(5.3b), proof of Theorem 5.5, p. 19 — prox_{γφ₂} ⊇ b − B argmin{g + (1/(2γ))‖B · + s − b‖²} for φ₂ = (Bg)(b − ·)
ProvedNonconvexDRS.ADMM.eq_5_3badmmimage-functionp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1proximal-mapping
Let , , , and , and set . If minimizes , then
This is the inclusion that turns the ADMM -update into the second proximal step of DRS.
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
/-- (5.3b), proof of Theorem 5.5, p. 19: with `φ₂ = (Bg)(b - ·)`,
`prox_{γφ₂}(s) ⊇ b - B argmin { g + (1/(2γ))‖B · + s - b‖² }`. -/
theorem eq_5_3b {n p : ℕ} (g : EuclideanSpace ℝ (Fin n) → EReal)
(B : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin p)) (b : EuclideanSpace ℝ (Fin p))
(γ : ℝ) (hγ : 0 < γ) (s : EuclideanSpace ℝ (Fin p)) (z : EuclideanSpace ℝ (Fin n))
(hz : ∀ z', g z + ((‖B z + s - b‖ ^ 2 / (2 * γ) : ℝ) : EReal) ≤
g z' + ((‖B z' + s - b‖ ^ 2 / (2 * γ) : ℝ) : EReal)) :
b - B z ∈ NonconvexDRS.Tight.proxSet (fun t => imageFn B g (b - t)) γ s := by sorry
end NonconvexDRS.ADMM
Source
Themelis & Patrinos, Douglas-Rachford splitting and ADMM for nonconvex optimization: tight convergence results, arXiv:1709.05747v4, p. 19, (5.3b) in the proof of Theorem 5.5
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.