There are two basic arguments which justify finding nearby stars: detailed
study of individual stars, and stellar statistics.
1. Individually, the nearest stars (of a particular class) are the brightest stars, and therefore permit the most intense scrutiny of physical characteristics, and star-to-star variations in those characteristics. We note in passing that the fact that there are differences to investigate became clear with the completion of the first successful measurements of stellar parallax: while the two components in Alpha Centauri are similar in brightness to the Sun (Henderson's parallax, 1839), Bessel's 1838 measurements for 61 Cygni showed that the fainter star in that pair is ~35 times fainter than the Sun, while Vega (parallax by Struve, 1841) is brighter by almost the same factor.
2. Statistically, the scientific justification for compiling a catalogue of the nearest stars is summarised succinctly by Kuiper (1942): apart from illuminating details of stellar evolution through their distribution in the Hertzsprung-Russell diagram, the nearest stars provide the basis for the determination of the stellar luminosity function, the mass-luminosity relation, the stellar contribution to the local mass density, the velocity distribution and the stellar multiplicity statistics. Add information on chemical abundance, and the nearby stars map out the metallicity distribution of the (local) Galactic disk, while age estimates make these stars probes of the local star formation history, and the variation of stellar kinematics (and other parameters) as a function of time.
To provide useful conclusions, statistical studies should be based on an underlying dataset which provides an unbiased, representative sample of the parent sample, in this case, stars in the Solar Neighbourhood. One of the most effective means of satisfying this criterion is to identify a complete, volume-limited sample of stars: that is, a set of stars defined solely by where they are in the Galaxy, not by any characteristic (velocity, abundance, luminosity) peculiar to the stars themselves. There are still inherent difficulties in taking results derived from analysis of that sample and extending those conlusions to the global Galactic stellar populations: high velocity stars, particularly members of the halo (Population II), spend relatively little time near the mid-plane of the disk, and are correspondingly under-represented in a local sample; similarly, the Sun lies over 100 parsecs from the nearest active star-forming region, and very young objects are correspondly lacking in the Solar Neighbourhood. However, if those biases can be characterised to at least some extent, then the net effects can be taken into account in subsequent analysis.
The nearest stars - the 5 parsec sample: In this type of analysis, the trick is defining a volume-limited sample which, if not complete, is at least incomplete in a well understood manner. A first step is to limit the sample to stars on the Sun's doorstep - set a distance limit of only a few parsecs. One of the first essays in this regard was by Hertzsprung (1922), who produced a list of 29 stars with measured parallaxes exceeding 0.2 arcseconds, placing them within 5 parsecs of the Sun. The 5 parsec limit became somewhat of a fiducial marker for the nearby star census. Hertzsprung's original catalogue corresponds to a star density of 0.055 stars/pc3, counting separately individual stars in multiple systems. van Maanen (1933) extended the sample to 39 stars (0.074 stars/pc3), but the sample had shrunk to 34 stars when van de Kamp took over the book-keeping (van de Kamp, 1940). van de Kamp's sample increased to 39 stars in 1945 and 42 stars in 1953.
Over the next decade, van de Kamp extended the distance limit to 5.2 parsecs, or 17 light years, and by the 1960s, the sample included 59 stars in 44 systems. These comprised 31 single stars (including the Sun), 11 binaries and two triple systems (van de Kamp, 1969). Since the 60s, the accretion rate has decreased: van de Kamp & Lippincott (1975) list 60 stars within 5.2 parsecs in their review article; the present total, including both new discoveries and the more accurate distances provided by the Hipparcos satellite, has reached 69 stars, but in only 40 systems. Setting the limit at 5 parsecs, the total is 63 stars in 37 stellar systems (including the Sun). Of these, the most recent additions are LHS 1565 (Henry et al, 1996), LP944-20 (Tinney, 1998) and LHS 2090 (Scholz et al, 2001). The first and last are mid-type M dwarfs (M5.5 and M6.5, respectively), and lie at distances of only ~4 parsecs; LP 944-20 is spectral type M9, lies at a distance of 4.97 parsecs and is a confirmed brown dwarf; together, these recent discoveries provide a clear hint that the goal of completeness may yet to be attained in surveying even the nearest `stars'. The current census corresponds to a star density of 0.12 stars/pc3, more than double the density derived by Hertzspring.
The 8 parsec sample: A distance limit of 5 parsecs offers advantages and disadvantages: the main advantage is high probability of detection; the primary disadvantage is small sample size. Given the advances in technology made in the last decades of the twentieth century, with the availability of all-sky surveys at wavelengths from X-ray through optical and near-infrared to radio, it has become possible to consider extending the scope of the nearest-star census. A distance limit of 8 parsecs marks a reasonable second step since the increase corresponds to quadrupling the survey volume over a 5-parsec sample. However, with that increase in distance goes a reduction in areal coverage: in brief, the southernmost stars have been surveyed less thoroughly for companions, and statistical analyses are therefore best confined to stars accessible from northern hemisphere observatories. The following table outlines the salient statistics for the current (23/6/2001) 5-parsec and 8-parsec sample, comparing the relative numbers of systems and individual stars north and south of decline -30o. The relative areal coverage is 3:1, with the "northern" (dec > -30o) sample covering 3-pi steradians; thus the ratios N(south)/N(north) should be close to 0.333
|8 vs 5||North||0.31+/-0.05||0.32+/-0.05|
A major factor in the incompleteness of the companion-star census at southern declinations is accessibility: the overwhelming majority of the Earth's landmass is north of the equator, and, as a result, so are the majority of the astronomical observatories. In particular, both Palomar Observatory, the site of the 48-inch Schmidt and Luyten's wide-field proper motion surveys, and Steward Observatory, site of Henry's (1991) speckle imaging survey of nearby stars, lie at latitude +33o. With the current emphasis on high-resolution adaptive optics, and the presence of active observatories in Chile (ESO, CTIO, Las Campanas), Australia (AAT) and South Africa (SAAO/yyy), this borealic bias should be remedied in the near future, partly as a result of programs initiated under the NStars aegeis; however, for the present,a line drawn at -30o declination is the most expedient approach.
The data listed in Table P1 can be used to derive a second statistic: we noted above that the volume embraced by an 8-parsec sample is approximately four times that of a 5-parsec survey (2144.6 vs 523.5 pc3). In that case, one would expect a factor of four increase in the number of stars/systems as rlim is increased from 5 to 8 parsecs; in fact, as Table P1 shows, the ratios are closer to 3:1. Does this represent residual incompleteness in the 8 parsec sample or a local overdensity in the 5 parsec sample? Henry et al (1994) favour the former interpretation; we tend to suspect the latter. Our analysis is given at this site , where we present data for our sample of stars (and brown dwarfs) within 8 parsecs of the Sun.
Further out and further up - not-quite-so-near stars: An alternative approach to increasing the statistical significance of nearby star studies is to set a larger distance limit for the sample as a whole, but vary that limit as a function of absolute magnitude of the individual stars in subsequent analysis. Kuiper (1942) was probably the first to adopt this approach, implicitly if not explicitly; moreover, his is the first sample to include a substantial number of stars with distances derived from spectroscopic parallaxes. [One might note that the luminosity function derived from the latter sample presages the over-abundance of M dwarfs which typfied similar analyses in the 1960s.] Kuiper's sample includes 254 stars with estimated or measured parallaxes in excess of 0.095 arcseconds.
Probably the prime exponent of this approach to constructing a local census was Wilhelm Gliese. His first compilation catalogue included basic data (astrometry, photometry and spectral types) and inferred parameters (luminosity, kinematics) for 1094 stars in 915 systems with trigonometric, photometric or spectroscopic parallaxes greater than 49 milliarcseconds, or distances of up to 20 parsecs from the Sun (Gliese, 1957: CNS1). These are the original Gliese stars, ordered in Right Ascension (equinox 1950) and designated Gl 1, Gl 2 ... Gl 915.
Extending the distance limit to 22 parsecs (parallaxes exceeding 44 mas), Gliese increased the sample size to 1890 stars in 1529 systems, and data for those stars were published in 1969 as the Catalogue of Nearby Stars (Gliese, 1969: CNS2). These stars were again catalogued in RA order, with the addtions designated as Gl nnn.1, Gl nnn.2 etc: that is, Gl 268.1 lies between Gl 268 and Gl 269 in RA; Gl 268.2 is east of Gl 268.1, but still west of Gl 269, etc. That publication prompted the then director of the Royal Greenwich Observatory and Astronomer Royal, R. v.d.r. Woolley, to tackle this issue, and RGO published its version of a nearby star census (to r=25 pc) the following year (Woolley, 1970). Subsequent observations have shown that many of the RGO stars (designated Wo nnnn) are well outwith the nominal 25 parsec limit.
In the years following the publication of the CNS2, surveys probed to greater depth, and follow-up observations, particularly of stars from Luyten's proper motion catalogues, revealed more candidates for the immediate Solar Neighbourhood. A decade after the completion of the CNS2, Gliese & Jahreiss (1979) published a supplement, including data for a further 462 stars: 294 had formal parallaxes exceeding 44 mas, and were identified as GJ 1nnn; a further 159 had incomplete data, but were likely to be within the 22 parsec distance limit, and were designated GJ 2nnn. The catalogue also included dat for nine new companions of previously-catalogued stars, and new data for 377 Gliese stars.
Data continued to accumulate during the 80s, and Gliese and Jahreiss continued to compile those results, producing an electronic version of a third version of the catalogue in the early 90s (Jahreiss & Gliese, 1991, pCNS3). The distance limit for that catalogue was extended from 22 to 25 parsecs, i.e. parallaxes > 39 milliarcseconds. As with the CNS1 and CNS2, distances are derived using trigonometric, photometric and spectroscopic techniques. However, the separate estimates were not averaged, as had been the case in CNS1 and CNS2: if the formal relative error in the trigonometric parallax was less than 14%, that value was adopted; spectroscopic and/or photometric estimates were adopted for stars with low accuracy astrometry. There are 3803 entries, including data for 3845 components of xxxx systems. This catalogue was never published officially, but has served as a reference for several large-scale projects, notably our PMSU survey, described further below. Jahreiss is currently working on the CNS4, incorporating new data, notably high accuracy astrometry from the Hipparcos satellite, together with the results of follow-up observations of the new infrared surveys (eg Delfosse et al, 1998; Scholz et al, 2001).
The nearby-star samples presented in the CNS1, CNS2 (and supplement) and pCNS3 are compilations, drawn from the astronomical literature. Each star in those catalogues has a particular characteristic which indicates a parallax (astrometric, photometric or spectroscopic) exceeding 40 milliarcseconds, and therefore a distance of less than 25 parsecs. However, data for individual stars and stellar systems are drawn from a wide range of sources, and span a wide range of quality; moreover, the different techniques used to estimate distances have themselves a variety of associated uncertainties. The situation is particularly acute at low luminosities, where many of the pCNS3 nearby-star candidates lacked even a spectral type estimate.
Given these circumstances, we embarked on a spectroscopic survey of the ~2400 stars in the pCNS3 which had nominal absolute magnitude estimates of MV > 7.5, and were not classed explicitly as degenerate white dwarfs. Our choice of spectroscopic rather than photometric observations was driven partly by pragmatism (quantitative spectroscopy is possible over a wider variety of conditions than quantitative photometry, while direct astrometry for so many targets imposes a prohibitive cost in telescope time), but also by the wider possibilities for probing stellar physics afforded by that technique. Spectroscopic bandstrength indices can not only provide quantitative estimates of effective temperature and luminosity, but also probe chemical abundance and chromospheric activity. These measurements provide a homogeneous set of distance estimates for late-type stars, permitting a self-consistent appraisal of membership the nearby star sample. The main observational results from our survey are presented in two papers - Reid, Hawley & Gizis (1995) and Hawley, Gizis & Reid (1996).
Figure P.1: Definition of the narrowband indices used to measure TiO, CaH and CaOH bandstrength in the PMSU survey
Our primary method of calibration uses narrowband spectroscopic indices, designed to measure the depth of the strong molecular absorption bands due to titanium oxide (TiO0, calcium hydride (CaH0 and calcium hydroxide (CaOH). Bandstrengths are determined by measuring the ratio between the flux within a small wavelength region centred on the feature in question, and a nearby pseudo-continuum point. These are essentially narrowband colour indices, but with the advantage that data are taken simultaneously at all wavelengths; thus, we only require accurate relative flux calibration, rather than the absolute calibration (and hence photometric conditions) demanded for conventional photometry. We chose to concentrate on the 6000 to 7500 Angstrom region of the spectrum, a region which encompasses several TiO features, notably the strong 7050 A bandhead, besides CaH and CaOH. Figure P.1 illustrates the techniques, and the following table gives the wavelength regions chosen for in-band and continuum points.
Figure P.2: The correlation between TiO5 bandstrength and spectral type for M dwarfs. Note the reversal at ~M6.5.
The TiO5 index is particularly useful for calibration purposes. Figure P.2 plots the relation between TiO5 and spectral type, calibrated using standard stars where the spectral type has been derived from inspection of the full spectrum (mainly stars from Kirkpatrick et al, 1991). Note that the TiO5 index reaches maximum strength at a spectral type of ~M6.5, becoming weaker at later spectral types. This reflects both saturation in the TiO band and the growing strength of additional absorption, mainly due to vanadium oxide, within the pseudo-continuum band. Formally, this means that TiO5 is double-valued with spectral type; other features in the spectrum, notably VO absorption, allow segregation of stars earlier or later than spectral type M6.
Figure P.3: Metallicity effects: CaH bandstrength as a function of TiO5 for disk dwarfs (open squares), sdM dwarfs (crosses) and esdM dwarfs (solid points). Data for the subdwarfs are from Gizis (1997).
Spectral type is correlated primarily with effective temperature. The strong correlation evident in Figure P.2, between TiO5 and overall type, emphasises how features vary together, in lock-step, with decreasing temperature - if all other parameters are approximately equal. All parameters, however, are not always equal; in particular, changes in chemical abundance can change significantly the emergent spectrum, at almost all temperatures. In the case of cool M dwarfs, the most striking effect is a change in the relative strength of the TiO and metal hydride bandstrengths; in part this reflects the fact that TiO is a double metal, hence doubly sensitive to decreasing metallicity; in part, it reflects the overwhelming presence of H, which locks up much of the available oxygen in water. The net result is that K-type metal-poor dwarfs retain strong features of MgH (near 5200 Angstroms), while later-type M subdwarfs have correspondingly strong CaH features (as illustrated in Figure P.1). The variations in the strength of these features has been quantified by Gizis (1997), and is illustrated in Figure P.3. Gizis separates the metal-poor M dwarfs into two sub-types: intermediate subdwarfs, sdM, which are likely to have an average abundance of [M/H]~-1, and extreme subdwarfs, esdM, which probably have a mean abundance closer to [M/H]=-2. The overwhelming majority of stars in the Solar Neighbourhood, and in the pCNS3, are disk dwarfs.
Figure P.4: MV as a function of TiO5 and CaH2 bandstrength. Magenta crosses mark disk dwarfs from the 8-parsec sample; cyan stars are sdM subdwarfs; green circles are esdM subdwarfs. Note that subdwarfs are significantly closer to the disk main sequence in the latter diagram, showing that CaH2 is less susceptible to abundance variations and a better absolute magnitude estimator
Since the bandstrengths are primarily temperature dependent, they offer the prospect of photometric parallax estimation, just as with conventional broadband colours. Indeed, as Figure P.4 shows, the hydride bands are more effective than conventional colours (save, perhaps, (R-I)), in being almost independent of metallicity. These calibrations therefore offer the possibility of estimating distances to all of the M dwarfs in the pCNS3.
The starting point for our survey was the sample of 2227 stars from the pCNS3 which were either listed explicitly as M dwarfs (spectral type M0 or later, or simply m), or which were identified as having absolute magnitudes MV > +8.0 and were not listed as white dwarfs. Of these stars, 1876 lie north of declination -30o, and those stars were discussed in Paper I (Reid et al, 1995 - PMSU1); the remaining 351 were included in PMSU2 (Hawley et al, 1996). Most of the observations for PMSU1 were made using the Palomar 60-inch telescope, supplemented by the 200-inch and Keck for the faintest stars, while the southern stars were observed from CTIO. Our spectra show that 61 of the PMSU1 stars and 30 PMSU2 stars are misclassified early-type stars, white dwarfs or M giants. We were unable to obtain data for 130 stars listed in PMSU1, since those stars are close companions of much brighter stars; the same holds for only a dozen stars in PMSU2. The relatively smaller number in the latter sample is more likely due to incompleteness in companion searches in the south than to a real deficit of companions. That incompleteness, and the absence of a deep proper motion survey similar to Luyten's work from Palomar, led us to restrict the main statistical analysis to the "northern" (or at least "accessible from the northern hemisphere") sample included in PMSU1. Some of those results are summarised here,
Figure P.5: Comparison between distance estimates in pCNS3 and subsequent measurements based on Hipparcos parallax data. the lower panel shows that there is no systematic offset as a function of MV
Before embarking on that summary, we briefly consider the impact of new data. Since the completion of our survey, new observations have become available for many stars, notably the astrometric results from the Hipparcos survey, while high spatial-resolution adaptive optics work and high spectral resolution radial velocity surveys have turned up many new companions to known nearby stars. The main result of the Hipparcos observationss has been to identify a significant number of pCNS3 stars with parallaxes of less than 39 mas, removing those stars from the 25-parsec sample. Figure P.5 compares pre- and post-Hipparcos distance estimates for pCNS3 stars. This result is not unexpected, since volume sampling effects will tend to produce this kind of bias in parallax measurements: since there are more stars with pi < pi0 than pi > pi0, symmetric errors will scatter a larger number of distant stars into a sample than nearby stars out of a sample. On the other hand, Hipparcos only added a small number of stars to the nearby sample, at least partly because the mission involved a pointed survey: Hipparcos targeted specific objects, becoming incomplete for stars fainter than V~8 (the limit is galactic latitude dependent) and observing very few stars fainter than V=12 (and none fainter than V=13). Thus, while Hipparcos could target all stars suspected a priori of being within 25 parsecs, it provides only inomplete coverage of stars not suspected of being within 25 parsecs, but which actually are. Thus, care is required in selecting a statistical sample for analysis.
Finally, more accurate radial velocities are now available for many stars - notably from Delfosse et al (1998) and from the P60 echelle data included in PMSU3 (Gizis et al, 2002). Those data, together with the revised distances, lead to revised estimates of space motions for several hundred stars.
We have incorporated all of these new measurements in revised versions of the relevant data tables from PMSU1 and PMSU2. Those tables are available at the bottom of this page, both as ascii files and in html format.
Our survey includes observations of 1684 confirmed M dwarfs north of decline -30o. Those stars include apparently-single M dwarfs, M dwarf companions of earlier-type stars and M dwarf binary, triple and quadruple systems. Figure P.6 plots the distance distribution as a function of absolute magnitude, making no distinction amongst these categories. Clearly, the sample becomes increasingly less complete with increasing distance at fainter absolute magnitudes. That bias needs to be taken into account before attempting statistical analysis of the sample.
Figure P.7: The run of system density with distance: these data are computed from the M-dwarf system sample, including only single M dwarfs and systems with M dwarf primary stars. The absolute magnitude refers to the brightest star in the system. The vertical line marks the distance chosen to represent the completeness limit. Note that there are only 4 systems in the MV=15.5 `complete' sample.
One method of setting compelteness limits is to look at the run of density with increasing distances as a function of absolute magnitude: we expect initial fluctuations, a flattening, and then a downturn as the sample becomes incomplete. Figure P.7 plots those data for the PMSU sample, where the densities are the number of systems per unit volume, grouped by the absolute magnitude of the brightest star in the system. The vertical lines mark the location of our adopted completeness limits. Those values are identical to the PMSU limits except for MV=9.5, where we have reduced the limit from 20 to 18 parsecs. The statistics are as follows:
|MV||dlim (pc)||Nsys||density||Nstars||density||N Hip|
We adopt the same distance limits for companion stars as primaries, although in principle the former might be expected to have more distant completeness limits, dependent more on the absolute magnitude of the primary. Note the sharp drop in the density profile for MV=15: the high local density is contributed by 4 systems. It is possible that the appropriate completeness limits is actually 8-9 parsecs (see our 8-parsec sample discussion). We also list the number of Hipparcos stars which fall within these distance limits but were not included in the pCNS3 (see Table 1D, PMSU2).
We have recently combined the revised PMSU sample with an Hipparcos-defined sample of stars with d < 25 parsecs and MV < 8.0, and used those data to determine the nearby-star luminosity function for -1 < MV < 18, the present-day mass function and the initial mass function. See this page for a full discussion.
Figure P.8: The stellar luminosity function: the blue histogram plots data for systems (M dwarf single stars and primaries); the magenta histogram includes all components. Green points (middle diagram) show an averaged photometric luminosity function; the points in the lowermost diagram plot data for Wielen's (1974) luminosity function.
Given a volume complete sample, we can determine systemic and stellar (star-by-star) volume densities, and those data are listed in the table. The resultant luminosity function is plotted in Figure P.8, and compared with data from photometric surveys (i.e. deep, moderately wide-field starcounts, using photometric parallaxes to estimate distances), and against Wielen's (1974) original analysis of the CNS2. The results are entirely compatible with our analysis from PMSU1: a broad maximum near MV=12.5, but not as pronounced as in the photometric studies. The overall densities are lower than those derived by Wielen (and note the large uncertainties in that analysis). The reasons for the difference with respect to the photometric analyses are discussed by Reid & Gizis (1997), and most likely stem from Malmquist effects introduced by the sharp jump in the (MV, (V-I)) diagram at (V-I)~3; the lower densities compare to Wielen reflect the gradual reduction of the sample as stars with over-estimated parallaxes are removed.
These data also permit investigation of the statistics of stellar motions, the local stellar kinematics. The (V, U) and (V, W) velocity distributions for the complete system samples are plotted in figure P.9, where U is the velocity towards the Galactic centre, V the velocity in the direction of rotation and W the velocity towards the North Galactic Pole.