inspired (again) by an inequality of Vasile Cirtoaje I propose my own conjecture :

Let $x,y>0$ such that $x+y=1$ and $n\geq 1$ a natural number then we have : $$x^{\left(\frac{y}{x}\right)^n}+y^{\left(\frac{x}{y}\right)^n}\leq 1$$

First I find it very nice because all the coefficient are $1$ .

I have tested with Geogebra until $n=50$ without any counter-examples.

Furthermore we have an equality case as $x=y=0.5$ or $x=1$ and $y=0$ and vice versa .

To solve it I have tried all the ideas here

My main idea was to make a link with this inequality (my inspiration) see here

So if you can help me to solve it or give me an approach...

...Thanks for all your contributions !

Little update

I think there is also an invariance as in question here Conjecture $a^{(\frac{a}{b})^p}+b^{(\frac{b}{a})^p}+c\geq 1$

Theoretical method

Well,Well this method is very simple but the result is a little bit crazy (for me (and you ?))

Well ,I know that if we put $n=2$ we can find (using parabola) an upper bound like

$$x^{\left(\frac{1-x}{x}\right)^2}\leq ax^2+bx+c=p(x)$$ And $$(1-x)^{\left(\frac{x}{1-x}\right)^2}\leq ux^2+vx+w=q(x)$$

on $[\alpha,\frac{1}{2}]$ with $\alpha>0$ and such that $p(x)+q(x)<1$

In the neightborhood of $0$ we can use a cubic .

Well,now we have (summing) :

$$x^{\left(\frac{1-x}{x}\right)^2}+(1-x)^{\left(\frac{x}{1-x}\right)^2}\leq p(x)+q(x)$$

We add a variable $\varepsilon$ such that $(p(x)+\varepsilon)+q(x)=1$

Now we want an inequality of the kind ($k\geq 2$):

$$x^{\left(\frac{1-x}{x}\right)^{2k}}+(1-x)^{\left(\frac{x}{1-x}\right)^{2k}}\leq (p(x)+\varepsilon)^{\left(\frac{1-x}{x}\right)^{2k-2}}+q(x)^{\left(\frac{x}{1-x}\right)^{2k-2}}$$

Now and it's a crucial idea we want something like :

$$\left(\frac{x}{1-x}\right)^{2k-2}\geq \left(\frac{1-(p(x)+\varepsilon)}{q(x)}\right)^y$$


$$\left(\frac{1-x}{x}\right)^{2k-2}\geq \left(\frac{1-q(x)}{p(x)+\varepsilon}\right)^y$$

Now it's not hard to find a such $y$ using logarithm .

We get someting like :

$$x^{\left(\frac{1-x}{x}\right)^{2k}}+(1-x)^{\left(\frac{x}{1-x}\right)^{2k}}\leq q(x)^{\left(\frac{1-q(x)}{q(x)}\right)^{y}}+(1-q(x))^{\left(\frac{q(x)}{1-q(x)}\right)^{y}}$$

Furthermore the successive iterations of this method conducts to $1$ because the values of the differents polynomials (wich are an approximation of the initial curve) tend to zero or one (as abscissa).

The extra-thing (and a little bit crazy) we can make an order on all the values.

My second question

Is it unusable as theoretical\practical method ?

  • $\begingroup$ Have you already tried the limit cases around 1 and 0.5? $\endgroup$ Jul 15, 2020 at 17:15
  • $\begingroup$ @Biggus Dickus Python, Does this help you to prove your conjecture ? $\endgroup$ Jul 15, 2020 at 20:57
  • $\begingroup$ @zeraoulia rafik interesting but it seems to be not sufficient to prove the entire inequality.Anyway with this I have a special case .(+1) for your question. $\endgroup$
    – Erik Satie
    Jul 17, 2020 at 13:37
  • $\begingroup$ Oups this not real number but natural number ah ! Anyway have a good day. :-) $\endgroup$
    – Erik Satie
    Jul 17, 2020 at 13:43
  • $\begingroup$ @ErikSatie: I would replace $n$ with a positive variable and try Method of Lagrange Multipliers to see if we can extract some thing. $\endgroup$
    – Bumblebee
    Dec 6, 2020 at 3:00

2 Answers 2


I am sorry for not proving your conjecture, but I thought of writing up my thoughts as it might help you.

Without loss of generality, we can say that $y \geq 0.5$. Let $q=y/x$ and because $y \geq 0.5$ we have $q\geq 1$ (you can do everything in reverse with $x \geq 0.5$ and $q\leq 1$). Let $a_0=x$ and $b_0=y$. Then we can write: $$ x^{\left( \frac{y}{x} \right)^n} + y^{\left(\frac{x}{y}\right)^n}=x^{q^n}+y^{\left(\frac{1}{q}\right)^n}=a_n+b_n, $$ with $$\begin{align} a_n &= a_{n-1}^q,\\ b_n &= b_{n-1}^{1/q}. \end{align}$$

It is easy to verify that with $b_n\geq0.5$, also $b_{n+1}\geq 0.5$. Now, let us assume that $a_{n-1} \leq 1-b_{n-1}$. For $n=1$, we have $a_0 \leq 1-b_0$ (more specifically, $a_0=1-b_0$). Because of that, we can write (remember that $q \geq 1$): \begin{align} a_{n+1} + b_{n+1} &= a_n^q+b_n^{1/q} \\ &\leq (1-b_n)^q + b_n^{1/q}. \end{align}

What is left, is to prove that $$ (1-b_n)^q+b_n^{1/q} \leq 1, \quad \forall \,\,\, b_n \in [0.5, 1], q\geq1 $$

This is where I got stuck, but perhaps you know how to continue from here. In the figure below, I plotted $(1-b)^q+b^{1/q}$ with $b$ on the interval $[0.5,1]$ for various values of $q$ and it seems that the result is always equal or below $1$.

enter image description here


Partial answer

Let us prove the case when $0 < x \le \frac{14}{33}$.

By Bernoulli inequality, we have $$y^{(x/y)^n} = (1 - x)^{(x/y)^n} \le 1 - x(x/y)^n.$$ It suffices to prove that $$x^{(y/x)^n} + 1 - x(x/y)^n \le 1$$ or $$\ln x \le \frac{- \ln[(y/x)^n]}{(y/x)^n - 1}.$$ Since $u \mapsto \frac{-\ln u}{u - 1}$ is strictly increasing on $u > 1$, it suffices to prove that $$\ln x \le \frac{- \ln[(y/x)]}{(y/x) - 1}.$$ With the substitution $x = \frac{1}{z}$, it suffices to prove that, for all $z \ge \frac{33}{14}$, $$z\ln(z) - \ln(z - 1) - 2\ln z \ge 0$$ which is true (easy).


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.