Organization of Talk:

1
http//www.phy.ornl.gov/tsi/

• Organization of Talk
• TSI Project Description
• TSI-ISIC Collaborations
• Ongoing TSI-SDM Projects
• A Look Ahead

2
(No Transcript)
3
Investigator Team

• Cross-Cutting Team
• Long-Term Collaborations
• Structured like SciDAC

TOPS

• Linear System/Eigenvalue Problem Solution
  Algorithms for Radiation Transport and Nuclear
  Structure Computation
• Dongarra (UT, ORNL)
• Saied (UIUC, NCSA)
• Saylor (UIUC, NCSA)

• Radiation Transport/
• Radiation Hydrodynamics
• Blondin (NC State)
• Bruenn (FAU)
• Hayes (UCSD)
• Mezzacappa (ORNL)
• Swesty (SUNYSB)

• Supernova Science
• Blondin
• Bruenn
• Fuller
• Haxton
• Hayes
• Lattimer
• Meyer (Clemson)
• Mezzacappa
• Swesty

TOPS
CCA PERC TSTT

• Nuclear Structure Computations
• for EOS and Neutrino-Nucleus/
• Nucleon Interactions
• Dean (ORNL, UT)
• Fuller (UCSD)
• Haxton (INT, Washington)
• Lattimer (SUNYSB)
• Prakash (SUNYSB)
• Strayer (ORNL, UT)

SDM

• Visualization
• Baker (NCSA)
• Toedte (ORNL)

4

• Goal
• Ascertain the explosion mechanism(s).
• Reproduce supernova phenomenology (element
  synthesis neutrino,
• gravitational wave, and gamma ray signatures
  neutron star kicks
• gamma ray burst connection)

• Relevance
• Dominant source of many elements in the
  Universe.
• Given sufficiently well developed models, serve
  as laboratories for
• fundamental nuclear and particle physics
  that cannot be explored
• in terrestrial laboratories.
• Driving application in computational science
  (radiation transport,
• hydrodynamics, nuclear physics, applied
  mathematics, computer
• science, visualization).

• Paradigm
• Result from stellar core collapse and
• bounce in massive stars.
• Radiatively driven (perhaps some are
• MHD driven, or both).

5
Convection

• Need Boltzmann Solution
• Need Angular Distribution
• Need Spectrum
• Gray Schemes Inadequate
• Spectrum Imposed
• Limited Angular Information
• (Few Moments)
• Parameterized
• (No First Principle Solution)
• The bar is high! (10 effects can
• make or break explosions.)

6
1D
1D
0D
0D
Neutrino Energy
Lightbulb
FLD
MGFLD
MGBT
D
Space
Burrows, Hayes, and Fryxell
Janka and Mueller
Mezzacappa et al.
1D
Herant et al.
TSI Year 1
TSI Year 2
2D
Swesty
Fryer and Heger
Past Transport in 2D Models D Diffusion FLD
Flux-Limited Diffusion MGFLD Multigroup
FLD MGBT Boltzmann Transport
TSI Year 3
TSI Year 2
3D
Gray Models
7
Explosion Mechanism Open Questions

• What is the Recipe for Explosion?

Neutrino Heating
Convection
General Relativity
Rotation
Magnetic Fields

• Are there multiple mechanisms?
• Neutrino-driven supernovae
• MHD-driven supernovae
• Supernovae driven by both neutrinos and MHD
  effects

• One mechanism for a class of stars?

• Is the mechanism tailored to the individual star?

8
Supernova Science
Hydrodynamics Explicit Differencing Reactive
Flows Newtonian General Relativistic
Nuclear, Weak Interaction Physics Thermodynamics (
Composition), Neutrino Sources and Interactions
Radiation Transport Implicit Differencing MGFLD Pr
econditioners Sparse System Solvers MGBT Precondit
ioners Sparse System Solvers (Matrix Free)
9
ISIC Collaborations TOPS

• Nonlinear Algebraic Equations
• Linearize
• Solve via Multi-D Newton-Raphson Method
• Large Sparse Linear Systems

Boltzmann Equation nonlinear integro-PDE

• Implicit Time Differencing
• Extremely Short Neutrino-Matter
• Coupling Time Scales
• Neutrino-Matter Equilibration
• Neutrino Transport Time Scales

Memory Requirements (assuming matrix-free
methods) 10s Gb up to 1/2 Tb
Progress Sparse Approximate Inverses for 2D
MGFLD (Saylor, Smolarski, Swesty J. Comp.
Phys.) ADI-Like Preconditioner for Boltzmann
Transport (DAzevedo et al. Precond 2001,
NLAA) AGILE-BOLTZRAN, V2D codes turned over to
TOPS for analysis and development.
10
ISIC Collaborations CCTTSS

• TSI Code
• F90 MPI Code
• Object-Oriented Design for Interoperability and
  Reuse
• Application Framework
• IBEAM Interoperability Based Environment for
  Adaptive Meshes
• NASA HPCC-Funded Project (PI Swesty)
• AMR PARAMESH

Goal Develop our framework to be
CCA-compliant. Initiated discussions with ANL,
LLNL, and ORNL members of CCTTSS.
11
ISIC Collaborations PERC

• Assess Code Performance on Parallel Platforms
• Identify Code Optimizations to Increase
  Performance

• TSI Code Suite
• Hydrodynamics
• VH-1 (PPM)
• ZEPHYR (Finite Difference)
• Neutrino Transport
• AGILE-BOLTZTRAN 1D General Relativistic Adaptive
  Mesh
• Hydrodynamics with 1D Boltzmann
  Transport
• V2D 2D MGFLD Transport Code
• V3D 3D MGFLD Transport Code (Under Development)
• 2D/3D Boltzmann Code (Under Development)

VH-1 numerical hydrodynamics algorithm scales
well.
Results for VH-1
12
ISIC Collaborations TSTT
Adaptive Quadratures (Direction Cosines) for
Multidimensional Radiation Transport

• Greatest challenge to completing 3D Boltzmann
• simulations is memory.
• Minimize number of quadratures to minimize
• memory needs while maintaining physical
• resolution. (Also important for 1D/2D MGBT.)
• Optimization Problem

Results for 1D Boltzmann Transport on Milne
Problem (DAzevedo)
Extended Core
Compact Core
13
Collaboration with Supporting Base Projects
Networking

• Identify Optimal Paths in Our Collaborative
  Visualization Server-Client Model
• Maximize Bandwidth along these Paths (Not
  Achieved Using Current Protocols)

• Participated in ORNL Workshop on DoE
  High-Performance Network RD
• and Applications
• Convey TSI Needs to Networking Team
• Participate in White Paper to Define and
  Develop Interface between Efforts

14
ISIC Collaborations SDM

• Use PROBE environment for staging data between
  simulation platforms and
• end-user visualization platforms.
• Develop new data analysis techniques/tools
  tailored to our application, allowing
• (a) data reduction and (b) discovery
  potential.
• Use of agent technology for distributed data
  analysis (data analysis must be
• done in parallel to achieve reasonable
  throughputs).

15

• Latest TSI 2D/3D Models
• Hydrodynamics only.
• Focused on understanding 2D/3D flow and its
• coupling to shock wave.
• Convectively stable.
• 2D model exhibits bipolar explosion (due to
• nonlinear flow-shock interaction).
• 3D model exhibits similar long-wavelength
• behavior. Key finding.
• New rolling flows identified.
• AAS Meeting Ap.J. Submitted

2D Model
3D Model
16
SDM Data Needs

• 3D Hydrodynamics Run
• 5 Variables (Density, Entropy, Three Fluid
  Velocities)
• 1024 X 1024 X 1024 Cartesian Grid
• 1000 Time Steps

20 Terabyte Dataset
17
SDM PROBE
Bulk Storage
IBM and Compaq Supercomputers
Production HPSS
Probe HPSS
CAVE
Data Reduction, pre-Vis Manipulation
Rendering
Stingray RS/6000 S80
Marlin RS/6000 H70
Origin 2000 Reality Monster
Other Probe Nodes
External Esnet Router
Utilize PROBE until data manipulations,
partitioning of manipulations, and bandwidths
are known. PROBE is adaptable!
18
SDM Data Analysis

• Data Reduction
• Scientific Discovery

Density distribution at last slice reconstructed
from 30 principal components. (First slice
reconstructed from 3!)
Original density distribution at final time
slice.
19
Integrate into collaborative visualization?
PCA Data Reconstruction
Server
Client
PCA Data Reduction
Networking Technologies
20
New Windows on the Universe?

• Scientific Discovery
• Can we use current data analysis tools to better
  understand and better
• quantify supernova physics?
• Can we develop new tools that will provide a new
  view of supernova physics?

21
SDM Agents
Team of Agents Divides Up Data

• Current data analysis techniques performed on
• a 10 Gb dataset would take 3 years to complete!
• Need for distributed data analysis.

• Agents perform analysis on subsets of data.
• Merge results via peer-to-peer agent
  collaboration and negotiation.

Both data analysis and visualization can employ
agent technology.

GUI/environment for the selection and
(distributed) use of data analysis tools and the
display of pre- and post-processed data.
22
SDM A Look Ahead

• We have a testbed!
• Existing 2D/3D datasets.
• Three TSI nodes NCSA, NCSU, ORNL.
• Testbed for collaborative visualization tools.
• Testbed for networking.
• PROBE being used to postprocess the data.
• PCA has been used successfully for data
  reduction.
• Agents have been used in a cross-platform demo
  utilizing this data.
• Continue to explore possibilities.
• Continue extensive interactions between TSI
  modelers and
• SDM data analysts.
• Can we integrate data analysis and agent
  technology (distributed data
• analysis) with collaborative visualization?
• Will existing tools/new tools lead to scientific
  discovery?
• New views on the data?
• Better quantification of supernova dynamics?