grimme-lab
Remove CPCM-X library
This PR removes dependency of CPCM-X library.
The reasons why:
- often leads to failure of CI tests that may affect users too
- no geometry optimizations
- API is used through files
While I see the stability problems with CPCM-X, it is an actively used feature of xtb and cannot just be removed. We are currently working on new solvation models, so maybe we can replace CPCM-X at some point in the future
actively used feature of xtb
Do you have statistics about CPCM-X + xtb combination?
We are currently working on new solvation models,
There is GBSA solvation model which supports grafients, while CPCM-X does not.
Our team uses CPCM-X a lot for various applications, like pKa prediction and other soon-to-be-released work, so this would be quite damaging for us. It's very useful to have accurate low-cost solvation energies, even without gradients.
Corin @corinwagen, thank you highlighting this use case, in general I would like to keep features available in xtb especially if there are active users. In case of CPCM-X there is currently no maintainer available for fixing bugs or even just creating new releases, if you are able to commit some time to (co-)maintain CPCM-X support in xtb, it would make the support of this feature much more feasible.
We also use CPCM-X for a few different applications at Schrodinger. I think we would be able to contribute towards maintaining this feature.
Does it make more sense to maintain this separate CPCM-X library or to reimplement CPCM-X into tblite? I may be misunderstanding, but it seems like it wouldn't be too extensive an addition to tblite once the ddX interface is added https://github.com/tblite/tblite/pull/235
The problem with implementing this in, e.g., tblite will be, that we fixed #1159 (and #1163). This means, that CPCM-X will not work with the new implementation (i.e., the results will likely be not usable without a rework and parameterization) and it would require us to ship two different implementations of the ddCOSMO scheme.
