The most successful metal/metal oxide catalysts currently available involve highly-dispersed, low-concentration metal atoms embedded in a metal oxide surface. Palladium metal supported on cerium oxide, is an important example of a highly active catalyst, with applications as an automotive three-way catalyst, in catalytic combustion, and as a solid oxide fuel cell anode material. The activity of these metal/metal oxide catalysts can be uniquely controlled by the support surface structure. Furthermore, these low-concentration metal catalysts have demonstrated significant resistance against sintering, a common multi-component catalyst degradation mechanism, thus indicating superior resistance to heating/cooling cycles and changes in redox enviroment. However, the structure of the active site is challenging to define at the atomistic scale. For the metal-ceria (M/CeO2) catalytic system, dynamic restructuring occurs under reaction conditions and both the ceria and metal structure alter reactivity.
So how then to explain the catalytic and performance behaviors of Pd on ceria? Three Investigators, A.C. van Duin and M. J. Janik of Pennsylvania State University and M. M. Batzill of the University of South Florida, hypothesize that mixed surface oxides of Ce1-xPdxO2-d may provide unique active sites with high activity and stability under certain reaction conditions. To confirm this and in order to fully develop the potential of Pd/CeOx and similar metal/metal oxide catalysts, they believe a detailed, atomistic-scale knowledge of the catalytic conversion mechanisms and the surface dynamics related to substrate-surface interactions is required for the Pd/ceria system. In a collaborative study, the PIs propose to utilize atomistic simulation with Reactive Force-Field (ReaxFF) and Density Functional Theory (DFT) approaches together with experimental surface science studies to investigate the dynamic structure and reactivity of Pd/CeO2 systems. The combined surface science and ReaxFF/DFT approach will provide detailed determination of the structure, stability, and activity of Ce1-xPdxO2-n mixed oxide surfaces. This will help answer questions about this catalyst system.
From the broader perspective, the combination of experimental and computational approaches applied to this complex catalytic system will advance the fundamental understanding of the effect of reducible oxide supports on catalyst stability and activity, as well as provide guidance towards the preparation of highly active M/CeO2 catalysts. Further, the development of an integrated, two-component simulation environment, which is validated against experiment is the outcome of this project. This collaboration between simulation and experiment will provide a roadmap for future catalytic research; the computational tools developed here are generally applicable, thus providing straightforward extension to other catalytic materials.
The research program also closely integrates education and outreach activities. Specifically, at PSU, courses for engineers on atomistic-scale simulation methods will be introduced, which will be complemented by lectures and tutorials on experimental techniques.
Over the coming years, material design faces a crucial challenge. Current practices in material design often depend on inspired guesses, followed by large-scale experiments to optimize material synthesis conditions or material properties. While these practices have obviously been successful, it is beginning to become more and more obvious that these concepts are running out of steam and that dramatic improvements in material properties will require more insight in the fundamental interactions at an atomistic-scale level – allowing, for example, materials to benefit maximally from enhancements related to nano-patterning. To enable such a fundamental interaction, better connections between atomistic-scale simulations and experiments are of foremost importance. Atomistic-scale simulations provide us the opportunity to design materials atom-by-atom – which is currently beyond practical experimental capabilities – and to evaluate trends in material properties. However, these simulations need validation – which ultimately comes from experiment. In this project, which linked two atomistic-scale simulation groups (Michael Janik and Adri van Duin’s groups at Penn State) with an experimental group (Matthias Batzill, University of Southern Florida) we were seeking to improve communication paths between atomistic-scale simulations and experiment, focusing on metal/metal oxide interfaces, which are crucial for the next generation of catalysts. To this end, we developed computational tools that enabled atomistic-scale simulations at unprecedented sizes and time-scales, thus enabling these methods, for the first time, to observe chemical reactions at complex metal/metal oxide interfaces. Figure 1 gives an example of the type of simulations enabled by our project, showing (Fig. 1a) first how a Pd-metal particle can transform into a partial oxide after adsorbing on a Ceria-surfaces and (Fig. 1b) how this Pd-oxide/Ceria interface provides a catalytic site with a unique activity towards hydrocarbon (in this case methane) activation. The simulations in Figure 1 were enabled by connecting Density Functional Theory (DFT) simulation methods – performed by the Janik group – to the ReaxFF reactive force field method – developed by the van Duin group. DFT is a well-established method for calculation accurate reaction energies and reaction barriers, however, the large computational expense associated with this method limits it application to relatively small systems and time scales – typically up to 100 atoms and several picoseconds of time. These size- and time restrictions typically limit DFT applications to highly idealized and symmetrized material configurations – which limits communication with experiments. The ReaxFF method – which is an empirical method, lacking the quantummechanical rigor of DFT – is several magnitudes faster than DFT, enabling applications to large systems (>> 10,000 atoms) and timescale (>> 1 nanosecond). However, at the beginning of this project the ReaxFF method was unproven for complex metal/metal oxide interfaces and its effective time-scale (>> 1 nanosecond), while substantially better than DFT, was still very small compared to typical experimental timescales (microseconds to hours). In this project, we successfully validated the ReaxFF method for Pd/TiO2 and Pd/Ceria interfaces – which are both of significant relevance to catalysts. Furthermore, we greatly enhanced the time-scales and pressure-ranges accessible to ReaxFF by developing a Grand-Canonical Monte Carlo (GCMC) simulation algorithm around the existing ReaxFF program – which allows us to essentially add or remove atoms from an ongoing ReaxFF simulation and evaluate the thermodynamics of this addition/removal step, thus allowing the structure of clusters and surfaces to evolve far faster than with standard ReaxFF-based molecular dynamics (MD) methods – which enforce very small time-steps (sub-femtoseconds). Understanding agglomeration of late transition metal atoms, such as Pd, on metal-oxide supports, such as TiO2, is critical for designing heterogeneous catalysts as well as for controlling metal/oxide interfaces in general. One approach for reducing particle sintering is to modify the metal oxide surface with hydroxyls that decrease ad-atom mobility. We study by scanning tunneling microscopy experiments, density functional theory (DFT) calculations, and Monte Carlo (MC) computer simulations, the atomistic processes of Pd-sintering on a hydroxyl-modified TiO2(011)-2×1 surface. The formation of small 1- to 3-atom clusters that are stable at room temperature is achieved on the hydroxylated surface, while much larger clusters are formed under the same conditions on a hydroxyl-free surface. DFT shows that this is a consequence of stronger binding of Pd-atoms adjacent to hydroxyls and increased surface diffusion barriers for Pd-atoms on the hydroxylated surface. DFT, kinetic MC, and ReaxFF-based NVT-MC simulations show that Pd-clusters larger than single Pd monomers can adsorb the hydrogen from the oxide surface and form Pd-hydrides. This depletes the surface hydroxyl coverage, thus allowing Pd to more freely diffuse and agglomerate at room temperature. Experimentally this causes a bi-modal cluster size distribution with 1-3 atom clusters prevalent at low Pd coverage, while significantly larger clusters become dominant at higher Pd concentrations. This study demonstrates that hydroxylated oxide surfaces can significantly reduce Pd-cluster sizes, thus enabling the preparation of surfaces populated with metal clusters comprised of 1-3 atoms.
