Preconditioned implicit solution of linear hyperbolic equations with adaptivity (Q596204)

An implicit method for the solution of time-dependent partial differential equations (PDEs) with grid adaptivity is developed and the stability of the adaptive scheme is analysed. The systems of linear equations which arise at each time step are solved by the generalized minimal residual (GMRES) method with preconditioning, and the analysis of the preconditioner shows that it will improve the convergence rate for stiff problems. This behaviour is observed in numerical experiments where only the slow time scale is present in the solution. The authors study the adaptive solution of linear hyperbolic PDEs and, in particular, the solution of problems with different time scales. Such problems are best solved by implicit time stepping methods where the solution vector satisfies a system of linear equations at each time step. These systems are solved using a preconditioned iterative method, resulting in an efficient solution procedure well suited to the space adaptivity. Earlier the authors have presented an implicit adaptive method for solving hyperbolic PDEs that is second-order accurate in both space and time. A structured grid is partitioned into predetermined blocks that are distributed dynamically over the available processors for a good load balance on a parallel computer. Data at block boundaries of neighbouring blocks are communicated by message passing. The discretisation in space is obtained via a cell-centered finite volume method with the fluxes computed using a centred or upwind scheme of second order. At block boundaries, ghost cells are added and a jump in the grid size is allowed. The unknowns in these cells are computed using interpolation in the neighbouring block. This combination of upwinding and one-sided interpolation ensures stability at the block boundaries. Time integration is done using a linear multistep method. A method which is explicit in time would require a time step proportional to the smallest space step in the grid due to the restrictions imposed by the Courant-Friedrichs-Levy (CFL) condition. By using implicit time stepping, however, the same large time step can be taken in all blocks as long as the discretisation error in time is small enough. Moreover, for problems with different time scales, it is the CFL condition for the fastest scale that would restrict the time step to be used with an explicit method; with an implicit method, if only the slowest time scale is of interest, a larger time step can be used without loss of solution accuracy. The discretisation error in space is estimated by comparing the discretisation on two different grids, and the grid size is adjusted so that a prescribed error tolerance on the local error is satisfied. At each time step, the numerical method described in the paper yields a linear system of equations. In the present paper, the authors investigate problems with different time scales, where the only slow scale processes are interested in. In such cases, a strongly nondiagonally dominant system of equations have to be solved and preconditioning is necessary to ensure fast convergence, so, in the paper a block preconditioner with semi-Toeplitz blocks is used. In the Section 2 the approximation in space and time together with the adaptive error control mechanism is described. It is also proved that the treatment at the block boundaries is stable for a model equation. In Sections 3 and 4, the iterative method and the preconditioner mentioned above are discussed. Spectral and asymptotic convergence analysis of this algorithm is presented in Section 5. Finally, in Section 6, the theoretical analysis with numerical experiments are corroborated.