Accelerated Materials Discovery by Active Machine Learning
Optimizing high-entropy alloy
•
High-performance, multi-metallic alloy nanoparticles were discovered using active machine learning and carbothermal shock experiments. The active learning model efficiently navigated through high-dimensional design space with more than 80,000 candidate compositions: the model discovered excellent precursor compositions within 100 experimental trials.
•
Currently, we are expanding our active learning methodologies to include complex yet chemically relevant data structures, e.g., molecules and formulations.
•
MD Simulations Beyond Main-Group Small Molecules
Multi-scale modeling of polymers
•
Microscopic and macroscopic behaviors of polymers can be modeled by multi-scale molecular dynamics simulations, where atomistic information is coarse-grained to allow access to macromolecular and continuum scales. In conjunction with experimental visualization by in-situ liquid-phase TEM, time-resolved conformation dynamics are matched to theoretical and computational results.
•
Recently, we are exploiting different simulation methods to model pattern formation processes in semiconductor manufacturing, verifying the theoretical limit of next-generation approaches to chip fabrication.
•
MLP modeling of organometallics
•
Organometallics form a rich class of materials, with direct chemical bonds between metal and main group elements. We are harnessing machine learning potential (MLP) to perform dynamics simulations of organometallic molecules, expanding the coverage of computational methods beyond main group chemistry. Molecule-level insights lead to improved understanding of chemical mechanisms, and eventually to novel design of high-performance functional materials.
Non-equilibrium Phenomena in Complex Fluids
Crossover in supercritical fluids
•
Combining molecular simulations and machine learning analysis, a novel theoretical framework is proposed to explain the anomalous behavior of supercritical fluids: as a non-uniform mixture of liquid-like and gas-like microstates. Diverging susceptibility of microstate fraction accurately points towards the macroscopic critical point, bridging between the atomistic and continuum scales. With the same scaling exponent, bulk behaviors of different fluid substances can be collapsed to a universal curve, proposing a unified framework to understand supercritical anomalies.
•
Recently, we are interested in near- and far-from-equilibrium phenomena in complex fluids, including long-lived dynamic structures and metastable liquid-liquid phase transitions.
•