We give a model for the performance impact on wireless systems of the limitations of certain resources, namely, the base-station power amplifier and the available OVSF codes. These limitations are readily modeled in the loss model formulation as a stochastic knapsack. A simple and well-known recurrence of Kaufman and Roberts allows the predictions of the model to be efficiently calculated. We discuss the assumptions and approximations we have made that allow the use of the model. We have included the model in Ocelot, a Lucent tool for modeling and optimizing cellular phone systems. The model is fast to compute, differentiable with respect to the relevant parameters, and able to model broad ranges of capacity and resource use. These conditions are critical to our application of optimization.

Metric spaces are very simple geometric structures: they comprise a set and a distance measure that obeys the triangle inequality. `epsilon`-nets, also called "low-dispersion sets" or "farthest-point samples", are subsets of metric spaces that are "well-distributed" in a certain sense. They can be found efficiently using a greedy algorithm. This talk will review that algorithm, and briefly survey applications of `epsilon`-nets in motion planning, nearest neighbor searching, building meshes to approximate curves and surfaces, and building triangulations to approximate functions.

This work addresses the problem of approximating a manifold by a simplicial mesh, and the related problem of building triangulations for the purpose of piecewise-linear approximation of functions. It has long been understood that the vertices of such meshes or triangulations should be "well-distributed," or satisfy certain "sampling conditions." This work clarifies and extends some algorithms for finding such well-distributed vertices, by showing that they can be regarded as finding `\epsilon`-nets or Delone sets in appropriate metric spaces. In some cases where such Delone properties were already understood, such as for meshes to approximate smooth manifolds that bound convex bodies, the upper and lower bound results are extended to more general manifolds; in particular, under some natural conditions, the minimum Hausdorff distance for a mesh with `n` simplices to a `d`-manifold `M` is

`\Theta((\int_M\sqrt{|\kappa(x)|}/n)^{2/d})`

as `n\rightarrow\infty`, where `\kappa(x)` is the Gaussian curvature at point `x\in M`. We also relate these constructions to Dudley's approximation scheme for convex bodies, which can be interpreted as involving an `\epsilon`-net in a metric space whose distance function depends on surface normals. Finally, a novel scheme is given, based on the Steinhaus transform, for scaling a metric space by a Lipschitz function to obtain a new metric. This scheme is applied to show that some algorithms for building finite element meshes and for surface reconstruction can be also be interpreted in the framework of metric space `\epsilon`-nets.

Given a set `S` of points in a metric space with distance function `D`, the nearest-neighbor searching problem is to build a data structure for `S` so that for an input query point `q`, the point `s\in S` that minimizes `D(s,q)` can be found quickly. We survey approaches to this problem, and its relation to concepts of metric space dimension. Several measures of dimension can be estimated using nearest-neighbor searching, while others can be used to estimate the cost of that searching. In recent years, several data structures have been proposed that are provably good for low-dimensional spaces, for some particular measures of dimension. These and other data structures for nearest-neighbor searching are surveyed.

Given a collection `S` of subsets of some set `U`, and `M \subset U`, the set cover problem is to find the smallest subcollection `C\subset S` such that `M` is a subset of the union of the sets in `C`. While the general problem is NP-hard to solve, even approximately, here we consider some geometric special cases, where usually `U = \RR^d`. Combining previously known techniques [BG,CF], we show that polynomial time approximation algorithms with provable performance exist, under a certain general condition: that for a random subset `R\subset S` and function `f()`, there is a decomposition of the complement `U\setminus\cup_{Y\in R} Y` into an expected `f(|R|)` regions, each region of a particular simple form. Under this condition, a cover of size `O(f(|C|))` can be found in polynomial time. Using this result, and combinatorial geometry results implying bounding functions `f(c)` that are nearly linear, we obtain `o(\log c)` approximation algorithms for covering by fat triangles, by pseudodisks, by a family of fat objects, and others. Similarly, constant-factor approximations follow for similar-sized fat triangles and fat objects, and for fat wedges. With more work, we obtain constant-factor approximation algorithms for covering by unit cubes in `\RR^3`, and for guarding an `x`-monotone polygonal chain.

Given an `n\times d` matrix `A` and an `n`-vector `b`, the `L_1` regression problem is to find the vector `x` minimizing the objective function `||Ax-b||_1`, where `||y||_1 \equiv \sum_i |y_i|` for vector `y`. This paper gives an algorithm needing `O(n\log n)d^{O(1)}` time in the worst case to obtain an approximate solution, with objective function value within a fixed ratio of optimum. Given `\epsilon>0`, a solution whose value is within `1+\epsilon` of optimum can be obtained either by a deterministic algorithm using an additional `O(n)(d/\epsilon)^{O(1)}` time, or by a Monte Carlo algorithm using an additional `O((d/\epsilon)^{O(1)})` time. The analysis of the randomized algorithm shows that weighted coresets exist for `L_1` regression. The algorithms use the ellipsoid method, gradient descent, and random sampling.

Let `S` be a set of `n` points in `d` dimensions. A `k`-set of `S` is a subset of size `k` that is the intersection of `S` with some open halfspace. This note shows that if the points of `S` are random, with a coordinate-wise independent distribution, then the expected number of `k`-sets of `S` is `O((k\log(en/k))^{d-1})2^d/(d-1)!`, as `k\log n->oo`, with a constant independent of the dimension.

We introduce a simple model of the effect of temporal variation in signal strength on active-set membership, for cellular phone systems that use the soft-handoff algorithm of IS-95a. This model is based on a steady-state calculation, and its applicability is confirmed by Monte Carlo studies.

Given a set `S` of `n` sites (points), and a distance measure `d`, the nearest neighbor searching problem is to build a data structure so that given a query point `q`, the site nearest to `q` can be found quickly. This paper gives a data structure for this problem; the data structure is built using the distance function as a "black box". The structure is able to speed up nearest neighbor searching in a variety of settings, for example: points in low-dimensional or structured Euclidean space, strings under Hamming and edit distance, and bit vector data from an OCR application. The data structures are observed to need linear space, with a modest constant factor. The preprocessing time needed per site is observed to match the query time. The data structure can be viewed as an application of a "kd-tree" approach in the metric space setting, using Voronoi regions of a subset in place of axis-aligned boxes.

This paper gives an algorithm for solving linear systems, using a randomized version of incomplete `LU` factorization together with iterative improvement. The factorization uses Gaussian elimination with partial pivoting, and preserves sparsity during factorization by randomized rounding of the entries. The resulting approximate factorization is then applied to estimate the solution. This simple technique, combined with iterative improvement, is demonstrated to be effective for a range of linear systems. When applied to medium-sized sample matrices for PDEs, the algorithm is qualitatively like multigrid: the work per iteration is typically linear in the order of the matrix, and the number of iterations to achieve a small residual is typically on the order of fifteen to twenty. The technique is also tested for a sample of asymmetric matrices from the Matrix Market, and is found to have similar behavior for many of them.

3G wireless systems such as 3G-1X, 1xEV-DO and 1xEV-DV provide support for a variety of high-speed data applications. The success of these services critically relies on the capability to ensure an adequate QoS experience to users at an affordable price. With wireless bandwidth at a premium, traffic engineering and network planning play a vital role in addressing these challenges. We present models and techniques that we have developed for quantifying the QoS perception of 1xEV-DO users generating FTP or Web browsing sessions. We show how user-level QoS measures may be evaluated by means of a Processor-Sharing model which explicitly accounts for the throughput gains from multi-user scheduling. The model provides simple analytical formulas for key performance metrics such as response times, blocking probabilities and throughput. Analytical models are especially useful for network deployment and in-service tuning purposes due to the intrinsic difficulties associated with simulation-based optimization approaches. We discuss the application of our results in the context of Ocelot, which is a Lucent tool for wireless network planning and optimization.

For 3G cellular networks, capacity is an important objective, along with coverage, when characterizing the performance of high-data-rate services. In live networks, the effective network capacity heavily depends on the degree that the traffic load is balanced over all cells, so changing traffic patterns demand dynamic network reconfiguration to maintain good performance. Using a four-cell sample network, and antenna tilt, cell power level and pilot fraction as adjustment variables, we study the competitive character of network coverage and capacity in such a network optimization process, and how it compares to the CDMA-intrinsic coverage-capacity tradeoff driven by interference. We find that each set of variables provides its distinct coverage-capacity tradeoff behavior with widely varying and application-dependent performance gains. The study shows that the impact of dynamic load balancing highly depends on the choice of the tuning variable as well as the particular tradeoff range of operation.

We give a simple analytic model of coverage probability for CDMA cellular phone systems under lognormally distributed shadow fading. Prior analyses have generally considered the coverage probability of a single antenna; here we consider the probability of coverage by an ensemble of antennas, using some independence assumptions, but also modeling a limited form of dependency among the antenna fades. We use the Fenton-Wilkinson approach of approximating the external interference `I_0` as lognormally distributed. We show that our model gives a coverage probability that is generally within a few percent of Monte Carlo estimates, over a wide regime of antenna strengths and other relevant parameters.

Given a set of points `P\subset R^d` and value `\epsilon>0`, an `\epsilon`-core-set `S \subset P` has the property that the smallest ball containing `S` is within `\epsilon` of the smallest ball containing `P`. This paper shows that any point set has an `\epsilon`-core-set of size `|~1/\epsilon~|`, and this bound is tight in the worst case. A faster algorithm given here finds an `\epsilon`-core-set of size at most `2/\epsilon`. These results imply the existence of small core-sets for solving approximate `k`-center clustering and related problems. The sizes of these core-sets are considerably smaller than the previously known bounds, and imply faster algorithms; one such algorithm needs `O(d n/\epsilon+(1/\epsilon)^{5})` time to compute an `\epsilon`-approximate minimum enclosing ball (1-center) of `n` points in `d` dimensions. A simple gradient-descent algorithm is also given, for computing the minimum enclosing ball in `O(d n / \epsilon^{2})` time. This algorithm also implies slightly faster algorithms for computing approximately the smallest radius `k`-flat fitting a set of points.

Given a set of points `P\subset \RR^d` and value `\epsilon>0`, an $\epsilon$-core-set `S \subset P` has the property that the smallest ball containing `S` is within `\epsilon` of the smallest ball containing `P`. This paper shows that any point set has an `\epsilon`-core-set of size `|~1/\epsilon~|`, and this bound is tight in the worst case. Some experimental results are also given, comparing this algorithm with a previous one, and with a more powerful, but slower one.

We present an algorithm based on lattice reduction for the fast decoding of diagonal differential modulation across multiple antenna. While the complexity of the maximum likelihood algorithm is exponential both in the number of antenna and the rate, the complexity of our approximate lattice algorithm is polynomial in the number of antennas and the rate. We show that the error performance of our lattice algorithm is very close to the maximum likelihood algorithm.

Given a set `S` of `n` sites (points), and a distance measure `d`, the nearest neighbor searching problem is to build a data structure so that given a query point `q`, the site nearest to `q` can be found quickly. This paper gives data structures for this problem when the sites and queries are in a metric space. One data structure, `D(S)`, uses a divide-and-conquer recursion. The other data structure, `M(S,Q)`, is somewhat like a skiplist. Both are simple and implementable. The data structures are analyzed when the metric space obeys a certain sphere-packing bound, and when the sites and query points are random and have distributions with an exchangeability property. This property implies, for example, that query point `q` is a random element of `S\cup\{q\}`. Under these conditions, the preprocessing and space bounds for the algorithms are close to linear in `n`. They depend also on the sphere-packing bound, and on the logarithm of the distance ratio `\upsilon(S)` of `S`, the ratio of the distance between the farthest pair of points in `S` to the distance between the closest pair. The data structure `M(S,Q)` requires as input data an additional set `Q`, taken to be representative of the query points. The resource bounds of `M(S,Q)` have a dependence on the distance ratio of `S\cup Q`. While `M(S,Q)` can return wrong answers, its failure probability can be bounded, and is decreasing in a parameter `K`. Here `K\leq |Q|/n` is chosen when building `M(S,Q)`. The expected query time for `M(S,Q)` is `O(K\log n)\log\upsilon(S\cup Q)`, and the resource bounds increase linearly in `K`. The data structure `D(S)` has expected `O(\log n)^{O(1)}` query time, for fixed distance ratio. The preprocessing algorithm for `M(S,Q)` can be used to solve the all-nearest-neighbor problem for `S` in `O(n(\log n)^2(\log\upsilon(S))^2)` expected time.

The technique of randomized incremental construction allows a variety of geometric structures to be built quickly. This note shows that once such a structure is built, it is possible to store the geometric input data for it so that the structure can be built again by a randomized algorithm even more quickly. Except for the randomization, this generalizes the technique of Snoeyink and van Kreveld that applies to planar problems.

Given points moving with constant, but possibly different, velocities, the minimum moving diameter problem is to find the minimum, over all time, of the maximum distance between a pair of points at each moment. This note gives a randomized algorithm requiring `O(n \log n )` expected time for this problem, in two and three dimensions. Also briefly noted is a randomized `O(n\log n\log\log n)` expected-time algorithm for the discrete 1-center problem in three dimensions; in this problem, a member `p` of a set `S` of points is desired, whose maximum distance to `S` is minimum over all points of `S`.

A simple idea for speeding up the computation of extrema of a partially ordered set turns out to have a number of interesting applications in geometric algorithms; the resulting algorithms generally replace an appearance of the input size `n` in the running time by an output size `A\leq n`. In particular, the `A` coordinate-wise minima of a set of `n` points in `R^d` can be found by an algorithm needing `O(nA)` time. Given `n` points uniformly distributed in the unit square, the algorithm needs `n+O(n^{5/8})` point comparisons on average. Given a set of `n` points in `R^d`, another algorithm can find its `A` extreme points in `O(nA)` time. Thinning for nearest-neighbor classification can be done in time `O(n\log n)\sum_i A_i n_i`, finding the `A_i` irredundant points among `n_i` points for each class `i`, where `n=\sum_i n_i` is the total number of input points. This sharpens a more obvious `O(n^3)` algorithm, which is also given here. Another algorithm is given that needs `O(n)` space to compute the convex hull of `n` points in `O(nA)` time. Finally, a new randomized algorithm finds the convex hull of `n` points in `O(n\log A)` expected time, under the condition that a random subset of the points of size `r` has expected hull complexity `O(r)`. All but the last of these algorithms has polynomial dependence on the dimension `d`, except possibly for linear programming.

This paper gives an algorithm for polytope covering: let `L` and `U` be sets of points in `R^d`, comprising `n` points altogether. A cover for `L` from `U` is a set `C\subset U` with `L` a subset of the convex hull of `C`. Suppose `c` is the size of a smallest such cover, if it exists. The randomized algorithm given here finds a cover of size no more than `c(\rboundp)`, for `c` large enough. The algorithm requires `O(c^2n^{1+\delta})` expected time. (In this paper, `\delta` will denote any fixed value greater than zero.) More exactly, the time bound is

`O(cn^{1+\delta}+c(nc)^{ 1/(1+\gamma/(1+\delta))}),`

where `\gamma\equiv 1/|__d/2__|`. The previous best bounds were `c O(\log n)` cover size in `O(n^d)` time [MiS]. A variant algorithm is applied to the problem of approximating the boundary of a polytope with the boundary of a simpler polytope. For an appropriate measure, an approximation with error `\epsilon` requires `c=O(1/\epsilon)^{(d-1)/2}` vertices, and the algorithm gives an approximation with `c(\apboundp)` vertices. The algorithms apply ideas previously used for small-dimensional linear programming. The final result here applies polytope approximation to the the post office problem: given `n` points (called sites) in `d` dimensions, build a data structure so that given a query point `q`, the closest site to `q` can be found quickly. The algorithm given here is given also a relative error bound `\epsilon`, and depends on a ratio `\rho`, which is no more than the ratio of the distance between the farthest pair of sites to the distance between the closest pair of sites. The algorithm builds a data structure of size `O(n(\log\rho)/\epsilon^{d/2}` in time `O(n^2(\log\rho))/\epsilon^d`. With this data structure, closest-point queries can be answered in `O(\log n)/\epsilon^{d/2}` time.

We give a practical and provably good Monte Carlo algorithm for approximating center points. Let `P` be a set of `n` points in `\R^d`. A point `c \in \R^d` is a `\beta`-center point of `P` if every closed halfspace containing `c` contains at least `\beta n` points of `P`. Every point set has a `1/(d+1)`-center point; our algorithm finds an `\Omega(1/d^2)`-center point with high probability. Our algorithm has a small constant factor and is the first approximate center point algorithm whose complexity is subexponential in `d`. Moreover, it can be optimally parallelized to require `O(\log^2d\log\log n)` time. Our algorithm has been used in mesh partitioning methods and can be used in constructing high breakdown estimators for multivariate datasets in statistics. It has the potential to improve results in practice for constructing weak `\epsilon`-nets. We derive a variant of our algorithm whose time bound is fully polynomial in `d` and linear in `n`, and show how to combine our approach with previous techniques to compute high quality center points more quickly.

This talk sketches: a convenient method for computing the volumes of Voronoi regions; a proof that "area-stealing" natural neighbor interpolation works; a scheme for smoother natural neighbor interpolation alternative to Sibson's method; the interpolation scheme used in the "finite volume element" method; and the observation that the minimax piecewise-linear interpolant of a convex function is the (lower) convex hull.

The problem of evaluating the sign of the determinant of a small matrix arises in many geometric algorithms. Given an `n\times n` matrix `A` with integer entries, whose columns are all smaller than `M` in Euclidean norm, the algorithm given here evaluates the sign of the determinant `\det A` exactly. The algorithm requires an arithmetic precision of less than `1.5n+2\lg M` bits. The number of arithmetic operations needed is `O(n^3)+O(n^2)\log\OD(A)/\beta`, where `\OD(A)|\det A|` is the product of the lengths of the columns of `A`, and `\beta` is the number of "extra" bits of precision,

`\min\{\lg(1/bb u)-1.1n-2\lg n-2,\lg N - \lg M - 1.5n - 1\},`

where `bb u` is the roundoff error in approximate arithmetic, and `N` is the largest representable integer. Since `\OD(A)\leq M^n`, the algorithm requires `O(n^3\lg M)` time, and `O(n^3)` time when `\beta=\Omega(\log M)`.

This paper shows that the `i`-level of an arrangement of hyperplanes in `E^d` has at most `((i+d-1), (d-1))` local minima. This bound follows from ideas previously used to prove bounds on `(\le k)`-sets. Using linear programming duality, the Upper Bound Theorem is obtained as a corollary, giving yet another proof of this celebrated bound on the number of vertices of a simple polytope in `E^d` with `n` facets.

This paper surveys some of the applications of randomization to computational and combinatorial geometry. Randomization provides a general way to divide-and-conquer geometric problems, and gives a simple incremental approach to building geometric structures. The paper discusses closest-point problems, convex hulls, Voronoi diagrams, trapezoidal diagrams of line segments, linear programming in small dimension, range queries, and bounds for point-line incidences and for `(\le k)`-sets. Relations to the Vapnik-Chervonenkis dimension, PAC-learnability of geometric concepts, and the Hough transform are briefly noted.

We prove four results on randomized incremental constructions (RICs):

• an analysis of the expected behavior under insertion and deletions,
• a fully dynamic data structure for convex hull maintenance in arbitrary dimensions,
• a tail estimate for the space complexity of RICs,
• a lower bound on the complexity of a game related to RICs.

We describe randomized parallel algorithms for building trapezoidal diagrams of line segments in the plane. The algorithms are designed for a CRCW PRAM. For general segments, we give an algorithm requiring optimal `O(A+n\log n)` expected work and optimal `O(\log n)` time, where `A` is the number of intersecting pairs of segments. If the segments form a simple chain, we give an algorithm requiring optimal `O(n)` expected work and `O(\log n\log\log n\log^**n)` expected time, and a simpler algorithm requiring `O(n\log^**n)` expected work. The serial algorithm corresponding to the latter is among the simplest known algorithms requiring `O(n\log^**n)` expected operations. For a set of segments forming `K` chains, we give an algorithm requiring `O(A+n\log^**n+K\log n)` expected work and `O(\log n\log\log n\log^** n)` expected time. The parallel time bounds require the assumption that enough processors are available, with processor allocations every `\log n` steps.

This paper gives a partitioning scheme for the geometric, planar traveling salesman problem, under the Euclidean metric: given a set `S` of `n` points in the plane, find a shortest closed tour (path) visiting all the points. The scheme employs randomization, and gives a tour that can be expected to be short, if `S` satisfies the condition that a random subset `R\subset S` has on average a tour much shorter than an optimal tour of `S`. This condition holds for points independently, identically distributed in the plane, for example, for which a tour within `1+\epsilon` of shortest can be found in expected time `nk^2 2^k`, where `k=O(\log\log n)^3/\epsilon^2`. One algorithm employed in the scheme is of interest in its own right: when given a simple polygon `P`, it finds a Steiner triangulation of the interior of `P`. If `P` has `n` sides and perimeter `L_P`, the edges of the triangulation have total length `L_PO(\log n)`. If this algorithm is applied to a simple polygon induced by a minimum spanning tree of a point set, the result is a Steiner triangulation of the set with total length within a factor of `O(\log n)` of the minimum possible.

This paper gives an algorithm for solving linear programming problems. For a problem with `n` constraints and `d` variables, the algorithm requires an expected

`O(d^2n)+(\log n)O(d)^{d/2+O(1)} +O(d^4\sqrt{n}\log n)`

arithmetic operations, as `n\rightarrow\infty`. The constant factors do not depend on `d`. Also, an algorithm is given for integer linear programming. Let `\varphi` bound the number of bits required to specify the rational numbers defining an input constraint or the objective function vector. Let `n` and `d` be as before. Then the algorithm requires expected

`O(2^ddn+8^dd\sqrt{n\ln n}\ln n) +d^{O(d)}\varphi\ln n`

operations on numbers with `d^{O(1)}\varphi` bits, as `n->oo`, where the constant factors do not depend on `d` or `\varphi`. The expectations are with respect to the random choices made by the algorithms, and the bounds hold for any given input. The technique can be extended to other convex programming problems. For example, an algorithm for finding the smallest sphere enclosing a set of `n` points in `E^d` has the same time bound.

We present an algorithm that triangulates a simple polygon on `n` vertices in `O(n log ^** n)` expected time. The algorithm uses random sampling on the input, and its running time does not depend on any assumptions about a probability distribution from which the polygon is drawn.

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 requires `O(A+n\log n)` expected time, where `A` is the number of intersecting pairs reported. The algorithm requires `O(n)` space in the worst case. Another algorithm computes the convex hull of `n` points in `E^d` in `O(n\log n)` expected time for `d=3`, and `O(n^{|__d/2__|})` expected time for `d>3`. The algorithm also gives fast expected times for random input points. Another algorithm computes the diameter of a set of `n` points in `E^3` in `O(n\log n)` expected time, and on the way computes the intersection of `n` unit balls in `E^3`. We show that `O(n\log A)` expected time suffices to compute the convex hull of `n` points in `E^3`, where `A` 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 `(\le k)`-sets, which are certain halfspace partitions of point sets, and give a simple proof of Lee's bounds for high order Voronoi diagrams.

We describe an algorithm for finding a minimum spanning tree of the weighted complete graph induced by a set of `n` points in Euclidean `d`-space. The algorithm requires nearly linear expected time for points that are independently uniformly distributed in the unit `d`-cube. The first step of the algorithm is the spiral search procedure described by Bentley, Weide, and Yao [BWY] for finding a supergraph of the MST that has `O(n)` edges. (The constant factor in the bound depends on `d`.) The next step is that of sorting the edges of the supergraph by weight using a radix distribution, or "bucket," sort. These steps require linear expected time. Finally, Kruskal's algorithm is used with the sorted edges, requiring `O(n\alpha(cn,n))` time in the worst case, with `c>6`. Since the function `\alpha(cn,n)` grows very slowly, this step requires linear time for all practical purposes. This result improves the previous best `O(n\log\log^**n)`, and employs a much simpler algorithm. Also, this result demonstrates the robustness of bucket sorting, which requires `O(n)` expected time in this case despite the probability dependency between the edge weights.

The problem of finding a rectilinear shortest path amongst obstacles may be stated as follows: Given a set of obstacles in the plane find a shortest rectilinear (`L_1`) path from a point `s` to a point `t` which avoids all obstacles. The path may touch an obstacle but may not cross an obstacle. We study the rectilinear shortest path problem for the case where the obstacles are non-intersecting simple polygons, and present an `O( n log^2 n )` algorithm for finding such a path, where `n` is the number of vertices of the obstacles. This algorithm requires `O(n log n )` space. Another algorithm is given that requires `O(n ( log n ) ^{ 3/2} )` time and space. We also study the case of rectilinear obstacles in three dimensions, and show that `L_1` shortest paths can be found in `O( n^2 log^ 3 n )` time.

This paper gives approximation algorithms for solving the following motion planning problem: Given a set of polyhedral obstacles and points `s` and `t`, find a shortest path from `s` to `t` that avoids the obstacles. The paths found by the algorithms are piecewise linear, and the length of a path is the sum of the lengths of the line segments making up the path. Approximation algorithms will be given for versions of this problem in the plane and in three-dimensional space. The algorithms return an `\epsilon`-short path, that is, a path with length within `(1+\epsilon)` of shortest. Let `n` be the total number of faces of the polyhedral obstacles, and `\epsilon` a given value satisfying `0<\epsilon\leq\pi`. The algorithm for the planar case requires `O(n\log n)/\epsilon` time to build a data structure of size `O(n/\epsilon)`. Given points `s` and `t`, an `\epsilon`-short path from `s` to `t` can be found with the use of the data structure in time `O(n/\epsilon+n\log n)`. The data structure is associated with a new variety of Voronoi diagram. Given obstacles `S\subset E^3` and points `s,t\in E^3`, an `\epsilon`-short path between `s` and `t` can be found in

`O(n^2\lambda(n)\log(n/\epsilon)/\epsilon^4 +n^2\log n\rho\log(n\log \rho))`

time, where `\rho` is the ratio of the length of the longest obstacle edge to the distance between `s` and `t`. The function `\lambda(n)=\alpha(n)^{O(\alpha(n)^{O(1)})}`, where the `\alpha(n)` is a form of inverse of Ackermann's function. For `\log(1/\epsilon)` and `\log\rho` that are `O(\log n)`, this bound is `O(n^2(\log^2n)\lambda(n)/\epsilon^4)`.

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 requires `O(s ^ {d + \epsilon})` expected preprocessing time to build a search structure for an arrangement of `s` hyperplanes in `d` 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 point `p`, the cell of the arrangement containing `p` can be found in `O( \log s)` worst-case time. (The bound holds for any fixed `\epsilon >0`, with the constant factors dependent on `d` and `\epsilon`.) 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 expected `O(n ^{|__d/2__|} )` time, where `n` is the total number of vertices of the two polytopes. This matches previous results [DoK] for the case `d=3` and extends them. Another algorithm samples points in the plane to determine their order `k` Voronoi diagram, and requires expected `O(s^{1+\epsilon}k)` time for `s` points. (It is assumed that no four of the points are cocircular.) This sharpens the bound `O(sk ^ 2 \log s)` for Lee's algorithm [Lee], and `O(s ^ 2 \log s + k(s-k) \log ^ 2 s)` for Chazelle and Edelsbrunner's algorithm [ChE]. Finally, random sampling is used to show that any set of `s` points in `E^3` has `O(sk^2\log ^ 8 s / ( \log \log s) ^ 6 )` distinct `j`-sets with `j\leq k`. (For `S \subset E^d`, a set `S'\subset S` with `|S'|=j` is a `j`-set of `S` if there is a halfspace `h^+` with `S'=S \cap h^+`.) This sharpens with respect to `k` the previous bound `O(s k ^ 5 )` [ChP]. The proof of the bound given here is an instance of a "probabilistic method" [ErS].

An algorithm for closest-point queries is given. The problem is this: given a set `S` of `n` points in `d`-dimensional space, build a data structure so that given an arbitrary query point `p`, a closest point in `S` to `p` can be found quickly. The measure of distance is the Euclidean norm. This is sometimes called the post-office problem [Kn]. The new data structure will be termed an RPO tree, from Randomized Post Office. The expected time required to build an RPO tree is `O(n^{|~d/2~| (1+ \epsilon )} )`, for any fixed `\epsilon > 0`, and a query can be answered in `O(\log n )` worst-case time. An RPO tree requires `O(n^{|~d/2~| (1+ \epsilon )} )` space in the worst case. The constant factors in these bounds depend on `d` and `\epsilon`. The bounds are average-case due to the randomization employed by the algorithm, and hold for any set of input points. This result approaches the `\Omega(n^{|~d/2~|} )` worst-case time required for any algorithm that constructs the Voronoi diagram of the input points, and is a considerable improvement over previous bounds for `d>3`. The main step of the construction algorithm is the determination of the Voronoi diagram of a random sample of the sites, and the triangulation of that diagram.

This dissertation reports a variety of new algorithms for solving closest-point problems.  The input to these algorithms is a set or sets of points in `d`-dimensional space, with an associated `L_p` metric. The problems considered are:

• The all nearest neighbors problem.  For point set `A`, find the nearest neighbors in `A` of each point in `A`.
• The nearest foreign neighbor problem.  For point sets `A` and `B`, find the closest point in `B` to each point in `A`.
• The geometric minimum spanning tree problem.  For point set `A`, find the minimum spanning tree for the complete weighted undirected graph associated with `A`, where the vertices of the graph correspond to the points of `A`, and the weight of an edge is the distance between the points defining the edge.

These problems arise in routing, statistical classification, data compression, and other areas.  Obvious algorithms for them require a running time quadratic in `n`, the number of points in the input.  In many cases they can be solved with algorithms requiring `O(n log ^{O(1)} n)` time.

In this work, approximation algorithms for some cases of these problems have been found.  For example, for the minimum spanning tree problem with the `L_1` metric, an algorithm has been devised that requires `O(n log^d (1/rho))` time to find a spanning tree with weight within `1+rho` of the minimum.  Several other algorithms have been found with time bounds dependent on `log(1/rho)` for attaining error `rho`.

Algorithms have also been found that require linear expected time, for independent identically distributed random input points with a probability density function satisfying weak conditions.  One such algorithm depends on the fact that under certain conditions, values that are identically distributed, but dependent, can be bucket sorted in linear expected time.

An algorithm has been found for the all nearest neighbors problem that requires `O(n log n)` expected time for any input set of points, where the expectation is on the random sampling performed by the algorithm.  This algorithm involves the construction of a new data structure, a compressed form of digital trie.

We present new algorithms for the all nearest neighbors problem:

Given a set `S` of `n` sites (points) in `d`-dimensional space, find the nearest neighbors in set `S` of each site in `S`.

Our results:

• An algorithm for solving the all nearest neighbors problem in `O(n\log\sigma)` time, where `\sigma` is the ratio of the distance between the farthest pair of sites to the distance between the closest pair of sites. A similar algorithm is described for finding all `k`'th nearest neighbors.
• An algorithm for building a celltree, a compressed form of digital trie, in `O(n\log n)` probabilistic time. The logarithm, floor, and bitwise exclusive-or functions are assumed available at unit cost.
• An algorithm for solving the all nearest neighbors problem in `O(n)` worst-case time, given a celltree for the sites.
• An algorithm for building a celltree in linear expected time, assuming the sites are independently identically distributed random variables, with an unknown probability density function obeying some very weak conditions.