Neural-network-enhanced evolutionary algorithm applied to supported metal nanoparticles

Research output: Contribution to journal/Conference contribution in journal/Contribution to newspaperJournal articleResearchpeer-review

  • E. L. Kolsbjerg
  • ,
  • A. A. Peterson, Brown Univ, Brown University, Sch Engn
  • ,
  • B. Hammer

We show that approximate structural relaxation with a neural network enables orders of magnitude faster global optimization with an evolutionary algorithm in a density functional theory framework. The increased speed facilitates reliable identification of global minimum energy structures, as exemplified by our finding of a hollow Pt-13 nanoparticle on an MgO support. We highlight the importance of knowing the correct structure when studying the catalytic reactivity of the different particle shapes. The computational speedup further enables screening of hundreds of different pathways in the search for optimum kinetic transitions between low-energy conformers and hence pushes the limits of the insight into thermal ensembles that can be obtained from theory.

Original languageEnglish
Article number195424
JournalPhysical Review B
Volume97
Issue19
Number of pages9
ISSN2469-9950
DOIs
Publication statusPublished - 16 May 2018

    Research areas

  • POTENTIAL-ENERGY SURFACES, GLOBAL OPTIMIZATION, STRUCTURE SENSITIVITY, GENETIC ALGORITHMS, MATERIALS DESIGN, CLUSTERS, SIMULATIONS, CATALYSIS, APPROXIMATION, REACTIVITY

See relations at Aarhus University Citationformats

ID: 127852340