Let $n$ be a positive integer and $x_1,...,x_n$ positive reals such that $x_1+\dots+x_n=1$. Prove that $$\prod_{i=1}^n\left(\frac{1}{x_i^2}-1\right)\ge (n^2-1)^n.$$
Now notice that if we apply Jensen's Inequality to $$f(x)=\ln\left(\frac{1}{x_i^2}-1\right)$$ We would get the exact inequality...except that $f$ is not convex on $(0,1)$. But I do know this,
Claim: If for all $k\le n$ we have $0< a_k\le 1/2$. Then, $$\prod_{i=1}^n\left(\frac{1}{a_i}-1\right)\ge \left(\frac{n}{a_1+...+a_n}-1\right)^n$$
The proof is not too hard. Just consider $$g(x)=\ln\left(\frac{1}{x}-1\right)$$
$g$ is convex on $(0,1/2)$ and you can finish with Jensen. Now what if we let $a_i=x_i^2$? Well, we would get $$\prod_{i=1}^n\left(\frac{1}{x_i^2}-1\right)\ge \left(\frac{n}{x_1^2+...+x_n^2}-1\right)^n\ge (n^2-1)^n$$
Because $x_1+...+x_n=1$ implies, $$\sum_{i=1}^n x_i^2\ge \frac{1}{n}\left(\sum_{i=1}^n x_i\right)^2=\frac{1}{n}$$ And we're done...Or are we? The argument is only valid if $x_i^2\le 1/2$ for all $i$. Well if there exists $x_i^2>1/2$ Can we still get the inequality?
By the way there is a way to solve the inequality without any use of Jensen. But I want to know if we can solve it like that.