pith. sign in

arxiv: 2606.00527 · v1 · pith:GPM6QEYEnew · submitted 2026-05-30 · 🪐 quant-ph

Finite-Sample Selected Covariance Spectra in Classical Shadows

classification 🪐 quant-ph
keywords localselectedcovarianceboundboundspaulishadowshadows
0
0 comments X
read the original abstract

We study finite-sample estimation of selected covariance matrices of classical-shadow outputs. For a general shadow-output vector, we consider its covariance matrix and a fixed selected compression. Our main theorem applies to arbitrary shadow protocols and gives an operator-norm error bound for the selected sample-centered empirical covariance. When the protocol-dependent constants appearing in this bound remain independent of the ambient system size, the required sample size is also independent of the ambient dimension. The proof combines matrix Bernstein concentration, an exact rank-one centering identity, and Weyl and Davis--Kahan perturbation bounds. We verify this bounded-output condition for local measurement settings. For general local product shadow protocols with fixed local dimension, finite-weight product observables lead to bounds controlled by support sizes and local reconstruction coefficients, not by the total number of tensor factors. Hence uniform bounds on selected set size, observable weight, and local reconstruction coefficients imply dimension-independent selected covariance estimation. For biased local Pauli shadows, we evaluate the relevant bound in closed form from the selected Pauli supports and local basis-selection probabilities. We also derive an exact covariance formula governed by Pauli compatibility and inverse-probability overlap factors, showing how measurement bias affects both diagonal variances and off-diagonal statistical couplings. A comparison with global Clifford shadows shows that this dimension-independent local behavior is not automatic for every shadow protocol.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.