Automated MUltiscale simulation environment

Abstract

Multiscale techniques integrating detailed atomistic information on materials and reactions to predict the performance of heterogeneous catalytic full-scale reactors have been suggested but lack seamless implementation. The largest challenges in the multiscale modeling of reactors can be grouped into two main categories: catalytic complexity and the difference between time and length scales of chemical and transport phenomena. Here we introduce the Automated MUltiscale Simulation Environment AMUSE, a workflow that starts from Density Functional Theory (DFT) data, automates the analysis of the reaction networks through graph theory, prepares it for microkinetic modeling, and subsequently integrates the results into a standard open-source Computational Fluid Dynamics (CFD) code. We demonstrate the capabilities of AMUSE by applying it to the unimolecular iso-propanol dehydrogenation reaction and then, increasing the complexity, to the pre-commercial Pd/In₂O₃ catalyst employed for the CO₂ hydrogenation to methanol. The results show that AMUSE allows the computational investigation of heterogeneous catalytic reactions in a comprehensive way, providing essential information for catalyst design from the atomistic to the reactor scale level.

Fig. 1. Schematics of: (a) heterogeneous catalysis and its main phenomena categorized by time and length scale, (b) AMUSE multiscale modeling workflow for heterogeneous catalysis.

Fig. 2. AutoProfLib workflow scheme. First, the optimized adsorption structures are converted to .xyz format. Next, the library encodes the geometric information into a molecular graph. The graph information is translated into a connectivity dictionary, which allows the comparison between all the surface intermediates to obtain the mechanism.

Fig. 3. Schematic representation of PyMKM. Taking as input the mechanism and the reaction energies, translated from the reaction energy profile, PyMKM subsequently: (a) recovers the stoichiometric matrix S and calculates the kinetic constants for each elementary step and the net rate, r_net, (b) solves automatically the resulting ordinary differential equations system, being the derivative in time of the surface coverage vector θ, and (c) outputs the rates, selectivity, apparent activation energy and reaction orders.

Fig. 4. Results for iso-propanol dehydrogenation. (a) Mechanism found with the AutoProfLib for iso-propanol dehydrogenation, (b) PyMKM estimation for the apparent activation energy of iso-propanol dehydrogenation on Co(0001) and Co(112̄0), and (c) CFD-derived iso-propanol conversion trend as function of time on Co(0001) and Co(112̄0) surfaces, depicted in orange and dark-green respectively.

Fig. 5. CO₂ hydrogenation results. (a) Reaction mechanism generated for CP_a structure with AutoProfLib, which is general for all In₂O₃-based catalysts, (b) apparent activation energy (filled bars) and selectivity (dots) towards methanol formation, estimated with PyMKM for all systems, compared to experiments (black bars for apparent activation energy, white dots for selectivity) and previous computational results (blue) for all cases, and (c) CFD selectivity results for Pd(111) (dashed lines) and CP_a (full lines) cases.