Solvent gets unbound in MD during QCG run

[codecollider]

Hi,

I'm currently testing QCG and because of some issues I'm testing the examples first. Specifically example 4 from: https://crest-lab.github.io/crest-docs/page/examples/qcg/example_4.html

I downloaded the coord files and ran it with crest pentanol.xyz --qcg benzene.coord --nsolv 25 --T 12 --gsolv --nclus 4 --fscal 0.65 --gbsa benzene

What looks strange to me is, that during the MD (folder tmp_MTD) the solvent (here benzene) is leaving the solute like no repulsive potential is present. Is this correct behaviour? I guess nci is used without explicitly turning it on, right?

I'm using CREST Version 3.0.2, Fri, 27 September 16:49:34, 09/27/2024 commit` (a71e303) and xTB 6.6.1.

I added the data from the (still running) simulation.

test.zip

[codecollider]

Actually i was using xTB 6.6.1 and the run crashed with Segmentation fault - invalid memory reference.
I tried 6.7.0 with the same error and now I try with option --legacy to see if that helps. (edit: no also crashed).

Is there any current binary combination of xTB and CREST that works with QCG ?

edit:

I tried xTB 6.6.1, 6.7.0 and the xtB-bleed and all give the same error.

[pprcht]

Hi, the unbound solvent is obviously not intended behavior...
Unfortunately, QCG implementations are quite in a bad shape and not well maintained in the head branch. Switching to 6.7.1 or the bleeding edge version of xtb is definitely necessary, but even then the routines (in CREST) are quite broken.
I came around to refactor the entire thing in this branch: https://github.com/pprcht/crest/tree/3.0.3-maintenance, but it is not yet merged into main. However, with this version at least all QCG examples worked as intended.

[codecollider]

Hi, the unbound solvent is obviously not intended behavior... Unfortunately, QCG implementations are quite in a bad shape and not well maintained in the head branch. Switching to 6.7.1 or the bleeding edge version of xtb is definitely necessary, but even then the routines (in CREST) are quite broken. I came around to refactor the entire thing in this branch: https://github.com/pprcht/crest/tree/3.0.3-maintenance, but it is not yet merged into main. However, with this version at least all QCG examples worked as intended.

Thanks! Then I have to wait and use COSMO for now as building is not an option.