Search

This article investigates the complex symmetric linear equations that appear in high-frequency electromagnetic-field analysis. As an efficient linear solver, we propose a mixed-precision iterative method with double- and double-double (DD)-precision floating-point numbers and an efficient implementation of DD-precision arithmetic with fused multiply-add instructions. Using the proposed method, we successfully reduce both iteration count and calculation time compared with the conventional method. Moreover, we demonstrate the relationship between multiple-precision arithmetic and the acceleration factor of the incomplete Cholesky factorization.

To effectively archive configuration data during molecular dynamics (MD) simulations of polymer systems, we present an efficient compression method with good numerical accuracy that preserves the topology of ring-linear polymer blends. To compress the fraction of floating-point data, we used the Jointed Hierarchical Precision Compression Number - Data Format (JHPCN-DF) method to apply zero padding for the tailing fraction bits, which did not affect the numerical accuracy, then compressed the data with Huffman coding. We also provided a dataset of well-equilibrated configurations of MD simulations for ring-linear polymer blends with various lengths of linear and ring polymers, including ring complexes composed of multiple rings such as polycatenane. We executed 109 MD steps to obtain 150 equilibrated configurations. The combination of JHPCN-DF and SZ compression achieved the best compression ratio for all cases. Therefore, the proposed method enables efficient archiving of MD trajectories. Moreover, the publicly available dataset of ring-linear polymer blends can be employed for studies of mathematical methods, including topology analysis and data compression, as well as MD simulations.

An iterative domain decomposition method is proposed for numerical analysis of three-dimensional (3D) linear magnetostatic problems taking the magnetic vector potential as an unknown function. The iterative domain decomposition method is combined with the preconditioned conjugate gradient (PCG) procedure and the hierarchical domain decomposition method (HDDM) which is adopted in parallel computing. Our previously employed preconditioner was the Neumann–Neumann (NN) preconditioner. Numerical results showed that the method was only effective for problems with a small number of subdomains. In this paper, we show its improvement with the balancing domain decomposition diagonal scaling (BDD-DIAG) preconditioner and show the asymptotic equivalence between BDD-DIAG and the simplified diagonal scaling (diag) preconditioner, which is derived from the following numerical evidences.

The authors investigate a high-accuracy analysis technique for the electromagnetic field in a human body by using the parallel finite-element method. In the high-frequency electromagnetic field analysis with anatomical human models, speed-up of calculation is one of the most important topics. In this paper, we implement the code and evaluate the performance of two kinds of iterative method for the interface problem in the domain decomposition method (DDM) by numerical experiments using large-scale human body models. As the results, the conjugate orthogonal conjugate residual method is very effective as the interface solver in the DDM method for the high-frequency electromagnetic field analysis using the anatomical human model.

This note briefly reports on the applicability of the local-in-time (LT) adjoint-based method to a large-scale topology optimization problem with unsteady thermal-fluid. The basic idea of the LT method is to divide a time-dependent optimization problem into reasonable subproblems to reduce memory cost. We demonstrate that the proposed method solves the large-scale topology optimization problem by incorporating the LT method, the lattice Boltzmann method, and parallel computing.

This paper deals with multiple-precision iterative methods for solving the complex symmetric linear equation derived from high-frequency electromagnetic field analysis using the edge finite element method. Double-precision iterative methods for solving this problem are slow to converge. Therefore, we implement a multiple-precision calculation, specifically, double-double (DD) precision numbers. The DD-precision complex numbers and arithmetic operations are implemented using an error-free transformation based on Knuth’s and Dekker’s algorithm. Moreover, some iterative methods with mixed precision are proposed to reduce computational cost. Using developed systems and the QD library that supports DD and quad-double arithmetics, some numerical methods are demonstrated experimentally and their performances are evaluated.

This paper proposes a novel file transfer tool to improve file transfer performance over Japan high performance computing and networking (JHPCN). We first develop a high-performance and flexible protocol (HpFP) for inter-datacenter transport network. An original HpFP is designed first for specified networks and puts more emphasis on latency and packet loss tolerances than fairness and friendliness, while an enhanced HpFP is more suitable for real network environments. Then, based on the enhanced HpFP, we implement a file transfer tool, called high-performance copy (HCP). The performance of our file transfer tool is evaluated between datacenters of JHPCN using real datasets collected from supercomputer resources. The results show that the HCP achieves higher throughput than traditional tool for file transfer over JHPCN.

This paper describes iterative methods for the high frequency electromagnetic analysis using the finite element method of Maxwell equations including displacement current. The conjugate orthogonal conjugate gradient method has been widely used to solve a complex symmetric system. However, the conventional method suffers from oscillating convergence histories in large-scale analysis. In this paper, to solve large-scale complex symmetric systems arising from the formulation of the E method, an iterative substructuring method like the minimal residual method is presented, and the performance of the convergence of the method is evaluated by numerical results. As the result, the proposed method shows a stable convergence behavior and a fast convergence rate compared to other iterative methods.

We have investigated a high-accuracy analysis for the electromagnetic field of numerical human body models using the finite-element method. The numerical human body models generated from computed tomography images are represented as voxel data, and composed of skin layers, blood vessels, bones, internal organs, and so on. In this paper, we propose a mesh smoothing technique to reduce the noise caused by reflection and scattering of the electric fields around material boundaries. Therefore, using a supercomputer, we successfully evaluated the electromagnetic field distribution inside the whole body model with a high accuracy.

This research was conducted to develop the three-dimensional (3D) finite element model of human skin for bio-heat transfer analysis. The skin burn was analyzed using Penne’s bio-heat equation, which has been adopted in many commercial finite element software. Burn injuries mostly occur due to heat transfer from hot object, hot liquids, cooking flames, and sometimes due to exposure to chemicals, electricity, and ionizing radiation. Depending upon the condition and duration of exposing, thermal burn may cause severe skin damage. The burn effect on human skin under the contact with a hot object or hot fluid was analyzed in this paper. The burn intensity in terms of degrees of burn was measured with different burning conditions and their corresponding time was graphically shown. Using the temperature profile obtained from the analysis, various methods of burn treatment were evaluated and compared. The results from this analysis will help to understand human skin burn under different burning conditions and treatment of different burn injuries.

We have been developing an open-source computer-aided engineering (CAE) software, ADVENTURE, which is a general-purpose parallel finite element analysis system and can simulate a large-scale analysis model with supercomputer. For supercomputer architecture such as an exa-scale system, to obtain high computational efficiency for software that requires large-scale numerical calculation data processing, a programming model that considers the hierarchical structure of hardware, such as a microprocessor and memory, is necessary. From this point of view, ADVENTURE system was developed using the hierarchical domain decomposition method (HDDM) as the basic technology for a large-scale data system. HDDM is technology developed by ourselves mainly for numerical analysis method. In particular, we have developed application-specific system software that can obtain high performance by focusing on simulation of continuum mechanics by finite element method (FEM) and particle method which are highly demanded by academic research and industry. We have developed four research items “DDM I/O (input/output) library,” “DDM solver library,” “DSL for continuum mechanics,” and “continuous mechanics simulator.” The software, which is the result of our research, is released as open-source software on the sub-project page in the ADVENTURE project homepage. In this chapter, some of those libraries are described in detail.

In numerical libraries for sparse matrix operations, there are many tuning parameters related to implementation selection. Selection of different tuning parameters could result in totally different performance. Moreover, optimal implementation depends on the sparse matrices to be operated. It is difficult to find optimal implementation without executing each implementation and thereby examining its performance on a given sparse matrix. In this study, we propose an implementation selection method for sparse iterative algorithms and preconditioners in a numerical library using deep learning. The proposed method uses full color images to represent the features of a sparse matrix. We present an image generation method for partitioning a given matrix (to generate its feature image) so that the value of each matrix element is considered in the implementation selection. We then evaluate the effectiveness of the proposed method by conducting a numerical experiment. In this experiment, the accuracy of implementation selection is evaluated. The training data comprise a pair of sparse matrix and its optimal implementation. The optimal implementation of each sparse matrix in the training data is obtained in advance by executing every implementation and getting the best one. The experimental results obtained using the proposed method show that the accuracy of selecting the optimal implementation of each sparse matrix is 79.5%.

In the present study, performance improvements of the particle search and particle interaction calculation steps constituting the performance bottleneck in the moving particle simulation method are achieved by developing GPU-compatible algorithms for many core processor architectures. In the improvements of particle search, bucket loops of the cell-linked list are changed to a loop structure having fewer local variables and the linked list and the forward star of particle search algorithms within a bucket are compared. In the particle interaction calculation, the problem of the ratio of particles within the interaction domain to the neighboring particle candidates being quite low is improved. By these improvements, a performance efficiency of 24.7 % can be achieved for the first-order polynomial approximation scheme using NVIDIA Tesla K20, CUDA-6.5, and double-precision floating-point operations.

This paper presents the three-dimensional finite element seismic response analysis of full-scale boiling water reactor BWR5 at Kashiwazaki-Kariwa Nuclear Power Station subjected to the Niigata-ken Chuetsu-Oki earthquake that occurred on 16 July 2007. During the earthquake, the automatic shutdown system of the reactors was activated successfully. Although the monitored seismic acceleration significantly exceeded the design level, it was found that there were no significant damages of the reactor cores or other important systems, structures and components through in-depth investigation. In the seismic design commonly used in Japan, a lumped mass model is employed to evaluate the seismic response of structures and components. Although the lumped mass model has worked well so far for a seismic proof design, it is still needed to develop more precise methods for the visual understanding of response behaviors. In the present study, we propose the three-dimensional finite element seismic response analysis of the...

In this paper, a partitioned coupling analysis system is developed for a numerical simulation of 3-dimensional fluid–structure interaction (FSI) problems, adopting an incompressible smoothed particle hydrodynamics (SPH) method for fluid dynamics involving free surface flow and the finite element method (FEM) for structural dynamics. A coupling analysis of a particle-based method and a grid-based method has been investigated. However, most of these are developed as a function-specific application software, and therefore lack versatility. Hence, to save cost in software development and maintenance, the open source software is utilized. Especially, a general-purpose finite element analysis system, named ADVENTURE, and a general-purpose coupling analysis platform, named REVOCAP_Coupler, are employed. Moreover, techniques of an interface marker on fluid–structure boundaries and a dummy mesh for fluid analysis domain are adopted to solve the problem that the REVOCAP_Coupler performs to unify two or more grid-based method codes. To verify a developed system, the dam break problem with an elastic obstacle is demonstrated, and the result is compared with the results calculated by the other methods.

We have been developing the ADVENTURE system, which is an open source CAE software based on parallel finite element method. Two of it‘s main modules are ADVENTURE_Thermal and ADVENTURE_Solid. A thermal-solid coupling system has been developed using these two modules. Three dimensional heat transfer and thermal stresses of a cooling stave in a blast furnace is modeled and analyzed using the developed system. The steady state heat conduction through the model is analysed using the ADVENTURE_Thermal module. Then the temperature distribution is considered as thermal load in the structural analysis. The ADVENTURE_Solid is used for this purpose. Both modules are featured with the parallel finite element method to solve large scale problems. Quite commonly, the matrix that participates in the finite element computation is sparse. Sparse matrix by definition, are populated primarily with many zeroes and thus special storage schemes need to be used to enable efficient storage and computational operations. Skyline storage scheme was used in the modules of old version. In this paper, several sparse matrix storage schemes that show the better performance compare to the Skyline storage are implemented in both modules. Impacts of sparse matrix storage schemes on the total execution time and required memory areevaluated in both thermal and solid analysis of a large scale model.

This paper describes a large-scale 3D magnetostatic analysis using the Domain Decomposition Method (DDM). To improve the convergence of an interface problem of DDM, DDM based on the Conjugate Residual (CR) method or the MINimal RESidual (MINRES) method is proposed. The CR or MINRES method improves convergence rate and shows stable convergence behavior in solving the interface problem, compared with the Conjugate Gradient (CG) method, and reduces the computation time for a large-scale problem with 100 million degrees of freedom.

This paper describes the Domain Decomposition Method (DDM) for high frequency electromagnetic field problems. As an efficient iterative method to solve a complex symmetric system arising from the formulation of the E method, a DDM based on a MINimal RESidual (MINRES)-like iterative procedure is presented. Numerical experiments are demonstrated to report a faster convergence compared to conventional methods.

This paper describes large-scale full-wave analyses of electromagnetic fields using numerical human body models. This is achieved by the use of the parallel finite element method with iterative domain decomposition. The numerical human body models provided by the National Institute of Information and Communications Technology (NICT) in Japan are composed of voxel data with all sides of 2mm and include skin layers, blood vessels, bones, internal organs, etc. distinguished by material properties. By using a High Performance Computing Infrastructure (HPCI), the user can evaluate the electromagnetic filed distribution inside the whole body model.