Collisions calculation getting stuck 62% 97%

[sohan]

Dear sir,
I am doing tdbse Shg on a system but my collision calc always get stuck at 62% or at 97%, I'm attaching coll input, log and report, please suggest a way so that my calculation can run smoothly, I am stuck in this step for days, i dont feel memory is an issue and it'll run successfully on single core but takes time to complete. Looking fwd to your valuable response.
regards
Sohan

[sohan]

Also I have configured and compiled 5.3.0 using

./configure --enable-dp --enable-mpi FC=ifort CC=icc

I hope this is not causing the mishap.
I have attached config log.
regards
sohan

[claudio]

Dear Sohan

please try to compile yambo_rt without double-precision and use it to calculate collisions,
it is much faster and collisions in single precision are perfect for real-time dynamics.

However do not forget that for the dynamics you need yambo_nl

best
Claudio

[sohan]

Dear Sir,
Thank you for the kind response, I'll have two different yambo with single and double precision and will give a try with single precision. Will report you the outcomes.
regards
Sohan

[sohan]

Dear sir,
The stuck issue is still there even after configuring with single precision. I have attached again report, input, log and config log. Please suggest any solution.
regards
sohan

[claudio]

Dear sohan

try to setup different parallelization scheme and see if it works, just do:
yambo_rt -X s -collisions -potential h+sex -V par

then you can try to parallelize:

RT_CPU= "1,4,1,4" # [PARALLEL] CPUs for each role
RT_ROLEs= "k,b,q,qp" # [PARALLEL] CPUs roles (k,b,q,qp)

or

RT_CPU= "1,1,1,16" # [PARALLEL] CPUs for each role
RT_ROLEs= "k,b,q,qp" # [PARALLEL] CPUs roles (k,b,q,qp)

and check if it works.
Then remember that your parameters are not converged you should increase at least
HARRLvcs= 3000 mHa # [HA] Hartree RL components
EXXRLvcs= 3000 mHa # [XX] Exchange RL components
CORRLvcs= 3000 mHa # [GW] Correlation RL components

and more bands in X.
best
Claudio

[sohan]

Dear sir,
There's one thing I noticed while running collisions calculation in parallel, that it will stuck for particular no. of processors and will run for one certain number of processors. In my case I was able to run collision calculation

mpirun -np N yambo_rt

when N= 1,2,3,5,11,19 and was getting stuck when N= 4,8,7,16,12
What could be the reason behind it?
regards
sohan

[Davide Sangalli]

It is an issue which was also noticed in the past and never solved so far.

At present we do not have a solution. We are checking the code in order to fix it.

Best,
D.