Is there an exact value to the minimum of the function $$f_k(x)=\underbrace{x^{x^{x^{.^{.^.}}}}}_{2k\,\text{times}}$$ as $k\to\infty$, where $k=1,2,3,\cdots$?

This visualisation in Desmos shows that the minimum is around $0.3792$.

I have specifically stated $2k$ as the function is strictly increasing for odd tetrations.

I am certain that it will not tend to $0$ since for every $x^x$ I add to the tetration, it decreases by $0.0004$, then $0.0003$ and $0.0002$.

Differentiation looks extremely tedious to do...

So is there a rigorous approach to find the exact value, if there exists one, of $\min(f_\infty(x))$?

If not, is there a method to find a lower bound?

  • $\begingroup$ Here, $f$ is a function of $x$. So: $$f_1(x)=x^x\quad f_2(x)=x^{x^{x^x}}\quad \cdots$$ and we want to find the minimum of $f_\infty(x)$. $\endgroup$ – TheSimpliFire Mar 24 '18 at 16:36
  • $\begingroup$ @orlp: The OP explicitly says "as $k\to\infty$", so the minimization is being done wrt $x$. I presume the domain is $x\gt 0$ $\endgroup$ – Prasun Biswas Mar 24 '18 at 16:37
  • $\begingroup$ Note that $e^{1/e} = 1.44466786101...$ $\endgroup$ – marty cohen Mar 24 '18 at 22:50
  • $\begingroup$ math.stackexchange.com/questions/2490769/… My answer shows a detailed graph and equations for the fixed points of $b^z$ and $b^{b^z}$ in the neighborhood of $b=\exp(-e)$ $\endgroup$ – Sheldon L Mar 25 '18 at 17:54

Claim: The minimum $1/e \approx 0.36788$ is achieved at $x = e^{-e} \approx 0.065988$.

We proceed by first establishing that the minimum is upper bounded by $1/e$ and then show that it is also lower bounded by $1/e$.

Proof of upper bound. Let $y_{0}=1$ and define the iteration $$ y_{n+1}=x^{y_{n}}. $$ Note that $$ y_{1}=x,\qquad y_{2}=x^{x},\qquad\text{etc.} $$ tends to the infinite tetration. It is known that this iteration converges (possibly to $\infty$) for all $x\geq e^{-e}$. Define $y_{\infty}$ as the function which associates to each $x \geq e^{-e}$ the limit $\lim_{n\rightarrow\infty}y_{n}$. Note that if $y_\infty(x)$ is finite, then $$ y_{\infty}(x)=\lim_{n\rightarrow\infty}y_{n+1}=\lim_{n\rightarrow\infty}x^{y_{n}}=y^{y_{\infty}(x)} $$ and hence $$ (y_{\infty}(x))^{1/y_{\infty}(x)}=x. $$ Solving for $y_{\infty}(x)$, $$ y_{\infty}(x)=\frac{W(\log(1/x))}{\log(1/x)}. $$ Let $x_0=1$ and define also the iteration $$ x_{n+1}=x^{x^{x_{n}}}. $$ Since it is monotone, this iteration converges (possibly to $\infty$) for all $x\geq0$. Define $x_{\infty}$ as the function which associates to each $x \geq 0$ the limit $\lim_{n\rightarrow\infty}x_{n}$. Since $x_{\infty}=y_{\infty}$ on $[e^{-e},\infty)$, $$ \min_{x\geq0}x_{\infty}(x)\leq y_{\infty}(e^{-e})=1/e. $$

Proof of lower bound. Fix $x$ and let $a=x_{\infty}(x)$ for brevity. Then, $$ x^{x^{a}}=a\implies x=a^{1/(x^{a})}\implies x^{a}=a^{a/(x^{a})}\implies(x^{a})^{x^{a}}=a^{a}. $$ Therefore, $$ a^{a}\geq\min_{u\geq0}u^{u}=e^{-1/e}, $$ which in turn implies $ a\geq1/e $.

  • $\begingroup$ $$b=\exp(-e) \approx 0.0659880358453$$ $$\lim_{n \to \infty} b \uparrow \uparrow n = \exp(-1)\approx 0.36787944117$$ This is the Op's minimum; for b in the range of 0..1. I will post links or answer with more detail later. $\endgroup$ – Sheldon L Mar 25 '18 at 15:39
  • $\begingroup$ See this related answer for the fixed points of $f(z)=b^z$ in the neighborhodd of $b=\exp(-e)$ math.stackexchange.com/questions/2490769/… See the graph and note that there is a parabolic fixed at $b>\exp(-e)$, and for $b_2>b$ the fixed points become complex conjugate pairs. $\endgroup$ – Sheldon L Mar 25 '18 at 15:52
  • $\begingroup$ Thank you for your contributions, parsiad and SheldonL. I have repetitively added more terms to the function and it does look like it converges to $1/e$! $\endgroup$ – TheSimpliFire Mar 25 '18 at 19:22
  • $\begingroup$ @TheSimpliFire: updated with an actual solution. $\endgroup$ – parsiad Mar 25 '18 at 21:53

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.