Proposition 5.2(iii), p. 18 — prox_{(Ch)/β} ⊇ C X_β
ProvedNonconvexDRS.ADMM.prop_5_2_iiiimage-functionnonconvex-optimizationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1proximal-mapping
Let , , and . For every ,
Minimizing over and then mapping by thus produces a proximal point of the image function; this is the mechanism behind both prox-inclusions of the ADMM–DRS equivalence.
Formalization Note The proximal mapping is the set proxSet with stepsize . The global hypothesis for all is dropped, which makes the statement stronger.
Preamble
import Mathlib import Definitions.Def_NonconvexSplitting_Shared_LimitingSubdiff import Definitions.Def_NonconvexDRS_ADMM_Setting open NonconvexSplitting.Shared open Filter Topology open scoped InnerProductSpace
Formal statement
namespace NonconvexDRS.ADMM
/-- Proposition 5.2(iii), p. 18: `prox_{(Ch)/β}(s) ⊇ C X_β(s)`, i.e. `C x_β` is a proximal point of
`(Ch)` with stepsize `1/β` at `s`, for every `s` and every `x_β ∈ X_β(s)`. -/
theorem prop_5_2_iii {n p : ℕ} (h : EuclideanSpace ℝ (Fin n) → EReal)
(C : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin p))
(β : ℝ) (hβ : 0 < β) (s : EuclideanSpace ℝ (Fin p))
(xβ : EuclideanSpace ℝ (Fin n)) (hxβ : xβ ∈ Xbeta h C β s) :
C xβ ∈ NonconvexDRS.Tight.proxSet (imageFn C h) (1 / β) s := by sorry
end NonconvexDRS.ADMM
Source
Themelis & Patrinos, Douglas-Rachford splitting and ADMM for nonconvex optimization: tight convergence results, arXiv:1709.05747v4, pp. 17–18, Proposition 5.2(iii)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.