abnormal SHG coefficent

[Dean]

Dear all,
I wan to calculate SHG coefficent by RT-BSE method. But, I get some abnormal SHG coefficent datas, such as 10^5 pm/v.
I find that the polarization is closely determained by the two parameters, NLtime and NLstep. Please see the figure.
My question is how to set these two parameters, NLtime and NLstep and then get the correct SHG coefficents?
Thanks in advance!

[claudio]

Dear

I see from the plot of your polarization, that there is a problem. The polarization should oscillate around zero and not go down.
Please make some test:
1) do you get the correct response in independent particle approximation?
2) how much smearing (dephasing) did you put in your calculation?

for the paramter NLTime and NLtstep,
try to reduce NLstep = 0.01 fs and to use a better integrator NLintegrator= "CRANKNIC"

let me know

best
Claudio

[Dean]

Dear Claudio,
Thanks for your reply.
This issue was solved by changing the version of Yambo from "Version 5.1.0 Revision 21422 Hash (prev commit) fde6e2a07" to "Version 5.1.0 Revision 22561 Hash (prev commit) 785b4fd6f".
Is ther any reference or test to set these two parameters, NLtime and NLstep and get the convergent SHG coefficents?
Best,

[claudio]

Dear Yimin

usually the default parameters are fine.
My advice is to use the CRANCKNIC integrator.
NLtime is fixed automatically from the smearing parameter, it is inversely proportional to the smearing,
Small smearing equal long runs.
you can make some test to decrease it and speed up calculation if you see that the result does not change

best
Claudio

[Dean]

Dear Claudio,
Thanks for your reply.
I use the CRANCKNIC integrator.
I will make some test.
Best,

[Dean]

Dear Claudio,
I got again abnormal SHG coefficent for another 2D materials, see below. This material show a small bandgap of about 0.5 eV.
I have used the latest version of Yambo(5.2.1). The input and output files are attahched. Please refer it.
Please give me some suggestions.
Thanks in advance!

[claudio]

Dear Yimin

I tried to run your calculation with slightly less converged parameters in such a way to have a fast result.

I found a correct X2, in the figure you can find the results IPA vs TD-SES for X2_yyy and X2_xyy

in attachment I put all input files, DFT and Yambo. (I used different pseudo and scissor operator, I did not have the GW results.)

please have a look and try to reproduce it, then you can increase cutoff, kpoints etc..

best
Claudio

ps use the last Yambo version on github, it is alwayer better

[Dean]

Dear Claudio,
Thanks for your help. I get resonable results by using your input files.
By comparing the input parameters, I think the abnormal SHG coefficent are caused by these parmeters, i.e., "NoDiagSC, ElecTemp= 0.0, BoseTemp=0 " in collisions and SHG calculations. I set these parmeters, because Yambo reports "metallic system", while it is a semiconductor for my system. So, I guess these parmeters, i.e., "NoDiagSC, ElecTemp= 0.0, BoseTemp=0 " will result in abnormal NL results?
Moreover, in QE runs, if I delete these parameters for a semicouductor, i.e.,"occupations='smearing', smearing='mv', degauss=0.0001," the following Yambo calculations shows correct outputs.
I wish these tips may be helpful for other users.
Best,

[Dean]

Dear Claudio,
After some tests, I find some unaccountable issues.
I can get resonable SHG coefficents, while the k-point of QE is set as 18*18*1, 15*15*1 or 12*12*1.....
But, abnormal SHG coefficents are produced with the same input files for Yambo, while the k-point of QE is set as 19*19*1, 20*20*1 or 21*21*1 .....
Please see the attached files.
Any suggestion is helpful. Thanks in advvance.
Best,

[claudio]

Thank you for the inputs, I will check them

best
Claudio