high performance computing on graphics processing units: hgpu.org

The purpose of this project is to develop a computer model to investigate the formation and life cycle of classical novae. A nova is an orbiting system consisting of a white dwarf and star. Over time, the white dwarf pulls hydrogen gas from the star which gathers onto the surface of the white dwarf (the accretion phase). Once the critical conditions are reached for thermonuclear runaway (TNR) the hydrogen gas on the surface of the white dwarf will ignite and explode as a rapidly expanding gas shell (the nova phase). Each particle in my computer simulation represents a volume of hydrogen gas and is initialized randomly in the outer shell of the companion star. The core of the star is modeled by a repulsive wall using a Wendland weight function. The motion of each particle is computed by solving Newton’s equations of motion. The forces on each particle include: gravity, centrifugal, Coriolis, friction, and Langevin. The friction and Langevin thermostat forces are used to model the viscosity and internal pressure of the gas. A rotating frame is used and the centrifugal and Coriolis forces are included to model the rotation. A velocity Varlet method with a one second time step is used to compute velocities and positions of the particles. A new particle recycling method was developed which was critical for computing an accurate and stable accretion rate (the rate at which particles stick to the white dwarf) and keeping the particle count reasonable. TNR occurs with only around 500 particles with the recycling method, compared to 500,000 without particle recycling. I used C++ with OpenCL and MPI to parallelize my simulations which were run on two Nvidia GTX 580s. The simulations used between 32,768 and 1 million particles and each run required between 1 and 10 hours to complete. I used my model to investigate how parameters such as the white dwarf mass, star mass and separation will affect the evolution of the nova. These parameters affect the accretion rate, the frequency of the nova explosions and light curves (the Nova’s visual magnitude versus time). My computed light curve results were fit to the observed light curve from the 2010 U Scorpii nova outburst. The program parameters were then varied to determine the effects they had on the light curve and frequency of nova explosions. When a small white dwarf mass was used, the light curve peaked higher and cooled off slower. The opposite was observed for a heavier white dwarf mass; the light curve peaked lower and cooled off faster. As the white dwarf mass is increased from 0.4 to 1.4 solar masses, the time period between explosions decreases dramatically from 100,000 to 1 yr. My simulations also show that the companion star blocks the expanding gas shell leading to an asymmetrical expanding shell. The results of my simulations compared well with the professional 1D and 3D hydrodynamic simulations and observed experimental data. I also investigated the effects of other parameters on my simulation results, such as: the time step, number of particles, Langevin Keff, friction coefficient, and particle mass.