Let's consider the interval $[0,1]$ in the same way that we constructed the Cantor set, we can use the same idea, but instead of removing in the step $n$ middle open intervals of length $\frac{1}{3^n}$ we remove of length $\frac{1}{5^n}$. Call each step of this construction $C_n$ i.e

$C_0 = [0,1]$

$C_1= C_0-\left(\frac{1}{2}-\frac{1}{10},\frac{1}{2}+\frac{1}{10}\right)=\left[0,\frac{2}{5}\right]\cup \left[\frac{3}{5},1\right]$

And then to construct $C_2$ consider the middle point of $\left[0,\frac{2}{5}\right]$ which is $\frac{2}{10}$ , and remove from that interval the open interval $\left(\frac{3}{20}-\frac{1}{2\cdot 5^2},\frac{3}{20}+\frac{1}{2\cdot 5^2}\right)$ , the same with the interval $\left[\frac{7}{10},1\right]$ .

(It's the same as the usual Cantor construction) So we define $ C = \bigcap C_n$. Let's define the function $f_n$ as the characteristic function of the set $C_n$ , and $f$ as the characteristic function of the set $C$.

Prove that $$ \mathop {\sup }\limits_{m \geqslant n} \int\limits_0^1 {\left| {f_n \left( x \right) - f_m \left( x \right)} \right|\,\mathrm dm} \leqslant \frac{1} {3}\left( {\frac{2} {5}} \right)^n $$ and then prove that under any modification of the function $f$ on a set of measure zero, is not Riemann integrable.

I'm very scared of this problem, I don't know how to prove the inequality please help me, I'm a little dizzy . Thanks for the EDIT


Let $\{I_{k,j}:j=1,\ldots,2^k\}$ be the set of open intervals we are throwing away on $k$-th step of our construction. Note that $m(I_{k,j})=5^{-k-1}$ for all $k\in\mathbb{N}$ and $j=1,\ldots,2^k$.

Let $$ C_0:=[0,1]\qquad\qquad C_{k+1}:=C_k\setminus\bigcup\limits_{j=1}^{2^k}I_{k,j} $$ Now we proceed to the proof of inequality. Define $f_n=\chi_{C_n}$, then for $m\geq n$ we have $$ \int_0^1|f_m(x)-f_n(x)|dm= \int_0^1|\chi_{C_m}(x)-\chi_{C_n}(x)|dm= \int_0^1\chi_{C_n\setminus C_m}(x)dm= m(C_n\setminus C_m)= m\left(\bigcup\limits_{k=n}^{m-1}\bigcup\limits_{j=1}^{2^k} I_{k,j}\right)= \sum\limits_{k=n}^{m-1}\sum\limits_{j=1}^{2^k}m(I_{k,j})= \sum\limits_{k=n}^{m-1}\sum\limits_{j=1}^{2^k}5^{-k-1}= \sum\limits_{k=n}^{m-1}2^k5^{-k-1} $$ Hence $$ \sup\limits_{m\geq n}\int_0^1|f_m(x)-f_n(x)|dm= \sup\limits_{m\geq n}\sum\limits_{k=n}^{m-1}2^k5^{-k-1}= \sum\limits_{k=n}^\infty 2^k5^{-k-1}=\frac{2^n 5^{-n-1}}{1-2\cdot 5^{-1}}=\frac{1}{3}\left(\frac{2}{5}\right)^n $$

We proceed to the second part. Define $$ C:=\bigcap\limits_{k=1}^\infty C_k $$ Its measure equals $$ m(C)= 1-m(C_0\setminus C)= 1-m\left(\bigcup\limits_{k=0}^\infty\bigcup\limits_{j=1}^{2^k}I_{k,j}\right)= 1-\sum\limits_{k=0}^\infty \sum\limits_{j=0}^{2^k} m(I_{k,j})=\\ 1-\sum\limits_{k=0}^\infty \sum\limits_{j=0}^{2^k} 5^{-k-1}= 1-\sum\limits_{k=0}^\infty 2^k 5^{-k-1}= 1-\frac{5^{-1}}{1-2/5}=\frac{2}{3} $$ Now note that $x\in C$ iff its representation in a fivefold number system have no digit $2$ at any place. Hence for a given $x\in C$ you can change some digit to $2$ to leave the set $C$. Moreover you can change digit with very big place number to get a number outside $C$ but very close to $x$. Thus for each $x\in C$ and each $\varepsilon>0$ there exist $\hat{x}\notin C$. This is equivalent to discontinuity of $f$ in point $x\in C$. So $f$ is discontinuous at any point of $C$.

Since $m(C)>0$ by Lebesgue critirion $f\notin\mathcal{R}([0,1])$, because set of discontinuities of $f$ have positive measure. Let $\hat{f}$ be a function equivalent to $f$. Assume $\hat{f}$ is Riemann interable, then it is bounded. Hence the function $\delta=\hat{f}-f$ is bounded. Since $\hat{f}$ is equivalent to $f$, then $\delta$ is non zero only on the set of zero measure. By Lebesgue criterion $\delta\in\mathcal{R}([0,1])$. Then $f=\hat{f}-\delta\in$$\mathcal{R}([0,1])$ as difference of two Riemann integrable functions. Contradiction, so $\hat{f}\notin\mathcal{R}([0,1])$

  • $\begingroup$ Thanks! your answer is so clear. $\endgroup$ – Gaston Burrull May 22 '13 at 18:08
  • $\begingroup$ @GastónBurrull, not at all :) $\endgroup$ – Norbert May 22 '13 at 18:34

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.