To address surface reaction network complexity using scaling relations machine learning and DFT calculations

Abstract

Surface reaction networks involving hydrocarbons exhibit enormous complexity with thousands of species and reactions for all but the very simplest of chemistries. We present a framework for optimization under uncertainty for heterogeneous catalysis reaction networks using surrogate models that are trained on the fly. The surrogate model is constructed by teaching a Gaussian process adsorption energies based on group additivity fingerprints, combined with transition-state scaling relations and a simple classifier for determining the rate-limiting step. The surrogate model is iteratively used to predict the most important reaction step to be calculated explicitly with computationally demanding electronic structure theory. Applying these methods to the reaction of syngas on rhodium(111), we identify the most likely reaction mechanism. Propagating uncertainty throughout this process yields the likelihood that the final mechanism is complete given measurements on only a subset of the entire network and uncertainty in the underlying density functional theory calculations.

Machine Learning for Computational Heterogeneous Catalysis

“…Ulissi et al . used Gaussian processes to predict adsorption energies, transition state scaling and rate‐limiting steps for the complex reaction network of the syngas reaction over Rh(111) . In a different study, Ulissi et al .…”

The Rise of Catalyst Informatics: Towards Catalyst Genomics

“…In the same fashion, high‐throughput calculations are another powerful approach for collecting catalyst data where the activation energy barrier can be evaluated and can thereby be used as data when conducting large‐scale catalyst screening ,. The reaction pathway and mechanisms have been a great mystery within catalysis due to the challenges faced when attempting to detect intermediate compounds with in experiments, yet first principles calculations paired with kinetic modeling allow for the ability to map and identify intermediate states during the reaction ,. Additionally, effective algorithms have been developed for creating the reaction map such as OPTIM and global reaction route mapping (GRRM) .…”

“…Given this, computational science has a major advantage for creating data within a short period of time. It has been demonstrated that machine learning can dramatically reduce the computational time for deriving parameters of catalytic relevance such as formation energy, transition state energy, and adsorption energy ,,. On the other hand, 30 years of oxidative coupling of methane experimental catalyst data from literature is collected and shown in Figure (a) and (b) .…”

Data Mining the C−C Cross‐Coupling Genome

“…Using machine learning (ML) to accelerate electronic structure methods has the potential to revolutionize the field of computational chemistry, having already demonstrated its promise by making significant headway in addressing several longstanding fundamental problems in chemistry and materials science over the past 15 years . In the field of computational catalysis, for example, recent work has tackled a diverse set of topics ranging from CO 2 uptake in metal organic frameworks to identifying structure‐activity relationships in Pt electrocatalysts for the oxygen reduction reaction to reaction mechanism discovery on rhodium surfaces . Such examples of ML only begin to scratch the surface .…”