6

Define $R_n=R_{n-1}+\frac{1}{R_{n-2}} $ with $R_0=R_1=1$

Define $K_n=\frac{1}{K_{n-1}}+K_{n-2} $ with $K_0=K_1=1$

These are all limits I've found using Python but no basis of proof for these limits and was wondering if anyone has experience with these.

$$\lim_{n\to\infty} \frac{R_{2n}^2}{K_{2n}^2}=\pi\tag{1}$$

$$\lim_{n\to\infty}\frac{R_{2n+1}^2}{K_{2n+1}^2}=\frac{4}{\pi}\tag{2}$$

$$\lim_{n\to\infty}\frac{R_{dn}}{R_{n}}=\sqrt{d}\tag{3}$$

For odd n, and even d

$$\lim_{n\to\infty}\frac{K_{dn}}{K_{n}}=\frac{2\sqrt{d}}{\pi}\tag{4}$$

Else

$$\lim_{n\to\infty}\frac{K_{dn}}{K_{n}}=\sqrt{d}\tag{5}$$

and

$$\lim_{n\to\infty}(K_{2n+1}^2-K_{2n-1}^2)=\pi; \;\; \lim_{n\to\infty}(R_{n+1}^2-R_{n}^2)=2\tag{6}$$

And I did the math a long time ago and lost my methods but I ended up getting the formula

$$R_{an}\approx F(R_n,a)=\frac{R_n^2\sqrt{a}+\sqrt{aR_n^4+2R_n^2(a-2)+a}}{2R_n}\tag{7}$$

and substituting R_n=1 you get the approximating (although weak) formula

$$R_n\approx\frac{\sqrt{n}}{2}+\sqrt{n-1}\tag{8}$$

I suppose my questions would be is there a better approximation formula? What happens when we generalize to ($R_0=a, R_1=b$) or ($K_0=a, K_1=b$) Is there a good approximation formula for $K_n$, and why does the sequence relations above converge to $\pi$

Larry Powell
  • 251
  • 1
  • 5

1 Answers1

2

We will only cover the case $K_0 = K_1 = R_0 = R_1 = 1$ here.

Let us look at sequence $K_n$ first. It turns out we can express $K_n$ in a closed form.
The key is following observation:

$$K_n = \frac{1}{K_{n-1}} + K_{n-2} \quad\iff\quad K_nK_{n-1} - K_{n-1}K_{n-2} = 1$$

Summing over them leads to

$$K_nK_{n-1} = K_1K_0 + \sum_{k=2}^n (K_kK_{k-1}-K_{k-1}K_{k-2}) = 1 + (n-1) = n $$

For even $n = 2k$, we have $$K_n = K_{2k} = K_0 \prod_{j=1}^k \frac{K_{2j}K_{2j-1}}{K_{2j-1}K_{2j-2}} = \prod_{j=1}^k \frac{2j}{2j-1} = \prod_{j=1}^k \frac{(2j)^2}{2j(2j-1)} = \frac{2^{2k} (k!)^2}{(2k)!}. $$ For odd $n = 2k+1$, we have $$K_n = K_{2k+1} = \frac{K_{2k+1}K_{2k}}{K_{2k}} = \frac{(2k+1)(2k)!}{2^{2k}(k!)^2} = \frac{(2k+1)!}{2^{2k}(k!)^2}$$

For large even $n = 2k$, Stirling's approximation tell us

$$K_{2k} = \frac{2^{2k}\left( \frac{k}{e}\right)^{2k}2\pi k}{\left(\frac{2k}{e}\right)^{2k}\sqrt{2\pi(2k)}}(1 + O(k^{-1})) = \sqrt{\pi k} + O(k^{-1/2}) $$ Together with $K_{2k+1}K_{2k} = 2k+1$, we find for large $n$, $\displaystyle\;K_n^2 = \begin{cases} \frac{\pi}{2} n, & n \text{ even }\\ \frac{2}{\pi} n, & n \text{ odd } \end{cases} + O(1)$.

Let us switch to sequence $R_n$, the recurrence relation $\displaystyle\;R_n = R_{n-1} + \frac{1}{R_{n-2}}$ tells us if $R_{n-1}, R_{n-2}$ is positive, so does $R_n$. Since $R_0 = R_1 = 1$ is positive, all $R_n$ is positive and $R_n$ is an increasing sequence.

Notice $R_2 = 2$ and for all $n > 2$,

$$R_n^2 = \left(R_{n-1} + \frac{1}{R_{n-2}}\right)^2 = R_{n-1}^2 + 2\frac{R_{n-1}}{R_{n-2}} + \frac{1}{R_{n-2}^2} = R_{n-1}^2 + 2 + \frac{2}{R_{n-2}R_{n-3}} + \frac{1}{R_{n-2}^2}\tag{*1} $$ We find for $n > 2$, we can bound $R_n^2$ from below as $$R_n^2 = R_2^2 + \sum_{k=3}^n \left( R_k^2 - R_{k-1}^2 \right) \ge 4 + \sum_{k=3}^n 2 = 4+2(n-2) = 2n$$

Please note that this inequality $R_n^2 \ge 2n$ is also valid at $n = 2$. Now substitute this lower bound back to $(*1)$, we find for $n > 4$

$$\begin{align} R_n^2 - 2n &= \sum_{k=3}^n(R_k^2 - R_{k-1}^2 - 2) = \sum_{k=3}^n \left(\frac{2}{R_{k-2}R_{k-3}} + \frac{1}{R_{k-2}^2}\right)\\ &\le \sum_{k=3}^n \frac{3}{R_{k-3}^2} \le 6 + \sum_{k=5}^n \frac{3}{2(k-3)} = \frac{9 + 3H_{n-3}}{2} \end{align} $$ where $H_n = \sum_{k=1}^n \frac{1}{k}$ is the $n^{th}$ harmonic number.

For large $n$, we have $H_n \approx \log n$. As a result,

$$R_n^2 = 2n + O(\log n)$$

This leads to

$$\begin{align} \lim_{k\to\infty}\frac{R_{2k}^2}{K_{2k}^2} &= \lim_{k\to\infty}\frac{2(2k) + O(\log k)}{\frac{\pi}{2}(2k) + O(1)} = \frac{4}{\pi}\\ \lim_{k\to\infty}\frac{R_{2k+1}^2}{K_{2k+1}^2} &= \lim_{k\to\infty}\frac{2(2k+1) + O(\log k)}{\frac{2}{\pi}(2k+1) + O(1)} = \pi \end{align} $$

This is pretty close to what OP has got except the limits for odd and even case has been swapped. I have computed the sequences $R_n$ and $K_n$ myself and their ratios agree with result here.

achille hui
  • 122,701