3
$\begingroup$

It is known that $$ \cos(\frac{x}{2})\cos(\frac{x}{4})\cos(\frac{x}{8})\dots = \frac{\sin x}{x} = O_{x \rightarrow \infty}(x^{-1}) $$

Is it true that $$ f(x) = \cos(\frac{x}{3})\cos(\frac{x}{9})\cos(\frac{x}{27})\dots = o_{x \rightarrow \infty} (1) ? $$ If so, what is the rate of convergence?

It seems to me that $f(x)$ converges to zero, but very slowly. For example $f(1081882100) \approx 0.27$. I guess the reason is that $f(x)$ is the Fourier transform of the uniform distribution on the Cantor set $C$ supported in $[-1/2, 1/2]$, which is highly irregular. To see this, let $X \sim \mathcal U (C)$, then by the self-similarity of $C$, we have $$ X \stackrel{(d)}{=} X/3 + Y $$ where $Y \sim \mathcal U(\pm 1/3)$ is independent of $X$. So $$ \mathbb E[e^{itX}] = \mathbb E[e^{itX/3}]\cos(x/3) $$ From which we obtain $$ \mathbb E[e^{itX}] = \cos(\frac{x}{3})\cos(\frac{x}{9})\cos(\frac{x}{27})\dots $$

Update. The answer is no by Noam Elkies. However now I want to ask the same question for $$ f_a(x) = \prod_{n \geq 1} \cos(\frac{x}{a^k}) $$ for $a>1$ and $a \neq 2$.

$\endgroup$
3
  • 7
    $\begingroup$ you will want to ask a new question, which can then be answered and accepted; combining two questions into one is not advised. $\endgroup$ Commented Jul 17, 2022 at 19:51
  • 1
    $\begingroup$ At the beginning I had in mind the general question but only asked for $a=3$ thinking an answer for this particular case works for all $a>2$. As it turns out the answer is simple but hard to extend to a continuous domain. It seems that the number-theoretic aspect of $a$ plays an essential role here. $\endgroup$ Commented Jul 17, 2022 at 21:30
  • 1
    $\begingroup$ I doubt this product has a closed form expression for general $a$. The closest expression I know is for $a=3$ $$\prod_{n=1}^\infty \frac{4\cos^2(x/3^k)-1}{3}=\frac{\sin x}{x}$$ $\endgroup$ Commented Jul 17, 2022 at 23:12

1 Answer 1

9
$\begingroup$

No. If $x = 3^n \pi$ then $|f(x)| = f(\pi) \neq 0$ (numerically it's about $0.466$), and $3^n \pi$ can be arbitrarily large.

Such a construction fails for $\prod_{k=1}^\infty \cos(x/2^k)$ because if some $x/2^k$ is nearly $\pi$, or more generally some odd multiple of $\pi$, then $x/2^{k+1}$ is nearly a half-integral multiple of $\pi$ and thus has a very small cosine.

[added later] For $f_a(x) := \prod_{k=1}^\infty \cos(x/a^k)$ with $a > 2$, the same construction does work if $a$ is an integer (consider $x = a^n \pi$), and more generally if $a$ is a Pisot-Vijayaraghavan number (which need not exceed $2$, e.g. if $a = (1+\sqrt5)/2$ then there are arbitrarily large $x$ such that $f_a(x)$ remains bounded away from zero).

$\endgroup$
2
  • 1
    $\begingroup$ I presume this means that this MSE answer (the red formula) is incorrect, since that does seem to converge to zero for large $x$. $\endgroup$ Commented Jul 17, 2022 at 19:49
  • 3
    $\begingroup$ @CarloBeenakker No, because that product has a finite radius of convergence (as it must because $\log f(x)$ is singular at odd multiples of $3\pi/2$). $\endgroup$ Commented Jul 17, 2022 at 20:07

You must log in to answer this question.

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