BSE Tutorial - WAVEDER file missing
Moderators: Global Moderator, Moderator
-
- Newbie
- Posts: 11
- Joined: Sun Nov 17, 2019 7:39 pm
BSE Tutorial - WAVEDER file missing
Dear VASP Community,
I am trying to reproduce the official BSE tutorial[1]. But the job is getting aborted due to missing WAVEDER file text[2].
I am using Si_BSEfatbands.tgz folder given with the tutorial for this job. I tried to do it manually as well but same results.
Can anybody please point, what am I missing?
Thank you.
[1] https://www.vasp.at/wiki/index.php/Plot ... ture_of_Si
[2] https://github.com/jatin19k/TEST_files/ ... 5/vasp.log
I am trying to reproduce the official BSE tutorial[1]. But the job is getting aborted due to missing WAVEDER file text[2].
I am using Si_BSEfatbands.tgz folder given with the tutorial for this job. I tried to do it manually as well but same results.
Can anybody please point, what am I missing?
Thank you.
[1] https://www.vasp.at/wiki/index.php/Plot ... ture_of_Si
[2] https://github.com/jatin19k/TEST_files/ ... 5/vasp.log
-
- Global Moderator
- Posts: 314
- Joined: Mon Sep 13, 2021 12:45 pm
Re: BSE Tutorial - WAVEDER file missing
Dear Jatin19k,
It looks like you are running your calculation with 64 CPU cores and the code automatically increases the number of bands to get at least one band per core. Since the DFT calculation is performed with KPAR=2, each k-point is calculated by 32 cores and, hence, 32 bands are actually written in the WAVECAR and WAVEDER files. BSE in VASP does not support k-point parallelization and the code automatically increases the number of bands to 64, which makes this BSE calculation incompatible with the WAVEDER file.
You can avoid this problem by performing this calculation with fewer CPU cores, for example 8 or 16. Otherwise, adjust NBANDS to make sure that WAVEDER is compatible with the BSE calculation.
It looks like you are running your calculation with 64 CPU cores and the code automatically increases the number of bands to get at least one band per core. Since the DFT calculation is performed with KPAR=2, each k-point is calculated by 32 cores and, hence, 32 bands are actually written in the WAVECAR and WAVEDER files. BSE in VASP does not support k-point parallelization and the code automatically increases the number of bands to 64, which makes this BSE calculation incompatible with the WAVEDER file.
You can avoid this problem by performing this calculation with fewer CPU cores, for example 8 or 16. Otherwise, adjust NBANDS to make sure that WAVEDER is compatible with the BSE calculation.
-
- Newbie
- Posts: 11
- Joined: Sun Nov 17, 2019 7:39 pm
Re: BSE Tutorial - WAVEDER file missing
Hi Alexey,
Thanks for a quick and informative reply. This solves the problem after lowering the cores count to 16.
Moreover, as per the tutorial webpage the job shall take 50 mins on 16 cores but it is taking greater than 24 hrs on XSEDE supercomputer and that too while consuming 220GB memory. To lower the memory needed I had reduced the # of kpoint to 4 but it looks like something is going wrong as you can see in below messages in the error file:
rm: cannot remove 'WAVECAR*': No such file or directory
rm: cannot remove 'WAVEDER*': No such file or directory
head: cannot open 'BSEFATBAND' for reading: No such file or directory
tail: cannot open 'BSEFATBAND' for reading: No such file or directory
mv: cannot stat 'BSEFATBAND': No such file or directory
Please look for shared OUTCAR[1] and log files[2] below.
Jatin
[1] https://github.com/jatin19k/TEST_files/ ... CAR.mBSE.4
[2] https://github.com/jatin19k/TEST_files/ ... 4/vasp.log
Thanks for a quick and informative reply. This solves the problem after lowering the cores count to 16.
Moreover, as per the tutorial webpage the job shall take 50 mins on 16 cores but it is taking greater than 24 hrs on XSEDE supercomputer and that too while consuming 220GB memory. To lower the memory needed I had reduced the # of kpoint to 4 but it looks like something is going wrong as you can see in below messages in the error file:
rm: cannot remove 'WAVECAR*': No such file or directory
rm: cannot remove 'WAVEDER*': No such file or directory
head: cannot open 'BSEFATBAND' for reading: No such file or directory
tail: cannot open 'BSEFATBAND' for reading: No such file or directory
mv: cannot stat 'BSEFATBAND': No such file or directory
Please look for shared OUTCAR[1] and log files[2] below.
Jatin
[1] https://github.com/jatin19k/TEST_files/ ... CAR.mBSE.4
[2] https://github.com/jatin19k/TEST_files/ ... 4/vasp.log
-
- Global Moderator
- Posts: 314
- Joined: Mon Sep 13, 2021 12:45 pm
Re: BSE Tutorial - WAVEDER file missing
I don't see any errors in these output files. Can you please provide the full log file? Also, note that if you change the number of k-points, the BSE matrix rank will be different and you need to adjust the lines that you extract with head and tail from BSEFATBAND.
-
- Newbie
- Posts: 11
- Joined: Sun Nov 17, 2019 7:39 pm
Re: BSE Tutorial - WAVEDER file missing
Hi Alexey,
That were final output files. In addition I am attaching DFT OUTCAR file here[1].
[1]
https://github.com/jatin19k/TEST_files/ ... TCAR.DFT.4
That were final output files. In addition I am attaching DFT OUTCAR file here[1].
[1]
https://github.com/jatin19k/TEST_files/ ... TCAR.DFT.4
-
- Global Moderator
- Posts: 314
- Joined: Mon Sep 13, 2021 12:45 pm
Re: BSE Tutorial - WAVEDER file missing
Both OUTCAR files look fine. The BSE log file suggests that BSEFATBAND is created in the BSE calculation: "Writing out 10 BSE eigenvectors", but head and tail do not see this file. Do you see BSEFATBAND in your directory? Perhaps you could remove all the postprocessing from the run script and just perform the VASP calculations to make sure that it produces all the necessary files.
-
- Newbie
- Posts: 11
- Joined: Sun Nov 17, 2019 7:39 pm
Re: BSE Tutorial - WAVEDER file missing
Surprise, it works but on another machine. I don't know, what's the difference, I used the same input files.
I still need to fix it on my first machine since that is very fast in comparison to second. Let me know what files I can share here from working machine to fix it on non-working machine.
One more question, can you please help me to understand how did you got the plot given at the end of tutorial[1].
I spend few hours on understanding this plot but hitting wall here.
The only thing in the tutorial is given, is "The sample output for this tutorial should look like the following:"
Thank you.
[1]
https://www.vasp.at/wiki/index.php/Plot ... ture_of_Si
I still need to fix it on my first machine since that is very fast in comparison to second. Let me know what files I can share here from working machine to fix it on non-working machine.
One more question, can you please help me to understand how did you got the plot given at the end of tutorial[1].
I spend few hours on understanding this plot but hitting wall here.
The only thing in the tutorial is given, is "The sample output for this tutorial should look like the following:"
Thank you.
[1]
https://www.vasp.at/wiki/index.php/Plot ... ture_of_Si
-
- Newbie
- Posts: 11
- Joined: Sun Nov 17, 2019 7:39 pm
Re: BSE Tutorial - WAVEDER file missing
To answer your above questions: there is no BSEFATBAND file generated and I modified the script to remove all post processing, still no BSEFATBAND.
-
- Global Moderator
- Posts: 314
- Joined: Mon Sep 13, 2021 12:45 pm
Re: BSE Tutorial - WAVEDER file missing
Are you using the same VASP executable on both machines?
Does the calculation crash when you run it without the postprocessing? Can you provide the output files?
The plot in the tutorial shows the band structure with the coupling coefficient |A| (circles) for pairs of valence-conduction states.
Does the calculation crash when you run it without the postprocessing? Can you provide the output files?
The plot in the tutorial shows the band structure with the coupling coefficient |A| (circles) for pairs of valence-conduction states.