Mon.Not.R.Astron.Soc.000,1–??(2013) Printed31January2014 (MNLATEXstylefilev2.2) Simulating star formation in Ophiuchus O. Lomax(cid:63)1, A. P. Whitworth1, D. A. Hubber2,3, D. Stamatellos4 and S. Walch5,6 1School of Physics and Astronomy, Cardiff University, Cardiff CF24 3AA, UK 2University Observatory, Ludwig-Maximilians-University Munich, Scheinerstr.1, D-81679 Munich, Germany 3Excellence Cluster Universe, Boltzmannstr. 2, D-85748 Garching, Germany 4Jeremiah Horrocks Institute, University of Central Lancashire, Preston, Lancashire, PR1 2HE, UK 4 5Physikalisches Institut der Universita¨t zu K¨oln, Zu¨lpicher Strasse 77, D-50937 Cologne, Germany 1 6Max-Planck-Institut fu¨r Astrophysik, Karl-Schwarzschild-Str. 1, 85748 Garching 0 2 n 31January2014 a J 0 ABSTRACT 3 Wehavesimulatedstarformationinprestellarcores,usingSPHandinitialconditions informedbyobservationsofthecoresinOphiuchus.Becausetheobservationsarelim- ] itedtotwospatialdimensionsplusradialvelocity,wecannotinferinitialconditionsfor R the collapse of a particular core. However, with a minimum of assumptions (isotropic S turbulence with a power-law spectrum, a thermal mix of compressive and solenoidal . h modes,acriticalBonnor-Ebertdensityprofile)wecangenerateinitialconditionsthat p match,inastatisticalsense,thedistributionsofmass,projectedsizeandaspectratio, - thermalandnon-thermalone-dimensionalvelocitydispersion,observedinOphiuchus. o The time between core-core collisions in Ophiuchus is sufficiently long, that we can r t simulate single cores evolving is isolation, and therefore we are able to resolve masses s a well below the opacity limit. We generate an ensemble of 100 cores, and evolve them [ withnoradiativefeedbackfromthestarsformed,thenwithcontinuousradiativefeed- back, and finally with episodic radiative feedback. With no feedback the simulations 2 producetoomanybrowndwarfs,andwithcontinuousfeedbacktoofew.Withepisodic v radiative feedback,boththe peak oftheprotostellar massfunction (at 0.2M )and 7 ∼ (cid:12) 3 the ratio of H-burning stars to brown dwarfs are consistent with observations. The 2 mass of a star is not strongly related to the mass of the core in which it forms. Low- .7 cmoarsessc(Mores (M(cid:38)co1rMe ∼)0u.1suMa(cid:12)lly) tfreangdmteonctoilnlatopsseevinertaolsoinbgjelectosbwjeitchts,awbhroearedams haisgshr-amnagses. 1 core (cid:12) 0 Key words: Stars:formation,stars:low-mass,stars:massfunction,stars:pre-Main- 4 Sequence, hydrodynamics. 1 : v i X r 1 INTRODUCTION In Turbulent Fragmentation, molecular clouds do not a collapse on a global scale, but turbulent flows within the The origin of the stellar initial mass function (IMF), and cloud produce dense cores (e.g. Padoan & Nordlund 2002; whyitappearstobeapproximatelyuniversal,remainmajor Hennebelle&Chabrier2008,2009),roughlybetween0.01pc problems in star formation. Two main scenarios have been and 0.1pc across. Cores are usually well separated from proposed to addresses these problems. one another, and those that are gravitationally bound col- InCompetitiveAccretion,partofamolecularcloudcol- lapse to form stars (e.g. Andre, Ward-Thompson & Bar- lapses, thereby providing a gas-rich environment in which sony 1993, 2000). Star formation in cores has been sim- stars form. The most massive stars preferentially accrete ulated numerically by Bate (1998, 2000); Horton, Bate the most gas; the least massive stars may be ejected from & Bonnell (2001); Matsumoto & Hanawa (2003); Good- thesystembyN-bodyinteractionsandceaseaccreting(e.g. win & Whitworth (2004); Delgado-Donate, Clarke & Bate Bonnell et al. 1997; Reipurth & Clarke 2001; Bonnell, Vine (2004);Delgado-Donateetal.(2004);Goodwin,Whitworth & Bate 2004; Bate & Bonnell 2005; Bate 2009a). This sce- &Ward-Thompson(2004,2006);Walchetal.(2009,2010); narionaturallyyieldsadistributionofstellarmassessimilar Walch, Whitworth & Girichidis (2012). in shape to the IMF (Kroupa 2001; Chabrier 2005), with a peak close to the cloud’s original Jeans mass. Recent millimetre and submillimetre observations of starformingregions(e.g.Motte,Andre&Neri1998;Testi& Sargent1998;Johnstoneetal.2000;Motteetal.2001;John- (cid:63) E-mail:[email protected] stoneetal.2001;Stankeetal.2006;Enochetal.2006;John- c 2013RAS (cid:13) 2 O. Lomax, A. P. Whitworth, D. A. Hubber, D. Stamatellos and S. Walch stone&Bally2006;Nutter&Ward-Thompson2007;Alves, Ophiuchusaregroupedintoclumps1,labelledOph-A,Oph- Lombardi&Lada2007;Enochetal.2008;Simpson,Nutter B1, Oph-B2, Oph-C, Oph-D, Oph-E and Oph-F. & Ward-Thompson 2008; Rathborne et al. 2009; Ko¨nyves We base the initial conditions for our simulations on etal.2010)haveshownthatthecoremassfunction(CMF), the observations of MAN98 and ABMP07. MAN98 present issimilarinshapetotheIMF,butshiftedupwardsinmass continuum observations of cores with a completeness limit byafactorof3to5.Fromthisitiscommonlyinferredthat of ∼ 0.1M . These data are complemented by observa- (cid:12) the shape of the IMF is directly inherited from the CMF, tionsofcorevelocitydispersionsinOphiuchusbyABMP07. i.e. each core collapses and converts roughly the same frac- There are more recent measurements of core masses and tionofitsmassintothesamenumberofprotostars.Holman sizes in Ophiuchus (e.g. Stanke et al. 2006; Simpson, Nut- et al. (2013) have shown that this is only compatible with ter & Ward-Thompson 2008), but they contain fewer cores the observed stellar multiplicity statistics if a typical core with an associated ABMP07 velocity dispersion than the spawns 4 or 5 stars. MAN98 data. Further details of the MAN98 and ABMP07 The L1688 cloud (hereafter Ophiuchus) is a region observations are given in the next two sections. wherestarsareformingincores.ObservationsofOphiuchus by Motte, Andre & Neri (1998) and Andr´e et al. (2007) 2.1 MAN98 (hereafter MAN98 and ABMP07) show that the cores are wellseparatedandunlikelytointeractdynamicallywithone MAN98 map dust emission at 1.3mm using the MPIfR another before they collapse. In this paper we use SPH to bolometerarrayontheIRAM30mtelescope.Thebeamsize simulate star formation in these cores, using initial condi- is 11(cid:48)(cid:48), corresponding to 1500AU at Ophiuchus, and they tions informed as closely as possible by the MAN98 and aresensitivetohydrogencolumn-densitiesN>1022Hcm−2. ABMP07observations.Alimitationofourapproachisthat Theirmapcoversanareaof∼480arcmin2 andincludesthe cores tend to be embedded in larger scale clumps and fil- Oph-A,Oph-B1,Oph-B2,Oph-C,Oph-D,Oph-EandOph- aments, and material that would accrete onto the cores as F clumps. MAN98 use a multi-scale wavelet analysis to ex- they collapse is not included in the simulations. tract 61 cores, 36 of which are resolved. Two dimensional Deriving reliable initial conditions for simulations of Gaussian distributions are fitted to the cores and the core corecollapse isa difficultinverseproblem. Coremasses can dimensionsaregivenbytheFWHMofthemajorandminor be estimated in a relatively straightforward way from ob- axes.Thecoreshavemassesbetween∼0.1M and∼3M , (cid:12) (cid:12) servedsubmillimetrecontinuumfluxes.However,theintrin- and dimensions between ∼1000AU and ∼20,000AU. sic three-dimensional sizes and shapes of cores can only be Thecoremassesandsizesusedinthispaperareslightly constrainedbyfittingtheparametersofacrediblemodelto differentfromthosegiveninMAN98.Themassofaprestel- observed data (see Lomax, Whitworth & Cartwright 2013, lar core is calculated using hereafter LWC13). The detailed three-dimensional motions S(λ)D2 within cores are even less well constrained by observations M = , (2.1) of radial velocity profiles. A key element of this paper is to core κ(λ)B(λ,T) generate an ensemble of cores which, in a statistical sense, whereS(λ)isthemonochromaticfluxfromthecoreatwave- resemble those observed in Ophiuchus. lengthλ,Disthedistancetothecore,κ(λ)isthemassopac- In Section 2 we review the data used to constrain the ity, and B(λ,T) is the Planck function at temperature T. distributionofcoremasses,sizesandvelocitydispersionsin Forλ=1.3mm,MAN98adoptκ(1.3mm)=0.005cm2g−1. Ophiuchus. We use these data to calibrate a multivariate The dust temperature estimates used by MAN98 have lognormal distribution from which we can draw representa- beenrevisedbyStamatellos,Whitworth&Ward-Thompson tivecoremasses,sizesandvelocitydispersions.InSection3 (2007). These temperatures, T and T are given man98 sww07 we describe how we set up the initial conditions for a core; in Table 1. The distance to Ophiuchus used by MAN98, this includes assigning an ellipsoidal shape to the core and D =160pc,hasalsobeenrevised,byMamajek(2008), man98 giving it a velocity field and a density profile. In Section to D = 139±6pc. To account for this, each mass given m08 4 we outline the SPH code used. For each core, we perform by MAN98 must be transformed thus: simulationswithnoradiativefeedback,continuousradiative B(1.3mm,T )D2 feedback, and episodic radiative feedback. In Section 5 we M → man98 m08 M , (2.2) present the results of the simulations and in Section 6 we man98 B(1.3mm,Tsww07)Dm2an98 man98 discuss and summarise the results. Similarly, each size must be transformed thus: D R → m08 R . (2.3) man98 D man98 man98 ThedusttemperaturesT arecoolerthanT ,which sww07 man98 2 OBSERVATIONS increases the estimated core masses. However, the revised Ophiuchus is a star forming region ∼140pc away from the distance Dm08 is less than Dman98, which decreases the core Sun (Mamajek 2008). It has been well observed in the mil- masses. The combined effect modestly increases the masses limetre and submillimetre continuum (e.g. Motte, Andre & of cores in Oph-A, Oph-E and Oph-F, whilst leaving the Neri 1998; Stanke et al. 2006; Simpson, Nutter & Ward- Thompson 2008), which traces the column density of dust, 1 MAN98 refer to these groups as “cores”, and to the smaller and allows observers to estimate the masses and sizes of densercondensationswithinthemas“clumps”.Wehaveadopted cores.Lada(1999)calculatesthatOphiuchushastotalmass thenownear-universaloppositeconventionthatthesmalldense Moph∼1to2×103M(cid:12) anddiameterDoph∼1pc.Thecoresin condensationsarecores,andtheyaregroupedinclumps. c 2013RAS,MNRAS000,1–?? (cid:13) Simulating star formation in Ophiuchus 3 Region Tman98 Tsww07 Mman98 2.3below,andTable3inonlinematerial)havefreefalltimes × (K) (K) t ∼ 4+4 ×104yrs. Thus, whereas interactions between FF −2 Oph-A 20 11 1.77 coresmayhavebeenimportantindeliveringthedistribution Oph-B 12 10 1.01 of core properties we observe today, the future evolution of Oph-C 12 10 1.01 these cores should not be much influenced by interactions. Oph-D 12 10 1.01 This conclusion is born out in our simulations, where the Oph-E 15 10 1.40 first protostar usually forms after about one freefall time, Oph-F 15 10 1.40 andanysubsequentstarformationtendstobeconcentrated Oph-J - 10 - inthevicinityofthisprotostar.Asafurtherprecaution,we Table 1. Dust temperatures from MAN98 and SWW07. The terminate the simulations after 2×105yrs. fourthcolumngivesthecorrectionfactorappliedtotheMAN98 Treating individual cores in isolation allows us to per- coremasseswhenweadoptT =Tsww07 andD=139pc. formsimulationswithsufficientlyhighresolutiontocapture the formation of protostars close to the opacity limit, and masses of cores in Oph-B, Oph-C and Oph-D essentially to perform many realisations, but it deprives us of the pos- unchanged (see Table 1). sibility of considering the effects of – for example – accre- After these adjustments have been made, the MAN98 tionflowsandtidalperturbationsfromthesurroundings.A CMF peaks at M ∼ 0.3M . This is lower than the large-scalesimulationtreatingthewholeofOphiuchuswith PEAK (cid:12) values found for the Ophiuchus cores by other groups, i.e. the same resolution is currently beyond our computational M ∼0.9M (Stanke et al. 2006) and M ∼0.4M resources,butitshouldbefeasibletosimulateawholeclump PEAK (cid:12) PEAK (cid:12) (Simpson,Nutter&Ward-Thompson2008).Thedifferences (e.g. Oph-A). inM mustbeattributedtoacombinationofthediffer- PEAK entwavelengthsused;thedifferentassumptionsmadeabout dustopacities,dusttemperatures,andsourcedistances;and 2.3 Lognormal distribution the different procedures used to identify cores, distinguish Together, MAN98 and ABMP07 provide 61 core masses, them from the background, and correct for incompleteness 36 mean core radii and 27 core velocity dispersions; 20 at low masses and poor statistics at high masses. We also coreshaveallthree.Weusethesedatatocalibrateamulti- notethatstudiesofotherstarformationregions,e.g.Perseus variate lognormal distribution of core properties. To within (Enochetal.2008),thePipeNebula(Rathborneetal.2009) the errors of small number statistics, and modulo the as- and Aquilla (Ko¨nyves et al. 2010) find CMFs with M PEAK sumption of a lognormal distribution, this reproduces sta- between ∼0.6M and ∼1.0M . We have therefore dou- (cid:12) (cid:12) tistically the observed properties of cores in Ophiuchus, bled the MAN98 masses, so that the peak of the CMF is and their correlations. Specifically, we draw values of x ≡ at M ∼0.6M , i.e. between the Stanke et al. (2006) PEAK (cid:12) (log(M),log(R),log(σnt)) from a probability density andSimpson,Nutter&Ward-Thompson(2008)values,and (cid:18) (cid:19) within the range found for other star formation regions. P(x)= 1 exp −1(x−µ)TΣ−1(x−µ) , (2π)3/2|Σ| 2 (2.6) 2.2 ABMP07 where ABMP07measurethewidthoftheN2H+(1–0)emissionline µM from 27 prestellar cores in Ophiuchus. Of these 27 cores, µ≡µ , (2.7) R 26 have mass estimates from MAN98. The line width gives µ c the radial velocity dispersion, σ , of the N H+ molecules 1d 2 and in the core. In general there are thermal and non-thermal contributionstothelinewidth,σ2 =σ2+σ2 .Thethermal σ2 ρ σ σ ρ σ σ contribution is given by 1d t nt Σ ≡ ρ σM σ M,Rσ2M R ρM,σnt σMσσnt . ρ M,RσMσR ρ σR σ R,σnσt2R σnt σt2 = mkbT , (2.4) M,σnt M σnt R,σnt R σnt σnt (2.8) mol Hereµ isthearithmeticmeanoflog(x),σ isthestandard x x wheremmol =4.8×10−23gisthemassofanN2H+molecule. deviation of log(x) and ρx,y is the Pearson’s correlation co- The non-thermal contribution, efficient of log(x) and log(y). The values of these terms are (cid:18) k T (cid:19)1/2 giveninTable2.Fig.1showstheprobabilitydensityfunc- σ = σ2 − b sww07 , (2.5) tionP(x)projectedthroughallthreeofitsdimensions.Su- nt 1d m mol perimposed are the observational data points and 100 data givesthemagnitudeofmacroscopicgasmotions,i.e.turbu- points drawn randomly from the distribution. lence, rotation and contraction or expansion. Since we are adopting the reduced temperatures from SWW07, we de- rivesomewhathighervaluesofσ thanABMP07,between nt 3 INITIAL CONDITIONS ∼0.05kms−1 and ∼0.3kms−1. ABMP07alsomeasurethebulkradialvelocitiesofcores While the mass, size and velocity dispersion of a core are in Ophiuchus. From the dispersion in the bulk radial veloc- relativelyeasytoquantifyfromobservations,thesamecan- ities of the cores within a clump, they estimate that, on not be said for the intrinsic three-dimensional shape, den- average, a core should collide with another core after a few sity profile and internal velocity field. Here we define the times 105yrs. In contrast, the individual cores (see Section methodology we adopt for assigning these properties. c 2013RAS,MNRAS000,1–?? (cid:13) 4 O. Lomax, A. P. Whitworth, D. A. Hubber, D. Stamatellos and S. Walch log(M/M ) log(σnt/kms−1) ⊙ -2.0 -1.5 -1.0 -0.5 0.0 0.5 -1.8 -1.6 -1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 4.0 4.0 3.8 3.8 3.6 3.6 3.4 3.4 AU) 3.2 3.2 AU) / / R R g( 3.0 3.0 g( o o l l 2.8 2.8 2.6 2.6 2.4 2.4 2.2 2.2 -1.6 -1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 -0.4 -0.4 log(σnt/kms−1) -0.6 -0.6 1)− -0.8 -0.8 1)− s s m m /k -1.0 -1.0 /k nt nt σ σ g( -1.2 -1.2 g( o o l l -1.4 -1.4 -1.6 -1.6 -1.8 -1.8 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 log(M/M ) ⊙ Figure 1. The multivariate lognormal distribution, P(x) where x=(log(M),log(R),log(σnt)). (a) shows the projections through log(σnt); (b) through log(M); and (c) through log(R). The concentric ellipses show the 1σ, 2σ and 3σ regions of the distribution. ThegreencirclesarerandomlydrawnpointsfromP(x).TheredsquaresaretheobservationaldatafromMAN98andABMP07.Because thereisalimtednumberofcasesforwhichmorethanoneoftheseparametersisknown,thedistributionofobervationalpointsisshifted relativetothefitteddistribution. Parameter Value generated with µM [log(M/M(cid:12))] -0.57 µ [log(R/AU)] 3.11 R µσnt [log(σnt/kms−1)] -0.95 A=1, σM [log(M/M(cid:12))] 0.43 B =exp(τGb), (3.1) σ [log(R/AU)] 0.27 σRσnt [log(σnt/kms−1)] 0.20 C =exp(τGc), ρ 0.61 M,R ρM,σnt 0.49 where Gb and Gc are random numbers drawn from a Gaus- ρ 0.11 R,σnt sian distribution with zero mean and unit standard devia- Table 2.Arithmeticmeans,standarddeviationsandcorrelation tion.ForOphiuchus,agoodfitisobtainedwithτ=0.6,and coefficientsoflog(M),log(R)andlog(σnt)forcoresinOphiuchus improvements to the fit obtained by introducing additional parameters do not appear to be justified (LWC13). Once the axes have been generated using Eqn. (3.1), they are re- ordered so that A≥B ≥C and normalised so that A=1. 3.1 Shapes Then, given a mean radius R from the distribution in Eqn. (2.6), the core axes are set to Inprojection,thecoresinOphiuchushaveirregular,andof- tenelongated,shapes,whichshownosignificantcorrelation with mass, size or velocity dispersion. LWC13 show that, if R A = , one assumes the intrinsic three-dimensional shape of a core core (BC)13 isarandomlyorientedtriaxialellipsoid,theobservedaspect (3.2) B =BA , ratioscanbefittedwithaoneparametermodel.Specifically, core core C =CA . therelativesizesoftheprincipalaxesoftheellipsoidcanbe core core c 2013RAS,MNRAS000,1–?? (cid:13) Simulating star formation in Ophiuchus 5 3.2 Density profile (k ,k ,k ), in the range 0 to 64. For a power spectrum x y z P ∝ k−α the energy density in k-space is E ∝ k−(α+2). Observationssuggest(e.g.Alves,Lada&Lada2001;Harvey k k Hence, with α=2, E ∝ k−4 and the energy is strongly etal.2001;Ladaetal.2008)thatdensecoresoftenapproxi- k concentratedtowardslowspatialfrequencies(i.e.longwave- matetothedensityprofileofacriticalBonnor-EbertSphere lengths).Theamplitude,a,andphase,ϕ,ofeachindividual (Bonnor 1956), i.e. ρ = ρ e−ψ(ξ), where ρ is the central C C mode, k, are set by generating a vector G of three random density,ψ istheIsothermalFunction,ξ isthedimensionless numbers drawn from a Gaussian distribution having zero radius, and the boundary is at ξ = 6.451. The FWHM of B mean and unit variance, and a vector R of three random the column-density through a critical Bonnor-Ebert Sphere numbers drawn from a uniform distribution between zero corresponds to ξ =2.424, so the density at (x,y,z) is FWHM and one. Then we put given by (cid:112) a(k) = E(k)G, (3.6) (cid:18) x2 y2 z2 (cid:19)1/2 ξ = ξ + + , (3.3) ϕ(k) = 2πR, (3.7) FWHM A2 B2 C2 core core core M ξ e−ψ(ξ) The corresponding contribution to the dimensionless veloc- ρ = core B , ξ<ξ , (3.4) ity field is given by 4πA B C ψ(cid:48)(ξ ) B core core core B vˆ(k) = a(k)cos(ϕ(k))+ia(k)sin(ϕ(k)), (3.8) where ψ(cid:48) is the first derivative of ψ. Using this density profile implies that the cores are in v(x) = 1 Re(cid:18)(cid:90) vˆ(k)eikxd3k(cid:19) ; (3.9) hydrostatic equilibrium, contained by an external pressure. (2π)2 For the simulated cores is this paper, this is generally not Eqn. (3.9) is evaluated numerically using the fast Fourier thecase.However,thedensityprofiledoesreproducetheflat transformlibraryFFTW(Frigo&Johnson2005)toproduce centralregionandpower-lawenvelopethatisoftenobserved values of the velocity field on a 1283 grid. in cores (e.g. Kirk, Ward-Thompson & Andr´e 2005; Roy et al. 2013). 3.3.2 Ordered motions We introduce large scale rotation and radial motion 3.3 Velocity fields into the velocity field by simply modifying the phases Spectroscopic observations of prestellar cores indicate that and symmetries of the k = 1 modes, i.e. k = theirvelocityfieldsmaycontainorderedrotation(e.g.Good- (1,0,0), (0,1,0)and(0,0,1). Specifically, we require that man et al. 1993), ordered inward and/or outward motion a(1,0,0) r ω −ω (e.g. Keto et al. 2006) and random turbulence (e.g. Dubin- x z y ski, Narayan & Phillips 1995). If the corresponding contri- a(0,1,0)=−ωz ry ωx , a(0,0,1) ω −ω r butions to the radial velocity dispersion are σ , σ and y x z rot rad (3.10) σ , the total is given by ϕ(1,0,0) π/2 π/2 π/2 turb σ2 =σ2 +σ2 +σ2 . (3.5) ϕ(0,1,0)=π/2 π/2 π/2 . nt rot rad turb ϕ(0,0,1) π/2 π/2 π/2 For the simplest case of a spherically symmetric core ro- Here (r ,r ,r ) are radial velocity amplitudes and x y z tating about an axis perpendicular to the line of sight, it (ω ,ω ,ω ) are rotational velocity amplitudes; the diame- x y z requires two parameters to specify the relative contribu- ter of the core is scaled to π, so that it corresponds to the tions to σ from rotation, isotropic pulsation and isotropic nt central half wavelength of these modified modes. The out- turbulence. An additional parameter is required if the ro- come of applying this modification is shown in Fig. 2. tation axis is not perpendicular to the line of sight. The The radial velocity amplitudes are proportional to the problembecomesevenmorecomplicatedifthecoreisellip- rates at which the core expands along the corresponding soidal,sincethepulsationsandturbulencearethenunlikely axis. For example, if r < 1, then the core is contracting x tobeisotropic;forexample,contractionislikelytobefastest along the x-axis, and if r >1, the core is expanding along x along the shortest axis (Lin, Mestel & Shu 1965). We have the x-axis. On the assumption that the core is contracting no way of estimating specific values for any of these pa- fastest along its shortest axis, and slowest along its longest rameters. To overcome this problem, we generate a random axis, we re-order the amplitudes so that r ≤ r ≤ r , and x y z turbulentvelocityfield,characterisedonlybythepowerex- orient the core so that its ellipsoidal axes, A≥B ≥C, are ponent (we adopt α=2; cf. Burkert & Bodenheimer 2000) respectivelyalignedwiththex,yandzaxes.Therotational and the ratio of compressive to solenoidal modes (we adopt amplitudesdefinetheangularvelocityω=(ω ,ω ,ω ),giv- x y z thethermalratio,i.e.M=1/2),butshiftthecentreofthe ing the axis and rate of core rotation; the axis of rotation coresothatitisatthecentreofthelargestmode;theenergy has a random direction. The values of r , r , r , ω , ω x y z x y inthelargestmodeistheninvestedinorderedrotationand and ω are independently drawn from a Gaussian distribu- z compression, whilst the rest is in smaller-scale turbulence. tion having zero mean and unit variance. This ensures that the energy in these large-scale components of the velocity field still subscribes to the same P ∝ k−2 power law as 3.3.1 Turbulent motions k the smaller-scale ones that deliver internal turbulence. We To construct a dimensionless turbulent velocity field, we stressthatonlythek=1modesaremodified.Allmodeswith work in a (2π)3 cubic domain, and populate all modes k>1aregeneratedasdescribedinSection3.3.1.Hencewe for which the wave-vector, k has integer components, have a velocity field composed of ordered rotation, ordered c 2013RAS,MNRAS000,1–?? (cid:13) 6 O. Lomax, A. P. Whitworth, D. A. Hubber, D. Stamatellos and S. Walch (a) (b) (c) (d) Figure 2. 2D velocity fields generated in an analogous manner to the 3D velocity fields described in Section 3.3.1 and 3.3.2. All four frames show the region π x π and π y π. From left to right: (a) a compressive velocity field with a(1,0)=( 1,0), −2 ≤ ≤ 2 −2 ≤ ≤ 2 − a(0,1)=(0, 1),ϕ(1,0)=(π,0)andϕ(0,1)=(0,π);(b)asolenoidalvelocityfieldwitha(1,0)=(0,1),a(0,1)=( 1,0),ϕ(1,0)=(0,π) − − and ϕ(0,1) = (π,0); (c) a turbulent field with modes generated as described in Section 3.3.1, but without the k=1 modes; (d) the additionofallthreefieldsgivingasuperpositionoflarge-scaleorderedmotionandsmallerturbulentmodes. radialmotionsand stochasticturbulentmotions,andbased the transport of cooling radiation against dust opacity and on just two parameters, α and M. molecular opacity, the sublimation of dust, the ionisation and dissociation of hydrogen and the ionisation of helium. It has been extensively tested, and reproduces very accu- 3.3.3 Interpolation rately the detailed results of Masunaga & Inutsuka (2000). To compute the initial velocity of an SPH particle, i, we assign it a position vector, 4.2 Accretion luminosity (cid:18) (cid:19) s = π xi , yi , zi , (3.11) Thedominantcontributiontotheluminosityofaprotostar i 2 Acore Bcore Ccore is from accretion, and estimate its dimensionless velocity by interpolation on fGM M˙ L (cid:39) (cid:63) (cid:63) , (4.1) thegridofcomputedvalues.Finallythedimensionlessveloc- (cid:63) R (cid:63) itiesoftheSPHparticlesarescaledsothattheir1Dvelocity where, f = 0.75 is the fraction of the accreted material’s dispersion matches a value of σ drawn from Eqn. (2.6). nt gravitationalenergythatisradiatedfromthesurfaceofthe protostar(therestispresumedtoberemovedbyradiation, andbybipolarjetsandoutflows; Offneretal.2009);M is (cid:63) 4 NUMERICAL METHOD the mass of the protostar; M˙ is the rate of accretion onto (cid:63) the protostar; and R = 3R is the approximate radius of (cid:63) (cid:12) 4.1 SPH code a protostar (Palla & Stahler 1993). We use the seren (Hubber et al. 2011) implementation In the sequel, each set of initial conditions is used to of grad-h SPH (Springel & Hernquist 2002; Price & Mon- run three simulations. In the first, radiative feedback from aghan 2004) with smoothing parameter η=1.2 (hence typi- sinksisneglected(NRF).Inthesecond,materialassimilated cally∼56neighbours).GravityiscalculatedwithaBarnes- byasinkispresumedtobeaccretedinstantaneouslybythe Hut tree and a gadget-style multipole acceptance crite- central protostar, and therefore delivers continuous radia- rion (Springel & Hernquist 2002). The code invokes mul- tive feedback (CRF). In the third, material assimilated by a tiple particle timesteps and a leapfrog drift-kick-drift inte- sinkisassumedtoaccumulateinaninneraccretiondisc,and grationscheme.Time-dependentartificialviscosityisimple- thenaccreteontothecentralstarinaburst,givingepisodic mented using the Morris & Monaghan (1997) formulation radiativefeedback(ERF).TheNRFsimulationsareperformed with α=1, α =0.1 and β =2α. purelyforcomparison,andourmainconcerniswiththedif- min All simulations are set up with m = 10−5M , so ferencebetweencontinuousandepisodicradiativefeedback. sph (cid:12) the mass resolution is ∼ 10−3M . A gas concentration In order for the disc around a protostar to fragment (cid:12) that is gravitationally bound and has density higher than gravitationally, at radius R, two criteria must be fulfilled. ρ = 10−9gcm−3 is replaced with a sink particle (Bate, First, gravity must be able to overcome thermal and cen- sink Bonnell & Price 1995) having radius R ≈ 0.2AU. Any trifugal support, sink SPH particles that subsequently enter this radius and are c(R)κ(R) gravitationally bound to the sink are assimilated by it. We Σ(R) >∼ πG , (4.2) have not used the more sophisticated NewSinks method, where Σ(R) is the local surface-density, c(R) the sound presentedbyHubber,Walch&Whitworth(2013),however, speed,andκ(R)theepicyclicfrequency;(Toomre1964).Sec- for this value of ρ , the two methods are reasonably well sink ond,thecondensingfragmentmustcoolfastenoughtoavoid converged. undergoing an adiabatic bounce and then being sheared Heating and cooling of the gas are treated with the apart, i.e. schemedescribedinStamatellosetal.(2007).Thistakesac- count of heating by cosmic rays and background radiation, t < t , (4.3) cool ∼ orb c 2013RAS,MNRAS000,1–?? (cid:13) Simulating star formation in Ophiuchus 7 where t is the gas cooling time, and t is the orbital cool orb period (Gammie 2001). Simulations that include instanta- neousaccretionandcontinuousradiativefeedback(e.g.Bate 1 NRF (a) 2009b; Krumholz 2006; Krumholz et al. 2010; Offner et al. 2009, 2010; Urban, Martel & Evans 2010) find that proto- )) M sthteelrleafrordeisdcos naoret sspoawwnarlmowt-hmaatssthHe-ybucranninnogtsftraargsmoernbtroawndn (log( 0.1 P dwarfs. However, there is observational evidence suggesting that accretion onto a protostar may sometimes be episodic, 0.01 1 ERF (b) i.e. it may occur intermittently, in short, intense bursts (e.g. Herbig 1977; Dopita 1978; Reipurth 1989; Hartmann & Kenyon 1996; Greene, Aspin & Reipurth 2008). In par- )) M ticular, FU Ori-type stars (e.g. Herbig 1977; Greene, Aspin og( 0.1 & Reipurth 2008; Peneva et al. 2010; Green et al. 2011; P(l Principe et al. 2013) exhibit large increases in their bright- ness that last for between a few and a few hundred years. The physical cause of episodic accretion is uncertain. The 0.01 1 CRF (c) critical issue is the duty-cycle, specifically the duration of the low-luminosity phase: if this is longer than the cooling timescale, typically a few times 103yrs, the disc can cool M)) and fragment. og( 0.1 (l For the purpose of the simulations presented here P we adopt the phenomenological ERF model of Stamatel- los, Whitworth & Hubber (2011, 2012) (hereafter SWH11), 0.01 which is based on the calculations by Zhu, Hartmann & 1 NRF (d) ERF Gammie (2009, 2010) and Zhu et al. (2010). In the outer CRF disc(outisidethesink),angularmomentumisredistributed )) bygravitationaltorques,allowingmaterialtospiralinwards M andintothesink.However,insidethesinkthematerialnear (log( 0.1 P the protostar is so hot that the inner disc becomes gravita- tionally stable, and there are no gravitational torques to transport angular momentum outwards. Consequently this 0.01 material cannot spiral inwards any further, and instead ac- -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 log(M/M ) cumulatesintheinnerdisc.Thiscontinuesuntiltheaccumu- (cid:12) lated material becomes so hot that it is ionised thermally, sufficiently to couple to the magnetic field. The magneto- rotational instability (MRI) then cuts in and the resulting Figure3.Protostarmassfunctions.Thetopthreeframes(a,b,c) torquesenablematerialtospiralinwardsandontothestar, givethemassfunctionsfromsimulationswith,respectively,NRF, giving a short burst of very high luminosity. The downtime ERF and CRF. The black histograms have bins equally spaced in – during which material is accumulating in the inner disc, log(M) and the red lines are kernel smoothed density functions. the luminosity is low, and the outer disc may be able to The bottom frame (d) shows the smoothed NRF, ERF and CRF densityfunctionsonasingleplot. fragment – is given by (cid:18) M (cid:19)2/3(cid:18) M˙ (cid:19)−8/9 t ∼13kyr (cid:63) sink ; (4.4) down 0.2M 10−5M yr−1 (cid:12) (cid:12) M˙ is the rate at which material flows into the sink. sink 5.1 Protostellar masses 5 RESULTS Fig.3shows,ashistogramsandsmoothedprobabilitydensi- Eachcoreisevolvedfor2×105yrs,whichislongerthanthe ties2, the protostellar mass functions from the simulations. freefall time of the least-dense core. Of the 100 cores sam- Both the NRF and ERF simulations deliver a wide range of pled from the lognormal distribution, 60 collapsed to form masses from a few Jupiter masses to between one and two at least one star; note that the simulations with different solar masses. The CRF simulations deliver roughly the same feedback are identical up until this point. The star forma- uppermasslimit,butnoprotostarsbelowabout50Jupiter tion efficiencies and numbers of protostars formed per core masses. In the NRF simulations, the mean efficiency (total aregiveninTable3intheonlinematerial.Inthefigureswe mass of stars formed divided by core mass) is 71%, with on use the naming convention NUM FBK, where NUM is the core average 3.3 stars and 3.4 brown dwarfs per core. In the ERF number (identifying the initial conditions) and FBK is the simulations,theefficiencyisagain71%,withonaverage3.1 feedback type. starsand1.5browndwarfspercore.IntheCRFsimulations, c 2013RAS,MNRAS000,1–?? (cid:13) 8 O. Lomax, A. P. Whitworth, D. A. Hubber, D. Stamatellos and S. Walch theefficiencyis59%,withonaverage1.6starsand0.1brown dwarfs per core. 1 NRF (a) Kroupa(2001) Chabrier(2005) Ratio of stars to brown dwarfs )) M Andersenetal.(2008)haveperformedasurveyofstarform- (log( 0.1 P ing regions and find that low mass stars outnumber brown dwarfs by a factor N(0.08M <M ≤1.0M ) 0.01 A= (cid:12) (cid:12) =4.3±1.6. (5.3) 1 ERF (b) N(0.03M <M ≤0.08M ) Kroupa(2001) (cid:12) (cid:12) Chabrier(2005) In the simulations presented here, A =2.2±0.3, A = nrf erf )) 3.9±0.6andA =17±8.ThisconfirmsthatNRFproduces M toomanybrowcnrfdwarfs,andsuggeststhatCRFproducestoo (log( 0.1 P few. ERF appears to produce about the right number. 0.01 Comparison to the IMF 1 CRF (c) Kroupa(2001) Chabrier(2005) InFig.4,wehaveplottedthemassdistributionsfromFig.3 with the Kroupa (2001; K01) and Chabrier (2005; C05) fits )) M ttohethCe0I5MIFM.FTahteMK0pCe10a5kIM=F0p.2eaMk(cid:12)s.atInMtphKe0ea1ks=imu0.l0a8tiMon(cid:12)s,wainthd P(log( 0.1 NRF, ERF and CRF, the mass distributions peak at Mnrf ≈ peak 0.1M , Merf ≈0.2M and Mcrf ≈0.4M , respectively. (cid:12) peak (cid:12) peak (cid:12) ThepeaksoftheNRFandERFmassdistributionsareingood 0.01 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 agreementwiththoseofK01andC05respectively;thepeak log(M/M ) of the CRF mass distribution appears to be rather too high. (cid:12) There are no high-mass cores in Ophiuchus, and therefore none of the simulated IMFs match the observed IMFs at Figure 4. As Fig. 3, but here the simulated protostellar mass high masses. functions (in black) are compared with the K01 and C05 IMFs (inredandbluerespectively).Thethreeframesshowprotostellar massfunctionforsimulationswith(a)NRF,(b)ERF,and(c)CRF. ThedashedpartsoftheK01andC05IMFsareinthestellarmass Summary: masses regime (M 0.08M(cid:12)) and the dotted parts are in the brown ≥ For masses less than ∼1M , the ERF simulations deliver dwarf mass regime (M < 0.08M(cid:12)), which is not as well con- (cid:12) strainedasthestellarregime.Allmassfunctionsarenormalised a protostellar mass function which is compatible with ob- sothat(cid:82)+∞P(log(M))dlog(M)=1. servedIMFs.Incontrast,theCRFsimulationsappeartopro- −∞ ducetoofewlow-massH-burningstarsandbrowndwarfs.In the ERF simulations each core produces on average N ≈ ERF 5.2 Fragmentation 4.6 stars with a star formation efficiency of η ≈ 0.7. ERF ThisisinagreementwiththestatisticalinferenceofHolman Mechanisms et al. (2013) that values of N =4.3±0.4 and η=1.0±0.3 In many simulations, the core collapses to form a central are required to reproduce the observed multiplicity statis- protostarsurroundedbyanaccretiondisc.Ifthediscissuf- tics (η >1 simply implies that the cores accrete additional ficiently massive and cold, it fragments to form additional massastheyformstars).IntheCRFsimulationsthenumber protostars, typically with masses smaller than the central of stars formed is significantly lower, N = 1.7, and the CRF protostar.AnexampleofthisisshowninFig.5.Weseethat efficiency is a little lower, η =0.59. CRF thetimescaleofdiscfragmentationisoforder103yrs.This mechanismiswelldocumented(e.g.Stamatellos,Hubber& Whitworth 2007; Stamatellos & Whitworth 2008, 2009a,b; 2 Thesmoothedprobabilitydensityisgivenby Thies et al. 2010; Stamatellos et al. 2011) and has been 1 (cid:88)N 1 (cid:18) (log(M) log(Mi))2(cid:19) shown to reproduce the observed properties of stars and P(log(M))= exp − . brown dwarfs rather well (e.g. Stamatellos & Whitworth N √2πh2 − 2h2 i=1 2009a). (5.1) In most simulations the central regions are fed by fila- IntheexpectationthatP(log(M))isroughlylognormal,weset mentary streams of material, and in some cases these fila- (cid:18)4σˆ5(cid:19)51 ments fragment before a central protostar forms; this also h= (5.2) 3N occurs on time scales of order 103yrs. An example of this is shown in Fig. 6. This core has a relatively high mass where σˆ is the standard deviation of log(M) (Silverman 1998). Thisisausefulmethodofextractingfromnoisyhistogramslarge (M=3.3M(cid:12)) and a low ratio of turbulent to gravitational scalefeaturesthatareotherwisehardtosee. energy(α =0.07).Thecorecollapsesintoafilamentbe- turb c 2013RAS,MNRAS000,1–?? (cid:13) Simulating star formation in Ophiuchus 9 Figure 5. Column density plots from simulation 002 ERF show- Figure 7. Column density plots from the 022 XXX simulations. ing disc fragmentation. From left to right, then top to bottom, The left-hand column gives snapshots from 022 NRF, the middle the times are t=17,18,19and20kyr. This core fragments into column from 022 ERF, and the right-hand column from 022 CRF. three stars and three brown dwarfs. This figure and subsequent Fromtoptobottom,thetimesaret=15,18and23kyr.022 NRF simulationsnapshotsarerenderedusingSplash(Price2007). formsfourstarsandninebrowndwarfs,022 ERFformsfourstars, and022 CRFformsasinglestar. simulations. Second, in the NRF simulations there is a sec- ondpeakaround∼0.01M .Themainpeaksarepopulated (cid:12) by protostars that form from core collapse. The secondary peak in the NRF simulations arises from prolific disc frag- mentation.Thisproducesanabundanceoflowmassobjects, manyofwhichareejectedfromthecore(e.g.Bate,Bonnell & Bromm 2002). Once ejected, accretion ceases and their masses are frozen at low values. In the ERF simulations, high-luminosity outbursts peri- odicallyraisethetemperatureofthegassurroundingapro- tostar,stabilisingitagainstgravitationalfragmentation.Ini- tially,theseoutburstsoccuratintervalsof∆t ∼104yrs, down whereasgravitationalinstabilitiesdevelopontimesscalesof ∆t ∼ 103yrs; therefore there are windows of opportunity gi for fragmentation between outbursts. However, as further protostars form, the intervals between outbursts decrease until eventually ∆t <∆t . Thereafter, further star for- down∼ gi mation is inhibited by regular luminosity outbursts. Thus Figure6.Columndensityplotsfromsimulation098 ERFshowing ERF allows fragmentation to occur, but then regulates the the filament fragmentation. From top to bottom, the times are process after a few protostars have formed. Because fewer t = 19,23,27and36kyrs. This core fragments into eight stars low-mass protostars are formed, there are fewer dynamical andtwobrowndwarfs. ejections,andmoremassisavailabletoaccreteontothecen- tral protostar (cf. Girichidis et al. 2012), so the main peak causeitisparticularlyelongated,withitslongestaxismore is enhanced and the secondary peak is removed. than twice the length of the other two. The filament then In the CRF simulations, once the first protostar has spawnstenprotostars.ThisconfirmstheresultsofGoodwin, formed, the remaining gas is continuously heated and very Whitworth&Ward-Thompson(2004),whoshowedthatlow littlefurtherfragmentationoccurs,sothesesimulationsusu- levelsofturbulencearesufficienttofragmentsphericalcores. allyformsingleprotostarsorbinaries.Fig.7showsanexam- pleofacoreevolvingwithNRF,ERFandCRF;theusualtrend isthatthemostprotostarsareformedwithNRF,followedby Feedback effects ERF, then CRF. Fig.7showsacomparisonofcolumndensityplotsfromsim- ulationsusingdifferentfeedbackmechanisms.Therearetwo Number of objects formed maindifferencesbetweentheprotostellarmassdistributions fromtheNRFandERFsimulations.First,themainpeakshifts Fig. 8 shows that the number of protostars formed per from∼0.1M intheNRFsimulations,to∼0.2M intheERF core is correlated with the core mass. Cores with masses (cid:12) (cid:12) c 2013RAS,MNRAS000,1–?? (cid:13) 10 O. Lomax, A. P. Whitworth, D. A. Hubber, D. Stamatellos and S. Walch 0.0 NRF NRF 10.0 (a) -0.5 E )(cid:12) OR M /NC M/? -1.0 N? 1.0 og( l -1.5 (a) 0.1 -2.0 ERF ERF 10.0 (b) -0.5 E )(cid:12) OR M /NC M/? -1.0 N? 1.0 og( l -1.5 (b) 0.1 -2.0 CRF CRF 10.0 (c) -0.5 E )(cid:12) OR M /NC M/? -1.0 N? 1.0 og( l -1.5 (c) 0.1 -2.0 -1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0 log(MCORE/M ) log(MCORE/M ) (cid:12) (cid:12) Figure 8. The number of protostars formed per core as a func- Figure 9. The distribution of protostar masses as a function tion of core mass. Error bars reflect the Poisson uncertainties in of core mass. The solid line gives the geometric mean mass and the numbers of cores and sinks, per bin. From top to bottom, standard error. The shaded area shows the geometric standard simulationsinvoking(a)NRF,(b)ERFand(c)CRF. deviation. The dashed line gives the arithmetic mean protostar mass.Fromtoptobottom,simulationsinvoking(a)NRF,(b)ERF and(c)CRF. Fragmentation: summary between 0.1M and 0.3M typically form single proto- (cid:12) (cid:12) AsnotedbyStamatellos,Whitworth&Hubber(2012),ERF stars, whereas cores with masses between 3M and 10M (cid:12) (cid:12) providesawindowofopportunityfordiscfragmentation.As typically produce more than ten protostars with ERF, and further protostars form, they too radiate and the window about five with CRF. For convenience, we will refer to cores eventually becomes too short to allow fragmentation. This with M ≤ 1M as low-mass cores, and cores with core (cid:12) meansthat,unlikeCRF,ERFpermitsbrowndwarfstoformby M >1M as high-mass cores. core (cid:12) disc fragmentation, but, unlike NRF, brown dwarf formation Fig. 9 shows the distribution of protostar masses as a is regulated to a level that agrees better with observation. function of core mass. In the NRF and ERF simulations, the Irrespectiveofthetypeoffeedbackinvoked,thesesimu- arithmetic mean mass is roughly constant over the entire range of core masses, viz. M¯nrf = 0.12M and M¯erf = lationsdonotsupportthenotionthattheshapeoftheIMF (cid:63) (cid:12) (cid:63) is inherited from the shape of the CMF. The mean mass 0.19M .AlsoshownonFig.9isthegeometricmeanofthe (cid:12) of a protostar formed in a low-mass core is approximately protostarmassesanditsstandarddeviation.Thisshowsthat the same as one formed in a high-mass core. The difference high-masscoresformawiderrangeofmassesthanlow-mass is that high-mass cores tend to produce more protostars, cores. In low-mass cores, there is little spread in the proto- roughly in proportion to their mass. star masses because most of the available gas goes into the Competitive accretion occurs in our simulations, but central protostar.A disc may form,but itis oftennot mas- only on the scale of individual cores having sufficient mass siveenoughtofragment.High-masscoreshavesufficientgas (M >M ) to form multiple protostars. to form more massive protostars, and there is often enough CORE (cid:12) gas remaining to build a massive disc, which can then frag- mentintolow-massprotostars.Finalmassesaredetermined 5.3 Limitations by competitive accretion for the remaing core mass. In simulations with CRF, low-mass cores tend to form SincethemostmassivecoreinOphiuchushasmass∼5M (cid:12) singleprotostars,andhigh-masscorestendtoformmultiple (and the next most massive ∼ 3M ), and since we evolve (cid:12) protostars, but in smaller numbers than with ERF. our cores in isolation (i.e. they do not accrete extra mate- c 2013RAS,MNRAS000,1–?? (cid:13)