New Algorithms for Optimal Online Checkpointing

摘要: Frequently, the computation of derivatives for optimizing time-dependent problems is based on integration adjoint differential equation. For this purpose, knowledge complete forward solution may be required. Similar information needed in context a posteriori error estimation with respect to given functional. In area flow control, especially three dimensional problems, it usually impossible keep track full due lack storage capacities. Further, many adaptive time-stepping procedures are toward efficient schemes time. Therefore, standard optimal offline checkpointing strategies not well suited that framework. paper we present two algorithms an online procedure determines checkpoint distribution fly. We prove these approaches yield distributions either or almost only small gap optimality. Numerical results underline theoretical results.