Machine Learning and Inverse Design for Sequence-Programmable, Multicomponent Materials

Speaker

William Jacobs

Assistant Professor
Princeton University
 

Abstract

Recent advances in physics-informed machine learning and optimization are transforming how we approach materials design. These tools are particularly powerful for sequence-dependent, multicomponent materials, whose properties are governed by high-dimensional spaces of sequences, compositions, and interactions. In this talk, I will present two examples that illustrate both the challenges and opportunities of inverse design. First, I will show how heteropolymer mixtures can be programmed to form many coexisting phases with user-specified compositions, yielding compartmentalized materials with selective partitioning. Second, I will demonstrate how active learning can be used to navigate trade-offs among multiple objectives, producing materials with optimized thermodynamic, kinetic, and transport properties. Together, these examples show how new theoretical and computational tools enable systematic exploration of complex design spaces, providing a practical route to materials design in previously inaccessible regimes.

Bio

William Jacobs obtained a B.S. in Physics and Engineering Science from the University of Virginia in 2010 and a Ph.D. in Theoretical Chemistry from the University of Cambridge in 2014. After completing a postdoc in Theoretical Chemistry and Biophysics at Harvard University, he began his independent career at Princeton University in 2019, where he is also affiliated with the department of Chemical and Biological Engineering, the Princeton Materials Institute, and the Biophysics program.