Abstract
Abstract
Information reconciliation (IR) in quantum key distribution (QKD) is usually treated as leaking the total amount of publicly disclosed reconciliation information. In weak-coherent-pulse decoy-state QKD, this can be overly conservative under the Gottesman–Lo–Lütkenhaus–Preskill (GLLP) tagging framework, because disclosures determined by tagged multi-photon bits that are already assumed known to an eavesdropper reveal no additional information and therefore need not be counted again as IR leakage. Here, we experimentally demonstrate a tighter IR-leakage estimation model that excludes this redundant contribution and bounds the additional information the eavesdropper acquires during reconciliation. The model is incorporated into asymptotic and composable finite-key analyses of decoy-state BB84 while leaving the constant privacy-amplification contribution and the underlying GLLP security assumptions unchanged. We experimentally evaluate the model using 2.5 GHz silicon-photonic chips with polarization-encoded decoy-state BB84 protocol under controlled channel loss and over a 55.41 km deployed metropolitan fiber link. The experimental results show a secure key rate increment of up to 78.82%, while the corresponding finite-key analysis indicates ~ 2.35 dB of additional tolerable system loss. These results experimentally demonstrate that tighter reconciliation-leakage accounting can improve the secure key rate performance of practical chip-based QKD systems without modifying the quantum hardware or the public communication required for reconciliation. Consequently, this methodology holds significant potential across diverse real-world quantum-secure communication scenarios, including high-demand metropolitan and intercity networks, satellite-to-ground QKD links, and next-generation quantum internet nodes.