Krylov Edge Spectroscopy of Symmetry-Protected Topological Phases

AI summary

Overview

Research area: Quantum many-body physics, specifically the detection and classification of one-dimensional symmetry-protected topological (SPT) phases using operator-space (Krylov/Lanczos) methods.

Technical level: Advanced. The paper assumes familiarity with SPT phases, projective edge algebras, Lieb–Robinson-type locality, Lanczos recursion in operator space, and Gibbs-state correlation decay.

Scope (one sentence): The paper introduces "Krylov edge spectroscopy," a protocol that turns the repeated commutators of a local boundary observable with a finite-range Hamiltonian into a semi-infinite hopping chain whose boundary weight, geometry ratios, and symmetry-resolved slow operators diagnose and classify 1D SPT phases without ground-state wavefunctions, entanglement spectra, or guessed string operators.

What This Paper Is About

Symmetry-protected topological phases in one dimension leave a fingerprint at open boundaries — the effective endpoint symmetry actions can anticommute even though the bulk symmetries commute — but that fingerprint is typically hidden from local static probes and requires either an entanglement spectrum, a hand-chosen nonlocal string operator, or prior knowledge of how symmetry splits between edges. The authors ask whether local boundary dynamics alone can reveal this protected edge memory and reconstruct the edge symmetry action. Their answer is a many-body operator-space protocol in which Lanczos recursion on a boundary observable produces an effective one-dimensional chain, and the normalizability of that chain's zero mode becomes an exact, testable criterion for protected boundary memory, with a separate symmetry-resolved search that reads off the topological label.

Key Contributions

1. An exact zero-frequency normalizability criterion. The boundary weight of the semi-infinite Krylov chain, 𝒵_∞ = μ_{O₀}({0}), is shown to be strictly positive if and only if the product sum ∑{m≥1} ∏{j=1}^{m}(b_{2j−1}/b_{2j})² converges (Theorem 1), giving a necessary-and-sufficient statement about persistent boundary memory rather than a finite-depth plateau alone.

2. Geometry and symmetry-based controls for "protected" memory. The protocol defines an open-versus-periodic ratio R_K = 𝒵_K^OBC/𝒵_K^PBC and a boundary-versus-bulk ratio B_K = 𝒵_K^boundary/𝒵_K^bulk, plus a symmetry-preserving boundary perturbation H(h_∂) = H + h_∂V_∂ with [V_∂, U_i] = 0, to distinguish a genuine projective edge from an accidental free endpoint.

3. A finite-depth leakage certificate. Proposition 2 constructs an explicit operator A_K from the Lanczos data whose commutator norm ε_K = b_{2K+1}|α_K|√𝒵_K bounds how much of the reconstructed operator lies away from zero frequency, via μ_{A_K}(|ω| ≥ Ω) ≤ ε_K²/Ω².

4. A symmetry-resolved classifier that recovers the cohomology label. Minimizing the normalized commutator stiffness κ(v) = v†Dv / v†Cv over symmetry-projected boundary operators in fixed charge sectors yields the slowest boundary operator; for ℤ_N × ℤ_N phases the recovered endpoint charge q* gives the SPT label as p = −q* mod N, verified independently by the algebra of a second endpoint operator.

Main Findings

• Cluster chain: topology appears as a Krylov localization transition. For H_cl(λ) with initial operator Z_0, the Lanczos hoppings are exactly b_{2m−1} = 2λ and b_{2m} = 2, giving 𝒵_K = (1 − λ²)/(1 − λ^{2K+2}) and a Krylov localization length ξ_Krylov = 1/|ln λ|. Throughout the SPT phase 0 ≤ λ < 1, 𝒵_K → 1 − λ²; exactly at λ = 1, 𝒵_K = 1/(K+1) and vanishes with depth. The localization-length exponent is ν_K = 1, coinciding with the Ising correlation-length exponent.

• Clock chain: the full projective class table is recovered numerically. For the ℤ₄ × ℤ₄ fixed point the classifier returns X_0, X_0Z_1, X_0Z_1², and X_0Z_1³ for true labels p = 0, 1, 2, 3 respectively, with measured charges q* = 0, 3, 2, 1, reproducing p = −q* mod 4 exactly. For p = 0 the anchored search already returns A_1 = X_0 at ℓ = 1.

• Spin-1 chains: finite β separates Haldane from trivial. At L = 8 and β = 5, the lowest nontrivial AKLT branch in sector (1,0) has κ_1^OBC = 1.12 at ℓ = 1, 6.20 × 10⁻² at ℓ = 2, and 7.19 × 10⁻³ at ℓ = 3, while the large-D (D = 3) control in sector (0,1) has 7.93, 3.13, and 1.77 at the same window sizes. The projective label p_{D₂} = 1 corresponds to the nontrivial Haldane class and p_{D₂} = 0 to the trivial class; a class is assigned only after detection succeeds.

• Increasing β enhances boundary contrast for AKLT but not for large-D. Figure 1 reports the spin-1 ratios at L = 8 as β runs from 0 to 5: the AKLT (D = 0) boundary contrast grows with β while the large-D (D = 3) response stays near unity.

• The cluster transition is visible in both geometry ratios. Figure 2 reports L = 12 as λ crosses λ = 1: R_K and B_K grow strongly with depth for λ < 1 and remain near or below unity for λ > 1.

• Size independence is exact at β = 0 and exponentially accurate at finite β. Proposition 1 states that at fixed Krylov depth K, once the generated support Λ_{2K} has not reached the far end or wrapped a ring, Δ_L b_n = Δ_L 𝒵_K = 0 at β = 0, and O(e^{−γ_β L}) for 0 < β < ∞ with a size-independent decay rate γ_β.

• A first hopping of zero is not sufficient evidence. The authors note that b_1 = 0 alone is inconclusive, since an ordinary conserved or symmetry-forbidden local operator can also have vanishing first hopping; in AKLT, b_1 is nonzero, showing the microscopic probe is not itself conserved, yet 𝒵_K^OBC stays sizable under D₂-preserving perturbations.

Methodology in Plain English

The starting point is a finite-range one-dimensional Hamiltonian, a Gibbs state ρ_β = e^{−βH}/Tr e^{−βH}, and a simple local operator on one boundary site. Taking repeated commutators of that operator with H generates a growing sequence of operators; the Lanczos recursion orthogonalizes them and produces two lists — normalized operators O_n and nonnegative hopping amplitudes b_n — so the whole many-body problem becomes a nearest-neighbor tight-binding chain in operator space, with a tridiagonal matrix T_K built from the b_n.

Detection reads the fate of a zero-energy mode on that chain. If the mode stays normalizable in the infinite-depth limit, the initial boundary operator retains weight at zero frequency forever, which is the signature of protected boundary memory. To check that the memory is genuinely tied to the boundary and not an artifact, the authors repeat the same run with periodic instead of open boundaries (R_K), with the probe moved into the bulk (B_K), and with a boundary perturbation that respects every protecting symmetry.

Classification is a separate search. Inside a boundary window of ℓ sites, the authors build a local operator basis, form the two matrices C_ab = (O_a|O_b)_S and D_ab = ([H,O_a]|[H,O_b])_S, and solve a generalized eigenvalue problem Dv = κCv after projecting onto fixed symmetry-charge sectors. The lowest-stiffness eigenvector is the slowest operator in that charge sector. To avoid assuming an answer, the anchored operator space is projected into every charge sector and q is scanned rather than fixed; ties are resolved by testing whether the recovered operator is unitary up to tolerance τ_U = 10⁻⁸, with coefficient iteration declared converged below 10⁻¹². The topological label is then read directly from the measured endpoint charge.

The numerics never require a ground-state wavefunction, an entanglement spectrum, or full-system exact diagonalization: at β = 0 the matrices C and D are exact local normalized traces, and at finite β a guarded local Gibbs calculation converges as the guard region grows. All benchmarks use only the Hamiltonian and the onsite symmetry matrices as inputs.

Why This Matters

Impact on research. The work turns a static, nonlocal diagnostic problem into a dynamical, local one. It establishes an ordered limit (system size L → ∞ at fixed Krylov depth K, then the K-flow), proves exact size independence at β = 0, and connects a topological transition to a localization-delocalization transition of a Krylov zero mode with the Ising exponent ν = 1 — an analytic link between bulk gap closing, edge physics, and operator-space localization in a solvable benchmark. It also avoids the usual need to guess a dressed edge or string operator, which is the practical bottleneck in numerical SPT studies.

Potential real-world applications (the paper reports benchmarks and proofs, not experimental demonstrations; the following are plausible domains implied by the models studied):

• Quantum simulation of Haldane/AKLT-type spin-1 chains, where the protocol's spin-1 D₂ classification could identify the edge spin-1/2 algebra from local quench or correlation data.
• Engineered Su–Schrieffer–Heeger, cluster, and clock-model devices, where the open-versus-periodic and boundary-versus-bulk ratios translate into a comparison of edge and bulk dynamics.
• Cold-atom and photonic lattice experiments with programmable boundaries, where the symmetry-preserving perturbation test could distinguish a genuine projective edge from a trivial bound state.
• Benchmarking and verification of quantum hardware, using the finite-depth leakage residual ε_K as a quantitative, tolerance-controlled certificate rather than a qualitative fit.

Industry relevance. The protocol is designed for locality: it needs only Gibbs marginals on a boundary window and its interaction collar, and it scales with Krylov depth rather than system size. That makes it structurally attractive for near-term quantum-simulation workflows where only local observables near an edge are accessible. The paper does not report an experimental implementation, hardware runtimes, or resource estimates, so industrial applicability remains a projection rather than a demonstrated result.

Future Directions

• Extend to fermionic, antiunitary, Floquet, and higher-dimensional boundary algebras. The conclusion explicitly lists these as suggested extensions; the present treatment covers one-dimensional bosonic ℤ_N × ℤ_N and D₂ cases.
• Push beyond fixed points and solvable limits. The clock and cluster benchmarks are exactly solvable representatives; how the classification window ℓ, the inverse temperature β, and the tie-resolution tolerance behave in interacting, non-integrable SPT chains is left open.
• Turn degeneracy handling into a general algorithm. The Supplemental Material treats degeneracies in the lowest-stiffness space and provides rules for selecting the physical endpoint and its charge, but robustness of the "unresolved" outcome when several sectors pass τ_U needs systematic study.
• Connect the Krylov localization exponent to other universality classes. The λ → 1 result ν_K = 1 matching the Ising exponent is established for the solvable cluster chain; whether Krylov localization exponents track bulk correlation-length exponents in other transitions is not reported.

Target Audience

Quantum many-body theorists and condensed-matter physicists working on topological phases, operator dynamics, and Krylov/Lanczos methods; researchers in quantum simulation and quantum information who need practical, local-observable diagnostics for topological order; and graduate students or advanced readers already comfortable with SPT classifications, projective edge algebras, and operator-space correlation functions. Readers seeking experimental protocols, resource estimates, or device-level demonstrations will not find them here.

Authors’ abstract

We introduce $Krylov$ $edge$ $spectroscopy$, a many-body operator-space protocol for detecting and classifying one-dimensional symmetry-protected topological phases from local boundary dynamics. A Hermitian boundary operator generates a semi-infinite Krylov hopping chain whose boundary weight obeys an exact zero-frequency normalizability criterion. Open-periodic, boundary-bulk, and symmetry-preserving boundary-perturbation tests identify protected boundary memory, while a finite-depth leakage residual certifies when an explicitly reconstructed operator is already near zero frequency. Classification minimizes the normalized commutator over symmetry-resolved boundary operators in fixed charge sectors. For bosonic $\mathbb{Z}_N \times \mathbb{Z}_N$ phases, the recovered endpoint charge gives the cohomology label. The method requires no explicit ground-state wavefunction, entanglement spectrum, or guessed dressed edge or string operator. For finite-range Hamiltonians, locality eliminates full-system exact diagonalization and guarantees thermodynamic-limit convergence at fixed Krylov depth. We demonstrate the protocol in cluster, clock, Haldane, and trivial spin-1 chains. In the exactly solvable cluster chain, the transition between the gapped topological and trivial phases manifests as a localization-delocalization transition of the Krylov edge mode on the Krylov chain. This transition occurs precisely at the bulk gap closing, and its localization-length exponent coincides with the Ising correlation-length exponent. Through operator Krylov dynamics, our work turns local boundary evolution into a direct spectroscopy of many-body topology.