Find all positive real number $\beta$,there are infinitely many relatively prime integers $(p,q)$ such that $$\left|\dfrac{p}{q}-\sqrt{2}\right|<\dfrac{\beta}{q^2}$$

maybe this problem background is Hurwitz's theorem: $$\left|\sqrt{2}-\dfrac{p}{q}\right|<\dfrac{1}{\sqrt{5}q^2}$$

so I guess my problem ? $\beta\ge\dfrac{1}{\sqrt{5}}$

and this problem is Germany National Olympiad 2013 last problem (1), see: http://www.mathematik-olympiaden.de/aufgaben/52/4/A52124b.pdf

  • 4
    $\begingroup$ The Hurwitz constant is the best global constant (i.e. valid for all numbers). There might be smaller constants for spesific numbers such as $\sqrt{2}$. One way to approach this is to write the inequality as $|p^2 - 2q^2| < \beta|p/q + \sqrt{2}|$ and study the solutions to $p^2 - 2q^2 = \pm 1$ etc. $\endgroup$ – Winther Sep 1 '14 at 3:30

Proof sketch. First write

$$\left|\frac{p}{q}-\sqrt{2}\right| = \frac{|p^2-2q^2|}{q^2\left|\frac{p}{q} + \sqrt{2}\right|}$$

The Pell equation $$p^2 - 2q^2 = 1$$

is known to have infinitely many solutions (should be proven) so

$$\left|\frac{p}{q}-\sqrt{2}\right| = \frac{|p^2-2q^2|}{q^2\left|\frac{p}{q} + \sqrt{2}\right|} = \frac{1}{q^2}\frac{1}{\left|\sqrt{2 + \frac{1}{q^2}} + \sqrt{2}\right|}$$

for infinitely many $p,q$ so all $\beta \geq \frac{1}{\sqrt{8}}$ are possible. Now try to show that it fails for $\beta < \frac{1}{\sqrt{8}}$. Let $p^2 - 2q^2 = k$ then $p/q = \sqrt{2 + k/q^2}$ and by inserting this into the inequality above show that it cannot hold for large enough $q$.



  • For all $p,q$ we have $$\left|\left(\frac pq-\sqrt2\right)\left(\frac pq+\sqrt2\right)\right|=\frac{|p^2-2q^2|}{q^2}\ge\frac1{q^2},$$ because $\sqrt2$ is irrational.
  • The Pell equations $p^2-2q^2=\pm1$ have infinitely many solutions $(p_n,q_n)$. For example those determined by $$(\sqrt2-1)^n=p_n-q_n\sqrt2.$$
  • When $p_n$ and $q_n$ are large, $p_n/q_n\approx\sqrt2$.
  • $\begingroup$ +1. The way of proving the infinitude of solutions by looking at $(\sqrt{2}-1)^n$ is very simple and elegant. Just out of curiosity, for any non-square $k$ then $(\sqrt{k}-\lfloor \sqrt{k}\rfloor)^n = p_n - q_n \sqrt{k}$ seems to do the same trick for $p^2 - k q^2 = 1$, or is there any reason why this does not work in that case? $\endgroup$ – Winther Sep 1 '14 at 4:44
  • 1
    $\begingroup$ @Winther (+1 to you was already there). That formula gives other solutions to a Pell equation when you have already found one solution. It does not work that simply always. For $k=6$ we have $\lfloor\sqrt6\rfloor=2$, and $$(\sqrt6-2)(\sqrt6+2)=6-2^2=2$$ instead of $1$. But $\sqrt6\approx 5/2$ and, indeed, $$(2\sqrt6-5)(2\sqrt6+5)=-1.$$ Therefore for all the pairs $(p_n,q_n)$ such that $p_n-q_n\sqrt6=(2\sqrt6-5)^n$ we have $$p_n^2-6q_n^2=(-1)^n.$$ $\endgroup$ – Jyrki Lahtonen Sep 1 '14 at 5:00
  • 2
    $\begingroup$ @Winther, find the solution to $U^2 - k V^2 = 1$ with smallest positive $V$ and also take $U > 0.$ Then, given any $x,y,$ we get $$ (U x + k V y)^2 - k (V x + U y)^2 = x^2 - k y^2 $$ gives a new solution to $x^2 - k y^2 = T$ with larger numbers, for target $T$ of interest. Theorem that the totality of solutions make a finite number of orbits under this action (including inverse). $\endgroup$ – Will Jagy Sep 1 '14 at 17:12
  • 2
    $\begingroup$ @WillJagy Great! Thanks to both of you. I haven't played with the Pell equation for like 10 years so its nice to relearn these things again:) $\endgroup$ – Winther Sep 1 '14 at 17:25

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.