7
$\begingroup$

In a high school reference book, I read a question asking which of $\frac{1+e^x}2$ and $\frac{e^x-1}x$ is larger when $x>0$.

Of course, the question is technically easy to answer. E.g. using the power series expansion of $e^x$, one immediately sees that the average height is greater than the slope of the secant line. Alternatively, one can answer the question by using the first and second derivatives of $x(1+e^x)-2(e^x-1)$.

Yet, answers like these seem too hacky. A more elegant one is to compare the derivatives of $\frac x2$ and $\tanh\frac x2=\frac{e^x-1}{e^x+1}$, but I don't expect a high school student to be aware of the use of hyperbolic tangent.

Since both $\frac{1+e^x}2$ and $\frac{e^x-1}x$ have geometric interpretations, I wonder if there is a more natural --- and probably geometrically minded --- answer. Any idea?

$\endgroup$
  • 2
    $\begingroup$ so $0<x<2$ right ?because $x>2$ seems to be obvious. $\endgroup$ – Seyhmus Güngören Mar 8 '18 at 2:19
  • $\begingroup$ @SeyhmusGüngören Actually I didn't notice that. Thanks for the simplification! $\endgroup$ – user1551 Mar 8 '18 at 2:21
  • $\begingroup$ The inequality is true for $x\ne0$, not only $x\gt0$. $\endgroup$ – robjohn Mar 9 '18 at 9:57
14
$\begingroup$

You might be interested in the strict formulation of the Hermite–Hadamard inequality—if $f$ is strictly convex on $[a,b]$ where $a<b$, then the integral of $f$ is strictly bounded above by its trapezoidal approximation: $$\int_a^bf(x)\mathrm{d}x< \tfrac{f(a)+f(b)}{2}(b-a)\text{.}$$ Set $a=0$, $b=x$, $f(x)=\mathrm{e}^x$. Then $$\mathrm{e}^x-1< x\left(\tfrac{\mathrm{e}^x+1}{2}\right)\text{.}$$ Since $f$ in this case is positive and smooth, it's pretty easy to illustrate the inequality graphically.

$\endgroup$
  • $\begingroup$ That's very nice indeed. +1 $\endgroup$ – user1551 Mar 8 '18 at 2:30
  • $\begingroup$ user1551 : I just realized that you asked for a strict inequality. Edited accordingly. $\endgroup$ – K B Dave Mar 8 '18 at 2:47
2
$\begingroup$

$$\frac{1+e^x}2>\frac{e^x-1}x$$ is equivalent to $$x+2>(2-x)e^x $$

These two functions, $x+2$ and $ (2-x)e^x $ have the same value and the same slope at $x=0$

for $x>0$, comparing the two derivatives, $1$ and $(1-x)e^x$ leads to comparing $ \frac {1}{1-x}$ and $e^x$

Note that on the interval $(0,1)$, $\frac {1}{1-x}= 1+x+x^2+...$ wins over $e^x = 1+x+x^2/2+x^3/6+..$

For $x\ge 1$ the $1>(1-x)e^x$ is straight forward.

$\endgroup$
1
$\begingroup$

Not a proof but a geometric interpretation: let $$\begin{align}O&=(0,0)&A&=(1,0)&P&=(x,y)\end{align}$$ where $x^2-y^2=1$ is the unit hyperbola and $x>0$. Let $B$ be the intersection of the line $x=1$ with $OP$. Then $$B=(1,\tfrac{y}{x})\text{.}$$ Now, triangle $\triangle OAB$ is contained completely within the hyperbolic sector $OAP$, so we have $$\text{Area}(\triangle OAB)<\text{Area}(OAP)\text{.}$$ Suppose the area of sector of $OAP$ is given by $\tfrac{\tau}{2}$. Then $$(x,y)=(\cosh\tau,\sinh\tau)\text{,}$$ and the area inequality above is $$\tfrac{1}{2}\tanh \tau < \tfrac{\tau}{2}\text{.}$$

$\endgroup$
  • $\begingroup$ Nice geometric interpretation. +1. But the fact that the area of the hyperbolic sector is $\tau/2$ may be inobvious. That said, I think it's a good alternative answer to your other one. $\endgroup$ – user1551 Mar 8 '18 at 7:58
  • $\begingroup$ I agree, it's not a good fit for modern definitions. I included it to illustrate that a student who understands $\tfrac{1}{2}\sin \theta\cos\theta<\tfrac{\theta}{2}<\tfrac{1}{2}\tan \theta$ can understand $\tfrac{1}{2}\tanh \tau < \tfrac{\tau}{2}<\tfrac{1}{2}\sinh\tau\cosh\tau$ even if they do not know any calculus. $\endgroup$ – K B Dave Mar 8 '18 at 8:59
1
$\begingroup$

Since we are interested in $0<x<2$, make the change $0<x=\ln t<2 \Rightarrow 1<t<e^2$. The inequality will take the form: $$\frac12 \ln t>\frac{t-1}{t+1}$$ At $t=1$ they are equal. The LHS function grow faster: $$\frac{1}{2t}>\frac{2}{(t+1)^2} \iff (t-1)^2>0.$$

$\endgroup$
  • $\begingroup$ This change of variables does tidy things up. +1 $\endgroup$ – user1551 Mar 8 '18 at 7:26
1
$\begingroup$

If $x\ne0$, then $$ \begin{align} \frac{\mathrm{d}}{\mathrm{d}x}\frac{e^x-1}{e^x+1} &=\frac{\mathrm{d}}{\mathrm{d}x}\left(1-\frac2{e^x+1}\right)\\[3pt] &=\frac{2e^x}{\left(e^x+1\right)^2}\\[3pt] &=\frac2{4+\left(e^{x/2}-e^{-x/2}\right)^2}\\ &\in\left(0,\frac12\right) \end{align} $$ Therefore, the Mean Value Theorem says $$ \frac{\frac{e^x-1}{e^x+1}-0}{x-0}\in\left(0,\frac12\right) $$ That is, $$ \frac{\ \frac{e^x-1}x\ }{\ \frac{e^x+1}2\ }\in(0,1) $$

$\endgroup$
0
$\begingroup$

One can manipulate the inequality and write it as follows:

$$4>f(x)=(e^x+1)(2-x)\quad for \quad x>0\quad\quad\quad (0)$$

The first derivative of the right hand side is

$$f^{'}(x)=e^x-xe^x-1<^?0\quad for \quad x>0\quad\quad\quad (1)$$

We know that for $x=0$ we have $f^{'}(0)=0$. What remains to be shown is whether $f^{''}$ is negative. If true, this means $f^{'}$ is decreasing, and a function which is decreasing for $x>0$ and $f^{'}(0)=0$ will satisfy $(1)$.

We have $$f^{''}(x)=-xe^x<0$$

Therefore $(1)$ is correct. If $(1)$ is correct, $(0)$ must also be correct because $f$ is decreasing with $f(0)=4$ for $x>0$.

$\endgroup$
0
$\begingroup$

The inequality is equivalent to $$\tanh{x\over2}={e^x-1\over e^x+1}<{x\over2}\qquad(x>0)\ .$$ As $\tanh0=0$ and $\>\tanh' u=1-\tanh^2 u<1$ $\>(u>0)$ the claim follows.

$\endgroup$

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.