Local Matrix Product Operators: Canonical Form, Compression, & Control Theory

TL;DR

This paper introduces a novel method for compressing local matrix product operators (MPOs) that represent sums of local terms, overcoming limitations of standard methods and extending applicability to the thermodynamic limit.

Contribution

It presents an 'almost Schmidt decomposition' for MPOs, enabling efficient compression and control, especially in the thermodynamic limit, with practical algorithms for large MPOs.

Findings

Method achieves $\\varepsilon$-close accuracy to MPS methods for finite MPOs.

Extends seamlessly to the thermodynamic limit where MPS techniques fail.

Provides practical algorithms suitable for large MPOs in DMRG applications.

Abstract

We present a new method for compressing matrix product operators (MPOs) which represent sums of local terms, such as Hamiltonians. Just as with area law states, such local operators may be fully specified with a small amount of information per site. Standard matrix product state (MPS) tools are ill-suited to this case, due to extensive Schmidt values that coexist with intensive ones, and Jordan blocks in the transfer matrix. We ameliorate these issues by introducing an "almost Schmidt decomposition" that respects locality. Our method is "$\varepsilon$-close" to the accuracy of MPS-based methods for finite MPOs, and extends seamlessly to the thermodynamic limit, where MPS techniques are inapplicable. In the framework of control theory, our method generalizes Kung's algorithm for model order reduction. Several examples are provided, including an all-MPO version of the operator recursion method (Lanczos algorithm) directly in the thermodynamic limit. All results are accompanied by practical algorithms, well-suited for the large MPOs that arise in DMRG for long-range or quasi-2D models.