collision OOM

[sunxl]

Dear all,

When I'm computing collisions and the memory is insufficient, the computation progress always gets stuck without reporting an error—it just continuously runs a futile task. At this point, I kill the process and switch to using more memory on a single core for the computation, which results in a warning.

[WARNING] Restarting from previous database

1、What I want to ask is, is it reliable to resubmit the collision task for continuation? Could the results undergo sudden changes due to task interruptions?
2、Moreover, I've noticed that the collision process always gets stuck when the progress reaches 97%. When I kill the task and resubmit it, the task shows as completed within seconds. In such circumstances, the resulting collision output file can be smoothly read during the next step of calculating SHG. Does this indicate that the collision step was completed correctly?

I appreciate any guidance and tutorial on that!

Best，
sunxl

[claudio]

Dear sunxl

here my advices:

1) use yambo_rt to calculate collisions and not yambo_nl that is in double precision, results will be the same

2) use multi-threads for example 2 or 4 threads this can help with the memory

3) start from a simpler calculations with less k-points and then increase them slowly

let me know
best
Claudio

ps in alternative we implemented approximated collisions here:
https://wiki.yambo-code.eu/wiki/index.p ... r_response
but you have to check if this approximation is good for your system