Everand Logo

Welcome to Everand!

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

    Only $11.99/month after trial. Cancel anytime.

    Splines and Variational Methods
    Splines and Variational Methods
    Splines and Variational Methods
    Ebook537 pages4 hours

    Splines and Variational Methods

    By P. M. Prenter

    5/5

    ()

    Read preview

    About this ebook

    One of the clearest available introductions to variational methods, this text requires only a minimal background in calculus and linear algebra. Its self-contained treatment explains the application of theoretic notions to the kinds of physical problems that engineers regularly encounter. The text’s first half concerns approximation theoretic notions, exploring the theory and computation of one- and two-dimensional polynomial and other spline functions. Later chapters examine variational methods in the solution of operator equations, focusing on boundary value problems in one and two dimensions. Additional topics include least squares and other Galerkin methods. Many helpful definitions, examples, and exercises appear throughout the book. A classic reference in spline theory, this volume will benefit experts as well as students of engineering and mathematics.
    LanguageEnglish
    PublisherDover Publications
    Release dateNov 26, 2013
    ISBN9780486783499
    Splines and Variational Methods

    Related to Splines and Variational Methods

    Titles in the series (100)

    View More
    View More

    Related ebooks

    Mathematics For You

    View More
    View More

    Related podcast episodes

    Related articles

    • Article

      Physicists Attack Math’s $1,000,000 Question

      Apr 4, 2017

      Physicists are attempting to map the distribution of the prime numbers to the energy levels of a particular quantum system.

      4 min read

    • Article

      Kolmogorov Complexity and Our Search for Meaning

      Aug 2, 2018

      Was it a chance encounter when you met that special someone or was there some deeper reason for it? What about that strange dream last night—was that just the random ramblings of the synapses of your brain or did it reveal something deep about your u

      8 min read

    • Article

      Choosing CFD carefully

      Jun 7, 2019

      Despite the progress made in computational methods in the past decade, there is still a level of compromise required by the CFD engineer in determining how best to model the real world. While the aerospace industry has developed highly specific and e

      3 min read

    • Article

      The Math That Takes Newton Into the Quantum World

      Feb 28, 2019

      In my 50s, too old to become a real expert, I have finally fallen in love with algebraic geometry. As the name suggests, this is the study of geometry using algebra. Around 1637, René Descartes laid the groundwork for this subject by taking a plane,

      9 min read

    • Article

      The Twin Prime Hero: Rags, riches, and fame in mathematics.

      Dec 29, 2016

      Yitang “Tom” Zhang spent the seven years following the completion of his Ph.D. in mathematics floating between Kentucky and Queens, working for a chain of Subway restaurants, and doing odd accounting work. Now he is on a lecture tour that includes st

      8 min read

    • Article

      When Beauty Gets In The Way Of Science

      Apr 18, 2019

      The biggest news in particle physics is no news. In March, one of the most important conferences in the field, Rencontres de Moriond, took place. It is an annual meeting at which experimental collaborations present preliminary results. But the recent

      7 min read

    • Article

      Quantum Mechanics 101

      Dec 5, 2019

      Quantum Mechanics 101
      7 min read

    • Article

      Numismatics and Nanotechnology

      Jan 20, 2023

      Nanotechnology involves science and applications on the scale of a nanometer, which is one-billionth of a meter. To give you some idea of size, a cube measuring one nano-meter on a side would hold 176 hydrogen atoms. It might be easier to visualize t

      2 min read

    • Article

      Are Neural Networks About to Reinvent Physics?

      Nov 21, 2019

      Can AI teach itself the laws of physics? Will classical computers soon be replaced by deep neural networks? Sure looks like it, if you’ve been following the news, which lately has been filled with headlines like, “A neural net solves the three-body p

      9 min read

    • Article

      Fanning The Flame

      Sep 13, 2023

      Fanning The Flame
      3 min read

    • Article

      The Universe Began With a Big Melt, Not a Big Bang

      Oct 5, 2017

      There are two tantalizing mysteries about our universe, one dealing with its final fate and the other with its beginning, that have intrigued cosmologists for decades. The community has always believed these to be independent problems—but what if the

      10 min read

    • Article

      Designing Landscape Architectural Education: Studio Ecologies for Unpredictable Futures

      Apr 30, 2023

      Back in 1973, in an article entitled “Resilience and Stability of Ecological Systems,” the Canadian ecologist C. S. Holling seminally contrasted urban infrastructure’s engineered fail-safe intentions with ecology’s safe-tofail “unpredictability.” Fif

      4 min read

    Related categories

    Reviews for Splines and Variational Methods

    5/5

    1 rating0 reviews

    What did you think?

    Tap to rate

    Review must be at least 10 words

      Book preview

      Splines and Variational Methods - P. M. Prenter

      Index

      1

      INTRODUCTORY IDEAS

      1.1A SIMPLY STATED PROBLEM

      1.2LINEAR SPACES

      1.3NORMED LINEAR SPACES

      1.4THE SPACE L2[a, b]

      1.5BASIS FOR A LINEAR SPACE

      1.6APPROXIMATING FROM FINITE DIMENSIONAL SUBSPACES

      1.1 A SIMPLY STATED PROBLEM

      The need for good techniques for the approximation of functions arises in many settings; one of these is the numerical solution of differential equations. For example, suppose you are given the differential equation

      subject to the boundary conditions

      where α and β are constant and a(t), b(t), and f(t) are functions of t defined on the interval [a, b]. Moreover, suppose you know that this equation, subject to the boundary conditions, has a unique solution x(t) which you would like to find. Our first problem then is

      PROBLEM 1

      How do we find x(t)?

      As those who have worked to any extent in ordinary differential equations know, the answer to this query is decidedly gloomy. In particular

      Answer. For most choices of a(t), b(t), and f(t) we cannot find x(t) exactly.

      This being the case, we compromise. Since we cannot find x(t) exactly, we can try to find it approximately. This leads us to a new problem.

      PROBLEM 2

      (t) to the solution x(t) of our differential equation?

      This problem is far more tractable, and there are many ways of answering it. All possible solutions depend in some way on the answer to yet another problem.

      PROBLEM 3

      Given a function x(t(t) make good approximations to x(t)?

      and to the companion problems

      PROBLEM 4

      What is meant by a good approximation?

      and

      PROBLEM 5

      (t) to a given function x(t)?

      Providing some answers to these simple questions and their two-dimensional analogs is precisely what this book is all about. Since linear spaces, subspaces, norms, and basis are notions fundamental to all we do in the sequel, we start with a minor digression to define these entities.

      1.2 LINEAR SPACES

      A real linear space X is simply a set of mathematical objects called vectors which add according to the usual laws of arithmetic and can be multiplied by real numbers in accord with the usual laws of arithmetic. Specifically to qualify as a real linear space, elements of X must satisfy the following conditions or axioms.

      For all x, y, and z, in X and for all real numbers α and β, αx is in X, x + y is in X, x + y = y + x, (x + y) + z = x + (y + z), 1·x = x, (α + β)x = αx + βx, α(βx) = (αβ)x, and α(x + y) = αx + αy. There exists a zero vector in X with the property that x = x for all x. Finally, for each x there is a unique vector – x, called the inverse of x, such that – x + x .

      If all these axioms are satisfied when multiplication is multiplication by complex numbers, we say X is a complex linear space. We usually deal with the real linear spaces.

      Examples abound. Among these are the set E² of vectors in the plane with addition defined as coordinatewise addition and the set C[a, b] of functions f(t) continuous on the closed interval [a, b]. In each case we must first define addition and scalar multiplication.

      EXAMPLE 1 THE REAL PLANE E²

      The set X = E² is simply the set

      of all ordered pairs of real numbers or vectors in a plane. Addition is coordinate-wise addition, and multiplication is coordinatewise multiplication. In particular, given a real number α and vectors x = (xl, x2) and y = (y1, y2), we define

      = (0, 0) is the zero vector (see Figure 1.1).

      The real linear space C[a, b] is much more interesting.

      EXAMPLE 2 THE SPACE C[a, b]

      The space C[a, b] is simply the set of all functions continuous on [a, b]. To define addition and scalar multiplication, let α be any real number and let f(t) and g(t) be two continuous functions from C[a, b]. We define f + g and αf in the usual way. That is,

      and

      Figure 1.1 Addition of vectors in the plane E².

      Noting that the sum of any two continuous functions is a continuous function and that a constant times a continuous function is again continuous, it is easy to verify that C[a, b] forms a real linear space. The zero vector of this space is the function that vanishes identically on the interval [a, b].

      A space to which we shall take frequent recourse is

      EXAMPLE 3 Pn[a, b], POLYNOMIALS OF DEGREE n.

      A real polynomial p(t) of exact degree n or less in one variable t is a function

      where a0, a1, …, an are given real numbers and t is a real variable. The polynomial p(t) is said to have exact degree n, if and only if an ≠ 0. Thus 2t³ + 5t – 1 is of degree 4 but is of exact degree 3. Let

      Thus Pn[a, b] is the set of all polynomials of exact degree n or less, defined on the interval [a, b]. Of course adding two polynomials of degree n or less produces a polynomial of degree n or less, whereas multiplying a polynomial of degree n or less by a constant gives a polynomial of degree n or less. Knowing this, it is easily checked that Pn[a, b] is a real linear space. The zero in this space is again the zero function, which is the same as the zero polynomial. Note too that since every polynomial is continuous, Pn[a, b]⊂C[a, b] (read Pn[a, b] is a subset of C[a, b]).

      We are interested in special subsets of linear spaces, which are known as subspaces. A subset M of a linear space X is called a linear subspace (or simply a subspace) of X if

      M is a subset of X and for each x and y in M and for any pair of scalars α and β, αx + βy belongs to M.

      For example, a real line passing through the origin is a linear subspace of the plane E². Also, the set Pn[a, b] is a subspace of C[a, b]. We use these concepts repeatedly in the sequel.

      EXERCISES

      1.Prove that C[a, b] is a linear space.

      2.Let P.C.[a, b] denote the set of all functions that are piecewise continuous on [a, b]. In particular, a function f(t) belongs to P.C.[a, b] if and only if it has at most a finite number of discontinuities on [a, b] and ∫ba[f(t)]² dt < ∞. Is P.C.[a, b] a linear space? Prove your answer. If f ∈ P.C.[a, b], does f′(t) ∈ P.C.[a, b]? Why? If f(t) is a simple jump function, such as

      does either f(t) or f′(t) belong to C[a, b]? Explain.

      3.Let X = {(x0, y0) + (x, y): x and y are real and x0 and y0 are constants}. Is X a linear subspace of E²? Explain.

      4.Let X = P.C.p[a, b] = {f(t): f(p–1)∈C[a, b] and f(p)∈P.C.[a, b]}. Is X a linear space? Why? If f ∈ X, does the fundamental theorem of calculus hold? In particular, does

      for all t, t0 in [a, b]? Explain.

      1.3 NORMED LINEAR SPACES

      Actually, we are interested in more than just linear spaces. We want linear spaces in which we can assign a notion of length ||x|| to each vector x in X. Such linear spaces are known as normed linear spaces, and the real number ||x|| is referred to as the norm of x. In particular, a norm is a real valued function defined on a linear space X having the following properties for each real number α and each pair of vectors x and y from X:

      1. ||x|| > 0 unless x = 0

      2. ||αx|| = |α| · ||x||

      3. ||x + y|| ≤ ||x|| + ||y|| (triangle inequality).

      The reader can check that these properties of a norm do coincide with usual geometric properties of length. Each of the linear spaces we have cited in our examples is easily made a normed linear space through an appropriate choice for || ||. For example, if X = E², the usual choice for the length or norm || ||2 of a vector x = (x1, x2) in the real plane is

      But there are many other choices available. For example, the function || ||1 defined by

      is a norm as is the function || ||∞, defined by

      To prove that each of these function is a norm, you must verify that each one satisfies each of the conditions 1, 2, and 3 required of a norm. For example, to see that || ||1 satisfies condition 1, note that ||x||1 = 0 implies |x1| + |x2| = 0. But this is possible if and only if x1 = x2 = 0. Thus ||x||1 = 0 if and only if x = (0, 0), the zero vector. Moreover, since ||x||1 is clearly nonnegative for all x, we see that condition 1 is satisfied. Conditions 2 and 3 follow readily from elementary properties of absolute values of real numbers.

      We are especially interested in the Tchebycheff or uniform norm on the space C[a, b].

      DEFINITION TCHEBYCHEFF NORM

      Let f(t)∈C[a, b]. The real-valued function || || defined by

      is known as the Tchebycheff norm of f.

      One must, of course, prove that this function actually is a norm. This is easily accomplished because of elementary properties of absolute value of real numbers. In particular, ||f|| ≥ 0, since |f(t)| ≥ 0 for all t in [a, b]. Also ||f|| = 0 only if |f(t)| =0 for all t in [a, b]. Thus ||f|| ≥ 0 unless f(t) ≡ 0 on [a, b]. Moreover, |αf(t)| = |α| · |f(t)| implies

      Thus 1 and 2 are clearly true. Condition 3 follows from the triangle inequality |f(t) + g(t)| ≤ |f(t)| + |g(t)| for absolute values of real numbers. Thus the Tchebycheff norm is indeed a norm.

      A second norm on C[a, b] which is also very important to us is the so-called L2 norm || ||2. In particular

      DEFINITION L2 NORM

      Let f(t)∈C[a, b], The real valued function || ||2 defined by

      is known as the L2 norm of f.

      Verification that || ||2 is a norm on C[a, b] is left as an exercise (see Exercise 2).

      Defining a norm on a linear space X introduces the companion notion of the distance ||x – y|| between two points x and y belonging to X. For example, if X = E², x = (1, 3) and y = (–2, 4), then

      On the other hand, if X = C[a, b] with the Tchebycheff norm, and f(t) and g(t) are any two functions belonging to X,

      To take a specific case, let f(t) = cos πt, g(t) = t² and X = C[0, 1]. Then

      (see Figure 1.2).

      Given a continuous function f(t) from C[a, b], it is helpful to consider the set of all functions g(t) from C[a, b] for which ||f – g|| < ∈. Such collections of functions, known as ∈-neighborhoods of f, are nice things—we can draw pictures of them. Suppose, for example, f(t) = sin2t + 2, where 0 ≤ t ≤ π, and X = C[0, π]. Then {g ∈ C[0, π]:||sin2t + 2 – g|| < ∈} is the family of all continuous functions g(t) from C[0, π] which thread through a tube of width 2∈, symmetric about the graph of sin 2t + 2 (see Figure 1.3). It is clear that |f – g| < ∈ if and only if the graph of g(t) threads through the shaded region about the graph of f(t). If g(t) were a function approximating f(t) and ||f – g|| < ∈, it appears g would be a good approximation to f if ∈ were small. We borrow this simple geometry to give an initial definition of a good approximation.

      DEFINITION

      Let f ∈C[a, b] with the Tchebycheff norm. A function g belonging to C[a, b] is a good approximation to f provided ||f – g|| < ∈ for a sufficiently small ∈.

      In some situations this definition may not be a sufficient measure of goodness of approximation. To see this, consider the function f(t∈ cos 4πt on the interval [0, 1], where ∈ is a very small constant. Then the function g(t) ≡ 0 on [0, 1] (the zero vector in C[0, 1]) is a good approximation to f by our definition, since

      However, note that y′(t) ≡ 0, and f′(t) = (–4π∈/2)sin4πt. Thus ||f′ – g′|| = ||f′|| = 2π∈. Similarly ||f″ – g″|| = 8π²∈, ||f″′ – g″′|| = 32π³∈, f(n) – g(n)|| = (4π)n∈/2, and so forth. Thus if you require a good approximation g to a given function f to be one that makes each of the quantities ||f – g||, ||f′ – g′||, …, ||f(n) – g(n)|| small, the definition just given simply will not do, since (4π)n∈/2 could be quite a large number even though ∈ was quite a small number. In this case, the way out of the difficulty is to retain the essential character of the definition of a good approximation, namely, ||f – g|| small, but change the linear space and the norm. For the example we have cited, you could take X = Cn[a, b], the space of n times continuously differentiable functions defined on [a, b]. For each f ∈ Cn[a, b] define the norm || ||n of f as

      Figure 1.2 Dotted line: g(t) = t²; solid line: f(t) = cosπt. ||f – g|| = 2. Graph of f and g.

      Figure 1.3 Graph of a tube of width 2∈ symmetric about sin2t + 2.

      You must, of course, check that this function really is a norm and that Cn[a, b] with addition and scalar multiplication defined as in C[a, b] really is a linear space. Then g would be a good approximation to f if ||f – g||n < ∈ for some small ∈.

      The Tchebycheff norm is not always available, and we sometimes have to use the L2 norm on C[a, b] (or on some larger space containing C[a, b]). The contrast in the geometry of these two spaces is well worth noting. For example, if f and g are any two functions from C[a, b] with the L2 norm, the distance ||f – g||2 between f and g has a geometric interpretation that is quite different from the distance between two continuous or bounded functions with the Tchebycheff norm. In particular, every student of elementary calculus knows that for continuous f and g

      is the "average distance" from f to g on the interval [a, b], The same interpretation applies when f and g are any two functions from L2[a, b] (see next section). In the sequel we are sometime forced to consider a function f to be a good approximation to a function g if ||f – g||2 < ∈ for some small ∈. To show just how much weaker a measure of good approximation this is than the Tchebycheff or uniform, norm ||f – g||, let f(t) ≡ 0 on [–1, 1] and let g²(t) be the roof function graphed in Figure 1.4. Then

      Thus

      However, no matter how small we make ∈, so long as it is positive we have

      Figure 1.4 Graph of g²(t).

      Thus f can be a very good approximate to g in the L2 norm but a perfectly dreadful approximation to g in the Tchebycheff norm. At times we simply have to live with this condition. In the theory of Fourier series, the problem has evoked some very beautiful theorems which are outside the scope of this book, including the well-known Féjer-Cesaro summation theorems.

      EXERCISES

      1.Let X = En = {(x1, x2, …, xn): x is real}. For each x ∈ X let

      Prove that each of these quantities is a norm. When i = 1, 2, 3, graph {x: ||x – a||i < 1}, where a is some fixed point of X.

      2.Let X = C[a, b], let f(t)∈C[a, b. Is || ||2 a norm for X? Prove it. Suppose (fn:n = 1, 2, …) is a sequence of functions in X such that lim fn(t) = f(t) exists for all t in [a, b]. Then we say fn converges pointwise to f. Must f(t) ∈ X? Does ||f – fn||2 → 0 as n → ∞? Explain.

      3.Let X = P.C.[a, b], the piecewise continuous functions f(t) on [a, b. Let (fn:n = 1, 2, …) be a sequence of functions from X for which

      . Does there exist a function f(t) ∈ X such that

      (a)(f(t):n = 1, 2, …) converges pointwise to f(t)?

      as n → ∞?

      Explain.

      4.Let x0 be a fixed vector in some normed linear space X. Let d > 0 be a real number and let K = {x : ||x – x0|| ≤ d}. Prove that K is closed and bounded.

      5.Let || || be a norm for a linear space X. Prove that g(x) = ||x|| is a continuous function on X.

      1.4 THE SPACE L2[a, b]

      There is a normed linear space of which mathematicians are especially enamoured (with good reason); many engineers are not formally familiar with it, however, even though they frequently solve many of their problems in precisely this setting. This is the space L2[a, b] of functions f(t), which are Lebesgue square integrable on [a, b]. To obtain this space, first define what is known as a set of measure zero (e.g., the rational numbers, any finite set of real numbers, and the integers) and a measureable function.

      A subset E of [a, b] is of measure zero if for each ∈ > 0 there exists a countable number of open intervals whose total length is less that ∈ which contains E. For example, the rational numbers E = {1/2, 1/3, …, 1/n, …} have measure zero, since

      and

      The irrational numbers in [0, 1], on the other hand, have measure 1.

      A function f(t) is measurable if it is the limit pointwise everywhere except on a set of measure zero of an increasing sequence fn(f) of step functions (such convergence is called convergence almost everywhere). Moreover, two functions f(t) and g(t) are considered to be the same function if f(t) = g(t) for all t in [a, b] except a set of measure zero. For example, L2[a, b] does not recognize any difference between the functions

      Having accomplished all this and having proved a few theorems, we define a new kind of integration known as Lebesgue integration, which differs somewhat from Riemann integration (the kind taught in first-quarter calculus). In particular, if fn(t) is an increasing sequence of step functions that converges almost everywhere to f(t), the Lebesgue integral ∫baf(t)dt of f(t) on [a, b] is defined by

      where the Riemann and Lebesgue integrals of step functions are the same. Formally then, we have

      DEFINITION L2[a, b]

      The set L2[a, b] is the collection of all measurable functions f(t) defined on [a, b] for which

      where integration is Lebesgue integration.

      One can now prove that the Riemann integral of a continuous function and of a function that is continuous except at a finite number of points (i.e., a jump function) is exactly the same as the Lebesgue integral of that function (i.e., C[a, b] is a linear subspace of L2[a, b]). What, if anything, have we gained by the introduction of the space L2[a, b]? A first answer to this question is that going to L2[a, b] enables us to integrate many more functions than would have been possible with the Riemann integral. For example, our function f(t)

      does not have a Riemann integral. However, it does have a Lebesgue integral and, in fact,

      The particular answer is not always altogether satisfying to very applied people. However, it is important and is linked to an even more compelling reason for turning to L2[a, b] rather than, say, C[a, b]. This answer depends on the norm for L2[a, b]. Before defining a norm on L2[a, b] one must ascertain that it is indeed a linear space. Defining addition and scalar multiplication of functions from L2[a, b] exactly as we did in C[a, b], it follows from the Lebesgue theory of integration that L2[a, b] is a linear space. In particular, one can prove that the sum of any two square integrable functions and a constant times a square integrable function is again square integrable. Put more concisely, if f, g ∈ L2[a, b] and α is any real number,

      The remaining axioms for a linear space then follow quite simply. The norm || ||2 for this linear space is defined as an integral. In particular

      DEFINITION L2 NORM

      Let f (t)∈L2[a, b]. The real valued function || ||2 defined by

      is known as the L2 norm of f.

      Proof that this function actually forms a norm depends on the Lebesgue theory. The student who has not formally studied this theory can get some feeling for the space by thinking of, say, functions f that are continuous (piecewise continuous); then try to prove which of the properties of a norm hold on this smaller space (you will find that all of them do). That is,

      THEOREM

      The real-valued function || ||2 defined by

      is a norm for the linear space C[a, b] as well as for the linear space L2[a, b].

      Our comments on the distance ||f – g||2 between two functions f and g from L2[a, b] are much the same as those given in Section 1.3 for functions from C[a, b] with the L2 norm. In particular, the weakness of the L2 norm as a measure of approximation in contrast with the Tchebycheff norm still obtains. However, it is only fair to point out that there are many simple functions possessing an L2 norm which have no Tchebycheff norm. An instance of this is given by the function

      We now return to our original question, "Why bother with L2[a, b]?" For example, why not simply use C[a, b], since || ||2 is a norm for this space as well. One reason is that there are sequences (fn(t)) of continuous functions for which ||fn – fm||2 → 0 as

      Enjoying the preview?
      Page 1 of 1