An analysis of a uniformly accurate spline difference method

From MaRDI portal





The paper discusses finite difference techniques to solve the singular perturbation problems of the class \[ L_{\varepsilon}y_{\varepsilon}\equiv y_{\varepsilon}(x) + p(x)y_{\varepsilon}'(x) - d(x)y_{\varepsilon}(x) =f(x), x \in (0,1) \] \[ By_{\varepsilon}\equiv (y_{\varepsilon}(0),y_{\varepsilon}(1))=(\alpha_{0},\alpha_{1}) \] where \(\varepsilon \in (0,1]\) is a small parameter, and \(p\), \(d\), \(f\) are assumed to be sufficiently smooth. Under assumption that for \(x \in [0,1]\) the functions \(p\) and \(d\) satisfy \(p(x)\geq p >0\), \(d(x) \geq 0\), \(L_{\varepsilon}\) satisfies the maximum principle and the above problem has the unique solution \(y_{\varepsilon}(x)\). The authoresses are interested in finite difference techniques that satisfy the criterion known as uniform convergence or uniform accuracy, that is \(\|u^{h} - y_{\varepsilon} \|\geq Ch^{r}\) (\(r>0\) is the convergence order, \(u^{h}\) is the approximate solution, \(y_{\varepsilon}\) - the exact solution, \(C\) is generic constant independent of \(\varepsilon\) and \(h\)). It is known that using standard finite difference or finite element methods to solve the above problem arises some difficulties when \(\varepsilon\) is small; they may be overcome by fitting properly meshes. The authoresses propose another approach, leading to a new scheme, derived by using as the collocation equation the auxiliary problem \[ \widetilde {L}_{y} \equiv \varepsilon y(x) +\widetilde{p} y'(x) - \widetilde{d}y(x) = \widetilde{f},\qquad \widetilde{B}_{y}\equiv (y(0),y(1))=(\alpha_{0},\alpha_{1}), \] where \(\widetilde{p}, \widetilde{d}, \widetilde{f}\) are piecewise polynomial (in the proposed scheme -- piecewise constant) approximations to the functions \(p(x), d(x)\) and \(f(x)\). This new scheme is second-order uniformly and classically (for fixed \(\varepsilon\)) convergent. It is proved in the paper that the errors for this scheme at the grid points are bounded by \(Mh^{2}\) when \(d(x)\equiv 0\), where \(M\) is a constant independent of \(\varepsilon\) and stepsize \(h\). The results of some numerical experiments suggest that this is valid also for \(d(x)\neq 0\). The proposed scheme is also compared with two another second-order accuracy schemes -- the well-known El-Mistikawy and Werle scheme and its improvement due to the both authoresses [Publ. Inst. Math., Nouv. Ser. 54, 144-155 (1993; Zbl 0809.65086)].











This page was built for publication: An analysis of a uniformly accurate spline difference method

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4261196)