Jordan Polygon Theorem/Lemma 1

From ProofWiki
Jump to navigation Jump to search

Lemma

Let $P$ be a polygon embedded in $\R^2$.

Denote the boundary of $P$ as $\partial P$.


$\R^2 \setminus \partial P$ is the union of at most two disjoint path-connected sets.


Proof

The polygon $P$ has $n$ sides, where $n \in \N$.

Denote the vertices of $P$ as $A_1, \ldots, A_n$.

Denote the sides of $P$ as $S_1, \ldots, S_n$, such that each vertex $A_i$ has adjacent sides $S_{i - 1}$ and $S_i$.

We use the conventions that $S_0 = S_n$, and $A_{n + 1} = A_1$.


Let $\ds \delta_i = \map d {S_i, \bigcup_{j \mathop = 1, \ldots, n: \size {i - j} > 1} S_j}$ be the Euclidean distance between a side $S_i$ and all sides not adjacent to $S_i$.

From Distance between Closed Sets in Euclidean Space:

$\delta_i > 0$

Put $\ds \delta = \min_{i \mathop = 1, \ldots, n} \delta_i$.


From Boundary of Polygon is Jordan Curve, it follows that $\partial P$ is equal to the image of a Jordan curve:

$\gamma: \closedint 0 1 \to \R^2$

that is a concatenation of $n$ paths:

$\gamma_1, \ldots, \gamma_n$

Each $\gamma_i$ is a line segment that joins its initial point $A_i$ and its final point $A_{i + 1}$.

Therefore the image of $\gamma_i$ is equal to the side $S_i$.


Let $\mathbf v_i = \dfrac {A_{i + 1} - A_i} {\map d {A_{i + 1}, A_i} }$ be the direction vector of $\gamma_i$ with norm $\norm {\mathbf v_i} = 1$.

Let $\mathbf w_i$ be the vector $v_i$ rotated $\dfrac \pi 2$ radians counterclockwise, so $\norm {\mathbf w_i} = 1$.




For any $\epsilon \in \openint 0 {\dfrac \delta 2}$, we intend to construct two Jordan curves $\sigma, \overline \sigma$ such that:

$\Img \sigma \cup \Img {\overline \sigma} = \set {q \in \R^2: \map d {q, \partial P} = \epsilon}$
Illustration of the Jordan curves $\sigma, \overline \sigma$ which both have distance $\epsilon$ to $\partial P$.
Note the rounded corners of the curves that occur in place of non-convex angles.


For $i \in \set {1, \ldots, n}$, initially define the $\sigma_i$ as the line segment that joins its initial point $A_i + \epsilon \mathbf w_i$ with its final point $A_{i + 1} + \epsilon \mathbf w_i$.


Suppose $\sigma_i$ and $\sigma_{i + 1}$ intersect at some point $p_{i + 1} \in R^2$.

Then re-define the two line segments so that:

the final point of $\sigma_i$ becomes $p_{i + 1}$

and:

the initial point of $\sigma_{i + 1}$ becomes $p_{i + 1}$.

Then define a path $\rho_i$ as the constant function $\map {\rho_i} t = p_{i + 1}$.

Otherwise, define a path $\rho_i$ with initial point $A_{i + 1} + \epsilon \mathbf w_i$ and final point $A_{i + 1} + \epsilon \mathbf w_{i + 1}$, such that:

the image of $\rho_i$ is part of the circumference of the circle with center $A_{i + 1}$ and radius $\epsilon$.


Define the path $\sigma: \closedint 0 1 \to \R^2$ as the concatenation:

$\sigma = \sigma_1 * \rho_1 * \sigma_2 * \rho_2 * \ldots \sigma_n * \rho_n$

Then $\sigma$ is a closed path, as $\sigma_1$ has initial point $A_1$ equal to the final point of $\rho_n$.


Each of the paths $\sigma_i$ and $\rho_i$ is injective.

For all $i, j \in \set {1, \ldots, n}$, $\sigma_i$ only intersects $\rho_{i - 1}$ and $\rho_i$ in their endpoints.

Also, $\sigma_i$ can only possibly intersect $\sigma_{i - 1}$ in its endpoint.

In that case the path $\rho_{i - 1}$ is constant.

Similarly, $\sigma_i$ can only possibly intersect $\sigma_{i + 1}$ in its endpoint.

In that case the path $\rho_{i + 1}$ is constant.

For $\size {i - j} > 1$:

let $x_i \in \Img {\sigma_i} \cup \Img {\rho_i}$
let $x_j \in \Img {\sigma_j} \cup \Img {\rho_j}$.

Let $p_i, p_j \in \partial P$ be two points such that:

$\map d {x_i, p_i} = \map d {x_j, p_j} = \epsilon$


Suppose $x_i \in \Img {\sigma_i}$.

Then we can put:

$p_i = x_i - \epsilon \mathbf w_i$

Suppose $x_i \in \Img {\rho_i}$.

Then we can put $p_i = A_{i + 1}$.


Then $x_i \ne x_j$, as:

\(\ds \map d {x_i, x_j}\) \(=\) \(\ds \map d {x_i, x_j} + \map d {x_i, p_i} + \map d {x_j, p_j} - 2 \epsilon\)
\(\ds \) \(\ge\) \(\ds \map d {p_i, p_j} - 2 \epsilon\) Triangle Inequality for Vectors in Euclidean Space
\(\ds \) \(>\) \(\ds \delta - 2 \epsilon\) as $p_i \in S_i, p_j \in S_j$, and $\map d {S_i, S_j} \ge \delta$
\(\ds \) \(=\) \(\ds 0\)

It follows that $\sigma_i$ and $\rho_i$ do not intersect $\sigma_j$ and $\rho_j$.

It follows that $\sigma$ is a Jordan curve.


Now for $i \in \set {1, \ldots, n}$, initially define the $\overline \sigma_i$ as the line segment that joins its initial point $A_i - \epsilon \mathbf w_i$ with its final point $A_{i + 1} - \epsilon \mathbf w_i$.

Proceed to define $\overline \rho_i$ and re-define $\overline \sigma_i$ similarly to what was done with $\sigma_i$ and $\rho_i$.

Finally, define $\overline \sigma$ as the concatenation:

$\overline \sigma = \overline \sigma_1 * \overline \rho_1 * \overline \sigma_2 * \overline \rho_2 * \ldots \overline \sigma_n * \overline \rho_n$

It follows that $\overline \sigma$ is a Jordan curve as above.


We now show that $\sigma$ and $\overline \sigma$ do not intersect.

First, $\sigma_i$ and $\overline \sigma_i$ do not intersect, as they are both line segments parallel to $S_i$ with a distance of $2 \epsilon$.

Second, $\rho_i$ and $\overline \rho_i$ do not intersect, as one path, say $\rho_i$, is constant with image equal to the crossing point of $\sigma_{i - 1}$ and $\sigma_i$.

The other path, say $\overline \rho_i$, is part of the circumference of the circle that joins the final point of $\overline \sigma_{i - 1}$ with the initial point of $\overline \sigma_i$.

Third, $\sigma_i$ and $\overline \sigma_{i + 1}$ do not intersect, since neither line segment crosses $\partial P$, which would be necessary if the two line segments should intersect.

Similarly, $\sigma_i$ and $\overline \sigma_{i - 1}$ do not intersect.

For all other combinations of $i, j \in \set {1, \ldots, n}$, $\sigma_i$ and $\rho_i$ do not intersect with $\overline \sigma_j$ and $\overline \rho_j$.

This follows as:

$\map d {\sigma_i, S_i} = \map d {\rho_i, S_i} = \map d {\rho_i, S_{i - 1} } = \epsilon$

and as $S_i$, or $S_{i - 1}$, are not adjacent sides to $S_j$, we have $\map d {S_i, S_j} > 2 \epsilon$.

Now we can use the Triangle Inequality for Vectors in Euclidean Space as above, to prove that the paths do not intersect.

$\Box$


Finally, let $q_1, q_2, q_3 \in R^2 \setminus \partial P$.

Let:

$\epsilon = \min \set {\map d {q_1, \partial P} / 2, \map d {q_2, \partial P} / 2, \map d {q_3, \partial P} / 2, \delta}$

For all $i \in \set {1, 2, 3}$, draw a line segment $\LL_i$ joining $q_i$ with any point on the boundary $\partial P$.


As $\map d {q_i, \partial P} > \epsilon$, it follows from the Intermediate Value Theorem that there is a point $x_i$ on $\LL_i$ such that $\map d {x_i, \partial P} = \epsilon$.

Then we have either:

$x_i \in \Img \sigma$

or:

$x_i \in \Img {\overline \sigma}$

when the Jordan curves $\sigma$ and $\overline \sigma$ gets defined from the new value of $\epsilon$.

It follows that at least two out of three of the points $q_1, q_2, q_3$ is path-connected to the same Jordan curve.

As a Jordan curve is a path, it follows that at least two of the points can be connected by a path.

Hence, $R^2 \setminus \partial P$ is the union of at most two disjoint path-connected sets.

$\blacksquare$


Sources



Jeff Ericsson: Computational Topology