# What's the limit of $\sqrt{2 + \sqrt{2-\sqrt{2+\sqrt{2-\sqrt{2-\sqrt{2 + …}}}}}}$?

Let's look at the continued radical

$R = \sqrt{2 + \sqrt{2-\sqrt{2+\sqrt{2-\sqrt{2-\sqrt{2 + ...}}}}}}$

whose signs are defined as $(+, -, +, -, -, + ,-, -, -,...)$, similar to the sequence $101001000100001...$, where

$1 = +$

$0 = -$

This radical seems to converge to a constant approximately equal to $1.567883...$.

The question is: Is it possible to find this limit $R$ in closed form?

Remark: In the article "On the periodic continued radicals of 2 and generalization for Vieta’s product", it is proved that a periodic sequence of signs composed of nested square roots of two converges to $2\sin(q\pi)$ for some rational number $q$. I have tried with non periodic sequences of plus and minus, and they also converge to numbers between $0$ and $2$. if this radical has a closed form, It can be the sine of an irrational multiple of $\pi$, since both are transcendental numbers.

• So one more minus sign after each positive sign? – Simply Beautiful Art May 7 '17 at 23:32
• One positive sign and a negative sign, one positive sign and two negative signs, one positive sign and three negative signs and so on. – user346361 May 7 '17 at 23:40
• To improve approximations, note that: $$1=\sqrt{2-\sqrt{2-\sqrt{2-\dots}}}$$ – Simply Beautiful Art May 7 '17 at 23:41
• Can this be expressed recursively? – MathematicsStudent1122 May 8 '17 at 2:04

For any function $f$ and sequence of functions $f_1, f_2, \cdots, f_n$, let $$\mathop{\bigcirc}_{k=1}^n f_k \stackrel{def}{=} f_1 \circ f_2 \circ \cdots \circ f_n \quad\text{ and }\quad f^{\circ n} \stackrel{def}{=} \mathop{\bigcirc}_{k=1}^n f = \underbrace{f\circ f \circ \cdots \circ f}_{n \text{ times}}$$ be a short hand of composing the functions in given order.

Consider following maps: \begin{align} \psi :&\quad [0,1] \ni \theta \quad\mapsto\quad 2\cos\left(\frac{\pi}{2}\theta\right) \in [0,2]\\ \phi_{\pm} :&\quad [0,2] \ni x\quad \mapsto\quad \sqrt{ 2 \pm x } \in [0,2] \end{align}

The infinite radical at hand can be interpreted as picking an arbitrary $x \in [0,2]$ and study following limit:

$$\lim_{n\to\infty} \left( \mathop{\bigcirc}_{k=1}^n \phi_{+}\circ \phi_{-}^{\circ k}\right)(x)$$ As functions, it is not hard to verify following equalities \begin{align} \psi^{-1}\circ\phi_{+}\circ\psi \;=&\quad [0,1] \ni \theta &\mapsto&\quad \frac{\theta}{2} \in [0,1]\tag{*1a}\\ \psi^{-1}\circ\phi_{-}\circ\psi \;=&\quad [0,1] \ni \theta &\mapsto&\quad 1 - \frac{\theta}{2} \in [0,1]\tag{*1b} \end{align} Let $\alpha = -\frac12.\,$ Apply $(*1b)$ $k$ times followed by $(*1a)$, we get \begin{align} \left(\psi^{-1}\circ\phi_{+}\circ\phi_{-}^{\circ k}\circ\psi\right)(\theta) &= \frac12\left(1 + \alpha + \alpha^2 + \cdots + \alpha^{k-1} + \alpha^k\theta\right) = \frac12\left(\frac{1 - \alpha^k}{1 - \alpha} + \alpha^k \theta\right)\\ &= \frac12\left(\frac23(1-\alpha^k) + \alpha^k\theta\right) = \frac13 + \alpha^{k+1}\left(\frac23 - \theta\right) \end{align} From this, we find

$$\left(\psi^{-1}\circ \left(\mathop{\bigcirc}_{k=1}^n \phi_{+}\circ\phi_{-}^{\circ k}\right)\circ\psi\right)(\theta) = \frac13 + \alpha^2\left[ \frac13 - \alpha^3\left[ \cdots \left[ \frac13 - \alpha^{n+1}\left[\frac23-\theta\right] \right] \cdots \right] \right]\\ = \frac13 \left [ 1 + \sum_{k=2}^n(-1)^k \alpha^{\frac{k(k+1)}{2}-1} \right] + (-1)^{n+1} \alpha^{\frac{(n+1)(n+2)}{2}-1}\left(\frac23-\theta\right)$$ Since $|\alpha| < 1$, the $\theta$ dependent term in above expression drops off exponentially.
Independent of choice of $\theta \in [0,1]$, we have $$\lim_{n\to\infty} \left(\psi^{-1}\circ \left(\mathop{\bigcirc}_{k=1}^n \phi_{+}\circ\phi_{-}^{\circ k}\right)\circ\psi\right)(\theta) = \frac13 \left[ 1 - 2\sum_{k=2}^\infty(-1)^k \alpha^{\frac{k(k+1)}{2}}\right]$$ This means the infinite radical is well defined and has value

$$2\cos\left[\frac{\pi}{6}\left(1 - 2\sum_{k=2}^\infty(-1)^k \alpha^{\frac{k(k+1)}{2}}\right)\right]$$

Numerically, this expression evaluates to $\approx 1.567883223337111$, matching the number mentioned in question.

• Would you mind explaining a bit more how you came up with the ideas presented? – Taufi May 8 '17 at 23:02
• @Taufi it comes down to the trigonometric identities $\sqrt{2+2\cos t} = 2\cos\frac{t}{2}$ and $\sqrt{2-2\cos t} = 2\sin\frac{t}{2} = 2\cos\left(\frac{\pi}{2}-\frac{t}{2}\right)$. If one works in $\theta = \frac{2}{\pi}\cos^{-1}\frac{x}{2}$ instead of $x$, the complicated functional composition of $x \mapsto \sqrt{2 \pm x}$ becomes linear operations in $\theta$ which is much easier to analysis. – achille hui May 8 '17 at 23:09
• @achillehui +1. Very good answer. Does this means that the closed form can be given by using infinite series and not using, for example, only a finite number of known functions? Is it possible to prove this? – user346361 May 8 '17 at 23:36
• @FractionalInquirer the series does look like a Jacobi theta's function except the range for $k$ to sum over doesn't match. It might be possible to express the series with some variant of the theta functions but I cannot figure that out. – achille hui May 8 '17 at 23:48
• I really like your notation you used for your definition at the beginning, is it your own original notation or have you ever seen it anywhere else? I used to brows the hell out of the books and magazines but had never came across this notation you are using. – Arjang May 9 '17 at 7:59

Let $R$ be the iterated square root in question

One easily check that any finite expression of the form $$\sqrt{2\pm\sqrt{2\pm\sqrt{2\pm\cdots}}}$$ is between $0$ and $2$. So $0\le R\le 2$.

Suppose that $a=2\cos t$. Then $$\sqrt{2+a}=\sqrt{2(1+\cos t)}=\sqrt{4\cos^2(a/2)}=2\cos\frac a2$$ and $$\sqrt{2-a}=\sqrt{2(1-\cos t)}=\sqrt{4\sin^2(a/2)}=2\sin\frac a2 =2\cos\left(\frac\pi2-\frac a2\right).$$ Therefore \begin{align} \cos^{-1}\frac R2&=\frac12\cos^{-1}\frac12\sqrt{2-\sqrt {2+\sqrt{2 -\sqrt{2-\sqrt{2-\cdots}}}}}\\ &=\frac{\pi}4-\frac14\cos^{-1} \frac12\sqrt {2+\sqrt{2 -\sqrt{2-\sqrt{2-\cdots}}}}\\ &=\frac{\pi}4-\frac18\cos^{-1} \frac12\sqrt{2 -\sqrt{2-\sqrt{2-\cdots}}}\\ &=\frac{\pi}4-\frac{\pi}{16}+\frac1{16}\cos^{-1} \frac12 \sqrt{2-\sqrt{2-\cdots}}\\ \end{align} etc. So we can get a series for $\cos^{-1}(R/2)$. I think it might be something like $$\frac\pi2\sum_{n=0}^\infty\frac{(-1)^n}{2^{n(n+1)/2}}$$ which is some sort of theta function.

• I'm sorry but how does $\sqrt{2(1+\cos{t})} = \sqrt{4\cos^{2}{(a/2)}}$? – VladeKR May 9 '17 at 7:16
• Shouldn't $2\cos(t/2)$ be absolute value? – VladeKR May 9 '17 at 8:03
• @LordSharktheUnknown The numerical value of the serie you gave is approximately $0.958691...$, while the infinite radical has value $1.567893...$, so that serie isn't the correct answer.. – user346361 Jul 1 '17 at 5:48

First aid.

$\sqrt {2_1 - \sqrt {2_2 \pm \sqrt {2_3 \pm \ldots \pm \sqrt {2_n }}}} = 2 \sin \left ( \dfrac {90 ^ \circ (2a + 1)} {2 ^ n} \right )$

$\sqrt {2_1 + \sqrt {2_2 \pm \sqrt {2_3 \pm \ldots \pm \sqrt {2_n }}}} = 2 \cos \left ( \dfrac {90 ^ \circ (2a + 1)} {2 ^ n} \right )$

If $n \le 2$ then $a = 0$

If $n \ge 2$ then $0 \le a \le 2 ^ {n - 2} - 1$

If $n, a$ are known, then the signs $s_t = \pm 1$ lying between the terms $2_2$ and $2_n$ are calculated from the relationships

$r = round(a / 2 ^ {n - t})$

$s_t = (- 1) ^ r , t = 2, 3, \ldots , n - 1$

if we match $+ = 0, - = 1$ , then the binary digits which are among the terms $2_2$ and $2_n$ form a number $b$ , which is closely related to the $a$ parameter:

● Binary representations of $a, b$ always have the same number of digits.

● Each value of $a$ is assigned to a single value of $b$, regardless of the $n$ value.