[gpaw-users] RPA calculations hangs after first energy

[Jens Jørgen Mortensen]

On 11/19/2014 01:45 AM, Thomas Stenbæk Jauho wrote:
> Hi,
>
> My calculations are divided into three scripts:
>
> 1) a DFT calculcation,
> 2) a diagonalization
> 3) and the RPA calcucaltion.
>
> The attached combination of scripts resulted in a failed RPA 
> calculation. I've also attached the RPA output from the failed 
> calculation.

I can't reproduce this.  Do I need your input file 
"../Structures/PBE_Na2O.cif" ?  I ran it with bulk silicon.

Jens Jørgen

>
> I have however fund that if I run the calculation again after a failed 
> run it will run successfully. It seems like it is only the first run 
> that is troublesome. My solution is thus to run a very short 
> calculation that purposely times out, and the run my calculations again.
>
> Best,
> Thomas
>
>
> On 2014-11-18 05:49, Jens Jørgen Mortensen wrote:
>> On 11/14/2014 08:52 PM, Thomas Stenbæk Jauho wrote:
>>> Hi,
>>>
>>> I have anoter RPA related question.
>>>
>>> After updating to the newest developer version of GPAW, some (but 
>>> not all) of my RPA calculations will only calculate one energy (i.e. 
>>> at 1 q-point, at 1 cut-off). Calculations I know should take less 
>>> than 1 cpu hour, ran out of time after 8 cpu hours.
>>>
>>> Investigating the issue I found that using .gpw files created before 
>>> my update resolves the issue. However as I don't have old .gpw files 
>>> for all my bulk systems, this approach helps for one or two systems. 
>>> And checking my files and backup I can confirm that I haven't 
>>> changed my way of generating the .gpw files used for RPA.
>>>
>>> Further trail and error found that removing the filename keyword 
>>> (used for restarts of the calculcation) solved the issue. This was 
>>> the only thing I've added since the update.
>>>
>>> It is beyond me what the core issue really is, since two very 
>>> different solutions work. I have attached my RPA scripts, as wells 
>>> the outputs.
>>>
>>> I am hoping that somebody has another way to resolve this issue, as 
>>> I would like to use the restart feature for my larger calculations.
>>
>> Can you send us a simple and complete example (from ground-state to
>> RPA-correlation) that fails?
>>
>> Jens Jørgen