(2bx) Automatic Reaction Mechanism Generation for Complex Systems Using Machine Learning and Computation

Automatic Reaction Mechanism Generation, Property Prediction, Quantum Chemistry, Reaction Mechanism Simulation and Analysis, Catalysis, Electrochemistry

Most chemical processes involve many elementary chemical reactions. It is often possible to fit locally valid kinetic expressions for overall reactions to experimental data in a postdictive sense. However, true predictive kinetic modeling requires a comprehensive list of elementary reactions and their associated kinetic parameters often called the chemical mechanism. Quantum chemistry and machine learning techniques can be combined to provide reliable estimates for kinetic parameters, although this can be challenging in in heterogeneous systems and condensed phases. Automatic mechanism generation software, such as the Reaction Mechanism Generator (RMG) enable efficient enumeration of mechanisms that can include hundreds to thousands of species. While this overall RMG framework has been applied very successfully in gas phase neutral chemistry, much less progress has been made in more complex systems such as in liquid phase, on metallic surfaces, with electrochemical systems, and systems involving large molecules.

Postdoctoral Project: “Creation of an Exascale Compatible Automated Workflow for Calculation of Surface and Gas-Surface Kinetics”

Under supervision of Judit Zador, Combustion Research Facility, Sandia National Laboratories

PhD Dissertation: “Automatic Generation and Analysis of Chemical Kinetic Mechanisms

Under supervision of William H. Green, Department of Chemical Engineering, Massachusetts Institute of Technology

Research Experience:

Within my academic career I have worked on virtually every aspect of computational kinetics: workflows for automatically calculating kinetic parameters using quantum chemistry; developing machine learning algorithms for estimating kinetic properties; developing improved algorithms for mechanism generation; and generation, analysis, and refinement of mechanisms. I developed mechanisms involving many different systems involving gas, liquid, and surface phases, multiple phases, and electrochemical reactions. I have worked on both independent and highly collaborative projects. During my PhD I was lead developer for the RMG organization and have extensive experience maintaining and developing software. Additionally, I have significant experience working on high performance computing resources.

To go over a handful of highlights from my work: I developed a new low-data, scalable, and human-readable machine learning algorithm with detailed uncertainty estimation for rate coefficient prediction that significantly improved accuracy compared to traditional methods; I developed a workflow for automatic calculation of rate coefficients for surface and gas-surface reactions using quantum chemistry; I developed a mechanism for methyl propyl ether oxidation and pyrolysis that I validated against experimental data from three different groups; I developed a more reliable and more accurate algorithm for computing phenomenological rate coefficients for pressure dependent reactions; I developed ReactionMechanismSimulator.jl a differentiable software for efficiently simulating and analyzing large kinetic mechanisms that I demonstrated to be appreciably faster than competing software; and lastly, I developed an algorithm for automatic analysis of chemical mechanisms that is able to automatically and efficiently identify important reactions, determine their associated pathways and determine why they are important in minutes even for very large mechanisms where traditional sensitivity calculations would take 18,000x longer.

Teaching Interests/Experience:

I am excited to teach chemical engineering courses, especially those focused on theory and computation. During my PhD I thoroughly enjoyed being a teaching assistant for the Numerical Methods in Chemical Engineering (Graduate) course at MIT. My duties included helping write and test homework assignments, giving recitations each week, tutoring students that were behind, and helping write and test exams. Over the course of my PhD I also mentored several graduate students within the Green group, helping teach them important concepts and techniques and providing general advice and support with their projects and PhD progression. Additionally, as lead developer for the RMG organization I gained extensive experience running workshops and giving lectures educating graduate students and postdocs on how to use software.

Future Direction:

As a faculty member I intend to employ the techniques I developed during my postdoc and PhD to develop automatic mechanism generation and quantum chemical techniques for complex systems such as those involving liquid and surface phases and those involving large molecules, molecular weight growth, and ions.

Reliably generating accurate mechanisms that involve more than a thousand reactions using current techniques can be challenging. To alleviate this and enable application of the technique to larger systems I will fuse machine learning and model analysis techniques I devised during my PhD to develop mechanism generation techniques that can be combined with current numerical approaches to generate large chemical kinetic mechanisms efficiently and reliably.

In particular, I would like to extend these techniques to electrochemical systems. I have done some proof-of-concept work on modeling the formation of the solid electrolyte interface (SEI) in batteries as a part of my PhD and postdoc. Quantitatively accurate modeling of SEI formation and growth is a highly relevant long-term target; however, it is uniquely challenging combining a liquid phase, an evolving surface phase, molecular weight growth, potentially large molecules and ions. I will use simpler relevant systems as stepping stones to develop the necessary technology. Important challenges in surface chemistry can be resolved with work in gas-phase heterogeneous catalysis. I will address challenges related to liquid phase solvation corrections and ions studying organic acid-base systems or homogeneous catalysis. Liquid phase heterogeneous catalysis and electrocatalysis provide industrially relevant stepping stones for studying liquid-surface interactions and electrochemical reactions. I will develop collaborations with experimentalists to validate the accuracy of developed mechanisms.

Selected Publications:

Johnson, M., Dong, X., Grinberg Dana, A., Chung, Y., Farina Jr, D., Gillis, R., Liu, M., Yee, N., Blondal, K., Mazeau, E., Grambow, C., Payne, A., Spiekermann, K., Pang, H.-W., Goldsmith, C. F., West, R., & Green, W. (n.d.). RMG Database for Chemical Property Prediction. Journal of Chemical Information and Modeling, 62(20), 4906–4915. https://doi.org/10.1021/acs.jcim.2c00965

Johnson, M. S., Nimlos, M. R., Ninnemann, E., Laich, A., Fioroni, G. M., Kang, D., Bu, L., Ranasinghe, D., Khanniche, S., Goldsborough, S. S., Vasu, S. S., & Green, W. H. (2021). Oxidation and pyrolysis of methyl propyl ether. International Journal of Chemical Kinetics, kin.21489. https://doi.org/10.1002/kin.21489

Johnson, M. S., & Green, W. H. (2022). A Machine Learning Based Approach to Reaction Rate Estimation. ChemRxiv. https://doi.org/10.26434/CHEMRXIV-2022-C98GC

Johnson, M. S., Gierada, M., Hermes, E. D., Bross, D. H., Sargsyan, K., Najm, H. N., & Zador, J. (2023). Pynta - An automated workflow for calculation of surface and gas-surface kinetics. https://doi.org/10.26434/CHEMRXIV-2023-2XPFQ

Johnson, M. S., Pang, H.-W., Payne, A. M., & Green, W. H. (2023). ReactionMechanismSimulator.jl: A Modern Approach to Chemical Kinetic Mechanism Simulation and Analysis. https://doi.org/10.26434/CHEMRXIV-2023-TJ34T

Johnson, M. S., Pang, H.-W., Liu, M., & Green, W. H. (2023). Species Selection for Automatic Chemical Kinetic Mechanism Generation. https://doi.org/10.26434/CHEMRXIV-2023-WWRQF

Johnson, M. S., McGill, C. J., & Green, W. H. (2022). Transitory Sensitivity in Automatic Chemical Kinetic Mechanism Analysis. https://doi.org/10.26434/CHEMRXIV-2022-ZSFJC

Johnson, M. S., & Green, W. H. (2022). Examining the Accuracy of Methods for Obtaining Pressure Dependent Rate Coefficients. Faraday Discussions. https://doi.org/10.1039/d2fd00040g

Johnson, M. S., Grinberg Dana, A., & Green, W. H. (2022). A workflow for automatic generation and efficient refinement of individual pressure-dependent networks. Combustion and Flame, 112516. https://doi.org/10.1016/j.combustflame.2022.112516

Liu, M., Grinberg Dana, A., Johnson, M. S., Goldman, M. J., Jocher, A., Payne, A. M., Grambow, C. A., Han, K., Yee, N. W., Mazeau, E. J., Blondal, K., West, R. H., Goldsmith, C. F., & Green, W. H. (2021). Reaction Mechanism Generator v3.0: Advances in Automatic Mechanism Generation. Journal of Chemical Information and Modeling, 61(6), 2686–2696. https://doi.org/10.1021/ACS.JCIM.0C01480/ASSET/IMAGES/LARGE/CI0C01480_00...

Dana, A. G., Johnson, M. S., Allen, J. W., Sharma, S., Raman, S., Liu, M., Gao, C. W., Grambow, C. A., Goldman, M. J., Ranasinghe, D. S., Gillis, R. J., Payne, A. M., Li, Y., Dong, X., Spiekermann, K. A., Wu, H., Dames, E. E., Buras, Z. J., Vandewiele, N. M., ... Green, W. H. (2023). Automated reaction kinetics and network exploration (Arkane): A statistical mechanics, thermodynamics, transition state theory, and master equation software. International Journal of Chemical Kinetics, 55(6), 300–323. https://doi.org/10.1002/kin.21637

Kavalsky, L., Hegde, V. I., Muckley, E. S., Johnson, M. S., Meredig, B., & Viswanathan, V. (2023). By how much can closed-loop frameworks accelerate computational materials discovery? Digital Discovery. https://doi.org/10.1039/D2DD00133K

Annevelink, E., Kurchin, R., Muckley, E., Kavalsky, L., Hegde, V. I., Sulzer, V., Zhu, S., Pu, J., Farina, D., Johnson, M., Gandhi, D., Dave, A., Lin, H., Edelman, A., Ramsundar, B., Saal, J., Rackauckas, C., Shah, V., Meredig, B., & Viswanathan, V. (2020). AutoMat: Accelerated Computational Electrochemical systems Discovery. https://doi.org/10.48550/arxiv.2011.04426

Keçeli, M., Elliott, S. N., Li, Y.-P., Johnson, M. S., Cavallotti, C., Georgievskii, Y., Green, W. H., Pelucchi, M., Wozniak, J. M., Jasper, A. W., & Klippenstein, S. J. (2019). Automated computational thermochemistry for butane oxidation: A prelude to predictive automated combustion kinetics. Proceedings of the Combustion Institute, 37(1), 363–371. https://doi.org/10.1016/J.PROCI.2018.07.113

Fridlyand, A., Johnson, M. S., Goldsborough, S. S., West, R. H., McNenly, M. J., Mehl, M., & Pitz, W. J. (2017). The role of correlations in uncertainty quantification of transportation relevant fuel models. Combustion and Flame, 180. https://doi.org/10.1016/j.combustflame.2016.10.014