# Getting different total magnetic moment in 'scf' and 'vc-relax' calculations in Quantum ESPRESSO

I am performing a spin polarized DFT calculations in Quantum Espresso on a Co2MnSi ferromagnetic crystal. For this I have performed these steps:
1) Performed vc-relax calculation to find its lattice constant. Here is some lines from output file:

CELL_PARAMETERS (alat= 10.00000000)
-0.531670336   0.000000000   0.531670336
0.000000000   0.531670336   0.531670336
-0.531670336   0.531670336  -0.000000000

ATOMIC_POSITIONS (alat)
Co       0.000000000   0.000000000  -0.000000000
Mn       0.265835168   0.265835168   0.265835168
Co       0.531670336   0.531670336   0.531670336
Si      -0.265835168  -0.265835168  -0.265835168
End final coordinates


Lattice constant found is as, a = (10/0.5)*0.531670336 = 10.63340672 Bohr and the final 'total magnetization' found at the end of the calculation is as follows:

  total magnetization       =     5.00 Bohr mag/cell
absolute magnetization    =     5.39 Bohr mag/cell/cell


2) Now I have plugged the value of lattice constant, (celldm(1)=10.63340672) in scf input file and performed the 'scf' calculation. But in the output file of scf, final total magnetization is different than the vc-relax output.

total magnetization       =     4.56 Bohr mag/cell
absolute magnetization    =     5.01 Bohr mag/cell


My Question:
Correct 'total magnetization' for this system is exact 5.00 (which I have got in the vc-relax calculation) but why I'm getting different total magnetization from the scf calculation? Shouldn't it be same cause this will affect the DOS calculation also? Also the total energy from two different calculation is different. Why is so? Am I performing the steps wrongly?

Thank you!

You can find the input and output files here

• Comments are not for extended discussion; this conversation has been moved to chat. – Thomas Dec 24 '20 at 15:21

Looking at your input and output files, I think the most likely explanation is a different local minimum in the energy, as mentioned by Andrew in the comments.

If you look at the final scf step in the vc-relax output, it uses the starting magnetization from the last bfgs step, not the initial values you used at the start of the vc-relax.

I would try two things:

1. Run a standalone scf calculation with the starting magnetizations used in the final scf step of the vc-relax calculation.
2. Do another vc-relax starting from your relaxed cell parameters. You can see in your separate scf calculation that the cell stress has increased a lot. You may converge to a different geometry and ground state than your first vc-relax. You can either compare the energies from the two calculations and choose the lower, or choose one based on the magnetizations on the atoms (if applicable), i.e. whether one is closer to the spin arrangement you expect or desire to study.

Hope this helps.

• But starting magnetization is used just to break the symmetry. Does it matters what value we are putting in it? Does changing the lattice constant (as obtained in vc relaxt output ) do not reduces the force and stress on system in the 'scf' calculation? – Ujjawal M. Dec 24 '20 at 4:29
• The initial guess for the magnetic moments always has a potential impact on what magnetic solution the calculation converges to. There are many local minima, each associated with a given set of magnetic moments. Ideally, you'd converge to the lowest energy magnetic configuration and that configuration would be consistent between calculations, but there is no guarantee this occurs. Your initial guess will (potentially) alter the magnetic state found, so it does indeed matter. – Andrew Rosen Dec 24 '20 at 5:28

Edit: This no longer seems to be the issue, as the atomic positions are apparently essentially unchanged, per @KevinJ.M.'s comment. I will nonetheless keep the answer here because it is still important to consider going forward.

It looks like you are getting a different magnetization because the structures are not identical. Your first structure is fully relaxed, including atomic positions and cell volume. You then took the optimized lattice constants, but not the optimized atomic positions, for the SCF calculation. This means the SCF calculation is not the local minimum energy structure you found with the geometry optimization. You should not necessarily expect the magnetic moments to be the same. In this case, I imagine your structure in the SCF calculation has a fairly high degree of stress on it, and the fact that it's not quite a local minimum in the potential energy surface is why you are getting a close (but not identical, and non-integer) net magnetic moment. Regardless, this also explains why your energies are different – the structure isn't the same!

If you aren't happy about the relaxed atomic positions, then you need to ask yourself why. You can't just neglect to update them because they don't look nice. Perhaps you want to enforce a given symmetry. Perhaps there is something missing from the structure that stabilizes the geometry. Perhaps the geometry you think you should have is not actually real at all! These are more problem-specific questions, not a DFT question at this point.

• Does it mean if I experimentally prepare the sample(say ideally) then it's crystal structure would be same as what I have got in the final bfgs calculation? – Ujjawal M. Dec 24 '20 at 5:49
• I strongly suspected this was the issue, @UjjawalM. If you have the experimental sample then you can enforce that space group potentially, but from a calculation you cannot know for sure that is what you will find. It could be kinetically trapped or stabilized by something you are not modeling. – Tristan Maxson Dec 24 '20 at 6:04
• His final structure from vc-relax and input file structures are the same to within 0.000001 angstrom. Do you think this explains a change in the ground state energy of 0.2 eV, and the difference in magnetization? – Kevin J. M. Dec 24 '20 at 6:08
• The code detects the cubic symmetries present and enforces this symmetry during the relaxation! There is no significant atomic relaxation in this vc-relax, only cell parameters. Look at the calculated forces on the atoms. – Kevin J. M. Dec 24 '20 at 6:12
• @KevinJ.M.-- if they are that close, then no I of course don't believe it's the cause. I was just going by what UjjawalM stated. I will edit my answer accordingly. – Andrew Rosen Dec 24 '20 at 6:51