Not sure if the following is easy to prove or not, I obtained the result using probabilistic arguments exclusively.

Define $x_1 = 2; x_2 = 4/3; x_3 = (8\cdot 6) / (7\cdot 5); x_4 = (16\cdot 14 \cdot 12 \cdot 10) / (15\cdot 13 \cdot 11 \cdot 9)$, and so on. Prove that

  • $x_k \rightarrow \sqrt{2}$.

  • $\sum_{k=1}^\infty (\log x_k)\cdot 2^{-k}= 1/2.$

A more challenging (unsolved problem): for which values of $k$ do we have $\log x_k < (\log 2)/2$ ? Maybe one can use the Stirling formula to solve this last problem.

  • $\begingroup$ I did some computations. Except for $k = 1$, we always have $\log x_k < (\log 2)/2$, that is, $x_k^2 < 2$. $\endgroup$ – Vincent Granville Aug 2 '19 at 1:06

Note that, for $k>1, x_k=\dfrac{y_k/y_{k-1}}{z_k/z_{k-1}}$ where $y_k=2^k(2^k-2)\cdot\cdot\cdot2=2^{2^{k-1}}(2^{k-1})!$

and $z_k=(2^k-1)(2^k-3)\cdot\cdot\cdot1=(2^k)!/y_k$.

Thus $x_k=\dfrac{y_k/y_{k-1}}{\dfrac{(2^k)!}{(2^{k-1})!}/\dfrac{y_{k-1}}{y_k}}= \dfrac{y_k^2}{y_{k-1}^2}\cdot\dfrac{(2^{k-1})!}{(2^k)!}$.

Substituting $y_k=2^{2^{k-1}}(2^{k-1})!$ and the corresponding expression for $y_{k-1}$ and simplifying yields $$x_k=\dfrac{2^{2^{k-1}}(2^{k-1})!^3}{(2^{k-2})!^2(2^k)!}.$$

Applying Stirling's formula ${\displaystyle n!\sim {\sqrt {2\pi n}}\left({\frac {n}{e}}\right)^{n}}$ then shows $x_k\sim\sqrt2$ after a lot of cancellation

[for example, $\sqrt{2\pi}^3$ in the numerator cancels $\sqrt{2\pi}^2\sqrt{2\pi}$ in the denominator, and

$\left(\dfrac1e\right)^{2^{k-1}3}$ in the numerator cancels $\left(\dfrac1e\right)^{2^{k-2}2}$$\left(\dfrac1e\right)^{2^{k}} $ in the denominator since $2^{k-1}3=2^{k-1}+2^k$].

  • $\begingroup$ Nice proof of the first bullet point. Does it help you to get the second one, or the inequality OP asks for? $\endgroup$ – Gerry Myerson Aug 2 '19 at 3:58
  • $\begingroup$ If I am not mistaken, the nice formula for $x_k$ does not seem to work for $k=1$. Am I wrong ? $\endgroup$ – Claude Leibovici Aug 2 '19 at 4:08
  • $\begingroup$ @ClaudeLeibovici: I think you are correct and I have inserted the qualification $k>1$; anyways, OP seems interested in asymptotics and not $k=1$ $\endgroup$ – J. W. Tanner Aug 2 '19 at 4:25
  • $\begingroup$ @GerryMyerson: thanks. it might, but I haven’t tried $\endgroup$ – J. W. Tanner Aug 2 '19 at 4:26
  • $\begingroup$ I am sorry, I did not see $k>1$ ! Nice answer and $\to +1$. $\endgroup$ – Claude Leibovici Aug 2 '19 at 4:41

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.