Machine Learning Models of Electrocatalysts for Ammonia Production
Abstract
Ammonia is an essential feedstock for producing fertilizers, pharmaceuticals, and other chemicals in both industry and agriculture. As such, it is imperative to build cost-effective strategies to replace the traditional nitrogen fixation process. The electrochemical nitrogen reduction reaction driven by economic catalysts is a promising strategy due to its efficiency, cleanliness, and sustainability. However, the conventional approach to discovering new catalysts relies on trial-and-error methods. Additionally, there is currently a lack of demonstrated examples for the computational design of electrochemical nitrogen reduction reaction devices to achieve practical applications. Therefore, the primary purpose of this project is to develop computational strategies for discovering novel, economic catalysts for ammonia production that exhibit high catalytic reactivity and selectivity comparable to that of nitrogenases.
This research project is supported by the FORMAS early-career grant, and initiated in January 2024 at Uppsala University. To achieve this aim, I will develop streamlined computational methods and machine learning models to aid in designing cost-effective catalysts and devices for electrochemical nitrogen reduction reactions. These predicted promising catalysts will be validated through interdisciplinary partnerships and utilized to engineer devices for electrochemical nitrogen reduction reactions. By achieving this goal, I will promote a sustainable path to agricultural feedstock production.
In undertaking the research project, we developed a computational workflow and toolkit for simulating reaction pathways of e-NRR catalyzed by metal-ligand complex directly from SMILES inputs. This approach integrates force field minimization, semi-empirical quantum mechanics, and DFT functionals. Using this workflow, we quantitatively calculated the geometric and electronic properties of these metal-ligand complexes using the Orca and Gaussian software on Tetralith.
In the following project, we will use these generated datasets of catalyst properties for the development of machine learning protocols. Additionally, we will develop efficient computational approaches for simulating reaction pathways of electrochemical nitrogen reduction reactions in a realistic environment. This involves comparing DFT functionals to calculate the key steps of a validated catalyst obtained from machine learning models. We will also evaluate quantum mechanics/molecular mechanics molecular dynamics and empirical valence bond methods for describing bond breaking of the catalysts in solution. Successful execution of this project section will necessitate HPC for computational method examination and simulation of the full reaction process in solution.
Since I started my researcher position at Uppsala University in January 2024, I have attracted one visiting PhD student, one visiting postdoc, four Master students, and four joint Master students. Thanks to the NAISS 2024/5-22 project, I am able to support my team in performing our research using high-performance computing resources. The continuation of HPC resources is of significance for the successful execution of this research project and leading a research team.
