p. 22 — 0 < (1 − √(1/κ))/(1 + √(1/κ)) < λ for λ ∈ (0, 1)
ProvedAccelVI.MVC.rate_boundsconvergence-ratemomentump2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let and . Then
So the rate of Momentum Value Computation is strictly better than the rate of plain Value Computation.
Preamble
import Mathlib import Definitions.Def_mm_basic import Definitions.Def_AccelVI_MVC_Setting
Formal statement
namespace AccelVI.MVC
/-- p. 22: 0 < (1 − √(1/κ))/(1 + √(1/κ)) < λ for λ ∈ (0, 1). -/
theorem rate_bounds (lam : ℝ) (h0 : 0 < lam) (h1 : lam < 1) :
0 < rateMVC lam ∧ rateMVC lam < lam := by sorry
end AccelVI.MVC
Source
Goyal & Grand-Clément, A First-Order Approach to Accelerated Value Iteration, arXiv:1905.09963v7, p. 22, sentence after the proof of Lemma B.2 (and Theorem 5.5, p. 10, last sentence)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.