Oscillating Energy for Ionic Relaxation - HSE functional

[shaun_tay]

Dear VASP community,

I am currently running HSE functional calculations for NaFeFe(CN)6 using Fe_pv, with the specific defined spin states of 3.8 and 5.0 for the first and second iron sites, respectively.

Initially, I optimized my structure using the GGA functional with the Gamma point, and this step was successful. However, when I switched to the HSE functional (without using the previous WAVECAR), the SCF calculations began to oscillate back and forth during the first ionic step and failed to converge.

Despite troubleshooting the issue by setting LREAL to Auto as suggested in the OUTCAR file, the problem still persists. I have attached my INCAR and input files for your reference.

I would greatly appreciate any advice or guidance on how to resolve this issue. Thank you all

INCAR
ALGO = Normal
EDIFF = 1e-05
EDIFFG = -0.02
ENAUG = 1360
ENCUT = 680
GGA = Pe
HFSCREEN = 0.2
IBRION = 2
ISIF = 3
ISMEAR = 0
ISPIN = 2
LAECHG = True
LASPH = True
LCHARG = True
LDAU = False
LELF = False
LHFCALC = True
LMAXMIX = 4
LMIXTAU = True
LORBIT = 11
LREAL = False
LVTOT = False
LWAVE = False
MAGMOM = 4*0.6 4*3.8 4*5.0 48*0.6
NELM = 200
NSW = 99
PREC = Accurate
PRECFOCK = Fast
SIGMA = 0.2

[henrique_miranda]

Hard to say what is the reason for this.
I don't see anything obviously wrong in your INCAR.

You say you relaxed the cell with GGA for the Gamma point, but in your HSE calculation you have a 2x2x2 k-point sampling.
Would running a Gamma point only calculation with HSE make sense as well?

Maybe you can try setting `PRECFOCK=Normal` instead of `PRECFOCK=Fast`
The advice to use LREAL=Auto is probably related to the relatively large size of your cell, but it is unlikely to affect the electronic convergence.

[shaun_tay]

Hi Henrique,

Thank you for the tip!

You say you relaxed the cell with GGA for the Gamma point, but in your HSE calculation you have a 2x2x2 k-point sampling.
Would running a Gamma point only calculation with HSE make sense as well?

That makes sense. However, I was able to run the structure with a 2 x 2 x 2 k-point sampling so I suspect that k-points is not the issue.

Maybe you can try setting `PRECFOCK=Normal` instead of `PRECFOCK=Fast`

The calculation did not seem to converge and is slower than `PRECFOCK=Fast`.

However, upon setting ALGO = Normal, IMIX = 1.0, and AMIX = 0.4, the energy values no longer oscillate but instead increase to a very large positive value. Do you have any advice on how I can approach this problem? Thank you!

[henrique_miranda]

You are using the Davidson to diagonalize the Hamiltonian (ALGO=Normal) and then mix the charge density, you might consider using conjugate gradient (ALGO=All) with TIME=0.05.
https://www.vasp.at/wiki/index.php/Trou ... onvergence

It is not guaranteed that it will work, but worth a try.

[shaun_tay]

Hi Henrique,

Thank you for your guidance!

You are using the Davidson to diagonalize the Hamiltonian (ALGO=Normal) and then mix the charge density, you might consider using conjugate gradient (ALGO=All) with TIME=0.05.

I was able to make some progress on convergence by referring to your comments and the website wiki/index.php/Troubleshooting_electronic_convergence. Despite these adjustments, convergence is still stuck in the first ionic step and remains slow.

I have tested adding a mixing of 25% Fock exchange, LHFCALC = TRUE, ISMEAR = 0, and SIGMA = 0.05. There were improvements where I achieved the 2nd ionic relaxation step. However, the first ionic step required 136 electronic steps to proceed to the second step which indicates slow convergence. May I have your advice on improving the convergence problem and speeding up the calculation? Thanks!