I'm trying to prove that $$\lim_{h \to 0}\frac{1}{h}\int_0^h{f(t)dt} = 0$$ where $$f(t) = \begin{cases}\cos{\frac{1}{t}} &\text{ if } t \neq 0\\ 0&\text{otherwise}\end{cases}.$$ Can someone give me a hint where to start? Darboux sums somehow seem to lead me nowhere.

NOTE: I cannot assume that $f$ has an antiderivative $F$.

  • $\begingroup$ Are you sure that this limit exists? $\endgroup$ – Crostul Nov 29 '15 at 12:59
  • $\begingroup$ Try starting with $|\cos \frac1t| \le 1$ if $t \ne 0$. $\endgroup$ – GNUSupporter 8964民主女神 地下教會 Nov 29 '15 at 13:00
  • $\begingroup$ I tried that and I got $\lim_{h \to 0}\frac{1}{h}\int_0^h{f(t)dt} \leq \lim_{h \to 0}\frac{1}{h}\int_0^h{dt} = 1$, so I still know nothing. $\endgroup$ – Andrei Kh Nov 29 '15 at 13:03

Consider $h>0$ then there is $n$ so that $ \frac{1}{(n+1)\pi +\pi/2} < h\le \frac{1}{n\pi + \pi/2}$. In this interval (called $I_n$), $\cos(1/t)$ is positive (resp. negative) if $n$ is odd (resp. even). Also

$$\int_0^h \cos\left(\frac 1t\right) dt = \sum_{k=n+1}^\infty \int_{I_k} \cos\left(\frac 1t\right) dt + \int_{\frac{1}{(n+1)\pi + \pi/2}}^h \cos\left(\frac 1t\right) dt$$

Note that the last term is bounded by $\frac{2}{\pi n^2}$. On the other hand, if we let $$a_k = \int_{I_k} \cos\left(\frac 1t\right) dt,$$ and $a_k$ is an alternating sequence and $a_k \to 0$. If $l> k$, then $|a_k| > |a_l|$ (see below). Thus we have

$$\left|\sum_{k=n+1}^\infty a_k\right| \le |a_{n+1}| \Rightarrow \left| \sum_{k=n+1}^\infty \int_{I_k} \cos\left(\frac 1t\right) dt\right| \le \int_{I_{n+1} }\left| \cos\left(\frac 1t\right) \right| dt \le \frac{2}{\pi n^2}. $$

As $h \in I_n$, $h > \frac{1}{(n+1)\pi + \pi/2}> \frac{1}{2\pi n} \Rightarrow \frac{1}{h} < 2\pi n$. Thus

$$\left|\frac 1h\int_0^h \cos \left(\frac 1t \right) dt \right| \le (2\pi n) \frac{4}{\pi n^2} = \frac{8}{n}. $$

As $h\to 0$, $n\to \infty$ and so the limit goes to zero.

Remark: If we do the substitution $u = 1/t$, then $$a_k = \int_{I_k} \cos\left(\frac 1t\right) dt = \int_{k\pi +\pi/2}^{(k+1)\pi + \pi/2} \frac{\cos u}{u^2} \mathrm du \Rightarrow |a_k| = \int_{k\pi +\pi/2}^{(k+1)\pi + \pi/2} \frac{|\cos u|}{u^2} \mathrm du.$$

Now $|\cos u|$ is $\pi$-periodic, $|a_k|$ is strictly decreasing.

  • $\begingroup$ Nice! But how do you prove that $|a_k| > |a_l|$ for $l >k$? $\endgroup$ – Andrei Kh Nov 29 '15 at 13:31
  • $\begingroup$ @AndreiKh Please see the edit. $\endgroup$ – user99914 Nov 29 '15 at 13:38
  • $\begingroup$ @JohnMa Please, how do we know that $a_k$ is an alternating sequence and that it converges to $0$. If I understand what you did correctly, $a_k$ was assumed to be an alternating sequence. Any justification to why it is so? Thanks $\endgroup$ – Jaynot Nov 18 '17 at 22:00
  • $\begingroup$ @JohnMa Also, if we change the question to $ \cos(\frac{2}{t}$, then the above argument works also. Am i right? $\endgroup$ – Jaynot Nov 18 '17 at 22:40

Partial integration \begin{align} \int \cos \bigg(\frac{1}{t} \bigg) dt &= \int t^2 \frac{1}{t^2}\cos \bigg(\frac{1}{t} \bigg)dt \\ &= -t^2 \sin \bigg(\frac{1}{t} \bigg) + \int 2 t \sin \bigg(\frac{1}{t} \bigg) dt \\ \end{align}


Let $g=\frac1h$ and $s=\frac1t$. Then integration by parts gives $$ \begin{align} \lim_{h\to0}\frac1h\int_0^h{\cos\!\left(\frac1t\right)\mathrm{d}t} &=\lim_{g\to\infty}g\int_g^\infty{\frac{\cos(s)}{s^2}\,\mathrm{d}s}\\ &=\lim_{g\to\infty}g\left[-\frac{\sin(g)}{g^2}+2\int_g^\infty\frac{\sin(s)}{s^3}\,\mathrm{d}s\right]\\ &=\lim_{g\to\infty}O\left(\frac1g\right)\\[6pt] &=0 \end{align} $$ Since $\sin(s)=O(1)$.

  • $\begingroup$ What did you do with the last integral? $\endgroup$ – Kamil Jarosz Dec 6 '15 at 10:05
  • $\begingroup$ The absolute value of the last integral is less than $2\int_g^\infty\frac1{s^3}\,\mathrm{d}s=\frac1{g^2}$ $\endgroup$ – robjohn Dec 6 '15 at 13:30
  • $\begingroup$ Indeed, cool solution +1 $\endgroup$ – Kamil Jarosz Dec 6 '15 at 13:39

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.