Yet Another Particle Simulation

March 27, 2025 ยท View on GitHub

Galaxy Simulation NVIDIA CUDA License

A high-performance astrophysics simulation modeling black hole migration resulting from the merger of a large host galaxy with a smaller galaxy. The simulation uses CUDA acceleration and Fourier transforms to compute gravitational interactions between 100 million particles, including two supermassive black holes (SMBHs). The project aims to study the final parsec problem in galaxy mergers.

image

Table of Contents

About

This project is part of UTSC PHYD57, focusing on galaxy evolution simulations. The model includes:

  • 100 million particles representing stars, dark matter, and gas.
  • 2 Supermassive Black Holes (SMBHs) resulted from recently merged galaxes
  • Fourier-based gravitational computations to resolve forces in O(NLog(N)) computations
  • CUDA-accelerated particle-grid interpolations and integration.

The simulation aims to observe the behavior of galaxies as they merge, with a particular focus on the final parsec problem, where the SMBHs slow down due to dynamical friction.

Features

  • CUDA Acceleration: All computations are GPU-accelerated
  • FFT-based Gravitational Solvers: Uses CUFFT to compute accelerations in Fourier space.
  • Integration Method:Leap Frog 2nd Order Symplectic Integrator

Dependencies

The following tools and libraries are required to build and run the simulation:

  • NVIDIA CUDA Toolkit (>=9.2)
  • PGI Fortran Compiler: For compiling CUDA Fortran code.
  • CUFFT Library: For GPU-accelerated Fourier transforms.

Installation

Clone the Repository

git clone https://github.com/Daniel-Harrington/YAPS.git
cd src

Compilation

Use the following command to compile the code:

pgf95 -mp -Mcuda=fastmath,cc35,cc50,cc60,fma,flushz,lineinfo -ta=nvidia -tp=haswell -O2 -fast -Minfo=all -mcmodel=medium Galaxy_Collison.f95 -o Galaxy_Collision -L/usr/local/cuda-9.2/lib64 -lcufft -lcupti

Usage

Running the Simulation

After successful compilation, run the executable to start the simulation:

./Galaxy_Collision

Input Parameters

  • Number of particles (N): 100,000,000 (adjustable in the source code).
  • Grid dimensions: Default nx=512, ny=512, nz=256.
  • Timestep (( \Delta t )): Default 0.1.