1: % ============================================================================

2: 

3: \documentclass{emulateapj}

4: 

5: \usepackage{natbib}

6: \usepackage{epsfig,color}

7: \usepackage{amssymb,amsmath}

8: \usepackage{multirow}

9: 

10: \shorttitle{Finding lenses near lenses}

11: \shortauthors{Newton et~al.}

12: 

13: \def\hon{$H_1$}

14: \def\htw{$H_2$}

15: \def\hth{$H_3$}

16: \def\hfo{$H_4$}

17: \def\hro{H_r^{opt}}

18: \def\hrr{H_r^{real}}

19: \def\thetaE{\theta_{\rm E}}

20: \def\zd{z_d}

21: \def\zs{z_s}

22: \def\Dd{D_d}

23: \def\Ds{D_s}

24: \def\Dds{D_{ds}}

25: \def\sigv{\sigma_V}

26: \def\sigstar{\sigma_*}

27: \def\sigSIE{\sigma_{\rm SIE}}

28: \def\mf{m_{\rm F814W}}

29: \def\S{Section}

30: 

31: % ============================================================================

32: \begin{document}

33: 

34: \title{Enhanced lensing rate by clustering of massive galaxies: \\

35: newly discovered systems in the SLACS fields}

36: 

37: \author{Elisabeth R. Newton\altaffilmark{1}}

38: \author{Philip J. Marshall\altaffilmark{1}}

39: \author{Tommaso Treu\altaffilmark{1,2}}

40: 

41: \altaffiltext{1}{Department of Physics, University of California,

42: Santa Barbara, CA 93106-9530}

43: \altaffiltext{2}{Alfred P. Sloan Research Fellow; Packard Fellow}

44: 

45: % ----------------------------------------------------------------------------

46: 

47: \begin{abstract}

48: 

49: Galaxy-scale strong gravitational lens systems are useful for a

50: variety of astrophysical applications. However, their use is limited

51: by the relatively small samples of lenses known to date. It is thus

52: important to develop efficient ways to discover new systems both in

53: present and forthcoming datasets. For future large high-resolution

54: imaging surveys we anticipate an ever-growing need for efficiency and

55: for independence from spectroscopic data. In this paper, we exploit

56: the clustering of massive galaxies to perform a high efficiency

57: imaging search for gravitational lenses.  Our dataset comprises 44

58: fields imaged by the Hubble Space Telescope (HST) Advanced Camera for

59: Surveys (ACS), each of which is centered on a lens discovered by the

60: Strong Lens ACS Survey (SLACS). We compare four different search

61: methods: 1) automated detection with the HST Archive Galaxy-scale

62: Gravitational Lens Survey (HAGGLeS) robot, 2) examining cutout images

63: of bright galaxies (BGs) after subtraction of a smooth galaxy light

64: distribution, 3) examining the unsubtracted BG cutouts, and 4)

65: performing a full-frame visual inspection of the ACS images.  We

66: compute purity and completeness and consider investigator time for the

67: four algorithms, using the main SLACS lenses as a testbed. The first

68: and second algorithms perform the best. We present the four new lens

69: systems discovered during this comprehensive search, as well as one

70: other likely candidate. For each new lens we use the fundamental plane

71: to estimate the lens velocity dispersion and predict, from the

72: resulting lens geometry, the redshifts of the lensed sources. Two of

73: these new systems are found in galaxy clusters, which include the

74: SLACS lenses in the two respective fields. Overall we find that the

75: enhanced lens abundance ($30^{+24}_{-8}$ lenses/degree$^2$) is higher

76: than expected for random fields ($12^{+4}_{-2}$ lenses/degree$^2$ for

77: the COSMOS survey). Additionally, we find that the gravitational

78: lenses we detect are qualitatively different from those in the parent

79: SLACS sample: this imaging survey is largely probing higher-redshift,

80: and lower-mass, early-type galaxies.

81: 

82: \end{abstract}

83: 

84: \keywords{%

85:  gravitational lensing  -- 

86:  techniques: miscellaneous  -- 

87:  surveys}

88: 

89: % ----------------------------------------------------------------------------

90:  

91: \section{Introduction}\label{sec:intro}

92: 

93: Strong gravitational lensing -- when the potential of a massive

94: foreground object causes the formation of multiple images of a

95: background source -- is a powerful tool for cosmological and

96: astrophysical research.  Applications include measuring the mass

97: distributions of dark and luminous matter, and measuring cosmological

98: parameters via the lens geometry or abundance \citep[see e.g.\ ][for a review]{Koc++06}.  

99: A homogenous, well-understood sample of lenses is

100: required for a statistically significant study, necessitating

101: large-scale, systematic surveys \citep{TOG84, Bol++06,Ina++08}.  The

102: $\sim200$ galaxy-scale lenses known today have been found through the

103: numerous selection algorithms described in the next paragraph, and

104: include a significant number of serendipitous discoveries. Although

105: this has enabled substantial progress, the number is still

106: the limiting factor for many applications.

107: 

108: Most searches so far have focused on either the source or lens population,

109: employing a range of different strategies.  Searches targeting potential sources

110: have included looking for multiply-imaged radio sources -- as in the

111: Cosmic Lens All-Sky Survey \citep[CLASS;][]{Mye++03, Bro++03} -- and

112: examining known quasars \citep[e.g.][]{Mao++97,Pin++03}. The Sloan

113: Digital Sky Survey (SDSS) Quasar Lens Search \citep{Ogu++06, Ina++08}

114: selects lensed quasar candidates using two algorithms, one based on

115: morphology and color for small-separation images, and one based only

116: on color.  Of the surveys targeting potential lenses, most have

117: involved inspection of high resolution images. \citet{Rat++99} and

118: \citet{Mou++07} selected candidates by eye from HST color images of

119: the Extended Groth Strip (EGS) fields.  Other authors have attempted

120: to pre-select massive galaxies by their optical magnitude and color,

121: and then examine the residuals after galaxy subtraction.  This was done in the

122: Great Observatories Origins Deep Survey (GOODS) lens search

123: \citep[][]{Fas++04} and the Cosmic Evolution Survey (COSMOS) lens

124: search \citep[]{Fau++08}. Most recently, \citet{J08} extended this

125: method to lower mass galaxies, but focused on viewing large-format

126: arrays of unsubtracted color galaxy cutout images.  The Sloan Lens ACS

127: Survey \citep[]{Bol++06, Bol++08a} candidates were selected based on

128: spectroscopic data indicating multiple redshifts in the spectrum of

129: early-type (and hence massive) galaxies, then classified using this

130: data and visual examination of high resolution HST images (before and

131: after lens galaxy subtraction).

132: 

133: Recently, several algorithms for automated lens detection have been

134: developed.  Those developed by \citet{Ala06}, \citet[]{S+B07} and

135: \citet{K+D08} look for arcs, a common feature in both weak and strong

136: lensing on group and cluster scales.  \citet{Est++07} and

137: \citet{Bel++07} suggest mining databases for blue objects near

138: potential lenses, since the most common sources are faint blue

139: background galaxies. The RingFinder \citep[]{Cab++07} applies the same

140: logic to smaller image separation lenses, subtracting a rescaled red

141: image from a blue cutout image to dig within the lens light

142: distribution; it then analyzes the shapes and positions of the

143: remaining residuals.  Most recently the HAGGLeS automated lens

144: detection ``robot'' \citep[]{Mar++08} attempts to model every object

145: (typically selected to be Bright Red Galaxies or BRGs) as a

146: gravitational lens, i.e.\ as a combination of background light from

147: the source that is consistent with having been multiply-imaged, and

148: residual foreground light from the lens galaxy. The result is the

149: robot's quantitative prediction of how a human would have classified

150: the candidate.

151: 

152: Fortunately, we are about to enter an era when orders of magnitude

153: increases in the number of known lenses will be possible. In the near

154: future, wide-field surveys such as the proposed Joint Dark Energy

155: Mission (JDEM) and Euclid space missions would provide $\sim10^3 -

156: 10^4$ square degrees of high resolution imaging data

157: \citep[]{Mar++08}.  Automation will be needed to examine such large

158: areas over manageable timescales, and to draw attention to those

159: systems which have a higher probability of being lenses. We also must

160: be prepared to proceed without the help of spectroscopic data.

161: 

162: We describe here just such a survey: based purely on imaging data, and with a

163: sufficient degree of automation. We compare the accuracy of four

164: methods of searching for strong gravitational lenses. Presented in

165: order of degree of automation, they are: (1) using the HAGGLeS robot

166: \citep{Mar++08}, (2) examining subtraction residuals \citep[e.g.][]{Fau++08}, (3) looking at galaxy cutouts

167: \citep[e.g.][]{Fas++04,J08}, and (4) performing a visual inspection of

168: entire fields \citep[e.g.][]{Mou++07}.

169: 

170: We additionally aim to make use of galaxy clustering in order to

171: improve the efficiency of our search, as suggested by

172: \citet[][]{Fas++06}.  The most likely lensing galaxies are massive

173: ellipticals with $0.3 < z < 1.3$ \citep[]{TOG84,Fas++04}: we may

174: expect that focusing on bright red galaxies (BRGs) will increase the

175: lens detection efficiency \citep[e.g.][]{Fas++04,Fau++08,Mar++08}. The

176: sources, meanwhile, are predominantly expected to be faint blue

177: galaxies (FBGs) at redshifts at or above 1 \citep[]{MBS05}.  Both of

178: these types of objects (potential lenses and potential sources) are

179: clustered, BRGs more strongly that FBGs.  We therefore expect that

180: strong gravitational lenses are also clustered; we anticipate finding

181: more new lenses by looking near known ones than we would otherwise.

182: For example, \citet[]{Fas++06} presents just such an occurrence: the

183: researchers discovered two additional lens candidates less than 40

184: arc-sec from the known lens B1608+656.

185: 

186: Therefore, as our dataset we use a subset of the SLACS HST/ACS fields.

187: The SLACS survey has discovered 70 definite galaxy-galaxy strong

188: lenses~\citep{Bol++08a} to date.  These definite lenses have clearly

189: identifiable lenses arcs or multiple images in addition to

190: spectroscopic data; due to the rigorous requirements for this

191: classification, we consider all to be confirmed lenses.  Of these, 63

192: are well-modeled by a singular isothermal ellipsoid (SIE) and have

193: lens and source redshifts along with F814W photometry for the lens

194: galaxy; most also have measured stellar velocity dispersions

195: \citep[]{Bol++08a}.  This makes the SLACS lenses the largest

196: homogenous sample of strong lenses to date.  Each of the fields we use

197: is centered on one of the known SLACS lenses (hereafter the ``main

198: lens''); we thus make use of gravitational lens clustering in every

199: field.  The majority of our fields were observed in just one filter,

200: meaning that we by necessity pre-select bright galaxies (BGs) rather

201: than BRGs. For the automated portion of our searches, we use the

202: HAGGLeS robot which, in addition to performing a quantitative lens

203: classification, creates a useful database of galaxy cutouts and

204: residuals.

205: 

206: The organization of this paper is as follows.  In

207: Section~\ref{sec:sample} we present the 44 ACS fields in our dataset,

208: and in Section~\ref{sec:methods} we outline the four search procedures

209: we used to find gravitational lenses in these images. To assess the

210: accuracy of these methods, we discuss each procedure's performance on

211: the main lenses in Section~\ref{sec:accuracy}.  We then present our

212: four new definite strong lensing systems and one likely candidate, and

213: investigate their physical properties and environments, in

214: Section~\ref{sec:results}.  After some discussion of our methods and

215: results in Section~\ref{sec:discussion}, we conclude in

216: Section~\ref{sec:conclusions}.  All magnitudes are in the AB system;

217: we assume a flat cosmology with $\Omega_m$=0.3, $\Omega_\Lambda$=0.7,

218: and H$_0$=100$h$ kms$^{-1}$Mpc$^{-1}$, where $h$=0.7 when necessary.

219: 

220: 

221: % ----------------------------------------------------------------------------

222:  

223: \section{Sample Selection and Object Detection}\label{sec:sample}

224: 

225: We use a subsample of the ACS F814W fields investigated in the Sloan

226: Lens ACS Survey

227: \citep[SLACS;][]{Bol++06,Tre++06,Koo++06,Gav++07,Bol++08a,Gav++08,Bol++08b,

228: Tre++08}, hereafter papers SLACS-I through VIII. The sample selection

229: and data reduction and analysis of the SLACS data is given in SLACS-I,

230: -IV and -V and will not be repeated here. We select the subset of 44

231: definite lenses whose ACS observations meet the following two

232: requirements: exposure time $>$1000 seconds through F814W, and lens

233: classification ``definite.''  Due to the demise of ACS, two filters

234: are not available across all fields; of the ACS filters used, F814W is

235: the most widely available in the SLACS fields. 14 fields were

236: additionally imaged through F555W.  We require long exposure time to

237: guarantee the ability to produce deep, high quality, cosmic ray-free

238: reduced images.

239: 

240: Our final sample comprises 44 uniformly-observed ACS fields, each

241: centered on a SLACS gravitational lens (a ``main lens'').  Each field

242: covers approximately 11 arcmin$^2$; thus our total survey area is

243: 0.134 square degrees.

244: 

245: Most lens-finding methods involve examining bright galaxies (BGs), as

246: being the most likely to be lensing galaxies \citep[e.g.][]{Fas++04,

247: Fau++08, Mar++08}.  Here this selection is done with magnitude and

248: size cuts on catalogs made with the

249: SExtractor\footnote{http://terapix.iap.fr/} software tool; only a

250: third of our fields have multi-filter ACS data so in order to maintain

251: consistency we do not use color selection even when it is possible.

252: We choose an apparent F814W magnitude ($\mf$; using Kron-like elliptical

253: apertures) limit of 22. Other limits were tried, but 22

254: was found empirically to be the best balance between efficiency and

255: the selection of all bright galaxies: with $\mf<22$, a total of 2399

256: BGs are selected, but with $\mf<23$, that number jumps to 5439

257: without adding significantly to the actual number of morphologically

258: early-type galaxies selected.  The minimum size selected is

259: FWHM=0$\farcs$215, in order to reject most stars.  We additionally use

260: the SExtractor ``flags'' parameter to limit the remaining

261: contaminants.

262: 

263: 

264: % ----------------------------------------------------------------------------

265: 

266: \section{Search Methods}\label{sec:methods}

267: 

268: We use four distinct methods to search for gravitational lenses.  They

269: are presented in order of degree of automation, where the first has the most

270: pre-selection and analysis, and the last none:

271: 

272: \begin{itemize}

273: \item Procedure 1: Using output from the HAGGLeS robot

274: \item Procedure 2: Inspection of subtraction residuals of bright galaxies

275: \item Procedure 3: Inspection of cutout images of bright galaxies

276: \item Procedure 4: Full-frame visual inspection of the ACS images

277: \end{itemize}

278: 

279: 

280: In general we expect our searches to be sensitive to

281: lensing events similar to, but fainter than, the SLACS main

282: lenses. With the BG selection in Procedures 1--3, and the innate

283: limits of the human eye in Procedure 4, we are biased towards lensing

284: events around large, luminous galaxies. However, empirically the main

285: lenses are the brightest objects (other than saturated stars) in the

286: fields, so we expect any other lenses we find to be, by necessity,

287: fainter. The main lenses also favor high magnification configurations

288: (Einstein rings) due presumably to their spectroscopic selection; we

289: expect to be somewhat preferential in this regard as well, on the

290: grounds that these distinctive cases will be easiest to identify

291: visually.

292: 

293: Before continuing, we first consider the question ``what is the

294: probability that this object is a gravitational lens?''  The

295: classification of lens candidates has varied among the major surveys

296: targeting potential lenses, as has the data on which these

297: classifications are based.  The COSMOS lens search \citep[]{Fau++08}

298: sample is subdivided into ``best systems'', which are deemed to have a

299: greater probability of being lenses than the remaining objects.  GOODS

300: \citep[]{Fas++04} selected candidates and voted to choose ``top

301: candidates.''  \citet[]{J08} groups potential lenses into three

302: categories: very likely or certain lenses, possible or probable

303: lenses, and not-lenses.  The SLACS survey \citep[]{Bol++08a} also

304: groups candidates into three categories (definite lenses, probable

305: lenses, and inconclusive/not lenses).  We note that the classification

306: of definite or probable lenses is not the same across surveys, and

307: therefore one has to be careful to impose similar quality criteria

308: when comparing inferred density of lens galaxies.

309: 

310: \input{tab1.tex}

311: 

312: In this work we follow \citet{Mar++08} and employ a 4-point subjective

313: classification scheme, outlined in Table~\ref{tab:class}: the classification 

314: parameter $H$ may range from 3 (definite lenses), through 2 (probable lenses)

315: and 1 (possible lenses), to $H=0$  (definitely not lenses).  Each of the 44

316: fields in our sample contains a confirmed  lens from the SLACS survey,

317: where here the grade is based on all available data, notably the  galaxy

318: spectrum (including anomalous high redshift emission lines) and clear

319: lensing-consistent residuals after lens galaxy light subtraction in all procured filters.  

320: We thus assign the classification parameter $H=3$ for each main lens. 

321: 

322: Humans (after some training) are adept at identifying gravitational

323: lenses by eye, using high resolution imaging data alone; we

324: effectively make an internal model for the lens and optimize

325: it. However, the amount of information available during each lens

326: search with which to do this may vary: in any given search we always

327: lack additional understanding that would aid us in identifying lenses.

328: When carrying out the four different search procedures, we therefore

329: assign each system a value $H_i$ (where $i=1,2,3$ or $4$): each of

330: these is our best guess as to the classification that a trained human

331: would have assigned the system, if they had been given only the data

332: presented in the $i^{\rm th}$ procedure.  (This results in four

333: ``re-classifications'' for each main lens.)  We consider the ``true''

334: classification $H$ to be the value we give an object when taking into

335: account \emph{all data}.

336: 

337: 

338: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

339: 

340: \subsection{Procedure 1: HAGGLeS robot}

341: 

342: In Procedure 1, we use the HAGGLeS robot, the automated lens detection program

343: developed by \citet[]{Mar++08}, to identify samples of lens candidates prior

344: to visual inspection.  The robot treats every object as if it were a

345: lens, models it, and then calculates how well the lens hypothesis works.  It

346: extracts and uses 6 arc-second square cutout images of each object, and then

347: focuses on the residuals made by subtracting off an elliptically symmetric

348: Moffat profile model for the putative lens galaxy light. The gravitational

349: potential of the lensing galaxy is assumed to be sufficiently well-described

350: by a singular isothermal sphere (SIS) plus external shear; this model is

351: fitted to the residual image. Disk-like features and bright neighboring

352: objects are masked before fitting.  When multi-filter data is available it is

353: used, and although the robot does not rely on color, with it performance

354: is expected to improve \citep[]{Mar++08}.  Based on the results of the

355: modeling process, the robot calculates and assigns a value of $H_r$ to each

356: object, where $H_r$ is the robot's estimate of the classification $H$ a human

357: would have given the system.  We summarize the human and robot classifications

358: in Table \ref{tab:class} \citep[after][]{Mar++08}.

359: 

360: The robot may have one of two characters, reflecting the prior

361: probability that an object is a lens.  The ``realistic'' character

362: robot might expect 0.1\% of objects to be lenses, and the

363: ``optimistic'' character robot might expect 60\% to be

364: \citep[]{Mar++08}.  The realistic robot is approximately in accordance

365: with current estimates of the fraction of strongly lensed galaxies

366: \citep[e.g.][]{Bol++06,Mou++07,Mar++08}, while the optimistic robot

367: has the advantage of being most inclusive by giving higher values of

368: $H_r$.  These two prior probability distributions are given

369: in Table \ref{tab:pdf} \citep[after][]{Mar++08}. \citet{Mar++08} note

370: that while the realistic robot produces lens samples of high purity -- and

371: correspondingly little need for human inspection -- a highly

372: complete search requires, at present, a more optimistic robot.

373: 

374: \input{tab2.tex}

375: 

376: We ran both the optimistic and realistic robots in order to verify

377: this last claim, but chose to use the optimistic robot to search for

378: new lenses, as our sample is small and we are most interested in

379: finding all lenses present in our sample. When comparing the

380: performances of both robots, we refer to the optimistic and realistic

381: robot classifications as $\hro$ and $\hrr$, respectively.  The main

382: goal of the comparison is to gain insight into the populations of

383: objects selected, and thus help improve the automated part of this

384: procedure for future surveys.

385: 

386: The output from the robot includes an individualized webpage for each object,

387: containing the cutout, subtraction residual, masked residual, and estimation

388: of Einstein radius.  This page also displays the minimal source  able to

389: produce the observed configuration, and the predicted image-plane residual

390: reconstructed from this minimal source \citep[see][for more details]{Mar++08}.   

391: We viewed the pages of objects for which $\hro > 1.5$ (indicating optimistically 

392: probable and definite lenses),  and gave each

393: object a human classification \hon, using the same data that was available to

394: the robot (namely, the cutout image and subtraction residual image), and the

395: lensmodel outputs produced by the robot.

396: 

397: 

398: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

399: 

400: \subsection{Procedure 2: Examining BG residuals}

401: 

402: Procedure 2 makes use only of the subtraction residuals produced by

403: the robot.  We examine the residuals of every BG initially selected in

404: the fields; all of the $6''$ by $6''$ subtraction residual cutouts for

405: a field are displayed in a grid for rapid viewing. When multi-filter

406: data is available, the residuals are shown in color. Looking at the 44

407: grids, we note all objects of interest; to each of these objects we

408: assign a human classification \htw, based solely on the subtraction

409: residual data.

410: 

411: 

412: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

413: 

414: \subsection{Procedure 3: Examining BG cutouts}

415: 

416: In this method, we inspect each of the $6''$ by $6''$ cutouts of the BGs

417: provided by the HAGGLeS robot. We again display the cutout images in a grid

418: for rapid viewing, making this procedure very similar to that employed by

419: \citet{J08}.  However, the stretch of the cutouts' image display is fixed for

420: all objects at a level appropriate to the EGS-type lenses \citep{Mou++07}.

421: This makes this method particularly sensitive to faint lensing events, and

422: somewhat insensitive to lensing by very bright galaxies.   We assign objects of interest a

423: human classification, this time \hth, using only the galaxy cutout on display.

424: 

425: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

426: 

427: \subsection{Procedure 4: Full-frame visual inspection}

428: 

429: During the full-frame visual inspection, each of the 44 fields' F814W ACS

430: images is viewed with ds9.\footnote{http://hea-www.harvard.edu/RD/ds9/}  We

431: initially set image parameters in each field so as to

432: be sensitive to the lensing events we expect to find: similar to, but fainter

433: than, the main lenses.  We do this by setting the intensity stretch such that the main lens is slightly over-saturated;

434: from this display we select objects of interest. For each object, we adjust the scale limits and intensity stretch

435: such that the potential lensing features are most apparent, then decide on a

436: classification. Procedure 4 differs from the previous three most notably in that all galaxies in the

437: field -- not just BGs -- are examined, and that their viewing parameters are

438: set individually. The human classification parameter assigned in this

439: procedure is denoted \hfo.

440: 

441: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

442: 

443: \subsection{Summary of Results}

444: 

445: During each procedure, any potential gravitational lens was marked and

446: assigned a human classification based only on the information

447: available during the search procedure by which it was found. Following

448: the completion of all searches, all available information was

449: considered holistically and objects were assigned a final human

450: classification $H$. The classification is based primarily on our

451: ability to recognize a typical lens geometry in the observed

452: morphology (in both the cutout and subtraction residual), on colors

453: and surface brightnesses consistent with lensing, and on the robot's

454: ability to model the object.  In total we discovered four new objects

455: with $H = 3$, and one object with $H = 2$. These new gravitational

456: lenses are presented in Section~\ref{sec:results}.

457: 

458: 

459: % ----------------------------------------------------------------------------

460: 

461: \section{Method Accuracy}\label{sec:accuracy}

462: 

463: Before investigating the properties of our newly-discovered lenses, we

464: first discuss the performance of the four search procedures in terms

465: of both their ability to find SLACS-type lenses (low redshift, high

466: magnification), and their contribution to the finding of the new

467: gravitational lens systems.  For each method, we calculate the purity

468: and completeness of the samples selected, considering our five best

469: systems and the SLACS main lenses together as lens systems to be

470: recovered.  We define purity as the percent of selected objects

471: (i.e. having classification parameter $H_i$ above some threshold),

472: that actually have final human classification $H$ greater than the

473: same threshold.  Only the SLACS main lenses and our 5 best systems

474: have $H > 1.5$, and all but one of these has $H = 3$.  Similarly,

475: completeness is defined as the percent of the objects with final human

476: class $H$ greater than some threshold that were given procedure

477: classification $H_i$ greater than the same threshold.  For Procedure

478: 1, any object that was not examined because $\hro < 1.5$ is considered

479: to have $H_1 = 0$.  Purity and completeness were calculated for each

480: procedure, and for both the realistic and optimistic robots. These

481: statistics are summarized in Table~\ref{tab:stats}; errors are of

482: order a few percent.

483: 

484: \input{tab3.tex}

485: 

486: Note that during each search procedure, the main SLACS lenses

487: themselves ae classified, resulting in values of \hon, \htw, \hth, and

488: $H_4$ for each.  A lens-by-lens comparison of classification values is

489: available in the Appendix, while data on the total numbers of main

490: lenses and new lens recovered by each search procedure are presented

491: in Table \ref{tab:methods}.  In this table we also include the amount

492: of time spent per field for each method, and include the performance

493: of the HAGGLeS robot (with no human inspection) for comparison.

494: 

495: \input{tab4.tex}

496: 

497: In the subsections that follow, we briefly discuss these results and their

498: implications for optimizing lens discovery in high resolution imaging surveys.

499: In the Appendix we show, for reference, some of the images that comprise the

500: input data to the first three search procedures, namely, cutout images of the

501: main lenses, and of the $\hrr > 2.5$ candidates.

502: 

503: 

504: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

505: 

506: \subsection{Procedure 1}\label{subsec:proc1a}

507: 

508: Inspecting all objects classified by the optimistic robot as $\hro >

509: 1.5$, we classified three out of the five new systems as $H_1 = 3$,

510: and one system as $H_1 = 2$.  The final system had been assigned a

511: robot classification parameter of $\hro < 1.5$ and so was not

512: examined. This failure is most likely due to the non-lensed light

513: remaining after the Moffat subtraction (see the Appendix for more

514: illustration of this).

515: 

516: Indeed, for the main lenses, we found that the unmasked disk features

517: are the most common cause of robot detection failures: we show 

518: the subtraction residuals and reconstructed images in the

519: Appendix.  The robot may fail to find a suitable model entirely 

520: in some extreme cases, such as when considering objects with very strong

521: disk components or bright companion galaxies (main lenses SDSSJ1029+0420, SDSJ1103+5322 and SDSSJ1416+5136).

522: We find overall that 8 of the main lenses, including the three cases just mentioned, are not detected at

523: $\hro > 1.5$ by the robot.  Interestingly, one of the 36 robot-detected

524: systems, and three of the 8 robot-rejected systems were

525: human-classified as $H_1 < 1.5$: given just the output from the robot, these lenses (SDSSJ1213+6708, and SDSSJ1016+3859

526: SDSSJ1029+0420, and SDSSJ1032+5322) would not have been identified as

527: lenses by a human inspector.

528: 

529: %Indeed, for the main lenses, we found that the unmasked disk features

530: %are the most common cause of robot detection failures. We show some

531: %subtraction residuals, and reconstructed images in

532: %Figure~\ref{fig:mainlenses}.  In these cases, the robot failed to find

533: %a suitable model entirely.  Two of these situations (main lenses

534: %SDSSJ1029+0420 and SDSJ1103+5322) were lens galaxies with very strong

535: %disk components, and the disk mask also masked the inner lensed

536: %images.  In the third system (main lens SDSSJ1416+5136), the masking

537: %of both a small, but bright, disk component and a bright companion

538: %galaxy masked the lensing event.  

539: 

540: The lens galaxy subtraction also seems to be the major cause of false

541: detections: ring-like and disk-like features left over from the Moffat

542: profile subtraction and then incompletely masked can be wrongly

543: interpreted by the robot as lenses.  Examples of such false positives

544: can be seen in the 10 objects classified by the realistic robot as

545: $\hrr > 1.5$ (in the Appendix).

546: 

547: Of the 36 robot-detected main lenses, only 20 were classified by the

548: optimistic robot as ``definite lenses'' ($\hro > 2.5$). The reasons

549: for these mis-classifications are a little more subtle. The probability

550: density functions (PDFs) used to calculate the $H_r$ values for

551: this research were determined by \citet[]{Mar++08} based on a training

552: set of EGS non-lenses and simulated lenses.  By overlaying the robot

553: model output for the SLACS main lenses on these PDFs, we can gain

554: insight into the cause of of the robot's mis-classifications.  In

555: Figure~\ref{fig:priors3} we show ${\rm Pr}(\mathbf{d}|H=3)$, and

556: overlay the $\mathbf{d}$-values for the main lenses (larger data

557: points). Here we clearly see that the PDF (contours approximating the

558: density of smaller, training set points) is not optimized for

559: SLACS-type (low redshift, high magnification) lenses: to the robot,

560: source magnitudes are surprisingly bright and arcs are of unusual

561: thickness.  These differences are to be expected, but we note that the

562: lenses we should expect to find with the HAGGLeS robot would therefore

563: be more similar to the EGS lenses in terms of apparent magnitude and

564: geometric configuration.  In a future wide-field search, where more

565: SLACS-type lenses may be present, it would be prudent to retrain the

566: robot on a wider variety of lenses -- and perhaps on the SLACS sample

567: itself. %should more SLACS lenses be desired

568: 

569: %in particular, the source magnitudes are surprisingly (to the

570: %robot) bright.  Many main lenses seem also to have been misclassified

571: %as a result of the robot assigning large uncertainty to the Einstein

572: %radius, an effect which likely due to the unusual thickness (from the

573: %robot's point of view) of the arcs, or to the incomplete galaxy

574: %subtractions discussed earlier.  

575: 

576: \begin{figure*}

577: \centering\epsfig{file=f1.eps,width=0.9\linewidth}

578: \caption{Pr(\bf{d}\vline $H$=3) derived from the robot outputs of the EGS training set (smaller points), overlain with the outputs of the SLACS main lenses (larger points).  Points correspond to objects with \hon=3 and the contours in this and the subsequent two figures are 68\% and 95\% CL.. After \citet[]{Mar++08}.}

579: \label{fig:priors3}

580: \end{figure*}

581: 

582: \begin{figure} 

583: \centering\epsfig{file=f2.eps,width=0.9\linewidth}

584: \caption{Comparison between the SLACS-measured Einstein radii (SLACS-V) and

585: those calculated by the HAGGLeS robot.  The systems for which robotic lens

586: modeling failed are represented by filled points. The one-to-one relation is

587: also shown to guide the eye; we find an rms scatter of 12\%.

588:  Errors on SLACS Einstein Radii are 2\% (SLACS-V).} \label{fig:cfrrein} 

589: \end{figure}

590: 

591: Having detected a SLACS main lens, how accurately does the HAGGLeS

592: robot model it?  As an illustration of the robotic models'

593: performance, we compare the robot-calculated Einstein radii ($\thetaE$)

594: to those determined in the SLACS papers (Figure \ref{fig:cfrrein}).

595: The 12 open data points are main lenses that were not well-modeled by

596: the robot; after removing these we find an rms scatter of 12\%. For

597: 60\% of the main lenses, then, the robot finds not only a successful

598: lens model, but one that is in very good quantitative agreement with

599: that inferred during the SLACS project.

600: 

601: 

602: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

603: 

604: \subsection{Procedure 2}

605: 

606: When examining the BG subtraction residual images, we classified all

607: four of the new $H = 3$ systems as $H_2 > 1.5$ or above; the $H = 2$

608: system was missed.  Einstein rings are particularly easy to identify,

609: resulting in a high proportion of the main lenses found.  However, we

610: occasionally lost some image context: some residuals are more easily

611: identified as being caused by lensing when viewed along side of the

612: lens galaxy cutout. When looking at both cutouts and subtraction

613: residuals, we are able to ask the question ``Is this structure a part

614: of the lensing galaxy or is it unique?''  However, since sources are

615: typically bluer than the lens galaxy, color residuals can aid us in

616: answering this question: we note that all but one main lens was

617: classified as $H_2 = 3$ when multiple filters were available.

618: 

619: 

620: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

621: 

622: \subsection{Procedure 3}

623: 

624: We found that Procedure 3 was not an appropriate method to use when looking

625: for occurrences like the SLACS lenses.  This is predominantly an issue of image

626: intensity scale and dynamic range. The scales used by the robot are set for

627: fainter lens galaxies than those in the SLACS sample, and as such the main

628: lenses often appear saturated: the lensed features were completely washed

629: out in all but 8 cases.   We consequently classified only two objects as

630: $H_3 = 3$,  and this is the only procedure in which we falsely identified an

631: object as a lens.  This procedure was found to be most effective when lensed

632: images are blue against a red galaxy: five of the eight main lenses given

633: $H_3 > 1.5$ were in color fields.   The issue of intensity scaling was noted by

634: \citet{J08}, who attempted to optimize the intensity scale on an object by

635: object basis, and hence performed rather better in terms of ``true'' lenses

636: recovered (\citeauthor{J08} recovered $\sim50\%$ of the lenses identified by

637: \citeauthor{Fau++08}~\citeyear{Fau++08}). However, the large dynamic range in

638: surface brightness of the lens and source galaxies makes this a very difficult

639: task.

640: 

641: 

642: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

643: 

644: \subsection{Procedure 4}

645: 

646: In Procedure 4 we found three out of four new $H = 3$ lenses, and only two of

647: the main lenses were not classified as $H_4 > 1.5$.  The greatest strength of

648: this method lies in the ability to vary the intensity scale of the image so as

649: to make apparent any lensed structures.  It is effectively the same as looking

650: at both a cutout and a subtraction residual: lensing can be viewed in the

651: context of the host galaxy.  Its weaknesses, however, are in the time

652: required  and the quantity of data presented: it is difficult to conduct a

653: thorough, mistake-free search when every object must be examined and the time

654: required is $\sim$30 minutes per field.

655: 

656: 

657: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

658: 

659: \subsection{Considerations and comparisons}

660: 

661: We find that procedures one and two are nearly identical in their

662: ability to find new lenses, while the latter performed marginally

663: better on the main lenses since all the objects were examined.

664: Procedure 3 we find unsuited to the task at hand; looking at the robot

665: output (Procedure 1) or subtraction residuals (Procedure 2) largely

666: circumvents the dynamic range problems of image display, and produces

667: superior results.  Procedure 4 is the most inclusive of the four

668: procedures as all galaxies (not just bright galaxies) are examined; it

669: is also the only procedure in which we would have a chance of finding

670: a ``dark lens.''  However, we found two additional lensing systems

671: when we pre-selected bright galaxies and our search took significantly

672: less time.  We hence find that looking only at the BG subsample does

673: not decrease the completeness of a survey, and could in fact improve

674: it due to a lower error-rate.

675: 

676: Taking into consideration efficiency, completeness, and purity, we

677: recommend the use of Procedure 1 and Procedure 2. There will be cases,

678: such as with eight of the main lenses and one of our new lenses, where

679: a lens will be missed by the current robot and thus by the human

680: following Procedure 1.  There are also certain lenses, particularly

681: very unusual ones, that the current robot will miss due to its

682: inability to model it.  For example, the naked cusp configuration that

683: often occurs with edge-on spirals produces three images blended into

684: an arc but no counter-image; without a counter image the current robot

685: will classify the system as a class 0.  Additionally, when the

686: environment plays a strong role -- in over-dense environments for

687: instance -- the simple SIS+external shear model used now may not be

688: sufficient. It therefore may be preferential to use a combination of

689: procedures 1 and 2, inspecting \emph{all} objects modeled by the

690: robot regardless of the modeling outcome.

691: 

692: 

693: % ----------------------------------------------------------------------------

694: 

695: \section{Results}\label{sec:results}

696: 

697: We have discovered four new definite gravitational lenses and one

698: promising lens candidate. These five systems were each found in at least one

699: of the four separate search procedures. Having taken into

700: consideration all available data--including spectroscopy for one

701: system--we assigned four objects true human classification $H = 3$

702: (our four new lenses), and only one $H = 2$ (our best candidate).  The

703: five systems were also classified during each of the procedures.

704: Classifications for potential lenses are referred to as \hon, \htw,

705: \hth, or $H_4$ according to procedure number; these values may differ

706: from $H$. We present the systems along with their classifications in

707: Table \ref{tab:candclass}. For simplicity, each lens is given a short

708: name that will be used in the remainder of this paper.

709: 

710: \input{tab5.tex}

711: 

712: \begin{figure*}[t]

713: % \centering\input{figs/models.tex}

714: \centering\epsfig{file=f3.eps,width=0.9\linewidth}

715: \caption{Lenses and lens models for new systems created by the HAGGLeS

716: robot.  We show the galaxy cutout (column 1), the subtraction residual

717: (column 2), the masked robot input image (column 3), and the

718: reconstructed images from the minimum source

719: required to produce the configuration  (column 4). 

720: The robot uses Moffat profile subtractions by

721: default; here we have improved upon this by using a B-spline

722: subtraction \citep[]{Bol++06}.  Cutouts are 6 arcseconds on a side.}

723: \label{fig:lensmodels}

724: \end{figure*}

725: 

726: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

727: 

728: \subsection{Improved Robotic Lens Models}

729: 

730: As shown in Section~\ref{sec:accuracy}, and noted by \citet{Mar++08},

731: the lens model parameters returned by the HAGGLeS robot are not always

732: accurate. We find that the principal cause of robot modeling (and

733: indeed classification) error is insufficient lens light

734: subtraction. Disks and irregular profile slopes both give rise to

735: significant symmetrical residuals that confuse the robot. For the

736: small number of high quality lens candidates identified in

737: section~\ref{sec:results}, we can solve this problem on a case by case

738: basis, and provide the robot with cleaner images to model, and thus

739: produce more accurate estimates of the lens candidates' Einstein radii.

740: 

741: We improved the lens galaxy light subtraction with the flexible

742: B-spline fitting approach developed by \citet{Bol++06}, after first

743: masking out the objects in the image identified as candidate

744: lensed arcs. This procedure leaves sharper lensed image residuals. We

745: then set all undetected pixels to zero as described in

746: \citet{Mar++08}, but also at this stage masked out all the features in

747: the B-spline residual map not identified as lensed arcs. We then

748: re-ran the lens-modeling part of the HAGGLeS robot, and took as our

749: final estimated Einstein radius the position of the peak of the Gaussian fit

750: to the source plane flux curve, as described in \citet{Mar++08}.  The

751: resulting lens models and their parameters are shown in

752: Figure~\ref{fig:lensmodels} and Table~\ref{tab:lenspars}.

753: 

754: We can use our results from \S~\ref{subsec:proc1a} to estimate the

755: accuracy of the Einstein Radii measured by the robot. As shown in

756: Figure~ \ref{fig:cfrrein}, we found that -- for well modeled lenses

757: -- robot- and SLACS-measured Einstein radii agree to within

758: 10\%. However, this can be considered an upper limit to our true

759: uncertainty, since we use improved subtractions for the new

760: systems. In practice, the robot's estimates using improved

761: subtractions are as good as the relatively simple models allow them to

762: be. To be conservative we adopt an error of 5\% on the robot's

763: Einstein Radii.

764: 

765: \input{tab6.tex}

766: 

767: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

768: 

769: \subsection{Photometry}

770: 

771: Available from SDSS\footnote[1]{http://cas.sdss.org/dr6/en/} are

772: photometric redshifts and apparent lens galaxy magnitudes; with the

773: exception of one (Frenchie), no spectra are available. To supplement the

774: SDSS data, we use the {\tt galfit} software \citep{Pen++02} to fit de 

775: Vaucouleurs models

776: to the Hubble F814W data and derive apparent magnitude and

777: circularized effective radii, listed in

778: Table~\ref{tab:lenspars}. After correcting for Galactic extinction, we

779: calculate the rest frame V-band magnitude from the observed F814W, a

780: conversion for which there is very little scatter between different

781: spectral types.  Details on surface photometry and K-color corrections

782: can be found in the paper by \citet{Tre++01b}. For the main SLACS

783: lenses we use data from \citet[]{Bol++08a}.

784: 

785: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

786: 

787: \subsection{Velocity dispersion and source redshift}

788: 

789: In this Section we use available spectroscopy and photometry to

790: estimate the velocity dispersion and source redshifts of the new lens

791: systems, using two physically motivated assumptions: i) early-type

792: lens galaxies lie on the fundamental plane \citep[hereafter FP][]{Dre++87,D+D87,Tre++06,Bol++08b}; ii) the ratio between stellar

793: velocity dispersion and that of the best fitting SIS is approximately constant.

794: 

795: We use the photometric redshift and evolution-corrected V band

796: luminosities to estimate the central stellar velocity dispersion

797: predicted by the FP relation:

798: 

799: \begin{equation}

800: \log R_{\rm e}  = a \log \sigma_{\rm e,2} + b \log I_e + c,

801: \label{eq:FP}

802: \end{equation}

803: 

804: \noindent

805: where $\sigma_{\rm e,2}$ is the stellar velocity dispersion corrected

806: to an aperture of radius half the effective radius (in units of 100

807: kms$^{-1}$), I$_{\rm e}$ is the effective surface brightness in units

808: of 10$^9$ L$_{\odot}$ kpc$^{-2}$, and R$_{\rm e}$ is the effective

809: radius in kpc. We adopt the coefficients $a=1.28$ $b=-0.77$ $c=-0.09$

810: derived in SLACS-V for the SLACS sample. The intrinsic

811: scatter of the fundamental plane dominate the uncertainty on the

812: estimated $\sigma_{\rm e,2}$ (0.05 dex).  Central stellar velocity

813: dispersions $\sigstar$ are obtained from $\sigma_{\rm e,2}$ using the

814: standard correction described in SLACS-V.

815:       

816: Additionally, SLACS-IV and -V found that $\sigstar$ is

817: correlated with $\sigSIE$, the velocity dispersion that best fits

818: the model of the lens as a singular isothermal ellipse (SIE).  For the

819: SLACS sample:

820: 

821: \begin{equation}

822: \langle \sigstar / \sigSIE \rangle =1.02\pm0.01.

823: \label{eq:ff}

824: \end{equation}

825: 

826: We may also calculate $\sigSIE$, assuming the Einstein

827: radius $\thetaE$, lens redshift, and source redshift are known:

828: 

829: \begin{equation}

830: % \sigma_{\rm SIE} = \left(\frac{c}{{\rm km\ s}^{-1}}\right) \sqrt{\left(\frac{1}{4\pi}\frac{\theta_E}{{\rm radians}}\right)\frac{D_S}{D_{DS}}},

831: \sigSIE = c \sqrt{\frac{\thetaE}{4\pi}\frac{\Ds}{\Dds}},

832: \label{eq:sSIE}

833: \end{equation}

834: 

835: \noindent where $\Ds$ and~$\Dds$ are, respectively, angular-diameter 

836: distances to the source galaxy and between the lens and source

837: galaxies, and $\thetaE$ is given in radians.

838: 

839: We combine Equations~\ref{eq:FP} to~\ref{eq:sSIE} to determine our

840: best estimate of the source redshift. To obtain the posterior

841: probability distribution function of~$\zs$, we assume that $\sigstar$

842: is log-normally distributed with scatter 0.06~dex, which is dominated

843: by the intrinsic scatter of the FP and of Eq~\ref{eq:ff}. We adopt

844: priors appropriate for the source population.  For the newly

845: discovered -- imaging selected -- lenses, we adopt as prior the

846: redshift distribution of faint galaxies in single orbit ACS-F814W data

847: as measured by the COSMOS survey. For the main SLACS lenses we adopt

848: the same prior, but truncated at $\zs < 1.5$, i.e.\ the highest redshift where

849: [\ion{O}{2}] is still visible within the observed wavelength range

850: covered by the SDSS spectrograph used for discovery. The results

851: change very little if a uniform prior is adopted instead.

852: 

853: The result of this calculation for the main lenses can be seen in

854: Figures~\ref{fig:cfrvzs} and~\ref{fig:zsnew}. The former compares the

855: estimated source redshifts to the known source redshifts for the main

856: SLACS sample. This sanity check indicates that our procedure is unbiased

857: and that the scatter is consistent with the estimated error bars. The

858: latter figure shows the posterior probability distribution function

859: for the source redshifts of the newly discovered lenses. As expected,

860: the posterior is asymmetric with a tail to high-z due to to the strong

861: dependency on the ratio of angular diameter distances on the source

862: redshift, when it approaches the lens redshift.

863: 

864: The estimated stellar velocity dispersions and source redshifts are

865: given in Table \ref{tab:lenspars}.  We include the spectroscopically

866: measured $\sigma_*$ when available, for comparison.

867: 

868: \begin{figure}

869: \centering\epsfig{file=f4.eps,width=0.9\linewidth}

870: \caption{Source redshift estimated from the FP method vs. the corresponding spectroscopic source

871: redshifts for the SLACS lenses. }

872: \label{fig:cfrvzs}

873: \end{figure}

874: 

875: \begin{figure}

876: \centering\epsfig{file=f5.eps,width=0.9\linewidth}

877: \caption{Source redshift posterior probability distribution function estimated for the newly discovered lenses using the FP method. }

878: \label{fig:zsnew}

879: \end{figure}

880: 

881: % ----------------------------------------------------------------------------

882:  

883: \section{Discussion}\label{sec:discussion}

884: 

885: We begin this section by summarizing the arguments that led to our

886: classification of the five newly discovered lens systems in

887: \S~\ref{sec:validity}.  We then discuss the environments in which the

888: lenses are found in \S~\ref{sec:env}, and in \S~\ref{sec:compare} we

889: compare the properties of the new systems to the SLACS main lenses.

890: 

891: %  - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

892:  

893: \subsection{Validity and Properties of Candidates}\label{sec:validity}

894: 

895: Our final classification is entirely imaging-based (except in Frenchie's

896: case, where an SDSS spectrum was available).  

897: We thus rely entirely on lens geometry, the appearance of the 

898: subtraction residuals,

899: and the robot's ability to model the lens light. For this reason our

900: standards are high: we require clearly-identified multiple images in

901: all cases, and a straightforwardly-modeled 

902: image configuration. When available,

903: colors are used to strengthen the case. As an additional sanity check,

904: we note that the Einstein Radii are consistent with those expected

905: from the FP and a simple SIE model, for sensible values of source

906: redshifts (see Table~\ref{tab:lenspars}). We now discuss each case

907: individually.

908: 

909: Danny's lensing galaxy is a large red elliptical, while the lensed

910: source is blue, as expected for lens systems.   The identified quad geometry is a typical lens geometry

911: and may be called ``cusp dominated'' \citep[e.g.][]{Koc++06}, as it occurs when the source lies

912: near to a cusp of the inner caustic.  Although we also noted a strong

913: disk component remaining after the initial Moffat profile subtraction, the

914: robot is able to effectively model Danny as an SIS+external shear (a

915: situation in which such a geometry would occur).

916: 

917: Sandy has two strong arcs on either side of the lens galaxy, consistent

918: with a double pattern produced by a source almost directly behind a

919: SIS; this is well modeled by the robot.  

920: We note a fainter peak in surface brightness above the

921: lens galaxy; with no obvious counter image,

922: we believe this object is likely a small satellite galaxy in the lens plane.

923: 

924: The two images comprising Kenickie's lensing event, and the center

925: of the lensing galaxy, are not all perfectly aligned: this suggests that

926: either the source does not lie quite on the optical axis of an 

927: ellipsoidal lens or

928: that external shear is present.  We note that the inner

929: image appears to have more curvature than the outer; this could be 

930: brought about

931: by unsubtracted lens-plane structure that a color image would rule out.

932: 

933: Frenchie, another double, has been imaged in two filters and has colors 

934: consistent with lensing.  This is

935: the only one of our systems backed by spectroscopic data.  We find a

936: spectroscopic redshift of $\zd = 0.104$. Also, a stellar velocity

937: dispersion is available from SDSS, corresponding to

938: $\sigstar = 248\pm17$ km\ s$^{-1}$ after aperture correction, in good

939: agreement with the value estimated via the FP technique ($226\pm27$

940: km\ s$^{-1}$)

941: 

942: Rizzo is able to be modeled as a double; however the morphologies of the two

943: identifiable images are not as well-matched as in the previous 4 cases. In the

944: lensing scenario, the source would lie only partly within the outer caustic of

945: the lens, with the more extended outer image being only partially

946: strongly-lensed.

947: 

948: %  - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

949: 

950: \subsection{Lens environments}\label{sec:env}

951: 

952: We expect lenses, as massive galaxies, to be clustered. Consistent

953: with this hypothesis, 3/4 new lenses are found at redshifts very

954: similar to those of the main SLACS lens in the field. In this section

955: we study the environment of the fields with more than one lens to

956: identify possible large scale structures, using the environment

957: measures of local and global density as defined in SLACS-VIII.

958: 

959: SLACS-VIII found no significant difference between the environment of

960: SLACS lenses and that of non-lensing, but otherwise identical, galaxies.

961: With the exception of Frenchie, the environments of the newly

962: discovered lenses are fairly typical of the overall distribution of

963: SLACS lens environments, with the field of Kenickie being slightly

964: under-dense, while that of Sandy being somewhat over-dense. Frenchie, as

965: previously mentioned, lies in a very over-dense environment, the

966: densest of all SLACS fields.

967: 

968: %They define $\Sigma_{10}$ as the surface density

969: %of galaxies within the tenth nearest neighbor, and $D_1$ as the

970: %density of galaxies within a radius $h^{-1}$.  The ``local'' and

971: %``global'' over-densities are the normalized $\Sigma_{10}$ and $D_1$,

972: %respectively. Values determined for the main lenses of each our five

973: %systems are shown in Table \ref{tab:env} \citep[after][]{Tre++08}. The

974: %average local density for all SLACS lenses was determined to be

975: %2.90$\pm$1.20 while the average global density was 1.10$\pm$0.09.

976: 

977: In fact, both Frenchie and Sandy appear associated with known

978: clusters.  Frenchie lies within SDSS-C41035, at redshift 0.106, which

979: places both Frenchie ($\zd = 0.104$) and its main lens ($\zd = 0.106$) as

980: members.  The cluster in Sandy's field, MaxBCGJ217.49493+41.10435, is

981: at redshift 0.270, and the main lens at $\zd = 0.285$ lies within the

982: cluster.  We have only photometric redshift for Sandy ($\zd = 0.32\pm0.01$), 

983: which places it slightly beyond the extent of the

984: cluster; however given the possible systematic errors in 

985: the photometric redshift we

986: deem it likely that Sandy is also a member of the field's cluster.

987:  

988: \input{tab7.tex}

989: 

990: In conclusion, the results suggest that lens fields associate with

991: galaxy over-densities are the most likely to present additional strong

992: lens phenomena. Only $12/70 = 17\%$ of all SLACS lenses are associated with

993: known clusters, but $50\%$ of the newly discovered ones are (2/5 when

994: including

995: the likely candidate).  With the present data it is hard to

996: disentangle the contribution to the boost in strong lens surface

997: density by the enhanced surface density of deflectors.

998: %from the extra convergence made available by cluster and group scale halos.

999: 

1000: %  - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

1001:  

1002: \subsection{Comparison to SLACS lenses}\label{sec:compare}

1003: 

1004: We anticipated discovering lensing events similar to, but fainter

1005: than, the SLACS main lenses; we find this to be largely true.  In

1006: Figure~\ref{fig:plotiz}, we plot apparent $i'$ magnitude against

1007: redshift for the main lenses and new systems.  We find that four of

1008: the new systems are less luminous than the SLACS lenses at the same redshift,

1009: but that Frenchie is comparable.  We show the distribution of stellar

1010: velocity dispersions in Figure~\ref{fig:historsig}; new systems are

1011: shaded.  

1012: 

1013: To interpret these histograms we need to take into account

1014: the redshift dependence of the properties of the SLACS main sample. At

1015: the lower redshifts, the sample is dominated by the more abundant,

1016: slightly less massive galaxies. At higher redshifts, the flux limit of

1017: the SDSS spectroscopic database leaves only the most massive

1018: galaxies. Keeping this in mind, we also plot the distributions for the

1019: SLACS sample in the same redshift range of all the new lenses save Frenchie ($>0.26$ is chosen because of a natural break in the

1020: redshift distribution; the choice of threshold does not influence our

1021: conclusions). Indeed, the newly identified lenses are less

1022: massive than the SLACS main lenses when both samples are restricted to

1023: $\zd > 0.26$. The average $\sigma_*$ for the two samples are

1024: respectively $243\pm35$ km\ s$^{-1}$ and $299\pm20$ km\ s$^{-1}$.

1025: 

1026: \begin{figure}

1027: \centering\epsfig{file=f6.eps, width=0.9\linewidth}

1028: \caption{Plot of apparent i-band magnitude vs.\ lens redshift for the

1029: SLACS main lenses and new systems.  We find that four of the five new

1030: systems are less luminous than the main lenses.}

1031: \label{fig:plotiz}

1032: \end{figure}

1033: 

1034: % - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

1035: 

1036: \subsection{Lens abundance}

1037: 

1038: In Table \ref{tab:rates}, we compare lensing rates (or abundances, in

1039: degree$^{-2}$) for four HST imaging surveys, including this one. The

1040: quoted uncertainties on the inferred rates delimit the Bayesian 68\%

1041: confidence interval, assuming Poisson statistics and a uniform prior

1042: PDF.  As noted previously, classification systems across surveys are

1043: not consistent, thus one must impose similar criteria when comparing

1044: the results of different surveys; naturally there will be differences

1045: in opinion.  In this work we do not rely on spectroscopic data to

1046: classify a lens as ``definite'' and instead require that lens

1047: morphology, surface brightness, and color (if available) are clearly

1048: identifiable with a typical lens geometry.  Due to the limited data

1049: available, we apply these criteria rigorously.  With the goal of

1050: applying similar criteria to all surveys considered, we take as

1051: definite lenses: the literature-confirmed lenses for MDS, the

1052: ``unambiguous'' candidates for AEGIS, the ``best'' candidates for

1053: COSMOS, and the $H = 3$ lenses for this work.

1054: 

1055: The largest survey of ``blank'' sky undertaken to date is the

1056: COSMOS survey \citep{Fau++08}, whose findings imply a measured lensing

1057: rate of around $12$ lenses per square degree. Since our data are of

1058: comparable depth (and in the same single filter), we adopt this as our

1059: fiducial value.  Given this lensing rate, we might expect to find 1.61

1060: lenses, instead of our observed 4. Assuming a uniform prior on the

1061: lensing rate we find that the inferred lensing rate from our survey is

1062: $30^{+24}_{-8}$ lenses/degree$^2$.

1063: 

1064: The uniform prior is somewhat unrealistic -- it does not down-weight

1065: the occasional high lens yields that could arise as statistical flukes

1066: from the long-tailed Poisson likelihood. A maximally conservative approach 

1067: would be to take the

1068: COSMOS rate as the mean of an exponential prior; in this case, we infer a lensing rate of

1069: $18^{+14}_{-5}$ lenses/degree$^2$.  We still find a

1070: significantly higher lens abundance than seen in the COSMOS survey:

1071: with the COSMOS prior the probability that the lensing rate in the

1072: SLACS fields is greater than 12~degree$^{-2}$ is 88\%. Relaxing the 

1073: assumption that our fields are similar in nature to those in

1074: COSMOS and reverting to the uniform prior, 

1075: we find that there is only a $\sim2\%$ chance that the lensing rate

1076: in the SLACS fields is less than the COSMOS rate of 12 lenses/degree$^2$. 

1077: 

1078: %The uniform prior is somewhat unrealistic -- it does not down-weight

1079: %the occasional high lens yields that could arise as statistical flukes

1080: %from the long-tailed Poisson likelihood. It also ignores the

1081: %experience in the previous lens surveys, and all other astrophysical

1082: %information.  A maximally conservative approach would be to take the

1083: %COSMOS rate as the mean of an exponential prior (appropriate for

1084: %situations where the mean of the prior distribution is known, but

1085: %nothing else). In this case, we infer a lensing rate of

1086: %$18^{+14}_{-5}$ lenses/degree$^2$. Even refusing to acknowledge the

1087: %fact that elliptical galaxies are clustered, and assuming that our

1088: %fields are comparable to those in the COSMOS survey, we still infer a

1089: %significantly higher lens abundance than seen in the COSMOS survey:

1090: %with the COSMOS prior the probability that the lensing rate in the

1091: %SLACS fields is greater than 12~degree$^{-2}$ is 88\%.

1092: % The lensing rate for this 

1093: % work, at 30 lenses/degree$^2$, is the highest of the high resolution imaging  

1094: % surveys considered. 

1095: % We additionally find that,  

1096: 

1097: Although no definitive conclusions can be drawn due to the small numbers and

1098: variety of surveys, the results from this project do support the hypothesis of

1099: \citet{Fas++06}

1100: that looking near known lenses increases the efficiency of finding new

1101: lenses.  

1102: % 

1103: % Support for this idea also comes from the serendipitous discovery of

1104: % two new lenses within a single ACS pointing, in the field of the known lens

1105: % B1608 \citep[an area of $\sim$11 sq. arcmin][]{Fas++06}.

1106: 

1107: % We calculate a lensing rate of 720

1108: % lenses/degree$^2$ for this field, with an upper limit of 2130 (if both new

1109: % systems are lenses), and a lower limit of 490 (if only one new system is a

1110: % lens). 

1111: %Although the lensing rate in this one field certainly cannot be

1112: %directly compared to those from dedicated lens searches, we consider it

1113: %additional evidence for the clustering of lenses.

1114: 

1115: \begin{figure}

1116: \centering\epsfig{file=f7.eps, width=0.9\linewidth}

1117: \caption{Histogram of central velocity dispersions for main lenses and new systems; new systems are shaded.  Main lens velocity dispersion were aperture-corrected from SDSS photometry, while we used the FP relation to calculate the values for new lenses.}

1118: \label{fig:historsig}

1119: \end{figure}

1120: 

1121: %\begin{figure}

1122: %\centering\epsfig{file=figs/historein.eps, width=\linewidth}

1123: %\caption{Histogram of Einstein radii for main lenses and new lenses; new systems are shaded.  Einstein radii for main lenses are from the SLACS papers; those for the new systems were calculated by the robot.}

1124: %\label{fig:historein}

1125: %\end{figure}

1126: 

1127: \input{tab8.tex}

1128: 

1129: % ----------------------------------------------------------------------------

1130: 

1131: \section{Conclusions}\label{sec:conclusions}

1132: 

1133: We have performed a highly efficient search for gravitational lenses

1134: based purely on imaging data.  Our search area consisted of 44 HST/

1135: ACS fields, each centered on a SLACS ``definite'' lens, therefore

1136: exploiting the expected clustering of gravitational lenses in each

1137: field.  We compared the purity, completeness, and investigator time

1138: for four different search methods.  These methods are comprised of:

1139: (1) use of the output from the HAGGLeS robot, (2) inspection of BG

1140: subtraction residual images, (3) inspection of BG cutout images, (4)

1141: full frame visual inspection of ACS fields.

1142: 

1143: Our main conclusions are:

1144: 

1145: \begin{itemize}

1146: 

1147: \item Taking into account efficiency as well as completeness and purity, we

1148: find that of the methods used, procedures 1 (using output from the

1149: HAGGLeS robot) and 2 (looking at subtraction residuals of bright galaxies) have the best performance.  

1150: In situations where the simple SIS+external shear model

1151: used by the robot may be insufficient -- such as in clusters or to

1152: find naked cusp configurations -- it may be most effective to use a

1153: combination of procedures 1 and 2, in order to inspect all objects

1154: modeled by the robot.  However, looking at only the bright galaxies (BGs) in

1155: the fields

1156: did not decrease the completeness of this particular survey, while

1157: doing so greatly improved efficiency.

1158: 

1159: \item We discovered four new strong lenses and one promising candidate in the

1160: course of our survey.  We find that 3/4 of these new lens systems have

1161: lens redshifts similar to those of their main lenses; additionally,

1162: two of the new lenses are found in clusters of galaxies 

1163: that also include their respective main lenses.

1164: 

1165: \item We find that 3/4 new systems are less luminous and 

1166: less massive

1167: than the SLACS lenses.  Overall, we are probing early-type galaxies at

1168: higher redshifts and lower masses than the SLACS survey. For these

1169: comparisons, we used the data from the improved robot models, available

1170: photometry, and the Fundamental Plane to estimate the central velocity

1171: dispersions and source redshifts for each of our new systems.

1172: 

1173: \item The lens abundance for this survey, $30^{+24}_{-8}$ lenses/degree$^2$ 

1174: (uniform prior), is markedly higher than the lensing rates for the three other

1175: HST surveys considered at comparable depths and resolution.  Despite

1176: the small numbers and variations in search methods, this result

1177: supports the idea that searching near known lenses increases the yield

1178: of a lens survey.

1179: 

1180: \end{itemize}

1181: 

1182: The HAGGLeS project is currently using a combination of procedures 1

1183: and 2 to search for new lenses in the HST archive; through efforts

1184: such as this and others, we will be able to refine our lens search

1185: techniques for future surveys covering much larger areas of the sky.

1186: The use of efficient and repeatable lens search methods will further

1187: us towards the goal of having a large, homogenous sample of strong gravitational

1188: lenses. Such a sample will enable us to calculate a lens-lens correlation

1189: function and constrain the statistical properties of halos containing

1190: lens galaxies.

1191: 

1192: 

1193: % ----------------------------------------------------------------------------

1194: 

1195: \acknowledgments

1196: 

1197: This paper builds on the work of the SLACS and HAGGLES

1198: collaborations. We are grateful to our SLACS and HAGGLES collaborators

1199: and friends -- Adam Bolton, Roger Blandford, Maru\v{s}a Brada\v{c},

1200: Scott Burles, Chris Fassnacht, Raph\"{ae}l Gavazzi, David Hogg, Leon

1201: Koopmans, Leonidas Moustakas, Eric Morganson, and Tim Schrabback-Krahe

1202: -- for their many insightful comments and suggestions throughout this

1203: project.  Support for HST programs

1204: \#10174, \#10587, \#10886, \#10676, \#10494, and \#10798 was

1205: provided by NASA through a grant from the Space Telescope Science

1206: Institute, which is operated by the Association of Universities for

1207: Research in Astronomy, Inc., under NASA contract NAS 5-26555.  

1208: %

1209: E.R.N.\ acknowledges partial financial support from the College of Creative Studies.

1210: %

1211: The work of P.J.M.\  was

1212: supported by the TABASGO foundation in the form of

1213: a research fellowship.

1214: %

1215: T.T.\ acknowledges support from the NSF through CAREER award NSF-0642621, by

1216: the Sloan Foundation through a Sloan Research Fellowship, and by the

1217: Packard Foundation through a Packard Fellowship.  This research has

1218: made use of the NASA/IPAC Extragalactic Database (NED) which is

1219: operated by the Jet Propulsion Laboratory, California Institute of

1220: Technology, under contract with the National Aeronautics and Space

1221: Administration. This project would not have been feasible without the

1222: extensive and accurate database provided by the Digital Sloan Sky

1223: Survey (SDSS).  Funding for the creation and distribution of the SDSS

1224: Archive has been provided by the Alfred P. Sloan Foundation, the

1225: Participating Institutions, the National Aeronautics and Space

1226: Administration, the National Science Foundation, the U.S. Department

1227: of Energy, the Japanese Monbukagakusho, and the Max Planck

1228: Society. The SDSS Web site is http://www.sdss.org/.  The SDSS is

1229: managed by the Astrophysical Research Consortium (ARC) for the

1230: Participating Institutions. The Participating Institutions are The

1231: University of Chicago, Fermilab, the Institute for Advanced Study, the

1232: Japan Participation Group, The Johns Hopkins University, the Korean

1233: Scientist Group, Los Alamos National Laboratory, the

1234: Max-Planck-Institute for Astronomy (MPIA), the Max-Planck-Institute

1235: for Astrophysics (MPA), New Mexico State University, University of

1236: Pittsburgh, University of Portsmouth, Princeton University, the United

1237: States Naval Observatory, and the University of Washington.

1238: 

1239: % ----------------------------------------------------------------------------

1240: 

1241: \clearpage

1242: 

1243: \appendix

1244: 

1245: \section{Supplementary Figures and Tables}

1246: 

1247: In this Appendix we give a more complete illustration of the HAGGLeS robot's

1248: performance when given the SLACS main lenses. In Figure~\ref{fig:classA} we

1249: show all objects classified by the \emph{realistic} robot as $\hrr > 2.5$:

1250: \citet{Mar++08} found this to give a sample with $\sim20\%$~completeness  but

1251: $\sim100\%$ purity. In the SLACS fields, 10 objects were classified as

1252: $\hrr > 2.5$, including 4 main lenses. Comparing to the human-classified results

1253: using the same input data (Procedure 2, Table~\ref{tab:stats}), this

1254: represents a completeness of $~\sim11\%$ and a purity of $40\%$. Some

1255: explanation for these differences are given in the main text. 

1256: 

1257: The classification of the SLACS main lenses is listed in full in

1258: Table~\ref{tab:mainlenses}. The survey cutout images of these systems, sorted

1259: into bins in $\hro$, are shown in Figure~\ref{fig:mainlensesA} ($\hro > 2.5$), 

1260: Figure~\ref{fig:mainlensesB} ($1.5 < \hro < 2.5$), and 

1261: Figure~\ref{fig:mainlensesX} ($\hro < 1.5$). For each system, we show the full

1262: cutout image as presented for inspection in Procedure~3, the lens

1263: galaxy-subtracted cutout image as presented for inspection in Procedure~2, and

1264: the lensed images and counter-images predicted by the HAGGLeS robot's best

1265: lens model.

1266: 

1267: \begin{figure*}[t]

1268: % \input{figs/classA.tex}

1269: \centering\epsfig{file=f8.eps,width=0.9\linewidth}

1270: \caption{Objects classified by the robot as $\hrr > 2.5$, including four main 

1271: lenses (first and second rows).  Cutouts are 6 arcseconds on a side.  

1272: We find common false identifications to be rings left by the Moffat 

1273: subtraction, unmasked disks, and nearby galaxies.\label{fig:classA}}

1274: \end{figure*}

1275: 

1276: \input{tab9.tex}

1277: 

1278: \begin{figure*}[t]

1279: % \input{figs/mainlensesA.tex}

1280: \centering\epsfig{file=f9.eps,width=0.9\linewidth}

1281: \caption{Robot output for SLACS main lenses with $\hro > 2.5$, including: 

1282: galaxy cutouts (col. 1,4), residuals after Moffat subtraction (col. 2,5), 

1283: and predicted images from the robot lens models (col. 3,6). 

1284: In this figure and the two following, cutouts are 6 arcsec on a side.\label{fig:mainlensesA}}

1285: \end{figure*}

1286: 

1287: \begin{figure*}

1288: \centering\epsfig{file=f10.eps,width=0.9\linewidth}

1289: \caption{Robot output for SLACS main lenses with $1.5 < \hro < 2.5$, including 

1290: galaxy cutouts, subtraction residuals, and predicted images from the 

1291: robot lens models.\label{fig:mainlensesB}}

1292: \end{figure*}

1293: 

1294: 

1295: \begin{figure*}

1296: \centering\epsfig{file=f11.eps,width=0.9\linewidth}

1297: \caption{Continuation of Figure~\ref{fig:mainlensesB}. \label{fig:mainlensesBb}}

1298: \end{figure*}

1299: 

1300: \begin{figure*}[t]

1301: \centering\epsfig{file=f12.eps,width=0.9\linewidth}

1302: \caption{Robot output for SLACS main lenses with $\hro < 1.5$, including 

1303: galaxy cutouts, subtraction residuals, and predicted images 

1304: from the robot lens models.\label{fig:mainlensesX}}  

1305: \end{figure*}

1306: 

1307: % ----------------------------------------------------------------------------

1308: 

1309: \bibliographystyle{apj}

1310: 

1311: \begin{thebibliography}{34}

1312: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi

1313: 

1314: \bibitem[{{Alard}(2006)}]{Ala06}

1315: {Alard}, C. 2006, ArXiv Astrophysics e-prints

1316: 

1317: \bibitem[{{Belokurov} {et~al.}(2007){Belokurov}, {Evans}, {Moiseev}, {King},

1318:   {Hewett}, {Pettini}, {Wyrzykowski}, {McMahon}, {Smith}, {Gilmore}, {Sanchez},

1319:   {Udalski}, {Koposov}, {Zucker}, \& {Walcher}}]{Bel++07}

1320: {Belokurov}, V., {Evans}, N.~W., {Moiseev}, A., {King}, L.~J., {Hewett}, P.~C.,

1321:   {Pettini}, M., {Wyrzykowski}, L., {McMahon}, R.~G., {Smith}, M.~C.,

1322:   {Gilmore}, G., {Sanchez}, S.~F., {Udalski}, A., {Koposov}, S., {Zucker},

1323:   D.~B., \& {Walcher}, C.~J. 2007, \apjl, 671, L9

1324: 

1325: \bibitem[{{Bolton} {et~al.}(2008{\natexlab{a}}){Bolton}, {Burles}, {Koopmans},

1326:   {Treu}, {Gavazzi}, {Moustakas}, {Wayth}, \& {Schlegel}}]{Bol++08a}

1327: {Bolton}, A.~S., {Burles}, S., {Koopmans}, L.~V.~E., {Treu}, T., {Gavazzi}, R.,

1328:   {Moustakas}, L.~A., {Wayth}, R., \& {Schlegel}, D.~J. 2008{\natexlab{a}},

1329:   \apj, 682, 964

1330: 

1331: \bibitem[{{Bolton} {et~al.}(2006){Bolton}, {Burles}, {Koopmans}, {Treu}, \&

1332:   {Moustakas}}]{Bol++06}

1333: {Bolton}, A.~S., {Burles}, S., {Koopmans}, L.~V.~E., {Treu}, T., \&

1334:   {Moustakas}, L.~A. 2006, \apj, 638, 703

1335: 

1336: \bibitem[{{Bolton} {et~al.}(2008{\natexlab{b}}){Bolton}, {Treu}, {Koopmans},

1337:   {Gavazzi}, {Moustakas}, {Burles}, {Schlegel}, \& {Wayth}}]{Bol++08b}

1338: {Bolton}, A.~S., {Treu}, T., {Koopmans}, L.~V.~E., {Gavazzi}, R., {Moustakas},

1339:   L.~A., {Burles}, S., {Schlegel}, D.~J., \& {Wayth}, R. 2008{\natexlab{b}},

1340:   ArXiv e-prints, 805

1341: 

1342: \bibitem[{{Browne} {et~al.}(2003){Browne}, {Wilkinson}, {Jackson}, {Myers},

1343:   {Fassnacht}, {Koopmans}, {Marlow}, {Norbury}, {Rusin}, {Sykes}, {Biggs},

1344:   {Blandford}, {de Bruyn}, {Chae}, {Helbig}, {King}, {McKean}, {Pearson},

1345:   {Phillips}, {Readhead}, {Xanthopoulos}, \& {York}}]{Bro++03}

1346: {Browne}, I.~W.~A., {Wilkinson}, P.~N., {Jackson}, N.~J.~F., {Myers}, S.~T.,

1347:   {Fassnacht}, C.~D., {Koopmans}, L.~V.~E., {Marlow}, D.~R., {Norbury}, M.,

1348:   {Rusin}, D., {Sykes}, C.~M., {Biggs}, A.~D., {Blandford}, R.~D., {de Bruyn},

1349:   A.~G., {Chae}, K.-H., {Helbig}, P., {King}, L.~J., {McKean}, J.~P.,

1350:   {Pearson}, T.~J., {Phillips}, P.~M., {Readhead}, A.~C.~S., {Xanthopoulos},

1351:   E., \& {York}, T. 2003, \mnras, 341, 13

1352: 

1353: \bibitem[{{Cabanac} {et~al.}(2007){Cabanac}, {Alard}, {Dantel-Fort}, {Fort},

1354:   {Gavazzi}, {Gomez}, {Kneib}, {Le F{\`e}vre}, {Mellier}, {Pello}, {Soucail},

1355:   {Sygnet}, \& {Valls-Gabaud}}]{Cab++07}

1356: {Cabanac}, R.~A., {Alard}, C., {Dantel-Fort}, M., {Fort}, B., {Gavazzi}, R.,

1357:   {Gomez}, P., {Kneib}, J.~P., {Le F{\`e}vre}, O., {Mellier}, Y., {Pello}, R.,

1358:   {Soucail}, G., {Sygnet}, J.~F., \& {Valls-Gabaud}, D. 2007, \aap, 461, 813

1359: 

1360: \bibitem[{{Djorgovski} \& {Davis}(1987)}]{D+D87}

1361: {Djorgovski}, S., \& {Davis}, M. 1987, \apj, 313, 59

1362: 

1363: \bibitem[{{Dressler} {et~al.}(1987){Dressler}, {Lynden-Bell}, {Burstein},

1364:   {Davies}, {Faber}, {Terlevich}, \& {Wegner}}]{Dre++87}

1365: {Dressler}, A., {Lynden-Bell}, D., {Burstein}, D., {Davies}, R.~L., {Faber},

1366:   S.~M., {Terlevich}, R., \& {Wegner}, G. 1987, \apj, 313, 42

1367: 

1368: \bibitem[{{Estrada} {et~al.}(2007){Estrada}, {Annis}, {Diehl}, {Hall}, {Las},

1369:   {Lin}, {Makler}, {Merritt}, {Scarpine}, {Allam}, \& {Tucker}}]{Est++07}

1370: {Estrada}, J., {Annis}, J., {Diehl}, H.~T., {Hall}, P.~B., {Las}, T., {Lin},

1371:   H., {Makler}, M., {Merritt}, K.~W., {Scarpine}, V., {Allam}, S., \& {Tucker},

1372:   D. 2007, \apj, 660, 1176

1373: 

1374: \bibitem[{{Fassnacht} {et~al.}(2006){Fassnacht}, {McKean}, {Koopmans}, {Treu},

1375:   {Blandford}, {Auger}, {Jeltema}, {Lubin}, {Margoniner}, \&

1376:   {Wittman}}]{Fas++06}

1377: {Fassnacht}, C.~D., {McKean}, J.~P., {Koopmans}, L.~V.~E., {Treu}, T.,

1378:   {Blandford}, R.~D., {Auger}, M.~W., {Jeltema}, T.~E., {Lubin}, L.~M.,

1379:   {Margoniner}, V.~E., \& {Wittman}, D. 2006, \apj, 651, 667

1380: 

1381: \bibitem[{{Fassnacht} {et~al.}(2004){Fassnacht}, {Moustakas}, {Casertano},

1382:   {Ferguson}, {Lucas}, \& {Park}}]{Fas++04}

1383: {Fassnacht}, C.~D., {Moustakas}, L.~A., {Casertano}, S., {Ferguson}, H.~C.,

1384:   {Lucas}, R.~A., \& {Park}, Y. 2004, \apjl, 600, L155

1385: 

1386: \bibitem[{{Faure} {et~al.}(2008){Faure}, {Kneib}, {Covone}, {Tasca},

1387:   {Leauthaud}, {Capak}, {Jahnke}, {Smolcic}, {de la Torre}, {Ellis},

1388:   {Finoguenov}, {Koekemoer}, {Le Fevre}, {Massey}, {Mellier}, {Refregier},

1389:   {Rhodes}, {Scoville}, {Schinnerer}, {Taylor}, {Van Waerbeke}, \&

1390:   {Walcher}}]{Fau++08}

1391: {Faure}, C., {Kneib}, J.-P., {Covone}, G., {Tasca}, L., {Leauthaud}, A.,

1392:   {Capak}, P., {Jahnke}, K., {Smolcic}, V., {de la Torre}, S., {Ellis}, R.,

1393:   {Finoguenov}, A., {Koekemoer}, A., {Le Fevre}, O., {Massey}, R., {Mellier},

1394:   Y., {Refregier}, A., {Rhodes}, J., {Scoville}, N., {Schinnerer}, E.,

1395:   {Taylor}, J., {Van Waerbeke}, L., \& {Walcher}, J. 2008, \apjs, 176, 19

1396: 

1397: \bibitem[{{Gavazzi} {et~al.}(2008){Gavazzi}, {Treu}, {Koopmans}, {Bolton},

1398:   {Moustakas}, {Burles}, \& {Marshall}}]{Gav++08}

1399: {Gavazzi}, R., {Treu}, T., {Koopmans}, L.~V.~E., {Bolton}, A.~S., {Moustakas},

1400:   L.~A., {Burles}, S., \& {Marshall}, P.~J. 2008, ArXiv e-prints, 801

1401: 

1402: \bibitem[{{Gavazzi} {et~al.}(2007){Gavazzi}, {Treu}, {Rhodes}, {Koopmans},

1403:   {Bolton}, {Burles}, {Massey}, \& {Moustakas}}]{Gav++07}

1404: {Gavazzi}, R., {Treu}, T., {Rhodes}, J.~D., {Koopmans}, L.~V., {Bolton}, A.~S.,

1405:   {Burles}, S., {Massey}, R., \& {Moustakas}, L.~A. 2007, ArXiv Astrophysics

1406:   e-prints

1407: 

1408: \bibitem[{{Inada} {et~al.}(2008){Inada}, {Oguri}, {Becker}, {Shin}, {Richards},

1409:   {Hennawi}, {White}, {Pindor}, {Strauss}, {Kochanek}, {Johnston}, {Gregg},

1410:   {Kayo}, {Eisenstein}, {Hall}, {Castander}, {Clocchiatti}, {Anderson},

1411:   {Schneider}, {York}, {Lupton}, {Chiu}, {Kawano}, {Scranton}, {Frieman},

1412:   {Keeton}, {Morokuma}, {Rix}, {Turner}, {Burles}, {Brunner}, {Sheldon},

1413:   {Bahcall}, \& {Masataka}}]{Ina++08}

1414: {Inada}, N., {Oguri}, M., {Becker}, R.~H., {Shin}, M.-S., {Richards}, G.~T.,

1415:   {Hennawi}, J.~F., {White}, R.~L., {Pindor}, B., {Strauss}, M.~A., {Kochanek},

1416:   C.~S., {Johnston}, D.~E., {Gregg}, M.~D., {Kayo}, I., {Eisenstein}, D.,

1417:   {Hall}, P.~B., {Castander}, F.~J., {Clocchiatti}, A., {Anderson}, S.~F.,

1418:   {Schneider}, D.~P., {York}, D.~G., {Lupton}, R., {Chiu}, K., {Kawano}, Y.,

1419:   {Scranton}, R., {Frieman}, J.~A., {Keeton}, C.~R., {Morokuma}, T., {Rix},

1420:   H.-W., {Turner}, E.~L., {Burles}, S., {Brunner}, R.~J., {Sheldon}, E.~S.,

1421:   {Bahcall}, N.~A., \& {Masataka}, F. 2008, \aj, 135, 496

1422: 

1423: \bibitem[{{Jackson}(2008)}]{J08}

1424: {Jackson}, N. 2008, ArXiv e-prints, 806

1425: 

1426: \bibitem[{{Kochanek}(2006)}]{Koc++06}

1427: {Kochanek}, C.~S. 2006, in Gravitational Lensing: Strong, Weak \& Micro, ed.

1428:   G.~Meylan, P.~Jetzer, \& P.~North, Lecture Notes of the 33rd Saas-Fee

1429:   Advanced Course (Springer-Verlag: Berlin)

1430: 

1431: \bibitem[{{Koopmans} {et~al.}(2006){Koopmans}, {Treu}, {Bolton}, {Burles}, \&

1432:   {Moustakas}}]{Koo++06}

1433: {Koopmans}, L.~V.~E., {Treu}, T., {Bolton}, A.~S., {Burles}, S., \&

1434:   {Moustakas}, L.~A. 2006, \apj, 649, 599

1435: 

1436: \bibitem[{{Kubo} \& {Dell'Antonio}(2008)}]{K+D08}

1437: {Kubo}, J.~M., \& {Dell'Antonio}, I.~P. 2008, \mnras, 385, 918

1438: 

1439: \bibitem[{{Maoz} {et~al.}(1997){Maoz}, {Rix}, {Gal-Yam}, \& {Gould}}]{Mao++97}

1440: {Maoz}, D., {Rix}, H.-W., {Gal-Yam}, A., \& {Gould}, A. 1997, \apj, 486, 75

1441: 

1442: \bibitem[{{Marshall} {et~al.}(2005){Marshall}, {Blandford}, \& {Sako}}]{MBS05}

1443: {Marshall}, P., {Blandford}, R., \& {Sako}, M. 2005, New Astronomy Review, 49,

1444:   387

1445: 

1446: \bibitem[{{Marshall} {et~al.}(2008){Marshall}, {Hogg}, {Moustakas},

1447:   {Fassnacht}, {Bradac}, {Schrabback}, \& {Blandford}}]{Mar++08}

1448: {Marshall}, P.~J., {Hogg}, D.~W., {Moustakas}, L.~A., {Fassnacht}, C.~D.,

1449:   {Bradac}, M., {Schrabback}, T., \& {Blandford}, R.~D. 2008, ArXiv e-prints,

1450:   805

1451: 

1452: \bibitem[{{Moustakas} {et~al.}(2007){Moustakas}, {Marshall}, {Newman}, {Coil},

1453:   {Cooper}, {Davis}, {Fassnacht}, {Guhathakurta}, {Hopkins}, {Koekemoer},

1454:   {Konidaris}, {Lotz}, \& {Willmer}}]{Mou++07}

1455: {Moustakas}, L.~A., {Marshall}, P., {Newman}, J.~A., {Coil}, A.~L., {Cooper},

1456:   M.~C., {Davis}, M., {Fassnacht}, C.~D., {Guhathakurta}, P., {Hopkins}, A.,

1457:   {Koekemoer}, A., {Konidaris}, N.~P., {Lotz}, J.~M., \& {Willmer}, C.~N.~A.

1458:   2007, \apjl, 660, L31

1459: 

1460: \bibitem[{{Myers} {et~al.}(2003){Myers}, {Jackson}, {Browne}, {de Bruyn},

1461:   {Pearson}, {Readhead}, {Wilkinson}, {Biggs}, {Blandford}, {Fassnacht},

1462:   {Koopmans}, {Marlow}, {McKean}, {Norbury}, {Phillips}, {Rusin}, {Shepherd},

1463:   \& {Sykes}}]{Mye++03}

1464: {Myers}, S.~T., {Jackson}, N.~J., {Browne}, I.~W.~A., {de Bruyn}, A.~G.,

1465:   {Pearson}, T.~J., {Readhead}, A.~C.~S., {Wilkinson}, P.~N., {Biggs}, A.~D.,

1466:   {Blandford}, R.~D., {Fassnacht}, C.~D., {Koopmans}, L.~V.~E., {Marlow},

1467:   D.~R., {McKean}, J.~P., {Norbury}, M.~A., {Phillips}, P.~M., {Rusin}, D.,

1468:   {Shepherd}, M.~C., \& {Sykes}, C.~M. 2003, \mnras, 341, 1

1469: 

1470: \bibitem[{{Oguri} {et~al.}(2006){Oguri}, {Inada}, {Pindor}, {Strauss},

1471:   {Richards}, {Hennawi}, {Turner}, {Lupton}, {Schneider}, {Fukugita}, \&

1472:   {Brinkmann}}]{Ogu++06}

1473: {Oguri}, M., {Inada}, N., {Pindor}, B., {Strauss}, M.~A., {Richards}, G.~T.,

1474:   {Hennawi}, J.~F., {Turner}, E.~L., {Lupton}, R.~H., {Schneider}, D.~P.,

1475:   {Fukugita}, M., \& {Brinkmann}, J. 2006, \aj, 132, 999

1476: 

1477: \bibitem[{{Peng} {et~al.}(2002){Peng}, {Ho}, {Impey}, \& {Rix}}]{Pen++02}

1478: {Peng}, C.~Y., {Ho}, L.~C., {Impey}, C.~D., \& {Rix}, H.-W. 2002, \aj, 124, 266

1479: 

1480: \bibitem[{{Pindor} {et~al.}(2003){Pindor}, {Turner}, {Lupton}, \&

1481:   {Brinkmann}}]{Pin++03}

1482: {Pindor}, B., {Turner}, E.~L., {Lupton}, R.~H., \& {Brinkmann}, J. 2003, \aj,

1483:   125, 2325

1484: 

1485: \bibitem[{{Ratnatunga} {et~al.}(1999){Ratnatunga}, {Griffiths}, \&

1486:   {Ostrander}}]{Rat++99}

1487: {Ratnatunga}, K.~U., {Griffiths}, R.~E., \& {Ostrander}, E.~J. 1999, \aj, 117,

1488:   2010

1489: 

1490: \bibitem[{{Seidel} \& {Bartelmann}(2007)}]{S+B07}

1491: {Seidel}, G., \& {Bartelmann}, M. 2007, \aap, 472, 341

1492: 

1493: \bibitem[{{Treu} {et~al.}(2008){Treu}, {Gavazzi}, {Gorecki}, {Marshall},

1494:   {Koopmans}, {Bolton}, {Moustakas}, \& {Burles}}]{Tre++08}

1495: {Treu}, T., {Gavazzi}, R., {Gorecki}, A., {Marshall}, P.~J., {Koopmans},

1496:   L.~V.~E., {Bolton}, A.~S., {Moustakas}, L.~A., \& {Burles}, S. 2008, ApJ, in

1497:   press, 806

1498: 

1499: \bibitem[{{Treu} {et~al.}(2006){Treu}, {Koopmans}, {Bolton}, {Burles}, \&

1500:   {Moustakas}}]{Tre++06}

1501: {Treu}, T., {Koopmans}, L.~V., {Bolton}, A.~S., {Burles}, S., \& {Moustakas},

1502:   L.~A. 2006, \apj, 640, 662

1503: 

1504: \bibitem[{{Treu} {et~al.}(2001){Treu}, {Stiavelli}, {M{\o}ller}, {Casertano},

1505:   \& {Bertin}}]{Tre++01b}

1506: {Treu}, T., {Stiavelli}, M., {M{\o}ller}, P., {Casertano}, S., \& {Bertin}, G.

1507:   2001, \mnras, 326, 221

1508: 

1509: \bibitem[{{Turner} {et~al.}(1984){Turner}, {Ostriker}, \& {Gott}}]{TOG84}

1510: {Turner}, E.~L., {Ostriker}, J.~P., \& {Gott}, III, J.~R. 1984, \apj, 284, 1

1511: 

1512: \end{thebibliography}

1513: 

1514: 

1515: \end{document}

1516: 

1517: % ============================================================================

1518: 

1519: