Low-Dimensional Projection of Reactive Islands in Chemical Reaction Dynamics Using a Supervised Dimensionality Reduction Method

Ryoichi Tanaka,Yuta Mizuno,Takuro Tsutsumi,Mikito Toda,Tetsuya Taketsugu,Tamiki Komatsuzaki

2024-03-07

Abstract:Transition state theory is a standard framework for predicting the rate of a chemical reaction. Although the transition state theory has been successfully applied to numerous chemical reaction analyses, many experimental and theoretical studies have reported chemical reactions with a reactivity which cannot be explained by the transition state theory due to dynamic effects. Dynamical systems theory provides a theoretical framework for elucidating dynamical mechanisms of such chemical reactions. In particular, reactive islands are essential phase space structures revealing dynamical reaction patterns. However, the numerical computation of reactive islands in a reaction system of many degrees of freedom involves an intrinsic challenge -- the curse of dimensionality. In this paper, we propose a dimensionality reduction algorithm for computing reactive islands in a reaction system of many degrees of freedom. Using the supervised principal component analysis, the proposed algorithm projects reactive islands into a low-dimensional phase space with preserving the dynamical information on reactivity as much as possible. The effectiveness of the proposed algorithm is examined by numerical experiments for Hénon-Heiles systems extended to many degrees of freedom. The numerical results indicate that our proposed algorithm is effective in terms of the quality of reactivity prediction and the clearness of the boundaries of projected reactive islands. The proposed algorithm is a promising elemental technology for practical applications of dynamical systems analysis to real chemical systems.

Chemical Physics

What problem does this paper attempt to address?

This paper mainly discusses the low-dimensional projection problem of reaction islands in chemical reaction kinetics. The traditional transition state theory has limitations in explaining the rate of certain chemical reactions because the dynamic effects may render its assumptions invalid. The dynamic system theory provides a theoretical framework to reveal the dynamic mechanisms of these complex reactions, in which the reaction island is a key phase space structure. The researchers propose a supervised dimension reduction algorithm for computing the reaction island of a multi-degree-of-freedom reaction system. The algorithm is based on supervised principal component analysis and aims to map the reaction island to a low-dimensional phase space while preserving reaction information as much as possible. Numerical experiments on the extended Hénon-Heiles system validate the effectiveness of the algorithm in predicting reactivity and clearly displaying the boundaries of the projected reaction island. The paper introduces a formulation for the dimension reduction problem, which aims to find a mapping that projects the high-dimensional structure of the reaction island onto a low-dimensional phase space while preserving reaction information. The proposed algorithm utilizes supervised principal component analysis, taking input data matrices and reaction label matrices, to determine the optimal low-dimensional representation by maximizing the dependence between two random variables. The numerical demonstration section showcases the application of the algorithm in Hénon-Heiles model systems with different degrees of freedom. The results show that compared to simple projection methods, the algorithm better preserves the dependence between position and reaction label, thereby enhancing the predictive quality of reactivity and clarity of the reaction island boundaries.

Low-Dimensional Projection of Reactive Islands in Chemical Reaction Dynamics Using a Supervised Dimensionality Reduction Method

Low-dimensional projection of reactivity classes in chemical reaction dynamics using supervised dimensionality reduction

The Invariant Constrained Equilibrium Edge Preimage Curve Method for the Dimension Reduction of Chemical Kinetics

Reduced Description of Complex Dynamics in Reactive Systems

Reaction Networks Resemble Low-Dimensional Regular Lattices

Selecting High-Dimensional Representations of Physical Systems by Reweighted Diffusion Maps

From High Dimensions to Human Insight: Exploring Dimensionality Reduction for Chemical Space Visualization

Reaction Coordinates by Nonlinear Dimensionality Reduction

The Geometry of Reaction Trajectories and Attracting Manifolds in Composition Space

From high-dimensional committors to reactive insights

On sparse identification of complex dynamical systems: A study on discovering influential reactions in chemical reaction networks

A hierarchical dynamical low-rank algorithm for the stochastic description of large reaction networks

Dimensionality reduction in bulk-boundary reaction-diffusion systems

Transport-chemistry coupling in the reduced description of reactive flows

Investigating the hyperparameter space of deep neural network models for reaction coordinates: Revisiting the solvent coordinate in alanine dipeptide isomerization

A global view of reactive coordinate manifolds from nonlinear dimensionality reduction

Reaction coordinate flows for model reduction of molecular kinetics

3-D Full-Field Reconstruction of Chemically Reacting Flow Towards High-Dimension Conditions Through Machine Learning

Decoupled Species and Reaction Reduction: An error-controlled method for Dynamic Adaptive Chemistry simulations

Unsupervised Assisted Directional Design of Chemical Reactions

FRDPC Subspace Construction Integrated with Bayesian Inference for Efficient Monitoring of Dynamic Chemical Processes