Equivariant learning of a transferable three-dimensional classical density functional
A single 3D-learned functional reproduced liquid structure, phase behavior, and confined-geometry forces without being trained on those targets.
Bingqing Cheng’s paper says the model learns an excess free-energy functional directly from three-dimensional equilibrium density fields while preserving spatial symmetry and variational consistency. It does not require free-energy or chemical-potential labels. The learned functional transfers across temperatures, system sizes, and statistical ensembles, then recovers structure factors, the equation of state, liquid-vapor coexistence, and interfacial broadening. In complex geometries, it predicts bridge-formation forces between colloids and adsorption in a gyroid pore. ArXiv · AI/CL/LG's note
Bingqing Cheng’s paper says the model learns an excess free-energy functional directly from three-dimensional equilibrium density fields while preserving spatial symmetry and variational consistency. It does not require free-energy or chemical-potential labels. The learned functional transfers across temperatures, system sizes, and statistical ensembles, then recovers structure factors, the equation of state, liquid-vapor coexistence, and interfacial broadening. In complex geometries, it predicts bridge-formation forces between colloids and adsorption in a gyroid pore. ArXiv · AI/CL/LG's note
score 5