## Geometric Applications of a Randomized Optimization Technique (1999)

Venue: | Discrete Comput. Geom |

Citations: | 52 - 7 self |

### BibTeX

@ARTICLE{Chan99geometricapplications,

author = {Timothy M. Chan},

title = {Geometric Applications of a Randomized Optimization Technique},

journal = {Discrete Comput. Geom},

year = {1999},

volume = {22},

pages = {547--567}

}

### Years of Citing Articles

### OpenURL

### Abstract

We propose a simple, general, randomized technique to reduce certain geometric optimization problems to their corresponding decision problems. These reductions increase the expected time complexity by only a constant factor and eliminate extra logarithmic factors in previous, often more complicated, deterministic approaches (such as parametric searching). Faster algorithms are thus obtained for a variety of problems in computational geometry: finding minimal k-point subsets, matching point sets under translation, computing rectilinear p-centers and discrete 1centers, and solving linear programs with k violations. 1 Introduction Consider the classic randomized algorithm for finding the minimum of r numbers minfA[1]; : : : ; A[r]g: Algorithm rand-min 1. randomly pick a permutation hi 1 ; : : : ; i r i of h1; : : : ; ri 2. t /1 3. for k = 1; : : : ; r do 4. if A[i k ] ! t then 5. t / A[i k ] 6. return t By a well-known fact [27, 44], the expected number of times that step 5 is execut...

### Citations

9119 | Introduction to algorithms
- Cormen, Leiserson, et al.
- 2001
(Show Context)
Citation Context ...r]g: Algorithm rand-min 1. randomly pick a permutation hi 1 ; : : : ; i r i of h1; : : : ; ri 2. t /1 3. for k = 1; : : : ; r do 4. if A[i k ] ! t then 5. t / A[i k ] 6. return t By a well-known fact =-=[27, 44]-=-, the expected number of times that step 5 is executed is given by the harmonic number 1 + 1=2 + \Delta \Delta \Delta + 1=rsln r + 1. Imagine that the numbers A[1]; : : : ; A[r] are not pre-computed b... |

1867 |
Computational geometry, an introduction
- PREPARATA, SHAMOS
- 1985
(Show Context)
Citation Context ...ets; unfortunately, in this case, the decision problem does not seem to be any easier to solve. An \Omega\Gamma n log n) lower bound for the diameter problem is known even in the two-dimensional case =-=[57]. Similar -=-arguments imply the same lower bound for the discrete 1-center problem (by a reduction from the "set equality problem" [9]). 6 Discussions We have obtained improved expected time bounds (in ... |

403 | Applications of random sampling in computational geometry ii
- Clarkson
- 1988
(Show Context)
Citation Context ...amming using construction of a (7=8)-cutting of size 4. The construction has been extended to higher dimensions (with worse constants) [56]. A much simpler randomized method was suggested by Clarkson =-=[21]-=-: take a random sample R ae H of size O((1=ffi) log(1=ffi)) and canonically triangulate the arrangement of R. Derandomization of this method and refinements on the constants are discussed in several p... |

236 | Applying parallel computation algorithms in the design of serial algorithms
- Megiddo
- 1983
(Show Context)
Citation Context ...of the decision algorithm. 1.2 Previous Approaches One of the most general approaches for reducing geometric optimization problems to their decision problems is parametric search, invented by Megiddo =-=[54]-=- (see [1, 4, 7, 12, 15, 17, 26, 34, 53, 59, 61, 64] for just a partial list of examples). The basic idea is to simulate the decision algorithm---compare the optimum with t---with the parameter t being... |

215 |
Lower bounds for algebraic computation trees
- Ben-Or
- 1983
(Show Context)
Citation Context ...for the diameter problem is known even in the two-dimensional case [57]. Similar arguments imply the same lower bound for the discrete 1-center problem (by a reduction from the "set equality prob=-=lem" [9]-=-). 6 Discussions We have obtained improved expected time bounds (in some cases, matching known worst-case lower bounds) for several geometric optimization problems using a common randomized technique.... |

198 | Linear programming in linear time when the dimension is xed - Megiddo - 1984 |

190 |
Small-dimensional linear programming and convex hulls made easy
- SEIDEL
- 1991
(Show Context)
Citation Context ...st k halfplanes of Hg: (A point q violates a halfplane h if q 62 h.) Set L k (H) = ; if k ! 0. Note that for k = 0, this reduces to a standard linear program (which has well-known linear-time methods =-=[22, 31, 55, 63]-=-). The difficulty for the general case k ? 0 lies in the fact that L k (H) (the so-called ( k)-level) need not be convex and may have several local minima. As Matousek [51] and Roos and Widmayer [60] ... |

126 |
Fundamental algorithms, volume 1 of The art of computer programming. Addison-Wesley, third edition edition
- Knuth
- 1997
(Show Context)
Citation Context ...r]g: Algorithm rand-min 1. randomly pick a permutation hi 1 ; : : : ; i r i of h1; : : : ; ri 2. t /1 3. for k = 1; : : : ; r do 4. if A[i k ] ! t then 5. t / A[i k ] 6. return t By a well-known fact =-=[27, 44]-=-, the expected number of times that step 5 is executed is given by the harmonic number 1 + 1=2 + \Delta \Delta \Delta + 1=rsln r + 1. Imagine that the numbers A[1]; : : : ; A[r] are not pre-computed b... |

125 | Ray shooting and parametric search
- Agarwal, Matouˇsek
- 1993
(Show Context)
Citation Context ...omputing rectilinear p-centers and discrete 1centers, and solving linear programs with k violations. 1 Introduction Consider the classic randomized algorithm for finding the minimum of r numbers minfA=-=[1]-=-; : : : ; A[r]g: Algorithm rand-min 1. randomly pick a permutation hi 1 ; : : : ; i r i of h1; : : : ; ri 2. t /1 3. for k = 1; : : : ; r do 4. if A[i k ] ! t then 5. t / A[i k ] 6. return t By a well... |

111 | Approximate nearest neighbor queries in fixed dimensions
- ARYA, MOUNT
- 1993
(Show Context)
Citation Context ...mming queries to halfspace range reporting. This reduction, when specialized to ray shooting, yields a different approach in which the preprocessing algorithm employs random sampling. (Arya and Mount =-=[8]-=- noted a similar reduction in the context of nearest neighbor searching.) In contrast, in the proof of Theorem 3.2, randomization is used in the query algorithm but not the preprocessing. The random s... |

105 | Las vegas algorithms for linear and integer programming
- Clarkson
- 1995
(Show Context)
Citation Context ...st k halfplanes of Hg: (A point q violates a halfplane h if q 62 h.) Set L k (H) = ; if k ! 0. Note that for k = 0, this reduces to a standard linear program (which has well-known linear-time methods =-=[22, 31, 55, 63]-=-). The difficulty for the general case k ? 0 lies in the fact that L k (H) (the so-called ( k)-level) need not be convex and may have several local minima. As Matousek [51] and Roos and Widmayer [60] ... |

105 |
Slowing down sorting networks to obtain faster sorting algorithms
- Cole
- 1987
(Show Context)
Citation Context ... need an efficient parallelization of the simulated decision algorithm, so that comparisons can be "batched." Running time typically increases by logarithmic factors, even when an improvemen=-=t by Cole [25]-=- is applicable. As many researchers have commented, the resulting algorithms tend to be complicated and impractical; see the survey by Agarwal and Sharir [3]. In contrast, our randomized reductions us... |

104 |
Probabilistic algorithms
- Rabin
- 1976
(Show Context)
Citation Context ...to three subproblems, each of size roughly 2n=3. Now apply Lemma 2.1 (with r = 3 and ff = 2=3). 2 Note: the above theorem easily extends to finding closest b-tuples, for any constant integer b. Rabin =-=[58]-=- was historically the first to apply randomized techniques to the standard closest-pair problem, where U is a fixed-dimensional space and d(\Delta; \Delta) is the Euclidean metric. Clarkson and Shor i... |

100 | Efficient algorithms for geometric optimization
- Agarwal, Sharir
- 1998
(Show Context)
Citation Context ...factors, even when an improvement by Cole [25] is applicable. As many researchers have commented, the resulting algorithms tend to be complicated and impractical; see the survey by Agarwal and Sharir =-=[3]. In contr-=-ast, our randomized reductions use the decision algorithms purely as "black boxes," avoid the extra logarithmic factors in the running time, and are easier to implement. A number of alternat... |

92 |
The upper envelope of Voronoi surfaces and its applications
- Huttenlocher, Kedem, et al.
- 1993
(Show Context)
Citation Context ...et of a fixed convex polygon enclosing k points [33]. 4.2 Matching Point Sets Inspired by pattern recognition applications such as in computer vision, numerous papers in computational geometry (e.g., =-=[18, 19, 20, 41]-=-) have studied the problem of matching point sets under a class of transformations by minimizing the Hausdorff distance. We investigate a specialized case where the point sets are two-dimensional, the... |

91 | Applications of parametric searching in geometric optimization
- Agarwal, Sharir, et al.
- 1994
(Show Context)
Citation Context ...ision algorithm. 1.2 Previous Approaches One of the most general approaches for reducing geometric optimization problems to their decision problems is parametric search, invented by Megiddo [54] (see =-=[1, 4, 7, 12, 15, 17, 26, 34, 53, 59, 61, 64]-=- for just a partial list of examples). The basic idea is to simulate the decision algorithm---compare the optimum with t---with the parameter t being the unknown optimum itself. In most instances, the... |

89 |
A combinatorial bound for linear programming and related problems
- SHARIR, WELZL
- 1992
(Show Context)
Citation Context ...niques are not unified and as straightforward to apply as parametric search, potential applications are sometimes missed. (One important exception though is the class of LP-type optimization problems =-=[65]-=-, where general randomized linear-time solutions have been developed; most of the problems considered in this paper do not fit into this class.) We hope that our randomized technique will partially re... |

86 |
Cutting hyperplanes for divide-and-conquer
- Chazelle
- 1993
(Show Context)
Citation Context ... a random sample R ae H of size O((1=ffi) log(1=ffi)) and canonically triangulate the arrangement of R. Derandomization of this method and refinements on the constants are discussed in several papers =-=[16, 40, 48, 52]-=-. (For a variant known as shallow cuttings [49], better bounds are still possible.) Since the size of the cutting is not important when applying Lemma 2.1 (as long as it is bounded by a constant), the... |

77 | Efficient partition trees
- Matoušek
- 1992
(Show Context)
Citation Context ... a random sample R ae H of size O((1=ffi) log(1=ffi)) and canonically triangulate the arrangement of R. Derandomization of this method and refinements on the constants are discussed in several papers =-=[16, 40, 48, 52]-=-. (For a variant known as shallow cuttings [49], better bounds are still possible.) Since the size of the cutting is not important when applying Lemma 2.1 (as long as it is bounded by a constant), the... |

74 | Geometric pattern matching under euclidean motion
- Chew, Goodrich, et al.
- 1997
(Show Context)
Citation Context ...et of a fixed convex polygon enclosing k points [33]. 4.2 Matching Point Sets Inspired by pattern recognition applications such as in computer vision, numerous papers in computational geometry (e.g., =-=[18, 19, 20, 41]-=-) have studied the problem of matching point sets under a class of transformations by minimizing the Hausdorff distance. We investigate a specialized case where the point sets are two-dimensional, the... |

71 |
Explicit Expanders and the Ramanujan Conjectures
- Lubotzky, Phillips, et al.
- 1986
(Show Context)
Citation Context ... the binary search [20, 39, 66]; that algorithm relies heavily on repeated weighted-median computations. In other instances, one can employ nontrivial explicit constructions of expander graphs (e.g., =-=[45]-=-) following a technique of Katz and Sharir [42, 43]. Without additional ideas, all of these techniques increase the running time by at least a logarithmic factor. Randomized techniques have also been ... |

69 | Output-sensitive results on convex hulls, extreme points, and related problems. Discrete Computational Geometry
- Chan
- 1996
(Show Context)
Citation Context ...ek [51] solves the problem in O(n log n + k 2 log 2 n) time, which is efficient only when k is roughly smaller than p n. The O(n log n) term can be improved to O(n log k) by a technique of the author =-=[14]-=-. We derive here the expected time bound O(n log n) for all values of k. Surprisingly, if k = O(n= log fi n) for a constant fi ? 1, our bound strengthens to O(n), matching the complexity of the standa... |

68 |
Reporting points in halfspaces
- Matousek
- 1992
(Show Context)
Citation Context ...anonically triangulate the arrangement of R. Derandomization of this method and refinements on the constants are discussed in several papers [16, 40, 48, 52]. (For a variant known as shallow cuttings =-=[49]-=-, better bounds are still possible.) Since the size of the cutting is not important when applying Lemma 2.1 (as long as it is bounded by a constant), the original method of Dyer and Megiddo or the ran... |

56 | Iterated nearest neighbors and finding minimal polytopes
- Eppstein, Erickson
- 1994
(Show Context)
Citation Context ...ing: ffl The minimum L1 -diameter k-point subset of a planar n-point set can be found in O(n log n) expected time for any 2sksn. The best previous algorithm for general k was by Eppstein and Erickson =-=[35]-=- and took O(n log 2 n) time. ffl The minimum L1 -Hausdorff distance between two planar n-point sets under translations can be found in O(n 2 log n) expected time. This problem was studied by Chew and ... |

52 |
An optimal-time algorithm for slope selection
- Cole, Salowe, et al.
- 1989
(Show Context)
Citation Context ...ision algorithm. 1.2 Previous Approaches One of the most general approaches for reducing geometric optimization problems to their decision problems is parametric search, invented by Megiddo [54] (see =-=[1, 4, 7, 12, 15, 17, 26, 34, 53, 59, 61, 64]-=- for just a partial list of examples). The basic idea is to simulate the decision algorithm---compare the optimum with t---with the parameter t being the unknown optimum itself. In most instances, the... |

52 |
Linear time algorithms for two- and three-variable linear programs
- Dyer
- 1984
(Show Context)
Citation Context ...onstraints. (Rectilinear applications are often more amenable, but certain non-rectilinear ones currently pose technical difficulties.) Note that unlike in traditional prune-and-search methods (e.g., =-=[31, 55, 56]-=-), we do not need an oracle to identify which subproblem contains the actual solution and requires recursion; our randomized technique will guide us to the right subproblem quickly, with the aid of th... |

48 |
Finding k points with minimum diameter and related problems
- Aggarwal, Imai, et al.
- 1991
(Show Context)
Citation Context ...s R 1 ; R 2 , we let R 1sR 2 denote the smallest rectangle enclosing their union. 4.1 Minimal k-Point Subsets Motivated by applications in clustering and statistical analysis, a number of researchers =-=[6, 10, 28, 33, 35, 50] have look-=-ed at problems of the type: given an n-point set P , compute a "minimal" kpoint subset. We illustrate our technique on one specific case, where the point set S is planar, and the measure of ... |

48 | Product range spaces, sensitive sampling, and derandomization
- Bronnimann, Chazelle, et al.
- 1993
(Show Context)
Citation Context ...ision algorithm. 1.2 Previous Approaches One of the most general approaches for reducing geometric optimization problems to their decision problems is parametric search, invented by Megiddo [54] (see =-=[1, 4, 7, 12, 15, 17, 26, 34, 53, 59, 61, 64]-=- for just a partial list of examples). The basic idea is to simulate the decision algorithm---compare the optimum with t---with the parameter t being the unknown optimum itself. In most instances, the... |

47 |
On geometric optimization with few violated constraints
- Matoušek
- 1995
(Show Context)
Citation Context ...d constraints can be solved in O(n log n) expected time in the feasible case for any 0sksn. (If k is not too large, the time bound is actually linear.) This type of problem was considered by Matousek =-=[51]-=- and Roos and Widmayer [60]; the previous time bound for this version for arbitrary k was O(n log 2 n). ffl Given an infeasible two-dimensional linear program with n constraints, the smallest number k... |

46 | Efficient collision detection for moving polyhedra
- Schömer, Thiel
- 1995
(Show Context)
Citation Context |

45 | Efficient randomized algorithms for some geometric optimization problems - Agarwal, Sharir - 1996 |

45 | Faster construction of planar two-centers
- Eppstein
- 1997
(Show Context)
Citation Context |

44 | Parallel algorithms for higherdimensional convex hulls
- Amato, Goodrich, et al.
- 1994
(Show Context)
Citation Context |

43 |
Static and dynamic algorithms for k-point clustering problems
- Datta, Lenhof, et al.
- 1995
(Show Context)
Citation Context ...s R 1 ; R 2 , we let R 1sR 2 denote the smallest rectangle enclosing their union. 4.1 Minimal k-Point Subsets Motivated by applications in clustering and statistical analysis, a number of researchers =-=[6, 10, 28, 33, 35, 50] have look-=-ed at problems of the type: given an n-point set P , compute a "minimal" kpoint subset. We illustrate our technique on one specific case, where the point set S is planar, and the measure of ... |

41 | An expander-based approach to geometric optimization
- Katz, Sharir
- 1993
(Show Context)
Citation Context ... relies heavily on repeated weighted-median computations. In other instances, one can employ nontrivial explicit constructions of expander graphs (e.g., [45]) following a technique of Katz and Sharir =-=[42, 43]-=-. Without additional ideas, all of these techniques increase the running time by at least a logarithmic factor. Randomized techniques have also been suggested as an alternative in several isolated cas... |

40 | A near-linear algorithm for the planar 2-center problem
- Sharir
- 1997
(Show Context)
Citation Context |

39 |
D.B.: The complexity of selection and ranking in X + Y and matrices with sorted columns
- Frederickson, Johnson
- 1982
(Show Context)
Citation Context ..., then an ordinary binary search is sufficient. For many rectilinear problems, the search space forms a matrix with sorted rows/columns, and one can use Frederickson and Johnson's selection algorithm =-=[37]-=- to carry out the binary search [20, 39, 66]; that algorithm relies heavily on repeated weighted-median computations. In other instances, one can employ nontrivial explicit constructions of expander g... |

35 | Decision tree complexity and betti numbers - YAO - 1997 |

33 | Fixed-dimensional linear programming queries made easy
- Chan
- 1996
(Show Context)
Citation Context ...ithout additional ideas, all of these techniques increase the running time by at least a logarithmic factor. Randomized techniques have also been suggested as an alternative in several isolated cases =-=[2, 10, 13, 15, 29, 34, 47, 50]-=-. Unfortunately, since these techniques are not unified and as straightforward to apply as parametric search, potential applications are sometimes missed. (One important exception though is the class ... |

31 |
Randomized optimal algorithm for slope selection
- Matousek
- 1991
(Show Context)
Citation Context ...ithout additional ideas, all of these techniques increase the running time by at least a logarithmic factor. Randomized techniques have also been suggested as an alternative in several isolated cases =-=[2, 10, 13, 15, 29, 34, 47, 50]-=-. Unfortunately, since these techniques are not unified and as straightforward to apply as parametric search, potential applications are sometimes missed. (One important exception though is the class ... |

31 |
Linear time algorithms for linear programming in R and related problems
- Megiddo
- 1983
(Show Context)
Citation Context ...onstraints. (Rectilinear applications are often more amenable, but certain non-rectilinear ones currently pose technical difficulties.) Note that unlike in traditional prune-and-search methods (e.g., =-=[31, 55, 56]-=-), we do not need an oracle to identify which subproblem contains the actual solution and requires recursion; our randomized technique will guide us to the right subproblem quickly, with the aid of th... |

30 |
width, closest line pair and parametric searching. Discrete Comput
- Diameter
- 1993
(Show Context)
Citation Context |

30 |
A randomized algorithm for slope selection
- Dillencourt, Mount, et al.
- 1992
(Show Context)
Citation Context ...ithout additional ideas, all of these techniques increase the running time by at least a logarithmic factor. Randomized techniques have also been suggested as an alternative in several isolated cases =-=[2, 10, 13, 15, 29, 34, 47, 50]-=-. Unfortunately, since these techniques are not unified and as straightforward to apply as parametric search, potential applications are sometimes missed. (One important exception though is the class ... |

30 | Rectilinear and polygonal ppiercing and p-center problems
- Sharir, Welzl
- 1996
(Show Context)
Citation Context ...his problem was studied by Chew and Kedem [20], who gave an O(n 2 log 2 n)-time algorithm. ffl The two-dimensional rectilinear 5-center problem can be solved in O(n log n) expected time. Recent works =-=[46, 62, 66]-=- gave an O(n log 2 n) time bound. ffl The two-dimensional linear programming problem with k violated constraints can be solved in O(n log n) expected time in the feasible case for any 0sksn. (If k is ... |

26 | Geometric Pattern Matching in d-Dimensional Space
- Chew, Efrat
- 2003
(Show Context)
Citation Context ...et of a fixed convex polygon enclosing k points [33]. 4.2 Matching Point Sets Inspired by pattern recognition applications such as in computer vision, numerous papers in computational geometry (e.g., =-=[18, 19, 20, 41]-=-) have studied the problem of matching point sets under a class of transformations by minimizing the Hausdorff distance. We investigate a specialized case where the point sets are two-dimensional, the... |

26 | A deterministic algorithm for the threedimensional diameter problem
- Matoušek, Schwarzkopf
- 1993
(Show Context)
Citation Context |

25 | The woodfuel problem
- Agarwal
- 1995
(Show Context)
Citation Context ...ime by a logarithmic factor. 4.3 Rectilinear p-Centers A class of facility location problems known as p-center problems has received much attention in the computational geometry literature (e.g., see =-=[5, 30, 34, 64, 66]-=-), due to applications in various areas such as operations research and clustering. We consider the case where p is a constant, the dimension is two, and the metric is L1 : Problem. Given an n-point s... |

24 |
An optimal algorithm for the (≤k)-levels, with applications to separation and transversal problems
- Everett, Robert, et al.
- 1996
(Show Context)
Citation Context ...exists a point in L k (H). In other words, we want to find a maximal consistent subset of halfplanes (of size n \Gamma k) [32, 51]. The dual is equivalent to the following line classification problem =-=[36, 51]-=-, with motivation from statistics: given n points in IR 2 each colored red or blue, find a line ` that minimizes the total number k of red points above ` and blue points below `. The problem is also r... |

23 | Optimal slope selection via cuttings
- Bronnimann, Chazelle
- 1994
(Show Context)
Citation Context ...s paper do not fit into this class.) We hope that our randomized technique will partially rectify this shortcoming. One other alternative to parametric searching that has been noted in the literature =-=[11]-=- involves the application of geometric cutting tools. Some of these tools (in a very elementary form) will be of use in several of our own randomized reductions for non-rectilinear problems. In Sectio... |

23 | More planar two-center algorithms
- Chan
- 1999
(Show Context)
Citation Context |