1
$\begingroup$

Hello!

I want to prove that $x = 0.5$ is the global maximum of the function

$f(x) = \frac{(1-a)^2e^{(2x\cdot(x-1)a^2)}}{(1-a)(e^{(2x\cdot(2x-1)a^2)}+e^{((2x-1)\cdot(2x-2)a^2)})-2(1-2a)e^{(4x\cdot(x-1)a^2)}}$

where $a\in(0,1)$ and $x\in[0,1]$. I tried to show this using "conventional" ways, but unfortunately the attempts have failed.

Thanks a lot!

Improve this question
$\endgroup$
0

1 Answer 1

Reset to default
0
$\begingroup$

I have to run out now, so will write only the hint as of now. Later I might complete the details. Let me do one case here, the other one probably follows similarly.

Let $a \in (1/2,1)$, and write $a=1/2 + \epsilon$, for an appropriate $\epsilon$.

Now consider the function $1/f(x)$, which is (after some cleanup) seen to be

\begin{equation*} \frac{2 e^{-2 (-1+x) x \left(\frac{1}{2}+\epsilon \right)^2} \left(e^{2 (-1+x) (-1+2 x) \left(\frac{1}{2}+\epsilon \right)^2} (1-2 \epsilon )+e^{2 x (-1+2 x) \left(\frac{1}{2}+\epsilon \right)^2} (1-2 \epsilon )+8 e^{(-1+x) x (1+2 \epsilon )^2} \epsilon \right)}{(1-2 \epsilon )^2} \end{equation*}

But this is a nonnegative (since $2\epsilon < 1$) combination of convex functions of $x$ (because we have exponentials, products are ok), and is thus itself convex. Now, you can numerically or directly optimize this function and see that its minimum occurs at $x=0.5$, which implies that the maximum of your function occurs at $x=0.5$. (assuming it is easy to guarantee that $f(x)$ is a positive function).

Improve this answer
$\endgroup$
13
  • $\begingroup$ Thank you for your respond!. But, I think that there is a little mistake. The term:$e^{-2(\frac{1}{2}+\epsilon)^2x(x-1)}$is not convex, and hence the product is also not surely convex. $\endgroup$
    – Josh
    Commented Jul 22, 2012 at 9:05
  • $\begingroup$ By the way, you can assume that $f$ is non-negative. $\endgroup$
    – Josh
    Commented Jul 22, 2012 at 9:08
  • 1
    $\begingroup$ Ok, i tried to open the product and "luckily" the product is convex! For $\alpha\in(0,\frac{1}{2})$ we can use the symmetry of $f$ around $\frac{1}{2}$. $\endgroup$
    – Josh
    Commented Jul 22, 2012 at 9:56
  • $\begingroup$ Actually, this function is a speceific case of the function: $f(x) = \frac{(\int_0^{1-a}\exp(x\cdot(x-1)\cdot(g(x+a)-g(x))^2)dx)^2}{\int_0^{1-a}(\exp(2x\cdot(2x-1)\cdot(g(x+a)-g(x))^2)+\exp((2x-1)\cdot(2x-2)\cdot(g(x+a)-g(x))^2))dx-2\int_a^{1-a}\exp(x\cdot(x-1)\cdot(g(x+a)-g(x-a))^2)dx}$ where $g$ is some continues function. This function again has a globel maximum at $x = \frac{1}{2}$. If we substitute $f(x) = x$, it can be seen that we get the original function. Do you have any idea? Thanks again $\endgroup$
    – Josh
    Commented Jul 22, 2012 at 9:57
  • $\begingroup$ @MarkT1: I think the general case might also yield to some nice "variable substitutions" but it will require some playing around with the formula. But it seems more likely that your general case may succumb better to arithmetic-geometric mean inequalities for bounding the solution. Another lemma that might be useful is: if $g$, $h$ are convex and both nondecreasing (or nonincreasing) and positive on an interval, then the product $gh$ is also convex. This lemma holds only on convex functions on $R$. $\endgroup$
    – Suvrit
    Commented Jul 22, 2012 at 19:00

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .