Stabilized rounded addition of hierarchical matrices

The efficiency of hierarchical matrices is based on the approximate evaluation of usual matrix operations. The introduced approximation error may, however, lead to a loss of important matrix properties. In this article we present a technique which preserves the positive definiteness of a matrix independently of the approximation quality. The importance of this technique is illustrated by an elliptic mixed boundary value problem with tiny Dirichlet part. Copyright © 2007 John Wiley & Sons, Ltd.     

Linear complexity algorithms for semiseparable matrices

Linear complexity algorithms are derived for the solution of a linear system of equations with the coefficient matrix represented as a sum of diagonal and semiseparable matrices. LDU-factorization algorithms for such matrices and their inverses are also given. The case in which the solution can be efficiently update is treated separately.This work was supported in part by the U.S. Army Research Office, under Contract DAAG29-83-K-0028, and the Air Force Office of Scientific Research, Air Force Systems Command under Contract AF83-0228.     

Linear complexity algorithm for semiseparable matrices

A new linear complexity algorithm for general nonsingular semiseparable matrices is presented. For symmetric matrices whose semiseparability rank equals to 1 this algorithm leads to an explicit formula for the inverse matrix.Supported in part by the NSF Grant DMS 9306357     

The preconditioned inverse iteration for hierarchical matrices

The preconditioned inverse iteration is an efficient method to compute the smallest eigenpair of a symmetric positive definite matrix M. Here we use this method to find the smallest eigenvalues of a hierarchical matrix. The storage complexity of the data‐sparse ‐matrices is almost linear. We use ‐arithmetic to precondition with an approximate inverse of M or an approximate Cholesky decomposition of M. In general, ‐arithmetic is of linear‐polylogarithmic complexity, so the computation of one eigenvalue is cheap. We extend the ideas to the computation of inner eigenvalues by computing an invariant subspace S of (M ? μI)² by subspace preconditioned inverse iteration. The eigenvalues of the generalized matrix Rayleigh quotient μ_M(S) are the desired inner eigenvalues of M. The idea of using (M ? μI)² instead of M is known as the folded spectrum method. As we rely on the positive definiteness of the shifted matrix, we cannot simply apply shifted inverse iteration therefor. Numerical results substantiate the convergence properties and show that the computation of the eigenvalues is superior to existing algorithms for non‐sparse matrices.Copyright © 2012 John Wiley & Sons, Ltd.     

On the additive complexity of GCD and LCM matrices

In the paper, the additive complexity of matrices formed by positive integer powers of greatest common divisors and least common multiples of the indices of the rows and columns is considered. It is proved that the complexity of the n × n matrix formed by the numbers GCD^r(i, k) over the basis {x + y} is asymptotically equal to rn log²n as n→∞, and the complexity of the n × n matrix formed by the numbers LCM^r(i, k) over the basis {x + y,?x} is asymptotically equal to 2rn log₂n as n→∞.     

Expansion of random field gradients using hierarchical matrices

We present two expansions for the gradient of a random field. In the first approach, we differentiate its truncated Karhunen-Loève expansion. In the second approach, the Karhunen-Loève expansion of the random field gradient is computed directly. Both strategies require the solution of dense, symmetric matrix eigenvalue problems which can be handled efficiently by combining hierachical matrix techniques with a thick-restart Lanczos method. (© 2011 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim)     

The group-theoretic complexity of subsemigroups of boolean matrices

We find the group-theoretic complexity of many subsemigroups of the semigroup B_n of n × n Boolean matrices, including Hall matrices, reflexive matrices, fully indecomposable matrices, upper triangular matrices, row-rank-n matrices, and others.     

On the Gröbner complexity of matrices

In this paper we show that if for an integer matrix A the universal Gröbner basis of the associated toric ideal I_A coincides with the Graver basis of A, then the Gröbner complexity u(A) and the Graver complexity g(A) of its higher Lawrence liftings agree, too. In fact, if the universal Gröbner basis of I_A coincides with the Graver basis of A, then also the more general complexities u(A,B) and g(A,B) agree for arbitrary B. We conclude that for the matrices A_3×3 and A_3×4, defining the 3×3 and 3×4 transportation problems, we have u(A_3×3)=g(A_3×3)=9 and u(A_3×4)=g(A_3×4)≥27. Moreover, we prove that u(A_a,b)=g(A_a,b)=2(a+b)/gcd(a,b) for positive integers a,b and .     

A hierarchical algorithm for making sparse matrices sparser

IfA is the (sparse) coefficient matrix of linear equality constraints, for what nonsingularT is ≡TA as sparse as possible, and how can it be efficiently computed? An efficient algorithm for thisSparsity Problem (SP) would be a valuable pre-processor for linearly constrained optimization problems. In this paper we develop a two-pass approach to solve SP. Pass 1 builds a combinatorial structure on the rows ofA which hierarchically decomposes them into blocks. This determines the structure of the optimal transformation matrixT. In Pass 2, we use the information aboutT as a road map to do block-wise partial Gauss-Jordan elimination onA. Two block-aggregation strategies are also suggested that could further reduce the time spend in Pass 2. Computational results indicate that this approach to increasing sparsity produces significant net reductions in simplex solution time.     

Component commonality via hierarchical orthogonal arrays and detailing economic decision matrices

Component commonality (CC) implies products using many common parts, desensitized to the range of product applications (noise), and meeting the functionality objectives of the product line. This paper lists a nine-step methodology for developing CC and applies it to a problem. These steps utilize the major concepts of analytical modeling, economic decision matrices (EDM), quality loss functions (QLS) for variates and weighted utilities, stochastic models, finite element (FE) simulations for concurrent engineering, and statistical design of experiments (DOE) for uncertainty in either application, statistics or managerial decisions. The details of the first six steps were illustrated in a previous paper by application to a problem involving a slider link subjected to an extreme range of “noise” (various inertia/pressure loadings). Six candidate designs of steel, aluminum and titanium were generated using an analytical model and a sensitivity study. The DOE utilized Taguchi's orthogonal arrays. These designs were ranked using cost, weight, and factors of safety with respect to yielding. A refining EDM with a three-part robustness criteria selected two candidates (best was steel, followed by aluminum) considering inner noise in the managerial decisions. In the current paper, the last three steps of the nine-step methodology are applied to these two candidates in order to obtain the “optimal” part for CC. The FE stress results are used with a modified Goodman fatigue criteria, and a stochastic model is developed based upon beta (strength) and three-parameter Weibull (stress) distributions. The model is then used in a detailing EDM to determine the stochastic reliability associated with a QLS defined with respect to fatigue reliability. A “fine-tuned” aluminum candidate is shown to meet a priori reliability requirements and have low-quality losses. However, both original candidates exhibited some high-quality losses, even though such losses were acceptable in the preceding refining EDM. The authors demonstrate that this loss of quality can be prevented if a fatigue criteria is used in both the refining and detailing EDM stages of the design process and, “warranty failures” are based on stochastic rather than deterministic definitions of maximum environmental conditions.     

Construction of a canonic basis for matrices and pencils of matrices

An algorithm is offered, which with insignificant modifications permits; 1) the finding of a canonic basis of the root sub space corresponding to a prescribed eigenvalue of a matrix; 2) the finding of chains of associated vectors to the eigenvectors corresponding to a prescribed eigenvalue of a regular linear pencil; 3) the finding of chains of generalized associated vectors corresponding to a prescribed eigenvalue of a regular kernel of a singular linear pencil of complete column rank of two matrices; 4) the finding of linearly independent polynomial solutions of a singular linear pencil. The algorithm consists in the construction of a finite sequence of certain auxiliary matrices the choice of which depends on the problem being solved and in the construction of a sequence of their null-spaces, enabling the obtaining of all necessary information on the unknown vectors of the canonic basis of the problem being solved.Translated from Zapiski Nauchnykh Seminarov Leningradskogo Otdeleniya Matematicheskogo Instituta im. V. A. Steklova AN SSSR, Vol. 90, pp. 46–62, 1979.     

Computational complexity of LCPs associated with positive definite symmetric matrices

Murty in a recent paper has shown that the computational effort required to solve a linear complementarity problem (LCP), by either of the two well known complementary pivot methods is not bounded above by a polynomial in the size of the problem. In that paper, by constructing a class of LCPs—one of ordern forn 2—he has shown that to solve the problem of ordern, either of the two methods goes through 2 ⁿ pivot steps before termination.However that paper leaves it as an open question to show whether or not the same property holds if the matrix,M, in the LCP is positive definite and symmetric. The class of LCPs in whichM is positive definite and symmetric is of particular interest because of the special structure of the problems, and also because they appear in many practical applications.In this paper, we study the computational growth of each of the two methods to solve the LCP, (q, M), whenM is positive definite and symmetric and obtain similar results.This research is partially supported by Air Force Office of Scientific Research, Air Force Number AFOSR-78-3646. The United States Government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright notation thereon.     

On the complexity of linear boolean operators with thin matrices

Under consideration is the problem of constructing a square Booleanmatrix A of order n without “rectangles” (it is a matrix whose every submatrix of the elements that are in any two rows and two columns does not consist of 1s). A linear transformation modulo two defined by A has complexity o(ν(A) − n) in the base {⊕}, where ν(A) is the weight of A, i.e., the number of 1s (the matrices without rectangles are called thin). Two constructions for solving this problem are given. In the first construction, n = p ² where p is an odd prime. The corresponding matrix H _p has weight p ³ and generates the linear transformation of complexity O(p ² log p log log p). In the second construction, the matrix has weight nk where k is the cardinality of a Sidon set in ℤ _n . We may assume that

On the complexity of decomposing matrices arising in satellite communication

Decomposing a square matrix into a weighted sum of permutation matrices, such that the sum of the weights becomes minimal, is NP-hard. This result justifies the heuristic approach to this problem proposed by several authors. An application of this problem arises from intercity communication via transmission satellites.     

The complexity of the realization of subdefinite matrices by gate schemes

Mathematical Notes -     

On the parallel complexity of hierarchical clustering and CC‐complete problems

Complex data sets are often unmanageable unless they can be subdivided and simplified in an intelligent manner. Clustering is a technique that is used in data mining and scientific analysis for partitioning a data set into groups of similar or nearby items. Hierarchical clustering is an important and well‐studied clustering method involving both top‐down and bottom‐up subdivisions of data. In this article we address the parallel complexity of hierarchical clustering. We describe known sequential algorithms for top‐down and bottom‐up hierarchical clustering. The top‐down algorithm can be parallelized, and when there are n points to be clustered, we provide an O(log n)‐time, n²‐processor Crew Pram algorithm that computes the same output as its corresponding sequential algorithm. We define a natural decision problem based on bottom‐up hierarchical clustering, and add this HIERARCHICAL CLUSTERING PROBLEM (HCP) to the slowly growing list of CC‐complete problems, thereby showing that HCP is one of the computationally most difficult problems in the COMPARATOR CIRCUIT VALUE PROBLEM class. This class contains a variety of interesting problems, and now for the first time a problem from data mining as well. By proving that HCP is CC‐complete, we have demonstrated that HCP is very unlikely to have an NC algorithm. This result is in sharp contrast to the NC algorithm which we give for the top‐down sequential approach, and the result surprisingly shows that the parallel complexities of the top‐down and bottom‐up approaches are different, unless CC equals NC. In addition, we provide a compendium of all known CC‐complete problems. © 2008 Wiley Periodicals, Inc. Complexity, 2008     

Construction of Williamson type matrices

Recent advances in the construction of Hadamard matrices have depeaded on the existence of Baumert-Hall arrays and four (1, ?1) matrices A B C Dof order m which are of Williamson type, that is they pair-wise satisfy

i) MN^T = NM^T , ∈ {A B C D} and

ii) AA^T + BB^T + CC^T + DD^T = 4mI_m .

It is shown that Williamson type matrices exist for the orders m = s(4 ? 1)m = s(4s + 3) for s∈ {1, 3, 5, …, 25} and also for m = 93. This gives Williamson matrices for several new orders including 33, 95,189.

These results mean there are Hadamard matrices of order

i) 4s(4s ?1)t, 20s(4s ? 1)t,s ∈ {1, 3, 5, …, 25};

ii) 4s(4:s + 3)t, 20s(4s + 3)t s ∈ {1, 3, 5, …, 25};

iii) 4.93t, 20.93t

for

t ∈ {1, 3, 5, … , 61} ∪ {1 + 2 ^a 10 ^b 26 ^c a b c nonnegative integers}, which are new infinite families.

Also, it is shown by considering eight-Williamson-type matrices, that there exist Hadamard matrices of order 4(p + 1)(2p + l)r and 4(p + l)(2p + 5)r when p ≡ 1 (mod 4) is a prime power, 8ris the order of a Plotkin array, and, in the second case 2p + 6 is the order of a symmetric Hadamard matrix. These classes are new.     

Construction of expanders and superconcentrators using Kolmogorov complexity

We show the existence of various versions of expander graphs using Kolmogorov complexity. This method seems superior to the usual probabilistic construction. It turns out that the best known bounds on the size of expanders and superconcentrators can be attained based on this method. In the case of (acyclic) superconcentrators we attain a density of about 34 edges/vertices. Furthermore, related graph properties are reviewed, like magnification, edge‐magnification, and isolation, and we develop bounds based on the Kolmogorov approach. © 2000 John Wiley & Sons, Inc. Random Struct. Alg., 17: 64–77, 2000