Statistical validation of PLATO 2.0 planet candidates Rodrigo F. Díaz Laboratoire d Astrophysique de Marseille José Manuel Almenara, Alexandre Santerne, Claire Moutou, Anthony Lehtuillier, Magali Deleuil PLATO 2.0 Science Workshop. ESA ESTEC, July 30th 2013.
Outline Planet Validation. Motivation in the context of PLATO 2.0 The Planet Analysis and Small Transit Investigation Software (PASTIS). Testing PASTIS on synthetic data. The radial velocity contribution.
10 3 Expected RV semi-amplitude [m/s] 10 2 10 1 10 0 10 1 10 2 SOPHIE 2km/s 5km/s 10km/s HARPS 2km/s 5km/s 10km/s 12 13 14 15 16 Magnitude Talks by F. Bouchy and A. Santerne
Planet Validation (alla BLENDER) Use all the information in the transit LC to constrain possible false positives (FPs). M2 = 1.0 Msun; d = 5mag Add additional constrains from other datasets: RV, AO, multi-band photometry,... Evaluate relative occurrence of planets to surviving blends (use Galactic models, current knowledge on mult. systems, etc.) Fressin et al. (2011) Spitzer
Planet Validation (the Bayesian way) Model comparison: based on the computation of the odds ratio. O ij = p(h i D, I) p(h j D, I) = p(h i I) p(h j I) p(d H i, I) p(d H j, I) Model prior ratio Bayes factor Hypotheses must be described by a model M with parameter vector θ. The Bayes factor is the ratio of the evidence for each model, defined as: p(d H, I) p(d M) = dθ p(θ M)p(D θ, M) Hi: hypothesis i D: data I: prior information The evidence is a k-dimensional integral, generally intractable!
Motivation in the context of PLATO 2.0 PLATO is expected to detect thousands of small-size planet candidates. Validation will be needed (unless reliance on Galactic priors is acceptable). RV confirmation is in principle be possible, but very time-demanding (see HARPS RV survey; talk by S. Udry). RV facilities likely limited to perform intense follow-up. Some stars will not be easily measured (fast rotators, hot stars,...). High S/N of many candidates (e.g. S/N ~ 150 for a single Earth-size planet transit on 1-yr orbit around a mv = 8 star). Possible strategy: focus on validated planets to complete characterization (mass, eccentricity, bulk density,...).
Planet Analysis and Small Transit Investigation Software Rigorousness (fully-bayesian approach) Flexibility (in the definition of FP scenarios) Speed (to be able to apply it to large samples)
The PASTIS data models Light curves (Kepler, CoRoT, Spitzer,...) GIE (Amores & Lepine) LDC (Claret) MCMC Full parameter posterior Spectral Energy Distribution (SDSS, 2MASS, WISE,...) Radial velocities RV, Bisector, FWHM (HARPS, SOPHIE) Model priors (Binary prop., planet prop.,...) SET (Dartmouth, Padova, Geneva,...) SAM (PHOENIX, ATLAS) Full details in Díaz et al. (2013, in prep.) Kepler1 jitter 0.000102 kr q 9.5e-05 8.7e-05 albedo2 ar incl 0.19 0.12 0.06 8.45 8.35 8.25 89.5 89.0 88.5 0.0702 0.0698 0.0694 0.0052 0.0035 0.0018 ua1 0.53 0.5 0.47 1.000004 1.000012 Kepler1 foot 1.000021 8.7e-05 Bayes factor O ij = p(h i D, I) p(h j D, I) = p(h i I) p(h j I) p(d H i, I) p(d H j, I) 9.5e-05 Kepler1 jitter 0.000102 0.06 0.12 0.19 albedo2 8.25 8.35 ar 8.45 BTPastis9 PLANET b0.00 Rpl0.0716 Mp1.00 SNR500 scenarioplanet Beta1.000000 88.5 89.0 incl 89.5 0.0694 0.0698 kr 0.0702 0.0018 0.0035 q 0.0052
Tests on synthetic data
Synthetic data Kepler data KIC11391018 KOI189 Q4; SC Noise level near the median for SC targets of Q4. Planet transiting candidate. Already checked. For simplicity, just use light curve. In some cases the other observables dominates (cf. CoRoT-16, Ollivier et al. 2012).
Synthetic data Kepler data KIC11391018 KOI189 Q4; SC Model P: the signal is produced by a transiting extrasolar planet. Model B: the signal is produced by a background eclipsing binary. + data model: jitter is an additional source of Gaussian error.
Synthetic data Transiting Planets Studied parameters able 4. Parameters for synthetic light curves Transiting Planet Planet Radius [R ] {1.0; 4.4; 7.8; 11.2} Impact Parameter b {0.0; 0.5; 0.75} Transit S/N {20; 50; 150; 500} Background Eclipsing Binary Mass Ratio {0.1; 0.3; 0.5} Impact Parameter b {0.0; 0.5; 0.75} Secondary S/N {2; 5; 7} Transit S/N independently varied from planet radius. b = 0.75 near median of KOI distribution..3. Modeling of systematic effects in the data Planet Radius n addition, we use a simple model of any potential systematic rrors in the data not accounted for in the formal uncertainties. e follow in the steps of Gregory (2005a), and model the addiional jitter noise of our data as a Gaussian-distributed variable Relative Flux 1.010 1.005 1.000 0.995 0.990 0.985 0.980 1.004 1.002 Table 5. Jump parameters and priors of the planet and BEB models use 1.000 to fit the synthetic light curves. 0.998 0.996 0.994 PLANET 0.992 model 0.990 k r = R 2 /R 1 Jeffreys(10 3, 0.5) 1.001 a R = a/r 1 Jeffreys(2.0, 100.0) Planet 1.000 albedo Uniform(0.0, 1.0) Mass ratio q Jeffreys(10 7,8 10 3 ) 0.999 Orbital 0.998 inclination i [deg] Sine(80, 90) Linear limb darkening coefficient Uniform(0.0, 1.0) 0.997 BEB model 1.00005 Primary M init [M ] Jeffreys(0.1, 10) Secondary M init [M ] Jeffreys(0.1, 5) Binary age Binary z [dex] Uniform(-2.5, 0.5) Binary distance [pc] Uniform(10, 10 4 ) 1.00000 0.99995 0.99990 0.99985 S/N = 50; b = 0.5 0.04 0.02 0.00 0.02 0.04 Impact parameter b Orbital Phase Uniform(0.0, 1.0) Target T eff [K] Normal(5770, 100)
Synthetic data S/N = 150; b = 0.5 Transiting Planets 1.000 0.995 0.990 Studied parameters able 4. Parameters for synthetic light curves Transiting Planet Planet Radius [R ] {1.0; 4.4; 7.8; 11.2} Impact Parameter b {0.0; 0.5; 0.75} Transit S/N {20; 50; 150; 500} Background Eclipsing Binary Mass Ratio {0.1; 0.3; 0.5} Impact Parameter b {0.0; 0.5; 0.75} Secondary S/N {2; 5; 7} Transit S/N independently varied from planet radius. b = 0.75 near median of KOI distribution..3. Modeling of systematic effects in the data Planet Radius n addition, we use a simple model of any potential systematic rrors in the data not accounted for in the formal uncertainties. e follow in the steps of Gregory (2005a), and model the addiional jitter noise of our data as a Gaussian-distributed variable Relative Flux 1.001 1.000 0.999 0.998 0.997 0.996 0.995 0.994 0.993 1.0005 Table 5. Jump parameters and priors of the planet and BEB models use to fit the synthetic light curves. PLANET model k r = R 2 /R 1 Jeffreys(10 3, 0.5) 1.0000 a R = a/r 1 Jeffreys(2.0, 100.0) 0.9995 Planet albedo Uniform(0.0, 1.0) 0.9990 Mass ratio q Jeffreys(10 7,8 10 3 ) 0.9985 Orbital inclination i [deg] Sine(80, 90) Linear limb darkening coefficient Uniform(0.0, 1.0) 0.9980 1.00002 BEB model 1.00000 Primary M init [M ] Jeffreys(0.1, 10) Secondary M init [M ] Jeffreys(0.1, 5) Binary age Binary z [dex] Uniform(-2.5, 0.5) Binary distance [pc] Uniform(10, 10 4 ) 0.99998 0.99996 0.99994 0.99992 0.99990 0.04 0.02 0.00 0.02 0.04 Impact parameter b Orbital Phase Uniform(0.0, 1.0) Target T eff [K] Normal(5770, 100)
Synthetic data S/N = 500; b = 0.5 Transiting Planets Studied parameters able 4. Parameters for synthetic light curves Transiting Planet Planet Radius [R ] {1.0; 4.4; 7.8; 11.2} Impact Parameter b {0.0; 0.5; 0.75} Transit S/N {20; 50; 150; 500} Background Eclipsing Binary Mass Ratio {0.1; 0.3; 0.5} Impact Parameter b {0.0; 0.5; 0.75} Secondary S/N {2; 5; 7} Transit S/N independently varied from planet radius. b = 0.75 near median of KOI distribution..3. Modeling of systematic effects in the data Planet Radius n addition, we use a simple model of any potential systematic rrors in the data not accounted for in the formal uncertainties. e follow in the steps of Gregory (2005a), and model the addiional jitter noise of our data as a Gaussian-distributed variable Relative Flux 1.000 0.998 0.996 0.994 0.992 0.990 0.988 1.000 Table 0.9995. Jump parameters and priors of the planet and BEB models use to fit 0.998 the synthetic light curves. 0.997 0.996 PLANET model 0.995 k1.0000 r = R 2 /R 1 Jeffreys(10 3, 0.5) a R = a/r 1 Jeffreys(2.0, 100.0) Planet 0.9995 albedo Uniform(0.0, 1.0) Mass 0.9990 ratio q Jeffreys(10 7,8 10 3 ) Orbital inclination i [deg] Sine(80, 90) 0.9985 Linear limb darkening coefficient Uniform(0.0, 1.0) BEB model 1.00000 0.99998 Primary M init [M ] Jeffreys(0.1, 10) Secondary M init [M ] Jeffreys(0.1, 5) Binary age Binary z [dex] Uniform(-2.5, 0.5) Binary distance [pc] Uniform(10, 10 4 ) 0.99996 0.99994 0.99992 0.04 0.02 0.00 0.02 0.04 Impact parameter b Orbital Phase Uniform(0.0, 1.0) Target T eff [K] Normal(5770, 100)
Results (I: PLANET simulations) b = 0.0 b = 0.50 b = 0.75 B10 < 3: inconclusive evidence 3< B10 < 20: positive evidence 20 < B10 < 150: strong evidence B10 > 150: very strong evidence
B10 < 3: inconclusive evidence Results (I: PLANET simulations) 3< B10 < 20: positive evidence 20 < B10 < 150: strong evidence B10 > 150: very strong evidence b = 0.0 b = 0.50 b = 0.75 Monotonic decrease of BPB with planet radius, and S/N of transit. Minimum BPB for b = 0.5. For b = 0.75, highest value of BPB. S/N = 20 and 500 (not plotted) are completely inside and outside of shaded are, respectively.
Synthetic data q = 0.3; b = 0.5 1.0005 1.0005 Background eclipsing 1.0000 1.0000 binaries Studied parameters { } { } Dilution 0.9995 0.9990 1.0005 0.9995 0.9990 0.04 0.02 0.00 0.02 0.04 1.0005 0.46 0.48 0.50 0.52 0.54 Background Eclipsing Binary Mass Ratio {0.1; 0.3; 0.5} Impact Parameter b {0.0; 0.5; 0.75} Secondary S/N {2; 5; 7} Relative Flux 1.0000 0.9995 Relative Flux 1.0000 0.9995 0.9990 0.9990 Secondary S/N: a way to quantify the dilution provided by the foreground target star. 1.0005 1.0000 0.9995 0.9990 0.04 0.02 0.00 0.02 0.04 1.0005 1.0000 0.9995 0.9990 0.46 0.48 0.50 0.52 0.54 Secondary S/N 0.04 0.02 0.00 0.02 0.04 Orbital Phase 0.46 0.48 0.50 0.52 0.54 Orbital Phase
B10 < 3: inconclusive evidence Results (II: BEB simulations) 3< B10 < 20: positive evidence 20 < B10 < 150: strong evidence B10 > 150: very strong evidence b = 0.0 b = 0.50 b = 0.75 Monotonic increase of BBP with mass ratio q, snr of the secondary, and decrease with impact parameter. BEB model is not strongly chosen for highest dilution level at any q or snr. Secondaries with snr = 7 lead to strong evidence for the BEB hypothesis (except q = 0.1; b = 0.75)
The radial velocity contribution Could RV measurements help us decide on some cases? Nordström et al. (2004) distribution of RV measurements in the Solar neighborhood used to BEB velocity. Velocity and bisector amplitudes are computed. Best-fit BEB model to PLANET simulations; RV amplitude clearly detectable.
Summary and Conclusions Planet validation is the only technique to establish the planetary nature of the smallest transiting candidates from CoRoT and Kepler. PASTIS correctly identifies both transiting planets and false positives if the signal is sufficiently high. Simulations of synthetic data show that BEB with a modest dilution are easily detectable as such. However, if the dilution is such that the secondary eclipse has signal-to-noise ratio ~ 2, the data cannot say much. For simulated PLANETS, only data from very-high-s/n transits provides strong support for the planet scenario. For the unresolved cases, the model priors will be the deciding factor. Other observations: AO, etc. Radial velocities can contribute in this point. Apparently more so in the case of planets than BEBs.