10
$\begingroup$

My attempt was to think about cohomology. But I think that there is some failure on my thought.

Suppose that the claim holds. Then $\mathbb{R}^2 = \cup_i S^1_i$. We know that $\mathbb{R}^2$ minus a circle has the comology isomorphic to $\mathbb{R}$. But does not make any difference take off a circle of such union, then the cohomology of $\mathbb{R}^2$ is not the trivial. What is a contradiction.

$\endgroup$
  • $\begingroup$ Why do you think a cohomological approach will be at all fruitful? $\endgroup$ – user98602 Feb 15 '16 at 19:18
  • $\begingroup$ I don't know, I know a little about cohomology, I don't know any way to solve the problem, this was the best could do. $\endgroup$ – L.F. Cavenaghi Feb 15 '16 at 19:22
13
$\begingroup$

Suppose you had such a decomposition. Pick a circle $C_1$. By Schoenflies it bounds a disc $D_1$. Inductively, pick a circle $C_n$ contained in $D_{n-1}$ such that the area (measure, if you like) of the disc $D_n$ it bounds is at most half the area of the area of $D_{n-1}$.

This is the key step in the proof, so we should carefully see why we can do this. For notational convenience I'm going to call $D=D_{n-1}$. First note that any disc properly contained in the interior of $D$ has strictly smaller area (its complement in the interior of $D$ is open hence has positive area). Now suppose I could not find a disc of arbitrarily small area bounded by one of our circles; then let the infimum of the areas of discs bounded by our circles be $t>0$. As above we see that we cannot actually achieve $t$, or else we would be able to achieve areas smaller than it by looking at circles inside the disc of area $t$. We will contradict this by showing that we can represent $t$.

Pick a sequence $S_n$ of circles with area at most $t+1/n$. Passing to a suvsequence if necessary and invoking the finite area of the disc I can assume the $S_n$ are nested downwards. Take the intersection of the discs they bound. By Cantor's intersection theorem this is nonempty; pick a point $x$ in it; because the circle containing $x$ must have been in the disc $S_n$ bounds for all $n$, that circle is contained in the infinite intersection; and so is the disc that circle bounds. But this disc must have area at most $t+1/n$ for all $n$, hence have area at most $t$, as desired.

The above discussion proved, recall, that we may always pick a circle $C_n$ contained in the previous, such that the area of the disc it bounds is at most half the previous. Now let's apply the previous argument again: take the intersection of all the $D_n$, it contains a disc, that disc has positive area, which is nonsense since its area is at most $\text{Area}(D_1)/2^n$ for all $n$. This contradicts our most crucial assumption: the existence of a decomposition into circles! Thus proven what we wanted to prove.

I see no way to do this using cohomological arguments. If you demand that the circles are a foliation of the plane it's much easier to prove impossibility.

$\endgroup$
  • 1
    $\begingroup$ I don't think I ever used that $D_i$ are discs in an essential way, so because Jordan generalizes in higher dinensions, this answer should apply to $(n-1)$-spheres in $n$-space. $\endgroup$ – user98602 Feb 15 '16 at 20:22
  • $\begingroup$ thanks a lot, nice approach, it is what I was looking for. Thank you! $\endgroup$ – L.F. Cavenaghi Feb 17 '16 at 17:14
9
$\begingroup$

Suppose it can be written as a union of circles. Now consider a circle $C^1$. let $c_1$ denote its center. Then there exists a circle $C^2$ containing $c_1$. Let $c_2$ be the center of $C^2$. Then there exists $C^3$ a circle containing $c_2$. Now consider the sequence of center of circles {$c_i$}. Then since {$c_i$} is a bounded sequence , so by Bolzano-Weierstrass property of $\mathbb R^2$, it has a convergent sequene which will converge to $c$. Now there should exists a cicle $C$ contatining $c$. But then $C$ will intersect $C^n$ for some large $n$( Because observe that radius of $C^{i+1}\leq 1/2$ radius of $C^i$). Contradiction.

$\endgroup$
  • 1
    $\begingroup$ @BalarkaSen See I edited it, 1st of all $C^{i+1}$ lies inside $C^i$ and there radius will converge to zero eventually. $\endgroup$ – Anubhav Mukherjee Feb 15 '16 at 19:53
  • 4
    $\begingroup$ This seems to only work if we assume that the circles are round, as opposed to just subspaces of $\Bbb R^2$ homeomorphic to $S^1$. $\endgroup$ – user98602 Feb 15 '16 at 20:01
  • 1
    $\begingroup$ This needs a bit of polishing; the post currently doesn't state that $C^{i+1}$ is contained in $C^i$ (i.e. contained in the disk bounded by $C^i$) but uses it to derive a contradiction nonetheless. $\endgroup$ – Danu Feb 15 '16 at 20:02
  • 3
    $\begingroup$ I do not think this answer should be downvoted. $\endgroup$ – user98602 Feb 15 '16 at 20:09
  • 1
    $\begingroup$ @Anubhav.K: I am unsure how to get the area to decrease like that without going to the pains in my answer. $\endgroup$ – user98602 Feb 15 '16 at 20:17

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.