Search NASA⌕ Search

SEARCH · Search NASA

Results for “sparse”

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.

At least 1,099 records · Page 61

Quantum chaos on edge

Recently, the physics of many-body quantum chaotic systems close to their ground states has come under intensified scrutiny. Such studies are motivated by the emergence of model systems exhibiting chaotic fluctuations throughout the entire spectrum [the Sachdev-Ye-Kitaev (SYK) model being a renowned representative] as well as by the physics of holographic principles, which likewise unfold close to ground states. Interpreting the edge of the spectrum as a quantum critical point, here we combine a wide range of analytical and numerical methods to the identification and comprehensive description of two different universality classes: the near edge physics of “sparse” and the near edge of “dense” chaotic systems. The distinction lies in the ratio between the number of a system's random parameters and its Hilbert space dimension, which is exponentially small or algebraically small in the sparse and dense case, respectively. Notable representatives of the two classes are generic chaotic many-body models (sparse) and invariant random matrix ensembles or chaotic gravitational systems (dense). While the two families share identical spectral correlations at energy scales comparable to the level spacing, the density of states and its fluctuations near the edge are different. Considering the SYK model as a representative of the sparse class, we apply a combination of field theory and exact diagonalization to a detailed discussion of its edge spectrum. Conversely, Jackiw-Teitelboim gravity is our reference model for the dense class, where an analysis of the gravitational path integral and random matrix theory reveal universal differences to the sparse class, whose implications for the construction of holographic principles we discuss. Published by the American Physical Society 2024

Altland, Alexander (ORCID:0000000229914805)↗

A method for volumetric imaging in radiotherapy using single x-ray projection

Purpose: It is an intriguing problem to generate an instantaneous volumetric image based on the corresponding x-ray projection. The purpose of this study is to develop a new method to achieve this goal via a sparse learning approach. Methods: To extract motion information hidden in projection images, the authors partitioned a projection image into small rectangular patches. The authors utilized a sparse learning method to automatically select patches that have a high correlation with principal component analysis (PCA) coefficients of a lung motion model. A model that maps the patch intensity to the PCA coefficients was built along with the patch selection process. Based on this model, a measured projection can be used to predict the PCA coefficients, which are then further used to generate a motion vector field and hence a volumetric image. The authors have also proposed an intensity baseline correction method based on the partitioned projection, in which the first and the second moments of pixel intensities at a patch in a simulated projection image are matched with those in a measured one via a linear transformation. The proposed method has been validated in both simulated data and real phantom data. Results: The algorithm is able to identify patches that contain relevant motion information such as the diaphragm region. It is found that an intensity baseline correction step is important to remove the systematic error in the motion prediction. For the simulation case, the sparse learning model reduced the prediction error for the first PCA coefficient to 5%, compared to the 10% error when sparse learning was not used, and the 95th percentile error for the predicted motion vector was reduced from 2.40 to 0.92 mm. In the phantom case with a regular tumor motion, the predicted tumor trajectory was successfully reconstructed with a 0.82 mm error for tumor center localization compared to a 1.66 mm error without using the sparse learning method. When the tumor motion was driven by a real patient breathing signal with irregular periods and amplitudes, the average tumor center error was 0.6 mm. The algorithm robustness with respect to sparsity level, patch size, and presence or absence of diaphragm, as well as computation time, has also been studied. Conclusions: The authors have developed a new method that automatically identifies motion information from an x-ray projection, based on which a volumetric image is generated.

60 APPLIED LIFE SCIENCES↗

Spectrotemporal CT data acquisition and reconstruction at low dose

Purpose: X-ray computed tomography (CT) is widely used, both clinically and preclinically, for fast, high-resolution anatomic imaging; however, compelling opportunities exist to expand its use in functional imaging applications. For instance, spectral information combined with nanoparticle contrast agents enables quantification of tissue perfusion levels, while temporal information details cardiac and respiratory dynamics. The authors propose and demonstrate a projection acquisition and reconstruction strategy for 5D CT (3D + dual energy + time) which recovers spectral and temporal information without substantially increasing radiation dose or sampling time relative to anatomic imaging protocols. Methods: The authors approach the 5D reconstruction problem within the framework of low-rank and sparse matrix decomposition. Unlike previous work on rank-sparsity constrained CT reconstruction, the authors establish an explicit rank-sparse signal model to describe the spectral and temporal dimensions. The spectral dimension is represented as a well-sampled time and energy averaged image plus regularly undersampled principal components describing the spectral contrast. The temporal dimension is represented as the same time and energy averaged reconstruction plus contiguous, spatially sparse, and irregularly sampled temporal contrast images. Using a nonlinear, image domain filtration approach, the authors refer to as rank-sparse kernel regression, the authors transfer image structure from the well-sampled time and energy averaged reconstruction to the spectral and temporal contrast images. This regularization strategy strictly constrains the reconstruction problem while approximately separating the temporal and spectral dimensions. Separability results in a highly compressed representation for the 5D data in which projections are shared between the temporal and spectral reconstruction subproblems, enabling substantial undersampling. The authors solved the 5D reconstruction problem using the split Bregman method and GPU-based implementations of backprojection, reprojection, and kernel regression. Using a preclinical mouse model, the authors apply the proposed algorithm to study myocardial injury following radiation treatment of breast cancer. Results: Quantitative 5D simulations are performed using the MOBY mouse phantom. Twenty data sets (ten cardiac phases, two energies) are reconstructed with 88 μm, isotropic voxels from 450 total projections acquired over a single 360° rotation. In vivo 5D myocardial injury data sets acquired in two mice injected with gold and iodine nanoparticles are also reconstructed with 20 data sets per mouse using the same acquisition parameters (dose: ∼60 mGy). For both the simulations and the in vivo data, the reconstruction quality is sufficient to perform material decomposition into gold and iodine maps to localize the extent of myocardial injury (gold accumulation) and to measure cardiac functional metrics (vascular iodine). Their 5D CT imaging protocol represents a 95% reduction in radiation dose per cardiac phase and energy and a 40-fold decrease in projection sampling time relative to their standard imaging protocol. Conclusions: Their 5D CT data acquisition and reconstruction protocol efficiently exploits the rank-sparse nature of spectral and temporal CT data to provide high-fidelity reconstruction results without increased radiation dose or sampling time.

60 APPLIED LIFE SCIENCES↗

A robust real-time surface reconstruction method on point clouds captured from a 3D surface photogrammetry system

Purpose: To develop a robust and real-time surface reconstruction method on point clouds captured from a 3D surface photogrammetry system. Methods: The authors have developed a robust and fast surface reconstruction method on point clouds acquired by the photogrammetry system, without explicitly solving the partial differential equation required by a typical variational approach. Taking advantage of the overcomplete nature of the acquired point clouds, their method solves and propagates a sparse linear relationship from the point cloud manifold to the surface manifold, assuming both manifolds share similar local geometry. With relatively consistent point cloud acquisitions, the authors propose a sparse regression (SR) model to directly approximate the target point cloud as a sparse linear combination from the training set, assuming that the point correspondences built by the iterative closest point (ICP) is reasonably accurate and have residual errors following a Gaussian distribution. To accommodate changing noise levels and/or presence of inconsistent occlusions during the acquisition, the authors further propose a modified sparse regression (MSR) model to model the potentially large and sparse error built by ICP with a Laplacian prior. The authors evaluated the proposed method on both clinical point clouds acquired under consistent acquisition conditions and on point clouds with inconsistent occlusions. The authors quantitatively evaluated the reconstruction performance with respect to root-mean-squared-error, by comparing its reconstruction results against that from the variational method. Results: On clinical point clouds, both the SR and MSR models have achieved sub-millimeter reconstruction accuracy and reduced the reconstruction time by two orders of magnitude to a subsecond reconstruction time. On point clouds with inconsistent occlusions, the MSR model has demonstrated its advantage in achieving consistent and robust performance despite the introduced occlusions. Conclusions: The authors have developed a fast and robust surface reconstruction method on point clouds captured from a 3D surface photogrammetry system, with demonstrated sub-millimeter reconstruction accuracy and subsecond reconstruction time. It is suitable for real-time motion tracking in radiotherapy, with clear surface structures for better quantifications.

60 APPLIED LIFE SCIENCES↗

TH-AB-202-08: A Robust Real-Time Surface Reconstruction Method On Point Clouds Captured From a 3D Surface Photogrammetry System

Purpose: Surface photogrammetry (e.g. VisionRT, C-Rad) provides a noninvasive way to obtain high-frequency measurement for patient motion monitoring in radiotherapy. This work aims to develop a real-time surface reconstruction method on the acquired point clouds, whose acquisitions are subject to noise and missing measurements. In contrast to existing surface reconstruction methods that are usually computationally expensive, the proposed method reconstructs continuous surfaces with comparable accuracy in real-time. Methods: The key idea in our method is to solve and propagate a sparse linear relationship from the point cloud (measurement) manifold to the surface (reconstruction) manifold, taking advantage of the similarity in local geometric topology in both manifolds. With consistent point cloud acquisition, we propose a sparse regression (SR) model to directly approximate the target point cloud as a sparse linear combination from the training set, building the point correspondences by the iterative closest point (ICP) method. To accommodate changing noise levels and/or presence of inconsistent occlusions, we further propose a modified sparse regression (MSR) model to account for the large and sparse error built by ICP, with a Laplacian prior. We evaluated our method on both clinical acquired point clouds under consistent conditions and simulated point clouds with inconsistent occlusions. The reconstruction accuracy was evaluated w.r.t. root-mean-squared-error, by comparing the reconstructed surfaces against those from the variational reconstruction method. Results: On clinical point clouds, both the SR and MSR models achieved sub-millimeter accuracy, with mean reconstruction time reduced from 82.23 seconds to 0.52 seconds and 0.94 seconds, respectively. On simulated point cloud with inconsistent occlusions, the MSR model has demonstrated its advantage in achieving consistent performance despite the introduced occlusions. Conclusion: We have developed a real-time and robust surface reconstruction method on point clouds acquired by photogrammetry systems. It serves an important enabling step for real-time motion tracking in radiotherapy. This work is supported in part by NIH grant R01 CA169102-02.

60 APPLIED LIFE SCIENCES↗

s-blas

This package includes the implementation for four sparse linear algebra kernels: Sparse-Matrix-Vector-Multiplication (SpMV), Sparse-Triangular-Solve (SpTRSV), Sparse-Matrix-Transposition (SpTrans) and Sparse-Matrix-Matrix-Multiplication (SpMM) for Single-node Multi-GPU (scale-up) platforms such as NVIDIA DGX-1 and DGX-2.

Li, Ang↗

ExaSGD: 2022 Kernel Thrust Activities

The Kernel Thrust milestone ADSE22-407 covers the development of device-capable optimization algorithms and solvers technologies required by the ExaSGD project’s software stack in order to solve security-constrained alternating current optimal power flow (SC-ACOPF) problems on emerging exascale architectures. To this extent, in FY22 the main objective of the Kernel Thrust was (i) provide sparse optimization solver that runs efficiently on hardware accelerator devices (i.e., NVIDIA and AMD GPUs) to perform intra-node computations, (ii) strengthen the reliability and increase the performance of the mixed-dense sparse (MDS) solver of HiOp for deployment on the FY22 target architectures, Summit and Crusher, and (iii) increase performance by improving the mathematical algorithm and refining the parallel MPI-based implementation of the coarse-grain parallel solver HiOp-PriDec for capabilities deployment on the FY22 target architectures, Summit and Crusher. This document presents the developments and contributions done by the Kernels Thrust Team in FY22 toward completion of the above-mentioned objectives. These contributions progressed along four main development (sub)thrusts: (1) Design and implementation of a sparse optimization solver for use on hardware accelerators; (2) Improvement of the mathematical algorithm and of the parallel implementation of HiOp-PriDec to ensure readiness and efficient coarse-grain parallelism for FY23 target exascale machine; and (3) Support Software and Application Development Thrusts of the exaSGD project in their deployment of the project’s software stack on AMD- and NVIDIA-based architectures. The development of the sparse optimization solver (thrust 1 above) was new in FY22 and resulted in a new sparse solver in HiOp (available as of version 0.6). The second development thrust was a continuation of the efforts from FY21 and improved the mathematical algorithm and the communication strategy of the HiOp-PriDec solver. The last developement thrust is a large collaborative effort. Namely, the project’s teams from multiple labs (LLNL, PNNL, ORNL, and NREL) performed large-scale demonstration of the ExaSGD software stack, namely the optimization solvers of HiOp interfaced with the modeling front-end ExaGO and the stochastic sampler PowerScenarios. These demonstration efforts solved large-scale instances of the SC-ACOPF challenge problem of medium network sizes (10, 000-bus system) and large number of contingencies on Summit (NVIDIA accelerators) and Crusher (AMD accelerators) systems at ORNL.

97 MATHEMATICS AND COMPUTING↗

Sparsity for Gradient-Based Optimization of Wind Farm Layouts

Optimizing wind farm layouts is an important step in designing an efficient wind farm. Optimizing wind farm layouts is also a difficult task due to computation times increasing with the number of turbines present in the farm. The most computationally expensive part of gradient-based optimization is calculating the gradient. In order to reduce the expense of gradient calculation, we performed a study on the use of sparsity in wind farm layout optimization. This paper presents the findings of the sparsity study and provides a method to use sparsity in wind farm layout optimization. We tested this sparsity method by optimizing multiple farms with sparse methods and compared the results to optimizations of the same farms using traditional methods. By using the sparse method to optimize multiple farms, we found that the objective results were comparable between sparse and traditional methods and that sparse methods were 4 times faster than traditional methods on average. We expect more speedups with improved methodology and larger wind farms. By using sparse methods, it is possible to solve the wind farm layout optimization problem more efficiently, thus allowing for a more thorough study of the wind farm layout design space without excessive computational costs. Further work is required to refine the method and prepare for testing on real-world wind farm layout applications.

gradient↗

Multiple View Zenith Angle Observations of Reflectance From Ponderosa Pine Stands

Reflectance factors (RF(lambda)) from dense and sparse ponderosa pine (Pinus ponderosa) stands, derived from radiance data collected in the solar principal plane by the Advanced Solid-State Array Spectro-radiometer (ASAS), were examined as a function of view zenith angle (theta(sub v)). RF(lambda) was maximized with theta(sub v) nearest the solar retrodirection, and minimized near the specular direction throughout the ASAS spectral region. The dense stand had much higher RF anisotropy (ma)dmurn RF is minimum RF) in the red region than did the sparse stand (relative differences of 5.3 vs. 2.75, respectively), as a function of theta(sub v), due to the shadow component in the canopy. Anisotropy in the near-infrared (NIR) was more similar between the two stands (2.5 in the dense stand and 2.25 in the sparse stand); the dense stand exhibited a greater hotspot effect than 20 the sparse stand in this spectral region. Two common vegetation transforms, the NIR/red ratio and the normalized difference vegetation index (NDVI), both showed a theta(sub v) dependence for the dense stand. Minimum values occurred near the retrodirection and maximum values occurred near the specular direction. Greater relative differences were noted for the NIR/red ratio (2.1) than for the NDVI (1.3). The sparse stand showed no obvious dependence on theta(sub v) for either transform, except for slightly elevated values toward the specular direction.

Johnson, Lee F.↗

Porting the Nonlinear Optimization Library HiOp to Accelerator-Based Hardware Architectures

While interior point method has been the centerpiece of nonlinear programming tools used in science and engineering, its reliance on linear solvers that can tackle sparse symmetric indefinite and highly ill-conditioned problems made it difficult to implement it effectively on hardware accelerators. HiOp optimization package attempts to provide an implementation of the interior point method suitable for hardware accelerators by compressing the original sparse problem to produce an underlying linear problem that is dense and of manageable size. Implementations of dense linear solvers are more mature and utilize hardware accelerators better than their sparse counterparts. There is a number of important domain problems, such as optimal power flow analysis for power grids, where the sparse problem can be effectively compressed and deploying dense linear solver within the interior point method can improve performance. Here we describe a portable implementation of HiOp optimization engine, which uses a linear solver from Magma library and runs entirely on hardware accelerators. To compress the problem, HiOp uses customized mixed dense-sparse linear algebra. All HiOp kernels are implemented using Umpire and RAJA portability libraries. We describe details of the implementation and discuss trade-offs between performance, portability and development cost.

97 MATHEMATICS AND COMPUTING↗

Sparsity and Nullity: Paradigms for Analysis Dictionary Learning

Sparse models in dictionary learning have been successfully applied in a wide variety of machine learning and computer vision problems, and as a result have recently attracted increased research interest. Another interesting related problem based on linear equality constraints, namely the sparse null space (SNS) problem, first appeared in 1986 and has since inspired results on sparse basis pursuit. In this work, we investigate the relation between the SNS problem and the analysis dictionary learning (ADL) problem, and show that the SNS problem plays a central role, and may be utilized to solve dictionary learning problems. Moreover, we propose an efficient algorithm of sparse null space basis pursuit (SNS-BP) and extend it to a solution of ADL. Lastly, experimental results on numerical synthetic data and real-world data are further presented to validate the performance of our method.

97 MATHEMATICS AND COMPUTING↗

Bi-sparsity pursuit: A paradigm for robust subspace recovery

The success of sparse models in computer vision and machine learning is due to the fact that, high dimensional data is distributed in a union of low dimensional subspaces in many real-world applications. The underlying structure may, however, be adversely affected by sparse errors. In this paper, we propose a bi-sparse model as a framework to analyze this problem, and provide a novel algorithm to recover the union of subspaces in the presence of sparse corruptions. We further show the effectiveness of our method by experiments on real-world vision data.

42 ENGINEERING↗

Convolutional neural network based non-iterative reconstruction for accelerating neutron tomography *

Abstract Neutron computed tomography (NCT), a 3D non-destructive characterization technique, is carried out at nuclear reactor or spallation neutron source-based user facilities. Because neutrons are not severely attenuated by heavy elements and are sensitive to light elements like hydrogen, neutron radiography and computed tomography offer a complementary contrast to x-ray CT conducted at a synchrotron user facility. However, compared to synchrotron x-ray CT, the acquisition time for an NCT scan can be orders of magnitude higher due to lower source flux, low detector efficiency and the need to collect a large number of projection images for a high-quality reconstruction when using conventional algorithms. As a result of the long scan times for NCT, the number and type of experiments that can be conducted at a user facility is severely restricted. Recently, several deep convolutional neural network (DCNN) based algorithms have been introduced in the context of accelerating CT scans that can enable high quality reconstructions from sparse-view data. In this paper, we introduce DCNN algorithms to obtain high-quality reconstructions from sparse-view and low signal-to-noise ratio NCT data-sets thereby enabling accelerated scans. Our method is based on the supervised learning strategy of training a DCNN to map a low-quality reconstruction from sparse-view data to a higher quality reconstruction. Specifically, we evaluate the performance of two popular DCNN architectures—one based on using patches for training and the other on using the full images for training. We observe that both the DCNN architectures offer improvements in performance over classical multi-layer perceptron as well as conventional CT reconstruction algorithms. Our results illustrate that the DCNN can be a powerful tool to obtain high-quality NCT reconstructions from sparse-view data thereby enabling accelerated NCT scans for increasing user-facility throughput or enabling high-resolution time-resolved NCT scans.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

BI-sparsity pursuit for robust subspace recovery

Here, the success of sparse models in computer vision and machine learning in many real-world applications, may be attributed in large part, to the fact that many high dimensional data are distributed in a union of low dimensional subspaces. The underlying structure may, however, be adversely affected by sparse errors, thus inducing additional complexity in recovering it. In this paper, we propose a bi-sparse model as a framework to investigate and analyze this problem, and provide as a result , a novel algorithm to recover the union of subspaces in presence of sparse corruptions. We additionally demonstrate the effectiveness of our method by experiments on real-world vision data.

47 OTHER INSTRUMENTATION↗

Adapting In Situ Accelerators for Sparsity With Granular Matrix Reordering

Neural network (NN) inference is an essential part of modern systems and is found at the heart of numerous applications ranging from image recognition to natural language processing. In situ NN accelerators can efficiently perform NN inference using resistive crossbars, which makes them a promising solution to the data movement challenges faced by conventional architectures. Although such accelerators demonstrate significant potential for dense NNs, they often do not benefit from sparse NNs, which contain relatively few non-zero weights. Processing sparse NNs on in situ accelerators results in wasted energy to charge the entire crossbar where most elements are zeros. To address this limitation, this paper proposes Granular Matrix Reordering (GMR): a preprocessing technique that enables an energy-efficient computation of sparse NNs on in situ accelerators. GMR reorders the rows and columns of sparse weight matrices to maximize the crossbars' utilization and minimize the total number of crossbars needed to be charged. The reordering process does not rely on sparsity patterns and incurs no accuracy loss. Finally, GMR achieves an average of 28% and up to 34% reduction in energy consumption over seven pruned NNs across four different pruning methods and network architectures.

97 MATHEMATICS AND COMPUTING↗

A High Performance Block Eigensolver for Nuclear Configuration Interaction Calculations

As on-node parallelism increases and the performance gap between the processor and the memory system widens, achieving high performance in large-scale scientific applications requires an architecture-aware design of algorithms and solvers. We focus on the eigenvalue problem arising in nuclear Configuration Interaction (CI) calculations, where a few extreme eigenpairs of a sparse symmetric matrix are needed. Here, we consider a block iterative eigensolver whose main computational kernels are the multiplication of a sparse matrix with multiple vectors (SpMM), and tall-skinny matrix operations. We then present techniques to significantly improve the SpMM and the transpose operation SpMM T by using the compressed sparse blocks (CSB) format. We achieve 3-4× speedup on the requisite operations over good implementations with the commonly used compressed sparse row (CSR) format. We develop a performance model that allows us to correctly estimate the performance of our SpMM kernel implementations, and we identify cache bandwidth as a potential performance bottleneck beyond DRAM. We also analyze and optimize the performance of LOBPCG kernels (inner product and linear combinations on multiple vectors) and show up to 15× speedup over using high performance BLAS libraries for these operations. The resulting high performance LOBPCG solver achieves 1.4× to 1.8× speedup over the existing Lanczos solver on a series of CI computations on high-end multicore architectures (Intel Xeons). We also analyze the performance of our techniques on an Intel Xeon Phi Knights Corner (KNC) processor.

97 MATHEMATICS AND COMPUTING↗

OFDM Radar Space-Time Adaptive Processing by Exploiting Spatio-Temporal Sparsity

We propose a sparsity-based space-time adaptive processing (STAP) algorithm to detect a slowly-moving target using an orthogonal frequency division multiplexing (OFDM) radar. We observe that the target and interference spectra are inherently sparse in the spatio-temporal domain. Hence, we exploit that sparsity to develop an efficient STAP technique that utilizes considerably lesser number of secondary data and produces an equivalent performance as the other existing STAP techniques. In addition, the use of an OFDM signal increases the frequency diversity of our system, as different scattering centers of a target resonate at different frequencies, and thus improves the target detectability. First, we formulate a realistic sparse-measurement model for an OFDM radar considering both the clutter and jammer as the interfering sources. Then, we apply a residual sparse-recovery technique based on the LASSO estimator to estimate the target and interference covariance matrices, and subsequently compute the optimal STAP-filter weights. Our numerical results demonstrate a comparative performance analysis of the proposed sparse-STAP algorithm with four other existing STAP methods. Furthermore, we discover that the OFDM-STAP filter-weights are adaptable to the frequency-variabilities of the target and interference responses, in addition to the spatio-temporal variabilities. Hence, by better utilizing the frequency variabilities, we propose an adaptive OFDM-waveform design technique, and consequently gain a significant amount of STAP-performance improvement.

adaptive waveform design.↗

PyAMG: Algebraic Multigrid Solvers in Python

PyAMG is a Python package of algebraic multigrid (AMG) solvers and supporting tools for approximating the solution to large, sparse linear systems of algebraic equations, Ax = b, where A is an n × n sparse matrix. Sparse linear systems arise in a range of problems in science, from fluid flows to solid mechanics to data analysis. While the direct solvers available in SciPy’s sparse linear algebra package (scipy.sparse.linalg) are highly efficient, in many cases iterative methods are preferred due to overall complexity. However, the iterative methods in SciPy, such as CG and GMRES, often require an efficient preconditioner in order to achieve a lower complexity. Preconditioning is a powerful tool whereby the conditioning of the linear system and convergence rate of the iterative method are both dramatically improved. PyAMG constructs multigrid solvers for use as a preconditioner in this setting. A summary of multigrid and algebraic multigrid solvers can be found in Olson (2015a), in Olson (2015b), and in Falgout (2006); a detailed description can be found in Briggs et al. (2000) and Trottenberg et al. (2001).

97 MATHEMATICS AND COMPUTING↗