Strange trend with large Fock exchange fractions

[guyohad]

Dear VASP developers,

I’ve noticed a very weird trend in VASP (version 6.4.2) when using large fractions of Fock exchange.

For LiF, using LHFCALC = .TRUE. with varying AEXX values, I calculated the following total energies:
For AEXX=1.0, I got E=-17.85 eV
For AEXX=0.99, I got E=-18.83 eV
For AEXX=0.98, I got E=-18.73 eV

Notice the significant drop in total energy when reducing Fock exchange by just 1% from 1.0 to 0.99, whereas the change from 0.99 to 0.98 is much smaller. When lowering AEXX further below 0.98 by 1%, the changes continue to be more gradual and reasonable.

I would appreciate any insights into whether this behavior is expected or if it indicates a bug or convergence issue.

I attach relevant input and output files.

Best regards,
Guy

[manuel_engel1]

Dear Guy,

Thanks for reaching out. What you are seeing is intended behavior. The AEXX tag has special behavior if it is set to exactly 1. The wiki mentions that

For AEXX=1.0, VASP switches off correlation by default (ALDAC=0.0, AGGAC=0.0, and AMGGAC=0.0) and thus runs a full Hartree-Fock calculation.

What this means is that you are not performing a hybrid-functional calculation anymore. Instead you are running a full Hartree-Fock calculation where the entire correlation part (that is otherwise part of the hybrid-functional calculation) is switched off. You can manually set ALDAC=1, AGGAC=1 and AMGGAC=1 to obtain results that are consistent with the other hybrid-functional calculations.

With regards to convergence, I do not see any convergence issues with your calculation. However, in general, we recommend to use ALGO=Damped or ALGO=All when running Hartree-Fock or hybrid-functional calculations.

[guyohad]

Dear Manuel,

Got it!
Thank you for the quick response.

Best,
Guy