Denote $\log x = \log_ex$. Let's consider the below function $$\frac{\log x}{x}$$. Apparently, It's maximum is $\frac{1}{e}$. and strictly increasing in $(0,e]$, strictly decreasing in $[e,+\infty)$. If we draw a line $y=a$, where $0<a<\frac{1}{e}$. It will have two intersection point $x_1,x_2$. the question is how to prove $$x_1x_2 > e^2$$

  • 3
    $\begingroup$ Here's what I've tried: When $\dfrac{\log x}{x}=a$, $x=e^{ax}.$ Hence, we have: $x_1=e^{ax_1},x_2=e^{ax_2}$. Thus, $x_1x_2=e^{a(x_1+x_2)}$. What have you tried? $\endgroup$ – user122283 May 24 '14 at 1:12
  • $\begingroup$ @SanathDevalapurkar I am not tried this method. My way is using Cauchy Inequality: Because $x_1x_2 > e^2$ equivalent to $\log (x_1x_2) > 2$. then I have $\log(x_1x_2) = \int_1^{x_1x_2} \frac{1}{t}\text{d}t \geq \left(\int_1^{x_1x_2}\frac{1}{\sqrt{t}}\text{d}t\right)^2 = (2\sqrt{x_1x_2}-2)^2 = 4(\sqrt{x_1x_2}-1)^2>2$. in order to satisfy the last inequality. we must have $x_1x_2 > \left(1+\frac{\sqrt{2}}{2}\right)^2$. the right hand side of the last inequality is less than $e^2$. I hope this way would solve this question.But up to now I can't go any further. $\endgroup$ – Laura May 24 '14 at 2:01
  • 1
    $\begingroup$ $\dfrac{\ln x}x=\dfrac{\ln y}y\iff\sqrt[x]x=\sqrt[y]y\iff x^y=y^x$. $\endgroup$ – Lucian May 24 '14 at 2:13
  • $\begingroup$ I proved this a number of years ago in sci.math, but I can't find it :( My proof essentially said that $\ln x/x$ gets steeper on the left side of $e$ than on the right. My proof was based on (iirc) the power series for the two curves compared term by term. Someone soon after came up with a better proof involving the derivatives. $\endgroup$ – marty cohen May 24 '14 at 5:10
  • $\begingroup$ Aha! I got a proof, below. This isn't this one I came up with, but seems to me to be similar to the other one I referred to. $\endgroup$ – marty cohen May 24 '14 at 5:48

I would change the independent variable to $t=\log x$. Then we consider the function $f(t)=te^{-t}$, which has maximum at $t=1$. (Indeed, $f'(t)=(1-t)e^{-t}$.) We take intersection points $t_1,t_2$ with horizontal line $y=a>0$, and want to prove that $t_1+t_2>2$.

Let's say $t_1<1<t_2$. Geometrically, $t_1+t_2>2$ means that of these two points, $t_2$ is further away from $1$. Which is intuitively clear, since $|f'|$, the rate of change of $f$, is smaller to the right of $1$. So it takes longer for $f$ to decrease from $1/e$ to $a$ when going to the right.

A formal computation to justify the above can look like this: $$ e^{-1}-a = \int_{t_1}^1 (1-t) e^{-t}\,dt > e^{-1} \int_{t_1}^1 (1-t) \,dt = \frac{(1-t_1)^2}{2e} $$ $$ e^{-1}-a = \int_{1}^{t_2} (t-1) e^{-t}\,dt < e^{-1} \int_{1}^{t_2} (t-1) \,dt = \frac{(t_2 - 1)^2}{2e} $$


Turn this around.

Suppose $1 < x < e$ and $y = e^2/x$. We want to show that $\ln y/y > \ln x/x$.

If this holds, since $\ln x/x$ is decreasing for $x > e$, then the value of $z> e$ such that $\ln z/z = \ln x/x$ must satisfy $z > y$ so that $z > e^2/x$ or $xz > e^2$.

$\ln y/y =\ln(e^2/x)/(e^2/x) =x(2-\ln x)/e^2 $ so we want $x(2-\ln x)/e^2 > \ln x/x$ or $x^2(2-\ln x) > e^2 \ln x $.

Let $f(x) =x^2(2-\ln x) - e^2 \ln x $ where $1 < x < e$. We want to show that $f(x) > 0$.

I'm going to look at $f$ and its successive derivatives and, I hope, show what I want. Here goes.

$f(1) = 2$ and $f(e) =e^2-e^2 =0 $, so the extreme values are OK.

$\begin{array}\\ f'(x) &=2x(2-\ln x)-x-e^2/x\\ &=4x-2x\ln x - x -e^2/x\\ &=3x-2x\ln x -e^2/x\\ \end{array} $

If we can show $f'(x) < 0$ we are done.

$f'(1) = 3-e^2 < 0$ and $f'(e) =3e-2e-e = 0 $ so, again, the extreme values are ok.

Let $g(x) = f'(x)= 3x-2x\ln x -e^2/x$. If we can show that $g(x)$ is increasing, then we are done.

$\begin{array}\\ g'(x) &=3-(2+2\ln x)+e^2/x^2\\ &=1-2\ln x+e^2/x^2\\ \end{array} $

$g'(1) = 1+e^2 > 0$ and $g'(e) = 0$ so the extreme values are OK.

$g''(x) = -2/x-2e^2/x^3 < 0$.

Finally, we have something definite!

Since $g''(x) < 0$ and $g'(e) = 0$, $g'(x) > 0$ for $1 < x < e$.

Since $g'(x) > 0$ and $g(e) = 0$, $f'(x) =g(x) < 0$ for $1 < x < e$.

Since $f'(x) < 0$ and $f(e) = 0$, $f(x) > 0$ for $1 < x < e$ (whew!).



Consider $x_1$ and $x_2$ as functions of $y\in(0,\frac1e)$, defined implicitly by $y=\frac{\log x_1}{x_1}$ and $y=\frac{\log x_2}{x_2}$ and $x_1<e<x_2$. Differentiating $\log x = xy$ implicitly wrt $y$ yields $\frac{x'}{x} = x + x'y$, and so $$ x' = \frac{x}{\frac1x-y} $$ Thus \begin{align*} (x_1x_2)' &= x_1' x_2 + x_1 x_2' \\ &= \frac{x_1x_2}{\frac1{x_1}-y} + \frac{x_1x_2}{\frac1{x_2}-y} \\ &= \frac{x_1x_2\big(\frac1{x_1}+\frac1{x_2} - 2y\big)}{\big(\frac1{x_1}-y\big)\big(\frac1{x_2}-y\big)} \end{align*} Since $x_1<e<x_2$, we have $$\frac1{x_1} - y = \frac1{x_1}(1-\log x_1) > 0 $$ and $$\frac1{x_2} - y = \frac1{x_2}(1-\log x_2) < 0 $$ and, by the inequalities of the harmonic, geometric, and logarithmic means, \begin{align*} \frac1{x_1}+\frac1{x_2} - 2y &> \frac2{\sqrt{x_1x_2}} - 2y \\ &> \frac{2(\log x_1 - \log x_2)}{x_1-x_2} - 2y \\ &= \frac{2(x_1y - x_2y)}{x_1-x_2} - 2y \\ &= 0 \end{align*} (The inequalities are strict because $x_1\ne x_2$.) Putting that all together yields $(x_1x_2)' < 0$, so $x_1x_2$ is a strictly decreasing function of $y\in (0,\frac1e)$; thus the value at $y=\frac1e$ is the (unique global) minimum, that is, $x_1x_2 > e^2$.

  • 2
    $\begingroup$ I give this answer because I'm tickled to have two chances in as many days to make use of the inequality of the geometric and logarithmic means. $\endgroup$ – user21467 May 24 '14 at 14:02

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.