It's a follow up generalization of An alternative lower bound for $\prod_{ i,j = 1}^n\frac{1+a_ia_j}{1-a_ia_j} $
Conjecture :
Let $f(x)$ be on $-1<x<1$ a strictly increasing function such that :
$1)f(x)=-f(-x)$
$2)f(1)=\infty$
$3)\forall x \in[0,1) f''(x)\geq 0,\forall x \in(-1,0] f''(x)\leq 0$
$4)\forall x\in (-1,1) f'''(x)>0$
Then for $a_i\in(-1,1)$ we have :
$$\sum_{i,j=1}^{n}f\left(a_{i}a_{j}\right)>n^{2}f\left(\sum_{i,j=1}^{n}\frac{a_{i}a_{j}}{n^{2}}\right)$$
As hint for a proof I used Fuch's inequality which is in itself an extension of Karamata's inequality for example in the case $n=2$ and $sgn(ab)=-1,1>a^2\geq b^2\geq ab>-1$ the vector $(a^2,b^2,-ab,-ab)$ majorize the vector $\left(\frac{a^{2}+b^{2}+2ab}{4},\frac{a^{2}+b^{2}+2ab}{4},-\frac{a^{2}+b^{2}+2ab}{4},-\frac{a^{2}+b^{2}+2ab}{4}\right)$ with weight $(1,1,-1,-1)$
For the general case remark that at some point the majorization decreases (decreasing minus increasing )and then at the end it's equal to zero .
Edit :
The draft is incorrect but we can use majorization in symmetric form see https://arxiv.org/abs/0803.2958
Is this draft correct ? If not have you a proof or a counter-example ?
The idea I had was to only focus on $[0,1]$ where $f$ is convex and try to prove we can extract $2k(n-k)$ of the $k^2 +(n-k)^2 $ which can be rearranged to majorize the negative $2k(n-k)$ numbers from the set ${a_ia_j\vert ,1\leq i,j\leq n}$, where we assume $a_1,a_2,\dots a_k$ are the negatives.
– dezdichado Apr 12 '23 at 17:34