Lemma 7 — second moments of Bernoulli likelihood ratios: E₀[dp_q/dp₀]² = 1 + ρ_n q²/(1 − ρ_n)
ProvedCPDNetworks.Lower.lemma_7hypothesis-testinglikelihood-ratiop2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let and . Write for the Bernoulli law with parameter and for the Bernoulli law with parameter , with densities , with respect to counting measure on . For a Bernoulli variable ,
The two identities compute the second moment of the likelihood ratio entry by entry in Step 3 of the proof of Theorem 2.
Formalization Note The expectation under is the two-term sum over (Bool), and the densities are the Bernoulli mass functions; on both points because .
Preamble
import Mathlib import Definitions.Def_HighDimProb_RandomMatrices_matrixOpNorm import Definitions.Def_CPDNetworks_Lower_Setting
Formal statement
namespace CPDNetworks.Lower
/-- **Lemma 7** (Enikeeva–Klopp, p. 34). Let `ρ_n ∈ (0, 1/2]` and `q, q₁, q₂ ∈ [−1, 1]`, and let
`p₀`, `p_q` be the Bernoulli laws with parameters `ρ_n` and `ρ_n(1 + q)`. For `X ∼ p₀`,
`E₀[(dp_q/dp₀)(X)]² = 1 + ρ_n/(1 − ρ_n) q²` and
`E₀[(dp_{q₁}/dp₀)(dp_{q₂}/dp₀)](X) = 1 + ρ_n/(1 − ρ_n) q₁q₂`. The densities w.r.t. counting measure
on `{0, 1}` (`Bool`) are the Bernoulli mass functions `bern`. -/
theorem lemma_7 (ρ q q₁ q₂ : ℝ) (hρ : 0 < ρ) (hρ2 : ρ ≤ 1 / 2)
(hq : -1 ≤ q ∧ q ≤ 1) (hq₁ : -1 ≤ q₁ ∧ q₁ ≤ 1) (hq₂ : -1 ≤ q₂ ∧ q₂ ≤ 1) :
∑ x : Bool, bern ρ x * (bern (ρ * (1 + q)) x / bern ρ x) ^ 2 = 1 + ρ / (1 - ρ) * q ^ 2 ∧
∑ x : Bool, bern ρ x *
(bern (ρ * (1 + q₁)) x / bern ρ x * (bern (ρ * (1 + q₂)) x / bern ρ x)) =
1 + ρ / (1 - ρ) * (q₁ * q₂) := by sorry
end CPDNetworks.Lower
Source
Enikeeva and Klopp, Change-point detection in dynamic networks with missing links, arXiv:2106.14470v2, p. 34, Lemma 7
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.