Re: Incomplete and/or broken PPA/Static diel. func. in G0W0

rnanune · Post by **rnanune** » Tue Dec 16, 2014 10:55 pm

Dear Yambo developers:

I am using Yambo 3.4.1 with the p2y interface to do some GW+BSE calculations. I have been trying to calculate the QP spectra of methyl-ammonium lead tri-iodide (CH3NH3PbI3) using the G0W0 self-energy and the plasmon-pole dynamical screening. Attached are my yambo input and report files. I've edited the report files to make them a little smaller so that I could upload them here.

I first calculated the static dielectric function using all the transferred momenta from the PWscf mean-field calculation. Then, I set up the yambo input file for the dynamical screening function using "yambo -p p -g n". However, in the middle of the calculation I got a program termination with the errors:

[ERROR] STOP signal received while in :[06.01] G0W0 : the P(lasmon) P(ole) A(pproximation)

[ERROR]Incomplete and/or broken PPA/Static diel. fun. database

I looked at the earlier forum threads addressing this same error:
viewtopic.php?f=14&t=166
viewtopic.php?f=14&t=326

In the input file, I am making sure that I am calculating the screening dielectric function at all the Q-points, so I am not sure why I am getting the above errors.

Thanks for your help.

Ravindra Nanguneri
Department of Chemistry & Biochemistry
University of Notre Dame
251 Nieuwland Science Hall, Notre Dame, IN 46556
USA

Post by **Daniele Varsano** » Tue Dec 16, 2014 11:17 pm

Dear Ravindra,
from what I can see in your report, you restarted a run from previous calculations. It looks that something went wrong in the restart, some corrupted database files or something similar. It is possible that some previous calculations went wrong for some reason.
Can you post the standard outputs (the l_em1d_ppa_HF_and_locXC_gw0_0* files of your runs) and the result of the ls ./SAVE/ directory and the output of yambo -D?

By the way, in order to perform GW calculations you do not need to calculate static screening.

Best,
Daniele

rnanune · Post by **rnanune** » Wed Dec 17, 2014 12:11 am

Dear Daniele,

Thanks for your reply. Attached are the log files and the output of "ls ./SAVE".

I did a few plasmon-pole/G0W0 runs before the current run for which I reporting the error, but those previous runs died due to memory shortage. I solved the memory problem by reducing the number of bands to sum over and the parameter EXXRLvcs. So maybe I need to delete the databases from the previous runs?

Ravindra Nanguneri
Department of Chemistry & Biochemistry
University of Notre Dame
251 Nieuwland Science Hall, Notre Dame, IN 46556
USA

Post by **Daniele Varsano** » Wed Dec 17, 2014 11:45 am

Dear Ravindra,

So maybe I need to delete the databases from the previous runs?

In line of principle no, as it is possible to restart non finished calculations, anyway I can see here some inconsistencies:
your input file contains 100 Xbands, while the previous unfinished run were done with 150 bands. This can be the reason of the problem you encountered.
The restart is thought to continue unfinished runs with the same input file.
At this stage, I suggest you to remove the ./SAVE//ndb.pp* files and restart your calculations.

A suggestion:
From your input:

Code: Select all

%QPkrange                    # [GW] QP generalized Kpoint/Band indices
  1|256| 20|60|
%

you aim to calculate a lot of quasi particle corrections: 41 bands for 256 k-points. My suggestion is to check convergences for few bands (e.g. 1 kpoint and few bands, the gap etc.) and once you have assessed the convergence parameter proceed to a production run with lot of bands and kpoints. For instance

Code: Select all

NGsBlkXp= 5            RL

5 lattice vectors in the inversion of the dielectric matrix quite surely are not enough and the computational load increase rapidly when raisning the G block. Next also the number of bands in the screening and G summation need to be checked.

Best,
Daniee

rnanune · Post by **rnanune** » Fri Dec 19, 2014 12:37 am

Dear Daniele,

I would like to ask if the number of transferred Q-points to be specified in the "yambo.in" input file is the number in the IBZ, or in the BZ? I have 256 Q points in the IBZ from the mean-field calculation, but during setup yambo reports 260 Q points in the IBZ in the "r_setup". I would like to know why this happens. Is it because Yambo is arranging the Q points in shells? Do you think this might also cause the error of "Incomplete and/or broken PPA/Static diel. func. in G0W0" if I always specify 256 Q points instead of 260 or 512 (the number of points in the unfolded BZ) in the input?

Thanks,
Ravindra Nanguneri
Department of Chemistry & Biochemistry
University of Notre Dame
251 Nieuwland Science Hall, Notre Dame, IN 46556
USA

Post by **Daniele Varsano** » Fri Dec 19, 2014 1:10 am

Dear Ravindra,
you have 256 K-points.
Then yambo calculate the q points |k=k'| and these are 260. They are reported in the r_setup and all the report files.
In the input for the GW you do not need to specify them, as all are needed to perform integrals in the Bz.
Please note that the variable:

Code: Select all

%QPkrange                    # [GW] QP generalized Kpoint/Band indices
  1|256| 20|60|
%

is meant to indicate the K points and bands you want to calculate the qp corrections (you can reduce it for convergence calculations!!), but the screening
is calculated for all the q points as integrals are needed to calculate the self energy.
For the static screening you can specify the q-points, as you can be interested in calculating electron energy loss function at finite q-points, but as I told you before the static screening is not needed for the GW calculations.
Best,
Daniele

rnanune · Post by **rnanune** » Tue Dec 23, 2014 4:47 pm

Dear Daniele,

My Plasmon Pole/G0W0 ran successfully after deleting "ndb.pp" of the previous runs from ./SAVE and reducing the number of k-points in QPkrange as you suggested. Now, I will increase the bands in the summation and size of the inverse dielectric matrix to check for convergence.

I would like to ask why there is no database like "ndb.em1d" generated in the Plasmon Pole/G0W0 runs. I thought the "yambo -p p -g n" run mode would generate the inverse dynamical dielectric matrix, but I don't see it in my folder "./SAVE". Also, I saw in the report that the most time is spent in calculating the dynamic PP dielectric matrix. I am assuming Yambo will calculate this matrix just once and use it for the GW bands for the subsequent runs.

A few more questions:
1) Why does yambo.in have two variables "QPkrange" and "QPerange"? If I specify only "QPkrange", is "QPerange" ignored?
2) What is the meaning of the variable "GbndRnge"? Is it the range of bands to be summed over for the calculation of W?
3) When I increase the bands summation and dielectric matrix block size parameters, should I deleted the "./SAVE/ndb.pp" from the old runs to avoid getting the same error "Incomplete and/or broken PPA/Static diel. func. in G0W0"?

Thanks,
Ravindra Nanguneri
Department of Chemistry & Biochemistry
University of Notre Dame
251 Nieuwland Science Hall, Notre Dame, IN 46556
USA

Post by **Daniele Varsano** » Wed Dec 24, 2014 10:10 am

Dear Ravindra,

My Plasmon Pole/G0W0 ran successfully..

Great, so there was some corrupted file.

why there is no database like "ndb.em1d" generated in the Plasmon Pole/G0W0 runs.

Because there is the PP database, so what you need for a GW run in this approximation, not the entire dynamical matrix, but essentially the information coming from two frequencies (plasmon pole model).

I am assuming Yambo will calculate this matrix just once

Yes this is the case.

1) Why does yambo.in have two variables "QPkrange" and "QPerange"? If I specify only "QPkrange", is "QPerange" ignored?

Two way to specify the states you aim to calculate corrections. You need to specify only once the other is ignored.

2) What is the meaning of the variable "GbndRnge"? Is it the range of bands to be summed over for the calculation of W?

It is the band summation entering in the Equation for the mass operator you can find here in the documentation.

3) When I increase the bands summation and dielectric matrix block size parameters, should I deleted the "./SAVE/ndb.pp" from the old runs to avoid getting the same error "Incomplete and/or broken PPA/Static diel. func. in G0W0"?

If you can increase the band summation GbndRnge you do not need to delete anything, as you use always the same screening. If you raise the bloxk size or the bands to calculate the screening you need to recalculate the databases. In theory you should not delete the files as Yambo should recognize that the old database are not suitable for the present run, recalculate and overwrite them. All these steps are then reported in the report as warnings I think that last time you had problem with unfinished jobs and corrupted files. .

Best,
Daniele

rnanune · Post by **rnanune** » Mon Jan 05, 2015 7:58 pm

Dear Daniele,

I have a question about G0W0, maybe it has been answered in some of the other posts: If the G0W0 run stops before finishing the calculation for all the k-points requested in "Qkrange", will yambo restart from the point where it stopped in the previous run, assuming I don't change any parameters in the input file?

Thanks,
Ravindra Nanguneri
Department of Chemistry & Biochemistry
University of Notre Dame
251 Nieuwland Science Hall, Notre Dame, IN 46556
USA

Post by **Daniele Varsano** » Mon Jan 05, 2015 8:05 pm

Dear Ravindra,
correct. You should have in the running directory a directory called RESTART. If you inspect the file contained inside it contains information on how many step have been done and the total number of step asked in the input.
Of course it is important to rerun exactly the same input file. Sometimes things can be wrong if there are some corrupted files, this can happen if the code stops while writing files, but it is very uncommon.
Best,
Daniele