[gpaw-users] Regarding 'RuntimeError: Atoms objects on different processors are not identical!'

Wed Mar 4 14:48:41 CET 2015

Hi Ask

The version we have is 0.10.0.11364. I will try the trunk and let you know 
the results

Kind regards
Yuriy

From:
Ask Hjorth Larsen <asklarsen at gmail.com>
To:
Yuriy Elesin <yuel at topsoe.dk>, 
Cc:
Morten Niklas Gjerding <mogje at fysik.dtu.dk>, gpaw-users 
<gpaw-users at listserv.fysik.dtu.dk>, Hanne Falsig <hafa at topsoe.dk>
Date:
03/03/2015 08:16 PM
Subject:
Re: [gpaw-users] Regarding 'RuntimeError: Atoms objects on different 
processors are not identical!'

I recently changed something having to do with that part.  You could try 
with trunk.  Which version are you using?

After the change, when the code runs the check it will allow for an error 
of 1e-8 (approx. square root of machine precision), then simply use the 
positions from the master process if it doesn't raise the error.

Best regards
Ask

2015-03-03 14:26 GMT+01:00 Yuriy Elesin <yuel at topsoe.dk>:
Thank you Morten and Ask! 

It has generated the output as a bunch of 
compare_atoms_r00**_positions.pickle files. Plus it generated one 
compare_atoms_fps_positions.pickle file. When I compare all the r00** 
files it indicates that positions on most of the cores are identical, 
except a few cores where the difference between positions generates one 
line with 4.33680869e-19 as difference: 
.... 
[  0.00000000e+00,   0.00000000e+00,   0.00000000e+00] 
[  0.00000000e+00,   4.33680869e-19,   0.00000000e+00] 
[  0.00000000e+00,   0.00000000e+00,   0.00000000e+00] 
.... 

It is always the same value, when the positions are different. But again 
out of 24 cores there are only 8 (I think) cores with different then the 
rest coordinates. 

The compare_atoms_fps_positions.pickle file has the following content: 

array([-8145271529702367328,  3281356470061976748,  3281356470061976748, 
        3281356470061976748,  3281356470061976748,  3281356470061976748, 
        3281356470061976748,  3281356470061976748,  3281356470061976748, 
       -8145271529702367328,  3281356470061976748,  3281356470061976748, 
        3281356470061976748,  3281356470061976748, -8145271529702367328, 
       -8145271529702367328,  3281356470061976748,  3281356470061976748, 
        3281356470061976748,  3281356470061976748, -8145271529702367328, 
        3281356470061976748,  3281356470061976748,  3281356470061976748]) 

Does it tell you anything? If it is just minor inaccuracies then how do we 
prevent the following error that seem to stop the calculation: 

RuntimeError: Atoms objects on different processors are not identical! 
GPAW CLEANUP (node 23): <type 'exceptions.RuntimeError'> occurred.  
Calling MPI_Abort! 

Kind regards 
Yuriy 

From: 
Morten Niklas Gjerding <mogje at fysik.dtu.dk> 
To: 
Yuriy Elesin <yuel at topsoe.dk>, Ask Hjorth Larsen <asklarsen at gmail.com>, 
Cc: 
gpaw-users <gpaw-users at listserv.fysik.dtu.dk> 
Date: 
03/03/2015 11:15 AM 
Subject: 
RE: [gpaw-users] Regarding 'RuntimeError: Atoms objects on different 
processors are not identical!'

Hi Yury.

I think that:

gpaw-python script.py --debug

is what you want.

KR. Morten. 

From: gpaw-users-bounces at listserv.fysik.dtu.dk [
gpaw-users-bounces at listserv.fysik.dtu.dk] on behalf of Yuriy Elesin [
yuel at topsoe.dk]
Sent: Tuesday, March 03, 2015 9:56 AM
To: Ask Hjorth Larsen
Cc: gpaw-users
Subject: Re: [gpaw-users] Regarding 'RuntimeError: Atoms objects on 
different processors are not identical!'

Hi Ask 

How do we enable the debug mode in GPAW? I tried gpaw-python with -d 
option but it says it is only for the parser, so no files with faulty 
positions were generated. 

Kind regards 
Yuiry 

From: 
Ask Hjorth Larsen <asklarsen at gmail.com> 
To: 
Hanne Falsig <hafa at topsoe.dk>, 
Cc: 
gpaw-users <gpaw-users at listserv.fysik.dtu.dk>, yuel at topsoe.dk 
Date: 
03/02/2015 04:03 PM 
Subject: 
Re: [gpaw-users] Regarding 'RuntimeError: Atoms objects on different 
processors are not identical!'

Hi Hanne

In trunk (and possibly in the old version; this I don't remember) the 
faulty positions will be dumped to files if you use debug mode.  Then you 
can check whether they completely disagree (something is wrong with the 
parallelization, probably of the XC functional) or whether there's just 
some numerical noise (some calculation is duplicated in a way that allows 
numerical inaccuracies across different cores).

Best regards 
Ask 

2015-03-02 11:30 GMT+01:00 Hanne Falsig <hafa at topsoe.dk>: 
Dear all, 

I get the following error message: 
RuntimeError: Atoms objects on different processors are not identical! 

I can see there have been a number of questions about this. Did someone 
find a solution  to this problem? 

My input is: 
calc = 
GPAW(xc='BEEF-vdW',kpts=[1,1,1],h=0.20,mixer=MixerSum(),spinpol=True, 
maxiter=300,occupations=FermiDirac(width=0.1),txt='gpaw.txt') 

We use gpaw version 0.10.0.11364 

Best regards, 
Hanne 
Hanne Falsig 
Research Scientist | Physical Analytical, R&D 

Haldor Topsoe A/S
Nymøllevej 55, DK-2800 Kgs. Lyngby
Phone (direct): +45 2275 4776 
Switchboard: +45 4527 2000 

Read more at www.topsoe.com 

Haldor Topsøe A/S: CVR: 41853816 This e-mail message (including 
attachments, if any) is confidential and may be privileged. It is intended 
only for the addressee. 
Any unauthorised distribution or disclosure is prohibited. Disclosure to 
anyone other than the intended recipient does not constitute waiver of 
privilege. 
If you have received this email in error, please notify the sender by 
email and delete it and any attachments from your computer system and 
records. 
HALDOR TOPSOE (www.topsoe.com) 

_______________________________________________
gpaw-users mailing list
gpaw-users at listserv.fysik.dtu.dk
https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users 

Haldor Topsøe A/S: CVR: 41853816 This e-mail message (including 
attachments, if any) is confidential and may be privileged. It is intended 
only for the addressee. 
Any unauthorised distribution or disclosure is prohibited. Disclosure to 
anyone other than the intended recipient does not constitute waiver of 
privilege. 
If you have received this email in error, please notify the sender by 
email and delete it and any attachments from your computer system and 
records. 
HALDOR TOPSOE (www.topsoe.com) 

Haldor Topsøe A/S: CVR: 41853816. This e-mail message (including attachments, if any) is confidential and may be privileged. It is intended only for the addressee.Any unauthorised distribution or disclosure is prohibited. Disclosure to anyone other than the intended recipient does not constitute waiver of privilege.If you have received this email in error, please notify the sender by email and delete it and any attachments from your computer system and records.HALDOR TOPSOE (www.topsoe.com)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20150304/04d7fea3/attachment-0001.html>