pipette
ENEnglish

Projection-Based Parareal for Nonlinear Initial-Value Problems

Chenyi Tan, Yuncheng Xu, Yehao Zhang, Yangfeng Su

Preprint

En palabras de los autores

Parareal is a parallel-in-time method for initial-value problems that has found extensive applications and given rise to many variants. Krylov-enhanced parareal is a variant of parareal that uses projection techniques to accelerate convergence substantially for linear problems. In this paper, we study a projection-based parareal algorithm for nonlinear initial-value problems with a modified iteration scheme and an arbitrary orthogonal projection. For a general nonlinear problem, we prove a convergence estimate for each iteration step on a fixed time grid. We analyze the errors of the modified scheme and the scheme of Krylov-enhanced parareal relative to the old error as the coarse grid is refined. For a coarse solver of order and step size , we prove error bounds of for the modified scheme and for that of Krylov-enhanced parareal at the first potentially inexact time point under mild assumptions. Numerical experiments on two nonlinear problems are consistent with the theoretical analysis.

Resultado principalLimitación que admiten los autores

Apareció: viernes, 16 de octubre. arXiv. Preprint, todavía sin revisión por pares.