This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Convex polygons and separation of convex setsthanks: Partially supported by Conacyt, México.

Eduardo Rivera-Campo Departamento de Matemáticas
Universidad Autónoma Metropolitana - Iztapalapa
[email protected]
Jorge Urrutia Instituto de Matemáticas
Universidad Nacional Autónoma de México
[email protected]
Abstract

We prove that for any collection FF of n2n\geq 2 pairwise disjoint compact convex sets in the plane there is a pair of sets AA and BB in FF such that any line that separates AA from BB separates either AA or BB from a subcollection of FF with at least n/18n/18 sets.

Keywords.- Convex polygon. Plane Compact Convex Set. Separating line.

1 Introduction

H. Tverberg [5] proved that for each positive integer kk, there is a minimum integer f(k)f(k) such that for every collection FF of f(k)f(k) or more plane compact convex sets with pairwise disjoint interiors, there is a line that separates one set in FF from a subcollection of FF with at least kk sets. R. Hope and M. Katchalski [3] showed that 3k+1f(k)12k13k+1\leq f(k)\leq 12k-1.

Later E. Rivera-Campo and J. Töröcsik [4] proved that in any collection FF of n5n\geq 5 pairwise disjoint compact convex sets in the plane, there is a pair of sets AA and BB such that any line that separates AA from BB separates either AA or BB from at least n+2830\frac{n+28}{30} sets in FF. In this paper we give a higher lower bound of n18\frac{n}{18} sets in FF for any collection FF of n2n\geq 2 sets.

2 Preliminary results

H. Edelsbrunner et al [1] proved the following theorem, see L. Fejes-Toth [2] for a related result.

Theorem 1.

Any collection of n3n\geq 3 compact, convex and pairwise disjoint sets in the plane may be covered with non-overlapping convex polygons with a total of not more than 6n96n-9 sides. Furthermore no more than 3n63n-6 distinct slopes are required.

We adapt part of the proof given in [1] to obtain the following result.

Lemma 1.

Let 𝒫={P1,P2,,Pn}\mathcal{P}=\{P_{1},P_{2},\ldots,P_{n}\} be a collection of n3n\geq 3 pairwise disjoint convex polygons in the plane. There exists a collection ={R1,R2,,Rn}\mathcal{R}=\{R_{1},R_{2},\ldots,R_{n}\} of pairwise non-overlapping convex polygons such that:

  1. 1.

    For i=1,2,,ni=1,2,\ldots,n, PiP_{i} is contained in RiR_{i}.

  2. 2.

    For i=1,2,,ni=1,2,\ldots,n, each side of RiR_{i} supports a side of PiP_{i}.

  3. 3.

    The total number of sides among polygons R1,R2,,RnR_{1},R_{2},\ldots,R_{n} is at most 9n99n-9.

Proof.

A side ss of a polygon Pi𝒫P_{i}\in\mathcal{P} is reducible with respect to 𝒫\mathcal{P} if the triangle tst_{s} (not containing PiP_{i}), bounded by ss and the lines supporting the sides of PiP_{i} incident to ss, does not intersect the interior of any another polygon PjP_{j}. Equivalently, a side ss of a polygon Pi𝒫P_{i}\in\mathcal{P} is reducible with respect to 𝒫\mathcal{P} if it vanishes before reaching the interior of a polygon PjP_{j} when it is translated in the direction orthogonal to ss and away from PiP_{i} (see Fig. 1).

Refer to caption
Figure 1: Polygon PP with a reducible side ss.

We modify 𝒫\mathcal{P} by growing, one at a time, polygons in 𝒫\mathcal{P} as follows: if some polygon Pi𝒫P_{i}\in\mathcal{P} has a reducible side ss with respect to 𝒫\mathcal{P} substitute PiP_{i} in 𝒫\mathcal{P} with polygon PitsP_{i}\cup t_{s}. Repeat this until no polygon in 𝒫\mathcal{P} has a reducible side. Denote by ={R1,R2,,Rn}\mathcal{R}=\{R_{1},R_{2},\ldots,R_{n}\} the family of polygons thus obtained.

Its is clear that \mathcal{R} satisfies conditions 1) and 2), we claim it also satisfies condition 3. To prove this consider a third family 𝒬={Q1,Q2,,Qn}\mathcal{Q}=\{Q_{1},Q_{2},\ldots,Q_{n}\} of convex polygons obtained from \mathcal{R} by further growing polygons R1,R2,,RnR_{1},R_{2},\ldots,R_{n} in the following way.

A side ss of a polygon RiR_{i} in \mathcal{R} is free with respect to \mathcal{R} if the open polygonal region, not containing RiR_{i}, which is determined by ss and the lines supporting the two sides of RiR_{i} adjacent to ss contains no points of polygons in \mathcal{R}.

Let CC be a circle containing all polygons in \mathcal{R}. Starting with the non free sides of polygons RiR_{i}\in\mathcal{R}, translate one at a time a side ss of a polygon RiR_{i}\in\mathcal{R} (also in the direction orthogonal to ss and away from RiR_{i}) as far as possible without intersecting the interior of another (possibly already grown) polygon RjR_{j} or the exterior of CC (see Fig. 2).

Refer to caption
Figure 2: Side ss of RiR_{i} reaches a polygon RjR_{j} or reaches circle CC.

Stop when no side can be translated as indicated and let 𝒬={Q1,Q2,,Qn}\mathcal{Q}=\{Q_{1},Q_{2},\ldots,\allowbreak Q_{n}\} where, for i=1,2,,ni=1,2,\ldots,n, QiQ_{i} is the polygon obtained from RiR_{i} in this manner. We claim that if CC is chosen large enough, then each polygon in 𝒬\mathcal{Q} contains at most 3 free sides and all non free sides of a polygon QiQ_{i} in 𝒬\mathcal{Q} are in contact with a polygon QjQ_{j} with iji\neq j.

By the choice of \mathcal{R} no side of a polygon RjR_{j} vanishes, therefore for i=1,2,,ni=1,2,\ldots,n the number of sides of RiR_{i} is equal to the number of sides of QiQ_{i}.

We define a plane graph GG with one vertex viv_{i} placed in the interior of each polygon Qi𝒬Q_{i}\in\mathcal{Q} and a vertex vn+1v_{n+1} placed outside circle CC. The edge set of GG consists of 6 types of edges as follows:

a) If a side of a polygon QiQ_{i} and a side of a polygon QjQ_{j} have a segment ll in common choose an arbitrary point pp in ll and draw an edge vivjv_{i}v_{j} through pp as in Fig. 3 (left).

b) If a vertex pp of a polygon QiQ_{i} lies in the interior of a side of a polygon QjQ_{j} and pp is not a vertex of a polygon in 𝒬\mathcal{Q} other than QiQ_{i}, then draw an edge vivjv_{i}v_{j} through pp as in Fig. 3 (center).

c) If two or more polygons Qi1,Qi2,,QirQ_{i_{1}},Q_{i_{2}},\ldots,Q_{i_{r}} share a vertex pp which lies in the interior of a side of a polygon QjQ_{j} and pp is not a point of any polygon QkQ_{k} other than Qi1,Qi2,,QirQ_{i_{1}},Q_{i_{2}},\ldots,Q_{i_{r}}, then there is a neighbourhood NN of pp containing no points of polygons QkQ_{k} with kj,i1,i2,,irk\neq j,i_{1},i_{2},\ldots,i_{r}. In this case draw a cycle with edges vjvi1,vi1vi2,vi2vi3,,vir1vir,virvjv_{j}v_{i_{1}},v_{i_{1}}v_{i_{2}},v_{i_{2}}v_{i_{3}},\ldots,v_{i_{r-1}}v_{i_{r}},v_{i_{r}}v_{j} as in Fig. 3 (right).

Refer to caption
Figure 3: Edge type a) (left), edge type b) (center) and edges type c) (right).

d) If three or more polygons Qi1,Qi2,,QirQ_{i_{1}},Q_{i_{2}},\ldots,Q_{i_{r}} share a vertex pp and pp is not a point of any polygon QkQ_{k} other than Qi1,Qi2,,QirQ_{i_{1}},Q_{i_{2}},\ldots,Q_{i_{r}}, then there is a neighbourhood NN of pp containing no points of any polygon QkQ_{k} with ki1,i2,,irk\neq i_{1},i_{2},\ldots,i_{r}. In this case draw a cycle with edges vi1vi2,vi2vi3,,vir1vir,virv1v_{i_{1}}v_{i_{2}},v_{i_{2}}v_{i_{3}},\ldots,v_{i_{r-1}}v_{i_{r}},v_{i_{r}}v_{1} as in Fig. 4 (left).

e) If two polygons QiQ_{i} and QjQ_{j} share a vertex pp and pp is not a point of any polygon in \mathcal{R} other than QiQ_{i} and QjQ_{j}, then draw an edge vivjv_{i}v_{j} through pp as in Fig. 4 (right).

Refer to caption
Figure 4: Edges type d) (left) and edge type e) (right).

f) For each free side ss of a polygon QiQ_{i} add an edge vivn+1v_{i}v_{n+1} drawn through an interior of point pp in ss.

Notice that edges of type f) are the only possible multiple edges of GG. Therefore GG is a plane multigraph with n+1n+1 vertices and m(3(n+1)6)+(x2+2x3)=(3n3)+(x2+2x3)m\leq(3(n+1)-6)+(x_{2}+2x_{3})=(3n-3)+(x_{2}+2x_{3}) edges, where for i=2,3i=2,3, xix_{i} denotes the number of polygons in 𝒬\mathcal{Q} with ii free sides.

Refer to caption
Figure 5: Graph GG with n=18n=18, x1=6x_{1}=6, x2=2x_{2}=2, me=1m_{e}=1 and mf=10m_{f}=10.

We say that an edge e=vivje=v_{i}v_{j} of GG intersects a side ss of a polygon in 𝒬\mathcal{Q} if ee is drawn through either an interior point of ss or through a vertex of ss. Each non free side among polygons in 𝒬\mathcal{Q} is intersected by at least one edge of GG of type a), b), c), d) or e) and each free side is intersected by an edge of type f). We claim that if an edge vivjv_{i}v_{j} of GG is of type e) then two of the 4 sides intersected by vivjv_{i}v_{j} are intersected by at least another edge of GG.


Proof of Claim. Let vivjv_{i}v_{j} be an edge of GG of type e). Then either two sides s,ts,t of QiQ_{i} (of Qj)Q_{j}) or one side ss of QiQ_{i} and one side tt of QjQ_{j} are such that the lines supporting ss and tt separate QiQ_{i} and QjQ_{j} (see Fig. 6).

Refer to caption
Figure 6: Lines supporting ss and tt separate QiQ_{i} and QjQ_{j}.

Without loss of generality we assume the line supporting a side ss of QiQ_{i} separates QiQ_{i} and QjQ_{j}. Let bb be the other side of QiQ_{i} incident in the common vertex pp of QiQ_{i} and QjQ_{j}. Then side bb must contain a point of a polygon QkQ_{k} other than QiQ_{i} and QjQ_{j}, otherwise bb could be translated away from QiQ_{i} in the direction orthogonal to bb which is not possible by the properties of 𝒬\mathcal{Q} (see Fig. 7). This means that side bb is intersected by an edge of GG other than edge vivjv_{i}v_{j}.

Refer to caption
Figure 7: Side bb could be translated away from QiQ_{i}.

Analogously a side cbc\neq b, incident with tt, is intersected by an edge of GG other than vivjv_{i}v_{j}. \square


Let TT denote the total number of sides among polygons in 𝒬\mathcal{Q}. Also let mm, mem_{e} and mfm_{f} denote the number of edges of GG, the number of edges of GG of type e) and the number of edges of GG of type f), respectively. Since polygons in 𝒬\mathcal{Q} contain at most 3 free edges, mf=x1+2x2+3x3m_{f}=x_{1}+2x_{2}+3x_{3}, where for i=1,2,3i=1,2,3 xix_{i} is the number of polygons in QQ containing ii free edges.

In order to bound the number of sides among polygons in 𝒬\mathcal{Q} we add the number of sides intersected by edges of GG and subtract the number of sides intersected by at least two edges of GG (2 sides for each edge of type e)). Considering that edges of types a), b), c), and d), intersect at most 3 sides; edges of type e) intersect 4 sides and edges of type f) intersect one side, we have:

T\displaystyle T (3(m(me+mf))+4me+mf)2me\displaystyle\leq(3(m-(m_{e}+m_{f}))+4m_{e}+m_{f})-2m_{e}
=3mme2mf\displaystyle=3m-m_{e}-2m_{f}
3m2mf\displaystyle\leq 3m-2m_{f}
3((3n3)+(x2+2x3))2(x1+2x2+3x3)\displaystyle\leq 3((3n-3)+(x_{2}+2x_{3}))-2(x_{1}+2x_{2}+3x_{3})
=9n92x1x2\displaystyle=9n-9-2x_{1}-x_{2}
9n9.\displaystyle\leq 9n-9.

Consider the following familly of plane geometric graphs GtG_{t}, t=1,2,t=1,2,\ldots given by Edelsbrunner et al [1] where it is used to show that the bounds in the number of sides and slopes on Theorem 1 are tight.

G1G_{1} is the graph with 77 vertices shown on Fig. 8 (left); for t1t\geq 1, Gt+1G_{t+1} is the graph obtained from GtG_{t} as in Fig. 8 (right). For k1k\geq 1 the graph GkG_{k} is a plane geometric graph with n=3kn=3k internal faces. The three inner-most faces are quadrilaterals while all other faces are hexagons.

Refer to caption
Figure 8: Left: Graph G1G_{1}. Right: Graph Gt+1G_{t+1} obtained from graph GtG_{t} (placed upside down inside the dotted triangle) .

Place an hexagon inside each of the three inner-most faces of GkG_{k} as in Fig. 9 (left), an octagon inside each outer-most internal face of GkG_{k} as in Fig. 9 (right) and a nonagon inside each other internal face of GkG_{k} as in Fig. 10.

Refer to caption
Figure 9: Hexagon inside inner-most face of GkG_{k}. Octagon inside outer-most internal face of GkG_{k}.
Refer to caption
Figure 10: Nonagon inside internal face of GkG_{k}

Since each face of GkG_{k} has an even number of sides, the placement of the polygons described above may be done in such a way that for each internal edge ee of GkG_{k} a side ses_{e} of the polygon RiR_{i} lying on one side of ee is parallel (and closed enough) to ee while a vertex vev_{e} of the polygon RjR_{j} lying on the other side of ee lies on ee as in Fig. 11.

Refer to caption
Figure 11: Polygons on both sides of an internal edge of GkG_{k}.

Let ={R1,R2,Rn}\mathcal{R}=\{R_{1},R_{2},\ldots R_{n}\} be the set of polygons described above. The total number of sides among polygons in \mathcal{R} is

m=3(6)+3(8)+(n6)(9)=9n12m=3(6)+3(8)+(n-6)(9)=9n-12

Each side among polygons in \mathcal{R} is relevant with respect to \mathcal{R}, therefore our bound on the number of sides in Lemma 1 is almost tight.

3 Main Results

In this section we present our main results.

Lemma 2.

Let 𝒫={P1,P2,,Pn}\mathcal{P}=\{P_{1},P_{2},\ldots,P_{n}\} be a collection of n3n\geq 3 pairwise disjoint convex polygons in the plane. There is a side ss of a polygon PiP_{i} such that the line supporting ss separates PiP_{i} from a subcollection \mathcal{F} of 𝒫\mathcal{P} with at least n18\frac{n}{18} polygons.

Proof.

Let ={R1,R2,Rn}\mathcal{R}=\{R_{1},R_{2},\ldots R_{n}\} be a collection of pairwise disjoint convex polygons satisfying conditions 1), 2) and 3) in Lemma 1 and let ={l1,l2,,lm}\mathcal{L}=\{l_{1},l_{2},\ldots,l_{m}\} be the (multi)set of lines supporting the sides of each polygon in \mathcal{R}. We include cc copies of a line ll if ll supports sides of cc polygons in \mathcal{R}. Therefore we can associate a unique polygon Ri(k)R_{i(k)} to each line lkl_{k}\in\mathcal{L} such that a side of Ri(k)R_{i(k)} is supported by lkl_{k}. For k=1,2,,mk=1,2,\ldots,m let HkH_{k}^{-} be the closed halfplane determined by lkl_{k} that does not contain Ri(k)R_{i(k)}.

Define a bipartite graph FF with a vertex vjv_{j} for each polygon RjR_{j}\in\mathcal{R} and a vertex wkw_{k} for each line lkl_{k}\in\mathcal{L}. Graph FF has an edge vjwkv_{j}w_{k} if polygon RjR_{j} is contained in HkH_{k}^{-}.

For each pair of polygons Ri,RjR_{i},R_{j} there is at least one side ss of one of them such that the line supporting ss separates RiR_{i} from RjR_{j}. Therefore graph FF has at least one edge for each pair {i,j}\{i,j\} with 1i<jn1\leq i<j\leq n. This implies that there is a vertex wkw_{k} whose degree in FF is at least (n2)/m(n2)/(9n9)=n/18\binom{n}{2}/m\geq\binom{n}{2}/(9n-9)=n/18.

This means that the line lkl_{k} separates polygon Ri(k)R_{i(k)} from at least n/18n/18 polygons in \mathcal{R}. By Property 2) in Lemma 1, line lkl_{k} supports a side of polygon Pi(k)P_{i(k)}.

Theorem 2.

In any collection 𝒞\mathcal{C} of n2n\geq 2 pairwise disjoint compact convex sets in the plane, there is a pair of sets AA and BB such that any line that separates AA from BB separates either AA or BB from a subcollection \mathcal{F} of 𝒞\mathcal{C} with at least n18\frac{n}{18} sets.

Proof.

Let n2n\geq 2 be an integer and 𝒞={C1,C2,,Cn}\mathcal{C}=\{C_{1},C_{2},\ldots,C_{n}\} be a collection of pairwise disjoint compact convex sets in the plane and let TT be a triangle containing all sets in 𝒞\mathcal{C} in its interior.

For any line ll let tlt_{l}^{-} and tl+t_{l}^{+} be the number of sets in 𝒞\mathcal{C} lying to the left and to the right of ll, respectively; for a horizontal line ll, tlt_{l}^{-} and tl+t_{l}^{+} are the number of sets in 𝒞\mathcal{C} lying above and bellow ll, respectively. For 1i<jn1\leq i<j\leq n let lijl_{ij} be a line separating CiC_{i} from CjC_{j} for which max{tlij,tlij+}max\{t_{l_{ij}}^{-},t_{l_{ij}}^{+}\} is as small as posible.

Each line lijl_{ij} determines two closed halfplanes HijH_{ij} and HjiH_{ji} containing CiC_{i} and CjC_{j}, respectively. For i=1,2,,ni=1,2,\ldots,n, let Pi=T(jinHij)P_{i}=T\cap({\bigcap}_{j\neq i}^{n}H_{ij}). Then For i=1,2,,ni=1,2,\ldots,n, PiP_{i} is a convex polygon that contains set CiC_{i} such that each side is supported by a line lijl_{ij} or by a side of TT.

Let 𝒫={P1,P2,,Pn}\mathcal{P}=\{P_{1},P_{2},\ldots,P_{n}\}. By Lemma 2, there is a side ss of a polygon PiP_{i} such that the line l(s)l(s) supporting ss separates PiP_{i} from a subcollection of 𝒫\mathcal{P} with at least n18\frac{n}{18} polygons.

Since no side of TT separates polygons in 𝒫\mathcal{P}, l(s)l(s) is of the form lijl_{ij}. By the choice of lijl_{ij} each line that separates sets CiC_{i} from CjC_{j} separates either CiC_{i} or CjC_{j} from at least n18\frac{n}{18} sets in 𝒞\mathcal{C}. ∎

References

  • [1] Herbert Edelsbrunner, Arch D. Robison and Xiao Jun Shen. Covering convex sets with non-overlapping polygons, Discrete Mathematics 81: 153 – 164, 1990.
  • [2] László Fejes Tóth. Illumination of convex discs, Acta Mathematica Academiae Scientiarum Hungaricae 29(3 - 4): 355 – 360, 1997.
  • [3] Rafael Hope and Meir Katchalski. Separating plane convex sets, Mathematica Scandinavica 66 (1): 44–46, 1990.
  • [4] Eduardo Rivera-Campo and Jeno Töröcsik. On separation of plane convex sets, European Journal of Combinatorics 14: 113 – 116, 1993.
  • [5] Helge Tverberg. A separation property of plane convex sets, Mathematica Scandinavica 45: 255–260, 1979.