Documente Academic
Documente Profesional
Documente Cultură
Two Dimensional Block Pulse Functions and Application to Solve Volterra-Fredholm Integral Equations with Galerkin Method
E. Babolian Department of Mathematics Tarbiat Moallem University, Tehran, Iran A. Jafari Shaerlar Department of Mathematics, Science and Research Branch Islamic Azad University, Tehran, Iran jafari shaerlar@yahoo.com
Abstract In this paper an ecient numerical analytical method for a class of linear mixed integral equations is presented. The exact solution is represented in the form of series. Its approximate solution is obtained by Bivariate Block Pulse series and a new numerical approximate method is obtained. The Bivariate Block Pulse Galerkin method for solving the linear Volterra-Fredholm integral equations of the second kind is examined.
Keywords: Bivariate Block pulse functions, Two Dimensional, VolterraFredholm Integral Equation, Galerkin method 1. Introduction We will consider the approximate solution of the Volterra-Fredholm integral equation of mixed type u(x, y ) = f (x, y ) +
y a
(1)
where u(x, y ) is an unknown function, the known functions f (x, y ) and F (x, y, s, t, u) are dened on D := [a, b] and S R where S := {(x, y, s, t)|a x, s b, (t, y ) } respectively and is a closed subset of R2 .
764
These equations arise in the theory of parabolic initial-boundary value problems (Fourier problems) and various physical, mechanical and biological problems. Moreover of the spatio-temporal development of an epidemic. For details see [1-3,7]. Actually few approximate methods for (1) are known. For the linear case some projection methods for numerical treatment of (1) are given in [4-6,8]. 2. Two dimensional block pulse functions Block pulse functions have been studied and applied extensively in the recent years as a basic set of functions for signal characterizations in systems science and control. This set of functions was rst introduced to electrical engineers by Harmuth in 1969. All these studies and applications show that block pulse functions may have denite advantages for problems involving integrals and derivatives due to their clearness in expressions and their simplicity in formulations. Block pulse functions are a set of orthogonal functions with piecewise constant values and are usually applied as a useful tool in the analysis, synthesis, identication and other problems of control and system science. We discussed the block pulse function which have single variable. These discussions can also be extended to functions which have two variables. To distinguish them, we usually call the block pulse functions containing one variable as one-dimensional(1D) block pulse functions and those containing two variables as two-dimensional(2D) block pulse functions. Similar to the 1D case, a set of 2D block pulse functions i1 ,i2 (t1 , t2 )(i1 = 1, 2, ..., m1 , i2 = 1, 2, ..., m2 ) are dened in the region of t1 (0, T1 ] and t2 (0, T2 ] as: 1 0 (i1 1)h1 t1 < i1 h1 , (i2 1)h2 t2 < i2 h2 otherwise
i1 ,i2 (t1 , t2 ) =
T1 T2 , h2 = m . where m1 , m2 are arbitrary positive integers, and h1 = m 1 2 Similar to the 1D case, the 2D block pulse functions are disjoint, i.e.
i1 ,i2 (t1 , t2 ) 0
i1 = j1 , i2 = j2 otherwise
h1 h2 0
i1 = j1 , i2 = j2 otherwise
765
in the region of t1 (0, T1 ] and t2 (0, T2 ], where i1 , j1 = 1, 2, ..., m1 , i2 , j2 = 1, 2, ..., m2 . Also similar to the 1D case, the 2D block pulse function set is complete when m1 and m2 approach innity. We can also expand a two-variable function f (t1 , t2 ) into its block pulse series:
m1 m2
f (t1 , t2 )
i1 =1 i2 =1
such that the mean square error between f (t1 , t2 ) and its block pulse series fm1 ,m2 (t1 , t2 ) is minimal: 1 = T1 T2
T1 0 0 T2
(f (t1 , t2 )
m1
m2
i1 =1 i2 =1
we can determine the values of the block pulse coecients. From d 2 = dfi1 ,i2 T1 T2
T1 0 0 T2 m1 m2
(f (t1 , t2 )
j1 =1 j2 =1
therefore the necessary condition of minimizing the mean square error yields the block pulse coecient fi1 ,i2 (i1 = 1, 2, ...., m1 , i2 = 1, 2, ...., m2): fi1 ,i2 = = 1 h1 h2
T1 0 i1 h1 (i1 1)h1 0 T2
1 h1 h2
i1 ,i2 (t1 , t2 ) 0
i1 = j1 , i2 = j2 otherwise
h1 h2 0
i1 = j1 , i2 = j2 otherwise
where i1 , j1 = 1, ...., m1 , i2 , j2 = 1, ...., m2 . hence we have d2 = dfi1 ,i2 dfj1,j2 or d2 = dfi1 ,i2 dfj1 ,j2
2 m1 m2 2 hh T1 T2 1 2
i1 = j1 , i2 = j2 otherwise i1 = j1 , i2 = j2 otherwise
holds for i1 , j1 = 1, ...., m1 , i2 , j2 = 1, ...., m2 , it is easy to verify the determinants for k = 1, ...., m1 , s = 1, ...., m2 :
Dk,s =
. . . . . .
. . . . . .
. . . . . . . . . . . .
Ai,j =
d2 dfki df1j
. . .
d2 dfki df2j
. . .
d2 dfki dfkj
. . .
where in Dk,s values of diagonal is ( d2 d2 d2 d2 d2 d2 d2 d2 , , ..., , , , ..., , ..., , ..., ) 2 2 2 2 2 2 2 2 df11 df21 dfk dfk df1 dfks s 1 df12 df22 2
2
so with denition dfi ,id dfj ,j we have Dk,s > 0. 1 2 1 2 This is the sucient condition for to be minimum the mean square error
767
under the block pulse coecients fi1 ,i2 i1 = 1, 2, ...., m1 , i2 = 1, 2, ...., m2.
3. Application of two dimensional block pulse functions to solve Volterra-Fredholm integral equations Consider the following linear integral equation of mixed type: u(x, y ) = f (x, y ) +
y c b a
(2)
where u(x, y ) is an unknown function, the known functions f (x, y ) and k (x, y, s, t) are dened on D := [a, b] [c, d] and S := {(x, y, s, t)|a x, s b, c t y d} respectively. Without loss of generality, we assume that on eq.(2) a = c = 0 and b = 1. It will become clear that the following analysis can be readily extended to every a, b, c, d . So, we will consider the following class of linear VolterraFredholm integral equation: u(x, y ) = f (x, y ) +
y 0 0 1
(3)
where the solution of this equation can be expressed by block pulse series:
m1 m2
u(x, y )
i=1 j =1
(4)
where 0 x, y 1 and m1 , m2 are positive integers. The sequence i,j is an orthogonal basis in L2 (D ) and constants ci,j are chosen such that , i,j = 0, i = 1...m1 , j = 1...m2 and the deviation function is dened in the following way (x, y ) = u(x, y ) then we get
T1 0 0 T2 T1 0 T2 0 0 T2 y 0 0 1 y 0 0 1
(5)
(6)
768
Putting block pulse series in the above equality and taking into account the orthogonality of the system, we obtain the following system of algebraic equations cl,m where Ni,j,l,m =
T1 0 0 T2 m1 m2 i=1 j =1
(8)
y 0 0
fl,m =
T1 0
(9)
and l, i = 1, ..., m1 , m, j = 1, ..., m2 . Hence from (8), we obtain (h1 h2 I M )C = GT where M is (m1 m2 ) (m1 m2 ) matrix and G, C are respectively 1 (m1 m2 ) and (m1 m2 ) 1 matrixes, and dened as
M=
Mi,j =
, C= and G= f1,1 f1,2 . . . f1,m2 . . . fm1 ,1 fm1 ,2 . . . fm1 ,m2 c1,1 c1,2 . . . c1,m2 . . . cm1 ,1 cm1 ,2 . . . cm1 ,m2
T
769
The relation (h1 h2 I M )C = GT corresponds to a system of (m1 m2 ) (m1 m2 ) linear algebraic equations with unknown Block pulse coecients, thus the unknown coecients ci,j (i = 1...m1 , j = 1...m2 ) can be computed from this fundamental linear Volterra-Fredholm integral equation in the Bivariate Block pulse series. 4. Numerical experiments In this section, we will give some examples to use the method given in the paper. All computations are performed by the Maple 9.5. Example 1.(From[10]) we consider the linear Volterra-Fredholm integral equation
y 1 1 4 1 xys2 t2 u(s, t)dsdt xy xy 5 + 15 16 0 0 2 where u(x, y ) = x + xy is the true solution. So by applying the method given in the paper with m1 = m2 = 2 we have
u(x, y ) = x2 + xy
(h1 h2 I M )1 =
11393/11392 1/11392 1/11392 1/11392 31/11392 11423/11392 31/11392 31/11392 3/11392 3/11392 11395/11392 3/11392 93/11392 93/11392 93/11392 11485/11392
the solution of this problem with Galerkin method U (x, y ) = 2125583 11647859 33712007 62341337 1,1 (x, y )+ 1,2 (x, y )+ 2,1 (x, y )+ 2,2 (x, y ) 58327040 174981120 174981120 174981120
References
[1] B.G. Pachpatte, On mixed Volterra-Fredholm integral equations, Indian J. Pure. Appl. Math., 17 (1986) 488-496. [2] O. Diekmann, Thesholds and traveling for the geographical spread of infection, J. Math. Biol., 6 (1978) 109-130.
770
[3] H.R. Thieme, A model for the spatial spread of an epidemic, J. Math. Biol., 4 (1977) 337-351. [4] J.P. Kauthen, Continuous time collocation method for Volterra-Fredholm integral equations, Numer. Math, 56 (1989) 409-424. [5] L. Hacia, On approximate solution for integral equations of mixed type, ZAMM Z. Angew. Math. Mech., 76 (1996) 415-416. [6] L. Hacia, Projection for integral equation in epidemic, J. Math. Model. Anal., 7(2) (2002) 229-240. [7] L. Hacia, On approximate solution of the Fourier problems, Demonst. Math., 12 (1979) 913-922. [8] H. Guoqiang, Z.liqing, Asyptotic error expansion for trapezoidal Nystrom method of linear Volterra-Fredholm integral equations, J. comput. Appl. Math., 51(3) (1994) 339-348. [9] M. Sezer, M. Kaynak, (Chebyshev series solution of linear dierential equations), Int. J. Math. Educ. Sci. Tech. 27 (1996) 607-618. [10] M. Hadizadeh, M. Asgary, (An ecient numerical approximation for the linear class of mixed integral equations), J. Appl. Math. comp. 167(2005) 1090-1100. Received: November, 2011