Astronomy&Astrophysicsmanuscriptno.6067 (cid:13)c ESO2008 February5,2008 Balmer jump temperature determination in a large sample of  ⋆,⋆⋆ low-metallicity H regions N.G.Guseva1,Y.I.Izotov1,P.Papaderos2,3,andK.J.Fricke3 1 MainAstronomicalObservatory,UkrainianNationalAcademyofSciences,Zabolotnoho27,Kyiv03680,Ukraine 2 InstitutodeAstrof´ısicadeAndaluc´ıa,Apdo.3004,18080Granada,Spain 3 InstituteforAstrophysics,Friedrich-Hund-Platz1,37077Go¨ttingen,Germany 7 0 Received ;Accepted 0 2 ABSTRACT n a Aims.ContinuingthesystematicdeterminationoftheelectrontemperatureofHregionsusingtheBalmerand/orPaschendisconti- J nuitiesbyGusevaetal.(2006)wefocushereon3.6mESOtelescopeobservationsofalargenewsampleof69Hregionsin45blue 2 compactdwarf(BCD)galaxies.Thisdatasetspansawiderangeinmetallicity(Z⊙/60<∼Z <∼Z⊙/3)and,combinedwiththesample of47HregionsfromGusevaetal.(2006),yieldsthelargestspectroscopicdataseteverusedtoderivetheelectrontemperaturein theH+zone. 1 Methods.InthesamewayasinGusevaetal.(2006)wehaveusedaMonteCarlotechniquetovaryfreeparametersandtocalculatea v seriesofmodelspectralenergydistributions(SEDs)foreachHregion.TheelectrontemperatureintheH+zoneswasderivedfrom 2 thebestfittingsyntheticandobservedSEDsinthewavelengthrange∼3200–5100Å,whichincludestheBalmerjump. 3 0 Results.Onthebaseofthepresentlargespectroscopicsamplewefindthatinhot(Te(H+)>∼11000K)Hregionsthetemperature of theO2+ zone, determined fromdoubly ionised oxygen forbidden lines, does not differ statisticallyfrom the temperature of the 1 H+ zone. Thus, weconfirmand strengthen thefinding by Guseva et al. (2006). Weemphasize that due toa number of modelling 0 assumptionsandtheobservationaluncertaintiesforindividualobjects,onlyalarge,homogeneoussample,astheoneusedhere,can 7 enableaconclusivestudyoftherelationbetweenT (H+)andT (O). 0 e e / Keywords.galaxies:irregular—galaxies:starburst—galaxies:ISM—galaxies:abundances h p - o 1. Introduction atureofarelativelylargesampleoflow-metallicityHregions r t usingtheBalmerandPaschenjumps. s Along-standingprobleminthestudyofHregionsisconnected a tothefactthattheelectrontemperaturesderivedfromthehydro- Differences between T (H+) and T (O ) were first dis- : e e v gen Balmer and Paschen discontinuity or from the UV helium cussed by Peimbert (1967), who introduceda conceptof tem- i X discontinuity are systematically lower than those derived from perature fluctuations in the nebulae. To quantify temperature the collisionallyexcitedoptical[O ] lineswhichare mostof- fluctuationsheusedtheparametert2,themeansquaretempera- r a ten used for the temperature determination (e.g. see review in turevariation.Itwasshowninmanystudiesofplanetarynebu- O’Delletal.,2003). laeandHregions(e.g. Peimbertetal.,1995;Estebanetal., 1998,1999,2002;Liuetal.,2000,2001)thattheheavyelement Electron temperature determinations based both on colli- abundancesderivedfromthe recombinationlines and from the sionallyexcitedlinediagnosticsandontheBalmerandPaschen collisionallyexcitedlinesareconsistentift2 isintherange0.02 jumps have been mainly applied to planetary nebulae (PNe) –0.10.Ontheotherhand,typicalvaluesoft2forphotoionization (e.g. Liuetal., 2000; Luoetal., 2001; Ruizetal., 2003; models of chemically and spatially homogeneous nebulae are Zhangetal.,2004;Peimbertetal.,2004;Wessonetal.,2005) and to nearly solar-metallicity H  regions in the Milky Way significantlylower,t2 = 0.00–0.02(e.g. Kingdon&Ferland, 1995;Pe´rez,1997). and several nearby galaxies with relatively high metallici- ties of 1/3 – 1/10 solar (e.g. Peimbert&Torres-Peimbert, Inrecentyearspoint-to-pointmeasurementsofthe electron 1992; Peimbertetal., 1993, 2000; Peimbert, 2003; temperature fluctuations have been done in several PNe and H Estebanetal., 1998; Garc´ıa-Rojasetal., 2004, 2005, 2006;  regions (Liu, 1998; Rubinetal., 2002; Krabbe&Copetti, Gonza´lez-Delgadoetal., 1994). A common result of these 2005; Rubinetal., 2003; O’Delletal., 2003). Only low am- studies is that, generally, T (H+) is lower than T (O ). Only e e plitude temperature variations were found across the nebulae. recentlyGusevaetal. (2006)havederivedtheelectrontemper- This is probably because the projected parameter t2 derived in s the point-to-pointmeasurements of the electron temperature is differentfromthetotalparametert2.Copetti (2006)hasshown Sendoffprintrequeststo:N.G.Guseva,[email protected] ⋆ Based on observations collected at the European Southern thatt2s givesonlyalowlimitoft2.Additionally,therearemany Observatory,Chile,ESOprogram76.B-0739. difficulties to obtain the temperature maps of the whole nebu- ⋆⋆ Tables 3 and 4 are only available in electronic form at the laewhenimagesandspectraobtainedwithdifferentspace-born CDS via anonymous ftp to cdsarc.u-strasbg.fr (130.79.128.5) or via and ground-based telescopes are combined (Luridianaetal., http://cdsweb.u-strasbg.fr/cgi-bin/qcat?J/A+A/ 2003a). 2 N.G.Gusevaetal.:Balmerjumptemperatureinlow-metallicityHregions Thus, the observationaldata for PNe and H  regions with much improved statistics, we took new spectra of H  regions relatively low electron temperatures ≤11000K certainly show withthe3.6mESOtelescope.Thesedataarecombinedwiththe systematic differencesbetweenthe temperaturesobtainedfrom onesobtainedbyGusevaetal. (2006). thecollisionallyexcitedlinesandfromtheBalmerandPaschen InSection2,wedescribetheobservationsanddatareduction jumps (see Fig.12 in Gusevaetal., 2006, for the data col- ofthenewspectroscopicdata.InSection3wecompareT (H+) e lectedfromliterature).On theotherhand,nosignificantdiffer- andT (O)forthenewBCDsampleandthosefromliterature. e encesbetween T (H+) and T (O ) were foundfor the higher- OurconclusionsaresummarizedinSection4. e e temperatureHregions(Gusevaetal.,2006).Thisconclusion is in agreement with the results obtained by Peimbertetal. (2002), who compared T (O ) and the electron temperature 2. Observationsanddatareduction e T (He+) in the He+ zone. Adoptingt2 = 0.01– 0.04,these au- e The sample observed with the 3.6m ESO telescope consists of thors found the temperature differencesof (1 – 3)% in H  re- 69low-metallicityHregionsin45BCDs.Theywereselected gions with T (O ) = 20000K and of (3 – 12)% in H  re- e mainly from the literature. Particularly, two galaxies were se- gions with T (O ) = 10000K. As T (He+) < T (H+) in PNe e e e lectedfromtheData Release 2ofthe Six-DegreeField Galaxy (Zhangetal., 2005) and T (He+) ≈ T (H+) in H  regions e e Redshift Survey (6dFGRS) (Jonesetal., 2005) and eighteen (Garc´ıa-Rojasetal., 2005), the differences between T (O ) e galaxies were selected from the Data Release 4 of the Sloan and T (H+) are expected to be not greater than (1 – 3)% in H e Digital Sky Survey (SDSS) (Adelman-McCarthyetal., 2006). regionswiththetemperaturesT (O)∼20000K. e The BCDs were chosen to span a large range of oxygenabun- Different mechanismsare proposed to explain the observa- dance, from about 1/60 to about 1/3 that of the Sun. Thus, we tionaldifferencesinT .Inparticular,densityandchemicalabun- e canstudythedependenceoftemperaturevariationswithmetal- dance inhomogeneities, ionization of nebulae by low-energy licity. Since the electron temperature of a H  region depends cosmic rays are considered (e.g. Torres-Peimbertetal., 1990; on its metallicity, H  regions in our BCD sample span also a Viegas&Clegg, 1994; Liuetal., 2000; Tsamisetal., 2004; large range of electron temperatures. The equivalent width of Tsamis&Pe´quignot, 2005; Giammanco&Beckman, 2005). theHβline,EW(Hβ),isameasureoftherelativecontributionof However, no definite conclusions have been made concerning theionisedgasemissiontothetotallight.Theselectedgalaxies themainmechanismforthesedifferences. coveralargerangeofEW(Hβ)from10Åto382Å. TheknowledgeofdifferencesinT isimportantfortheele- e All spectra were obtained with the EFOSC2 (ESO Faint mentabundancedetermination.Theheavyelementabundances Object Spectrograph and Camera) mounted at the 3.6m ESO are usually derived assuming that the temperature of the H+ telescope at La Silla in two observing runs, the first one, dur- zone is equal to the temperature of the O2+ zone. If instead, ingApril11–14,2005andthesecondone,duringOctober7– T (H+) is smaller than T (O ), then the heavy element abun- e e 9, 2005.The observingconditionswere photometricduringall danceswouldbeincreased.Forthedeterminationoftheprimor- nights.Forthespringobservationsweusedthegrism#14andthe dial He abundance from spectra of low-metallicity BCDs (e.g. grating600gr/mm.Thelongslitwith1′′×300′′wascenteredon Izotov&Thuan,2004)theknowledgeofthetemperaturestruc- the brightestpartofeach galaxy.Theaboveinstrumentalsetup ture of a H  region is especially important because the high gave a wavelength coverage of λλ3200–5083, a spectral reso- precisionof1 –2 %isrequiredfortheHe abundance.Thusto lutionof ∼6.2 Å (FWHM) anda spatialscale of 0′.′314pixel−1 estimatethesystematicerroroftheprimordialHeabundanceitis alongtheslitfortheused2×2pixelbinning. crucialtoinvestigatewhethertemperaturedifferencesareasim- During the fall observations the grism #07 and the grat- portantin low-metallicityBCDs as they are in high-metallicity HregionsandPNe. ing 600 gr/mm were used, resulting in a wavelength coverage of λλ3250–5200. These observations were carried out with a Gusevaetal. (2006)havedeterminedtheelectrontempera- ture of H+ zones from the Balmer jump in 23 H  regionsand 1′.′2×300′′slitcenteredonthebrightestpartofeachgalaxy.The fromthePaschenjumpin24Hregionsinthemetallicityrange spectralresolutionandspatialscalealongthisslitwere∼6.2Å (FWHM)and0′.′157pixel−1,respectively. from1/3to1/60ofsolar,based,respectively,onMultipleMirror Our sample galaxies were mostly observed at low airmass Telescope (MMT) and Sloan Digital Sky Survey (SDSS) data. <1.2. Spectroscopicobservationsat a higherairmass were car- They used Monte Carlo simulations, varying the electron tem- peratureintheH+zone,theextinctionoftheionisedgasandthat riedoutalongtheparallacticangle.Thus,nocorrectionsforat- mosphericrefractionhavebeenapplied.Thetotalexposuretime ofthestellarpopulation,therelativecontributionoftheionised of typically40– 60minutespergalaxywas splitup into 2 – 3 gastothetotalemissionandthestarformationhistorytofitthe subexposures to allow for a more efficient rejection of cosmic spectral energy distribution (SED) of the galaxies in the large rayhits.Threespectrophotometricstandardstarswereobserved wavelengthrange,whichincludestheBalmerandPaschendis- duringeachnightforfluxcalibration.Thejournaloftheobser- continuities.Thebestsetsoffreeparametershavebeenobtained vationsisgiveninTable1. from the minimizationof the deviationsbetween observedand modelledSEDs. Itwas foundthatthe temperaturesT (O) of ThedatareductionwascarriedoutwiththeIRAF1 software e the O2+ zonesdeterminedfrom the nebular to auroralline flux package. This includes bias subtraction, flat–field correction, ratio of doublyionised oxygen[O ] λ(4959+5007)/λ4363do cosmic-ray removal, wavelength calibration, night sky back- notdiffer,inastatisticalsense,fromthetemperaturesT (H+)of ground subtraction, correction for atmospheric extinction and e the H+ zonesdeterminedfrom Balmer and Paschen jumps. On absolute flux calibrationof the two–dimensionalspectrum.For each night the sensitivity curve was derived from averaging of theotherhand,Gusevaetal. (2006)haveemphasizedthat,due threestandardstars. to large observationaluncertaintiesand modellingassumptions forindividualobjects,onlya statistical studyofa largesample 1 IRAFistheImageReductionandAnalysisFacilitydistributedby ofHregionscanallowfordefiniteconclusions. theNationalOpticalAstronomyObservatory,whichisoperatedbythe Thereforein orderto studytheTe(O) vs.Te(H+) relation AssociationofUniversitiesforResearchinAstronomy(AURA)under overalargemetallicityrange,betweenZ⊙/60andZ⊙/3,andwith cooperativeagreementwiththeNationalScienceFoundation(NSF). N.G.Gusevaetal.:Balmerjumptemperatureinlow-metallicityHregions 3 temperatureT (O ). The latter was obtained from the relation e betweenT (O)andT (O)ofIzotovetal. (2006b).The3.6m e e spectracoveredonlythebluewavelengthregion,so thatthe [S ] λ6717,6731emission lines can notbe used for the electron number density determination. Therefore we have adopted N e = 100 cm−3. The electron temperatures T (O ) and T (O ) e e for the high- and low-ionisation zones in H  regions respec- tively,theionisationcorrectionfactors(ICFs)andtheionicand total heavy element abundances for oxygen, neon and iron are given in Table 4 excluding data for 16 H  regions published inthepapersbyIzotovetal. (2006a);Papaderosetal. (2006b) andPapaderosetal.(2006,inpreparation). Relevantparameterstothepurposeofthepapersuchasthe equivalentwidthEW(Hβ)oftheHβemissionline, theelectron temperature t (O ) of the O2+ zone, the electron temperature e t (H+) of the H+ zone, the oxygen abundance 12 + log O/H e Fig.1. Best fit model SED to the redshift- and extinction- andtheextinctioncoefficientC(Hβ),obtainedfromtheBalmer corrected observed spectrum of UM 462c. The model SED is decrement,are collectedin Table2 forthe entiresampleofthe calculatedassumingthattheHregionisionisation-bounded.It 69Hregionsobservedwiththe3.6mESOtelescope. isseenthattheredpartofspectrumisfittedquitewellwhereas the modelled SED underestimates the flux shortward of the Balmerjump.Thisislikelyduetotheleakageofionisingpho- 3.2.ElectrontemperatureTe(H+) tonsfromtheHregion. To derive the electron temperature T (H+) we use the same e methodasthatdescribedindetailbyGusevaetal. (2006).The One-dimensional spectra of the bright H  regions in each method is based on the determination of the electron temper- ature T (H+) of H+ zones by fitting a series of model SEDs galaxy were corrected for interstellar extinction using the red- e to the observed SEDs and finding the best fits. Each fit is per- dening curve by Whitford (1958) and for redshift, derived formed over the whole observed spectral range shortward of fromtheobservedwavelengthsoftheemissionlines.Redshift- correctedspectraareshowninFig.3. ∼ λ(5100 – 5200)Å, which includes the Balmer jump region Emission line fluxes were measured using Gaussian pro- (λ3646Å).BesidestheelectrontemperatureTe(H+)whichcon- file fitting. The errors of the line flux measurements were cal- trols the magnitude of the Balmer jump, the shape of the SED culated from the photon statistics of non-flux-calibrated spec- dependsonseveralotherparameters.AseachSEDisthesumof tra. They have been propagated in the calculations of the ele- both stellar and ionised gas emission, its shape dependson the mental abundance errors. The observed relative emission line relativebrightnessofthesetwocomponents.InBCDs,thecon- fluxes F(λ)/F(Hβ) and fluxes I(λ)/I(Hβ) corrected for inter- tributionoftheionisedgasemissioncanbeverylarge.However, stellar extinction and underlying stellar absorption, equivalent the EWs of hydrogen emission lines never attain the theoreti- widthsEWsofemissionlines,extinctioncoefficientsC(Hβ),ob- cal values for pure ionized gas emission. This implies a non- servedHβfluxesF(Hβ),andequivalentwidthsofthehydrogen negligible contribution of stellar emission in all sample galax- absorptionlinesarelisted inTable3.Notallgalaxiesobserved ies.Wethereforeparametrizetherelativecontributionofgaseous withthe3.6mtelescopeareincludedinTable3,butallofthem emissiontothestellaronebytheequivalentwidthEW(Hβ). are usedin the subsequentanalysis.We excludedfromTable 3 The shape of the spectrumdependsalso on reddening.The eight H  regions with an oxygen abundance of 12 + log O/H extinction coefficient for the ionised gas C(Hβ) has been ob- ≤7.6.Theemissionlinefluxes,electrontemperatures,ionicand tainedfromtheobservedhydrogenBalmerdecrement.Herewe totalelementabundancesfortheseHregionsarepresentedin assumethatthedifferencesbetweentheobservedandtheoretical Papaderosetal.(2006,inpreparation),aspartofadetailedspec- recombinationhydrogenBalmerlineratiosareonlyduetored- troscopic and photometric study of extremely metal-deficient dening and underlying stellar absorption. However, additional star-forming galaxies. We neither include in Table 3 the two source of differences, collisional excitation of hydrogen lines, extremelymetal-pooremission-linegalaxiesSDSSJ2104–0035 couldplaysomeroleinthehotHregions(Stasin´ska&Izotov, andJ0133+0052discoveredinSDSS.Allobservationaldatafor 2003; Luridiana,etal., 2003b). Collisional excitation may in- thesegalaxieswerepublishedbyIzotovetal. (2006a).Finally, crease the flux of the Hα emission line by <∼ 5% above the re- Table 3 does not include the extremely metal-deficient BCDs combinationvalue.Theeffectislessimportant,<∼2%,fortheHβ SBS 0335–052Wand SBS 0335–052E. A spectroscopic study emissionline.Wefindthatignoringcollisionalexcitationofhy- ofthesetwosystemsispresentedinPapaderosetal. (2006b). drogen lines does not change significantly the derived T (H+). e As the Hα line is not present in our spectra, we decreased by 2%thefluxofHβlinemeasuredinspectraofseveralhottestH 3. Results  regions from our sample. Such correction results in decreas- 3.1.ElectrontemperatureT (O)andelementabundances ing C(Hβ) by ∼ 0.07. Then the derived Te(H+) is not changed e morethanby∼1%ascomparedtothecasewiththemeasured TheelectrontemperatureT ,ionicandtotalheavyelementabun- fluxoftheHβline.Therefore,forthewholesamplewedecided e danceswere derived followingIzotovetal. (2006b). In partic- nottotakeintoaccountcollisionalexcitationofhydrogenlines ular,fortheO2+ andNe2+ ionswe adoptthe temperatureT (O and use the C(Hβ) obtained from the measured hydrogen line e ) derived from the [O ] λ4363/(λ4959 + λ5007) emission fluxes. We have no direct observational constraint for the red- line ratio. The O+ and Fe++ abundanceswere derived with the deningforthestellarcomponent,whichcoulddifferfromC(Hβ). 4 N.G.Gusevaetal.:Balmerjumptemperatureinlow-metallicityHregions Fig.2.(a)ExampleoftheparameterspaceexploredwiththeMonteCarlotechnique(2253solutionsoutof105 simulations)tofit the spectrumofthe BCD UM 311.The parametersshownare the equivalentwidth ofthe Hβ emissionline EW(Hβ),the agesof the old and youngstellar populations,t(old) and t(young),the mass ratio b of the old-to-youngstellar population,the extinction coefficientforthestellaremissionC(Hβ) ,theextinctioncoefficientfortheionisedgasC(Hβ) ,andtheelectrontemperature stars gas t (H+)=10−4T (H+)intheH+zone.Theparameterσisanestimatorofthegoodnessofthefit.(b)and(c)arethesameas(a)except e e thatonlythe64and10bestMonteCarlorealizationsareshown. Therefore,forsimplicity,weadoptincalculationsthatbothex- the EW(Hβ) expected for pure gaseous emission. The gaseous tinctions C(Hβ) and C(Hβ) , respectively for the ionised continuum emission is calculated following Aller (1984) and gas stars gasemissionandthestellarcomponentemission,arerandomly includes hydrogen and helium free-bound, free-free, and two- variedin thenarrowrangearoundC(Hβ).Finally,the SED de- photonemission.Inourmodelsitisalwayscalculatedwiththe pendsonthestarformationhistoryoftheBCD. electrontemperatureT (H+)andwiththeHe/Habundanceratio e We have carried out a series of Monte Carlo simula- derived from the H  region spectrum. The observed emission tions to reproduce the SED in each H  region of our sam- linescorrectedforreddeningandscaledusingtheabsoluteflux ple. To calculate the contribution of stellar emission to the of the Hβ emission line were added to the calculated gaseous SEDs, we have adopted a grid of the Padua stellar evo- continuum. lution models by Girardietal. (2000)2 with heavy element Given that T (H+) is not necessarily equal to T (O ), we e e mass fractions Z = 0.0001, 0.0004,0.001, 0.004,0.008. Using chosetovaryitintherange(0.7–1.3)×T (O).Weadoptfor e these data we have calculated with the package PEGASE.2 therangeoftheallowedEW(Hβ)valuesbetween0.95and1.05 (Fioc&Rocca-Volmerange,1997)agridofinstantaneousburst times its nominal value. As for the extinction, we allow both SEDsinawiderangeofages,from0.5Myrto15Gyr.Wehave C(Hβ) and C(Hβ) to vary in the range of (0.8 – 1.2) of stars gas adopteda stellar initialmassfunctionwith a Salpeter slope,an C(Hβ),obtainedfromtheBalmerdecrement. uppermass limitof100 M⊙ anda lowermass limitof0.1 M⊙. We assumethattheyoungstellar populationwasformedin Following Gusevaetal. (2006) we could approximate the star a recent burst of star formation with an age t(young) between formationhistory in BCDs by two short bursts of differentage 0.3 and 10 Myr. For the age of the older stars t(old) we adopt butequaldurationsandwith differentstrengths.Theoldstellar values between 10 Myr and 15 Gyr. The contribution of each populationisusuallyapproximatedbycontinuousstarformation burst to the SED is defined by the ratio of the masses of stel- (e.g. Gusevaetal.,2001,2003a,b,c).However,thecontribution larpopulationsformedrespectivelyintheoldandyoungbursts, oftheoldstellarpopulationtothelightofthebrightHregions b = M(old)/M(young), which we vary between 0.01 and 100. considered in this paper is always small, not exceeding a few For each H  region we computed a series of synthetic SEDs percentin the optical range. Therefore,the approximationof a usingagridofPaduamodelswiththeheavyelementmassfrac- short burst for the old stellar population is acceptable for our tion, that is closest to the heavy element mass fraction of the modelling,asitwasshownbyGusevaetal. (2006).Thecontri- ionisedgas.Werun105MonteCarlomodelsforeachHregion butionofgaseousemissiontothetotalemissionisscaledbythe varyingsimultaneouslyt(young),t(old),b,T (H+),C(Hβ) and e gas ratiooftheobservedequivalentwidthoftheHβemissionlineto C(Hβ) . The first four parameters are used to calculate the stars model EW(Hβ). Only those solutions were further considered 2 http://pleiadi.pd.astro.it. for the SED fitting in which the modelled EW(Hβ) falls in the N.G.Gusevaetal.:Balmerjumptemperatureinlow-metallicityHregions 5 Fig.4. (a)Comparisonofthet (H+)derivedfromfittingtheBalmerjumpandtheSEDwiththet (O)=10−4T (O)derivedfrom e e e thenebulartoaurorallinefluxratio[O]λ(4959+5007)/λ4363.Ineachpanel,thedashedlinedenotesequaltemperatures,thesolid lineisthelinearregressionobtainedbythelikelihoodmethod(Pressetal.,1992),dottedlinesdenote1σdispersionsofoursample H  regionsaroundthe regressionline. Error barsare the rootmeansquare temperaturedeviationsfromthe mean of the 10 best MonteCarlosimulations.3.6mdataareshownbylargefilledcircles.Twoobjects,Pox186andUM462c,whichcouldnotbefitted byionisation-boundedmodelaredenotedbysquares.Foreachofthesetwogalaxiesweshowbyopensquaresthesolutionsfrom the ionisation-boundedmodeland by filled squaresthe solutionsfromthe modelwith escapingLyc photons.The derivedt (H+) e fromtheBalmerjumpin23low-metallicityHregions(MMTdata)andthePaschenjumpin24low-metallicityHregions(SDSS data)areshownbylargeandsmallopencircles,respectively(Gusevaetal.,2006).Forcomparisonwehavealsoplottedbydotsin (a)thedataofLiuetal. (2004),Zhangetal. (2004),Wessonetal. (2005)andKrabbe&Copetti (2005)for58GalacticPNe.Stars showthedataforH regionsfromEstebanetal. (1998),Garc´ıa-Rojasetal. (2004, 2006),Peimbert&Torres-Peimbert (1992), Peimbertetal. (2000),Peimbert (2003)andGonza´lez-Delgadoetal. (1994).(b)Thispanelisidenticaltopanel(a)exceptthatonly H regionswithsmalldispersionsfort (H+)(σ[t (H+)]/t (H+) <0.1)for3.6mdata(presentpaper)anddatafromGusevaetal. e e e (2006)(MMTandSDSSdata)areshown.(c)Thesameasin(a)butonlyHregionswithlargeequivalentwidthsandsmallerrors fort (H+)andt (O)areincluded(σ[t (H+)]/t (H+)<0.1;σ[t (O)]/t (O)<0.013;EW(Hβ)>80Å). e e e e e e range 0.95 – 1.05 of the observed one. Typically, the number we choose to average. No systematic differences were found of such solutionsis ∼ 10 – 100 times less than the total of 105 when the number of best fits varies between a few and a cou- simulations. ple of ten (Gusevaetal., 2006). Hence, we adoptfor t (H+) = e 10−4T (H+) the mean of the 10 best-fitting solutions weighted In the same way as in Gusevaetal. (2006) we used the σ e byσ. statistic to quantifythe goodnessof each model’sfit to the ob- WenotethatthebimodalσdistributionforsomeHregions served SED. For each Monte Carlo realization we computed a does notallow to derivea uniqueT (H+) in those objects. The mean deviation σ between the observed and modelled spectra. e averageT (H+)ofthe10best-fittingsolutionsfortheseHre- Wecalculateineachspectrumtherootmeansquaredeviationσ e j gionscan significantly differ from the actualone. On the other foreach of the five continuumregionsselected to be devoidof hand,thedispersionoftheaverageT (H+)sinHregionswith lineemissionandabsorption,asσj = qPiN=j1(foibs− fmiod)2/Nj, thebimodalσdistributionislargerefleectingtheuncertaintiesof whereN isthenumberofthepointsineachparticularspectral the temperaturedetermination.Therefore,in the followingdis- j interval.Thenσ= σ /5. cussionwewillusedifferentsamples:thetotalsample(Fig.4a) j P and the samples with the sharp minimum in the σ distribution AsshowninFig.2a,forallMonteCarlorealizationstheage (Figs.4b,c). t(old)andt(young)andmassratioboftheoldandyoungstellar populations,andtheelectrontemperatureT (H+)intheH+zone e spanawiderange.However,ifonlysolutionswiththelowestσs 3.3.Comparisonoft (H+)andt (O) e e aretakenintoaccountthentherangeoftheelectrontemperature intheH+ zoneformanyHregionsfromoursampleisconsid- Figure3showsbyathicksolidlinethebest-fitSED(SEDwith erablynarrowed(Fig.2band 2c).Ontheotherhand,forsome the smallest σ) superimposed on the redshift- and extinction- H  regions the minimum in the distribution of σ is not sharp correctedobservedspectrumof eachH  regionin the sample, andeventwominimainthedistributionofσareobtainedwith includingtheHregionsdiscussedbyIzotovetal. (2006a)and significantlydifferentT (H+)(see alsoFig.9 in Gusevaetal., Papaderosetal.(2006,inpreparation). e 2006). The T (H+) of each H  region is derived as the aver- Thegalaxiesare arrangedin orderof decreasingHβequiv- e ageoftheelectrontemperaturesobtainedforseveralbest-fitting alent width, i.e. in order of decreasing contribution of gaseous SEDs.WealsoderivethedispersionaroundtheaverageT (H+) emissionrelativetostellaremission.Itisseenthat,ineverycase, e oftheelectrontemperaturesobtainedfromthesamebest-fitting themodelSEDfitswelltheobservedoneoverthewholespectral SEDs. Thensmalldispersionscorrespondto sharpsinglemini- range,includingtheBalmerjumpregion.Theseparatecontribu- muma in the distributions of σ. We have checked how the de- tionsfromthestellarandionisedgascomponentsareshownonly rivedT (H+)doesdependontheexactnumberofbestsolutions forSBS0335–052ESEbythinsolidlines. e 6 N.G.Gusevaetal.:Balmerjumptemperatureinlow-metallicityHregions InFig.4a wecomparet (H+) andt (O) =10−4 T (O). andlikelyshowsLyαemissioninthe UV. Thesepropertiesled e e e The 3.6m data for the sample of 69 H  regions in 45 BCDs Gusevaetal. (2004)toconcludethattheHregioninPox186 areshownbylargefilledcirclesand,inthecaseofPox186and maybedensity-bounded.IfthisisindeedthecasethenEW(Hβ) UM462c,bysquares.Theelectrontemperaturet (H+)foreach does not measure the starburst age anymore. Its inclusion into e galaxyisthemeanofthetemperaturesderivedfromthe10best ourmodellingprocedurewouldthenresultinanincorrectrepro- Monte Carlo realizationsweightedby their σs (Table 2). Error ductionofthestarformationhistoryandthestellarSED.While bars are the root mean square temperature deviations from the this assumption is less critical in the case of Pox 186 because mean of the 10 best Monte Carlo simulations. The dashed line of the high EW(Hβ) and hence the dominant contribution of denotes equal temperatures, while the solid line is a linear re- gaseous emission to the total emission of this system, it is im- gressiont (H+)=1.078×t (O)–0.054obtainedwiththelike- portantforthecaseofaHregionwithlowerEW(Hβ),suchas e e lihood method (Pressetal., 1992) for all 116 H  regionsthat UM462c.ThisisillustratedinFig.1whereweshowthemodel takes into account errors in t (H+) and t (O ) in each object. SED which best fits the observed spectrum of UM 462c. It is e e Dotted lines illustrate 1σ deviationsof the sample H  regions seenthattheionisation-boundedmodeloftheHregionfailsto from the linear regression. The linear regression is almost not reproduce the observations regardless of the adopted synthetic changediftheoutlyingobjectwiththehighestelectrontemper- SED.Specifically,themodelsunderestimatetheSEDblueward aturet (O) isexcluded.ItisseenfromFig.4athatthelinear oftheBalmerjump,whereasredwardofittheyprovideasatis- e regression(solidline)isslightlydifferentfromthelineofequal factorilyfit.Themostlikelyreasonforthisdiscrepancyisleak- temperatures (dashed line) because of the small dispersion of ageofionisingphotonsfromtheHregion. t (H+)fortheHregionMrk71No.1(Gusevaetal.,2006)and, ThediscrepancybetweenobservedandmodelledSEDscan e therefore,itslargeweightinthelinearregressiondetermination. be eliminated,however,assuming a H  regionmodelwith es- Ontheotherhand,ifequaldispersionsoft (H+)areadoptedfor capingLyc photons,which impliesa largernumberof massive e allHregions,thelinearregressiondeviatesverylittlefromthe ionising stars than the number needed to account for the ob- lineofequaltemperatures.Allthesedifferencesaresmallerthan servedEW(Hβ)inthecaseofanionisation-boundedHregion. the1σdeviationsofHregionsfromthelinearregression(dot- Thus,MonteCarlosimulationsassumingaH regionwith es- tedlines).Therefore,weconcludethatnosystematicdifference capingLyc photonsrequirea youngerstellar populationwith a betweent (O)andt (H+)isevidentfromFig.4a. larger production rate of ionising photons, compared to mod- e e Our 3.6m observations in Fig. 4a are supplemented by the elsassuminganionisation-boundedH region.Asa resultthe data frompreviousstudies.The resultsoft (H+) determination stellarSEDbluewardoftheBalmerjumpisactuallylargerthan e from Balmer jump for 23 low-metallicity H  regions (MMT what fits to EW(Hβ) imply.In practice,in orderto accountfor data) and Paschen jump for 24 low-metallicity H  regions thefractionofphotonsescapingfromtheHregionweneedto (SDSSdata)fromGusevaetal. (2006)areshownbylargeand introduceanadditionalparameter f.Then,themodelledSEDof small open circles respectively. In Fig. 4a we also plot some the ionised gas emission and EW(Hβ) must be multiplied by a other data from the literature. By stars are shown the H  re- factor(1– f)tomatchtheobservedones. gions:M17(Peimbert&Torres-Peimbert,1992),Orionnebula InFig.4awe showbyopensquaresthet (H+) forPox186 e (Estebanetal.,1998),NGC3576(Garc´ıa-Rojasetal.,2004),S andUM462cobtainedfromtheionisation-boundedmodeland 310(Garc´ıa-Rojasetal.,2004, 2006) intheGalaxy;30Dorin by filled squares the t (H+) obtained from the model with es- e LargeMagellanic Cloud (Peimbert, 2003); NGC 346 in Small capingLycphotons.Itisseenthatthet (H+)obtainedfromthe e Magellanic Cloud (Peimbertetal., 2000); and H  regions A latter models for both H  regions (filled squares) are in bet- and B in the BCD Mrk 71 (Gonza´lez-Delgadoetal., 1994). ter agreement with the data for other H  regions. This is also By dots in Fig. 4a we plot the data for 58 Galactic PNe by evident from Fig. 3 where we plot the best-fit SEDs obtained Liuetal. (2004),Zhangetal. (2004),Wessonetal. (2005)and forthese twosystemsfrommodelswithescapingLycphotons. Krabbe&Copetti (2005).Thedatacollectedfromliteratureex- TheseSEDsreproducequitewelltheobservedspectraovertheir tendthet (H+)–t (O)relationtolowertemperatures. whole wavelength range, including the region blueward of the e e InFig.4bareshownonly41Hregionswithsmalldisper- Balmerjump. sionsoft (H+) (σ[t (H+)]/t (H+) <0.1)for3.6mdata(present The leakage of Lyc photons could be present in other H e e e paper)andfordatafromGusevaetal. (2006)(MMTandSDSS  regions from our sample. This may introduce an additional data). These data are fitted by a linear regression t (H+) = sourceofuncertaintiesinthedeterminationoft (H+).However, e e 1.085×t (O ) – 0.050 (solid line) that is very similar to the thefactthatmajorityofHregionsarewellfittedbythemod- e linear regressionobtainedfor the whole sample of 116 H  re- elswithnon-escapingLycphotonssuggeststhatingeneral f is gion.Ifthe outlyingpointwith the highestt (O ) is excluded small.Ourconclusionissupportedbymulti-wave-bandstudies e thelinearregressionisalmostnotchanged.Thus,fromtheanal- byLeithereretal. (1995).Theyfindthatlessthan3%ofioniz- ysisofthesubsampleofHregionswithsharpestminimainthe ingphotonsescapefromlocalstarburstgalaxies.Morerecently, σdistributionwereachthesameconclusionasthatbasedonthe forseverallocalstarburstgalaxies,Heckmanetal. (2001)esti- total sample: there is no significant differencesbetween te(H+) mate f <∼6%andBergvalletal. (2006)find f ∼4%–10%for andt (O).However,wecannotexcludesmalldifferencesof≤ anotherlocalstarburst.ThedirectdetectionofLymancontinuum e 1000K,correspondingtothetemperaturefluctuationparameter emissionfrom14high-redshiftstar-forminggalaxiesshowsthat t2≤0.02. only2of14studiedgalaxieshaveasignificantemissionbelow SpecialemphasiswasgiventotheanalysisofafewHre- theLymanlimit(Shapleyetal.,2006).Therefore,forallourH gions which are most deviant from the line of equal electron regionsexceptforPox186andUM462cwedecidedtoapply temperatures in Fig. 4a. Two of these H  regions, UM 462c themodelswiththesmallestnumberoffreeparametersadopting and Pox 186,are labeled in Fig. 4a. One of these sources, Pox f =0. 186, has been studied previously by Gusevaetal. (2004). No Inspection of Fig. 4b also reveals that some of the H  re- H  21 cm emission was detected in this galaxy.Its H  region gionsfromthe3.6mdatasamplewiththehighesttemperatures is characterisedbya veryhighO2+/O+ abundanceratioof ∼20 [(te(H+)∼1.9andte(O)>∼2.0)]areoffsetfromthelinearre- N.G.Gusevaetal.:Balmerjumptemperatureinlow-metallicityHregions 7 gressionandfromthelineofequaltemperatures,atvarianceto We also identify and discuss cases of H  regions where H regionswith alowert (H+).Theseoffsetgalaxiesarelow- ionisation-boundedmodels fail to fit the observed SEDs. Such e excitation H  regions with weak [O ]λ4363 emission lines. modelstypicallyunderestimatetheobject’sSEDbluewardofthe Therefore, the determination of t (O ) is more uncertain in Balmerdiscontinuity.Thiscanbe,however,plausiblyaccounted e those galaxies. Indeed, the number of deviant points from the forbyphotoionisationmodelsallowingforthepartialleakageof line of equal temperatures are decreased if only galaxies with LycphotonsfromaHregion. high equivalentwidths of Hβ (EW(Hβ) > 80Å) and small t (O e )andt (H+)errorsareconsidered(Fig.4c). Acknowledgements. Y.I.I.andN.G.G.thankthehospitality oftheInstitute e forAstrophysics(Go¨ttingen),andthesupportoftheDFGgrantNo.436UKR WiththeadditionalPNedataaninterestingtrendappearsin 17/25/05.P.P.wouldliketothankGaspareLoCurto,LorenzoMonaco,CarlosLa Fig.4a.WhilethePNedatascattersnicelyoneithersideofthe Fuente,EduardoMatamorosandthewholeESOstaffattheLaSillaObservatory line of equal temperatures for te(O ) >∼ 1.1, just as the BCD fortheirsupport.AlltheauthorsacknowledgetheworkoftheSloanDigitalSky data, the t (H+) – t (O ) relationshipcurvesdownfor objects Survey(SDSS)team.FundingfortheSDSShasbeenprovided bytheAlfred withte(Oe)<∼1.1,ewithte(H+)beingsystematicallylowerthan PS.pSalcoeaAndFmouinnidsatrtaiotino,nt,htehePaNrtaitciiopnaatilnSgcIiennsctietuFtioounnsd,athtieonN,atthioenUa.lSA.eDreopnaaruttmicesnatnodf te(O).ThereareseveralcoolHregionscollectedfromliter- Energy,theJapaneseMonbukagakusho,andtheMaxPlanckSociety.TheSDSS atureandshownbystars,whichalsofollowthetrenddelineated Websiteishttp://www.sdss.org/. by PNe. All other H  regionswith t > 1.0 collected fromlit- e erature do not deviate from line of equal temperatures, similar to the sample of 116 H  regions in 88 emission-line galaxies References collectedfrom3.6m,MMTandSDSSdata. Adelman-McCarthy,J.K.,Augu¨eros,M.A.,Allam,S.S.,etal.2006,ApJS,162, Thus, we confirm the previous finding by Gusevaetal. 38 (2006)thatthetemperatureoftheOzoneisequal,withinthe Aller,L.H.1984,PhysicsofThermalGaseousNebulae,Dordrecht:Reidel errors,tothetemperatureoftheH+zoneintheHregionswith BerOg¨vstalliln,,NG..,2Z0a0c6k,rAiss&oAn,,E44.,8A,5n1d3ersson, B.-G.,Arnberg, D.,Masegosa, J.,& Te(H+) >∼ 11000K. We emphasize that, due to the large obser- Copetti,M.V.F.2006,A&A,453,943 vationaluncertaintiesforeachindividualobjectandanumberof Esteban,C.,Peimbert,M.,Torres-Peimbert,S.,&Escalante,V.1998,MNRAS, modellingassumptions,onlyastatisticalstudyofalarge,homo- 295,301 geneoussampleofHregionsallowsustofirmlyestablishthe Esteban,C.,Peimbert,M.,Torres-Peimbert,S.,Garc´ıa-Rojas,J.,&Rodr´ıguez, M.1999,ApJS,120,113 relationbetweenT (H+)andT (O). e e Esteban,C.,Peimbert,M.,Torres-Peimbert,S.,Rodr´ıguez,M.2002,ApJ,581, 241 Fioc,M.,&Rocca-Volmerange,B.1997,A&A,326,950 4. Conclusions Garc´ıa-Rojas, J., Esteban, C., Peimbert, M., Rodr´ıguez, M., Ruiz, M. T., & Peimbert,A.2004,ApJS,2004,153,501 Garc´ıa-Rojas, J., Esteban, C., Peimbert, A., Peimbert, M., Rodr´ıguez, M., & Basedon3.6mESOspectroscopicdata,wehavedeterminedin thispaperthetemperaturesoftheH+zonesin69Hregionsof Ruiz,M.T.2005,MNRAS,362,301 Garc´ıa-Rojas, J., Esteban, C., Peimbert, M., Costado, M.T., Rodr´ıguez, M., 45bluecompactdwarf(BCD)galaxies.Followingtheprocedure Peimbert,A.,&Ruiz,M.T.2006,MNRAS,368,253 ofGusevaetal. (2006),wehaveusedaMonteCarlotechnique Giammanco,C.,&Beckman,J.E.2005,A&A,437,L11 tovaryfreeparametersandtocalculateaseriesofmodelspec- Girardi,L.,Bressan,A.,Bertelli,G.,&Chiosi,C.2000,A&AS,141,371 tralenergydistributions(SEDs).TheelectrontemperatureofH+ Gonza´lez-Delgado, R.M.,Perez,E.,Tenorio-Tagle,G.,etal.1994,ApJ,437, 239 zones were derived from best-fit SEDs to the observed spec- Guseva,N.G.,Izotov,Y.I.,Papaderos,P.,Chaffee,F.H.,Foltz,C.B.,Green,R. trainthewavelengthrange∼3200–5100Å,whichincludesthe F.,Thuan,T.H.,Fricke,K.J.,&NoeskeK.G.2001,A&A,378,756 Balmerjumpregion.Thesenewdataarecombinedwithasam- Guseva,N.G.,Papaderos,P.,Izotov,Y.I.,Green,R.F.,Fricke,K.J.,Thuan,T. pleof47HregionsfromGusevaetal. (2006)inordertostudy H.,&NoeskeK.G.2003a,A&A,407,75 Guseva,N.G.,Papaderos,P.,Izotov,Y.I.,Green,R.F.,Fricke,K.J.,Thuan,T. systematictrendsintheelectrontemperaturedeterminationover H.,&NoeskeK.G.2003b,A&A,407,91 awiderangeinoxygenabundance(7.13≤12+log(O/H)≤8.36) Guseva,N.G.,Papaderos,P.,Izotov,Y.I.,Green,R.F.,Fricke,K.J.,Thuan,T. and Hβ equivalent width (10Å ≤EW(Hβ) ≤397Å). This is the H.,&NoeskeK.G.2003c,A&A,407,105 largestspectroscopicsampleoflow-metallicityHregionsever Guseva,N.G.,Papaderos,P.,Izotov,Y.I.,Noeske,K.G.,&Fricke,K.J.2004, A&A,421,519 compiledinordertoaddressthelong-standingquestionofhow Guseva,N.G.,Izotov,Y.I.,&Thuan,T.X.2006,ApJ,644,890 the electron temperature of H+ zones is related to that derived Heckman,T.M.,Sembach,K.R.,Meurer,G.R.,Leitherer,C.,Calzetti,D.,& fromcollisionallyexcitedforbiddenlines. Martin,C.L.2001,ApJ,558,56 WeconcludethatinHregionswithanelectrontemperature Izotov,Y.I.,&Thuan,T.X.2004,ApJ,602,200 Izotov,Y.I.,Papaderos,P.,Guseva,N.G.,Fricke,K.J.&Thuan,T.X.2006a, Te(H+)>∼11000KthetemperatureoftheO2+ zone,determined A&A,454,137 fromthenebulartoaurorallinefluxratioofdoublyionisedoxy- Izotov,Y.I.,Stasin´ska,G.,Meynet,G.,Guseva,N.G.,&Thuan,T.X.2006b, gen [O ] λ(4959+5007)/λ4363is, within the errors, equal to A&A,448,955 thetemperatureoftheH+zonedeterminedfromfittingtheSED Jones,D.H.,Saunders,W.,Read,M.A.,&Colless,M.2005,PASA,22,277 Kingdon,J.B.&Ferland,G.J.1995,ApJ,450,691 includingthatintheBalmerjumpwavelenghtregion.However, Krabbe,A.C.,&Copetti,V.F.2005,A&A,443,981 we could not exclude small differences of ≤ 1000 K between Leitherer,C.,Ferguson,H.C.,Heckman,T.M.,&Lowenthal,J.D.1995,ApJ, T (H+) and T (O ) correspondingto the temperature fluctua- 454,L19 e e tionsparametert2≤0.02. Liu,X.-W.1998,MNRAS,295,699 Liu,X.-W.,Storey,P.J.,Barlow,M.J.,Danziger,I.J.,Cohen,M.,&Bryce,M. Consequently,temperaturedifferencescouldnotbe of con- 2000,MNRAS,312,585 cern for the determination of the primordial He abundance, as Liu, X.-W., Luo, S.-G., Barlow, M. J., Danziger, I. J., & Storey, P. J. 2001, it relies mainly on the hot low-metallicity H  regions. Thus, MNRAS,327,141 onthebase ofa newlargesampleoflow-metallicityBCDs we Liu,Y.,Liu,X.-W.,Luo,S.-G.,&Barlow,M.J.2004,MNRAS,353,1231 Luo,S.-G.,Barlow,M.J.,Danziger,I.J.,&Storey,P.J.2001,MNRAS,327,141 confirm and strengthen the previous finding by Gusevaetal. Luridiana,V.,Pe´rez,E.,&Cervin˜o,M.2003a,AJ,125,3196 (2006),basedontheanalysisandmodellingoftheBalmerand Luridiana,V.,Peimbert,A.,Peimbert,M.,&Cervin˜o,M.2003b,ApJ,592,846 Paschenjumpinasmallersampleoflow-metallicityHregions. O’Dell,C.R.,Peimbert,M.,&Peimbert,A.2003,AJ,125,2590 8 N.G.Gusevaetal.:Balmerjumptemperatureinlow-metallicityHregions Papaderos,P.,Izotov,Y.I.,Guseva,N.G.,Thuan,T.X.,&Fricke,K.J.2006, A&A,454,119 Peimbert,A.2003,ApJ,584,735 Peimbert,A.,Peimbert,M.,&Luridiana,V.2002,ApJ,565,668 Peimbert,M.1967,ApJ,150,825 Peimbert,M.,&Torres-Peimbert,S.1992,Rev.Mex.Astron.Astrofis.,24,155 Peimbert,M.,Storey,P.J.,&Torres-Peimbert,S.1993,ApJ,414,626 Peimbert, M., Torres-Peimbert, S., & Luridiana, V. 1995, Rev.Mex.Astron. Astrofis.,31,131 Peimbert,M.,Peimbert,A.,&Ruiz,M.T.2000,ApJ,541,688 Peimbert,M.,Peimbert,A.,Ruiz,M.T.,&Esteban,C.2004,ApJS,150,431 Pe´rez,E.1997,MNRAS,290,465 Press, W. H., Teukolsky, S. A., Vetterling, W. T.,& Flannery, B. P., 1992, NumericalRecipesinC,TheArtofScientificComputing/SecondEdition/, CambrigeUniversityPress Rubin,R.H.,Bhatt,N.J.,Dufour,R.J.,etal.2002,MNRAS,334,777 Rubin,R.H.,Martin,P.G.,Dufour,R.J.,etal.2003,MNRAS,340,362 Ruiz,M.T.,Peimbert,A.,Peimbert,M.,&Esteban,C.2003,ApJ,595,247 Shapley,A.E.,Steidel,C.C.,Pettini,M.,Adelberger,K.L.,&Erb,D.K.2006, ApJ,651,688 Stasin´ska,G.,&Izotov,Y.I.2003,A&A,397,71 Torres-Peimbert,S.,Peimbert,M.,&Pen˜a,M.1990,A&A,233,540 Tsamis,Y.G.,&Pe´quignot,D.2005,MNRAS,364,687 Tsamis,Y.G.,Barlow,M.J.,Liu,X.-W.,Storey,P.J.,&Danziger,I.J.2004, MNRAS,353,953 Viegas,S.&Clegg,R.E.1994,MNRAS,271,993 Wesson,R.,Liu,X.-W.,&Barlow,M.J.2005,MNRAS,362,424 Whitford,A.E.1958,AJ,63,201 Zhang,Y.,Liu,X.-W.,Liu,Y.,&Rubin,R.H.2005,MNRAS,358,457 Zhang,Y.,Liu,X.-W.,Wesson,R.,Storey,P.J.,Liu,Y.,&Danziger,I.J.2004, MNRAS,351,935 N.G.Gusevaetal.:Balmerjumptemperatureinlow-metallicityHregions 9 Table2.Comparisonoft (O)aandt (H+)bforthesamplegalaxies e e Galaxy EW(Hβ) te(O) te(H+) 12+logO/H C(Hβ)c SBS0335-052ESE 382. 1.98±0.05 1.90±0.10 7.28 0.00 Pox186 358. 1.67±0.02 2.00±0.07 7.75 0.08 HS2236+1344a 343. 2.05±0.02 1.96±0.06 7.49 0.14 UM461a 335. 1.66±0.02 1.77±0.32 7.76 0.22 J104457.80+035313.1a 318. 1.96±0.02 1.91±0.08 7.44 0.31 Tol1214-277 314. 2.04±0.03 1.92±0.06 7.50 0.14 UM570 310. 1.83±0.02 1.83±0.13 7.72 0.14 CGCG032-017 308. 1.38±0.01 1.48±0.21 8.09 0.27 Pox108 306. 1.31±0.02 1.31±0.17 8.11 0.08 UM559 306. 1.60±0.02 1.55±0.12 7.72 0.06 SBS0335-052#7 299. 1.97±0.06 1.91±0.06 7.21 0.14 UM442 289. 1.30±0.01 1.29±0.09 8.10 0.14 Mrk1329 288. 1.05±0.01 1.26±0.06 8.33 0.03 Mrk1236a 283. 1.21±0.01 1.30±0.08 8.17 0.14 Tol2146-391 277. 1.63±0.02 1.59±0.32 7.78 0.18 CGCG007-025a 274. 1.58±0.02 1.83±0.27 7.81 0.20 HS2134+0400 262. 2.02±0.04 1.86±0.20 7.41 0.25 J001434.98-004352.0 260. 1.39±0.04 1.31±0.28 7.97 0.12 J232420.34-000625.0 249. 1.41±0.01 1.55±0.22 7.95 0.20 SBS0335-052E#3b 249. 2.04±0.03 2.07±0.22 7.31 0.45 J125305.96-031258.9 248. 1.38±0.01 1.34±0.21 8.01 0.36 J120122.30+021108.3 248. 1.81±0.03 1.70±0.27 7.49 0.09 SBS0335-052E#3a 246. 2.02±0.03 2.05±0.22 7.31 0.43 Tol65a 244. 1.76±0.02 1.83±0.17 7.54 0.10 UM311 243. 0.99±0.01 1.00±0.04 8.36 0.22 J130432.27-033322.1 233. 1.08±0.01 1.20±0.08 8.24 0.12 J024815.95-081716.5a 233. 1.32±0.01 1.13±0.19 8.04 0.04 Pox4a 220. 1.39±0.01 1.11±0.23 8.01 0.08 Pox120a 215. 1.53±0.01 1.76±0.22 7.88 0.21 HS2236+1344b 215. 1.86±0.02 1.80±0.10 7.58 0.22 CGCG007-025c 214. 1.56±0.02 1.49±0.25 7.78 0.02 UM462a 212. 1.39±0.01 1.45±0.22 8.00 0.52 J230210.00+004938.8 212. 1.80±0.03 1.91±0.09 7.63 0.00 Tol2138-405 208. 1.39±0.02 1.08±0.12 7.99 0.39 J210455.31-003522.2N 203. 2.01±0.08 1.87±0.18 7.26 0.00 J011340.44+005239.1 193. 2.29±0.30 2.55±0.04 7.17 0.00 J013352.56+134209.3 178. 1.74±0.02 1.62±0.33 7.56 0.04 Pox4b 177. 1.43±0.01 1.51±0.18 7.87 0.00 IC4662a 172. 1.27±0.01 1.34±0.19 8.09 0.44 J143053.50+002746.3 168. 1.30±0.01 1.21±0.15 8.07 0.15 IC4662c 166. 1.25±0.01 1.25±0.14 8.16 0.33 Pox120b 158. 1.37±0.02 1.18±0.20 7.99 0.14 g2252292-171050 158. 1.59±0.02 1.74±0.24 7.82 0.10 UM461b 147. 1.58±0.02 1.79±0.23 7.77 0.18 Mrk1236c 147. 1.22±0.04 0.98±0.03 8.05 0.00 J144805.36-011057.7 146. 1.39±0.01 1.44±0.24 8.00 0.21 IC4662b 140. 1.27±0.01 1.32±0.07 8.10 0.14 J223036.79-000636.9 133. 1.71±0.03 1.86±0.30 7.64 0.34 Pox105 131. 1.53±0.02 1.49±0.27 7.85 0.04 UM462d 130. 1.45±0.03 1.52±0.26 7.86 0.23 UM462c 127. 1.48±0.02 1.11±0.15 7.88 0.00 J024815.95-081716.5b 125. 1.35±0.04 1.36±0.18 7.99 0.00 CGCG007-025b 122. 1.58±0.02 1.67±0.19 7.75 0.06 J162410.12-002202.5 122. 1.20±0.01 1.21±0.15 8.16 0.16 UM462e 112. 1.59±0.05 1.67±0.26 7.77 0.00 J013452.00-003854.4 111. 1.49±0.15 1.53±0.26 7.85 0.16 SBS0335-052W 109. 1.97±0.25 2.01±0.25 7.13 0.02 Tol65b 96. 1.69±0.04 1.74±0.31 7.59 0.14 SBS0335-052E#4,5 89. 2.11±0.04 1.90±0.13 7.27 0.26 UM462b 85. 1.42±0.02 1.44±0.15 7.92 0.21 CGCG007-025d 69. 1.68±0.04 1.72±0.27 7.65 0.00 ESO338-IG04 59. 1.46±0.02 1.44±0.27 7.90 0.02 Mrk1236b 56. 1.17±0.02 1.14±0.22 8.12 0.07 J141454.13-020822.9 48. 1.89±0.01 1.98±0.23 7.32 0.08 IC4662d 44. 1.25±0.02 1.12±0.28 8.07 0.14 J010746.56+010352.0 40. 1.56±0.23 1.57±0.25 7.68 0.12 g0405204-364859 23. 2.08±0.25 1.95±0.21 7.34 0.02 J104457.80+035313.1b 20. 1.94±0.21 1.81±0.32 7.45 0.14 IC4662e 10. 1.31±0.01 1.30±0.23 8.08 0.06 ate(O)=Te(O)/10000K,derivedfromthe[O]λ(4959+5007)/λ4363emissionlineratio. bte(H+)=Te(H+)/10000,derivedastheweightedmeantemperaturefromthe10bestfitsoftheSED. cExtinctioncoefficientisderivedfromobservedhydrogenBalmerdecrement. N.G.Gusevaetal.:Balmerjumptemperatureinlow-metallicityHregions,OnlineMaterialp1 Online Material