Hartree-Fock energy in ACFDTR calculations

[benoit]

Hello,

I tried to compare low-scaling ACFDT calculations in VASP 6.2 :
- 1st case: without specifying NBANDS in the INCAR in order to run everything in a single step.
- 2nd case: with successive calculations (DFT, exact diagonalization to compute empty states and RPA)

Both calculations give me similar RPA energies but the HF energies differ (slightly):

• 1st case:

Code: Select all

  
  HF-free energy      FHF    =       -18.48397649 eV
  HF+RPA corr. energy TOTEN  =       -30.39044900 eV
  HF+E_corr(extrapolated)    =       -30.76356554 eV
  ACFDT corr.    EXHF_ACFDT  =         0.00000000 eV (see jH, gK, PRB 81, 115126)
  

• 2nd case:

Code: Select all

  HF-free energy      FHF    =       -18.53965407 eV
  HF+RPA corr. energy TOTEN  =       -30.44612658 eV
  HF+E_corr(extrapolated)    =       -30.81924313 eV
  ACFDT corr.    EXHF_ACFDT  =        -0.00000000 eV (see jH, gK, PRB 81, 115126)
  

If I check against a separate HF calculation, I get a value very close to the 2nd case. I have attached the input files and OUTCAR.
Do you have an idea what causes this difference?

Benoit

[merzuk.kaltak]

Dear Benoit,

I have run the two jobs you posted and do not see any discrepancy with vasp-6.2.0 and the Intel-Compilers of 2020 (see attachment).
I doubt that any other compiler stack will have a problem.
Please check weather you have started the two jobs from a clean directory.

[benoit]

Dear Merzuk,

Thanks a lot for your answer. Actually the discrepancy seems to occur only when I use more than 1 OpenMP thread per MPI rank but I haven't investigated it further.
I will try to have a look at it in more details to make sure that it is not related with my compilation.

Kind regards,

Benoit