Akira Nukada | Searcharxiv

arXiv2026

Radiative Efficiency Enhancement by Electromagnetic Energy Dissipation in Strongly Magnetized Supercritical Accretion Flows around Kerr Black Holes

We investigate how black hole spin and the amount of magnetic flux affect electromagnetic energy dissipation and radiation transport in supercritical accretion flows. For this purpose, we perform general relativistic radiation magnetohydrodynamic simulations of MAD and SANE accretion flows with different black hole spins. In the high-spin MAD model, we find that a fraction of the electromagnetic energy extracted by the Blandford--Znajek mechanism is dissipated near the disk surface in the vicinity of the black hole. This dissipation significantly contributes to the generation of radiative energy and enhances the luminosity. The time-averaged radiative efficiency reaches $\eta_{\rm rad}=0.60$, which is much larger than $0.088$ for the non-spinning black hole case and $0.21$ for the weak-magnetic-flux case, corresponding to the SANE state. As a result, the effective trapping radius, defined as the radius at which the outward radiative luminosity becomes equal to the inward radiative luminosity, is $r_{\rm trap}=2.5r_{\rm g}$, comparable to the ISCO radius. This value is significantly smaller than $12r_{\rm g}$ for the non-spinning case and $8.5r_{\rm g}$ for the SANE state. These results suggest that the amount of magnetic flux accumulated on the black hole can affect the radiative properties of supercritical accretion flows and should therefore be considered, in addition to black hole mass, spin, and mass accretion rate, when interpreting observed luminosities and spectra.

astro-ph.HE↗

arXiv2019

Batched Sparse Matrix Multiplication for Accelerating Graph Convolutional Networks

Graph Convolutional Networks (GCNs) are recently getting much attention in bioinformatics and chemoinformatics as a state-of-the-art machine learning approach with high accuracy. GCNs process convolutional operations along with graph structures, and GPUs are used to process enormous operations including sparse-dense matrix multiplication (SpMM) when the graph structure is expressed as an adjacency matrix with sparse matrix format. However, the SpMM operation on small graph, where the number of nodes is tens or hundreds, hardly exploits high parallelism or compute power of GPU. Therefore, SpMM becomes a bottleneck of training and inference in GCNs applications. In order to improve the performance of GCNs applications, we propose new SpMM algorithm especially for small sparse matrix and Batched SpMM, which exploits high parallelism of GPU by processing multiple SpMM operations with single CUDA kernel. To the best of our knowledge, this is the first work of batched approach for SpMM. We evaluated the performance of the GCNs application on TSUBAME3.0 implementing NVIDIA Tesla P100 GPU, and our batched approach shows significant speedups of up to 1.59x and 1.37x in training and inference, respectively.

cs.DC↗