Lemma 4.2, p. 1493 — for q̃ = λq + (1 − λ)e_N and λ ≠ 0, ‖a − â‖_{q̃,𝒜} = (1/|λ|)‖a − â‖_{q,𝒜}
ProvedRiskUncSets.InnerApprox.lemma_4_2Let , let be a centrally symmetric generator, let have sample mean , and let with . Then the vector satisfies
where is the Minkowski functional (12) of .
So, in the norm of a fixed centrally symmetric generator, mixing with shrinks or enlarges the unit ball by exactly the factor ; maximizing therefore maximizes the inner approximation in Theorem 4.5.
Formalization Note The statement is written for ranging over . Two corrections to the printed (13) are made and disclosed: the right-hand side uses , not (the page prints on both sides, which would make the identity false for ), and is assumed because the right-hand side contains . When has empty interior the page's two sides are both off ; with Mathlib's gauge both are the junk value there, so the identity remains consistent.
import Mathlib import Definitions.Def_RiskUncSets_InnerApprox_Setting noncomputable section
namespace RiskUncSets.InnerApprox
/-- Lemma 4.2, (13): `‖a − â‖_{q̃,𝒜} = (1/|λ|) ‖a − â‖_{q,𝒜}` for `q̃ = λq + (1−λ)e_N`,
`q ∈ Δ̂ᴺ_sym`, `λ ≠ 0`. -/
theorem lemma_4_2 {N n : ℕ} (hN : 0 < N) (q : Fin N → ℝ)
(hq : q ∈ symRestrictedSimplex N) (a : Fin N → Fin n → ℝ)
(lam : ℝ) (hlam : lam ≠ 0) :
∀ w : Fin n → ℝ, normQ (mix q lam) a w = (1 / |lam|) * normQ q a w := by sorry
end RiskUncSets.InnerApprox
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.