Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

Levi Convergence Theorem for step functions

  1. Jun 24, 2009 #1
    Hi Everyone,

    I am currently reading Apostol's "Mathematical Analysis" to introduce myself to some advanced calculus techniques I will need in probability. Usually the proofs in this book are fairly straight forward, without too many missed steps, which is why I think this book is great for self study. I was reading about the Levi's convergence theorem for step functions and didn't like the proof that Apostol gives. To me it seems unnecessarily long and counter-intuitive. I think I have an alternative proof which is simpler to understand and shorter. But I started doubting myself, why would he present a long winded proof, when there is a simpler available - maybe something in my proof is wrong and I just can't spot it? That is why I want to share it here, maybe someone would be able to quickly see an error in my reasoning. Thank you very much in advance for anyone caring to taking a look.

    Let [tex]\{s_n\}[/tex] be a sequence of step functions such that:

    1). for every [tex]x \in I[/tex], [tex]s_{n+1}(x) \ge s_n(x)[/tex]

    2). [tex]$\lim_{n \to \infty} \int_I s_n$[/tex] exists.

    Then [tex]\{s_n\}[/tex] converges almost everywhere on [tex]I[/tex] to a limit function [tex]f[/tex].

    Since integrals of [tex]\{s_n\}[/tex] converge, they must be bounded. Suppose that
    [tex]$\lim_{n \to \infty} \int_I s_n < M$[/tex]. Let [tex]D[/tex] be a set of points x of I such that [tex]s_n(x)[/tex] diverges. Given any [tex]$\epsilon$[/tex], define:
    [tex]B_n = \{x | s_n(x) > \frac{2M}{\epsilon}\}[/tex]
    Then we have:

    [tex]B_n \subseteq B_{n+1} [/tex]

    [tex]D \subseteq \bigcup_{n} B_n [/tex]

    Also, each [tex]B_n[/tex] is a finite collection of intervals. Define [tex]|B_n|[/tex] to be the sum of the lengths of these intervals.

    Now for any n:

    [tex]M > \int_I s_n \ge \int_{B_n} s_n > \int_{B_n}\frac{2M}{\epsilon} = \frac{2M}{\epsilon}\int_{B_n} 1 = \frac{2M}{\epsilon}|B_n|[/tex]

    So that [tex] |B_n| < \epsilon/2 [/tex]. From this we get that [tex]$\lim_{n\to\infty} |B_n| \le \epsilon/2$[/tex].
    Since [tex]|\bigcup_{n=1}^m B_n| = |B_m|[/tex], [tex]D[/tex] is covered by a countable collection of intervals the sum of whose lengths is less than any [tex]\epsilon[/tex], therefore [tex]D[/tex] has measure zero.
    From this it follows that [tex]s_n[/tex] is bounded almost everywhere, and so converges almost everywhere on I to some function [tex]f[/tex].
    Last edited: Jun 25, 2009
  2. jcsd
Share this great discussion with others via Reddit, Google+, Twitter, or Facebook

Can you offer guidance or do you also need help?
Draft saved Draft deleted