Negation of the Gelbrich projection (false as formalized)
ProvedWassersteinDRO.Gelbrich.projection_ambiguity_set_falseNegation of WassersteinDRO.Gelbrich.projection_ambiguity_set (a0c00f9b), which is false as formalized.
The formalized Proposition 1 claims the image of the type-2 Wasserstein ambiguity set under Q -> (meanVector Q, covarianceMatrix Q) is contained in the mean-covariance uncertainty set U_e(muhat, SigmaHat). But the nominal PN is not assumed to have finite second moments, and covarianceMatrix is a Bochner integral (junk value 0 on non-integrable entries), so it can be non-positive-semidefinite — which meanCovarianceUncertaintySet explicitly forbids via a PosSemidef conjunct.
Counterexample: m = 2, Xi = univ, eps = 1, muhat = 0, PN = law of (sign X, X) for symmetric X with density proportional to 1/(1+|x|^3) (so E|X| < inf = E[X^2]). Then meanVector PN = (0,0) and covarianceMatrix PN = !![1, c; c, 0] with c = E|X| > 0, det = -c^2 < 0, hence not PSD. PN is in ambiguitySet (diagonal coupling gives W_2(PN,PN) = 0 <= 1) but (meanVector PN, covarianceMatrix PN) is rejected by the PosSemidef conjunct of U. The first conjunct fails at Q = PN. The paper's Proposition 1 assumes finite second moments; the formalization does not.
import Mathlib import Definitions.Def_WassersteinDRO_Gelbrich_ambiguitySet import Definitions.Def_WassersteinDRO_Gelbrich_meanVector import Definitions.Def_WassersteinDRO_Gelbrich_covarianceMatrix import Definitions.Def_WassersteinDRO_Gelbrich_meanCovarianceUncertaintySet set_option autoImplicit false open MeasureTheory open scoped ENNReal
namespace WassersteinDRO.Gelbrich
/-- Proposition 1 (Projection of `B_{ε,2}(P̂N)` onto the mean-covariance space),
Kuhn et al. 2019, p. 16-17, is FALSE as formalized: the nominal distribution
`P̂N` is not assumed to have finite second moments, so `covarianceMatrix PN`
(a Bochner integral, junk value `0` on non-integrable entries) can fail to be
positive semidefinite — which `meanCovarianceUncertaintySet` explicitly requires
via its `PosSemidef` conjunct. Concretely, with `m = 2`, `Ξ = univ`, `ε = 1`,
`μ̂ = 0`, the law `PN` of `(sign X, X)` for symmetric `X` with
`E|X| < ∞ = E[X²]` satisfies `PN ∈ ambiguitySet 1 2 univ PN`
(the diagonal coupling gives `W₂(PN,PN) = 0 ≤ 1`),
`(meanVector PN, covarianceMatrix PN) = (0, !![1, E|X|; E|X|, 0])`, and
`!![1, E|X|; E|X|, 0]` has negative determinant, hence is not positive
semidefinite — so the first conjunct fails at `Q = PN`. The paper's Proposition 1
assumes finite second moments; the formalization does not. -/
theorem projection_ambiguity_set_false :
¬∀ {m : ℕ} (ε : ℝ) (Ξ : Set (EuclideanSpace ℝ (Fin m)))
(PN : Measure (EuclideanSpace ℝ (Fin m))) (μhat : EuclideanSpace ℝ (Fin m))
(SigmaHat : Matrix (Fin m) (Fin m) ℝ)
(hμ : meanVector PN = μhat) (hS : covarianceMatrix PN = SigmaHat),
(∀ Q ∈ ambiguitySet ε 2 Ξ PN,
(meanVector Q, covarianceMatrix Q) ∈ meanCovarianceUncertaintySet ε μhat SigmaHat):= by sorry
end WassersteinDRO.Gelbrich