On numerical methods and error estimates for degenerate fractional convection-diffusion equations (Q2510401)

The authors propose and analyze numerical methods for solving integro partial differential equations of the type: \[ \partial_t u + \text{div}f(u) = {\mathcal L}^\mu [A(u)] \text{ for all } (x,t) \in \mathbb R^d \times (0,T) \text{ and } u(x,0) = u_0(x) \] for all \(x \in \mathbb R^d\), where \[ {\mathcal L}^\mu[\phi](x) = \int_{| z | > 0} \phi(x+z) - \phi(x) - z \cdot \nabla\phi(x) \pmb{1}_{| z | < 1} (z) \text{d} \mu(z) \] for smooth bounded functions \(\phi\) (\(\pmb{1}\) denotes the indicator function). At first, the case \(f \equiv 0\) is considered. The presented numerical method is based on a finite volume discretization for the discretization in space and forward or backward difference approximations for discretization in time. The presented method is an extension of a method described by the authors earlier [Ann. Inst. Henri Poincaré, Anal. Non Linéaire 28, No. 3, 413--441 (2011; Zbl 1217.35204)]. It is shown that the proposed methods are conservative and monotone. Several a priori estimates for the numerical solutions as well as the existence and uniqueness of the solution of the numerical schemes are proved. A new \textit{N. N. Kuznetsov}-type theory (see [U.S.S.R. Comput. Math. Math. Phys. 16 (1976), No. 6, 105--119 (1978; Zbl 0381.35015)]) for getting error estimates and convergence rates for the proposed numerical schemes is developed. Finally, it is discussed how the results obtained in the case \(f \equiv 0\) can be extended to the case \(f \not\equiv 0\).