The MOSDEF Survey: Broad Emission Lines at z=1.4-3.8
William R. Freeman, Brian Siana, Mariska Kriek, Alice E. Shapley, Naveen Reddy, Alison L. Coil, Bahram Mobasher, Alexander L. Muratov, Mojegan Azadi, Gene Leung, Ryan Sanders, Irene Shivaei, Sedona H. Price, Laura DeGroot, Dušan Kereš
DDraft version October 11, 2017
Preprint typeset using L A TEX style emulateapj v. 01/23/15
THE MOSDEF SURVEY: BROAD EMISSION LINES AT z = 1 . − . William R. Freeman , Brian Siana , Mariska Kriek , Alice E. Shapley , Naveen Reddy , Alison L. Coil ,Bahram Mobasher , Alexander L. Muratov , Mojegan Azadi , Gene Leung , Ryan Sanders , Irene Shivaei ,Sedona H. Price , Laura DeGroot , Duˇsan Kereˇs Draft version October 11, 2017
ABSTRACTWe present results from the MOSFIRE Deep Evolution Field (MOSDEF) survey on broad flux fromthe nebular emission lines H α , [N II ], [O III ], H β , and [S II ]. The sample consists of 127 star-forminggalaxies at 1 . < z < .
61 and 84 galaxies at 2 . < z < .
80. We decompose the emission linesusing narrow (FWHM <
275 km s − ) and broad (FWHM >
300 km s − ) Gaussian components forindividual galaxies and stacks. Broad emission is detected at > σ in <
10% of galaxies and the broadflux accounts for 10-70% of the total flux. We find a slight increase in broad to narrow flux ratiowith mass but note that we cannot reliably detect broad emission with FWHM <
275 km s − , whichmay be significant at low masses. Notably, there is a correlation between higher signal-to-noise (S/N)spectra and a broad component detection indicating a S/N dependence in our ability to detect broadflux. When placed on the N2-BPT diagram ([O III ]/H β vs. [N II ]/H α ) the broad components of thestacks are shifted towards higher [O III ]/H β and [N II ]/H α ratios compared to the narrow component.We compare the location of the broad components to shock models and find that the broad componentcould be explained as a shocked outflow, but we do not rule out other possibilities such as the presenceof an AGN. We estimate the mass loading factor (mass outflow rate/star formation rate) assumingthe broad component is a photoionized outflow and find that the mass loading factor increases as afunction of mass which agrees with previous studies. We show that adding emission from shocked gasto z ∼ z ∼ INTRODUCTION
Rest-frame optical nebular emission lines such as H α ,[N II ], [O III ], H β , and [S II ] are diagnostics of physicalproperties of galaxies such as star formation rate (SFR)(Kennicutt 1998, Shivaei et al. 2015), dust extinction(Kashino et al. 2013, Steidel et al. 2014, Reddy et al.2015), electron density (Osterbrock 1989, Hainline et al.2009, Bian et al. 2010, Sanders et al. 2016), and metal-licity (Pettini & Pagel 2004, Erb et al. 2006, Sanderset al. 2015). Consequently, galaxies fall into well-definedpatterns in emission line diagnostic diagrams such as theN2-BPT diagram ([O III ]/H β vs. [N II ]/H α ) and theS2-BPT diagram ([O III ]/H β vs. [S II ]/H α ) (Baldwin email: [email protected] * Based on data obtained at the W.M. Keck Observatory,which is operated as a scientific partnership among the Cali-fornia Institute of Technology, the University of California, andNASA, and was made possible by the generous financial supportof the W.M. Keck Foundation. Department of Physics & Astronomy, University of Cali-fornia, Riverside, 900 University Avenue, Riverside, CA 92521,USA Astronomy Department, University of California, Berkeley,CA 94720, USA Department of Physics & Astronomy, University of Califor-nia, Los Angeles, 430 Portola Plaza, Los Angeles, CA 90095,USA Center for Astrophysics and Space Sciences, University ofCalifornia, San Diego, 9500 Gilman Dr., La Jolla, CA 92093-0424, USA Steward Observatory, University of Arizona, Tucson, AZ85721, USA Max-Planck-Institut fr extraterrestrische Physik, Giessen-bachstr. 1,D-85737 Garching, Germany Department of Physics, The College of Wooster, Wooster,OH 44691 et al. 1981). The position of a galaxy on these diagramsis determined by its underlying physical conditions suchas electron density, hardness of ionizing radiation, AGNpresence, and metallicity (Kewley et al. 2013, Shapleyet al. 2015, Coil et al. 2014, Sanders et al. 2016).Galaxies at z ∼ z ∼ a r X i v : . [ a s t r o - ph . GA ] O c t increased star formation associated with merging couldbe driving the outflows. Estimates of physical proper-ties of galaxies from emission lines typically assume theemission originates in the HII regions of galaxies. If asignificant portion of the flux originates in a broad, out-flowing component, this might influence the estimates ofgalaxy physical properties.It is important to understand the effects of broad emis-sion on measurements of physical properties of galax-ies, particularly at z ∼
2, where they have higher starformation rate (SFR) (Reddy & Steidel 2009, Madauet al. 1998, Madau & Dickinson 2014), higher gas frac-tions (Daddi et al. 2010, Swinbank et al. 2011, Tacconiet al. 2013), and are more compact (Trujillo & Pohlen2005;Shen et al. 2003, Barden et al. 2005) compared togalaxies at z ∼
0. The higher SFR and smaller sizes leadto a larger star formation surface density (Σ
SFR ) whichmay result in an outflow (Ostriker & Shetty 2011, New-man et al. 2012). Studies of broad emission at z ∼ SFR ) of 1.0 M (cid:12) /yr/kpc (Newman et al. 2012). However, previous studies ofbroad emission at this redshift are based on a small sam-ple (Newman et al. 2014), are done on galaxies with highsSFR (Shapiro et al. 2009, Newman et al. 2012), or fo-cus on galaxies with AGN (F¨orster Schreiber et al. 2014,Genzel et al. 2011, Leung et al. 2017). A study of broad-ened emission with a large sample of typical galaxies at z ∼ − ) survey (Kriek et al. 2015) to study broadenedemission for a large sample of z ∼ − ∼ . < z < .
80. The data in theMOSDEF survey allow for measurements of broad emis-sion on a large sample of star-forming galaxies (as wellas AGN, presented in Leung et al. 2017). The goal ofthis paper is to measure or place limits on a broad com-ponent in star-forming galaxies in order to determine theamount of broad emission in typical z ∼ − http://mosdef.astro.berkeley.edu/ with Ω m = 0 .
3, Ω λ = 0 .
7, and H = 70 km s − Mpc − .All magnitudes are given in the AB system (Oke & Gunn1983). The wavelengths of all emission lines are in vac-uum. OBSERVATIONS, REDUCTION, AND GALAXYPROPERTY MEASUREMENTS
Observations, Reduction, and Sub-sample Selection
In this work, we use the first two years of data from theMOSDEF survey (Kriek et al. 2015) where we obtainednear-infrared spectra for ∼ . < z < .
80. The MOSDEF sur-vey targets galaxies in the AEGIS, COSMOS, GOODS-N, GOODS-S, and UDS extragalactic legacy fields whichhave extensive ancillary data including Chandra, Spitzer,Herschel,
HST , VLA, and ground based optical/near-IRdata.One-dimensional spectra were extracted using customIDL software called
BMEP , as described in the Appendix. BMEP was tested with output from the MOSDEF team’scustom 2D reduction, the MOSFIRE Data ReductionPipeline , and the 2D optical spectra from the Keck LowResolution Imaging Spectrometer (LRIS, Oke et al. 1995,Rockosi et al. 2010). For the MOSDEF data, both op-timally weighted and unweighted spectra were extractedfor each object, and we use the optimally weighted spec-tra for this analysis. The optimal extraction algorithmfollows Horne (1986) but is modified to be able to extractfractions of pixels (see the Appendix). To determine theweighting profile, center, and width of each object, wefit a Gaussian to the profile of each object in each fil-ter. The profile was determined by summing flux onlyat those wavelengths with high S/N in either the con-tinuum or emission lines. Using high S/N areas of thespectra creates clean weighting profiles for the optimalextraction since wavelengths with little or no signal areexcluded.Galaxies in the MOSDEF Survey are split into 3 red-shift bins, 1 . < z < .
70, 2 . < z < .
61, and2 . < z < .
80 that were each observed using a differentfilter set in order to maximize efficiency of detecting mul-tiple rest-optical emission lines of interest (see Kriek et al.2015, for details). We combine the 1 . < z < .
70 and2 . < z < .
61 galaxies into a single sample (hereafterthe z ∼ α and [O III ]. There may be some evolution betweenthe galaxies at these two redshift ranges, but withoutcombining them, broad emission is extremely difficult todetect. The 2 . < z < .
80 galaxies do not have cover-age of H α and are stacked separately (hereafter the z ∼ III ] detec-tions and 394 H α detections. We create a sub-samplewhere we remove galaxies for which it may be difficult toaccurately measure the broad emission or have AGN (dis- Source code and installation instructions available at: https://github.com/billfreeman44/bmep cussed below). When cleaning the sample, we considerthe H α and [O III ] detections separately except whenconsidering galaxy-wide effects which are mergers andAGN presence. • We remove 50 [O
III ] and 35 H α detections wherethe galaxy was an IR, X-ray, or both IR and X-ray detected AGN (see Coil et al. 2014 and Azadiet al. 2017). AGN are a possible source of outflows(Leung et al. 2017), and removing them allows usto isolate the effects of star formation on outflows. • We remove 20 [O
III ] and 26 H α detections wherethe galaxy would be classified as an AGN basedon z ∼ z ∼ z ∼ • We remove 182 [O
III ] detections where S/N < III ] and 122 H α detections where S/N < α . • We remove 22 [O
III ] and 41 H α detections wherethe [O III ] or H α emission line was on or near brightsky lines. • We remove 14 [O
III ] and 17 H α detections wherethe galaxy appears to be undergoing a merger asindicated from the images or spectra of nearby ob-jects. Mergers may have complicated kinematicsand may not be well fit by our fitting method(described in Section 3.1). We determined whichgalaxies were mergers by inspecting both imagesand spectra by eye. If the galaxy was very mis-shapen or there was a nearby companion we re-moved it. In the spectra, if there were two profilesthat overlapped then those galaxies were removed. • We remove 7 [O
III ] and 8 H α detections where the[O III ] or H α emission is near the edge of wave-length coverage and the shape of the profile is dif-ficult to determine. • We remove 5 [O
III ] and 7 H α detections for whichwe measured a FWHM >
275 from a single Gaus-sian. Some galaxies have broad lines simply fromrotation and velocity dispersion. To isolate thebroad emission, we restrict the narrow emission tohave a FWHM <
275 km s − as described in Sec-tion 3.1. Including galaxies with FWHM >
275 kms − creates false positives because the narrow com-ponent does not properly fit the narrow emission.The final sample has 216 unique galaxies with 203[O III ] measurements and 138 H α measurements. Thereare 125 galaxies with both an H α and [O III ] detection.We create stacks (discussed in Section 3.2) and the stacksat z ∼ α , [O III ], [S II ], [N II ],and H β , which results in 113 galaxies in the z ∼ z ∼ <
10 M (cid:12) yr − have been removed. The right side of Figure 1 shows theSFR vs. stellar mass diagram for our sample along withthe star-forming main sequence for MOSDEF galaxies inthe 2 . < z < .
61 range from Shivaei et al. (2015).Because we removed galaxies that have an H α or [O III ]S/N <
10, our sample may be incomplete for galaxiesbelow the main sequence and at low stellar mass.
Stellar Population Properties
We estimate the physical parameters for our sample,including stellar mass, SFR and age by comparing thephotometric SEDs with stellar population synthesis mod-els (Conroy et al. 2009) using the stellar populationfitting code FAST (Kriek et al. 2009). We assume aChabrier (2003) initial mass function (IMF) and the dustreddening curve from Calzetti et al. (2000). We use spec-troscopic redshifts from the MOSDEF survey and broad-band and mediumband photometric catalogs assembledby the 3D-HST team (Skelton et al. 2014) spanning ob-served optical to mid-infrared wavelengths. We includea template error function to account for the mismatchin less constrained sections of the spectrum. For a fulldescription of the stellar population modeling proceduresee Kriek et al. (2015).When available, we derive SFRs based on the H α emis-sion line by correcting for Balmer absorption (using theSED) and dust extinction (using the Balmer decrementof H α /H β ), then converting the H α luminosity into aSFR (Kennicutt 1998), adjusted for a Chabrier (2003)IMF (see Shivaei et al. 2015 for more details). Becausegalaxies in the z ∼ . α ,we use SED fitting to determine their SFRs. MEASURING THE BROAD COMPONENT
In this section, we describe the technique for measur-ing the broad emission line components for individualgalaxies. We also describe how we create stacks.
Fitting galaxies
We aim to measure an underlying broad componentof emission lines of galaxies. By assuming that emissionlines are composed of narrow and broad components withGaussian profiles. This section describes the fitting pro-cess as well as constraints on parameters.The H α , [N II ], and [S II ] lines are in the same filter(H if at z ∼ . z ∼ .
3) while [O
III ] and H β fall into a different filter (J if at z ∼ .
5, H if at z ∼ . z ∼ . . Additionally the see-ing may vary between different filters. We do not include[S II ] in fits of individual galaxies because it is too faint tomeasure the broad component, however it is included infits of stacks (see Section 3.2). For individual galaxies at1 . < z < .
61 we fit H α and [N II ] λλ . < z < . III ] λ III ] λ β simultaneously.For each set of lines, we perform two preliminary fitsand one final fit. The first preliminary fit uses a singleGaussian to fit each emission line using MPFIT , a non-linear least squares fitting code (Markwardt 2009). We N u m be r o f G a l a x i e s * )0.00.51.01.52.02.53.0 Log ( S F R ) Figure 1.
Left: Histograms of galaxy redshifts in the MOSDEF survey. The solid blue histogram is the full sample, the red is the sampleafter removing all galaxies with a S/N <
10 in H α (for z < .
3) or in [O
III ] (for z > . . < z < .
61 galaxies and purple triangles are 2 . < z < . use this single Gaussian fit to subtract off a linear con-tinuum and normalize the data so the fitted flux densityof the peak of the brightest line for each set of lines (H α or [O III ]) is unity, respectively. Next, we fit the dataagain using
MPFIT but this time each emission line is fitwith two Gaussians, one broad and one narrow. We usethe resulting values and errors of this second fit as ini-tial values for the final fit which is done with a customMarkov Chain Monte Carlo (MCMC) code
MPMCMCFUN that uses the Metropolis-Hastings algorithm (Metropo-lis et al. 1953, Hastings 1970). The errors in the secondpreliminary fit are used as the parameter jump ampli-tudes for the final fit. The final MCMC fit is necessarybecause it offers a better characterization of errors. Thisis especially relevant in these fits because we are fittingone emission line with two Gaussian components and thecorrelation between parameters may be significant. Inwhat follows, we use the subscripts S, B, and N to dis-tinguish parameters for the single, broad, and narrowcomponents, respectively.When fitting multiple emission line components, weconstrain the FWHM B , FWHM N , broad component shift(∆v), constant background, and the narrow componentredshift to be the same for each line. This leaves eachsingle emission line with two free parameters, broad am-plitude (A B ) and narrow amplitude (A N ). Two excep-tions to this are the [N II ] λ II ] λ III ] λ III ] λ .
93 and 2 .
98 respectively according to atomic physics(Osterbrock 1989). Therefore, each fit has five free pa-rameters shared by each line (FWHM B , FWHM N , ∆v,narrow component redshift, and constant background)and two free parameters for each line (A N and A B ).The resulting best-fit parameters are likely to dependon the chosen limits. For instance, not placing a mini-mum on the FWHM N can result in an unphysically nar-row emission line. Also, not placing a minimum on theFWHM B can result in the broad component not beingrepresentitive of broadened emission. Therefore, we placephysically motivated restrictions on all free parameters.For individual galaxies, we restrict the FWHM N so thatit cannot be lower than the average FWHM of skylinesin that particular filter and mask. For galaxies smaller Source code and installation instructions available at: https://github.com/billfreeman44/mpmcmcfun than the slit width, it is possible that the FWHM of thenarrow component is smaller than that of skylines. Inmost cases, the seeing is not smaller than the width ofthe slit (0 . (cid:48)(cid:48) N tobe much lower than the width of the sky lines. We alsorestrict the FWHM N to be less than 275 km s − . Forthis sample, we have removed galaxies where FWHM S islarger than 275 km s − (see Section 2.1).In order to properly study outflows, we must be cer-tain that the broad components measure a kinematicallydistinct feature from the rotation of the host galaxy. Inother words, the broad flux must not be an artifact froma better fit to the narrow emission by using two Gaus-sian components. Therefore, we restrict the minimumFWHM B to be a larger value than could be reason-ably fit using only a single Gaussian component. Ac-cordingly, we set the minimum FWHM B to be 300 kms − which provides some separation in the velocities ofthe narrow and broad components. Typical FWHMs forionized outflows from star-forming galaxies are 300-600km s − (Newman et al. 2012, Genzel et al. 2011, Woodet al. 2015). Some studies have measured galactic out-flow speeds > − , but these are typically as-sociated with AGN (Shapiro et al. 2009, Genzel et al.2014a, F¨orster Schreiber et al. 2014). Since we have re-moved known AGN, we do not expect outflows of suchhigh velocity. The upper limit of FWHM B is set to 850km s − which is the typical maximum velocity deducedby the blue-shifted interstellar absorption lines in the restframe UV of z ∼ ±
100 km s − of the expected value. Thebroad component shift is the same for each line. Noobjects that had significant detections of the broad com-ponent ran into this limit. Other studies typically findshifts of <
100 km/s (Newman et al. 2012, Wood et al.2015).The amplitudes of the narrow H α and [O III ] compo-nents are constrained to be between 0.2 and 1.05 (the
Table 1
Constraints of the FitsParameter Minimum MaximumFWHM S a varies b S c S d -100 +100FWHM N varies b N N -100 +100FWHM B
300 850A B -0.3 0.8∆v B -100 +100 a FWHM in units of km s − Set to the average FWHM of skylinesin each mask c Relative to the maximum flux of theline d Center shift in units of km s − . Neg-ative values imply blueshifts peak of these lines were normalized to unity from thesingle Gaussian fit) and the broad component amplitudeis constrained to -0.3 and 0.8. Since some galaxies donot show any signs of broad emission, the best value forthe FWHM B could be at or near zero for these galaxies.In these cases, it is still useful to put limits on the broademission. Therefore, we allow the FWHM B to be nega-tive to fully sample the parameter space and set properupper limits on the FWHM B . In cases where the bestvalue for the FWHM B is less than zero, we interpret thisgalaxy as having no significant broad emission but stillshow the upper limit. For [N II ], [S II ], and H β we scalethe restrictions to the relative peak of each line. All ofthe constraints are listed in Table 1.Figure 2 shows fits for four galaxies that exhibit thestrongest evidence for a broad component. It is clearthat a single Gaussian does not fit these galaxies well asevidenced by the “wave” pattern that is present in theresiduals. The pattern shows the single fits underesti-mate flux at the peak, overestimate the wings, and under-estimate the base. This pattern is particularly evident inthe [O III ] lines of COSMOS-12015, and COSMOS-13015and in the H α lines of GOODS-N-12024 and GOODS-N-7231.The broad flux only dominates a small fraction of theline at high velocities. If the broad flux component isnot approximately Gaussian, then the fitted broad fluxmight be different from what we measure. Other studiesthat analyze galaxies with higher signal to noise find thatthe broad emission is typically well fit by a Gaussian(Newman et al. 2012, Shapiro et al. 2009, Genzel et al.2011). Making Stacks of Spectra
The broad component is difficult to separate from thenarrow emission. The galaxies in Figure 2 were chosenbecause they show the strongest evidence for broad emis-sion. The faint, high velocity wings of the broad com-ponent are difficult to distinguish from the noise for themajority of the individual galaxies. In order to achieve ahigher S/N, we create stacks of galaxies in bins of stellarmass such that each stack has approximately the samenumber of galaxies. To create each stack, we interpolatethe flux for each galaxy to a common rest-frame wave-length grid, subtract off any continuum, convert each spectrum from flux density to luminosity density, divideby the total luminosity of either H α or [O III ] depend-ing on the wavelength coverage of the stack, and thensum each spectrum with no weighting. To avoid addingsignificant noise from sky line subtraction residuals, weremove pixels associated with sky lines where the errorspectrum is above 1.5 × the median error. The error inthe stacked spectrum is calculated by making the stack200 times but using input spectra with added Gaussiannoise according to the associated error spectrum for eachindividual object; the error is calculated by taking thestandard deviation of the 200 stacks at each wavelength.The z ∼ SF R . These stacks are nottruly independent from the stacks by mass because theseproperties correlate with mass. We chose to use mass forthe primary analysis in this work because, of the physicalparameters we considered, mass is the only parameterthat is not estimated by using the H α emission line whichmay be influenced by broad emission.The line fitting process for stacks is the same as de-scribed in Section 3.1 with some exceptions. The doublet[S II ] λλ α and [N II ].For the lower limit of FWHM N , we use the average sky-line FWHM for each galaxy, which is 80 km s − . The fitto each stack is plotted in Figure 3. This figure showsH α and [N II ] for the stacks at 1 . < z < .
61. For bothH α and [O III ] in all stacks, the amplitude of the broadcomponent is significant at the > σ confidence level. Assessment of False Positives
The broad component dominates the line only at thehighest velocities, which is also where the S/N is thelowest. Here we test the fitting process to show thatmeasured broad line parameters are consistent with sim-ulated input parameters.For this test, we take a single Gaussian, add noise, andfit the Gaussian using the method described in Section3.1. Since this idealized Gaussian has no actual broadcomponent, any broad component that we measure is afalse positive. We performed this test on 200 simulatedemission lines, each with 10 different FWHMs between75 and 275 km s − , which span the range of measurednarrow components from the MOSDEF sample. We usedthe same resolution and wavelength as for an H α linefor a z ∼ . α lineranges between 10 and 300. The assumption of constantnoise is an appropriate approximation of the error fora single emission line in the actual spectra because weare limited by the bright sky rather than Poisson noisefrom the object. The noise does increase as a functionof wavelength, particularly in the K band, but this testwas only on a single emission line and the error does notchange much over the span of a single emission line.We find that only 11%, 1%, and 0.05% of simulatedgalaxies have a false positive of 1 σ , 2 σ , or 3 σ respec-tively. These are lower than the expected rates based onGaussian statistics, which are 16%, 2%, and 0.1%. Theslightly lower values of the test result are because we al-low the narrow peak to exceed 1.0 (the max is 1.05). Theaverage broad component is slightly less than 0 whichcreates an offset in the number of 1 σ , 2 σ , or 3 σ detec- −0.20.00.20.4 −0.20.00.20.4 N o r m a li z ed F l u x [OIII]12015COSMOS H β H α , [NII] −0.20.00.20.4 N o r m a li z ed F l u x [OIII]13015COSMOS H β H α , [NII] −0.20.00.20.4 N o r m a li z ed F l u x [OIII]12024GOODS−N H β H α , [NII] 4995 5005 5015 Wavelength (Å) −0.20.00.20.4 N o r m a li z ed F l u x [OIII]7231GOODS−N 4850 4860 4870 Wavelength (Å) H β α , [NII] Figure 2.
Four example fits for individual spectra. Each row is one object and each column from left to right is [O
III ], H β , and H α .The field and 3D HST v2.1 catalog ID is in the upper left. Each line is normalized such that the strongest line peak is unity. The singleGaussian fit is shown in green. The overall fit for the narrow+broad fits for each stack is shown in purple with the broad component forthis fit shown in red. The error spectrum is shown as a dotted blue line. The two gray lines show the residuals for the single Gaussianfit (top) and the narrow+broad fit (bottom). The horizontal solid black lines show the amount each residual is offset. A skyline has beenmasked for GOODS-N 12024 at 5004 ˚A and at 4877 ˚A. −0.10.00.10.20.30.4 −0.10.00.10.20.30.4 N o r m a li z ed F l u x Mass=9.56 χ r,s =0.875 χ r,d =0.775H α −0.10.00.10.20.30.4 N o r m a li z ed F l u x Mass=9.56 χ r,s =0.875 χ r,d =0.775[NII] −0.10.00.10.20.30.4 N o r m a li z ed F l u x Mass=9.56 χ r,s =0.876 χ r,d =0.831[OIII] −400 0 +400 Velocity (Km/s) −0.10.00.10.20.30.4 N o r m a li z ed F l u x Mass=9.56 χ r,s =0.876 χ r,d =0.831H β Mass=9.95 χ r,s =1.01 χ r,d =0.915H α Mass=9.95 χ r,s =1.01 χ r,d =0.915[NII] Mass=9.95 χ r,s =1.06 χ r,d =0.883[OIII] −400 0 +400 Velocity (Km/s) Mass=9.95 χ r,s =1.06 χ r,d =0.883H β Mass=10.3 χ r,s =1.44 χ r,d =0.966H α Mass=10.3 χ r,s =1.44 χ r,d =0.966[NII] Mass=10.3 χ r,s =1.25 χ r,d =1.08[OIII] −400 0 +400 Velocity (Km/s) Mass=10.3 χ r,s =1.25 χ r,d =1.08H β Figure 3.
Stacks of galaxies showing both the single Gaussian and narrow+broad component fits. The rows show H α ,[N II ], [O III ], andH β lines from top to bottom. The columns show each stack with the stellar mass increasing from left to right. The line colors have thesame meaning as in Figure 2. χ r,s is the reduced χ for the single fit and The χ r,d the reduced χ for the narrow+broad fit. tions. Another result of this test is that the fraction offalse positives did not change as a function of width ornoise added.In addition to false positives in individual spectra,there is a possibility of creating an artificial broad com-ponent when making the stacks. We performed severaltests to ensure that in creating the stacks, we did not alsocreate an artificial broad component. The first test takes50 Gaussians of random FWHM between 75 and 250 kms − , adds noise, and creates a stack as described in Sec-tion 3.2. Each added Gaussian has a constant amount ofnoise across each wavelength element which is similar tothe level in actual spectra except for skylines which arenot included in these simulated stacks. We found no evi-dence of introducing false positives when creating stacks.We repeat this test and add a random shift between ± × − which is ∼ .
13 ˚A at these redshifts.These stacks also failed to produce false positives.These tests have shown that we do not expect falsepositives to be an issue when using the fitting methoddescribed in Section 3.1. We have also shown that creat-ing stacks of galaxies does not introduce a broad emissionsignature. RESULTS
In this section, we discuss the broad flux measured inindividual and stacked spectra. We discuss the physicalinterpretation of these measurements in the subsequentsection.
Broad Flux Ratio
After fitting each galaxy and stack, we parameterizethe broad emission we measured as a broad to narrowflux ratio (broad flux / narrow flux, BFR). We chosethis parameterization because other studies have usedthis and using the same parameterization allows for easycomparision (e.g. Newman et al. 2012). The BFR isalso used to estimate the mass loading factor (Section5.3). The other natural parameterization, broad flux tototal flux, can be calculated as (broad flux/total flux) =1 / (1 + BFR − ).The left side of Figure 4 shows the BFR measured fromthe H α line as a function of mass. For individual galax-ies, there are 10 detections with > σ significance out of138 galaxies (7%). For the stacks and the galaxies withdetections, the broad flux accounts for 10-70% of thetotal flux in nebular emission lines. The MOSDEF mea-surements for BFR are consistent with the measurementsfrom Newman et al. (2012) who did a similar analysisfor galaxies at the same mass range. The details of thefits are in Table 2. The small differences in the numberof significant detections in this study and Leung et al.(2017) for the same sample can be attributed to slightdifferences in codes used to fit the data.The right side of Figure 4 shows the BFR measuredfrom the [O III ] lines as a function of mass. For [O
III ]there are 21 detections with > σ significance out of201 galaxies (10%). For the stacks and galaxies withdetections, the broad flux accounts for 20-50% of thetotal flux in nebular emission lines. For the z ∼ z ∼ <
275 km s − andreliably detecting low velocity broad emission is difficult(discussed in detail in Section 5.3). Non-measurement ofthe broad fluxes for the lowest mass galaxies may intro-duce a bias in the BFR vs. stellar mass relation. Ad-ditionally, the [O III ] broad emission at z ∼ × M (cid:12) , and the z ∼ α line asa function of S/N. Here, we define the S/N as the fittedflux by a single Gaussian divided by the error in the fluxfor either H α or [O III ] λ α , 66% of galaxieswith S/N >
70 have broad component detections butonly 1.6% of galaxies with S/N <
70 have detections.For [O
III ], 32% of galaxies with S/N >
45 have broadcomponent detections but only 5% of galaxies with S/N <
45 have detections. These two thresholds were chosenby eye to emphasize the S/N dependence on detecting thebroad component. It is easier to detect broad emission in[O
III ] than in H α because we include the [O III ] λ III ] λ z ∼ Broad and Narrow Component Line Ratios
As described in Section 3.1, we fit narrow and broadcomponents to the [O
III ], H β , H α , [N II ], and [S II ] emis-sion lines in stacked spectra. From this analysis, we areable to calculate the [N II ]/H α , [S II ]/H α , and [O III ]/H β ratios and place each component on the N2-BPT andS2-BPT diagrams. Figure 6 shows the N2-BPT and S2-BPT diagram for the low, medium, and high mass stacks.We do not include individual galaxies here because therewere not enough 3 σ detections of the broad componentsof H β , [N II ], and [S II ] and we could not create robustline ratios. The blue dashed line is measured from Kew-ley et al. (2013) for local galaxies. The orange dashedline is measured from Shapley et al. (2015) for z ∼ . H α Figure 4.
The BFR as a function of mass for H α (left) and [O III ] (right). Red squares are galaxies with a broad component detection of > σ significance with 1 σ error bars plotted. Orange triangles are 3 σ upper limits for galaxies with < σ significance. Blue stars show theBFR of the z ∼ z ∼ σ error bars from the fit and the horizontal dashed lines show the range of points included. Figure 5.
The BFR as a function of S/N for H α (left) and [O III ] (right). Red squares are galaxies with a broad component detection of > σ significance with one sigma error bars plotted. Orange triangles are 3 σ upper limits for galaxies with < σ significance. Vertical linesare drawn at S/N=70 for H α and S/N=45 for [O III ]. For H α , 66% of galaxies with S/N >
70 have broad component detections but only1.6% of galaxies with S/N <
70 have detections. For [O
III ], 32% of galaxies with S/N >
45 have broad component detections but only 5%of galaxies with S/N <
45 have detections. The location of the vertical lines was chosen by eye to emphasize the dependence of detectingbroad flux and S/N. The dependence of the detection of the broad flux on S/N implies that the 10% detection rate is a lower limit. mated from the stellar population synthesis model fit tothe SED of each galaxy (Reddy et al. 2015). For eachstack, we estimate the Balmer absorption by calculatingthe average absorption for each galaxy in the stack. Thisgives us an estimate for the total fraction of flux that wasabsorbed but no information about the shape. Withoutknowing the exact shape/width of the absorption feature,we do not know how much of the correction should beapplied to the narrow feature and how much should beapplied to the broad feature. Therefore, we calculate the Balmer absorption correction assuming the broad com-ponent is affected by 0, 33, 66, and 100% of the Balmerabsorption and the narrow component is affected by 100,66, 33, and 0% respectively. This gives a general ideaof the most the Balmer absorption could affect each lineratio. In Figure 6, the 0% and 100% absorption casescorrespond to the hollow point and the solid point fur-thest from the hollow point, respectively. For the singleGaussian fits (square points) there is only one solid pointbecause the Balmer emission is not split between narrow0and broad components and the magnitude of the Balmerabsorption correction is unambiguous. The shape of theBalmer absorption may also affect the fits in a mannerthat is difficult to predict.In Figure 6, the ratios from the single Gaussian fits(squares) are lower than results from previous MOSDEFstudies (Shapley et al. 2015). This can be explained bythe fact that we required a S/N >
10 for the H α and[O III ] lines. This requirement preferentially removedlower mass galaxies which are typically more offset fromthe local relation (Shapley et al. 2015).The narrow component ratios tend to lie more towardsthe z ∼ z ∼ .
3. The broadcomponents of the narrow+broad fits (diamonds) lie inthe composite region or above the Kewley et al. (2001)line. The Balmer correction is large for the [O
III ]/H β ratio and this makes it difficult to conclude if the broadcomponents have higher [O III ]/H β than their narrowcounterparts. The Balmer correction is smaller for the[N II ]/H α ratio and it is clear that the broad componentshave higher [N II ]/H α than the narrow components evenafter Balmer absorption correction.The right side of Figure 6 shows the S2-BPT diagramfor each stack and for each component. The [S II ] linetypically has less flux than the [N II ] line making measur-ing the broad component more difficult. We are only ableto place 1 σ limits on the broad components of the [S II ]line in stacks. Nevertheless, these limits are consistentwith a higher [S II ]/H α ratio for the broad components.In addition to the fits, we calculated line ratios by in-tegrating the flux in several velocity bins with respect tothe centroid of the line. This measurement provides anon-parametric estimation of the line ratios that is inde-pendent of any model and is shown in the Appendix. Wefound that the higher velocity bins generally had higher[N II ]/H α and [O III ]/H β ratios which is consistent withwhat we measure with the fits. DISCUSSION
In this section we discuss possible origins of the broadflux emission that can explain the offset line ratios ofthe broad component compared to the narrow compo-nent. We consider shocks (Section 5.1) and low lumi-nosity AGN (Section 5.2). We also interpret the broademission as an outflow and estimate the mass loadingfactor for the stacks (Section 5.3).
Shocks
Emission line ratios from shocks differ from ratios inphotoionized gas. Shock-heated gas can become ionizedby high-energy photons from the shock or excited bycollisions. Emission line ratios shift in the presence ofshocks and the magnitude and direction of the shift de-pends on the metallicity, electron density, magnetic field,and shock velocity (Allen et al. 2008). Shocked emissiontends to have higher [N II ]/H α and [S II ]/H α ratios rel-ative to what is produced in photoionized HII regions(Allen et al. 2008). Since the broad components in Fig-ure 6 have higher [N II ]/H α ratios than the narrow com-ponents or single Gaussian fits, this may indicate thepresence of shocks. In this section, we investigate if thebroad emission can be explained by shocks by creatingthe N2-BPT and S2-BPT diagrams using data from the shock models by Allen et al. (2008) and comparing thesemodels to the broad emission line ratios.The shock models simulated emission line ratiosfor shocked gas, the precursor to the shock, and ashock+precursor which combines the shock and precur-sor components. The precursor is material that is pho-toionized by the shock but not directly shocked itsself.Because we do not spatially resolve the emission fromthese galaxies, we are unable to separate the differentcomponents of the shock. Therefore, we compare ourmeasurements of the broad emission to the combinedshock+precursor ratios.Allen et al. (2008) measured shock+precursor emis-sion line ratios for two sets of models, one at a fixedelectron density with varying metallicity (n e = 1 cm − at log(O/H)+12 of 8.03, 8.35, 8.44, and 8.93), and an-other at fixed metallicity with varying electron density(log(O/H) + 12 = 8 .
93 at n e = 1 , , , − ).We restrict the models shown to those that have a mag-netic field strength at pressure equipartition. The shockvelocity for the models range from 100 − − , butwe only show shock velocities of 200-500 km s − basedon the velocities measured in Table 2.In Figure 7, we show the shocked models for the N2-BPT and S2-BPT diagrams. The top row shows the ef-fect of changing metallicity on shocked diagnostic ratios,and the bottom row shows the effect of changing density.The galaxies in this sample (with S/N of [N II ] >
3) havea median metallicity of log(O/H) + 12 = 8 .
43 with 80%of galaxies between 8 . < log(O/H) + 12 < .
59 calcu-lated using the [N II ]/H α ratio as in Sanders et al. (2015).The electron density of the MOSDEF galaxy sample at2 . < z < .
61 is 290 +88 − cm − (Sanders et al. 2016).This was calculated using the entire [S II ] line and theelectron density of the material causing the broad emis-sion may be different. Newman et al. (2012) measured adensity of 10 +590 − cm − from a stack of 14 galaxies, andthis value is consistent with our assumption of 290 cm − .Since none of the simulations span exactly the range ofmetallicities and densities of the MOSDEF galaxies, weare forced to extrapolate between the effects of metallic-ity and density. We highlight the point that is the bestmatch to the metallicity, electron density, and shock ve-locity in green. This green point corresponds to the shockmodel which has ( v = 300 km s − , log(O/H)+12 = 8.44,n e = 1 cm − ) in the top row and ( v = 300 km s − ,log(O/H)+12 = 8.93, n e = 100 cm − ) in the bottomrow.In Figure 7, there is a strong metallicity dependenceon the [N II ]/H α ratio and there is almost no changeas a function of electron density except at the high-est density (n e = 1000 cm − ) where [N II ]/H α de-creases. The broad lines measured from stacks are con-sistent with the [log(O/H)+12=8.44, n e = 1 cm − ] and[log(O/H)+12=8.35, n e = 1 cm − ] points. The shockmodel that best matches the physical parameters (greenpoint) is very near the broad emission line ratios. There-fore, it is feasible that the positions of the broad com-ponents in the N2-BPT diagram can be explained byshocks.In the top row of Figure 7, the model that best matches http://cdsweb.u-strasbg.fr/~allen/shock.html −1.5 −1.0 −0.5 0.00.20.40.60.81.01.2−1.5 −1.0 −0.5 0.0log ([NII]/H α )0.20.40.60.81.01.2 l og ( [ O III]/ H β ) Low Mass stackMed Mass stackHigh Mass stackBalmer Corrected
NarrowComp.BroadComp. SingleFit −1.0 −0.8 −0.6 −0.4 −0.2 0.0log ([SII]/H α ) Low Mass stackMed Mass stackHigh Mass stackBalmer Corrected
BroadComp. NarrowComp. SingleFit
Figure 6.
The N2-BPT and S2-BPT diagram for z ∼ . σ and galaxies with S/N < σ limits and are marked by arrows. The blue dashed line is measured from Kewley et al.(2013) for local galaxies. The orange dashed line is measured from Shapley et al. (2015) for z ∼ . the MOSDEF data has a higher [S II ]/H α ratio than anyof the broad emission. This may be due to the limitationthat the models with varying metallicity have an electrondensity of 1 cm − . The models in the bottom row showthat [S II ]/H α decreases as electron density increases.Since the electron density of the MOSDEF galaxy sampleis 290 +88 − cm − the models in the top row would likelyshift to lower [S II ]/H α ratios at higher electron densities.Therefore, it is possible that the positions of the broadcomponents in the S2-BPT diagram can be explained byshocks.From these line ratios, we conclude that it is possi-ble that the broad emission is a result of shocked emis-sion. This explaination does not rule out other sourcesof emission such as AGN (discussed in Section 5.2) andphotoionized outflows (discussed in Section 5.3). AGN
When creating the sample presented in this work, weremoved all X-ray, IR, and optically identified AGN be-cause our goal is to study star formation driven outflows,and AGN are also known to drive outflows at z ∼ II ]/H α and [O III ]/H β ratios because of harder ionization coming from the ac-cretion disk, which is consistent with the line ratios wefind in the broad component.With integral field spectroscopy it is possible to cre-ate spatially resolved line ratios (Newman et al. 2014,Wright et al. 2010), which can be used to determine if a galaxy in the “composite” region of the N2-BPT dia-gram has an AGN. Using spatially-resolved emission linemaps, Newman et al. (2014) found some galaxies that liein the composite region of the BPT diagram host AGN.The cores of these galaxies lie in the AGN region whilethe outer edges lie in the star-forming region. The high[N II ]/H α and [O III ]/H β ratios of the core indicatedthe presence of an AGN that would not be detected inspectra of low spatial resolution.It is unlikely that we are detecting outflows driven pri-marily by AGN because AGN-driven outflows have moreextreme kinematics compared to star-formation drivenoutflows. AGN driven outflows are typically 500-5000km/s (Genzel et al. 2014a, F¨orster Schreiber et al. 2014,Leung et al. 2017) which is much faster than typical out-flow velocities from star-forming galaxies (300-550 km/s)(Shapiro et al. 2009, Newman et al. 2012). The velocitydifference between the narrow and broad components inAGN driven outflows is 100-500 km/s (Leung et al. 2017)while star-formation driven outflows typically have veloc-ity offsets of <
100 km/s (Newman et al. 2012).Although we have removed all X-ray, IR, and opticallyidentified AGN from this sample, we can not completelyrule out some contribution to the emission lines fromlow mass, low luminosity black holes. We used an ex-tremely conservative cutoff for optically identified AGNcandidates. If low-luminosity AGN mixing is a signifi-cant source of emission in the stacks, then the presenceof AGN would have to be extremely widespread among z ∼ z ∼ z ∼
0, weak AGNwould not be bright enough to significantly change lineratios (Coil et al. 2014). It seems much more likely thatshocks could be commonplace at z ∼ −1.5 −1.0 −0.5 0.0 0.00.20.40.60.81.01.2 −1.5 −1.0 −0.5 0.0 log([NII]/H α )0.00.20.40.60.81.01.2 l og ( [ O III]/ H β ) log(O/H)+12=8.03, n e =1log(O/H)+12=8.35, n e =1log(O/H)+12=8.44, n e =1log(O/H)+12=8.93, n e =1
500 km/s400 km/s300 km/s200 km/s −1.2 −1.0 −0.8 −0.6 −0.4 −0.2 0.0 0.2log([SII]/H α ) −1.5 −1.0 −0.5 0.0 log([NII]/H α )0.00.20.40.60.81.01.2 l og ( [ O III]/ H β ) log(O/H)+12=8.93, n e =1000log(O/H)+12=8.93, n e =100log(O/H)+12=8.93, n e =10log(O/H)+12=8.93, n e =1 500 km/s400 km/s300 km/s200 km/s −1.2 −1.0 −0.8 −0.6 −0.4 −0.2 0.0 0.2log([SII]/H α ) Figure 7.
Shock models by Allen et al. (2008) for the N2-BPT and S2-BPT diagrams overlaid on the line ratios measured from the stackswhich are shown using the same symbols as in Figure 6. The models in the top row change in metallicity and the models in the bottom rowchange in electron density (in the legend, the units for electron density are cm − ). The green point corresponds to the shock model whichhas ( v = 300 km s − , log(O/H)+12 = 8.44, n e = 1 cm − ) in the top row and ( v = 300 km s − , log(O/H)+12 = 8.93, n e = 100 cm − ) inthe bottom row. This point is the best match to the metallicity, electron density, and shock velocity for the entire sample. SFRs, instead of AGN being ubiquitous. If there is anyAGN contribution it is likely small.
Outflows
One interpretation of the broad component is that ittraces ionized outflowing materials (Bland & Tully 1988,Heckman et al. 1990, Gergeev 1992, Phillips 1993, Lehn-ert & Heckman 1996, Veilleux et al. 2001, Colina et al.2004, Westmoquette et al. 2007, Westmoquette et al.2008, Shapiro et al. 2009, Newman et al. 2012, Rupke &Veilleux 2013, Genzel et al. 2014b, Feruglio et al. 2015,Leroy et al. 2015). In this section, we interpret the broadcomponent as a photoionized outflow, calculate the massloading factor η (outflow mass rate/SFR), and compareto other observations as well as simulations.Using some assumptions about the outflow velocity, ra-dius, temperature, and density we can convert the BFRinto an estimate of η . We adopt the outflow model fromGenzel et al. (2011). This model assumes the broad com-ponent to be photoionized and the emission of H α to bea result of case-B recombination. The model assumes a spherical outflow with a constantvelocity (cf. Steidel et al. 2010). The mass outflow rate,˙ M out can be calculated as:˙ M out = 1 . m H γ H α n e (cid:18) L Hα F broad F narrow (cid:19) V out R out (1)where m H is the atomic mass of hydrogen, V out is thevelocity of the outflow, R out is the radius, γ H α is theH α emissivity, n e is the electron density in the outflow, L Hα is the total extinction corrected H α luminosity, and F broad /F narrow is the BFR.We attempt to measure each component of Equation1 from our stacks (as described below), but sometimeswe do not have sufficient signal to do so. For physicalparameters we cannot estimate, we adopt values fromNewman et al. (2012) (hereafter N12) who preformed asimilar analysis on 27 star-forming galaxies at z ∼ ∼
90 M (cid:12) /yr on average)which may lead to physical differences. However, N12 is3currently the most similar study to ours with measure-ments of the parameters in Equation 1, and we use theirvalues when we are unable to measure them from ourdata.The electron density for the outflow can be measuredusing the broad [S II ] λ II ] λ II ] lines is lowwhich results in a large measurement uncertainty in theratio. We are unable to constrain the density using thebroad component from this work. We adopt the valueused by N12 of 50 +550 − cm − .The term V out /R out is the inverse of the characteristictimescale of the outflow. In an attempt to measure theradius of the outflow we made a stack of the 2D spectraand attempted to find broad flux in the spatial direction(e.g. Martin 2006, Leung et al. 2017). We were unable tomeasure a spatially extended component in the stackedspectrum. For R out we adopt the value of 3 kpc as mea-sured by N12. This value is reasonable given the angularsize at this redshift is ∼ − and our best see-ing is ∼ . (cid:48)(cid:48)
6. For V out , we use the “maximum” velocityof the outflow defined as V max = | ∆v B - 2 σ B | (Genzelet al. 2011, Wood et al. 2015). This value representsthe velocity of the outflow if one assumes the outflow isspherically symmetric with a constant velocity.We use an H α emissivity of 3 . × − erg cm s − which assumes an electron temperature of T e = 10 K.If we use the Kennicutt (1998) relation betweenSFR and H α luminosity corrected for a Kroupa IMF(SFR[M (cid:12) yr − ] = 7 . × − L Hα [ergs s − ]), we candivide Equation 1 by SFR and simplify to: η ≈ . (cid:18)
50 cm − n e (cid:19) (cid:18) V out
300 km s − (cid:19) (cid:18) R out (cid:19) (cid:18) F broad F narrow (cid:19) (2)Figure 8 shows η calculated for each stack and thevalues are listed in Table 2. The error calculation in-cludes measurement uncertainties from the BFR and theFWHM B . We do not include errors in the radius, elec-tron density, and temperature assumed, and includingthese errors would increase the error on the mass load-ing factor by at least an order of magnitude. Figure 8also includes the mass loading factor measured by N12.The measurements from N12 are higher than those fromthe MOSDEF stacks despite having similar BFR mea-surements. N12 use V out = 400 km s − but we useV out = | ∆v B - 2 σ B | which results in a lower velocitycompared to N12 by 50-100 km s − .We compare our results to the FIRE cosmologicalgalaxy formation simulations with explicit stellar feed-back (Hopkins et al. 2014, Muratov et al. 2015). Inter-estingly, η increases as a function of mass which is con-trary to what we expect from the FIRE simulations (Mu-ratov et al. 2015) but are consistent with the results fromN12. This difference is likely explained by our inabilityto detect low velocity outflows. The speed of outflowsincreases as a function of SFR and galaxy stellar masswhich is seen in observations (Martin 2005, Weiner et al.2009) and simulations (Muratov et al. 2015, Christensenet al. 2015). At low outflow speeds (FWHM B <
275 km
Full sample stacksN12 stacksFIRE Simulations
Figure 8.
The mass loading factor as a function of mass. The z ∼ η from N12 are shown as black circles. We also include the η vs. mass relationship found from the FIRE simulations (Muratovet al. 2015). The conversion from BFR to η is described in detailin Section 5.1. s − ), the emission from the outflow may be indistinguish-able from the emission from the HII regions. To quantifythe detectability of low velocity outflows we created twotests using simulated spectra where we can control theBFR and FWHM B of galaxies used to make stacks. Wecan then measure the BFR and FWHM B of the stacksand check if they are representative of the input galaxyparameters.In the low velocity test (Test 1), the input FWHM B increase from 100 to 800 km s − between stellar masses of10 M ∗ and 10 M ∗ . For Test 1, galaxies below 10 M ∗ have velocities below 300 km s − . In the high velocitytest (Test 2), the input FWHM B increase from 275 to550 km s − as a function of stellar mass. Both tests usethe same distribution of BFRs and a FWHM N of 200 kms − . The slope of the BFR vs. M ∗ for the tests is thesame as the slope of η vs. M ∗ form FIRE simulations(eq. 8 from Muratov et al. (2015)). We normalize thisrelation using simplified version of our equation 2, η =C ∗ BFR and determine constant C using our highest massstack (BFR=0.68 at log M ∗ =10.4), where the effects oflow velocity outflows should be minimal. We use thedistribution of SFRs and stellar masses of the sample(shown in Figure 1) to make stacks by mass and by SFR.The results of these tests are shown in Figure 9. InTest 1, stacks underestimated the BFR for galaxies inthe low and medium mass stacks. The stacks in Test 1show an increase in BFR as a function of mass despitethe input galaxies having a decrease with mass. The lowand medium mass stacks contain 100% and 55% galaxieswith FWHM B <
275 km s − respectively. The FWHMmeasured for these stacks is too large compared to theinput galaxy values although this is expected because wedo not allow the FWHM B parameter to go below 300 kms − (see Section 3.4). The stacks in the test show that we4could underestimate the extent of broad flux when theFWHM B is <
275 km s − for a large fraction of galaxieswithin the stack. We cannot measure broad flux at lowvelocities with the measurements made in this work butthis is possible with the data in the FIRE simiulation.In Figure 10 we show the fraction of outflowing ma-terial above 275 km s − versus stellar mass as expectedfrom FIRE simulations. To calculate these fractions weuse the fit to median velocities as a function of halo cir-cular velocity from Muratov et al. (2015). We assume lognormal velocity distribution at a given circular velocitythat matches the 25-75% velocity distribution range inMuratov et al. (2015) (their Figure 8). This enables usto calculate the expected fraction of outflows above 275km s − . We convert the circular velocity to halo massand then to stellar mass using relations from Behroozi etal. 2013 and show the calculated fraction as a functionof M ∗ for several different redshifts. Direct comparisonof our results with FIRE simulation is difficult becausethey measure outflows at 1/4 of the virial radius and in-clude all possible outflow phases. Nevertheless, if the out-flows in the observed high-z galaxies are similar to thosein FIRE simulations, it is clear that the high-velocitythreshold for broad component will miss the bulk of theoutflows in lower mass galaxies. To summarize, the massloading factor in the low mass galaxy stack in Figure 8 iscompatible with the results from the FIRE simulationsonly if the outflows for low mass galaxies ( 1 < M (cid:12) )have low velocities (FWHM B <
275 km s − ).The mass loading factor for the highest mass stack( η = 1 . +0 . − . ) is lower than the predicted value fromFIRE at that mass ( η = 2 . B measuredfor this stack is 340 ±
30 km s − so the difficulty of de-tecting low velocity outflows should not be a factor (Ta-ble 2). Using a smaller electron density or smaller radiusin Equation 2 would bring these into better alignmenthowever there is no evidence to justify such changes. Analternate explanation for why η in Figure 8 is lower thanexpected is that some fraction of the outflow is neutraland not visible as a broad H α component. Outflowsmeasured in the H α line are a measure of the ionizedcomponent of the outflow, but outflows are multi-phaseand have neutral, ionized, and dusty components (Leroyet al. 2015, Wood et al. 2015, Feruglio et al. 2015). Somestudies of a small number of local galaxies have mea-sured the neutral phase to have 9 − × as much mass asthe ionized phase (Wood et al. 2015, Mart´ın-Fern´andezet al. 2016). Additionally, Martin (2006) measured bothNa I absorption and H α emission in 18 ultra-luminousinfrared galaxies and found that there was no correlationbetween the strength of the Na I absorption and the ex-tended H α emission. An undetected neutral componentmay account for the factor of 2 difference between whatis measured here and the FIRE simulations.The assumptions that the broad component is an out-flow and that the broad component is shocked are notmutually exclusive. The broad component could be ashocked outflow. If we assume a 100% collisionally-ionized outflow, the mass loading factor would be smallerby a factor of ∼ η at this B F R (Mass)1001000 F W H M [ k m s − ] (SFR) Test 1 inputTest 2 inputTest 1 Stack FitTest 2 Stack Fit
Figure 9.
Results from two tests where we added broad compo-nents to simulated spectra, created stacks, and fit the stacks usingthe method described in Section 3.1. The BFR for both tests areidentical, but the FWHM B for Test 1 ranges from 100-900 km/sand for Test 2 ranges from 260-500 km/s. The input BFR andFWHM B are linear with respect to mass. The left column showsfits to stacks by mass and the right column shows fits to stacksby SFR. When the FWHM of the broad component is below 275km/s the BFR is underestimated. redshift (Newman et al. 2012). We assumed the elec-tron density and geometry of the outflow were the sameas those of other studies (Newman et al. 2012) since wewere not able to measure these parameters with our sam-ple. At low masses, η increases as a function of masswhich is contradictory to the results of the FIRE simu-lations (Muratov et al. 2015, Hopkins et al. 2014) wherea decrease with mass is predicted. This difference is bestexplained by our inability to detect contributions fromlow velocity (FWHM <
275 km s − ) broad componentsas shown in Figure 9. Another feasible explanation isthat the outflows have a large neutral component whichis not detected because broad H α emission is sensitive tothe ionized component of the outflow. IMPLICATIONS OF BROADENED EMISSION ONESTIMATING PHYSICAL PROPERTIES OF GALAXIES
Nebular emission lines provide a means of estimatingphysical properties of galaxies such as dust extinction,metallicity, electron density, and ionization parameter.However, most of the calculations assume the emission iscoming from photoionized HII regions within the galaxy.If the broad components we have measured here are aresult of shocks, then the inclusion of this flux will affectline ratios and measurements. In this section, we aim toanswer the question: Is it possible that the changes inline ratios between z ∼ z ∼ The O32, R23, O3N2, and N2 line ratios
The abbreviations introduced in this section are:N2 = [NII]/H α O32 = [OIII] λλ λλ λλ α )R23 = ([OIII] λλ λλ β )5 M star ( M fl ) f r a c t i o n o f m a ss f l u x a b o v e k m / s z=1.0z=1.5z=2.0z=2.5 Figure 10.
The fraction of mass flux above 275 km/s as a functionof stellar mass calculated based on the outflow velocity scalingrelations of FIRE simulations from Muratov et al. 2015
For the MOSDEF sample, Shapley et al. (2015) andSanders et al. (2016) showed that the z ∼ z ∼ II ]. Specifically, the galaxies were offset inthe N2-BPT, O32 vs. O3N2, and O32 vs. N2 diagramsand did not show any significant offset in the O32 vs.R23 and S2-BPT diagrams. While there is much spec-ulation, there is no definitive explanation for the offsetin diagrams that include [N II ] (eg. Steidel et al. 2014,Steidel et al. 2016, Masters et al. 2014, Masters et al.2016, Shapley et al. 2015, Sanders et al. 2016). In thissection, we test if the offsets in the diagnostic diagramscould be caused only by adding the emission from shocksto the z ∼ z ∼ z ∼ z ∼ +88 − cm − Sanders et al. 2016, log(O/H) +12 = 8 .
43, Sanders et al. 2015, and shock velocity of 300km s − ). These models are shown as green points. Wethen add the SDSS distribution and shocked data pointtogether assuming a BFR of 0.4 (which is the averageBFR for the stacks by H α and corresponds to 29% of thetotal flux being shocked) and plot the result as a greenline. These SDSS+shocks models represent what localgalaxies would look like with the addition of the bestfitting shock model from Figure 7.The SDSS+shock data in Figure 11 generally showgood agreement with the z ∼ e = 100 cm − SDSS+shock models (bottom row) are higher than expected but couldbe explained because these models are at a metallicityof log(O/H)+12 = 8.93 which is higher than the aver-age z ∼ z ∼ z ∼ z ∼ z ∼ z ∼ II ]/H α , and the 3 σ lim-its are consistent with the lowest electron density models.Therefore, we conclude that all of the electron densitiesin the SDSS+shock models would match the z ∼ z ∼ z ∼ z ∼ e = 1 − − are plausible. This does notprove that the offset is caused by shocks, only that theyare a possibility. Our results do not rule out contribu-tion from AGN as a driver of these offsets since AGNoccupy similar regions of diagnostic line-ratio diagramsas shocks. The N2-BPT diagram
A great deal of study has been done on the cause of theoffset of z ∼ z ∼ z ∼ z ∼ l og ( O ) log(O/H)+12=8.03, n e =1log(O/H)+12=8.35, n e =1log(O/H)+12=8.44, n e =1log(O/H)+12=8.93, n e =1
500 km/s400 km/s300 km/s200 km/s
SDSS galaxiesz~2 high N2−BPT offsetz~2 low N2−BPT offsetSDSS+shock model l og ( O ) log(O/H)+12=8.93, n e =1000log(O/H)+12=8.93, n e =100log(O/H)+12=8.93, n e =10log(O/H)+12=8.93, n e =1 500 km/s400 km/s300 km/s200 km/s Figure 11.
Same as Figure 7 but for the O32 vs. R23, O32 vs. O3N2, and O32 vs. N2 diagrams. The tan points and solid black linehave the same meaning as in Figure 7. The blue lines show a running median or linear fit to galaxies in the MOSDEF sample that aremore offset than average (compared to SDSS galaxies) in the N2-BPT diagram, and the red lines show a running median or linear fit togalaxies that are below the average MOSDEF galaxy offset from Sanders et al. (2016). The green point corresponds to the shock modelwhich has (log(O/H)+12 = 8.44, n e = 1 cm − ) in the top row and (log(O/H)+12 = 8.93, n e = 100 cm − ) in the bottom row. The greenline combines the local SDSS data with the green point with 29% of the flux coming from shocks. our stacks. Since AGN have similar line ratios to shocksthe same argument holds that low-luminosity AGN (in-stead of shocks) could explain the offset of the z ∼ The S2-BPT diagram
As shown in Figure 7, the shocked [S II ]/H α ratios thatbest match the properties of the MOSDEF galaxies areoffset to higher values than the SDSS data. If shocksare the cause of the offset in the N2-BPT diagram, thenone might also expect an offset in the S2-BPT diagramas well. However, there is no measured offset betweenthe SDSS and the z ∼ II ]/H α ratio and contribution from diffuse ionized gas.The electron density of the MOSDEF sample is 290 +88 − cm − , and the shocked line ratios for that particulardensity would lie between the 100 and 1000 cm − shockmodels. The SDSS galaxies also lie between the 100 and 1000 cm − shock models (see Figure 7). It is possiblethat the shocked [S II ]/H α ratio for and electron densityof 290 cm − lies close to the SDSS distribution. If thisis the case, including the shocked emission would notchange the [S II ]/H α ratios of the z ∼ II ]/H α ratioand the photoionized [S II ]/H α ratio could explain thelack of an offset in the S2-BPT diagram.Another explaination for no offset in the S2-BPT dia-gram could be because of less contribution from diffuseionized gas at z ∼ z ∼
0. The fraction ofH α emission from diffuse ionized gas decreases as Σ Hα increases (Oey et al. 2007). As emission from diffuse ion-ized gas decreases, the [S II ]/H α ratio decreases while[O III ]/H β stays the same (Zhang et al. 2017, Sanderset al. 2017). Since galaxies at z ∼ Hα which implies they will have less con-7 −1.5 −1.0 −0.5 0.0 0.00.20.40.60.81.01.2 −1.5 −1.0 −0.5 0.0 log([NII]/H α )0.00.20.40.60.81.01.2 l og ( [ O III]/ H β ) SDSS+shock model (log(O/H)=8.93, n e =100 )SDSS+shock model (log(O/H)=8.44, n e =1 )z~2 galaxies from Shapley et al. (2015) Figure 12.
The N2-BPT diagram with the SDSS+shock andbroad component predictions. The blue, orange, and black lineshave the same meaning as in Figure 6. The diamonds are the broadcomponents from Figure 6. The green lines are the SDSS+shockmodels. The dashed green line is the high electron density modeland the solid green line is the low metallicity model. The greencircle corresponds to the shock model which has ( v = 300 kms − , log(O/H)+12 = 8.44, n e = 1 cm − ) and the green squarecorresponds to the shock model which has ( v = 300 km s − ,log(O/H)+12 = 8.93, n e = 100 cm − ). These points are the bestmatch to the metallicity, electron density, and shock velocity forthe entire sample. tribution from diffuse ionized gas if these local relationshold at z ∼
2. Local galaxies with high Σ
SF R do lie atlower [S II ]/H α at a given [O III ]/H β than those with lowΣ SF R on the S2-BPT diagram (Masters et al. 2016). If z ∼ II ]/H α ratio at a given [O III ]/H β ratio.Therefore, the lack of an offset in the S2-BPT diagramcould be because less contribution from diffuse ionizedgas causes the narrow emission to lie at lower [S II ]/H α than average z ∼ II ]/H α due to shocks. In this scenario, theseeffects compete with each other and ultimately canceleach other out, resulting in no net offset in the S2-BPTdiagram. Estimating SFR from H α The presence of shocks can also affect measurementsmade from single emission lines such as SFR from H α . Ifthe broad emission should be removed when calculatingSFR then not doing so would overpredict the SFR. Giventhe measured BFRs, SFRs would be overpredicted by 15,40, and 68%, respectively in our low, medium, and highstellar mass stacks. However, given the large number ofsystematic uncertainties in these calculations (extinctioncurves, nebular vs. continuum extinction, initial massfunctions, and star formation histories), a 15-70% offsetcould go undetected. Furthermore, despite only detect-ing a BFR of 0.15 in the lowest mass bin, the contribu-tion from broad emission may be higher because of ourinablility to detect broad emission with FWHM <
300 km/s. CONCLUSION
We present results from the MOSFIRE Deep Evolu-tion Field (MOSDEF) survey on broad emission fromthe nebular emission lines H α , [N II ], [O III ], H β , and[S II ]. After removing known AGN, merging galaxies, andlines affected by skylines, we study broad flux by fittingthe emission lines of individual galaxies and stacks usingnarrow and broad Gaussian components. The broad fluxaccounts for 10-70% of the flux in nebular emission lineswhen detected. For individual galaxies, we find no cor-relations between the BFR as a function of mass, SFR,sSFR, or Σ SFR , but there is a strong correlation withhigher S/N galaxies and a broad component detection.We calculate [S II ]/H α , [N II ]/H α , and [O III ]/H β lineratios for the narrow components, broad components,and the single Gaussian fits. Compared to what onewould obtain using a single Gaussian, the broad com-ponents have higher [N II ]/H α and [O III ]/H β line ra-tios. When placed on the BPT diagram (Figure 6) thebroad components for stacks lie within the compositestar-forming/AGN region. We compare the locations ofthe broad component line ratios to shock models fromAllen et al. (2008) and conclude that the broad emissioncould be explained by shocks. The locations of the broadcomponents could also be explained by contribution fromlow-luminosity AGN that may have been included in thestack.We estimate the mass loading factor and we find gen-erally good agreement with other measurements of η atthis redshift (Newman et al. 2012). At low masses, η increases as a function of mass. This result is contra-dictory to the results of the FIRE simulations (Mura-tov et al. 2015, Hopkins et al. 2014) where a decreasewith mass is predicted. This difference is best explainedby our inability to detect contributions from low veloc-ity (FWHM <
275 km s − ) broad components as shownin Figure 9. Another feasible explanation is that theoutflows have a large neutral component which is notdetected because broad H α emission is sensitive to theionized component of the outflow.We combine the shock models from Allen et al. (2008)with local line ratios from SDSS and calculate wherethese galaxies would lie on several emission line diagnos-tic diagrams. We compare these SDSS+shock models tothe emission line properties of z ∼ z ∼ z ∼
2. If weadd a 29% shocked component to SDSS data at z ∼
0, theN2-BPT, O32 vs. O3N2, and O32 vs. N2 diagrams havesimilar offset line ratios to the observed z ∼ z ∼
2. The lack of an offsetin the S2-BPT diagram seen at z ∼ z ∼ z ∼ z ∼ z ∼ − z ∼ ACKNOWLEDGMENTS
We thank Bili Dong for commenting on an early versionof this work and discussing details about current andfuture work with the FIRE simulation. We thank theMOSFIRE instrument team for building this powerfulinstrument.This work would not have been possible without the3D-HST collaboration, who provided us the spectro-scopic and photometric catalogs used to select our tar-gets and to derive stellar population parameters. Weare grateful to I. McLean, K. Kulas, and G. Mace fortaking observations for us in May and June 2013. Weacknowledge support from an NSF AAG collaborativegrant AST-1312780, 1312547, 1312764, and 1313171, andarchival grant AR-13907, provided by NASA through agrant from the Space Telescope Science Institute.This work is also based on observations made with theNASA/ESA Hubble Space Telescope (programs 12177,12328, 12060-12064, 12440-12445, 13056), which is oper-ated by the Association of Universities for Research inAstronomy, Inc., under NASA contract NAS 5-26555.NAR is supported by an Alfred P. Sloan Research Fel-lowship.The data presented in this paper were obtained at theW.M. Keck Observatory, which is operated as a scientificpartnership among the California Institute of Technol-ogy, the University of California and the National Aero-nautics and Space Administration. The Observatory wasmade possible by the generous financial support of theW.M. Keck Foundation. The authors wish to recognizeand acknowledge the very significant cultural role andreverence that the summit of Mauna Kea has alwayshad within the indigenous Hawaiian community. We aremost fortunate to have the opportunity to conduct ob-servations from this mountain.
APPENDIX
ONE-DIMENSIONAL EXTRACTION SOFTWARE:
BMEP
The MOSDEF team has written software to handle the2D and 1D reduction process. The 2D code is described in Kriek et al. (2015) and the 1D extraction code,
BMEP ,is described here. In general, BMEP can be used to ex-tract spectra from any rectified 2D spectroscopic dataincluding the MOSFIRE Data Reduction Pipeline.The 2D reduction code outputs 2D spectra that arecombined, flat-corrected, cleaned of cosmic rays, and rec-tified. We have designed a 1D extraction program thatcan optimally extract spectra, help the user find the pri-mary object, create a redshift catalog, and “blindly” ex-tract spectra where there were no obvious emission linesor continuum.
Using
BMEP
Reduced 2D spectra have two dimensions: spatial andwavelength. The overall goal of
BMEP is to optimally ex-tract 1D integrated spectra which sums the flux over thespatial dimension and leaves the user with flux vs. wave-length. The first step in extracting spectra is to findthe primary object.
BMEP draws a line over the object’sexpected position which helps distinguish the primaryobject from serendipetous objects. Next, the user inter-actively bins the data in the wavelength direction to cre-ate a flux profile. Finally, the user fits a Gaussian to theprofile. The center and width of this Gaussian determinethe spatial region which to sum as well as the weightingprofile for an optimal extraction (Horne 1986). In someobjects with high S/N the profile is non-Gaussian, andthe user can choose to weight by the actual profile insteadof the Gaussian fit.Although the above process sounds simple, it is difficultto determine which wavelengths to bin to create the high-est S/N spatial profile. Many galaxies at high redshifthave bright emission lines compared to their continuum(Stark et al. 2013). Summing all wavelengths results ina spatial profile dominated by noise which makes find-ing the center and width of the object impossible. Thebenefit of using
BMEP is the ability to interactively createthe best profile by adding or removing wavelengths whencreating the spatial profile. Additionally, some galaxiesdo have continuum but summing all wavelengths resultsin a noisy spatial profile because skylines would be in-cluded. The user can enable a “continuum mode” whichdoes not include skylines in the spatial profile by remov-ing wavelengths where the variance is higher than themedian variance.After extraction, the spectrum is plotted and can beinspected. The locations where the user clicked are alsodrawn in red on the plot. In noisy spectra, this allows theuser to easily find emission lines in the 1D spectra. Oncean emission line is found, the user can fit the line to aGaussian, input which line it is, and calculate a redshift.All emission lines and calculated redshifts are saved in acatalog. A separate program consolidates all the lines fitfor each object and calculates the most likely redshift.In cases where an object had no obvious emission linesor continuum in the 2D spectrum, a “blind” extractionwas performed. For objects with no signal in any bands,the blind extraction uses the expected position of theobject calculated from the mask file and uses the sameextraction width as the star’s width in each filter. Forobjects with signal in one or more bands, the blind ex- Source code and installation instructions available at: https://github.com/billfreeman44/bmep Table 2
Measurements from the Stack FitsParameter a Avg. b Range c BFR d BFR mine
BFR maxf
FWHM Bg ∆V Bh η i (H α )- z ∼ +0 . − . ±
200 -9.0 ±
20 0.26 +0 . − . (H α )- z ∼ +0 . − . ±
60 16. ±
10 0.64 +0 . − . (H α )- z ∼ +0 . − . ±
30 -10. ± +0 . − . ([O III ])- z ∼ ± .
028 0.029 0.20 390 ±
100 -63 ±
40 -([O
III ])- z ∼ ± .
12 0.21 0.92 300 ±
20 -18 ± III ])- z ∼ ± .
17 0.060 1.1 300 ±
30 8.5 ± III ])- z ∼ ± .
14 0.32 1.2 340 ±
30 -13 ± III ])- z ∼ ± .
16 0.08 0.85 300 ±
100 27 ±
20 -([O
III ])- z ∼ ± .
37 0.1 1.8 320 ±
80 25 ±
20 - a Parameter by which the stack was created. b The geometric mean of the galaxies included. c Mass range of galaxies included. d Broad Flux Fraction of the stack and 1 σ errors. e The 3 σ lower limit on the BFR. f The 3 σ upper limit on the BFR. g The FWHM of the broad component (km s − ) and the 1 σ error. h The velocity offset between the peaks of the broad and narrow components (km s − ). A negative value indicates a blueshift.The 1 σ error in the velocity offset is included. g The mass loading factor (see Section 5.1). traction used the average extraction widths and centersfrom filters where there was signal from the object. Thewidths from each filter are corrected for seeing differ-ences. These blind spectra allow us to put upper limitson emission lines for spectra if we know the redshift. Cur-rently,
BMEP is only able to read in MOSFIRE mask filesfor the blind extraction and would need to be modifiedto be able to read in mask files of a different format.
Sub-pixel Extraction Equations
The optimal extraction used in
BMEP is based on Horne(1986). While testing the software, we compared extrac-tions of a bright object done by several different users.Some extractions differed in extracted flux by 2-4% at allwavelengths for some users. We traced the cause of thisto rounding differences between two extraction profiles.The extraction width is determined by a Gaussian fit tothe profile and in some cases, the extracted widths wouldbe 1 pixel different because we rounded the extraction tothe nearest whole pixel. The optimal extraction of Horne(1986) sums over an integer number of pixels in the spa-tial direction to calculate the flux at each wavelength. Tofix this, we created a sub-pixel optimal extraction algo-rithm. This algorithm extracts the central pixels exactlythe same as in Horne (1986) but adds a fraction of a pixelat each end.We base the sub-pixel algorithm on Equation 8 fromHorne (1986). However, it is simplified for MOSFIREreduction because there is no sky subtraction or cosmicray rejection needed as these are done during the 2Dreduction. The equations from Horne (1986) with thesesimplifications are: x (cid:48) b = R ( c − w ) x (cid:48) t = R ( c + w ) x (cid:48) t (cid:88) x (cid:48) b D = F (cid:48) box (A1) x (cid:48) t (cid:88) x (cid:48) b V = V (cid:48) box (A2) (cid:80) x (cid:48) t x (cid:48) b P D/V (cid:80) x (cid:48) t x (cid:48) b P /V = F (cid:48) opt (A3) (cid:80) x (cid:48) t x (cid:48) b P (cid:80) x (cid:48) t x (cid:48) b P /V = V (cid:48) opt (A4)Unnumbered equations are defining relationships orvariables. Bold letters indicate functions. R is the roundfunction, c is the center of the object from the Gaussianfit to the profile, w is half the width to extract, D is thethe flux in one pixel, V is the variance in one pixel, P is the weighting profile, x b is the pixel at the bottom ofthe profile, and x t is the pixel at the top of the profile.The weighting profile comes from the Gaussian fit to thespatial profile. F (cid:48) box is the boxcar flux, V (cid:48) box is the boxcarvariance, F (cid:48) opt is the optimal flux, and V (cid:48) opt is the optimalvariance for the Horne (1986) algorithm that does not in-clude sub-pixel corrections. We remove the wavelengthsubscript for simplification.We extend this equation to extract a fraction of eachpixel. The central region of extraction is extracted thesame as in Horne (1986), then a fraction of the outer pix-els are added to this flux. First we calculate the rangewhich the flux is extracted in the same manner as equa-tions A1-A4. This is between xb (cid:48) and xt (cid:48) which are cal-culated as follows:0 x b = L ( c − w + 1) x t = L ( c + w ) L is the “Floor” function. Next, calculate the “remain-der” from the bottom ( R b ) and the top ( R t ): R b = 1 − ( x b − x (cid:48) b ) R t = x t − x (cid:48) t Now we calculate the weighting for sub-pixel extractionon the edges: P t = P ( xt + 1) R t P b = P ( xb − R b The boxcar extraction for the sub-pixel algorithm is: B = D ( xb − R b T = D ( xt + 1) R bx t (cid:88) x b D + B + T = F box (A5) x t (cid:88) x b V + V b R b + V t R t = V box (A6)Where B and T are the flux from the bottom and toppixels to be added to the central region. For the optimalextraction this extra flux is: B = P b D ( x b − R b V ( x b − T = P t D ( x t + 1) R t V ( x t + 1))Calculate sub-pixel flux and variance:( (cid:80) x t x b P D/V ) + B + T (cid:80) x t x b P /V + P b /V b + P t /V t = F opt (A7)( (cid:80) x t x b P ) + P b + P t (cid:80) x t x b P /V + P b /V b + P t /V t = V opt (A8)If the spatial range to extract are integers, then R b = 1, R t = 0, x (cid:48) b = x b + 1, and x (cid:48) t = x t , and one can recoverthe original equations from Horne (1986).Figure 13 shows a comparison of the sub-pixel opti-mal vs. normal optimal. This figure was produced byselecting a flat, featureless section of a star that has nosky lines. Within this region, we calculated the averageflux and S/N, then we varied the extracted width. Asone might expect, the Horne extraction has steps wherethe width rounds to the next pixel and the sub-pixel ex-traction is smooth. Though the jumps in flux are severewhen the width is small, the steps flatten out as widthincreases. At the width where we extract (2x FWHM),the jumps between steps is quite small, at worst around A v e r age S i gna l t o N o i s e A v e r age S i gna l t o N o i s e N o r m a li z ed F l u x Figure 13.
A comparison between sub-pixel and the Horne (1986)extractions. The solid line is the sub-pixel extraction and thedashed is the Horne (1986) extraction. This plot is made by firstextracting a star normally, then looking for a section of the spec-trum that is featureless showing no absorption features, emissionlines, or sky lines. Next, the spectrum is extracted using widthsbetween 1.5 and 5.0 pixels in increments of 0.1 pixels. The pointsare plotted as the lines in the figure above. Because each star hasa slightly different width, we convert the width in pixels to a widthin “sigma”.
RESULTS OF NON-PARAMETRIC RATIO ESTIMATION
When fitting the broad flux, we assumed the broadflux is Gaussian in shape. To measure broad emissionregardless of shape, we calculate line ratios using the fluxat different velocities from the centroid of each line forstacks from Figure 6. The results are shown in Figure 14.This provides a non-parametric measurement of the lineratios as a function of velocity. The high velocity pointsare typically higher than the Kauffmann et al. (2003) linewhich is a similar trend to the broad component ratios inFigure 6. Since the non-parametric measurements havesimilar results to the fits, it is reasonable to use the fitsin our analysis. REFERENCES
Allen, M. G., Groves, B. A., Dopita, M. A., Sutherland, R. S., &Kewley, L. J. 2008, ApJS, 178, 20Azadi, M., Coil, A. L., Aird, J., et al. 2017, ApJ, 835, 27Baldwin, J. A., Phillips, M. M., & Terlevich, R. 1981, PASP, 93, 5Barden, M., Rix, H.-W., Somerville, R. S., et al. 2005, ApJ, 635,959Bian, F., Fan, X., Bechtold, J., et al. 2010, ApJ, 725, 1877Bland, J., & Tully, B. 1988, Nature, 334, 43Brinchmann, J., Pettini, M., & Charlot, S. 2008, MNRAS, 385,769Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682Chabrier, G. 2003, PASP, 115, 763 −1.5 −1.0 −0.5 0.00.20.40.60.81.0−1.5 −1.0 −0.5 0.0log ([NII]/H α )0.20.40.60.81.0 l og ( [ O III]/ H β ) Low Mass stackMed Mass stackHigh Mass stack0−50 km/s50−100 km/s100−150 km/s150−350 km/s
Figure 14.
BPT diagram created by integrating the flux overdifferent velocities.Christensen, C. R., Dav´e, R., Governato, F., et al. 2015, ArXive-prints, arXiv:1508.00007Coil, A. L., Aird, J., Reddy, N., et al. 2014, ArXiv e-prints,arXiv:1409.6522Colina, L., Arribas, S., & Clements, D. 2004, ApJ, 602, 181Conroy, C., Gunn, J. E., & White, M. 2009, ApJ, 699, 486Cowie, L. L., Barger, A. J., & Songaila, A. 2016, ApJ, 817, 57Daddi, E., Bournaud, F., Walter, F., et al. 2010, ApJ, 713, 686Erb, D. K., Shapley, A. E., Pettini, M., et al. 2006, ApJ, 644, 813Feruglio, C., Fiore, F., Carniani, S., et al. 2015, A&A, 583, A99F¨orster Schreiber, N. M., Genzel, R., Newman, S. F., et al. 2014,ApJ, 787, 38Genzel, R., Newman, S., Jones, T., et al. 2011, ApJ, 733, 101Genzel, R., F¨orster Schreiber, N. M., Rosario, D., et al. 2014a,ApJ, 796, 7Genzel, R., F¨orster Schreiber, N. M., Lang, P., et al. 2014b, ApJ,785, 75Gergeev, S. G. 1992, Ap&SS, 197, 77Hainline, K. N., Shapley, A. E., Kornei, K. A., et al. 2009, ApJ,701, 52Hastings, W. K. 1970, Biometrika, 57, 97Heckman, T. M., Armus, L., & Miley, G. K. 1990, ApJS, 74, 833Hirschmann, M., Charlot, S., Feltre, A., et al. 2017, ArXive-prints, arXiv:1706.00010Hopkins, P. F., Kereˇs, D., O˜norbe, J., et al. 2014, MNRAS, 445,581Horne, K. 1986, PASP, 98, 609Jones, T., Martin, C., & Cooper, M. C. 2015, ApJ, 813, 126Kashino, D., Silverman, J. D., Rodighiero, G., et al. 2013, ApJL,777, L8Kauffmann, G., Heckman, T. M., Tremonti, C., et al. 2003,MNRAS, 346, 1055Kennicutt, Jr., R. C. 1998, ARA&A, 36, 189Kewley, L. J., Dopita, M. A., Leitherer, C., et al. 2013, ApJ, 774,100Kewley, L. J., Dopita, M. A., Sutherland, R. S., Heisler, C. A., &Trevena, J. 2001, ApJ, 556, 121Kewley, L. J., Zahid, H. J., Geller, M. J., et al. 2015, ApJL, 812,L20Kriek, M., van Dokkum, P. G., Labb´e, I., et al. 2009, ApJ, 700,221Kriek, M., Shapley, A. E., Reddy, N. A., et al. 2015, ApJS, 218,15Lehnert, M. D., & Heckman, T. M. 1996, ApJ, 462, 651Leroy, A. K., Walter, F., Martini, P., et al. 2015, ApJ, 814, 83Leung, G. C. K., Coil, A., Azadi, M., et al. 2017, ArXiv e-prints,arXiv:1703.10255Liu, X., Shapley, A. E., Coil, A. L., Brinchmann, J., & Ma, C.-P.2008, ApJ, 678, 758Madau, P., & Dickinson, M. 2014, ARA&A, 52, 415Madau, P., Pozzetti, L., & Dickinson, M. 1998, ApJ, 498, 106 Markwardt, C. B. 2009, in Astronomical Society of the PacificConference Series, Vol. 411, Astronomical Data AnalysisSoftware and Systems XVIII, ed. D. A. Bohlender, D. Durand,& P. Dowler, 251Martin, C. L. 2005, ApJ, 621, 227—. 2006, ApJ, 647, 222Mart´ın-Fern´andez, P., Jim´enez-Vicente, J., Zurita, A., Mediavilla,E., & Castillo-Morales, ´A. 2016, MNRAS, arXiv:1605.00407Masters, D., Faisst, A., & Capak, P. 2016, ApJ, 828, 18Masters, D., McCarthy, P., Siana, B., et al. 2014, ApJ, 785, 153McLean, I. S., Steidel, C. C., Epps, H. W., et al. 2012, in Societyof Photo-Optical Instrumentation Engineers (SPIE) ConferenceSeries, Vol. 8446, Society of Photo-Optical InstrumentationEngineers (SPIE) Conference Series, 0Metropolis, N., Rosenbluth, A. W., Rosenbluth, M. N., Teller,A. H., & Teller, E. 1953, The Journal of Chemical Physics, 21,1087Muratov, A. L., Keres, D., Faucher-Giguere, C.-A., et al. 2015,ArXiv e-prints, arXiv:1501.03155Newman, S. F., Genzel, R., F¨orster-Schreiber, N. M., et al. 2012,ApJ, 761, 43Newman, S. F., Buschkamp, P., Genzel, R., et al. 2014, ApJ, 781,21Oey, M. S., Meurer, G. R., Yelda, S., et al. 2007, ApJ, 661, 801Oke, J. B., & Gunn, J. E. 1983, ApJ, 266, 713Oke, J. B., Cohen, J. G., Carr, M., et al. 1995, PASP, 107, 375Osterbrock, D. E. 1989, Astrophysics of gaseous nebulae andactive galactic nucleiOstriker, E. C., & Shetty, R. 2011, ApJ, 731, 41Pettini, M., & Pagel, B. E. J. 2004, MNRAS, 348, L59Phillips, A. C. 1993, AJ, 105, 486Reddy, N. A., & Steidel, C. C. 2009, ApJ, 692, 778Reddy, N. A., Kriek, M., Shapley, A. E., et al. 2015, ArXive-prints, arXiv:1504.02782Rockosi, C., Stover, R., Kibrick, R., et al. 2010, in Proc. SPIE,Vol. 7735, Ground-based and Airborne Instrumentation forAstronomy III, 77350RRupke, D. S., Veilleux, S., & Sanders, D. B. 2005, ApJS, 160, 115Rupke, D. S. N., & Veilleux, S. 2013, ApJ, 768, 75Sanders, R. L., Shapley, A. E., Zhang, K., & Yan, R. 2017, ArXive-prints, arXiv:1708.04625Sanders, R. L., Shapley, A. E., Kriek, M., et al. 2015, ApJ, 799,138—. 2016, ApJ, 816, 23Shapiro, K. L., Genzel, R., Quataert, E., et al. 2009, ApJ, 701,955Shapley, A. E., Coil, A. L., Ma, C.-P., & Bundy, K. 2005, ApJ,635, 1006Shapley, A. E., Reddy, N. A., Kriek, M., et al. 2015, ApJ, 801, 88Shen, S., Mo, H. J., White, S. D. M., et al. 2003, MNRAS, 343,978Shivaei, I., Reddy, N. A., Shapley, A. E., et al. 2015, ArXive-prints, arXiv:1507.03017Skelton, R. E., Whitaker, K. E., Momcheva, I. G., et al. 2014,ApJS, 214, 24Stark, D. P., Auger, M., Belokurov, V., et al. 2013, MNRAS, 436,1040Steidel, C. C., Erb, D. K., Shapley, A. E., et al. 2010, ApJ, 717,289Steidel, C. C., Strom, A. L., Pettini, M., et al. 2016, ApJ, 826, 159Steidel, C. C., Rudie, G. C., Strom, A. L., et al. 2014, ApJ, 795,165Strom, A. L., Steidel, C. C., Rudie, G. C., et al. 2017, ApJ, 836,164Swinbank, A. M., Papadopoulos, P. P., Cox, P., et al. 2011, ApJ,742, 11Tacconi, L. J., Neri, R., Genzel, R., et al. 2013, ApJ, 768, 74Trujillo, I., & Pohlen, M. 2005, ApJL, 630, L17Veilleux, S., Cecil, G., & Bland-Hawthorn, J. 2005, ARA&A, 43,769Veilleux, S., Shopbell, P. L., & Miller, S. T. 2001, AJ, 121, 198Weiner, B. J., Coil, A. L., Prochaska, J. X., et al. 2009, ApJ, 692,187Westmoquette, M. S., Clements, D. L., Bendo, G. J., & Khan,S. A. 2012, MNRAS, 424, 4162