Faster D3 for MACE

[LesTuche]

While using MACE through ASE I noticed that enabling dispersion correction leads to an almost order of magnitude slowdown in relaxation time compared to running without dispersion.

From the runtime warning it seems to me that this:

torch_dftd3_calculator.py:98: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at /pytorch/torch/csrc/utils/tensor_new.cpp:254.)

Seems to cause the slowdown.

Since torch-dftd is not actively maintained any more I wonder whether it would be possible to switch the backend to the https://github.com/dftd3/tad-dftd3 library as a maintained alternative?

Thank you very much for your help

[ilyes319]

Hey @LesTuche, we have been working on integrating the Alchemi (from Nvidia) D3. @IsaacParker30 did you make a PR already to support that in torch-dft-d3?

[LesTuche]

Great to hear, looking forward to it.

[IsaacParker30]

No. I think @gabor1 and @WillBaldwin0 suggested other stuff was being worked on atm.
I can link my branch here though for now:
https://github.com/IsaacParker30/torch-dftd/tree/alchemi