15

$\gamma=0.57721566...$

$\phi=\frac{1+\sqrt5}{2}$

Let,

$$A=\sum_{n=1}^{\infty}\arctan\left(\frac{(e+e^{-1})(\phi^{\frac{2n-1}{2}}+\phi^{\frac{1-2n}{2}})}{e^2+e^{-2}-1+\phi^{2n-1}+\phi^{1-2n}}\right)$$ $$B=\sum_{n=1}^{\infty}\arctan\left(\frac{(\gamma+\gamma^{-1})(\phi^{\frac{2n-1}{2}}+\phi^{\frac{1-2n}{2}})}{\gamma^2+\gamma^{-2}-1+\phi^{2n-1}+\phi^{1-2n}}\right)$$

(1)

$$A=B=\pi$$

We found (1) accidentally while trying to search for $\pi$ in term of $\arctan(x)$ most idea are from Dr Ron Knott his Fibonacci site.

Can anyone prove (1)

  • 2
    Would you mind to link the site you mentioned? – Ng Chung Tak May 27 '16 at 11:32
  • @NgChungTak, it might be this page: http://www.maths.surrey.ac.uk/hosted-sites/R.Knott/Fibonacci/fibpi.html – Barry Cipra May 27 '16 at 16:25
  • Our followin paper on continued fractions and cot function may be relevant: http://www.math.washington.edu/~reu/papers/2008/william/layered.pdf – DVD Jun 01 '16 at 01:04
  • @DVD It is an interesting paper, but why do you think it is relevant? – Marco Cantarini Jun 08 '16 at 19:26
  • Here is a generalization of your problem. Define $$f(z) = \sum_{n=1}^\infty\arctan\left(\frac{z y_n}{z^2 + y_n^2 - 5}\right)$$ then $A = f(e+e^{-1})$ and $B = f(\gamma + \gamma^{-1})$. I think we should have that $$f(z) = \left{\matrix{\pi & z > \sqrt{3-\sqrt{5}}\0 & z^2 < \sqrt{3-\sqrt{5}}\-\pi & z < \sqrt{3-\sqrt{5}}}\right.$$ I checked this numerically and it seems to hold. By bet is that a clever application of the arctan addtion formula giving rise to some telescoping series is the way to go to prove this. – Winther Jun 10 '16 at 02:36
  • I think you did not mean to put the outer square root radical in the middle branch of your expression @Winther. Also $\sqrt{3-\sqrt{5}}=(\sqrt{10}-\sqrt{2})/2$. – Oscar Lanzi Jun 10 '16 at 19:15
  • Relevant, because of the constants and functions in the formula and that it was discovered by accident... – DVD Jun 11 '16 at 02:16

1 Answers1

17

For any $a \ge 1$, let $f_n(a)$ be the expression

$$f_n(a) \stackrel{def}{=}\frac{(a+a^{-1})(\phi^{n-1/2} + \phi^{1/2-n})}{a^2 + a^{-2} - 1 + \phi^{2n-1} + \phi^{1-2n}}$$

The sums at hand equal to $\sum\limits_{n=1}^\infty \tan^{-1}f_n(e)$ and $\sum\limits_{n=1}^\infty \tan^{-1}f_n(\gamma^{-1})$ respectively.

For any $n \ge 0$, let $p_n = \phi^{n-1/2} - \phi^{1/2-n}$. It is easy to verify

  • $p_{n+1} - p_{n-1} = (\phi - \phi^{-1})(\phi^{n-1/2} + \phi^{1/2-n}) = \phi^{n-1/2} + \phi^{1/2-n}$
  • $p_{n+1} p_{n-1} = (\phi^{n+1/2} - \phi^{-1/2-n})(\phi^{n-3/2} - \phi^{3/2-n}) = \phi^{2n-1} + \phi^{1-2n} - 3$

Let $u = a + a^{-1}$. Notice

$$\frac{\frac{u}{p_{n-1}} - \frac{u}{p_{n+1}}}{ 1 + \frac{u}{p_{n-1}} \frac{u}{p_{n+1}} } = \frac{u(p_{n+1}-p_{n-1})}{u^2 + p_{n+1}p_{n-1}} = \frac{(a+a^{-1})(\phi_{n-1/2} + \phi_{1/2-n})}{a^2 + a^{-2} - 1 + \phi^{2n-1} + \phi^{1-2n}} = f_n(a) $$ There exists integers $N_n$ such that $$\tan^{-1} f_n(a) = \tan^{-1} \left(\frac{\frac{u}{p_{n-1}} - \frac{u}{p_{n+1}}}{ 1 + \frac{u}{p_{n-1}} \frac{u}{p_{n+1}} }\right) = \tan^{-1}\frac{u}{p_{n-1}} - \tan^{-1}\frac{u}{p_{n+1}} + N_n \pi $$

When $n > 1$, $p_{n+1} > p_{n-1} > 0$, it is easy to see $N_n = 0$.
When $n = 1$, we have $$ \begin{cases} f_1(a), \frac{u}{p_2} > 0\\ \frac{u}{p_0} < 0 \end{cases} \quad\implies\quad \begin{cases} \tan^{-1}f_1(a), \tan^{-1}\frac{u}{p_2} \in (0,\frac{\pi}{2})\\ \tan^{-1}\frac{u}{p_0} \in (-\frac{\pi}{2},0) \end{cases} \quad\implies\quad N_1 = 1 $$ This implies $$\begin{align} & \sum_{n=1}^\infty \tan^{-1}f_n(a)\\ = & \sum_{m=1}^\infty \left( \tan^{-1}f_{2m-1}(a) + \tan^{-1}f_{2m}(a)\right)\\ = & \pi + \lim_{N\to\infty}\sum_{m=1}^N\left[\left( \tan^{-1}\frac{u}{p_{2m-2}} - \tan^{-1}\frac{u}{p_{2m}}\right) + \left( \tan^{-1}\frac{u}{p_{2m-1}} - \tan^{-1}\frac{u}{p_{2m+1}}\right)\right]\\ = & \pi + \lim_{N\to\infty} \left[ \left(\tan^{-1}\frac{u}{p_0} + \tan^{-1}\frac{u}{p_1}\right) - \left(\tan^{-1}\frac{u}{p_{2N}} + \tan^{-1}\frac{u}{p_{2N+1}}\right) \right]\\ = & \pi + \left(\tan^{-1}\frac{u}{p_0} + \tan^{-1}\frac{u}{p_1}\right) \end{align} $$ Notice $p_0 = -p_1$, the last bracket vanishes. As a result, $$\sum_{n=1}^\infty \tan^{-1}f_n(a) = \pi\quad\text{ for all } a \ge 1$$

As pointed out by @Winther, we can relax the condition from $a \ge 1$ to $a > 0$.
This is because for $a \in (0,1)$, $f_n(a) = f_n(a^{-1})$.

We can generalize this a little bit. Let $g_n(u)$ be the expression

$$g_n(u) \stackrel{def}{=} \frac{u(\phi^{n-1/2} + \phi^{1/2-n})}{ u^2 + (\phi^{n-1/2} + \phi^{1/2-n})^2 - 5}$$

We have $f_n(a) = g_n(a + a^{-1})$. If we adopt the convention that $\tan^{-1}(\mathbb{R}) = \left(-\frac{\pi}{2}, \frac{\pi}{2} \right]$,
i.e $\tan^{-1}\infty = \frac{\pi}{2}$ and $\tan^{-1}x < 0$ for $x < 0$, we have

$$\sum_{n=1}^\infty \tan^{-1}g_n(u) = \begin{cases} \pi, & u_c \le u\\ 0, & 0 < u < u_c \end{cases} \quad\text{ where }\quad u_c = \sqrt{3-\sqrt{5}} $$

For all $u > 0$, the analysis is essentially the same as above. When $u < u_c$, there is one place we need to adjust the argument. Namely, $g_1(u)$ becomes $-ve$ and this forces the corresponding $N_1$ to vanish. At the end, when $u < u_c$, this causes the $\pi$ term disappear from above sum.

achille hui
  • 122,701