Can someone suggest a way to calculate the maximum with respect to $x \ge 1$ of:
$$f(x)=\frac{1}{x!} \frac{1}{1-c^{1/\binom{x+n-1}{n-1}}}.$$
The constants $c$ and $n$ are parameters such that $c \in (0,1)$ and $n \in \mathbb{N}$.
If it helps, we can assume $x \in \mathbb{N}$ but it is not necessary.
I've tried looking at f(x+1)/f(x) which does simplifies things enough to solve it.
Working it out a bit, I suspect the maximum is for $x$ such that $\lfloor\sqrt{n} \rfloor -1 \le x \le \lfloor\sqrt{n} \rfloor$.
-
1$\begingroup$ What's the source of this problem? $\endgroup$– Janne KokkalaCommented Oct 2, 2014 at 9:49
-
4$\begingroup$ I'd use the gamma function, which is nice and smooth, and you can use calculus on it. en.wikipedia.org/wiki/Gamma_function $\endgroup$– Dima PasechnikCommented Oct 2, 2014 at 10:23
-
$\begingroup$ To JiK:The source of the problem is a bound on polynomial interpolation. The full explanation is not too long but it is to long too fit here. More infomation about where this type of function occure when studying interpolation bounds can be found here: Yu A Brudnyi and MI Ganzburg. On an extremal problem for polynomials in n variables. Mathematics of the USSR-Izvestiya, 7(2):345, 1973. $\endgroup$– gilCommented Oct 2, 2014 at 12:57
-
$\begingroup$ I echo Dima's comment, especially because $\Gamma$ is log-convex $\endgroup$– SuvritCommented Oct 2, 2014 at 17:55
1 Answer
In a quite large range of the parameters we can approximate the fraction by a Taylor series to obtain $$ f(x) = \frac{1}{x!}\frac{1}{\frac{-\log c}{\binom{x+n-1}{n-1}} + \mathcal{O}\left(\frac{\log^2 c}{\binom{x+n-1}{n-1}^2}\right)} = \left(1+ \mathcal{O}\left(\frac{\log^2 c}{\binom{x+n-1}{n-1}^2}\right)\right)\frac{\binom{x+n-1}{n-1}}{-x!\log c}, $$ which is valid for $-\log c<\binom{x+n-1}{n-1}$. Hence the function to be optimized is essentially $g(x)=\frac{1}{x!}\binom{x+n-1}{n-1}$. Now $\frac{g(x+1)}{g(x)}=\frac{x+n}{x(x+1)}$, hence the maximum is attained around $x=\sqrt{n}$. If $x$ differs from this optimum by more than 1, then $\frac{g(x+1)}{g(x)}$ differs from 1 by more than $\frac{1}{x}$, hence the error term in our formula for $f$ is negligible provided that $-\log c$ is small compared to $n^{\sqrt{n}}$. Whether this is enough depends on what your parameters evetually are. If $c$ is much closer to 0, then the maximum will be at much smaller values of $x$.
-
$\begingroup$ Any argument why it's a unique maxima ? Thanks, Gil. $\endgroup$– gilCommented Nov 3, 2014 at 13:35