1 / 28

Multi-Scale Challenge

A New Paradigm for Time Integration of Multi-Scale Systems: Self-Adaptive Event-Driven Simulation Computational Group at SciberQuest, Inc. Solana Beach, CA in collaboration with Georgia Tech. Multi-Scale Challenge. Large disparity in spatial and temporal scales due to system inhomogeneity

tallis
Download Presentation

Multi-Scale Challenge

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. A New Paradigm for Time Integration of Multi-Scale Systems:Self-Adaptive Event-Driven SimulationComputational Group at SciberQuest, Inc. Solana Beach, CA in collaboration with Georgia Tech

  2. Multi-Scale Challenge • Large disparity in spatial and temporal scales due to system inhomogeneity • Encompasses many scientific fields (e.g., space physics, fusion, climate modeling, biology, materials science, etc.)

  3. 3D Global Hybrid-PIC Simulation of Earth’s Magnetosphere: Karimabadi et al., 2004

  4. Computational Load

  5. What is Wrong with Time Stepping? • DEGENERACY: • Idle Systems: • Stiff Systems GOAL: Update micro-states (particles, fields etc.) with rates determined by local physical time scales • SOLUTION: • For df/dt=S solve dt/df=1/S by predicting Dt based on rate of change, df/dt and threshold (quantum) value, Df • Monitor and correct solution behavior via causal (not parametric) time dependencies

  6. Self-Adaptive DES • Fast (no redundant computation) • Accurate (validated against TDS) • Stable (even in super-Courant regimes) • Multi-D (extendable) • Parallel (MPI) • Genericframework (PDE,PIC) • Ideal for nonuniform meshes (AMR, unstructured, mapped)

  7. Discrete Event Simulation (DES)

  8. Event Processing Advance: update solution Synchronize: propagate change Schedule: predict new update

  9. Convection-Diffusion-Reaction

  10. What About Flux Conservation? “It is what makes time travel possible – the flux capacitor.” - Dr. Brown, “Back to the Future”.

  11. Flux Synchronization

  12. Linear Diffusion-Reaction

  13. Linear Advection

  14. Heat Wave(D=const, )

  15. Nonlinear Diffusion ( )

  16. Diffusion-Convection

  17. PIC-DES Simulation • Particles are pushed with individual (time varying) Dt’s • Fields are updated with rates proportional to local frequencies • PIC and field micro-states are synchronized via “wake-up” calls

  18. PIC-Field Synchronization

  19. Hybrid-DES Model • A and E are cell-centered, B is face-centered • Use (through ) to predict • Get A-quantum, • Integrate A asynchronously (through ) • Get • Push particles with

  20. Weak Fast Shock (M=2, )

  21. Rotational Discontinuity (M=0, )

  22. Intermediate Shock (M=1.05, )

  23. Strong Fast Shock (M=6, )

  24. Parallel Issues PTDS: key metric is scaling with the number of processors (performance) PDES: issue of key metric more complex (resolution + performance)

  25. Preemptive Event Processing (PEP) • Generic approach to parallelization of physics-based DES codes • Event pipeline minimizes inter-processor communication • Ideal for high-resolution computing • Load balancing is based on Event Distribution Function (EDF)

  26. Summary • Elimination of time degeneracy leads to novel, event-driven algorithms with superior performance metrics: - Accuracy - Stability - Speed • Applicable to full-PIC, CFD/MHD and Vlasov simulation models • Ideal in combination with nonuniform meshes

More Related