Incorrect pressure at energy minimum with rVV10

[aaron_kaplan1]

This concerns calculations with meta-GGAs + rVV10. Sample input/output files using SCAN + rVV10 to determine the equilibrium geometry of graphite are attached.

When relaxing the cell directly (ISIF = 3), VASP successfully finds a zero pressure state. However, if an equation-of-state fitting is performed first, followed by an ISIF = 4 relaxation (to refine the equilibrium geometry), my collaborators and I are finding that:

1. There is a substantial external pressure at the true energy minimum (-9.15 kB in the attached files)
2. The direct relaxation with ISIF = 3 gets close to the true energy minimum, but overestimates the minimum (the difference is small, about 0.002 eV)

For calculations with SCAN alone, the agreement is much closer, with zero pressure at the true energy minimum. We've found the same trend when using other meta-GGAs with rVV10. We can't test this with GGAs + rVV10, as this isn't supported currently in VASP, though please correct me if I'm wrong.

ENCUT is set sufficiently high (1200 eV) to minimize Pulay stresses during the ISIF=3 relaxation, and there should be negligible Pulay stress from ISIF = 4, since the cell volume is held fixed there.

It's also worth noting that if the final geometry of the ISIF = 3 calculation is used as input to an ISIF = 4 relaxation, this still finds a zero pressure state near the energy minimum, but not the true energy minimum.

Could this be a bug in the stress tensor elements calculated with rVV10?

[martin.schlipf]

Thank you for reporting this to us, I could reproduce the behavior you described and will investigate its origin.

[martin.schlipf]

The bug is now confirmed and we are working with our collaborators to fix it. We added a warning to the vdW-dF section that stress in rVV10 is broken.

[john_low1]

The comment added to vasp documentation states:

"NOTE: As of vasp.6.2 (and prior version) the stress tensor seems to be broken for rVV10. It is correct for other vdW-DF though."

I thought the version numbers of vasp distributions have three digits. The most current version is 6.2.1. The above comment implies that the RVV10 stress tensor in vasp 6.2.* and vasp 6.1.* is broken.

Does this mean that the RVV10 stress tensor in vasp.6.1.* is broken as well?

I believe that VASP.6.0 was a beta version which we are not currently allowed to use under our license agreement.

VASP 5.4.4.pl2 is the only version prior to 6.1.* available for download. Could someone confirm that the stress tensor for rVV10 is not broken in VASP 5.4.4.pl2?

I apologize for my confusion.

Thank You!

John Low
Argonne National Laboratory

[martin.schlipf]

This bug will be fixed in the next VASP release. We still need to run some tests on it, to make sure that nothing else is broken.

All earlier version will not have this fix so you cannot use them for that particular combination.