THE HANKEL MATRIX METHOD FOR GAUSSIAN QUADRATURE IN 1 AND 2 DIMENSIONS

Similar documents
Lecture 14: Quadrature

Numerical Integration

Definite integral. Mathematics FRDIS MENDELU

Z b. f(x)dx. Yet in the above two cases we know what f(x) is. Sometimes, engineers want to calculate an area by computing I, but...

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

The Regulated and Riemann Integrals

Definite integral. Mathematics FRDIS MENDELU. Simona Fišnarová (Mendel University) Definite integral MENDELU 1 / 30

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

Numerical Analysis: Trapezoidal and Simpson s Rule

Best Approximation. Chapter The General Case

NUMERICAL INTEGRATION

III. Lecture on Numerical Integration. File faclib/dattab/lecture-notes/numerical-inter03.tex /by EC, 3/14/2008 at 15:11, version 9

Chapter 5. Numerical Integration

Numerical quadrature based on interpolating functions: A MATLAB implementation

APPROXIMATE INTEGRATION

Math& 152 Section Integration by Parts

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

1 The Lagrange interpolation formula

Unit #9 : Definite Integral Properties; Fundamental Theorem of Calculus

Lecture 20: Numerical Integration III

Sections 5.2: The Definite Integral

3.4 Numerical integration

7.2 The Definite Integral

Definite Integrals. The area under a curve can be approximated by adding up the areas of rectangles = 1 1 +

Orthogonal Polynomials

6.5 Numerical Approximations of Definite Integrals

1 The Riemann Integral

Theoretical foundations of Gaussian quadrature

Chapter 6 Notes, Larson/Hostetler 3e

Numerical integration

Review of Calculus, cont d

Lecture 19: Continuous Least Squares Approximation

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

Properties of Integrals, Indefinite Integrals. Goals: Definition of the Definite Integral Integral Calculations using Antiderivatives

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 ),

The Riemann Integral

Math 131. Numerical Integration Larson Section 4.6

Lecture Note 4: Numerical differentiation and integration. Xiaoqun Zhang Shanghai Jiao Tong University

Abstract inner product spaces

Lecture 1: Introduction to integration theory and bounded variation

Euler, Ioachimescu and the trapezium rule. G.J.O. Jameson (Math. Gazette 96 (2012), )

COT4501 Spring Homework VII

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

n f(x i ) x. i=1 In section 4.2, we defined the definite integral of f from x = a to x = b as n f(x i ) x; f(x) dx = lim i=1

DOING PHYSICS WITH MATLAB MATHEMATICAL ROUTINES

Tangent Line and Tangent Plane Approximations of Definite Integral

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

The Fundamental Theorem of Calculus. The Total Change Theorem and the Area Under a Curve.

Overview of Calculus I

Chapters 4 & 5 Integrals & Applications

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

Exam 2, Mathematics 4701, Section ETY6 6:05 pm 7:40 pm, March 31, 2016, IH-1105 Instructor: Attila Máté 1

Lecture 14 Numerical integration: advanced topics

Midpoint Approximation

Numerical Analysis. 10th ed. R L Burden, J D Faires, and A M Burden

Big idea in Calculus: approximation

Review of basic calculus

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

LECTURE 19. Numerical Integration. Z b. is generally thought of as representing the area under the graph of fèxè between the points x = a and

How can we approximate the area of a region in the plane? What is an interpretation of the area under the graph of a velocity function?

1 Error Analysis of Simple Rules for Numerical Integration

The area under the graph of f and above the x-axis between a and b is denoted by. f(x) dx. π O

The Fundamental Theorem of Calculus

Construction of Gauss Quadrature Rules

Numerical Integration

Undergraduate Research

Math 61CM - Solutions to homework 9

Section 4.8. D v(t j 1 ) t. (4.8.1) j=1

Chapter 2. Numerical Integration also called quadrature. 2.2 Trapezoidal Rule. 2.1 A basic principle Extending the Trapezoidal Rule DRAWINGS

Improper Integrals, and Differential Equations

1 The fundamental theorems of calculus.

Chapter 3 Polynomials

1 Part II: Numerical Integration

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

B.Sc. in Mathematics (Ordinary)

Numerical integration. Quentin Louveaux (ULiège - Institut Montefiore) Numerical analysis / 10

different methods (left endpoint, right endpoint, midpoint, trapezoid, Simpson s).

ODE: Existence and Uniqueness of a Solution

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

Lecture 23: Interpolatory Quadrature

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

Improper Integrals. Type I Improper Integrals How do we evaluate an integral such as

Section 17.2 Line Integrals

Fundamental Theorem of Calculus

Solutions to Assignment #8

Lecture 12: Numerical Quadrature

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

7.2 Riemann Integrable Functions

Math 8 Winter 2015 Applications of Integration

31.2. Numerical Integration. Introduction. Prerequisites. Learning Outcomes

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) =

Section 7.1 Integration by Substitution

AN INTEGRAL INEQUALITY FOR CONVEX FUNCTIONS AND APPLICATIONS IN NUMERICAL INTEGRATION

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

4.4 Areas, Integrals and Antiderivatives

Week 10: Riemann integral and its properties

Properties of the Riemann Integral

f(x) dx, If one of these two conditions is not met, we call the integral improper. Our usual definition for the value for the definite integral

a < a+ x < a+2 x < < a+n x = b, n A i n f(x i ) x. i=1 i=1

Chapter 7 Notes, Stewart 8e. 7.1 Integration by Parts Trigonometric Integrals Evaluating sin m x cos n (x) dx...

Transcription:

THE HANKEL MATRIX METHOD FOR GAUSSIAN QUADRATURE IN 1 AND 2 DIMENSIONS CARLOS SUERO, MAURICIO ALMANZAR CONTENTS 1 Introduction 1 2 Proof of Gussin Qudrture 6 3 Iterted 2-Dimensionl Gussin Qudrture 20 4 References 22 5 Experiments using Gussin Qudrture Rule on [, b] ppendix-1 6 Experiments using Gussin Qudrture Rule with non-stndrd Mesures ppendix-2 7 Experiments compring Gussin Qudrture, the Trpezoid Rule, And Simpson s Rule ppendix-3 8 Experiments using Gussin Qudrture Rule On Disjoint Intervls ppendix-4 9 Experiments using Iterted Gussin Qudrture Over A Rectngle [, b] [c, d] ppendix-5 1 Introduction The Fundmentl Theorem of Clculus llows one to compute the definite integrl of function over n intervl [, b] by using nti-derivtives Once the ntiderivtive of the function is found, then it is evluted t the end points of the intervl For exmple, suppose we hve continuous function f(x) on[, b] If F (x) isnntiderivtive of f(x), ie F (x) =f(x) for ll x in [, b], then f(x)dx = F (b) F () 0 This project ws crried out during the Spring, 2008 nd Summer I, 2008 semesters with Professor Lwrence A Filkow s the mentor During Spring, 2008 the project ws sponsored by SUNY/NSF Allince for Minority Prticiption t SUNY New Pltz The New Pltz AMP Director is Professor Stcie Nunes from the Deprtment of Physics During Summer I, 2008 the project ws sponsored under Ntionl Science Foundtion Grnt (number) The uthors thnk the sponsors for their support 1

For mny functions f(x) we re not ble to compute n nti-derivtive in closed form, so we cn t use the Fundmentl Theorem of Clculus For exmple f(x) =e x2 does not hve n nti-derivtive in closed form Insted, we will use numericl methods to estimte f(x)dx Gussin Qudrture is one such method The purpose of our reserch is to find wy of estimting complicted definite integrls using the Gussin Qudrture rule There re other numericl integrtion methods s well, like Simpson s rule nd the Trpezoid rule In some of the experiments, we will compre the efficiency of Gussin Qudrture to the efficiencies of Simpson s Rule nd the Trpezoid Rule The ide behind Gussin Qudrture is tht given n integer M =2n +1 > 0, we cn find points x 1 x n in [, b], nd positive weights w 1 w n,sotht (11) p(x)dx = p w i p(x i ) i=1 for every polynomil p(x) with deg p M Then, for ny continuous function f(x) on the intervl [, b], we cn pproximte the integrl f(x)dx by the Gussin Qudrture rule Q(f) := n i=1 w if(x i ), ie, (12) f(x)dx Q(f) = w i f(x i ) i=1 To explin why this pproch is effective, we hve to consider the Weierstrss Approximtion Theorem (WAT) This theorem sys tht given continuous function 2

f(x) on the intervl [, b] ndsmllɛ>0, we cn find polynomil p(x) sotht f(x) p(x) <ɛfor ll x in [, b] Butwelsohvetotkeintoccountthtto get the best results from (11) nd (12), the degree of the polynomil p(x) inwat must be less thn the chosen M for the Gussin Qudrture Rule If deg p > M, the pproximtion in (12) my not be stisfctory Our definition of Gussin Qudrture sys tht Q(f) = p i=1 w if(x i ) Wht the Weierstrss Approximtion Theorem suggests is tht Q(f) f(x)dx To see this, consider the error, (13) E := Q(f) f(x)dx For every polynomil p, we hve E = Q(f) Q(p)+Q(p) Q(f) Q(p) + Q(p) Now suppose tht deg p M Tht mens tht f(x)dx f(x)dx Q(p) = p(x)dx So, (14) Q(p) f(x)dx 3

= = p(x)dx f(x)dx (p(x) f(x))dx (15) p(x) f(x) dx If p(x) f(x) <ɛthroughout [, b], then the expression in (15) is t most = ɛdx, which is equl to ɛ(b ) So we hve (16) Q(p) f(x)dx ɛ(b ) Now let s consider Q(f) Q(p) Thisisequlto w i f(x i ) w i p(x i ) = w i (f(x i ) p(x i )) w i (f(x i ) p(x i ) w i ɛ = ɛ w i So we hve: = ɛ(b ) (since w i = Q(1) = 1dx = b ) (17) Q(f) Q(p) ɛ(b ) 4

Now we cn show tht the error (13) cn be mde smll, s follows Suppose we re given n intervl [, b], continuous function f(x) on[, b], ɛ>0,ndletp(x) be given by the Weierstrss Approximtion Theorem, ie p(x) f(x) <ɛ( x b) If we choose M deg p nd pply the Gussin Qudrture Rule using this M, then from (17) nd (14) we hve E = Q(f) f(x)dx is Q(f) Q(p) + Q(p) f(x)dx ɛ(b )+ɛ(b ) =2ɛ(b ), nd since ɛ is smll, E is lso smll Unfortuntely, for smll ɛ, deg p is often very lrge, so it is not prcticl to use Gussin Qudrture with M deg p Wht the experiments will show is tht strting with M = 1,sM increses we cn chieve 6 plce ccurcy in estimting f(x)dx by using firly smll vlues of M, though the smllest M tht gives 6 plce ccurcy depends on severl fctors, such s the intervl length b nd the complexity of f(x) 5

2 Proof of Gussin Qudrture We will derive n implementtion of Gussin Qudrture bsed on mtrix positivity nd Lgrnge interpoltion, s described in [HK, pge 115] Recll the following fcts bout integrls (21) (f(x)+g(x))dx = f(x)dx + g(x)dx (22) αf(x)dx = α f(x)dx The sme rules pply to Gussin Qudrture: (23) Q(f(x)+g(x)) = Q(f(x)) + Q(g(x)) (24) Q(αf(x)) = αq(f(x)) To prove (23), we hve Q(f(x)+g(x)) = = w i (f(x i )+g(x i )) (w i f(x i )+w i g(x i )) 6

= w i (f(x i )) + w i (g(x i )) = Q(f(x)) + Q(g(x)) The proof of (24) is similr Equtions (21) - (24) show tht it is enough to prove Gussin Qudrture for p(x) =x i (0 i M), ie, we must find points nd weights s in (11) such tht (25) β j := x j dx = Q(x j ):= w i x j i, (0 j M) Exmple 21 We will illustrte this reduction to monomils with p(x) = 0 + 1 x Applying (21) nd (22) we hve, ( o + 1 x)dx = 0 1dx+1 xdx If (25) holds, then the lst expression equls 0 Q(1) + 1 Q(x), nd by (23)-(24), this equls Q( 0 1+ 1 x), ie, ( 0 + 1 x)dx = Q( 0 + 1 x) Exmple 22 Throughout this section we shll illustrte the method of Gussin Qudrture in detil for the cse when m =3,n =1,ndtheintervlis[0, 1] We need to find points x 0,x 1 in [0, 1] nd positive weights w 0,w 1 such tht: 7

(26) β 0 := 1 = w 0 x 0 + w 1 x 1 (27) β 1 := 1/2 =w 0 x 0 + w 1 x 1 (28) β 2 := 1/3 =w 0 x 2 0 + w 1 x 2 1 (29) β 3 := 1/4 =w 0 x 3 0 + w 1 x 3 1 defined s In the sequel we will show how to solve (26)-(29) To estblish Gussin Qudrture we need to look t the Hnkel mtrix H H = β 0 β 1 β n β 1 β 2 β n+1 β 2 β 3 β n+2 β n β n+1 β 2n We will denote the columns of H by 1,t,, t n We lso will consider the vector v :=t n+1 =(β n+1,,β 2n+1 ) t 8

It is known tht H is positive nd invertible Tht mens H, > 0 whenever 0 Another wy to know tht the mtrix H is positive nd invertible is to show tht the nested determinnts of the corners re ll positive If H i is the i i upper left hnd corner of H, we must show tht det H i > 0(1 i n +1) Exmple 23 In our exmple, we hve H = 1 1/2 1/2 1/3 Then det(h 1 )= 1 nd det(h 2 )= 1/3 1/4 =1/12 > 0SoH>0 Since H is invertible, there is n unique vector =( 0,, n ) such tht (210) H = v, ie, = H 1 v, 9

or equivlently, (211) 0 β 0 + + n β n = β n+1 0 β 1 + + n β n+1 = β n+2 0 β n + + n β 2n = β 2n+1 So t n+1 = 0 1+ 1 t+ + n t n Nowletp(t) =t n+1 ( 0 1+ 1 t + + n t n ) Exmple 24 In our exmple, we hve 1 1/2 1/2 1/3 0 1 = 1/3 1/4 so 0 = 1/3 H 1 1 1/4 1/3 1/2 1/3 =12 1/2 1 1/4 nd we find 0 = 1/6, 1 =1 10

So we hve p(t) =t 2 ( 1/6+t) =t 2 t +1/6 It is known tht p(t) hs exctly n + 1 distinct rel roots in [, b] (see the proof of Proposition 33 nd Theorem 41 (iv) => (iii) in [CF] Denote these roots by x 0 x n Exmple 25 In our exmple, the roots of t 2 t +1/6 =0re t =1± 1 4(1)(1/6)/2 =1± (5/3)/2, so x 0 07886751346 nd x 1 002113248654 Note tht both x 0 nd x 1 re in [0, 1] Now let s look t the Vndermonde mtrix V defined s V = x 0 0 x 0 n x 1 0 x 1 n x n 0 x n n It is known tht becuse x 0,,x n re distinct, then V is invertible, ie, det(v ) 0[HK, pge 115] 11

Exmple 26 In our exmple, V = 1 1 x 0 x 1 So det(v )=x 1 x 0 0 Since V is invertible, there is n unique vector, ω := (ω 0 ω n ), such tht (212) Vω= β 0 β 1 β n, ie, (213) 1 1 x 0 x n x 0 2 x 2 n x n 0 x n n ω 0 ω n = β 0 β 1 β n 12

(213) is equivlent to the following system: (214) β 0 = ω 0 + + ω n (= Q(1)) β 1 = ω 0 x 0 + + ω n x n (= Q(x)) β n = ω 0 x n 0 + + ω n x n n (= Q(x n )) We clim tht x 0,,x n nd ω 0,,ω n solve the Gussin Qudrture system (25) Exmple 27 In the exmple we hve, 1 1 x 0 x 1 ω 0 ω 1 = β 0 β 1, or equivlently w 0 + w 1 = β 0, nd w 0 x 0 + w 1 x 1 = β 1, where w 0 0623893 nd w 1 0376107 The Gussin Qudrture system (25) is equivlent to the system of equtions in (214) together with the following system: 13

(215) β n+1 = ω 0 x n+1 0 + + ω n x n+1 n (= Q(x n+1 )) β n+2 = ω 0 x n+2 0 + + ω n x n+2 n (= Q(x n+2 )) = β 2n+1 = ω 0 x 2n+1 0 + + ω n x 2n+1 n (= Q(x 2n+1 )) Since (214) is stisfied from (213), we re going to focus on (215) Consider the first eqution of (215), β n+1 = ω 0 x n+1 0 + + ω n x n+1 n We hve β n+1 = x n+1 dx nd Q(x n+1 ) w i x n+1 i Since ech x i is root of t n+1 = 0 + 1 t + + n t n,wehve Q(x n+1 )= w i x n+1 i 14

= w i ( 0 1+ 1 x i + + n x n i ) = 0 wi + 1 wi x i + + n wi x n i = 0 β 0 + 1 β 1 + + n β n (by (214)) By the first eqution of (211), the lst expression is equl to β n+1, so we conclude tht (216) Q(x n+1 )=β n+1 For the second eqution of (215), we hve Q(x n+2 ) w i x n+2 i Since ech x i is root of t n+1 = 0 + 1 t + + n t n, x i is lso root of t n+2 = 0 t + + n t n+1 Sowehve Q(x n+2 )= w i x n+2 i = w i ( 0 x i + + n x n+1 i ) 15

= 0 wi x i + + n wi x n+1 i = 0 β 1 + + n β n+1 (by (214) nd (216)) By the second eqution of (211) the lst expression is equl to β n+2,sowe conclude tht (217) Q(x n+2 )=β n+2 Following the sme procedure, we re ble to prove the rest of the equtions in the set (215) This completes the proof tht the points x i,,x n nd the weights w i,,w n stisfy the Gussin Qudrture system (11) From the bove discussion, we hve distinct points x 0,,x n in [, b] nd weights w 0,,w n such tht (218) Q(p) := w i p(x i )= p(x)dx (deg p 2n +1) To complete the proof of Gussin Qudrture s in (11) we must still show tht ech w i > 0 In wht follows we will use Lgrnge polynomils to prove tht ech w i > 0 Lgrnge Interpoltion sys tht given distinct points x 0,,x n nd given numbers y 0,,y n there is unique polynomil p(x) withdeg p = n, such tht p(x i )=y i (0 i n) 16

Exmple 28 Consider Lgrnge Interpoltion with n=1 The Lgrnge polynomil of degree 1 such tht p(x 0 )=y 0 nd p(x 1 )=y 1 is (219) p 1 (x) := y 0(x x 1 ) (x 0 x 1 ) + y 1(x x 0 ) (x 1 x 0 ) The curve y = p(x) is the line connecting (x 0,y 0 )nd(x 1,y 1 ) The Lgrnge polynomil p 1 (x) in (219) is the bsis for the Trpezoidl Rule In this rule, we subdivide [, b] inton equl subintervls [x i,x i+1 ](0 i n + 1,x 0 =, x n = b) On intervl [x i,x i+1 ], we pproximte f(x) by the line connecting (x i,f(x i )) to (x i+1,f(x i+1 )); this line is the grph of the Lgrnge polynomil p 1 (x) such tht p 1 (x i )=y i := f(x i ) nd p 1 (x i+1 )=y i+1 := f(x i+1 ) We then pproximte x i+1 x i grph of p 1 between x i nd x i+1 : f(x)dx by the re of the trpezoid determined by the xi+1 x i f(x)dx 1 2 (x i+1 x i )(f(x i )+f(x i+1 )) Letting h = x i+1 x i (0 i n 1), we my express the Trpezoidl Rule by 17

(220) n 1 f(x)dx = h f( + ih)+h( f()+f(b) ) 2 i=1 Exmple 29 For n =2,wehvedistinctpointsx 0,x 1,x 2,ndvluesy 0,y 1,y 2 The Lgrnge polynomil p(x) looks like the following: (221) p 2 (x) := y 0(x x 1 )(x x 2 ) (x 0 x 1 )(x 0 x 2 ) + y 1(x x 0 )(x x 2 ) (x 1 x 0 )(x 1 x 2 ) + y 2(x x 0 )(x x 1 ) (x 2 x 0 )(x 2 x 1 ) The curve y = p 2 (x) is the unique prbol pssing through (x 0,y 0 ), (x 1,y 1 ), (x 2,y 2 ) The Lgrnge polynomil p 2 (x) in (221) is the bsis for Simpson s Rule In this rule, we pproximte y = f(x) by piecewise-prbolic curve y = g(x) on[, b], using p 2 (x) for ech piece of g(x) We cn then pproximte f(x)dx by g(x)dx Ifwe let the function f be tbulted t points x 0,x 1 nd x 2, eqully spced by distnce h, nd we let y i = f i := f(x i )(0 i 2), then Simpson s rule sys tht x2 x0 +2h x 0 f(x)dx = x0 +2h x 0 f(x)dx x 0 p 2 (x)dx = 1 3 h(f 0 +4f 1 + f 2 ) 18

If we use n double-intervls with eqully spced points x 0,x 1,,x 2n, then by using the bove method on ech double-intervl nd dding up, we get (222) f(x)dx h n 1 3 ( (2f( +2ih)+4f( +(2i +1)h)) + f(b) f()) The generl formul for the Lgrnge polynomil p(x) such tht p(x i )=y i (p i n) is cler from (219) nd (221) If q(x) is nother polynomil of deg n such tht q(x i )=y i (0 i n), then (p q)(x i )=0 Since deg p q n nd p q hs n +1 distinct roots, it follows tht p q = 0, ie p = q So there is unique Lgrnge polynomil of deg n Now we return to Gussin Qudrture, nd the clim tht ω i > 0(0 i n) We fix j nd we let p(x) p j (x) be the Lgrnge polynomil of degree n such tht p(x i )=0fori j nd p(x j )=1 Now consider q(x) =p(x) 2 Since deg q =2n (< 2n + 1), then from (218), (223) q(x)dx = w i q(x i ) Note tht q(x) 0nd q(x j )=1, so q(x)dx > 0 19

Therefore, 0 < q(x)dx = w i q(x i )=w j, so we conclude tht w j > 0(0 j n) This completes the proof of the Gussin Qudrture rule (11) 3 Iterted 2-Dimensionl Gussin Qudrture In 1-dimensionl Gussin Qudrture we showed tht given n>0nd n intervl [, b], there exist points x 0,,x n in [,b] nd positive weights ω 0,,ω n, such tht p(x)dx = n w ip(x i )(deg p 2n +1) Now we re going to use 1-dimensionl Gussin Qudrture to find rule tht works in two dimensions We re given rectngle in the plne, R =[, b] [c, d], nd continuous function f(x, y) onr We wnt to pproximte f(x, y)dxdy R Given n, letx 0,,x n, ω 0,,ω n be the Gussin Qudrture points nd weights for [, b], so (11) holds Let y 0,,y n, s 0,,s n be the Gussin Qudrture points nd weights for [c, d], so tht d c q(y)dy = s j q(y j )(deg q 2n +1) j=0 Now for function f(x, y), defined on rectngle R, let Q(f) := s j w i f(x i,y j ) j=0 20

For polynomil p(x, y), let p y (x) =p(x, y)( polynomil in x with y fixed), nd let p x (y) =p(x, y)( polynomil in y with x fixed) We re clming tht if p(x, y) is polynomil, with deg p y (x) 2n +1nd deg p x (y) 2n +1, then Q(p) = d p(x, y)dxdy( c R p(x, y)dxdy) If we fix x nd consider g(y) :=p x (y) =p(x, y), then deg g 2n +1, so which is the sme s d c g(y)dy = s j g(y j ), j=0 H(x) := d c p(x, y)dy = s j p(x, y j )( x b) j=0 So we hve d c p(x, y)dydx = = ( d j=0 p x (y)dydx = c s j p(x, y j ))dx H(x)dx = s j p(x, y j )dx (deg p(x, y j ) 2n +1) j=0 21

= j=0 s j w i p(x i,y j )= s j w i p(x i,y j ) j=0 = Q(p(x, y)) So Q(p) = R p(x, y)dxdy whenever deg p x(y) 2n +1 nd deg p y (x) 2n +1 For generl function f(x, y), tht is defined nd continuous on R, we my pproximte f(x, y)dxdy by Q(f) R In ppendix-5 we will illustrte this pproximtion with numericl exmples References [CF] RE Curto, LA Filkow, Recursiveness, positivity, nd truncted moment problems, Houston Journl of Mthemtics 17(1991), 603-635 [HK] Hoffmn, RKunze, Liner Algebr, Prentice-Hill, 1961 Deprtment of Computer Science, Stte University of New York, New Pltz, NY 12561, USA 22