Proof of Theorem 3.7, p.37 — tilting by with preserves epi-convergence
ProvedRockafellianRelax.tilt_epiConvergesoptimizationp2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1stochastic-optimizationvariational-analysis
Let with epi-converging to , and let in . Then the tilted functions
Epi-convergence is (3.8)–(3.9) of p.9. This turns the hypothesis of Theorem 3.7 into epi-convergence of the objectives of the approximating problems (3.7) to that of the relaxed problem (3.2).
Differences from the source:
- generalization: f the exact Rockafellian to which the f^ν epi-converge (limit-exact hypothesis of Theorem 3.7) → any f : ℝᵐ × ℝⁿ → R̄ to which the f^ν epi-converge; reason: the step uses only (3.8) and (3.9); the paper's f is one such f
- encoding: ℝᵐ × ℝⁿ, values in R̄ = [−∞, ∞], ⟨y, u⟩ → (Fin m → ℝ) × (Fin n → ℝ) with the product topology, EReal, y ⬝ᵥ u; reason: Lean convention; same instances (the product topology has the same convergent sequences as the Euclidean norm)
- encoding: ν ∈ ℕ = {1, 2, …} → ν ∈ Lean's ℕ from 0; reason: Lean convention; a shift of the index changes no limit
Preamble
import Mathlib import Definitions.Def_RockafellianRelax_tilt import Definitions.Def_PerspEpi_SeqEpiConv open Filter Topology
Formal statement
namespace RockafellianRelax
/-- **Tilting preserves epi-convergence.** Royset, Chen & Eckstrand, arXiv:2204.04762v4, Proof of
Theorem 3.7, p.37: "Consider the functions given by f̃ν(u, x) = fν(u, x) − ⟨yν, u⟩ and
f̃(u, x) = f(u, x) − ⟨ȳ, u⟩. It follows directly from (3.8) and (3.9) that the epi-convergence of fν
to f guarantees that f̃ν epi-converges to f̃."
If `fν` epi-converges to `f` on `ℝᵐ × ℝⁿ` and `yν → ȳ`, then `fν − ⟨yν, ·₁⟩` epi-converges to
`f − ⟨ȳ, ·₁⟩`.
**Formalization Note.** Epi-convergence is the sequential (3.8)–(3.9) (`PerspEpi.SeqEpiConv`) on
the product topology. In the paper `f` is the exact Rockafellian of the limit-exact hypothesis; the
statement holds for every `f`. -/
theorem tilt_epiConverges {m n : ℕ} (F : ℕ → (Fin m → ℝ) × (Fin n → ℝ) → EReal)
(f : (Fin m → ℝ) × (Fin n → ℝ) → EReal) (ybar : Fin m → ℝ) (y : ℕ → Fin m → ℝ)
(hy : Tendsto y atTop (𝓝 ybar)) (hF : PerspEpi.SeqEpiConv F f) :
PerspEpi.SeqEpiConv (fun ν => tilt (F ν) (y ν)) (tilt f ybar) := by sorry
end RockafellianRelax
Source
Royset, Chen & Eckstrand, Rockafellian Relaxation and Stochastic Optimization under Perturbations, arXiv:2204.04762v4, Proof of Theorem 3.7, p.37
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.