Corollary 1 — robust optimal values and solution sets over f-divergence balls are consistent
ProvedGenEmpLik.Consistency.corollary_1Let be i.i.d. random elements of a separable metric space with law and let be the empirical distribution of the first . Let be nonempty and compact, let be lower semicontinuous on , with measurable for every and continuous on for every . Let satisfy Assumption A, , and let Assumption E hold. Define the robust and population objectives
and the solution sets and . Then, in outer probability,
So the optimal value of the distributionally robust problem is a consistent estimate of the population optimal value, and its minimisers eventually lie arbitrarily close to the population minimisers: robustness costs nothing in consistency under these conditions.
Formalization Note Convergence in outer probability is stated as: for every , , where for an arbitrary set is Mathlib's outer measure (no measurability assumed). takes values in and the second event is . Samples are 0-based, is the law of , and is robustMean f ρ (ℓ(x;ξ_0),…,ℓ(x;ξ_{n-1})). Nonemptiness of is the paper's standing assumption (p. 1); nonemptiness of the solution sets is not assumed.
import Mathlib import Definitions.Def_GenEmpLik_Coverage_AssumptionA import Definitions.Def_GenEmpLik_Consistency_AssumptionE import Definitions.Def_GenEmpLik_Expansion_robustMean import Definitions.Def_GenEmpLik_Consistency_argminSet import Definitions.Def_GenEmpLik_Consistency_inclDist open MeasureTheory ProbabilityTheory Filter Topology open scoped ENNReal
namespace GenEmpLik.Consistency
/-- Corollary 1 (arXiv:1610.03425v3, p. 17). Let `ξ₀, ξ₁, …` be i.i.d. with law `P₀`, let `f`
satisfy Assumption A, `ρ ≥ 0`, let `X ⊂ ℝ^d` be nonempty and compact, and let `Ξ` be a
separable metric space. Let `ℓ` be lower semicontinuous on `X × Ξ`, `ℓ(·; ξ)` continuous on
`X` for every `ξ`, `ℓ(x; ·)` measurable for `x ∈ X`, and let Assumption E hold. With
`F̂_n(x) = sup { E_P[ℓ(x; ξ)] : P ≪ P̂_n, D_f(P ‖ P̂_n) ≤ ρ/n }` and `F(x) = E_{P₀}[ℓ(x; ξ)]`,
`inf_X F̂_n − inf_X F → 0` and `d_⊂(argmin_X F̂_n, argmin_X F) → 0` in outer probability. -/
theorem corollary_1 {Ω Ξ : Type*} [MeasurableSpace Ω] [MeasurableSpace Ξ] [MetricSpace Ξ]
[SecondCountableTopology Ξ]
{μ : Measure Ω} [IsProbabilityMeasure μ] {d : ℕ}
(f : ℝ → EReal) (hf : GenEmpLik.Coverage.AssumptionA f) (ρ : ℝ) (hρ : 0 ≤ ρ)
(ξ : ℕ → Ω → Ξ) (hξm : ∀ i, Measurable (ξ i)) (hξind : iIndepFun ξ μ)
(hξid : ∀ i, IdentDistrib (ξ i) (ξ 0) μ μ)
(X : Set (EuclideanSpace ℝ (Fin d))) (hX : IsCompact X) (hXne : X.Nonempty)
(ℓ : EuclideanSpace ℝ (Fin d) → Ξ → ℝ) (hℓm : ∀ x ∈ X, Measurable (ℓ x))
(hℓlsc : LowerSemicontinuous (fun xs : X × Ξ => ℓ xs.1 xs.2))
(hcont : ∀ s, ContinuousOn (fun x => ℓ x s) X)
(hE : AssumptionE (μ.map (ξ 0)) X ℓ) :
(∀ δ : ℝ, 0 < δ → Tendsto (fun n : ℕ => μ {ω |
δ < |(⨅ x : X, GenEmpLik.Expansion.robustMean f ρ (fun i : Fin n => ℓ x (ξ i ω)))
- ⨅ x : X, ∫ s, ℓ x s ∂(μ.map (ξ 0))|}) atTop (𝓝 0)) ∧
(∀ δ : ℝ, 0 < δ → Tendsto (fun n : ℕ => μ {ω |
ENNReal.ofReal δ < inclDist
(argminSet X (fun x => GenEmpLik.Expansion.robustMean f ρ (fun i : Fin n => ℓ x (ξ i ω))))
(argminSet X (fun x => ∫ s, ℓ x s ∂(μ.map (ξ 0))))}) atTop (𝓝 0)) := by sorry
end GenEmpLik.Consistency
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.