ypp, Bands interpolation problem

[Fabiof]

Dear all,

I have done a calculation on monolayer MoS2 to obtain the GW gap correction.
The command that i used is

Code: Select all

yambo -d -k hartree -g n -p p

I tested the convergence on all necessary variables.

After that i used the post-processor ypp to obtain the band structure in the path Gamma-K-M-Gamma.
Without using the variable GfnQPdb= "E < GoWo700-RL/db.QP", i obtain a band structure that is consistent with the band structure of my DFT calculations.

The problems arise when i use the variable GfnQPdb= "E < GoWo700-RL/db.QP" to obtain the qp energy corrections. The band structure seems to be dependent of the variable
GfnQP_N.

First problem: If i use GfnQP_N = 1, the band structure has a very strange structure. Very different from the expected one. Same happens if i use GfnQP_N = 10.

Second problem: When i start to use values GfnQP_N = 25 the band structure has the expected structure but if i increase the variable GfnQP_N, the vlaue of gap tends to the DFT value (1.53 eV) .

Here is the data:

GfnQP_N , Gap (eV)
25, 2.34
50, 1.94
100, 1.74
200, 1.64
500, 1.58
1000, 1.56
2000, 1.55
4000, 1.54


I would like to know why this happens? What is the function of GfnQP_N? Why the band interpolation is so dependent of this variable? why it tends to the DFT value?

I have attached some input and report files.

[Davide Sangalli]

Dear Fabio,
the variable which controls the QP corrections has to be imporved.

Try to use:

GfnQPdb= "E < GoWo700-RL/db.QP uki"
GfnQP_N=1

Hope it helps.
D.

[Fabiof]

Dear Davide,

I tested your suggestion, but there is no difference between

GfnQPdb= "E < GoWo700-RL/db.QP uki"
GfnQP_N=1

or

GfnQPdb= "E < GoWo700-RL/db.QP"
GfnQP_N=1

the qp energies are the same.

[Fabiof]

I tested Silicon and i have the same problem.

The only difference is that with GfnQP_N=1 the band structure has the expected structure, but if i increase GfnQP_N, the gap values tends to the DFT value.

[Davide Sangalli]

Dear Fabio,
some more details. My last reply was indeed too short.

When you do this kind of run the code basically is ready to do 2 interpolations in k-space:
1 - An interpolation of the QP corrections from the one calculated to all the k-points included in your SAVE directory
2 - An interpolation from the k-points in your SAVE directory to the path in k space you specified

The interpolation 1 is a nearest neighbour interpolation.
The interpolation 2 is based on Fourier function and is more involved.

The variable GfnQP_N controls the number of neighbour used in the nearest neighbour interpolation (interpolation 1).
Suppose you want the QP correction at a point k where you did not compute it.
If GfnQP_N=1 the code will use the value computed at the nearest k-points.
If GfnQP_N=2 the code will use the average of the value at the 2 nearest k-points.
If GfnQP_N=3 the code will use the average of the value at the 3 nearest k-points.
and so on ...
So if you use too many neighbours, the code will basically do an average on all the k-points.

In particular if you computed the QP corrections for all k-point in your "SAVE k-grid" it is useless to set GfnQP_N higher than 1
and there should be not the interpolation 1. Thus I guess it is ok that for Silicon works with GfnQP_N=1
(Probably we could code to do an average based on the distance to avoid this problem when increasing GfnQP_N ... )

However there is still something I do not understand for the case of MoS2.
I'll try to investigate the problem a bit more and I''l let you know ..

D.

[Fabiof]

Dear Davide,

Thanks for the detailed answer. I tested the hexagonal BN and it has the same problem of MoS2.

Im doing some calculations changing some variables in DFT and yambo inputs to see if i can "avoid" the problem. I will let you know if something works.


Fabio

[Fabiof]

Dear Davide

I just want to let you know that i have solved the problem for BN hexagonal (convergence issues).

Regarding MoS2, my bandstructures are much better now, but i still have an indirect gap (with GoWo and PPA).



Fabio

[Fabiof]

Dear Davide

I think i know what is the main problem that originates the indirect gap at K point in bandstructure of MoS2. It seems that Mo Pseudopotential has to include valence 4s and 4p states in its generation. This states are important to describe the exchange part.

I'm going to try to generate the PP containing that states to verify if it has some influence.

Fabio

[Davide Sangalli]

Dear Fabio,
thank you for the update.

Best,
D.

[Fabiof]

Dear all,

I'm having a problem trying to do a ypp band interpolation after a PPA calculation with MoS2 full relativistic.
I have no problems after a COHSEX calculation.

I have these warning after the command ypp

Code: Select all

 <01s> [WARNING]Bands range reduced
 <01s> Number of K-points in the circuit :  4
 <01s> [06.01] External QP corrections (G)
 <01s> [QP@G] E<../PPA100-5/ndb.QP[ PPA XG:285 Xb:1  100 Scb:1  100]
 <01s> [WARNING]Fit order on E [c] reduced to 0
 <01s> [WARNING]Global and response function energy levels have different metallic character

The result is a o-build and o-interpolated files with the same energy for the different K-points at conduction bands (after the b18):

Code: Select all

#   |k|        b13        b14        b15        b16        b17        b18        b19        b20        kx         ky         kz
#
    0.000000  -2.808954  -2.808954  -2.687519  -2.687519  -0.693666  -0.693666   1.940277   1.940277   0.000000   0.000000   0.000000
     0.04415   -2.82786   -2.82328   -2.63373   -2.63580   -0.68543   -0.68154    1.94028    1.94028    0.01667    0.03333    0.00000
     0.08829   -2.88999   -2.87895   -2.50880   -2.51189   -0.68801   -0.67688    1.94028    1.94028    0.03333    0.06667    0.00000
     0.13244   -2.97089   -2.95265   -2.46597   -2.45793   -0.71617   -0.68474    1.94028    1.94028    0.05000    0.10000    0.00000
     0.17659   -3.09982   -3.07635   -2.46147   -2.43409   -0.80285   -0.75241    1.94028    1.94028    0.06667    0.13333    0.00000
     0.22074   -3.27964   -3.25221   -2.36663   -2.32406   -0.96350   -0.91316    1.94028    1.94028    0.08333    0.16667    0.00000

Do you know what is the problem? The files are attached.

Thanks,

Fabio