[
  {
    "id": "ARR-2026-0JN84GQ1MD8FW9WR-v1",
    "type": "article",
    "title": "Sharp separable information contraction for reset dynamics",
    "author": [
      {
        "literal": "Lluis Eriksson"
      }
    ],
    "issued": {
      "date-parts": [
        [
          2026,
          10,
          27
        ]
      ]
    },
    "publisher": "AIRR.SCIENCE",
    "archive": "AIRR.SCIENCE",
    "archive_location": "ARR-2026-0JN84GQ1MD8FW9WR",
    "version": "v1",
    "genre": "Research preprint",
    "URL": "https://airr.science/papers/ARR-2026-0JN84GQ1MD8FW9WR/versions/v1/",
    "note": "Research preprint; ARR-2026-0JN84GQ1MD8FW9WR v1. Not peer reviewed by AIRR. Working paper; not admitted to the AIRR accepted collection.",
    "abstract": "We determine the exact mutual-information retention supremum of a channel that mixes the identity with replacement by an arbitrary finite-dimensional state. For separable inputs, PPT inputs, and inputs satisfying a unilateral reduction inequality, the same sharp coefficient is lambda^2/[lambda+(1-lambda)lambda_min(tau)], uniformly over every finite quantum reference. A rational separable family with a single-qubit reference approaches the bound, with a simultaneous estimate controlling finite relative entropy. This separates quantum-reference retention from its classical-reference counterpart without requiring input entanglement. For serial mixtures of unitary transmission and pure replacement, the coefficient depends exactly on the smallest eigenvalue of accumulated replacement noise. Fewer than d stages cannot improve the worst-input coefficient below the product transmission weight; an explicit d-stage schedule attains the optimum at fixed product weight. For continuous Hamiltonian dynamics with a fixed pure reset target, strict improvement below the transmission weight is equivalent to cyclicity of the reset vector. An exact qubit formula shows that increasing the constant drive can worsen worst-input correlation erasure at fixed reset rate and time. We also give necessary Schmidt-number thresholds and exact examples where entangled inputs exceed the separable benchmark even for entanglement-breaking noise. These are normalized retention results, not capacity or positive-rate preservation formulas. Novelty beyond the selected antecedents remains provisional."
  }
]