New Nanoporous Materials

Peng Bai, associate professor of chemical and biomolecular engineering, seeks to discover new nanoporous materials that are capable of selectively separating mixed compounds. To do so, he is building a machine learning framework that can both rapidly identify the optimal material structure for a given application and predict if such a structure is feasible to synthesize in reality. 

He compares it to a highly sophisticated sponge. “If you put a sponge in oily water, both oil and water will be absorbed, but depending on the application, what we want is for the sponge—the nanopores—to only absorb one of the components.” As such, nanoporous materials have been investigated for a wide range of applications such as water harvesting and separation of critical elements. 

What the “sponge” material will absorb is highly dictated by its structure and shape. With this in mind, Bai and his team are encoding materials by viewing them as 3D analogues of images for their machine learning framework. This project is being funded in part by the Defense Advanced Research Projects Agency (DARPA) Young Faculty Award that Bai received last Fall. 

Bai leads a research lab at UMass Amherst focused on bridging computational modeling, machine learning, and materials science to accelerate the discovery of energy- and environment-related materials. The group integrates insights from chemical engineering, condensed matter physics, and artificial intelligence to tackle complex problems in materials design. The lab has previously contributed to advancements in heterogeneous catalysis, membrane separation, and solid-state electrolytes—all areas in which nanoporous materials play a central role. 

The Materials Genome Initiative—a U.S. effort launched in 2011 to accelerate the discovery and deployment of advanced materials through data and computational tools—has led to the proliferation of large materials databases. However, whether these materials can feasibly be synthesized is hard to determine. “They all look reasonable even to a seasoned synthetic chemist—the computationally predicted materials have the same atomic arrangement and bonding geometry exactly like known materials—but a machine learning model was able to tell 99% of the computationally predicted materials apart from the real materials,” says Bai. 

He also sees this project as a way to explore the integration of machine learning tools with existing bodies of knowledge: “From a research point of view, a very exciting aspect is how we can couple what we know from physics-based thermodynamic nowledge with data-based machine learning tools.”