How to get metal oxide slab converge to specific AFM configuration

[JohnnyTsien]

Hi, I'm doing catalytic reactions on a magnetic metal oxide surface. For bulk material, it's AFM, and my calculation results is consistent with experiment and repeatable.

But when it comes to slab calculation, I was trying to set MAGMOMs to make it AFM as well. The problem is surface atoms have uncompensated spins, so the magnetic configuration is not exactly the same as that in the bulk materials. I tried several different sets of MAGMOMs, they all converged... but to different energies of course.

I guess I should choose the configuration with lowest energy as some literature reported. But the difference between different converged AFM configurations is subtle, while the energy difference can be as large as 0.6 eV.

As I said, I am doing chemistry on the surface, so the molecular adsorption energy is important to me.

My question is how can I insure the slab converge to similar magnetic configuration when where is some adsorbate on the surface? I mean otherwise I will get an energy difference with a magnetic effect, but not purely adsorption energy.

Any discussion and advise would be highly appreciated, I've struggled in this problem for weeks.

Thanks,
Johnny

PS: What I have tested are colinear magnetic structures

[Danny]

Maybe something to consider is the fact that your adsorbate might change the magnetic configuration of the surface, in reality, which should be represented in your calculation. In that case the change in energy of the system due to changing magnetic configuration could be seen as part of the adsorption energy.
If you don't want the magnetic effect include you might want to work the other way around: start from your slab+adsorbate and then calculate the slab alone with the magnetic configuration obtained in your slab+adsorbate calculation...this however might not be the groundstate of your slab, but then again that is not what you are looking for.

Cheers
Danny

[forsdan]

When you have a lot of competing magnetic configuration, there is usually not much to do except trial and error. Play around with MAGMOM, the mixing parameters, the convergence algorithm (including increasing NELMDL), and explore which kind of configurations that are available. Changing the k-point grid can also change the convergence behaviour. I have also notice that changing the smearing can help in certain cases (once the solution has converged to the specific configuration you just switch back to the old smearing starting from the converged results). So you have to figure out which settings that work for your specific system.

Regarding the adsorbate-surface interaction, the worst case scenario is that the adsorbate influence the magnetic configuration is such a way that convergence to the AFM configuration inside the slab can't be recovered for your slab thickness (or it will be extremely difficult). Adding more layers can then be a possible option.


Cheers,
/Dan

[JohnnyTsien]

Thank you Danny and forsdan! If I understand your comments correctly, the adsorption energy should be the ground state of surface-adsorbate minus the ground state of the separate slab and molecule. So I have to play around MAGMOM to find the ground state by hand?... I mean, in practice it could be difficult, since there are a lot of magnetic configurations, it could be finite number, but I don't know how could I look into every case exhaustively.

I am confused about the roles of mixing parameters, algorithms and k-point mesh. If I can get the system converging quickly, should I still play around with these parameters? Or can they affect the final converged state?

What's more I was just using colinear settings, if I use non-colinear calculation, I guess I would get even more competing state?

Or am I in the wrong track, or should I just set the all the MAGMOMs to a little larger positive values and make it converge to FM state. That's what we get if no initial magnetic moments are set by user.

Again, thank you very much!

[forsdan]

The general procedure is:

i) Find the ground state of the slab, where the slab is sufficiently thick so that the bulk configuration is recovered in the middle of the slab.

ii) Find the ground state of the free adsorbate.

iii) Find the ground state of the slab+adsorbate, where the slab is sufficiently thick so that the bulk configuration is recovered in the middle of the slab.

If iii) requires additional layers, the same number of layers should be used in i). Then you evaluate the adsorption energy as the difference between slab+adsorbate and the separate slab and molecule calculations.

To find the ground state IF several local competing minima (corresponding to different magnetic configurations) are presence is a matter of patience and trial and error. IF such a situation is presence you can first play around with the MAGMOM tag (which specifies an initial guess for the magnetic configuration). This can help in many situations. However, this is not always sufficient. Changing in particular the mixing parameters and the diagonalization algorithm can then be an option, since they change the way the density is updated during the self-consistency loop. In this way the convergence can occur to another local minima (i.e. another magnetic configuration). NOTE: The convergence to the global minima is never garantueed to 100% for spin-polarized calculations unless there is only one well-fined minima.

So the bottom-line is yes, the settings can effect your final magnetic configuration if there are several competing minima.

Regarding the non-collinear calculations: it will depend on your system and the size of your supercell. The latter must be able to resolve the non-collinear configuration in that case. But in theory you can have even more possible configurations if you allow for non-collinear calculations (the difference in energy may however be small). I would however just concentrate on the collinear calculations unless there is evidence that non-collinearity is relevant for your system.

So if your system seems to exhibit a lot of different configurations you should explore a sufficent number of different possibilites so that you are confident that you have obtained the ground state of both the slab+adsorbate as well as the separate slab. If this takes a lot of time is another issue, but it's what should be done in my opinion.

In principle there is also always a possibility to do as Danny suggest above and evaluate the adsorption energy for the same magnetic configuration (not necessarily the ground state) to get an estimate, but then one should be confident that the neteffect of the change in magnetism is the small or neglectable.

Does the above clarify my response?

/Dan






<span class='smallblacktext'>[ Edited Wed Apr 07 2010, 04:48PM ]</span>

[JohnnyTsien]

Thank you for your help!