For a natural number $n$, let $$ \begin{equation} \beta_n(z)=\frac{(1+z)^n+(1-z)^n}{(1+z)^n-(1-z)^n}. \end{equation} $$ Then the coefficients of the numerator and denominator of $\beta_n$ are binomial. For example: $$\begin{equation} \beta_4(z)=\frac{z^4+6z^2+1}{4z^3+4z}=\frac{1}{4} z+\cfrac{1}{\frac{4}{5}z+\cfrac{1}{\frac{25}{16}z+\cfrac{1}{\frac{16}{5} z}}}. \end{equation} $$ Is there a simple formula for the coefficients of the continued fraction for any $n$?


  • the fact that all the coefficients are positive numbers follows from complex analysis, since $\Re\beta_n(z)>0$ for $\Re z >0$

  • computer calculation shows that the factorizations of the coefficients consist of small primes, less than $2(n+1)$

  • the motivation for the problem is the rational approximation of the square root function, since $(z\beta_n)(z^2)\approx z$ for $\Re z >0$

  • a relevant discussion: Binary eigenvalues matrices and continued fractions


Recall by De Moivre's Theorem

\begin{equation} \tan n\theta = \frac{n \tan\theta - C(n,3) \tan^3 \theta + \cdots}{1 - C(n,2)\tan^2\theta+ C(n,4)\tan^4 \theta-\cdots} \end{equation}

Let $z = i\tan w$. Substituting and applying this result $$ \beta_n(z) = \beta_n(i\tan w) = -i \cot nw $$ By 1 or 2 which clearly hold for a complex domain

$\displaystyle \tan(nx) = \cfrac{n\tan x}{1 -\cfrac{(n^{2} - 1^{2})\tan^{2}x}{3 -\cfrac{(n^{2} - 2^{2})\tan^{2}x}{5 -\cfrac{(n^{2} - 3^{2})\tan^{2}x}{7 -\cdots}}}}$

with final term $\dfrac{(n^{2} - (n - 1)^{2})\tan^{2}x}{(2n - 1)}$. Simple operations show

\begin{align} \beta_n(z) = -i \cot nw &= {\frac{1}{ni\tan w} +\cfrac{n^{-1}(n^{2} - 1^{2})i \tan w}{3 -\cfrac{(n^{2} - 2^{2})\tan^{2}w}{5 -\cfrac{(n^{2} - 3^{2})\tan^{2}w}{7 -\cdots}}}} \\ &= {\frac{1}{nz} +\cfrac{n^{-1}(n^{2} - 1^{2})z}{3 +\cfrac{(n^{2} - 2^{2})z^2}{5 + \cfrac{(n^{2} - 3^{2})z^2}{7 +\cdots}}}} \\ &= {\frac{z^{-1}}{n} +\cfrac{n^{-1}(n^{2} - 1^{2})}{3z^{-1} +\cfrac{(n^{2} - 2^{2})}{5z^{-1} + \cfrac{(n^{2} - 3^{2})}{7z^{-1} +\cdots}}}} \\ &= {a_1z^{-1} +\cfrac{1}{a_2z^{-1}+\cfrac{1}{a_2z^{-1} + \cfrac{1}{a_3z^{-1} +\cdots}}}} \end{align}

Where $a_1 = \frac{1}{n}$, $a_2 = 3\frac{n}{n^2-1},a_3 = 5\frac{n^2-1}{n(n^2-2^2)},a_4 = 7\frac{n(n^2-2^2)}{(n^2-1)(n^2-3^2)},\cdots$ to $n$ terms. In general

\begin{equation} a_k = (2k-1)\Big(\frac{n}{n^2-1}\frac{n^2-2^2}{n^2-3^2}\frac{n^2-4^2}{n^2-5^2}\cdots\Big)^{(-1)^k} \end{equation}

to $[k/2]$ fractional products, $[x]$ being the ceiling function, where if $k$ is odd the last term has denominator $1$.

Substituting $1/z$ for $z$ now gives an explicit solution of the problem by using $\beta_{2m}(z) = \beta_{2m}(z^{-1})$ and $\beta_{2m+1}(z)= \beta_{2m+1}(z^{-1})^{-1}$.

To verify your case substituting $n=4$ returns $a_1 = 1/4, a_2 = 4/5$, $a_3 = 25/16$, and $a_4 = 16/5$.

Edit: The form of $a_k$ can be simplified. For even $n$ and even $k$ \begin{equation} a_k = (2k-1) \frac{(n-(k-2))(n-(k-4))\cdots n \cdots (n+k-4)(n+k-2)}{(n-(k-1))(n-(k-3))\cdots(n+k-3)(n+k-1)} = (2k-1)\Bigg(\frac{2^{k-1}((n+k)/2-1)!}{((n-k)/2)!}\Bigg)^2 \frac{(n-k)!}{(n+k-1)!} = \frac{2^{2(k-1)}(2k-1)}{n+k-1}{n-k \choose (n-k)/2}{n+k-2 \choose (n+k-2)/2}^{-1} \end{equation} Setting $k=n$, $a_n = 2^{2(n-1)}((n-1)!)^2/(2n-2)!$ giving a power of $2$ in the numerator. For even $n$ and odd $k$ take the reciprocal of $a_{k+1}$ and multiply by $(2k-1)(2k+1)/(n^2-k^2)$

\begin{equation} a_k = \frac{2k-1}{2^{2k}(n-k)}{n-k-1 \choose (n-k-1)/2}^{-1}{n+k-1 \choose (n+k-1)/2} \end{equation}

Treating odd $n$ and odd $k$ separately because of $n=k=1$ \begin{align} a_k &= (2k-1)\frac{(n-(k-2))(n-(k-4))\cdots(n+k-4)(n+k-2)}{(n-(k-1))(n-(k-3)\cdots n \cdots (n+k-3)(n+k-1)} \\ &= (2k-1)\Bigg(\frac{2^{k-1}((n+k-2)/2)!}{((n-k)/2)!}\Bigg)^2 \frac{(n-k)!}{(n+k-1)!}\\ &= \frac{2^{2(k-1)}(2k-1)}{n+k-1}{n-k \choose (n-k)/2}{n+k-2 \choose (n+k-2)/2}^{-1} \end{align} For odd $n$ even $k$ take the reciprocal of $a_{k+1}$ and multiply by $(2k-1)(2k+1)/(n^2-k^2)$ \begin{equation} a_k = \frac{2k-1}{2^{2k}(n-k)}{n-k-1 \choose (n-k-1)/2}^{-1}{n+k-1 \choose (n+k-1)/2} \end{equation}

  • $\begingroup$ Thank you, it's good that your continued fraction terminates... $\endgroup$ – DVD Oct 12 '18 at 19:37
  • $\begingroup$ Excellent work! The computer calculation shows that $a_n$ has a power of $2$ in the numerator. Can you explain that? $\endgroup$ – DVD Oct 15 '18 at 20:25
  • $\begingroup$ Please see the following question with my answer for the application of your result: math.stackexchange.com/questions/2941564/… $\endgroup$ – DVD Oct 15 '18 at 21:55
  • 1
    $\begingroup$ You may wish to verify the factorial coefficients for odd $n$. $\endgroup$ – Jonny Boy Oct 16 '18 at 3:16
  • 1
    $\begingroup$ Apologies, I corrected a subtle error in the odd coefficients. $\endgroup$ – Jonny Boy Oct 18 '18 at 21:02

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.