Machine-Learned Potentials and First-Principles Simulation of Condensed Matter
This community develops computational tools to predict the physical behavior of solids and liquids, specifically using machine learning to accelerate atomistic simulations and first-principles calculations to determine electronic, magnetic, and thermal properties.
The work centers on constructing interatomic potentials and neural network models that replace expensive quantum mechanical calculations for large-scale molecular dynamics. Recurring themes include the prediction of phase transitions, thermal expansion, and mechanical stability in alloys, metallic glasses, and rare-earth compounds. Researchers frequently apply density functional theory and Monte Carlo methods to map crystal structures and electronic band gaps under varying pressure and temperature. The output is a library of predictive models and simulated datasets that describe how materials deform, conduct heat, or change phase, providing the foundational data required for high-throughput materials screening and the design of functional solids.
The community’s output is most concentrated in uranium research, where it represents 3.0% of all tracked papers for that element, and in plutonium research, where it accounts for 3.4%. It also holds a significant share of cerium research at 3.0%. The largest absolute number of papers within this group is found in nickel research, with 1,960 entries.
The community comprises 60,249 papers, with the highest publication volumes in Physical Review B, Journal of Applied Physics, and Physical Review.
Recent work focuses on the development of universal neural network potentials for energetic materials and heterogeneous catalysis, as well as the application of machine learning to accelerate the discovery of energy materials and the simulation of complex phase behaviors in Laves phase compounds.