 |
Center for Nonlinear Studies |
Further advances in technology increasingly depend on our ability to predict how materials behave across scales, from angstrom-level bond breaking to bulk rheological response. One effective way to bridge these scales is to use simplified descriptions of matter through coarse-grained (CG) modeling, in which many atoms are grouped into fewer effective CG units that capture the dominant large-scale behavior. Machine learning (ML) methods have recently been shown to outperform traditional physics-based models in the accuracy of these descriptions. State-of-the-art models perform well for small molecules and simple fluids, but extending them to polymers, biomolecules, and heterogeneous systems remains a challenge. Most ML architectures rely on distance-based message-passing, where the interactions between effective point-like particles are inferred solely from local geometry. Without orientation information, they are unable to separate bonded and non-bonded interactions and instead learn an averaged response, which compromises structural integrity in systems with directional bonding or rigidity and limits generalizability across molecular sizes, phases, and chemical environments. Moreover, choosing the scale of CG units forces a trade-off between preserving local chemical detail and capturing large-scale structure. In this talk, I will outline the scientific context underlying this problem, and present strategies we propose to address it. These include introducing internal variables, allowing bonding patterns to emerge dynamically through probabilistic learning, and highlighting related efforts moving in similar directions. Overall, we aim to develop generalizable structure-aware models that retain bonded interactions and molecular identity, strengthening LANL’s capacity for predictive modeling of complex materials.Bio: Feranmi Victor Olowookere is a Ph.D. candidate in Chemical Engineering at the University of Alabama (UA), advised by Prof. Heath Turner. His research develops computational frameworks for molecular modeling of solvated polymers, such as PVC, to advance plastic waste upcycling and materials design. Victor earned a B.S. in Chemical Engineering, summa cum laude, from the University of Lagos, Nigeria, in 2022, after which he moved to the United States the same year to begin a fully funded Ph.D. program at UA and has since completed an M.S. in Chemical Engineering, summa cum laude. Throughout his doctoral studies, he has published several peer-reviewed papers as lead author and delivered multiple presentations at national and international conferences. His work has earned national and institutional recognition, including the AIChE Computational Molecular Science and Engineering Forum Graduate Student Award, the American Institute of Chemists Award, competitive fellowships at Los Alamos National Laboratory and UA, and departmental honors such as Outstanding Research by a Doctoral Student. Beyond his research, Victor has served the academic community as a mentor, guest lecturer, judge, and ambassador.
Host: Emily Shinkle