[gpaw-users] mpi-problem

[Torsten Hahn]

Am 20.01.2011 um 16:13 schrieb Jens Jørgen Mortensen:

> On Thu, 2011-01-20 at 12:18 +0100, Torsten Hahn wrote:
>> 
>> Am 20.01.2011 um 11:59 schrieb Jens Jørgen Mortensen:
>> 
>>> On Thu, 2011-01-20 at 10:09 +0100, Torsten Hahn wrote:
>>>> Dear Jussi,
>>>> 
>>>> i attached a minimum example which shows the problem. The error
>>>> occurs in most cases right after the first scf-cycle is finished
>>>> during structure optimization.
>>> 
>>> On how many cores do you run this calculation?
>>> 
>> 
>> 
>> 32 cores
> 
> It also fails for me on 32 cores.  I haven't found the bug, but I have
> discovered that the code crashes while it is calculating the forces.  I
> also found that if I comment the line "setups={None:'hgh'}" it seems to
> work.  Could you check that this works for you also?
> 
> Jens Jørgen
> 
> PS: What is the reason you want to use HGH pseudopotentials?

If i do not use 'hgh' pseudopotentials it seems to work.

I investigated a bit further. The code breaks with ncores >= 8. For me this looks like a race condition. Should we open a ticket for this ?

There was no special reason for using this. I looked only for a way to have a reasonable 'fast' way to look if the rest of the calculation script (not send to this list) works and i therefore looked for a low-end setup.

Best,
Torsten.