CHARACTERIZING EXTRASOLAR PLANETS WITH MULTI-COLOR PHOTOMETRY

Transcription

1 CHARACTERIZING EXTRASOLAR PLANETS WITH MULTI-COLOR PHOTOMETRY By KNICOLE DAWN COLÓN A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2012

2 c 2012 Knicole Dawn Colón 2

3 For my parents, Dave and Debbie. You always told me to reach for the stars! 3

4 ACKNOWLEDGMENTS Many people have supported me in different ways over the years, and this dissertation would not have been possible without that support! I thank my family (Mom, Dad, Dave, Kim and all her little ones, and Teenee - may she rest in peace) for their continued support throughout all my endeavors. You have truly helped make me who I am today, and I love you all so much! I would like to thank my best friends, Kelly Wells and Kristina Pengler, for always believing in me and for always knowing how to say the right thing in any situation. Ashley Bright (Maloney), Erin Crupi, Michelle Blakely (Vihonski) - you all have been great friends to me and kept me sane through our crazy college years. Much love goes out to T8 and South 4! I also want to acknowledge the physics crew (n00bs!) at TCNJ - without you I definitely would not be here today! I also would like to thank all the astronomy graduate students and postdocs at UF for their support over the years, including Justin Crepp, Michelle Edwards, Paola Rodriguez Hidalgo, Dimitri Veras and Dave Vollbach. You all especially helped me survive my first year of graduate school (and I will never forget that I won the Late Night Cookie Time award that year!). I especially would like to acknowledge Dan Capellupo, who has been a truly great friend ever since I started at UF (and I can t believe we are going to be halfway across the world from each other!). I would like to thank my awesome physics professors at TCNJ for their support during college and for encouraging me to pursue astronomy. I am sincerely grateful for the support my advisor, Dr. Eric Ford, has given me through the years and for his seemingly endless knowledge about everything in astronomy. I would also like to acknowledge my many collaborators for all their help and advice, and I want to give special recognition to the staff at the GTC for their help and patience as we learned how to use a new telescope and instrument together. Finally, I want to thank Philip Jackson for always being there for me and for always being able to make me laugh. You provided me with motivation when I needed to get 4

5 work done, but you also reminded me to have fun whenever I became too stressed about work. I am grateful every day for your encouragement and support. I love you! 5

6 TABLE OF CONTENTS page ACKNOWLEDGMENTS... 4 LIST OF TABLES... 9 LIST OF FIGURES ABSTRACT CHAPTER 1 INTRODUCTION CHARACTERIZING TRANSITING EXTRASOLAR PLANETS WITH NARROW-BAND PHOTOMETRY AND GTC/OSIRIS Observations Data Reduction Light-Curve Analysis Results Planetary Parameters Light-Curve Residuals Transit Colour Discussion PROBING POTASSIUM IN THE ATMOSPHERE OF HD 80606B WITH TUNABLE FILTER TRANSIT SPECTROPHOTOMETRY FROM THE GRAN TELESCOPIO CANARIAS Observations In-Transit and Out-of-Transit Observations Data Reduction and Analysis Results Effects of Earth s Atmosphere Limb-Darkening Effects Transit Colour Discussion Interpretation of Light-Curve Shape Comparison to Previous Observations Lack of a K I Line Core Planetary Atmosphere Models Change in Apparent Radius with Wavelength Absorption by an Exosphere Possibility of Other Absorbers Absorption by a Wind Potential Systematics

7 Excluding Telluric Absorption Excluding Instrumental Effects Possible Non-Planetary Astrophysical Effects Conclusion Future Prospects A SEARCH FOR METHANE IN THE ATMOSPHERE OF GJ 1214B VIA NARROW-BAND TRANSMISSION SPECTROPHOTOMETRY Observations July 22 Transit August 28 and 29 Transit and Baseline Observations June 11 Transit Data Reduction Light Curve Analysis Results Results from the 2010 July 22 Transit Results from the 2010 August 28 Transit Results from the 2011 June 11 Transit Results from All Transits Discussion Variability due to GJ 1214b s Atmosphere Variability due to Stellar Activity Variability due to Earth s Atmosphere Conclusions VETTING KEPLER PLANET CANDIDATES WITH MULTICOLOR PHOTOMETRY FROM THE GTC: IDENTIFICATION OF AN ECLIPSING BINARY STAR NEAR KOI Observations Data Reduction and Analysis Results Discussion CONSTRAINING THE FALSE POSITIVE RATE FOR KEPLER PLANET CANDIDATES WITH MULTI-COLOR PHOTOMETRY FROM THE GTC Target Selection Observations KOI KOI KOI KOI Data Reduction Light Curve Analysis Results

8 6.5.1 KOI KOI KOI KOI Discussion Comparison to Theoretical Studies Comparison to Observational Studies Conclusion SUMMARY AND CONCLUSIONS REFERENCES BIOGRAPHICAL SKETCH

9 Table LIST OF TABLES page 2-1 Relative transit photometry of TrES-2 and TrES System parameters of TrES-2 and TrES Absolute transit photometry of HD from January Absolute out-of-transit photometry of HD from April Relative transit photometry of HD from January Normalized transit photometry of HD from January Relative out-of-transit photometry of HD from April Time-averaged flux ratios and noise estimates for HD Time-averaged flux ratios and noise estimates for HD (outlying absolute fluxes excluded) Absolute out-of-transit photometry of HD from January Relative out-of-transit photometry of HD from January Normalized photometry of KOI Normalized photometry of KIC Properties of Kepler targets Photometric precisions achieved from GTC observations of Kepler targets Best-fit model parameters for Kepler targets

10 Figure LIST OF FIGURES page 2-1 Transit light curves for TrES Transit light curves for TrES Standard deviation of time-binned residuals for TrES Absolute photometry of HD and HD from January Absolute photometry of HD and HD from April Transit light curves for HD Uncorrected and corrected light curves for HD from around mid-transit Correlations between transit flux ratios and parameters for HD Uncorrected and corrected light curves for HD from April Comparisons of the corrected light curves for HD from around mid-transit Histograms of normalized flux ratios for HD from around mid-transit Standard deviation of time-binned in-transit residuals for HD Standard deviation of time-binned out-of-transit residuals for HD Observed and model spectra of HD 80606b around the K I line Transit color of HD Standard deviation of time-binned transit color for HD Mean transit colors for HD Baseline observations of GJ 1214 from August Transit light curves for GJ 1214 from July Transit light curves for GJ 1214 from August Transit light curves for GJ 1214 from June Combined transit light curves for GJ Planet-star radius ratios measured from observations of GJ 1214 in July Planet-star radius ratios measured from observations of GJ 1214 in August Planet-star radius ratios measured from observations of GJ 1214 in June

11 4-9 Planet-star radius ratios measured from combined observations of GJ Light curves for KOI Field of view around KOI Light curves for KIC Colors for KOI 565, KIC , and a hypothetical unresolved system Number of planet candidates and eclipsing binaries discovered by Kepler Radius versus orbital period of Kepler planet candidates Signal-to-noise ratio per transit as a function of orbital period of Kepler planet candidates Kepler magnitude versus stellar effective temperature for Kepler planet-hosting stars Field of view around KOI Transit light curves for KOI Transit light curves for KOI Transit light curves for KOI Transit light curves for KOI Cumulative distribution functions of the Galactic latitude of planet-hosting stars and eclipsing binaries discovered by Kepler Cumulative distribution functions of the Galactic latitude of specific subsets of planet-hosting stars and eclipsing binaries discovered by Kepler Number of eclipsing binaries discovered by Kepler as a function of orbital period and Galactic latitude

12 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy CHARACTERIZING EXTRASOLAR PLANETS WITH MULTI-COLOR PHOTOMETRY Chair: Eric B. Ford Major: Astronomy By Knicole Dawn Colón August 2012 Over the past twenty years, nearly 800 planets have been discovered orbiting stars other than the Sun. The discovery of these extrasolar planets (or simply, exoplanets) has led to a renewed interest in planet formation and evolution, as many exoplanets have properties that are nothing like those of the planets found in the Solar System. A subset of exoplanets are known to transit, or pass in front of, their host star, which provides a unique opportunity to measure how their radius changes with wavelength. Such measurements can be used to study the atmospheres of exoplanets, since changes in the measured radius can indicate absorption of stellar photons by the exoplanet atmosphere. Finding a significant change in the radius with wavelength can also indicate that a planet candidate is not a planet at all, but is instead an eclipsing binary star composed of two stars with different temperatures and therefore colors. With over 200 confirmed transiting exoplanets and NASA s Kepler mission s recent discovery of over 2000 transiting exoplanet candidates, detailed investigations into the properties of exoplanetary atmospheres and false positive rates for planet search surveys can now be conducted. To aid these investigations, I developed a novel technique of using the Optical System for Imaging and low Resolution Integrated Spectroscopy (OSIRIS) installed on the 10.4 meter Gran Telescopio Canarias (GTC) to acquire near-simultaneous, multi-color, narrow-band photometry of exoplanet transits. 12

13 I first used this technique to observe the transits of the hot-jupiters TrES-2b and TrES-3b, from which I reached some of the best photometric precisions ( mmag) achieved to date using a ground-based telescope. I subsequently used this technique to measure a 4.2% change in the apparent planetary radius of the giant exoplanet HD 80606b during transit between wavelengths that probe potassium. I hypothesize that the excess absorption is due to potassium in a high-speed wind being driven from the exoplanet s exosphere. This was one of the first detections of potassium in an exoplanet atmosphere. In a similar study, I compared the transit depths for the super-earth GJ 1214b as measured in and out of a predicted methane absorption feature, but I was not able to confirm or refute the presence of methane in GJ 1214b s atmosphere due to the significant impact that stellar variability had on the measurements. Finally, I used the measured color change during transit to identify three short-period Kepler planet candidates as false positives and validate two as planets. These results test recent predictions of the false positive rates for Kepler candidates and suggest that stellar eclipsing binaries significantly contaminate short-period planet candidates. These results demonstrate the capability of the GTC for constraining the properties of transiting planets, which in turn allows us to better understand how different types of planets form and evolve. 13

14 CHAPTER 1 INTRODUCTION In the past decade, the field of extrasolar planet (or exoplanet) research has grown extraordinarily. Ground- and space-based surveys have discovered nearly 800 planets orbiting stars other than the Sun. This includes over 200 planets that transit, or pass in front of, their host star and therefore can have their radius measured. These planets are the focus of this dissertation. While a majority of these have been discovered by wide-field ground-based surveys, space-based missions like COROT and Kepler are quickly catching up (together, they have produced over 80 discoveries of transiting planets to date). Regardless of the discovery method, the tremendous increase in the number of known exoplanets has made it increasingly important to make the best possible use of follow-up resources. As demonstrated by Colón & Ford (2009), large ground-based telescopes are capable of contributing significantly to photometric follow-up efforts even for planets discovered by, e.g., Kepler, which has achieved superior photometric precisions capable of detecting transiting planets with radii smaller than the Earth s (Borucki et al., 2009; Fressin et al., 2012; Gautier et al., 2012; Muirhead et al., 2012). This is because relatively high-precision photometric follow-up can be conducted at near-infrared wavelengths (rather than in a single white broadband filter), where the degeneracy between the transit impact parameter (the minimum distance from the center of the planet to the center of the star when projected on the sky plane; also equivalent to a cos i/r ) and stellar limb darkening parameters can be minimized (due to minimization of the effects of stellar limb darkening at redder wavelengths). As a result, the stronger constraints on the impact parameter improve measurements of the planet-star radius ratio (or, simply the radius ratio, defined as R p /R and which is approximately equal to the square root of the transit depth, F /F, in the absence of stellar limb darkening) and transit duration. In turn, the improved constraints on the radius ratio allow for precise comparisons of the radius ratio as measured at different 14

15 wavelengths. As demonstrated by the studies presented in this dissertation, measuring precise radius ratios as a function of wavelength is extremely important for, for example, atmospheric studies of exoplanets as well as identifying the true nature of planet candidates. In this dissertation, I focus on using a novel technique to achieve high-precision observations of exoplanet transits. Specifically, I use near-simultaneous narrow-band multi-color photometry from the 10.4 m Gran Telescopio Canarias (GTC) to characterize both known and candidate transiting planets. While such a technique is inefficient for planet searches, it provides another avenue towards reaching the high photometric precisions necessary for some studies (e.g., atmospheric studies). In Chapter 2 (Colón et al., 2010), I describe some of the first scientific observations conducted with the GTC, which began scientific operations in March 2009 and is currently the world s largest, ground-based, fully steerable, single-aperture, optical telescope. Specifically, I use the Optical System for Imaging and low Resolution Integrated Spectroscopy (OSIRIS) (Cepa et al., 2000, 2003) installed on the GTC. OSIRIS includes two pixel E2V CCDs, which provide a maximum unvignetted field of view of arcmin 2. There are a suite of filters available with OSIRIS, including broadband order sorter filters that were custom made for OSIRIS and that are narrower than Sloan filters (which have a full-width at half-maximum of about nm, compared to nm for the OSIRIS filters). There is also a tunable filter (TF) imaging mode, which currently allows the user to specify custom bandpasses with a central wavelength of nm and a full-width at half-maximum of nm. 1 Thus, bandpasses can be specifically chosen to avoid water vapor absorption and skyglow so as to minimize 1 A blue end of the TF is expected to be available in the future, which would effectively increase the scientific return from the GTC. In particular, the blue TF would allow for additional atmospheric features to be observed, including the Na I feature that has been previously detected in some exoplanet atmospheres. 15

16 effects from Earth s atmosphere. Also, with the GTC/OSIRIS, it is possible to observe in multiple filters nearly simultaneously, as there is minimal dead time due to switching between filters. Therefore, the GTC/OSIRIS allows for fast, narrow-band, high-precision spectrophotometry of exoplanet transits, which allows us to (1) probe the composition and other properties of the atmospheres of exoplanets and (2) measure the transit color of planet candidates. Furthermore, this technique is extremely efficient, considering that multiple bandpasses can be observed nearly simultaneously within a single transit event. As discussed above, there is now a significant sample of known transiting planets. These discoveries allow for a unique investigation of the atmospheres of exoplanets, as the physical characteristics of an exoplanetary atmosphere can be probed by transmission spectroscopy or spectrophotometry observed against the spectrum of the host star (e.g. with the GTC/OSIRIS). If there is absorption of stellar photons in the exoplanetary atmosphere, this leads to a larger apparent size of the planet at the absorbing wavelengths (Brown, 2001). However, such observations require extremely high precision observations, as the excess absorption is typically expected to cause a decrease in the measured in-transit flux ratio of much less than 0.1%. Early models focused on the atmospheres of hot-jupiters, as these short-period, giant planets were the first transiting planets discovered. The models predicted absorption, particularly from the alkali metals sodium (Na I) and potassium (K I) (e.g., Seager & Sasselov, 2000; Brown, 2001; Hubbard et al., 2001). While the strongest absorber is expected to be Na I (λλ589.6, nm), it is not within the current observing modes on GTC/OSIRIS. However, K I (λλ769.9, nm) is expected to be the second strongest transmission spectrum signature in the optical wavelength range and is readily within the wavelength range that the GTC/OSIRIS can probe. To date, there have been several detections of absorption due to Na I in exoplanetary atmospheres (e.g. Moutou et al., 2001; Charbonneau et al., 2002; Winn et al., 2004; Narita et al., 2005; Redfield et al., 16

17 2008; Sing et al., 2008a; Snellen et al., 2008; Langland-Shula et al., 2009), but these detections have been primarily made in just two exoplanets, HD b and HD b. I discuss one of the first detections of K I in a different exoplanet s atmosphere made using the GTC/OSIRIS TF in Chapter 3 (Colón et al., 2012). 2 While hot-jupiters orbiting bright stars are prime targets for atmospheric studies, the discovery of super-earths orbiting M dwarf stars opened an entirely new area of atmospheric research. One of the more well-studied super-earths is GJ 1214b, which was discovered in a ground-based survey by Charbonneau et al. (2009). Due to the small radius of the host star, the transit depth is nearly 1.5%, which immediately lends itself to making GJ 1214b an excellent candidate for atmospheric studies. Furthermore, Miller-Ricci & Fortney (2010) modeled its atmosphere and predicted that GJ 1214b is required to have a significant atmosphere based on its observed mass and radius. However, the composition of the atmosphere is debated (in particular, whether or not methane is present), so many recent efforts have been made to observe GJ 1214b s atmosphere via transmission spectroscopy (e.g. Bean et al., 2010, 2011; Croll et al., 2011; Crossfield et al., 2011; Désert et al., 2011; Berta et al., 2012; de Mooij et al., 2012). In this dissertation, I take the same approach used in Chapter 3 (Colón et al., 2012), and I use the GTC/OSIRIS TF to perform transmission spectrophotometry of GJ 1214b in order to search for absorption due to methane, the results of which are presented in Chapter 4. Atmospheric studies as described above are critical for understanding how hot-jupiters and super-earths form and evolve. However, it is admittedly difficult to grasp the overall picture of planet formation and evolution when such studies focus on one planet at a time. On the other hand, the Kepler mission is providing unprecedented 2 A second detection of K I was reported by Sing et al. (2011) around the same time as the study presented in Chapter 3 (Colón et al., 2012). 17

18 views into the bulk properties of extrasolar planetary systems due to its discovery of over 2000 planet candidates (Batalha et al., 2012). Of course, a majority of these discoveries are unconfirmed, meaning that they do not have independent confirmation such as an associated mass measurement that confirms their planetary status. A recent theoretical study has predicted that as many as 95% of these candidates are true planets and not false positives (e.g., a blend with a stellar eclipsing binary either in the background/foreground or bound to the target star) (Morton & Johnson, 2011), but this study does not take into account that different subsets of Kepler targets may have different false positive rates associated with them. In particular, planet candidates with short orbital periods (P < 3 d) may be significantly contaminated by eclipsing binary stars, as there is a rapid rise in the number of eclipsing binaries that have been discovered by Kepler at these periods (Prša et al., 2011; Slawson et al., 2011). While preliminary results from an observational study that is currently being conducted with warm-spitzer (Désert et al., 2012) support the findings of Morton & Johnson (2011), in this dissertation I present multi-color observations of five short-period Kepler planet candidates that support the idea that there is a correlation between the false positive rate and orbital period, particularly at the shortest orbital periods. Since planetary transits should be largely achromatic when observed at different wavelengths (excluding the small color changes due to stellar limb darkening), it is possible to use the observed transit color to identify candidates as either false positives or validated planets. Specifically, in Chapter 5 (Colón & Ford, 2011), I present GTC/OSIRIS TF photometry of a super-earth-size Kepler planet candidate which resulted in the identification of an eclipsing binary star as the source of the transit signal. In Chapter 6 (Colón et al., accepted), I present observations of four more candidates, two of which were identified as false positives and two that were validated as planets. Thus, I conclude that those candidates with particularly short periods are significantly contaminated by a population 18

19 of eclipsing binaries, which in turn has implications regarding the properties of the bulk population of known transiting planets and candidates. To summarize, in this dissertation I present a novel technique of using the GTC/OSIRIS to acquire high-precision, narrow-band, multi-color photometry of exoplanet transits (Chapter 2). I then present observations that use this technique to search for absorption due to potassium in the atmosphere of a giant exoplanet (Chapter 3). In Chapter 4, I present similar observations that probe the methane content in the atmosphere of a super-earth-size planet. Next, in Chapters 5 and 6, I present multi-color observations of Kepler planet candidates, with the goal of using the transit color to identify the true nature of the planet candidates. Lastly, in Chapter 7, I summarize the results and conclusions from each of the previous chapters. Additional details regarding all the topics discussed in this chapter can be found in the corresponding chapters in this dissertation. Finally, I note that Chapters 2, 3, 5 and 6 in this dissertation are, or aim to be, self-contained journal articles. Chapters 2 and 3 are published in the Monthly Notices of the Royal Astronomical Society (MNRAS) journal (Colón et al., 2010, 2012). Chapter 5 is published in the Publications of the Astronomical Society of the Pacific (PASP) journal (Colón & Ford, 2011). Chapter 6 has been accepted to MNRAS in collaboration with E. Ford and R. Morehead. 19

20 CHAPTER 2 CHARACTERIZING TRANSITING EXTRASOLAR PLANETS WITH NARROW-BAND PHOTOMETRY AND GTC/OSIRIS Space-based missions like CoRoT and Kepler have the photometric capability to detect transiting planets with radii not much larger than the Earth s, commonly referred to as super-earths (Borucki et al., 2009; Léger et al., 2009). While space-based observatories have provided the highest photometric precisions, ground-based follow-up observations play an essential role in confirming detections of transiting planets and characterizing the planets orbits, interiors, compositions and atmospheres (Torres et al., 2008; Johnson et al., 2009). In particular, because CoRoT s and Kepler s white light observations are affected by stellar limb darkening (LD), there can be strong correlations among the stellar LD parameters, the transit impact parameter and the transit duration (which depends on the eccentricity and pericentre direction). High-precision photometry conducted at far-red, near-infrared (NIR) or mid-infrared wavelengths can break this degeneracy, providing more precise measurements of a planet s orbital and physical parameters (Colón & Ford, 2009). Observing at these wavelengths also allows for reduced differential atmospheric effects, which can further improve the quality of the transit light curve (LC). The detection of transiting Earth-size planets around solar-like stars requires such high photometric precision that astronomers had long assumed that characterizing such planets could only be done from space to avoid the deleterious effects of Earth s atmosphere (Borucki et al., 1985). Narrow-band photometry with large telescopes provides an alternative path towards high photometric precision. While inefficient for planet searches, near-simultaneous narrow-band observations provide additional opportunities for the characterization of known planets (or previously identified planet candidates), such as the measurement of atmospheric absorption. In this paper we describe a novel observational technique for high-precision transit photometry, using near-simultaneous observations in multiple narrow bandpasses 20

21 (Section 2.1). We describe our data reduction in Section 2.2 and LC analysis in Section 2.3. In Section 2.4, we present results for TrES-2 and TrES-3 and demonstrate that this technique can achieve a very high photometric precision. Finally, in Section 2.5, we compare our results to those found in the literature and discuss the implications for studying the atmospheric composition of giant planets and for characterizing the bulk properties of super-earth-size planets, including those in the habitable zone of main-sequence stars. 2.1 Observations The 10.4-m Gran Telescopio Canarias (GTC) 1 is located at the Observatorio del Roque de los Muchachos, on the island of La Palma. While the GTC began scientific operations in 2009 March, commissioning of the telescope and first light instruments is ongoing. We describe some of the first scientific observations with the GTC using the Optical System for Imaging and low Resolution Integrated Spectroscopy (OSIRIS) in the tunable filter (TF) imaging mode (Cepa et al., 2000, 2003). OSIRIS includes two pixel E2V CCDs which provide a maximum unvignetted field of view of arcmin 2. OSIRIS offers a suite of filters, including a TF which allows the user to specify custom bandpasses with a central wavelength of nm and a full width at half-maximum (FWHM) of nm. Note that the TF bandpass is not uniform across the field of view, with the effective wavelength decreasing radially outwards from the centre. Therefore, for the wavelengths used in these observations, a difference of 10 nm exists between the tuned wavelength at the optical centre and the wavelength observed at 4 arcmin from the optical centre (i.e. near one edge of the CCD). We observed one transit each of two giant extrasolar planets, TrES-2b and TrES-3b, using GTC/OSIRIS. The field of view was chosen so that (1) the target and a primary reference star were observed at the same wavelength using the same CCD and (2)

22 several additional secondary reference stars were observed with the same CCD but at different distances from the optical centre (and thus at different wavelengths). Due to the restrictions in CCD orientation, different chips in the OSIRIS CCD mosaic were used for the TrES-2 and TrES-3 observations. During each transit, observations alternated between two bandpasses centred on and nm at the location of the target. These wavebands were chosen to minimize atmospheric effects by minimizing water vapour absorption and skyglow. Each observation was followed by 33.4 s of dead time for readout. We use 1 1 binning but utilize a fast readout mode (500 khz) in order to decrease the dead time between exposures. Recent and future CCD controller upgrades will reduce the dead time between exposures. The telescope was also defocused to increase efficiency and to reduce the impact of pixel-to-pixel sensitivity variations. Even with a slight defocus, the resulting point-spread functions of the stars were fairly well defined (i.e. not doughnut-shaped). The TrES-2 (V 11.4) observations took place under photometric conditions on 2009 June 25 (dark time) from 1:58 to 5:43 UT, during which the airmass ranged from 1.07 to The defocused FWHM of the target varied from 1.6 to 2.8 arcsec ( pixels) during the observations, while the actual seeing was 1.2 arcsec. The autoguider system kept the images aligned within a few pixels, with the target s centroid coordinates shifting by less than 4 pixels in the x-direction and less than 6 pixels in the y-direction. We used 80-s exposures, resulting in an overall cadence of 3.78 min for each bandpass. Four of the images were excluded from our analysis due to tracking problems. The last 16 min of out-of-transit (OOT) data were affected by twilight and are not included in the analysis. The TrES-3 (V 12.4) observations took place under photometric conditions during grey time, starting at 23:53 UT on 2009 August 10 and continuing until 2:54 UT the next morning. During the observations, the airmass ranged from 1.13 to 2.22, the 22

23 defocused FWHM of the target varied from 1.4 to 2.1 arcsec ( pixels) and the target s centroid coordinates shifted by less than 3 pixels in the x-direction and less than 5 pixels in the y-direction. The exposure time was 120 s, yielding a 5.11-min observing cadence for each bandpass. Problems with the telescope caused our observations to begin shortly after ingress had already begun. 2.2 Data Reduction Standard IRAF procedures for bias subtraction and flat-field correction were used for the TrES-2 observations. However, during preliminary analysis, we found that the total number of counts in the dome flat fields (with a given exposure time) was decreasing with time. Thus, for flat fielding, we include only those dome flats taken after the first 30 min of the lamp being turned on, by which time the lamp intensity had stabilized. For TrES-2, we combine 40 (out of 60) dome flats for each observed wavelength (790.2 and nm). After summing over either set of flats, the average number of counts per pixel was For the TrES-3 observations, we perform similar reductions to TrES-2, but we only use 45 of the 95 total dome flats taken for each wavelength. After summing over multiple flats, the average number of counts per pixel was ( ) for (794.4) nm. For the TrES-3 data reduction, we replace the median bias frame with the median of a series of dark frames taken with the same exposure time as the observations (albeit taken several months after the actual observations took place), since early observations revealed a higher than expected dark current. We note that subtracting darks was beneficial only for the TrES-3 data, based on the root mean square (rms) scatter of the OOT LC for TrES-3. The opposite is true for TrES-2; i.e. the rms scatter of the OOT LC was smaller when bias frames were subtracted (instead of dark frames). This difference is likely due to the fact that the dark frames for both targets were taken in 2009 October, rather than in the exact same conditions as the transit observations (i.e. several months after each transit was observed). Therefore, each data set was affected differently by its respective set of dark 23

24 frames, resulting in the darks proving to be useful only for the TrES-3 observations. Recent hardware upgrades have reduced the dark current. Due to the TF s small bandpass and position-dependent wavelength, all observations contain sky (OH) emission rings. Therefore, we performed sky subtraction on all images based on the IRAF package TFred, 2 which estimates the sky background, including rings due to sky emission. We performed aperture photometry on each target and reference star using the standard IDL routine APER. 3 We repeated this analysis using a range of aperture radii and measured the rms scatter of the flux ratio (target over sum of references) outside of transit in each colour. After considering the results in both bandpasses, we adopted an aperture radius of 44 pixels (5.6 arcsec) for stars in the field of TrES-2 and 38 pixels (4.8 arcsec) for stars in the field of TrES-3. Note that the use of TFred removes the need for a sky annulus, so we did not input one in the APER routine. To account for any atmospheric extinction, we fit linear airmass trends to each reference star s LC (computed by dividing the flux for a given star by the sum of the flux from all the other reference stars). For each reference star, we discarded points that resulted in a flux ratio greater than 3σ from the mean (typically one and at most two points per reference star). We computed the reference flux at each time as the weighted sum of the flux of the remaining reference stars. The flux in the target aperture was divided by the reference to compute the final LCs. We found that using an ensemble of six to eight reference stars resulted in a smaller rms OOT scatter than using just the primary reference, despite the fact that the secondary references were observed at different wavelengths and that the reference stars in the field of TrES-2 (TrES-3) ranged over 2 (3) mag in brightness. Since most of the reference stars were fainter than the 2 Written by D. H. Jones for the Taurus Tunable Filter, previously installed on the Anglo-Australian Telescope; reduc.html

25 targets, we conclude that adding additional faint references helps to reduce the photon noise for a larger ensemble and balances stellar variability within different reference stars. The uncertainties in the flux ratio were estimated by computing the quadrature sum of the photon noise for the target and the reference ensemble [median values are (0.339) and (0.249) mmag for TrES-2 and (0.433) and (0.214) mmag for TrES-3 for (794.4) nm], the uncertainty in the sum of sky background and dark current for the target and the reference ensemble [median values are (0.0737) and (0.105) mmag for TrES-2 and (0.0840) and (0.0667) mmag for TrES-3 for (794.4) nm] and the scintillation noise for the target and primary reference [median values are (0.0630) mmag for TrES-2 and (0.0766) mmag for TrES-3 for (794.4) nm]. The uncertainty in the sum of sky background and dark current was estimated by performing aperture photometry on the sky frames produced by TFred at the specific location of each target and reference star. The scintillation noise was estimated using the relation given by Dravins et al. (1998), based on Young (1967). We caution that this is an empirical relation and has not been tested for large telescopes at excellent sites like La Palma. Thus, we consider the expression for scintillation noise to be only a rough estimate. Nevertheless, it demonstrates that scintillation is expected to be only a very small contribution to the total error budget, thanks to the narrow filter bandpass. Readout and digitization noise as well as flat-field noise is negligible compared to photon noise and is not included in the calculation of the uncertainties in the flux ratio, so we include estimates for these noise sources (as based on relations given by Southworth et al., 2009) only for reference. The median readout noise is 7.44 electrons per pixel or (0.0123) mmag for TrES-2 and 7.35 electrons per pixel or (0.0158) mmag for TrES-3 for (794.4) nm. The median flat-field noise is 6.07e-4 (6.07e-4) electrons per pixel or ( ) mmag for TrES-2 and 1.07e-3 (7.63e-4) electrons per pixel or ( ) mmag 25

26 for TrES-3 for (794.4) nm. Based on the relation given by Howell (2006), we find the median total uncertainties in the flux ratio for each exposure to be (0.443) mmag for TrES-2 and (0.499) mmag for TrES-3 for (794.4) nm. We note that our estimated uncertainties are somewhat larger than the measured residual rms given in Section While the estimated and measured precision are marginally consistent, it is possible that we overestimated the uncertainty in individual measurements. We perform the above analysis for both bandpasses and each target. The resulting photometric time series is reported and shown (with some corrections; Section 2.4) in Table 2-1 (full version available online Supporting Information) and Figures 2-1 and Light-Curve Analysis Before fitting models to our LCs, we first applied the external parameter decorrelation (EPD) technique (e.g. Bakos et al., 2007, 2010) to each LC in order to remove any systematic trends that are correlated with the following parameters: the x and y centroid coordinates of the target on the image frames, the sharpness of the target s profile [approximately equal to (2.35/FWHM) 2 ] and the airmass. We then performed the following analysis on the decorrelated LCs. We fit the flux ratio with a standard planet transit model that includes a quadratic LD law (Mandel & Agol, 2002). We parametrize the LC model using the time of mid-transit (t 0 ), impact parameter (b a cos i/r ), transit duration (from first to fourth contact; D), planet-star radius ratio (p R p /R ), average OOT flux ratio, a linear slope for the OOT flux ratio (α) and two LD coefficients (c 1 u 1 + u 2 and c 2 u 1 u 2 ), where u 1 and u 2 are, respectively, the linear and quadratic LD coefficients of Mandel & Agol (2002). We used the publicly available code mpfitfun 4 to perform Levenberg-Marquardt minimization of χ 2 to identify a best-fitting model for the transit photometry. The initial 4 craigm/idl/idl.html 26

27 guesses for b, D and p are based on estimates (and their uncertainties) from Holman et al. (2007) and Sozzetti et al. (2007) for TrES-2b and Gibson et al. (2009) and Sozzetti et al. (2009) for TrES-3b. The initial guesses for t 0 were based on the ephemeris of Rabus et al. (2009) for TrES-2 and Sozzetti et al. (2009) for TrES-3. In each case described below, we repeat the local minimization using an array of initial guesses for the modeled parameters based on the published uncertainties and concluded that the results of the non-linear model fitting were not sensitive to our initial guesses. Similarly, we tested several sets of values for c 1 and c 2 based on the theoretical LD coefficients estimated for each star. Specifically, both LD coefficients were computed for our specific bandpasses for a grid of stellar models using PHOEBE (Prša & Zwitter, 2005). We interpolated in (T eff, log g, [Fe/H]) to estimate c 1 and c 2 for each of several sets of stellar parameters. We adopted stellar parameters from Sozzetti et al. (2007) for TrES-2 and from Sozzetti et al. (2009) for TrES-3, but we note that these abundance measurements are systematically lower than those determined by Ammler-von Eiff et al. (2009). Valenti & Fischer (2005) emphasize that different studies can yield different results due to, e.g., the analysis techniques and/or stellar models used, but both Sozzetti et al. (2007, 2009) and Ammler-von Eiff et al. (2009) use very similar techniques to estimate stellar parameters. Therefore, we chose to interpolate LD coefficients over the range of the stellar parameters plus their measurement uncertainties as given by Sozzetti et al. (2007, 2009), since their analysis used slightly higher resolution spectra than used by Ammler-von Eiff et al. (2009). As an extra precaution, we confirmed that their measurement uncertainties were realistic by comparing them to the adopted uncertainties given by Valenti & Fischer (2005), which were based on the analysis of observations of the solar spectrum as reflected by the asteroid Vesta. Most previous studies use two-parameter LD laws and fit for one or both LD parameters. Southworth (2008) investigated the effects of various LD models and assumptions. He concluded that a linear LD law was often inadequate, so a non-linear 27

28 law should be used, but that two-parameter fits were typically highly degenerate. In the cases of TrES-2 and TrES-3, two-parameter LD models are particularly degenerate, since both have a large impact parameter and the planet only probes the stellar surface brightness near the stellar limb. Southworth (2008) also found that including uncertainty in LD parameters was important to obtain realistic uncertainties for transit model parameters. Therefore, Southworth (2008) recommended holding c 2 fixed and fitting for c 1. Initially, we tried this approach using separate LD parameters for each star and each bandpass and fitted for c 1. We found that the best-fitting LD coefficients for one star and our two bandpasses could differ significantly, even though stellar atmosphere models predict similar LD coefficients in the two nearby bandpasses. Therefore, we adopt an alternative approach. For both TrES-2 and TrES-3, we model the LCs using four different scenarios. First, we fit separate models to the flux ratios at the two wavelengths (Columns 1 and 2 of Table 2-2), fixing both c 1 and c 2 at a pair of self-consistent values based on PHOEBE models using a single set of spectroscopic parameters. The transit and LD parameters in these first two scenarios need not necessarily be self-consistent. Comparing these first two models allows us to evaluate the sensitivity of our results to the two different bandpasses. Secondly, we correct each LC based on the previous results for the fitted slope and mean OOT flux and fit a single model to the LCs for both bandpasses simultaneously (Column 3 of Table 2-2). In this scenario, the transit and planetary parameters (t 0, b, D, p) are forced to be the same for both bandpasses (self-consistent if we neglect the possibilities of wavelength-dependent planet radius and contamination from background light). For this and the subsequent scenario, we fix all four LD parameters (two for each bandpass) at self-consistent values based on PHOEBE models. Finally, we fit a single model to the two (corrected) LCs but allow for separate values of D and p for each LC (Column 4 of Table 2-2). For the final scenario, we continue to force the fit to have the same t 0 and b for both bandpasses since these 28

29 should be the same regardless of the bandpass used for observations, but allow the models in Column 4 to have different values of p and D in the two bandpasses so as to account for a potential difference in the planet radius in the two bandpasses or blending of a putative binary star (Section 2.5). Since we have observations of a single transit in multiple bands and impose additional constraints that all four LD parameters be self-consistent, one would expect our method to result in a larger rms scatter about the model than the fitting procedures used by previous authors and developed for analyzing transit observations at a single waveband. The primary differences in the model parameters among the four scenarios can be traced to different sets of LD parameters. As Southworth (2008) demonstrated, fixing both LD coefficients at their theoretical values could produce measurement uncertainties that are too small. To account for the uncertainty in the LD model, we repeat each of the analyses varying the spectroscopic parameters by the published uncertainties and thus the calculated LD coefficients. We report model parameters for the best-fitting model with the smallest χ 2 value, but present their parameter uncertainties based on the complete set of best-fitting values and uncertainties estimated for all allowed LD parameters (Table 2-2). For example, we estimate the upper error bar for a given parameter as the result of subtracting its best-fitting value (as given by the model with the smallest χ 2 value) from the maximum sum of a fit value and its associated measurement uncertainty, considering the full set of models computed for the different LD coefficients. To further investigate the measurement errors, we applied the Prayer Bead method (e.g. Désert et al., 2009). Specifically, we construct synthetic LCs by calculating the residuals from the initial best-fitting LC model, performing a circular shift and adding the shifted residuals back to the best-fitting LC model. Then, we conducted the above analysis on the synthetic data sets. We use the dispersion of the fit parameters to estimate the effects of any other systematic noise sources not removed by the EPD 29

30 technique. We discuss the results from this analysis as well as the formal 1σ errors (as determined from the covariance matrix) for the best-fitting parameters in the following section for the case presented in Column Results We show the flux ratios and the best-fitting models (after correcting for slopes/trends based on results given in Column 4 of Table 2-2) in Figures 2-1 and 2-2, respectively. Note that the time series given in Table 2-1 was corrected based on results from Column 4 of Table 2-2 as well. The best-fitting LC parameters for TrES-2 and TrES-3 are given in Table 2-2 and described in Section We analyze the residuals and discuss the photometric precision in Section Planetary Parameters For the joint analysis presented in Column 4 of Table 2-2 (which we adopt as our baseline model), the best-fitting parameter values and their formal 1σ uncertainties for TrES-2 are t 0 = ± (HJD), b = ± , D (790.2nm) = ± d, D (794.4nm) = ± d, p (790.2nm) = ± and p (794.4nm) = ± Similarly, for the joint analysis of TrES-3, the best-fitting parameter values and their formal 1σ uncertainties are t 0 = ± (HJD), b = ±0.0073, D (790.2nm) = ± d, D (794.4nm) = ± d, p (790.2nm) = ± and p (794.4nm) = ± We note that for both TrES-2 and TrES-3, these uncertainties are typically smaller than those in Table 2-2, since those also account for uncertainty in the LD model. Additionally, the 1σ errors for the median best-fitting parameters from the Prayer Bead analysis for both TrES-2 and TrES-3 are slightly larger than, but still comparable to, the formal 1σ uncertainties, with the largest deviations occurring for the uncertainty in the radius ratio. We also computed the rms of the best-fitting parameter values from the Prayer Bead analysis as an additional check on our measurement errors. We find that 30

31 accounting for the uncertainty in the LD model sufficiently accounts for the distribution of the best-fitting parameters as derived from the Prayer Bead analysis. For both TrES-2 and TrES-3, the best-fitting values for each parameter are consistent between the different models presented in Table 2-2 but only after accounting for uncertainties in the LD model. The most notable difference in our results with different modeling procedures is for TrES-3 when comparing the impact parameter in Columns 2 and 3 (analysis for the nm LC and the joint analysis with common planet parameters) with the other two models for TrES-3. This is at least partly due to the near degeneracy between impact parameter and LD model. Indeed, the LD coefficients for the models in Columns 2 and 3 are significantly different from the best values for the other two cases. While we were able to place tight constraints on the impact parameter for TrES-2b, the precision for TrES-3b is significantly reduced, since we are modeling an incomplete LC and have no data prior to ingress. The best-fitting slopes and their formal uncertainties as measured in the individual LCs before performing the joint analyses are ± and ± d 1 for the and nm LCs for TrES-2, and ± and ± d 1 for TrES-3. While the estimated slopes for TrES-2 are fairly consistent, the difference in the estimated slopes is much larger for TrES-3 than for TrES-2. Note that the measured slopes account for the effects of both differential atmospheric extinction (due to some reference stars being observed at a different wavelength than the target and primary reference) and real astrophysical variability in the colour of the target and/or reference stars. Our observations are not generally able to disentangle the two potential causes of a slope. If the slope were primarily due to differential extinction, then observations of multiple transits should produce consistent results. On the other hand, if stellar variability is significant, then each transit would need a separate slope parameter. In the case of TrES-3, the large difference in slopes between the two bandpasses suggests that differential extinction is not responsible. However, we caution that the 31

32 measurement of the slope around transit is likely affected by the lack of pre-transit data. Thus, we emphasize the importance of acquiring complete transit LCs as well as extended, uninterrupted OOT baseline data. For TrES-2, the uncertainties in the OOT flux (as determined from the individual LC analyses) were ( ) for the (794.4) nm observations, based on 26 (25) exposures totaling 35 (33) min of integration time. The uncertainties for the OOT flux for TrES-3 are slightly larger, perhaps due to the lack of pre-transit observations and covariance with the slope. However, the TrES-3 observations were more stable overall and allowed for a single baseline of OOT data that was longer than either the pre- or post-transit baselines for the TrES-2 observations. Both observations demonstrate the capability for very high-precision measurements that could enable the detection of super-earth-sized planets around solar-like stars and/or the characterization of the atmospheres of giant planets orbiting bright stars (Sections and 2.5) Light-Curve Residuals We computed the residual flux by subtracting the best-fitting models given by the joint analysis presented in Column 4 of Table 2-2 from the data (Figures 2-1b and c and 2-2b and c). Note that some of the residuals show evidence of additional systematic noise sources that were not removed with the EPD technique. While we do not know the origin of these systematics, we note that the uncertainty estimates from the Prayer Bead analysis are consistent with our best-fitting parameters (as discussed in Section 2.4.1). We estimate the photometric precision to be 343±45 and 412±43 parts per million (ppm) for observations of TrES-2 at and nm, based on the rms deviation of the residuals. Despite high airmass (up to 2.22), the TrES-3 observations produced a precision of 470±64 and 424±57 ppm at and nm, respectively. Because the estimated precisions for both targets are smaller than (or comparable to, based on the upper limits computed from the expected standard deviations) the estimated 32

33 measurement uncertainties (Section 2.2), we take them to be consistent with the theoretical limit. Due to the long, uninterrupted OOT baseline obtained for TrES-3, we take the TrES-3 observations as the best-case scenario to achieve high precisions consistent with the theoretical limit and present the standard deviation of the time-binned residuals for the TrES-3 LCs in Figure 2-3. The residuals for the nm LC are consistent with the trend expected for white Gaussian noise, while the residuals for the nm LC deviate at binning factors larger than 5. When binning the uncorrelated nm residuals over longer observing times ( 40 min), we estimate a precision of 146 ppm, which is sufficient to detect the transit of a super-earth-size planet ( 1.3 R ) orbiting a solar-like star or an Earth-size planet orbiting a star smaller than 0.7 R (assuming 120 min of observations between second and third contact and 3σ confidence level) Transit Colour In Figures 2-1(d) and 2-2(d), we present the colour of the residual fluxes. The colour was computed by averaging each pair of points in the nm LC residuals and dividing by the flux residuals of the nm LC. We find no significant colour deviation in either the TrES-2 or TrES-3 system. We estimate rms precisions of 519 and 502 ppm for TrES-2 and TrES-3, respectively, which are both slightly larger than, but still consistent with, the precisions estimated for the individual LCs. We found no significant difference in the colours in transit and OOT for TrES-2. For TrES-3, there was a slight slope in the colour during transit, but we do not consider this to be significant given the large uncertainty in the slopes fitted to the individual LCs due to the lack of pre-transit observations. 2.5 Discussion The first exoplanet observations from the GTC provided excellent photometric precision, despite problems with the telescope, high airmass and relatively poor atmospheric conditions. 33

34 Our results for TrES-3 are consistent with previous high-precision observations (Gibson et al., 2009; Sozzetti et al., 2009). Our measurements of planet-star radius ratio, impact parameter and transit duration bracket those of previous studies. Our measured transit time has an uncertainty of 9.5 s and occurs 44 s after that predicted by Sozzetti et al. (2009) and 14 s after the updated ephemeris of Gibson et al. (2009). Gibson et al. (2009) conducted a transit timing analysis of TrES-3b, and they find no evidence of transit timing variations. Given the differences between various ephemerides and the possibility for stellar variability to contribute additional timing noise, we do not consider such a difference significant. Our results for TrES-2b are also in good agreement with Holman et al. (2007) and Sozzetti et al. (2007). However, we note that all reported values for the radius ratio could be affected by a binary companion to TrES-2, which has a separation of ± arcsec from the primary star and i 3.7 mag as reported by Daemgen et al. (2009). While the companion star is clearly included within our aperture, our results do not account for the flux of the companion in our analysis. In this case, our model parameter p serves as a depth parameter which can be related to the actual planet-star radius ratio for a given amount of contamination (Daemgen et al., 2009). If we assume that the primary-secondary star flux ratio in our bandpasses is similar to that in i, then the planet-star radius ratio would increase by 1.6 per cent relative to our estimates in Table 2-2. In principle, the different colour of a blended star could result in different transit depths when observed in multiple different bandpasses. While our results are suggestive of such a transit depth difference, the difference is not statistically significant. Given the high precision of our observations and a possible feature in the in-transit data (around mid-transit), we suspect that the apparent differences in p may be due to 34

35 variations in the stellar surface brightness. 5 In light of these potential complications, near-simultaneous photometry in multiple bandpasses could be particularly useful for recognizing when a companion or background object is affecting the photometry. Several recent studies have considered the possibility of variations in the transit times and/or durations of TrES-2b (Mislis & Schmitt, 2009; Rabus et al., 2009; Raetz et al., 2009; Mislis et al., 2010; Scuderi et al., 2010). Our best-fitting transit time has a measurement uncertainty of 11 s and is offset from the predicted ephemeris of Holman et al. (2007) by 4 min, of Rabus et al. (2009) by 142 s, of Raetz et al. (2009) by 16 s and of Scuderi et al. (2010) by 64 s. Given the differences in various ephemerides, observing bandpasses and LD models, we do not consider the transit time offset to be significant evidence for transit timing variations. This is consistent with the recent transit timing analysis of Raetz et al. (2009). While Rabus et al. (2009) raised the possibility of sinusoidal variations in the transit ephemeris due to an exomoon, they do not find results of the magnitude proposed by Mislis & Schmitt (2009). Due to TrES-2b s large impact parameter, the transit duration is very sensitive to changes induced by additional planets or other bodies in the system (Miralda-Escudé, 2002). Mislis & Schmitt (2009) and Mislis et al. (2010) claim that the transit duration decreased by 3 min between 2006 and 2008, and they argue that a third body is a natural explanation for this change. If we include uncertainty due to unknown LD models, we cannot definitively rule out the shorter durations suggested by Mislis & Schmitt (2009) and Mislis et al. (2010). When interpreting putative differences in transit duration/impact parameter, one should be mindful of differences in the various filters and the stellar LD model used (Scuderi et al., 2010). Indeed, Scuderi et al. (2010) report 5 Indeed, previous studies have found other indications of possible stellar activity, including variations in the OOT flux (O Donovan et al., 2006b; Raetz et al., 2009). Our observations show no signs of a putative second dip (Raetz et al., 2009) but only extend for 1 h after the end of transit. 35

36 no significant change in the orbital inclination/transit duration when they compare their observations with those of O Donovan et al. (2006b) (taken in the same filter). Since our observations use unique bandpasses, it is not (yet) possible to compare them to previous observations with the same bandpass. Since TrES-2 is in the Kepler field, Kepler observations should soon shed light on this matter, as the observations will cover longer time-scales and therefore multiple transits all within the same passband. In summary, we find that large ground-based observatories are capable of achieving high-precision differential photometry. Systematic effects due to variable atmospheric extinction can be minimized by the use of a small far-red or NIR bandpass chosen to avoid sky absorption lines. Thus, large, ground-based observatories can help characterize super-earth-size planets discovered by ongoing transit searches, including planets in the habitable zone of main-sequence stars for which transits last several hours. We anticipate that this innovative technique for high-precision photometry will enhance the ability of current and future large optical/nir observatories to study the properties of both Earth-like planets (e.g. size, density and orbit) and giant planets. For example, the high photometric precision could allow for the detection of the occultation of hot transiting giant planets like TrES-3b at optical wavelengths. While the occultation depth probes the temperature and dynamics of the planet s atmosphere, the time and duration of occultation provide a powerful probe of the orbit s eccentricity. The technique is even more powerful when observations in multiple bandpasses can be obtained nearly simultaneously and thus during the same transit. Such observations could help verify that the transit of a candidate planet is not due to stellar variability plus rotation by observing in multiple wavelengths that reduce the effects of stellar variability and/or enhance the contrast of spots or plage. Similarly, nearly simultaneous observations in multiple narrow bandpasses could be used to characterize the atmospheres of giant planets. For example, the Hubble Space Telescope and the Hobby-Eberly Telescope (HET) have used transmission spectroscopy of 36

37 HD b and HD b to measure Na I absorption in the planetary atmospheres (Charbonneau et al., 2002; Redfield et al., 2008). Future GTC observations could be used to perform similar measurements of atmospheric features such as Na I and K I absorption in planets transiting bright stars, especially once planned improvements in the OSIRIS CCD controller software reduce the dead time between exposures. Considering the increasing numbers of known transiting exoplanets, we hope that this novel technique will be an additional useful tool to help characterize a variety of planetary properties. 37

38 Table 2-1. Relative photometry of TrES-2 and TrES-3. This is a sample of the full table, which is available with the online version of the paper (Supporting Information). Observed λ (nm) HJD Relative Flux Uncertainty TrES TrES Note. The time stamps included here are for the times at mid-exposure, and the relative flux has been corrected for slopes/trends in the LCs (Section 2.4). 38

39 Table 2-2. System parameters of TrES-2 and TrES-3. Parameter Value a b λ 1 λ 2 (λ 1 + λ 2 ) c (λ 1 + λ 2 ) d TrES-2 t (HJD) b D (d) λ 1 : λ 2 : p λ 1 : λ 2 : c 1 (fixed) λ 1 : λ 1 : λ 2 : λ 2 : c 2 (fixed) λ 1 : λ 1 : λ 2 : λ 2 : TrES-3 t (HJD) b D (d) λ 1 : λ 2 : p λ 1 : λ 2 : c 1 (fixed) λ 1 : λ 1 : λ 2 : λ 2 : c 2 (fixed) λ 1 : λ 1 : λ 2 : λ 2 : a We label the bandpass centred on nm as λ 1 for simplicity. b We label the bandpass centred on nm as λ 2 for simplicity. c The best-fitting parameters were determined from the joint analysis of the two LCs for each target, with only the LD coefficients determined separately for the two wavelengths. See the text for additional details. d Same as in Column 3, but also with D and p determined separately for each wavelength.

40 Figure 2-1. Normalized LCs (a), residuals (b), (c), and colour (d) for nearly simultaneous observations at and ± 2.0 nm of TrES-2 as observed on UT 2009 June 25. The filled circles are observations and the lines in panel (a) show the best-fitting models. In panel (a), the nm LC has been arbitrarily offset by Panels (b) and (c) show residuals from the fits for the and nm LC. The colour of the residuals is shown in panel (d). 40

41 Figure 2-2. Same as in Figure 2-1, but for TrES-3 as observed on UT 2009 August 10. In panel (a), the nm LC has been arbitrarily offset by

42 Figure 2-3. Standard deviation of time-binned residuals for TrES-3 as a function of the number of data points per bin (n). The blue square (red triangle) symbols are the standard deviations of the (794.4) nm binned LC residuals. The solid curves show the trend expected for white Gaussian noise (n 1/2 ). Each exposure was separated by 5.11 min; thus, the present observations only allow us to test time-scales of up to 40 min due to the limited duration of the observations. 42

43 CHAPTER 3 PROBING POTASSIUM IN THE ATMOSPHERE OF HD 80606B WITH TUNABLE FILTER TRANSIT SPECTROPHOTOMETRY FROM THE GRAN TELESCOPIO CANARIAS Discoveries of extrasolar planets which transit their host star provide valuable opportunities to measure the physical properties of exoplanetary atmospheres. The physical characteristics of an exoplanetary atmosphere can be probed by transmission spectroscopy observed against the spectrum of the host star. Seager & Sasselov (2000), Brown (2001) and Hubbard et al. (2001) developed models that predicted such absorption, particularly from Na I, K I and other alkali metals. Subsequent refinements of such models have confirmed that in the optical wavelength regime the strongest lines are expected from the Na I resonance lines (λλ 589.6, nm) and the K I resonance lines (λλ 769.9, nm) (e.g. Barman, 2007; Fortney et al., 2010). 1 In the optical, the cores of the atomic features of Na I and K I are relatively narrow. For this reason, medium to high resolution spectrographs can be used to compare the in-transit stellar spectrum to the out-of-transit (OOT) stellar spectrum. The absorption of stellar photons in the exoplanetary atmosphere leads to excess absorption in the in-transit stellar spectrum when compared to the OOT spectrum. In photometric observations, this leads then to deeper transits and a larger apparent size of the planet at the absorbing wavelengths (Brown, 2001), with variations of order the atmospheric scale height (Fortney, 2005). Such measurements in strong optical transitions can also constrain the atmospheric metallicity, rainout of condensates, distribution of absorbed stellar flux and photoionization of atmospheric constituents. 1 We caution that these lines are most prominent for hot Jupiter like planets with a certain range of atmospheric temperatures. Atmosphere models generated for HD 80606b at the time of transit [based on Fortney et al. (2010)] do not predict a significant K I absorption feature, due to the low equilibrium temperature of 500 K. We refer the reader to Section for further discussion. 43

44 The first detection of absorption due to an exoplanetary atmosphere came from Na I observations of HD b using the Space Telescope Imaging Spectrograph (STIS) onboard the Hubble Space Telescope (HST) (Charbonneau et al., 2002). Unfortunately, the subsequent failure of the STIS instrument prevented similar observations for more than 5 years. Thus, attention was directed towards making such observations from the ground (e.g. Moutou et al., 2001; Winn et al., 2004; Narita et al., 2005). The second detection of absorption due to an exoplanetary atmosphere, this time from the ground, was also made of Na I in observations of HD b using the 9.2-m Hobby-Eberly Telescope (HET) (Redfield et al., 2008). Further detections of Na I in the atmosphere of HD b were made using archival data from the 8.2-m Subaru Telescope (Snellen et al., 2008), from HST by Sing et al. (2008a) and from Keck by Langland-Shula et al. (2009). The recent repair of STIS and installation of the Cosmic Origins Spectrograph (COS) onboard HST has enabled new optical and ultraviolet transmission spectrum observations of exoplanetary atmospheres, extended exospheres and auroral emission (e.g. Linsky et al., 2010; Fossati et al., 2010; France et al., 2010). Comparing the surprisingly weak Na I absorption in HD b (Charbonneau et al., 2002; Knutson et al., 2007) to the three times stronger Na I absorption of HD b (Redfield et al., 2008) suggests that the two planets have different atmospheric structures. Theorists have suggested numerous mechanisms such as adjustments to the metallicity, rainout of condensates, distribution of absorbed stellar flux or photoionization of sodium (Fortney et al., 2003; Barman, 2007). In particular, Barman et al. (2002) suggested that non-local thermodynamic equilibrium Na level populations were the cause of the weak Na feature observed in HD b, and a reanalysis of the Knutson et al. (2007) data by Sing et al. (2008a,b) suggested that Na condensation or Na photoionization in HD b s atmosphere was the best explanation for matching the data, given the Na line shapes they derived. It is clear that comparisons of the atmospheric properties of different transiting planets will be critical to understanding the 44

45 atmospheric properties of exoplanets as a whole. Although still small, the list of detected atoms and molecules is growing. In addition to Na I, several molecules have been detected, primarily in the infrared, with both space-based and ground-based platforms, including CO, CO 2,H 2 O and CH 4 (Swain et al., 2008; Swain et al., 2009; Snellen et al., 2010). Other HST observations using the Advanced Camera for Surveys (ACS) did not detect K I in HD b (Pont et al., 2008). If detections of constituents in the extended exosphere are included, then H I, CII, OI, Mg II and other metals have also been detected (Vidal-Madjar et al., 2003, 2004; Fossati et al., 2010; Linsky et al., 2010). Each new detection provides not only compositional information, but also another window into the physical properties of the exoplanetary atmosphere (e.g. condensation, wind speed and photoionization). Even though atmosphere models do not predict a significant K I feature in HD 80606b, it remains of great interest to observationally determine the level of K I absorption in its atmosphere, since K I is generally predicted to be the second strongest transmission spectrum signature in the optical wavelength range. Further, Na I and K I probe different layers of the atmosphere. Measurements of K I can test the hypothesis that the low abundance of Na I on HD b may be due to a high-altitude layer of clouds or haze. Finding low abundance for both Na I and K I would be consistent with either the cloud hypothesis or with the photoionization hypothesis, as both are very easy to ionize. Finding that only Na I is significantly depleted would point to alternative models with complex atmospheric chemistry (e.g. incorporation into grains, odd temperature structure, unexpected mixing patterns). Finally, in principle, future observations could probe temporal variability of Na I and K I due to high-speed, high-altitude winds and/or differences in the leading and trailing limb (Fortney et al., 2010). All of the above atmospheric studies were based on observations using high-resolution spectrographs. Here, we describe a new technique that utilizes fast, narrow-band spectrophotometry with the Optical System for Imaging and low Resolution Integrated 45

46 Spectroscopy (OSIRIS) installed on the 10.4-m Gran Telescopio Canarias (GTC) to probe the composition and other properties of the atmospheres of exoplanets that transit bright stars (Section 3.1). Fast line spectrophotometry can be much more efficient (e.g. 34% with GTC/OSIRIS) than typical high-resolution spectrographs ( 1-2 per cent) thanks to the use of a tunable filter (TF) rather than diffraction gratings. Further, this technique has the potential to be less sensitive to several systematic noise sources, such as seeing variations that cause line variations in wide spectrograph slits (specifically in non-fiber fed spectrographs), atmospheric variations (since reference stars will be observed simultaneously) and/or flat-fielding errors (since on- and off-line data are obtained at the same detector location). Thus, spectrophotometry with a TF technique is particularly well suited for observing a narrow spectral range of atomic absorption features, without suffering from the inefficiencies or potential systematic uncertainties of high-resolution spectrographs. Here we present results of such observations of the 2010 January transit of HD 80606b using the GTC and the OSIRIS TF imager. HD 80606b was originally discovered by radial velocity observations (Naef et al., 2001) and was remarkable due to its very high eccentricity (e = 0.93). Only several years later did Spitzer and ground-based observations reveal that the planet passes both behind and in front of its host star (Laughlin et al., 2009; Fossey et al., 2009; Garcia-Melendo & McCullough, 2009; Moutou et al., 2009). Spectroscopic observations revealed that the angular momentum axis of the stellar rotation and that of the orbital plane are misaligned (Moutou et al., 2009; Pont et al., 2009; Winn et al., 2009). Given the infrequent transits and long transit duration ( 12 h), follow-up observations are quite challenging. Winn et al. (2009), Hidas et al. (2010) and Shporer et al. (2010) were able to characterize transits of HD 80606b with longitudinally distributed networks of ground-based observatories, and Hébrard et al. (2010) observed the 2010 January transit using the Spitzer spacecraft. 46

47 The Spitzer observations constrain the thermal properties of the planet s atmosphere (Laughlin et al., 2009; Hébrard et al., 2010). To the best of our knowledge, the observations presented here are the first to attempt to detect atmospheric absorption by HD 80606b. While existing atmosphere models predict that HD 80606b would not have any significant K I feature due to its high surface gravity and cold atmosphere at the time of transit (e.g. Section 3.3.4), our observations test this prediction. Even though models do not predict a K I feature, exoplanet observations have a track record of unexpected discoveries. Furthermore, in principle, depending on the atoms/molecules found in the atmosphere, these observations could yield information about how the planet cools, independent of any observations of the thermal phase curve of this system. In principle, transmission spectroscopy also provides a way to characterize transiting planets in eccentric orbits, which either do not pass behind their host star or which are too cool to detect via occultation when they do pass behind the star. Finally, we note that HD is one of the best systems for making very precise spectrophotometric measurements. HD is the brightest of the transiting planet host stars which have a comparably bright reference star very nearby ( 20 arcsec). Also, the long duration between the second and third points of contact ( 6 h) of HD 80606b provides time to collect a large amount of in-transit data in a single transit. Thus, we expect that all else (e.g. observing conditions) being equal, HD 80606b permits the most precise spectrophotometric measurements of any known system (at least with observations of a single transit). This paper presents extremely precise measurements of the variation in HD 80606b s apparent radius with wavelength near the K I feature, which in turn can help us test the predictions of atmosphere models. Section 3.1 describes our observations and data analysis procedures. We describe the results of our observations in Section 3.2. In Section 3.3 we interpret the results, and we summarize our conclusions and discuss the future prospects for the method in Sections 3.4 and

48 3.1 Observations HD and its nearby companion (HD 80607) are both bright G5 dwarves of a similar magnitude (V 9) and colour. On three nights, we measured the flux of both HD (target) and HD (reference) simultaneously. We cycled through a set of four wavelengths throughout the observations. On the night of 2010 January 13-14, the planet was in transit for the duration of our observations, and we measure an in-transit flux ratio of HD to HD for each wavelength. We repeated the observations on 2010 January 15 and 2010 April 4, when the planet was not transiting HD 80606, allowing us to measure the OOT flux ratio of HD to HD for each wavelength. Our results (Section 3.2) are based on the ratio of in-transit flux ratio (target over reference) to OOT flux ratio (target over reference). Any changes in the Earth s atmosphere from one night to the next should affect both the target and reference star similarly. By making differential measurements of the colour during the same transit and at similar atmospheric conditions, this method allows for extremely precise measurements of the transit depth at different wavelengths. While night-to-night variability in the atmospheric conditions or either of the stars could cause a systematic scaling of the transit depth measurements, the relative wavelength dependence of the apparent planet radius is largely insensitive to either of these potential systematics. We refer the reader to Sections and for further discussion In-Transit and Out-of-Transit Observations We observed a partial transit of HD 80606b on 2010 January and acquired baseline data on 2010 January 15 and 2010 April 4 to establish the OOT flux ratios. For our observations, we used the TF imaging mode of the OSIRIS instrument installed on the 10.4-m GTC, which is located at the Observatorio del Roque de los Muchachos on the island of La Palma (Cepa et al., 2000, 2003). In the TF mode, the user can specify custom bandpasses with a central wavelength of nm and a full width at half-maximum (FWHM) of nm. The effective wavelength decreases radially 48

49 outward from the optical centre; because of this effect, we positioned the target and its reference star at the same distance from the optical centre and on the same CCD chip. The observed wavelengths described below refer to the location of the target (and reference) on the CCD chip. During the transit observations and baseline observations on 2010 January 15, exposures of the target and its reference star cycled through four different wavelengths (all with a FWHM of 1.2 nm): one on the predicted core of the K I line ( nm); one to the blue side ( nm) and two redwards of the K I feature ( and nm). As the tunings for the TF are set by the order sorter (OS) filter used, our bluest wavelength is the bluest wavelength we could observe at in the wing of the K I line and still observe within the same OS filter as the on-line wavelength (i.e. at the location of the core of the K I line). We then chose two wavelengths redwards of the K I line in order to sample more of the structure/wings around the K I line. The reddest bandpass was chosen since we expect to see (for a typical hot Jupiter) a maximum difference between the flux ratio in the on-line bandpass and around that reddest bandpass. In order to maximize the signal-to-noise ratios in the on-line wavelength and in the reddest off-line wavelength, in each sequence we observed on-line three times, at the reddest off-line wavelength two times and at the other off-line wavelengths one time each. During the transit, the observing sequence from the GTC was as follows: , , , , , and nm (repeat). We emphasize that these wavelengths were chosen to be around the location of the K I feature in HD 80606b s atmosphere. In order to observe on the K I feature (which has a rest wavelength of nm) in the frame of the planet, we accounted for the Doppler shifts due to the Earth s motion around the Sun, the system s radial velocity and the planet s non-zero radial velocity during transit [ 59.6 km s 1 based on velocities from Winn et al. (2009)]. After accounting for these effects, the observed wavelengths in the frame of the planet are redshifted by 0.16 nm to nm (on-line) and , 49

50 and nm (off-line). The observed wavelengths in the frame of the star are essentially the same as observed on Earth due to the small systemic velocity of the HD planetary system and the Earth s small barycentric velocity on the night of the transit. For the remainder of the paper, we report the wavelengths as observed in the frame of the planet when discussing results from the transit observations. A similar sequence as described above was used for the baseline observations taken on 2010 April 4, but the observed wavelengths were corrected for the Doppler shift due to the planet s orbital velocity on that specific date ( 23.9 km s 1 ) in order to match the wavelengths observed during the transit. Thus, the wavelengths observed on 2010 April 4 (from the GTC) are nm (on-line) and , and nm (off-line). Transit observations of HD 80606b began at 22:28 UT on 2010 January 13 (during ingress) and ended at 7:15 UT on 2010 January 14 (around the beginning of egress and including astronomical twilight), during which the airmass ranged from 1.08 to The observing conditions were photometric, with a clear sky and a dark moon. No data were taken between 5:20 and 5:50 UT on 2010 January 14 due to recalibration of the TF during that time. The actual seeing varied between 0.7 and 0.9 arcsec during the transit observations, but we used a slight defocus to increase efficiency and reduce the impact of pixel-to-pixel sensitivity variations. Therefore, the defocused FWHM of the target varied from 0.9 to 2.3 arcsec (7-18 pixels) during the transit. For the portion of the light curve used in our analysis (Section 3.1.2), the FWHM was much more stable than is indicated by the range given above, with a typical value between 10 and 14 pixels and a mean value of 12 pixels. Even with an autoguiding system, the target s centroid coordinates shifted by 9-10 pixels over the course of the night. We used 1 1 binning and a fast readout mode (500 khz) to readout a single window of pixels (located on one CCD chip) in order to reduce the dead time between exposures. This window is equivalent to a field of view of arcsec 2, so the only stars in our field 50

51 were HD and a single reference star, HD Each individual observation was followed by an average dead time of less than 4 s for readout and to switch between TF tunings. We used 10 s exposures, resulting in an overall cadence of about 14 s for each observation. Due to the short exposure time used, the sky background level was low enough that we did not need to discard any images taken during astronomical twilight. Baseline observations were taken from 5:50 to 7:10 UT (i.e. also through the beginning of astronomical twilight) on 2010 January 15, but the data were highly scattered, so we do not include it in our primary analysis. 2 Additional baseline observations took place on 2010 April 4 from 21:30 (including the end of astronomical twilight) to 0:00 UT. The observing conditions were photometric and taken during grey time, using the same set-up as the in-transit observations described above. During the observations, the airmass ranged from 1.08 to 1.20, and the actual seeing varied between 1.4 and 1.6 arcsec ( pixels), so the telescope was not intentionally defocused. The target s centroid coordinates shifted by 5-8 pixels during the observations. The exposure time was changed from the initial exposure time of 10 s to 8 s and then again to 11 s to counteract variations in the seeing as well as increasing airmass while avoiding saturation and maintaining a high number of counts. In our analysis, we discard the 10-s data because a majority of the images were saturated. We tested using the OOT flux ratios from the 8 and 11 s data individually in our analysis and found that they produced very similar results. Thus, we combine the 8 and 11 s data to establish the final OOT flux ratios (Section 3.1.2) and to achieve the longest usable baseline possible Data Reduction and Analysis Observations taken with OSIRIS prior to 2010 mid-march suffered from a higher than expected level of dark current despite the short exposure times used. Therefore, 2 Section

52 we used standard IRAF procedures for bias and dark subtraction and flat-field correction for the 2010 January transit observations of HD We note that the flat-fields for these observations did not produce the pattern of having the total number of counts in the dome flat-fields decreasing with time as seen by Colón et al. (2010), so we use almost all (65 out of 75) dome flats for each observed wavelength in our analysis (the 10 dome flats not included in the analysis were overexposed). A new dewar fixed the problems with the dark current before the 2010 April observations took place, so for the baseline data we performed standard bias subtraction and flat-field correction (combining all 133 flats taken for each observed wavelength) and did not need to subtract dark frames. Because of the very small readout window used for our observations, our images do not contain the sky (OH) emission rings that occur due to the TF s small bandpass and position-dependent wavelength. Therefore, we performed simple aperture photometry on the target and reference star using the standard IDL routine APER 3 for a range of aperture radii. We measured the rms scatter of the flux ratio (equal to the target flux divided by the reference flux) at the bottom of the transit (for the 2010 January data) and for the individual 8 and 11 s data taken OOT (in 2010 April) in each bandpass. We considered the results for each bandpass and adopted an aperture radius of 28 pixels (3.6 arcsec) for the in-transit data and 32 pixels (4.1 arcsec) for the OOT data, as these were the aperture radii that typically yielded the lowest rms scatter. The radii of the sky annulus used for the reduction of both data sets were pixels in order to completely avoid any flux from the target or reference star. We have included the results of our aperture photometry in Tables 3-1 and 3-2 and illustrate the results in Figures 3-1 and 3-2. As illustrated, the flux in each bandpass displayed large variations during 3 Landsman 1993; 52

53 parts of the observations (particularly during parts of the transit), and we take this into consideration in our analysis (Section 3.2.1). We present the raw in-transit light curves in Figure 3-3, which were computed by dividing the flux in the target aperture by the flux in the reference star aperture and then normalizing by the weighted mean OOT flux ratio for each bandpass (Section 3.2 for details on the computation of the mean flux ratios). In an attempt to reduce systematic trends seen in our transit light curves, we applied the external parameter decorrelation (EPD) technique (e.g. Bakos et al., 2007, 2010) to each transit and baseline light curve. Note that for the transit light curve, we only applied EPD to the 4 h centered around mid-transit, or 3:36 UT on 2010 January 14, as estimated by Shporer et al. (2010). 4 Specifically, we decorrelated each individual light curve against the following parameters: the centroid coordinates of both the target and reference, the sharpness of the target and reference profiles [equivalent to (2.35/FWHM) 2 ] and the airmass. As illustrated in Figure 3-4, EPD removed most of the correlations in the in-transit data. For reference, we show the correlations between the in-transit data and the target s FWHM and centroid coordinates both before and after EPD has been applied in Figure 3-5. For the baseline data, we performed EPD for the 8- and 11-s data series separately, but we then combined the two data sets to compute the weighted mean flux ratio and its uncertainty for each bandpass as described in Section 3.2. The results of the decorrelation for the OOT data are illustrated in Figure 3-6. As a result of applying EPD, the rms scatter in each of the bandpasses improved by as much as 25 per cent, but decorrelating the light curves against the above parameters did not completely remove the systematics that are seen in our data. In a further attempt to remove systematics, we also tried a quadratic decorrelation against the sharpness of 4 This ephemeris is in between that given by Winn et al. (2009) and Hébrard et al. (2010). The choice of ephemeris used does not significantly affect our results. 53

54 the target and reference profiles, as that was the only parameter that showed a possible residual systematic pattern after EPD was applied. However, the quadratic decorrelation did not reduce systematics in our light curves any further. We discuss other potential sources of systematics in detail in Section Because our goal is to compare the depths of the transit in each bandpass, the rest of our analysis focuses on the data from the bottom of the transit as presented in Figure 3-3 and highlighted in Figure 3-7 i.e. the 4 h centered around mid-transit. Note that the light curves shown in Figure 3-7 have been corrected using EPD. We also discarded points that had a flux ratio greater than 3σ from the mean of the bottom of the transit light curve. This resulted in discarding four points from the reddest light curve ( nm). We also discarded several exposures from each wavelength that were unusable due to saturation. The different panels in Figure 3-7 illustrate the deviation between the magnitude of the on-line flux ratios and each of the off-line flux ratios, which will be discussed in detail in Sections 3.2 and 3.3. We estimated the uncertainties in the flux ratios by computing the quadrature sum of the photon noise for HD and HD 80607, the uncertainty in the sum of the sky background (and dark current, for the in-transit observations) and the scintillation noise for the two stars. We assume Poisson statistics to compute the uncertainty in the sky background, and the noise due to scintillation was estimated from the relation given by Dravins et al. (1998), based on Young (1967). We caution that this empirical relation might overestimate scintillation for large telescopes located at excellent sites such as La Palma. Regardless, the relation demonstrates that scintillation is still a small contribution to the total error budget for these observations. The flat-field noise is also negligible compared to the photon noise, so we do not include it in our determination of the measurement uncertainties. Based on the relation given by Howell (2006), which computes the standard deviation of a single measurement in magnitudes and includes a correction term between the error in flux units and the error in magnitudes, we find 54

55 the median total uncertainties in the flux ratio for each exposure to be 0.538, 0.532, and mmag at , , and nm (over the bottom of the transit), respectively. The rms of the transit light curve is comparable, but slightly larger, with values of 0.585, 0.667, and mmag for those wavelengths. The median total uncertainties for the OOT observations are calculated in a similar way, but the uncertainties for the 8- and 11-s data sets were scaled by the flux ratios for each respective set in order to compute a weighted uncertainty. Thus, the median total (weighted) uncertainties in the flux ratio are 0.657, 0.650, and mmag, while the estimated rms is quite comparable, with values of 0.562, 0.605, and mmag for , , and nm, respectively. The complete photometric time series for each bandpass of the in-transit data (uncorrected and unnormalized) is reported in Table 3-3, while the photometric time series (both before and after EPD was applied) for the transit bottom and the April observations are reported in Tables 3-4 and 3-5. The weighted mean flux ratios for both the in-transit and OOT data (Section 3.2) are given in Table 3-6, along with their uncertainties. 3.2 Results As illustrated in Figure 3-7, we can see by eye a hint of a deviation between the in-transit flux ratios observed at the on-line wavelength and the red off-line wavelengths, but no clear deviation is seen when compared to the bluest off-line wavelength. Despite evidence of time-correlated systematics in our data, we emphasize that the error bars shown in Figure 3-7 are binned error bars, which illustrate that our measurement uncertainties are larger than any residual systematics present in the light curves and that the deviations in the flux ratios between the different bandpasses are real. We refer the reader to our discussion of possible systematic sources in Section In Figure 3-8, we plot histograms of the (unbinned) flux ratios at the bottom of each of the transit light curves, where the flux ratios have been normalized against the mean 55

56 OOT flux ratio for each respective wavelength. These histograms further illustrate that the flux ratios for the on-line and bluest off-line light curves are comparable, but the red off-line flux ratios (particularly for the reddest light curve) clearly lie at slightly higher values compared to the on-line flux ratios, indicating a smaller apparent planetary radius at those wavelengths. Ideally, when one has access to either a complete or partial transit light curve and baseline data acquired immediately before or after the transit event, one can fit a model to the data and estimate the transit depth from the model results. Due to the very long duration of HD 80606b s transit, we were not able to acquire baseline data on the night of the transit, thereby making this type of analysis impractical. However, thanks to several recent campaigns to observe a complete transit of HD 80606b and establish accurate orbital and physical parameters for this system via light-curve modeling (Winn et al., 2009; Hébrard et al., 2010; Hidas et al., 2010; Shporer et al., 2010), we do not need to fit a model to our partial light curve to achieve the goals of this paper. Instead, we consider only the middle 4 h of the transit light curve in our analysis (compared to the full duration of the bottom of the transit, which is 6 h), thereby minimizing systematic effects of stellar limb darkening (LD) as the strongest LD occurs during ingress, egress and right after/before ingress/egress. Further, since we do not know the LD model for this star to the precision of our observations, adding such a model would not be useful for this study. Thus, we assume that LD is the same over all our bandpasses and that the transit ephemeris, impact parameter and transit duration do not vary with wavelength. The only parameter of which we assume changes with wavelength is the apparent planet radius (R p ). To investigate how the apparent planet radius changes with wavelength, we simply compute the weighted mean in-transit flux ratio [< δf /F >, which is proportional to the planet-to-star radius ratio, (R p /R ) 2 ], and its uncertainty for each wavelength. 56

57 Specifically, we compute the weighted mean as < δf /F >= n w i F i i=1 (3 1) n w i i=1 where the weights, w i, are equal to 1/(βσ i ) 2. Here, σ i is the estimated photometric uncertainty weighted by some wavelength-specific factor (β) in order to account for the presence of any red noise in each individual bandpass. To illustrate the effect of red noise on our measurements and the need for a re-weighting factor, the standard deviations (σ N ) of the in-transit and OOT time-binned flux ratios are shown in Figures 3-9 and 3-10 as a function of binning factor (N) for each bandpass. The theoretical trend expected for white Gaussian noise ( N 1/2 ) is plotted as a solid curve, and we can see that for the in-transit data the rms deviates from the theoretical curve at large binning factors, indicating that red noise is present in most bandpasses (being the least significant in the bluest bandpass). However, for the OOT data, our photometry appears to be generally consistent with the photon limit (although the bluest light curve suffers from small number statistics). Following methods used by e.g. Pont et al. (2006) and Winn et al. (2007), we calculated explicit estimates for both the white (σ w ) and red (σ r ) noise in each bandpass by solving the following system of equations: σ 2 1 = σ 2 w + σ 2 r (3 2) σ 2 N = σ2 w N + σ2 r. (3 3) The re-weighting factor, β, is then computed as σ r /(σ w / N). Based on our fits to the red and white noise, we computed a re-weighting factor for each bandpass and applied it as stated above. We imposed a minimum value for β of 1, particularly for cases where red noise was negligible. 57

58 The uncertainties for the OOT flux ratio are also weighted by the flux ratio, F i, since two different exposure times were used during the OOT observations. Finally, the uncertainty on the weighted mean is computed as 1 σ <δf /F > =. (3 4) n w i We include the uncertainty on the weighted mean OOT flux ratio in our calculation of the mean normalized in-transit flux ratio and its uncertainty. The resulting spectrum of HD 80606b (the normalized weighted mean in-transit flux ratios as a function of wavelength) is shown in Figure 3-11, and it clearly illustrates a difference between the flux ratios for the bluest bandpasses and those for the reddest bandpasses. While we find no significant difference between the flux ratios measured at and nm, we measure differences of 3.02 ± and 8.09 ± between observations at and and nm. We list the weighted mean in-transit flux ratios (normalized by the weighted mean OOT flux ratios) as well as the weighted mean OOT flux ratios and their uncertainties in Table 3-6. In this table, we also include our fits to the white and red noise, as well as our estimates for β. When calculating the normalized in-transit flux ratio and its uncertainty, we also included the re-weighted uncertainty for the mean OOT flux ratio in our calculation. The error bars for the flux ratios given in Table 3-6 and shown in Figure 3-11 also take red noise into account Effects of Earth s Atmosphere We consider the effect of random atmospheric variations (e.g. clouds) during the night of the transit as well as during the April baseline observations. As mentioned in Section 3.1.2, large variations in the absolute flux of both the target and reference were observed towards the beginning and the end of the transit observations, with a few large fluctuations around the middle of the observations as well. Thus, to check if our i=1 58

59 measured in-transit flux ratios were affected by these fluctuations, we computed the weighted mean in-transit flux ratio for each bandpass after excluding outlying absolute flux measurements from our analysis. We specifically excluded any points that were greater than 3σ away from the mean of the flattest part of the spectrum measured for each bandpass and each star. After excluding outlying points from both the in-transit and April baseline data, we found that the new spectrum for HD 80606b shows a very similar shape as the original spectrum, albeit with the flux ratio in the reddest bandpass differing the most from the original spectrum. However, we still measure a significant difference between the flux ratios in the on-line and reddest bandpasses. These results are included in Table 3-7 and shown in Figure 3-11 as the solid circles Limb-Darkening Effects So far our analysis has assumed that LD is the same between our different bandpasses, so LD should not affect the mean flux ratios for each bandpass differently. However, in principle, there is also the possibility that LD coefficients vary significantly in and out of narrow spectral lines. To investigate the possibility that our spectrum s signature is a result of our probing in and out of HD s stellar spectral lines, we have computed quadratic LD coefficients for each of our bandpasses for a grid of stellar models [using PHOEBE; Prša & Zwitter (2005)]. We then generated theoretical limb-darkened light curves for each bandpass using the standard planet transit model of Mandel & Agol (2002). We used stellar parameters and uncertainties for HD as given by Winn et al. (2009) to estimate a range of LD coefficients to use in our models. We also input planetary parameters and uncertainties for HD 80606b as given by Hébrard et al. (2010). After computing light-curve models for different combinations of LD coefficients and planetary parameters, we computed the mean model flux ratio over the bottom of each transit light curve (the 4 h centered around mid-transit). We include the resulting model spectrum in Figure 3-11 as solid squares. This particular spectrum was computed based on using a median set of LD coefficients, but all the model results 59

60 were similar over the range of LD coefficients used. The median linear and quadratic LD coefficients (u 1, u 2 ) are (0.392,0.229), (0.388,0.233), (0.391,0.230) and (0.376,0236) for the , , and nm bandpasses. While small differences in LD exist between the different bandpasses, the mean model flux ratios differed by only a very small amount (< ) between the different bandpasses. From this, we conclude that LD is most likely not the cause of the large variations in our observed spectrum. However, we note that PHOEBE (as well as other LD codes) has not been calibrated in and out of narrow spectral lines. We also note that the models show that the bottom of the light curve is in fact not flat due to LD. However, based on our calculation of the mean model flux ratio over the limb-darkened transit bottom for each bandpass, this should not affect the magnitude of the variations we measure in our observed spectrum. Due to LD effects, the overall normalization of the spectrum may be affected Transit Colour In Figure 3-12 we present the colour of the normalized in-transit flux ratios, computed by dividing each point in the off-line bandpasses by the average of each pair of on-line points around those off-line points. We find that the colour between the bluest bandpass and the on-line bandpass is consistent with zero, with a mean value of 6.30± (computed following the method described in Section 3.2). The mean colour of the nm and on-line bandpasses is 3.57± , and the mean colour between the reddest and on-line bandpasses is 8.99± We also present the standard deviation of each colour for a number of binning factors in Figure We find that the trend for each colour is consistent with having only white noise in each of our colours. This is also confirmed by fitting the white and red noise explicitly for each colour. Considering that the red noise is estimated to be less than for each colour, white noise clearly dominates the uncertainties in the transit colour. 60

61 As explored in Section 3.2.1, we also compute mean colours after excluding outlying absolute flux measurements from our analysis. After excluding those data points, we estimate the mean colours between each off-line and the on-line bandpasses to be 1.79± , 3.54± , and 6.92± (from bluest to reddest). Both these mean colours and those discussed above are plotted in Figure The colours are comparable between the two data sets, with the colour of the reddest bandpass having the only measurable difference between the two sets. Furthermore, Figure 3-14 illustrates that not only is there a significant change in the colour during transit, but also that the magnitude of the change is equivalent to a large change in the apparent planet radius. At the reddest wavelengths, we clearly measure a change of over 3 per cent (and as much as 4.2 percent, based on the flux ratios that do not exclude outlying absolute flux measurements) in the apparent radius of the planet compared to the planet s apparent radius in the on-line bandpass. Overall, these colours agree with the magnitude and direction of the differences measured between the weighted mean in-transit flux ratios for the different bandpasses (Section 3.2). Furthermore, the differences between the colour of the bluest to on-line bandpasses and the reddest to on-line bandpasses has greater than 5σ significance. Since our colour measurements match the magnitude and direction of the differences in the flux ratios as measured from our spectrum, we conclude that our measured spectrum of HD 80606b s atmosphere is real and that the differences in the flux ratio are significant. 3.3 Discussion Interpretation of Light-Curve Shape First, we compare our light curve (integrated over all bandpasses) to simultaneous observations from Spitzer (Hébrard et al., 2010) and other ground-based observatories (Shporer et al., 2010). In particular, Hébrard et al. (2010) identified a bump in the in-transit light curve that occurred within the hour before their estimated time of 61

62 mid-transit and pondered whether it could be due to an exomoon or spot crossing. Under the exomoon hypothesis, the magnitude of the bump should be wavelength-independent. If the bump were due to a spot, then one would expect an even greater feature in the optical. We do not find any evidence for a coincident bump [regardless of whether we adopt the ephemeris of Hébrard et al. (2010) or Shporer et al. (2010)]. Thus, the bump is unlikely to be due to either an exomoon or star-spot. If anything, we find possible evidence of a bump occurring after mid-transit, but this feature was not detected by Hébrard et al. (2010). If we assume our candidate bump is not a result of instrumental systematics, and we compare our candidate bump as observed in the different wavebands, we find that the size of the putative bump is smallest in the bluest bandpass, providing further evidence against a star-spot. Furthermore, since the magnitude of the bump varies slightly for each bandpass, this provides additional evidence against the existence of an exomoon. Future high-precision, multi-wavelength observations could help provide additional constraints on the light-curve shape Comparison to Previous Observations Next, we note that our measured in-transit flux ratios differ slightly from the flux ratio given by Hébrard et al. (2010). This is at least partly due to the different bandpasses used. There could also be a systematic uncertainty in the overall normalization of our transit depths. If our goal had been to measure the transit depth precisely, we would have required observations taken just before and after the transit event. In this case, ground-based observations spanning the full transit were not feasible due to the extremely long transit duration. Thus, we normalized our in-transit light curves by OOT observations taken on a different night. While our observations resulted in a very high precision for differential measurements of the transit depth in each bandpass, a change in the observing conditions between nights could result in the transit depths all being affected by a common scaling factor. 62

63 To confirm that the change in the apparent planetary radius with wavelength is based on a robust estimate of the OOT flux ratio despite using baseline observations separated by four months from the transit observations, we estimated the weighted mean in-transit flux ratios as before, but normalized them against the lower quality OOT data taken on 2010 January 15. For reference, we include the results of the aperture photometry for this data set in Table 3-8 and the flux ratios before and after EPD in Table 3-9. We found that despite the large scatter in that OOT data, the normalized in-transit flux ratio (and therefore, the apparent radius of the planet) still changes significantly with wavelength and maintains the same shape as shown in Figure We conclude that the large change in transit depth from the and nm bandpasses to the and nm bandpasses is a robust result. We also emphasize that the atmosphere was much more stable during the April observations than the January baseline observations, so we still rely on the April baseline observations for our primary analysis. We also tested whether our results were sensitive to the aperture radius used for photometry. We tried a variety of annuli for the apertures for both the in-transit and OOT data sets. In all cases, we see the trend of increasing flux ratio with wavelength and found very similar results to those presented here. The only difference occurred for the largest apertures, in which case the weighted mean in-transit flux ratio on the K I feature is slightly smaller than the flux ratio at the bluest wavelength. Even for this choice of apertures, the fluxes in the and nm bandpasses are not significantly different (though the error bars are slightly larger) Lack of a K I Line Core As illustrated in Figure 3-11, there is no significant difference between the observations acquired in the core of the K I line and slightly to the blue. Given the 1.2-nm FWHM, a Doppler shift of 200 km s 1 would be needed to shift the line core out of the on-line bandpass. This is greater than the escape speed from HD 80606b 63

64 ( 121 km s 1 ). Thus, we place a 3σ limit on the strength of the K I line core of (for our 1.2-nm FWHM bandpass). By itself, the lack of a line core is most naturally explained by a lack of K I at the altitudes probed by transmission spectrophotometry. This could occur if (1) there is a significant bulk underabundance of potassium, (2) the potassium has condensed into clouds and/or molecules, (3) there is a cloud or haze layer above the region capable of causing significant potassium absorption, and/or (4) the potassium has been photoionized (Fortney et al., 2003). In the previous case of HD b, theoretical investigations of the unexpectedly weak Na I absorption showed that the observed feature depth is particularly sensitive to the extent of cloud formation (Fortney et al., 2003). In the case of HD 80606b, the highly eccentric orbit results in flash heating near pericenter and extreme temperature variations over the orbital period. At the time of transit, the star-planet separation is 0.3 AU, so the equilibrium temperature is 500 K. Based on Spitzer observations, cooling is sufficiently rapid that the planet is expected to have cooled between pericenter and transit (Laughlin et al., 2009). Thus, both sodium and potassium are predicted to have condensed into clouds. Thus, we conclude that the lack of a K I core could easily be due to potassium having condensed into clouds before the time of transit Planetary Atmosphere Models In an attempt to model our observations, we considered both a conventional 1D cold atmosphere model (Fortney et al., 2010) (solid line in Figure 3-11) and a similar model, but with arbitrary additional heating to raise the effective temperature by 500 K (dotted line in Figure 3-11). Both models have been normalized to the stellar radius estimated by Hébrard et al. (2010), and assume a star-planet separation of 0.3 AU (i.e. the distance between the star and planet when the planet transits). Chemical equilibrium and a standard pressure-temperature profile for HD 80606b are assumed. In the cold atmosphere model, the planet s (apparent) radius at 10 bar was adjusted to 64

65 match the radius measured by Hébrard et al. (2010) at 4.5 µm. In the hot atmosphere model, the temperatures in the upper atmosphere range from 300 to 500 K, even with the additional heating. The higher temperature increases the observed planetary radius at all wavelengths, and slightly increases the peak to trough distance of the features, but the planet s radius was not adjusted to match the radius from Hébrard et al. (2010). At these temperatures, most of the potassium is expected to have formed condensates, significantly reducing the K I absorption feature. As the inset in Figure 3-11 illustrates, neither the hot atmosphere model nor the cold atmosphere model predicts a significant feature due to K I absorption Change in Apparent Radius with Wavelength While we do not detect the K I core, we find relatively large differences (3.57 ± and 8.99 ± ) between the colours of the on-line bandpass and the bandpasses to the red ( and nm). Clouds and hazes would suppress both the core and wings of the absorption feature. A similar observation for a typical hot Jupiter could be readily interpreted as strong absorption in the wings of the potassium line due to absorption by pressure broadened potassium at lower altitudes, while potassium at higher altitudes has been photoionized (Fortney et al., 2003). However, in our observations, the magnitude of the difference in absorption at the two blue and two red wavelengths appears too large for such a model. One could expect such observations to probe the lower atmosphere over 10 scale heights (H), from a pressure of 100 mbar to 1 microbar. Assuming the planet has reached a thermal equilibrium for the star-planet distance at the time of transit and a 500K upper atmosphere temperature, the scale height would be H 20 km. Thus, one might expect to see changes in the apparent radius of the planet on the order of 200 km. Our observations suggest a much larger change in the apparent radii (up to 4.2 per cent or 2900 km) when comparing observations in the K I line core and the reddest bandpass. The scenario described above would suggest that these observations probed

66 scale heights in the atmosphere of HD 80606b, or pressures of less than bar, which is well into the exosphere. Such a large number of scale heights is not realistic, implying that the absorption is originating from a part of the atmosphere much hotter than 500 K. Fortunately, the temperature is expected to rise rapidly to thousands of Kelvin above one planetary radius (Yelle, 2004) Absorption by an Exosphere Based on the model of Section 3.3.4, we would estimate that our observations have probed 145 scale heights in the atmosphere of HD 80606b, or a pressure of less than bar. However, these estimates assume an atmospheric temperature of 500 K. Yelle (2004) finds a steep rise in the temperature from 350 to K from 1 R p to 1.1 R p for a planet at 0.1 AU from the Sun. If we use their model as a rough guideline, and if we assume a temperature of 2000 K between 1 and 1.04 R p for HD 80606b, the 2900 km measured change in the apparent radius would imply that the observations probed 36 scale heights, or to a pressure of less than bar. Regardless of whether we assume a temperature of 500 or 2000 K, the implied pressures are indicative of those that would exist in an exosphere. The models and opacity database of Section are not complete for the temperature and pressures of the exosphere. The opacity database used extends to temperatures of 2600 K and 1 microbar and is not intended to describe opacity sources in an exosphere or wind (e.g. Vidal-Madjar et al., 2003, 2008; Ballester et al., 2007; Ehrenreich et al., 2008; Lecavelier Des Etangs et al., 2008, 2010; Ben-Jaffel & Sona Hosseini, 2010). To the best of our knowledge, an exospheric model that predicts the location and strength of absorption features arising from the exosphere does not exist. We hope that our observations will stimulate theoretical models for the observable effects of exoplanet exospheres on transmission spectroscopy and spectrophotometry. Given the planet s high surface gravity and any reasonable choice of planetary parameters, a 4.2 per cent change in the planet s apparent radius requires a very 66

67 dramatic change in the pressure at which the slant optical depth reaches unity, between 770 and 777 nm. Thus, we conclude that absorption at high altitude and temperature is the most likely explanation for the large change in the apparent planet radius Possibility of Other Absorbers Next, we consider whether another absorber might be responsible for the observed change in apparent planet radius. Methane can be active in this region of the spectrum. However, methane would be unstable at the high temperatures of an exosphere or wind. Both of the models in Section include methane at all temperatures at which it would be stable, around <1000 K. In the wavelength regime that we observed, the opacity of methane is largest at 778 nm and smallest at 769 nm, so its presence would produce the opposite trend from what is shown in the data. The observed wavelengths were also chosen to avoid water vapor (which is also unstable at high temperatures). We are not aware of any other absorber which could explain the large change in apparent planet radius, and consider K I the most likely absorber. Nevertheless, we cannot rule out the possibility that HD 80606b s exosphere possesses an absorber that is something other than K I on account of the incomplete opacity database Absorption by a Wind If the 4.2 per cent change in the apparent radius is due to absorption by K I at high altitude, then it is not obvious why the observations on the K I core ( nm) are not significantly different from the observations slightly to the blue ( nm). One possibility is that the line core was shifted out of the on-line bandpass. Given the 1.2-nm FWHM, this would require a Doppler shift of 200 km s 1. A blueshift of 225 km would place the core halfway between the and nm bandpasses. A somewhat smaller Doppler shift plus Doppler broadening might also reduce the signal strength. In any case, the velocities required would be greater than the escape speed from HD 80606b ( 121 km s 1 ). 67

68 While a velocity exceeding the escape speed is somewhat concerning, it is not out of the question for a wind being driven from the exosphere. In fact, similar observations of other planets also appear to find an unexpectedly large Doppler shift. Specifically, a large blueshift has been found in all cases; e.g. Redfield et al. (2008) found an unexpected blueshift of the core of the Na I absorption for HD b. Snellen et al. (2010) detected a 2 km s 1 blueshift in the upper atmosphere of HD b with observations of CO. Additionally, Holmström et al. (2008) reported Lyman-α absorption around HD b at wavelength offsets corresponding to velocities of several 100 km s 1, but there was no information about the Lyman-α core as it is not observable due to Earth s geo-corona. Much like our observations of HD 80606b, there is considerable uncertainty regarding the origin of the absorption and Doppler shift for HD b (Lecavelier Des Etangs et al., 2008, 2010; Ben-Jaffel & Sona Hosseini, 2010). Proposed mechanisms include radiation pressure and interaction with a stellar wind (e.g. Tian et al., 2005; García Muñoz, 2007; Murray-Clay et al., 2009; Ekenbäck et al., 2010), and in particular we note that models of HD b s atmosphere match observations better if it is assumed that the line core is obscured. Our observations could be explained if a similar mechanism operates on HD 80606b and heavy elements (i.e. potassium) are mixed into the wind. In the case of HD 80606b, the dynamics of the exosphere and any planetary wind is almost certainly quite complex. The planet has the largest semimajor axis of any confirmed transiting planet (0.455 AU), but it follows such a highly eccentric orbit (e = 0.93) that the star-planet separation of HD 80606b at periastron is 2/3 that of HD b. Thus, HD 80606b experiences strong and rapid heating of the atmosphere near pericenter. The large and rapid changes in the incident stellar flux and temperature as well as the stellar wind flux could lead to episodic mass-loss following each pericenter passage (Laughlin et al., 2009). Based on the observed X-ray flux (Kashyap et al., 2008) and mass-loss correlation (Wood et al., 2005), HD could have a mass-loss rate 68

69 as much as 100 times stronger than HD , providing a much stronger stellar wind to drive a wind from HD 80606b. The rapid contraction and expansion of the Roche lobe around each pericenter could further complicate the dynamics of the exosphere and planetary wind Potential Systematics Excluding Telluric Absorption The usual suspect in ground-based observations is variability in the telluric absorption. At our observed wavelengths there is very little absorption. The only two species that contribute any appreciable absorption are water and oxygen. In particular, there is a lack of absorption from carbon dioxide or methane in our observed bandpasses. Oxygen is generally well mixed in the atmosphere. Thus, we expect any variability due to oxygen has been removed in our data reduction procedure, which normalizes each observation of HD by the flux of HD taken at the same time and using the same bandpass. Thus, we rule oxygen absorption out as a potential systematic. Since water can be very anisotropically distributed in the atmosphere, one could worry that the 20-arcsec separation between HD and HD might allow for variations in the water absorption that are not removed by calibration. However, the two bandpasses to the red of K I were specifically chosen to be at wavelengths that avoid water absorption. Thus, even in the scenario that the on-line and blue bandpasses were contaminated by water absorption, we still measure a 2.7 per cent change in the apparent planet radius between the two reddest bandpasses (both of which should be substantially free of telluric absorption). From this, we conclude that our primary result of measuring a large change in the apparent radius with wavelength is not the result of variable telluric water absorption. However, in an effort to confidently rule variable telluric absorption out as a potential source for systematics, we construct an alternative model for the spectrum based on 69

70 changing the level of water vapor absorption. Specifically, we integrated our bandpasses over high-resolution model transmission spectra for telluric water vapor and oxygen. Using the TERRASPEC code (Bender et al., in preparation), we computed model transmission spectra for two different airmasses (representing the mean airmass over the transit bottom as observed in January and the mean airmass during the baseline data observed in April) and three different water vapor levels (1, 5 and 10 mm). We then integrated our bandpasses over each spectra and computed the relative transmission for the different bandpasses for every possible combination of water vapor towards HD and HD We integrated over the appropriate bandpasses for each set of observations, as the bandpasses used in April were centered at slightly different wavelengths than for the January observations. Our goal was to determine if the transmission spectrum would have a similar signature as our observed spectrum if there was a difference in the water vapor towards HD and HD during either or both of the January and April observations. For example, we took the integrated transmission for a water vapor level of 10 mm (towards HD 80606) divided by the integrated transmission for a water vapor level of 1 mm (towards HD 80607) based on the mean airmass in January. Then, we divided that result by a similar ratio based on the spectra for the April observations. We computed this ratio for all combinations of water vapor and compared the results. Realistically, the water vapor was most likely below 6 mm for both the January and April observations [based on García-Lorenzo et al. (2010)], but we approach this issue with much caution and therefore discuss the results for the 10-mm water vapor level as well. From our results, we can make several arguments against variable water vapor absorption and/or the different wavelengths observed in January and April being the cause of our spectrum s signature. First, since our measurements are multiply differential (comparing the target to the reference in-transit to the target to the reference OOT), we minimize any such effects from our measurements. Secondly, even if the 70

71 water vapor column towards HD and HD differed by an average of 10 mm on one of the nights, it would result in a difference of only per cent (if the January water vapor differed by 10 mm) or per cent (if the April water vapor differed by 10 mm) between the reddest and on-line bandpasses. This difference in transmission based on the wavelengths observed in April is less than half of the actual measured difference between the flux ratios in these two bandpasses. Further, the separation between HD and HD is only 20 arcsec, so it is extremely unlikely that the time-averaged water column towards the two stars would differ by 10 mm. Further, an untenably large water column, inconsistent with the observational conditions, would be required to explain the observed difference of between the on-line and reddest bandpasses. Thirdly, even if the average water column towards the two stars did differ by that much on one of the nights, the resulting colours differ from what we observe, i.e. the hypothesis that our measurements are primarily due to atmospheric variability would predict the two bluest bandpasses to be comparable in some cases and differ largely in others, while the two reddest bandpasses are comparable in all cases. While we do observe the flux ratios in the two bluest bandpasses to be comparable, we see a significant difference between the flux ratios for the two reddest bandpasses. Further, the magnitude of the differences between the bluest and reddest bandpasses is observed to be much larger than what the difference would be if they were caused by variable atmospheric absorption. Thus, we estimate that for the four different bandpasses, the effect of variable atmospheric absorption would be less than (0.0024, , , per cent) [ < mm of H 2 O towards HD > - < mm of H 2 O towards HD > ]/[10 mm of H 2 O] at the wavelengths and airmass observed at in January, or less than (0.004, 0.027, , per cent) multiplied by the same ratio given above at the wavelengths and airmass observed at in April. However, assuming that the difference in transmission is neglible between the target and reference for both the January and April observations, 71

72 then the fact that the stars were observed at different wavelengths and airmasses on those nights should be irrelevant. Finally, we note that spectrophotometry using a narrow-band TF is much less prone to systematics than spectroscopic observations. The lack of a slit, the simultaneous use of a very good reference star, rapid switching between bandpasses and the multiply differential nature of our measurement should all minimize the effects of telluric absorption. While the OH lines are variable, the sky subtraction in our data reduction process removes the emission to a high degree of precision. Finally, we see no evidence, in our atmospheric transmission models, of absorbers that could account for the signal detected. To first order, the effects of atmospheric extinction are corrected by measuring flux ratios relative to HD We expect negligible second-order differential extinction, since the target and reference stars are of the same spectral type and separated by only 20 arcsec. Since the magnitude of this effect scales as the square of the filter bandpass, our use of such a narrow bandpass further minimizes second-order differential extinction, allowing this technique to be applied to other targets with reference stars that differ in temperature. We do not consider differential extinction to be a viable explanation for the effect seen in Figure Nevertheless, we performed an additional check, in which we do not perform relative photometry between the target and reference. We compare the ratio of the absolute flux of the reference star in the reddest bandpass and the bandpass centered on K I as measured on the night of the transit to the same ratio as measured on the night the OOT observations were taken (2010 April 4). We estimate a ratio of ± , equivalent to a colour deviation of 1.5 per cent between the two nights. This provides an upper bound on the effects of atmospheric variability, including differential extinction. The accuracy of our primary analysis should be considerably higher thanks to the use of relative photometry to correct for atmospheric variability. 72

73 Excluding Instrumental Effects With TF imaging, the photons for each observed wavelength land on the same pixels, eliminating concerns about spatial variations in the flat-fielding. However, the normalization of the flux measurements is affected by the wavelength dependence of the pixel sensitivity. To minimize this effect, we took dome flat-fields for each observed wavelength and corrected the science frames taken at each wavelength with their respective flat-fields. Furthermore, we guard against possible non-uniformity in the shutter motion, which could result in the systematic effect of producing slightly different exposure times for the target and reference star, depending on where they are located on the CCD chip. This systematic effect is more noticeable for shorter exposure times, so we guard against it by following an observing sequence that repeats after seven exposures, so that the subsequent set of exposures occurs with the shutter motion in the opposite direction. Depending on the orientation of the TF, the observed wavelength can drift due to the rotation of the instrument during the observations. For our observations, the TF was tuned before observations, in the middle of the transit and at the end of the observations. No drifts larger than 0.1 nm occurred. Finally, we conducted a thorough investigation into the possibility of saturation and/or non-linearity as a source of systematic effects. A majority of the peak counts during our observations were well below the saturation threshold ( ADUs), and for standard observing modes linearity is guaranteed up to ADUs, so non-linearity should not be an issue. However, as we use a non-standard observing mode on OSIRIS in order to read out the CCDs at the highest rate possible (and thereby greatly reduce dead time), it is worthwhile to investigate whether non-linearity is an issue. Therefore, we discuss here several checks for non-linearity, where we arbitrarily assume that counts ( ADUs, based on the gain of 1.46 e per ADU) is the level at which non-linearity might begin. 73

74 First, we checked if the average number of counts from the flat-fields taken for each bandpass had a linear dependence with exposure time, as we had taken flat-fields at several different exposure times. We fit a line to all measurements of the mean flat-field counts (for five different exposure times), and then we fit another line to the data but excluded measurements that were near or above counts. To see if including measurements at higher counts resulted in non-linearity, we compared the slopes and y-intercepts of the two best-fitting lines. After comparing the best-fitting solutions between the different bandpasses and for the different series of flats taken in January and April, we find that it is not obvious that any one set of flats displays significant non-linearity compared to the others. Secondly, we investigated fitting a quadratic function to the flat-field counts for both the January and April flat-fields. After comparing the best-fitting coefficients for the different bandpasses, we found that at least one set of coefficients deviated significantly from the coefficients for the other bandpasses. While there might be an obvious outlier in terms of one bandpass that might be affected by non-linearity for each set of flats, the supposed outlier is different for the January and April flats. We again conclude that it is not obvious which, if any, of our bandpasses is displaying significant non-linearity. Thirdly, we investigated the possibility of non-linearity by computing the colour (between each off-line bandpass and the on-line bandpass) and seeing how it varied with respect to the average on-line flux per pixel (estimated by dividing the total absolute on-line target flux by the target s FWHM squared). We computed the median colour for exposures where the average flux per pixel was below and for exposures where the average counts were above We then estimated the difference in the median colour for those two sets of exposures. We found that in the near-red bandpass ( nm), non-linearity most likely does not play a role, as the median colours below and above the count level differ by an insignificant amount. However, in the bluest and reddest colours, we do see a slight correlation, with the median colours 74

75 differing by comparable amounts. This is not what we might expect to see if non-linearity were causing systematic effects in our observed spectrum, since we do not observe a comparable colour difference in the in-transit flux ratio between the blue-on-line and reddest-on-line colours. We also computed the colour deviations for the April baseline data, and we found that the colours below and above the count mark are slightly larger than the in-transit colour deviations (but these were computed using a combined data set for two different exposure times, which could affect these estimates). Regardless, we still find that the smallest difference in the colours is in the near-red bandpass, and the differences are comparable for the bluest and reddest bandpasses, even though in both the observed in-transit and OOT flux ratios we see the smallest difference in the flux ratio between the bluest bandpasses and the largest between the on-line and reddest bandpass. In summary, we conducted several checks for non-linearity. We conclude that any effects of non-linearity are either too insignificant to affect our photometry or they are not correlated with the data Possible Non-Planetary Astrophysical Effects Lastly, we consider potential astrophysical systematics such as stellar variability. Observations in all four bandpasses were obtained during the same transit. If observations using different bandpasses had been made during different transits, then the interpretation would be ambiguous, as stellar variability (e.g. spots that the planet does not necessarily pass over) could result in apparent changes in the in-transit flux ratio. For the large change in apparent radius to be due to stellar variability, there would need to be a 4.2 per cent change in the colour of either the target or reference star. Such large variability over a small range of wavelengths is a priori unlikely for solar-like stars (Hébrard et al., 2010). However, as suggested by the referee, we have estimated how spotted HD 75

76 80606 would have to be to cause a difference of in the flux ratios in the on-line and reddest bandpass. We computed the blackbody flux for HD (T eff 5572 K), then integrated the flux over each bandpass to estimate the total flux observed in each bandpass. We then completed similar calculations for a spot assuming a temperature 1000 K cooler than HD and a spot radius equal to the planet s radius. After computing the ratio of the integrated spot flux to the integrated star flux for some N spots, we found that about 26 spots with the above properties would have to exist on the surface of HD during the transit observations in order to produce a difference in the on-line and reddest flux ratios of about That is equivalent to having 26 per cent of HD s surface covered with spots. Even if the systematic trends we see in our transit light curves are due to spots coming in and out of view on the surface of the star, the per cent of the stellar surface covered by spots is unlikely to be as much as 26 per cent. Furthermore, if the star was this spotted, we should also see a difference in the flux ratio between the two bluest bandpasses of over , yet the difference we observe is less than We conclude that it is possible for spots to account for some of the variations we measure, but that HD is very unlikely to be spotted enough to cause the magnitude of variations we measure. In fact, Wright et al. (2004) measured values of S HK = and log R HK = 5.09 for HD 80606, which indicate that the star is quite inactive. Also, Hébrard et al. (2010) monitored HD and determined it is not an active star. Specifically, they estimate that the star is photometrically stable at the level of a few mmag in the optical range on the time-scale of several weeks. Despite these statements, they still attribute the bump in their light curve to a spot on the stellar surface. Considering the precision of our observations (much better than 1 mmag), it is possible that we observed flux variations that they did not have the precision to. 76

77 As a final comment, we note that spots could affect the normalization of the overall spectrum, as the spectrum could need to be scaled downward (i.e. decrease the flux ratios or increase the transit depths) to account for the effect of spots. However, the shape of the spectrum would remain the same, unless over 26 per cent of the star s surface was covered with spots during transit. As noted by the referee, a large, long-lived polar spot could exist on HD 80606, which would not induce large photometric variations but could still affect our photometry. Or, both HD and HD or HD alone could be spotted and cause the observed variations. Due to the possible variable nature of HD (or HD 80607), we encourage future OOT observations of HD and HD to determine if such variability is common. 3.4 Conclusion In summary, our observations do not match existing models, due to two basic observations. We find a large change in apparent planet radius with wavelength, but do not observe a significant difference where the K I line core would be expected. Our observations place a strong limit on the strength of the line core (unless it has been Doppler-shifted by 100 km s 1 ), yet imply large variations in radius over wavelengths usually dominated by K I absorption. In the absence of other viable absorbers, absorption by K I remains the most viable explanation. The atmospheric scale height of HD 80606b at transit ( 20 km) is significantly smaller than that of HD b and HD b, yet the variation in radius is larger than that of HD b (Sing et al., 2008a). One possible model is absorption by potassium that is part of a high-speed wind coming off the exosphere. While high-speed winds have been observed for other exoplanets, the mechanism for powering such winds is unclear. We encourage further theoretical investigations to improve models for transmission spectroscopy of exoplanet exospheres in general and the specific challenge of HD 80606b. 77

78 Finally, we have investigated several potential sources of systematic effects. There is no simple or obvious source causing the systematics in our data. Further, any systematics introduced by the sources we have investigated here produce neither the same signature as our observed spectrum nor the same magnitude of difference as that of our measured flux ratios. While we are confident that none of these possible sources of systematic effects causes the shape of our observed spectrum, we still allow the possibility that one or some combination of these systematics may affect our measurements and/or the overall normalization of the observed spectrum. We also acknowledge that the target was observed at a slightly different set of wavelengths in January as compared to April. While the difference in wavelengths is small ( 0.1 nm), there is still the possibility that this could result in small differences in either the telluric absorption or stellar spectra, which in turn could cause the observed spectrum that we have attributed to absorption from the atmosphere of HD 80606b. As a final note, we highly encourage follow-up transit observations of HD 80606b to confirm the signal measured here. We note that the next partial transit observable from La Palma occurs on 2012 March 3, during which observations pre-transit through the complete first half of the transit will be possible. 3.5 Future Prospects Future transit observations at wavelengths around K I in HD 80606b are possible, but require considerable patience due to the long orbital period (111 d). Observations of the transit depth around other absorbing species could test the exosphere and wind models. Similar observations of other planets would enable a comparison of K I strength in both the wing and core as a function of star and planet properties. We note that shortly before submission, we became aware of independent, but similar, observations of another exoplanet (Sing et al., 2011). Both these and future observations of additional exoplanets will enable comparisons of the atmospheric composition and structure, as well as studies of potential correlations with other planet or host star properties. Such 78

79 observations would also help improve the interpretation of the existing HD 80606b observations. Currently, only the OSIRIS red TF ( nm) is available at the GTC. Once the blue TF is available for scientific observations, it will be possible to observe additional atmospheric features, including the Na I feature previously detected for HD b and HD b. The large aperture of the GTC makes it practical to perform similar observations of several fainter host stars. Thus, we look forward to future observations of a large sample of transiting planets. The striking diversity of exoplanets suggests that it will be fruitful to compare Na I and K I observations to identify trends with stellar and planet properties. Despite the complex interpretation of these observations, the very high precision obtained with the OSIRIS narrow-band TF imager opens up new avenues of research for large ground-based observatories. Indeed, the measured precision exceeds that of Spitzer (Hébrard et al., 2010) and even the HST observations for the given bandpass (Pont et al., 2008). Thus, ground-based observations can now characterize the atmospheres of giant planets using spectrophotometry. The photometric precision is also sufficient to measure emitted and/or scattered light during occultation at multiple near-infrared wavelengths that could improve constraints on atmosphere models of short-period giant planets. By providing high-precision photometry at multiple wavelengths during a single transit, the technique could also contribute to the confirmation of transiting planet candidates, such as those identified by Kepler (Borucki et al., 2010a). The technique could also improve measurements of the impact parameter and thus orbital inclination (Colón & Ford, 2009). This would be particularly valuable for systems with multiple transiting planets (Steffen et al., 2010), for which the orbital evolution depends on the relative inclination of the orbits (Ragozzine & Holman, 2010). 79

80 Since all Neptune and super-earth-sized planets will have relatively low surface gravities, they can make good targets for transmission spectroscopy. Despite a smaller transit depth than giant planets, the potentially large atmospheric scale height can lead to a substantial signal in transmission (Charbonneau et al., 2009), particularly for Neptune-sized planets orbiting sub-solar-mass stars and/or super-earth-sized planets orbiting low-mass stars. Previously, it has generally been assumed that the Earth s atmosphere will prevent ground-based facilities from achieving the high precisions necessary to measure biomarkers on super-earth-sized planets and that the James Webb Space Telescope will provide the first opportunity to characterize atmospheres of super-earths (Deming et al., 2009). If the challenges of Earth s atmosphere could be overcome, then ground-based observatories have several advantages (e.g. much larger collecting area, more modern and sophisticated instrumentation, ability to adjust and upgrade instruments). These observations demonstrate that ground-based narrow-band photometry on large telescopes can deliver the precision necessary to characterize super-earth-size planets around bright, nearby, small stars. We encourage astronomers to consider a future generation of instruments specifically designed for high-precision transit observations, which may allow the characterization of super-earth-sized planets in upcoming large ground-based observatories [e.g. Giant Magellan Telescope (GMT), Thirty Meter Telescope (TMT) and Extremely Large Telescope (ELT)]. 80

81 Table 3-1. Absolute transit photometry from 2010 January 13. λ (nm) HJD F target F ref Note. The wavelengths included in the table are the observed wavelengths in the frame of the planet (see text for additional details). The time stamps included here are for the times at mid-exposure. F target and F ref are the absolute flux measurements of HD and HD The full table is included online (Supporting Information), while a portion is shown here so the reader can see the formatting of the table. 81

82 Table 3-2. Absolute OOT photometry from 2010 April 4. λ (nm) t exp (s) HJD F target F ref Note. Columns are similar to Table 3-1, except the wavelengths included in the table are the wavelengths as observed from the GTC (see text for additional details). The second column contains the exposure time for the observations, as observations based on two different exposure times were included in our analysis. The full table is available online (Supporting Information), and a portion is shown here so the reader can see the formatting of the table. 82

83 Table 3-3. Relative transit photometry. λ (nm) HJD F ratio Uncertainty Note. The wavelengths included in the table are the observed wavelengths in the frame of the planet (see text for additional details). The time stamps included here are for the times at mid-exposure. F ratio represents the relative flux ratio between the target and reference star (i.e. F target /F ref ). The full table is available online (Supporting Information), and a portion is shown here so the reader can see the formatting of the table. 83

84 Table 3-4. Normalized photometry from around mid-transit. λ (nm) HJD F ratio F ratio Uncertainty (raw) (corrected) Note. The wavelengths included in the table are the observed wavelengths in the frame of the planet (see text for additional details). The time stamps included here are for the times at mid-exposure. The flux ratios are presented both before (raw) and after (corrected) EPD was applied. The flux ratios have also been normalized to the weighted mean OOT flux ratio (Table 3-6). The full table is available online (Supporting Information), and a portion is shown here so the reader can see the formatting of the table. 84

85 Table 3-5. Relative OOT photometry from 2010 April 4. λ (nm) t exp (s) HJD F ratio F ratio Uncertainty (raw) (corrected) Note. The wavelengths included in the table are the wavelengths as observed from the GTC (see text for additional details). The time stamps included here are for the times at mid-exposure. The flux ratios are presented both before (raw) and after (corrected) EPD was applied. The full table is available online (Supporting Information), and a portion is shown here so the reader can see the formatting of the table. 85

86 Table 3-6. Time-averaged flux ratios and noise estimates. λ E (nm) λ P (nm) λ S (nm) < δf /F > σ <δf /F > σ w σ r β In-transit Out-of-transit Note. λ E is the observed wavelength from the GTC (i.e. from the Earth), λ P is the observed wavelength in the frame of the planet and λ S is the observed wavelength in the frame of the star. Values for λ P are not given for the OOT observations, as the planet was not transiting and was therefore not technically observed. The in-transit ratios refer to the relative flux ratio between the target and reference that has been normalized to the weighted mean OOT flux ratios (given at the bottom of the table).

87 Table 3-7. Time-averaged flux ratios and noise estimates (outlying absolute fluxes excluded). λ E (nm) λ P (nm) λ S (nm) < δf /F > σ <δf /F > σ w σ r β In-transit Out-of-transit Note. Same as in Table 3-6, but the flux ratios listed here are those computed after excluding outlying absolute flux measurements from the analysis.

88 Table 3-8. Absolute OOT photometry from 2010 January 15. λ (nm) HJD F target F ref Note. Columns are similar to Table 3-1, except the wavelengths included in the table are the wavelengths as observed from the GTC (see text for additional details). The full table is available online (Supporting Information), and a portion is shown here so the reader can see the formatting of the table. 88

89 Table 3-9. Relative OOT photometry from 2010 January 15. λ (nm) HJD F ratio F ratio Uncertainty (raw) (corrected) Note. The columns are similar to Table 3-5. The full table is available online (Supporting Information), and a portion is shown here so the reader can see the formatting of the table. 89

90 Figure 3-1. Absolute fluxes of HD (a) and HD (b) as measured on 2010 January The different light curves represent the fluxes as measured nearly simultaneously in the different bandpasses, with the black, blue, brown and red light curves representing the , , and nm data. These data have not been corrected for airmass or decorrelated in any way. Note the break in the data around 2 h after mid-transit due to recalibration of the TF. The vertical solid lines indicate the expected beginning and end of the transit, and the vertical dotted lines indicate the end of ingress and the beginning of egress [based on durations estimated by Hébrard et al. (2010) and the transit ephemeris from Shporer et al. (2010)]. The vertical dashed lines indicate the 4 h interval around mid-transit that our analysis focused on (see text for further details). 90

91 Figure 3-2. Similar to Figure 3-1, but for the OOT data taken the night of 2010 April 4. Note that the discontinuity in the fluxes around is due to a change in the exposure time (from 8 to 11 s). 91

92 Figure 3-3. Transit light curves as observed nearly simultaneously in different bandpasses on 2010 January The on-line light curve ( nm) is shown in black, and the off-line light curves (768.76, and nm) are shown in blue, brown and red. The flux ratio for each bandpass has been normalized to the weighted mean OOT flux ratio estimated from the baseline data acquired in 2010 April, but the data have not been corrected for airmass or decorrelated in any way. The off-line light curves have been arbitrarily offset by 0.006, and 0.018, and error bars are not shown for clarity. The vertical solid, dotted and dashed lines are the same as in Figure

93 Figure 3-4. Relative in-transit flux ratio normalized to the relative OOT flux ratio as measured on 2010 April 4. The relative flux before (a) and after (b) EPD was applied is shown. The different colors represent the flux ratios as measured in the different bandpasses, with the colors the same as in Figure 3-3. Note that EPD was only applied to the 4 h centered around mid-transit (i.e. the bottom of the transit light curve). The data shown have not been binned, but the different light curves have been offset arbitrarily for clarity. 93

94 Figure 3-5. Correlations between the normalized in-transit flux ratio and the target FWHM and x and y centroid coordinates, before (left-hand column) and after (right-hand column) EPD has been applied. All four bandpasses are shown in each panel, with the colours the same as in Figure 3-3. Similar results were obtained when decorrelating the data against the reference parameters but are not shown here. 94

95 Figure 3-6. Relative OOT flux ratio as measured on 2010 April 4. The relative flux before (a) and after (b) EPD was applied is shown. The different colors represent the flux ratios as measured in the different bandpasses, with the colors the same as in Figure 3-3. Note the small break in the data around where the exposure time was changed. The data have not been binned, but the different light curves have been offset arbitrarily for clarity. 95

96 Figure 3-7. Corrected light curves for observations of the bottom of the transit as observed nearly simultaneously in different bandpasses on 2010 January In each panel, the black points illustrate the measurements taken in the on-line ( nm) bandpass. We also show measurements taken in each of the off-line bandpasses (768.76, , nm) in each of the respective panels (a, b, c) for comparison to the on-line flux ratios. The data shown here have been decorrelated. The colours and normalizations are the same as in Figure 3-3, but no offsets have been applied. Here, we have binned the data and error bars simply for clarity. 96

97 Figure 3-8. Histograms of normalized flux ratios from the bottom of the transit light curve as shown in Figure 3-4(b). The histograms were generated using a bin size of 0.5 mmag. Each panel compares the on-line flux ratios with the off-line flux ratios. In each panel, the black (solid) histograms represent the nm (on-line) light curve. The blue (dotted), brown (dashed) and red (dot-dashed) histograms are for the , and nm light curves and are shown in panels (a), (b) and (c), respectively. Panel (d) shows the histograms for all four wavelengths for further comparison. 97

98 Figure 3-9. Standard deviation of the time-binned flux ratio measurements from the bottom of the transit [e.g. as shown in Figure 3-4(b)] as a function of the number of data points per bin (N). Panels (a), (b), (c) and (d) show the standard deviations for the binned , , and nm light curves. The amount of binning that could be performed varies for each light curve since the different wavelengths were observed a different number of times in a given observing sequence (Section 3.1.1). The solid line in each panel represents the trend expected for pure white Gaussian noise ( N 1/2 ), normalized to the unbinned standard devation measured in our data. The dotted lines represent the trend for Gaussian noise when normalized to the theoretical noise for our observations. The dashed curves are models fitted to the standard deviation that include both white and red noise. The effect of red noise is obvious in all bandpasses. 98

99 Figure Standard deviation of the time-binned OOT flux ratio measurements from 2010 April [e.g. as shown in Figure 3-6(b)] as a function of the number of data points per bin (N). Panels (a), (b), (c) and (d) show the standard deviations for the binned , , and nm light curves. The solid line in each panel represents the trend expected for pure white Gaussian noise ( N 1/2 ). The dotted lines represent the trend for Gaussian noise when normalized to the theoretical noise for our observations. The dashed curves are models fitted to the standard deviation that include both white and red noise. Compared to the in-transit observations, red noise has a very minimal effect here. Deviations below the curve are likely due to small number statistics. These results demonstrate that narrow-band ground-based observations can provide very high precision differential photometry. For a given bandpass, the combined precision exceeds that of Spitzer (Hébrard et al., 2010) or HST observations (Pont et al., 2008). To the best of our knowledge, these represent the highest precision photometry for a 1.2-nm bandpass for ground or space observations. 99

100 Figure Normalized weighted mean in-transit flux ratio versus observed wavelength (in the frame of the planet). The open triangles represent the flux ratios as computed for each light curve described in Sections 3.1 and 3.2. The solid circles represent the flux ratios computed after excluding outlying absolute flux values for each star from the analysis (Section 3.2.1). Note that the solid circles have been offset by 0.25 nm for clarity. The vertical error bars include a factor to account for the effects of red noise in both the in-transit and OOT data. The error bars in the horizontal direction indicate the FWHM of each bandpass. The solid squares represent the mean in-transit flux ratios estimated from limb-darkened transit light curve models for HD 80606b. The lines show the predictions of planetary atmosphere models (Section 3.3.4). The inset figure shows the atmosphere models on a small vertical scale. While LD or night-to-night variability (of Earth s atmosphere or either star) could affect the overall normalization, the observed change in the flux ratio with wavelength is robust. 100

101 Figure Colours of the normalized in-transit flux ratios. The different panels show the colour as computed between each off-line bandpass and the on-line bandpass (after binning the on-line data to the number of points in each of the off-line bandpasses). The dashed line in each panel illustrates where the colour equals zero. The data has not been explicitly offset, and there are no obvious systematics seen in any of the colours. 101

102 Figure Standard deviation of the time-binned colour measurements from the bottom of the transit (as shown in the different panels in Figure 3-12). The different panels show the standard deviations for the different colours as presented in the panels in Figure 3-12, with panels (a), (b) and (c) respectively showing the standard deviations for the nm, nm and nm colours. The solid line in each panel represents the trend expected for pure white Gaussian noise ( N 1/2 ). The dotted lines represent the trend expected for Gaussian noise when normalized to the unbinned theoretical uncertainties for these observations. There is no obvious presence of red noise at large binning factors. 102

103 Figure Mean colour of the in-transit flux ratios as computed between each off-line bandpass and the on-line bandpass. The open triangles represent the colours as computed in Section and illustrated in Figure The solid circles represent the colours computed after excluding outlying absolute flux measurements for each star from the analysis (Section 3.2.1). The error bars represent the 1σ uncertainties. The dashed line illustrates where the colour equals zero. We arbitrarily set this point equivalent to an apparent planet radius of 1 (i.e. we let the measured radius in the on-line bandpass be the baseline radius of HD 80606b). The mean colours around the nm bandpass are essentially equal for both sets of points, so the two data points appear as one. 103

104 CHAPTER 4 A SEARCH FOR METHANE IN THE ATMOSPHERE OF GJ 1214B VIA NARROW-BAND TRANSMISSION SPECTROPHOTOMETRY The discovery of super-earths, planets with masses between 1.5 and 10 M, opened an entirely new field of exoplanet research. Previously, astronomers had been surprised by the discovery of hot-jupiters, Jupiter-size planets in extremely close orbits around their host stars. Such discoveries inspire new theories, new models, and new observations, so that we may try to understand how objects like super-earths and hot-jupiters form and evolve. Furthermore, investigations of such planets help us better understand why our own Solar System does not contain any such planets. One of the more well-studied super-earths is GJ 1214b, which was discovered in a ground-based transit survey (MEarth) by Charbonneau et al. (2009). Because GJ 1214b orbits an M dwarf star, it has a larger planet-star radius ratio than most other super-earths discovered to date. This results in a transit depth that is nearly 1.5%, which immediately lends itself to making GJ 1214b an excellent candidate for atmospheric studies. Furthermore, based on its observed mass and radius, it was believed that GJ 1214b was required to have a significant atmosphere (Miller-Ricci & Fortney, 2010). However, there is a degeneracy in the models due to GJ 1214b s density and irradiation (Rogers & Seager, 2010). As a result, it is predicted that GJ 1214b is either composed of a rocky/ice core surrounded by a hydrogen-rich atmosphere, a water/ice core with a water vapor atmosphere, or a rocky core with a thin atmosphere formed by outgassing. Recent studies have attempted to constrain GJ 1214b s atmosphere through transmission spectroscopy (e.g., Bean et al., 2010, 2011; Croll et al., 2011; Crossfield et al., 2011; Désert et al., 2011; Berta et al., 2012; de Mooij et al., 2012). In a majority of these studies, it was found that GJ 1214b has a flat, featureless spectrum. With no evidence of any significant features either in optical ( nm) or near-infrared ( µm) wavelengths. It is believed that the lack of significant features supports the 104

105 presence of either a heavy, metal-rich atmosphere or thick clouds/hazes that produce a constant level of absorption across a large range of wavelengths. One exception is a study conducted by Croll et al. (2011). Specifically, they reported a significantly deeper transit depth at 2.15 µm, a wavelength where methane would be a viable source of opacity. To reconcile these studies, we present narrow-band photometry of three transits of GJ 1214b around a predicted methane absorption feature found in the optical (predicted based on the models from Miller-Ricci & Fortney, 2010). The broad methane feature that we focus on is found around 890 nm and is predicted to cause additional absorption during transit at a level of 0.1% (assuming a hydrogen-rich atmosphere). In 4.1, we describe our observations (acquired using the OSIRIS instrument installed on the 10.4 m GTC). In 4.2 and 4.3, we describe our data reduction and analysis procedures. We present our results in 4.4, and we discuss the implications of our results on future atmospheric studies of planets like GJ 1214b in 4.5. Finally, we conclude with a summary of our findings in Observations We observed three transits of GJ 1214b, during which we acquired photometry of GJ 1214 and three nearby reference stars (hereafter known as Ref 1, Ref 2, and Ref 3). We also acquired additional baseline (out-of-transit) photometry of the same four stars on a night following one of the transit events. For all observations, we used the tunable filter (TF) imaging mode on GTC/OSIRIS to acquire near-simultaneous photometry in two narrow bandpasses (FWHM = 1.2 nm) by alternating between the two bandpasses during the observations. The TF imager allows for custom bandpasses with a central wavelength between nm and a FWHM of nm to be specified. Therefore, we specifically chose our bandpasses so that one was located in the continuum, to the left of where we expect the methane absorption feature to be. The other bandpass was located within the assumed methane absorption feature, 105

106 around the expected peak of the absorption. The final observing sequence that we used was to take two exposures in the continuum bandpass, two exposures in the methane bandpass, and repeat. The specific bandpasses used are given in the following sections, as they varied slightly for the different observations. As described in Colón et al. (2010), another feature of the TF imaging mode is that the effective wavelength decreases radially outward from the optical center, so we attempted to position the target and a single primary reference star (i.e., one most comparable in brightness to the target) at the same distance from the optical center so as to observe both stars at the same wavelengths. 1 The other reference stars were then observed at slightly different wavelengths than the target due to their being at different distances from the optical center. All observations used 1 1 binning, a fast pixel readout rate of 500 khz, as well as a single window located on one CCD chip. The size of the window varied slightly for each observation, but was chosen to be large enough so as to contain the target and all three reference stars. Further details regarding each specific transit observation are given in the following sections July 22 Transit We observed a transit of GJ 1214b on 2010 July 22 under photometric conditions and during bright time. Observations began at 00:26 UT and ended at 02:11 UT. The airmass ranged from 1.25 to The actual seeing varied between 0.9 and 1.4 arcsec, but a slight defocus was used in order to avoid saturation. The exposure time (for both bandpasses) was set to 120 s, with a corresponding 22 s of dead time following each exposure. The dead time includes the time to read out a window 1 Due to technical issues, the positioning for some of the observations was not as expected, and the target and a single reference star were not always observed at the same exact wavelengths. See the following sections for further details. 106

107 of pixels located in CCD2 of OSIRIS, which contained the target and the three reference stars. The target and primary reference star (Ref 1) were positioned at distances of 2.86 and 3.32 arcmin from the optical center, which resulted in the stars being observed at different wavelengths. Therefore, while the target was observed at nm and nm, the primary reference star was observed at nm and nm. Likewise, the other two reference stars in the field were observed at slightly different wavelengths than the target. These observations were conducted in queue (service) mode August 28 and 29 Transit and Baseline Observations Observations of the 2010 August 28 transit of GJ 1214b took place from 22:00 UT (2010 August 28) to 00:30 UT (2010 August 29), during bright time and under mostly clear conditions (though some dust was present). The airmass ranged from about 1.26 to The seeing was variable throughout the observations, so the telescope was defocused and the exposure time was changed to avoid saturation. An exposure time of 100 s was initially used, but when the seeing became worse, the exposure time was increased to 150 s (before the transit ingress). Finally, the exposure time was increased again to 200 s towards the end of the observations (after transit egress). The dead time following each exposure was typically about 22 s, which included reading out a single window of pixels containing the target and reference stars, located in CCD2 of OSIRIS. As for the first transit we observed ( 4.1.1), the target and a primary reference star were positioned at slightly different distances from the optical center (3.0 and 3.3 arcmin). This resulted in the target being observed at wavelengths of nm and nm, while the primary reference was observed at nm and nm. We note that based on the observations presented in 4.1.1, in which we found that the primary reference (Ref 1) appeared to be variable, we chose a different star (Ref 3) as our primary reference for these observations. Finally, there are some small gaps in the data towards the beginning of the observations due to minor technical issues. Also, 107

108 there is some vignetting in the last few images due to the low elevation of the telescope, so we exclude these from our analysis. Specifically, all the images with an exposure time of 200 s were discarded. Additional baseline (out-of-transit) photometry was acquired on the night of 2010 August 29. The same set up was used for these observations as was used for the observations conducted on the night of 2010 August 28. Observations began at 21:17 UT (2010 August 29) and ended at 00:20 UT (2010 August 30), during which the airmass ranged from 1.17 to An exposure time of 150 s was used for these observations, to match the exposure time used during a majority of the transit observations from the previous night. As in the previous night s observations, some vignetting occurred towards the end of the observing run, so we exclude these later images from our analysis. We note that both of these observations were also conducted in queue (service) mode June 11 Transit Observations of the 2011 June 11 transit of GJ 1214b were conducted in visitor mode. The conditions were clear, and observations took place in bright time. Observations began at 23:40 UT (2011 June 10) and ended at 02:50 UT (2011 June 11), during which time the airmass ranged from about 1.09 to 1.19 and the seeing varied between 1.0 and 1.1 arcsec. A single window of pixels containing the target and reference stars was read out. Unlike the previous observations, the stars were located in CCD1 of OSIRIS. An exposure time of 100 s was used, with a corresponding 19 s of dead time following each exposure. Also unlike our previous observations, we were able to set the target and a primary reference star (Ref 3, the same primary reference as in the August 2010 observations) at the same distance from the optical center (3.2 arcmin). Thus, both the target and primary reference were observed at wavelengths of nm and nm. Finally, we note that for the duration of the observations, there was a problem with the GTC s primary mirror. One segment of the mirror would not stack with 108

109 the other segments. However, as this problem had the same effect on all the stars (i.e., each star had a very small copy of itself located towards the bottom right of the star on the CCD), we assume that the photometry would not be affected by this problem as aperture photometry would still include the photons from the unstacked segment. 4.2 Data Reduction For all our data sets, we used standard IRAF procedures for bias subtraction and flat-field correction. For the flat-field correction, we used dome flats that were taken after each observation and for each filter setting. However, since the dome lights do not provide a uniform illumination, we added an illumination correction to the final flat-field images for each data set. As discussed in Colón et al. (2010) and Colón & Ford (2011), the OSIRIS TF produces images with sky (OH) emission rings. Therefore, we performed sky subtraction using the IRAF package TFred 2 prior to performing aperture photometry. This task specifically measures the sky background while including the rings due to sky emission. After running this task on all the science images, we used standard IDL routines to perform aperture photometry on the target and the three reference stars in the field. We tested several different size apertures for each data set, and we set our final aperture to be the one that yielded the smallest scatter in the baseline (out-of-transit) flux ratios (computed as the target flux divided by the sum of the reference star fluxes). We considered the individual results for each data set and applied the same aperture to each filter (for a given transit observation). For the transit observations in July 2010, August 2010, and June 2011, we used apertures of 42, 45, and 52 pixels (approximately 5.3, 5.7, and 6.6 arcsec). The aperture used for the baseline observations from August 2010 was 56 pixels, equivalent to 7.1 arcsec. We did not need to include a sky annulus in the aperture photometry process because the 2 Written by D. H. Jones for the Taurus Tunable Filter, previously installed on the Anglo-Australian Telescope; reduc.html 109

110 TFred task had already removed the sky background. Once aperture photometry had been completed for each data set, we continued with the light curve analysis described in the following section. 4.3 Light Curve Analysis We computed light curves for each bandpass and for each observation separately by dividing the total flux measured within the target aperture by the total flux of the reference star(s) for a given bandpass/observation. Since we had only three reference stars in our field, we computed light curves for the target relative to each of the reference stars as well as to different combinations of the reference stars. When using more than one reference star, we computed the total weighted flux of the reference star ensemble, where we weighted the flux from each reference star after discarding any outlying flux measurements. Since all the reference stars were typically located at different distances from the optical center than the target (except for the third transit, observed in June 2011, where the primary reference star was indeed located at the same distance from the optical center as the target), we opted to test all possible light curves in order to compare the results from each one and to determine if any specific reference star(s) provided a more stable light curve than the others. 3 After computing the relative flux ratios, we normalized each light curve to the mean baseline (out-of-transit) flux ratio measured for each bandpass. We then corrected each light curve against changes in airmass. Also, for the June 2011 transit, we removed an additional quadratic trend that was present over the duration of the observations. Finally, we performed external parameter decorrelation (e.g., Bakos et al., 2007, 2010) against the x and y centroid coordinates of the target and the sharpness (i.e. the FWHM) of the target s profile. We note that the corresponding observation times for each light curve were computed from 3 We also used the baseline observations from August 2010 to check the stability of GJ 1214 relative to each reference star. See the following section for further details. 110

111 the UTC timestamps given in the image headers. The UTC time at mid-exposure was then converted to Barycentric Julian Date in Barycentric Dynamical Time (BJD TDB) via an online calculator. 4 In order to compute the photometric uncertainties on each data point, we consider the photon noise of the target and the reference star(s) used to compute the light curves, the noise in the sky background around the target and the reference(s), and scintillation noise. We find the median photometric uncertainties for all the light curves to typically be between 0.5 and 0.6 mmag. In all cases, the photon noise of either the target or Ref 1 dominates the errors, but in general the target and each reference star have a comparable level of photon noise. Finally, we investigated whether red noise has a significant presence in our data by computing the standard deviation of the flux ratios after binning the data. In general, we found that our data tend to follow the trend expected if only white Gaussian noise were present. However, for the observations from July 2010, there appeared to be some residual systematics in the light curves due to the inclusion of Ref 1 (the one we used as a primary reference for that observation). We attribute this red noise primarily to variability in that star, as that specific reference star consistently produced light curves that were highly variable from transit to transit. As a result, we investigate the effect of these potential residual systematics in our data via a prayer-bead analysis (e.g., Désert et al., 2009), and we discuss our results in further detail below. For the three transit observations, we fit synthetic models to each corrected light curve. Specifically, we followed an approach similar to that taken by Colón et al. (2010), and we used the planetary transit light curve models from Mandel & Agol (2002) to fit limb-darkened models to our data. For each light curve, we fixed the impact 4 As described in Eastman et al. (2010). The calculator can be found at 111

112 parameter (a cos i/r ) and limb darkening coefficients (c 1 and c 2 ) 5 to values given in the literature. The reason for this is that due to the relatively long exposure times used ( s) in comparison to the short transit duration (< 1 h), we acquired very little data, if any, during the transit ingress and egress, which makes it difficult to constrain parameters such as the impact parameter and limb darkening. We then fit for the following parameters: time of mid-transit (t 0 ) transit duration (from first to fourth contact; τ) planet-star radius ratio (R p /R ) baseline flux ratio (linear) baseline slope The limb darkening coefficients were based on the coefficients for the Sloan z filter as reported by Carter et al. (2011). During the fitting process, we kept the coefficients fixed at self-consistent values, but we did test a range of values based on the uncertainties given by Carter et al. (2011). The values that we used for the impact parameter were taken from Berta et al. (2011), as were the initial guesses for the mid-transit time, transit duration, and radius ratio. We also tested a range of initial guesses for these parameters based on the uncertainties given in Berta et al. (2011). We identified best-fitting models via a Levenberg-Marquardt minimization scheme. 6 We modeled the light curves for each individual transit observation following a similar procedure as Colón et al. (2010). We then applied a similar analysis to light curves that were generated by combining the data from all three transit observations. 5 Here, we define the limb darkening coefficients as c 1 u 1 + u 2 and c 2 u 1 u 2, where u 1 and u 2 are the linear and quadratic limb darkening coefficients. 6 We specifically used mpfitfun, which is publicly available at craigm/idl/idl.html 112

113 To summarize the procedure here, we fit models to each light curve for each bandpass and each transit observation individually, then used the models to correct the data and discard any points lying greater than 3σ from the residuals. We then fit models to the corrected light curves in a joint analysis, where we forced the mid-transit time, transit duration, baseline flux ratio and baseline slope to be the same for both light curves (for a given transit). However, we allowed the different light curves to have different values for the radius ratio and limb darkening coefficients. We then corrected the individual light curves again, this time based on the results from the joint analysis. A final joint analysis was applied to the final corrected light curves. During this final stage, we also performed a prayer-bead analysis. Specifically, we performed a circular shift on the residuals for each light curve and generated synthetic light curves by adding the shifted residuals back to the best-fit model. We applied the joint analysis to each synthetic light curve, and we compare the dispersion of the best-fit parameters to the formal 1σ errors on each fitted parameter. This allows us to investigate the effect of any additional systematic noise sources in the data, as discussed above. We repeated this process for the combined light curves, where we considered all three transits together. We also performed this analysis for all the different light curves that we had computed for GJ 1214 that were based on different combinations of reference stars. We present the results from our analysis in the following section. 4.4 Results In this section, we first describe the overall results from all our observations, and then we discuss the results for each individual transit observation in detail in Since we only had three reference stars in our field, we first considered the results for our baseline observations from August 2010 to determine whether any specific reference star or combination of reference stars was obviously more stable than others. In Figure 4-1, we present the light curves for GJ 1214 relative to each combination of reference stars from the August 2010 baseline observations. Visual 113

114 inspection reveals that the light curves are all fairly scattered, but we find that using multiple reference stars helps to reduce the scatter and level out variability due to either stellar activity and/or non-ideal weather conditions (particularly in the case of these observations, some dust was present and could have affected our photometry). However, including Ref 1 specifically tends to introduce the most scatter, so we conclude that we should generally disregard results that are based on the inclusion of Ref 1. This is further supported by our other observations, as discussed below. We present the light curves and the corresponding best-fit models for each transit observation in Figures 4-2, 4-3, and 4-4. In Figure 4-5, we show the light curves generated from the combination of all the transit observations. In each figure, we again show light curves for GJ 1214 relative to each combination of reference stars. Also, for ease of comparison, all of these figures are plotted with the same ranges on the x- and y-axes. As already illustrated in Figure 4-1, we find that when we include Ref 1 in the analysis, the light curves tend to be more scattered, particularly in the case of the August 2010 transit (Figure 4-3). Therefore, in the following sections, we focus on the results from the following cases, which we later refer to as cases 2, 3, and 6: GJ 1214 normalized to Ref 2, GJ 1214 normalized to Ref 3, and GJ 1214 normalized to Ref 2 and Ref 3. Before we discuss the specific results from each transit observation, in Figures 4-6, 4-7, 4-8, and 4-9 we present the best-fit radius ratios for each observation and each individual light curve. As before, for ease of comparison, all figures have the same ranges on the x- and y-axes. The best-fit values and corresponding 3σ error bars that are shown are based on the model with the smallest χ 2 value. While not illustrated here, we also compare the formal errors with the scatter in the fitted parameters, computed from the full set of models generated in the prayer-bead analysis. We find that the distribution of the best-fit radius ratios over all permutations is generally comparable to the formal 1σ errors found from the models fit to the first two transit observations 114

115 (July 2010 and August 2010). However, for the third transit (June 2011), we find that the distribution is scattered, and is sometimes smaller or larger than the formal errors, depending on which reference star is included in the analysis. There is no trend as to which reference star causes a skewed distribution of best-fit radius ratios. Based on visual inspection of the light curves (Figure 4-4), we find that there is an unusual feature present in each of the light curves at nm, which suggests that activity in GJ 1214 may be the culprit of such a feature. Despite this, we believe the 3σ error bars shown in Figures are largely representative of the true errors in the fitted radius ratios. While the error bars may in fact be larger in some cases, we do not expect them to be significantly larger than what is shown in these figures. While not shown here, we constructed similar figures for the impact parameter, the transit duration, and the mid-transit time, and we found that (a) a smaller impact parameter was generally preferred, (b) the transit duration tended to match values from the literature (within a few minutes), and (c) the measured mid-transit time typically deviated from the expected transit ephemeris by less than one minute. In the following sections, we discuss the results from each specific transit observation, as well as from the combination of all three transits Results from the 2010 July 22 Transit As illustrated in Figures 4-2 and 4-6, for these observations we consistently measure transit depths (and therefore radius ratios) that are larger in the bluer bandpass (i.e., the one located in the continuum). Specifically, we find that the radius ratios as measured in our two bandpasses consistently differ by slightly more than 3σ. We find this result even when we discount measurements that include Ref 1 (which we found to be variable from observation to observation). Also, looking at just the results for cases 2, 3, and 6, we find that the radius ratios measured in each bandpass are consistent between the different cases, and that the radius ratios measured in the bluer bandpass are consistent with that reported by Berta et al. (2011). This is interesting, as none 115

116 of the references were observed at the same wavelength as the target in this transit observation. This suggests that for these reference stars, observing them at different wavelengths than GJ 1214 has a negligible effect on our results. Regardless, our finding that the radius ratios are larger in the continuum bandpass contradicts what we would expect if methane were present and absorbing in GJ 1214b s atmosphere Results from the 2010 August 28 Transit In Figure 4-3, we clearly see that Ref 1 introduced a large amount of scatter during these observations. Therefore, we disregard any results based on Ref 1. Looking at just cases 2, 3, and 6, we find that in cases 2 and 6, there appears to be no significant difference in the transit depths or radius ratios between the two bandpasses. However, in case 3, we find that the radius ratio in the red bandpass (i.e., the methane band) is larger than that in the blue bandpass at a level slightly greater than 3σ. Correspondingly, the radius ratio measured in the red bandpass is consistent between all three cases, while the radius ratio found for the blue bandpass is consistent only for cases 2 and 6. However, despite any consistencies, none of the measured radius ratios for cases 2, 3, and 6 match the value reported by Berta et al. (2011). In fact, all our radius ratios are smaller than that found by Berta et al. (2011). These results suggest that there may some stellar activity in Ref 3 that is affecting the light curve for GJ 1214 and/or that in this case, the fact that none of the references were observed at the same wavelength as the target is important. Also, because we consistently measure a smaller radius ratio than Berta et al. (2011), it may be that GJ 1214 happened to have fewer star spots on its surface during this observation than it usually does (assuming it typically has some number of spots on its surface). Having fewer star spots would imply that GJ 1214 appeared slightly brighter than it typically does, so that the fractional loss of light due to the planet during transit would decrease, resulting in a shallower transit and therefore a smaller measured planet-star radius ratio (we also refer the reader to Carter et al., 2011 for further discussion on the effects of star spots on the transit light curve for GJ 1214b). 116

117 4.4.3 Results from the 2011 June 11 Transit In this transit observation, we see an anomalous feature present in each of the light curves acquired in the redder bandpass (Figure 4-4). The feature is not indicative of a star spot crossing, as the light curve appears to get slightly deeper around the beginning of the transit egress. Also, the feature is most prominent when including Ref 1, so we again consider only cases 2, 3, and 6. As the feature is less visible in these cases, we do not discard any of these points in the fitting process. Regardless of the source of the anomalous feature, we find that the radius ratio is consistently larger in the redder bandpass at a level greater than 3σ (Figure 4-8), which suggests possible additional absorption due to methane in GJ 1214b s atmosphere. Furthermore, in cases 3 and 6, we measure a radius ratio in the blue bandpass that matches that from Berta et al. (2011). This matches our results from the first transit observation ( 4.4.1). However, we find a different radius ratio in the blue bandpass for case 2 as well as much larger radius ratios in the red bandpass compared to those measured in the first transit observation. As in the second transit observation, it might be that we can attribute these differences to varying stellar activity (e.g., different numbers of star spots on GJ 1214 s surface). Considering that we are finding different results from each of our transit observations, we believe that stellar activity may be an extremely significant factor in our observations. We discuss this in further detail in Results from All Transits Despite the different observing conditions for the three transits that we observed, we investigate the results from combining all three transits together. Our goal was to compare these results with the results found for the individual transit observations. In this case, we find that the measured radius ratios are highly correlated with the individual reference stars (i.e., cases 1, 2, and 3). However, when including more than one reference star in the analysis, we find that the radius ratios are consistently slightly larger than that found by Berta et al. (2011). Furthermore, the radius ratios between 117

118 the two bandpasses tend to be consistent for all cases where multiple references are included. From this combined analysis, we cannot confirm or refute the presence of methane at these wavelengths. As in 4.4.3, we conclude that stellar variability (likely in the target and references) has a significant effect on our measurements. 4.5 Discussion We observed three transits of GJ 1214b using narrow-band photometry, with the goal of searching for extra absorption due to methane in GJ 1214b s atmosphere. From our analysis, we find that all three transits yield inconsistent results in regards to our search for methane absorption. Specifically, for the first transit (July 2010), we unexpectedly measure a slightly larger planet-star radius ratio in a bandpass that should be located in the continuum (to the left of the predicted methane absorption feature). For the second transit (August 2010), we measure consistent radius ratios between our two bandpasses when using some reference stars, but we also find a case where the radius ratio is larger in a bandpass that should be located on a methane absorption feature. Regardless, these radius ratios do not match values from the literature, as they are significantly smaller. For the third transit (June 2011), we consistently find a larger radius ratio in a bandpass that is located in the predicted methane absorption band. While we measure a significant difference in the radius ratios, which could indicate extra absorption due to methane in GJ 1214b s atmosphere, the values for the radius ratios for each bandpass vary quite a bit depending on which reference stars are included in the analysis. Finally, when we consider all three transits together, we consistently find no significant difference between the radius ratios measured in our bandpasses (except for the case where we only use Ref 1, but that star has appeared to have significant transit-to-transit variability). Ultimately, we find that the results from each transit observation are inconsistent. However, we must consider the fact that each observation was conducted under different conditions (most relevant is that in two of the three transits, no reference stars were 118

119 observed at the same wavelengths as GJ 1214). If we consider the best-case scenario, which would be the case where we observed GJ 1214 and another reference star at the same wavelength, then that leads us to the third transit observation (specifically the third panel in Figures 4-4 and 4-8), where we observed GJ 1214 and Ref 3 at the same wavelength. For this specific case, we measure radius ratios that are different by more than 3σ, with the larger radius ratio being measured in the bandpass that is located in the predicted region of methane absorption. Furthermore, the radius ratio in the continuum bandpass is an excellent match with values from the literature. This suggests that when all our observing criteria were met, we were indeed able to find potential evidence of methane in GJ 1214b s atmosphere. However, while such a result would have a significant impact on the composition of GJ 1214b s atmosphere, we cannot dismiss the fact that our other transit observations reveal the highly variable nature of such studies. In the following sections, we discuss three potential causes of variability in our measured radius ratios. We specifically focus on variability within GJ 1214b (the planet, specifically its atmosphere), GJ 1214 (i.e., activity due to star spots), and Earth (i.e., our own atmosphere). While we do not discuss instrumental effects as a source of variability here, we are confident that we have already accounted for the most significant effects ( 4.3) and that any residual systematics due to the instrument would have a negligible effect on our results (e.g., Colón et al., 2012) Variability due to GJ 1214b s Atmosphere Based on early models for GJ 1214b s atmosphere (Miller-Ricci & Fortney, 2010), we estimate the expected signal due to methane absorption in our bandpasses would be 0.1% (assuming a hydrogen-rich atmosphere). Looking again at our best-case scenario (case 3 for the transit observation from June 2011), there we measured planet-star radius ratios of ± and ± in our blue (continuum) and red (methane) bandpasses. This translates to a change in the 119

120 flux ratios between our two bandpasses at a level of 0.29%, which is nearly three times larger than the expected signal. We consider that if methane were present and absorbing in GJ 1214b s atmosphere, the strength of the absorption line could be time-variable if there are, for example, high-speed and high-altitude winds in the atmosphere. However, to be as variable as ranging from 0.1% to nearly 0.3% in absorption is unlikely, unless there was much more methane present in the atmosphere than predicted. We conclude that while we may have measured absorption due to methane, the level of absorption we measured is likely far too large for there not to be other sources of variability affecting our results. Furthermore, considering that recent studies support a flat, featureless spectrum for GJ 1214b (e.g., Bean et al., 2010; Berta et al., 2012), we believe that as a whole GJ 1214b s atmosphere is likely quite stable Variability due to Stellar Activity GJ 1214 has been previously shown to have some level of activity due to spots and flares (e.g., Kundurthy et al., 2011), so it is not surprising that we find inconsistent results from transit-to-transit. However, our observations were acquired at fairly red wavelengths, where the contrast between star spots and the stellar photosphere is minimized. Therefore, we might expect to find more consistent results from transit to transit than we actually did, unless the number of spots present on GJ 1214 s surface varies significantly. Together, this suggests that perhaps the reference stars and/or GJ 1214 are more variable than previously believed ( 4.5.3). We note that it is also possible that our observations probed a part of the GJ 1214 s spectrum that is highly variable. For instance, TiO is typically present in M4.5 stars around 880 nm (Kaler, 2011), and, if variable, could have affected our measurements of the planet-star radius ratios. In any case, stellar activity surely plagues studies such as the one presented here Variability due to Earth s Atmosphere We expect that any variability due to Earth s atmosphere is removed in our data reduction procedure, where we normalize each observation of GJ 1214 by the flux 120

121 of a reference star taken at the same time. However, we have to consider that in the case of our first two transit observations, GJ 1214 and each of the reference stars were observed at different enough wavelengths that any variability due to species in Earth s atmosphere may not have been entirely removed. To investigate this, we consider the light curve for Ref 2 normalized to Ref 3 (after correcting for changes in airmass), as they are the two most stable references in our field. Even though they were observed at different wavelengths, we use this light curve to determine how stable the Earth s atmosphere was during each observation. Specifically, we find that for the July 2010 observations, the Ref 2/Ref 3 light curve is quite stable, and shows no anomalous features due to either variability in the references or the Earth s atmosphere. The scatter in the light curve is also smaller than that measured for GJ 1214 (outside of the transit event), so we conclude that any effects from Earth s atmosphere were sufficiently removed in the first transit we observed. In the August 2010 observations, we find that the Ref 2/Ref 3 light curve in the bluer bandpass increases slightly around the time of mid-transit, though the light curve is fairly stable in the redder bandpass for the duration of the observations. Assuming the flux from Ref 2 and Ref 3 was stable, then this feature could be a result of variability in Earth s atmosphere at that time. Indeed, this could be the source of the smaller apparent planet-star radius ratio measured in the bluer bandpass for the case where GJ 1214 was normalized to only Ref 3 (Figures 4-3 and 4-7). But, considering that GJ 1214 and Ref 3 were observed at more similar wavelengths than GJ 1214 and Ref 2 were, it would seem more likely that the light curve for GJ 1214 normalized to Ref 2 would have had an anomalous radius ratio. We suggest that instead, the feature that we find in the Ref 2/Ref 3 light curve is actually due to variability in one of the reference stars and not the Earth s atmosphere. Furthermore, we believe that the feature is due to Ref 2, because we find that the Ref 2/Ref 3 light curve for the third observation (June 2011) also showed an anomalous feature in the bluer bandpass that looked very much like a flare. Given that the flux ratios increased sharply 121

122 (around the end of the transit egress), we conclude that the flux of Ref 2 increased relative to Ref 3, and therefore Ref 2 is the active reference star in both the August 2010 and June 2011 observations. However, it is interesting that the features seen in the Ref 2/Ref 3 light curves occur only in the bluer bandpass, and the light curves in the redder bandpass were stable. Even if such variability in the reference stars light curve is due to Earth s atmosphere rather than the reference stars themselves, we do not believe that we can disentangle whether the Earth s atmosphere or one of the references has a greater effect on our measurements of the planet-star radius ratio. 4.6 Conclusions In summary, we find that we can reach very-high precisions with the GTC/OSIRIS tunable filter, even for a star as faint and active as GJ Furthermore, the precisions that we reach are suitable for atmospheric studies. In this case, we attempted to search for methane in GJ 1214b s atmosphere by measuring the radius ratios in two narrow bandpasses around the location of a predicted methane feature. After acquiring observations of three transits of GJ 1214b and finding that the observations do not yield consistent results regarding the presence of absorption due to methane, we conclude that variability has a significant effect on such studies, regardless of whether the variability comes from variable absorption in an exoplanet s atmosphere, stellar activity, or variable absorption in Earth s atmosphere. 122

123 Figure 4-1. Light curves for GJ 1214 from baseline observations acquired in August In each panel, the blue and red points represent observations of GJ 1214 at nm and nm. Recall that for these observations, no reference star was observed at the same exact wavelength as GJ The horizontal dotted line in each panel illustrates a baseline flux ratio of 1.0. The different panels show the light curves for GJ 1214 when normalized to different combinations of reference stars and are labeled according to which references were used to compute the light curve shown in each specific panel. Note that these light curves were corrected for trends in airmass, the shape of the target s profile, and the centroid coordinates of the target, but no other corrections were made so that we could ascertain the effects of different references on the target. 123

124 Figure 4-2. Light curves and residuals from observations of the 2010 July 22 transit of GJ 1214b. In each panel, the blue and red points represent observations of GJ 1214 at nm and nm. Again, recall that for these observations, no reference star was observed at the same exact wavelength as GJ The blue and red solid curves are the corresponding best-fit models. The horizontal dotted line in each panel indicates the level at which the residuals were offset (for clarity). As in Figure 4-1, the different panels show the light curves for GJ 1214 when normalized to different combinations of reference stars and are labeled according to which references were used to compute the light curve shown in each specific panel. 124

125 Figure 4-3. Same as Figure 4-2, but for observations of the 2010 August 28 transit of GJ 1214b. For this transit, the blue and red points represent observations of GJ 1214 at nm and nm. 125

126 Figure 4-4. Same as Figure 4-2, but for observations of the 2011 June 11 transit of GJ 1214b. As for the observations presented in Figure 4-3, the blue and red points represent observations of GJ 1214 at nm and nm. Also, for this transit, Ref 3 was the only reference observed at the same wavelength as GJ

127 Figure 4-5. Similar to Figures 4-2, 4-3, and 4-4, but here we show the light curves and the corresponding best-fit models generated from combining all three transit observations from July 2010, August 2010, and June

128 Figure 4-6. Best-fit planet-star radius ratios as determined from models fit to the 2010 July 22 transit light curves (Figure 4-2). In each panel, the horizontal error bars illustrate the width of the filters used, and the vertical error bars are the formal 3σ errors as determined from the models (see text for further details). Also shown in each panel is a horizontal dashed line, which indicates the radius ratio reported by Berta et al. (2011), and which we consider to be a representative radius ratio for most values that have been reported in the literature. The corresponding horizontal dotted lines illustrate the uncertainty in their radius ratio. As in previous figures, each panel illustrates results from different light curves computed based on different combinations of reference stars. 128

129 Figure 4-7. Same as Figure 4-6, but for radius ratios as determined from models fit to the 2010 August 28 transit light curves (Figure 4-3). 129

130 Figure 4-8. Same as Figure 4-6, but for radius ratios as determined from models fit to the 2011 June 11 transit light curves (Figure 4-4). 130

131 Figure 4-9. Similar to Figures 4-6, 4-7, and 4-8, but for radius ratios as determined from models fit to all three transit observations (Figure 4-5). 131

132 CHAPTER 5 VETTING KEPLER PLANET CANDIDATES WITH MULTICOLOR PHOTOMETRY FROM THE GTC: IDENTIFICATION OF AN ECLIPSING BINARY STAR NEAR KOI 565 At present, there are over 180 confirmed transiting planets, but only 10% are estimated to be Neptune-size or smaller. 1 The Kepler space mission, which launched in 2009, is responsible for the discovery of a majority of the known transiting Neptuneand super-earth-size planets. Some of the small planets Kepler has discovered to date include Kepler-4b (a Neptune-size planet; Borucki et al., 2010b), Kepler-9d (a super-earth-size planet in a system with two Saturn-size planets; Holman et al., 2010; Torres et al., 2011), Kepler-10b (a rocky planet; Batalha et al., 2011), and Kepler 11-b, c, d, e, f, g (six Neptune- to super-earth-size planets; Lissauer et al., 2011a). Further, Kepler recently discovered over 1000 additional small planet candidates (R p < 6R ; Borucki et al., 2011b). While 80-95% of these candidates are expected to be true planets (Borucki et al., 2011b; Morton & Johnson, 2011), identifying which are false positives remains a challenge. The main sources of false positives are background (or, rarely, foreground) eclipsing binaries (EBs) or hierarchical multiple systems. Due to Kepler s large point-spread function ( 6 ), the flux from a star that has an eclipsing stellar or planetary companion can be blended with the flux of Kepler s target star if the two stars are spatially co-aligned with each other. In these cases, it appears that the target star has a transiting companion. High-resolution ground-based imaging (adaptive optics or speckle imaging), spectroscopy, and Kepler s centroid analysis help to eliminate many blends, but can struggle in cases where the blended system is separated by less than 0.1 from the target (e.g., Borucki et al., 2011b). Further, radial velocity (RV) follow-up is very time-consuming for the typical Kepler target, which may be fainter than V 14 (Batalha 1 The Extrasolar Planets Encyclopedia; 132

133 et al., 2010). In the absence of RV measurements, the detection of secondary transits or differences in the depths of individual transits can also be used to help rule out a blended system. Here, we consider an alternative technique that (to the best of our knowledge) was first discussed by Tingley (2004) and first demonstrated observationally by O Donovan et al. (2006a), which is to rule out blends by measuring the transit depth in different bandpasses. This is possible because, as shown by Tingley (2004), the color change during a transit event (i.e., the difference in the transit depth measured at different wavelengths) increases as the color between the different components of a blend increases. Therefore, observations acquired in multiple bandpasses can be used to reject a planet candidate if the measured transit depths in different bandpasses differ significantly, which could indicate, for example, a blend with a stellar EB of a different spectral type from the target star (Tingley, 2004; O Donovan et al., 2006a; Torres et al., 2011). We note that the COROT space telescope even has a prism built in for the purpose of vetting COROT planet candidates with multicolor photometry, further demonstrating the value of such a technique (Auvergne et al., 2009; Deeg et al., 2009). Here, we describe multiwavelength observations acquired with the Optical System for Imaging and low Resolution Integrated Spectroscopy (OSIRIS) installed on the 10.4 m Gran Telescopio Canarias (GTC) that we used to determine the true nature of (Kepler Object of Interest) KOI KOI (Kepler magnitude of 14.3) was presented by Borucki et al. (2011a) as a super-earth-size planet candidate, with an estimated planet radius of 1.6 R, orbiting a R star with a period of 2.34 days. However, the recent article by Borucki et al. (2011b) lists KOI as most likely being a false positive, as Kepler measured a centroid shift of 8 to the north of KOI 565, indicating that the star that is actually dimming is located on a pixel that is offset from the position of KOI 565. Given that we did not have this information at the time of the observations 2 Also known as KIC in the Kepler Input Catalog. 133

134 presented here, we operate under the assumption that we did not know whether KOI was a true planet or false positive. In this case, it was possible for us to resolve the true source of the transit signal that contaminated Kepler s photometry of KOI 565. We also present measurements of the color of KOI 565 and several nearby stars during the predicted transit event, which independently confirm that KOI 565 is in fact not the true host of the transit signal. More importantly, we show that the false positive would have been identified even if the separation between the stars was too small to either spatially resolve them or allow for the measurement of a centroid shift. Our approach offers an efficient false-positive identification method that is highly complementary to the multi-color follow-up photometry that is currently being conducted with Spitzer at infrared wavelengths (e.g., Fressin et al., 2011) as well as to other ground-based follow-up techniques. We describe our observations in 5.1 and the data reduction and light curve (LC) analysis in 5.2. In 5.3 we present our results and demonstrate that color photometry from the GTC can be used to help identify false positives from transit surveys. Finally, in 5.4, we conclude with a summary of our results and a discussion of our plans for future observations of additional Kepler planet candidates with the GTC. 5.1 Observations We observed the target and several nearby stars around the predicted time of the transit event on 2010 September 19 using the OSIRIS tunable filter (TF) imager installed on the 10.4 m GTC. With the TF imager, custom bandpasses with a central wavelength between nm and a FWHM of nm can be specified. In this observing mode, the effective wavelength decreases radially outward from the optical center, so we positioned the target and a primary reference star (i.e., most comparable in brightness with the target) at the same distance from the optical center so that both stars would be observed at the same wavelengths. Several secondary reference stars were also observed, but they were all at different distances from the optical center 134

135 and thus were observed at slightly different wavelengths. During the observations, we alternated between two bandpasses centred on and nm (at the location of the target on the CCD chip) and with FWHM of 2.0-nm. These bandpasses were specifically chosen as they minimize effects of telluric absorption and emission and yield extremely high differential precisions, as demonstrated by Colón et al. (2010). We used 1 1 binning, a fast pixel readout rate of 500 khz, and read out a single window (containing the target and several reference stars) located on one CCD chip of pixels (equivalent to 3 6 or 56% of the CCD chip) in order to decrease the dead time between exposures. Due to the faintness of the target (V 14.3) and the narrow bandpasses used, the exposure time (for both filter settings) was set to 180 s, with each exposure followed by approximately 21 s of dead time. The observations began at 21:52 UT on 2010 September 19 (during bright time) and ended the following morning at 01:55 UT. There were thin cirrus clouds around the time of observations. The air mass ranged from 1.07 to The actual seeing was better than 1.0, but the telescope was intentionally defocused to reduce pixel-to-pixel sensitivity variations, so the defocused FWHM of the target varied between ( pixels). The telescope s guiding system kept the images aligned within a few pixels during the observations, with the target s centroid coordinates shifting by less than 2 pixels in either direction. The predicted midtransit time based on the ephemeris and orbital period from Borucki et al. (2011a) was 23:59 UT ( BJD) on 2010 September 19. However, Borucki et al. (2011b) presented an updated ephemeris and orbital period, so the transit event we observed occurred 135 min later than initially predicted; because of this, our observations ended before midtransit. It should be noted that the updated ephemeris from Borucki et al. (2011b) is still consistent with the uncertainty in the original ephemeris from Borucki et al. (2011a). Furthermore, the ephemerides given in Borucki et al. (2011a) were based on 43 days of observations, while Borucki et al. 135

136 (2011b) cited ephemerides based on a much longer time baseline, thus allowing for significantly more precise constraints on the ephemerides. 5.2 Data Reduction and Analysis Standard IRAF procedures for bias subtraction and flat-field correction were used. In total, 95 dome flats were taken for each filter setting. We note that the dome lights do not produce a uniform illumination, so we added an illumination correction to the final flat-field image. Due to the narrow filters used and position-dependent wavelength, all images contain sky (OH) emission rings. Therefore, we performed sky subtraction on all images using the IRAF package TFred, 3 which measures the sky background while including the rings due to sky emission. We then performed aperture photometry on the target and several reference stars using standard IDL routines. We tested several different size apertures and based our final choice of aperture on that which yielded the smallest scatter in the relative flux ratios (i.e., the target flux divided by the total reference flux) outside of transit. The final aperture radius used in our analysis is 23 pixels ( 2.9 ). No sky annulus was needed, due to the use of TFred, which automatically removes the sky background. These procedures were performed for each filter separately, but we considered the results for each bandpass and used the same aperture for each filter. We discarded three-four images taken in each bandpass, due to errors in the reduction process that prohibited us from performing aperture photometry on these images. LCs for each bandpass were computed for KOI 565 by dividing the flux measured within the target aperture by the total weighted flux of an ensemble of reference stars. We used six reference stars in total to compute the reference ensemble flux. Even though these reference stars were located at different distances from the optical center, 3 Written by D. H. Jones for the Taurus Tunable Filter, previously installed on the Anglo-Australian Telescope; reduc.html 136

137 we found that using an ensemble of reference stars, rather than just a single reference star, greatly improved the signal-to-noise ratio (S/N) of our observations. Each LC was then normalized to the mean baseline (out-of-transit) flux ratio as measured in each bandpass. We corrected each LC against changes in the air mass, and in order to account for any additional systematics in the LCs, we performed external parameter decorrelation (e.g., Bakos et al., 2007, 2010) against each of the following parameters: the X and Y centroid coordinates of the target on the image frames and the sharpness of the target s profile [equivalent to (2.35/FWHM) 2 ]. The photometric uncertainties in the relative flux ratios include the photon noise of the target and the reference ensemble, the noise in the sky background around the target and references, and scintillation noise. We calculated the median photometric uncertainties to be mmag for both the and nm LCs, where the photon noise of the target dominates the errors. We also investigated the possibility of red noise within our data by computing the standard deviation of the flux ratios after averaging the data over several different bin sizes, and we found that our data follow the trend expected for white Gaussian noise. 5.3 Results We present the results of our photometry in Table 5-1 and the corresponding LCs for KOI 565 in Figure Based on the parameters of the candidate planet transiting KOI 565 given by Borucki et al. (2011a), we expected to measure a transit depth of approximately 182 ppm (parts per million). Due to the somewhat low S/N of our observations (the result of using very narrow bandpasses to observe a faint target while maintaining a reasonable exposure time), our photometric precisions (0.978 mmag) 4 We note the observation times given are the Barycentric Julian Dates in Barycentric Dynamical Time (BJD TDB), computed from the Julian Dates using the calculator found at (Eastman et al., 2010). The ephemerides given by Kepler are in the same time coordinate system. 137

138 were insufficient to detect such a shallow transit. However, we did not need very high precisions to determine that another star near the target was the true source of the dimming that Kepler observed in KOI 565. We followed similar procedures as described in 5.2 to compute the LCs of several stars within 20 of the location of the target that might have contaminated Kepler photometry of KOI 565, and we visually inspected their LCs to see if any showed a transit-like signal around the predicted time of the transit. For reference, we present the field of view around KOI 565 in Figure 5-2. From our analysis of stars near KOI 565, we determined that a star (KIC ) approximately 15 to the north of KOI 565 is the true source of the transit signal, as we observed a significant decrease (> 15%) in the brightness of that star at the time of the predicted transit event. We present the LCs for KIC in Figure 5-3, and the photometry is also given in Table 5-2. Although we were not able to observe a complete LC, based on both the minimum depth and shape of the LC, we deduce that KIC is a stellar EB. Thus, our identification of KIC as a stellar EB that contaminated Kepler s photometry of KOI 565 is consistent with the magnitude and direction of the centroid shift, as well as the eclipse ephemeris from the Kepler data (Borucki et al., 2011b). The EB therefore has an eclipse ephemeris of BJD and orbital period of days (as determined by Borucki et al., 2011b). Based on the stars colors and relative brightnesses, we infer that KIC is a background (rather than foreground) EB. Next, we consider the colors (790.2 nm nm) of the stars during the transit event. In Figure 5-4, we present the colors of KOI 565 and KIC We also consider the color for an unresolved system, simulating a scenario in which the target star and EB are physically associated and thus could not be spatially resolved so that all their light is combined. For this case, we combine the flux from KOI 565 with the flux from KIC in each bandpass and then compute the color from the LCs of the 138

139 unresolved system. 5 Letting λ 1 be the apparent magnitude in the nm bandpass and letting λ 2 be the magnitude in the nm bandpass, we calculated the color indices as λ 1 λ 2 = 2.5 log F λ 1 F λ2, (5 1) where we have taken the average of each pair of flux ratios in the nm LC (F λ1 ) and divided by the corresponding points in the nm LC (F λ2 ). As illustrated in Figure 5-4, we do not measure an appreciable change in the color of KOI 565 during the transit event, but the color of KIC shows a significant color change. To compare the color changes for each star directly, we calculate the weighted mean colors and their uncertainties for the interval before the transit (i.e., the interval to the left of the leftmost dashed line in Figure 5-4) and compare those to the same values measured during the partial transit event. For KOI 565, we calculate the difference in the mean colors to be 6.64 ± , which is consistent with there being no difference in the colors at a level of 1σ. On the contrary, the mean colors of KIC differ at a confidence level of 8.3σ, with a difference of 77.9 ± In the case of the hypothetical unresolved system, where we imagine the target and EB to be physically bound so the projected separation between the two stars is undetectable and all their light is added together, we would still measure an appreciable change in the color, with a difference of 47.9 ± at a significance of 3.8σ. The fact that we measure this large of a color change over such a narrow wavelength regime ( 4 nm) during the transit event clearly indicates a nonplanetary source of the color change, i.e., a stellar EB composed of two stars with very different temperatures. For comparison, Colón et al. (2010) found no appreciable difference between the in-transit and out-of-transit colors as measured 5 Based on the separation of the target and the EB, it is most likely that only a portion of the light from the EB was blended with the light from KOI 565 (Figure 5-2). 139

140 from the same bandpasses used here for either TrES-2 or TrES-3, both of which host Jupiter-size planets. 5.4 Discussion In this article we have presented observations of a Kepler planet candidate acquired in two very narrow bandpasses. From our observations, we identified a nearby stellar EB in eclipse at the predicted time of the transit of the Kepler candidate. We also used our observations to measure the change in the color of a hypothetical unresolved source (composed of the Kepler target and the stellar EB) during the transit event. The identification of a nearby stellar EB and the measured color change during the transit event separately identify the Kepler candidate as a false positive, thus confirming the findings of Borucki et al. (2011b). Based on the LCs of the resolved stars, we deduce that some of the light from a background stellar EB (KIC ) contaminated the photometry of KOI 565 to mimic the transit of a super-earth-size planet around KOI 565. The technique we describe in this article is complementary to other follow-up observations of transiting planet candidates that are currently being conducted. For example, Spitzer is also being used for follow-up of Kepler planet candidates with a wide infrared bandpass (e.g., Fressin et al., 2011), and the COROT space telescope has a prism built in for the purpose of detecting changes in the color during transits of COROT planet candidates via three wide optical channels (Auvergne et al., 2009; Deeg et al., 2009). However, while Spitzer and COROT will go offline in the near future, our approach of acquiring ground-based transit photometry nearly simultaneously in narrow optical bandpasses can be used indefinitely. When compared with multicolor photometry acquired with other ground-based telescopes, our technique has the advantage of being able to acquire multicolor photometry in a single transit observation. We note that some other ground-based instruments are capable of similar observations, e.g., the Simultaneous Quad IR Imager (SQIID; Ellis et al., 1993) at Kitt Peak National Observatory and ULTRACAM (Dhillon et al., 2007) at the William Herschel Telescope. 140

141 However, the GTC/OSIRIS has a unique combination of a large field of view, a superior collecting area, and a wide selection of filters; combined, these allow very efficient high-precision multicolor photometry of faint Kepler targets. Multicolor photometry with the GTC is thus a useful tool for identifying false positives in transit surveys, since the magnitude of the color change during transit can be used to identify not only background (or foreground) EB stars, but also physical triple star systems, which are difficult to reject, as they dilute the transit depth and result in a negligible centroid shift. In the case of physical triples that are difficult to resolve spatially (e.g., with separations of less than 0.1 ), high-precision multicolor photometry can be useful, as a measurable color change during transit could indicate a blend with a stellar EB of a different spectral type from that of the target star (Tingley, 2004; O Donovan et al., 2006a; Torres et al., 2011). Morton & Johnson (2011) predict a slightly higher false-positive rate for Kepler planet candidates, due to physical triples versus background EBs. This is in part because physical triples can often mimic the transit depth of Neptune-size planet candidates (Morton & Johnson, 2011), and Neptune-size planet candidates dominate the candidates discovered by Kepler (Borucki et al., 2011b). Thus, high-precision multicolor photometry may be particularly useful for rejecting false positives within the class of Neptune-size planet candidates. A blend with a star that is hosting a transiting Jupiter- or Neptune-size planet will be more difficult to reject with multicolor photometry, as the magnitude of the color change during transit will be much smaller than for a blend with a stellar EB. The most difficult scenario to reject with multicolor photometry is a hierarchical triple system, where a physically associated companion star has a planetary companion. Our measurements presented in this article would not have been sensitive to a blend with a star hosting a Jupiter-size planet. However, we emphasize that our results were based on observations in two narrow bandpasses with central wavelengths that differed by only 4 nm. Future observations similar to those here could be conducted using OSIRIS s broadband filters. 141

142 These broadband filters are much wider than those allowed by the TF imaging mode, but are still narrower than typical Sloan filters by a factor of 2-4, so they still reduce effects of differential extinction and variable atmospheric absorption. The advantage of using slightly wider filters is to ensure that high S/N are achieved even for fainter targets (V 14-15) while maintaining a reasonable exposure time (less than a few minutes). Further, a larger wavelength regime can be covered by observing in, for example, a bluer filter (e.g., 666 nm) and a redder filter (e.g., 858 nm), which can enhance the change in the color during a transit event. Observing in a red filter, in particular, will also help reduce stellar limb-darkening (compared with the broad optical bandpass used by Kepler; Colón & Ford, 2009), so that for candidates determined to be true planets, measuring the transit depth in a red bandpass will improve estimates of the planet radius and density (and thus bulk composition, assuming a certain mass range for the candidate planet; Valencia et al., 2007). While TF imaging is particularly well suited for high-precision transit photometry of brighter targets (e.g., Colón et al., 2012, 2010), it is not ideal for fainter stars, due to the longer exposure times required to get a high S/N. The use of OSIRIS s broader filters, which cover a larger wavelength regime, is thus one possible way to boost the S/N to reject blends with background, foreground, or physically associated stars hosting transiting planets. With transit surveys like Kepler and COROT actively searching for and finding new planet candidates, it will be vital to use all the tools at hand to reject false positives and determine the true nature of the candidate planets. These observations demonstrate that multicolor photometry from the GTC is one additional tool that can help with the identification of false positives in the coming years. 142

143 Table 5-1. Normalized photometry of KOI 565. BJD Flux ratio Uncertainty λ = nm

144 Table 5-1. Continued BJD Flux ratio Uncertainty λ = nm a The time stamps included here are the Barycentric Julian Dates in Barycentric Dynamical Time (BJD TDB) at mid-exposure. The flux ratios included here are those that have been corrected using external parameter decorrelation and normalized to the baseline (out-of-transit) flux ratios ( 5.2). 144

145 Table 5-2. Normalized photometry of KIC BJD Flux ratio Uncertainty λ = nm

146 a The columns are the same as in Table 5-1. Table 5-2. Continued BJD Flux ratio Uncertainty λ = nm

147 Figure 5-1. Normalized LCs for nearly simultaneous observations at 790.2±2.0 nm (circles) and 794.3±2.0 nm (squares) of KOI 565 as observed on 2010 September 19. The nm LC has been offset for clarity. The vertical dashed lines indicate (from left to right) the predicted beginning of ingress and mid-transit time (based on Borucki et al., 2011b). See the electronic edition of the PASP for a color version of this figure. 147

148 Figure 5-2. Image from GTC/OSIRIS observations at nm containing part of the field of view around KOI 565. This is only a small portion of the observed field of view, so the six reference stars used in our analysis are not shown here. The target is located at α = 19 h 17 m s, δ = , and KIC is located at α = 19 h 17 m s, δ = The small circles around KOI 565 and KIC indicate the size of the apertures used in our photometry (r 23 pixels 2.9 ). The larger circle around KOI 565 has a radius of 8 and is included simply to illustrate which stars are located within a distance 8 from the target, as Borucki et al. (2011b) measured a centroid shift of 8 to the north of the target. The position of the star causing the centroid shift is typically at a slightly further distance, so the position of KIC relative to KOI 565 is consistent with measurements from Kepler. See the electronic edition of the PASP for a color version of this figure. 148

149 Figure 5-3. Similar to Figure 5-1, but for KIC While only a partial eclipse was observed, the minimum depth of the LC is comparable with what is expected for a stellar EB. See the electronic edition of the PASP for a color version of this figure. 149

150 Figure 5-4. Colors as computed between the and nm observations of (a) KOI 565, (b) a stellar EB (KIC ), and (c) for an unresolved system (the target light combined with the light from the EB). In each panel, the vertical dashed lines indicate (from left to right) the approximate beginning of ingress and the mid-transit time (based on Borucki et al., 2011b). The vertical scale is the same for each panel for ease of comparison. There is no change in the color seen for KOI 565, but for the EB as well as the hypothetical unresolved system, we measure an appreciable difference in the color during the transit event. 150

151 CHAPTER 6 CONSTRAINING THE FALSE POSITIVE RATE FOR KEPLER PLANET CANDIDATES WITH MULTI-COLOR PHOTOMETRY FROM THE GTC The Kepler space mission has discovered, to date, 61 transiting planets as well as over 2,000 planet candidates and 2,000 eclipsing binary stars (Batalha et al., 2012; Prša et al., 2011; Slawson et al., 2011). 1 With such a vast number of planet candidates, it can be difficult to decide which to focus follow-up efforts on. Recent studies have tried to address this issue by estimating the false positive rate for the Kepler sample. Based on the list of 1,235 Kepler planet candidates published by Borucki et al. (2011b), it has been predicted that as many as 95% of these candidates are true planets (Morton & Johnson, 2011). However, previous studies have not taken into account how different subsets of Kepler targets may have different false positive rates. For example, there is a rapid rise in the number of detached eclipsing binary stars that have been discovered by Kepler at orbital periods of less than 3 days, and such systems can mimic planetary transits (Prša et al., 2011; Slawson et al., 2011). This suggests that there may be corresponding changes in the false positive rate with orbital period. Because the probability of observing a transit event increases as the orbital period of the planet decreases, many of Kepler s planet candidates have short periods. Thus it is necessary to be cautious when estimating false positive rates over the whole Kepler sample. While observational studies with warm-spitzer support predictions of low false positive rates over the entire Kepler sample (Désert et al., 2012), biases in target selection can affect observationally-constrained false positive rates ( 6.6). We also note that a recent imaging study has found that nearly 42% of their sample of 98 Kepler planet candidate hosts has a visual or bound companion within 6 arcseconds of the target star (Lillo-Box 1 Up to date catalogs can be found at 151

152 et al., in preparation). Such studies emphasize the need for follow-up imaging to exclude blend scenarios imitating planet candidates or contaminated transit depths. The false positive scenarios we consider in this paper are those that result from stellar eclipsing binaries that are either in the background (or, in rare cases, foreground) or bound to the target star and are not always easy to identify with Kepler due to the flux from the different stars being blended together within Kepler s photometric aperture ( 6 arcsec). As discussed by, e.g., Colón & Ford (2011), different techniques can be used to eliminate many, but not all, blends. Multi-color transit photometry is an efficient method for recognizing blends that cannot be spatially resolved, as measuring the transit depth in different bandpasses (i.e. the transit color) allows one to test the planet hypothesis. This is possible since the magnitude of the transit color changes as long as the blended stars have significantly different colors. Colón & Ford (2011) presented multi-color transit photometry of a Kepler target, KOI , that was first announced by Borucki et al. (2011a) to be a super-earth-size planet candidate but later recognized as a likely false positive due to measurements of a centroid shift away from the location of the target on the CCD during transit (Borucki et al., 2011b). In Colón & Ford (2011), we used near-simultaneous multi-color observations acquired using the narrow-band tunable filter imaging mode on the Optical System for Imaging and low Resolution Integrated Spectroscopy (OSIRIS) installed on the 10.4-m Gran Telescopio Canarias (GTC) to confirm that KOI is indeed a false positive, as we both resolved a stellar eclipsing binary 15 arcsec from the target and measured a color change in the unresolved target+eclipsing binary system. Thus, Colón & Ford (2011) demonstrated the capability of the GTC/OSIRIS for efficient vetting of planet candidates via its capabilities for near-simultaneous multi-color photometry within a single transit event. In this paper, we present observations of four Kepler planet candidates specifically selected to have small radii and short orbital periods (R < p 5 R and P < 6 days). 152

153 Measuring the false positive rate for this extreme subset of planet candidates will allow us to test if there is a correlation between the false positive rate and different planetary and stellar properties. As in Colón & Ford (2011), the observations presented here were acquired with the GTC/OSIRIS. However, these observations used broadband filters in lieu of the narrow-band tunable filters in order to collect more photons and to obtain greater wavelength coverage (and thereby probe greater color differences). In 6.1 we discuss our target selection criteria, and we discuss the corresponding observations for our four targets in 6.2. In 6.3 and 6.4 we describe our data reduction and light curve analysis procedures, and we present results for each target in 6.5. We include a discussion of our results and how they relate to the distribution of eclipsing binaries that have been discovered by Kepler as well as previous estimates of the false positive rate for the Kepler sample in 6.6. In particular, we discuss theoretical estimates from Morton & Johnson (2011) and observational constraints from studies by Désert et al. (2012) and Santerne et al. (2012). Finally, we summarize our results and conclusions in 6.7, and we also discuss our plans for future observations of additional Kepler targets with the GTC. 6.1 Target Selection Recent studies (e.g., Lissauer et al., 2011b, 2012) have demonstrated that a majority of the planet candidates in Kepler s multi-planet candidate systems should in fact be real planets. Considering these studies, for our program we target only those candidates found in single systems as determined by Borucki et al. (2011b). From the sample of candidates in single systems, we selected targets based on the following criteria: orbital period (P) < 6 days planet radius (R p ) < 6 R transit depth (at center of transit; δ) > 500 ppm transit duration (first to fourth contact; τ) < 2.5 h 153

154 Kepler magnitude (Kp) < 15.5 vetting flag > 1 As discussed in the previous section, there is a significant rise in the number of detached eclipsing binary stars compared to planet candidates at short orbital periods. This is also illustrated in Figure 6-1, where we show histograms of the number of planet candidates, detached eclipsing binaries, and all other eclipsing binaries (i.e. all binaries that are not listed as detached in Slawson et al., 2011) as a function of orbital period. Therefore, we focus on constraining the false positive rate only for planet candidates with short orbital periods, where the presence of eclipsing binaries is greatest. Our decision to focus on small candidates is due to there being a dominant population of Neptune-size or smaller candidates in the Kepler sample. We set constraints on the transit depth, transit duration and Kepler magnitude due to limitations of our observing technique, so as to measure the transit depth and color precisely. Furthermore, we want to do this while maintaining a reasonable observing cadence, hence the limits on the Kepler magnitude of our targets. Finally, we exclude candidates that have vetting flags of 1, as those have been previously confirmed as planets. Along with the above constraints on the planetary and stellar properties, we have several observational constraints due to limitations of the GTC/OSIRIS. First, we rule out candidates that have no observable transits from the location of the GTC (which is located on La Palma at the Observatorio del Roque de los Muchachos). Then, we exclude those that did not have multiple transits observable during bright or grey time (during the 2011 observing season). We also exclude transit events that occur outside altitudes of degrees, so as to avoid high airmass and vignetting that occurs at high altitudes due to limitations of the GTC dome. 154

155 Finally, we did not check the Kepler light curves 2 for each target that fit all of the above criteria prior to acquiring observations. However, as we discuss in 6.7, for future observations we plan to check the Kepler light curves for secondary eclipses and V-shaped eclipses, either of which could indicate a priori that the target is not a transiting planet but instead a stellar eclipsing binary. These selection criteria led us to acquire observations of four planet candidates between April and September 2011: KOI (Kepler Object of Interest) , , and In Table 6-1 we list some of the properties of each of our targets as determined by Borucki et al. (2011b). Borucki et al. (2011b) flagged KOI as possibly having ellipsoidal variations. KOI has the shortest period of all the planet candidates announced by Borucki et al. (2011b). In Figures 6-2, 6-3 and 6-4 we illustrate different properties of our targets in comparison to the sample of 1,235 KOIs and the corresponding 997 host stars from which our targets were chosen. The sample of KOIs observed with warm-spitzer (Désert et al., 2012) is also illustrated in these figures for comparison (we discuss the Spitzer sample in more detail in 6.6.1). Our targets and the corresponding observations are described in detail in 6.2 below. 6.2 Observations We acquired photometry of each target and several nearby reference stars around the predicted time of a transit event. For each observation, we used the GTC/OSIRIS to acquire near-simultaneous multi-color photometry by alternating between two broadband order sorter filters 3 that were custom made for OSIRIS: 666±36 nm and 858±58 nm. Both bandpasses were specifically chosen so as to minimize effects of telluric absorption and emission while also allowing for ample wavelength coverage. In order to decrease dead time, all observations used 1 1 binning, a fast pixel readout 2 Available at

156 rate of 500 khz and a single window located on one CCD chip. The size of the window varied for each observation, but each window was large enough to contain the target and several reference stars. We note that all observations except those for KOI 1187 were conducted in queue (service) mode. In the following sections, we describe specific details regarding each target and its respective observations KOI We observed the transit event of KOI on 2011 April 13 under clear conditions and during dark time, with observations beginning at 03:10 UT and ending at 06:12 UT, during which time the airmass ranged from 1.66 to The exposure time was 50 s (for both bandpasses), with 40 s of dead time following each exposure. Seeing varied between 1.2 and 2.0 arcsec until about 05:00 UT, at which point the seeing improved to < 1 arcsec and a slight defocus was introduced in order to avoid saturation. A diffuse, dark band was present in all the images (a result of a very bright star located just outside the CCD), so the target was positioned on the CCD such that it was outside this region and also avoided bad pixels in the center of the CCD chip. During the observations, the target s centroid coordinates shifted by < 9 pixels in the x-direction and 2 pixels in the y-direction. Several images were lost due to technical issues, and a few of the images towards the end of the observations were discarded due to the beginning of twilight KOI Observations of the 2011 September 13 transit of KOI took place from 21:48 UT (2011 September 13) to 01:30 UT (2011 September 14). Observations occurred during bright time and under photometric conditions, and the airmass ranged from 1.07 to The exposure time was set to 40 s, with a corresponding 20 s of dead time following each exposure. The seeing roughly ranged from arcsec throughout the observations, though at the beginning of the observations the seeing conditions changed drastically enough from one image to another that some of the reference stars 156

157 were saturated. We explicitly exclude any saturated reference stars from our analysis ( 6.4). The seeing stabilized and improved after about 23:00 UT, enough so that a slight defocus was implemented to avoid saturation. The target s centroid coordinates shifted by < 4 pixels in either direction during the observations KOI The 2011 September 11 transit of KOI was observed under photometric conditions and during bright time. Observations began at 21:37 UT on 2011 September 11 and ended at 00:00 UT on 2011 September 12. The airmass ranged from 1.05 to 1.22 and the seeing was stable between 0.6 and 0.8 arcsec. A slight defocus was implemented, yielding a defocused FWHM of arcsec. An exposure time of 10 s was used, with 20 s of dead time between exposures. As in observations of KOI ( 6.2.1), a dark band caused by a bright star outside the CCD window was present in all images, so we placed the target star appropriately far away from the band so that the photometry would not be affected. The centroid coordinates of the target shifted by less than 4 and 2 pixels in the x- and y-directions, respectively. Twice during the observations the primary mirror segments lost alignment and produced distorted images: first around 22:30 UT and again at 00:00 UT. We discarded a few images from the first instance, but the issue with the mirror could not be readily fixed during the second instance, and the observations were forced to end early, around the time of the transit egress KOI We observed the 2011 June 12 transit event of KOI , with observations taking place from 22:31 UT (2011 June 11) to 02:01 UT (2011 June 12). Observations took place during bright time and under clear conditions, and the airmass ranged from 1.90 to Seeing was excellent and ranged from 0.6 to 0.9 arcsec during the night. Due to the excellent seeing conditions, the exposure time was set to 5 s. There were

158 s of dead time following each exposure. During the observations, the target s centroid coordinates shifted by < 4 pixels in either direction. 6.3 Data Reduction We used standard IRAF procedures for bias subtraction and flat-field correction (using dome flats taken for each bandpass) for each target. Due to nonuniform illumination by the lamp, we added an illumination correction to the final flat-field image for each target (except KOI 225). The illumination correction was performed using the IRAF task mkillumflat within the noao.imred.ccdred package, which removes the large scale illumination pattern from the flat field by smoothing the flat field image. For KOI 225, the final flat-field for the 666 nm filter showed a strong gradient after performing an illumination correction, so we chose to use the flat-field image as it was. To be consistent, we did not perform an illumination correction on the 858 nm final flat-field for KOI 225 either. We used the IDL Astronomy User s Library 4 implementation of DAOPHOT (Stetson, 1987) to perform aperture photometry on each target and several nearby reference stars as well as on other potential sources of the transit signal (i.e. stars within 20 arcsec of each target). For each target, we tested different apertures and chose a final aperture based on that which resulted in the smallest scatter in the baseline (out-of-transit) flux ratios (i.e., the target star flux divided by the ensemble reference star flux). For KOI 225, 420, 526 and 1187 our final aperture was 25, 23, 15 and 14 pixels (equivalent to approximately 3.2, 2.9, 1.9 and 1.8 arcsec). Sky background subtraction took place during the aperture photometry process, with annuli chosen to be far enough away from each star that the flux from a given star would not be included within the sky annulus. While these procedures were performed for each bandpass separately, we considered the results for each bandpass and used the same aperture and sky annulus for each data set for a given target. Note that in many cases a smaller

159 aperture and sky annulus were used for the stars within 20 arcsec of a given target due to the small separations between some of the stars and the potential for blending. For reference, we present a reduced science image showing the field of view around KOI 1187 in Figure 6-5. Once aperture photometry and sky subtraction were completed, we proceeded with light curve analysis as discussed in 6.4 below. 6.4 Light Curve Analysis For each target, we computed light curves for each bandpass by dividing the total flux measured from the target by the total weighted flux of an ensemble of reference stars, where the flux from each reference star was weighted accordingly after discarding outlying flux values (resulting from either technical issues or variable sky conditions). During the analysis, several reference stars were found to be obviously variable or saturated, so we excluded those stars from the analysis. This resulted in us using 10, 4, 2 and 9 reference stars in the field near KOI 225, 420, 526 and 1187 to compute the reference ensemble flux. Next, each light curve for each target was normalized to the mean baseline flux ratio as measured in each bandpass. The light curves were then corrected for changes in airmass as well as for drifts in the centroid coordinates of the target on the CCD and the sharpness of the target s profile [(2.35/FWHM) 2 ]. The latter corrections were done via external parameter decorrelation (e.g., Bakos et al., 2007, 2010). We note that due to the lack of data for KOI 526 during the transit egress and post-transit, our attempts to correct those light curves produced skewed results. Therefore, all further analysis for KOI 526 is based on the uncorrected light curves. The corresponding observation times for each light curve for each target were computed from the UTC timestamps given in the image headers. We converted the UTC time at mid-exposure to Barycentric Julian Dates in Barycentric Dynamical Time (BJD TDB) via an online calculator described in Eastman et al. (2010) in order to match 159

160 the time coordinate system that the ephemerides from Borucki et al. (2011b) are given in. 5 The photometric uncertainties were computed from the photon noise of the target star and the reference star ensemble, the noise in the sky background around the target and each reference used in the ensemble, and scintillation noise. The resulting median photometric uncertainties for each light curve for each target are given in Table 6-2. In each case, the photon noise of the target is the primary source of error. Finally, in order to investigate the presence of red noise in our data, we computed the standard deviation of the flux ratios for different bin sizes and found that all our data follow the trend expected for white (Gaussian) noise. Despite this result, we also investigate potential residual systematics via a prayer-bead analysis, which we discuss below. After correcting the light curves for each target (except for KOI 526 as discussed above), we proceeded to fit synthetic models to each light curve, following the approach taken by Colón et al. (2010). We assumed each observed transit event was a planetary transit, and we used the planetary transit light curve models from Mandel & Agol (2002) to fit limb-darkened models to our data. Specifically, for each target, we fit for the following parameters: time of mid-transit (t 0 ) transit duration (first to fourth contact; τ) impact parameter (a cos i/r ) planet-star radius ratio (R p /R ) two limb darkening coefficients (c 1 and c 2 ) baseline flux ratio (linear) baseline slope 5 The calculator is available at 160

161 The limb darkening coefficients that we fit for are defined as c 1 u 1 +u 2 and c 2 u 1 u 2, where u 1 and u 2 are linear and quadratic limb darkening coefficients. For KOI 526, we note that we held the transit duration fixed to the value from Borucki et al. (2011b) in order to fit a model to our partial transit. Initial guesses for the mid-transit time, transit duration, impact parameter and radius ratio were based on the values and their corresponding uncertainties as given in Borucki et al. (2011b). Values for the limb darkening coefficients were interpolated from the Claret & Bloemen (2011) models for the Sloan r and z filters and are based on the stellar parameters given in Borucki et al. (2011b). Due to the potentially larger than estimated uncertainties in the stellar parameters (e.g., Borucki et al., 2011b), we did not allow the limb darkening coefficients to be completely free parameters in the fitting process. Rather, we kept the coefficients fixed at self-consistent values during the fitting process, but we tested a range of fixed values for the coefficients. Similarly, we tested a range of initial guesses for the other parameters based on the uncertainties for the planetary and stellar parameters. Best-fitting models were identified via a Levenberg-Marquardt minimization scheme. 6 Our light curve fitting procedure for each target is as follows. First, we fit models to each light curve individually, corrected the data against the best-fit baseline flux ratio and slope, subtracted the best-fit models and discarded any data points lying greater than 3σ from the residuals. Then, we used the corrected light curves and fit models to them in a joint analysis, where we forced the impact parameter, transit duration, mid-transit time, baseline flux ratio and baseline slope to be the same for both light curves. However, we allowed different values for the radius ratio and limb darkening coefficients to be fitted to the different light curves. The results from the joint analysis were then used to correct 6 We specifically used mpfitfun, which is publicly available at craigm/idl/idl.html 161

162 the individual light curves as in the first step described above. A final joint analysis was then applied to the final corrected light curves. During this final stage, we also performed a prayer-bead analysis as was done in Colón et al. (2010). Specifically, we performed a circular shift on the residuals for each light curve (computed after removing the best-fit model) and constructed synthetic light curves by adding the shifted residuals back to the best-fit model. The joint analysis described above was applied to each synthetic light curve, and we use the dispersion of the best-fit parameters to calculate uncertainties on each parameter. This accounts for any additional systematic noise sources in the data. We present results from our light curve analysis in Results We present the light curves and the corresponding best-fit models for each target in Figures 6-6, 6-7, 6-8 and 6-9. While not shown here, we had also generated light curves for the potential sources of each transit signal (i.e. stars that could have been blended with the target within Kepler s aperture) following a similar procedure as described above. We generated light curves for 10, 3, 3 and 6 stars within 20 arcsec of KOI 225, 420, 526 and Upon visual inspection, we found that none of these light curves showed a transit signal during the time of the transit event, indicating that the transits that we observed either occur due to an object transiting the target star or an object transiting an unresolved star that is blended with the target star. We also present the transit color (666 nm 858 nm) of each target in the bottom panel of Figures , which was computed by taking the average of each pair of flux ratios in the 666 nm light curve and dividing by the corresponding points in the 858 nm light curve. Therefore, in Figures , a positive color indicates a redder transit and a negative color indicates a bluer transit. In Table 6-3, we report best-fit model parameters for each target based on the model with the smallest χ 2 value, but we note that the parameter uncertainties are based on the full set of best-fit values and their corresponding uncertainties as found 162

163 during the prayer-bead analysis. Specifically, for a given parameter, the upper error bar is a result of subtracting the best-fit value (from the model with the smallest χ 2 value) from the maximum sum of a fitted value and its associated measurement uncertainty as determined from the full set of models computed during the prayer-bead analysis, and likewise for the lower error bar. In general, we find that the formal 1σ errors for the best-fit parameters are comparable to those errors computed based on the results from the prayer-bead analysis, which indicates that any residual systematics in the data (i.e. those not removed by external parameter decorrelation or airmass corrections) have a negligible effect on our results. Furthermore, the distribution of each best-fit light curve parameter over all permutations is smaller than the uncertainties given in Table 6-3, which further indicates that the results presented here are robust. Finally, as suggested by the referee, we consider whether we can use astrometry from our images to provide additional constraints on the properties of the targets we observed. A preliminary astrometric analysis does not suggest that any of the KOIs we observed are due to a blend with another star that was resolved by our observations. Furthermore, despite the large aperture used by Kepler, the astrometric precisions from Kepler are extremely high (< arcseconds for a single 30 min exposure; Monet et al., 2010), and to the best of our knowledge, no significant centroid shifts were measured by Kepler for any of our KOIs. We discuss specific results for each target individually in the following sections KOI The light curves presented in Figure 6-6 are prominently V-shaped, which alone suggests a possible non-planetary transit event. Considering the shape of the light curves and the significant difference in the transit depths as measured in the GTC bandpasses, we deduce that KOI is most likely a stellar eclipsing binary system. Specifically, we find that the best-fit planet-star radius ratios differ at a level of > 11σ, which is a clear indication that this object is not a planet. The transit depth in the Kepler 163

164 bandpass is also significantly different from those measured in the GTC bandpasses. In particular, the transit depth is notably different between the Kepler bandpass and the blue GTC bandpass (around 666 nm), even though they probe fairly similar wavelength regimes. We primarily attribute this difference to dilution in the Kepler photometry, since there is a relatively bright star near KOI 225 that appears to have slightly contaminated one of the pixels used in the Kepler photometry. 7 As noted in Borucki et al. (2011b), this target s light curve had possible ellipsoidal variations, also suggesting that the system contains an eclipsing binary. Furthermore, after our observations had been conducted, Slawson et al. (2011) listed this target in their eclipsing binary catalog. 8 Finally, Ofir & Dreizler (2012) recently conducted an independent analysis of the Kepler data set and identified KOI as a false positive due to the detection of significant differences between the odd and even eclipse events. Since we measure a deeper eclipse in the bluer GTC bandpass, this indicates that during the eclipse more blue light is blocked, and therefore the secondary (eclipsing) component is redder than the primary star. In Table 6-3, we provide an updated ephemeris and eclipse duration, though we note that likely due to the non-planetary transit light curve shape, the best-fit model yielded an impact parameter that reached the upper boundary limits during the fitting process. Therefore, the resulting best-fit light curve parameters are likely somewhat skewed, but this does not change the significance of our results. Slawson et al. (2011) report a period that is twice as long as that found by Borucki et al. (2011b). If correct, then either what Borucki et al. (2011b) believed were transit 7 As determined from target pixel files downloaded from 8 We refer the reader to Slawson et al. (2011) for further details on how some KOIs were rejected as planet candidates and subsequently added to the eclipsing binary catalog. 164

165 events were actually a combination of primary and secondary eclipses or the eclipsing binary has a large enough inclination that all eclipse events are primary eclipses and the secondaries are unobservable from our line-of-sight. If we assume that the period found by Slawson et al. (2011) is correct (and was determined, for instance, by measuring different depths for successive transits, which led to this KOI being rejected), then it is necessary to consider how this affects any correlation between the orbital period and the false positive rate for Kepler targets. We refer the reader to 6.6 for further discussion KOI As illustrated in Figure 6-7, the transit depths as observed in the GTC bandpasses are comparable, and there is not a significant change in the measured color during transit. The planet-star radius ratios measured from the GTC light curves are consistent within 2.8σ, and we did not resolve any potential sources of the transit signal within 20 arcsec of the target, so we deduce that this target is not a false positive and is instead a validated planet. 9 The best-fit parameters from our light curve models are given in Table 6-3 and include an updated ephemeris, transit duration, and impact parameter. We note that the impact parameter given in Borucki et al. (2011b) has an associated uncertainty of 1, and Batalha et al. (2012) report an impact parameter of 0.57 with a relatively large uncertainty of 0.52, so we provide a much stronger constraint on this parameter here. Finally, based on our measured planet-star radius ratios and the revised stellar radius (0.69 R compared to the value of 0.83 R reported by Borucki et al., 2011b) given in Batalha et al. (2012), we find KOI to have a radius that is between 4.15 and 4.87 R (slightly larger than the value of 3.65 R found by Batalha et al., 2012). 9 We use the term validated to mean that the planet candidate is most likely a planet based on our observations but it is not a confirmed planet because there is not independent evidence (e.g., Doppler or transit timing variations) for the planet model. 165

166 6.5.3 KOI Despite only observing a partial transit with the GTC (as illustrated in Figure 6-8), we find that the transit depths clearly match each other, with planet-star radius ratios that are consistent within 1.2σ. While the Kepler depth is much shallower, there appears to be no significant contamination from nearby stars in the Kepler photometry (see footnote in 6.5.1). Given that even the transit depths in the Kepler and the blue GTC bandpass are significantly different, and that dilution appears to not be the source, we believe this object warrants further investigation beyond the scope of this paper. For instance, it is possible that stellar variability in the target star or a blended star impacted both the GTC and Kepler s measured transit depths. 10 Regardless, since the depths between the GTC bandpasses are consistent, this target still passes our validation test. As in the case for KOI , we found no stars within 20 arcsec of the target that showed a transit signal at the expected time of the transit, so we also validate KOI as a planet. Table 6-3 includes an updated ephemeris and impact parameter (recall that the transit duration was held fixed due to fitting only a partial transit). Again, as for KOI , Borucki et al. (2011b) found an uncertainty of 1 on the impact parameter, and Batalha et al. (2012) report an impact parameter of 0.80±0.34, so our observations and models provide a much stronger constraint on the impact parameter, which we find has an associated uncertainty of Based on our measured planet-star radius ratios and assuming a stellar radius of 0.92 R (from Batalha et al., 2012, and slightly larger than the value of 0.80 R reported by Borucki et al., 2011b), we calculate that KOI has a radius between 6.37 and 7.16 R.We note that this is over twice the radius of 3.11 R measured by Batalha et al. (2012). 10 Kepler light curves for this target show baseline variability at a level of 1%. 166

167 6.5.4 KOI As in the case of KOI , the light curves for KOI (shown in Figure 6-9) appear to be fairly V-shaped. Furthermore, visual inspection of the GTC light curves shows that there is clearly a significant difference in the GTC transit depths. However, we find that the best-fit planet-star radius ratios only differ at a level of 1.8σ, likely due to the large uncertainty in the radius ratio measured for the 858 nm light curve. We attribute this large uncertainty to a combination of the relatively poor photometric precisions achieved for this target combined with the degeneracy between the impact parameter (measured to be nearly equal to 1) and the planet-star radius ratio. To reconcile these measurements with what we find visually, we compute weighted mean colors and their uncertainties for the in-transit and out-of-transit data. We find a mean in-transit color of ± and a mean out-of-transit color of ± , which differ significantly at a level of 5.8σ. Therefore, despite the consistent measured radius ratios, there is an obviously significant color change during the transit event, so we argue that KOI is in fact a false positive and not a planet. Contrary to KOI , the transit of KOI is deeper in the redder GTC bandpass than in the bluer GTC bandpass. This implies that during the eclipse, more red light is blocked than blue light, and the smaller (eclipsing) component is bluer than the primary star. In this case, KOI 1187 may consist of an evolved giant star that is redder and several magnitudes brighter than the eclipsing star. Slawson et al. (2011) also list this target in the eclipsing binary catalog as a rejected KOI (which we were also not aware of prior to observing this target). Also, just as for KOI , Slawson et al. (2011) found an orbital period for KOI that is twice as long as that found by Borucki et al. (2011b). As discussed in above and below in 6.6, such results have ramifications on any correlation between the orbital period and false positive rate. 167

168 6.6 Discussion Of the four Kepler planet candidates presented in this paper, we identified two as false positives and provide further evidence supporting the planet hypothesis for two candidates. This suggests a false positive rate that is much higher than has been previously predicted for the Borucki et al. (2011b) KOI catalog (Morton & Johnson 2011; 6.6.1), so we consider what could cause such a large false positive rate for our sample. Referring back to Figure 6-1, we see that our two false positives, KOI and KOI , have shorter orbital periods than the two KOIs we validated as planets. Therefore, our findings suggest that the false positive rate for the Kepler sample varies significantly with orbital period and is largest at the shortest periods (P < 3 days), which is what one would expect a priori due to the rise in the number of detached eclipsing binaries at these short periods (Figure 6-1). We find no significant correlation between the false positive rate and planet radius or other properties of the host star. Figure 6-2 illustrates that there seems to be no trend in the false positive rate with planet radius, as the KOIs we identified as false positives have the smallest and largest apparent radii (as measured by Kepler) in our sample. Similarly, there appears to be no trend with the effective temperature of the host star, although we note that both false positives were also the faintest targets in our sample, as illustrated in Figure 6-4 ( 6.6.1). Finally, we emphasize that based on the stellar parameters from the KIC, all our targets are likely FGK dwarfs, which are the primary targets of the Kepler mission. One factor that must be considered when making such conclusions is that we do not take into account the level of uncertainty in the stellar parameters that are drawn from the KIC. Any uncertainty in the stellar parameters would obviously affect the distribution of the planetary radii and stellar magnitudes and temperatures discussed here. However, if the stellar radii in the KIC have been systematically overor under-estimated, then all planetary radii would simply scale up or down, and our 168

169 conclusions about the lack of a correlation between the false positive rate and planet radius would remain the same. 11 Additionally, there is the issue that if a given planet candidate is actually an eclipsing binary with an orbital period that is twice as long as was initially expected, then this would imply that we are not probing planet candidates with orbital periods of less than 6 days, but instead we are actually probing a sample of eclipsing binaries with periods as long as 12 days. However, since there is still a comparable population of eclipsing binaries and planet candidates with longer periods (e.g. out to at least 10 days, as illustrated in Figure 6-1), we believe this period discrepancy would not affect our conclusion that a population of short-period eclipsing binaries can significantly contaminate short-period planet candidates. Candidates that are initially identified as planets but are actually eclipsing binaries will tend to have shorter apparent periods (their periods will double when they are identified as eclipsing binaries), which strengthens our argument for contamination of the planet sample at short periods, since this effect would further reduce the ratio of planets to eclipsing binaries at periods of less than approximately 3 days. Finally, we note that with the technique presented here, it is not possible for us to identify cases where a planet candidate is actually a binary star composed of two stars that have the same temperature, as there would be no measurable color change during the transit event. This suggests that some of our potentially validated planets could still be false positives. However, Kepler s photometry should be able to distinguish if the transit depths differ between every other transit (i.e. odd-even transit depths), which would identify a candidate as a false positive. In cases where a significant odd-even 11 Some high signal-to-noise ratio eclipsing binaries are more likely to have correctly determined radii, so some short-period planet candidates that are actually eclipsing binaries are correspondingly more likely to have accurate radii. 169

170 ratio is measured, this can also be used to reconcile the issue with the orbital periods described above. To the best of our knowledge, no significant odd-even ratio was found for our validated planets, KOI and KOI Recently, a new list of Kepler planet candidates was announced, bringing the total number of candidates to 2,321 (Batalha et al., 2012). As illustrated in Figures and 6-10, the new catalog follows the same general distribution of orbital periods, planet radii, stellar magnitudes, temperatures, and Galactic latitudes as the Borucki et al. (2011b) catalog, but there is notably a greater number of smaller and shorter-period planet candidates in the new catalog. This serves to emphasize the need to observationally constrain the false positive rate for such small, short-period planets. We note that this new catalog was generated using improved vetting metrics, so it should have a higher fidelity than previous ones (i.e., the Borucki et al. 2011b catalog). However, the KOIs from the Borucki et al. (2011b) catalog have not been vetted against the improved metrics described in Batalha et al. (2012), so both KOI and KOI are still included in the updated catalog as potential planet candidates. Also, Batalha et al. (2012) remark that potential KOIs are not vetted against the shape of their light curves, so planet candidates with V-shaped transit light curves (such as KOI and KOI ) are not immediately rejected. Given ongoing parallel efforts to identify planet candidates and KOIs and that Slawson et al. (2011) previously identified both KOI and KOI as eclipsing binaries, we recommend that the eclipsing binary catalog be consulted before following up any given planet candidate Comparison to Theoretical Studies A recent study by Morton & Johnson (2011) provided theoretical estimates of the false positive rate for Kepler planet candidates. Morton & Johnson (2011) specifically estimated that nearly 90% of the 1,235 candidates presented by Borucki et al. (2011b) had a false positive probability of less than 10%. For our targets in particular, Morton & Johnson (2011) computed the following false positive probabilities: 0.01 (KOI ), 170

171 0.04 (KOI ), 0.03 (KOI ) and 0.03 (KOI ). It is interesting to note that the target with the lowest false positive probability is one that ended up being a false positive. However, we note that the false positive probabilities from Morton & Johnson (2011) are only valid for non V-shaped transit signals. Therefore, considering that both KOI and KOI appear to have somewhat obviously V-shaped transits, the low false positive probabilities computed for these targets by Morton & Johnson (2011) are not appropriate. A recent study by Morton (2012) improves upon the analysis from Morton & Johnson (2011) by taking the transit shape into account when computing the false positive probability for a given target. From their new analysis, Morton (2012) calculate a false positive probability of > 0.99 for KOI and 0.76 for KOI , which is consistent with our observations. The caveat about the transit shape also affects the overall assumed false positive rate for the Kepler sample, as Morton & Johnson (2011) did not separate planet candidates with V-shaped transit signals from those with non-v-shaped transit signals. This means that we should not interpret their findings as 90-95% of Kepler planet candidates are planets. However, we can use their averaged false positive probabilities as a starting point for the overall false positive rate for the Kepler sample. Despite any caveats, we estimate the probability of detecting 2 or 3 false positives out of 4 or 5 targets observed. Assuming a binomial distribution and a false positive rate of 10% for Kepler planet candidates (Morton & Johnson, 2011), we estimate that the probability of detecting 2 false positives from a sample of 4 targets is less than 1% (and less than 5% for detecting 3 false positives from 5 targets). Thus, we use this to emphasize that the false positive rate for current Kepler planet candidates with radii less than 5 R and orbital periods less than 6.0 days is likely much higher than 10%. In regards to the full Kepler sample, Morton & Johnson (2011) found that the false positive probability varied with the depth of the transit as well as the magnitude and Galactic latitude of the target star, but they did not investigate how the probability 171

172 might vary with orbital period or transit duration (though they claim that such properties would serve to decrease their probability estimates and thus their estimates are upper limits). Specifically, our results support their finding that the false positive rate increases slightly with magnitude, as our two faintest targets ended up being false positives. They also find a general increase in the false positive probability with increasing transit depth (though there are some local minima and maxima; e.g., Figure 7 in Morton & Johnson 2011) and with decreasing Galactic latitude. In our case, the target with the largest transit depth (as measured in the Kepler bandpass) was KOI , which we validated as a planet. While we find no obvious correlation in our sample between Kepler s measured transit depths and the false positive rate, dilution is an important factor that has to be considered (e.g., as illustrated by our light curves for KOI in Figure 6-6), as some of the depths measured by Kepler may be underestimated. In regards to Galactic latitude, in Figure 6-10 we show the cumulative distribution of the Galactic latitudes for the Borucki et al. (2011b) KOI list, the Batalha et al. (2012) KOI list and the Slawson et al. (2011) catalog of eclipsing binaries, which illustrates that the samples have the same general distribution. Furthermore, our false positives are indeed located at lower latitudes, which supports the argument by Morton & Johnson (2011) that there is a slight increase in the false positive rate with decreasing Galactic latitude. We present a similar figure in Figure 6-11, where we show the cumulative distribution of the Galactic latitudes for the Kepler eclipsing binaries along with separate distributions for stars that host short-period (P 2 days) and long-period (P > 2 days) planet candidates. We see that the distribution for the stars that host short-period candidates matches the eclipsing binary distributions more so than the sample of long-period candidates. Therefore, Figure 6-11 further illustrates the likely contamination of short-period planet candidates by eclipsing binaries. In Figure 6-12, we illustrate how the number of eclipsing binaries varies with Galactic latitude, and we find that overall the fraction of Kepler targets that are eclipsing binaries is about the same as a function of 172

173 Galactic latitude. This further implies a nearly consistent presence of eclipsing binaries that could plague the Kepler planet candidate sample Comparison to Observational Studies Désert et al. (2012) have used warm-spitzer follow-up of KOIs to find a low false positive rate consistent with the estimates from Morton & Johnson (2011). Overall, our sample includes shorter period planets than their sample, as well as fainter targets (excluding their sample of M dwarf stars). Given that only planet candidates with the shortest orbital periods (<< 2 days) are likely to be significantly contaminated by non-detached eclipsing binaries, it is not entirely surprising that Désert et al. (2012) found a low false positive rate for their sample of planet candidates with periods of days. Recently, Santerne et al. (2012) presented a Doppler study of short-period giant Kepler planet candidates. They targeted a sample of 46 Kepler planet candidates with transit depths greater than 0.4%, orbital periods less than 25 days, and host stars brighter than 14.7 magnitude (Kp). Based on their radial velocity follow-up observations, Santerne et al. (2012) estimate a false positive rate of 34.8±6.5% for this subset of Kepler planet candidates. While we probed a different (and smaller) population of Kepler planet candidates, our observed false positive rate of 50% (with large uncertainty due to our small sample size) for small, short-period Kepler planet candidates is supported by their observations. The study by Santerne et al. (2012) also supports our argument that different populations of Kepler targets likely have different false positive rates associated with them. 6.7 Conclusion We acquired multi-color transit photometry of four small ( R ), short-period ( days) Kepler planet candidates with the GTC. Based on the transit color, we identified two candidates as false positives. For two, we find further evidence supporting the planet hypothesis, consistent with validated planets. We also remind 173

174 the reader of KOI , which was a planet candidate observed in a similar fashion and that was also found to be a false positive (albeit it was selected from the first KOI catalog published by Borucki et al. 2011a; Colón & Ford 2011). While we find a high false positive rate (2/4 or 3/5, if we include KOI 565) in our small sample, we caution that this is likely not representative of the entire sample of Kepler planet candidates, due to the small number of targets we observed and the specific properties of these candidates (e.g. the orbital period and size). Nevertheless, our results demonstrate the importance of considering these properties when evaluating the false positive probability of specific systems. While our findings seem to contradict the theoretical estimates from Morton & Johnson (2011), the low false positive rate that they estimate is based on the assumption that all candidates had passed preliminary false-positive vetting metrics based on Kepler photometry and astrometry. Thus, if we consider the obviously V-shaped transits for KOI and KOI to imply a false positive nature for these KOIs, then according to Morton & Johnson (2011) the low false positive probabilities for these targets are not accurate. The false positive rate for our sample is also much larger than the observational constraints from Désert et al. (2012) that predict that the false positive rate is much less than 10%. This is likely partly a result of different targets being probed by the different studies. The recent study by Santerne et al. (2012) further supports this idea, as they found a 35% false positive rate for short-period giant Kepler planet candidates. We plan to continue observing small, short-period KOIs with the GTC in order to improve the sample size of our study. The observations presented here, as well as future observations with the GTC, greatly complement follow-up of KOIs done with warm-spitzer as well as other observatories. As we can expect the number of Kepler planet candidates to continue to increase, we can use results from all such studies to pinpoint which targets are the best to follow-up in order to maximize the science output from the Kepler mission. 174

175 175 Table 6-1. KOI Properties. Star Planet Candidate KOI KIC Kp T eff (K) b (deg) P (days) τ (hr) δ (ppm) R p (R ) R p /R All values are from Borucki et al. (2011b). The KIC number refers to the Kepler Input Catalog number for each target. Note that b is the Galactic latitude of the KOI host star, τ is the transit duration and δ is the transit depth as measured in the Kepler bandpass.

176 Table 6-2. Photometric Precisions. KOI σ 666nm (ppm) σ 858nm (ppm)

177 177 Table 6-3. Best-Fit Model Parameters. KOI t 0 τ a cos i/r R p /R R p /R c 1 c 2 c 1 c 2 (BJD ) (hr) (666 nm) (858 nm) (666 nm) (666 nm) (858 nm) (858 nm) (fixed) Note that the time of mid-transit, t 0, is technically given in BJD TDB (Barycentric Julian Date in Barycentric Dynamical Time). See text for further details.

178 Figure 6-1. Histograms illustrating the number of Kepler planet candidates (solid blue line; based on Borucki et al., 2011b) and detached and all other eclipsing binaries (dotted red and dashed black lines, respectively; based on Slawson et al., 2011) as a function of orbital period. We also show a histogram (dash-dot green line) based on the recently released list of 2,321 KOIs from Batalha et al. (2012). The righthand y-axis shows the percentage of systems relative to the total number of target stars observed by Kepler in Q1. Note that only systems with orbital periods of up to 10 days are shown. While there is a substantial number of other (primarily semi-detached and overcontact) eclipsing binaries at the shortest orbital periods (which may be less likely to be flagged as potential planets), there is still a greater-to-comparable number of short-period detached eclipsing binaries compared to short-period planet candidates, which indicates the potential for eclipsing binaries to infiltrate the short-period planet candidate sample. The orbital periods of the KOIs we observed are indicated with vertical long-dashed gray lines and are labeled accordingly. 178

179 Figure 6-2. Radius versus orbital period of Kepler planet candidates. The locations of the four KOIs we observed are indicated with filled red squares and are labeled accordingly. The locations of the targets observed with warm-spitzer by Désert et al. (2012) are indicated with filled blue squares. The filled gray circles and the open gray circles respectively represent all KOIs presented by Borucki et al. (2011b) and Batalha et al. (2012) out to orbital periods of 200 days. Note the prominence of KOIs smaller than 6 R. 179

180 Figure 6-3. Signal-to-noise ratio (SNR) per transit as a function of orbital period for Kepler planet candidates. The SNR per transit for each KOI was computed from the transit depth, the transit duration and either the 6-hour combined differential photometric precision (CDPP) from Q3 data (for the Borucki et al., 2011b list) or the 6-hour CDPP from Q1-Q6 data (for the Batalha et al., 2012 list). The colors and symbols are the same as in Figure

181 Figure 6-4. Kepler magnitude (Kp) versus effective temperature for the 997 KOI host stars published in Borucki et al. (2011b) and the additional 926 KOI host stars published in Batalha et al. (2012). The colors and symbols are the same as in Figures 6-2 and

182 Figure 6-5. Image of the field of view around KOI 1187 as acquired using the GTC/OSIRIS and the 666/36 nm order sorter filter. The target is labeled accordingly and is contained within a black circle equivalent to twice the aperture used in the reduction process. The unlabeled stars contained in black circles are the nine reference stars used in our analysis. The larger red circle around KOI 1187 has a radius of 20 arcsec, and the six brightest stars located within this circle (excluding the target) are the stars that we investigated as potential sources of the transit signal. 182