The best rational approximations $p/q$ to an irrational $\alpha$ are defined by the property $$ \left|\alpha - \frac{p}{q}\right| < \left|\alpha - \frac{p'}{q'}\right| $$ for all $q' \leq q$. The approximants $p/q$ are found by simply truncating the continued fraction expansion.

The "most" irrational number is the Golden ratio $\phi$, which is defined by the property that for any given $N$, it has the most good approximations which satisfy $q < N$.

Furthermore, for (i) algebraic and (ii) almost all irrational numbers, they satisfy the bound $$ \left|\alpha - \frac{p}{q}\right| > \frac{1}{q^{2+\epsilon}} $$ for any $\epsilon > 0$ and $q$ sufficiently large.


I am interest in known generalisations of these results to the approximation of multiple irrationals.

I have found a generalisation of part of the final result, which is provided by the Subspace theorem. The subspace theorem has the following corollary: for $D$ rationally independent algebraic numbers $(\alpha_1, \alpha_2, \ldots \alpha_D)$, $$ \left|\alpha_d - \frac{p_d}{q}\right| > \frac{1}{q^{1+1/D+\epsilon}} $$ for any $\epsilon > 0$, and $q$ sufficiently large.


My questions are:

  • Is there a commonly used corresponding definition of the best rational approximations $(p_1/q,p_2/q \ldots p_D/q)$ to the irrational tuple $(\alpha_1, \alpha_2, \ldots \alpha_D)$? (generalising the first equation above)
  • If there is a good definition, is there a better method than exhaustive search for finding the rational approximations $p_d/q$ to the irrational tuple $\alpha_d$? (generalising the truncated continued fraction expansion)
  • For a given $D$ is there a known "most irrational" tuple $(\alpha_1, \alpha_2, \ldots \alpha_D)$ in the sense that there are the maximal number of good approximations satisfying $q<N$ for any $N$? (generalising the Golden ratio)
  • $\begingroup$ Although "truncations" of continued fractions are best rational approximations, there are some additional fractions related to the continued fraction expansion which are also best rational approximations. It's been awhile since I looked at the literature on simultaneous rational approximations, but I recall no analogy to the continued fraction algorithm (for multiple targets). $\endgroup$ – hardmath Aug 21 at 22:56
  • 1
    $\begingroup$ @hardmath Indeed, my understanding is that there really aren't any good algorithms. I think that there is a way of finding good simultaneous approximations using a combination of continued fraction expansion (to get initial approximants), then massaging these a bit further with a lattice basis reduction algorithm such as LLL. I have a colleague who wrote his PhD thesis on this---I'll give him a ping and see if he has any insight. $\endgroup$ – Xander Henderson Aug 21 at 23:22
  • $\begingroup$ @hardmath yes you are right, I should have written $|q \alpha < p| < | q' \alpha < p'|$ for the first equation and similarly multiplied by $q$ for all other equations. $\endgroup$ – ComptonScattering Aug 22 at 0:35
  • $\begingroup$ @XanderHenderson thank you $\endgroup$ – ComptonScattering Aug 22 at 0:36
  • 1
    $\begingroup$ Try: math.tamu.edu/~doug.hensley/SimultaneousDiophantine.pdf $\endgroup$ – O. S. Dawg Aug 23 at 0:43

To the best of my knowledge, there does not exist an analog of the continued fraction algorithm to do this. However, as my colleague Xander Henderson pointed out, you can use a popular lattice basis reduction algorithm, known as the LLL algorithm, to calculate simultaneous Diophantine approximations. Here is a link to the original paper on LLL that got me started (see Proposition 1.39). link to paper

  • $\begingroup$ Thanks eddie, I will digest this and may well come back to ask further questions. $\endgroup$ – ComptonScattering Aug 22 at 20:16

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.