[gpaw-users] memory and gpw-file problems

[Georg Kastlunger]

Dear Mailing list,

I have a little problem regarding the memory consumption when I am 
trying to do a gridrefinement of 4 for the all electron density. 
Unfortunately the routine always crashes because of memory overflow, so 
I wanted to ask you if you know how the refinement and writing of the 
output is parallelized. Till now i know that every core gets the whole 
grid (i.e every value on every grid point), which leads to a very high 
memory consumption.
Does anyone know how I could solve this problem?
Doing a gridrefinement of 2 does not lead to any crashes, but I would 
like to have the gridrefinement of 4 for my bader analysis.

The second and somehow related issue is an error I get using the delta 
scf routines (In the present example I used the Molecularorbitals class).
The problem I am facing right now is that as it seems the gpw-files that 
are written from some delta scf calculations cannot be used as a restart 
point, when I am trying to get the all electron density.
Restarting form these files leads to the following error message:

#####################################################################################
python: c/extensions.h:40: gpaw_malloc: Assertion `p != ((void *)0)' failed.
/work/sge/default/spool/r15n15/job_scripts/1516372: line 13: 11011 
Aborted                 python write_density.py
######################################################################################

In this case i didn't use an mpi run since I just wanted to show you the 
error. Does anyone know what this problem means and how i can solve it 
or work around it?
The script I am using to get this error is very simple:

####################
from ase import *
from gpaw import *
import sys

atoms,calc=restart('the_bad_file.gpw')


rho = atoms.calc.get_all_electron_density(gridrefinement=4) * Bohr**3
write('Fortunately_a_density.cube', atoms, data=rho)
#####################

As you can see both problems are relevant in this script. Since the 
system on which I am applying this script is computationally quite 
demanding, I would really like to have the possibility to start from the 
gpw file instead of recalculating everything in every try.
I have also tried to apply the script to smaller systems calculated with 
delta scf, where the error does not appear.

If anyone was facing the same problems and/or was able to solve it I 
would be grateful for an advice.

Thanks in advance,
Georg