Documentation

MIPRE.Background.Repetition.CommutingRepetition.Prerounding.Core

theorem CommutingRepetition.greedy_core {Ω : Type u_1} {ι : Type u_2} [Fintype Ω] [Fintype ι] [DecidableEq ι] (law : FiniteEventLaw Ω) (wins : ιΩBool) {δ : } (hδ0 : 0 < δ) (hδ1 : δ < 1) ( : 0 < law.eventMass (FiniteEventLaw.winEvent wins Finset.univ)) (hn : Real.log (1 / law.eventMass (FiniteEventLaw.winEvent wins Finset.univ)) / δ < (Fintype.card ι)) :

Greedy core (node 1.2.1; 05_prerounding.tex, lem greedy-core). With θ = P(W_[n]) > 0, 0 < δ < 1, and log(1/θ)/δ < n, there is a proper subset D ⊊ [n] with |D| ≤ log(1/θ)/δ, P(W_D) ≥ θ, and average conditional success (1/(n−|D|)) ∑_{i∉D} P(W_i | W_D) ≥ 1 − δ. All conditionals are defined because W_[n] ⊆ W_D gives P(W_D) ≥ θ > 0.

theorem CommutingRepetition.core_parameters {ε γ θ p : } {n Dcard : } (hε0 : 0 < ε) (hε1 : ε 1) ( : γ ε / 8) (hℓ : 0 ) (hn : 1 n) ( : Real.exp (-(γ * n)) < θ) (hθp : θ p) (hp1 : p 1) (hD : Dcard Real.log (1 / θ) / (ε / 4)) :
Dcard < 4 * γ * n / ε 4 * γ * n / ε n / 2 n / 2 < n - Dcard Real.log (1 / p) < γ * n (Real.log (1 / p) + Dcard * ) / (n - Dcard) < 2 * γ * (1 + 4 * / ε) 2 * γ * (1 + 4 * / ε) 8 * γ * ((ε + ) / ε)

Core parameters and bounds (node 1.2.2; 05_prerounding.tex, eqs p-q-m, eta-definition, core-basic-bounds, eta-gamma-bound). With δ = ε/4, m = n − |D|, p = P(W_D) ∈ [θ, 1], t₀ = log(1/p), s₀ = |D|·ℓ, η = (t₀ + s₀)/m, the pre-rounding hypotheses θ > e^{−γn}, 0 < ε ≤ 1, γ ≤ ε/8 give |D| < 4γn/ε ≤ n/2 (so m > n/2), t₀ < γn, and η < 2γ(1 + 4ℓ/ε) ≤ 8γ(ε + ℓ)/ε. The q ≥ 1 − ε/4 part of eq core-basic-bounds is not restated here: it is verbatim the third conjunct of greedy_core at δ = ε/4 (node 1.2.2 lists it as a "put" item) and is consumed directly from there.