Numerical piecewise approximate solution of Fredholm integro-differential equations by the Tau method S. Mohammad Hosseini a, * , S. Shahmorad b a Department of Mathematics, Tarbiat Modarres University, P.O. Box 14115-175, Tehran, Iran b Department of Applied Mathematics, Tabriz University, Tabriz, Iran Received 1 October 2003; received in revised form 1 December 2004; accepted 8 February 2005 Available online 19 March 2005 Abstract A general form of numerical piecewise approximate solution of linear integro-differential equations of Fredholm type is discussed. It is formulated for using the operational Tau method to convert the differential part of a given integro-differential equation, or IDE for short, to its matrix representation. This formulation of the Tau method can be useful for such problems over long intervals and also can be used as a good and simple alternative algorithm for other piecewise approximations such as splines or collocation. A Tau error estimator is also adapted for piecewise application of the Tau method. Some numerical examples are con- sidered to demonstrate the implementation and general effect of application of this (segmented) piecewise Chebyshev Tau method. Ó 2005 Elsevier Inc. All rights reserved. Keywords: Piecewise approximate; Piecewise Tau method; Integro-differential equations; Segmented Tau method 1. Introduction We extend (see [1–4]), the operational approach of the Tau method (see [5]), to the numerical solution of general form of linear Fredholm and Volterra integro-differential equations and obtain accurate results, except for the problems defined over a long interval. 0307-904X/$ - see front matter Ó 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.apm.2005.02.003 * Corresponding author. E-mail address: [email protected](S.M. Hosseini). www.elsevier.com/locate/apm Applied Mathematical Modelling 29 (2005) 1005–1021
17
Embed
Numerical piecewise approximate solution of Fredholm integro-differential equations by the Tau method
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Numerical piecewise approximate solution of Fredholmintegro-differential equations by the Tau method
S. Mohammad Hosseini a,*, S. Shahmorad b
a Department of Mathematics, Tarbiat Modarres University, P.O. Box 14115-175, Tehran, Iranb Department of Applied Mathematics, Tabriz University, Tabriz, Iran
Received 1 October 2003; received in revised form 1 December 2004; accepted 8 February 2005Available online 19 March 2005
Abstract
A general form of numerical piecewise approximate solution of linear integro-differential equations ofFredholm type is discussed. It is formulated for using the operational Tau method to convert the differentialpart of a given integro-differential equation, or IDE for short, to its matrix representation. This formulationof the Tau method can be useful for such problems over long intervals and also can be used as a good andsimple alternative algorithm for other piecewise approximations such as splines or collocation. A Tau errorestimator is also adapted for piecewise application of the Tau method. Some numerical examples are con-sidered to demonstrate the implementation and general effect of application of this (segmented) piecewiseChebyshev Tau method.� 2005 Elsevier Inc. All rights reserved.
Keywords: Piecewise approximate; Piecewise Tau method; Integro-differential equations; Segmented Tau method
1. Introduction
We extend (see [1–4]), the operational approach of the Tau method (see [5]), to the numericalsolution of general form of linear Fredholm and Volterra integro-differential equations and obtainaccurate results, except for the problems defined over a long interval.
0307-904X/$ - see front matter � 2005 Elsevier Inc. All rights reserved.doi:10.1016/j.apm.2005.02.003
In this method, we only replace the operator matrix representation for the differential part ofthe equation using the operational Tau method. However, a full matrix formulation of the Taumethod [3] can also be implemented, for which the details appear in another paper. The remainderof the paper is organized as follows:
Section 2, is devoted to introducing the problem and some preliminary results concerning theTau method. In Section 3, details of the formulation of the new method are explained. In Section4, the way of implementing the method is demonstrated. Section 5 provides some numericalresults used to clarify the efficiency of the method.
2. The problem and some preliminary results of the Tau method
Let us consider the general linear Fredholm integro-differential equation
DyðxÞ � kZ b
akðx; tÞyðtÞdt ¼ f ðxÞ; x 2 ½a; b� ð2:1Þ
with m independent conditions
Xmk¼1
cð1Þjk yðk�1ÞðaÞ þ cð2Þjk y
ðk�1ÞðbÞh i
¼ cj; j ¼ 1; . . . ; m; ð2:2Þ
where cj is constant and m is the order of the differential operator D with polynomial coefficientspi(x)
D ¼Xmr¼0
prðxÞdr
dxr; prðxÞ ¼
Xarj¼0
prjxj ¼ p
rX ; ð2:3Þ
where ar is the degree of pr(x), pr¼ ðpr0; pr1; . . . ; prar ; 0; 0; . . .Þ and X ¼ ð1; x; . . . ÞT.
k(x,t) is a bivariate polynomial or its bivariate Tau polynomial approximation of degree pairs,say, (k1,k2). Similarly, f(x) is a polynomial or a suitable polynomial approximation of it.
The operational Tau method is generally based on three simple matrices
g ¼
0 0 0 0 � � �1 0 0 0 � � �0 2 0 0 � � �... ..
. ... ..
. . ..
0BBBB@
1CCCCA; l ¼
0 1 0 0 � � �0 0 1 0 � � �0 0 0 1 � � �... ..
. ... ..
. . ..
0BBBB@
1CCCCA; i ¼
0 1 0 0 � � �0 0 1
20 � � �
0 0 0 13
� � �
..
. ... ..
. ... . .
.
0BBBB@
1CCCCA:
In this paper, we have only used g, l for the differential part of (2.1). See [5] for more applica-tion of these three matrices. The matrices g, l have the following effects:
Lemma 2.1. If yn(x) = anX, with an = (a0,a1, . . . , an, 0,0, . . .), then
dynðxÞdx
¼ angX ; xynðxÞ ¼ a
nlX :
Let V = {vi(x):i = 0,1, . . .} be a polynomial base given by V = VX where V is a nonsingular lowertriangular matrix and V �1 its inverse.
Theorem 2.2. For any linear differential operator D defined by (2.3) and any series y(x) = bV, b =(b0,b1, . . .) or y(x) = aX, a = (a0,a1, . . .) we have
DyðxÞ ¼ aYx
X ¼ bYv
V ; ð2:4Þ
where
Yx
¼Xmi¼0
gipiðlÞ;Yv
¼ VYx
V �1: ð2:5Þ
3. Formulation of the piecewise method
Let
a ¼ x0 < x1 < � � � < xn ¼ b
be a partition of [a,b], and
½xi�m; xðiþ1Þ�m�;
i = 0,1, . . ., p � 1, with p ¼ n
m, be subintervals of [a,b], on which we want to approximate the solu-tion y(x) of (2.1) and (2.2), as pieces of polynomials yi(x) of degree m with unknown coefficientsyi0, . . .,yim, i.e.,
yðxÞ �
Pmj¼0
y0jv�0jðxÞ; x 2 ½x0; xm�;
Pmj¼0
y1jv�1jðxÞ; x 2 ½xm; x2m�;
..
. ...
Pmj¼0
yp�1;jv�p�1jðxÞ; x 2 ½xn�m; xn�;
8>>>>>>>>>>><>>>>>>>>>>>:
ð3:1Þ
where the basis functions v�ijðxÞ are shifted to suitable polynomials (such as Chebyshev or Legen-dre) of degree j in [xi·m,x(i+1)·m] for j = 0,1, . . .. We set
V �i ¼ v�i0ðxÞ; v�i1ðxÞ; . . .
� �T ð3:2Þ
so that expansion of the functions k(x,t), f(x) and y(x) for x 2 [xi·m, x(i+1)·m], t 2 [tj·m, t(j+1)·m]and i, j = 0,1, . . ., p � 1, can be expressed as
in each subinterval of the form [xi·m,x(i+1)·m]. Since there are p subintervals of this form, the num-ber of required unknowns and corresponding linear equations will be (m + 1) · p. Note that thenumber of equations that can be obtained from continuity conditions (3.17) is equal to (p � 1) · m.We also obtain m equations from supplementary conditions (3.21). So, the total number of equa-tions that can be obtained from the continuity and supplementary conditions will be p · m. There-fore, we need
ðmþ 1Þ � p � p � m ¼ ðmþ 1� mÞ � p;
other linear equations that must be obtained from (3.14).Therefore, we have the following three steps:
Step 1. For i,j = 0, . . ., p � 1, l1 = 0, . . ., m and l2 = 0, . . ., m � m, set
ðGijÞl1;l2 ¼Qvj
�kAij
!l1;l2
; if i ¼ j;
�kðAjiÞl1;l2 ; if i 6¼ j;
8>><>>:
ðFjÞl2 ¼ ðf
jÞl2 :
Step 2. For i = 0, . . ., p � 1, l1 = 0, . . ., m and l2 = m � m + 1 . . .,m, set
ðGi�1;i�1Þl1;l2 ¼ ðCi�1Þl1;l2 ;
ðGi;i�1Þl1;l2 ¼ ðCiÞl1;l2 ;
ðFi�1
Þl2 ¼ 0;
where Ci�1 and Ci are matrices corresponding to continuity conditions.Step 3. For l1 = 0, . . ., m and l2 = 1, . . ., m, set
by a Gaussian elimination method in which, to avoid direct inversion of diagonal blocks, we haveapplied the QR factorization, providing the possibility of taking advantage of numerical stabilityof QR method.
4. How to implement this piecewise method
To show the structure of final system (3.22), we give block matrices Gij and other related vectorsfor
See Table 1 for further numerical results of this example.
5. Error estimation
When the solution is not known, the need for an error estimator presents itself as a vital com-ponent for any given algorithm. To this end, one can follow the same lines of a similar discussionin [1] and extend the method for piecewise approximation. Let us call em(x) = y(x) � ym(x) the‘‘mth-order Tau error function’’ which is to be approximated by the same method of Tau.For an integro-differential equation ID(y) = 0 with conditions B(y) = b0, the Tau problemID(ym) = Hm(x) with B(ym(x)) = b0 is associated which is defined by the same integro-differentialoperator ID; Hm(x) is a polynomial of degree m chosen for the exact solution ym(x) to be a poly-nomial of a prescribed degree. Subtracting the equations related to the Tau problem from those ofthe exact problem one obtains the Tau error problem ID(em(x)) = �Hm(x), with B(em(x)) = 0.
Like the original problem we proceed in the same manner to estimate em(x) with the Taumethod. We seek a polynomial approximation em,n(x) for n > m which provides the Tau errorestimation and is denoted by ‘‘Est.Err’’ in the following Table 6. The terms ‘‘exact’’, ‘‘Tau’’,and ‘‘Ext.Err’’ denote exact solution, Tau approximate solution, and difference between thosevalues, respectively.
In this section, numerical results for some examples are given to clarify the accuracy of the pre-sented method (see Tables 1–5). Table 5, shows the absolute errors of an unconditionally stablemethod for second-order Fredholm integro-differential equations reported by Garey et al. (see[6]), applied to Examples 1, 2, providing a good comparison with the numerical results, obtainedby the method presented in this paper (see Tables 1,2). We also provide Tau error estimations inTable 6.
Example 1 [6, Example 1]
TableMaxim
m
3510
TableMaxim
m
3510
TableMaxim
m
3510
y00ðxÞ ¼ 9yðxÞ þ e�15 � 1
3þZ 5
0
yðtÞdt; x 2 ½0; 5�;
yð0Þ ¼ 1;
y0ð0Þ ¼ �3:
The exact solution is y(x) = e�3x.For numerical results see Table 1.
The exact solution is yðxÞ ¼ cosðxÞ.For numerical results see Table 4.
6.1. Examples for the Tau estimator
To clarify the efficiency and use of the Tau estimator, we consider again the test Example 4. Thenumerical results for (m = 5,n = 7) are given in Table 6. The results confirm that Ext.err andEst.err are in good agreement.
Example 5. We consider a test problem in which the coefficients are not polynomial and againconfirms that the Tau method is capable of dealing with such problems without having to facevery much computational effort. Again, well agreement between exact and estimated errors isillustrated in Table 7.
TableThe e
s
[�1.0�1.�0.�0.�0.�0.�0.
[�0.5�0.�0.�0.�0.�0.0.00
[0.00,0.000.100.200.300.400.50
[0.50,0.500.600.700.800.901.00
exy 00ðxÞ þ cosðxÞy 0ðxÞ þ sinðxÞyðxÞ þZ 1
�1
eððxþ1ÞtÞyðtÞdt
¼ ðcosðxÞ þ sinðxÞ þ exÞex þ 2sinhðxþ 2Þ
xþ 2; x 2 ½�1; 1�;
yð�1Þ þ yð1Þ ¼ eþ 1=e;
yð�1Þ � y0ð�1Þ þ yð1Þ ¼ e:
The exact solution is y(x) = exp(x).
7rror-estimation ‘‘Est.err’’ for Example 5, (m = 9, n = 11)
The formulation of the piecewise Tau method, or segmented Tau, was given. We demonstratedthe implementation and accuracy of the method for some examples with different degrees and seg-mentations. Although in the classical Tau method one is required to use an initial approximation ofthe nonpolynomial coefficient functions appearing in the equation, this should not cause anyone tohesitate about applying the Taumethod and taking advantage of its features that have already beenintroduced through different papers during last 30 years. However, in some examples that we haveconsidered, the coefficients and the kernels are nonpolynomials which have been replaced by easilyobtained interpolating polynomials based on zeros of the Chebyshev polynomials. Comparisonswith some other well known methods have often been reported in some related articles, but herewe would particularly like to comment on the application of the classical collocation method. In[9] it has been shown that collocation approximations of any given degree can be simulated throughthe Tau method by using a special perturbation term. In doing so, the collocation method acquiresthe permanence property of the Taumethod and consequently the number of arithmetic operationsrequired to compute collocation approximations can be reduced considerably.
We finally have adapted a Tau estimator to estimate the error of approximations on each seg-ment. The results given in Tables 6 and 7 confirm the efficiency of the introduced Tau errorestimator.
References
[1] S.M. Hosseini, S. Shahmorad, Numerical solution of a class of integro-differential equations with the Tau methodwith an error estimation, Appl. Math. Comput. 136 (2003) 559–570.
[2] S.M. Hosseini, S. Shahmorad, Tau numerical solution of Fredholm integro-differential equations with arbitrarypolynomial bases, Appl. Math. Model 27 (2003) 145–154.
[3] S.M. Hosseini, S. Shahmorad, A matrix formulation of the tau for the Fredholm and Volterra linear integro-differential equations, Korean J. Comput. Appl. Math. 9 (2) (2002) 497–507.
[4] S. Shahmorad, Numerical solution of a class of integro-differential equations by the Tau method, Ph.D Thesis,Tarbiat Modarres University, Tehran, 2002.
[5] E.L. Ortiz, H. Samara, An operational approach to the Tau method for the numerical solution of nonlineardifferential equations, Computing 27 (1981) 15–25.
[7] A.M. Wazwaz, A First Course in Integral Equations, World Scientific, River Edge, NJ, 1997.[8] L.M. Delves, J.L. Mohamed, Computational Methods for Integral Equations, Cambridge University Press,
Cambridge, 1985.[9] M.K. EL-Daou, E.L. Ortiz, A recursive formulation of collocation in terms of canonical polynomials, Computing