,%yielding%a%dustktokstellar%mass%ratio% 10 K6 K10 K 5 %and%a%dustktokgas%mass%ratio%3.2k13 10 K6.%If%I%Zw%18%is%a%reasonable%analog%of%

Transcription

1 Dust%May%Be%More%Rare%Than%Expected%in%Metal%Poor%Galaxies% DavidB.Fisher 1,2,AlbertoD.Bolatto 1,RodrigoHerrera9Camus 1,BruceT. Draine 3,JessicaDonaldson 1,FabianWalter 4,KarinM.Sandstrom 4,Adam K.Leroy 5,JohnCannon 6,andKarlGordon 7 % Affiliations%% % 1 DepartmentofAstronomy,LaboratoryforMillimeter9waveAstronomyandJoint SpaceInstitute,UniversityofMaryland,CollegePark,MD20742,USA 2 CentreforAstrophysicsandSupercomputing,SwinburneUniversity,POBox218, Hawthorn,Victoria3122Australia 3 DepartmentofAstrophysicalSciences,PrincetonUniversity,Princeton,NJ08544, USA 4 Max9PlankInstitutfürAstronomie,Königstuhl17D969117,Heidelberg,Germany 5 NationalRadioAstronomyObservatory,Charlottesville,VA22903,USA 6 DepartmentofPhysicsandAstronomy,MacalesterCollege,1600GrandAvenue, SaintPaul,MN55105,USA 7 SpaceTelescopeScienceInstitute,MD21218,USA % `Normal %galaxies%observed%at%z>6,%when%the%universe%was%<1%billion%years%old,% thus%far%show%no%evidence 1,2,3 %of%the%cold%dust%that%accompanies%star%formation% the% local% Universe,% where% the% dustktokgas% mass% ratio% is% ~1%.% A% prototypical% example% is% 'Himiko'% (z=6.6),% which% a% mere% 840% Myr% after% the% Big% Bang% is% forming% stars% at% a% rate% of% 30K100% M % yr K1,% yielding% a% mass% assembly% time% M*/SFR~ %yr.%himiko%is%estimated%to%have%a%low%fraction%(2k3%%of%the% Solar% value)% of% elements% heavier% than% helium% (metallicity),% and% although% its% gas% mass% cannot% be% asserted% at% this% time% its% dustktokstellar% mass% ratio% is% constrained 3 %to%be%<0.05%.%the%local%galaxy%i% Zw%18,%with% a%metallicity% ~4%% solar 4 % and% forming% stars% less% rapidly% than% Himiko% but% still% vigorously% for% its% mass%(m*/sfr~ %yr) 5,%is%also%very%dust%deficient%and%perhaps%one%of%the% best% analogues% of% primitive% galaxies% accessible% to% detailed% study.% Here% we% report%observations%of%dust%emission%from%i%zw%18%from%which%we%determine% its%dust%mass%to%be%450k1800%m,%yielding%a%dustktokstellar%mass%ratio% 10 K6 K10 K 5 %and%a%dustktokgas%mass%ratio%3.2k13 10 K6.%If%I%Zw%18%is%a%reasonable%analog%of% Himiko,% then% Himiko's% dust% mass% is% 50,000% M,% a% factor% of% 100% below% the% current% upper% limit.% These% numbers% are% considerably% uncertain,% but% if% most% highkz% galaxies% are% more% like% Himiko% than% like% the% quasar% host% SDSS%

2 J ,% then% the% prospects% for% detecting% the% gas% and% dust% in% them%are%much%poorer%than%hitherto%anticipated.% Therecentstudy 7 ofhfls3,a maximumstarburst atz=6.3,providesanexampleof agalaxywithalargeamountofdust(mdust 10 9 M ),andadust9to9gasmassratio (D/G 0.01)anddust9to9starmassratio(D/M* 0.04)morelikenearbystarbursting galaxies.thisgalaxyhasanastonishingstarformationrateof 3,000M yr 91,and converts its gas into stars at rates 2,000 times that of typical galaxies, properties thatarerareevenforthegasrichhigh9zgalaxies.frequentlyobservationsofdust and molecules in high9redshift galaxies tend to target those with bright active galaxies 8,likeHFLS3orJ Massivegalaxieslikethesearewell knowntoberareatallredshifts.forthose"normal"galaxieswheredeepsub9mm observations have been performed, currently only upper limits for both [CII] and sub9millimeter continuum exits 1,2. These observational limits suggest that in the first800myrofgalaxyevolution,galaxieswithverylittledustandlowmetallicity, likehimiko,aremoretypical.resultsfromstellarpopulationsanalysesindeedshow that high redshift galaxies have very little evidence of dust extinction 9. Understandingofthephysicalconditionsunderwhichstarsformintheseprimitive systems,however,canonlycomefromthestudyoflocalanalogues. Locatedatadistanceof18Mpc 10,IZw18isthearchetypalstar9formingverylow metallicity 4 galaxy (12+log(O/H)=7.17, or 1/30 Solar metallicity). I Zw 18 is gas rich 11,12,13 (MHI= M and MH M ) and actively star forming 5 (SFR=0.05±0.02M yr 91 ).Comparedtoitsstellarmass,M* M,thisgalaxy hasaveryhighgasfraction(mgas/mstar+gas 2/3).IZw18iscurrentlyundergoinga starburst phase 14, and despite its active star formation there is no detected CO emissionindicativeofmoleculargas 12 inizw18.thelowestmetallicitydetectionof CO was recently reported in the dwarf galaxy WLM 15 (12+log(O/H) 7.8 or 1/8 Solar).UnlikeWLM,IZw18hasaveryactivestar9formingenvironmentwhichlikely photodissociatesco.thesepropertiesmeanthatitisamongtheclosestanalogues to primitive high9redshift galaxies, although I Zw 18 contains a more significant populationofevolvedstarsthanmaybefoundintheearlyuniverse. NotethatthedifferenceinstellarmassbetweenIZw18(oranylocallowmetallicity galaxy)andobservedhigh9redshiftgalaxiesislarge.becauseofthesmallerpotential well,identicalstarburstscaninprinciplemoreeasilydrivedustandmetal9enriched gasoutizw18thanoutofhimikoforexample.nonetheless,izw18andgalaxies like it remain our best candidates for the study of metal9poor, starbursting environments. UsingHerschelPACS,wemeasurethefluxofIZw18tobe21.1mJyat100μm,with anuncertaintyinflux(calculatedbyplacingaperturesrandomlyinthemap)of±2.9 mjy (signal9to9noise, S/N~7) and a calibration uncertainty of 10% (±2.1 mjy). At 160μmwemeasureafluxof5.6mJy,andafluxuncertaintyinthemapof±1.3mJy (S/N~4)andcalibrationuncertaintyof±0.6mJy.(MapsareshowninFig.1andour

3 procedure isdiscussed in the Supplementary Information.) Together with these detections, we use a number of ancillary data sources to construct a full infrared SED for modeling the dust and star formation. We employ data from Spitzer covering 3.6, 4.5, 5.8, 8.0, 24 and 70 μm 16,12 as well as an IRS spectrum 14 (see SupplementaryInformation). The dust mass we determine from models (shown in Figure 2) with mixed dust graintemperaturesismdust= M (seesupplementaryinformationfora complete discussion of the uncertainties).% % Modified black body models are also commonly used for fitting dust SEDs, although they yield unrealistically low dust masses because of the assumption of a single temperature. For comparison, a modified blackbody model with ν 1.5 Bν(Tdust) and mass9emissivity κ200 = 6.37 cm 2 g 1,reproducesthefluxat70and160μmforTdust=70KandMdust 250M.We findthatasignificantmassofcolddustwitht>15kwouldbedetectedbyour160 μm flux(see Supplementary Information). Independent of the assumed model, the dustmassnecessarytoexplainthesedobservedinizw18isextremelylow. Under the simplest assumptions decreasing the amount of heavy elements, which constitute the dust particles, results in a proportional decrease in the dust9to9gas ratio. This scenario is frequently assumed in cosmological models of star formation 17.ObservationalconstraintsforthisrelationshipintheearlyUniverseare scarce. I Zw 18 provides a probe of such environments, and our measurements directlyconstraintherelationshipbetweendust9to9gasratioandmetallicityatvery low metallicities 12,18,19 (12+log(O/H) 8). We find that I Zw 18 falls roughly two orders of magnitude below the linear correlation between metallicity and dust9to9 gas ratio (Figure 3). The distance to the linear relation is very significant, approximatelyfourtimeslargerthanthespreadinthedata,andmuchlargerthan theerrorbarsonthemeasurement.usingdust9to9gasratiosmeasuredonlyintheir emitting region typically results in a linear relationship even at low metallicity 18, 12+log(O/H)~8,butnotinthecaseofIZw18whereDGRlocalisstillafactorof38 belowthelinearrelationship. IZw18standsoutinthelocalgalaxypopulationbecauseithasanenvironmentthat is both star bursting and also lacking heavy elements. Our results suggest that in starburstinggalaxieswithverylowmetallicitiesthedust9to9gasratioisdetermined by more than just the availability of heavy elements. The essentially dust9free characterofnascentgalaxies 1,2,3,likeHimiko,thereforelikelyreflectsacombination of low metallicity and the balance of the dust production and destruction mechanisminastarburstenvironment,whichacttogethertokeepthedust9to9gas ratioverylowinthesegalaxies. Thesearepropertiesthatarecommonatveryhighredshift(z>6),andconsequently we expect those primitive galaxies to exhibit very low dust masses compared to theirstarformationratesandstellarmasses.theratioofmdust9to9sfr(figure4)ini

4 Zw18ismorethantwoordersofmagnitudelowerthantypicalinlocalgalaxies,and an order of magnitude lower than observed in the z=293 starbursts. Even when normalizedbyitsstellarmassorsfr,thedustmassofizw18isextremelysmall comparedtobothlocalandz=293galaxies.infigure4weshowthatthedifference betweeninterpretingthedustmassofhimikousingtheupperlimitonthe1.2mm flux 3, or using I Zw 18 as an analogue has very significant impact on the inferred properties.withconsiderableuncertainty,wecanscalethestellarmassofhimiko withthedust9to9stellarmassratioofizw18,andplaceitonfig.4withadustmass of 50,000M.Ifthedusttemperatureis40K 3,wecalculateHimikowouldhavea fluxdensityof0.5μjyatearthat260ghz,whichwouldrequireseveraltensofdays ofintegrationwiththecompleteatacamalargemillimeterarray(alma)todetect it.maximumstarburstslikehfls3areveryrare,evenintheearlyuniverse 7 ( 1 Gpc 93 ),whereasbluedust9poor drop9out galaxiesaremuchmorecommonathigh redshifts 9 (10 93 Mpc 93 ). This implies that the ISM of I Zw 18 may indeed be representativeoftheprimitivegalaxypopulationintheearlyuniverse.ifthisisthe case the prospects for detecting dust emission at z>6 will likely be limited to the unusualevolvedsourceslikehfls3andj Walter,F.etal. EvidenceforLowExtinctioninActivelyStar9formingGalaxiesat z>6.5 Astrophys.J.752,93998(2012) 2. Kanekar,N.;Wagg,J.;RamC.R.;Carilli,C.L. ASearchforCII158μmLine EmissioninHCM6A,aLyαEmitteratz=6.56 Astrophys.J.,771,L20925(2013) 3. Ouchi,M.etal. AnIntenselyStar9FormingGalaxyatz~7withLowDustand MetalContentRevealedbyDeepALMAandHSTObservations arxiv (2013) 4. Skillman,EvanD.,Kennicutt,RobertC.,Jr. Spatiallyresolvedopticalandnear9 infraredspectroscopyofizw18 TheAstrophys.J.411, (1993) 5. Cannon,J.M.,Skillman,E.D.,Garnett,D.R.,Dufour,R.J. DustinIZw18from HubbleSpaceTelescopeNarrowbandImaging Astrophys.J,565, (2002) 6. Walter,F.etal. Moleculargasinthehostgalaxyofaquasaratredshiftz=6.42 Nature,424, (2003) 7. Riechers,D.A.etal. Adust9obscuredmassivemaximum9starburstgalaxyata redshiftof6.34 Nature496, (2013) 8. Vieira.J.D.etal. DustystarburstgalaxiesintheearlyUniverseasrevealedby gravitationallensing Nature495, (2013) 9. Bouwens,R.J.etal. UV9ContinuumSlopesof>4000z~498Galaxiesfromthe HUDF/XDF,HUDF09,ERS,CANDELS9South,andCANDELS9NorthFields arxiv (2013) 10. Aloisi,A.,etal. IZw18RevisitedwithHSTACSandCepheids:NewDistanceand Age Astrophys.J.,667,.L1519L154(2007)

5 11. vanzee,l.,westpfahl,d.,haynes,m.p.,salzer,j.j. TheComplexKinematicsof theneutralhydrogenassociatedwithizw18 Astron.J115, (1998) 12. Herrera9Camus,R,etal. Dust9to9gasRatiointheExtremelyMetal9poorGalaxyI Zw18 TheAstrophys.J.752, (2012) 13. Lelli,F.;Verheijen,M.;Fraternali,F.;Sancisi,R. Dynamicsofstarburstingdwarf galaxies:izw18 Astron.&Astrophys.537,13926(2012) 14. Wu,Y.etal. DustintheExtremelyMetal9PoorBlueCompactDwarfGalaxyIZw 18:TheSpitzerMid9infraredView TheAstrophys.J662, (2007) 15. Elmegreen,B.G.,Rubio,M.,Hunter,D.A.,Verdugo,C.,Brinks,E.,Schruba,A. Carbonmonoxideincloudsatlowmetallicityinthedwarfirregulargalaxy WLM Nature,495, (2013) 16. Engelbracht,C.W.,etal. MetallicityEffectsonDustPropertiesinStarbursting Galaxies TheAstrophysJ678, (2008) 17. Gnedin,NickolayY.,Kravtsov,AndreyV. EnvironmentalDependenceofthe Kennicutt9SchmidtRelationinGalaxies TheAstrophys.J.,728,889108(2011) 18. Draine,B.T.,Li,A. InfraredEmissionfromInterstellarDust.IV.TheSilicate9 Graphite9PAHModelinthePost9SpitzerEra TheAstrophysJ.657, (2007) 19. Galametz,M.,Madden,S.C.,Galliano,F.,Hony,S.,Bendo,G.J.,Sauvage,M Probingthedustpropertiesofgalaxiesuptosubmillimetrewavelengths.II. Dust9to9gasmassratiotrendswithmetallicityandthesubmmexcessindwarf galaxies Astron.&Astrophys.,532,56974(2011) 20. Leroy,A.,Bolatto,A.,Stanimirovic,S.,Mizuno,N.,Israel,F.,Bot,C. TheSpitzer SurveyoftheSmallMagellanicCloud:Far9InfraredEmissionandColdGasinthe SmallMagellanicCloud TheAstrophys.J.,658, (2007) 21. Draine,B.T.etal. DustMasses,PAHAbundances,andStarlightIntensitiesinthe SINGSGalaxySample 663, (2007) 22. Kennicutt,R.C.,Jr.etal. Dust9correctedStarFormationRatesofGalaxies.I. CombinationsofHαandInfraredTracers TheAstrophysJ.703, (2009) 23. Skibba,Retal. TheEmissionbyDustandStarsofNearbyGalaxiesinthe HerschelKINGFISHSurvey 738,899(2011) 24. Fisher,D.B.,Bolatto,A.,Drory,N.,Combes,F.,Blitz,L.,Wong,T. TheMolecular GasDensityinGalaxyCentersandhowitConnectstoBulges TheAstrophys.J. 764, (2013) 25. Magnelli,B.,etal. AHerschelviewofthefar9infraredpropertiesof submillimetregalaxies Astron.&Astrophys.,539, (2012) Supplementary%Informationislinkedtotheonlineversionofthepaperat % Acknowledgements% DBF, ADB, and RHC acknowledge support from University of MarylandandtheLaboratoryforMillimeterAstronomyandNSF9AST ADB acknowledges partial support from a CAREER grant NSF9AST , NSF9 AST , and from a Research Corporation for Science Advancement Cottrell Scholaraward.BTDacknowledgespartialsupportfromNSF9AST J.M.C.is

6 supported by NSF grant AST KMS acknowledges support from a Marie CurieInternationalIncomingfellowship.PACShasbeendevelopedbyaconsortium of institutes led by MPE(Germany) and including UVIE(Austria); KU Leuven, CSL, IMEC (Belgium); CEA, LAM (France); MPIA (Germany); INAF9IFSI/OAA/OAP/OAT, LENS, SISSA (Italy); IAC (Spain). This development has been supported by the funding agencies BMVIT (Austria), ESA9PRODEX (Belgium), CEA/CNES (France), DLR(Germany),ASI/INAF(Italy),andCICYT/MCYT(Spain).Wearethankfultoboth therefereesandeditorsforhelpfulcommentsonthismanuscript. % Author%Contributions:%DBFandADBwrotethetextofboththeproposalandthis manuscript.dbfandrhcperformeddetailedcalculations.jdreducedtheherschel data.btdmodeledthesed.aklandfwobtainedandreducedcoobservations.jc obtainedthehαfluxforizw18.allauthorsparticipatedindiscussionofresultsand helpedwithrevisionofthemanuscript. Author%InformationCorrespondenceandrequestsformaterialsshouldbe addressedtodbf(dfisher@swin.edu.au)orab(bolatto@astro.umd.edu). FIG.1. The%100%μm%and%160%μm%images%of%I%Zw%18.Inthisfigureweshowthe first λ = 100 μm far9infrared (FIR) detection of dust emission in I Zw 18, and the marginal detection of I Zw 18 at 160 μm. These new observations (PID: OT_dbfisher_1) were obtained with Herschel PACS. White contours show N(HI) = 0.7,1.4and cm 2 fromtheverylargearray(vla)map 11.Thebeamsizeof the HI map is arcsec. For display purposes, the pixel size of the infrared mapsisresampledtomatchthepixelsizeofthehimap.at100μmizw18isclearly detected, and well matched to the center of the HI gas contours. The emission we detectinbothfarinfraredfiltersiscontainedwithinasmallregion(15 or1.3kpc). We note that the off target peaks in the 160 μm map are not noise, they are all coincidentwithpeaksinthe100μmmap,andarethereforemostlikelybackground targets.at160μmwedetectemissionatthe3σlevelthatisconsistentwiththepeak ofbothhiandtheiremissionat24,70and100μm,andwhichweattributetoizw

7 18.Thepeakandextentoftheemissioninourimagesiscoincidentwiththatofthe HαemissionfromHSTimages 5,andalsothepeakofHIemission Draine & Li (2007) Cold Dust Component DL07 + Cold Flux (mjy) m FIG.2. The%far%infrared%spectral%energy%distribution%of%I%Zw%18.Bluesquares representfluxesmeasuredbyiracandmipsonboardspitzer 16,12@.@Thebluelineisa smoothedirsspectrum 14.TheopentrianglesareupperlimitsfromHerschelSPIRE. The red circles show our new Herschel PACS data at 100 and 160 μm. Error bars represent 1σ uncertainties. We fit the SED (spectral energy distribution) with modelsbasedonamixtureofcarbonaceousandsilicatedustgrains 18,whichassume a distribution of grain sizes matching that of the Milky Way. This is a commonly adopted dust model, allowing direct comparison to dust masses in the literature, andisnotexpectedtointroducelargeerrorsinthedustmassestimate 18.Mostof the modeled dust is heated by a single starlight intensity (Umin), but a fraction is heatedbyapower9lawdistributionofintensities,withanadjustableslope(α)and uppercut9off(umax).thebest9fitdustmodeltothesedofizw18,showninfigure 2,returnsthefollowingvalues:α=2.4,Umin=100,and<U> 200,forUmax=10 7 and no polycyclic aromatic hydrocarbons, with U in units of the radiation field in the vicinityofthesun.thesolidgreylinerepresentsthebest9fitdustmodel.thedotted linerepresentsacold(t=20k)dustmodelwithamassof1000m andκthatis proportionatetoν 1.5.Thedashedlinerepresentsthelinearcombinationofthecold componentandthedustmodel.wefindthatincludingthecoldcomponentofdust increases the discrepancy between the model and the 160 μm flux to 2.3σ. Our assumptionofafactoroftwouncertaintyonthedustmass,therefore,accountsfor thepossibilityofthecoldcomponent.

8 SMC WLM Local Galaxies I Zw 18 (open = IR area) log(o/h) FIG. 3. The dustktokgas% ratio% of% galaxies% compared% to% metallicity% for% local% galaxies% and% I% Zw% 18.TheLocalGalaxysampleconsistsofdust9to9gasratiosfrom two recent papers 21,19 the sample is representative of typical disk and dwarf galaxiesinthelocaluniverse.forsomegalaxiesinthelocalsampleeitherhiorh2 arenotavailableintheliterature.inthiscaseweestimatethetotalgasmasswithan empiricalcorrelationrelatingmoleculartoatomicgas,m(h2)@=@0.008@m(hi) 1.2.Those galaxies are plotted as open circles. We also include two nearby, well9known low metallicitygalaxieswlm 15 (purplecircle)andthesmc 20 (bluecircle).theerrorbars in the upper left corner represent the median 2σ for the Local galaxy sample. A common assumption is that dust9to9gas ratio in galaxies scales linearly with metallicity 17.Wethereforeshowalinearbisectinglinesettomatchthelocalgalaxy sample, the dashed lines represent the ±2σ root9mean9square scatter around this bisector.izw18isaclearoutlierfromthiscorrelation.notethedistinctionbetween bothsmcandwlmfromizw18isthestarburstingnatureofizw18.ifizw18is representative of starbursting low9metallicity environments, this implies that the dustmassislowerthanwemayexpectinprimitivegalaxiesoftheearlyuniverse.

9 log(m [M ]) dust (a) z=2-3 Starbursts Local Galaxies (b) I Zw 18 Himiko HFLS log(sfr [M yr ]) log(m stars[m ]) FIG.4. Dust%mass%versus%star%formation%rate%and%stellar%mass%for%local%disks,% highkredshift% star% bursts% and% I% Zw% 18.% Here we compare dust masses to star formation rates(a) and stellar masses(b) of a sample of normal and starbursting galaxies 21,19,22,23,24, including I Zw 18. The definitions of symbols are the same in both panels. The red diamond represents I Zw 18. Squares represent 1<z<3 starbursts 25.Thebluetrianglerepresentsthecurrentupper9limitonthedustmass ofhimiko,thebluediamondshowsthedustmassofhimikoifithasasimilardust9 to9stellarmassratioasizw18.solidlinesrepresentslinearrelationshipsmatching the local galaxies, the dashed lines represent 1/100 (short dashes) and 1/1000 (long dashes) of the local sample. The error bars in both panels represent the median 2σ error bar for the sample. Note that very significant differences exist between I Zw 18 and the other starbursting systems shown here; the distant starburstshavestarsformationratesfiveordersofmagnitudehigherthanizw18, andmanyofthemhavesolarmetallicity 25.IZw18isseentobeaclear,extreme, outlier toward lower ratios of dust9to9sfr and dust9to9gas when compared to typical nearby galaxies, and the dust mass is even lower per unit SFR than star burstinggalaxies.asindicatedbytheextremedifferenceinthecurrentupperlimit ofhimiko,andthepredicteddustmassusingthedust9to9starmassratioofizw18, those observations of the highest redshift galaxies may be significantly over estimatingthedustmass.

10 Supplementary%Information% % Flux@Measurements:ThedataarereducedinHIPE usingstandardmethods.to measure the flux we choose a circular region centered on the peak of the 100 μm image.theradiusischosentocorrespondtothepointatwhichthe100μmsurface brightnessisequivalenttothebackgroundnoise(15 ).Thebeamsizeofourmapsis 8 and12 at100 μm and160 μm respectively. The background is removed by fittinga2dimensionalplanetothemap.weapplyanaperturecorrection,according to the Herschel documentation, of and to the 100 and 160 μm photometryrespectively. ThemeasurementuncertaintyontheHerschelfluxesisacombinationofanassumed zero point uncertainty and also the uncertainty introduced from the background noise.basedonourowncomparisonoffluxesfrompacsandmips,weassumean estimateof10%forthecalibrationuncertainty.tomeasuretheuncertaintyinthe maps we generate a set of apertures equal in size to our flux measurement (15 ) thatarerandomlyplacedtofillthemap,excludingtheregionweidentifyasizw18. Wefindthat,onaverage,theexpectedfluxofa15 apertureis2.9mjyinthe100μm map and 1.26 mjy in the 160 μm map. We also note that there is a source of emissionthatisverynearizw18inboththe100and160μmmaps.wecannot knowforcertainifthisispartoftheizw18ornot,sowecomputethefluxinaflux of24 inradius,whichincludestheextraemission.wefindthatthisintroducesan uncertaintyof2.7mjy@at100μmand1.6mjyat160μm.thisiscomparabletothe uncertaintyfromrandomlyplacedapertures,andthereforeisaccountedforinour error bars. In our maps the calibration and measurement uncertainties contribute comparable amounts to the total uncertainty. The total uncertainty is therefore computedbyaddinginquadraturethecalibrationandimageuncertainties.

11 We use Hα fluxes obtained from HST mapping 5. The Hα flux is measured in a continuumsubtractedimage,andintegratedoveraregionthatiscollocatedwiththe IRemittingregion.FormoredetailsabouttheHαwedirectthereadertoitsoriginal publication 5.Weconvertthisfluxtoastarformationratewitharecentcalibration 27. We use VLA mapping 11 to measure the HI mass of the galaxy. The CO(190) measurementsofizw18returnanupperlimit 12 ofthemoleculargasmassofmh M, using a galactic conversion factor for CO flux9to9molecular gas mass. Alternatively, we can estimate the molecular gas mass by assuming that I Zw 18 followsasimilarstarformationlawasothergalaxies 28,suchthatMmol ( ) SFR, and thus Mmol 10 M. Indeed when we adopt the metallicity dependent conversionfactor 29 theupperlimittothemoleculargasmassismh2@ M.In thispaperweuse M asthemoleculargasmassofizw18.wenotethatthis amountstoonly30%ofthetotalgasmass,andhas,therefore,averysmalleffecton ourmainresult(infigure3). Wefindthatthefluxinthe4.5μmisheavilydominatedbythestarlightemission andnotbybr@αlineemission.wedeterminethisbyscalingthehαfluxby (10 4 KCaseB 30 ).WefindthattheBrαlineaccountsforroughly 2%ofthefluxin IRACband2. Fitting@the@Dust@Model@to@the@SED:@InthefitstheIRAC,MIPSandPACSdatapointsare weightedbyafactorof5andtheirsdatapointsaregivenaweightof1;thisisdone to produce and even weighting across all wavelength ranges. Otherwise the software would only seek to fit the IRS data. Also, all points are weighted by the inverseoftheuncertaintysquared.infigure2thereisaslightdifferencebetween the flux at 8μm using either IRAC or IRS data. One simple explanation is that this discrepancyisduetoimperfectionsintheiracchannel3and4zeropoints.inlight ofthis,wedonotincludethesechannelsinthefit.insteadthefitisconstrainedby theirsdataaloneoverthesewavelengths.inspectionofpublishedirsedsofother galaxies shows that these discrepancies are not extremely uncommon. For completeness,wealsofitthefullsed,includingbothiracbands3&4andallother data. We find that this does not significantly affect the dust mass, Mdust(all data) = 881M. We set qpah=0 in the model. This is consistent with previous observations of low metallicitygalaxies.wenotethatthepahmassintheqpah=0modelisnotexactly zero, rather the size distribution of carbonaceous grains is set to go to zero smoothly, and therefore has a very small amount of material in the range of sizes thatcorrespondtopahgrains.(itisforthisreasonthatafewpahlinesarepresent inourdustmodelsed.) The dust mass uncertainty comes from three sources: errors in the flux measurements, the scatter in the fit values, and the systematic uncertainty in the model.weestimatetheuncertaintyonthedustmassfromthefluxmeasurements by combining the error bars shown in Fig. 2. The uncertainty is dominated by the errorinthe160μmmeasurement.weran20independentrealizationsofthemodel,

12 withgaussiannoiseadded,andtheresultingscatterindustmassis 20%.Including sub9mmfluxes(λ>160μm)inlowmetallicitydwarfgalaxies,andincludingacold dust population in models, systematically increases the dust mass 19 by roughly a factor of 1.5. In Fig. 2 open triangles represent upper limits measured from Herschel. Only upper limits exist for the sub9mm flux of I Zw 18. Conversely, assumingamilkywaydustdistributionmayoverestimatethedustmass 31 inalow metallicity environment by a factor of 2. For physical dust models, differences in assumptions about grain size distributions, compositions and mass emissivities introduceuncertaintyinthedustmassatthefactoroftwolevel 18.Toaccountfor thesepossiblesystematicuncertaintiesweassumethatthedustmassisuncertain byafactorof2. InFigure2weshowtheeffectofincludingadustcomponentwithT=20Kandmass Mcold=1,000M.Thefitteddustmodelisalreadyhigherthanthefluxat160μmby 1.3σ, and including this cold dust component increases the discrepancy to 2.3σ. Nonetheless,amuchcolderdustcomponentwouldbeimpossibletorejectwiththe existing data, and in fact a T=10 K component of Mcold=10,000 M would be undetectable in our observations. As stated above, such a component would likely be outside of the IR emitting region, given the high radiation fields. Recent evidence 32 fromhst/cosmicoriginsspectrographsuggeststhatinthehienvelope the metallicity is likely even lower than that in the center of the galaxy, with pockets of pristine gas, free of any metals. Such observations suggest that it is unlikelythatsignificantamountsofcolddustexistinthehienvelopesurrounding theiremittingregion. ThelowercutofffortheradiationfieldinIZw18isUmin=100.Thishighvalue,five timeslargerthanthelargestumintypicallyfoundinnearbygalaxies 18,suggestsan absence of quiescent regions in I Zw 18 over a region that is roughly 3 kpc in diameter. Typical values of <U> are in samples of nearby galaxies 18. The average dust grain in I Zw 18 experiences a radiation field that is many times stronger than in a typical galaxy, and even more extreme than in distant starbursts 25. IRS@ The coverage of the IRS spectrum is a slit that runs through the centerofirregion 13.Theslitisonly~4"wideandthereforedoesnotobserveallthe extended,lowsurfacebrightnessemission.nonetheless,thesliteasilyincludesthe brightestpartsofthehαmap.@ AllmetallicitiesaredisplayedinthePilyugin&Thuan(2005)calibration Ott,S.AstronomySocietyofthePacificConferenceSeriesVol434Astronomical anddataanalysissoftwareandsystemsxixed.y.mizumoto,k.i.morita,&m. Ohishi (2010)

13 27. Hao,Cai9Na,Kennicutt,R.C.,Johnson,B.D.,Calzetti,D.,Dale,D.A.,Moustakas,J. Dust9correctedStarFormationRatesofGalaxies.II.CombinationsofUltraviolet andinfraredtracers TheAstrophys.J.741, (2011) 28. Bigiel,F.etal. AConstantMolecularGasDepletionTimeinNearbyDisk Galaxies TheAstrophys.J.730,L13917(2011) 29. Leroy,A.K.etal. TheCO9to9H2ConversionFactorfromInfraredDustEmission acrossthelocalgroup TheAstrophys.J.,737,12925(2011) 30. Lebouteiller,V.;Heap,S.;Hubeny,I.;Kunth,D. Chemicalenrichmentand physicalconditionsinizw18 Astron.&Astrophys.553,32948(2013) 31. Draine,B.T.,Salpeter,E.E. Destructionmechanismsforinterstellardust The Astrophys.J.231, (1979) 32. Brown,T.M.,Heap,S.R.,Hubeny,I.,Lanz,T.,Lindler,D. IsolatingClusterswith Wolf9RayetStarsinIZw18 Astrophys.J.,579,L759L78(2002) 33. Pilyugin,LeonidS.;Thuan,T.X. OxygenAbundanceDeterminationinHII Regions:TheStrongLineIntensities9AbundanceCalibrationRevisited The Astrophys.J.631, (2005)