13

I want to prove that $\mathbb R ^n $ without a finite number of points is simply connected for $n\geq 3$. Let $X$ be that finite set of points. My idea is to prove this by induction on cardinality of $X$.

Base case with $|X|=1$ follows from the fact that $S^n$ is simply connected and it is a deformation retract of $\mathbb R^{n+1}$ without a point.

For inductive case with $n\geq 2$, I want to prove that there are two distinct parallel affine planes $P $ and $Q $ such that:

1) neither $P$ nor $Q$ intersect $X$;

2) if we denote with $A_+,A_-$ the connected components of $\mathbb R^n \setminus P\cup X$ and with $B_+,B_-$ the connected components of $\mathbb R^n \setminus Q\cup X$, such that $A_+$ contains $Q$ and $B_+$ contains $P$, then $X$ intersects $A_-$ and $B_-$ in at least one point.

(Maybe it is less complicate to prove that closure of $A_+ \cap B_+ $ does not contain any point of $X$)

Then I can apply inductive hypotesis, and $A_+$ and $B_-$ are simply connected. Moreover, $A_+ \cup B_+ = \mathbb R^n \setminus X$ and $A_+ \cap B_+$ is arc-connected, so I can apply Van Kampen.

How can I prove the existence of such two planes?

Ervin
  • 1,275
  • 1
    I haven't carefully checked your strategy, but existence of hyperplanes is easy: Since $X$ is finite, only finitely many real numbers appear as first coordinate of a point of $X$, so the hyperplane ${x_{1} = c}$ misses $X$ for all but finitely many $c$. – Andrew D. Hwang Mar 12 '15 at 15:56
  • What is 'Van Kampen'? – copper.hat Mar 12 '15 at 16:10
  • http://en.wikipedia.org/wiki/Seifert%E2%80%93van_Kampen_theorem – Ervin Mar 12 '15 at 17:25
  • 3
    just some little error to correct: it is $S^{n-1}$ that is a deformation retract of $\mathbb{R}^n$ minus one point, not $S^n$ – sure Mar 13 '15 at 11:43

4 Answers4

17

If you're happy with deformation retraction arguments, you can do this more quickly than that. Put disjoint balls around each point, and join them with thin paths in some order. I claim that $\mathbb{R}^n$ minus these points deformation retracts to the boundaries of these balls together with these thin paths; that is, $\mathbb{R}^n$ minus $k$ points is homotopy equivalent to a wedge sum of $k$ copies of $S^{n-1}$.

Qiaochu Yuan
  • 419,620
  • beautiful simple idea – sure Mar 13 '15 at 11:47
  • 1
    Could you briefly describe what the retraction looks like? The points inside the balls project onto the boundaries, but what about the other points? – CuriousKid7 Aug 05 '18 at 07:33
  • Can't you just say that given any 2 paths you enclose them in a sphere, move all the punctures outside the sphere using a continuous map, keeping the paths in the sphere, and then use a straight line homotopy between the paths, which is defined as the sphere is convex. The composition of the homeomorphism and the homotopy is a homotopy, hence the space is simply connected as we have a homotopy between any two paths. – Ivan Aug 01 '19 at 07:58
  • More directly you can just define the straight line homotopy. If this intersects with any of the points modify it to avoid them - you only need to do this a finite number of times which is trivial - when the line hits a small enough sphere around the point you can replace the line inside the sphere by the great circle between the two intersections. – Ivan Aug 01 '19 at 08:01
  • Can you provide a hint on how to construct the deformation retraction? – Srinath Prasad Oct 07 '21 at 08:49
5

Here is a longwinded way involving tedious path surgery.

First we need to show that $Y=\mathbb{R}^n\setminus X$ is path connected. Pick $x,y \in Y$.Choose $d \neq 0$ such that $d \bot (x-y)$. Pick $\lambda \in \mathbb{R}$, $t \in [0,1]$ and let $p_\lambda(t) = x + t(y-x) + (1-|2t-1|) \lambda d$. (It is easy to see that $p_\lambda$ is the polygonal path $(x,{x+y \over 2}+ \lambda d, y)$.)

Since $x-y,d$ are linearly independent, it is easy to see $p_{\lambda_1}=p_{\lambda_2}$ iff $\lambda_1 = \lambda_2$ . In fact, if $t_1, t_2 \in (0,1)$, then $p_{\lambda_1}(t_1) = p_{\lambda_2}(t_2)$ iff $\lambda_1 = \lambda_2$ and $t_1 = t_2$.

Each $p_\lambda$ is a path between $x,y$ and at most $|X|$ of these can intersect $X$. Since there are an uncountable number of such paths, there is at least one path joining $x,y$ hence the set is path connected and so is connected.

Now suppose $x_0 \in Y$ and $\gamma:[0,1] \to Y$ is a closed path based at $x_0$. Since $Y$ is open, we see that $\gamma$ is homotopic to a polygonal closed path in $Y$ also based at $x_0$. Hence we may take $\gamma$ to be polygonal, that is, straight lines joining a finite number of points $x_0=\gamma_0,...,\gamma_m = x_0$.

Now consider the finite collection of points $A=\{\gamma_k\} \cup X$. Pick a hyperplane $H$ passing through $x_0$ such that the orthogonal projections onto $H$ of the points in $A$ are distinct. (There are only a finite number of orientations of the hyperplane such that two points in $A$ project to the same point.) Let $\Pi$ be the orthogonal projection operator (projects onto the subspace parallel to $H$), and let $h$ be the normal of the hyperplane.

Since $Y$ is open, we see that $B(x_0,\epsilon) \subset Y$ for some $\epsilon>0$. Let $H_\eta = H+ \{ \eta h\}$. By choosing $\eta$ sufficiently small, we can shift the hyperplane such that it intersects $B(x_0,\epsilon)$ but passes through none of the points $X$.

Let $\phi$ be the (affine) orthogonal projection onto $H_\eta$.

Let $B= \{ y | \Pi y = \Pi x \text{ for some } x \in X \}$ (a finite collection of lines perpendicular to $H_\eta$). By construction, none of the points $\gamma_k$ lie in $B$, but it is possible that some segment $[\gamma_i, \gamma_{i+1}]$ intersects $B$.

Suppose a segment $[\gamma_i, \gamma_{i+1}]$ intersects $B$. Pick a direction $d$ that is perpendicular to $\gamma_{i+1}-\gamma_{i}$ and $h$ (this is where $n\ge 3$ comes in). As above, define $p_\lambda(t) = \gamma_i + t(\gamma_{i+1}-\gamma_{i}) + (1-|2t-1|) \lambda d$, and let $N= \{ \lambda | p_\lambda([\gamma_i, \gamma_{i+1}]) \cap B \neq \emptyset \}$. Note that $N$ is finite, hence there is some $\delta>0$ such that $p_\lambda([\gamma_i, \gamma_{i+1}])$ does not intersect $B$ for $\lambda \in (0,\delta]$. Hence $p_\lambda([\gamma_i, \gamma_{i+1}])$ does not intersect $X$ for $\lambda \in [0,\delta]$. Hence we can continuously modify the path $\gamma$ by adding the point ${\gamma_i +\gamma_{i+1} \over 2}+ \delta d$ while remaining in $Y$. Repeat this process for all segments that intersect $B$. Hence the original path is homotopic in $Y$ to a curve that does not intersect $B$.

The points $x_0, \phi(x_0)$ are in $B(x_0,\epsilon)$ and since the ball is convex, we can see that the modified curve is homotopic in $Y$ to the same curve with the points $x_0, \phi(x_0)$ prepended (that is, the points on the path are $x_0, \phi(x_0), x_0=\gamma_0, ...$). In a similar manner, add the points $\phi(x_0), x_0$ to the end of the path.

The modified path looks like $x_0, \phi(x_0), \gamma_1, ...,\gamma_n, \phi(x_0), x_0$ (the $\gamma_i$ are the modified points).

Now consider the map $\theta_t(x) =(1-t)x+ t \phi(x)$ apply the map to the portion of the curve $\phi(x_0), \gamma_1, ...,\gamma_n, \phi(x_0)$. Hence the modified path is homotopic in $Y$ to the path $x_0, \phi(x_0), \phi(\gamma_1), ...,\phi(\gamma_n), \phi(x_0), x_0$, and since the points $\phi(x_0), \phi(\gamma_1), ...,\phi(\gamma_n), \phi(x_0)$ lie in the convex set $H_\eta \subset Y$ the curve is homotopic to the curve $x_0, \phi(x_0), x_0$, and since the ball is convex, this curve is homotopic in $Y$ to the constant curve $t \mapsto x_0$.

copper.hat
  • 172,524
  • Isn't the OP looking to prove $\mathbf{R}^{n} \setminus X$ is simply connected...? – Andrew D. Hwang Mar 12 '15 at 16:18
  • @user86418: Thanks! I need to learn to read first! – copper.hat Mar 12 '15 at 16:19
  • Major revision of answer to wrong question. – copper.hat Mar 13 '15 at 03:45
  • (+1): Here's my understanding of the strategy. Show there exists an affine hyperplane $H$ such that the finite set $X$ is disjoint from the cylinder of lines orthogonal to $H$ and hitting the image of $\gamma$; push $\gamma$ to $H$ along this cylinder (prepending and postpending segments from $x_{0}$ to $H$ to keep $\gamma$ based at $x_{0}$), then homotop inside the convex set $H$, and finally homotop to a constant by retracting a segment. (There are some nifty, memorable tricks in here, e.g., the quadratic splines! :) – Andrew D. Hwang Mar 13 '15 at 13:33
  • @user86418: That's the idea. Definitely not my area, so I am sure I am missing some standard terminology & results! – copper.hat Mar 13 '15 at 14:49
1

Actually, the first comment on my question helped me to find a simple solution.

The proof is by induction on $|X|$

If $|X|=1$, then WLOG we can suppose $X=\{0\}$: then $\mathbb R^n \setminus \{0\} $ deforms into $S^{n-1}$ that is simply connected for $n\geq 3$.

If the thesis is true for $|X|<k$, let us prove that the thesis is true for $|X|=k$. WLOG, we can suppose that $X=\{p_1 ,..., p_k\}$, with $(p_i)_1\leq (p_{i+1})_1 $ (if $x\in \mathbb R^n$, then by definition $(x)_1$ is the first coordinate of $x$ with respect to canonical basis). We can suppose also that there exists $i$ such that $(p_i)_1<(p_{i+1})_1$. There is no loss of generality because $k\geq 2$ and so there are at least two points of $X$ that have distinct coordinates. Define then $\delta = \frac {(p_{i+1})_1 - (p_{i})_1}{3}$, and name $A $ the open set of points such that first coordinate is $>(p_{i+1})_1 - 2\delta$ and $B$ the open set of points such that first coordinate is $<(p_i)_1 +2\delta$. Then $A$ and $B$ are homeomorphic to $\mathbb R ^n $ and $A$ and $B$ both intersect $X$ in less than $k$ points. Then, by inductive hypotesis, $A\setminus X$ and $B\setminus X$ are both simply connected; moreover, $A\cap B = A\setminus X \cap B\setminus X =$ a convex set, and $A\cup B \setminus X = \mathbb R ^n \setminus X$. Then we can apply Van Kampen.

Ervin
  • 1,275
1

Definition of simply connected only requires trivial fundamental group together with path-connectedness. The argument for both cases is intuitively the same, just utilize an extra dimension to go around the holes.

If it so happens that the straight line between two points passes through a hole, cut the line before and after the hole and append a half-circle around it. Inductively this gives a path between any two points in the desired space.

For the triviality of the loop classes, consider locally around one of the holes. Fit a plane to the largest segment of the loop possible and if the hole is coplanar, homotope in any "normal" direction (relative to the plane) to go around. Induct on the number of holes and that should be it. This is not very rigorous, but it can probably be parameterized if one tries hard enough.

Kevin
  • 11