Page 2 of 3

Re: Q dependence of exciton eigenvalues

Posted: Wed Oct 27, 2021 1:57 pm
by DavidPolito93
Dear all,

In the folder where I performed BSE computations I don't have the ndb.BS_PAR_Q1 database.
I have for Q=1:

ndb.BS_diago_Q1 and ndb.BS_Q1_CPU_x (where x goes from 1 to 119 in my case)

How can I obtain the ndb.BS_PAR_Q1 database?

Best,

Davide

Re: Q dependence of exciton eigenvalues

Posted: Wed Oct 27, 2021 2:04 pm
by Daniele Varsano
Dear Davide,

please sign your post with your affiliation, it is a rule of the forum and you can do once for all filling the signature in the user profile.
The BS database are CPU dependent. You can either use just 1CPU (but this could be prohibitive), or you need to use the parallel I/O.
In order to do so you need to compile the code using in the configure the command:

--enable-hdf5-par-io

I never used it, but if I'm not wrong this should do the job, maybe other expert on parallel I/O can add more or correct me.

Best,
Daniele

Re: Q dependence of exciton eigenvalues

Posted: Thu Oct 28, 2021 3:24 pm
by DavidPolito93
Dear Daniele,

Sorry, I forgot to put the affiliation.

I have done as suggested by Fulvio: I re-compiled Yambo (more precisely version 5.0.4) with the option: -enable-hdf5-par-io

The p2y works and also the initialization of the SAVE database.

However when I run the GW step, it crushes. I am sending in attachment the input, output, the LOG and the error of my machine.

It seems it has some problems with the HDF5 after he finishes with dipoles. What's going wrong?

Best,
Davide
-----------------------------------------------------
Post-doctoral fellow
Institut des NanoSciences de Paris
Sorbonne Université, CNRS
4 place Jussieu,
75252 PARIS Cedex 05

Re: Q dependence of exciton eigenvalues

Posted: Thu Oct 28, 2021 3:44 pm
by Daniele Varsano
Dear David,
this seems an error from the H5FScache.c :

Code: Select all

 yambo: H5FScache.c:1100: H5FS__cache_sinfo_deserialize: Assertion `old_tot_space == fspace->tot_space' failed., 
honestly I do not have a clue on what is happening here, someone else maybe has more experience and can provide some hints.
Mot probably not related to you problem please note that in your script the intel/mkl module is not loaded:

Code: Select all

ERROR: Unable to locate a modulefile for 'intel-mkl/2020.5'
Next, I had a look to our input and I suggest you to not use the terminator for X as it is memory intensive and does not save much, while the terminator on G is highly suggested.

Best,
Daniele

Re: Q dependence of exciton eigenvalues

Posted: Thu Oct 28, 2021 4:01 pm
by DavidPolito93
Dear Daniele,

Thanks for your suggestions. Should I write something in the compilation topic or should I wait here if somebody has an answer?

Best,

Davide

Re: Q dependence of exciton eigenvalues

Posted: Thu Oct 28, 2021 4:16 pm
by Daniele Varsano
Dear Davide,

I'm not 100% sure this is a compilation issue. Let's hope that someone has a clue.
What you can do in the meanwhile is to calculate the dipoles without the parallel netcdf support and the BSE using it, in this way we can see if it is a particular problem of the dipole section.
Best,

Daniele

Re: Q dependence of exciton eigenvalues

Posted: Thu Oct 28, 2021 7:36 pm
by Davide Sangalli
Dear all,
hard to say from the files.

Looking at the log file the job is interrupted when trying to write the ndb.dipoles database.
Can you check if the database was created and what it contains?
Also, maybe in one of the log files there is a specific error message
Tray to run

Code: Select all

$ grep ERROR LOG/l-*
The error message I see is a general HDF5 message.
Attach the config/report file from your compilation as well.

D.

Re: Q dependence of exciton eigenvalues

Posted: Fri Oct 29, 2021 9:27 am
by DavidPolito93
Dear Davide,

So:

1) In my folder "g0w0" I have the dipole database "ndb.dipoles". If I do "ncdump -h ndb.dipoles" I get:

Code: Select all

ncdump: ndb.dipoles: ndb.dipoles: NetCDF: HDF error
2) By doing "grep ERROR LOG/l-*" I get:

Code: Select all

LOG/l-g0w0_HF_and_locXC_gw0_rim_cut_em1d_ppa_CPU_100:P100-r4i6n29: [ERROR] STOP signal received while in[05] Dipoles
LOG/l-g0w0_HF_and_locXC_gw0_rim_cut_em1d_ppa_CPU_100:P100-r4i6n29: [ERROR] Writing File ./g0w0//ndb.dipoles; Variable  NOT DEFINED; NetCDF: HDF error
LOG/l-g0w0_HF_and_locXC_gw0_rim_cut_em1d_ppa_CPU_101:P101-r4i6n30: [ERROR] STOP signal received while in[05] Dipoles
And this is for all the others cpus except for the first one apparently.

3) I am attaching the report file

Best,

Davide

Re: Q dependence of exciton eigenvalues

Posted: Tue Nov 02, 2021 9:45 am
by Davide Sangalli
It looks like it is the issue with asynchronous I/O
Try to replace the file

Code: Select all

src/dipoles/DIPOLE_IO.F
with the one attached and recompile yambo.

Best,
D.

Re: Q dependence of exciton eigenvalues

Posted: Tue Nov 02, 2021 4:47 pm
by DavidPolito93
Dear Davide,

Thank you very much! Now everything works and I finally have also the ndb.BS_PAR* database :)

Best,

Davide