[abridged] We carry out numerical simulations to gauge the Gaia potential for precision astrometry of exoplanets orbiting a sample of known dM stars within 30 pc from the Sun. (1) It will be possible to accurately determine orbits and masses for Jupi
ter-mass planets with orbital periods in the range 0.2<P<6.0 yr and with an astrometric signal-to-noise ratio > 10. Given present-day estimates of the planet fraction f_p around M dwarfs, 100 giant planets could be found by Gaia around the sample. Comprehensive screening by Gaia of the reservoir of 4x10^5 M dwarfs within 100 pc could result in 2600 detections and as many as 500 accurate orbit determinations. The value of f_p could then be determined with an accuracy of 2%, an improvement by over an order of magnitude with respect to the most precise values available to-date; (2) in the same period range, inclination angles corresponding to quasi-edge-on configurations will be determined with enough precision (a few percent) so that it will be possible to identify intermediate-separation planets which are potentially transiting within the errors. Gaia could alert us of the existence of 10 such systems. More than 250 candidates could be identified assuming solutions compatible with transit configurations within 10% accuracy, although a large fraction of these (85%) could be false positives; (3) for well-sampled orbits, the uncertainties on planetary ephemerides, separation and position angle, will degrade at typical rates of < 1 mas/yr and < 2 deg/yr, respectively; (4) planetary phases will be measured with typical uncertainties of several degrees, resulting (under the assumption of purely scattering atmospheres) in phase-averaged errors on the phase function of 0.05, and expected uncertainties in the determination of the emergent flux of intermediate-separation (0.3<a<2.0 AU) giant planets of 20%. [abridged]
We use detailed simulations of the Gaia observations of synthetic planetary systems and develop and utilize independent software codes in double-blind mode to analyze the data, including statistical tools for planet detection and different algorithms
for single and multiple Keplerian orbit fitting that use no a priori knowledge of the true orbital parameters of the systems. 1) Planets with astrometric signatures $alphasimeq 3$ times the single-measurement error $sigma_psi$ and period $Pleq 5$ yr can be detected reliably, with a very small number of false positives. 2) At twice the detection limit, uncertainties in orbital parameters and masses are typically $15%-20%$. 3) Over 70% of two-planet systems with well-separated periods in the range $0.2leq Pleq 9$ yr, $2leqalpha/sigma_psileq 50$, and eccentricity $eleq 0.6$ are correctly identified. 4) Favorable orbital configurations have orbital elements measured to better than 10% accuracy $> 90%$ of the time, and the value of the mutual inclination angle determined with uncertainties $leq 10^{degr}$. 5) Finally, uncertainties obtained from the fitting procedures are a good estimate of the actual errors. Extrapolating from the present-day statistical properties of the exoplanet sample, the results imply that a Gaia with $sigma_psi$ = 8 $mu$as, in its unbiased and complete magnitude-limited census of planetary systems, will measure several thousand giant planets out to 3-4 AUs from stars within 200 pc, and will characterize hundreds of multiple-planet systems, including meaningful coplanarity tests. Finally, we put Gaia into context, identifying several areas of planetary-system science in which Gaia can be expected to have a relevant impact, when combined with data coming from other ongoing and future planet search programs.
