BRMIX: very serious problems the old and the new charge density differ

[NIDAL@UWM.EDU]

Hello,

I am running VASP 6.3.0 compiled with the Intel one API Base and HPC toolkits on my dual node Intel Xeon CPU with 40 cores. My system comprises of the perovskite-type LiLaTiO lattice with the composition Li5La9(TiO3)16. I have prepared the POSCAR from scratch on VESTA using the structure parameters from previous research (https://pubs.acs.org/doi/10.1021/cm0709469). I am using vaspkit 1.31.1 to generate my simulation input files:

• I am assembling my POTCAR using the following PBE pseudopotentials: Li_sv, La, Ti_pv, and O.
• I am generating KPOINTS for SCF calculation using Irreducible K-Points with Gamma scheme by applying a K-Point resolution of 0.02.
• The above step also generates an INCAR in which I have performed a few modifications by adding

  Code: Select all

  NCORE = 4

  , enabling

  Code: Select all

  ENCUT = 520

  , and

  Code: Select all

  PREC = Accurate

  .

I am running the simulation using the command:

Code: Select all

mpirun -np 40 vasp_std

, which runs into the following problem after a few steps:

Code: Select all

DAV:   8    -0.652994305258E+03   -0.28196E+01   -0.18222E+01 53640   0.129E+01BRMIX: very serious problems
 the old and the new charge density differ
 old charge density:   562.00551 new  562.16602
    0.389E+00

So, I am terminating the execution; otherwise, it would eventually run into more serious errors and fails to converge.

I am encountering this problem irrespective of my K-Point Resolution, ENCUT, ISPIN, or PREC. I even tried using various GGE and GW pseudopotentials, but the problem persists. Could anyone please help me understand what is going wrong here? I am attaching my simulation files too.

Regards,
Phani Nunna

[marie-therese.huebsch]

Hi Phani Nunna,

Thank you for submitting your question with such attention to completeness. I have had a look at your input files and wonder if you have carefully considered the use of ADDGRID = True. There are some caveats to it and it should not be used as a default. In particular, if the Fourier component of the charge density at the maximum G corresponding to the fine grid (given by NGXF×NGYF×NGZF) is finite, then the values of the charge density will alternate between positive and negative values on the ultra-fine grid (which has 8 times more points than the fine grid).

Could you please try if the problem persists without setting ADDGRID = True?

Best regards,
Marie-Therese

[NIDAL@UWM.EDU]

Hello Marie-Therese,

Thank you for the insightful explanation. I have discovered that my problem was machine specific. I could run the simulation fine without errors on other computers (I tried on an HPC cluster and my laptop). I still wanted to see if your advice would help me overcome the error on my workstation. So I commented out the line:

Code: Select all

# ADDGRID = .TRUE.

I also received a suggestion from a colleague to decrease the values of AMIX and BMIX. Accordingly, I also applied the tags

Code: Select all

AMIX = 0.2 

and

Code: Select all

BMIX = 0.6

at the end of my INCAR.

I tried applying either or both of these suggestions, but the problem persists. Then for some reason, I decided to run the VASP simulation on fewer cores. Even without performing any of the above modifications, the simulation ran fine without any errors for the following command:

Code: Select all

mpirun -np 32 vasp_std

It was still conducive to knowing what these tags could do for my simulation.

Regards,
Phani Nunna

[alex]

Hello,

have a look here: wiki/index.php/NELMDL

The idea is to have reasonably converged orbitals, where the charge is derived from. I'm using

NELMDL = -20

with very good results (and have your error only if I really messed up the POSCAR ;-) ).

Good luck!

alex