radial basis functions

In this investigation, a numerical method for solving nonlinear two-dimensional Volterra integral equations is presented. This method uses radial basis functions (RBFs) constructed on scattered points as a basis in the discrete collocation method. Therefore, the method does not need any background mesh or cell structure of the domain. All the integrals that appear in this method are approximated by the composite Gauss-Legendre integration formula. This method transforms the source problem into a system of nonlinear algebraic equations. Error analysis is presented for this method. Finally, numerical examples are included to show the validity and efficiency of this technique.

The stable Gaussian radial basis function (RBF) interpolation is applied to solve the time and space-fractional Schrödinger equation (TSFSE) in one and two-dimensional cases. In this regard, the fractional derivatives of stable Gaussian radial basis function interpolants are obtained. By a method of lines, the computations of the TSFSE are converted to a coupled system of Caputo fractional ODEs. To solve the resulting system of ODEs, a high-order finite difference method is proposed, and the computations are reduced to a coupled system of nonlinear algebraic equations, in each time step. Numerical illustrations are performed to certify the ability and accuracy of the new method. Some comparisons are made with the results in other literature.

‎In this paper‎, ‎we propose a radial basis function partition of unity (RBF-PU) method to solve sparce optimal control problem governed by the elliptic equation‎.‏ The objective function, in addition to the usual quadratic expressions, also includes an ‎L1-norm‎‎‎ of the control function to compute its spatio sparsity. ‎Meshless methods based on RBF approximation are widely used for solving PDE problems but PDE-constrained optimization problems have been barely solved by RBF methods‎. RBF methods have the benefits of being versatile in terms of geometry, simple to use in higher dimensions, and also having the ability to give spectral convergence. ‎In spite of these advantages‎, ‎when globally RBF collocation methods are used‎, ‎the interpolation matrix becomes dens and computational costs grow with increasing size of data set‎. ‎Thus‎, ‎for overcome on these problemes RBF-PU method will be proposed‎. ‎RBF‎ -‎PU methods reduce the computational effort‎. ‎The aim of this paper is to solve the first-order optimality conditions related to original problem‎.‎‎‎

In the present paper, the relatively new method of Radial Basis Function-Generated Finite Difference (RBF-FD) is used to solve a class of Partial Differential Equations (PDEs) with Dirichlet and Robin boundary conditions. For this approximation, Polyharmonic Splines (PHS) are used alongside Polynomials. This combination has many benefits. On the other hand, Polyharmonic Splines have no shape parameter and therefore relieve us of the hassle of calculating the optimal shape parameter. As the first problem, a two-dimensional Poisson equation with the Dirichlet boundary condition is investigated in various domains. Then, an elliptic PDE with the Robin boundary condition is solved by the proposed method. The results of numerical studies indicate the excellent efficiency, accuracy and high speed of the method, while for these studies, very fluctuating and special test functions have been used.

In this article the Caputo time- and Riesz space-fractional Fokker-Planck equation (TSFFPE) is solved by the stable Gaussian radial basis function (RBF) method. By a spatial discretization and using the Riesz fractional derivative of the stable Gaussian radial basis function interpolants computed in [23], the computations of TSFFPE reduced to a system of fractional ODEs. A high order finite difference method is presented for this system of ODEs, and the computations are converted to a system of linear or nonlinear algebraic equations, in each time step. In the nonlinear case, these systems can be easily solved by the Newton iterative method. Numerical illustrations are performed to confirm the accuracy and efficiency of the presented method. Some comparisons are made with the results in other literature.

In this paper, the application of the Fifth-order Meshless Local Petrov-Galerkin Method in solving the linear partial differential-algebraic equations (PDAEs) was surveyed. The Gaussian quadrature points in the domain and on the boundary were determined as centers of local sub-domains. By governing the local weak form in each sub-domain, the compactly supported radial basis functions (CS-RBFs) approximation was used as the trial function and the Heaviside step function was considered as the test function. The proposed method was successfully utilized for solving linear PDAEs and the numerical results were obtained and compared with the exact solution to investigate the accuracy of the proposed method. The sensitivity to different parameters was analyzed and a comparison with the other methods was done.

The radial basis functions (RBFs) meshless method has high accuracy for the interpolation of scattered data in high dimensions. Most of the RBFs depend on a parameter, called shape parameter which plays a significant role to specify the accuracy of the RBF method. In this paper, we present three algorithms to choose the optimal value of the shape parameter. These are based on Rippa’s theory to remove data points from the data set and results obtained by Fasshauer and Zhang for the iterative approximate moving least square (AMLS) method. Numerical experiments confirm stable solutions with high accuracy compared to other methods.

This paper aims to advance the radial basis function method for solving space-time variable-order fractional partial differential equations. The fractional derivatives for time and space are considered in the Coimbra and the Riemann-Liouville sense, respectively. First, the time-variable fractional derivative is discretized through a finite difference approach. Then, the space-variable fractional derivative is approximated by radial basis functions. Also, we advance the Rippa algorithm to obtain a good value for the shape parameter of the radial basis functions. Results obtained from numerical experiments have been compared to the analytical solutions, which indicate high accuracy and efficiency for the proposed scheme.

The radial basis functions (RBFs) methods were first developed by Kansa to approximate partial differential equations (PDEs). The RBFs method is being truly meshfree becomes quite appealing, owing to the presence of distance function, straight-forward implementation, and ease of programming in higher dimensions. Another considerable advantage is the presence of a tunable free shape parameter, contained in most of the RBFs that control the accuracy of the RBFs method. Here, the solution of the two-dimensional system of nonlinear partial differential equations is examined numerically by a Global Radial Basis Functions Collocation Method (GRBFCM). It can work on a set of random or uniform nodes with no need for element connectivity of input data. For the timedependent partial differential equations, a system of ordinary differential equations (ODEs) is derived from this scheme. Like some other numerical methods, a comparison between numerical results with analytical solutions is implemented confirming the efficiency, accuracy, and simple performance of the suggested method.

In this paper, we first present a nonlinear structural model for pricing mortgage-backed securities. These derivatives are considered to be the primary cause of the 2008 financial crisis that was raised in the United States. We focus our work on pass-through mortgages, which pay both the principal and interest to the investors. We begin our work by introducing the factors that affect the market of mortgage-backed securities. Then, by applying some assumptions and conditions to the parameters of the initial model, and without the loss of generality, we show that this model can be greatly simplified. We focus our attention on how the change in interest rates can affect the value of mortgage-backed securities. Various numerical methods can be used to solve the reduced model that is achieved. We ‎adapt the mesh-less method of radial basis functions to solve the reduced model. The numerical results indicate that the method that we have used can capture the market trends in a specific interval.

‎‎‎‎‎In this paper, we present a numerical technique to deal with the one-dimensional forward-backward heat equations. First, the physical domain is divided into two non-overlapping subdomains resulting in two separate forward and backward subproblems, and then a meshless method based on multiquadric radial basis functions is employed to treat the spatial variables in each subproblem using the Kansa’s method. We use a time discretization scheme to approximate the time derivative by the forward and backward finite difference formulas. In order to have adequate boundary conditions for each subproblem, an initial approximate solution is assumed on the interface boundary, and the solution is improved by solving the subproblems in an iterative way. The numerical results show that the proposed method is very useful and computationally efficient in comparison with the previous works.

In this study, a radial basis functions (RBFs) method for solving nonlinear timeand space-fractional Fokker-Planck equation is presented. The time-fractional derivative is of the Caputo type, and the space-fractional derivatives are considered in the sense of Caputo or Riemann-Liouville. The Caputo and Riemann-Liouville fractional derivatives of RBFs are computed and utilized for approximating the spatial fractional derivatives of the unknown function. Also, in each time step, the time-fractional derivative is approximated by the high order formulas introduced in [6], and then a collocation method is applied. The centers of RBFs are chosen as suitable collocation points. Thus, in each time step, the computations of fractional Fokker-Planck equation are reduced to a nonlinear system of algebraic equations. Several numerical examples are included to demonstrate the applicability, accuracy, and stability of the method. Numerical experiments show that the experimental order of convergence is 4 − α where α is the order of time derivative.

In this investigation, we solve the Caputo's fractional parabolic partial integro-differential equations (FPPI-DEs) by Gaussian-radial basis functions (G-RBFs) method. The main idea for solving these equations is based on RBF which also provides approaches to higher dimensional spaces.In the suggested method, FPPI-DEs are reduced to nonlinear algebraic systems. We propose to apply the collocation scheme using G-RBFs to approximate the solutions of FPPI-DEs. Error analysis of the proposed method is investigated. Numerical examples are provided to show the convenience of the numerical schemes based on the G-RBFs. The results reveal that the method is very efficient and convenient for solving such equations.

In this paper, we propose a meshless regularization technique for solving an optimal shape design problem (OSD) which consists of constructing the optimal configuration of a conducting body subject to given boundary conditions to minimize a certain objective function. This problem also can be seen as the problem of building a support for a membrane such that its deflection is as close as possible to 1 in the subset D of the domain. We propose a numerical technique based on the combination of the method of fundamental solutions and application of the Tikhonov’s regularization method to obtain stable solution. Numerical experiments while solving several test examples are included to show the applicability of the proposed method for obtaining the satisfactory results.

This paper presents an approximate method to solve a class of fractional partial differential equations (FPDEs). First, we introduce   radial basis functions (RBFs) combined with wavelets.  Next, we obtain fractional integral operator (FIO) of wavelets-radial basis functions (W-RBFs) directly.  In the next step, the W-RBFs and their FIO  are used to transform the problem under consideration into a  system of algebraic equations, which can be simply solved to achieve the solution of the problem.   Finally, some numerical examples are presented to illustrate the efficiency and accuracy of the method.

We propose an efficient mesh-less method for functional integral equations. Its convergence analysis has been provided. It is tested via a few numerical experiments which show the efficiency and applicability of the proposed method. Attractive numerical results have been obtained.

In this paper‎, ‎we decide to select the best center nodes‎ ‎of radial basis functions by applying the Multiple Criteria Decision‎ ‎Making (MCDM) techniques‎. ‎Two methods based on radial basis‎ ‎functions to approximate the solution of partial differential‎ ‎equation by using collocation method are applied‎. ‎The first is based‎ ‎on the Kansa's approach‎, ‎and the second is based on the Hermite‎ ‎interpolation‎. ‎In addition‎, ‎by choosing five sets of center nodes‎: ‎Uniform grid‎, ‎Cartesian‎, ‎Chebyshev‎, ‎Legendre and‎ ‎Legendre-Gauss-Lobato (LGL) as alternatives and achieving the error‎, ‎the condition number of interpolation matrix and memory time as‎ ‎criteria‎, ‎rating of cases with the help of PROMETHEE technique is‎ ‎obtained‎. ‎In the end‎, ‎the best center nodes and method is selected‎ ‎according to the rankings‎. ‎This ranking shows that Hermite‎ ‎interpolation by using non-uniform nodes as center nodes is more‎ ‎suitable than Kansa's approach with each center node.

We propose a new approach for solving nonlinear Klein–Gordon and sine-Gordon equations based on radial basis function-pseudospectralmethod (RBF-PS). The proposed numerical method is based on quasiinterpolation of radial basis function differentiation matrices for thediscretization of spatial derivatives combined with Runge–Kutta time stepping method in order to deal with the temporal part of the problem.The method does not require any linearization technique; in addition, a new technique is introduced to force approximations to satisfy exactlythe boundary conditions. The introduced scheme is tested for a number of one- and two-dimensional nonlinear problems. Numerical results andcomparisons with reported results in the literature are given to validate the presented method, and the reported results show the applicabilityand versatility of the proposed method.