Physical meaning of principal component analysis for classical lattice systems with translational invariance

Abstract

We explore the physical implications of applying principal component analysis (PCA) to translationally invariant classical systems defined on a $d$-dimensional hypercubic lattice. Using Rayleigh-Schrödinger perturbation theory, we demonstrate that the principal components are related to the reciprocal lattice vectors of the hypercubic lattice, and the corresponding eigenvalues are connected to the discrete Fourier transform of the sampled configurations. From a different perspective, we show that the PCA in question can be viewed as a numerical method for computing the ensemble average of the squared moduli of the Fourier transform of physical quantities. Our results also provide a way to determine approximately the principal components of a classical system with translational invariance without the need for matrix diagonalization.

Published in Phys. Rev. E

References in corpus (6)

• Discovering Phases, Phase Transitions and Crossovers through Unsupervised Machine Learning: A critical examination
• Unsupervised machine learning account of magnetic transitions in the Hubbard model
• Unsupervised Learning of Frustrated Classical Spin Models I: Principle Component Analysis
• Quantitative analysis of phase transitions in two-dimensional XY models using persistent homology
• Monte Carlo simulations of bosonic reaction-diffusion systems
• Principal component analysis of absorbing state phase transitions