Tensor-Network Analysis of Root Patterns in the XXX Model with Open Boundaries

TL;DR

The paper develops and applies a tensor-network framework to analyze root patterns in the open-boundary XXX model with general boundary fields, addressing the challenging regime where U(1) symmetry is broken. By combining the zero-root (t-W) method with the inhomogeneous T-Q relation and leveraging DMRG with MPO representations, it resolves the structure of both zero and Bethe roots for large systems (up to N≈100). Zero roots organize into bulk strings, boundary strings, and additional roots, while Bethe roots split into regular, line, arc, and paired-line classes; the latter exhibit distinct geometric configurations and evolve across boundary-parameter-induced crossovers, including a restoration of U(1) symmetry as boundary fields are tuned. The work demonstrates that the t-W framework yields simpler root topologies than off-diagonal Bethe Ansatz in this context and provides a scalable numerical pipeline that can be extended to excited states and other integrable models, offering precise insights into thermodynamic properties and symmetry-breaking effects in quantum integrable systems.

Abstract

The string hypothesis for Bethe roots represents a cornerstone in the study of quantum integrable systems, providing access to physical quantities such as the ground-state energy and the finite-temperature free energy. While the $t-W$ scheme and the inhomogeneous $T-Q$ relation have enabled significant methodological advances for systems with broken $U(1)$ symmetry, the underlying physics induced by symmetry breaking remains largely unexplored, due to the previously unknown distributions of the transfer-matrix roots. In this paper, we propose a new approach to determining the patterns of zero roots and Bethe roots for the $Λ-θ$ and inhomogeneous Bethe ansatz equations using tensor-network algorithms. As an explicit example, we consider the isotropic Heisenberg spin chain with non-diagonal boundary conditions. The exact structures of both zero roots and Bethe roots are obtained in the ground state for large system sizes, up to ($N\simeq 60$ and $100$). We find that even in the absence of $U(1)$ symmetry, the Bethe and zero roots still exhibit a highly structured pattern. The zero roots organize into bulk strings, boundary strings, and additional roots, forming two dominant lines with boundary-string attachments. Correspondingly, the Bethe roots can be classified into four distinct types: regular roots, line roots, arc roots, and paired-line roots. These structures are associated with a real-axis line, a vertical line, characteristic arcs in the complex plane, and boundary-induced conjugate pairs. Comparative analysis reveals that the $t-W$ scheme generates significantly simpler root topologies than those obtained via off-diagonal Bethe Ansatz.

Figures (34)

• Figure 1: Tensor network diagram of the transfer matrix: horizontal bonds denote the auxiliary space labeled 0, and vertical bonds represent the quantum spaces labeled $1$ through $N$.
• Figure 2: Tensor network representation of the Hamiltonian $H$. The network has $N$ input and $N$ output indices, each corresponding to a local Hilbert space of dimension $d$.
• Figure 3: Here $E_{0}(\leq E_{1}\leq E_{2} \dots)$ denotes the lowest variational energy obtained by DMRG, which approximates the true ground-state eigenvalue of $H$. The shaded region below $H$ indicates the MPS representation of the approximate ground state corresponding to this energy.
• Figure 4: Tensor network representation of the Hamiltonian $H$ in MPO form. Each tensor represents a local operator with one physical index on the top and bottom, and virtual indices connecting neighboring sites.
• Figure 5: Transformation of the double-row transfer matrix into an MPO. Each local tensor $W_i$ on the right is obtained by contracting the vertical pair of $R$-matrices at site $i$ in the double-row structure, along with the adjacent boundary tensors at the first and last sites.