Clarification Regarding Optical Calculations

[scanmat_centre]

We are currently performing test optical properties calculations for elemental systems such as Fe, Ti, and Al using VASP, WIEN2k, and RSPt to check for consistency. While analyzing the imaginary part of the dielectric function, we observed inconsistency in the results obtained from VASP, RSPt and WIEN2k (See enclosed figure).
Specifically, in the energy range below 3 eV, the absorptive part of the dielectric tensor calculated using RSPt shows noticeably lower values, whereas WIEN2k and VASP yield slightly higher values that are in close agreement with each other. However, beyond 3 eV, all three methods produce results that are in good agreement. But, in the case of semiconductors (Si) all three codes are agreeing well.
We have taken care to ensure that the k-point sampling, broadening parameters, and convergence criteria are as consistent as possible across all three codes. Despite this, the low-energy deviation persists between these codes.
We would greatly appreciate any insights you might have regarding this discrepancy. Could this be related to the basis set involved or the way optical matrix elements are handled between these codes?
For Optical calculations, we are unable to use the tetrahedron method (ISMEAR = -4/-5) for metals, or is there any way to use the tetrahedron method for optics?
We are attaching the input files (GGA) and the output plots for the absorptive part of the dielectric tensor for your reference.

Thanking you in advance for your time and consideration.

Best Regards

Abhishek K G
Jayendran S

SCANMAT Centre.

[michael_wolloch]

Dear Abhishek K G and Jayendran S,

Thanks for sharing your results. Indeed, integration of the imaginary part of the dielectric function using the tetrahedron method is not supported for metallic systems, so you cannot use ISMEAR -4 or -5 for these calculations.

I have no experience with RSPt at all, and my knowledge of WIEN2K is also quite limited, so I cannot really tell you why the low-energy results differ. A colleague of mine (who is very experienced with WIEN2K) told me that he has heard that good basis set construction for FP-LMTO is a lot more difficult than for FP-(L)APW, so your RSPt basis set might be to blame for the deviations of that code. However, this is just a guess.

You could benchmark your VASP results with other approaches within VASP that go beyond the independent particle picture. For example, GW calculations, BSE calculations (also check out our tutorials on BSE), or time evolution. However, all of these will be more computationally expensive than the calculations that you ran.

I will move this question to the "From users for users" section of the Forum and invite other people to comment on the issue. Are there some VASP users with RSPt experience out there? It is your time to shine now

Please feel free to ask more questions if you have practical or conceptual problems with VASP's handling of optical properties.

All the best, Michael