Automated Mechanism Generation Using Linear Scaling Relationships and Sensitivity Analyses Applied to Catalytic Partial Oxidation of Methane

Abstract

Kinetic parameters for surface reactions can be predicted using a combination of DFT calculations, scaling relations, and machine learning algorithms; however, construction of microkinetic models still requires a knowledge of all the possible, or at least reasonable, reaction pathways. The recently developed Reaction Mechanism Generator (RMG) for heterogeneous catalysis, now included in RMG version 3.0, is built upon well-established, open-source software that can provide detailed reaction mechanisms from user-supplied initial conditions without making a priori assumptions. RMG is now able to estimate adsorbate thermochemistry and construct detailed microkinetic models on a range of hypothetical metal surfaces using linear scaling relationships. These relationships are a simple, computationally efficient way to estimate adsorption energies by scaling the energy of a calculated surface species on one metal to any other metal. By conducting simulations with sensitivity analyses, users can not only determine the rate limiting step on each surface by plotting a "volcano surface" for the degree of rate control of each reaction as a function of elemental binding energies, but also screen novel catalysts for desirable properties. We investigated the catalytic partial oxidation of methane to demonstrate the utility of this new tool and determined that an inlet gas C/O ratio of 0.8 on a catalyst with carbon and oxygen binding energies of -6.75 eV and -5.0 eV, respectively, yields the highest amount of synthesis gas. Sensitivity analyses show that while the dissociative adsorption of O 2 has the highest degree of rate control, the interactions between individual reactions and reactor conditions are complex, which result in a dynamic rate-limiting step across differing metals. File list (2) download file view on ChemRxiv LSRs in RMG-Cat preprint.pdf (1.72 MiB) download file view on ChemRxiv supplementaryinfo.pdf (374.47 KiB)

Quantifying the Impact of Parametric Uncertainty on Automatic Mechanism Generation for CO2 Hydrogenation on Ni(111)

“…Thermodynamic properties of the adsorbates on the Ni facet were estimated via LS relations 35 (see eq 1 ), which were recently implemented in RMG by Mazeau et al, 45 based on reference values for Pt(111) obtained via BEEF–vdW calculations in ref ( 46 ). The binding energy of an adsorbate is estimated via where ΔE Pt AX is the binding energy of the adsorbate AX* in the Pt(111) database, where X represents any adsorbate that binds through A*, ΔE Ni A is the binding energy of the adatom A*, ΔE Pt A is the analogous property for Pt(111), and the slope γ is related to the degree of saturation for the adsorbate.…”

Autonomous Reaction Network Exploration in Homogeneous and Heterogeneous Catalysis

“…High-throughput studies can be accelerated by exploiting also surrogate models, i.e., efficient, empirical models that can produce property predictions such as adsorption energies, albeit less accurately than a first-principles-based model such as density functional theory (DFT) [ 110 ]. Surrogate models can be scaling relationships [ 111 – 113 ], physical descriptors [ 114 – 117 ], or machine learning (ML) models trained on physical or structural descriptors [ 118 – 134 ]. Furthermore, they can be enhanced by stability analysis to save computing time on unstable materials [ 135 ].…”

Autonomous Reaction Network Exploration in Homogeneous and Heterogeneous Catalysis

“…The latest release of the graph-based reaction mechanism generator (RMG) by Green and co-workers [152 ] features additional graph rules for surfaces, in which the surface is treated as a single graph node with which every other node can form bonds with. The authors applied this approach to methane dry reforming on Ni (111) [171 ], for which their algorithm found many of the reactions of an established mechanism [172 ]. However, their approach was limited to predefined reaction types, the adsorption energies were based on literature values or group additivity for missing literature data, and the reaction energy barriers were derived from scaling relationships from the literature.…”

“…It implements an interface to the atomistic simulation environment (ASE) [174 ] to find adsorption sites and explore reaction paths of adsorbates with their growing string method (GSM) [162 , 175 ]. Maeda et al have also explored reactions of adsorbates on (111) surfaces [176 -178 ] with their artificial force induced reaction (AFIR) approach [179 ]. While both approaches, GSM and AFIR, are versatile and general, the application studies were limited to low-index surfaces with a completely constrained slab.…”

“…In a different approach, Liu and co-workers [194 ] sampled a reaction on a Cu (111) surface, namely the water gas shift reaction. They constrained two of three layers and found the reaction with their enhanced sampling technique called stochastic surface walking (SSW).…”