Parallel solution of ODEs

From MaRDI portal





[For the entire collection see Zbl 0745.00048.] The author speculates on the maximum parallelism that can be achieved in initial value problems if the concern is utilization of parallelism to achieve high speed rather than efficiency. The first part of this paper deals with parallelism across time: different processors are assigned to different points in time. One linear ordinary differential equation (ODE) can be solved in \(O(\log(N))\) time, but if the ODE is a system of \(M\) equations the total time is \(O(\log(N)\log(M))\). For nonlinear ODEs a new method, the blended waveform method, is described and analyzed. The second part of this paper deals with parallelism across space: different processors are assigned to each equation or subset of equations. The major limitation of such a method is communication. The amount of information that must be transmitted is examined.











This page was built for publication: Parallel solution of ODEs

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4026270)