Halo Mass Dependence of HI and OVI Absorption: Evidence for Differential Kinematics
Nigel L. Mathes, Christopher W. Churchill, Glenn G. Kacprzak, Nikole M. Nielsen, Sebastian Trujillo-Gomez, Jane Charlton, Sowgat Muzahid
aa r X i v : . [ a s t r o - ph . GA ] J u l D RAFT VERSION A PRIL
3, 2018
Preprint typeset using L A TEX style emulateapj v. 04/17/13
HALO MASS DEPENDENCE OF H I AND O VI ABSORPTION: EVIDENCE FOR DIFFERENTIAL KINEMATICS N IGEL
L. M
ATHES , C HRISTOPHER
W. C
HURCHILL , G LENN
G. K
ACPRZAK , N
IKOLE
M. N
IELSEN ,S EBASTIAN T RUJILLO -G OMEZ , J ANE C HARLTON , AND S OWGAT M UZAHID Draft version April 3, 2018
ABSTRACTWe studied a sample of 14 galaxies (0 . < z < .
7) using
HST /WFPC2 imaging and high-resolution
HST /COSor
HST /STIS quasar spectroscopy of Ly α , Ly β , and O VI λλ , . ≤ log( M h / M ⊙ ) ≤ .
2, lie within D = 300 kpc of quasar sightlines, probing out to D / R vir = 3. When thefull range of M h and D / R vir of the sample are examined, ∼
40% of the H I absorbing clouds can be inferred tobe escaping their host halo. The fraction of bound clouds decreases as D / R vir increases such that the escapingfraction is ∼
15% for D / R vir < ∼
45% for 1 ≤ D / R vir <
2, and ∼
90% for 2 ≤ D / R vir <
3. Adopting themedian mass log M h / M ⊙ = 11 . VI absorption is found in only ∼
40% of theH I clouds in and around lower mass halos as compared to ∼
85% around higher mass halos. For D / R vir < ∼ ∼ ≤ D / R vir <
2, the escape fractions are ∼
55% and ∼
35% for lower mass and higher mass halos,respectively. For 2 ≤ D / R vir <
3, the escape fraction for lower mass halos is ∼ R vir of their host galaxies and that the kinematics aredominated by outflows. Our finding of “differential kinematics” is consistent with the scenario of “differentialwind recycling” proposed by Oppenheimer et al. We discuss the implications for galaxy evolution, the stellarto halo mass function, and the mass metallicity relationship of galaxies. Keywords: galaxies: halos — quasars: absorption lines INTRODUCTIONCharacterizing the baryonic gas processes within and sur-rounding galaxies is central to understanding their formationand evolution. Quantifying the spatial extent, kinematics, and,in particular, the recycling and/or escape fraction of circum-galactic gas are of primary importance in that they place directobservational constraints on simulations of galaxies and pro-vide insights into the workings of galaxy evolution.High-resolution spectroscopy of quasars, which provide abackground luminous source, and high-resolution imaging ofthe foreground galaxies provides the data necessary for exam-ining the kinematics of galactic gas and its geometric distri-bution with respect to the galaxy projected orientation.In general, the gas structures in and around galaxies can bedivided into three broad categories: the interstellar medium(ISM), the circumgalactic medium (CGM), and the inter-galactic medium (IGM). The CGM, being the gas reservoirthat interfaces with the star-forming ISM, outflowing stellar-driven winds, and the accreting IGM, may contain up to50% of the baryonic mass bound to galaxies (Tumlinson et al.2011) and account for up to 50% of the baryons unac-counted for in galaxy dark matter halos (Werk et al. 2014).As such, the CGM may play the most critical role in govern-ing the properties of galaxies (e.g., Oppenheimer et al. 2010;Churchill et al. 2013b), including regulatory physics leadingto the observed stellar mass function (e.g., Behroozi et al.2013) and the stellar mass-ISM metallicity relationship (e.g.,Tremonti et al. 2004).The physical extent of the CGM and the transition zone New Mexico State University, Las Cruces, NM 88003 Swinburne University of Technology, Victoria 3122, Australia Australian Research Council Super Science Fellow The Pennsylvania State University, University Park, PA 16802 between the CGM and IGM are currently open questions.Studies by Steidel et al. (2010), Prochaska et al. (2011), andRudie et al. (2012) indicate a transition from the CGM to theIGM at log N (H I ) ≃
14 and a projected distance of ∼
300 kpcfrom galaxies at z ∼ .
5. At this redshift, log N (H I ) ≃ ρ H / ¯ ρ H ≃ .
5, whereas at z ≃
0, this overdensity would suggest a CGM/IGM transitionat log N (H I ) ≃
13 (see Davé et al. 1999). Indeed, Ford et al.(2013) show that, in the over-dense regions hosting galaxies,the extent of the H I at fixed column density increases withvirial mass, suggesting that a single fiducial physical size forthe CGM may not apply across the entire mass range of galax-ies; it may be more appropriate to scale CGM properties rela-tive to the virial radius (e.g., Churchill et al. 2013a,b)Cosmological hydrodynamic simulations indicate that thevirial mass may dictate the temperature history, density,and mode of IGM accretion (Birnboim & Dekel 2003;Kereš et al. 2005; Dekel & Birnboim 2006; Kereš et al.2009; van de Voort et al. 2011; Fumagalli et al. 2011;van de Voort & Schaye 2012). The recycling timescale ofwind material back into the ISM may depend upon galaxyvirial mass according to what Oppenheimer et al. (2010)call “differential wind recycling”. Their simulations suggestthat the recycling time of wind material through the CGMcould be shorter for higher mass halos, and possibly longerthan the Hubble time for the lowest mass halos, and that thisbehavior may be key for understanding the galaxy stellarmass function in the low mass range. As deduced from thesimulations, differential wind recycling is primarily due togreater hydrodynamic (not gravitational) deceleration of windmaterial in higher mass halos due to their being embeddedin denser gas environments, leading to diminished recyclingtimes (also see Oppenheimer & Davé 2008).As probed by Mg II absorption (see Nielsen et al. 2013a,b, M ATHES ET AL .and references therein), the observed projected absorptionprofile, covering fraction, and physical extent of the coolgas component of the CGM combine to suggest that thecool/warm CGM exhibits a self-similar radial behavior withvirial mass (Churchill et al. 2013a,b). Mg II absorption prop-erties behave self-similarly with D / R vir , the projected dis-tance of the absorption from the galaxy relative to the virialradius. Stocke et al. (2013) also report behavior that can beinterpreted as self-similarity in that, once impact parameter isscaled by R vir , the cool/warm CGM gas properties show little-to-no variation as a function of projected distance, gas kine-matics, and galaxy luminosity (however, see Werk et al. 2014,who report weak anti-correlations in the cloud hydrogen num-ber densities and ionization parameters). If virial mass influ-ences the global physics of the CGM, then the CGM/IGMboundary and ISM/stellar formation physics of galaxies mayfundamentally be related to the dark matter overdensity pro-file of halos within the virial radius.Simulations of starbursts predict that outflowing gas willpreferentially escape along the galaxy minor axis (e.g.,Strickland et al. 2004). Infalling gas is predicted to prefer-entially accrete in the galactic plane and kinematically tracegalaxy rotation (e.g., Stewart et al. 2011), consistent withthe observations of Steidel et al. (2002) and Kacprzak et al.(2010). Consistent with these predictions, Mg II absorption ismost commonly found along the projected minor and majoraxes of galaxies (Bouché et al. 2012; Kacprzak et al. 2012a).Bordoloi et al. (2011) report that, on average, larger Mg II equivalent widths are found along the projected minor axisas compared to the projected major axis. The distribution ofH I , traced using Ly α absorption, appears to be more uniformwith respect to the galaxy projected axis (e.g., Stocke et al.2013).The kinematics of CGM/IGM absorption with respect tohost galaxy escape velocity places direct observational con-straints on the influence of hydrodynamic and/or gravitationaldeceleration of CGM gas and provides insights into the plau-sibility of a mass dependent wind recycling scenario. For D <
150 kpc (corresponding to D / R vir < M h / M ⊙ ) > .
3, CGM gas is predominantly bound.Stocke et al. (2013) find that the majority of the CGM withinthe projected virial radius also appears bound, but outside ofthe projected virial radius, velocities can exceed the escapevelocity.Simulations, such as those by Kereš et al. (2005) andby van de Voort & Schaye (2012), predict that more mas-sive halos have higher hot gas mass fractions. The metal-enriched hot phase of the CGM, which can be traced byO VI absorption, may be a reservoir of a significant bary-onic mass (Tumlinson et al. 2011). The hot phase mayserve as a “coronal” hydrostatic region surrounding galax-ies that strongly governs the formation and destruction ofthe cool/warm CGM “clouds” (Mo & Miralda-Escude 1996;Maller & Bullock 2004; Dekel & Birnboim 2006). On theother hand, gas traced by O VI absorption may arise in multi-phase gas structures (Prochaska et al. 2004; Cooksey et al.2008).Although O VI absorbing gas has been extensively stud-ied in the Galactic, extragalactic, and IGM environments(Savage et al. 2002, 2003; Richter et al. 2004; Sembach et al.2004; Lehner et al. 2006; Danforth & Shull 2008; Tripp et al.2008; Thom et al. 2011; Tumlinson et al. 2011; Muzahid2014), the potential important role of CGM gas beckons fur- ther exploration of O VI absorption around galaxies in or-der to address basic questions such as: what is the physi-cal extent of O VI absorbing gas around galaxies, and whereis the CGM/IGM transition region of the hot phase? Can aCGM/IGM transition region, or boundary, be observationallydiscerned? Is the transition region halo mass dependent? Arethere trends in the hot phase with respect to a galaxy’s virialradius? Does the hot phase have a preferred geometrical dis-tribution around galaxies, similar or dissimilar to what is seenfor the cold/warm phase? Is there kinematic evidence for thedifferential wind recycling scenario?In order to address these questions, we examine the kine-matics and spatial distribution of H I and O VI column den-sities in the CGM/IGM and their absorption kinematics fora small sample of galaxies. This paper is structured as fol-lows: In Section 2 we discuss the sample selection, the data,and data analysis. In Section 3, we examine the spatial ex-tent and geometry of the H I and O VI absorbing gas. In Sec-tion 4, we compare the H I and O VI kinematics and examinethe spatial and virial mass dependence of the CGM with re-spect to halo escape velocity. In Section 5, we discuss ourfindings and in Section 6 we summarize our results and dis-cussion. Throughout, we adopt a Λ CDM cosmological modelwith h = 0 .
7, where h = H /
100 km s - Mpc - , with Ω m = 0 . Ω Λ = 0 . SAMPLE SELECTION, DATA, AND ANALYSIS2.1.
Sample Selection
We have assembled a sample of 14 galaxies in the fieldsof UV bright quasars with high resolution
HST imaging andultraviolet spectra. We impose four primary criteria for thegalaxy sample: (1) each galaxy must be intervening to a back-ground quasar within a projected distance of 300 kpc from theline of sight and must have a spectroscopic redshift measure-ment, (2) each galaxy must be imaged with the
Hubble SpaceTelescope ( HST ), (3) a
HST /COS and/or STIS spectrum ofthe background quasar is available that, at a minimum, coversthe redshifted Ly α , Ly β , and O VI λλ , ± - of the associated foreground galaxy red-shift, and (4) the foreground galaxy must not reside in a groupor cluster environment to the extent that the data provides suchinformation.The 300 kpc projected distance allows us to study theCGM out to the extent probed by Steidel et al. (2010),Prochaska et al. (2011), and Rudie et al. (2012) and to ex-tend beyond the 150 kpc range probed by Tumlinson et al.(2011, 2013). The HST /WFPC2 images provide the spa-tial information required to measure galaxy morphologicalparameters and determine galaxy orientations relative to thequasar line of sight. The
HST /COS and/or STIS spectroscopyallows Voigt profile decomposition of the Ly α , Ly β , andO VI λλ , ± - (based upon redshift). If other galaxies existin the field within this velocity window, we then require thegalaxies lie farther than a projected distance of 600 kpc fromthe quasar line of sight.The quasar fields from which the galaxy sample is drawnwere surveyed by Ellingson & Yee (1994), Lanzetta et al.(1995), Le Brun et al. (1996), Chen et al. (2001), and IFFERENTIAL K INEMATICS Table 1
Journal of Observations(1) (2) (3) (4) (5)Quasar Instrument Filter/Grating Exp. Time PID[s]Q0405 - HST /WFPC2 F702W 2400 5949
HST /COS G130M+G160M 20,749 11541Q0454 - HST /WFPC2 F702W 1200 5098
HST /COS G160M 2778 12252
HST /COS G160M 1849 12466
HST /COS G185M 74,410 12536Q1001 + HST /WFPC2 F702W 2400 5949
HST /COS G130M+G160M 12,988 12,038Q1136 - HST /WFPC2 F702W 2100 6919
HST /COS G130M 7751 12275Q1216 + HST /WFPC2 F702W 2100 6619
HST /COS G130M+G160M 10,702 12025Q1259 + HST /WFPC2 F702W 2100 6919
HST /WFPC2 F702W 2100 6919
HST /COS G130M+G160M 20,383 11541Q1317 + HST /WFPC2 F702W 4700 5984
HST /COS G160M+G185M 22,971 11667Q1704 + HST /WFPC2 F702W 2400 5949
HST /STIS E140M 22,155 8015
Johnson et al. (2013). We note that the fields have been stud-ied for different science goals employing different facilitiesto varying degrees of completeness. A detailed discussion ofthe application of the galaxy selection criteria for each field ispresented in Appendix A. Here, we briefly summarize the sur-veys. The galaxies observed by Ellingson & Yee (1994) haveredshifts 0 . < z < . . < z < . HST /FOS spectraof the quasar had been obtained for the
HST
Key Project (cf.,Bahcall et al. 1993).The resulting redshift range of our galaxy sample is de-termined exclusively by the UV spectral coverage of the
HST /COS and STIS observations and not by any a priori red-shift cuts. Using the above four selection criteria, we com-piled a sample of 14 galaxies spanning the redshift range of0 . ≤ z ≤ .
67 with impact parameters from 60 ≤ D ≤ Galaxy Imaging and Photometric Properties
All
HST /WFPC2 images were obtained using the F702Wband. We adopted the reduced and calibrated images from theWFPC-2 Associations Science Products Pipeline (WASPP ).The galaxy apparent Vega magnitudes, m F702W , were de-termined using 1 . σ isophotes from Source Extractor(Bertin & Arnouts 1996). From the galaxy centroids, we http://archive.stsci.edu/hst/wfpc2/pipeline.html compute the galaxy offset from the quasar (arcsec) andthe galaxy-quasar sightline impact parameter (kpc). Wecomputed AB r -band absolute magnitudes, M r , by k -correcting the observed F702W magnitudes following themethod of Nielsen et al. (2013b). To determine galaxy virialmasses, M h , we performed halo abundance matching (e.g.,Behroozi et al. 2010; Trujillo-Gomez et al. 2011) followingthe method of Churchill et al. (2013b), in which we matchthe distribution of the maximum circular velocity of halos inthe Bolshoi N -body cosmological simulation of Klypin et al.(2011) to the COMBO-17 r -band luminosity function ofWolf et al. (2003). Galaxy virial radii are then computed from M h using the relation of Bryan & Norman (1998). Uncertain-ties in the viral masses and virial radii, which are on the or-der 10%, originate from the scatter in the virial mass circularvelocity distribution function (see Churchill et al. 2013b, fordetails).Quantified galaxy morphological parameters were mea-sured using GIM2D (Simard et al. 2002) following the meth-ods of Kacprzak et al. (2011). GIM2D models the two-dimensional brightness profiles of the galaxies and computesthe inclination, i , and the position angle, Φ , on the sky. Weadopt the formalism that i = 0 ◦ is face-on and i = 90 ◦ is edge-on. We translate the position angle to an “azimuthal angle”defined such that for Φ = 0 ◦ the quasar sightline lies along theprojected major axis, and for Φ = 90 ◦ it lies along the galaxyprojected minor axis.In Table 2, columns (1) and (2) list the quasar field and thegalaxy spectroscopic redshift. Columns (3)–(5) list the galaxyoffsets relative to the quasar and the galaxy impact parameter,respectively. Columns (6) and (7) list the galaxy HST /WFPC2F702W apparent magnitude (Vega) and the r -band absolutemagnitude (AB). Columns (8) and (9) list the virial mass andvirial radius of the galaxy. Columns (10) and (11) list thegalaxy azimuthal angle and inclination.The galaxy WFPC/F702W apparent magnitudes range from22 . ≥ m F702W ≥ .
8. Absolute r -band magnitudes range from - . ≥ M r ≥ - .
2. The range of galaxy inclinations and az-imuthal angles are 18 ◦ ≤ i ≤ ◦ and 6 ◦ ≤ Φ ≤ ◦ , respec-tively. The virial masses range from 10 . ≤ log( M h / M ⊙ ) ≤ .
2, with virial radii between 70 ≤ R vir ≤
225 kpc. The me-dian virial mass is log( M h / M ⊙ ) = 11 . Quasar Spectra and Absorption Properties
The
HST /COS spectra were reduced and flux calibrated us-ing the CalCOS pipeline (V2.11). Vacuum and heliocentriccorrections, dispersion alignment, and co-addition of individ-ual exposures were performed using software developed bythe COS team (also see Narayanan et al. 2011).Reduction and calibration of the E140M HST /STIS spec-trum (for Q1704 + . ′′ × . ′′ slit) was per-formed using the standard STIS pipeline (Brown et al. 2002).Further details are discussed in Narayanan et al. (2005). Con-tinuum fitting for both the HST /COS and
HST /STIS data setswas conducted using the interactive SFIT task in IRAF fol-lowing the methods described in Sembach & Savage (1992).We then refined higher order continuum fits using our owncode, F ITTER (Churchill et al. 2000). http://casa.colorado.edu/ ∼ danforth/science/cos/costools.html IRAF is distributed by the National Optical t Astronomy Observatory,which is operated by the Association of Universities for Research in Astron-omy (AURA) under cooperative agreement with the National Science Foun-dation. M ATHES ET AL . Table 2
Galaxy Properties(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)Quasar z gal ∆ α ∆ δ D m
F702W M r log( M h ) R vir Φ i [arcsec] [arcsec] [kpc] [Vega] [AB] [M ⊙ ] [kpc] [deg] [deg]Q0405 -
123 0.1534 - . - . - .
21 11.8 + . - . + - + . - . + . - . Q0405 -
123 0.2978 31 . - . - .
86 12.2 + . - . + - + . - . + . - . Q0405 -
123 0.4100 2 . - . - .
37 11.2 + . - . + - + . - . + . - . Q0454 - . - . - .
35 12.0 + . - . + - + . - . + . - . Q1001 + - . - . - .
49 10.8 + . - . + - + . - . + . - . Q1001 + . - . - .
59 11.2 + . - . + - + . - . + . - . Q1136 - - . . - .
43 11.0 + . - . + - + . - . + . - . Q1136 - . - . - .
42 11.7 + . - . + - + . - . + . - . Q1216 + . - . - .
11 11.7 + . - . + - + . - . + . - . Q1259 + . - . - .
46 11.2 + . - . + - + . - . + . - . Q1259 + - . . - .
96 11.9 + . - . + - + . - . + . - . Q1317 + . - . - .
15 12.1 + . - . + - + . - . + . - . Q1704 + . . - .
85 12.0 + . - . + - + . - . + . - . Q1704 + - . - . - .
16 11.6 + . - . + - + . - . + . - . Table 3
Absorption Properties(1) (2) (3) (4) (5) (6) (7) (8) (9) a (10) a Quasar z gal z abs W r (Ly α ) W r (Ly β ) W r (O VI ) log N (H I ) log N (O VI ) v ( - ) v ( + ) [Å] [Å] [Å] [cm - ] [cm - ] [km s - ] [km s - ]Q0405 -
123 0.1534 0.1530 0.547 ± ± ± ± ± -
575 212Q0405 -
123 0.2978 0.2977 0.343 ± ± ± ± ± -
194 87Q0405 -
123 0.4100 0.4059 0.966 ± ± ± ± ± - - - ± ± ± ± ± -
230 248Q1001 + ± ± ± ± ± -
768 12Q1001 + ± ± ± ± ± -
556 50Q1136 - ± < < ± < -
347 49Q1136 - ± ± ± ± ± -
456 293Q1216 + ± ± ± ± ± -
254 268Q1259 + ± < ± ± ± -
307 441Q1259 + ± < < ± < -
100 99Q1317 + ± ± ± ± ± -
514 344Q1704 + ± ± ± ± ± -
133 80Q1704 + < < < < < -
100 99 a The velocities correspond to the Ly α absorption. We searched the quasar spectra for Ly α , Ly β , andO VI λλ , ± - of theidentified galaxy redshift. We adopt the objective detectionmethods of Schneider et al. (1993) using the 5 σ uncertaintyin the equivalent width spectrum. Once an absorption featureis identified, we use the methods of Churchill et al. (1999) andChurchill & Vogt (2001) to measure the velocity extremes ofthe absorption, v ( - ) and v ( + ) , the rest-frame equivalent widths, W r , and the optical depth mean system absorption redshifts, z abs , using the Ly α absorption feature.For an absorption feature to be adopted as a detection, weapply the criterion of a 3 σ equivalent width significance level,i.e., W r ≥ σ W r , otherwise we quote 3 σ W r as the upper limiton W r . The quoted uncertainties in the measured equivalent widths account for both the pixel statistical uncertainty andthe systematic uncertainty due to the choice of continuumfit. The latter assumes a mean continuum placement uncer-tainty of 30% of the mean pixel statistical uncertainty (seeSembach & Savage 1992). Depending on the signal-to-noiseratio of the spectral region, the continuum uncertainty yieldsa 5–20 mÅ systematic uncertainty in the equivalent width.To verify the identity of Ly α absorption, we examinewhether the associated Ly β absorption is formally detected.Further verification is obtained by detection of the O VI dou-blet; however, not all Ly α absorption has detected Ly β and/ormetal-line absorption. In the case of the O VI doublet, we donot require that both members of the doublet are formally de-tected in order to identify either O VI λ VI λ IFFERENTIAL K INEMATICS Figure 1.
Galaxy and absorption data for the galaxies at z gal = 0 . z gal = 0 . z gal = 0 . - HST /WFPC2image with an arrow pointing in the direction of the quasar sightline. The galaxy impact parameter, inclination, and azimuthal angle are listed. We also showthe observed spectra and Voigt profile fits for the associated absorption. Plotted in black is the normalized, continuum-fitted UV spectrum and in green is the1 σ uncertainty spectrum. Overlaid in red is the VP model fit for each transition. Red ticks mark the centroid of each VP component. Velocities are rest-framerelative to the associated galaxy systemic velocity. absorption (either due to line blending or the λ α transition was detected for all but one of the galaxy-absorberpairs. For all but four galaxy-absorber pairs, Ly β was de-tected, but O VI was detected for three of these four pairs. Foronly one pair, no Ly α , Ly β , nor O VI was detected.To quantify the absorption column densities and kinemat-ics, we fitted the absorption line using Voigt profile (VP)decomposition. Each VP component is described by threephysical parameters, the column density, Doppler b param-eter, and the velocity center. We employed the code M INFIT (Churchill & Vogt 2001), adopting the philosophy of enforc-ing the minimum number of statistically required VP compo-nents to model an absorption system (i.e., simultaneously toall transitions). Details of the fitting procedure are describedin Evans (2011). We tie the velocity centers for each VP com-ponent across all ions. We also assume the line broadeningis dominated by a Gaussian turbulent component; therefore,each VP component at a given velocity has the same Doppler b parameter. The latter assumption is motivated by the simula-tions of Oppenheimer & Davé (2009) in which sub-resolutionturbulence was required for modeling O VI absorbing gas inorder to reproduce the observed column density and b param-eter distributions. The one exception is z gal = 0 . + b parameters.Though Ly γ is detected for several, but not all of the ab-sorption systems, we quote VP model results only from simul-taneous fits to the Ly α and Ly β transitions (except in threesystems in which only Ly α was detected). Muzahid (2014)showed that Ly α and Ly β absorption primarily traces the high ionization phase where O VI arises, whereas the higherorder Lyman series line profiles are dominated by the lowerionization phase giving rise to C II , Si II , C III , and Si
III ab-sorption. Thus, by omitting higher order Lyman series lineswe do not lose information on the O VI phase. Most impor-tantly, by omitting higher order H I Lyman series lines, evenwhen they are detected, we present a uniform analysis of theabsorption systems. In Appendix B, we compare the derivedH I column densities, N (H I ), obtained from Ly α +Ly β VPmodels and Ly α +Ly β +Ly γ VP models for systems for whichLy γ is also detected. The exercise demonstrates that the in-clusion of Ly γ does not discernibly alter our derived N (H I )values.In the case of line blending (overlapping absorption fromtransitions associated with systems at other redshifts), whenpossible, we carefully decompose the lines using the proce-dure illustrated in Appendix C. The deblending technique isdesigned to recover the shape of the profile for the target tran-sition.Results of the absorption line analysis are listed in Table3. Column (1) gives the quasar field. Columns (2) and (3)list the spectroscopic galaxy redshift and the absorption red-shift. Columns (4)–(6) list the measured rest-frame equivalentwidths of the Ly α , Ly β , and O VI λ I and O VI columndensities, which are the sums of the VP components in eachsystem. Columns (9) and (10) list the maximum blueward andredward velocity limits of the Ly α absorption profiles.The range of rest-frame equivalent widths is 0 . ≤ W r (Ly α ) ≤ .
54 Å and 0 . ≤ W r (1031) ≤ .
45 Å, corre-sponding to the system total column densities ranges 13 . ≤ M ATHES ET AL . Figure 2.
The system total N (H I ) [sky-blue points], N (O VI ) [green], and ratio N (O VI ) / N (H I ) [magenta] plotted against D [panels (a) and (c)] and against D / R vir [panels (b) and (d)]. Open circles with downward arrows represent upper limits. The uncertainties in D are less than 1 kpc. The uncertainties in D / R vir are shown in panel (f) only. log N (H I ) ≤ . . ≤ log N (O VI ) ≤ .
7, respectively.The maximum blueward and redward velocities of the Ly α absorption are v ( - ) = - - and v ( + ) = 440 km s - , re-spectively.2.4. Presentation of Galaxy-Absorber Pairs
In Figure 1, we show three of the 14 galaxy-absorber pairsin our sample. The remaining 11 galaxies and their associatedabsorption are presented in Appendix A.For each galaxy-absorber pair, we present a portion of the
HST /WFPC2 image centered on the galaxy and the Ly α ,Ly β , and O VI λλ , D , inclination, i , and azimuthal angle, Φ . The Voigt profilemodels of the absorption lines are the red curves superim-posed on the black data. The velocity centroid of each VPcomponent is shown by the red ticks above the continuumnormalized spectra. The velocity zero-point of each spectrumis taken to be galaxy systemic velocity. CGM EXTENT AND GEOMETRY3.1.
Spatial Behavior
In Figure 2, we present the spatial behavior of the systemtotal H I column density, N (H I ), and the system total O VI col-umn density, N (O VI ). The system total column densities arethe sums of the individual Voigt profile component columndensities. We color the N (H I ) points sky-blue, the N (O VI )points green, and the ratio N (O VI ) / N (H I ) magenta. Upperlimits are shown as open circles with downward arrows.In Figure 2(a), we plot N (H I ) and N (O VI ) versus impactparameter, D . The total N (H I ) is typically log N (H I ) = 14 outto D ∼
300 kpc (we note that one system has a stringent up-per limit of log N (H I ) < N (H I ) systemsare found at D <
100 kpc, there is no statistical trend between N (H I ) and D . A Kendall- τ rank correlation test, which in-cludes upper limits, yields a 1 . σ consistency with the nullhypothesis of no correlation.We find N (O VI ) out to ∼
290 kpc to a limit of log N (O VI ) =12 .
8. We note that the detection at D = 292 kpc (inQ0454 -
132 at z = 0 . VI absorption out to ∼
260 kpc. Similar to the be-havior with N (H I ), we find no statistically significant trendbetween N (O VI ) and D (only a 2 . σ trend).In Figure 2(b), we plot N (H I ) and N (O VI ) versus D / R vir .There is a higher average value and a broader spread in N (H I )for D / R vir < D / R vir >
1, with log h N (H I ) i = IFFERENTIAL K INEMATICS Figure 3. (a) The system total N (H I ) [sky-blue] and N (O VI ) [green] plottedagainst inclination i . (b) The ratio N (O VI ) / N (H I ) [magenta] plotted againstinclination. Open circles with downward arrows represent upper limits. Theuncertainties in the inclination are shown in panel (b) only. . ± . h N (H I ) i =14 . ± .
5. outside the projected virial radius. In We findthat the N (O VI ) values inside and outside of the projectedvirial radius lie within one standard deviation of the averagelog h N (O VI ) i = 14 . ± .
44. We do not see a larger disper-sion in N (O VI ) at D / R vir < I .The ratio N (O VI ) / N (H I ) is shown in Figures 2(c) and 2(d)as a function of D and D / R vir , respectively. Due to the flatspatial distribution of N (O VI ) and the higher dispersion of N (H I ) at D / R vir <
1, the spread in N (O VI ) / N (H I ) is a factorof ≃ σ ( D / R vir ≤
1) = 1 . σ ( D / R vir >
1) = 0 . N (O VI ) / N (H I ) inside and out-side the projected virial radius are driven by the spatial be-havior of H I . Inside the projected virial radius, N (H I ) hasa larger dispersion than outside this region, whereas N (O VI )has a small dispersion both inside and outside the projectedvirial radius. The O VI column densities inside the projectedvirial radius appear to be similar to those outside this region,whereas the quantity of H I can be variable.3.2. Galaxy Orientation
To ensure we have a fair sample with which we can examinethe geometric distribution of H I and O VI absorption aroundthe galaxies, we performed Kolmogorov-Smirnov (KS) teststo determine whether the observed distributions of galaxy az-imuthal angles, Φ , and inclinations, i , are consistent with be-ing drawn from the distributions expected for an unbiasedsample. For Φ , the KS test probability is P (KS) = 0 .
62 andfor i the probability is P (KS) = 0 .
46. We conclude that both Φ and i are consistent with unbiased distributions.In Figure 3(a) and 3(b), we show the system total N (H I ), N (O VI ), and N (O VI ) / N (H I ), respectively, versus inclination.The uncertainties in the inclination measurements derivedfrom GIM2D are shown only in Figure 3(b). A Kendall- τ test (including limits) yields no trend in the N (H I ) distribution Figure 4. (a) The system total N (H I ) [sky-blue] and N (O VI ) [green] plottedagainst azimuthal angle, Φ . (b) The ratio N (O VI ) / N (H I ) [magenta] plottedagainst azimuthal angle. Open circles with downward arrows represent upperlimits. The uncertainties in the azimuthal angle are shown in panel (b) only. (1 . σ ), the N (O VI ) distribution (0 . σ ), nor in the distributionof the ratio N (O VI ) / N (H I ) (0 . σ ) with galaxy inclination.Whether a galaxy is observed with a face-on or an edge-onorientation, there appears to be a relatively flat distribution ofH I and O VI column densities.In Figure 4(a) and 4(b), we present the system total N (H I ), N (O VI ), and N (O VI ) / N (H I ), respectively, versus azimuthalangle, Φ . The uncertainties in the azimuthal angle measure-ments derived from GIM2D are shown only in Figure 4(b) butapply for all panels. A Kendall- τ test (including limits) yieldsno statistical signature for a correlation between H I columndensity and Φ (0 . σ ). We do note that the N (H I ) values ap-pear to increase toward the projected major axis ( Φ = 0 ◦ ) andthe projected minor axis ( Φ = 90 ◦ ).To crudely estimate the degree to which the two largest N (H I ) absorbers (which are most closely aligned with theprojected axes) may be outliers of the distribution of H I ab-sorbers, we calculated the mean and standard deviation of the N (H I ) excluding the two largest N (H I ) absorbers. We ob-tained log h N (H I ) i = 14 . ± .
74. The two largest N (H I ) ab-sorbers lie at 2 . σ and 5 . σ from h N (H I ) i . For this smallsample, the N (H I ) of the system most closely aligned with theprojected minor axis is an outlier of the N (H I ) distribution inthat it has a significantly larger column density. A statisticalsignature for an H I column density enhancement toward theprojected major axis is less convincing.We further examine the apparent trend for increasing N (H I )toward the projected major and minor axes by symmetricallyfolding the azimuthal angle about 45 ◦ , such that 0 ◦ corre-sponds to alignment along either the major or the minor pro-jected axis and 45 ◦ corresponds to a 45 ◦ azimuthal angle withrespect to either axis. A Kendall- τ test (including limits) onthe folded distribution yields no statistical signature for ananti-correlation between N (H I ) and angular separation awayfrom the galaxy projected axes (1 . σ ). If a trend exists, oursample is too small to reveal a statistical significance. M ATHES ET AL . Figure 5.
Projected azimuthal and D / R vir cloud locations for (a) H I and (b) O VI . The horizontal axis represents the projected major axis and the vertical axisrepresents the projected minor axis. The dashed lines represent curves of constant D / R vir . Data points are colored according to the total system column density.Column density upper limits are plotted as open circles. In the case of O VI , the azimuthal distribution of N (O VI )appears uniform, or flat, for all galaxy orientations. Com-puting the mean and standard deviation (omitting limits), weobtain log h N (O VI ) i = 14 . ± .
42 and find no outlying mea-surements with detected O VI absorption.For the ratio N (O VI ) / N (H I ), the two absorbers within ± ◦ of the major and minor axes show a smaller ratio com-pared to the data in the range 10 ◦ ≤ Φ ≤ ◦ . Excludingthese two absorbers and the two absorbers with upper lim-its, the mean and standard deviation is log h N (O VI ) / N (H I ) i = - . ± .
45. Compared to this distribution, the Φ ≃ ◦ ab-sorber is a ∼ σ outlier and the Φ ≃ ◦ absorber is a ∼ σ outlier. Since the N (O VI ) distribution is flat with azimuthalangle, whereas the N (H I ) distribution exhibits higher valuesnear the projected axes, the smaller N (O VI ) / N (H I ) ratios inthese two absorbers are driven by their higher N (H I ). We in-fer that the chemical and/or ionization conditions of the hotCGM are globally distributed, if patchy, for all azimuthal an-gles more than ± ◦ away from the projected major and mi-nor axes (acknowledging some variation as suggested by thefew upper limits).3.3. Distance and Orientation
For visualization purposes, in Figure 5(a) and Figure 5(b),we illustrate the relationship between the two-dimensionalprojected location of the absorbing gas and the system total N (H I ) and N (O VI ), respectively. The projected geometric po-sition of the quasar sightlines are computed with respect to thevirial radius using the relations D / R vir cos( Φ ) for projectionalong the galaxy major axis, and D / R vir sin( Φ ) for projectionalong the galaxy minor axis. Data point colors correspond toabsorber column density according to the color scale on theright. Upper limits are plotted as open circles.Most galaxies probed at D / R vir > N (H I ) ≃
14 lo-cated at intermediate azimuthal angles, between 10 ◦ and 90 ◦ .In turn, the galaxies probed at D / R vir < I and O VI gas is distributed around galaxies out to greaterthan 2 R vir , with higher column density H I aligned near thegalaxy projected major and minor axes for D < R vir . KINEMATICS AND VIRIAL MASSCharacterizing the velocity distribution of the hot CGM isinstrumental for determining the physical origin and fate ofboth H I and O VI absorbing gas in galactic environments. Ifthe material is outflowing, velocity flows less than the galaxyhalo escape velocity might trace gas that is likely to recycleback into the ISM and fuel star formation, whereas velocityflows greater than the escape velocity might leave the CGMpermanently and chemically enrich the IGM. If the materialis infalling from the IGM (or from satellite merging), the ve-locity distribution would provide insights into mechanisms ofhow such gas mixes with the hot CGM or eventually accretesinto the ISM. 4.1. Completeness
The detection threshold sensitivity for absorption is not uni-form from absorber to absorber due to varying signal-to-noiseratios, S / N , of the quasar spectra. Thus, for example, weakabsorption at high relative velocity that could be detected ina high S / N spectral region for one system, might not be de-tectable for a different system appearing in a lower S / N spec-tral region. In conducting our kinematic analysis, we first ex-amine the non-uniformity of the detection sensitivity.In Figure 6, we present the cumulative distribution (CDF)of the 3 σ equivalent width detection limits for both Ly α and O VI λ σ equivalent width uncertainties for unresolved lines (cf.,Schneider et al. 1993; Churchill et al. 2000) averaged over ± - relative to the galaxy redshift assuming un-resolved absorption lines. The sample is 100% completefor absorption features greater than W r (Ly α ) = 0 .
062 Å and W r (1031) = 0 .
032 Å, indicating that we generally have higher S / N spectral coverage for O VI absorption. All but two of theindividual absorption features in the sample have measuredequivalent widths above the 100% completeness level. Forour analysis, we removed these two features. IFFERENTIAL K INEMATICS Figure 6.
Cumulative distribution function of the equivalent width detection limit in (a) the Ly α absorption, and (b) the O VI λ W r (Ly α ) = 0 .
062 Å and W r (1031) = 0 .
032 Å. Only one Ly α absorption feature and one O VI λ Figure 7. (a) The velocity offsets of the H I absorbing Voigt profile compo-nents “clouds” in a system with respect to the highest N (O VI ) cloud in thesystem as a function of D / R vir . The colors of the data points are based uponthe cloud N (H I ) as given by the color bar legend. (b) The velocity offset be-tween the highest N (O VI ) cloud and the highest N (H I ) cloud in a system asa function of D / R vir . In both panels, the dotted line at v OVI - v HI = 0 indicatesno velocity offset. Kinematic Alignment of H I and O VI In Figure 7(a), we plot the velocity offset between the Voigtprofile component “clouds” with the highest N (O VI ) and eachH I absorbing cloud as a function of D / R vir . The dotted line at v OVI - v HI = 0 indicates no velocity offset between the H I cloudsand the highest N (O VI ) cloud. Most H I and O VI absorbingclouds are clustered within ∼
500 km s - .The data also reveal that the highest N (O VI ) cloud in a sys-tem does not align kinematically with the highest N (H I ) cloudin every case. To further illustrate, we plot the velocity off- set between the highest N (O VI ) cloud and the highest N (H I )cloud in a system in Figure 7(b). In 5 of 10 systems with de-tected O VI absorption, we observe a velocity offset betweenthe bulk of the neutral hydrogen and the bulk of the O VI .In three of these cases, the velocity offset is ∼
100 km s - .Since a velocity offset implies spatially separated absorbingclouds, we can infer physically distinct phases of gas (differ-ent densities, temperatures, and metallicities) in roughly halfof the sightlines through the CGM, as probed using O VI andH I as a tracer of the gas phase. Examining absorption fromthe low ions such as C II , Si II , etc., would be instrumental inexamining whether this is the case. Unfortunately, the mean N (H I ) for our sample is roughly 1.5 dex below the thresholdwhere low ion metals can be detected in spectra with moder-ate signal-to-noise ratios (Hellsten et al. 1997).4.3. Kinematics and Escape Velocity
In Figure 8, we plot the velocity difference between theindividual Voigt profile component “clouds” and the galaxysystemic velocity, ∆ v = v cld - v gal = c ( z cld - z gal ) / (1 + z gal ), asa function of halo mass, M h . Data point colors denote the D / R vir location of the absorption, where sky-blue correspondsto D / R vir ≤
1, green between 1 < D / R vir ≤
2, and D / R vir > ∼
70% of H I absorb-ing components lie within ±
200 km s - of the galaxy sys-temic velocity, ∼
80% lie within ±
300 km s - , and ∼
90% liewithin ±
500 km s - . For O VI , we find ∼ ∼ ± ± ±
500 km s - , respectively.To investigate whether clouds have relative line of sight ve-locities that exceed or do not exceed the escape velocity ofthe halo in which they reside, we computed the escape veloc-ity for each galaxy (cf., Steidel et al. 2010), v ( R ) = 2 GM h R ln[1 + c ( R / R vir )]ln(1 + c ) - c / (1 + c ) , (1)for each galaxy at R = R vir , 2 R vir , and 3 R vir , where R is galac-tocentric distance, and superimposed the results on Figure 8as colored lines corresponding to the three values of R . ANavarro et al. (1997) (NFW) dark matter halo profile is as-sumed with mean concentration parameter, c ( M h , z gal ), com-puted from the relation of Bullock et al. (2001). The curvesare not smooth with increasing virial mass due to the redshiftdependence of the concentration parameter.The galactocentric distances of the clouds are not known,only constrained to lie at R ≥ D . Assuming cloud galactocen-0 M ATHES ET AL . Figure 8.
Individual Voigt profile component (cloud) velocity offsets with respect to the galaxy systemic velocity as a function of virial mass, M h . (a) H I clouds.(b) O VI clouds. Data are colored by D / R vir bins. Vertical error bars are dominated by a ≃
30 km s - uncertainty in the galaxy redshift. The colored lines are theescape velocities, v esc , for each galaxy computed using Equation 1 for assumed cloud galactocentric distances R = R vir , 2 R vir , and 3 R vir . At a given halo mass,points of a given color are to be compared to v esc values having the same colored line. Figure 9.
The absolute relative velocity of the Voigt profile “cloud” velocities with respect to the galaxy normalized to the escape velocity, | ∆ v / v esc | , as afunction of virial and stellar mass, M h and M ∗ . (a) H I clouds. (b) O VI clouds. Data are colored by D / R vir bins using the same convention as for Figure 8. Theescape velocity is computed assuming the clouds reside at galactocentric distance R = D , which yields lower limits on | ∆ v / v esc | . Representative error bars for | ∆ v / v esc | are provided for the lowest virial mass galaxy and account for the ≃
30 km s - uncertainty in the galaxy redshift and in the virial mass propagatedthrough Equation 1. The horizontal dashed line at | ∆ v / v esc | = 1 represents cloud offset velocities equal to the halo escape velocity. Histograms (right subpanels)provide the number of clouds in equal logarithmic bins for all virial masses. The fraction of clouds with velocity offsets greater than the halo escape velocityincreases with decreasing stellar and halo mass. tric distances at multiples of the virial radius, and comparingsame-colored points and curves, we find ∼
40% of H I cloudcomponents have relative velocities in excess of the galaxyescape velocity. Roughly 60% of the clouds that reside out-side the virial radius (green and magenta data) have velocitiesexceeding the escape velocity, whereas only ∼
15% of clouds,if they reside at the virial radius (sky-blue data/lines), havegreater than escape velocities.4.4.
Differential Kinematics
In Figure 9, we show the absolute relative velocity of theabsorption with respect to the galaxy normalized to the es-cape velocity, | ∆ v / v esc | , versus virial mass. We also showthe galaxy stellar mass, M ∗ , based upon the stellar mass tohalo mass functions of Moster et al. (2010). Data points arecolored by D / R vir using the same designations as in Fig-ure 8. For this exercise, we assume R = D for the absorbingclouds, which provides the upper limit to the escape veloc- ity. Thus, the points plotted in Figure 9 are lower limits on | ∆ v / v esc | . Characteristic error bars are shown on the left-mostdata points in each panel. Points that lie above the dotted lineat | ∆ v / v esc | = 1 are clouds that have velocities in excess of thegalaxy escape velocity and, if outflowing, could be unbound.The histogram on the right shows the total number of cloudsin | ∆ v / v esc | bins for all halo masses.For H I absorption, we computed the fraction of clouds with | ∆ v / v esc | ≤
1, i.e., those that can be inferred to be gravita-tionally bound to their host halo. We divide the sample ofclouds into several subsamples based upon D / R vir and virialmass, M h . The D / R vir ranges are (0 , , , . M h < . M ⊙ , M h > . M ⊙ , and“all” M h , where M h = 10 . M ⊙ is the median virial mass ofthe sample. Comparing the “bound fraction” in each of thesesubsamples allows a differential characterization of H I kine-matics. IFFERENTIAL K INEMATICS Table 4
Bound Fractions of H I Clouds(1) (2) (3) (4) D / R vir all M h M h > . M h < . range [M ⊙ ] [M ⊙ ] [M ⊙ ]0 < D / R vir ≤ . + . - . . + . - . . + . - . < D / R vir ≤ . + . - . . + . - . . + . - . < D / R vir ≤ . + . - . · · · . + . - . < D / R vir ≤ . + . - . . + . - . . + . - . The H I cloud bound fractions are presented in Table 4. Col-umn (1) lists the D / R vir range, and columns (2), (3), and(4) list the bound fractions for the three mass ranges. Thebound fractions are n / ( n + n ), where n is the number ofclouds with | ∆ v / v esc | ≤ n is the number of clouds with | ∆ v / v esc | >
1. The quoted uncertainties assume a binomialdistribution (see Gehrels 1986) and were computed using in-complete β functions for a confidence level of 84.13% (singlesided 1 σ ). We remind the reader that we measure line of sightvelocities. Despite having constraints on the galaxy inclina-tions and azimuthal angles on the sky, deprojecting the gas ve-locities is an intractable problem due to significant uncertaintyin the true gas motions which are a result of the complex in-terplay between outflow geometry, environmental conditions,and inflow dynamics.Examining column (2) of Table 4, we find that the boundfraction decreases as D / R vir increases. In other words, theproportion of clouds with greater than escape velocities in-creases with increasing projected distance relative to the virialradius. On average, ∼
40% of the clouds could be inferred tobe escaping the halo for D / R vir ≤ D / R vir range, higher mass halos have a largerfraction of bound clouds than do lower mass halos. Or, alter-natively, the fraction of clouds with | v | > v esc is larger in lowermass halos than in higher mass halos. On average, ∼
75% ofthe clouds have | v | > v esc for lower mass halos, whereas only ∼
10% of the clouds have | v | > v esc for higher mass halos.When interpreting the trends presented in Table 4, we mustbe careful to consider the selection effect that, in small sam-ples characterized by a pre-selected D range, lower mass halosare preferentially probed at larger D / R vir because they havesmaller virial radii than larger mass halos. Thus, the majorityof the absorbers in the M h < . M ⊙ subsample are probedat D / R vir >
1. The trend that the bound fraction decreases as D / R vir increases may be enhanced in our sample due to the in-crease in the relative number of lower mass galaxies at larger D / R vir . This is corroborated by the result that higher mass ha-los have a larger fraction of bound clouds than do lower masshalos in each finite D / R vir range, which likely does not sufferfrom any selection bias and is a more robust finding.Examining O VI absorbers, we find associated O VI absorp-tion in only ∼
40% of the H I clouds in and around lower masshalos as compared to ∼
85% around higher mass halos. Giventhe flat H I column density distribution for our sample, thelower number of detected O VI clouds in lower mass halossuggests conditions favoring higher O VI column densities areless common out to D ≃
300 kpc of lower mass halos than forhigher mass halos. For O VI absorbers, as shown in Figure 9(b), the clouds havea bound fraction of 0 . + . - . for all halo masses in the sample.For higher mass halos, the bound fraction is 1 . + . - . and forlower mass halos the bound fraction is 0 . + . - . . We thus caninfer that O VI absorbing gas is more common in higher masshalos and is primarily bound to the halo, whereas O VI absorb-ing gas is less commonly found in the vicinity of lower masshalos and only half of the O VI absorbing clouds are bound. DISCUSSION5.1.
Spatial Extent of the Hot CGM
Our small sample of 14 galaxy-absorber pairs is similar tothat of the COS-Halos project (Tumlinson et al. 2013), butwith a few differences. First, our sample probes H I and O VI absorption out to 300 kpc, whereas the COS-Halos sampleprobes out to 150 kpc. Second, our sample covers a slightlysmaller range of virial mass, 10 . ≤ log( M h / M ⊙ ) ≤ .
2, ascompared to COS-Halos, 11 . ≤ log( M h / M ⊙ ) ≤ .
3. Wethus probe to 0.5 dex lower in virial mass, but do not probethe highest full decade of the COS-Halos sample. The redshiftcoverage is roughly identical to that of COS-Halos ( z . . z = 0 . a priori pre-disposition to H I absorption in the back-ground quasar spectra. We do not have the data to estimatethe specific star formation rates of the galaxies in our sam-ple. However, since Tumlinson et al. (2013) conclude thereis very weak evidence for a difference in the detection fre-quency of H I between “star-forming” and “passive” galaxies,there should be little-to-no ambiguity in comparing the neu-tral hydrogen between samples.To a 3 σ equivalent width detection sensitivity of W r (Ly α ) =0 .
05 Å (100% completeness, corresponding to log N (H I ) = 13for b = 30 km s - ), we find H I absorption is present out to300 kpc for 13 of 14 galaxies in our sample, indicating thatH I gas is clearly present out to ∼ D / R vir <
1, we measure a mean system total column densityof log h N (H I ) i = 15 . ± .
6, which is in good agreement withthe column densities found in this region for the COS-Halossample, which probes out to D / R vir ≃ .
75. For D / R vir > h N (H I ) i = 14 . ± .
5, a value ≃ . D / R vir ≃ I , possibly due to a changes in cloud den-sities, sizes, or ionization conditions.The behavior of H I that we described above is consistentwith the conclusion drawn from Tumlinson et al. (2013), thatthere is significant “evolution” in the H I properties betweenthe regions D <
200 kpc and outside this region, as basedupon their comparison with several other studies of H I ab-sorption around galaxies (see their Section 5.1). In our smallersample representing the lower mass range of COS-Halos, this“evolution”, or transition, appears to set in at D ≃
100 kpc.This could imply that lower mass halos have smaller phys-ical extent than higher mass halos (also see Churchill et al.2013b; Ford et al. 2013). Consistent with the conclusions ofTumlinson et al. (2013) and Stocke et al. (2013), we also in-fer that, based upon the behavior of H I absorption, the virialradius appears to be a transition region between the CGM andthe IGM. Our data suggest this region is quite extended andis not abrupt. However, this does not preclude that possibilitythat metals from the ISM are being transported through theCGM to several virial radii and out to the IGM.2 M ATHES ET AL .To a 3 σ detection sensitivity of W r (1031) = 0 .
032 Å (100%completeness, corresponding to log N (O VI ) = 13 . b =20 km s - ), we find O VI absorption is present out to the im-pact parameter limit of our survey ( ∼
300 kpc) for 11 of 14galaxies in our sample. This corresponds to O VI absorptionas far out as D / R vir ≃ .
7, which could suggest that O VI en-richment for the host galaxy as far as ∼ N (O VI ) with D nor with D / R vir , though the five detections at D / R vir < N (O VI ) absorbers.Out to their survey limit of D = 150 kpc ( D / R vir ≃ . VI absorption with log N (O VI ) ≥ . VI absorption and the star formingproperties of the galaxies in our smaller sample. However,it would be interesting to do so given our deeper detectionsensitivity to O VI absorption. COS-Halos is roughly 20%complete (9/42) for detections below log N (O VI ) = 14 . N (O VI ) = 14 .
2, whereas we are 100%complete to log N (O VI ) = 13 .
5. Of interest is that the “pas-sive” galaxies are among the most massive galaxies in theCOS-Halos sample and lie in the mass range not representedin our sample. If we speculate that this implies our lowermass galaxies are drawn from the same population as thestar-forming population represented in the COS-Halos sur-vey, then we have found examples where the CGM of starforming galaxies can have detectable absorption weaker thanlog N (O VI ) = 14 . N (O VI ) / N (H I ) is consistent with being flat out to D = 290 kpcand D / R vir ≃ .
7. However, in terms of individual clouds(VP components), there is a higher incidence of O VI absorp-tion in H I clouds for higher mass halos than in lower masshalos. Dividing the sample by the median virial mass oflog M h / M ⊙ = 11 .
5, O VI absorption is found in only ∼ I clouds in the around lower mass halos as comparedto ∼
85% around higher mass halos. Since the system to-tal N (H I ) is fairly flat, the smaller fraction of detected O VI clouds in lower mass halos suggest conditions favoring O VI are less common out to D ≃
300 kpc of lower mass halos thanfor higher mass halos.5.2.
Geometric Distribution of the Hot CGM
Using Mg II absorbers, Kacprzak et al. (2012a) andBouché et al. (2012) have shown cool/warm CGM gas is morefrequently found to be aligned with either the galaxy projectedmajor or minor axes. Bordoloi et al. (2011) finds strongerMg II is preferentially aligned with the projected minor axis.Kacprzak et al. (2011) find that Mg II equivalent widths cor-relate with galaxy inclination when scaled by the impact pa-rameter. Based upon these results, the authors have sug-gested wind driven material may be responsible for the en-hanced absorption strengths aligned with the galaxy minoraxes, whereas the inclination correlation may indicate a planerdistribution (also seen as a projected major axis alignment) ofabsorbing gas, perhaps accreting from the IGM.For our sample, our data suggest that higher H I columndensity gas is preferentially found within ± ◦ of the majorand minor axes (inside the projected virial radius). However,this is not a statistically significant result.In the case of O VI , as seen in Figure 4(b), the azimuthaldistribution of N (O VI ) is statistically consistent with beingflat. However, the ratio N (O VI ) / N (H I ) for the two absorbers within ± ◦ of the major and minor axes have the smallestvalues and are statistical outliers (to better than 4 σ ) as com-pared to the values in the range 10 ◦ ≤ Φ ≤ ◦ . This wouldsuggest that, on average, the chemical and/or ionization con-ditions of the hot CGM are fairly uniform in their geomet-rical distribution around galaxies for azimuthal angles morethan ± ◦ away from the projected major and minor axes (ac-knowledging some variation as suggested by the fact that notall H I clouds exhibit O VI absorption).The higher H I column density gas clouds with lower N (O VI ) / N (H I ) ratios, but with typical N (O VI ), that re-side within ± ◦ of the major and minor axes, may reflectthe presence of multi-phase gas at these geometric projec-tions, with most of the H I associated with higher density,lower ionization gas. The ionization conditions, densities,and temperatures of the Φ ≃ ◦ (minor axis aligned) ab-sorber (Q1317 + z = 0 . Φ ≃ ◦ (major axis aligned) absorber(Q1136 - z = 0 . II λ III λ II λλ , III λ III and Si
III are clearly detected in absorption, C II may be weakly de-tected, and Si II resides in a very noisy region of the spectrum.We might infer that the two absorbers within ± ◦ of themajor and minor axes have larger N (H I ) columns becausethey are multi-phase systems, whereas the systems at greaterangular separation from the galaxy projected axes trace thehot CGM. This would be consistent with the findings ofKacprzak et al. (2012a), who report an increase in the fre-quency of Mg II absorbers with azimuthal locations alignedwith the projected major and minor axes. Unfortunately, the N (H I ) values for the remaining absorbers in our sample areroughly 1.5 dex below the threshold where low ion metalscan be detected, even for solar metallicity gas (Hellsten et al.1997). 5.3. Interpreting the Kinematics
Constraining the galactocentric distances of the absorbingclouds is central to interpreting the differential behavior in thebound fraction of H I clouds (see Figure 9 and Table 4). Thedata place only lower limits of R ≥ D on their galactocen-tric distance and R / R vir ≥ D / R vir on their distances relativeto the virial radius; we cannot definitively determine whethera given absorber arises in the IGM beyond R = 4 R vir , whereHubble flow begins to dominate the peculiar velocities (seeCuesta et al. 2008).Based upon SPH simulations, Oppenheimer & Davé (2009)argue that weaker O VI absorbers, log N (O VI ) ≃
14, may tracethe old high-metallicity regions of the IGM and that manyof these absorbers are not dynamically associated with thegalaxy closest in projection on the sky. They find that manyof the absorbers could have originated from a different galaxyat an earlier epoch and show (see their Figure 15) that weakO VI absorbers could arise between 1 ≤ R / R vir ≤
10, where R is the galactocentric distance. This range corresponds to100 kpc ≤ R ≤ | ∆ v | , is due toHubble flow (with zero peculiar velocity), we can estimate R . IFFERENTIAL K INEMATICS Figure 10.
The proper galactocentric distances, R , to the absorbing clouds assuming that the observed cloud-galaxy velocity offsets are pure line of sight Hubbleflow, i.e., | ∆ v | = | v HF (los) | . (a) R versus v HF (los) for z gal = 0 .
15 and z gal = 0 .
65, illustrating Eq. 2. At each z gal , the curves are independent of D for R ≫ D . Thehorizontal line at R = D max = 290 kpc indicates the maximum impact parameter of the sample. (b) R as a function of | v HF (los) | / v esc for the individual clouds inour sample. Points are colored by their D / R vir location using the color scheme employed for Figures 8 and 9. Filled points are clouds for which O VI absorptionis detected with the H I absorption and open points are H I only clouds. The vertical line is | v HF (los) | / v esc = 1. (c) R / R vir as a function of | v HF (los) | / v esc . Thevertical line is | v HF (los) | / v esc = 1 and the horizontal line is R / R vir = 4, the spatial location around a halo where Hubble flow dominates over dark matter accretion. The observer line of sight Hubble flow velocity for a cloudwith impact parameter D at galactocentric distance R from agalaxy at z gal is v HF (los) = H E ( z gal ) R q - ( D / R ) , (2)where E ( z ) = √ Ω m (1 + z ) + Ω Λ . In Figure 10(a), we plot R asa function of v HF (los) for z gal = 0 .
15 and z gal = 0 .
65, whichbracket the redshifts of our sample. When D / R ≪
1, thecurves are independent of D .In Figure 10(b), we plot R as a function of | v HF (los) | / v esc forthe individual clouds in our sample by assuming that the ob-served velocity offset of the cloud is pure line of sight Hubbleflow, i.e., | ∆ v | = | v HF (los) | . Points are colored by their D / R vir location using the color scheme employed for Figures 8 and 9.Filled points are clouds for which O VI absorption is detectedwith the H I absorption, and open points are H I only clouds.If the H I and O VI absorbing clouds are interpreted in thecontext of the predictions of Oppenheimer & Davé (2009)[see their Figure 15], we would expect the clouds residewithin R ≤ ∼ ∼ VI absorption would be predicted to reside between1 Mpc and 6 Mpc. Our O VI absorbing clouds, which havelog N (O VI ) ≃
14, would not be analogues of the “dynamicallyunassociated” O VI absorbers of Oppenheimer & Davé (2009)if they reside at R > R / R vir as a function of | v HF (los) | / v esc . UsingAMR cosmological simulations, Cuesta et al. (2008) showedthat the influence of the halo gravitational potential on darkmatter particles extends no farther than R ≃ R vir for halomasses ranging from 10 ≤ log M h / M ⊙ ≤
14; Hubble flowdominates for R / R vir > ∼
70% of the clouds and ∼
60% of the clouds with The radial Hubble flow velocity of a source at z s for an observer at z o is v HF = cE ( z o ) D c ( z o , z s ) / (1 + z o ), where D c ( z o , z s ) is their radial comovingseparation. O VI absorption reside at R / R vir >
4. The 16 clouds with | v HF (los) | / v esc > ≤ R / R vir ≤ VI absorption would reside between15 ≤ R / R vir ≤ R = 15 R vir from their identified galaxies, or (2)the clouds are in fact associated with their host galaxies andthat the velocity offsets are peculiar velocities due to physi-cal and dynamical processes within R ≃ R vir . As we discussbelow, our exercise leaves very little room for ambiguity be-tween these very different scenarios.The surveys of Tripp et al. (2001, 2006), andTumlinson et al. (2005, 2011) have shown that the near-est projected neighboring galaxies are within 200 kpcof O VI absorbers. Stocke et al. (2006) finds that, forlog N (O VI ) ≥ .
2, the median distance of O VI absorbersfrom the nearest projected galaxy is 350-500 kpc for L ∗ galaxies and 200-270 kpc for 0 . L ∗ galaxies. In addition,we note that if O VI absorbers with column densities in theregime detected in our survey are in the IGM at Mpc distancefrom galaxies, then the covering fraction of O VI absorbersshould have no dependence on galaxy property. However,precisely the opposite is observed in that Tumlinson et al.(2011) reports a reduced frequency of O VI absorbers in thevicinity of galaxies with lower specific star formation rates,at least for log N (O VI ) ≥ . D = 150 kpc. Finally,we argue that, if the transition from the CGM to the IGMoccurs at an overdensity of log ρ H / ¯ ρ H ≃ .
5, as indicatedby the observations of Steidel et al. (2010), Prochaska et al.(2011), and Rudie et al. (2012) and cosmological simulationssuch as those of Davé et al. (1999), then the IGM at z < . N (H I ) .
13. The individualclouds we are studying have log N (H I ) ∼
14 correspondingto log ρ H / ¯ ρ H ≃ .
3, suggesting that they reside in the regimeof R / R vir . ATHES ET AL .the halos of their host galaxies, we still cannot directly dis-tinguish whether the clouds are outflowing or inflowing fromthe data themselves. However, simple gravitational energyconserving physical arguments can be invoked to show thatinfalling material is not expected to have velocity offsets withrespect to the galaxy that exceed the halo escape velocity.First, material does not fall into halos from infinity, but fromthe “Eulerian sphere”, a region with a ∼ Λ CDM cosmological simulationssupport such expectations (e.g., Cuesta et al. 2008). Thoughthe infall velocities of a non-negligible fraction of the infallingdark matter particles exceed the circular velocity at the virialradius, virtually none exceed the escape velocity. Second,gas experiences hydrodynamic forces that act to decelerateinfalling gas. v r a d ( k m s − ) V esc V circ R vir l og ( M ga s ) [ M ⊙ k p c − / ( k m s − ) ] Figure 11.
Radial velocity versus galactocentric radius of gas mass in unitsof solar masses per unit kpc per unit velocity (see the color bar) showing theoutflow (positive velocities) and inflow (negative velocities) into a simulatedgalaxy with M h ≃ × M ⊙ . The vertical dotted line is the virial ra-dius. The solid green curves represent the escape velocity as computed fromEquation 1 and the blue curves represent the circular velocity. Note that thevelocity of infalling gas does not exceed the escape velocity. The kinematic behavior of gas can be examined in hydrody-namic + N-body Λ CDM cosmological simulations that com-pare various stellar feedback recipes (Trujillo-Gomez et al.2014). We examined the gas kinematics in a simulated galaxyfrom the work of Trujillo-Gomez et al. (2013). We use themodel spRP_40, which has virial mass log M h / M ⊙ = 11 . - ] as a functionof radial velocity and galactocentric radius and present the re-sults in Figure 11. The dotted vertical line provides the loca-tion of the virial radius and the solid curves provide the cir-cular velocity and the escape velocity as a function of galac-tocentric distance. Outflowing material (positive velocities) isseen to have radial velocities exceeding the escape velocity,whereas inflowing material (negative velocities) is always in-falling with radial velocities roughly a factor of ≃ . Implications of Differential Kinematics
Having argued that the absorbing clouds in our sample arebest interpreted as residing within R ≃ R vir of their galax-ies and that those clouds with | v | > v esc are outflowing fromthe galaxy, we now explore the implications for galaxy evo-lution in light of our finding of differential kinematics (seeSection 4.4).By differential kinematics, we are referring to the result inwhich we find that the lower mass subsample has a smallerfraction of bound clouds than the higher mass subsample and,for all masses, the bound fraction decreases as D / R vir in-creases. Summarizing, dividing the sample into lower massand higher mass halos, we find that for D / R vir <
1, lower masshalos have an escape fraction of ∼ ∼ v esc cannot achieve a distance greater than theirturn-around radius, as we probe further out from the galaxy,we would naturally find that the fraction of higher velocityclouds increases.For our higher mass subsample, 11 . < log( M h / M ⊙ ) ≤ . I and O VI absorbing clouds are consistent with the findings of COS-Halos (Tumlinson et al. 2011, 2013). However for our lowermass subsample, 10 . ≤ log( M h / M ⊙ ) < .
5, we find higherescape fractions, and this holds for all D / R vir bins. This im-plies that galaxies with halo masses of log( M h / M ⊙ ) < . expel a larger portion of their winds to the IGM than dohigher mass galaxies. It also implies that wind recyclingwould characteristically be more common in higher massgalaxies than in lower mass galaxies. Thus, our result of differential kinematics has implica-tions for the recycling of wind material as a function of halomass, and in fact, is consistent with the “differential windrecycling” scenario proposed by Oppenheimer et al. (2010).Wind recycling serves as a third mechanism, in additionto “cold” and “hot” mode accretion (e.g., Kereš et al. 2005;Dekel & Birnboim 2006), for gas accretion into the ISM forfueling star formation. In the differential wind recycling sce-nario, the wind recycling time decreases with increasing halomass, flattening toward the highest masses. Towards lowerhalo mass, the recycling time exceeds the Hubble time, so thatlower mass galaxies would not experience wind recycling buthave their star formation fueled primarily through cold mode
IFFERENTIAL K INEMATICS
Comparing to Wind Models
Here, we undertake an exercise to estimate the degree towhich the realization of the data from our sample may beconsistent with simple wind models. We investigate threeconstant-velocity wind models using the Monte Carlo tech-nique. We employ the two-dimensional distribution of datapresented in Figure 9. i.e., | ∆ v | / v esc vs. M h , to constrainwhether the models are statistically inconsistent or are notinconsistent with the realization of our data. Our aim is todetermine the degree to which the paucity of data points with | v (los) | / v esc > | v (los) | / v esc < v w , independent of galaxyhalo mass, (2) a random wind velocity ranging from 0 km s - to a maximum velocity, v w , also independent of galaxy halomass, and (3) the vzw wind model of Oppenheimer et al.(2010), in which the wind velocity scales with the stellar ve-locity dispersion σ ∗ ( r ), where r is the radius at which thewinds are launched. The vzw wind model is therefore halomass dependent.The model is one dimensional in which the wind veloci-ties are plane-parallel and randomly oriented at some angle, θ , with respect to the observer’s line of sight. There is signif-icant uncertainty concerning the orientation of galactic windswith respect to galaxy inclination and position angle (i.e. ori-entation on the sky). Therefore, we adopt this simple modelusing an unweighted distribution of random angles in hopesof capturing the stochastic effects of varying galactic outflow conditions without introducing extra free parameters and/orpossible model biases.For the first two models, we varied v w over the range 100 to1500 km s - . For the vzw model (also known as “ momentumdriven winds”), the wind velocity is given by v w = 3 σ ∗ ( r ) p f L - , (3)where f L is the luminosity factor. FollowingOppenheimer et al. (2010), we adopt f L = 2. The stel-lar velocity dispersion at the radius where the winds arelaunched is given by σ ∗ ( r ) = r - Φ ( r ) , (4)where Φ ( r ) is the gravitational potential evaluated at the windlaunch radius. We assume an NFW profile for which Φ ( r ) = - π G ρ r s ln(1 + r / r s ) r / r s , (5)where r s is the scale radius, and ρ is given by ρ = M ( R vir )4 π r s [ln(1 + c ) - c / (1 + c )] , (6)where c is the concentration parameter. Note that M ( R vir ) cor-responds to our measurement M h . The concentration parame-ter is both halo mass and redshift dependent; for this exercise,we adopt the median redshift of the sample, z = 0 . M h , in the range of thesample galaxies, 10 . < log( M h / M ⊙ ) < .
2, from which wecompute the virial radius R vir . We then generate a wind orien-tation in the range 0 ◦ < θ < ◦ and an impact parameter inthe range 57 < D <
292 kpc (the range of the sample). Wethen compute the escape velocity at the galactocentric dis-tance equal to the impact parameter D , reproducing the v esc employed for Figure 9. For the constant velocity wind model,we assign a value to v w . For the random velocity model, weassign a maximum value of v w and then multiply by a ran-dom U (0 ,
1) deviate. In the case of the momentum-driven vzw wind model, we specify the launch radius of the wind andcompute v w from Equation 3. Finally, we determine the lineof sight “observed” velocity | v (los) | = v w · cos( θ ), from whichwe compute the ratio | v (los) | / v esc .For a given wind model, we generate 100,000 realizations(galaxy/wind pairs). From these pairs, we randomly draw 41galaxy/wind pairs but enforce that the 41 pairs match the num-ber of data points on Figure 9 with D / R vir ≤
1, 1 < D / R vir ≤ < D / R vir ≤ . ≤ log M h / M ⊙ ≤ .
2. Thus, thetwo-dimensional distribution of halo mass and D / R vir of the41 galaxy/wind pairs emulates that of the observed data onFigure 9. On the | v (los) | / v esc – M h plane, we then computethe two-dimensional KS statistic between the galaxy/windpairs and the data points in order to quantify the degree towhich the distribution of wind model points is inconsistentwith the distribution of observed points. We adopt the crite-rion that the model points are inconsistent with the data when P (KS) ≤ . σ ) or higherconfidence level.We repeat the entire process for 100,000 trials, each timecalculating the two-dimensional KS probability comparingthe model data to the observed data. Finally, we compute the6 M ATHES ET AL .fraction of model trials for which P (KS) ≤ . σ level). As thisfraction, f ( P KS < . Figure 12.
The fraction of realizations, f ( P KS < . σ level versus the wind velocity, v w , for three differentwind models. Panel (a) shows a linearly scaled zoom-in of the upper regionof panel (b). For the constant wind velocity model, v w is the wind veloc-ity. For the random wind velocity model, v w is the maximum wind velocity,which can range from 0 km s - to v w . The vzw wind model (right panels) iscomputed for wind launch radii over the range r = 1–10 kpc in steps of 1 kpc.The upper point corresponds to r = 1 kpc and the lower point corresponds to r = 10 kpc. The random wind models with wind velocities peaking around ∼
500 km s - are most frequently consistent with the data. In Figures 12(a) and 12(b), we plot f ( P KS < . v w for the constant velocity and random velocitywind models. For the random wind model, v w represents themaximum value of the constant velocity wind. Panel (a) isa linearly scaled zoom-in of the upper portion of panel (b).Where the curves have f ( P KS < . ≥ .
9, the distributionof the model data is ruled out at the 3 σ level for 90% ormore of the realizations. The right-hand panels show f ( P KS < . vzw wind model for ten different launch radiiranging from r = 1 to 10 kpc in steps of 1 kpc. For this model,the value of f ( P KS < . f ( P KS < . r = 1kpc).For the constant velocity wind model, f ( P KS < . ≥ . v w ≥
650 km s - . That the range of velocitiesbelow this value is less frequently inconsistent with the datais not outside of expectations, since 90% of the H I absorbingcloud velocities lie within ∆ v = ±
500 km s - . Note that it isvery rare for the realizations to be inconsistent with the datafor v w ≃
200 km s - , the value within which 70% of all H I absorbing cloud velocity offsets lie with respect to the galaxy.For the random velocity wind model, f ( P KS < . ≥ . v w ≃
100 km s - . Recall that in this model, thewind velocity of any given galaxy/wind pair falls in the range 0 km s - to the maximum velocity, v w . We find f ( P KS < . < . v w >
100 km s - ,indicating that the random velocity wind model cannot beruled as being inconsistent with the data for more than 90%of the realizations over the range 100 < v w ≤ - .For the vzw wind model, f ( P KS < . ≥ .
90 occurs forlaunch radii r ≤ vzw wind model cannotbe ruled as inconsistent with the data; however, for reasonablyphysical launch radii, 90% of the realizations are inconsistentwith the data to the 3 σ level. Generally, the vzw wind modeldoes not exhibit a compelling signature for being consistentwith the data.That the random velocity wind model can be consistentwith the data suggests that, even in a small sample (14 galax-ies with 41 absorbing clouds), highly variable wind velocitiesfrom galaxy to galaxy can give rise to the differential kine-matics we inferred from the data. This remains consistentwith our previous statement that one possible explanation fordifferential kinematics is that (in real galaxies) the higher thelaunch velocity of the wind, the further from the galaxy theabsorbing clouds potentially travel, so that as we probe fur-ther out from a galaxy we observe a higher fraction of highervelocity clouds.If the cloud velocities are decelerated dynamically inhigher mass galaxies, as found in the simulations ofOppenheimer et al. (2010), then higher wind velocities wouldbe more frequently observed in the outer extended CGM oflower mass galaxies, as we have inferred for our sample. Inthe wind model of Chelouche & Bowen (2010), the wind ve-locity is proportional to the star formation rate in the galaxydisk. We do not have estimates of the star formation ratesfor the galaxies in our sample. But, we note that variationsin the star formation rate from galaxy to galaxy would mani-fest in their model in a manner similar to our random veloc-ity wind model. We also note, generally, that lower (stellar)mass galaxies tend to have higher specific star formation ratesthan higher mass galaxies (cf., Schawinski et al. 2014), and itseems reasonable that specific star formation rate would cor-relate with wind velocity. CONCLUSIONSWe have presented an analysis of the spatial and geo-metric distribution and kinematics of H I and O VI absorp-tion surrounding 14 galaxies within a projected distance of D = 300 kpc of background quasars. The galaxies are imagedusing HST /WFPC2 and their morphological and orientationparameters have been measured using GIM2D. The absorp-tion is measured in
HST /COS or
HST /STIS quasar spectra.We have focused our analysis on the Ly α and Ly β transi-tions, and O VI λλ , . ≤ z ≤ .
67, and an impact parameter range 60 ≤ D ≤
290 kpc. Thegalaxy virial masses range from 10 . ≤ log( M h / M ⊙ ) ≤ . ≤ R vir ≤
225 kpc.The median virial mass is log( M h / M ⊙ ) = 11 .
5. The range of D / R vir spans from 0 .
45 to 2 .
75. The range of galaxy inclina-tions and azimuthal angles are 18 ◦ ≤ i ≤ ◦ and 6 ◦ ≤ Φ ≤ ◦ , respectively.6.1. Spatial and Geometric Distributions
We first highlight some general results with regards to thespatial and geometric distribution of H I and O VI absorbing IFFERENTIAL K INEMATICS N (H I ) systems are found at D <
100 kpc, there is no statistical trend between N (H I ) and D .Over the range 100 ≤ D ≤
300 kpc, the system total N (H I )is typically log N (H I ) ≃
14. For all D , the mean and dis-persion is log h N (H I ) i = 14 . ± .
61. We detect O VI asfar as D ∼
290 kpc (our sample maximum), to a 3 σ limitof log N (O VI ) = 12 .
8. The distribution of N (O VI ) is effec-tively flat as a function of D , showing no statistically sig-nificant trend between N (O VI ) and projected distance fromthe host galaxy. The mean and dispersion is log h N (O VI ) i =14 . ± . N (H I ) for D / R vir < D / R vir >
1, withlog h N (H I ) i = 15 . ± . h N (H I ) i = 14 . ± . N (O VI ) with D and the higher averagevalue and dispersion of N (H I ) at D / R vir <
1, the dispersion in N (O VI ) / N (H I ) is a factor of ≃ N (H I ), N (O VI ), or N (O VI ) / N (H I ) and galaxy inclination.Statistically, there is no correlation between these quantitiesand azimuthal angle. However, in our small sample, N (H I ) islargest when probed nearest to the project axes of the galaxyand decreases as the azimuthal angle increases away from theprojected axes.4. Within D = 300 kpc, there is a higher incidence of O VI absorption in higher mass halos than in lower mass halos, us-ing the sample median of log M h / M ⊙ = 11 . VI absorption in only ∼
40% of the H I clouds in and aroundlower mass halos as compared to ∼
85% around higher masshalos. Since the system total N (H I ) is fairly flat, the smallerfraction of detected O VI clouds in lower mass halos suggestconditions favoring O VI is less common out to D = 300 kpcof lower mass halos than for higher mass halos, but that thephysical conditions of the gas are not dissimilar.In summary, the highest N (H I ) clouds reside within the pro-jected virial radius and are found at azimuthal angles closelyaligned with the galaxy projected axes. It could be that for D / R vir <
1, the H I in our sample is reflecting the presenceof a cool/warm gas phase preferentially found along the pro-jected galaxy axes, such is observed for Mg II absorption(Bordoloi et al. 2011; Kacprzak et al. 2012a; Bouché et al.2012). The data we have in hand cannot definitively ad-dress the presence of a cool/warm phase. Overall, it ap-pears that there is a transition in the behavior of H I absorp-tion in the regime of D / R vir ∼
1, in which higher systemtotal N (H I ) is found inside the projected virial radius andlog N (H I ) ≃
14 outside the projected virial radius at least asfar as D / R vir ≃
3. For O VI absorption, the distribution ofthe system total N (O VI ) is flat for all D / R vir out to at least D / R vir ∼ .
8. O VI absorbers are more common in the CGMof higher mass halos out to D ≃
300 kpc. Altogether, N (O VI )shows no preferred geometric dependencies, suggesting thatregions of hot CGM gas are quite globally distributed.6.2. Differential Kinematics
The main result of this paper is differential behavior in thefraction of bound clouds (individual VP components) as afunction of both virial mass, M h , and virial radius, R vir . Wecalled this behavior “differential kinematics”. These findingsare shown in Figure 9 and Table 4. Figure 9 shows the ab- solute relative velocity of the Voigt profile “cloud” velocitieswith respect to the galaxy normalized to the escape velocity, | ∆ v / v esc | , as a function of virial and stellar mass, M h and M ∗ .Table 4 lists the fraction of clouds that can be inferred to bebound to the host halo as a function of M h and D / R vir . Theinterpretation relies heavily upon the inference (presented inSection 5.3) that clouds with | ∆ v / v esc | > I and O VI absorbing clouds are clustered within ∼
500 km s - . In ∼
50% of the systems with detected O VI absorption, we observe a velocity offset between the bulk ofthe H I and the bulk of the O VI , as defined by the highestcolumn density Voigt profile components in a system. In threeof these cases, the velocity offset is ∼
100 km s - . The datasupport the idea that the log N (H I ) ≃
14 regime of the CGMrepresents various gas conditions as inferred from H I and O VI absorption, even though the system total N (H I ), N (O VI ), and N (O VI ) / N (H I ) show little variation from system to system.2. When the full range of M h and D / R vir of the sample areexamined, ∼
40% of the H I absorbing clouds can be inferredto be escaping their host halo. Segregating the sample intofinite ranges of D / R vir , the fraction of bound clouds decreasesas D / R vir increases such that the escaping fraction is ∼ D / R vir < ∼
45% for 1 ≤ D / R vir <
2, and ∼
90% for 2 ≤ D / R vir <
3. That is, averaged over all M h , the fraction of H I absorbing clouds that could be escaping the galaxy increaseswith increasing D / R vir .3. Dividing the sample into lower mass and higher mass ha-los, where the dividing virial mass is the median of the sam-ple, log M h / M ⊙ = 11 .
5, we find that the lower mass subsam-ple has a smaller fraction of bound clouds in each of the threeaforementioned D / R vir ranges. For D / R vir <
1, lower masshalos have an escape fraction of ∼ ∼ ≤ D / R vir < ∼
55% and ∼
35% for lower massand higher mass halos, respectively. For 2 ≤ D / R vir <
3, theescape fraction for lower mass halos is ∼
90% (higher masshalos were not probed in this range in our sample).4. We demonstrated that the absorbing clouds are likely tobe outflowing winds, since their kinematics are not consistentwith infall kinematics, based upon feedback simulations. Weshowed that the absorbing gas is likely to reside within 4 R vir of the galaxies, also based upon simulations and the dynam-ics of Hubble flow. We explored three constant velocity windmodels to explore the degree to which the observed charac-teristics of differential kinematics are inconsistent with thesemodels. We find that the most consistent constant wind ve-locity model is that with random winds velocities in the range300 ≤ v w ≤
800 km s - , and suggest that specific star forma-tion rate, from galaxy to galaxy, coupled with higher dynam-ical deceleration of the gas in higher mass halos, may be in-strumental in explaining differential kinematics.Differential kinematics may be an observational signaturesupporting the theoretical scenario of differential wind recy-cling proposed by Oppenheimer et al. (2010). If so, differ-ential kinematics would be an important finding that shouldbe verified and further characterized with additional obser-vations. It is becoming well accepted that wind recyclingthrough the CGM is an important regulating process forgalaxy evolution and may, to a large degree, control the shapeof the stellar to halo mass function and the mass-metallicity8 M ATHES ET AL .relationship of galaxies.We thank Ben Oppenheimer for helpful and insightful dis-cussions on the details of his wind simulations. NLM, CWC,and SM were supported mainly through grant HST-GO-13398and JCC and SM were partially supported by grant HST-AR-12644, both provided by NASA through the Space TelescopeScience Institute, which is operated by the Association of Uni-versities for Research in Astronomy (AURA) under NASAcontract NAS 5-26555. ST-G was supported through the Re-search Enhancement Program awarded to CWC provided byNASA’s New Mexico Space Grant Consortium (NMSGC).NLM and NMN were partially supported through NMSGCGraduate Research Fellowships. NMN was also partially sup-ported through a three-year Graduate Research EnhancementGrant (GREG) sponsored by the Office of the Vice Presidentfor Research at New Mexico State University.APPENDIX A. INDIVIDUAL QUASAR FIELDSAll spectroscopic redshift data for the galaxies analyzed inthis paper come from one of five different sources. (1) Veryearly work was conducted by Ellingson & Yee (1994), whoemployed the MARLIN/LAMA Multiobject Spectrograph onthe Canada-France-Hawaii Telescope (CFHT). They cite a78% completeness level for successful spectroscopic identi-fication of observed galaxies with m r ≤ .
5, and 49% forthe fields overall. (2) A survey by Lanzetta et al. (1995) usesthe Kast Spectrograph on the Lick Observatory 3-m tele-scope. This study is 37% complete for m r < .
5, withlimiting magnitudes of m r = 23 . m r = 23 . + + m r = 22 .
5. (4)An
HST imaging survey, using FOS quasar spectroscopy, wasconducted by Chen et al. (1998, 2001), targeting the fieldsstudied by Lanzetta et al. (1995) and additional fields forwhich much of the details are to appear in Chen et al. (2001,in preparation). Estimates on the completeness and magni-tude limits for quasar fields using only these observations willbe made from published data on a field-by-field basis below.Finally, (5), Johnson et al. (2013), performed detailed spec-troscopic follow-up observations of the galaxies in the fieldof Q0405 - L > . L ∗ galaxiesat impact parameters less than 100 kpc and a 75% complete-ness level for L > . L ∗ galaxies at impact parameters lessthan 500 kpc.Galaxy and absorber data can be found in Tables 2 and 3,respectively. Galaxy image footprints and analyzed spectrafor Ly α , Ly β , and O VI λλ , The Field Toward Q0405 - This field was first spectroscopically surveyed byEllingson & Yee (1994) and has had follow-up observa-tions published in Johnson et al. (2013). Nearly all ofthe redshifts measured by Ellingson & Yee (1994) have been revised. Some galaxies associated with absorbers inChen et al. (2001), which uses the Ellingson redshifts, havechanged significantly. The galaxy measured at z gal = 0 . z gal = 0 . z gal = 0 . z gal = 0 . z gal = 0 . z gal = 0 . z QSO = 0 .
57 that arenot considered in this study due to their likely physicalconnection to the quasar. In addition, there are four galaxypairs whose absorption cannot be disentangled (the first at z gal = 0 . z gal = 0 . z gal = 0 . z gal = 0 . z gal = 0 . z gal = 0 . z gal = 0 . z gal = 0 . z gal =0 . , . . The Field Toward Q0454 - The galaxy identifications come from Chen et al. (1998),but the spectroscopic survey of the field remains unpublished.From Chen et al. (1998), we estimate that the survey limit-ing magnitude is m r ∼ .
8. There are two galaxies in thefield within ∼
300 km s - of each other ( z gal = 0 . z gal = 0 . α , Ly β , and O VI ).We examine only the galaxy-absorber pair for the galaxy at z gal = 0 . The Field Toward Q1001 + This field was spectroscopically surveyed by Lanzetta et al.(1995). There are only a few bright galaxies near the quasar,allowing straight-forward identification of galaxy-absorberpairs. We note that the galaxy at z gal = 0 . z gal = 0 . . The Field Toward Q1136 - As with the Q0454 - m r ∼ .
3. There are three galaxies clustered around z gal ≃ .
36 which are excluded from our study due also to a lack ofUV spectral coverage of Ly α , Ly β , and O VI absorption inthe COS spectrum.We examine galaxy-absorber pairs at redshifts z gal = 0 . . The Field Toward Q1216 + The galaxy identifications come from Chen et al. (2001),but the detailed spectroscopic survey of the field remains un-published. From Chen et al. (2001), we estimate that the sur-vey limiting magnitude is m r ∼ .
6. There is only one galaxyidentified that has a redshift lower than that of the quasar withthe required spectral coverage in the COS spectrum.We examine the only galaxy-absorber pair at redshift z gal =0 . IFFERENTIAL K INEMATICS Figure A1.
Same as Figure 1, but for the galaxy at z gal = 0 . - z gal = 0 . z gal = 0 . + Figure A2.
Same as Figure 1, but for the galaxies at z gal = 0 . z gal = 0 . - z gal = 0 . + ATHES ET AL . Figure A3.
Same as Figure 1, but for the galaxies at z gal = 0 . z gal = 0 . + z gal = 0 . + Figure A4.
Same as Figure 1, but for the galaxies at z gal = 0 . z gal = 0 . + IFFERENTIAL K INEMATICS
The Field Toward Q1259 + The galaxy identifications come from Chen et al. (2001),but the detailed spectroscopic survey of the field remains un-published. From Chen et al. (2001), we estimate that the sur-vey limiting magnitude is m r ∼ . z gal = 0 . . The Field Toward Q1317 + This field was spectroscopically surveyed by Le Brun et al.(1996). They report two pairs of galaxies in this field ( z gal =0 . . z gal = 0 . . z ≃ .
54 galaxies falls in the gaps inthe COS spectrum. Chen et al. (2001) also studied this field.The most recent work is from Churchill et al. (2012), whostudied the z gal = 0 . z gal = 0 . z gal = 0 . z gal = 0 . The Field Toward Q1704 + This field was spectroscopically surveyed by bothLanzetta et al. (1995) and Le Brun et al. (1996).Ambiguity in this field exists for the galaxies identified at z gal = 0 . z gal = 0 . z gal = 0 . D = 260 kpc).Chen et al. (2001) do not identify an absorber with the z gal =0 . D ∼
530 kpc, has a red-shift nearly coincident with absorber at z abs = 0 . z gal = 0 . z gal = 0 . . B. COLUMN DENSITIESFor this work, we present H I column densities using onlythe Ly α and Ly β transitions. For roughly half of our samplethe Ly γ is also available for fitting. As such, requiring Ly γ for the fits would significantly reduce our sample size.To ensure that the Voigt profile fits using Ly α and Ly β onlyare not systematically skewed relative to fits using Ly α , Ly β ,and Ly γ , we compared the fits with and without Ly γ for thesubsample that has Ly γ coverage.In Figure B1, we present the H I column densities derivedfrom Ly α and Ly β only fits and Ly α , Ly β , and Ly γ fits.The resulting column densities are virtually identical for non-saturated lines. Even in the saturated higher column densitylines, the resulting column densities are highly consistent witha one-to-one correlation. We thus have validated that omittingthe Ly γ transition provides no skew in the resulting H I col-umn densities. C. DEBLENDINGIn two cases, we identified absorption components blendedwith other absorption features from a different redshift. Here,we illustrate our deblending technique.
Figure B1.
Voigt profile column density results for log N (H I ). On the x-axisis the resultant column density measured using only Ly α and Ly β . On they-axis is the result using Ly α , Ly β , and Ly γ . The dotted line shows a one-to-one correlation. Including the Ly γ with the fit has very little impact on themeasured H I column density. The first case occurs in the Ly β line associated with the z gal = 0 . - β isblended with N V λ z abs = 0 . V λ V λ V λ - z gal = 0 . VI λλ , VI λ β at z abs = 0 . VI λ ǫ at z abs = 0 . β and Ly ǫ ).REFERENCES Bahcall, J. N., Bergeron, J., Boksenberg, A., et al. 1993, ApJS, 87, 1Behroozi, P. S., Conroy, C., & Wechsler, R. H. 2010, ApJ, 717, 379Behroozi, P. S., Marchesini, D., Wechsler, R. H., et al. 2013, ApJ, 777, L10Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393Birnboim, Y., & Dekel, A. 2003, MNRAS, 345, 349Bordoloi, R., Lilly, S. J., Knobel, C., et al. 2011, ApJ, 743, 10Bouché, N., Hohensee, W., Vargas, R., et al. 2012, MNRAS, 426, 801Brown, T. et al. 2002,
HST
STIS Data Handbook, version 4.0, ed. B.Mobasher, (Baltimore: STScI)Bryan, G. L., & Norman, M. L. 1998, ApJ, 495, 80Bullock, J. S., Kolatt, T. S., Sigad, Y., et al. 2001, MNRAS, 321, 559Chelouche, D., & Bowen, D. V. 2010, ApJ, 722, 1821Chen, H.-W., Lanzetta, K. M., Webb, J. K., & Barcons, X. 1998, ApJ, 498,77—. 2001, ApJ, 559, 654Churchill, C. W., Kacprzak, G. G., Steidel, C. C., et al. 2012, ApJ, 760, 68Churchill, C. W., Mellon, R. R., Charlton, J. C., et al. 2000, ApJS, 130, 91Churchill, C. W., Nielsen, N. M., Kacprzak, G. G., & Trujillo-Gomez, S.2013a, ApJ, 763, L42Churchill, C. W., Rigby, J. R., Charlton, J. C., & Vogt, S. S. 1999, ApJS,120, 51Churchill, C. W., Trujillo-Gomez, S., Nielsen, N. M., & Kacprzak, G. G.2013b, ArXiv e-printsChurchill, C. W., & Vogt, S. S. 2001, AJ, 122, 679
ATHES ET AL . Figure C1.
Deblending of Ly β for absorption associated with the galaxy at z gal = 0 . - β line is blended with N V λ z abs = 0 . IFFERENTIAL K INEMATICS Figure C2.
Deblending of O VI λλ , z gal = 0 . - VI λ β at z abs = 0 . VI λ ǫ at z abs = 0 ..