Kernel based high order ``explicit'' unconditionally stable scheme for nonlinear degenerate advection-diffusion equations (Q2302400)

In the present article, authors propose a high-order numerical scheme for solving a class of nonlinear degenerate parabolic equations written as \[ \partial_t u+\partial_x f(u)=\partial_{xx}g(u), \] where \(g'(u)\geq 0\) and \(g'(u)\) can vanish for some values of \(u\). This type of equations arises in a wide range of applications, containing for example radiative transport or porous medium flow. It has similar properties as hyperbolic conservation laws, including in particular possible existence of nonsmooth solutions. It is therefore important to consider a high order approximation for this type of equations. Moreover, since the problem is parabolic, a classical explicit time discretization would imply a constraining restriction on the time step to ensure stability, while classical implicit methods would lead to invert some operator at each time steps. To overcome these issues, an approach based on the method of lines transpose (\(MOL^T\)) is considered. The idea is to first discretize the time, here with an explicit strong-stability-preserving Runge-Kutta (SSP RK) method, and then to solve a boundary value problem at each discrete time levels. Let us mention that the same ideas where used in [\textit{A. Christlieb} et al., J. Comput. Phys. 327, 337--367 (2016; Zbl 1422.65432)] for advection equations, the main difference being that in this previous article, an implicit SSP RK method was used instead of an explicit one. Then, the spatial derivatives are represented as infinite series, in which each term relies on a special kernel based formulation of the solution. This approach makes the method effectively implicit at each stage of the explicit SSP RK method, but without the need of invert any operator. Finally, a WENO methodology is applied, but may not be sufficient to suppress solution overshoots in the case of nonsmooth problems. To enhance robustness of the method, a nonlinear filter is introduced, constructed thanks to the smoothness indicators provided by the WENO methodology, and then not increasing the computational cost. Unconditional stability of the proposed numerical scheme is rigorously proved in the linear case with periodic boundary conditions, for order \(k=1,\,2,\,3\). The stability for the more general nonlinear case is not theoretically established, but numerically confirmed. Several numerical 1D and 2D test cases, including porous medium equation, Buckley-Leverett equation and strongly degenerate parabolic problems, are presented. Numerical results demonstrate efficiency of the proposed method, high order accuracy and ability to produce nonoscillatory shock transitions. Moreover, it appears that the CFL number can be chosen arbitrarily large, leading only to a loss of accuracy and not a stability issue.