Theorem 8, p. 17 — accelerated rate for the strongly convex ODE
ProvedNesterovODE.StrongCvx.theorem_8convergence-ratenesterov-odep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1strong-convexity
Fix real numbers and with . There exists a constant depending only on and such that, in every dimension, for every , , , minimizer , initial point , solution of (17), and ,
This gives the paper's improved polynomial rate when the damping and strong convexity are combined. The constant is uniform over functions, dimensions, strong-convexity parameters, smoothness parameters, initial data, solutions, and time.
Formalization Note The condition is implicit in the paper's denominator and threshold time. The statement ranges over every solution in the ODE regularity class, avoiding dependence on an unproved uniqueness assertion for general .
Preamble
import Mathlib import Definitions.Def_NesterovODE_StrongCvx_Setting
Formal statement
namespace NesterovODE.StrongCvx
/-- Theorem 8: a constant uniform in dimension, smoothness, strong convexity, and trajectory. -/
theorem theorem_8 (r α : ℝ) (ha : 2 ≤ α) (hr : α ≤ 2 * r / 3) :
∃ C : ℝ, ∀ (n : ℕ) (f : NesterovODE.WellPosed.E n → ℝ) (mu : ℝ) (L : NNReal),
0 < mu → 0 < L → InSMuL mu L f →
∀ (x₀ xstar : NesterovODE.WellPosed.E n), (∀ y, f xstar ≤ f y) →
∀ (X V : ℝ → NesterovODE.WellPosed.E n), IsSolution f r x₀ X V →
∀ t : ℝ, 0 < t →
f (X t) - f xstar ≤
C * ‖x₀ - xstar‖ ^ 2 / (mu ^ ((α - 2) / 2) * t ^ α) := by sorry
end NesterovODE.StrongCvx
Source
Su, Boyd, Candès, A Differential Equation for Modeling Nesterov's Accelerated Gradient Method, arXiv:1503.01243v2, p. 17, Theorem 8
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.