Spectral function shifted from W90 bands

[HenryScottx]

Hi there, I'm learning TRIQS and came accross one question. I've run One-Shot calculation and have plotted spectral function and W90 on the same graph, however the spectral function is shifted above approximately by the amount of my Fermi energy which I specified also in the .INP file. I've tried putting "add_mu_tb=True" but putting it or not doesn't change my result. The shape matches at different k-points, but my question was only on this "shifting".

Thanks in advance!

[the-hampel]

Hi @HenryScottx,

Let's see if we can nail down the problem you have. If I understand correctly, you subtracted the fermi level when running the w90 converter correct? So in your first DMFT iteration your chemical potential should be roughly 0.

For the post-processing:
How large is your value mu_tb that you pass to get_dmft_bands? Do the wannier bands look correct when plotting compared to a wannier90 output?

Then the script outputs always this line:

μ=15.6588 eV set for calculating A(k,ω)

Does the value you encounter there matches your expectation, i.e. corresponds to the value in the last DMFT iteration?

When adding add_mu_tb you should also see this line:

Adding mu_tb to DMFT μ; assuming DMFT was run with subtracted dft μ.

yes?

If you can provide these answers I can maybe help you more.

Best,
Alex

[the-hampel]

Hi @HenryScottx,
I played a bit around with the parameters and I think you need much more cycles to converge the calculation. I was able to run your TiO example with these solver settings:

length_cycle = 400
n_warmup_cycles = 10000
n_cycles_tot = 1.0e+7
off_diag_threshold = 1e-10
imag_threshold = 1e-6

This converged in roughly 15 iterations and I was able to obtain the following continued self-energies from MaxEnt:


with these settings:

Sigma_real_freq = sigma_maxent.main(external_path=h5_file,
                                    omega_min=-10, omega_max=10,
                                    maxent_error=0.02, iteration=None,
                                    n_points_maxent=201,
                                    n_points_alpha=60,
                                    analyzer='LineFitAnalyzer',
                                    n_points_interp=3001,
                                    n_points_final=2001,
                                    continuator_type='inversion_dc')[0]

The self-energies now look physical. However, as expected the self-energies are very flat showing almost no frequency dependence. So we expect only a static shift in the result spectral function. I compared the calculation with a Hartree soltution that gives only a static self-energy:

You can see that both spectral functions are very similar giving almost only static shifts of the Wannier bands. The shift is a bit stronger in HF (left). So it is almost overkill to do DMFT for this system. Anyway, you should be able to reproduce my results with the posted settings. Be also advised that dcdmft=True/False will influence the splitting between Ti-d and O-p orbitals a lot. This flag decides whether the DC potentital is updated during DMFT with current impurity occupations (True) or kept at the DFT level (False).

[HenryScottx]

Dear Alex and Sophie and team,
I would like to thank you very much for your prompt replies and willing to go beyond to help me solve the issue that I raised. I have noted all of the suggestions, I have increased the parameters and let it run for a large number of iterations. Since I have limited performance on my virtual machine, this will take so much time, so I thought to let you guys know in the meantime that I have appreciated your help.

I'm confident that by:

• Selecting 22 bands consisting of 10 Ti-d (above Fermi) and 12 O-p (below Fermi)
  

• Treating O-p as uncorrelated in the Wannier90 converter
  

• Using off_diag_threshold parameter
  

• Increasing the n_warmup_cycles, n_cylces_tot, iterations
  

the results will definitely improve and be similar to yours. I was running for the past few days, but my computer accidentally switched off, this is the reason of my delayed response to your previous message; so I'm running again.

Sincerely,
Henry

[dijasila]

Hello,
Can any one help me to get the spin polarized spectral function? When I use add_spin= True in w90_dict, I get this error, "AssertionError: Dimension mismatch between L and self". My system is ferromagnetic and I am doing wannier90 calculations with spinors=true..
My hope is to get the spectral functions for spin up and down case and plot it against tb bands. I would really appreciate if you can guide and help.

[the-hampel]

Dear @dijasila,

Could you please open a new discussion for this? I don't see how this is related to the topic discussed here. I am happy to help you there.

The key here is to understand, that the the option add_spin=True will add an extra spin degree of freedom for each wannier function. However, if you created your Wannier functions already with spinors=True you have already wannier functions for each spin in the wannier HR file. Hence, you don't want to add again a spin degree of freedom: set 'add_spin': False in the w90_dict when calling get_dmft_bands and set the number of Wannier functions to the correct value. Does that change anything?

Again please do not respond here, and open a new discussion. Thank you!

[dijasila]

Hi @the-hampel, I have opened another discussion thread, can you please answer my questions there?