Lemma 2.10 — SRT and expected SRT functions are convex
ProvedKallMayerSLP.SimpleRecourse.lemma_2_10p2o-gran-per-chapterp2o-plan-bookp2o-v1simple-recoursestochastic-programming
Let be real with , let be the SRT function, and let be the expected SRT function of a real random variable with finite mean. Then
- is convex on for every fixed ;
- is convex on for every fixed ;
- is convex on .
Convexity of is what makes the simple recourse problem a convex program, and convexity of is what Jensen and Edmundson–Madansky bounds for rest on.
Differences from the source:
- encoding: a random variable ξ̃ with distribution ℙ_ξ̃ → a probability measure P on ℝ (the law of ξ̃); reason: every quantity in the statement depends on ξ̃ only through its law; same instances
- encoding: 𝔼_ξ̃[ξ̃] provided to exist (Definition 2.3) → Integrable (fun ξ => ξ) P; reason: existence of a finite mean is integrability of the identity under the law
Preamble
import Mathlib import Definitions.Def_KallMayerSLP_SimpleRecourse_srt import Definitions.Def_KallMayerSLP_SimpleRecourse_expectedSRT open MeasureTheory
Formal statement
namespace KallMayerSLP.SimpleRecourse
/-- **Lemma 2.10 (convexity of SRT and ESRT functions).**
Kall & Mayer, *Stochastic Linear Programming* (2005), Chapter 3, Lemma 2.10, p.229 (PDF 238),
with Definition 2.3, p.228 (PDF 237). Let `φ` be an SRT function (`α + β ≥ 0`) and `Φ` the
corresponding expected SRT function. Then `φ(z, ·)` is convex in `ξ` for any fixed `z ∈ ℝ`,
`φ(·, ξ)` is convex in `z` for any fixed `ξ ∈ ℝ`, and `Φ(·)` is convex in `z`.
**Formalization Note.** `P` is the distribution of `ξ̃`, a probability measure on `ℝ`; `hP` is the
book's standing proviso "`𝔼_ξ̃[ξ̃]` provided to exist" of Definition 2.3. -/
theorem lemma_2_10 (α β γ : ℝ) (hαβ : 0 ≤ α + β) (P : Measure ℝ) [IsProbabilityMeasure P]
(hP : Integrable (fun ξ => ξ) P) :
(∀ z : ℝ, ConvexOn ℝ Set.univ (fun ξ => srt α β γ z ξ)) ∧
(∀ ξ : ℝ, ConvexOn ℝ Set.univ (fun z => srt α β γ z ξ)) ∧
ConvexOn ℝ Set.univ (expectedSRT α β γ P) := by sorry
end KallMayerSLP.SimpleRecourse
Source
Kall & Mayer, Stochastic Linear Programming (2005), Chapter 3, Lemma 2.10, p.229 (PDF 238)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.