I discovered ELK earlier this week and I've been looking at the effect of spin-orbit coupling (SOC) on bandstructures. GeTe is an example of a material where these effects are very strong. So far, I've been able to find the correct ground state atomic positions and find exactly the same increase in band gap as other calculations have reported. I'm very impressed with how easy this was to do in this code!
GeTe also shows the Rashba effect, where the top of the valance band splits at a special k-point in the material. This energy splitting has been calculated in VASP as 227 meV. In my ELK calculation though, this effect seems to be weaker and I calculated a splitting of only 64 meV. Still, the right feature does seem to be present in my band plot.
1) Does anyone have any suggestions of why this might be the case, or know which convergence parameters are particularly important to look at in this context?
2) I'm from a plane-wave background and I don't really understand what convergence parameters ELK has. In particular, I couldn't find a keyword to set the plane wave energy cut off. Is there one, or is this handled in a different way in ELK?
Thanks for reading,
Tom
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Here are the two bandstructures I should of probably included in the first place. As you can see from these plots, the VB is not splitting as significantly in the ELK calculation. The energy splitting is defined as the difference in energy between the CBM of the two separate parts of the conduction band.
in the FAQ there's this comment on convergence parameters:
7.1 Which paramemters should I check for convergence?
rgkmax, gmaxvr, lmaxapw, lmaxvr, nempty, lradstp are the most important ones. Checking for convergence with respect to the k-point mesh is mandatory, as in any DFT calculation. Depending on the mesh, swidth has to be chosen appropriately. A small swidth is necessary for accurate gaps and magnetic moments.
There is an easy way of increasing all important cutoffs to better values: Set 'highq' or 'vhighq' to true. This will give you a much slower but also much more accurate calculation.
Elk is an FLAPW code, so there are some more cutoffs than just the plane wave energy (or, equivalently, the G-vector length). rgkmax gives you this cutoff for the interstitial region, but the regions inside the muffin-tins are described with a spherical harmonics plus radial functions basis - that's what the 'A' means in FLAPW (full-potential linearized augmented plane-waves). So there are many more cutoffs that describe how this muffin-tin region is treated and all of them can be important.
In the end, a difference with VASP may remain. In PAW the spin-orbit coupling is not treated as accurately as in LAPW, so I would take the fully converged LAPW value as the reference.
Markus
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thank you very much for your reply and summary of the FLAPW method. I have found the vhighq flag and find that my results barely change when using this parameter. My belief is that this means my basis set is most likely converged.
I will take another look at the swidth parameter, as I don't believe that this parameter is affected by vhighq. Am I correct in thinking that decreasing the smearing should improve the quality of my results, as long as the SCF cycles are able to find the correct solution?
It is my understanding (perhaps showing my ignorance of VASP's SOC implementation) that the calculations I am comparing against are PW only, but using PAW to generate the norm-conserving pseudopotentials used. If this is true and my results are converged, are you saying that you believe the ELK value is more accurate?
All the best,
Tom
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
if I remember correctly, VASP applies SOC only inside the augmentation spheres. Naturally, SOC is largest there, but this might introduce some error. Also, I don't know how large the influence of using a scalar-relativistic PAW with SOC is on the final results.
So overall I think that an FLAPW calculation with SOC is more accurate, but I can not definitely comment on how good or bad the SOC in VASP is. Maybe there are more experienced VASP users here, who can comment on that.
Regards,
Markus
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi ELK users,
I discovered ELK earlier this week and I've been looking at the effect of spin-orbit coupling (SOC) on bandstructures. GeTe is an example of a material where these effects are very strong. So far, I've been able to find the correct ground state atomic positions and find exactly the same increase in band gap as other calculations have reported. I'm very impressed with how easy this was to do in this code!
GeTe also shows the Rashba effect, where the top of the valance band splits at a special k-point in the material. This energy splitting has been calculated in VASP as 227 meV. In my ELK calculation though, this effect seems to be weaker and I calculated a splitting of only 64 meV. Still, the right feature does seem to be present in my band plot.
1) Does anyone have any suggestions of why this might be the case, or know which convergence parameters are particularly important to look at in this context?
2) I'm from a plane-wave background and I don't really understand what convergence parameters ELK has. In particular, I couldn't find a keyword to set the plane wave energy cut off. Is there one, or is this handled in a different way in ELK?
Thanks for reading,
Tom
Addendum:
Here are the two bandstructures I should of probably included in the first place. As you can see from these plots, the VB is not splitting as significantly in the ELK calculation. The energy splitting is defined as the difference in energy between the CBM of the two separate parts of the conduction band.
Last edit: Thomas Durrant 2015-07-15
Dear Thomas,
in the FAQ there's this comment on convergence parameters:
7.1 Which paramemters should I check for convergence?
rgkmax, gmaxvr, lmaxapw, lmaxvr, nempty, lradstp are the most important ones. Checking for convergence with respect to the k-point mesh is mandatory, as in any DFT calculation. Depending on the mesh, swidth has to be chosen appropriately. A small swidth is necessary for accurate gaps and magnetic moments.
There is an easy way of increasing all important cutoffs to better values: Set 'highq' or 'vhighq' to true. This will give you a much slower but also much more accurate calculation.
Elk is an FLAPW code, so there are some more cutoffs than just the plane wave energy (or, equivalently, the G-vector length). rgkmax gives you this cutoff for the interstitial region, but the regions inside the muffin-tins are described with a spherical harmonics plus radial functions basis - that's what the 'A' means in FLAPW (full-potential linearized augmented plane-waves). So there are many more cutoffs that describe how this muffin-tin region is treated and all of them can be important.
In the end, a difference with VASP may remain. In PAW the spin-orbit coupling is not treated as accurately as in LAPW, so I would take the fully converged LAPW value as the reference.
Markus
Dear Markus,
Thank you very much for your reply and summary of the FLAPW method. I have found the vhighq flag and find that my results barely change when using this parameter. My belief is that this means my basis set is most likely converged.
I will take another look at the swidth parameter, as I don't believe that this parameter is affected by vhighq. Am I correct in thinking that decreasing the smearing should improve the quality of my results, as long as the SCF cycles are able to find the correct solution?
It is my understanding (perhaps showing my ignorance of VASP's SOC implementation) that the calculations I am comparing against are PW only, but using PAW to generate the norm-conserving pseudopotentials used. If this is true and my results are converged, are you saying that you believe the ELK value is more accurate?
All the best,
Tom
Dear Thomas,
if I remember correctly, VASP applies SOC only inside the augmentation spheres. Naturally, SOC is largest there, but this might introduce some error. Also, I don't know how large the influence of using a scalar-relativistic PAW with SOC is on the final results.
So overall I think that an FLAPW calculation with SOC is more accurate, but I can not definitely comment on how good or bad the SOC in VASP is. Maybe there are more experienced VASP users here, who can comment on that.
Regards,
Markus