(27b) Automatic Transition State Searches for On-The-Fly Kinetic Calculations

Computing advances have enabled rule-based automatic generation of kinetic models of large reaction networks to replace the laborious and fallible manual methods. We propose an automatic transition state (TS) generation procedure as the first step to calculate important unknown reaction rates for the automatic Reaction Mechanism Generator (RMG). Currently in RMG, thermodynamic and kinetic parameters are preferentially sourced from experimentally derived or theoretically calculated rates, but these are often unavailable given the large number of required parameters, so the model is completed with estimates derived from group additivity. Quantum chemical calculations have recently been implemented in RMG to calculate thermodynamic parameters on-the-fly when the group additive approximations are not suitable, e.g. fused polycyclic rings; we present a similar approach for transition states.

We apply distance geometry to automatically generate reactant and product structures along the reaction coordinate, and use these as inputs for semi-empirical electronic structure double-ended search algorithms, such as quasi-synchronous transit, to produce a TS estimate. The placement of the reactive atoms plays a key role in the success of this method. A further optimization step finds the saddle point geometries, which have successfully been used as inputs to density function theory (DFT) calculations. Semi-empirical calculations were chosen to generate estimates for the DFT calculations to improve the computational efficiency of the procedure.

The automatic geometry estimation method has been tested on hydrogen abstraction reactions, with the resulting geometries used to start DFT optimizations with M06-2X/6-31+g(d,p). At the time of writing, 194 of the 338 reactions tested (all H-abstraction from C/H/O species in the NIST kinetics database) successfully generated a TS estimate; most of the failures were for abstraction by hydrogen radicals, and have been attributed to the nature of the reaction coordinate.

TS geometries found with DFT have shown, as expected, trends in the distances based on the functional groups of the reactions involved. Applying these trends to a consistent and repeatable procedure will enable direct estimation of the TS geometries, which could also be generated via the distance geometry method. Direct estimation of the TS geometry is expected to increase the success of the automatic procedure, especially in the case of small molecule (H radical) reactions.