I am reviewing Ramanujan's proof of Bertrand's Postulate which can be found here.

At step #7, he writes: "But it is easy to see that..."

$\log\Gamma(x) - 2\log\Gamma\left(\frac{1}{2}x + \frac{1}{2}\right) \le \log[x]! - 2\log\left[\frac{1}{2}x\right]!$

As $\Gamma(x+1) = [x]!$, this seems reasonable enough to me.

My question is: why did Ramanujan choose $\frac{1}{2}$. It seems like this would be true for any $n \ge 0$ where:

$\log\Gamma(x) - 2\log\Gamma\left(\frac{1}{2}x + n\right) \le \log[x]! - 2\log\left[\frac{1}{2}x\right]!$

Is that correct?



  • 1
    $\begingroup$ Having cracked a book and taken out my magnifying glass, I feel that you may have mis-stated (7). I would be the last to down-vote for a typo... $\endgroup$
    – daniel
    Mar 23 '13 at 23:33
  • $\begingroup$ Thanks for the tip. I'll check it and update my question. $\endgroup$ Mar 23 '13 at 23:58
  • $\begingroup$ @daniel: I checked it and it's possible that 1/3 may be 1/2. It does look like 1/3 to me. Is there any other area where you see a mistake? That's the only ambiguity that I see. $\endgroup$ Mar 24 '13 at 0:02
  • $\begingroup$ Thanks! I've change the detail to 1/2. Thanks very much for the correction! :-) $\endgroup$ Mar 24 '13 at 0:08
  • 2
    $\begingroup$ For those interested, here is an easier-to-read html version of Ramanujan's proof of Betrand's Postulate. $\endgroup$ Mar 24 '13 at 0:54

Inequality (7) is delicate. $\log x!-\log\Gamma(x)$ is a little bigger than $2\log (x/2)!-2\log\Gamma(x/2+1/2),$ which makes the lower bound tight. $\log\Gamma(x+1)$ is no different from $\log x!$ so he relies on $2\log\Gamma(x/2+1/2)$ being slightly less than $2\log(x/2)!$ to achieve the upper bound.

In (8) he refers to the Stirling approximation for the bounds in (7), so there may have been an element of computational convenience in the choice 1/2. Once he establishes the numerical bound in (8),(9) he has no further use for the $\log\Gamma$ functions used in (7).

There may be room for a constant other than 1/2 but anything you do will have an effect at both ends. $2\log\Gamma(x/2+3/4)$ might be closer to and still less than $2\log(x/2)!$ but will the left inequality hold? It appears not.

Whether 1/2 is optimal I am not sure but I would guess so. When you examine the asymptotic behavior of $\log\Gamma$ the derivative becomes important. Question 94582, though I can't link to it at the moment, concerned this property. The factor 1/2 in argument of $\log\Gamma$ (and 2 outside it) make it something to check as an explanation of the choice of 1/2.

Edit: In light of this question it appears that $1/2$ is needed to give desired asymptotic behavior to the l.h.s. and r.h.s. For completeness:

$PG(x) :=$ Polygamma(x). From (7):

$$\ln\Gamma(x) -2\ln\Gamma\left(\frac{x+1}{2}\right)\leq \cdot \leq \ln\Gamma\left(x+1\right)-2\ln\Gamma\left(\frac{x+1}{2}\right) $$

$$\frac{d}{dx}\left(\ln\Gamma(x)-2\ln\Gamma\left(\frac{x+1}{2}\right)\right)= PG(x)-PG\left(\frac{x+1}{2}\right)\sim \ln 2^-$$

$$\frac{d}{dx}\left(\ln\Gamma\left(x+1\right)-2\ln\Gamma\left(\frac{x+1}{2}\right)\right)= PG\left(x+1\right)-PG\left(\frac{x+1}{2}\right)\sim \ln 2^+$$

So $\log[x]! -2\log[x/2]!$ is trapped between two functions which approach $\log 2$ from above and below. The 1/2 is from $\frac{x+1}{2}$ and is essential to the asymptotic limits.


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.