arXiv:2609.27992
Beyond Shor's Conjecture: Optimal Quantum Measurements at Fixed Error
Jinbo Wang, Qihang Wang, Pengcheng Hou, Kun Chen
Abstract
We prove Shor's conjecture: an orthogonal measurement maximizes the information obtainable from any two quantum states in finite dimension. We then show that an error constraint can require three outcomes. For a class of full-rank qubit pairs, the best orthogonal basis switches abruptly as the error penalty increases. Near the switch, three outcomes yield more information at the same error than any orthogonal measurement or random mixture of them. The extra outcome distinguishes confidence levels in one guess; every outcome is used.
AI Summary
The paper asks when projective measurements suffice to extract information from two quantum states, and whether this remains true under a guessing-error constraint. It gives a constructive proof of Shor’s conjecture in arbitrary finite dimension, including singular states: a rank-one projective measurement attains accessible information. Projective sufficiency also follows from an earlier measured-$f$-divergence theorem; the direct construction additionally classifies all optimal finite-outcome measurements. The construction maximizes a concave objective over one Hermitian matrix. Operator concavity of the logarithm supplies an information upper bound, while Gibbs’ inequality shows that the optimizer’s eigenbasis attains it. Every optimal POVM refines its spectral projectors and is statistically equivalent on the ensemble under classical postprocessing. A multistate extension attains its bound exactly when the optimized posterior operators commute. For full-rank qubit pairs, the paper studies $I-\lambda P_e$ and proves that three outcomes suffice. If two distinct projective bases tie at positive $\lambda$, a three-outcome POVM strictly outperforms every projective measurement and random mixture thereof. A sufficient condition forces such a basis switch. Proofs reduce the problem to scalar moment optimization and dual contact certificates; the extra outcome distinguishes confidence levels within one guess, without discarding results. Certified advantages occur in $5.5\%$ of 20,000 sampled pairs, with median maximal gap $3.4\times10^{-4}$ nat among gapped pairs. Gains are usually small. The switch condition is not necessary; higher-dimensional outcome counts, extension to rank-deficient pairs, and whether advantage windows can be disconnected remain open.
The use of generative AI in analytical checks and in reviewing the numerical protocols and results is disclosed at the end of the main text.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 33 pages, 6 figures, 4 tables. Substantially expanded to include information-error trade-offs, a three-outcome measurement advantage, and generalized losses. Title and author list updated. The original version was publicly available on Zenodo on 13 August 2026: https://doi.org/10.5281/zenodo.21911604