Lemma 5.2, p. 24 — for monotone -Lipschitz minorants on a compact set, iff
ProvedConvexSDDP.Det.lemma_5_2convergenceconvex-analysislipschitzp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be compact, convex, , and a sequence of convex -Lipschitz functions . Fix an integer and suppose
(the left inequality for , the right one for all ). Then for every sequence in ,
In the proof of Theorem 2.1 it is applied with , and to pass from the current approximation to the previous one (p. 12).
Formalization Note "For any integer " is a fixed . In Lean k - κ is truncated subtraction, which differs from the page only for ; the monotonicity hypothesis is required only for and no limit sees finitely many indices. The functions are real valued, as in the lemma; convexity is on all of and the Lipschitz property is global (LipschitzWith α).
Preamble
import Mathlib open Filter Topology
Formal statement
namespace ConvexSDDP.Det
theorem lemma_5_2 {d : ℕ} (X : Set (EuclideanSpace ℝ (Fin d))) (hX : IsCompact X)
(f : EuclideanSpace ℝ (Fin d) → ℝ) (hf : ConvexOn ℝ Set.univ f)
(fk : ℕ → EuclideanSpace ℝ (Fin d) → ℝ) (α : NNReal)
(hfk : ∀ k, ConvexOn ℝ Set.univ (fk k) ∧ LipschitzWith α (fk k))
(κ : ℕ) (hmono : ∀ k, κ ≤ k → ∀ y ∈ X, fk (k - κ) y ≤ fk k y)
(hle : ∀ k, ∀ y ∈ X, fk k y ≤ f y)
(xs : ℕ → EuclideanSpace ℝ (Fin d)) (hxs : ∀ k, xs k ∈ X) :
Tendsto (fun k => f (xs k) - fk k (xs k)) atTop (𝓝 0) ↔
Tendsto (fun k => f (xs k) - fk (k - κ) (xs k)) atTop (𝓝 0) := by sorry
end ConvexSDDP.Det
Source
Girardeau, Leclère & Philpott, On the Convergence of Decomposition Methods for Multistage Stochastic Convex Programs, author's version hal-01208295v1, p. 24, Lemma 5.2
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.