§3.2, p. 17 — 1/√(χ²(ν, π) + 1) ≤ Σ_x ν(x)√(π(x)/ν(x)) = ⟨ν|π⟩
ProvedQuantumMC.Cooling.inv_sqrt_chiSq_add_one_le_overlapchi-squaredfidelityjensen-inequalityp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1quantum-sample
Let and be probability distributions on a finite set , with for every . Then
where and are the quantum sample states and is their (complex) inner product. Terms with contribute to the middle sum.
Combined with a bound on , this gives a lower bound on the overlap of quantum samples.
Formalization Note The positivity of is implicit on the page (the display divides by ) and is stated as a hypothesis. In Lean, evaluates to when , which matches the sum over the support of . The equality with is stated in .
Preamble
import Mathlib import Definitions.Def_QuantumMC_Cooling_Setting open scoped ENNReal InnerProductSpace
Formal statement
namespace QuantumMC.Cooling
/-- §3.2, p. 17 (Jensen display): for probability distributions `ν ≥ 0` and `π > 0` on `Ω`,
`1/√(χ²(ν, π) + 1) ≤ ∑_x ν(x) √(π(x)/ν(x)) = ⟨ν|π⟩`. -/
theorem inv_sqrt_chiSq_add_one_le_overlap {Ω : Type*} [Fintype Ω] (ν π : Ω → ℝ)
(hν0 : ∀ x, 0 ≤ ν x) (hν : ∑ x, ν x = 1) (hπpos : ∀ x, 0 < π x) (hπ : ∑ x, π x = 1) :
1 / Real.sqrt (chiSq ν π + 1) ≤ ∑ x, ν x * Real.sqrt (π x / ν x) ∧
((∑ x, ν x * Real.sqrt (π x / ν x) : ℝ) : ℂ) = ⟪qsample ν, qsample π⟫_ℂ := by sorry
end QuantumMC.Cooling
Source
Montanaro, Quantum speedup of Monte Carlo methods, arXiv:1504.06987v3, §3.2, p. 17, display after the first sentence
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.