Classical states are as close as their mixing measures overlap: for
ProvedBosonicLearn.classical_mixture_overlap_boundbosonic-gaussian-statesquantum-informationquantum-learning
Data processing for classical states (mixtures of coherent states).
Let be probability measures on phase space and , the corresponding mixtures of coherent states, specified through their characteristic functions , . If a measure satisfies and , then
With the right-hand side is the total variation distance of and .
Role. Used with BosonicLearn.gaussian_shift_overlap in BosonicLearn.lemma_18_classical.
Formalization Note are probability measures on EuclideanSpace ℝ (Fin n × Fin 2) (interleaved quadrature coordinates), is Mathlib's charFun, and a state is identified with the mixture through BosonicLearn.state_ext_characteristic.
Preamble
import Mathlib import Definitions.Def_BosonicLearn_traceDist import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P14 open QAlgorithms.Papers.Independent.Dependency50.P33 (State traceSize) open QAlgorithms.Papers.Independent.Claimed50.P47 (Occupation) open QAlgorithms.Papers.Independent.Support50.P14 open MeasureTheory ProbabilityTheory open scoped ENNReal Matrix
Formal statement
namespace BosonicLearn
/-- **Two classical states are as close as their mixing measures overlap.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, Lemma 18, p. 60, and the proof of Theorem 8, p. 62 (data processing for classical mixtures of coherent states).
Let `σ, σ'` be the mixtures of coherent states over probability measures `γ, γ'` on phase space
(`χ_σ(ξ) = γ̂(ξ) e^{−|ξ|²/4}`, `χ_{σ'}(ξ) = γ̂'(ξ) e^{−|ξ|²/4}`). If a measure `κ` satisfies `κ ≤ γ` and `κ ≤ γ'`, then
`½‖σ' − σ‖₁ ≤ 1 − κ(ℝ²ⁿ)`. (With `κ = γ ∧ γ'` the right-hand side is the total variation distance of `γ, γ'`.)
**Formalization Note.** `γ, γ'` are probability measures on `EuclideanSpace ℝ (Fin n × Fin 2)` and `γ̂` is `charFun`. -/
theorem classical_mixture_overlap_bound {n : ℕ} (γ γ' κ : Measure (EuclideanSpace ℝ (Fin n × Fin 2)))
[IsProbabilityMeasure γ] [IsProbabilityMeasure γ'] (hκ : κ ≤ γ) (hκ' : κ ≤ γ')
(σ σ' : State (Occupation (Fin n)))
(hσ : ∀ ξ : Fin n × Fin 2 → ℝ, oscillatorCharacteristic σ.coefficients ξ =
charFun γ (WithLp.toLp 2 ξ) * ((Real.exp (-(∑ a, ξ a ^ 2) / 4) : ℝ) : ℂ))
(hσ' : ∀ ξ : Fin n × Fin 2 → ℝ, oscillatorCharacteristic σ'.coefficients ξ =
charFun γ' (WithLp.toLp 2 ξ) * ((Real.exp (-(∑ a, ξ a ^ 2) / 4) : ℝ) : ℂ)) :
traceDist σ'.coefficients σ.coefficients ≤ ENNReal.ofReal (1 - (κ Set.univ).toReal) := by sorry
end BosonicLearn
Source
Chen, Anna Mele, Anna Mele, Preskill, When are bosonic Gaussian states classical to learn?, arXiv:2609.26705, Lemma 18 (first-moment perturbation), Eq. (204), p. 60; proof of Theorem 8, p. 62