Eq. (3.22), p. 292 — Φ*_{s+1} + (α/2)‖x_s − v_{s+1}‖² = (1 − 1/√κ)Φ*_s + (α/2)(1 − 1/√κ)‖x_s − v_s‖² + f(x_s)/√κ
OpenConvexOptAlg.NesterovStrong.eq_3_22accelerated-gradientconvex-optimizationestimate-sequencep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-bookp2o-v1
Let , , , let , (in the role of ) and the points be arbitrary, and let , and be as in (3.17) and (3.21). Then for every ,
This identity, obtained by evaluating at , gives a closed recursion for the minimum values .
Formalization Note Like eq_3_21_form, the identity is algebraic and is stated for any sequence of points and any map .
Preamble
import Mathlib import Definitions.Def_OnlineConvexOpt_ConvexBasics_StronglyConvexOn import Definitions.Def_ConvexOptAlg_NesterovStrong_Defs open scoped InnerProductSpace
Formal statement
namespace ConvexOptAlg.NesterovStrong
/-- Bubeck, proof of Theorem 3.18, Eq. (3.22), p. 292: for any `α > 0`, any `f`, gradient map
`g`, `β`, any sequence of points `x_s` and every `s ≥ 1`,
`Φ∗_{s+1} + (α/2)‖x_s − v_{s+1}‖² = (1 − 1/√κ)Φ∗_s + (α/2)(1 − 1/√κ)‖x_s − v_s‖² + (1/√κ) f(x_s)`. -/
theorem eq_3_22 {n : ℕ} (f : EuclideanSpace ℝ (Fin n) → ℝ)
(g : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin n)) (α β : ℝ)
(hα : 0 < α) (x : ℕ → EuclideanSpace ℝ (Fin n))
(s : ℕ) (hs : 1 ≤ s) :
PhiStar f g α β x (s + 1) + α / 2 * ‖x s - v g α β x (s + 1)‖ ^ 2 =
(1 - 1 / Real.sqrt (kappa α β)) * PhiStar f g α β x s +
α / 2 * (1 - 1 / Real.sqrt (kappa α β)) * ‖x s - v g α β x s‖ ^ 2 +
1 / Real.sqrt (kappa α β) * f (x s) := by sorry
end ConvexOptAlg.NesterovStrong
Source
Bubeck, arXiv:1405.4980v2, proof of Theorem 3.18, Eq. (3.22), p. 292