Convergence analysis of a \textit{periodic-like} waveform relaxation method for initial-value problems via the diagonalization technique (Q2326376): Difference between revisions

The authors present a parallel-in-time method (also called a time-parallel solver) for the solution technique of the ordinary differential equations (ODE) of the type \[ \dot{u}(t) = f(t, u(t)) \quad \forall t>0, \quad u(0) = u_0. \] The time-parallel solver developed in the article uses a waveform relaxation technique that generates a function sequence which converges to the solution of the original ODE in the limit. Each element of the aforementioned function sequence is a solution to a period-like ODE and therefore the application of a diagonalization technique is very suitable. Such diagonalization yields a natural parallel-in-time method for the solution of each iterate and therefore the original ODE. The idea behind applying the diagonalization on these iterates is to avoid the difficulties arising from application of the diagonalization technique on the original ODE. Indeed, when the diagonalization is applied to the original ODE, it is hard to balance the diagonalization roundoff errors and the discretization error. The authors analyze the resulting algorithm, and in particular derive bounds on the roundoff error of the diagonalization procedure as well as convergence factor of the algorithm both at the discrete and continuous level. Finally the authors provide numerical experiments supporting the theoretical results.