The Canonical Luminous Blue Variable AG Car and Its Neighbor Hen 3-519 Are Much Closer than Previously Assumed

Eruptive mass loss exhibited by luminous blue variables (LBVs) is thought to be important for the evolution of massive stars (Humphreys & Davidson 1994; Smith & Owocki 2006; Smith 2014), but the exact role LBVs play and the physics of their instability has been challenging to understand. The class of LBVs defined in the Milky Way and Magellanic Clouds (LMC/SMC) is thought to be responsible for some extragalactic non-supernova (SN) transients (Smith et al. 2011; Van Dyk & Matheson 2012). The episodic mass loss of LBVs has also been a reference point for interpreting the dense circumstellar material (CSM) around Type IIn supernovae (SNe IIn), which is thought to be indicative of extreme pre-SN eruptions (Smith 2014).

The diverse collection of objects known collectively as LBVs was first proposed as a group by Conti (1984), and the now standard interpretation of these stars and their role in evolution was established through the 1980s and 1990s. The review by Humphreys & Davidson (1994) provides a summary of the traditional view of LBVs, which has a few key tenets:

1. LBVs are thought to represent a brief transitional phase in the evolution of the most massive stars, between the main sequence O-type stars and the H-deficient Wolf-Rayet (WR) stars. Because they have already lost some mass and because their core luminosity has increased, their L/M ratio is high. This proximity to the classical Eddington limit, combined with Fe opacity, leads to a violent instability in the envelope that somehow triggers runaway Geyser-like mass loss. The heavy mass loss of LBVs, in the single-star view, is essential to remove the H envelope to form WR stars.

2. Observed S Doradus-type variations, as exemplified most clearly by AG Car in the Milky Way and R127 in the LMC (Wolf & Stahl 1982; Stahl et al. 1983), are temperature variations at constant bolometric luminosity. The cool (visibly bright) state is caused by a pseudo-photosphere that develops in an optically thick wind, which occurs at the same temperature regardless of luminosity (Davidson 1987; Humphreys & Davidson 1994). In their quiescent (hot) phase, all LBVs reside on the diagonal "S Doradus instability strip" (Wolf 1989) on the Hertzsprung–Russell (HR) diagram.

3. The strong mass loss of LBVs halts their redward evolution, preventing them from becoming red supergiants (RSGs). This explains the observed absence of high-luminosity RSGs above log(L/L_⊙) = 5.8, due to instability and mass loss in evolved massive single stars.

4. Some LBVs suffer dramatic "giant eruptions" where huge amounts of mass can be lost. Although there is a formalism to explain the mechanism of this mass loss with super-Eddington continuum-driven winds (Owocki et al. 2004), the underlying reason why single stars suddenly exceed the classical Eddington limit by factors of 5–10 has remained unexplained. Nevertheless, this tremendous mass loss is observed and is likely to be important in the evolution of massive stars.

Several aspects of this traditional view, however, have unraveled with time. In particular, the important conjecture that S Dor variations are major mass-loss events caused by optically thick winds appears to be wrong, and steady super-Eddington winds might not be the dominant driving mechanism in many giant LBV eruptions (see review by Smith 2014). Moreover, the recognition that SNe IIn have progenitors consistent with LBVs (see, e.g., Mauerhan et al. 2013) casts doubt on the idea that LBVs are only a brief transitional phase before the start of core He burning. Investigations with quantitative spectroscopy (de Koter et al. 1996; Groh et al. 2009a, 2009b) disproved the conjecture that S Dor brightening events are caused by pseudo-photospheres in optically thick winds (Davidson 1987). The mass-loss rates of S Dor maxima are not high enough to make such extreme pseudo-photospheres, and so they are more likely to be caused by envelope inflation or pulsation (Graefener et al. 2012). Also, bolometric luminosities during S Dor eruptions are not necessarily constant (Groh et al. 2009a). Similarly, the idea that giant-eruption maxima are caused by pseudo-photospheres in super-Eddington winds is challenged by light-echo spectra of η Carinae (Rest et al. 2012; Prieto et al. 2014; although see Owocki & Shaviv 2016), by detailed analysis of the ejecta around η Car that are better matched by an explosive event (Smith 2006b, 2008, 2013), and the fact that many extragalactic giant LBV eruptions are relatively hot, rather than cool, at peak luminosity (Smith et al. 2011; Mauerhan et al. 2015). Last, as discussed by Smith & Tombleson (2015), their isolation from massive O-type stars suggests that they are much older than expected. Smith & Tombleson (2015) suggested that they are largely products of binary evolution and not a transitional state in the lives of the most massive single stars.

Here we add another possible wrinkle to the unraveling story of LBVs. One of the canonical, classical high-luminosity LBVs in the Milky Way is AG Carinae; from a direct measurement of its parallax from Gaia this star appears to be much closer—and therefore much less luminous—than previously thought. Even though AG Car is a defining S Doradus variable, this closer distance moves it well below the S Doradus instability strip established for LBVs in the LMC (Wolf 1989). Although the distance to LBVs in the LMC is not that uncertain, the distance to Galactic LBVs has always been precarious. Below we discuss AG Car and the three other Galactic LBV-like stars that are included in the Gaia first data release.

The Gaia first data release (DR1) has recently provided trigonometric parallaxes and proper motions for the ∼2 million stars previously observed by Hipparcos and Tycho-2, with a typical precision in the parallax of ∼0.25 mas (Gaia Collaboration et al. 2016). These parallaxes are enabling a number of investigations not previously possible, such as direct measurement of the basic properties of all planets and their stellar hosts (Stassun et al. 2016). These fundamental new measurements also provide an opportunity to reassess the nature of these LBVs.

The data from Gaia DR1 and from the literature that we use in this study are described in Section 2, which in particular includes parallaxes and proper motions for the four Milky Way LBVs Hen 3-519, HD 168607, HR Car, and AG Car. Section 3 presents the principal results of this study, namely, the distances, space motions, and other basic characteristics of these LBVs, as well as the uncertainty in these values. Section 4 discusses the implications of our findings, especially the distances for AG Car and Hen 3-519, which we find to be much closer than previously thought. We conclude with a brief summary of our results in Section 5.

We searched the Gaia DR1 catalog for all of the Milky Way LBVs and LBV candidates listed in Smith & Tombleson (2015), which yielded parallaxes and proper motions for four: Hen 3-519, HD 168607, HR Car, and AG Car (see Table 1). From the parallax, π, a distance may be straightforwardly computed via d = 1/π. The parallaxes of the four LBVs span the range 0.34–1.02 mas, corresponding to distances spanning the range ≈1–3 kpc.

Table 1.  Data Used for the LBVs in Our Study Sample

Name	Hipp. ID	TYC ID	Gaia DR1						Astraatmadja & Bailer-Jones (2016)
			π (mas)	σπ (mas)	μα (mas yr−1)	${\sigma }_{{\mu }_{\alpha }}$ (mas yr−1)	μδ (mas yr−1)	${\sigma }_{{\mu }_{\delta }}$ (mas yr−1)	d (pc)	σd (pc)	d5% (pc)	d95% (pc)
Hen 3-519	...	8958-1166-1	0.796	0.575	−4.282	1.799	3.904	0.853	1600	859	750	3575
HD 168607	89956	...	1.016	0.282	0.404	0.102	−1.314	0.063	1156	346	754	1891
HR Car	50843	...	0.342	0.239	−5.813	0.069	2.902	0.065	2267	973	1407	4607
AG Car	53461	...	0.400	0.225	−4.861	0.079	1.923	0.092	1953	726	1317	3705

Download table as:  ASCIITypeset image

However, because the parallax measurement errors are relatively large (28%–72%), the distances and their uncertainties are expected to depend on the adopted prior (see, e.g., Bailer-Jones 2015). Therefore, we have also retrieved the distances computed by Astraatmadja & Bailer-Jones (2016) using a more realistic prior distribution based on simple but empirically motivated stellar density distributions for the Milky Way. These distances, which span the range 1.2–2.3 kpc (Table 1), are consistent within the uncertainties with those obtained simply via 1/π, but with somewhat smaller uncertainties due to the more informative prior used. Thus we prefer the Astraatmadja & Bailer-Jones (2016, hereafter, ABJ) distances in what follows but note that our primary conclusions do not depend strongly on this choice.

The Gaia DR1 release notes state that the parallaxes may possess uncharacterized systematic uncertainties of up to 0.3 mas, and recommend adding an additional 0.3 mas to the reported measurement uncertainty. Consequently, ABJ provide a version of their catalog that includes this additional 0.3 mas error. However, Stassun & Torres (2016b) used a set of nearby, benchmark eclipsing binary stars (Stassun & Torres 2016a) to quantify the systematic error, finding it to be −0.25 ± 0.05 mas in the sense that the Gaia parallaxes are systematically too small (distances too long). Jao et al. (2016) found a similar offset among a sample of nearby, high-proper-motion stars in the solar neighborhood. However, at larger distances, Stassun & Torres (2016a) found that the systematic offset vanishes for π ≲ 1 mas (d ≳ 1 kpc). This is corroborated by Casertano et al. (2016), who found excellent agreement between the Gaia distances and a large sample of Galactic Cepheids at d ∼ 2 kpc. Lindegren et al. (2016) and Sesar et al. (2016) similarly use samples such as RR Lyrae stars at intermediate distances to argue that no correction is needed beyond ∼1 kpc. Finally, Davies et al. (2017) uses red clump stars over a large range of galactic distances to broadly confirm the above findings, with an offset similar to that of Stassun & Torres (2016a) for nearby stars that vanishes at d ≳ 1.2 kpc. Therefore, since the distances of the LBVs in our study sample are all greater than ∼1 kpc (Table 1), we do not apply any offset to the parallaxes and moreover do not add any further systematic error to the reported measurement uncertainties.

HD 168607: This LBV is located only about 1' away from HD 168625, which is well-known for its SN 1987A-like triple-ring nebula (Smith 2007). HD 168607 is considered an LBV based on its characteristic variability, whereas its neighbor HD 168625 is usually considered an LBV candidate (Sterken et al. 1999). Both are found in the outskirts of the star-forming region M17 and are thought to be part of the larger Ser OB1 association at ∼2.2 kpc (Chentsov & Gorda 2004). A distance of ∼2.2 kpc is usually adopted in the literature (van Genderen et al. 1992), although distances of 1.2–2.8 kpc have been proposed (Robberto & Herbst 1998; Pasquali et al. 2002). The Gaia DR1 distance to HD 168607 from Table 1 is 0.98 ± 0.27 kpc (π = 1.02 ± 0.28 mas) or 1.16 ± 0.35 kpc (π = 0.87 ± 0.26 mas) using the ABJ parallaxes. This is on the low end of previously adopted values that are usually close to 2 kpc, but not too far off. The 95% upper limit from the ABJ parallaxes is 1.9 kpc. This is unlikely to prompt a major revision in the interpretation of this object. We postpone a detailed discussion until the higher precision that will be available in the next Gaia data release. HD 168607 appears to have a low transverse velocity of only a few km s⁻¹ indicated by its Gaia proper motion.

HR Car: The distance adopted in the literature for HR Car is usually 5 ± 1 kpc (e.g., van Genderen et al. 1991; Groh et al. 2009a), placing it among the low-luminosity group of LBVs (see, e.g., Smith et al. 2004). The Gaia distance in Table 1 is 2.92 ± 2.04 kpc (π = 0.34 ± 0.24 mas) or 2.27 ± 0.97 kpc (π = 0.44 ± 0.19 mas) according to the ABJ parallaxes, which is lower than previously assumed, but the 95% upper limit of 4.6 kpc is arguably consistent with the usual estimate. Given that HR Car is also now known to have a resolved wide companion about 3 mas away (Boffin et al. 2016), which may complicate the measured parallax, we do not advocate a major revision of its distance or luminosity at this time. Again, we postpone a detailed discussion pending the higher precision that will be available in the next Gaia data release. The Gaia absolute proper motion of HR Car is about 6.5 ± 0.16 mas yr⁻¹ (PA ≈ 297°), which, at a distance of ∼2.5 kpc, translates to a transverse velocity of about 72 km s⁻¹.

AG Car: Although AG Car is seen in projection amid the Car OB1/OB2 association, located at 2–2.5 kpc, a larger distance has usually been adopted in the literature, making AG Car one of the most luminous stars in the Milky Way. The larger distance is based on its radial velocity relative to the local standard of rest (LSR) compared to the Galactic rotation curve, as well as its high line-of-sight extinction, from which Humphreys et al. (1989) derived a likely distance of 6.4–6.9 kpc. However, the new Gaia distance is 2.50 ± 1.41 kpc (π = 0.40 ± 0.23 mas) or 1.95 ± 0.73 kpc (π = 0.51 ± 0.19 mas) according to the ABJ parallaxes, with a 95% upper limit of 3.7 kpc (Table 1). This would appear to rule out the larger distance above 6 kpc derived by Humphreys et al. (1989), instead suggesting membership in the closer Car OB1/OB2 association after all. If even approximately correct, this much closer distance has profound implications for our interpretation of AG Car, and consequences for our understanding of LBVs in general, as discussed in the next section. At a distance of ∼2 kpc, the measured Gaia absolute proper motion of 5.2 ± 0.26 mas yr⁻¹ (to the west/northwest; PA ≈ 292°) would suggest a transverse velocity of about 50 km s⁻¹.

Hen 3-519: Hen 3-519 is located very close in the sky to AG Car (about 20' away) and would appear to be part of the same Car OB1/OB2 association, but like AG Car, previous authors have generally favored a very large distance near 8 kpc. The large 8 kpc distance was proposed by Davidson et al. (1993) based on the large line-of-sight extinction inferred from UV data. Smith et al. (1994) found a slightly lower reddening, but also favored a large distance near 8 kpc based on the LSR velocity implied by nebular emission and interstellar absorption lines in high-resolution spectra. Again, this would make Hen 3-519 an extremely luminous star. Like AG Car, however, parallax once again indicates a much smaller distance. The Gaia distance is 1.26 ± 0.91 kpc (π = 0.80 ± 0.58 mas) or 1.60 ± 0.86 kpc (π = 0.63 ± 0.34 mas) according to the ABJ parallaxes, with a 95% upper limit from the ABJ parallaxes of 3.6 kpc (Table 1). This would appear to rule out the larger distance of 8 kpc proposed by Davidson et al. (1993), instead suggesting membership in the closer Car OB1/OB2 association, just like its neighbor AG Car. At a distance of ∼2 kpc, the measured Gaia absolute proper motion of 5.8 ± 2.7 mas yr⁻¹ (PA ≈ 312°) would suggest a transverse velocity of 56 km s⁻¹. This is, again, very similar in magnitude and direction to the motion of AG Car, although with larger uncertainty.

The three targets in Carina all exhibit a similar absolute proper motion of roughly 5 mas yr⁻¹ to the west/northwest. This corresponds to apparent motion in a direction along the Galactic plane. It is roughly consistent with the direction and magnitude of longitude drift expected for Galactic rotation on circular orbits in the solar neighborhood. At ∼2 kpc, objects on the near side of the Carina Arm are at roughly the same radius from the Galactic Center as the Sun. At a much larger distance of 6–8 kpc and a location on the far side of the Carina Arm, the expected proper motions should be less, so perhaps the improved precision of future Gaia data releases will provide a more definitive constraint from proper motion.

As a further check on the veracity of the Gaia distances for these stars, we used the sample of O stars in the Carina star-forming region⁴ from Smith (2006a), which are all expected to be at a common distance of ∼2.2 kpc (π ∼ 0.45 mas) on the basis of multiple lines of evidence, including the nebular expansion of η Car itself (see, e.g., Smith 2006b). Forty-three of the stars in that study are present in the Gaia DR1, and in Figure 1 we present the distribution of their parallaxes using both the direct Gaia DR1 parallaxes and the Milky Way prior-based parallaxes from ABJ. The former gives a mean distance of 2.06 ± 0.41 kpc and the latter gives 1.86 ± 0.05 kpc. The median Gaia parallax error for these stars is 63%, comparable to that for the four LBVs in our study sample.

Zoom In Zoom Out Reset image size

Thus, the Gaia distances for the Carina O stars via a simple 1/π estimator are consistent with expectation; the distances from the Milky Way prior-based method (ABJ) are perhaps slightly underestimated. In any event, this check provides a measure of validation that the Gaia distances for our target LBVs—also luminous stars at expected distances ∼2 kpc or greater—should be reliable (see also Casertano et al. 2016; Jao et al. 2016; Stassun & Torres 2016b; Davies et al. 2017). If so, the implications are important, as discussed next.

Figure 2 shows LBVs on the HR Diagram, adapted from previous studies as noted in the caption. The red points show the luminosities for AG Car, HR Car, HD 168607, and Hen 3-519 using previously adopted distances in the literature, whereas the blue points show how the luminosities are lowered if the nearer Gaia DR1 distances are adopted. For Hen 3-519, widely different luminosities are given for the same 8 kpc distance by Davidson et al. (1993) and Smith et al. (1994) based on different assumed values of $E(B-V)$. We adopt an average of the two here, although this difference is small compared to the factor of 16 reduction in luminosity that results from the nearer distance.

Zoom In Zoom Out Reset image size

As noted above, the nearer distance for HR Car is not in such severe disagreement with previous values, and for HD 168507, the distance agrees within the quoted uncertainty with previous estimates. It is interesting, though, that at the nearer 2.3 kpc distance, the hotter state of HR Car is almost coincident on the HR Diagram with the progenitor of SN 1987A. Speculative comparisons between SN 1987A's progenitor and LBVs have been discussed before (Smith 2007). Future Gaia data releases will provide higher precision in the parallax and proper motion, so we postpone a more detailed discussion of HR Car and HD 168607 until then. For now, we discuss the basic implications of the substantially smaller distances for AG Car and Hen 3-519, where the Gaia distances—even with the relatively large uncertainty of DR1—appear to be inconsistent with previous estimates, and where the possible implications for our understanding of LBVs are more severe. We acknowledge that the following discussion may need to be revisited if the improved Gaia DR2 measurements show the adopted DR1 values and uncertainties to be substantially incorrect.

4.1. AG Car as a Prototypical S Doradus Variable

4.2. Hen 3-519 at Low Luminosity

We report the parallax distances and proper motions of the Galactic LBV stars AG Car, HR Car, and HD 168607, and the LBV candidate Hen 3-519, resulting from the Gaia DR1 first data release. These are the only Galactic LBVs included in DR1. The distances to all four objects are closer than traditionally assumed in the literature, suggesting lower intrinsic luminosities.

1. For HD 168607, the reduction in distance is about a factor of 2, but the new distance is consistent with previous values within the uncertainty. We therefore postpone an evaluation of its properties until the higher precision that will be available later from Gaia.

2. For HR Car, the implied distance of ∼2.3 kpc is more than a factor of 2 lower than the traditional value of 5 ± 1 kpc, but the upper limit to the distance of 4.6 kpc is still marginally consistent with the old value. It is interesting that at this closer distance, HR Car's position on the HR diagram is nearly identical to the progenitor of SN 1987A.

3. The prototypical, classical LBV star AG Carinae has a Gaia parallax indicating a distance of only ∼2 kpc—three times closer than previously thought, making it nine times less luminous. The upper limit to its distance is inconsistent with the usually adopted value of 6.4–6.7 kpc (Humphreys et al. 1989). It now lies along a track of a 25 M_⊙ star, rather than a 90–100 M_⊙ star. If correct, this has dramatic implications for the interpretation of this star and its nebula. In particular, AG Car may have gone through a previous RSG phase, and it is likely to be a product of binary interaction. Moreover, since AG Car is regarded as a defining member of the S Doradus class and a classical high-luminosity LBV, its lower luminosity and mass have profound consequences for our traditional view of LBVs in general.

4. For Hen 3-519, the distance implied by the Gaia parallax is also about 2 kpc, which is ∼4 times closer than previously thought (making the star 16 times less luminous). This moves it far away from the traditional locus of LBVs on the HR diagram. Its origin and evolutionary state remain unclear, but its nebula may be the product of previous RSG mass loss or binary interaction.

Given that these stars were thought to be among the most luminous stars in the Milky Way and have shaped our views of LBVs for the last 30 years, it seems likely that Gaia distances for the remaining LBVs may instigate a re-evaluation of our standard view of LBVs. More precise values of the parallax for these four objects and for a larger number of Galactic LBVs will be available soon; the results reported here may hint at a coming upheaval in our understanding of massive-star evolution, even if they are regarded as preliminary.

An obvious avenue for further investigation is whether there are similar low-luminosity stars in the LMC or other nearby galaxies that have LBV-like spectra or variability, but that have evaded detection because they are faint. On the other hand, this does not undermine the properties of existing LBVs known in the LMC and SMC, since their distances are not so uncertain; the S Dor strip still seems to work for them. It will be interesting to investigate why there are stars that appear to be classical S Doradus-like LBVs in the Milky Way that are so far below the traditional S Dor instability strip. Perhaps metallicity plays a key role, or perhaps samples in the LMC are incomplete at lower luminosity.

We thank Megan Kiminki for double-checking the Gaia distances for O-type stars in the Carina Nebula and for helpful discussions. Support for N.S. was provided by the National Science Foundation (NSF) through grants AST-1210599 and AST-1312221 to the University of Arizona. K.G.S. acknowledges partial support from NSF PAARE grant AST-1358862. This work has made use of data from the European Space Agency (ESA) 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.