Efficient Location of Transition State Structures Via the Freezing String Method

Transition states along ab initio potential energy surfaces provide important descriptions of molecular reactivity. However, determining these saddle points in highly complex molecular systems remains challenging. Common approaches to finding transition states in chemical systems include string methods, which usually come with significant computational cost. In order to alleviate the cost of such methods, we propose the freezing string method (FSM) to rapidly generate guess transition state structures that can be further refined by standard eigenvector following algorithms. Eigenvector following algorithms are typically reliable only when a suitable approximation to the transition state structure is available. In order to provide this guess, FSM creates a string of structures connecting reactant and product geometries using a combination of linear synchronous transit and gradients from simulation. This method, which will be elucidated in detail, requires many fewer gradient computations than typical string methods. The performance of FSM will be demonstrated for a variety of chemical transformations, including reactions in zeolites.