Global Transition State Search - The Transit-Guided Quasi-Newton (STQN) Methods

In many cases it is difficult to guess a good starting point for the local gradient-based transition state optimizer while the simple scanning approach doesn't work (e.g. due to the complex reaction mechanism). For these cases Gaussian offers two variants of the transit-guided quasi-Newton (STQN) method which first interpolate between reactant and product structures in a non-linear way and then initiate a local transition state optimization from the point closest to the approximate transition state structure. The two variants available in Gaussian are used with the keywords:




The first version QST2 only requires the reactant and product structures as input, while the QST3 version needs structures for the reactant, the product, and an approximate transition state (in that order). In both cases the reactant and product structures must be described with all atoms in the same order. Regardless of the actual coordinates used in the input file, the actual interpolation and transition state searches will be performed in redundant internal coordinates. Using the isomerization from HCN to CNH again as an example, the QST2 input file would be:

Please observe that a full geometry specification with comment line and charge and multiplicity must be provided for each of the structures. In this particular case the QST2 method fails due to the very unfavorable choice of redundant internal coordinates made by Gaussian. Instead of moving the hydrogen atom around the C-N moiety, the QST2 coordinates try to move the hydrogen atom through the two central atoms. This will, of course, never lead to a meaningfull starting point for the transition state search. The QST3 method is more successful in this instance, the input file being:

The third structure in the input file corresponds to what has been used as the starting point for the ts optimization before. The transition state is located in this case within 9 optimization cycles. In contrast to transition state optimizations using the local methods discussed before, a full calculation of the Hessian is not required in this case.