so I was doing my calculus homework and ran into this tricky question.

Determine the total number of critical points of the function $f(x)=(x+e^x)^k$, where $k>0$ is an integer

So I got the derivative is $f'(x)=k(x+e^x)^{k-1} \times (1+e^x)$, but I couldn't find the point, however when I look at the graph there is a point around $0.567$ or W($1$), so I am a bit lost.


  • $\begingroup$ you mean $f(x)=(x+e^x)^k$? $\endgroup$ – mengdie1982 Jul 11 '18 at 8:19
  • $\begingroup$ Yes, that is what I meant. $\endgroup$ – Michael M Jul 11 '18 at 8:20
  • $\begingroup$ Using iterations i got $-0.567$. $\endgroup$ – Chris2018 Jul 11 '18 at 8:26
  • 1
    $\begingroup$ It should be $\color{red}{-}W(1)$ $\endgroup$ – Claude Leibovici Jul 11 '18 at 9:06

$$f'(x)=k(x+e^x)^{k-1} \times (1+e^x)=0 $$

has only one solution which is where $x+e^x=0$ and that is the point that you want to approximate.

The answer should be negative so $x=0.567$ is problematic.

| cite | improve this answer | |


Notice $$f'(x)=k(x+e^x)^{k-1}(1+e^x).$$

Let $f'(x)=0$. Then we obtain $$x+e^x=0$$

This equation has no closed-form solution. But by graphing $y=-x$ and $y=e^x$, you may intuitively find there exists only one root over $(-1,0)$.

enter image description here

Moreover, we may prove this fact. Since $f''(x)=1+e^x>0.$ Then $f'(x)$ is rigorously increasing over $(-\infty,+\infty)$. Thus, there exists at most one solution for $f'(x)=0$. Notice that $f'(-1)=-1+e^{-1}<0$ and $f'(0)=0+e^0>0$. Thus, by intermediate value theorem, we may claim the fact.

| cite | improve this answer | |
  • $\begingroup$ Thank you for the explanation! $\endgroup$ – Michael M Jul 12 '18 at 0:17

At critical points $f'(x)=0$ so

$0=k(x+e^x)^{k-1} \times (1+e^x)$
k and $(1+e^x)$ are positive so $0=x+e^x$
$x_n=-e^{x_{n+1}}$ so $x \approx -0.56714329$

| cite | improve this answer | |

Given \begin{align} f'(x)&=k(x+\exp(x))^{k-1} \times (1+\exp(x)) \end{align}

the function $f$ have no critical points for $k=1$ and one critical point for $k>1$ at

\begin{align} x&=-\operatorname{W}(1) \approx -0.567143290409784 . \end{align}


We need to solve $f'(x)=0$, so since it is given that $k>0$ and we know that $\exp(x)>0$ for all $x\in \mathbb{R}$, we can ignore the factor $(1+\exp(x))$, and all we left to consider is when

\begin{align} (x+\exp(x))^{k-1}&=0 . \end{align}

for $k=1$ we have

\begin{align} f(x)&=x+\exp(x) \\ f'(x)&=1+\exp(x)>0 \quad \forall x\in\mathbb{R} . \end{align}

For $k>1$:

\begin{align} x+\exp(x)&=0 ,\\ -x&=\exp(x) ,\\ -x\exp(-x)&=1 \tag{1}\label{1} . \end{align}

At this point we have an equation of the form $u\exp(u)=v$, and a long time ago it was suggested to write an exact solution of this equation as $u=\operatorname{W}(v)$, where $\operatorname{W}$ is the Lambert W function, which for equations of the form $u\exp(u)=v$ plays exactly the same role as the function $\ln$ plays for equations of the form $\exp(u)=v$: to express the exact solution of $\exp(u)=v$, we write $u=\ln(v)$.

So, to continue with \eqref{1}, we have \begin{align} \operatorname{W}(-x\exp(-x))&=\operatorname{W}(1) ,\\ -x&=\operatorname{W}(1) ,\\ x&=-\operatorname{W}(1) \tag{2}\label{2} . \end{align}

Now, we need to recall that when dealing with $\operatorname{W}$ (very much like with quadratic equation, for example) we can face with from 0 to at most two real solutions, which have special names $\operatorname{W_0}$ and $\operatorname{W_{-1}}$.

The number of solutions and their range is solely defined by the argument $v$ in $\operatorname{W}(v)$:

\begin{align} v&<-\tfrac1{\mathrm{e}}\quad\text{no real solutions} ;\\ -\tfrac1{\mathrm{e}}< v&<0\quad\text{two real solutions:} \quad\operatorname{W_0}(v), \operatorname{W_{-1}}(v), \text{ for which holds}\quad \operatorname{W_{-1}}(v)<-1 <\operatorname{W_0}(v)<0 ;\\ v&\ge0 \quad\text{just one real solution} \quad\operatorname{W_0}(v)\ge0 ;\\ v&= -\tfrac1{\mathrm{e}}\quad\text{the case when} \quad\operatorname{W_0}(v)= \operatorname{W_{-1}}(v)=-1 . \end{align}

Thus, in our case \eqref{2}, the argument of $\operatorname{W}$ is $1$, that means, we have just one real solution,

\begin{align} x&=-\operatorname{W_0}(1) . \end{align}

| cite | improve this answer | |
  • $\begingroup$ Hi I am still a bit confused about how did you get -W(1) from the f '(x). $\endgroup$ – Michael M Jul 12 '18 at 0:18
  • $\begingroup$ @Michael M: See the expanded steps explained. $\endgroup$ – g.kov Jul 12 '18 at 5:26

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.