Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I'm giving some lectures on continued fractions to high school and college students, and I discussed the standard theorem that, for a real number $\alpha$ and integers $p$ and $q$ with $q \not= 0$, if $|\alpha-p/q| < 1/(2q^2)$ then $p/q$ is a convergent in the continued fraction expansion of $\alpha$. Someone in the audience asked if 2 is optimal: is there a positive number $c < 2$ such that, for every $\alpha$ (well, of course the case of real interest is irrational $\alpha$), when $|\alpha - p/q| < 1/(cq^2)$ it is guaranteed that $p/q$ is a convergent to the continued fraction expansion of $\alpha$?

Please note this is not answered by the theorem of Hurwitz, which says that an irrational $\alpha$ has $|\alpha - p_k/q_k| < 1/(\sqrt{5}q_k^2)$ for infinitely many convergents $p_k/q_k$, and that $\sqrt{5}$ is optimal: all $\alpha$ whose cont. frac. expansion ends with an infinite string of repeating 1's fail to satisfy such a property if $\sqrt{5}$ is replaced by any larger number. For the question the student at my lecture is asking, an optimal parameter is at most 2, not at least 2.

share|cite|improve this question
up vote 4 down vote accepted

2 is optimal. Let $\alpha=[a,2,\beta]$, where $\beta$ is a (large) irrational, and let $p/q=[a,1]=(a+1)/1$. Then $p/q$ is not a convergent to $\alpha$, and $${p\over q}-\alpha={1\over2-{1\over \beta+1}}$$ which is ${1\over(2-\epsilon)q^2}$ since $q=1$.

More complicated examples can be constructed. I think this works, though I haven't done all the calculations. Let $\alpha=[a_0,a_1,\dots,a_n,m,2,\beta]$ with $m$ and $\beta$ large, let $p/q=[a_0,a_1,\dots,a_n,m,1]$. Then again $p/q$ is not a convergent to $\alpha$, while $$\left|{p\over q}-\alpha\right|={1\over(2-\epsilon)q^2}$$ for $m$ and $\beta$ sufficiently large.

share|cite|improve this answer
Gerry: I agree with the first answer, although (perhaps you felt the same way) it seemed kind of cheap to get away with an integer for $p/q$. In your second example I didn't want to check the general case, but I checked $\alpha = [1,m,2,\beta]$. Then $p/q = (m+2)/(m+1)$, with $q = m+1$ and I find that $\alpha - p/q = A/(2q^2)$, where $A = 2(m+1)(\beta+1)/((2m+1)\beta+m) = (1 + 1/\beta + 1/m + 1/(2m\beta))/(1 + 1/(2\beta) + 1/(2m))$, which indeed is nearly 1 (from above) when the integer $m$ and irrational $\beta$ are both sufficiently large. Thus $A = 1+\delta$ for small $\delta$. Thanks! – KCd Jul 26 '12 at 20:40
Thanks for asking the question. I suspect one can strengthen the result by finding a single $\alpha$ such that there is an infinite sequence of $p_n/q_n$, none a convergent, with $\alpha-(p_n/q_n)=(1+\delta_n)/(2q^2)$, $\delta_n$ going to zero. I'm thinking $\alpha=[1,b_1,2,b_2,2,b_3,2,\dots]$ with $b_1\lt b_2\lt b_3\dots$. Perhaps the $b_i$ have to increase rapidly. – Gerry Myerson Jul 26 '12 at 23:51

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.