How to accelerate convergence in ground state calculations?

Total energy, DFT+U, BigDFT,...

Moderator: amadon

Post Reply
tangds
Posts: 8
Joined: Mon Jan 06, 2020 9:59 am

How to accelerate convergence in ground state calculations?

Post by tangds » Thu Sep 10, 2020 1:50 pm

Dear all,

I am running a series calculations of ground state with 108 atoms using 48 cores (in a single node). The settings "autoparal and paral_kgb" are both used in input file. To get the convergence "toldfe=1e-10", about 50 steps are needed, which is quite large. Is it possible to accelerate convergence?

Best,
Daosheng

ebousquet
Posts: 434
Joined: Tue Apr 19, 2011 11:13 am
Location: University of Liege, Belgium

Re: How to accelerate convergence in ground state calculations?

Post by ebousquet » Mon Sep 14, 2020 5:40 pm

Dear Daosheng,
It depends on the system, some can be more pathological than others.
Yiu can find some hints about how to improve convergence of the SCF in the "We2 – Torrent (PDF)" link of the last Abinit school here:
http://palata.fzu.cz/abinitschool/index.php?item=home

Flags that have to be tunes are diemix (diemixmag if magnetic), diemac, nline, etc.

Best wishes,
Eric

Post Reply