Four connected directions
Research
Machine learning and first-principles theory for atomistic discovery, predictive modeling and materials design.
View full-size figure ↗Learning atomic interactions and exploring reconstructed surfaces with machine-learning potentials.
Active learning for atomistic discovery.
Machine-Learning Atomistic Simulation
머신러닝 원자 시뮬레이션
Finding a useful atomic structure begins with deciding which part of the energy landscape to explore. We develop and fine-tune machine-learning interatomic potentials from first-principles data, using active learning to identify unfamiliar environments and select new calculations. This feedback between learning and simulation enables broader structure searches and larger-scale atomistic studies.
Active learning · Foundation potentials · Fine-tuning · Structure exploration
View full-size figure ↗Experimental TiO₂ reconstruction and simulated water dissociation at BiVO₄ interfaces are shown as separate examples.
ML-guided exploration of working interfaces.
Dynamic Surfaces & Interfaces
동적 표면과 계면
Working surfaces can differ fundamentally from ideal crystal terminations. We use machine-learning potentials to search for reconstructions, then refine the resulting structures with first-principles calculations and investigate their interfaces through atomistic simulation. This connects a wider search of possible surfaces to the local coordination, adsorption and proton-transfer processes that govern their chemistry.
ML-guided surface search · DFT refinement · Interface dynamics
View full-size figure ↗Combining kinetic models, atomic-scale calculations and experiments to study catalyst stability.
Data-driven prediction of catalyst stability.
Catalytic Activity & Stability
촉매 활성과 안정성
A catalyst must sustain activity as its atomic sites and particle sizes evolve. We connect first-principles reaction mechanisms with data-driven models to investigate activity and degradation together. Hybrid modeling combines physical kinetics, atomistic calculations and experimental observations to predict particle ripening and assess uncertainty, while mechanistic studies resolve the active sites and dissolution pathways that shape catalyst lifetime.
Hybrid modeling · Uncertainty quantification · Active sites · Ripening & dissolution
View full-size figure ↗Tungsten-bronze structures and optical responses, alongside surface oxidation and hydrogen evolution on VSe₂. The separate ML panel illustrates a published prediction method.
Learning structure–property relationships.
Functional Materials Design
기능성 소재 설계
Designing a functional material requires navigating many possible compositions and atomic arrangements. We combine machine-learning property prediction and candidate screening with first-principles calculations to investigate structure–property relationships and identify promising materials. Electronic, spin, ionic and optical responses provide complementary targets for connecting atomic-scale design choices to useful function.
ML property prediction · Candidate screening · Structure–property relationships
Join CCML
Undergraduate research
& graduate study
Interested in graduate study? Prior coding or computational chemistry experience is not required.