I have recently read the famous Weierstrass Non differentiable function which is continuous everywhere but nowhere differentiable.

But now my question is:

Given a countable set $S (\subset \mathbb{R})$ , can we construct a function which is continuous everywhere but non differentiable only at points of S..?

My Attempt:
If $S$ finite say, $S=\{c_1,c_2,...,c_n\}$ Then the following function is the desired one:$$f(x)=\sum_{i=1}^{n}|x-c_i|$$

But I cannot construct the function for a arbitrary Enumerable set $S.$

Can we generalize the question for an arbitrary set $S$?
Please Help.

Thank you...!!

  • $\begingroup$ You can make an easy example of a set which has corners periodically by just defining the function on, say, $[-1,1]$ and arranging so that it takes equal values on the ends, and then copy-pasting it. Maybe you can use this to handle the general case? I'm not sure - just a thought. $\endgroup$ Dec 26, 2016 at 5:09
  • $\begingroup$ Sir, I think I got your point a little...but Is there any explicit example...??@ A Yerger $\endgroup$
    – sigma
    Dec 26, 2016 at 5:14
  • $\begingroup$ If you want to write down an explicit formula for such a thing that won't be defined piecewise, it's done with Fourier series. But I have thought of another example which modifies your own, and is also discontinuous on the integers. $f(x) = \sum_{n \in \mathbb Z} \frac{1}{2^{|n|}}|x-n|$. This should be okay since as $n$ moves away from $0$, we shrink the contributions geometrically. $\endgroup$ Dec 26, 2016 at 5:19
  • $\begingroup$ But sir this is an particular case. and I have also construct a function which is non differentiable at only $\mathbb{Z}$ but continuous every where.....But I cannot extend this construction for the arbitrary set $S$. In fact, for $S={a_1, a_2, a_3,.......}$....!!!! $\endgroup$
    – sigma
    Dec 26, 2016 at 5:24
  • $\begingroup$ Is the construction possible...?? Intuitively it seems to me that this is possible (as u noted in ur first comment) ....but I can't.... $\endgroup$
    – sigma
    Dec 26, 2016 at 5:26

2 Answers 2


The same construction can be generalized. Let $a_1,a_2,\dots$ be the elements of $S$.

Then the function $f(x)=\sum\limits_{n=1}^\infty \dfrac{|x-a_n|}{2^n \max(|a_n|,1)}$ is differentiable at all points except the points at $S$.

  • $\begingroup$ Can you prove it? We can try to work it out if you'd like. $\endgroup$
    – Asinomás
    Dec 26, 2016 at 5:40
  • $\begingroup$ How does the series Converge....?? $\endgroup$
    – sigma
    Dec 26, 2016 at 5:50
  • $\begingroup$ hmm, good point. we'll adjust it. $\endgroup$
    – Asinomás
    Dec 26, 2016 at 5:56
  • $\begingroup$ Please tell how does the series converge now...??@ J F Hidalgo $\endgroup$
    – sigma
    Dec 26, 2016 at 6:04
  • $\begingroup$ well, each summand is smaller than $|\frac{x}{a_n}-1|2^{-n}|$ and $|x-a_n|2^{-n}$. If $|a_n|\leq 1$ it is smaller than $(|x|+1)|2^{-n}$ and otherwise it is smaller than $(|x|+1)2^{-n}$ $\endgroup$
    – Asinomás
    Dec 26, 2016 at 6:08

Yes. See this answer, on the kinds of sets where a function $f$ may fail to be differentiable. Such a set must be a $G_{\delta\sigma}$ set, meaning it must be an infinite, countable union of $G_{\delta}$ sets (which in turn are countable intersections of open sets).

Notice that every closed set is $G_{\delta}$ and in particular, every singleton is $G_{\delta}$:


Hence, every countable set is $G_{\delta\sigma}$: $C=\cup_{n\in\mathbb{N}}\{c_n\}$.

The paper linked in the answer describes how one may construct continuous functions $f:\mathbb{R}\longrightarrow\mathbb{R}$ whose set of non-differentiability is a $G_1\cup G_2$, where $G_1$ is any $G_{\delta}$ and $G_2$ is a $G_{\delta\sigma}$ with measure zero. In particular, notice that every countable set is both $G_{\delta\sigma}$ and has measure zero, so every countable set may be written as some $G_1\cup G_2$.


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.