On the Numerical Solution of the 3-Dimensional Fractional Diffusion Equation in the shifted Grunwald estimate form I. I. Gorial Department of Mathematics, Address Ibn Al-Haithem College Education, Baghdad University, Email [email protected] ICM 2012, 11-14 March, Al Ain u(x,y,zR,t) =3(x,y,t), for x0 x xR , y0 y yR and 0 t Τ ABSTRACT In this paper, a numerical solution of the 3-dimensional fractional diffusion equation has been presented. The algorithm for the numerical solution for this equation is based on implicit finite difference method. The consistency, unconditional stability, and convergence of the fractional order numerical method are described. The numerical method has been applied to solve a practical numerical example and the results have been compared with exact solution. The results were presented in tables using the MathCAD 12 software package when it is needed. The implicit finite difference method appeared to be effective and reliable in solving the 3dimensional fractional diffusion equation. Keywords: Fractional derivative, implicit Euler method, fractional diffusion equation, stability, convergence. The u x , u y , u z fractional derivatives by the shifted Grunwald estimate formulae [ 2, 7]: u ( x, y , z , t ) 1 x (x) i 1 g k 0 u ( x, y , z , t ) 1 y (y ) u ( x, y , z , t ) 1 z (z ) 1 INTRODUCTION Various fields of science and engineering deal with dynamical systems that can be described by fractional partial differential equations, for example, systems biology, chemistry and biochemistry applications due to anomalous diffusion effects in constrained environments. However, effective numerical methods and numerical analysis for fractional partial differential equations are still in their infancy, [1, 2, 3, 4, 5, 6, 7, 11, 13,14,15]. In this work, we consider the 3-dimensional fractional Diffusion equation: u ( x, y, z, t ) u ( x, y, z, t ) u ( x, y, z, t ) a ( x, y , z ) b( x, y, z ) t x y c ( x, y , z ) where a , b , c and are known functions of x, y and z, and 1 is a known function of y, z and t, 2 is a known function of x,z and t, 3 is a known function of x,y and t. , and are given fractional number. q is a known function of x, y ,z and t. u ( x, y , z , t ) q ( x, y , z , t ) z (1) subject to the initial condition u(x,y,z,0)= (x,y,z), for x0 x xR, y0 y yR and z0 zzR (2) and the boundary conditions u (x0,y,z,t)= 0, for y0 y yR , z0 z zR and 0 t Τ u(x,y0,z,t)= 0, for x0x xR, z0 z zR and 0 t Τ u(x,y,z0,t)= 0, for x0 x xR, y0 y yR and 0 t Τ (3) u(xR,y,z,t) = 1(y,z,t), for y0 y yR , z0 z zR and 0 t Τ u(x,yR,z,t) =2(x,z,t), for x0 x xR , z0 z zR and 0 t Τ ,k u isk11, j , f O(x) j 1 g k 0 ,k f 1 g k 0 ,k u is,j1 k 1, f O(y ) (4) u is,j1, f k 1 O(z ) In this paper, an implicit finite difference approximation for the 3-dimensional fractional diffusion equation is presented. The unconditional stability and convergence of the implicit finite difference approximation are analyzed and finally, we will present example to show the efficiency of our numerical method . 2 FINITE DIFFERENCE METHOD FOR SOLVING THE 3-DIMENSIONAL FRACTIONAL DIFFUSION EQUATION In this section, we use the finite difference method to solve the initial and boundary value problem 3-dimensional fractional diffusion equation (1)-(3). The finite difference method starts by dividing the x-interval [x0, xR] into n subintervals to get the grid points xi = x0 + ix, where x xR x0 n and i = 0,1,…,n. and we divide the y-interval [y0, yR] into m subintervals to get the grid points yj = y0+jy, where y y R y 0 m and j=0,1,…,m. also we divide the z-interval [z0, zR] into p subintervals to get the grid points zg =z0+fz, where z z R z 0 p and f = 0,1,…,p. Also, the t-interval [0,T ] is divided into M subintervals to get the grid points ts = st, s = 0,…, M, where t T M . Now, we evaluate eq. (1) at (x i, y j, ,zf ,t implicit Euler method to get s+1) and we use the 1 I. I. Gorial ui0, j , f i , j , f , i=0,…, n , j=0,…, m and f = 0,1,…,p. u ( xi , y j , z f , ts 1 ) u ( xi , y j , z f , ts ) t u ( xi , y j , z f , t s 1 ) b ( xi , y j , z f ) y a( xi , y j , z f ) c ( xi , y j , z f ) u ( xi , y j , z f , ts 1 ) u0s, j , f 0, j=0,…, m , f =0,1,…,p and s=1,…,M uis,0, f 0, i=0, … , n , f =0,1,…,p and s=1,…,M (5) uis, j ,0 0, i=0, …, n , j=0,…, m and s=1,…,M x u ( xi , y j , z f , t s 1 ) z q( xi , y j , z f , t s 1 ) Use fractional derivative of the shifted Grunwald estimate to the , , - the fractional derivative eq.(4), to reduce eq.(5) as in the following form: u is,j1, f u is, j , f i 1 1 a ( xi , y j , z f ) x t g k 0 ,k j 1 1 b ( xi , y j , z f ) y g 1 z g c ( xi , y j , z f ) u isk11, j , f k 0 f 1 k 0 u ,k ,k s 1 i , j k 1, f u is,j1, f k 1 q( xi , y j , z f , t s 1 ) uis,j1,f uis, j , f t ai , j , f x v i 1 g ,k uisk11, j , f k 0 ci , j , f z f 1 g k 0 ,k bi , j , f y j 1 g k 0 ,k uis,j1k 1, f u Rs , j , f sj , f , j=0, …, m, f =0,1,…,p and s=1,…,M uis, R, f is, f , i=0, …, n, f =0,1,…,p and s=1,…,M uis, j ,R is, j , i=0, …, n, j=0,…, m and s=1,…,M where i , j , f ( xi , y j , z f ), sj , f ( y j , z f , ts ) and is, j ( xi , y j , ts ) 3 STABILITY OF THE FINITE DIFFERENCE 3DIMENSIONAL FRACTIONAL DIFFUSION EQUATION We consider the following fractional partial difference operator: , x uis,j1, f ai , j , f uis,j1, f k 1 qis,j1,f i 1,..., n 1 , j 1,..., m 1, f 1,..., p 1 , s 0,..., M (6) u , y uis,j1, f bi , j , f u( xi , y j , z f , t s ), ai , j , f a( xi , y j , z f ), bi , j , f b( xi , y j , z f ), , z uis,j1, f ci , j , f ci , j , f c( xi , y j , z f ), qis,j1,f q( xi , y j , z f , t s 1 ), ( 1) ( k 1) g , k (1) k k! g ,k ( 1) k g , k (1) k , k=0,1,2,… , k=0,1,2,… The resulting equation can be implicitly solved for uis,j1, f ai , j , f ci , j , f t z t x i 1 g k 0 f 1 g k 0 ,k ,k uisk11, j , f bi , j , f t y s 1 i, j, f u j 1 g k 0 ,k uisk11, j , f t y t z j 1 g k 0 ,k f 1 g k 0 ,k u is,j1 k 1, f uis,j1, f k 1 (1 t , x t , y t , z )uis,j1,f uis, j , f qis,j1,f t to give -order uis,j1k 1, f uis,j1, f k 1 uis, j , f tqis,j1, f (8) eq.(8) may be written in form (1 t , x )(1 t , y )(1 t , z )uis,j1,f uis, j , f tqis,j1,f (9) Then (7) ETOU Also form the initial condition and boundary conditions one can get 2 ,k With these definitions, the implicit difference scheme (7) may be written in the following compact form: k! k! k 0 are O(y ) and O (z ) approximation of the and Grunwald shifted fractional derivatives term, respectively. ( 1)( k 1) k=0,1,2,…and , ( 1) ( k 1) i 1 g which is an O(x) approximation to the th fractional derivative. Similarly, the following fractional partial difference operators are defined. Where s i, j , f t x s 1 U R s s 1 Where E (1 t ,x ), (10) On the Numerical Solution of the 3-Dimensional Fractional T (1 t , y ), U ks1,1k23 [u1s,k11,k32 , u 2s ,k11 ,3k2 , , u ns 11,k31 ,k2 ]T , O (1 t ,z ), U ks1 ,k2 t Qks1,1k2 [u1s,k1 ,k2 q1s,k11,k2 t , u 2s,k1 ,k2 q2s,k11 ,k2 t ,, u ns1,k1 ,k2 qns11,k1 ,k2 t ]T and Ci , j Therefore the resulting matrix entries U [u1s,1,1 , u2s,1,1 , , uns1,1,1 , u1s, 2,1 , u2s, 2,1 , , uns1, 2,1 , , u1s,m1,1 , u2s,m1,1 , , uns1,m1,1 s j 1,, n 1 are defined by u1s,1, 2 , u2s,1,2 ,, uns1,1,2 , u1s, 2, 2 , u2s,2, 2 ,, uns1,2,2 ,, u1s,m1, 2 , u2s,m1, 2 ,, uns1,m1,2 Ci, j u1s,1, p1 , u2s,1, p1 , , uns1,1, p1 , u1s, 2, p1 , u 2s, 2, p 1 , , uns1, 2, p 1 , , u1s,m1, p1 , u s 2 ,m1, p 1 , , u s T n 1,m1, p 1 and the vector R ] s 1 is the forcing term . Hence, we obtain the fractional implicit scheme at time (1 t , x )uis,j1,f3 uis, j , f tqis,j1,f (1 t , y )uis,j ,2f 3 uis,j1,f3 1 i , k1 , k 2 g ,1 i , k1 , k 2 g , 2 i , k1 , k 2 g , 0 i , k1 , k 2 g , j 1 ji for for j i 1 for j i 1 for j i 1 where the coefficients t s 1 : i ,k 1 ,k 2 (11) (12) and (1 t , z )uis,j1,f uis,j ,2f 3 for i 1,2, , n 1 and (13) Thus, we require three steps to solve the third-dimensional fractional diffusion equation in one time step. t x ai ,k1 ,k2 To illustrate this matrix pattern, we list the corresponding equations for the rows i =1, 2 and n-1: 1,k1 ,k2 g ,2 u0s,k11 /,k32 (1 1,k1 ,k2 g ,1 )u1s,k11,/k32 1,k1 ,k2 g ,0 u2s,k11 /,k32 u1s,k1 ,k2 tq1s,k11,k2 2, k1 , k 2 g ,3u0s,k11 ,/k32 2, k1 , k 2 g , 2u1s,k11,/k32 (1 2, k1 , k 2 g ,1 )u2s,k11 ,/k32 2, k , k g ,0u3s,k1 ,/k3 u2s, k , k tq2s,k1 , k 1 2 1 2 1 2 1 2 Firstly: if (yj, zf ) is fixed we will obtain an intermediate solution uis,j1,f3 from (11). Second: if (xi, zf ) is fixed we will obtain an intermediate solution u is,j ,2f 3 from (12). Third: if (xi, yj ) is fixed we will obtain an intermediate solution from (13) using information compiled during Second step. n 1, k1 , k 2 g , nu0s,k11/,3k 2 n 1, k1 , k 2 g , 2uns 12/,3k1 , k 2 (1 n 1, k , k g ,1uns 11,/k3 , k n 1, k , k g ,0uns,k1/,3k uns 1, k , k tqns 11, k , k 1 2 1 1 2 1 2 1 2 1 2 According to the Greshgorin theorem [9], the eigenvalues of the matrix C lie in the union of the circles centered at ri Now, we must prove each one-dimensional implicit system defined by the linear difference eqs. (11), (12) and (13) is unconditionally stable for all 1 < , , < 2. Theorem 3.1. The implicit system defined by the linear difference eqs.(11),(12) and (13) is unconditionally stable for all 1 < , , < 2. 2 ci ,i with radius n c l 0 l i i ,l . Here we have ci ,i 1 i ,k1 ,k2 g ,1 1 i ,k1 ,k2 and Proof: At each grid point yk1, for k1 1,, m 1 , and zk2 , for k 2 1,, p 1 , the system of equation defined by (11). This system of equations may be written as n i 1 l 0 l i l 0 l i ri ci ,l i ,k1 ,k2 g ,i l 1 i ,k1 ,k2 i ,k1 ,k2 Ck1 ,k2 U ks1,1k23 U ks1 ,k2 t Qks1,1k2 With strict inequality holding true when is not an integer. This implies that the eigenvalue of the matrix C k k has a real-part larger where than 1, and therefore a magnitude larger than 1. Hence the spectral 1, 2 3 I. I. Gorial 1 radius of the matrix C k ,k is less than 1.This proves that the meth1 2 od is stable [10]. Ai , j And with the same method above; the results of equations system, defined by (11), can be defined as: 1 k1 , k 2 , f g ,1 k1 , k 2 , f g , 2 k1 , k 2 , f g , 0 k1 , k 2 , f g , j 1 ji for for j i 1 for j i 1 for j i 1 S k1 ,k2 U ks1,k22/ 3 U ks1,1k2/ 3 , where the coefficients where k ,k U ks1,1k 2 [u ks1,11,k2 , u ks1,12,k 2 ,, u ks1,1m 1,k 2 ]T , 1 U ks1,1k 2/ 3 [u ks1,11,/k32 , u ks1,12/, k32 ,, u ks1,1m/31,k 2 ]T , and S k ,k is the matrix of coefficients, and is the sum of a lower 1 2 triangular matrix and a super diagonal matrix at the grid point xk1 for k1 1,, n 1 and zk2 for k 2 1,, p 1 . Therefore the resulting matrix entries S i , j for i 1,2, , m 1 and j 1, , m 1 are defined by 1 k1 , j , k 2 g ,1 k1 , j , k 2 g , 2 k1 , j , k 2 g , 0 k1 , j , k 2 g , j 1 S i, j 1 2 1 1 k is less than 1, this proves that s2 / 3 k1 , k 2 tral radius of the matrix E Since U U s2 / 3 k1 , k 2 Ak1 ,k2 [u s 1 k1 , k 2 ,1 [u ,u s 1 k1 , k 2 , 2 s2 / 3 k1 , k 2 ,1 ,u is xk1 for , is less than 1, and hence the spec- is less than 1. spectral radius of the matrix and O are in the un- 1 T and in U n O 1 are less than 1. 0 in U 0 results in an error given by n ( ETO) n 0 s 1 T k1 , k 2 , m 1 , , u T ion of the Greschgorin disks for the matrices S k and O . Again the argument of Theorem 3.2 may be applied to show that the 0 at time t n ] , When the matrices E , s2 / 3 T k1 , k 2 , m 1 k1 1,, n 1 ] , and O yk2 for k 2 1,, m 1 . Therefore the resulting matrix entries Ai , j for i 1,2,, m 1 and j 1,, m 1 are defined by 1 T and O commute, E i.e. 1 , T 1 and commute, we have n E n T n O n 0 As the spectral radius of each matrix E than one, it follows that n , 4 1 1 Note that eq.(9) implies that an error , , u s2 / 3 k1 , k 2 , 2 E Similarly, the eigenvalues of the matrix where s 1 k1 , k 2 commute. value of the inverse matrix Now, resulting the system of equations defined by (12) is then defined by: U O E ,T theorem 3.1, it follows that every eigenvalue of the matrix E has a real-part larger than 1. Therefore, the magnitude of every eigen- the method is stable [10]. Ak1 , k2 U conditionally stable for 1< , , < 2 if the matrices E k are in the union of the Greschgorin disks for the matrices C k . Applying the argument of part larger than 1, and therefore a magnitude larger than 1. Hence s 1 / 3 k1 , k 2 Theorem 3.2. The implicit-Euler method, defined by (9), is un- The eigenvalues of the matrix for j i 1 So, and in the same way, eigenvalue of the matrix S k has a realthe spectral radius of the matrix S therefore a magnitude larger than 1. Hence the spectral radius of the matrix Ak1 is less than 1, and hence this system is also uncon- Proof. for j i 1 t y 2 t z ck1 ,k2 , f ji for j i 1 where the coefficients k , j ,k bk , j ,k f So, eigenvalue of the matrix Ak has a real-part larger than 1, and and for 2, E n 1 0, T , n T 1 and 0, O O n 1 is less 0 , as On the Numerical Solution of the 3-Dimensional Fractional where 0 denotes the zero matrix . Therefore, the stability of the implicit Euler method follows. c ( x, y , z ) 1..5u ( x, y, z, t ) q( x, y, z, t ) z1..5 With the coefficient function: 4 CONSISTENCY AND CONVERGENT OF THE FINITE DIFFERENCE 3-DIMENSIONAL FRACTIONAL DIFFUSION EQUATION To obtain the consistency of the 3-dimensional fractional diffusion equation, note that the time difference operator in (8) has alocal truncation error of order O ( t ), and the three space difference operators in (8) have local truncation errors of orders O(x), O(y ) and O ( z ) respectively. Similar to Lemma 2.1 in paper of Meerschaert et al., (2006), below we can in our paper, we have the results: f ( x, y, z ) x y f ( x, y, z ) (x y ) x y f ( x, y, z ) x z f ( x, y, z ) (x z ) x z f ( x, y, z ) y z f ( x, y, z ) (y z ) y z then f ( x, y, z ) x y z f ( x, y, z ) (x y z ) x y z a ( x, y , z ) (1.4) x 8 (1 x) 7 , 2 b( x, y, z ) (0.2) y 8 (1 y) 7 , c( x, y, z ) (2.5) z 7 (1 z ) 7 , and the source function: q( x, y, z, t ) x 7.4 (1 x) 7 yz 3e 3t x 2 y 7.2 (1 y) 7 z 3e 3t x1.5 y(1 y) 7 z 7 e 3t x 7.4 (1 x) 7 yz 3 e 3t x 2 y 7.2 (1 y) 7 z 3 e 3t x1.5 y(1 y) 7 z 7 e 3t 3x 2 yz 3 e 3t subject to the initial condition u (x,y,z,0) = x2yz3, 0 x 1, 0 < y < 1, 0 < z < 1 and the boundary conditions u (0,y,z,t) = 0, 0 < y < 1, 0 < z < 1, 0 t 0.025 u (x,0,z,t) = 0, 0 < x < 1, 0 < z < 1, 0 t 0.025 u (x,,y,0,t) = 0, 0 < x < 1, 0 < y < 1, 0 t 0.025 u (1,y,z,t) = yz3e-3t, 0 < y < 1, 0 < z < 1, 0 t 0.025 u (x,1,z,t) = x2z3e-3t, 0 < x < 1, 0 < z < 1, 0 t 0.025 u (x,,y,1,t) = x2ye-3t, 0 < x < 1, 0 < y < 1, 0 t 0.025 Note that the exact u(x,y,z,t) = x2yz3e-3t. solution to this problem is: Table1 and 2 show the numerical solution using the implicit finite difference approximation. From table 1 and 2, it can be seen that that good agreement between the numerical solution and exact solution. Which leads to the 3-dimensional fractional diffusion equation with order O(t ) O(x) O(y ) O(z ) . We show above that implicit Euler method is consistent and unconditionally stable, then by Laxs equivalence theorem, [12], it convergence at the rate O(x y z t ) . Table 1: The numerical solution of example using the finite difference method (x 0.2, y 0.2, z 0.2, t 0.0125) x = y =z t Numerical Solution Exact Solution |uex -uapprox.| 0.2 0.0125 6.169E-5 0.61644E -4 4.55573 E -8 0.4 0.0125 3.949 E-3 0.39452E -2 3.75567 E -6 0.6 0.0125 0.045 0.44939 E -1 6.12012 E -5 0.8 0.0125 0.253 0.25296 5.04363 E -4 0.2 0.0250 5.947 E-5 0.59376E-4 9.44171 E -8 0.4 0.0250 3.807 E-3 0.38000E -2 6.96268 E -6 0.6 0.0250 0.043 0.43285 E -1 2.84800 E -4 0.8 0.0250 0.244 0.24320 7.97612 E -4 5 NUMERICAL TEST EXAMPLE In this section, we implement the proposed method to solve a 3dimensional the fractional diffusion equation (1). Also, a comparison with numerical solution and exact solution, which is based on the implicit finite difference approximation of fractional derivative, is given. Example: In this example, we with 1.6 , 1.8 , 1.5 , of the form: consider (1) u ( x, y, z, t ) 1.6u ( x, y, z, t ) 1.8u ( x, y, z, t ) a( x, y, z ) b( x, y, z ) 1.6 t x y1.8 5 I. I. Gorial Table 2: The numerical solution of example using the finite difference method. (x 0.25, y 0.25, z 0.25, t 0.0125) x = y =z t NumericalSolution Exact solution |uex -uapprox.| 0.25 0.0125 2.353 E – 4 0.23515 E -3 1.45113E-7 0.50 0.0125 0.015 0.15050 E -1 4.99128 E-5 0.75 0.0125 0.172 0.17143 5.72087 E-4 0.25 0.0250 2.269 E – 4 0.22650E -3 4.00125 E-7 0.50 0.0250 0.015 0.14496 E -1 5.04008 E-4 0.75 0.0250 0.165 0.16512 1.18409 E-4 Table 3: Maximum error for the numerical solution of example using the finite difference method. x = y = z t Maximum Error 0.20 0.0125 0.0002848 0.25 0.0125 0.0001184 6 CONCLUSIONS In this paper (1) Numerical method for solving the 3-dimensional twosided fractional Diffusion equation has been described and demonstrated. (2) The implicit Euler method is proved to be unconditionally stable and converges. (3) Numerical examples indicate the convergence of the solution with exact results. REFERENCES Miller K. and Ross B. (1993) An Introduction to the Fractional Calculus and Fractional Differential Equations. Wiley and Sons, New York. Meerschaert M.M. and Tadjeran C. (2006) Finite difference ap proximations for two-sided space fractional partial differential equations. Appl. Numer. Math., Vol. 56, No. 1, PP. 80–90. Podlubny I. (1999) Fractional Differential Equations. Academic Press, New York. Roop J.P. (2005) Computational aspects of FEM approximation of fractional advection dispersion equations on bounded domains in R2. J. Comput. Appl. Math. Meerschaert M. .M. and Tadjeran C. (2007) A second-order accurate numerical method for the two-dimensional fractional diffusion equation . Journal of Computational Physics, Vol. 220, PP. 813–823. Lin R. and Liu F(2004) Analysis of Fractional- Order Numerical Method for the Fractional Relaxation Equation. J. comp. Appl. Math., Vol.91, PP. 198-210. 6 Meerschaert M.M. et al. (2006) Finite difference methods for two dimensional fractional dispersion equation. J. Comput. Phys., Vol. 211, PP. 249–261. Meerschaert M. .M. and Tadjeran C. (2004) Finite difference approximations for fractional advection-dispersion flow equations. J. Comput. Appl. Math., Vol.172, PP. 65–77. Isaacson E. and Keller H.B. (1966) Analysis of Numerical Methods”, Wiley, New York. Varga R. (1962) Matrix Iterative Analysis”, Prentice Hall, New Jersey. Liu Q. et al. (2009) Numerical simulation for the 3D seepage fllow with fractional derivatives in porous media. IMA Journal of Applied Mathematics,Vol.74 PP.201-229. Smith D.D. Numerical Solution of Parial Differential Equations: Finite Difference Methods. Oxford Applied Mathematics and Computing Science Series, Oxford University. Joaquín Q.M. and .Santos B.Y. (2011) An Explicit Difference Method for Solving Fractional Diffusion and Diffusion-Wave Equations in the Caputo Form. Journal of Computational and Nonlinear Dynamics,Vol..6. Yuste S.B. and Acedo L. (2005) On an Explicit Finite Difference Method for Fractional Diffusion Equations. SIAM. Soc.Ind.Appl.Math. J.Numer.Anal., 42, PP. 1862–1874. Morton K.W. and. Mayers D.F, (1994) Numerical Solution of Partial Differential Equations. Cambridge University Press, Cambridge, UK.
© Copyright 2024 Paperzz