Theorem 3.1(i), p. 483 — x ∈ dom h solves min(g − h) if and only if ∂_εh(x) ⊂ ∂_εg(x) for every ε > 0
ProvedTaoAnDCA.GlobalOpt.theorem_3_1_iLet be proper, lower semicontinuous and convex, and assume the inclusions
where are the conjugate functions. Consider the d.c. program
with the convention , and let be its set of solutions. For let denote the -subdifferential of a convex function at .
Then for every ,
This is the global optimality condition for d.c. programs, first established by Hiriart-Urruty; the paper gives a short proof through the d.c. duality. It characterizes global minimizers of a nonconvex objective by inclusions between convex objects.
Formalization Note The paper states (i) for every ; its proof argues for . The statement is restricted to because for the set is empty, so the right-hand side holds vacuously while , and the unrestricted statement would be false. The restriction loses nothing: every solution lies in , and for both sides are false. The -subdifferential is empty outside the domain, as in the paper's definition; the difference uses the convention (dcSub). The standing assumptions ( and the inclusions (3), assumed throughout the paper) are hypotheses; finiteness of is not assumed.
import Mathlib import Definitions.Def_TaoAnDCA_GlobalOpt_Setting open TaoAnDCA.GlobalOpt
namespace TaoAnDCA.GlobalOpt
theorem theorem_3_1_i {n : ℕ} (g h : EuclideanSpace ℝ (Fin n) → EReal)
(hgh : DCStanding g h) (x : EuclideanSpace ℝ (Fin n)) (hx : x ∈ effDom h) :
x ∈ primalSol g h ↔ ∀ ε : ℝ, 0 < ε → epsSubdiff h ε x ⊆ epsSubdiff g ε x := by sorry
end TaoAnDCA.GlobalOpt
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.