Electron-Phonon coupling 2D-MoS2 (EPW5.1)

[SaraFiore]

Dear experts,
I am trying to obtain the electron-phonon coupling matrix elements g(q,k) of a monolayer of MoS2

For the seek of simplicity I reported a detailed description of the inputs and of the plot here : https://paper.dropbox.com/doc/e-ph-coupling-ML-MoS2-EPW-5.1--AapH9Z046bS0GI5t3aJK2htEAQ-thdVyYUu33A1Ss1nvdJfn

My references are (see section GOAL of the dropbox paper)
- Two-dimensional Frohlich interaction in transition-metal dichalcogenide monolayers: Theoretical modeling and first-principles calculations
Sohier,Calandra, Mauri DOI: 10.1103/PhysRevB.94.085415
- Phonon-limited mobility in n-type single-layer MoS2 from first principles. Kaasbjerg, Kristen; Thygesen, Kristian S.; Jacobsen, Karsten
DOI: 10.1103/PhysRevB.85.115317

Unfortunately I get weird results (see section Result with EPW 5.1 and section EPW input files)

I tried to perform the same calculation with QuantumESPRESSO (see section Result with Quantum ESPRESSO) i.e.single q-point calculations
with quiet satisfactory results.

I performed all the checks that came to my mind (see section Check with EPW)
- plotting the decay files: decay.dynmat, decay.epmat_wanep , decay.epwane , decay.H , decay.P
- comparing the electron and phonon band structure with independent results obtained with quantum espresso
and they seems good

I have then a couple of questions that may be the reason of the discrepancy

1)I got

Code: Select all

 ??error internal error, cannot bracket Ef  

??so I set Ef manually in the middle of the gap ~ -0.84 eV
??from nscf.out highest occupied, lowest unoccupied level (ev): -1.6587 0.0596
??Is this approach correct?

2) ??Why despite my filqf ='path.dat' (see section EPW input files) is composed by 251 points, in the output I get (see section EPW output files - epw_g.out)
??

Code: Select all

 A selecq.fmt file was found but re-created because selecqread == .false. 
??     We only need to compute       72 q-points

??the starting (0.0 0.0 0.0) and ending (0.33, 0.33, 0.0) point provided in the output correspond to the ones I picked, although I do not understand the reason of this discrepancy.

3) ??What is the meaning of plotting decay.epmat_wanep column 1 vs 3(=Rp). Does my result (4th pane in the first figure of section Check with EPW) make sense?

Thank you in advance for any help!!

Regards

Sara Fiore
ETH - Zurich
PhD student

[Zhe_L]

Dear Sara,

In my opinion,
(1) You don't need to set the Ef in the middle of the gap. If you want to calculate an N-type MoS2, a Ef near the conductive band might be better.

(2) You need a larger fsthick. EPW filter those q points that both E_k and E_k+q are within Ef +/- fsthick for at least one k point. So a large enough fsthick(such as 15eV) is useful in your case.

(3) The convergence of g(R_e, R_p) also confuses me a lot and I am looking forward to the ideas of others.

Best regards
Zhe

[SaraFiore]

Dear Zhe,
thank you for your reply.

1) using Quantum ESPRESSO with a small smearing in the nscf file, I get the value of the Fermi energy

Code: Select all

grep "Fermi" nscf_smearing.out
>> the Fermi energy is    -1.5849 ev

2) It is not clear to me what you meant...

I reported here https://paper.dropbox.com/doc/ML-MoS2-E ... Ym8f9E0zw0

what I obtained changing only the last epw run (i.e. after the wannierization step)

I tried with Ef = -1.5849

and

- fsthick = 1.0 , 5.0, 10.0, 15.0
- filqf = path of 50 or 250 q points along [Gamma, K] (see dropbox file)
- filkf = K or Gamma (see dropbox file)

the result it is always the same bad result

I hope somebody can help me

Thank you in advance for any help

regards

Sara Fiore
ETH - Zurich
PhD student

[Zhe_L]

Dear Sara,

(1) Fermi energy depends on doping. The smearing method can only calculate the undoped case. Actually, it is the relation between Ef and carrier concentration that matters.

(2) Sorry for my poor expression. I mean, if you set fsthick = 15.0, the problem below will disappear.

Code: Select all

     A selecq.fmt file was found but re-created because selecqread == .false. 
?     We only need to compute       72 q-points

The program will calculate all q points that you provide.

Besides, It could be better to fix k at K where CBM and VBM are, instead of Gamma.

Best regards
Zhe

[SaraFiore]

Dear Zhe,
thank you for your answer, although

1) moving the Ef close to the CB gives exactly the same result as if it is close to the VB

2) Yes, it works, but I do not get the reason why (I'm sorry..)

Best regards

Sara
ETH - Zurich
PhD student

[Zhe_L]

Dear Sara,

Sorry, I misunderstood your purpose before.

(1)If you just want to calculate the e-ph matrix, the Ef is meaningless. It works on the calculation of mobility by changing the Fermi-Dirac distribution.

Maybe you can check the wannierized band structure and the convergence of e-ph in real space, i.e. g(R_e, R_p) to find the reason.

(2)fsthick(E_fs) is used to restrict the k and q points and improve the speed of calculation. Only those k states whose energy E_k within (E_f-E_fs, E_f+E_fs) (called fermi shell) are calculated. As for q points, both E_k and E_k+q should be within this range. The q points to be calculated is called 'q-window'. If you increase E_fs, some E_k+q could get into fermi shell and q-window will include more q points.

I don't know if I make myself clear. In fact, I'm just starting to discuss on the internet in English and trying to improve my expression. Therefore, I feel so sorry for my poor English.

Best Regards

Zhe

[SaraFiore]

Dear Zhe,
Thank you for your time and reply and yes, your explanation is clear

I will try to increase the q grid in the phonon calculation, hoping that this will give a better result of decay.epmat_wanep vs Rp

with a 771 q-grid my result is the one reported in section Check with EPW (fig 6) in https://paper.dropbox.com/doc/e-ph-coupling-ML-MoS2-EPW-5.1--AbQgl_3N_UWuYZcs4UzLuC1nAQ-thdVyYUu33A1Ss1nvdJfn
Pretty bad, I would say.

I assume the result I should get is the one reported in fig 2 of https://www.sciencedirect.com/science/article/pii/S0010465510003218?via%3Dihub#fg0020 (EPW: A program for calculating the electronphonon coupling using maximally localized Wannier functions. https://doi.org/10.1016/j.cpc.2010.08.027)

I hope the issue will be solved increasing the grid

I'll let you know

Thank you

Sara
ETH - Zurich
PhD student

[Zhe_L]

Dear Sara,

I find that you used different types of pseudopotential(PP) for Mo(NC) and S(US).

Maybe you could check the band structure and replace one of PPs.

Best Regards

Zhe

[SaraFiore]

Dear Zhe,

the reason why I am using different PP is because I am using the one reported in this library https://www.materialscloud.org/discover/sssp/table/efficiency (which are the one tested for quantum espresso, see here https://www.materialscloud.org/discover/2dstructures/details/MoS2)

I tried to use only Norm Conservative, but the results were worse.

I am currently trying to use a denser q-grid and as latest chance I'll follow what suggested here http://epwforum.uk/viewtopic.php?f=6&t=406&p=1452 by @sponce

I have never looked to much at 2D materials yet but I know that some cutoff might be required. This is not implemented at the moment.
If its not too expansive, I would suggest you to try those coarse grid to see if it improves:

nk1 = 16
nk2 = 16
nk3 = 2
!
nq1 = 16
nq2 = 16
nq3 = 2

If I recall correctly, somebody in our group did that and it helped him for its 2D materials.
In principle you should not need this for a 2D materials but in practice it might required since there is no 2D truncation in the code.

If these 2 options do not work out, I'll try to look for other pseudo potential (although the band structure looked good)

Thank you again for your time

Sara
ETH - Zurich
PhD student