1: %\documentclass[12pt,preprint]{aastex}
2: \documentclass[apjl]{emulateapj} \usepackage{apjfonts}
3:
4: \shortauthors{Darnley et al.} \shorttitle{The Angstrom Project Alert
5: System}
6:
7: \begin{document}
8:
9: \title{The Angstrom Project Alert System: real-time detection of
10: extragalactic microlensing}
11:
12: \author{M.J.~Darnley\altaffilmark{1}, E.~Kerins\altaffilmark{1,2},
13: A.~Newsam\altaffilmark{1},
14: J.P.~Duke\altaffilmark{1}, A.~Gould\altaffilmark{3},
15: C.~Han\altaffilmark{4}, M.A.~Ibrahimov\altaffilmark{5},
16: M.~Im\altaffilmark{6}, Y.-B.~Jeon\altaffilmark{7},
17: R.G.~Karimov\altaffilmark{5}, C.-U.~Lee\altaffilmark{7},
18: B.-G.~Park\altaffilmark{7} }
19: \altaffiltext{1}{Astrophysics Research Institute, Liverpool John
20: Moores University, Twelve Quays House, Birkenhead, Merseyside CH41
21: 1LD, United Kingdom}
22: \altaffiltext{2}{Jodrell Bank Observatory, University of
23: Manchester, Macclesfield, Cheshire SK11 9DL, United Kingdom}
24: \altaffiltext{3}{Department of Astronomy, Ohio
25: State University, 140 West 18th Avenue, Columbus, OH 43210}
26: \altaffiltext{4}{Department of Physics, Chungbuk National University,
27: Chongju 361-763, Korea}
28: \altaffiltext{5}{Ulugh Beg Astronomical
29: Institute, Uzbek Academy of Sciences, Tashkent, Uzbekistan}
30: \altaffiltext{6}{Department of Physics and Astronomy, FPRP, Seoul National
31: University,
32: Seoul 151-742, Korea}
33: \altaffiltext{7}{Korea
34: Astronomy and Space Science Institute, Hwaam-Dong, Yuseong-Gu, Daejeon
35: 305-348, Korea}
36:
37: %\submitted{Submitted to The Astrophysical Journal}
38:
39: \begin{abstract}
40: The Angstrom Project is undertaking an optical survey of stellar
41: microlensing events across the bulge region of the Andromeda Galaxy
42: (M31) using a distributed network of two-meter class telescopes. The
43: Angstrom Project Alert System (APAS) has been developed to identify in
44: real time candidate microlensing and transient events using data from
45: the Liverpool and Faulkes North robotic telescopes. This is the first
46: time that real-time microlensing discovery has been attempted outside
47: of the Milky Way and its satellite galaxies. The APAS is designed to
48: enable follow-up studies of M31 microlensing systems, including
49: searches for gas giant planets in M31. Here we describe the APAS and
50: we present a few example light curves obtained during its commissioning
51: phase which clearly demonstrate its real-time capability to identify
52: microlensing candidates as well as other transient sources.
53: \end{abstract}
54:
55: \keywords{gravitational lensing -- galaxies: individual (M31) -- techniques:
56: photometric}
57:
58: \section{Introduction}
59:
60: Gravitational microlensing has proved itself as a powerful tool for
61: planetary, stellar and Galactic studies \citep{pac96,beau06,gou06}. It
62: has also been used to look for dark matter in the Andromeda Galaxy
63: (M31) where the stellar crowding is much more severe
64: \citep{cal03,rif03,dej04}. Many of the exciting discoveries made by
65: microlensing within our own Galaxy are possible only because
66: microlensing is alerted in real time by survey teams such as OGLE \citep{sumi06}
67: and MOA \citep{sumi03}, allowing follow-up networks such MicroFUN
68: \citep{gou06} and PLANET/RoboNet \citep{beau06} to
69: target candidate events with high time resolution. The requirement of
70: both rapid and robust data processing has until now prevented the
71: development of an alert system for M31 microlensing.
72:
73: The Andromeda Galaxy Stellar Robotic Microlensing (Angstrom) Project
74: aims to use microlensing to probe the geometry and stellar mass
75: function of the M31 bulge \citep{ker06}. Since many of these events
76: are expected to have durations of just a few days, the survey employs a
77: distributed network of telescopes to enable round-the-clock monitoring
78: of the M31 bulge. Two of the telescopes within the network are fully
79: robotic, allowing data to be collected and processed rapidly. The
80: Angstrom Project is taking advantage of this to develop a real-time
81: data processing pipeline as part of a microlensing and transients
82: alert system. The Angstrom Project Alert System (APAS) is designed to
83: enable follow-up studies of candidate events in M31 in a similar vein
84: to those conducted within our own Galaxy. \cite{chu06} has highlighted
85: how such an alert system could enable microlensing discovery of gas
86: giant planets in M31 itself. Other benefits of such a system would be
87: a greater ability to identify and characterize binary microlensing
88: events, which are estimated to comprise around 15\% of all M31 events
89: \citep{bal01}, and the possibility of measuring finite source effects
90: which can be used to constrain lens and source parameters.
91:
92: In this {\em Letter}\/ we describe the design and implementation of
93: the APAS and we show some early results from its commissioning phase
94: during the 2006/7 Angstrom observing season. The results illustrate
95: the potential of the system for real-time microlensing and transient
96: discovery.
97:
98: \section{Angstrom Project Observations}
99:
100: The Angstrom Project\footnote{http://www.astro.livjm.ac.uk/angstrom/}
101: employs a network of 2m-class telescopes. These include the robotic 2m
102: Liverpool Telescope (LT) on La~Palma, the robotic 2m Faulkes Telescope
103: North (FTN) in Hawaii, the 1.8m telescope at Bohyunsan Observatory in
104: Korea, the 2.4m telescope at MDM in Arizona and, recently, the
105: 1.5m at Maidanak Observatory in Uzbekistan. The telescope network
106: provides excellent round-the-clock coverage of the M31 bulge during
107: the observing season, which lasts from August through to
108: February. Angstrom aims to obtain upwards of three observations per
109: 24-hour period in order to be able to detect and characterize
110: short-lived microlensing events typically induced by low-mass M dwarf
111: stars in the M31 bulge \citep{ker06}.
112:
113: The APAS primarily utilizes incoming data from the two robotic
114: telescopes, the LT and FTN. These telescopes operate autonomously,
115: executing programs that are queued by an intelligent scheduler
116: program \citep{fras06}. Images obtained by these telescopes are automatically
117: pre-processed and are made available via the web typically within fifteen
118: minutes of observation. Both telescopes employ identical optical
119: cameras with a 4.6~arcminute field of view and 2k$\times$2k CCD
120: arrays. Angstrom observations are conducted primarily in the
121: Sloan-like $i$ band on both telescopes, typically with a sequence of
122: exposures totalling around 30~mins. Each exposure within the sequence
123: is short (no more than 200~secs) in order to avoid saturating the core
124: of the M31 bulge. The individual exposures are then stacked to form a
125: single 30~min exposure image for each epoch. More details can be found
126: in \cite{dar07}, hereafter D07, which describes the main processing
127: pipeline for the whole telescope network.
128:
129: In addition to LT and FTN data the APAS uses archive data from the
130: POINT-AGAPE dark matter microlensing survey of M31
131: \citep{cal03}. POINT-AGAPE used the wide-field camera of the 2.5m
132: Isaac Newton Telescope on La~Palma to survey a 0.6~deg$^2$ area of the
133: M31 disk and bulge. Around 65\% of the LT/FTN reference field overlaps with
134: the POINT-AGAPE data. For the APAS we have reprocessed all Sloan
135: $i$-band data from the POINT-AGAPE survey obtained between 1999 and
136: 2001. We use it to provide an extended data baseline for the subset of
137: light curves lying in both Angstrom and POINT-AGAPE survey regions.
138:
139: \section{The Angstrom Project Alert System (APAS)}
140:
141: M31 presents a challenging target for real-time microlensing
142: discovery. At a distance of 780~kpc it is two orders of magnitude more
143: distant than typical microlensing sources towards the Galactic bulge
144: where the majority of Galactic microlensing alerts occur. This means
145: that larger telescopes and longer exposures are required to obtain
146: sufficiently sensitive photometry. It also means that we have to deal
147: with stellar fields which are far more crowded and where the typical
148: microlensing source is completely unresolved at baseline (see
149: Figure~\ref{images}). This is the
150: so-called pixel-lensing regime.
151:
152: \begin{figure*}
153: \epsscale{1.} \plotone{f1.eps}
154: \caption{(a) LT $i$-band image of a two arcminute region of the Angstrom
155: M31 bulge field. (b) Corresponding difference image obtained after
156: convolving and subtracting a reference image, showing variable objects
157: as black and white spots. The residual of the imperfectly subtracted
158: core of the M31 bulge is visible in the top right-hand corner. (c) The
159: corresponding significance map of variable sources visible above the
160: background. Variable source blending is evident across the
161: map. Blackened regions denote masked areas. (d) Stars within a two
162: arcminute region of an OGLE-III Milky Way bulge field shown for
163: comparison with the crowding levels seen in the variable source map in
164: (c). Whilst all objects in (c) have variable flux only a handful of the stars visible in (d) are variables. OGLE image is from the OGLE Early Warning System \citep{uda03}
165: for alert OGLE-2005-BLG-172. }
166: \label{images}
167: \end{figure*}
168:
169: The challenge of developing an alert system for an M31 pixel-lensing
170: survey differs in some key respects from that of a Galactic
171: microlensing alert system. One downside is that towards the core of
172: the M31 bulge we reach the variable star crowding limit. Reliable
173: photometry for individual variable objects becomes difficult even when
174: using difference imaging \citep{ala98}. This
175: means that very often more than one variable source contributes to the
176: light curve at a given position, as illustrated in the variable source
177: map shown in panel~(c) of
178: Figure~\ref{images}. However, variable signals in the M31
179: bulge must necessarily involve a significant source flux change in
180: order to be detectable above the very bright M31 bulge surface
181: brightness. Therefore there are fewer classes of variable stars that
182: are sufficiently bright and variable to provide potential false
183: positive signals.
184:
185: With these potential difficulties in mind, we have adopted a pragmatic
186: approach in developing the APAS. We do not require it to autonomously
187: and robustly flag up high quality microlensing alerts. Rather the APAS
188: presents a modest list of light curves that exhibit transient or
189: microlensing-like behavior. The list is then reviewed by a human
190: observer who makes the final decision on whether to issue an alert for
191: any of the signals. The role of the APAS is to present to the observer
192: via a web-based interface a compact list (typically no more than a few
193: dozen light curves) of the currently most interesting transient signals
194: from a continuously updated database of around 20000 light curves. The
195: observer is then in a position to quickly review ongoing signals of
196: potential interest. The APAS also records the positions of all
197: variable signals in the immediate neighborhood of each light curve in
198: the list so that the observer can check and quantify any light curve
199: contamination.
200:
201: The APAS comprises two stages: a robust real-time data reduction
202: pipeline and an event classification scheme. We now describe both of these
203: stages in more detail.
204:
205: \subsection{Real-time data reduction}
206: \label {rtdr}
207:
208: The Angstrom Data Analysis
209: Pipeline (ADAP) is responsible for reducing all
210: Angstrom data from both robotic and non-robotic telescopes, and is described in
211: detail in D07. The nominal mode of operation of the ADAP is to process the
212: full dataset offline at the end of each observing season. However we
213: also employ the ADAP to process a 1k$\times$1k re-binned copy of the
214: 2k$\times$2k robotic telescope images that can be processed in real time.
215: The pixel size of the re-binned data is still only $0.''26$ and so adequately
216: samples the typical seeing disk ($1''-1.''5$). The ADAP offline processing is
217: described fully in D07, whereas here we concentrate on the real-time processing
218: of the re-binned data which forms part of the APAS. Many of these stages are
219: similar or identical to the ADAP offline processing so we keep our
220: description brief and refer the reader to D07 for further details.
221:
222: Firstly, an initial
223: quality control score is made of each frame based upon the shape and size of
224: the point-spread function (PSF). Images which fail this check are dropped from
225: the
226: subsequent analysis as their inclusion would result in unacceptable
227: difference image quality. For the surviving images the APAS performs defect
228: masking, cosmic ray rejection and image alignment. Then the APAS de-fringes each
229: frame using an iterative method described in D07. Images taken within the same
230: run (i.e sequentially at one epoch) are then PSF matched and stacked using the
231: AngstromISIS difference image engine, which is a modified version of a subset of
232: routines from the ISIS image subtraction package \citep{ala98}. AngstromISIS is
233: described in detail in D07. The image stack selected to be the reference is then
234: convolved and subtracted from the other image stacks to create a sequence of
235: difference images. These are then used to produce one likelihood map of variable
236: sources for each epoch, using an indicator of statistical significance similar
237: to that of
238: \cite{cash79}. Sources from within each likelihood map are then identified and
239: matched and an updated master list of variable sources is compiled. PSF-fitting
240: photometry is performed on the difference images and the resulting
241: photometry is stored within a MySQL database.
242: The time between observation and adding new
243: photometric points to the database is about two hours.
244:
245: This pipeline is
246: applied both in real time to the LT and FTN data and offline on regions
247: of legacy POINT-AGAPE data which overlap
248: with the Angstrom field. The LT and FTN are
249: essentially identical telescopes and the observations were taken using
250: matched filter systems, so a simple linear scaling is sufficient to deduce
251: relative light curve scalings and offsets for these telescopes. The POINT-AGAPE
252: $i$-band data is matched to LT/FTN $i$-band photometry by fitting a sample of
253: periodic variables present in both the Angstrom and POINT-AGAPE fields.
254:
255: \subsection{Candidate classification}
256:
257: APAS event classification is initiated each time
258: a new robotic observation is added to the light curve database. The APAS
259: employs a
260: number of selection criteria to classify
261: each light curve into one of seven different categories.
262:
263: Each object's light curve is separated by telescope (LT, FTN or POINT-AGAPE),
264: with each
265: telescope being initially processed separately. The APAS deems an object to
266: be valid if it contains at
267: least seven LT light curve points from the first (2004/5) and second
268: (2005/6) Angstrom
269: observing seasons. It must also contain at least seven FTN light curve points
270: from
271: the second Angstrom season. If an object does not meet these criteria
272: it is classed as {\em Void} and will never be processed again by the
273: APAS. If a light curve does not contain at least seven
274: points in the current (2006/7) observing season then the object is marked as
275: {\em Invalid}. Such objects are reviewed again when new
276: observations are added to the database.
277:
278: For remaining objects a baseline is estimated for each telescope by
279: determining the median light curve flux. The APAS then looks for indications of
280: transient behavior. Any point that is at
281: least $5\sigma$ above the baseline is considered ``significant''. For each
282: object the APAS then investigates any
283: clusterings of significant points (``peaks'') for each telescope. A
284: peak is defined as a clustering of at least five significant points,
285: all lying within the same observing season. Within the peak some leeway is
286: permitted - up to three consecutive non-significant points are admitted
287: within the peak provided they are followed by a significant point within the
288: same season and provided their presence coincides with worsening
289: seeing.
290:
291: Objects containing no peaks are classed as {\em Flat}. These
292: objects will be re-analyzed and possibly re-classified when new observations
293: are added to the database.
294: If an object contains no peaks in the current season but contains at
295: least one peak within a previous
296: season then it is classed as {\em Variable}, undergoing
297: re-analysis when additional observations are obtained.
298: Objects containing peaks in the current
299: season are checked for the presence of peaks in previous seasons. The
300: APAS must allow for the fact that the crowding of variables may superpose
301: variability on the baseline of transient and microlensing events \citep[for example candidate PA-99-N1][]{pau03}. Accordingly,
302: it does not necessarily classify objects as variable if they show variability
303: on previous seasons as well as in the current one. Instead, if peaks are seen in
304: previous seasons then the object is classified as variable unless a peak in the
305: current
306: season is at least 150 reference image counts and is at least
307: 50\% greater than the highest amplitude peak in any previous season. Other than
308: these the only remaining objects are those which show only peaks during the
309: current season.
310:
311: The APAS now considers all surviving objects to be transients and
312: flags them as {\em Followed}. The FTN and POINT-AGAPE (where available)
313: light curves for these objects are now scaled to the reference image
314: of the LT data and a
315: reduced Paczy\'nski profile is then
316: fitted with a
317: flux $F(t)=B+\Delta F/\sqrt{1+12[(t-t_{0})/t_{\rm FWHM}]^2}$,
318: where $t$ and $t_{0}$ are the epochs of observation and
319: maximum brightness, $B$ is the baseline flux, $\Delta F = F(t_0)-B$ is the
320: maximum flux deviation and $t_{\rm FWHM}$
321: is the duration of the full-width at half-maximum. The values of
322: $B$ and $\Delta F$ are telescope and bandpass dependent.
323: We require that the best fit $t_{0}$
324: does not occur
325: before the first observation of the current season and that it does
326: not occur after the anticipated start of the next season. The FWHM of
327: the event must be in the range of $1~\mbox{day} \leq t_{\rm FWHM}\leq365$ days.
328: Finally we require that the amplitude of the event be
329: positive. Objects passing these criteria are
330: classed as {\em Alerts}. Objects that fail any of these criteria
331: remain classed as {\em Followed}, unless they were previously classified as an
332: alert in which case they are re-classed as {\em Old}.
333:
334: The APAS utilizes a web-based interface to report results.
335: Whilst allowing a user to examine the light curves of all alerted
336: and followed objects it also displays light curves of all
337: objects within a three arcsecond vicinity of each alert. This is helpful in
338: deciding whether an object is a true alert but may have a baseline
339: contaminated by a nearby variable source.
340:
341: \subsection{Results from the commissioning phase}
342:
343: \begin{figure*}
344: \begin{centering}
345: \includegraphics[width=2.85cm,angle=270]{f2a}
346: \includegraphics[width=2.85cm,angle=270]{f2b}\\
347: \includegraphics[width=2.85cm,angle=270]{f2c}
348: \includegraphics[width=2.85cm,angle=270]{f2d}\\
349: \includegraphics[width=2.85cm,angle=270]{f2e}
350: \includegraphics[width=2.85cm,angle=270]{f2f}\\
351: \end{centering}
352: \caption{Three examples of alerts selected by the
353: APAS. LT photometry is shown in black and FTN in red. (a) A short
354: time-scale ($t_{\rm FWHM} \simeq 1$~day) high signal-to-noise ratio microlensing
355: candidate. (b) A lower
356: signal-to-noise ratio microlensing candidate at the borderline of identification
357: by
358: APAS. Both microlensing candidates exhibit periodic variability in their
359: baselines due to nearby variable sources. (c) Classical Nova
360: 2006~8\protect\footnote{Rochester naming convention
361: http://www.rochesterastronomy.org/snimages/index.html} first
362: announced by \cite{atel884}. This light curve is easily distinguishable
363: from microlensing by its asymmetric profile. Right-hand panels show zooms of
364: the events and the solid line is the best-fitting reduced Paczy\'nski curve.}
365: \label{lc}
366: \end{figure*}
367:
368: \begin{table}
369: \caption{A snapshot statistical summary from the APAS.}
370: \center{\begin{tabular}{lll}
371: \hline
372: \hline
373: Object & Number of & Percentage\\
374: status & objects & of total \\
375: \hline
376: Void & $6,311$ & $34.0\%$\\
377: Invalid & $54$ & $0.3\%$\\
378: Flat & $9,931$ & $53.6\%$\\
379: Variable & $2,197$ & $11.8\%$\\
380: Follow & $12$ & $0.1\%$\\
381: Old & $0$ & $0.0\%$\\
382: Alert & $38$ & $0.2\%$\\
383: \hline
384: \end {tabular}}
385: \label{alert_table}
386: \end{table}
387:
388: The Angstrom Project is currently in its third observing
389: season and this is the first season in which we have implemented the APAS.
390: Table~\ref{alert_table} contains a snapshot summary of statistics from a
391: recent run of the APAS. Around $12\%$ of the light curve catalogue
392: is classified as variable or transient. The bulk of the rest of the
393: light curves are either ``Flat'' because they do not show significant
394: light curve variation, or ``Void'' because they do not comprise enough
395: data points in previous observing seasons. Reassuringly, the number of variables
396: flagged up by the APAS is in line with the variable number density derived from
397: an analysis of three years of data from the wide area POINT-AGAPE survey
398: \citep{an04}. This indicates that, whilst probably not optimal, the APAS
399: nonetheless competes in terms of sensitivity with previous non real-time
400: pipelines. We can expect that the offline ADAP analysis of the un-binned data
401: will be more sensitive still. APAS also succeeds in presenting an easily
402: manageable number of alerted and followed light curves for a human observer to
403: review.
404:
405: Three light curves from the subset of 38 listed alerts are shown in
406: Figure~\ref{lc}. Clearly, these examples represent credible microlensing and
407: nova candidates. The APAS therefore demonstrates the capability to provide a
408: rapid, robust and credible extragalactic alert system. The first candidate in
409: particular appears to be a very short high signal-to-noise ratio microlensing
410: event.
411: Such events tend to be sensitive to finite source effects and this event is
412: currently the subject of a separate study \citep{duke07}.
413:
414: \section{Discussion}
415:
416: We have successfully implemented the Angstrom Project Alert System (APAS), which
417: performs real-time reduction and identification of extragalactic transient
418: sources and
419: microlensing events. This is the first system of its type to process
420: extragalactic observations. The APAS has been commissioned during the third
421: season of Angstrom observations and is already displaying the capability
422: to identify credible microlensing and nova candidates. During the remainder of
423: the current 2006/7 observing season and in the following observing seasons we
424: aim to use the APAS to flag up candidate sources for higher cadence follow-up in
425: order to obtain detailed light curves of novae outbursts, and to identify
426: exotic microlensing systems.
427:
428: \acknowledgments
429:
430: The LT is operated by Liverpool JMU on behalf of the UK
431: Particle Physics and Astronomy Research
432: Council (PPARC). FTN is operated by the Las
433: Cumbres Observatory Global Telescope network. Data from FTN was
434: obtained through the RoboNet-1.0 Consortium Open Time programme. Pre-
435: processed data from the Isaac Newton Telescope is courtesy of the POINT-AGAPE
436: team.
437: EK and JPD are supported, respectively, by an Advanced Fellowship
438: and a PhD studentship from PPARC.
439: CH and B-GP are supported by grant C00072 of the Korea Research Foundation.
440: Y-BJ and C-UL acknowledge support from the Korea Astronomy and Space
441: Science Institute. MAI and RGK acknowledge the permanent technical and financial support
442: of the Maidanak Observatory observations from the Consortium of the
443: Korean Universities operating under the MOU between the Consortium and
444: UBAI.
445:
446:
447: \begin{thebibliography}{99}
448: %\frenchspacing
449:
450: %\bibitem[]{}
451:
452: \bibitem[Alard \& Lupton(1998)]{ala98} Alard, C. \& Lupton, R., 1998, ApJ,
453: 503, 325
454:
455: \bibitem[An et al.(2004)]{an04}
456: An, J., et al., 2004, MNRAS, 351, 1071
457:
458: \bibitem[Baltz \& Gondolo(2001)]{bal01} Baltz, E. \& Gondolo, P., 2001,
459: ApJ, 559, 41
460:
461: \bibitem[Beaulieu et al.(2006)]{beau06} Beaulieu, J., et al., 2006,
462: Nat, 439, 437
463:
464: \bibitem[Burwitz et al.(2006)]{atel884} Burwitz, V., et al., 2006, The
465: Astronomer's Telegram, 884, 1
466:
467: \bibitem[Calchi~Novati et al.(2003)]{cal03} Calchi~Novati, S., et al.,
468: 2003, A\&A, 405, 851
469:
470: \bibitem[Cash(1979)]{cash79}
471: Cash, W., 1979, ApJ, 228, 939
472:
473: \bibitem[Chung et al.(2006)]{chu06} Chung, S.-J., et al., 2006, ApJ,
474: 650, 432
475:
476: \bibitem[Darnley et al.(2007)]{dar07} Darnley, M.J., et al., 2007, in
477: preparation (D07)
478:
479: \bibitem[de~Jong et al.(2004)]{dej04} de~Jong, J.T.A., et al., 2004,
480: A\&A, 417, 461
481:
482: \bibitem[Duke et al. (2007)]{duke07} Duke, J.P., et al., 2007, in preparation
483:
484: \bibitem[Fraser(2006)]{fras06}
485: Fraser, S., 2006, Astronomische Nachrichten, 327, 779
486:
487: \bibitem[Gould et al.(2006)]{gou06} Gould, A., et al., 2006, ApJ, 644,
488: L37
489:
490: \bibitem[Kerins et al.(2006)]{ker06} Kerins, E., et al., 2006, MNRAS,
491: 365, 1099
492:
493: \bibitem[Paczy\'nski(1996)]{pac96} Paczy\'nski, B., 1996, ARA\&A, 34,
494: 419
495:
496: \bibitem[Paulin-Henriksson et al.(2003)]{pau03}
497: Paulin-Henriksson, S., et al., 2003, A\&A, 405, 15
498:
499: \bibitem[Riffeser et al.(2003)]{rif03} Riffeser, A., et al., 2003, ApJ,
500: 599, L17
501:
502: \bibitem[Sumi et al.(2003)]{sumi03}
503: Sumi, T., et al., 2003, ApJ, 591, 204
504:
505: \bibitem[Sumi et al.(2006)]{sumi06}
506: Sumi, T., et al., 2006, ApJ, 636, 240
507:
508: \bibitem[Udalski(2003)]{uda03} Udalski, A., 2003, AcA, 53, 291
509: \end{thebibliography}
510:
511: \end{document}
512: