Collaborative Research: EAGER: ADAPT: Machine Learning Thermodynamic Speed Limits for Dynamic Materials

With support from the Chemical Theory, Models and Computational Methods (CTMC) program in the Division of Chemistry and the Office of Multidisciplinary Activities (OMA), Jason R. Green of the University of Massachusetts Boston and Igor Mezic of the University of California-Santa Barbara will work to advance the fundamental understanding of how to regulate transformations of energy in chemically-active materials.

To benefit applications across the energy, biomedical, and healthcare industries, it is necessary to design materials that execute functional behaviors on chosen time scales. Predicting these dynamical processes requires new theoretical methods to simultaneously navigate their large design space, control the timing of dynamical functions, and regulate the dissipation of energy.

This project aims to address this need by combining machine learning and physical theory to create new methods for the design and optimization of functional materials with tailored optical, mechanical, or photonic properties on finely tuned time scales. Coupled to these scientific aims, the project will collaboratively create an active learning curriculum to teach chemists the statistical techniques of data science and contribute to the training of a diverse AI(artificial intelligence)-aware workforce.

Materials chemistry now aims to create dissipative materials that function dynamically, forming patterns and generating work on finite time scales. Recent experiments have taken the first steps to identify chemical systems that drive transient formation of materials structures. However, further progress requires navigating their large design space and regulating flows of energy from the nanoscale up.

Machine learning has potential to guide experiments and accelerate this process but is not yet able to optimize the energy efficiency and timed delivery of structure. The proposed project will address this challenge by strategically incorporating recent advances in statistical mechanics into predictive models from machine learning. The specific objectives will be to (i) construct the data-driven dynamics of active hydrogels with techniques from AI, (ii) show that thermodynamic speed limits can be cast as optimally predictive models in machine learning, and (iii) implement these speed limits as design principles for maximizing yield and minimizing dissipation.

The project includes dedicated activities to develop strength in STEM (science, technology, engineering and mathematics) at the intersection of data science and theoretical chemistry and to broaden participation in STEM through targeted outreach.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.