10
$\begingroup$

Let $^na$ denote the $n$th tetration of $a$, so that $^0a=1$ and $$^{n+1}a=a^{^na}$$ for $n=0,1,\dots$. (For complex $x$ and $y$, here we use the definition $x^y:=e^{y\ln x}$, where $\ln$ is the principal branch of the logarithm.)

It appears that $^9(-\sqrt2)$ is very close to $1$, but not exactly $1$ — so that the sequence $\big(^n(-\sqrt2)\big)$ is almost(?) periodic: $$^9(-\sqrt2)-1\approx(4.99+1.51\, i)\times10^{-45}.$$

Is this a mere coincidence or is there an explanation for this?

$\endgroup$
11
  • $\begingroup$ It seems like the $^{n}(-\sqrt{2})$ is small in absolute value until $n=6$, then for $n=7$ it gets out of hand ($^{7}(-\sqrt{2})\sim-33+29i$), so that $^{8}(-\sqrt{2})$ is almost $0$ and $^{9}(-\sqrt{2})$ is almost $1$. In general if at some point $a_n=^{n}(-\sqrt{2})$ satisfies that $a_n\cdot\ln(-\sqrt{2})$ has a very negative real part, the same thing is going to happen, so at least it is not such a big coincidence $\endgroup$
    – Saúl RM
    Commented Nov 22, 2022 at 2:57
  • 1
    $\begingroup$ As soon as this sequence gets its real part sufficiently big and negative, the next one will be very close to zero, and, consequently, second next will be almost equal to 1. Tetration of a (more or less random) complex number looks pretty much like a random process to me (...until it's not random, as I pointed out), so I'd totally expect it to fall somewhat deep into negative half-plane at some point. Also I'm doubtful about "periodicity"; even things as tiny as $10^{-45}$ will quickly blow up under iterated exponents $\endgroup$
    – Denis T
    Commented Nov 22, 2022 at 3:11
  • 1
    $\begingroup$ @GottfriedHelms : Thank you for your comment. Unfortunately, I don't understand it. In particular, it is not clear to me what you mean by "reversed the operation to that of $\log()$". In your expression (definition?) for $\log_b(z,2)$ there seems to be no $b$. The meaning of your table is also unclear to me. $\endgroup$ Commented Nov 22, 2022 at 16:56
  • 1
    $\begingroup$ About repunits and frozen digits in tetration with a high hyperexponent, let me share a couple of papers that I've recently written, proving that integer tetration is the only hyperoperator which is characterized by a constant "congruence speed" for any base that is not a multiple of $10$: arxiv.org/abs/2208.02622 [1] and also arxiv.org/abs/2210.07956 [2]. In [1], it has been shown that the congruence speed of any given base depends only on the $15$ solutions of the equation $y^5=y$ in the commutative ring of $10$-adic integers. $\endgroup$ Commented Nov 22, 2022 at 23:58
  • 1
    $\begingroup$ Thus, tetration has a very stable tail of frozen digits and repunits occur quite often, at the end (e.g., see [1], Equations (6)&(7), second line). $\endgroup$ Commented Nov 23, 2022 at 0:03

1 Answer 1

22
$\begingroup$

This is not a huge coincidence: the idea is that the sequence $a_n={}^{n}(-\sqrt{2})$ has small norm until $n=6$, then it gets out of hand for $n=7$ ($a_7\sim-33+29i$), so that $a_8=e^{a_7\ln(-\sqrt{2})}$ is almost $0$ and $a_9$ is almost $1$.

In general when sequences of this type get out of hand (increase exponentially in norm), it seems likely that at some point you will raise $e$ to a number with a big negative real part so a similar phenomenon will happen.

In fact, as you say the sequence is almost periodic, in the sense that $a_{9n}$ converges to some point close to $1$ when $n$ goes to infinity. Let $k=\ln(-\sqrt{2}):=\pi i+\frac{1}{2}\ln(2)$ and let $f(x)=e^{kx}$, so that $a_{n+1}=f(a_n)$. Then $f'(x)=kf(x)$, and by the chain rule we get that $(f^n)'(x)=k^n\prod_{i=1}^nf^i(x)$.

Now let $g(x)=f^9(x)$. We know that $|g(1)-1|<10^{-40}$, so by the proof of the Banach fixed point theorem, to check that $g^n(1)$ converges to some point close to $1$ it is enough with the following proposition:

Proposition: If $x\in B(1,10^{-30})$, then $|g'(x)|<\frac{1}{2}$.

To prove it, note that $|g'(x)|=|k|^9\lvert\prod_{i=1}^9f^i(x)\rvert\leq10^6\prod_{i=1}^9|f^i(x)|$, so we just have to estimate $|f^i(x)|$. It is also easy to check that $|f^i(x)|<1.49$ for $i=1,\dots,6$. Also, for any $x$ with $|x|<1.5$ we have $|f'(x)|=|kf(x)|\leq4|f(x)|\leq4e^{3.5\cdot1.5}<1000$.

This implies by induction that $|f^i(x)-f^i(1)|\leq1000|f^{i-1}(x)-f^{i-1}(1)|$ for $i\leq6$, so $|f^6(x)-f^6(1)|\leq10^{-15}$, so $|f^7(x)-f^7(1)|<10^{-10}$ and $|f^8(x)|=|e^{kf^7(x)}|\sim |e^{-103-95i}|<|10^{-40}|$. So, finally, $|g'(x)|\leq10^6\prod_{i=1}^9|f^i(x)|\leq10^6\cdot1.49^7\cdot100\cdot10^{-40}\leq\frac{1}{2}$, as we wanted.

$\endgroup$
2
  • 1
    $\begingroup$ A nice addition about the convergence! $\endgroup$ Commented Nov 22, 2022 at 17:41
  • 1
    $\begingroup$ It is not the first time that I've seen these kind of questions involving tetration, but this is a final answer which could be copy-pasted and holds for almost all of them. Very clear explanation! $\endgroup$ Commented Nov 23, 2022 at 0:18

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .