A finite difference method for a class of two-point boundary value problems over a semi-infinite range (Q1081285)

We examine the three-point finite difference discretization of the two- point boundary value problem: \(-y''+f(x)y=g(x)\), \(0\leq x<\infty\), \(y(0)=y(\infty)=0\). Our purpose here is to study the resulting infinite tridiagonal linear system. We first show that, under suitable assumptions, the infinite linear system possesses a unique solution provided inf f(x)\(=\epsilon >0\). We show that the \(\ell_{\infty}\)-norm of the discretization error is bounded by \(Ch^ 2\) where h is the step- size and C is a constant independent of h. We assume that \(| g(x)| \leq Me^{-\alpha x}\), for suitable positive constants M and \(\alpha\). We then obtain a bound for the numerical solution \(\tilde y_ n\), at \(x_ n\), and the bound involves only h, \(\epsilon\), M and \(\alpha\). From this bound we conclude that \(\tilde y_ n\to 0\) as \(n\to \infty\). An interesting application of this bound is that we can obtain an a priori estimate for n so that for this n the numerical solution \(\tilde y_ n\) at \(x_ n\) is ''almost zero'' in the sense that \(| \tilde y_ n| \leq \delta\) for a preassigned \(\delta\). In other words, this bound provides an a priori estimate for n for truncating the infinite linear system (or equivalently, for truncating the semi-infinite range at \(x_ n)\) so that \(\tilde y_ n\) is within the tolerance \(\delta\).