Abstract
Atomistic modeling has long been a cornerstone of materials science, yet achieving both quantum accuracy and the scale needed to explore complex materials has remained a central challenge. Machine learning (ML) is helping overcome this challenge, enabling simulations that are simultaneously accurate, transferable, and efficient enough to screen vast chemical spaces and directly guide materials design and experimental synthesis.
In this talk, I will illustrate how integrating machine learning interatomic potentials with generative algorithms and finite-temperature thermodynamic modeling creates a tightly coupled theory-to-synthesis pipeline for the discovery of new materials with extreme and unconventional transport properties. In close collaboration with experimental labs, we are developing computational tools that enable the predictive synthesis of new intermetallic compounds. Leveraging systematically benchmarked universal ML models with atomistic thermodynamic calculations over ~104 ternary and quaternary compounds, we have identified promising clathrate and Zintl phases, with predictions directly informing synthetic efforts. Notable results include the discovery of new pseudoclathrates Ba2Zn5(As, Sb)6, new phosphide clathrates with promising thermoelectric properties, and a thorough exploration of the Ba-In-Sb chemical space that led to the synthesis of six new compounds, among which BaIn4Sb4 features ultralow thermal conductivity.
I will also present a broad computational search for materials with extremely high thermal conductivity. Combining generative algorithms with fast, quantum-accurate ML models and extending the exploration from carbon-based solids to hundreds of thousands of crystals across the full periodic table, we identify 34 carbon polymorphs with room-temperature thermal conductivity exceeding 800 Wm-1K-1, 18 semiconducting compounds surpassing silicon, and binary metallic systems combining anomalously high electrical and vibrational thermal conductivity, a combination not previously reported.
Davide Donadio is a professor of chemistry at the University of California, Davis (UC Davis). He holds an MS in physics (1998) and a PhD in materials science (2003) from the University of Milan. His academic journey includes postdoctoral positions at the Swiss Federal Institute of Technology (ETH) Zurich and UC Davis, as well as leading an independent junior research group at the Max Planck Institute for Polymer Research in Germany. Before joining UC Davis, he served as an Ikerbasque Professor at the Donostia International Physics Center in Spain. His research focuses on advancing atomistic and first-principles simulation methods to explore the physicochemical properties of solids and liquids, including ice, aqueous solutions, and energy materials. He is an elected fellow of the American Association for the Advancement of Science and of the American Physical Society.