(85), App. E, p. 49 — the κ⁽²⁾ inverse bound T⁻¹(y) ≤ y^{log l/log u} (y ≥ 1), ≤ y otherwise
ProvedSelfStructIS.LogEff.eq_85importance-samplinginverse-map-boundsp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be the self-structuring transformation with exponent and , so that with . For every and every with , componentwise,
This is the counterpart of Lemma A.4, used in Appendix E to rerun the efficiency proof for the model-free exponent.
Formalization Note The paper states (85) inside the proof for , where is the setting of Table 1's row (); that range is assumed. The bound is stated for every preimage of .
Preamble
import Mathlib import Definitions.Def_HeavyTailNV_Tail_Model import Definitions.Def_SelfStructIS_LogEff_Setting import Definitions.Def_SelfStructIS_LogEff_Sampler open MeasureTheory Filter Topology
Formal statement
namespace SelfStructIS.LogEff
theorem eq_85 (d : ℕ) (ρ u l : ℝ) (hl : 1 < l) (hlu : l ≤ u)
(y : Fin d → ℝ) (hy : ∀ i, 0 ≤ y i) (x : Fin d → ℝ) (hx : Tmap .two ρ u l x = y) :
∀ i, x i ≤ if 1 ≤ y i then y i ^ (Real.log l / Real.log u) else y i := by sorry
end SelfStructIS.LogEff
Source
Deo, Murthy, Achieving Efficiency in Black-Box Simulation of Distribution Tails with Self-Structuring Importance Samplers, arXiv:2102.07060v3, p. 49, App. E, (85)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.