Multilayer atomic cluster expansion for semi-local interactions

Abstract

Traditionally, interatomic potentials assume local bond formation supplemented by long-range electrostatic interactions when necessary. This ignores intermediate range multi-atom interactions that arise from the relaxation of the electronic structure. Here, we present the multilayer atomic cluster expansion (ml-ACE) that includes collective, semi-local multi-atom interactions naturally within its remit. We demonstrate that ml-ACE significantly improves fit accuracy compared to a local expansion on selected examples and provide physical intuition to understand this improvement.

BIP: Boost Invariant Polynomials for Efficient Jet Tagging

“…The ACE framework was recently extended as a general framework for equivariant message passing [40,41] and has proven highly successful for modeling interatomic interactions [42,43] typically outperforming other approaches despite employing much shallower architectures. This work suggests that BIP model could also be extended to a geometric deep-learning framework which would likely further improve the already excellent accuracy of the BIP model.…”

Atomic cluster expansion and wave function representations

“…The effective single particle functions could be extended and refined further by adapting recent extensions of ACE in natural ways to include message passing to electrons 4,5 along similar lines as in the PauliNet archtecture 10 .…”

“…It is a complete representation 2 , which means that any conceivable form of the interaction between atoms can in principle be represented. The framework has been extended to include magnetism and charge transfer, to handle vectorial and tensorial objects 3 as well as message passing networks 4,5 . Efficient implementations 6 and parameterization software [7][8][9] are available.…”