§3, PART 1, pp. 8–9 —
ProvedScenarioExact.PartOne.tail_eq_choose_mul_integralLet be a probability space and consider the uncertain program (1.1) with , , convex closed and convex closed , such that every finite instance has a unique solution , and assume the problem is fully-supported. Let be the distribution (3.1) of the violation of and the violation probability (Definition 1.1). Then for every and every ,
The paper partitions the event according to the indexes of the support constraints. Combined with (3.2) and the integration-by-parts identity, this yields (2.3).
Formalization Note. The paper's comes from the indicator of , so it is taken over the half-open interval ; may have an atom at in this generality. The integral is the lower Lebesgue integral against violationLaw, and the identity is an equality of extended nonnegative reals. The solution of the program with constraints is θs N, which under existence and uniqueness is the goal theorem's θstar. Measurability of the event is not assumed; it follows from the joint measurability of the constraint relation and the measurability of the solution maps, which are assumed.
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
namespace ScenarioExact.PartOne
/-- §3, PART 1, pp. 8–9: for a fully-supported problem,
`ℙ^N{V(x*_N) > ε} = binom(N, d) ∫_ε^1 (1 − α)^{N−d} F(dα)`, the integral over `(ε, 1]`. -/
theorem tail_eq_choose_mul_integral
{d : ℕ} {Δ : Type*} [MeasurableSpace Δ]
(P : Measure Δ) [IsProbabilityMeasure P]
(c : EuclideanSpace ℝ (Fin d)) (Θ : Set (EuclideanSpace ℝ (Fin d)))
(Θδ : Δ → Set (EuclideanSpace ℝ (Fin d)))
-- Assumption 1 / (1.1): `𝒳` and every `𝒳_δ` are convex and closed
(hΘ_convex : Convex ℝ Θ) (hΘ_closed : IsClosed Θ)
(hΘδ_convex : ∀ δ, Convex ℝ (Θδ δ)) (hΘδ_closed : ∀ δ, IsClosed (Θδ δ))
-- Assumption 1: for every number `k` of constraints and every sample, `P_k` has a unique solution
(hexu : ∀ (k : ℕ) (ω : Fin k → Δ), ∃! θ, IsSolution c Θ Θδ ω θ)
-- measurability ("assumed for granted", p. 4): the constraint relation is jointly measurable
(hmeas : MeasurableSet {p : EuclideanSpace ℝ (Fin d) × Δ | p.1 ∈ Θδ p.2})
-- `θs k ω` is the solution `x*_k` of `P_k` with the sampled constraints `ω`; measurable in `ω`
(θs : (k : ℕ) → (Fin k → Δ) → EuclideanSpace ℝ (Fin d))
(hθs : ∀ (k : ℕ) (ω : Fin k → Δ), IsSolution c Θ Θδ ω (θs k ω))
(hθs_meas : ∀ k, Measurable (θs k))
-- Definition 2.3: the problem is fully-supported
(hfs : FullySupported c Θ Θδ P)
{N : ℕ} (hd : 1 ≤ d) (hN : d ≤ N) (ε : ℝ) (hε0 : 0 ≤ ε) (hε1 : ε ≤ 1) :
Measure.pi (fun _ : Fin N => P) {ω | ε < violation P Θδ (θs N ω)} =
(N.choose d : ℝ≥0∞) *
∫⁻ α in Set.Ioc ε 1, ENNReal.ofReal ((1 - α) ^ (N - d)) ∂(violationLaw P Θδ θs) := by sorry
end ScenarioExact.PartOne
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.