Applications of random sampling in computational geometry,II

We use random sampling for several new geometric algorithms. The algorithms are Las Vegas, and their expected bounds are with respect to the random behavior of the algorithms. These algorithms follow from new general results giving sharp bounds for the use of random subsets in geometric algorithms. These bounds show that random subsets can be used optimally for divide-and-conquer, and also give bounds for a simple, general technique for building geometric structures incrementally. One new algorithm reports all the intersecting pairs of a set of line segments in the plane, and requiresO(A+n logn) expected time, whereA is the number of intersecting pairs reported. The algorithm requiresO(n) space in the worst case. Another algorithm computes the convex hull ofn points inE ^d inO(n logn) expected time ford=3, andO(n ^[d/2]) expected time ford>3. The algorithm also gives fast expected times for random input points. Another algorithm computes the diameter of a set ofn points inE ³ inO(n logn) expected time, and on the way computes the intersection ofn unit balls inE ³. We show thatO(n logA) expected time suffices to compute the convex hull ofn points inE ³, whereA is the number of input points on the surface of the hull. Algorithms for halfspace range reporting are also given. In addition, we give asymptotically tight bounds for (k)-sets, which are certain halfspace partitions of point sets, and give a simple proof of Lee's bounds for high-order Voronoi diagrams.     

New applications of random sampling in computational geometry

This paper gives several new demonstrations of the usefulness of random sampling techniques in computational geometry. One new algorithm creates a search structure for arrangements of hyperplanes by sampling the hyperplanes and using information from the resulting arrangement to divide and conquer. This algorithm requiresO(s ^d+ ) expected preprocessing time to build a search structure for an arrangement ofs hyperplanes ind dimensions. The expectation, as with all expected times reported here, is with respect to the random behavior of the algorithm, and holds for any input. Given the data structure, and a query pointp, the cell of the arrangement containingp can be found inO(logs) worst-case time. (The bound holds for any fixed >0, with the constant factors dependent ond and .) Using point-plane duality, the algorithm may be used for answering halfspace range queries. Another algorithm finds random samples of simplices to determine the separation distance of two polytopes. The algorithm uses expectedO(n ^[d/2]) time, wheren is the total number of vertices of the two polytopes. This matches previous results [10] for the cased = 3 and extends them. Another algorithm samples points in the plane to determine their orderk Voronoi diagram, and requires expectedO(s ¹⁺ k) time fors points. (It is assumed that no four of the points are cocircular.) This sharpens the boundO(sk ² logs) for Lee's algorithm [21], andO(s ² logs+k(s–k) log² s) for Chazelle and Edelsbrunner's algorithm [4]. Finally, random sampling is used to show that any set ofs points inE ³ hasO(sk ² log⁸ s/(log logs)⁶) distinctj-sets withjk. (ForS E ^d , a setS S with |S| =j is aj-set ofS if there is a half-spaceh ⁺ withS =S h ⁺.) This sharpens with respect tok the previous boundO(sk ⁵) [5]. The proof of the bound given here is an instance of a probabilistic method [15].A preliminary version of this paper appeared in theProceedings of the 18th Annual ACM Symposium on Theory of Computing, Berkeley, CA, 1986.     

A new approach to the dynamic maintenance of maximal points in a plane

A pointp _i=(x _i, y_i) in thex–y plane ismaximal if there is no pointp _j=(x _j, y_j) such thatx _j>x_i andy _j>y_i. We present a simple data structure, a dynamic contour search tree, which contains all the points in the plane and maintains an embedded linked list of maximal points so thatm maximal points are accessible inO(m) time. Our data structure dynamically maintains the set of points so that insertions takeO(logn) time, a speedup ofO(logn) over previous results, and deletions takeO((logn)²) time.The research of the first author was partially supported by the National Science Foundation under Grant No. DCR-8320214 and by the Office of Naval Research on Contract No. N 00014-86-K-0689. The research of the second author was partially supported by the Office of Naval Research on Contract No. N 00014-86-K-0689.     

Maintaining the minimal distance of a point set in polylogarithmic time

A dynamic data structure is given that maintains the minimal distance in a set ofn points ink-dimensional space inO((logn) ^k log logn) amortized time per update. The size of the data structure is bounded byO(n(logn) ^k ). Distances are measured in the MinkowskiL _t -metric, where 1 t . This is the first dynamic data structure that maintains the minimal distance in polylogarithmic time for fully on-line updates.This work was supported by the ESPRIT II Basic Research Actions Program, under Contract No. 3075 (project ALCOM).     

Covering times of random walks on bounded degree trees and other graphs

The motivating problem for this paper is to find the expected covering time of a random walk on a balanced binary tree withn vertices. Previous upper bounds for general graphs ofO(|V| |E|)⁽¹⁾ andO(|V| |E|/d _min)⁽²⁾ imply an upper bound ofO(n ²). We show an upper bound on general graphs ofO( |E| log |V|), which implies an upper bound ofO(n log² n). The previous lower bound was (|V| log |V|) for trees.⁽²⁾ In our main result, we show a lower bound of (|V| (log _{d max} |V|)²) for trees, which yields a lower bound of (n log² n). We also extend our techniques to show an upper bound for general graphs ofO(max{E T_i} log |V|).     

Quasi-optimal range searching in spaces of finite VC-dimension

The range-searching problems that allow efficient partition trees are characterized as those defined by range spaces of finite Vapnik-Chervonenkis dimension. More generally, these problems are shown to be the only ones that admit linear-size solutions with sublinear query time in the arithmetic model. The proof rests on a characterization of spanning trees with a low stabbing number. We use probabilistic arguments to treat the general case, but we are able to use geometric techniques to handle the most common range-searching problems, such as simplex and spherical range search. We prove that any set ofn points inE ^d admits a spanning tree which cannot be cut by any hyperplane (or hypersphere) through more than roughlyn ^1–1/d edges. This result yields quasi-optimal solutions to simplex range searching in the arithmetic model of computation. We also look at polygon, disk, and tetrahedron range searching on a random access machine. Givenn points inE ², we derive a data structure of sizeO(n logn) for counting how many points fall inside a query convexk-gon (for arbitrary values ofk). The query time isO(kn logn). Ifk is fixed once and for all (as in triangular range searching), then the storage requirement drops toO(n). We also describe anO(n logn)-size data structure for counting how many points fall inside a query circle inO(n log² n) query time. Finally, we present anO(n logn)-size data structure for counting how many points fall inside a query tetrahedron in 3-space inO(n ^2/3 log² n) query time. All the algorithms are optimal within polylogarithmic factors. In all cases, the preprocessing can be done in polynomial time. Furthermore, the algorithms can also handle reporting within the same complexity (adding the size of the output as a linear term to the query time).Portions of this work have appeared in preliminary form in Partition trees for triangle counting and other range searching problems (E. Welzl),Proc. 4th Ann. ACM Symp. Comput. Geom. (1988), 23–33, and Tight Bounds on the Stabbing Number of Spanning Trees in Euclidean Space (B. Chazelle), Comput. Sci. Techn. Rep. No. CS-TR-155-88, Princeton University, 1988. Bernard Chazelle acknowledges the National Science Foundation for supporting this research in part under Grant CCR-8700917. Emo Welzl acknowledges the Deutsche Forschungsgemeinschaft for supporting this research in part under Grant We 1265/1-1.     

A new duality result concerning voronoi diagrams

A new duality between order-k Voronoi diagrams inE ^d and convex hulls inE ^d+1 is established. It implies a reasonably simple algorithm for computing the order-k diagram forn points in the plane inO(k ² n logn) time and optimalO(k(n–k)) space.Research was supported by the Austrian Fond zur Foerderung der wissenschaftlichen Forschung.     

Rectilinear shortest paths in the presence of rectangular barriers

In this paper we address the following shortest-path problem. Given a point in the plane andn disjoint isothetic rectangles (barriers), we want to construct a shortestL ₁ path (not crossing any of the barriers) from the source point to any given query point. A restricted version of this problem (where the source and destination points are knowna priori) had been solved earlier inO(n ²) time. Our approach consists of preprocessing the source point and the barriers to obtain a planar subdivision where a query point can be located and a shortest path connecting it to the source point quickly transvered. By showing that any such path is monotone in at least one ofx ory directions, we are able to apply a plane sweep technique to divide the plane intoO(n) rectangular regions. This leads to an algorithm whose complexity isO(n logn) preprocessing time,O(n) space, andO(logn+k) query time, wherek is the number of turns on the reported path. If only the length of the path is sought,O(logn) query time suffices. Furthermore, we show an (n logn) time lower bound for the case where the source and destination points are known in advance, which implies the optimality of our algorithm in this case.A preliminary version of this paper appeared in theProceedings of the First Symposium on Computational Geometry (1985).Supported in part by CNPq-Conselho Nacional de Desenvolvimento Científico e Tecnológico (Brazil).Supported in part by the National Science Foundation under Grants MCS 8420814 and ECS 8340031.     

Some simple in-place merging algorithms

The theoretical presentation and analysis is given for two families of simple in-place merging algorithms and their limiting cases. The first family merges stably inO(k·n) time andO(n ^1/k ) additional space with a limiting case running inO(n logn) time and constant space. The second family merges unstably inO (k ·n) time andO(log ^k n) space with a limiting case running inO(nG(n)) time and constant space. HereG(n) is the leastk such thatF(k) n whereF(0)=1 andF(i)=2 ^F(i–1) fori1. Each algorithm gives rise to a corresponding merge sort.     

Constructing arrangements optimally in parallel

We give two optimal parallel algorithms for constructing the arrangement ofn lines in the plane. The first nethod is quite simple and runs inO(log² n) time usingO(n ²) work, and the second method, which is more sophisticated, runs inO(logn) time usingO(n ²) work. This second result solves a well-known open problem in parallel computational geometry, and involves the use of a new algorithmic technique, the construction of an -pseudocutting. Our results immediately imply that the arrangement ofn hyperplanes in ^d inO(logn) time usingO(n ^d) work, for fixedd, can be optimally constructed. Our algorithms are for the CREW PRAM.This research was supported by the National Science Foundation under Grants CCR-8810568 and CCR-9003299, and by the NSF and DARPA under Grant CCR-8908092.     

Partitioning arrangements of lines II: Applications

In this paper we present efficient deterministic algorithms for various problems involving lines or segments in the plane, using the partitioning algorithm described in a companion paper [A3]. These applications include: (i) anO(m ^2/3 n ^2/3 · log^2/3 n · log^/3 (m/n)+(m+n) logn) algorithm to compute all incidences betweenm points andn lines, where is a constant <3.33; (ii) anO(m ^2/3 n ^2/3 · log^5/3 n · log^/3 (m/n)+(m+n) logn) algorithm to computem faces in an arrangement ofn lines; (iii) anO(n ^4/3 log^(+2)/3 n) algorithm to count the number of intersections in a set ofn segments; (iv) anO(n ^4/3 log^{( + 2)/3} n) algorithm to count red-blue intersections between two sets of segments, and (v) anO(n ^3/2 log^/3 n) algorithm to compute spanning trees with low stabbing number for a set ofn points. We also present an algorithm that, given set ofn points in the plane, preprocesses it, in timeO(nm log^+1/2 n), into a data structure of sizeO(m) forn lognmn ², so that the number of points ofS lying inside a query triangle can be computed inO((n/m) log^3/2 n) time.Work on this paper has been supported by Office of Naval Research Grant N00014-87-K-0129, by National Science Foundation Grant DCR-83-20085, and by grants from the Digital Equipment Corporation and the IBM Corporation. A preliminary version of this paper appears in theProceedings of the 5th ACM Symposium on Computational Geometry, 1989, pp. 11–22.     

Efficient Computation of Rectilinear Geodesic Voronoi Neighbor in the Presence of Obstacles

In this paper we present an algorithm to compute the rectilinear geodesic voronoi neighbor of an arbitrary query pointqamong a setSofmpoints in the presence of a set ofnvertical line segment obstacles inside a rectangular floor. The distance between a pair of points α and β is the shortest rectilinear distance avoiding the obstacles in and is denoted by δ(α, β). The rectilinear geodesic voronoi neighbor of an arbitrary query pointq,RGVN(q) is the pointp_i ∈ Ssuch that δ(q, p_i) is minimum. The algorithm suggests a preprocessing of the elements of the setsSand inO((m + n)log(m + n)) time such that for an arbitrary query pointq, theRGVNquery can be answered inO(log(m + n)) time. The space required for storing the preprocessed information isO(n + m log m). If the points inSare placed on the boundary of the rectangular floor, a different technique is adopted to decrease the space complexity toO(m + n). This technique works even if the obstacles are rectangles instead of line segments. Finally, the parallelization of the preprocessing steps for the latter algorithm is suggested, which takesO(log³(m + n)) time, usingO((m + n)^1.5/log²(m + n)) processors andO(log(m + n)) query time.     

On the approximation of positive operators and the behaviour of the spectra of the approximants

LetT be a positive linear operator on the Banach latticeE and let (S _n ) be a sequence of bounded linear operators onE which converge strongly toT. Our main results are concerned with the question under which additional assumptions onS _n andT the peripheral spectra (S _n ) ofS _n converge to the peripheral spectrum (T) ofT. We are able to treat even the more general case of discretely convergent sequences of operators.     

Triangulating point sets in space

A setP ofn points inR ^d is called simplicial if it has dimensiond and contains exactlyd + 1 extreme points. We show that whenP containsn interior points, there is always one point, called a splitter, that partitionsP intod + 1 simplices, none of which contain more thandn/(d + 1) points. A splitter can be found inO(d ⁴ +nd ²) time. Using this result, we give anO(nd ⁴ log_1+1/d n) algorithm for triangulating simplicial point sets that are in general position. InR ³ we give anO(n logn +k) algorithm for triangulating arbitrary point sets, wherek is the number of simplices produced. We exhibit sets of 2n + 1 points inR ³ for which the number of simplices produced may vary between (n – 1)² + 1 and 2n – 2. We also exhibit point sets for which every triangulation contains a quadratic number of simplices.Research supported by the Natural Science and Engineering Research Council grant A3013 and the F.C.A.R. grant EQ1678.     

Dynamic Text Indexing under String Updates

This paper generalizes the dynamic text indexing problem, introduced in [15], to insertion and deletion of strings. The problem is to quickly answer on-line queries about the occurrences of arbitrary pattern strings in a text that may change due to insertion or deletion of strings from it. To treat strings as atomic objects, we provide new sequential techniques and related data structures, which combine the suffix tree with the naming technique used in parallel computation on strings. We also introduce a geometric interpretation of the problem of finding the occurrences of a pattern in a given substring of the text. As a result, the algorithm allows the insertion in the text of a stringSinO(|S| log(n + |S|)) amortized time, and the deletion from the text of a stringSinO(|S| log n) amortized time, wherenis the length of the current text. A pattern search requiresO(p log p + upd ( + log p) + pocc) worst-case time, wherepis the length of the pattern andpoccis the number of its occurrences in the current text, obtained after the execution ofupdupdate operations. This solution requiresO(n² log n) space, which is not initialized.We also provide a technique to reduce the space toO(n log n), yielding a solution that requiresO((p + upd) log p + pocc) query time in the worst-case,O(|S| log^3/2(|S| + n)) amortized time to insert a stringSin, andO(|S| log^3/2n) amortized time to delete a stringSfrom the current text.Furthermore, we use our techniques to solve the novel on-line dynamic tree matching problem that requires the on-line detection of the occurrences of an arbitrary subtree in a forest of ordered labeled trees. The forest may change due to insertion or deletion of subtrees or by renaming of some nodes. Such a problem is solved by a simple reduction to the dynamic text indexing problem.     

On-line coloring of perfect graphs

Lovász, Saks, and Trotter showed that there exists an on-line algorithm which will color any on-linek-colorable graph onn vertices withO(nlog^(2k–3) n/log^(2k–4) n) colors. Vishwanathan showed that at least (log ^k–1 n/k ^k ) colors are needed. While these remain the best known bounds, they give a distressingly weak approximation of the number of colors required. In this article we study the case of perfect graphs. We prove that there exists an on-line algorithm which will color any on-linek-colorable perfect graph onn vertices withn ^10k/loglogn colors and that Vishwanathan's techniques can be slightly modified to show that his lower bound also holds for perfect graphs. This suggests that Vishwanathan's lower bound is far from tight in the general case.Research partially supported by Office of Naval Research grant N00014-90-J-1206.     

On the extreme spectral properties of Toeplitz matrices generated byL 1 functions with several minima/maxima

In this paper we are concerned with the asymptotic behavior of the smallest eigenvalue ₁ ⁽ⁿ⁾ of symmetric (Hermitian)n ×n Toeplitz matricesT _n (f) generated by an integrable functionf defined in [–, ]. In [7, 8, 11] it is shown that ₁ ⁽ⁿ⁾ tends to essinff =m _f in the following way: ₁ ⁽ⁿ⁾ –m _f 1/n ^2k . These authors use three assumptions:A1)f –m _f has a zero inx =x ₀ of order 2k.A2)f is continuous and at leastC ^2k in a neighborhood ofx ₀.A3)x =x ₀ is the unique global minimum off in [–, ]. In [10] we have proved that the hypothesis of smoothnessA2 is not necessary and that the same result holds under the weaker assumption thatf L ¹[–, ]. In this paper we further extend this theory to the case of a functionf L ¹[–, ] having several global minima by suppressing the hypothesisA3 and by showing that the maximal order 2k of the zeros off –m _f is the only parameter which characterizes the rate of convergence of ₁ ⁽ⁿ⁾ tom _f .     

A sparse graph almost as good as the complete graph on points inK dimensions

A setV ofn points ink-dimensional space induces a complete weighted undirected graph as follows. The points are the vertices of this graph and the weight of an edge between any two points is the distance between the points under someL _p metric. Let ε≤1 be an error parameter and letk be fixed. We show how to extract inO(n logn+ε ^−k log(1/ε)n) time a sparse subgraphG=(V, E) of the complete graph onV such that: (a) for any two pointsx, y inV, the length of the shortest path inG betweenx andy is at most (1+∈) times the distance betweenx andy, and (b)|E|=O(ε^−k n).     

Points and triangles in the plane and halving planes in space

We prove that for any setS ofn points in the plane andn ^3−α triangles spanned by the points inS there exists a point (not necessarily inS) contained in at leastn ^3−3α/(c log⁵ n) of the triangles. This implies that any set ofn points in three-dimensional space defines at most halving planes. Work on this paper by Boris Aronov and Rephael Wenger has been supported by DIMACS under NSF Grant STC-88-09648. Work on this paper by Bernard Chazelle has been supported by NSF Grant CCR-87-00917. Work by Herbert Edelsbrunner has been supported by NSF Grant CCR-87-14565. Micha Sharir has been supported by ONR Grant N00014-87-K-0129, by NSF Grant CCR-89-01484, and by grants from the U.S.-Israeli Binational Science Foundation, the Israeli National Council for Research and Development, and the Fund for Basic Research administered by the Israeli Academy of Sciences.     

Pseudorandom generators for space-bounded computation

Pseudorandom generators are constructed which convertO(SlogR) truly random bits toR bits that appear random to any algorithm that runs inSPACE(S). In particular, any randomized polynomial time algorithm that runs in spaceS can be simulated using onlyO(Slogn) random bits. An application of these generators is an explicit construction of universal traversal sequences (for arbitrary graphs) of lengthn ^O(logn).The generators constructed are technically stronger than just appearing random to spacebounded machines, and have several other applications. In particular, applications are given for deterministic amplification (i.e. reducing the probability of error of randomized algorithms), as well as generalizations of it.This work was done in the Laboratory for Computer Science, MIT, supported by NSF 865727-CCR and ARO DALL03-86-K-017