A&A 508, 10111020 (2009)
Transit timing analysis of the exoplanets TrES1 and TrES2^{}
M. Rabus^{1,2}  H. J. Deeg^{1,2}  R. Alonso^{3,4}  J. A. Belmonte^{1,2}  J. M. Almenara^{1,2}
1  Instituto de Astrofísica de Canarias, vía Lactea s/n, 38205 La Laguna, Spain
2  Departamento de Astrofísica, Universidad de La Laguna, 38205 La Laguna, Tenerife, Spain
3  Laboratoire d'Astrophysique de Marseille (UMR 6110), Technopôle de MarseilleÉtoile, 13388 Marseille, France
4  Observatoire de Genève, Université de Genève, 1290 Sauverny, Switzerland
Received 1 April 2009 / Accepted 28 August 2009
Abstract
Aims. The aim of this work is a detailed analysis of transit
light curves from TrES1 and TrES2, obtained over a period of three to
four years, in order to search for variabilities in observed
midtransit times and to set constraints on the presence of additional
third bodies.
Methods. Using the IAC 80 cm telescope, we observed
transits of TrES1 and TrES2 over several years. Based on these new
data and previously published work, we studied the observed light
curves and searched for variations in the difference between observed
and calculated (based on a fixed ephemeris) transit times. To model
possible transit timing variations, we used polynomials of different
orders, simulated OC diagrams corresponding to a perturbing third
mass, and we used sinusoidal fits. For each model we calculated the residuals and the false alarm probability (FAP).
Results. For TrES1, we can exclude planetary companions (>1
)
in the 3:2 and 2:1 MMRs having high FAPs based on our transit
observations from the ground. Likewise, a light time effect caused,
e.g., by a 0.09
mass star at a distance of 7.8 AU is possible. As for TrES2, we find a better ephemeris of
Epoch
and a good fit for a sine function with a period of 0.2 days,
compatible with a moon around TrES2 and an amplitude of 57 s, but
it is not a uniquely low
value that would indicate a clear signal. In both cases, TrES1 and
TrES2, we are able to put upper constraints on the presence of
additional perturbers masses. We also conclude that any sinusoidal
variations that might be indicative of exomoons need to be confirmed
with higher statistical significance by further observations, noting
that TrES2 is in the fieldofview of the Kepler Space Telescope.
Key words: planetary systems  methods: Nbody simulations  techniques: photometric
1 Introduction
In 1992, the first exoplanets around the pulsar PSR B1257+12 were discovered by detections of anomalies in the pulsation period (Wolszczan & Frail 1992). Similarly, by studying variations in the times of transit, transiting exoplanets give another possibility of finding additional companions, even down to Earth masses. There are different mechanisms causing these variations. For one the gravitational influence of a perturbing body can alter the orbital period of the transiting planet (Agol et al. 2005; Holman & Murray 2005). This effect is particularly sensitive to detections of additional bodies in meanmotion resonances with the transiting planet, or of moons around that planet. For another, a perturbing mass in an orbit larger than the transiting planet can cause the ``star  transiting planet'' system to wobble around the barycentre and alter the observed periodicity, something that is known as the lighttime effect (Schneider 2005; Doyle & Deeg 2004; Irwin 1959; Schneider & Doyle 1995). These anomalies are reflected in the times of transit. Using a linear ephemeris and several observations of transits, it is therefore possible to get the difference OC between the observed transit times and the calculated ones from the ephemeris. From this difference we can learn about perturbations on strict orbital periodicity caused by a possible companion object that has not been detected yet.
Motivated by the possible detection of lowmass companions around transiting planets, we started a longterm campaign to observe transits of the exoplanets TrES1 (Alonso et al. 2004) and TrES2 (O'Donovan et al. 2006) over several years. Here we present the results of our observations and a study of the transit timing variations that were found.
Searches for planetmass objects from the light time effect were first proposed for lowmass eclipsing binaries by Doyle et al. (1998) and have been performed indepth on the system CMDraconis, with ambiguous results to date (Morales et al. 2009; Deeg et al. 2008,2000). They were soon followed by several studies of timing effects in transiting planet systems: TrES1 (Steffen & Agol 2005), HD 189 733 (MillerRicci et al. 2008b), HD 209 458 (Agol & Steffen 2007; MillerRicci et al. 2008a), GJ 436 (Bean & Seifahrt 2008; Alonso et al. 2008) and CoRoT1b (Bean 2009). In all cases the authors could only constrain the parameters of a potential second planet.
The first search for variability in the transit times of TrES1 was conducted by Steffen & Agol (2005). They fitted perturbing planets in different orbits, using transit timing observations from Charbonneau et al. (2005). Steffen & Agol (2005) did not find any strong evidence of a third object. Winn et al. (2007) observed three additional transits of TrES1; they found that the transit times of these three observations occurred progressively later. However, as mentioned by the authors, these three measurements were not enough to make any firm conclusions. Recently, Raetz et al. (2009b) has reported a transit timing analysis of TrES1 with the only result of being an improved ephemeris. Previous studies of transit timings for TrES2 have been performed by Holman et al. (2007) and Raetz et al. (2009a), neither finding evidence of an unseen perturbing planet. Very recently Mislis & Schmitt (2009) analysed transit time durations between 2006 and 2008, indicating a possible change of inclination.
In Sect. 2 we describe the procedure to fit our observations to a transit model and to obtain the observed midtransit times and their error estimates. These values are then used in Sect. 3 to interpret the OC diagram by means of different models. Finally we discuss the results in Sect. 4.
2 Light curve fitting
All observations reported here were performed with the IAC 80 cm telescope (IAC80) at the Teide Observatory, Tenerife. We used the same observing and analysis procedure for both TrES1 and TrES2. For further investigation described here, we only used observations where a complete transit light curve was acquired. A detailed description of the analysis leading from the telescope images to the light curves can be found in Rabus et al. (2009), where a study of flux variations during transits caused by starspots or additional transiting planets for TrES1 was presented. We decorrelated the light curves against the airmass by subtracting a quadratic polynomial fit. Due to a slight defocusing of the telescope during observations, the centroids are not well defined and decorrelation against the target's detector position is not applicable. However, the defocusing and spread of the flux over several pixels will lower the noise correlated with the target's detector position.
For TrES1 we obtained eight useful transit observations (Table 2) over a period of three years and for TrES2 five observations (Table 3) over a period of two years. In order to measure the observed midtransit times, we first created a template of the transit event from a folding in phase and by binning (6point bins, with a mean size of 66 s for TrES1 and 90 s for TrES2) of all observations of the respective transiting planet (Fig. 1). The standard deviations inside the individual bins were 1.63 mmag for TrES1 and 1.79 mmag for TrES2, whereas the standard deviation outside the transit part of the binned light curve is 0.7 mmag for TrES1 and 0.8 mmag for TrES2.
We used these phased and binned light curves to create a model of the transit light curve, using the formalism from Giménez (2006) and the simplexdownhill fitting algorithm (Press et al. 1992). We fitted for the planetary and stellar radii ratio, k, the sum of the projected radii, rr, and orbital inclination, i, while we kept the eccentricity fixed at zero and the limb darkening coefficient fixed, as obtained by the tables of Claret et al. (1995), assuming quadratic coefficients. The bestfit models are plotted over phased data in Fig. 1 and the bestfit parameters are shown in Table 1. We can see that our light curve model parameters are consistent with the parameters from Southworth (2008) for both TrES1 and TrES2.
Table 1: Comparison between bestfit values of this work and parameters from Southworth (2008) for transit light curve models of TrES1 (upper values) and TrES2 (lower values).
Table 2: Overview of TrES1 transits used in this work.
Table 3: Same as Table 2, for TrES2.
Figure 1: Phased and binned light curve of all IAC80 observations of TrES1 ( top) and TrES2 ( bottom) transits with the solid (red) line being the bestfit model light curve. 

Open with DEXTER 
We then shifted in time our model with the fixed bestfit parameters
against each individual observed light curve and calculated the
residuals for each shift. The fit with the minimum
value gave us the observed midtransit time. We estimated the timing error within a 68% (1)
confidence interval of the
values, given by the range where the
residuals increases by 1 over the minimum value. We also calculated the timing precision of each individual light curve,
,
by propagating their respective photometric precision, ,
using the equation (Doyle & Deeg 2004):
where L(t_{i}) is the stellar brightness at t_{i}, is the cadence and the sum goes over all brightness values within the eclipse event. Comparing the estimated error, based on the 1 interval, with the error calculated using Eq. (1) we found a good agreement; both error values are indicated in Tables 2 and 3. Generally the estimated 1 errors are higher than the propagated errors, but in four cases the error based on the 68% confidence interval were lower. Our timing errors are possibly underestimated due to correlated noise. However, we do not modify them in order to stay consistent with the timing measurements and associated errors that we took from the literature.
Finally we subtracted the observed midtransit times from the calculated ones, obtaining the OC values. The calculated midtransit times for TrES1 were obtained from the ephemeris Epoch (Winn et al. 2007) and for TrES2 we used Epoch (Holman et al. 2007).
In the following transit timing analysis we included several published midtransit observations, by Charbonneau et al. (2005), Winn et al. (2007) and Hrudková et al. (2009) for TrES1 (Table 2), and for TrES2 by Holman et al. (2007) (Table 3). We considered only OC times with errors below 60 s, which led to the rejection of some OC values from Charbonneau et al. (2005). In the OC residuals of TrES1 we also removed two outliers at Epoch 255 and 358 that have been identified by Rabus et al. (2009) as transits with possible starspots. Several further transit times of TrES2 and TrES1 were recently reported by Raetz et al. (2009b,a), respectively, and for TrES2 by Mislis & Schmitt (2009), but we did not include these in our study, since they had individual errors and an internal scatter several times larger than the data included in this study. The transit midtime corresponding to Epoch 0 of TrES2, reported by O'Donovan et al. (2006) has also been removed due to its high error of >60 s. We obtained for TrES1 an OC diagram spanning four years, with 16 points (Fig. 2) and for TrES2 a diagram spanning two years, with 8 points (Fig. 3).
Figure 2: Difference between calculated (based on ephemeris from Winn et al. 2007) and observed midtransit times for TrES1. Filled dots are OC values obtained with the IAC80, squares are taken from Charbonneau et al. (2005), triangles from Winn et al. (2007) and asterisk are from Hrudková et al. (2009). The lines show different polynomial fits, where the solid black line indicates a linear fit, the longdashed (red) line a quadratic and the shortdashed (blue) line a cubic polynomial. The dotted (red) lines show the fits corresponding to the variation of the quadratic term within 1 confidence limits. 

Open with DEXTER 
Figure 3: Similar as Fig. 2 (based on ephemeris from Holman et al. 2007) including only polynomials fits and corresponding 1 confidence limits, but for TrES2. Filled dots are OC values obtained with the IAC80 and squares are taken from Holman et al. (2007). 

Open with DEXTER 
3 Transit timing analysis
Two mechanisms have been brought forward that may cause deviations of transit times from strict periodicity. For one the light time effect and for another the influence of a third body on the transiting planet's intrinsic periodicity.
3.1 Search for the light time effect
We first consider the light time effect, where the barycentre of the binary system, in our case the system ``startransiting planet'', is offset against the barycentre formed by an additional third body. This will cause the light to travel a longer or shorter distance to the observer and hence the observer will see a different period, where the intrinsic period is unknown.
For the analysis of the OC times, we fitted three polynomials of different orders to the OC values, namely a linear,
,
a parabolic,
,
and a cubic polynomial,
,
where E is the Epoch number and
the fitted polynomial coefficients. We also analysed the case of fixed original ephemeris (
)
and the case of maintaining the original period and fitting only for a constant offset in OC (
). Table 4 shows the obtained bestfit coefficient for TrES1 and TrES2 respectively, and Figs. 2 and 3
show a plot of the respective polynomials. In order to quantify the
improvement of the different fits, we calculated the false alarm
probability (FAP), that is the significance level of the fit quality
improvement and indicates the probability of making a type I error.
Therefore, we first calculated the Fvalues according to equation:
where are residuals from the lower order fit, are residuals from the higher order fit, and are the corresponding degrees of freedom (Table 4). We then used the Fvalue to calculate the significance using the IDLroutine MPFTEST from the Markwardt IDL library, that gives the probability for a value drawn from the Fdistribution to equal or exceed the given Fvalue. For TrES1, FAPs were calculated against the original ephemeris, against the offsetonly case and against the linear polynomial. We see that the original ephemeris is unlikely to be the best solution with the linear and quadratic fits being the most likely descriptions, having low residuals and a low FAP. For TrES2 a linear trend in the OC residuals (Fig. 3) is apparent. The statistical analysis shows no clear preference between a linear or a quadratic polynomial (Table 4). The distribution of the observations of TrES2 into three groups, acting as pivot points, does not support fits of orders higher than the quadratic one. Therefore, we fitted no cubic polynomial to the TrES2 data.
As for the linear coefficients
and
these are without relevant physical meaning, but indicate a slightly different ephemeris than used, where
is an offset of the midtransit time at epoch 0 and a
correction to the period of the ephemeris. However, we can give a
physical meaning to the quadratic coefficient. The quadratic term (), gives the system's acceleration along the line of sight by using the equation (Deeg et al. 2008):
where c is the speed of light and P is the observed period.
In order to estimate the error of the quadratic term, we stepped through different values of , fitting for and . The values which increased the bestfit residuals by 2.3 gave the 1 confidence limits (see Table 4); dotted lines in Figs. 2 and 3 show the corresponding fits at these limits. Our best fit parabola model for TrES1 gave a quadratic term of s. Solving Eq. (3) we obtained an acceleration of m s^{2} and s^{1}. The cubic term would indicate a constant change in acceleration; however the cubic fit is also less likely than the quadratic one and will not be further discussed.
For the quadratic solution of TrES2 we obtained a bestfit value of smaller than its error. This high error is consistent with a low significance of the quadratic solution, therefore we support the linear case, with a 1 upper limit for accelerations of m s^{2}.
Considering the clear linear trend of OC times in Figs. 2 and 3, and using the coefficients from the linear fit, we indicate here an improved ephemeris for TrES1, given by:
where the values in parenthesis give the uncertainty in the last two digits and for TrES2:
3.2 Search for a perturbation of the intrinsic period
The other cause for transit timing variations could be a perturbation of the intrinsic planet period P' caused by the gravitational influence of a third body on the transiting planet. Regarding such perturbations, there exist no analytical equations that describe the gravitational influence on a transiting planet caused by an undetected third body. Generally, Nbody simulations are used, iterating over a large orbital parameter space for a maximum possible mass range, see e.g. Agol et al. (2005); Agol & Steffen (2007); Holman & Murray (2005); MillerRicci et al. (2008b); Steffen & Agol (2005); Nesvorný & Morbidelli (2008); MillerRicci et al. (2008a).
In order to find the parameter space of third bodies compatible with the observed transit times, we created a numerical 2dimensional simulation of a three body system by integrating over the equations of motion. We considered the problem in two dimensions, assuming that the orbits of the exoplanetary system and its respective perturber are coaligned. We further considered the problem in a heliocentric frame, with the star in the centre of the coordinate system, meaning that we neglect the lighttime effect in this context. To integrate the equation of motion we used the BurlischStoer algorithm (Press et al. 1992) with a 1 s time step and an accuracy of 10^{10}.
Table 4: Comparison of different residuals for TrES1 (upper values) and TrES2 (lower values).
We considered the problem for an inner and an outer perturber separately and neglected the 1:1 mean motion resonance (MMR). Each simulation started with a perturber 0.005 AU away from the transiting planet, with an initial zero eccentricity and with a phase shift towards the transiting planets between 0 and 315 in steps of 45. We advanced the simulations in orbital steps of 0.001 AU and used 100 mass steps from 1 to 100 for the perturbing object. We simulated 1000 transits, corresponding to approximately eight years coverage for TrES1 and seven years coverage for TrES2. We used a mass of 0.61 M_{J} for TrES1 and a mass of 0.88 for its host star (Torres et al. 2008), and a mass of 1.2 M_{J} for TrES2 and 0.98 for its host star (Sozzetti et al. 2007). After obtaining the simulated OC diagrams, we applied a Fourier transformation to the synthetic OC diagrams (MillerRicci et al. 2008b,a) and derived the maximum obtained OC amplitudes as a function of the perturber's semimajor axis; showing them for masses of 10, 20 and 40 in Fig. 4, where also the MMRs are indicated by vertical lines. Figures 5 and 6 show some examples of synthetic OC diagrams for TrES1.
Figure 4: Maximum OC amplitudes of a 2 dimensional Nbody simulation for the system TrES1 ( upper two panels) and TrES2 ( lower two panels) caused by a perturber during 1000 transits plotted against the semimajor axes of the perturber. The coloured solid lines correspond to different perturbing masses, namely 10 (red), 20 (blue) and 40 (green). Left side: inner perturber, right side: outer perturber. Vertical lines indicate the MMRs. 

Open with DEXTER 
Figure 5: Example synthetic OC diagrams for TrES1 and a perturber of 30 at 0.053 AU and with different phase shifts between them. Note that the appearance of several lines is caused by an aliasing effect between consecutive transits. For a given transit number there is only one OC time. 

Open with DEXTER 
Figure 6: Same as Fig. 5, but for a perturber distance of 0.062 AU (2:1 MMR). 

Open with DEXTER 
To establish potential third body orbital parameters, we fitted for each perturber distance the observed OC diagram to all synthetic ones that had been generated for masses of 1100 (Fig. 7). We left three parameters open in the fits: one parameter to shift the OC diagrams in Epoch, corresponding to find the best moment for the first real observation (E=0) within the simulated data, and two parameters describing linear deviations between simulated and modelled OC diagrams, that have no relevant physical meaning.
For the bestfit masses, we see a general trend to higher masses for perturbers closer to the host star and at larger semimajor axis, and we can also identify the meanmotion resonances, where the best fit for a consistent perturber indicates lower masses (vertical lines in Fig. 7). At larger orbits for the perturber, the bestfit perturber's mass increases and it increases more steeply for TrES2 than for TrES1. This is due to TrES2 having a higher mass than TrES1 and being closer to the central star. The high peaks at the 2:3 and 2:1 MMRs for TrES1 reveal that a >1 planet in that configuration might have been detected in our data set.
Again, in order to quantify the improvement of the simulated OC fits in the lower MMRs and the bestfit simulated OC against the linear polynomial, we calculate the FAPs. In these cases it is better not to use Eq. (2) to calculate the Fvalue, since this equation is better suited for cases where one expects small changes for additional parameters, whereas for the simulated OC values, we expect big changes. Therefore, we used to calculate the Fvalue for the MPFTESTroutine. In Table 4, we show the and FAPs for some special cases of our simulated OC values.
For TrES1 we found that most of the values for the simulated OC models were below the one of the linear fit, but based on the available data, we can exclude the case of 2:3 and 2:1 MMR for planets more massive than 1 with high FAPs. We found a best fit for a perturber with a mass of 2 and a semimajor axis of 0.05 AU, but this peak is not outstandingly low. Similarly, for TrES2, all values for the OC models are below the one from the linear fit. We can identify some low peaks, with the lowest at 0.049 AU ( ) for a perturbing mass of 18 . Figure 8 shows for TrES2 the OC simulation for the lowest with the observed OC residuals overplotted. Given that none of the simulated perturbers indicate a uniquely low , we do not find support for any of them.
In order to calculate an upper mass limit above which we can reject perturber masses due to a high increase in , we first set a threshold to 50.2 for TrES1 and 28.5 for TrES2. These thresholds correspond to a 90% FAP against the linear fit using the MPFTESTroutine. We then increased the masses for each semimajor axis until the residuals reached the previously established threshold, which gave us the upper mass limit. Masses above the threshold can be rejected with a FAP higher than 90% (Fig. 7).
Figure 7: Masses of third bodies with the lowest (solid black line), based on the best fit of the observed OC values to the synthetic OC values for each semimajor axis, and upper mass limits resulting in a FAP of 90% (dasheddotted black line) with respect to the linear fit for each semimajor axis. Red lines show the lowest values (dashed red line) corresponding to the bestfit masses and the value of the linear fit is indicated by a horizontal dasheddotted red line. The upper plot is for TrES1 and the lower plot for TrES2. Mean motion resonances are indicated by vertical dotted lines. 

Open with DEXTER 
3.3 Search for a sinusoidal transit timing variability
In order to search for any sinusoidal periodicity in the data with periods on the order of the campaign duration and shorter, we fitted amplitude and phase of a sinusoidal function to the OC residuals after subtracting the linear fit with the parameters of Table 4. A sinusoidal transit timing variation might be caused by an exomoon around the transiting planet (Doyle & Deeg 2004; Simon et al. 2007; Sartoretti & Schneider 1999; Kipping 2009). For this fit we stepped through different trial periods between 1500 days ( cycl/d) and 0.1 days (10 cycl/d) for TrES1 and for TrES2 between 600 days ( cycl/d) and 0.1 days (10 cycl/d) in steps of 0.00001, 0.0001, 0.001, 0.01 and 0.1 cycl/d. For each trial period we then logged the residuals at the bestfitting amplitude and phase.
From Fig. 9, showing the result of the sine fit, we see that we lack outstanding peaks of low values for TrES1 and TrES2, being it very unlikely that a real sinusoidal signal has more than one period. However, we also calculated the FAPs of the best sinusoidal fit against the linear polynomial, as described in Sect. 3.2 and show the results in Table 4 for comparison. We note that the best sinusoidal fit for TrES1 with a period of 16.7 days and an amplitude of 25 s (Fig. 10, upper graph) does not give a significant improvement against the linear polynomial, having a lower residual but a high FAP, but for TrES2 we found a good sinusoidal fit with a FAP of 1% for a period of 0.2 days and an amplitude of 57 s (Fig. 10, lower graph).
4 Discussion
For TrES1 we obtained eight new OC values and made use of 10 previously published values of which we removed two observations with a high probability of containing star spots. The standard deviation of the OC values is 33 s and the maximum OC deviation from the established ephemeris (Winn et al. 2007) is 60 s. Similarly, for TrES2 we observed five transits and made use of three additional published midtransit times (Holman et al. 2007). The transit timing estimation at epoch 0 for TrES2 had an error greater than 60 s and was hence removed. After correcting the ephemeris, the standard deviation of this OC diagram is 43.9 s and the maximum OC deviation, is 66.5 s.
Figure 8: Simulated OC diagram of TrES2 and the best fit perturber with a mass of 18 and an orbit of 0.051 AU. The red dots correspond to the observations and the squares to the corresponding simulated OC values. 

Open with DEXTER 
In order to search for a light time effect, we fitted different
polynomials. Assuming the validity of the quadratic function for
TrES1, we can use the quadratic term to estimate a possible
acceleration and its corresponding 1
error of
s^{1},
the negative value indicating a decelerating system. Comparing this
deceleration to the acceleration for the solar system of a few 10^{19} s^{1} (Zakamska & Tremaine 2005) and CM Draconis of 10^{17} s^{1} (Deeg et al. 2008),
the acceleration of TrES1 could be several orders of magnitudes
larger, but is not well constrained given the current data. Following Deeg et al. (2008), the minimum mass of a third body causing a given acceleration in dependence of its lateral distance
is given by
We note that this is independent of the mass of the accelerating system. Using the distance to TrES1 of 157 pc (Sozzetti et al. 2004) and above value for , we obtain the minimum mass of a possible third body as a function of angular separation from TrES1 of
This relation allows the identification of nearby objects found in any future highresolution imaging; that is, any possibly found object has to fulfil Eq. (7) in order to be a potential source of the observed acceleration. For example, for a third object at a distance of 0.05 the object's mass has to be at least 0.09 in order to cause the previously mentioned acceleration ( ). The angular separation of 0.05 translates into an orbit for a third object of 7.8 AU, corresponding to a period of 21.8 years. This stellar object would have caused a semiamplitude in the radial velocity measurements of 1.0 km s^{1}. However, the observation span of the available radial velocity measurements is only 49 days (Alonso et al. 2004; Laughlin et al. 2005), that is 0.6% of the exemplary third object's period. Therefore, this object would not have been detected in existing radial velocity measurements. This example also shows that the obtained value for the acceleration is reasonable.
A second approach has been assuming perturbing objects at nearby orbits. Then for TrES1, perturbers with a masses >1 at the 2:3 and 2:1 MMRs are very unlikely due to the high residual with a high FAP against the linear fit. However, none of the simulated perturbers indicated a uniquely low peak. Therefore, we do not find support for any perturber, but we established upper mass limits above which we would obtain a FAP of 90% against the linear fit. For TrES2, we obtained some low peaks; none of them in a loworder MMR. For the bestfit simulated OC we obtained a perturber of 18 and 0.049 AU. Due to the residuals below the one from the linear fit, none of the perturbers could be excluded and again we established upper mass limits for perturbers in the TrES2 system.
Figure 9: Results of the sine fitting as function of trial frequencies. The red shaded areas show possible frequencies which might have been introduced by a moon. Top: TrES1, bottom: TrES2. 

Open with DEXTER 
Figure 10: Best fit sine function (solid line) and observed OC values with error bars for TrES1 and TrES2, phased with period of that fit. This period is 0.06 cycles/day or 16.7 days for TrES1 and 4.8 cycles/day or 0.21 days for TrES2. 

Open with DEXTER 
Steffen & Agol (2005) did an analysis of transit times of TrES1, but they found no convincing evidence of a second planet. We also found generally a higher near to the loworder MMRs. Steffen & Agol (2005) gave upper mass limits for additional planets in the system TrES1, whereas we give additionally the bestfitting mass for any orbital distance. In general, their upper mass limit for additional planets decreases closer to the transiting planet, similar to our findings in Fig. 7. We also found that additional planets with masses above 10 at the MMRs are very unlikely, that is consistent with the results from Steffen & Agol (2005), indicating that at the loworder MMRs and near zero eccentricity, the mass of the additional planet has to be below 10 .
Assuming sinusoidal transit timing variations, we note that we find a good OC fit for TrES2 with a FAP of 1.1%. However, since we find several good periods beyond the best one of 0.2 days, we conclude that we need more observations and maybe with higher precision in order to confirm any one of them as a possible exomoon. Whereas for TrES1, we found no evidence that a sinusoidal function improved the fit significantly with respect to the linear one.
Sartoretti & Schneider (1999), Doyle & Deeg (2004), Simon et al. (2007) and recently Kipping (2009)
discussed the possibility of detections of moons around extrasolar
planets using a timing offset induced by the wobble of the planet
around the planetmoon barycentre. For a given timing offset
the mass of the possible moon
can be estimated using (Sartoretti & Schneider 1999):
where is the semimajor axis of the planet, the mass of the transiting planet, a_{M} semimajor axis of the moon and the period of the planet. A moon around the transiting planet should have a semimajor axis which is between the Roche limit and the Hill radius (Kipping 2009). The Roche limit is given by:
where is the density of the planet and the density of the moon. As can be seen from Eq. (9), if , the Roche limit is inside the planet. This is most likely the case for a rocky moon around a gas giant planet. Therefore, we do not further consider the Roche limit as lower orbit limit, but the planetary radius. On the other hand, the Hill radius is given by:
where M_{*} is the mass of the star. Using the radius of the transiting planets, i.e. R_{J} or and Eq. (10), we obtain possible orbital periods for a moon around the transiting planets of 4.2 h < < 15 h for TrES1 and 3.9 h < < 12 h for TrES2. If we consider only the corresponding period ranges in Fig. 9 (red shaded areas), we also find no clear peak there for TrES1, but for TrES2 the best peak of 0.2 days is in the range for an exomoon. Using Eq. (8) we obtain a possible moon mass of 52 for the bestfit amplitude of this peak.
We note that Barnes & O'Brien (2002) gave an analytical expression for an upper mass limit of possible moons. Generally, the more massive the moon the shorter its lifetime. Lubow et al. (1997) estimated that closein orbiting planets spin down into synchronous rotation very quickly. Therefore we use the procedure from Barnes & O'Brien (2002) to estimate the upper mass limit of a moon around TrES1 and TrES2. We obtain an upper mass limit around TrES1 of the order of 10^{6} and around TrES2 of the order of 10^{7} , similar to one obtained for HD 209 458b by Barnes & O'Brien (2002). This means that moons with masses greater than 10^{6} will not have survived until now. This upper mass limit is clearly way below our detection limit for masses causing transit timing variations in the system TrES1 and TrES2. The closest moon at this upper mass limit would cause a timing amplitude of the order of 10^{6} s, which is not detectable even from space by several orders of magnitude.
Transit duration investigations on TrES2 performed by Mislis & Schmitt (2009) indicated a decrease of 3 min. between 2006 and 2008 in the duration. However, in our transit timing analysis, we do not find nonlinear deviations of OC times with a similar magnitude. While we can not provide any firm detection, we can put upper mass limits which are consistent with our observations for TrES1 and TrES2, respectively. However, due to the gaped data, it might have been possible that we missed important points in the observed OC diagram, like e.g. a transit timing measurement at the highest amplitude at the meanmotion resonances. We need at least about 3 years of continuous transit observations and with several transit observations per year in order to avoid the missing of windows of highamplitude OC deviations from perturbing bodies, with the maximum amplitudes being observable only during a few transit events, see Fig. 6. But even for lowmass perturbers such amplitudes should be easily measurable from the ground at meanmotion resonances. On the other hand, outside the meanmotion resonances, we need higher precision than obtainable from the ground for detections of the small amplitudes caused by a perturbing body of planetary mass, something that may be expected from the current satellite missions in operation. Any transit timing variation in TrES2 may be expected to be confirmed in the near future by observations done with the Kepler Space Telescope.
AcknowledgementsThis article publishes observations made with the IAC80 telescope operated by the Instituto de Astrofísica de Canarias in the Observatorio del Teide. We thank the students and staff of the Teide Observatory and Angel de Vicente Garrido for his help with the IAC Supercomputing facility CONDOR. M. R. likes to thank the European Association for Research in Astronomy (EARA) for their support by a EARA  Marie Curie Early Stage Training fellowship. Part of this work was financed by grant ESP20076548000202 by the Spanish investigation and development ministry. We also thank Tristan Guillot and the anonymous referee for valuable comments that led to a significant improvement of the analysis presented here.
References
 Agol, E., & Steffen, J. H. 2007, MNRAS, 374, 941 [NASA ADS] [CrossRef]
 Agol, E., Steffen, J., Sari, R., et al. 2005, MNRAS, 359, 567 [NASA ADS] [CrossRef]
 Alonso, R., Brown, T. M., Torres, G., et al. 2004, ApJ, 613, L153 [NASA ADS] [CrossRef]
 Alonso, R., Barbieri, M., Rabus, M., et al. 2008, A&A, 487, L5 [NASA ADS] [EDP Sciences] [CrossRef]
 Barnes, J. W., & O'Brien, D. P. 2002, ApJ, 575, 1087 [NASA ADS] [CrossRef]
 Bean, J. L. 2009, A&A, 506, 369 [EDP Sciences] [CrossRef]
 Bean, J. L., & Seifahrt, A. 2008, A&A, 487, L25 [NASA ADS] [EDP Sciences] [CrossRef]
 Charbonneau, D., Allen, L. E., Megeath, S. T., et al. 2005, ApJ, 626, 523 [NASA ADS] [CrossRef]
 Claret, A., DiazCordoves, J., & Gimenez, A. 1995, A&AS, 114, 247 [NASA ADS]
 Deeg, H. J., Doyle, L. R., Kozhevnikov, V. P., et al. 2000, A&A, 358, L5 [NASA ADS]
 Deeg, H. J., Ocaña, B., Kozhevnikov, V. P., et al. 2008, A&A, 480, 563 [NASA ADS] [EDP Sciences] [CrossRef]
 Doyle, L. R., & Deeg, H.J. 2004, in Bioastronomy 2002: Life Among the Stars, ed. R. Norris, & F. Stootman, IAU Symp., 213, 80
 Doyle, L. R., Deeg, H., Jenkins, J. M., et al. 1998, in Brown Dwarfs and Extrasolar Planets, ed. R. Rebolo, E. L. Martin, & M. R. Zapatero Osorio, ASP Conf. Ser., 134, 224
 Giménez, A. 2006, A&A, 450, 1231 [NASA ADS] [EDP Sciences] [CrossRef]
 Holman, M. J., & Murray, N. W. 2005, Science, 307, 1288 [NASA ADS] [CrossRef]
 Holman, M. J., Winn, J. N., Latham, D. W., et al. 2007, ApJ, 664, 1185 [NASA ADS] [CrossRef]
 Hrudková, M., Skillen, I., Benn, C., et al. 2009, in IAU Symp., 253, 446
 Irwin, J. B. 1959, AJ, 64, 149 [NASA ADS] [CrossRef]
 Kipping, D. M. 2009, MNRAS, 392, 181 [NASA ADS] [CrossRef]
 Laughlin, G., Wolf, A., Vanmunster, T., et al. 2005, ApJ, 621, 1072 [NASA ADS] [CrossRef]
 Lubow, S. H., Tout, C. A., & Livio, M. 1997, ApJ, 484, 866 [NASA ADS] [CrossRef]
 MillerRicci, E., Rowe, J. F., Sasselov, D., et al. 2008a, ApJ, 682, 586 [NASA ADS] [CrossRef]
 MillerRicci, E., Rowe, J. F., Sasselov, D., et al. 2008b, ApJ, 682, 593 [NASA ADS] [CrossRef]
 Mislis, D., & Schmitt, J. H. M. M. 2009, A&A, 500, L45 [NASA ADS] [EDP Sciences] [CrossRef]
 Morales, J. C., Ribas, I., Jordi, C., et al. 2009, ApJ, 691, 1400 [NASA ADS] [CrossRef]
 Nesvorný, D., & Morbidelli, A. 2008, ApJ, 688, 636 [NASA ADS] [CrossRef]
 O'Donovan, F. T., Charbonneau, D., Mandushev, G., et al. 2006, ApJ, 651, L61 [NASA ADS] [CrossRef]
 Press, W. H., Teukolsky, S. A., Vetterling, W. T., et al. 1992, Numerical recipes in FORTRAN. The art of scientific computing (Cambridge: University Press, c1992, 2nd edn.)
 Rabus, M., Alonso, R., Belmonte, J. A., et al. 2009, A&A, 494, 391 [NASA ADS] [EDP Sciences] [CrossRef]
 Raetz, S., Mugrauer, M., Schmidt, T. O. B., et al. 2009a, Astron. Nachr., 330, 459 [NASA ADS] [CrossRef]
 Raetz, S., Mugrauer, M., Schmidt, T. O. B., et al. 2009b, Astron. Nachr., 330, 475 [NASA ADS] [CrossRef]
 Sartoretti, P., & Schneider, J. 1999, A&AS, 134, 553 [NASA ADS] [EDP Sciences] [CrossRef]
 Schneider, J. 2005, in The LightTime Effect in Astrophysics: Causes and cures of the OC diagram, ed. C. Sterken, ASP Conf. Ser., 335, 191
 Schneider, J., & Doyle, L. R. 1995, Earth Moon and Planets, 71, 153 [NASA ADS] [CrossRef]
 Simon, A., Szatmáry, K., & Szabó, G. M. 2007, A&A, 470, 727 [NASA ADS] [EDP Sciences] [CrossRef]
 Southworth, J. 2008, MNRAS, 386, 1644 [NASA ADS] [CrossRef]
 Sozzetti, A., Yong, D., Torres, G., et al. 2004, ApJ, 616, L167 [NASA ADS] [CrossRef]
 Steffen, J. H., & Agol, E. 2005, MNRAS, 364, L96 [NASA ADS]
 Sozzetti, A., Torres, G., Charbonneau, D., et al. 2007, ApJ, 664, 1190 [NASA ADS] [CrossRef]
 Torres, G., Winn, J. N., & Holman, M. J. 2008, ApJ, 677, 1324 [NASA ADS] [CrossRef]
 Winn, J. N., Holman, M. J., & Roussanova, A. 2007, ApJ, 657, 1098 [NASA ADS] [CrossRef]
 Wolszczan, A., & Frail, D. A. 1992, Nature, 355, 145 [NASA ADS] [CrossRef]
 Zakamska, N. L., & Tremaine, S. 2005, AJ, 130, 1939 [NASA ADS] [CrossRef]
Footnotes
 ... TrES2^{}
 Photometric data for TrES1 and TrES2 are only available in electronic form at the CDS via anonymous ftp to cdsarc.ustrasbg.fr (130.79.128.5) or via http://cdsweb.ustrasbg.fr/cgibin/qcat?J/A+A/508/1011
All Tables
Table 1: Comparison between bestfit values of this work and parameters from Southworth (2008) for transit light curve models of TrES1 (upper values) and TrES2 (lower values).
Table 2: Overview of TrES1 transits used in this work.
Table 3: Same as Table 2, for TrES2.
Table 4: Comparison of different residuals for TrES1 (upper values) and TrES2 (lower values).
All Figures
Figure 1: Phased and binned light curve of all IAC80 observations of TrES1 ( top) and TrES2 ( bottom) transits with the solid (red) line being the bestfit model light curve. 

Open with DEXTER  
In the text 
Figure 2: Difference between calculated (based on ephemeris from Winn et al. 2007) and observed midtransit times for TrES1. Filled dots are OC values obtained with the IAC80, squares are taken from Charbonneau et al. (2005), triangles from Winn et al. (2007) and asterisk are from Hrudková et al. (2009). The lines show different polynomial fits, where the solid black line indicates a linear fit, the longdashed (red) line a quadratic and the shortdashed (blue) line a cubic polynomial. The dotted (red) lines show the fits corresponding to the variation of the quadratic term within 1 confidence limits. 

Open with DEXTER  
In the text 
Figure 3: Similar as Fig. 2 (based on ephemeris from Holman et al. 2007) including only polynomials fits and corresponding 1 confidence limits, but for TrES2. Filled dots are OC values obtained with the IAC80 and squares are taken from Holman et al. (2007). 

Open with DEXTER  
In the text 
Figure 4: Maximum OC amplitudes of a 2 dimensional Nbody simulation for the system TrES1 ( upper two panels) and TrES2 ( lower two panels) caused by a perturber during 1000 transits plotted against the semimajor axes of the perturber. The coloured solid lines correspond to different perturbing masses, namely 10 (red), 20 (blue) and 40 (green). Left side: inner perturber, right side: outer perturber. Vertical lines indicate the MMRs. 

Open with DEXTER  
In the text 
Figure 5: Example synthetic OC diagrams for TrES1 and a perturber of 30 at 0.053 AU and with different phase shifts between them. Note that the appearance of several lines is caused by an aliasing effect between consecutive transits. For a given transit number there is only one OC time. 

Open with DEXTER  
In the text 
Figure 6: Same as Fig. 5, but for a perturber distance of 0.062 AU (2:1 MMR). 

Open with DEXTER  
In the text 
Figure 7: Masses of third bodies with the lowest (solid black line), based on the best fit of the observed OC values to the synthetic OC values for each semimajor axis, and upper mass limits resulting in a FAP of 90% (dasheddotted black line) with respect to the linear fit for each semimajor axis. Red lines show the lowest values (dashed red line) corresponding to the bestfit masses and the value of the linear fit is indicated by a horizontal dasheddotted red line. The upper plot is for TrES1 and the lower plot for TrES2. Mean motion resonances are indicated by vertical dotted lines. 

Open with DEXTER  
In the text 
Figure 8: Simulated OC diagram of TrES2 and the best fit perturber with a mass of 18 and an orbit of 0.051 AU. The red dots correspond to the observations and the squares to the corresponding simulated OC values. 

Open with DEXTER  
In the text 
Figure 9: Results of the sine fitting as function of trial frequencies. The red shaded areas show possible frequencies which might have been introduced by a moon. Top: TrES1, bottom: TrES2. 

Open with DEXTER  
In the text 
Figure 10: Best fit sine function (solid line) and observed OC values with error bars for TrES1 and TrES2, phased with period of that fit. This period is 0.06 cycles/day or 16.7 days for TrES1 and 4.8 cycles/day or 0.21 days for TrES2. 

Open with DEXTER  
In the text 
Copyright ESO 2009