6
$\begingroup$

I was wondering whether anyone would be so kind as to briefly check my proof? I am supposed to prove the statement without using any theorems which would render the proof trivial.

If $\displaystyle\int_a^b f$ exists then $f$ has infinitely many points of continuity in $[a,b]$

For the sake of contradiction suppose $f$ has finitely many. It suffices to show that there is an interval $[u,v]\subset [a,b]$ over which $f$ is not Riemann integrable. Pick any $[u,b]\subset [a,b]$ such that $f$ is discontinuous everywhere in $[u,v]$. For any finite partition $D=\{x_1\cdots x_n\}$ of $[u,v]$ let:

$$s(f,D)=\sum_i (x_{i+1}-x_i)\inf_{x\in[x_{i},x_{i+1}]}f\quad\text{and}\quad S(f,D)=\sum_i (x_{i+1}-x_i)\sup_{x\in[x_{i},x_{i+1}]}f$$ We prove $\sup_D s(f,D) <\inf_D S(f,D)\;(1)$. To achieve this consider an arbitrary chain: $$D_0\subset D_1\subset \cdots$$ Pick any $I_0\subset I_1\subset \cdots$ where $I_k=[u_k,v_k]$ and $I_k$ is an interval of the partition $D_k$. Without loss of generality $\bigcap_i I_i=z$. Given that $f$ is discontinuous everywhere, there exists $\epsilon>0$ such that for any $\delta>0$ there exists $y$ such that $|z-y|<\delta$ yet $|f(z)-f(y)|\geq \epsilon$.

This immediately implies $\displaystyle\sup_{I_n} f>\inf_{I_n} f$ and hence $\displaystyle\sup_{D_n} s(f,D)<\inf_{D_n} S(f,D)$.

This is valid for any sequence $D_n$ so claim $(1)$ holds, contradicting the integrability of $f$.

$\endgroup$
4
  • $\begingroup$ I wouldn't write "$|z-y|<\delta \implies$" rather, "$|z-y|<\delta$, yet...". Also "This immediately implies..." (Why?) $\endgroup$
    – Pedro Tamaroff
    May 1 '14 at 0:27
  • $\begingroup$ @PedroTamaroff Thanks, adjusted. Is the argument reasonable otherwise? $\endgroup$ May 1 '14 at 0:29
  • $\begingroup$ You want to take $I_0\supset I_1\supset\cdots$; too. You wrote without loss of generality $\bigcap I_i=\color{red}{\{}z\color{red}{\}}$. This seems unnecessary, it suffices there is one there. $\endgroup$
    – Pedro Tamaroff
    May 1 '14 at 0:52
  • $\begingroup$ A function is Riemann integrable iff it's set of discontinuity points has zero measure. $\endgroup$ May 1 '14 at 2:50
3
$\begingroup$

One can also give a direct proof. Suppose $f:[0,1]\to \Bbb R$ is Riemann integrable.

Lemma Suppose that for a partition $P$; $\displaystyle D(f,P)=\sum_{k=0}^n (M_k-m_k)\Delta x_k<\varepsilon$. Then $M_k-m_k<\varepsilon $ for some $k$.

Proof Else $\displaystyle\varepsilon =\sum_{k=0}^n \varepsilon\Delta x_k\geqslant \sum_{k=0}^n (M_k-m_k)\Delta x_k$, proving the contrapositive.

Obs More generally, we get $M_k-m_k<(b-a)\varepsilon$ if the interval of integration is $[a,b]$.

We proceed to the proof of your claim.

Notation For a set $S$, we denote $o(f,S)=\sup\limits_{t,t'\in S}|f(t)-f(t')|$.

With this lemma at hand, we may start rolling: take $\varepsilon =1$. We know there exists a partition $P_1=\{0=x_0,x_1,\ldots,x_n=1\}$ with $D(f,P_1)<1$, since $f$ is Riemann integrable. By the lemma, there is a subinterval $I_i=[x_{i-1},x_i]$ where $o(f,I_i)<1$. If $i\neq 1$ or $n$, let $K_1=I_i$. Else, pick $x_0<x_0'<x_1$ or $x_{n-1}<x_n'<x_n$, and replace this in the endpoints of $K_i$. Now $f$ is integrable over $K_1$, so there exists a partition $P_2$ for which the sum in $K_1$, $D(f,P_2)<\frac 1 2$, which gives some subinterval $K_2$ where $o(f,K_2)<\dfrac 1 2|K_1|<\dfrac 1 2 $, since $|K_1|<1$. Again, we replace endpoints if necessary. Since we're shrinking the domain, in any case the oscillation descreases, so we're safe. Inductively, we can construct a sequence of strictly nested closed bouned intervals $$K_1\supsetneq K_2\supsetneq K_2\supsetneq\cdots$$ where $o(f,K_i)<\frac 1 i$. By Cantor, there is $x_0\in\bigcap\limits_{j\geqslant 1} K_j$. I claim $f$ is continuous at $x_0$. (Proof?)

Now consider $[0,x_0],[x_0,1]$. Note that by construction $x_0\neq 0,1$, so repeating this on $[0,x_0]$ we obtain $x_1\neq x_0$ where $f$ is continuous, and so on. Using this, we may also show

Claim The set where $f$ is continuous is dense in $[0,1]$.

$\endgroup$
1
  • 3
    $\begingroup$ For the interval $[a,b]$, don't we get $M_k-m_k<\frac{\epsilon}{(b-a)}$ in the lemma? $\endgroup$ Sep 8 '16 at 9:52

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.