(112), p. 29 — (z − prox_{ηg}(z))/η ∈ ∂g(prox_{ηg}(z))
ProvedNAGFlow.APGM.eq_112convex-analysisp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1proximal-operatorsubdifferential
Let be a real Hilbert space, proper, closed and convex, and . For every , if , that is, minimises over , then
where is the subdifferential (105): iff for all .
This is the optimality condition of the proximal subproblem; it is what turns the proximal-gradient step into a gradient-like step in §7.2.
Formalization Note. is encoded by its domain and its real values on ; is part of the prox predicate, and the subgradient inequality is required for (it is automatic for , where ).
Preamble
import Mathlib import Definitions.Def_NAGFlow_APGM_Setting
Formal statement
namespace NAGFlow.APGM
/-- The proximal optimality condition (112) (Luo & Chen, arXiv:1909.03145v4, §7.2.1, p. 29). Let
`g : V → ℝ ∪ {+∞}` be proper, closed and convex (encoded by `(g, D)`, `D = dom g`), `η > 0` and
`z ∈ V`. If `p = prox_{ηg}(z)`, then `(z − p)/η ∈ ∂g(p)`. -/
theorem eq_112 {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V]
(g : V → ℝ) (D : Set V) (hg : IsProperClosedConvex g D) (η : ℝ) (hη : 0 < η)
(z p : V) (hp : IsProx g D η z p) :
(1 / η) • (z - p) ∈ subdiff g D p := by sorry
end NAGFlow.APGM
Source
Luo & Chen, arXiv:1909.03145v4, §7.2.1, (112), p. 29; (8), p. 2; (105), p. 26
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.