My Community

Hi all,

I’m working on a spin–orbit coupling (SOC) calculation using VASP for an isolated organic molecule with metal, and I’ve run into some questions regarding the role of initial magnetic moments.

We don’t have a strong prior understanding of this molecule’s magnetic structure — it may or may not exhibit spin polarization, and its symmetry is low. During testing, I’ve found that changing the MAGMOM values (i.e., the initial magnetic moments) affects the final electronic state with LNONCOLLINEAR = .TRUE. and LSORBIT = .TRUE. are enabled, giving an total energy difference up to 50 meV and different magnetization projected by LORBIT = 11.

This raises a few related questions:
1. How should initial magnetic moments (MAGMOM) be interpreted in SOC + noncollinear setups?
I understand they define the starting spin direction (with SAXIS), but what exactly are they, how they play a role in the DFT equations and to what extent do they bias or constrain the final result?
2. Is there a recommended or systematic way to explore different initial conditions for molecules where the spin state is unknown?
3. In your experience, how reliable are final magnitization in SOC runs, especially when convergence is sensitive to the starting MAGMOM?

I’d really appreciate any insight or best practices from those who’ve worked with SOC in finite systems (molecules or clusters). I’m especially interested in understanding how to set this up properly without inadvertently steering the result through initial choices.

Thanks in advance!

Best,
Simon

Dear Simon,

In principle the starting values for MAGMOM should not affect the final result so much, although they might make it harder to converge if you start from a completely wrong guess. Could you post here some examples that I can use to reproduce this issue? The INCAR, POSCAR, POTCAR, KPOINTS files would be enough, but it would be great if you can also provide me with some reference OUTCAR files.

Regarding your questions:

• The initial MAGMOM is used to provide a starting point for the calculation. Since you use LNONCOLLINEAR = .TRUE. and LSORBIT = .TRUE., VASP has to converge the electronic density and the magnetisation by using Pauli's equation. This is much harder then converging just the density, so a starting point that is close to the expected behaviour can speed up convergence. In principle they should not constrain the final result, but in some cases you might end up in a local minimum that is not the ground-state.

• In principle you should do a quick test to check if your system is susceptible to changing starting points. Again, if you know more or less how your system should behave, it can really speed up the calculation.

• Calculations should be reliable within the standard precision given by DFT (5-10 meV).
  

My guess is that it will be easier to check if there's something wrong with your setup, or if this is an issue with the system, or even with VASP, if you provide me with the input files.

Kind regards,
Pedro

Hi,

Thank you so much for your response — I’ve attached a zip file containing the input files (INCAR, POSCAR, POTCAR, KPOINTS) as requested.

In my test, I observed that the final magnetization changes from roughly -0.003 (on the z-axis) when using MAGMOM = 0 on all atoms, to around 0.093 (x), 0.099 (y), 0.099 (z) when using MAGMOM = 1 1 1 uniformly. I’m trying to understand whether this level of variation is physically meaningful, or if it’s within the typical numerical noise or convergence sensitivity expected in SOC calculations.

Also, if you don’t mind elaborating a bit more on the following:
You mentioned that VASP converges magnetization using Pauli’s equation under LSORBIT = .TRUE. — could you clarify what this entails in the context of PAW and how it relates to the relativistic terms?
How exactly does SOC enter the PAW framework in VASP, and how can we verify that a given potential (e.g., for Hg) includes the required relativistic components?

I really appreciate your help — I’m trying to understand the underlying formalism better.

Best,
Simon