I have a simple question the rational approximation of real vectors.

Dirichlet's simultaneous approximation theorem states:

Given any $d$ real numbers $\alpha_1,\ldots,\alpha_d$ and for every natural number $N \in \mathbb{N}$, there exists an integer denominator $q \leq N$, and $d$ integer numerators $p_1,\ldots,p_d \in \mathbb{Z}$, such that:

$$ \Bigg|\alpha_i - \frac{p_i}{q}\Bigg| < \frac{1}{qN^{1/d}} \text{ for } i=1,\ldots,d $$

I am hoping to simplify this theorem for the case that $\alpha_i \in (0,1)$.

My questions are:

1) If $\alpha_i \in (0,1)$, then can we say that there is a value of $N$ that is large enough so that the rational approximation $\frac{p_i}{q}$ is also confined to the interval $(0,1)$. If so, what is this value?

2) Assuming that there is a value of $N$ such that the rational approximation $\frac{p_i}{q}$ is within $(0,1)$, then it should follow that $p_i < q$ for all $i$. Since the initial theorem states that $q \leq N$, we know that $p_i < q \leq N$ for all $i$. In light of this, can I simply omit the $N$ and restate the theorem as follows:

Given $d$ real numbers $\alpha_1,\ldots,\alpha_d$ such that $|\alpha_i|<1$ for $i=1,\ldots d$, then there exist an integer denominator $q \in \mathbb{N}$ and $d$ integer numerators $p_1,\ldots,p_d < q$ such that:

$$ \Bigg|\alpha_i - \frac{p_i}{q}\Bigg| < \frac{1}{q^{1+\frac{1}{d}}} \text{ for } i=1,\ldots d $$

The key difference here is that I have used the fact $q \leq N$ in order to replace $\frac{1}{qN^{1/d}}$ with $\frac{1}{q^{1+1/d}}$.

Alternatively if anyone has a proof of the simultaneous approximation theorem (for reals), that would be helpful as well (I can't seem to find one anywhere).

  • 1
    $\begingroup$ The theorem doesn't say "there exists $N$"; it says, "for every $N$". $\endgroup$ Mar 21 '14 at 22:37
  • $\begingroup$ Thanks! I just changed it in the theorem. You might notice that I did not change this in my proposed simplifications. This is because I would be happy knowing that there exists $N$ instead of knowing that the result holds for every $N$ (though it would be cool if the latter were true). $\endgroup$
    – Elements
    Mar 21 '14 at 23:04

To restrict the approximation theorem to real numbers between $(0,1)$, notice that if $\alpha>1$, then $\exists \beta\in\mathbb{R},k\in\mathbb{Z}$ such that $\beta\in (0,1)$, and $\beta+k=\alpha$. In this case, we have: $$\begin{align} \left|\alpha-\frac{p}{q}\right| &=\left|\alpha-\frac{p}{q} +k-k\right| \\ &=\left|(\alpha+k)-\left(k+\frac{p}{q}\right)\right| \\ &=\left|\beta-\frac{p+kq}{q}\right|\\ \\ &=\left|\beta-\frac{p'}{q}\right| \end{align}$$

This is allowed since the denominator doesn't change and the numerator is allowed to depend on $\alpha_i$

To see that we can always find a rational approximation $\frac{p_i}{q}$ within $(0,1)$ simply pick $q$ large enough such that $0<\alpha-q^{-1}<\alpha+q^{-1}<1$, as if this holds and $\frac{p_i}{q}$ is not in the desired range than there exists a closer approximation.

  • $\begingroup$ Thanks for this answer! I see what you did. However, I think that this only shows that we can use the theorem to produce a rational approximation for a real number $\alpha \in (0,1)$ - right? What I was asking is if knowing that $\alpha \in (0,1)$ also guarantees that there exists a rational approximation $\frac{p'}{q} \in (0,1)$ (i.e. $p' \leq q$). If not, is there some way that we can modify the theorem so as to obtain a rational approximation in $(0,1)$ as well? $\endgroup$
    – Elements
    Mar 24 '14 at 21:06
  • $\begingroup$ Added an answer to thT $\endgroup$ Mar 26 '14 at 4:24

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.