Using Chebyshev's polynomials for solving Fredholm integral equations of the second kind

From MaRDI portal





The paper addresses the problem of computing the solution of nonlinear integral Fredholm equations of second kind, \[ \phi(x) = f(x) + \lambda \int_a^b\mathcal{K}(x,t)\mathcal{H}(\phi)(t)\mbox{d}\, t,\qquad a\le x \le b, \] where \(\mathcal{H}\) stands for the Nemytskii operator. The approach the authors have opted for in this paper is based on the Newton iterative method of type \[ \phi_{n+1} = \phi_n -(\mathcal{T}'(\phi_n))^{-1}\mathcal{T}(\phi_n),\qquad n\ge 0, \] which gives rise to a continuous/infinite-dimensional solution, instead of numerical discretizations/finite-dimensional solutions. The Newton method should be applied to the equality \(\mathcal{F}(\phi)=0\) where \[ \mathcal{F}(\phi)(x) =\phi(x) - f(x) - \lambda \int_a^b \mathcal{K}(x,t)\mathcal{H}(\phi)(t)\mbox{d}\, t,\qquad a\le x \le b, \] whose theoretical solution is \(\phi\). However this method has a well-known drawback: the computation of the inverse of the derivative of an operator \((\mathcal{T}'(\phi_n))^{-1}\). In the present case, the situation is even worse since such an operator would be an integral operator. If the kernel \(\mathcal{K}\) is separable (and with certain regularity) some results are known. The authors focus on the following problem: finding the solution of a quite similar equation \(\mathcal{G}(\phi)=0\), where \[ \mathcal{G}(\phi)(x) =\phi(x) - f(x) - \lambda \int_a^b p_m(x,t)\mathcal{H}(\phi)(t)\mbox{d}\, t,\qquad a\le x \le b. \] Here \(p_m\) stands for a Chebyshev polynomial of degree \(m\). The theoretical solution is denoted by \(\phi^*\). The approach used leads to a separable kernel due to the nature of \( p_m \); therefore, previous methods can be adapted. It turns out that the polynomial \( p_m \) can approximate \(\mathcal{K}\) as closely as desired by increasing its degree. Moreover, the Runge-Kutta phenomenon is avoided thanks to the use of Chebyshev collocation points. All of this enables the authors to establish the main (and highly relevant) contribution of the paper. Specifically, they prove that the iterative Newton method for solving \(\mathcal{G}(\phi) = 0\) is convergent under very weak regularity assumptions on \(\mathcal{H}\), namely, only Lipschitz continuity is required. In addition, they show that the better the approximation of \( p_m \) to \(\mathcal{K}\), the closer the approximation \(\phi^*\) is to the true solution \(\phi\), thus allowing an approximation of \(\phi\) as accurate as desired. Finally, the authors provide an algorithm for the practical implementation of the theoretical method, along with several illustrative examples that demonstrate the superior performance of the proposed approach.











This page was built for publication: Using Chebyshev's polynomials for solving Fredholm integral equations of the second kind

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7021437)