Proof of Theorem 10, p. 14 — F_β(x, ·) is finite and convex
ProvedRUCVaR.MinFormula.F_finite_convexconvexitycvarminimization-formulap2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be a probability measure on , let the loss be integrable, and let . Then for every the positive part is integrable, so
is finite, and is convex on .
These are the first two assertions of Theorem 10; they make the one-sided derivatives of available.
Formalization Note "Finite" is stated as integrability of for every , since a real-valued Lean integral is for a non-integrable function. Only integrability of the single loss and are assumed; the remaining standing assumptions of §2 are not needed and are dropped, which makes the statement stronger.
Preamble
import Mathlib import Definitions.Def_MultistageStochastic_RiskFunctional import Definitions.Def_RUCVaR_MinFormula_Setting open MeasureTheory Filter Topology
Formal statement
namespace RUCVaR.MinFormula
/-- Proof of Theorem 10, p. 14: `F_β(x, ·)` is finite and convex. -/
theorem F_finite_convex {m n : ℕ} (P : Measure (Fin m → ℝ)) [IsProbabilityMeasure P]
(f : (Fin n → ℝ) → (Fin m → ℝ) → ℝ) (x : Fin n → ℝ) (hint : Integrable (f x) P)
(β : ℝ) (hβ1 : β < 1) :
(∀ α : ℝ, Integrable (fun y => max (f x y - α) 0) P) ∧
ConvexOn ℝ Set.univ (Fbeta P f β x) := by sorry
end RUCVaR.MinFormula
Source
Rockafellar & Uryasev, Conditional value-at-risk for general loss distributions, Research Report #2001-5, Univ. of Florida, April 4, 2001, p. 14, proof of Theorem 10, (27)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.