arXiv:2609.38306
Making the most of leftovers: Improved privacy amplification for quantum key distribution
Matthew Simon Tan, Bartosz Regula, Marco Tomamichel
Abstract
The amount of secret key that can be obtained from a quantum key distribution run depends on both the physically observed error rates and the mathematical bounds used to certify security. For finite datasets, conservative bounds force users to discard a substantial fraction of the potentially available key. Here we further refine and extend the privacy amplification bounds achievable through the recent leftover hash lemma of Regula and Tomamichel [arXiv:2603.04493] and incorporate them into the security analysis of quantum key distribution based on entropic uncertainty relations. This improves on state-of-the-art key rates in finite-block regimes, certifying more secret key from the same experimental data without changes to the protocol, and outperforming techniques based on entropy accumulation. The results illustrate how sharper mathematical estimates can directly increase the usable output of a quantum communication system.
AI Summary
The paper asks how much finite-key QKD performance improves when privacy amplification uses sharper bounds compatible with conventional smooth min-entropy security proofs. It refines a recent measured-smooth-entropy leftover hash lemma and incorporates it into an entropic-uncertainty-relation analysis of ideal BB84. For exact-collision $2^*$-universal hashing, it proves $\Delta\leq\varepsilon^2+\tfrac12\sqrt{(|\mathcal Z|-1)\kappa(\varepsilon)2^{-H_{\min}^{\varepsilon}(X|E)}}$, with $1\leq\kappa\leq4/3$ and $\kappa\to1$ as $\varepsilon\to0$. Relative to the conventional lemma, smoothing costs $\varepsilon^2$ rather than $2\varepsilon$; relative to the recent precursor, the entropy-dependent term improves from a cube root to a square root. Ordinary 2-universal hashing incurs at most one additional bit of key-length loss. The proof constructs a Hermitian approximation from purified-distance smoothing and bounds its Bures norm, establishing a sharper comparison with measured smooth collision divergence. For QKD, the analysis combines this bound with exact hypergeometric parameter-estimation probabilities and binomial completeness tails, then optimises protocol parameters. At security $10^{-10}$, completeness error $0.01$, QBER $1\%$, and $10^4$ retained rounds, certified bits per round increase from $0.385$ with the old lemma to $0.456$; the compared Rényi entropy-accumulation calculation gives $0.433$. At $10^6$ rounds, these rates are $0.747$, $0.761$, and $0.756$. The numerical results assume ideal BB84 and fitted reconciliation leakage; some short-block calculations extrapolate those fits beyond their calibration conditions. The entropy-accumulation comparison uses a different sampling protocol. Authentication costs are excluded, so positive output does not establish net key expansion.
Error found: The abstract's broad claim of outperforming entropy-accumulation techniques is supported only for the particular refined REAT implementation and performance models studied. The tabulated EUR improvements use separately optimised protocol parameters and therefore do not quantify gains for one fixed protocol and experimental dataset, although the theorem supports fixed-protocol improvements. The EUR–REAT benchmark also differs in sampling, completeness accounting, and reconciliation assumptions, so the reported advantage cannot be attributed exclusively to the improved privacy-amplification bound.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 14+13 pages