Suppose that $f:[0,1] \to \mathbb{R}$ is defined by $f(x) = 1$ when $x = \frac{1}{n}$ for some positive integer $n$ and $f(x) = 0$ otherwise. How can I prove that $f$ is Riemann integrable on $[0,1]$?

  • 3
    $\begingroup$ Cover each $x = \frac1n$ by interval of size $\frac{\epsilon}{2^n}$ i.e. consider intervals around $\frac1n$ as $\left(\frac1n - \frac{\epsilon}{2^{n+1}}, \frac1n + \frac{\epsilon}{2^{n+1}} \right)$ $\endgroup$ – user17762 Dec 9 '11 at 2:51
  • $\begingroup$ @Sivaram Perhaps I am missing something, but I am unsure how to proceed from your hint. Doesn't Riemann integration consider only finite partitions of $[0,1]$? $\endgroup$ – Srivatsan Dec 9 '11 at 2:55
  • 2
    $\begingroup$ @Srivatsan: This should go with the previous hint. Choose $N$ such that $\frac1N < \frac{\epsilon}{2}$. For all $n < N$, cover each $x = \frac1n$ by interval of size $\frac{\epsilon}{2^{n+1}}$ i.e. consider intervals around $\frac1n$ as $\left(\frac1n - \frac{\epsilon}{2^{n+2}}, \frac1n + \frac{\epsilon}{2^{n+2}} \right)$ and for the rest use an interval of the form $[0,\epsilon/2)$ $\endgroup$ – user17762 Dec 9 '11 at 2:59
  • $\begingroup$ You need to use these intervals to show that the set of discontinuities has measure zero. math.ncku.edu.tw/~rchen/Advanced%20Calculus/… $\endgroup$ – Henry Shearman Dec 9 '11 at 3:00

Here's a more elementary argument. We show that for every tagged partition $P$ of $[0,1]$ of norm at most $\varepsilon \gt 0$, the lower and upper Riemann sums $L(f; P)$ and $U(f;P)$ of $f$ satisfy $$ 0 \leqslant L(f, P) \leqslant U(f;P) \leqslant 2\sqrt{\varepsilon} . \tag{1} $$ The Riemann integrability of $f$ follows as a consequence. The lower bound is obvious; we show only the upper bound of $O(\sqrt{\varepsilon})$ here.

Fix a parameter $t \in (0,1)$ (to be determined later). Without loss of generality With a modest loss in generality, assume that $t$ is the endpoints of two intervals in $P$. [I leave it to the OP to remove this assumption.] The strategy is to bound the contributions of the two types of intervals individually:

  • Those that fall to the left of $t$ (i.e., those that are a subset of $[0, t]$). The contribution from these intervals to the Riemann sum is at most $t$, the total length of the left part (since the function is always bounded above by $1$ in magnitude).

  • Those that fall to the right of $t$ (i.e., those that are a subset of $[t, 1]$). Of all such intervals, at most $1/t$ of them contain a point of the form $1/n$. Therefore, all but $1/t$ of them contribute zero to the Riemann sum. Since each of these intervals is of length at most $\varepsilon$, the total contribution of these to the Riemann sum is at most $\varepsilon \cdot \frac{1}{t} = \frac{\varepsilon}{t}$.

Adding the two contributions above, we get that the total Riemann sum is at most $t + \frac{\varepsilon}{t}$. Picking $t = \sqrt{\varepsilon}$ to optimise this upper bound, we get $(1)$.


The discontinuities of the function are in the form $x=\frac{1}{n}$. Using the given hint by Sivaram Ambikasaran in the comments to construct covering intervals, you can show that the set


has measure zero which satisfies the Lebesgue criterion for integrability.

  • 4
    $\begingroup$ Applying the Lebesgue criterion for this problem seems like an overkill. :-) $\endgroup$ – Srivatsan Dec 9 '11 at 3:01
  • 2
    $\begingroup$ This is a nice and probably the efficient way to go provided the OP knows Lebesgue integral in the first place. $\endgroup$ – user17762 Dec 9 '11 at 3:07
  • $\begingroup$ Probably is overkill, but the interval construction works very nicely in showing that the sets have measure zero. $\endgroup$ – Henry Shearman Dec 9 '11 at 3:08
  • 2
    $\begingroup$ The extra work to get a finite partition is so small, why not go ahead and do it? $\endgroup$ – GEdgar Dec 9 '11 at 20:07

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.