Validation Study of Fundamental Properties of Type Ia Supernovae Models
I'm a third year graduate student in Electrical Engineering, working with Andrew Packard in the Berkeley Center for Control and Identification (BCCI). My research focuses on stability and robustness verification for large-scale nonlinear systems. The 3 day short course in Parallel Programming piqued my initial interest in parallel programming. After taking this course, I expect to gain experience with parallel architectures and programming. Ideally, I would like to potentially gain insight as to how to exploit parallelism in my particular research.
Astronomers and Physicists believe that studying supernova will shed light on the mystery of dark energy and the expanding universe. In order to better understand supernova, there must first be a better understanding of the way in which stars explode. Past experiments have detonated white dwarf stars in two-dimensional sets, but it remained to be successfully simulated in three-dimensions. In 2007, the Flash Center team at University of Chicago achieved this goal.
The three-dimensional simulations indicated that the stars detonate in a supersonic process resembling diesel-engine combustion, which confirmed what the team already suspected from previous tests. This process is called gravitationally confined detonation (GCD). The new 3D white dwarf simulation shows the formation of a flame bubble near the center of the star. The bubble, initially measuring approximately 10 miles in diameter, rises more than 1,200 miles to the surface of the star in one second. In another second, the flame crashes into itself on the opposite end of the star, triggering a detonation.
The scientists found that the detonation conditions were robustly reached in their 3D simulations for a range of initial conditions and resolutions. Recent observations imply a compositional structure that is qualitatively consistent with that expected from these simulations. The behavior of turbulence is different in 3D than in 2D, and the cylindrical symmetry of the 2D simulations might enhance the focusing of the surface flow that triggers the detonation in the GCD model.
Platform and Tools
This 512 processor job was run on supercomputers using distributed memory platform. The group programmed the simulations using FLASH code, a parallel simulation code developed specifically for simulations of astrophysical thermonuclear flashes. FLASH is modular, in that it has been designed to allow users to configure initial and boundary conditions, change algorithms, and add new physical effects with minimal effort. It is adaptive in that it uses the PARAMESH library to manage a block-structured adaptive grid, placing resolution elements only where they are needed most. It is parallel, by using the Message-Passing Interface (MPI) library to achieve portability and scalability on a variety of different message-passing parallel computers.
Simulations ran at Lawrence Berkeley National Lab (LBNL) on Bassi (Top 500 rank peaked at 53 in November of 2005), LBNL on Seaborg (Top 500 rank peaked at 5 in June 2003), and at Lawrence Livermore National Lab (LLNL) on ASC Purple ( Top 500 rank peaked at 4 in November of 2006. Livermore Computing Center computer scientists assisted in making the Flash code run efficiently during the long simulation runs
The computer science team on the FLASH project is devoted to supporting FLASH in the critical areas of scalable performance and I/O, numerical libraries, distributed computing and advanced visualization techniques. As far as scalable performance, the group continues support of the MPI infrastructure work through MPICH and ROMIO; the performance visualization with SLOG-2 and Jumpshot; and parallel I/O libraries with MPI-IO, HDF-5, and NetCDF. Given that they project is operating in the FLASH code infrastructure, performance may be limited so that the problem adheres to the particular infrastructure.