Typical Correlation Length of Sequentially Generated Tensor Network States

TL;DR

This paper investigates the typical correlation length in random tensor network states generated sequentially, revealing exponential decay of correlations and dimension-dependent effects of bond dimension on correlation length.

Contribution

It introduces ensembles of sequentially generated tensor network states in 1D and 2D, analyzing their correlation decay and deriving the typical correlation length behavior.

Findings

Correlation length depends only on spatial dimension.

Exponential decay of correlations in typical states.

Bond dimension increases correlation length in 1D, decreases it in 2D.

Abstract

The complexity of quantum many-body systems is manifested in the vast diversity of their correlations, making it challenging to distinguish the generic from the atypical features. This can be addressed by analyzing correlations through ensembles of random states, chosen to faithfully embody the relevant physical properties. Here, we focus on spins with local interactions, whose correlations are extremely well captured by tensor network states. Adopting an operational perspective, we define ensembles of random tensor network states in one and two spatial dimensions that admit a sequential generation. As such, they directly correspond to outputs of quantum circuits with a sequential architecture and random gates. In one spatial dimension, the ensemble explores the entire family of matrix product states, while in two spatial dimensions, it corresponds to random isometric tensor network states. We extract the scaling behavior of the average correlations between two subsystems as a function of their distance. Using elementary concentration results, we then deduce the typical case for measures of correlation such as the von Neumann mutual information and a measure arising from the Hilbert-Schmidt norm. We find for all considered cases that the typical behavior is an exponential decay (for both one and two spatial dimensions). We observe the consistent emergence of a correlation length that depends only on the underlying spatial dimension and not the considered measure. Remarkably, increasing the bond dimension leads to a higher correlation length in one spatial dimension but has the opposite effect in two spatial dimensions.