Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

On internet I found some recreational problems as $$3=\sqrt{x+\sqrt{x+\sqrt{x+\sqrt{x+...}}}}$$ $$\frac{1}{2}=\frac{1}{x+\frac{1}{x+...}}$$ $$2=x^{x^{x^{...}}}$$ And the trick to solve them was just to reuse the equation itself inside the equation, so for example obtaining $3=\sqrt{x+3}$ in the first example above, and so on. But this "trick" reminded me of the well known pseudoparadox we could obtain manipulating divergent series as for example the Bolzano's $$1-1+1-1+1-1+...$$ that setting $S=1-1+1-1+1-1+...$ become $$S=1-1+1-1+1-1+...=1-(1-1+1-1+1-1+...)=1-S$$ and therefore $1-1+1-1+1-1+...=S=1/2$. Obviously the error is in assuming that this series has a precise value, naming it $S$.

We obtain other oddities by generalizing the equations above so that $a=\sqrt{x+\sqrt{x+\sqrt{x+\sqrt{x+...}}}}$ would imply $x=a(a-1)$, $a=\frac{1}{x+\frac{1}{x+...}}$ would imply $x=\frac{1}{a}-a$ and $a=x^{x^{x^{...}}}$ would imply $x=a^{\frac{1}{a}}$, that would be quite strange identities for lots of values of $a$.

Then I feel the need to show that the equations above make sense. I think the most obvious way to formalize them was to define them as limits, exempli gratia $$\sqrt{x+\sqrt{x+\sqrt{x+\sqrt{x+...}}}}=\lim_{n\rightarrow \infty}\underbrace { \sqrt{x+\sqrt{x+\sqrt{x+...+\sqrt{x}}}} }_{n \text{ times} }$$

Then, trying to generalize the problem, we want to study the convergence of the sequence of "recursive" function $$F_0(x)=x \\ F_n(x)=f(x,F_{n-1}(x))$$ For example, we obtain the example at the beginning by respectively setting $f(x,y)=\sqrt{x+{y}}$,$f(x,y)=\frac{1}{x+y}$ and $f(x,y)=x^y$.

In case we manage to prove the convergence, naming $l$ the limit of the sequence, we would have that $f(x,l)=l$.

I've tried to study the sequence $x,x^x,x^{x^x},x^{x^{x^{x}}},...$, for $x \in (0,1)$; after noticing that $\underbrace {x^{x^{x^{x}}} }_{2n \text{ times} }\leq \underbrace {x^{x^{x^{x}}} }_{2n+1 \text{ times} }$ and $\underbrace {x^{x^{x^{x}}} }_{2n+1 \text{ times} }\geq \underbrace {x^{x^{x^{x}}} }_{2n+2 \text{ times} }$, I suspected that for very small value, this succession does not converge, somehow oscillating... but I was not able to prove it.

My questions are:

  • Is there some related theory about "infinite equations" as discussed here (that is, if my formalization is correct, the convergence of $F_n(x)=f(x,F_{n-1}(x))$)?
  • What about the special case of $2=x^{x^{x^{...}}}$? Is there a way to prove or disprove the convergence for $x\in (0,1)$?
share|cite|improve this question
Fixed point theorems come to mind... – Asaf Karagila Sep 1 '12 at 0:56
To answer your first question: In a broad way, ergodic theory ( comes to mind as well. – JavaMan Sep 1 '12 at 1:44
up vote 3 down vote accepted

For your example $x^{x^\ldots}$, for $x \in (0,1)$ the function $f(y) = x^y$ is decreasing on $(0,\infty)$ with limits $1$ and $0$ as $y \to 0+$ and $y \to \infty$, so there is a unique fixed point $L(x)$. This is not $2$, it is $-{\frac {{\it LambertW} \left( -\ln \left( x \right) \right) }{\ln \left( x \right) }}$.

Now we have to investigate stability. We have $f'(L(x)) = \exp(-LambertW(-\ln(x))) \ln(x)$. This is an increasing function on $(0,1)$, and is $-1$ at $x = x_0 \approx 0.06598803585$. So for $1 > x > x_0$ the fixed point is stable: from any starting point close enough to $L(x)$ the sequence will converge to $L(x)$. In fact, it appears that this will be the case for any starting point in $(0,\infty)$. For $x < x_0$ the fixed point is unstable, and the sequence will never converge to $L(x)$ unless it happens to hit exactly $L(x)$.

EDIT: That $x_0 = e^{-e}$. For this value, the fixed point is $1/e$.

share|cite|improve this answer

Assume $0<x<1$. As seen above, if $x^{x^{x^\cdots}}$ converges at all, then it converges to a number $a$ such that $x^a=a$, that is a fixpoint of the map $f\colon (0,1)\to(0,1)$, $y\mapsto x^y = e^{y\ln x}$. Since $f'(y) = x^y\ln x$, we have $|f'(y)|<1$ at least when $\frac1e<x<1$; then we find a unique fixpoint $a$. However, if $0< x<\frac 1 e$ I am no longer sure; maybe one can ensure contraction after a few steps?

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.