Problem Let $(a_k)_{k=1}^\infty$ be the sequence of values of $\mathbb{Q}\cap[0,1]$ (which is countable set). Let $g:\mathbb{R}\rightarrow\mathbb{R}$ by $g=\sum_{k=1}^\infty {1 \over k}\cdot \chi_{a_k}$, that is, $g(x)=0$ if $x\notin\mathbb{Q}\cap[0,1]$ and $g(a_k)={1\over k}$ for all $k\in\mathbb{N}$. Show $g$ is Riemann integrable.

(Note: My attempt is below, but I was told that there are many ways to prove this efficiently.)

Attempted Work My main problem is understanding the function $g$. I thought at first glance that it was the harmonic series, which is unbounded and so then not integrable, but I was told to think more carefully, but I'm stumped for some reason. I think it's the way how its defined twice to me: seems contradictory definition. But regardless, I went ahead to try...

Proof so far: Let $g$ be defined as above. For $g(a_k)={1\over k}$ we have that $g$ is bounded with compact support in $(a_k)\subseteq\mathbb{Q}\cap[0,1]$. Now let $g={1 \over k}={1\over k^m}+{k^{m-1}-1\over k^m}$ for $k=2,...,\infty$, where $f_1={1\over k^m}$ is bounded above (by 1) and $f_2={k^{m-1}-1\over k^m}$ is bounded by some $\epsilon_k$>0. Both $f_1$ and $f_2$ have thier support in $(a_k)$. $f_1$ is Riemann integrable (is this circular logic?) and $||f_2||_\infty\leq\epsilon_k$. Hence, by a Theorem about Lower and Upper Riemann Sums provided, $g$ is Riemann integrable. (I know this last part leaves much to be desired)

Lemma provided Basically we were provided a nameless lemma that said if we could show that:

(1) $f_1$ and $f_2$ existed such that $g=f_1+f_2$ with $f_1$ and $f_2$ bounded with support on $(a_k)$ and

(2) $f_1$ is Riemann integrable and $||f_2||_\infty\leq\epsilon$ for all $\epsilon>0$.

Then $g$ is Riemann integrable.

Any tips, advice or help is really appreciated. Thanks so much in advance!


I think it is easier to work from the definition here.

It should be clear that for any partition $\pi$, we have $L(g,\pi) = 0$.

Pick $\epsilon>0$. Choose $N$ such that $\frac{1}{N} < \frac{1}{2} \epsilon$. It should be clear that we can pick a partition $\pi$ of $[0,1]$ such that the points $a_1,...,a_N$ are contained in intervals of total length $\frac{1}{2} \epsilon$.

Let $\{I_k\}$ be the intervals of the partition $\pi$, and let ${\cal I} \subset \{I_k\}$ be the collection of intervals containing any point $a_1,...,a_N$.

Then \begin{eqnarray} U(g,\pi) &=& \sum_k (\sup_{x \in I_k} g(x)) \ l(I_k) \\ &=& \sum_{I \in {\cal I}} (\sup_{x \in I} g(x)) \ l(I) +\sum_{I \in {\cal I}^c} (\sup_{x \in I} g(x)) \ l(I) \\ &\le& \sum_{I \in {\cal I}} (1) \ l(I) + \sum_{I \in {\cal I}^c} (\frac{1}{N} l(I)) \\ & \le& \frac{1}{2} \epsilon + \frac{1}{N} \\ &<& \epsilon \end{eqnarray} It follows that $g$ is integrable and $\int_0^1 g(x)dx = 0$.

Note: You could view the above as writing $g = f_1+f_2$, where $f_1 = \sum_{k=1}^N \frac{1}{k} 1_\{a_k\}$, and $f_2 = g-f_1$. It should be clear that $\|f_2\|_\infty < \frac{1}{2} \epsilon$, however, I think it is simpler to do the above.

  • $\begingroup$ Thanks a lot! I agree that your definition using partitions is probably much simpler. $\endgroup$ Jun 13 '13 at 7:00

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.