Machine Learning for Atomistic Simulation and Materials Discovery
This community develops computational tools that use machine learning to predict the behavior of atoms and molecules, replacing slow quantum-mechanical calculations with faster, data-driven models for simulating materials.
The work centers on creating "interatomic potentials" and "force fields" that allow for large-scale "molecular dynamics" simulations. Researchers frequently apply "density functional theory" and "first-principles calculations" to generate the training data for these models. Key techniques include "graph neural networks" and "active learning" to improve prediction accuracy for "crystal structure" and "electronic structure." Specific applications involve modeling "transition metal" oxides, "oxygen vacancies," and "thin films," with a strong focus on "surface" science and "adsorption" phenomena. The goal is to bridge the gap between quantum accuracy and computational efficiency for discovering new functional materials.
The largest share of this research is found in cerium, accounting for 1.3% of all cerium research, and in oxygen, which represents 0.2% of oxygen research. Oxygen also contributes the highest raw paper count within this group, with 1,134 papers.
The community comprises 9,468 papers, publishing most frequently in The Journal of Physical Chemistry C, npj Computational Materials, and The Journal of Chemical Physics.
Recent work focuses on developing general-purpose neural network potentials for energetic materials and advanced modeling, as well as using AI agents for autonomous discovery in energy materials and electrocatalysis.