Re: ICORELEVEL Ti L-edge spectra

[joebiv98]

Hi,

I am trying to perform a series of Ti L-edge XAS calculations for SrTiO3 using the ICORELEVEL module available in VASP using an HSE06 xc functional. Since spin-orbit coupling (and therefore multipole splitting) is important when simulating the Ti L-edge, I believe only ICORELEVEL=1 (the initial state approximation) can be used. I am using the GW POTCAR files. I've also observed that the output spectra strongly changes every time I perform the calculation, regardless of initial configuration, and do not match well with a previous report: https://journals.aps.org/prresearch/abs ... h.5.013199. Would someone be able to help me with identifying what could be improved here?

Here is my input file:

LHFCALC= .TRUE.
AEXX = 0.25
HFSCREEN= 0.2
ALGO = All
PRECFOCK = FAST
HFLMAX = 4
LDIAG = .TRUE.
LSUBROT = .TRUE.
ENCUT = 520
NELM = 200
EDIFF = 5E-6
PREC = Accurate
LWAVE = .TRUE.
LASPH = .TRUE.
LREAL= .FALSE.
LMAXMIX = 4

ISMEAR = 0
SIGMA = 0.05
LORBIT = 11
ICORELEVEL = 1
CLNT = 2
CLN = 2
CLL = 1
CLZ = 1.0
CH_LSPEC = .TRUE.
CH_SIGMA = 0.001
CH_NEDOS = 2500
LSORBIT = .TRUE.
NBANDS = 208

NCORE = 4

Additionally, I wanted to run a benchmark and compare to a GW+BSE core level calculation for the Ti L-edge as was done in the attached report. I haven't been able to find any documentation for how to approach these types of calculations via VASP, though I have been able to find one report doing so for some K-edge spectra: https://arxiv.org/abs/2206.11544 I would appreciate any assistance on this end too. Thank you!

[joebiv98]

I realize I did not attach the input files. I have attached them here.

Input Files.zip

171eaba9d4ec814235ce64ce6ce5c2d8.png

The spectra was convolved with a Lorentzian during post-processing. The main issue I'm encountering is that the spectra changes significantly each time I run the calculation and it does not properly capture spin-orbit splitting of the L2,3 edges. This serves as a benchmark for some other systems I am studying (which I cannot disclose), but I am encountering the same issue there too.

[andreas.singraber]

Hello!

Please apologize this very late answer... since this is a rather complex question I will have to discuss this with my colleagues, please stay tuned...

Best,
Andreas Singraber

[alexey.tal]

Dear joebiv98,

Let me first address your question about the spin-orbit coupling. In VASP the spin-orbit coupling for the core states is not supported. But one can introduce the splitting in the post-processing. It is a common issue when the core is frozen (for example see this paper on XAS with pseudopotentials 10.1103/PhysRevB.107.205148)

If you are using the initial state approximation, you don't account for the excitonic effects and you should compare your result to the IP spectrum in Fig. 4 of the reference (10.1103/PhysRevResearch.5.013199). Actually, the SOC can be included in the final state approximation (ICORELEVEL=2), so I'm not sure I understand why you choose the initial state approximation.

I've also observed that the output spectra strongly changes every time I perform the calculation, regardless of initial configuration, and do not match well with a previous report:

Could you show an example of this change? If the spectrum is different every time you run the calculation, it could mean that the calculation is not properly converged.

Additionally, I wanted to run a benchmark and compare to a GW+BSE core level calculation for the Ti L-edge as was done in the attached report. I haven't been able to find any documentation for how to approach these types of calculations via VASP, though I have been able to find one report doing so for some K-edge spectra: https://arxiv.org/abs/2206.11544

The BSE code for the core-states has not been release yet. But the splitting due to the SOC in the core states is not included in the BSE code either.