Theorem 2.2, p. 435 — dual representation (2.6) of a proper lsc convex risk function; (A2), (A3), (A4) read off 𝒜 = dom(ρ*)
ProvedConvexRiskFn.Dual.theorem_2_2Let be a measurable space, a locally convex space of -measurable functions , and a linear space of finite signed measures paired with by , compatibly with the topology of , and satisfying condition (C). Let be proper, lower semicontinuous and convex, with conjugate and dual domain . Then
and moreover
- (A2) is monotone () if and only if every is a nonnegative measure;
- when contains the constant function (so that ): (A3) is translation equivariant (, ) if and only if for every ;
- (A4) is positively homogeneous (, ) if and only if
The theorem generalizes the dual representations of coherent and convex risk measures (Artzner et al., Delbaen, Föllmer–Schied, Cheridito et al.) to paired locally convex spaces: each axiom of a risk function corresponds to a property of the measures in the dual domain.
Formalization Note Condition (C) is the standing assumption of §2 and is a hypothesis; it is used in (i). Part (ii) is stated for every element that is the constant function , with written ; the other parts do not require to contain constants. Values are in EReal, where , so measures outside would contribute to a supremum.
Retired 2026-10-27: a Lean-vs-paper fidelity audit found this statement weaker than, or narrower than, the cited result. A corrected statement will be resubmitted.
import Mathlib import Definitions.Def_ConvexRiskFn_Dual_Setting
namespace ConvexRiskFn.Dual
/-- Theorem 2.2, p. 435: let `ρ : 𝒳 → ℝ̄` be proper, lower semicontinuous and convex, and let
`𝒜 = dom(ρ*)`. Then `ρ(X) = sup_{μ ∈ 𝒜} {⟨μ, X⟩ − ρ*(μ)}` for all `X` (2.6), and
(i) (A2) holds iff every `μ ∈ 𝒜` is nonnegative; (ii) (A3) holds iff `μ(Ω) = 1` for every
`μ ∈ 𝒜`; (iii) (A4) holds iff `ρ(X) = sup_{μ ∈ 𝒜} ⟨μ, X⟩` for all `X` (2.7).
Condition (C) is the standing assumption of §2. Part (ii) is stated for every `one ∈ 𝒳` that is
the constant function `1` (needed for (A3) to make sense); parts (2.6), (i), (iii) do not need
constants in `𝒳`. -/
theorem theorem_2_2 {Ω : Type*} [MeasurableSpace Ω] {𝒳 : Type*} [AddCommGroup 𝒳] [Module ℝ 𝒳]
[TopologicalSpace 𝒳] [IsTopologicalAddGroup 𝒳] [ContinuousSMul ℝ 𝒳] [LocallyConvexSpace ℝ 𝒳]
(S : PairedSpaces Ω 𝒳) (hC : CondC S)
(ρ : 𝒳 → EReal) (hp : IsProper ρ) (hlsc : LowerSemicontinuous ρ) (hcvx : A1 ρ) :
(∀ X, ρ X = ⨆ μ ∈ dualDom S ρ, ((pair μ (S.toFun X) : ℝ) : EReal) - conj S ρ μ) ∧
(A2 S ρ ↔ ∀ μ ∈ dualDom S ρ, 0 ≤ μ) ∧
(∀ one : 𝒳, S.toFun one = (fun _ => 1) →
(A3 one ρ ↔ ∀ μ ∈ dualDom S ρ, μ Set.univ = 1)) ∧
(A4 ρ ↔ ∀ X, ρ X = ⨆ μ ∈ dualDom S ρ, ((pair μ (S.toFun X) : ℝ) : EReal)) := by sorry
end ConvexRiskFn.Dual
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.