Overview
Modern advances in quantum-chemical calculations and machine learning have ushered in a new age of computational materials science. While there is much excitement about accelerating the design and discovery of new materials, this is often viewed from the binary lens of a material being stable or unstable. Reality, however, is much more complex. Many of the most useful and interesting materials are not in their thermodynamic ground state at all. Instead, these materials can be metastable: higher-energy, kinetically trapped structures whose distinctive physical and chemical properties can make or break their utility.
The Rosen Research Group consists of quantum-chemical engineers who combine high-throughput computing and machine learning to study materials at the edge of stability. Our work spans two broad areas:
- Materials pushed to their thermodynamic and kinetic limits, wherein we predict whether a material can be synthesized, how it decomposes, and the conditions under which it may exist.
- Materials pushed to their electronic limits, wherein we predict how electrons can occupy unusual states, resulting in entirely new kinds of chemical reactivity.
Our work is motivated by applications in energy and sustainability science, including heterogeneous catalysis, energy storage and conversion, and separation processes.
Source: C&EN
Modeling Thermodynamic and Kinetic Stability Limits
In recent years, computational methods have advanced to the point where we can identify promising materials for a wide range of applications. However, a major question is often left unanswered—how do we know if a newly proposed material is experimentally realizable, and what factors dictate its temperature- and pressure-dependent stability? Historically, stability has been judged based on idealized, static crystal structures and energies at 0 K. We develop and apply machine learning interatomic potentials to better bridge the gap between theory and experiment by predicting the decomposition mechanisms and finite-temperature stability of metastable materials under experimentally relevant operating conditions.
Engineering Exotic Electronic States for Unique Chemical Reactivity
Heterogeneous catalysis—which underlies the production of most manufactured chemicals—is often limited by intrinsic tradeoffs between key reaction steps, which can be traced back to how electrons transfer to and from reacting species. One way to escape these constraints is to design materials with exotic electronic structure properties that break from conventional design rules. We are using quantum-mechanical calculations to understand how such properties can arise in chemically diverse materials, such as electrides: materials in which electrons reside in the void spaces of a crystal structure, where they can serve as reactive sites in their own right. We are also developing new classes of electron-aware machine learning models that move beyond the atom-centered representations conventionally used to describe materials.
Source: Matter
Source: PBS
Democratizing Computational Materials Science
To fully embrace the modern age of machine learning and artificial intelligence, there is a substantial need for large-scale, high-quality data. We curate open datasets of quantum-mechanical properties for materials and develop open-source scientific software to democratize complex computational materials chemistry workflows for the broader community to build upon. We believe that the tools and data behind the work that we do is as important to share as the results themselves.