Discover millions of ebooks, audiobooks, and so much more with a free trial

Only $11.99/month after trial. Cancel anytime.

Eigenvalues in Riemannian Geometry
Eigenvalues in Riemannian Geometry
Eigenvalues in Riemannian Geometry
Ebook507 pages3 hours

Eigenvalues in Riemannian Geometry

Rating: 0 out of 5 stars

()

Read preview

About this ebook

The basic goals of the book are: (i) to introduce the subject to those interested in discovering it, (ii) to coherently present a number of basic techniques and results, currently used in the subject, to those working in it, and (iii) to present some of the results that are attractive in their own right, and which lend themselves to a presentation not overburdened with technical machinery.
LanguageEnglish
Release dateNov 7, 1984
ISBN9780080874340
Eigenvalues in Riemannian Geometry

Related to Eigenvalues in Riemannian Geometry

Titles in the series (96)

View More

Related ebooks

Mathematics For You

View More

Related articles

Reviews for Eigenvalues in Riemannian Geometry

Rating: 0 out of 5 stars
0 ratings

0 ratings0 reviews

What did you think?

Tap to rate

Review must be at least 10 words

    Book preview

    Eigenvalues in Riemannian Geometry - Isaac Chavel

    affection

    Preface

    Isaac Chavel

    The subject of the geometry of the Laplace operator has undergone a veritable explosion in the past fifteen years, characterized by a dizzying wealth and variety of subject matter and techniques. The only systematic treatment available, with a Riemannian geometric viewpoint, is the lecture-notes volume by Berger–Gauduchon–Mazet [1], which appeared in 1970. It therefore seemed to me, ten years later, that an updated introduction would be most desirable.

    The basic goals of the book are: (i) to introduce the subject to those interested in discovering it, (ii) to coherently present a number of basic techniques and results, currently used in the subject, to those working in it, and (iii) to present some of the results that are attractive in their own right, and which lend themselves to a presentation not overburdened with technical machinery.

    I have made no attempt to give an exhaustive definitive account of the subject with a grand overview—in fact, I have resisted any temptation to do so. More particularly, I have restricted the subject matter to the Laplacian acting on functions (except for an appendix, graciously contributed by J. Dodziuk, on the Laplacian on forms), to questions concerning eigenvalues of the Laplacian for compact manifolds and domains with compact closure, and to the study of the associated questions concerning the heat equation. (The study of the heat equation for noncompact manifolds was viewed as a natural extension and application of the study of Dirichlet heat kernels of domains with compact closure.) Even in this area, I did not present, fully, all the topics deserving a detailed presentation. In some cases I settled for remarks on the results and, in others, for references (e.g., the inequalities of Hayman and Osserman relating the inradius of a domain to its lowest Dirichlet eigenvalue). In still other instances, I simply relied on the bibliography to put the reader on the track to related topics.

    I also add that there is no overarching philosophy, nor big questions point-of-view, to my presentation. The hallmark of the subject, to my mind, is the lively and variegated interaction of the fields of topology and Riemannian geometry with the fields of partial differential equations, probability, and number theory; and it is precisely this quality that I have tried to capture in these pages.

    The first chapter introduces the Laplacian or, more precisely, the Laplace–Beltrami operator, acting on functions on Riemannian manifolds, summarizes the basic facts of the existence of eigenvalues, and their associated Weyl formulas, and then presents the classical Rayleigh characterization of eigenvalues, followed by the max–min method.

    The second chapter gives a discussion of the basic examples, emphasizing, in some detail, the Dirichlet and Neumann eigenvalues of geodesic disks in simply connected constant curvature space forms. The third chapter then develops the comparison arguments with which to estimate the lowest Dirichlet eigenvalue of a geodesic disk in a Riemannian manifold in terms of the corresponding eigenvalue in a constant curvature space form, the constant in question being an upper or lower bound of the sectional curvatures of the geodesic disk of the original manifold. These estimates are then applied to derive the theorems of Obata and Topogonov. A summary of information about the exponential map, Jacobi fields, and geodesic spherical coordinates is presented in the beginning of Chapter III.

    Chapters IV and V are devoted to isoperimetric inequalities. We start with the celebrated Faber–Krahn solution to the Rayleigh conjecture and with a number of variations of the argument. Then we discuss the Cheeger and isoperimetric constants and their effect on estimating eigenvalues and eigenfunctions. Chapter V is devoted to estimating the Cheeger and isoperimetric constants in terms of Riemannian data of the underlying manifold. It starts with the contributions of M. Berger and J. L. Kazdan toward the solution of the Blaschke conjecture, which are then applied, by arguments of C. B. Croke, to the estimation of the Cheeger and isoperimetric constants.

    In Chapters VI–IX we focus on the heat equation. We give the existence theorems for eigenvalues (of the closed and Dirichlet eigenvalue problems) and the convergence of temperature distributions (for large times) to steady state, following the arguments of A. N. Milgram and P. C. Rosenbloom, which, in turn, require the existence of the heat kernel, for which we use the construction of S. Minakshisundaram. (The advantage of this approach is that the most advanced spectral theorem required is the one for compact operators on Hilbert space.) Also included are the Weyl formulas for these two eigenvalue problems. We then study the heat kernel on noncompact Riemannian manifolds, starting with the uniqueness and existence theorems of J. Dodziuk, followed by the comparison theorems of J. Cheeger and S. T. Yau, and concluding with the estimates of J. Cheeger, M. Gromov, and M. Taylor.

    In Chapter IX we present the work of E. A. Feldman and myself on topological perturbations of the underlying manifold with negligible spectral effects. Besides my obvious interest in these questions, I consider these results quite suitable for the book, as they draw on much of the earlier material, and as they provide an appropriate context for introducing considerations of Brownian motion to the study of the heat kernel and, thereby, the study of eigenvalues.

    The next two chapters study compact surfaces of constant negative curvature. In Chapter X we study the interaction of the geometry–topology of the surface with low eigenvalues, using the arguments of P. Buser, B. Randol, and S. Wolpert–R. Schoen–S. T. Yau. These arguments require only Chapters I–IV. (In fact, one might construct a convenient introductory short-course from Chapters I–V followed by Sections 1, 3, and 4 of Chapter X.) The whole chapter contrasts sharply with the elegant and powerful theory of the Selberg trace formula, presented in Chapter XI by B. Randol. One enters, here, to a completely fresh enterprise revealing a development of the subject of eigenvalues along totally different lines from those featured earlier in the book.

    The final chapter is what the title says it is: Miscellanea—those matters that did not fit smoothly into the flow of the text, but which I did not want to exclude. The Appendix is an essay, by J. Dodziuk, summarizing some of the basic features of the theory of the Laplacian on forms—most notably, the heat equation approach to index theorems.

    The bibliography is far from complete, but I think that it is sufficiently serviceable to the reader wishing to continue exploring the subject. I only hope that there are no gross injustices in the inclusion–exclusion of references, and in the attribution of results.

    To my knowledge, earlier surveys and collections of articles are as follows: (i) Volumes 16(1970), 23(1973), 27(1975), and 36(1980), of the Proceedings of American Mathematical Society Symposia in Pure Mathematics, with Volume 36 completely devoted to the geometry of the Laplace operator. (ii) The books Polya–Szegö [1], Bandle [1], and the survey articles of Payne [1] and Osserman [3; 4], on isoperimetric inequalities. (iii) The articles of M. Berger in Volumes 16 and 27 of the above AMS Proceedings, the survey articles of Simon-Wissner [1], Li [4], the lecture-notes Berger–Gauduchon–Mazet [1], the bibliography Bérard-Berger [1], and the broad survey article of S. T. Yau [3] on partial differential equations in differential geometry.

    It is a pleasure to thank the geometers and analysts of the doctoral faculty of the City University of New York for their extended help and discussions since I came to C.U.N.Y. in 1970—J. Dodziuk, E. A. Feldman, S. Kaplan, B. Randol, and R. Sacksteder. A special thanks goes to J. Dodziuk for his appendix on the Laplacian on forms.

    I am especially grateful to B. Randol for his chapter on the Selberg trace formula and am proud that his essay is included in this book.

    And, finally, I am delighted to acknowledge my debt to Edgar A. Feldman, with whom I have worked on this subject for more years than it is wise to announce. There is hardly a worthwhile insight that I brought to the pages of this book in which he does not have a major share.

    Riverdale, New York

    July 1984

    Chapter I

    The Laplacian

    In this chapter we present the basic definitions and facts to be used in the subsequent chapters. Few proofs are presented here—our main interest is in the interplay between the eigenvalues of the Laplacian and the global geometric invariants of the underlying manifold, and we wish to get to this topic as quickly as possible. Thus, for example, we shall consider the heat equation later, in some detail, and content ourselves in Section 4 with the formal calculation connecting the phenomena of eigenvalues and diffusion. Only in Section 5 do we start sketching the arguments. Nearly all the background material is available in the book references. (We note that the max–min arguments are based on the classic: Courant–Hilbert [1, Vol. I, Chap. VI].)

    1 Definitions and Preliminaries

    We let M be an n–dimensional, n 1, connected, C∞, Riemannian manifold. Should M have a boundary ∂M, we shall assume that M is oriented, and that (unless otherwise noted) ∂M is also C∞.

    For each point p M, the tangent space to M at p will be denoted by Mp; and the tangent bundle, that is, the union of all the tangent spaces of M endowed with its natural differentiable structure, will be denoted by TM.

    into manifolds are usually referred to as paths.

    Given p in our manifold M and a C¹ real-valued function f defined on a neighborhood of p, then to each ξ ∈ Mp is associated the directional derivative of f at p in the direction ξ, denoted by ξf, and defined by

    (1)

    where ω(t) is any path in M satisfying ω(0) = p and ω′(0) = ξ. The map Mp given by ξ → ξf is linear. For functions f, h one has

    (2)

    (3)

    The Riemannian metric on M associates to each p M an inner product on Mp, which we denote by〈, 〉. The associated norm will be denoted by ||. The Riemannian metric is C∞ in the sense that if X, Y are C∞ vector fields on M, then 〈X, Y〉 is a C∞ real-valued function on M.

    Definition 1

    Given a real-valued Ck, k 1, function f on M, we define the gradient of f, grad f, to be the vector field on M for which

    (4)

    for all ξ ∈ TM.

    There is no question of the existence of grad fξf is linear on each tangent space. The calculation below (cf. (22)) shows that grad f is a Ck−1 vector field. One has for functions f, h,

    (5)

    (6)

    Whereas the differentiation of functions on a manifold is naturally determined by the differentiable structure, the differentiation of vector fields, on the other hand, is not naturally determined but involves the choice of a connection, that is, a rule which associates to each p M, ξ ∈ Mp, and C¹ vector field X defined on a neighborhood of pξ X Mp satisfying

    (7)

    (8)

    where X, Y are C¹ vector fields, and f is a C¹ real-valued function, all defined on a neighborhood of pξ X be linear in ξ ∈ Mp, and that for X, Y CX Y is a Cξ X is traditionally referred to as the covariant derivative of X with respect to ξ.

    The Riemannian metric on M does determine a unique connection, called the Levi–Civita connection, when one adds the requirements

    (9)

    where [, ] is the Lie bracket of the indicated vector fields, and

    (10)

    for all C¹ vector fields X, Y on M, and ξ ∈ TM.

    Definition 2

    Given a Ck1, vector field X on M, define the real-valued function the divergence of X, div X, by

    (11)

    where ξ ranges over Mp.

    The divergence of X is a Ck−1 function on M; and for the function f, and vector fields X, Y on M, we have

    (12)

    (13)

    Definition 3

    For any Ck 2, function f on M we define the function the Laplacian of f, Δf, by

    (14)

    One has that Δf Ck-2; and for functions f, h we have

    (15)

    (16)

    (17)

    We now calculate the expressions of the above operators in local coordinates. Let U be an open set in M, and x: U n a diffeomorphism of U n, that is, a chart on M. Then associated to the chart are n coordinate vector fields, written as ∂/∂xj or as ∂j, j = 1, …, n. The directional derivative determined by ∂j satisfies

    (18)

    where p is any point in U and f is any differentiable function defined on a neighborhood of p. For each p U, the vectors {∂1(p), …, ∂n(p)} span Mp. Therefore, for

    (19)

    and f C¹, we have

    (20)

    For the given Riemannian metric, define

    (21)

    where j, k = 1, …, n, and (henceforth) det denotes the determinant. Then

    for all ξ. From (4) and (20) we have

    (22)

    To calculate the divergence, one certainly has n, i, j, k = 1, …, n, known as Christoffel symbols, determined by

    (23)

    on U. Thus for ξ given by (19) and X given by

    (24)

    we have, using (7) and (8),

    (25)

    Therefore

    (26)

    .

    Recall that if X is given by (24) and Y is given by

    (27)

    then

    (28)

    One immediately has, by setting X = ∂j and Y = ∂k, that (9) is equivalent to

    (29)

    for all j, k, l = 1, …, n. If one also sets ξ = ∂i then (10) becomes

    (30)

    from which one deduces, with (29), by a standard argument,

    (31)

    In particular, we have, by (26) and (31),

    that is,

    (32)

    In the above calculation we have let tr denote the trace, ∂jG the matrix obtained from G by differentiating each of the entries with respect to the jth coordinate. To go from the second line to the third, one uses the standard formula for differentiating determinants.

    Finally, (22) and (32) combine to imply

    (33)

    2 Green’S Formulas

    M is our given Riemannian manifold. With the Riemann metric is associated an integration theory in which (i) the function f is measurable if, for every chart x: U n on M, f x−1 is measurable on the image of U n, (ii) for every covering {xα: Un : α ∈ I}, where I is some set, of M by charts with subordinate partition of unity {ϕα : α ∈ I}, the Riemannian measure on M is given by the density

    (34)

    where dx¹α … dxnα is the density of Lebesgue measure on xα(Un, and gα is the determinant defined in (21) for the chart xα: Undx¹ … dxn on the domain U is independent of the mapping function x. The partition of unity is then the device with which the measure is defined globally on M.

    Formula (32) admits the following interpretation: Given the vector field X on M, let {Φt} denote the induced flow on M. Fix any compact set K in MThus div X measures the infinitesimal distortion of volume by the flow generated by X.

    Divergence Theorem (I)

    If X is a C¹ vector field on M with compact support, then

    (35)

    Green’s Formulas (I)

    Let h C¹, f C² be functions on M such that h(grad f) has compact support. Then

    (36)

    If we also assume that h C² and both f, h have compact support, then

    (37)

    To derive Green’s formula from the divergence theorem, one simply lets X = h(grad f) and substitutes (16) into (35). Equation (37) follows easily from (36).

    Now assume that M has boundary ∂M, with induced Riemannian metric and measure, the density of the measure being denoted by dA. Let v denote the outward unit normal vector field on ∂M.

    Divergence Theorem (II)

    Let X be a vector field which is C. Then

    (38)

    Green’s Formulas (II)

    Let h C), f C) such that h(grad f. Then

    (39)

    If we also have h C) and both f, h , then

    (40)

    3 Basic Facts for Eigenvalue Problems

    We let L²(M) be the space of measurable functions f on M On L²(M) we have the usual inner product, and induced norm, given by

    (41)

    for f, h L²(M). With the inner product, L²(M) is a Hilbert space.

    Our fundamental interest is in the following eigenvalue problems.

    Closed Eigenvalue Problem: Let M be compact, connected. Find all real numbers λ for which there exists a nontrivial solution ϕ ∈ C²(M) to

    (42)

    Neumann Eigenvalue Problem: For ∂M compact and connected, find all real numbers λ for which there exists a nontrivial solution ϕ ∈ C²(M) ∩ C) to (42), satisfying the boundary condition

    (43)

    on ∂M (recall: v is the outward unit normal vector field on ∂M).

    Dirichlet Eigenvalue Problem: For ∂M compact and connected, find all real numbers λ for which there exists a nontrivial solution ϕ ∈ C²(M) ∩ C) to (42), satisfying the boundary condition

    (44)

    on ∂M.

    Mixed Eigenvalue Problem: For ∂M compact and connected, N an open submanifold of ∂M, find all real numbers λ for which there exists a nontrivial solution ϕ ∈ C²(M) ∩ C¹(M N) ∪ C) to (42), satisfying the boundary conditions

    (45)

    The desired numbers λ are referred to as eigenvalues of Δ, and the vector space of solutions of (42) for a given eigenvalue λ [eq. (42) is linear in ϕ], its eigenspace. The elements of each eigenspace are called eigenfunctions.

    Theorem 1

    and each associated eigenspace is finite dimensional. Eigenspaces belonging to distinct eigenvalues are orthogonal in L²(M), and L²(M) is the direct sum of all the eigenspaces. Furthermore, each eigenfunction is C.

    We first note that as soon as we know that the eigenfunction ϕ ∈ C²(M) ∩ C), then its eigenvalue λ must be nonnegative. Indeed one sets f = h = ϕ and applies the appropriate Green formula [viz., (36) or (39)] to obtain

    (46)

    From 1 = 0, and in the Dirichlet and mixed (N ≠ ∂M1 > 0.

    We also note that the orthogonality of distinct eigenspaces is a direct consequence of the Green formulas (37) and (40). Indeed, let ϕ, ψ be eigenfunctions of the respective eigenvalues λ, τ. Then

    and the remark follows.

    We refer to the dimension of each eigenspace as the multiplicity of the eigenvalue. with each eigenvalue repeated according to its multiplicity.

    If ϕ1, ϕ2, … is an orthonormal sequence (in L²(M)) of eigenfunctions so that ϕj is an eigenfunction of λj for each j = 1, 2, …, then ϕ1, ϕ2, … is a complete orthonormal sequence of L²(M). In particular, for f L²(M) we have

    (47)

    in L²(M), and

    (48)

    These last two formulas are referred to as Parseval identities.

    Weyl’s Asymptotic Formula: (Weyl [1]): In each of the above eigenvalue problems, let Nλ. Then

    (49)

    as λ → + ∞, where ωn n, and vol M is (henceforth) the volume of M. In particular,

    (50)

    as k → + ∞.

    For now, we just consider Weyl’s formula when n will have its usual Riemannian structure, and, for the boundary value eigenvalue problems, we will have M equal to some compact interval, say, [0, L]. For any function f on M = [0, L], Δf = f″, and eq. (42) is

    (51)

    If we consider the Dirichlet eigenvalue problem on [0, L] then the associated boundary condition is

    (52)

    from which one has, for k = 1, 2, …,

    (53)

    and

    (54)

    Thus we obtain equality in the Weyl formula for all k.

    For the Neumann eigenvalue problem the boundary conditions are

    (55)

    from which one has

    (56)

    and the Weyl formula is again satisfied.

    In the case of the mixed eigenvalue problem one has boundary conditions

    (57)

    or

    (58)

    and

    (59)

    Weyl’s formula follows immediately.

    Finally, if we set

    (60)

    ² of length L, parametrized with respect to arc length along the circle, and the eigenfunctions of the circle consist of the L-periodic solutions of (51) (cf. Section II.2 for more details).

    Enjoying the preview?
    Page 1 of 1