Local diabatization (LD) propagates electronic amplitudes using time overlaps of adiabatic states and is robust to wave function phase arbitrariness, root character transport, and rotations within near-degenerate manifolds. However, when truncated dynamical bases are not numerically supported by the consecutive-geometry overlaps, polar orthogonalization underlying the LD approach can enforce unitary evolution in unresolved directions, leading to unphysical population transfer. This work introduces an active space screening protocol based on singular values of time overlap matrices. Candidate spaces proposed from persistent energy gaps and state character are accepted only when every singular direction remains above the user-defined threshold and can then be tested for physical convergence across the nested active spaces. Three-state and 30-state models, controlled overlap failure tests, and atomistic (TiO2)2 and C20 trajectories show that formal rank and increasing state count alone do not guarantee reliable LD dynamics. Smaller, well-supported spaces can outperform the nominal full space when the latter is ill-conditioned.
No takes yet. Share an insight, caveat, or question.
Alexey V. Akimov (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: