Initial estimate for minimum energy pathways and transition states using velocities in internal coordinates

Abstract: Many algorithms for finding reaction pathways require an initial estimate of the minimum energy path (MEP). Most estimation methods use a variational approach and thus must be seeded from an even simpler path, such as one generated by Cartesian interpolation. Often, care must be taken to avoid atomic intersections in this seed path, and the estimator itself may potentially converge to multiple undesirable local minima. As an alternative we form an initial estimate by numerically integrating a velocity vector field that is projected from redundant internal coordinates into the Cartesian manifold. We compare this method to the image dependent pair potential, the geodesic method, and linear Cartesian interpolation using three test cases: the rotation of a methyl group in ethane, HCN$\to$HNC tautomerization, and HONO elimination from dimethylnitramine. In the first test case, a zero-temperature string calculation seeded with our method converges to the MEP in significantly fewer geometry and SCF cycles than any of the others, while in the second, only the geodesic method slightly outperformed ours. In the third test case, we used the midpoint of each path as an initial guess for a transition state calculation. Our midpoint was geometrically the closest to the true transition state and converged in the fewest geometry and SCF cycles.