Efficiency of the TPSSh Functional

[oliver_conquest]

Dear VASP Community

I have been calculating the relaxed structure of 40-50 atom molecules using the TPSSh functional and found calculation times to be 30-40 times longer compared to PBE and 4-6 time longer compared to TPSS (maybe this is to be expected). However, a colleague who has done calculations with identical systems using the TPSSh functional in Gaussian has found calculation times to be much faster, by approximately 10 times.

I ran some tests and found:
- Using ALGO = D converged approximately twice as fast as ALGO = A
- ALGO = D does not converge for charged systems but ALGO = A does converge (better at treating orbitals)
- Changing AMIX and BMIX does not seem to improve electronic convergence (as confirmed in the wiki)

The functional tags I used in my INCAR file are:
METAGGA = TPSS
LHFCALC = .TRUE.
AEXX = 0.1
AGGAX = 0.9
LASPH = .TRUE.

Discovered in the OUTCAR file that METAGGA = F (non-selfconsistent MetaGGA calc.) should it be set to true since TPSS is a self-consistent meta-GGA functional?

Would changing the TIME tag improve electronic convergence of a charged molecule system when using the Damped algorithm with the TPSSh functional, or is it best I stick to the conjugate gradient algorithm?

Do you have any other recommendations for reducing calculation times with the TPSSh functional?

Kind regards,
Ollie

[ferenc_karsai]

You can also try LMIXTAU=.TRUE., this sends also the kinetic energy through the mixer. Maybe this speeds up the electronic convergence.

[oliver_conquest]

Thank you for your reply Ferenc,

It seems setting the LMETAGGA = .TRUE. tag in the INCAR file changes the METAGGA setting in the OUTCAR file to METAGGA = T. Do you have any insight into whether this tag is required to be set?

I ran a test and found setting LMETAGGA = .TRUE. does not seems to change the calculation except for the tag in the OUTCAR file.

Another person in the forum had a similar question some time ago but it was not answered: forum/viewtopic.php?f=4&t=15113&p=15113 ... GGA#p15113

Kind regards,
Ollie

[oliver_conquest]

I have checked the 'main.f', 'metagga.f', paw.f and radial.f (location of the subroutine) source code files for VASP 5.4.4 and it seems the LMETAGGA tag is used as a conditional for the subroutine that calculates the metaGGA XC energy component for the core charge density on the radial grid when doing GW calculations. Since I was just doing standard SC calculations with ALGO = D or A then I'd expect setting the LMETAGGA tag either .TRUE. or .FALSE. wouldn't contribute anything extra to my calculation.

I would be interested to hear if other people know more about this or if I have missed something.

Kind regards,
Ollie