20
Simple Radiative Simple Radiative Transfer in Decomposed Transfer in Decomposed Domains Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

Embed Size (px)

DESCRIPTION

3 Radiative Transfer in Decomposed Domains RT important for optically thin media Diffusion approximation(s) deficient RT is a highly non-local problem Difficult to reconcile with domain decomposition

Citation preview

Page 1: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

Simple Radiative Transfer in Simple Radiative Transfer in Decomposed DomainsDecomposed Domains

Tobi HeinemannÅke Nordlund

Axel Brandenburg

Wolfgang Dobler

Page 2: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

2

The Pencil CodeThe Pencil Code

• High order finite difference code for MHD– 6th order in space, 3rd order in time– Memory and cache efficient

• Typical applications– MHD turbulence– Convection– Accretion discs

• Massive parallelization with MPI (Message Passing Interface)

Page 3: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

3

Radiative Transfer in Radiative Transfer in Decomposed DomainsDecomposed Domains

• RT important for optically thin media• Diffusion approximation(s) deficient• RT is a highly non-local problem• Difficult to reconcile with domain

decomposition

Page 4: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

4

The Transfer Equation & The Transfer Equation & ParallelizationParallelization

Analytic Solution:Processors

Page 5: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

The Transfer Equation & The Transfer Equation & ParParaallelizationllelization

Analytic Solution:

Ray direction

Intrinsic Calculation

Processors

Page 6: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

The Transfer Equation & The Transfer Equation & ParParaallelizationllelization

Analytic Solution:

Ray direction

Communication

Processors

Page 7: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

The Transfer Equation & The Transfer Equation & ParParaallelizationllelization

Analytic Solution:

Ray direction

Communication

Processors

Page 8: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

The Transfer Equation & The Transfer Equation & ParParaallelizationllelization

Analytic Solution:

Ray direction

Communication

Processors

Page 9: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

The Transfer Equation & The Transfer Equation & ParParaallelizationllelization

Analytic Solution:

Ray direction

Communication

Processors

Page 10: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

The Transfer Equation & The Transfer Equation & ParParaallelizationllelization

Analytic Solution:

Ray direction

Communication

Processors

Page 11: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

The Transfer Equation & The Transfer Equation & ParParaallelizationllelization

Analytic Solution:

Ray direction

Communication

Processors

Page 12: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

The Transfer Equation & The Transfer Equation & ParallelizationParallelization

Analytic Solution:

Ray direction

Communication

Processors

Page 13: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

The Transfer Equation & The Transfer Equation & ParallelizationParallelization

Analytic Solution:

Ray direction

Communication

Processors

Page 14: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

The Transfer Equation & The Transfer Equation & ParallelizationParallelization

Analytic Solution:

Ray direction

Processors

Intrinsic Calculation

Page 15: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

15

Details about the Details about the implementationimplementation

• Plasma composed of H and He• Only hydrogen ionization• Only H- opacity, calculated analytically No need for look-up tables• Ray directions determined by grid geometry No interpolation is needed

Page 16: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

16

Preliminary ResultsPreliminary Results• 2D model of surface convection

– Started from uniform initial state

Page 17: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

17

Preliminary ResultsPreliminary Results

• 3D model of sunspot– Started from Nordlund-Stein snapshot– Uniform initial magnetic field added

Page 18: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

18

Preliminary ResultsPreliminary Results

• 3D model of sunspot

Bottom Surface

Page 19: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

19

Timing resultsTiming results

• With 6 rays, and with ionization: 42.7 s/pt/st• With 2 rays, and with ionization: 37.6 s/pt/st• No radiation, but with ionization: 19.6 s/pt/st• No radiation, and no ionization: 8.7 s/pt/st• Ionization 2.3 times slower!• Radiation either 1.9 or 2.2 times slower.

Page 20: Simple Radiative Transfer in Decomposed Domains Tobi Heinemann Åke Nordlund Axel Brandenburg Wolfgang Dobler

20

ConclusionsConclusions

The method• is conceptually simple• is robust (analytic expressions, not limited

by table bounds)• has the potential to scale well in parallel

environments