Uniform matrix product states with a boundary

Canonical forms are central to the analytical understanding of tensor network states, underpinning key results such as the complete classification of one-dimensional symmetry-protected topological phases within the matrix product state (MPS) framework. Yet, the established theory applies only to uniform MPS with periodic boundary conditions, leaving many physically relevant states beyond its reach. Here we introduce a generalized canonical form for uniform MPS with a boundary matrix, thus extending the analytical MPS framework to a more general setting of wider physical significance. This canonical form reveals that any such MPS can be represented as a block-invertible matrix product operator acting on a structured class of algebraic regular language states that capture its essential long-range and scale-invariant features. Our construction builds on new algebraic results of independent interest that characterize the span and algebra generated by non-semisimple sets of matrices, including a generalized quantum Wielandt’s inequality that gives an explicit upper bound on the blocking length at which the fixed-length span stabilizes to an algebra. Together, these results establish a unified theoretical foundation for uniform MPS with boundaries, bridging the gap between periodic and arbitrary-boundary settings, and providing the basis for extending key analytical and classification results of matrix product states to a much broader class of states and operators.

💡 Research Summary

**
The paper “Uniform matrix product states with a boundary” addresses a fundamental gap in the theory of matrix product states (MPS). While canonical forms and fundamental theorems are well‑established for translationally invariant MPS with periodic boundary conditions (PBC), many physically important one‑dimensional quantum states—such as W‑states, Dicke states, domain‑wall states, and states arising from the algebraic Bethe ansatz—cannot be represented as uniform PBC MPS with a constant bond dimension. Instead, they admit an exact description as uniform MPS with an additional boundary matrix, a structure the authors denote as MPS‑X.

Key Contributions

1. Definition of Stable MPS‑X
   The authors introduce the notion of stability: an MPS‑X is stable if, after blocking a finite number of sites (L_{\text{stab}}), the linear span of all products of the local tensors of length (\ell) (denoted (\mathcal{A}(\ell))) coincides with the algebra generated by the original tensors, (\operatorname{Alg}({A_i})). This condition guarantees that the long‑range “backbone” of the state becomes independent of system size and remains invariant under coarse‑graining. All familiar examples (GHZ, W, Dicke, domain‑wall) satisfy stability, while non‑stable MPS‑X exhibit size‑dependent coefficients that can be treated separately.

2. Algebraic Regular Language States (Algebraic RLS)
   Building on the concept of regular language states (RLS) introduced in earlier work, the authors define algebraic RLS by allowing complex weights on symbols via an operator (\hat S(m)). An alphabet (\Sigma) is split into an infinite‑repetition part (\Sigma_\infty) and a finite‑occurrence part (\Sigma_f). By assigning complex amplitudes to words in a regular language, they obtain a flexible class of states that includes weighted GHZ, W, and multi‑excitation Dicke states as special cases. These algebraic RLS capture the essential long‑range structure of stable MPS‑X.

3. Algebraic Structure of Non‑Semisimple Matrix Sets
   The paper provides a thorough analysis of the span and algebra generated by a set of matrices that is not semisimple (i.e., may contain nilpotent blocks). Two central results are proved:

   • Block‑Invertibility after Blocking: By blocking a sufficient number of sites, any set of matrices can be transformed into a block‑diagonal form where each block is invertible within its subspace. This yields a block‑invertible matrix product operator (MPO) that acts on the backbone.
   • Generalized Quantum Wielandt Inequality: Extending the classical Wielandt bound, the authors show that there exists an explicit upper bound (L_{\text{W}} \le C D^4) (with (D) the bond dimension and (C) a constant) such that for all (\ell \ge L_{\text{W}}) the span (\mathcal{A}(\ell)) stabilizes to the full algebra. This bound holds even when the generating set contains nilpotent components, providing a concrete stopping criterion for blocking procedures.

4. Generalized Canonical Form (gCF) for MPS‑X
   Leveraging the algebraic results, the authors construct a generalized canonical form (gCF) for any stable MPS‑X. The gCF decomposes the state into two independent layers:

   • Upper Layer: A block‑invertible uniform MPO (M) that encodes all short‑range entanglement and is gauge‑equivalent to an invertible matrix product operator.
   • Lower Layer (Backbone): An algebraic RLS (|L_N\rangle) that carries the scale‑invariant, long‑range features (phases, topological order, etc.). The overall state is written as (|\psi_N\rangle = M,|L_N\rangle). This mirrors the well‑known pb‑cCF for PBC MPS but now accommodates off‑diagonal blocks introduced by the boundary matrix.

5. Matrix‑CF and Freedom of Representation
   The authors introduce an intermediate matrix canonical form (matrix‑CF) that performs a simultaneous block‑upper‑triangularization using an invertible matrix (P). They then refine the representation with site‑dependent gauge matrices (Z_j) to achieve the final gCF. A detailed analysis of the residual gauge freedom shows:

   • No negligible blocks can be removed without altering the physical state.
   • Additive gauge freedoms are eliminated by fixing the normalization of the algebraic RLS.
   • Different choices of the backbone basis correspond to the same physical subspace.
   • The remaining freedom is precisely characterized by similarity transformations within each block, analogous to the fundamental theorem for PBC MPS.

6. Illustrative Example and Practical Implications
   The paper works out the W‑state explicitly: with bond dimension 2, a boundary matrix (X) and local tensors (A_0, A_1) are given. By blocking two sites, the span stabilizes, the algebra becomes block‑invertible, and the state is expressed in gCF form. This demonstrates that standard tensor‑network algorithms (DMRG, TEBD) can be extended to handle MPS‑X by first bringing the tensors to gCF, then applying existing contraction and optimization techniques on the upper MPO while the lower algebraic RLS remains analytically tractable.

Broader Impact

The generalized canonical form bridges the conceptual gap between periodic and open‑boundary MPS, providing a rigorous foundation for analyzing states that naturally require a boundary matrix. It opens the door to:

• Extending classification schemes of 1D symmetry‑protected topological phases to systems with arbitrary boundaries.
• Designing MPO‑based quantum cellular automata and quantum circuits that incorporate non‑trivial boundary effects.
• Developing more efficient numerical methods for scar states, Bethe‑ansatz wavefunctions, and other exotic excitations that are otherwise inaccessible to standard uniform PBC MPS techniques.
• Applying the algebraic framework to higher‑dimensional PEPS or to tensor‑network representations of quantum channels where non‑semisimple structures appear.

In summary, the paper delivers a comprehensive theoretical toolkit—combining algebraic matrix analysis, a generalized Wielandt bound, and a two‑layer canonical decomposition—that elevates uniform MPS with boundaries to the same level of analytical control as their periodic counterparts, thereby substantially broadening the applicability of tensor‑network methods in quantum many‑body physics.