Let $(A,\mathcal{F},\mu)$ be finite measure space and $\{f_n\}$ a sequence of finite real measurable functions so that $f_n\rightarrow f$ a.e. We say $f_n\rightarrow f$ almost uniformly if $\epsilon>0$, there is $E\subseteq A$ such that $f_n \rightarrow f$ uniformly on $E^c$ and $\mu(E)<\epsilon$.

I want to show that $f_n\rightarrow f$ almost uniformly implies convergence in $\mu$. For this, suppose not. Then $$\exists \eta,\epsilon>0:\forall N\in \mathbb{N}:\exists n>N:\mu(\mid f_n-f\mid\geq\epsilon)\geq \eta, $$ i.e., for infinitely many points $n\in \mathbb{N}$. From the definition of almost uniform convergence, $\exists E:\mu(E)<\eta$ and $f_n\rightarrow f$ uniformly on $E^c$. Contradiction.


It seems intuitive to me. But how to deduce this contradiction precisely? I know that if $x\in E$, then it must satify the negation of uniform convergence which is $$\exists \epsilon>0:\forall N\in \mathbb{N}:\exists n>N:\mid f_n-f\mid\geq\epsilon.$$ Now, $x$ may not be in $\{f_n \text{ does not converge in measure to } f \}$ if $\mu(\mid f_n(x)-f(x)\mid\geq\epsilon)<\eta$. So I conclude that $$\{f_n \text{ does not converge in measure to } f \}\subseteq E$$ implying that $\eta>\mu(E)\geq \eta$; a contradiction.

My argument seems right but also very inefficient. How could you express this idea as clean as possible?



Here's a way without going for contradiction: Let $\epsilon >0$ be fixed and consider some $\delta >0$.

By almost uniform convergence, there exists some $E$ with $\mu(E)\leq \delta$ and some $N$ such that $n\geq N\implies \forall x\in E^c, |f_n(x)-f(x)|< \epsilon$.

For $n\geq N$, note the inclusion $(|f_n-f|\geq \epsilon) \subset E$, hence $\mu((|f_n-f|\geq \epsilon))\leq \mu(E)\leq \delta$.

Hence $\forall \delta>0, \exists N, n\geq N \implies \mu((|f_n-f|\geq \epsilon))\leq \delta$. Thus $ \mu((|f_n-f|\geq \epsilon)) \to 0$.

  • $\begingroup$ Thanks! The inclusion $\{\mid f_n-f\mid\geq \epsilon\}\subset E$ that I don't understand. What is this infinity-norm that you wrote? $\endgroup$ – Celine Harumi Aug 23 '19 at 20:30
  • $\begingroup$ @Danmat I removed the infinity norm (which denotes the sup-norm), do you understand now ? $\endgroup$ – Gabriel Romon Aug 23 '19 at 20:32
  • $\begingroup$ It has something to do with the fact that uniform convergence implies pointwise convergence? And then the complement of the latter is contained in the complement of the former. $\endgroup$ – Celine Harumi Aug 23 '19 at 20:38
  • $\begingroup$ @Danmat I just wrote the definition of uniform convergence on $E^c$. Pointwise convergence is not used in the proof. $\endgroup$ – Gabriel Romon Aug 23 '19 at 20:40
  • $\begingroup$ But, $E$ is constituted of elements in $A$ such that $f_n \rightarrow f$ does not converge uniformly, right? $\endgroup$ – Celine Harumi Aug 23 '19 at 20:42

This is proved more succinctly by a direct proof.

Given $\varepsilon > 0,$ let $E$ be as in the definition of almost-uniform convergence. Then there is some $N$ such that $n \geq N$ implies $$\mu(E^c \cap \{|f_n-f|\geq\varepsilon\}) = 0$$ What does this imply for $\mu(|f_n-f|\geq \varepsilon)?$

  • $\begingroup$ How to conclude that $\{\mid f_n-f\mid \}\subseteq E$? $\endgroup$ – Celine Harumi Aug 23 '19 at 20:55
  • $\begingroup$ @Danmat Forget about $E$ for a second. On $E^c,$ the functions converge uniformly. This means that there will be some $N$ such that $n \geq N$ implies $|f_n - f| < \varepsilon.$ In particular, on $E^c,$ $$n \geq N \implies \{|f_n - f| \geq \varepsilon\} = \emptyset.$$ This directly means $$E^c \cap \{|f_n - f| \geq \varepsilon\} = \emptyset.$$ If you want, then you can use the property of sets that $$A^c \cap B = \emptyset \iff B \subseteq A$$ $\endgroup$ – Brian Moehring Aug 23 '19 at 21:05
  • $\begingroup$ Very nice! Now, it's all clear to me! $\endgroup$ – Celine Harumi Aug 23 '19 at 21:11

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.