NUMERICAL SOLUTION OF NONLINEAR SINE - ResearchGate

By submitting this extended abstract to IMC44 I confirm that (i) I and
any other coauthor(s) are responsible for its content and its originality;
(ii) any possible coauthors agreed to its submission to IMC44.
NUMERICAL SOLUTION OF NONLINEAR
SINE-GORDON EQUATION WITH LOCAL RBF-BASED
FINITE DIFFERENCE COLLOCATION METHOD
YAQUB AZARI AND GHOLAMREZA GARMANJANI AND HAMED RABIEI1
1
School of Mathematics, Iran University of Science and Technology
Narmak, Tehran 16844-13114, Iran
[email protected]
[email protected]
[email protected]
Abstract. This paper presents the local radial basis function
based on finite difference (LRBF-FD) for the sine-Gordon equation. Some advantages of the proposed method are that this method
is mesh free unlike finite difference (FD) and finite element (FE)
methods, and its coefficient matrix is sparse and well-conditioned
as compared with the global RBF collocation method (GRBF).
Numerical results show that the LRBF-FD method has good accuracy as compared with GRBF.
1. Introduction
We consider the generalized one-dimensional sine-Gordon equation
∂ 2u
∂u
∂ 2u
+
ОІ
=
− f (x, t, u), a < x < b, t ≥ 0,
∂t2
∂t
∂x2
with the initial conditions
(1.1)
u(x, 0) = h(x), ut (x, 0) = g(x), a ≤ x ≤ b,
2010 Mathematics Subject Classification. 65M22.
Key words and phrases. Sine-Gordon Equation, Local RBF-based finite difference (LRBF-FD), Global RBF collocation.
1
2
AZARI AND GARMANJANI AND RABIEI
and with boundary conditions
u(a, t) = p(t), u(b, t) = q(t), t ≥ 0.
(1.2)
In this paper, we develop the local RBF based finite difference (LRBFFD) for the solution of Eq. (??) and the results are compared with
the solutions obtained by global RBF collocation (GRBF). In the local
RBF method, the function derivatives are approximated at each node
in terms of the function value on a set of points in local stencils. This
makes flexibility in choosing these stencils [?].
1.1. RBF interpolation. RBFs are widely used for scattered data
interpolation. Given (xj , uj ) j = 1, ..., N with xj в€€ Rd , uj в€€ R, the
standard interpolation problem is to find an interpolant as
N
О»j П†( x в€’ xj ),
uЛњ(x) =
(1.3)
j=1
where П† is a radial basis function and . denotes the Euclidean distance
between points x, xj , and λj , 1 ≤ j ≤ N are unknown RBF coefficients.
The following conditions are imposed on the approximation uЛњ(x),
uЛњ(xj ) = u(xj ), j = 1, . . . , N,
leading to a system of equations which can be written in matrix form
as AО» = u,
where A = (П†( xi в€’ xj )), i, j = 1, . . . , N,
T
T
.
u = u(x1 ) . . . u(xN )
, О» = О»1 . . . О»N
The existence and uniqueness of solution of the above system are guaranteed when П† is a positive definite function.
1.2. The Local RBF method based on finite difference(LRBFFD). Let L is differential operator. The finite difference method approximates derivatives of a function u(x) at point x = xi by
n
Lu(xi ) в€ј
=
w(i,j) u(xj )
(1.4)
j=1
where w(i,j) are called the finite difference(FD) weights at node points
xi . The unknown coefficients w(i,j) are obtained by using polynomial interpolation or Taylor series[?]. In the RBF-FD approach the
weights of the FD formulas are obtained by using the RBF interpolation technique. To derive these weights the interpolant (??) is
n
considered in a Lagrangian form as s(x) =
j=1 П€j (x)u(xj ), where
П€j (xk ) = Оґjk , k = 1, ..., N .
Using this method, we determine the approximated derivative for each
LRBF-FD METHOD FOR SINE-GORDON EQUATION
3
interior node in a local support region. Support region for interior node
is defined as
support region (xi ) = {xj | 0 ≤ xj − xi ≤ R} i = 2, ..., N − 1.
To derive LRBF-FD formula, for example at the node x1 , we approximate the differential operator using the Lagrangian from of the RBF
interpolant i.e.,
n
Lu(x1 ) ≈ Ls(x1 ) =
LП€j (x1 )u(xj ),
(1.5)
j=1
where n is the number of the nodes in support region of x1 . From (??)
and (??) we get w(1,i) = LП€i (x1 ). The number of points in support
regions can be variable and this property makes the method flexible.
Once the LRBF-FD method is applied to discretize the spatial derivatives in the governing Equation (??) and approximating RBF weights
for L ≡ ∂x∂ 2 we obtain at any interior node xi , i = 2, ..., N − 1,
n
un+1
в€’ 2uni + unв€’1
un+1
в€’ unв€’1
i
i
i
i
+ОІ
= Оё(
w(i,j) un+1 (xj ))
2
(dt)
2(dt)
j=1
(1.6)
n
w(i,j) un (xj )) в€’ f (xi , tn , uni ),
+(1 в€’ Оё)(
j=1
where n is the number of nodes in support region. Equation (??) is
written for each interior node. Substitruting the function value u1 , uN
from the boundary conditions (??) leads to a system of equations which
can be written in matrix form as Au = B, where u is the vector of the
unknown function values at whole of the interior nodes at t = tn+1 and
B is the vector of the right hand of the Equation(??) along with the
boundary terms.Note that matrix A is sparse and well-conditioned and
can hence be effectively inverted.
2. Main results
We present some numerical results to illustrate the applicability and
justify the accuracy and efficiency of our presented method. Some
RBFs contain the parameter c that is shape parameter which plays an
important role in the application of RBF methods. There is experimental evidence that shows the accuracy of RBF methods significantly
depend on the amount of this shape parameter, but choosing of the
optimal shape parameter for any equation is open problem yet [?]. In
4
AZARI AND GARMANJANI AND RABIEI
Table 1. The errors of resulting solution of LRBF-FD
and GRBF methods with N = 40, dt = 0.001 ,and Оё = 12
t
1
2
3
4
5
R = 1h
1.77e-3
1.39e-3
8.09e-4
3.74e-4
1.28e-4
LRBF в€’ F D
R = 3h
8.05e-5
6.98e-5
4.74e-5
2.86e-5
1.60e-5
GRBF
R = 5h
6.48e-6
4.37e-6
2.92e-6
1.76e-6
9.85e-7
8.17e-4
4.12e-4
2.55e-4
1.92e-4
1.34e-4
order to analyze the error of the method, we consider Lв€ћ error as
Lв€ћ = u(x, t) в€’ uN (x, t)
в€ћ
= max |u(xi , t) в€’ uN (xi , t)|,
1≤i≤N
where uN is numerical solution.
Example 2.1. We consider Equation (??) with a = 0, b = 1, ОІ = 1
and f (x, t, u) = 2 sin u+ПЂ 2 exp(в€’t) cos(ПЂx)в€’2 sin(exp(в€’t)(1в€’cos(ПЂx))),
u(x, 0) = 1 − cos(πx) 0 ≤ x ≤ 1
with the initial conditions
ut (x, 0) = −1 + cos(πx) 0 ≤ x ≤ 1
The exact solution is derived in [?] as u(x, t) = exp(в€’t)(1 в€’ cos(ПЂx)).
The computed solutions are compared with the exact solution at equidistant points.The Lв€ћ errors are tabulated in Table (??). It can be seen
that numerical results obtained from LRBF-FD method is in good
agreement with exact solution in contrast to the GRBF method.
References
1. M. Cui, Fourth-Order Compact Scheme for the One-Dimensional Sine-Gordon
Equation, Numer. Methods Partial Differential Equations. 25 (2009), no. 3, 685–
711.
2. B. Fornberg, Calculation of weights in finite difference formulas, SIAM Reviews,
40(1998)685-691.
3. j. Rashidinia, Y. Azari, G. Garmanjani, Numerical solution of the nonlinear
Fredholm integral equations of the second kind by radial basis functions, The 43re
Iranian Mathematics Conferencs, University of Tabriz, Tabriz, Iran. (2012).
4. C. Shu, H. Ding, K.S. Yeo, Local radial basis function-based differential quadrature method and its application to solve two-dimensional incompressible NavierStokes equations, Comput. Methods Appl. Mech. Engrg. 192(2003)941954.