Multi-stage waveform Relaxation and Multisplitting Methods for Differential Algebraic Systems
This work addresses the need for efficient numerical solvers for differential algebraic equations, but the contribution appears incremental as it extends existing waveform relaxation methods with additional decomposition stages.
The paper introduces multi-stage waveform relaxation and multisplitting methods for solving differential algebraic equations, using outer and inner iterations to decouple the system and enable parallel computation. A preliminary experiment on a DAE system demonstrates the approach, but no quantitative results are reported.
We are motivated to solve differential algebraic equations with new multi-stage and multisplitting methods. The multi-stage strategy of the waveform relaxation (WR) methods are given with outer and inner iterations. While the outer iterations decouple the initial value problem of differential algebraic equations (DAEs) in the form of $A \frac{d y(t)}{dt} + B y(t) = f(t)$ to $M_A \frac{d y^{k+1}(t)}{dt} + M_1 y^{k+1}(t) = N_1 y^k(t) + N_A \frac{d y^{k}(t)} + f(t)$, where $A = M_A - N_A$, $B = M_1 - N_1$. The inner iterations decouple further $M_1 = M_2 - N_2$ and $M_2 = M_3 - N_3$ with additional iterative processes, such that we result to invert simpler matrices and accelerate the solver process. The multisplitting method use additional a decomposition of the outer iterative process with parallel algorithms, based on the partition of unity, such that we could improve the solver method. We discuss the different algorithms and present a first experiment based on a DAE system.