§3.3, p. 10 — the partial derivative ∂F/∂w_j of the multilinear extension is E[h(Z)|Z_j = 1] − E[h(Z)|Z_j = 0]
ProvedDiffNeighbors.Bias.first_partialdifferences-in-neighborsmultilinear-extensionp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1partial-derivative
Let and let
be its multilinear extension, which for is with independent . For every coordinate and every ,
where is the multilinear extension, at , of .
In the proof of Theorem 1 this is applied to (and ): it identifies the gradient term of the first-order Taylor approximation of around .
Formalization Note The partial derivative is stated as HasDerivAt of at .
Differences from the source:
- generalization: F¹ of node i, a function of the |N_i| neighbour treatments → the multilinear extension of any h : {0,1}^ι → ℝ, at any w ∈ ℝ^ι, for any coordinate j; reason: specializes with h = condOutcome f i true, which under Assumption 1 depends only on z_{N_i}, so its extension is the paper's F¹
- encoding: ∂F/∂w_j and E[f¹(Z)|Z_j = b] → HasDerivAt of t ↦ mlExt h (update w j t) at w j, with value mlExt (h ∘ update · j true) w − mlExt (h ∘ update · j false) w; reason: the partial derivative of F in its j-th coordinate; conditioning on Z_j = b sets z_j := b by independence
- encoding: units [N] = {1,…,N}, treatment vectors z ∈ {0,1}^N, an undirected graph without self-loops, d = max_i |N_i| → a finite type ι with N = card ι, z : ι → Bool with 1/0 read through ind, G : SimpleGraph ι, N_i = G.neighborFinset i, d = G.maxDegree; reason: a simple graph is exactly an undirected graph without self-loops; same instances
Preamble
import Mathlib import Definitions.Def_DiffNeighbors_Bias_Model
Formal statement
namespace DiffNeighbors.Bias
open Finset
/-- **First partial derivative of the multilinear extension**, §3.3, p. 10:
`∂F/∂w_j (w) = E_w[h(Z) | Z_j = 1] − E_w[h(Z) | Z_j = 0]`, where `E_w` is the expectation under
independent Bernoulli(`w_l`) coordinates. -/
theorem first_partial {ι : Type*} [Fintype ι] [DecidableEq ι] (h : (ι → Bool) → ℝ) (w : ι → ℝ)
(j : ι) :
HasDerivAt (fun t => mlExt h (Function.update w j t))
(mlExt (fun z => h (Function.update z j true)) w
- mlExt (fun z => h (Function.update z j false)) w) (w j) := by sorry
end DiffNeighbors.Bias
Source
Peng, Ye, Zheng, Differences-in-Neighbors for Network Interference in Experiments, arXiv:2503.02271v1, p. 10, §3.3 (proof of Theorem 1), display '∂F¹/∂w_j = … = E[f¹(Z)|Z_j = 1] − E[f¹(Z)|Z_j = 0]'
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.