5

Let $f:[a,b] \to \mathbb R$ be a function Riemann integrable over $[a,b]$ . It is known that if we change its values at finitely many points of $[a,b]$ , then the changed function still remains Riemann integrable on $[a,b]$ . My question is , suppose we change its values at countably infinitely many points of $[a,b]$ such that the set of this points is not dense in $[a,b]$ ; then is it true that this changed function is Riemann integrable on $[a,b]$ ?

1 Answers1

7

Let's say $E\subset [a,b]$ is "Riemann unnoticeable" (RU) if the values of any $f\in \mathcal R [a,b]$ on $E$ can be changed arbitrarily and still leave us with a function in $\mathcal R [a,b].$ Claim: $E$ is RU iff $m(\overline E )=0.$

Proof: Suppose $m(\overline E )=0.$ Set $U= [a,b]\setminus \overline E.$ Then $U$ is open in $[a,b].$ Let $f\in \mathcal R [a,b].$ Then $f$ is continuous a.e. in $[a,b]$, hence a.e. in $U.$ Since $U$ is open, that won't change no matter how we change $f$ on $E$ (or $\overline E$ for that matter). Because $U$ has full measure, the changed function is continuous a.e. in $[a,b].$ Thus all such changes leave us in $\mathcal R [a,b],$ hence $E$ is RU.

Suppose $m(\overline E )>0.$ Case 0: $E$ contains an interval. Then obviously $E$ is not RU. Case 1: $E$ contains no interval, but $\overline E$ contains an interval $I.$ Then $E\cap I, I\setminus E$ are both dense in $I.$ Suppose $f \equiv 0.$ Change $f$ to $\chi_E.$ Then $\chi_E$ is discontinuous at each point of $I,$ a set of postitive measure. Thus $\chi_E \not \in \mathcal R [a,b].$ Hence $E$ is not RU. Case 2: $\overline E$ contains no interval $I.$ Again set $U= [a,b]\setminus \overline E.$ Then $U$ is dense in $[a,b].$ Take $f \equiv 0$ and change it to $\chi_E.$ Since each $x \in \overline E$ is accessible from $E$ and from $U,$ $\chi_E$ is discontinuous at $x.$ Hence $E$ is not RU.

zhw.
  • 105,693
  • 2
    That's one of the main results in the theory of Riemann integration: If $f$ is bounded on $[a,b],$ then $f$ is Riemann integable on $[a,b]$ iff $f$ is continuous at a.e. $x\in [a,b].$ – zhw. Aug 21 '15 at 17:11
  • 1
    Frankly I never saw such a fine statement: can you give a reference ? – Tony Piccolo Aug 22 '15 at 13:32
  • This is Lebesgue's theorem on Riemann integration; we don't really need much measure theory for this - just measure $0$ stuff. This is discussed here for instance (click "show" for an outline of the proof). https://en.wikipedia.org/wiki/Riemann_integral#Integrability – zhw. Aug 22 '15 at 15:57
  • I referred to your answer, not to your comment. – Tony Piccolo Aug 22 '15 at 16:26
  • In that case my reference is my answer. I had never thought about this before. – zhw. Aug 22 '15 at 16:36
  • I was sure. It seems to me you are a professional, so I profit $\dots$ How is a similar statement for the Lebesgue integral ? – Tony Piccolo Aug 22 '15 at 16:42
  • 1
    $E$ is LU iff $m(E) = 0.$ – zhw. Aug 22 '15 at 16:52
  • Why don't you consider the closure of $E$ ? – Tony Piccolo Aug 22 '15 at 16:54
  • The Lebesgue theory makes this all clear. – zhw. Aug 22 '15 at 17:12
  • 2
    Another way of viewing this proof: if you modify a function on $E$, only elements of $\overline{E}$ can leave the set of continuity points. That and the Lebesgue criterion give one direction. For the other direction you basically have to prove that there is some modification on $E$ such that all members of $\overline{E}$ do leave the set of continuity points. – Ian Aug 22 '15 at 21:08
  • Just to be clear (since I barely understand concept of measure). What you're saying is $f \in R[a,b]$ with $\int_a^b f =I$ if and only if the set of discontinuities of $f$ on $[a,b]$ has measure $0$. But you don't comment anything on the value of integral? Does it change or not? For example, if I change values of $f$ at countably infinite points, then what I understand from this answer is $f$ remains Riemann integrable but the what about value of integral? Would it it still be equal to $I$? – William Aug 27 '22 at 09:36
  • 1
    @William No, changing $f$ at countably many points can change whether it's Riemann integrable. (This follows from the result above, since a countable set can have a closure of positive measure...) – David C. Ullrich Aug 27 '22 at 12:26