MARA: Continuous SE(3)-Equivariant Attention for Molecular Force Fields

Francesco Leonardi, Boris Bonev, Kaspar Riesen · 2026

Read MARA: Continuous SE(3)-Equivariant Attention for Molecular Force Fields by Francesco Leonardi, Boris Bonev, Kaspar Riesen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02671.txt.

Description

Francesco Leonardi and 2 co-authors add continuous spherical attention to machine-learned interatomic force fields, in a 2026 paper. Such force fields now power accurate atomistic runs, but most rely on fixed angular expansions that limit how freely local shapes can be weighted. Attention, if made rotation-aware, could weight each neighbor's share on the fly.

Their module extends spherical attention from tasks about rotation alone to the full molecular symmetry group, using neighbors' angular and radial coordinates. It handles equivariant interactions at low cost, meaning predictions rotate as they should with the molecule. The module plugs into current models such as MACE with no structural changes. Across benchmarks it sharpened energy and force predictions, cut high-error events, and raised robustness. The authors argue the method is a general operator for atomistic models, not a one-model trick.

Catalog record

CREATORFrancesco Leonardi, Boris Bonev, Kaspar Riesen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-02671

Browse the full library · About the Tower of Babel