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

New Lower Bounds for Tverberg Partitions with Tolerance in the Plane

Sergey Bereg Department of Computer Science, University of Texas at Dallas. This research is supported in part by NSF award CCF-1718994.    Mohammadreza Haghpanah
Abstract

Let PP be a set nn points in a dd-dimensional space. Tverberg’s theorem says that, if nn is at least (k1)(d+1)+1(k-1)(d+1)+1, then PP can be partitioned into kk sets whose convex hulls intersect. Partitions with this property are called Tverberg partitions. A partition has tolerance tt if the partition remains a Tverberg partition after removal of any set of tt points from PP. Tolerant Tverberg partitions exist in any dimension provided that nn is sufficiently large. Let N(d,k,t)N(d,k,t) be the smallest value of nn such that tolerant Tverberg partitions exist for any set of nn points in d\mathbb{R}^{d}. Only few exact values of N(d,k,t)N(d,k,t) are known.

In this paper we establish a new tight bound for N(2,2,2)N(2,2,2). We also prove many new lower bounds on N(2,k,t)N(2,k,t) for k2k\geq 2 and t1t\geq 1.

1 Introduction

The classical Tverberg’s theorem [18] states that a sufficiently large set of points in d\mathbb{R}^{d} can be partitioned into subsets such that their convex hulls intersect. For the history and recent advances around Tverberg’s theorem, we refer the reader to [3, 4, 6, 13].

Theorem 1 (Tverberg [18]).

Any set PP of at least (k1)(d+1)+1(k-1)(d+1)+1 points in d\mathbb{R}^{d} can be partitioned into kk sets P1,P2,,PkP_{1},P_{2},\dots,P_{k} whose convex hulls intersect, i.e.

i=1k𝖼𝗈𝗇𝗏(Pi).\bigcap_{i=1}^{k}\mathsf{conv}(P_{i})\neq\emptyset. (1)

In 1972, David Larman proved the first tolerant version of Tverberg’s theorem: any set of 2d+32d+3 points in d\mathbb{R}^{d} can be partitioned into two sets AA and BB such that their convex hulls intersect with tolerance one, i.e., for any point xdx\in\mathbb{R}^{d}, 𝖼𝗈𝗇𝗏(A{x})𝖼𝗈𝗇𝗏(B{x})\mathsf{conv}(A\setminus\{x\})\cap\mathsf{conv}(B\setminus\{x\})\neq\emptyset. García-Colín [8] generalized it for any tolerance. Soberón and Strausz [17] generalized it further for partitions into many sets. The general problem can be stated as follows.

Problem (Tverberg partitions with tolerance). Given positive integers d,k,td,k,t, find the smallest positive integer N(d,k,t)N(d,k,t) such that any set PP of N(d,k,t)N(d,k,t) points in d\mathbb{R}^{d} can be partitioned into kk subsets P1,P2,,PkP_{1},P_{2},\dots,P_{k} such that for any set YPY\subset P of at most tt points

i=1k𝖼𝗈𝗇𝗏(PiY).\bigcap_{i=1}^{k}\mathsf{conv}(P_{i}\setminus Y)\neq\emptyset. (2)

We call a partition of PP into kk subsets P1,P2,,PkP_{1},P_{2},\dots,P_{k} tt-tolerant if condition (2) holds for any set YY of at most tt points. Several upper bounds for N(d,k,t)N(d,k,t) are known [10, 12, 14, 17]. Some lower bounds for N(d,k,t)N(d,k,t) are known [9, 2, 15].

Theorem 2 (Ramírez-Alfonsín [2]).

For any d4d\geq 4

N(d,2,1)5d3+3.N(d,2,1)\geq\left\lceil\frac{5d}{3}\right\rceil+3. (3)
Theorem 3 (García-Colín and Larman [9]).
N(d,2,t)2d+t+1.N(d,2,t)\geq 2d+t+1. (4)
Theorem 4 (Soberón [15]).
N(d,k,t)k(t+d/2+1).N(d,k,t)\geq k(t+\lfloor d/2\rfloor+1). (5)

In contrast to Tverberg’s theorem, most bounds are not known to be tight. The only known tight bounds are the following. Larman [12] proved N(d,2,1)=2d+3N(d,2,1)=2d+3 for d3d\leq 3. Forge, Las Vergnas and Schuchert [7] proved N(4,2,1)=11N(4,2,1)=11. For all k2,t1k\geq 2,t\geq 1 and dimension one, N(1,k,t)=k(t+2)1N(1,k,t)=k(t+2)-1 by Mulzer and Stein [14]. In this paper we establish a new tight bound for N(2,2,2)N(2,2,2).

Theorem 5.

N(2,2,2)=10N(2,2,2)=10.

We also improve the bound (5) for the plane.

Theorem 6.

Let cc be a positive integer. For any integers k2ck\geq 2c and tct\geq c

N(2,k,t)k(t+2)+c.N(2,k,t)\geq k(t+2)+c. (6)

Remark. The bound of Theorem 6 can be stated without parameter cc

N(2,k,t)k(t+2)+min(t,k/2).N(2,k,t)\geq k(t+2)+\min(t,\lfloor k/2\rfloor). (7)

If d=2d=2 and k=2k=2 then Equation (6) provides a lower bound N(2,2,t)2(t+2)+1N(2,2,t)\geq 2(t+2)+1. We further improve the bound for k=2k=2 and t3t\geq 3.

Theorem 7.

For any t3t\geq 3, N(2,2,t)2t+6N(2,2,t)\geq 2t+6.

Several upper bounds for N(d,k,t)N(d,k,t) are known [8, 10, 16]. For example, Soberón [16] proved N(d,k,t)=kt+O(t)N(d,k,t)=kt+O(\sqrt{t}) for fixed kk and dd. Therefore, it is interesting to find lower and upper bounds for P(d,k,t)=N(d,k,t)ktP(d,k,t)=N(d,k,t)-kt. Theorems 5,6, and 7 provide new lower bounds for P(d,k,t)P(d,k,t) for d=2d=2.

Remark. Recently, we improved some lower bounds using computer pograms [5]. For example, N(2,2,t)2t+7N(2,2,t)\geq 2t+7 for 5t105\leq t\leq 10.

The rest of the paper is organized as follows. Section 2 introduces some observations and lemmas that will be used in our proofs. Theorems 6 and 7 are proven in Section 3. Theorem 5 is proven in Section 4. Finally, we discuss the results and open problems in Section 5.

2 Preliminaries

We start with a simple observation.

Observation 8.

Every part in a tt-tolerant partition has size at least t+1t+1.

The following lemma is simple, yet very useful in proving that a partition is not tt-tolerant.

Lemma 9.

Let P1,P2,,PkP_{1},P_{2},\dots,P_{k} be a partition of a finite set PP in d\mathbb{R}^{d} and let XX be a subset of a set PiP_{i} such that

𝖼𝗈𝗇𝗏(X)𝖼𝗈𝗇𝗏(PX)=\mathsf{conv}(X)\cap\mathsf{conv}(P\setminus X)=\emptyset (8)

and |X|=m,|Pi|=t+m|X|=m,|P_{i}|=t+m. Then the partition of PP is not tt-tolerant.

Proof.

Let Y=PiXY=P_{i}\setminus X. Then |Y|=t|Y|=t. Therefore

j=1k𝖼𝗈𝗇𝗏(PjY)=\bigcap_{j=1}^{k}\mathsf{conv}(P_{j}\setminus Y)=\emptyset (9)

by Equation 8 since one of the sets in the intersection is XX and every other set is a subset of PXP\setminus X. ∎

Two special cases of Lemma 9.

Lemma 10.

A partition P1,P2,,PkP_{1},P_{2},\dots,P_{k} of a finite set PP in d\mathbb{R}^{d} is not tt-tolerant if one of the following conditions holds.

  1. (1)

    A set PjP_{j} contains a vertex of 𝖼𝗈𝗇𝗏(P)\mathsf{conv}(P) and |Pj|=t+1|P_{j}|=t+1.

  2. (2)

    A set PjP_{j} contains two points pi,pi+1p_{i},p_{i}+1 such that pi,pi+1p_{i},p_{i}+1 is an edge of 𝖼𝗈𝗇𝗏(P)\mathsf{conv}(P) and |Pj|=t+2|P_{j}|=t+2.

Proof.

The first claim follows from Lemma 9 by setting X={pi}X=\{p_{i}\} where piPjp_{i}\in P_{j} is a vertex of 𝖼𝗈𝗇𝗏(P)\mathsf{conv}(P). The second claim follows from Lemma 9 by setting X={pi,pi+1}X=\{p_{i},p_{i+1}\}. ∎

3 Proofs of Theorems 6 and 7

First, we prove Theorem 6.

Proof of Theorem 6.

Construct a set PP of k(t+2)+c1k(t+2)+c-1 points in the plane as follows. Let QQ be a regular (k(t+2)1)(k(t+2)-1)-gon with center at the origin OO. Place k(t+2)1k(t+2)-1 points q1,,qk(t+2)1q_{1},\dots,q_{k(t+2)-1} at the vertices of QQ and cc points p1,,pcp_{1},\dots,p_{c} close to the origin. More specifically, we assume that |p1|,,|pc|<d|p_{1}|,\dots,|p_{c}|<d where dd is the distance from the origin to the line p1pkp_{1}p_{k}. Let V={q1,,qk(t+2)1}V=\{q_{1},\dots,q_{k(t+2)-1}\}. We assume that the vertices of QQ are sorted in clockwise order, see Figure 1.

Refer to caption
Figure 1: A point set for Theorem 6.

We show that any partition of PP into P1,P2,,PkP_{1},P_{2},\dots,P_{k} is not tt-tolerant. By Observation 8, we assume |Pi|t+1|P_{i}|\geq t+1 for all ii. If |Pi|=t+1|P_{i}|=t+1 for some ii, then PiP_{i} contains a point qjq_{j} since t+1>ct+1>c. The partition is not tt-tolerant by Lemma 10(1).

It remains to consider the case where all sets in the partition have size at least t+2t+2. At most c1c-1 sets in the partition may have size t+3\geq t+3. Since k2ck\geq 2c there exists a set PiP_{i} of size t+2t+2 such that PiVP_{i}\subset V. Since |V|=k(t+2)1|V|=k(t+2)-1, there exist two points qaq_{a} and qa+jq_{a+j} with jk1j\leq k-1 such that the vertices between qaq_{a} and qa+jq_{a+j} are in QPiQ\setminus P_{i} (we assume here that qx=qyq_{x}=q_{y} if xy(modk(t+2)1)x\equiv y\pmod{k(t+2)-1}). There is a set Pm,miP_{m},m\neq i such that none of these vertices are in PmP_{m}. Therefore the partition is not tt-tolerant since 𝖼𝗈𝗇𝗏{qa,qb}𝖼𝗈𝗇𝗏(Pm)=\mathsf{conv}\{q_{a},q_{b}\}\cap\mathsf{conv}(P_{m})=\emptyset. Note that the distance from the origin to any point in {p1,,pc}\{p_{1},\dots,p_{c}\} is smaller than the distance from the origin to the line qaqbq_{a}q_{b}. ∎

Next we prove Theorem 7.

Refer to caption
Figure 2: A point set for Theorem 7.
Proof of Theorem 7.

Construct a set PP of 2t+52t+5 points in the plane as follows. First, place 2t+22t+2 points in convex position P={p1,p2,,p2t+2}P^{\prime}=\{p_{1},p_{2},\dots,p_{2t+2}\} where points are in clockwise order. Then place three points Q={q1,q2,q3}Q=\{q_{1},q_{2},q_{3}\} inside the convex hull of PP^{\prime} such that point qi,i=1,2,3q_{i},i=1,2,3 is close to edge p2i1p2ip_{2i-1}p_{2i} (it is required that qiq_{i} is inside triangles p2i2p2i1p2i\bigtriangleup p_{2i-2}p_{2i-1}p_{2i} and p2i1p2ip2i+1\bigtriangleup p_{2i-1}p_{2i}p_{2i+1}), see Figure 2. We show that any partition of PP into P1,P2P_{1},P_{2} is not tt-tolerant.

Without loss of generality, we assume |P1||P2||P_{1}|\leq|P_{2}|. Therefore, |P1|t+2|P_{1}|\leq t+2. By Observation 8, we assume |P1|t+1|P_{1}|\geq t+1. Suppose if |P1|=t+1|P_{1}|=t+1. Since t3t\geq 3, P1P_{1} contains a point of PP^{\prime}. The partition is not tt-tolerant by Lemma 10(1).

It remains to consider the case where |P1|=t+2|P_{1}|=t+2. If set P1P_{1} does not contain a point qiq_{i}, then P1PP_{1}\subset P^{\prime} and |P1||P|/2|P_{1}|\geq|P^{\prime}|/2. Then P1P_{1} contains two consecutive points of PP^{\prime}. The partition is not tt-tolerant by Lemma 10(2). Therefore P1QP_{1}\cap Q\neq\emptyset.

Suppose qiP1q_{i}\in P_{1} for some ii. If p2i1P1p_{2i-1}\in P_{1} or p2iP1p_{2i}\in P_{1} then the partition is not tt-tolerant using X={qi,p2i1}X=\{q_{i},p_{2i-1}\} or X={qi,p2i}X=\{q_{i},p_{2i}\}, respectively, and Lemma 9. Therefore, we assume that both points p2i1p_{2i-1} and p2ip_{2i} are in P2P_{2}.

Let m=|P1{q1,q2,q3}|m=|P_{1}\cap\{q_{1},q_{2},q_{3}\}|. Then m{1,2,3}m\in\{1,2,3\}. By the previous argument, set P2P_{2} contains at least mm pairs of consecutive points of PP^{\prime} (the pair p2i1,p2ip_{2i-1},p_{2i} for each point qiP1q_{i}\in P_{1}). Note that |PP1|=t+2m|P^{\prime}\cap P_{1}|=t+2-m and |PP2|=t+m|P^{\prime}\cap P_{2}|=t+m. We will show that set P1P_{1} contains a pair of consecutive points of PP^{\prime}. Then, the partition is not tt-tolerant by Lemma 10(2).

If m=1m=1, then |PP1|=|PP2|=t+1|P^{\prime}\cap P_{1}|=|P^{\prime}\cap P_{2}|=t+1 and set P2P_{2} contains a pair of consecutive points of PP^{\prime}. Then set P1P_{1} contains a pair of consecutive points of PP^{\prime}.

Suppose m=2m=2, say qi,qjP1q_{i},q_{j}\in P_{1} where i<ji<j. Then set P2P_{2} contains p2i1,p2i,p2j1p_{2i-1},p_{2i},p_{2j-1}, and p2jp_{2j}. There are two intervals p2i+1,,p2j2p_{2i+1},\dots,p_{2j-2} and p2j+1,,p2i2p_{2j+1},\dots,p_{2i-2} in PP^{\prime} containing points of P1P_{1}. Note that each interval contains an even number of points. If set P1P_{1} does not contain a pair of consecutive points of PP^{\prime}, then set P1P_{1} contains at most half of points in each interval. Then |PP1|<t|P^{\prime}\cap P_{1}|<t. This contradiction implies that set P1P_{1} contains a pair of consecutive points of PP^{\prime}.

Finally, if m=3m=3, then there exists only one interval p7,p8,,p2t+2p_{7},p_{8},\dots,p_{2t+2} and set P1P_{1} must contain a pair of consecutive points of PP^{\prime}; otherwise |PP1|<t+1|P^{\prime}\cap P_{1}|<t+1. The theorem follows. ∎

4 Proof of Theorem 5

We found a special configuration of nine points to prove a lower bound for Theorem 5. For this, we checked point configurations of all order types for n=9n=9. Order types, introduced by Goodman and Pollack [11], are useful for characterizing the combinatorial properties of point configurations. An order type of a set of points p1,p2,,pnp_{1},p_{2},\dots,p_{n} in general position in the plane is defined using a mapping that assigns each triple of integers i,j,ki,j,k with 1i<j<kn1\leq i<j<k\leq n the orientation (either clockwise or counter-clockwise) of the triple pi,pj,pkp_{i},p_{j},p_{k}. Two point sets P1P_{1} and P2P_{2} have the same order type if there is a bijection π\pi from P1P_{1} to P2P_{2} preserving this map, i.e., for any three distinct points a,b,ca,b,c in P1P_{1}, the orientation of a,b,ca,b,c and the orientation of π(a),π(b),π(c)\pi(a),\pi(b),\pi(c) are the same. Aichholzer et al.  [1] established that there are 158,817 order types for n=9n=9. We developed a program for testing each of these point sets whether it admits a 2-tolerant partition into two sets. The proof of the tolerance of a configuration found by the program turns out to be not simple even using the tools from Section 2.

Refer to caption
Figure 3: A set SS of nine points for Lemma 11.
Lemma 11.

There exists a set of nine points in the plane that does not admit a 2-tolerant partition into two sets.

Refer to caption
Figure 4: Case 2a. The points of AA are red and the points of BB are black. The shaded polygons are the convex hulls of ACA-C and BCB-C.
Proof.

Consider a set SS of nine points shown in Fig.3 where S=PQ,P={p1,,p6}S=P\cup Q,P=\{p_{1},\dots,p_{6}\} and Q={q1,q2,q3}Q=\{q_{1},q_{2},q_{3}\}. We will prove that any partition of S=ABS=A\cup B is not 2-tolerant, i.e. there exists a set C={a,b}SC=\{a,b\}\subseteq S such that

𝖼𝗈𝗇𝗏(AC)𝖼𝗈𝗇𝗏(BC)=.\mathsf{conv}(A\setminus C)\cap\mathsf{conv}(B\setminus C)=\emptyset.

We can assume that |A|<|B||A|<|B|. By Observation 8, we assume |A|3|A|\geq 3. Suppose that |A|=3|A|=3. If A=QA=Q, then take C={p1,p2}C=\{p_{1},p_{2}\}. If AQA\neq Q, then the partition of SS is not 2-tolerant by Lemma 10(1).

It remains to analyze partitions with |A|=4|A|=4. By Lemma 10(2), we assume that AA does not contain two consecutive points pi,pi+1p_{i},p_{i+1} (assuming p7=p1p_{7}=p_{1}). For example, this implies that AA contains at least one point of QQ. Consider the following cases depending on |AP||A\cap P|.

Case 1. Set AA contains exactly one point of PP, say pip_{i}. Then QAQ\subset A. One of the sets {p1,p2,p3}\{p_{1},p_{2},p_{3}\}, {p3,p4,p5}\{p_{3},p_{4},p_{5}\}, {p4,p5,p6}\{p_{4},p_{5},p_{6}\}, say set XX, does not contain pip_{i}. The partition is not 2-tolerant by Lemma 9 using XX.

Refer to caption
Figure 5: (a) Case 2b. (b) Case 2c.

Case 2. Set AA contains exactly two points of PP, i.e. A={pi,pj,qc,qd}A=\{p_{i},p_{j},q_{c},q_{d}\} for some 1i<j61\leq i<j\leq 6 and 1c<d31\leq c<d\leq 3.

Case 2a. Suppose i=1i=1. By our assumption j{2,6}j\notin\{2,6\}. We can assume that j3j\neq 3 using X={p4,p5,p6}X=\{p_{4},p_{5},p_{6}\} and Lemma 9. So, j{4,5}j\in\{4,5\}.

We also can assume that q1Bq_{1}\in B using X={p1,q1}X=\{p_{1},q_{1}\} and Lemma 9. Therefore {q2,q3}A\{q_{2},q_{3}\}\subset A. If j=4j=4 (so A={p1,p4,q2,q3}A=\{p_{1},p_{4},q_{2},q_{3}\}) then take C={p1,p3}C=\{p_{1},p_{3}\}, see Fig.4(a). If j=5j=5 (so A={p1,p5,q2,q3}A=\{p_{1},p_{5},q_{2},q_{3}\}) then take C={p6,q2}C=\{p_{6},q_{2}\}, see Fig.4(b).

Refer to caption
Figure 6: Case 3a.

Case 2b. Suppose i=2i=2. Then j4j\geq 4. We can assume j4j\neq 4 using X={p1,p5,p6}X=\{p_{1},p_{5},p_{6}\} and Lemma 9. We can assume j6j\neq 6 using X={p3,p4,p5}X=\{p_{3},p_{4},p_{5}\} and Lemma 9. Thus, j=5j=5. We can assume q1Aq_{1}\in A using X={p1,p6,q1}X=\{p_{1},p_{6},q_{1}\} and Lemma 9.

If q3Aq_{3}\in A then A={p2,p5,q1,q3}A=\{p_{2},p_{5},q_{1},q_{3}\} we can use X={p3,p4,q2}X=\{p_{3},p_{4},q_{2}\} and Lemma 9. If q2Aq_{2}\in A then A={p2,p5,q1,q2}A=\{p_{2},p_{5},q_{1},q_{2}\} and we can use C={p1,p5}C=\{p_{1},p_{5}\}, see Fig.5(a).

Case 2c. Suppose i,j{3,4,5,6}i,j\in\{3,4,5,6\}. We can assume q1,q2Aq_{1},q_{2}\in A using X={p1,p2,x}X=\{p_{1},p_{2},x\} for x{q1,q2}x\in\{q_{1},q_{2}\} and Lemma 9. Since pip_{i} and pi+1p_{i+1} cannot be in AA together, there are three options. If p3,p5Ap_{3},p_{5}\in A, take C={p4,q1}C=\{p_{4},q_{1}\}, see Fig.5(b). If p4,p6Ap_{4},p_{6}\in A then we can use X={p1,p2,p3}X=\{p_{1},p_{2},p_{3}\} and Lemma 9. If p3,p6Ap_{3},p_{6}\in A then we can use X={p3,q2}X=\{p_{3},q_{2}\} and Lemma 9.

Case 3. Set AA contains exactly three points of PP, i.e. {p1,p3,p5}A\{p_{1},p_{3},p_{5}\}\subset A or {p2,p4,p6}A\{p_{2},p_{4},p_{6}\}\subset A.

Case 3a. Suppose {p1,p3,p5}A\{p_{1},p_{3},p_{5}\}\subset A. If q1Aq_{1}\in A then we can use X={p1,q1}X=\{p_{1},q_{1}\} and Lemma 9. If q2Aq_{2}\in A, take C={p2,p5}C=\{p_{2},p_{5}\}, see Fig.6(a). If q3Aq_{3}\in A, take C={p3,p6}C=\{p_{3},p_{6}\}, see Fig.6(b).

Case 3b. Suppose {p2,p4,p6}A\{p_{2},p_{4},p_{6}\}\subset A. If q1Aq_{1}\in A, take C={p1,p4}C=\{p_{1},p_{4}\}, see Fig.7(a). If q2Aq_{2}\in A, take C={p3,p6}C=\{p_{3},p_{6}\}, see Fig.7(b). If q3Aq_{3}\in A, take C={p2,p5}C=\{p_{2},p_{5}\}, see Fig.7(c). The lemma follows. ∎

Refer to caption
Figure 7: Case 3b.
Corollary 12.

N(2,2,2)10N(2,2,2)\geq 10.

Soberón and Strausz [17] proved an upper bound N(d,k,t)(k1)(t+1)(d+1)+1N(d,k,t)\leq(k-1)(t+1)(d+1)+1 which implies N(2,2,2)10N(2,2,2)\leq 10. Therefore N(2,2,2)=10N(2,2,2)=10.

5 Further Discussion and Open Problems

Improving the bounds for Tverberg partitions with tolerance is an interesting problem. There are many triples (d,k,t)(d,k,t) with a gap between known lower and upper bounds. Finding sharp bounds for N(d,k,t)N(d,k,t) is a challenging problem. For example, Larman [12] conjectured that N(d,2,1)=2d+3N(d,2,1)=2d+3 for any d1d\geq 1. The conjecture is open for d5d\geq 5. For dimension d=5d=5, the best known upper bound is N(5,2,1)13N(5,2,1)\leq 13 by Larman [12] and the best known lower bound is N(5,2,1)12N(5,2,1)\geq 12 by García-Colín and Larman [9].

Another possibility for a new sharp bound is for d=2,k=2d=2,k=2 and tolerance t=3t=3. The lower bound is N(2,2,3)12N(2,2,3)\geq 12 by Theorem 7. The best upper bound is N(2,2,3)13N(2,2,3)\leq 13 by Soberón and Strausz [17]. Based on our computer experiments, we conjecture that N(2,2,3)=12N(2,2,3)=12. One way to improve the upper bound for d=2,k=2d=2,k=2 and t=3t=3 is to study the following core lemma in the proof of the upper bound for N(d,k,t)N(d,k,t) [17].

Lemma 13.

Let p1p\geq 1 and r0r\geq 0 be integers, SpS\subset\mathbb{R}^{p} be a set of n=p(r+1)+1n=p(r+1)+1 points a1,a2,,ana_{1},a_{2},\dots,a_{n} and GG be a group with |G|p|G|\leq p. If there is an action of GG in a set SpS^{\prime}\subset\mathbb{R}^{p} which is compatible with SSS\subset S^{\prime}, then for each aia_{i} there is a giGg_{i}\in G such that the set {g1a1,g2a2,,gnan}\{g_{1}a_{1},g_{2}a_{2},\dots,g_{n}a_{n}\} captures the origin with tolerance rr.

For d=2,k=2d=2,k=2 and tolerance 3, Lemma 13 uses n=13n=13 points since p=(k1)(d+1)=3p=(k-1)(d+1)=3. Is this bound for nn sharp?

5.1 Order types and tolerant Tverberg partitions

Our proof of Theorem 5 uses the set of nine points shown in Figure 3. It was computed by a program that we developed for testing the tolerance of a given point set. We have applied this program to 158,817 order types for n=9n=9 which are provided by Aichholzer et al.  [1]. Our program found that 155,115 point sets admit a 2-tolerant partition into two sets. Only 3,702 point sets do not admit a 2-tolerant partition into two sets and therefore, can be used as examples for proving the lower bound N(2,2,2)10N(2,2,2)\geq 10.

We analyzed these 3,702 order types and found that their convex hulls have sizes between three and six. We count the number of different order types for each size of the convex hull, see Table 1. Figures 8, 9, and 10 show some order types with 5, 4, and 3 points on the convex hull, respectively. It follows from our computer experiments that any set of nine points in the plane with at least seven points on the convex hull admits a 2-tolerant partition into two sets. It is interesting to investigate how the size of convex hull of a point set affects lower bounds of N(d,k,t)N(d,k,t).

hh h=3h=3 h=4h=4 h=5h=5 h=6h=6 h=7h=7 h=8h=8 h=9h=9
NhN_{h} 1303 1554 769 76 0 0 0
Table 1: 3,702 order types of nine points that do not admit a 2-tolerant partition into two sets. The point sets are partitioned using the size of the convex hull hh. The size of each group is NhN_{h}.
Order type 1874
Order type 2163
Figure 8: Two point sets of nine points each with five points on the convex hull. The point sets have different order types. Both point sets do not admit a 2-tolerant partition into two sets.
Order type 94078
Order type 8005
Figure 9: Two point sets of nine points each with four points on the convex hull. The point sets have different order types. Both point sets do not admit a 2-tolerant partition into two sets.
Order type 158483
Order type 103224
Figure 10: Two point sets of nine points each with three points on the convex hull. The point sets have different order types. Both point sets do not admit a 2-tolerant partition into two sets.

References

  • [1] O. Aichholzer, F. Aurenhammer, and H. Krasser. Enumerating order types for small point sets with applications. Order, 19:265–281, 2002.
  • [2] J. L. R. Alfonsín. Lawrence oriented matroids and a problem of mcmullen on projective equivalences of polytopes. Eur. J. Comb., 22(5):723–731, 2001.
  • [3] I. Bárány, P. V. M. Blagojevic, and G. M. Ziegler. Tverberg’s theorem at 50: extensions and counterexamples. Notices Amer. Math. Soc., 63(7):732–739.
  • [4] I. Bárány and P. Soberón. Tverberg’s theorem is 50 years old: a survey. Bulletin of the American Mathematical Society, 55(4):459–492, 2018.
  • [5] S. Bereg and M. Haghpanah. Algorithms for Radon partitions with tolerance. In 6th Internat. Conf. on Algorithms and Discrete Applied Mathematics, pages 476–487, 2020.
  • [6] J. De Loera, X. Goaoc, F. Meunier, and N. Mustafa. The discrete yet ubiquitous theorems of carathéodory, helly, sperner, tucker, and tverberg. Bulletin of the American Mathematical Society, 56(3):415–511, 2019.
  • [7] D. Forge, M. Las Vergnas, and P. Schuchert. 10 points in dimension 4 not projectively equivalent to the vertices of a convex polytope. European Journal of Combinatorics, 22(5):705–708, 2001.
  • [8] N. García-Colín. Applying Tverberg type theorems to geometric problems. PhD thesis, University College of London, 2007.
  • [9] N. García-Colín and D. G. Larman. Projective equivalences of kk-neighbourly polytopes. Graphs and Combinatorics, 31(5):1403–1422, 2015.
  • [10] N. García-Colín, M. Raggi, and E. Roldán-Pensado. A note on the tolerant Tverberg theorem. Discrete & Computational Geometry, 58(3):746–754, 2017.
  • [11] J. E. Goodman and R. Pollack. Multidimensional sorting. SIAM J. Comput., 12(3):484–507, Aug. 1983.
  • [12] D. G. Larman. On sets projectively equivalent to the vertices of a convex polytope. Bulletin of the London Mathematical Society, 4(1):6–12, mar 1972.
  • [13] J. Matoušek. Lectures on discrete geometry, volume 212. Springer, 2002.
  • [14] W. Mulzer and Y. Stein. Algorithms for tolerant Tverberg partitions. Int. J. Comput. Geometry Appl., 24(4):261–274, 2014.
  • [15] P. Soberón. Equal coefficients and tolerance in coloured Tverberg partitions. Combinatorica, 35(2):235–252, 2015.
  • [16] P. Soberón. Robust Tverberg and colourful Carathéodory results via random choice. Combinatorics, Probability and Computing, 27(3):427–440, 2018.
  • [17] P. Soberón and R. Strausz. A generalisation of Tverberg’s theorem. Discrete & Computational Geometry, 47(3):455–460, 2012.
  • [18] H. Tverberg. A generalization of Radon’s theorem. J. London Math. Soc., 41:123–128, 1966.