arXiv.itsnick.co

an independent field guide to 23 million papers from arXiv & OpenAlex — not affiliated with arXiv.org

Author

Sanja Singer

15 papers · 2004–2020

2004: 12005: 22006: 02007: 12008: 02009: 02010: 42011: 02012: 22013: 02014: 02015: 02016: 02017: 02018: 02019: 32020: 2
2004peak 4 in 20102020

Frequent co-authors

Papers

showing 1–15 of 15 RSS

Efficient Implementation of the Nelder–Mead Search Algorithm

Saša Singer, Sanja Singer

Numerical Analysis · Applied Numerical Analysis & Computational Mathematics · · cited 75× · DOI ↗ · more like this ·

Abstract The Nelder–Mead or simplex search algorithm is one of the best known algorithms for unconstrained optimization of non–smooth functions. Even though the basic algorithm is quite simple, it is implemented in many different ways. Apart from some minor computational details, the main difference between various implementations lies in the selection of convergence (or termination) tests, which are used to break the iteration process. A fairly simple efficiency analysis of each iteration step reveals a potential computational bottleneck in the domain convergence test. To be efficient, such a test has to be sublinear in the number of vertices of the working simplex. We have tested some of the most common implementations of the Nelder–Mead algorithm, and none of them is efficient in this sense. Therefore, we propose a simple and efficient domain convergence test and discuss some of its…

A GPU-based hyperbolic SVD algorithm

Vedran Novakovic, Sanja Singer

math.NA · · cited 38× · arXiv ↗ · more like this ·

A one-sided Jacobi hyperbolic singular value decomposition (HSVD) algorithm, using a massively parallel graphics processing unit (GPU), is developed. The algorithm also serves as the final stage of solving a symmetric indefinite eigenvalue problem. Numerical testing demonstrates the gains in speed and accuracy over sequential and MPI-parallelized variants of similar Jacobi-type HSVD algorithms. Finally, possibilities of hybrid CPU--GPU parallelism are discussed.

A Kogbetliantz-type algorithm for the hyperbolic SVD

Vedran Novaković, Sanja Singer

math.NA · · cited 5× · arXiv ↗ · more like this ·

In this paper a two-sided, parallel Kogbetliantz-type algorithm for the hyperbolic singular value decomposition (HSVD) of real and complex square matrices is developed, with a single assumption that the input matrix, of order $n$, admits such a decomposition into the product of a unitary, a non-negative diagonal, and a $J$-unitary matrix, where $J$ is a given diagonal matrix of positive and negative signs. When $J=\pm I$, the proposed algorithm computes the ordinary SVD. The paper's most important contribution -- a derivation of formulas for the HSVD of $2\times 2$ matrices -- is presented first, followed by the details of their implementation in floating-point arithmetic. Next, the effects of the hyperbolic transformations on the columns of the iteration matrix are discussed. These effects then guide a redesign of the dynamic pivot ordering, being already a well-established pivot…

Implicit Hari--Zimmermann algorithm for the generalized SVD on the GPUs

Vedran Novaković, Sanja Singer

math.NA · · cited 4× · arXiv ↗ · more like this ·

A parallel, blocked, one-sided Hari--Zimmermann algorithm for the generalized singular value decomposition (GSVD) of a real or a complex matrix pair $(F,G)$ is here proposed, where $F$ and $G$ have the same number of columns, and are both of the full column rank. The algorithm targets either a single graphics processing unit (GPU), or a cluster of those, performs all non-trivial computation exclusively on the GPUs, requires the minimal amount of memory to be reasonably expected, scales acceptably with the increase of the number of GPUs available, and guarantees the reproducible, bitwise identical output of the runs repeated over the same input and with the same number of GPUs.

Skew-Symmetric Differential qd Algorithm

Sanja Singer, Sa?a Singer

Computational Theory and Mathematics · Applied Numerical Analysis & Computational Mathematics · · cited 3× · DOI ↗ · more like this ·

Differential qd (dqd) algorithm with shifts is probably the fastest known algorithm which computes eigenvalues of symmetric tridiagonal matrices with high relative accuracy. In this paper we will construct a similar algorithm for computing eigenvalues of skew-symmetric matrices, which is based on implicit usage of both the QR and the symplectic QR factorizations. If we apply this algorithm to tridiagonal skew-symmetric matrices, we obtain the skew-symmetric dqd algorithm. This algorithm also enjoys high relative stability. However, incorporation of shifts is much harder then in the symmetric case, and yet to be implemented. Finally, the standard algorithm for computing the eigenvalues of tridiagonal skew-symmetric matrices can also be interpreted in the context of the skew-symmetric dqd algorithm. (© 2005 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim)

Nearly optimal scaling in the SR decomposition

Heike Fassbender, Miroslav Rozloznik, Sanja Singer

math.NA · · cited 0× · arXiv ↗ · more like this ·

In this paper we analyze the nearly optimal block diagonal scalings of the rows of one factor and the columns of the other factor in the triangular form of the SR decomposition. The result is a block generalization of the result of the van der Sluis about the almost optimal diagonal scalings of the general rectangular matrices.

The antitriangular factorization of skew-symmetric matrices

Sanja Singer

math.NA · · cited 0× · arXiv ↗ · more like this ·

In this paper we develop algorithms for orthogonal similarity transformations of skew-symmetric matrices to simpler forms. The first algorithm is similar to the algorithm for the block antitriangular factorization of symmetric matrices, but in the case of skew-symmetric matrices, an antitriangular form is always obtained. Moreover, a simple two-sided permutation of the antitriangular form transforms the matrix into a multi-arrowhead matrix. In addition, we show that the block antitriangular form of the skew-Hermitian matrices has the same structure as the block antitriangular form of the symmetric matrices.

Orthosymmetric block rotations

Sanja Singer

Computational Theory and Mathematics · Electronic Journal of Linear Algebra · · cited 0× · DOI ↗ · more like this ·

Rotations are essential transformations in many parts of numerical linear algebra. In this paper, it is shown that there exists a family of matrices unitary with respect to an orthosymmetric scalar product J, that can be decomposed into the product of two J-unitary matrices-a block diagonal matrix and an orthosymmetric block rotation. This decomposition can be used for computing various one-sided and two-sided matrix transformations by divide-and-conquer or treelike algorithms. As an illustration, a blocked version of the QR-like factorization of a given matrix is considered.

Perturbation bounds for singular values of matrices with singletons

Sanja Singer

Computational Theory and Mathematics · Journal of Mathematical Inequalities · · cited 0× · DOI ↗ · more like this ·

This paper considers bounds for the singular values of a matrix with a row-or columnsingleton if such an element is perturbed. A multiplicative approach to such perturbations may lead to sharper bounds than the usual additive Weyl bounds. These bounds immediately give the interlace property for the singular values. These results are used for perturbation bounds if a tridiagonal matrix with a zero diagonal is perturbed in the last pair of off-diagonal elements. Such perturbation bounds are then applied to bound perturbations of the eigenvalues and the first component of the eigenvectors, which has a direct application in the computation of the Gaussian quadrature formulae for compression splines.

Three-Level Parallel J-Jacobi Algorithms for Hermitian Matrices

Sanja Singer, Sasa Singer, Vedran Novakovic +3 more

math.NA · · cited 0× · arXiv ↗ · more like this ·

The paper describes several efficient parallel implementations of the one-sided hyperbolic Jacobi-type algorithm for computing eigenvalues and eigenvectors of Hermitian matrices. By appropriate blocking of the algorithms an almost ideal load balancing between all available processors/cores is obtained. A similar blocking technique can be used to exploit local cache memory of each processor to further speed up the process. Due to diversity of modern computer architectures, each of the algorithms described here may be the method of choice for a particular hardware and a given matrix size. All proposed block algorithms compute the eigenvalues with relative accuracy similar to the original non-blocked Jacobi algorithm.

Novel Modifications of Parallel Jacobi Algorithms

Sanja Singer, Sasa Singer, Vedran Novakovic +2 more

math.NA · · cited 0× · arXiv ↗ · more like this ·

We describe two main classes of one-sided trigonometric and hyperbolic Jacobi-type algorithms for computing eigenvalues and eigenvectors of Hermitian matrices. These types of algorithms exhibit significant advantages over many other eigenvalue algorithms. If the matrices permit, both types of algorithms compute the eigenvalues and eigenvectors with high relative accuracy. We present novel parallelization techniques for both trigonometric and hyperbolic classes of algorithms, as well as some new ideas on how pivoting in each cycle of the algorithm can improve the speed of the parallel one-sided algorithms. These parallelization approaches are applicable to both distributed-memory and shared-memory machines. The numerical testing performed indicates that the hyperbolic algorithms may be superior to the trigonometric ones, although, in theory, the latter seem more natural.

The LAPW method with eigendecomposition based on the Hari--Zimmermann generalized hyperbolic SVD

Sanja Singer, Edoardo Di Napoli, Vedran Novaković +1 more

math.NA · · arXiv ↗ · more like this ·

In this paper we propose an accurate, highly parallel algorithm for the generalized eigendecomposition of a matrix pair $(H, S)$, given in a factored form $(F^{\ast} J F, G^{\ast} G)$. Matrices $H$ and $S$ are generally complex and Hermitian, and $S$ is positive definite. This type of matrices emerges from the representation of the Hamiltonian of a quantum mechanical system in terms of an overcomplete set of basis functions. This expansion is part of a class of models within the broad field of Density Functional Theory, which is considered the golden standard in condensed matter physics. The overall algorithm consists of four phases, the second and the fourth being optional, where the two last phases are computation of the generalized hyperbolic SVD of a complex matrix pair $(F,G)$, according to a given matrix $J$ defining the hyperbolic scalar product. If $J = I$, then these two phases…

Estimates for the Spectral Condition Number of Cardinal B-Spline Collocation Matrices (Long version)

Vedran Novakovic, Sanja Singer, Sasa Singer

math.NA · · arXiv ↗ · more like this ·

The famous de Boor conjecture states that the condition of the polynomial B-spline collocation matrix at the knot averages is bounded independently of the knot sequence, i.e., depends only on the spline degree. For highly nonuniform knot meshes, like geometric meshes, the conjecture is known to be false. As an effort towards finding an answer for uniform meshes, we investigate the spectral condition number of cardinal B-spline collocation matrices. Numerical testing strongly suggests that the conjecture is true for cardinal B-splines.