arXiv:2610.19922
Quantum Folding in Chebyshev-Based Quantum Subspace Diagonalization
Seung Park, Dongni Chen, Stefano Chesi, Kyunghyun Baek, Mahn-Soo Choi
Abstract
Quantum subspace diagonalization (QSD) with a Krylov space spanned by Chebyshev polynomials of the Hamiltonian is a natural ground-state method for fault-tolerant quantum computers, but every guarantee it offers---convergence rate, noise floor, and the cost of turning its classical output back into a physical state---is controlled by the overlap of the reference state with the true ground state, which decays exponentially with system size for any product state and non-trivial Hamiltonian. We propose quantum folding: for a system governed by a local Hamiltonian, seed the QSD with the product of the ground states of its subsystems, which differs from the target only through the interface terms. We prove a nonasymptotic error bound for the Chebyshev QSD that is linear in the measurement noise, by transplanting Kirby's effective-Hamiltonian analysis to the Chebyshev--Krylov basis, and identify the block-encoding rescaling of the Hamiltonian as a parameter of the guarantee. Exact simulations of antiferromagnetic Heisenberg chains show that the folded reference state reduces the Krylov dimension for a given accuracy severalfold and lowers the error under noise by two to five orders of magnitude relative to the Néel state. We show how the Ritz vector is prepared as a physical state and that the ground-state generation probability obeys a tight ceiling set by the overlap, so that recursive folding prepares ground states at polynomial cost. Taken together, these results allow for a scalable QSD framework that avoids the exponentially small ground-state overlaps associated with conventional product-state initialization. Finally we compute the fold fidelity itself: a slow power law for the gapless spin-1/2 chain, a size-independent constant for the gapped spin-$1$ chain, and an exact closed form converging to 1/4 for the AKLT chain.
AI Summary
Quantum subspace diagonalization (QSD) depends strongly on the reference-state ground overlap, which can be exponentially small for product-state initialization. This paper studies “quantum folding”: initializing Chebyshev–Krylov QSD with a product of subsystem ground states and recursively preparing larger systems. The analysis adapts an existing effective-Hamiltonian technique to obtain a nonasymptotic energy-error upper bound linear in projected-matrix noise. For normalized noise $\eta$, its simplified form is $O(\|H\|[(\|H\|/\Delta+D)\eta+\rho_D]/|\gamma'_0|^2)$. The convergence factor $\rho_D$ decays exponentially in $D\Delta/\Omega$ generally, or $D\sqrt{\Delta/\Omega}$ with sufficiently tight lower-edge alignment. Exact Heisenberg-chain simulations, reaching 24 spin-$1/2$ sites and 12 spin-1 sites, report severalfold reductions in required Krylov dimension and energy errors two to five orders of magnitude below Néel initialization under Gaussian moment noise. LCU/QSVT constructions prepare the Ritz state, with thresholding controlling postselection cost. Spectral filtering obeys the tight generation-probability ceiling $pF\le|\gamma_0|^2$. DMRG finds fold overlaps approximately $1.09L^{-0.107}$ for open spin-$1/2$ chains and approaching $0.22$ for spin-1 Heisenberg chains; an exact AKLT expression approaches $1/4$. The recursive cost analysis gives polynomial overlap overhead for constant-overlap gapped chains, but quasi-polynomial overhead for power-law overlaps. It assumes efficient filtering and appropriate edge-sector selection. General overlap lower bounds are not proved; higher-dimensional interfaces retain area-dependent suppression. The fitted critical exponent may have logarithmic corrections, and simulations exclude coherent walk-operator errors. Practical block-encoding normalization also loses the edge-aligned acceleration.
Central claim may not hold: For the gapless chain, the paper's own recursive preparation cost contains exp(Theta((log L)^2)), contradicting its unconditional polynomial-cost conclusion. The favorable overlaps demonstrated for particular one-dimensional chains do not establish scalable preparation for local Hamiltonians generally. The proposed Ritz-value inputs do not certify the true spectral separation required by the preparation filter, leaving its matching success-probability guarantee conditional.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 31 pages, 11 figures; some error bounds improved, some figures regenerated in v2