Building Materials for the Future
Postdoctoral Researcher Uses Machine Learning to Accelerate Materials Discovery
- Working with Professor of Chemistry Davide Donadio, postdoctoral researcher Frank Cerasoli uses computer simulations to model materials at the molecular level, with the hope of discovering new materials that can advance our technologies.
- Cerasoli studies crystal structures called clathrates, a class of inclusion compound that can trap other atoms or molecules in its cage-like structure.
- Specifically, he's interested in inorganic clathrates due to their potential as thermoelectric materials.
Beneath the concrete world discernible to our senses is a world of building blocks. A world of molecules, and beneath that, atoms. The organization of these individual parts dictates the properties of materials.
For Frank Cerasoli, a UC Davis postdoctoral researcher working with Professor of Chemistry Davide Donadio, the geometry of this hidden world is what initially drew him to practice science.
“The visual imagination is one of the most powerful tools we have,” said Cerasoli, who uses computer simulations to model materials at the molecular level. “It’s estimated that there are up to a googol of real materials that could exist, but how can we really access these things? Surely, we don’t have a chance of synthesizing all of them.”
As our technologies advance, so too must the materials that build them. Theoretical materials discovery holds immense promise to revolutionize this endeavor. And Cerasoli is at the ground floor of this work.
Building crystal structures
Cerasoli studies crystal structures called clathrates, a class of inclusion compound that can trap other atoms or molecules in its cage-like structure. In the second half of the 20th century, scientists started synthesizing and studying inorganic clathrates due to their potential as thermoelectric materials.
“These inorganic clathrates offer a lot of interesting properties, such as superconductivity," Cerasoli said. “They are promising thermoelectrics, which can harvest electricity from temperature differences.”
“If your phone gets too hot, maybe you can recycle that heat into electricity and store the wasted energy,” he added, noting a potential application of inorganic clathrate materials. “This is kind of the primary goal of the thermoelectric regime.”
Because of the unique thermoelectric properties of inorganic clathrates, scientists are seeking ways to hasten the materials discovery process. In the 1970s, scientists employed a computational method called density functional theory to hypothesize new materials. But that method is computationally expensive, according to Cerasoli.
Recent advances in machine learning, however, have enabled computational materials scientists like Cerasoli to reduce the computational requirements for predicting new materials and their chemical properties while also maintaining accuracy.
“What I do now is sort of a hybrid approach where I combine density functional theory calculations, the solid framework that we know and love, with machine learning methods,” Cerasoli said.
Since starting his research, Cerasoli has simulated roughly 15,000 clathrate compounds. Some are useable, others are not, but regardless, Cerasoli uses those simulated compounds as a training set to improve the prediction capability of his clathrate-generating machine learning models.
“No result is just sent under the bridge to run downstream,” he said. “We capture everything, reintegrate it into our models and try to do better.”
The goal is to accurately predict the stability and energetic favorability of these simulated clathrates and pass that information to collaborators who can synthesize the materials and test them in the real world.
“This is really a data-driven process,” Cerasoli said. “These computational tools really provide a route to identifying potential materials or shrinking the subspace of possible materials. We’re removing configurations that would never be physical, so really cleaving down the realm of possible materials into something that is more tangible.”
The path to UC Davis
With a year under his belt at UC Davis, Cerasoli has been reflecting on his experience. He praised Donadio, his lab colleagues and the University of California system for its strong scientific network. The two met when Cerasoli was a doctoral student at the University of North Texas where they were both attending a materials simulation boot camp. Donadio mentioned his lab had an opening for a postdoctoral scholar specializing in materials discovery.
“I didn’t think twice,” Cerasoli said of the following opportunity.
For Cerasoli, UC Davis has provided a proving ground for his work.
“Computational materials science is an experimental science in a lot of ways,” Cerasoli said. “You really have this experimental platform that you can test against nature, test against reality. Building models that simulate the real world is the interesting component to me.”
To learn more, visit the Donadio Lab website.