ASPECTSOFTURBULENCEINASTROPHYSICS

R.Walder
(1,2)
,D.Folini
(2)
,J.Favre
(3)
(1)
ObservatioredeStrasbourg,ruedel'universit11,67000Strasbourg,France
(2)
InstituteofAstronomy,ETHZentrum,8092Zurich,Switzerland
(3)
SwissNationalSupercomputingCenterCSCS,viaCantonale,6928Manno,Switzerland
walder@astro.phys.ethz.ch,folini@astro.phys.ethz.ch,jfavre@cscs.ch
Abstract
Turbulenceinastrophysicsisakeyingredientforthe
dynamical understanding of many objects. Birth, life,
anddeathofstars,asonlyoneexample,cruciallydepend
onit.Wegiveashortoverviewofthedifferentturbulent
regimes encountered in different astrophysical objects.
Wedescribesomeaspectsofsupersonicturbulence,i.e.
turbulencewithrootmeansquareMachnumberclearly
above1anddissipationismostlyduetoshocks.Finally,
wediscusssomeissuesrelatedtonumericalrealization.
1 Introduction
Turbulence is ubiquitous in astrophysics. It has its
peculiarities, however, bringing new, interesting facets
into the field of research. The first peculiarity is its
dynamical interaction with other physical processes
under nearly all circumstances. Magnetic and radiative
fields,gravity,nuclear,andchemicalreactionssubstant
ially contribute to the momentum and energy balance.
The second peculiarity is the eminent dynamical role
turbulenceplaysfortheevolutionofmanyastrophysical
objects. A third peculiarity results from the fact that
theoreticalornumericalresultsarehardtoconfrontwith
laboratorymeasurementsduetotheextreme,nonterrest
rial conditions under which the physical and chemical
processes take place. Indeed, the essentially single
sourceofinformationastrophysicistshaveonturbulence
in outer space is the light from the object, either as a
pictureormerelyasaspectrum.Abundantassuchdata
are, they contain only indirect information (like the
widthofspectrallines)integratedalongtheentirelineof
sight.Consequently,theanalysisofthisdatawithregard
toturbulenceisaresearchfieldinitsownright(1,2).
2 Turbulenceinastrophysics:anoverview
Aniceoverviewofdifferentaspectsofturbulencein
astrophysicsprovidetheconferenceproceedings(3,4,5).
2.1Stellarstructureandevolution
Tounderstandthechemicalevolutionoftheuniverse
stellar evolution must be understood, as essentially all
chemical elements beyond Beryllium are produced by
nuclear reactions in stars (6). Essential for the under
standing of stellar evolution is the transport of energy
and angular momentum within stars. Energy produced
bynuclearfusionistransportedoutwardsinessentially
twoways:convectionandradiation.Angularmomentum
is also transported by turbulent convection, losses (by
stellar winds) or gains (by accretion from a binary
companion)atthestellarsurfacethusaffecttherotation
profile throughout the star. Magnetic fields can
substantially enhance the efficiency of the transport.
Turbulencealsoplaysacrucialrolefortheproductionof
magneticfieldsinastellardynamo(5).Turbulencealso
transports newly produced chemical elements from the
central region of the star to its outer layers (7). The
associatedchangeinchemicalcompositionoftheouter
layers can affect the mass loss and thus the further
evolution of the star. The mutual interactions of the
differentprocesses,whichhappenondifferentscalesand
in different locations within the star, have not yet
emerged into a unique, complete picture. A lot of
progress has, however, been achieved in the last few
years. The far reaching consequences sketched at the
beginningoftheparagraphmakethepropertreatmentof
turbulence, turbulentmixing, and the interactionofthe
turbulence with the radiation and/or magnetic field of
thestarakeytopicinastrophysics.
2.2Starformation
Starformationtoday takesplace only in molecular
clouds: cold (10K), high density (10
2
10
4
part./cm
3
)
regionsofinterstellarspace(220pc,1pc=3*10
18
cm)
thatare molecularinnatureandcontain10
2
10
4
solar
masses.Fromobservationsitisfurtherknownthatthese
cloudsareturbulent.Compressibleturbulenceisthought
to regulate the efficiency at which stars form in such
clouds. Turbulent pressure prevents global collapse of
the cloud, which would result in a much higher star
formation efficiency than what is observed. Locally,
turbulence leads to density enhancements from which
individual stars or small groups of stars can form by
cooling and subsequent gravitational collapse. Whether
the mass distribution of the newly formed stars (the
initial mass function or IMF) is directly set by the
turbulenceisnotclear.Thequestionisof fundamental
importanceastheinitialmassofastariscrucialforits
evolution (e.g. life time and synthesis of chemical
elements).Anotherimportantquestionisthedrivingof
theturbulence,theenergysourceandinjectionscale.For
reviewsandfurtherreferenceswereferto(810).
2.3Accretion,winds,andjets
Accretion of mass and angular momentum onto a
star or a black hole is one of the crucial processes in
astrophysics. Turbulence plays a decisive role in this
process. Conservation of angular momentum hinders
directinfallofmatterontoaccretingobjects.Accretion
disks form in which the fluid parcels have nearly
Keplerian orbits. The orbital velocity is highly super
sonic(uptoMach100).Massisbroughtinwardsbythe
transportofangularmomentumoutwards.Onemechan
ism for this is viscous dissipation provided by turbul
ence. Commonly it is believed that magnetic shear
provides the mechanism that excites and drives the
turbulence (1113). Strong magnetic fields can be
generatedinaneffectivedynamo(5)atthesametime.
Spiral shocks and disk winds provide an additional
sourceofangularmomentumtransport.
Outflowsormassejectionsarenotonlyessentialfor
stellarevolutionbutalsoforthechemicalenrichementof
the universe. Moreover, they are principally linked to
highenergyeventssuchassupernova, rayburstsand
quasars.Often,theyaredrivenbyexternalforces:mom
entum transfer from photons by scattering in spectral
linesorondustgrains.Anotherimportantmechanismis
centrifugalaccelerationalongmagneticfieldlineswhich
fixes the matter to rigid rotation (e.g. winds and jets
from accretion disks). Turbulence in outflows pres
umably originates from unstable driving (14). Most
likelythisprocessleadstoahighlyfragmentedmedium,
highdensityknots,somekindofsupersonicturbulence
describedinsection3.
2.4Supernovae
We distinguish between two different classes of
supernovae which both provide 10
51
ergs explosion
energyand10
49
ergsinlightemission.Inbothclasses,
turbulenceseemstoplayaprincipalrole.
Thefirstclassisathermonuclearrunaway(typeIain
the phenomenological classification), which burns
withinlessthanasecondabout1.4solarmassesofox
ygenandcarbontoiron.Theturbulentreactiveburning
front probably develops from a combustion to a det
onationregime(seee.g.RpkeandHillebrandtin(4)).
The second class results from the gravitational
collapseoftheironcoreofamassivestarattheveryend
of its evolution. Most of the released gravitational
energyisstoredinneutrini.Onlyabout1permilleofthe
energyisinitiallyintheflow.Withinlessthanasecond
aftercollapseandsubsequentstabilizationofthecorein
aneutronstar,theneutrinistarttodumpenergyintothe
gas,leadingtostrongconvection.Thestillfallingrestof
thestargetsshocked.Aspointedoutby(15)and(16),
this can trigger a rapidly growing vorticalacoustic
cycle: vorticity is generated at the kinked accretion
shock, transported into the core, where it is scattered.
Thescatteringproducesstrongsoundwaveswhichkink
theshockevenmorewheninteractingwithit.Moreover,
thesoundwavesstronglyinfluencetheenergydumping
intothegassincetheheating/coolingbytheneutriniis
proportionaltoT
6
(T:Temperature).Alltheseprocesses
combine to a so far very poorly understood gravo
neutrinomagnetohydrodynamical turbulence which ev
entuallytriggerstheexplosionoftheentirestar.
3AspectsofsupersonicTurbulence
Weusetheterm'supersonicturbulence'forturbulence
with root mean square Mach number larger than one,
M
rms
>1.Turbulenceinthisregimeclearlydeviatesfrom
the incompressible picture and is also much less
understood.Mostprogresshasbeenmadeforisothermal
conditions (=1) on which we focus in the following.
The results we are going to summarize are based on
numerical simulations for the most part. Isothermal
supersonicturbulenceisthoughttobeagoodfirstorder
model for the interstellar medium (ISM, Section 2.3)
and, partly, stellar winds and jets (Section 2.4). An
importantaspectofthedrivingofthisturbulenceisthe
collision of two flows (e.g. faster and slower parts of
radiatively driven stellar winds, windwind, supernova
ejectaISM). We discuss a very simple model of such
collisionzonesinSection3.2.
3.1Isothermalsupersonicturbulence
!solleimal sueisonic luiLulence is claiacleiised
Ly lle iesence ol slocls, sliong densily and velocily
conliasls, and a alcly aeaiance ol lle llow (liguie
1). Densily las a log-noimal disliiLulion (liguie 2;
leie =1 is ciucial, (17,18)), coiielalion will velocily
is essenlially zeio. Slocls accounl loi al leasl 7O ol
lle lolal eneigy loss (19,2O). !n diiven luiLulence, lle
sliongesl slocls dissiale mosl ol lle eneigy, wlile in
decaying luiLulence eneigy is dissialed Ly a laige
numLei ol weal slocls. !n lle lalei case, lle linelic
eneigy decays as l
-l
,will O.85 < l < 1.2 (21). Tle
alcly salial allein loimed Ly lle slocls scales
will lle eneigy injeclion scale (8). !oi lle velocily
sliucluie lunclions ol diiven isolleimal luiLulence, a
unilying desciilion liom suLsonic lo sueisonic las
Leen ioosed (22,23). Only one aiamelei, lle
Hausdoill dimension D ol lle mosl dissialive
sliucluie, las lo Le adaled as a lunclion ol Mims,
liom D=1 (lilamenls) in lle suLsonic case lo D=2
(sleels) in lle sueisonic case (Mims = 1O). \ndei
non-isolleimal condilions, values u lo D=2.3 lave
Leen ieoiled (24).
3.2Anexample:the2Dslab
Whereasmoststudiesofsupersonicturbulencesofar
arebasedoncomputationsinaperiodiccubicboxwith
artificial driving, we report now on a first systematic
investigationofsupersonicisothermalturbulencedriven
by colliding hypersonic flows (25). Such a scenario is
both realistic and relevant. Stellar winds and stellar
debris from exploded stars are among the most signif
icantsourcesofenergyfortheISM.Preliminaryresults
ofturbulenceincollidinghypersonicflowswhichincl
udestrong(radiative)coolingcanbefoundin(2629).
Figure1showstheturbulentinteractionzonesfrom
severalsimulations.Suchzonesareshowntobelinearly
unstable (30). In each simulation, the flows are anti
parallelbutotherwise identical.The solutiononly dep
endsononesingleparameter,theupstreamMachnumb
er M
u
, ranging from 5.5 to87. The interaction zone is
supersonically turbulent with characteristics as disc
ussedin thelastparagraph:1<M
rms
< 16, lognormal
densitydistribution(figure2),patchyappearance.
Assuming selfsimilarity in infinite space and no
correlationbetweendensityandvelocity,(25)performed
a dimensional analysis within the frame of isothermal
Euler equations. The analysis reveals the mean density
in the turbulent interaction zone to be independent of
M
u
while M
rms
scales linearly with M
u.
Numerically
includingviscousdissipationimplicitly(seesection4)
thisresultisconfirmedand<>~30
u
, M
rms
~0.2M
up.
Thisis insharpcontrasttothesituationin1D,whereit
canbeshownthat<>~
u
M
u
2.
andM
rms
=0.
Figure1:Supersonicallyturbulentinteractionzoneof
headoncollidingflows.Upperrow:densityforthecase
ofM
u
=20atanearlytime(left)andalatetime(right).
Lowerrow:velocitydivergenceforM
u
=10flows(left
andforM
u
=30(right).Thesizeofthepatchesandthe
wiggelingoftheboundingshocksarecloselylinkedand
dependonM
rms
andthewidthoftheslab.
The analysis also brings a deeper insight into the
natureoftheturbulence. Ondimensionalgrounds, the
dissipated energy per unit volume and time must be
proportional to
diss
v
diss
3
l
diss
1
. For the average column
integrated value (average width of the slab: l
cdl.
) this
results in
diss
~
diss
v
diss
3
l
diss
1
l
cdl.
. With no other spatial
scale and velocity scale present, it is then naturally to
assumel
diss
~l
slab,,
andv
diss
~v
rms,
leadingto
diss
~
diss
v
rms
3
,
or
diss
~
diss
M
u
3
.Acloseranalysisgivenin(18)reveals
averyweakMachnumberdependenceof
diss
.
Thenumericalresultscoincidewiththisanalysis(figures
1,3). Thedissipationscalel
diss
andtheautocorrelation
length of the shock wiggeling l
corr
nicely scale linearly
among each other and with the averaged width of the
slab,l
cdl
.AlllengthsscalewithM
u
0.6
(figure3). This
also confirms the eyecatching distribution of the
patches and the shockwiggeling which can be taken
formfigure1.Thepatchesandtheshockwigglesarebig
for large slabs and low Machnumbers and small for
small slabs and high Machnumbers. The dissipation
scale in this supersonically turbulent regime is clearly
relatedtothemeandistancebetweenshocksandnottoa
viscousdissipationlength.Thegoodcoincidenceofthe
dimensionalanalysisbasedontheEulerequationswith
the numerical results which include implicitly viscous
dissipationindeedprovesthe'Eulerlike'characterofthe
turbulenceinthisregime.
Figure2:Lognormaldensitydistributionofturbulent
flow field shown in Figure 1. The slight deviation
indicatesprobablyaslightdecayoftheturbulence.
Figure3:Thelengthscalecharacterizingthedissipation,
l
ekin
(l
diss
inthetext)andthespatialscaleofthewiggeling
oftheboundingshocks(l
corr
)scalelinearlywiththe
widthoftheslab(l
cdl
)andM
u
0.6
.Shownarethescalings
againstthemassaccumulatedintheslab(l(N))forall
runsofoursample.Inprintsoftheinitialconditionsdie
outatapproximatelyl(N)~15,thefinitesizeofthe
computationalboxstarttoaffecttheresultsat
approximatelyl(N)=70.
4 Aspectsofthenumericalrealization
In astrophysical hydrodynamical simulations, both
high resolution finite volume shock capturing schemes
(HRFV)suchasPPMandstatisticalmethods(socalled
'smoothedparticlehydrodynamics',SPH,seee.g.(8)for
areview)arewidelyused.Otherprocesseslikegravity,
radiation,andmagneticfieldsaremostlyincludedusing
fractionalstepmethods.Validatingselfgravitatingmag
netoradiationhydrodynamical simulations is difficult
andmostofthecodessofararenotvalidatedinascient
ificallyproperway.
In astrophysical simulations, Adaptive Mesh Refin
ement(AMR)hasproventobeveryuseful.AMRhelps,
forexample,toresolveturbulentburninglayersinstars,
which are although geometrically thin often the
sourceofaconvectiveoverturnoflargepartsofthestar.
A turbulent molecular cloud starts to collapse due to
gravityandcoolingatanarbitrarymomentoftimeatan
arbitrarylocationinspace,andonewouldliketofollow
thatcollapseover34ordersofmagnitudeinspaceina
computationally well resolved manner. However, in
connection with turbulence, AMR may also be a nuis
ance.
Compared to engineering, dynamical subgrid scale
models(SGS)arelesswidelyusedinastrophysics.The
reason for this lies mostly in the presence of shocks
and/orotherphysicalprocessesforwhichnoproperSGS
hasbeendevelopedsofar.Anotherreasonisthenotor
ious difficulty to combine SGS with AMR. Only
recently,someattemptsaremadetouseflamelets(e.g.
Ewald and Peter in (4)) in simulating reactive burning
fronts in supernovae and to combine dynamical SGS
models with AMR (see e.g. Niemeyer, Schmit and
Klingenbergin(4),(31)).
Computationsofsupersonicturbulencemostlyusethe
MILES approach (Monotone Integrated Large Eddy
Simulation). As pointed out in (32) and (33), HRFV
schemesactimplicitlyasSGSmodels.Inparticular,the
turbulentenergyispassedcorrectlyandwithoutenergy
jam from large scale to small scales, thus correctly
mimicking the turbulent cascade. A larger study (34)
confirms this for a large variety of schemes, such as
MUSCL, (W)ENO, and Jameson. It also shows,
however, the clear limitation of this approach. Many
small scale flow features are suppressed by the large
diffusivity of these schemes. The further use of an
explicitdynamicalSGSmodelevenmakesthesituation
more worse. Having no other tools at hand, all the
simulationsdiscussedinSection3neverthelessusedthis
approach. However, none of the only very few grid
studies have shown convergence. Figure 4 shows the
ratioofM
rms
ascomputedonafineandcoarsegridfor
set of simulations used in section 3.2. Although 1280
and2560cellsareused,thetwoM
rms
differbyabout15
percent.Remarkably,thefinerdiscretizationshowsmore
dissipation. In our view, this results from the better
resolutionofshocks,themainsourceofdissipation,ona
finergrid.
Figure4:DifferenceinM
rms
ascomputedonafineanda
coarsegridforoursetofruns.
Clearly,thesituationdemandsremedy. Theauthors
make currently an attempt to develop appropriate
methodswhichhopefullyhelptoovercomethesketched
problemsincomputationandvalidation.
5Conclusions
Observations and theoretical arguments strongly
support the view that turbulence is a keyprocess in
astrophysics. The properties and evolution of a wide
varietyofastrophysicalobjectscanonlybeunderstoodif
turbulence is taken properly into account. 'Properly' in
astrophysics signifies that the turbulence is mostly
compressible, even supersonic, and usually interwoven
with other physicalprocesses. Attemptstomodel these
facets of turbulence have started only recently, in the
wake of thesteadilyincreasingcomputer power. More
appropriate numerical tools still have to be developed.
They must be able to cope with both, the chaotic
dynamicsofshocksandthesubsonicturbulentcascade.
Moreover, they must properly take into account the
effectofexternalfields(radiation,magnetic,gravity)on
the momentumand energybalance. Astrophysics isan
excellentlabforstudyingmanyfacetsofturbulenceand
we predict a 'golden age' of turbulence research in
astrophysics. To go along this path, we strongly
encourage the interchange of models and results with
othercommunities.WewouldliketothankERCOFTAC
tomakethishappeningandpossible.
References
[1] Stutzki,J.,Bensch,F.,Heithausen,A.,Ossenkopf,
V., and Zielinsky, M. (1998), On the fractal
structureofmolecularclouds,A&A
*
336,697720.
[2] Ossenkopf, V. and Mac Low, M.M. (2002),
Turbulent velocity structure in molecular clouds,
A&A
*
390,307326.
[3] Falgarone,E.andPassot,T.(eds.),(2003),Turbul
ence and magnetic fields in astrophysics, Lecture
NotesinPhysics614,SpringerVerlagBerlin.
[4] Kupka,F.andHillebrandt,W.(eds.)(2005),Inter
disciplinary aspects of turbulence, conference
proceedings (/www.mpagarching.mpg.de/mpa/
publications/proceedings/proceedingsen.html).
[5] Proctor,M.R.E.etal.(1995),SolarandPlanetary
Dynamos,PublicationsoftheNewtonInstitute.
[6]Woosley,S.E.,Heger,A.,andWeaver,T.A.(2002),
The evolution and explosion of massive stars,
Reviewsofmodernphysics74,10151072.
[7] Meynet, G. and Maeder, A. (2005), Rotation and
mixing in massive stars: principles and
uncertainties,ASPConferenceSeries337,1527.
[8] Mac Low, M.M. and Klessen, R. S. (2004),
Controlofstarformationbysupersonicturbulence,
Reviewsofmodernphysics76,126194.
[9] Elmegren,B.G. andScalo,J.(2004),Interstellar
turbulenceI:observationsandprocesses,Ann.Rev.
ofAstronomyandAstrophysics42,211273.
[10] Scalo, J. and Elmegren, B. G. (2004), Interstellar
turbulence II: implications and effects, Ann. Rev.
ofAstronomyandAstrophysics42,274316.
[11] Shakura, N.I. and Sunyaev, R.A. (1973), Black
holesinbinarysystems.Observationalappearance,
A&A*24,337355.
[12]Balbus,S.A.andHawley,J.F.(1998),Instability,
turbulence, and enhanced transport in accretion
disks,Reviewsofmodernphysics70,153.
[13]Hawley,J.F,Balbus,S.A.,andStone,J.M.(2001),
A Magnetohydrodynamic Nonradiative Accretion
FlowinThreeDimensions,ApJ**,L49L52.
[14] Owocki, S.P., Castor, J.I., Rybicki, G.B. (1988),
Timedependent models of radiatively driven
stellarwinds.INonlinearevolutionofinstabilities
forapureabsorptionmodel,ApJ**335,914930.
[15]Blondin,J.M.,Mezzacappa,A.,andDeMarino,C.
(2003),Stabilityofstandingaccretionshocks,with
aneyetowardcorecollapsesupernovae,ApJ
**
584,
971980.
[16] Foglizzo, T., Scheck, L., Janka, H.T. (2005),
Neutrinodriven convection versus advection in
core collapse supernovae, submitted to A&A
*
,
astroph/0507636(http://fr.arxiv.org/).
[17]Passot,T.andVzquezSemadeni,E.(1998),Dens
ity probability distribution in 1dimensional poly
tropicgasdynamics,Phys.Rev.E58,450145001.
[18]Scalo,J.,VzquezSemadeni,E.,Chappell,D.,and
Passot, T. (1998), On the probability density
function of galactic gas. I. Numerical simulations
andthesignificanceofthepolytropicindex,ApJ
**
504,835853.
[19] Smith, M.D., Mac Low, M.M., and Zuev, J.M.
(2000), The shock waves in decaying supersonic
turbulence,A&A
*
356,287300.
[20] Smith, M.D., Mac Low, M.M., and Heitsch, F.
(2000), Thedistributionofshockwavesindriven
supersonicturbulence,A&A
*
362,333341.
[21] MacLow, M.M., Klessen, R.S., Burkert, A., and
Smith,M.D.(1998),Kineticenergydecayratesof
supersonic and superAlfvenic turbulence in star
formingclouds,Phys.Rev.Lett.80,27542757.
[22] Boldyrev, S. (2002), KolmogorovBurgers model
forstarformingturbulence,ApJ
**
569,841845.
[23] Padoan, P., Jimenez, R., Nordlund, ., and Bold
yrev, S. (2004), Structure function scaling in
compressible superAlfvenic MHD turbulence,
Phys.Rev.Lett.92,19.
[24] Kritsuk, A.G. and Norman, M.L. (2004), Scaling
relations for turbulence in the multiphase
interstellarmedium,ApJ
**
601,L55L88.
[25] Folini, D. and Walder, R. (2006), Supersonic
turbulence in shockbound interaction zones I:
symmetricsettings,A&A
*
,inpress(astroph).
[26] WalderR.andFoliniD.(2000),Onthestabilityof
colliding flows: radiative shocks, thin shells, and
supersonic turbulence, Astrophysics and Space
Science274,343352.
[27]Walder,R.andFoliniD.(1998), Knots,filaments
andturbulenceinradiativeshocks,A&A
*
330,L21.
[28] Heitsch, F., Burkert, A., Lee, W., Slyz, A.D, and
Devriendt, J.E.G. (2005), Formation of Structure
in Molecular Clouds: A Case Study, ApJ** 633,
L113L116.
[29]VzquezSemadeni,E.,Ryu,D.,Passot,T.,Gonz
lez, R., and Gazol, A. (2006), Molecular Cloud
Evolution.I.MolecularCloudandThinColdNeut
ralMediumSheetFormation,ApJ**643,245259.
[30] Vishniac, E.T. (1994), Nonlinear instabilities in
shockboundedslabs,ApJ
**
428,186208.
[31] Kritsuk, A.G., Norman, M.L., Padoan, P. (2006),
Adaptive mesh refinement for supersonic
molecularcloudturbulence,ApJ
**
638,L25L28.
[32]Boris, J.P,Grinstein,F.F,Oran, E.S., andKolbe,
R.L (1992), New insights into large eddy simul
ation,Fluid.Dynam.Res.10,199228.
[33] Porter, D.H., Pouquet, A., and Woodward, P.R.
(1994),Kolmogorovlikespectraindecayingthree
dimensional supersonic flows, Phys. Fluids 6,
21332142.
[34]Garnier,E.,Mossi,M.,Sagaut,P.,Comte,P.,and
Deville,M.(1999),Ontheuseofshockcapturing
schemes for large eddy simulation, Journal of
ComputationalPhysics153,273311.
*A&A=AstronomyandAstrophysics
**ApJ=AstrophysicalJournal