Lemma 4.1 — Kim's coefficients (25) with , , are feasible for (D)
ProvedAccelPPM.FPR.lemma4_1_dual_feasiblep2o-batch-p200bp2o-gran-per-chapterp2o-plan-paperp2o-v1performance-estimationproximal-point-methodsemidefinite-programming
Let . Take Kim's step coefficients
and the dual variables
Then is a feasible point of the dual problem (D) for : all dual variables are nonnegative and
Consequently is a feasible point of the joint problem (HD) , and .
This dual certificate is the source of the accelerated rate of the fixed-point residual.
Formalization Note A feasible point of (HD) is a pair (, dual variables) feasible for (D), so the single statement IsDualFeasible N kimCoeff a b c covers both halves of "a feasible point of (D) and (HD)". is the function fun i => 2 * ((i:ℝ) - 1) * i / N^2, with cast to before subtracting; only is read.
Preamble
import Mathlib import Definitions.Def_AccelPPM_FPR_kimCoeff import Definitions.Def_AccelPPM_FPR_boundBD
Formal statement
namespace AccelPPM.FPR
/-- **Lemma 4.1** (Kim, arXiv:1905.05149v4, p. 8). For every `N ≥ 1`, the step coefficients
`h = (25)` (`kimCoeff`) together with the dual variables (26),
`a_i = 2(i − 1)i/N²` (`i = 2, …, N`), `b_N = 2/N`, `c = 1/N²`,
form a feasible point of the dual problem (D), hence (with `h`) of (HD). -/
theorem lemma4_1_dual_feasible (N : ℕ) (hN : 1 ≤ N) :
IsDualFeasible N kimCoeff (fun i => 2 * ((i : ℝ) - 1) * (i : ℝ) / (N : ℝ) ^ 2)
(2 / (N : ℝ)) (1 / (N : ℝ) ^ 2) := by sorry
end AccelPPM.FPR
Source
Kim, Accelerated proximal point method for maximally monotone operators, arXiv:1905.05149v4, p. 8, Lemma 4.1 (Eqs. (25), (26))
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.