SLIDE 1
1
Yuri Omelchenko SciberQuest, San Diego, CA Collaborators:
- H. Karimabadi, H. X. Vu (SciberQuest)
2
ASTRONUM-2013, July 1-5, 2013, Biarritz, France 1 2 3 - - PowerPoint PPT Presentation
ASTRONUM-2013, July 1-5, 2013, Biarritz, France 1 2 3 - - PowerPoint PPT Presentation
Yuri Omelchenko SciberQuest, San Diego, CA Collaborators: H. Karimabadi, H. X. Vu (SciberQuest) ASTRONUM-2013, July 1-5, 2013, Biarritz, France 1 2 3 Bowshock/foreshock physics Magnetosheath turbulence Ionospheric outflow 4
SLIDE 2
SLIDE 3
3
SLIDE 4
4
Bowshock/foreshock physics
Magnetosheath turbulence
Ionospheric outflow
Effect of O+ in magnetotail
Ring currents
Turbulence in the magnetotail
Transport /formation of boundaries
SLIDE 5
Solar Wind Inflow Boundary Condition Simulation Domain 1 20 x 30 x 30 RE
3
20 RE 100 RE 1 RE =130 c/ωp
Ω-1=0.5sec
1 RE =6 c/ωp
Ω-1=1.5sec
5
SLIDE 6
Magnetosphere is a Multiscale Coupled System:
Spatial scales vary from centimeters to 200 RE (span of 1011 spatial scales!) Temporal scales vary from less than milliseconds to days (span of 108 temporal scales!) Electron physics: e.g., controls reconnection rate Ion physics: e.g., accounts for formation of boundaries, transport, energization Dynamic M-I coupling: still missing in kinetic simulations
- Requires yottaflops (1024) and beyond
SLIDE 7
SLIDE 8
8
SLIDE 9
- Time-Driven (Stepped) Simulation (TDS)
stability issues (for dt > CFL)
diffusion/dispersion issues (for dt < < CFL)
inactive regions are still time-stepped
local time stepping has synchronization issues
- Discrete-Event Simulation (DES)
updates are driven by physical changes (= > speedup)
changes are always limited (= > accuracy/stability)
arbitrary grids may be considered
adaptive synchronization via event preemption Q: Can we advance solution in time asynchronously in accordance with locally varying time scales?
SLIDE 10
Generalized equation: Event scheduling: Cells have different “trajectories” Cells have different ∆t’s
SLIDE 11
Dt 2Dt
T: CELL # 1 2 3 4 5 6 7 .. S(T) = Time-driven simulation updates the entire system Event-driven simulation updates active cells only S(0) S(Dt) S(2Dt)
update only active cells
S(0)
T1 T2
S(T1) S(T2)
update only active cells update all cells update all cells
T:
SLIDE 12
- 1. For df/ dt = R schedule events: for each “state” predict Δt
based on its trajectory, f(t) and accuracy threshold, Δf.
- 2. Update synchronize[preempt?]reschedule events.
SLIDE 13
SLIDE 14
1 4
SLIDE 15
1 5
PIC Δt Field Δt
SLIDE 16
SLIDE 17
SLIDE 18
1 8
SLIDE 19
t=1/Ωi t=3/Ωi t=7/Ωi
SLIDE 20
t=1/Ωi t=3/Ωi t=7/Ωi
SLIDE 21
t=1/Ωi t=3/Ωi t=7/Ωi
SLIDE 22
SLIDE 23
SLIDE 24
SLIDE 25
SLIDE 26
Density Tperp Tpar
SLIDE 27
Field Δt PIC Δt
SLIDE 28
Load Load imbalance
SLIDE 29
►Hybrid simulations are revealing new features of the
solar wind interaction with the magnetosphere and laboratory plasmas.
►Disparate time scales in global 3D simulations can be
addressed with DES. First parallel 3D DES runs have just been performed on 200 cpus.
►DES makes possible new advances in global hybrid
simulations by enabling: (i) model coupling (e.g., I-M), (ii) new electron physics (e.g., X-Hybrid).
► DEScan be applied to MHD, CFD, PIC(turbulence,
unstructured grids, higher-order methods, etc).
SLIDE 30
30