Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2, p. 13 — the level set score (S′)^Γ (11) is consistent for the α level set L(f_Y ∗ w; α)

Proved
MultivarScores.LevelSet.theorem_2

by mikedeng1 · Oct 14, 2026 · Mathlib 0df444a (Lean v4.33.1)

consistent-scoring-functionlevel-setsp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1probabilistic-forecastingscoring-rules

Let PYP_YPY​ be a probability measure on Rd\mathbb R^dRd, the distribution of the observation YYY, with smoothed density (fY∗w)(z)=∫w(z−s) dPY(s)(f_Y*w)(z)=\int w(z-s)\,dP_Y(s)(fY​∗w)(z)=∫w(z−s)dPY​(s). Let w,h:Rd→Rw,h:\mathbb R^d\to\mathbb Rw,h:Rd→R be measurable, with w≥0w\ge0w≥0, h≥0h\ge0h≥0, hhh integrable, and (s,z)↦w(z−s)h(z)(s,z)\mapsto w(z-s)h(z)(s,z)↦w(z−s)h(z) integrable with respect to PY⊗LdP_Y\otimes\mathcal L^dPY​⊗Ld. Let α>0\alpha>0α>0 and let A⊆RdA\subseteq\mathbb R^dA⊆Rd be a Borel set. With the level set score

(S′)Γ(A,y;w,h,α)=∫Rd(α−w(z−y))1{z∈A}h(z) dz(11)(S')^\Gamma(A,y;w,h,\alpha)=\int_{\mathbb R^d}\bigl(\alpha-w(z-y)\bigr)\mathbb 1\{z\in A\}h(z)\,dz \qquad (11)(S′)Γ(A,y;w,h,α)=∫Rd​(α−w(z−y))1{z∈A}h(z)dz(11)

and the α\alphaα level set L(fY∗w;α)={z:(fY∗w)(z)≥α}L(f_Y*w;\alpha)=\{z : (f_Y*w)(z)\ge\alpha\}L(fY​∗w;α)={z:(fY​∗w)(z)≥α} of (9),

EPY[(S′)Γ(L(fY∗w;α),∙;w,h,α)]≤EPY[(S′)Γ(A,∙;w,h,α)].\mathbb E_{P_Y}\bigl[(S')^\Gamma(L(f_Y*w;\alpha),\bullet;w,h,\alpha)\bigr]\le\mathbb E_{P_Y}\bigl[(S')^\Gamma(A,\bullet;w,h,\alpha)\bigr].EPY​​[(S′)Γ(L(fY​∗w;α),∙;w,h,α)]≤EPY​​[(S′)Γ(A,∙;w,h,α)].

In words, (11) is a consistent scoring function for the α\alphaα level set of fY∗wf_Y*wfY​∗w: no Borel set has a smaller expected score than the true level set. This is the first claim of Theorem 2; it gives the level set analogue of a proper scoring rule, with the CDF level set score (13) and the density level set score (12) as special cases.

Formalization Note The product-integrability hypothesis pins down Assumption 1 and Theorem 2's "(S′)Γ(A,∙)(S')^\Gamma(A,\bullet)(S′)Γ(A,∙) is finite Ld\mathcal L^dLd-a.e. and (S′)Γ(A,∙)fY(∙)(S')^\Gamma(A,\bullet)f_Y(\bullet)(S′)Γ(A,∙)fY​(∙) is integrable". With w,h≥0w,h\ge0w,h≥0, Tonelli bounds ∬w(z−y)1B(z)h(z) dz dPY(y)\iint w(z-y)\mathbb 1_B(z)h(z)\,dz\,dP_Y(y)∬w(z−y)1B​(z)h(z)dzdPY​(y) by ∬w(z−y)h(z) dz dPY(y)\iint w(z-y)h(z)\,dz\,dP_Y(y)∬w(z−y)h(z)dzdPY​(y) for every Borel BBB at once, including L(fY∗w;α)L(f_Y*w;\alpha)L(fY​∗w;α), for which the comparison also needs it, and λ(B)≤∫h<∞\lambda(B)\le\int h<\inftyλ(B)≤∫h<∞ handles the α\alphaα-term. A single PYP_YPY​ replaces the class Vd\mathcal V^dVd: the statement is pointwise in PYP_YPY​, and Vd\mathcal V^dVd enters only through Assumption 1. The paper allows www to be a local Borel measure (Assumption 1(a)); here www is a measurable function, a disclosed specialization that excludes only the Dirac kernel w=δ0w=\delta_0w=δ0​ (the density level set score (12)).


Retired 2026-10-27: a Lean-vs-paper fidelity audit found this statement weaker than, or narrower than, the cited result. A corrected statement will be resubmitted.

Preamble
import Mathlib
import Definitions.Def_MultivarScores_LevelSet_Setting
Formal statement
namespace MultivarScores.LevelSet

open MeasureTheory

theorem theorem_2 {d : ℕ} (w h : MultivarScores.L2Rule.E d → ℝ) (PY : Measure (MultivarScores.L2Rule.E d)) [IsProbabilityMeasure PY]
    (hw_meas : Measurable w) (hw_nonneg : ∀ z, 0 ≤ w z)
    (hh_meas : Measurable h) (hh_nonneg : ∀ z, 0 ≤ h z) (hh_int : Integrable h)
    (hwh_int : Integrable (fun p : MultivarScores.L2Rule.E d × MultivarScores.L2Rule.E d => w (p.2 - p.1) * h p.2) (PY.prod volume))
    (α : ℝ) (hα : 0 < α) (A : Set (MultivarScores.L2Rule.E d)) (hA : MeasurableSet A) :
    ∫ y, scoreΓ w h α (levelSet (MultivarScores.L2Rule.smooth PY w) α) y ∂PY ≤ ∫ y, scoreΓ w h α A y ∂PY := by sorry

end MultivarScores.LevelSet
Source
Meng, Taylor, Ben Taieb, Li, Scores for Multivariate Distributions and Level Sets, arXiv:2002.09578v6, p. 13, Theorem 2 (first two sentences), display (11); p. 12, (9) and the definition of consistency; p. 32, Assumption 1
Human review
  • Endorsed by Shuze Chen · Oct 14, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 14, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me