A deep understanding of how materials perform in energy applications is essential for developing better technologies, from catalysts to batteries. However, predicting and analyzing the behavior of these materials, particularly their reaction kinetics, is challenging due to complex chemical reactions, material degradation, and evolving performance over time. Traditional computational methods, such as density functional theory (DFT), kinetic Monte Carlo (kMC), and machine learning (ML), are commonly used to study these processes but each has limitations when used alone. In this work, we present a combined computational approach using DFT and kMC specifically aimed at effectively analyzing reaction kinetics in energy applications. DFT provides fundamental kinetic parameters, including reaction energies and activation barriers, and offers insights into electronic properties and atomic-level interactions. However, DFT alone cannot fully capture kinetic behavior over extended time scales or account for environmental conditions realistically. To overcome this limitation, we integrate DFT-calculated kinetic parameters into kMC simulations
[1]-[2]. The kMC model allows us to simulate the time-dependent evolution of reaction processes by tracking events such as changes in surface coverage, adsorbate interactions, and turnover frequencies under various conditions, including temperature and pressure. This combined framework provides a more complete picture of the reaction kinetics, bridging the gap between atomic-level calculations and realistic operating environments. Even with this progress, one critical challenge remains: modeling the long-term degradation of materials and the resulting shift in performance still demands enormous computational resources. For example, as a material structure slowly changes over time, its reaction kinetics can also evolve, especially under varying external conditions like temperature and pressure. Simulating these gradual, long-timescale processes using only DFT and kMC can be prohibitively slow. To address this, we introduce a long short-term memory (LSTM) neural network into our computational framework. By coupling kMC-generated time-series data with LSTM, we can efficiently predict how materials degrade and how their kinetic properties evolve over time without having to simulate each step in full detail. This LSTM-accelerated kMC approach captures complex time-dependent behaviors such as surface degradation and shifting activity levels, enabling long-term predictions with significantly reduced computational costs. This integrated DFT-kMC-LSTM
[3] framework not only provides a more complete picture of reaction kinetics but also offers a practical path toward identifying stable and high-performing materials for a wide range of energy applications.
References
[1] Lee, C. H.; Pahari, S.; Sitapure, N.; Barteau, M. A.; Kwon, J. S. I., DFT-kMC Analysis for Identifying Novel Bimetallic Electrocatalysts for Enhanced NRR Performance by Suppressing HER at Ambient Conditions Via Active-Site Separation. ACS Catal 2022, 12(24), 15609-15617.
[2] Lee, C. H.; Pahari, S.; Barteau, M. A.; Kwon, J. S. I., Exploring dynamics in single atom catalyst research: A comprehensive DFT-kMC study of nitrogen reduction reaction with focus on TM aggregation. Appl. Catal., B, 2024, 358, 124434.
[3] Lee, C. H.; Pahari, S.; Sitapure, N.; Barteau, M. A.; Kwon, J. S. I., Investigating high-performance non-precious transition metal oxide catalysts for nitrogen reduction reaction: a multifaceted DFT–kMC–LSTM approach. ACS Catal 2023, 13(13), 8336-8346.
