Dft_tools INP file

[AQuantumGuy]

Dear TRIQS community,
I am stuck for my DMFT calculation. I am working with Titanium Dioxide (rutile phase) TiO2 and I've successfully run my calculations for Quantum Espresso and Wannier90.

For Wannier90 I've calculated and plotted 22 Wannier bands which comply very similarly to QE results. I'm focusing on Ti-d and O-p bands both above and below the fermi level. Now I'm trying to write to INP file to run my calculations on TRIQS, but I'm not sure how to write it. I've seen from the documentation page, but I'm still confused, I'd appreciate if someone can help me write it.
The grid I've run my calculations is 14 14 14

Thank you for any help!

[phibeck]

Hi,

could you please paste what you have so far for the seedname.inp file, together with the projections card of your Wannier90 input? Just as a quick summary: In seedname.inp you can choose among those projections defined in your Wannier90 input. However, if you decide not to include any of those states in your DMFT calculations, they must be defined last in the Wannier90 projections.

[AQuantumGuy]

Hi,
I actually don't know exactly how to construct the inp file. I know we describe the "specification of the k-mesh", "electron density", "number of atoms", "atom, sort, l, dim, SO, irep", but I'm having troubles to understand what shall I input in these.

So far I understand that I shall input " 0 14 14 14" at the beginning.

My .win file code is:

`begin unit_cell_cart
ang
-0.0000000000 -0.0000000000 2.9692030000
4.6532720000 -0.0000000000 0.0000000000
0.0000000000 4.6532720000 0.0000000000
end unit_cell_cart

begin atoms_cart
ang
Ti 0.5000000000 0.5000000000 0.5000000000
Ti 0.0000000000 0.0000000000 0.0000000000
O 0.5000000000 0.1954200000 0.8045800000
O 0.5000000000 0.8045800000 0.1954200000
O 0.0000000000 0.3045800000 0.3045800000
O 0.0000000000 0.6954200000 0.6954200000
end atoms_cart

! system
num_wann = 22
num_bands = 70
mp_grid 14 14 14

! job control
iprint = 2

! plotting
wannier_plot = false
wannier_plot_supercell = 3
bands_plot = true
bands_num_points = 100
bands_plot_format = gnuplot
write_hr = true
write_u_matrices = true

! disentanglement
dis_win_min = 0
dis_win_max = 20
dis_num_iter = 1000
dis_conv_tol = 1.0E-12

! wannierisation
num_iter = 1000
conv_window = 5
conv_tol = 1e-12

! pp tools
kpath = false
kpath_task = bands
kpath_bands_colour = spin
kpath_num_points=500
begin kpoint_path
G 0.00000 0.00000 0.00000 X 0.50000 0.00000 0.00000
X 0.50000 0.00000 0.00000 M 0.50000 -0.50000 0.00000
M 0.50000 -0.50000 0.00000 G 0.00000 0.00000 0.00000
G 0.00000 0.00000 0.00000 Z 0.00000 0.00000 0.50000
Z 0.00000 0.00000 0.50000 R 0.50000 0.00000 0.50000
R 0.50000 0.00000 0.50000 A 0.50000 -0.50000 0.50000
A 0.50000 -0.50000 0.50000 Z 0.00000 0.00000 0.50000
end kpoint_path

kslice = false
kslice_task = fermi_lines
fermi_energy = 8.6430
kslice_2dkmesh = 200 200
kslice_corner = 0.00 0.00 0.00
kslice_b1 = -0.25 0.25 0.25
kslice_b2 = 0.25 -0.25 0.25

dos = true
dos_project = 1
!dos_energy_min = 8.
!dos_energy_max = 13.
dos_kmesh = 50
!dos_adpt_smr = false
!dos_smr_type = gauss
!dos_smr_fixed_en_width = 0.05

begin projections
Ti:d
O:p
end projections

begin kpoints`

[phibeck]

Please read this carefully https://triqs.github.io/dft_tools/latest/guide/conv_W90.html I do not know what you want to compute in the end. Based on your Wannier90 input, possible options are the following:

0 14 14 14 # specification of the k-mesh
some float # electron density (this is the filling you want to have. Probably partially filled Ti, perhaps you want to add oxygen?)
2 # number of atoms (or 6, if you want to add oxygen)
0 0 2 5 0 0 # atom, sort, l, dim, SO, irep (d-shell corresponds to l=2, all 5 orbitals, so dim=5)
1 0 2 5 0 0 # atom, sort, l, dim, SO, irep (I assume here that this Ti atom is symmetry-equivalent to Ti atom 1 (sort=0), verify!)
*** the next four lines you might add if you want to include the oxygen p states
2 1 1 3 0 0 # atom, sort, l, dim, SO, irep (p-shell corresponds to l=1, all 3 orbitals, so dim=3)
3 1 1 3 0 0 # atom, sort, l, dim, SO, irep (I assume here that all O atoms are symmetry-equivalent to O atom 1 (sort=1), verify!)
4 1 1 3 0 0 # atom, sort, l, dim, SO, irep
5 1 1 3 0 0 # atom, sort, l, dim, SO, irep

Please specify if anything is unclear.

[AQuantumGuy]

Hi, just tried updating the script for plotting, but I still get the same result. Like I said, the band lines perfectly resemble W90 ones, but are scaled smaller and I can't find out what might be the reason. Any suggestion would be appreciated!

Sincerely,

[phibeck]

Okay. For the updated plotting script you should have encountered some issues that have to do with renaming variables. Can you confirm you encountered these so we can be sure the updating of the plotting script worked?

Could you please share an image of the self-energy you used (should be on the real frequency axis), and remove the tag w_mesh in the dictionary that contains the information about the self-energy, then try again?

[AQuantumGuy]

Hi, I just realized that it did not update the file and I've just manually updated it and as Alexander stated, I now do get error stating missing argument "mu_tb". Whats the best way to proceed on this?

Sincerely,

[the-hampel]

Yes now you have the newest version. You have to change in the w90_dict the name of the paramter mu to mu_tb. Also there are some other changes. The function now returns only 3 values. Please remove the dft_mu output value. You can find a working example here: https://triqs.github.io/solid_dmft/tutorials/correlated_bandstructure/plot_correlated_bands.html . If the result looks still weird, consider adding the paramter add_mu_tb=True to the get_dmft_bands function call.

[AQuantumGuy]

Hi, tried following the advice, I haven't got the plot, but now I receive the following error:
Number of Wannier orbitals: 22 and self-energy target_shape (1, 1) does not match

The self-energy graph is shown here and
here
From my W90 calculations, I do obtain 22 bands.

Sincerely,

[phibeck]

Hi, okay so this clarifies the issues better. Currently, there seem to be at least two problems:

1. your self-energy is computed for a single orbital, while the program expects a 22x22 block for this to work. Did you do your DMFT calculations for all 22 bands or only a subset of these? The self-energy needs to have the same shape as the number of Wannier functions. So in case you have multiple blocks or impurities, they will have to be merged. We can discuss how this can be achieved, but we first need to know what you did for the DMFT calculations. Perhaps you could upload the output of your DMFT calculation, that would be helpful.
2. the plot of your self-energy is a Matsubara self-energy, i.e. on the imaginary axis. Did you use analytic continuation? I suppose the answer is yes, otherwise the code should have triggered another error. Could you show a plot of the real-frequency self-energy?

[AQuantumGuy]

With old plot_correlated_bands.py and NOT putting Fermi energy in seedname.inp file, I get this new error when plotting spectral function:
Operands could not be broadcast together with shapes (1000, 5, 5) (10, 10)

[the-hampel]

Which part of constructing a larger BlockGf is not clear? Maybe this helps: Imagine you want to create a BlockGf S_w_upfold with two blocks, each having a target_shape of 5x5 (or whatever number of wannier orbitals you have in a wannier90_hr.dat) and inheriting the mesh of a continued self-energy S_w_cont with shape 2x2. Let's first create the large target BlockGf:

prototype = GfReFreq(mesh= S_w_cont.mesh, target_shape=[5,5])  # creates an 0.0 initialized Gf object with shape 5x5
S_w_upfold = BlockGf(name_list = ['up_0', 'down_0'], block_list = [prototype, prototype], make_copies = True)

Let's assume in our example that the first 2x2 block of up/down should be diagonal and degenerate holding information from S_w_cont from the first impurity of your DMFT calculation matching the first two orbitals in your wannier90 setup. We can fill the first 2x2 blocks of S_w_upfold with the following lines:

for block, gf in S_w_upfold:
    S_w_upfold[block][0,0] << S_w_cont[block][0,0]
    S_w_upfold[block][1,1] << S_w_cont[block][1,1]

Now the first 2 orbitals in S_w_upfold are filled with the continued self-energy and the rest is 0.0. If you have a second impurity that is equivalent to the first and has not been treated by solid_dmft, i.e. the next 2 wannier orbitals are the same but belong to a different site, you can also duplicate the self-energies to these obitals:

for block, gf in S_w_upfold:
    S_w_upfold[block][2,2] << S_w_cont[block][0,0]
    S_w_upfold[block][3,3] << S_w_cont[block][1,1]

The fifth orbital of S_w_upfold in this example is empty and can represent uncorrelated wannier states in your system.

This is exactly what you need to do, but just with different number of orbitals. Additionally you might need to subtract some double counting from some orbitals. The get_dmft_bands function will automatically subtract the DC from DMFT for ALL orbitals in Sigma. So you might need to add it to some of your orbitals.

I hope this helps a bit. Please let me know which part of creating a larger self-energy is unclear. I will also update you here once I fixed the script to load all information from the archive automatically. For now this has to be done manually.

• Alex

[AQuantumGuy]

So from what you explained, first you created a BlockGF with two blocks both of size 5x5. Then you focused only on the first block and you filled diagonally until the 4th orbitals and leaving 5th as zero.

Since the error I get is "Number of Wannier orbitals: 10 and self-energy target_shape (5, 5) does not match.", I need to create a target_shape (10, 10), right?

When I run "Sigma_imp_iw" I get the following (WHEN I DON'T PUT FERMI ENERGY in seedname.inp file)

Green Function Sigma_freq_0 composed of 2 blocks: 
 Greens Function Sigma_freq_0_up_0 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (5, 5): 
 
 Greens Function Sigma_freq_0_down_0 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (5, 5): 

From what I understand, I do need to construct a 10x10 GF where in the diagonal entries I place the analytically continued 1x1 self-energies for each orbital of all my states. When I run "Sigma_imp_iw" I get the following (WHEN I PUT FERMI ENERGY in seedname.inp file)

Green Function Sigma_freq_0 composed of 10 blocks: 
 Greens Function Sigma_freq_0_up_0 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (1, 1): 
 
 Greens Function Sigma_freq_0_up_1 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (1, 1): 

 Greens Function Sigma_freq_0_up_2 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (1, 1): 
 
 Greens Function Sigma_freq_0_up_3 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (1, 1): 

 Greens Function Sigma_freq_0_up_4 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (1, 1) : 
 
 Greens Function Sigma_freq_0_down_0 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (1, 1): 

 Greens Function Sigma_freq_0_down_1 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (1, 1): 

 Greens Function Sigma_freq_0_down_2 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (1, 1): 

 Greens Function Sigma_freq_0_down_3 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (1, 1): 

 Greens Function Sigma_freq_0_down_4 with mesh Matsubara Freq Mesh of size 502, Domain: Matsubara domain with beta = 10, statistic = Fermion, positive_only : 0 and target_shape (1, 1): 

So you can see that I do actually get the GF that you guys suggested (when putting Fermi energy) and plotting using the previous plot_spectral function code. Just a reminder, when plotting the spectral functions, these are a bit different from W90 ones. Example, here is the result when I do only 3 dmft_iteractions.

[the-hampel]

This GF has 10 1x1 blocks instead of 1 10x10 block. I do not see what the Fermi flag in the w90 converter has to do with that? btw. you can use it either way and specify this later by passing to get_dmft_bands the following flag add_mu_tb=False/True . Anyway, it seems everything is now working? Or do you have a question?

[AQuantumGuy]

I understand, but considering still where I need to modify my GF to make it 10x10GF. I do need to construct a 10x10 GF, however since I have a 5x5 block of "up_0" and "down_0", how do I fill the rest of the data of 5x5 to fill the whole 10x10 GF?

When I run the analytical continuation, the first few lines are:

Finished reading h5 archive. Found 2 impurities.
Created sigma continuator of type "inversion_sigmainf"
Starting run of maxent now.
Continuing impurities with blocks:
- Imp 0: ['up_0', 'down_0']
- Imp 1: ['up_0', 'up_1', 'up_2', 'up_3', 'down_0', 'down_1', 'down_2', 'down_3' ]
- Appending
- Appending
- .....

Does this mean that I need to construct 10x10 GF with the first two orbitals for all 'up_' cases? If that is actually the case, could you remind me how can I access and input Imp 0 data (up_0) in first two block then moving into Imp 1 (up_0) for next position, etc.

[the-hampel]

Hi @AQuantumGuy ,

We adapted the postprocessing script of solid_dmft to take care of the transformation from impurity to lattice self-energy in case the user uses the wannier90 interface without entanglement:
https://github.com/TRIQS/solid_dmft/blob/DEV_PCB/python/solid_dmft/postprocessing/plot_correlated_bands.py
please copy this file and replace your local version of it. Then you can use the script as described in the tutorial without changes and it should automatically convert the Sigma stored in the h5 from the Maxent run to the correct blockstructure matching the size of the wannier90_hr.dat file.