Utility Lemma 13 — if X ≤ Y a.s. then E|X| ≤ |E X| + 2 E|Y|
ProvedLearnStability.ERMLOO.utility_lemma13Let and be integrable real random variables on a probability space such that almost surely. Then
The inequality turns a bound on a signed expectation, together with a one-sided pointwise majorant, into a bound on an expected absolute value; it is the step from on-average generalization to generalization in Lemma 14.
Formalization Note. Integrability of and is assumed (the paper leaves it implicit); without it Lean's integrals default to .
import Mathlib open MeasureTheory
namespace LearnStability.ERMLOO
theorem utility_lemma13 {Ω : Type*} [MeasurableSpace Ω] {μ : Measure Ω} [IsProbabilityMeasure μ]
(X Y : Ω → ℝ) (hX : Integrable X μ) (hY : Integrable Y μ) (hXY : X ≤ᵐ[μ] Y) :
∫ ω, |X ω| ∂μ ≤ |∫ ω, X ω ∂μ| + 2 * ∫ ω, |Y ω| ∂μ := by sorry
end LearnStability.ERMLOO
Read-back
What the Lean code literally says, in plain math · claude-opus-5-5
The setting is a probability space and two real-valued functions on , both -integrable, with -almost everywhere. The statement asserts
that is, . No sign condition is placed on , and no independence or other relation between and is assumed beyond almost everywhere.
Degenerate cases:
- Nonempty space. is necessarily non-empty, since is a probability measure.
- almost surely. Then almost surely, and the claim reads .
- Integrability. Both functions are assumed integrable, so every integral is a genuine one.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.