# SOC calculations in VASP

I have a question regarding spin-orbit coupling (SOC) calculation in VASP.

For applying SOC in any system, should tags related to SOC be included in relaxation or in the SCF cycle? If it is in the SCF cycle, then for the calculation of formation energy, should one consider the ground state energy of the system from relaxation output or from the SCF (SOC included) output?

• Does SOC = Spin-Orbit Coupling and SCF = Self-Consistent Field? Not all readers here are experts, I'm a dummy in fact but don't tell anybody.
– uhoh
Commented Aug 9, 2022 at 0:39

Sometimes the forces will be similar with and without SOC, so you can relax without SOC, and then do a single point calculation on the relaxed structure with SOC. When you do that calculation, you will be able to see if the forces from the structure relaxed without SOC are similar to the forces calculated on the static structure with SOC, and that could give you a sense for the impact of relaxing with and without SOC.

For the magnetic moments, it’s similar. Sometimes, the site integrated or cell integrated magnetization magnitude will be similar with and without SOC. Other times it won’t be. Really, you just have to check it if you are hoping that you can get away without including SOC in your system since it can vary from system to system.

I would suggest doing the following:

1. Relax the atomic positions of the initial and final states without SOC, but with collinear magnetism turned on if you have a magnetic system
2. Calculate the formation energy or binding energy by subtracting the total energies without including SOC
3. Run a single point calculation from the relaxed geometries with SOC
4. Calculate the formation energy or binding energy by subtracting the total energies from the single point calculation while including SOC
5. Compare the forces from the last relaxation step and the SOC single point calculation and also the magnitude of the magnetization between the two to see how similar they are

One reminder is that if you do a variable cell relaxation, you will want to do a single point calculation at the end even without SOC. This is because of Pulay stress, which you can read about in the VASP manual.

If you find that the forces, binding energies, or magnetizations are pretty different with and without SOC, you might want to go back to your original structure and relax it with SOC. This would probably mainly be relevant in large systems with flexible degrees of freedom (like molecules or porous structures) to make sure that you don’t end up relaxing to a different local minimum with and without SOC.

If you have a simpler geometry, like many inorganic systems that are relatively small, then you could also try doing a relaxation without SOC, then turning SOC on and continuing the relaxation if the forces in the single point SOC calculation are high. This approach can also be used with systems with more complicated geometry (and in fact might be needed for computational efficiency), but you in theory run the risk of ending up in a local minimum that is different from the local minimum that you would get if you just relaxed the original structure with SOC the whole time.

• I gave my +1 long ago, but I wanted to bring this new question to your attention: mattermodeling.stackexchange.com/q/9912/5 Do you think you can answer it? Commented Nov 2, 2022 at 16:16
• I don't think I understand the question in that post. Hopefully the other answer that you linked to in that post (the one talking about modifying the source code) will be able to help, because if the question is something like "how do I artificially vary the strength of SOC in VASP", then I do not know the answer to that
– AGS
Commented Nov 2, 2022 at 22:48
• Thanks! Yea I wasn't sure what precisely the user wanted to know (as I mentioned in my first comment), but since the user joined SE today just to ask that question, I want them to have a good experience here so that they come back more often and spread good words about us to their colleagues (we still need to grow a lot more). For this reason I wanted to try to get the question answered, and I have also written the same comment on the other post's answer. Thanks again for trying to help! Commented Nov 2, 2022 at 23:24
• I've substantially improved the question now, after learning more from the OP: mattermodeling.stackexchange.com/q/9912/5. Commented Nov 3, 2022 at 3:15