Connecting Galactic Outflows and Star Formation: Inferences from H-alpha Maps and Absorption Line Spectroscopy at 1 < z < 1.5
DDraft version February 23, 2021
Typeset using L A TEX twocolumn style in AASTeX63
Connecting Galactic Outflows and Star Formation:Inferences from H α Maps and Absorption Line Spectroscopy at (cid:46) z (cid:46) . ∗† Nikolaus Z. Prusinski ,
1, 2
Dawn K. Erb , and Crystal L. Martin The Leonard E. Parker Center for Gravitation, Cosmology and Astrophysics, Department of Physics, University ofWisconsin-Milwaukee, 3135 N Maryland Avenue, Milwaukee, WI, 53211, USA Cahill Center for Astronomy and Astrophysics, California Institute of Technology, MC 249-17, Pasadena, CA 91125, USA Department of Physics, University of California Santa Barbara, Santa Barbara, CA 93106, USA (Dated: February 23, 2021)
Submitted to AJABSTRACTWe investigate the connection between galactic outflows and star formation using two independentdata sets covering a sample of 22 galaxies between (cid:46) z (cid:46) . . The HST
WFC3/G141 grismprovides low spectral resolution, high spatial resolution spectroscopy yielding H α emission line mapsfrom which we measure the spatial extent and strength of star formation. In the rest-frame near-UV,Keck/DEIMOS observes Fe II and Mg II interstellar absorption lines, which provide constraints on theintensity and velocity of the outflows. We compare outflow properties from individual and compositespectra with the star formation rate (SFR) and SFR surface density ( Σ SFR ), as well as the stellarmass and specific star formation rate (sSFR). The Fe II and Mg II equivalent widths (EWs) increasewith both SFR and Σ SFR at (cid:38) σ significance, while the composite spectra show larger Fe II EWsand outflow velocities in galaxies with higher SFR, Σ SFR , and sSFR. Absorption line profiles of thecomposite spectra further indicate that the differences between subsamples are driven by outflowsrather than the ISM. While these results are consistent with those of previous studies, the use of H α images makes them the most direct test of the relationship between star formation and outflows at z > to date. Future facilities such as the James Webb Space Telescope and the upcoming Extremely LargeTelescopes will extend these direct, H α -based studies to lower masses and star formation rates, probinggalactic feedback across orders of magnitude in galaxy properties and augmenting the correlations wefind here. Keywords:
Galaxy evolution (594), Galaxy formation (595), High-redshift galaxies (734), Starburstgalaxies (1570) [email protected]@[email protected] ∗ This work is based in part on observations taken by the 3D-HSTTreasury Program (GO 12177 and 12328) with the NASA/ESA
Hubble Space Telescope , which is operated by the Associationof Universities for Research in Astronomy, Inc., under NASAcontract NAS5-26555. † Some of the data presented herein were obtained at the W. M.Keck Observatory, which is operated as a scientific partnershipamong the California Institute of Technology, the University ofCalifornia and the National Aeronautics and Space Administra-tion. The Observatory was made possible by the generous finan-cial support of the W. M. Keck Foundation. INTRODUCTIONGalaxy evolution is driven through the baryon cycle:cool gas flows into the galaxy from the cosmic web and isconverted into stars, the most massive of which quicklydie, expelling their metal-enhanced baryons into the in-terstellar medium (Péroux & Howk 2020). These ejectacan be propelled to the circumgalactic or intergalacticmedia through galactic winds. Although the generaloutline is clear, a detailed understanding of the processesinvolved remains an open issue in modern astrophysics(Veilleux et al. 2020). In this study, we focus on therelationship between galactic winds and star formation. a r X i v : . [ a s t r o - ph . GA ] F e b Prusinski, Erb & Martin
Galaxies with intense star formation are observed tohave powerful outflows of gas; however, the primarydriving mechanisms remain uncertain. Active galacticnuclei (AGN) (Faucher-Giguère & Quataert 2012), en-ergy injection from supernovae (SNe) (Leitherer et al.1999; Veilleux et al. 2005), cosmic rays (Grenier et al.2015), and radiation pressure (Murray et al. 2011) gen-erate and sustain galactic outflows. At a given time,many of these mechanisms are occurring simultaneouslyand they are likely to interact in complex and nonlinearways that depend on the type of galaxy (Hopkins et al.2012). These outflows transfer energy and momentumfrom the centers of galaxies to large radii (Bouche et al.2007; Ménard et al. 2011; Chevalier & Clegg 1985; Mur-ray et al. 2005). In doing so, the outflows may cause adepletion in the availability of cool gas and star forma-tion may be quenched (Tremonti et al. 2007; Hopkinset al. 2008; Gabor et al. 2011).The global star formation rate (SFR) reached its peakat z ∼ and since then, has been steadily decreasing(Hopkins & Beacom 2006; Bouwens et al. 2007). A de-crease in the rate of cool gas accretion onto galaxies mayexplain this drop in star formation (Kereš et al. 2005),although a more robust understanding of galactic feed-back is required to constrain the physical processes driv-ing these changes. Simulations of feedback (e.g. Nelsonet al. 2015; Genel et al. 2015; Sales et al. 2010) and how itregulates galactic disk formation (e.g. Brooks et al. 2009;Minchev et al. 2015; Sales et al. 2012; Übler et al. 2014)together with observations of gas flows at < z < provide insight into galaxy and baryon cycle evolutionduring this critical period.Galactic winds at z ∼ are typically traced by rest-frame UV absorption lines backlit by the stellar contin-uum (Weiner et al. 2009; Rubin et al. 2010; Prochaskaet al. 2011; Erb et al. 2012; Kornei et al. 2012; Mar-tin et al. 2012; Bordoloi et al. 2014). Cool ( T ∼ K) outflowing gas appears blueshifted with respect tothe systemic velocity of the galaxy. These outflows arecommonplace in starburst galaxies across cosmic time,from z ∼ (Heckman et al. 1990; Strickland & Stevens2000; Martin & Bouché 2009; Soto et al. 2012), to z ∼ (Weiner et al. 2009; Rubin et al. 2010; Zhu et al. 2015),to z (cid:38) (Steidel et al. 2010; Shapley 2011; Schreiberet al. 2019).Observations and simulations have suggested a starformation rate surface density ( Σ SFR ) threshold neededto drive outflows of 0.1 M (cid:12) yr − kpc − , although manyfactors (e.g. galaxy escape velocity, inclination angle,etc.) contribute to the presence and detectability of out-flows (Heckman 2002; Murray et al. 2011; Kornei et al.2012). Outflow velocities have been observed to increase with galaxy mass and star formation rate, suggestingthat higher mass galaxies have more ambient gas andenergy from SNe and radiation pressure (SNe rate andluminosity both scale with SFR) and therefore sustainfaster outflows than their lower mass counterparts (Mar-tin 2005; Rupke et al. 2005; Chisholm et al. 2015). Inaddition, the equivalent widths of interstellar absorptionlines associated with outflows are observed to increasewith stellar mass, SFR, and SFR surface density (Weineret al. 2009; Rubin et al. 2010; Martin et al. 2012; Bor-doloi et al. 2014).In this paper, we observe star formation and outflowproperties for a sample of galaxies at < z < . .We probe interstellar Mg II and Fe II absorption withKeck/DEIMOS and use WFC3/G141 grism data fromthe Hubble Space Telescope to construct H α emissionline maps that trace the spatial distribution of star for-mation for each object. We determine the equivalentwidths of the absorption lines along with outflow veloc-ities of the gas flows. Star formation rate surface den-sities for the same objects are ascertained through areameasurements of the highest surface brightness regionsof the H α maps. By combining the two data sets, wedirectly compare the structure of star formation to theoutflow properties.The paper is organized as follows. We describe thejoint dataset in Section 2. Section 2.1 shows how H α emission line maps were constructed, Section 2.1.1 ex-plains how star formation rate surface densities were cal-culated, and Section 2.2 shows how outflow propertieswere measured. In Section 3, we discuss the correla-tions between star formation and outflow properties. InSection 3.1, we form composite absorption line spectra,and in Section 4, we summarize and discuss our results.Throughout this paper, we use the Salpeter (1955) IMFand adopt the Planck Collaboration et al. (2016) cos-mology with H = 67 . km s − Mpc − , Ω m = 0 . and Ω Λ = 0 . . In this cosmology, at the median red-shift of the sample ( z = 1 . ), 1 (cid:48)(cid:48) corresponds to 8.5kpc. OBSERVATIONS, DATA REDUCTION, ANDMEASUREMENTSIn order to measure star formation and outflow veloc-ities at (cid:46) z (cid:46) . , we constructed a joint data setcomprising rest-frame near-UV absorption line spectraand rest-frame optical grism spectra to measure emis-sion lines. Keck/DEIMOS detected Fe II and Mg II absorption lines which provided outflow velocities, andthe H α emission line observed with HST
WFC3/G141traces star formation. onnecting Galactic Outflows and Star Formation . . . Log (M ? /M (cid:12) ) N Low MassHigh Mass . . . Log [SFR/(M (cid:12) yr − )] N Low SFRHigh SFR − . − . . . . Log (sSFR/Gyr − ) N Low sSFRHigh sSFR − . − . . Log [ Σ SFR /(M (cid:12) yr − kpc − )] N Low Σ SFR
High Σ SFR
Figure 1.
Distributions of galaxy properties for the 22 galaxy sample. Stellar mass (top left), star formation rate (top right),specific star formation rate (bottom left), and star formation rate surface density (bottom right) are shown. As described inSection 3.1, the full sample is split into two 11 object subsamples based on the median value of a given parameter. The blueand red histograms correspond to the low and high subsamples respectively from which composite spectra are formed.
The galaxies in this sample were selected from theCosmic Assembly Near-IR Deep Extragalactic LegacySurvey (CANDELS), specifically the Extended GrothStrip (EGS; α = δ = +52:51:00) and the Cos-mic Evolution Survey (COSMOS; α = δ = +02:20:36) fields (Grogin et al. 2011; Koekemoer et al.2011). The sample was chosen using data from the Skel-ton et al. (2014) photometric catalog such that eachgalaxy had (1) a SFR > (cid:12) yr − as measured byspectral energy distribution (SED) fitting from the pho-tometric catalog; (2) . (cid:46) z phot (cid:46) . at 99% confi-dence so that the systemic redshift z sys could be mea-sured from the [O II ] λλ , doublet and absorp-tion lines from Fe II ( ∼ − Å) and Mg II (2800Å) were visible; and (3) R (cid:46) in order to obtain spec-tra with continuum S/N sufficient to measure absorptionlines in individual objects in one night of observation.Starting from the full COSMOS and EGS catalogs, weeliminated objects with unreliable photometry or poorlyconstrained SED fits, which when combined with theabove three criteria reduced the 29,791 (36,699) objectsin the COSMOS (EGS) field to a sample of 520 (446) galaxies. Of the three requirements, the magnitude cuteliminated the largest number of objects from the sam-ple. Because our targets are bright, the catalog fromwhich they are drawn is highly complete over our mag-nitude range of interest; Skelton et al. (2014) estimatea 90% completeness level at H F160W = 25 . , while thefaintest object in our final sample is two magnitudesbrighter than this with H F160W = 23 . . We therefore ex-pect our parent sample to be highly complete, but notethat our selection criteria may eliminate dusty galaxieswith SFR > (cid:12) yr − but R > .Ground-based observations for 84 of these galaxieswere conducted on 2015 March 26 and 27 using the DEepImaging Multi-Object Spectrograph (DEIMOS) on theKeck II telescope. DEIMOS is a medium-resolution op-tical spectrometer with spectral coverage from 4000 Å to10,500 Å (Faber et al. 2003). Out of the 84 galaxy sam-ple at < z < . , 47 objects had significant absorptionline detections. These lines include Fe II λ , λ , λ , λ , λ , and Mg II λλ , , whichtrace the outflow velocities of interstellar gas. DEIMOSobservations were conducted using the 600 lines mm − Prusinski, Erb & Martin grating with one slitmask per field and 1 (cid:48)(cid:48) slits. The dis-persion was 0.65 Å pixel − and the spectral resolutionFWHM was 4.6 Å or ∼ km s − as measured fromthe widths of night sky lines. Total exposure times forthe EGS and COSMOS fields were 8.79 hours and 9.04hours respectively. The airmass ranged between 1.05and 1.31, and the average seeing was ∼ . (cid:48)(cid:48) α emission in the 3D-HST grism survey (Mom-cheva et al. 2016; van Dokkum et al. 2011; Brammeret al. 2012). Each of the galaxies in the EGS and COS-MOS fields has WFC3 F140W+G141 direct and grismobservations from two visits with an average exposuretime of ∼ s. The F140W filter and G141 grism haveoverlapping wavelength coverage from ∼ − Å, with H α visibility from z ∼ . to z ∼ . (2.75 Gyrof cosmic time). Excluding objects with strong contam-ination (shown in Figure 2 and discussed in Section 2.1)and retaining objects with significant H α emission andeither Fe II or Mg II absorption line detections led to a fi-nal sample of 22 galaxies (14 in COSMOS and 8 in EGS).Of those 22 objects, 18 have significant Fe II detections,20 have significant Mg II detections, and 16 have detec-tions of both Mg II and Fe II absorption features. Themedian redshift of the sample is . with standard de-viation . , and the galaxies generally fall on the starforming main sequence for that redshift (Speagle et al.2014).Figure 1 shows the range of galaxy physical parame-ters for the 22 object sample. We observe a median stel-lar mass (M ∗ ) of . × M (cid:12) , a median SFR of 10.6M (cid:12) yr − , a median specific star formation rate (sSFR)of 1.0 Gyr − , and a median SFR surface density of 0.4M (cid:12) yr − kpc − .2.1. Rest-frame Optical Data from HST
The HST/WFC3 grism images were reduced using theGrism Redshift and Line Analysis (
Grizli ) pipeline(see Brammer 2019; Abramson et al. 2020; Wang et al.2019 for descriptions of Grizli ). We input WFC3 di-rect images paired with G141 grism images contain-ing dispersed 2D spectra for each of the objects in thefield. In addition, we supply the spectroscopic redshiftof the target galaxy given by the [O II ] emission lineobserved in the Keck/DEIMOS spectra. Grizli firstpre-processes the G141 exposures by performing astro-metric alignment, background subtraction, flat-fielding,and extracting visit-level catalogs and segmented im-ages from the corresponding direct image. Using the https://github.com/gbrammer/grizli/ AstroDrizzle software (Gonzaga 2012), the pipelinereturns drizzled mosaics of the direct and grism images.
Grizli then makes continuum and contamination mod-els using a polynomial fit and extracts 1D and 2D spec-tra. Examples of the 2D spectra are shown in Figure2. The next step is the creation of H α emission line maps.These are possible because of WFC3’s high spatial res-olution (0.136 (cid:48)(cid:48) ) and G141’s low ( R ∼ ) point-sourcespectral resolution. A G141 grism spectrum compriseshigh spatial resolution images placed in series on theWFC3 detector in 46 Å increments. These exposurescover the wavelength range (1075-1700 nm) of G141 andare placed sequentially on the WFC3 detector. To makethe emission line maps, a spectral template is fit to thedata using the given spectroscopic redshift of the galaxy.The maps are constructed by subtracting the stellar con-tinuum model from the 2D spectrum; the remaining fluxcomes from emission features (Nelson et al. 2016a). Thedirect image is used to map the spatial distribution ofthe emission line from the 2D spectrum to an 80 × (cid:48)(cid:48) , thiscorresponds to 8 (cid:48)(cid:48) × (cid:48)(cid:48) or 68 ×
68 kpc at z = 1 . . InFigure 3, we plot the central (cid:48)(cid:48) × (cid:48)(cid:48) region of each H α emission line map. The emission line in these maps ap-pears as an image of the galaxy taken at the wavelengthof the line. Some of these images contain contaminationfrom other spectra in the field (see left panel of Fig-ure 2). We retained objects for which we were able toexclude regions with strong contamination, so that thecontamination did not significantly impact the flux fromthe H α line.We apply two corrections to the H α flux in these maps:an [N II ] correction and an extinction correction. Dueto the low spectral resolution of the G141 grism, theH α and [N II ] λ lines are blended, so we observethe combined flux of the two lines. After convertingthe galaxy masses in the photometric catalog (Skeltonet al. 2014) to a Salpeter (1955) IMF, we estimate the[N II ] emission as a function of galaxy mass using the[N II ]/H α mass-metallicity relation from Yabe et al.(2014). We then subtract the estimated [N II ] contri-bution from the total observed flux. The median [N II ]contribution across the sample is ∼ of the total flux.For the extinction correction, we assume the Calzettiet al. (2000) extinction law and use A V values from thephotometric catalog to apply an extinction correctionto each of the H α fluxes; the median H α flux correctionfactor is 1.52.We note that the adoption of a single value for A V assumes that the extinction does not vary across thegalaxy, which may not be the case (e.g. Wang et al. onnecting Galactic Outflows and Star Formation Figure 2.
Examples of 2D spectra taken by the WFC3/G141 grism on
HST . The
Grizli pipeline produces a fully reduced 2Dspectrum (top panel), model (middle panel), and subtracts the continuum to produce a line-only spectrum (bottom panel). Onthe left, COS 10318 has H α emission visible at 1.27 µ m, while COS 19180 (right) has [O III ] and H α emission lines present at1.11 and 1.45 µ m respectively. The dark bands in the spectrum of COS 10318 are contamination from another source in thefield also visible in the H α map in Figure 3. α emis-sion likely mitigates this effect. It is also probable thatthe value of A V determined from SED fitting of the spa-tially integrated stellar continuum is not the value ap-propriate to the strongest H α emission, since nebularemission is generally more attenuated than the contin-uum (Calzetti et al. 2000; Reddy et al. 2015) and ex-tinction gradients measured from the Balmer decrementtend to peak in the centers of galaxies (Nelson et al.2016b). Since our galaxy sample does not have H β emis-sion measured over the same regions as H α we are unableto quantify this effect, but the potential impact is thatthe extinction corrections are underestimated and there-fore the SFRs and SFR surface densities may be higherthan reported.2.1.1. Calculation of Star Formation Rate SurfaceDensities
With dust and [N II ]-corrected H α fluxes in hand, wenow seek to measure the galaxy area and compute starformation rate surface densities. Many previous studies(e.g. Rubin et al. 2010; Bordoloi et al. 2014; Heckmanet al. 2015; Heckman & Borthakur 2016) have used anarea proportional to πr , where r is the half-light ra-dius measured from space-based rest-frame UV imaging.Others have made minor variations to this method suchas using the galaxy’s semimajor axis (Rubin et al. 2014)as opposed to the half-light radius; however, becausestar formation often occurs in small, separated clumps,sizes measured in this way may overestimate the area ofthe regions most likely to produce outflows (Rubin et al.2010).Kornei et al. (2012) take a more refined approach tomeasuring the SFR surface density, targeting regions ofa galaxy with active star formation. For the subset oftheir sample in the redshift range for which the Kenni-cutt (1998) conversion between UV luminosity and starformation can be applied, they calculate a “clump” areaby selecting pixels above a Σ SFR threshold of 0.1 M (cid:12) yr − kpc − , since these are most likely to contributeto driving outflows. They then parameterize this areameasurement by calculating a scale factor between the“clump” area and the area calculated using πr P , where r P is the Petrosian radius. They find that the median“clump” area is 74% of the area corresponding to thePetrosian radius, and then systematically define the areaof each object to be the brightest region of the galaxycontaining 74% of the flux within the Petrosian radius.Because our study has the advantage of using H α im-ages, we can define areas that directly measure the re-gions of strongest star formation by converting the H α luminosity of each pixel to SFR via the Kennicutt (1998)relation. On the other hand, the H α emission line mapshave much lower S/N than the broadband images usedto measure sizes in previous studies, and the S/N ofindividual pixels with Σ SFR ∼ (cid:12) yr − kpc − isgenerally low ( ∼ ). For several reasons, we opt notto simply measure the area corresponding to the pix-els above a specified SFR surface density threshold assuggested by Kornei et al. (2012). First, although thereis some observational and theoretical justification for athreshold of Σ SFR = 0 . M (cid:12) yr − kpc − , this value isnot robustly determined and we therefore prefer not toimpose a somewhat arbitrary threshold onto the data.The use of a constant threshold also does not take thevarying noise properties of the images into account, andbecause the S/N of the pixels near the proposed thresh-old is low it also results in the inclusion of significantnoise.Instead of using pixels above a particular threshold,we adopt a technique motivated by aperture photom-etry and define an optimal H α aperture designed tomaximize the S/N of the integrated flux. In the caseof background-dominated observations, the radius thatmaximizes the S/N of the enclosed flux is directly re-lated to the characteristic scale of the surface brightnessprofile; for an exponential profile, the optimal radius is1.8 h , where h is the scale length, while for a Gaussian Prusinski, Erb & Martin
COS 05191 COS 09419 COS 10318 COS 11696COS 12589 COS 13739 COS 14214 COS 15852COS 16172 COS 19180 COS 22785 COS 25161COS 25927 COS 27042 EGS 02986 EGS 12858EGS 14380 EGS 18959 EGS 24533 EGS 26680EGS 27539 EGS 29026
Figure 3.
F140W direct images (left panel) and H α line maps (right panel) of the objects in the sample. Each image is 4 (cid:48)(cid:48) on a side with the red contour indicating the area maximizing H α S/N. The WFC3/G141 PSF has a FWHM of 0 . (cid:48)(cid:48)
136 and isdenoted by the white circle in the top left plot. All plots are on a log color scale. The direct images have a scale ranging between × − and × − erg s − cm − Å − while the H α emission line maps have a scale between × − and × − ergs − cm − . onnecting Galactic Outflows and Star Formation σ or 0.67 FWHM (see Appendix A for aderivation).We iteratively determine this optimal region bysmoothing the emission line and error images slightly,using a range of smoothing kernels (typically 1–2 pix-els). For each smoothed image, we create a series ofapertures of varying sizes by selecting the pixels fallingabove a range of S/N thresholds in the smoothed data,and then measure the enclosed flux and its uncertaintyfor each of these apertures on the original, unsmoothedimages. Finally, we choose the aperture that maximizesthe H α S/N in the original data. As expected, thismethod tracks the highest surface brightness regions,but has the advantages of not imposing a semi-arbitrarySFR surface density threshold and still allowing fainterpixels to contribute. We find that the median aperturesize resulting from this method matches that found byinstead including all pixels above a threshold of Σ SFR > . M (cid:12) yr − kpc − , although not all pixels withinour apertures are above this threshold. The optimal H α apertures are shown by red contours in Figure 3. Forcomparison, these apertures are 40–90% of the galaxysizes measured from the direct F140W images by the Grizli pipeline. As a result, H α fluxes measured usingthe full-light apertures from the direct images are ∼ . times larger than those of the maximal S/N method.Once the sizes of the H α regions are defined, we con-vert the H α luminosity within each aperture to the starformation rate of the galaxy using the Kennicutt (1998)relation, and find the star formation rate surface den-sity by dividing the SFR by the area of the aperture.We note that there is a systematic uncertainty associ-ated with the conversion of H α luminosity to SFR, whichKennicutt (1998) estimated to be ∼ from the vari-ation in other published models and calibrations at thetime. The resulting median Σ SFR for the 22 galaxy sam-ple is 0.4 M (cid:12) yr − kpc − ; this is higher than that ofKornei et al. (2012) because the lower S/N H α imagesare less effective in tracing fainter star formation. How-ever, our sample spans more than an order of magnitudein Σ SFR , and we also find that four objects have a SFRsurface density less than 0.18 M (cid:12) yr − kpc − , and one(EGS 12858) has Σ SFR < . M (cid:12) yr − kpc − .For comparison with the literature, we also calculateSFR surface densities using the methodology of Bor-doloi et al. (2014), who adopt SFRs from SED fittingand define the SFR surface density to be Σ SFR , B = SFR / πR / , where R / is the half-light radius. UsingSFRs from the Skelton et al. (2014) photometric cata-log and half-light radii from the measurements of van derWel et al. (2012), we find that the median SFR surfacedensities corresponding to our optimal apertures are a factor of ∼ . times larger than the values of Σ SFR , B forthe sample. We discuss how correlations between theseSFR surface densities and our outflow-related quantitiescompare with the optimal aperture results in Section 3.The specific star formation rate of each galaxy is com-puted by dividing the SFR by the stellar mass deter-mined from the SED in the 3D-HST catalog (Skeltonet al. 2014). Uncertainties on the stellar mass and A V are provided by R. Skelton (private communication),which we propagate into our error calculations below.To establish uncertainties on each of our measured pa-rameters, 350 Monte Carlo simulations were run for eachgalaxy. Each pixel in the emission line map was per-turbed by a random amount drawn from a Gaussian dis-tribution of width equal to the uncertainty on that pixel,and the measurements of area and SFR described abovewere repeated on each perturbed H α image. The errorfor each quantity was then estimated from the widthof the 68% confidence interval in the resulting distribu-tions. The resulting measurements and uncertainties aregiven in Table 1.2.2. Rest-frame Near-UV Keck Spectroscopy
The Keck/DEIMOS spectra were reduced using theDEEP2 reduction pipeline (Newman et al. 2013; Cooperet al. 2012) , which produces one-dimensional airwavelength-calibrated galaxy, inverse variance, and skyspectra. We then refined the wavelength calibration us-ing the night sky lines, removed the instrumental signa-ture using standard stars observed on the same night asthe science data, and rebinned the spectra by a factor oftwo to increase the S/N; see Erb et al. (2012) for moredetails. The final spectra have a median S/N of 7.6 perresolution element over the wavelength range for whichwe measure absorption features.We measured systemic redshifts for the sample of22 galaxies by fitting a double Gaussian to the [O II ] λλ , emission lines, and then shifted the spec-tra to the rest frame. We normalized the spectra overthe wavelength range 2300–2850 Å, as this interval con-tains the Fe II and Mg II spectral features of interest.Following a slight modification of the procedure outlinedby Rix et al. (2004), we defined a series of continuumwindows and fit a spline curve to the median fluxes ineach of the windows. Each spectrum was then dividedby the best fit to its continuum. The uncertainty onthe points to be fit was defined as the standard devia-tion of the mean of the fluxes in each window. For aconservative estimate of the uncertainty associated with Prusinski, Erb & Martin N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII*
Rest Wavelength (˚A) N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* COS12589 z=1.073 N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII*
Rest Wavelength (˚A) N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* EGS14380 z=1.076 . . . N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* . . . N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII*
Rest Wavelength (˚A) . . . N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* EGS24533 z=1.281 N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII*
Rest Wavelength (˚A) N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* COS05191 z=1.225 N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII*
Rest Wavelength (˚A) N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* COS15852 z=1.189 . . . . N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* . . . . N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII*
Rest Wavelength (˚A) . . . . N o r m a li z e d F l u x FeII FeII FeII FeII FeII MgII MgIIFeII* FeII* FeII* FeII* FeII*FeII* EGS18959 z=1.392
Figure 4.
Six representative DEIMOS spectra (dark gray line), with the range over which significantly-detected ( ≥ σ ) absorption lines are measured over-plotted in black and shaded in gray. The orange dashed line shows the total error calculatedby summing the statistical and normalization uncertainties in quadrature. In the top two panels, there are only Mg II and Fe II detections respectively. In the left figure of the middle panel, there are many Fe II lines in addition to the Mg II doublet. In themiddle right, there is strong Mg II emission line filling, which occurs significantly in all of the bottom four panels as well. In thebottom row, a noisy spectrum is shown with a correspondingly large error spectrum (especially in the 2300–2400 Å range). Onthe bottom right, the galaxy has high Mg II S/N but no other lines are detected. onnecting Galactic Outflows and Star Formation ≈ ± across the sample. Thecontinuum-normalized spectra were used for the analysisthat follows.We then measured the equivalent widths (EWs), ve-locity centroids ( ∆ v ), and maximum blueshifted veloc-ities ( v max ) for all of the absorption lines of interest.Equivalent widths are measured by direct integrationover the region for which the flux is below the contin-uum, and the maximum blueshifted velocity is definedas the velocity corresponding to the wavelength wherethe absorption feature first meets the continuum on theblue side of the line. Uncertainties in each of these quan-tities were determined from the 68% confidence intervalsof 1000 Monte Carlo simulations.Defining a detection as a ≥ σ measurement of theequivalent width, we detect at least one Fe II line in18 galaxies (typically Fe II λ , detected in 17 of 18sources with Fe II detections). Fifteen of those 18 ob-jects have at least two Fe II lines detected, with bothFe II λ and Fe II λ detected in 13 of the 15objects with two or more Fe II detections. In four galax-ies, we detect some or all of the bluer lines (Fe II λ , λ λ ) in addition to Fe II λ and Fe II λ .For the Mg II λλ , absorption system, Mg II λ is detected in 20 objects, and 17 of those 20 have3 σ detections of both lines. Representative examples ofthe spectra are shown in Figure 4.The commonly-detected interstellar absorption linesin star-forming galaxies at z (cid:38) are typically stronglysaturated, such that the equivalent width is largely de-termined by the covering fraction and velocity rangeof the absorbing gas (e.g. Shapley et al. 2003, Erbet al. 2012). In addition the interstellar medium (ISM)and outflow components of the absorption features areblended at the resolution of our spectra; we discuss thisfurther in Section 3.1 when looking at the line profilesof composite spectra. We confirm that for a given ob-ject the Fe II equivalent widths are generally consistentwithin the uncertainties, indicating that the lines are in-deed saturated. We therefore adopt a weighted averageof the equivalent widths and velocities of the detectedFe II features, using the inverse variance of each mea-surement as the weight and the inverse of the squareroot of the sum of the weights as the uncertainty in theaverage. For Mg II , we use only the Mg II λ line forequivalent width and velocity calculations, since the blueside of Mg II λ is often blended with Mg II λ . Equivalent width and velocity measurements from theDEIMOS spectra are shown in Table 2. RESULTSThe complete sample of objects is shown in Figure 3,with F140W direct images (left) and H α emission linemaps (right) for each galaxy. The area that maximizesthe S/N of the integrated H α flux is over-plotted witha red contour. We find that the regions of strongeststar formation are not necessarily contiguous (e.g. EGS12858), and distinct regions with higher H α surface den-sities may exist away from the center. From the 22object sample, 21 objects have Σ SFR > (cid:12) yr − kpc − , with a median SFR surface density of 0.37 M (cid:12) yr − kpc − .We test for the presence of outflows or inflows by de-termining whether the centroids of the Fe II absorptionlines are consistent with the systemic velocity. At the1 σ level, 14 objects have outflows and three have in-flows, while the remaining five objects have velocity off-sets consistent with zero. Requiring a 3 σ threshold re-sults in six objects with outflows and no inflows. K-Stests indicate that each of these subsamples is consistentwith being drawn from the same parent distribution interms of stellar mass, SFR, and SFR surface density. Weuse the Fe II centroid velocities because the Fe II absorp-tion lines are less susceptible to emission line filling. Forthe four galaxies with Fe II non-detections, we use theMg II λ centroid instead. Four of the six objectswith 3 σ detections of outflows are measured with Fe II ,indicating that outflow detections are not dominated bythe effects of Mg II emission line filling.In Figures 5, 6, and 7 we compare the stellar mass,SFR, sSFR, and Σ SFR against EW, ∆ v , and v max re-spectively for all objects in the dataset. We test for cor-relations using the Spearman correlation coefficient ρ ,and determine the significance, σ , which represents thenumber of standard deviations from the null hypothesis(no correlation between quantities). We find that thestrength of Fe II and Mg II absorption increases withthe star formation rate surface density: both the Fe II and Mg II equivalent widths are correlated with Σ SFR at the 3.4 σ level, with similar correlation coefficients of ρ = 0 . and ρ = 0 . respectively.We also that find Mg II equivalent width is positivelycorrelated ( ρ = 0 . ) with the star formation rate atthe 3.2 σ level. In addition, Fe II EW has a marginalcorrelation ( ρ = 0 . , σ = 2 . ) with the SFR. Thesecorrelations suggest that galaxies with more star for-mation drive more gas to a broader range of velocities,thereby increasing the equivalent width of the absorp-tion lines. Finally, we find a weak correlation between0 Prusinski, Erb & Martin
Table 1.
Galaxy Masses and Star Formation RatesObject z sysa M ∗ b SFR c sSFR c Σ SFRc ( M (cid:12) ) (M (cid:12) yr − ) (Gyr − ) (M (cid:12) yr − kpc − )COS 05191 . ± . . ± . . ± . . ± .
76 0 . ± . COS 09419 . ± . . ± . . ± . . ± .
86 0 . ± . COS 10318 . ± . . ± . . ± . . ± .
62 0 . ± . COS 11696 . ± . . ± . . ± . . ± .
37 1 . ± . COS 12589 . ± . . ± . . ± . . ± .
75 0 . ± . COS 13739 . ± . . ± . . ± . . ± .
24 0 . ± . COS 14214 . ± . . ± . . ± . . ± .
05 0 . ± . COS 15852 . ± . . ± . . ± . . ± .
50 0 . ± . COS 16172 . ± . . ± . . ± . . ± .
41 0 . ± . COS 19180 . ± . . ± . . ± . . ± .
45 0 . ± . COS 22785 . ± . . ± . . ± . . ± .
68 0 . ± . COS 25161 . ± . . ± . . ± . . ± .
42 0 . ± . COS 25927 . ± . . ± . . ± . . ± .
27 0 . ± . COS 27042 . ± . . ± . . ± . . ± .
19 0 . ± . EGS 02986 . ± . . ± . . ± . . ± .
13 0 . ± . EGS 12858 . ± . . ± . . ± . . ± .
05 0 . ± . EGS 14380 . ± . . ± . . ± . . ± .
24 0 . ± . EGS 18959 . ± . . ± . . ± . . ± .
15 0 . ± . EGS 24533 . ± . . ± . . ± . . ± .
78 0 . ± . EGS 26680 . ± . . ± . . ± . . ± .
21 0 . ± . EGS 27539 . ± . . ± . . ± . . ± .
26 0 . ± . EGS 29026 . ± . . ± . . ± . . ± .
33 0 . ± . Note — a Systemic redshift ( z sys ) from [O II ] λλ , measured by Keck/DEIMOS. b Stellar masses (M ∗ ) from 3D-HST photometric catalog (Skelton et al. 2014) with uncertaintiesfrom R. Skelton (private communication), converted to a Salpeter (1955) IMF. c Star formation rates (SFR), specific star formation rates (sSFR), and star formation ratesurface densities ( Σ SFR ) computed from
HST
WFC3/G141 H α emission line maps (see Figure3). Mg II maximum outflow velocity and the SFR surfacedensity ( ρ = − . ), with 2.7 σ significance. We findno correlations between between outflow velocity andthe sSFR, nor any correlations between stellar mass andEW or outflow velocity.For a more direct comparison with the literature, wetest the SFR surface densities Σ SFR , B computed usingthe Bordoloi et al. (2014) method (discussed in Section2.1.1) for correlations with our outflow-related quan-tities. The significant correlations between Fe II andMg II EW and the SFR surface density discussed aboveare not present when Σ SFR , B is used; we find EW and ∆ v to be uncorrelated ( σ ≤ . ) with Σ SFR , B for bothFe II and Mg II . On the other hand, we observe a . σ correlation between Mg II maximum outflow velocity and Σ SFR , B , slightly more significant than the . σ cor-relation seen above using our optimal aperture method.The lack of correlation between equivalent width andstellar mass provides further insight into the correlationsdescribed above. If the EW is dominated by the ISMcomponent, we would expect to observe an increase inequivalent width with increasing mass and velocity dis-persion; since we do not see this increase, we concludethat changes in the Mg II and Fe II EW across the sam-ple are likely to be due more to the outflow componentthan the ISM. The Mg II and Fe II EW vs. SFR cor-relations we detect are also then likely to be primarilydetermined by the outflow. We discuss the relative con-tributions of the ISM and outflow components furtherin Section 3.1 below. onnecting Galactic Outflows and Star Formation Table 2.
Absorption Line Measurements from Keck/DEIMOSObject Fe II EW a Mg II λ EW Mg II λ EW Fe II ∆ v b Mg II ∆ v c Fe II v maxd Mg II v maxe (Å) (Å) (Å) (km s − ) (km s − ) (km s − ) (km s − )COS 05191 . ± . . ± . . ± . − ± − ± − ± − ± COS 09419 . ± . . ± . . ± . − ± − ± − ± − ± COS 10318 · · · . ± . . ± . · · · − ± · · · − ± COS 11696 . ± . . ± . . ± . − ± − ± − ± − ± COS 12589 . ± . · · · · · · − ± · · · − ± · · · COS 13739 . ± . . ± . . ± . ± − ± − ± − ± COS 14214 . ± . . ± . . ± . ± − ± − ± − ± COS 15852 . ± . . ± . . ± . − ± − ± − ± − ± COS 16172 . ± . . ± . . ± . ± − ± − ± − ± COS 19180 . ± . · · · · · · − ± · · · − ± · · · COS 22785 . ± . . ± . · · · − ± − ± − ± − ± COS 25161 . ± . . ± . · · · − ± − ± − ± − ± COS 25927 . ± . . ± . . ± . − ± − ± − ± − ± COS 27042 . ± . . ± . . ± . − ± − ± − ± − ± EGS 02986 . ± . . ± . . ± . ±
69 107 ± − ± − ± EGS 12858 . ± . . ± . · · · ± − ± − ± − ± EGS 14380 · · · . ± . . ± . · · · − ± · · · − ± EGS 18959 · · · . ± . . ± . · · · − ± · · · − ± EGS 24533 . ± . . ± . . ± . − ± − ± − ± − ± EGS 26680 · · · . ± . . ± . · · · − ± · · · − ± EGS 27539 . ± . . ± . . ± . − ± − ± − ± − ± EGS 29026 . ± . . ± . . ± . ± − ± − ± − ± Note — a Weighted average of equivalent widths from all significantly detected Fe II absorption lines. b Weighted average of Fe II centroid velocities from all significantly detected Fe II lines. c Mg II centroid velocities computed using Mg II λ . d Weighted average of Fe II maximum velocities from all significantly detected Fe II lines. e Mg II maximum velocities computed using Mg II λ . We also tested subsets of the 22 object sample forcorrelations between the stellar population parametersand the outflow quantities. In particular, we calculatedSpearman correlation coefficients and determined theirsignificance for the 14 object and 6 object subsampleswith 1 σ and 3 σ outflows respectively. In both cases, nosignificant correlations ( σ ≤ . ) were detected betweenany of the quantities, probably due to the small numberof objects in these subsets.Below, we discuss the results from composite absorp-tion line spectra constructed from low and high subsam-ples of galaxy physical properties.3.1. Composite Spectra
We created composite spectra by dividing the samplein half by stellar mass, SFR, sSFR, and SFR surface density and constructing the median spectrum of eachsubsample (we refer to the subsamples as low mass, highmass, low SFR, etc.). The spectra are shown Figure 8,with error spectra computed from the standard devia-tions of 100 bootstrap resamples of each subsample. Thedistributions of the low and high subsets are shown bythe blue and red histograms respectively in Figure 1.For the low (high) subsamples, we find a median stellarmass of . × M (cid:12) ( . × M (cid:12) ), a median SFR of5.9 M (cid:12) yr − (18.1 M (cid:12) yr − ), a median specific SFR of0.5 Gyr − (1.4 Gyr − ), and a median SFR surface den-sity of 0.23 M (cid:12) yr − kpc − (0.43 M (cid:12) yr − kpc − ). Thedifference between the medians of the low and high stel-lar mass, SFR, and sSFR composites is a factor of ∼ ,while the medians of the low and high Σ SFR samples dif-fer by a factor of ∼ . We also note that the separation2 Prusinski, Erb & Martin
EW (˚A) M a ss ( M (cid:12) ) Mg IIFe II
EW (˚A) S F R ( M (cid:12) y r − ) ρ = 0 . , . σρ = 0 . , . σ EW (˚A) − − s S F R ( y r − ) EW (˚A) − Σ S F R ( M (cid:12) y r − k p c − ) ρ = 0 . , . σρ = 0 . , . σ Figure 5.
From left to right, we plot stellar mass, star formation rate, specific star formation rate, and star formation ratesurface density against Fe II (orange triangles) and Mg II λ (blue circles) equivalent widths. As described in the text,there are significant correlations between Mg II EW and SFR, Mg II EW and Σ SFR , and Fe II EW and Σ SFR , and a marginalcorrelation between Fe II EW and SFR as well. Spearman correlation coefficients and significances are shown in the plots forwhich there are significant correlations, color coded by absorption line. −
200 0∆ v (km s − ) M a ss ( M (cid:12) ) Mg IIFe II −
200 0∆ v (km s − ) S F R ( M (cid:12) y r − ) −
200 0∆ v (km s − ) − − s S F R ( y r − ) −
200 0∆ v (km s − ) − Σ S F R ( M (cid:12) y r − k p c − ) Figure 6.
From left to right, we plot stellar mass, star formation rate, specific star formation rate, and star formation ratesurface density vs. the Fe II and Mg II velocity centroids ∆ v . Symbols are as in Figure 5, and we find no significant correlationsbetween any of the quantities ( σ ≤ . ). − − v max (km s − ) M a ss ( M (cid:12) ) Mg IIFe II − − v max (km s − ) S F R ( M (cid:12) y r − ) − − v max (km s − ) − − s S F R ( y r − ) − − v max (km s − ) − Σ S F R ( M (cid:12) y r − k p c − ) ρ = − . , . σ Figure 7.
From left to right, we plot stellar mass, star formation rate, specific star formation rate, and star formation ratesurface density vs. the Fe II and Mg II maximum velocity v max , defined as the velocity where the absorption line reaches thecontinuum on the blue side of the line. The symbols remain the same as in Figures 5 and 6. There is a marginal correlationbetween SFR surface density and Mg II maximum outflow velocity (correlation coefficent and significance shown in the figure),and no other correlations are found ( σ ≤ . ). onnecting Galactic Outflows and Star Formation N o r m a li z ed f l u x Fe II Fe II Fe II* Fe II*Fe II*
Mg II Mg II
Low MassHigh Mass2580 2590 2600 2610 2620 26300.00.51.01.5 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 SFR
High
SFR
Figure 8.
Composite spectra constructed from subsets of galaxies based on the galaxy properties of Figure 1. The left columnof panels shows Fe II transitions at ∼ Å, while the right column shows the Mg II λλ , doublet. From top to bottom,composite spectra are binned by stellar mass, SFR, specific SFR, and SFR surface density. The color scheme is the same asFigure 1, with red and blue corresponding to the low and high respective subsamples of a particular quantity. Uncertainties areshown as faded bars behind the spectra. Prusinski, Erb & Martin
Table 3.
Composite Spectra PropertiesComposite a Fe II EW b Mg II λ EW Fe II ∆ v c Mg II ∆ v d Fe II v maxe Mg II v maxf (Å) (Å) (km s − ) (km s − ) (km s − ) (km s − )Low Mass . ± . . ± . − ± − ± − ± − ± High Mass . ± . . ± . − ± − ± − ± − ± Low SFR . ± . . ± . − ± − ± − ± − ± High SFR . ± . . ± . − ± − ± − ± − ± Low sSFR . ± . . ± . − ± − ± − ± − ± High sSFR . ± . . ± . − ± − ± − ± − ± Low Σ SFR . ± . . ± . − ± − ± − ± − ± High Σ SFR . ± . . ± . − ± − ± − ± − ± Note — a See Figure 1 for distributions and ranges of each subsample. b Weighted average of Fe II λ and Fe II λ equivalent widths. c Weighted average of Fe II centroid velocities from Fe II λ and Fe II λ . d Mg II centroid velocities computed using Mg II λ . e Weighted average of Fe II maximum velocities from Fe II λ and Fe II λ . f Mg II maximum velocities computed using Mg II λ . between the bins is larger than the uncertainties on thebinned quantities by factors ranging from 2.3 (sSFR) to6.7 (stellar mass).For each physical quantity, we measured equivalentwidths, centroid velocities, and maximum velocities forboth of the composite spectra, using only the Fe II λ , Fe II λ , and Mg II λλ , lines be-cause the lines at redder wavelengths are covered bymost (21/22) or all of the individual spectra. We takethe weighted average of the Fe II λ and Fe II λ lines for the most direct comparison to the individualspectra. The absorption line measurements from thecomposite spectra are reported in Table 3.To see how the outflow parameters change with galaxyphysical properties, we compare the difference in equiva-lent widths, centroid velocities, and maximum velocitiesfor each low and high subsample (e.g. the Fe II high massEW vs. Fe II low mass EW). Overall, we see the mostsignificant differences between subsets when measuringthe Fe II absorption features, likely because the use ofthe weighted average of the Fe II λ and Fe II λ lines decreases the uncertainties. In contrast, the Mg II EW and velocity differences between low and high sub-samples tend to remain small with respect to the Fe II measurements. This difference may be because we useonly Mg II λ for the calculations (as with the indi-vidual spectra) which tends to have higher uncertaintieswith respect to the Fe II weighted averages and is com-plicated by emission line filling. On that note, the most significant difference betweenthe composites is found when comparing the Fe II equiv-alent widths of the high and low star-formation-relatedquantities (SFR, sSFR, and Σ SFR ), with the two SFRsubsamples showing the largest difference of . ± . Å. We find almost no difference between the low andhigh stellar mass subsets across all three absorption linemeasurements, which suggests that the changes in thevelocity and equivalent width of the specific star forma-tion rate subsamples are primarily due to changes in theSFR rather than the mass. Excluding the low and highSFR subsamples, the Mg II EWs also have differencesconsistent with zero. These results roughly support theFe II EW vs. SFR correlation as well as the Fe II EW vs. Σ SFR correlation we found by looking at the individualobjects.In comparing the outflow velocities between the lowand high subsamples, we again find that the Fe II mea-surements vary more between subsamples than the Mg II velocities. The Fe II centroid velocities are consistentlyfaster in all of the high SFR-related composites, withthe largest difference of − ± km s − found betweenthe low and high SFR subsamples. We also find a signif-icantly faster Fe II maximum velocity in the high SFRsample. The only significant velocity-related differencein the Mg II composite measurements is found for theSFR surface densities, for which the centroid velocity isfaster in the higher Σ SFR sample. onnecting Galactic Outflows and Star Formation II and Fe II absorption lines changebetween subsamples. In the second row of Figure 8 (thecomposites based on SFR, which show the largest differ-ences between high and low subsamples), we note thatthe red wings of the absorption lines have very simi-lar profiles. This suggests that the velocity dispersionof the ISM is similar in the two subsets, although thespectral resolution of ∼ km s − likely prevents usfrom fully resolving the ISM component. Larger differ-ences are found on the blue sides of the lines, implyingthat the difference between the spectra of the two sub-samples is largely determined by the outflow component.This, along with the higher covering fractions indicatedby the larger depth of the lines throughout the absorp-tion profile in the high SFR subsample (especially onthe blue sides of the lines), points to stronger outflowcomponents associated with higher SFRs.We summarize our study and discuss the results andfuture prospects below. SUMMARY AND DISCUSSIONWe have investigated the relationship between galac-tic outflows and star formation by compiling a jointdataset of 22 star-forming galaxies at (cid:46) z (cid:46) . with rest-frame near-UV absorption line spectrafrom Keck/DEIMOS and H α emission line maps fromWFC3/G141 grism spectra taken as part of the 3D-HST survey. All of the objects have at least a 3 σ de-tection of the H α emission line; 18 have Fe II absorp-tion, and 20 have Mg II significantly detected. Thesample has a median mass and standard deviation of log( M ∗ /M (cid:12) ) = 10 . ± . , and a median redshift of . with standard deviation . . Our primary results areenumerated below.1. We used the grism data and the Grizli pipelineto construct 1D and 2D spectra (see Figure 2 forsample 2D spectra) and H α emission line maps foreach of the 22 objects. We determined the sizes ofthe regions of strongest star formation from theH α maps, choosing the area that maximizes theS/N of the integrated H α flux (Figure 3). We ex-plained this method of area measurement in detailin Section 2.1.1. With the H α luminosities andsizes, we compute star formation rates, specificstar formation rates, and star formation rate sur-face densities. These quantities and their respec-tive distributions are listed in Table 1 and plottedin Figure 1. 2. From the DEIMOS spectra, we measured Fe II andMg II equivalent widths, centroid velocities, andmaximum velocities (the velocity where the fluxreaches the continuum on the blue side of the line)for each of the objects. These measurements aregiven in Table 2, and a representative sample ofthe spectra is shown in Figure 4.3. The results from the H α and absorption line mea-surements are combined in Figures 5, 6, and 7, inwhich we plot stellar mass, SFR, sSFR, and Σ SFR against EW, centroid velocity ( ∆ v ), and maximumvelocity ( v max ) respectively. Spearman correla-tion coefficients and their significances are com-puted for each of the relationships. The Fe II andMg II equivalent widths are both positively corre-lated with the star formation rate surface densityat the ∼ . σ level, and the Mg II equivalent widthalso increases with the star formation rate with σ = 3 . . Marginal correlations are found betweenFe II EW and the SFR ( σ = 2 . ), and Mg II max-imum outflow velocity versus SFR surface density( σ = 2 . ). There are no significant correlations( σ < ) between the other plotted quantities.4. Composite spectra were formed by splitting thedataset into low and high subsamples based onstellar mass, SFR, sSFR, and SFR surface den-sity (Figure 8). For each low and high subset,we measured Fe II and Mg II equivalent widths,centroid velocities, and maximum velocities (seeTable 3). For all of the star-formation-relatedquantities (SFR, sSFR, and Σ SFR ), the Fe II ab-sorption lines show significantly larger equivalentwidths and centroid velocities for the high subsetsrelative to the low subsamples. We find that theFe II and Mg II absorption lines in the high SFRcomposites have stronger blue wings, supportingthe hypothesis that the increase in EW seen withSFR in the spectra of individual objects is due toan increase in the strength of the outflow compo-nent with SFR.Given the complex physics of galactic outflows, manyauthors have attempted to constrain their driving mech-anisms by identifying relationships between galacticproperties (e.g. mass, SFR) and outflow-related quanti-ties such as wind velocity. In the local universe ( z ∼ ),Martin (2005), Rupke et al. (2005), Chisholm et al.(2015), and Heckman et al. (2015) have observed out-flows in relatively small ( ∼ ) samples of star-forminggalaxies. These samples cover a wide range in mass( ∼ – M (cid:12) ) and SFR ( ∼ . to nearly 1000 M (cid:12) Prusinski, Erb & Martin yr − ), and the inclusion of dwarf galaxies with lowmasses and star formation rates has been crucial to thedetection of trends between galaxy and outflow proper-ties. Multiple studies have extended these observationsto z ∼ (e.g. Weiner et al. 2009, Rubin et al. 2010, Ko-rnei et al. 2012, Martin et al. 2012, Bordoloi et al. 2014,Rubin et al. 2014); however, the higher redshift sam-ples span only the upper end of parameter space (stellarmasses (cid:38) M (cid:12) , SFRs (cid:38) – M (cid:12) yr − ) and oftenrely on the coadding of large numbers of spectra.From these observations, several trends have been de-tected that link star formation to outflow characteristics.Various studies have detected a shallow increase in out-flow velocity with star formation rate, finding relation-ships similar to v ∼ SFR . (Martin 2005; Rupke et al.2005; Weiner et al. 2009; Chisholm et al. 2015; Trainoret al. 2015). Outflow velocities have also been foundto generally increase with SFR surface density, both lo-cally (Heckman et al. 2015) and at higher redshifts (Ko-rnei et al. 2012), although this is not always observed(Chisholm et al. 2015; Rubin et al. 2014). Most studiesof galactic outflows at z ∼ have observed correlationsbetween Fe II and Mg II equivalent widths and stellarmass, SFR, or SFR surface density (Martin et al. 2012;Weiner et al. 2009; Rubin et al. 2010; Bordoloi et al.2014; Rubin et al. 2014). Given that the observed Mg II and Fe II transitions are optically thick, the equivalentwidth is primarily determined by the covering fractionand velocity distribution of the interstellar gas (with apotential additional contribution from emission fillingin the case of Mg II ). The observed correlations withequivalent width then support a general model in whichgalaxies with more intense star formation drive outflowswith a higher covering fraction to a wider range of ve-locities.Our results are consistent with this model and withprevious studies at z ∼ in the sense that the strongestcorrelations we observe between individual objects arerelated to the Fe II and Mg II equivalent width, whichincrease with SFR and SFR surface density. The com-posite spectra in particular highlight the relationshipbetween star formation and outflows, with larger Fe II equivalent widths and higher velocities seen in the highersubset of all the spectra based on SFR-related quanti-ties. These findings echo previous results at z ∼ whichobserve ties between the SFR and SFR surface densityand outflow velocity (Kornei et al. 2012; Heckman et al.2015; Heckman & Borthakur 2016), confirming the closeconnection between feedback and star formation.The sample studied here spans slightly more than anorder of magnitude in stellar mass, from – × M (cid:12) , and within this relatively narrow range we find no correlations with outflow properties measured from in-dividual spectra. Most of the composite spectra alsoshow no differences in velocity when divided by mass,although we do find that the Fe II maximum outflowvelocity is higher in the low mass subsample, perhapscontrary to expectations. The lack of a relationship be-tween mass and outflow velocity in the individual ob-jects may be unsurprising given previous findings thatthese quantities are only weakly correlated (Heckmanet al. 2015; Chisholm et al. 2015). The lack of corre-lation between mass and equivalent width in both theindividual and composite spectra may be more unex-pected, since most other studies at z ∼ have observedthat EW increases with stellar mass (Weiner et al. 2009;Rubin et al. 2010; Martin et al. 2012; Kornei et al. 2012;Bordoloi et al. 2014; Rubin et al. 2014); however, thesestudies are based on larger samples covering a widerrange in mass.We can make a more quantitative comparison betweenour sample and previous results by assuming the scalingrelation v max ∼ SFR . , as found in previous studies.Using the median values of v max and SFR in our sam-ple to normalize the relationship, our full range of SFRsthen predicts a velocity range of v max ∼ – kms − across the full sample, much wider than observed;however, two-thirds of our sample lies in the narrowerrange of < SFR /(M (cid:12) yr − ) < , and within thisrange the observed scatter in v max is comparable to theexpected variation due to SFR. This suggests that theintrinsic scatter in the velocity–SFR relationship is toolarge for it to be detected within the range of star for-mation rates probed here. We note that although Heck-man et al. (2015) observe a strong correlation betweenoutflow velocity and SFR in the local universe, there isnearly an order of magnitude variation in outflow veloc-ity at a given SFR at the upper end of their sample.Although our small range of star formation rates likelyprevents us from finding a correlation between outflowvelocity and SFR in the individual objects, we do finda relationship between Fe II maximum outflow velocityand SFR in the composite spectra that is consistent withprevious studies: using the median SFRs for the low andhigh SFR subsets along with their respective maximumvelocities, we find that Fe II v max ∼ SFR . . Since com-posite spectra improve the measurement of weak fea-tures, the maximum outflow velocity can be measuredmore robustly, which may explain why this correlation isnot seen in the individual objects. We also note that us-ing the Fe II centroid velocities ∆ v rather than the max-imum velocities v max results in the significantly strongerrelationship ∆ v ∼ SFR . . onnecting Galactic Outflows and Star Formation (cid:12) yr − kpc − . One possible explanationfor this may be that outflows are present, but are colli-mated and pointed away from our line of sight; evidencefor non-spherical outflow geometries has been found inother studies at similar redshifts (Kornei et al. 2012;Martin et al. 2012; Bordoloi et al. 2014; Rubin et al.2014). If the absence of detectable outflows is an orien-tation effect, we would expect these four galaxies to bedisks observed roughly edge-on.To assess this possibility, we turn to the catalog ofstructural measurements of CANDELS galaxies (van derWel et al. 2012), from which we find that these fourobjects have axis ratios between 0.43 and 0.79, indis-tinguishable from the rest of the sample. We also findthat these four galaxies are compact, all with effectiveradii below the sample median of 2.6 kpc; however, theirSFRs are not particularly high, with all but one belowthe sample median, meaning that the high SFR surfacedensities are due more to small sizes than high SFRs.The compact nature of these objects limits our abilityto derive constraints on their geometry and orientation,particularly given the fact that the WFC3 F160W im-ages used in the van der Wel et al. (2012) catalog are afactor of ∼ lower in spatial resolution than the ACSF606W and F814W images used by Kornei et al. (2012)and Bordoloi et al. (2014).While this study is limited by both sample size anddynamic range, it demonstrates the novel combinationof direct measurements of star formation at high redshiftvia H α emission line maps and direct measurements ofoutflows with deep absorption line spectroscopy. Oursample pushes the limits of current technologies: withexisting ground-based facilities, full night integrationtimes ( ∼ hrs) are required to obtain sufficiently highS/N spectra of even bright galaxies at z (cid:38) , whilethe time that would be required to obtain space-basedH α maps of galaxies fainter than those in our sampleis currently impractical. However, upcoming facilitieswill enable significant advances in both aspects of thisstudy. The future extremely large telescopes (ELTs) will be able to obtain similarly high S/N rest-UV spectra ofbright galaxies with a fraction of the time, and withlonger exposures they will extend absorption line stud-ies to fainter or more distant objects.When the James Webb Space Telescope (JWST) flies,it will conduct deep galaxy surveys at < z < , trac-ing star formation across ∼ Gyr of cosmic time andlooking at earlier and more distant objects than
HST isable to see. Onboard, the Near-Infrared Imager and Slit-less Spectrograph (NIRISS) will be analogous to Hub-ble’s WFC3 camera. One of the four NIRISS observingmodes enables wide field slitless spectroscopy over theentire field of view, using one or both of the telescope’sgrisms and blocking filters to isolate wavelength inter-vals between 0.8 and 2.2 µm . With JWST ’s increasedcollecting area, it will then be possible to map regionsof star formation across a broader range of galaxies at (cid:46) z (cid:46) . JWST and the upcoming 30-m class tele-scopes will observe galaxies with lower masses and starformation rates, correlating outflows and star formationto measure galactic feedback across orders of magnitudein galaxy properties.ACKNOWLEDGMENTSThe authors thank the referee for a thorough and con-structive report, as well as Gabriel Brammer for use-ful discussions and support with
Grizli . N.Z.P. wassupported by the University of Wisconsin-Milwaukee’sOffice of Undergraduate Research through the Supportfor Undergraduate Research Fellows (SURF) Award andSenior Excellence in Research Award (SERA). D.K.E.and N.Z.P. are supported by the US National ScienceFoundation (NSF) through the Faculty Early CareerDevelopment (CAREER) Program grant AST-1255591and the Astronomy & Astrophysics grant AST-1909198.C.L.M. is supported by NSF grant AST-1817125. Theauthors wish to recognize and acknowledge the very sig-nificant cultural role and reverence that the summit ofMaunakea has always had within the indigenous Hawai-ian community. We are most fortunate to have the op-portunity to conduct observations from this mountain.
Facilities:
HST (WFC3/G141), Keck:II (DEIMOS).
Software:
Grizli (Brammer 2019), astropy (The Astropy Collaboration et al. 2013, 2018),
AstroDrizzle (Gonzaga 2012), DEEP2 ReductionPipeline (Newman et al. 2013; Cooper et al. 2012).APPENDIX8
Prusinski, Erb & Martin A. DERIVATION OF THE APERTURE OF HIGHEST SIGNAL-TO-NOISEWe compute the optimal apertures that maximize the signal-to-noise ratio for background-limited sources withGaussian and exponential surface brightness profiles as referenced in Section 2.1.1 (the optimal radius of a Gaussianprofile is a useful quantity for aperture photometry). We assume a two-dimensional circular surface brightness profilewith radial form denoted by g ( r ) , g ( r ) = exp (cid:16) − rh (cid:17) exponential profile exp (cid:18) − r σ (cid:19) Gaussian profile , (A1)where h and σ are the scale length and standard deviation of the exponential and Gaussian distributions respectively.We omit numerical prefactors in Equation A1, since they do not affect the final result.The signal S ( R ) received from the source is the total number of photons collected by the detector, or equivalently,the number of photoelectrons produced. For a source with enclosed area A , S ( R ) is given by (cid:90) A g ( r ) d A , so for ourassumed profiles S ( R ) = 2 π (cid:90) R g ( r ) r d r. (A2)The variance of the signal is S ( R ) , assuming Poisson statistics.The signal-to-noise of the observation is then given by the CCD equation (Merline & Howell 1995): S/N = S ( R ) (cid:114) S ( R ) + n pix (cid:16) n pix n B (cid:17) (cid:16) N B + N D + N R + G σ f (cid:17) . (A3)The number of pixels inside the photometric and background apertures correspond to n pix and n B respectively, whilethe final term in the denominator contains the per pixel background noise ( N B ), dark current ( N D ), read noise ( N R ),gain in electrons/ADU ( G ), and A/D conversion error ( σ f ).We assume that the photometric aperture is small compared to the aperture used to determine the background, thus n pix /n B (cid:28) . To simplify the expression, we combine the non-Poisson sources of noise into a single per-pixel noise σ b .For a circular aperture of radius R , the background variance is then πR σ b , and the total variance is σ = S ( R ) + πR σ b . (A4)This leads to a signal-to-noise ratio S/N = S ( R ) (cid:112) S ( R ) + πR σ b = 2 π (cid:90) R g ( r ) r d r (cid:115) π (cid:90) R g ( r ) r d r + πR σ b , (A5)which we seek to maximize.Under the assumption of background-dominated observations, the Poisson noise is negligible compared to the back-ground σ b , and the signal-to-noise ratio is then S/N ≈ S ( R ) (cid:112) πR σ b = 2 π (cid:90) R g ( r ) r d r (cid:112) πR σ b . (A6)In order to maximize the S/N, we numerically solve ∂ ( S/N ) ∂R = 0 with the assumption that S ( R ) (cid:28) πR σ b for boththe exponential and Gaussian profiles. The resulting radii corresponding to the highest S/N apertures are then: R max = (cid:40) . h exponential profile . σ Gaussian profile . (A7) onnecting Galactic Outflows and Star Formation R max = 0 . FWHM.REFERENCES