = 1xi, i {1, , n}. (2.53)
Take a ball B in Rn of radius R centered at the origin so that vol(B) =V ; then
R =
(V
n
)1/n.
By using the weighted isoperimetric inequality proved in [14], we have
M
n
i=1
x2i
B
n
i=1
x2i (2.54)
= area(B) R2
= nV
(V
n
)1/n.
Substituting (2.54) into (2.52), one gets (2.51). If the equality holds in(2.51), then the inequalities (2.52) and (2.54) must take equality sign.
eigenvaluecm2013/9/2page 28
i
i
i
i
i
i
i
i
28 [CAP. 2: ISOPERIMETRIC INEQUALITIES FOR EIGENVALUES
It follows that the position vector x = (x1, , xn) when restrictedon M satisfies
x =: (x1, ,xn) = 1(x1, , xn).
Also, it is well known that
x = (n 1)H,
whereH is the mean curvature vector of M . Consider now the func-
tion g = |x|2 : M R. Observing that H is normal to M , we inferthat
wf = 2w, x = 2(n 1)1
w,H = 0, w (M).
Thus f is a constant function and so M is a hypersphere.
eigenvaluecmb2013/9/2page 29
i
i
i
i
i
i
i
i
Chapter 3
Universal Inequalities
for Eigenvalues
3.1 Introduction
Payne, Polya and Weinberger proved that the Dirichlet eigenvaluesof the Laplacian for R2 satisfy the bound [77], [77].
k+1 k 2
k
k
i=1
i, k = 1, 2, (3.1)
This result easily extends to Rn as
k+1 k 4
kn
k
i=1
i, k = 1, 2, (3.2)
Many interesting generalizations of (1.3) have been done during thepast years, e. g., in [3], [4], [9], [20], [21], [22], [23], [25], [26], [27], [29],[30], [33], [41], [42], [43], [44], [45], [46], [47], [50], [52], [68], [69], [90],[98]. In 1991, Yang [98] proved the following much stronger result:
29
eigenvaluecm2013/9/2page 30
i
i
i
i
i
i
i
i
30 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
Theorem 3.1. The Dirichlet eigenvalues of the Laplacian of Rn satisfy the inequality
k
i=1
(k+1 i)(k+1
(1 +
4
n
)i
) 0, for k = 1, 2, . (3.3)
The inequality (3.3), as observed by Yang himself, and as laterproved, e. g., in [3], [4], [9], is the strongest of the classical inequal-ities that are derived following the scheme devised by Payne-Polya-Weinberger. Yangs inequality provided a marked improvement foreigenvalues of large index. Recently, some Yang type inequalities oneigenvalues of the clamped plate problem, the buckling problem, thepolyharmonic operator and some other type eigenvalue problems havebeen proved. This chapter is devoted to prove some of the univer-sal inequalities in this subarea. Since the method in proving Yangsinequality has been widely generalized in obtaining various universalinequalities for eigenvalues, we end this section by proving Yangsinequality.
Proof of Theorem 3.1. Let uk be the orthonormal eigenfunctioncorresponding to the kth eigenvalue k, i.e. uk satisfies
uk = kuk, in uk| = 0,uiuj = ij .
(3.4)
Let x1, , xn be the standard coordinate functions in Rn. For anyfixed p = 1, , n, put g = xp and define i by
i = gui k
j=1
aijuj , aij =
guiuj = aji. (3.5)
It is easy to see that
iuj = 0, for i, j = 1, , k. (3.6)
Letting
bij =
ujg ui,
eigenvaluecm2013/9/2page 31
i
i
i
i
i
i
i
i
[SEC. 3.1: INTRODUCTION 31
from Greens formula, we derive
jaij =
g(uj)ui = 2bij + iaij
and so
2bij = (i j)aij . (3.7)Since
i = igui + 2g ui +k
j=1
jaijuj ,
we have
|i|2 = i
2i 2
ig ui. (3.8)
On the other hand, from the definition of i, (3.5) and (3.6), wederive
2
ig ui (3.9)
= 2
gg uiui + 2k
j=1
aij
ujg ui
= 1 +
k
j=1
(i j)a2ij .
From the mini-max principle, we obtain
(k+1 i)
2i 1 +k
j=1
(i j)a2ij . (3.10)
Multiplying (3.9) by (k+1 i)2 and taking sum on i from 1 to k,we obtain
k
i=1
(k+1 i)2 +k
i,j=1
(i j)(k+1 i)2a2ij (3.11)
= 2k
i=1
(k+1 i)2
ig ui.
eigenvaluecm2013/9/2page 32
i
i
i
i
i
i
i
i
32 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
By aij = aji, bij = bji, we have
2k
i=1
(k+1 i)2
ig ui (3.12)
=
k
i=1
(k+1 i)2 4k
i=1
(k+1 i)b2ij w.
Multiplying (3.10) by (k+1 i)2 and taking sum on i from 1 to k,we infer
k
i=1
(k+1 i)3
2i
k
i=1
(k+1 i)2 4k
i=1
(k+1 i)b2ij = w.
Fromuij = 0 for all i, j = 1, , k, we have, for arbitrary con-
stants dij ,
w2
=
(2
k
i=1
(k+1 i)2
ig ui)2
4k
i=1
(k+1 i)32i
k
i=1
(k+1 i)1/2g ui k
j=1
dijuj
2
4wk
i=1
(k+1 i)|g ui|2 +
k
j=1
dijuj
2
2k
j=1
dij(k+1 i)1/2ujg ui
.
eigenvaluecm2013/9/2page 33
i
i
i
i
i
i
i
i
[SEC. 3.1: INTRODUCTION 33
Then we have
w 4k
i=1
(k+1 i)(uixp
)2
+4
2k
i,j=1
(k+1 i)1/2bij +k
i,j=1
d2ij
.
Putting dij = (k+1 i)1/2bij , we obtain
w 4k
i=1
(k+1 i)
(uixp
)2 4
k
i,j=1
(k+1 i)b2ij (3.13)
and so we infer
k
i=1
(k+1 i)2 4k
i=1
(k+1 i)
(uixp
)2. (3.14)
Summing over p, we obtain
k
i=1
(k+1 i)2 4
n
k
i=1
(k+1 i)
|ui|2 (3.15)
=4
n
k
i=1
(k+1 i)i.
Yangs inequality has been generalized to bounded domains incomplete submanifolds in Euclidean space. That is, we have
Theorem 3.2 ([20], [41]). Let be a bounded domain in ann-dimensional complete Riemannian manifold Mn isometrically im-mersed in RN . Then the Dirichlet eigenvalues of the Laplacian of satisfy the inequality
k
i=1
(k+1 k)2 4
n
k
i=1
(k+1 k)(i +n2
4||H||2), (3.16)
where H is the mean curvature vector field of Mn and ||H||2 =sup |H|2.
eigenvaluecmb2013/9/2page 34
i
i
i
i
i
i
i
i
34 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
3.2 Eigenvalues of the Clamped Plate
Problem
Let us generalize Yangs method to prove universal inequalities foreigenvalues of the clamped plate problem on Riemannian manifolds.
Theorem 3.2 ([94]). Let M be an n-dimensional completeRiemannian manifold and let be a bounded domain with smoothboundary in M . Denote by the outward unit normal of and leti the i-th eigenvalue of the problem:
{2u = u in ,
u| = u
= 0.(3.17)
i) If M is isometrically immersed in Rm with mean curvaturevector H, then
k
i=1
(k+1 i)2 (3.18)
1n
{k
i=1
(k+1 i)2(n2H20 + (2n+ 4)
1/2i
)}1/2
{
k
i=1
(k+1 i)(n2H20 + 4
1/2i
)}1/2,
where H0 = sup |H|.ii) If there exists a function : R and a constant A0 such
that
|| = 1, || A0, on , (3.19)
eigenvaluecmb2013/9/2page 35
i
i
i
i
i
i
i
i
[SEC. 3.2: EIGENVALUES OF THE CLAMPED PLATE PROBLEM 35
then
k
i=1
(k+1 i)2 (3.20)
{
k
i=1
(k+1 i)2(A20 + 4A0
1/4i + 6
1/2i
)}1/2
{
k
i=1
(k+1 i)(2
1/4i +A0
)2}1/2
.
iii) If there exists a function : R and a constant B0 suchthat
|| = 1, = B0, on , (3.21)
then
k
i=1
(k+1 i)2 {
k
i=1
(k+1 i)2(61/2i B20)}1/2
(3.22)
{
k
i=1
(k+1 i)(4
1/2i B20
)}1/2.
iv) If admits an eigenmap f = (f1, f2, , fm+1) : Sm(1)corresponding to an eigenvalue , that is,
f = f, = 1, ,m+ 1,m+1
=1
f2 = 1,
then
k
i=1
(k+1 i)2 {
k
i=1
(k+1 i)2(6
1/2i +
)}1/2(3.23)
{
k
i=1
(k+1 i)(4
1/2i +
)}1/2,
eigenvaluecm2013/9/2page 36
i
i
i
i
i
i
i
i
36 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
where Sm(1) is the unit m-sphere.
Theorem 3.2 can be deduced from a general result.
Lemma 3.1 ([89]). Let i, i = 1, , be the i-th eigenvelue of theproblem (3.17) and ui the orthonormal eigenfunction correspondingto i, that is,
2ui = iui in ,
ui| = ui
= 0,Muiuj = ij , i, j = 1, 2, .
(3.24)
Then for any smooth function h : R and any > 0, we havek
i=1
(k+1 i)2
u2i |h|2 (3.25)
k
i=1
(k+1 i)2
{u2i (h)2 2ui|h|2ui
+4((h ui)2 + uihh ui)}
+k
i=1
(k+1 i)
(h ui +
uih
2
)2.
Proof of Lemma 3.1. For i = 1, , k, consider the functionsi : R given by
i = hui k
j=1
rijuj ,
where
rij =
huiuj .
Since i| = i
= 0 and
uji = 0, i, j = 1, , k,
eigenvaluecm2013/9/2page 37
i
i
i
i
i
i
i
i
[SEC. 3.2: EIGENVALUES OF THE CLAMPED PLATE PROBLEM 37
it follows from the mini-max inequality that
k+1 i
2i2i
. (3.26)
We have
i2i (3.27)
=
i
2(hui) k
j=1
rijjuj
=
i2(hui)
= i||i||2 k
j=1
rijsij
+
hui((uih) + 2(h ui) + 2h (ui) + hui),
where ||i||2 =2i and
sij =
uj((uih) + 2(h ui) + 2h (ui) + hui).
Multiplying the equation 2ui = iui by huj , we have
huj2ui = ihuiuj . (3.28)
Changing the roles of i and j, one gets
hui2uj = jhuiuj . (3.29)
Subtracting (3.28) from (3.29) and integrating the resulted equation
eigenvaluecm2013/9/2page 38
i
i
i
i
i
i
i
i
38 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
on , we get
(j i)rij (3.30)
=
(hui2uj huj2ui)
=
((hui)uj (huj)ui)
=
((uih+ 2h ui)uj (ujh+ 2h uj)ui)
=
uj((uih) + 2(h ui) + hui + 2(ui) h)= sij ,
Observe that
hui((uih) + 2(h ui) + 2h (ui) + hui)
=
(u2i (h)2 + 4(|h ui|2 + uihh ui) 2ui|h|2ui).
It follows from (3.26), (3.27) and (3.30) that
(k+1 i)||i||2 (3.31)
(u2i (h)2 + 4(|h ui|2 + uihh ui) 2ui|h|2ui)
+
k
j=1
(i j)r2ij .
Set
tij =
uj
(h ui +
uih
2
); (3.32)
eigenvaluecm2013/9/2page 39
i
i
i
i
i
i
i
i
[SEC. 3.2: EIGENVALUES OF THE CLAMPED PLATE PROBLEM 39
then tij + tji = 0 and
(2)i(h ui +
uih
2
)(3.33)
=
(2huih ui u2ihh) + 2k
j=1
rijtij
=
u2i |h|2 + 2k
j=1
rijtij .
Multiplying (3.33) by (k+1 i)2 and using the Schwarz inequality,we get
(k+1 i)2
u2i |h|2 + 2k
j=1
rijtij
(3.34)
= (k+1 i)2
M
(2)i(h ui +
uih
2
)
= (k+1 i)2
M
(2)i
(h ui +
uih
2
)
k
j=1
tijuj
(k+1 i)3||i||2
+(k+1 i)
M
h ui +
uih
2
k
j=1
tijuj
2
= (k+1 i)3||i||2
+(k+1 i)
(h ui +
uih
2
)2
k
j=1
t2ij
Substituting (3.31) into (3.34) and summing over i from 1 to k and
eigenvaluecm2013/9/2page 40
i
i
i
i
i
i
i
i
40 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
noticing rij = rji, tij = tji, we get
k
i=1
(k+1 i)2
u2i |h|2 2k
i,j=1
(k+1 i)(i j)rijtij
k
i=1
(k+1 i)2
(u2i (h)2 + 4((h ui)2 + uihh ui)
2ui|h|2ui) +k
i=1
(k+1 i)
((h ui)2 +
uih
2
)2
k
i,j=1
(k+1 i)(i j)2r2ij k
i,j=1
(k+1 i)
t2ij ,
which implies (3.25).
Proof of Theorem 3.2. Let {ui}i=1 be the orthonormal eigenfunc-tions corresponding to the eigenvalues {i}i=1 of the problem (3.17).
i) Let x, = 1, ,m, be the standard coordinate functions ofR
m. Taking h = x in (3.25) and summing over , we have
k
i=1
(k+1 i)2m
=1
u2i |x|2 (3.35)
k+1
i=1
(k+1 i)2m
=1
(u2i (x)2 + 4((x ui)2
+uixx ui) 2ui|x|2ui)
+
k
i=1
(k+1 i)
m
=1
(x ui +
uix2
)2,
Since M is isometrically immersed in Rm, we have
m
=1
|x|2 = n
eigenvaluecm2013/9/2page 41
i
i
i
i
i
i
i
i
[SEC. 3.2: EIGENVALUES OF THE CLAMPED PLATE PROBLEM 41
which implies that
m
=1
u2i |x|2 = n (3.36)
Also, we have
(x1, , xm) (x1, ,xm) = nH, (3.37)
m
=1
(x ui)2 =m
=1
(ui(x))2 = |ui|2 (3.38)
and
m
=1
xx ui =m
=1
xui(x) = nH ui = 0. (3.39)
Substituting (3.36)-(3.39) into (3.35), we get
n
k
i=1
(k+1 i)2 (3.40)
k
i=1
(k+1 i)2
(n2u2i |H|2 + 4|ui|2 2nuiui)
+
k
i=1
(k+1 i)
(|ui|2 +
n2u2i |H|24
)
k
i=1
(k+1 i)2(n2H20 + (2n+ 4)1/2i )
+
k
i=1
(k+1 i)
(
1/2i +
n2H204
).
Here in the last inequality, we have used the fact that |H| H0 and
|ui|2 =
uiui (
u2i
)1/2(
(ui)2
)1/2=
1/2i .(3.41)
eigenvaluecm2013/9/2page 42
i
i
i
i
i
i
i
i
42 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
Taking
=
ki=1(k+1 i)
(
1/2i +
n2H204
)
ki=1(k+1 i)2(n2H20 + (2n+ 4)
1/2i )
1/2
,
one gets (3.18).ii) Substituting h = into (3.25) and using (3.19) and the Schwarz
inequality, we get
k
i=1
(k+1 i)2 (3.42)
k
i=1
(k+1 i)2
{u2i ()2 + 4(( ui)2 + ui ui)
2uiui} +k
i=1
(k+1 i)
( ui +
ui
2
)2
k
i=1
(k+1 i)2
(A20u2i + 4(|ui|2 +A0|ui||ui|) 2uiui)
+
k
i=1
(k+1 i)
(|ui|2 +A0|ui||ui| +
A20u2i
4
).
Substituting (3.41) and
|ui||ui| (
u2i
)1/2(
|ui|2)1/2
1/4i
into (3.42), we get
k
i=1
(k+1 i)2
k
i=1
(k+1 i)2(A20 + 4A01/4i + 61/2i )
+
k
i=1
(k+1 i)
(
1/4i +
A02
)2.
eigenvaluecm2013/9/2page 43
i
i
i
i
i
i
i
i
[SEC. 3.2: EIGENVALUES OF THE CLAMPED PLATE PROBLEM 43
Taking
=
ki=1(k+1 i)
(
1/4i +
A02
)2
ki=1(k+1 i)2(A20 + 4A0
1/4i + 6
1/2i )
1/2
,
we obtain (3.20).
iii) Introducing h = into (3.25) and using (3.21), we have
k
i=1
(k+1 i)2
k
i=1
(k+1 i)2
(B20u2i + 4(|ui|2 +B0ui ui) 2uiui)
+
k
i=1
(k+1 i)
(|ui|2 +B0ui ui +
B20u2i
4
)
k
i=1
(k+1 i)2(61/2i B20) +k
i=1
(k+1 i)
(
1/2i
B204
),
where in the last inequality, we have used the fact that
ui,ui = 1
2
u2i = B202.
Taking
=
ki=1(k+1 i)
(
1/2i
B204
)
ki=1(k+1 i)2(6
1/2i B20)
1/2
,
we obtain (3.22).
iv) Taking the Laplacian of the equation
m+1
=1
f2 = 1
eigenvaluecm2013/9/2page 44
i
i
i
i
i
i
i
i
44 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
and using the fact that
f = f, = 1, ,m+ 1,
we have
m+1
=1
|f|2 = .
It then follows by taking h = f in (3.25) and summing over that
k
i=1
(k+1 i)2
k
i=1
(k+1 i)2
(2u2i + 4
m+1
=1
(f ui)2 2uiui)
+
k
i=1
(k+1 i)
(m+1
=1
(f ui)2 +2u2i
4
)
k
i=1
(k+1 i)2(2 + 61/2i )
+
k
i=1
(k+1 i)
(
1/2i +
2
4
).
We get (3.23) by taking
=
ki=1(k+1 i)
(
1/2i +
4
)
ki=1(k+1 i)2
(6
1/2i +
)
1/2
.
Here are some examples of manifolds supporting the functions onthe whole manifolds as stated in items ii)-v) of Theorem 1.1.
Example 3.1. Let M be an n-dimensional Hadamard manifoldwith Ricci curvature satisfying RicM (n 1)c2, c 0 and let
eigenvaluecm2013/9/2page 45
i
i
i
i
i
i
i
i
[SEC. 3.2: EIGENVALUES OF THE CLAMPED PLATE PROBLEM 45
: [0,+) M be a geodesic ray, namely a unit speed geodesicwith d((s), (t)) = t s for any t > s > 0. The Busemann functionb corresponding to defined by
b(x) = limt+
(d(x, (t)) t)
satisfies |b | 1(Cf. [10], [49]). Also, it follows from Theorem 3.5 in[84] that |b | (n1)c2 on M . Thus any Hadamard manifold withRicci curvature bounded below supports functions satisfying (3.19).
Example 3.2. Let (N, ds2N ) be a complete Riemannian manifoldand define a Riemannian metric on M = R N by
ds2M = dt2 + 2(t)ds2N , (3.43)
where is a positive smooth function defined on R with (0) = 1.The manifold (M,ds2M ) is called a warped product and denoted byM = RN . It is known thatM is a complete Riemannian manifold.
Set = et and consider the warped product M = R et N .Define : M R by (x, t) = t. One can show that
|| = 1, = 1 n. (3.44)That is, a warped product manifold M = Ret N admits functionssatisfying (3.21).
Let Hn be the n-dimensional hyperbolic space with constant cur-vature 1. Using the upper half-space model, Hn is given by
Rn+ = {(x1, x2, , xn)|xn > 0} (3.45)
with metric
ds2 =dx21 + + dx2n
x2n(3.46)
One can check that the map : R et Rn1 given by(t, x) = (x, et)
is an isometry. Therefore, Hn admits a warped product model, Hn =R et Rn1.
Example 2.3. Any compact homogeneous Riemannian manifoldadmits eigenmaps to some unit sphere for the first positive eigenvalueof the Laplacian [70].
eigenvaluecm2013/9/2page 46
i
i
i
i
i
i
i
i
46 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
3.3 Eigenvalues of the Polyharmonic
Operator
The method of proving universal bounds for eigenvalues of the clampedplate problem can be generalized to the eigenvalue problem of poly-harmonic operators.
Theorem 3.3 ([55]). Let M be an n-dimensional compact Rie-mannian manifold with boundary M (possibly empty) Let l be apositive integer and let i, i = 1, , be the i-th eigenvalue of theproblem (1.7) and ui be the orthonormal eigenfunction correspondingto i, that is,
()lui = iui in M,ui|M = ui
M
= = l1uil1
M
= 0,
Muiuj = ij , for any i, j = 1, 2, .
(3.47)
Then for any function h Cl+2(M) Cl+1(M) and any positiveinteger k, we have
k
i=1
(k+1 i)2
M
u2i |h|2 (3.48)
k
i=1
(k+1 i)2
M
hui(()l(hui) ihui)
)
+
k
i=1
(k+1 i)
h,ui +uih
2
2
,
where is any positive constant.
Proof. For i = 1, , k, consider the functions i : M R givenby
i = hui k
j=1
rijuj , (3.49)
where
rij =
M
huiuj . (3.50)
eigenvaluecm2013/9/2page 47
i
i
i
i
i
i
i
i
[SEC. 3.3: EIGENVALUES OF THE POLYHARMONIC OPERATOR 47
Since
i|M =i
M
= = l1il1
M
= 0
and
M
uji = 0, i, j = 1, , k,
it follows from the mini-max inequality that
k+1
M
2i (3.51)
M
i()li
= i||i||2 +
M
i(()li ihui
)
= i||i||2 +
M
i(()l(hui) ihui
)
= i||i||2 +
M
hui(()l(hui) ihui
)
k
j=1
rijsij ,
where
sij =
M
(()l(hui) ihui
)uj .
Notice that if u Cl+2(M) Cl+1(M) satisfies
u|M =u
M
= = l1u
l1
M
= 0, (3.52)
then
u|M = u|M = u|M = (u)|M = = m1u
M
= (m1u)M
= 0, when l = 2m
and
u|M = u|M = u|M = (u)|M = = m1uM
= (m1u)M
= mu|M = 0, when l = 2m+ 1.
eigenvaluecm2013/9/2page 48
i
i
i
i
i
i
i
i
48 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
We can then use integration by parts to conclude that
M
uj()l(hui) =
M
hui()l(uj) = jrij ,
which gives
sij = (j i)rij . (3.53)
Set
pi(h) = ()l(hui) ihui;
then we have from (3.51) and (3.53) that
(k+1 i)||i||2
M
ipi(h) (3.54)
=
M
huipi(h) +k
j=1
(i j)r2ij .
Set
tij =
M
uj
(h ui +
uih
2
); (3.55)
then tij + tji = 0 and
M
(2)i(h ui +
uih
2
)= wi + 2
k
j=1
rijtij , (3.56)
where
wi =
M
(hu2i h 2huih ui). (3.57)
eigenvaluecm2013/9/2page 49
i
i
i
i
i
i
i
i
[SEC. 3.3: EIGENVALUES OF THE POLYHARMONIC OPERATOR 49
Multiplying (3.56) by (k+1 i)2 and using the Schwarz inequalityand (3.54), we get
(k+1 i)2
wi + 2k
j=1
rijtij
= (k+1 i)2
M
(2)i
(h ui +
uih
2
)
k
j=1
tijuj
(k+1 i)3||i||2
+(k+1 i)
M
h ui +
uih
2
k
j=1
tijuj
2
= (k+1 i)3||i||2
+(k+1 i)
h ui +uih
2
2
k
j=1
t2ij
(k+1 i)2
M
huipi(h) +k
j=1
(i j)r2ij
+(k+1 i)
h ui +uih
2
2
k
j=1
t2ij
.
Summing over i and noticing rij = rji, tij = tji, we inferk
i=1
(k+1 i)2wi 2k
i,j=1
(k+1 i)(i j)rijtij
k
i=1
(k+1 i)2
M
huipi(h)
+
k
i=1
(k+1 i)
h,ui +uih
2
2
k
i,j=1
(k+1 i)(i j)2r2ij k
i,j=1
(k+1 i)
t2ij .
eigenvaluecm2013/9/2page 50
i
i
i
i
i
i
i
i
50 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
Using (3.48), we can obtain universal inequalities for eigenvaluesof the problem (1.7) when M is a bounded domain in Rn or Sn(1).
Theorem 3.4 ([55]). Let be a bounded domain in Rn andDenote by i the i-th eigenvelue of the eigenvalue problem:
{()lu = u in ,u| = u
= = l1ul1
= 0.(3.58)
Then we have
k
i=1
(k+1 i)2 (3.59)
(
4l(n+ 2l 2)n2
)1/2( k
i=1
(k+1 i)2 (l1)/li
)1/2
(
k
i=1
(k+1 i)1/li
)1/2
Proof. Let x1, x2, , xn be the standard Euclidean coordinatefunctions of Rn. Let ui be the i-th orthonormal eigenfunction cor-responding to the eigenvalue i of the problem (3.58), i = 1, ;then
()l(xui) = ixui + 2l(1)lx (l1ui) (3.60)
Taking h = x in (3.48), we infer for any > 0 that
k
i=1
(k+1 i)2
k
i=1
(k+1 i)2
2l(1)lxuix (l1ui)
+1
k
i=1
(k+1 i) ||x ui||2.
eigenvaluecm2013/9/2page 51
i
i
i
i
i
i
i
i
[SEC. 3.3: EIGENVALUES OF THE POLYHARMONIC OPERATOR 51
Summing over , we have
n
k
i=1
(k+1 i)2 (3.61)
2lk
i=1
(k+1 i)2n
=1
(1)lxuix (l1ui)
+1
k
i=1
(k+1 i)
|ui|2.
By induction, we infer
ui()kui k/li , k = 1, , l. (3.62)
Since
l1(xui) = 2(l 1)x (l2ui) + xl1ui,
we have
xuix (l1ui) (3.63)
=
xuil1x ui
=
l1(xui)x ui
=
(2(l 1)x (l2ui) + xl1ui
)x ui.
On the other hand,
xuix (l1ui) (3.64)
=
l1ui div(xuix)
=
l1ui(|x|2ui + xx ui).
eigenvaluecm2013/9/2page 52
i
i
i
i
i
i
i
i
52 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
Combining (3.63) and (3.64), we obtain
xuix (l1ui) (3.65)
=
M
{(l 1)x (l2ui)x ui
1
2l1ui|x|2ui
}
Observe that
n
=1
(1)lxuix (l1ui) (3.66)
=
(1)l{
(l 1)(l2ui) ui n
2ui
l1ui}
=(l 1 + n
2
)
ui()l1ui
(l 1 + n
2
)
(l1)/li .
Substituting (3.62) and (3.66) into (3.61), one gets
n
k
i=1
(k+1 i)2 (3.67)
l(n+ 2l 2)k
i=1
(k+1 i)2 (l1)/li +1
k
i=1
(k+1 i)1/li .
Taking
=
{ ki=1 (k+1 i)
1/li
l(n+ 2l 2)ki=1 (k+1 i)2
(l1)/li
}1/2,
we get (3.59).
Let l be a positive integer and for p = 0, 1, 2, ..., define the poly-nomials Fp(t) inductively by
{F0(t) = 1, F1(t) = t n,Fp(t) = (2t 2)Fp1(t) (t2 + 2t n(n 2))Fp2(t), p = 2, .
eigenvaluecm2013/9/2page 53
i
i
i
i
i
i
i
i
[SEC. 3.3: EIGENVALUES OF THE POLYHARMONIC OPERATOR 53
(3.68)
Set
Fl(t) = tl al1tl1 + + (1)l1a1t+ (n)l. (3.69)
Theorem 3.4 ([55]). Let i be the i-th eigenvalue of the eigen-value problem:
{()lu = u in ,u| = u
= = l1ul1
= 0,
where is a compact domain in Sn(1). Then we have
k
i=1
(k+1 i)2 (3.70)
1n
{k
i=1
(k+1 i)2(a+l1i
l1l + + a+1 i
1l + a+0
)}1/2
{
k
i=1
(k+1 i)(n2 + 4
1/2i
)}1/2,
where a+j = max{0, aj}.
Proof. As before, let x1, x2, , xn+1 be the standard coordinatefunctions of Rn+1; then
Sn(1) = {(x1, . . . , xn+1) Rn+1;
n+1
=1
x2 = 1}.
It is well known that
x = nx, = 1, , n+ 1. (3.71)
Taking the Laplacian of the equationn+1
=1 x2 = 1 and using (3.71),
we get
n+1
=1
|x|2 = n. (3.72)
eigenvaluecm2013/9/2page 54
i
i
i
i
i
i
i
i
54 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
Let ui be the i-th orthonormal eigenfunction corresponding to theeigenvalue i, i = 1, 2, . For any > 0, by taking h = x in(3.48), we have
k
i=1
(k+1 i)2
u2i |x|2
k+1
i=1
(k+1 i)2
xui(()l(xui) ixui)
+1
k
i=1
(k+1 i)
x ui +uix
2
2
Taking sum on from 1 to n+ 1 and using (3.72), we get
n
k
i=1
(k+1 i)2 (3.73)
k+1
i=1
(k+1 i)2n+1
=1
xui(()l(xui) ixui)
+1
k
i=1
(k+1 i)n+1
=1
x ui +uix
2
2
.
It is easy to see that
n+1
=1
x ui +uix
2
2
(3.74)
=
n+1
=1
((x ui)2 nx uiuix +
n2u2ix2
4
)
=n2
4+
|ui|2
n2
4+
1/li .
For any smooth functions f, g on , we have from the Bochner for-
eigenvaluecm2013/9/2page 55
i
i
i
i
i
i
i
i
[SEC. 3.3: EIGENVALUES OF THE POLYHARMONIC OPERATOR 55
mula that
(f g) (3.75)= 22f 2g + f (g) + g (f)
+2(n 1)f g,
where
2f 2g =n
s,t=1
2f(es, et)2g(es, et),
being e1, , en orthonormal vector fields locally defined on . Since
2x = xI,
we infer from (3.75) by taking f = x that
(x g) (3.76)= 2xg + x (g) + (n 2)x g= 2xg + x (( + (n 2))g).
For each q = 0, 1, , thanks to (3.71) and (3.76), there are polyno-mials Bq and Cq of degrees less than or equal to q such that
q(xg) = xBq()g + 2x (Cq()g). (3.77)
It is obvious that
B0 = 1, B1 = t n, C0 = 0, C1 = 1. (3.78)
It follows from (3.71), (3.76) and (3.77) that
q(xg) = (q1(xg)) (3.79)
= (xBq1()g + 2x (Cq1()g))= x(( n)Bq1() 4Cq1())g
+2x ((Bq1() + ( + (n 2))Cq1())g).
Thus, for any q = 2, , we have
Bq() = ( n)Bq1() 4Cq1(), (3.80)Cq() = Bq1() + ( + (n 2))Cq1(). (3.81)
eigenvaluecm2013/9/2page 56
i
i
i
i
i
i
i
i
56 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
Consequently,
Bq() (3.82)
= (2 2)Bq1() ( + n 2)Bq1() 4Cq1()= (2 2)Bq1() (2 + 2 n(n 2))Bq2()
+4[Bq2() + ( + n 2)Cq2() Cq1()]= (2 2)Bq1() (2 + 2 n(n 2))Bq2(), q = 2, .
Since (3.78) and (3.82) hold, we know that Bq = Fq, q = 0, 1, .It follows from (3.77) and the divergence theorem that
xui(()l(xui) ixui) (3.83)
=
xui((1)l (xBl()ui + 2x (Cl()ui)) ixui
)
=
xui((1)l
(x(
l al1l1 + + (n)l)ui+2x (Cl()ui)) ixui)
=
(1)lxui(x(al1l1 + + (n)l)ui + 2x (Cl()ui)
)
Summing on , one has
xui(()l(xui) ixui) (3.84)
=
ui(1)l(al1l1 + + (n)la0)ui
= al1
ui()l1ui + + a1
ui()ui + nl
u2i
a+l1(l1)/li + + a+1
1/li + n
l.
eigenvaluecm2013/9/2page 57
i
i
i
i
i
i
i
i
[SEC. 3.4: EIGENVALUES OF THE BUCKLING PROBLEM 57
Substituting (3.74) and (3.84) into (3.73), we infer
n
k
i=1
(k+1 i)2
k
i=1
(k+1 i)2(a+l1(l1)/li + + a+1
1/li + n
l)
+1
k
i=1
(k+1 i)(
1/li +
n2
4
).
Taking
=
ki=1(k+1 i)
(
1/li +
n2
4
)
ki=1(k+1 i)2(a+l1
(l1)/li + + a+1
1/li + n
l)
1/2
,
we get (3.70).
3.4 Eigenvalues of the Buckling Problem
Let Rn and consider the problem{
2u = u,u| = u
= 0(3.85)
which is used to describe the critical buckling load of a clamped platesubjected to a uniform compressive force around its boundary.
Payne, Polya and Weinberger [77] proved
2/1 < 3 for R2.
For Rn this reads
2/1 < 1 + 4/n.
Subsequently Hile and Yeh [51] reconsidered this problem obtainingthe improved bound
21
n2 + 8n+ 20
(n+ 2)2for Rn.
eigenvaluecmb2013/9/2page 58
i
i
i
i
i
i
i
i
58 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
Ashbaugh [3] proved :
n
i=1
i+1 (n+ 4)1. (3.86)
This inequality has been improved to the following form [54]:
n
i=1
i+1 +4(2 1)n+ 4
(n+ 4)1.
Cheng and Yang introduced a new method to construct trial functionsfor the problem (3.85) and obtained the following universal inequality[28]:
k
i=1
(k+1 i)2 4(n+ 2)
n2
k
i=1
(k+1 i)i. (3.87)
It has been proved in [88] that for the problem (1.5) ifM is a boundedconnected domain in an n-dimensional unit sphere, then the followinginequality holds
2k
i=1
(k+1 i)2 (3.88)
k
i=1
(k+1 i)2(i +
2(i (n 2))4(i + n 2)
)
+1
k
i=1
(k+1 i)(i +
(n 2)24
),
where is any positive constant.The inequality (3.87) has been generalized to eigenvalues of buck-
ling problem of arbitrary orders. That is, we have
Theorem 3.5 ([54]). Let l 2 and let i be the i-th eigenvalueof the following eigenvalue problem:
{()lu = u in ,u| = u
= = l1ul1
= 0.(3.89)
eigenvaluecm2013/9/2page 59
i
i
i
i
i
i
i
i
[SEC. 3.4: EIGENVALUES OF THE BUCKLING PROBLEM 59
where is a bounded domain with smooth boundary in Rn. Then fork = 1, , we have
k
i=1
(k+1 i)2 (3.90)
2(2l2 + (n 4)l + 2 n)1/2
n
{k
i=1
(k+1 i)2(l2)/(l1)i
}1/2
{
k
i=1
(k+1 i)1/(l1)i
}1/2.
Before proving theorem 3.5, let us recall a method of constructingtrial functions developed by Cheng-Yang (Cf. [28], [54]). LetM be ann-dimensional complete submanifold in an m-dimensional Euclideanspace Rm. Denote by the canonical metric on Rm as well as thatinduced on M . Let and be the Laplacian and the gradientoperator of M , respectively. Let be a bounded connected domainof M with smooth boundary and let be the outward unit normalvector field of . For functions f and g on , the Dirichlet innerproduct (f, g)D of f and g is given by
(f, g)D =
f g.
The Dirichlet norm of a function f is defined by
||f ||D = {(f, f)D}1/2 =(
|f |2)1/2
.
Consider the eigenvalue problem
{()lu = u in ,u| = u
= = l1ul1
= 0.(3.91)
Let
0 < 1 2 3 ,
eigenvaluecm2013/9/2page 60
i
i
i
i
i
i
i
i
60 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
denote the successive eigenvalues, where each eigenvalue is repeatedaccording to its multiplicity.
Let ui be the i-th orthonormal eigenfunction of the problem (3.91)corresponding to the eigenvalue i, i = 1, 2, , that is, ui satisfies
()lui = iui in ,ui| = ui
= l1ul1
= 0,
(ui, uj)D =ui,uj = ij , i, j.
(3.92)
For k = 1, , l, let k denote the k-th covariant derivative operatoron M , defined in the usual weak sense via an integration by partsformula. For a function f on , the squared norm of kf is definedas
kf2 =
n
i1, ,ik=1
(kf(ei1 , , eik)
)2, (3.93)
where e1, , en are orthonormal vector fields locally defined on .Define the Sobolev space H2l () by
H2l () = {f : f, |f |, ,lf
L2()}.Then H2l () is a Hilbert space with respect to the norm || ||l,2:
||f ||l,2 =(
(l
k=0
|kf |2))1/2
. (3.94)
Consider the subspace H2l,D() of H2l () defined by
H2l,D() =
{f H2l () : f | =
f
= l1u
l1
= 0
}.
The operator ()l defines a self-adjoint operator acting on H2l,D()with discrete eigenvalues 0 < 1 k for the bucklingproblem (3.91) and the eigenfunctions {ui}i=1 defined in (3.92) forma complete orthonormal basis for the Hilbert space H2l,D(). If H2l,D() satisfies (, uj)D = 0, j = 1, 2, , k, then the Rayleigh-Ritz inequality tells us that
k+1||||2D
()l. (3.95)
eigenvaluecm2013/9/2page 61
i
i
i
i
i
i
i
i
[SEC. 3.4: EIGENVALUES OF THE BUCKLING PROBLEM 61
For vector-valued functions F = (f1, f2, , fm), G = (g1, g2, , gm) : Rm, we define an inner product (F,G) by
(F,G) =
m
=1
fg.
The norm of F is given by
||F || = (F, F )1/2 ={
m
=1
f2
}1/2.
Let H21() be the Hilbert space of vector-valued functions given by
H21()
={F = (f1, , fm) : Rm; f, |f| L2(), = 1, ,m
}
with norm
||F ||1 =(||F ||2 +
m
=1
|f|2)1/2
.
Observe that a vector field on can be regarded as a vector-valuedfunction from to Rm. Let H21,D() H2l () be a subspace ofH2l () spanned by the vector-valued functions {ui}i=1, which forma complete orthonormal basis of H21,D(). For any f H2l,D(), wehave f H21,D() and for anyX H21,D(), there exists a functionf H2l,D() such that X = f .
Proof of Theorem 3.5. With notations as above, we consider nowthe special case that is a bounded domain in Rn. Let us decomposethe vector-valued functions xui as
xui = hi +Wi, (3.96)
where hi H l2,D(), hi is the projection of xui in H21,D()and Wi H21,D(). Thus we have
Wi| = 0,
Wi u = 0, u H2l,D() (3.97)
eigenvaluecm2013/9/2page 62
i
i
i
i
i
i
i
i
62 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
and from the discussions in [28] and [88] we know that
div Wi = 0. (3.98)
For each = 1, , n, i = 1, , k, consider the functions i : R, given by
i = hi k
j=1
aijuj , (3.99)
where
aij =
xui uj = aji. (3.100)
We have
i| =i
= l1il1
= 0, (3.101)
(i, uj)D =
i uj = 0, j = 1, , k. (3.102)
It follows from the Rayleigh-Ritz inequality that
k+1
|i|2 (3.103)
D
i()li, = 1, , n, i = 1, , k.
It is easy to see that
()li = (1)ll1 (ui, + xui) +k
j=1
aijjuj
eigenvaluecm2013/9/2page 63
i
i
i
i
i
i
i
i
[SEC. 3.4: EIGENVALUES OF THE BUCKLING PROBLEM 63
and so
i()li (3.104)
=
i(1)ll1 (ui, + xui)
=
hi(1)ll1 (ui, + xui) k
j=1
aij
uj()lhi
=
hi(1)ll2 (ui, + xui) k
j=1
aij
hi()luj
=
hi(1)l((l2ui), + l2(xui)) +k
j=1
jaij
hiuj
=
(1)l(ui, + xui)((l2ui), + l2(xui))
k
j=1
jaij
hi,uj
=
(1)l(ui, + xui)((2l 3)(l2ui), + xl1ui))
k
j=1
jaij
hi,uj
=
(1)l((2l 3){ui,(l2ui), + xui(l2ui),)
+ui,xl1ui + x
2ui
l1ui} k
j=1
ja2ij
Since
l1(xui) = 2(l 1)(l2ui), + xl1ui,
eigenvaluecm2013/9/2page 64
i
i
i
i
i
i
i
i
64 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
we have
xui(l1ui), =
xuil1ui, (3.105)
=
l1(xui)ui,
=
(2(l 1)(l2ui), + xl1ui
)ui,.
On the other hand, it holds
xui(l1ui), =
l1ui(ui + xui,). (3.106)
Combining (3.105) and (3.106), we obtain
xui(l1ui), (3.107)
=
M
{(l 1)(l2ui),ui,
1
2ui
l1ui
}
Hence
xui,l1ui =
ui(l1ui + x(
l1ui),) (3.108)
=
M
{(l 1)(l2ui),ui, +
1
2ui
l1ui
}
and consequently, we have
xui(l2ui), =
xuil2ui, (3.109)
=
l2(xui)ui,
=
ui,(2(l 2)(l2ui), + xl1ui
)
=
{(l 3)(l2ui),ui,
1
2ui
l1ui
}.
eigenvaluecm2013/9/2page 65
i
i
i
i
i
i
i
i
[SEC. 3.4: EIGENVALUES OF THE BUCKLING PROBLEM 65
Also, one has
uix2ui (3.110)
=
x2|ui|2 2
xuiui,
=
x2|ui|2 +
u2i ,
x2uil1ui (3.111)
=
ui(x2
l1ui)
=
ui(2l1ui + x
2
lui + 4x(l1ui),)
=
ui(2l1ui + (1)l1ix2ui + 4x(l1ui),).
Combining (3.107), (3.110) and (3.111), we get
x2uil1ui = 4(l 1)
(l2ui),ui, (3.112)
+(1)l1i{
x2|ui|2 +
u2i
}.
Substituting (3.109), (3.111) and (3.112) into (3.104), one gets
i()li (3.113)
=
(1)l{(l + 1)uil1ui + (2l2 4l + 3)(l2ui),ui,
}
+i
{
x2|ui|2
u2i
}
k
j=1
ja2ij .
It is easy to see that
||xui||2 = ||hi||2 + ||Wi||2 (3.114)
eigenvaluecm2013/9/2page 66
i
i
i
i
i
i
i
i
66 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
and
||hi||2 = ||i||2 +k
j=1
a2ij , (3.115)
Combining (3.103), (3.113), (3.114) and (3.115), we infer
(k+1 i)||i||2 (3.116)
(1)l{(l + 1)uil1ui + (2l2 4l + 3)(l2ui),ui,
}
i(||ui||2 ||Wi||2) +k
j=1
(i j)a2ij ,
Observe that (xui) = uix + xui H21,D(). For Ai =(xui hi), we have
uix = Ai Wi (3.117)and so
||ui||2 = ||uix||2 = ||Wi||2 + ||Ai||2.Because of (ui,,Wi) = 0, it follows that
2||ui,||2 = 2
Ai ui,
1/(l1)i ||Ai||2 +1
1/(l1)i
||ui,||2
which gives
i||Ai||2 2l2l1
i ||ui,||2 + l3l1
i ||ui,||2 (3.118)Introducing (3.118) into (3.116), we get
(k+1 i)||i||2 (3.119)
(1)l{(l + 1)uil1ui + (2l2 4l + 3)(l2ui),ui,
}
2(l2)/(l1)i ||ui,||2 + (l3)/(l1)i ||ui,||2
+
k
j=1
(i j)a2ij ,
eigenvaluecm2013/9/2page 67
i
i
i
i
i
i
i
i
[SEC. 3.4: EIGENVALUES OF THE BUCKLING PROBLEM 67
Since
2
xui ui,
= 2
u2i, + 2
xui,ui
= 2
u2i, + 2
ui(xui,)
= 2
u2i, + 2
uix(ui), + 4
uix ui,
= 2
u2i, 2
ui(ui + xui,) 4
ui,div(uix)
= 2
u2i, + 2 2
xui,ui 4
u2i,
= 2
u2i, + 2 + 2
ui (xui,)
= 2 + 2
xui ui,,
we have
2
xui ui, = 1. (3.120)
Set
dij =
ui, uj ;
then dij = dji and we get
1 = 2
xui ui, (3.121)
= 2
hi ui,
= 2
i ui, 2k
j=1
aijdij .
eigenvaluecm2013/9/2page 68
i
i
i
i
i
i
i
i
68 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
Thus, we have
(k+1 i)2
1 + 2k
j=1
aijdij
(3.122)
= (k+1 i)2
2i,
ui, k
j=1
dijuj
(k+1 i)3||i||2 +1
(k+1 i)
||ui,||2 k
j=1
d2ij
,
where is any positive constant. Substituting (3.119) into (3.122),we get
(k+1 i)2
1 + 2k
j=1
aijdij
(k+1 i)2(
(1)l{(l + 1)uil1ui
+(2l2 4l + 3)(l2ui),ui,}
2l2l1
i ||ui,||2 + l3l1
i ||ui,||2 +k
j=1
(i j)a2ij
+1
(k+1 i)
||ui,||2 k
j=1
d2ij
,
eigenvaluecm2013/9/2page 69
i
i
i
i
i
i
i
i
[SEC. 3.4: EIGENVALUES OF THE BUCKLING PROBLEM 69
Summing on i from 1 to k, we infer
k
i=1
(k+1 i)2 2k
i,j=1
(k+1 i)(i j)aijdij
(
k
i=1
(k+1 i)2(
(1)l{(l + 1)uil1ui
+(2l2 4l + 3)(l2ui),ui,}2(l2)/(l1)i ||ui,||2 +
(l3)/(l1)i ||ui,||2
)
k
i,j=1
(k+1 i)(i j)2a2ij
+1
k
i=1
(k+1 i)||ui,||2 k
i,j=1
(k+1 i)d2ij
,
which gives
k
i=1
(k+1 i)2
k
i=1
(k+1 i)2(
(1)l{(l + 1)uil1ui
+(2l2 4l + 3)(l2ui),ui,}
2l2l1
i ||ui,||2 + l3l1
i ||ui,||2)
+1
k
i=1
(k+1 i)||ui,||2,
eigenvaluecm2013/9/2page 70
i
i
i
i
i
i
i
i
70 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
Taking sum for from 1 to n, we get
nk
i=1
(k+1 i)2
k
i=1
(k+1 i)2(
(1)l{n(l + 1)uil1ui
+(2l2 4l + 3)(l2ui) ui}
2l2l1
i + l3l1
i
n
=1
||ui,||2)
+1
k
i=1
(k+1 i)n
=1
||ui,||2
= k
i=1
(k+1 i)2(2
l2l1
i + l3l1
i
n
=1
||ui,||2
+(2l2 + (n 4)l + 3 n)
ui()l1ui)
+1
k
i=1
(k+1 i)n
=1
||ui,||2.
But
k
=1
||ui,||2 =
k
=1
ui,ui,
=
k
=1
ui,(ui),
=
k
=1
ui,ui
=
(ui)2
=
ui2ui,
eigenvaluecm2013/9/2page 71
i
i
i
i
i
i
i
i
[SEC. 3.4: EIGENVALUES OF THE BUCKLING PROBLEM 71
where ui, =2uix2
. Therefore,
nk
i=1
(k+1 i)2
k
i=1
(k+1 i)2(2
l2l1
i + l3l1
i
ui2ui
+(2l2 + (n 4)l + 3 n)
ui()l1ui)
+1
k
i=1
(k+1 i)
ui2ui.
Observe that
ui()l1ui l2l1
i ,
ui2ui
1l1
i .
Thus we have
n
k
i=1
(k+1 i)2
(2l2 + (n 4)l + 2 n)k
i=1
(k+1 i)2l2l1
i
+1
k
i=1
(k+1 i)1
l1
i .
Taking
=
{ki=1(k+1 i)
1l1
i
}1/2
{(2l2 + (n 4)l + 2 n)ki=1(k+1 i)2
l2l1
i
}1/2 ,
we get (3.90).
eigenvaluecm2013/9/2page 72
i
i
i
i
i
i
i
i
72 [CAP. 3: UNIVERSAL INEQUALITIES FOR EIGENVALUES
Recently, Cheng-Yang have strengthened the inequality (3.87) tothe following form:
k
i=1
(k+1 i)2 4(n+ 43
)
n2
k
i=1
(k+1 i)i. (3.123)
The inequality (3.90) has been improved by Cheng-Qi-Wang-Xia veryrecently in [24]:
n
k
i=1
(k+1 i)2 (3.124)
k
i=1
i(k+1 i)2(
2l2 +
(n 14
3
)l +
8
3 n
)
l2l1
i
+
k
i=1
1
i(k+1 i)
1l1
i ,
where {i}ki=1 is any positive non-increasing monotone sequence.For eigenvalues of the buckling problem on spherical domains, the
inequality (3.88) has also been improved (Cf. [28])) and generalizedto buckling problem of arbitrary orders (Cf. [54], [24]).
eigenvaluecm2013/9/2page 73
i
i
i
i
i
i
i
i
Chapter 4
Polya Conjecture and
Related Results
4.1 Introduction
Let Rn be a bounded open set and consider the eigenvalueproblem of the Dirichlet Laplacian on :
{u+ u = 0 in ,u| = 0.
(4.1)
In 1960, Polya [81] showed that for any plane covering domain in R2 (those that tile R2) the Weyl asymptotic relation (1.) is in facta one-sided inequality (his proof also works for Rn-covering domains)and conjectured, for any bounded domain Rn, the inequality
k C(n)(k
||
)2/nk, with C(n) = (2)
2
2nn
. (4.2)
Polyas conjecture has been a central problem about eigenvalues andmany important developments have been made during the past years.In 1982, Li-Yau [72] showed the lower bound
k
i=1
i nkC(n)
n+ 2
(k
||
)2/n(4.3)
73
eigenvaluecm2013/9/2page 74
i
i
i
i
i
i
i
i
74 [CAP. 4: POLYA CONJECTURE AND RELATED RESULTS
which yields an individual lower bound on k in the form
k nC(n)
n+ 2
(k
||
)2/n. (4.4)
Similar bounds for eigenvalues with Neumann boundary conditionshave been proved in [62], [63] and [65]. It was pointed out in [66]that (4.3) also follows from an earlier result by Berezin [11] by theLegendre transformation. The inequality (4.3) has been improved byMelas:
Theorem 4.1 ([75]). For any bounded domain Rn and anyk 1, the eigenvalues of the problem (4.1) satisfy the inequality
k
i=1
i nkC(n)
n+ 2
(k
||
)2/n+ d(n)k
||I()
(4.5)
where the constant dn depends only on the dimension and I() =minaRn
|x a|2dx is the moment of inertia of .
Proof. Fix a k 1 and let u1, , uk be an orthonormal setof eigenfunctions of (4.1) corresponding to the set of eigenvalues1, , k. We consider the Fourier transform of each eigenfunction
fj() = uj() = (2)n/2
uj(x)eixdx. (4.6)
From Plancherels Theorem, we know that f1, , fk is an orthonor-mal set in Rn. Bessels inequality implies that for every Rn
k
j=1
|fj()|2 (2)n
|eix|dx = (2)n|| (4.7)
and
k
j=1
|fj()|2 (2)n
|ixeix|dx = (2)nI(). (4.8)
Since uj | = 0 it is easy to see that
Rn
||2|fj()|2d =
|uj |2 = j . (4.9)
eigenvaluecm2013/9/2page 75
i
i
i
i
i
i
i
i
[SEC. 4.1: INTRODUCTION 75
Setting
F () =
k
j=1
|fj()|2; (4.10)
then 0 F () (2)n||,
|F ()| 2(k
j=1
|fj()|2)1/2(k
j=1
|fj()|2)1/2 (4.11)
2(2)n
||I(), Rn,
Rn
F ()d = k (4.12)
and
Rn
||2F ()d =k
j=1
j . (4.13)
Let F () = (||) denote the decreasing radial rearrangement of F .By approximating F we may assume that the function : [0;+) [0, (2)n||] is absolutely continuous. Setting (t) = |{F > t}| =|{F > t}| the co-area formula (1.17) implies that
(t) =
(2)n||
t
{F=s}|F |1dsds. (4.14)
Observe that F is radial and so ((s)) = |{F > (s)}| = nsnwhich gives nns
n1 = ((s))(s) a.e. It follows from (4.11),(4.14) and the isoperimetric inequality that
((s)) =
{F=(s)}|F |1ds (4.15)
1|{F = (s)}| 1nnsn1
eigenvaluecm2013/9/2page 76
i
i
i
i
i
i
i
i
76 [CAP. 4: POLYA CONJECTURE AND RELATED RESULTS
and so for almost every s
(s) 0, (4.16)
where
= 2(2)n||I().
From (4.12) and (4.13), we have
k =
Rn
F ()d =
Rn
F ()d (4.17)
= nn
0
sn1(s)ds
and
k
j=1
j =
Rn
|xi|2F ()d (4.18)
Rn
|xi|2F ()d
= nn
0
sn+1(s)ds.
We need an elementary lemma.
Lemma 4.1([75]) Let n 1, , A > 0 and : [0,+) [0,+)be decreasing (and absolutely continuous) such that
(s) 0,
0
sn1(s)ds = A. (4.19)
Then
0
sn+1(s)ds 1n+ 2
(nA)n+2
n (0)2n +
A(0)2
6(n+ 2)2. (4.20)
Applying Lemma 4.1 to the function with A = (nn)1k, =
2(2)n||I() we get in view of (4.12) and (4.13) that
j=1
j n
n+ 2 2nn k
n+2n (0)
2n +
ck(0)2
(n+ 2)2(4.21)
eigenvaluecmb2013/9/2page 77
i
i
i
i
i
i
i
i
[SEC. 4.2: THE KROGERS THEOREM 77
where c is any constant such that 0 < c < 16 . Obs