Generalized spline interpolation of functions with large gradients in boundary layers (Q2207518)

Boundary layer problems are a fundamental issue in applied mathematics and numerical analysis. One standard approach which comes from approximation theory uses piecewise polynomials -- splines -- as a scheme to model the sought solutions. A typical way to compute these splines is collocation. The collocation can take place at equally spaced or scattered data points along the boundary. One problem that is addressed in this article is that of possible steep gradients that occur along such a boundary layer. The authors' idea is to decompose the approximant into a spline part that makes the routine approximation plus another part \(\Phi\) that takes care of the large gradients. This \(\Phi\) could be an exponential function \(\exp(-m\cdot/\epsilon)\) for example. Another one is \(\Phi(x)=\sqrt{x+\epsilon}\) which is proposed in this work. (\(\epsilon\) is a small positive number in both cases.) The reason for using such a decomposition is that the (usually twice-)differentiable (cubic) splines are not adapted to mimicking these steep gradients. On the other hand, the approximations that are generated are exact on the \(\Phi\) part so no accuracy is lost by the decomposition, and it can be employed to full effect when computing the interpolants. The paper shows unique existence of such spline collications and convergence and stability properties are established.