I have been doing some excercises on total variation when the following questions came up to my mind:

(1) Let $f$ be continuous on the interval $[0,1]$ and be of bounded variation. Is it true that its total vatiation function $TV(f_{[0,x]})$ is uniformly continuous? i.e. Is it true that for $\forall\space\epsilon>0$, $\exists\space\delta>0$, such that for arbitrary interval $[a,b]$ with $|b-a|<\delta$, we have $TV(f_{[a,b]})<\epsilon$?

(2) Alternatively, if it is not uniformly continuous, can I say that for $\forall\space{}x\in[0,1]\text{ and } \forall\space\epsilon>0$ , $\exists\space\text{ nondegenerate interval }I \text{ such that }x\in{}I\text{ and } TV(f_{I})<\epsilon$?

Thank you!

  • 1
    $\begingroup$ Hint: What is the total variation function of a monotonous function? $\endgroup$ – Did May 12 '12 at 11:46
  • $\begingroup$ @Didier I think it should be the absolute value of the difference of end point values? But I still don't see how I can use that on this question... $\endgroup$ – Vokram May 12 '12 at 12:16
  • $\begingroup$ Hence, for example, if $f(0)=0$ and $f$ is nondecreasing and if $g(x)=$TV$(f_{[0,x]})$, then $g=\ldots$ $\endgroup$ – Did May 12 '12 at 12:53
  • $\begingroup$ @Didier yeah in that case $g$ would be continuous. Hmm but sorry if I'm getting your point really slow here, I did not assume f is monotone in the question. Are you saying that I should use Jordan's decomposition and express it as the difference of two such functions? But I think that way I will get an inequality instead of an equality and there's no guarantee they are continuous any more $\endgroup$ – Vokram May 12 '12 at 13:10
  • $\begingroup$ This shows is that the regularity of $x\mapsto$TV$(f_{[0,x]})$ is at most the regularity of $f$, in general. $\endgroup$ – Did May 12 '12 at 13:31

Since a continuous function on a compact set is uniformly continous, you're actually asking whether the total variation of a continuous function is continuous.

Let $\epsilon\gt0$ and $x_0\in(0,1]$ be given. Since the total variation is non-decreasing, it suffices to find a point $x\lt x_0$ with $TV(f_{[x,x_0]})\lt\epsilon$ to show that the total variation is left-continuous, and thus by symmetry also right-continuous and hence continuous.

Pick some point $x_1\lt x_0$. By definition there is a partition of $[x_1,x_0]$ such that the sum of absolute differences of function values over the partition is within $\epsilon/2$ of $TV(f_{[x_1,x_0]})$. Since $f$ is continuous, we can find a point $x$ between the last intermediate point of the partition and $x_0$ such that $|f(x)-f(x_0)|\lt\epsilon/2$. Refining the partition with this point doesn't decrease its sum of absolute differences. Now the sum of absolute differences in the partition up to $x$ is within $\epsilon/2+\epsilon/2=\epsilon$ of $TV(f_{[x_1,x_0]})$, and thus so is $TV(f_{[x_1,x]})$; hence $TV(f_{[x,x_0]})\lt\epsilon$ as required.

  • 3
    $\begingroup$ Nice! However, I had hard time understanding the last sentence, so maybe it is worth to supply a few more details as follows: Let $V$ denote the variation over the partition chosen within $\epsilon/2$ of $TV(f_{[x_1,x_0]})$ as described above. Then $$TV(f_{[x_1,x_0]})<V+\epsilon/2.$$ Also, $V-|f(x)-f(x_0)|$ is some variation over interval $[x_1,x]$, so we have: $$TV(f_{[x_1,x]})\ge V-|f(x)-f(x_0)|.$$ Finally, $TV(f_{[x_1,x_0]})=TV(f_{[x_1,x]})+TV(f_{[x,x_0]})$, so that: $$TV(f_{[x,x_0]}) = TV(f_{[x_1,x_0]}) - TV(f_{[x_1,x]})<(V+\epsilon/2) - (V-|f(x)-f(x_0)|)<\epsilon/2+\epsilon/2=\epsilon.$$ $\endgroup$ – mathreader Jul 7 '13 at 0:09
  • 1
    $\begingroup$ Perhaps the downvoter would care to explain the downvote? $\endgroup$ – joriki Jul 27 '16 at 22:55
  • $\begingroup$ @mathreader actually $V-|f(x)-f(x_0)|$ is no more than some variation over $[x_1, x]$. But the proof follows the way you said. +1 for the comment $\endgroup$ – Filburt Feb 19 '17 at 20:31

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.