While studying the space $\mathbb{D}[0,1]$ of right continuous functions with left hand limits (i.e. càdlàg functions) on $[0,1]$, I came across the following theorem:

Theorem. If $f$ is càdlàg on $[0,1]$, it is bounded.

My proof attempt: I am aware that if a function has both left and right hand limits on $[0,1]$, then the set of discontinuities is at most countable. Hence I tackled this in two parts, one where the discontinuities are finite, the other infinite.

I got the finite discontinuity one. But I am stuck at the infinite discontinuity part. My guess is that there is something special about the countable discontinuities (e.g. they cannot be dense in $[0,1]$) and somewhere, I'll have to use the sequential compactness property to get a contradiction, but I am unable to collect my ideas.

I request any starting hints on this. A sketch of the proof would also be appreciated.

  • 2
    $\begingroup$ Hint: a locally bounded function is bounded on compact sets. $\endgroup$ – 23rd Jul 16 '13 at 9:54

Billigsley gives an excellent one-line proof. I am repeating it here. First note that this Lemma is true :

For every $\epsilon >0$, $ \exists$ partition $ 0=t_0<t_1<\ldots,t_k=1$ such that for the set $S_i = [t_i,t_{i+1})$, we have $\sup_{s,t \in S_i} |f(s)-f(t)| <\epsilon$ for all $i$.

This Lemma is easily proved. Once Lemma is seen to be true, choose an $\epsilon >0$. Then the bound on $f$ is simply $(\sum_{l=1}^{k} J_{l} ) + \epsilon \times k$ (first term is the amount of jumps which occur at the $k$ points which were obtained using the Lemma, while the second term is the increment of $f$ within these intervals.


Thanks for your help everyone. I think I got it. Kindly check my proof for errors if any:

Proof: $\forall x \in [0,1]$, fix $\epsilon > 0$. Now because at every point, both left and right hand limits exist(except at $0$ and $1$, in which case only one of the following will be there),

$$\exists \delta_x,\eta_x > 0$$ $$x_0 \in (x-\delta_x,x) \Rightarrow |f(x_0) - f(x^-)| < \epsilon$$ $$x_0 \in (x,x+\eta_x) \Rightarrow |f(x_0) - f(x)| < \epsilon$$

where $f(x^-)$ is the left hand limit of the function at x and since we have right continuity, $f(x^+) = f(x)$.

Let $O_x = (x-\delta_x,x)$ and $V_x = (x,x+\eta_x)$, $$\bigcup_{x \in [0,1]} O_x \cup V_x$$ is an open cover for $[0,1]$. Since $[0,1]$ is compact, it has a finite subcover. Hence $$ [0,1] \subseteq \left(\bigcup_{k=1}^n O_{x_k}\right) \cup \left(\bigcup_{k=n+1}^m V_{x_k}\right)$$

For $k=1...n$, if $x \in O_{x_k}$, then $|f(x)-f(x^-_k)| < \epsilon \Rightarrow |f(x)| < |f(x^-_k)| + \epsilon$

If $M_1 = \max_{k=1...n} |f(x_k^-)|$, then $|f(x)| < M_1 + \epsilon$. (Only if $x$ belongs to any of the "$O$" balls)

Similarly, by defining $M_2 = \max_{k=n+1...m} |f(x_k)|$, and following similar steps, we get $|f(x)| < M_2 + \epsilon$. (For x belonging to "$V$" balls).

So the bound for $x \in [0,1]$ is, $$|f(x)| < \max(M_1,M_2) + \epsilon$$ Thus f is bounded. $\blacksquare$

I am interested in other elegant proofs if there are any. Thanks again.


If $f$ is unbounded, there will be an infinite sequence of points $(x_n)_{n\in\mathbb{N}}$ such that $\lim_{n\to\infty}x_n=x\in[0,1]$ and $\lim_{n\to\infty}f(x_n)=\infty$. Split the subsequence into two, one on the left and one on the right of $x$. One of them must be an infinite sequence. Suppose both of them are infinite (the other cases are similar). Call them $(u_n)_{n\in\mathbb{N}}$ and $(v_n)_{n\in\mathbb{N}}$ respectively. Then $u_n\to x^-,\ v_n\to x^+$ and $\lim_{n\to\infty}f(u_n)=\lim_{n\to\infty}f(v_n)=\infty$, but this is impossible because càdlàg functions have left and right limits everywhere.

  • $\begingroup$ Thanks for editing. And for this useful hint. $\endgroup$ – Gautam Shenoy Jul 16 '13 at 11:19
  • $\begingroup$ You probably meant to write f(x) instead of just $x$, is that correct? $\endgroup$ – JohnK Jul 12 at 13:36
  • $\begingroup$ @JohnK I meant that both $f(u_n)$ and $f(v_n)$ tend to infinity. $\endgroup$ – user1551 Jul 12 at 14:14
  • $\begingroup$ Thanks for the correction. $\endgroup$ – JohnK Jul 12 at 15:22

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.