Error using "mpirun " for SHG

[sabrine]

Dear Yambo developer and user
i want to calculate the SHG using yambo for 1ML MoSe2. I use a cluster to do such a calculation .

But i have a problem with 'mpirun ' . In fact to use yambo_nl i have load the following module:

module load gcc/8.3.0
module load intel/21.1.1
module load openmpi/4.1.0
module load yambo/5.1.1

compile the calculation give the following error (please see the file attached)

<09m-28s> [DIP] Writing dipoles header
[ERROR] STOP signal received while in[04] Dipoles
[ERROR] Writing File ./SAVE//ndb.dipoles; Variable NOT DEFINED; Permission denied


In the r_setup (please see the file attached) is shown that it is serial calculation, even though i load openmpi/4.1.0 and i run it in parallel

Version 5.1.0 Revision 21422 Hash (prev commit) fde6e2a07
Branch is
Serial+HDF5_IO Build

Note that It compile without any error for 'serial' calculation

The problem is it in the module that i load or in the installation of yambo in the cluster?

Thanks alots.

[myrta gruning]

To clarify, when you say that the calculation is completed without errors when running it in serial, are you referring to the same calculation for the dipoles ran at a different time with a serial-only version code or to the setup calculation which is always run in serial?

[sabrine]

Dear myrta gruning
sorry , i dont undrestand well your question .
what i did is first i run the calculation using 'mpirun' so i get the error .
Then , I wanted to check if it is the problem in the 'mpirun 'or in the input so in separate folder i do the process from the binning p2y, yambo_nl, ypp ...etc but i run it in serial this time, so it compile without any error .
Best.

[myrta gruning]

Thank you Sabrine.

To recap, you ran twice the same calculation, using the same cluster, the same installation of the code and having loaded the same modules, but with the difference that in one case you ran the calculation in parallel and once in serial.

Can you attach the reports (you only attached the log, starting with l_ rather than r_) for both the calculation that ended successfully and the one that failed?

Also, does the failure happen only for this system? Did you try e.g. with one of the system of the tutorial?

[sabrine]

Dear Myrta
Thanks for your help .yes exactly as you said i ran twice the same calculation, using the same cluster, the same installation of the code and having loaded the same modules, but with the difference that in one case you ran the calculation in parallel and once in serial.


Please find attached the report file for both parallel ( in which it give an error) and in serial (work well).

For your question about the system. Actually it is not the first time that i did such a calculation .Usually i use my own laptop (it have 16 processor ) and it work well for this system (MoSe2) . The problem start when i use a cluster because now i need to converge (use sufficient K point) also i want to include excitonic effect and the memory of my own laptop is not enough.

Best.