Masaharu Matsumoto

The power generation characteristics of a disk-shaped magnetohydrodynamic generator driven by rotating detonation with potassium-seeded hydrogen–air mixture are examined via quasi-two-dimensional numerical simulations. For unburned gas injection in the outward and inward directions, the rotating direction of the detonation wave is set so that the azimuthal velocity enhances the radial electromotive force. The simulation results show that the power output in the outward flow configuration exceeds that of the inward flow configuration. In the outward flow configuration, the radial gas velocity around the contact surface is the main contributor to the power generation, while in the inward flow configuration, the low gas velocity in almost the entire region and the vortices around the contact surface lead to poor power output. The characteristics in not only the power output mode but also the power input mode in which the voltage is applied externally between the electrodes are investigated. Under the positive load voltage, high enough for providing negative current, the Lorentz force occurs in the same direction as the propagation of the detonation wave; otherwise, the Lorentz force is induced in the opposite direction of the detonation wave.

Some deep learning-based methods, Deep Galerkin Method (DGM), Physics Informed Neural Networks (PINNs), and others, have been suggested for solving partial differential equations. In these methods, a loss function for training a deep neural network is formulated, so that differential operators, boundary conditions, and initial conditions of the intended partial differential equation are satisfied. This study applied a deep learning-based method to solve viscous and resistive magnetohydrodynamic (MHD) equations. In particular, the deep learning-based method is applied to the 1D Brio-Wu shock wave problem, 2D Balsara-Spicer MHD Rotor problem, 2D Orszag-Tang MHD Vortex problem, and 2D magnetosphere problem. The approximated solution obtained by the deep learning-based method generally agrees with that obtained by conventional numerical methods.

In this study, we evaluate the performance of sparse matrix-vector multiplication (SpMV) using the compressed row storage (CRS) format on CPUs and MIC. We focus on the relationship between OpenMP scheduling and performance. The performance of SpMV is measured using various OpenMP scheduling settings and the results are analyzed, which show that OpenMP scheduling has a considerable effect on the performance of SpMV. We confirm that some scheduling settings resulted in performance improvements compared with default scheduling for particular matrices. The results of the evaluation show that the performance of SpMV is improved by up to 1.57 times compared with SPARC64 IXfx, 2.47 times compared with Xeon Ivy Bridge-EP, and 2.26 times compared with Knights Corner. Next, we modify the SpMV function of OpenATLib, an auto-tuned numerical library, to consider the scheduling of optimization as an additional SpMV implementation. We measure the performance of the GMRES solver and obtain performance improvements of up to 11.4%. These results will help to improve the performance of various numerical calculation applications.