SEARCH · Engineering Papers
Results for “Cholesky factorization.”
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Task-parallel Sparse Incomplete Cholesky Factorization using Kokkos Portable APIs.
Abstract not provided.
Efficient multitasking of Choleski matrix factorization on CRAY supercomputers
A Choleski method is described and used to solve linear systems of equations that arise in large scale structural analysis. The method uses a novel variable-band storage scheme and is structured to exploit fast local memory caches while minimizing data access delays between main memory and vector registers. Several parallel implementations of this method are described for the CRAY-2 and CRAY Y-MP computers demonstrating the use of microtasking and autotasking directives. A portable parallel language, FORCE, is used for comparison with the microtasked and autotasked implementations. Results are presented comparing the matrix factorization times for three representative structural analysis problems from runs made in both dedicated and multi-user modes on both computers. CPU and wall clock timings are given for the parallel implementations and are compared to single processor timings of the same algorithm.
Positive semidefinite tensor factorizations of the two-electron integral matrix for low-scaling ab initio electronic structure
Tensor factorization of the 2-electron integral matrix is a well-known technique for reducing the computational scaling of ab initio electronic structure methods toward that of Hartree-Fock and density functional theories. The simplest factorization that maintains the positive semidefinite character of the 2-electron integral matrix is the Cholesky factorization. In this paper, we introduce a family of positive semidefinite factorizations that generalize the Cholesky factorization. Using an implementation of the factorization within the parametric 2-RDM method [D. A. Mazziotti, Phys. Rev. Lett. 101, 253002 (2008)], we study several inorganic molecules, alkane chains, and potential energy curves and find that this generalized factorization retains the accuracy and size extensivity of the Cholesky factorization, even in the presence of multi-reference correlation. The generalized family of positive semidefinite factorizations has potential applications to low-scaling ab initio electronic structure methods that treat electron correlation with a computational cost approaching that of the Hartree-Fock method or density functional theory.
Factorization Of Positive Definite, Banded Hermitian Matrices
Report discusses application of Cholesky factorization algorithm to positive definite, banded Hermitian matrices. Begins by extending Cholesky factorization algorithm to cover uniformly-partitioned, banded, positive definite matrices of rank n that is real symmetric or Hermitian. Then two stratagems given for use of algorithm in concurrent-processing system in which N less than it has to be to enable factorization of matrix in as few serial steps as possible and where uniformly high efficiency expected from all processing elements. One of major purposes of this and related studies to maximize speedup and efficiency in system of concurrent-data-processing elements.
Direction-Preserving and Schur-Monotonic Semi-Separable Approximations of Symmetric Positive Definite Matrices}
For a given symmetric positive definite matrix A {element_of} R{sup N x N}, we develop a fast and backward stable algorithm to approximate A by a symmetric positive-definite semi-separable matrix, accurate to a constant multiple of any prescribed tolerance. In addition, this algorithm preserves the product, AZ, for a given matrix Z {element_of} R{sup N x d}, where d << N. Our algorithm guarantees the positive-definiteness of the semi-separable matrix by embedding an approximation strategy inside a Cholesky factorization procedure to ensure that the Schur complements during the Cholesky factorization all remain positive definite after approximation. It uses a robust direction-preserving approximation scheme to ensure the preservation of AZ. We present numerical experiments and discuss potential implications of our work.
Numerical algorithms for finite element computations on concurrent processors
The work of several graduate students which relate to the NASA grant is briefly summarized. One student has worked on a detailed analysis of the so-called ijk forms of Gaussian elemination and Cholesky factorization on concurrent processors. Another student has worked on the vectorization of the incomplete Cholesky conjugate method on the CYBER 205. Two more students implemented various versions of Gaussian elimination and Cholesky factorization on the FLEX/32.
A Nonstationary and Non-Gaussian Moving Average Model for Solar Irradiance
Historically, power has flowed from large power plants to customers. Increasing penetration of distributed energy resources such as solar power from rooftop photovoltaic has made the distribution network a two-way-street with power being generated at the customer level. The incorporation of renewables introduces additional uncertainty and variability into the power grid. Distribution network operation studies are being adapted to include renewables; however, such studies require high quality solar irradiance data that adequately reflect realistic meteorological variability. Data from satellite-based products are spatially complete, but temporally coarse, whereas solar irradiances exhibit high frequency variation at very fine timescales. We propose a new stochastic method for temporally downscaling global horizontal irradiance (GHI) to 1 min resolution, but we do not consider the spatial aspect due to limited availability of the in situ irradiance measurements. Solar irradiance's first and second-order structures vary diurnally and seasonally, and our model adapts to such nonstationarity. Empirical irradiance data exhibits highly non-Gaussian behavior; we develop a nonstationary and non-Gaussian moving average model that is shown to capture realistic solar variability at multiple timescales. We also propose a new estimation scheme based on Cholesky factors of empirical autocovariance matrices, bypassing difficult and inaccessible likelihood-based approaches. The model is demonstrated for a case study of three locations that are located in diverse climates through the United States. The model is compared against competitors from the literature and is shown to provide better uncertainty and variability quantification on testing data.
symPACK: A GPU-Capable Fan-Out Sparse Cholesky Solver
Sparse symmetric positive definite systems of equations are ubiquitous in scientific workloads and applications. Parallel sparse Cholesky factorization is the method of choice for solving such linear systems. Therefore, the development of parallel sparse Cholesky codes that can efficiently run on today’s large-scale heterogeneous distributed-memory platforms is of vital importance. Modern supercomputers offer nodes that contain a mix of CPUs and GPUs. To fully utilize the computing power of these nodes, scientific codes must be adapted to offload expensive computations to GPUs. We present symPACK, a GPU-capable parallel sparse Cholesky solver that uses one-sided communication primitives and remote procedure calls provided by the UPC++ library. We also utilize the UPC++ "memory kinds" feature to enable efficient communication of GPU-resident data. We show that on a number of large problems, symPACK outperforms comparable state-of-the-art GPU-capable Cholesky factorization codes by up to 14x on the NERSC Perlmutter supercomputer.
Learning constitutive relations using symmetric positive definite neural networks
In this work, we present a new neural-network architecture, called the Cholesky-factored symmetric positive definite neural network (SPD-NN), for modeling constitutive relations in computational mechanics. Instead of directly predicting the stress of the material, the SPD-NN trains a neural network to predict the Cholesky factor of the tangent stiffness matrix, based on which the stress is calculated in incremental form. As a result of this special structure, SPD-NN weakly imposes convexity on the strain energy function, satisfies the second order work criterion (Hill's criterion) and time consistency for path-dependent materials, and therefore improves numerical stability, especially when the SPD-NN is used in finite element simulations. Depending on the types of available data, we propose two training methods, namely direct training for strain and stress pairs and indirect training for loads and displacement pairs. We demonstrate the effectiveness of SPD-NN on hyperelastic, elasto-plastic, and multiscale fiber-reinforced plate problems from solid mechanics. The generality and robustness of SPD-NN make it a promising tool for a wide range of constitutive modeling applications.
A Comparison of Linear Solvers for Resolving Flow in Three-Dimensional Discrete Fracture Networks
We compare various methods for resolving steady flow within three-dimensional discrete fracture networks, including direct methods, Krylov subspace methods with and without preconditioning, and multi-grid methods. We compared the performance of the methods based on compute times and scaling of the solution as a function of the number of grid nodes and log-variance of the hydraulic aperture. The methods are applied to three test cases: (a) variable density of networks with a truncated power-law distribution of fracture lengths, (b) a fixed network composed of monodisperse fracture sizes but varied permeability/aperture heterogeneity, (c) and a network based on field site in Nevada, US. We chose these cases to allow us to study the impact of the mesh size and flow properties, as well as to demonstrate our conclusions on a large-scale, realistic problem (more than 40 million mesh nodes). A direct solution using Cholesky factorization outperformed other methods for every example but was closely followed in performance by some algebraic multigrid (AMG) preconditioned Krylov subspace methods. Among the Krylov methods, conjugate gradients (CG) with an AMG preconditioner performs the best. Generally, Cholesky factorization is recommended, but CG with an AMG preconditioner may be suitable for very large problems beyond 40 million nodes where the entire linear system cannot reside in memory.
Efficient use of high performance computers for integrated controls and structures design
Simply transporting design codes from sequential-scalar computers to parallel-vector computers does not fully utilize the computational benefits offered by high performance computers. By performing integrated controls and structures design on an experimental truss platform with both sequential-scalar and parallel-vector design codes, conclusive results are presented to substantiate this claim. The efficiency of a Cholesky factorization scheme in conjunction with a variable-band row data structure is presented. In addition, the Lanczos eigensolution algorithm has been incorporated in the design code for both parallel and vector computations. Comparisons of computational efficiency between the initial design code and the parallel-vector design code are presented. It is shown that the Lanczos algorithm with the Cholesky factorization scheme is far superior to the sub-space iteration method of eigensolution when substantial numbers of eigenvectors are required for control design and/or performance optimization. Integrated design results show the need for continued efficiency studies in the area of element computations and matrix assembly.
Randomized Householder-Cholesky QR Factorization with Multisketching
Explore the source record for details and available documents.
Parameterizing correlations between hydrometeor species in mixed-phase Arctic clouds
Mixed-phase Arctic clouds, like other clouds, contain small-scale variability in hydrometeor fields, such as cloud water or snow mixing ratio. This variability may be worth parameterizing in coarse-resolution numerical models. In particular, for modeling processes such as accretion and aggregation, it would be useful to parameterize subgrid correlations among hydrometeor species. However, one difficulty is that there exist many hydrometeor species and many microphysical processes, leading to complexity and computational expense.Existing lower and upper bounds (inequalities) on linear correlation coefficients provide useful guidance, but these bounds are too loose to serve directly as a method to predict subgrid correlations. Therefore, this paper proposes an alternative method that is based on a blend of theory and empiricism. The method begins with the spherical parameterization framework of Pinheiro and Bates (1996), which expresses the correlation matrix in terms of its Cholesky factorization. The values of the elements of the Cholesky matrix are parameterized here using a cosine row-wise formula that is inspired by the aforementioned bounds on correlations. The method has three advantages: 1) the computational expense is tolerable; 2) the correlations are, by construction, guaranteed to be consistent with each other; and 3) the methodology is fairly general and hence may be applicable to other problems. The method is tested non-interactively using simulations of three Arctic mixed-phase cloud cases from two different field experiments: the Indirect and Semi-Direct Aerosol Campaign (ISDAC) and the Mixed-Phase Arctic Cloud Experiment (M-PACE). Benchmark simulations are performed using a large-eddy simulation (LES) model that includes a bin microphysical scheme. The correlations estimated by the new method satisfactorily approximate the correlations produced by the LES.
The ijk forms of factorization methods. II - Parallel systems
The paper considers the 'ijk forms' of LU and Cholesky factorization on certain parallel computers. This extends an earlier analysis for vector computers. Attention is restricted to local memory systems with processors that may or may not have vector capability. Special attention is given to bus architectures but qualitative analyses are given for other interconnection systems.
Parallel linear equation solvers for finite element computations
The overall objective of this research is to develop efficient methods for the solution of linear and nonlinear systems of equations on parallel and supercomputers, and to apply these methods to the solution of problems in structural analysis. Attention has been given so far only to linear equations. The methods considered for the solution of the stiffness equation Kx=f have been Choleski factorization and the conjugate gradient iteration with SSOR and Incomplete Choleski preconditioning. More detail on these methods will be given on subsequent slides. These methods have been used to solve for the static displacements for the mast and panel focus problems in conjunction with the CSM testbed system based on NICE/SPAR.
Comparison of Factorization-Based Filtering for Landing Navigation
This paper develops and analyzes methods for fusing inertial navigation data with external data, such as data obtained from an altimeter and a star camera. The particular filtering techniques are based upon factorized forms of the Kalman filter, specifically the UDU and Cholesky factorizations. The factorized Kalman filters are utilized to ensure numerical stability of the navigation solution. Simulations are carried out to compare the performance of the different approaches along a lunar descent trajectory using inertial and external data sources. It is found that the factorized forms improve upon conventional filtering techniques in terms of ensuring numerical stability for the investigated landing navigation scenario.