[Paper Review] GALAXY package for N-body simulation

J. A. Sellwood|arXiv (Cornell University)|Jun 25, 2014

Astronomy and Astrophysical Research20 citations

TL;DR

The GALAXY package is a highly optimized N-body simulation code for modeling isolated, collisionless stellar systems such as galaxies, using particle-mesh methods with multiple force computation techniques. It achieves 50 to 200 times higher performance than the widely used Gadget-2 tree code for certain applications, particularly large-scale spherical and disk-like systems, while maintaining high accuracy through advanced grid-based gravitational solvers and on-the-fly analysis tools.

ABSTRACT

This posting announces public availability of the GALAXY software package developed by the author over the past 40 years. It is a highly efficient code for the evolution of (almost) isolated, collisionless stellar systems, both disk-like and ellipsoidal. In addition to the N-body code galaxy, which offers eleven different methods to compute the gravitational accelerations, the package also includes sophisticated set-up and analysis software. This paper gives an outline of the contents of the package and provides links to the source code and a comprehensive on-line manual. While not as versatile as tree codes, the particle-mesh methods in this package are shown, for certain restricted applications, to be between 50 and 200 times faster than a widely-used tree code.

Motivation & Objective

• To provide a high-performance, publicly available N-body simulation package for isolated stellar systems, particularly galaxies.
• To overcome the computational bottleneck in collisionless simulations by maximizing efficiency through optimized particle-mesh methods.
• To support accurate initial condition generation and real-time analysis for complex multi-component systems like disks with bulges and halos.
• To enable researchers to achieve higher-resolution simulations with improved numerical fidelity despite the code's steep learning curve.
• To offer a free, open-source alternative under the GNU GPL with long-term maintainability and extensibility.

Proposed method

• The core N-body code, galaxy, uses 11 distinct particle-mesh methods to compute gravitational accelerations, including spherical and cylindrical grid configurations with harmonic decomposition.
• It employs FFT-based solvers from FFTPAK and supports both 2D and 3D grid geometries with variable harmonic modes (e.g., l ≤ 4, m ≤ 8) for force accuracy.
• On-the-fly analysis is implemented to record key dynamical quantities (e.g., energy, angular momentum, density profiles) at each timestep, minimizing storage and enabling compact data output.
• The package includes specialized initialization tools (e.g., begin, start) for generating equilibrium initial conditions with multiple mass components, including disks, bulges, and halos.
• It supports both single-processor and MPI-parallel execution via separate executables (e.g., galaxy_mpi), enabling scalable performance on multi-core systems.
• The code integrates with external libraries: FFTPAK for FFTs, PGPLOT for plotting, and optionally NAG for advanced features, with fallbacks for open-source alternatives.

Experimental results

Research questions

• RQ1 How can particle-mesh methods be optimized to achieve extreme performance in collisionless N-body simulations of isolated galaxies?
• RQ2 To what extent does the GALAXY package outperform established tree-code methods like Gadget-2 in terms of computational efficiency for specific galaxy models?
• RQ3 Can on-the-fly analysis in N-body simulations significantly reduce data storage requirements without sacrificing scientific utility?
• RQ4 What are the practical limitations and risks of using a high-performance but complex code like GALAXY for new users in astrophysical research?
• RQ5 How does the performance advantage of GALAXY scale with particle count and number of processors across different system geometries?

Key findings

• The GALAXY code runs approximately 50 times faster than Gadget-2 for spherical Plummer models across all tested particle counts and processor counts.
• For thickened Kuzmin-Kutuzov disk models, the performance advantage of GALAXY increases with particle count, reaching over 200 times faster than Gadget-2 at N ≈ 10^8 particles.
• The particle-mesh method in GALAXY exhibits minimal performance degradation with increasing N due to fixed grid overhead, enabling efficient large-scale simulations.
• On-the-fly analysis reduces data storage needs significantly, as results are saved compactly at each analysis step without requiring post-processing.
• The code achieves high performance even on single processors, with the potential for massive speedups when used in parallel via MPI-enabled executables.
• Despite its speed, the package requires significant user expertise—users are advised to spend weeks mastering it before producing publishable results.