Lemma 3 — the Bregman monotone conjugate is finite and differentiable
ProvedBregmanPPA.IneqMult.lemma3bregman-functionmonotone-conjugatep2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1
If is a Bregman function whose zone and gradient image each contain the strictly positive orthant, then its monotone conjugate is closed, proper, convex, everywhere finite, and differentiable:
This is the well-posedness result for the objective and gradient appearing in the multiplier recursion.
Formalization Note The conjugate itself is extended-real valued, with explicit exclusion of both infinite values; differentiability is then stated for its real-valued version.
Preamble
import Mathlib import Definitions.Def_BregmanPPA_IneqMult_Program open Filter Topology InertialFB.IFB
Formal statement
namespace BregmanPPA.IneqMult
/-- Lemma 3, p. 216: the monotone conjugate is closed, proper, convex, finite, and smooth. -/
theorem lemma3 {m : ℕ} (S : Set (E m)) (h : E m → ℝ)
(hh : BregmanPPA.Convergence.IsBregmanFunction S h)
(hS : posOrthant m ⊆ S)
(him : posOrthant m ⊆ gradient h '' S) :
LowerSemicontinuous (monoConj h) ∧
IsProperFn (monoConj h) ∧
IsConvexFn (monoConj h) ∧
(∀ z : E m, monoConj h z ≠ ⊤ ∧ monoConj h z ≠ ⊥) ∧
Differentiable ℝ (fun z : E m => (monoConj h z).toReal) := by sorry
end BregmanPPA.IneqMult
Source
Eckstein, Nonlinear proximal point algorithms using Bregman functions, with applications to convex programming, Math. Oper. Res. 18(1) (1993), p. 216, Lemma 3, https://doi.org/10.1287/moor.18.1.202
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.