|Diplom (Physikal. Chemie)||McMaster University, Hamilton, Canada (2005)|
|M.Sc. (Surface Science)||McMaster University, Hamilton, Canada (2007)|
|Ph.D.||Department of Materials Science & Engineering, University of Toronto, Canada (2008-2012)|
|Postdoc||Alexander von Humbold Stipendiat, Anorganische Chemie, Fritz-Haber Institut der Max-Planck Gesellschaft, Berlin (2012-2017)|
|Group leader||MPI CEC (Heterogenous Reactions) (since 2017)|
The Surface Analytics Group in the department of Heterogeneous Reactions focuses on analysis of chemically active materials (i.e. catalysts) in reactive environments, as well as the design and development of novel catalysts based on new design paradigms, such as single-atom alloying. The approach is to understand the underlying physics of the material properties, how these properties come to be and how they play a role in catalytic reactions. As the group's research is centered around experimental observation of these phenomena, several advanced insitu spectroscopic, microscopic and spectromicroscopic methods are used. These methods generate large quantities of data that require ever evolving data analysis techniques to extract useful knowledge from the data. For this reason, the group is heavily involved in developing new analysis algorithms and applying modern machine learning algorithms to the experimental data obtained.
Alloy catalysts are a promising class of materials that could enable fine tuning of catalytic properties and the development of novel catalysts. However, the behavior of these materials when exposed to reactions conditions is still in its infancy.
One the phenomena that occur when exposing alloy catalysts to an oxidative atmosphere is the formation of meta-stable surface species. We have found that AgCu alloys in epoxidation conditions for 2D Cu-oxides that are the 2D analogues to Cu2O. These structures are found to be correlated with ethylene epoxide selectivity, and could play an important role in similar catalytic reactions.
We recently found that some metal combinations give rise to an unique electronic structure when mixed together in dilute amounts. In particular, when small quantities of Cu are added to Ag, free-atom-like d-states are formed in the metal’s valence band. These states resemble electronically, the states for free atoms. We have found that a number of alloys exhibit this property, including AgPd and AgMn. We have synthesized AgPd alloy catalysts, and have experimentally confirmed the existence of free-atom-like Pd states in AgPd single-atom-alloys. We have found that Pd exhibits different chemical behavior when present as a free-atom-like state compared to bulk Pd.
In order to examine the wide array of possible alloy compositions, and to relate spectroscopic investigations with catalytic properties, one requires a general means of synthesizing high surface area alloys. Two approaches have been employed to synthesize nanometer-sized alloy particles: 1) laser ablation of bulk alloys, 2) polyol synthesis. The laser ablation method has the benefit that one can start from bulk alloy foils, produced using well-known traditional metallurgical methods. The down side is that the rate of particle production is rather slow, resulting in small yields. We have successfully synthesized AgCu particles using this method. Alternatively, one can use the polyol method, where one heats metal precursors in a polyol such as ethylene glycol, to reduce the metal precursors into metal nanoparticles. The method is simple to employ, but is restricted in the metal combinations that can be formed. We have successfully used the method to make CuPt nanoparticles.
Modern experimental methods generate large volumes of data. The datasets can be complex and high dimensional. In order to analyze the data as fast as it is generated, and to extract the most knowledge from the data as possible, we utilize many modern data science, chemometrics and machine learning algorithms. As an example, a spectromicroscopy map (i.e. a single data set, that is generated in 30 minutes of measurement time) generates an array of 120 000 XPS spectra. We have developed clustering algorithms for rapidly analyzing these large data sets. The algorithms are based on k-means clustering, principle component analysis, non-negative matrix factorization, hierarchical clustering, DB-SCAN and spectral clustering. The scripts developed with these methods allow us to perform nearly live data analysis during data collection.
Research groups collect large quantities of diverse data over the years. Many researchers contribute to these
collections of data. In order for the complete body of data to be useful for large scale analytics, as well as being useful for future generations of scientists, the data must be well organized and stored with all necessary metadata. To this end, it is necessary to develop an information model for the data. We have developed an information model for in-situ XPS data that enables one to query a large collection of XPS data according to material type, spectrum type and conditions. This data model will be implemented in the coming FAIR data initiatives.
The in-depth interpretation of experimental data requires high certainty in the validity of the results. Yet experimental methods are continuously evolving. Thus methods continuously need to be evaluated. We have invested substantial effort into validating the modern in-situ XPS methods. For instance, we have found that the presence of a gas phase introduces artifacts to the photoemission signal, such that a modified background fitting routine needs to be employed in order to correctly quantify spectra.
Additionally, analysis of XPS spectra where complex line shapes are involved, requires the use of reference standards. However, we have found that when heating a sample to elevated temperatures, or close to phase
transitions, the reference spectra measured at room temperature in ultra-high vacuum are no longer valid. We have developed a large collection of reference spectra for oxides with complex line shapes, such as cobalt, iron, manganese, titanium and nickel oxides at elevated pressures, elevated temperatures and close to phase transitions. These data sets are additionally being used as training sets for neural network models to enable rapid analysis of in-situ XPS spectra.