SOLUTIONS OF HW8 MINGFENG ZHAO April 20, 2011 1. [Problem 11, in Page 92] Let µ be a positive measure on X. A collection of functions {fα }α∈A ⊂ L1 (X, µ) is called uniformly integrable if for every > 0, there exists δ > 0 such that whenever µ(E) < δ, we have Z fα dµ < , for all α ∈ A E Show that a. Any finite subset of L1 (X, µ) is uniformly integrable. 1 1 1 b. If {fn }∞ n=1 is a sequence in L (X, µ) that converges in the L metric to f ∈ L (X, µ), then {fn }∞ n=1 is uniformly integrable. Proof. a. Let f1 , f2 , · · · , fn ∈ L1 (X, µ). For each fi , 1 ≤ i ≤ n, by Corollary 3.6, in Page 89, we know for any > 0, there exists δi > 0, 1 ≤ i ≤ n such that whenever µ(E) < δi , we have Z fi dµ < , 1 ≤ i ≤ n. E For the above > 0, take δ = min {δ1 , δ2 , · · · , δn }. Whenever µ(E) < δ, then we know µ(E) < δ ≤ δi for all 1 ≤ i ≤ n, which implies Z fi dµ < for all 1 ≤ i ≤ n. E Therefore, {f1 , f2 , · · · , fn } is uniformly integrable. 1 1 1 b. Let {fn }∞ n=1 be a sequence in L (X, µ) that converges in the L metric to f ∈ L (X, µ), that is, Z |fn (x) − f (x)|dµ = 0 lim n→∞ X 1 2 MINGFENG ZHAO So for any > 0, there exists N ≥ 1 such that whenever n ≥ N + 1, we have Z |fn (x) − f (x)|dµ < X . 2 Since f ∈ L1 (X, µ), so by the Corollary 3.6, in Page 89, we know for the above > 0, there exists δN > 0 such that whenever µ(E) < δN , we have Z f (x)dµ < . 2 E So for all n ≥ N + 1, we have Z Z f (x)dµ = (fn (x) − f (x))dµ + Z fn (x)dµ E E E Z Z ≤ (fn (x) − f (x))dµ + f (x)dµ E E Z ≤ E < Z |fn (x) − f (x)|dµ + f (x)dµ E + 2 2 = . For f1 , f2 , · · · , fN , which is a subset in L1 (X, µ), by the result of the part a., we know for the above 0 > 0, there exists δN > 0 such that Z fn dµ < for all 1 ≤ n ≤ N . E 0 Now for the above > 0, take δ = min {δN , δN } > 0, we know whenever µ(E) < δ, we have Z fn dµ < for all n ≥ 1. E Therefore, {fn }∞ n=1 is uniformly integrable. 2. [Problem 17, in Page 93] Let (X, M, µ) be a measure space, N be a sub σ-algebra of M, and ν = µ|M , such that (X, N , ν) is a σ-finite measure space. If f ∈ L1 (X, µ), then there exists g ∈ L1 (X, ν) SOLUTIONS OF HW8 (thus g is N -measurable) such that R E f dµ = R E 3 gdν for all E ∈ N , and g 0 is another such function, then g = g 0 ν-a.e. Proof. Let f ∈ L1 (X, µ), define the singed measure µ induced by f on M given by for any E ∈ M, we have Z µ(E) = f dµ. E And let ν = µ|N , we know that ν is a signed measure on N . So for E ∈ N , such that ν(E) = 0, then we have µ(E) = 0, so µ(E) = R E f dµ = 0, which implies that ν ν. Since ν is σ-finite, then by the Lebesgue-Radon-Nikodym theorem 3.8, in Page 90, we know that there exists an extended ν-integrable function g : X → R such that dν = gdν. Therefore, for all E ∈ N , we have Z Z f dµ = µ(E) = ν(E) = E gdν. E If g 0 ∈ L1 (X, ν) is another function such that for all E ∈ N , we have Z Z f dµ = E gdν. E This implies that dν = g 0 dν. By the Lebesgue-Radon-Nikodym theorem 3.8, in Page 90, we know g = g 0 ν-a.e. Remark 1. Here, I assume (X, N , ν) is a σ-finite measure space, of course, this implies that (X, M, µ) is a σ-finite measure space. But the book does not assume this, and the book just assumes (X, M, ν) is a σ-finite measure space. But I think this is not enough, here I give a counterexample. For example X = R, M = BR , and µ =Lebesgue measure on BR . Let N = {φ, R} be the trivial σ-subalgebra of M, and ν = µ|N . Consider the function f (x) = Z Z ∞ f (x)dx = R −∞ x2 1 x2 +1 1 dx = π. +1 on R. We know 4 MINGFENG ZHAO On the other hand, we know any N -measurable simple function g on R must be of the form g(x) ≡ c for some c ∈ R. So R R for all x ∈ R g(x)dν = c · ν(R), which implies that if g ∈ L1 (X, ν), then c = 0. Hence for all ν-integrable function g, we know g ≡ 0 on R. So we can not find ν-integrable function g such that Z Z f dµ = E gdν. E for all E ∈ N . Department of Mathematics, University of Connecticut, 196 Auditorium Road, Unit 3009, Storrs, CT 06269-3009 E-mail address: mingfeng.zhao@uconn.edu