DOE Funds AI-Driven Catalyst Discovery Project as Part of Genesis Mission
On October 8, the U.S. Department of Energy (DOE) announced a second round of funding for its Genesis Mission, a nationwide initiative to use AI to tackle pressing challenges in science and technology. Among the projects that will receive funding is a Caltech effort that will use AI to help identify new and better catalysts, materials that speed up chemical reactions.
The catalysis project is led by the DOE's SLAC National Accelerator Laboratory in collaboration with Caltech and Lawrence Berkeley National Laboratory. It is one of six newly announced phase I projects that will receive short-term seed funding to explore feasibility. The DOE also announced 12 new phase II funding awards for larger projects that are ready to scale up.
Catalyst Discovery
To produce useful products from chemical reactions, scientists are always searching for the most effective catalysts. The Liquid Sunlight Alliance (LiSA), a DOE-funded Energy Innovation Hub led by Caltech's Harry Atwater, the Otis Booth Leadership Chair of the Division of Engineering and Applied Science, is specifically looking for catalysts that can drive reactions to convert water, carbon dioxide, and nitrogen into liquid fuels and valuable chemicals.
"This Genesis mission project brings together laboratory-based accelerated materials discovery with X-ray beamline measurements to speed the pace of knowledge generation about how catalysts enable pathways for chemical fuel synthesis," says Atwater, who is also the Howard Hughes Professor of Applied Physics and Materials Science.
To fully understand how certain catalysts work and whether they can be improved, scientists with LiSA have used X-ray beamlines at SLAC's Stanford Synchrotron Radiation Lightsource (SSRL) to measure changes in the physical structures of catalysts as they become active and spur on chemical reactions. That work, though, is extremely time-consuming; although millions of possible catalyst compositions exist—even miniscule changes in the chemical composition can dramatically affect how a catalyst works—only a handful have been tested in this way.
The Genesis-funded project aims to use AI and automated experimentation to speed up the process, enabling the rapid testing and identification of the best catalysts for particular chemical reactions, including the efficient and economically viable creation of liquid fuels.
The catalysis project, "Closed-Loop Autonomous Discovery of Mechanistic Activity-Selectivity-Stability Rules in Catalysis," will build a closed-loop system in which an AI agent will compare competing ideas about how catalysts work, identify what evidence is missing to validate (or disprove) those ideas, and then choose the next experiments that should be conducted.
At SSRL, the system will combine real-time X-ray measurements with electrochemistry, computer models, and published knowledge. Each result will feed back into the system and guide the next measurement. Phase I of the project will test whether this approach can enable scientific understanding with fewer experiments than standard approaches. Faster insight into how catalysts work can help researchers develop more efficient and durable processes for energy and chemical production while advancing the broader vision of autonomous laboratories that can help scientists test ideas on faster timescales.
"The AI weighs the scientific evidence, decides which measurement would tell us the most, and that decision becomes a real experiment at SSRL," said Dimosthenis Sokaras, the project's principal investigator and director of the Chemistry and Catalysis Division at SSRL, in an SLAC press release. "The result comes straight back into the reasoning process. That closed loop is a key step toward autonomous laboratories."
As part of the new project, Caltech researchers will prepare libraries of hundreds to thousands of catalyst compositions to test with SLAC's X-ray beam, says Caltech's Joel Haber, a member of the professional staff and team lead with LiSA. "We can measure how well the material is performing as an electrocatalyst at the same time we're measuring the structure of the catalyst. So, we will be able to see what changes when it turns on and becomes effective as an electrocatalyst and then correlate the structure with that performance."
That data will be combined with electrochemical measurements from labs at Caltech along with information from computational chemists in the collaboration to constantly update the AI model.
Two Caltech-led Genesis Mission projects were announced in the first round of DOE funding for the initiative in July. At the same time, a "programmable cloud laboratory" led by Caltech and funded by the National Science Foundation (NSF) was announced as part of a network of AI-enabled automated laboratories that represent a core NSF contribution to the nationwide Genesis Mission.