DISCOVERY OF FIVE CANDIDATE ANALOGS FOR η CARINAE IN NEARBY GALAXIES

The last stages of the evolution of the most massive (M ≳ 30 M_⊙) stars may be dominated by episodic large mass ejections (e.g., Humphreys & Davidson 1984; Smith 2014). This leads to dust condensing out of the ejecta, obscuring the star in the optical but revealing it in the mid-infrared (mid-IR) as the absorbed UV and optical photons are re-emitted at longer wavelengths (e.g., Kochanek et al. 2012). The best known example is η Carinae (η Car), which contains one of the most massive (100–150 M_⊙) and most luminous (∼5 × 10⁶ L_⊙) stars in our Galaxy (e.g., Robinson et al. 1973). Its Great Eruption in the mid-1800s led to the ejection of ∼10 M_⊙ of material (Smith et al. 2003) now seen as a dusty nebula around the star. While ongoing studies are helping us further analyze the Great Eruption (see, e.g., Rest et al. 2012; Prieto et al. 2014), deciphering the rate of such events and their consequences is challenging because no analog of this extraordinary laboratory for stellar astrophysics (in terms of stellar mass, luminosity, ejecta mass, time since mass ejection, etc.) has previously been found.

A related puzzle is the existence of the Type II superluminous supernovae (SLSNe-II) that are plausibly explained by the SN ejecta colliding with a massive shell of previously ejected material (e.g., SN 2006gy; Smith et al. 2007). A number of SNe, such as the Type Ib SN 2006jc (Pastorello et al. 2007) and the Type IIn SN 2009ip (e.g., Mauerhan et al. 2013; Pastorello et al. 2013; Prieto et al. 2013), have also shown transients that could be associated with mass ejections shortly prior to the final explosion. However, the relationship between these transients and η Car or other LBVs surrounded by still older, massive dusty shells (e.g., Smith & Owocki 2006) is unclear.

There are presently no clear prescriptions for how to include events like the Great Eruption into theoretical models. Even basic assumptions—such as whether the mass loss is triggered by the final post-carbon ignition phase as suggested statistically by Kochanek et al. (2012) or by an opacity phase-transition in the photosphere (e.g., Vink et al. 1999) or by interactions with a binary companion (e.g., Soker 2005)—are uncertain. Studies of possible mass-loss mechanisms (e.g., Shiode & Quataert 2014) are unfortunately non-prescriptive on either rate or outcome. Observationally, we are limited by the small numbers of high-mass stars in this short evolutionary phase and searching for them in the Galaxy is complicated by having to look through the crowded, dusty disk and distance uncertainties. Obtaining a better understanding of this phase of evolution requires exploring other galaxies.

We demonstrated in Khan et al. (2010, 2011, 2013) that searching for extragalactic self-obscured stars utilizing Spitzer images is feasible, and in Khan et al. (2015a) we isolated an emerging class of 18 candidate self-obscured stars with L_bol ∼ 10^5.5–6.0L_⊙ (M_ZAMS ≃ 25–60M_⊙) in galaxies at ∼1–4 Mpc. We have now expanded our search to the large star-forming galaxies M51, M83, M101, and NGC 6946 (distance ≃ 4–8 Mpc). We picked these galaxies because they have high star formation rates (total SFR_Hα ≃ 6.9M_⊙ yr⁻¹; mainly based on Kennicutt et al. 2008) and hosted significant numbers of core-collapse supernovae (ccSNe) over the past century (total of 20; e.g., Botticella et al. 2012), indicating that they are likely to host a significant number of evolved high-mass stars.

In this Letter, we announce the discovery of five objects in these galaxies that have optical through mid-IR photometric properties consistent with the hitherto unique η Car as it is presently observed. In what follows, we describe our search method (Section 2), analyze the physical properties of the five potential η Car analogs (Section 3), and consider the implications of our findings (Section 4).

At extragalactic distances, an η Car analog would appear as a bright, red point source in Spitzer IRAC (Fazio et al. 2004) images, with a fainter optical counterpart due to self-obscuration. Given enough absorption, the optical counterpart could be undetectable. Building on our previous work (Khan et al. 2011, 2013, 2015a, 2015b), we relied on these properties to identify the η Car analog candidates. For M51 (D ≃ 8 Mpc; Ferrarese et al. 2000), M83 (D ≃ 4.61 Mpc; Saha et al. 2006), and M101 (D ≃ 6.43 Mpc; Shappee & Stanek 2011), we used the full Spitzer mosaics available from the Local Volume Legacy Survey (Dale et al. 2009), and for NGC 6946 (D ≃ 5.7 Mpc; Sahu et al. 2006), we used those from the Spitzer Infrared Nearby Galaxies Survey (Kennicutt et al. 2003).

We built Vega-calibrated IRAC 3.6–8 μm and MIPS (Rieke et al. 2004) 24 μm point-source catalogs for each galaxy following the procedures described in Khan et al. (2015b). We use point-spread function (PSF) photometry at 3.6 and 4.5 μm, a combination of PSF and aperture photometry (preferring PSF) at 5.8 μm, and only aperture photometry at 8.0 and 24 μm as the PSF size and PAH emission both increase toward longer wavelengths. For all sources, we determine the spectral energy distribution (SED) slope a (λL_λ ∝ λ^a), the total IRAC luminosity (L_mIR), and the fraction f of L_mIR that is emitted in the first three IRAC bands. Following the selection criteria established in Khan et al. (2013)—L_mIR > 10⁵ L_⊙, a > 0, and f > 0.3—we initially selected ∼700 sources from our mid-IR point-source catalogs.

We examined the IRAC images to exclude the sources associated with saturated, resolved, or foreground objects and utilized the VizieR⁶ web service to rule out spectroscopically confirmed non-stellar sources and those with high proper motions. We inspected the 3.6–24 μm SEDs of the remaining sources to identify the ones that most closely resemble the SED of η Car and then queried the Hubble Source Catalog (HSC,⁷ Version 1) to exclude those with bright optical counterparts (m ≲ 20 mag, implying L_opt ≳ 1.5–6 × 10⁵ L_⊙). These steps produced a short-list of ∼20 sources for which we retrieved archival HST images and the associated photometry from the Hubble Legacy Archive (HLA⁸ ). Since the HST and Spitzer images sometimes have significant (∼10) astrometric mismatches, we utilized the IRAF GEOMAP and GEOXYTRAN tasks to locally align the HST and Spitzer images with uncertainties ≲01. We then searched for the closest optical counterpart within a matching radius of 03.

We identified five sources with mid-IR SEDs closely resembling that of η Car and optical fluxes or flux limits ∼1.5–2 dex fainter than their mid-IR peaks. We will refer to these sources as η Twins-1, -2, -3, -4, and -5. We find one source each in M51 (η Twin-1), M101 (η Twin-2), and NGC 6946 (η Twin-3), and two sources in M83 (η Twins-4, -5). We identified HST counterparts of η Twins-1, -2, and -4 within the 03 matching radius. For η Twin-3, no HST source is cataloged within the matching radius, so we visually identified the closest location of flux excess at ∼03 and used simple aperture photometry techniques to measure the I-band flux and the B- and V-band flux upper limits. For η Twin-5, although a cataloged HST source exists within the 03 matching radius, we selected a different source at 035 as the more likely photometric match because it is also a bright HST J-band source. Table 1 lists the photometry of these sources, Figure 1 shows their IRAC 3.6 μm and HST I-band images, and Figure 2 shows their SEDs. η Twins-1, -4, and -5 are Hα emitters, and η Twin-2 is a UV source (see Table 1).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We have UBVR variability data for M51, M101, and NGC 6946 from the LBC survey for failed supernovae (Gerke et al. 2015). We analyzed 21/26/37 epochs of data spanning a 7.1/7.2/8 year period for M51/M101/NGC 6946 with the ISIS image subtraction package (Alard & Lupton 1998). We did not detect any significant optical variability at the locations of η Twins-1, -2, or -3.

Cutri et al. (2012) identified η Twin-2 as a WISE point source, and we use their 12 μm flux measurement as an upper limit for SED models (Section 3). Johnson et al. (2001) report an optically thick free–free radio source located 049 from η Twin-3, and Hadfield et al. (2005) identified a source with Wolf–Rayet spectroscopic signature 154 from η Twin-4. We could not confirm if these sources are reasonable astrometric matches to the IRAC locations. η Twins-4 and -5 were cataloged by Williams et al. (2015) but were not flagged as massive stars.

We fit the SEDs of these five sources using DUSTY (Ivezic & Elitzur 1997) to model radiation transfer through a spherical medium surrounding a blackbody source, which is also a good approximation for a combination of unresolved non-spherical/patchy/multiple circumstellar shells. We considered models with either graphitic or silicate dust (Draine & Lee 1984). The models are defined by the stellar luminosity (L_*), stellar temperature (T_*), V-band optical depth (τ_V), dust temperature at the inner edge of the shell (T_d), and shell thickness ζ = R_out/R_in. We embedded DUSTY inside a Markov Chain Monte Carlo driver to fit each SED by varying T_*, τ_V, and T_d with L_* determined by a χ² fit for each model. We fix ζ = 4 since its exact value has little effect on the results (Khan et al. 2015a), limit T_* to a maximum value of ∼50,000 K, set the minimum flux uncertainty to ∼10% (0.1 mag), and do not account for distance uncertainties.

The best-fit model parameters determine the radius of the inner edge of the stellar-ejecta distribution (R_in). The mass of the shell is ${M}_{e}=4\pi {R}_{\mathrm{in}}^{2}{\tau }_{V}/{\kappa }_{V}$ (scaled to a visual opacity of κ_V = 100 κ_v100 cm² g⁻¹), and the age estimate for the shell is t_e = R_in/v_e (scaled as v_e = 100 v_e100 km s⁻¹), where we can ignore R_out to the zeroth order. Table 2 reports the parameters of the best-fit models, and Figure 2 shows these models. The integrated luminosity estimates depend little on the choice of dust type and are in the range of L_* ≃ 10^6.5–6.9L_⊙. We also fit the SEDs using Castelli & Kurucz (2004) stellar atmosphere models instead of blackbodies. Since these resulted in similar parameter estimates, we only report the blackbody results.

Table 2.  Best-fit SED Models

ID			χ2/(m + n)	τV	Td	T*	$\mathrm{log}\left({R}_{\mathrm{in}}\right)$	log Lbol	Me	te
					(K)	(K)	(cm)	(L⊙)	(M⊙)	(years)
Graphitic
η Twin-1			151/(11 + 0)	4.34	386	27,610	17.53	6.842	32.04	1087
η Twin-2			92/(10 + 1)	2.51	404	50,120	17.37	6.554	8.81	750
η Twin-3			87/(6 + 2)	16.67	337	5570	17.44	6.920	78.76	870
η Twin-4			113/(8 + 0)	2.24	394	50,110	17.39	6.544	8.59	783
η Twin-5			269/(9 + 0)	3.61	375	23,530	17.37	6.504	12.62	748
Silicate
η Twin-1			130/(11 + 0)	6.34	603	37,740	16.97	6.890	3.53	299
η Twin-2			205/(10 + 1)	4.22	641	50,120	16.79	6.573	0.99	194
η Twin-3			69/(6 + 2)	34.30	424	4730	16.71	6.839	5.78	164
η Twin-4			136/(8 + 0)	3.94	622	50,120	16.80	6.555	0.97	199
η Twin-5			307/(9 + 0)	4.57	760	37,520	16.59	6.449	0.44	124

Note. The format χ²/(m + n) indicates the goodness of fit χ², the number of flux measurements m used to determine the luminosity, and the number of upper limits n added to the estimate of χ² once the luminosity is determined.

Download table as:  ASCIITypeset image

Generally, the best fits derived for graphitic dust require lower optical depths, lower dust temperatures, and larger shell radii, leading to higher ejecta masses and age estimates. For η Twins-2 and -4, the stellar temperature estimates reach the allowed maximum of ∼50,000 K. The best-fit models of η Twin-1 and -5 also require the presence of a hot star, but with temperatures lower than the allowed maximum (∼27,600/37,750 K and ∼23,500/37,500 K for graphitic/silicate dust). Constrained by the low optical flux, the best-fit models of η Twin-3 require the presence of a cool star (∼5000 K). For η Twins-2, -4, and -5, the best fits derived for graphitic dust had lower χ², and for for η Twins-1 and -3, the best fits derived for silicate dust have lower χ². Considering these models, the η Car analog candidates appear to be embedded in ∼5–10 M_⊙ of warm (∼400–600 K) obscuring material ejected a few centuries ago.

Figure 3 contrasts the bolometric luminosities and ejecta mass estimates of these five objects with the relatively less luminous sources we identified in Khan et al. (2015a). The five new sources form a distinct cluster close to η Car in the ${L}_{\mathrm{bol}}-{M}_{\mathrm{ejecta}}$ parameter space, whereas the previously identified dusty-star candidates from Khan et al. (2015a) are more similar to the Galactic OH/IR star IRC+10420 (e.g., Tiffany et al. 2010) or M33's Variable A (e.g., Humphreys et al. 1987).

Zoom In Zoom Out Reset image size

To an extragalactic observer located in one of the targeted galaxies surveying the Milky Way with telescopes similar to the HST and Spitzer, η Car's present-day SED would appear nearly identical to the extragalactic η Car analog candidates we found. The Carina nebula is ∼25 in extent (Smith & Brooks 2007), corresponding to ∼25 at our most distant galaxy (M51 at 8 Mpc). While this would not be resolved by Spitzer, it would be easily resolved by HST. Because more compact clusters are not uncommon, in Khan et al. (2013), we considered whether dusty clusters can hide η Car like stars and if we would confuse unresolved star clusters with η Car analogs. In general, a cluster sufficiently luminous to hide an evolved ≳30 M_⊙ star has hosted many luminous stars with strong UV radiation fields and winds, which will generally clear the cluster of the gas and dust needed to produce strong mid-IR emission over the timescale that even the most massive star needs to evolve away from the main sequence. Moreover, emission from warm circumstellar ejecta peaks between the IRAC 8 μm and MIPS 24 μm bands and then turns over, as seen in all of our candidates, unlike emission from colder intra-cluster dust that generally peaks at longer wavelengths.

A significant majority of massive stars are expected to be in multiple-star systems (e.g., Sana & Evans 2011), as is the case for η Car (e.g., Damineli 1996; Mehner et al. 2010). This is a minor complication, affecting luminosity estimates by at most a factor of 2, and mass estimates even less. Assuming all the candidates we have identified are real analogs of η Car, then our galaxy sample (including the Milky Way) contains N_c = 6 η Car–like stars. Based on the ratio of star formation rates (2 versus 10 M_⊙ yr⁻¹), our original sample of seven galaxies would be expected to have ∼1 η Car analog, which is statistically consistent with not finding one in Khan et al. (2015a).

If we expand our simple rate estimates from Khan et al. (2015a), our N_c = 6 sources implies an eruption rate over the 12 galaxies (7 previous, 4 in this work, and the Milky Way) of ${F}_{e}=0.033{t}_{d200}^{-1}$ yr⁻¹ (0.016 yr${}^{-1}\lt {F}_{e}{t}_{d200}\lt 0.059$ yr⁻¹ at 90% confidence), where t_d ≃ 200 years is a rough estimate of the period over which our method would detect an η Car–like source. For comparison, the number of ccSNe recorded in these galaxies over the past 30 years is 10 (mainly based on Botticella et al. 2012) for an SN rate of F_SN = 0.33 yr⁻¹. This implies that the rate of η Car–like events is a fraction f = 0.094 (0.040 < f < 0.21 at 90% confidence) of the ccSN rate.

If there is only one eruption mechanism and the SLSNe-II are due to ccSNe occurring inside these dense shells, then the ratio of the rate of η Car–like events and SLSNe-II, r_SLSN/r_η = t_SLSN/t_η, is the ratio of the time period t_SLSN during which the shell is close enough to the star to cause an SLSN to the time period t_η over which shell ejections occur. With r_SLSN ∼ 10⁻³ of the core-collapse rate (Quimby et al. 2013), we must have that t_SLSN/t_η ∼ 10⁻². A typical estimate is that t_SLSN ∼ 10–10² years, which implies t_η ∼ 10³–10⁴ years, consistent with the properties of the massive shells around luminous stars observed in our own Galaxy and suggesting that the instabilities driving the eruptions are linked to the onset of carbon burning (Kochanek 2011). This would also imply the existence of "superluminous" X-ray ccSNe, where an older shell of material is too distant and low density to thermalize the shock heated material but is still dense enough for the cooling time to be faster than the expansion time. Such events should be ∼10 times more common than optical SLSNe-II. If the eruptions driving SLSNe-II are only associated with later and shorter burning phases (e.g., as in Shiode & Quataert 2014), then there must be two eruption mechanisms, and the vast majority of η Car analogs will not be associated with the SLSN-II mechanism.

We identified the five potential η Car analogs by specifically focusing on finding sources that most closely resemble the SED of present-day η Car. The reason that the SEDs of these five sources are so remarkably similar to each other is by design. We have not closely studied the less luminous mid-IR sources that may belong to the class of candidate self-obscured stars we identified in Khan et al. (2015a), and some of the sources that we excluded because they have relatively bright optical counterparts may be evolved high-mass stars with older, lower optical-depth shells. It is readily apparent that a closer scrutiny of our mid-IR catalogs should reveal richer and more diverse populations of evolved massive stars. This in turn will let us better quantify the abundance of those stars and constrain the rates of mass ejection episodes and mass loss from massive stars prior to their death by core collapse.

The η Car analog candidates we identified can be studied at greater detail with the James Webb Space Telescope (JWST; e.g., Gardner et al. 2006), taking advantage of its order-of-magnitude-higher spatial resolution. These sources are luminous in the 3.6–24 μm wavelength range where the JWST will be most sensitive. They are rare laboratories for stellar astrophysics and will be very interesting extragalactic stellar targets for spectroscopic study with JWST's mid-IR instrument (Rieke et al. 2015). This will give us an unprecedented view of these most massive self-obscured stars, letting us study their evolutionary state and the composition of their circumstellar ejecta.

We thank the referee for helpful comments. R.K. is supported by a JWST Fellowship awarded through the NASA Postdoctoral Program. S.M.A. is supported by a Presidential Fellowship at The Ohio State University. K.Z.S. is supported in part by NSF grant AST-151592. C.S.K. is supported by NSF grant AST-1515876. This research has made use of observations made with the Spitzer Space Telescope, which is operated by the JPL and Caltech under a contract with NASA; observations made with the NASA/ESA Hubble Space Telescope and obtained from the Hubble Legacy Archive, which is a collaboration between the STScI/NASA, ST-ECF/ESA, and the CADC/NRC/CSA; and the VizieR catalog access tool, CDS, Strasbourg, France.