Let $$f(x) = \frac{1}{2}+\sum_{k=1}^\infty \frac{\mbox{sgn}(\sin kx)}{2^{k+1}} .$$ Here $\mbox{sgn}$ represents the sign function. Many simple integer and rational values of $x$ result in $f(x)$ very closely approximating some simple rational numbers, and you don't have to spend much time to identify plenty of them. Yet it seems obvious that if $x$ is rational, then $f(x)$ is irrational. One number that stands out is $$x = 10^5 + \frac{1}{10}\cdot\Big(\frac{3}{5}\Big)^2.$$

Surprisingly, $f(x)$ is almost equal to $\frac{2}{3}$, as the first $12,897$ binary digits of both numbers agree. Just after that, they disagree. You don't need a sophisticated algorithm to check this. Just compute $\mbox{sgn}(\sin kx)$ for $k=1, 2, \cdots, 12,897$. These signs alternate perfectly depending on whether $k$ is odd or even, just like the binary digits of $\frac{2}{3}$.


I started to have some doubts about the fact that if $x$ is rational, then the sequence $z_k = \mbox{sgn}(\sin kx)$ can not be periodic. Can someone prove that I am right, and that this weird number $x$ is just a coincidence, not leading to periodicity anyway. Do you have some explanation for these coincidences for so many different $f(x)$ values: very often, 20 or 30 binary digits match those of a simple rational, sometimes 40 and even 87 digits for the number $x=10^5$ itself -- but no pattern for $x=10^5-1, x=10^5-\frac{1}{10}, \mbox{ or } x = 10^5+1$. A pattern again for $x=2\cdot 10^5$ and for so many other numbers, starting with $x=1$ resulting in $f(x)=0.11111113\cdots$ (in base $10$).


Another number leading to almost periodicity is $x=\log_2 3$ resulting in $f(x) = 2/5$ (almost). But $x=\sqrt{2}/2$ does not yield the same spectacular result. It is a hit and miss.

Finally, try $x=\frac{355}{113}$. The first $11776655$ binary digits of $f(x)$ are identical to those of $\frac{2}{3}$. Not only $11776655$ is large, but even more surprising, look at the base-$10$ digits of $11776655$: two $1$, two $7$, two $6$, two $5$. Note that if you concatenate the base-$10$ digits of $355$ with those of $113$, you get $113355$.


The reason for the near-periodic behavior of the choice $x = 10^5 + \frac{9}{250}$ has to do with the fact that $$\frac{x}{\pi} \approx 31831.00007753496977,$$ which is almost an integer with error $\epsilon \approx 0.000077 < 10^{-4}$. Moreover, $$\frac{1}{\epsilon} \approx 12897.4062021729,$$ and now you can see why this many terms are needed.

The above also suggests that if you can find some choice of $x$ such that $$\frac{x}{\pi} - \left\lfloor \frac{x}{\pi} \right\rfloor$$ is extremely tiny, you can make this phenomenon extend to as large of a value of $k$ as you please. It just so happens that the particular choice $10^5 + \frac{9}{250}$ is also close to a round number in base 10.

  • $\begingroup$ Based on your answer, I updated my post, now with $x=\frac{355}{113}$ and the first $11776655$ binary digits of $f(x)$ equal to those of $\frac{2}{3}$. $\endgroup$ – Vincent Granville Sep 2 '19 at 15:31

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.