I understand that this theorem works for complete metric spaces, but I have been studying this purely for normed vector spaces. I want to check my heuristic understanding of the theorem and proof:

Statement of theorem:

Let $(V, ||.||)$ be a complete normed vector space and $U \subset V$ a closed subset of the space. Let $f: U \rightarrow U$ be a contraction mapping such that $\exists K \in (0,1)$ such that $\forall u, v \in V, ||f(u) - f(v)|| \leq K||u-v||$ .

Then there is a unique $w \in U$ such that $f(w) = w$

The proof constructs a sequence by taking any $u_0 \in U$ and doing $f(u_n) = u_{n+1}$.

My understanding:

Since $K <1$ and we are mapping each time between $U$, we are mapping a 'smaller' and 'smaller' such that eventually $f(u_n)$ will just map to itself thus reaching a fixed point.

And this limit exists because this sequence $u_n$ is Cauchy and in a Banach space all Cauchy sequences will converge to some limit.

My questions

Why are we free to take any $u_0 \in U$? Does our choice of $u_0$ affect the convergence rate?

Why is it necessarily the case that the limit is unique? And how is this affected by the map $f?$

Does the theorem still apply for $K =1?$

What applications are there to this theorem?

  • $\begingroup$ To answer your first question, yes the choice of $u_0$ affects the convergence rate. If $u_0$ is the fixed point then the $f(u_0) -u_0=0$. $\endgroup$ – fred Apr 13 '16 at 18:17

The fixed point is unique because if $f(u)=u$ and $f(v)=v$ then $$ ||u-v||=||f(u)-f(v)||\leq K||u-v|| $$ and since $K<1$ this is only possible if $u=v$.

If $K=1$ then either uniqueness or existence can fail. For instance, the map $f(u)=u$ satisfies the hypotheses with $K=1$, but every point is a fixed point. And if $V=\mathbb{R}$ and $f(x)=x+\frac{1}{1+e^x}$, then $0\leq f^{\prime}(x)<1$ for all $x$ (so the contraction property holds with $K=1$ by the mean value theorem), but $f$ has no fixed point.

The most important application I'm aware of is the Picard-Lindelof theorem, which establishes the existence and uniqueness of solutions to a class of ordinary differential equations.

| cite | improve this answer | |
  • $\begingroup$ Thanks, is there any case where $K =1$ but both uniqueness and existence still hold? Or is there a proof of why this can not happen? $\endgroup$ – Kevin Apr 13 '16 at 18:29
  • $\begingroup$ If $X$ is a compact metric space and $f:X\to X$ satisfies $d(f(x),f(y))<d(x,y)$ for all $x\neq y\in X$, then it's still possible to show that $f$ has a unique fixed point. My memory is that the proof is very different from the case $K<1$. $\endgroup$ – carmichael561 Apr 13 '16 at 18:32
  • $\begingroup$ Thank you. Would you say my heuristic understanding is correct, or do you have a more intuitive way to think about the theorem? $\endgroup$ – Kevin Apr 13 '16 at 18:34
  • $\begingroup$ Your understanding seems correct to me. $\endgroup$ – carmichael561 Apr 13 '16 at 18:35

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.