Is there any way to constrain the magnetic moments in a collinear calculation?

[xiancai]

Dear VASP developers,

I am trying to obtain the charge populations during a spin-polarized FPMD simulation. Since VASP cannot directly output the charge populations during a MD simulation, I need to extract some frames from the FPMD trajectory and perform single-point calculations for these frames. However, for the input MAGMOM values of a given frame, regardless of whether I use (1) the “magnetization (x)” values of the current frame from the OUTCAR file of the FPMD simulation, (2) the “magnetization (x)” values of the previous FPMD frame, or (3) 1.5 times (1.5×) the “magnetization (x)” values of the current frame, the output MAGMOM values from the single-point calculation always differ from those in the OUTCAR file of the FPMD simulation.

Given that the final magnetic state is highly sensitive to the initial MAGMOM values, I want to constrain the magnetic moments (as the “magnetization (x)” values from the OUTCAR file of the FPMD simulation) in the single-point calculations. I have attempted to use the I_CONSTRAINED_M tag, but it did not influence the results. I would greatly appreciate any suggestions. Thank you in advance!

Sincerely,
Hua Tian

P.S. Personally, I recommend that VASP could consider enabling the direct output of charge populations during FPMD simulations, as this feature would be useful in many situations and could save significant time and effort.

[ferenc_karsai]

In theory using the same starting MAGMOM in during an MD step and a single shot calculation should lead to the same results. However during the MD you have a different starting charge density as compared to a single shot calculation and if this strongly influences your final results it means you have many local electronic minima in which your calculation is trapped and seems to be converged.

So it's an electronic minimization problem and for electronically better converged results you need to anyway do single shot calculations with improved parameters for convergence (consider the MD as only obtaining the ionic configurations). Don't constrain the moments. Please set the following in the INCAR file for your calculations:
1 ) Set ALGO=ALL.
5 ) Set LASPH=.TRUE. to ensure that the non-spherical contributions from PAW spheres are added correctly.
6 ) For PAW potentials set LMAXMIX=6 to ensure correct mixing of the high-l components.
7 ) Reduce the mixing parameters:
AMIX = 0.2
BMIX = 0.0001
AMIX_MAG = 0.8
BMIX_MAG = 0.0001

I will forward your suggestion on having the possibilty to obtain the charge occupations at each step, but I think the reason why such quantities are not available is the huge computational overhead and the significant increase of the amount of output. Minimizing the amount of output became recently espially important due to the very long trajectories from machine learning MD calculations.

[xiancai]

Dear Dr. Karsai,

Thank you very much for your detailed reply!

Following your guidance, I have obtained some convergent results from the calculations initiated with different MAGMOM values. However, it seems that the calculated magnetization (x) and energy values are still strongly influenced by the initial MAGMOM values. The results are tabulated in the enclosed figure. Could you please review the enclosed INCAR file and point out any areas that need improvement? Additionally, are the settings you suggested (or those in the enclosed INCAR file) applicable to non-spin-polarized calculations? Thank you!

Sincerely,
Hua Tian