Carl Tannler

Astro 420 Exoplanet Detection through Radial Velocity Measurements

There are many methods that have been developed in order to detect exoplanets. One of the most prominent methods used today is the radial velocity detection method. The theory behind this method is that while a planet is orbiting around a star, the star will move in its own small orbit in response to the gravity of the orbiting planet, and due to conservation of momentum. This will create small variations in the stars radial velocity with respect to Earth. When the spectral lines are determined, the radial velocity of the star can be determined via the Doppler Effect. The variations in the radial velocity measurements are used to confirm the presence of an orbiting planet. Even though the velocity of the star around the center of mass is very small compared to that of the planet, modern spectrometers such as the High Accuracy Radial Velocity Planet Searcher (HARPS Spectrometer) can measure velocity variations down to 1 m/s or less. Using the periodic radial velocity variations, it is possible to determine six parameters in the spectroscopic orbit of the star (Wright. 2012) that will determine the size, shape, and distance of the stars orbit. Among the six parameters determined from the radial reflex motion of a star, information about the orbiting planets mass can also be determined. If there is more than one orbiting planet in the system, then the number of Keplerian parameters of the system determined from the radial velocity data is 5n+1 (Wright and Howard 2009), where n is the number of planets in the system. The other parameters include: P, the period of the planets orbit; K, the semiamplitude of the radial velocity signal; e, the eccentricity of the orbit; , the argument of periastron; tp, a date of periastron passage; and , the apparent radial velocity of the center of mass of the system.

When looking at the radial velocity signal of a planet, it turns out to have the form, Vr = KF(t,e,*,T0,P), where * is the longitude of periastron of the star, and T0 is the phase of the radial velocity curve. When taking all of these parameters in account, and also including the inclination of the system, a relation between the physical

PK 3 (1 e 2 ) properties of the system is, 2G

M 3 sin 3 i p (M p + M * ) 2

Mp is the mass of the planet, M* is the mass of the star, and i is the inclination angle. The right side of the equation is known as the mass function of the system. If one assumes a uniform and dense sampling of a radial velocity curve over a time that is long compared to the period P, with total number of observations, N, each with uncertainty RV, the total signal-to-noise ratio can be determined,

( S / N ) RV g (e, * ) N

K RV

When the orbit of a planet is circular, g = 2-1/2, but is usually a function of e for e 0.6. When eccentricities increase, g declines steadily. When the planets have a period larger than the duration of the observations, the detectability depends also on the period and
1 phase of the planet, and generally decreases with increasing period, ( S / N ) RV P

(Wright 2012; Eisner & Kulkarni 2001; Cumming 2004). This means that typically, detecting a planet via radial velocity requires a precision of RV << KN1/2, and for Mp <<

 P M*, the semiamplitude K is, K = 2G

M p sin i M*
2 3

(1 e 2 )

. As an example, to detect

a Jupiter-mass planet in a 11.8 yr orbit, with e = 0, around a solar mass star, which yields K (12.5 m/s) sini, will take a few dozen observations with a precision of a few m/s (Wright 2012). One method that is used for both detecting and characterizing the orbits of extrasolar planets is a Bayesian Markov chain Monte Carlo (MCMC) algorithm (Gregory 2005), or the Bayesian analysis. An MCMC can explore all parameter possibilities that have significant probability, which eliminates the need to carry out a separate search for orbital periods. The Bayesian analysis will yield, depending on the number of parameters contained in the model, the relative probability of each of the models in question. The Bayesian analysis looks at two questions of interest, the number of planets involved in the stars reflex motion, and the orbital parameters of each planet detected. The ground work for Bayesian analysis is Bayes theorem which is p(Mi|D, I) = p(Mi|I)p(D|Mi, I)/p(D|I), where p(Mi|D, I) is the posterior probability, or the probability under relevant evidence, that model Mi is true, given data D and background information I. As an example of the Bayesian analysis, let M0 be the model for a star moving away or toward an observer with an unknown velocity. Using the radial velocities published by Tinney et al. (2003), an arbitrary zero point is determined by a template observation, and then a constant velocity V is assumed for M0 with 1 parameter. Next, model M0s assumes an extra noise term in addition to the known uncertainties, with 1+s parameters. The extra noise term is added incase the star actually has two planets instead

of the assumed one planet. Now let M1 be a model with non constant V + an elliptical orbit and 6 parameters, M1s is a non constant V + an elliptical orbit + an extra noise term with 6+s parameters, and M2 has V + 2 elliptical orbits with 11 parameters, and the trend follows as more orbits are assumed. Then to usefully claim detection of a planet, one wants p(M1|D, I) > both p(M0|D, I) and p(M0s|D, I). If p(M1|D, I) > p(M0|D, I) but not > p(M0s|D, I), then the noise component doesnt account for the reflex motion due to a planet. If detection of a second planet is to be determined, then p(M2|D, I) > both p(M1| D, I) and p(M1s|D, I). Using this information, the predicted radial velocity fi , for one planet and 6 unknowns is fi = V + K{cos[(ti + P) + ] + ecos}. V is a constant velocity, K is the

velocity amplitude which is equal to

2a sin i P (1 e 2 )
1 2

, where a is the semimajor axis of the

orbit, and is the fraction of the orbit, prior to data collecting, at which periastron occurred. This means that P is the number of days prior to ti = 0, and (ti + P) is known as the true anomaly. This method is capable of fitting a portion of an orbital period, which means that search periods longer than the duration of the data are possible. The MCMC algorithm used in this Bayesian analysis can result in huge savings in time by eliminating many trial runs to manually establish a proposed set of uncertainties. If one uses the data from Tinney et al. (2003), and the Bayesian analysis, it can be shown that the one-planet model M1s has the highest probability. The model, M1s, finds three possible orbits with periods of 128, 190, and 376 days, and also shows that the 128 day orbit has the smallest residuals, and the 376.2 day orbit is the most probable. The original data in Tinney et al. (2003) has the 190 day orbit as the most probable, which

with the Bayesian analysis is the least probable. This means that the Bayesian analysis of radial velocity data has better results when determining the orbital parameters of planets. When collecting data on systems in question, typical stellar spectral lines detected by stars, have the centroid at the fixed equivalent width and the centroid can be measured
32 12 with a precision of V N eff (Wright 2012), where V is the effective velocity width

of the spectral line, and Neff is the effective number of photons in the line. In order to maximize the precision, the lines need to be well-resolved, which means that the instrumental velocity resolution has to be less than the intrinsic velocity width of the star. A typical width of a spectral feature from a slowly rotating star is of order a few km/s meaning that the needed resolving powers are R = / 105, which is comparable to the resolving power of a typical high resolution astronomical echelle spectrograph. Echelle is a type of grating used in spectrographs such as HARPS. Usually, the velocity precision per line is insufficient to detect planets, so averaging over many lines is needed. This also means that the wavelength calibration must be more precise than the velocity precisions, and it must be stable over many times the orbital period. An example of the radial velocity detection method comes from data collected from HARPS, and the detection of two super-Earth planets in the G1 581 system which had already been known to harbour a hot Neptune around an M3 dwarf (Udry 2008). M dwarfs are good candidates for high precision radial velocity searches because the lower mass makes detection of light planets easier. At first, 20 high resolution HARPS spectra of G1 581 were available, and the residuals from a 1-planet keplerian solution showed a tentative 2nd signal at a frequency of 1/13 d-1. Since there were only a limited number of observations, the low amplitude of the 13-day velocity variation had only modest

significance which then prompted 30 more high resolution observations using HARPS. Using all 50 high-precision HARPS radial velocities confirmed a 5.36-d period planet. The 5.36-d period was modeled poorly by a single keplerian planet model with high residuals. This and the 13-days periodogram peak gives motivation for the investigation of a 2-planet model. The 2-planet model gives a second planet with an eccentricity e 0.28 0.06, with a period of 12.895 days. The 2nd planet has a mass of about 5.03 M , with a semimajor axis a = 0.073 AU, and planetary radius of about 1.5 R , orbiting just outside of the hot Neptune. Since the periodogram has a significant power around 84 days, a 3-planet model was examined in order to prove the existence of a third planet. The third planet model changes the parameters of the hot Neptune, and the super-Earth slightly (lower eccentricities). The third planet then has an 83.6 day period, an eccentricity of e = 0.2, a mass of 7.7 M , and a semimajor axis a = 0.25 AU. This puts the planet close to the out edge of the habitable zone of the system. With the addition of this 3rd planet, 5 free parameters were added lowering residuals and giving accurate results for planetplanet gravitational interactions. Thus making the 3-planet models the most accurate for the system. Overall the radial velocity detection method has been extremely successful as is shown by the large number of exoplanets detected with this method. Spectrometers with high resolution are needed to obtain adequate information to prove the presence of a planet. Doing an in depth analysis of the spectral lines of the star and the reflex variations, planet models can be created and tested. The Bayesian analysis is an excellent method to test planet models around systems. The radial velocity detection method has

been the most successful exoplanet detection method and will continue to be until another method arises that can surpass it.

Sources 1. Cumming, A. 2004, MNRAS, 354, 1165 2. Cumming, A., 2008, arXiv:astro-ph/0408470 3. Eisner, J. A., & Kulkarni, S. R. 2001, ApJ, 550, 871 4. Gregory, P. C., 2005, ApJ, 1198, 1214 5. Tinney, C. G., Butler, R. P., Marcy, G. W., Jones, H. R. A., Laughlin, G., Carter, B. D., Bailey, J. A., OToole, S., 2006, ApJ, 594, 599 6. Udry, S., Bonfils, X., Delfosse, X., Forveille, T., Mayor, M., Perrier, C., Bouchy, F., Lovis, C., Pepe, F., Queloz, D., Bertaux J. L., 2008, arXiv:0704.3841 7. Wright, J. T., Gaudi, B. S. 2012, arXiv:1210.2471 8. Wright, J. T., Howard, A. D., 2009, ApJ, 205, 215