A predictive model of surface adsorption in dissolution on transition metals and alloys

Abstract

Surface adsorption is often coupled with surface dissolution and is generally unpredictable on alloys due to the complicated alloying and dissolution effects. Herein, we introduce the electronic gradient and cohesive...

A First-Principles Approach to Modeling Surface Site Stabilities on Multimetallic Catalysts

“…In models based on intrinsic properties , inspired by the NA model and the seminal work of Hammer and Nørskov, ,, the adsorption strength is correlated to the spatial extent of metal d-orbitals with respect to adsorption distances. Leaning on this theoretical foundation, Gao and Li et al , linked electronic descriptors with valence electrons and electronegativity when introducing their semiempirical models based on geometric and electronic gradient descriptors. The models of Gao and Li et al , offer both fast and inexpensive screening while proficiently accounting for compositional variations larger than those of the current incarnation of the α-parameter scheme introduced herein.…”

“…Leaning on this theoretical foundation, Gao and Li et al , linked electronic descriptors with valence electrons and electronegativity when introducing their semiempirical models based on geometric and electronic gradient descriptors. The models of Gao and Li et al , offer both fast and inexpensive screening while proficiently accounting for compositional variations larger than those of the current incarnation of the α-parameter scheme introduced herein. Our approach fills an important role, however, as it is fully grounded in physical arguments and seeks to derive the binding energy from its exact expression.…”

“…Similarly, models based on machine-learning also require large data sets to accurately train and design a framework for MMAs; in addition, they typically lack physical interpretation. Deriving modified methods from the Newns-Anderson (NA) and d-band models have shown particularly promising results, as exemplified by the semiempirical approach of Gao and Li et al, 64,65 66 which is that existing theoretical approaches like the d-band, NA, and coordination number (CN)-based models are inadequate when describing interactions on complex metal surfaces. This was attributed to the inability to account for effects due to perturbations in the electronic structure introduced by alloying and effects explicitly introduced by the adsorbate.…”

“…Similarly, models based on machine-learning also require large data sets to accurately train and design a framework for MMAs; in addition, they typically lack physical interpretation. Deriving modified methods from the Newns-Anderson (NA) and d-band models have shown particularly promising results, as exemplified by the semiempirical approach of Gao and Li et al, , which integrates the d-band model, the muffin-tin-orbital theory, and a coordination-based model. The model has made significant strides in capturing variations up to the first-nearest neighbor, providing valuable insights into surface site stability and displaying impressive predictions for metallic materials across the periodic table.…”

Latent Variable Machine Learning Framework for Catalysis: General Models, Transfer Learning, and Interpretability

“…The lack of accuracy of linear scaling relations on the specific data sets we study here has been noted previously. ,, Furthermore, many ML-based studies have predicted energetics on alloys using electronic structure properties; ,− however, these studies most often either require density functional theory (DFT)-calculated features, which makes the screening several orders of magnitude slower, or only consider a relatively small range of adsorbates and/or metals, limiting their generality. Additionally, fitting a model to multiple species has been performed in previous work ,, but true transfer learning, in which a pretrained model is used to accelerate training for a new application, has rarely been used across adsorbates, across computational setups, or to experimental data.…”