RT @marceldotsci: ✨preprint alert: "learning long-range representations w/ equivariant messages" in which we get into the fray of long-rang….

Many thanks to everyone who helped with this development effort, in particular @PhilipLoche , Fillipo Bigi, @jwasci, @MicheleCeriotti and many others at @lab_COSMO, their help was invaluable in getting this project out and useful!. 10/n, n=10!.

We are also open to adding other representations you use for your research, please let us know what you would like to see! 9/n.

Documentation is available here (, in a rough state but complete. The next step for us will be to improve this documentation, and then bring GPU acceleration in the mix! 8/n.

As usual, everything is open source (, and you can now use `pip install featomic` or `pip install featomic-torch` to install it! 7/n.

Featomic output is using the metatensor format, keeping the data as sparse as possible to minimize memory usage; and fully annotating the representation with metadata to help you understand the complex objects you are manipulating. 6/n.

We also provide an integration with TorchScript to use your machine learning models based on featomic inside LAMMPS/GROMACS/PLUMED/. without keeping Python around (more on this coming soon!) 5/n.

The code is parallelized, able to compute gradients with respect to positions, cell and strain; and available from multiple languages: Python of course, but also C, C++ and Rust. 4/n.

We also provides tools to build more complex representations in Python, such as high body order equivariants (i.e. ACE) and multi-centers equivariants, for example to describe properties of pairs of atoms when learning an Hamiltonian matrix. 3/n.

Featomic implements the SOAP PowerSpectrum and related representations, such as the 2 body radial spectrum, and the equivariant spherical expansion. We also implemented LODE long range equivariant features; and many more! You can find everything here: 2/n.

Happy new year everyone! After more than three years of work, and splitting out a whole separate package, I am extremely happy to announce the first full public release of featomic, a package to compute representations for atomistic machine learning! 1/n.

RT @marceldotsci: New: Fast and flexible range-separated models for atomistic machine learning, spearheaded by Kevin and Philip, with many….

RT @marceldotsci: are you using vesin yet? it's really good!.

RT @lab_COSMO: Hello! I'm posting this both here and on the better place, let's see where it gets more re-posts 😇. We are looking for a res….

See for benchmarks, the code is pretty fast on CPU!.

Vesin provides a Python API, as well as a C API, intended to be dropped inside your own code in a single file. We also expose TorchScript bindings for your ML models that need to compute their own neighbor list.

I'm getting into the habit of announcing new code releases. Today's release is version 0.3.0 of vesin (, a small library to compute neighbor/pair lists for atomistic systems! Now with an option to sort the pairs and integration with metatensor models.

RT @lab_COSMO: It is time for jack-o-molecule 🎃! Happy chemiscoping

RT @lab_COSMO: Then, we added some cosmetic options for the map viewer, to hide point outlines, or to draw a line to show a trajectory. Noo….

RT @lab_COSMO: First, @andrew_tarzia added preliminary support for defining bonds manually, which is useful when displaying non-standard st….