Quasars Probing Quasars IX. The Kinematics of the Circumgalactic Medium Surrounding z ~ 2 Quasars
DDraft version March 23, 2018
Preprint typeset using L A TEX style emulateapj v. 12/16/11
QUASARS PROBING QUASARS IX. THE KINEMATICS OF THE CIRCUMGALACTIC MEDIUMSURROUNDING Z ∼ Marie Wingyee Lau , J. Xavier Prochaska , Joseph F. Hennawi Draft version March 23, 2018
ABSTRACTWe examine the kinematics of the gas in the environments of galaxies hosting quasars at z ∼ (cid:46)
300 km s − and average offsets from the systemic redshift (cid:46) |
100 km s − | . Westack the background quasar spectra at the foreground quasar’s systemic redshift to study the meanabsorption in C II , C IV , and Mg II . We find that the mean absorptions exhibit large velocity widths σ v ≈
300 km s − . Further, the mean absorptions appear to be asymmetric about the systemic redshifts.The mean absorption centroids exhibit small redshift relative to the systemic δv ≈ +200 km s − , withlarge intrinsic scatter in the centroid velocities of the individual absorption systems. We find theobserved widths are consistent with gas in gravitational motion and Hubble flow. However, while theobservation of large widths alone does not require galactic-scale outflows, the observed offsets suggestthat the gas is on average outflowing from the galaxy. The observed offsets also suggest that theionizing radiation from the foreground quasars is anisotropic and/or intermittent. Keywords: galaxies: clusters: intracluster medium – galaxies: formation – galaxies: halos – intergalac-tic medium – quasars: absorption lines – quasars: general INTRODUCTION
Galaxy formation and evolution are driven by the flowsof gas into and out of their interstellar medium. Cur-rent theories demand that star-forming galaxies maintainthese flows. Gas accretes, cools, and adds to the fuel sup-ply, while star formation feedback heats gas, blows it outof galaxies, and regulates star formation (for a review seeSomerville & Dav´e 2015).Direct observations of galactic flows are difficult to ac-quire. Detecting the presence of the gas is itself challeng-ing. Either the gas mass is too small, or the gas density istoo low for the detection of line-emission, e.g. 21 cm, Ly α ,or H α from H I . Resolving the kinematics and establish-ing the mass flux pose an even greater challenge. Thesechallenges are accentuated for distant, young galaxies,where flows of gas are predicted to prevail (Kereˇs et al.2009; Fumagalli et al. 2011). Therefore, with rare ex-ceptions, (e.g., Cantalupo et al. 2014; Hennawi et al.2015), the community has relied on absorption-line spec-troscopy to detect and characterize the gas surroundinggalaxies (e.g., Bergeron & Boisse 1991; Steidel et al. 2010;Prochaska et al. 2011; Tumlinson et al. 2013). In a pre-vious paper of the Qussars Probing Quasars series (Lauet al. 2016, , hereafter QPQ8), we measured the velocityfield for C II IV II systems and 10 C IV systems, we measuredthe velocity interval that encompasses 90% of the totaloptical depth, ∆ v , and the 1 σ dispersion relative to Email: [email protected] Department of Astronomy and Astrophysics, UCO/Lick Ob-servatory, University of California, 1156 High Street, Santa Cruz,CA 95064 Department of Physics, Broida Hall, University of California,Santa Barbara, CA 93106-9530 the profile centroid, σ v . The median ∆ v is 555 km s − for C II − for C IV σ v is 249 km s − for C II − forC IV δv , are often positive, with thesign convention that positive velocities indicate a redshiftfrom the systemic.With absorption-line spectroscopy of backgroundsightlines, other researchers also have had success in char-acterizing the flows of gas around galaxies. Rakic et al.(2012) found a net large-scale inflow around star-forminggalaxies, or a Kaiser effect for gas on 1–2 Mpc scales. Hoet al. (2017) found gas spiraling inward near the diskplane of star-forming galaxies on <
100 kpc scales. John-son et al. (2015) studied the CGM surrounding z ∼ v of the Mg II ab-sorption is less than the average of the QPQ8 C II ab-sorption. In Muzahid et al. (2015), an absorber is foundwith ∆ v smaller than 555 km s − in O VI and N V ,and still smaller for other ions. The Zahedy et al. (2016)sample likewise has average ∆ v smaller than that ofQPQ8.A significant limitation of absorption-line analysis oftransverse sightlines, especially regarding galactic-scaleflows, is the inherent symmetry of the experiment. One a r X i v : . [ a s t r o - ph . GA ] M a r generally lacks any constraint on the distance of the gasalong the sightline. Positive or negative velocities withrespect to the galaxy may be interpreted as gas flow-ing either toward or away from the system. “Down-the-barrel” observations break this symmetry, and havegenerally provided evidence for flows away from galaxies(Rupke et al. 2005; Martin 2005; Weiner et al. 2009; Ru-bin et al. 2014). However, these data are frequently atlow spectral resolution which limits one’s sensitivity toinflowing gas.In this paper (hereafter QPQ9), we examine the flowsof gas in the environments of massive galaxies host-ing quasars. Our approach leverages a large datasetof quasar pairs (Hennawi et al. 2006, hereafter QPQ1)to use the standard techniques of absorption-line spec-troscopy with background quasars. These quasar pairshave angular separations that correspond to less than300 kpc projected separation at the foreground quasar’sredshift. Our previous publications from these quasarpairs have established that these galaxies are surroundedby a massive, cool, and enriched CGM (QPQ5, QPQ6,QPQ7: Prochaska et al. 2013b,a, 2014). We have col-lected a sample of 148 background spectra that are pairedwith foreground quasars with precisely measured red-shifts. Among the sightlines in the QPQ9 sample, 13have spectral resolution R > I absorption on scales larger thanthe CGM, and gave similar arguments on anisotropy orintermittence.We adopt a ΛCDM cosmology with Ω M = 0 . , Ω Λ =0 .
74, and H = 70 km s − Mpc − . Distances are properunless otherwise stated. When referring to comovingdistances we include explicitly an h − term and followmodern convention of scaling to the Hubble constant h = H / (70 km s − Mpc − ). THE EXPERIMENT
Table 1
Properties of the Projected Quasar Pairs in the QPQ9 sampleForeground Quasar z fg Line for z a fg Background Quasar z bg BG Quasar Instrument R ⊥ (kpc) g UV J003423.05 − − − − − − − − − − − − α J083757.13+383722.4 2.251 LRIS 89 8609J083854.52+462124.4 1.7596 MgII J083852.94+462137.6 2.163 BOSS 184 673J084158.47+392120.0 2.0414 [OIII] J084159.26+392139.0 2.214 LRIS 183 1514J084511.89+464135.5 1.6295 MgII J084509.64+464113.0 1.898 BOSS 283 135
Table 1 — Continued
Foreground Quasar z fg Line for z a fg Background Quasar z bg BG Quasar Instrument R ⊥ (kpc) g UV J085019.43+475538.5 1.8164 MgII J085021.17+475516.0 1.891 BOSS 249 392J085151.38+522901.6 1.9738 MgII J085154.53+522910.6 2.031 BOSS 262 265J085249.45+471423.1 1.6468 MgII J085248.55+471419.3 1.688 BOSS 87 7078J085358.36 − − − α J101003.47+403754.9 2.505 BOSS 191 5421J101323.89+033016.0 1.9401 MgII J101322.23+033009.1 2.273 BOSS 219 426J101753.38+622653.4 1.6528 MgII J101750.44+622648.2 2.738 BOSS 184 922J101947.11+494835.8 1.6224 MgII J1019470+494849.1 1.652 LRIS 117 881J102007.23+611955.0 1.7909 MgII J102010.05+611950.3 2.387 BOSS 180 1817J102259.33+491125.8 1.9757 MgII J102259.97+491151.7 2.469 BOSS 231 212J102821.26+240121.8 1.8709 MgII J102822.18+240057.4 2.414 BOSS 240 834J103443.62+085702.0 1.6395 MgII J103442.26+085645.7 2.766 BOSS 233 700J103628.12+501157.9 2.0097 MgII J103630.52+501219.8 2.228 BOSS 271 1198J103857.37+502707.0 3.1325 [OIII] J103900.01+502652.8 3.236 ESI 233 3567J103946.92+454716.0 1.8644 MgII J103945.58+454707.4 2.456 BOSS 148 675J104244.84+650002.7 1.9876 MgII J104245.14+645936.7 2.124 BOSS 227 703J104435.62+313950.7 1.7062 MgII J104434.76+313957.7 2.377 BOSS 115 1873J104955.01+231358.2 1.8439 MgII J104953.97+231401.3 2.171 BOSS 129 1813J105221.77+555253.5 1.9989 MgII J105218.36+555311.3 2.278 BOSS 293 846J105246.45+641832.2 1.6429 MgII J105251.42+641838.5 2.936 BOSS 286 127J111339.86+330604.8 1.8913 MgII J111337.84+330553.3 2.413 BOSS 243 853J111850.44+402553.8 1.9257 MgII J111851.45+402557.6 2.317 BOSS 106 1946J112858.89+644440.4 1.6561 MgII J112854.14+644427.4 2.217 BOSS 289 149J113852.65+632934.0 1.8855 MgII J113851.73+632955.6 2.625 BOSS 196 1912J114435.54+095921.7 2.9734 [OIII] J114436.65+095904.9 3.16 MIKE-Red 189 2914J114439.51+454115.8 1.687 MgII J114442.48+454111.3 2.592 BOSS 275 78J114546.54+032236.7 1.7664 MgII J114546.22+032251.9 2.011 MagE 139 779J115253.09+150706.5 1.7883 MgII J115254.97+150707.8 3.349 BOSS 237 622J115457.16+471149.3 1.6819 MgII J115458.69+471209.9 1.947 SDSS 226 60J115502.45+213235.5 1.9551 MgII J115504.25+213254.0 2.695 BOSS 277 397J115529.49+463413.1 1.6491 MgII J115528.75+463442.9 2.329 BOSS 270 326J115533.62+393359.2 1.6118 MgII J115531.32+393415.4 2.555 BOSS 272 742J120224.68+074800.3 1.6613 MgII J120226.48+074739.7 2.767 BOSS 296 584J120417.47+022104.7 2.436 [OIII] J120416.69+022110.0 2.532 HIRES 112 2710J120856.94+073741.2 2.1708 MgII J120857.16+073727.3 2.616 MagE 123 3853J121159.88+324009.0 1.978 MgII J121201.69+324013.3 2.273 BOSS 209 1046J121344.28+471958.7 1.8371 MgII J121343.01+471931.0 3.275 BOSS 260 491J1215590+571616.6 1.93 [OIII] J121558.82+571555.5 1.964 BOSS 184 614J121657.82+152706.6 1.9473 MgII J121657.00+152712.7 2.318 BOSS 116 4225J122514.29+570942.3 1.8953 MgII J122517.89+570943.7 2.224 BOSS 255 330J123143.01+002846.3 3.2015 [OIII] J123141.73+002913.9 3.308 GMOS-S 271 1490J124632.33+234531.2 1.9937 MgII J124632.19+234509.5 2.573 BOSS 188 1886J124846.05+405758.2 1.8265 MgII J124846.97+405820.9 2.463 BOSS 219 897J130124.74+475909.6 2.194 H α J130125.67+475930.8 2.765 SDSS 199 4932J130605.19+615823.7 2.1089 H α J130603.55+615835.2 2.175 LRIS 141 1761J130714.79+463536.6 1.6226 MgII J130716.07+463511.2 2.248 BOSS 251 137J131341.32+454654.6 1.6878 MgII J131342.78+454658.2 2.241 BOSS 139 1098J132514.97+540930.6 2.0507 MgII J132511.07+540927.0 3.235 BOSS 298 514J133026.12+411432.0 2.0645 MgII J133023.67+411445.9 2.217 BOSS 271 384
Table 1 — Continued
Foreground Quasar z fg Line for z a fg Background Quasar z bg BG Quasar Instrument R ⊥ (kpc) g UV J133924.02+462808.2 1.8539 MgII J133922.31+462749.2 3.391 BOSS 226 940J134650.08+195235.2 2.0697 MgII J134648.19+195253.1 2.523 BOSS 278 884J135306.35+113804.7 1.6315 MgII J135307.90+113805.5 2.431 BOSS 213 8963J135849.71+273806.9 1.9008 MgII J135849.54+273756.9 2.127 LRIS 89 1765J140208.01+470111.1 1.9161 [OIII] J140209.52+470117.8 2.269 BOSS 140 1437J140918.01+522552.4 1.8808 MgII J140916.98+522535.3 2.109 SDSS 170 583J141337.18+271517.1 1.6905 MgII J141337.96+271511.0 1.965 BOSS 105 609J142003.67+022726.7 3.617 [OIII] J142004.12+022708.8 4.191 ESI 144 2291J142054.42+160333.3 2.0221 [OIII] J142054.92+160342.9 2.057 MagE 104 7811J142215.57+465230.7 1.748 MgII J142214.63+465254.6 2.338 BOSS 225 639J142758.89 − − − a The emission-line analyzed for measuring z fg . The goal of our experiment is to measure the averagevelocity fields of the absorption from C + , C , and Mg + ions associated with the CGM of the galaxies hosting z ∼ , we analyze a subset of systemsthat pass within transverse separation R ⊥ <
300 kpcfrom a foreground quasar with z fg > .
6. We restrictthe sample to foreground quasars with redshift measuredfrom Mg II III ] 5007, or H α emission, giving aprecision of 300 km s − or better and an average offsetfrom the systemic redshift of |
100 km s − | or less. Ac-cording to Shen et al. (2016a), the [O III ] emission-lineredshifts have the smallest scatter (intrinsic scatter andmeasurement error combined) of 68 km s − about thesystemic redshift, and we analyze the sub-sample with[O III ] redshifts separately. The [O
III ] line has an av-erage blueshift of 48 km s − about the systemic redshift,which has been added when we compute the redshift ofthe line. The scatter and average offset of [O III ] red-shifts reported by Shen et al. (2016a) is consistent withthe numbers reported by Boroson (2005) using a largerbut lower redshift sample. Systemic redshifts measuredfrom Mg II have a precision of 226 km s − according toShen et al. (2016a), and we have taken into account their reported median blueshift of 57 km s − of Mg II from thesystemic. We note that Richards et al. (2002) reported amedian redshift of 97 km s − of Mg II from [O III ] using alarger but lower redshift sample. In QPQ8, we quantifiedthe precision of H α to be 300 km s − and the median off-set from the systemic redshift is close to zero, consistentwith the velocity shifts measured by Shen et al. (2011).Although H β is a narrow emission-line, we do not con-sider its redshift sufficiently reliable for use as systemicredshift. H β redshifts have a large scatter about thesystemic ≈
400 km s − , and a large average offset aboutthe systemic ≈
100 km s − (Shen et al. 2016a, QPQ8).Our line-centering algorithm calculates the mode of a linegiven by 3 × median − × mean, applied to the upper60% of the emission, while Shen et al. (2016a) calculatesthe peak of a line. We expect that our line-centeringalgorithm gives emission redshifts very comparable tothe Shen et al. (2016a) algorithm, however. Shen et al.(2016a) states that the difference between the peak andthe centroid of an emission-line is not significant exceptfor the broad line H β , which we do not use in redshiftmeasurements. To quantify the above, we further obtainindividual measurements of centroids and peaks in theShen et al. (2016a) sample through private communica-tion. We found there is essentially no difference betweenusing the centroid versus using the peak for [O III ] emis-sion redshifts, and there is on average 50 km s − differ-ence for Mg II . We may expect the difference between z f g CII1.52.02.53.03.5 z f g CIV0 100 200 300 R (kpc)2.02.53.0 z f g MgII 2.02.53.03.54.04.5 l og g UV Figure 1.
These panels summarize properties of the QPQ9dataset. The QPQ survey selects quasar pairs of R ⊥ <
300 kpc and z fg > .
6. Assuming that the foreground quasars emit isotropicallyand at a distance equal to the impact parameter, the enhancementin the UV flux relative to the extragalactic UV background, g UV ,can be estimated. Large symbols correspond to foreground quasarswith the most precise redshift measurement from [O III ] 5007, whilesmall symbols correspond to z fg measurements from Mg II α , or H β emission. The top panel shows quasar pairs with cov-erage of C II z fg in the background quasar spectra. Themiddle panel shows pairs with coverage of C IV II the mode and the peak is even smaller. Hence, we arguethat the average systemic bias corrections measured inShen et al. (2016a) may be self-consistently applied toour measured emission-line redshifts to obtain systemicredshifts.We further add to the QPQ dataset with quasar pairsselected from the public dataset of igmspec (Prochaska2017), which includes the spectra from the quasar cat-alogs based upon the Sloan Digital Sky Survey SeventhData Release (Schneider et al. 2010) and the TwelfthData Release (Pˆaris et al. 2017). We only select pairswith z fg measurable using a robust Mg II > − , to ex- https://github.com/specdb/igmspec clude physically associated binary quasars. The cut onvelocity difference is motivated by the typical redshiftuncertainty of ≈
500 km s − of the background quasars.In QPQ8, it was required that the observed wavelengthsof the metal ion transitions fall outside the Ly α forestof the background quasar. In this paper, we exclude asmall window around the Ly α emission, in additional tothe Ly α forest, from analysis. For stacked profile analy-sis, a good estimate of the continuum level is necessary.In QPQ8 we found that absorption associated to theforeground quasar occurs within ± − around z fg . Therefore, it is desirable to keep a ≈ ± − window relatively free of contamination from Ly α for-est. Taking into account the redshift uncertainties, wedecide that at least one transition among C II IV II z fg must lie redward of(1215 . × (1 + z bg ) ˚A, for a pair to be includedin the analysis.Furthermore, we include only those spectra with av-erage signal-to-noise ratio (S/N) exceeding 5.5 per rest-frame ˚A in a ± − window centered on the ob-served wavelengths of the metal ion transitions. This cri-terion is a compromise between maximizing sample sizeversus maintaining good data quality on the individualsightlines. We find that S/N > . ± − around a consid-ered metal ion transition does not overlap with strongatmospheric O bands. The O A- and B-band span7595–7680 ˚A and 6868–6926 ˚A respectively.Table 1 lists the full QPQ9 sample. In Table 2, we firstlist the sample size, the median z fg , and the median R ⊥ ofthe quasar pairs that survive the above selection criteriafor C II IV II z fg measured from [O III ]. ANALYSIS
Stacked Profiles
We create composite spectra that average over the in-trinsic scatter in quasar environments, continuum place-ment errors, and redshift errors. The individual spectraof background quasars are shifted to the rest-frame of theforeground quasars at the transitions of interest. Eachspectrum has been linearly interpolated onto a fixed ve-locity grid centered at z fg with bins of 100 km s − . Fora velocity bin of this size, it is unnecessary to smooththe data to a common spectral resolution. The individ-ual spectra are then combined with a mean or a medianstatistic. Spectra with broad absorption-line systems andmini-broad absorption-line systems are excluded. Badpixels in the individual spectra have been masked be-fore generating the composites. Since each quasar pairgives an independent probe of the CGM, each pair hasan equal weighting in the stacked profiles, i.e. we do notweight the spectra by the measured S/N near the metalion transitions. Scatter in the stacked spectra is domi-nated by randomness in the CGM rather than scatter inthe flux of individual observations. The mean statisticof the individual spectra yields a good estimate of the N o r m a li ze d F l ux CII, mean CII, median0.850.900.951.00 N o r m a li ze d F l ux CIV, mean CIV, median2000 1000 0 1000 2000Relative Velocity (km/s)0.850.900.951.00 N o r m a li ze d F l ux MgII, mean 2000 1000 0 1000 2000Relative Velocity (km/s)MgII, median
Figure 2.
Mean and median absorption centered at C II IV II − relative to z fg , and are shown inthick, blue. For the doublets, a second Gaussian with a fixed meanseparation and a tied standard deviation is included in the model-ing. The absorptions frequently exhibit large velocity widths. Theblue dashed lines mark the centroids, which show small positivevelocity offsets from z fg . The 1 σ modeling error for the disper-sion and the centroid of the C II mean stack are 27 km s − and25 km s − respectively, which arealso the typical modeling errors of the other stacks. average absorption and preserves equivalent width. Themedian statistic is less sensitive to outliers. However, themedian opacity at any velocity channel is rather small,since the discrete absorbers are spread throughout theentire velocity window. A pixel is not affected unlessthere are more than 50% of quasar pairs with absorptionat its velocity. The analysis on the median velocity fieldis thus subject to larger uncertainty. In the followingwe present stacked spectra using both the mean and themedian statistic.In Figure 2, we present mean and median stacks ofC II IV II II mean stack. C IV and Mg II are doublet transitionsand it is more challenging to analyze their kinematics.Two results are evident in Figure 2: (i) the mean C II stack exhibits excess absorption spanning a large velocitywidth; (ii) the mean absorption is likely skewed towardpositive velocities.Visually, there are two absorption components. Onecomponent is the uniform depression in the continuumlevel in the stack resulting from absorbers unassociatedwith the foreground quasars. The other componentcomes from absorbers associated with the foregroundquasars and distribute around their systemic velocities.To model the absorption, we introduce a Gaussian profilewhile allowing a constant “pseudo-continuum” to vary. We perform χ minimization with each channel givenequal weight. From the best-fit to the data, we measurethe 1 σ dispersion of the stack to be 293 km s − and thecentroid of the C II stack to be +232 km s − . The dis-persion suggests extreme kinematics, while the centroidsuggests an asymmetry that contradicts the standard ex-pectation. The median stack, on the other hand, showsweaker absorption, and the Gaussian model has a moreuncertain dispersion and a centroid with smaller offset.To test whether the dispersion in the average ab-sorption associated with the foreground quasars is well-captured by a Gaussian model, we also calculate thedispersion separately using the standard deviation for-mula. Specifically, we apply the standard deviation for-mula to a ± − window surrounding the absorp-tion centroid, while fitting a continuum to the rest of the ± − window for stacking. The dispersion mea-sured with the above method is essentially the same asthat found by fitting a Gaussian. One may also spec-ulate on the existence of a broader absorption compo-nent hidden in the depressed continuum. We try re-placing the constant absorption component with a broadGaussian component in our modeling. We find thatthis second Gaussian component has an amplitude of0 . ± . σ disper-sion of 3614 ± − , i.e. wider than the entire ve-locity window for stacking. This weak and very broadcomponent has no effect on the width and the centroid ofthe narrow, associated Gaussian component. We there-fore consider that one single Gaussian component is agood description of the average absorption associatedwith the foreground quasars. Moreover, narrow asso-ciated absorbers of background quasars will not preferthe systemic velocities of the foreground quasars, andhence will not affect the average absorption measuredfor the foreground quasars. Their contribution to theaverage absorption should result in a tilt in the pseudo-continuum, which is insignificant.We also create mean and median stack for the sub-sample with [O III ] redshifts and model the absorptionwith Gaussian best-fit. The C II mean stack for this sub-sample has a dispersion of 330 km s − , and a centroid at+235 km s − , consistent with the full sample.To model the mean and median absorption ofC IV II − and 769 km s − respectively), and tie the dis-persion of the two lines in a doublet. We allow the dou-blet ratio to vary from 2:1 to 1:1. The centroid is degen-erate with the doublet ratio in our modeling. This degen-eracy is more pronounced for C IV , although we find thata double ratio closer to 2:1 better captures the absorptionat the central few pixels. We state that the modeling ofthe doublets is presented for consistency check, but theanalysis focuses on C II . The modeling results show thatthe velocity fields of C IV and Mg II are consistent withC II , i.e. large dispersion and centroid is skewed towardpositive velocities.The above analyses are summarized in Table 2. TheGaussian models normalized to pseudo-continuum areoverplotted on the data stacks in Figure 2. Table 2
Summary of the Data and AnalysisMeasure C II IV II z fg R ⊥
157 191 1841 σ dispersion of mean stack (km s − ) 293 ±
87 303 ±
44 295 ± − ) +232 ±
98 +11 ±
63 +215 ± .
97 0 .
98 0 . σ dispersion of median stack (km s − ) 137 ±
569 218 ±
71 276 ± − ) +119 ±
240 +189 ±
81 +121 ± .
99 0 .
99 0 . z fg R ⊥
183 122 1121 σ dispersion of mean stack (km s − ) 330 ±
103 255 ±
41 235 ± − ) +235 ±
118 +130 ±
77 +210 ± .
98 0 .
98 0 . σ dispersion of median stack (km s − ) 103 ±
236 211 ±
71 175 ± − ) +256 ±
191 +210 ±
111 +244 ± .
99 0 .
98 0 . N o r m a li ze d F l ux CII, mean
Figure 3.
The same C II mean stack shown in the first panel ofFigure 2 is shown in thin, black. We overplot in dashed, limegreenan absorption profile with σ v = 554 km s − , which is larger thanthe observed width by three times the standard deviation in thebootstrap analysis. Motions that produce a velocity width largerthan this can be ruled out. We overplot in dot dashed, limegreenan absorption profile with σ v = 214 km s − , which is smaller thanthe observed width by three times the modeling error. Unless grav-itational and Hubble flows together with redshift error broadeningproduce a velocity width smaller than this, extra dynamical pro-cesses (e.g. outflows) will not be required to explain the observedwidth. In thick, green, we overplot the Gaussian absorption modelof the Monte Carlo simulations generated from a purely clusteringargument. The model from clustering analysis is multiplied to thepseudo-continuum level of the stack of the observational data, andbroadened by the mean redshift error in the data. While this modelhas a dispersion within modeling error of the dispersion in the data.the centroid of the stack of the data appears to be redshifted fromthe systemic. A model with only gravitational motions and Hubbleflows cannot explain this putative asymmetry. Interpretation of the large velocity fields
Under the assumption that the intrinsic dispersion andthe redshift uncertainty add in quadrature to give theobserved width, we solve for the intrinsic dispersion inthe C II mean stack. For the full QPQ9 sample, withthe mean σ fullerror( z ) = 189 km s − , we recover σ fullintrinsic =224 km s − . For the sub-sample with [O III ] redshifts, we M halo /M )0.20.30.40.50.6 W C II ( Å ) + Figure 4.
Probability distributions of the parameters W CII and M halo . The plot shows the degeneracy between W CII and M halo inrecovering the intrinsic width of the absorption profile. We markcontours for points that produce an absorption profile of width thatis 1, 2, and 3 times the modeling error away from the observed in-trinsic width. The intrinsic velocity width corresponding to typicalQPQ halo mass is marked with a plus sign, and is contained withinthe 2 σ contour. recover σ [OIII]intrinsic = 323 km s − .To assess the statistical significance of the observed dis-persion, we perform a bootstrap analysis by randomly re-sampling from the full sample 10000 times. We introducea Gaussian absorption profile to model each bootstrap re-alization. We find that a width of σ v >
554 km s − wouldbe larger than the observed by three times the scatter inthe bootstrap realizations (overplotted in Figure 3). Inother words, taking into account the redshift errors, mo-tions in addition to gravitational and Hubble flows thatwill produce an intrinsic dispersion σ v >
521 km s − canbe ruled out.On the contrary, σ rms <
70 km s − together with Hub-ble velocities and broadening by redshift errors will resultin a velocity width that is more than three times the mod-eling error away from the observed width (overplotted inFigure 3). This implies that, unless the characteristic M halo < . M (cid:12) , additional dynamical processes arenot required to explain the observed width.Eftekharzadeh et al. (2015) measured the clustering ofquasars in the range 2 . < z < . . < z < .
2. They estimated thatthese quasars are hosted by dark matter halos withmass M halo = 10 . M (cid:12) and M halo = 10 . M (cid:12) re-spectively. If dark matter halos hosting QPQ9 quasarshave a characteristic mass 10 . M (cid:12) and follow an NFWprofile (Navarro et al. 1997) with concentration param-eter c = 4, at z ≈ − . Tormen et al. (1997) found that the max-imum circular velocity is ≈ . σ rms . Hence, the average line-of-sight rms velocity typ-ical of QPQ9 halos is σ rms = 246 km s − . In QPQ8, weestimated the probability of intercepting a random op-tically thick absorber is 4%, and clustering would onlyincrease that to 24%. Although motions due to Hub-ble flows do not dominate, they nevertheless contributeto the observed dispersion. We can investigate whethergravitational motions and Hubble flows are sufficient toreproduce the dispersion in the data, using Monte Carlomethods to simulate the absorption signals.Since C II systems arise in optically thick absorbers,we may adopt the clustering analysis results of QPQ6.In the absence of clustering, the expected number of ab-sorbers per unit redshift interval for Lyman limit sys-tems, super Lyman limit systems, and damped Ly α systems are respectively (cid:96) LLSIGM ( z ) ≈ . z ) / (1 +2 . . , (cid:96) SLLSIGM ( z ) ≈ . z ) / (1 + 2 . . , and (cid:96) DLAIGM ( z ) ≈ . z ) / (1 + 2 . . . The quasar-absorber correlation functions for Lyman limit systems,super Lyman limit systems, and damped Ly α sys-tems are respectively ξ LLSQA ( r ) = ( r/ (12 . h − Mpc)) − . , ξ SLLSQA ( r ) = ( r/ (14 . h − Mpc)) − . , and ξ DLAQA ( r ) =( r/ (3 . h − Mpc)) − . . For each quasar pair, we calcu-late the expected number of optically thick absorberswithin ± − at a distance R ⊥ from the fore-ground quasar and at z fg . Then we generate 1000 mocksightlines. The number of absorbers for each mock spec-trum is randomly selected from a Poisson distributionwith mean equal to the expected number calculated asabove. The absorbers are randomly assigned Hubble ve-locities, with a probability distribution according to thequasar-absorber correlation functions. The absorbers arerandomly assigned additional peculiar velocities drawnfrom a normal distribution with mean equal to 0 km s − and scatter equal to σ rms . For each absorber, we assumea rest equivalent width for C II W CII and a Gaussian ab-sorption profile. We repeat the above procedure for all 40quasar pairs, and create a mean stack of the 40000 mockspectra generated. We fit a Gaussian absorption profilemultiplied to a constant continuum level to model thestack of mock spectra. We adjust the W CII adopted forthe absorbers until the amplitude of the best-fit Gaus-sian of the stack of mock spectra matches the ampli-tude of the stack of the observational data. We find that W CII ≈ . W CII or line profile for one absorber.In Figure 3, we show a comparison of the observa-tional data stack and the Gaussian absorption model ofthe Monte Carlo simulations. n the figure, the Gaus-sian absorption model is broadened by the mean red-shift error by adding it in quadrature to the dispersion in the model. The resulting stack of mock spectra hasa 1 σ dispersion of 282 km s − , about 2 times the model-ing error away from the intrinsic dispersion in the C II mean stack for the full sample, and about 2 times themodeling error away from the intrinsic dispersion in thestack of the sub-sample with [O III ] redshifts as well.One may also consider whether absorbers in Hubble flowand cosmological distances show peculiar velocities thatare typical for quasar-mass halos, and adopt a different σ rms accordingly. The best-studied coeval, more typ-ical star-forming galaxies are the Lyman-break galax-ies. Their clustering strength implies a characteristichalo mass of M halo ≈ . M/ M (cid:12) (Bielby et al. 2013;Malkan et al. 2017), corresponding to σ rms = 114 km s − .If we adopt this smaller σ rms insead,for absorbers outsidethe loosely defined quasar CGM boundary, at (cid:38)
300 kpc,the stack of mock spectra will have a 1 σ dispersion of247 km s − . This is again within modeling errors of theobserved intrinsic dispersion. We also test for the sen-sitivity of this measured dispersion to the correlationfunctions adopted. The QPQ6 clustering analysis is per-formed on only the strongest absorber near z fg , and a low R ⊥ sightline may in fact intercept more than one opti-cally thick absorber. We double the number of absorbersfor each mock sightline, and find the measured dispersiononly increases by several km s − . Thus, before the pu-tative asymmetry is confirmed, the hypothesis that theobserved velocity width is only produced by a combina-tion of gravitational motions and Hubble flows cannot beruled out. Extra dynamical processes are not necessaryto explain the large velocity fields.Although we acknowledge the possibility that ourGaussian model does not capture all the powers at ex-treme velocities, we also caution that occasional largekinematic offsets do not make a strong case against grav-itational motions. Firstly, quasars occasionally inhabitextremely overdense environments such as protoclusters(e.g., Hennawi et al. 2015). Secondly, the probability ofintercepting a random, unassociated absorber is boostedby large-scale clustering. Hence occasional extreme ve-locities would not necessitate outflows. Our preferredkinematic measure is thus the dispersion in the averageabsorption.In Figure 4, We show the probability distributions ofthe degenerate parameters W CII and M halo in recoveringthe intrinsic width of the absorption profile. We requirethat the amplitude of the absorption is reproduced within3 times its modeling error, and mark contours for pointsin ( W CII , M halo )-space that produce an absorption profileof width within 1, 2, and 3 times the modeling error of theobserved intrinsic width. From the figure, a higher W CII means the M halo that will reproduce the observed widthis higher. If there are no extra dynamical processes, theintrinsic velocity width corresponding to typical QPQhalo mass is contained within the 2 σ contour. Interpretation of the asymmetric absorption
We quote the standard deviation in the bootstrap re-alizations to be the scatter in the centroids of the data.The scatters ≈
100 km s − are comparable to the mea-sured offsets, indicating large intrinsic variation in quasarCGM environments (see also QPQ8). In Figure 5, weshow the distribution of the absorption centroids frombootstrapping on the C II mean stack. We find that97% of the centroids are at positive velocities. Weplace a generous 3 σ upper limit to the small offset ofthe centroid from z fg for the C II mean absorption at δv < +526 km s − . Given the large intrinsic scatter, wedo not attempt to explore whether there exists relativeasymmetry among the C II , C IV , and Mg II absorption.One may ask whether absorption from C II* II II II* − . In the higher resolution data from QPQ8,we identify nine C II -bearing subsystems where absorp-tion from C II* II II to C II* equivalent width ratio is 0.2. Amongthese nine subsystems, only one, labeled J1420+1603F inQPQ8, has a C
II* II II* mayonly bias the C II centroid by ≈ +40 km s − . Hence,contamination from C II*
III ] emission of z ∼ z quasars. To test for this potential sourceof bias, we create another mean stack at C II III ] redshifts by a redshift measured fromthe more symmetric Mg II or H α emission when avail-able. We are able to replace for 11 out of the 15 systemswith [O III ] redshifts in the original sample. The newstack is similiar in velocity structure and again shows apositive offset ≈ +303 km s − . We thus conclude thatour algorithm for measuring redshifts is not severely bi-ased by the blue wing of the [O III ] emission-line.Motivated by the study of Mg II absorbers surrounding z ∼ II . < z fg < . z fg measured by Hewett & Wild (2010).For quasars with z fg < .
84, as redshift determinationis dominated by [O
III ] emission, a shift of +48 km s − is applied to bring the emission-line redshift to the sys-temic. For quasars with 0 . < z fg < .
6, as redshiftdetermination is dominated by Mg II emission, a shiftof +57 km s − is applied. There are 233 pairs selected,with a median z fg of 0.90 and a median R ⊥ of 208 kpc.We present the mean stack in Figure 6. The absorptionis weaker than the z ∼ − ±
379 km s − and a dispersion of 172 km s − . Theaverage offset from 0 km s − is much smaller than theoffsets in the z ∼ II stack for lower redshift suggests a different cen-troid, we consider that systematic biases are unlikely toexplain the asymmetry signal in the z ∼ DISCUSSION
600 400 200 0 200 400 600Centroid Velocity (km/s)02004006008001000 F r e qu e n c y CII, mean
Figure 5.
Histogram of the absorption centroids of 10000 boot-strap realizations of the data sample for the C II mean stack. 97%of the centroids are positively offset from z fg . N o r m a li ze d F l ux MgII, z ∼ . , mean Figure 6.
Mean stack at Mg II z ∼ .
9. Line-style coding is the same as in Figure 2.The centroid is approximately at 0 km s − , and the absorption isweaker than the main QPQ9 sample. At low redshift, both symmetric and asymmetric ion-ization cones around quasars and AGNs are observed. Inthe extended emission-line region of 4C37.43, most of the[O
III ] emission is blueshifted (Fu & Stockton 2007), butthere are counter examples in less extended sources inFu & Stockton (2009). Recently, there has been Fabry-Perot interferometric data for less extended narrow-lineregions in more nearby sources (Keel et al. 2015, 2017),which show mostly symmetric velocity fields and gas dis-tributions.In the following, we explore two possible explanationsfor the non-dynamical processes that provides the puta-tive asymmetry. The explanations arise from a trans-verse proximity effect (QPQ4), which is the suppressionof opacity in background sightlines passing close to fore-ground quasars.One possibility is an asymmetric radiation field thatpreferentially ionizes the gas moving toward the observer,where the quasar is known to shine. Alternatively, theasymmetric radiation field may preferentially ionizes thegas at smaller Hubble velocity than the quasar. In Fig-ure 7, we show a cartoon of a quasar that is blockedin the direction pointing away from the observer. Thegas observed in absorption preferentially lies behind thequasar. Roos et al. (2015) and Gabor & Bournaud (2014)performed simulations of a high-redshift disk galaxy in-cluding thermal AGN feedback and calculated radiativetransfer in post-processing. They found the ionizationradiation is typically asymmetric, due to either a denseclump that lies on one edge of the black hole or the blackhole’s location being slightly above the disk. Faucher-Gigu`ere et al. (2016) represents the only simulation sofar that is able to reproduce the substantial amount ofcool gas in quasar-mass haloes. We eagerly await theirgroup to compare the fraction of gas in inflows and out-flows.0
FIG. 7 A cartoon showing a unipolar quasar. The gas observed in low- to intermediate-ionabsorption preferentially lies behind the quasar, and is shadowed from the ionizing radiation. Q C C Q C C C ⋎ ⬮ ⬮ ⋎ ⬮ ⬮
300 kpcC C C C + C + C + C + C + C + FIG. 8 A cartoon showing the fi-nite lifetime of quasar episodesas an explanation to the asym-metric absorption. The setup onthe left shows that the foregroundquasar has not been shining longenough for its ionizing radiation toreach the gas behind it, when thelight from the background quasarreaches. The setup on the rightshows the scenario after an amountof time comparable to the lighttravelling time across CGM scale.Gas in front of the foregroundquasar has been ionized, by thetime the light from the backgroundquasar reaches. Q C C Q C C C ⋎ ⬮ ⬮ ⋎ ⬮ ⬮
300 kpcC C C Q C Q C C ⋎ ⬮ ⬮ ⋎ ⬮ ⬮
300 kpcC C + C + C + C + C + C + IV absorption. The redshiftedC IV may be regarded as an intermediate ion, and ablueshifted absorption signal needs to be searched in ahigher ion.We test the case of Hubble flows versus galactic-scale outflows in producing the putative asymmetry.Anistropic emission is degenerate with intermittent emis-sion in their asymmetric light-echo in the observer’sframe. We first consider the scenario where the quasar’slifetime is infinite, and the observed ansymmetric ab-sorption is only produced by anisotropic emission. Wetry implementing an arbitrary quasar opening angle inour Monte Carlo simulations to reproduce an absorp-tion centroid that is redshifted from the systemic. InQPQ6, we argued that the observed anisotropic cluster-ing of optically thick systems around quasars demandsthat a quasar’s radiation field must affect optically thicksystems on Mpc scales. Hence in this test, we set the in-cidence of absorbers happening within the quasar open-ing angle to be zero, for absorbers at all disances fromthe quasar. In the case without outflows, even if we setthe opening angle to be 180 ◦ , i.e. absorption only hap-pens at positive velocities, the mean absorption centroidwould merely reach ≈ +85 km s − . Hence, the observed δv ≈ +200 km s − shift in the mean absorption cannotbe produced by asymmetric distribution in Hubble ve-locities alone. This suggests the presence of an outflowcomponent to account for extra asymmetry in line-of-sight velocities. In order to produce the observed intrin-sic dispersion and centroid of the mean absorption byC II , we must add a radial outflow speed of ≈
450 km s − to the absorbers and a unipolar quasar opening angleof ≈ ◦ . Next, we consider another scenario wherethe quasar is isotropic, and the asymmetric absorptionis only produced by short episodic lifetime. We furthermake the assumption the luminosity is constant during aquasar episode. In the quasar’s rest frame, the light echowould be observed as a spherical region with the quasar at the origin. In the observer’s frame, due to finite lighttravel time, the light-echo would not be spherically sym-metric. The light-echo from the quasar at its luminosity t yr earlier traces a paraboloid with the quasar at thefocus and the vertex t/ ≈
420 km s − and the quasar must have shinedfor ≈ . × yr. For the anisotropy-only scenario, theopening angle deduced is rather large compared to litera-ture findings, which give 30 ◦ –90 ◦ (e.g., Trainor & Steidel2013; Borisova et al. 2016). For the intermittence-onlyscenario, if the quasars are on averaged observed nearthe middle of the episode, the lifetime deduced is some-what small compared to other existing constraints fromobservations and simulations, which give 10 –10 yr (e.g.,Martini 2004; Hopkins et al. 2005). We thus speculatethat the asymmetric absorption is the result of a combi-nation of anisotropic and intermittent emission.We note that, Turner et al. (2017) conclude that theclustering of low to intermediate ions around Lyman-break galaxies in velocity space is most consistent withgas that is inflowing on average. The different conclu-sion from our analysis may originate from the highermasses of our systems, the presence of quasar-driven ou-flows (e.g., Greene et al. 2012), and/or starburst-drivenoutflows that are correlated with the presence of quasars(e.g., Barthel et al. 2017).Motivated by the asymmetry found in metal ion ab-sorption in the CGM using precise z fg measurements,and the asymmetry found by Kirkman & Tytler (2008)in H I on larger scales, we are assembling a sample ofquasar pairs with precise z fg measurements to study thisasymmetry in H I (J. F. Hennawi et al. 2018, in prepa-ration). In conclusion, we observe large and positivelyskewed velocity fields in absorption, of metal ions in theCGM of z ∼ APPENDIX
LINE-OF-SIGHT ABSORPTION
In the previous QPQ papers, we argued that optically thick absorbers in the vicinity of quasars are distributedanisotropically. We now have the means to show this anisotropic clustering explicitly. Given that the techniques for2 N o r m a li ze d F l ux CII, f/g, mean0.900.951.001.05 N o r m a li ze d F l ux MgII, f/g, mean5000 0 5000Relative Velocity (km/s)0.900.951.001.05 N o r m a li ze d F l ux CIV, f/g, mean
Figure 9.
Mean stacks of the foreground quasar spectra at C II IV II II andMg II , the mean absorption is weaker than that in the background stacks, and there is no evidence for an excess at z fg . The stack for C IV ,which includes line-of-sight absorbers at all distances, shows a large, blueshifted mean velocity field. stacking spectra are established, it is straightforward to apply the same techniques to stack the foreground quasarspectra. In Figure 9, we present mean stacks of foreground quasar spectra for the QPQ9 sample. We require thatthe spectra survive a S/N cut of 5.5 per rest-frame ˚A at C II IV II z fg . In contrast tothe large equivalent widths exhibited in the stacks of background spectra, C II and Mg II mean absorption alongthe line-of-sight to the foreground quasars is weaker, and an excess at z fg is absent. This supports a scenario wherethe ionizing radiation of the foreground quasars are anisotropic and/or intermittent. For C IV IV absorption has been well studied as narrow associated absorption line systems (e.g.,Wild et al. 2008). We note that, in the z ∼ V absorption is found in proximity to thehost quasars. REFERENCES Adelberger, K. L. 2004, ApJ, 612, 706Baldwin, J. A. 1977, ApJ, 214, 679Barthel, P., Podigachoski, P., Wilkes, B., & Haas, M. 2017, ApJ,843, L16Bielby, R., Hill, M. D., Shanks, T., Crighton, N. H. M., Infante,L., Bornancini, C. G., Francke, H., H´eraudeau, P., Lambas,D. G., Metcalfe, N., Minniti, D., Padilla, N., Theuns, T.,Tummuangpak, P., & Weilbacher, P. 2013, MNRAS, 430, 425Bergeron, J., & Boisse, P. 1991, Advances in Space Research, 11,241Borisova, E., Lilly, S. J., Cantalupo, S., Prochaska, J. X., Rakic,O., & Worseck, G. 2016, ApJ, 830, 120Boroson, T. 2005, AJ, 130, 381Cantalupo, S., Arrigoni-Battaia, F., Prochaska, J. X., Hennawi,J. F., & Madau, P. 2014, Nature, 506, 63Churchill, C. W., Kacprzak, G. G., Steidel, C. C., Spitler, L. R.,Holtzman, J., Nielsen, N. M., & Trujillo-Gomez, S. 2012, ApJ,760, 68Eftekharzadeh, S., Myers, A. D., White, M., Weinberg, D. H.,Schneider, D. P., Shen, Y., Font-Ribera, A., Ross, N. P., Paris,I., & Streblyanska, A. 2015, MNRAS, 453, 2779Faucher-Gigu`ere, C.-A., Feldmann, R., Quataert, E., Kereˇs, D.,Hopkins, P. F., & Murray, N. 2016, MNRAS, 461, L32Fu, H., & Stockton, A. 2007, ApJ, 666, 794—. 2009, ApJ, 690, 953Fumagalli, M., Prochaska, J. X., Kasen, D., Dekel, A., Ceverino,D., & Primack, J. R. 2011, MNRAS, 418, 1796Gabor, J. M., & Bournaud, F. 2014, MNRAS, 441, 1615 Gauthier, J.-R. 2013, MNRAS, 432, 1444Greene, J. E., Zakamska, N. L., & Smith, P. S. 2012, ApJ, 746, 86Hennawi, J. F., Prochaska, J. X., Burles, S., Strauss, M. A.,Richards, G. T., Schlegel, D. J., Fan, X., Schneider, D. P.,Zakamska, N. L., Oguri, M., Gunn, J. E., Lupton, R. H., &Brinkmann, J. 2006, ApJ, 651, 61Hennawi, J. F., Prochaska, J. X., Cantalupo, S., &Arrigoni-Battaia, F. 2015, Science, 348, 779Hewett, P. C., & Wild, V. 2010, MNRAS, 405, 2302Ho, S. H., Martin, C. L., Kacprzak, G. G., & Churchill, C. W.2017, ApJ, 835, 267Hopkins, P. F., Hernquist, L., Martini, P., Cox, T. J., Robertson,B., Di Matteo, T., & Springel, V. 2005, ApJ, 625, L71Johnson, S. D., Chen, H.-W., & Mulchaey, J. S. 2015, MNRAS,452, 2553Keel, W. C., Lintott, C. J., Maksym, W. P., Bennert, V. N.,Chojnowski, S. D., Moiseev, A., Smirnova, A., Schawinski, K.,Sartori, L. F., Urry, C. M., Pancoast, A., Schirmer, M., Scott,B., Showley, C., & Flatland, K. 2017, ApJ, 835, 256Keel, W. C., Maksym, W. P., Bennert, V. N., Lintott, C. J.,Chojnowski, S. D., Moiseev, A., Smirnova, A., Schawinski, K.,Urry, C. M., Evans, D. A., Pancoast, A., Scott, B., Showley,C., & Flatland, K. 2015, AJ, 149, 155Kereˇs, D., Katz, N., Fardal, M., Dav´e, R., & Weinberg, D. H.2009, MNRAS, 395, 160Kirkman, D., & Tytler, D. 2008, MNRAS, 391, 1457Lau, M. W., Prochaska, J. X., & Hennawi, J. F. 2016, ApJS, 226,253