Support vector machines for learning reactive islands.

Area of Science:

• Dynamical Systems and Chaos Theory
• Machine Learning Applications
• Computational Physics

Background:

• Phase space transport governs dynamical system behavior.
• Reactive islands, formed by manifolds of unstable periodic orbits, quantify transition dynamics.
• Traditional methods require computing unstable periodic orbits and their manifolds.

Purpose of the Study:

• To develop a machine learning framework for learning governing phase space structures.
• To focus on identifying reactive islands in two degrees-of-freedom Hamiltonian systems.
• To provide a direct method for finding reactive islands.

Main Methods:

• Utilizing a machine learning framework, specifically support vector machines.
• Applying the framework to data from trajectories of Hamilton's equations.
• Directly learning the boundaries between distinct dynamical behaviors.

Main Results:

• Support vector machines are effective for identifying phase space boundaries.
• The developed method directly finds reactive islands.
• The approach was successfully applied to the Hénon-Heiles Hamiltonian system.

Conclusions:

• The machine learning framework offers an efficient alternative for identifying reactive islands.
• This method simplifies the analysis of transition dynamics in Hamiltonian systems.
• Further exploration of sampling and learning strategies is warranted.