Constraining the metallicities, ages, star formation histories, and ionizing continua of extragalactic massive star populations
J. Chisholm, J.R. Rigby, M. Bayliss, D.A. Berg, H. Dahle, M. Gladders, K. Sharon
DDraft version May 14, 2019Typeset using L A TEX twocolumn style in AASTeX62
Constraining the metallicities, ages, star formation histories,and ionizing continua of extragalactic massive star populations ∗ J. Chisholm, J.R. Rigby, M. Bayliss, D.A. Berg, H. Dahle, M. Gladders, and K. Sharon University of California–Santa Cruz, 1156 High Street, Santa Cruz, CA, 95064, USA Observational Cosmology Lab, NASA Goddard Space Flight Center, 8800 Greenbelt Rd., Greenbelt, MD 20771, USA MIT Kavli Institute for Astrophysics and Space Research, 77 Massachusetts Ave., Cambridge, MA 02139, USA Department of Astronomy, The Ohio State University, 140 W. 18th Avenue, Columbus, OH 43202, USA Institute of Theoretical Astrophysics, University of Oslo, P.O. Box 1029, Blindern, NO-0315 Oslo, Norway Kavli Institute for Cosmological Physics, University of Chicago, 5640 South Ellis Ave., Chicago, IL 60637, USA Department of Astronomy, University of Michigan, 500 Church St., Ann Arbor, MI 48109, USA
ABSTRACTWe infer the properties of massive star populations using the far-ultraviolet stellar continua of 61 star-forminggalaxies: 42 at low-redshift observed with
HST and 19 at z ∼ (cid:12) and are similar to the measured nebular metallicities. Wequantify the ionizing continua using the ratio of the ionizing flux at 900Å to the non-ionizing flux at 1500Åand demonstrate the evolution of this ratio with stellar age and metallicity using theoretical single burst models.These single burst models only match the inferred ionizing continua of half of the sample, while the other halfare described by a mixture of stellar ages. Mixed age populations produce stronger and harder ionizing spectrathan continuous star formation histories, but, contrary to previous studies that assume constant star formation,have similar stellar and nebular metallicities. Stellar population age and metallicity affect the far-UV continuain different and distinguishable ways; assuming a constant star formation history diminishes the diagnosticpower. Finally, we provide simple prescriptions to determine the ionizing photon production efficiency ( ξ ion )from the stellar population properties. ξ ion inferred from the observed star-forming galaxies has a range oflog( ξ ion ) = . − . − that depends on the stellar population age, metallicity, star formation history,and contributions from binary star evolution. These stellar population properties must be observationallydetermined to accurately determine the number of ionizing photons generated by massive stars. Keywords: binaries: general – dark ages, reionization, first stars – galaxies: abundances – galaxies: starburst –stars: abundances – stars: massive INTRODUCTIONO-stars, which have masses >15 M (cid:12) and lifetimes <10 Myr,are the the only main-sequence stars hot enough to generatea significant number of ionizing photons ( λ < [email protected] ∗ Based on observations made with the NASA/ESA Hubble Space Tele-scope, obtained from the Data Archive at the Space Telescope Science In-stitute, which is operated by the Association of Universities for Research inAstronomy, Inc., under NASA contract NAS 5-26555. state (Strömgren 1939; Seyfert 1943; Baldwin et al. 1991)and chemical evolution (Tinsley 1980) of star-forming galax-ies. The emission lines trace the most recent star formationand measure the rate at which stars form (Kennicutt 1998;Kennicutt & Evans 2012). These observations describe howgalaxies build up their stellar mass (Brinchmann et al. 2004;Noeske et al. 2007; Elbaz et al. 2007) and how star formationevolves with cosmic time (Madau et al. 1999; Madau & Dick-inson 2014). Massive stars impact more than just their hostgalaxies: the ionizing photons produced by the earliest starsmay have been sufficient to reionize the universe (Ouchi et al.2009; Robertson et al. 2013, 2015; Finkelstein et al. 2019).Ionizing photons from massive stars generate the fundamen- a r X i v : . [ a s t r o - ph . GA ] M a y Chisholm et al.tal observables which describe the formation and evolution ofstar-forming galaxies. As such, determining how stars pro-duce ionizing photons is fundamental to understanding galaxyformation and evolution.Stellar ionizing photons are challenging to directly observebecause neutral hydrogen within galaxies efficiently absorbsionizing photons. Nearly all inferences about the flux andspectral shape of the stellar ionizing continua have been madeeither from emission lines that have been reprocessed throughnebular gas adjacent to massive stars, or from the techniqueof stellar population synthesis. Stellar population synthesisconstructs a model stellar spectrum by first determining a hy-pothetical stellar population (with a given age, composition,and star formation history) and then creating a theoreticalspectrum of that stellar population using model stellar atmo-spheres. The stellar age, metallicity and star formation historyare inferred by constructing models with a range of these pa-rameters and using statistical methods to determine whichpopulation values best match the observed spectrum. Largelibraries of rest-frame optical spectra, from surveys such asthe Sloan Digital Sky Survey (Alam et al. 2015), have revolu-tionized population synthesis at optical wavelengths (Bruzual& Charlot 2003; Maraston 2005; Conroy 2013).Stellar population synthesis of the most massive stars can,in principle, constrain the ionizing continua of massive stars.However, massive stars have largely featureless optical spectrathat do not change appreciably with stellar metallicity or age.Therefore, optical stellar population synthesis has a temporalresolution on the order of 10-100 Myr when B-stars, withsignificant Balmer absorption features, begin to appear inoptical spectra. In contrast, the O-stars that produce themajority of the ionizing photons have much shorter lifetimesof 2-10 Myr.The ideal wavelength range to capture the rapid tempo-ral evolution of massive stars is the rest-frame far-ultraviolet(FUV). The FUV contains spectral features of massive stars,namely stellar wind lines (Walborn et al. 1985; Howarth &Prinja 1989; Lamers & Cassinelli 1999; Walborn et al. 2002;Pellerin et al. 2002), which have been observed in star-forminggalaxies over most of cosmic time (Kinney et al. 1993; Heck-man et al. 1998; Pettini et al. 2002; Leitherer et al. 2011; Stei-del et al. 2016; Rigby et al. 2018b). The shape and strengthof these spectral features strongly depend on both the age andmetallicity of the stellar populations (Leitherer et al. 1995,1999; Smith et al. 2002), enabling FUV spectral synthesis todetermine the population properties of massive stars.Both stellar physics and stellar population properties dic-tate the production of ionizing photons. In spectral popu-lation synthesis, the stellar models amass the complicatedunderlying stellar physics of the individual stellar propertiesand evolution which leads to the observed stellar continuum.These vital stellar physical properties include the initial mass function (IMF; Salpeter 1955; Kroupa 2001; Chabrier 2003),stellar rotation (Meynet & Maeder 2000; Levesque et al. 2012;Leitherer et al. 2014), and the stellar evolution tracks that mayinclude interactions among binary stars (Meynet et al. 1994;Leitherer et al. 1995; Eldridge & Stanway 2009; Stanwayet al. 2016). Ultimately, stellar spectral population synthesisis founded upon the individual stellar models. The success orfailure of the stellar synthesis relies upon the models properlyincorporating the crucial stellar physics.Predominately, this paper focuses on using stellar models toconstrain the stellar population properties such as age, metal-licity, and star formation history. We then use these propertiesto infer their ionizing continua. More massive stars must behotter to counteract their intense gravity and remain in hy-drostatic equilibrium. Increased stellar temperatures producebluer spectra and fully ionized stellar atmospheres. Both ef-fects lead to the production of a copious amount of ionizingphotons. These massive stars rapidly exhaust the hydrogenin their cores and have much shorter lifetimes than coolerstars. Consequently, ionizing photons are only produced bythe youngest and most massive stars. Further, since hydrogenis highly ionized in their photospheres, metals are the mainopacity source of ionizing photons in massive stars. Thus,lower metallicity stars produce significantly more ionizingphotons than stars of similar ages but higher metallicities.The stellar age and metallicity must be observationally con-strained to determine the number of ionizing photons gener-ated by massive stars.The 2–10 Myr lifetimes of the most massive stars are only1–10% of the dynamical timescales of galaxies. Thus, therelative proportion of massive stars depends on when starswere formed and how many stars formed at each epoch. Thisis referred to as the star formation history. A starburst galaxyis typically defined as recently forming a large fraction ofthe total stellar mass, thus, their star formation histories aretypically assumed to be nearly a delta-function of a singleburst (McQuinn et al. 2010a). Meanwhile, the entire disksof normal star-forming galaxies have more moderate, nearlyconstant, star formation histories that generate new massivestars at a nearly constant rate (Leitherer et al. 1995). A con-stant star formation history always has a component of youngmassive stars capable of producing ionizing photons that isdiluted by the older population. Nature is unlikely to complywith these simplified star formation histories, and the true starformation histories are assuredly somewhere between thesetwo extremes (McQuinn et al. 2010b). To understand therelative strength of the youngest stellar populations and theirrole in producing ionizing photons, there must be an obser-vationally motivated method to determine the star formationhistory.In this paper, we perform FUV stellar population synthe-sis to constrain the age, metallicity, star formation history,roperties and ionizing continua of extragalactic massive stars populations 3and ionizing continua of extragalactic massive star popula-tions. We fit the non-ionizing FUV continua of a sample of61 low and moderate redshift star-forming galaxies as a linearcombination of single age, fully theoretical stellar continuummodels. We infer the light-weighted ages, metallicities, andthe ionizing continua of the massive star populations fromthese fits. We compare the stellar and nebular metallicities(Section 5.1) and explore the inferred ionizing continua ofthe stellar populations (Section 5.2). The star formation his-tories are derived by comparing the inferred stellar continuumfits to single burst models (Section 5.3). We test the obser-vational differences between populations that contain binarystars (Section 5.4) and illustrate how the stellar continuum fitspredict the total number of ionizing photons (Section 5.7).Throughout this paper we follow the literature conventionand assume that stellar solar metallicity is 0.02 (Leithereret al. 1999, 2010; Stanway & Eldridge 2018b). It is debatedwhether solar abundance is actually higher or lower than thisvalue (Nieva & Przybilla 2012; Villante et al. 2014), but weretain the 0.02 value used in stellar models because it deter-mines the stellar evolution tracks and stellar wind profiles. Wetake the solar gas-phase metallicity to be 12+log(O/H) = 8.69and Z neb (cid:12) = . λ units (erg s − cm − Å − ).The equivalent widths of absorption lines are defined to bepositive; emission lines are defined to be negative. DATA2.1.
Moderate redshift galaxies
MegaSaura data
Here we predominately display spectra of nineteen star-forming galaxies from project MegaSaura: The MagellanEvolution of Galaxies Spectroscopic and Ultraviolet Refer-ence Atlas (Rigby et al. 2018a). The extended MegaSaurasample includes the brightest southern lensed galaxies foundin the Red-sequence Cluster Survey (RCS; Gladders & Yee2005), the Sloan Giant Arcs Survey (SGAS; Bayliss et al.2011), the South Pole Telescope (SPT; Schaffer et al. 2011),and the ESA
Planck survey (Planck Collaboration et al. 2014,2016). These surveys found z ∼ z ∼ > − . < z < . R = − , or 0.5Å at 1500Å) andSNR =
21 (Rigby et al. 2018a). The spectra were correctedfor Milky Way reddening in the observed frame using theCardelli et al. (1989) attenuation curve and the dust mapsfrom Green et al. (2015). We normalized the spectra to themedian of the flux in the line-free region of 1267–1276Å inthe rest-frame. The MegaSaura spectra contain all of thestrong stellar features that constrain the stellar fits at a highSNR with a resolution similar to the stellar continuum models.Due to the superior combination of wavelength coverage andsensitivity, we use the MegaSaura sample as our main sampleinstead of the HST/COS sample introduced below.2.1.2.
Moderate-redshift stacked data
While the individual MegaSaura spectra have high SNR,many of the important stellar features are extremely weak. Byaveraging many observations together (often called ‘‘stack-ing’’), the SNR increases by a factor of √ N , where N is thenumber of spectra included in the stack. Consequently, acomposite provides an average spectrum of an ensemble ofgalaxies at an extremely high SNR (see Section 5.5).This stacking procedure has demonstrated the average FUVspectrum of galaxies at moderate redshifts (Shapley et al.2003; Steidel et al. 2016; Rigby et al. 2018b; Steidel et al.2018). We use two recent stacks: (1) the MegaSauralensed galaxies (Rigby et al. 2018b) and (2) a stack of 30star-forming, field galaxies at z ∼ . = = Low-redshift galaxies
Our low-redshift sample consists of spectra from recentobservations of 2 low-metallicity galaxies (PID: HST-GO-15099, PI: Chisholm) and the compilation from Chisholmet al. (2016) which are 40 local star-forming galaxies at Chisholm et al.0 . < z < . Cosmic Origins Spectrograph (COS; Green et al. 2012)on the
Hubble Space Telescope (HST). The data were com-piled from eight different HST programs, and we include theHST program IDs and references in Table 4. The spectrawere processed through CalCOS v2.20.1, reduced followingthe procedures in Wakker et al. (2015), binned by 20 pixels(0.2Å, or 48 km s − at 1240Å), and convolved to the resolu-tion of the starburst99 models (0.4Å). We normalized thespectra near 1270Å, similar to the MegaSaura observations.These spectra are also corrected for Milky Way reddeningin the observed frame. The COS observations are typicallyonly made with one grating (G130M), such that the averagerest-frame wavelength coverage is from 1150-1450Å. Thisspectral regime contains many, but not all, of the stellar fea-tures that define the stellar age and metallicity (see Section 4).As such, we display the MegaSaura sample throughout thispaper rather than the narrow HST/COS wavelength range.2.3. Host galaxy properties
The 61 galaxies studied here sample a wide range inhost galaxy properties. Table 3 & 4 give literature nebu-lar metallicity values, measured as 12+log(O/H) and referredto as Z neb , which were determined using rest-frame opti-cal emission lines. The ‘‘gold-standard’’ nebular metallic-ity method, the direct method, uses the temperature sensi-tive [O
III ] 4363Å emission line to determine the emission-line emissivities, which directly translates into oxygen abun-dances. However, [O
III ] 4363Å is a weak emission linethat is challenging to observe in faint galaxies and is substan-tially weaker in higher metallicity regions. In the absence of[O
III ] 4363Å detections, calibration techniques have beendeveloped using strong nebular emission lines to infer theionization structure and nebular metallicity. These strong-line abundances are easily observed, however, the inferredabsolute abundances from different calibration methods canbe discrepant by as much as 0.7 dex (Kewley & Ellison 2008).We have used direct metallicities whenever possible, how-ever, the [O
III ] 4363Å emission line is faint and rarely ob-served at high-redshift; accordingly, the MegaSaura Z neb val-ues are calculated using the Pettini & Pagel (2004) [N II ]/H α calibration (table 2 of Rigby et al. 2018a). Meanwhile, opticalspectra of the entire low-redshift COS sample are not publiclyavailable and the literature values have used different strong-line calibrations, precluding a uniform metallicity analysis.Of these, the O3N2 method from Pettini & Pagel (2004),which uses the ([O III ] 5007/H β )/([N II ] 6583/H α ) ratio, isthe most common empirical metallicity calibration used forour sample. We have also calculated 12+log(O/H) using thedirect method for three galaxies in the sample following themethods of Berg et al. (2019). Consequently, 12+log(O/H)(or Z neb ) is not uniformly calculated and may include sys- tematic calibration uncertainties (Kewley & Ellison 2008).The low-redshift galaxies span a factor of 50 in Z neb from0.03–1.5 Z (cid:12) (correspond to 12+log(O/H) = . − . unresolved stellar popula-tion. In other words, each spectrum samples multiple young,UV-bright massive stars. At high redshifts, the physical scaledepends on the lensing magnification, which varies from 2-200 (Sharon et al. 2019), such that the MegaSaura spectraprobe multiple star-forming regions within the same galaxy(see the analysis in Bordoloi et al. 2016). Similarly, COS is afixed circular aperture spectrograph with a 2 . (cid:48)(cid:48) z = . − . STELLAR CONTINUUM MODELING3.1.
Fitting procedure
We fit the stellar continua of both the MegaSaura andlow-redshift samples by assuming that the observed spec-tra are combinations of multiple bursts of single age, singlemetallicity stellar populations. The light from these stellarpopulations then propagated through an ambient interstellarmedium which attenuated the stellar continuum to producethe observed spectral shape. We fit this with a uniform dustscreen model as: F obs ( λ ) = − . k ( λ ) Σ i X i M i ( λ ) , (1)where E(B-V) is the stellar attenuation parameter, k ( λ ) is thereddening curve from Reddy et al. (2016), and X i is the linearcoefficient multiplied by the i th single age stellar populationmodel, M i . Each M i corresponds to a single age and singlemetallicity (Z ∗ ) fully theoretical stellar continuum model (seeSection 3.2). Thus, k ( λ ) , E(B-V), X i and M i completelydescribe the shape and spectral features of the observed stellarcontinuum.We chose the Reddy et al. (2016) attenuation law becauseit is observationally defined down to 950Å, closer to the ion-izing continuum than other models (e.g. Calzetti et al. 2000).This is important because we are predominately interested ininferring the ionizing continua of massive stars. We testedthe effect that changing the attenuation law has on the de-rived stellar properties and found it to most strongly affectthe inferred E(B-V) values which were 0.01 mag redder, onaverage, using the Calzetti et al. (2000) law versus the Reddyet al. (2016) law.roperties and ionizing continua of extragalactic massive stars populations 5We fit the entire observable wavelength regime between1220-2000Å. We did not include wavelengths below 1220Ådue to strong Ly α features and the Ly α forest. The avail-able rest-frame wavelength regime for each galaxy dependson the observational setup and the redshift of the galaxy. Wemasked out ±
500 km s − around strong ISM absorption andemission lines as well as absorption from foreground sys-tems at lower redshifts (Rigby et al. in preparation). Weoptimized this masking velocity interval by studying the indi-vidual ISM features at our spectral resolution. One exceptionis the C IV −
500 to +
50 km s − in order to include the crucial C IV P-Cygni emission (Section 4.1.1). Further, we manually maskedout regions that are not stellar continuum features, such asabnormally large ISM absorption, Milky Way features, orsky-emission lines. We then fit for the X i and E(B-V) of eachsingle age, single metallicity, fully theoretical stellar contin-uum model in Equation 1 using MPFIT (Markwardt 2009).3.2. Stellar models
The theoretical stellar continuum models ( M i ( λ ) ) are keyto the spectral population synthesis. We used both single starmodels (Leitherer et al. 1999, 2010, 2014) and models thatinclude binary evolution (Eldridge et al. 2017; Stanway & El-dridge 2018b) to quantify the effect that binary evolution hason the ionizing and non-ionizing continua of massive stars(Section 5.4). We chose the stellar atmosphere models belowbecause they are the most comparable to each other (Eldridgeet al. 2017) and have the most observationally motivated massoutflow rates (Leitherer et al. 2010). For both models we as-sumed a standard Kroupa initial mass function (IMF; Kroupa2001) with a broken power-law with a high-(low-) mass expo-nent of 2.3 (1.3) and a high-mass cut-off of 100 M (cid:12) . Steidelet al. (2016) demonstrated that the high-mass cut-off weaklyimpacted the spectral synthesis fits to FUV stellar continuaby showing that the best-fit stellar continua did not changedrastically using either a 300 M (cid:12) or 100 M (cid:12) cut-off (theirfig. 7).Star light between 1200-2000Å is dominated by young,massive O-stars. Consequently, we used fully theoreticalstellar models with young ages corresponding to 1, 2, 3, 4, 5,8, 10, 15, 20 and 40 Myr. At 1270Å, a 20 Myr stellar popula-tion of a given initial mass is nearly two orders of magnitudefainter than a 1 Myr stellar population, while older popula-tions are fainter still (Leitherer et al. 1999; Eldridge et al.2017). Moreover, the UV stellar continua of older stellarpopulations evolve more slowly with time, such that there aresmall spectral differences between a 50 and 100 Myr stellarpopulation (fig. 5 in de Mello et al. 2000). Finally, the ef-fective temperature of B-star populations greater than 40 Myrdrops below the 20,000 K threshold where high-resolutionstellar templates are computed using the WM-Basic code (Leitherer et al. 2010). Thus, the selected age range includesmodels of the most luminous O-stars whose spectral featuresvary rapidly with time at sufficient spectral resolution to re-solve these important spectral features.We use the five Z ∗ models that are available from theGeneva stellar atmospheres (Meynet et al. 1994): 0.05, 0.2,0.4, 1.0, and 2.0 Z (cid:12) . Combined, each observed spectrum is fitwith 50 fully theoretical stellar models and one free parameterfor the dust attenuation, for a total of 51 total free parameters.Other stellar physics impact the production of ionizing pho-tons, such as rapid rotation (Levesque et al. 2012; Leithereret al. 2014; Choi et al. 2017) and a varying (or stochasticallypopulated) IMF (Leitherer et al. 1995; Rigby & Rieke 2004;Crowther 2007). However, the stellar population synthesisroutines only include two Z ∗ (0.14 Z (cid:12) and 1 Z (cid:12) , but seethe recent extension to 0.02 Z (cid:12) from Groh et al. 2019), whichsamples metallicity too coarsely for our fitting. Consequently,binary models are the only alternative stellar model that wediscuss below.3.2.1. The fiducial case: starburst99 single-star models
We used the fully theoretical starburst99 models withGeneva atmospheres that incorporate high-mass loss rates(Meynet et al. 1994) as our fiducial model. These modelshave a spectral resolution of 0.4Å, which match the spec-tral resolution of the MegaSaura spectra. We convolved themodels with a Gaussian to match the observed spectral reso-lution of each individual MegaSaura spectrum, as measuredfrom the optical sky emission lines (Rigby et al. 2018a), andresampled the stellar models onto the wavelength grid of theobservations. Similarly, we convolved the higher resolutionHST/COS data to the 0.4Å spectral resolution of the star-burst99 models from the spectral resolution measured fromthe Milky Way absorption lines (Chisholm et al. 2016). Eachmodel spectrum was normalized to the median flux densitybetween 1267–1276Å.The starburst99 stellar models were created using theWM-Basic method (Pauldrach et al. 2001) and densely sam-ple the high-mass portion of the Hertzsprung-Russell diagramup to temperatures of 20,000 K. WM-Basic does not calculatehigh-resolution models below these temperatures (Leithereret al. 2010). Consequently, we chose the ten stellar ages be-tween 1–40 Myr listed above with stellar temperatures greaterthan 20,000 K. These models include Wolf-Rayet (WR) starsusing the Potsdam Wolf-Rayet code (PoWR; Sander et al.2015), but the evolutionary tracks predict that few if any WRstars are present in low-metallicity stellar population, suchthat the WR spectra are rarely incorporated into 0.2–0.4 Z (cid:12) starburst99 models (Leitherer et al. 2018).3.2.2.
The binary evolution case: bpass models
Chisholm et al. A r b i t r a r y F l u x [ F λ ] STARBUST99BPASS
N V O V Si IV He II
Si IIIC IIIC III C III C IV Si III
Starburst99BPASS
Figure 1.
The fully theoretical, rest-frame FUV stellar continuum of five starburst99 (blue) and bpass (red) single burst models with ametallicity of 0.4 Z (cid:12) . Each spectrum shows a different age, with age decreasing from 1 Myr at the top to 15 Myr at the bottom (labeled onthe right). Prominent stellar wind (orange) and photospheric (green) lines are labeled near the models where the features are strongest. Thedisplayed models are a subset of the 50 stellar continuum models that determine the stellar population properties and demonstrate the spectralvariations in the rest-frame FUV with stellar population age.
We also used the Binary Population and Spectral Synthesis(bpass) v2.2.1 models , which include binary star evolution(Stanway & Eldridge 2018b). bpass models have a largermetallicity range, but, for consistency, we used the same fiveZ ∗ available from the Geneva models. bpass models use acustom set of O-star models created with WM-Basic at 1Åresolution for O-stars with temperatures greater than 25,000 K(Eldridge et al. 2017). Temperatures less than this have thebaselv3.1 and C3K models which have spectral resolutionof 20Å below 1500Å (Westera et al. 2002; Le Borgne et al.2003; Conroy & van Dokkum 2012; Conroy et al. 2014).Therefore, bpass models are lower resolution when ages are https://flexiblelearning.auckland.ac.nz/bpass/9.html greater than 20 Myr for any metallicity and when ages aregreater than 15 Myr for metallicities greater than 0.4 Z (cid:12) (seeFigure 1). This spectral resolution is too low to diagnose manyof the narrow B-star features of older stellar populations andcannot be used to distinguish older stellar populations. Forthis reason, we chose the starburst99 models as our fiducialmodel. We return to this issue in Section 5.4.2.3.3. The nebular continuum
Young massive stars produce large amounts of ionizingphotons which produce free-free, free-bound, and two-photonnebular continuum emission. The nebular continuum heavilycontributes to the total continuum flux at young ages, lowmetallicities, and redder wavelengths (Steidel et al. 2016;Byler et al. 2018). For a stellar continuum metallicity ofroperties and ionizing continua of extragalactic massive stars populations 7
Figure 2.
Comparison of the fitted light fractions for each stellar ageused in Equation 1 to determine the fit to the FUV spectra for twogalaxies: RCS Knot E (left panel) and S1527+0652 (right panel).The total light-weighted ages are given as the dashed vertical lines.RCS Knot E has a very young light-weighted age, 2.5 Myr, with allof the light coming from three ages: 2, 3, and 4 Myr. S1527+0652has an older light-weighted age and the fitted models have a mix ofold and young populations. Each bar is color-coded by the stellarmetallicity of the corresponding model. (cid:12) and a stellar age of 1 Myr, the nebular continuum is25% of the stellar continuum at 2000Å.We created a nebular continuum for each age, metallicity,and stellar model by processing the stellar continuum modelsthrough cloudy v17.0 (Ferland et al. 2013). We assumedthat the gas-phase metallicity and stellar metallicity were thesame (Section 5.1), an ionization parameter of log(U) = − . n H =
100 cm − . We produced a nebular continuum foreach stellar population, added the output nebular continua tothe stellar models, and normalized by the flux between 1267–1276Å. The inclusion of the nebular continuum producesredder stellar models than before, which has a pronouncedimpact on the fitted E(B-V) of young stellar populations.We tested the effect that different ionization parametershave on the fitted stellar ages and metallicities by also cre-ating models with log(U) = − − − − χ of the resultant fits do not changestatistically for the different log(U) values. Consequently, weadopted a midrange log(U) = − Stellar population parameters derived from the fits
The observed stellar continua are fit by statistically deter-mining the linear multiplicative coefficient, X i , for the 50single age, single metallicity stellar models ( M i ; Equation 1).These linear coefficients can take any value greater than orequal to 0 and MPFIT determines the linear combination thatbest fits the observed stellar continuum. In practice, the codetypically assigns X i = X i ). Thus, each property derivedbelow is a ‘‘light weighted’’ property. First, the light fraction( f i ) that each model contributes to the total intrinsic flux at1270Å is defined as: f i = X i Σ i X i . (2)Secondly, the light-weighted age at 1270Å is defined as:Age = Σ i X i Age i Σ i X i . (3)The light-weighted ages of RCS Knot E and S1527+0652 areindicated as vertical lines in the left panel of Figure 2. Finally,we computed the light-weighted stellar metallicity as: Z s = Σ i X i Z i Σ i X i . (4)These three parameters describe the properties of the observedstellar populations. The uncertainties on these parameterswere derived by varying the observed flux density at everywavelength by a random Gaussian kernel with width equalto the flux uncertainty at that wavelength. We then recalcu-lated the ages and Z ∗ , tabulated each value, and repeated theprocedure 100 times. The standard deviation of each age andZ ∗ distribution is the uncertainty on the age and Z ∗ , respec-tively. We include the f i , Z ∗ , and light-weighted ages in theelectronic version of the Appendix.All of these stellar population properties (age and metal-licity) are light-weighted at 1270Å; they cannot be directlycompared to similar properties derived at other wavelengths.Younger stars produce relatively more light at bluer wave-lengths than older stars, biasing the light from young starsto bluer wavelengths. The ages derived from full SED mod-eling using optical and near-infrared observations will inher-ently return older ages than we estimated because optical lightcomes from older stars.An important measure of the ionizing continuum is the ra-tio of the intrinsic flux density at 900Å to the flux densityat 1500Å (F /F ). Observations compare the observedF /F to the intrinsic F /F to determine the fractionof ionizing photons that escape galaxies (Steidel et al. 2001).We estimate the intrinsic F /F by extending the stel-lar population fits to bluer wavelengths than the observationsand removing the contributions from dust attenuation (setting Chisholm et al.E(B-V) = 0 in Equation 1). The high-resolution starburst99models are only defined at > , and 1495–1506Å, for F .Ionizing photons with higher energies create high-ionization gas (e.g. O ++ ). We also inferred the stellar fluxdensity between 510–540Å (F ) and between 280–320Å(F ) to determine how many high-energy photons a givenstellar populations produces. F (with photon energies of24 eV) probes photons that singly ionize oxygen, but do notionize helium. Meanwhile, F (photon energies of 41 eV)probes photons that doubly ionize oxygen and singly ionizehelium, but do not doubly ionize helium. These wavelengthswere carefully chosen to probe the peak of the stellar SEDs,while avoiding contributions from strong stellar absorptionand emission features (see Figure 1).Finally, all derived parameters are either flux density ratios(e.g., F /F ) or derived from normalized spectra (e.g.,stellar age). This means that the stellar population parametersare independent of the intrinsic luminosity, which depends onthe magnification from gravitational lensing. RELATING SPECTRAL FEATURES AND INFERREDSTELLAR POPULATION PROPERTIESOften times the ages of stellar populations are deduced us-ing the broadband UV through IR SED shapes. While theSED shape provides important age information, it is often de-generate with metallicity and dust attenuation. By contrast,fitting the stellar spectral features with theoretical stellar tem-plates simultaneously determines the age, metallicity, anddust attenuation of the stellar populations. Consequently, thelight-weighted ages and metallicities derived above are drivenby spectral features which are less degenerate than the spectralshape alone (see Figure 1 and Table 1).The two main types of FUV stellar features are strong, broadstellar wind P-Cygni features and weak stellar photosphericabsorption lines. Both types of features can be contaminatedby neighboring ISM absorption, and require high SNR andmoderate spectral resolutions to resolve. In the followingtwo subsections, we discuss both types of spectral featuresindividually, and illustrate how the individual features relateto the inferred stellar population ages and metallicities. Thepurpose of these subsections is not to advocate for determin-ing the stellar ages and metallicities using single features, butrather to demonstrate that the stellar properties inferred fromthe full spectral fits are entirely consistent with the trends ofstellar spectral features.4.1.
Stellar wind features
Table 1.
Prominent stellar featuresLine bpass Age starburst99 Age[Myr] [Myr]Stellar wind linesN V − − V − − IV −
15 3 − IV −
15 1 − II −
20 3 − III −
15 5 − III − III
III − III − V − −
15 1 − V − −
15 1 − V − − IV − −
10 3 − II − III − −
40 10 − (cid:12) bpass and starburst99 model, respectively. The most notable stellar features in the FUV are thebroad blueshifted absorption and redshifted emission profiles(called P-Cygni profiles; orange labels in Figure 1). TheseP-Cygni profiles arise from strong winds that are radiativelydriven off of stellar photospheres (Castor et al. 1975; Lamers& Cassinelli 1999). The terminal velocity, ionization struc-ture, and mass-outflow rates sensitively depend on the stellarluminosity (Castor et al. 1975; Lamers & Leitherer 1993;Lamers et al. 1995; Leitherer et al. 1995; Puls et al. 1996;Kudritzki & Puls 2000). The terminal velocity describes themaximal velocity extent of the absorption component, whilethe mass-loss rate determines the depth of the profile. Inturn, these establish the shape of the P-Cygni absorption andemission.Whether a given ion is observed as a P-Cygni profile inthe wind depends on the ionization structure of the stellarwind. The peak ionization stages for stellar winds typicallyare the C V , N IV , O IV , and Si V states (Lamers & Cassinelli1999), none of which have resonant transitions in the rest-frame FUV. Alternatively, the presence of adjacent ionizationstages with P-Cygni profiles (N V , C IV , or Si IV ) providesinformation on the stellar temperature of the most luminousstars and, by inference, the stellar population age. The age ofa stellar population can be inferred from the strength of theobserved P-Cygni transitions: O V and N V have the highestionization states and are strongest in stars with lifetimes ofroperties and ionizing continua of extragalactic massive stars populations 92-3 Myr, while C IV and Si IV are lower ionization and peakin stars with lifetimes near 5 Myr.Stellar temperature, or age, is not the sole determinant of thestellar wind profiles. Metals in the photospheres of hot starsabsorb continuum photons which is what accelerates the gasoff the stellar surface. Consequently, the stellar metallicitydetermines both the acceleration and mass outflow rate of thestellar wind (Lamers & Cassinelli 1999; Vink et al. 2001).The terminal velocities and mass-loss rates of O-stars with > . (cid:12) have been empirically determined to scale as Z . ∗ andZ . ∗ , respectively, (Leitherer et al. 1992; Vink et al. 2001).Lower metallicity stellar winds of resolved individual starshave not been observed, consequentially the mass outflowrate and terminal velocity relations have been extrapolated tolower metallicities. These relations illustrate how stellar windP-Cygni absorption profiles scale with stellar metallicity.In the next five sub-sections we walk through the individ-ual P-Cygni lines in the FUV. Each subsection explores thetheoretical and observed wind features and their relationshipto the inferred stellar population age and metallicity. In Sec-tion 4.3 we conclude that the N V P-Cygni and He II emissionare strong in very young stellar populations, while the shapeof the C IV P-Cygni profile changes both with stellar age andmetallicity (see Figure 3). Conversely, the Si IV of our sam-ple is dominated by interstellar absorption, and the O V lineis not observed. Collectively, the stellar wind profiles mimicthe inferred stellar ages and metallicities.4.1.1. The C IV P-Cygni feature
The C IV feature is strong, broad, and has a P-Cygni profilefor all of the MegaSaura galaxies (there is not C IV cov-erage for many of the low-redshift COS spectra). Further,the C IV P-Cygni profile is sufficiently broad that stellar andinterstellar components can easily be separated with moder-ate spectral resolution (see shaded regions in Figure 4 andFigure 5). The C IV profile probes stellar outflows from 1-10 Myr populations and the absorption component distinctlyvaries with Z ∗ (Figure 3). This makes it an ideal diagnosticof stellar age and metallicity.Different portions of the C IV profile depend on eitherZ ∗ or stellar age (left panels of Figure 3). At a constant5 Myr age (upper left panel), the C IV absorption of thestarburst99 models deepen and broaden as Z ∗ increasesfrom 0.4 Z (cid:12) (blue line) to 1 Z (cid:12) (gold line) to 2 Z (cid:12) (pinkline), because the C IV mass-loss rate strongly increases withZ ∗ . Conversely, the models predict that the C IV emissionis nearly independent of Z ∗ . Meanwhile, at a constant 1 Z (cid:12) metallicity (lower left panel), the modeled C IV absorption isindependent of stellar age, but the emission strongly peaks atages <5 Myr (blue line). In summary, the C IV absorptionvaries with Z ∗ and the C IV emission varies with stellar age. Constant Age: 5 MyrConstant Z * : 1 Z ☉ ☉ ☉ ☉ ☉ ☉ ☉ N o r m a li z ed F l u x N o r m a li z ed F l u x Figure 3.
The dependence of the C IV V ∗ ) of 2 (red), 1 (gold), and0.4 Z (cid:12) (blue). The bottom two panels show three starburst99models with a constant Z ∗ = 1 Z (cid:12) but with varying stellar ageof 8 (red), 5 (gold) and 3 Myr (blue). Z ∗ does not change theN V profile (upper right panel), but the emission profile stronglypeaks for ages <5 Myr (lower right panel). Z ∗ strongly impactsthe C IV absorption (wavelengths less than 1550Å), but weaklyimpacts the C IV emission (upper left panel). Conversely, the stellarage strongly impacts the C IV emission (wavelengths greater than1550Å) but weakly impacts the C IV absorption (lower left panel).By eye, the C IV and N V profiles of S0033+0242 are best-fit betweenthe blue and gold profiles in each panel. This is consistent with thelight-weighted age and metallicity of 5 ± . . ± .
04 Z (cid:12) . The temporal evolution of C IV found in the models isclearly corroborated by the light-weighted ages. Figure 4shows three MegaSaura spectra ordered in descending light-weighted stellar age and each have similar light-weightedmetallicities (0 . ± .
06 Z (cid:12) ). Overplotted in red is a sin-gle age starburst99 model with a population age nearestto the inferred light-weighted age. The C IV emission isstrongest in the youngest populations and adequately matches0 Chisholm et al. ISM absorption
Si II*
RCS Knot ERCS Knot GS1226+2152 N o r m a li z ed F l u x Figure 4.
The dependence of the C IV wind profile on stellarage. Each panel shows an observed galaxy spectra (black line)ordered by descending fitted light-weighted ages of 2 . ± .
1, 11 ± .
8, and 26 ± . . ± .
04 Z (cid:12) ). The colored curves are theoretical models. A single age,0.2 Z (cid:12) starburst99 theoretical model with an age nearest to thefitted light-weighted age is overplotted in red. The multiple age,multiple metallicity bpass (blue) and starburst99 (gold) fits arealso shown. The single age models describe the overall shape ofthe younger C IV profiles (e.g. RCS Knot E and G), while themultiple age fits are required to capture the full details of the olderprofiles (e.g. S1226+2152). Gray regions correspond to interstellarC IV absorption and Si II ∗ emission. the triangular wind emission from the RCS Knot E spectrum.Meanwhile, RCS Knot G, a different region from the samegalaxy, has weaker P-Cygni emission (although note the nar-row nebular C IV emission), similar to an 8 Myr single agepopulation. Finally, S1226+2152 has an even older inferredlight-weighted age; the C IV P-Cygni feature has nearly disap-peared and has been replaced by a flat C IV feature with nar-row, interstellar absorption. The derived light-weighted agesof each spectrum are consistent with the by-eye P-Cygni vari-ation: Knot E, Knot G, and S1226+2152 have starburst99light-weighted ages of 2.5, 11, and 26 Myr, respectively.While the single age models reflect the C IV profiles of theyoungest populations, they do not describe all of the galaxy- ISM absorption
Si II* N o r m a li z ed F l u x Single age 0.2 Z ☉ SB99 multiple ageBPASS multiple ageSingle age 0.4 Z ☉ SB99 multiple ageBPASS multiple ageSingle age 1.0 Z ☉ SB99 multiple ageBPASS multiple age
RCS Knot U(0.29 Z ☉ )S0033+0242(0.84 Z ☉ )Sunburst Arc(0.55 Z ☉ ) Figure 5.
The dependence of the C IV stellar wind profile on stellarmetallicity (Z ∗ ) at nearly constant stellar age (4 ± ∗ of 0 . ± .
07, 0 . ± .
04, and 0 . ± .
04 Z (cid:12) . TheC IV stellar absorption increases with increasing Z ∗ . The coloredcurves correspond to various models. A 5 Myr starburst99 modelis shown in red in each panel with Z ∗ closest to the inferred light-weighted Z ∗ . The multiple age, multiple metallicity bpass (blue) andstarburst99 (gold) fits are also shown. At these young ages, thesingle age models describe the overall shape of the C IV lines. Grayregions correspond to interstellar C IV absorption and Si II ∗ emis-sion. to-galaxy P-Cygni variations in the older populations. TheC IV absorption from RCS Knot G is deeper at bluer velocitiesthan the 8 Myr single age model, while S1226+2152 has asmall amount of redshifted emission. The observed stellarpopulations are not single age populations, but the multipleage fits capture the different populations (blue and gold linesin Figure 4 are the starburst99 and bpass fits). While asingle age population largely describes the C IV emissionfrom the youngest populations, older populations require amix of both young and old stars to match the observations.Metallicity has a similarly strong impact on the shape of theobserved C IV profile, but this time on the absorption portionof the P-Cygni profile. The upper left panel of Figure 3 showsthat Z ∗ mostly impacts the depth and width of the absorptionroperties and ionizing continua of extragalactic massive stars populations 11 N o r m a li z ed F l u x N o r m a li z ed F l u x C IIIC III C IIIC IIISunburst Arc (3 Myr, 0.5 Z ☉ ) S0033+0242(5 Myr, 0.8 Z ☉ )S1429-1202(15 Myr, 0.6 Z ☉ ) S1527+0652(21 Myr, 0.5 Z ☉ )Starburst99BPASS Broad Ly a Figure 6.
The observed N V V windprofile transitions from a strong P-Cygni profile in the upper left toa nearly flat continuum in the lower right as the light-weighted agedeclines. The four stellar populations have similar Z ∗ but differentlight-weighted ages, emphasizing that the N V line chiefly dependson stellar age (see the right panel of Figure 3). The starburst99(blue) and bpass (gold) fits are included in each panel. The dashedgray line denotes the C III component. This is further illustrated in Figure 5 which showsthree MegaSaura spectra with nearly constant stellar age (5,3, and 5 Myr) but with increasing stellar metallicity. At0.3 Z (cid:12) (top panel), the C IV absorption only reaches a depthof 0.6 in normalized flux units. Increasing the metallicity to0.6 Z (cid:12) (middle panel) creates a pronounced, broad P-Cygniabsorption profile that reaches 0.5 in normalized flux units.Finally, by 0.8 Z (cid:12) (bottom panel) the stellar wind dominatesthe C IV spectral region and the absorption reaches 0.4 innormalized flux units. The stellar emission does not stronglychange with Z ∗ , as the stellar emission peaks near 1 . ± . IV P-Cygni profile.To summarize: the C IV profile strongly varies with theinferred light-weighted stellar population properties. The C IV P-Cygni absorption depends on the Z ∗ and the emissiondepends on the stellar age (Figure 3). The multiple age andmultiple metallicity fits to the C IV P-Cygni profiles mimicchanges in the inferred stellar age (Figure 4) and metallicity(Figure 5). 4.1.2.
The N V P-Cygni feature
The N V +++ , but as thestellar temperature increases with decreasing Z ∗ , N +++ gas isheated into the N + state. This heating produces relativelymore gas in the N + ionization state for low Z ∗ winds thanhigher Z ∗ winds and nearly balances the decreasing metal-licity (Kudritzki 1998; Lamers & Cassinelli 1999; Leithereret al. 2010). The models in the upper right panel of Figure 3show the negligible N V variation with Z ∗ , while strong ab-sorption and emission only occurs at ages young enough toproduce N V stellar winds (bottom right panel of Figure 3).N V only arises from very young (<5 Myr) stellar populations.A strong N V profile is detected in most FUV spectra inFigure 6. The exceptions are the oldest stellar populationswhich do not show P-Cygni profiles in either the C IV or theN V ionization state (e.g. S1527+0652). Weak C IV and anon-detection of N V indicates that there is not currently ayoung (<8 Myr) stellar population in these older galaxies (seethe discussion in Section 5.3). In contrast, the normalizedflux of the N V profile from the Sunburst Arc varies by afactor of 2.5 from the absorption depth to the emission peak,illustrating the strong N V P-Cygni profiles in populationswith very young light-weighted ages. This trough-to-peakratio decreases with increasing inferred stellar population agein Figure 6: it is 2.2 for S0033+0242 with an age of 5 Myrand 1.4 for S1429-1202 with an age of 15 Myr.4.1.3.
The Si IV P-Cygni feature
P-Cygni Si IV is generally observed in supergiants (evolvedstars) or in metal-rich stars because the dominant ionizationstate in main-sequence stellar winds is the Si + state. Thus,the Si +++ ionization state is too low of a temperature to tracethe bulk of the O-star wind (the opposite behavior as N V above; Walborn & Panek 1984). As O-stars evolve into super-giants, their stellar photospheres expand and the stellar windsbecome denser, causing the dominant ionization state, Si + ,to recombine into Si +++ . These denser winds produce promi-nent Si IV P-Cygni features in the spectra of evolved O-stars(Drew 1989; Pauldrach et al. 1990). Similarly, the ionizationstructure also shifts towards lower ionization stages as largerstellar metallicities produce relatively more Si +++ in their stel-lar winds. Thus, the stellar models predict that Si IV P-Cygniprofiles are only strong in stellar populations with a lowerionization structure, either from higher metallicity stars or anout-sized contribution of evolved stars.2 Chisholm et al. N o r m a li z ed F l u x + C on s t an t Starburst99 BPASS Sunburst ArcS1226+2152S0108+0624Cosmic Eye(29Myr)(26 Myr)(7 Myr)(3 Myr)
Si IV 1393 Si IV 1402
Figure 7.
The Si IV IV spectral regime, are shadedin gray. An 8 Myr single age model is included over the S0108+0624spectra in light-blue to demonstrate that bpass models predict strongP-Cygni profiles at these ages, while the starburst99 models donot. Strong Si IV IV is rarely observed to have a stellar P-Cygni profile. Theentirety of the observed Si IV absorption can be explained bynarrow interstellar absorption that reaches nearly zero flux.After accounting for interstellar absorption (gray regions inFigure 7), the observed Si IV region is nearly featureless.The lack of strong P-Cygni Si IV suggests that the stellarpopulations within our sample are either metal poor (see Sec-tion 5.1) or not dominated by evolved stars. We conclude thatthe observed Si IV regions are dominated by interstellar ab-sorption and that Si IV does not strongly vary with the stellarpopulation properties.4.1.4. The He II emission feature He II II emissionlines in both the optical and FUV that are strongest in highermetallicity stars (Schaerer & Vacca 1998).He II in Figure 8 is not observed as a broad P-Cygni pro-file like N V or C IV . For the oldest light-weighted popu-lations, He II is a weak absorption line (equivalent widthof 0 . ± .
04Å in S1226+2152). Conversely, the youngestpopulations (Sunburst Arc, RCS Knot E, S0033+0242, andRCS Knot U) have a broad triangular shaped He II emissionprofile (see the arrows in Figure 8). The He II emission re- N o r m a li z ed F l u x + C on s t an t Starburst99BPASSSunburst ArcS1226+2152S0108+0624Cosmic Eye(29 Myr)(26Myr)(7 Myr)(3 Myr)
He II 1640
Figure 8.
The observed He II II region better than the starburst99models. However, neither model adequately fits J0108+0624 northe Sunburst Arc (bottom two spectra). J0108+0624 has narrowemission (96 km s − ) that is likely nebular in origin, while theSunburst Arc has broad emission (379 km s − ) that is likely stellar inorigin. The arrows above the Sunburst Arc’s spectrum emphasize thebroad emission feature. The other young populations have similarlybroad He II emission. sembles the redshifted triangular C IV emission profile butwithout the blueshifted absorption component that createsthe P-Cygni profile. There is not He II absorption becauseHe II α ), therefore the stellar wind is optically thin to theHe II recombination emission. The He II emission equiv-alent width from the Sunburst Arc spectrum is − − , consistent with a 3 Myr, 0.5 Z (cid:12) Wolf-Rayet model (Schaerer & Vacca 1998). The presenceof a broad He II emission profile strongly suggests that thespectrum is dominated by a < > σ ), narrow He II emission (equiv-alent width of − . ± .
08Å and FWHM = 96 km s − , whichis resolved by the 69 km s − spectral resolution). This nar-row emission appears nebular in origin when compared tothe broad Wolf-Rayet feature of the Sunburst Arc (comparethe bottom and second-to-bottom spectra in Figure 8). ThisHe II emission is at the weak end of the range of He II equiv-alent widths seen in local dwarf galaxies ( − . − . II emission in Section 5.4.4.The multiple age fits to the He II region are fairly poor. Thisis especially true for galaxies with strong WR emission (e.g.the Sunburst Arc). The bpass models fit the He II region sub-stantially better than the starburst99 models do, althoughroperties and ionizing continua of extragalactic massive stars populations 13they still do not match the WR features of younger popula-tions. While both stellar models include WR models, neithermodel appears to produce a sufficient number of WR starsto match the broad He II emission observed in the youngeststellar populations (e.g., Leitherer et al. 2018).Overall, we find broad Wolf-Rayet He II emission in stellarpopulations with the youngest light-weighted ages and weakHe II absorption in older populations.4.1.5. The O V wind feature With an ionization potential of 117 eV, O V K). This feature is an unambiguous indicator of theyoungest stars and disappears once the stellar population agespast 3 Myr (see Figure 1 and Table 1).While O V is an ideal tracer of the most massive stars, itis not strongly observed in any of the MegaSaura or COSspectra. This may indicate that there is not a significantextremely massive stellar population ( ∼
300 M (cid:12) ), but morelikely it indicates that the winds of the most massive starsare significantly clumpier–thus denser–than assumed in themodel atmospheres. Denser, clumpier winds lead to lowerionization stellar winds and reduced O V profiles (Bouretet al. 2003; Leitherer et al. 2010).4.2. Photospheric absorption lines
The broad, pronounced stellar wind features, fully dis-cussed in the previous section, arise as intense radiation fieldsaccelerate gas off of stellar surfaces. However, gas within thestellar photospheres also absorbs stellar radiation. These pho-tospheric absorption lines are narrower and weaker than thestellar wind lines because the photosphere is relatively static,but provide similarly robust indicators of stellar age and Z ∗ (de Mello et al. 2000). Since photospheric gas is denser thaninterstellar gas, photospheric lines typically arise from excitedstates that are easily separated from ISM lines.High-ionization photospheric lines, like S V III andSi
III photospheric lines at 1247Å (see Figure 6) and be-tween 1295–1299Å (specifically C
III N o r m a li z ed F l u x S1226+2152(26 Myr)S0900+2234(7Myr)Sunburst Arc(3 Myr)
ISM absorption
O I Si II
Figure 9.
A set of photospheric absorption lines (1290–1300Å)from three galaxies: S1226+2152 (top panel), S0900+2234 (middlepanel), and the Sunburst Arc (bottom panel). The four photosphericabsorption lines are marked by dashed gray lines. These lines origi-nate in the photospheres of B-type stars with ages greater than 8 Myrand strengthen with increasing light-weighted age (printed under-neath each name). Gray regions indicate possible ISM absorptionfrom O I II versus 10-30 eV for the features near 1420Å; Leitherer et al.2011). The 1290Å lines blend with neighboring O I andSi II ISM absorption features at low spectral resolution, how-ever, at moderate resolution the Si
III
III strengthens in stellar pop-ulations older than 5-10 Myr. The Si
III . ± .
02Å for S1226+2152 with anage of 26 Myr, to 0 . ± .
15 for S0900+2234 with an age of7 Myr, to undetected (0 . ± . III . ± .
21, 0 . ± .
07, and 0 . ± . III
Figure 10.
The change in flux ( ∆ F λ ) incurred by increasing theage from 2 Myr to 8 Myr of a 0.2 Z (cid:12) population (gold line) orincreasing the stellar metallicity from 0.05 Z (cid:12) to 0.4 Z (cid:12) of a 5 Myrpopulation (blue line). The strong spectral features discussed in thetext are labeled, but significant continua changes occur outside ofthese spectral features in "featureless" regions (see Table 2). There are multiple Fe photospheric lines in the FUV thatarise from the Fe
III , Fe IV , and Fe V ionization states (Ta-ble 1; Nemry et al. 1991; de Mello et al. 2000; Rix et al.2004). These Fe lines (specifically Fe V in O-stars) blanketthe spectral regions, overlap in wavelength, and form large-scale continuum features that change the shape of the stellarcontinuum. These Fe features are included in the full spectralfitting, however, even at the high SNRs of the MegaSaurasample, the Fe lines are very weak.The power and utility of the stellar photospheric linesis limited because the photospheric lines are significantlyweaker than the stellar wind features, requiring extremelyhigh SNR >
20 observations. At the SNR and spectral reso-lution of our observations, the most prominent photosphericlines are near 1299Å and these features indicate the presenceof stars with ages > Summary: Individual spectral features are consistentwith inferred stellar ages and metallicities
Throughout this section we have compared the light-weighted stellar population properties (age and metallicity)to individual stellar features, but in all cases the inferredstellar properties were determined from the entire observedwavelength regime. In Figure 10 we plotted the change influx density ( ∆ F λ ) that occurs due to varying the stellar age(gold) or metallicity (blue). The largest ∆ F λ arises from thestrong stellar wind and photospheric lines that we emphasizedabove. We demonstrated this by integrating the (cid:113) ∆ F λ withinspecific wavelength regions to determine the total flux changeattributable to individual spectral regions (Table 2). The N V Table 2.
The integrated (cid:113) ∆ F λ incurred when changing stellarpopulation propertiesRegion Spectral Feature within Region ∆ Age ∆ Z (cid:12) V III IV IV (cid:12) starburst99 modelfrom 2 Myr to 8 Myr (gold line in Figure 10) integrated over the re-gion in Column 1. Column 4 gives the same value but for increasingthe stellar metallicity from 0.05 Z (cid:12) to 0.4 Z (cid:12) for a 5 Myr population(blue line in Figure 10). profile changes almost exclusively with age and is weakly de-pendent on Z ∗ . Similarly, the Si IV region largely depends onZ ∗ and hardly depends on stellar age (compare Column 3 and4 in Table 2). While the notable stellar wind features dom-inate Figure 10, small spectral features occur outside theseregions that on aggregate differentiate between stellar proper-ties. For instance, the region between 1420–1520Å is devoidof prominent stellar wind features, but the shape of the stellarcontinuum noticeably changes over 100Å due to the stellarpopulation age and Z ∗ . The 1420–1520Å region changes theflux comparably to the broad C IV P-Cygni wind feature andthe Si
III IV wind region.This implies that "featureless" regions of stellar continuumhold significant power to determine stellar age and Z ∗ . Thepower in these regions stems from the fact that they are notfeatureless, but crucially contain weak photospheric metalfeatures like Fe III , Fe IV , and Fe V (Nemry et al. 1991; deMello et al. 2000; Rix et al. 2004). In other words, the pre-vious sections were not advocating to use a single feature todetermine the stellar properties, rather the sections illustratedthat the spectral features consistently and distinctly vary ac-cording to the inferred stellar population properties from thefull stellar continuum.Young stellar populations produce the most ionizing pho-tons. Thus, it is important to have spectral features thatdistinguish immense sources of ionizing photons. There arespectral indices that are unique to very young (<7 Myr) stel-lar populations: (1) strong and broad C IV and N V P-Cygnifeatures (Figure 3), (2) broad (>300 km s − ) He II emission(Figure 8), and (3) undetected Si III
Figure 11.
Comparison of the light-weighted stellar metallicity(Z ∗ ) derived from the starburst99 fitting to the gas-phase nebularmetallicity (Z neb ). The points are color-coded according to the threesamples: blue are low-redshift galaxies, orange are the individual z ∼ neb was calculated using the direct method (circles) or a strong-linecalibration (squares). Note, only 8 of the 19 MegaSaura galaxieshave a measured Z neb . Two error bars, one at 0.3 Z (cid:12) and one at1.2 Z (cid:12) , are shown in the bottom of the plot using the median Z ∗ error and the 0.4 dex 12+log(O/H) spread between Z neb diagnosticssuggested by Kewley & Ellison (2008). A one-to-one line is shownin gray: Z ∗ is largely equivalent to Z neb . spectral signatures suggest that the stellar population is dom-inated by stars younger than 7 Myr. Older stellar populationsappear to have combinations of both young and old stars,obscuring single trends in spectral features (see Section 5.3). DISCUSSION5.1.
Nebular gas and massive stars have similar metallicities
The stellar metallicity is a fundamental galaxy propertythat is traditionally challenging to measure, but crucially de-termines the production of ionizing photons. Previous ob-servations have either used spectral indices (e.g., Rix et al.2004; Leitherer et al. 2011; Byler et al. 2018) or full spectralsynthesis (e.g., Pettini et al. 2000; Kudritzki et al. 2012; Stei-del et al. 2016, 2018; Hernandez et al. 2019) to estimate themetallicities of young stellar populations. Here, we explore the relation between the light-weighted Z ∗ derived from thefull spectral synthesis to the observed gas-phase metallicitiesderived from rest-frame optical nebular emission lines (Z neb ).Figure 11 shows that the light-weighted Z ∗ and the Z neb arecorrelated at the 8 σ significance (with a Pearson’s correlationcoefficient of 0.87). The metallicities scatter about the one-to-one line, indicating that the young massive stars have asimilar metallicity as the surrounding nebular gas. Both thelow and high-redshift galaxies are near the one-to-one line;this relationship does not appear to evolve with redshift. Thereis a dearth of points above 1 Z (cid:12) , largely as a selection effectof both samples. While we have used direct metallicitieswhenever possible (see the discussion in Section 2.3), wehave used different metallicity calibrations which can have a0.4 dex scatter in their relative metallicity calibrations. Thelargest outlier from a one-to-one relationship is NGC 3256,in the lower right quadrant with Z ∗ = 0 . ± .
13 Z (cid:12) andZ neb = 1 . ± .
25 Z (cid:12) (including calibration uncertainties;Engelbracht et al. 2008). Consequently, the Z neb of the largestoutlier is still consistent with Z ∗ at the 1.3 σ significance level.Further, we measure the residual standard error of the trendto be 0.16 Z (cid:12) , or 0.17 dex in 12+log(O/H) at the median Z neb of 0.4 Z (cid:12) . This residual standard error is consistent withthe full 0.4 dex spread found by Kewley & Ellison (2008)between the Pettini & Pagel (2004) and Kobulnicky & Kewley(2004) metallicity calibrations (the two most commonly usedstrong-line methods here). Thus, the observed dispersion inFigure 11 is entirely consistent with the spread of the different12+log(O/H) calibration methods.Z ∗ is surprisingly consistent for the lowest metallicity popu-lations. The stellar mass-loss rate, and in turn the stellar windprofile, is only empirically constrained for Z ∗ > 0.2 Z (cid:12) and the0.05 Z (cid:12) starburst99 models instead rely on an extrapolationof the mass-loss rates to lower metallicities (Leitherer et al.1992; Vink et al. 2001). The close metallicity correspondencebelow 0.2 Z (cid:12) suggests that the stellar wind extrapolation ad-equately reproduces the observed stellar wind profiles andtheir mass-loss rates. Z neb for 1 Zw 18 is lower than 0.05 Z (cid:12) ,the lowest Meynet et al. (1994) atmospheric model, leadingto a possible over-estimate of the light-weighted Z ∗ for thisgalaxy.The strong relationship between Z neb and Z ∗ in Figure 11suggests that Z ∗ robustly determines the metallicities of galax-ies. A robust metallicity indicator is need at high-redshiftsas the crucial, yet extremely faint, metallicity sensitive lines,like [O III ] 4363Å, are redshifted out of the optical (Yuan& Kewley 2009; James et al. 2014a; Sanders et al. 2016).Recent work has attempted to calibrate Z neb using rest-frameUV emission lines (Pérez-Montero & Amorín 2017; Byleret al. 2018), but these calibrations depend on rather uncertainassumptions of the carbon-to-oxygen abundance that are notyet well calibrated. Fortunately, Figure 11 implies that ob-6 Chisholm et al.
200 400 600 800 1000 1200 1400 1600Rest Wavelength [Å]24681012 O ff s e t F l u x [ F λ ] Starburst99BPASS 𝑭 𝑭 𝑭 𝑭 𝑭 𝑭 𝑭 𝑭 𝑭 𝑭 𝑭 Figure 12.
The ionizing continua of five different single age, fully theoretical stellar models with a constant metallicity of 0.4 Z (cid:12) . We plot bothstarburst99 single star models (blue lines) and bpass models which include binary star evolution (red). The stellar populations are orderedby descending age, with the youngest (2 Myr) population at the top and the oldest (15 Myr) population at the bottom. Each age is offset by aconstant for presentation purposes. Each population is normalized by the flux density at 1500Å (F ; dashed line) such that zero flux is givenby the dot-dashed line. The gray regions highlight the three regions that we measure the ratio of the ionizing to non-ionizing flux density toquantify the shape of the ionizing continuum: F /F , F /F , and F /F (the flux density at 300, 525, and 900Å relative to theflux density at 1500Å). F /F steadily decreases as the stellar population ages. servations from the ground using upcoming large telescopesand the rest-frame FUV stellar continuum may constrain thechemical enrichment of galaxies to redshifts of z ∼ neb and Z ∗ indicates that the gassurrounding high-mass stars is not instantaneously metal-enriched by massive stars. In other words, it takes longerthan the inferred lifetimes of the massive star populations toincrease the metallicity of adjacent ISM gas. Newly synthe-sized metals are ejected from star-forming galaxies as veryhot supernovae ejecta and galaxies must take longer than the10 yr timescales observed here to fully mix these newly syn-thesized metals (Kobulnicky & Skillman 1997). In fact, wedid not find a statistical correlation between Z neb and the stel-lar age, implying that gaseous enrichment either occurs on longer timescales than observed here, or that the enrichmentof a single stellar population is within the scatter of Figure 11.5.2. The ionizing continua of massive stars
The non-ionizing FUV continua of massive stars are richwith features that diagnose the stellar population age andmetallicity. The primary goal of this paper is to use theseinferred light-weighted metallicities and ages to constrain theionizing continua of massive stars. The same massive starsthat produce the FUV spectral features also produce the un-seen ionizing continua; these spectral features are the mostdirect probe of the ionizing continua because they only dependon the massive star properties (unlike nebular emission lineswhich also depend on the gaseous properties). We thereforeuse the theoretical stellar models to develop simple prescrip-roperties and ionizing continua of extragalactic massive stars populations 17
Figure 13.
The temporal variation of the ratio of the ionizing to non-ionizing flux density at three ionizing wavelengths (900, 525, and 300Åfrom left to right) of a theoretical, 0.4 Z (cid:12) , single burst stellar population. We include both starburst99 (blue squares) and bpass models (goldcircles). Young stellar populations emit more ionizing photons than non-ionizing photons. The flux ratio of a bpass model with a constant starformation law and a 0.05 Z (cid:12) metallicity is included as a dashed red line.
Figure 14.
The variation of the ionizing to non-ionizing flux density ratios with stellar metallicity (Z ∗ ). The three panels show these ratios atthree wavelengths (900, 525, and 300Å from left to right) for a theoretical 5 Myr single burst stellar population. We include both starburst99(blue squares) and bpass models (gold circles). Low metallicity bpass models produce significantly more ionizing photons, especially at 300Å,due to the effects of binary star evolution (Section 5.4.1). tions for how the ionizing continuum varies with stellar prop-erties (Section 5.2.1) and then infer the ionizing continua fromthe FUV observations using the stellar fits (Section 5.2.2). Werefer to the ionizing continua determined from the models asthe inferred ionizing continua to emphasize that we do notdirectly observe the ionizing continua. Throughout this sub-section we only focus on the starburst99 fits; we return tothe bpass fits in Section 5.4.5.2.1. Stellar population properties determine the ionizingcontinua of massive stars
The full ionizing continuum will likely never be observedbecause foreground neutral hydrogen efficiently absorbs thesephotons. Observations are typically fortunate to observe theionizing continuum at a single wavelength, which is most feasible at 900Å. This ionizing flux density is then normal-ized by the observed non-ionizing continuum to control forstellar mass and star formation contributions. Thus, the lit-erature typically quantifies the ionizing continuum throughthe ratio of the flux at 900Å (ionizing) to the flux at 1500Å(non-ionizing), or F /F (Steidel et al. 2001). In thissubsection, we use the starburst99 fully theoretical modelsto explore how the stellar properties determine this ratio aswell as flux density ratios at other ionizing to non-ionizingwavelengths.Figure 12 shows the theoretical ionizing continua of singleburst, 0.4 Z (cid:12) , starburst99 stellar populations at five differ-ent ages. At 2 Myr the intrinsic ionizing flux density at 900Åis actually 1.7 times more luminous than the non-ionizingflux density at 1500Å. This is because the blackbody spec-8 Chisholm et al.trum of a 43,000 K object, or a 40 M (cid:12) star, peaks at 670Å.F /F steadily decreases as the stellar population ages,from 1 at 3 Myr to 0 at 15 Myr. Clearly, the stellar age cru-cially determines the stellar ionizing flux density of singlebursts.We use three different ionizing to non-ionizing flux den-sity ratios to quantify the shape of the ionizing continuum(F /F , F /F , F /F ). As hinted in Figure 12,the left panel of Figure 13 shows a smooth temporal trendof F /F with age at a fixed metallicity. Higher energyphotons (probed by the flux density ratio of F /F ) havea similarly strong temporal evolution (middle panel of Fig-ure 13), but shifted to earlier ages, such that only stars youngerthan 5 Myr are sufficiently hot to emit appreciably at 525Å.The ionizing continua of the youngest stars peak near 525Åwhere F is three times more luminous than F . Finally,the highest energy photons (probed by the flux density ratioof F /F ) are exclusively produced by stellar populationswith ages less than 2 Myr (right panel of Figure 4). The fluxdensity ratios of theoretical starburst99 models show thatthe stellar age correlates with the shape and strength of theionizing continua.The stellar metallicity (Z ∗ ) also impacts the shape of theionizing continuum as measured by the flux density ratios(Figure 14; Smith et al. 2002). At a constant stellar age of5 Myr, the F /F increases by a factor of 3 as Z ∗ decreasesby a factor of 40 (left panel of Figure 14). Similarly, F /F and F /F decrease by a factor of three and twelve fromZ ∗ = 0.05 to 2 Z (cid:12) , respectively. The ionizing to non-ionizingflux density ratios also quantify the impact of Z ∗ on the shapeand strength of the stellar ionizing continua.Figure 15 illustrates that both the stellar population age andmetallicity impact the ionizing continuum. A multivariate,robust M-estimator linear regression (Huber 1981) finds arelationship between these three variables for the single burststarburst99 models (only including ages <
20 Myr) asF F = ( . ± . ) × −( . ± . ) (cid:16) Age (cid:17) −( . ± . ) (cid:16) Z ∗ Z (cid:12) (cid:17) . (5)This relation determines the ratio of the intrinsic ionizing fluxdensity at 900Å to the non-ionizing flux density at 1500Å fora single age burst of star formation given the stellar populationage and metallicity.In Figure 15, we use Equation 5 to overplot three curvesof constant Z ∗ onto the full starburst99 grid. The curvesgenerally approximate the variations in the models, but therestill exists real variations at a given age. F /F evolvessignificantly at the youngest ages and small measurement er-rors lead to large F /F errors. The median error of theestimated age and Z ∗ is 13% and 11% of the estimated values,respectively. For an estimated age of 3 Myr and metallicityof 0.4 Z (cid:12) , with these median uncertainties, there is a 15% Figure 15.
The theoretical evolution of the ionizing to non-ionizingflux density ratio (F /F ) with stellar age of a single burst star-burst99 model. All five starburst99 metallicities are included (asindicated by the color-bar). After 10 Myr a single burst producesan insignificant amount of ionizing photons. The colored curves aresingle metallicity trends, using Equation 5, for three metallicitiescolor-coded the same as the points (light-blue, tan, and gold are 1,0.4, and 0.05 Z (cid:12) , respectively). uncertainty on F /F , including calibration errors. Themedian MegaSaura SNR, 21, is high for restframe FUVobservations, but extremely high-quality observations are re-quired to determine F /F , even with a 15% error.The F /F scales strongly with both stellar age andmetallicity because both impact the stellar temperature. Asstellar populations age, their stellar temperatures decreasewhich increases the fraction of neutral hydrogen within thestellar atmospheres. This is observed in the optical whereBalmer absorption lines increase in older B and A-stars. Thus,the rapidly evolving F /F ratio with stellar age and Z ∗ in Figure 15 probes the increasing X(H )/X(H + ) ionizationfraction within the stellar atmospheres with decreasing stel-lar temperature. By the time the stellar population ages to15 Myr, the temperature has dropped below 20,000 K andthere is sufficient neutral hydrogen within the stellar atmo-spheres to absorb all of the ionizing photons. Thus, F /F traces the hydrogen ionization fraction within the stellar at-mospheres.5.2.2. Inferred ionizing continua from observed galaxies
We inferred the ionizing continua of the observed massivestar populations by extending the stellar fits blueward beyondthe observations. The fits are linear combinations of mul-tiple single age models, and therefore the inferred ionizingroperties and ionizing continua of extragalactic massive stars populations 19 N o r m a li z ed F l u x N o r m a li z ed F l u x N o r m a li z ed F l u x
200 400 600 800 1000 1200 1400 1600Rest Wavelength [Å]0.00.51.01.52.02.53.0 N o r m a li z ed F l u x 𝐅 𝐅 RCS Knot E (2 Myr)Sunburst Arc(3 Myr)S0108+0624(7 Myr)S1226+2152(26 Myr)SB99BPASS
RCS Knot E (2 Myr) 𝐅 𝐅 Figure 16.
The inferred ionizing continua, normalized at 1500Å,for four galaxies: RCS Knot E (with a starburst99 light-weightedage of 2 Myr; top panel), the Sunburst Arc (3 Myr; second panel),S0108+0624 (7 Myr; third panel), and S1226+2152 (26 Myr; bottompanel). Both the starburst99 (blue line) and bpass models (goldline) are plotted to illustrate the large differences between the inferredpopulations at older light-weighted ages. The spectra are orderedby increasing light-weighted age. The RCS Knot E inferred bpasscontinuum is included in gray in all lower panels to illustrate howthe ionizing continuum changes as the stellar population ages. continua may be more complicated than the single burst rela-tions presented in Section 5.2.1. How much do the inferredcontinua depart from a single burst model?In Figure 16, we show the inferred ionizing continua offour galaxies that span the full age range of the sample. Fo-cusing on only the starburst99 fits (blue lines), the generaltrend found in Figure 12 is reproduced: the inferred ioniz-ing continua at 900Å decreases with fitted stellar age fromRCS Knot E (top panel; 2 Myr), to the Sunburst Arc (top-middle panel; 3 Myr), to S0108+0624 (bottom-middle panel;7 Myr), and finally S1226+2152 (bottom panel; 26 Myr).The galaxies with the strongest inferred ionizing continua arethe youngest galaxies with the strongest stellar wind profiles(Section 4). At wavelengths blueward of 900Å, the starburst99 in-ferred ionizing continua have substantially different mor-phologies. The ionizing continuum of RCS Knot E increasesfrom 900 to 525Å, the Sunburst Arc stays relatively flat from900 to 525Å, while the ionizing continua of S0108+0624 andS1226+2152 both decline with decreasing wavelength. Theinferred shapes of the ionizing continua constrain the totalnumber of ionizing photons produced by the stellar popula-tion and the hardness of the nebular emission spectra (seeSection 5.7). In this parlance, RCS Knot E has a particularlyhard ionizing spectrum, while S1226+2152 has a relativelysoft spectrum.The inferred F /F qualitatively follows the generaltrends outlined in Section 5.2.1. RCS Knot E is the youngestpopulation and has a F /F that is 5 times larger than theoldest galaxy, S1226+2152. However, there are large depar-tures in the inferred F /F values from the single burstmodels: S1226+2152 has a light-weighted age of 26 Myr,yet it still has a F /F = .
3. S1226+2152 has an oldstellar population, the C IV profile is nearly flat (Figure 4),yet the inferred ionizing continuum is 500 times strongerthan a single burst model with the same age and metallic-ity (F /F = 0.0006). Similarly, S1527 + /F of 0.2 and 0.1, respectively. These valuesare 50 and 400 times larger than predicted by the single burstmodels. This over-production of ionizing photons extends toall populations older than 9 Myr.Whether the inferred ionizing continua agree with the sin-gle burst models splits the observed stellar populations intotwo star formation histories: a single burst and a mixed agepopulation (Figure 17). Bursts have stellar ages less than8.6 Myr, while mixed age populations have ages greater than8.6 Myr. This divides the full sample into 31 burst-dominatedstellar populations and 30 mixed age populations. There isnot a statistical redshift dependence as both the low-redshiftsample and the MegaSaura sample are nearly split evenly.We overlay the single population model tracks of Figure 15onto the inferred F /F in Figure 17 to show that youngpopulations have single burst-dominated ionizing spectra. Allof the burst-dominated non-ionizing spectra have the spectralproperties outlined in Section 4.3 for young stellar popula-tions: strong N V and C IV P-Cygni features, broad He II emission (when the transition is within the wavelength cover-age), and non-detected Si III photospheric features.The spectra of the second population are a mixture ofold and young stellar populations. The stellar continua ofthese stellar populations are complex, with contributions fromyoung stars shaping the stellar winds and B-stars contribut-ing photospheric absorption features.This creates an averagedionizing continuum that is non-zero due to contributions from0 Chisholm et al.
Figure 17.
The inferred ratio of the ionizing to non-ionizingflux density (F /F ) versus the inferred age from the multi-ple age starburst99 stellar continuum fits to the MegaSaura (goldsquares) and low-redshift (blue circles) samples. The curves are thesingle burst starburst99 models at three metallicities from Fig-ure 15 (blue is 1 Z (cid:12) , tan is 0.4 Z (cid:12) , and gold is 0.05 Z (cid:12) ). Theobserved points are described by two star formation histories: a sin-gle burst of star formation that aligns with the single burst modelsat young ages and a mixed age population. The gray line is the fit tothe mixed age population (Equation 14). The red dashed line is theF /F value of a 0.05 Z (cid:12) constant star formation bpass modeland the red triangle is the inferred value from the starburst99 fitto the Steidel et al. (2016) spectra. massive O-stars, but diluted by contributions from older pop-ulations.The spectral differences between the two populations areseen by comparing the C IV features of the burst-dominatedRCS Knot E and the mixed age S1226 + IV absorp-tion and emission profiles, similar to the single age star-burst99 model. Meanwhile, single burst models do not fitS1226 + IV profile under-estimates the C IV absorption and emission, indicating a weakO-star population. Besides this weak O-star contribution,the pronounced photospheric absorption clearly indicates anolder B-star population (Figure 9). The observed spectralfeatures demonstrate that as the inferred light-weighted ageincreases, a single stellar population ceases to dominate thestellar continuum, rather the stellar light becomes a mix ofyoung and old spectral features.In Figure 17, the F /F of the mixed age populationstrongly (5.5 σ , Pearson’s correlation coefficient of − . F = ( . ± . ) − ( . ± . ) × (cid:18) Age1 Myr (cid:19) . (6)This relationship is overplotted on the observations in Fig-ure 17 as a gray dot-dashed line. There is not a statisticallysignificant trend between F /F and Z ∗ , such that the sta-tistical significance of the trend decreases if we introduce amultivariate fit. Equation 6 analytically quantifies the evolu-tion of the strength of the ionizing continua of older mixedaged stellar populations.We investigated the origin of mixed age populations byenvisioning that the observations of individual star-formingregions sample a random mixture of stellar populations atvarious ages. This can be analytically tested using a ran-dom assortment of the theoretical stellar models. The modelgrid contains more young stars because their spectral fea-tures vary on shorter timescales, consequently, for modelingpurposes, we only included starburst99 models with agesless than 4 Myr or greater than 10 Myr. We then simulatedwhether a burst occurred using a binomial distribution witha uniform probability of success for each model. We chosethe probability for success to be 15% such that the expectedvalue of the number of bursts matched the number of modelstypically included in the fits ( ∼ M i ) = 1), weassigned a random light-fraction drawn from a Gaussian dis-tribution. This process was repeated for all possible stellarmodels and the sum of all light-fractions was normalized tototal one. We multiplied the stellar models by the randomizedlight-fractions and summed the synthetic stellar populationsaccording to Equation 1. We inferred the light-weighted ageand F /F using the same method as for the observations.We then repeated this process one million times to create astatistical sample of random mixtures of stellar populations.This sample of synthetic observations is hereafter referred toas the MCMC sample. We found a strong correlation (Pear-son’s correlation coefficient of − .
81) between the age andthe F /F of this random mixed age population to be F F = . − . × (cid:18) Age1 Myr (cid:19) . (7)This relationship is statistically similar to the relationship in-ferred from the mixed age population in Equation 6. Thissuggests that older inferred light-weighted ages are compos-ites of multiple epochs of stellar populations within a singleaperture. A single stellar population does not dominant theFUV light of mixed age populations, rather the FUV light isbroadly distributed distributed over many ages. The resul-tant composite is a detailed, light-weighted mixture of thedifferent epochs of star formation.roperties and ionizing continua of extragalactic massive stars populations 21The relative proportions of this age mixture are determinedby the light-fraction of each model at a given age, or f age (where f age is the light-fraction of each model age; see Equa-tion 1). As a simplified example, assume that the light at1270Å of a 0.4 Z (cid:12) stellar population comes 70% from a2 Myr population ( f = .
7) and 30% from a 25 Myr popula-tion ( f = . /F from Equation 5 is1.05. These inferred values are on the age-F /F mixedage relationship of Equation 6. The 8.9 Myr stellar contin-uum has a deep N V profile from the 2 Myr population, butweak photospheric lines from the 25 Myr population. Thishypothetical stellar population resembles RCS Knot G in Fig-ure 4 which is dominated by light-fractions of f , f , and f of 0.30, 0.43, and 0.20, respectively. If the assumed light-fractions of the toy model are reversed, the light-weighted ageincreases to 18 Myr and F /F decreases to 0.45. Theyounger population contributes less to the integrated stellarlight and the stellar wind lines of this older hypothetical pop-ulation are now flatter, resembling S1429-1202 in Figure 6.The observed stellar populations are more complicated thanthese toy examples, but they illustrate the evolution of theionizing continua of mixed age populations.The ionizing continuum of a mixed age population looksvery different than that of a single burst. A single burst8 Myr, 0.4 Z (cid:12) , starburst99 population has F /F = . /F = .
05, but the mixed age populationenvisioned above has F /F = . /F = . six timesmore photons per F than a single burst of the same age.Moreover, mixed age populations produce a large numberof extremely hard ionizing photons due to their out-sizedpresence of very young stellar populations (in this example f = . /F that is still three times larger than a singleburst 3 Myr stellar population, among the youngest in oursample, even though the stellar age inferred from the non-ionizing continuum is five times older. Mixed age stellarpopulations produce dramatically more ionizing photons thansuggested by their light-weighted ages.The fitted light fractions illustrate the difference betweena population dominated by a single burst and a mixture ofbursts (see Figure 2). RCS Knot E has a burst-dominatedstellar spectra with a light-weighted age of 2 . ± . f = ± f and f with 31 and44%, respectively). The age distribution of burst populations is relatively simple and clustered, but the light from mixedage populations is broadly distributed over many stellar ages.A principle worry is whether the derived continuum fits,and the associated mixed age populations, are robust. Theerrors on the stellar properties were derived by varying theobserved flux of the stellar continuum by a random Gaussianwith a width equal to the observed flux uncertainty and thenrefitting the stellar continuum. This process was repeated 100times and the standard deviation of the ensemble was takenas the stellar property uncertainty. We find that the SNR ofthe parameter scales with the spectral SNR at the 3 σ signif-icance (p-value of 0.0006, Kendall’s τ value of 0.65). As ageneral guide, we find that the median SNR of the inferredstellar population age is 0.44 times the observed spectral SNRper resolution element. Thus, a spectral SNR per resolutionelement of 22 (7) is required to determine the stellar age atthe 10 σ (3 σ ) significance level. In Section 5.2 we also notedthat a spectral SNR of 21 estimates the F /F with a 15%uncertainty. Thus, a SNR ≈
20 is a general guideline to ro-bustly determine the stellar population properties and to inferthe ionizing continua.If the age mixtures were arbitrary mixtures that were notconstrained by spectral properties, then we would measurea different fitted age for each iteration and the inferred ageand metallicity uncertainty would be largely unconstrainedand uncorrelated with the spectral SNR. We conclude that theproperties of both single bursts and mixed age populationsare relatively robust and constrained by spectral signatures.5.3.
Mixed age versus continuous star formation histories
In Section 4 we emphasized that the O and B-star spectralfeatures reflect changes in the inferred stellar population ageand metallicity. In Figure 3, we showed that the stellar windlines are sensitive to the age of the stellar population. Youngpopulations show strong C IV P-Cygni emission, while olderstellar populations have more moderate C IV features. TheN V wind feature only depends on age (Figure 3) and we seestrong galaxy-to-galaxy N V variations (Figure 6). Short-lived Wolf-Rayet stars create a broad He II feature that isonly seen in the youngest (<5 Myr) spectra (Figure 8). Sim-ilarly, older stellar populations have weak, but distinct, pho-tospheric absorption features that are absent in the youngeststellar populations (Figure 9). We concluded that there isreal and quantifiable age variation within the observed stellarcontinua.The stellar continua vary from galaxy-to-galaxy, but not al-ways as predicted from a single stellar population. The stellarwind profiles from populations with older light-weighted agesrequire a mixture of ages (see S1226+2152 in Figure 4). Thismixture extends to the ionizing continua, where Figure 17demonstrated that the inferred ionizing spectra are dividedinto two star formation histories: single bursts and mixed2 Chisholm et al.age populations. A mixed age stellar population containssome assortment of young (<5 Myr) and old (>8 Myr) stellarpopulations that varies from galaxy-to-galaxy.Constant star formation is the canonical alternative to asingle burst star formation history. A constant star forma-tion history assumes that every year a fixed number of starsare formed by consistently populating the IMF. When aver-aged over a sufficiently long temporal baseline ( >
100 Myr,similar to the ages of older B-stars), continuous star forma-tion is a good physical description of the star formation his-tory of evolved star-forming galaxies, such as normal spirals.However, it is unlikely that stars with lifetimes 1-10% ofthe dynamical time of galaxies will form at a constant rate.However, in principle, a constant star formation is a singular,simplified mixed age solution that can be recovered with theproper light-fractions.A constant star formation history has a fixed number of O-stars which produce a fixed number of ionizing photons. Thered lines in Figure 13 show the ionizing flux density ratios ofa 0.05 Z (cid:12) bpass constant star formation model. This constantstar formation law is similar to a mixed age population with anage of 12 Myr (Figure 17), but a continuous star formation his-tory under-predicts the ionizing continuum of a 8 Myr mixedage population by a factor of 2, and over-predicts the ionizingcontinuum of a 25 Myr mixed age population by a factor of3 (Figure 17). The number of ionizing photons produced bya constant star formation history only matches the inferredionizing continua of the mixed age fits near 12 Myr, and aconstant star formation history does not match the ionizingcontinua at other stellar ages.Only Z ∗ can change the stellar continuum of a constant starformation history. Predominately age sensitive tracers, suchas N V and the ionizing continuum, have fixed strengths ina constantly star-forming model. A constant star formationmodel may fit part of the C IV absorption profile well, butover (or under) estimate the N V profile because the age hasbeen fixed (see Figure 3). The constant star formation modelfails to simultaneously fit the different stellar wind profiles ofvery young populations (e.g. the Sunburst Arc in Figure 6)or older populations that lack stellar wind signatures (e.g.S1527+0652 in Figure 6).Unlike continuous star formation histories, FUV spectralfeatures determine the mixed age star formation histories andobservationally constrain the relative contributions of youngand old stars. By accounting for the dominance of youngstellar populations to certain ionizing continua, mixed agestar formation histories generate significantly more ionizingphotons, at higher energies, than continuous star formationhistories. The different star formation histories alter the ion-izing continua by factors of 2–3 and must be determined toaccurately infer the number of ionizing photons produced bystellar populations. 5.4. Impact of binary evolution on the stellar continua
Most massive stars form in binaries (Sana et al. 2012), andmass can be transferred between the two stars if they aresufficiently close. The star that receives the mass is rejuve-nated and its main-sequence lifetime is extended (Eldridge &Stanway 2009; Götberg et al. 2017). Mass transfer typicallyoccurs when one star is in an evolved stage and expands to fillthe Roche lobe. Consequently, binary star evolution extendsthe duration that stars spend in late evolutionary phases suchas the Wolf-Rayet phase (Vanbeveren et al. 2007; Georgyet al. 2012). By increasing the pathways to form evolvedstars, populations with binary evolution emit more ionizingphotons and have harder ionizing spectra than populationswith only single star evolution (Stanway et al. 2016; Götberget al. 2017, 2018). Binary evolution establishes a signifi-cant evolved population at lower metallicities because lowermetallicity stars are hotter and must expand more in their laterevolutionary phases. On the other side of mass transfer, thestar that donates its hydrogen envelope is hypothesized to re-main as an extremely hot helium core with a black body thatpeaks blueward of 912Å, possibly as an X-ray binary (VanBever & Vanbeveren 2000; Mirabel et al. 2011; Fragos et al.2013; Götberg et al. 2017). These processes add an additionalsource of extreme ionizing radiation.Binary star evolution drastically increases the number ofionizing photons produced by a stellar population. Can therest-frame FUV observationally constrain the importance ofbinary evolution? Here we discuss the observational dif-ferences between bpass and starburst99 models in theirnon-ionizing spectra (Section 5.4.1), the differences betweenthe derived stellar population properties (Section 5.4.2), andthen compare the ionizing continua of the two evolution mod-els (Section 5.4.3). We conclude that the non-ionizing con-tinua alone cannot discern the impact of binary evolution,and explore alternative observational methods to distinguishbetween binary and single star models (Section 5.4.4).5.4.1.
Does binary evolution change the observed non-ionizingstellar continuum? starburst99 and bpass predict comparable non-ionizingcontinua (see Figure 1). The two use similar O-star librariesand produce largely identical non-ionizing spectral features.The stellar wind shapes (e.g. N V and C IV ) slightly differ,but this reflects the subtle differences in the O-star modelsrather than the impact of binary evolution (Leitherer et al.2010; Conroy et al. 2014; Eldridge et al. 2017). The SNRsof individual galaxies are typically too low to distinguishbetween the two stellar models (compare the blue and goldlines in Figure 4–6). However, the starburst99 models fitthe N V and C IV wind features of the high SNR (SNR = 103per resolution element) MegaSaura stack marginally betterthan the bpass models (Rigby et al. 2018b).roperties and ionizing continua of extragalactic massive stars populations 23bpass and starburst99 have significant differences in theirolder populations. The C III and Si
III
III and Si
III ab-sorption features is likely due to the inadequate spectral res-olution, not binary star evolution.The broad stellar He II emission is more prominent in 5-8 Myr bpass models than starburst99 models (Figure 1;Steidel et al. 2016). The bpass models do fit the He II re-gion substantially better than the starburst99 models, butthe overall fit is still occasionally poor at young ages (Fig-ure 8). This is likely because binary evolution provides anadditional pathway to create and rejuvenate WR-stars suchthat the bpass models contain more evolved stars, but eitherthe WR evolution or atmosphere models still need refinement.The improved fit of the He II region may indicate that binarystars contribute to the FUV light, but even the bpass fits donot completely describe the He II region.Similarly, the increased population of evolved massive starsproduces Si IV P-Cygni that are stronger than the C IV fea-tures at ages of 5–10 Myr (Figure 1). Differences betweenSi IV stellar wind profiles are a key prediction of models thatdo and do not include binary evolution. Si IV is not promi-nently observed as a stellar P-Cygni profile at low Z ∗ , butFigure 7 shows that the fitted Si IV profiles are statisticallysimilar whether bpass or starburst99 models are used. Eventhough the single age bpass models show broad Si IV P-Cygniprofiles, the multiple age fits do not indicate Si IV P-Cygnifeatures. This is because the population is a mixed age pop-ulation with young (<5 Myr) and old (>10 Myr) populationsrather than a single age 8 Myr population. Thus, the non-ionizing stellar continua do not disentangle the impact ofbinary star evolution on the ionizing continua.5.4.2.
Does binary evolution impact the inferred light-weightedstellar ages and metallicities?
The light-weighted metallicities and ages derived withbpass and starburst99 in Figure 18 are similar. However,bpass models have a lower spectral resolution at wavelengths < < V stellar wind profiles. We tested that the reduced wave-length coverage causes the poor fitting of the COS data by Figure 18.
Comparison of the derived stellar population propertiesof the MegaSaura sample computed using a linear combination ofeither starburst99 (SB99; x-axis) or bpass (y-axis) stellar models.
Top Panel:
Comparison of the light-weighted stellar metallicities(Z ∗ ). Z ∗ generally follows the one-to-one trend (gray dashed line). Bottom Panel:
Comparison of the light-weighted stellar age. Thereis a slope change near 15 Myr largely because high spectral resolutionbpass stellar templates do not exist at all wavelengths past this age(Figure 1). Consequently, the oldest inferred bpass ages are youngerthan the inferred starburst99 ages. Even still, there is a qualitativeagreement such that old inferred bpass populations are also oldstarburst99 populations. < σ significance) and ages (5 σ significance) scale along the one-to-one relationship in Figure 18. The ages are similar foryoung populations (<15 Myr), while bpass models system-atically estimate younger ages when starburst99 ages are>15 Myr. This split happens at the same age as the degrada-tion of the bpass spectral resolution discussed above. Thus,we conclude that, while the poor bpass spectral resolutioninhibits a firm conclusion, the ages and Z ∗ derived using thedifferent stellar models are similar. In the next section, wecompare the bpass ionizing continua, but we note that theinferred bpass ages are up to 30% younger for the oldestpopulations.5.4.3. Binary evolution strongly impacts the ionizing continua
The C IV (Figure 4 & 5) and N V (Figure 6) regions can-not discriminate between binary and single star populations.While bpass models fit the He II region better than star-burst99 models (Figure 8), neither successfully reproducesthe full He II spectral shape. We concluded that the fittednon-ionizing continua cannot determine the contribution ofbinary star evolution to the stellar continuum.Conversely, the bpass modeled ionizing continua drasti-cally differ from the starburst99 continua. This is mostextreme for older and lower metallicity stellar populations(Figure 13 & 14). At young ages (<3 Myr), the productionof ionizing photons is similar regardless of the binary evo-lution model because young stars have not yet evolved offthe main-sequence. Meanwhile at older ages, such as 8 Myr,there is a factor of two difference in F /F between bpassand starburst99 models. This difference is similar to thedifference between a burst and a 8 Myr mixed age population(Section 5.2.2).The modeled ionizing continua differ most drastically athigher energies (right two panels of Figure 13). A single10 Myr starburst99 burst has negligible F /F = × − , but a population that includes binary evolution producesa staggering F /F = .
20 (the furthest right circle in theright panel of Figure 12). Populations with binary evolutionhave substantially harder ionizing spectra due to the enhancedcontribution of evolved stars.Thus, the inferred bpass ionizing continua are typicallystronger and harder than those inferred from starburst99(blue versus gold curves in Figure 16). While the bpassmodels exhibit the same qualitative trend of decreasing F /F with inferred stellar age, there are important dif-ferences between the ionizing continua inferred with bpassand starburst99. RCS Knot E (top panel) has the youngestMegaSaura population (2.5 Myr) and the inferred ionizingcontinuum is similar for both starburst99 and bpass mod-els. Conversely, the Sunburst Arc, which is only 0.5 Myrolder, has similar F /F values, but the F /F andF /F are 1.4 and 4.9 times larger for the bpass mod-els than the starburst99 models, respectively. The oldestgalaxy, S1226 + /F , butF /F , F /F , and F /F are 1.4, 2.0, and 7.2times larger for the bpass models than the starburst99 mod-els.S0108 + /F , F /F , and F /F are 2, 4, and 12 timeslarger for the bpass fits than starburst99. Surprisingly, thebpass models in Figure 16 predict larger F /F for theolder population of S0108 + How to observationally distinguish binary and single starpopulations
The previous sub-sections demonstrated that the non-ionizing stellar continua cannot quantify the impact of bi-nary star evolution on the integrated stellar spectra, but thereare extreme differences in their respective ionizing continua.While observations suggest that massive stars typically formin binaries, the exact binary parameters–mass ratios, perioddistributions, and remnant masses–are largely unknown andmust be assumed to create the bpass models (Eldridge et al.2017). Therefore, the impact of binary evolution on the stellarcontinuum is observationally unconstrained. Observationaltests must break this degeneracy.Direct observations of the extreme-UV continuum (EUV;wavelengths <912Å) are a definitive method to break this de-generacy. With up to a factor of 12 flux density differencebetween the bpass and starburst99 modeled ionizing con-tinua, direct observations would indisputably determine theimpact of binary star evolution on the ionizing continuum.Further, there are strong stellar wind lines in the EUV (e.g.N
III
Figure 19.
The age variation of the theoretical stellar continuumflux density ratio at 300Å (F ) and 900Å (F ) for starburst99(gold lines) and bpass (blue line) single burst models. Solid linesare 0.05 Z (cid:12) stellar metallicities and dashed lines correspond to 1 Z (cid:12) stellar metallicities. Photons at 300Å create O ++ and photons at900Å create O + , thus the F /F ratio probes the ionization frac-tion of X(O ++ )/X(O + ). Full photoionization modeling is required toexplore the actual nebular emission line ratios ([O III ]/[O II ]). star-burst99 models only produce extreme amounts of hard ionizingphotons within the first 5 Myr and at low metallicities. bpass mod-els produce hard ionizing photons at later ages and all metallicities.Older, higher metallicity bpass models actually produce relativelymore high-energy photons than younger metal poor populations. than single star populations (Figure 16). The intrinsic LyCbreak (near 912Å) is also different for 5–10 Myr populationsdue to the higher stellar temperatures of binary populations(Figure 12). However, the EUV of an O-star dominated stellarpopulation has never been observed.Steidel et al. (2016) broke this degeneracy using nebu-lar emission lines. For an ionization bounded nebula (orwhere the gas absorbs all of the ionizing photons), the rela-tive strength of nebular emission lines at different ionizationstates (e.g., [O III ] 5007Å versus [O II ] 3727Å) traces theintrinsic hardness of the stellar spectrum, and H and He re-combination lines trace the intensity (normalization) of theionizing continuum (modulo the escape of ionizing photonsand the effects of dust). Constraining whether the inferredbpass or starburst99 ionizing continua better reproduce theobserved nebular emission line structure will illuminate theimportance of binary evolution.Comparing [O III ] and [O II ] emission lines may resolvethe different ionizing continua of the two evolutionary mod-els. bpass models create more more O ++ gas relative to low ionization, O + , gas than single star models (Stanway et al.2014). We illustrate this by comparing the F /F ratio ofthe starburst99 (blue lines in Figure 19) and bpass models(gold lines in Figure 19). Photons at 300Å produce O ++ gas,while photons at 900Å only generate O + ; F /F probesthe formation of O ++ and O + gas which emit [O III ] and[O II ], respectively. This is meant to illustrate the possibilityof observing differences in the [O III ]/[O II ] ratio with stellarage, but full photoionization modeling is required to com-pare to the observed emission lines. Models without binaryevolution produce O ++ at low metallicities and young ages,while bpass models produce large X(O ++ )/X(O + ) ratios at allmetallicities and ages. In fact, high Z ∗ bpass models producelarger X(O ++ )/X(O + ) ratios than the youngest low-metallicitysingle star models. Correlations between X(O ++ )/X(O + ) andstellar population properties may constrain the importance ofbinary evolution.Individual emission lines may also break the degeneracy be-tween binary and single-star models. Nebular He II emissionrequires extremely high-energy photons ( λ < II II + II emission line (Figure 8; equiv-alent width of − . ± . + II .5.5. Fits to stacked MegaSaura spectra
Stacking, or averaging many spectra together, increasesthe SNR of the composite spectrum. Often it is assumedthat a stacked spectrum probes the average properties of theunderlying population. The combination of the high-qualityindividual MegaSaura spectra and the stacked MegaSauraspectrum allows us to test this hypothesis.By weighting the individually measured stellar ages andmetallicities by the SNR of their respective MegaSaura spec-trum at 1500Å, we estimated the stellar age and metallic-ity of the MegaSaura ensemble to be 14 . ± . . ± .
03 Z (cid:12) (9 . ± . . ± .
02 Z (cid:12) for bpass).We then fit the stacked MegaSaura spectrum and estimateda light-weighted age and metallicity of 12 . ± . . ± .
01 Z (cid:12) (9 . ± . . ± .
01 Z (cid:12) for bpass).Thus the inferred stellar properties of the stack are statisticallysimilar to the ensemble averages.The SNR weighted average F /F of the individualfits is 0 . ± .
42 and the inferred F /F from the stack6 Chisholm et al.is 0.47. Meanwhile, if we input the fitted light-weightedage of the stellar population into the F /F relationship(Equation 6), we expect F /F = .
77. We find that thefitted stellar properties of the MegaSaura stacked spectrareasonably match the underlying population averages.5.6.
Comparison to previous work
The most directly comparable work to our stellar popula-tion synthesis fitting is Steidel et al. (2016). Those authorsused a stack of 30 galaxies at z ∼ . ∗ , IMFs, and stellar models(bpass versus starburst99) to the observed stellar contin-uum, but always used a continuous star formation history.Low metallicity ( ∼ .
05 Z (cid:12) ) continuous star formation mod-els fit the stellar continuum best, even though the measurednebular metallicities were Z neb = 0.29–0.49 Z (cid:12) .The various continuous star formation stellar continuummodels were then used as the ionizing source within cloudyphotoionization models to predict the nebular emission linesand compare these predictions to the observed values. Con-tinuous star formation starburst99 models produced aninsufficient number of ionizing photons to match the ob-served classical Baldwin et al. (1981) diagnostics such as[O
III ] 5007Å/H β and [N II ] 6585Å/H α . Similarly, bpassmodels with Z ∗ = Z neb failed to reproduce these nebularemission-line ratios.However, Steidel et al. (2016) could reproduce the emis-sion lines if Z ∗ was five times smaller than Z neb . Steidel et al.(2016) argued that rather than a lower Z ∗ , the α /Fe relativeabundance ratio of z ∼ ∗ that is smaller than the actualmetallicity, and the stars would produce more ionizing pho-tons because Fe is the chief opacity source in the atmospheresof massive stars. Thus, they concluded that the non-solar α /Ferelative abundances produced an illusion that the stars had fivetimes fewer metals than the gas.We fit the Steidel et al. (2016) stack with the methods out-lined above, and derived a light-weighted Z ∗ = 0.37 ± .
04 Z (cid:12) using starburst99 models (0 . ± .
02 Z (cid:12) using bpass).The inferred starburst99 stellar population age of the Stei-del et al. (2016) stack is 12.6 Myr, implying that it is a mixedage population. These metallicities are similar to the nebularmetallicities of Z neb = 0.29–0.49 calculated using the directand strong-line methods, implying that the stars and gas havesimilar metallicities (we plot both the direct and strong-lineZ neb calculations in Figure 11).While the inferred light-weighted Z ∗ is meant to conveya total metal content, the strong stellar wind lines, whichare only observed through α -element transitions in the FUV,could hide discrepant α /Fe abundance ratios (Steidel et al.
200 400 600 800 1000 1200 1400 1600Rest Wavelength [Å]0.00.51.01.52.02.53.03.5 N o r m a li z ed F l u x Mixed age BPASS, Z * ≈ Z neb BPASS constant, Z * = 0.2Z neb BPASS constant, Z * = Z neb Mixed age SB99, Z * ≈ Z neb 𝐅 C IV
Figure 20.
Comparison of the inferred stellar continua of the stackedspectrum from Steidel et al. (2016) using a linear combination ofsingle age bpass (in orange with a light-weighted age of 10 Myrand 0.22 Z (cid:12) ) and starburst99 (dark blue with a light-weighted ageand metallicity of 13 Myr and 0.37 Z (cid:12) ). The 0.05 Z (cid:12) (1/5 × Z neb )constant star formation model that reproduces the nebular emissionstructure is shown in red, while the constant star formation modelwith Z ∗ = Z neb , which does not reproduce the nebular emissionlines, is in light blue. The constant and mixed age bpass modelshave similar shapes, even though their metallicities differ by a factorof 4. The light-weighted Z ∗ of the mixed age fits are similar tothe measured nebular metallicity of 0.29–0.49 Z (cid:12) (Steidel et al.2016). Mixed age populations produce a strong and hard ionizingcontinuum, while having a similar metallicity as the nebular gas. ∗ as when we fit the entirespectrum (0.43 and 0.37 Z (cid:12) for starburst99 and bpass, re-spectively). Thus, even in stellar spectral regions dominatedby Fe absorption features, we still infer that Z ∗ and Z neb aresimilar.The only difference between the inferred Z ∗ from this workand Steidel et al. (2016) is the assumed star formation history.As emphasized in Section 5.3, a constant star formation his-tory predetermines a fixed number of young and older stars,each with their own distinct spectral signatures. Since youngpopulations have Fe IV and Fe V photospheric lines (Figure 1)and slightly older populations generally have Fe III lines (fig. 5in de Mello et al. 2000), a continuously star-forming pop-ulation is always expected to have Fe
III , Fe IV , and Fe V absorption. The assumption of constant star formation mayforce the Steidel et al. (2016) fits toward lower metallicitiesin order to balance the strength of the Fe absorption acrossroperties and ionizing continua of extragalactic massive stars populations 27different ionization states. By contrast, a mixed age stellarpopulation can reduce the Fe absorption lines by either reduc-ing the metallicity or by selecting the proper combination ofstellar population ages to match the observed Fe lines (i.e. anolder population without Fe IV and Fe V or a younger pop-ulation without Fe III ). Thus, we emphasize that the a prioriassumed star formation history (constant versus mixed age)has a profound impact on the inferred stellar metallicity.Even though the light-weighted Z ∗ derived from our mixedage fits to the stellar continuum are four times larger, theC IV (cid:12) continuous star formationmodel (orange and red lines Figure 20) and weaker than the0.3 Z (cid:12) constant star formation model (light blue line). Asdiscussed in Section 5.3, continuous star formation modelsremove possible age variations, leaving Z ∗ as the only leverto influence the C IV P-Cygni shape. The mixed age fits tothe stacked data have light-weighted ages of 12 . ± . . ± . IV absorption.Figure 20 shows the inferred ionizing continuum from thefit to the Steidel et al. (2016) stack. This figure stronglyemphasizes that mixed age populations produce substantiallymore ionizing photons, even at higher metallicities, than con-tinuous star formation models. The starburst99 fit (darkblue line) has a weaker ionizing continuum than the bpassfit (orange line), and the starburst99 fit is more similar tothe bpass Z ∗ = Z neb continuous star formation model (light-blue), which failed to reproduce the nebular emission. Assuggested by Steidel et al. (2016), the starburst99 fits areunlikely to generate sufficient ionizing photons to match thenebular emission.The bpass mixed age fit to the stack (orange curve in Fig-ure 20) produces more ionizing photons, with a similar spec-tral shape, as the low metallicity continuous star formationmodel which reproduced the nebular emission structure (redcurve). The youngest stellar populations produce the mostionizing photons (Figure 15), thus, allowing younger popula-tions to have larger light-fractions increases the total numberof ionizing photons produced by a stellar population (see Sec-tion 5.7). Consequently, a mixed age stellar population pro-duces more ionizing photons than a continuous star formingmodel, while having a similar stellar and nebular metallicity.5.7. The production efficiency of ionizing photons bymassive star populations
Throughout this paper we have described the ionizing con-tinua using F /F , a measure of the ionizing continuumat a single wavelength. However, the principle goal of thispaper is to use the FUV stellar continua to determine a morephysically important parameter: the total number of ionizing Figure 21.
The scaling of the ionizing photon production efficiency( ξ ion ; the number of ionizing photons per FUV luminosity) for asingle burst starburst99 stellar population with age (x-axis) andmetallicity (Z ∗ ; color-bar). Younger stellar populations produce anorder of magnitude more ionizing photons per FUV luminosity thanolder stellar populations. The blue, tan, and gold curves are singlemetallicity relations for ξ ion using Equation 10 with Z ∗ of 1, 0.4, and0.05 Z (cid:12) , respectively. photons produced by massive star populations ( Q ). Q scaleslinearly with the star formation rate, such that populationsforming more stars generate more ionizing photons (Madauet al. 1998; Kennicutt 1998; Kennicutt & Evans 2012; Madau& Dickinson 2014). Therefore, the ionizing photon produc-tion efficiency, ξ ion , compares the capacity of different stellarpopulations to produce ionizing photons at a given FUV lumi-nosity. The ionizing photon production efficiency is definedfrom the stellar models as ξ ion [ photons Å erg − ] = Q [ photons s − ] L [ erg Å − s − ] , (8)or in the more common units as ξ ion [ photon Hz erg − ] = Q [ photon s − ] L [ erg Hz − s − ] . (9)When multiplied by the reddening corrected FUV luminosity,L , ξ ion determines the total number of ionizing photonsgenerated by a stellar population.Figure 21 shows that the stellar population parameters, par-ticular stellar age, determine ξ ion . A 2 Myr, 0.4 Z (cid:12) star-burst99 stellar population produces 50 times more ionizingphotons per L than a similar 10 Myr population. Similarly,a 0.05 Z (cid:12) , 2 Myr population produces 3 times more ionizing8 Chisholm et al. Figure 22.
The intrinsic number of ionizing photons per FUV lu-minosity generated by a theoretical stellar population ( ξ ion ) versusthe ratio of the flux density at 900Å to the flux density at 1500Å(F /F ). The blue points are single burst starburst99 pop-ulations and the gray line is the best-fit relation to these points(Equation 12). The relation found from the MCMC starburst99mixed age population is shown by the red line (Equation 14). Mod-els that include binary evolution are gold circles. The upper x-axisshows the stellar age of a 0.4 Z (cid:12) model: at young ages (<5 Myr)all models produce a similar number of ionizing photons per FUVluminosity. photons than a 2 Z (cid:12) population of the same age. These num-bers stress the relative importance of age for the productionof ionizing photons. The multivariate relationship between ξ ion and the starburst99 stellar population properties is ξ ion [ photon Å erg − ] = (cid:16) . × (cid:17) − . Age1 Myr − . Z ∗ (cid:12) , (10)or ξ ion [ photon Hz erg − ] = (cid:16) . × (cid:17) − . Age1 Myr − . Z ∗ (cid:12) . (11)This relation powerfully illustrates that the total number ofionizing photons, Q , can be determined from the stellar pop-ulation age, metallicity, and the extinction-corrected L .As with the F /F relations, the models vary from thefitted relation at a given age (the curves in Figure 21 are sin-gle metallicity relations), but Equation 10 and Equation 11characterize the overall evolution of the efficiency of ionizingphoton production with an observed stellar age and metallic-ity.Equation 10 & 11 have exceptionally similar scalings withage and Z ∗ as F /F does in Figure 15. In fact, F /F strongly scales with ξ ion (Figure 22) aslog (cid:16) ξ ion [ Å erg − ] (cid:17) = . + . × log (cid:18) F F (cid:19) . (12)The strong relationship between F /F and ξ ion may comeas a surprise, but it is simply related by changes in stellar tem-perature. As the stellar temperature decreases with increasingstellar age, the hydrogen within the stellar atmospheres be-comes more neutral and the stellar Lyman break increases(compare the 2 Myr to 3 and 5 Myr models in Figure 12).Thus, F /F effectively measures the hydrogen ioniza-tion fraction within the stellar atmosphere, which is relatedto the stellar temperature by both the stellar age and metal-licity. Since stars are to first order complicated black bod-ies, it follows that the integrated ionizing energy emitted bymassive stars, ξ ion , must strongly scale with the stellar tem-perature (or proxies thereof, such as F /F ) through theStephan-Boltzmann law. Thus, the analysis of the variation ofF /F with age and Z ∗ from Section 5.2.1 directly extendsto ξ ion .Populations with binary evolution produce a flatter rela-tionship between ξ ion and F /F (orange circles in Fig-ure 22), because binary populations produce more evolvedstars, which in turn have higher temperatures at older ages. ξ ion from bpass models scales aslog (cid:16) ξ ion [ Å erg − ] (cid:17) = . + . × log (cid:18) F F (cid:19) . (13)At F /F = 0 . ∼
10 Myr), populations withbinary evolution produce three times more total ionizing pho-tons than populations without binary evolution. However,at ages < at older light-weighted ages than asingle burst star formation history (red line in Figure 22),due to the outsized contribution of younger populations (Sec-tion 5.3). Using the same MCMC sample as Section 5.2.2,there is a strong correlation between ξ ion and F /F forthe mixed age starburst99 population:log (cid:16) ξ ion [ Å erg − ] (cid:17) = . + . × log (cid:18) F F (cid:19) . (14)These three cases illustrate that at older light-weighted ages ξ ion strongly depends on both the star formation history andimpact of binary evolution.Finally, we compare the ionizing photon production effi-ciencies of the observed stellar spectra in Figure 23. The ξ ion values are determined by integrating the entire ionizingroperties and ionizing continua of extragalactic massive stars populations 29 Figure 23.
The inferred ionizing photon production efficiency ( ξ ion )from integrated the ionizing continuum of the stellar continuumfits versus the inferred light-weighted stellar age using the star-burst99 (blue squares) and bpass models (orange circles). Only theMegaSaura spectra were fit with bpass models due to the resolu-tion discrepancy. The blue, tan, and gold curves are single burst Z ∗ starburst99 curves (Equation 10) with Z ∗ of 1, 0.4, and 0.05 Z (cid:12) ,respectively. The red dashed line is the relationship from the MCMCstarburst99 mixed age models. continuum, divided by the photon energy, of the FUV stellarcontinuum fits. Galaxies that we defined as having a singleburst star formation history in Section 5.2.2 closely track thesingle burst ξ ion relations of Equation 10 (blue, tan, and yel-low curves), while mixed age populations follow the curvedefined by the synthetic MCMC population (red curve).The median log( ξ ion [Å erg − ]) of the observed sampleand the starburst99 models is 13 . ± . . ± . ξ ion [Hz erg − ])) and the full log( ξ ion [Å erg − ]) range is12.2–13.6 (24.4–25.7 for log( ξ ion [Hz erg − ])). The galaxieswithin our sample have over an order of magnitude range inthe number of ionizing photons that high-mass stars produceper FUV luminosity. ξ ion strongly varies from galaxy-to-galaxy; accurate stellar continuum modeling must determinehow efficiently ionizing photons are produced. Perhaps mostimportantly, the range of inferred ξ ion shows that the numberof ionizing photons cannot be estimated from the observedFUV luminosity to better than an order of magnitude withoutestimating the stellar population age and metallicity.The measured ξ ion values are largely consistent with valuesinferred in the literature. Kennicutt & Evans (2012) used aconstant star formation history, 1 Z (cid:12) starburst99 model toderive expressions for the star formation rate of normal lo-cal galaxies. These models have log( ξ ion [Å erg − ]) = 13.0(log( ξ ion [Hz erg − ]) = 25.1), which is statistically consistentwith the median value of our sample. While this assumed ξ ion value agrees with the median of our sample, it will not agree on a galaxy-by-galaxy basis, to which a detailed anal-ysis of the massive star properties is required. At higherredshifts, it is typically presumed that more extreme objects,with log( ξ ion [Hz erg − ]) > . − .
3, are required to reion-ize the early universe if 20% of their ionizing photons escapethe interstellar medium (Robertson et al. 2013; Bouwens et al.2016; Finkelstein et al. 2019). Half of our sample, 31 of 61(51%), have ξ ion greater than the value typically assumed forcosmic reionization (log( ξ ion [Hz erg − ]) > . ξ ion [Hz erg − ] = 25.6 measured from their C III ] 1909Åemission lines (Schaerer et al. 2018). These ξ ion values areconsistent with the youngest 2–3 Myr stellar populations inboth of our samples.The bpass fits to the MegaSaura spectra indicate that bi-nary evolution increases the duration of the peak ξ ion , butnot the peak ξ ion itself (orange circles versus blue squares inFigure 23). At the youngest ages, the inferred ξ ion from thestarburst99 and bpass fits are very similar; binary evolutiondoes not increase the maximum inferred ξ ion . Consequently,the youngest stellar populations, which produce the majorityof the ionizing photons, have consistent ξ ion values regardlessof the chosen binary evolution model. Thus, binary evolutiondoes not increase the maximum ξ ion , rather it elongates theperiod which stellar populations produce their maximum ξ ion .If binary evolution is important at high redshifts, it would in-crease the total number of ionizing photons that high-massstellar populations produced over their entire lifetimes, andmake it easier for stars to reionize the universe.A surprising prediction from the bpass models in Figure 23is that the ξ ion values of 4-7 Myr populations do not decreasealong the single burst tracks of Figure 21. Rather, the WR-stars in populations with binary evolution keep the ξ ion ele-vated, and nearly constant at the maximum ξ ion value, overthe 10 Myr lifetime of O-stars. Comparing the inferred ion-izing continua of RCS Knot E and S0108+0624 in Figure 16demonstrates this behavior: the two galaxies have substan-tially different light-weighted ages but similar bpass ionizingcontinua. Thus, the number of ionizing photons produced bya stellar population with binary evolution does not vary untilthe stellar population is older than 10 Myr. As mentionedabove, this could substantially increase the total number ofionizing photons generated over the integrated lifetime of stel-lar populations within the epoch of reionization. This is alsoa testable observation of the impact of binary evolution usingthe nebular emission lines.Determining the total number of ionizing photons intrinsi-cally produced by a given stellar population is a key strength0 Chisholm et al.of FUV spectral synthesis. This analysis infers the total num-ber of ionizing photons, as well as their spectral distribution,using observable spectral features from the same massivestars that emit the ionizing photons. In theory, spectral syn-thesis promises to be an extremely powerful tool for a wideassortment of extragalactic applications, from understandingthe generation of nebular emission lines to uncovering thesources of cosmic reionization. In practice, determining thetotal number of ionizing photons requires precise constraintson the stellar population age, metallicity, star formation his-tory, and the impact of binary star evolution (Figure 21 andFigure 22). We demonstrated throughout this paper that mas-sive star spectral features constrain the age, metallicity, andstar formation history. However, we also emphasized thatthe non-ionizing FUV stellar continuum alone cannot con-strain the impact of binary evolution. Binary evolution doesnot dramatically impact the number of ionizing photons pro-duced per FUV luminosity of the youngest stellar populations(<4 Myr), but it significantly increases the number of ionizingphotons produced by older populations. Without constrain-ing the impact of binary evolution, there is up to a factor of 7uncertainty in the total number of ionizing photons producedby older stellar populations. SUMMARYWe have studied the massive star populations in a sam-ple of 61 star-forming galaxies comprised of 42 at z < . z ∼ λ r = − λ r < IV stellar wind line varieswith the inferred stellar age and metallicity in different anddistinguishable ways (Figure 3), such that the C IV emissiondepends on stellar age (Figure 4) and the absorption dependson metallicity (Figure 5). The N V line strongly depends onage and is nearly independent of metallicity (Figure 6). Veryyoung stellar populations have broad ( ∼
400 km s − ) He II emission from Wolf-Rayet stars, while older populations havenegligible He II features (Figure 8). Finally, old populationshave C III and Si
III photospheric absorption features, whilethese features are undetected in younger populations (Fig-ure 9).The light-weighted stellar metallicity is strongly correlatedwith (8 σ ), and consistent with, the nebular metallicity (Fig-ure 11). The metallicity relationship does not depend onredshift, as we find z ∼ /F , depends on stellar populationproperties, because 900Å is the most common wavelength toobserve the ionizing continuum. Theoretical stellar modelsdemonstrate that the strength and spectral shape of the ioniz-ing continuum of a single burst strongly depends on the stellarage (Figure 12 and Figure 13) and metallicity (Figure 14). Amultivariate relationship between F /F and both the ageand metallicity (Equation 5) describes most of the variationin F /F for a single burst of star formation (Figure 15).We then inferred the ionizing continuum of the observedstellar populations by extrapolating the fits to the observedstellar continua. Stellar populations with older light-weightedages have weaker inferred ionizing continua (Figure 16).However, only half (31 of 61) of the sample follows the sin-gle burst F /F relations (Figure 17). The other half ofthe sample has a mixture of stellar populations at multipleages. The mixed age populations have stronger inferred ion-izing continua at older ages than prescribed by a single burstmodel. The inferred F /F of these galaxies is stronglylinearly anti-correlated (5.5 σ ) with the light-weighted age(Equation 6). These two populations are split at a light-weighted age of 8.6 Myr.Mixed age populations have stronger ionizing continuathan single bursts of a similar age due to the presence ofvery young stellar populations (Figure 17). We emphasizedthe differences between mixed age populations and the typi-cally assumed continuous star formation history (Section 5.3).Chiefly, the mixed age continua depend on the relative mixtureof age and metallicity. Thus, age sensitive spectral features(C IV , N V , and He II ) can vary, and the ionizing continuumcan correspondingly increase or decrease. Mixed age stel-lar populations produce significantly more ionizing photonsthan both a single burst with the same light-weighted age anda continuous star formation history, while having the samemetallicity as the nebular gas (Figure 20). This is in starkcontrast with previous work that assume a constant star for-mation history and we emphasized that the only difference isthe assumed star formation history.Our fiducial stellar models are single star starburst99models mainly because these provided sufficient spectral res-olution at all wavelengths (see the discussion in Section 3.2.2).We also fit the MegaSaura sample with stellar models thatinclude binary evolution (bpass models), but do not find sig-nificant differences between the starburst99 and bpass fitsroperties and ionizing continua of extragalactic massive stars populations 31to the FUV features (compare the blue and gold lines in Fig-ure 3–6). Similarly, the light-weighted metallicities and agesare similar, as long as we only consider young ages for whichthe bpass spectral resolution is adequate (Figure 18). Themost significant difference is that bpass models produce upto 12 times more high-energy ionizing photons than star-burst99 models at older ages and lower metallicities (Fig-ure 15 and Figure 16). Consequently, the number of ionizingphotons produced by a stellar population depends on the ob-servationally un constrained impact of binary evolution.We provide scaling relations for the total number of ioniz-ing photons produced per FUV luminosity ( ξ ion ). ξ ion stronglydepends on the age, metallicity, star formation history, andpresence of binary stars (Figure 21 and Figure 22). Theinferred ξ ion values from our observations have an order ofmagnitude scatter that strongly depends on stellar popula-tion properties (Figure 23). We emphasize that ξ ion timesthe extinction corrected FUV luminosity determines the totalnumber of ionizing photons produced by a stellar population.While fitting the FUV stellar continuum promises to be a pow-erful tool in determining the total number of ionizing photonsproduced by high-mass stars, without constraining the impactof binary evolution there is a factor of 7 uncertainty in thetotal number of ionizing photons produced by older stellarpopulations.The rest-frame FUV spectral features of massive stars mir-ror the production of ionizing photons. Stellar populationsynthesis can estimate many properties required to infer thenumber of ionizing photons produced by stars and determinethe source of cosmic reionization. This type of modeling iscurrently infeasible for galaxies within the epoch of reionizia-tion, but observations of the stellar continuum will be possiblewith future 30 m class telescopes and the James Webb SpaceTelescope . Stellar population synthesis of the most massive stars, as presented here, will be the most direct method toconstrain the number of ionizing photons produced by themost distant star-forming galaxies.We thank the anonymous referee for careful reading theoriginal manuscript and their insightful suggestions. Wethank Chuck Steidel for kindly providing his stacked spectrafor our comparison. JC appreciates Christy Tremonti’s earlystellar continuum fitting guidance and discussions through-out this project. JC thanks Joe Cassinelli for early discussionsthat provided the fundamental stellar physics that inspired asubstantial portion of this project. JC is grateful for help-ful discussions and clarifications from Claus Leitherer. Wethank X. Prochaska for insightful comments and clarificationsthat greatly improved the scope and clarity of the paper. JCthanks the Ohio State University for their hospitality whilewriting portions of this paper. JR is grateful for helpful dis-cussions with George Sonneborn and Sally Heap. JR and DBacknowledge discussion and ideas presented at the CarnegieSymposium in honor of Leonard Searle, on the topic of ‘‘Un-derstanding Nebular Emission in High-Redshift Galaxies,’’held at the Carnegie Observatories in Pasadena in July 2015.This paper includes data gathered with the 6.5 meter Magel-lan Telescopes located at Las Campanas Observatory, Chile.We thank the staff of Las Campanas for their dedicated ser-vice, which has made possible these observations. We thankthe telescope allocation committees of the Carnegie Observa-tories, The University of Chicago, The University of Michi-gan, Massachusetts Institute of Technology, and Harvard Uni-versity, for supporting the MegaSaura project over severalyears of observing. This paper includes data from observa-tions made with the Nordic Optical Telescope, operated bythe Nordic Optical Telescope Scientific Association at theObservatorio del Roque de los Muchachos, La Palma, Spain,of the Instituto de Astrofisica de Canarias.APPENDIXHere we give the tables of the derived stellar parameters for the MegaSaura (Table 3) and low-redshift samples (Table 4). Wethen give the table of the ionization production efficiency ( ξ ion ) for both samples in Table 5 and Table 6. In the electronic versionof the paper we also include the models used to make the fits, the light fractions of each fit, and the fitted stellar continua for eachgalaxy.2 Chisholm et al. Table 3.
Stellar continuum properties of the MegaSaura spectra(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)Galaxy name z E(B-V) E(B-V) Z neb Z ∗ Z ∗ Age Age F /F SpectralSB99 BPASS SB99 BPASS SB99 BPASS SB99 Reference[mag] [mag] [Z (cid:12) ] [Z (cid:12) ] [Z (cid:12) ] [Myr] [Myr]RCS-0327-1326 Knot E 1.7034 0.31 0.29 0.34 c ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± c ± ± ± ± ± ± ± ± ± ± a ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± b ± ± ± ± ± ± ± ± ± ± c ± ± ± ± ± d ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± f ± ± ± ± ± ± ± ± ± ± e ± ± ± ± ± neb ) from the literature. Column 6 and 7 give the light-weighted stellar metallicities (Z ∗ ) using starburst99 and bpass,respectively. Column 8 and 9 gives the light-weighted stellar age using starburst99 and bpass, respectively. Column 10 gives the inferred fluxdensity ratio at 900Å and 1500Å. Column 11 gives references for the galaxies and spectra. All spectra will be fully introduced in Rigby et al.(in preparation).Metallicity references: (a) Bian et al. (2010), (b) Hainline et al. (2009), (c) Rigby et al. (2011), (d) Rigby et al. (2018a), (e) Stark et al. (2008),(f) Wuyts et al. (2012).Spectral references: (1) Bayliss et al. (2011), (2) Belokurov et al. (2007), (3) Bordoloi et al. (2016), (4) Dahle et al. (2016), (5) Diehl et al.(2009), (6) Hennawi et al. (2008), (7) Koester et al. (2010), (8) Marques-Chaves et al. (2017), (9) Rigby et al. (2014), (10) Rigby et al. (2018a),(11) Rivera-Thorsen et al. (2017, 2019), (12) Smail et al. (2007), (13) Wuyts et al. (2012). roperties and ionizing continua of extragalactic massive stars populations 33 Table 4.
Stellar continuum properties of the low-redshift HST/COS Spectra(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)Galaxy name z E(B-V) Z neb Z ∗ Age F /F PID Z neb
Metallicity Spectra[mag] [Z (cid:12) ] [Z (cid:12) ] [Myr] Ref Calibration RefJ1416+1223 0.1231 0.14 0.60 0.60 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± z ). Column 3 gives the fitted stellar attenuation from the starburst99 fits to the stellar continua (E(B-V)). Column 4 givesthe literature nebular metallicity of each galaxy (Z neb ). Column 5 gives the inferred stellar metallicity from the starburst99 fits to the stellarcontinua (Z ∗ ). Column 6 gives the inferred stellar age from the starburst99 fits to the stellar continua. Column 7 is the inferred ratio of theflux at 900Å to the flux at 1500Å. Column 8 is the HST project ID for each spectra. Column 9 is the literature reference for each Z neb . Column10 is the calibration method used to determine Z neb . Codes for metallicity calibration are: PP04 (Pettini & Pagel 2004), KK04 (Kobulnicky &Kewley 2004), EP84 (Edmunds & Pagel 1984), D02 (Denicoló et al. 2002), and M13 (Marino et al. 2013). Column 11 gives literature referencesfor each spectra.Spectral references are coded as: (1) Alexandroff et al. (2015), (2) Borthakur et al. (2014), (3) Fox et al. (2013), (4) France et al. (2010), (5)Hayes et al. (2014), (6) Heckman et al. (2015), (7) James et al. (2014b), (8) Leitherer et al. (2014), (9) Östlin et al. (2014), (10) Rivera-Thorsenet al. (2015), (11) Wofford et al. (2013). Table 5. ξ ion values for the MegaSaura sample.(1) (2) (3)Galaxy name log( ξ ion ) SB99 log( ξ ion ) BPASSlog([photon Å erg − ]) log([photon Å erg − ])RCS-0327-1326 Knot E 13.53 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ξ ion ) for the MegaSaura sample. Column 2 gives the values inferred using the starburst99 modelsand Column 3 gives the values for the bpass models. The galaxies are listed in descending light-weighted age inferred using the starburst99models. roperties and ionizing continua of extragalactic massive stars populations 35 Table 6. ξ ion values for the low-redshift HST/COS sample(1) (2)Galaxy name log( ξ ion )log([photon Å erg − ])J1416+1223 13.54 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ξ ion ) for the low-redshift HST/COS galaxies calculated using the starburst99models. The tables is in descending age order. Abbott D. C., Conti P. S., 1987, ARA&A, 25, 113Alam S., et al., 2015, ApJS, 219, 12Alexandroff R. M., Heckman T. M., Borthakur S., Overzier R.,Leitherer C., 2015, ApJ, 810, 104Asplund M., Grevesse N., Sauval A. J., Scott P., 2009, ARA&A,47, 481Baldwin J. A., Phillips M. M., Terlevich R., 1981, PASP, 93, 5Baldwin J. A., Ferland G. J., Martin P. G., Corbin M. R., CotaS. A., Peterson B. M., Slettebak A., 1991, ApJ, 374, 580Bayliss M. B., Gladders M. D., Oguri M., Hennawi J. F., SharonK., Koester B. P., Dahle H., 2011, ApJL, 727, L26Belokurov V., et al., 2007, ApJL, 671, L9Berg D. A., Skillman E. D., Henry R. B. C., Erb D. K., Carigi L.,2016, ApJ, 827, 126Berg D. A., Erb D. K., Auger M. W., Pettini M., Brammer G. B.,2018, ApJ, 859, 164Berg D. A., Erb D. K., Henry R. B. C., Skillman E. D., McQuinnK. B. W., 2019, arXiv e-prints,Bian F., et al., 2010, ApJ, 725, 1877Bordoloi R., Rigby J. R., Tumlinson J., Bayliss M. B., Sharon K.,Gladders M. G., Wuyts E., 2016, MNRAS, 458, 1891Borthakur S., Heckman T. M., Leitherer C., Overzier R. A., 2014,Science, 346, 216Bouret J.-C., Lanz T., Hillier D. J., Heap S. R., Hubeny I., LennonD. J., Smith L. J., Evans C. J., 2003, ApJ, 595, 1182Bouwens R. J., et al., 2016, ApJ, 833, 72Brinchmann J., Charlot S., Heckman T. M., Kauffmann G.,Tremonti C., White S. D. M., 2004, ArXiv Astrophysics e-prints,Bruzual G., Charlot S., 2003, MNRAS, 344, 1000Byler N., Dalcanton J. J., Conroy C., Johnson B. D., LevesqueE. M., Berg D. A., 2018, ApJ, 863, 14Calzetti D., Armus L., Bohlin R. C., Kinney A. L., Koornneef J.,Storchi-Bergmann T., 2000, ApJ, 533, 682Cardelli J. A., Clayton G. C., Mathis J. S., 1989, ApJ, 345, 245Castor J. I., Abbott D. C., Klein R. I., 1975, ApJ, 195, 157Chabrier G., 2003, PASP, 115, 763Chisholm J., Tremonti C. A., Leitherer C., Chen Y., Wofford A.,2016, MNRAS, 457, 3133Choi J., Conroy C., Byler N., 2017, ApJ, 838, 159Conroy C., 2013, ARA&A, 51, 393Conroy C., van Dokkum P., 2012, ApJ, 747, 69Conroy C., Graves G. J., van Dokkum P. G., 2014, ApJ, 780, 33Cortijo-Ferrero C., et al., 2017, MNRAS, 467, 3898Crowther P. A., 2007, ARA&A, 45, 177Dahle H., et al., 2016, A&A, 590, L4Denicoló G., Terlevich R., Terlevich E., 2002, MNRAS, 330, 69Diehl H. T., et al., 2009, ApJ, 707, 686Drew J. E., 1989, ApJS, 71, 267Edmunds M. G., Pagel B. E. J., 1984, MNRAS, 211, 507 Elbaz D., et al., 2007, A&A, 468, 33Eldridge J. J., Stanway E. R., 2009, MNRAS, 400, 1019Eldridge J. J., Stanway E. R., Xiao L., McClelland L. A. S., TaylorG., Ng M., Greis S. M. L., Bray J. C., 2017, Publications of theAstron. Soc. of Australia, 34, e058Engelbracht C. W., Rieke G. H., Gordon K. D., Smith J.-D. T.,Werner M. W., Moustakas J., Willmer C. N. A., Vanzi L., 2008,ApJ, 678, 804Ferland G. J., et al., 2013, rmxaa, 49, 137Finkelstein S. L., et al., 2019, arXiv e-prints,Fox A. J., Richter P., Wakker B. P., Lehner N., Howk J. C., BenBekhti N., Bland-Hawthorn J., Lucas S., 2013, ApJ, 772, 110Fragos T., Lehmer B. D., Naoz S., Zezas A., Basu-Zych A., 2013,ApJL, 776, L31France K., Nell N., Green J. C., Leitherer C., 2010, ApJL, 722, L80Georgy C., Ekström S., Meynet G., Massey P., Levesque E. M.,Hirschi R., Eggenberger P., Maeder A., 2012, A&A, 542, A29Gladders M. D., Yee H. K. C., 2005, ApJS, 157, 1Gonzalez-Delgado R. M., Perez E., Diaz A. I., Garcia-VargasM. L., Terlevich E., Vilchez J. M., 1995, ApJ, 439, 604Götberg Y., de Mink S. E., Groh J. H., 2017, A&A, 608, A11Götberg Y., de Mink S. E., Groh J. H., Kupfer T., Crowther P. A.,Zapartas E., Renzo M., 2018, A&A, 615, A78Green J. C., et al., 2012, ApJ, 744, 60Green G. M., et al., 2015, ApJ, 810, 25Groh J. H., et al., 2019, arXiv e-prints, p. arXiv:1904.04009Hainline K. N., Shapley A. E., Kornei K. A., Pettini M.,Buckley-Geer E., Allam S. S., Tucker D. L., 2009, ApJ, 701, 52Hayes M., et al., 2014, ApJ, 782, 6Heckman T. M., Robert C., Leitherer C., Garnett D. R., van derRydt F., 1998, ApJ, 503, 646Heckman T. M., Alexandroff R. M., Borthakur S., Overzier R.,Leitherer C., 2015, ApJ, 809, 147Hennawi J. F., et al., 2008, AJ, 135, 664Hernandez S., et al., 2019, arXiv e-prints,Howarth I. D., Prinja R. K., 1989, ApJS, 69, 527Huber P. J., 1981, Robust statistics. WileyIzotov Y. I., Thuan T. X., Lipovetsky V. A., 1997, ApJS, 108, 1Izotov Y. I., Worseck G., Schaerer D., Guseva N. G., Thuan T. X.,Fricke A. V., Orlitová I., 2018, MNRAS, 478, 4851James B. L., Tsamis Y. G., Walsh J. R., Barlow M. J.,Westmoquette M. S., 2013, MNRAS, 430, 2097James B. L., et al., 2014a, MNRAS, 440, 1794James B. L., Aloisi A., Heckman T., Sohn S. T., Wolfe M. A.,2014b, ApJ, 795, 109Kehrig C., Vílchez J. M., Pérez-Montero E., Iglesias-Páramo J.,Brinchmann J., Kunth D., Durret F., Bayo F. M., 2015, ApJL,801, L28Kennicutt Jr. R. C., 1998, ARA&A, 36, 189 roperties and ionizing continua of extragalactic massive stars populations 37
Kennicutt R. C., Evans N. J., 2012, ARA&A, 50, 531Kewley L. J., Ellison S. L., 2008, ApJ, 681, 1183Kinney A. L., Bohlin R. C., Calzetti D., Panagia N., Wyse R. F. G.,1993, ApJS, 86, 5Kobulnicky H. A., Kewley L. J., 2004, ApJ, 617, 240Kobulnicky H. A., Skillman E. D., 1996, ApJ, 471, 211Kobulnicky H. A., Skillman E. D., 1997, ApJ, 489, 636Koester B. P., Gladders M. D., Hennawi J. F., Sharon K., Wuyts E.,Rigby J. R., Bayliss M. B., Dahle H., 2010, ApJL, 723, L73Kroupa P., 2001, MNRAS, 322, 231Kudritzki R. P., 1998, in Aparicio A., Herrero A., Sánchez F., eds,Stellar astrophysics for the local group: VIII Canary IslandsWinter School of Astrophysics. p. 149Kudritzki R.-P., Puls J., 2000, ARA&A, 38, 613Kudritzki R.-P., Urbaneja M. A., Gazak Z., Bresolin F., PrzybillaN., Gieren W., Pietrzyński G., 2012, ApJ, 747, 15Lamers H. J. G. L. M., Cassinelli J. P., 1999, Introduction to StellarWinds. Cambridge, UK: Cambridge University PressLamers H. J. G. L. M., Leitherer C., 1993, ApJ, 412, 771Lamers H. J. G. L. M., Snow T. P., Lindholm D. M., 1995, ApJ,455, 269Le Borgne J.-F., et al., 2003, A&A, 402, 433Leitherer C., Robert C., Drissen L., 1992, ApJ, 401, 596Leitherer C., Chapman J. M., Koribalski B., 1995, ApJ, 450, 289Leitherer C., et al., 1999, ApJS, 123, 3Leitherer C., Ortiz Otálvaro P. A., Bresolin F., Kudritzki R.-P., LoFaro B., Pauldrach A. W. A., Pettini M., Rix S. A., 2010, ApJS,189, 309Leitherer C., Tremonti C. A., Heckman T. M., Calzetti D., 2011,AJ, 141, 37Leitherer C., Ekström S., Meynet G., Schaerer D., Agienko K. B.,Levesque E. M., 2014, ApJS, 212, 14Leitherer C., Byler N., Lee J. C., Levesque E. M., 2018, ApJ, 865,55Levesque E. M., Leitherer C., Ekstrom S., Meynet G., Schaerer D.,2012, ApJ, 751, 67López-Sánchez Á. R., Esteban C., García-Rojas J., 2006, A&A,449, 997Madau P., Dickinson M., 2014, ARA&A, 52, 415Madau P., Pozzetti L., Dickinson M., 1998, ApJ, 498, 106Madau P., Haardt F., Rees M. J., 1999, ApJ, 514, 648Maraston C., 2005, MNRAS, 362, 799Marble A. R., et al., 2010, ApJ, 715, 506Marino R. A., et al., 2013, A&A, 559, A114Markwardt C. B., 2009, in Bohlender D. A., Durand D., Dowler P.,eds, Astronomical Society of the Pacific Conference Series Vol.411, Astronomical Data Analysis Software and Systems XVIII.p. 251 ( arXiv:0902.2850 )Marques-Chaves R., et al., 2017, ApJL, 834, L18 Marshall J. L., et al., 2008, in Ground-based and AirborneInstrumentation for Astronomy II. p. 701454( arXiv:0807.3774 ), doi:10.1117/12.789972McQuinn K. B. W., et al., 2010a, ApJ, 721, 297McQuinn K. B. W., et al., 2010b, ApJ, 724, 49Meynet G., Maeder A., 2000, A&A, 361, 101Meynet G., Maeder A., Schaller G., Schaerer D., Charbonnel C.,1994, A&AS, 103, 97Mirabel I. F., Dijkstra M., Laurent P., Loeb A., Pritchard J. R.,2011, A&A, 528, A149Moustakas J., et al., 2011, preprint, ( arXiv:1112.3300 )Nanayakkara T., Brinchmann J., The MUSE Collaboration 2018,arXiv e-prints,Nemry F., Surdej J., Hernaiz A., 1991, A&A, 247, 469Nieva M.-F., Przybilla N., 2012, A&A, 539, A143Noeske K. G., et al., 2007, ApJL, 660, L43Östlin G., et al., 2014, ApJ, 797, 11Ouchi M., et al., 2009, ApJ, 706, 1136Pauldrach A. W. A., Kudritzki R. P., Puls J., Butler K., 1990, A&A,228, 125Pauldrach A. W. A., Hoffmann T. L., Lennon M., 2001, A&A, 375,161Pellerin A., et al., 2002, ApJS, 143, 159Pérez-Montero E., Amorín R., 2017, MNRAS, 467, 1287Pettini M., Pagel B. E. J., 2004, MNRAS, 348, L59Pettini M., Steidel C. C., Adelberger K. L., Dickinson M.,Giavalisco M., 2000, ApJ, 528, 96Pettini M., Rix S. A., Steidel C. C., Adelberger K. L., Hunt M. P.,Shapley A. E., 2002, ApJ, 569, 742Planck Collaboration et al., 2014, A&A, 571, A29Planck Collaboration et al., 2016, A&A, 594, A27Puls J., et al., 1996, A&A, 305, 171Reddy N. A., Steidel C. C., Pettini M., Bogosavljević M., 2016,ApJ, 828, 107Rigby J. R., Rieke G. H., 2004, ApJ, 606, 237Rigby J. R., Wuyts E., Gladders M. D., Sharon K., Becker G. D.,2011, ApJ, 732, 59Rigby J. R., Bayliss M. B., Gladders M. D., Sharon K., Wuyts E.,Dahle H., 2014, ApJ, 790, 44Rigby J. R., et al., 2018a, AJ, 155, 104Rigby J. R., et al., 2018b, ApJ, 853, 87Rivera-Thorsen T. E., et al., 2015, ApJ, 805, 14Rivera-Thorsen T. E., Östlin G., Hayes M., Puschnig J., 2017, ApJ,837, 29Rivera-Thorsen T. E., et al., 2019, arXiv e-prints, p.arXiv:1904.08186Rix S. A., Pettini M., Leitherer C., Bresolin F., Kudritzki R.-P.,Steidel C. C., 2004, ApJ, 615, 98Robertson B. E., et al., 2013, ApJ, 768, 718 Chisholm et al.