Efficiency of AIMMD in highly diffusive regimes with long reactive trajectories

Determine whether Artificial Intelligence for Molecular Mechanism Discovery (AIMMD) maintains computational efficiency during reaction-coordinate refinement for highly diffusive processes characterized by extremely long reactive trajectories, such as salt nucleation, when repeated generation of reactive trajectories is required.

Background

The commentary notes a practical challenge for TPS-based approaches, including AIMMD, in settings where processes are highly diffusive and reactive trajectories become very long. In such cases, iterative refinement of the reaction coordinate requires repeatedly generating long reactive trajectories, potentially impacting efficiency.

The authors explicitly state that it remains to be tested to what extent AIMMD preserves efficiency under these slow dynamical regimes, highlighting an unresolved question about the method’s practical performance in systems like salt nucleation.

References

For highly diffusive processes, such as salt nucleation, reactive trajectories can be extremely long, which raises the question about the efficiency of repeatedly generating such trajectories during RC refinement. The extent to which AIMMD can maintain its efficiency in these slow dynamical regimes remains to be fully tested.

— Path Sampling for Rare Events Boosted by Machine Learning  (2602.05167 - Minh et al., 5 Feb 2026) in Conclusion, second paragraph