arXiv:2610.20482
Quantum Spectral Estimation with Multiple Initial States: Improved Error and Cost Scaling
Lingwei Zhang, Zhiyan Ding
Abstract
We study quantum spectrum and eigenstate property estimation using multiple initial states and generalized Hadamard tests. Building on the QFAMES (quantum filtering and analysis of multiplicities in eigenvalue spectra) framework, we introduce a new method that samples evolution times independently for each initial-state pair. Under the stated assumptions, when statistical error determines the sampling cost, the total Hamiltonian simulation cost required for fixed-accuracy eigenvalue and observable estimation is independent of the number of initial states, up to logarithmic factors. This improves scalability when many initial states are needed to explore the target eigenspaces. The method also reduces the cost of multiplicity recovery and retains Heisenberg-limited scaling for eigenvalue estimation. Numerical experiments support the predicted statistical error scalings.
AI Summary
The paper asks how the cost of spectral characterization using multiple trial states scales with the numbers $L,R$ of left and right initial states. It modifies QFAMES, which estimates dominant eigenvalue clusters, their multiplicities, and spectra of observables projected onto their eigenspaces. The change is to sample evolution times independently for each state pair, rather than reuse times across all pairs; filtering, peak search, singular-value thresholding, and generalized eigenvalue estimation remain unchanged. Writing $a=\bar p_{\min}$ and $b=\bar p_{\mathrm{tail}}$ for normalized dominant and tail overlaps, the eigenvalue-estimation simulation-time bound becomes $\widetilde{\mathcal O}(\max\{1/(ab\epsilon),\sqrt{LR}/(a^2\epsilon),LRb/(a\epsilon)\})$, versus $\widetilde{\mathcal O}(LR/(ab\epsilon))$ for shared sampling. It is independent of $L,R$ when $\sqrt{LR}\le a/b$, and retains inverse-precision scaling. Multiplicity recovery costs $\widetilde{\mathcal O}(\max(L,R)/(a^2\Delta))$ rather than quadratic state-pair dependence, in the stated regime. For exactly degenerate clusters, no tail support, constant dominant overlap, and $L=R$, projected-observable estimation costs $\widetilde{\mathcal O}(\|O\|_2^2/(\epsilon_O^2\Delta))$ when $\epsilon_O\lesssim\|O\|_2/L$. The guarantees use independent-entry concentration and refined projected-matrix perturbation bounds; controlled toy-model simulations support the statistical trends. Assumptions include separated clusters, dominant signals, well-conditioned overlap matrices, and a bounded-outcome observable measurement model. State-count independence holds only within specified parameter ranges: minimum per-pair sampling eventually restores an $LR$ cost term. Nonzero cluster widths restrict admissible precision. Costs count Hamiltonian evolution time; broader physical-model tests and adaptive per-entry sampling are left for future work.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 3 figures. Code and numerical results available at https://github.com/lingwei-z/qfames-error-scaling