I'm doing a practice problem which asks to prove that the sequence defined by $x_{n+1} = e^{-x_n}$ is convergent (or rather "study the convergence of $(x_n)$"). So I'd like to try and find some sufficient condition on $x_0$ for the sequence to converge.

I can see that $e^{-x}$ is $k$-lipschitzian with $k<1$ on $[a, \infty[$ for $a>0$. But the problem is that $e^{-x}$ does not map $[a, \infty[$ into itself. I started trying to find values of $a, b$ such that $[e^{-b}, e^{-a}]\subseteq[a, b]$, but then I wondered if maybe there was some simpler approach that I was missing.

  • 2
    $\begingroup$ Do you mean $x_{n+1}=e^{-x_n}$? $\endgroup$ – Wojowu Dec 27 '15 at 15:24
  • $\begingroup$ @Wojowu Yes. ${}$ $\endgroup$ – Jack M Dec 27 '15 at 15:25
  • $\begingroup$ $x\mapsto e^{-x}$ maps $[0,1]$ into itself. $\endgroup$ – PhoemueX Dec 27 '15 at 15:30
  • $\begingroup$ @PhoemueX But $e^{-x}$ is only $1$-lipschitzian on $[0, 1]$. $\endgroup$ – Jack M Dec 27 '15 at 15:31

The solution is a corollary of another Banach fixed point theorem: see here Essentially, if we have contraction with constant $k=1$ and compactness (instead of completeness) the result will follow.

For any $x_0$ we have $x_1 \geq 0$ and so $x_2 \leq e^0 = 1$. A trivial induction shows that $x_n \in [0,1]$ for $n\geq 2$ As @Jack M has pointed out in the comments, $e^{-x}$ is indeed 1-lipschitz on $[0,1]$ so the conditions of the theorem in the link I provided are met: $K = [0,1]$. Thus, the result follows.

| cite | improve this answer | |
  • $\begingroup$ So to show $x_n\to a$ (where $a$ is the unique fixed point), we use that $x_n$ has a convergent subsequence, show that the limit of this subsequence must be $a$ with the usual argument, then use the decreasing nature of $|x_n - a|$ to prove that the existence of a convergent subsequence implies the convergence of $(x_n)$ itself? $\endgroup$ – Jack M Dec 27 '15 at 16:48
  • $\begingroup$ Yes, and you can use that $a$ is the unique solution of $x = e^{-x}$ for the last part of the argument. $\endgroup$ – Milen Ivanov Dec 27 '15 at 16:58

I like Milen's answer and upvoted it. Here's an approach that uses the standard contraction mapping theorem.

Let $f(x)=\exp(-x)$ and let $F(x)=f(f(x))$. We compute \begin{align} F'(x) &= e^{-x-e^{-x}} \\ F''(x) &= e^{-x-e^{-x}} \left(e^{-x}-1\right). \end{align} Note that $F''(x)<0$ for $x>0$. Thus $F'$ is decreasing on $[0,\infty)$ and $$F'(x)<F'(0)=\frac{1}{e}<1$$ for $x\in[0,\infty)$. It follows that $F$ is a contraction on $[0,\infty)$ and, thus, has a unique fixed point in $[0,\infty)$, say $x_0$. I claim that your recursively defined sequence converges to $x_0$.

Let $a$ be a real number with $0<a<x_0$ and let $b=f(a)$. Our observations on $F$ imply that $a<f(b)<x_0<f(a)<b$. In particular, $f$ maps $[a,b]$ into itself. Furthermore, it's easy to see that $|f'(x)|<|f'(a)|<1$ on $[a,b]$ so that $f$ is a contraction on $[a,b]$. Thus, iteration of $f$ must converge to $x_0$ when we start in $[a,b]$. Finally, every starting point in $\mathbb R$ must eventually land in some interval of the form $[a,f(a)]$.

| cite | improve this answer | |
  • 1
    $\begingroup$ Very nice trick, to consider $f(f(x))$ to get a Lipshitz constant $1/e$ $\endgroup$ – Milen Ivanov Dec 27 '15 at 17:00

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.