equivalent conditions for uniform integrability


Let (Ω,ℱ,μ) be a measure spaceMathworldPlanetmath and S be a boundedPlanetmathPlanetmathPlanetmath subset of L1⁢(Ω,ℱ,μ). That is, ∫|f|⁢𝑑μ is bounded over f∈S. Then, the following are equivalentMathworldPlanetmathPlanetmathPlanetmathPlanetmathPlanetmath.

  1. 1.

    For every ϵ>0 there is a δ>0 so that

    ∫A|f|⁢𝑑μ<ϵ

    for all f∈S and μ⁢(A)<δ.

  2. 2.

    For every ϵ>0 there is a K>0 satisfying

    ∫|f|>K|f|⁢𝑑μ<ϵ

    for all f∈S.

  3. 3.

    There is a measurable functionMathworldPlanetmath Φ:ℝ→[0,∞) such that Φ⁢(x)/|x|→∞ as |x|→∞ and

    ∫Φ⁢(f)⁢𝑑μ

    is bounded over all f∈S. Moreover, the function Φ can always be chosen to be symmetricPlanetmathPlanetmath and convex.

So, for bounded subsets of L1, either of the above properties can be used to define uniform integrability. Conversely, when the measure space is finite, then conditions (2) and (3) are easily shown to imply that S is bounded in L1.

To show the equivalence of these statements, let us suppose that ∫|f|⁢𝑑μ<L for f∈S.

(1) implies (2)

For ϵ>0, property (1) gives a δ>0 so that ∫A|f|⁢𝑑μ<ϵ whenever f∈S and μ⁢(A)<δ. Choosing K>L/δ, Markov’s inequalityMathworldPlanetmath gives

μ(|f|>K)≤K-1∫|f|dμ≤L/K<δ

and, therefore, ∫|f|>K|f|⁢𝑑μ<ϵ.

(2) implies (3)

For each n=1,2,…, property (2) gives a Kn satisfying

∫(|f|-Kn)+⁢𝑑μ≤∫|f|>Kn|f|⁢𝑑μ≤2-n.

Without loss of generality, the Kn can be chosen to be increasing to infinityMathworldPlanetmathPlanetmath, so we can define Φ⁢(x)=∑n(|x|-Kn)+. Then,

∫Φ⁢(f)⁢𝑑μ=∑n∫(|f|-Kn)+⁢𝑑μ≤∑n2-n=1.

(3) implies (1)

First, suppose that ∫Φ⁢(f)⁢𝑑μ<M for f∈S. For ϵ>0, the condition that Φ⁢(x)/|x|→∞ as |x|→∞ gives a K>0 such that Φ⁢(x)/|x|≥2⁢M/ϵ whenever |x|>K. Setting δ=ϵ/2⁢K,

∫A|f|⁢𝑑μ≤∫|f|>K|f|⁢𝑑μ+K⁢μ⁢(A)<(ϵ/2⁢M)⁢∫|f|>KΦ⁢(f)⁢𝑑μ+K⁢δ<ϵ/2+ϵ/2=ϵ.

whenever μ⁢(A)<δ and f∈S.

Title equivalent conditions for uniform integrability
Canonical name EquivalentConditionsForUniformIntegrability
Date of creation 2013-03-22 18:40:17
Last modified on 2013-03-22 18:40:17
Owner gel (22282)
Last modified by gel (22282)
Numerical id 7
Author gel (22282)
Entry type Theorem
Classification msc 28A20