The Hubble Catalog of Variables
K. Sokolovsky, A. Bonanos, P. Gavras, M. Yang, D. Hatzidimitriou, M. I. Moretti, A. Karampelas, I. Bellas-Velidis, Z. Spetsieri, E. Pouliasis, I. Georgantopoulos, V. Charmandaris, K. Tsinganos, N. Laskaris, G. Kakaletris, A. Nota, D. Lennon, C. Arviset, B. Whitmore, T. Budavari, R. Downes, S. Lubow, A. Rest, L. Strolger, R. White
TThe
Hubble
Catalog of Variables K. Sokolovsky , , ,(cid:63) , A. Bonanos ,(cid:63)(cid:63) , P. Gavras , M. Yang , D. Hatzidimitriou , , M. I.
Moretti , , A. Karampelas , I. Bellas-Velidis , Z. Spetsieri , , E. Pouliasis , , I. Georgantopoulos , V. Charmandaris , K. Tsinganos , N. Laskaris , G. Kakaletris , A. Nota , , D. Lennon , C. Arviset , B. Whitmore , T. Budavari , R. Downes , S. Lubow , A. Rest , L. Strolger , and R. White IAASARS, National Observatory of Athens, Vas. Pavlou & I. Metaxa, 15236 Penteli, Greece Sternberg Astronomical Institute, Moscow State University, Universitetskii pr. 13, 119992 Moscow, Russia Astro Space Center of Lebedev Physical Institute, Profsoyuznaya Str. 84/32, 117997 Moscow, Russia Department of Physics, National and Kapodistrian University of Athens, 15771 Ilissia, Greece INAF-Osservatorio Astronomico di Capodimonte, Salita Moiariello, 16, 80131 Napoli, Italy Athena Research and Innovation Center, Artemidos 6 & Epidavrou, 15125 Maroussi, Greece Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA European Space Agency, Research and Scientific Support Department, Baltimore, MD 21218, USA European Space Astronomy Centre, Camino bajo del Castillo, Urbanizacion Villafranca del Castillo, Vil-lanueva de la Cañada, 28692 Madrid, Spain The Johns Hopkins University, Baltimore, MD 21218, USA
Abstract.
We aim to construct an exceptionally deep ( V (cid:46)
27) catalog of variable ob-jects in selected Galactic and extragalactic fields visited multiple times by the
HubbleSpace Telescope (HST). While HST observations of some of these fields were searchedfor specific types of variables before (most notably, the extragalactic Cepheids), we at-tempt a systematic study of the population of variable objects of all types at the magnituderange not easily accessible with ground-based telescopes. The variability timescales thatcan be probed range from hours to years depending on how often a particular field hasbeen visited. For source extraction and cross-matching of sources between visits we relyon the
Hubble
Source Catalog which includes 10 objects detected with WFPC2, ACS,and WFC3 HST instruments. The lightcurves extracted from the HSC are corrected forsystematic e ff ects by applying local zero-point corrections and are screened for bad mea-surements. For each lightcurve we compute variability indices sensitive to a broad rangeof variability types. The indices characterize the overall lightcurve scatter and smooth-ness. Candidate variables are selected as having variability index values significantlyhigher than expected for objects of similar brightness in the given set of observations.The Hubble
Catalog of Variables will be released in 2018.
The
Hubble Space Telescope (HST) is great for deep imaging thanks to the unique combination oflow sky background, sharp point spread function (PSF) and wide field of view (compared to ground- (cid:63) [email protected] (cid:63)(cid:63) [email protected] a r X i v : . [ a s t r o - ph . S R ] J u l ased adaptive optic systems; [13]). It is sensitive to ultraviolet light not accessible from the ground.Some fields were imaged by the HST multiple times opening a window to time-domain studies at faintmagnitudes and high spatial resolution. High resolution imaging is important to overcome confusionin dense star fields, like the ones found in nearby galaxies and globular cluster cores. A number ofdedicated variability studies were conducted with the HST including [1, 2, 5, 9, 10, 17, 25].While the HST has superb internal astrometric precision, obtaining accurate absolute astrometryhas been a challenge in the previous years. The positional information provided by the observatory’sattitude control system is limited by the position accuracy of individual Guide Star Catalog ([14]) stars.The Hubble
Source Catalog (HSC, [27]) solves this problem by cross-matching sources detected inindividual HST visits ([3]) and matching the brightest detected sources to deep reference catalogs:PanSTARRS, SDSS and 2MASS. The Gaia catalog will be used as reference in future HSC versions.The HSC provides access to a uniform reduction of the majority of publicly available imagesobtained with the WFPC2, ACS / WFC, WFC3 / UVIS, and WFC3 / IR instruments. The HSC is veryinhomogeneous: it includes 112 instrument-filter combinations; some filters are more popular thanothers. Most fields were observed only few times and the time between visits varies. The current HSCversion 2 has 89 fields containing 1 . ≥
25 times. The HSC is based on visit-combined imagesthat are deeper than individual exposures and mostly clean of cosmic rays. However, that comes at aprice: information about changes on timescales shorter than one visit (that may last a few hours) isaveraged-out and the resulting number of independent measurements is by a factor of a few smallerthan the number of HST exposures of the field.We aim to define a set of algorithms that will detect and validate candidate variable sources withinthe HSC, producing the
Hubble
Catalog of Variables (HCV). This is a work in progress that will leadto the release of the first version of the HCV in 2018. The initial project overview is given by [8].
It is common in optical photometry that measurement errors are not accurately known. The contribu-tion of random background variations and photon noise to the brightness measurements uncertaintycan be estimated easily for CCD observations. However, the hard-to-quantify residual systematice ff ects (pixel-to-pixel sensitivity variations, charge transfer ine ffi ciency, blending with nearby stars)limit the photometric accuracy for brighter objects. Some measurements get corrupted by cosmic rayhits, CCD cosmetic defects, incorrect background estimation near frame edges and other undetecteddata processing anomalies. Since the majority of field stars are not variable at a few per cent level, weutilize them to estimate typical photometric accuracy in a given dataset as a function of magnitude.Ground-based photometry was used by [22] to compare a number of variability indices that char-acterize “how variable” a lightcurve appears by quantifying its scatter and smoothness. We extendthis work by comparing the variability indices listed in Table 1 to simulations based on HSC data.We inject variability with random amplitude into HSC lightcurves of non-variable objects using thetechnique described by [22]. The simulations confirm that the interquartile range (IQR) of the mea-sured magnitudes, m i together with the inverse von Neumann ratio that characterizes the lightcurvesmoothness 1 /η = N (cid:80) i = ( m i − ¯ m ) / N − (cid:80) i = ( m i + − m i ) (where ¯ m is the mean magnitude and N is the numberof measurements) can recover a broad range of variability patterns and are robust against individualoutlier measurements. These indices do not depend on the estimated errors (Table 1) that may be un-reliable. Figure 1 presents IQR and 1 /η indices as a function of magnitude in one of the simulations.Figure 2 illustrates how the variability detection e ffi ciency of the indices changes as a function of thenumber of points in a lightcurve. I n t e r qu a r til e r a ng e - I Q R ( m a g ) F606W (mag)allvariableF cut-off at 8.1 σ / η F606W (mag)allvariableF cut-off at 6.3 σ Figure 1.
Variability indices characterizing scatter (left panel) and lightcurve smoothness (right panel) plotted asa function of magnitude. The indices are computed for HSC lightcurves of non-variable objects in the M31 halofield [2]. Crosses mark the objects in which artificial non-periodic variability was injected as described in [22].Selecting objects above the solid line as candidate variables results in the maximum F -score. Dotted line showsthe median value of a variability index as a function of magnitude. F m a x NSimulated variability type: psd σ IQR1/ η Figure 2.
The e ffi ciency of variable objects selectioncharacterized by the maximum (over all possiblecut-o ff thresholds) F -score as a function of the numberof lightcurve points, N , for three variability indices: σ ,IQR and 1 /η (see the references in Table 1). Based onsimulated non-periodic variability injected in HSClightcurves (M4 field, [17]). For the definition of F -score see https: // en.wikipedia.org / wiki / F1_score
Table 1.
Variability indices computed by the HCV pipeline.
Index Errors Ref. Index Errors Ref.
Indices quantifying lightcurve scatter time-weighted Stetson’s J time (cid:88) [7]reduced χ statistic – χ (cid:88) [4] clipped Stetson’s J clip (cid:88) [22]weighted std. deviation – σ (cid:88) [12] Stetson’s L index (cid:88) [24]median abs. deviation – MAD [29] time-weighted Stetson’s L time (cid:88) [7]interquartile range – IQR [22] clipped Stetson’s L clip (cid:88) [22]robust median stat. – RoMS (cid:88) [19] consec. same-sign dev. – Con. [28]norm. excess variance – σ (cid:88) [16] excursions – E x (cid:88) [18]norm. peak-to-peak amp. – v (cid:88) [21] autocorrelation – l [11] Indices quantifying lightcurve smoothness inv. von Neumann ratio – 1 /η [20]Welch-Stetson index – I (cid:88) [26] excess Abbe value – E A [15]Stetson’s J index (cid:88) [24] S B statistic (cid:88) [6] D ec . [ d e g ] R.A. [deg]
Figure 3.
Lightcurve pre-processing. Left: spatial distribution of HSC objects in the M4 field [17]. Highlightedare the objects used to determine local magnitude zero-point correction for the object marked with the cross.Right: lightcurve of that object before (top) and after (bottom) applying the local correction. The circles markidentified outlier points. The two lightcurves are shifted along the magnitude axis for clarity.
The HSC data are imported and grouped according to the observed field. The measurements of objectsnear frame edges or obtained with an old version of the image processing pipeline are excluded asunreliable. For each object all its measurements in a given filter that pass the above selection arecollected to construct a lightcurve. Each lightcurve is fitted with a straight line using robust regressionand outliers from the fit are identified. If a high percentage of measurements obtained during somevisit are identified as outliers in the corresponding lightcurves, all measurements associated withthis visit are discarded (bad image). For each of the remaining visits, for each object a local zero-point correction is computed as the mean di ff erence between the measured magnitudes and the onespredicted by the robust line fit for all objects within a specified radius around the object being corrected(Fig. 3). This should compensate for the residual large-scale sensitivity variations across the image.A set of corrected HSC lightcurves of objects observed in a given field with the same instrument-filter combination is the basic unit of variability search. For each lightcurve in a set we computevariability indices and select as candidate variables the objects that have variability index values sig-nificantly higher than the typical value for their brightness. Magnitude-dependent cuts in robust in-dices IQR and 1 /η are currently used to select candidate variables (Fig. 1). We are investigating waysto e ffi ciently combine information captured by all indices listed in Table 1 by means of the principalcomponent analysis and machine learning.The HCV pipeline is implemented in J ava and parallelized using the A pache S park framework.The variability indices implementation in the pipeline is consistent with their implementation in V a ST([23]) which is also used for lightcurve visualization and testing while a dedicated HCV visualizationsoftware is being developed. • The HCV is a catalog of variable objects derived from the HSC. It will be released in 2018. • The catalog is very heterogeneous due to the nature of the HSC dataset. It covers selected fields inthe Galaxy, the Local Group and beyond.
The HCV is very deep, it ventures into poorly explored region of variability parameter space. • Data pre-processing and variability detection techniques used for HCV are applicable to other multi-epoch surveys.
Acknowledgments : This work is supported by ESA under contract No. 4000112940.
References [1] Bernard E. J., et al., 2013, MNRAS, 432, 3047[2] Brown T. M., Ferguson H. C., Smith E., Kimble R. A., Sweigart A. V., Renzini A., Rich R. M.,2004, AJ, 127, 2738[3] Budavári T., Lubow S. H., 2012, ApJ, 761, 188[4] de Diego J. A., 2010, AJ, 139, 1269[5] Dolphin A. E., et al., 2001, ApJ, 550, 554[6] Figuera Jaimes R., Arellano Ferro A., Bramich D. M., Giridhar S., Kuppuswamy K., 2013, A&A,556, A20[7] Fruth T., et al., 2012, AJ, 143, 140[8] Gavras P., et al., 2017, arXiv:1703.00258[9] Ho ff mann S. L., et al., 2016, ApJ, 830, 10[10] Je ff ery E. J., et al., 2011, AJ, 141, 171[11] Kim D.-W., Protopapas P., Alcock C., Byun Y.-I., Khardon R., 2011, ASPC, 442, 447[12] Kolesnikova D. M., Sat L. A., Sokolovsky K. V., Antipin S. V., Samus N. N., 2008, AcA, 58,279[13] Lanzerotti L. J., Assessment of options for extending the life of the
Hubble Space Telescope :final report:final report