5
$\begingroup$

let $x\in (0,1)$, and $k$ be postive intgers,such $k\ge 2$,

I conjecture following inequality maybe hold? $$x\cdot \dfrac{(1-x)^{k-1}}{(k+1)^{k-2}}+\dfrac{(1-2x)^k}{k^k}\le \dfrac{1}{(k+2)^{k-1}}$$ creat by wang yong xi

This is my attempt

when $k=2$,then inequality can be written as $$x\cdot (1-x)+\dfrac{(1-2x)^2}{4}\le\dfrac{1}{4}$$ it is obviously true.

when $k=3$then inequality can be written as $$\dfrac{x(1-x)^2}{4}+\dfrac{(1-2x)^3}{27}\le\dfrac{1}{25}$$ or $$-\dfrac{(5x-1)^2(5x+8)}{2700}\le 0$$ it is clearly true.

when $k=4$ it's equivalent $$\dfrac{x(1-x)^3}{25}+\dfrac{(1-2x)^4}{256}-\dfrac{1}{216}\le 0$$ or $$\dfrac{(6x-1)^2(108x^2+12x-125)}{172800}\le 0$$ it is clearly

when $k=5$, it's equivalent $$x\cdot\dfrac{(1-x)^4}{216}+\dfrac{(1-2x)^5}{5^5}-\dfrac{1}{7^4}=-\dfrac{(7x-1)^2(185563x^3-181202x^2-127589x+156384)}{1620675000}<0$$ But I can't prove for any postive intgers $k$.and I have found $$LHS-RHS=[(k+2)x-1]^2\cdot h(x,k)$$.so we must prove $h(x,k)\le 0$ Thanks

$\endgroup$
5
  • 1
    $\begingroup$ Note that it is an equality at $x=1/(k+2)$. Use the derivatives of RHS-LHS to investigate whether this is indeed a minimum. $\endgroup$ Commented Oct 25, 2017 at 11:33
  • $\begingroup$ @inequality, can you write $h(x,k)$ explicitly? $\endgroup$
    – Ron P
    Commented Oct 25, 2017 at 11:39
  • $\begingroup$ @JulesLamers derivative at $x=1/(k+2)$ is zero, but it does not immediately yield the inequality $\endgroup$ Commented Oct 25, 2017 at 11:45
  • 1
    $\begingroup$ Using the polsturm(P,[0,1]) command of Pari/GP, you can check in a few seconds that the inequality is true up to at least $k=500$. $\endgroup$ Commented Oct 25, 2017 at 11:51
  • $\begingroup$ @HenriCohen Likewise using Reduce in Mathematica $\endgroup$ Commented Oct 25, 2017 at 11:53

1 Answer 1

8
$\begingroup$

Consider only $k>2$. Denote $$f(x)=x\cdot \frac{(1-x)^{k-1}}{(k+1)^{k-2}}+\frac{(1-2x)^k}{k^k},$$ we need to prove that $f(x)\leqslant f(\frac1{k+2})$ for $x\in [0,1]$. We have $f'(\frac1{k+2})=0$ and $$f'(0)=(k+1)^{2-k}-2k^{1-k}=k^{2-k}\left(\left(1+\frac1k\right)^{2-k}-\frac2k\right)>0$$ by Bernoulli inequality $(1+x)^a>1+ax$ for $a=2-k$, $x=1/k$. Also $f(1)$ equals $\pm f(0)$. It means that the maximal value of $f$ on $[0,1]$ is attained at an interior point $a\in (0,1)$, and thus $f'(a)=0$.

I claim that $f'$ has unique root (multiplicity counted) on $(0,1/2]$, and this root is $\frac1{k+2}$, and at most one root on $[1/2,1)$. In both cases the only possible maximum point is $\frac{1}{k+2}$ (the second extremal point would be a local minimum)

We have $$f'(x)=(1-x)^{k-2}(1-kx)(k+1)^{2-k}-2(1-2x)^{k-1}k^{1-k}=0.$$ Denote $y=\frac{1-x}{1-2x}$, then $x=\frac{1-y}{1-2y}$, $1-kx=\frac{(k-2)y-(k-1)}{1-2y}$. Our equation $f'(x)=0$ in terms of $y$ rewrites as $$y^{k-2}((k-2)y-(k-1))(k+1)^{2-k}+2k^{1-k}=0.$$ LHS is monotone in $y$ for $y\in (1,\infty)$ (corresponds to $x\in (0,1/2)$) and for $y<0$ (corr. $x\in (1/2,1)$), that implies the above claim.

$\endgroup$
5
  • $\begingroup$ I think the Bernoulli inequality $(1+x)^a>1+ax$, the $a$ must $a>1$. but $2-k<0$ $\endgroup$
    – math110
    Commented Oct 25, 2017 at 13:32
  • $\begingroup$ it works for negative $a$ too (and for $a\in (0,1)$ with the opposite sign). the function $(1+x)^a$ is convex, and we compare it with the tangent at $x=0$. $\endgroup$ Commented Oct 25, 2017 at 13:35
  • $\begingroup$ ah,Nice Thanks, I have understand,+1 $\endgroup$
    – math110
    Commented Oct 25, 2017 at 13:45
  • $\begingroup$ why the maximum point is interior $(0.1/2]$? maybe is interior on $[1/2,1)$.because there is a root on $[1/2,1)?$ can you explain more detail? Thanks $\endgroup$
    – math110
    Commented Jun 3, 2018 at 16:06
  • $\begingroup$ @functionsug if $x_0\geqslant 1/2$ is another root of $f'$, we have $f'<0$ on $[\frac1{k+2},x_0]$, thus it can not be a maximum point $\endgroup$ Commented Jun 8, 2018 at 3:51

You must log in to answer this question.

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