Can someone provide an example of $X$ being a non-continuous random variable with continuous cumulative distribution function?

For instance:

$X$ is discrete if it takes (at most) a countable number of values.

$X$ is continuous (or absolutely continuous) if its law $P^X$ admits a density $f(x)$.

Note: A random variable don't have to be necessarily discrete or continuous; just take a cumulative distribution function that is non-constant and continuous except in $0$. Then $X$ is neither continuous nor discrete.

I know that to ensure that $X$ is continuous, we need to ask $F_X \in C^1$, as $F_X \in C^0$ does not suffice.

I would then like to see a non continuous random variable with continuous cdf


A simple example is $X$ uniformly distributed on the usual Cantor set, in other words, $$X=\sum_{n\geqslant1}\frac{Y_n}{3^n},$$ for some i.i.d. sequence $(Y_n)$ with uniform distribution on $\{0,2\}$.

Other examples are based on binary expansions, say, $$X=\sum_{n\geqslant1}\frac{Z_n}{2^n},$$ for some i.i.d. sequence $(Z_n)$ with any nondegenerate distribution on $\{0,1\}$ except the uniform one.

These distributions have no density with respect to Lebesgue measure, even partly, since $P(X\in C)=1$ for some Borel set $C$ with zero Lebesgue measure. They have no atom either, in the sense that $P(X=x)=0$ for every $x$.

  • $\begingroup$ This seems awesome. Can you add at least an hint (or a link) on how to compute their cdf, though? :) $\endgroup$ – Ant Jul 23 '14 at 16:36
  • $\begingroup$ Which one? // Have you no idea how to do it? $\endgroup$ – Did Jul 23 '14 at 18:37
  • $\begingroup$ For what concerns $X$: since $Y_n$ can only take the values $0, 1, 2$, it means that at every addition we make we are going to specify one of the $3^n$ partition of the cantor set. (of course, considering only $n$ cuts). As $n \to \infty$, the probability that $X$ lies in the cantor set is $1$, but it is known that the cantor set has measure $0$, so $X$ is not continuos. Why is $P(X=x) = 0$ though? $\endgroup$ – Ant Oct 4 '14 at 18:19
  • 1
    $\begingroup$ Let me suggest to revise the construction of the Cantor set, which has 2^n parts, not 3^n, after n cuts. Relatedly, Y_n is never 1, only 0 or 2. $\endgroup$ – Did Oct 4 '14 at 19:16
  • $\begingroup$ Oh right I got confused, thanks! . Can you offer an hint for $P(X=x)=0$? $\endgroup$ – Ant Oct 4 '14 at 21:10

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.