Schwarz Waveform Relaxation for a Neutral Functional Partial Differential Equation Model of Lossless Coupled Transmission Lines
Shu-Lin Wu,Ting-Zhu Huang
DOI: https://doi.org/10.1137/110860975
IF: 2.968
2013-01-01
SIAM Journal on Scientific Computing
Abstract:In this paper, we investigate the convergence behavior of theSchwarz waveform relaxation (SWR) algorithm for partial neutralfunctional differential equations which arise from losslesscoupled transmission lines. We first analyze the algorithm atcontinuous level with transmission conditions of Dirichlet andRobin type. We show that the convergence rate of the algorithmcan be significantly improved by using the Robin transmissioncondition, provided the involved parameter is chosen properly. Then we discretize the model problem in space and investigatethe SWR algorithm at the semidiscrete level. It would be desirable ifthe convergence could be further improved by considering theinfluence of the space discretization, since the semidiscrete SWRalgorithm is closer to the fully discretized algorithm. Forthis point, we show that for the Dirichlet transmission conditionthe convergence rate of the SWR algorithm analyzed at the continuousand semidiscrete levels is equal. However, for the Robintransmission condition, using the parameter derived atthe semidiscrete level can both ameliorate and deteriorate theconvergence rate of the SWR algorithm in practical computations,compared to the one derived at the continuous level. In particular,when a relatively large space discretization size $\Delta x$ is used, the acceleration ability of the parameter derived at the semidiscretelevel outperforms the one derived at continuous level, while if the mesh parameter $\Delta x$ is small, the latter outperforms the former.The obtained results can also predict how the convergence rate ofthe algorithm with Robin transmission condition behaves withrespect to the time discretization size $\Delta t$ and the delayargument $\tau$. Numerical results are provided to validate ourtheoretical conclusions. (An erratum is attached.)