AI learns transferable 3D density functional from equilibrium data

Researchers have developed a method to learn a transferable three-dimensional classical density functional directly from equilibrium density fields. This approach preserves spatial symmetry and variational consistency, and does not require free-energy or chemical-potential labels for training. The learned functional demonstrates transferability across different temperatures, system sizes, and statistical ensembles, accurately reproducing structure factors, equations of state, and liquid-vapor coexistence without being explicitly trained on these properties. The method was applied to complex 3D geometries, predicting phenomena such as non-monotonic forces in colloidal interactions and adsorption within gyroid pores, showcasing its ability to connect microscopic liquid structure to macroscopic thermodynamic behavior. AI

IMPACT Enables more accurate and efficient simulations in materials science and condensed matter physics.

RANK_REASON Academic paper detailing a new methodology for learning scientific models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Classical density functional theory methods in soft and hard matter
• colloid
• equation of state
• equilibrium density fields
• gyroid pore
• Hugging Face
• liquid--vapor coexistence
• thermodynamics

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →