Fundamental multiscale models are increasingly being used to describe complex systems. Microkinetic models, which consider a detailed surface reaction mechanism containing all relevant reactions, are prototypical multiscale models. These multiscale microkinetic models employ results from fundamental techniques such as density functional theory (DFT) and transition state theory (TST), along with semi-empirical methods such as bond order conservation (BOC), and couple them with larger scale models (kinetic Monte Carlo or mean field, and reactor scale models) to build predictive models. The predictive capabilities of these models make them suitable candidates for catalyst design. In this work, we describe a novel optimization-based technique, which calculates the optimal values required for atomic descriptors (e.g., binding energies of adsorbates), and then calculates alloys or single metal catalysts that best approach the optimal values. A key advantage of this method is that the models used account explicitly for the coupling between the atomic scale descriptors and the macroscale environment (type of reactor, operating conditions), leading to the optimal design of catalysts under actual conditions of operation. We demonstrate the overall framework for several prototype examples, the ammonia decomposition, the water gas shift, and the selective oxidation of hydrogen.

While this method offers great promise in the rational design of catalysts, it requires the multiscale models used to have quantitative predictive power. This requirement is not always met for models developed purely from first principles. The computational effort in calculating all parameters of a multiscale model for real systems from first principles is prohibitive, and parameter uncertainty still limits full quantitative capabilities of these models. This motivates the development of rational model-based techniques in order to refine uncertain parameters and assess the global model robustness in the entire experimental parameter space. We describe physics-aided methods (based on sensitivity, partial equilibrium, and most abundant reactive intermediate analyses) and statistics-based methods (A, D, and E optimal designs) for the design of experiments, and demonstrate them for the catalytic decomposition of ammonia on ruthenium to produce hydrogen. We show that D optimal and sensitivity-based designs are most promising, and generate conditions that delineate important chemistry. We also develop novel informatics methods to identify optimal regions of the operating space to conduct ‘insightful' experiments. These methods couple, for the first time, the effect of macroscopic scales and microscopic ones in the design of experiments and catalysts.

Finally, we describe results for reactor scale optimization in the same framework, in particular through the optimization of feed location and the introduction of membrane based separation.