Active Site Representation in First-Principles Microkinetic Models: Data-Enhanced Computational Screening for Improved Methanation Catalysts

Abstract

Computational screening based on first-principles microkinetic modeling has evolved into a widespread tool for catalyst discovery. Efficiently exploiting various scaling relations, this approach draws its predictive character from reliable adsorption energies, typically calculated with density-functional theory (DFT). In prevalent screening approaches, the concomitant computational costs are kept tractable through the use of reductionist microkinetic models that only resolve a minimalistic amount of active site motifs at the catalyst surface. Here, we scrutinize this common practice by systematically comparing the screening predictions for the CO methanation reaction when using microkinetic models that resolve an increasing amount of sites, up to the full consideration of all high-symmetry sites at stepped transition metal (TM) and binary TM alloy catalysts. Apart from generally overestimating the catalytic activity, the simplified models fail to identify a most promising class of layered bimetallic alloys as their insufficient representation of the catalyst surface does not allow them to correctly capture the rate-determining step. Only the full microkinetic model provides this proper mechanistic basis for the screening. The excessive amount of predictive-quality adsorption energetics required for this model is obtained from a compressed sensing descriptor that once trained readily provides these data for a new material from a single DFT calculation of the clean surface. With the current methodological advances in areas such as compressed sensing and machine learning, and the concurrent availability of cheap adsorption energetics for a wide range of possible catalyst materials, there is thus no reason to continue to use simplistic microkinetic models in computational catalyst screening.

Identifying Outstanding Transition-Metal-Alloy Heterogeneous Catalysts for the Oxygen Reduction and Evolution Reactions via Subgroup Discovery

“…In this paper, we applied the SGD approach to identify the most relevant atomic, bulk and surface properties—as well as rules associated to those parameters—describing outstanding SGs of transition-metal surface sites. In particular, we demonstrated this approach using a data set of DFT-calculated adsorption energies [ 8 , 24 ] by searching for surface sites (i) that present optimal range of oxygen binding strength for the ORR or (ii) that deviate the most from the linear-scaling relations between O and OH adsorption energies that impose a limit to the OER performance. The SGs rules not only hint at the relevant underlying physicochemical processes that govern the local statistically exceptional behavior, but are also suitable for guiding the design of challenging bimetallic alloys.…”

“…We analyze a data set containing 95 oxygen (atomic O) adsorption energies, which were calculated with DFT using the van der Waals-corrected BEEF-vdW exchange–correlation functional in previous publications. [ 8 , 24 ] Eleven transition metals and several adsorption sites of different surfaces for which (meta)stable oxygen adsorption is observed were included in our analysis (Fig. 1 B).…”

Autonomous Reaction Network Exploration in Homogeneous and Heterogeneous Catalysis

“…Highthroughput 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][112][113], physical descriptors [114][115][116][117], or machine learning (ML) models trained on physical or structural descriptors [118][119][120][121][122][123][124][125][126][127][128][129][130][131][132][133][134]. Furthermore, they can be enhanced by stability analysis to save computing time on unstable materials [135].…”

Toward Better and Smarter Batteries by Combining AI with Multisensory and Self‐Healing Approaches

“…[42][43][44] Recent advances using ML-based approaches like sure independence screening and sparsifying operator has led to the development of more reliable microkinetic models that can capture a more comprehensive mechanistic understanding. [45] Important developments are also made for ML potentials capable of describing the spatio-temporal evolution of reactive liquid-solid/amorphous systems with an unprecedented number of atoms and timespan of simulations. Recent work by Deringer et al, using GAP molecular-dynamics simulations on 100 000 silicon atoms were capable of accessing the time scales needed for prediction of distinct electronic features that can be compared directly to ultrafast spectroscopic techniques, and the experimentally relevant length scales for description of (poly-)crystallization in amorphous silicon.…”