Theoretische Untersuchung der strukturellen Eigenschaften von Kupferclustern auf Zinkoxid
Theoretische Chemie: Moleküle, Materialien, Oberflächen
Zusammenfassung der Projektergebnisse
Metal clusters supported at oxide surfaces play an important role in heterogeneous catalysis. A prominent example is the Cu/ZnO catalyst used in the synthesis of methanol, one of the most important large-scale base products in chemical industry. Understanding the structure of the catalyst at the atomic level is a mandatory condition for gaining deeper insights into the nature of the active sites and the reaction mechanism of methanol formation from synthesis gas, a mixture of H2, CO and CO2. The goal of this project has been the structural characterization of the Cu/ZnO catalyst by atomistic computer simulations using different types of model systems covering complementary aspects. While electronic structure methods like density-functional theory (DFT) have been shown to provide reliable results for a wide range of systems, the high computational costs make a direct application of first principles methods prohibitively expensive. For this reason, in this project a high-dimensional neural network potential (HDNNP) has been constructed based on DFT reference data, which can provide the energy and forces of complex systems and allows to extend the time and length scales of atomistic simulations by several orders of magnitude without a significant loss in accuracy. Several new algorithms have been developed in the project to increase the performance of atomistic simulations and to analyze large amounts of data. An efficient selection of important reference structures needed for the construction of the HDNNP has been enabled by a new bin and hash algorithm that is also applicable to the selection of descriptors used as input in machine learning potentials and to assess the quality of a given data set. The performance of Monte Carlo simulations has been improved by the development of a flexible grid algorithm that results in a higher acceptance rate without biasing the simulation outcome. To characterize the structural properties of ZnO-supported copper clusters, the HDNNP has been combined with a variety of simulation techniques, which have been applied to different types of model systems. The most stable structures of small copper clusters have been identified employing genetic algorithms showing that even for small particles Cu(110) and Cu(111)-like interface patterns are emerging. The inclusion of the substrate relaxation has been found to be essential in these simulations. For medium-sized and large clusters Monte Carlo methods like basin hopping, as well as simulated annealing, have been performed. The shape and element-distribution of large a-brass particles has been investigated using Monte Carlo simulations in the Semi-Grand Canonical Ensemble. We found that zinc atoms tend to accumulate at the surfaces of the particles, which thus exhibit a high zinc contents. The interior of the brass particles is mainly composed of copper, although for certain compositions ordered phases resembling the bulk phases of a-brass have been observed.
Projektbezogene Publikationen (Auswahl)
- A Bin and Hash Method for Analyzing Reference Data and Descriptors in Machine Learning Potentials
M. L. Paleico and J. Behler
- A flexible and adaptive grid algorithm for global optimization utilizing basin hopping Monte Carlo, J. Chem. Phys. 152 (2020) 094109
M. L. Paleico and J. Behler
(Siehe online unter https://doi.org/10.1063/1.5142363) - Global Optimization of Copper Clusters at the ZnO(10-10) Surface Using a DFT-based Neural Network Potential and Genetic Algorithms, J. Chem. Phys. 135 (2020) 054704
M. L. Paleico and J. Behler
(Siehe online unter https://doi.org/10.1063/5.0014876) - Properties of α-Brass Nanoparticles I: Neural Network Potential Energy Surface, J. Phys. Chem. C 124 (2020) 12682
J. Weinreich, A. Römer, M. L. Paleico and J. Behler
(Siehe online unter https://doi.org/10.1021/acs.jpcc.0c00559)
