Quasi-classical trajectory analysis with isometric feature mapping and locally linear embedding: deep insights into...

From the perspective of data analysis, finding and determining reaction paths from the quasi-classical trajectories of a polyatomic reaction system are equivalent to finding low-dimensional manifolds embedded in a high-dimensional space. Two manifold learning methods, isometric feature mapping and locally linear embedding, are applied to the analysis of reaction trajectories, which are calculated by the quasi-classical trajectory approach on a newly developed accurate quartet state NH₃⁺(⁴A) potential energy surface for a multichannel reaction NH⁺ + H₂→ N + H₃⁺/NH₂⁺ + H. The results show that isometric feature mapping can clearly identify different reaction paths from the reactive trajectories, and the locally linear embedding is better for the classification of non-reactive trajectories, and both of them facilitate quantitative analysis. With the help of trajectory analysis, the competition between the two H-atom abstraction reactions can be attributed to two different capture paths.