Effective-size facts (proofs of Observation 2 and Lemma 6, p. 12) — 0 ≤ β_k(Y) ≤ 1 for Y ∈ [0,1], and β_k(Y) ≥ E[Y]
ProvedStochLoadBal.Makespan.effSize_factseffective-sizejensen-inequalityp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be an integer.
- If is a random variable with values in , then
- If is integrable and , then
Part 1 applied to truncated parts gives in Observation 2; part 2 (Jensen's inequality) gives for the truncated load of each machine in the proof of Lemma 6.
Formalization Note The page calls Jensen's inequality "for concave functions"; the inequality used is Theorem 5 for the convex function . For , and part 2 is an equality.
Preamble
import Mathlib import Definitions.Def_StochLoadBal_Makespan_Model
Formal statement
namespace StochLoadBal.Makespan
open MeasureTheory ProbabilityTheory
/-- The two effective-size facts used in the proofs of Observation 2 and Lemma 6 (p. 12), for
`k ≥ 1`: (a) a random variable `Y` with values in `[0, 1]` has `0 ≤ β_k(Y) ≤ 1`;
(b) every integrable `Y` with `E[e^{(log k) Y}] < ∞` has `E[Y] ≤ β_k(Y)` (Jensen). -/
theorem effSize_facts {Ω : Type} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
(k : ℕ) (hk : 1 ≤ k) :
(∀ Y : Ω → ℝ, Measurable Y → (∀ ω, 0 ≤ Y ω ∧ Y ω ≤ 1) →
0 ≤ effSize P k Y ∧ effSize P k Y ≤ 1) ∧
(∀ Y : Ω → ℝ, Measurable Y → Integrable Y P →
Integrable (fun ω => Real.exp (Real.log k * Y ω)) P →
∫ ω, Y ω ∂P ≤ effSize P k Y) := by sorry
end StochLoadBal.Makespan
Source
Gupta, Kumar, Nagarajan & Shen, Stochastic Load Balancing on Unrelated Machines, arXiv:1904.07271v1, p. 12, proof of Observation 2 and proof of Lemma 6 (unnumbered)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.