Unsupervised Techniques to Detect Quantum Chaos

TL;DR

This paper demonstrates that unsupervised neural networks, specifically Self-Organizing Maps, can detect quantum chaos directly from Hamiltonian matrices without requiring eigenvalue computations, offering a computationally efficient alternative.

Contribution

It introduces a novel method using unsupervised neural networks to identify quantum chaos directly from Hamiltonian matrices, bypassing traditional spectral analysis.

Findings

Neural networks detect transition from integrable to chaotic spectral statistics.

Self-Organizing Maps successfully identify quantum chaos without eigenvalue calculation.

Method reduces computational complexity in quantum chaos detection.

Abstract

Conventional spectral probes of quantum chaos require eigenvalues, and sometimes, eigenvectors of the quantum Hamiltonian. This involves computationally expensive diagonalization procedures. We test whether an unsupervised neural network can detect quantum chaos directly from the Hamiltonian matrix. We use a single-body Hamiltonian with an underlying random graph structure and random coupling constants, with a parameter that determines the randomness of the graph. The spectral analysis shows that increasing the amount of randomness in the underlying graph results in a transition from integrable spectral statistics to chaotic ones. We show that the same transition can be detected via unsupervised neural networks, or more specifically, Self-Organizing Maps by feeding the Hamiltonian matrix directly into the neural network, without any diagonalization procedure.