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

Only $11.99/month after trial. Cancel anytime.

Variational Methods for Eigenvalue Problems: An Introduction to the Methods of Rayleigh, Ritz, Weinstein, and Aronszajn
Variational Methods for Eigenvalue Problems: An Introduction to the Methods of Rayleigh, Ritz, Weinstein, and Aronszajn
Variational Methods for Eigenvalue Problems: An Introduction to the Methods of Rayleigh, Ritz, Weinstein, and Aronszajn
Ebook343 pages4 hours

Variational Methods for Eigenvalue Problems: An Introduction to the Methods of Rayleigh, Ritz, Weinstein, and Aronszajn

Rating: 5 out of 5 stars

5/5

()

Read preview

About this ebook

The importance of eigenvalue theory in pure and applied mathematics, and in physics and chemistry, makes it incumbent on students to understand the various methods of approximate calculation of eigenvalues. It is especially important to develop such methods in a general and theoretical manner, if only to avoid missing opportunities for particular applications. This book does just that, approaching the topic from a purely mathematical standpoint.
Because variational methods are particularly well adapted to successive approximation, this book gives a simple exposition of such methods, not only of the familiar Rayleigh-Ritz method, but especially of the related methods — the Weinstein method, Weinstein-Aronszajn method, and others.
To make the book accessible to a broad range of students, little mathematical knowledge is presupposed beyond the elements of calculus. Where specialized knowledge is required — as it is in the discussion of direct methods in the calculus of variations and the theory of completely continuous operators in Hilbert space — the requisite material is developed in full.
The first nine chapters, written in elementary style, discuss the general theory of variational methods with special reference to the vibrating plate. In the last chapter, the information gained thereby is extended, in a less elementary way, to more general cases. Exercises are provided throughout to illuminate the ideas and methods developed in the text.
LanguageEnglish
Release dateMay 24, 2012
ISBN9780486165806
Variational Methods for Eigenvalue Problems: An Introduction to the Methods of Rayleigh, Ritz, Weinstein, and Aronszajn

Related to Variational Methods for Eigenvalue Problems

Titles in the series (100)

View More

Related ebooks

Physics For You

View More

Related articles

Reviews for Variational Methods for Eigenvalue Problems

Rating: 5 out of 5 stars
5/5

1 rating0 reviews

What did you think?

Tap to rate

Review must be at least 10 words

    Book preview

    Variational Methods for Eigenvalue Problems - S. H. Gould

    INDEX

    INTRODUCTION

    WE shall be chiefly concerned with the methods of Rayleigh—Ritz and of Weinstein for approximate calculation of eigenvalues. The concept of an eigenvalue is of great importance in both pure and applied mathematics. A physical system, like a pendulum, a vibrating string, or a rotating shaft has connected with it certain numbers characteristic of the system, namely the period of the pendulum, the frequencies of the various overtones of the string, the critical angular velocities at which the shaft will buckle, and so forth. The German word eigen means characteristic and the hybrid word eigenvalue is used for characteristic numbers in order to avoid confusion with the many other uses in English of the word characteristic. The desire to avoid a half-German, half-English word has led some mathematicians to use other names for eigenvalues. But the word eigenvalue has by now gained such a firm place for itself, not only in mathematics but in physics and chemistry, that no change can be made. There can be no doubt that eigenvalue will soon find its way into the standard dictionaries. In fact, it ought to be there already. The English language has many such hybrids; for example, liverwurst.

    Since the present book as a whole is about mathematics rather than physics, the differential equations of the vibrating string, rod, membrane, plate, etc., are taken for granted. These equations, and their generalizations throughout the book, will be of elliptic type (compare Chapter X, section 2). For their physical derivation the reader may consult any standard text on the differential equations of mathematical physics, e.g. Webster 2. This form of reference, namely the author’s name followed by a numeral, is used throughout to indicate the bibliography at the end of the book.

    The methods under discussion depend on the fact that the desired eigenvalues may be considered from two points of view, differential and variational. From the differential point of view, eigenvalues appear as the possible values for a parameter in a differential equation; from the variational they are the maxima or minima of certain expressions, a variational problem being taken to be any problem of finding extreme values, i.e. maxima or minima. If we are unable to find the exact eigenvalues of a differential problem we cannot easily proceed with successive approximations. But if the problem can be put into a corresponding variational form, it can then be modified in either of two ways and thereby changed, in many cases, into a problem that can be solved.

    In the first place, we can introduce more stringent prescribed conditions. Then the desired minimum will be raised, or at least not lowered, and we will obtain upper bounds for the eigenvalues of the original problem. This is the essence of the Rayleigh—Ritz method. On the other hand, we can weaken the prescribed conditions, so that the minimum will be lowered, or at least not raised, and we will get lower bounds for the desired eigenvalues. This is the essence of the Weinstein method.

    A combination of the two methods will confine the eigenvalues within close bounds, but the question remains open whether our successive approximations actually converge to the desired results. It was to answer this question, as well as certain questions of existence, that the ideas of Aronszajn were developed. These ideas, discussed in later chapters of the book, involve the concepts of Hilbert space and functional completion, but it must be emphasized that, although more abstruse developments of this kind widen the applicability of the methods and are indispensable for further theoretical progress, still the original methods of Rayleigh—Ritz and Weinstein can be used by the practical computer without reference to them. An example is given in section 9 of Chapter VI. Moreover, it is hoped that the detailed way in which the theory of Hilbert space is harnessed to a numerical problem, namely to evaluate or bound eigenvalues for relatively simple physical problems, will be attractive to applied mathematicians.

    The procedure for changing a differential problem into corresponding variational form is clearly illustrated in the finite-dimensional case, that is, for the top or pendulum in contrast to continuous media like strings or plates. Chapters I and II therefore deal with this case. Chapters III and VI give the classical methods of Rayleigh—Ritz and Weinstein in their original form. Chapter VIII presents what may be called the equivalent problem for Hilbert space, where existence and convergence are readily proved. Then Chapter IX shows how, on the basis of concepts introduced in Chapters IV, V, and VII, we can prove the equivalence of the two problems, namely the classical differential problem for, say, the clamped plate, and the corresponding problem for a linear operator in Hilbert space. For the clamped plate, this equivalence was first proved in the joint paper of Aronszajn and Weinstein 1.

    In the final Chapter X we discuss the ideas of Aronszajn for more general differential problems.

    I

    SYSTEMS VIBRATING WITH A FINITE NUMBER OF DEGREES OF FREEDOM

    1. Differential equations for small vibrations about a position of stable equilibrium. The motion of a system with a finite number of degrees of freedom, such as a top or a pendulum, is easy to visualize and forms a helpful guide to the more complicated motion of elastic media, such as strings, membranes, or plates. Moreover, the methods of greatest importance for the latter type of problem depend on the results of the present chapter. For example, the Rayleigh-Ritz method consists of successive approximation to continuous media by the use of systems with a finite number of degrees of freedom, and in the Weinstein method too, as we shall see, a fundamental role is played by finite-dimensional systems.

    The position of a system with n degrees of freedom is defined at any time t by the values of n parameters or generalized coordinates q1, q2,..., qn, so that solving the problem of motion consists of finding these parameters qi as functions of t. We consider only conservative systems, that is, systems with a potential energy, which implies, in particular, that there are no frictional forces. We also assume that there are no external forces varying with the time. The physical characteristics of the system then enable us to express its kinetic and potential energies in terms of qi , and shall see that such a motion must be a superposition of n simple, periodic motions whose amplitudes, assuming them to be sufficiently small, are arbitrary but whose frequencies are completely determined by the physical properties of the system.

    equal to zero and arrange, if necessary, by the addition of suitable constants to the coordinates qi that this minimum is reached for the values q1 = q2 = ... = qn = 0.

    Then, by the well-known theorem of Dirichlet (see, e.g., Ames and Murnaghan 1) the position q1 = q2 = ... = qn = 0 is one of stable equilibrium in the sense that if the values of qi are sufficiently small for t = 0, they will remain small for every subsequent time.

    , we have

    (i = 1, 2,..., n),

    the notation indicating that the partial derivatives are to be taken at the point q1 = q2 = ... = qn = 0.

    we may write

    up to terms of higher order, which may be neglected.

    is represented by a quadratic form (homogeneous polynomial of second degree) in the position coordinates qi,

    where the coefficients

    are known constants and the summation is taken, as everywhere in this chapter, from 1 to n.

    Since aik = aki, is symmetric, is positive definite, that is, it vanishes for

    q1 = q2 = ... = qn = 0

    but is otherwise positive.

    To see that this is true, we set up rectangular axes and let x1, x2, x3 be the coordinates of a particle p of mass m. Since the xi are functions of the qi, we have

    of the particle p are functions of the position coordinates qi. Thus the kinetic energy of the whole system, being the sum of the kinetic energies of its particles, may be written in the same form, namely

    If we now develop each of the functions bik(q1,...,qn) in a Taylor series about q1 = q2 = ... = qn = 0, we may neglect every term except the constant bik(0, 0,..., 0), since any term of first or higher degree in qi which is at least of third degree in the small numbers qi Writing bik for bik(0, , as desired. Here the bik = bki is positive definite.

    The actual motion of the system, as determined by its kinetic and potential energies, is given by the classical equations of Lagrange (see, e.g., Webster 1), which in this case reduce to

    (i = 1, 2,..., n).

    and the second with respect to qi. We may rewrite them so that all partial derivatives are taken with respect to qi , which is the same form in the qi

    Then

    so that

    and the Lagrange equations become

    (i = 1, 2,..., n).

    Finding the motion of the physical system thus consists of solving this system of n linear ordinary differential equations for the qi as functions of t. Difficulty arises only because each equation of the system contains all the variables qi. It will be our object to introduce new variables xi by means of a reversible linear transformation

    with constants qik for which det∣qiktake the simple (so-called canonical) forms

    with constants λi. depending on the known coefficients bik and aikis positive definite, these constants λi must be positive.

    In terms of the parameters xi the Lagrange equations are

    or

    ẍi+λixi = 0.

    Since each of these equations contains a single variable xi, it may be solved at once. The general solution is

    where the ai and θi are constants of integration.

    The particularly simple and important case (see next section) in which a1= ... = ah−1, = ah+1 = ... = an = 0, while ah ≠ 0, is called the hth eigenvibration of the system. Its equations of motion are given by

    x1 = 0,   x2 = 0,   ...,   xh = ah sin νh(t+θh),   ...,   xn = 0,

    or in terms of the original parameters qi by

    from which it is seen that in the hth eigenvibration, for any h, all the n coordinates qi(t) oscillate with the same frequency and phase, and with amplitudes proportional to q1h, q2h,..., qnh, where the qih are the coefficients of our desired transformation.

    The importance of eigenvibrations is due not only to such well-known physical applications as avoidance of resonance (see Chapter II, section 10) but also to the fact that any possible motion of the system is described by the equations

    (k = 1, 2,..., n)

    and is therefore simply a superposition of eigenvibrations.

    EXAMPLES

    Example 1. Two masses 2m1 and 2m2 are connected by springs as indicated in Fig. 1. The spring constants are 2(k1−k2) and 2k2, a spring being said to have spring constant k if, when its elongation is e, it exerts a restoring force ke. The two masses, which slide without friction along the horizontal axis, are slightly displaced from their position of equilibrium and we wish to examine the subsequent motion. Letting q1(t) and q2(t) denote the displacements of the masses we find, after an elementary calculation of the kinetic and potential energies, that

    so that the Lagrange equations are

    FIG. 1.

    To take a special case, we set m1 = m2 = 1, k1 = 5, k2 = 2, so that

    Let us now make the following transformation (the method of calculating its coefficients is the subject of this chapter; see, e.g., section 11):

    Then it is readily verified that, as desired,

    so that λ1 = 1, λ2 = 6 and the Lagrangian equations become 1+x1 = 0 and 2+6x2 = 0, whose general solution is x1 = a1sin(t+θ1) and

    Thus the solution in terms of the original parameters qi is given by

    Example 2. Consider the motion of a string stretched along the x-axis with its ends fastened at x = 0 and x = l. The problem becomes n-dimensional if we suppose that n beads, each of mass m, have been fastened to the string at the points x = a = l/(n+1), 2a,..., na, and that the mass of the string can be neglected. We also assume that each of the beads moves in a horizontal straight line in the y-direction (e.g. the beads are supported on a smooth horizontal table), so that our problem consists of finding the displacement perpendicular to the string, call it qi(t), of each bead as a function of t.

    is easily calculated as follows. Since we are dealing with small displacements, we may assume that the tension of the string, say S, is constant, so that the only force acting on a bead is the y-component of the difference in tension of the adjacent parts of the string.

    FIG. 2.

    But this component of the tension in the segment Bi Bi−1 (see Fig. 2) is equal to S cos Bi−1BiPi = (S/a)(qi−qi−1), so that the force Fi on the ith bead is given by

    (i = 1, 2,..., n; q0 = qn = 0).

    Thus the potential energy is represented by

    is seen at once to be given by

    now appears as

    and the Lagrange equations are

    (i = 1, 2,..., n).

    Let us now make the transformation qi = ∑ qikxk, with

    qik = 2m−1(n+1)−1 sin{πik(n+1)−1

    take on the canonical forms

    with

    Thus the general solution of the problem is given by

    (k = 1, 2,..., n),

    or, in terms of the original parameters qi, by qi = ∑ qikxk.

    EXERCISES

    FIG. 3.

    1. A homogeneous bar OA of length 2a and mass m1 turns freely in the vertical plane about the fixed point O, while a second bar AB of length 2b and mass m2 turns in the same plane about A (see Fig. 3). Taking the angles q1, q.

    Answer.

    FIG. 4.

    2. Consider the double torsional pendulum constructed as follows (see Fig. 4). Two heavy horizontal disks D1 and D2 are suspended on a light vertical wire whose ends P and Q are fixed. Let k1 and k2 be the torsional constants as indicated in the diagram. (A spring is said to have torsional constant k if, when turned through an angle q, it exerts a restoring torque kq.) Let I1 and I2 be the moments of inertia of the disks D1 and D2 about the axis PQ. Letting q1 and q.

    Answer.

    is a strict minimum at the origin q1 = ... = qn = 0, then the equilibrium is stable in the sense that, after choice of ∈ > 0, a number δ > 0 can be found such that, if qi(0) and qi(0) are less than δ for

    i = 1, 2,..., n,

    then qi(tremain less than ∈ for all t. (See, e.g., Ames and Murnaghan 1.)

    BIBLIOGRAPHICAL NOTE

    The important principle of superposition of small motions was first advanced by Daniel Bernoulli. The above application of generalized coordinates to a system with finitely many degrees of freedom was invented by Lagrange. The systematic introduction into physics of the eigenvibrations of such a system is due to Thompson and Tait 1, who called them normal vibrations, a name still in common use. For the material of the present chapter see Routh 2, Rayleigh 1, Courant-Hilbert 1, Webster 1, or any standard text on mechanics. The simultaneous reduction of two quadratic forms to canonical form can be carried out in two different ways. A first method, which is purely algebraic and can be most concisely presented in terms of matrices (see, e.g. Perlis 1), is not suited to our purposes. A second method, used by Rayleigh and others, depends on variational principles and is discussed in detail in the present chapter.

    2. Normal coordinates. Eigenvibrations. If we succeed in finding the above transformation from the qi coordinates to the xi, that is, if we find the constants qik, then, no matter how it may be set in motion, the features of the system represented by the coordinates xi (called its normal coordinates) will vary in a simple harmonic motion xi = ai sin νi(t+θi) with amplitude ai and with frequency νi, where νi

    Enjoying the preview?
    Page 1 of 1