SPIRou Input Catalogue: global properties of 440 M dwarfs observed with ESPaDOnS at CFHT

Abstract

Present and future high-precision radial-velocity spectrometers dedicated to the discovery of low-mass planets orbiting low-mass dwarfs need to focus on the best selected stars to make an efficient use of telescope time. In the framework of the preparation of the SPIRou Input Catalogue (SPIC), the CoolSnap program aims at screening M dwarfs in the solar neighbourhood against binarity, rapid rotation, activity, etc. To optimize the selection, this paper describes the methods used to compute effective temperature, metallicity, projected rotation velocity of a large sample of 440 M dwarfs observed in the visible with the high-resolution spectropolarimeter Echelle SpectroPolArimetric Device for the ObservatioN of Stars (ESPaDOnS) at Canada–France–Hawaii Telescope. It also summarizes known and newly discovered spectroscopic binaries, and stars known to belong to visual multiple systems. A calibration of the projected rotation velocity versus measured line widths for M dwarfs observed by the ESPaDOnS spectropolarimeter is derived, and the resulting values are compared to equatorial rotation velocities deduced from rotation periods and radii. A comparison of the derived effective temperatures and metallicities with literature values is also conducted. Finally, the radial-velocity uncertainty of each star in the sample is estimated, to narrow down the selection of stars to be included into the SPIC.

1 INTRODUCTION

Dwarf stars of spectral type M were the coolest stellar objects known until the discovery of field brown dwarfs (Becklin & Zuckerman 1988; Rebolo, Zapatero Osorio & Martín 1995; Nakajima et al. 1995) and the creation of new spectral types L, T, and Y (Martin et al. 1997; Kirkpatrick et al. 1999; Martín et al. 1999; Kirkpatrick 2000). M dwarfs are the most numerous stars in our Galaxy, amounting to about two-thirds in number and about 40 per cent in stellar mass (Kirkpatrick et al. 2012). They were not known from ancient astronomers, as none of them is visible to the naked eye: the brightest one, Gl 825, has a V magnitude of 6.7 and is an M0V, sometimes classified as K7V. Therefore, M stars were a good example of ‘invisible’ matter, later recognized as a major contributor to the stellar mass of our Galaxy.

Although they share a common spectral class, they display a wide range in properties: their masses span a range of about a factor 9, similar to the range spanned by B, A, F, G, and K stars altogether. Similarly, their bolometric luminosities span a range of 200. Their global properties vary a lot along the sub-classes from M0 to M9, crossing the limit between stars and brown dwarfs, and as other spectral types, they display a large variety of ages, from pre-main-sequence stars of a few Myr to very old stars, with a corresponding range of radius and therefore gravity for a given mass. They also belong to different star populations (Galactic disc and halo), being classified as dwarfs, subdwarfs, extreme subdwarfs, and even ultra subdwarfs according to their metallicity (Lépine, Rich & Shara 2007).

Lépine & Gaidos (2011) estimate that there are about 11 900 M dwarfs brighter than J = 10 in the whole sky. But given their wide range in absolute magnitudes, it is difficult to translate this figure to a given number of M dwarfs within a given distance limit, for instance 25 pc. All the early M dwarfs (up to M3.5V) will then be counted, but not the later spectral type ones. There is no current complete catalogue of late M dwarfs up to a given distance.

In addition, it is well known that M dwarfs display a range of activity, rotational velocity, and magnetic properties (West et al. 2004; Reiners 2007; Kiraga & Stepien 2007; Donati et al. 2008; Morin et al. 2008a,b, 2010, 2011; Irwin et al. 2011; Reiners, Joshi & Goldman 2012; West et al. 2015; Newton et al. 2017 among others), that is further investigated in this study and companion papers (Moutou et al. 2017; Malo et al., in preparation) . Although this class of stars was somehow neglected in the past due to their faintness at optical wavelengths, it started to emerge with the advent of near-infrared sky surveys, DEep Near-Infrared Southern Sky Survey (DENIS) (Epchtein et al. 1999) and Two Micron All Sky Survey (2MASS, Skrutskie et al. 2006), which opened the way to near-infrared spectrometers. As the small mass and radius of M dwarfs were favourable to reveal their planetary companions, and with the additional benefit that their habitable zones lie close enough to the star to allow discoveries of habitable planets, large surveys of these stars began (e.g. Bonfils et al. 2013; Delfosse et al. 2013).

In the framework of the preparation of the new near-infrared high-resolution spectropolarimeter SPIRou (Donati et al., in Deeg & Belmonte (2018)), to be installed at Canada–France–Hawaii Telescope (CFHT) in 2018, members of the SPIRou team decided in 2014 to embark upon an observational snapshot program of M dwarfs, nicknamed CoolSnap, using the ESPaDOnS visible high-resolution spectropolarimeter at CFHT (Donati et al. 1997). The goal of this survey is a better knowledge of M dwarfs selected as prime targets to search for planetary-mass objects in the habitable zone before their inclusion into the SPIRou Input Catalogue (SPIC). The selection criteria used to build the CoolSnap sample are described in Malo et al. (in preparation). Their activity and magnetic properties are described in Moutou et al. (2017). Here, we concentrate on the global properties of the observed stars, such as effective temperature, metallicity, rotational velocity, and binarity. These properties are important for our selection, as we want to avoid stars that are too active, fast-rotating objects, close multiple systems, which will all prevent us from detecting low-mass planets orbiting these stars.

Other near-infrared spectrographs are currently under development, such as Habitable-zone Planet Finder (HPF, Mahadevan et al. 2012), Calar Alto high-Resolution search for M dwarfs with Exoearths with Near-infrared and optical Echelle Spectrographs (CARMENES, Quirrenbach et al. 2014), or GIAno and haRPS (GIARPS, Claudi et al. 2016). These projects can benefit from our study, as we benefited for instance from the CARMENCITA catalogue (CARMENes Cool dwarf Information and daTa Archive, Alonso-Floriano et al. 2015; Cortés-Contreras et al. 2017).

This paper is organized as follows: Section 2 describes how the stars were selected to build a sample of 440 M dwarfs. Section 3 describes spectroscopic binaries either discovered during these observations or already known, and more generally the multiplicity of systems to which stars in our sample belong. Section 4 explains how spectral type, effective temperature and metallicities are derived for our sample, and the limitation of the mcal method introduced by Neves et al. (2014), and used to measure these properties . Section 5 describes how projected rotation velocities are derived from the width of the Least-Squares Deconvolution (LSD) profile obtained from the observed spectra with an M2 template. Section 6 concludes about stars which are good candidates for radial velocity (RV) search of low-mass planets using the SPIRou near-infrared spectropolarimeter, from the point of view of the parameters measured in this study. Finally, Section 7 summarizes this work and link it to the other two papers in this series, namely Moutou et al. (2017) and Malo et al. (in preparation).

2 SAMPLE AND OBSERVATIONS

We performed our initial compilation of M dwarfs based on the following studies (see Malo et al., in preparation for more details):

• An all-sky catalogue of bright M dwarfs (Lépine & Gaidos 2011), which consists of 8889 K7–M4 dwarfs with J < 10. This sample is based on the ongoing proper-motion survey using the |$\small{\small{SUPERBLINK}}$| software. Spectral types are estimated from the V − J colour index.

• A catalogue of bright (K < 9) M dwarfs (Frith et al. 2013), which consists of 8479 K7–M4 dwarfs. This catalogue rests on the Position and Proper Motion eXtended-L (PPMXL) proper-motion survey.

• An all-sky catalogue of nearby cool stars CONCH–SHELL (Catalogue Of Nearby Cool Host-Stars for Habitable ExopLanets and Life, Gaidos et al. 2014), which consists of 2970 nearby (d < 50 pc), bright (J < 9) M- or late K-type dwarf stars, 86 per cent of which have been confirmed by spectroscopy. This sample is also selected from the proper-motion survey described in (i), combined with spectra and photometric colour criteria.

• A sample of spectroscopically confirmed nearby M dwarfs (Newton et al. 2014), which consists of 447 M dwarfs with measured metallicities, RVs, and spectral types from moderate resolution (R ∼ 2000) near-infrared spectroscopy. This sample is drawn from the MEarth survey (Irwin et al. 2011).

• A southern sample of M dwarfs within 25 pc (Winters et al. 2015), which consists of 1404 M0–M9.5 dwarfs with 6.7 < V < 21.4. This sample is based on the REsearch Consortium On Nearby Stars (RECONS) program and supplemented by observations at the Cerro Tololo Inter-american Observatory / Small and Moderate Aperture Research Telescope System Consortium (CTIO/SMARTS) 0.9 m telescope.

• The CARMENES input catalogue of M dwarfs (Alonso-Floriano et al. 2015), which consists of 753 spectroscopically confirmed K–M stars.

• A northern sample of mid-to-late M dwarfs from the MEarth project (Newton et al. 2016), which consists of 387 nearby dwarfs with measured rotation periods.

This compilation leads to an all-sky sample of about 14 000 K5–M9 stars. Since SPIRou will be installed at CFHT (latitude 20°), we restrict our sample to stars observable with declination north of −30°, which gives a final sample of 10 142 stars.

We applied to this initial sample a merit function computed from the star flux in H band and the expected RV amplitude produced by a 3 Earth mass planet orbiting it in the Habitable Zone, which in turn depends upon mass, radius, and temperature of the star, to select the 150 highest merit stars to be observed. Details about this merit function are given in Malo et al. (in preparation).

Observations were conducted with the ESPaDOnS spectropolarimeter (Donati et al. 1997) at the CFHT 3.6 m telescope on top of Maunakea (Hawaii), which provides a wide optical range from 367 to 1050 nm in a single shot at a resolving power of 65 000 (polarimetry) or 68 000 (pure spectroscopy in the so-called star plus sky mode, with one fibre on the target and one on the sky: we call it ‘S+S’ hereafter). Data are reduced using the libre-esprit software (Donati et al. 1997). LSD (Donati et al. 1997) is then applied to all the observations, to take advantage of the large number of lines in the spectrum and increase the signal-to-noise ratio (S/N per 2.6 km s⁻¹ pixel) by a multiplex gain of the order of 10. We used a mask of atomic lines computed with an atlas local thermodynamic equilibrium (LTE) model of the stellar atmosphere (Kurucz 1993a). The final mask contains about 4000 moderate to strong atomic lines with a known Landé factor. This set of lines spans a wavelength range from 350 to 1082 nm. The use of atomic lines only for the LSD masks relies on former studies of early and mid-M dwarfs (Donati et al. 2006).

More details about the CoolSnap observations¹ and the data reduction are given in Moutou et al. (2017) and Malo et al. (in preparation). For the purpose of this paper, let us just state that two high S/N spectra (S/N ∼100 at 800 nm) are taken for each M star of the sample (typically M0–M6), separated by several days or weeks, in order to assess possible changes in the magnetic activity or in the heliocentric radial velocity (HRV). We observed 280 spectra in polarimetric mode for 118 stars. Removing four stars initially selected for the CoolSnap sample and observed, but for which classification issues (they most certainly are not M dwarfs) were later discovered, leads to 114 genuine M dwarfs in the CoolSnap sample. The four rejected stars are listed in Table 1 for completeness.

In addition to our own measurements, we searched the ESPaDOnS archive in polarization mode at the Canadian Astronomy Data Center (CADC²) from 2005 to 2015 (inclusively) and found 839 spectra for 71 additional M dwarfs (and 10 spectra for two stars in the CoolSnap sample, namely Gl 411 and Gl 905). The two samples have different characteristics, the stars from the archive often being active and rapid rotators and generally having a large number of spectra, while the CoolSnap sample is limited to two spectra taken at different epochs for each star.

Finally, we also searched the ESPaDOnS archives for stars observed in the purely spectroscopic S+S mode. We found 785 spectra for 255 additional stars, raising the total sample of M dwarfs observed with ESPaDOnS to 440.

Spectra of stars belonging to the complementary samples (polarimetric and spectroscopic) have generally been published, but we reanalyse them to derive their effective temperature, metallicity, and projected rotation velocity in a consistent way.

3 MULTIPLE SYSTEMS

Binarity (and higher multiplicity) is common among stars. Many techniques have been devised to disentangle physical association from apparent projection on the sky. A good historical review is given by Dommanget & Nys (2000a). For our purpose, multiplicity may be important for the following reasons:

• we may discover that an object initially identified as a single star is in fact a close binary. The selection criterion may then be invalidated when the magnitude or colour encompasses both stars;

• if the components are too close to be separated in the fibre entrance of the spectrograph, both spectra are recorded and the object may then reveal as a single- or double-line spectroscopic binary;

• even when the separation is large enough, and assuming that the system is physical, the planet formation mechanism may have been affected by the binarity;

• wide multiple physical systems composed of an FGK primary and an M secondary allow a calibration of the metallicity of the M dwarf, assuming that it shares the same metallicity as the primary component of the system (see e.g. Bonfils et al. 2005).

The release of originally Global Astrometric Interferometer for Astrophysics (GAIA) data (DR1 and soon DR2) will allow us to confirm the status of the binaries in our sample, and discard the optical systems which are not physical. GAIA will certainly also discover new astrometric binaries in this sample. However, it is still important for future observations to know whether a star has a close companion, since the light from the companion may contribute significantly to the measured flux, which may affect the measured parameters (magnitudes, colours, etc.).

It is obvious that only a fraction of these systems may affect our observations or the future detection of planetary systems orbiting the stars of our sample. As the fibre diameter is 1.58 arcsec for ESPaDOnS and 1.33 arcsec for SPIRou, binaries separated by less than 1 arcsec will contaminate the observed spectrum. Components separated by more than 2 arcsec should be easy to separate under reasonable seeing. However, at this separation, some parameters may still be affected, such as visual or near-infrared magnitudes.

On another hand, physical separations matter in the rate of formation of planetary systems. Therefore, close physical multiplicity of the stellar system may affect the formation of planets. More details are given in Thebault & Haghighipour (2014).

In order to identify the physical systems (visual or spectroscopic) in our catalogue, we started to build a catalogue of multiple systems involving M dwarfs. We defer to a future publication details and statistics about this catalogue, for instance a confirmation of physical systems based on future released data from GAIA (DR2 and following), and an evaluation of the multiplicity rate among M stars, compared to earlier spectral types, based on a complete distance-limited sample.

3.1 Spectroscopic binaries

Spectroscopic binaries are easily identified when two peaks appear in the LSD profile (SB2). Sometimes, only one component is visible in the spectrum (generally because the other component is much fainter), and we have an SB1. Given that the accuracy of the HRV measured by ESPaDOnS and reduced with libre-esprit (Donati et al. 1997) is about 20–30 m s⁻¹ (Moutou et al. 2007), SB1 are revealed when the RV corrected to the heliocentric reference frame HRV significantly differs between the two spectra. In Table A1, we list the stars in our sample which have been observed and revealed themselves as SB2 (21 stars including uncertain ones), or even SB3 (2 stars), together with already known spectroscopic binaries (28 SB2, 4 SB3, two quadruple systems SB1+SB2 and SB2+SB2), which should have been excluded when assembling the observational sample.

Among the 57 SB listed in this table, about one half also appears in Table A2, because they belong to multiple systems with both visual and spectroscopic components.

In Table 2, we list the stars in our sample which have been observed and revealed themselves as SB1 (two stars), together with already known single-line spectroscopic binaries (eight stars), also missed when assembling the sample or discovered by others during our survey. RV variations may also be due to activity-induced rotational modulation for stars with strong magnetic fields, rather than binarity. A few special cases with discrepant or anomalous results are listed in Table 3.

3.2 Visual multiple systems detected by imagery

As stated above, it is important to know whether a star in our survey belongs to a physical multiple system. Unfortunately, there is no recent compilation of such systems. Rather than just checking for the multiplicity status of the stars in our sample, we embarked into a parallel project of listing all multiple systems involving an M dwarf, in order to get better statistics, not biased by the selection process which led to our sample. For this purpose, we surveyed the literature for physical systems detected by imagery, including adaptive optics, coronagraphy, or lucky imaging observations of M dwarfs.

We started by checking the information provided by the Washington Double Star Catalogue (Mason et al. 2001), in its constantly updated online version at CDS (hereafter WDS), the Catalogue of Visual Double Stars observed by the Hipparcos satellite (Dommanget & Nys 2000a,b), the Catalog of Physical Multiple Stars (MSC, Tokovinin 1997), the Catalog of Common Proper-Motion Companions (hereafter CPM) to Hipparcos stars (Gould & Chanamé 2004), and the Catalogue of Faint Companions to Hipparcos stars (Lépine & Bongiorno 2007). We then surveyed the literature for additional binary stars or additional information on the systems described in the above references. Finally, some optical binaries were discovered by us at the telescope, using images from the guider.

The compilation used in this paper is not complete, as we preferred waiting for the second release of GAIA in 2018 April, to discard unphysical multiple systems or components when GAIA measures discrepant parallaxes or proper motions. In its present version, it contains 671 multiple systems, among which 393 have an M dwarf primary. We used this limited version for investigating the multiplicity of stars in our sample of 440 M dwarfs. The resulting table is given in Appendix A.

4 MEASURE OF SPECTRAL TYPE, EFFECTIVE TEMPERATURE, AND METALLICITY

4.1 Spectral type

We estimate the spectral type of our stars from a measurement of the TiO₅ spectral index at 713 nm, as defined and calibrated in Reid, Hawley & Gizis (1995). It is well adapted to the range of spectral types of our sample, at least up to M6.5V. Standard numerical values are adopted, from −1 for K7V, 0 for M0V to 6 for M6V. The correlation with the V − K_s colour is clear, as displayed in Fig. 1. Some stars with an earlier spectral type than our M0 limit (negative spectral indices) or for which we could not measure the spectral type are listed in Table 4. As the limit between spectral classes K7V and M0V is somewhat fuzzy, we prefer not to exclude those stars a priori, without a clear confirmation of a K spectral type. The value of V − K_s may help, as the average value of 25 M0 stars in our sample is 3.65 ± 0.02. Other outliers are generally close visual binaries, where the photometry may be contaminated. They are listed in Table 5.

4.2 The mcal method

Three important parameters used to characterize stars are effective temperature, metallicity, and gravity. In case of M dwarfs, they are notoriously difficult to measure, especially because no continuum exists in the optical spectrum. There is a long list of publications dealing with several methods to measure mainly the two first, without reaching definite conclusions, for instance Bonfils et al. (2005), Woolf & Wallerstein (2005), Casagrande, Flynn & Bessell (2008), Önehag et al. (2012), Rojas-Ayala et al. (2012), and Rajpurohit et al. (2013).

In this work, we chose to use the mcal method of measurement described in Neves et al. (2014). In short, it is based on measurements of pseudo-equivalent widths of lines in high-resolution optical spectra obtained by Bonfils et al. (2013) using the High Accuracy Radial velocity Planet Searcher (HARPS) spectrometer, which are then correlated to known values of T_eff and [Fe/H] from Casagrande et al. (2008) and Neves et al. (2012), respectively. A caveat is that gravity is not used in this correlation, so young stars with low gravity probably get assigned a biased temperature and metallicity.

For this study, we started by using the Neves et al. (2014) calibration: the calibrating T_eff values come from Casagrande et al. (2008) T_eff versus colour relations. The authors enhanced the infrared flux method (Blackwell et al. 1990), to apply it to M dwarfs by adding information from the optical range. Their new method is called MOITE (Multiple Optical-Infrared TEchnique). In this method, the bolometric flux comes from optical and infrared photometry for about 80 per cent, and the rest comes from Phoenix models³ described for instance in Hauschildt et al. (1999).

Although this method allows one to derive metallicities, Neves et al. (2014) prefer to use their own metallicity values. These are based on a technique pioneered by Bonfils et al. (2005). It starts with binary stars where the primary component is a star of spectral type F, G, or K which has a spectroscopically measured metallicity, and the secondary is an M dwarf assumed to share the metallicity of the primary. These binary M dwarfs serve in turn to calibrate an M_k versus V − K_s colour–magnitude diagram: the main-sequence locus at an average metallicity is identified, and the colour or absolute magnitude shift from this locus gives a measure of the metallicity of new M stars. Subsequently, Johnson & Apps (2009) corrected the calibration for metal-rich M stars, and Schlaufman & Laughlin (2010) refined that latter calibration. Mann et al. (2013a) compiled and measured metallicities of solar-type primaries in 112 wide binary systems involving an M dwarf secondary. Maldonado et al. (2015) used a similar method to mcal to calibrate stellar parameters of 53 M dwarfs observed with HARPS.

Neves et al. (2012) refined once more over Schlaufman & Laughlin (2010). Using this calibration, Neves et al. (2013) computed the metallicity of all the M dwarfs in the Bonfils et al. (2013) sample and Neves et al. (2014) selected the more suitable for their calibration of the pseudo-equivalent widths versus metallicity and effective temperatures. Their table 2 contains 65 stars, and the calibrating values are given in the columns labelled [Fe/H]_N12 and Teff_C08. It should be noted that some of these values differ from the similar previous table A.1 in Neves et al. (2013), probably because of a change in the adopted V magnitude of the star, which in turn produces a change in the distance to the main-sequence locus and therefore of its computed T_eff from colours.

4.3 Limits of the method

Not all spectra are usable when applying the mcal method. Some spectra have low S/N, giving an ill-defined peak in the LSD profile, or an inaccurate RV. Two stars (vB8 and vB10) have nine spectra each in the Polar archive (published in Morin et al. (2010), with S/N between 68 and 107, but they have very late spectral types (M7V and M8V, respectively) outside of the calibration range of the method. In polarimetric mode, we are therefore working on 1090 spectra taken with a large enough S/N (typically, S/N per 2.6 km s⁻¹ pixel >100), for 182 stars, removing the two very-late dwarfs mentioned above and 2MASS J09002359+215054, which only has one spectrum with an S/N of 30 in the CoolSnap sample. Similarly, some S+S spectra have a low S/N which does not meet our original quality criterion for polarimetry (S/N >100). We only exploited S+S spectra of good quality (well-detected LSD profile, correct RV), reducing the number of useful spectra to 706 for 298 stars (including 45 with polarimetric spectra too), which added to the 182 stars with useful polarimetric spectra leads to a total of 435 stars which can a priori be used to measure global parameters.

But in fact, as explained by Neves et al. (2014), some very active stars are not suitable to the measurement of T_eff and [Fe/H] by this method. As many stars in the ESPaDOnS archive are active, this can drastically reduce the sample of stars where those parameters can be measured. To identify very active stars, the method measures an Hα index as defined in Gomes da Silva et al. (2011). A small value of about 0.03 corresponds to inactive stars. The adopted cut-off is a value of 0.25, roughly corresponding to a luminosity ratio log L_Hα/L_bol of −4.0, above which Hα and magnetic flux become independent of the rotation rate, as shown in Reiners et al. (2009). According to this cut-off between saturated (or very active) and non-saturated stars, our CoolSnap sample contains 10/113, the polarimetric archive 40/69, and the spectroscopy archive 146/253 very active stars, for which metallicity and effective temperatures cannot be reliably measured by the mcal method. An additional 33 non-saturated stars are spectroscopic binaries, for which the method does not work properly either (see above). Finally, a few non-saturated rapid rotators are not well suited either to this technique, as the measurement of pseudo-equivalent widths is affected by the broadening of the lines due to rotation, and the calibration therefore returns too low temperatures. We do not consider measured effective temperatures and metallicities for 20 non-saturated stars with a vsin i larger than 8 km s⁻¹ . We are left with 192 stars on which comparisons with other methods can be secured.

The main source of accurate T_eff comes from the work of Boyajian et al. (2012), who measure M dwarf radii using the Center for High Angular Resolution Astronomy (CHARA) interferometer. They then compute the bolometric flux from multiband photometry and derive a value of T_eff. This seems to be a straightforward method, if the template spectra fitted to the photometry are reliable. Mann, Gaidos & Ansdell (2013b) argue that when compared to their actual low-resolution spectra, there are systematic differences, leading to underestimated bolometric fluxes and temperatures. Finally, Mann et al. (2015) use the same method to measure the bolometric flux, but use the Cosmological Impact of the First STars (CIFIST) team suite of the BT-Settl version of the PHOENIX atmosphere models (Allard et al. 2013), to measure T_eff and derive the corresponding radii.

Both Mann et al. (2015) and Rajpurohit et al. (2013), who measured T_eff by fitting BT-Settl synthetic spectra, show that T_eff values from Casagrande et al. (2008) are too low due to the assumption that M dwarf can be treated as black bodies beyond 2 μm. As the Casagrande et al. (2008) temperature scale is used in the original mcal method used by Neves et al. (2014), it is important to confirm this result. For this purpose, we compared the original Neves et al. (2014) calibration to other sources of measurements, for instance Woolf & Wallerstein (2005, 2006), who use CaH₂ and TiO₅ molecular band strength indices, Önehag et al. (2012), Lindgren, Heiter & Seifahrt (2016), and Lindgren & Heiter (2017), who fit synthetic spectra to high-resolution Very Large Telescope / CRyogenic high-resolution InfraRed Echelle Spectrograph (VLT-CRIRES) spectra in the J band, which are free from large molecular-band contributions, or Rojas-Ayala et al. (2012), who measure equivalent widths of CaI and NaI lines in the near-infrared and a spectral index quantifying the absorption due to H₂O opacity. We found a faire agreement for the metallicities (within 0.2 dex), but the effective temperatures obtained using the original calibration are systematically low by about 200 K.

We therefore adopt the Mann et al. (2015) T_eff scale while retaining the Neves et al. (2014) metallicity scale. We modified the original mcal code to recompute the coefficients of the calibration relations using the more recent and accurate source of T_eff. The code contained a revised table of 68 calibrators, adding three stars to Neves et al. (2014, Table 2: Gl 388, Gl 551, and Gl 729). Among these calibrators, only 29 have T_eff and [Fe/H] values in Mann et al. (2015). We therefore use these 29 stars with Mann et al. (2015) T_eff (ranging from 3056 to 3848 K) and Neves et al. (2014) [Fe/H] (ranging from −0.51 to 0.19 dex) to recalibrate the matrices given in Neves et al. (2014). The median differences between Mann et al. (2015)- and Neves et al. (2014)-based calibrations are: ΔT_eff = 180 ± 80 K and Δ[Fe/H] = 0.04 ± 0.12 dex. Similarly, median differences between the Mann et al. (2015)-based calibration and Rojas-Ayala et al. (2012) for 21 stars in common are: ΔT_eff = 240 ± 170 K and Δ[Fe/H] = 0.08 ± 0.11 dex. This confirms the offset of about 200 K in temperature and the fair agreement in metallicity.

The list of stars used in this comparison is given in Table A3 in Appendix A. Spectroscopic binaries have been removed from this comparison: SB2 have double lines which probably affect the determination of the continuum, and there is a risk to mix both components in the measurements of the lines. SB1 are a priori more immune, but the secondary may affect the line depth, which is used in the determination of both effective temperature and metallicity in the mcal method.

Promising new techniques to derive effective temperature, metallicity, and gravity of M dwarfs have been pioneered by Rajpurohit et al. (2013), using high-resolution stellar spectra and up-to-date model atmospheres. They look for the best combination of the three parameters used as an input to generate BT-Settl synthetic spectra which reproduce the observed spectra. We are in the process of applying this method described in Rajpurohit et al. (2017) to our spectra. Unfortunately, preliminary results show a good agreement only for effective temperatures, but no correlation for metallicities. An example of fitted spectrum is given in Fig. 2. A more thorough comparison of our results with BT-Settl synthetic spectra will be deferred to a future paper.

Finally, a similar comparison using specific wavelength windows in which line parameters were corrected to provide an optimal fit to some standard stars with known parameters is also in progress (Kulenthirarajah et al., in preparation).

4.4 Comparison of results

Fig. 3 shows a comparison of our effective temperatures to corresponding values from Mann et al. (2015). We adopt their uncertainty on T_eff as listed (typically 60 K) and a quadratic sum of the uncertainty returned by mcal and a systematic uncertainty of 60 K for our measurements, based on the observed dispersion between the two sets. The agreement is not surprising as our recalibration of mcal method is based on 29 effective temperatures from Mann et al. (2015, green points), but we have more measured stars (red points) and not all 29 calibrators have an ESPaDOnS spectrum. After rejecting three outliers from the sample (LHS 1723, Gl 297.2B, and HH And=Gl 905), the mean difference between the two systems computed from 57 stars is T_eff (this work) – T_eff (reference) = 20 ± 12 K with an rms of 90 K. Given that Mann's temperatures have a typical uncertainty of 60 K, it shows that our effective temperatures should have a similar accuracy, and we therefore adopt a systematic uncertainty of 60 K for our measurements.

We also compare our results to the work of Maldonado et al. (2015), who use a similar method to mcal to estimate effective temperatures and metallicities. We adopt their uncertainty on T_eff as listed (typically 68 K) and a quadratic sum of the uncertainty returned by mcal and a systematic uncertainty of 60 K for our measurements. Unfortunately, their sample is limited to early-type stars, but the agreement with our effective temperatures is also satisfactory, as can be seen in Fig. 4 (mean difference, this work minus Maldonado et al. 2015: +16 ± 17 K, σ = 64 K).

For the metallicity comparison, Fig. 5 displays the results from Mann et al. (2015) compared to ours. We adopt an uncertainty on the [Fe/H] values from their paper (typically 0.08 dex), and a quadratic sum of the uncertainty returned by mcal and a systematic uncertainty of 0.10 dex, based on the observed dispersion between the two sets. This is a more meaningful comparison than for effective temperatures, as Mann's metallicities have not been used in our recalibration. It shows a generally good agreement, but some of our metallicities seem too high. These correspond to K7V–M0V stars, which have effective temperatures slightly out of our calibration domain. Rejecting the same three stars, the mean difference between the two systems is [Fe/H] (this work) – [Fe/H] (reference) = 0.014 ± 0.020 dex with an rms of 0.15 dex. Given that Mann et al. (2015) claim an accuracy of 0.08 dex, our accuracy would be about 0.13 dex. However, removing four K7V–M0V stars with discrepant metallicities still gives a negligible offset of −0.021 ± 0.011 dex, but with an rms of 0.08 dex. We therefore adopt a systematic uncertainty of 0.10 dex for our values of [Fe/H] when the effective temperatures fall within the limits of our calibration (3056–3848 K), to be added quadratically to the generally negligible uncertainty returned by mcal.

A more independent comparison for metallicities has been made with the results obtained by Terrien et al. (2012), who measure equivalent widths of Na, Ca, and K lines in the near-infrared (H and K bands), and correct for effective temperature effects using H₂O indices. Thirty-three non-active stars were found in common with Terrien et al. (2015), and the comparison is displayed in Fig. 6. We adopt a uniform uncertainty of 0.11 dex on the [Fe/H] values from Terrien et al. (2015), as stated in their paper, and a quadratic sum of the uncertainty returned by mcal and a systematic uncertainty of 0.10 dex. The agreement is satisfactory (mean difference, this work minus Terrien et al. 2015: −0.055 ± 0.026 dex, σ = 0.15 dex).

5 MEASURE OF THE PROJECTED ROTATION VELOCITY

In order to measure the rotation of these stars from our polarimetric observations, we need a calibration of the rotational velocity of M dwarfs from the observed width of the LSD profile given by the LSD technique, described in Donati et al. (1997).

We use M dwarfs of known vsin i from the literature for which high-resolution spectra have been obtained with ESPaDOnS, most of them from archival data and some from the CoolSnap program itself. We have combined both polarimetric and S+S spectra, assuming that the spectral resolution is the same (in fact 65 000 versus 68 000).

5.1 Sample and measurement techniques

We based our compilation of vsin i values from the literature on the catalogue of 334 M dwarfs in Reiners et al. (2012). We only retained stars with a measured value of vsin i, not those with an upper limit. We then added a few stars from Reiners & Basri (2007), Donati et al. (2008), Morin et al. (2008b), Reiners et al. (2009), and Morin et al. (2010) which were missing from the 2012 compilation. Very recently, Reiners et al. (2017) published a spectroscopic survey of 324 M dwarfs, where resolved values of vsin i are listed for 78 stars. This allowed us to revise old values of vsin i and add new calibrators.

Cross-matching the 440 M dwarfs observed with ESPaDOnS in our sample with the list of vsin i calibrators, we end up with 62 common stars with vsin i values ranging from 1.0 to 55.5 km s⁻¹ . Removing two stars which are SB2 (Gl 268 and Gl 735) gives 60 calibrators listed in Table 6.

To calibrate our vsin i measurements, we used three approaches: a first approach uses a calibration of vsin i versus the observed width of the LSD profile, taking into account an intrinsic width which depends on the spectral type of the star. This is the approach adopted by Delfosse et al. (1998) for the ELODIE fiber-fed echelle spectrograph at Observatoire de Haute-Provence (OHP), Melo, Pasquini & De Medeiros (2001) for the Fiber-fed Extended Range Optical Spectrograph (FEROS) spectrograph, Boisse et al. (2010) for Spectrographe pour l‘Observation des PHenomenes des Interieurs stellaires et des Exoplanetes (SOPHIE) at OHP, Houdebine & Mullan (2015), both for SOPHIE and HARPS at European Southern Observatory, La Silla. We find that the intrinsic width, defined as the lower envelope of the observed width, slightly depends on the spectral type. However, it has to be recalled that we use a single template spectrum (mask) for all the stars that we correlate with the observed spectrum. So any mismatch between the actual spectral type of the star and the spectral type of the mask (M2) translates into a modification of the LSD profile.

A second approach uses an FeH line at 995.0334 nm to better estimate the intrinsic broadening of the line due to rotation. This line is insensitive to gravity and magnetic field (Reiners 2007) and should give a more direct comparison among stars of different spectral types than the LSD profiles. The measurement quality, however, is worse than when thousands of lines are used.

Finally, a third approach uses a few slow rotators for which the value of vsin i is known from the literature, and a high S/N polarimetric spectrum taken with ESPaDOnS exists. By broadening the LSD profile of a calibrator using different values of vsin i and comparing to the observed spectrum of a given star, we can then select the best calibrator and deduce the best value of vsin i reproducing the observed spectrum. This assumes that rotation is the main contributor to the width of the LSD profile, which means that we assume that convective turbulence and magnetic field broadening can be neglected. All methods better work for stars where the projected rotational velocity has a significant impact on the global line broadening.

5.2 First approach: measure of the LSD profile

The LSD software (Donati et al. 1997) uses a line list built from an atlas9 LTE model (Kurucz 1993a,b) matching the properties of M2 stars, which contains about 5000 atomic lines weighed by their intensity. The multiplex gain is about 10 in S/N.

5.2.1 Variation of σ_° with spectral type

A necessary step in the calibration of vsin i from the width of the LSD profile is to estimate at each spectral type the minimum value of the width which can be measured. We measure the width of the LSD profile by fitting a Gaussian profile and measuring the value of σ, and we use the V − K_s colour as a quantitative estimate of the spectral type of the stars in our sample. We reject spectra with an S/N lower than 30. A diagram of σ versus V − K_s is displayed on Fig. 7 and clearly shows an accumulation of points at small values of σ. The minimum value of σ could be measured as the mode of the distribution in colour bins. In practice, we fit a lower envelope by eye, and it can be seen that it fits both polarimetric measurements (black points) and S+S spectra (orange points). The minimum value of this lower envelope is about 4 km s⁻¹ , corresponding to a full width at half-maximum of the LSD profile of 9 km s⁻¹ . It is obtained at a V − K_s of about 5, corresponding roughly to an M4 spectral type. For earlier- or later-spectral types, the minimum values are higher.

5.2.2 Calibration of vsin i versus σ

Table 6 gives a list of the 60 stars used to calibrate these relations. Stars with an * have not been used in the calibration of the FeH relation (see below). When an uncertainty is not given in the reference of vsin i, we adopt 10 per cent of vsin i, with a minimum value of 1.5 km s⁻¹ .

We then plot the literature measurements of vsin i versusΔ in Fig. 8. The largest rotator (GJ 3789 at vsin i = 55.5 km s⁻¹ ) does not fit well the trend and is then rejected in order not to bias the calibration. Gl 412B is a clear outlier (strong magnetic slow rotator) and is removed too before the fit. Finally, we could not measure the value of σ for three stars because their spectra have an S/N smaller than 30, and one star has a σ value slightly smaller than the adopted σ_° for its colour.

5.3 Second approach: measure of the 995.0334 nm FeH line

We have selected two FeH lines recommended by Reiners (2007), because the continuum is well defined around 1 μm and these two lines are insensitive to gravity and magnetic effects. However, Reiners (2007) used the Coudé Echelle Spectrograph (CES) at La Silla Observatory (Chile), which has a resolution of 200 000. ESPaDOnS in polarimetric mode has a typical resolution of 65 000 and one of the Reiners’ line is blended in our spectra. We therefore only measure the FeH line at 995.0334 nm (air wavelength), which is very well defined in most of our spectra.

We fit a Gaussian with a linear baseline to this line, and estimate the quality of the fit using various criteria. In some cases, the fit produces spurious results, for instance for spectroscopic binaries, low S/N spectra, K dwarfs where the FeH lines tend to disappear, etc. The criteria are:

• the wavelength shift with respect to the expected position must be smaller than 0.02 nm;

• the value of the χ² per degrees of freedom must be smaller than 0.7;

• the signal must be in absorption and its amplitude must be large enough compared to the noise: after fitting the Gaussian profile, we subtract it from the spectrum and measure the residual noise: we accept a line if the ratio of its amplitude to the noise is larger than 3;

• finally, we reject the fit when the σ is smaller than 1 pixel or much larger than the corresponding σ of the LSD profile by a factor 3.

With these criteria, about 865 of our 1900 spectra provide a valuable fit of the 995.0334 nm FeH line.

A comparison of the LSD profile widths σ to the corresponding values for the FeH line is shown as a histogram of the corresponding broadening in Fig. 9, displayed as the ratio of the widths. It appears that in average the LSD profile is about twice larger than a single FeH line. We checked that this ratio does not significantly depend of the colour of the star.

This result is confirmed by an analysis of a BT-Settl synthetic spectrum at T_eff = 3500 K, [Fe/H]=0.0, and log g = 5.0, where we measure an average line width of 0.24 nm for 3 Ti I lines around 974 nm, and 0.11 nm for 2 FeH lines around 993 nm. A possible interpretation of this difference in line widths between atomic lines and molecular FeH lines comes from the low dissociation energy of FeH, namely 1.63 eV. So the molecule will be dissociated in regions where the turbulence is strong. A quick calculation gives a corresponding collision velocity of 2.4 km s⁻¹ . Higher velocity collisions would destroy the molecule and reduce the pressure broadening accordingly.

A similar diagram to Fig. 7 for the FeH line is displayed in Fig. 10 and shows a lower envelope which is flatter than for the LSD profile width and not defined very accurately, as a single line measurement is noisier than the LSD profile. This envelope is fit by equation (4):

5.4 Third approach: convolution with slow rotator templates

A different technique consists in using a few slow rotators with high S/N spectra obtained with ESPaDOnS and for which the value of vsin i is well measured by high S/N spectra at higher resolution. The method is described in details in Malo et al. (2014b) and uses six calibrators, ranging in spectral type from M1.0 to M3.5, listed in Table 7. The reference values of vsin i all come from Reiners (2007), who used very high-resolution spectra (200 000) from the CES spectrograph at La Silla Observatory, which ensures reliability and homogeneity. The S/N of the ESPaDOnS spectrum used as template is given in the last column of Table 7, and is measured per CCD pixel at 810 nm on the intensity spectrum.

For each calibrator, we artificially broaden its spectrum using different values of vsin i, and for each star in our sample, we look for the best fit of its spectrum among the library of broadened spectra of the calibrators. We then adopt as the value of vsin i for this spectrum the best match.

A comparison of the results of this technique with the value of vsin i calibrated from the measure of the width of the LSD profile gives a good agreement at intermediate projected rotation velocity (typically from 4 to 30 km s⁻¹ ). For slower rotators, there are differences due both to the calibration of σ_° for the LSD profile method, and the adopted template RV for the template method. For rapid rotators (and a few specific stars such as Gl 412B), non-Gaussian LSD profiles affect both methods and lead to differences between the two approaches too.

5.5 Adopted projected rotation velocity

From the three methods exposed above, we adopt a value of vsin i which is defined as follows, where vsin i_LSD is obtained from the calibrated LSD intensity profile, vsin i_FeH from the FeH line, and vsin i_c from the template convolution:

• All three methods are used and compared for each star, with the goal of obtaining a single value per star with an error bar representative of data quality, measurement dispersion, and calibration uncertainties.

• The median value of the three measurement is adopted, when vsin i_LSD is larger than 3 km s⁻¹ (resolved profiles) and vsin i_FeH is measured.

• When vsin i_LSD is smaller than 3 km s⁻¹ (unresolved profiles), vsin i_c is not included in the adopted value calculation.

• When vsin i_LSD was found smaller than 2 km s⁻¹ , we estimate that the rotation profile is unresolved in ESPaDOnS spectra and such values are reported as ‘<2’.

For stars with a strong magnetic field, vsin i_FeH from the FeH line should be preferred over the other two methods, as it is insensitive to the magnetic field. However, as the measurement is based on a single line it is more noisy, and in addition these stars are generally rapid rotators, which makes the line blended with nearby lines.

6 DISCUSSION

6.1 Comparison between projected and equatorial rotation velocities

We found about 150 stars in our sample with a known rotation period, either measured from time series photometry or spectroscopy of chromospheric indicators (Suárez Mascareño et al. 2015). We did not use rotation periods deduced from spectroscopic measurements when they are converted from chromospheric indicators such as |$R^{\prime }_{\rm HK}$| or projected rotation velocities vsin i. Uncertain values are given in parentheses. From this period and the adopted radius of the star, we compute the equatorial rotation velocity, using v_eq = 50.59 R/P_rot, where v_eq is in  km s⁻¹ , R in |$\mathcal {R}^{\rm N}_{{\odot }}$| (assumed to be 695 700 km from Prša et al. 2016), and P_rot in days. An alternative approach pioneered e.g. by Donati et al. (2008) consists in comparing Rsin i to the adopted radius, under the hypothesis that vsin i is measured more accurately than R, at least for rapid rotators. In our case, vsin i depends on the adopted calibrations and averaging process, so it is probably not more accurate than the star radius, which is estimated from the star colour V − J by a relation that we calibrated on interferometrically measured radii from Boyajian et al. (2012).

We divided our sample into two parts; the slow rotators (v_eq < 3 km s⁻¹ ), for which we want to check that small equatorial velocities are confirmed by a small value of vsin i from our measurement, and the resolved rotators, for which our measurement of vsin i should be smaller than the computed v_eq. Both tables are given in Appendix A.

We confirm that the calculus of v_eq from the estimated radius and measured P_rot agrees with our measured value of vsin i for average inclinations: about two-thirds of the expected slow rotators are not resolved with our spectrograph (vsin i < 2 km s⁻¹ ). Those having a measured value of vsin i may indicate that our calibration is slightly inaccurate (supposedly resolved projected rotation velocities are in fact upper limits). In a few cases, it may be due to a metallicity effect in the calibration of equations (1) and (4), which has not been taken into account and may affect metal-poor and metal-rich stars (see Melo et al. 2001, for an explanation of this expected effect).

However, in about half the cases of resolved rotators, vsin i taken at face value is larger than v_eq. It is not unexpected that the distribution of sin i is biased towards larger values, as there is an observational bias against low inclination systems where photometric variations are more difficult to detect. However, the magnitude of the effect is too large to be attributed to this bias. This surprising effect has already been evidenced by e.g. Reiners et al. (2012, see their fig. 10), who attribute it to possibly inaccurate photometric rotation periods. We can also add inaccurate radii, for instance for young stars, as we use a mean relation only valid for old stars. But these inaccuracies can probably only explain a few cases, not the majority.

6.2 From fundamental properties to radial-velocity uncertainty

Using the measured effective temperatures and collected apparent magnitudes in the H band, it was then possible to estimate the potential of SPIRou observations for this sample of stars. When effective temperatures were not available, we used first the values from Mann et al. (2015), then their equation (7) deriving T_eff from the V − J colour index and a correction for unknown metallicity based on the J − H colour index, with coefficients given in their table 2. We then use an exposure simulator for SPIRou to estimate the S/N obtained in a typical visit of 600 s integration time, with median seeing conditions of Maunakea (0.6 arcsec in the H band). Then, from the S/N estimates, we used the RV content as calculated in Figueira et al. (2016) to estimate the range of RV uncertainties per visit. The quantity depends upon the rotational velocity, the effective temperature, and the performance of telluric corrections, in addition to the S/N. In Fig. 12, we show two extreme conditions for each star where an effective temperature and rotational velocity are available: the conservative configuration where all regions contaminated by telluric lines more than 2 per cent depth are masked, and the optimistic configuration where these telluric lines are almost completely corrected for (see details in Figueira et al. 2016, their cases 2 and 3). It is difficult, at this point, to predict where telluric corrections with SPIRou will stand: the proposed method is a Principal Component Analysis (PCA)-based approach using a library of observed telluric spectra in varying conditions (Artigau et al., 2014); its performance in real conditions still needs to be assessed. As a first estimate, we used the RV uncertainty calculated for a rotational velocity of 1 (respectively, 10) km s⁻¹ for all stars having a vsin i less than (respectively, greater than) 5 km s⁻¹ , which explains why data points are not covering the parameter space randomly.

Finally, as the RV uncertainty is derived by photometric band, we applied the correction factor found for Barnard's star between models and observations of this M4 star (Artigau et al. submitted). These correction factors enhance the contribution of the H and K bands with respect to the bluer part of the spectrum; it is not yet known how they vary across the spectral type of M stars and with their metallicity.

Fig. 12 shows that an RV uncertainty of 1 m s⁻¹ is achieved for all slowly rotating stars brighter than an H magnitude of 7 in 600 s. Fainter stars, or faster rotators, would need a longer exposure time to achieve this precision. When the conservative approach of telluric masking is used, the limit drops by almost two magnitudes, showing the importance of devoting telescope time and pipeline development efforts to recover the stellar signal in these contaminated area. Finally, it seems that stars rotating at more than 10 km s⁻¹ will never achieve the 1 m s⁻¹ level, even when perfect telluric corrections are applied, down to an H magnitude of 4.5. This must be taken into account when considering the targets for planet searches.

6.3 Multiplicity and planet formation

Among the 153 systems listed in Table A2, more than half (88) have an apparent separation smaller than 2.0 arcsec, preventing in most cases a clear separation of the two components with our instruments, the fibre of which have diameters of 1.6 arcsec (ESPaDOnS) and 1.2 arcsec (SPIRou). Thebault & Haghighipour (2014) warn that RV surveys aiming at exoplanet detection reject binary systems and therefore prevent from getting information about the planet formation in such systems. They mention a physical separation of about 100 au below which the planet formation is affected. For the above-mentioned limit in angular separation (2.0 arcsec), this corresponds to a distance from Earth to the multiple system of 50 pc.

A more complete statistics has been drawn from our catalogue of multiple systems involving an M dwarf. Among 669 systems, 111 have a physical separation smaller than 100 au (assuming they are all physical systems). Among those, 28 are close enough to have an angular separation larger than 2.0 arcsec. This means that our observational constraints typically reject 75 per cent of the interesting sample where planet formation may be affected by the binarity.

Spectroscopic binaries are also rejected from most samples, especially SB2. In our sample of 440 M dwarfs, we listed 55 SB2 already known or discovered by us, a rate of 12.5 per cent. About a third of them are also close visual binaries (angular separation smaller than 2.0 arcsec), allowing a good determination of their physical properties.

In summary, about 80 per cent of interesting multiple systems for constraining the planet formation mechanism are lost to the size of the spectrograph fibers, linked to the atmospheric seeing.

7 SUMMARY AND CONCLUSION

In this paper, we have been reporting on a sample of 440 M dwarfs observed with the ESPaDOnS spectropolarimeter at CFHT. 114 of them correspond to observations conducted by our team in the framework of the CoolSnap collaboration. Two other papers (Moutou et al. 2017; Malo et al., in preparation) report additional results from this program. Another 71 stars observed in polarimetric mode and 255 in spectroscopic mode (S+S) were extracted from the ESPaDOnS archive at CADC and cover the whole set of observations of M dwarfs conducted at CFHT between 2005 and 2015.

From this homogeneous set of observations, we measured spectral type using the TiO₅ index, effective temperatures and metallicities using the mcal method when the star is not active (Hα index smaller than 0.25, see Section 4). We checked that our values generally agree with measurements obtained from similar or different methods in the literature.

As part of a larger project to identify multiple systems involving M dwarfs, we list all the stars in our sample belonging to such system, without limit on the separation. We also identify new spectroscopic binaries from our observations and summarize those already known from the literature.

We calibrate the measurement of the projected rotation velocity from the width of the LSD profile. This calibration is valid for other observations of late-type dwarfs observed with the ESPaDOnS spectropolarimeter.

Finally, we estimate the RV content for each star of our sample, in order to select those which are expected to display the smallest RV uncertainty possible with SPIRou. This work participates to the effort of selecting the targets for low-mass planet search using the new high-velocity precision near-infrared spectropolarimeter SPIRou. In the first paper, Moutou et al. (2017) defined a merit function based on the star activity; in this paper, we discarded close binaries and estimated the expected RV uncertainty; in the final paper of this series, Malo et al. (in preparation) use the present measurements of T_eff and [Fe/H] to refine the planet-detection merit function used to define the initial sample, and combines it to the other merit function and selection criteria to finally select the best sample of targets for the new SPIRou instrument.

Acknowledgements

The authors wish to recognize and acknowledge the very significant cultural role that the summit of Maunakea has always had within the indigenous Hawaiian community. We are most grateful to have the opportunity to conduct observations from this mountain.

We are sincerely grateful to the anonymous referee, whose careful reading and suggested additions and corrections helped and clarified the paper.

PF gratefully acknowledges Simon Prunet's help with the python codes and the entire CFHT ‘Ohana’ for offering excellent conditions of work.

This research has made use of:

• the Set of Identifications, Measurements and Bibliography for Astronomical Data (SIMBAD) data base and the VizieR catalogue access tool, operated at CDS, Strasbourg, France. The original description of the SIMBAD data base was published in A&AS, 143, 9 (2000), and of the VizieR service in A&AS, 143, 23 (2000);

• data products from the 2MASS, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation;

• data products from the AAVSO Photometric All Sky Survey, funded by the Robert Martin Ayers Sciences Fund and the National Science Foundation;

• data from the European Space Agency mission Gaia (http://www.cosmos.esa.int/gaia), processed by the Gaia Data Processing and Analysis Consortium (DPAC) (http://www.cosmos.esa.int/web/gaia/dpac/consortium). Funding for the DPAC has been provided by national institutions, in particular the institutions participating in the Gaia Multilateral Agreement;

• the facilities of the CADC operated by the National Research Council of Canada with the support of the Canadian Space Agency;

• the Washington Double Star Catalog maintained at the U.S. Naval Observatory.

This research has made use of the Brazilian time at CFHT through the agreement between the Brazilian Ministry of Science Technology Innovation and Communications (MCTIC) and the CFHT. EM acknowledges the support of CNPq, under the Universal grant process number 443557/2014-4, and also the support of FAPEMIG, under project number 01/2014-23092.

XD, TF, and FA received funding from the French Programme National de Physique Stellaire and the Programme National de Planétologie of CNRS (INSU). This work has been partially supported by the Labex OSUG@2020. The computations of atmosphere models were performed in part on the Milky Way supercomputer, which is funded by the Deutsche Forschungsgemeinschaft through the Collaborative Research Centre (SFB 881) ‘The Milky Way System’ (sub-project Z2) and hosted at the University of Heidelberg Computing Centre, and at the Pôle Scientifique de Modélisation Numérique at the École Normale Supérieure in Lyon, and at the Gesellschaft für Wissenschaftliche Datenverarbeitung Göttingen in collaboration with the Institut für Astrophysik Göttingen. This is also based on observations obtained at the CFHT which is operated by the National Research Council of Canada, the Institut National des Sciences de l'Univers of the Centre National de la Recherche Scientifique of France, and the University of Hawaii.

Footnotes

REFERENCES

SUPPORTING INFORMATION

Supplementary data are available at MNRAS online.

Table A6. Master list of data for the whole sample (440 dwarfs and 447 entries): number N of measured spectra, number n of rejected ones if any, instrument mode (P for polarimetry and/or S for S+S), spectral type (from TiO₅ index), V − K_s colour, H magnitude, HRV, projected rotational velocity and error in  km s⁻¹ (<2 if not resolved), Hα index, [Fe/H] and error, effective temperature and error in K, source code, RV uncertainty, and binarity flag (SB1, SB2, close VB).

Please note: Oxford University Press is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

APPENDIX A:

Long tables of the paper are given in theses appendices.

A1 Table of double-line spectroscopic binaries

In Table A1 are listed all the SB2 systems in our sample, detected in this work and from the literature.

A2 Table of multiple systems involving M dwarfs in our sample

In Table A2, we list 153 multiple systems from this compilation and involving at least one of the M dwarfs of our sample, detected by imagery, with the level of multiplicity and the component we measured in parentheses following the WDS notation; we also give the most recent projected separation and the corresponding position angle, or the semimajor axis when the orbit is known; in that case, the position angle is listed as ‘sma’; we list the physical status of the system (common proper motion and orbital monitoring), the year as given in the WDS, or the reference of discovery when more recent. For multiplicity larger than 2, we also list the separations and position angles for each pair composing the system (or the semimajor axes when ‘sma’ is listed as position angle), with a classical notation used to define the targeted pair (AB, Aab, Bab, etc.), following the WDS when possible. Twenty-three stars listed in Tables A1 and 2, which are only spectroscopic binaries are not repeated in this table, but spectroscopic binaries belonging to visual systems of higher multiplicity are included.

A3 Comparison of effective temperatures and metallicities between the present work and a reference (Mann et al. 2015)

Table A3 gives a comparison between our results for T_eff and [Fe/H] using the mcal method, with state of the art reference values taken from Mann et al. (2015).

A4 Comparison of equatorial and projected rotation velocities

Table A4 gives the comparison of equatorial rotation velocities computed from photometric rotation periods and radii, with projected rotation velocities for the slow rotators.

Table A5 gives the same comparison for the resolved rotators (v_eq > 3 km s⁻¹ ). For LP 193-584, the rotation period from Hartman et al. (2011) is uncertain and therefore given in parentheses, as well as the affected value of v_eq. For NLTT 3478, the very large difference between v_eq and vsin i would imply an improbable small value of the inclination. The photometric period should therefore be measured again.

A5 Master table of properties for the stars in our sample

A summary of the measurements for the whole sample of 440 stars is given in Table A6, an extract of which (0 < RA < 2 h) is given here. There are actually 447 entries, as observations for some stars in the initial sample but later rejected are listed, and some close binaries appear with two different entries, one for each component. For each star are listed the number of observed spectra, including those finally rejected for the measurement of metallicity and effective temperature, the spectroscopic mode of observation (polarimetric or S+S), the spectral type from the TiO₅ index, the V − K_s colour, the H magnitude from the 2MASS PSC, the HRV averaged over all spectra for the star (not given for SB2), the projected rotational velocity and its error (or <2 km s⁻¹ when unresolved), the Hα index (above 0.25, T_eff and [Fe/H] cannot be reliably measured by the mcal method), our mean value of [Fe/H] and T_eff for the inactive stars (Hα index <0.25), the predicted uncertainty of the RV assuming a full correction of telluric lines (see Section 6), and a binarity flag (SB1 for a single-line spectroscopic binary, SB2 for a multiple-lines spectroscopic binary, and VB for a visual binary with a projected separation smaller than 2.0 arcsec). The uncertainties on T_eff and [Fe/H] are computed from the individual internal uncertainties returned by the mcal method. They do not reflect systematic uncertainties associated with this method. Our measurements of T_eff and [Fe/H] are given the source code 1. When they are not available, we used the values listed in Mann et al. (2015) with a source code 2, or values without error derived using their equation (7) and coefficients in Table 2 with a source code 3. The full table is available online. The first page is displayed here to illustrate the format.