§3.1, p. 8 — S(P, y) = S′(P, y) + ∫ w²(z − y) h(z) dz: the scores (2) and (3) differ by a term depending on y only
ProvedMultivarScores.L2Rule.S_eq_S_prime_addp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1proper-scoring-rulescoring-rules
Let be a distribution on , a kernel, a weight and . Suppose , and are Lebesgue integrable. Then the L² scoring rules (2) and (3) satisfy
The last term depends on only, not on ; this is the sense in which the paper calls (3) "equivalent" to (2): both rules have the same divergence.
Formalization Note The three integrability hypotheses make each of the three integrals finite; without them a Bochner integral is and the identity can fail. They are the pointwise forms of Assumption 1(b)/(c).
Preamble
import Mathlib import Definitions.Def_MultivarScores_L2Rule_Setting
Formal statement
namespace MultivarScores.L2Rule
open MeasureTheory
/-- (2)–(3) relation, §3.1, p. 8: `S(P, y) = S′(P, y) + ∫ w²(z − y) h(z) dz`, where the last term
depends on `y` only. -/
theorem S_eq_S_prime_add {d : ℕ} (P : Measure (E d)) (w h : E d → ℝ) (y : E d)
(h_sq_smooth : Integrable (fun z => smooth P w z ^ 2 * h z))
(h_sq_w : Integrable (fun z => w (z - y) ^ 2 * h z))
(h_cross : Integrable (fun z => smooth P w z * w (z - y) * h z)) :
S P w h y = S' P w h y + ∫ z, w (z - y) ^ 2 * h z := by sorry
end MultivarScores.L2Rule
Source
Meng, Taylor, Ben Taieb, Li, Scores for Multivariate Distributions and Level Sets, arXiv:2002.09578v6, p. 8, §3.1, the sentence between (2) and (3)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.