Minimising the needs of follow-up observations for PLATO - the central role of planet-validation tools - Alexandre Santerne Marie Curie Fellow Instituto de Astrofísica e Ciências do Espaço Universidade do Porto alexandre.santerne@astro.up.pt supported by the European Union under a Marie Curie Intra-European Fellowship for Career Development with reference FP7-PEOPLE-2013-IEF, number 627202.
The planet-validation technique, in short Almenara et al. (2011) etic transit light curve generated from a planetary model, shown in green. The red curve is the best fit odel. Both models are compatible with the data points (black dots with error bars). s intensive computation resources (to compute up to 7 10 8 models), they use 1024 processors leiades cluster. Then, they construct maps for two of the free parameters: distance and mass star in the binary for a background/foreground eclipsing binary scenario, mass of the secondary psing binary and the mass of the primary star for the triple stellar system, planetary radius and the transiting star for the star-star-planet triple scenario. The statistics in this maps is based nce between the considered scenario and the best model of the star-transiting planet scenario. regions are obtained using the number of free parameters as the number of degrees of freedom, outside the 3 contour is excluded. But the general approach is not conceptually correct, as in approach model comparison is not possible. This method has not been proved to be statistically g, for example, simulated data. Additional observations (radial velocity, high resolution image optics, transits observed in the infrared with Spitzer) add, a posteriori, constraints in the statistic from the 2 di erence. These maps are used to constrain the allowed magnitude range of the inside the 3 contour and allowed by the additional observations). Then, they use the Besançon ure models (Robin et al. 2003) to count background/foreground stars in the allowed magnitude, taking into account the star counts and the probability of each scenario they compute a false the star-transiting planet scenario. If this false alarm rate is small enough, the planet is said to ressin et al. 2011). bound & unbound
The planet-validation Scenario 0 302 technique, 128 39 70 in 161 short Almenara et al. (2011) etic transit light curve generated from a planetary model, shown in green. The red curve is the best fit odel. Both models are compatible with the data points (black dots Pwith S error bars). i D, I 1. Model all astrophysical the mean uncertainty < the jitter. s intensive computation resources (to compute up to 7 10 8 models), they use 1024 processors P S leiades cluster. Then, they construct maps for two of the free parameters: j D, I distance and mass star in the binary for a background/foreground eclipsing binary scenario, mass of the secondary false-positive scenarios psing binary and the mass of the primary star for the triple stellar system, planetary radius and S the transiting star for the star-star-planet triple scenario. The statistics in this maps is based 2. Constrain scenarios using i I P D S i, I nce between the considered scenario and the best O ij model = of the star-transiting planet scenario. regions are obtained using the number of free parameters as the number of degrees of freedom, available data S outside the 3 contour is excluded. But the general approach is not j I P conceptually R D S correct, j, I as in approach model comparison is not possible. This method has not been proved to be statistically 3. Evaluate each scenario g, for example, simulated data. Additional observations (radial velocity, high resolution image optics, transits observed in the infrared with Spitzer) add, a posteriori, constraints in the statistic from the 2 di probability S erence. These maps are used to constrain the allowed i I magnitude R i range of the inside the 3 contour and allowed by the additional O ij observations). = Then, they i use the Besançon ure models (Robin et al. 2003) to count background/foreground stars in the allowed S magnitude, taking into account the star counts and the probability of each scenario j I they compute j a false the star-transiting planet scenario. If this false alarm rate is small enough, the planet j is said to ressin et al. 2011). > of the dataset, computed without including wrms Kepler SED RV V span FWHM [ppm] [mmag] [m.s 1 ] [m.s 1 ] [m.s 1 ] Scenario 1 301 142 37 68 113 Scenario 2 303 136 43 70 109 Scenario 3 303 139 51 65 319 < > 264 60 26 52 103 4.5. Bayesian statistical comparison of the scenarios and planet validation We analysed in Sects. 4.1, 4.2, 4.3, and 4.4 the same dataset considering four di erent scenarios. To quantify which scenario is best supported by the data, we computed for each pair of scenarios the odds ratio O ij between the scenarios i and j, as defined in the Bayesian statistics. O ij = (3) (4) S i, I P S j, I P bound & unbound D i, S i, I d i, (5) D j, S j, I d j presented in Tuomi & Jones (2012) an (2014). However, since the four scen have nearly the same number of free p these limitations do not significantly a The probability distribution of th for scenario 1 against all other scen 9 (upper panel) and given in Table 5 strong evidence (as defined by Kass odds ratio greater than 150), or decisiv Je reys 1961, for an odds ratio grea 1 compared with scenarios 0, 2 and rejected in favour of scenario 1 (see can easily be explained by the fac reproduce the observed FHWM. Sce supported by the data than scenario 5). This might be surprising since sc data quite well. However, to reprod the transit light curve and the bisecto fine-tuned, as illustrated by the unce stellar mass (nearly 5%) or the secon These statistical uncertainties are inde star. Since scenario 2 needs a fine-tu reproduce the data, it is therefore le which requires much less fine-tuning penalised by the Occam s razor, as of Gregory (2005). Finally, scenario observed drift in the FWHM and i than scenario 1 which does reproduce Table 5). Moreover, we expect the odd scenarios to significantly increase as obtained with a longer timespan, by in or not, of the FWHM variation. The lower panel of figure 9 show scenario, assuming that 3X
of the two transits. Transit models (red curves) smoothed to the 29.426-min cadence are overplotted. These two signals are unambiguously detected in each of the eight quarters The planet-validation technique, in short of Kepler data, and have respective signal-to-noise ratios of 23.6 and 18.5, which cannot be due to stellar variability, data treatment or aliases from the other transit signals 4. Besançon Galactic Model data constraints Fressin et al. (2012)
of the two transits. Transit models (red curves) smoothed to the 29.426-min cadence are overplotted. These two signals are unambiguously detected in each of the eight quarters The planet-validation technique, in short of Kepler data, and have respective signal-to-noise ratios of 23.6 and 18.5, which cannot be due to stellar variability, data treatment or aliases from the other transit signals 4. Besançon Galactic Model data constraints Fressin et al. (2012) Probability of each astrophysical scenario
of the two transits. Transit models (red curves) smoothed to the 29.426-min cadence are overplotted. These two signals are unambiguously detected in each of the eight quarters The planet-validation technique, in short of Kepler data, and have respective signal-to-noise ratios of 23.6 and 18.5, which cannot be due to stellar variability, data treatment or aliases from the other transit signals 4. Besançon Galactic Model data constraints Fressin et al. (2012) Probability of each astrophysical scenario If the planet scenario is significantly the most likely scenario validated planet
PASTIS: Díaz, Almenara, Santerne, et al. 2014, MNRAS, 441, 983 Santerne, Díaz, Almenara, et al., 2015, in prep. Planet Analysis and Small Transit Investigation Software
Díaz, Almenara, Santerne, et al. 2014, MNRAS, 441, 983 Santerne, Díaz, Almenara, et al., 2015, in prep. PASTIS: Planet Analysis and Small Transit Investigation Software In numbers: ~ 23 000 lines of code Computation time needed (4yrs of Kepler data): 1k - 5k CPU hours (planet analysis) 5k - 25k CPU hours (validation) Python (v2.7) code with C++, fortran, and IDL routines eq. ~3x full-time PhD / post-doc for 3 years (2010 - now)
Today s view of PLATO follow-up, step by step Validation only planned on RV-boring stars* *RV-boring stars: stars not suitable for precise RV observations, thus not suitable for precise planet mass determination
Planet-validation tools: optimising the follow-up of PLATO PLATO products: LC, Δα, Δδ, Δν Compute probability of each scenario GAIA products: π, L, Av, BG density Other data: Spectra, RVs,... 1. Define the parameter space compatible with the PLATO data for all scenarios 2. Define which follow-up observation is the most efficient to further constrain them Bound Unbound
dots). The Kepler photometry phase-binned in 30-min intervals (blue dots with 1 standard error of the mean (s.e.m.) error bars) for Kepler-20 e (a) and Kepler-20 f (b) is Example of how to optimise the follow-up of PLATO with displayed as a function of time, with the data detrended 4 and phase-folded at the period PASTIS of the two transits. Transit models (red curves) smoothed to the 29.426-min cadence are overplotted. These two signals are unambiguously detected in each of the eight quarters of Kepler data, and have respective signal-to-noise ratios of 23.6 and 18.5, which cannot be due to stellar variability, data treatment or aliases from the other transit signals 4. Kepler-20e Fressin et al. (2012)
dots). The Kepler photometry phase-binned in 30-min intervals (blue dots with 1 standard error of the mean (s.e.m.) error bars) for Kepler-20 e (a) and Kepler-20 f (b) is Example of how to optimise the follow-up of PLATO with displayed as a function of time, with the data detrended 4 and phase-folded at the period PASTIS of the two transits. Transit models (red curves) smoothed to the 29.426-min cadence are overplotted. These two signals are unambiguously detected in each of the eight quarters of Kepler data, and have respective signal-to-noise ratios of 23.6 and 18.5, which cannot be due to stellar variability, data treatment or aliases from the other transit signals 4. HARPS SPHERE Kepler-20e Fressin et al. (2012)
dots). The Kepler photometry phase-binned in 30-min intervals (blue dots with 1 standard error of the mean (s.e.m.) error bars) for Kepler-20 e (a) and Kepler-20 f (b) is Example of how to optimise the follow-up of PLATO with displayed as a function of time, with the data detrended 4 and phase-folded at the period PASTIS of the two transits. Transit models (red curves) smoothed to the 29.426-min cadence are overplotted. These two signals are unambiguously detected in each of the eight quarters of Kepler data, and have respective signal-to-noise ratios of 23.6 and 18.5, which cannot be due to stellar variability, data treatment or aliases from the other transit signals 4. HARPS SPHERE Kepler-20e Fressin et al. (2012)
dots). The Kepler photometry phase-binned in 30-min intervals (blue dots with 1 standard error of the mean (s.e.m.) error bars) for Kepler-20 e (a) and Kepler-20 f (b) is Example of how to optimise the follow-up of PLATO with displayed as a function of time, with the data detrended 4 and phase-folded at the period PASTIS of the two transits. Transit models (red curves) smoothed to the 29.426-min cadence are overplotted. These two signals are unambiguously detected in each of the eight quarters of Kepler data, and have respective signal-to-noise ratios of 23.6 and 18.5, which cannot be due to stellar variability, data treatment or aliases from the other transit signals 4. HARPS SPHERE Kepler-20e Fressin et al. (2012)
dots). The Kepler photometry phase-binned in 30-min intervals (blue dots with 1 standard error of the mean (s.e.m.) error bars) for Kepler-20 e (a) and Kepler-20 f (b) is Example of how to optimise the follow-up of PLATO with displayed as a function of time, with the data detrended 4 and phase-folded at the period PASTIS of the two transits. Transit models (red curves) smoothed to the 29.426-min cadence are overplotted. These two signals are unambiguously detected in each of the eight quarters of Kepler data, and have respective signal-to-noise ratios of 23.6 and 18.5, which cannot be due to stellar variability, data treatment or aliases from the other transit signals 4. HARPS SPHERE Only SPHERE is relevant to constrain this scenario Both HARPS and SPHERE might be useful to constrain this scenario Kepler-20e Fressin et al. (2012)
Role of planet-validation tool in PLATO The PASTIS planet-validation tool could evaluate the probability of all astrophysical scenarios and constrain their parameters.
Role of planet-validation tool in PLATO The PASTIS planet-validation tool could evaluate the probability of all astrophysical scenarios and constrain their parameters. These constraints could be used to define the best follow-up strategy to rule out the false positives and secure the planet detection.
Role of planet-validation tool in PLATO The PASTIS planet-validation tool could evaluate the probability of all astrophysical scenarios and constrain their parameters. These constraints could be used to define the best follow-up strategy to rule out the false positives and secure the planet detection. Then, it would be possible to quantify the probability of a given observation / given instrumentation to detect a false-positive scenario, if the candidate is not a planet.
Limitations of planet-validation tools CPU-intensive Need reduced light-curves (no stellar activity, no TTVs, no instrumental effects) otherwise it takes even more time. More efficient with GAIA products, stellar mean density from asteroseismology, complete list of BG contaminants (PIC-CC)
Limitations of planet-validation tools CPU-intensive Need reduced light-curves (no stellar activity, no TTVs, no instrumental effects) otherwise it takes even more time. More efficient with GAIA products, stellar mean density from asteroseismology, complete list of BG contaminants (PIC-CC) Could be used at least on the most expensive candidates + RV-boring stars
To-Do list (for the next 10 ~9.1 years) Improve & fasten the code (some ideas - on going) Make it more automatic Include more observable (e.g. centroids) Characterise the follow-up instruments [WG 14x] Define priorities [WG 113]
The role of amateur astronomers: even decrease (professional) telescope time Possible role of amateurs: screen out EBs 1995 2014 characterise giant planets characterise EB of 2m telescope + ELODIE 27cm telescope + home-made R=50k spectro CBP Amateurs have almost unlimited access of telescope time Buil, Santerne et al., in prep.
Take-home messages Planet-validation tools like PASTIS could optimise the follow-up observations CPU-time consuming unlikely to be applied to all POIs (PLATO Objects of Interest) could be applied only to the most expensive planets (small & cool planets) Amateur astronomers could also participate substantially to the (RV) follow-up efforts.