Bio: Elif Ertekin is an Andersen Faculty Scholar, Associate Professor, and Associate Head for Graduate Programs in the Mechanical Science and Engineering Department at the University of Illinois at Urbana-Champaign. She is a faculty affiliate of the National Center for Supercomputing Applications (NCSA) and the Materials Research Laboratory (MRL). Her research interests center on the theory and modeling of materials, with an emphasis on probabilistic and stochastic methods. She focuses on developing a microscopic understanding of atomic and electronic scale processes in materials, with applications areas in thermal transport, energy conversion, and defect chemistry. She received BS degrees in Mathematics and in Engineering Science and Mechanics from Penn State, a PhD in Materials Science and Engineering from UC Berkeley, and she carried out post-doctoral work at the Berkeley Nanoscience and Nanoengineering Institute and the Massachusetts Institute of Technology. She is an Associate Editor for the Journal of Applied Physics and a Divisional Associate Editor for
Physical Review Letters.

Physical Mechanisms or Learned Patterns? Reconciling First-Principles Models with Machine Learning for Predictive Materials

Predictive materials simulation has long been rooted in first-principles descriptions of physical mechanisms, grounded in quantum mechanics but limited by tractable length scales, sampling challenges, and the accuracy-cost tradeoff. Today, machine-learning methods seek to transform materials science by revealing patterns in data extending beyond conventional modeling. My talk will explore how these two paradigms, mechanistic simulation and data-driven learning, can act synergistically to accelerate materials discovery and understanding. I will begin by outlining what first-principles simulations can currently achieve and where their limitations arise, using examples from our work in thermoelectrics, wide-band-gap semiconductors, ion-transport materials, and structural alloys. Building on this foundation, I will show how machine-learning approaches, when designed with materials-specific considerations such as symmetries and invariances, can enhance traditional methods. Examples include symmetry-aware generative models for inorganic crystalline solids and machine-learning solutions to the many-body electronic-structure problem that rival high-accuracy quantum methods. Together, these examples highlight how integrating mechanisms and patterns can help advance predictive materials simulations.\

The MICDE 2025-26 Seminar Series is open to all.

This seminar is organized by the Michigan Institute for Computational Discovery & Engineering (MICDE) and the Department of Mechanical Engineering. Prof. Ertekin will be hosted by Prof. Chenhui Shao, Associate Professor of Mechanical Engineering.

This is an in-person event. This seminar will not be recorded!

Graduate Certificate in Computational Discovery and Engineering, and MICDE fellows, please use this form to record your attendance.

Questions? Email [email protected]