5
$\begingroup$

Let $f$ be a real-valued Lebesgue measurable function on $\mathbb{R}$. Prove that there exist Borel measurable functions $g$ and $h$ such that $g(x)=h(x)$ almost everywhere and $g(x)\le f(x) \le h(x)$ for every $x \in \mathbb R$. I know that $f$ is measurable since there exists a sequence of simple function that converges to $f$. I have no further idea how to tackle this problem.

$\endgroup$
  • $\begingroup$ yea I also got confused. $\endgroup$ – user145993 Sep 29 '15 at 23:33
  • $\begingroup$ I am familiar with the usual statement of this theorem: you need $g(x)\le f(x)\le h(x)$, not $g(x)<f(x)<h(x)$, or else it trivially fails, as @user251257 points out. $\endgroup$ – Mario Carneiro Sep 30 '15 at 0:00
  • $\begingroup$ @Mario you are right, Then, can you do it for g(x)≤f(x)≤h(x) $\endgroup$ – user145993 Sep 30 '15 at 0:13
  • $\begingroup$ Can $g$ and $h$ take infinite values? $\endgroup$ – quartermind Mar 13 '16 at 10:43
3
$\begingroup$

It seems that this is in fact not possible: there exists a measurable function $f:\mathbb R\to\mathbb R$ for which one cannot find any Borel function $h:\mathbb R\to\mathbb R$ such that $f\leq h$ everywhere.

Here is a (hopefully correct) example. Since my previous answer was wrong, I would be glad to know if this one is OK...

Take a perfect set $K\subseteq\mathbb R$ with Lebesgue measure $0$. The family of all perfect subsets of $K$ has the cardinality of the continuum; so (using the axiom of choice), one can enumerate the perfect subsets of $K$ as $(L_\alpha)_{\alpha<\mathfrak c}$. Then, one can define by transfinite induction a family $(C_\alpha)_{\alpha<\mathfrak c}$ of countable (infinite) sets such that $C_\alpha\subseteq L_\alpha$ and $C_\alpha\bigcap \left(\bigcup_{\beta<\alpha} C_\beta\right)=\emptyset$ for all $\alpha<\mathfrak c$. Indeed, if the $C_\beta$ have been found for all $\beta<\alpha$, then the set $Z_\alpha:=\bigcup_{\beta<\alpha} C_\beta$ has cardinality less than $\mathfrak c$, so $L_\alpha\setminus Z_\alpha$ is infinite because $L_\alpha$ has cardinality $\mathfrak c$, and hence $L_\alpha\setminus Z_\alpha$ contains a countable infinite set $C_\alpha$.

Now, enumerate each $C_\alpha$ as $C_\alpha=\{ x_{\alpha,n};\; n\in\mathbb N\}$, and define a function $f:\mathbb R\to\mathbb R$ as follows : $f(x_{\alpha,n}):=n$ for all $\alpha,n$, and $f(x)\equiv0$ outside $\bigcup_{\alpha<\mathfrak c} C_\alpha$. This is a measurable function because $f(x)\equiv 0$ outside $K$ (and hence almost everywhere). On the other hand, the function $f$ has the following property : it is not bouded above on any perfect subset of $K$.

Assume that one can find a Borel function $h:\mathbb R\to\mathbb R$ such that $f\leq h$. Since $h$ is real-valued, we have $K=\bigcup_{n\in\mathbb N} B_n$, where $B_n:=\{ x\in K;\; h(x)\leq n\}$. Since $K$ is uncountable, at least one of the sets $B_n$ must be uncountable; and since $B_n$ is a Borel set (because $h$ is Borel), it must contain some perfect set $L$ (by a well known theorem). Then $h$ is bounded above on $L$, a contradiction since $h\geq f$ and $f$ is not bounded above on $L$.

$\endgroup$
  • 1
    $\begingroup$ Yes, this is right. Though, there is bit easier way to build counterexample: take function $f\colon [0,1] \rightarrow \mathbb R$ which is unbounded on every set of positive measure (non-linear additive function fits), and "rescale" $f$ to uncountable Borel set of zero measure $K$ by Borel isomorphism between $[0,1]$ and $K$. On $\mathbb R \setminus K$ assume our function is zero. If $K$ is Cantor set the isomorhism can be written explicitly. But in fact, your solution uses similar ideas. $\endgroup$ – quartermind Mar 13 '16 at 13:10
  • $\begingroup$ @quartermind OK, thanks! Why didn't you post an answer, since apparently you knew it? $\endgroup$ – Etienne Mar 13 '16 at 14:31
  • $\begingroup$ Laziness (joke). I wanted to get clarification from the asker. $\endgroup$ – quartermind Mar 13 '16 at 14:48
0
$\begingroup$

Warning. What follows does not provide an answer to the question.

First, assume that $f$ is an indicator function, say $f=\mathbf 1_A$ where $A$ is a measurable. You can find Borel sets $B,C$ such that $B\subseteq A\subseteq C$ and $C\setminus B$ has measure $0$; then take $g:= \mathbf 1_B$ and $h:=\mathbf 1_C$.

Next, assume that $f$ is a simple function, say $f=\sum_{i=1}^N \alpha_i\mathbf 1_{A_i}$, where the sets $A_i$ are measurable and $\alpha_i\in\mathbb R$. By the first case, one can choose Borel functions $g_i$ and $h_i$ such that $g_i=h_i$ almost everywhere and $g_i\leq \mathbf 1_{A_i}\leq h_i$ if $\alpha_i\geq 0$, whereas $h_i\leq \mathbf 1_{A_i}\leq g_i$ if $\alpha_i<0$. Then take $g:=\sum_i\alpha_i g_i$ and $h:=\sum_i\alpha_i h_i$.

Finally, assume that $f$ is an arbitrary measurable function. One can find a sequence of simple measurable functions $(f_n)$ such that $f_n\to f$ everywhere. For each $n$, choose Borel functions $g_n$ and $h_n$ such that $g_n\leq f_n\leq h_n$ and $g_n=f_n=h_n$ almost everywhere. Now define $\widetilde g:=\limsup g_n$ and $\widetilde h_n:=\liminf h_n$. These are Borel extended real-valued functions, and $\widetilde g\leq f\leq\widetilde h$ because $g_n\leq f_n\leq h_n$ for all $n$ and $f_n\to f$. Moreover, we have $\widetilde g=f=\widetilde h$ almost everywhere because $g_n=f_n=h_n$ almost everywhere for each $n\in\mathbb N$ and $f_n\to f$. Now, set $E:=\{ \widetilde g=-\infty\}\cup\{\widetilde h=+\infty\}$, a Borel set of measure $0$ because $f$ is real-valued and $g=f=h$ a.e. Define $g:=\widetilde g$ on $\mathbb R\setminus E$ and $g:=0$ on $E$; and likewise $h:=\widetilde h$ on $\mathbb R\setminus E$ and $h:=0$ on $E$.

Unfortunately, this doesn't work since, as pointed out by @quartermind, one does not know that $g\leq f\leq h$ on $E$.

$\endgroup$
  • $\begingroup$ You're welcome! $\endgroup$ – Etienne Oct 4 '15 at 20:42
  • $\begingroup$ $g \le f \le h$ might not hold on $E$. $\endgroup$ – quartermind Mar 12 '16 at 23:56
  • $\begingroup$ @quartermind You're absolutely right. Thanks for pointing out the mistake. $\endgroup$ – Etienne Mar 13 '16 at 9:02
  • $\begingroup$ @quartermind Could you please check my second answer? $\endgroup$ – Etienne Mar 13 '16 at 12:30
  • $\begingroup$ @user145993 My answer was wrong (as pointed out by @quartermind). I posted another one, which says that in fact one cannot find the functions $g$ and $h$. $\endgroup$ – Etienne Mar 13 '16 at 17:21

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.