Reproduce “Cartesian atomic cluster expansion for machine learning interatomic potentials” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bingqing Cheng
Bingqing Cheng developed the Cartesian Atomic Cluster Expansion (CACE), a machine learning interatomic potential that enforces rotational symmetry entirely in Cartesian coordinates, bypassing spherical harmonics. It demonstrated competitive accuracy, exceptional stability in molecular dynamics simulations up to 5000 K, and strong extrapolation capabilities across diverse materials, including 25-element high-entropy alloys and unseen elements.
Key insights
- It is possible to maintain full rotational symmetry in MLIPs by operating entirely within Cartesian coordinates, providing a mathematically equivalent and simpler alternative to spherical harmonic-based methods.
- Systematic construction of invariant features in Cartesian space can inherently generate a compact, polynomially independent basis set, improving efficiency and avoiding redundancies found in other ACE frameworks.
- Combining a Cartesian-based symmetry approach with modern MLIP components like element embeddings and message passing results in potentials with high accuracy, exceptional stability in MD simulations, and strong extrapolation capabilities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.