We consider only the set $M$ of a.e. essentially locally bounded measurable functions $[0, 1] \to \mathbb R$. Here $m(S)$ denotes the Lebesgue measure of $S$.

Let $f$ be measurable. For every $e$ in $(0, 1]$, by Lusin’s theorem, we can write our measurable function as continuous on $[0, 1]-H$, and horrid on a set $H$ of measure $e$. How does “horrid” vary with $e$?

One way to quantify “horrid” is to ask how discontinuous the function is on $H$. Inspired by this, we calculate the average pointwise oscillation of the function of $H$. Formally this is the integral of the essential oscillation of $f$ on $H$ divided by $m(H)$. Since oscillation is upper semi continuous, it is integrable. Further we take the infimum over all such $H$ of measure less than or equal to $e$.

Thus $$ O(f, e) \mathrel{:=} \inf_{\substack{m(H) \le e,\\ f\in C^0[0, 1] \setminus H}}\left\{\ \frac{1}{m(H)} \int\limits_{x \in H} \lim_{d \to 0}\ \inf_{m(G) = 0} \sup_{\substack{y, z \in B_d (x)\setminus G}} \lvert f(y) - f(z)\rvert\mathrm{d}x\right\}. $$

The end result is that for every $e$, we get a function $O(f): (0, 1] \to [0, \infty) $ describing how horrible the discontinuity behaviour is on the best behaved $H$ we can find.


Call a function $f$ tame if $O(f, e) = 0$ for all $e$. Is it true that a function is tame iff it agrees a.e. with a function that is continuous a.e.?

  • $\begingroup$ Possibly Jack Brown's 1995 survey paper Restriction theorems in real analysis (preprint version here) could be of use, at least in pointing you to possibly relevant literature. $\endgroup$ – Dave L Renfro Mar 20 '19 at 8:57
  • $\begingroup$ I am not sure that I understood the definition, what is $O(f,e)$ for $f=\chi_{(0,1/2]}$? $\endgroup$ – Fedor Petrov Mar 20 '19 at 12:01
  • $\begingroup$ This would be zero for all e, since for any such H with m(H) = e, we have that the oscillation on H is 0 a.e. (everywhere except 0 and 1/2). So the integral is 0. $\endgroup$ – James Baxter Mar 20 '19 at 12:47
  • $\begingroup$ I hate to ask a question so similar to one that was just answered, but I feel that I'm still missing something here (probably just one of those days :) ). If we took $f=\chi_{\mathbb{Q}\cap [0,1]}$, what would we get for $O(f,e)$? $\endgroup$ – Gary Moon Mar 22 '19 at 1:53
  • $\begingroup$ This function is discontinuous everywhere with oscillation 1, so the integral would be 1 as well. $\endgroup$ – James Baxter Mar 22 '19 at 6:51

A counterexample to this problem can be constructed as follows. Take a sequence $(K_n)_{n\in\omega}$ of pairwise disjoint nowhere dense compact sets $K_n\subset[0,1]$ of positive Lebesgue measure $\lambda(K_n)>0$ such that $\sum_{n=0}^\infty\lambda(K_n)=1$. Consider the function $f:[0,1]\to [0,1]$ defined by $$ f(x)=\begin{cases}\frac1{2^n}&\mbox{if $x\in K_n$ for some $n\in\omega$;}\\ 0&\mbox{otherwise}. \end{cases} $$

It is easy to see that the function $f$ is not continuous a.e.

On the other hand, for every $\varepsilon >0$, we can choose $n\in\mathbb N$ so large that $\frac1{2^n}<\varepsilon$ and $\sum_{i>n}\lambda(K_i)<\varepsilon$. Then the set $H=[0,1]\setminus \bigcup_{i\le n}K_i$ has measure $\lambda(H)<\varepsilon$ and $f$ has oscillation $\le \frac1{2^n}<\varepsilon$ at points of the open set $H$ (because $f(H)\subset [0,\frac1{2^n}]$).

| cite | improve this answer | |
  • $\begingroup$ Nice! A well deserved bounty for you sir. $\endgroup$ – James Baxter Mar 24 '19 at 8:51
  • $\begingroup$ Well actually, the oscillation is counted over the whole set, not just on H. But i believe this bug might be fixable? $\endgroup$ – James Baxter Mar 24 '19 at 8:52
  • $\begingroup$ @JamesBaxter Thanks! So, quick reaction! $\endgroup$ – Taras Banakh Mar 24 '19 at 8:52
  • $\begingroup$ @JamesBaxter The set $H$ is open, so the oscillation of $f$ and $f{\restriction}H$ are the same at points of $H$. $\endgroup$ – Taras Banakh Mar 24 '19 at 8:53
  • $\begingroup$ Oh, right I missed that. Very nice.. $\endgroup$ – James Baxter Mar 24 '19 at 8:53

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.