This question arose from here. I am interested to find a nice proof about the convergence of $${^n}a=\underbrace{a^{a^{\ .^{\ .^{\ .^a}}}}}_{n\ \text{times}}.$$ I find with google a necessary and sufficient condition to have the convergence is $\frac{1}{e^e} \leq a \leq e^{1/e}$ but the part for $a\le1$ need some ugly work.

Does anyone have an elegant/slick proof ?

  • $\begingroup$ For finite $n$ I imagine it'll always converge, do you mean infinite $n$? $\endgroup$ – frogeyedpeas Aug 7 '14 at 18:27
  • 1
    $\begingroup$ @frogeyedpeas Sure, I mean $\lim\limits_{n\to\infty}{^n}a$. $\endgroup$ – Free X Aug 7 '14 at 18:30

Does anyone have an elegant/slick proof ?

Like in the answer given to the link you quote by Kiril, the fastest proof is given by fixed point iteration. The function you iterate, is:


The fixed points of $g_c(x)$ are given in terms of the Lambert $W$ function as:


The fixed point condition (for attraction hence convergence) is therefore:

$$|g'(x_0)|\le 1\Rightarrow$$ $$|-W(-\ln(c))|\le 1 (*)$$

Now consider the function:


$$(*)\Rightarrow W(-\ln(c))\in [-1,1]\Rightarrow$$ $$m(W(-\ln(c)))\in m([-1,1])\Rightarrow$$ $$m(W(-\ln(c)))\in [-e^{-1},e]\Rightarrow$$ $$-\ln(c)\in[-e^{-1},e]\Rightarrow$$ $$c\in[e^{-e},e^{1/e}]$$

Fixed point iteration, therefore, takes care of all cases where $c\in(e^{-e},e^{1/e})$ (open interval). The two end points have to be checked separately, because the fixed point condition is inconclusive there, being possibly $\pm1$.

The upper end point ($e^{1/e}$) is easy using standard Calculus (Consider the sequence $a_n=g_{e^{1/e}}^{(n)}(1)$ and prove that it converges to $e$, etc.). The lower end point ($e^{-e}$), is a bit tricky, but can also be done using Calculus, and you are done.


Firstly if $$a \le -1$$ then we are guaranteed $$1 \le |a^a| \le 0$$ so if the value converges it is definitely bounded

For $$-1 < a \le 0$$ we have that $$ |a^a| \rightarrow 1 $$ so again boundedness isn't an issue

For $$0 < a \le 1$$ the same reasoning holds to ensure boundedness

Now for convergence notice the following. If we assume the tower converges to a finite value, that is:

$$ a^{a^{a^{a^{\vdots}}}} = n$$

Then we can take logarithms on both sides and use the power to product rule to find

$$ a^{a^{a^{a^{\vdots}}}}\log(a) = \log(n)$$

In other words:

$$n \log(a) = \log(n)$$

Exponentiate both sides

$$ a^n = n$$


$$a = n^{\frac{1}{n}}$$

Now we can use calculus to find the maximum and minimal values that $n^{\frac{1}{n}}$ take on and this gives us the range of possible $a$ among the real numbers if we want $n$ to be among the real numbers.

  • $\begingroup$ $n^{\frac {1}{n}}$ has doesn't have a positive minimum, then how would this explain the $e^{-e}$ part? $\endgroup$ – Dhvanit Apr 13 at 4:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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