ERROR ANALYSIS OF TRAPEZOID RULE, SIMPSON’S RULE, AND ROMBERG METHOD M.Sc. Graduate Seminar By Kumama Regassa August 2012 Haramaya University ERROR ANALYSIS OF TRAPEZOID RULE, SIMPSON’S RULE, AND ROMBERG METHOD A Graduate seminar Submitted to the College of Natural and Computational Sciences, Department of Mathematics, School of Graduate Studies HARAMAYA UNIVERSITY In partial Fulfillment of the Requirement for the Degree of MASTER OF SCIENCE IN MATHEMATICS (NUMERICAL ANALYSIS) By Kumama Regassa Advisor Getinet Alemayehu(PhD) August 2012 Haramaya ii SCHOOL OF GRADUATE STUDIES HARAMAYA UNIVERSITY As member of the Examination Board of the Final M.Sc. Open Defense, we certify that we have read and evaluated this Graduate Seminar prepared by Kumama Regassa Entitled:’’ERROR ANALYSIS OF TRAPEZOID RULE, SIMPSON’S RULE AND ROMBERG METHOD’’ and recommended that it be accepted as fulfilling the Graduate Seminar requirement for the Degree of Master of Science in Mathematics (Numerical Analysis) _________________ _____________ Name of chairman Signature ______________________ _____________ External examiner Signature ___________________ _____________ Internal examiner Signature _____________ Date ____________ Date ____________ Date Final approval and acceptance of the Graduate Seminar is contingent upon the submission of the final copy of the Graduate Seminar to the Council of Graduate Studies (CGS) through the Department Graduate Committee (DGC) of the candidate’s department. I hereby certify that I have read this Graduate Seminar prepared under my direction and recommend that it be accepted as fulfilling the Graduate Seminar requirement. Getinet Alemayehu (PhD) _____________ ___________ Name of Advisor Signature Date iii PREFACE The purpose of this seminar is to discuss errors on trapezoid rule, Simpson’s rule, and Romberg method during approximating definite integral using numerical approximations. The seminar has two parts. The first part has the introduction which introduces the concept of numerical integration and some necessary topics that are essential in the main body The second part is the main body of the seminar, which deals with trapezoid rule, Simpson’s rule, and Romberg method. In this part error and bounds of the error during approximations are presented. iv AKNOWLEDGEMENT First of all I would like to thank the almighty GOD for changing my dream to true as the first chapter of my life. I extend my heartfelt gratitude to my advisor Dr.Getinet Alemeyehu for his positive, valuable professional guidance, constructive comments, suggestions, and encouragements from the beginning up to the end of the study. Zemach Gudeto and kadir Abdella for their cooperativeness in borrowing me lap top. Last but not the least; I would like to thank my wife Fenet Deferew, my brother Getu Regassa my brother Bizuneh Regassa, my grandmother Damile Dadi, my friend Abiyot Nano, Milisha Cheneke, and my staff for their encouragement and support, both morally and financial. v TABLE OF CONTENTS PREFACE iv AKNOWLEDGEMENT v TABLE OF CONTENTS vi 1. INTRODUCTION AND PRELIMINARY CONCEPTS 2 1.1. Introduction 2 1.2. Objective 3 1.3. Methodology 3 1.4. Preliminary concepts 3 2. TRAPEZOID RULE, SIMPSON RULE, AND ROMBERG METHOD Error! Bookmark not defined. 2.1. Trapezoidal rule 8 2.2. Simpson`s rule 13 2.2.1. Simpson`s one-third rule 13 2.3. Simpson’s 3/8 rule 21 2.4. Romberg method 27 2.4.1. Romberg method for the trapezium rule 28 2.4.2. Romberg method for the Simpson’s 1/3 rule 30 3. SUMMARY 37 4. REFERENCE 39 vi 1. INTRODUCTION AND PRELIMINARY CONCEPTS 1.1. Introduction The process of finding the function π(π₯) if its derivative is given is called antiderivative. π The fundamental theorem of calculus gives us an exact formula for computing ∫π π(π₯)ππ₯, provided we can find an antiderivative for f. This method of evaluating definite integral is called the analytic method. However, there are times when this is difficult or impossible. In these cases it is usually good enough to find an approximate or numerical solution. Trapezoid rule, Simpson’s rule, and Romberg method are used for approximating a definite integral provided that the integrand is integrable. The integral is approximated by linear combinations of the values of π(π₯) at the tabulated points as π πΌ = ∫π π(π₯)ππ₯ ≈ ∑ππ=0 ππ π(π₯π ) = π0 π(π₯0 ) + π1 π(π₯1 ) + π2 π(π₯2 ) + β― + ππ π(π₯π ) Where the tabulated points π₯π ′π are called abscissas, π(π₯π )′ π are called the ordinates ππ ′π are called the weights of integration. We define the error of the approximation for the given method as π π πΈπ (π) = ∫ π(π₯)ππ₯ − ∑ ππ π(π₯π ) π π=0 Since we cannot always calculate exactly what the error is, we look instead for the bound on the error. There are several reasons for carrying out numerical integration. The integrand π(π₯) may be known only at certain points such as obtained by sampling or a formula for the integrand may be known, but it may be difficult or impossible to find an anti derivative which is elementary functions or it may be possible to find an antiderivative symbolically but it may be the case if the antiderivatives is given as an infinite series or product, or if its evaluation requires a special function which is not available. Numerical integration methods can generally be described as combing evaluations of the integrand to get an approximation to the integral. 2 1.2. Objective This seminar intends to explore the following specific objectives: ο To present trapezoid rule, Simpson’s rule and Romberg method ο To present general formula for the error term ο To present proof of error in trapezoid rule and Simpson’s rule ο To present approximation of definite integral using trapezoid rule, Simpson’s rule and Romberg method 1.3. Methodology The seminar would be involved collecting facts in some books available in libraries and searched in the internet regularly. The necessary information obtained would be recorded. 1. A detail study of trapezoid rule, Simpson’s, and Romberg method considered. 2. Notes of concepts and facts made to find out possible relationships of some numerical methods and the corresponding errors. 3. The investigator collected books, and prepared reports then made a frequent contact with the advisor from the beginning to the end of the seminar work. 4. The necessary concepts added and the unnecessary concepts reduced by the recommendation of advisor. 1.4. Preliminary concepts Fundamental theorem of calculus Let π(π₯) be continuous on [π, π]. If πΉ(π₯) is any antiderivatives of π(π₯), then π ∫ π(π₯)ππ₯ = πΉ(π) − πΉ(π) π Mean value theorem for the integral Iπ π ∈ πΆ[π, π], π is integrable on [π, π] and π(π₯) does not change sign on [π, π], then there exists a number c in (π, π) with π π ∫ π(π₯)π(π₯)ππ₯ = π(π) ∫ π(π₯)ππ₯ π π 3 Intermediate value theorem If π ∈ πΆ[π, π] and π is any real number between π(π) andπ(π), then there exists a number c in (π, π) for which π(π) = π Forward difference operator β The nth forward difference of π(π₯π ) is defined as π! βπ π(π₯π ) = ∑ππ=0(−1)π π!(π−π)! ππ+π−π Newton’s forward interpolation formula Let the function π¦ = π(π₯) take the valuesπ¦0 , π¦1 , π¦2 , π¦3 , …, π¦π corresponding to the values π₯0 , π₯1 , π₯2 , …, π₯π of x. let these values of x be equally spaced such that π₯π = π₯0 + πβ (π = 0,1, … ) Assuming π¦(π₯) to be polynomial of nth degree in x and p is real number, then the formula π¦(π₯) = π¦(π₯0 + πβ) = π¦π = π¦0 + πβπ¦0 + π(π−1)…(π−π+1) π! π(π−1) 2! β2 π¦0 + π(π−1)(π−2) 3! β3 π¦0 + β― + βπ π¦0 is called Newton’s forward interpolation formula. Lagrange polynomial error formula Let π: [π, π] → β be (n+1) times continuously differentiable, then the remainder πΈπ (π) = π − ππ (π) For polynomial interpolation with n+1 distinct points π₯0 , π₯1, π₯2 … π₯π ∈ [π, π] can be represented in the form (πΈπ π)(π₯) = π π+1 (π) (π+1)! ∏ππ=0( π₯ − π₯π ), π₯ ∈ [π, π] for some π ∈ [π, π] depending on π₯ Order of a method An integration method is said to be of order p if it produces exact results, that is πΈπ (π) = 0, for all polynomials of degree less than or equal to p but not for p+1. That is, it produces exact results for π(π₯) = 1, π₯, π₯ 2 , …, π₯ π π This implies that πΈ(π₯ π ) = ∫π π₯ π ππ₯ − ∑ππ=0 ∧π π₯ππ = 0 for m = 0, 1, 2… p The error term is obtained for π(π₯) = π₯ π+1 π π πΈπ (π) = ∫π π(π₯)ππ₯ − ∑ππ=0 ∧π π(π₯π ) = (π+1)! π π+1 (π) a<π<b π We define c = ∫π π₯ π+1 ππ₯ − ∑ππ=0 ∧π π₯ππ+1 Where c is called error constant, and ∧π be weight of integration. 4 Errors In any numerical computation we come across the following types of error: 1. Inherent errors: Errors which are already present in the statement of the problem before its solution. 2. Rounding errors: errors that arise from the process of rounding off the numbers during computation. 3. Truncation errors are caused by using approximate results or on replacing an infinite process by finite one. 4. Absolute, Relative and Percentage errors. If π₯ is the true value of a quantity and π₯′ is its approximate value, then | π₯ − π₯ ′ | is called the absolute error ( πΈπ ).The relative error(πΈπ ) is defined by π₯−π₯ ′ πΈπ = | π₯ | And the percentage error(πΈπ ) is defined by 5 π₯−π₯ ′ πΈπ = 100πΈπ = 100 | π₯ | 2. TRAPEZOID RULE, SIMPSON RULE, AND ROMBERG METHOD 2.1. Newton – cotes formula Let π(π₯) be continuous on the interval [π, π] and its antiderivative πΉ(π₯) is known then the definite integral of π(π₯) from a to b may be evaluated using Newton – Leibnitz formula. π ∫π π(π₯)ππ₯ = πΉ(π) − πΉ(π) (1) Where πΉ′(π₯) = π(π₯) However, computation of the definite integral by (1) becomes difficult or practically impossible when, the anti derivative πΉ(π₯) cannot be found by elementary means or is too involved or, the integrand π(π₯) is specified in tabular form. Thus, the basic integration rule is to replace π(π₯) by a simple polynomial π(π₯) , say Newton interpolation polynomial in [π, π] π π ∫π π(π₯)ππ₯ ≈ ∫π π(π₯)ππ₯ Therefore (2) We define the error of approximation for a given method as π π πΈπ (π) = ∫π π(π₯)ππ₯ − ∫π π(π₯)ππ₯ (3) Where π(π₯) is polynomial representing the function y = π(π₯) in the interval[π, π]. We have also π ∫π π(π₯)ππ₯ = ∑ππ=0 ππ π(π₯π ) = π0 π(π₯0 ) + π1 π(π₯1 ) + π2 π(π₯2 ) + …+ ππ π(π₯π ) Where the tabulated points π₯π ′s are called abscissas,π(π₯π )’s are called the ordinates and ππ ’s are called the weights of the integration. Hence we can define error of approximation for the given method as π πΈπ (π) = ∫π π(π₯)ππ₯ − ∑ππ=0 ππ π(π₯π ) 6 (4) Since we cannot always calculate exactly what the error is, we look instead for the bound on the error, The bound for the error term is given by |π| |πΈπ (π)| ≤ (π+1)! max | π π+1 (π₯)| (7) π≤π₯≤π Before discussing Trapezoid rule and Simpson’s rule let derive a general formula from which these rules are obtained which is called Newton–Cote’s quadrature formula π I = ∫π π(π₯)ππ₯ Let Where π(π₯) takes the valueπ¦0 , π¦1 , π¦2 ,…π¦π for π₯ = π₯0 , π₯1 , π₯2 …π₯π Observe figure below π¦ = π(π₯) π¦0 π¦1 π¦2 π¦π−1 π₯0 π₯1 π₯2 π¦π π₯π−1 π₯π Figure 1 As shown in figure 1 above Let us divide the interval [π, π] into n sub – intervals of width β so that π = π₯0 , π₯1 = π₯0 + β, π₯2 = π₯0 + 2β, … , π₯π = π₯0 + πβ = π. Then π₯ +πβ πΌ = ∫π₯ 0 0 π(π₯)ππ₯ π = β ∫0 π(π₯0 + πβ)ππ (8) (By puttingπ₯ = π₯0 + πβ, ππ₯ = βππ were π is any real number) Now by Newton forward interpolation formula we have: π(π₯0 + πβ)= π¦0 + πβπ¦0 + π(π−1)(π−2)(π−3) 4! β4 π¦0 +…+ π(π−1) 2! β2 π¦0 + π(π−1)(π−2) 3! π(π−1)(π−2)(π−3)…(π−π−1) π! β3 π¦0 + βπ π¦0 (9) Now substituting (9) in (8) we have π π πΌ = β ∫0 π(π₯0 + πβ)ππ = β ∫0 [π¦0 + πβπ¦0 + π(π−1)(π−2)(π−3) 4! π = β ∫0 [π¦0 + πβπ¦0 + β4 π¦0 + β― + π2 −π 2! β2 π¦0 + π(π−1) 2! β2 π¦0 + π(π−1)(π−2)(π−3)…(π−π−1) π! π3 −3π2 +2π 3! 7 β3 π¦0 + π(π−1)(π−2) 3! β3 π¦0 + βπ π¦0 + β― ] ππ π4 −6π3 +11π2 −6π 4! β4 π¦0 + β― ] ππ π π = β [∫0 π¦0 ππ + βπ¦0 ∫0 πππ + β2 π¦0 2! π ∫0 (π2 − π )ππ + β3 π¦0 3! π ∫0 (π3 −3π2 + 2π)ππ + β4 π¦0 4! π ∫0 (π4 − 6π3 +11π2 − 6π)ππ + β―] = β [ππ¦0 + π2 2 βπ¦0 + π = πβ [π¦0 + 2 βπ¦0 + π2 (2π−3) 12 π(2π−3) 12 π2 (π−2)2 β2 π¦0 + β2 π¦0 + π5 24 π(π−2)2 24 3π4 β3 π¦0 + ( 5 − π4 β3 π¦0 + ( 5 − 3π3 2 + 2 + 11π2 3 11π3 3 − 3π2 ) − 3π) β4 π¦0 4! β4 π¦0 4! +β―] +β―] (10) The last formula refers to us Newton–Cotes quadrature formula. From this formula we can deduce a general formula for 1) Trapezoidal rule 2) Simpson one-third rule 3) Simpson three-eight rule 2.1.1. Trapezoidal rule Putting π = 1 in (10) and taking the curve through (π₯π , π¦π ) and (π₯π+1 , π¦π+1 ), πππ π = 0,1,2,3 … as a straight line as shown in figure 2 below that is a polynomial of first order so that differences of order higher than first become zero, we get π y π = π₯0 π₯1 π₯2 π₯π = π x Figure 2 π₯ 1 β β β 1 ∫π₯ π(π₯)ππ₯ = β (π¦0 + 2 βπ¦0 ) = 2 (2π¦0 + π¦1 − π¦0 ) = 2 (π¦0 + π¦1 ) = 2 (π(π) + π(π)) 0 Thus the value of the integral on [π, π] is given by π ∫π π(π₯)ππ₯ = π−π 2 (π(π) + π(π)) (11) This is called trapezoidal rule for approximating the value of integral on the single sub-interval. Thus for π sub-intervals with nodes π = π₯0 , π₯1 , π₯2 ,..., π₯π = π we have approximation of the integral similar to (11) as π₯ β 1 ∫π₯ π(π₯)ππ₯ = 2 (π¦0 + π¦1 ) 0 π₯2 ∫π₯ π(π₯)ππ₯ 1 β = 2 (π¦1 + π¦2 ) π₯ β 3 ∫π₯ π(π₯)ππ₯ = 2 (π¦2 + π¦3 ) 2 . . . . . . 8 π₯π−1 ∫π₯ π−2 β π(π₯)ππ₯ = 2 (π¦π−2 + π¦π−1 ) π₯π ∫π₯ β π−1 π(π₯)ππ₯ = 2 (π¦π−1 + π¦π ) Adding the integrals on the left side and formulas on the right down ward we have π₯ π₯ π₯ π₯ π₯ 1 2 3 π−1 π(π₯)ππ₯ + ∫π₯ π π(π₯)ππ₯ ∫π₯ π(π₯)ππ₯ + ∫π₯ π(π₯)ππ₯ + ∫π₯ π(π₯)ππ₯ + …+ ∫π₯ 0 1 β 2 β π−2 β π−1 β β = 2 (π¦0 + π¦1 ) + 2 (π¦1 + π¦2 ) + 2 (π¦2 + π¦3 ) + …+ 2 (π¦π−2 + π¦π−1 ) + 2 (π¦π−1 + π¦π ) β β β β β β β β = 2 π¦0 + 2 π¦1 + 2 π¦1 + 2 π¦2 + 2 π¦2 + 2 π¦3 + … + 2 π¦π−2 + 2 π¦π−1 + β β 2 2 β β π¦ + 2 π¦π 2 π−1 = π¦0 + βπ¦1 + βπ¦2 + βπ¦3 + …+ βπ¦π¦0 π−2 + βπ¦π−1 + π¦π β = 2 [π¦0 + 2(π¦1 + π¦2 + π¦3 + β― + π¦π−2 + π¦π−1 ) + π¦π ] Using the properties of definite integral on bounded continuous functions we get π₯ β 0 2 π ∫π₯ π(π₯)ππ₯ = [π¦0 + 2(π¦1 + π¦2 + π¦3 + β― + π¦π−2 + π¦π−1 ) + π¦π ]( say) (12) This last equation refers to us a general formula for composite trapezoidal rule Theorem 1 (Error of trapezoidal rule for single sub-interval) Let π: πΆ[π, π] → β be twice continuously differentiable. Then the error for the trapezoidal rule can be represented in the form π ∫π π(π₯)ππ₯ − π−π 2 [π(π) + π(π)] = − β3 12 π ′′ (π), with some π ∈ [π, π] and β = π − π Proof: let πΏ1 π denote the linear interpolation of π at the interpolation points π₯0 = π and π₯1 = π. By construction of the trapezoidal rule we have that the error π πΈ1 (π) = ∫π π(π₯)ππ₯ − π−π 2 [π(π) + π(π)] π π (13) Is given by πΈ1 (π) = ∫π [π(π₯) − (πΏ1 π)(π₯)] = ∫π (π₯ 2 − ππ₯ − ππ₯ + ππ) π(π₯)−(πΏ1 π)(π₯) ππ₯ (π₯−π)(π₯−π) Since the first factor of the integrand is non-positive on [π, π]and since by L’Hopital`s rule the second factor is continuous, from the mean value theorem for the integrals we obtain that πΈ1 (π) = π(π§)−(πΏ1 π)(π§) π ∫ (π₯ (π§−π)(π§−π) π π Now ∫π (π₯ − π)(π₯ − π)ππ₯ = − − π)(π₯ − π)ππ₯ For some π§ ∈ [π, π] β3 6 (15) With aid of Lagrange polynomial error formula for linear interpolation we have π ′′ (π) 2 = π(π§)−(πΏ1 π)(π§) (π§−π)(π§−π) (16) 9 (14) Substituting (16) and (15) in (14) we have β3 πΈ1 (π) = − 12 π ′′ (π) π π−π Now from (13) we have πΈ1 (π, π₯) = ∫π π(π₯)ππ₯ − [π(π) + π(π)] = − 2 β3 12 π ′′ (π) β Corollary The bound for the error is given by |πΈ1 (π)| ≤ (π−π)3 12 π2 , where π2 = max |π ′′ (π₯)| π≤π₯≤π Theorem 2 (error of composite trapezoidal rule) Let π: πΆ[π, π] → β be twice continuously differentiable. Then the error for the composite trapezoidal rule is given by π πΈ1 = ∫π π(π₯)ππ₯ − πβ (π) = − (π−π) 12 β2 π′′(π), for some π ∈ [π. π] Proof: Consider π π₯ π₯ π₯ 1 2 1 2 3 ∫π π(π₯)ππ₯ − πβ (π) = ∫π=π₯ π(π₯)ππ₯ + ∫π₯ π(π₯)ππ₯ + ∫π₯ π(π₯)ππ₯ + 0 π₯ β …+ ∫π₯ π π(π₯)ππ₯ − 2 π−1 π₯ 1 = ∫π=π₯ π(π₯)ππ₯ − 0 β 2 [π¦0 + 2(π¦1 + π¦2 + π¦3 + β― + π¦π−2 + π¦π−1 ) + π¦π ] π₯ β 1 2 [π¦0 + π¦1 ] + ∫π₯ 2 π(π₯)ππ₯ − π₯ …+ ∫π₯ π π(π₯)ππ₯ − π−1 β3 β3 π₯ β 2 2 [π¦1 + π¦2 ] + ∫π₯ 3 π(π₯)ππ₯ − β 2 [π¦2 + π¦3 ] + [π¦π−1 + π¦π ] β3 β3 = (− 12 π′′(π1 )) + (− 12 π′′(π2 )) + (− 12 π′′(π3 )) + …+ (− 12 π′′(ππ )) By theorem 1 Were π ≤ π1 ≤ π2 ≤ π3 ≤ …≤ ππ ≤ π β3 = − 12 [∑π=π π=1 π′′(ππ )] (17) From n(min π′′( π₯)) ≤ ∑π=π π=1 π′′(ππ ) ≤ π (max π′′(π₯)) and the π₯∈[π,π] π₯∈[π,π] continuity of π′′ we conclude that there exists π ∈ [π, π] such that π′′( π) = ′′ ∑π=π π=1 π (ππ ) π ⇒ π π′′( π) = ∑π=π π=1 π(ππ ) (18) Substituting (18) in (17) we have π ∫π π(π₯)ππ₯ − πβ (π) = β3 − 12 [∑π=π π=1 π′′(ππ )] =− π Therefore ∫π π(π₯)ππ₯ − πβ (π) = − β3 ′′ = − 12 π π (π) = − ( π−π 3 ) π 12 π π′′(π) = − (π−π)3 12π2 π′′(π) (π − π) 2 β π′′(π) 12 (π−π) 12 β2 π′′(π) for some π ∈ [π, π] 10 β 1 Example1 evaluate ∫0 π π₯ 2 ππ₯, using the trapezoid rule with2, 4, 8, and 16 equal subintervals. Solution: with π = 2, 4 and 8, we have the following step lengths and nodal points π = 2: π = 4: π = 8: β= π−π π β= β= π−π π π−π π 1 = 2. The nodes are 0, 0.5, and 1.0. 1 = 4.The nodes are 0, 0.25, 0.5, 0.75, and 1.0. 1 = 8.The nodes are 0, 0.125, 0.25, 0.375, 0.5, 0.675, 0.75, 0.875, and 1.0. We have the following tables of values. Table1.1 π=2 π₯ 0 0.5 1 π₯2 0 0.25 1 π(π₯) 1 1.284 2.718282 = ππ₯2 025 Table 1.2 π=4 π₯ 0 0.25 0.5 0.75 1 π₯2 0 0.0625 .25 0.5625 1 π(π₯) 1 1.064494 1.284025 1.755055 2.718282 = ππ₯2 Table 1.3 π π₯ 0 =8 0.12 0.25 0.375 0.5 0.625 0.75 0.875 1 0.0625 0.1406 0.25 0.3906 0.562 0.7656 1 25 5 25 5 π₯2 0 0.01 5625 π(π₯) = ππ₯2 1 25 1.01 1.06449 1.1509 1.2840 1.4779 1.755 2.1503 2.71 5748 4 93 25 04 055 38 Table 1.4 11 8282 π = 16 π₯ 0 0.0625 0.125 0.187 0.25 5 π₯2 0 0.312 0.3 0.4 5 75 375 0.5 0.5625 0.00390 0.0156 0.035 0.06 0.097 0.1 0.1 0.2 0.3164 6 25 156 25 656 40 914 5 06 62 06 5 π(π₯) 1 = ππ₯2 1.00391 1.0157 1.035 1.06 1.102 1.1 1.2 1.2 1.3721 4 48 781 4494 583 50 109 84 87 99 50 02 2 π₯ 0.62 0.6875 0.75 5 π₯2 0.5625 0.812 0.87 0.937 5 5 5 0.660 0.76 0.878 156 5625 906 5 1 0.39 0.47265 1 0625 6 π(π₯) 1.47 1.60424 1.7550 1.935 2.15 2.408 2.7 = ππ₯2 7904 9 55 094 0338 264 18 28 2 Now we compute the value of the integral. β For π = 2: πΌπ = 2 [π(0) + 2π(0.5) + π(1)] = 0.25[1 + 2(1.284025) + 2.718282] = 1.571583 β For π = 4: πΌπ = 2 [π(0) + 2(π(0.25) + π(0.5) + π(0.75)) + π(1)] = 0.125[1 + 2(1.064494 + 1.284025 + 1.755055) + 2.718282] = 1.490679For π = 8: β πΌπ = 2 [π(0) + 2(π(0.125) + π(0.25) + π(0.375) + π(0.5)π(0.675) + π(0.75)) + π(1)] β = 2 [1 + 2(1.015748 + 1.064494 + 1.150993 + 1.284025 + 1.477904 + 1.755055 + 2.150338) + 2.718282] = 1.469712 12 β For π = 16: πΌπ = 2 [π(0) + 2(π(0.0625) + π(0.125) + π(0.1875) + π(0.25) + π(0.3125) + π(0.375) + π(0.4375) + π(0.5) + π(0.5625) + π(0.625) + π(0.6875) + β π(0.75) + π(0.8125) + π(0.875) + π(0.9375)) + π(1)] = 2 [1 + 2(1.003914 + 1.015748 + 1.035781 + 1.064494 + 1.102583 + 1.150993 + 1.210951 + 1.284025 + 1.372187 + 1.477904 + 1.604249 + 1.755055 + 1.935094 + 2.150338 + 2.408264) + 2.718282] = 1.464422 Remark1 ο Trapezoidal Rule is the simplest quadrature formula. ο The formula is applicable to both even and odd numbers of subintervals ο Geometrical interpretation of trapezoid rule is the area under the curve π¦ = π(π₯) is replaced by the area of a trapezium. And hence the total area under the curve from π₯0 to π₯π is replaced by the sum of the area of n trapeziums. ο The method is order one. 2.1.2. Simpson`s rule 2.1.2.1. Simpson`s one-third rule Putting π = 2 in Newton-Cote`s quadrature formula (10) and taking the curve through (π₯0 , π¦0) , (π₯1 , π¦1) and (π₯2 , π¦2) as a parabola that is a polynomial of second order so that differences of order higher than second vanish, we get π₯ 2 2 ∫π₯ π(π₯)ππ₯ = 2β [π¦0 + 2 βπ¦0 + 2(2(2)−3) 12 0 π π₯ 1 β β2 π¦0 ] = 2β [π¦0 + βπ¦0 + 6 β2 π¦0 ] = 3 [π¦0 + 4π¦1 + π¦2 ] β β Thus ∫π π(π₯)ππ₯ = ∫π₯ 2 π(π₯)ππ₯ = 3 [π¦0 + 4π¦1 + π¦2 ] = 3 [π(π) + 4π( 0 π+π 2 ) + π(π) ] (19) This formula is called Simpson`s one-third rule To derive a formula for composite Simpson’s one third rule let divide the interval [π, π] as the following sub-intervals [π = π₯0 , π₯2 ] ,[π₯2 , π₯4 ] ,[π₯4 , π₯6 ] ,…,[π₯π−2 , π₯π ] assuming the curve passing through each sub-interval be polynomial function of degree two which assumes the same value as a function at the end points of the sub-intervals and middle points as shown in figure below 13 π¦0 π¦1 π¦2 π = π₯0 π₯1 π₯2 π¦3 π¦4 π₯3 π₯4 π₯5 π₯6 π₯π Figure 3 π₯ β Now ∫π₯ 2 π(π₯)ππ₯ = 3 [π¦0 + 4π¦1 + π¦2 ] by Simpson’s one-third 0 rule π₯4 ∫ π(π₯)ππ₯ = π₯2 π₯ β [π¦ + 4π¦3 + π¦4 ] 3 2 β 6 ∫π₯ π(π₯)ππ₯ = 3 [π¦4 + 4π¦5 + π¦6 ] . 4 π₯π ∫π₯ . . . . . . β π−2 π(π₯)ππ₯ = 3 [π¦π−2 + 4π¦π−1 + π¦π ] π Being even Adding all integrals on the left side and adding all formulas on the right side we have π₯ π₯ π₯ π₯ 2 4 6 π ∫π₯ π(π₯)ππ₯ + ∫π₯ π(π₯)ππ₯ + ∫π₯ π(π₯)ππ₯ + …+ ∫π₯ π(π₯)ππ₯ 0 β 2 β = 3 [π¦0 + 4π¦1 + π¦2 ] + 3 [π¦2 + 4π¦3 + π¦4 ] + 4 β 3 π−2 β [π¦4 + 4π¦5 + π¦6 ] + β― + [π¦π−2 + 4π¦π−1 + 3 π¦π ] β = 3 [π¦0 + 4π¦1 + π¦2 + π¦2 + 4π¦3 + π¦4 + π¦4 + 4π¦5 + π¦6 + β― + π¦π−2 + 4π¦π−1 + π¦π ] β = 3 [π¦0 + 4π¦1 + 2π¦2 + 4π¦3 + 2π¦4 + 4π¦5 + π¦6 + β― + π¦π−2 + 4π¦π−1 + π¦π ] β = 3 [π¦0 + 4(π¦1 + π¦3 + π¦5 + β― + π¦π−1 ) + 2(π¦2 + π¦4 + π¦6 + β― + π¦π−2 ) + π¦π ] Now using the properties of definite on continuous and bounded function π we have π₯ β π ∫π₯ π(π₯)ππ₯ = 3 [π¦0 + 4(π¦1 + π¦3 + π¦5 + β― + π¦π−1 ) + 2(π¦2 + π¦4 + π¦6 + β― + π¦π−2 ) + π¦π ] 0 = πβ (π) (say) (20) This formula is known as composite Simpson’s rule for Simpson’s one third rule. 14 Theorem 3 Let π: πΆ[π, π] → β be four times continuously differentiable. Then the error for Simpson’s one third rule can represented in the form π β πΈ2 (π, π₯) = ∫π π(π₯)ππ₯ − 3 [π(π) + 4π( π+π 2 β5 ) + π(π) ] = − 90 π 4 ( π) π−π For some π ∈ [π, π] and β = 2 Proof: using definition of the error given in (4) and taking π(π₯) = 1, π₯, π₯ 2 , π₯ 3 , π₯ 4 we have For: π π−π π(π₯) = 1 : πΈ2 (π, π₯) = ∫π ππ₯ − 6 =π−π– [π(π) + 4π( π−π 6 π 1 = 2 π−π (b2 − a2 ) − = 1 2 6 π−π 6 ) + π(π) ] π+π [π(π) + 4π ( π+π [π + 4 ( (b2 − a2 ) − 2 (1 + 4 + 1) = π − π − (π − π) = 0 π(π₯) = π₯ : πΈ2 (π, π₯) = ∫π π₯ππ₯ − For: π+π 2 2 ) + π(π)] ) + π] (π−π)[3π+3π] 6 = 3b2 − 3a2 − 3ab − 3b2 + 3a2 + 3ab = 0 π π(π₯) = π₯ 2 : πΈ2 (π, π₯) = ∫π π₯ 2 ππ₯ − For: 1 (b3 − a3 ) − 3 = = = π3 3 − π−π (b3 − a3 ) − π3 3 − ππ 2 3 − 6 ππ2 3 = 1 4 1 (b4 − a4 ) − 4 − (b4 − a4 ) − = π−π 6 π−π 6 [π(π) + 4π( π+π 2 ) + π(π) ] π+π 2 ) + b2 ] 2 [2a2 + 2ππ + 2b2 ] π3 3 + π3 π = For: 2 6 6 [a2 + 4 ( π(π₯) = π₯ 3 : πΈ2 (π, π₯) = ∫π π₯ 3 ππ₯ − For: = 1 π−π π−π 3 + π−π 6 ππ2 3 + ππ 2 3 =0 [π(π) + 4π( π+π 2 ) + π(π) ] π+π 3 [a3 + 4 ( 2 ) + b3 ] 1 [a3 + 2 (a3 + (a3 + 3ππ 2 + 3ππ2 +b3 ) + b3 ] 1 4 π−π 3 (a + ππ 2 + ππ2 +b3 ) (b − a4 ) − 4 4 π 4 π4 ππ3 π 2 π2 ππ 3 π 4 π4 ππ3 π 2 π2 ππ 3 − − − − − + + + + =0 4 4 4 4 4 4 4 4 4 4 π π(π₯) = π₯ 4 : πΈ2 (π, π₯) = ∫π π₯ 4 ππ₯ − = 1 (b5 − a5 ) − 5 π−π 6 π−π 6 π+π 4 [a4 + 4 ( 15 [π(π) + 4π( 2 ) + b4 ] π+π 2 ) + π(π) ] = = 1 (b5 − a5 ) – 5 1 (b5 − a5 ) − 5 = π5 5 − = π5 5 + π−π π−π 24 6 π−π 6 [a4 + [a4 + (π+π)4 4 + b4 ] a4 +4π3 π+6π2 π2 +4ab3 +b4 4 + b4 ] (5a4 +4π3 π + 6π2 π 2 +4ab3 + 5b4 ) 24(π 5 −π5 )−5(π−π)[5a4 +4π3 π+6π2 π2 +4ab3 +5b4 ] 120 π−π [−24(a4 +π3 π + π2 π 2 +ab3 + b4 ) + 25a4 +20π3 π + 30π2 π 2 +20ab3 + 25b4 ] 120 π−π [−24a4 −24π3 π − 24π2 π 2 −24ab3 − 24b4 + 25a4 +20π3 π + 30π2 π 2 +20ab3 =− 120 =− + 25b4 ] π−π = − 120 [π4 − 4π3 π + ππ2 π 2 − 4π3 π + ππ2 π 2 − 4ππ 3 + π 4 ] =− =− π−π (π − π)4 120 (π−π)5 120 ≠0 (21) Hence, the Simpson’s one-third rule integrates exactly polynomial of degree less than or equal to three. Therefore, the method is of order 3. Now the error is obtained for (π₯) = π₯ 4 , using error formula (6) we have π π πΈ2 (π, π₯) = 4! π 4 (π), Where π = ∫π π₯ 4 ππ₯ − π−π 6 π+π [π(π) + 4π ( As given in (21) above, π = − π Therefore πΈ2 (π, π₯) = 4! π 4 (π) = − =− =− π β5 90 π 4 (π) 2 ) + π(π)] (π−π)5 120 (π−π)5 π 4 (π) 120 4! (2β)5 π 4 (π) 120 24 By β = β Hence πΈ2 (π, π₯) = ∫π π(π₯)ππ₯ − 3 [π(π) + 4π( π−π 2 π+π 2 and π ≤ π ≤ π β5 ) + π(π) ] = − 90 π 4 (π) β Note: the method is one order higher than expected, since we have approximated π(π₯) by polynomial of degree 2 only. 16 Theorem 4 Let π: πΆ[π, π] → β be four times continuously differentiable. Then the error for the composite Simpson’s one-third rule is given by π π−π πΈ2 (π, π₯) = ∫π π(π₯)ππ₯ − πβ (π) = − 180 β4 π 4 (π) For some π ∈ [π, π] Proof: π π₯ π₯ π₯ 0 2 4 Consider ∫π π(π₯)ππ₯ − πβ (π) = ∫π₯ 2 π(π₯)ππ₯ + ∫π₯ 4 π(π₯)ππ₯ + ∫π₯ 6 π(π₯)ππ₯ + …+ π₯π ∫π₯ π(π₯)ππ₯ π−2 β − 3 [π¦0 + 4(π¦1 + π¦3 + π¦5 + β― + π¦π−1 ) + 2(π¦2 + π¦4 + π¦6 + β― + π¦π−2 ) + π¦π ] π₯ π₯ π₯ 0 2 4 π₯ β = ∫π₯ 2 π(π₯)ππ₯ + ∫π₯ 4 π(π₯)ππ₯ + ∫π₯ 6 π(π₯)ππ₯ + …+ ∫π₯ π π(π₯)ππ₯ − 3 [π¦0 + 4π¦1 + 2π¦2 + 4π¦3 + π−2 2π¦4 + 4π¦5 + π¦6 + β― + π¦π−2 + 4π¦π−1 + π¦π ] π₯ π₯ β π₯ β = ∫π₯ 2 π(π₯)ππ₯ − 3 [π¦0 + 4π¦1 + π¦2 ] + ∫π₯ 4 π(π₯)ππ₯ − 3 [π¦2 + 4π¦3 + π¦4 ] + ∫π₯ 6 π(π₯)ππ₯ − 0 2 β 3 β5 [π¦4 + 4π¦5 + π¦6 ] + β5 4 π₯ …+ ∫π₯ π π(π₯)ππ₯ π−2 β − 3 [π¦π−2 + 4π¦π−1 + π¦π ] β5 β5 = − 90 π 4 (π1 ) + (− 90 π 4 (π2 )) + (− 90 π 4 (π3 )) + β― + (− 90 π 4 (ππ )) By theorem 3 Where, π = π₯0 ≤ π1 ≤ π₯2 ,π₯2 ≤ π2 ≤ π₯4 ,…,π₯π−2 ≤ ππ ≤ π₯π = π β5 = − 90 [π 4 (π1 ) + π 4 (π2 ) + π 4 (π3 ) + β― + π 4 (ππ )] β5 4 = − 90 ∑π π=1 π (ππ ) (22) 4 4 From π min π 4 (π₯) ≤ ∑π π=1 π (ππ ) ≤ π max π (π₯) π₯∈[π,π] π₯∈[π,π] And the continuity of π 4 we conclude that there exists π ∈ [π, π] such that π 4 (π) = 4 ∑π π=1 π (ππ ) π 4 Implies Nπ 4 (π) = ∑π π=1 π (ππ ) (23) Substituting (23) in (22) we get β5 π ∫π π(π₯)ππ₯ − πβ (π) = − 90 Nπ 4 (π) 17 =− π−π 5 ) π ( 90 Nπ 4 (π) By β = π−π π = π−π 2π π−π ( 2π )5 =− Nπ 4 (π) 90 =− 1 (π − π)5 Nπ 4 (π) 90 25 π 5 =− =− 1 (π − π)5 4 π (π) 90 25 π 4 1 (π − π)(π − π)4 4 π (π) 90 2(2π)4 =− π (π − π)β4 4 π (π) 180 Hence πΈ2 (π, π₯) = ∫π π(π₯)ππ₯ − πβ (π) = − (π−π)β4 180 π 4 (π) β The bound on the error is given by πΈ2 (π, π₯) ≤ (π − π)β4 π4 180 Or πΈ2 (π, π₯) ≤ (π − π)β5 π 2880π 4 4 π4 = max |π 4 (π₯) | and πβ = Where π≤π₯≤π π−π 2 Remark: ο Simpson’s 1/3 rule and composite Simpson’s 1/3 rule are of order 3. ο The number of subintervals is even positive integer. ο As N increases, the error decrease. ο Geometrical interpretation of Simpson’s 1/3 rule is that the area under the curve is reπ placed by 2 arcs of second degree polynomials, or parabolas with axes taken in the vertical direction. 18 Example 2 1 2 Find the approximate value of πΌ = ∫0 π π₯ ππ₯, using Simpson’s 1/3 rule with 4, 8, and 16 equal subintervals and find bounds of the errors. Solution: with π = 2π = 4, 8, πππ 16 or π = 2, 4, 8 we have the following step lengths and nodal points. π = 2: π = 4: β= π = 8: β= π−π 2π π−π 2π β= π−π 2π 1 = 4.The nodes are 0, 0.25, 0.5, 0.75, and 1.0. 1 = 8.The nodes are 0, 0.125, 0.25, 0.375, 0.5, 0.675, 0.75, 0.875, and 1.0. 1 = 16.The nodes are 0, 0.0625, 0.125, 0.1875, 0.25, 0.3125, 0.375, 0.4375, 0.5, 0.5625, 0.625, 0.675, 0.6875, 0.75, 0.875, 0.9375 and 1.0. We have the following tables of values. Table 2.1 π=4 π₯ 0 0.25 0.5 0.75 1 π₯2 0 0.0625 .25 0.5625 1 π(π₯) 1 1.064494 1.284025 1.755055 2.718282 Table 2.2 π π₯ 0 0.125 0.25 0.375 0.5 0.675 0.75 0.875 1 =8 π₯2 0 0.015625 0.0625 0.14062 0.25 0.45562 0.5625 0.76562 1 5 π(π₯) 1 1.015748 5 5 1.0644 1.15099 1.28402 1.47790 1.75505 2.15033 94 3 5 4 5 8 19 2.718282 Table2.3 π = 16 π₯ 0 And 0.06 0.1 0.1 0.2 0.3 25 25 875 5 12 π(π₯) = ππ₯ 0.375 0.437 0.5 5 0.562 5 5 2 π₯2 0 0.00 0.0 0.0 0.0 0.0 0.140 0.191 3906 156 351 625 97 625 406 25 56 0.25 0.316 406 65 6 π(π₯) 1 1.00 1.0 1.0 1.0 1.1 1.150 1.210 1.28 1.372 3914 157 357 644 02 993 951 402 188 48 81 94 58 5 3 π₯ 0.6 0.68 0.7 0.8 0.8 0.9 25 5 125 75 37 75 1 5 π₯2 0.3 0.47 0.5 0.6 0.7 0.8 90 625 601 656 78 56 25 90 2656 62 5 π(π₯) 1 6 1.4 1.60 1.7 1.9 2.1 2.4 2.718 77 550 350 503 08 282 55 94 38 26 4249 90 4 4 Now, we compute value of the integral. π = 2π = 4: = β πΌ1 = 3 [π(0) + 4(π(0.25) + π(0.75)) + 2π(0.5) + π(1)] 0.25 [1 + 4(1.064494 + 1.755055) + 2(1.284025) + 2.718282] 3 = 1.463711 20 β π = 2π = 8: πΌ2 = 3 [π(0) + 4(π(0.125) + π(0.375) + π(0.675) + π(0.875)+ = 2(π(0.25) + π(0.5) + π(0.75)) + π(1)] = 0.125 [1 + 4(1.015748 + 1.150993 + 1.477904 + 2.150338) 3 + 2(1.064494 + 1.284025 + 1.755055) + 2.718282] = 1.462409 β π = 2π = 16: πΌ2 = 3 [π(0) + 4(π(0.0625) + π(0.1875) + π(0.3125) + π(0.4375) + π(0.5625) + π(0.6875) + π(0.8125) + π(0.9375) + 2(π(0.125) + π(0.25) + π(0.375)) + π(0.5) + π(0.625) + π(0.75) + π(0.875)) + π(1)] = 0.125 3 [1 + 4(1.003914 + 1.035781 + 1.102583 + 1.210951 + 1.372187 + 1.604249 + 1.935094 + 2.408264 + 2(1.015748 + 1.064494 + 1.150993 + 1.284025 + 1.477904 + 1.755055 + 2.150338) + 2.718282] = 1.462656 2.1.2.2. Simpson’s 3/8 rule Putting π = 3 in Newton-Cotes formula and taking the curve through (π₯π , π¦π ): π = 0,1,2,3 as polynomial of third order so that differences above the third order vanish, we get π π₯ 3 3 1 3 3 ∫π π(π₯)ππ₯ = ∫π₯ π(π₯)ππ₯ = 3β(π¦0 + 2 βπ¦0 + 2 β2 π¦0 + 8 β3 π¦0 ) = 8 β(π¦0 + 3π¦1 + 3π¦2 + π¦3 ) 0 π 3 Thus ∫π π(π₯)ππ₯ = 8 β(π¦0 + 3π¦1 + 3π¦2 + π¦3 ) (24) This is called Simpson’s 3/8 rule. Similarly we can derive composite Simpsons 3/8 rule. π₯3 3 ∫ π(π₯)ππ₯ = β(π¦0 + 3π¦1 + 3π¦2 + π¦3 ) 8 π₯0 π₯6 3 ∫ π(π₯)ππ₯ = β(π¦3 + 3π¦4 + 3π¦5 + π¦6 ) 8 π₯3 π₯9 3 ∫ π(π₯)ππ₯ = β(π¦6 + 3π¦7 + 3π¦8 + π¦9 ) 8 π₯6 π₯π ∫π₯ π−3 . . . . . . 3 π(π₯)ππ₯ = 8 β(π¦π−3 + 3π¦π−2 + 3π¦π−1 + π¦π ) 21 Adding the right and left expressions we get π₯ π₯ 0 3 π₯ π₯ 3 3 6 9 π ∫π₯ π(π₯)ππ₯ + ∫π₯ π(π₯)ππ₯ + ∫π₯ π(π₯)ππ₯ + β― + ∫π₯ π(π₯)ππ₯ = 8 β(π¦0 + 3π¦1 + 3π¦2 + π¦3 ) + 6 3 π−3 3 3 β(π¦3 + 3π¦4 + 3π¦5 + π¦6 ) + 8 β(π¦6 + 3π¦7 + 3π¦8 + π¦9 ) + β― + 8 β(π¦π−3 + 3π¦π−2 + 3π¦π−1 + 8 π¦π ) = (π¦0 + 3π¦1 + 3π¦2 + π¦3 ) + (π¦3 + 3π¦4 + 3π¦5 + π¦6 ) + (π¦6 + 3π¦7 + 3π¦8 + π¦9 ) + β― + (π¦π−3 + 3π¦π−2 + 3π¦π−1 + π¦π ) = 3 β(π¦0 + 3(π¦1 + π¦2 + π¦4 + π¦5 + β― + π¦π−1 ) + 2(π¦3 + π¦6 +π¦9 + β― + π¦π−3 ) + π¦π ) 8 Since π(π₯) is continuous we have π₯ 3 π ∫π₯ π(π₯)ππ₯ = 8 β(π¦0 + 3(π¦1 + π¦2 + π¦4 + π¦5 + β― + π¦π−1 ) + 2(π¦3 + π¦6 +π¦9 + β― + π¦π−3 ) + π¦π ) 0 This is called composite Simpson’s 3/8 rule. Theorem 5 Let π: β[π, π] → β be four times continuously differentiable then the error for Simpson’s rule can be represented in the form π πΈ3 (π, π₯) = ∫π π(π₯)ππ₯ − For some π ∈ [π, π] and β = 3β 8 [π(π) + 3π ( 2π+π 3 π+2π ) + 3π ( 3 ) + π(π) ] = − 3β5 80 π 4 (π) π−π 3 Proof: Consider the error expression π πΈ3 (π, π₯) = ∫π π(π₯)ππ₯ − 3β 8 2π+π [π(π) + 3π ( 3 π+2π ) + 3π ( 3 ) + π(π) ] For π(π₯) = 1, π₯, π₯ 2 , π₯ 3 , π₯ 4 , evaluate πΈ3 (π, π₯). π Now for π(π₯) = 1 : πΈ3 (π, π₯) = ∫π 1ππ₯ − 3β 8 2π+π [π(π) + 3π ( 3 π+2π ) + 3π ( 3 ) + π(π) ] π−π 3( 3 ) π−π [1 + 3(1) + 3(1) + 1] = π − π − = (π − π) − 8=0 8 8 π For:π(π₯) = π₯: πΈ3 (π, π₯) = ∫π π₯ππ₯ − = = π 2 −π2 2 − π−π 8 3β 8 2π+π [π(π) + 3π ( 3 ) + 3π ( π+2π 3 ) + π(π) ] π 2 − π2 π − π 2π + π π + 2π − [π + 3 ( ) + 3( ) + π] 2 8 3 3 [π + 2π + π + π + 2π + π] = = 22 π 2 −π2 2 − π−π 8 [4π + 4π] = π 2 −π2 2 − π 2 −π2 2 =0 π For: π(π₯) = π₯ 2 : πΈ3 (π, π₯) = ∫π π₯ 2 ππ₯ − 3β 8 2π+π [π(π) + 3π ( 3 ) + 3π ( π+2π 3 ) + π(π) ] π 3 − π3 3( π − π 2π + π 2 π + 2π 2 2 = − ) [π + 3( ) + 3( ) + π2 ] 3 8 3 3 3 = π 3 − π3 π − π 2 4π2 + 4ππ + π 2 π2 + 4ππ + 4π 2 − [π + + + π2] 3 8 3 3 = π 3 −π3 3 − π−π 3π2 +4π2 +4ππ+π 2 +π2 +4ππ+4π2 +3π2 [ 8 π 3 −π3 = = 3 π 3 −π3 3 = 3 − 3 π2 +ππ+π 2 3 ] ) π 3 − π3 π 3 − π3 − =0 3 3 π 3β 8 [π(π) + 3π ( 2π+π 3 π+2π ) + 3π ( 3 ) + π(π) ] π 4 − π4 3( π − π) 3 2π + π 3 π + 2π 3 − [π + 3 ( ) + 3( ) + π3 ] 4 8 3 3 3 = = [ 8 −(π − π)( For π(π₯) = π₯ 3 : πΈ3 (π, π₯) = ∫π π₯ 3 ππ₯ − = π−π 8π2 +8ππ+8π 2 ] π 4 − π4 (π − π) 3 (2π + π)3 (π + 2π)3 − [π + + + π3] 4 8 9 9 π 4 − π4 (π − π) 9π3 + 8π3 + 12π2 π + 6ππ 2 + π 3 + π3 + 6π2 π + 12ππ 2 + 8π 3 + 9π 3 − [ ] 4 8 9 π 4 − π4 (π − π) 18π3 + 18π2 π + 18ππ 2 + 18π 3 = − [ ] 4 8 9 π 4 − π4 (π − π) [2π3 + 2π2 π + 2ππ 2 + 2π 3 ] = − 4 8 = π 4 − π4 (π − π) 3 [π + π2 π + ππ 2 + π 3 ] − 4 4 = π 4 − π4 π 4 − π4 − =0 4 4 π For π(π₯) = π₯ 4 : πΈ3 (π, π₯) = ∫π π₯ 4 ππ₯ − = π5 5 = − π5 5 − 3β 8 [π(π) + 3π ( 2π+π 3 π+2π ) + 3π ( 3 ) + π(π) ] (π−π) 27π4 +16π4 +32π3 π+24π2 π 2 +8ππ 3 +π 4 +π4 +8π3 π+24π2 π2 +32ππ 3 +16π4 +27π 4 [ 8 (π−π)(π 4 +π3 π+π 2 π2 +ππ3 +π4 ) 5 27 − (π−π) 216 [44π4 + 40π3 π + 48π2 π 2 + 40ππ 3 + 44π 4 ] 23 ] =− (π − π) (−216π 4 − 216π 3 π − 216π 2 π2 − 216ππ3 − 216π4 + 220π4 + 200π3 π 1080 + 240π2 π 2 + 200ππ 3 + 220π 4 ) (π − π) (4π4 − 16π3 π + 24π2 π 2 − 16ππ 3 + 4π 4 ) 1080 (π − π) 4 (π − 4π3 π + 6π2 π 2 − 4ππ 3 + π 4 ) =− 270 =− =− (π−π) 270 (π − π)4 = − (π−π)5 270 =− (3β)5 270 =− 35 β5 270 9 = − 10 β5 ≠ 0 9 Now by definition of error constant π = − 10 β5 Now the error term for the expression is, π 9 3 πΈ3 (π, π₯) = 4! π 4 (π) = − 10×4×3×2×1 β5 π 4 (π) = − 80 π 4 (π) π Thus πΈ3 (π, π₯) = ∫π π(π₯)ππ₯ − 3β 8 [π(π) + 3π ( 2π+π 3 π+2π ) + 3π ( 3 ) + π(π) ] = − 3β5 80 π 4 (π) β The bound for the error term is given by, |πΈ3 (π, π₯)| ≤ − 3β5 80 M4 Where π4 = max | π 4 (π)| π≤π₯≤π Remark: ο The number of sub-intervals must be divisible by three. ο Its error constant c is larger than the Simpson 1/3 rule. ο The number of nodal points must be odd. ο Geometrical interpretation of Simpson’s 3/8 rule is the area under the curve in which the π curve is replaced by 3 arcs of third degree polynomial. 24 Theorem 6(error of composite 3/8 rule) Letπ: [π, π] → β be four times continuously differentiable. Then the error for the composite Simpson’s rule is given by π ∫π π(π₯)ππ₯ − πβ (π) = − (π−π)β4 80 π 4 (π) For some π ∈ [π, π] and β = π−π π = π−π 3π Proof: π π₯ π₯ π₯ 0 3 6 π₯ Consider ∫π π(π₯)ππ₯ − πβ (π) = ∫π₯ 3 π(π₯)ππ₯ + ∫π₯ 6 π(π₯)ππ₯ + ∫π₯ 9 π(π₯)ππ₯ + β― + ∫π₯ π π(π₯)ππ₯ − π−3 3 β(π¦0 + 3(π¦1 + π¦2 + π¦4 + π¦5 + β― + π¦π−1 ) + 2(π¦3 + π¦6 +π¦9 + β― + π¦π−3 ) + π¦π ) 8 π₯ π₯ π₯ 0 3 6 π₯ 3 = ∫π₯ 3 π(π₯)ππ₯ + ∫π₯ 6 π(π₯)ππ₯ + ∫π₯ 9 π(π₯)ππ₯ + β― + ∫π₯ π π(π₯)ππ₯ − 8 β [(π¦0 + 3π¦1 + 3π¦2 + π¦3 ) + π−3 (π¦3 + 3π¦4 + 3π¦5 + π¦6 ) + (π¦6 + 3π¦7 + 3π¦8 + π¦9 ) + β― + (π¦π−3 + 3π¦π−2 + 3π¦π−1 + π¦π )] π₯ π₯ 3 3 = ∫π₯ 3 π(π₯)ππ₯ − 8 β (π¦0 + 3π¦1 + 3π¦2 + π¦3 ) + ∫π₯ 6 π(π₯)ππ₯ − 8 β (π¦3 + 3π¦4 + 3π¦5 + π¦6 ) + 0 3 π₯9 ∫π₯ π(π₯)ππ₯ 6 3 − 8 β (π¦6 + 3π¦7 + 3π¦8 + π¦9 ) + β― + π₯π ∫π₯ π(π₯)ππ₯ π−3 3 − 8 β (π¦π−3 + 3π¦π−2 + 3π¦π−1 + π¦π ) = (− 3 80 β5 π 4 (π1 )) + (− 3 80 3 β5 π 4 (π2 )) + (− 80 β5 π 4 (π3 )) + β― + (− 3 80 β5 π 4 (ππ )) Where, π = π₯0 ≤ π1 ≤ π₯3 ,π₯3 ≤ π2 ≤ π₯6 , π₯6 ≤ π2 ≤ π₯9, …,π₯π−2 ≤ ππ ≤ π₯π = π =− =− 3 80 3 80 β5 [π 4 (π1 ) + π 4 (π2 ) + π 4 (π3 ) + β― + π 4 (ππ )] 4 β 5 ∑π π=1 π ( ππ ) ( 25) 4 4 4 From π min π 4 (π₯) ≤ ∑π π=1 π ( ππ ) ≤ π max π (π₯) and the continuity of π π₯∈[π,π] π₯∈[π,π] We conclude that there exists π ∈ [π, π] such that π 4 (π) = 4 ∑π π=1 π (ππ ) π 4 ππ 4 (π) = ∑π π=1 π ( ππ ) Implies Substituting (26) in (25) we get, π ∫π π(π₯)ππ₯ − πβ (π) = − 3 3 80 β5 (ππ 4 (π)) π−π 5 = − 80 ( 3π ) ππ 4 (π) 3 (π−π)5 = − 80 35 π 5 1 (π−π)5 ππ 4 (π) = − 80 34 π 4 π 4 (π) = − 25 (π−π) (π−π)4 80 34 π 4 π 4 (π) = − (π−π) 80 π−π 4 ( 3π ) π 4 (π) =− (π − π) 4 4 β π (π) 80 π Therefore ∫π π(π₯)ππ₯ − πβ (π) = − 1 (π−π) 80 β4 π 4 (π) β 2 Example 2 Find the approximate value of πΌ = ∫0 π π₯ ππ₯, using Simpson’s 3/8 rule with 6, and 12 equal subintervals and find bounds of the errors. Solution: with π = 3π = 6 πππ 12 or π =2 and 4 we have the following step lengths and nodal points. π = 2, β = π = 4: β= π−π 3π π−π 3π 1 = 6.The nodes are 0, 0.1667, 0.3333, 0.5001, 0.6668, 0.8334, and 1.0. 1 = 12.The nodes are 0, 0.0833, 0.1666, 0.2499, 0.3332, 0.4165, 0.4998, 0.5831, 0.6664, 0.7497, 0.8330, 0.9163, and 1.0. We have the following tables of values. Table 3.1 π=6 π₯ 0 0.1667 0.3333 0.5001 0.6668 0.8334 1 π₯2 0 0.0278 0.1111 0.2501 0.4446 0.6946 1 π(π₯) 1 1.0282 1.1175 1.2842 1.5599 2.0029 2.7183 And π(π₯) = π π₯2 Table 3.2 π = 12 π₯ 0 0.833 0.1666 0.2499 And π(π₯) π₯2 0 0.0069 0.0278 0.0624 = ππ₯2 π(π₯) π₯ 1 0.749 1.0069 1.0282 1.0644 08330 0.9163 1 0.6944 0.8396 1 2.0025 2.3154 2.7183 7 π₯2 0.562 0 π(π₯) 1.754 2 26 0.3 0.416 0.499 0.58 332 5 8 31 0.1 0.173 0.249 0.34 110 4 8 00 1.1 1.189 1.283 1.40 174 3 8 49 0.6664 0.4441 1.5591 Now, we compute value of the integral. π = 3π = 6: πΌ1 = 3β 8 [π(0) + 3(π(0.1667) + π(0.3333) + π(0.6668) + π(0.8334)) + 2π(0.5001) + π(1)] = 0.5 [1 + 3(1.0282 + 1.1175 + 1.5599 + 2.0029) + 2(1.2842) + 2.718282] = 8 =1.463554 π = 3π = 12: πΌ2 = 3β 8 [π(0) + 3(π(0.0833) + π(0.1666) + π(0.3332) + π(0.4165) + π(0.5831) + π(0.6664) + π(0.8330) + π(0.9163) + 2(π(0.2499) + π(0.4998) + π(0.7497)) + π(1)] = 0.25 [1 + 3(1.0069 + 1.0282 + 1.1174 + 1.1893 + 1.4049 + 1.5591 + 2.0025 + 2.3154) 8 + 2(1.0644 + 1.2838 + 1.7542) + 2.718282] = 1.462318 |πΈ3 | ≤ (π − π) 4 β max |π 4 (π₯)| π≤π₯≤π 80 1 For n = 6, πΈ3 ≤ 80π₯64 76e=0.001993 1 For n=12, πΈ3 ≤ 80π₯124 76e=0.000124 2.2. Romberg method While computing the value of the integral with a particular step length, the values of the integral with a particular step length, the values of the integral obtained earlier by using larger step length were not used. Further, convergence may be slow. Romberg method is a powerful tool which uses the method of extrapolation. We compute the value of the integral with a number of step lengths using the same method. Usually, we start with a course step length, then reduce the step lengths and compute the value of the integral again. The sequence of these values converges to the exact value of the integral. Romberg method uses these values of the integral obtained with various step lengths, to refine the solution such that the new values are of higher order. That is, as if the results are obtained using a higher order method than the order of the method used. The extrapolation method is derived by studying the error of the method that is being used. Let us derive the Romberg method for the trapezium and Simpson’s rules. 27 2.2.1. Romberg method for the trapezium rule Let the integral π πΌ = ∫ π(π₯)ππ₯ π be computed by the composite trapezium rule. Let πΌ denote the exact value of the integral and πΌπ denote the value obtained by the composite trapezium rule. The error, πΌ − πΌπ , in the composite trapezium rule in computing the integral is given by πΌ − πΌπ = π1 β2 + π2 β4 + π3 β6 + β― Or πΌ = πΌπ + π1 β2 + π2 β4 + π3 β6 + β― Where, π1 , π2 , π3 , … are independent of β to illustrate the extrapolation procedure, first consider two error terms πΌ = πΌπ + π1 β2 + π2 β4 Let πΌ be evaluated using two step lengths β and πβ, 0 < π < 1 Let these values be denoted by πΌπ (β) andπΌπ (πβ). The error equations become πΌ = πΌπ (β) + π1 (β2 ) + π2 (β4 ) (26) πΌ = πΌπ (πβ) + π1 π 2 β2 + π2 π 4 β4 (27) From (26) we obtain πΌ − πΌπ (β) = π1 (β2 ) + π2 (β4 ) (28) From (27), we obtain πΌ − πΌπ (πβ) = π1 π 2 β2 + π2 π 4 β4 (29) Multiply (28) by π 2 to obtain π 2 (πΌ − πΌπ (β)) = π1 π 2 β2 + π2 π 2 β4 (30) Eliminating π1 π 2 β2 from (29) and (30), we obtain (1 − π 2 )πΌ − πΌπ (πβ) + π 2 πΌπ (β) = π2 π 2 β4 (π 2 − 1) Solving for πΌ, we obtain πΌ= πΌπ (πβ)−π 2 πΌπ (β) (1−π 2 ) − π2 π 2 β4 Note that the error term on the right hand side is now of order π(β4 ). Neglecting the π(β4 ) error term, we obtain the new approximation to the value of the integral as πΌ = πΌ1 (β) = πΌπ (πβ)−π2 πΌπ (β) (31) (1−π 2 ) We note that this value is obtained by suitably using the values of the integral obtained with step lengths β and πβ, 0 < π < 1 this computed result is of order, π(β4 ), which is higher than the 28 1 order of the trapezium rule, which is of π(β2 ). For π = 2, that is, computations are done with β step lengths β and 2, the formula (31) simplifies to πΌπ (1) (β) = β 2 1 4 1 1− 4 πΌπ ( )− πΌπ (β) β 2 4πΌπ ( )−πΌπ (β) = (32) 3 β β β In practical applications, we normally use the sequence of step lengths β, 2, 22 , 23 ,... β β Suppose, the integral is computed using the step lengths β, 2, 22 . Using the results obtained with β β the step lengths 2, 22 , we get β πΌπ (1) (2) β = 1 β πΌπ ( )− πΌπ ( ) 4 4 2 1 1− β β 4πΌπ ( )−πΌπ ( ) 4 2 3 = 4 (33) β Both the results πΌπ (1) (β) and πΌπ (1) (2) are of order, π(β4 ). Now, we can eliminate the π(β4 ) terms of these two results to obtain a result of next higher order π(β6 ). 1 1 The multiplicative factor is now (2)4 = 16. The formula becomes πΌπ (2) (β) ≈ β 2 1 16 1 1− 16 πΌπ (1) ( )− πΌπ (1) (β) = β 2 16πΌπ (1) ( )−πΌπ (1) (β) 15 (34) Therefore, we obtain the Romberg extrapolation procedure for the composite trapezoid rule as πΌπ (π) (β) ≈ β 2 4π −1 4π πΌπ (π−1) ( )−πΌπ (π−1) (β) , π = 1,2,3, … (35) where πΌπ (0) (β) = πΌπ (β) The computed result is of order π(β2π+2 ). A general expression for Romberg integration can also be written as I οI I k , j ο½ I k ο1, j ο«1 ο« k ο1, j ο«k1ο1 k ο1, j , k ο³ 2 (36) 4 ο1 The index k represents the order of extrapolation. For example, k ο½ 1 represents the values obtained from the regular trapezoidal rule, k ο½ 2 represents the values obtained using the true error ο¨ ο© estimate as O h 2 , etc. The index j represents the more and less accurate estimate of the integral. The value of an integral with a j ο« 1 index is more accurate than the value of the integral with a j index. β β The extrapolations using three step lengths, β, 2, 22 are given in the table below 29 Table 2.2.1 Step lengths β Value of πΌ Value of πΌ Value of πΌ π(β2 ) π(β4 ) π(β6 ) πΌπ (β) πΌπ β 2 β πΌπ ( ) 2 β 4 πΌπ (4) (1) (β) = β 4πΌπ (2) − πΌπ (β) 3 β 4πΌπ ( ) − πΌπ (β) (1) β 2 πΌπ ( ) = 2 3 πΌπ (2) (β) = β 16πΌπ (1) (2) − πΌπ (1) (β) 15 β Note that the most accurate values are the values at the end of each column 2.2.2. Romberg method for the Simpson’s 1/3 rule We can apply the same procedure as in trapezium rule to obtain the Romberg’s extrapolation procedure for the Simpson’s 1/3 rule. Let πΌ denote the exact value of the integral and πΌπ denote the value obtained by the composite Simpson’s 1/3 rule. The error, πΌ − πΌπ , in the composite Simpson’s 1/3 rule in computing the integral is given by πΌ − πΌπ = π1 β4 + π2 β6 + π3 β8 + β― Or πΌ = πΌπ + π1 β4 + π2 β6 + π3 β8 + β― (36) As in trapezium rule, to illustrate the extrapolation procedure first consider two error terms πΌ = πΌπ + π1 β4 + π2 β6 (37) Let πΌ be evaluated using two step lengths β and πβ, 0< π < 1 Let these values be denoted by πΌπ (β) and πΌπ (πβ). The error equation become πΌ = πΌπ (β) + π1 β4 + π2 β6 (38) πΌ = πΌπ (πβ) + π1 π 4 β4 + π2 π 6 β6 (39 30 From (38), we obtain πΌ − πΌπ (β) = π1 β4 + π2 β6 (40) From (39), we obtain πΌ − πΌπ (πβ) = π1 π 4 β4 + π2 π 6 β6 (41) Multiply (40) by π 4 to obtain π 4 (πΌ − πΌπ (β)) = π1 π 4 β4 + π2 π 4 β6 (42) Eliminating π1 π 4 β4 from (41) and (42), we obtain (1 − π 4 )πΌ − πΌπ (πβ) + π 4 πΌπ (β) = π2 π 4 β6 (π 2 − 1) Note that the error term on the right hand side is now of order π(β6 ). Solving for πΌ, we obtain πΌπ (πβ) − π 4 πΌπ (β) π2 π 4 6 πΌ= − β (1 − π 4 ) 1 + π2 Neglecting the π(β6 ) error term, we obtain the new approximation to the value of the integral as πΌ ≈ πΌπ (1) (β) = πΌπ (πβ)−π4 πΌπ (β) (43) (1−π 4 ) Again we note that this value is obtained by suitably using the values of the integral obtained with step lengths β an πβ, 0 < π < 1. this computed result is of order, π(β6 ), which is higher than the order of the Simpson’s 1/3 rule, which is of π(β4 ). 1 β For π = 2, that is, computations are done with step lengths β and 2, the formula (43) simplifies to πΌπ (1) (β) ≈ β 1 πΌπ ( )−( )4 πΌπ (β) 2 2 1 (1−( )4 2 = β 2 16πΌπ ( )−πΌπ (β) 16−1 = β 2 16πΌπ ( )−πΌπ (β) (44) 15 β β β In practical applications, we normal use the sequence of step lengths β , 2, 22 , 23 , … β β Suppose, the integral is computed using the step lengths β , 2, 22 . Using the results obtained with β β step lengths 2, 22 , we get πΌπ (1) β 1 β β β β β πΌπ ( ) − ( )4 πΌπ ( ) 16πΌπ ( ) − πΌπ ( ) 16πΌπ ( ) − πΌπ ( ) β 4 2 2 4 2 4 2 ( )≈ = = 1 4 2 16 − 1 15 (1 − (2) Both the results β πΌπ (1) (β) and πΌπ (1) ( ) are of order π(β6 ). Now, we can eliminate π(β6 ) terms of these two re2 1 sults to obtain a result of next higher order π(β8 ). The multiplicative factor is now 0(2)6. The formula becomes 31 πΌπ (2) (β) ≈ β 1 πΌπ (1) (2) − (2)6 πΌπ (1) (β) 1 (1 − (2)6 = β 64πΌπ (1) (2) − πΌπ (1) (β) 64 − 1 = β 64πΌπ (1) (2) − πΌπ (1) (β) 63 Therefore, we obtain the Romberg extrapolation procedure for the composite Simpson’s 1/3 rule as πΌπ (π) (β) ≈ 4π+1 πΌπ (π−1) β ( )−πΌπ (π−1) (β) 2 4π+1 −1 , π = 1,2, … WhereπΌπ (0) (β) = πΌπ (β). The computed result is of order π(β2π+4 ) β The extrapolations using three step lengths β, 2, Step β 22 , are given below. Table 2.2.2 Value of πΌ Value of Value of πΌ π(β6 ) πΌ π(β8 ) π(β4 ) Length β πΌπ (β) β 2 β πΌπ ( ) 2 β 4 β πΌπ ( ) 4 πΌπ (1) (β) = β 16πΌπ (2) − πΌπ (β) 15 πΌπ πΌπ (1) (2) (β) = β β 16πΌπ (4) − πΌπ (2) β ( )= 2 15 Remark: ο The computation continued till values are close to each other ο The most accurate values are at the end of each column. 1 2 Example 4 evaluate I = ∫0 ex dx with 4, 8 and 16 subintervals using i. Trapezoid rule ii. Simpson’s rule iii. Romberg method as improvement of trapezoid rule and Simpson’s rule iv. Error bound Solution: i. Done in example 1 ii. Done in example 2 32 β 64πΌπ (1) (2) − πΌπ (1) (β) 63 iii. The approximation using the trapezium rule to the integral with various values of the step lengths were obtained 1 1 h = 4 , N = 4: I = 1.490679; h = 8 , N = 8 : I = 1.469712; 1 h = 16 , = 16 : I = 1.46442 Using Romberg method for the trapezium rule we have 1 I (1) (4) = 1 I (1) (8) = 1 I (2) ( ) 4 = 1 8 1 4 4I( )-I( ) 1 16 4(1.469712)-1.490679 = 3 1 8 4I( )-I( ) 4(1.46442)-1.469712 = 3 1 8 3 1 4 16I(1) ( )-I(1) ( ) 15 3 = = 1.462723 = 1.462656 16(1.462656)-1.462723 15 = 1.462652 The approximation using the Simpson’s 1/3 rule to the integral with various values of the step size were obtained in example 2 as follows 1 1 h = 4 , n = 2N = 4 : I = 1.463711; h = 8 , n = 2N = 8: I = 1.462409 1 h = 16 , n = 2N = 16 : I = 1.462656 Now using Romberg methods for Simpson’s 1/3 rule we have 1 8 1 4 I (1) (4) = 16I( )-I( ) 1 I (1) (8) 16I( )-I( ) 1 1 = I (2) (4) = 1 16 16(1.462409)-1.463711 = 15 1 8 15 1 8 = 1 4 64I(1) ( )-I(1) ( ) 63 15 16(1.462656)-1.462409 15 = = 1.462322 = 1.462672 64(1.462672)-1.462322 63 = 1.462678 To find error bound 2 2 f(x) = ex Implies f '' (x) = (2 + 4x 2 )ex ; f 4 (x) = (12 + 48x 2 + 16x 4 )ex 2 max |f '' (x)| = max (12 + 48x 2 + 16x 4 )ex = 76e a≤x≤b 0≤x≤1 Now using error bound on the trapezium rule 2 max |f '' (x)| = max (2 + 4x 2 )ex = 6e 0≤x≤1 0≤x≤1 Therefore |E1 (f, x)| ≤ 1 |E1 (f, x)| ≤ 12N2 |E1 (f, x)| ≤ 12N2 1 (b-a) 3 12N2 M2 where M2 = max |f '' (x)| a≤x≤b 6e = 0.084946 for N = 4 6e = 0.021236 for N = 8 33 2 1 |E1 (f, x)| ≤ 12N2 6e = 0.005309 for N = 16 Error bound for the Simpson’s 1/3 rule is given by |E2 (f, x)| ≤ (b-a) 5 2880N4 M4 where M4 = max |f 4 (x)| and Nh = a≤x≤b 1 Now |E2 (f, x)| ≤ 2880(24 ) 76e = 0.004483 |E2 (f, x)| ≤ |E2 (f, x)| ≤ 1 2880(84 ) 2 for n = 2N = 4 76e = 0.000280 for n = 2N = 8 76e = 0.0000175 for n = 2N = 16 2880(44 ) 1 b-a Table 4.3 Romberg method for trapezium rule Step length 1 4 1 8 Value of I O(h2 ) Value of I O(h4 ) Value of I O(h6 ) 1.490679 1.462723 1.469712 1.462656 1.462672 1 16 1.46442 Table 4.4 Romberg method for Simpson’s rule Value of I O(h4 ) Step length 1 4 1 8 Value of I O(h6 ) 1.463711 1.462322 1.462409 1.462672 1 16 Value of I O(h8 ) 1.462656 34 1.462678 Example 5 The vertical distance in meters covered by a rocket from t ο½ 8 to t ο½ 30 seconds is given by ο¦ οΆ 140000 ο© οΉ x ο½ ο² ο§ο§ 2000 ln οͺ ο 9.8t ο·ο·dt οΊ ο«140000 ο 2100t ο» οΈ 8ο¨ 30 Use the trapezoidal rule and Romberg’s rule to find the distance covered. Use the 1, 2, 4, and 8 sub intervals (segments) Solution Table 4.5 n From Table 2.6, the needed rule are I 1,1 ο½ 11868 1 2 4 8 Trapezoidal Rule 11868 11266 11113 11074 values from the trapezoidal I 1, 2 ο½ 11266 I 1,3 ο½ 11113 I 1, 4 ο½ 11074 where the above four values correspond to using 1, 2, 4 and 8 segment trapezoidal rule, respectively. To get the first order extrapolation values, I 2,1 ο½ I 1, 2 ο« ο½ 11266 ο« I 1, 2 ο I 1,1 3 11266 ο 11868 3 ο½ 11065 Similarly I 2, 2 ο½ I 1,3 ο« 35 I 1,3 ο I 1, 2 3 ο½ 11113 ο« 11113 ο 11266 3 ο½ 11062 I 2,3 ο½ I1, 4 ο« ο½ 11074 ο« I1, 4 ο I1,3 3 11074 ο 11113 3 ο½ 11061 For the second order extrapolation values, I 3,1 ο½ I 2, 2 ο« ο½ 11062 ο« I 2, 2 ο I 2,1 15 11062 ο 11065 15 ο½ 11062 Similarly I 3, 2 ο½ I 2,3 ο« ο½ 11061 ο« I 2,3 ο I 2, 2 15 11061 ο 11062 15 ο½ 11061 For the third order extrapolation values, I 4,1 ο½ I 3, 2 ο« ο½ 11061 ο« I 3, 2 ο I 3,1 63 11061 ο 11062 ο½ 11061m 63 36 Table 2.2.3 shows these increasingly correct values in a tree graph. Table 3 Improved estimates of the value of an integral using Romberg integration. First Order 1-segment 2-segment 4-segment 8-segment Second Order Third Order 11868 11266 11065 68 11113 11062 68 11074 11061 868 11062 868 11061 868 37 11061 868 3. SUMMARY π Numerical integration approximate a definite integral ∫π π(π₯)ππ₯ with different methods, The approximation of the integral by trapezoid rule gives error, π πΈ1 = − 12 β3 π ′′ (π) = − (π−π) 12 β2 π ′′ (π) Where π is number of subintervals. And πβ = π − π , π < π < π, π ′′ (π) is the largest value of the second order derivative of the function on [π, π]. Error bound of trapezoid is given by: |πΈ1 | ≤ π−π 12 β2 |π ′′ (π)| π Error of Simpson’s 1/3 rule in approximating ∫π π(π₯)ππ₯ is given by: π−π πΈ2 = − 180 β4 π 4 (π), where, πβ = π − π and π multiple of two, π < π < π and π 4 (π) is the largest value of the fourth order derivative of the function on [π, π]. Error bound of Simpson’s 1/3 is given by: |πΈ2 | ≤ π−π 180 β4 |π 4 (π)| π Error of Simpson’s 3/8 rule in approximating ∫π π(π₯)ππ₯ is given by: π πΈ2 = − 80 β5 π 4 (π), where, πβ = π − π and π multiple of three, π < π < π and π 4 (π) is the largest value of the fourth order derivative of the function on [π, π]. The bound on the error of Simpson’s 3/8 rule is given by: |πΈ2 | ≤ π 80 β5 |π 4 (π)| 1 The Romberg extrapolation procedure for the composite trapezoid rules for π = 2 is given by: β πΌπ (π) (β) ≈ 4π πΌπ (π−1) ( )−πΌπ (π−1) (β) 2 4π −1 , π = 1,2,3, … where πΌπ (0) (β) = πΌπ (β) 38 (35) 1 The Romberg extrapolation procedure for the composite Simpson’s 1/3 rule for π = 2 is given by: πΌπ (π) (β) ≈ 4π+1 πΌπ (π−1) β ( )−πΌπ (π−1) (β) 2 4π+1 −1 , π = 1,2, … where πΌπ (0) (β) = πΌπ (β). The error of Romberg method is the modification of the other errors hence with high accuracy than others. On the successive application of Romberg method on trapezoid rule and Simpson’s rule primitive errors are modified by order two that is, error order has the form, β(2π+2) .From this seminar we get that analytical non-integrable, or impossible functions can be approximated using numerical methods. From these approximation formulas Romberg method is better than trapezoid rule and Simpson’s rule in general. Simpson method is better than trapezoid rule in most cases but for periodic functions trapezoid rule is better than Simpson’s rule. 39 4. REFERENCE Atkinson, L.V., P.J., Harley, and J.D., Hudson, 1988. Numerical methods with fortran 77, Addison-wesley publishing company. Chapra, S.C., and R.P. Canale, 1988, numerical methods for engineers, McGraw.Hill companies. De, P.K., 2006, Computer based numerical methods and statistical techniques, Cbs publishers and distributors. Faires, J.D and R.L. Burden, 2002. Numerical methods volume I, Brooks Cole Grewal, B.S., 1991. Numerical methods in engineering and science, Khan Publishers. Kress, R., 1997. Graduate Texts in mathematics, Springer- Verlag, New York publishers. Iyengar, S.R.K., and R.K., 2009, Numerical methods, New age international publishers. Kaw, A., February 9, 2016, http://numericalmethods.eng.usf.edu. Ralston, A. and P. Rabinowitz, 1978, First course in numerical analysis, Dover publishers. 40