Accessing scrambling using matrix product operators

TL;DR

This paper introduces a matrix product operator method to efficiently compute out-of-time-ordered correlators in one-dimensional quantum systems, capturing the early growth of scrambling with modest computational resources.

Contribution

It presents a novel tensor network approach to calculate OTOCs throughout the early growth phase, challenging the belief that such methods are limited to short times.

Findings

MPO approximation accurately captures early OTOC growth

Heisenberg operators with low entanglement enable efficient computation

Universal form for OTOC dynamics near the wavefront is proposed

Abstract

Scrambling, a process in which quantum information spreads over a complex quantum system becoming inaccessible to simple probes, happens in generic chaotic quantum many-body systems, ranging from spin chains, to metals, even to black holes. Scrambling can be measured using out-of-time-ordered correlators (OTOCs), which are closely tied to the growth of Heisenberg operators. In this work, we present a general method to calculate OTOCs of local operators in local one-dimensional systems based on approximating Heisenberg operators as matrix-product operators (MPOs). Contrary to the common belief that such tensor network methods work only at early times, we show that the entire early growth region of the OTOC can be captured using an MPO approximation with modest bond dimension. We analytically establish the goodness of the approximation by showing that if an appropriate OTOC is close to its initial value, then the associated Heisenberg operator has low entanglement across a given cut. We use the method to study scrambling in a chaotic spin chain with $201$ sites. Based on this data and OTOC results for black holes, local random circuit models, and non-interacting systems, we conjecture a universal form for the dynamics of the OTOC near the wavefront. We show that this form collapses the chaotic spin chain data over more than fifteen orders of magnitude.