11

Let $\sinh^{\circ n}x$ be the $n$-fold iteration of $\sinh x$. I'm interested in the limit $$ \lim_{n\to\infty}\sinh^{\circ n}\sqrt{\dfrac{3}{n}}. $$ This limit exists because the general term is decreasing: it suffices to show that $$ \sqrt{\dfrac{3}{n}}>\sinh\sqrt{\dfrac{3}{n+1}},\quad\forall n\in\mathbb{N}^*, $$ which can be deduced from $$ \operatorname{arsinh}x=\ln(x+\sqrt{x^2+1})>\dfrac{\sqrt{3}x}{\sqrt{x^2+3}},\quad\forall x>0. $$ The following graph is the plot for the first million terms.

So I would like to know if the limit can be determined and, if the limit is $0$, can we say something about the asymptotic behavior of $\sinh^{\circ n}\sqrt{\dfrac{3}{n}}$? Any help appreciated.

Sangchul Lee
  • 181,930
Jianing Song
  • 2,545
  • 5
  • 25
  • 2
    https://math.stackexchange.com/q/1449281/42969, https://math.stackexchange.com/q/4785678/42969 – Martin R Oct 30 '23 at 21:43
  • 1
    @MartinR Sorry, but could you please say how these two questions are related to mine? I was aware of both of them, and that iterated $\sin$ doesn't seem to be the same kind of sequence as in this question. – Jianing Song Oct 30 '23 at 22:03
  • You can try and see if a proof similar to https://math.stackexchange.com/q/1449343 would work in your case. – Gary Oct 31 '23 at 00:18
  • Just out of curiosity : what syntax did you use to be able to compute for so large values of $n$ ? Thanks – Claude Leibovici Oct 31 '23 at 04:03
  • @ClaudeLeibovici I used PARI :) – Jianing Song Oct 31 '23 at 04:09
  • @Gary Well, I've tried to estimate $\dfrac{1}{x_k^2}-\dfrac{1}{x_{k+1}^2}$ for $x_k:=\sinh^{\circ k}\sqrt{\dfrac{3}{n}}$. We have $\dfrac{1}{x_k^2}-\dfrac{1}{x_{k+1}^2}=\dfrac{1}{x_k^2}-\dfrac{1}{\sinh^2 x_k}\ge\dfrac{1}{x_n^2}-\dfrac{1}{\sinh^2 x_n}$ for $k=0,\cdots,n-1$, so $\dfrac{n}{3}-\dfrac{1}{x_n^2}\ge n\left(\dfrac{1}{x_n^2}-\dfrac{1}{\sinh^2 x_n}\right)$, or $x_n^2\left(\dfrac{1}{3}-\dfrac{1}{x_n^2}+\dfrac{1}{\sinh^2 x_n}\right)\ge\dfrac{1}{n}$. – Jianing Song Oct 31 '23 at 04:11
  • @Gary This is currently what I got, and it doesn't seem to be very useful: the solution to $x^2\left(\dfrac{1}{3}-\dfrac{1}{x^2}+\dfrac{1}{\sinh^2 x}\right)=\dfrac{1}{n}$ tends to $0$ as $n\to\infty$ ... We do have $\dfrac{x^2}{15}\ge \dfrac{1}{3}-\dfrac{1}{x^2}+\dfrac{1}{\sinh^2 x}$, so $x_n\ge\sqrt[4]{\dfrac{15}{n}}$. – Jianing Song Oct 31 '23 at 04:12
  • 1
    @JianingSong: I had not noticed that your question is about iteration of the hyperbolic sine, my fault! But I will keep my above comment, perhaps those threads inspire someone to find a solution to this problem. – Martin R Nov 02 '23 at 12:29
  • 1
    It seems like this limit only exists and is nonzero with the constant 3; with a larger constant it appears to be infinite and with a smaller constant it appears to be zero. (But I am computing values of $x_k$ by a naive numerical algorithm so I can't guarantee my computations are accurate.) – Michael Lugo Nov 02 '23 at 14:05
  • 2
    @MichaelLugo: Unless I made an error, $\sinh^{\circ n}(x) \le \sqrt{\frac{3x^2}{3-nx^3}}$ holds if $nx^2 < 3$, which shows that the limit of $\sinh^{\circ n}\sqrt{\dfrac{a}{n}}$ is zero if $0 < a < 3$. Unfortunately, this does not seem to help in the case $a=3$. – Martin R Nov 02 '23 at 14:22
  • 1
    Adding another parameter: it also appears that $\lim_{n \to \infty} \sinh^{\circ bn} \sqrt{a/n}$ is zero if $ab < 3$, infinite if $ab > 3$, and as above if $ab = 3$. – Michael Lugo Nov 02 '23 at 14:58
  • @MartinR The limit for $0<a<3$ is a special case of this result :) – Jianing Song Nov 03 '23 at 15:57

1 Answers1

5

In this answer, we establish the following result:

$$ \bbox[color:navy;padding:8px;border:1px dotted navy;]{ \sinh^{\circ n}\sqrt{\frac{3}{n}} \sim \sqrt{\frac{5}{\smash[b]{\log n}}} \quad\text{as}\quad n \to \infty } \tag{$\diamond$} $$

The lemma below plays the key role in our analysis:

Lemma. Define $f(x) = \frac{3}{\sinh^2(\sqrt{3/x})}$. There exists a constant $C > 0$ such that $$ x - 1 \leq f(x) \leq x \tag{1} $$ and $$ x - 1 + \frac{3}{5x} - \frac{C}{x^2} \leq f(x) \leq x - 1 + \frac{3}{5x} \tag{2} $$ for all $x > 0$.

Let us first see how this lemma implies the main result. For each fixed $n$, we define $(n_k)_{0\leq k \leq n}$ so that $n_0 = 0$ and $n_{k} = f(n_{k-1})$ for all $1 \leq k \leq n$. This sequence is designed so that

$$ \sinh^{\circ k}\sqrt{\frac{3}{n}} = \sqrt{\frac{3}{\smash[b]{n_k}}}. $$

Then by $\text{(1)}$, for all $0 \leq k \leq n$ we have

$$ n - k \leq n_k \leq n. $$

Plugging this to the second part of $\text{(2)}$, for all $1 \leq k \leq n$ we get

\begin{align*} n_k &= n_0 + \sum_{j=0}^{k-1} (x_{j+1} - x_j) \\ &\leq n + \sum_{j=0}^{k-1} \left( -1 + \frac{3}{5(n - j)} \right) \\ &= n - k + \frac{3}{5}(H_n - H_{n-k}), \tag{3} \end{align*}

where $H_m = 1 + \frac{1}{2} + \cdots + \frac{1}{m}$ is the $m$th harmonic number. Then plugging this to the first part of $\text{(2)}$,

\begin{align*} n_k &\geq n + \sum_{j=0}^{k-1} \left( -1 + \frac{3}{5(n - j) + 3(H_n - H_{n-j})} - \frac{C}{(n - j)^2} \right). \tag{4} \end{align*}

Then from $(3)$, we have

$$ n_n \leq \frac{3}{5}H_n \sim \frac{3}{5}\log n. $$

Also, from $(4)$ and with $C' = C \sum_{k=1}^{\infty} \frac{1}{k^2}$, we get

\begin{align*} n_n &\geq \sum_{j=0}^{n-1} \frac{3}{5(n - j) + 3(H_n - H_{n-j})} - C' \\ &\geq \sum_{k=1}^{n} \frac{3}{5k + 3(\log n - \log k)} - C' \\ &\geq \int_{1}^{n} \frac{3 \, \mathrm{d}t}{5t + 3(\log n - \log t)} - C'. \end{align*}

In the last step, we utilized $H_n - H_k = \frac{1}{k+1} + \cdots + \frac{1}{n} \leq \int_{k}^{n} \frac{\mathrm{d}t}{t} = \log n - \log k $ and the fact that $t \mapsto 5t - 3\log t$ is positive and increasing for $t \geq 1$. Then substituting $u = \frac{1}{\log n}\log t$,

\begin{align*} n_n &\geq (\log n) \int_{0}^{1} \frac{3 \, \mathrm{d}u}{5 + 3(1 - u)n^{-u}\log n} - C' \\ &\sim \frac{3}{5}\log n \end{align*}

by the dominated convergence theorem. Therefore $n_n \sim \frac{3}{5}\log n$ and the desired conclusion follows by plugging this to $\sinh^{\circ n}\sqrt{\frac{3}{n}} = \sqrt{\frac{3}{\smash[b]{n_n}}}$. $\square$


Proof of Lemma. We assume $x > 0$ throughout this proof.

The second part of $\text{(1)}$ is the easiest one. From the inequality $\sinh t \geq t$ for $t \geq 0$, it follows that

\begin{align*} f(x) \leq \frac{3}{\left[ (3/x)^{1/2} \right]^2} = x. \end{align*}

Next, note that

$$ \sinh^2 t = \frac{\cosh(2t) - 1}{2} = \sum_{k=1}^{\infty} \frac{2^{2k-1}}{(2k)!} t^{2k} \leq \sum_{k=1}^{\infty} \frac{1}{3^{k-1}} t^{2k} = \begin{cases} \frac{t^2}{1 - t^2/3}, & 0 < t < \sqrt{3} \\ \infty, & t \geq \sqrt{3} \end{cases}. $$

Hence, $\frac{1}{\sinh^2 t} \geq \frac{1}{t^2} - \frac{1}{3}$ for all $t > 0$ and it follows that

$$ f(x) \geq 3 \left( \frac{1}{(\sqrt{3/x})^2} - \frac{1}{3} \right) = x - 1, $$

proving the second part of $\text{(1)}$.

Now we move on to proving $\text{(2)}$. The first part is evident from the asymptotic formula

\begin{align*} f(x) &= \frac{3}{\left[ (3/x)^{1/2} + \frac{1}{6}(3/x)^{3/2} + \frac{1}{120}(3/x)^{5/2} + \mathcal{O}(x^{-7/2}) \right]^2} \\ &= x - 1 + \frac{3}{5x} + \mathcal{O}(x^{-2}) \quad \text{as}\quad x \to \infty \end{align*}

For the second part, from $ \sinh t \geq t + \frac{t^3}{6} + \frac{t^5}{120}$, $t \geq 0$, it follows that

\begin{align*} f(x) &\leq \frac{3}{\left[ (3/x)^{1/2} + \frac{1}{6}(3/x)^{3/2} + \frac{1}{120}(3/x)^{5/2} \right]^2} \\ &= x - 1 + \frac{3}{5x} - \frac{2200x^3 + 1365x^2 + 315x + 27}{5x (40x^2 + 20x + 3)^2} \\ &\leq x - 1 + \frac{3}{5x}, \end{align*}

completing the proof of the lemma. $\square$

Sangchul Lee
  • 181,930
  • 1
    Genius, thanks so much for detailed elaboration! For one's information, we have $C=\dfrac{2}{7}$ here (of course, just for curiosity) by considering the expansion $\operatorname{csch}^2x=-(\coth x)'=\displaystyle\sum^{\infty}_{n=0}\dfrac{(-1)^n\times2(2n-1)\zeta(2n)}{\pi^{2n}}x^{2n-2}$, $0<|x|<\pi$. – Jianing Song Nov 03 '23 at 11:38