Let $f:[a,b]\to\mathbb{R}$ be continuous except at $\{x_1,x_2,\cdots,x_n,\cdots\}$, $g:[0,1]\to[a,b]$ be a continuous function. My question is: can the points of discontinuity of $f\circ g$ have full measure?

The set of points of discontinuity of $f\circ g$ is included in $\bigcup^\infty_{n=1} \partial g^{-1}(x_n)$, where $\{\partial g^{-1}(x_n)\}_{n\ge 1}$ is a countably infinite collection of disjoint closed sets with empty interior. I'm well aware that a subset of $[0,1]$ with full measure can be written as a countable union of closed sets with empty interior (there are many simple constructions), but I don't know if there is a subset of $[0,1]$ with full measure that is a countable union of disjoint closed sets with empty interior. Also, even there is a subset $E$ of $[0,1]$ with full measure such that $E = \bigcup^\infty_{n=1} C_n$, where $\{C_n\}$ are disjoint closed sets with empty interior, can we find $\{x_1,x_2,\cdots,x_n,\cdots\}$ and define a continuous function $g$ such that $g^{-1}(x_n) = C_n$?

Edit: Note that $f\circ g$ cannot be a.e. discontinuous if $f$ has only finitely many points of discontinuity, since $\bigcup^{N}_{n=1} \partial g^{-1}(x_n)$ would be closed and, by Baire category theorem, have empty interior, so it cannot have full measure. On the other hand, the set of points of discontinuity of $f\circ g$ can have measure arbitrarily close to 1, even if $f$ is just discontinuous at one point.


1 Answer 1


Here is an example:

We use an auxiliary continuous function $\chi: [0, 1] \to [0,1]$ such that $\chi(0) = \chi(1) = 0$, $\chi(1/2) = 1$, and $\chi^{-1}(\mathbb{Q})$ has full measure.

To construct $g: [0,1] \to [0,1]$, take a map $t: \mathbb{Z}_+ \to \mathbb{Q} \cap [0,1]$ such that each rational in $[0,1]$ appears in the image infinitely many times. Consider $$g = \sum_{n = 1}^{\infty} c_n\chi_{I_n}$$ where $\chi_{I_n}$ is a scaling of $\chi$ function such that its support is $I_n =[t_n - \ell_n, t_n + \ell_n]$, and $c_n \in \mathbb{Q}^+$ with $c_n \leq 2^{-n}$ and $\ell_n \leq 2^{-n}q_n^{-3}$, $q_n$ being the denominator of $t_n$. Then by uniform convergence, $g$ is continuous. Furthermore, $g$ has the following property:

  1. By choosing $c_n$ appropriately, $g(\mathbb{Q})$ can be disjoint from $\mathbb{Q}$. To achieve this, for each $q \in \mathbb{Q} \cap [0,1]$, let $n \in S_q$ be the set of $n$ such that $I_n$ is centered at $q$. $q$ lies in the intervals $\{I_n\}_{n \in S_q}$ plus finitely many other intervals $\{J_m\}$. Thus, it suffice to take the $c_n$ for $n \in S_q$ such that $$\sum_{n \in S_q} c_n$$ is outside the $\mathbb{Q}$-span of $\{1, \chi_{J_m}(q)\}_m$.

  2. $g^{-1}(\mathbb{Q})$ has measure $1$, as each number in its complement must either appear in infinitely many $I_n$, or is in $\chi_{I_n}^{-1}(\mathbb{R} \backslash \mathbb{Q})$ for some $n$, both of which have measure zero.

Now let $f(p/q) = 1/q$ for any $p/q \in \mathbb{Q}$, and let $f \equiv 0$ outside $\mathbb{Q}$. Then $f$ is continuous except on $\mathbb{Q}$, but $f \circ g$ is discontinuous on $g^{-1}(\mathbb{Q})$.

Note: $\chi$ itself can be constructed similar to the cantor set. Given a tenary representation $0.a_1a_2\cdots$ of a number in $[0,1]$, let $i$ be the first such that $a_i = 1$, then let $$\eta(a) = \sum_{j = 1}^{i - 1} 2^{-j - 1}a_j + 2^{-i}.$$ Then $\eta(0) = 0, \eta(1) = 1$, and $\eta^{-1}(\mathbb{Q})$ is full measure. Glue two copies of $\eta$ together to get $\chi$.

  • $\begingroup$ Thanks for your answer! What I don't stand is that $x$ appears in only finitely many $I_n$ implies that $g(x)\in A$. $\chi_n(x)$ can be strictly between $(0,1)$, right? So I don't understand why should $g(x)\in A$ be true in this case. The same applies for the property 1, $q\in \mathbb{Q}\cap [0,1]$ appears in infinitely many $[t_n-2^{-n-1},t_n+2^{-n-1}]$ does not seem to guarantee that $g(q)$ is not of the form $\frac{k}{2^l}$ since $\chi_n(q)$ can be strictly between $(0,1)$. What am I missing here? $\endgroup$ May 9 at 5:04
  • $\begingroup$ I have edited the answer to address these issues. $\endgroup$ May 9 at 14:14
  • $\begingroup$ Bravo! Brilliant answer indeed. Just one comment: is it better to change to exponent in $q^{-2}_n$ to $q^{-(2+\varepsilon)}_n$, to guarantee that the numbers appear in infinitely many intervals have zero measure? Because I've heard that, for any irrational $x$ there exists infinitely many $p,q$ such that $|x - p/q|<1/q^2$. $\endgroup$ May 10 at 12:54
  • $\begingroup$ @JianingSong . A stronger result is there are infinitely many $p,q\in\Bbb Z$ such that $|x-p/q|<1/(q^2\sqrt 5)$, which is the best possible, in the sense that in the case $x=(1+\sqrt 5)/2$, if $k>\sqrt 5$ then there are only finitely many $p,q\in \Bbb Z$ with $|x-p/q|<1/(kq^2)$. $\endgroup$ May 14 at 19:27

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.