arXiv:2609.22420
Bootstrap certification of string order in quantum spin chains
Sagnik Banerjee, Haoyu Guo, Debanjan Chowdhury
Abstract
The many-body bootstrap certifies ground-state properties by minimizing energy over correlators constrained only by positivity, without any variational wavefunction. Its cost, however, grows exponentially with the size of the operators involved, placing the long-range string correlators that diagnose topological phases out of reach. We overcome this by treating string operators as primary objects: bare and endpoint-dressed strings satisfy a closed operator algebra with one another and with local words, yielding a semidefinite program whose cost grows only polynomially with string length. A single computation then yields estimates of string correlators of all lengths. Rigorous two-sided bounds are obtained for each target string once the ground-state energy is pinned within a window. When benchmarked against density-matrix renormalization group computations on the cluster Ising and spin$-1$ Heisenberg chains, the method certifies the nonlocal string order of the cluster and Haldane phases directly from the Hamiltonian. Our rigorous certification of string order parameters reveal that the sharpness of the bounds is set by which operators enter the calculation, not by their length alone. By exploiting algebraic closure of strings and local words, our framework lifts spatially extended observables into polynomially tractable bootstrap variables, opening up a route to wavefunction-free certification of nonlocal order in quantum many-body systems.
AI Summary
Many-body bootstrap methods bound ground-state properties using positivity of correlator moment matrices, but including all operator words up to a long string’s length is exponentially costly. This paper asks whether nonlocal string correlators diagnosing one-dimensional symmetry-protected topological phases can instead be bounded with polynomial cost in string length. The construction augments a fixed local-operator basis with bare strings of on-site Hermitian involutions, $u_i^2=\mathbb I$, and endpoint-dressed strings. Canonical product reductions relate local and string moments without an expansion exponential in string length. Translation symmetry block-diagonalizes the moment matrix. One energy-minimization semidefinite program produces estimates for all represented string lengths; separate minimization and maximization programs certify each target observable using an energy window between the bootstrap lower bound and a DMRG variational upper bound. Benchmarks cover transverse-field Ising chains with $N=30$, cluster Ising chains with $N=20$, and spin-1 Heisenberg chains with $N=10$. The reported intervals exclude zero for SPT string correlators in the cluster and Haldane benchmarks. Enlarging the local basis or choosing endpoint operators informed by reduction identities tightens the bounds. Minimal $3\times3$ positivity constraints explain string pinning near ordered limits and give an interval-width ceiling $\Delta S^z(\ell)\lesssim\delta\ell^2$ when the nearest-neighbor correlator is $1-\delta$ and $\ell\sqrt\delta\ll1$. Bounds loosen near criticality and depend on basis selection, whose systematic optimization remains open. The demonstrations certify finite-chain correlators rather than establish thermodynamic-limit string order. Certification uses DMRG energy upper bounds, and energy-window intervals also encompass compatible low-energy states.
Central claim may not hold: The paper certifies nonzero finite-chain string correlators but supplies no size-uniform bound or controlled thermodynamic-limit argument certifying the nonzero string order parameters of the cluster and Haldane phases. For the periodic spin-1 singlet, the full-chain den Nijs–Rommelse string equals the nearest-neighbor correlation across the boundary, so the N=10 string-length scan does not establish the claimed approach to the thermodynamic string-order plateau.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- Main text: 9 pages, 5 figures. Supplementary material: 8 pages, 5 figures