(4.13)–(4.15) — the -variant is equivalent to , and is convex or concave
ProvedKallMayerSLP.ICC.eq_4_13_to_4_15convexityintegrated-chance-constraintsp2o-gran-per-chapterp2o-plan-bookp2o-strengthenedp2o-v1polyhedrastochastic-programming
Let have law , suppose the expected values of and exist, and put , . For every and every , the three conditions
are equivalent. Moreover:
- , so is the evaluation of at ;
- is convex for ;
- is concave for ;
- for , is linear-affine.
The sign of is what makes the constraint convex or not, and it is why the polyhedral representation below needs .
Differences from the source:
- generalization: ξ is a random vector in ℝʳ, t(ξ), h(ξ) affine in ξ (1.2), law ℙ_ξ → Ξ any measurable space, P any probability measure on Ξ, t, h arbitrary functions with integrable components; reason: the book says (1.2) is not used in the chapter; Ξ = ℝʳ with P = ℙ_ξ gives the book's case
- generalization: α prescribed, with the remark that α ∈ [0, 1] suffices → every real α; reason: the identities hold for all α; [0, 1] is a special case
Preamble
import Mathlib import Definitions.Def_KallMayerSLP_ICC_K import Definitions.Def_KallMayerSLP_ICC_H import Definitions.Def_KallMayerSLP_ICC_KAlpha open MeasureTheory Matrix
Formal statement
namespace KallMayerSLP.ICC
/-- **The `α`-variant (4.13)–(4.15) and the shape of `K_α`.**
Kall & Mayer, *Stochastic Linear Programming* (2005), Chapter 2, (4.13)–(4.15), p.153 (PDF 163),
and p.154 (PDF 164). The inequality `𝔼_ξ[ζ⁺] ≤ α 𝔼_ξ[|ζ|]` (4.13) can be equivalently written as
`(1 − 2α)𝔼_ξ[ζ⁺] + α 𝔼_ξ[ζ] ≤ 0` (4.14) or as `(1 − 2α)𝔼_ξ[ζ⁺] + α(t̄x − h̄) ≤ 0` (4.15). Also
`K_α = (1 − α)K − αH`, the evaluation of `ρ^α_sic = (1 − α)𝔼[ϑ⁺] − α𝔼[ϑ⁻]` at `ζ(x, ξ)`; `K_α` is
convex for `α < ½`, concave for `α > ½`, and linear-affine for `α = ½`.
**Formalization Note.** No sign restriction on `α` is needed; the book's remark that `α ∈ [0, 1]`
suffices is a remark on which `α` give interesting constraints. `ht`, `hh` are the standing
assumption of §4.1 (p.149). -/
theorem eq_4_13_to_4_15 {Ξ : Type*} [MeasurableSpace Ξ] {n : ℕ} (P : Measure Ξ)
[IsProbabilityMeasure P] (t : Ξ → Fin n → ℝ) (h : Ξ → ℝ)
(ht : ∀ j, Integrable (fun ξ => t ξ j) P) (hh : Integrable h P) (α : ℝ) :
(∀ x, (K P t h x ≤ α * ∫ ξ, |zeta t h x ξ| ∂P ↔
(1 - 2 * α) * K P t h x + α * ∫ ξ, zeta t h x ξ ∂P ≤ 0) ∧
((1 - 2 * α) * K P t h x + α * ∫ ξ, zeta t h x ξ ∂P ≤ 0 ↔
(1 - 2 * α) * K P t h x + α * ((∫ ξ, t ξ ∂P) ⬝ᵥ x - ∫ ξ, h ξ ∂P) ≤ 0)) ∧
(∀ x, KAlpha P t h α x = (1 - α) * K P t h x - α * H P t h x) ∧
(α < 1 / 2 → ConvexOn ℝ Set.univ (KAlpha P t h α)) ∧
(1 / 2 < α → ConcaveOn ℝ Set.univ (KAlpha P t h α)) ∧
(α = 1 / 2 → ∀ x, KAlpha P t h α x = α * ((∫ ξ, t ξ ∂P) ⬝ᵥ x - ∫ ξ, h ξ ∂P)) := by sorry
end KallMayerSLP.ICC
Source
Kall & Mayer, Stochastic Linear Programming (2005), Chapter 2, (4.13)–(4.15), p.153 (PDF 163), and p.154 (PDF 164)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.