p. 4 (after (2.3)) — (2.3) implies g′(x̄, d) ≥ ε‖d‖ for some ε > 0 and all d ∈ T_Θ(x̄)
ProvedSAARate.Sharp.dirDeriv_ge_eps_normconvex-analysisdirectional-derivativep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1sharp-minimumtangent-cone
Let be convex, let be convex, let , and suppose that
Then there is such that
Here is the directional derivative of at in direction and is the tangent cone to at . Positivity on nonzero tangent directions is thereby upgraded to a uniform linear lower bound; on the unit sphere it gives , the margin the proof of Theorem 2.1 needs before passing to the sample average.
Formalization Note. Stated for an arbitrary finite convex (the page states it for ). The tangent cone is Mathlib's posTangentConeAt Θ x̄, the closure of for convex . At both sides are .
Preamble
import Mathlib import Definitions.Def_SAARate_Sharp_Setting
Formal statement
namespace SAARate.Sharp
open MeasureTheory ProbabilityTheory Filter Topology
theorem dirDeriv_ge_eps_norm {m : ℕ} (g : E m → ℝ) (hg : ConvexOn ℝ Set.univ g)
(Θ : Set (E m)) (hΘ : Convex ℝ Θ) (xbar : E m) (hx : xbar ∈ Θ)
(h23 : ∀ d ∈ posTangentConeAt Θ xbar, d ≠ 0 → 0 < dirDeriv g xbar d) :
∃ ε > 0, ∀ d ∈ posTangentConeAt Θ xbar, ε * ‖d‖ ≤ dirDeriv g xbar d := by sorry
end SAARate.Sharp
Source
Shapiro & Homem-de-Mello, On Rate of Convergence of Optimal Solutions of Monte Carlo Approximations of Stochastic Programs, preprint (SPEPS copy, edoc.hu-berlin.de), p. 4, sentence after (2.3): "it follows from (2.3) that f′(x̄, d) ≥ ε‖d‖ for some ε > 0 and all d ∈ T_Θ(x̄)"
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.