§3, PART 1, p. 8 — is the unique solution of (3.4)
ProvedScenarioExact.PartOne.moment_uniquenessmoment-problemp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1probability
Fix an integer . The identities
have exactly one solution among probability distributions on , namely . Precisely:
- Existence. The distribution with distribution function on , i.e. density , satisfies (3.4): for every ,
- Uniqueness. If is a probability measure on with and for every , then for every .
The paper cites the uniqueness of the solution of a moment problem for a distribution with bounded support (Shiryaev, Probability, Chapter II, §12.9, Corollary 1). Combined with (3.4) it gives (3.2).
Formalization Note. Pure measure theory, with no scenario objects. "Distribution with finite support" in the paper means a distribution supported in the bounded interval , entered as . Integrals are lower Lebesgue integrals over (as extended nonnegative reals), the same convention as (3.3) and (3.4). The hypothesis is the paper's ( is the size of ).
Preamble
import Mathlib import Definitions.Def_ScenarioApproach_Generalization_violation import Definitions.Def_ScenarioApproach_Generalization_scenarioProgram import Definitions.Def_ScenarioApproach_Generalization_supportConstraint import Definitions.Def_ScenarioExact_PartOne_Basic open MeasureTheory ScenarioApproach.Generalization open scoped ENNReal
Formal statement
namespace ScenarioExact.PartOne
/-- §3, PART 1, p. 8, sentence after (3.4): `F(α) = α^d` solves (3.4), and it is the only
distribution on `[0, 1]` that does. -/
theorem moment_uniqueness {d : ℕ} (hd : 1 ≤ d) :
-- existence: the law with density `d α^{d-1}` on `[0, 1]` (distribution function `α^d`) solves (3.4)
(∀ m : ℕ, d ≤ m →
(m.choose d : ℝ≥0∞) *
∫⁻ α in Set.Icc (0 : ℝ) 1, ENNReal.ofReal ((1 - α) ^ (m - d) * ((d : ℝ) * α ^ (d - 1))) =
1) ∧
-- uniqueness: every probability law on `[0, 1]` solving (3.4) has distribution function `α^d`
∀ μ : Measure ℝ, IsProbabilityMeasure μ → μ (Set.Icc (0 : ℝ) 1)ᶜ = 0 →
(∀ m : ℕ, d ≤ m →
(m.choose d : ℝ≥0∞) * ∫⁻ α in Set.Icc (0 : ℝ) 1, ENNReal.ofReal ((1 - α) ^ (m - d)) ∂μ = 1) →
∀ α ∈ Set.Icc (0 : ℝ) 1, μ (Set.Iic α) = ENNReal.ofReal (α ^ d) := by sorry
end ScenarioExact.PartOne
Source
Campi & Garatti, The exact feasibility of randomized solutions of uncertain convex programs, SIAM J. Optim. 19(3) (2008); authors' final manuscript, p. 8, §3, PART 1, sentence after (3.4)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.