Crest 3 QCG, final xTB optimisations not using the available cores

[andrewtarzia]

As in title, a QCG run seems to be hanging after adding the requested number of solvents, but looking at htop I see the following command is running:

xtb cluster.xyz --gfn2 --opt best.xyz best_after_gen.xyz cluster.coord cluster.xyz cluster_cavity.coord final_structures.xyz optimized_structures.xyz solute solute_cavity.cood solute_cut.cood .........

As I understand, this is the final xtb optimisation of cluster.xyz, right? However, it seems to only be using 1 thread. Is there a way to change this?

My crest version: Version 3.0, Sat Apr 6 18:06:37 UTC 2024 commit (d321183) compiled by 'runner@fv-az778-216'

Crest command: crest input_solute.xyz --qcg input_solvent.xyz --nsolv 10 -xnam xtb --alpb water -chrg 0 -uhf 0 --gfn2 -T 6 --optlev crude --nopreopt --fixsolute -I det_control.in

[andrewtarzia]

An alternative option is to turn off this final optimisation, which I would actually prefer. Is that possible?

[pprcht]

I've tried to fix some things with regards to this issue in #300 and the continuous release.
The final optimization can be turned off now with -no_final_opt_gfn2

[andrewtarzia]

Hey @pprcht , thank you for this! --no_final_opt_gfn2 seems to avoid this issue. However, cluster_optimized.xyz is still written to file and seems to be very slightly different to cluster.xyz. Is that expected?

[pprcht]

From my understanding of the code, yes, that seems to be normal. @cplett is the author and maintainer of QCG.
What I read from the routine is that cluster.xyz is put forth for one final optimization to produce cluster_optimized.xyz. Even when cluster.xyz is a minimum already, the optimization will take one or two tiny steps. My guess is that's what you're seeing.

Have you by any chance checked if #300 also resolves your issue #294?

[andrewtarzia]

Ok, thank you again! I will update on #294 in that conversation.

Andrew

[pprcht]

I'm closing the issue for now, if something else related to this comes up feel free to comment here again.