Higher SCF energies after relaxation

[mtardieux]

Dear forum members,

I've been working on calculations involving supercells and have encountered some issues.

Despite receiving the message "reached required accuracy - stopping structural energy minimization," the scf energy of my system after relaxation is higher than the one before the relaxation process.

Does anyone have any insights or suggestions regarding this discrepancy? Any help would be greatly appreciated.

Thank you,
Marie

[fabien_tran1]

Hi,

Could you please provide the input files and the OUTCAR of your calculation?

[mtardieux]

Hi, thank you very much for your answer.
Please find attached the link to the OUTCAR and the input files of my calculation.

https://www.dropbox.com/scl/fo/k69jwuj2 ... eab1a&dl=0

[fabien_tran1]

I can not access the files. Could you please create a .zip file of your files, as mentioned here: https://www.vasp.at/forum/viewtopic.php?t=17928

[fabien_tran1]

I have been able to access your files. From the free energies printed in the OUTCAR (grep for "free energy" with two blanks between "free" and "energy"), one can see that the relaxation ran properly. The free energies of the relaxation are
free energy TOTEN = -607.49277291 eV
free energy TOTEN = -620.92557732 eV
free energy TOTEN = -622.81222772 eV
As expected the free energy of the final structure is the lowest, i.e., the most negative.

[mtardieux]

Yes, indeed. Thank you.
But I've also checked the 'TOTAL-FORCE (eV/Angst)' and it looks like they are still high on certain atoms. Is there any way to decrease the forces and the total drift?

[fabien_tran1]

In order to reduce the forces on the atoms you need to optimize their positions, which is not what you did. You used ISIF=7, which corresponds to the optimization of only the cell volume (see https://www.vasp.at/wiki/index.php/ISIF). A non-zero drift may be due to numerical inaccuracy/noise, which can be reduced by increasing ENCUT, however I would not worry too much about the drift for the moment.

[mtardieux]

Actually, I also took this step after relaxing the volume, and yes, I see a decrease in the forces after relaxing the atomic positions.
However, what I did next was to perform an SCF calculation (with magnetization) using the new CONTCAR. What surprises me is that the SCF energy after the entire relaxation process is higher than the energy (also with magnetization) before the relaxation.

[fabien_tran1]

What surprises me is that the SCF energy after the entire relaxation process is higher than the energy (also with magnetization) before the relaxation.

If you want some help, then you need to provide the input and OUTCAR files (and please not via dropbox, but as a zip file attached to the message).

[mtardieux]

Yes sure, thank you very much for your help.
Here is the zip file with the scf calculation before the relaxation and the one after.

[fabien_tran1]

Indeed, the energy before relaxation is more negative which is not what one should expect. However, in order to have an idea of what is going on, I would need to see also the OUTCAR of the calculation of the relaxation process. Do you still have this calculation?

[mtardieux]

I attached the zip file. During the relaxation process I did several steps.
1) relax the volume
2) relax Mn and Sb positons from CONTCAR of 1)
3) relax Te positons from CONTCAR of 2)
4) relax the volume again from CONTCAR of 3)

[fabien_tran1]

The relaxation process is fine, since from step 1 to step 4 the energy is more and more negative:
step 1: -622.43095203
step 2: -629.27034902
step 3: -649.43636289
step 4: -649.45685423

The problem is that the two other calculations (before_relax and after_relax) were done with magnetism and spin-orbit coupling (SOC), while the relaxation process was done without, which is inconsistent. In general, one should expect magnetism/SOC to have some influence on the relaxation process, which here seems to be quite important since the initial geometry leads to a lower energy than with the final geometry.

Thus, the proper way to follow would be to relax the geometry with magnetism and SOC turned on, which is unfortunately much more expensive. A compromise may be to do the relaxation with collinear magnetism (ISPIN=2) without SOC.

[mtardieux]

Thank you for your assistance and detailed response! It is very helpful. I better understand now the potential importance of magnetism and SOC influence on the relaxation process.
I will try to perform relaxation following your suggestions