Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Automated Framework for Analyzing Structural Evolution in On-the-fly Non-adiabatic Molecular Dynamics Using Autoencoder and Multiple Molecular Descriptors

Hangxu Liu, Yifei Zhu, Zhenggang Lan

TL;DR

The paper tackles the challenge of automatically identifying key reaction coordinates in nonadiabatic molecular dynamics by introducing an automated framework that combines nonlinear dimensionality reduction via an Autoencoder with clustering and information entropy. It processes on-the-fly TSH trajectory data using six molecular descriptors (Cartesian, RIC, IDM, MBTR, SOAP, AEV), selects effective descriptors, and identifies distinct reaction channels through DBSCAN clustering in latent space. Information entropy then links reduced coordinates to each channel to reveal the active molecular motions driving photochemical pathways. Validated on keto isocytosine and methaniminium cation, the approach yields interpretable, channel-specific coordinates and aligns with established mechanistic insights, offering a robust, generalizable tool for automated analysis of excited-state dynamics.

Abstract

A major challenge in nonadiabatic molecular dynamics is to automatically and objectively identify the key reaction coordinates that drive molecules toward distinct excited-state decay channels. Traditional manual analyses are inefficient and rely heavily on expert intuition, creating a bottleneck for interpreting complex photochemical processes. To overcome this, we introduce a fully automated machine-learning framework that directly extracts these coordinates from on-the-fly trajectory surface hopping data. By combining an Autoencoder for nonlinear dimensionality reduction with clustering and information entropy analysis, our method autonomously maps reaction channels and pinpoints their governing structural motions. When applied to keto isocytosine and the methaniminium cation, the framework objectively revealed invovled reaction channels and corresponding active coordinates with high efficiency and accuracy. This work establishes an effective paradigm for mechanistic insight in excited-state dynamics, transforming raw trajectory data into clear, interpretable reaction mechanisms.

An Automated Framework for Analyzing Structural Evolution in On-the-fly Non-adiabatic Molecular Dynamics Using Autoencoder and Multiple Molecular Descriptors

TL;DR

The paper tackles the challenge of automatically identifying key reaction coordinates in nonadiabatic molecular dynamics by introducing an automated framework that combines nonlinear dimensionality reduction via an Autoencoder with clustering and information entropy. It processes on-the-fly TSH trajectory data using six molecular descriptors (Cartesian, RIC, IDM, MBTR, SOAP, AEV), selects effective descriptors, and identifies distinct reaction channels through DBSCAN clustering in latent space. Information entropy then links reduced coordinates to each channel to reveal the active molecular motions driving photochemical pathways. Validated on keto isocytosine and methaniminium cation, the approach yields interpretable, channel-specific coordinates and aligns with established mechanistic insights, offering a robust, generalizable tool for automated analysis of excited-state dynamics.

Abstract

A major challenge in nonadiabatic molecular dynamics is to automatically and objectively identify the key reaction coordinates that drive molecules toward distinct excited-state decay channels. Traditional manual analyses are inefficient and rely heavily on expert intuition, creating a bottleneck for interpreting complex photochemical processes. To overcome this, we introduce a fully automated machine-learning framework that directly extracts these coordinates from on-the-fly trajectory surface hopping data. By combining an Autoencoder for nonlinear dimensionality reduction with clustering and information entropy analysis, our method autonomously maps reaction channels and pinpoints their governing structural motions. When applied to keto isocytosine and the methaniminium cation, the framework objectively revealed invovled reaction channels and corresponding active coordinates with high efficiency and accuracy. This work establishes an effective paradigm for mechanistic insight in excited-state dynamics, transforming raw trajectory data into clear, interpretable reaction mechanisms.

Paper Structure

This paper contains 19 sections, 10 equations, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. TSH Dynamics
  4. Descriptors
  5. Cartesian Coordinates
  6. Redundant Internal Coordinates (RICs)
  7. Inversed Distance Matrix (IDM)
  8. Many-Body Tensor Representation (MBTR)
  9. Smooth Overlap of Atomic Positions (SOAP)
  10. Atomic Environment Vectors (AEV)
  11. Autoencoder
  12. Clustering
  13. Information Entropy
  14. Implementation
  15. Results
  16. ...and 4 more sections

Figures (8)

  • Figure 1: Schematic illustrations of the Autoencoder architecture.
  • Figure 2: The S0 minimum structures of (a) keto isocytosine and (b) methaniminium cation.
  • Figure 3: Dimensionality reduction results of descriptors via the Autoencoder for Keto isocytosine model. Different descriptors are employed: (a) Cartesian coordinates; (b) RIC; (c) IDM; (d) MBTR; (e) SOAP; (f) AEV.
  • Figure 4: (a) DBSCAN results of RIC descriptors ($\varepsilon$ = 0.15 and $\mathrm{MinPts}$ = 8). (b)-(e) The typical molecular structures of Cluster 1-4.
  • Figure 5: (a),(g),(m),(s) The distribution of information entropy of Cluster 1-4. (b)-(f) The distribution of D(4,1,5,3), D(3,2,8,4), D(5,1,4,8), A(1,4,8) and D(6,1,5,3) of Cluster 1 respectively. (h)-(l) The distribution of D(5,1,4,8), D(4,1,5,3), A(1,4,8), D(6,1,5,3) and D(3,2,8,4) of Cluster 2 respectively. (n)-(r) The distribution of D(4,1,5,3), D(5,1,4,8), D(6,1,5,3), A(1,4,8) and D(6,1,4,8) of Cluster 3 respectively. (s)-(x) The distribution of A(2,8,4), R(2,3), R(2,8), R(7,8) and A(4,8,7) of Cluster 4 respectively.
  • ...and 3 more figures