Proposition 7.2 — convexity, subadditivity and positive homogeneity of pass to
ProvedKallMayerSLP.RiskMeasures.proposition_7_2convex-analysisp2o-gran-per-chapterp2o-plan-bookp2o-v1risk-measuresstochastic-programming
Let be a linear space of real random variables on and . Let be an -dimensional random vector and a random variable such that for every , and put
Then:
- if is convex on , then is convex on ;
- if and is subadditive, then is subadditive;
- if and is positive homogeneous, then is positive homogeneous.
The properties of hold for every probability distribution of the data, so a convex risk measure always yields a convex stochastic programming model.
Formalization Note is a submodule of the functions ; is given on all functions and only its restriction to is assumed convex, subadditive or positively homogeneous. No probability measure is needed.
Differences from the source:
- encoding: ρ : Υ → ℝ → ρ : (Ω → ℝ) → ℝ with hypotheses on its restriction to the submodule Υ; reason: every function on Υ extends to all functions, and V only evaluates ρ on Υ; same instances
- encoding: ξ ≡ 0 → ξ = 0 as a function; reason: the book's identity of random variables
- encoding: x ∈ ℝⁿ, ηᵀx → x : Fin n → ℝ, η ω ⬝ᵥ x; reason: Lean convention; same instances
Preamble
import Mathlib import Definitions.Def_KallMayerSLP_RiskMeasures_IsPosHomogeneous import Definitions.Def_KallMayerSLP_RiskMeasures_IsSubadditive open MeasureTheory open scoped ENNReal
Formal statement
namespace KallMayerSLP.RiskMeasures
/-- **Proposition 7.2** (properties of `ρ` pass to the evaluation function `V`).
Kall & Mayer, *Stochastic Linear Programming* (2005), Chapter 2, §7, Prop. 7.2, p.182 (PDF 192). Let `Υ` be a linear space of random variables and
`ρ : Υ → ℝ` a real-valued function on `Υ`. Assume that `ηᵀx − ξ ∈ Υ` for all `x` and let
`V(x) := ρ(ηᵀx − ξ)`. Then
a) if `ρ` is convex then `V` is convex;
b) if `ξ ≡ 0` and `ρ` is subadditive then `V` is subadditive;
c) if `ξ ≡ 0` and `ρ` is positive homogeneous then `V` is positive homogeneous.
**Formalization Note.** `Υ` is a submodule of the functions `Ω → ℝ` and `ρ` a function on
`Ω → ℝ` of which only the restriction to `Υ` is constrained; convexity, subadditivity and positive
homogeneity of `ρ` are those of this restriction (a function on the linear space `Υ`).
`x ∈ ℝⁿ` is `Fin n → ℝ`, `ηᵀx` is `η ω ⬝ᵥ x`; `ξ ≡ 0` is `ξ = 0` as a function. No measure
enters. -/
theorem proposition_7_2 {Ω : Type*} (Υ : Submodule ℝ (Ω → ℝ)) (ρ : (Ω → ℝ) → ℝ)
{n : ℕ} (η : Ω → Fin n → ℝ) (ξ : Ω → ℝ)
(hΥ : ∀ x : Fin n → ℝ, (fun ω => η ω ⬝ᵥ x - ξ ω) ∈ Υ) :
(ConvexOn ℝ Set.univ (fun ϑ : Υ => ρ ϑ) →
ConvexOn ℝ Set.univ (fun x : Fin n → ℝ => ρ (fun ω => η ω ⬝ᵥ x - ξ ω))) ∧
(ξ = 0 → IsSubadditive (fun ϑ : Υ => ρ ϑ) →
IsSubadditive (fun x : Fin n → ℝ => ρ (fun ω => η ω ⬝ᵥ x - ξ ω))) ∧
(ξ = 0 → IsPosHomogeneous (fun ϑ : Υ => ρ ϑ) →
IsPosHomogeneous (fun x : Fin n → ℝ => ρ (fun ω => η ω ⬝ᵥ x - ξ ω))) := by sorry
end KallMayerSLP.RiskMeasures
Source
Kall & Mayer, Stochastic Linear Programming (2005), Chapter 2, §7, Proposition 7.2, p.182 (PDF 192)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.