The DES Bright Arcs Survey: Hundreds of Candidate Strongly Lensed Galaxy Systems from the Dark Energy Survey Science Verification and Year 1 Observations

Gravitational lensing occurs because the trajectory of photons from a distant object is deflected while passing through the gravitational field of a less-distant massive object along the line of sight with the observer. We call the more distant object a "source" and the less-distant object a "lens." If the source, the lens, and the observer are sufficiently separated and collinear, and if the mass of the lens is sufficiently large, then the apparent shape of the source can be noticeably distorted. Indeed, the source can appear as an extended arc or ring or even appear multiple times around the lens. This effect is called "strong" gravitational lensing.

Strong gravitational lens systems provide opportunities to study both astrophysics and cosmology. These systems provide an opening for studying properties of distant galaxies. Because the surface brightness of a source is unchanged during lensing, magnification of the source provides amplification of the image flux and allows studies of details that would otherwise be unresolved or too faint for ground-based investigation (e.g., Kostrzewa-Rutkowska et al. 2014). Sources with relatively large redshift are used for studies of star formation and metallicity in young galaxies (e.g., Bayliss et al. 2014). Studies of the lens systems, whether galaxies, groups, or clusters, provide information on their mass distribution, including the dark matter (Koopmans et al. 2009; Wiesner et al. 2012; Treu & Ellis 2014; Newman et al. 2015). Special cases of strong lens systems can be used to study cosmology. For instance, for lensed time-varying sources such as galaxies that contain quasars, the different appearances of the source may have differing times-of-flight and this information can be used (Refsdal 1964; Blandford & Narayan 1992) to extract the expansion history between source, lens, and observer (Schechter et al. 1997; Suyu et al. 2013, 2017; Birrer et al. 2016; Bonvin et al. 2017). Lens systems with multiple sources at differing redshifts can provide (Link & Pierce 1998; Gavazzi et al. 2008; Jullo et al. 2010; Collett & Auger 2014) complementary information (Collett et al. 2012) about the expansion history, independent of the Hubble constant.

While individual strong lens (SL) systems provide details of the characteristics of the lens and source objects, studies of statistically large samples of strong lensing systems have been considered as probes of the growth of structure and cosmology (Meneghetti et al. 2013). Realistic simulations (Li et al. 2016) of SL systems make it possible to compare (Xu et al. 2016) large samples with theoretical expectations. While the computations can easily generate O(10,000) or more simulated strong lensing systems, samples of more than a few dozen actual strong lens candidates from a single survey are scarce.

A number of automated methods for identifying strong lens candidates have been developed. These include a search for elongated objects (Alard 2006), an Arcfinder (Seidel & Bartelmann 2007) that identifies SL candidates associated with galaxy clusters or groups, analysis of third-order moments of galaxy shapes (Kubo & Dell'Antonio 2008) to find systems with arcs in the Deep Lens Survey (Wittman et al. 2006), principal component analysis (PCA) to identify (Joseph et al. 2014; Paraficz et al. 2016) SL systems with complete or nearly complete Einstein rings, and Deep Learning (Lanusse et al. 2017) and neural network (de Bom et al. 2017; Petrillo et al. 2017) analysis of galaxy shapes. Another new method, YattaLens (Sonnenfeld et al. 2017), identifies galaxy–galaxy lens candidates with arc-like features by modeling the source and lens galaxies and subtracting the lens galaxy from the image. Attaining large samples of real lenses with a variety of morphologies is important for vetting and testing the automated lens-finding algorithms, particularly for identification of group and cluster-scale SL systems. While these automated techniques show promise and could improve the statistical analysis of SL systems, traditionally productive searches have required labor-intensive techniques including visual scanning of candidate systems.

Wide-field surveys present a rich data sample in which to look for strong lens systems. The 1.64 deg² Hubble Space Telescope COSMOS survey field yielded 67 galaxy–galaxy lens candidates (Faure et al. 2008). The SDSS data yielded 19 confirmed systems to the Sloan Bright Arcs Survey (Allam et al. 2007; Diehl et al. 2009; Kubo et al. 2009, 2010; Lin et al. 2009), more than 30 confirmed and 50 additional candidate lenses to the CASSOWARY survey (Belokurov et al. 2009; Pettini et al. 2010; Stark et al. 2013), and 68 new galaxy clusters with giant arcs (Wen et al. 2011). The CFHTLS-Strong Lensing Legacy Survey (More et al. 2012) sample includes 54 systems with promising lenses, including 12 giant arcs, found in 150 deg² using the Arcfinder method. The Blanco Cosmology Survey yielded one serendipitous discovery (Buckley-Geer et al. 2011). Gavazzi et al. (2014) provides 49 confirmed strong lens systems identified using RingFinder on CFHTLS data. Crowdsourcing (Marshall et al. 2016) has led to discovery of 29 promising and 59 total (More et al. 2016) new strong lens systems in the CFHTLS data. Three different methods, including YattaLens, were used to search the Hyper Suprime-Cam Subaru Strategic Program (HSC SSP) images. The program (Sonnenfeld et al. 2017) yielded 333 candidates from an area of 442 deg². The HSC SSP sample is comparable in size and complementary to the result of this paper, as their candidates are principally galaxy–galaxy lenses with a small Einstein radius. Previous searches of the Dark Energy Survey data initially yielded six confirmed strongly lensed galaxies (Nord et al. 2016) in the early DES data, and more recently yielded eight more (B. Nord et al. 2017, in preparation), and four gravitationally lensed quasars (Agnello et al. 2015; Lin et al. 2017; Ostrovski et al. 2017). Other SL systems discovered using the DECam imager include the Canarias Einstein Ring (Bettinelli et al. 2016).

Searches of massive galaxy clusters have yielded many strong lenses. A search (Hennawi et al. 2008) of 240 massive galaxy clusters yielded 16 strong lens systems with >10'' radius and 21 additional SL candidates, where the lensing interpretation is based on the morphology of the systems. The South Pole Telescope identified (Reichardt et al. 2013) massive clusters using the Sunyaev–Zel'dovich effect (inverse Compton scattering of the cosmic microwave background radiation off hot electrons in the intergalactic medium within the cluster) (Sunyaev & Zel'dovich 1972) in a 2500 sq. deg. field that overlaps the same field that is presented in this paper. Many of the clusters have strong lens systems apparent in optical imaging follow-up observations (Staniszewski et al. 2009; Song et al. 2012; Aravena et al. 2013; Bleem et al. 2015). These are all compiled in one paper (Bleem et al. 2015). One of these was previously reported and studied in Buckley-Geer et al. (2011). Others were also found and reported (Menanteau et al. 2010a, 2012) in the ACT survey data.

The master lens database (L. Moustakas & J. Brownstein 2017, in preparation) lists⁴² 657 strong lens candidates, in three grades, to date. Ongoing and upcoming surveys will discover many more. Predictions for the number of lenses depend on the depth and area of the survey and range from a few thousand for the full Dark Energy Survey to more than a hundred thousand for near-future surveys (Oguri & Marshall 2010; Collett 2015).

In this paper we report the discovery of 348 previously unreported (and 26 additional) strong gravitational lens candidates from the Dark Energy "Science Verification" and "Year 1" data. This paper is organized as follows. In Section 2 we describe the Dark Energy Survey Science Verification and Year 1 observations and catalogs. In Section 3 we describe our strong gravitational lens search procedures. In Section 4 we describe the results from the searches and provide the properties of the candidate lens systems. We highlight some of the systems that have notable properties. Finally, in Section 5 we recapitulate the results and provide prospects for the analysis of the full DES wide-field.

The Dark Energy Survey is in the midst of imaging 5000 sq. deg. of the southern galactic cap using the Dark Energy Camera (DECam) (Flaugher et al. 2015), which is operated on the 4 m Victor M. Blanco Telescope at Cerro Tololo Interamerican Observatory (CTIO) near La Serena, Chile.

DECam installation was completed in 2012. There followed a period of commissioning the new instrument and recommissioning the telescope. Science verification (SV) spanned 79 nights or half-nights from 2012 November 1, to 2013 February 22. The main SV wide-field (WF) survey areas amounted to ∼250 sq. deg. at non-uniform depth and data quality. The median i-band limiting magnitude for extended⁴³ objects (10σ) was 23.0. In a subset of the area, amounting to about 150 sq. degs., the survey is more than a half magnitude deeper and comparable to what we expect in the final 5-year long survey (The Dark Energy Survey Collaboration 2016). This was accomplished by observing each part of those fields 10 times in each of the 5 filters: the g, r, i, z, and Y-bands. The exposure times varied with fields and were usually of 90 s duration, with most Y-band exposures taken with 45 s duration. The DES observing footprint, including the location of the SV fields, is described in The Dark Energy Survey Collaboration (2016) and is shown in Figure 1.

Zoom In Zoom Out Reset image size

The first full observing season, Year 1 (Y1), spanned 119 nights or half-nights from 2013 August 31 to 2014 February 9. The Y1 wide-field (WF) survey observations were concentrated in two areas: one of about 150 sq. deg. near the celestial equator that included a part of SDSS Stripe 82 (Annis et al. 2014), and a much larger region of roughly 1800 sq. deg. from −60° to −40° decl. that overlapped the area mapped in microwaves by the South Pole Telescope (Carlstrom et al. 2011). Generally, we observed those fields four times in each of the five filters: the g, r, i, z, and Y-bands. The exposures were of 90 s duration for the g through z bands and 45 s for the Y-band. The average FWHM of the point-spread function (PSF) for Y1 wide-survey exposures in the r, i, z bands was 0.94 arcsec, while the FWHM for the g, Y-bands was 1.17 arcsec. The i-band limiting magnitude for extended objects (10σ) was 22.9 (Drlica-Wagner et al. 2017). In addition to the wide-field survey, DES performed a time-domain ("supernova") survey during the same time period, visiting 10 fields in the g, r, i, and z-band filters with an approximately weekly cadence and at much greater depth (Kessler et al. 2015) than the wide-field survey. More details of the operations, data collection procedures, and observing results are available (Diehl et al. 2014). Figure 1 also shows the DES Y1 and SN fields.

The data were processed by the Dark Energy Survey Data Management (DESDM) system (Mohr et al. 2012; R. Gruendl et al. 2017, in preparation) in three pipelined stages: single-epoch "detrending," photometric calibration, and coaddition. The detrending operation removes the instrumental signature from the individual exposures. This includes corrections for cross-talk between amplifiers on the CCDs, subtraction of the bias, removal of the overscan and masking of "bad" pixels, application of a flat-field frame, an illumination correction determined on a CCD by CCD basis, a correction for the pupil ghost, a sky-background subtraction, and an artifact (cosmic ray) removal. Single-epoch catalogs were produced using PSFex (Bertin 2011) and SExtractor (Bertin & Arnouts 1996). Astrometric calibration is performed by matching bright stars on each exposure to reference stellar catalogs using scamp (Bertin 2006). Next, a photometric calibration is made. The "Global Calibration Module" starts with a list of exposures taken under photometric conditions (i.e., no extinction due to clouds or atmospheric dust), determines the magnitudes of many stars in each filter, and propagates that information across the many non-photometric overlapping exposures to determine a zero-point for each CCD in each exposure. Relative photometry of better than 2% rms accuracy was achieved. The relative photometric calibration was tied to an AB absolute system through targeted observations of bright spectrophotometric standards, again at about the 1%–2% level. Finally, the exposures in each filter were coadded using SWarp (Bertin et al. 2002) in 10,000 by 10,000 pixel "tiles" 072 on a side. SExtractor was then rerun on these coadded tiles to form catalogs of objects. A weighted combination of the coadded r + i + z "detection" tiles was used for identifying objects. The separation or "deblending" of closely positioned (or even overlapping) objects is a challenge, where the goal is to balance completeness against the spurious separation of features within a single galaxy. The deblending was performed using the detection images. The standard SExtractor 2.0 algorithm, which we used for the deblending, is not optimized for closely spaced lenses and sources or those in dense galaxy cluster cores (Zhang et al. 2014). The object catalogs contain the list of objects, their shapes, and their astrometric and photometric properties calculated from the coadd tile for each filter. Model magnitudes are fit to galaxies using a PSF derived from each coadd tile. Unless noted otherwise, the SExtractor MAG_AUTO magnitudes are the primary measures of coadd flux used in further analysis. There were typically 25,000 to 40,000 objects in the catalog of a full area tile.

The 580 sub-catalogs from SV are called "SVA1," where the A1 stands for "Annual Release #1." The 3778 sub-catalogs from Y1 are called "Y1A1." In both SVA1 and Y1A1 many of the sub-catalogs are made from incompletely observed tiles; these are typically from along the boundaries of the fields. The SVA1 catalog⁴⁴ contains 46M objects. The Y1A1 catalog contains 140M objects. Additional details about the Y1A1 WF processing and catalog can be found in Drlica-Wagner et al. (2017).

We applied several different techniques, described below, to search for SL systems using the SVA1 and Y1A1 data. The different techniques have some common elements. For each technique we created a list of potential lens systems. These lists were loaded onto the "DES Science Portal," a tool for visualizing the DES fields that can also provide catalog information about the objects. We used the Portal to produce small, 3-color (g, r, and i-band) cutout images, typically 55'' × 55'', centered on each of the systems. Several people, either scientists with experience identifying SL systems or students trained to do so, scanned pages of cutouts. At least two people scanned every cutout among the lists. Each page required 30 to 60 s to scan, depending on the speed of the scanner. Figure 2 shows a sample page of candidates as seen on the Portal. Potential SL candidates were identified by the occurrence of an apparent arc, or a pattern of arc-like knots or objects suggestive of an instance of strong lensing. It was not required that the potential sources or lenses that we identified were part of the selection that caused the cutout to be made in the first place. Interesting candidates were flagged for further evaluation. Some bright or particularly interesting candidates were immediately designated for further study. This initial process occurred over a period of about a year and a half.

Zoom In Zoom Out Reset image size

Each search uncovered unique, new systems, as well as some that eventually became very familiar, these having been "discovered" multiple times. Eventually, as further effort would lead to diminishing returns—mostly in the form of fainter systems, we stopped creating new searches. After we decided to terminate further searches, short lists of systems identified as candidates were compiled and all re-ranked, over a period of a few days, by a team of five scientists. Each person assigned a score of 0 to 2 to each system—0 points if the system was thought to not be a SL candidate, 1 point if it might be, and 2 points if the system was expected to be an instance of strong lensing. The maximum summed score that a system could attain was a 10. Systems with a total score of at least 3 were taken as the final list for this paper. The candidate rankings of 3 to 10 span the range from "possible" to "probable" to "definite" SL systems, with rankings consistent with those used in the Master Lens Database (L. Moustakas & J. Brownstein 2017, in preparation) and other graded samples of similar SL candidates.

We did not apply our search techniques to any samples of simulated strong lenses.

In this section we describe the four separate search procedures and the number of candidates that each produced.

3.1. A Search around Galaxy Clusters Identified by the South Pole Telescope

3.2. "Blue Near Anything Knot" Searches

We searched the SV and Y1 catalogs for SL candidates using a "Blue Near Anything" (BNA) algorithm, originally motivated in Kubik (2007). This algorithm aimed at identifying strong lensing of star-forming Lyman break galaxies and Lyα-emitting galaxies lensed by massive luminous red galaxies (LRGS).

We developed the BNA algorithm using the SV catalogs. The procedure was performed on a single coadd tile at a time and is illustrated in Figure 3. First a list of candidate lens galaxies is created. The criteria for a galaxy to be in the list of possible lenses are that at least one of the r-band, i-band, or z-band magnitudes is less than 21. Selection criteria on Sextractor outputs removed galaxies that were faint, objects that were not well deblended, objects that are likely to be stars, and artifacts left over from objects with saturated pixels. There were typically 4000 to 5000 candidate lens galaxies per tile. Next we formed a list of source candidates. The criteria for an object to be in the list of possible sources are that at least one of the the magnitudes for the g-band, r-band, or i-band must be less than 21, and that the object is not poorly deblended or contains saturated pixels. We did not make a star-galaxy separation because we wanted to preserve the possibility of identifying strongly lensed quasars for which the appearance of the knots are star-like (Reed et al. 2015). A color selection was applied to select blue source candidates; we required that g − r < 1.0 and that r − i < 1.0. There were typically 2000 to 3000 candidate source objects per tile. Next, for each object in the lens list we identified the objects in the source list that were within 8'' of the lens. Then we identified the largest set of those sources, associated with a given lens candidate object, that each had a similar color, where the "similar" requirement was that $| {\rm{\Delta }}(g-r)|$ and $| {\rm{\Delta }}(r-i)|$ both be less than 0.25 magnitudes. This corresponds to about three times the uncertainty in the color of a given source object at the faintest allowed magnitude. Although we do not impose color cuts on the lens selection, this algorithm predominantly finds blue-colored source galaxies lensed by red galaxies and galaxy clusters. For visual scanning, we kept any system that had two or more matched source objects. We refer to this as the BNA 2+ search. There were 11,539 such systems found in the 580 tile catalogs.

Zoom In Zoom Out Reset image size

For the SV data, we reprocessed the coadd tiles using the afterburner GAIN deblending technique (Zhang et al. 2014). The technique searches for blended sources that are not associated with the already cataloged objects. Searching considers image intensity peaks, image intensity gradient, and also image segmentation area. The photometry measurement is performed after evaluating the light contamination from neighboring sources. Though GAIN found only 1% more source and lens candidate objects than the DESDM algorithm, the number of lens candidates that were matched within 8'' of a source candidate was 4% higher and there were more source matches per lens candidate. Repeating the BNA algorithm resulted in the identification of 14,258 candidate lens systems. We removed candidate SL systems identified in the GAIN catalogs that were within 10'' of any candidate system identified in the DESDM catalogs. A total of 3281 candidate lens systems remained.

The BNA algorithm used for Y1A1 was similar to that used for SV. There were minor changes for the first pass through the data. The criteria to eliminate both artifacts from the list of lens and source candidates was strengthened. A total of 43,598 systems were identified that had at least two or more matched source objects. These were scanned for lens systems as described above. Later we ran the BNA algorithm again, this time with the source and lens object magnitude limits raised from 21.0 to 21.5. This time a total of 74,624 systems were identified. Removing any within 10'' of the previous list (of 43,598) left 31,964 candidate lens systems to scan. The combination of these two Y1A1 2+ knot searches yielded 211 candidates in the SL short list. Of these, 96 had a rank of 3 or above.

Having noticed that our BNA 2+ search was vulnerable to missing systems where only one source object had been identified, we implemented a search for "One Knot" lens candidates, referred to as BNA 1K. In order to leave a list of candidates that were short enough to scan, we applied more restrictive criteria to the lens and source object selection. In addition to the criteria listed for the BNA algorithm, we required that candidate lens objects contain at least one half of their i − band flux within a radius of <184 (7 pixels), have a ratio of the length of the major-to-minor axis <7, and that g −r > 0.7 and r −i > 0.3. The cuts on the flux radius and major-to-minor axis ratio remove artifacts such as diffraction spikes from stars, satellite trails, and deblended pieces of large nearby galaxies. Finally, we required that the magnitude for the r-band, i-band, or z-band be less than 20. These criteria restricted the lens list to bright red galaxies. The source list selection criteria was the same as in the BNA algorithm, but with the magnitude limited to objects brighter than 20.5 in the g, r, or i-bands. The maximum matching radius was reduced to 6''. Finally, we eliminated the fainter of any system that was within 10'' of any other system. The SV data yielded 35,012 candidates. This was reduced to 18,010 for scanning by requiring that the lens system be north of decl. = −60° to avoid the crowded Large Magellanic Cloud. There were 132,725 candidates in Y1A1 and it was not necessary to require that they were north of decl. = −60° because we had stayed away from the Large Magellanic Cloud during the Y1 observations. The BNA  1K searches added an additional 14 systems from SVA1 and 107 systems from Y1A1 to the short list. The other searches described here had not identified 81 of these. The combined SV plus Y1 BNA 1K searches yielded 75 systems with a rank of 3 or more. Of these, 36 were uniquely discovered by the BNA 1K search.

The combined BNA 2+ and BNA 1K searches formed a final BNA list of 153 SL candidates with a rank of 3 or more. Their mean rank was 4.9 (out of 10).

3.3. Search around redMaPPer Galaxy Clusters and redMaGiC Galaxies

3.4. "Red Near Anything Knot" Searches

The ranked lists from the various searches were combined. We identified 374 lens system candidates with a ranking of 3 or greater. A total of 348 are presented for the first time. Figure 4 contains a histogram of the rank for the systems that had a rank of 3 or more. We found some candidate systems in more than one search; with some systems being identified in every search that we performed. Figure 5 shows the Venn diagram of the systems indicating the overlap between the search techniques. Table 1 shows the number of objects searched, scanned, and found.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 2 provides the system name, the algorithms that identified the candidates for scanning, the rank as given by the experts, the distance (radius) of the source(s) from the presumed lens center, and other names for the system from previous references to it. Figures 6–16 show a 3-color cutout of each system. At the top of each image is a unique label, formed from the position, for each system. The most prominent galaxy, with the source(s) centered on it, is taken as the lens. The putative lens is centered in the image and is labeled with a letter "A." There are some systems where there is not a single galaxy to assign as the principal lens. For those systems we labeled the additional lensing objects with additional letters, e.g.,: "B," or "C." All of the lens objects are found in the DESDM SV and/or Y1A1 catalogs. Source objects are labeled on the cutout images with numbers, e.g.,: "1," "2," etc. Because it was not required that the potential sources or lenses that we identified be part of the selection that caused the system to be selected for scanning in the first place, some systems have no sources identified in the DESDM catalog. In addition, some objects may not have been in the catalog because of problems with the deblending noted in Section 2. Many of these missing sources are also identified by the number on the cutout so that the reason for the ranking is made apparent. Table 3 shows details, extracted from the catalog, for each object identified in each system for the first two pages (out of many) of systems. The full table is provided as a supplemental file, as is a copy of Table 2. These details include the identification mark on the cutout, the R.A., decl., g, r, i, z, and Y-band magnitudes (not corrected for Galactic extinction), and the photometric redshift (corrected for Galactic extinction). For those sources where there is no catalog information available, we supply the R.A. and decl. only.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Photometric redshifts (photo-z's) were computed using the "DESDM" artificial neural network method, as originally described in Oyaizu et al. (2008a, 2008b) and later vetted on DES data by Sánchez et al. (2014). The photometric redshift distributions for the lenses and for the sources are plotted in Figure 17. We expect that the lens photo-z's should be reasonably well-estimated, given that our lens samples consist predominantly of red galaxies, which have strong 4000 Å break features that yield better photo-z measurements. However, we caution that our sources, which are typically fainter blue objects, will have photo-z's that are subject to larger uncertainties. An important factor is that imperfect object deblending in these candidate lensing systems (where objects tend to be close in angular separation) will result in photometry errors that affect the photo-z measurements for the fainter sources more than those for the brighter lenses. Moreover, the bluer source galaxies have weaker spectral break features that will lead to larger photo-z errors, as well as possible catastrophic mistakes. Thus, the source photo-z distribution shown in Figure 17 may not be reliable. In particular, we see that the source photo-z distribution noticeably lies below the lens photo-z distribution at the lowest redshifts. While part of this may result from foreground objects contaminating our candidate source sample, it may also be due to catastrophic photo-z errors scattering true higher-redshift (z ≳ 1) blue source galaxies to erroneously low photo-z values.

Zoom In Zoom Out Reset image size

For each system, we measure an average radius of the source images, with respect to the primary lens. The uncertainty on the mean is drawn from the standard deviation on the mean, summed in quadrature with the pixel scale of DES, 0263. The pixel scale represents the resolution of DES images, which we use as a minimum uncertainty. The average radius of source images is an approximation for the Einstein radius, and is identical to that when the true source position is directly behind the lens. The image separation distribution is sensitive to a number of inputs such as the halo mass, the lens mass distribution, and the source redshift. It therefore contains information about the cosmological parameters and various scaling relations between galaxy properties and halo mass and can be measured from galaxy to cluster scales (Oguri 2006; More et al. 2016). Figure 18 is the distribution of the radii.

Zoom In Zoom Out Reset image size

In Section 3.1 we noted that the SPT Collaboration had identified (Bleem et al. 2015) 48 strong lens systems in the SPT data. We found 18 of those in the searches described here. DES did not observe in the locations of 14 of the SPT SL systems during SV or Y1. We do see evidence of strong lensing at the location of 3 of the SPT lenses that we did not identify as strong lensing systems in our searches. The sources, which appear very faint, did not pass the magnitude selection criteria, so we did not scan cutouts for those positions. For the 13 remaining SPT lenses, the DES images do not show any evidence of lensing. The sources are presumably too faint to be identified in the DES coadded data.

We cannot quantify the purity of our strong lens candidate sample because of the presently limited statistics of the follow-up results in B. Nord et al. (2017, in preparation), except to note that a few low-ranked systems (3 or 4), as well as a few higher-ranked systems, are already confirmed and that the higher-ranked systems are expected to have a higher purity than systems with lower rank.

4.1. Notable Systems

We report the results of several searches of the DES SV and Y1 imaging data for strong gravitational lens systems. These searches cover roughly 2000 sq.-deg. and used a combination of techniques. We searched the positions of known SPT and DES galaxy clusters, and we searched the DES catalogs for spatial matches of potential lens and source candidates. For all of the searches we produced a short list of candidates and then evaluated cutouts to identify the most promising systems based on color and morphology. A total of 388,017 cutouts were evaluated. We then assigned those a rank that quantifies our confidence, on those bases, that the system is a potential strong gravitational lens. We provide the R.A. and decl., the magnitudes and photometric properties of the lens and source objects, and the distance (radius) of the source(s) from the lens center for each system. Of the 374 that we found, 350 are presented for the first time. Some of these are striking systems with giant arcs. Some have red-colored sources. Two have both blue and red candidate sources at differing distance from the candidate lens. Using a Gemini Large and Long Program⁴⁵ over two years we have spectroscopically confirmed 13 of the systems presented here (Nord et al. 2016; B. Nord et al. 2017, in preparation; Collett et al. 2017; Lin et al. 2017), which is 3.5% of the sample. It is clear that the abundance of candidates means that within DES we are only able to follow-up a small number of the systems. Eventually, we expect most of these will be studied in more detail.

This large catalog of strong lens candidates, presented from a single search effort using uniform data, provides hundreds of ranked strong lensing candidate systems. We expect the variety of configurations will make it useful and valuable as a training set for future crowdsourced searches and future automated searches. This catalog also underscores the need for and importance of crowdsourced or automated lens modeling techniques (Birrer et al. 2015; Küng et al. 2015) being developed.⁴⁶

We have recently completed re-processing of the DES data from the first three observing seasons. This will add about 3000 additional sq.-degs. to be searched. We have therefore decided to release this catalogs of strong lens candidates from the DES SV and Y1 catalogs.

We are grateful for the extraordinary contributions of our CTIO colleagues and the DECam Construction, Commissioning and Science Verification teams in achieving the excellent instrument and telescope conditions that have made this work possible. The success of this project also relied critically on the expertise and dedication of the DES Data Management group.

Funding for the DES Projects has been provided by the U.S. Department of Energy, the U.S. National Science Foundation, the Ministry of Science and Education of Spain, the Science and Technology Facilities Council of the United Kingdom, the Higher Education Funding Council for England, the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign, the Kavli Institute of Cosmological Physics at the University of Chicago, the Center for Cosmology and Astro-Particle Physics at The Ohio State University, the Mitchell Institute for Fundamental Physics and Astronomy at Texas A&M University, Financiadora de Estudos e Projetos, Fundação Carlos Chagas Filho de Amparo à Pesquisa do Estado do Rio de Janeiro, Conselho Nacional de Desenvolvimento Científico e Tecnológico and the Ministério da Ciência, Tecnologia e Inovação, the Deutsche Forschungsgemeinschaft and the Collaborating Institutions in the Dark Energy Survey.

The Collaborating Institutions are Argonne National Laboratory, the University of California at Santa Cruz, the University of Cambridge, Centro de Investigaciones Energéticas, Medioambientales y Tecnológicas-Madrid, the University of Chicago, University College London, the DES-Brazil Consortium, the University of Edinburgh, the Eidgenössische Technische Hochschule (ETH) Zürich, Fermi National Accelerator Laboratory, the University of Illinois at Urbana-Champaign, the Institut de Ciències de l'Espai (IEEC/CSIC), the Institut de Física d'Altes Energies, Lawrence Berkeley National Laboratory, the Ludwig-Maximilians Universität München and the associated Excellence Cluster Universe, the University of Michigan, the National Optical Astronomy Observatory, the University of Nottingham, The Ohio State University, the University of Pennsylvania, the University of Portsmouth, SLAC National Accelerator Laboratory, Stanford University, the University of Sussex, Texas A&M University, and the OzDES Membership Consortium.

The DES data management system is supported by the National Science Foundation under grant No. AST-1138766. The DES participants from Spanish institutions are partially supported by MINECO under grants AYA2012-39559, ESP2013-48274, FPA2013-47986, and Centro de Excelencia Severo Ochoa SEV-2012-0234. Research leading to these results has received funding from the European Research Council under the European Unions Seventh Framework Programme (FP7/2007-2013), including ERC grant agreements 240672, 291329, and 306478.

We are also grateful to Phillips Academy students Aiden Driscoll, Charles Stacey, Ashley Scott, and Sabine Nix for their contributions to the scanning effort.