the explore project. i. a deep search for filetors that determine the eﬃciency and the number of...

THE EXPLORE PROJECT. I. A DEEP SEARCH FOR TRANSITING EXTRASOLAR PLANETS

G.Mallen-Ornelas,1,2

S. Seager,3H. K. C. Yee,

2,4D.Minniti,

2,5Michael D. Gladders,

4,6

G.M.Mallen-Fullerton,7and T. M. Brown

8

Received 2002March 13; accepted 2002 September 13

ABSTRACT

Planet transit searches promise to be the next breakthrough for extrasolar planet detection and will bringthe characterization of short-period planets into a new era. Every transiting planet discovered will have ameasured radius, which will provide constraints on planet composition, evolution, and migration history.Together with radial velocity measurements, the absolute mass of every transiting planet will be determined.In this paper we discuss the design considerations of the Extrasolar Planet Occultation Research(EXPLORE) project, a series of transiting planet searches using 4 m class telescopes to continuously monitora single field of stars in the Galactic plane in each �2 week observing campaign. We discuss the general fac-tors that determine the efficiency and the number of planets found by a transit search, including time sam-pling strategy and field selection. The primary goal is to select the most promising planet candidates forradial velocity follow-up observations. We show that with very high photometric precision light curves thathave frequent time sampling and at least two detected transits, it is possible to uniquely solve for the mainparameters of the eclipsing system (including planet radius), based on several important assumptions aboutthe central star. Together with a measured spectral type for the star, this unique solution for orbital parame-ters provides a powerful method for ruling out most contaminants to transiting planet candidates. For theEXPLORE project, radial velocity follow-up observations for companion mass determination of the bestcandidates are done on 8 m class telescopes within 2 or 3 months of the photometric campaigns. This same-season follow-up is made possible by the use of efficient pipelines to produce high-quality light curves withinweeks of the observations. We conclude by presenting early results from our first search, EXPLORE I, inwhich we reached better than 1% rms photometric precision (measured over a full night) on �37,000 starswith 14:5 � I � 18:2.

Subject headings: planetary systems — surveys — techniques: photometric

1. INTRODUCTION

The discovery of giant extrasolar planets in the mid-1990susing radial velocity (RV) techniques (see, e.g., Marcy,Cochran, &Mayor 2000) heralded a new era in the study ofplanetary systems, and to date�100 extrasolar giant planetshave been discovered.9 RV searches produced the com-pletely unexpected discovery of massive planets in few-day

period orbits, such as 51 Peg b (Mayor & Queloz 1995). Todate, 17 systems with orbital distances of less than 0.1 AUand periods of a few days have been found. The existence ofa class of close-in giant planets shows that planetary systemscan be radically different from our own. The discovery ofclose-in giant planets sparked much theoretical work onplanet formation and migration scenarios to explain theproximity of giant planets to the parent star, such as plane-tesimal scattering (see, e.g., Murray et al. 1998), planet-diskor binary star interactions (see, e.g., Lin, Bodenheimer, &Richardson 1996; Holman, Touma, & Tremaine 1997), anddynamical instabilities in multiple giant planet systems(Rasio & Ford 1996).

The existence of a significant population of close-in extra-solar giant planets (CEGPs) makes the method of findingplanetary systems via transits of their parent star very prom-ising; the probability that a given planet will show transits isinversely proportional to its orbital distance and relativelylarge for CEGPs around main-sequence stars (�10%).Moreover, for planets with periods of 3–4 days, it is possibleto detect two or more transits via high photometric preci-sion light curves that span a relatively small number ofnights. A photometric precision of 1%, which can be rou-tinely achieved with CCD cameras, is sufficient to detectgiant planets around Sun-like stars (see Fig. 1). The adventof wide-field CCD mosaic cameras greatly increases the effi-ciency of the transit search method, since a very large num-ber of stars can be monitored at once.

The principal motivation for the transit search method isthe possibility of characterizing planets in a way not possi-ble with current RV surveys: radius. Transiting planets are

1 Princeton University Observatory, Peyton Hall, Princeton, NJ 08544;[email protected]; and Departamento de Astronomıa yAstrofısica, Pontificia Universidad Catolica de Chile, Casilla 306,Santiago 22, Chile.

2 Visiting Astronomer, Cerro Tololo Inter-American Observatory,National Optical Astronomy Observatory (NOAO). NOAO is operated bythe Association of Universities for Research in Astronomy, Inc., undercontract with the National Science Foundation.

3 Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540.Current address: The Carnegie Institution of Washington, Department ofTerrestrial Magnetism, 5241 Broad Branch Road NW, Washington, DC20015; [email protected].

4 Department of Astronomy and Astrophysics, University ofToronto, 60 St. George Street, Toronto, ON M5S 3H8, Canada;[email protected].

5 Departamento de Astronomıa y Astrofısica, Pontificia UniversidadCatolica de Chile, Casilla 306, Santiago 22, Chile; [email protected].

6 Current address: The Carnegie Observatories, 813 Santa BarbaraStreet, Pasadena, CA 91107; [email protected].

7 Universidad Iberoamericana, Prolongacion Paseo de la Reforma 880,Edificio F, Segundo Piso, Col. Lomas de Santa Fe, 01200 Mexico, D.F.,Mexico; [email protected].

8 High Altitude Observatory/National Center for AtmosphericResearch, P.O. Box 3000, Boulder, CO 80307; [email protected].

9 See Extrasolar Planets Catalog,http://www.obspm.fr/encycl/catalog.html.

The Astrophysical Journal, 582:1123–1140, 2003 January 10

# 2003. The American Astronomical Society. All rights reserved. Printed in U.S.A.

1123

currently the only ones whose radii can be determined(based on transit depth and stellar radius). The planet radiuscan be measured to better than 10% precision with follow-up ultra–high precision photometry on transiting planets(see, e.g., Brown et al. 2001; Cody & Sasselov 2002). Aradius measurement is necessary to constrain the planetevolution and migration history and also provides con-straints on planet composition and atmosphere throughevolutionary models. The radii of CEGPs are especiallyinteresting for planetary physics because of the evolution ofthe planet in proximity to the central star and because of theas yet unknown migration timescales. Very interestingrecent work (Guillot & Showman 2002) has shown that theobserved radius of transiting planet HD 209458b (Brown etal. 2001) disagrees with the theoretically predicted radius by30% for their preferred evolutionary models. This impliesthat there are atmospheric or interior physical processestaking place that are not currently known. The determina-tion of planet radius as a function of stellar type and in dif-ferent stellar environments will be a major step forward inplanetary characterization and in understanding giantplanet physics.

Follow-up observations of planets found from transitsurveys can be very important for planet confirmation andcharacterization. RV planet mass measurements arerequired in order to confirm transit candidates as actualplanets. Mass measurements for transiting planets are facili-tated by the fact that the phase and period for the system areknown in advance, and therefore it is possible to conductobservations at times of maximum RV amplitude. For faint

stars (V � 18), it is currently possible to measure masses orplace upper limits as low as a couple of MJ (see, e.g., G.Mallen-Ornelas et al. 2003, in preparation). Brown et al.(2001) give a list of interesting follow-up studies possible fortransiting planets around bright stars. For fainter stars,large planetary moons could be detected from transit tim-ing. More importantly, transit searches have the possibilityof probing new regions of parameter space, compared tocurrent RV planet searches. For example, fainter stars canbe monitored in a photometric planet transit search, allow-ing planets to be found in more distant environments andorbiting intrinsically smaller stars. Transit searches areunbiased with respect to unusual spectral characteristics,which can lead to unexpected discoveries and help constrainplanet formation models.

One confirmed transiting extrasolar planet, HD 209458b,is currently known (Charbonneau et al. 2000; Henry et al.2000). HD 209458b was discovered by the RV technique,and follow-up photometry determined that it transits itsparent star. More than 20 groups around the world are cur-rently using photometry to search for transiting extrasolarplanets. Many different environments are being or havebeen searched, including the globular cluster 47 Tuc (Gilli-land et al. 2000), open clusters of different metallicities(Howell et al. 1999; Mochejska et al. 2002; Quirrenbach etal. 2000; Street et al. 2000, 2002; Burke et al. 2002), fieldstars with magnitude d13 (Borucki et al. 2001; Brown &Charbonneau 2000), and 13dId17:75 stars in the directionof the Galactic center (Udalski et al. 2002b; see also Udalskiet al. 2001a). There have also been several conceptual papersabout where and how to look for transits, including earlypapers by Struve (1952), Rosenblatt (1971), and Borucki &Summers (1984), a thorough and prescient paper by Giam-papa, Craine, & Hott (1995), and later papers after the dis-covery of CEGPs, including cluster search strategies (Janes1996) and searches toward the Galactic bulge with 10 mclass telescopes (Gaudi 2000).

Although no planets found from a transit search have yetbeen confirmed with a mass detection from RV follow-up,many planet candidates now exist and are being followed upfor mass confirmation. Notably, the OGLE III search(Udalski et al. 2002a, 2002b) has made public a list of over50 stars with small transiting companions available to theastronomical community for follow-up. The Vulcan project(Borucki et al. 2001; Jenkins, Caldwell, & Borucki 2002)and the Extrasolar Planet Occultation Research(EXPLORE) project (this paper; G. Mallen-Ornelas et al.2003, in preparation) have also produced planet candidatesthat are being followed up with RV observations. Althoughit is disappointing that the transit search method has not yetresulted in any confirmed planets, a steep learning curve, aswell as some lead-up time, is expected for this kind of enter-prise. It is not clear howmuch of the current low planet yieldresults from the nature of the problem (e.g., because of anintrinsically low frequency of close-in planets in the types ofstars surveyed so far) and how much from technical chal-lenges faced by different transit surveys. One of the limitingfactors so far is that many transit surveys have not moni-tored enough stars with sufficiently high photometric preci-sion. This paper has the goal of outlining the necessary stepsfor a successful transit search with a CCD mosaic cameraon a 2–4 m telescope, and in addition it describes the mainfalse positives that can be ruled out with high-precision,high time sampling photometry and spectral typing of stars.

Fig. 1.—Photometric precision requirements for detecting transits ofplanets with different radii, as a function of stellar radius. Diagonal linesindicate the combination of planet and star radii (y- and x-axes, respec-tively) that will result in transit depths (R2

p=R2�) of 0.01, 0.003, 0.001, and

0.0003mag. IfN data points are available during transit, a photometric pre-cision equal to the transit depth will result in an N1=2 � transit detection.For example, a 1% rms photometric precision will generally be sufficient todetect all planet/star radius combinations that fall above the top diagonalline.

1124 MALLEN-ORNELAS ET AL. Vol. 582

This paper presents a framework for the design of asearch for transiting planets around field stars and presentsearly results from the EXPLORE project. The EXPLOREproject is a set of transit searches using wide-field mosaiccameras on 4 m class telescopes, with follow-up RV mea-surements on 8 m class telescopes. We start in x 2 by describ-ing a very useful property of a high-quality light curve withtwo or more flat-bottomed transits: there is a unique solu-tion for planet and star parameters, as long as certainimportant conditions are met. This unique solution can beused to obtain a clean set of planet transit candidates andhence is one of the main motivations for our experimentaldesign. When designing a transit survey, it is important toconsider the frequency of transiting planets and the transit-ing planet detection probability. Section 3 presents adescription of the factors that affect the number of planetsdetected and an estimate of the number of planets expectedin a transit survey. The issues considered in x 3 are of a gen-eral nature and can be applied to the design of any planettransit search. When designing a specific planet transitsearch, a number of interrelated choices must be made inthe observational design; in x 4, we continue the discussionof survey design by describing the specific aspects of theEXPLORE project strategy. In x 5, we present early resultsfrom the first search in our project, the EXPLORE I search,conducted in 2001 June at the Cerro Tololo Inter-AmericanObservatory (CTIO) 4 m telescope. We summarize and con-clude in x 6.

2. THE UNIQUE SOLUTION OF A TWO-TRANSITLIGHT CURVE

One of the most attractive aspects of transit searches isthat much can be learned about a system with an orbitingcompanion from a good light curve showing two or moreeclipses. Here we describe for the first time how a light curvewith two or more flat-bottomed eclipses can in principle beused to derive a unique solution of orbital parameters andcompanion radius, given certain conditions. The unique so-lution provides a powerful method to select the best planetcandidates to be followed up for mass determination. Spe-cifically, the stellar mass M�, stellar radius R�, companionradius Rp, orbital distance D, and orbital inclination i canbe uniquely derived from a light curve with two or moreeclipses if the following conditions are met:

1. The light curve has an extremely high photometric pre-cision and high time sampling.2. The eclipses have flat bottoms (in a bandpass where

limb darkening is negligible), which implies that the com-panion is fully superimposed on the central star’s disk.3. There are no secondary eclipses (i.e., the brightness of

the companion is negligible compared to that of the centralstar).4. The period can be derived from the light curve (e.g.,

the two observed eclipses are consecutive).5. The light comes from a single star, rather than from

two or more blended stars.6. The central star is on the main sequence.7. The mass of the companion is negligible compared to

that of the central star (Mp5M�).8. The orbit is circular (expected for CEGPs because of

their short tidal circularization timescales).

If the above conditions are met, the five parameters M*,R*, Rp, D, and i can be uniquely derived from the five equa-tions below. For simplicity, the equations presented hereassume that R�5D. The five equations are the transitdepth,

DF ¼ Rp

R�

� �2

; ð1Þ

the relation between the inclination of the orbit and theshape of the transit light curve, as parameterized by theratio of the duration of the transit’s flat bottom tflat to thetotal transit duration tT ,

tflattT

� �2

¼1� Rp=R�

� �� 2� D=R�ð Þ cos i½ �2

1þ Rp=R�� 2� D=R�ð Þ cos i½ �2

; ð2Þ

the total transit duration,

tT ¼ PR��D

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ Rp

R�

� �2

� D

R�cos i

� �2s

; ð3Þ

Kepler’s third law,

P2 ¼ 4�2D3

GM�; ð4Þ

and the mass-radius relation for (Sun-like) main-sequencestars,

M� ¼ f R�ð Þ � R�M�R�

: ð5Þ

The following observable quantities are measured fromthe light curve and are used to solve the system of equations:the period P, the total transit duration tT , the flat eclipsebottom duration tflat, and the transit depth DF . Determiningthe five parameters M*, R*, Rp, D, and i from the system ofequations and observables is a useful shortcut, but in prac-tice the final errors and parameters are directly derived froma fit to the light curve. We note that five model parameterscan be extracted from the four observable quantitiesbecause of the assumed stellar mass-radius relation (eq. [5]),which provides a constraint that does not depend on obser-vations specific to a particular star. Analyzing the errorpropagation through equations (1)–(5) shows that the mostcritical inputs are tflat, tT , and the mass-radius relation. Forsystem parameters similar to those of HD 209458, errors of10 minutes in tflat or tT , or a 20% error in assumed radius ata given mass, leads to errors of about 30% in M� and cos iand about 15% inR�,Rp, andD.

The presence of significant limb darkening will have aneffect on the transit depth and shape (see Fig. 2). The flatbottom of the transit will become progressively morerounded when viewed at increasingly shorter wavelengths;moreover, a central transit will be deeper than R2

p=R2�, since

a larger fraction of the stellar light is coming from a smallerarea of the star. The shape of the transit can still be used toconstrain the orbital parameters if an appropriate limb-darkening model is adopted. However, given the extraparameters and the uncertainty in the adopted limb-darken-ing model, it is preferable to use light curves taken at longwavelengths, so that limb darkening is minimized (Fig. 2).

If any of the assumptions listed at the beginning of thissection are incorrect for a given light curve, then the derived

No. 2, 2003 EXPLORE PROJECT. I. 1125

parameters (M*, R*, Rp, D, and i) will also be incorrect. IfM� and R� can be obtained from spectral classification ofthe star, then the above system of five equations andunknowns will be overconstrained and can be used to checkthe correctness of the assumptions about the system. Forexample, when two transits are separated by an odd numberof nights they might not be consecutive transits, since atransit might have occurred during the day at time P=2. Ingeneral, gaps in time coverage can lead to missed eclipsesand the resulting period aliasing. If period aliasing is sus-pected, the actual period can be determined from the abovesystem of equations plus a spectral type. Even when onlyone high-quality transit is detected, the three parameters P,Rp, and i can still be constrained from equations (1)–(5),provided the spectral type—and hence M� and R�—isknown. Another crucial example of the usefulness ofobtaining a spectral type is the case of a light curve from anunresolved triple system in which two of the stars form aneclipsing binary system. In this case, solving the above equa-tions under the assumption that the light is coming fromonly one star will generally give a solution for R� and M�that is inconsistent with the spectral type. Thus, by comple-menting a two-transit light curve with spectra or evenbroadband colors, much can be learned about an eclipsingsystem, provided the photometry has very high precisionand high time sampling. Further details, error simulations,and applications of the unique solution to a transit lightcurve are presented in Seager &Mallen-Ornelas (2003).

3. GENERAL CONSIDERATIONS FOR THE DESIGNOF A TRANSIT SURVEY

In this section, we discuss the factors that should motivatethe basic design of any transit survey. The two most impor-tant broad considerations for designing a successful surveyare (1) finding planets and (2) providing useful statistics ofplanet frequency and characteristics. In broad terms, the

number of planets found by a transit search will be deter-mined by (1) the frequency of close-in planets around starsin the survey, (2) the probability of having a geometricalignment that shows transits, (3) the number of stars sur-veyed, (4) the photometric precision, and (5) the windowfunction of the observations. The last three elements can becontrolled by the survey strategy. The following subsectionsdiscuss the five factors listed above and their significance forsurvey strategy.

3.1. Planet Frequency and Detection Probability

An estimate of the fraction of field stars that have transit-ing short-period planets is useful for designing a transitsearch. The frequency of transiting planets for a givenensemble of main-sequence stars of similar metallicity, age,and environment can be approximately written as

Fp ¼ZZZ

Pp R�;Rp;D� �

Pg R�;Dð Þ dR� dRp dD : ð6Þ

Pp is the probability distribution that a star of radius R� hasa planet of radius Rp with an orbital distance D, and is pre-cisely what a good survey should aim to measure. Pp is alsolikely dependent on stellar metallicity, age, and environ-ment. Pp is currently not known, because only a small num-ber of CEGPs have been found to date by RV planetsearches. Pg is the geometric probability that a planet willoccult its parent star as seen from Earth; Pg � R�=D for anensemble of randomly oriented systems with circular orbitsand Rp5R�5D. A simple estimate of the frequency Fp oftransiting close-in giant planets (Pd4:5 days) can beobtained by assuming that all isolated stars have the samefrequency of CEGPs as isolated Sun-like stars and adoptingPp � 0:007 (Butler et al. 2000) and the correspondingPg � 0:1. Assuming that we can detect planets only aroundisolated stars and adopting a binary fraction of 1

2, we getFp ¼ 0:00035. In other words, we expect one in 3000 stars tohave a transiting close-in giant planet, with large uncer-tainty. The uncertainty comes from two sources. First, Pp ismeasured only for nearby, isolated Sun-like stars with plan-ets ofMp ’ MJ; moreover, this Pp estimate comes from sur-veys that suffer from limited statistics and selection effectsthat are difficult to characterize. Second, it might be possibleto detect transiting planets around a star in a binary system,depending on the brightness ratio of the stars and on thephotometric precision.

In practice, the fraction of stars with planets actually dis-covered by a transit search will likely be much less thanFp � 1=3000. The number of detected planets Np willdepend crucially on the window function of the observa-tionsW, the photometric precision �m, the time sampling ofthe observations �t, and the number of stars monitoredN R�ð Þ. The number of detected planets can be schemati-cally written as

Np W ; �mð Þ ¼ZZZ

N R�ð ÞPp R�;Rp;D� �

Pg R�;Dð Þ

� Pdet R�;Rp;D; �m; �t� �

� Pvis W ;D;R�ð Þ dR� dRp dD : ð7Þ

Here Pdet is the probability of detecting a transit of depthRp=R��

2, given a photometric precision �m and time inter-val between photometric data points �t and assuming thatthe transit occurs during the observations. For a given pho-

Fig. 2.—Solar limb darkening dependence of a central (i ¼ 90) planettransit light curve. The planet has R ¼ 1:4 RJ (approximately that of HD209458b), and the star has R ¼ R�. The solid curve shows a transit lightcurve with limb darkening neglected. The other curves, from top to bottom(at time = 0), show central transit light curves with solar limb darkening forwavelengths 3, 0.8, 0.55 and 0.45 lm. Although the transit depth changes atdifferent wavelengths, the ingress and egress slopes do not change signifi-cantly. The ingress and egress slopes mainly depend on the time it takes theplanet to cross the stellar limb.


tometric precision, the significance of the detection willincrease as the square root of the number of photometricdata points during transit; this number depends on the timesampling of the observations �t and the transit duration(which is dependent on D, M�, R�, Rp, and orbital inclina-tion i and will generally be close to 2–3 hr for close-in plan-ets orbiting Sun-like stars). For a given photometricprecision, transits for a planet of a given sizeRp will be moreeasily detected around stars of smaller radius R�. Con-versely, higher photometric precision will enable the detec-tion of smaller planets or planets around larger stars (seeFig. 1). Note that a Jupiter-sized planet transiting a Sun-sized or smaller star will have transits of depthRp=R�� 2

e1% and will thus be easily detected in well-

sampled 1% photometric precision light curves; i.e., for 1%photometric precision light curves, Pdet ¼ 1 for Jupiter-sized planets transiting Sun-like or smaller stars.

Pvis is the probability that at least two full transits willoccur during an imaging campaign; observing at least twotransits is required in order to measure the orbital periodand confirm the transit. Pvis is a function of orbital period P,transit length tT , duration of observations each night, andnumber of observing nights (described by the window func-tion W ). Note that P and tT are in turn functions of D, R*,andM* (or simplyD andR� for main-sequence stars). Sincewe ultimately seek to measure Pp, a good characterizationof Pvis is essential to determine the frequency of planetsaround different types of stars. Figure 3a showsPvis for three

Fig. 3.—(a) Probability Pvis of detecting transiting planets with different orbital periods. Quantity Pvis is calculated with the requirement that two transitsmust be observed. Consecutive nights of 10.8 hr each night are assumed. At each period all phases are considered; the difficulty of detecting some phases isexpressed by the dips in the curves. For example, at integer periods some transits will always occur during the day and will not be detectable. The different sym-bols are for transit searches of different total number of nights: 21 nights (triangles), 14 nights (bars), and the actual time coverage of the EXPLORE I search(dotted line). The vertical line indicates the lower period limit of known CEGPs. (b) Mean Pvish i as a function of number of consecutive nights in an observingrun. The solid line is for the requirement to detect two transits and the dashed line that for one transit. (c) Efficiency of the Pvish i per night. For a two-transitrequirement (solid line), the most efficient observing length (for 10.8 hr each night) is around 21 nights, but the efficiency is similar for runs of 16–30 nights. Fora single transit requirement, the efficiency curve (dashed line) decreases monotonically, because each additional night does not add new transit detections, butthe totalPvis for one night is nonetheless tiny (see [b]).


different cases, all with the requirement that two full transitsare observed. Shown are periods of 2–5 days, although itshould be noted that there are no known extrasolar planetswith periods below 3 days.10 The case of Pvis for 21 consecu-tive nights (triangles) can be used to illustrate the behaviorof Pvis. For planets of 2–3 day periods, most orbital phaseswould result in at least two transits occurring during night-time over the 21 days of observations, and the correspond-ing Pvis is therefore unity. Planets with longer periods willhave a smaller number of transits occur during the 21 dayspan of the run. Consequently, as the period increases twotransits will be visible at night for a smaller fraction of orbi-tal phases, resulting in generally lower Pvis for longer peri-ods. The downward spikes in Pvis at integer day periods(clearly visible for the 21 day case) illustrate the limitationsof a nighttime transit search done from a single observatory,since a percentage of the transits with integer day periodswill always occur during daylight. The effects of changingthe length of the observing run are illustrated by the othercurves shown in Figure 3a: the bars correspond to 14 con-secutive nights, and the dotted line corresponds to the actualtime coverage of the EXPLORE I transit search at CTIO in2001 June (11 nights, but only the equivalent of six nightshad good weather). Note that the Pvis simulations consider10.8 hr of continuous observing each night. However,observing for 10.8 hr each night of the run might not be pos-sible for all combinations of field declination and observa-tory latitude, because of the sliding of sidereal timethroughout the run. For an alternative definition and dis-cussion of Pvis, see Borucki et al. (2001) and Giampapa et al.(1995).

When planning a transit search, it is important to con-sider the mean planet detection efficiency Pvish i per observ-ing run and the mean planet detection efficiency per nightPvish i=N as a function of observing run length, as shown inFigures 3b and 3c. The solid line in each graph correspondsto the requirement that two transits be visible, while thedashed line corresponds to a requirement of one visibletransit and is shown as a reference; in both cases Pvish i is cal-culated for planets with 3–4.5 day periods. As one wouldexpect, the mean planet detection efficiency Pvish i increasesmonotonically as a function of observing-run length. Thisefficiency increase is very steep for runs of d25 days. Theeffective planet detection efficiency per night, Pvish i=N, hasa broad peak around 21 days. Thus, for a site with perfectweather, it would generally be most efficient to distributeobserving runs in blocks of 3 weeks. Note that Pvish i=Ndecreases very sharply for runs lasting less than 1 week. Alsonote that for a single transit detection, Pvis;1

� =N is highest

for the shortest observing runs and decreases monotonicallyfor longer runs, since the extra nights will result in repeattransit observations, which will not increase Pvis;1

� any

further.Based on equation (7), our estimates of Pvis, and our esti-

mate of Fp ¼ 0:00035 for planets of 3–4.5 day periods, wecan calculate the expected number of transit detections in a�2 week observing campaign (a reasonable limit for ashared 4 m class telescope). We estimate that during a per-fect 13 night run, Pvish i ¼ 0:3, and thus one transiting close-in giant planet will be discovered for every �10,000 stars

withd1% precision light curves. For a perfect 17 night run,Pvish i increases to 0.5, and one transiting planet detectionshould be expected for every �6000 stars observed withd1% photometric precision.

3.2. Maximizing Pvis

Maximizing Pvis is a main consideration for maximizingobserving efficiency. A useful transit search will be one thatproduces a clean set of candidates with minimal contamina-tion from false positive planet detections. In this section, weconsider possible strategies for allocating a given limitednumber of observing hours in the context of trying to pro-duce the cleanest set of planet candidates. Strategies canrange from carrying out a few observations per night over alarge number of nights to carrying out all observations in asingle observing run of consecutive nights. We focus on ascenario in which the telescope is not dedicated to transitsearches (e.g., a shared national or international facility).We argue that when observing time is limited to a few weeksin a given season, the cleanest set of transit candidates willbe obtained by conducting observations in one contiguousblock.

A transiting close-in planet with a period of �3–4 dayswill typically have very few transits occurring at night dur-ing a 2–3 week observing campaign. A well-sampled eclipselight curve and reliable detection are most easily achievedwith high-cadence observations in which an eclipse can bedetected in a single night (i.e., without folding the lightcurve). In principle, it is also possible to detect the 1% dipsin the star’s flux caused by a transiting planet as a periodicsignal in sparse observations taken over many nights or evenweeks, as long as extremely high photometric precision isachieved from night to night. Unless the observations havea very long baseline and many hundreds of data points areobtained for each star, the resulting phased light curve willmost likely not have enough in-transit data points to enablea measurement of the shape of the eclipse. Knowing theshape of the eclipse is critical for ruling out common con-taminants with periodic 1% dips, such as grazing binaries(see x 4.6; Seager & Mallen-Ornelas 2003). Such contami-nants can introduce many false positive detections to the listof planet candidates, making follow-up very inefficient andstatistical analysis very difficult. Therefore, when only a fewweeks of telescope time are available, it is best to obtainhigh-cadence observations in which well-sampled transitsare detectable in a single night.

If a transit is to be detected within a single night, it is bestto detect a full transit rather than a partial transit, sincethere are common systematic errors in the photometry thatcan mimic the beginning or end of a transit. Full transits arebest for determining transit length and shape, which are nec-essary to constrain system parameters and find good planetcandidates for follow-up (x 4.6). Full transits are only visiblewhen the middle of the transit is within the middle L� tT hrof the observations, where L is the number of hours of con-tinuous observations and tT is the total transit length inhours. This implies that observations should be taken for asmany continuous hours as possible. For example, considera series of observations lasting 4 hr at a time. Such a strategywould be extremely inefficient for finding transits, since a 3hr transit (the typical length for a close-in planet around aSun-like star) could only be detected if it was centered dur-ing the middle hour, i.e., just 25% of the observing time. In

10 See Extrasolar Planets Catalog, http://www.obspm.fr/encycl/catalog.html.


contrast, the transit detection efficiency for a series of 11 hrobservations is much higher, since a 3 hr transit would bedetected if centered in the middle 8 hr of observations, or73% of the observing time. Figure 4 illustrates the impor-tance of scheduling full nights of observations by comparingPvis for 227 hr of observations scheduled in either 21 fullnights of 10.8 hr or 76 nightly 3 hr segments.

It is best to schedule all nights in a contiguous block, sothat the field of choice is visible for as long as possible eachnight throughout the run; if the allocated nights are spreadout over many weeks, changes in sidereal time will cause thefield to be visible for only part of the night, thus decreasingPvis. In an ideal case, observations would be done from sev-eral observatories spread out in longitude, or from a loca-tion where it is possible to obtain continuous coverage ofthe field.

Period aliasing will be a severe problem if a given numberof nights are split into two observing runs separated by ayear instead of in one long block, since the period will beeffectively unknown for any systems that have only onetransit detected in each of the observing runs. Without aperiod determination, the characteristics of the eclipsingsystem will be unknown, and the interpretation of an RVfollow-up (with only a few data points) will be severelylimited.

We have argued in favor of allocating observing time inone contiguous block, which has the advantage that ahigh photometric precision is easier to achieve in a single

night rather than across several different nights. This strat-egy is well suited to the constraints of using large telescopesin shared national or international facilities, where a realis-tic time allocation in one season is limited to 2–3 weeks.Note that once the best planet candidates have been identi-fied, it is very useful to conduct additional observations inshort sets spread out over several weeks or months in orderto obtain an accurate period measurement and confirm thetransits.

An alternative strategy to the one we advocate abovewould be to take less frequent observations of the field overmany weeks or months and fold the light curve with a vari-ety of periods in order to search for transits. This strategy ismost suitable for private telescopes that can be dedicated toa photometric campaign over many weeks. By observing thefield for many seasons, the total number of photometricdata points taken in this way would eventually be the sameas in the strategy with contiguous high time sampling obser-vations, resulting in the same effective time sampling. Thisstrategy can gain statistical certainty by summing phaseddata from different transits, but it requires extremely highphotometric precision from night to night. The advantageof having sparse observations over a long baseline, however,is that the period can be determined to very high accuracy.For a good example of this strategy, see the OGLE IIIplanet transit search (Udalski et al. 2002b).

4. A FIELD TRANSIT SURVEY DESIGN: THEEXPLORE PROJECT

The general framework presented in the previous sectioncan be used in the design of any transit survey. In order todesign a specific transit search, a large number of interre-lated issues must be considered. In this section, we discussmany relevant factors that affect transit survey design anddescribe the specific choices made by the EXPLORE proj-ect. The EXPLORE project is a series of searches for transit-ing planets around Galactic plane stars. The mainconsiderations for the EXPLORE program design are tomaximize N R�ð Þ, Pdet, and Pvis (see eq. [7]) and minimizefalse positive detections, to obtain a high yield of actualplanets among the transit candidates. We maximize N R�ð Þby using 4 m class telescopes with large-format CCDmosaiccameras to look in the Galactic plane. We maximize Pdet

and minimize false positive detections by carrying out high-precision photometry with high time sampling on a singlefield with mostly main-sequence stars. We maximize Pvis bymonitoring the selected field for as many consecutive nightsas possible, for as long as possible each night. As part of theprogram design, we reduce the data and find planet candi-dates within a few weeks of the observations; this allows usto follow up planet candidates with RV measurements inthe same season, before orbital phase information is lost.The following subsections explain the details of theEXPLORE project experimental design.

4.1. Instrument Selection

From the estimates of transiting planet frequency and Pvis

presented in x 3.1, it is clear that many thousands of starsmust be monitored in order to detect a single planet. Themost efficient approach is to use a wide-field instrument toobserve an area in the sky with a high density of stars. Notethat the total number of pixels in the detector is crucial for

Fig. 4.—Comparison of Pvis for two different observing strategies: full-night or partial-night observations. Triangles show the fiducial model: Pvis

for 21 consecutive nights, with 10.8 hr each night (the same curve as in Fig.3a). (a) Pvis from the fiducial model compared to Pvis from a transit searchwith the same two-transit requirement and the same total observing timespread out over 76 nights with 3 hr per night. Almost no full transits aredetected twice in the partial-night, piecewise strategy. (b) Same as (a), butwith the requirement relaxed to detect four half-transits. Under the relaxedrequirement, both search strategies will detect a larger percentage of theexisting transits. However, the piecewise search will still find far fewer tran-sits than the 21 night search. As discussed in the text, the piecewise transitsearch strategy has additional serious practical problems, such as false par-tial transits caused by systematic errors in the photometry.


determining how many stars can be monitored with highphotometric precision (<1%). The relationship between thefield size and the telescope aperture and instrument effi-ciency is also very important, since a high time sampling isessential in order to pick the best planet candidates withoutmajor contamination from grazing binaries and blendedstars (see x 4.6). The EXPLORE project currently hassearches using theMOSAIC II camera at the CTIO 4m tele-scope (8K� 8K pixels, 360 � 360 field of view) and theCFH12K camera at the 3.6 m Canada-France-Hawaii Tele-scope (CFHT; 8K� 12K pixels, 280 � 420 field of view).Both searches have a high time sampling, with photometricmeasurements everyd3 minutes.

It is useful to outline some benefits of using a large tele-scope to conduct a narrow-angle transit search on relativelyfaint stars. First, deep transit searches contain a proportion-ately larger number of low-mass, low-luminosity stars thanshallow searches; in particular, deep transit searches probeintrinsically fainter stars than most ongoing RV searches.For a search conducted with a large-format mosaic CCDcamera on a 2–4 m telescope, a very large number of starscan be observed at once with good photometric precisionand time sampling. The large number of stars directlyincreases the chances of finding a transiting planet. Observ-ing many stars at once also helps improve the precision ofthe relative photometry. The point-spread function (PSF) isusually well sampled in the case of large telescopes, whichresults in higher photometric precision than is possible withthe typically undersampled PSF of small telescopes withlarge fields of view. Furthermore, there is less concern aboutpossible systematic errors introduced by differential extinc-tion across the relatively small field of a deep survey than inthe large-angle field of a small telescope. Finally, going tofainter apparent magnitudes reduces the proportion ofintrinsically bright and distant giant stars in the observedsample, leaving a larger fraction of main-sequence stars use-ful for finding planets. Conversely, there are two main dis-advantages associated with using a large telescope: (1) it ismore difficult to get large amounts of telescope time toimprove time coverage; and (2) since the stars are fainter, an8–10 m class telescope is required for the RV follow-up, andthe range of feasible follow-up studies is much more limitedthan for bright stars.

4.2. Field and Filter Selection

Lower main-sequence stars are the only stars in whichtransits by Jupiter-sized planets are easily detected, since forlarger stars the Rp=R�

� �2dip caused by a transiting planet

will be much smaller than 1%. In this section we discuss ourchoices of field and filter, both aimed at obtaining the largestnumber of lower main-sequence stars observable with betterthan 1% photometric precision.

In order to get a large number of lower main-sequencestars, we look at the Galactic plane. With large-formatCCD mosaic cameras on 4 m class telescopes, it is possibleto find Galactic plane fields with 100,000–500,000 starsdetected in 1–2 minutes of integration. The initial considera-tion for picking a field is the time of the year, since theGalactic plane must be visible from the observatory in ques-tion. For a given site, the best combination of long nightsand good weather is important for choosing the best monthfor the observations. It is also important to have the Galac-tic plane at a declination that will make it visible for as many

hours as possible in a given night. After observing time isallocated, we preselect an area in the Galactic plane by con-sidering the following: (1) sidereal time matched to rightascension, so that the field transits the meridian in the mid-dle of the night at the center of the observing run; (2) lowdust content, based on the dust map by Schlegel, Fink-beiner, & Davis (1998) and the CO map by Dame, Hart-mann, & Thaddeus (2001); (3) high stellar number countsfrom the USNO-A2.0 catalog (Monet et al. 1998) and theDigitized Sky Survey 2.11

In order to choose the best field within the preselectedregion, we take BVRI test images of several fields, using thesame instrument to be used for the actual search. We com-pute number counts and construct color-magnitude (CM)and color-color diagrams for each field and choose the fieldwith the best combination of the following factors: highestproportion of lower main-sequence stars, uniform and lowdust extinction, and smallest number of bright stars that sat-urate large areas in the CCD. As an example, Figure 5shows a CM diagram of one chip in the field of theEXPLORE I search at the CTIO 4 m telescope. An addi-tional consideration could possibly be to choose anuncrowded stellar field, so that the photometric precision isnot adversely affected by crowding. In practice, however,most Galactic plane fields are not significantly crowded forthe 1–2 minute exposures and 4 m class telescopes used inthe EXPLORE project. Moreover, effective photometryalgorithms have been developed to handle crowded-field rel-ative photometry (see, e.g., difference imaging by Alard &

Fig. 5.—CM diagram for chip 1 from the EXPLORE I field. The giantbranch and main sequence are clearly visible, showing that the fraction ofgiant stars is small. Most stars fainter than magnitude I ¼ 18 are too faintfor better than 1% photometry, but faint-star data points in the light curvescan be co-added to get better precision.

11 The Digitized Sky Survey was produced at the Space Telescope Sci-ence Institute under US government grant NAGW-2166.


Lupton 1998; Wozniak 2000; Udalski et al. 2002b). Figure 6shows an example of a small region (1/1400 of the totalarea) in the EXPLORE I field, which is located in the Galac-tic plane at l ¼ �27=8, b ¼ �2=7. Note that a large fractionof the stars are relatively isolated and that a significant frac-tion of the image area is free from stars, which permits agood determination of the sky level (crucial for faint-starphotometry).

An important consideration is to minimize the number ofgiant stars, since they are too large to be useful for planettransit detection [since RJ=R�ð Þ25 1%] and a major sourceof contamination in shallow (mag d 13), wide-field transitsurveys (W. Borucki 2001, private communication; D.Latham 2001, private communication). In the case of theEXPLORE project, we minimize the proportion of contam-inating giant stars in our sample by observing the Galacticplane with deep images (e.g., 15dId20). A giant star wouldhave to be nearly outside the Galaxy in order to have thefaint apparent magnitude of most stars in our survey; forexample, a K5 giant with an apparent magnitude of I � 17(assuming 1 mag of extinction in I ) would be at a distanceof 54 kpc, where the Galactic stellar density is extremelylow. Also, we select fields away from the bulge, in order toavoid bright giant stars, which would often saturate theCCD and increase crowding in the field.

Using a very red filter allows us to maximize the numberof lower main-sequence stars (i.e., relatively small stars), forwhich transits are most easily detected for a given sizeplanet. Specifically, using an I-band filter increases thecounts of stars of later type than the Sun by a factor of 2–6

over using an R-band filter (for a fixed magnitude range).Observing in the I band minimizes the effects of absorptionby interstellar dust, as compared to bluer bandpasses.Finally, the choice of the I band produces light curves withthe least significant limb darkening among standard BVRIfilters (see Fig. 2); this is extremely useful for selecting eclip-ses with clear flat bottoms (see x 4.6.1) and for deriving thebest transit parameters without significant dependence onuncertain limb-darkening models (see x 2).

4.3. Photometric Precision and Time Sampling

High photometric precision and high time sampling arecrucial in order to identify the best set of transiting planetcandidates with minimal contamination, as was described inx 2. In the EXPLORE project, we achieve both high photo-metric precision and high time sampling by monitoring asingle field throughout the observing run, with exposurestaken every �3 minutes. Care is taken so that the field posi-tion does not shift in the CCD, bymaking small adjustmentsto the pointing throughout the night, and average net shiftsin the field position are kept to less than 100. This is done inorder to minimize photometric errors introduced by resid-ual differences in the CCD response across the chip that arenot completely taken away by the flat-field correction andto simplify the photometry pipeline algorithm.

Observing a single field is the strategy that achieves thehighest time sampling. A more complicated strategy, inwhich several fields are monitored at once by switching fromfield to field, would significantly decrease time sampling ofeach field. Also, setting up the position and guiding of eachfield many times throughout the night would likely lead to alarge waste of observing time. Another alternative strategy,switching fields only once or twice throughout the night,would mean that each field would only be observed for 3–5hr at a time; this would significantly reduce the transit detec-tion efficiency Pvis, as discussed in x 3.2.

We have developed a customized pipeline to performhigh-precision photometry of faint stars in dense fields witha well-sampled PSF. Full details of our photometry pipelinewill be described in H. Yee et al. (2003, in preparation), andwe present a brief summary of the algorithm here. A key fea-ture of our high-precision photometry algorithm is the useof relatively small apertures (about a factor of 2–3 largerthan the seeing disk, i.e., a diameter of 200–300) for measuringthe flux. This stems from the requirement to minimize thecontribution of sky noise for stars that are not significantlybrighter than the sky (as faint as I � 19). The crucial consid-eration in obtaining high-precision relative photometrywhen using such small apertures is the exact placement ofthe center of the aperture relative to the centroid of the stars.We achieve this high-precision aperture placement by usingan iterative sinc-shift algorithm to resample each star so thatthe central 3� 3 pixels sample the PSF symmetrically (Yee1988). A photometric growth curve for each object is thenderived, using integer pixel apertures on the resampledimage of the star. The resampling is equivalent to placing allthe stars within the photometry aperture in an identicalmanner, allowing for relative photometry to be carried outusing much smaller apertures than is customary. The photo-metric measurements of each star are then put on a relativesystem, by comparing them with a set of reference starsdetermined using an iterative algorithm to find the moststable stars in a given region of the CCD. Light curves are

Fig. 6.—Sample 6000 � 6000 square of a survey image. The wholeMOSAIC II field is 360 � 360, or �1400 times this area. Note that a largefraction of the stars are relatively isolated and that a significant fraction ofthe image area is free from stars, which still permits a good determinationof the sky level. The arrows point to three stars with different magnitudes.The two at I ¼ 16:2 and 18.0 span the range of most of our stars suitablefor high-precision photometry. The star with I ¼ 19:9 is too faint for betterthan 1% photometry in a single frame.


finally produced based on the relative photometry. Ex-amples of the high-quality light curves achieved using thefirst version of our pipeline are shown in Figures 7 and 8and further discussed in x 5.

4.4. Data Reduction Strategy

A key feature of the EXPLORE project is that the datareduction and analysis are done on a short timescale. Inorder to produce light curves within 1–2 weeks of the end ofan observing run, we have developed a pipeline that runs ona dedicated computer cluster. Our pipeline consists of cus-tom-written programs to do image preprocessing, aperturephotometry, and relative photometry and to generate lightcurves. The only steps that currently require significanthuman intervention are visual verification of the automaticobject finding performed with the program PPP (Yee 1991)to create a star catalog and finding the best parameters forthe relative photometry. The latter step will eventually bedone automatically as well. The main bottleneck that cur-

rently prevents us from reducing data in real time (which isour eventual goal) is the long time it takes to read the rawdata tapes written at the telescope into the computer clusterwhere the data are reduced. A main motivation for the fastdata reduction is that follow-up RV observations of transit-ing planet candidates are best interpreted if done in the sameseason when the phase of the orbit is known. For a 2 weekor shorter observing run, the baseline for determining theperiod is small, so that typical errors in the period will accu-mulate over a year and the phase will likely be lost.

4.5. Follow-up Radial VelocityMeasurements

Late M dwarfs (M 80 MJ), brown dwarfs (13 MJ <M < 80 MJ), and gas giant planets (M � 13 MJ) are all ofsimilar sizes because of a competition between Coulombforce effects (R � M1=3) and electron degeneracy pressureeffects (R � M�1=3) (Hubbard, Burrows, & Lunine 2002).Hence, transits alone are not enough to determine that atransiting companion is actually a planet, even if the radius

Fig. 7.—Two examples of the high time sampling and high photometric precision light curves from the EXPLORE I search. The rows correspond to differ-ent nights of data, with the night numbers listed at the far right. Note that the photometric precision varies from night to night and on some nights is especiallypoor because of clouds and bad seeing. The dome was closed on a large portion of nights 1, 5, and 11 and for all of night 2. There is a one-night gap in the timeassignment between nights 1 and 2. Left: Many low-amplitude variable stars, such as this � Scuti star, are found in our data set. Right: Light curve of a likelygrazing, eclipsing binary star from the EXPLORE I search. The round bottom and very sloped ingress and egress are indicative of a grazing binary system.The photometric precision and high time sampling of our data are good enough to rule out grazing binary stars, a common contaminant in other transit sur-veys (W. Borucki 2001, private communication). Note the scale on the y-axis, which clearly shows that our relative photometry reaches a precision of consider-ably better than 1%.


is constrained to be �0.1–0.15 R�. RV measurements aretherefore needed to determine the mass, and thus the nature,of the orbiting companion. RV measurements are also use-ful to rule out grazing binaries and other possible contami-nants that mimic the transit signature, which can becommon in the case of noisy light curves (x 4.6). The transitsearch method with follow-up RV confirmation is verypowerful, because every planet found has a measured radiusand an absolute mass. Obtaining a mass measurement fortransiting planet candidates is facilitated by the fact that theorbital period and phase are known a priori, and thereforeobservations can be conducted at predetermined times suchthat the RV differences are maximized. In the case of faintstars, where planet masses lower than 1–2 MJ cannot bedetected because of limitations in the currently achievableRV precision, it is still possible to determine a mass upperlimit of a fewMJ (see, e.g., G. Mallen-Ornelas et al. 2003, inpreparation). An actual mass measurement is generallyrequired for confirmation of the presence of a planet. How-ever, a mass upper limit could still be used to make a casefor the presence of a planet as long as all possible contami-

nants to the transit signature can be ruled out with confidence(see x 4.6).

The amplitude of the RV variations of a star in the pres-ence of a less massive companion in a circular, edge-on orbitis

K ¼ 2�G

P

� �1=3 m2

m2=31

: ð8Þ

Here P is the period andm1 andm2 are the primary and sec-ondary masses, respectively. Because transiting planetorbits are seen almost completely edge-on (i � 90), the fullRV variation is along the line of sight. A G2 V star(M ¼ M�) with an 80 MJ M dwarf and a 13 MJ browndwarf companion with an orbital distance D ¼ 0:05 AU(corresponding to P ¼ 4:08 days) will show RV amplitudesof 10.1 and 1.6 km s�1, respectively. Thus, both M dwarfsand brown dwarfs are very easy to rule out with better than500 m s�1 RV precision (even for stars more massive thanG2 V and for stars with planets in slightly longer periodorbits). A simulated example is shown in Figure 9. Note that

Fig. 8.—Two eclipsing binary star systems from the EXPLORE I search.Left: The eclipse depth in this light curve is the same (�2%) as that of a giant planettransiting a Sun-sized star. However, the long transit duration (�7% of the total orbital period), together with the 2.2 day orbital period, indicates that the pri-mary star has a very large radius, and therefore the companion is too large to be a planet. Right: Light curve showing two flat-bottomed eclipses. The eclipsedepth is 3%, and this star appears to have a spectral type of early K. The flat bottom of this light curve means that the companion is fully superimposed on theprimary during the transit. As discussed in the text, the companion in this system is not a planet.


an RV precision of 500 m s�1 is easily attainable with anechelle spectrograph on an 8 m class telescope, even for therelatively faint stars (Id18) in the EXPLORE project.

We note that planet searches that use RV measurementsto find planets reach a precision of a few meters per second(see, e.g., Butler et al. 1996; Pepe et al. 2000). This level ofprecision is important when one is trying to detect possibleperiodic changes in the RV and measure orbital parameters,but is not necessary when trying to distinguish variations ofwidely different amplitudes for a system with a knownperiod and phase. RV follow-up confirmation of transitcandidates can be extremely efficient. As shown in Figure10, only a handful of RV points at a judiciously chosen timeare needed to constrain the companion’s mass, as long asthe period and phase are known. Knowing the transitingcompanion’s orbital phase is very important when interpret-ing the RV measurements. A small error in the period mea-surement from a two-transit discovery light curve willrapidly accumulate with each orbit, to give a phase errorthat increases linearly with time. For instance, in 1 year aplanet with a 3� 0:007 day period (i.e., a 10 minute uncer-tainty) will have an accumulated error of 0.85 days, or 0.3 inphase. Thus, for a�2 week observing run (with only a short

baseline for period determination) it is best to do follow-upobservations in the same season the discovery light curve istaken, since otherwise a second imaging run will be requireda year after the discovery observations simply to recover thephase.

4.6. Minimizing Potential Contamination to theTransit Signature

It is important to select the very best candidates for theRV follow-up in order to have a high yield of planets. Thethree main characteristics intrinsic to transiting planet lightcurves are (1) they show very shallow eclipses; (2) the eclip-ses have a flat bottom in a bandpass where limb darkening isnegligible; and (3) there is no secondary eclipse. Four differ-ent types of systems could be confused with a transitingplanet: grazing eclipsing binary stars; an eclipsing binarysystem consisting of a large primary star with a small stellarcompanion; an eclipsing binary star contaminated by thelight of a third blended star; and a transiting brown dwarfor late M dwarf. This section discusses the first three typesof possible contaminants and some ways to differentiatethem from bona fide planet transit light curves before theRV follow-up. Contaminants of the fourth type (browndwarfs or late M dwarfs) can only be distinguished by RVfollow-up observations, but they are of great interest in theirown right.

4.6.1. Ruling Out Grazing Eclipsing Binaries

At certain orbital inclinations, a grazing eclipsing binarystar can produce the sought-after drop in brightness of 1%when a small part of the companion crosses the primarystar. If the stars are of similar surface brightness, or if onestar has a much larger surface brightness than the other,then it might not be possible to discern any secondary eclip-ses in the data. Hence, these very shallow eclipses can be themajor cause of false positive planet candidates in sometransit searches (W. Borucki 2001, private communication;D. Latham 2001, private communication). Even though theeclipse depth might be the same as that of a transitingplanet, a grazing eclipse from a binary star system has a dif-ferent shape. As illustrated in Figure 11, a triangular lightcurve with a rounded bottom is indicative of a grazingbinary system, since the stellar companion only partiallyoverlaps the primary star’s disk. In contrast, a transitingplanet has a flat-bottomed eclipse, which indicates that theeclipsing companion is entirely superimposed on the disk ofthe primary star. Note that for a small range of orbital incli-nations, the transiting planet is never fully superimposed onthe primary star and produces an eclipse with a shape anddepth very similar to those of a grazing binary (Fig. 11b).However, a partial transit geometry is rare for Rp5R�, andin most cases the depth of a partial transit will be much lessthan 1%. Thus, for practical purposes, even if partial planettransits are not included in the follow-up RVmeasurements,they can be accounted for statistically.

High time sampling and high-precision photometry arerequired in order to determine the shape of the eclipse andthus distinguish between the shallow, round eclipses causedby grazing binary stars and eclipses with flat bottoms thatmight be caused by transiting planets. Distinguishingbetween the two types of eclipses is easiest when the lightcurve is taken in a bandpass that is not severely affected bylimb darkening (Fig. 2).

Fig. 9.—Model RV curves for companions of different mass orbiting asolar-mass star withD ¼ 0:05 AU (corresponding to P ¼ 4:08 days). Over-plotted are simulated RV data points for a 2.5 MJ companion with 100 ms�1 rms noise and well spaced in phase. The adopted error bar is 100 m s�1,approximately what is attainable with hour-long exposures on faint stars(V � 18) with 8 m class telescopes (G.Mallen-Ornelas et al. 2003, in prepa-ration). The two dotted lines in each panel show the uncertainty in orbitalphase corresponding to accumulated 20 minute errors in the 4 day periodover 4 months. It is evident from the top two panels that even with an RVprecision of�500m s�1, transits due to a stellar companion or brown dwarfcompanion can be easily ruled out. The y-axes are different for each panel;the dotted lines look similar in each panel because the RV amplitude scaleslinearly with companionmass (eq. [8]).


To distinguish grazing binaries from transit candidates,the EXPLORE project takes observations at a very highrate (every d3 minutes) and uses an I-band filter, so thatlimb darkening is not significant. Figure 7 (right-hand panel)shows an example of a grazing binary system light curvefrom the EXPLORE I search, as evidenced from the roundbottom and highly sloped ingress and egress of the eclipses.Although grazing binaries can be trivially ruled out by fol-low-up RV measurements, it is essential to have flat-bot-tomed eclipse candidates for a high yield of actual planetsamong the candidates chosen for follow-up.

4.6.2. Ruling Out Eclipsing Binary SystemsWith a LargePrimary Star

A small star eclipsing a large star can have the sameeclipse depth as a Jupiter-sized planet eclipsing a Sun-sizedstar. For example, an M4 dwarf eclipsing an F0 star willcause a 1% deep eclipse with a flat bottom. A secondaryeclipse is a definite indicator of an eclipsing binary star sys-tem, regardless of the eclipse depth. However, if the surfacebrightness ratio of the primary to secondary star is large,the resulting secondary eclipse will not be visible in the lightcurve. A binary star system with a large primary is easy torule out from the length of the eclipse alone. Figure 8 (left-hand panel) shows a clear example of a 2% deep eclipse forwhich the Jupiter-sized planet/Sun-like star hypothesis canbe immediately ruled out. The eclipse has a 2.2 day periodand lasts 5.5 hr, which is much longer than an eclipse causedby a planet with the same period orbiting a solar-type orsmaller star. Another way to rule out eclipsing binaries witha large primary is to consider the unique solution of a light

curve with two or more transits. With a light curve of suffi-cient photometric precision and time sampling, the stellarsize and mass can be derived using the five equations in x 2.Note that in the case of a giant star, the eclipse will be muchlonger than for a main-sequence star of the same mass; solv-ing the five equations in x 2 for an evolved star using themain-sequence mass-radius relation will give an upper limitto the actual R� (Seager & Mallen-Ornelas 2003). This willlikely be enough to rule out the planet hypothesis, and it canbe confirmed using the color of the star. Alternatively, abinary star system with a large primary can also be ruled outby spectral classification of the star.

4.6.3. Ruling Out the Presence of a Contaminating Blended Star

A flat-bottomed and relatively deep eclipse from a com-panion star fully superimposed on its larger primary willappear shallower if light from a third blended star is presentin the light curve. The contaminating blended star could bepresent because of a chance alignment with the eclipsingbinary system or, more likely if the field is not too crowded,the contaminating star could be a component of an unre-solved multiple star system.

The unique solution of a light curve with two or moretransits can be used to identify an eclipse contaminated by ablended star. The length of the ingress or egress is set by acombination of Rc=R� and the projected impact parameterD=R�ð Þ cos i at which the companion crosses the center ofthe stellar disk (where Rc is the radius of the eclipsing com-panion, R� is the radius of the central star, D is the orbitaldistance, and i is the orbital inclination). A 1% eclipse withan ingress and egress that are long compared to the total

Fig. 10.—Simulated example, showing that only a few RV measurements are needed to constrain the mass of a close-in extrasolar giant planet if the phaseof the eclipsing system is known. Solid lines show a model RV curve for a 2.5MJ planet orbiting a solar-mass star at 0.05 AU (which corresponds to a 4.08 dayperiod). (a)–(c) Theoretical RV points with added Gaussian noise of � ¼ 100 m s�1. Panels (a)–(c) show 5, 10, and 30 RV points, respectively. (d )–( f ) Same as(a)–(c), but with a noise of 50 m s�1.


eclipse duration can be produced by the following two cases:(1) a planet crossing a Sun-sized star with impact parameterD=R�ð Þ cos id1 (i.e., the planet transits near the stellarlimb), and (2) a small star eclipsing a Sun-sized star, withadditional light from a blended star contaminating the lightcurve. In case 1, the long ingress and egress are due to thefact that the planet transits close to the limb and is partiallysuperimposed on the stellar disk for a relatively long time.This is illustrated in Figure 12 for the i ¼ 86 case (top linein Fig. 12a, dashed line in Figs. 12b and 12c). Note that onlya very small range of inclinations will result in a transit witha proportionally long ingress/egress; therefore, for bonafide planets, having a transit with a long ingress/egress ismuch less likely than having one with a short ingress/egress.In case 2, the long ingress and egress are due to the fact thata larger companion will necessarily take a long time to com-pletely cross the primary star’s limb, even for a centraleclipse (middle line in Fig. 12a). Normally, cases 1 and 2would not be confused, because the larger companion incase 2 will produce a much deeper eclipse than the small

companion in case 1. However, if the light curve is contami-nated with additional light from a bright blended star, theobserved eclipse depth will be reduced, and thus case 2 canmimic the shallow transit in case 1. The surface brightnessratio of the primary and secondary stars in the eclipsingbinary system in case 2 can easily be large enough so thatthe secondary eclipse is lost in the photometric noise of theblended light curve.

The case of an eclipsing binary system plus a blended starcan be ruled out by using a spectral type to complement theunique solution to the equations in x 2, provided that thelight curve has two eclipses with good photometric precisionand time sampling. In the presence of a blended star, theunique solution gives a stellar mass and radius that are dif-ferent from the mass and radius derived from the spectraltype. Specifically, the unique solution will give an errone-ously large primary star. This is because the inferred inclina-tion of the orbit will result in a solution in which the planettransits close to the stellar limb and therefore seems to begoing across a relatively small length of the primary star. Aspectral type that is inconsistent with the primary star’smass and radius as derived from the unique solution to thelight curve is therefore a strong indication that there is acontaminating blended star.

Even without a spectral type, the planet hypothesis can beruled out from the light curve alone in the case of an eclips-ing binary system plus a blended star, based on the overesti-mated primary radius. The large stellar radius, togetherwith the measured �1% transit depth, will usually give acompanion radius too large to be a planet. In other words,for a given eclipse length, the inferred stellar radius will bemuch larger than the true radius, and the system will appearto be a large star with a smaller stellar companion transitingvery close to the stellar limb. The uncertainties in the param-eters derived from the unique solution can be large, and insome blended cases the companion can appear to have a sizethat is almost compatible with that of a giant planet. There-fore, obtaining a spectral type prior to the RV follow-up isvery worthwhile, especially when the ingress and egress of atransit are long compared to the total transit duration.

If the data are noisy, it might not be possible to rule out acontaminating star from the light curve, even if a spectraltype is available. In this case, RV follow-up observationscan be used to rule out a planet by identifying two compo-nents in the spectrum: (1) a constant-velocity componentcoming from the blended star, and (2) a component fromthe primary star in the eclipsing binary system that willexhibit large RV changes with the period and phase corre-sponding to the transits. If a bright guide star or laser guid-ing is available, it is also possible to test the blendhypothesis by obtaining a very high resolution image withan adaptive optics system to look for close companions.Typical stars in the EXPLORE search are 1–2 kpc away, soa resolution of 0>05 would be adequate to detect compan-ions at 50–100 AU.

4.6.4. Ruling Out Other Sources of Contamination

Stellar secular variability might be thought of as a con-cern in the search for transits. However, a transit signal isvery different from most intrinsic variability of a star. Inparticular, note that a transiting planet causes a drop inbrightness during only a small percentage (<5%) of the totaltime. A large spot on the star’s surface, for example, can

Fig. 11.—Comparison of a planet transit light curve and a grazing,eclipsing binary star light curve. The planet has R ¼ 1:4 RJ orbiting a starwithR� ¼ R�, at an orbital distance ofD ¼ 0:05 AU, with a correspondingperiod of 4.08 days. The grazing binary system is composed of two identicalSun-like stars with twice the planet’s period. (a) The eclipsing binary starsystem light curve (thin solid line), with solar limb darkening at I, has a graz-ing angle of 85=11, chosen to match a planet transit (at i ¼ 90) depth in theI band (dashed line). Also shown are transit light curves neglecting limbdarkening (thick solid line) and with solar limb darkening at 0.45 lm (dottedline). (b) The grazing, eclipsing binary system light curve (thin solid line) hasa grazing angle 85=024, chosen to match the depth of a partial planet transit(dot-dashed line; the same partial planet transit curve is shown by the dot-dashed line in Fig. 12c). Panel (a) illustrates that with high-precisionphotometry and good time sampling, it should be possible to distinguish afull planet transit from a grazing binary star eclipse. Panel (b) shows that inthe case in which a planet produces a partial transit, the light curve can benearly identical to that produced by a grazing binary star. However, sinceRp5R� for giant planets around Sun-like stars, partial planet transitsare rare.


cause a periodic drop in stellar brightness with the stellarrotation period. However, the probability is low that thecombination of the spot’s position on the star and the incli-nation of rotational axis will conspire to produce a drop inbrightness for a time much shorter than half of the rota-tional period. Moreover, a star with one large spot is alsolikely to exhibit other variability. Observations at differentwavelengths should distinguish variability due to star spotsfrom a relatively gray planet transit. See Giampapa et al.(1995) for a thorough discussion of this. Another concernmight be confusion arising from brown dwarf or M dwarfeclipses; however, these are of great interest themselves andwill be revealed by follow-up RV observations.

5. THE EXPLORE I TRANSIT SEARCH

We now turn to early results from the first EXPLOREtransit search. The EXPLORE I search was conducted dur-ing 11 nights on the CTIO 4 m telescope with the MOSAICII camera, on 2001 May 30 and June 1–10. The MOSAIC IIcamera is made up of eight 2K� 4K thinned CCDs with 15lmpixels, which corresponds to 0>27 pixel�1 and a 360 � 360

field of view at the CTIO 4 m prime focus. We observed asingle field near the Galactic plane (l ¼ �27=8, b ¼ �2=7,� = 16h27m28 900, � =�5252040>0 [J2000.0]) with 100,000stars down to I ¼ 18:2 and �350,000 stars to I ¼ 21. Atthe latitude of CTIO, it was possible to observe theEXPLORE I field through the whole night, which resultedin as much as 10.8 hr of coverage on good nights. A total of�1800 images were obtained, although �200 of these wereof very low quality because of poor weather conditions.The EXPLORE I field is located in an area of low and

uniform dust extinction and sits in the Norma patch of theOGLEmicrolensing search (Udalski, Kubiak, & Szymanski1997).12 It will thus be possible to study long-term variabil-ity of objects in our sample when the OGLE data becomepublic.

We had clear weather for approximately six nights andhad variable thick clouds and rain or fog during the rest ofthe run. For a percentage of the stars, however, the datataken during bad weather still yielded useful photometry.Our typical exposure time was 60 s for good conditions andwas adjusted up to compensate for cloud cover or down ifthe sky was too bright because of moonlit clouds. The detec-tor read time plus overhead was 101 s, so we typicallyobtained one photometric data point every 161 s, or 2.7minutes. Out of the 100,000 stars to I � 18:2 that were ini-tially processed, we have so far produced 37,000 light curveswith 0.2%–1% rms over a good night (Fig. 13). Figure 14shows a histogram of the number of stars with photometricprecision better than 0.5%, 1%, and 1.5%. Our best candi-dates tend to have �0.5% photometry and 15dId17.Figures 7 and 8 show sample light curves.

The high photometric precision and high time samplingof our data are crucial for distinguishing between transitsand grazing eclipsing binary stars (see x 4.6.1) and necessaryfor deriving the system parameters from the light curve itself(see x 2; Seager & Mallen-Ornelas 2003). Typical transitlengths for known close-in planets would be 2.5–3 hr (e.g.,HD 209458b), yielding �50 points at minimum light to0.2%–1% precision (e.g., 7 � detections per full transit with

Fig. 12.—(a) Schematic diagram of transiting planets with different orbital inclinations. First through fourth contacts are indicated. For lower inclinations,the eclipses are shorter and ingress and egress are longer than for central transits. Planet and star are to scale for Rp ¼ 1:4 RJ and R� ¼ R�. (b) Transit lightcurves for the three inclinations shown in (a), for parameters Rp ¼ 1:4 RJ, R� ¼ R�, andD ¼ 0:05 AU. The orbital inclinations, from top to bottom are, 85,86, and 90. The top curve is round-bottomed because the planet is only partially transiting the star (see [a]). At all other orbital inclinations the transit lightcurve is flat, indicating that the planet is fully superimposed on the parent star. (c) Same as (b), but with solar limb darkening at 0.8 lm adopted.

12 See http://bulge.princeton.edu/~ogle/ogle2/fields.html.


�m ¼ 0:002 0:01). Note that a transit of an HD 209458b–like planet (Rp ¼ 1:35 RJ;

13 Brown et al. 2001; see also Cody& Sasselov 2002) around a K0 star (typical for our field) willproduce an easily detectable 2.6% deep transit.

The data were reduced as outlined in x 4.3, and candidateswere found by visual examination within 7 weeks after theobserving run. The fast processing and examination of thedata were done so that we could obtain RV follow-up meas-urements on our best candidates before phase informationwas lost. The final analysis of the data will be done with anautomatic transit detection program and extensive tests todetermine detection thresholds and completeness. A pre-liminary estimate, based on the window function for thepoor weather conditions during our run, results in �1expected planet for the 37,000 stars examined so far. A stat-istical analysis and constraints on close-in giant planet fre-quency will follow in a later paper. We followed up ourthree most promising candidates in 2001 September with 19hr of Director’s Discretionary Time on the VLT+UVES.Details are discussed in G. Mallen-Ornelas et al. (2003, inpreparation).

We found several systems with eclipses of 1.5%–3%depth. Three of these systems each had two flat-bottomed

Fig. 13.—Photometric precision from the EXPLORE I search on the CTIO 4 mMOSAIC camera. Log rms error vs. Imagnitude for a good chip (left) anda not-so-good chip (right; with shallower electron wells), on a full good night (bottom) and a full mediocre night (top; variable clouds and variable bad seeing)in our 2001 June run. Points below the horizontal line correspond to stars with better than 1% rms photometry that are suitable for planet detection. On a badnight (not shown), there are no light curves with this precision. The dense clusters of points at bright magnitudes and the dense clusters of points with log rms�0.8 are due to saturated stars.

13 WhereRJ is Jupiter’s equatorial radius.Fig. 14.—Histogram of the number of stars with rms photometric preci-

sion better than 0.5% (solid line), 1% (dotted line), and 1.5% (dashed line).


eclipses (EXP1 J1628-52c07s24763, EXP1 J1628-52c07s18161, and EXP1 J1628-52c01s52805). In addition,we found several eclipsing systems in which the noise wastoo great to determine the eclipse shape. Of the three starsshowing two eclipses with flat-bottoms, the first (EXP1J1628-52c07s24763) is clearly ruled out as a planet candi-date because the transit is too long, implying a large primary(Fig. 8, left-hand panel). The second system (EXP1 J1628-52c07s18161) has a noisy light curve, and more data areneeded to confirm the transits.

The third system (EXP1 J1628-52c01s52805), shown inFigure 8 (right-hand panel), has two high-quality eclipseswith clear flat bottoms. Based on follow-up RV data, wenow know that the eclipse is not caused by a planet (G.Mallen-Ornelas et al. 2003, in preparation), but it is still aninteresting system worth discussing. As is seen in Figure 8(right-hand panel), the light curve has two 3% deep, flat-bot-tomed eclipses due to a companion in a 2.23 day periodorbit, and there is no sign of a secondary eclipse at P/2. Theclear flat-bottomed eclipses indicate that the companiondisk is fully superimposed on the parent star. BVRIphotometry and a classification spectrum indicate that thisis an early K star on the main sequence (M� � 0:8 M� andR � 0:85 R�). A straightforward interpretation would bethat the eclipsing companion is a 1.4 RJ planet orbiting at 8stellar radii from the parent star.

This third system’s eclipse ingress and egress are of nearlythe same duration as the flat part, which could be because ofa planet with an orbital inclination such that it transits closeto the stellar limb (Fig. 12). However, the almost equallength of the ingress/egress and flat part of the light curve isa cause for concern: applying the uniqueness criteria for atwo-transit light curve (x 2) gives the stellar radius of an Fstar, while the spectral type indicates a K star. As was dis-cussed in x 4.6.3, this is an indication of the likely presenceof light from a blended companion diluting the light curveand making an otherwise deep eclipse appear shallow. SinceEXP1 J1628-52c01s52805 was our cleanest two-transit lightcurve, it was included in the follow-up RV observations.The observations revealed a contaminating star with con-stant RV, blended with a fainter star showing RV variationswith a 2.23 day period and a �60 km s�1 amplitude (see G.Mallen-Ornelas et al. 2003, in preparation). Thus, for thiseclipsing system, the planet hypothesis was ruled out.

6. SUMMARY AND CONCLUSION

The first successful detection of a planet by the transitmethod will mark a huge step forward in planet detectionand characterization. Transit searches will allow planets tobe found around a variety of stellar types and in a variety ofenvironments. Every planet discovered by the transitmethod will have a measured radius and absolute mass(together with RV follow-up measurements) that will pro-vide key constraints for planetary formation and evolutionmodels. In order to find a transiting planet, many thousandsof stars must be monitored for e2 weeks with high photo-metric precision and high time sampling. Although manygroups around the world are conducting transit searchesand a few groups have produced transiting planet candi-dates, no planets have yet been confirmed by a mass mea-surement. In addition to searching for planets, transitsearches are providing databases of variable star lightcurves with unprecedented time sampling and photometric

precision. We note that these databases will also beextremely useful for finding moving objects, such as aster-oids, finding short-duration microlensing events, such asthose from free-floating planets, and generating deep imagesof stellar fields.

The EXPLORE project is a series of transit searchesaround stars in the Galactic plane, using wide-field, large-format CCD cameras on 4 m class telescopes. We have pre-sented the EXPLORE transit search strategy, whichinvolves monitoring a single low Galactic latitude field con-tinuously with high-precision photometry for as many con-secutive nights as possible. We have shown that continuousmonitoring is the most efficient way to find planet transitcandidates from the ground when only a few weeks of tele-scope time are available. One of the key aspects of our strat-egy is high time sampling, which allows us to use the uniquesolution of a light curve that has at least two flat-bottomedeclipses in order to select the very best planet candidates(e.g., by ruling out grazing binary stars). A unique aspect ofour strategy among current planet transit searches is that wefollow up our transit candidates with RV measurements inthe same season in which they were discovered. This is keyto interpreting the RV observations, because only a few RVpoints are needed to rule out brown dwarfs and M dwarfs ifthe phase is known. The EXPLORE team has so far con-ducted a southern search with the CTIO 4 m telescope (2001June), with RV follow-up done on the VLT, as well as anorthern search with the CFHT 3.6 m telescope (2001December), with RV follow-up done on the Keck telescope.

We have reported early results of the EXPLORE I search,which used the CTIO 4m telescope andMOSAIC II camerafor 11 nights in 2001 May–June, out of which about six hadexcellent weather. We have reached a photometric precisionof 0.2%–1%, with points every 2.7 minutes for a sample of37,000 stars in the Galactic plane at l ¼ �27=8, b ¼ �2=7(� = 16h27m28 900, � =�5252040>0 [J2000.0]). For thisnumber of stars and our limited time coverage, we expect tofind �1 transiting planet. We have followed up three planettransit candidates with VLT UVES in 2001 September. Theresults of the UVES follow-up are described in G. Mallen-Ornelas et al. (2003, in preparation).

We thank the following people for valuable contributionsto this project: George Hau, Sara Ellison, Jon Willis, Lau-rent Eyer, Fred Courbin, Beatriz Barbuy, and Mike Wev-rick. We also thank Bohdan Paczynski, Robert Lupton,Scott Gaudi, and David Charbonneau for useful conversa-tions. We thank the Red-Sequence Cluster Survey project(Howard Yee, Mike Gladders, Felipe Barrientos, and PatHall) for obtaining test field images during their CTIO runsfor selecting the EXPLORE I field. Andrzej Udalski, Prze-myslaw Wozniak, and Bohdan Paczynski kindly providedus with OGLE color images, which were helpful in selectingthe EXPLORE I field. We thank David Spergel for lettingus use Princeton’s Beowulf cluster (Fluffy), which was essen-tial for the EXPLORE I project. We would like to thankboth the US and Chile CTIO time allocation committees forgenerous allocations. The staff at CTIO were very helpful inaccommodating a run that pushed the telescope dataacquisition to its limits. G.M.-O. thanks Scott Tremaine forvery valuable discussions, advice, and encouragement. G.M.-O. thanks John Bahcall and the IAS for generous sup-port during a visit when much of this work was carried out.S. S. thanks John Bahcall for his very strong interest and


support for this project. G. M.-O. is supported in part byFundacion Andes and FONDECYT, and S. S. is supportedby the W. M. Keck Foundation. The research of H. K. C.Y. andM.D. G. is supported in part by NSERC and a grant

from the University of Toronto. D. M. is supported byFONDAP Center for Astrophysics (Project 15010003) andFONDECYT. The National Center for AtmosphericResearch is supported by the National Science Foundation.

REFERENCES

Alard, C., & Lupton, R. H. 1998, ApJ, 503, 325Borucki, W. J., Caldwell, D., Koch, D. G., Webster, L. D., Jenkins, J. M.,Ninkov, Z., & Showen, R. 2001, PASP, 113, 439

Borucki, W. J., & Summers, A. L. 1984, Icarus, 58, 121Brown, T.M., & Charbonneau, D. 2000, in IAU Symp. 202, Planetary Sys-tems in the Universe: Observation, Formation, and Evolution, ed.A. Penny, P. Artymowicz, A.-M. Lagrange, & S. Russell (San Francisco:ASP), in press

Brown, T. M., Charbonneau, D., Gilliland, R. L., Noyes, R. W., &Burrows, A. 2001, ApJ, 552, 699

Burke, C. J., DePoy, D. L., Gaudi, B. S., Marshall, J. L., & Pogge, R. W.2002, in ASP Conf. Ser., Scientific Frontiers in Research on ExtrasolarPlanets, ed. D. Deming & S. Seager (San Francisco: ASP), in press(astro-ph/0208305)

Butler, R. P., Marcy, G. W., Fischer, D. A., Vogt, S., Tinney, C. G., Jones,H. R. A., Penny, A. J., & Apps, K. 2000, in IAU Symp. 202, PlanetarySystems in the Universe: Observation, Formation, and Evolution, ed.A. Penny, P. Artymowicz, A.-M. Lagrange, & S. Russell (San Francisco:ASP), in press

Butler, R. P., Marcy, G. W., Williams, E., McCarthy, C., Dosanjh, P., &Vogt, S. S. 1996, PASP, 108, 500

Charbonneau, D., Brown, T. M., Latham, D. W., &Mayor, M. 2000, ApJ,529, L45

Cody, A.M., & Sasselov, D. D. 2002, ApJ, 569, 451Dame, T.M., Hartmann, D., & Thaddeus, P. 2001, ApJ, 547, 792Gaudi, B. S. 2000, ApJ, 539, L59Giampapa,M. S., Craine, E. R., &Hott, D. A. 1995, Icarus, 118, 199Gilliland, R. L., et al. 2000, ApJ, 545, L47Guillot, T., & Showman, A. P. 2002, A&A, 385, 156Henry, G. W., Marcy, G. W., Butler, R. P., & Vogt, S. S. 2000, ApJ, 529,L41

Holman,M., Touma, J., & Tremaine, S. 1997, Nature, 386, 254Howell, S. B., Everett, M. E., Esquerdo, G., Davis, D. R., Weidenschilling,S., & van Lew, T. 1999, in ASP Conf. Ser. 189, Precision CCD Photom-etry, ed. E. R. Craine, D. L. Crawford, & R. A. Tucker (San Francisco:ASP), 170

Hubbard,W. B., Burrows, A., & Lunine, J. I. 2002, ARA&A, 40, 103Janes, K. 1996, J. Geophys. Res., 101, 14853

Jenkins, J.M., Caldwell, D. A., & Borucki,W. J. 2002, ApJ, 564, 495Lin, D. N. C., Bodenheimer, P., & Richardson, D. C. 1996, Nature, 380,606

Marcy, G.W., Cochran,W. D., &Mayor,M. 2000, in Protostars and Plan-ets IV, ed. V. Mannings, A. P. Boss, & S. S. Russell (Tucson: Univ.Arizona Press), 1285

Mayor,M., &Queloz, D. 1995, Nature, 378, 355Mochejska, B. J., Stanek, K. Z., Sasselov, D. D., & Szentgyorgyi, A. H.2002, AJ, 123, 3460

Monet, D., et al. 1998, The PMM USNO-A2.0 Catalog (Washington:USNO)

Murray, N., Hansen, B., Holman, M., & Tremaine, S. 1998, Science, 279,69

Pepe, F., et al. 2000, Proc. SPIE, 4008, 582Quirrenbach, A., Cooke, J., Mitchell, D., & Safizadeh, N. 2000, in IAUSymp. 202, Planetary Systems in the Universe: Observation, Formation,and Evolution, ed. A. Penny, P. Artymowicz, A.-M. Lagrange, &S. Russell (San Francisco: ASP), in press

Rasio, F. A., & Ford, E. 1996, Science, 274, 954Rosenblatt, F. 1971, Icarus, 14, 71Schlegel, D. J., Finkbeiner, D. P., &Davis,M. 1998, ApJ, 500, 525Seager, S., &Mallen-Ornelas, G. 2003, ApJ, in pressStreet, R. A., et al. 2000, in IAU Symp. 202, Planetary Systems in theUniverse: Observation, Formation, and Evolution, ed. A. Penny,P. Artymowicz, A.-M. Lagrange, & S. Russell (San Francisco: ASP), inpress

———. 2002, in ASP Conf. Ser., Scientific Frontiers in Research on Extra-solar Planets, ed. D. Deming & S. Seager (San Francisco: ASP), in press(astro-ph/0208154)

Struve, O. 1952, Observatory, 72, 199Udalski, A., Kubiak,M., & Szymanski, M. 1997, Acta Astron., 47, 319Udalski, A., Zebrun, K., Szymanski, M., Kubiak, M., Soszynski, I.,Szewczyk, O., Wyrzykowski, n., & Pietrzynski, G. 2002a, Acta Astron.,52, 115

Udalski, A., et al. 2002b, Acta Astron., 52, 1Wozniak, P. R. 2000, Acta Astron., 50, 421Yee, H. K. C. 1988, AJ, 95, 1331———. 1991, PASP, 103, 396

1140 MALLEN-ORNELAS ET AL.

the explore project. i. a deep search for filetors that determine the eﬃciency and the number of...

Documents