OGLE-IV Real-Time Transient Search
L.Wyrzykowski, Z.Kostrzewa-Rutkowska, S.Kozlowski, A.Udalski, R.Poleski, J.Skowron, N.Blagorodnova, M.Kubiak, M.K. Szymanski, G.Pietrzynski, I.Soszynski, K.Ulaczyk, P.Pietrukowicz, P.Mroz
aa r X i v : . [ a s t r o - ph . GA ] S e p ACTA ASTRONOMICA
Vol. (2014) pp. 0–0 OGLE-IV Real-Time Transient Search
Ł. Wyrzykowski , , Z. Kostrzewa-Rutkowska , S. Kozłowski ,A. Udalski , R. Poleski , , J. Skowron , N. Blagorodnova , M. Kubiak ,M. K. Szyma´nski , G. Pietrzy´nski , , I. Soszy´nski , K. Ulaczyk ,P. Pietrukowicz , P. Mróz Warsaw University ObservatoryAl. Ujazdowskie 4, 00-478 Warszawa, Poland(lw,zkostrzewa,simkoz)@astrouw.edu.pl Institute of Astronomy, University of CambridgeMadingley Road, CB3 0HA Cambridge, UK Department of Astronomy, Ohio State University, 140 W. 18th Ave., Columbus, OH43210, USA Universidad de Concepción, Departamento de Astronomia, Casilla 160-C, Concepción,Chile
Received Month Day, Year
ABSTRACTWe present the design and first results of a real-time search for transients within the 650 sq. deg.area around the Magellanic Clouds, conducted as part of the OGLE-IV project and aimed at detect-ing supernovae, novae and other events. The average sampling of about 4 days from September toMay, yielded a detection of 238 transients in 2012/2013 and 2013/2014 seasons. The superb photo-metric and astrometric quality of the OGLE data allows for numerous applications of the discoveredtransients.We use this sample to prepare and train a Machine Learning-based automated classifier for earlylight curves, which distinguishes major classes of transients with more than 80% of correct answers.Spectroscopically classified 49 supernovae Type Ia are used to construct a Hubble Diagram withstatistical scatter of about 0.3 mag and fill the least populated region of the redshifts range in theUnion sample. We investigate the influence of host galaxy environments on supernovae statistics andfind the mean host extinction of A I =0.19 ± A V =0.39 ± Key words: surveys, supernovae, novae, transients ol. 64
1. Introduction
In the last decade wide-field instruments installed on medium-sized telescopeshave opened a new window in the time-domain astronomy. Hundreds of thousandsof new variable stars have not only been found ( e.g. , Bramich et al. et al. et al. et al. et al. (2014) found an RR Lyrae-type star, which overthe course of a few years, changed the mode of pulsation from double to single.Another remarkable example is a merger of a contact binary which resulted in aspectacular explosion (Tylenda et al. i.e. , transient events. Among the recent projectsaiming at unbiased large scale observations of large fractions of the sky are SDSS-Stripe82 (Sako et al. et al. et al. et al. e.g. , Wyrzykowski and Hodgkin 2012, Blagorodnova et al. e.g. , Riess et al. et al. et al. et al. e.g. , Phillips1993, Prieto et al. e.g. , mass of the host galaxy or its metallic-ity on the standardization process of the SNe ( e.g. , Childress et al. et al. et al. e.g. , Poznanski2009).Wide-field and long-term observations increase the number of supernovae of
A. A. well known types, but also increase chances for detecting rare and unusual exam-ples of supernovae. Exotic supernovae are being found in both the cores ( e.g. , Mattila et al. e.g. , Maguire et al. e.g. , Quimby et al. e.g. , Gezari et al.
2. Observations and detection pipeline
The Optical Gravitational Lensing Experiment (OGLE) has started in 1992 asone of the first generation microlensing surveys (OGLE-I: 1992-1995). Since 1997the OGLE survey started using a new 1.3 m Warsaw Telescope with a first gen-eration camera (for technical details see Udalski et al. et al. ×
2k pixel E2V detector with 15 µ m pixels, giving the 0.26 arcsec/pixel scaleat the focus of the Warsaw Telescope. OGLE-IV uses only two filters, Johnson-Cousin I - and V - bands, however, vast majority of observations are carried out inthe I filter.OGLE has always been among the largest variability surveys providing hun-dreds thousands variable objects of all types and various transients. These wereprimarily microlensing events, found in thousands every year toward the GalacticCenter, which are also used for finding extrasolar planets ( e.g. , Udalski et al. et al. e.g. , Wyrzykowski etal. e.g. , Skowron et al. et al. ol. 64 et al. (2012). Black-outlined are the fields searched for transients inKozłowski et al. (2013).other rare transient objects ( e.g. , Tylenda et al. via the Early Warning System (EWS, Udalski et al. et al. et al. et al. A. A.
Figure 2: OGLE-IV fields as in Fig.1 displaying positions of all transients discov-ered in years 2012-2014. Large points show all spectroscopically observed tran-sients, while small dots show the remaining unconfirmed objects, visually classifiedinto classes (See text and Table 1).search for on-going supernovae and other transients in the OGLE-IV data collectedin vicinity of the Magellanic Clouds. Fig. 1 shows the map of the OGLE-IVfields observed around the Large and Small Magellanic Clouds (LMC and SMC,respectively) and the Magellanic Bridge (MBR) and Fig. 2 displays all detectedtransients in years 2012–2014. Most of the 475 fields have been observed regularlysince 2010. 300 fields were processed in real-time since October 2012 (marked asgray in Fig. 1 and Fig. 2) and the remaining 175 were added to the pipeline inJune 2013.The observing season for the Magellanic Clouds System for the OGLE tele-scope runs from late July until March, i.e. , more than eight months, depending onthe region of the System. The region around the SMC is observed for longer periodof time that other regions, mostly due to lesser overlap with other OGLE programsin the Bulge and Galactic Disk. The mean cadence (Fig. 3) also varies throughthe season, depending of the region. The SMC and MBR sections are typicallyobserved with frequency as high as 2 days, whereas the LMC parts can only be ob-served with 5 days cadence at best. Note in Fig. 3 that the overall mean cadence hasdegraded slightly in 2013/2014 seasons, compared to 2012/2013, due to increase inthe covered area.Fig. 4 shows in its left panel a distribution of measured seeing on individual im- ol. 64
Figure 4: Left panel: distribution of seeing as measured for individual frames inseasons 2012/2013 and 2013/2014 in three different regions (LMC/MBR/SMC).The median seeing was about 1.4 arc sec for all observations. Right panel: dis-covery magnitude of OGLE-IV transients from seasons 2012/2013 and 2013/2014,indicating completeness by ∼
20 mag in I -band. A. A. ages used in the transient search in seasons 2012/2013 and 2013/2014. The medianseeing was about 1.4 arc sec. There was a significant number of of observations car-ried under superb seeing condition of about 1 arc second. The right panel of Fig.4 shows a distribution of discovery magnitude of all transients detected in seasons2012/2013 and 2013/2014. The completeness of detection reaches down to ∼ All the data reduction stages take place at the telescope site in near real-time.After de-biasing and flat-fielding, the data is processed by the OGLE real timephotometric pipeline (Udalski 2003) which uses the Difference Imaging Analysis(DIA) technique, fine-tuned to the OGLE data and using the Wo´zniak (2000) im-plementation of Alard and Lupton (1998) algorithm. The key element in the DIAmethod is a set of good quality reference images, which before each subtraction areconvolved to match a given image.The static database of objects is generated prior to the real-time processingand uses references images, which are obtained by stacking several high qualityimages obtained under excellent seeing conditions (better than 1 arc second). Forthe detection of stellar objects and determination of the reference image fluxes allthe reference images were analyzed with DoPhot (Schechter et al. i.e. , new sources.The search pipeline is run every day, after the data reduction of the previousnight is finished, typically before Chilean noon. Then, among 475 fields we selectthose which were observed last night and in those data we investigate new objects,returned by the subtraction pipeline. We select those subtraction residuals whichare of positive sign, i.e. , are caused by a brightening. The detection threshold for thebrightenings is relatively high to avoid a flood of artifacts. The match to Gaussianprofile of the PSF profile of brightening must exceed 0.7 (1.0 means perfect match)to classify it as a candidate new object.In order to assure robust detections and avoid numerous cosmic rays (note, weonly take one frame per field during each observing sequence), we also require thatthe residuals are present on at least two subsequent frames at the same location.This, therefore, naturally limits possibilities for very early detections of transientswhile they are still young. However, because our sampling is on average 2-5 days(see Fig. 3), our detectability time-frame is still relatively quick. ol. 64 i.e. , ignore all brighter objects) in order to avoid numerous variable stars, whichare typically brighter in the Magellanic Clouds.
For objects selected from both “new” and “old” channels, typically about a cou-ple thousand candidates, we generate small cutouts from the subtracted image takenat the brightest epoch. Those small imagettes are then fed into an automated im-age recognition classifier. The classifier is based on a Self-Organizing Maps (SOM)technique ( e.g. , Wyrzykowski and Belokurov 2008), trained on several thousands ofsmall thumbnail subtraction images. The resulting SOM has 5 × e.g. , Soszy´nski et al. (2002), Poleski et al. (2011). After the training, thecells were visually inspected and labeled according to their thumbnail. About 30%of cells contained images of good subtractions of new stellar-like objects. The re-maining were either bad subtractions, caused by misalignment of images, or effectsof high proper motion of stars. The SOM classifier usually reduced the number ofcandidates by a factor of two, removing the most obvious and common artifacts.However, many weirdly shaped artifacts or elongated and twisted cosmic rays stillremained as candidates for transients and those were removed at the final stage byvisual inspection of both the images and the light curves.We have to note here that the OGLE pipeline’s new objects selection criteria arequite stringent, requiring relatively good match to the Gaussian profile of the PSF.This results in a natural limitation on detected transients: for an image obtained un-der typical seeing condition of Las Campanas Observatory (about 1.2 arc secondsfor the OGLE telescope), the limiting detection magnitude is about 20.0 mag in the I -band. For images taken with exceptionally good seeing (less than 0.9 arc seconds)the detection threshold can go down to 21 mag. However, the OGLE telescope canreach mag ∼
22 in a 150 s exposure with reasonable signal-to-noise. Therefore, thesecond stage of the detection pipeline, which runs on manually selected candidatesfor transients (typically a couple per day) we perform a detailed difference imag-ing photometry, optimized to the known position of the transient. This allows usto obtain better quality photometry, as well as perform forced photometry of theflux of the transient when it is very faint. It also provides the timestamps of non-detection of the transient over all frames collected for that region, what allows toconfirm the transient nature of the detection. Because of the data and processinglimitations, we typically run the verification forced photometry of each transient on
A. A.
Figure 5: Visualization of nodes of the Self-Organizing Map trained on hundreds ofcut-outs from the subtracted images. The map had 5 × i.e. , more deepthan a single frame). Additionally, we perform the cross-match with the WISEinfrared catalog (Wright et al. et al. (2011) andKozłowski et al. (2012) for disentangling between AGNs and galaxies. Given allthe above information we make the final decision about the nature of the transient.If it is likely to be a supernova or a classical nova we give it a name followingthe pattern: OGLE-year-SN-number, or OGLE-year-NOVA-number, respectively.Possible dwarf nova detections or AGN activity are not reported on discovery, asthey are more persistent than other transients in their nature and, moreover, can bemore effectively searched for in the archival data set ( e.g. , Kozłowski et al. et al. et al. via theweb-site updated more or less daily: http://ogle.astrouw.edu.pl/ogle4/transients/ For each transient the photometry in the I - and V -band is provided as well as thefinding chart, subtracted image and the false-color reference image. The photom-etry available on the real-time active web-pages is roughly calibrated to within 0.2mag, and the calibrated light curves are obtained after the event is gone and isarchived. The OGLE-IV photometry is tied to precisely calibrated OGLE Photo- ol. 64 et al. , typically in batches once a week or less often, depending on the number ofnew detections.Table 1 provides information on all OGLE-IV transients found by the Tran-sients Detection System in real-time over the period of two seasons 2012/2013and 2013/2014, with their basic information. The table presents all transients inorder of discovery and contains the following columns: id of the transient (ID);internal OGLE-IV database id (DBID) in format field.chip.star number ; equato-rial coordinates RA J2000 . and DEC J2000 . ; discovery date as Julian date; discoverymagnitude in I -band; Astronomers Telegram number with the discovery (ATEL for the nearest galaxy; ATel num-ber with the spectroscopic classification (ATEL spec. R Ser (Offset ∗ , see Section 4); half-light radius of the host obtained in galfit model (seeSection 4); classification of the nearest galaxy using WISE colors after Assef et al. http://ogle.astrouw.edu.pl/ogle4/transients/archive2012-2014 Kozłowski et al. (2013) searched for supernovae and other transients in thearchival OGLE-IV data from years 2010–2012. They also provided the cumulativenumber of expected supernovae down to a given peak I -band magnitude using bestavailable rates for most SNe types. Comparing our real-time detections to thosepredictions, we estimate our detection efficiency to be about 50% down to 19 magand about 15% above 20 mag, what is naturally lower than the archival search(83% and 38%, respectively). The lower efficiency is expected, given the fact thatthe real-time search relies on much shorter light curves available at the time ofdetection, hence only the most robust objects are typically selected as transients inorder to keep the sample as pure as possible. http://ned.ipac.caltech.edu/ A . A . ID DBID RA
J2000 . DEC
J2000 . Discovery Disc. ATEL phot. prob. spec. z ATEL Offset Offset Offset ∗ R Ser
WISE commentOGLE-20.. HJD-2450000 I mag o l . ID DBID RA
J2000 . DEC
J2000 . Discovery Disc. ATEL phot. prob. spec. z ATEL Offset Offset Offset ∗ R Ser
WISE commentHJD-2450000 I mag A . A . ID DBID RA
J2000 . DEC
J2000 . Discovery Disc. ATEL phot. prob. spec. z ATEL Offset Offset Offset ∗ R Ser
WISE commentHJD-2450000 I mag o l . ID DBID RA
J2000 . DEC
J2000 . Discovery Disc. ATEL phot. prob. spec. z ATEL Offset Offset Offset ∗ R Ser
WISE commentHJD-2450000 I mag A . A . ID DBID RA
J2000 . DEC
J2000 . Discovery Disc. ATEL phot. prob. spec. z ATEL Offset Offset Offset ∗ R Ser
WISE commentHJD-2450000 I mag o l . ID DBID RA
J2000 . DEC
J2000 . Discovery Disc. ATEL phot. prob. spec. z ATEL Offset Offset Offset ∗ R Ser
WISE commentHJD-2450000 I mag A. A.3. Classification of transients
During the real-time detection in years 2012-2014 the transients were classi-fied only by a human observer and were split into two classes of candidates: super-novae or novae, typically relying on contextual information (primarily by checkingif there was a galaxy-like object nearby) and the observed amplitudes. More robustand detailed classification can be performed either via spectroscopy while the tran-sient is still on-going, or after the event is over and the full light evolution can beused for distinguishing transient classes. Below we describe those two channels.
OGLE-IV transients appeared on the web-site after being found in the last-night’s data (typically sooner than 12h after the last observation) and were imme-diately available for the astronomical community for the follow-up. Spectroscopicfollow-up was carried out mostly by the Public ESO Survey PESSTO (Valenti etal. et al. et al. et al. via WiseRep repository (Yaron et al. ol. 64
18 19 20 21 22 300 350 400 450 500 550 I [ m a g ] OGLE-2012-SN-050IIn 18 19 20 21 22 500 550 600 650OGLE-2013-SN-047IIL 18 19 20 21 22 550 600 650 I [ m a g ] HJD-2456000.0OGLE-2013-SN-104Ia 18 19 20 21 22 650 700 750HJD-2456000.0OGLE-2014-SN-004IIP
Figure 6: Examples of various spectroscopically determined supernova typesamong transients detected by the OGLE-IV Transients Detection System.8
A. A.
18 19 20 21 22 650 700 I [ m a g ] OGLE-2014-SN-001 15 16 17 18 19 700 750 800HJD-2456000.0OGLE-2014-NOVA-01 16 17 18 19 20 21 650 700OGLE-2013-NOVA-03 18 19 20 21 22 700 750 I [ m a g ] HJD-2456000.0OGLE-2014-SN-023 11 12 13 14 15 16 17 18 600 650 700 750HJD-2456000.0OGLE-2013-NOVA-02 16 17 18 19 20 21 22 550 600 650 700OGLE-2013-NOVA-01
Figure 7: Examples of Novae and Dwarf Novae detected by the OGLE-IV Tran-sients Detection System.
Thanks to its relatively high cadence, OGLE-IV provides well sampled lightcurves of supernovae and other transients in the I -band and more sparse lightcurves in V -band. Therefore, the data can reveal characteristic features allowingus to distinguish between different supernova classes and cataclysmic variables.In particular, obtaining an early preliminary classification of a transient candidatecould be helpful in allocating limited spectroscopic follow-up resources. Within theOGLE-IV Transients Detection System pipeline we typically detect transients nearor just after their maxima, and there are at least a couple of data points available, aswell as the time of the last non-detection. Here we present and test the performanceof an automated light curve classifier, which uses such incomplete early data as in-put. This classifier will be implemented in the processing and detection chain insubsequent observing seasons.In order to build a training set for the classifier we used both spectral classifi-cation and (somewhat subjective) visual classification of the full light curves andseparated our findings into five major classes of transients: supernovae Type Ia (Ia),core-collapse supernovae of Type IIn and IIp, dwarf novae (DNe) and classical no-vae (CNe). The visual inspection was carried out by multiple experienced observersand relied on identifying crucial features of each of the classified types ( e.g. , secondmaximum in SNe Type Ia), as well as contextual information ( e.g. , presence of a ol. 64 comment column. The visual classification wasattempted on the entire set of transients, however, still many objects remained clas-sified as unknown . Among those visually classified, we selected about a dozenfrom each class to construct the training set. We would like to emphasize here theadvantages of the OGLE data collected in the I -band, which allow to maintain highpurity of the photometric classification thanks to clear second maximum present inSNe Type Ia and well sampled light curves.Because there were only a couple of Dwarf Novae found by the Transient De-tection System (usually discarded during the detection process and not reported onthe web-site), for training we used a set of synthetic light curves, generated from alinear rise and exponential decline model of a DN outburst based on several dozensof DNe found in the OGLE-III data by Skowron et al. (2009). Several light curveswere excluded from the training set because they had no pre-maximum data or thelast non-detection date was unknown (for the very first transients). The training setfor the automated classifier comprised of 63 objects of Type Ia and II SNe, CNeand about a hundred of simulated DNe, adjusted to the OGLE-IV sampling. Eachlight curve in the training set was trimmed at its maximum to mimic its typical ap-pearance at the discovery epoch. For such data we computed the following set offeatures: • slope1- slope in mag/d before the max, • mag1- the maximum observed brightness, • time2max- time to reach the maximum from the last non-detection, • rise- difference in mag between the max and the detection level.We classified the I -band photometry using a Random Forest classifier (Breiman et al. . Random Forest (RF) takes asan input a set of values, which can represent any feature of the light curve. Wetrained the RF model on the training set and then classified all light curves withtrained model. The highest ranked (winner) class and its probability are provided inTable 1in the phot.class column, along with its probability ( prob . column). Note,the light curve classification was performed in the observer frame, however, thishad a negligible impact on the result as the redshifts of most of our extragalactictransients were ranging from z=0.05–0.15.There are in total 238 transients in our table, however, if we exclude all tran-sients with uncertain visual classification or with not enough data points before themaximum brightness, we are left with 196 objects. Further on, if we exclude objectsclassified outside of our five classes ( e.g. , AGNs), we are left with 143 classifiable Weka, version 3.6.5, developed at the University ofWaikato in New Zealand A. A. transients. Among those, 120 (84%) were assigned a class in agreement with thespectral and visual classifications. This is a very promising result, especially giventhe fact, that in some cases there was just 1 data point between a non-detection andthe maximum brightness. Overall, the performance of the classifier is good enoughto provide not only distinction between CVs and SNe, but also between thermonu-clear and core-collapse supernovae and their major subtypes. The sample of CNeused for training of the classifier should, however, be extended in future, with newdetections, but also possibly with the OGLE-IV data from the Galactic bulge (Mróz et al. e.g. , OGLE-2012-SN-006 classified spectroscop-ically as Ibn (Pastorello et al. in prep.), or OGLE-2013-SN-066 which was likelyan AGN flare, the classifier returned the winning class from the five trained classes,however, usually the broad probability distribution function (PDF) indicated theuncertainty of the classification. The highest value in the PDF and correspondingclass are shown in Table 1 for those objects.In the future, in order to increase the capabilities of the classifier, the trainingset should be expanded with more examples of cataclysmic variable outbursts andalso a wider variety of Type II supernovae. Nevertheless, we have shown thatapplying a very simple feature-based classification, we were able to reasonablywell reproduce the classification from spectra or full light curve inspection. Thisclassification schema will be implemented within the OGLE-IV pipeline in futureobserving seasons.
4. Supernovae Environments
The majority of the SNe from the OGLE sample are located within or close togalaxies. The host-galaxy properties and the supernova environment is a plausiblesource of inhomogeneity in SN properties ( e.g. , Maguire et al. etal. n free to vary using the gal f it software (Peng etal. R Ser . InFig. 8 we present examples of fitted galaxies (OGLE-IV reference image, the bestfit model, and the residuals). The positions of supernovae were derived using thesubtracted images from the DIA pipeline (which, by definition of the difference ol. 64 -20 -10 0 10 2020100-10-20 [ a rc s e c ] OGLE-2014-SN-014 -10 -10 0 10 2010100-10-20 -20 -10 0 10 2020100-10-20 − . − . . . . -10 -5 0 5 10[arcsec]1050-5-10 [ a rc s e c ] OGLE-2014-SN-006 -10 -5 0 5 10[arcsec]1050-5-10 -10 -5 0 5 10[arcsec]1050-5-10 − . − . . . . Figure 8: Examples of gal f it models of the light of the host galaxies. The leftmostpanel shows the original images from the OGLE-IV reference images, the middleare the models and the residuals are on the right. The scale of the residuals isnormalized to one sigma. The upper example shows a galaxy with spiral arms, notwell modeled by gal f it , whereas the lower example is a galaxy well approximatedby a single component ellipsoidal model.imaging, were registered on the same grid as the reference images), with the resid-uals fitted with the Gaussian profile. The angular offset between the supernova andits host’s nucleus is listed in Table 1, as well as the projected separation in kpc forcases with known distance ( via redshift). Fig. 11 shows the distribution of angu-lar distances for 148 transients with successful gal f it models in arc seconds. Wealso added a distribution of 110 transients classified as very likely supernovae inthe visual and spectroscopic classification. The distance distribution clearly showsthat OGLE-IV Transient Detection System program is finding most of its transientsnear or on top of the cores of the galaxies. This is achieved thanks to superb imagequality of the OGLE survey and a dedicated difference imaging software fine-tunedto the survey’s data.Galaxies vary in morphology and size, hence in order to obtain a more ho-mogenous picture of the distribution of supernova separations we normalized thedistance between a SN and its host using the Sérsic radius, taking into account theellipticity and orientation of the galaxy. Table 1 contains the Offset ∗ , which iscomputed across isophotes of the galaxy light profile, such that, e.g. , for an edge-on galaxy, a supernova located below or above the most of the light of the galaxy2 A. A. -20 -10 0 10 20-20-1001020 [ a rc s e c ] OGLE-2013-SN-032 -20 -10 0 10 20-20-1001020 -20 -10 0 10 20-20-1001020-20 -10 0 10 20-20-1001020 [ a rc s e c ] OGLE-2013-SN-004 -20 -10 0 10 20-20-1001020 -20 -10 0 10 20-20-1001020-20 -10 0 10 20[arcsec]-20-1001020 [ a rc s e c ] OGLE-2012-SN-051 -20 -10 0 10 20-20-1001020 -20 -10 0 10 20-20-1001020
Figure 9: Examples of spectroscopically confirmed supernovae separated by morethan 3 R ∗ SER from the host center. From left to right: reference image, maximumbrightness image, subtraction image. The red cross marks the centroid position onthe subtracted image and the blue cross is the center of the galaxy from the gal f it model. ol. 64 -20 -10 0 10 20-20-1001020 [ a rc s e c ] OGLE-2014-SN-024 -20 -10 0 10 20-20-1001020 -20 -10 0 10 20-20-1001020-20 -10 0 10 20-20-1001020 [ a rc s e c ] OGLE-2013-SN-126 -20 -10 0 10 20-20-1001020 -20 -10 0 10 20-20-1001020-20 -10 0 10 20[arcsec]-20-1001020 [ a rc s e c ] OGLE-2013-SN-124 -20 -10 0 10 20-20-1001020 -20 -10 0 10 20-20-1001020
Figure 10: Examples of apparently host-less supernovae with spectral confirmationas Type Ia. From the left to right: reference image, maximum brightness image,subtraction image. Red crosses mark the position of the centroid of the subtractedobject.will have its Offset ∗ larger than another supernova at the same angular distance,but located along the galaxy disk. Fig. 12 shows examples of spectroscopicallyconfirmed supernovae which were detected at a galactocentric distance larger than3 Sérsic radii computed across isophotes ( R ∗ SER ). Fig. 10 shows supernovae clas-sified as Type Ia for which the host was not found within 30 arc seconds fromthe transient. In most cases the host was probably too faint and was not detectedon OGLE reference images (depth ∼
22 mag, which corresponds to M I <-16 mag),however in some cases the host could have been located at a much larger separation( e.g. , OGLE-2013-SN-006).In the Fig. 11 we present the distribution of separations between the transient4 A. A.
Figure 11: Distribution of distance between the transient and the galaxy centre oflight in arcsec and OGLE-IV pixels (0.26 arcsec) - gray: all 148 models and blackline: 110 visual confirmed SNe. About half of the transients from 26 found withinradius of 1 pixel (0.26 arcsec) from galaxy center are most likely AGNs (based onspectra, previous variability or WISE colors), but we still find a significant numberof SNe in the galaxy centers. ser ]spectral Ia EllSab 0 1 2 3 4 5 6 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 Offset* [R ser ]spectral CC EllSab
Figure 12: Distribution of distance between the transient and the galaxy centre oflight normalized with Sérsic radius for: visually and spectroscopically confirmedType Ia and CC SNe. The transients are also divided by the host galaxy type - spiralor elliptical - based on WISE colors classification. ol. 64 et al. (2010) and Kozłowski et al. (2013). We notice that the majority of OGLE SNeare found in spiral galaxies. The distribution follows the light in galaxy and halfof the events lies within a single Sérsic radius. Deficit of supernovae below 0.5half-light radii is mostly caused by a selection bias of the spectroscopic follow-upobservations, which tend to avoid nuclear transients.
Inclination plays an important role in SN detection. Spiral galaxies with higherinclination have higher surface brightness and higher extinction along the line ofsight. Therefore, we should expect an observational bias toward finding more su-pernovae toward less inclined galaxies. We have used the galaxy short-to-long axisratio q = b / a obtained from gal fit to compute the galaxy inclination for the sampleof OGLE spiral galaxies. The usual Hubble (1926) formula has been used, adopt-ing the median short-to-long axis ratio g = .
22 derived by Unterborn and Ryden(2008) and the correction of +3 ◦ from Aaronson et. al. (1980). i = cos − "s(cid:18) q − g − g (cid:19) + ◦ (1)The resulting normalized distribution of OGLE SN hosts inclinations is shownin Fig. 13, along with the normalized distribution of a random selection fromthe overall galaxy population. We observe a gap in the low inclination regime( i.e. , face-on galaxies). This effect has already been observed in other surveys (Lea-man et. al. 2011) and it is associated with lack of precision when measuring themajor and minor axis. In the case of randomly oriented galaxy sample, we shouldexpect a uniform distribution in sin( i ); however, the precision issue makes thisassumption no longer valid.To analyze the distribution of inclination angles for the SN host population,removing any possible systematic bias associated with the uncertainty in the incli-nation angle, we compared two populations: the first containing SN host galaxiesand the second containing a random sample of galaxies ( i.e. , contained in the OGLEfield LMC571, as published in Soszy´nski et al. gal f it . In order to account for spiral galaxies only, we selected only thegalaxies with Sérsic index n < ( i ) for both populations is plotted in the right panel of Fig. 13. The figure,6 A. A.
Figure 13: Left: Histogram of computed inclinations for a SN host galaxy popu-lation (blue thick line), and histogram for a random sample of spiral galaxies (thinblack line). Right: Cumulative normalized distribution of sin(i) for both popula-tions.contrary to our expectations, shows an excess of objects with higher inclinations(approximately higher than 45 ◦ ) among the SN hosts, meaning that the edge-onorientation are more frequent among SN hosts. In order to quantify this effect, werun a Komolgorov-Smirnov two population test, which provides a p -value=0.1078.This result means that with a significance of 10% we can not rule out the null hy-pothesis that the two populations are identical. The conclusion is that there is no astatistically significant host galaxy bias in our sample. The main source of uncertainty in the galacto-centric distance measurementcomes from the galaxy light modeling. Positions of transients are derived from theDIA subtracted images and are typically known to better than a fraction of a pixel.In order to assess uncertainty of the nucleus position we used a sample of AGNs,which are located at the centers of galaxies. We selected the AGNs based on theirmid-IR colors as measured by WISE (Assef et al. gal f it . We only considerednearby AGNs ( z < .
15) where the host galaxy was clearly visible on the OGLEreference images. For the selected sample of 16 AGNs (see Fig. 14) the measuredoffsets in the x and y coordinates were smaller than 0.5 pixel (0.13 arc sec). Asthis verification method also included the uncertainty in the transient position asmeasured in difference images (DIA), we find that the overall error budget for our ol. 64 GALFIT -X DIA Y GALFIT -Y DIA
Figure 14: Difference in x and y coordinate between gal f it -measured center ofa galaxy hosting a central AGN and a DIA-based position of the residuals at themaximum brightness of the AGN. The sample comprises of 16 AGNs, for whichthe h D x i = − . ± . h D y i = − . ± .
379 pixels. 1 pixel is 0.26arc seconds.distance determination using gal f it and the DIA is about 0.13 arc seconds for atypical transient of 18-19 mag.
Interestingly, there seem to be at least a dozen of transients found within thecenters of their hosts (within 1 pixel of the center) which were not classified as su-pernovae, neither spectroscopically nor visually. Partially, the reason for the deficitof spectroscopically observed transients near the centers of galaxy is the naturalbias of the follow-up groups, which tend to avoid taking spectra near the centers ofgalaxies due to centering and contamination with host galaxy light. Nevertheless,the excess of central transients in Fig. 11 is clearly visible. Those are probablyfor the most part flares or other photometric activity of AGNs, with the most obvi-ous examples being OGLE13-088 and OGLE13-090, for which their AGN naturewas also confirmed with spectroscopy. We also used WISE color cuts to classifytransients as potential AGNs following the method of Assef et al. (2010), shown inTable 1 under WISE column. Some of the central transients, however, do not seemto correspond to an AGN identified within the nucleus of the host, e.g. , OGLE13-066 or OGLE13-033. Among those transients located within one pixel from thenucleus is one (OGLE13-071) with a light curve resembling that of a Tidal Disrup-tion Event (TDE, e.g. , Gezari et al.
A. A.
18 19 20 21 250 300 I [ m a g ] HJD-2456000.0OGLE-2013-SN-003 20 21 22 350 400HJD-2456000.0OGLE-2013-SN-056 19 20 21 550 600HJD-2456000.0OGLE-2013-SN-071
Figure 15: Nuclear transients of not obvious nature found within 1 pixel (0.26arcsec) from the cores of their hosts.conclude upon their nature with no spectroscopic follow-up, however, we can ex-pect similar transients being found in the OGLE data in the following seasons, andhope for real-time detections of TDEs and other exotic nuclear transients, allowingtheir detailed studies.
We note that among all OGLE transients, there were two pairs of supernovaewhich exploded in the same host galaxy. Such cases are important for studyingsupernovae environments and the supernova rates ( e.g. , Thöne et al. -20 -10 0 10 20[arcsec]-20-1001020 [ a rc s e c ] -20 -10 0 10 20[arcsec]-20-1001020 Figure 16: Supernova factories: galaxies with two supernovae found within one andtwo years, respectively. Left: Type IIn OGLE-2013-SN-017 (green) and type IbOGLE-2014-SN-014 (red). Right: OGLE-2011-SN-034 (green) (from Kozłowski et al. ol. 64 i.e. , 5.3 kpc at z=0.043. SN OGLE13-017 was classified as a TypeIIn (Inserra et al. et al. et al.
5. Cosmology with OGLE Supernovae
A key goal of most supernova surveys is detecting Type Ia supernovae, whichare known to be “standardizable candles”, and hence can be used for cosmologicalstudies of the expansion of the Universe ( e.g. , Riess et al. et al. et al. et al. z < .
14 with a median value of z=0.076.The absolute magnitude of the SN observed at magnitude m is described by theequation: M = m − µ − A MW − A H − K ( z ) , (2)where µ is the distance modulus, A MW and A H are the extinctions in the MilkyWay and in the SN host galaxy in the I -band, respectively. The single filter K -correction, which accounts for brightness difference due to redshifted spectrum,was adopted from the Magellanic Bridge supernova sample in the OGLE-IV (Kozłowski et al. L CDM cosmological model from the Planck mission withparameters: H =
68 km/s/Mpc, W M = . W L = .
69 (Planck collaboration2013). We took into account the Milky Way extinction toward each SN usingGalactic Extinction maps from Schlafly and Finkebeiner (2011) (retrieved via theNASA/IPAC Extragalactic Database). Due to lack of color light curves we were notable to fit the color for our light curves, hence the host galaxy extinction remainedas the only unknown parameter.The distance moduli were also derived from the light curve fitting and herewe used the empirical method for fitting multi-color light curves of Type Ia SNedescribed by Prieto et al. (2006). This method relies on the calibrated relationbetween the absolute magnitudes at maximum light and the post maximum declinerate D m (brightness change from maximum to 15 days post maximum) in BV RI A. A. I [ m a g ] [day]OGLE-2014-SN-016-2-1 0 1 2 0 20 40 60 80 V -I [ m a g ] I [ m a g ] [day]OGLE-2014-SN-002-2-1 0 1 2 0 20 40 60 80 V -I [ m a g ] I [ m a g ] [day]OGLE-2013-SN-148-2-1 0 1 2 0 20 40 60 80 V -I [ m a g ] I [ m a g ] [day]OGLE-2013-SN-098-2-1 0 1 2 0 20 40 60 80 V -I [ m a g ] Figure 17: Example light curves of Type Ia supernovae from OGLE-IV along withtheir models from Prieto et al. (2006) (black solid line). The time axis shows daysfrom the I -band maximum. The inset in each panel shows the available V − I datawith the model. Supernova OGLE-2013-SN-148 exhibits a very deep dip betweenthe two peaks and its model is poorly fitted. Also its color around maximum issignificantly redder than other supernovae, indicating a significant amount of ex-tinction.filters. By fitting the parameter D m and using the linear relation between theabsolute magnitude at maximum and the post maximum decline rate we obtainedthe absolute magnitude, and hence the distance moduli for our supernovae. Forfitting we used the well sampled I -band light curves and, where available, V -banddata. In Fig. 17 we show examples of OGLE Type Ia supernovae and the best fittedtemplate from Prieto et al. (2006).In Fig. 18 we present the Hubble Diagram for 49 Type Ia SNe from the OGLE ol. 64 D m (Prieto et al. L CDM cosmology model with Planck parameters(Planck collaboration, 2013) and obtained the scatter of 0.315 mag in residuals.Table 7 presents the results of the D m fits to the light curves of 49 Type Ia SNe inthe OGLE sample with the columns with D m value, the distance modulus derivedfrom D m , residuals on the Hubble diagram for the Planck cosmological models,the isophotal offset between a SN and the galaxy center (where the light model wasavailable) in units of the Sérsic radius.The Hubble Diagram residuals exhibit relatively low scatter of 0.315 mag, de-spite the fact that the results relied on single-band light curves. However, our lightcurves were in most cases well covered from before the maximum until the super-nova disappeared, allowing for good template fitting. The residuals exhibit clearly asystematic positive offset, which is expected, as we did not include any host galaxyextinction in the distance moduli calculations. Ignoring other more subtle effectson the residuals, like the host mass or metallicity, we can therefore use them to in-fer the amount of the host extinction for each of the supernovae. The most strikingoutlier is supernova OGLE13-148, which exhibits the most significant deviationfrom the expected brightness by more than 1 mag. Also, a single V − I measure-ment near the first peak indicates the host extinction was very large in this case, butthe finding chart shows that this supernova appeared on top of a well-pronouncedspiral arm of a large galaxy. On the other hand, as seen in Fig. 17, the model of thelight curve of OGLE13-148 matches well everywhere except for the dip betweenthe two I -band peaks. However, the significant dip by more than 1 mag is hard toexplain even by somewhat higher extinction, and this object requires more detailedstudies.In Fig. 19 we show the offsets between supernovae locations and their hostgalaxy centers against their residuals on the Hubble Diagram. We notice a sub-tle increase in the residuals while approaching the normalized center of the hostgalaxy, however, after about two half-light radii the residuals tend to agree withzero. Ignoring the larger scatter in the residuals below 0.1 R Ser and a few negativeresiduals, likely due to inaccuracies in galaxy core subtraction, we can see that inrange from 0.1 to 2 R Ser the systematic offset can be approximated as constant. As-suming it is all caused by the host extinction we can derive a mean value of the ex-tinction for this range of normalized galactocentric separations of A I = . ± . A I and A V from Rieke and Lebofsky (1985)we derive A V = . ± .
21 mag, in agreement with Galbany et al. (2012) whoobtained the mean value of A V = . ± .
02 mag for the SDSS sample of TypeIa SNe in spiral galaxies. They have also derived a linear dependence of extinctionon the projected distance from the host galaxy center, however, we do not see thistrend in our data.2
A. A. -0.5 0 0.5 1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 m - m L C D M z
32 34 36 38 40 m (H=68 km/s/Mpc, W M =0.31, W L =0.69) Figure 18: Top: Hubble diagram for the OGLE sample: the distance modulus–red-shift relation ( µ L CDM ) of the assumed L CDM cosmological model from Planck.Bottom: Residuals from the assumed L CDM cosmological model as a function ofredshift. Mean offset and its rms is 0 . ± . et al.
6. Summary
Since 2010, the OGLE-IV survey has been annually discovering ∼
150 tran-sients in the regions of the sky around the Magellanic Clouds. The OGLE-IV Tran-sient Detection System surveys 650 deg of the sky to a depth of I < z ∼ ol. 64 -1-0.8-0.6-0.4-0.2 0 0.2 0.4 0.6 0.8 0 0.5 1 1.5 2 2.5 3 3.5 4 m - m L C D M Offset* [R ser ] (H=68 km/s/Mpc, W M =0.31, W L =0.69) Figure 19: Residuals on the Hubble Diagram as a function of the offset betweenthe SN location and the host galaxy nucleus in units of Sérsic radius. We showthe results for 33 Type Ia SNe for which the hosts were detected and successfullymodeled with gal f it . The host extinction in the range between 0.1 and 2 half-lightradii computed from the mean residuals is A I = . ± .
10 mag (dashed line)assuming Planck cosmological parameters. Central transients (distance below 0.1)exhibit significant scatter in residuals most likely due to inaccuracies in galaxy coresubtraction.4
A. A.
Figure 20: Hubble Diagram with OGLE-IV (black dots) and Union 2.1 sample(gray) of supernovae Type Ia along with Planck (flat) and open cosmological model.OGLE-IV supernovae were corrected for extinction based on their distance fromthe host center. ol. 64 ID z A MW D m µ µ − µ L CDM
Offset*OGLE-20.. [mag] [mag] [mag] [R
Ser ]14-SN-024 0.1 0.244 0.839 38.403 0.02814-SN-021 0.039 0.113 1.183 36.556 0.321 1.4514-SN-019 0.04 0.113 1.260 35.905 -0.39 0.0914-SN-016 0.07 0.287 1.299 37.536 -0.01914-SN-010 0.056 0.109 1.247 37.649 0.604 0.0814-SN-002 0.10 0.063 1.289 38.114 -0.261 0.6813-SN-156 0.14 0.055 1.234 39.090 -0.074 0.0413-SN-148 0.043 0.150 1.368 37.841 1.38613-SN-147 0.099 0.113 1.281 38.016 -0.339 0.0113-SN-141 0.05 0.189 1.307 36.829 0.034 0.0513-SN-136 0.080 0.065 0.860 38.026 0.171 1.1913-SN-130 0.09 0.095 1.052 38.738 0.60313-SN-129 0.08 0.028 1.337 37.904 0.049 2.4013-SN-126 0.06 0.051 1.246 37.202 -0.00313-SN-124 0.13 0.065 1.191 38.785 -0.213-SN-123 0.08 0.078 1.457 37.139 -0.716 0.0813-SN-120 0.07 0.113 1.644 37.628 0.07313-SN-118 0.07 0.113 0.882 37.757 0.202 2.4813-SN-109 0.088 0.036 1.096 38.013 -0.062 0.2913-SN-099 0.028 0.478 0.953 35.812 0.317 0.5513-SN-098 0.06 0.044 1.068 37.347 0.142 0.3813-SN-096 0.11 0.087 1.158 38.478 -0.117 1.7413-SN-080 0.103 0.076 1.096 38.766 0.321 1.6713-SN-075 0.08 0.038 1.117 37.748 -0.107 1.3213-SN-073 0.091 0.057 1.257 38.151 -0.004 0.8913-SN-070 0.043 0.029 1.734 36.297 -0.15813-SN-057 0.10 0.032 1.265 38.163 -0.212 0.8813-SN-051 0.07 0.024 1.107 37.456 -0.099 0.9213-SN-044 0.07 0.077 1.069 37.581 0.026 0.5813-SN-043 0.14 0.031 1.025 39.354 0.189 1.2313-SN-041 0.10 0.034 1.192 38.176 -0.199 0.7213-SN-040 0.09 0.033 1.219 38.039 -0.095 1.0613-SN-039 0.08 0.056 1.132 38.007 0.152 0.8913-SN-032 0.09 0.035 1.097 38.083 -0.052 3.7613-SN-018 0.067 0.113 1.441 37.418 -0.037 0.1013-SN-015 0.088 0.051 1.714 37.819 -0.25713-SN-009 0.060 0.060 1.243 37.265 0.06 0.7913-SN-004 0.06 0.140 1.415 36.982 -0.223 3.0113-SN-001 0.09 0.026 1.266 37.559 -0.576 3.1112-SN-051 0.11 0.045 1.482 38.650 0.05512-SN-049 0.07 0.056 1.303 37.477 -0.078 2.7712-SN-046 0.111 0.050 1.182 38.480 -0.13512-SN-044 0.06 0.034 1.182 36.910 -0.29512-SN-040 0.014690 0.113 1.889 34.354 0.279 0.1912-SN-032 0.063 0.113 1.216 37.299 -0.01612-SN-014 0.059 0.113 1.069 37.312 0.14712-SN-009 0.013966 0.046 1.063 34.076 0.11112-SN-007 0.059438 0.103 1.293 37.139 -0.04612-SN-005 0.076 0.042 1.207 37.687 -0.058 1.89
Table 7: Parameters of the 49 Type Ia supernovae modeled using Prieto et al. (2006)templates. A MW is the Milky Way extinction in the I -band. D m is the parameterobtained from the template fits, µ is the derived distance modulus, µ − µ L CDM arethe HD residuals for different cosmologies and Offset ∗ is the galacto-centric dis-tance of a supernova in units of the Sersic radius (only for hosts with good gal f it models).6 A. A. tions.We presented the data for all 238 transients found in real-time in years 2012-2014, among which a significant fraction were classified spectroscopically. Super-novae Type Ia were used to construct the Hubble Diagram with relatively smallintrinsic scatter. Systematics in the residuals were also used to derive the meanvalue of host extinction in the range up to two half-light radii.Projected galactocentric distances of most of the transients were measured, andwe showed that OGLE-IV is capable of detecting nuclear transients with good ef-ficiency. We presented a few examples of interesting central transients of unknownnature.The OGLE-IV Transient Detection System is expected to continue in the future,extending the sample of well covered supernovae light curves and providing moreinteresting cases for detailed studies.
Acknowledgements.
We thank Drs M. Fraser, D. Poznanski, H. Campbellfor their continued support and help in preparation of this manuscript. We wouldlike to express our gratitude to the members of the spectroscopic follow-up groups,especially the PESSTO team, CSP (in particular to E. Y. Hsiao, G. H. Marion andM. Phillips) and M. Childress (ANU).We also would like to thank the fellows of the Polish Children’s Fund, JakubAniulis, Jakub Banaszak and Jakub Mnich, for their involvement in this work.This work made use of observations collected at the European Organisation forAstronomical Research in the Southern Hemisphere, Chile as part of PESSTO, (thePublic ESO Spectroscopic Survey for Transient Objects Survey) ESO program ID188.D-3003.This research has made use of the NASA/IPAC Extragalactic Database (NED)which is operated by the Jet Propulsion Laboratory, California Institute of Technol-ogy, under contract with the National Aeronautics and Space Administration.This publication makes use of data products from the Wide-field Infrared Sur-vey Explorer (WISE), which is a joint project of the University of California,Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology,funded by the National Aeronautics and Space Administration.The OGLE project has received funding from the European Research Coun-cil under the European Community’s Seventh Framework Programme (FP7/2007-2013)/ERC grant agreement no. 246678 to AU. This work has been supported bythe Polish Ministry of Science and Higher Education through the program “IdeasPlus” award No. IdP2002 000162 to IS.REFERENCES
Aaronson, M., Mould, J. and Huchra, J. 1980,
ApJ , , 655.Alard, C. and Lupton, R. H. 1998, ApJ , , 325.Arcavi, I. et al. ApJL , , L30. ol. 64 Assef, R. J. et al.
ApJ , , 970.Assef, R. J. et al. ApJ , , 56.Blagorodnova, N., Koposov, S. E., Wyrzykowski, Ł., Irwin, M. and Walton, N. A. 2014, MNRAS , , 327.Bramich, D. M. et al. MNRAS , , 887.Breiman, L., Last, M. and Rice, J. 2003, Statistical Challenges in Astronomy , , 243.Campbell, H. et al.
ApJ , , 88.Childress, M. et al. ApJ , , 108.Drake, A. J. et al. ApJ , , 870.Drake, A. J. et al. ApJS , , 9.Fraser, M. et al. MNRAS , , 1312.Galbany, L. et al. ApJ , , 125.Gezari, S. et al. Nature , , 217.Hadjiyska, E. et al. IAU Symposium , , 324.Hamuy, M. et al. PASP , , 266.Hubble, E. P. 1926, ApJ , , 321.Inserra, C. et al. ATel , , 1.Kozłowski, S., Kochanek, C. S. and Udalski, A. 2011, ApJS , , 22.Kozłowski, S. et al. ApJ , , 27.Kozłowski, S. et al. Acta Astron. , , 1.Law, N. M. et al. PASP , , 1395.Maguire, K. et al. MNRAS , , 747.Marion, G. H. et al. ATel , , 1.Mattila, S. et al. ApJ , , 111.Mowlavi, N. et al. arXiv , , 1.Mróz, P. et al. MNRAS , , 784.Ofek, E. O. et al. ApJ , , 104.Pan, Y.-C. et al. MNRAS , , 1391.Peng, C. Y., Ho, L. C., Impey, C. D. and Rix, H.-W. 2002, AJ , , 2.Perlmutter, S. et al. ApJ , , 565.Phillips, M. M. 1993, ApJL , , L105.Pietrukowicz, P. et al. Acta Astron. , , 379.Planck Collaboration 2013, ArXiv , , .Poleski, R., Soszy´nski, I., Udalski, A., Szyma´nski, M. K., Kubiak, M., Pietrzy´nski, G., Wyrzykowski,Ł. and Ulaczyk, K. 2011, Acta Astron. , , 199.Poleski, R. et al. ApJ , , 47.Poznanski, D. et al. ApJ , , 1067.Prieto, J. L., Rest, A. and Suntzeff, N. B. 2006, ApJ , , 501.Quimby, R. M. et al. Nature , , 487.Rieke, G. H. and Lebofsky, M. J. 1985, ApJ , , 618.Riess, A. G. et al. AJ , , 1009.Sako, M. et al. ApJ , , 162.Schechter, P. L., Mateo, M. and Saha, A. 1993, PASP , , 1342.Schlafly, E. F. and Finkbeiner, D. P. 2011, ApJ , , 103.Skowron, J., Wyrzykowski, Ł., Mao, S. and Jaroszy´nski, M 2009, MNRAS , , 999.Smartt, S. et al. ESO Messanger , , 50.Soszy´nski, I. et al. Acta Astron. , , 143.Soszy´nski, I. et al. Acta Astron. , , 219.Soszy´nski, I. et al. Acta Astron. , , 21.Soszy´nski, I. et al. Acta Astron. , , 1.Sullivan, M. et al. ApJ , , 102.Suzuki, N. et al. ApJ , , 85. A. A.
Thöne, C. C., Michałowski, M. J., Leloudas, G., Cox, N. L. J., Fynbo, J. P. U., Sollerman, J., Hjorth,J. and Vreeswijk, P. M. 2009,
ApJ , , 1307.Tylenda, R. et al. A&A , , A16.Udalski, A. and Ogle III Survey Team 2004, IAU Circ. , , 2.Udalski, A., Szymanski, M., Kaluzny, J., Kubiak, M., Mateo, M., Krzeminski, W. and Paczynski, B.1994, Acta Astron. , , 227.Udalski, A., Kubiak, M. and Szyma´nski, M. 1997, Acta Astron. , , 319.Udalski, A. et al. ApJL , , L109.Udalski, A., Szyma´nski, M. K., Szoszy´nski, I. and Poleski, R. 2008a, Acta Astron. , , 69.Udalski, A., Soszy´nski, I., Szyma´nski, M.K., Kubiak, M., Pietrzy´nski, G., Wyrzykowski, Ł.,Szewczyk, O., Ulaczyk, K., and Poleski. R. 2008b, Acta Astron. , , 89.Udalski, A., Soszy´nski, I., Szyma´nski, M.K., Kubiak, M., Pietrzy´nski, G., Wyrzykowski, Ł.,Szewczyk, O., Ulaczyk, K., and Poleski. R. 2008c, Acta Astron. , , 329.Udalski, A. 2003, Acta Astron. , , 291.Udalski, A. 2004, IAU Circ. , , 1.Unterborn, C. T. and Ryden, B. S. 2008, ApJ , , 976.Wozniak, P. R. 2000, Acta Astron. , , 421.Wright, E. L. et al. AJ , , 1868.Wyrzykowski, Ł. and Belokurov, V. 2008, American Institute of Physics Conference Series , ,201.Wyrzykowski, Ł. and Hodgkin, S. 2012, IAU Symposium , , 425.Wyrzykowski, Ł. et al. ArXiv ,1405.3134