Necessity of full relative-error design randomness for structured shadow estimation

Determine whether the full randomness required for relative-error approximate designs is necessary for structured shadow-estimation tasks beyond stabilizer-state fidelities, and characterize when task-specific shadow guarantees can replace conventional approximate-design conditions.

Background

The shallow phase shadow protocol demonstrates that efficient global estimation of stabilizer-state fidelities can be achieved with a sparse Clifford-IQP ensemble that fails to form an arbitrarily accurate relative-error state 2-design. Its second-look principle exploits the structure of stabilizer observables to control bias without requiring generic design randomness.

The discussion explicitly generalizes this issue to other structured shadow-estimation tasks. The unresolved question is whether relative-error approximate-design randomness is genuinely necessary in those settings, or whether suitably task-adapted ensembles can provide rigorous guarantees with substantially less randomness and circuit depth.

References

Second, and more broadly, this work raises the question of whether the full randomness needed for relative-error approximate designs is genuinely necessary for other structured shadow-estimation tasks.

— Constant-depth global shadow estimation  (2609.10408 - Zhang et al., 9 Sep 2026) in Discussion section, first paragraph beginning “Our results point to several compelling directions”

Several practical and theoretical questions remain open. First, our analysis assumes lossless propagation and ideal photon-number-resolving detection; incorporating non-unit efficiencies and dark counts is important for applications to large photonic processors. Second, extending the framework beyond collision-free inputs to higher mode occupations, and to continuous-variable states such as those used in Gaussian boson sampling , remains open. Finally, for general correlated states, it would be useful to reduce the shot-to-shot reconfiguration required by the randomized protocol, for example through unitary designs or partial derandomization, or to determine whether non-trivial bounds can be obtained from only a small number of fixed interferometers.

— Optimal interferometric certification of multi-photon indistinguishability  (2609.37620 - Robbio et al., 29 Sep 2026) in Section Conclusion