Fat Polygonal Partitions with Applications
to Visualization and EmbeddingsThanks: A preliminary version of this paper appeared in the Proceedings of the 24th ACM Symposium on Computational Geometry [18].

Let $d:=\diam(P)$ and let $uv$ be a diagonal of $P$ that has length $d$. Consider the bounding box $B$ of $P$ that has two edges parallel to $uv$. We call the edge of $B$ parallel to and above $uv$ its top edge, and the edge of $B$ parallel to and below $uv$ its bottom edge. Let $r$ be a vertex of $P$ on the top edge of $B$ and let $s$ be a vertex on its bottom edge—see Fig. 4.

Let $e_{1}$ and $e_{2}$ be the edges of $P$ incident to $r$, and let $e_{3}$ and $e_{4}$ be the edges incident to $s$. Let the angles $\alpha_{1},\ldots,\alpha_{4}$ be defined as in Fig. 4. We distinguish two cases.

• •

  Case (a): none of $e_{1},e_{2},e_{3},e_{4}$ are parallel.
  By condition (i), this implies that the angles any two edges make is at least $\phi$. Hence, we have

  	$\displaystyle\phi$	$\displaystyle\leq\myangle(e_{1},e_{3})\leq\max(\alpha_{1},\alpha_{3}),$		(1)
  	$\displaystyle\phi$	$\displaystyle\leq\myangle(e_{2},e_{4})\leq\max(\alpha_{2},\alpha_{4}),$		(2)
  	$\displaystyle\phi$	$\displaystyle\leq\myangle(e_{1},e_{4})\leq\alpha_{1}+\alpha_{4},$		(3)
  	$\displaystyle\phi$	$\displaystyle\leq\myangle(e_{2},e_{3})\leq\alpha_{2}+\alpha_{3}.$		(4)

  By (3) we have $\alpha_{1}+\alpha_{4}\geq\phi$. Now assume without loss of generality that $\alpha_{1}\geq\phi/2$. If $\alpha_{2}\geq\phi/2$ as well, then

  $$\area(uvr)\geq(d^{2}/4)\cdot\sin(\phi/2).$$

  Since $uvr\subset P$, this implies that

  $$\ar(P)\leq\frac{d^{2}}{(d^{2}/4)\cdot\sin(\phi/2)}=O(1/\phi).$$

  If $\alpha_{2}<\phi/2$, then we use (2) and (4) to conclude that $\alpha_{4}\geq\phi$ and $\alpha_{3}\geq\phi/2$. Hence, we now have $\area(uvs)\geq(d^{2}/4)\cdot\sin(\phi/2)$, which implies that $\ar(P)=O(1/\phi)$.

• •

  Case (b): some edges in $e_{1},e_{2},e_{3},e_{4}$ are parallel.
  By conditions (i)–(iii), two edges of $P$ can be parallel only if they are contained in opposite edges of $U$. Hence, $|rs|\geq 1$. Moreover, since $uv$ defines the diameter we have $|uv|\geq|rs|$. Let $\alpha:=\myangle(uv,rs)$, as illustrated in Fig. 5.

  If $\alpha\geq\min\{\phi,\pi/4\}$, this is easily seen to imply that $\area(P)\geq\area(urvs)=\Omega(\phi)$, which means that $\ar(P)=O(1/\phi)$. Now consider the case where $\alpha<\phi$ and $\alpha<\pi/4$. We show that this leads to a contradiction. Let $x:=uv\cap rs$ and assume without loss of generality that, as in Figure 5, we have $\alpha=\angle rxv$. Then $\myangle(e_{1},e_{3})\leq\alpha<\phi$. By conditions (i)–(iii) this can only happen if $e_{1}$ and $e_{3}$ are contained in opposite edges of $U$, if we assume that $r$ and $s$ are chosen such that they maximize the distances between $r$ and $s$, which can be done without loss of generality. We have $\myangle(e_{3},rs)\leq\myangle(uv,rs)=\alpha\leq\phi<\pi/4$. However, it is impossible to place two points $r$ and $s$ on opposite sides of $U$ such that $\myangle(e_{3},rs)<\pi/4$. The smallest angle one can obtain is $\pi/4$.

∎

The proof is by induction on $k$.

For $k=0$, we have $\nu=\myroot(\mathcal{T}^{\prime})$ and $P(\nu)$ is a unit square. Hence, $P(\nu)$ is $(\pi/2)$-separated and therefore also $(\pi/6)$-separated.

For $k>0$ we argue as follows. By the induction hypothesis, the polygon $P(\mu)$ corresponding to the parent $\mu$ of $\nu$ is $(\pi/(2k+4))$-separated. Moreover, by construction it has at most $4+(k-1)=k+3$ edges. Consider the sorted (circular) sequence of angles that these edges make with the $x$-axis. By the pigeon-hole principle, there must be two adjacent angles that are at least $\pi/(k+3)$ apart. Hence, the cut $c$ that is chosen to partition $P(\mu)$ makes an angle at least $\pi/(2k+6)$ with all edges of $P(\mu)$, and by the induction hypothesis, all the other angles are at least $\pi/(2k+4)$. ∎

We distinguish two cases, depending on whether $a\leq 1/k^{2}$ (that is, we are cutting off a relatively small subpolygon) or not.

Case 1: $a\leq 1/k^{2}$. Let $\phi$ be the smallest angle of $P$, and let $v$ be a vertex of $P$ whose interior angle is $\phi$. Since $P$ has $k$ vertices, we have

$$\phi\leq\pi\left(1-\frac{2}{k}\right).$$

Let $\ell$ be the angular bisector at $v$. Consider the cut $c$ orthogonal to $\ell$ such that $\area(P_{1})=a\cdot\area(P)$, where $P_{1}$ is the subpolygon induced by $c$ having $v$ as a vertex—see Figure 6(a) for an illustration. Let $P_{2}$ be the other subpolygon. Clearly, $P_{1}$ and $P_{2}$ are convex polygons of the required area with at most $k+1$ vertices each. Therefore, it remains to bound the aspect ratios of $P_{1}$ and $P_{2}$.

Since $P_{2}\subset P$, we have

	$\displaystyle\ar(P_{2})$	$\displaystyle=\frac{\diam(P_{2})^{2}}{\area(P_{2})}\leq\frac{\diam(P)^{2}}{(1-a)\cdot\area(P)}=\frac{\ar(P)}{1-a}<\ar(P)\left(1+2a\right)$
		$\displaystyle<\ar(P)\left(1+\frac{2}{k^{2}}\right)<\ar(P)\left(1+\frac{1}{k}\right).$

We next bound $\ar(P_{1})$. Let $x_{1},x_{2}$ be the two endpoints of the cut $c$, and let $t$ be the distance between $x_{1}$ and $x_{2}$. Let $h$ be the distance from $v$ to $c$. We distinguish between two subcases.

Case 1.1: $t\geq h/k^{2}$. Since $P$ is convex, the triangle $vx_{1}x_{2}$ is contained in $P_{1}$. Therefore,

$$\area(P_{1})\geq h\cdot t/2\geq h^{2}/(2k^{2}).$$

On the other hand, since $c$ is normal to the bisector of the angle of $v$, it follows that $P_{1}$ is contained inside a rectangle of width $h$ and height $H$, with

	$\displaystyle H$	$\displaystyle\leq 2\cdot h\cdot\tan(\phi/2)\leq 2\cdot h\cdot\tan\left(\frac{\pi(1-2/k)}{2}\right)$
		$\displaystyle\leq 2\cdot h/\tan(\pi/k)\leq 2\cdot h\cdot k/\pi.$

Thus, $\diam(P_{1})<h(1+2\cdot k/\pi)$. It follows that

$$\ar(P_{1})=\frac{\diam(P_{1})^{2}}{\area(P_{1})}<\frac{(h+2\cdot h\cdot k/\pi)^{2}}{h^{2}/(2k^{2})}<k^{5}.$$

Case 1.2: $t<h/k^{2}$. Let $\ell_{1}$ be the line passing through $v$ and $x_{1}$, and let $\ell_{2}$ be the line passing through $v$ and $x_{2}$. Let $\gamma$ be the angle between $\ell_{1}$ and $\ell_{2}$. Observe that $P_{2}$ is contained between $\ell_{1}$ and $\ell_{2}$. Therefore, if $u$ is the point in $P_{2}$ farthest away from $v$ we have

$$\frac{\gamma}{2\pi}\pi|uv|^{2}\geq\area(P_{2}),$$

where $|uv|$ denotes the length of the segment $uv$. It follows that

$$\diam(P)^{2}\geq|uv|^{2}\geq\frac{2}{\gamma}(1-a)\cdot\area(P).$$

Therefore,

$$\ar(P)=\frac{\diam(P)^{2}}{\area(P)}\geq\frac{2}{\gamma}\left(1-a\right)\geq\frac{2}{\gamma}\left(1-\frac{1}{k^{2}}\right).$$

We now give an upper bound on the diameter of $P_{1}$. Assume without loss of generality that $|vx_{2}|\geq|vx_{1}|$. Consider a segment $y_{1}y_{2}$ parallel to $x_{1}x_{2}$ with $y_{1},y_{2}\in\bd P$ such that $y_{1}y_{2}$ lies between $x_{1}x_{2}$ and $v$—see Figure 6(b). Let $h^{\prime}$ be the distance between $v$ and $y_{1}y_{2}$. We first argue that $|y_{1}y_{2}|\leq 2t$.

Assume for the sake of contradiction that $|y_{1}y_{2}|>2t$. Let $g_{1}$ be the line passing through $y_{1}$ and $x_{1}$, and let $g_{2}$ be the line passing through $y_{2}$ and $x_{2}$. Observe that since $|y_{1}y_{2}|>|x_{1}x_{2}|$, the lines $g_{1}$ and $g_{2}$ intersect in a point $w$ such that $P_{2}$ is contained in the triangle $x_{1}x_{2}w$. Furthermore, the polygon $vy_{1}x_{1}x_{2}y_{2}$ is contained in $P_{1}$. If $h^{\prime}\geq h/2$, then the area of the triangle $vy_{1}y_{2}$ is greater or equal to the area of the triangle $x_{1}x_{2}w$. Therefore, $\area(P_{1})\geq\area(P_{2})$, contradicting the fact that $a\leq 1/k^{2}$. If, on the other hand, $h^{\prime}<h/2$, then the area of the quadrilateral $y_{1}x_{1}x_{2}y_{2}$, is greater than the area of the triangle $x_{1}x_{2}w$, again implying that $\area(P_{1})\geq\area(P_{2})$, a contradiction. Therefore, we obtain that $|y_{1}y_{2}|\leq 2t$.

It now follows that any point $q\in P_{1}$ is at distance at most $2t$ from the line segment $vx_{2}$. Moreover, we have $t<h/k^{2}\leq|vx_{2}|/k^{2}$. Hence,

$$\diam(P_{1})=\max_{q,q^{\prime}\in P_{1}}|qq^{\prime}|\leq 2t+|vx_{2}|+2t\leq|vx_{2}|\left(1+\frac{4}{k^{2}}\right).$$

Let $x^{*}$ be the point on the line segment $x_{1}x_{2}$ that is closest to $v$. Since $|vx_{2}|\geq h$, we have

$$\area(P_{1})\geq\frac{\gamma}{2\pi}\pi|vx^{*}|^{2}\geq\frac{\gamma}{2}(|vx_{2}|-t)^{2}\geq\frac{\gamma}{2}|vx_{2}|^{2}\left(1-\frac{1}{k^{2}}\right).$$

Therefore,

	$\displaystyle\ar(P_{1})$	$\displaystyle=\frac{\diam(P_{1})^{2}}{\area(P_{1})}\leq\frac{2}{\gamma}\cdot\frac{(1+4/k^{2})^{2}}{1-1/k^{2}}\leq\ar(P)\frac{(1+4/k^{2})^{2}}{(1-1/k^{2})^{2}}$
		$\displaystyle\leq\ar(P)\cdot(1+6/k^{2})^{2}\leq\ar(P)\cdot(1+2/k)^{2}$
		$\displaystyle\leq\ar(P)\cdot(1+6/k).$

Case 2: $a>1/k^{2}$.

Case 2.1: $\ar(P)\leq k^{6}$. In this case any cut giving the two subpolygons $P_{1}$ and $P_{2}$ the required areas works. Indeed,

$$\ar(P_{1})=\frac{\diam(P_{1})^{2}}{\area(P_{1})}\leq\frac{\diam(P)^{2}}{a\cdot\area(P)}\leq k^{2}\cdot\ar(P)\leq k^{8}$$

and

$$\ar(P_{2})=\frac{\diam(P_{2})^{2}}{\area(P_{2})}\leq\frac{\diam(P)^{2}}{(1-a)\cdot\area(P)}\leq 2\cdot\ar(P)\leq 2\cdot k^{6}<k^{7}.$$

Case 2.2: $\ar(P)>k^{6}$. Pick points $v_{1},v_{2}\in P$, such that $|v_{1}v_{2}|=\diam(P)$. For each $z\in[0,\diam(P)]$, let $\ell(z)$ be a line normal to $v_{1}v_{2}$ that is at distance $z$ from $v_{1}$ and intersects $P$. Note that $\ell(0)$ contains $v_{1}$ and $\ell(\diam(P))$ contains $v_{2}$. Define $f(z)$ to be the length of the intersection of $P$ with $\ell(z)$. Observe that

$$\area(P)=\int_{z=0}^{\diam(P)}f(z)dz$$

Pick $s_{1},s_{2}\in[0,\diam(P)]$, so that

$$a\cdot\area(P)=\int_{z=0}^{s_{1}}f(z)dz=\int_{z=\diam(P)-s_{2}}^{\diam(P)}f(z)dz.$$

Let $Q_{1}$ be the part of $P$ that is contained between $\ell(0)$ and $\ell(s_{1})$. Similarly, let $Q_{2}$ be the part of $P$ that is contained between $\ell(\diam(P)-s_{2})$ and $\ell(\diam(P))$. Clearly, both $Q_{1}$ and $Q_{2}$ are convex polygons with at most $k+1$ vertices.

First, we will show that

$$\min\left\{\frac{\area(Q_{1})}{s_{1}},\frac{\area(Q_{2})}{s_{2}}\right\}\leq\frac{\area(P)}{\diam(P)}$$

Assume for a contradiction that both $\area(Q_{1})/s_{1}$ and $\area(Q_{2})/s_{2}$ are greater than $\area(P)/\diam(P)$. It follows that there exist $z_{1}\in[0,s_{1}]$ and $z_{2}\in[\diam(P)-s_{2}]$ such that $f(z_{1})>\area(P)/\diam(P)$ and $f(z_{2})>\area(P)/\diam(P)$. Since $P$ is convex, $f$ is a bitonic function. Therefore, for each $z\in[z_{1},z_{2}]$, we have $f(z)>\area(P)/\diam(P)$. It follows that

$$\area(P)=\area(Q_{1})+\area(Q_{2})+\area(P\setminus(Q_{1}\cup Q_{2}))>\frac{\area(P)}{\diam(P)}\cdot\diam(P),$$

a contradiction.

We can therefore assume without loss of generality that

$$\frac{\area(Q_{1})}{s_{1}}\leq\frac{\area(P)}{\diam(P)}.$$

Note that this implies

$$s_{1}\geq a\cdot\diam(P).$$

We set $P_{1}=Q_{1}$, and $P_{2}=P\setminus Q_{1}$. It remains to bound $\ar(P_{1})$ and $\ar(P_{2})$.

By the convexity of $P$, we have

$$\area(P)\geq\max_{z\in[0,\diam(P)]}f(z)\cdot\diam(P)/2.$$

Since $\ar(P)>k^{6}$, it follows that

$$\max_{z\in[0,\diam(P)]}f(z)\leq 2\cdot\frac{\area(P)}{\diam(P)^{2}}\cdot\diam(P)<\frac{2}{k^{6}}\cdot\diam(P).$$

This implies that $P$ is contained inside a rectangle with one edge of length $\diam(P)$ parallel to $v_{1}v_{2}$, and one edge of length $\frac{4}{k^{6}}\cdot\diam(P)$ normal to $v_{1}v_{2}$. Thus,

$$\diam(P_{1})\leq s_{1}+\frac{4}{k^{6}}\cdot\diam(P).$$

Let $\sigma_{1}$, $\sigma_{2}$ be the two points where $\ell(s_{1})$ intersects $\bd P$. Let $\zeta_{1}$, $\zeta_{2}$, be the lines passing through $v_{1}$ and $\sigma_{1}$, and $v_{1}$ and $\sigma_{2}$, respectively. Let also $\sigma_{1}^{\prime}$ and $\sigma_{2}^{\prime}$ be the points where $\zeta_{1}$ and $\zeta_{2}$, respectively, intersect $\ell(\diam(P))$—see Figure 7.

By the convexity of $P$ and $P_{1}$, we have

$$\area(P_{1})\geq\area(v_{1}\sigma_{1}\sigma_{2})=\left(\frac{s_{1}}{\diam(P)}\right)^{2}\cdot\area(v_{1}\sigma_{1}^{\prime}\sigma_{2}^{\prime})\geq\left(\frac{s_{1}}{\diam(P)}\right)^{2}\cdot\area(P).$$

Since $\area(P_{1})=a\cdot\area(P)$, it follows that $s_{1}\leq\sqrt{a}\cdot\diam(P).$ Using that $a>1/k^{2}$ we can now derive

	$\displaystyle\ar(P_{1})$	$\displaystyle=\frac{\diam(P_{1})^{2}}{\area(P_{1})}\leq\frac{(s_{1}+4\cdot\diam(P)/k^{6})^{2}}{\area(P_{1})}$
		$\displaystyle\leq\frac{(\sqrt{a}\cdot\diam(P)+4\cdot\diam(P)/k^{6})^{2}}{a\cdot\area(P)}$
		$\displaystyle<\frac{\diam(P)^{2}}{\area(P)}\cdot\left(1+\frac{4}{k^{6}\sqrt{a}}\right)^{2}\leq\ar(P)\cdot\left(1+\frac{8}{k^{4}}+\frac{16}{k^{16}}\right)$
		$\displaystyle\leq\ar(P)\cdot\left(1+\frac{1}{k}\right).$

Since $f$ is bitonic, it follows that

$$\min_{z\in[s_{1},\diam(P)-s_{2}]}f(z)\geq\min\left\{\max_{z\in[0,s_{1}]}f(z),\max_{z\in[\diam(P)-s_{2},\diam(P)]}f(z)\right\}.$$

Therefore,

$$\frac{\area(P_{2})}{\diam(P)-s_{1}}\geq\frac{\area(P_{1})}{s_{1}}.$$

Because $P_{2}$ is contained in a rectangle with one edge of length $\diam(P)-s_{1}$ parallel to $v_{1}v_{2}$, and one edge of length $\frac{4}{k^{6}}\cdot\diam(P)$ normal to $v_{1}v_{2}$, we have

$$\diam(P_{2})\leq\diam(P)-s_{1}+\frac{4}{k^{6}}\cdot\diam(P).$$

Putting everything together, we get

	$\displaystyle\ar(P_{2})$	$\displaystyle=\frac{\diam(P_{2})^{2}}{\area(P_{2})}\leq\frac{(\diam(P)\cdot(1+4/k^{6})-s_{1})^{2}}{(1-a)\cdot\area(P)}$
		$\displaystyle\leq\ar(P)\cdot\left(\frac{1+4/k^{6}-a}{\sqrt{1-a}}\right)^{2}\leq\ar(P)\cdot\left(1+4\cdot\frac{\sqrt{2}}{k^{6}}\right)^{2}$
		$\displaystyle\leq\ar(P)\cdot\left(1+\frac{1}{k^{2}}\right)^{2}\leq\ar(P)\cdot\left(1+\frac{3}{k^{2}}\right)$
		$\displaystyle\leq\ar(P)\cdot\left(1+\frac{1}{k}\right).$