A Rényi-2 converse for POVM outcome probabilities:
ProvedBosonicLearn.povm_renyi2_conversebosonic-gaussian-statesquantum-informationquantum-learning
The Cauchy–Schwarz substitute for Fano's inequality and the Holevo bound in the proof of Theorem 9.
Let be a POVM on ( countable) with a measurable outcome space, a density operator, and a diagonal density operator with . If
then for every measurable set of outcomes .
Proof idea: , Cauchy–Schwarz for the Hilbert–Schmidt inner product, and for .
Formalization Note The first conclusion records that is a density operator, so that the Born rule of FockPOVM applies to it; probabilities are (E.law · B).toReal.
Preamble
import Mathlib import Definitions.Def_BosonicLearn_FockPOVM open QAlgorithms.Papers.Independent.Dependency50.P33 (State IsDensity traceSize) open MeasureTheory open scoped ENNReal Matrix
Formal statement
namespace BosonicLearn
/-- **A Rényi-2 (χ²) converse for POVM outcome probabilities.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, §8, proof of Theorem 9, Eqs. (229)–(234), pp. 66–67, where Fano's inequality and the Holevo bound are applied to the
ensemble; this is the Cauchy–Schwarz substitute for that step.
Let `E` be a POVM on `ℓ²(I)`, `ρ` a density operator and `σ = diag(p)` a diagonal density operator with `p_i > 0`.
If `q = ∑_{i,j} |ρ_{ij}|² / √(p_i p_j) = Tr(ρ σ^{−1/2} ρ σ^{−1/2})` is finite, then for every measurable set `B` of
outcomes
`Pr_ρ(B)² ≤ Pr_σ(B) · q`.
**Formalization Note.** `Tr(E ρ) = Tr(σ^{1/4} E σ^{1/4} · σ^{−1/4} ρ σ^{−1/4})`, Cauchy–Schwarz for the Hilbert–Schmidt
inner product, and `Tr(σ^{1/2} E σ^{1/2} E) ≤ Tr(σ E)` for `0 ⪯ E ⪯ 1`. The first conclusion records that `diag(p)` is a
density operator, so that the Born rule of `E` applies to it. -/
theorem povm_renyi2_converse {I O : Type} [DecidableEq I] [Countable I] [MeasurableSpace O]
(E : FockPOVM I O) (ρ : Matrix I I ℂ) (hρ : IsDensity ρ)
(p : I → ℝ) (hp : ∀ i, 0 < p i) (hp1 : HasSum p 1)
(q : ℝ) (hq : HasSum (fun x : I × I => ‖ρ x.1 x.2‖ ^ 2 / Real.sqrt (p x.1 * p x.2)) q)
(B : Set O) (hB : MeasurableSet B) :
IsDensity (Matrix.diagonal fun i => ((p i : ℝ) : ℂ)) ∧
(E.law ρ B).toReal ^ 2 ≤ (E.law (Matrix.diagonal fun i => ((p i : ℝ) : ℂ)) B).toReal * q := by sorry
end BosonicLearn
Source
Chen, Anna Mele, Anna Mele, Preskill, When are bosonic Gaussian states classical to learn?, arXiv:2609.26705, §8, proof of Theorem 9, pp. 65–67 (Eqs. (229)–(234): Fano's inequality and the Holevo bound, replaced here)