Implement k-points sampling about tblite HOT 6 OPEN

Implementing k-point sampling here in tblite would be a really cool feature, but with the integration in DFTB+ we already have such an implementation with both k-points and spin channels available. Therefore, this is currently less of a priority for me.

If more projects start using tblite's high-level API to evaluate xTB singlepoints for periodic systems this might be worth revisiting.

from tblite.

For me it was easiest to get a good k-point implementation by integrating tblite in DFTB+ and leverage the available k-point aware SCF. For an implementation in tblite we can rely on an external library, but it has to be available in the SCF (so in Fortran) to be useful.

from tblite.

You can use xTB in DFTB+ with k-point sampling instead of building supercells.

Also, there is no limit in systemsize, but since we are using a shared memory parallelization you have to provide enough memory for all the threads using the OMP_STACKSIZE environment variable, otherwise it the OMP stack might overflow and the program crash.

from tblite.

Thanks for the comment. Sounds like it's a feature. I'll try building DFTB+.

from tblite.

Hi @awvwgk as a follow up I got dftb+ to compile and for the same system described above it looks like convergence is pretty rapid with a pretty small MP kpoint mesh. However, the converged results do not agree with the largest supercell energy reported above. The gamma point and molecular energies do agree to high precision. Do you have some example crystals with GFN2-xTB energies that you believe are converged? It would be useful for me to make sure I'm getting the right answers.

from tblite.

Implementing k-point sampling here in tblite would be a really cool feature, but with the integration in DFTB+ we already have such an implementation with both k-points and spin channels available. Therefore, this is currently less of a priority for me.

If more projects start using tblite's high-level API to evaluate xTB singlepoints for periodic systems this might be worth revisiting.

I have been testing tblite, DFTB+, and CP2k for a while. tblibe as a stand-alone calculator is the most attractive. Yet, it does luck some features, like k-point sampling.
How difficult is to implement k-points?
How about linking the corresponding module to https://pypi.org/project/kpLib/?

from tblite.