dfpt_nstwf.F90 bug message

[jzwanzig]

Hi, I'm running electric field perturbation calculations of the type I've done many times before--general chain of calculations looks like

1) GS, saving WFK

2) DDK, rfdir 1 1 1, reading WFK from 1), saving 1WF..

3) rfelfd 3, rfdir 1 1 1, reading WFK from 1) and 1WF from 2

Usually this works but now for one strained cell I get a crash at the the point in 3) where the SCF cycle is converged and the system is reading in the DDK wavefunctions; the error message is

--- !BUG
src_file: dfpt_nstwf.F90
src_line: 250
mpi_rank: 0
message: |
For isppol = 1, ikpt = 1 and idir = 1
the number of plane waves in the ddk file is equal to 7285
while it should be 7360
...

I understand what this means, and could find it in the code, but I don't understand quite how it could happen or what to do about it. (I do have ecutsm set to a non-zero value, is that a problem? It hasn't been before...)

thanks for any help,

Joe

follow-up--

ngfft is being set differently in the runs of the response function calculation--

GS: ngfft 45 45 60
DDK: ngfft 45 45 60
RFELFD: ngfft 48 45 60

quite odd because the cell is primitive orthorhombic in all 3 cases (of course).

I am now running with ngfft set in the input file to 45 45 60 ("manu militari" in the parlance of Matthieu Verstraete).

more follow up: manually setting ngfft doesn't help, code still fails in the same way.

[mverstra]

Hi Joe!

love it when you cite me...

Do you understand why the fft grid changes (ecutsm or different set of k points or k+q points?) I suspect internally somewhere it is still using a different grid. Anything fishy in your system (non collinear magnetism, parallelism...) - do you have a minimal input file that crashes and runs in a few minutes?

cheers