Performance of cluster expansions of coverage-dependent adsorption of atomic oxygen on Pt(111)

David J. Schmidt, Wei Chen, C. Wolverton, William F. Schneider*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

93 Scopus citations

Abstract

A density functional theory (DFT) database of 66 Pt(111)/O formation energies is presented. We fit this database of formation energies to a range of cluster expansions (CEs) of systematically increasing size and flexibility. We find that the performance of the CE depends upon the property or properties of interest. Pair-wise CEs with up to third nearest neighbor interactions poorly predict all metrics. CEs with five to eight pairwise interactions and one to two triplet interactions predicted formation energies and most ground states accurately but predicted average and differential adsorption energies with modest errors. A larger CE captures average and differential adsorption energies as well as formation energies and ground states. The choice of figures in the CEs is also examined. Pair-wise figures and the linear, 1-1-3, triplet are necessary to obtain CEs that qualitatively reproduce the examined properties; however, other figures are more interchangeable. The electronic and strain components of the adsorbate-adsorbate interactions is studied by comparing a CE of DFT formation energies in which atoms were not allowed to relax to the CEs of the relaxed surface. On an unrelaxed Pt surface, interactions are shorter-ranged interactions and more repulsive at first nearest neighbor separation.

Original languageEnglish (US)
Pages (from-to)264-273
Number of pages10
JournalJournal of Chemical Theory and Computation
Volume8
Issue number1
DOIs
StatePublished - Jan 10 2012

ASJC Scopus subject areas

  • Computer Science Applications
  • Physical and Theoretical Chemistry

Fingerprint

Dive into the research topics of 'Performance of cluster expansions of coverage-dependent adsorption of atomic oxygen on Pt(111)'. Together they form a unique fingerprint.

Cite this