convergence test

[samaneh]

Dear Developers
I have done DFT calculations with "nbnd=200"
can I set the number of BndsRnXd to 1| 400 in yambo input file
and run yambo
or first I have to do the DFT calculations with "nbnd=400" ???
and then run yambo with "BndsRnXd 1| 400" ???
Thank you
S. Ataei, PhD student of physics at University of Tehran, Iran

[myrta gruning]

Dear Samaneh

yambo gets all infos on wavefunctions from dft, so it is the latter: You need to perform a larger dft (nscf) calculation with 400 bands.
You can set the number of bands in yambo to a lower value though.
When performing the convergence you may want to use a number of kpoints lower than the converged value and to consider the gw correction on only few selected states.

Best
m

[samaneh]

Dear Dr. Myrta,
Thank you for your helpful comments
Best regards
Samaneh

[Zahra Taghipour]

Dear Developers and Users

Hello

I am doing the convergence test for my case and I wonder if what I'm doing is correct:

I get a good band gap for these parameters:
K_pnt = 13
GbndRnge = 1000,
bandRnXn = 1000,
NGsBlk = 73
ExxRLvcs = 73
FFT 70'000

Then, I restart the computation (by setting the same output for all) while decreasing only one of the parameters.

I have 2 questions:
1- I know that the optimized parameter is dependent on the other parameters. Therefore, how confident can I be about my procedure, i.e. can I optimize them, independently?
2- Is restarting OK ? Yambo does not do the computation from scratch if I use the same name as an output_name for all the convergency tests, which makes it really fast.

I appreciate your comments
Zahra
PhD student, CHTM, University of New Mexico, Albuquerque, NM, USA

[Daniele Varsano]

Dear Zahra,
1. Some parameters are not independent (for instances bandRnXn and NGsBlk), while others are independent (ExxRLvcs controls the exchange part only). Anyway converging parameters one-by-one looks me a correct procedure.
2.Yambo read existing database and depending on the variable has changed input can read the database and give you a warning or decide that the changed variable imply to repeat the calculation with the new variable.

You can see it in the report file, the read database and the one recalculated and written by scratch are indicated with (RD ad WR). The variable of the databases, header, are reported and indicated with WRN or ERR the variable different in input. If you find an ERR you will see that te database is recalculated with the new values of the variables found in input.

Best,
Daniele

[Zahra Taghipour]

Dear Daniele,
Thank you for your response.

1. Some parameters are not independent (for instances bandRnXn and NGsBlk)

So, in the case of two parameters that are not independent, is one-by-one procedure still correct?
Could you tell me if these two are the only dependent parameters, or there are any other ones?

You can see it in the report file, the read database and the one recalculated and written by scratch are indicated with (RD ad WR). The variable of the databases, header, are reported and indicated with WRN or ERR the variable different in input. If you find an ERR you will see that te database is recalculated with the new values of the variables found in input.

Actually, I restarted a calculation changing FFT, ExxRLvcs, and NGsBlk, while keeping the rest unchanged. The calculation did not last so long and this made me wonder if the result was reliable.
So, I assume there is no harm in restarting the calculation, even with some ERR in the report.


I also had this question about Transition Dipole that is being calculated at the very beginning. Usually, this takes a long time. However, I removed the **kb** files, as it is recommended here, and the calculation was done in half an hour!
I know this question is not related to the topic, but I was wondering if you could tell me if this Dipole measurement is also affecting the GW result? i.e. When do we need to keep the *kb* files?

Best,
Zahra Taghipour,
PhD at UNM, USA

[Daniele Varsano]

Dear Zahara,
1. Yes. By producing different convergence curves of one parameter with respecr the other. See eg fog 1 of Prb 84 241201 (2011).
An Err in the reports wheb reading databases implies that the database is recalculated and overwritten. From a careful reading of the report you should see what is kept and what is recalculated. You can also use the -J option to store databases in different directories.
About the commutator of the nom local part of pseudos in dipoles it something system dependent and should be tested anyway i would say that it should affect more the oscillator strengths in optical spectra than a gw quasiparticle calculation.
Best,
Daniele