Abstract de la publi numéro 11385

We consider the problem of efficiently computing a set of entries in the inverse of a sparse matrix in the context of out-of-core direct solvers, where the factors are stored on disk. In this computational setting, the main issue is how to partition the requested entries into blocks such that memory requirements are kept limited while the cost of loading factors from the disk is reduced. In this talk, we investigate several partitioning models and methods, propose algorithmic contributions, and present a few open problems.