Page 1 of 1

Time consuming in writing wavefunction

Posted: Thu Apr 08, 2021 12:44 am
by hongyang_ma
Hi,

I'm a user of VASP 6.1. I'm currently working on a 1000 atoms system using HSE06.
The speed of the calculation is not that slow. If I start the calculation from an existing WAVECAR, each SCF can converge within 3 cycles and this process takes around 260 mins. However, the issue is that after each SCF cycle, the process for writing wavefunction is very compuationally costly which takes around 200 mins for each SCF cycle. And if I do not write the WAVECAR by setting "LWAVE = .FALSE.", the first SCF cycle of the next restarted calculation with setting "ISTART = 0" will take around 35 cycles (~2500 mins) to converge. So restarting from an existing WAVECAR is necessary for me too.
So my question is that in VASP, is there any usefull keywords to speed up the writing process of WAVECAR? And is it possible to write the wavefunction only in the last SCF cycle rather than each SCF cycle (such as putting a keyword in the STOPCAR to constrol the writing of wavefunction)?

Thanks & Regards,
Hongyang

Re: Time consuming in writing wavefunction

Posted: Mon Apr 12, 2021 10:39 am
by ferenc_karsai
Usually the WAVECAR is only written at the end of the SCF cycle. For hybrid calculations this is different, there the WAVECAR file is written at every step. This turned out to be important in the past to be able to continue within the SCF calculations if they didn't finish. Usually (at smaller structures than yours) it is beneficial.

Currently there is no option to only write the WAVECAR file at the end of the calculation. But you can manually modify the code for yourself.
In the main.F at line number 4164 (in the VASP 5.4.4 version) there is the following line:
wrtwave: IF ( IO%LWAVE .AND. ( INFO%LSTOP .OR. LTMP .OR. LHFCALC) ) THEN

Just remove the ".OR.LHFCALC". Then the WAVECAR file will be only written in the last step.

Re: Time consuming in writing wavefunction

Posted: Mon Apr 12, 2021 10:53 am
by hongyang_ma
I think this may be not an option for me because the vasp is compiled by the HPC staff in the supercomputer and it the access is shared with other vasp users. I don't have access for editing the file.. But thank you all the same.

Re: Time consuming in writing wavefunction

Posted: Mon Apr 12, 2021 1:34 pm
by alex
Hi hongyang_ma,

I'd suggest you ask the user support to make you an extra version with the changes above. The HPC staff normally is very helpfull.
And in your case the change comes with a performance plus. So they should love it.

Cheers,

alex

Re: Time consuming in writing wavefunction

Posted: Mon Apr 12, 2021 8:54 pm
by hongyang_ma
Hi Alex,

You are right. I will give it a try. Thanks.

Cheers,
Hongyang