Yambo Community Forum

Dear All,

I am trying to do BSE calculation with rim_cut for 2D mos2 unit cell using yambo version 4.0.2 Rev 96.
I have generated the input file for bse using During runtime, the program is always giving me a segmentation fault at the calculation.
I am attaching the necessary input files. Can you please let me know where I am doing wrong ?

Regards

/Soumyajyoti

Dear Soumyajyoti,

I cannot see any mistake in your input.
The only think I can see that can cause problem is:
try to replace with:
Beside that, even if not all related with the seg. fault:
1) 10000 random Q points are quite few (you can see also in the report when comparing the volume of Bz)
2) X_all_q_CPU= "4 1 4 8" I would avoid parallelization on "q", the calculation results quite unbalanced (we are working on that)
3) BS_CPU= "16 1 8" , try to parallelize on K the most possible, next on eh
4) BSE bands: quite large number of bands, You will have a huge excitonic matrix, did you check that it is needed? If I'm not wrong Nc*Nv*Nk=32*24*2601~2 Millions, I'm not sure this is manageable.
5) BndsRnXs: Bands in \chi have to fulfill a closure relation, it is safer to include all the occupied bands.

Try to change the CUTGeo, if the problem persist we will need to reproduce the error.

And can you also try to see if the code does not stop when removing the rim_cut from the input, to be sure that the problem is in the building of the cutoff potential?

Best,

Daniele

Dear Daniele,

Thanks for your reply.

I have tried , but still does not help.

Regarding your other comments.
1. I have now used 1000000 random Q points while trying rim cut
2. Removed parallelization on "q" and using "1 4 4 4" for "q k c v"
3. using BS_CPU = "16 4 1" (Although I got a warning "[WARNING] n_eh_CPU > 1 in a system with symmetries. This is not efficient, in particular if you have many kpts. [WARNING] It is suggested to first distribute on "k" and "t". Maximum value for both is: 243) Should I remove parallelization on eh also ? What will be best strategy for parallelization here ?
4. I have reduced the BSEbands to Nc=8, Nv=8.
5. BndsRnXs: I am using all occupied bands and same number of unoccupied bands.

Using the above suggestions I have done two calculations. One with rim_cut and another without rim_cut.
With rim_cut I got the same segmentation fault I got before. However, without rim_cut the code ran and gave me a OOM error during BSE kernel loop (M 8.808 Gb] Alloc BS_T_group_X_oscillators ). I guess I need to increase number of core more to reduce per core memory use.

I am currently doing another run without rim cut with same parameters and reducing Nc=4 and Nv=4 and it is running so far.

Attaching the inputs, reports and LOG files

Regards

Soumyajyoti

Dear Soumyajyoti,
thanks for reporting:

3. using BS_CPU = "16 4 1" (Although I got a warning "[WARNING] n_eh_CPU > 1 in a system with symmetries. This is not efficient, in particular if you have many kpts. [WARNING] It is suggested to first distribute on "k" and "t". Maximum value for both is: 243) Should I remove parallelization on eh also ? What will be best strategy for parallelization here ?

Ok, sorry my fault. Try to distribute on k the most you can, next ot t.

OOM error during BSE kernel loop (M 8.808 Gb] Alloc I guess I need to increase number of core more to reduce per core memory use.

Yes, otherwise use less task for single node in order to have more memory available.

We will check on the rim_cut, I suspect that will be not straightforward to spot the problem: t is first time this problem is reported. We would need to reproduce the problem: Can you send us your quantum espresso input files together with the pseudos?

Best,
Daniele

Dear Daniele,

Here is the quantum espresso input files. I am using PWSCF v.5.3.0

Regards
Soumyajyoti

Dear Soumyajyoti,
thank you we will look at it as soon as possible.

Best,

Daniele