Helps needed for reproducing the absorption spectrum

[kle2797]

Dear developers,

I have been trying to reproduce the absorption spectrum of rutile. So I was following the parameters in the following paper, specifically figure 3, but could not get a similar one.

To be consistent with the paper, I have checked the k-grids (8x8x12), the cut-off for V (BSENGexx) and W (BSENGBlk) and used 1 valence, 1 conduction bands for the BSE at Q=1. My result only got a strong peak near 4 eV with a shoulder on the lower energy, which I could resolve a bit better by lowering the Damping range (BDmRange) but that is not nearly the same.

I also did other calculations with more bands and larger cut-offs, but the results seem not changing much, so I think it could be some other parameters. Could you please help me understand why this happened? I have attached the input, the report and the output here. Hope that will help.

Best,

[Daniele Varsano]

Dear Khoa Le,
in order to reproduce the results of Kang and Hybertsen you should be sure to have the same geometry and pseudo-potential (at least same accuracy). To my experience the rutile spectrum is rather sensible to the used pseudo-potential, so please check if you have semi-core electrons etc..
Note also the that GW correction in the paper you indicate has been calculated with a contour deformation technique.

Anyway, besides these details, the main problem here is that you are using a slepc iterative solver calculating 100 eigenvalues/eigenvectors. Then the spectrum is calculated using this number of states and this is the reason why you cannot see the tail of the spectrum.

You can either use a full diagonalization, BSSmod="d" (you have a small matrix so this is not a problem), or, if you are not interested in the eigenvectors you can use the haddock procedure BSSmod='h'.

Next verify also the k points sampling and the number of bands entering in the BSE as from the paper it is possibile they used a 12x12x18 grid and a number of bands such that Ec-Ev ~ 15 eV

Best,
Daniele

[kle2797]

Thank you for the reply,

I see the that fully diagonalize the Hamiltonian gave a much better spectrum.

Next verify also the k points sampling and the number of bands entering in the BSE as from the paper it is possibile they used a 12x12x18 grid and a number of bands such that Ec-Ev ~ 15 eV

I will make sure to do these checks. Also, in order to include only Ec-Ev ~ 15 eV, is setting the variable "BSEEhEny" to:

Code: Select all

% BSEEhEny
  0.00000 | 15.00000 |         eV

the correct way to do so?
Best,
Khoa

[Daniele Varsano]

Dear Khoa,

yes you can use the variable BSEEhEny, this will select all the transitions in the range compatibles with the BSE bands included.
Please note that the namelist needs to be closed with "%" symbol.

You can also look at the KS energies (reported both in QE and Yambo report) and see how many bands you need to include to reach that max Ec-Ev energy.

Best,
Daniele