(334as) Computational Approaches Fueling Clean Materials Chemistry - from First Principles Calculations to Data Driven Discovery and Materials Design

In the search for cheaper and energetically favorable materials for sustainable applications, there is a desire to catalyze reactions at room temperature and pressure with high active surface area for a number of chemistries. Designing new catalysts in turn improve photovoltaics, chemicals, enzymes, polymers, oil and gas, and automobile industries. MXenes, a new class of 2D materials similar to graphene, are tunable through physical and chemical modifications to their surfaces and have been considered as catalysts for chemical reactions for a decade now.^1,2 For the past decade, learning methods have been used to predict and infer chemical and physical properties through analyzing datasets and building atomistic potentials.³

In this poster, I will discuss approaches to performing high throughput computational screening on 2D materials to generate catalytic data based on these chemical and physical alterations to their structure. Computational based discovery requires the use of combined efforts of generating data and applying advanced techniques to understand and make data useful. Previous studies demonstrate our ability to compute data using density functional theory on MXenes for a number of chemical reactions crucial to reaching a clean energy economy.^4–6 With this newly generated data, I perform data analyses of both computed and tabulated features through (a) examining correlations in the features and relationships to our target properties and descriptors of chemical reactions, based on changes to the material’s stoichiometry and geometry; and (b) using data science pipelines to generate simple models for structure-property relationships and infer features that point to how our chemistries change due to materials tuning. The methods involved training machine learning models ranging from linear regression and lasso method to neural networks and decision trees through hyperparameter tuning, validation tests, and feature selection through singular value decompositions.

This work provides not only invaluable data to the 2D materials and catalysis communities but crucial analyses and fundamental understanding into how we can design materials and substances with high chemical activity through altering their electronic structure. The study also extends as a guideline to the scientific community on how to treat structured data in an appropriate manner to extract crucial insights needed to design new systems and improve upon existing ones. As a fifth year PhD candidate, the goal is to share how analytical, technical, and professional experience as a researcher, healthcare consultant, masters and undergraduate student mentor, and involved graduate student form the perfect candidate for business development and data related roles in industry.

Research Interests: materials design, catalysis, data science, electrochemistry, density functional theory

Teaching Interests: active learning, flipped classrooms, electronic note-taking, physical chemistry, reactor design, kinetics

References

1. Gogotsi, Y. & Barsoum, M. W. Two-Dimensional Transition Metal Carbides. ACS Nano 6, 1322–1331 (2012).
2. Seh, Z. W. et al. Two-Dimensional Molybdenum Carbide (MXene) as an Efficient Electrocatalyst for Hydrogen Evolution. ACS Energy Lett. 1, 589–594 (2016).
3. Kitchin, J. R. Machine learning in catalysis. Nat. Catal. 1, 230–232 (2018).
4. Handoko, A. D. et al. Tuning the Basal Plane Functionalization of Two-Dimensional Metal Carbides (MXenes) To Control Hydrogen Evolution Activity. ACS Appl. Energy Mater. 1, 173–180 (2018).
5. Johnson, L. R. et al. MXene Materials for the Electrochemical Nitrogen Reduction-Functionalized or Not? ACS Catal. 10, 253–264 (2020).
6. Jin, D. et al. Computational Screening of 2D Ordered Double Transition-Metal Carbides (MXenes) as Electrocatalysts for Hydrogen Evolution Reaction. J. Phys. Chem. C 124, 10584–10592 (2020).