different mBJ result on a cubic system.

[yujia_teng]

Dear developer,
I'm working on a cubic system with VASP 6.5.1. The (110) and (010) direction of cubic should be equivalent, so if spin points along these two directions, the result should be same. In GGA result, indeed it's same. However, the resulted mBJ band and fermi surface is different between these two cases. The electronic step convergence behavior is also different. I tried my best but cannot figure out where I did it wrong. The relevant input/output files and figures are attached.

I use primitive cell, set spin points along (110) and (010) Cartesian direction with SAXIS, and symmetry is turned off so all k points are listed (This is unnecessary because I was misled by external tools). I also checked with symmetry on with new KPOINTS_OPT method, and nothing changes. For spin along (101) direction, it's harder to converge. It cannot converge with default DAV algorithm, and must use ALGO=Damped, which takes about 500 steps to converge. The attached file just include the last step after repeating reading WAVECAR to restart. But for (110) direction, it converges well with default DAV algorithm (using ALGO=Damped also takes many steps, and ALGO=All gives wrong result although converges faster).

Best,
Yujia

[fabien_tran1]

Hi,

I have repeated your calculations, but with different parameters (e.g., ALGO=All and a different k-mesh). I can see that the results with the two different SAXIS can slightly differ depending on the functional (I compared the total energies and magnetic moments, but did not plot the band structures). This difference is basically zero with LDA, but larger with GGA and MGGAs (PBE, R2SCAN, MBJ, ...).

The main reason is probably the sensitivity of the results on details of the initial conditions for starting the calculations. For some technical reasons the calculations with different SAXIS may possibly start with different initial wavefunctions and density, leading at the end to slightly different results. This is not so surprising in non-collinear calculations with f-electrons, since such systems may have an electronic-configuration space with multiple shallow minima close in energy. Anyway, the differences that you obtained for the band structures seem small and probably not important.

I think that you did not make any real mistake, but your two calculations are not very clean since they were done with different ISMEAR and ENCUT, which may increase the difference. In general, try to have clean INCAR files. Furthermore, as already mentioned in a previous topic, the comments in INCAR should be commented out with #, otherwise they may not be considered as comments and perturb the reading of the INCAR. For instance, this should be

Code: Select all

LORBIT=11                             #for band structure, set to 12

[yujia_teng]

Dear fabien,
Thanks for the explanation. It's very clear and there're no more questions. And since a previous topic, I've paid more attention to simplify INCAR and use correct comments of new calculations. I should also keep consistent parameters.

Best,
Yujia