Porting the grid-based 3D+3V hybrid-Vlasov kinetic plasma simulation Vlasiator to heterogeneous GPU architecturesBattarbee, Markus; Papadakis, Konstantinos; Ganse, Urs; Hokkanen, Jaro; Kotipalo, Leo; Pfau-Kempf, Yann; Alho, Markku; Palmroth, Minna
doi: 10.1088/1742-6596/2997/1/012010pmid: N/A
Vlasiator is a space plasma simulation code which models near-Earth ion-kinetic dynamics in three spatial and three velocity dimensions. It is highly parallelized, modeling the Vlasov equation directly through the distribution function, discretized on a Cartesian grid, instead of the more common particle-in-cell approach. Modeling near-Earth space, plasma properties span several orders of magnitude in temperature, density, and magnetic field strength. In order to fit the required six-dimensional grids in memory, Vlasiator utilizes a sparse block-based velocity mesh, where chunks of velocity space are added or deleted based on the advection requirements of the Vlasov solver. In addition, the spatial mesh is adaptively refined through cell-based octree refinement. In this paper, we describe the design choices of porting Vlasiator to heterogeneous CPU/GPU architectures. We detail the memory management, algorithmic changes, and kernel construction as well as our unified codebase approach, resulting in portability to both NVIDIA and AMD hardware (CUDA and HIP languages, respectively). In particular, we showcase a highly parallel block adjustment approach allowing efficient re-ordering of a sparse velocity mesh. We detail pitfalls we have overcome and lay out a plan for optimization to facilitate future exascale simulations using multi-node GPU supercomputing.
Prefacedoi: 10.1088/1742-6596/2997/1/011001pmid: N/A
The Center for Space Plasma and Aeronomic Research (CSPAR) at the University of Alabama in Huntsville (UAH) and Maison de la Simulation, a joint laboratory at the French Alternative Energies and Atomic Energy Commission (CEA), National Center for Scientific Research (CNRS), University of Paris-Saclay, and University of Versailles-St Quentin, organized the 16th International Conference on Numerical Modeling of Space Plasma Flows (ASTRONUM-2024) on July 1—6, 2024 in La Rochelle, France.The Program Committee consisted of Nikolai Pogorelov (University of Alabama in Huntsville/CSPAR, USA, chair), Edouard Audit (CEA/Maison de la Simulation, Gif-sur-Yvette, France, co-chair), Amitava Bhattacharjee (Princeton Plasma Physics Laboratory, USA), Phillip Colella (Lawrence Berkeley National Laboratory, USA), Tomoyuki Hanawa (Chiba University, Japan), Maria Elena Innocenti (Ruhr University Bochum, Germany), Kanya Kusano (Nagoya University, Japan), Dongwook Lee (University of California, Santa Cruz), Jon Linker (Predictive Science Inc., USA), Anthony Mezzacappa (University of Tennessee, Knoxville, USA), Andrea Mignone (University of Turin, Italy), and Gary P. Zank (University of Alabama in Huntsville, USA).The conference attracted over 60 scientists from 10 countries representing different branches of the plasma simulation community. The distinctive feature of this conference is a combination of diverse research topics, all of which are essential for performing high-resolution, continuum mechanics and particle, simulations of physical phenomena in space physics and astrophysics. Among such topics were software packages for modeling and analyzing plasma flows; advanced numerical methods for space and astrophysical flows; large-scale fluid-based, kinetic, and hybrid simulations; turbulence and cosmic ray transport; and magnetohydrodynamics. The discussed applications included cosmology and galaxy formation, supernova explosions, physics of the Sun-heliosphere-magnetosphere environments, space weather, the interstellar medium and star formation, stellar physics, experimental plasma physics, astrophysical accretion, numerical methods for ideal and non-ideal, relativistic and nonrelativistic MHD, etc. The proceedings volume is structured so that it covers all of these topics.The structure of the ASTRONUM conference series is based on the idea that modelers working in seemingly distant fields should have an opportunity to share their scientific achievements with the broad community of computational scientists performing numerical experiments. As in previous ASTRONUM meetings, we were interested in physical systems that are coupled across a multiplicity of spatial and temporal that incorporate diverse physical processes.The contributors to this volume are both young researchers and renowned experts in space physics and astrophysics, applied mathematics, and computational physics. This volume describes the application of numerical methods and the algorithms themselves, allowing us to discuss the challenges that theory imposes on numerical schemes for solving partial differential equations describing collisional and collisionless processes in space and astrophysical plasmas.We would like to thank the participants who submitted their papers to Proceedings of ASTRONUM-2024 and especially to those who reviewed manuscripts thus ensuring the high quality of this publication. We also are grateful to Valérie Belle (CEA, Maison de la Simulation, France) and Bruno Thooris (CEA, IRFU, France) for the excellent management of the conference.The volume will be useful to graduate and postgraduate students majoring in space physics, astrophysics, numerical, engineering, and applied mathematics. It is also aimed at specialists in applied mathematics, and various fields of physics that involve flows of partially ionized plasmas at both the collisional and collisionless levels.Nikolai V. Pogorelov, Edouard Audit, and Gary P. ZankFebruary 20, 2025
Peer Review Statementdoi: 10.1088/1742-6596/2997/1/011002pmid: N/A
All papers published in this volume have been reviewed through processes administered by the Editors. Reviews were conducted by expert referees to the professional and scientific standards expected of a proceedings journal published by IOP Publishing.• Type of peer review: Single Anonymous• Conference submission management system: Morressier• Number of submissions received: 20• Number of submissions sent for review: 19• Number of submissions accepted: 19• Acceptance Rate (Submissions Accepted / Submissions Received × 100): 95• Average number of reviews per paper: 1• Total number of reviewers involved: 19• Contact person for queries:Name: Nikolai PogorelovEmail: [email protected]: The University of Alabama in Huntsville
HelioCubed: A High Order Inner Heliosphere Simulation Code with a Mapped Cubed Sphere Grid and Adaptive Mesh RefinementSingh, Talwinder; Colella, Phillip; Van-Straalen, Brian; Bozhart, Chris; Pogorelov, Nikolai V.
doi: 10.1088/1742-6596/2997/1/012019pmid: N/A
We present the HelioCubed, a high-order magnetohydrodynamic (MHD) code designed for modeling the inner heliosphere. The code is designed to achieve 4th order accuracy both in space and in time. In addition, HelioCubed can perform simulations on mapped grids, such as those based on cubed spheres, which makes it possible to overcome stability limitations caused by the geometrical singularity at the polar axis of a spherical grid, thus enabling substantially larger time steps. HelioCubed has been developed using the high-level Proto library, ensures performance portability across CPU and GPU architectures, and supports back-end implementations, e.g., CUDA, HIP, OpenMP, and MPI. The code is compatible with the HDF5 library, which facilitates seamless data handling for simulations and boundary conditions derived from semi-empirical and MHD models of the solar corona. While presenting the results of preliminary simulations, we demonstrate that our simulations are indeed performed with 4th order of accuracy. Our approach ensures that HelioCubed solves the MHD equations preserving the radial flow to machine round-off error even on cubed-sphere grids. Solar wind simulations are performed using the boundary conditions provided by the Wang–Sheeley–Arge coronal model of the ambient solar wind. It also allows us to to simulate coronal mass ejections using observation-driven flux rope models. These capabilities make HelioCubed a versatile and powerful tool to advance heliophysics research and space weather forecasting.
Approximating Convective Urca Cooling in a Simmering White DwarfBoyd, B; Calder, A C; Townsley, D M; Zingale, M
doi: 10.1088/1742-6596/2997/1/012006pmid: N/A
Type Ia supernovae are bright thermonuclear explosions that play important roles in many areas of astronomy such as cosmology and galaxy evolution. The near Chandrasekhar mass white dwarf is a potential progenitor for these supernovae. This model entails a white dwarf accreting material from a companion and gaining mass to the point of igniting carbon fusion in the core. The onset of carbon fusion, called the simmering phase, drives convection and alters the evolution of the white dwarf as it approaches the thermonuclear explosion. A key factor during this phase is the convective Urca process which links convection with weak nuclear reactions that leak energy from the star. To study the effects of the convective Urca process, it is vital to accurately model the turbulent convection in the core. This necessitates 3D hydrodynamic simulations which are computationally expensive. As a point of comparison and to aid in exploring initial conditions, we use the “quick mixing” approximation, which assumes convective mixing is efficient enough to produce a uniform composition in the convection zone. Utilizing this approximation, we can predict the ratio of the A = 23 Urca pair as well as the resulting neutrino loss rates without running full 3D simulations. We compare the results of a 3D hydrodynamic simulation, run using the low Mach number hydrodynamic code MAESTROeX, to the quick mixing calculation. Additionally, we investigate how varying the size of the convection zone influences the convective Urca process and sets approximate bounds on reasonable initial conditions.
GRMHD simulations of accretion processes onto merging massive black hole binaries embedded in a slab of gasFedrigo, G; Cattorini, F; Giacomazzo, B; Colpi, M
doi: 10.1088/1742-6596/2997/1/012004pmid: N/A
We conduct general relativistic magnetohydrodynamic simulations of merging equal-mass spinning black holes within an equatorial thin slab of magnetized gas. The gas slab’s rest-mass density follows a Gaussian profile symmetric about the equatorial plane, and is initially either stationary or Keplerian in rotational support. Our study investigates configurations where the black holes are nonspinning, have spins aligned with the orbital angular momentum, or have misaligned spins. As part of our diagnostic approach, we monitor both the accretion of matter onto the black hole horizons and the Poynting luminosity. During the inspiral phase, configurations with nonzero spin exhibit modulations in the mass accretion rate that correspond to the orbital frequency and its harmonics. Frequency analysis suggests these modulations may be a typical characteristic of inflows around merging binaries. Unlike previously studied binary models, none of our current simulations display a significant post-merger increase in the mass accretion rate, suggesting that peak luminosity may not be observable at the merger time in future electromagnetic detections.
Algorithmic Advancements for High-Order Self-Gravitating HydrodynamicsHanawa, Tomoyuki; Dean Mullen, Patrick
doi: 10.1088/1742-6596/2997/1/012012pmid: N/A
Self-gravity plays a key role in the formation and evolution of many astronomical objects. Though gravity is often dominant at large scales, other forces (e.g., gas pressure gradients, radiation, and/or magnetic fields) often compete. It is therefore essential for numerical simulations to evaluate their interplay accurately and robustly. Hanawa & Mullen [1] derived a 4th-order accurate finite volume scheme to solve the equations of self-gravitating hydrodynamics on a uniform Cartesian grid. In this work, we supply improvements to the algorithm that (1) mitigate spurious gravitational circulation and (2) greatly simplify the evaluation of the high order corrections. The proposed algorithm provides the gravitational acceleration (ρg) and the gravitational energy release (ρv · g) as source terms for the hydrodynamic equations, all while preserving conservation of linear momentum. Spurious heating and/or cooling associated with truncation error in the numerical evaluation of the gravitational energy release decreases in proportion to the fourth power of the cell width. We demonstrate fourth order convergence on smooth problems (e.g., 3D inclined sound wave propagation and 3D equilibria). An application test tracks the spherical collapse of a polytrope by an imposed, sudden decrease of the central gas pressure; a bounce and second collapse (associated with a spherical accretion shock) are robustly captured by the high order algorithm.
A Gaussian process method with a neural network for compressible shock capturingLee, Dongwook; Waterhouse, Henry; DeGrendele, Christopher
doi: 10.1088/1742-6596/2997/1/012018pmid: N/A
We present a machine learning-based shock-capturing approach called the GP-NN method. The new GP-NN method is similar to the existing a posteriori GP-MOOD method in the sense of combining high- and low-order solutions away from and near shocks, respectively, to balance the accuracy and stability of the solution. However, GP-NN is different from GP-MOOD in that GP-NN is an a priori method from the shock-capturing perspective. In GP-NN, we delve into integrating machine learning within the GP-MOOD framework to replace the a posteriori MOOD shock detection mechanism with a new a priori neural network (NN) shock detector. Using NN to distinguish shock cells from non-shock cells, our a priori GP-NN approach achieves an unparalleled equilibrium between high-order accuracy and stability, dynamically adjusting to local flow conditions. Most importantly, the NN-based a priori shock-capturing approach allows to mitigate two major drawbacks in the a posteriori MOOD shock detection mechanism such as workload imbalance in parallel computing and computational inefficiency in implicit solves. This sophisticated strategy optimizes numerical accuracy and stability while maintaining the overall predictive capabilities to simulate shock-dominant flows.
Network model of magnetohydrodynamic turbulenceBeck, Benjamin; Müller, Wolf-Christian
doi: 10.1088/1742-6596/2997/1/012002pmid: N/A
This contribution deals with a reduced model of fully developed homogeneous magnetohydrodynamic (MHD) turbulence which aims at delivering consistent energy distributions based on prescribed energy injection and dissipation mechanisms which might come from a super-ordinate numerical description. The Fourier-space model exploits a particular symmetry that is evident in the Elsässer formulation of the incompressible MHD equations, i.e. the individual and detailed ideal conservation of the Elsässer energies, E±(k) = |z±(k)|2/2 with z±(k) = v(k)±b(k). The ensuing simplification of nonlinear triad interaction to pair exchanges between Fourier modes allows to formulate the nonlinear dynamics of MHD turbulence as a superposition of pair exchanges in a highly flexible network of mode inter-connections. The resulting model is efficient, simple and highly flexible with regard to the incorporation of spatial and dynamical anisotropies or the inclusion of phenomenological concepts. This is demonstrated by the dynamical generation of Kolmogorov, Iroshnikov-Kraichnan and anisotropic Goldreich-Sridhar energy spectra.