Extreme-Scale Stochastic Particle Tracing for Uncertain Unsteady Flow Visualization and Analysis

Area of Science:

• Fluid Dynamics
• Computational Science
• Data Visualization

Background:

• Estimating transport behaviors in uncertain unsteady flows is computationally intensive.
• Traditional methods require numerous Monte Carlo simulations for particle tracing, limiting scalability.
• Visualizing and analyzing uncertain flow fields presents significant challenges.

Purpose of the Study:

• To present an efficient and scalable solution for estimating uncertain transport behaviors using stochastic flow maps (SFMs).
• To reduce the computational cost associated with computing flow maps from uncertain flow fields.
• To enable effective visualization and analysis of uncertain unsteady flows.

Main Methods:

• Decoupling time dependencies in SFMs to process shorter sub-time intervals independently.
• Employing adaptive refinement to minimize the number of simulation runs.
• Implementing parallel processing over tasks (particle packets) for MPI/thread hybrid programming.
• Utilizing a task model that supports CPU/GPU coprocessing.

Main Results:

• Demonstrated significant reduction in computational cost for estimating stochastic flow maps.
• Achieved high efficiency through parallelization and CPU/GPU coprocessing.
• Showcased scalability on supercomputers (Mira and Titan), tracing billions of particles rapidly.

Conclusions:

• The proposed method offers an efficient and scalable approach to analyze uncertain unsteady flows.
• Stochastic flow maps provide a powerful tool for visualizing and understanding complex fluid dynamics.
• The computational advancements enable the analysis of large-scale particle transport with unprecedented speed.