Core-collapse supernovae

On the 3D supernova simulations

Andrzej Odrzywo lek

Dept. of General Relativity & Astrophysics Jagiellonian University Cracov

Wed, 8.12.2010, 10:15

Core-collapse supernovae

stars have masses in the range of 0.08 . . . ∼100 M

massive star is by definition star that will explode in core-collapse event

limiting lower range is 8 ± 1 M (Smartt 2009); beyond that mass core-collapse of the iron core is inevitable

main idea (Zwicky, 1939) is simple: collapse leads to neutron star formation, gravitational binding energy of 100 B (10

erg ) is stored in form of degenerate neutrino Fermi sea; somehow ∼1% of this energy is transferred to stellar envelope leading to supernova event

from astronomical point of view event is visible as type Ib/c, IIb, II-P,

II-L, IIn supernova/hypernova or long (Gamma Ray Burst)

Why supernova simulations are so important?

1

every single atom (except H & He) of the Earth, including our bodies has been produced and expelled in supernova explosion

2

supernovae are main agents of Galactic evolution

3

supernovae are exclusive producers of the neutron stars and stellar mass range black holes

4

all of the modern physics (neutrinos, relativity, QCD, nuclear physics, atomic physics, reactive hydro etc.) is important

5

simulations are on frontier of the hardware/software/algorithmic capabilities

6

nearby supernovae are main targets for neutrino and gravitational wave astronomy

7

far supernovae and GRB are cosmological distance/evolution

indicators

Simulation of the stellar lifecycle

1

Stellar birth [3D]

2

Stellar evolution (few Myrs, final Si burning stage few days, Odrzywolek&Heger 2010) [1D]

3

Gravitational collapse (GR1D) (-100 . . . 100 ms)

4

Neutrino-driven stage (0 . . . 2 s) [2D,3D] [PNS evolution, 2D]

5

hydrodynamic expansion stage t < 5 h [2D]

6

nebular (remnant) stage

At least 5 completely different codes required!

Stellar evolution - Kippenhahn diags

Core-collapse

collapse is ,,canonically spherically symmetric”

role of the rotation is still unclear, typically neglected

collapsed model: protoneutron star in the center + shock at some radius

collapsed model is an initial value data for subsequent 3D simulation

GR1D (stellarcollapse.org, http://arxiv.org/abs/1011.0005)

Why 3D ?

Physical space-time is 3+1 dimensional!

1D 2D 3D

Newtonian gravity trivial full monopole^∗

GR radial radial (TOV) radial

Turbulence/convection NO WRONG YES

Symmetry spherical cylindrical/equatorial none

SASI NO YES YES

NS kick NO z-axis only any

neutrino irradiation min. enhanced max.

rotation NO pure any

magnetic field NO NO YES

time 1 hour/CPU 1 day /10 C PU 1 month/100 CPU

CPU hours 1 500 10⁵

∗Procedure for 3D potential ready (P.Mach&AO, 2010). 3D simulations queued.

Initial and boundary conditions

Initial model

mapping 1D → 3D (or 2D) into spherical coordinates

PNS excised from the grid and replaced by boundary conditions (L

, point mass) at ,,surface” because of the CFL [= cell size/(c

+ v ) ] condition

L

(t) is a free parameter of the model (PNS evolution, SN1987A neutrinos, energy budget)

input parameters: (i) total energy radiated as neutrinos (ii)

time-dependence of the neutrino luminosity

Example of the 3D grid

Example of the 3D grid

Neutrinos

Light bulb model

neutrino radiation field is assumed to be radial with energy spectrum prescribed analytically

neutrinos are absorbed leading to energy-momentum exchange with matter

energy deposition by neutrinos might be as high as few tens MeV/baryon (thermonuclear energy is few MeV/baryon)

ν

, ¯ ν

flux changes proton-to-neutron ratio (neutron excess, Y

)

Our hardware

Deszno

3D simulations become possible thanks to successful proposal of prof.

E. Malec; technical specs by P. Mach.

Deszno (supercomputer) = 6 × complex

complex (SMP machine, single operating system, 96 physical CPU) complex = 4 × node (24 CPU mainboards)

node = 4 × 6-core E7450 2.4 GHz 12MB L3 cache complex = 256 GB RAM (2.66 GB/core)

Comparison:

Deszno complex: 96 × 2.4 GHz Xeon, 256 GB RAM

Software: physics

conservative multi-fluid hydro with shocks: PPM + HLLE + CMA spherical coords: 450x56x120 ' 3 × 10

cells

nuclear reactions: 18 species NSE + Bader-Deulhard ODE neutrinos : light bulb at the coordinate center

protoneutron star ,,pinned” at the center of the grid

accretion included; co-moving grid due to momentum conservation gravitation: TOV + poison3d (in progress)

time evolution: Strang splitting (XYZ–ZYX), explicit, 2nd order

Software: technicals

FORTRAN code; at least several years of development, mainly at MPA Garching

code is vectorized (SSE, Intel AVX) and OpenMP-parallelized (NUMA-aware)

software lifetime  hardware lifetime

CPU/Memory affinity required: (KMP AFFINITY, numactl, /proc/cpuset, ???) → HW/SW tunning in progress:

R. Marcinek&Deszno users

Simulation: timeline

15 Sep 1 Oct 15 Oct 1 Nov 15 Nov 1 Dec Now

0.0 0.5 1.0 1.5 2.0

t@secD

Analysis of the 3D results?

supercomputer: device converting compute-bound problem into I/O problem

3D supernova simulation on Deszno yields 1.8 TB (terabytes) of data no existing hardware at IFUJ is capable of data processing and visualization, except Deszno itself !

LLNL VisIt (https://wci.llnl.gov/codes/visit/)

interactivity, realtime 3D graphics, animation and stereo rendering is required to understand results obtained !

Comment: Any output from 3+1 simulation is a set of genuine non-trivial

functions of four variables. It is really hard to understand what is going

on. We need novel methods of analysis, or at least well-posed questions to

Summary

two successful 3D models were computed; one is already transferred as initial data for further processing

much more time-consuming higher resolution runs are in progress poison3D models are queued

new initial model (core-collapse) calculations are in progress

recently SN1979C (type II) supernova remnant observations revealed 8 M black hole produced from exploded 20 M star: this is a challenge for 3D simulations

We have just started! Prepare form more.