0
$\begingroup$

I'm reading an English translation of Khinchin's Continued Fractions and I may have found an error in Theorem 1, page 4.

Khinchin observes that if we simplify a finite continued fraction $[a_0; a_1, ... a_n]$ as $p/q$, and the continued fraction $[a_1; a_2, ... a_n]$ as $p'/q'$, then because of $p/q = a_0 + \frac 1 {p'/q'}$, we can choose values of $p$ and $q$ such that:

$$p = a_0 p' + q', \quad q= p' \tag {*}$$

And uses this to recursively define a standard representation of a finite continued fraction in the form $p/q$.

Khinchin defines $[a_0; a_1, ... a_k]$ as the $k$-th convergent of a continued fraction $[a_0; a_1, ... a_n]$ where $n\geq k$.

Theorem 1 states, where $p_k/q_k$ is the standard representation of the $k$-th convergent:

$$p_k = a_k p_{k-1} + p_{k-2}$$ $$q_k = a_k q_{k-1} + q_{k-2}$$

Now Khinchin writes $p'_r/q'_r$ for the $r$-th convergent. This is where I first got suspicious... isn't he already writing $p_r/q_r$ for the that? Then he says:

On the basis of the formulas in $(*)$, $$p_n = q_0p'_{n-1} + q'_{n-1}$$ $$q_n=p'_{n-1}$$

What? That isn't what $(*)$ says at all! That recurrence relates $[a_0; a_1, ... a_n]$ to $[a_1; a_2, ... a_n]$, here Khinchin is trying to relate $[a_0; a_1, ... a_n]$ to $[a_0; a_1, .. a_{n-1}]$. It seems to me that Khinchin or the translator is using the word "convergent" wrongly. The recurrence relations forming the basis of Theorem 1 seem to be true when "convergent" is interpreted using the definition that I gave above, but I only verified that for $k=2$.

$\endgroup$
1
$\begingroup$

in the middle of page 3, not numbered, he says

$$ [a_0; a_1, \ldots, a_n] = [a_0; r_1] = a_0 + \frac{1}{r_1}. $$ Here, $$ r_1 = [a_1; a_2, \ldots, a_n] $$

which is consistent with your (*), this being formula (6) on page 4.

Meanwhile, I have gotten good use out of these for years, I recommend you get accustomed to this slang for writing the convergents to some $\sqrt N.$ Notice how the first two convergents are the (legal) $0/1$ and the illegal $1/0.$ Here, for convergent $p/q,$ the integer directly below is $p^2 - 13 q^2.$

$$ \sqrt {13} $$

$$ \begin{array}{cccccccccccccccccccccccccccccc} & & 3 & & 1 & & 1 & & 1 & & 1 & & 6 & & 1 & & 1 & & 1 & & 1 & & 6 & \\ \frac{0}{1} & \frac{1}{0} & & \frac{3}{1} & & \frac{4}{1} & & \frac{7}{2} & & \frac{11}{3} & & \frac{18}{5} & & \frac{119}{33} & & \frac{137}{38} & & \frac{256}{71} & & \frac{393}{109} & & \frac{649}{180} & & \frac{4287}{1189} \\ \\ & 1 & & -4 & & 3 & & -3 & & 4 & & -1 & & 4 & & -3 & & 3 & & -4 & & 1 & & -4 \end{array} $$

$\endgroup$
  • $\begingroup$ Yes, but the situation to which formula (6) is applied in the proof of Theorem 1 doesn't seem to be of that form. $\endgroup$ – Jack M May 12 '15 at 21:01
  • $\begingroup$ @JackM, the important thing is Theorem 1. I recommend you practice with some square roots of positive integers, as i have edited into my answer, then see if you can go back and justify (6). All that is needed to find the "digits" above, the 3;1,1,1,1,6,1,1,1,1,6 is a calculator that takes a positive real number, takes its fractional part, takes the reciprocal of that, and reports that as the new real number. The "digits," the $a_n,$ are the integer parts of those. $\endgroup$ – Will Jagy May 12 '15 at 21:09
  • $\begingroup$ @JackM, meanwhile, I think Khinchin is correct in every particular, but I can live with it if (6) turns out to be wrong. I know how to do these by hand and how to prove theorems with them. $\endgroup$ – Will Jagy May 12 '15 at 21:12
  • $\begingroup$ @WillJagy, what is the bottom line in your computation? When I compute convergents, I write down everything except for that part, and I can't tell what you're doing there. To be clear, I'm talking about the line: 1, -4, 3, -3, ... $\endgroup$ – G Tony Jacobs Jul 1 '15 at 2:40
  • 1
    $\begingroup$ Ah, I see now where you explained that. Thank you. It's the number-field norm of the algebraic integer $p+q\sqrt{13}$. $\endgroup$ – G Tony Jacobs Jul 1 '15 at 3:04
1
$\begingroup$

The problem turned out to be a very simple reading error on my part.

In fact Khinchin's $p'_r/q'_r$ is not the $r^{\rm th}$ convergent of the original continued fraction

$$[a_0; a_1, ... a_n],$$

but of the continued fraction:

$$[a_1; a_2, ..., a_n],$$

in which case his claim is obvious.

$\endgroup$

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.