aiqbench analysis
FlashCart introduces a novel architecture for equivariant interatomic potentials that significantly accelerates atomistic simulations by optimizing GPU kernels and Cartesian tensor products. This approach allows models to achieve higher accuracy with far fewer parameters compared to standard transformer-based methods, effectively pushing the efficiency frontier for machine-learned materials modeling.
FlashCart: Fast Cartesian Tensor Products for Equivariant Interatomic Potentials
arXiv:2610.06409v1 Announce Type: new Abstract: Machine-learned interatomic potentials extend atomistic simulations beyond the length- and timescales accessible to electronic-structure methods. However, the computational cost of equivariant architectures limits the local correlations they can represent in practice and therefore their achievable accuracy. Here we introduce FlashCart, which makes higher-order correlations affordable by combining generated GPU kernels with an architecture that recursively builds equivariant features and compresses them to a fixed width at each step. We express tensor products in independent Cartesian components and symbolically simplify them and their derivatives, producing fused kernels that often outperform optimized spherical counterparts. We then show that increasing correlation order improves accuracy more efficiently than increasing width, depth, or tensor rank. On SPICE-MACE-OFF, FlashCart models advance the measured accuracy-efficiency frontier: a model with $5.6$ million parameters achieves lower energy and force errors and $10\times$ faster inference than a transformer with $189$ million parameters.
Abstract:Machine-learned interatomic potentials extend atomistic simulations beyond the length- and timescales accessible to electronic-structure methods. However, the computational cost of equivariant architectures limits the local correlations they can represent in practice and therefore their achievable accuracy. Here we introduce FlashCart, which makes higher-order correlations affordable by combining generated GPU kernels with an architecture that recursively builds equivariant features and compresses them to a fixed width at each step. We express tensor products in independent Cartesian components and symbolically simplify them and their derivatives, producing fused kernels that often outperform optimized spherical counterparts. We then show that increasing correlation order improves accuracy more efficiently than increasing width, depth, or tensor rank. On SPICE-MACE-OFF, FlashCart models advance the measured accuracy-efficiency frontier: a model with $5.6$ million parameters achieves lower energy and force errors and $10\times$ faster inference than a transformer with $189$ million parameters.
Machine Learning (cs.LG)
arXiv:2610.06409 [cs.LG]
(or arXiv:2610.06409v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.06409 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 236 words
Related coverage
A Cartesian-3j Framework for Machine Learning Interatomic Potentials
arXiv · Jun 3, 2026
E3J: An Efficient and Open-Source Backend for Euclidean Equivariant Operations on GPU and TPU
arXiv · Sep 30, 2026
BranchIP: Learning Adaptive Equivariant Computation for Interatomic Potentials
arXiv · Oct 2, 2026
LunaTP: Enabling Efficient and Scalable Tensor Product for Equivariant MLIPs
arXiv · Oct 14, 2026