Developing mechanistic models, identifying informative experiments for model validation, and computational catalyst design – which are key steps towards a comprehensive computational analysis of a heterogeneous catalytic system – entails optimizing a desired objective. Consequently, these can be formulated as nonlinear optimization problems subject to constraints imposed by the reaction network, thermochemistry, and catalyst properties.
To this end, in this talk, we discuss broadly our efforts in developing a general framework based on nonlinear optimization that allows for parameter estimation, model-based experimental design, and catalyst design. Specifically, our framework formulates a nonlinear objective that includes minimizing model-data mismatch, maximizing surface coverage or reaction flux, and maximizing reaction turnover frequency or selectivity subject to constraints that represent the microkinetic model. In this context, we consider both: (a) a nonlinear programming (NLP) reformulation of a continuously stirred tank reactor (CSTR) microkinetic model, operating at steady-state, and (b) a traditional formulation of a microkinetic model as a system of differential algebraic equations (DAE). Using our formulation, we show that global optimal solutions of large scale parameter estimation problems can be explored with robustness and orders of magnitude speed-up over off-the-shelf software. Further, NLP formulation allows for rapid optimization at different reaction conditions and is also applicable for reactions with many independent reactivity descriptors. These results suggest an ambitious method to elucidate reaction mechanisms and design and search for improved catalysts.