The activity-normalized sub-Weibull class (Definition 9.1)
DefinitionLawLocalComp_Supplement_IsSubWeibulllaw-compressionmonotone-normsp2o-gran-per-chapterp2o-plan-paperp2o-v1probabilitystochastic-loads
Fix rational and . For a nonconstant finite nonnegative law put , , and . satisfies the activity-normalized sub-Weibull promise if
Point masses are included in the class separately.
The class constrains only the tail of the active part of , relative to its activity-normalized mean ; it requires no lattice support and no hazard-rate structure. Theorem 9.3 determines the sharp atom order for this class.
Formalization Note For nonconstant , and , so and (44) is written as . , and are the base mission's published baseline, activity and activeMean; is the real power.
Definition code
import Mathlib
import Definitions.Def_LawLocalComp_baseline
import Definitions.Def_LawLocalComp_activeMean
namespace LawLocalComp
namespace Supplement
/-- **The activity-normalized sub-Weibull class (Definition 9.1).**
Mahboub–Rahgoshay, *Optimal Law-Local Compression for Stochastic Loads*, STOC 2027
submission manuscript of Oct. 29, 2026 (sha256 5f71501c…aa873), Definition 9.1, (44), p.16: "Fix rational α > 0, A ≥ 1. For nonconstant X, put b = min supp X,
R = X − b, p = P(R > 0), ν = E[R | R > 0]. The activity-normalized sub-Weibull promise is
P(R ≥ Aνu | R > 0) ≤ 2^{−u^α} (u ≥ 1). Point masses are included separately."
`IsSubWeibull α A X`: `X` is a point mass, or for every real `u ≥ 1`,
`P(X − b ≥ A ν u) ≤ 2^{−u^α} · p` with `b = baseline X`, `p = activity X b = P(X > b)` and
`ν = activeMean X b = E[X − b | X > b]`.
**Formalization Note.** For a nonconstant law `p > 0` and `ν > 0`, so `A ν u > 0` and the event
`{R ≥ A ν u}` lies inside `{R > 0}`; the conditional bound (44) is therefore the unconditional bound
multiplied by `p`. `u^α` is `Real.rpow`. The support need not be a lattice. -/
def IsSubWeibull (α A : ℚ) (X : Law) : Prop :=
(∃ t : ℝ, X.w = Finsupp.single t 1) ∨
∀ u : ℝ, 1 ≤ u →
∑ x ∈ X.w.support with (A : ℝ) * activeMean X (baseline X) * u ≤ x - baseline X, X.w x ≤
(2 : ℝ) ^ (-(u ^ (α : ℝ))) * activity X (baseline X)
end Supplement
end LawLocalComp
Source
Mahboub–Rahgoshay, Optimal Law-Local Compression for Stochastic Loads, STOC 2027 submission (manuscript of Oct. 29, 2026, sha256 5f71501c…aa873), Definition 9.1, (44), p.16