Physical Meaning of Principal Component Analysis for Lattice Systems with Translational Invariance

Su-Chan Park

Published 2024-10-30Version 1

We seek for the physical implication of principal component analysis (PCA) applied to lattice systems with phase transitions, especially when the system is translationally invariant. We present a general approximate formula for a principal component as well as all other eigenvalues and argue that the approximation becomes exact if the size of data is infinite. The formula explains the connection between the principal component and the corresponding order parameter and, therefore, the reason why PCA is successful. Our result can also be used to estimate a principal component without performing matrix diagonalization.