Two-level adaptation for Adaptive Multipreconditioned FETI

This article introduces two strategies to reduce the memory cost of Adaptive Multipreconditioned FETI method (AMPFETI) while preserving its capability solve ill conditioned systems efficiently. Their common principle is gather search directions into aggregates which are frequently adapted in order achieve best compromise between decrease solver error and computational resources employed. The methods assessed on weak scalability studies highly heterogeneous problems up 10,368 cores half a billion unknowns, ill-conditioned industrial applications, related numerical homogenization solid propellant simulation multiperforated aircraft combustion chamber.