9
$\begingroup$

Let $X \subset \mathbb{R}^n$ be a union of countably many affine subspaces of $\mathbb{R}^n$ with dimension less or equal $n-2$.

Prove that $\mathbb{R}^n \setminus X$ is path connected.

Assume $n = 2$. Then we need to show, that the plane with countably many points removed is path connected.

Take $x_1, x_2 \in \mathbb{R}^2 \setminus X$. We find two nonparallel lines in $\mathbb{R}^2$ going through $x_1$ and $x_2$ which do not cross $X$. It can be done because the cardinality of the set of all lines through $x_i$ is $|\mathbb{R}|$ and the cardinality of $X$ is just $|\mathbb{N}|$. These two lines intersect at some point and we get a path between $x_1$ and $x_2$ in $\mathbb{R}^2 \setminus X$.

Now I want to reduce the initial problem to the case with $n=2$. In order to do so I need to

prove that for any two points $x_1$ and $x_2$ in $\mathbb{R}^n \setminus X$ there is a plane in $\mathbb{R}^n$ containing both these points and intersecting $X$ in at most countably many points.

How can I do that?

  • 0
    Maybe some measure theory? There's a map from $S^{n-2}$ to the set of hyperplanes joining $x_1$ and $x_2$, I think. If the set of all the "bad" hyperplanes corresponds to a measure-zero set in $S^{n-2}$, then there must be a "good" hyperplane.2017-02-23
  • 0
    @AkivaWeinberger I understand how it should work. When $n = 3$, then we can rotate the plane containing $x_1$ and $x_2$ only around the line connecting these points. Then we have $|\mathbb{R}|$ different positions for the plane and again only $|\mathbb{N}|$ of them are "bad". However when $n > 3$ we can rotate the plane in many different directions and it is hard to understand which positions of the plane are "bad". It looks like for $n=4$, bad positions are represented by finitely many lines on the unit sphere. How can I do it more or less formally?2017-02-23
  • 0
    @Akiva Weinberger You speak about "bad hyperplanes" but here it is "bad $n-2$ dimensional spaces".2017-02-23
  • 1
    @JeanMarie True. Sorry.2017-02-23
  • 0
    In the case $n=2$ there are lines $l_1$ and $l_2$ thru $x_1,x_2,$ respectively, that are not parallel, and that are not subsets of $X$. But that does not imply that $l_1$ and $l_2$ are disjoint from $X$, so it does not follow that the union of the line-segments $ x_1p $ and $px_1$ (where $\{p\}=l_1\cap l_2$) is disjoint from $X.$2017-03-07
  • 0
    @user254665 That is not the argument the OP made.2017-03-08

3 Answers 3

0

Proof for $\mathbb R^3:$ Suppose $L_1,L_2, \dots$ are lines in $\mathbb R^3.$ Assume $p,q\in \mathbb R^3 \setminus (\cup L_k).$ WLOG, we can take $p=(0,0,1),q=(0,0,-1).$

For each $m>0$ let $P_m$ denote the plane spanned by $p$ and $(1,m,0).$ Let $S$ be the unit sphere, and define the unit circle $C_m = P_m\cap S.$ Now each $L_k$ can intersect $S$ in at most two points. It follows that $\cup L_k$ can intersect only countably many $C_m.$ But there are uncountably many $C_m.$ Thus there must exist $m$ such that $(\cup L_k)\cap C_m= \emptyset$ (in fact there are uncountably many such $m$). For any such $m$ we have an entire circle $C_m$ connecting $p$ to $q$ within $\mathbb R^3 \setminus (\cup L_k).$

0

As is suggested in the comments it is possible to do what you are suggesting, but it will require some measure theory.

In particular you have a $\mathbb{P}^{n-2}(\mathbb{R})$ worth of planes through $x_1$ and $x_2$. For any dimension $d \le n-2$ subspace $V$ of $\mathbb{R}^n$ those planes through $x_1$ and $x_2$ that intersect $V$ in a line cut out a subvariety of $\mathbb{P}^{n-2}(\mathbb{R})$ of strictly smaller dimension (cut out by some determinants vanishing) and in particular this has measure zero. The countable union of measure zero sets has measure zero so in particular we can find a plane through $x_1$ and $x_2$ that meets all the subspaces in $X$ in dimension at most 1, as desired.

I'll note that if you want to avoid measure theory and just use facts about cardinality you can still solve the overall problem, but by a slightly different strategy.

Note that for each dimension $n-2$ subspace $V$ in $X$ there is exactly $1$ hyperplane containing $V$ and $x_1$ (and the same for $x_2$). In particular for cardinality reasons we can find nonparallel hyperplanes through $x_1$ and $x_2$ that meet all the subspaces in $X$ in dimension at most $n-3$.

But now we can apply induction to get that the complement of $X$ in each hyperplane is path connected. So to make a path from $x_1$ to $x_2$ we can just find a path from $x_1$ to some point in the intersection of these two hyperplanes, and then a path from that to $x_2$

0

Let $\{ A_k \}_{k = 1}^\infty$ be affine subspaces of $\mathbb{R}^n$ of dimension $n-2$, and take distinct points $x, y \in \mathbb{R}^n$. Translating, one can assume $x = 0$ and $0 \notin A_k \forall k \in \mathbb{N}$.

There is a homeomorphism $\Phi: \mathcal{G} \rightarrow \mathbb{P}^{n-2} ( \mathbb{R} )$ where $\mathcal{G}$ is the space of $n-2$ dimensional affine spaces intersecting $\mathbb{R}y$. We construct this map as follows. Let $U = \mathbb{R} y$ and $V = U^{\perp}$. $V$ is an $n-1$ dimensional vector space. Take an $n-2$ dimensional vector space $W$ containing $y$. $W$ is uniquely determined by $W \cap V$, since $W = W \cap (V \oplus U) = (W \cap V ) \oplus (W \cap U) = (W \cap V) \oplus U$. There is a vector space isomorphism $T : V \rightarrow \mathbb{R}^{n-1}$. $T(W \cap V)$ is then a subspace of $\mathbb{R}^{n-1}$. Obviously \begin{align*} dim_{\mathbb{R}}(T(W \cap V)) = dim_{\mathbb{R}}( W \cap V) = dim_{\mathbb{R}} ( W ) - dim_{\mathbb{R}} ( U ) = n - 2 \end{align*} Thus $T(W \cap V)^{\perp}$ is a line through the origin of $\mathbb{R}^{n-1}$. Define $\Phi (W) = T(W \cap V)^{\perp}$.

The following is rather straightforward, so I leave it as an exercise:

Exercise: For each $k \in \mathbb{Z}$ the set $\mathcal{G}_k = \{ z \in \mathbb{P}^2 ( \mathbb{R} ) : |\Phi^{-1}(z) \cap A_k| \leq 1 \} $ is open and dense in $\mathbb{P}^2 (\mathbb{R})$.

The claim follows therefore from the following lemma:

Lemma: Baire Category Theorem Let $\mathcal{X}$ be a locally compact Hausdorff space and take dense open sets $\{ U_i \}_{i = 1}^\infty$ in $\mathcal{X}$. Then $\cap_{i = 1}^\infty U_i$ is dense in $\mathcal{X}$.

Spaces with this property are called Baire spaces- spaces in which the intersection of countably many dense sets are dense. See here: https://en.wikipedia.org/wiki/Baire_space


Edit: here is an elaboration on how to finish the proof from the lemma, as requested: from the exercise we know that $\mathcal{G}_k$ are open and dense in $\mathbb{P}^2 (\mathbb{R})$. Since $\mathbb{P}^2 ( \mathbb{R} )$ is compact and Hausdorff, $\mathcal{H} = \cap_{i = 1}^\infty \mathcal{G}_k$ is dense by the Baire Category Theorem. But this is much stronger than we need. The upshot is that $\mathcal{H}$ is nonempty. Take $z \in \mathcal{H}$. Then $|\Phi^{-1} (z) \cap A_k| \leq 1 \forall k \in \mathbb{N}_{\geq 1}$, so that $\left|\Phi^{-1} (z) \cap \bigcup_{k \in \mathbb{N}_{\geq 0}} A_k \right|$ is countable. $\Phi^{-1} (z)$ is therefore the desired plane.

  • 0
    Could you please explain how "The claim follows therefore from the following lemma"?2017-03-11
  • 0
    Ok. I made some edits.2017-03-11