Double grid in Yambo-4.5.2

[jjmm1974]

Dear all

I have tried the Double grid technique following this tutorial:
http://www.attaccalite.com/speed-up-die ... ith-yambo/

I did it successfully for yambo-4.1.2 (rev. 120), but I am facing problems with yambo-4.5.2 (rev. 9122). It seems that the algorithm is finding something that it did not expect. Please, have a look to the attached r- (original, coarse grid) and r-_01 (attempt for double grid) files. I am also appending the input file.

Just to double check, this is what I did:
1) scf + nscf + BSE (with QP corrections) in a coarse grid containint 30 k's.
2) scf + nscf in the double grid system, calculating only the bands involved in BSE corrections.
3) p2y -w in the double grid *.save folder -> Yielded SAVE without wave functions.
4) ypp -m + path to the folder where the above SAVE is.
5) ypp -> Yielded ndb.Double_Grid in the original SAVE folder.

Could you please give me any hint? Thanks in advance.

Juanjo

P.S. Happy New Year. It will be easy to be better than 2020, no doubt.

[claudio]

Dear Juanjo

I don't know what can have generated your error message.

I tried to repeat the calculations of the double-grid with yambo 5.0 that you can find
in the github now:

https://github.com/yambo-code/yambo

and it works both serial and parallel.

My advices are:
1) generate input file for the BSE with
yambo -o b -k sex -y i -X s (or -X p after GW calculations)

2) do not use too many processor for the parallelization, but put some of them in openmp
by setting for example

export OMP_NUM_THREADS=4

3) use parallel algebra in the BSE, just add -V par in the input file generation and set

BS_nCPU_LinAlg_INV =4 # [PARALLEL] CPUs for Linear Algebra (if -1 it is automatically set)
BS_nCPU_LinAlg_DIAGO=4 # [PARALLEL] CPUs for Linear Algebra (if -1 it is automatically set)


see if it works.

Notice that double-grid approach is convenient when you have not too strong bound excitons, as for example Silicon, AlAs, GaAs etc...
Otherwise you can consider other approaches as Haydock solver for the spectra (-y h) and SLEPc for the excitonic wave-function (-y s)

best
Claudio

[jjmm1974]

Hi, Claudio
Thanks a lot for your response.
I already got the BSE spectra with a grid fine enough, but wanted to check the Double Grid method. I will upgrade to yambo-5.0 and will check again.
Best
Juanjo