Request: share forces and velocities

[SiLiKhon]

Dear Authors,

I've been reviewing the shared dataset and noticed that only the atomic positions are available for the trajectories. Would it be possible to include the forces and velocities as well? I think that including them in model's input may be quite useful for inference scenarios where short molecular dynamics trajectories are extended with machine learning models.

Thank you for your work and assistance!

Best,
Artem

[recisic]

Hi @SiLiKhon, we also have the forces and velocities for the trajectories, but we could only upload the positions due to Zenodo’s file size limits. If you have a specific use case for the full trajectory data, we can arrange a transfer through our compute clusters.

[SiLiKhon]

Hi @recisic, thanks for your reply and apologies for my slow reaction.

I didn't realize there was such a limitation on Zenodo. Since we probably don't need the forces for each frame, I can likely just re-run MACE-MP-0 on some of the shared images. This wouldn't give me velocities of course, but maybe that's fine for now. Would it possible though to share maybe 1-2 trajectories with forces, so that I validate that MACE is giving me the same output? I believe that would make a really small file that's attachable here.

Our use case is similar to the one from your paper, although we hypothesize that forces also contain relevant information for predicting diffusion at larger time scales. We have a small study done by my student. I think it be curious to compare our approaches, and especially check the generalizability across different data (we used sevennet for our trajectories).