I am working on this equation, and I think I need to consider Rolle's theorem. According to a teacher, I need to use IMV as well. Any ideas?

$e^{-kx}+e^{-x} = x$

The value of $k$ is that, wherein the exponential $f(x) = e^{-kx}+e^{-x}$ is tangent to $g(x) = x$.

  • $\begingroup$ For every $k \geq 0$, the graphs of functions $f(x) = e^{-kx} + e^{-x}$ and $g(x) = x$ intersect in one and only one point. $\endgroup$ – jgsmath Jul 2 '17 at 20:02
  • $\begingroup$ Yes. :) How would one do it with $k < 0$? It is double-valued. $\endgroup$ – mov0021 Jul 3 '17 at 1:13

Take the derivatives and set equal...



Clearly $k\ge0$ is not possible, so we look at $k=-n<0$


$$u=e^x\implies nu^{n+1}=1+u$$

When $n$ is even, then there are two solutions for $u$ and when $n$ is odd, there is only one solution.

(Prove with intermediate value theorem)

Then check if $f(x)=g(x)$ at those points.

  • $\begingroup$ The OP did not state that $k$ has to be integral. $\endgroup$ – John Bentin Jul 2 '17 at 20:46

If $k>0$, left side is monotonously decreasing to 0 and right side increasing, so it clearly always has 1 and only 1 real root. You can use Rolle's theorem for this case to prove it rigorously.

For $k<0$ left side is a convex function and it will cross $y=x$ either never (huge negative $k$), twice ($k$ just below zero), or (in a single $k$ inbetween when it has a double root). If it's a double root, it matches both in value and derivative. So the critical value of $k$ satisfies

$$e^{-kx}+e^{-x}=x$$ $$-ke^{-kx}-e^{-x}=1$$

which is two equations for variables $k$ and $x$.

Plotting a few cases for $k$, I fond the critical value to be somewhere around $-0.36$ which naturally gets me thinking of $k=-e^{-1}$ but isn't quite correct. In fact, the pair of equations is transcendental and has no reason of having a closed form solution.


I don't think this can be expressed only with Lambert's function. Here's my try. Setting $n=-k$, above can be converted to

$$e^{nx}=\frac{1+x}{1+n}=u$$ $$e^{-x}=\frac{nx-1}{n+1}=v$$ where $x=u+v$. The last one can be converted to $$ve^v=e^{-u}\rightarrow v=W(e^{-u})$$ I hoped that substitutio to $u$ and $v$ variables would fully separate the variables and enable me to solve for $u$ and $v$ separately, but the $nx=(u+v)(1+v)/u$ can't be nicely simplified to something that would lead to a solvable equation. There may be better substitutions but I don't think the system has enough symmetry for this to be successful.

This process, however, does lead to a nice iterative scheme which relates to the guessed approximations. The iteration begins with initial condition:

$$u_0=e, \quad v_0=0$$ These are consistent with the guess $n\approx e^{-1}$, $x\approx 1/n \approx e$, $nx\approx 1$.

The iterative step is:

$$u_{k+1}=e^{(1+v_k)(u_k+v_k)/u_k},\quad v_{k+1}=W(e^{-u_{k+1}})$$

This converges fairly quickly to $u=2.915444361$, $v=0.05146225679$, and therefore $x=2.96690661781$ and $n=0.36065248606699$.

  • $\begingroup$ Yes, my friend. I love trascendental. Something around -0.360524. I still cannot find it exactly. $\endgroup$ – mov0021 Jul 2 '17 at 20:17
  • $\begingroup$ You won't "find" it, unless you mean numerically or in terms of an infinite recursion. It might involve Lambert's function, but even that's buried in another layer of equations. $\endgroup$ – orion Jul 2 '17 at 20:24
  • $\begingroup$ Why cannot I find it? Yes. Indeed it requires W-function, but that should be not as chaotic as numerical methods. In fact, the homework is to produce an analytical result which puts numerical methods aside. $\endgroup$ – mov0021 Jul 2 '17 at 21:33
  • $\begingroup$ W-function is numerical function, it just got a name because it's used a lot (on the inside, it's doing one of the iterative procedures, or maybe a rational approximation). But yes, I know what you want, I'll be playing with it and see if I can express this in terms of W. $\endgroup$ – orion Jul 3 '17 at 6:30

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.