Researchers have developed new methods to improve the accuracy and scalability of machine learning interatomic potentials (MLIPs) for materials modeling. One approach, DPA4C-LR, analytically separates long-range electrostatic forces and Born effective charges into a global calculation while preserving local MLIP properties, enabling simulations of up to 5.2 million atoms. Another method frames the decision of when to use expensive first-principles calculations (like DFT) versus cheaper MLIPs as a risk-controlled selection problem using a conformal risk control framework, which aims to keep MLIP errors below a user-defined threshold. AI
IMPACT These advancements could accelerate materials discovery and simulation by improving the efficiency and reliability of MLIPs, reducing reliance on computationally expensive first-principles methods.
RANK_REASON The cluster contains two academic papers detailing new methods for machine learning interatomic potentials.
Read on Hugging Face Daily Papers →
- Atom-centered symmetry functions for constructing high-dimensional neural network potentials
- Conformal Risk Control
- density functional theory
- Hugging Face
- Smooth Overlap of Atomic Positions
- arXiv
- Diamond
- DPA4C-LR
- Li_{0.9}NbO_{0.9}Cl_{4.1}
- nvalchemi
- Pb(Mg_{1/3}Nb_{2/3})O3
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
