Lattice constant differ in all-in-one and step-by-step

[samuelekaputrapayong_masan]

Dear Developers,

Thank you for your continued support.
I recently performed RPA calculations with the R2SCAN orbital for the bulk polyethylene lattice constant.

The equilibrium lattice constant obtained from the all-in-one RPAR calculation differs from the step-by-step calculation by around 1 Å, which is far too much for a simple numerical error.
This difference comes from the EXX. I see that in the newest VASP version (6.5.2), there is LFOCKSTD option that will force EXX to be the same for either type of calculation.
But does this mean that my previous calculations with all-in-one RPAR are wrong?

I attached a sample OUTCAR from both calculations and the EOF plot for your reference.

I really look forward to your guidance regarding this matter.

With best regards,
Samuel Payong Masan

EOF plot: https://drive.google.com/file/d/15MxPpi ... drive_link
OUTCAR all_in_one (I have started it from converged DFT calculation so you well see only one electron step here): https://drive.google.com/file/d/13rRE5N ... drive_link
OUTCAR step by step: https://drive.google.com/file/d/1bt3tOy ... drive_link

[merzuk.kaltak]

Hello Samuel Payong Masan,

would you please share your input files that have been used for the OUTCAR files?
Without this additonal information I can only speculate.

The most obvious issue is the following.
It seems that the all-in-one calculation has read a wavefunction (ISTART =1 is found in OUTCAR_all_in_one) that most probably differs from the wavefunction used to calculate the EXX contribution in OUTCAR_step_by_step.
Starting from a fresh working directory, the all-in-one approach should yield very similar EXX energies as the step-by-step approach.
The discrepancy should be small.

For reference: LFOCKSTD does not cause such a big difference in the exact exchange contribution to the total energy (FHF) in the all-in-one RPA calculation compared to the evaluation of EXX using ALGO=Eigenval, provided the very same wavefunction has been used. Nevertheless, we suggest to switch it on, since it treats one-center terms for the exchange contribution rigorously.

A larger discrepancy can be expected by PRECFOCK; the FFT grid for exchange routines. As you see OUTCAR_step_by_step has PRECFOCK=Normal, while OUTCAR_all_in_one has PRECFOCK=fast. You can set the former for the all-in-one approach to be more consistent with the step-by-step result.
In fact, one should obtain an almost perfect match between the step-by-step EXX energy and the all-in-one approach if following tags are set in the latter:

Code: Select all

PRECFOCK=Normal
LFOCKSTD=T

[samuelekaputrapayong_masan]

Dear Dr. Merzuk Kaltak,

Thank you very much for your response. I am sorry for my very late reply.

I have uploaded input files as well as poscar and kpoints files in the following link.
https://drive.google.com/drive/folders/ ... drive_link

Just to make sure my calculation flow is correct, what I have done is as follows.
1. Perform a normal DFT calculation until requaired convergence of 1E-8 is achieved. Here I use a small number of cpu cores.
2. Copy WAVECAR to two seperate directory: RPA_all_in_one and EXX
3. Start RPA_all_in_one and EXX calculation. Here I use a bigger number of CPUs, which differ from one another.

From these steps, I got FHF from all_in_one of -488.04647587 eV and EXX of -110.52042996 eV.

I have checked again that the WAVECAR inside RPA_all_in_one and EXX are identical.

With best regards,
Samuel

[merzuk.kaltak]

Dear Samuel,

Kindly refrain from uploading proprietary POTCAR files to publicly accessible platforms such as Google Drive. These files are subject to copyright restrictions. The preferred procedure for sharing these files within error reports is to compress them into a password-protected zip archive and upload them directly to the designated forum.

Regarding the issue you encountered:
The "all-in-one" computational mode requires an exact diagonalization of the Kohn-Sham Hamiltonian, setting the number of bands (NBANDS) to the maximum number of plane waves available, prior to initiating the Random Phase Approximation (RPA) step. This prerequisite appears to be absent from your current computational workflow.
Therefore, it is recommended that you insert an additional computational step immediately following Step 1 in your established workflow. This new step should utilize the existing INCAR file, supplemented by the inclusion of the following parameters at the beginning of the file:

Code: Select all

ALGO = EXACT
NELM = 1
NBANDS = 3696

The value "3696" corresponds to the maximum number of plane waves per node, as specified at the end of the relevant line in your OUTCAR-dft file:

Code: Select all

maximum and minimum number of plane-waves per node :      3696     3638