Transition-State Search Suggestions

Searching for a transition state is one of the more difficult tasks for molecular modeling software. In contrast to geometry optimization, in which moving downhill in energy to the nearest minimum does not require that any particular path be followed, in transition-state searching it is essential that the correct reaction coordinate be followed uphill to the desired transition state. Over the years we have accumulated experience with transition-state searching to make the following recommendations.

  • Use the QST search method, in preference to LST or the “standard” method. The QST method requires the most user effort because it requires three input structures, but it is the most likely to find the desired transition-state structure because these three structures allow the optimizer to make the best choice of Hessian eigenvector to follow.

  • Prepare the three QST input structures carefully. The extra time spent in preparing high-quality input will be more than compensated by reduced calculation time.

    Start by building your best guess at the TS structure. Use all of your chemical intuition about how molecular structures change as they undergo reactions. Not only do bonds lengthen and shorten, but bond angles also change as atoms change from planar hybridization to pyramidal or vice versa, and torsional motion occurs to avoid steric clashes or to improve orbital overlap. Make all of the changes that you think will distinguish the TS structure from that of the reactant or product. Another option is to perform a geometry optimization in which you constrain those coordinates whose values you feel that you can estimate fairly well, and let everything else relax. The final structure can be used as your guess at the transition state.

    Once you have the guess TS structure, duplicate it twice in the Project Table. Modify the duplicates to form the reactant and product structures. Do not optimize these structures. In order for the QST interpolation scheme to work well, it is essential not to use optimized product or reactant structures, because those structures lie too far away from the transition state structure to provide information about the reaction coordinate. Instead, try to build reactant and product structures that are part of the way downhill from the TS structure. When you modify bond lengths to create the reactant and product structures, the adjustments should be about 0.1 angstrom. When you change bond angles, the adjustments should be about 5 degrees. Torsional changes should be about 10-15 degrees.

  • List any bonds that are created or broken in a connect section in the input file (see Specifying Bonds for Internal Coordinates with a connect Section). This section ensures that the specified coordinates are used by the optimizer, which can greatly reduce the number of steps to find the transition state. Specifying these bonds is critical when bonds are broken or formed. Without this section, bonds that are too far outside the normal range are not generated as internal coordinates, and therefore are not used to guide the search.

  • Label coordinates in the coord section of the input file as active with the #active flag, to designate them as active coordinates (see Active Coordinates in the Jaguar Input File). Active coordinates are used in transition state vetting, Hessian refinement, and selection of a Hessian eigenvector to follow.

  • In the Optimization tab of the Jaguar panel, choose the Quantum Mechanical option from the Initial Hessian option menu, or set inhess=4 in the gen section of the input file. This is an expensive option, particularly for very large structures, but it also provides the best possible initial Hessian. All of the other initial Hessian algorithms are designed for optimizations, not transition state searches, and in particular the Hessians produced by these algorithms are always positive definite (no negative eigenvalues).

  • Since TS searches are often slow to converge and can sometimes fail to find the desired TS, a two-stage approach can be employed. Use a relatively small basis set and cheap level of theory, like B3LYP/6-31G*, to get a transition state or close to it. Then switch to a more accurate method like M06-2X/cc-pVTZ(-f) for the final refinement. If the transition state found by the cheap method is very different from your starting guess structure, then it might be a good idea to build new product and reactant structures based on the new transition state structure.

  • For atom-transfer reactions, it is usually better to perform a simple relaxed geometry scan rather than a TS search. This is because the reaction coordinate in this situation is completely dominated by the distance coordinate involving the atom being transferred. If your scan increment is small enough, you can accurately locate the transition state.