numerical simulations of the mri: the effects of dissipation coefficients
DESCRIPTION
Numerical simulations of the MRI: the effects of dissipation coefficients. S.Fromang CEA Saclay, France J.Papaloizou ( DAMTP, Cambridge, UK) G.Lesur ( DAMTP, Cambridge, UK), T.Heinemann (DAMTP, Cambridge, UK). Background: ESO press release 36/06. Setup. The shearing box (1/2). z. x. - PowerPoint PPT PresentationTRANSCRIPT
Numerical simulations of Numerical simulations of the MRI: the effects of the MRI: the effects of dissipation coefficientsdissipation coefficients
S.Fromang CEA Saclay, France
J.Papaloizou (DAMTP, Cambridge, UK)G.Lesur (DAMTP, Cambridge, UK),
T.Heinemann (DAMTP, Cambridge, UK)
Background: ESO press release 36/06
Setup
The shearing box (1/2)
H
H H
x
yz
• Local approximation• Code ZEUS (Hawley & Stone 1995)• Ideal or non-ideal MHD equations• Isothermal equation of state• vy=-1.5x
• Shearing box boundary conditions• (Lx,Ly,Lz)=(H,H,H)
Magnetic field configurationZero net flux: Bz=B0 sin(2x/H) Net flux: Bz=B0
x
z
The shearing box (2/2)
Transport diagnostics
• Maxwell stress: TMax=<-BrB>/P0
• Reynolds stress: TRey=<vrv>/ P0
• =TMax+TRey
Small scale dissipation
• Reynolds number: Re =csH/• Magnetic Reynolds number: ReM=csH/• Magnetic Prandtl number: Pm=/
The issue of convergence
(Nx,Ny,Nz)=(128,200,128)Total stress: =2.0 10-3
(Nx,Ny,Nz)=(256,400,256)Total stress: =1.0 10-3
(Nx,Ny,Nz)=(64,100,64)Total stress: =4.2 10-3
Fromang & Papaloizou (2007)
The decrease of with resolution is not a property of the MRI. It is a numerical artifact!
Code ZEUSZero net flux
Numerical dissipation
Numerical resisitivity(Nx,Ny,Nz)=(128,200,128)
No explicit dissipation included
BUT: numerical dissipation depends on the flow itself in ZEUS…
Residual-k2B(k)2
Fourier Transform and dot product with the FT magnetic field:
=0 (steady state) Balanced by numerical dissipation (k2B(k)2)
ReM~30000 (~ Re)
Pm=/=4, Re=3125(Nx,Ny,Nz)=(128,200,128)
Maxwell stress: 7.4 10-3
Reynolds stress: 1.6 10-4
Total stress: =9.1 10-3
Residual-k2B(k)2
balanced by numerical dissipation
Explicit dissipation
Statistical issues at large scale
Varying the resolution(Nx,Ny,Nz)=(128,200,128)
Maxwell stress: 7.4 10-3
Reynolds stress: 1.6 10-3
Total stress: =9.1 10-3
(Nx,Ny,Nz)=(256,400,256)
Maxwell stress: 9.4 10-3
Reynolds stress: 2.1 10-3
Total stress: =1.1 10-2
(Nx,Ny,Nz)=(64,100,64)
Maxwell stress: 6.4 10-3
Reynolds stress: 1.6 10-3
Total stress: = 8.0 10-3
Good agreement
but…
Residual-k2B(k)2
Numerical & explicit dissipation comparable!
Code comparison: Pm=/=4, Re=3125
ZEUS : =9.6 10-3 (resolution 128 cells/scaleheight) NIRVANA : =9.5 10-3 (resolution 128 cells/scaleheight)SPECTRAL CODE: =1.0 10-2 (resolution 64 cells/scaleheight)PENCIL CODE : =1.0 10-2 (resolution 128 cells/scaleheight)
Good agreement between different numerical methods
NIRVANASPECTRAL CODE
PENCIL CODEZEUS
Fromang et al. (2007)
Code comparison: Pm=/=4, Re=3125
ZEUS : =9.6 10-3 (resolution 128 cells/scaleheight) NIRVANA : =9.5 10-3 (resolution 128 cells/scaleheight)SPECTRAL CODE: =1.0 10-2 (resolution 64 cells/scaleheight)PENCIL CODE : =1.0 10-2 (resolution 128 cells/scaleheight)
Good agreement between different numerical methods
NIRVANASPECTRAL CODE
PENCIL CODEZEUS
Fromang et al. (2007)
RAMSES
=1.4 10-2
(resolution 128 cells/scaleheight)
Zero net flux: parameter survey
Flow structure: Pm=/=4, Re=6250(Nx,Ny,Nz)=(256,400,256)
Density Vertical velocity By component
QuickTime™ et undécompresseur codec YUV420
sont requis pour visionner cette image.
Movie: B field lines and density field (software SDvision, D.Polmarede, CEA)
Schekochihin et al. (2007)Large Pm case
Velocity Magnetic field
Effect of the Prandtl number
Take Rem=12500 and vary the Prandtl number….
(Lx,Ly,Lz)=(H,H,H)(Nx,Ny,Nz)=(128,200,128)
increases with the Prandtl number No MHD turbulence for Pm<2
Pm=/=4Pm=/= 8Pm=/= 16
Pm=/= 2
Pm=/= 1
Pm=/=4
(Nx,Ny,Nz)=(128,200,128)
Re=3125
Total stress=9.2 ± 2.8 10-3
Total stress=7.6 ± 1.7 10-3
(Nx,Ny,Nz)=(256,400,256)
Re=6250
By in the (x,z) plane
Pm=4, Re=12500
Total stress=2.0 ± 0.6 10-2
(Nx,Ny,Nz)=(512,800,512)
BULL cluster at the CEA ~500 000 CPU hours (~60 years)
1024 CPUs (out of ~7000)2106 timesteps600 GB of data
No systematic trend as Re increases…
Power spectra
Re=3125 Re=6250
Re=12500
Kinetic energy
Magnetic energy
Summary: zero mean field case
• Transport increases with Pm• No transport when Pm≤1• Behavior at large Re, ReM?
Fromang et al. (2007)
Transition
Pm=3
Pm=4
Pm=2.5
~4.510-3
(Lx,Ly,Lz)=(H,H,H)(Nx,Ny,Nz)=(128,200,128)
Re=3125
Vertical net flux
The mean field caseLesur & Longaretti (2007)
- Pseudo-spectral code, resolution: (64,128,64)- (Lx,Ly,Lz)=(H,4H,H)- =100
Pm
1
min
max
Critical Pm?Sensitivity on Re, ?
Flow structure Pm=/>>1
Viscous length >> Resistive length
Schekochihin et al. (2007)
Velocity Magnetic field
Pm =/ <<1
Viscous length << Resistive length
Schekochihin et al. (2007)
Velocity Magnetic field
vz Bz vz BzRe=800 Re=3200
Relation to the MRI modes
Growth rates of the largest MRI mode
No obvious relation between and the MRI linear growth rates
Conclusions & open questions• Include explicit dissipation in local simulations of the MRI:
resistivity AND viscosity Zero net flux AND nonzero net flux an increasing function of Pm Behavior at large Re is unclear
?
MHD turbulence
No turbulenceRe
Pm
• Vertical stratification? Compressibility (see poster by T.Heinemann)?• Global simulations? What is the effect of large scales?• Is brute force the way of the future? Numerical scheme?• Large Eddy simulations?
Pm
1
min
max
Critical Pm?Sensitivity on Re, ?