16
$\begingroup$

I want to prove the monotone convergence theorem using Fatou's lemma (and its reverse) as exercise, and I need a check; I will use also the following properties of limit inferior and limit superior:

Let $f,g: D \to \mathbb{R}$ be functions. Then if $\lim_{x \to c} f(x)$ exists in $\tilde{\mathbb{R}}=\mathbb{R} \cup \{-\infty, + \infty \}$ we have $$\liminf_{x \to c} (f(x) +g(x))=\lim_{x \to c} f(x) + \liminf_{x \to c} g(x)$$ and the same for limit superior.

Statement. Let $(X, \mathcal{M}, \mu)$ be a measure space. Assume that $f_0 \le f_1 \le f_2 \le \dots$ is an increasing sequence of functions in $L^{+}(X)$ ($=$ the set of all extended real valued positive measurable functions), such that $f_n \uparrow f$ pointwise. Then $$ \int_X f = \lim_{n \to \infty} \int_X f_n $$

Proof. $g_n=(f-f_n)$ is a sequence of positive measurable functions, and then I can apply Fatou's lemma; it is $$\int_X \liminf_n (f-f_n) d\mu \le \liminf_n \int_X (f-f_n) d \mu$$Using the property above we have that $$\int_X \liminf_n (f-f_n) d\mu =\int_X (f - \liminf_n f_n) d \mu=0$$ and $$\liminf_n \int_X (f-f_n) d \mu=\liminf_n \left[ \int_X f d\mu - \int_X f_n d\mu \right]=\int_X f d \mu - \liminf_n \int_X f_n d \mu$$ So it is $$\int_X f d \mu \ge \liminf_n \int_X f_ d \mu \ge \int_X \liminf_n f_n d \mu=\int_X f d \mu$$using the lemma on $f_n$, which implies $$\int_X f d \mu = \liminf_n \int_X f_n d \mu$$Now: if $$\begin{split} \int_X f d \mu=\infty & \ \longrightarrow \ \underbrace{\liminf_n \int_X f_n d \mu}_{=\infty} \le \limsup_n \int_X f_n d \mu = \infty \\ & \ \longrightarrow \ \int_X f d \mu = \lim_n \int_X f_n d \mu = \infty \end{split}$$ and if $$\int_X f d \mu < \infty$$ We can apply, in a similar way as above, the reverse of Fatou's lemma considering that $f-f_n \le f - f_0$ and that $f-f_0$ is integrable ( - here I've used the fundamental hypothesis of $(f_n)_{n \in \mathbb{N}}$ increasing). We obtain $$ \int_X f d \mu=\limsup_n \int_X f_n d \mu $$and then the thesis.

What do you think about it?

Thank you very much.

$\endgroup$
  • 1
    $\begingroup$ You can apply Fatou's Lemma to a less generalized version which essentially states $$ \int_X f \leq \lim\inf \int_X f_n $$ if $\{f_n\} \rightarrow f$ pointwise a.e. on $X$. Then you can apply the properties of monotonicity, and complete the proof by stating $$ \lim\sup \int_X f_n \leq \int_X f. $$ Your method of proof is extremely useful when you get to the dominated convergence theorem, so keep that in mind. $\endgroup$ – user1876508 Nov 18 '13 at 1:43
8
$\begingroup$

I think your proof is basically fine, but it looks to me as if there are a few places where you were a bit careless.

$$\int_X \liminf_n (f-f_n) d\mu =\int_X (f - \liminf_n f_n) d \mu=0$$

While what you wrote here is all technically true, your choice of $\int (f - \liminf f_n)$ as the intermediary expression is unnatural because it suggests that you used the assertion $$\liminf (-a_n) = - \liminf a_n,$$ which is false in general. Instead, $$\liminf (-a_n) = -\boldsymbol\limsup a_n.$$ So in our case, it would be more straightforward to argue that $$\int \liminf(f - f_n) = \int(f - \limsup f_n) = \int(f-f) = 0.$$

$$\liminf_n \int_X (f-f_n) d \mu=\liminf_n \left[ \color{maroon}{\int_X f d\mu - \int_X f_n d\mu} \right] = \color{maroon}{\int_X f d \mu - \liminf_n \int_X f_n d \mu}$$

Aside from another potential mix-up of $\liminf$ and $\limsup$, note that the two highlighted expressions may not be well-defined because they could take the form $\infty - \infty$. You would need to prove more carefully that $$ \liminf \int (f - f_n) \geq 0 \implies \int f \geq \limsup \int f_n \geq \liminf \int f_n. $$


A cleaner approach was brought up by user1876508 in their comment. We have, using Fatou's lemma along the way, that $$ \int f = \int \liminf f_n \leq \liminf \int f_n \leq \limsup \int f_n. $$ For all $n$, we can deduce from $f_n \leq f$ that $\int f_n \leq \int f$. It follows that $$\limsup \int f_n \leq \int f.$$ Hence $\lim \int f_n$ exists and is equal to $\int f$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.