# Quantum v4.39 — A certified noisy repeater witness Local candidate on `2437a0830903a1ad673b4e8c330219b3d22ee4ad` (v4.38), branch `quantum-noisy-witness-v439`. The base is local, not published main. The separately published baseline remains `a42c09d7` (v4.21–v4.36). [Source](../Verification/QuantumNoisyRepeaterWitness.lean) · [Knowledge card](../knowledge/quantum-physics/QuantumNoisyRepeaterWitness.en.md) · [Inventory](quantum-v4.39-inventory.json) ## Fixed model and statement The witness uses constant output-pair post-depolarization quality gamma=19/20, initial Visibility 4/5 and chronological schedule 16 then 16. The spatial step is the existing ideal pairSwap followed by the noisy binary BBPSSW tree. This module changes neither the channel nor the v4.38 operator and accumulator. The first level's depth 16 strictly exceeds 4/5; every smaller depth stays below 4/5. Starting from that particular first output, the second level's depth 16 strictly exceeds the first output; every smaller second depth stays below it. The depth-15 second output even stays below the original 4/5. The targets are therefore explicitly different at the two levels. These are input-specific minimum depths. They do not prove global optimality among all schedules, a discontinuous depth law as a function of noise, or a minimum expected resource cost. The ideal schedule 4 then 2 is a separate previously checked witness, not a noisy alternative with the same guarantee. ## Interval proof rather than a floating-point assumption The source proves monotonicity of noisyRecurrence gamma on [1/4,1] whenever gamma>=0. A general interval-step lemma combines that monotonicity with proved lower and upper endpoint inequalities and preservation of the physical domain. Two chains of 16 steps use outward-rounded rational endpoints on a 10^-8 grid. Every endpoint transition is discharged by norm_num inside Lean. No external numeric result is trusted, no native_decide is used, and no proof limit is raised. The first initial fidelity is exactly 73/100. The second initial fidelity is bounded from the proved first visibility enclosure, not a decimal approximation substituted for the actual first state. | Quantity | Certified lower bound | Certified upper bound | | --- | --- | --- | | First-level F at depth 15 | 0.84911037 | 0.84911044 | | First-level F at depth 16 | 0.85169269 | 0.85169275 | | Second-level F at depth 15 | 0.84975023 | 0.84975040 | | Second-level F at depth 16 | 0.85222959 | 0.85222975 | | First output visibility | 0.8022568 | 0.8022572 | | Second output visibility | 0.8029727 | 0.8029731 | The decimals in this table denote exact terminating rationals. Both actual swapped inputs lie inside the supercritical working window. Strict monotonicity of each fidelity orbit bounds every depth below 16 by its depth-15 value. This turns the certified endpoint bounds into quantified minimum-depth statements. ## Operator and resource consequences The witness proves LocalQualityGain from the two strict gains, and both nonempty prefixes strictly exceed initial quality. The v4.38 refinement identifies the actual unnormalized matrix with p*isotropic(secondLevelVisibility), proves PSD, 0