(191j) Progress in the Understanding of Chemical Interactions through Atomic Scale Modeling – Applications in Catalyst Design

In this presentation, I will exemplify recent progresses in the theoretical description of chemical interactions based on my current research activities. This will be discussed in the perspective of heterogeneous catalysis and surface chemistry, with focus on the fundamental challenges in modeling and how these can be addressed. In particular, I will present methods that allow for the modeling of realistic material surfaces and reaction conditions for catalytic systems under operation than. These methods pave the way toward in-silico guidance in materials design.

Central to many materials properties is the ability to form chemical bonds on surfaces. This includes catalytic activity, but also corrosion resistance, nanotoxicity, and biocompatibility. Herein, different examples will be given on physically motivated and transparent approaches for efficient estimations of local surface interactions taken from my recent research. To highlight the impact of these methods, I will present results from applications of broad general interest including the design of bi- and multi-metallic catalytic active sites for nitrogen oxide decomposition; the tailoring of copper nanostructures for selectivity in electrocatalytic carbon dioxide reduction; and the rationalization of the activity of metal and metaloxide nanoclusters in methane oxidation.

Three methods for modeling surface interaction will be discussed and compared:

• First, I will describe a promising strategy for predicting the interaction of an adsorbate and a surface of arbitrary complexity by matching local properties determined for the separated compounds.^[1-5] In this approach, we evaluate spatially-resolved properties that each describe different modes of interaction such as charge-transfer, electrostatics, and dispersion. Through the proper weighting of these properties, the interaction energy of a given adsorbate-surface pair can be accurately estimated. This can be efficiently carried out for structurally diverse surfaces spanning metal nanoparticles, alloys, and metaloxides. Applications range from understanding aromatic substitution reactions to predictions of active bimetallic single-site PdMO_x catalysts for methane conversion into value-added compounds.
  
  
• Second, a coordination-based approach will be discussed that is able to capture variations in the local adsorption energies onto multi-metal surface motifs as a function of local structure and composition. The method is referred to as the alpha-parameter scheme, where pair-wise parameters are trained to represent the average atom-wise bond-formation energy between two metal atoms leading to a given coordination number. Because the coordination environment is readily computed even for large structures, systems up to micrometer scale can be treated. Catalyst design in NOx decomposition over Pt-based alloys as well as in the tailoring of nanostructured copper for electrochemical upgrading of CO₂ are examples of applications.^[6,7]
  
  
• Third, a careful derivation of surface interactions on metal alloys from the Newns-Anderson model will be introduced.^[8] It allows for the accurate description of chemisorption energies by accounting not only for the influence of the surface on the electronic structure of the adsorbate, but also for adsorbate-induced effects on the surface states. Variations in the capacity of a site to respond to the perturbation induced by the adsorbate is described by properties of the d-band of the site and its nearest neighbors, as well as by electronic properties of the adsorbate. The model is fully generalizable over the transition metals, but relies on high-quality estimation of the d-band center and upper band edge of the surface site. To make the prediction of these properties more efficient, we train a machine-learning model. This uses pDOS-information from bulk alloy structures and geometrical information from the surface to accurately estimate atom-projected d-band centers and upper band edges of surface atoms.^[9] Together, the physics-based model and the machine-learned surrogate model exemplifies a promising path towards simultaneously utilizing physical insight and computer science.

The motivation to my work stems from the grand challenges in our society, including the efficient utilization of natural resources. The computational-aided rational design of durable, active, and selective catalysts tailored for a given process is an essential part of the transition to a sustainable future. Putting my work on a timescale and in relation to the efforts of other groups around the world, it is obvious that we can only reach the ambitious goals of our society if we collaborate. I believe the work that I and my coworkers have carried out adds a few pieces towards accelerated understanding of the fundamentals of surface chemistry and to the progress towards atomic-scale guidance in catalysts innovation.

References:
[1] T Brinck, JH Stenlid, Adv. Theory Sim., 2, 1800149 (2019)
[2] JH Stenlid, T Brinck, J. Am. Chem. Soc., 139, 11012-11015 (2017)
[3] M Görlin, JH Stenlid, S Koroidov, H-Y Wang, et al., Nat. Commun., 11, 1-11 (2020)
[4] JH Stenlid, T Brinck, J. Org. Chem., 82, 3072-3083 (2017)
[5] JH Stenlid, AJ Johansson, T Brinck, Phys. Chem. Chem. Phys., 21, 17001-17009 (2019)
[6] JA Gauthier, JH Stenlid, F Abild-Pedersen, M Head-Gordon, AT Bell, ACS Energy Lett., 6, 3252-3260 (2021)
[7] JH Stenlid, V Streibel, TS Choksi, F Abild-Pedersen, ChemRxiv, doi:10.26434/chemrxiv-2021-lmlmg (2021)
[8] S Saini, JH Stenlid, F Abild-Pedersen, submitted
[9] JH Stenlid, S Saini, K Winther, J Voss, F Abild-Pedersen, in preparation