Proposition 3.1 — for ε = γ = 0 and ρ > 0 the worst-case conditional expected loss equals sup_{y∈𝒴} ℓ(y)
ProvedCondPortOT.SingUninf.proposition_3_1Let (covariates) and (returns) be Polish spaces with Borel -algebras, let , , , be the data with empirical measure , let , and let be the optimal transport ball of radius around for the ground cost . Suppose Assumption 2.4 holds with fiber radius and that is not isolated for . Then for every and every loss ,
with both sides in .
This is the case of the robustified conditional portfolio problem (4), with the mass constraint taken as the strict inequality . The result is negative: the worst-case conditional loss does not depend on the data and equals the robust (support-function) value, so conditioning on a single covariate value with an optimal transport ball is uninformative.
Formalization Note stands for at an arbitrary fixed decision , so the theorem quantifies over all functions ; no measurability, boundedness or integrability is assumed, and the conditional expectation uses the extended integral. "" is the fiber , equal to under Assumption 2.4(i). The hypothesis that is not isolated (for every some has ) is an addition to the paper: it is Assumption 2.4(iii) at when the boundary of is taken in an ambient Euclidean space, as the paper's remark on p. 11 suggests, and the proof uses it; with the boundary taken in itself, Assumption 2.4(iii) is vacuous at an isolated point, where the statement fails.
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_CondPortOT_SingUninf_Setting open MeasureTheory
namespace CondPortOT.SingUninf
theorem proposition_3_1 {X Y : Type*} [TopologicalSpace X] [PolishSpace X] [MeasurableSpace X]
[BorelSpace X] [TopologicalSpace Y] [PolishSpace Y] [MeasurableSpace Y] [BorelSpace Y]
(DX : X → X → ℝ) (DY : Y → Y → ℝ) {N : ℕ} (hN : 0 < N) (xh : Fin N → X) (yh : Fin N → Y)
(x0 : X) (hA : Assumption24 DX DY x0 0 xh) (hH1 : NotIsolated DX x0)
(ρ : ℝ) (hρ : 0 < ρ) (L : Y → ℝ) :
worstCasePos DX DY xh yh x0 0 ρ L = ⨆ y, (L y : EReal) := by sorry
end CondPortOT.SingUninf
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.