Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

For real $x,$ it's well-known that $$\Gamma^{-1}(x)\sim\frac{\log x}{\log\log x}$$

So a natural question is to bound $$G(x)=\Gamma^{-1}(x)\frac{\log\log x}{\log x}$$ which of course is 1 + o(1). Interestingly, its value is near 2 (that is, far from its asymptotic value) for many useful values of x: for example, $1.8<G(x)<2.1$ for $14<x<10^{77}.$

It seems that $G$ has a maximum near 3637.133905003816072848664328 of 2.01119450670696919822787997170113557148977275... and to decrease very slowly thereafter.

Question 1: Is the above the unique maximum?

Question 2: Is there an $x>5$ such that $G(x)<1$?

Question 3: Is there a useful factor or secondary term that makes this approximation more precise for useful values of x? I'm being intentionally vague on this point—if I knew exactly what I was looking for I probably wouldn't need to ask. :) For example, had I asked an analogous question about the prime-counting function, telling me about Li would be better than just giving the next asymptotic term $cx/\log^k x.$

share|cite|improve this question
I don't have the time to look now, but here's an idea: how about studying the graph of $\arctan(G(\tan(x)))$ within $(0,\pi/2)$? – J. M. May 3 '11 at 17:50
Can you provide a reference for your "well-known" fact? – GEdgar May 3 '11 at 19:09
Perhaps study $G(\Gamma(y)) = y \log\log\Gamma(y)/\log\Gamma(y)$. – GEdgar May 3 '11 at 19:15
@GEdgar: I can give sources using it, but I don't know of any addressing it directly -- it's folklore. – Charles May 3 '11 at 20:12
up vote 2 down vote accepted

I got a better approximation by fiddling with Stirling's formula out of Abramowitz and Stegun, just using the first term $$ \log \Gamma(z) \sim z \log z, $$ take $t = \Gamma(z)$ and the approximation $$ z_1 = \frac{L_1}{L_2} + \frac{L_1 L_3}{L_2^2}, $$ where $L_1 = \log t, \; \; L_2 = \log \log t, \; \; L_3 = \log \log \log t.$ I get $$ z_1 \log z_1 = \log t - \frac{L_1 L_3^2}{L_2^2} + \frac{L_1 L_3}{L_2^2} + smaller $$ which is an improvement on your folklore result, as the unwanted terms are actually smaller than the next term $z$ in the fuller $$ \log \Gamma(z) \sim z \log z - z - \frac{1}{2} \log z + \frac{1}{2} \log {2 \pi} + \log \left(1 + \frac{1}{12 z} + \frac{1}{288 z^2}- \cdots \right) $$ You ought to be able to do something with this for your original question.

In particular, it says your $$ G(t) \sim 1 + \frac{\log \log \log t}{\log \log t}, $$ meaning change is so slow that the limit is invisible to a computer, your experimental bounds may well be correct but certainty will be hard to come by.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.