The MICDE PhD Student Seminar Series showcases the research of students in the Ph.D. in Scientific Computing. Lunch will be served. These events are open to the public, but we request that all who plan to attend register in advance. Planned sessions will be canceled if no one signs up to present, and registered attendees will be notified.

If you have any questions, please email [email protected].

Register to attend

Accelerating Molecular Simulations with Multilayer Machine-Learned Interatomic Potentials

Machine-learned interatomic potentials (ML-IAPs) enable near-quantum-accurate molecular simulations at significantly lower computational cost than first-principles methods. However, accurately modeling covalently bonded systems requires resolving both highly featured short-range interactions and smoother long-range interactions, creating a trade-off between model complexity and computational efficiency. We address this challenge with a multilayer representation that decomposes the potential energy surface into separate short-range and long-range models, implemented within the ChIMES ML-IAP framework. Evaluated on propane, water, and reactive C/O systems spanning diverse thermodynamic conditions, the multilayer approach matches the accuracy of conventional single-layer models while reducing computational cost by more than an order of magnitude. A hyperparameter sensitivity analysis further provides practical guidelines for model design. These results establish multilayer ChIMES as a general framework for substantially improving ML-IAP efficiency without compromising predictive accuracy.

Awwal Oladipupo (Chemical Engineering and Scientific Computing)

Awwal Oladipupo is a Ph.D. candidate in Chemical Engineering and Scientific Computing at the University of Michigan. His research focuses on developing machine learning-enabled atomistic simulation methods, multiscale modeling frameworks, and scientific software to accelerate the discovery and design of materials for energy, environmental, and chemical applications, with an emphasis on improving the accuracy, scalability, and computational efficiency of molecular simulations.