If a function is monotonic on set E. Is f differentiable almost everywhere? I have proved for case E closed bounded or open intervals, hence all open sets. But in general I am not able to figure it out.

And this I know that derivative of f at isolated points is not defined(or is infinity). But there can be atmost countable no. of isolated points. So what we can say about general E whether it contains isolated points or not.

  • $\begingroup$ I am not sure I understand. Are you assuming $\mathrm{dom}(f)=E$? $\endgroup$ – Andrés E. Caicedo Dec 3 '14 at 3:48
  • $\begingroup$ Anyway, assuming that's what you meant, can you extend $f$ to a function monotone on $\bar E$ and then further to a function monotone everywhere? $\endgroup$ – Andrés E. Caicedo Dec 3 '14 at 3:55
  • $\begingroup$ @AndresCaicedo If it is possible then how my problem will get solve? As someone has pointed the way of extension. Yes I am assuming dom(f) = E $\endgroup$ – Sushil Dec 3 '14 at 13:58

First, you should note that the set of isolated points of $E$ is countable. This is in fact a general property of $\mathbb{R}$

Theorem: Let $E$ be a subset of $\mathbb{R}$ and let $F$ be the set of isolated points of $\mathbb{R}$. Then $F$ is at most countable.

Proof: Suppose otherwise, that is, that $F$ is uncountable. Then there exists some interval $[k,k+1]$ such that $F\cap[k,k+1]$ is uncountable. For each $x\in F\cap [k,k+1]$, choose a rational number $q_x$, $0<q_x<1$ such that $(x-2q_x,x+2q_x)\cap F=\varnothing$. Since the set $\left\{q_x:x\in F\cap[k,k+1]\right\}$, then there exists some $q$ such that $X=\left\{x:q_x=q\right\}$ is uncountable, in particular infinite. The choice of $q_x$ implies that the sets $(x-q,x+q)$ are all disjoint for $x\in X$, and they are all contained in $[k-1,k+2]$. Therefore, we constructed an infinite family of disjoint intervals of length $2q$, all of which are contained in the bounded interval $[k-1,k+2]$, a contradiction. QED

(Probably, there is a nicer proof of this theorem somewhere in this site.)

Therefore, we should not worry about the isolated points of $E$ when analysing derivatives: the set of isolated points has null measure.

A trick that works here is to extend your function $f$ to an interval containing $E$. We can do this in the following manner:

Let $E\subseteq\mathbb{R}$ and $f:E\to\mathbb{R}$ be monotonic. The function $\hat{f}:(\inf E,\sup E)\to\mathbb{R}$ given by $\hat{f}(x)=\sup_{y\in E,y\leq x}f(y)$ is an extension of $f$ (if $\sup E$ or $\inf E\in E$, define $\hat{f}(\sup E)=f(\sup E)$ or $\hat{f}(\inf E)=f(\inf E)$).

Another extension is given by $\overline{f}(x)=\inf_{y\in E,y\geq x}f(x)$. In fact, you can check that if $g$ is any other extension of $f$ defined on $[\inf E,\sup E]\cap E$, then $\hat{f}(x)\leq g(x)\leq\overline{f}(x)$ for all $x$.

Alternatively, you can prove this with Zorn's Lemma, but the argument is basically the same: Zorn's lemma gives you a maximal extension of $f$ to a monotonic function $\widetilde{f}:F\to \mathbb{R}$ defined on some subset $F\supseteq E$. To show that $F$ is an interval you apply the argument above and extend $\widetilde{f}$ to some interval containing $F$. Maximality implies that $F$ is that interval.

Now, about your question of differentiability of $f$: For almost every point $x$ of $(\inf E,\sup E)$, the function $\hat{f}$ is differentiable at $x$. But we also know that almost every point of $E$ is not isolated. Using these two fact, we conclude that almost every point $x$ of $E\cap(\inf E,\sup E)$ is not an isolated point of $E$, and $\hat{f}$ is differentiable at $x$. You can then check that for such $x$, $f$ is differentiable at $x$, and $f'(x)=\hat{f}'(x)$.

Therefore, $f$ is differentiable at almost every point of $E$.

  • $\begingroup$ Are points of differentiability of f and f^ are same? And I think f^ you defined for case f is increasing $\endgroup$ – Sushil Dec 3 '14 at 13:46
  • $\begingroup$ Using Zorn's Lemma can you please give some reference $\endgroup$ – Sushil Dec 3 '14 at 13:46
  • $\begingroup$ @Sushil I put lots of details. I hope it gets clearer now. $\endgroup$ – Luiz Cordeiro Dec 3 '14 at 21:16

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.