1

Let $0<x<0.5$ or $0.5<x<1$ defines the following function : $$f(x)=x^{4(1-x)^2}+(1-x)^{4x^2}$$

Claim : The function $g(x)=\exp(|1-f(x)|)-1$ is a logarithmically concave function

Or : The function $h(x)=\ln(g(x))=\ln(\exp(|1-f(x)|)-1)$ is a concave function

The second derivative is very ugly and I prefer this form

Furthermore I try to use the definition of a logarithmically concave function but it gives a lot of difficulties .

I try also the Popoviciu's inequality but it requires to be continuous so I think it collapses .

If someone have a good idea it will be great .

Thanks a lot for your time and patience .

Edit

An example : If we use Popoviciu's inequality where $x=0.5-\varepsilon$ and $y=a$ and $z=0.5-a$ where $a\in[0.1,0.4]$ and $0<\varepsilon<10^{-2}$ we have :

$$\frac{h(0.5-\varepsilon)+h(a)+h(0.5-a)}{3}+h\Big(\frac{0.5-\varepsilon+a+0.5-a}{3}\Big)$$ Or : $$\frac{h(0.5-\varepsilon)+h(a)+h(0.5-a)}{3}+h\Big(\frac{1-\varepsilon}{3}\Big)$$

And :

$$\frac{2}{3}\Big(h\Big(\frac{0.5-\varepsilon+a}{2}\Big)+h\Big(\frac{0.5-\varepsilon+0.5-a}{2}\Big)+h\Big(\frac{0.5-a+a}{2}\Big)\Big)$$

Or:$$\frac{2}{3}\Big(h\Big(\frac{0.5-\varepsilon+a}{2}\Big)+h\Big(\frac{1-\varepsilon-a}{2}\Big)+h\Big(\frac{0.5}{2}\Big)\Big)$$

Now when $\varepsilon$ tends to $0$ the first quantity tends to $-\infty$ and the second is finite so it prove the partially the claim.

I have this :

Defines :$$p(x)=2\sqrt{x^{4(1-x)^2}(1-x)^{4x^2}}$$ And:$$q(x)=\ln(\exp(|1-p(x)|)-1)$$

Then $$h''(x)\leq q''(x)\leq 0$$

Barackouda
  • 3,879
  • What makes you think that $g$ is logarithmically concave? Why would that be an interesting result? (Is it an attempt to solve https://math.stackexchange.com/q/2176571/42969?) – Martin R Jan 22 '20 at 14:15
  • @MartinR Yes it's a try to prove the inequality in your link . – Barackouda Jan 22 '20 at 14:38
  • @Macavity You can check that the second derivative or $h''(x)$ is less than $-100$ – Barackouda Jan 22 '20 at 14:40
  • @Macavity around the maximum of $h''(x)$ we have : https://www.wolframalpha.com/input/?i=second+derivative+ln%28exp%28%7C1-%28x%5E%284%281-x%29%5E2%29%2B%281-x%29%5E%284x%C2%B2%29%29%7C%29-1%29+for+x%3D0.28 – Barackouda Jan 22 '20 at 14:42
  • Plot $h(x)$ do you think this is concave in the interval? https://www.wolframalpha.com/input/?i=Plot+ln%28exp%28%7C1-%28x%5E%284%281-x%29%5E2%29%2B%281-x%29%5E%284x%C2%B2%29%29%7C%29-1%29+for+0%3Cx%3C1 – Macavity Jan 22 '20 at 15:02
  • @Macavity: $x =0.5$ was excluded in the question, so my guess would be that they are trying to prove the claim in $(0,1/2)$ and in $(1/2, 1)$ separately. Of course that should be clarified in the question itself. – Martin R Jan 22 '20 at 15:33
  • @Martin, that’s possible. However if the inequality referred to is the purpose, then it needs to hold across the interval, else there will be cases across the two sub-intervals where it cannot be used. – Macavity Jan 22 '20 at 15:36
  • @The.old.boy There are two subtle different definitions for log-concavity of a function $F(x)$ (see wikipedia or wolfram):
    1. $F > 0$ and $\log F(x)$ is concave.

    2. $F(x) \ge 0$ but $F(c)=0$ for some $c \in \mathrm{dom} F$, and $F(tx + (1-t)y) \ge F(x)^t F(y)^{1-t}$ for all $x, y\in \mathrm{dom} F$ and $0< t < 1$.

    For this case, one can not take logarithm $\log F(x)$.

    For you problem, only if $f(x) \le 1$ for $0 < x < \frac{1}{2}$, you can consider the log-concavity of $h(x)$. If you want to take derivative for $h(x)$, you need $f(x) < 1$ for $0 < x < \frac{1}{2}$.

    – River Li Feb 08 '20 at 02:35
  • @RiverLi Thanks for the advice but it's not sufficient to have $f(x)\neq 1$? – Barackouda Feb 10 '20 at 15:30
  • @The.old.boy Before you start to analyze the log-concavity of $g(x)$, you have to prove that $f(x) < 1$ for $0 < x < \frac{1}{2}$. – River Li Feb 10 '20 at 15:48

0 Answers0