pseudo-spectral methods for hpc of mhd turbulence · 2016-11-14 · why pseudo-spectral? 1)...
TRANSCRIPT
Pseudo-spectral methods for HPC of MHD turbulence
Luca BiferaleDept. of Physics and INFN University of Rome “Tor Vergata”
HPC Methods for Computational Fluid Dynamics and Astrophysics @Cineca
h"ps://www.fisica.uniroma2.it/~biferale/HPC-LEAP.html
WHYPSEUDO-SPECTRAL?1) MULTI-SCALESACCURACY+EXPONENTIALACCURACYFORDERIVATIVES2) POTENTIALTOOLTOPERFORMEXPERIMENTSINSILICO:EXACTCONTROLOFTHEDEGREES-OF-FREEDOM
• Toomanyturbulences?• Canwedisentangleuniversalfromnon-universalproperXes?• CanweunderstanduniversalproperXes?• Does‘compuXng’mean‘understanding’?(Computoergosum?)
smallparXcles:drag,addedmass,li]force,etc...
TurbulenceorTurbulences?
Flows with additives:Advection-diffusion-reaction of passive scalar/vectors (temperature, magnetic field, chemical reactions, etc...)Advection-diffusion of active scalars/vectors (convection, magnetic dinamo)Polymers (drag reduction) Bubbles/Droplets (two phase flows, rain formation, etc...)Swimmers (cooperative hydrodynamical interactions)
tracer bubble heavy
+boundarycondiXons
temperaturemagneXcfield
+periodicboundarycondiXons
3DCASE:MAINLYUNSOLVED!
HOMOGENEOUS&ISOTROPICMHDTURBULENCE
-homogeneous-isotropic-Gaussian-white-noiseinXme-large-scale
L.B.,G.Boffe"a,A.Celani,B.Devenish,A.Lano"eandF.ToschiPhys.Rev.Le+.93,064502,2004
K41 prediction
Multifractal prediction
Multifractal prediction
mean-field (k41) prediction
ACCELERATION
Reynolds number ~ (Non-Linear)/(Linear terms)
Fully Developed Turbulence:1. Strongly non-linear & non-perturbative system
COMPLEXPHYSICSWITHSIMPLEFLOWS
2. Out of Equilibrium (non perturbative)Dissipative anomaly 3. Small-scales PDFs strongly non-Gaussian
Anomalous scaling
COMPLEXPHYSICSWITHSIMPLEFLOWS
1. inerXalrangeofscales:powerlaw(anomalous)2. extensionincreaseswithReynolds!
4. Many-body problem:
THE ENERGY CASCADE:
COMPLEXPHYSICSWITHSIMPLEFLOWS
Number crunching approach: computo ergo sum.
state-of-the-art DNS (Kaneda’s group):
Isotropic, homogeneous Fully Periodic FlowsPseudo-Spectral Methods.
Resolution 12000^3
Reynolds : 10^7,Storage of 1 velocity configuration: 30 Tbyte
Moral: easy to saturate any computing power(present and/or future)
astrophys. flow atmosph. flow laboratory flow
VISCOUSTERMEXACTLYINTEGRATEDADAMS-BASHFORTH2°ORDER
WHATABOUTMHD????
HTTP://WWW.P3DFFT.NET
WHYPSEUDO-SPECTRAL?1) MULTI-SCALESACCURACY+EXPONENTIALACCURACYFORDERIVATIVES2) POTENTIALTOOLTOPERFORMEXPERIMENTSINSILICO:EXACTCONTROLOFTHEDEGREES-OF-FREEDOM
• Toomanyturbulences?• Canwedisentangleuniversalfromnon-universalproperXes?• CanweunderstanduniversalproperXes?• Does‘compuXng’mean‘understanding’?(Computoergosum?)
LORENTZ
ADVECTION+STRETCHING
LORENTZ
STRETCHING+ADVECTION
+ boundarycondiXons
KinemaXcs+DissipaXonareinvariantunderRotaXon+TranslaXonNon-universalstaXsXcalbehaviour<->Anisotropy
Smallscalesvslargescales
Turbulentjet 3dConvecXveCell ShearFlow
Thesimplestsetof0-ranktensor(SCALAR)observable:
LongitudinalStructureFuncXons
Arad, V. L’Vov I. Procaccia PRE 59, 6753 (1999). Arad et al. PRL 82, 5040 (1999).
Arad et al. PRL 81, 5330 (1998).
3d rotation
Setof3n*(2j+1)EigenfuncXonsofgroupofrotaXonsin3d:
DecomposiXonintermsof(irreducible)invariantsubset-labelledbyq,j=0,1,2,…
n-ranktensorwhichdepends
ona3dvector
rotaXonalinvariantoperator
FOLIATION!!!
+ so(3) ->
J=0
J=2
J=4 Largescalephysics:
allsectorscoupledby
forcingterms
scaling?
WorkingHypothesis
projecXononeachsectorhasauniversalscalingexponent,dependingonthatsectoronly.
Dependencyonlargescalephysicsshowsuponlyinprefactors
Purepowerlawsonlyineachseparetedsector:
MatchingInfra-RedboundarycondiXons:
prefactor cannot be universal
Aboutuniversalityofscalingexponentsnothingcanbesaidrigorously,atleastfortheNSeqs.
RecoveryofIsotropySmall-ScalesUniversality
WeperformedaDNSofaRandom-KolmogorovFlow
PeriodicboundarycondiXons
256x256x256
Hyperviscosity
HomogeneousbutAnisotropic
L.B. and F. Toschi, PRL 86, 4831 (2001)
L.B. I. Daumont, A. Lanotte and F. Toschi. PRE. 66, 056306 (2002)
isotropicsector
beforeso(3)decomposiXon
Localslopes
ComparisonofscalingproperXes:isotropicsector(j=0,m=0)vsundecomposedstructurefuncXon
xyz
Arad et al. PRL 82, 5040 (1999)
scalingism-independent
dimensional
j=6
j=4
j=0
Recoveryofisotropy
L.B. I. Daumont, A.S. Lanotte and F. Toschi PRE. 66, 056306 (2002)
ExtendingLumley’sanisotropictheory
PARTIALREFERENCELISTOFINTERESTFORTHISMINI-COURSE
MulLfractalsEulerian&LagrangianTurbulence:A.Arneodo,R.Benzi,J.Berg,L.Biferaleetal.Phys.Rev.Le".100,254504(2008)R.Benzi,L.Biferale,G.Paladinetal.Phys.Rev.Le".67,2299(1991)L.Biferale,G.Boffe"a,A.Celanietal.Phys.Rev.Le".93,064502(2004)L.Biferale,G.Boffe"a,A.Celanietal.Phys.Fluids.17,021701(2005)L.Biferale,E.Bodenschatz,M.Cencinietal.PhysFluids20,065103(2008)Eulerian&LagrangianTurbulence:U.Frisch.‘Turbulence’(CambridgeUniversityPress,1996)R.Benzi,L.Biferale,R.Fisheretal.Journ.FluidMech.653,221(2010)L.Biferale,E.Bodenschatz,M.Cencinietal.PhysFluids20,065103(2008)G.Falkovich,K.Gawȩdzki,M.Vergassola.Rev.Mod.Phys.73,913(2001)T.Gotoh,D.FukayamaandT.Nakano.Phys.Fluids14,1065(2002)T.Ishihara,T.Gotoh,andY.Kaneda.Ann.Rev.FluidMech.41,165(2009)V.L’vovandI.Procaccia.Phys.Rev.E52,384(1995)V.L'vovandI.Procaccia.Phys.Rev.Le".76,2896(1996)J.P.L.C.SalazarandL.R.Collins.Ann.Rev.FluidMech.41,405(2009)B.SawfordandP.K.Yeung.Phys.Fluids23,091704(2011)K.R.Sreenivasan.Ann.Rev.FluidMech.25,435(1997)F.ToschiandE.Bodenschatz.Ann.Rev.FluidMech.41,375(2009)P.K.Yeung.Ann.Rev.FluidMech.34,115(2002)T.Baebler,L.Biferale.andA.S.Lano"ePhysRevERap.Comm85025301R2(012)InerLalParLcles:J.Bec,L.Biferaleetal.J.FluidMech.550,349(2006);J.BecetalPhys.Rev.Le".98,084502(2007)L.Biferale,J.Bec,M.Cencinietal.Phys.Fluids18,081702(2006)Anisotropies:L.BiferaleandI.Procaccia.Phys.Rep.414,43(2005)L.BiferaleandF.Toschi.Phys.Rev.Le".86,4831(2001)L.Biferale,E.Calzavarini,A.S.Lano"eetal.PhysicaA338,194(2004)K.Chang,G.P.BewleyandE.Bodenschatz.physics.flu-dynarXiv:1102.1197(2011)B.Jacob,L.Biferale,G.Iugoetal.Phys.Fluids16,4135(2004)S.Kurien,K.R.Sreenivasan.Phys.Rev.E62,2206(2000)X.Shen,Z.Warha].Phys.Fluids14,2432(2002)