(488e) Using Algorithmic Differentiation for the Top Down Identification of Intrinsic Kinetic Parameters from Transient Data

Computational chemistry is widely used to obtain an intrinsic understanding of the kinetics of heterogeneous catalysis using a "bottom up" approach starting from an atomistic description. However, determining the atomistic configuration of the real active site can be challenging for complex materials. Here we present an alternative application of computational techniques that enables extracting these intrinsic kinetic parameters from the "top down" using transient kinetic experimental data. The temporal analysis of products (TAP) reactor system is one promising transient kinetic approach that is advantageous because it allows direct testing of a packed bed consisting of a complex catalytic material. However, simulating a TAP reactor requires the numerical solution of a reaction-diffusion partial differential equation, making it difficult to fit micro-kinetic models to TAP data. One promising tool to combat this issue is algorithmic differentiation, a method to calculate accurate derivatives while avoiding the pitfalls and expense of symbolic or numeric methods. Techniques for utilizing adjoint operators to obtain automatic derivatives of finite element simulations have recently been implemented in some software packages. A combination of finite element methods and adjoint operations offers a promising route forward for time-dependent data analysis.
A Python package has been developed and built around FEniCS, a finite-element simulation code, and Dolfin-Adjoint, an algorithmic differentiation code. The integration of general micro-kinetic models is achieved using input files, enabling users to easily simulate complex reaction mechanisms. This program allows users to compute sensitivity analyses around observables of interest, as well as fit kinetic parameters through the adjoint approach. This work provides a novel tool that makes this process more efficient, flexible, and accessible to researchers interested in analyzing time-dependent data from diverse transient techniques.