Adaptive spline interpolation for Hamilton-Jacobi-Bellman equations (Q2497779)

The authors study an adaptive discretization technique using cubic spline interpolation and the performance of adaptive spline interpolation in semi-Lagrangian discretization schemes for Hamilton-Jacobi-Bellman (H-J-B) equations, respectively. The approach turns out to be efficient in case of smooth solutions which is investigated analytically and illustrated by a numerical example. Main result: For smooth solutions adaptive spline approximations perform very well and can considerably improve the results obtained with both adaptive low order and nonadaptive high order methods. For nonsmooth solutions (the authors always understand the solutions of the H-J-B equation in the viscosity solutions sense) numerical instabilities may occur which can seriously affect and even destroy the convergence of the scheme. The local approximation properties of cubic splines on locally refined grids by a theoretical analysis are investigated. How the proposed method performs in practice is shown by the numerical examples. Finally the authors (using those examples) also illustrate numerical stability.