Terascale Supernova Initiative

1
Terascale Supernova Initiative
Discovering New Dynamics of Core-Collapse
Supernova Shock Waves
John M. Blondin NC State University
Scientific Discovery through Advanced Computing
2
(No Transcript)
3
A Long History of Computational Physics

• 1966 Colgate and White
• Neutrino-Driven prompt explosion
• 1985 Bethe and Wilson
• Shock reheating via neutrino energy
  deposition
• 1992 Herant, Benz, and Colgate
• Convective instability above
  neutrino-sphere

4
The Modern Picture
5
It all starts with core collapse
6
First generation of 2D SN models hinted at a
low-order asymmetry in the shock wave at late
times (100s of msec after bounce).
Burrows, Hayes Fryxell 1995
7
Dynamics of the Supernova Shock Wave
When, Where and How is spherical symmetry is
broken?
8
Modeling post-bounce shock
9
SN Code Verification
Houck and Chevalier 1992 Blondin and Mezzacappa
2005
This post-bounce model provides an opportunity to
verify supernova codes against the results of a
linear perturbation analysis.
10
Spherical Accretion Shock Instability
11
SASI
Standing pressure waves within the cavity of a
spherical accretion shock are amplified with each
oscillation. The shock becomes significantly
distorted after only a few periods. In
core-collapse supernovae, SASI will operate in
conjunction (competition?) with neutrino-driven
convection.
12
(No Transcript)
13
Must move to 3D!
This initial SASI discovery with axisymmetric 2D
simulations pointed to the obvious need for
models in full 3D.
To better understand the challenges of 3D, let us
first look at the process of discovery for the
initial 2D models.
14
Hurdles for Large-Scale 3D
Not a problem
Simulation code Floating points Data output Data
transport Visualization and analysis
Thank you DOE
It works
Does not work
I cant see!
15
First Results SASI Exists in 3D

• 3D Cartesian grid
• 100 Million zones
• 100s of processors
• 100s of GB in full run

With data stuck on the West Coast, this was
science in the dark!
16
Science Begins with Data
Scientific discovery is done with interactive
access to data.

• Must have interactive access on a large-memory
  computer for analysis and visualization.
• Must have high bandwidth in accessing the data.
• Must have sufficient storage to hold data for
  weeks/months.

17
Interactive Visualization of TB Datasets
A commodity linux cluster provides all the must
haves. Data is sliced into slabs and stored on
local disks on the cluster nodes. EnSight Gold
provides an easy visualization solution,
including remote client-server operation and
collaboration.
18
(No Transcript)
19
We have jumped the hurdle, but there is much more
to be gained. Current research in scientific
visualization is providing glimpses of very
powerful new techniques. The next step is to get
these tools into the hands of application
scientists so they can explore their data.
20
LoRS tools / IBP depots
21
Data Flow Continues to Evolve
Analysis Cluster
Supercomputer
(flops)
(interactive)
Parallel analysis and vis on distributed data
Run Simulation On 100s to 1000s of cpus
22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
Forecast looks challenging

• Current full physics models in 2 spatial
  dimensions (2562 ) produce 70 GB per run.
• Current limited physics models in 3 spatial
  dimensions (10003) produce 4 TB per run.
• We know this problem must be attacked in 3D with
  accurate nuclear physics and neutrino transport.
  With advances in code development and computing
  platforms, we are looking at PB datasets in the
  near future!