Theoretical foundations of Gaussian quadrature

Similar documents
The Regulated and Riemann Integrals

1 The Lagrange interpolation formula

Chapter 3 Polynomials

Orthogonal Polynomials

Abstract inner product spaces

Construction of Gauss Quadrature Rules

1. Gauss-Jacobi quadrature and Legendre polynomials. p(t)w(t)dt, p {p(x 0 ),...p(x n )} p(t)w(t)dt = w k p(x k ),

Numerical integration

Best Approximation in the 2-norm

Physics 116C Solution of inhomogeneous ordinary differential equations using Green s functions

Math 4310 Solutions to homework 1 Due 9/1/16

CMDA 4604: Intermediate Topics in Mathematical Modeling Lecture 19: Interpolation and Quadrature

Lecture 14: Quadrature

Best Approximation. Chapter The General Case

20 MATHEMATICS POLYNOMIALS

Discrete Least-squares Approximations

Advanced Calculus: MATH 410 Notes on Integrals and Integrability Professor David Levermore 17 October 2004

Numerical Analysis. Doron Levy. Department of Mathematics Stanford University

MATH34032: Green s Functions, Integral Equations and the Calculus of Variations 1

ODE: Existence and Uniqueness of a Solution

Sturm-Liouville Eigenvalue problem: Let p(x) > 0, q(x) 0, r(x) 0 in I = (a, b). Here we assume b > a. Let X C 2 1

Notes on length and conformal metrics

Chapter 14. Matrix Representations of Linear Transformations

Math Solutions to homework 1

Math 61CM - Solutions to homework 9

Bases for Vector Spaces

NUMERICAL INTEGRATION. The inverse process to differentiation in calculus is integration. Mathematically, integration is represented by.

Numerical Integration

MAA 4212 Improper Integrals

Lecture 6: Singular Integrals, Open Quadrature rules, and Gauss Quadrature

Elementary Linear Algebra

Quadratic Forms. Quadratic Forms

3.4 Numerical integration

Lecture 19: Continuous Least Squares Approximation

DEFINITION The inner product of two functions f 1 and f 2 on an interval [a, b] is the number. ( f 1, f 2 ) b DEFINITION 11.1.

MORE FUNCTION GRAPHING; OPTIMIZATION. (Last edited October 28, 2013 at 11:09pm.)

Lecture 1. Functional series. Pointwise and uniform convergence.

Matrices, Moments and Quadrature, cont d

Orthogonal Polynomials and Least-Squares Approximations to Functions

MATH 174A: PROBLEM SET 5. Suggested Solution

CAAM 453 NUMERICAL ANALYSIS I Examination There are four questions, plus a bonus. Do not look at them until you begin the exam.

Interpolation. Gaussian Quadrature. September 25, 2011

1 The Riemann Integral

Math 1B, lecture 4: Error bounds for numerical methods

p(x) = 3x 3 + x n 3 k=0 so the right hand side of the equality we have to show is obtained for r = b 0, s = b 1 and 2n 3 b k x k, q 2n 3 (x) =

Lecture 3. Limits of Functions and Continuity

UNIFORM CONVERGENCE. Contents 1. Uniform Convergence 1 2. Properties of uniform convergence 3

(4.1) D r v(t) ω(t, v(t))

p-adic Egyptian Fractions

Presentation Problems 5

W. We shall do so one by one, starting with I 1, and we shall do it greedily, trying

Advanced Computational Fluid Dynamics AA215A Lecture 3 Polynomial Interpolation: Numerical Differentiation and Integration.

Lecture 17. Integration: Gauss Quadrature. David Semeraro. University of Illinois at Urbana-Champaign. March 20, 2014

c n φ n (x), 0 < x < L, (1) n=1

Review of basic calculus

Convex Sets and Functions

Numerical Methods I Orthogonal Polynomials

Lecture 3 ( ) (translated and slightly adapted from lecture notes by Martin Klazar)

Lecture Notes: Orthogonal Polynomials, Gaussian Quadrature, and Integral Equations

Review of Riemann Integral

Continuous Random Variables

Review of Calculus, cont d

MATRICES AND VECTORS SPACE

Inner-product spaces

Phil Wertheimer UMD Math Qualifying Exam Solutions Analysis - January, 2015

LECTURE 3. Orthogonal Functions. n X. It should be noted, however, that the vectors f i need not be orthogonal nor need they have unit length for

Hilbert Spaces. Chapter Inner product spaces

8 Laplace s Method and Local Limit Theorems

MTH 5102 Linear Algebra Practice Exam 1 - Solutions Feb. 9, 2016

Math 270A: Numerical Linear Algebra

1. On some properties of definite integrals. We prove

1 1D heat and wave equations on a finite interval

1.3 The Lemma of DuBois-Reymond

Chapter 3. Vector Spaces

f(x)dx . Show that there 1, 0 < x 1 does not exist a differentiable function g : [ 1, 1] R such that g (x) = f(x) for all

Advanced Calculus: MATH 410 Uniform Convergence of Functions Professor David Levermore 11 December 2015

The Riemann-Lebesgue Lemma

63. Representation of functions as power series Consider a power series. ( 1) n x 2n for all 1 < x < 1

Review of Gaussian Quadrature method

Part IB Numerical Analysis

SOLUTIONS FOR ANALYSIS QUALIFYING EXAM, FALL (1 + µ(f n )) f(x) =. But we don t need the exact bound.) Set

g i fφdx dx = x i i=1 is a Hilbert space. We shall, henceforth, abuse notation and write g i f(x) = f

Infinite Geometric Series

HERMITE-HADAMARD-TYPE INEQUALITIES FOR GENERALIZED CONVEX FUNCTIONS

1 Probability Density Functions

Riemann Sums and Riemann Integrals

QUADRATURE is an old-fashioned word that refers to

Numerical Integration. 1 Introduction. 2 Midpoint Rule, Trapezoid Rule, Simpson Rule. AMSC/CMSC 460/466 T. von Petersdorff 1

7.2 The Definite Integral

ARITHMETIC OPERATIONS. The real numbers have the following properties: a b c ab ac

Farey Fractions. Rickard Fernström. U.U.D.M. Project Report 2017:24. Department of Mathematics Uppsala University

Introduction to Numerical Analysis

Riemann Sums and Riemann Integrals

Introduction to Some Convergence theorems

STUDY GUIDE FOR BASIC EXAM

COSC 3361 Numerical Analysis I Numerical Integration and Differentiation (III) - Gauss Quadrature and Adaptive Quadrature

Operations with Polynomials

Handout: Natural deduction for first order logic

Matrices and Determinants

A REVIEW OF CALCULUS CONCEPTS FOR JDEP 384H. Thomas Shores Department of Mathematics University of Nebraska Spring 2007

Transcription:

Theoreticl foundtions of Gussin qudrture 1 Inner product vector spce Definition 1. A vector spce (or liner spce) is set V = {u, v, w,...} in which the following two opertions re defined: (A) Addition of vectors: u + v V, which stisfies the properties (A 1 ) ssocitivity: u + (v + w) = (u + v) + w u, v, w in V ; (A 2 ) existence of zero vector: 0 V such tht u + 0 = u u V ; (A ) existence of n opposite element: u V ( u) V such tht u + ( u) = 0; (A 4 ) commuttivity: u + v = v + u u, v in V ; (B) Multipliction of number nd vector: αu V for α R, which stisfies the properties (B 1 ) α(u + v) = αu + αv α R, u, v V ; (B 2 ) (α + β)u = αu + βu α, β R, u V ; (B ) (αβ)u = α(βu) α, β R, u V ; (B 4 ) 1u = u u V. Definition 2. An inner product liner spce is liner spce V with n opertion (, ) stisfying the properties () (u, v) = (v, u) u, v in V ; (b) (u + v, w) = (u, w) + (v, w) u, v, w in V ; (c) (αu, v) = α(u, v) α R, u, v, w in V ; (d) (u, u) 0 u V ; moreover, (u, u) = 0 if nd only if u = 0. Exmple. The stndrd inner product of the vectors u = (u 1, u 2,..., u d ) R d nd v = (v 1, v 2,..., v d ) R d is given by (u, v) = d u i v i. 1

Exmple. Let G be symmetric positive-definite mtrix, for exmple 5 4 1 G = (g ij ) = 4 7 0. 1 0 Then one cn define sclr product corresponding to G by (u, v) := d d u i g ij v j. j=1 Remrk. In n inner product liner spce, one cn define the norm of vector by The fmous Cuchy-Schwrz inequlity reds u := (u, u). (u, v) u v. Think bout the mening of this inequlity in R. Exercise. Find the norm of the vector u = (, 0, 4) using the stndrd inner product in R nd then by using the inner product in R defined through the mtrix G. A very importnt exmple. Consider the set of ll polynomils of degree no greter thn 4, where the opertions ddition of vectors nd multipliction of number nd vector re defined in the stndrd wy, nmely: if P nd Q re such polynomils, P (x) = p 4 x 4 + p x + p 2 x 2 + p 1 x + p 0, Q(x) = q 4 x 4 + q x + q 2 x 2 + q 1 x + q 0, then their sum, P + Q is given by (P + Q)(x) = (p 4 + q 4 )x 4 + (p + q )x + (p 2 + q 2 )x 2 + (p 1 + q 1 )x + (p 0 + q 0 ), nd, for α R, the product αp is defined by (αp )(x) = (αp 4 )x 4 + (αp )x + (αp 2 )x 2 + (αp 1 )x + (αp 0 ). Then this set of polynomils is vector spce of dimension 5. One cn tke for bsis in this spce the set of polynomils E 0 (x) := 1, E 1 (x) := x, E 2 (x) := x 2, E (x) := x, E 4 (x) := x 4. This, however, is only one of the infinitely mny bses in this spce. For exmple, the set of vectors G 0 (x) := x 1, G 1 (x) := x + 1, G 2 (x) := x 2 + x +, 2

G (x) := x + x 2 4, G 4 (x) := x 4 x 2x is perfectly good bsis. (Note tht I clled G 0, G 1,..., G n vectors to emphsize tht wht is importnt for us is the structure of vector spce nd not so much the fct tht these vectors re polynomils.) Any vector (i.e., polynomil of degree 4) cn be represented in unique wy in ny bsis, for exmple, the polynomil P (x) = x 4 5x 2 + x + 7 cn be written s P = E 4 5 E 2 + E 1 + 7 E 0, or, lterntively, s P = G 4 G + 4 G 2 11 G 1 + 6 G 0. 2 Inner product in the spce of polynomils One cn define n inner product structure in the spce of polynomils in mny different wys. Let V n (, b) stnd for the spce of polynomils of degree n defined for x [, b]. Most of the theory we will develop works lso if = nd/or b =. Let w : [, b] R be weight function, i.e., function stisfying the following properties: () the integrl w(x) dx exists; (b) w(x) 0 for ll x [, b], nd w(x) cn be zero only t isolted points in [, b] (in prticulr, w(x) cnnot be zero in n intervl of nonzero length). We define sclr product in V n (, b) by (P, Q) := P (x) Q(x) w(x) dx ; (1) if the intervl (, b) is of infinite length, then one hs to tke w such tht this integrl exists for ll P nd Q in V n (, b). Let V n (, b; w) stnds for the inner product liner spce of polynomils of degree n defined on [, b], nd sclr product defined by (1). Exmple. The Legendre polynomils re fmily of polynomils P 0, P 1, P 2,... such tht P n is polynomil of degree n defined for x [ 1, 1], with leding coefficients equl to 1 ( leding re the coefficients of the highest powers of x) nd such tht P n nd P m re orthogonl for n m in the sense of the following inner product: (P n, P m ) = 1 1 P n (x) P m (x) dx. In other words, the polynomils P 0, P 1, P 2,..., P n constitute n orthogonl bsis of the spce V n ( 1, 1; w(x) 1). Here re the first severl Legendre polynomils: P 0 (x) = 1, P 1 (x) = x, P 2 (x) = x 2 1, P (x) = x 5 x, P 4(x) = x 4 6 7 x2 + 5,....

Sometime Legendre polynomils re normlized in different wy: P 0 (x) = 1, P1 (x) = x, P2 (x) = 1 2 (x2 1), P (x) = 1 2 (5x x), P4 (x) = 1 8 (5x4 0x 2 + ),... ; check tht P n is proportionl to P n for ll the polynomils given here. Exercise. Check tht ech of the first five Legendre polynomils is orthogonl to ll other Legendre polynomils in the exmple bove. Exmple. The Hermite polynomils re fmily of polynomils H 0, H 1, H 2,... such tht H n is polynomil of degree n defined for x R, normlized in such wy tht (H n, H n ) = 2 n n! π nd (H n, H m ) = 0 for n m, where the inner product is defined s follows: (H n, H m ) = H n (x) H m (x) e x2 dx. In other words, the polynomils H 0, H 1, H 2,..., H n constitute n orthogonl bsis of the spce V n (, ; e x2 ). Here re the first five Hermite polynomils: H 0 (x) = 1, H 1 (x) = 2x, H 2 (x) = 4x 2 2, H (x) = 8x 12x, H 4 (x) = 16x 4 48x 2 +12. Gussin qudrture Theorem 1. Let w be weight function on [, b], let n be positive integer, nd let G 0, G 1,..., G n be n orthogonl fmily of polynomils with degree of G k equl to k for ech k = 0, 1,..., n. In other words, G 0, G 1,..., G n form n orthogonl bsis of the inner product liner spce V n (, b; w). Let x 1, x 2,..., x n be the roots of G n, nd define L i (x) := n j=1, j i x x j x i x j for i = 1, 2,..., n. Then the corresponding Gussin qudrture formul is given by where I(f) := f(x) w(x) dx I n (f) := w i := L i (x) w(x) dx. w i f(x i ), The formul I n (f) hs degree of precision exctly 2n 1, which mens tht I n (x k ) = I(x k ) for k = 0, 1,..., 2n 1, but there is polynomil Q of degree 2n for which I n (Q) I(Q). 4

Proof: We will prove tht the qudrture formul I n (f) = w i f(x i ) (where the weights w i re given by the formul bove) hs degree of precision exctly 2n 1 in three steps: Step 1: The degree of precision of the qudrture formul is n 1. Let V n 1 stnd for the liner spce of ll polynomils of degree not exceeding n 1 defined for x [, b]. We need to prove tht our qudrture formul is exct for ny R V n 1. The polynomils L i (x), i = 1, 2,..., n re Lgrnge polynomils of degree n 1 such tht L i (x k ) = δ ik for ll k = 1, 2,..., n. This implies tht R(x i ) L i (x) is the Lgrnge polynomil of degree n 1 tht interpoltes the polynomil R t the n points x 1, x 2,..., x n. Reclling tht R is polynomil of degree of degree n 1, we conclude tht R(x) = R(x i ) L i (x) x [, b]. Then we hve I(R) = = R(x) w(x) dx = [ ] R(x i ) L i (x) w(x) dx [ ] R(x i ) L i (x) w(x) dx = R(x i ) w i = I n (R), therefore the qudrture formul is exct for ny polynomil R of degree up to nd including n 1, or, in other words, the degree of precision of the qudrture formul is n 1. Step 2: The degree of precision of the qudrture formul is 2n 1. Let P be polynomil of degree 2n 1, i.e., P V 2n 1. Divide P by G n (which is polynomil of degree n to obtin P (x) = Q(x) G n (x) + R(x), where Q V n 1 nd R V n 1. Since the polynomils G 0, G 1, G 2,..., G n 1 form bsis of the liner spce V n 1, we cn write Q in the form n 1 Q(x) = q i G i (x). 5

Then I(P ) = I(Q G n + R) = = [ b n 1 = 0 + [Q(x) G n (x) + R(x)] w(x) dx ] q i G i (x) G n (x) w(x) dx + R(x) w(x) dx = I(R), R(x) w(x) dx becuse ll polynomils G 0, G 1,..., G n 1 re orthogonl to G n (with respect to the sclr product defined with the weight function w): (G i, G n ) = G i (x) G n (x) w(x) dx = 0 i = 0, 1, 2,..., n 1. Now recll tht R is polynomil of degree n 1 to conclude tht I(R) = I n (R) ccording to wht we proved in Step 1. Since x i re roots of the polynomil G n, it follows tht P (x i ) = Q(x i ) G n (x i ) + R(x i ) = 0 + R(x i ) = R(x i ), which implies tht I n (R) = I n (P ). Combining these results, we obtin I(P ) = I n (P ) P V 2n 1. Step : The degree of precision of the qudrture formul is 2n 1. We lredy know tht the degree of precision is t lest 2n 1, so to prove tht it is exctly 2n 1, it is enough to find one polynomil of degree 2n for which the qudrture formul is not exct. Note tht the polynomil G 2 n(x) hs degree exctly 2n, nd tht I(G 2 n) = G 2 n(x) w(x) dx > 0 (2) becuse the integrnd, G 2 n(x) w(x), is nonnegtive nd cn be zero only t isolted points ( polynomil like G 2 n hs t most 2n roots, nd w cn vnish only t isolted points by the definition of weight function). But, since x i re roots of G n for i = 1, 2,..., n, the qudrture formul gives I n (G 2 n) = w i G 2 n(x i ) = 0. () Compring (2) nd (), we see tht the qudrture formul is not exct for G 2 n. 6

In the proof of the bove theorem, we implicitly used tht G n hs exctly n simple roots, ll of which re inside the intervl (, b). This property is estblished rigorously in the following lemm. Lemm 1. Let w be weight function on [, b], let n be positive integer, nd let G 0, G 1,..., G n be fmily of polynomils such tht the degree of G k is equl to k for ech k = 0, 1,..., n, nd (G i, G j ) = 0 if i j, where the sclr product is defined by (1). Then for ech k = 0, 1, 2,..., n, the polynomil G k hs exctly k rel roots which re simple nd lie in the intervl (, b). Proof: First we will prove the existence of rel roots for G k, nd then we will count those roots. Step 1: Existence of rel roots of G k. Since the degree of G 0 is zero, it follows tht G 0 (x) = α x [, b] for some constnt α 0. (If you don t understnd why α 0, look t pge 87 of the book.) Then, for k 1, 0 = G 0 (x) G k (x) w(x) dx = α G k (x) w(x) dx, where the first equlity holds due to orthogonlity. However, since w is positive on [, b] except possibly being equl to 0 t isolted points, nd the integrl G k(x) w(x) dx is equl to zero, we conclude tht the function G k must chnge sign in (, b). Since G k is continuous function (ll polynomils re continuous functions), the Intermedite Vlue Theorem gurntees tht G k hs rel root somewhere in (, b). Step 2: Counting the roots of G k. Suppose tht G k chnges sign t exctly j points r 1, r 2,..., r j in (, b) such tht < r 1 < r 2 < < r j < b. Without loss of generlity, ssume tht G k (x) > 0 for x (, r 1 ). Then G k lterntes sign on (r 1, r 2 ), (r 2, r ),..., (r j, b). Define the uxiliry function Z(x) := ( 1) j j (x r j ) V j. By construction, Z(x) nd G k (x) hve the sme sign for ll x [, b]. From this it follows tht Z(x) G k (x) w(x) dx > 0. (4) 7

Suppose tht j < k. Since the polynomils G 0, G 1, G 1,..., G j form bsis for V j, there exist constnts z 0, z 1, z 2,..., z j such tht Z(x) = j z i G j (x). Substituting this expression into (4), we obtin [ b j ] Z(x) G k (x) w(x) dx = z i G j (x) G k (x) w(x) dx = j z i G j (x) G k (x) w(x) dx = 0, where we hve used tht j < k, which implies tht the polynomils G j nd G k re orthogonl. The lst equlity clerly contrdicts (4)! Hence the ssumption tht j < k ws wrong, i.e., we must hve tht j k. However, since the degree of the polynomil G k is k, G k cnnot hve more thn k roots, which implies tht j = k. Exmple: As n exmple of ppliction of the bove Theorem, let us rederive the qudrture formul we derived in clss, nmely, 1 ( f(x) dx f 1 ) ( ) 1 + f. 1 This formul cn be obtined by using the Legendre polynomils P k defined on pge. Tke the polynomils P 0 (x) = 1, P 1 (x) = x, nd P 2 (x) = x 2 1, which re n orthogonl bsis in the liner spce V 2 ( 1, 1; w(x) 1). Since n = 2, we expect tht the formul tht we will obtin will hve degree of precision 2n 1 = 2(2) 1 =. As in the Theorem bove, define x 1 = 1 nd x 2 = 1 to be the zeros of the polynomil P 2, define the polynomils (of degree n 1 = 2 1 = 1) nd the weights w 1 = 1 1 L 1 (x) w(x) dx = L 1 (x) = x x 2 x 1 x 2, L 2 (x) = x x 1 x 2 x 1, 1 1 x x 2 x 1 x 2 1 dx = 1 x 1 x 2 nd, similrly, w 2 = 1 1 L 2(x) w(x) dx = = 1, to obtin I 2 (f) = 2 w i f(x i ) = f ( 1 ) ( ) 1 + f ( ) x 2 1 2 x 2x. x= 1 = 1, As n exercise, check tht the degree of ccurcy of this qudrture formul is indeed. 8