Binary Stars Can Provide the "Missing Photons" Needed for Reionization

0.36 MB·

by  Xiangcheng Ma

Mon.Not.R.Astron.Soc.000,1–6(0000) Printed30March2016 (MNLATEXstylefilev2.2)
Binary Stars Can Provide the “Missing Photons” Needed for
Reionization
Xiangcheng Ma,1∗ Philip F. Hopkins,1 Daniel Kasen,2,3 Eliot Quataert,2 Claude-André
Faucher-Giguère,4 Dušan Kereš5 Norman Murray6† and Allison Strom7
1TAPIR,MC350-17,CaliforniaInstituteofTechnology,Pasadena,CA91125,USA
6 2DepartmentofAstronomyandTheoreticalAstrophysicsCenter,UniversityofCaliforniaBerkeley,Berkeley,CA94720
1 3LawrenceBerkeleyNationalLaboratory,1CyclotronRoad,Berkeley,CA94720
0 4DepartmentofPhysicsandAstronomyandCIERA,NorthwesternUniversity,2145SheridanRoad,Evanston,IL60208,USA
2 5DepartmentofPhysics,CenterforAstrophysicsandSpaceSciences,UniversityofCaliforniaatSanDiego,9500GilmanDrive,LaJolla,CA92093
6CanadianInstituteforTheoreticalAstrophysics,60StGeorgeStreet,UniversityofToronto,ONM5S3H8,Canada
r 7DepartmentofAstrophysics,MC249-17,CaliforniaInstituteofTechnology,Pasadena,CA91125,USA
a
M

8 Draftversion30March2016
2

] ABSTRACT
A Empiricalconstraintsonreionizationrequiregalacticionizingphotonescapefractions f (cid:38)
esc
G 20%,butrecenthigh-resolutionradiation-hydrodynamiccalculationshaveconsistentlyfound
muchlowervalues∼1–5%.Whilethesemodelsincludestrongstellarfeedbackandadditional
.
h processes such as runaway stars, they almost exclusively consider stellar evolution models
p basedonsingle(isolated)stars,despitethefactthatmostmassivestarsareinbinaries.Were-
-
o visitthesecalculations,combiningradiativetransferandhigh-resolutioncosmologicalsimu-
r lationsfromtheFIREproject.Forthefirsttime,weuseastellarevolutionmodelthatincludes
t aphysicallyandobservationallymotivatedtreatmentofbinaries(theBPASSmodel).Binary
s
a masstransferandmergersenhancethepopulationofmassivestarsatlatetimes((cid:38)3Myr)after
[ starformation,whichinturnstronglyenhancesthelate-timeionizingphotonproduction(es-

peciallyatlowmetallicities).Thesephotonsareproducedafterfeedbackfrommassivestars
2
v has carved escape channels in the ISM, and so efficiently leak out of galaxies. As a result,
9 thetime-averaged“effective”escapefraction(ratioofescapedionizingphotonstoobserved
5 1500Åphotons)increasesbyfactors∼4–10,sufficienttoexplainreionization.Whileimpor-
5 tantuncertaintiesremain,weconcludethatbinaryevolutionmaybecriticalforunderstanding
7 theionizationoftheUniverse.
0
. Keywords: binaries:general–stars:evolution–galaxies:formation–galaxies:high-redshift
1
–cosmology:theory
0
6
1
:
v
i 1 INTRODUCTION there is no confirmed Lyman continuum (LyC) detection, neither
X fromindividualgalaxiesnorfromstackedsamples,implyingupper
Theescapefraction(f )ofhydrogenionizingphotonsfromhigh-
ar redshiftstar-forminggeaslcaxiesisperhapsthemostimportantandyet limits of fesc =1–3% (e.g. Leitet et al. 2011, 2013; Bridge et al.
2010; Siana et al. 2010). Even at z∼3, many earlier reports of
most poorly understood parameter in understanding the reioniza-
LyCdetectionfromLymanbreakgalaxies(LBGs)andLyαemit-
tionhistory.Modelsofcosmicreionizationsuggestthat f (cid:38)20%
esc ters(LAEs)haveproventobecontaminatedbyforegroundsources
(e.g. Kuhlen & Faucher-Giguère 2012; Finkelstein et al. 2012;
(e.g.Sianaetal.2015)andalow f about5%hasbeenderived
Robertsonetal.2013,2015)isneededtomatchtheopticaldepth esc
fromsomegalaxysamplesatthisredshift(e.g.Iwataetal.2009;
ofelectronscatteringinferredfromcosmicmicrowavebackground
Boutsiaetal.2011).
(CMB)measurements(e.g.Hinshawetal.2013;PlanckCollabora-
Moreover, the latest generation of cosmological hydrody-
tionetal.2014),assumingthatmostoftheionizingphotonscome
fromstar-forminggalaxiesbrighterthanM =−13. namic simulations predict fesc to be no more than a few percent
UV
However,suchahigh f isproblematicinthecontextofboth ingalaxiesmoremassivethan109M(cid:12) inhalomassatz>6(e.g.
esc
observationsandtheory.Fromthelocaluniversetoredshiftz∼1, Wiseetal.2014;Kimm&Cen2014;Paardekooper,Khochfar&
Dalla Vecchia 2015; Ma et al. 2015). These simulations include
detailed models of ISM physics, star formation, and stellar feed-
∗ E-mail:[email protected] back,incontrasttoearlygenerationsofsimulationswhichtended
† CanadaResearchChairinAstrophysics. toover-predict fesc byanorderofmagnitude,owingtomoresim-
(cid:13)c 0000RAS
2 X.Maetal.
plisticISMmodels(seeMaetal.2015,andreferencestherein).The
Table1.Simulationsanalyzedinthispaper.
low f inthesesimulationsisduetothefactthatnewlyformed
esc
stars, which dominate the intrinsic ionizing photon budget, begin
life buried in their birth clouds, which absorb most of the ioniz- Name mb (cid:15)b mdm (cid:15)dm Mvir M∗ MUV
ingphotons.Bythetimelowcolumndensityescapechannelsare (M(cid:12)) (pc) (M(cid:12)) (pc) (M(cid:12)) (M(cid:12)) (ABmag)
z5m09 16.8 0.14 81.9 5.6 7.6e8 3.1e5 -10.1
clearedintheISM,themassivestarshavebeguntodieandthepre-
z5m10mr 1.1e3 1.9 5.2e3 14 1.5e10 5.0e7 -17.5
dictedionizingphotonluminosityhasdroppedexponentially.Stel-
z5m11 2.1e3 4.2 1.0e4 14 5.6e10 2.0e8 -18.5
lar populations older than 3Myr have order unity photon escape
fractions,but–accordingtosinglestellarevolutionmodelssuchas Notes.Initialconditionsandgalaxypropertiesatz=6.
STARBURST99(Leithereretal.1999)–thesestarsonlycontribute (1)Name:Simulationdesignation.
asmallfractionoftheintrinsicionizingphotonbudget(Maetal. (2)mb:Initialbaryonicparticlemass.
2015). (3)(cid:15)b:Minimumbaryonicforcesoftening.Forcesofteningisadaptive.
Therefore,thereappearstobeafactorof∼4–5fewerionizing (4)mdm:Darkmatterparticlemassinthehigh-resolutionregions.
photonspredicted,comparedtowhatisneededtoionizetheUni- (5)(cid:15)dm:Minimumdarkmatterforcesoftening.
verse. Several solutions have been proposed. For example, Wise (6)Mvir:Halomassoftheprimarygalaxyatz=6.
et al. (2014) suggested that tiny galaxies that are much fainter (7)M∗:Stellarmassoftheprimarygalaxyatz=6.
thanM =−13mayplayasignificantroleinreionization,since (8)MUV:GalaxyUVmagnitude(absoluteABmagnitudeat1500Å).
UV
f increases quickly from 5% to order unity for halo mass be-
esc
low108.5M .However,othershavenotedthattherequirednum-
(cid:12) whichisrequiredtoexplaintheobserveddifferencesbetweenvari-
beroftinygalaxieswouldimplyahugepopulationofMilkyWay
ousnebularemission-linepropertiesofmetal-poor,youngergalax-
satelliteswhichhavenotbeenobserved(seeBoylan-Kolchin,Bul-
iesatz∼2–3andlocalgalaxies(Steideletal.2014,foramorede-
lock&Garrison-Kimmel2014;Grausetal.2016).Conroy&Krat-
tailedstudyseeStrometal.,inpreparation,Steideletal.,inprepa-
ter (2012) proposed that runaway OB stars can boost f ; how-
esc ration). In Ma et al. (2015), we performed Monte Carlo radiative
everbothKimm&Cen(2014)andMaetal.(2015)showedthat
transfer(MCRT)calculationsonasuiteofcosmologicalhydrody-
inhigh-resolutionsimulationstheseproduceamarginaleffect,in-
namicsimulationsandshowedthatthetime-averaged f isabout
esc
creasing fesc systematically by a factor of only ∼1.2 (far short 5%forgalaxiesofhalomassesfrom109–1011M atz=6using
ofthe(cid:38)4required).Amoreradicalalternativeistoinvokenon- (cid:12)
thesingle-starevolutionmodelsfromSTARBURST99.Importantly,
stellarsourcesforreionization,forexampleAGN(seee.g.Madau
we showed that the results were robust to the resolution of both
&Haardt2015).Thisreliesonrecentobservations(e.g.Giallongo
theradiativetransfercalculationandthehydrodynamics(oncesuf-
etal.2015)suggestingmuchhighernumberdensitiesoffaintAGN
ficientresolutionforconvergencewasreached),tovariationsofthe
athighredshiftthanpreviouslythought(e.g.Hopkins,Richards&
starformationandstellarfeedbackmodel,andeventotheinclusion
Hernquist2007).
oflargepopulationsofrunawaystars.Wewillshowhere,however,
But there are gaps in our understanding of stellar evolution.
thattheinclusionofbinaryevolutioneffectsincreasesthepredicted
Onekeyfactorthatisusuallynotconsideredinstandardstellarpop-
escapefractionssubstantially,reconcilingthemwithconstraintson
ulationmodelsistheeffectofbinaryinteraction.Masstransferbe-
reionization.Wedescribethesimulationandradiativetransfercode
tweenbinarystars,andbinarymergers,caneffectivelyincreasethe
inSection2,presenttheresultsinSection3,andconcludeinSec-
numberofhigh-massstarsatlatertimesafterstarformation.Also,
tion4.
massive,rapidly-rotatingstarsproducedviamasstransferundergo
We adopt a standard flat ΛCDM cosmology with cosmo-
quasi-homogeneousevolutionifthemetallicityissufficientlysub- logicalparametersH =70.2kms−1Mpc−1,Ω =0.728,Ω =
0 Λ m
solar.Thesestarsarehotterandtheirsurfacetemperatureincreases
1−Ω =0.272,Ω =0.0455,σ =0.807andn=0.961,consis-
Λ b 8
as they evolve (e.g. Eldridge & Stanway 2012). All of these can
tentwithobservations(e.g.Hinshawetal.2013;PlanckCollabora-
substantiallyincreasethenumberofionizingphotonsproducedat
tionetal.2014).
latetimes,comparedtowhatisexpectedfromsingle-starevolution
models(e.g.deMinketal.2014).Recently,Stanway,Eldridge&
Becker(2016)pointedoutthattheemissivityofionizingphotons
from high-redshift galaxies, inferred from their UV luminosities, 2 METHOD
wouldbehigherbyafactorof∼1.5usingstellarevolutionmodels
Inthiswork,westudytheeffectofbinaryevolutionon f using
esc
thataccountforbinaryinteraction.Furthermore,binaryevolution
three galaxies from a suite of cosmological zoom-in simulations
doesnotjustproducemoreionizingphotons,butitmayalsosub-
presentedinMaetal.(2015).Thesimulationandradiativetransfer
stantiallyincreasetheescapefractions(Maetal.2015).
areidentical.Weonlyreplacethestellarevolutionmodelusedfor
In this Letter, we explore in more detail the effect of binary
thepost-processingradiativetransfercalculations.Thisislikelyto
interaction on ionizing photon production and escape by repeat-
resultinalowerlimitontheimpactofbinarieson f ,becausewe
esc
ingthecalculationdescribedinMaetal.(2015)usingtheBinary
donotincludetheenhancedradiativefeedbackduetobinariesin
PopulationandSpectralSynthesis(BPASS)modelofstellarpop-
oursimulation.Webrieflyreviewthemethodologyhere,butrefer
ulation evolution1 (Eldridge, Izzard & Tout 2008; Eldridge et al,
toMaetal.(2015)formoredetails.
inpreparation).Thesemodelsarecalibratedtoobservationsoflo-
The simulations are part of the Feedback in Realistic Envi-
calstellarpopulations,andreproducetheobservedmultiplicitydis- ronment project2 (FIRE; Hopkins et al. 2014). They are run us-
tributions (Eldridge, Izzard & Tout 2008). Moreover, the BPASS
ing GIZMO (Hopkins 2015) in P-SPH mode, which adopts a La-
modelpredictsstellarpopulationswithaharderionizingspectrum,
grangianpressure-entropyformulationofthesmoothedparticlehy-
1 http://bpass.auckland.ac.nz 2 http://fire.northwestern.edu
(cid:13)c 0000RAS,MNRAS000,1–6
BinaryStarsCanExplainReionization 3
drodynamics(SPH)equationsthatimprovesthetreatmentoffluid- calculation (Ma et al. 2015). The MCRT code we use is derived
mixing instabilities (Hopkins 2013). Galaxy properties at z=6 fromtheMCRTcodeSEDONA(Kasen,Thomas&Nugent2006),
forthethreesimulationsusedinthiswork(z5m09,z5m10mr,and butfocusesonradiativetransferofhydrogenionizingphotons.The
z5m11) are listed in Table 1. Our simulations span halo masses MCRTmethodissimilartothatdescribedinFumagallietal.(2011,
from 109–1011M at z=6. These galaxies lie on the low-mass 2014).N =3×107photonpacketsareisotropicallyemittedfrom
(cid:12) star
extrapolationsoftheobservedstellarmass–halomassrelationand thelocationofstarparticles,samplingtheirionizingphotonbud-
SFR–stellarmassrelationatz>6(Maetal.2015).Atlowerred- gets.AnotherN =3×107photonpacketsareemittedfromthe
UVB
shifts (where observations exist at these masses), the simulations boundaryofthecomputationaldomaininamannerthatproduces
havebeenshowntoreproduceobservedscalingrelationsandchem- a uniform, isotropic ionizing background with intensity given by
icalabundances(Hopkinsetal.2014;Maetal.2016),properties Faucher-Giguèreetal.(2009).TheMCRTcodeincludesphotoion-
ofgalacticoutflowsandcircum-galacticabsorbers(Muratovetal. ization,collisionalionization,recombination,anddustabsorption
2015; Faucher-Giguère et al. 2015, 2016), and abundances and andusesaniterativemethodtoreachphotoionizationequilibrium.
kinematicsofobserved(local)dwarfsinthismassrange(Oñorbe Thenumbersofphotonpacketsanditerationareselectedtoensure
etal.2015;Chanetal.2015). convergence.
In the simulations, gas follows an ionized-atomic-molecular
cooling curve from 10–1010K, including metallicity-dependent
fine-structureandmolecularcoolingatlowtemperaturesandhigh- 3 RESULTS
temperature metal-line cooling for 11 separately tracked species
(Wiersma, Schaye & Smith 2009). We do not include a primor- We use Q˙ion(t) to represent the ionizing photon production rate
dialchemistrynetworknorconsiderPopIIIstarformation,butap- (numberofionizingphotonsproducedpersecond)and
ipzlaytiaonmsettaatlelisciatyrefldoeotreromfiZne=d 1fo0l−lo4wZi(cid:12)ng.AKtaetza,chWteiminebsetregp,&theHieornn-- Qion(>t)=(cid:90) ∞Q˙ion(t(cid:48))dt(cid:48) (1)
t
quist(1996)andcoolingratesarecomputedfromacompilationof
to represent the number of ionizing photons produced after time
CLOUDY runs,includingauniformbutredshift-dependentphoto-
t.InFigure1,weshowQ˙ (t)(upperpanel),thefractionofion-
ionizing background tabulated in Faucher-Giguère et al. (2009), ion
izing photons emitted after time t, Q (>t)/Q (>0) (middle
and an approximate model of photo-ionizing and photo-electric ion ion
panel),andtheratiobetweenionizingluminosityandtheluminos-
heatingfromlocalsources.Gasself-shieldingisaccountedforwith
ityat1500Å,
alocalJeans-lengthapproximation,whichisconsistentwiththera-
diativetransfercalculationsinFaucher-Giguèreetal.(2010).The (cid:82)912ÅL dλ
on-the-fly calculation of ionization states is consistent with more ξ = 10Å λ (2)
ion
λL (1500Å)
accurate post-processing radiative transfer calculations (Ma et al. λ
2015). (bottompanel),asafunctionofage,ofaninstantaneouslyformed
WefollowthestarformationcriteriainHopkins,Narayanan starclusterofmass106M ,forseveralstellarpopulationmodels
(cid:12)
& Murray (2013) and allow star formation to take place only fromBPASS.WeadoptaKroupa(2002)IMFwithslopesof−1.3
in dense, molecular, and self-gravitating regions with hydrogen from 0.1–0.5 M and −2.35 from 0.5–100M , consistent with
(cid:12) (cid:12)
number density above a threshold nth =50 cm−3. Stars form at thatusedinthesimulation.WeshowtheBPASSmodelatmetal-
100%efficiencyperfree-falltimewhenthegasmeetsthesecrite- licityZ=0.001(Z=0.05Z ,black),thelowestmetallicityavail-
(cid:12)
ria,andthereisnostarformationelsewhere.Becausewerequire ableandtheclosesttooursimulations,forbothsingle-starevolu-
star-forminggastobeself-gravitating,itseffectivedensityiseven tion(dotted)andbinaryevolution(solid)models.Wealsocompare
higherthanthefiducialdensitythresholdweadoptinoursimula- thosewithZ=0.02(Z=Z ,cyan)modelsfromBPASS.Wenote
(cid:12)
tions.Weemphasizetheimportanceofresolvingtheformationand thattheSTARBURST99models(notshown),whicharethedefault
destructionofindividualstar-formingregionsinaccuratelypredict- model in Ma et al. (2015), are nearly identical to the single-star
ing fesc,asstressedalsoinotherstudies(e.g.Kimm&Cen2014; modelfromBPASSatbothmetallicities.
Paardekooper, Khochfar & Dalla Vecchia 2015; Ma et al. 2015). Theionizingphotonproductionratesinthesingle-starandbi-
Simulationsusingunphysicallylownthfailtoresolvethisandtend narymodelsareverysimilarforthefirst3Myr,butstarttodiffer
toover-predict fesc byanorderofmagnitude(seeMaetal.2015, significantlyafter3MyratZ=0.05Z(cid:12),withthebinarymodelpro-
andreferencetherein). ducing an order of magnitude more ionizing photons by 10Myr.
The simulations include several different stellar feedback Also, in the binary model, the production of ionizing photons is
mechanisms, including (1) local and long-range momentum flux more extended. For example, almost 60% (20%) of the ionizing
fromradiativepressure,(2)energy,momentum,massandmetalin- photons are produced after 3Myr (10Myr), while this fraction is
jection from SNe and stellar winds, and (3) photo-ionization and 40%(1%)insingle-starmodel.Theselate-timephotonscanescape
photo-electric heating. We follow Wiersma et al. (2009) and in- moreeasilysooneshouldexpectthemtomakeabigdifferenceon
cludemetalproductionfromType-IISNe,Type-IaSNe,andstel- f (asconfirmedbyMCRTcalculationsbelow).However,atso-
esc
lar winds. Every star particle is treated as a single stellar popu- lar metallicity, these fractions are much lower and the difference
lationwithknownmass,age,andmetallicity,assumingaKroupa betweensingle-starandbinarymodelsislesssignificant.
(2002) initial mass function (IMF) from 0.1–100 M(cid:12). The feed- Toillustratetheeffectsofbinaries,werunourMCRTcodeon
backstrengthsaredirectlytabulatedfromSTARBURST99. themaingalaxyinourz5m10mrsimulation(a∼1010M(cid:12) haloat
Foreverysnapshot,wemapthemaingalaxyontoaCartesian z=6)andcompute f usingbothsingle-starandbinaryBPASS
esc
gridofsidelengthLequaltotwovirialradiiandwithNcellsalong models with Z =0.05Z . The results are presented in Figure 2.
(cid:12)
eachdimension.WechooseN=256forz5m09andz5m10mrand Linesandsymbolsshowtheinstantaneousvalueandtime-averaged
N=300forz5m11,sothatthecellsizel=L/N variesbutisal- valuesover∼100Myr,respectively.Dottedlinesandopensymbols
wayssmallerthan100pc.ThisensuresconvergenceoftheMCRT representthesingle-starmodel,whilesolidlinesandfilledsymbols
(cid:13)c 0000RAS,MNRAS000,1–6
4 X.Maetal.
z
53
7.8 7.4 7.0 6.6 6.2 5.8
52
]
1−
[s 51 10-1
)
t
(
n
˙gQio50 fesc
o
l 49 10-2
48
1 2 3 4 6 8 10 20 30
10-3
0.8
0.4
)
0
> 0.3
0.6
(
n
Qio 0.2
/
>t) 0.4 ξion
(
n
Qio0.2 0.1
0.08
0 0.06
1 2 3 4 6 8 10 20 30
100
10-1 10-1
t) eff
ξ(ion10-2 Binary Z=0.05Z fesc,
Single Z=0.05Zfl 10-2
fl
10-3 Binary Z=Z BPASS binary
fl
Single Z=Z BPASS single
fl
10-4 10-3
1 2 3 4 6 8 10 20 30 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 1.05
Age [Myr] Cosmic time [Gyr]
Figure1.Top:Ionizingphotonproductionrate,Q˙ion(t)asafunctionofage Figure2.Top:Trueionizingphotonescapefraction fesc,inourz5m10mr
fora106 M(cid:12) starcluster,predictedbydifferentstellarevolutionmodels. simulation(a∼1010M(cid:12)haloatz=6)asafunctionofredshift(orcosmic
Middle:Fractionofionizingphotonsemittedaftertimet.Bottom:Ratio time).Linesshowtheinstantaneousvaluesineachsnapshot,symbolsare
ofionizingluminosityto1500Åluminosity,ξion asafunctionofagefor time-averagedin100Myrintervals.Middle:ξion(asFig.1)asafunctionof
thesamestarcluster.Weshowbothsingle-starmodels(dotted)andbinary time.Bottom:Effectiveescapefraction, fesc,eff(Equation3)asafunction
models(solid)atmetallicitiesZ=0.05Z(cid:12)(black)andZ=Z(cid:12)(cyan),re- oftime.Inthesingle-starmodel, fesc(cid:46)5%mostofthetime,insufficient
spectively.Includingbinariesleadstomoremassivestarsatlatetimes(from forreionization.Accountingforbinaryeffectsboostsξionbyafactor∼1.5
masstransferandmergers),whichdramaticallyenhancestheionizingpho- – considerable but insufficient to explain reionization. But it also boosts
tonproductionaftert∼3Myr.About60%(20%)oftheionizingphotons fesc byfactors∼3–6becausetheionizingphotonsproducedlater(after
areemittedafter3Myr(10Myr)inthelow-metallicitybinarymodel,while t(cid:38)3Myr)preferentiallyescape,sothe“effectiveescapefraction” fesc,eff
thisfractionismuchlowerinsingle-starmodelaswellasatsolarmetallic- isincreasedbyfactors∼4–10andreachesthe∼20%valuesneededto
ity.Thedifferencebetweensingle-starandbinarymodelsislesssignificant explainreionization.
atsolarmetallicity.STARBURST99models,whichalsoignorebinaries,are
nearlyidenticaltotheBPASSsingle-starmodelsatbothmetallicities.
Theinstantaneous f ishighlytime-variable,associatedwith
esc
represent the binary model. From top to bottom, the three panels stochastic formation and destruction of individual star-forming
show f (the “true” fraction of ionizing photons that escape the clouds (consistent with several other studies Wise et al. 2014;
esc
galaxyvirialradius),ξ ,andthe“effective”escapefractionfrom Kimm & Cen 2014; Paardekooper, Khochfar & Dalla Vecchia
ion
z=5.5–8.Theeffectiveescapefractionisdefinedas 2015).Forsingle-starmodels, fesc isbelow5%mostofthetime,
becauseyoungstarsareburiedintheirbirthclouds,whichprevent
ξ
fesc,eff= fesc(cid:104)ξ i(cid:105)on , (3) almost all ionizing photons from escaping. Most of the photons
ion single that escape come from stellar populations with age ∼3–10Myr,
whichistheratiooftheescapingionizingfluxto1500Åflux,rel- buttheyonlycontributeaverysmallfractionoftheintrinsicioniz-
ativetowhatwouldbecomputedusingsingle-starmodels. f ingphotonsinsingle-starmodels(Maetal.2015).However,atall
esc,eff
simplyequals f forsingle-starmodels,whileforbinarymodels, times,thebinarymodelpredictssignificantlyhigher(factors∼3–
esc
italsoaccountsforthechangeofξ relativetosingle-starmodels. 6) values for f . We also find that ξ is boosted by a factor of
ion esc ion
(cid:13)c 0000RAS,MNRAS000,1–6
BinaryStarsCanExplainReionization 5
z stellarevolutionmodelsignoringbinaries.Theselater-timephotons
7.8 7.4 7.0 6.6 6.2 5.8
easilyescape,collectivelyincreasingtheescapefractionandion-
z5m09 izingphotonproductionratedramaticallyfromhigh-redshiftlow-
10-1 metallicitygalaxies.
For single-star evolution models, we predict f below 5%
esc
most of the time, less than what is required (∼20%) for cosmic
eff reionization.However,whenaccountingforbinaryeffects, fesccan
fesc, 10-2 beboostedbyfactorsof∼3–6andξioncanbeboostedbyafactorof
1.5.Therefore,the“effective”escapefraction(theratioofescaped
ionizingphotonfluxto1500Åflux)canbeboostedbyfactorsof∼
4–10.Forthemoremassivegalaxiesinoursimulation,thisbrings
10-3
them into good agreement with the values required to ionize the
Universe.
100
z5m11 We emphasize that the most important change relative to
single-star models is not in the absolute photon production rate,
butitstime-dependence,becausephotonsemittedafter3Myrcan
10-1 much more easily escape star-forming complexes once feedback
frommassivestarshasdestroyedthedensebirthcloud.Moreover,
eff
fesc, wtheathtahveeseexrehsauulsttsivaerleyntoesttseednisnitiaveprteovoiouurssstaturdfyor(mMaatioent aaln.d20s1te5l)-
10-2 lar feedback models. For example, increasing the strength of all
feedback (radiation, stellar winds, SNe) per star particle relative
BPASS binary
toourfiducialmodelsimplyleadstoself-regulationatlowerstar
BPASS single
formation rates, giving an identical prediction for f . Likewise,
esc
10-3 increasingthedensitythresholdforstarformation,re-distributing
0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 1.05
Cosmic time [Gyr] ionizingphotonstofewerbutmoreluminousparticles,increasing
theionizingphotonproductionrateusedforfeedbackinthecode,
Figure3.Effectiveescapefractionasafunctiontimeforz5m09andz5m11. allproducesimilarpredictionsfor f .
esc
Likez5m10(Figure2),binarystellarmodelsboost fesc,effbyfactorsof∼
Nevertheless,thebinaryfractioninhigh-redshiftgalaxiesand
4–10.Inmoremassivegalaxies,themean fesc,effreaches∼20%,sufficient
the details of binary evolution are both uncertain, so our results
forreionization.
arenotdefinitive.Theydo,however,demonstratethepotentialfor
binaryevolutiontoreconcileempiricalconstraintsonreionization
∼1.5,consistentwithStanway,Eldridge&Becker(2016).Multi- bystarlightwithobservationsandsimulations.Inprinciple,these
plyingthetwofactors,wefindthattheeffectiveescapefractionis modelscanbeconfrontedbytheobservedGRBratesatthesered-
boostedbyfactorsof∼4–10,withmostofthecontributioncom- shifts(e.g.Kistleretal.2009;Wyitheetal.2010),althoughlarge
ingfromtheincreased f .Averagedovertheentireredshiftrange uncertainties remain. In addition, the BPASS model does not in-
esc
z=5.5–8thatweconsiderhere,accountingforbinaryeffectsin- clude stellar rotation before binary interaction, which may also
creasesthetrueionizingescapefraction f from6%to14%and significantlyincreasetheintrinsicionizingphotonproductionrate
esc
increases f from6%to∼20%.Thisisconsistentwithwhatis (e.g. Topping & Shull 2015). Rotation likely has a smaller effect
esc,eff
requiredinempiricalreionizationmodels. on f because most of the extra ionizing photons it predicts are
esc
InFigure3,weshowtheeffectiveescapefractionasafunction producedlessthan3Myrafterstarformation.
oftimeforz5m09andz5m11galaxies.Similartoz5m10mr(Figure
Wehaverepeatedourradiativetransfercalculationoncosmo-
2),binarymodelsboosttime-averaged fesc,effbyfactorsof∼4–10. logicalsimulationsofMilkyWay-massgalaxiesatz=0(Z∼Z(cid:12))
Inthelowestmasshalo(z5m09), f isstilllow,becausehalo
esc,eff from the FIRE project (see Hopkins et al. 2014, for details). We
gas is largely neutral close to the galaxy in such low-mass sys- findthatbinariesappeartobeenhancing f byonlyafactor∼1.5
esc
tems,whichconsumesalargefractionoftheionizingphotons.In
atsolarmetallicity.Thisisexpectedsincebinaryeffectstendtobe
moremassivegalaxieslikez5m10mrandz5m11, f canreach
esc,eff weakerathighermetallicities(alsoseeFigure1),foratleastthree
(cid:38)20%.
reasons:(1)thenumberofionizingphotonsdecreasessignificantly
asstellaratmospheresarecooler,(2)quasi-homogenousevolution
ceasestoapplyaboveZ=0.004(Z=0.2Z ),and(3)stellarwinds
(cid:12)
become stronger, reducing the lifetime of massive stars and sup-
4 DISCUSSIONANDCONCLUSIONS
pressingthemasstransferedbetweenbinaries.Inaddition,theab-
In this work, we study the effect of binary evolution on ionizing solutetime-averaged f doesnotexceed∼3%inthesegalaxies,
esc
photonproductionandescapeinhigh-redshiftgalaxies,usingthree consistentwithobservationalconstraintsinthelocalUniverse(see
high-resolution cosmological simulations from the FIRE project. referencesinSection1).Itappearsthatthesegalaxiesareforming
Thesimulatedgalaxiesarearoundthemassestimatedtodominate stars in a more “calm”, less-bursty mode compared to the high-
re-ionization(haloM =109–1011M atz=6).Usingdetailed redshift dwarfs here (e.g. Sparre et al. 2015), and maintain much
halo (cid:12)
radiativetransfercalculations,weshowthatrecentstellarevolution largerreservoirsofneutralgasintheirgalacticdisks,whichleadsto
models which account for mass transfer and mergers in binaries muchlargerabsorptionevenofphotonsproducedbyintermediate-
(specifically, the BPASS model) produce significantly more ion- agemassivestars.Adetailedstudywillbepresentedinaseparate
izingphotonsforstellarpopulationsolderthan3Myrcomparedto paper(Suetal.,inpreparation).
(cid:13)c 0000RAS,MNRAS000,1–6
6 X.Maetal.
ACKNOWLEDGMENTS HopkinsP.F.,RichardsG.T.,HernquistL.,2007,ApJ,654,731
IwataI.etal.,2009,ApJ,692,1287
WethankChuckSteidelforhelpfuldiscussionsandtherefereefor
KasenD.,ThomasR.C.,NugentP.,2006,ApJ,651,366
usefulcomments.WealsothankJohnBeacomandMikeShullfor
KatzN.,WeinbergD.H.,HernquistL.,1996,ApJS,105,19
helpful suggestions after the paper was submitted to arXiv. The
KimmT.,CenR.,2014,ApJ,788,121
simulationsusedinthispaperwererunonXSEDEcomputational
KistlerM.D.,YükselH.,BeacomJ.F.,HopkinsA.M.,Wyithe
resources(allocationsTG-AST120025,TG-AST130039,andTG-
J.S.B.,2009,ApJ,705,L104
AST140023).TheanalysiswasperformedontheCaltechcompute
KroupaP.,2002,Science,295,82
cluster “Zwicky” (NSF MRI award #PHY-0960291). Support for
KuhlenM.,Faucher-GiguèreC.-A.,2012,MNRAS,423,862
PFH was provided by an Alfred P. Sloan Research Fellowship,
LeitetE.,BergvallN.,HayesM.,LinnéS.,ZackrissonE.,2013,
NASA ATP Grant NNX14AH35G, and NSF Collaborative Re-
A&A,553,A106
searchGrant#1411920andCAREERgrant#1455342.D.Kasen
Leitet E., Bergvall N., Piskunov N., Andersson B.-G., 2011,
issupportedinpartbyaDepartmentofEnergyOfficeofNuclear
A&A,532,A107
PhysicsEarlyCareerAward,andbytheDirector,OfficeofEnergy
LeithererC.etal.,1999,ApJS,123,3
Research,OfficeofHighEnergyandNuclearPhysics,Divisionsof
MaX.,HopkinsP.F.,Faucher-GiguèreC.-A.,ZolmanN.,Mura-
NuclearPhysics,oftheU.S.DepartmentofEnergyunderContract
tovA.L.,KerešD.,QuataertE.,2016,MNRAS,456,2140
No. DE-AC02-05CH11231 and by the NSF through grant AST-
MaX.,KasenD.,HopkinsP.F.,Faucher-GiguèreC.-A.,Quataert
1109896.D.KerešwassupportedbyNSFgrantAST-1412153and
E.,KerešD.,MurrayN.,2015,MNRAS,453,960
funds from the University of California, San Diego. CAFG was
MadauP.,HaardtF.,2015,ApJ,813,L8
supportedbyNSFthroughgrantsAST-1412836andAST-1517491,
MuratovA.L.,KerešD.,Faucher-GiguèreC.-A.,HopkinsP.F.,
by NASA through grant NNX15AB22G, and by STScI through
QuataertE.,MurrayN.,2015,MNRAS,454,2691
grant HST-AR-14293.001-A. EQ was supported by NASA ATP
OñorbeJ.,Boylan-KolchinM.,BullockJ.S.,HopkinsP.F.,Kereš
grant 12-APT12-0183, a Simons Investigator award from the Si-
D.,Faucher-GiguèreC.-A.,QuataertE.,MurrayN.,2015,MN-
monsFoundation,andtheDavidandLucilePackardFoundation.
RAS,454,2092
Paardekooper J.-P., Khochfar S., Dalla Vecchia C., 2015, MN-
RAS,451,2544
PlanckCollaborationetal.,2014,A&A,571,A16
REFERENCES
RobertsonB.E.,EllisR.S.,FurlanettoS.R.,DunlopJ.S.,2015,
BoutsiaK.etal.,2011,ApJ,736,41 ApJ,802,L19
Boylan-Kolchin M., Bullock J. S., Garrison-Kimmel S., 2014, RobertsonB.E.etal.,2013,ApJ,768,71
MNRAS,443,L44 SianaB.etal.,2015,ApJ,804,17
BridgeC.R.etal.,2010,ApJ,720,465 SianaB.etal.,2010,ApJ,723,241
Chan T. K., Kereš D., Oñorbe J., Hopkins P. F., Muratov A. L., SparreM.,HaywardC.C.,FeldmannR.,Faucher-GiguèreC.-A.,
Faucher-GiguèreC.-A.,QuataertE.,2015,MNRAS,454,2981 MuratovA.L.,KerešD.,HopkinsP.F.,2015,ArXive-prints
ConroyC.,KratterK.M.,2012,ApJ,755,123 StanwayE.R.,EldridgeJ.J.,BeckerG.D.,2016,MNRAS,456,
de Mink S. E., Sana H., Langer N., Izzard R. G., Schneider 485
F.R.N.,2014,ApJ,782,7 SteidelC.C.etal.,2014,ApJ,795,165
EldridgeJ.J.,IzzardR.G.,ToutC.A.,2008,MNRAS,384,1109 ToppingM.W.,ShullJ.M.,2015,ApJ,800,97
EldridgeJ.J.,StanwayE.R.,2012,MNRAS,419,479 WiersmaR.P.C.,SchayeJ.,SmithB.D.,2009,MNRAS,393,99
Faucher-GiguèreC.-A.,FeldmannR.,QuataertE.,KeresD.,Hop- WiersmaR.P.C.,SchayeJ.,TheunsT.,DallaVecchiaC.,Torna-
kinsP.F.,MurrayN.,2016,ArXive-prints toreL.,2009,MNRAS,399,574
Faucher-GiguèreC.-A.,HopkinsP.F.,KerešD.,MuratovA.L., WiseJ.H.,DemchenkoV.G.,HalicekM.T.,NormanM.L.,Turk
QuataertE.,MurrayN.,2015,MNRAS,449,987 M.J.,AbelT.,SmithB.D.,2014,MNRAS,442,2560
Faucher-GiguèreC.-A.,KerešD.,DijkstraM.,HernquistL.,Zal- WyitheJ.S.B.,HopkinsA.M.,KistlerM.D.,YükselH.,Beacom
darriagaM.,2010,ApJ,725,633 J.F.,2010,MNRAS,401,2561
Faucher-Giguère C.-A., Lidz A., Zaldarriaga M., Hernquist L.,
2009,ApJ,703,1416
FinkelsteinS.L.etal.,2012,ApJ,758,93
Fumagalli M., Hennawi J. F., Prochaska J. X., Kasen D., Dekel
A.,CeverinoD.,PrimackJ.,2014,ApJ,780,74
FumagalliM.,ProchaskaJ.X.,KasenD.,DekelA.,CeverinoD.,
PrimackJ.R.,2011,MNRAS,418,1796
GiallongoE.etal.,2015,A&A,578,A83
GrausA.S.,BullockJ.S.,Boylan-KolchinM.,WeiszD.R.,2016,
MNRAS,456,477
HinshawG.etal.,2013,ApJS,208,19
HopkinsP.F.,2013,MNRAS,428,2840
HopkinsP.F.,2015,MNRAS,450,53
Hopkins P. F., Kereš D., Oñorbe J., Faucher-Giguère C.-A.,
QuataertE.,MurrayN.,BullockJ.S.,2014,MNRAS,445,581
Hopkins P. F., Narayanan D., Murray N., 2013, MNRAS, 432,
2647
(cid:13)c 0000RAS,MNRAS000,1–6