Fixed Quantum Measurements That Work Blind: Sharp Bounds on How Much Fisher Information Survives When You Can't Tune Your Detector

Proves the tightest possible guarantees on information retention for fixed measurements in quantum metrology, using projective designs. The Fisher floor approaches ½ — the single-copy quantum limit — as design order grows.

Imagine you're a quality-control inspector who must use one fixed ruler to measure every product on the line — screws, bolts, washers, springs — without knowing what's coming next. Your ruler can't be optimal for everything, but you need a guarantee: no matter what arrives, you'll capture at least some minimum fraction of the information a perfectly tuned instrument would get. This paper asks: what is that minimum fraction for quantum measurements, and which measurement designs hit it exactly? The committed claim: for projective t-designs in dimension d ≥ 2, the authors determine the sharp lower bound on the 'Fisher floor' — the worst-case fraction of quantum Fisher information retained over all pure states and all tangent directions — for order t = 2 and every odd t ≥ 3. At even orders the bounds are nearly tight. The Fisher floor approaches 1/2 as design order grows, which is the fundamental single-copy quantum limit. Finite weighted designs achieve these bounds. The ladder here is primarily internal to quantum metrology theory rather than benchmarked against a competing numerical method. The relevant prior art is the general theory of quantum Fisher information and its bounds under symmetric measurements (Hayashi, Zhu and collaborators, and the quantum state tomography literature). The key advance over prior work is that these bounds are proved to be *sharp* — not just valid, but the tightest possible given only the design order as a constraint. The companion paper on constant-number-of-bases metrology forms the experimental counterpart. Architecturally, this is a pure mathematical physics result. The tools are representation theory (Schur-Weyl duality, irreducible decomposition of tensor-product spaces), spectral analysis of measurement operators, and probabilistic arguments for random subsampling. The core trick is connecting the symmetry properties enforced by projective t-designs to spectral constraints on the Fisher information matrix, then optimizing over the worst case. No quantum hardware, no simulation — pencil, paper, and combinatorial algebra. Integrity is high for a theory paper. The results are mathematical proofs, not numerical experiments. The sharpness claim means the bounds are achieved by explicit constructions (finite weighted designs), so they're not merely upper/lower bound sandwiches with a gap. The probabilistic sampling result — that subsampling a 3-design yields a measurement with linearly many outcomes and a dimension-independent Fisher floor with probability 1 − O(1/d) — is a concrete, falsifiable prediction for experimentalists. No cherry-picking is possible when your result is a theorem. The practical upshot is a recipe for pure-state estimation: take any 3-design, subsample it, and you get a measurement with O(d) outcomes that separates all pure states in Hellinger distance. Feed N copies into maximum-likelihood estimation and you achieve mean infidelity O(d/N) uniformly — optimal up to a constant. This is a concrete, implementable protocol, not an existence theorem. The gap to watch is experimental realization. The theory says 3-designs suffice; constructing efficient 3-designs in high dimensions and implementing them on real quantum hardware remains nontrivial. The authors do not demonstrate this on hardware or in simulation — the next paper in the sequence would naturally target either numerical validation at moderate d or a proof-of-concept experiment. Whether the constant hidden in O(d/N) is practically competitive with adaptive schemes at realistic N and d is an open empirical question the theory alone cannot answer.