Proof of Theorem 1, p. 32 — Fubini step: ∫∫ g(z) w(z − s) h(z) dz dP_Y(s) = ∫ g(z) (f_Y ∗ w)(z) h(z) dz
ProvedMultivarScores.L2Rule.integral_cross_termconvolutionfubinip2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1scoring-rules
Let be a probability distribution on and let be measurable. If is integrable with respect to , then
where .
This is the step of the proof of Theorem 1 that "holds by Fubini's theorem and the definition of convolution": it turns the expectation over the realization of the cross term of the expanded score into an integral against the smoothed density . With it is the passage from the second to the third line of the divergence computation.
Formalization Note The product integrability is the form Assumption 1(c) takes for this cross term; it is what makes Fubini's theorem applicable.
Preamble
import Mathlib import Definitions.Def_MultivarScores_L2Rule_Setting
Formal statement
namespace MultivarScores.L2Rule
open MeasureTheory
/-- Fubini step in the proof of Theorem 1, p. 32: for a distribution `P_Y`,
`∫ (∫ g(z) w(z − s) h(z) dz) dP_Y(s) = ∫ g(z) (f_Y ∗ w)(z) h(z) dz`. -/
theorem integral_cross_term {d : ℕ} (PY : Measure (E d)) [IsProbabilityMeasure PY]
(g w h : E d → ℝ) (hg : Measurable g) (hw : Measurable w) (hh : Measurable h)
(hint : Integrable (fun p : E d × E d => g p.2 * w (p.2 - p.1) * h p.2) (PY.prod volume)) :
∫ s, (∫ z, g z * w (z - s) * h z) ∂PY = ∫ z, g z * smooth PY w z * h z := by sorry
end MultivarScores.L2Rule
Source
Meng, Taylor, Ben Taieb, Li, Scores for Multivariate Distributions and Level Sets, arXiv:2002.09578v6, p. 32, proof of Theorem 1, second and third lines of the divergence display
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.