Electronic convergence influenced by parallelisation

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

Moderator: amadon

Post Reply
Posts: 6
Joined: Tue Jul 03, 2018 2:32 pm

Electronic convergence influenced by parallelisation

Post by Annelinde » Mon Jul 09, 2018 12:04 pm


I'm doing calculations on graphene nanoribbons (Armchair, 6 carbon atoms wide, edges pacified with hydrogen) and have some issues with the electronic convergence. I ran autoparal and chose the most efficient parallelisation (parallelisation on kpoints, bands and FFT grid), and with these settings I can't get the system to electronically converge.
I accidently ran a calculation with these parallelisation flags turned off, which results in default kpoint parallelisation alone, and the electronic convergence was a breeze (around 40 electronic steps needed).
I ran this calculation in both ABINIT 8.6.3 and 8.8.3.
I enclosed the outfiles (of ABINIT 8.6.3. , but I can provide the files of 8.8.3 as well) of these two calculations and the in-file.

I noticed that some internal parameters like mgfft, nfft and npw are different in these two cases.

What is the reason that I can't seem to perform a SCF with full parallelisation?

Thanks in advance

Outfile of a SCF with parallelisation over kpoints, bands and FFT grid
(69.67 KiB) Downloaded 184 times
Input file of the SCF. Tags governing the autoparallisation are either switched on or off completely to result in respectively the autoparal.out file and the kparal.out file.
(1.62 KiB) Downloaded 184 times
Outfile of a SCF with parallelisation over kpoints
(55.95 KiB) Downloaded 183 times

Post Reply