Complexity of Computing the Local Dimension of a Semialgebraic Set

The paper describes several algorithms related to a problem of computing the local dimension of a semialgebraic set. Let a semialgebraic set V be defined by a system of k inequalities of the formf  ≥  0 with f  R [ X₁, ,X_n ], deg(f)  < d , andx   V . An algorithm is constructed for computing the dimension of the Zariski tangent space to V at x in time (kd)^O(n). Let x belong to a stratum of codimension l_xin V with respect to a smooth stratification ofV . Another algorithm computes the local dimension dim_x(V) with the complexity (k(l_x +  1)d)^O(l_x2n). Ifl  = max_x  Vl_x, and for every connected component the local dimension is the same at each point, then the algorithm computes the dimension of every connected component with complexity (k(l +  1)d)^O(l2n). If V is a real algebraic variety defined by a system of equations, then the complexity of the algorithm is less thankd^O(l2n) , and the algorithm also finds the dimension of the tangent space to V at x in time kd^O(n). Whenl is fixed, like in the case of a smooth V , the complexity bounds for computing the local dimension are (kd)^O(n)andkd^O(n) respectively. A third algorithm finds the singular locus ofV in time (kd)^O(n2).     

Exact Upper Bound on the Mean of the Product of Many Random Variables with Known Expectations

In practice, in addition to the intervals x _i = [x _i , x _i] of possible values of inputs x ₁, ..., x _n, we sometimes also know their means E _i. For such cases, we provide an explicit exact (= best possible) upper bound for the mean of the product x ₁ ... x _n of positive values x _i.     

Exponential lower bounds for depth three Boolean circuits

We consider the class of unbounded fan-in depth three Boolean circuits, for which the bottom fan-in is limited by k and the top gate is an OR. It is known that the smallest such circuit computing the parity function has gates (for k = O(n ^1/2)) for some , and this was the best lower bound known for explicit (P-time computable) functions. In this paper, for k = 2, we exhibit functions in uniform NC ¹ that require size depth 3 circuits. The main tool is a theorem that shows that any circuit on n variables that accepts a inputs and has size s must be constant on a projection (subset defined by equations of the form x _i = 0, x _i = 1, x _i = x _j or x _i = ) of dimension at least log(a/s)log n. Received: April 1, 1997.     

Computing Population Variance and Entropy under Interval Uncertainty: Linear-Time Algorithms

In statistical analysis of measurement results it is often necessary to compute the range of the population variance when we only know the intervals of possible values of the x _i . While can be computed efficiently, the problem of computing is, in general, NP-hard. In our previous paper “Population Variance under Interval Uncertainty: A New Algorithm” (Reliable Computing 12 (4) (2006), pp. 273–280) we showed that in a practically important case we can use constraints techniques to compute in time O(n · log(n)). In this paper we provide new algorithms that compute (in all cases) and (for the above case) in linear time O(n). Similar linear-time algorithms are described for computing the range of the entropy when we only know the intervals of possible values of probabilities p _i . In general, a statistical characteristic ƒ can be more complex so that even computing ƒ can take much longer than linear time. For such ƒ, the question is how to compute the range in as few calls to ƒ as possible. We show that for convex symmetric functions ƒ, we can compute in n calls to ƒ.     

Empirical Bayes estimation of the scale parameter in a Pareto distribution

Let f(xθ) = αθ^αx^−(α+1)I(x>θ) be the pdf of a Pareto distribution with known shape parameter α>0, and unknown scale parameter θ. Let {(X_i, θ_i)} be a sequence of independent random pairs, where X_i's are independent with pdf f(xα_i), and θ_i are iid according to an unknown distribution G in a class of distributions whose supports are included in an interval (0, m), where m is a positive finite number. Under some assumption on the class and squared error loss, at (n + 1)th stage we construct a sequence of empirical Bayes estimators of θ_n+1 based on the past n independent observations X₁,…, X_n and the present observation X_n+1. This empirical Bayes estimator is shown to be asymptotically optimal with rate of convergence O(n^−1/2). It is also exhibited that this convergence rate cannot be improved beyond n^−1/2 for the priors in class .     

Mosaic-Skeleton approximations

If a matrix has a small rank then it can be multiplied by a vector with many savings in memory and arithmetic. As was recently shown by the author, the same applies to the matrices which might be of full classical rank but have a smallmosaic rank. The mosaic-skeleton approximations seem to have imposing applications to the solution of large dense unstructured linear systems. In this paper, we propose a suitable modification of brandt's definition of an asymptotically smooth functionf(x,y). Then we considern×n matricesA _n =[f(x _i ⁽ⁿ⁾ ,y _j ⁽ⁿ⁾ )] for quasiuniform meshes {x _i ⁽ⁿ⁾ } and {y _j ⁽ⁿ⁾ } in some bounded domain in them-dimensional space. For such matrices, we prove that the approximate mosaic ranks grow logarithmically inn. From practical point of view, the results obtained lead immediately toO(n logn) matrix-vector multiplication algorithms. The work was supported in part by the Russian Fund of Basic Research and also by the Volkswagen-Stiftung.     

Interpolation that leads to the narrowest intervals and its application to expert systems and intelligent control

In many real-life situations, we want to reconstruct the dependencyy=f(x ₁,…, x_n) from the known experimental resultsx _i ^(k) , y^(k). In other words, we want tointerpolate the functionf from its known valuesy ^(k)=f(x ₁ ^(k) ,…, x _n ^(k) ) in finitely many points $\bar x^{(k)} = (x_1^{(k)} , \ldots ,x_n^{(k)} )$ , 1≤k≤N There are many functions that go through given points. How to choose one of them? The main goal of findingf is to be able to predicty based onx _i. If we getx _i from measurements, then usually, we only getintervals that containx _i. As a result of applyingf, we get an interval y of possible values ofy. It is reasonable to choosef for which the resulting interval is the narrowest possible. In this paper, we formulate this choice problem in mathematical terms, solve the corresponding problem for several simple cases, and describe the application of these solutions to intelligent control.     

Dynamic Hotlinks

Consider a directed rooted tree T=(V,E) representing a collection V of n web pages connected via a set E of links all reachable from a source home page, represented by the root of T. Each web page i carries a weight w _i representative of the frequency with which it is visited. By adding hotlinks, shortcuts from a node to one of its descendants, we are interested in minimizing the expected number of steps needed to visit pages from the home page. We give the first linear time algorithm for assigning hotlinks so that the number of steps to access a page i from the root of the tree reaches the entropy bound, i.e. is at most O(log (W/w _i )) where W=∑ _i∈T w _i . The best previously known algorithm for this task runs in time O(n ²). We also give the first efficient data structure for maintaining hotlinks when nodes are added, deleted or their weights modified, in amortized time O(log (W/w _i )) per update. The data structure can be made adaptive, i.e. reaches the entropy bound in the amortized sense without knowing the weights w _i in advance.     

Bounded dilation maps of hypercubes into Cayley graphs on the symmetric group

LetG andH be graphs with |V(G)≤ |V(H)|. Iff:V(G) →V(H) is a one-to-one map, we letdilation(f) be the maximum of dist _H (f x),f(y)) over all edgesxy inG where dist _H denotes distance inH. The construction of maps fromG toH of small dilation is motivated by the problem of designing small slowdown simulations onH of algorithms that were originally designed for the networkG. LetS(n), thestar network of dimension n, be the graph whose vertices are the elements of the symmetric group of degreen, two verticesx andy being adjacent ifx o (1,i) =y for somei. That is,xy is an edge ifx andy are related by a transposition involving some fixed symbol (which we take to be 1). Also letP(n), thepancake network of dimension n, be the graph whose vertices are the elements of the symmetric group of degreen, two verticesx andy being adjacent if one can be obtained from the other by reversing some prefix. That is,xy is an edge ifx andy are related byx o (1,i(2,i-1) ⋯ ([i/2], [i/2]) =y. The star network (introduced in [AHK]) has nice symmetry properties, and its degree and diameter are sublogarithmic as functions of the number of vertices, making it compare favorably with the hypercube network. These advantages ofS(n) motivate the study of how well it can simulate other parallel computation networks, in particular, the hypercube. The concern of this paper is to construct low dilation maps of hypercube networks into star or pancake networks. Typically in such problems, there is a tradeoff between keeping the dilationsmall and simulating alarge hypercube. Our main result shows that at the cost ofO (1) dilation asn→ ∞, one can embed a hypercube of near optimum dimension into the star or pancake networksS(n) orP(n). More precisely, lettingQ (d) be the hypercube of dimensiond, our main theorem is stated below. For simplicity, we state it only in the special case when the star network dimension is a power of 2. A more general result (applying to star networks of arbitrary dimension) is obtained by a simple interpolation. This author's research was done during the Spring Semester 1991, as a visiting professor in the Department of Mathematics and Statistics at Miami University.     

Diagonal polynomials and diagonal orders on multidimensional lattices

HereR andN denote the real numbers and the nonnegative integers, respectively. Alsos(x)=x ₁+···+x _n whenx=(x ₁, …,x _n) inR ⁿ. A mapf:R ⁿ →R is call adiagonal function of dimensionn iff|N ⁿ is a bijection ontoN and, for allx, y inN ⁿ, f(x)<f(y) whens(x)<s(y). Morales and Lew [6] constructed 2 ⁿ⁻² inequivalent diagonal polynomial functions of dimensionn for eachn>1. Here we use new combinatorial ideas to show that numberd _n of such functions is much greater than 2 ⁿ⁻² forn>3. These combinatorial ideas also give an inductive procedure to constructd _n+1 diagonal orderings of {1, …,n}.     

Fast Algorithms for the Density Finding Problem

We study the problem of finding a specific density subsequence of a sequence arising from the analysis of biomolecular sequences. Given a sequence A=(a ₁,w ₁),(a ₂,w ₂),…,(a _n ,w _n ) of n ordered pairs (a _i ,w _i ) of numbers a _i and width w _i >0 for each 1≤i≤n, two nonnegative numbers ℓ, u with ℓ≤u and a number δ, the Density Finding Problem is to find the consecutive subsequence A(i ^*,j ^*) over all O(n ²) consecutive subsequences A(i,j) with width constraint satisfying ℓ≤w(i,j)=∑ _r=i ^j w _r ≤u such that its density is closest to δ. The extensively studied Maximum-Density Segment Problem is a special case of the Density Finding Problem with δ=∞. We show that the Density Finding Problem has a lower bound Ω(nlog n) in the algebraic decision tree model of computation. We give an algorithm for the Density Finding Problem that runs in optimal O(nlog n) time and O(nlog n) space for the case when there is no upper bound on the width of the sequence, i.e., u=w(1,n). For the general case, we give an algorithm that runs in O(nlog ² m) time and O(n+mlog m) space, where and w _min=min  _r=1 ⁿ w _r . As a byproduct, we give another O(n) time and space algorithm for the Maximum-Density Segment Problem. Grants NSC95-2221-E-001-016-MY3, NSC-94-2422-H-001-0001, and NSC-95-2752-E-002-005-PAE, and by the Taiwan Information Security Center (TWISC) under the Grants NSC NSC95-2218-E-001-001, NSC95-3114-P-001-002-Y, NSC94-3114-P-001-003-Y and NSC 94-3114-P-011-001.     

Checking If There Exists a Monotonic Function That Is Consistent with the Measurements: An Efficient Algorithm

In many problems in science and engineering ranging from astrophysics to geosciences to financial analysis, we know that a physical quantity y depends on the physical quantity x, i.e., y = f(x) for some function f(x), and we want to check whether this dependence is monotonic. Specifically, finitely many measurements of x_i and y = f(x) have been made, and we want to check whether the results of these measurements are consistent with the monotonicity of f(x). An efficient parallelizable algorithm is known for solving this problem when the values x_i are known precisely, while the values y_i are known with interval uncertainty. In this paper, we extend this algorithm to a more general (and more realistic) situation when both x_i and y_i are known with interval uncertainty.     

A Primal-Dual Approximation Algorithm for Partial Vertex Cover: Making Educated Guesses

We study the partial vertex cover problem. Given a graph G=(V,E), a weight function w:V→R ⁺, and an integer s, our goal is to cover all but s edges, by picking a set of vertices with minimum weight. The problem is clearly NP-hard as it generalizes the well-known vertex cover problem. We provide a primal-dual 2-approximation algorithm which runs in O(nlog n+m) time. This represents an improvement in running time from the previously known fastest algorithm. Our technique can also be used to get a 2-approximation for a more general version of the problem. In the partial capacitated vertex cover problem each vertex u comes with a capacity k _u . A solution consists of a function x:V→ℕ₀ and an orientation of all but s edges, such that the number of edges oriented toward vertex u is at most x _u k _u . Our objective is to find a cover that minimizes ∑ _v∈V x _v w _v . This is the first 2-approximation for the problem and also runs in O(nlog n+m) time. Research supported by NSF Awards CCR 0113192 and CCF 0430650, and the University of Maryland Dean’s Dissertation Fellowship.     

Finding the Extrema of a Distributed Multiset

We consider the problem of finding the extrema of a distributed multiset in a ring, that is, of determining the minimum and the maximum values,x_minandx_max, of a multisetX= {x₀,x₂, ...,x_n−1} whose elements are drawn from a totally ordered universeUand stored at thenentities of a ring network. This problem is unsolvable if the ring size is not known to the entities, and it has complexity Θ(n²) in the case of asynchronous rings of known size. We show that, in synchronous rings of known size, this problem can always be solved inO((c+ logn) ·n) bits andO(n·c·x^1/c) time for any integerc> 0, wherex= Max{|x_min|, |x_max|}. The previous solutions requiredO(n²) bits and the same amount of time. Based on these results, we also present a bit-optimal solution to the problem of finding the multiplicity of the extrema.     

On ACC

We show that every languageL in the class ACC can be recognized by depth-two deterministic circuits with a symmetric-function gate at the root and AND gates of fan-in log ^O(1) n at the leaves, or equivalently, there exist polynomialsp _n (x ₁ ,..., x _n ) overZ of degree log ^O(1) n and with coefficients of magnitude and functionsh _n :Z{0,1} such that for eachn and eachx{0,1} ⁿ ,XL ^(x) =h _n (p _n (x ₁ ,...,x _n )). This improves an earlier result of Yao (1985). We also analyze and improve modulus-amplifying polynomials constructed by Toda (1991) and Yao (1985).     

Nonlocal quasi-Newton method for solving convex variational inequalities

Conclusion In the optimization problem [f ₀(x)│h_i(x)<-0,i=1,…,l] relaxation of the functionf ₀(x)+Nh⁺(x) does not produce, as we know [6, 7], α_k=1 in Newton's method with the auxiliary problem (5), (6), whereF(x)=f _0′(x). For this reason, Newton type methods based on relaxation off ₀(x)+Nh⁺(x) are not superlinearly convergent (so-called Maratos effect). The results of this article indicate that if (F(x)=f _0′(x), then replacement of the initial optimization problem with a larger equivalent problem (7) eliminates the Maratos effect in the proposed quasi-Newton method. This result is mainly of theoretical interest, because Newton type optimization methods in the space of the variablesx ∈R ⁿ are less complex. However to the best of our knowledge, the difficulties with nonlocal convergence arising in these methods (choice of parameters, etc.) have not been fully resolved [10, 11]. The discussion of these difficulties and comparison with the proposed method fall outside the scope of the present article, which focuses on solution of variational inequalities (1), (2) for the general caseF′(x)≠F′ ^T(x). Translated from Kibernetika i Sistemnyi Analiz, No. 6, pp. 78–91, November–December, 1994.     

Output-Sensitive Algorithm for Computing β-Skeletons

The β-skeleton is a measure of the internal shape of a planar set of points. We get an entire spectrum of shapes by varying the parameter β. For a fixed value of β, a β-skeleton is a geometric graph obtained by joining each pair of points whose β-neighborhood is empty. For β≥1, this neighborhood of a pair of points p _i ,p _j is the interior of the intersection of two circles of radius , centered at the points (1−β/2)p _i +(β/2)p _j and (β/2)p _i +(1−β/2)p _j , respectively. For β∈(0,1], it is the interior of the intersection of two circles of radius , passing through p _i and p _j . In this paper we present an output-sensitive algorithm for computing a β-skeleton in the metrics l ₁ and l _∞ for any β≥2. This algorithm is in O(nlogn+k), where k is size of the output graph. The complexity of the previous best known algorithm is in O(n ^5/2logn) [7]. Received April 26, 2000