CDS&E: Elucidating the Structure and Catalytic Activity of Nanoparticles Under Catalytic Conditions Using Ab Initio Machine Learning Force Fields

University Of Alabama Tuscaloosa, Tuscaloosa AL

Investigators

Abstract

The manufacture of commercial and consumer chemical products relies heavily on catalytic reaction processes that consume a significant fraction of our nation's energy resources and are a major contributor to the emission of green-house gases. These catalysts often are composed of a metallic catalyst nanoparticle attached to a support of different material and can take on a nearly infinite number of configurations, shapes, compositions, and process operating conditions. To probe this vast parameter space for the optimal catalyst system purely by experimentation would be impossible and so efficient simulation tools are needed to explore catalyst behavior at the atomistic level. This proposal seeks to develop these simulation tools with emphasis on understanding how catalyst nanoparticle shapes change during realistic reactor operating conditions; this will be made possible by the proposal’s plan to improve the computational efficiency of molecular dynamics simulations using advanced machine learning methods. The computational models will be rigorously compared against known experimental benchmarks to guide simulator development and improve its prediction accuracy. The catalysts discovered using the simulation tools developed in this research program will contribute to the decarbonization of the chemical processes and will play an important role in developing circular chemical economies. The proposed research also will create opportunities to educate the next generation of researchers and industry leaders. Undergraduate students will learn programming skills that will increase their competitiveness in emerging job fields, and their immersive research experiences will prepare them for positions at top graduate schools and careers in higher education. Macroscopic renderings of catalysts designed with this software will be generated by 3D printing to demonstrate to the public the role computations play in accelerating catalyst design. This proposal seeks to develop and apply ab initio machine learning force fields (AIMLFF) to simulate nanoparticle (NP) catalysts under realistic reaction conditions and to help elucidate the nature of catalytic active sites. This proposed research will specifically address these challenges by hypothesizing that when AIMLFFs are trained on common structural features of periodic density functional theory (DFT) calculations that the community has identified as meaningful representations of NP catalysts, AIMLFF will be able to model NP catalysts directly under reaction-relevant conditions. This work will address critical questions related to the accuracy of the AIMLFFs by making comparisons to benchmark-quality microscopic and calorimetric measurements available in the literature, and will develop a general understanding of how the shape of metal NPs and available binding sites depend on the species of metal and the support under temperature and pressure conditions representative of reaction conditions. The proposed research will also develop an improved understanding of the relationship between catalytic activity and the evolution of NPs in comparison with high-quality X-ray measurements. Fundamental knowledge will be gained on how the equilibrium shape and defect density of supported metal NPs change over realistic reaction times for systems that are too large for current simulators. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.