Definitions and some motivation:

Let $\mathcal B$ be the set of bounded measurable functions from $[0, 1]$ to $\mathbb R$. Denote by $\mathcal N$ the set of measurable subsets of $[0, 1]$ with Lebesgue measure $0$.

Given a function $f \in \mathcal B$, define the function $\mathcal Of$ by

$\mathcal Of(x) := \inf_{N \in \mathcal N} \lim_{\delta \to 0} \sup_{y, z \in B_\delta (x) \setminus N}\ |f(y) - f(z)|$.

Thanks to Lusin’s theorem, we know that we can modify $f$ on an arbitrarily small set and get a continuous function, and so we force the oscillation to be $0$ everywhere. But can we force it to be whatever we want, and in a controlled way?


Does there exist, for any given $f, g \in \mathcal B$ and $\varepsilon > 0$, a function $f’ \in \mathcal B$ such that the following conditions are satisfied?

i) $f’ = f$ everywhere except for a set of measure at most $\varepsilon$.

ii) $\mathcal Of’ = \mathcal Og$.

iii) $\mathcal O(f’ - f) \leq |\mathcal O(g) - \mathcal O(f)|$.

Remark: Condition (iii) intuitively says that we do not modify the values of $f$ any more than strictly necessary.

Note: All functions are genuine functions and not equivalence classes modulo null sets of such.

  • $\begingroup$ Yes, that’s true! But I think you can modify $f$ on a small (less than measure $\varepsilon$) interval around $0$ to achieve any value of $\mathcal Of’(0)$. Of course this is only at one point though.. $\endgroup$
    – Nate River
    Apr 19 '21 at 7:31
  • $\begingroup$ Yes Martin Hairer, but I don’t see immediately how this answers to the OP. $\endgroup$ Apr 19 '21 at 7:32
  • $\begingroup$ I don’t think it’s meant to be an answer... @A.DellaCorte $\endgroup$
    – Nate River
    Apr 19 '21 at 7:33

Let $f,g: [0,1] \to \mathbf{R}$ be two given, bounded measurable functions and $\epsilon > 0$ be an arbitrary constant. By Lusin's theorem there exist continuous functions $F,G: [0,1] \to \mathbf{R}$ so that together \begin{equation} \lvert \{ F \neq f \} \rvert + \lvert \{ G \neq g \} \rvert < \epsilon. \end{equation}

In other words $F - f$ and $G - g$ are zero except on a set of measure at most $\epsilon$. Note also that $\mathcal{O}(G) = \mathcal{O}(F) = 0$ by continuity, so they can be added and subtracted with impunity. Therefore we may set \begin{equation} f' = F + g - G = f + (F - f) + (g - G). \end{equation}

This function has

  • $\mathcal{O}(f') = \mathcal{O}(F + g - G) = \mathcal{O}(g)$ because $F,G$ are continuous,
  • $\{ f' \neq f \} \subset \{ g \neq G \} \cup \{ f \neq F \}$ so $\lvert \{ f' \neq f \} \rvert < \epsilon$,
  • $\mathcal{O}(f - f') = \mathcal{O}(f - F - g + G) = \mathcal{O}(f - g)$ and taking absolute values this is bounded by $\lvert \mathcal{O}(f) - \mathcal{O}(g) \rvert$ by the triangle inequality.
  • $\begingroup$ Very slick solution! Thank you. $\endgroup$
    – Nate River
    Apr 19 '21 at 8:47

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.