11

In here, I saw that $$x^{x^{x^{x^{x^{x^{x^{.{^{.^{.}}}}}}}}}}$$ exists as a real number (convergent) if and only if $$x\in[e^{-e}, e^\frac{1}{e}].$$ How can I prove this??

Bumblebee
  • 18,220
  • 5
  • 47
  • 87
  • 6
    Have a look at http://en.wikipedia.org/wiki/Tetration. What you are asked has been established by Euler. – Claude Leibovici Jan 03 '15 at 10:16
  • 3
    Hint: $y=x^y$. – Lucian Jan 03 '15 at 10:36
  • 1
    You should define this infinite power tower as $$\lim_{n\to\infty} a_n$$ where $$a_0 = 1 \qquad \text{and} \qquad a_n=x^{a_{n-1}}$$ This lets you formulate the question precisely, unambiguously (otherwise it's not clear what exactly an infinite power tower is). – Zubin Mukerjee Jan 03 '15 at 10:41
  • 2
    @Lucian Assuming convergence of $y$, we can say $$y=x^y$$ – Zubin Mukerjee Jan 03 '15 at 10:42
  • @ZubinMukerjee: Correct. $($I was referring to how one might determine the limits of the interval of convergence$)$. – Lucian Jan 03 '15 at 19:16
  • 1
    These may be equivalent, but it is far from clear that one is the same question as the other, and the answers do not seem to match. – trlkly Jan 06 '18 at 04:41

3 Answers3

7

Using Knuth's up-arrow notation,

$$\newcommand{\W}{\operatorname{W}}x^{x^{x^{x^{x^{x^{x^{.{^{.^{.}}}}}}}}}} = \lim_{n \to \infty} x\uparrow\uparrow n$$

Let $\displaystyle y = \lim_{n \to \infty} x\uparrow\uparrow n$ and let us assume that it does converge.

Then, $\displaystyle x^y = \lim_{n \to \infty} x \uparrow\uparrow(n+1)$

As $n \to \infty$, $n + 1 \to \infty$ (Roughly speaking $\infty \pm n \to \infty$)

Hence, $\displaystyle \lim_{n \to \infty} x\uparrow\uparrow n \equiv \lim_{n \to \infty} x \uparrow\uparrow(n+1) \implies \boxed{\displaystyle y = x^y}$

A trivial solution here is $(x,y) = (1,1)$

The variables $x$ and $y$ in $y = x^y$ are not separable (we cannot isolate the any of the variables here on to any one side ) in this equation by usual algebraic manipulations.

However, we can use the Lambert $\W$ function to isolate the variables, $$y = - \frac{\W(-\ln x)}{\ln x}$$

Considering the real values of $\W$ and Using the fixed point iteration method,
it can be seen that $\displaystyle x\in\left[\frac1{e^{e}}, e^\frac{1}{e}\right]$

Good Luck :)

Alice Ryhl
  • 7,853
Nick
  • 6,804
  • For anyone like me not seeing how to get to $y = - \frac{W(-\ln x)}{\ln x}$. The $W$ function satisfies $W(z) = w$ if and only if $z = we^w$, so we want to get to something of this form. Start with $y = x^y$, divide by $x^y$ to get $yx^{-y} = 1$, and rewrite as $ye^{-y\ln x } = 1$. Now multiply by $-\ln x$ to match the exponent: $-y\ln x\ e^{-y\ln x} = -\ln x$. This is in the desired form, and we get $-y \ln x = W(-\ln x)$, ergo $y = \frac{W(-\ln x)}{-\ln x}$. – Anakhand Feb 20 '24 at 09:38
7

The equation

$$x^{x^{x\cdots}}$$

is not well defined, we have to find some way to define it to find its value.

There are two sensible definitions:

The first is to say that the thing in the exponent is actually the same as the entire thing, which is written as

$$y=x^y$$

And solving for $y$ here would give the solution.

To solve this we need a special multivalued function called the Lambert $\newcommand{\W}{\operatorname{W}}\W$ function, which satisfies $z=\W(z)\cdot\exp(\W(z))$ for all complex $z$. The solution is then

$$y=-\frac{\W(-\log x)}{\log x}$$

If we restrict attention to real-valued $\W$, the function is defined only for $x \ge −1/e$, and is double-valued on $(−1/e, 0)$.

Since the argument is $-\log x$, which is a decreasing, and $-\log(e^{1/e})=-\frac1e$, our solution is only defined when $x\le e^{1/e}$ and when the logarithm is defined. Therefore, under this definition, the function is defined (not necessarily uniquely) and real when $x\in[0,e^{1/e}]$


There is however another definition:

Define the sequence $a_n$

$$a_0=x\qquad\qquad a_n=x^{a_{n-1}}$$

And the value were looking for is

$$\lim_{n\to\infty}a_n$$

Which can be written as $$\lim_{n\to\infty}x\uparrow\uparrow n$$

In Nicks answer, it is explained why this is the same as the definition $y=x^y$

Anakhand
  • 2,572
Alice Ryhl
  • 7,853
1

As Lucian mentioned above, I have considered the function $$f(y)=y^{\frac{1}{y}}$$ $$\lim_{y\to 0^+}f(y)=0$$ $$\lim_{y\to \infty}f(y)=1$$ Also $$f'(y)=y^{\frac{1}{y}}\Big(\dfrac{1}{y^2}-\dfrac{\ln y}{y^2}\Big).$$ Using this, we have maximum of $f(y)$ occurs at $y=e.$ Hence $$f(y)\le e^{\frac{1}{e}},\,\,\,\,\,\, \forall y\ge 0.$$ But I did not have an idea about obtaining the lower bound $e^{-e}.$

Bumblebee
  • 18,220
  • 5
  • 47
  • 87