Berlin 2018 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
O: Fachverband Oberflächenphysik
O 16: Heterogeneous Catalysis: Theory
O 16.7: Talk
Monday, March 12, 2018, 16:45–17:00, MA 141
Cheap and Accurate Descriptors for the Computational Screening of Bimetallic Catalysts Identified Using Compressed-Sensing — •Mie Andersen1, Sergey V. Levchenko2, Matthias Scheffler2, and Karsten Reuter1 — 1Theoretical Chemistry, Technische Universität München, Germany — 2Fritz-Haber-Institut der Max-Planck-Gesellschaft, Berlin, Germany
Computational screening for new and improved catalyst materials requires cheap and accurate descriptors for energy trends in surface adsorption. To this end, simple and intuitive physical models such as the d-band model or scaling relations have proven highly successful for transition metal catalysts [1]. Here we apply SISSO [2], a recently developed compressed-sensing method, to identify more accurate descriptors for surface adsorption using a training data set of adsorption energies of atomic and molecular adsorbates on the fcc(211) facets of elemental transition metals. The descriptors are constructed as non-linear functions of primary features, which can be obtained at low computational cost from the clean metal surface. Examples of primary features are d- and sp-projected density of states at the Fermi level, d-band filling, d-band center, and the Pauling electronegativity of the metal atom. We illustrate the superior predictive performance of the found descriptors on a test data set of adsorption energies on alloy surfaces and apply the new model to the computational screening of bimetallic catalysts for methanol synthesis.
[1] F. Abild-Pedersen, Catal. Today 272, 6 (2016)
[2] R. Ouyang et al., arXiv:1710.03319