9
$\begingroup$

Question 1:

Let $x_{i}>0$, ($i=1,2,\cdots,n$) and such that $$x_{1}+x_{2}+\cdots+x_{n}=\pi.$$ Show that $$ \dfrac{\sin{x_{1}}\sin{x_{2}}\cdots\sin{x_{n}}}{\sin{(x_{1}+x_{2})}\sin{(x_{2}+x_{3})}\cdots\sin{(x_{n}+x_{1})}}\le\left(\dfrac{\sin{\frac{\pi}{n}}}{\sin{\frac{2\pi}{n}}}\right)^n $$ Question 2 (may not hold): if $f''(x)\le 0,x\in I$, can we prove the following inequality? $$ \begin{split} f(x_{1}+x_{2})+&f(x_{2}+x_{3})+\ldots+f(x_{n}+x_{1})+nf\left(\dfrac{x_{1}+x_{2}+\ldots+x_{n}}{n}\right)\\ &\ge f(x_{1})+f(x_{2})+\ldots+f(x_{n})+nf\left(\dfrac{2(x_{1}+x_{2}+\ldots+x_{n})}{n}\right), \end{split} $$ where $x_{i}\in I$, $i=1,2,3\ldots,n$. I tried everything, but failed.


As an example of Question 2, consider $f(x)=\ln{\sin{x}}$, $0<x<\pi$. Since $$f''(x)=-\csc^2{x}<0$$ it suffices to prove that $$ \begin{split} f(x_{1}+x_{2})+f(x_{2}+x_{3})+&\ldots+f(x_{n}+x_{1})+nf\Big(\dfrac{\pi}{n}\Big)\\ &\ge f(x_{1})+f(x_{2})+\ldots+f(x_{n})+nf\Big(\dfrac{2\pi}{n}\Big) \end{split} $$ or $$ \begin{split} f(x_{1}+x_{2})+&f(x_{2}+x_{3})+\ldots+f(x_{n}+x_{1})+nf\left(\dfrac{x_{1}+x_{2}+\ldots+x_{n}}{n}\right)\\ &\ge f(x_{1})+f(x_{2})+\ldots+f(x_{n})+nf\left(\dfrac{2(x_{1}+x_{2}+\ldots+x_{n})}{n}\right). \end{split}$$ In other words,if $f''(x)\le 0$, can we prove following inequality? $$ \begin{split} f(x_{1}+x_{2})+&f(x_{2}+x_{3})+\ldots+f(x_{n}+x_{1})+nf\left(\dfrac{x_{1}+x_{2}+\ldots+x_{n}}{n}\right)\\ &\ge f(x_{1})+f(x_{2})+\ldots+f(x_{n})+nf\left(\dfrac{2(x_{1}+x_{2}+\ldots+x_{n})}{n}\right)? \end{split} $$

$\endgroup$
7
  • $\begingroup$ It is certainly not that general. The last inequality may fail for a general concave function. $\endgroup$
    – fedja
    May 29, 2019 at 1:18
  • $\begingroup$ The questions arise: what for? where could it be applied? $\endgroup$
    – user64494
    May 29, 2019 at 7:33
  • $\begingroup$ math.stackexchange.com is a right forum for such type questions. $\endgroup$
    – user64494
    May 29, 2019 at 7:42
  • $\begingroup$ did you check the inequality $f(x+y)-\frac12(f(x)+f(y))+f(\pi/n)-f(2\pi/n)\geqslant c(x+y-2\pi/n)$ for $c=f'(2\pi/n)-\frac12 f'(\pi/n)$? $\endgroup$ May 29, 2019 at 9:38
  • 1
    $\begingroup$ "and the numerator grows" Why? $\endgroup$
    – fedja
    Jun 21, 2019 at 2:44

2 Answers 2

3
$\begingroup$

It is actually quite a cute problem except it is only 1/3-analysis and 2/3 elementary geometry. The analysis part is that if you have $2$ positive numbers $A,B$, then for any $p,q$ with $\frac 1p+\frac 1q=1$, we have $A+B=\frac{pA}{p}+\frac{qB}q\ge p^{1/p}q^{1/q}A^{1/p}B^{1/q}$ (Young) and if $pA=qB$, you have equality.

The geometry part is Ptolemy's theorem.

Now let $d,d'$ be some consecutive diagonals of order $k\in[2,n-2]$ (sides are order $1$ (or $n-1$), diagonals spanning $2$ sides are order $2$ (or $n-2$, etc.) in an inscribed $n$-gon. Let $a,b$ be the diagonals of orders $k-1,k+1$ and $c,c'$ be the sides so that $d,d'$ are the diagonals of the quadrilateral with the sides $a,c,b,c'$. We have $$ dd'=ab+cc'\ge p^{1/p}q^{1/q}(ab)^{1/p}(cc')^{1/q} $$ and, most importantly, we can choose $p,q$ depending on $k$ only so that we have an identity for the regular $n$-gon. Multiplying over all choices of the pair $d.d'$, we get an inequality of the kind $$ {\prod}_k\ge c_k\left[{\prod}_{k-1}{\prod}_{k+1}\right]^{\alpha_k}{\prod}_1^{1-2\alpha_k} $$ where $\prod_k$ is the product of diagonals of order $k$, $c_k>0$ and $\alpha_k\in(0,\frac 12)$ are some appropriate numbers.

The rest is trivial. If you want more analytic flavor, just iterate this inequality like crazy until everything except $\prod_1$ wears out on the RHS, but you can also do it by completely elementary means (you have some fancy kind of log-concavity here). The upshot is that you get an inequality $\prod_k\ge C_k\prod_1$ in which the regular $n$-gon produces an identity. Now just take $k=2$.

I'm answering here on MO because the MSE thread is cluttered enough already, but your best bet for asking such stuff is, probably, AoPS.

$\endgroup$
3
  • $\begingroup$ Apparently, there are infinitely many different points where the inequality is achieved, at least in the case for $n = 4$ and $n=5.$ I am kind of failing to see how these points will be obtained from your solution. $\endgroup$
    – dezdichado
    Jul 3, 2019 at 21:05
  • $\begingroup$ @dezdichado For $n=4$ it is easy (see my response to Fedor Petrov in the comments to the question). $n=5$ is a bit trickier to understand. $\endgroup$
    – fedja
    Jul 3, 2019 at 23:34
  • $\begingroup$ I actually worked out the configuration for $n = 5,$ but exhausting all of the equilibrium points was rather tedious. $\endgroup$
    – dezdichado
    Jul 4, 2019 at 19:43
2
$\begingroup$

The answer to your Question 2 is negative. E.g., let $f(x)=-\max(0,x-1)$ (or a smooth concave approximation to $f(x)$), $n=3$, $x_1=1/2,x_2=0,x_3=1$. Then your inequality becomes $-1/2\ge0$, which is false.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.