(2aj) Sustainable Chemistry with Machine Learning and Multi-Scale Simulations

Developing sustainable and novel chemical processes is core to humankind’s efforts to mitigate the effects of climate change and create a safer environment for future generations. My research focuses on using computational tools to understand the unknown chemistry behind these complex chemical processes, which often involves experimental collaborations. I started my research career while pursuing my M.S. degree at IIT Madras. During this time, I studied pollutant (NOx) formation in biodiesel powered combustion engines. According to the EPA, transportation sector accounts for 38% of human generated CO₂ emissions in 2021. Renewable fuels along with other sustainable mobility solutions such as electric vehicles could significantly reduce these CO₂ emissions. I developed a reaction mechanism for biodiesel combustion that predicted the experimentally observed NOx formation using computational fluid dynamics (CFD) modelling. This work helped us understand the factors governing NOx formation in engines. During my Ph.D. at Penn State, I continued my research interest in biofuels, but at an atomistic scale. I used classical reactive molecular dynamics (RMD) simulations to explore the reaction kinetics of novel aviation biofuels. These simulations allowed us to identify the best biofuel candidates based on their properties for aviation applications. We also modeled the effect of strong electric fields on jet fuel oxidation reactions which is important for proposed hybrid thermal-electrical propulsion systems. These research projects helped me gain expertise in molecular dynamics simulations and proficiency in quantum chemistry and continuum simulations.

During my postdoc at Princeton University, I was able to leverage this expertise to model the recycling of plastics in a novel thermochemical reactor using RMD. This reactor was used to recycle common plastic polymers into their monomers through pyrolysis in a catalyst-free process. My modeling of the plastic pyrolysis process enabled our experimental collaborators to understand and optimize the recycling process. We obtained one of the highest monomer yields from a catalyst-free plastic recycling process and these results have recently been published in Nature. Another focus of my postdoc research has been plasma catalytic synthesis of NH₃, which promises to be an alternative to the CO₂intensive Haber-Bosch process. I have shown through density functional theory (DFT) calculations that surface charge could play an important role in plasma catalysis. Hence, surface charge could potentially be used to control the NH₃synthesis process. Although the atomistic simulation techniques described above are useful tools, they are limited by either computational expense (DFT) or accuracy (classical RMD). Machine learning (ML) RMD potentials can help us overcome this barrier. I have been actively working on developing ML potentials for H₂ and NH₃ chemistry. Our primary results with H₂ combustion ML potentials show an order of magnitude higher accuracy compared to classical RMD. Using these results, I helped my advisor write a recently approved DOE grant. This funding will help me continue the development of machine learning potentials.

Despite their unprecedented accuracy, machine learning potentials incur large computational cost during their training. Looking towards the future, my research group will focus on changing this status quo by taking advantage of machine learning techniques to use existing computational chemistry datasets for training the RMD potentials. If accurate RMD potentials can be trained using existing datasets, that will significantly reduce the computational cost for discovering unknown chemical pathways in complex chemical environments. Another sub-area of focus would be predictive and quantitative multi-scale modeling. For example, my group will quantitatively model non-equilibrium catalytic synthesis of chemicals such as NH₃. These non-equilibrium processes have shown the promise to replace conventional CO₂intensive processes. I have had a chance to collaborate with numerous experimental colleagues to model the physics behind their observations in the areas of biofuels, plastic pyrolysis and 2D material synthesis. My group will leverage my expertise in multi-scale simulations to do the same in areas of electrified manufacturing, semiconductors and sustainable fuels. These modelling advances would accelerate our progress towards finding sustainable methods to replace the current methods.

Teaching interests

Along with research, I am also excited about teaching opportunities. My philosophy is to use a combination of well-designed course material, frequent feedback and active student participation to create an enjoyable learning experience in my class. My first exposure to teaching was being a teaching assistant for a programming lab while pursuing my master’s degree. While this was an incredible learning opportunity, I observed during the course that a section of students consistently struggled with the first few assignments since they were not up to speed with the programming experience that the course instructor expected from them. To avoid a similar situation, I would like my classes to be accessible to people with different backgrounds and interest levels. Hence, when I designed and taught 7 weeks of a graduate course on ‘Practical Molecular Simulations’, I created a non-graded assignment to gauge the student readiness level. This assignment greatly helped me adjust my lecture plan during those 7 weeks. To give an example of active student participation, my lectures devoted half of the time to discussing concepts and rest of the half to students doing small examples based on these concepts using the university computer resources. This structure was based on the feedback that I received from a 90-minute guest lecture that I delivered to a different graduate class during my first semester as a postdoc. Based on the obtained feedback, it was evident that the students enjoyed the interactive nature of the class. In this course, I focused on parts of molecular dynamics, statistical mechanics, and thermodynamics. I intend to expand these lectures into a full graduate course by including basics of quantum chemistry as well as machine learning potentials. With my background, I will also be interested in teaching undergraduate or graduate level thermodynamics, numerical/computational methods, heat and/or mass transfer and reaction engineering.

I believe that a combination of my past research experience in multi-scale simulation techniques (CFD, RMD, DFT, ML) will help me in guiding new researchers. Collectively, we can make the proposed work a reality and aid in the development of sustainable chemical processes. I am interested in joining an institution and department with a strong research background, commitment to teaching along with an interest in fostering research that helps us achieve a sustainable future.