Ecut Convergence and Parallelism  [SOLVED]

Total energy, geometry optimization, DFT+U, spin....

Moderator: bguster

Locked
Patrick
Posts: 4
Joined: Wed Jan 13, 2016 5:01 pm

Ecut Convergence and Parallelism

Post by Patrick » Wed Jan 13, 2016 6:01 pm

Hi All,

I am attempting to perform ecut convergence tests for a system at finite temperature requiring a (relatively) large number of bands. I've noticed that when I go to run the code in parallel I find error messages of the form:

--- !WARNING
message: |
Number of bands mband 1024 > number of planewaves mpw = 218
The number of eigenvectors cannot be greater that the size of the Hamiltonian!
Action: decrease nband or, alternatively, increase ecut
src_file: gstate.F90
src_line: 404
...

The value for mband agrees with the nband in my input file, however mpw given here does not seem to agree with my ecut (115 Hartrees ~ 10^4 plane waves for my box size). I've noticed that when running with fewer cores this mpw value goes up to the point that this warning disappears. I've read the 'Large-scale ab initio calculations based on three levels of parallelisation' by F. Bottin et al as recommended by the Abinit documentation, but am still confused as to how mpw is calculated and its relation to ecut.

Most importantly, given that I do not encounter this warning for fewer cores, is it therefore the case that I might have a lower ecut (or even a faster calculation) when running on fewer cores?

Many Thanks,
Patrick

User avatar
gmatteo
Posts: 291
Joined: Sun Aug 16, 2009 5:40 pm

Re: Ecut Convergence and Parallelism  [SOLVED]

Post by gmatteo » Wed Jan 13, 2016 7:21 pm

The test on the number of bands was not implemented correctly when paral_kgb == 1.
You can ignore the warning reported in the log file.

Patrick
Posts: 4
Joined: Wed Jan 13, 2016 5:01 pm

Re: Ecut Convergence and Parallelism

Post by Patrick » Thu Jan 28, 2016 1:17 pm

Ok, many thanks gmatteo!

Locked