A&A 474, 293-299 (2007)
X. Bonfils1 - M. Mayor2 - X. Delfosse3 - T. Forveille3 - M. Gillon2 - C. Perrier3 - S. Udry2 - F. Bouchy4 - C. Lovis2 - F. Pepe2 - D. Queloz2 - N. C. Santos1,2,5 - J.-L. Bertaux6
1 - Centro de Astronomia e Astrofísica da Universidade de Lisboa, Observatório Astronómico de Lisboa, Tapada da Ajuda, 1349-018 Lisboa, Portugal
2 - Observatoire de Genève, 51 ch. des Maillettes, 1290 Sauverny, Switzerland
3 - Laboratoire d'Astrophysique, Observatoire de Grenoble, Université J. Fourier, BP 53, 38041 Grenoble, Cedex 9, France
4 - Institut d'Astrophysique de Paris, CNRS, Université Pierre et Marie Curie, 98bis Bd Arago, 75014 Paris, France
5 - Centro de Astrofísica, Universidade do Porto, Rua das Estrelas, 4150-762 Porto, Portugal
6 - Service d'Aéronomie du CNRS, BP 3, 91371 Verrières-le-Buisson, France
Received 9 January 2007 / Accepted 18 April 2007
Context. How planet properties depend on stellar mass is a key diagnostic of planetary formation mechanisms.
Aims. This motivates planet searches around stars that are significantly more massive or less massive than the Sun, and in particular our radial velocity search for planets around very low-mass stars.
Methods. As part of that program, we obtained measurements of GJ 674, an M 2.5 dwarf at d=4.5 pc. These measurements have dispersion much in excess of their internal errors. An intensive observing campaign demonstrates that the excess dispersion is due to two superimposed coherent signals, with periods of 4.69 and 35 days.
Results. These data are described well by a 2-planet Keplerian model where each planet has a 11 M minimum mass. A careful analysis of the (low-level) magnetic activity of GJ 674, however, demonstrates that the 35-day period coincides with the stellar rotation period. This signal therefore originates in a spot inhomogeneity modulated by stellar rotation. The 4.69-day signal, on the other hand, is caused by a bona-fide planet, GJ 674b. Conclusions. Its detection adds to the growing number of Neptune-mass planets around M-dwarfs and reinforces the emerging conclusion that this mass domain is much more populated than the Jovian mass range. We discuss the metallicity distributions of M dwarf with and without planets and find a low 11% probability that they are drawn from the same parent distribution. Moreover, we find tentative evidence that the host star metallicity correlates with the total mass of their planetary system.
Key words: stars: individual: GJ 674 - stars: planetary systems - stars: late-type - techniques: radial velocities
The first planet found to orbit an M dwarf, GJ 876b (Marcy et al. 1998; Delfosse et al. 1998), was only the 9th exoplanet discovered around a main sequence star. Besides showing that Jupiter-mass planets can form at all around very low-mass stars, its discovery suggested that they might be common, since it was found amongst the few dozen M dwarfs that were observed at that time. Despite these early expectations, no other M dwarf was reported hosting a planet until 2004, though a second planet (GJ 876c, - Marcy et al. 2001) was soon found around GJ 876 itself.
In 2004, the continuous improvement of the radial-velocity techniques resulted in the quasi-simultaneous discovery of three Neptune-mass planets around Ara ( - Santos et al. 2004), Cnc ( - McArthur et al. 2004), and GJ 436 ( - Maness et al. 2006; Butler et al. 2004). Of those three, GJ 436b orbits an M dwarf, which put that spectral class back on the discovery forefront. It was soon followed by another two: a single planet around GJ 581 ( - Bonfils et al. 2005b) and a very light ( ) third planet in the GJ 876 system (Rivera et al. 2005). As a result, planets around M dwarfs today represent a substantial fraction (30%) of all known planets with .
Even with GJ 849b ( - Butler et al. 2006) now completing the inventory of M-dwarf planets found with radial-velocity techniques, the upper range of planet masses remains scarcely populated. This contrasts both with the (still very incompletely known) Neptune-mass planets orbiting M dwarfs and with the Jovian planets around Sun-like stars. At larger separations, microlensing surveys similarly probe the frequency of planets as a function of their mass. That technique has detected four putative planets that are probably orbiting M dwarfs : OGLE235-MOA53b ( - Bennett et al. 2006; Bond et al. 2004), OGLE-05-071Lb ( - Udalski et al. 2005), OGLE-05-390Lb ( - Beaulieu et al. 2006), and OGLE-05-169Lb ( -Gould et al. 2006). Two of these four planets very likely have masses below 0.1 . Given the detection bias of that technique towards massive companions, this again suggests that Neptune-mass planets are much more common than Jupiter-mass ones around very low-mass stars.
Here we report the discovery of a 11 M planet orbiting GJ 674 every 4.69 days. GJ 674b has the 5 lowest mass of the known planets, and coincidentally is also the 5 planetary system centered on an M dwarf. Its detection adds to the small inventory of both very low-mass planets and planets around very low-mass stars. After reviewing the properties of the GJ 674 star (Sect. 2), we briefly present our radial velocity measurements (Sect. 3) and their Keplerian analysis (Sect. 4). A careful analysis of the magnetic activity of GJ 674 (Sect. 5) assigns one of the two periodicities to rotational modulation of a stellar spot signal and the other one to a bona fide planet. We conclude with a brief discussion of the properties of the detected planet.
Its photometry ( ; - Cutri et al. 2003; Turon et al. 1993) and parallax imply absolute magnitudes of and . GJ 674's J-K color (=0.86 - Cutri et al. 2003) and the Leggett et al. (2001) color-bolometric relation result in a K-band bolometric correction of BCK=2.67, and in a 0.016 L luminosity. The K-band mass-luminosity relation of Delfosse et al. (2000) gives a mass and the Bonfils et al. (2005a) photometric calibration of the metallicity results in . The moderate X-ray luminosity ( - Hünsch et al. 1999) and Ca II H & K emission depict a modestly active M dwarf (Fig. 1). Its UVW galactic velocities place GJ 674 between the young and old disk populations (Leggett 1992), suggesting an age of .
Last but not least, since we are concerned with radial velocities, the high proper motion of GJ 674 ( - ESA 1997) changes the orientation of its velocity vector along the line of sight (e.g. Kürster et al. 2003) to result in an apparent secular acceleration of . At our current precision, this acceleration will not be detectable before another decade.
|Figure 1: Emission reversal in the Ca II H line of GJ 674 ( top) and GJ 581 ( bottom). Within our sample, GJ 581 has one of the weakest Ca II emissions and illustrates a very quiet M dwarf. GJ 674 has much stronger emission and is moderately active.|
|Open with DEXTER|
Table 1: Observed and inferred stellar parameters for GJ 674.
We observed GJ 674 without interlaced Thorium-Argon light to obtain cleaner spectra for spectroscopic analysis, at some small cost in the ultimate Doppler precision. Since June 2004, we have gathered 32 exposures of 900 s each with a median S/N ratio of 90. Their Doppler information content, evaluated according to the prescriptions of Bouchy et al. (2001), is mostly below 1 . Our internal errors also include, in quadrature sum, an "instrumental'' uncertainty of for the nightly drift of the spectrograph (since we do not use the ThAr lamp to monitor it) and the measurement uncertainty of the daily wavelength zero-point calibration. We did benefit from the recent improvements in the HARPS wavelength calibration, which is now stable to (Lovis & Pepe 2007).
A constant radial velocity gives a very high reduced chi-square value ( ) for the time series, which reflects a dispersion ( ) well above our internal errors (Fig. 2). This prompted a search for an orbital (Sect. 4) and/or magnetic-activity (Sect. 5) signal.
|Figure 2: Upper panel: radial-velocity measurements of GJ 674 as a function of time. The high dispersion ( ) and chi-square value ( ) betray a (coherent or incoherent) signal in the data. Bottom panel: the Lomb-Scargle periodogram of the velocities has prominent power excess around P = 4.69 days (downward arrow), which indicates that much of the excess dispersion reflects a coherent signal with a period close to that value. The second-highest peak, at 1.27 day, is a one-day alias of the 4.69 day period ( ).|
|Open with DEXTER|
A Lomb-Scargle periodogram (Press et al. 1992) of the velocity measurements shows a narrow peak around 4.69-day (Fig. 2). Adjustment of a single Keplerian orbit demonstrates that it is best described by an M planet ( ) revolving around GJ 674 every days in a slightly eccentric orbit ( ). The residuals around this low-amplitude orbit ( ) have a dispersion of 3.27 (Fig. 3), still well above our measurement errors, and the reduced chi-square per degree of freedom is . A periodogram of the residuals indicates that much of this excess dispersion stems from a broad power peak centered around 35 days, prompting us to perform a 2-planet fit.
|Figure 3: Upper panel: radial velocities of GJ 674 (red filled circles) phase-folded to the 4.6940 days period of the best 1-planet fit (curve). The dispersion around the fit ( ) and its reduced chi-square ( per degree of freedom) indicate that a single planet does not describe the data very well. Middle panel: radial-velocity residuals of the 1-planet fit Bottom panel: the Lomb-Scargle periodogram of the residuals shows a broad peak centered around 35 days.|
|Open with DEXTER|
We searched for 2-planet Keplerian solutions with Stakanof (Tamuz, in prep.), a program that uses genetic algorithms to efficiently explore the large parameter space of multi-planet models. Stakanof quickly converged to a 2-planet solution that describes our measurements much better than the single-planet fit ( , per degree of freedom - Fig. 4). The orbital parameters of the 4.69-day planet change little from the 1-planet fit, except for the eccentricity, which increases to . Its mass is revised down to , and the period hardly changes, day. The second planet would have a day period, an eccentricity, and a minimum mass of . Such periods would correspond to semi-major axes of 0.04 and 0.15 AU. Those are enough disjointed that mutual interactions can be neglected over observable time scales, and that the system would be stable on longer time scales.
The low dispersion around the solution and the lack of any significant peak in the Lomb-Scargle periodogram of its residuals shows that our current radial-velocity measurements contain no evidence of an additional component.
|Figure 4: Top two panels: radial velocity measurements phased to each of the two periods, after subtraction of the other component of our best 2-planet Keplerian model. Third panel: residuals of the best 2-planet fit as a function of time (O-C, Observed minus Computed). Bottom panel: lomb-Scargle periodogram of these residuals.|
|Open with DEXTER|
|Figure 5: Upper panel: differential photometry of GJ 674 as a function of time. The star clearly varies with a 1.3% amplitude. Bottom panel: the periodogram of the GJ 674 photometry exhibits significant power excess peaking at 35 days (small black arrow).|
|Open with DEXTER|
We obtained photometric measurements with the CCD camera of the Euler Telescope (La Silla) during 21 nights between September 2 and October 19, 2006. GJ 674 was observed through a VG filter that, among those available, optimizes the flux ratio between GJ 674 and its two brightest reference stars. This relatively blue filters also happens to have good sensitivity to spots on cool stars such as GJ 674. To minimize atmospheric scintillation noise, we took advantage of the low stellar density to defocus the images to , so that we could use longer exposure times. The increased read-out and sky-background noises from the larger synthetic aperture, which we then had to use, remain negligible compared to both stellar photon noise and scintillation.
We gathered 14 to 75 images per night with a median exposure time of 20 s. We used the Sept. 24 data, which have the longest nightly time base, to tune the parameters of the IRAF DAOPHOT package and optimise the set of reference stars (HD 157931, CD 4611534, and 7 anonymous fainter stars) to minimise the dispersion in the GJ 674 photometry for that night. These parameters were then fixed for analysis of the full data set. The nightly light curves for GJ 674 were normalized by that of the sum of the references, clipped at 3- to remove a small number of outliers, and then averaged to one measurement per night to examine the long-term photometric variability of GJ 674. GJ 674 clearly varies with a 1.3% amplitude and a (quasi-)period close to 35 days (Fig. 5). To verify that this variability does not actually originate in one of the reference stars, we repeated the analysis alternately using HD 157931 alone as the reference star and the average of the 8 other references. Both light curves are very similar to Fig. 5.
The photometric observations are consistent with the signal of a single spot, within the limitations of their incomplete phase coverage: the variations are approximately sinusoidal, and their 0.2-0.3 radian phase shift from the corresponding radial velocity signal closely matches the difference expected for a spot. The spot would cover 2.6% of the stellar surface if completely dark, corresponding to a radius for a circular spot.
|Figure 6: Upper panel: differential radial velocity of GJ 674, corrected for the signature of the 4.69 days planet in our 2-planet Keplerian fit, as a function of the H (red filled circles) and Ca II H&K (green filled squares) spectral indices defined in the text. Bottom right panels: the Ca II H+K and H indexes phased to the longer period of the 2-planet keplerian model. Bottom left panels: Power density spectra of the spectroscopic indexes. A clear power excess peaks at 34.8 days (vertical dashed lines).|
|Open with DEXTER|
Like the well-known Mt. Wilson S index (Baliunas et al. 1995),
our Ca II H+K index is defined as
This H+K index varies with a clear period of 34.8 days (Fig. 6). Within the combined errors, this is consistent with both the photometric period and the longer radial-velocity period. The phasing of the chromospheric index and the photometry is such that lower photometric flux matches higher Ca II emission, as expected if active chromospheric regions hover over photospheric spots.
The imprint of a spot is also seen when the radial-velocity is plotted as a function of the spectral index value. During a star's rotation, the spot crosses the sub-observer meridian two times, once on the hemisphere facing the observer (with a maximal projected area) and once on the opposite hemisphere (with a minimal projected area, and possibly unseen because of its latitude and/or the stellar inclination). Since the rotational velocity cancels on the sub-observer meridian, the phases with index extrema correspond to radial-velocity minima. At intermediate phases, the spot induces positive or negative radial-velocity shifts depending if the masked area is on a rotationally blue- or red-shifted part of the star. At the end of the day, the spectral index traces a closed loop as a function of radial velocity.
Chromospheric filling-in of the photospheric H absorption has similarly been found to be a powerful activity diagnostic for M dwarfs. Kürster et al. (2003) find that in Barnard's star it correlates linearly with the radial-velocity variations and interpret that finding as evidence that active plage regions inhibit the convective velocity field. The variation pattern in GJ 674 definitely differs from a linear correlation between H and the radial-velocity residuals, so it needs a different explanation.
Similar to Kürster et al. (2003), we define our
The chromospheric indices vary by factors of 2 and 1.3 (for our specific choices of continuum windows) and are thus much more contrasted than the photometry. They do not, however, vary as smoothly with phase as the photometry, perhaps due to (micro-)flares, somewhat reducing their value as diagnostics of spot-induced radial velocity variations. These measurements on the other hand require no new observation, and they undoubtly reinforce the spot interpretation.
In Sect. 4 we have shown that our 32 radial-velocity measurements of GJ 674 are described well by two Keplerian signals, as illustrated by the low reduced chi-square of that model. The above analysis (Sect. 5) however demonstrates that the rotation period of GJ 674 coincides with the longer of the two Keplerian periods. Both the stellar flux and the Ca II H+K emission vary with that period, implying that the surface of GJ 674 has a magnetic spot. This spot must induce radial-velocity changes, with the observed phase relative to the photometric signal. As a consequence, some, and probably all, of the 35-day radial-velocity signal must originate in the spot. Planet-induced activity through magnetic coupling (e.g. Shkolnik et al. 2005) should be an alternative explanation of the correlation, but here it is not a very attractive one: the inner planet is at least as massive as the hypothetical 35-day planet and would, at least naively, be expected to have stronger interactions with the magnetosphere of GJ 674. The 4.69-day period, however, is only seen in the radial velocity signal, and it has no photometric or chromospheric counterpart.
The 1-planet fit, which effectively treats the activity signal as white noise, results in a minimum mass of for GJ 674b. The 2-planet fit by contrast filters out this signal. That filtering obviously uses a physical model that is not completely appropriate, but that remains preferable to handling a (partly) coherent signal as white noise. We therefore adopt the corresponding estimate of the minimum mass, .
At 0.039 AU from its parent star, the temperature of GJ 674 b is 450 K. Planets above a few Earth masses planets can, but need not, accrete a large gas fraction, leaving its composition - whether mostly gaseous or mostly rocky - unclear. The orbital eccentricity might shed light on the structure of GJ 674b, if confirmed by additional measurements. Rocky and gaseous planets have rather different dissipation properties, and significant eccentricity at the short period of GJ 674 b needs a high Q factor, unless it is pumped by an additional planet at a longer period (e.g. Adams & Laughlin 2006). For now, the stellar activity leaves the statistical significance of the eccentricity slightly uncertain, so we prefer to stay clear of overinterpreting it.
One can additionally note that the two stars that host giant planets, GJ 876 and GJ 849, occupy the metal-rich tail of the M dwarf metallicity distribution, with GJ 849 almost as metal-rich as the most metal-rich star of the comparison sample. The next most metal-rich of the M dwarfs with planets, GJ 436, has an additional long-period companion (P>6 yr) that might well be a giant planet (Maness et al. 2006), which would then strengthen that trend. If confirmed by additional data, this would validate the theoretical predictions (Benz et al. 2006; Ida & Lin 2004) that only Jovian-mass planets are more likely to form around metal-rich stars. Current observations are consistent with this prediction, but not yet very conclusively so (Udry et al. 2006).
|Figure 7: Upper panel: metallicity distributions of 44 M dwarfs without known planets (gray shading) and of the 5 M dwarfs known to host planets (black shading). Bottom panel: corresponding cumulative distributions (solid and dashed lines, respectively).|
|Open with DEXTER|
Much recent theoretical work has gone into examining how planet formation depends on stellar mass. Within the "core accretion'' paradigm, Laughlin et al. (2004) and Ida & Lin (2005) predict that giant planet formation is inhibited around very low-mass stars, while Neptune-mass planets should inversely be common. Within the same paradigm, but assuming that M dwarfs have denser protoplanetary disks, Kornet et al. (2006) predict instead that Jupiter-mass planets become more frequent in inverse proportion to the stellar mass. Finally, Boss (2006) examines how planet formation depends on stellar mass for planets formed by disk instability, and concludes that the frequency of Jupiter-mass planets is independent of stellar mass, as long as disks are massive enough to become unstable.
To date, none of the 300 M dwarfs scrutinized for planets by the various radial-velocity searches (Endl et al. 2006; Bonfils et al. 2006; Butler et al. 2006) has been found to host a hot Jupiter. Conversely, GJ 674b is already the 4th hot Neptune. Although that cannot be established quantitatively yet, these surveys are likely to be almost complete for hot jupiters, which are easily detected. Hot-Neptune detection, on the other hand, is definitely very incomplete. Setting aside this incompleteness for now, simple binomial statistics show that the probability of finding none and 4 detections in 300 draws of the same function is only 3%. There is thus a 97% probability that hot Neptunes are more frequent than hot Jupiters around M dwarfs. Accounting for this detection bias in more realistic simulations (Bonfils et al. in prep.) obviously increases the significance of the difference. Planet statistics around M dwarfs therefore favor the theoretical models that, at short periods, predict more Neptune-mass planets than Jupiter-mass planets.
Table 2: Keplerian parameterization for GJ 674b and GJ 674's spot.
We are grateful to the anonymous referee for constructive comments. X.B. and N.C.S. acknowledge support from the Fundação para a Ciência e a Tecnologia (Portugal) in the form of fellowships (references SFRH/BPD/21710/2005 and SFRH/BPD/8116/2002) and a grant (reference POCI/CTE-AST/56453/2004). The photometric monitoring was performed on the EULER 1.2 meter telescope at La Silla Observatory. We are grateful to the SNF (Switzerland) for its continuous support. This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France.
As demonstrated by Saar & Donahue (1997), the bisector analysis loses much of its diagnostic power when applied to slow rotators. In simulations of the impact of star spots on radial-velocity and bisector measurements, they found that, for a given spot configuration, the radial velocity varies linearly with , while the bissector span varies as . The bissector signal therefore decreases faster with decreasing rotational velocities than the radial-velocity signal, and disapears faster in measurement noise. For GJ 674 we measure a very low rotation velocity ( ). It is therefore not surprising that the correlation between the bissector span and radial velocity is weak (Fig. A.1) and not statistically significant.