Lemma 2 — effective-size Chernoff bound: Σ_i β_k(Y_i) ≤ b implies Pr[Σ_i Y_i ≥ c] ≤ (1/k)^(c−b)
ProvedStochLoadBal.Makespan.lemma_2chernoff-boundconcentrationeffective-sizep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be an integer and let be independent real random variables with . If
then for every real ,
This is the upper-bound property of effective sizes: a total effective size at most makes a load above have probability at most . It gives the per-machine tail bound of Observation 3.
Formalization Note The lemma is stated for integers , where Definition 1's formula applies; for the right side is and the statement is empty. The exponential-moment hypothesis makes the logarithm of a finite moment generating function rather than a junk value.
Preamble
import Mathlib import Definitions.Def_StochLoadBal_Makespan_Model
Formal statement
namespace StochLoadBal.Makespan
open MeasureTheory ProbabilityTheory
/-- Lemma 2 (Upper Bound, p. 6): for an integer `k ≥ 2` and independent random variables
`Y_1, …, Y_n` with finite exponential moments `E[e^{(log k) Y_i}]`, if `Σ_i β_k(Y_i) ≤ b` then
`Pr[Σ_i Y_i ≥ c] ≤ (1/k)^{c - b}` for every real `c`. -/
theorem lemma_2 {Ω : Type} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
(n : ℕ) (Y : Fin n → Ω → ℝ) (hmeas : ∀ i, Measurable (Y i)) (hind : iIndepFun Y P)
(k : ℕ) (hk : 2 ≤ k)
(hint : ∀ i, Integrable (fun ω => Real.exp (Real.log k * Y i ω)) P)
(b c : ℝ) (hb : ∑ i, effSize P k (Y i) ≤ b) :
(P {ω | c ≤ ∑ i, Y i ω}).toReal ≤ (1 / (k : ℝ)) ^ (c - b) := by sorry
end StochLoadBal.Makespan
Source
Gupta, Kumar, Nagarajan & Shen, Stochastic Load Balancing on Unrelated Machines, arXiv:1904.07271v1, p. 6, Lemma 2
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.