3
$\begingroup$

I am looking for bounds on the double factorial for even and odd $n \in \mathbb{N}$, defined as

$$ n!! = n \cdot (n-2) \cdot (n-4) \dots $$

For example, $9!! = 9 \cdot 7 \cdot 5 \cdot 3 \cdot 1$ and $8!! = 8 \cdot 6 \cdot 4 \cdot 2$, see Double factorial. It is clear for me that

$$ n! = (n!!)((n-1)!!) $$

holds, such that if I have some bounds for the factorial

$$ b^-_n \leq n! \leq b^+_n $$

I can obtain bounds for the double factorial

$$ \frac{b^-_n}{(n-1)!!} \leq n!! \leq \frac{b^+_n}{(n-1)!!} $$

which can be used recursively, e.g.

$$ \frac{b^-_1 b^-_3 b^-_5}{b^+_2 b^+_4} \leq 3!! \frac{b^-_5}{b^+_4} \leq \frac{b^-_5}{4!!} \leq 5!! \leq \frac{b^+_5 b^+_3 b^+_1}{b^-_4 b^-_2} \ . $$

I have used bounds $b^\pm_n$ based on Stirling's approximation, see Speed of convergence and error estimates, but due to the recursive use, the estimate is too far away for large $n$.

Question: do you know better (explicit) bounds for the double factorial?

Edit (2018-06-28): In order to provide a clearer view concerning Alex's comment, gammatester comment and and Gerry's answer (see his answer), consider the factorial bounds, see Stirling's approximation $$ b^\pm_n = c^\pm_n \beta_n \ , \quad \beta_n = n^{n+1/2} e^{-n} \ , \quad c^-_n = \sqrt{2\pi} \ , \quad c^+_n = e \ . $$ (It should be remarked that alternative bounds are also given by $c^-_n = \sqrt{2\pi} e^{1/(12n+1)}$ and $c^+_n = \sqrt{2\pi} e^{1/(12n)}$, see Speed of convergence and error estimates). The double factorial of even numbers, as remarked by Alex (see his comment), can be expressed and bounded by the factorial bounds as $$ (2k)!! = k! \cdot 2^k \Rightarrow 2^k b^-_{k} \leq (2k)!! \leq 2^k b^+_{k} \quad k \in \mathbb{N} \ . $$ The asymptotic $$\alpha_n = n^{(n+1)/2} e^{-n/2}$$ of Vaclav Kotesovec pointed at by Gerry fulfills for even $n=2k, k \in \mathbb{N}$ $$ 2^k b^-_{k} = \sqrt{\pi}\alpha_{2k} $$ such that the asymptotic delivers lower and upper bounds ($2^k b^+_k = \sqrt{2e}\alpha_{2k}$) for even $n=2k$. Interestingly, for odd $n = 2k+1$, the Vaclav's asymptotic seems to fulfill (I just checked for $k=0,1,2,3,\dots,10^4$ in Mathematica) $$ \sqrt{2} \alpha_{2k+1} \leq (2k+1)!! \leq \sqrt{e} \alpha_{2k+1} $$ but I am do not know, if this always holds, or something better than $e$ for an upper bound is possible. Alternatively, you can use the bounds based on the relations to the factorial, see Alex's comment, $$ (2k+1)!! = \frac{(2k+1)!}{k! \cdot 2^k} \geq \frac{b^-_{2k+1}}{k! \cdot 2^k} \geq \frac{b^-_{2k+1}}{b^+_k \cdot 2^k} \ . $$ If you are interested in the use of the $\Gamma(x)$ functions, as suggested by gammatester (see his comment), you may want to use the results of the work of Necdet Batir or Necdet Batir (2017). Thanks to all for the suggestions.

|cite|improve this question
$\endgroup$
  • 1
    $\begingroup$ Since $(2n)!!=n!·2^n$ and $(2n+1)!!=\dfrac{(2n+1)!}{(2n)!!}=\dfrac{(2n+1)!}{n!·2^n}$, it reduces to find better bounds for $m!$. $\endgroup$ – Saad Jun 26 '18 at 12:47
  • 1
    $\begingroup$ How about using bounds for the Gamma function or the ordinary factorial with the relations \begin{align*} (2k)!! &= 2^k k!\\ (2k+1)!! &= (2k+1)!/(2^k k!) = \frac{2^{k+1}}{\sqrt{\pi}}\,\Gamma\left(k+\tfrac{3}{2}\right) \end{align*} $\endgroup$ – gammatester Jun 26 '18 at 12:49
  • $\begingroup$ @AlexFrancisco thank you, you are right, I have to get better bounds for the factorial. $\endgroup$ – Mauricio Fernández Jun 26 '18 at 13:12
  • $\begingroup$ @gammatester hmmmm, .... that would be intersting, specially using dlmf.nist.gov/5.6#E8 (last equation) $\endgroup$ – Mauricio Fernández Jun 26 '18 at 13:15
  • $\begingroup$ Any thoughts on my answer, Mauricio? $\endgroup$ – Gerry Myerson Jun 28 '18 at 3:17
1
$\begingroup$

The double factorial is tabulated at http://oeis.org/A006882 where there are links to the literature and the asymptotic expression $cn^{(n+1)/2}e^{-n/2}$ where $c=\sqrt{\pi}$ if $n$ is even, $\sqrt2$ if $n$ is odd, attributed to Vaclav Kotesovec.

|cite|improve this answer
$\endgroup$
  • 1
    $\begingroup$ Hi Gerry, I took a look at Veclav's asymptotic. I updated my question, you may want to have a look. Thanks for the reference. It actually gave me an idea for another problem :D $\endgroup$ – Mauricio Fernández Jun 28 '18 at 10:38

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.