(3.6), p. 15 — E(t) ≤ (1+√(μs))(f(X)−f(x⋆)) + ‖Ẋ‖² + 3μ‖X−x⋆‖² + (3s/4)‖∇f(X)‖²
ProvedHighResODE.NAGSCODE.eq_3_6high-resolution-odelyapunov-functionnesterovp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let , , and , and let be any curves in . The Lyapunov function (2.4),
satisfies, at every time (with , ),
It is the upper bound on that, combined with the dissipation bound (3.5), turns the derivative estimate into the rate of Lemma 3.1.
Formalization Note. The inequality is pointwise in and uses only ; it is therefore stated for arbitrary , arbitrary curves and , which contains the page's setting (where and solves (1.11)) as a special case. is the velocity curve .
Preamble
import Mathlib import Definitions.Def_HighResODE_NAGSCODE_Setting open scoped InnerProductSpace
Formal statement
namespace HighResODE.NAGSCODE
theorem eq_3_6 {n : ℕ} (f : EuclideanSpace ℝ (Fin n) → ℝ) (μ s : ℝ) (hμ : 0 ≤ μ) (hs : 0 ≤ s)
(xs : EuclideanSpace ℝ (Fin n)) (X V : ℝ → EuclideanSpace ℝ (Fin n)) (t : ℝ) :
lyap f μ s xs X V t ≤ (1 + Real.sqrt (μ * s)) * (f (X t) - f xs) + ‖V t‖ ^ 2
+ 3 * μ * ‖X t - xs‖ ^ 2 + 3 * s / 4 * ‖gradient f (X t)‖ ^ 2 := by sorry
end HighResODE.NAGSCODE
Source
Shi, Du, Jordan & Su, Understanding the Acceleration Phenomenon via High-Resolution Differential Equations, arXiv:1810.08907v3, p. 15, proof of Lemma 3.1, (3.6)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.