PBC Question about tblite HOT 6 CLOSED

The issue present in grimme-lab/xtb#257 should be resolved in this implementation, but I'm currently not implementing GFN0-xTB here. For grimme-lab/xtb#491 there is still the halogen bonding energy term for GFN1-xTB missing, but it will not raise an error if those bonds are present. (implemented with awvwgk@778f1c1).

I tested the implementation quite thoroughly and are pretty confident that I got it right. What is missing is some quality of life stuff, like a release, full documentation and binary packages. If you feel brave enough, you can try to compile the library in DFTB+ (see dftbplus/dftbplus#792) or build the Python extension module to use it in ASE. Feedback for both is more than welcome.

from tblite.

I have been able to run tblite from the command line but I am stuck when trying to build the extension module for python. I tried pip and meson both result with this error when trying to run a sp Intel MKL FATAL ERROR: Cannot load libmkl_avx2.so or libmkl_def.so. One note on the meson build for the python module — I had to change the version of tblite here to 0.1.0. I have been using intel compilers with MKL. What compilers/LA backend have you used to build the python extension? Let me know if you have any ideas.

from tblite.

To build Python extension modules with MKL the normal MKL libraries are not usable, instead the real time variant has to linked with -Dlapack=mkl-rt.

from tblite.

To close the loop on this, I tested the python extension module with ASE. Not a ton of feedback — worked as advertised once I got it installed. I tested PBCs with translations and rotations and everything I saw was consistent. I did not test systems with halogen bonding. Any plans to include GFN0? (I ask because GFN1 and GFN2 have convergence issues for some of systems we are interested in)

from tblite.

Thanks for the feedback, much appreciated.

Any plans to include GFN0? (I ask because GFN1 and GFN2 have convergence issues for some of systems we are interested in)

Currently not, GFN0-xTB is still a half-backed parametrization with known issues. Maybe I will provide the possibility to construct something like GFN0-xTB, but I think the way forward is to provide the possibility for creating tailored parametrizations with tblite.

from tblite.

Currently not, GFN0-xTB is still a half-backed parametrization with known issues. Maybe I will provide the possibility to construct something like GFN0-xTB, but I think the way forward is to provide the possibility for creating tailored parametrizations with tblite.

I see. From our perspective, GFN0 isn't perfect but it is fast and always gives an answer which are both good for delta learning. I also think creating tailored parameterization would be very interesting.

Feel free to close

from tblite.