0801.1503/ms.tex
1: %%% Astron. Nachrichten, instructions for authors coding with LaTeX2e. V2005.1
2: %%% -MWL- 2006-01-13 auf Verlagswunsch wieder altes Bibliographie-Format
3: %%% 
4: 
5: \documentclass[mathleft
6: % early%
7: ]{an}
8: \usepackage{graphicx}
9: \usepackage{times}
10: %\usepackage{cite}
11: \overfullrule5pt
12: \begin{document}
13: 
14: % The following seven commands are intended for editorial usage and should be ignored by
15: % the author(s).
16: \Pagespan{789}{}% Document's page range. 
17: % If second parameter is left empty, the last page is computed automatically.
18: \Yearpublication{2006}%
19: \Yearsubmission{2005}%
20: \Month{11}%   
21: \Volume{999}%  
22: \Issue{88}% 
23: % \DOI{This.is/not.aDOI}% 
24: 
25: \def\aap{{\it Astr.~Ap.}}     %Astronomy & Astrophysics%
26: \def\nat{{\it Nature}}      %Nature%
27: \def\et{{et\thinspace al}}    %et al.%
28: 
29: \title{Transient Detection and Classification}
30: 
31: \author{A.C.Becker\inst{1}\fnmsep\thanks{Corresponding author:
32:   \email{becker@astro.washington.edu}\newline}
33: }
34: \titlerunning{Transient Detection and Classification}
35: \authorrunning{A.C. Becker}
36: \institute{
37: University of Washington, Department of Astronomy, Box 351580, Seattle WA 98195-1580
38: }
39: 
40: \received{}
41: \accepted{}
42: \publonline{later}
43: 
44: \keywords{surveys -- methods: data analysis -- techniques: image processing}
45: 
46: \abstract{ I provide an incomplete inventory of the astronomical
47:   variability that will be found by next--generation time--domain
48:   astronomical surveys.  These phenomena span the distance range from
49:   near--Earth satellites to the farthest Gamma Ray Bursts.  The
50:   surveys that detect these transients will issue alerts to the
51:   greater astronomical community; this decision process must be
52:   extremely robust to avoid a slew of ``false'' alerts, and to
53:   maintain the community's trust in the surveys.  I review the
54:   functionality required of both the surveys and the telescope
55:   networks that will be following them up, and the role of VOEvents in
56:   this process.  Finally, I offer some ideas about object and event
57:   classification, which will be explored more thoroughly by other
58:   articles in these proceedings.  }
59: 
60: \maketitle
61: 
62: 
63: \section{Introduction}
64: Next generation surveys such as
65: Pan-STARRS\footnote{http://pan-starrs.ifa.hawaii.edu} and
66: LSST\footnote{http://www.lsst.org} promise to open up the time domain
67: of astronomical variability to the general community as a service.
68: This will allow the global study of all on--going phenomena in
69: real--time, enabling both the small aperture amateur astronomer and
70: the large aperture professional.  It also places great responsibility
71: on the surveys themselves to provide a reliable stream of information.
72: 
73: The scope of the planned data release is unprecedented; 10$^{5-6}$
74: transient ``alerts'' are predicted to be generated on a nightly basis
75: by the LSST alone.  This will place a huge burden on follow--up
76: networks in the very near future.  Undertaking follow--up of these
77: alerts could easily consume {\it all} of the available global
78: telescope resources unless intelligent decisions are made about which
79: events to focus on.  The volume of the data streams will preclude this
80: decision from being made by a human.
81: 
82: To enable intelligent, autonomous follow--up systems, the VOEvent
83: protocol has been developed as a means of automatically conveying
84: information between astronomical resources.  To take advantage of this
85: stream, Heterogeneous Telescope Networks (HTNs) are being implemented
86: to undertake follow--up of alerts.  This purpose of this conference is
87: to address the practical implementation of this marriage.  This paper
88: will address the types of astronomical variability the surveys will
89: have to sort through, the types of alerts that the surveys will
90: generate, and present various ideas on the classification of these
91: alerts.  I will emphasize that the surveys must do more than generate
92: alerts; they must also listen to the follow--up results of the
93: community or risk retaining outdated classifications based upon their
94: own limited data.
95: 
96: \section{Sources of Astronomical Variability}
97: The exciting part about surveys such as LSST is that they intend to
98: detect, classify, and release information on {\it all} variability
99: found whether it be photometric or astrometric.  This provides a
100: technical challenge for the surveys in terms of autonomy and
101: reliability that has been approached but not yet demonstrably met in
102: precursor efforts (Bailey \et. 2007; Becker \et. 2004).
103: 
104: Modern surveys use image subtraction techniques to remove the static
105: portion of their images, leaving only the residual flux of objects
106: that have varied in brightness or position.  In these difference
107: images, the astronomical signal is almost exclusively objects
108: elongated due to astrometric motion, or positive or negative point
109: sources that have varied in brightness or position.  I provide below
110: an incomplete listing of astronomical phenomena expected to be found
111: in these images, starting with the foreground of astrometrically
112: variable objects and ending with the most distant of cosmological
113: explosions.
114: 
115: \subsection{Astrometric Variability}
116: Earth--orbiting satellites provide the least interesting (for most
117: scientists) but most destructive foreground.  They move at an angular
118: velocity of order $10^4 \arcsec$ s$^{-1}$, and completely traverse a
119: single field--of--view in a typical exposure time.  Their image
120: signature is a nearly infinitely elliptical streak.  Inactive
121: satellites may also be tumbling, which yields a periodic sequence of
122: flashes along the satellite's vector.  Most photometry software is not
123: designed to accurately model such a trail, and thus will deblend the
124: trail into numerous elongated sources.  This sequence of detections
125: may be identified in a database using e.g. a Hough transform (Storkey
126: \et. 2004).
127: 
128: Solar system objects move with an apparent angular velocity that is a
129: combination of their own spatial velocity (the dominant term for
130: nearby objects) and the Earth's reflex motion (dominant for distant
131: objects).  These objects may appear elongated depending on their
132: apparent angular motion, angle from opposition, and exposure time.
133: %Elongated objects have the profile of the image's point spread
134: %function (PSF) in the dimension perpendicular to motion, and the PSF
135: %convolved with a line along the direction of motion.  
136: Near--Earth objects (NEOs) have closest approaches of order 0.001 AU,
137: where their apparent sky motion is as fast as $1 \arcsec$ s$^{-1}$.
138: Main belt asteroids are found between 1 and 2 AU from Earth, orbiting
139: between Mars and Jupiter.  Typical angular velocities are $10 \arcsec$
140: hr$^{-1}$.  Trans--Neptunian objects (TNOs), or Kuiper belt objects
141: (KBOs), orbit within or beyond the orbit of Neptune near 40 AU.
142: Moving at $\sim 1 \arcsec$ hr$^{-1}$, they appear as single--epoch
143: point sources in all but the longest astronomical exposures, and are
144: difficult to distinguish from background transients.  An ensemble of
145: moving objects imaged nightly will yield a ``new'' detection per
146: object per image.  These single--epoch detections may be efficiently
147: linked in\-to orbits (Kubica \et. 2007), with any orphaned detections
148: potentially background variability.  Main Belt asteroids and TNOs are
149: concentrated strongly in the ecliptic plane, which may be used as a
150: prior to disentangle them from background cosmological transients.
151: Beyond the Kuiper belt, there are very few Solar System objects known.
152: The Oort cloud of comets is thought to exist between $10^4$ and $10^5$
153: AU, however no objects are currently known at this distance.  This
154: transition region reflects the boundary from objects that primarily
155: reflect the Sun's radiation to objects that produce their own.
156: 
157: Beyond this regime, astrometric motion is only noticeable in the
158: nearest or most rapidly moving of our stellar neighbors.  The extreme
159: example of this is Barnard's Star, whose apparent angular motion is
160: $\sim 10 \arcsec$ yr$^{-1}$.  The difference imaging signature of high
161: proper motion objects is a dipole whose nodes grow more separate as a
162: function of time, over the timescale of years.  Since dipoles are also
163: a classical signature of image subtraction failure, these may be
164: difficult to distinguish from systematic noise until an ensemble of
165: difference images is examined.
166: 
167: \subsection{Photometric Variability}
168: Beyond the solar neighborhood, all variability will have the higher
169: order moments of the image's PSF (a noteworthy exception are supernova
170: light echoes, which leave large, low surface brightness features
171: evolving radially from a central point; Rest \et. 2005).  Since these
172: variables all appear as point sources, contextual and temporal
173: information are required to classify the nature of each event.  I
174: summarize some of the expected source populations below.
175: 
176: Planetary transit searches are undertaken on nearby, apparently
177: bright, stars out to a distances of a couple of kilo--parsecs
178: (Pollacco \et. 2006).  Their lightcurves are characterized by minute
179: (several percent), periodic (several days) decrements in the host star
180: brightness as it is transited by its exoplanet.
181: 
182: At similar distances, a foreground ``fog'' of flaring M--dwarf stars
183: has been found (Kulkarni \& Rau 2006).  This comprises low--mass stars
184: that are too faint to be seen in a single image but which may flare
185: suddenly, yielding short timescale ($\sim 1000$ s) apparently hostless
186: transients (Becker \et. 2004).  The all--sky rate of these may be up
187: to tens of millions per year, making them a significant foreground to
188: cosmological variability.
189: 
190: At a distance of 8 kpc, the Galactic center is used to backlight a
191: foreground microlensing population, yielding (typically) symmetric,
192: unique source brightenings (Paczynski 1991).  Durations of these
193: events can be anywhere from $10^0$ to $10^3$ days.  Because of their
194: angular coverage and temporal sampling, microlensing surveys have also
195: yielded a wealth of information on stellar variability of all types.
196: At around 50 kpc, the Large and Small Magellanic Clouds are targeted
197: for microlensing by objects in our Galactic halo (Paczynski 1986),
198: with typical timescales of $10^2$ days.  Individual stars in these
199: galaxies are able to be resolved from the ground and photometered,
200: although the blending can be quite severe.
201: 
202: RR Lyrae are periodically pulsating horizontal branch stars that may
203: be recognized from their lightcurve shapes.  Their periods can range
204: from 0.2 to 2 days.  Because of their well--studied period--luminosity
205: relationship, their apparent magnitude distribution can be used to
206: infer Galactic structure (Sesar \et. 2007).  They have been found in
207: our Galactic halo out to 100 kpc; LSST expects to probe Galactic
208: accretion structure using RR Lyrae out to 400 kpc.
209: 
210: At a distance of 1000 kpc, M31 is also targeted by microlensing
211: surveys, but the lensed stars are unable to be resolved and difference
212: imaging techniques are necessary (Uglesich \et. 2004).
213: 
214: The HST Key Project measured lightcurves of Cepheid variable
215: stars in nearby galaxies out to 10 Mpc to measure the Hubble constant
216: (Freedman \et. 2001).  Cepheids are intrinsically several magnitudes
217: brighter than RR Lyrae, meaning they can be seen to larger distances.
218: They also have well calibrated period--luminosity relationships, with
219: periods of 5 to 30 days.
220: 
221: Starting at around 100 Mpc (a redshift of z $\sim$ 0.1), a wealth of
222: nearby supernovae have been observed, starting with the Calan-Tololo
223: sample of Type Ia supernova (Hamuy \et. 1996).  Because supernovae are
224: typically enmeshed in the light of their host galaxies, image
225: subtraction techniques are required to photometer the supernova light
226: uniquely.  Contextually, supernovae are commonly (but not always)
227: found near an extended host galaxy.  Their lightcurves experience a
228: steep rise ($\sim 20$ days) and a gradual fall over $\sim 100$ days
229: characteristic of their subtype (Ia, Ib/c, IIp, etc...).
230: 
231: Just beyond this sample, around a redshift of 0.1, the closest Gamma
232: Ray Bursts have been found (GRB 031203; Gotz \et. 2003).  Their
233: temporal evolution is much faster than supernovae, rising and falling
234: within hours to days, making optical discovery of these phenomena
235: extremely di\-fficult.
236: 
237: Medium redshift supernova surveys like the SDSS--II Supernova Survey
238: find events out to z $\sim 0.4$ (Frieman \et. 2007).  Higher redshift
239: supernova surveys, including both ESSENCE (Wood-Vasey \et. 2007) and
240: the Supernova Le\-gacy Survey (SNLS; Astier \et. 2006), detect Ia events
241: between 0.3 $<$ z $<$ 1.0, while the highest redshift supernova
242: surveys involving the HST find Ia supernovae out to $z \sim 1.4$
243: (Riess \et. 2007).  At their most distant, supernova are only visible
244: for several days around their peak.
245: %
246: Finally, the highest redshift cosmological transient, GRB 050904, was
247: found at $z \sim 6.3$ (Haislip \et. 2005).
248: 
249: The overall extent of astronomical variability is clearly enormous,
250: and individual surveys have typically been commissioned to address a
251: subset of the above whole.  However, surveys such as LSST anticipate
252: not only {\it detecting} and {\it classifying} the above phenomena,
253: but doing so in real--time.
254: 
255: \section{Alerts and Classification of Sources}
256: The general paradigm for alert generation is that a new event will be
257: recognized by a real--time survey pipeline, and the survey will
258: subsequently release an ``alert'' describing the detection (and
259: possibly supporting characterization) observations.  The format of
260: these alerts will be as VOEvent packets.  The decision to take action
261: based upon a VOEvent is undertaken by ``intelligent agents'' as a part
262: of each HTN.  Depending upon the science goals of each HTN, different
263: agents will make different decisions based upon the same information.
264: These actions will depend upon the event classifications of the survey
265: and potentially of the agent itself.  I summarize below some of the
266: requirements necessary for this model to succeed.
267: 
268: \subsection{Alerts and Followup}
269: A VOEvent packet includes {\tt inference} fields, where the survey
270: lists to the best of its abilities the classification of the event, as
271: well as the {\tt probability} that the event is of this class.  These
272: alerts may be {\it urgent} in nature, suggesting immediate follow--up.
273: The HTN resources decide to target or not based on the {\tt inference}
274: and {\tt probability}, and their particular science goals.  The
275: surveys and follow--up networks will also be releasing more prosaic
276: {\it informational} alerts that do not necessarily require action.
277: These should be released each time an object that has been alerted on
278: has been followed up.
279: 
280: Each HTN's intelligent agent may be assumed to be autonomous from the
281: others.  Consequently, these different agents may come to different
282: conclusions about the true nature of the event given the same
283: information.  This could easily lead to asynchronous/conflicting
284: evolution of knowledge between networks.  
285: 
286: The instantaneous state of knowledge about an alert can be extracted
287: from its ensemble of VOEvents by a citation mechanism that links
288: multiple observations together -- they are federated by mutual
289: citation.  As an alternative, the concept of a ``broker'' has been
290: introduced representing an agent who centralizes and disseminates this
291: information.  As with the surveys, the broker's survival requires
292: engendering trust from its subscribers.  Surveys may play a hybrid
293: role in this model, releasing VOEvents on the entirety of their alerts
294: but also serving as brokers by releasing more descriptive alerts
295: (including e.g. fit parameters) on particular subsets of events.
296: 
297: \subsection{Source Classification}
298: To release accurate alerts with a minimum of false positives, i.e. to
299: have trustworthy {\tt inference} in VOEvents, model templates of event
300: behavior must be built beforehand.  To first order, there are two
301: levels of classification requiring image--based (spatial) and
302: lightcurve--based (temporal) models.
303: 
304: \subsubsection{Spatial Classification}
305: The morphological classification of flux in an image is a
306: well--studied problem.  All astronomical images have a characteristic
307: point spread function (PSF), which is the transfer function of a point
308: source through the atmosphere, telescope optics, and detecting
309: instrument.  Objects sharper than the PSF are not likely to be real
310: astronomical phenomena; objects broader than the PSF may be noise
311: artifacts or resolved objects such as galaxies or comets.  Moving
312: objects will have the profile of the PSF in the dimension
313: perpendicular to motion, and the PSF convolved with a line along the
314: direction of motion
315: 
316: Classification in difference images occurs through comparison of the
317: residual flux with the PSF profile.  A spatial model of the PSF is
318: typically built from the data itself.  This step requires direct
319: access to the images where the variability was detected.  Given the
320: potential bandwidth and disk access requirements of the alternative,
321: this step is almost exclusively the responsibility of the survey.
322: 
323: \subsubsection{Temporal Classification}
324: Sets of detections may be linked into a lightcurve through mutual
325: citation or by a broker.  To recognize a given event as a certain
326: class of phenomena, data models must first be built that span the
327: parameter space characterizing the phenomena.  The lightcurve data are
328: then compared to each template lightcurve in a probabilistic sense,
329: determining which model (if any) they fit best.  These ideas are
330: explored more fully in other articles in these proceedings (Bloom,
331: Mahabal, Bailey).
332: 
333: A complication for these next generation surveys is that a data model
334: incomplete at the e.g. $1\%$ (or $0.1\%$, or even $0.01\%$) level will
335: result in an unacceptably large number of falsely classified alerts.
336: The models need to be nearly bullet--proof.  To build these models,
337: both theoretical and experimental priors should be taken into account.
338: However, the observational data best span the range of actual (as
339: opposed to expected) phenomenology.  This suggests that an ideal use
340: of existing datasets is to build these models {\it before} they are
341: required by LSST and Pan-STARRS.  A prime example of phenomenological
342: model building is in the description of supernova Ia lightcurves
343: (e.g. Guy \et. 2007).
344: 
345: One successful implementation of lightcurve classification driving
346: real--time decision making is by the SDSS-II Supernova Search (Sako
347: \et. 2007).  The evolving light\-curves are fit to various supernova
348: lightcurve models after each epoch of observation.
349: %Currently, spectroscopic follow--up is required to confirm the
350: %redshift and type of each supernova; since the survey is searching for
351: %Type Ia events, the sample targeted for spectroscopic follow--up
352: %should have a high a fraction of Ia events as possible.  
353: SDSS-II find that after 2--3 lightcurve epochs, approximately $90\%$
354: of objects {\it photometrically} classified as Ia end up being Ia.
355: While their analysis does not examine the efficiency of this process
356: (how many Ia are missed), it is nevertheless an encouraging precursor
357: effort.
358: 
359: \subsection{Data Access and Computational Requirements}
360: In the process of alerting and event classification, there are
361: multiple stages of computation and inference undertaken by members of
362: this community.  The separation of responsibilities emerges by
363: examining which aspects of the data are required to address each
364: problem, and who has easiest access to it.
365: 
366: As concrete examples, plausible alerts on moving objects include : a
367: known object may soon be lost, or its error ellipse is growing at an
368: unacceptable rate; an object is brighter/fainter than past behavior
369: would indicate; or the likelihood of this object impacting Earth is
370: significant.  Each of these alerts requires computation and knowledge
371: exchange.  For the first, the future cadence of the survey is
372: required; this sort of computation is best done by the survey.  For
373: the second, the past behavior of the object must be accessible or
374: queryable.  This is a requirement on the survey or on the federation
375: of all data on the object.  The last use case requires significant
376: computational resources to project all moving objects into the future.
377: This computation is most likely to get done (with high latency) by a
378: motivated user.
379: 
380: In the use case of a newly detected transient, it is unclear who (if
381: anyone) will have the final say in classification of the event.
382: Especially in the early portion of the lightcurve, the inference will
383: be changing rapidly as new data come in.  A broker seems most useful
384: at this critical stage, tying all data together into a coherent
385: inference.
386: 
387: Finally, there are the cases of known objects whose current behavior
388: is unexpected.  Examples include exotic deviations in an on--going
389: microlensing event due to a planet orbiting the lens, stellar flares
390: from low--mass stars, and deviations in the timings of transiting
391: systems due to the gravitational influence of other unseen planets.
392: In these cases, real--time lightcurve fitting is required (an
393: extensive task), not likely to happen by the survey unless it also
394: chooses to serve as a broker.
395: 
396: \section{Conclusions}
397: I have detailed some of the interplay between surveys and follow--up
398: networks in time--domain astronomy, highlighting their potential roles
399: and responsibilities.  While the concept of data federation using
400: VOEvent's citation mechanism may work, it may also lead to different
401: inferences by different resources.  An alternative is to broker the
402: evolving ``truth'' regarding an event by trusted agents in the system.
403: These agents have the potential to fully realize (or scuttle) the
404: successful interplay between surveys and follow--up HTNs.
405: 
406: The surveys and agents each play a role in the ultimate classification
407: of each event.  I emphasize that the surveys should also {\it listen}
408: to the network, and decide if and how to allow these external
409: resources to influence their internal event classifications.  Finally,
410: the need for accurate and precise data models is commensurate with the
411: data flow that will be compared against them.  These models should be
412: built sooner rather than later using currently existing data sets, to
413: ensure that the promise of the first several years of these surveys is
414: not lost to faulty or inaccurate alerts.
415: 
416: 
417: %\begin{thebibliography}{}
418: %  \bibitem{} Author1, A.B., Author2, C.D.: 2001, AN 322, 1
419: %  \bibitem{} Author3, E.F., Author4, G.H.: 2001, AN 322, 10
420: %  \bibitem{} Author5, I.: 2001, AN 322, 20
421: %  \bibitem{} Author6, J.: 2001, AN 322, 30
422: %\end{thebibliography}
423: 
424: %\bibliography{ms}
425: %\bibliographystyle{plain}
426: %\input{ms.bbl}
427: 
428: \begin{thebibliography}{10}
429: 
430: \bibitem{2006A&A...447...31A}
431: P.~{Astier} \et. : 2006,
432: \newblock {\em \aap} 447, 31--48
433: 
434: \bibitem{2007ApJ...665.1246B}
435: S.~{Bailey} \et. : 2007,
436: \newblock {\em \apj} 665, 1246--1253
437: 
438: \bibitem{2004ApJ...611..418B}
439: A.~C. {Becker} \et. : 2004,
440: \newblock {\em \apj} 611, 418--433
441: 
442: \bibitem{2001ApJ...553...47F}
443: W.~L. {Freedman} \et. : 2001,
444: \newblock {\em \apj} 553, 47--72
445: 
446: \bibitem{2007arXiv0708.2749F}
447: J.~A. {Frieman} \et. : 2007,
448: \newblock {\em ArXiv e-prints} 0708.2749
449: 
450: \bibitem{2003GCN..2459....1G}
451: D.~{Gotz} \et. : 2003,
452: \newblock {\em GRB Coordinates Network} 2459
453: 
454: \bibitem{2007A&A...466...11G}
455: J.~{Guy} \et. : 2007,
456: \newblock {\em \aap} 466, 11--21
457: 
458: \bibitem{2005GCN..3914....1H}
459: J.~{Haislip} \et. : 2005,
460: \newblock {\em GRB Coordinates Network} 3914
461: 
462: \bibitem{Hamuy-Calan96}
463: M.~{Hamuy} \et. : 1996,
464: \newblock {\em \aj} 112, 2391
465: 
466: \bibitem{2007Icar..189..151K}
467: J.~{Kubica} \et. : 2007,
468: \newblock {\em Icarus} 189, 151--168
469: 
470: \bibitem{2006ApJ...644L..63K}
471: S.~R. {Kulkarni} and A.~{Rau}. : 2006,
472: \newblock {\em \apjl} 644, L63--L66
473: 
474: \bibitem{1986ApJ...304....1P}
475: B.~{Paczynski}. : 1986,
476: \newblock {\em \apj} 304, 1--5
477: 
478: \bibitem{1991ApJ...371L..63P}
479: B.~{Paczynski}. : 1991,
480: \newblock {\em \apjl} 371, L63--L67
481: 
482: \bibitem{2006PASP..118.1407P}
483: D.~L. {Pollacco} \et. : 2006,
484: \newblock {\em \pasp} 118, 1407--1418
485: 
486: \bibitem{2005Natur.438.1132R}
487: A.~{Rest} \et. : 2005,
488: \newblock {\em \nat} 438, 1132--1134
489: 
490: \bibitem{2007ApJ...659...98R}
491: A.~G. {Riess} \et. : 2007,
492: \newblock {\em \apj} 659, 98--121
493: 
494: \bibitem{2007arXiv0708.2750S}
495: M.~{Sako} \et. : 2007,
496: \newblock {\em ArXiv e-prints} 0708.2750
497: 
498: \bibitem{2007arXiv0704.0655S}
499: B.~{Sesar} \et. : 2007,
500: \newblock {\em \aj} 134, 2236--2251
501: 
502: \bibitem{2004MNRAS.347...36S}
503: A.~J. {Storkey} \et. : 2004,
504: \newblock {\em \mnras} 347, 36--51
505: 
506: \bibitem{2004ApJ...612..877U}
507: R.~R. {Uglesich} \et. : 2004,
508: \newblock {\em \apj} 612, 877--893
509: 
510: \bibitem{2007astro.ph..1041W}
511: W.~M. {Wood-Vasey} \et. : 2007,
512: \newblock {\em \apj} 666, 694--715
513: 
514: \end{thebibliography}
515: 
516: 
517: \end{document}
518: 
519: 
520: