issues with convergence when setting NCORE=1

[sergey_lisenkov1]

Hello,
I need to compute a polarization (using Berry phase) of the material and this is a 2 step process. SCF calculations (NCORE=24) is converged, and then I restarted the calculations with LCALCPOL=T and NCORE=1. And despite starting from WAVECAR and CHGCAR, I ran into convergence issues:

Code: Select all

found WAVECAR, reading the header
  number of bands has changed, file:    32 present:   192
  trying to continue reading WAVECAR, but it might fail
 INCAR ok, starting setup
 POSCAR ok, starting setup
 FFT: planning ... GRIDC
 FFT: planning ... GRID_SOFT
 FFT: planning ... GRID
 reading WAVECAR
 random initialization beyond band           32
 the WAVECAR file was read successfully
 charge-density read from file: unknown                                 
 entering main loop
       N       E                     dE             d eps       ncg     rms          rms(c)
 WARNING: Sub-Space-Matrix is not hermitian in DAV            6
  1.967360880904889E-002
 WARNING: Sub-Space-Matrix is not hermitian in DAV            7
 -1.543244903816308E-003
 WARNING: Sub-Space-Matrix is not hermitian in DAV            8
 -8.001534786381991E-004

OSZICAR file:

Code: Select all

       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1    -0.164778699282E+10   -0.16478E+10   -0.16478E+10  7680   0.715E+01
DAV:   2    -0.146046009312E+11   -0.12957E+11   -0.14603E+11  7680   0.378E+02
DAV:   3    -0.826630916205E+12   -0.81203E+12   -0.82601E+12  7680   0.353E+02
DAV:   4    -0.500556256871E+11    0.77658E+12   -0.50047E+11  7680   0.410E+02
DAV:   5    -0.893897555555E+09    0.49162E+11   -0.89182E+09  7296   0.334E+02

I tried so many tweaks, but nothing helps. I use 6.5.1 version of VASP.

My input files are attached.

[max_liebetreu]

Hello Sergey,

Thank you for reaching out to us on the forum!

First off, the files you provided are insufficient to narrow down the core issue. Could you please also attach the OUTCAR?

Thank you, and best regards,

[sergey_lisenkov1]

Hi Max,

Thanks for reply. I attached OUTCAR file.

Error messages are different depending on ALGO used. When I do ALGO=Normal and NBANDS = 96 (so there would be no discrepancy between NBANDS in NCORE=24 and NCORE=1 runs), I get FOCK_ACE_ error in LAPACK ZPOTRF. When I use ALGO=All, then everything diverges.
It is strange, because the structure is large band gap dielectric.

Thanks,
Sergey

[max_liebetreu]

Hi Sergey,

We looked at your setup, and were able to achieve convergence when commenting out the parts of your INCAR related to LIBXC:

Code: Select all

LHFCALC = .TRUE.
METAGGA = LIBXC
AEXX = 0.50
LIBXC1 = HYB_MGGA_XC_R2SCAN50

Could you please try the same (commenting out these flags) on your end, and let us know if you still see any errors then?

We also noticed a few oddities, which you may or may not be aware of and which might or might not be intentional on your part:

• Your ENCUT seems very high. Unless your exchange correlation requires it, checking ENMAX in your POTCAR suggests a value of 311 should suffice.
• You are setting IBRION=2 + NSW=0, which leads to 0 structure optimization steps and should be equivalent to IBRION=-1.
• NELMDL=-16 should be 0 when starting from WAVECAR. Setting it to something else might interfere with your results.

We are still investigating your error with the LIBXC flags, and will get back to you on that.

Best regards,

[sergey_lisenkov1]

Hi Max,

Thank you. I know I can achieve convergence without using hybrid functional R2SCAN50. But this is the point - I need to use this functional.
Also, I'm confused why number of bands increased four times, when NCORE is set to 1. Is it normal?

Thanks,
Sergey

[max_liebetreu]

Hi Sergey,

It is possible something is off with your previous run. In order to isolate the problem, could you please try adjusting some of your INCAR tags:

Code: Select all

LHFCALC = .TRUE.
METAGGA = LIBXC
AEXX    = 0.50
LIBXC1 = MGGA_X_R2SCAN
LIBXC2 = MGGA_C_R2SCAN
ENCUT = 400

and then restart your run without reading WAVECAR and CHGCAR, from scratch?

Best regards,

[sergey_lisenkov1]

Yes, it worked if I set ENCUT=400 and start from scratch (no wavecar/chgcar were used)

[max_liebetreu]

Hi Sergey,

In that case, we suggest you simply continue your computations with these settings, which should be equivalent to what you have used:

Code: Select all

LHFCALC = .TRUE.
METAGGA = LIBXC
AEXX    = 0.50
LIBXC1 = MGGA_X_R2SCAN
LIBXC2 = MGGA_C_R2SCAN

You can set the ENCUT back to 800, but that should not make a difference.

Best regards,