MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Some of the fundamental results in analysis (inverse function theorem, existence and uniqueness of solutions to ODEs) have slick proofs using the idea of a contraction. So, it seems plausible to me that one might be motivated to study a "contraction space":

I'll define a contraction space as a metric space $(X,d)$ such that there is at least one fixed point of any function $f:X\rightarrow X$ with the property that for all $x,y$ we have $d(f(x),f(y)) \le \frac12 d(x,y)$.

Here's the question: is every contraction space complete?

I feel like the answer should be yes. The only approach I can see for proving this, though, is to find some way to take a Cauchy sequence $(a_n)$ and construct a contraction on the whole space taking each $a_i$ to some $a_j$ with $j > i$, but even for the case of Cauchy sequences in the rationals I don't see an obvious approach.

A related question: suppose we take an arbitrary metric space, and add a fixed point for every contraction the same way we add limits to Cauchy sequences. Is the resulting space then a contraction space?

share|cite|improve this question
up vote 10 down vote accepted

There is a "contraction space" which is not complete. For example, consider a metric $d$ on $[1,+\infty)$ such that for $x,y\in[n,n+1]$ where $n\in\mathbb N$ one has $d(x,y)=2^{-n}|x-y|^{1/n}$ (other distances are defined by gluing the segments together). The completion is obtained by adding one point at $+\infty$.

But there is no map contracting this space to $+\infty$. Indeed, any Lipschitz map $f$ with $f(x)>n+1$ for some $x\in[n,n+1]$ must be a constant on $[n,n+1]$, so there will be a finite fixed point.

Added. Here is a counter-example to the second question.

Let $X=(\mathbb R,d)$ from the above example. Define $Y=(\mathbb R,d')$ similarly by setting $d'(x,y)=2^{-n}|x-y|$ for $x,y\in[n,n+1]$. Then $Y$ is isometric to $[0,1)$. Consider the disjoint union of $X$ and $Y$. For $t\in [0,+\infty)$, denote by $t_X$ and $t_Y$ the copies of $t$ in $X$ and $Y$. For every $t$, attach an arc $\gamma_t$ of length $\ell_t:=10\cdot 2^{-t}$ connecting $t_X$ to $t_Y$. Let $Z$ denote the union of $X$, $Y$ and all these arcs. We have yet defined distances on $X$, on $Y$ and on every arc $\gamma_t$. The metric on $Z$ is defined as the maximal metric bounded above by these metrics on these subsets. It is easy to see that $X$ and $Y$ are embedded into $Z$ isometrically, as well as sufficiently short intervals of the arcs $\gamma_t$ (e.g. their halves are sufficiently short).

The counter-example is the space $Z\setminus Y$. Its completion is $Z\cup\{+\infty\}$. The space cannot be contracted to $+\infty$ for the same reasons as above: the arcs $\gamma_t$ added to $X$ form a tree and don't help to go around weird parts of $X$.

On the other hand, $Z\setminus Y$ can easily be contracted to any point of $Y$ because $\gamma_t$ nearby $t_Y$ is isometric to a straight line segment $(0,\epsilon_t)$. Just map every point of $Z\setminus X$ to a point in this segment in such a way that the distance to $t_Y$ gets multiplied by a small positive constant (like $\epsilon_t/100$).

Thus adding "contraction points" to $Z\setminus Y$ yields $Z$. But $Z$ can be contracted to $+\infty$ - just project everything to $Y$ and contract there.

share|cite|improve this answer
It seems that a circle without a point is also a counter-example to the first question. – Petya Apr 23 '10 at 19:38
No, fold it twice and contract the half-circle. – Sergei Ivanov Apr 23 '10 at 19:43
@Petya In fact, a (planar) Cook continuum without a point would be such a counterexample. – Ady Apr 23 '10 at 22:29

I think should be helpful.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.