Adaptive Parallel Methods for Reactive Fluid Dynamics and Particle Dynamics Driven by Applications to Galaxy Formation Problems

Objective: The project aims at simulating the physics of galaxy formation. This is a Grand Challenge problem for several reasons: 1) It is fully 3D, requiring about 512^3 resolution; thus, it needs to be done on a massively parallel processor (MPP) machine with several GFLOPS capability. 2) It requires multifluid modeling and also multiphase modeling. 3) It has a significant dynamic range, thus requiring the use of adaptive mesh refinement (AMR). 4) Fluid solvers and particles need to be used together. 5) All these steps need to be done in an MPP environment.

Approach: The fluid solvers take the form of higher-order Godunov schemes. The Poisson solvers take the form of particle-particle particle-mesh (P3M) and multigrid techniques. This is all brought together using the object-oriented C++ language.

Accomplishments:

• Balsara working on a paper with R. Melhem on parallelization of fluid algorithms. It represents HPF-based strategies to complement C++ efforts.
• Finished load-balancing algorithm for P3M on the CRAY T3D in cooperation with B. He and R. Melhem.
• Worked out a natural way to represent P3M in HPF.
• Tested C++/A++ version of RIEMANN++ and verified that it works.
• Making slight modifications required to accommodate auto-parallelizing version running under the auto-parallelizing language P++.
• Tested and verified AMR++ with dynamic adaptation for Poisson problem.
• Working on incorporating AMR++ and RIEMANN++ together.
• Achieved several GFLOPS of performance with optimal scaling on both fluid and P3M algorithms.
• Quinlan participated in the formulation of an HPC++ standard.
• P++ has been chosen as a de facto Phase I standard for HPC++.
• Simulated the formation of a galactic interstellar medium in a self-consistent way.
• Carried out multiphase simulations of starburst galaxies.

Significance: Understanding how our solar system formed addresses one of the stated goals of NASA and provides insight into the conditions necessary for the formation of planets. Applying this knowledge to other stellar systems will enable us to better assess the frequency of other solar systems. This type of theoretical work complements the observational searches that are underway to detect other solar systems.

Status/Plans: Achieving several GFLOPS of performance demonstrates the feasibility of portable standard code and independence from a particular MPP architecture. The fact that simulations have been carried out means that the full range of input physics has been put in. That P++ has been chosen as a phase I standard means that NASA-sponsored research has become an international standard in the parallel computation community!

Point of Contact:

Dinshaw S. Balsara
National Center for Supercomputing Applications
University of Illinois at Urbana-Champaign
u10956@ncsa.uiuc.edu
217-244-1481