Higher order time stepping methods for subdiffusion problems based on weighted and shifted Grünwald-Letnikov formulae with nonsmooth data (Q2188041)

The authors propose two higher-order time stepping methods for solving subdiffusion problems. The Caputo time fractional derivatives are approximated by using the weighted and shifted Grünwald-Letnikov formulae introduced recently. The proposed time stepping methods have essentially \(\mathcal{O}(k^2)\) and \(\mathcal{O}(k^3)\) as optimal convergence orders, respectively, for both smooth and nonsmooth data. The error estimates are proved by directly bounding the approximation errors of the kernel functions, for both homogeneous and inhomogeneous cases. Employing Laplace transform techniques, it is shown that the error estimates are even suitable for the more general case involving an elliptic operator if it satisfies the resolvent estimate. Numerical illustrations are given to demonstrate how the numerical results are consistent with the proven theoretical results.