Erasure extremality and rigidity for binary information combining with quantum side information
Suppose you know how uncertain two independent bits remain given their respective quantum records. How much can the two records together hide about whether the bits agree? This article proves the erasure upper-bound conjecture of Hirche and Reeb (Conjecture VII.2) and answers the inverse question: when the parity is almost as uncertain as possible, what must one of the original records look like? Why erasure is worst. The decisive feature is all-or-nothing reliability. For an erasure record the squared posterior contrast \(U\) takes only the values 0 and 1, so \(U^n=U\) at every positive integer order. Partial reliability puts weight between these endpoints. When both records have such weight, their reliability moments decrease together with order, giving a strictly positive contribution to the surviving parity information. In a quantum record the proof uses the full family of binary decision tests to express these moments. Restricting the joint test to spectral blocks leaves a second, nonnegative contribution governed by the Bogoliubov–Kubo–Mori entropy curvature. Both contributions refer to the same physical pair. An exact erasure on either side makes both vanish, however its independent partner encodes the bit. Compute the sharp parity bound from two entropies. Let \(s_i=H(X_i\mid B_i)\) be the conditional von Neumann entropy of bit \(X_i\) given its memory \(B_i\), in bits. For arbitrary bit priors and finite-dimensional quantum memories, the parity entropy satisfies \[s_\oplus=H(X_1\oplus X_2\mid B_1B_2)\le s_1+s_2-s_1s_2.\] At equality, at least one record splits into three orthogonal sectors: a shared sector in which the bit is fair, and two exclusive sectors in which the label is known exactly. One such record suffices for every independent partner. Figure 1 gives a benchmark: at \(s_1=s_2=1/2\), erasure leaves 0.750 bits of parity uncertainty. Two binary symmetric records at those same input entropies leave 0.71354 bits. Partial clues can preserve more information about parity even when all signal states commute. Use the gap to infer source structure. Given the joint parity entropy as well, compute \(G=s_1+s_2-s_1s_2-s_\oplus\). Let \(d_i\) be the half trace distance from the entire bit–memory state of source \(i\) to the nearest erasure source on its own memory, allowing the comparison prior to vary. With \(c=1/(2\ln 2)\) and \(k=c(1-c)\), Theorem 2 gives \[\min(d_1,d_2)\le 30(G/k)^{1/4}.\] This also gives a test in the opposite direction. If both records are known to be at distance at least \(d\) from every erasure source, the entropy gap must be at least \(k(d/30)^4\). The guarantee holds uniformly in the memory dimensions. The fourth-root exponent is optimal in both half trace distance and purified distance. At two fair priors, the half-trace prefactor can be reduced to less than 8 while preserving the comparison prior. Use noncommutation to improve the bound for fair bits. Write \(P_i,Q_i\) for the two normalized signal states of record \(i\) and set \(\gamma_i=\lVert[P_i,Q_i]\rVert_{\mathrm{HS}}^2/4\). The total gap obeys \(G\ge C_*\gamma_1\gamma_2\), with sharp coefficient \(C_*\simeq0.793946\) bits. Thus lower bounds on both commutator squares sharpen the parity ceiling by a calculable amount. Two qubit records approach the coefficient. The proof separates the information retained through partial reliability from the entropy curvature lost by restricting the joint quantum test. The curvature contribution alone is at least \(8\gamma_1\gamma_2/(45\ln 2)\) bits. The sharp coefficient \(C_*\) prices the combined gap. When the coefficients can be used. At first-bit prior \(41/61\) with a fair partner, an explicit pair gives a normalized gap of 0.53235413 using the prior-weighted commutators (Section 4.3). Thus the fair-input coefficient cannot simply be carried across priors. For the distance comparison, the sharp regular tangent ratio in purified distance is approximately 0.241976, whereas a finite pair of full-rank qubit sources has ratio below 0.222626 (Figure 3). Local perturbations identify the exponent but can overestimate the entropy cost of a finite departure from erasure. Files and verification. The 93-page PDF contains the proofs and three figures. The entropy upper bound and fourth-root recovery bound have analytic proofs. The accompanying ZIP gives the verification programs for the sharp fair-input coefficient (Appendix F), the finite full-rank witness (G), and the regular tangent comparison (H). Its README identifies the command and computational requirements for each check.
Authors
- Zixuan He
Institutions
- University of Glasgow (GB)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24377764
- Primary Topic
- Quantum Information and Cryptography
- Type
- preprint