Parallel Iterative Time Domain Decompositions Based on Deferred Correc- Tion Methods for Solving Ode Systems

The development of large computational resources leads to search for parallel implementation not only based on space decomposition. In this talk we will propose to combine time domain decomposition and the simple deferred correction. The deferred correction [1, 2] builds a perturbated problemwhich the exact solution is known. The defect difference between the perturbated solution and the solution of the original problem, is computed by a process increasing the time accurary at each iteration. The iterative correction is computed sequentially on each time steps. To introduce parallelism, we gather the time steps into subdomains distributed among processors. A parallel pipe is then defined as: a processor computes its correction, sends it to the next subdomain and starts its next iteration. Hence a cyclic distribution of the time subdomains is introduced to reduced the startup of the first step of the iterated correction process. Numerical results and parallel efficiency will be presented.