Let $\alpha \in \mathbb{R}$ be a fixed (positive) number. For each $k \in \mathbb{N}$ we choose $\varepsilon_k >0$ with the property that $\lim_k \varepsilon_k =0$.

If $\alpha \in \mathbb{R} \setminus \mathbb{Q}$, we know that there exist infinitely many $q_k \in \mathbb{Z}$ such that $$ \frac{1}{2} - \varepsilon_k \leq \{q_k \alpha \} \leq \frac{1}{2}, $$ since $\{q \alpha\}$ is dense in $[0,1)$. Here $\{ z \}$ denotes the fractional part of a number $z$, i.e. $\{z\} = z -[z]$. Can we also choose $q_k$ such that $q_k = O(\varepsilon_k^{-1})$? In other words, can we select $q_k$ so that $\limsup_k \varepsilon_k q_k < \infty$?

It seems to me that the answer is not trivial, since we are, roughly speaking, trying to single out a slow sequence of approximation. Maybe this holds only for some particular (kind of) irrational number $\alpha$.

Edit. It seems that there is a positive result if one requires the weaker condition $k/\varepsilon_k \geq q_k$, see here. Of course $q_k$ can be much larger than we need.

  • $\begingroup$ What does $\lim_k$ mean? $\endgroup$ – joro May 8 '16 at 11:10
  • 1
    $\begingroup$ @joro $\lim_k = \lim_{k \to +\infty}$, the only possible limit as $k \in \mathbb{N}$. $\endgroup$ – Siminore May 8 '16 at 11:18
  • $\begingroup$ I think this should be true using some form of the pigeonhole principle, like Dirichlet's approximation theorem. $\endgroup$ – benblumsmith May 8 '16 at 11:29
  • $\begingroup$ @benblumsmith Do you think it is so easy? Take for instance $\varepsilon_k = 2^{-k}$. Then we want $q_k \lesssim 2^k$. I don't see how a pigeonhole principle can be invoked. How can we bound, a priori, the largeness of the integer $q_k$? In some sense there are too few integers below $2^k$... $\endgroup$ – Siminore May 8 '16 at 11:32

For each positive integer $N$, let $A_N$ be the set of $\alpha$ such that for every positive integer $k$ there is a positive integer $q_k$ with $q < N \epsilon_k^{-1}$ and $1/2 - \epsilon_k \le \{q_k \alpha\} \le 1/2$.

Note that if $\alpha = b/c$ is rational with $c$ odd, $1/2 - \epsilon_k \le \{q_k \alpha\} \le 1/2$ is impossible unless $2 c \epsilon_k \ge 1$. So $A_N$ is a closed set that contains no rationals with odd denominator, and thus is nowhere dense. By the Baire category theorem, the union of these is a meagre set. In particular, there exist irrationals not in that union. For such $\alpha$, you can't have $q_k = O(\epsilon_k^{-1})$.

  • $\begingroup$ It is very interesting. I suspected that this could be true only for most irrationals. However I am unable to make a reasonable guess. $\endgroup$ – Siminore May 9 '16 at 19:15
  • $\begingroup$ Another thing I suspect is that the constant $C$ in the upper bound $q_k \leq C \varepsilon_k^{-1}$ has a role. There seem to be (partial) results with a small $C$. $\endgroup$ – Siminore May 10 '16 at 8:52

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.