I find it very strange that $$ e^x \geq x^e \, \quad \forall x \in \mathbb{R}^+.$$

I have scratched my head for a long time, but could not find any logical reason. Can anybody explain what is the reason behind the above inequality? I know this is math community, but I'd appreciate a more intuitive explanation than a technical one.

  • 1
    $\begingroup$ It is not always true as others have pointed out but, as $x$ gets larger and larger then $e^x$ increases very rapidly since its exponent is inceasing and $e$ itself is greater than one. $x^e$ however is always raising some number to a fixed power ($e$ in this case). So raising a large number to a fixed power generally gives you a smaller result than raising a small number ($\gt1$) to a large power due to the way exponential numbers grow. $\endgroup$
    – Mufasa
    Jan 20 '16 at 18:35
  • $\begingroup$ @Mufasa, that argument explains why $e^x\geq x^e$ when $x$ is big. But it doesn't help us when $x$ is between $0$ and $e$. $\endgroup$
    – Asinomas
    Jan 20 '16 at 18:36
  • 8
    $\begingroup$ Why on earth is this question being downvoted? $\endgroup$
    – Brian Tung
    Jan 20 '16 at 18:40
  • $\begingroup$ I upvoted because I knew it was true and always forgot to prove it, so now I finally proved it. $\endgroup$
    – Asinomas
    Jan 20 '16 at 18:46
  • 6
    $\begingroup$ @MorganRodgers: I'm upvoting, partly in compensation, and partly because I don't agree with your characterization. There's plenty more context here than in many questions that don't get downvoted: It's not a homework question, and they're looking for intuition as to why this property holds. I think the wording of looking for a non-mathematical explanation is unfortunate, and may be misleading some people; I think what they mean is, they don't want an explanation that is rigorous but lacks intuitive explanatory power. $\endgroup$
    – Brian Tung
    Jan 20 '16 at 18:49

Not a rigorous proof at all, but you could take the $xe$th root of both sides, and then we have

$$ e^{1/e} \geq x^{1/x} $$

That $x^{1/x}$ attains a maximum at $x = e$ can be shown fairly straightforwardly.

ETA2: A picture is worth—well, a lot of words, if not quite a thousand:

enter image description here

ETA: OK, a discussion of why $x^{1/x}$ attains a maximum at $x = e$. To obtain the actual value requires calculus, but we can get some intuition for the fact that it attains a maximum somewhere near $e$, as follows.

Let $f(x) = x^{1/x} = \sqrt[x]{x}$. For $x = 1$, we clearly have $f(x) = 1$. For $x = 2$, we have $f(x) = \sqrt{2} > 1$. So far, it's going up.

Now, for $x = 4$, we have $f(x) = \sqrt[4]{4} = \sqrt{2} = f(2)$. And for $x$ really large—let's say, $x = 256$, we have $f(x) = \sqrt[256]{256} = \sqrt[128]{16} = \sqrt[64]{4} = \sqrt[32]{2}$. We may not know exactly what the $32$nd root of $2$ is, but it's clearly quite a bit less than the square root of $2$. (It's about $1.02190$.) However, no matter how large $x$ is, $f(x)$ is always the $x$th root of a number greater than $1$, so $f(x)$ must be greater than $1$.

The general shape we have, then, is a function that (over the range from $1$ to $\infty$) starts at $1$, rises to a peak at some point between $2$ and $4$, and then falls off toward $1$ again (though never arriving at it). We can even find a value of $x$ in the interval $(2, 4)$ for which $f(x) > \sqrt{2}$. That is,

$$ f(\sqrt{8}) = \sqrt{8}^{1/\sqrt{8}} $$

How does this compare to $f(2) = f(4) = \sqrt{2}$? At first glance, it is difficult to say. However, if we cube $f(\sqrt{8})$, we get

$$ [f(\sqrt{8})]^3 = \sqrt{8}^{3/\sqrt{8}} $$

Note that that exponent is greater than $1$ (since $3 = \sqrt{9} > \sqrt{8}$), so

$$ [f(\sqrt{8})]^3 = \sqrt{8}^{3/\sqrt{8}} > \sqrt{8} = [f(2)]^3 = [f(4)]^3 $$

so (again, not rigorously)

$$ f(\sqrt{8}) > \sqrt{2} = f(2) = f(4) $$

The actual value is about $1.44426$.

As I said previously, proving that $x^{1/x}$ attains a maximum specifically at $x = e$ requires calculus, but it seems that may not be exactly what you want? Hoping the above serves as some kind of intuitive basis.

\begin{align} \frac{d}{dx} x^{1/x} & = \frac{d}{dx} e^{\frac{\ln x}{x}} \\ & = \left( \frac{d}{dx} \frac{\ln x}{x} \right) e^{\frac{\ln x}{x}} \\ & = \frac{1-\ln x}{x^2} e^{\frac{\ln x}{x}} \\ & = \frac{1-\ln x}{x^2} x^{1/x} \end{align}

and $f(x)$ attains a maximum when that derivative is equal to $0$, which happens when $1 = \ln x$, which happens when $x = e$. (There might be a bit of circularity there—I have to give that some thought—but that's how you obtain the maximum.)

  • $\begingroup$ @BrainTung Yes, Please explain why this is so ? $\endgroup$
    – Neeraj
    Jan 20 '16 at 18:56
  • $\begingroup$ The derivative of $ g(x)=\log (x^{1/x})=(\log x)/x$ is $(1-\log x)/x^2$, which is positive for $x<e$ and negative for $x>e$. So $g(x)$ ha a unique maximum at $x=e$ . So $x^{1/x}=e^{g(x)}$ has a unique maximum at $x=e$. $\endgroup$ Jan 20 '16 at 19:16
  • $\begingroup$ @user254665: Yup, just added it as you were typing, I think. $\endgroup$
    – Brian Tung
    Jan 20 '16 at 19:18
  • $\begingroup$ @BrainTung Thanks for explaining in so simple words. I wanted something more of explanation rather than just mathematical equation and this is exactly what you have been provided. $\endgroup$
    – Neeraj
    Jan 20 '16 at 19:34
  • $\begingroup$ A very good answer all around - and it's probably worth pointing out that answers can't really get much more intuitive than this simply because there isn't really an 'intuitive' definition of $x^e$ - any such definition is going to have to pass through a chunk of calculus (limits, in particular) and so the other tools of calculus 'should' be available for an answer. $\endgroup$ Jan 20 '16 at 20:29

Let $f:\mathbb{R}^+\to\mathbb{R}$ given by $f(x)=e^x-x^e$.

Then $f^\prime(x)=e^x-ex^{e-1}=e(e^{x-1}-x^{e-1})$ and $f^{\prime\prime}(x)=e(e^{x-1}-(e-1)x^{e-2})=e(e^{x-1}-ex^{e-2}+x^{e-2})$. Thus $f^{\prime}(e)=0$ and $f^{\prime\prime}(e)=e^{e-1}>0$. So $f$ has a local minimim in $e$.

Since $\lim_{x\to\infty}f(x)=\infty$, thus $f$ attains his absolut minimum on $x=e$, which is $0$.

Then $f(x)\ge 0$ for all $x\in\mathbb{R}^+$.


Here is a different approach, just for the sake of variety, which could be made more rigorous:

Consider the tangent to the curve $y=\ln x$ at the point $x=e,y=1$ is $$y-1=\frac 1e(x-e)\implies y=\frac xe$$

We know that the curve is concave so lies below the tangent except at the tangent point.

Therefore, $$\ln x\leq\frac xe$$ $$\implies x^e\leq e^x$$


The reasons is that the exponential function($something^x$) grows much faster than the power function ($x^{something}$)

So in general you would expect that for every $a, b > 1$something like this hold

$$a^x \ge x^b$$ at least for $x$ big enough

Now I don't think it's particularly meaningful that for the special case $a = b = e$ the inequality holds for every $x \ge 0$. It can be shown easily with basic calculus in any case, but maybe that won't satisfy your intuition :)


I'd appreciate a more intuitive explanation than a technical one.

Well on that, try to consider the behaviour of $2^x$. Every time I increase x by one I get another (binary) digit out of my expression. So I quickly get a huge number. If I double $x$ I get way more digits.

Now compare that with $x^2$. Every time I increment $x$ I will not generally get an extra digit from this expression. In fact as $x$ gets really large there may be no difference in the number of digits between two "adjacent" values of the expression.

For example the number of binary digits required to represent $1000^2$ is the same as the number to represent $1001^2$. In fact it's 10 binary digits for both. But $2^{1000}$ requires one thousand binary digits. So a huge difference in growth.

So we see an exponentiation function grows much faster than the square function.

If binary digits aren't your "thing" then use base $10$ and compare $10^x$ and $x^{10}$. It's the same argument.

Now if you consider your specific request about $e^x$ and $x^e$ you should see, intuitively, that this is a similar situation in terms of relative growth. We no longer have the convenience of binary digits, but it's pretty clear the same rationale holds.


$$e^x \ge x^e ,\quad \forall x \in \mathbb{R}^+$$ $$\ln e^x \ge \ln x^e$$ $$ x\ln e \ge e \cdot \ln x$$ $$ x \ge e \cdot \ln x$$

Let $f(x) = x-e\cdot \ln x.$ Using derivatives to find an absolute minimum, you'll find one at $x=e$, where $f(e)=0$. So at $[0, e)\cup (e, \infty)$ we have $f(x)>0 \implies x \gt e \cdot \ln x,$ and at $x=e$ we have $f(x)=0 \implies x = e \cdot \ln x.$

$$\therefore e^x \ge x^e ,\quad \forall x \in \mathbb{R}^+.$$


Throughout, $\log$ denotes the natural logarithm.

If $x$ and $y$ are positive real numbers, then \begin{align*} x^{y} \leq y^{x} &\quad\text{if and only if}\quad y\log x \leq x \log y \\ &\quad\text{if and only if}\quad \frac{\log x}{x} \leq \frac{\log y}{y}. \end{align*} The final inequality holds for all positive real $x$ if and only if $y$ is an absolute maximum of the function $$ \phi(u) = \frac{\log u}{u},\quad u > 0. $$ It turns out the function $\phi$ has a unique maximum, at $u = e$. This tells you two things:

  • $x^{e} \leq e^{x}$ for all $x > 0$, and

  • $b = e$ is the only positive real number with the property that $x^{b} \leq b^{x}$ for all $x > 0$.


The number $e$ is somewhat accidental, but there is a simple graphical explanation that applies to many other two-variable functions more general than $x^y$.

The function $f(x,y)=x^y$ has the property that the graph of $f(x,y)=f(y,x)$ is a transverse (in fact, perpendicular) intersection of the line $y=x$ with a curve that is the graph of a continuous function on some interval of the real line. The intersection point $x=y=E$ will have either $f(E,x) \geq f(x,E)$ for all $x$, or the opposite inequality for all $x$.

Why? The line and the curve divide the region above and below the interval into 4 parts, and in the interior of each part the difference $D = f(x,y) - f(y,x)$ has constant sign. Opposite parts have the same sign, therefore a horizontal or vertical line through $E$ have $D$ of one sign.

For the function in this question $x^y$ it is not hard to derive that $E=e$ is the intersection point. It can be done without calculus using the parameterization of $x^y=y^x$ as $x = (1+\frac{1}{w})^w$ , $y = (1 + \frac{1}{w})^{w+1}$, which at $w = \infty$ gives $x=y=e$.

Which parts of this generalize to other functions? Properties that are important for this $f$ are that the vertical cross-sections $D(a,x)$ are convex and the horizontal sections are negatives of the verticals. The graph of $D = f(x,y)-f(y,x)$ is a saddle surface with positive and negative curvatures are each point; the Hessian is of signature $(1,1)$ everywhere. I do not have a general characterization but it is possible that once a point on the diagonal is found with $(E,E)$ a local minimum of $D(x,y)$ in the vertical direction, and the saddle condition holds, then $f(E,x) - f(x,E)$ does not change sign.


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.