Open jtkrogel opened 2 years ago
The Localized-ACE approach introduced in this paper saves CPU time in hybrid calculations and makes the algorithm more robust. The authors apply hybrid calculations up to 565 atoms with negligible inaccuracies introduced by the approximation.
I believe it is triggered by localization_thr
keyword in pw.x
.
This issue is for discussion on best practices to achieve speedup for individual SCF iterations, particularly for plane-wave hard potentials.
This includes parallelization and any accelerating input parameters for particular cases.
Once #4 is reasonably settled, the resolution of this issue is key to reduce overall time to solution and increase scientific productivity.