Professor Sijia Dong is using machine learning to uncover the mysteries of molecules and materials. In a recent publication in Chemical Science, Dr. Dong and co-authors developed a new computational strategy that can be used to understand how complex materials behave when they are exposed to sun light.
This kind of research is instrumental across scientific fields—understanding everything from photosynthesis to photovoltaic cells (aka the building blocks of solar panels) rely on computational simulations.
“Theory and computation are essential in understanding light-matter interactions, to explain experimental results, and to provide key insights in designing materials. We want to be able to not only simulate light-matter interactions but also consider conditions comparable to experimental conditions in the simulations, such as at finite temperature. Neither is easy,” Dr. Dong detailed. During her postdoctoral work at the Argonne National Laboratory (ANL), Dong worked with Prof. Giulia Galli of the University of Chicago and ANL and assistant scientist Dr. Marco Govoni of ANL to simplify the solution of quantum mechanical equations to describe the way light is absorbed by solids, liquids, molecules, and various other interfaces.
“Currently people don’t have a feasible way to simulate everything exactly,” said Dong, “so what the field of theoretical and computational chemistry does is to make approximations in physical equations so that the properties of the molecules and materials we care about can be simulated in a reasonable amount of time with reasonable accuracy.”
Involving machine learning was Dong’s way of improving the efficiency of the simulation process. In this work, for a solid-liquid interface (such as an electrode-liquid interface found in batteries), using the original method the calculation would take 240 computing cores over several days to solve the equations. With this novel method, it only takes one computing core and several hours, speeding up the process 200 times.
From here, Dong and her new Northeastern lab utilize machine learning techniques to improve physics-based computational methods for other corners of chemical and materials research. “We usually start from an application that is interesting and important, and then we try to figure out what computational methods or tools can help us solve that problem,” Dong said.
Despite this success and the potential widespread application, Dong has identified challenges in theoretical and computational research in chemistry and materials science. First and foremost is communication, or lack thereof.
“Nowadays science needs multidisciplinary efforts,” Dong said. “Then you either make the other people understand your language or you invent a new language that everybody can understand. Each field has its own body of knowledge that cannot be learned in a few minutes or hours.”
Although computational quantum chemistry seems like a daunting field, Dong said that a lot of people give up before they even give themselves a chance. Many researchers outside of this field create a mental block around things they have not learned, and thereby prevent themselves from ever learning it.
“Sometimes it is just a psychological mindset preventing people from joining technical fields, not their ability,” Dong said. She has set out to fix that problem by proposing a 5000-level computational course designed to help students find ways that these methods apply to their own interests and experiments. The students would learn a foundation of computational chemistry, including both physics-based and data-driven methods, and applying these concepts to project proposals derived from real-world research problems. This course would help build a community of researchers who can effectively communicate and work with each other.
Nonetheless, Dr. Sijia Dong’s Theoretical and Computational Chemistry Lab will continue to develop high-powered design strategies to better understand the loose puzzle pieces of the chemical world.
“My colleagues are very supportive in my process of establishing my own lab,” Dong said. Professor Dong is happy to be working with motivated and hard-working undergraduate students, graduate students, and postdoctoral scholars to develop more state-of-the-art computational methods.