A Catalog of GALEX Ultraviolet Emission from Asymptotic Giant Branch Stars

Our sample of AGB stars is derived from numerous AGB samples found in the literature and was originally compiled by Ramstedt et al. (2012) to search for X-ray detections associated with AGB stars. As in Ramstedt et al. (2012), there are a total of 468 unique AGB stars in our sample: 286 M-type Miras from Little-Marenin & Little (1990); 171 AGB stars from the samples of Schöier & Olofsson (2001), González Delgado et al. (2003), and Ramstedt et al. (2006); plus 11 mixed-type stars from the sample of Sahai et al. (2008). The C-type (C/O > 1) star sample is nearly complete out to 500 pc. The S-type (C/O ∼ 1) star sample is nearly complete out to 600 pc. The completeness of the M-type (C/O < 1) sample is not well investigated.

The GALEX mission performed a two-band survey of the UV sky. Using a dichromatic beam splitter, GALEX simultaneously observed far-UV (FUV; ${\lambda }_{\mathrm{eff}}\sim 1528$ Å; 1344–1786 Å) and near-UV (NUV; ${\lambda }_{\mathrm{eff}}\sim 2310$ Å; 1771–2831 Å) in surveys with different depths. The All-Sky Imaging Survey (AIS) had a typical exposure of ∼150 s, and the Medium Imaging Survey (MIS) had a typical exposure of ∼1500 s. The FUV and NUV detectors were photon counting microchannel plates with ∼12 fields of view with images virtually binned to 15 square pixels. The spatial resolution is 43 in FUV and 53 in NUV. Limiting magnitudes for the AIS are ∼19.9 and ∼20.8 mag in the FUV and NUV, respectively, for the typical AIS exposure time. For MIS, the limiting magnitudes for typical MIS exposure times are ∼22.7 mag in the NUV and FUV (Morrissey et al. 2007). GALEX could also perform slitless grism spectroscopy to disperse the FUV and NUV emission. As described in further detail in Morrissey et al. (2007), spectroscopic observations place a grism into the converging beam of the telescope to simultaneously disperse all sources onto the detector plane. According to Morrissey et al. (2007), the usable ranges of the grism spectra are 1300–1820 Å and 1820–3000 Å in the FUV and NUV, with average resolutions of 8 Å and 20 Å, respectively. In 2009 May, the FUV detector ceased functioning, but the NUV detector continued functioning well beyond the NASA-led phase, which ended in 2011. In all, the GALEX mission made nearly 300 million UV measurements, all of which are available via the MAST data archive.

To supplement our study of the GALEX observations of AGB stars, we collected photometric data from across the electromagnetic spectrum for all the AGB stars considered using SIMBAD and VizieR tools. Since many of our stars are bright and exhibit long-period variations, they are often targets of the American Association of Variable Star Observers (AAVSO). We collected AAVSO light curves that span the GALEX mission lifetime (2003 May 28 to 2013 June 28) from the AAVSO International Database.⁶ Additionally, when available, we have gathered Hipparcos parallax measurements with signal-to-noise ratios above 1.5 for our entire sample (van Leeuwen 2007).⁷ To estimate the selective extinction for all bandpasses considered (see Section 4.1), we used the ATLAS9 stellar atmosphere models of Castelli & Kurucz (2004).

The characteristics of detections and nondetections among the GALEX-AGB sample can provide insight into the nature of GALEX-detected UV emission from AGB stars. In Figure 3 we compare the optical and NIR photometric properties of detected and undetected AGB stars. In each panel we display the cumulative distribution of apparent brightness for several optical and infrared photometric bands of the detected and undetected AGB stars.⁸ The distributions in Figure 3 suggest that GALEX-detected AGB stars are approximately 2 mag brighter, on average, than those that are not detected. Given the relatively narrow range of temperatures and bolometric luminosities of AGB stars (Vassiliadis & Wood 1993), the apparent magnitudes of our AGB stars are a first-order indication of their relative distances. Hence, the patterns seen in Figure 3 further suggest that the AGB stars undetected in the UV by GALEX are more distant, on average, than UV-detected stars. This notion is further supported by the fact that ∼44% of the detected AGB stars have Hipparcos parallax measurements (van Leeuwen 2007) with signal-to-noise ratios >1.5, while only ∼5% of the undetected AGB stars have similarly significant parallax measurements. For the observed J-band apparent magnitudes, which are the least affected by interstellar medium (ISM) extinction of the bands considered, there is clear evidence that the fluxes are proportional to the distance to the AGB stars (Figure 4) and the brighter stars are more easily detected in the UV. However, ISM extinction increases toward shorter wavelengths and will influence the overall UV detectability.

Zoom In Zoom Out Reset image size

Next, we consider the distribution of the AGB stars (detections and nondetections) in galactic coordinates. In Figure 5, we present the distributions in galactic latitude, b, and longitude, l. In galactic latitude, the AGB stars appear to be normally distributed with $\bar{b}\sim 0^\circ \pm 10^\circ$ and FWHM ∼ 50°. The nondetections follow a similar distribution but with a narrower FWHM (∼25°) compared to the entire sample. In contrast, Figure 5 shows that galactic latitudes of the detected AGB stars display a bimodal distribution with a dip in the number of detected stars near b ∼ 0. Although GALEX initially avoided low galactic latitudes, more of the galactic plane was observed toward the end of the extended mission (Bianchi 2014). Galactic scale heights based on the Hipparcos-derived distance of detected and undetected AGB stars (see Figure 4) suggest that the bimodal distribution of AGB star detections with galactic latitude is not due to the poor coverage of the galactic plane. Specifically, we find that AGB stars with high galactic latitudes, $| b| \geqslant 45^\circ$, are more readily detected at larger distances than those AGB stars with lower galactic latitudes, $| b| \leqslant 45^\circ$. Such behavior suggests that high galactic latitudes are more favorable sightlines for detections. This, in turn, suggests that ISM extinction, which increases more rapidly with distance for sightlines in the galactic plane, is responsible for the decline in AGB star detection fraction with galactic latitude. Indeed, that such a trend is due to dust is consistent with studies of larger unbiased samples of GALEX sources such as Bianchi et al. (2011).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To mitigate the influence of ISM extinction in subsequent analysis, we use the extinction estimates provided by the GALEX source catalog and listed in Table 1 to deredden the observed magnitudes. A caveat of such an approach is that the GALEX extinction estimates are based on galactic dust maps (Schlegel et al. 1998) intended to give total Galactic extinction for a given line of sight through the Milky Way; hence, these extinction estimates are likely to overestimate the extinction for the closest AGB stars that are near the galactic plane. Based on the galactic distribution of detections and nondetections, such a reddening correction will be problematic for a majority of undetected sources and <6% of the detected sources. An additional caveat is that the circumstellar material that surrounds an AGB star is still a factor depending on the site of UV emission. To estimate the selective extinction in a given bandpass, ${A}_{\mathrm{BP}}/{E}_{{\rm{B}}-{\rm{V}}}$, we followed the procedure described in a GALEX study of hot stars (Bianchi 2011) for Milky-Way-type dust with R_V = 3.1. However, given the nature of our sample, we used cooler stellar atmosphere models (${T}_{\mathrm{eff}}\lt 5000$ K) when estimating bandpass-selective extinctions. For ${A}_{\mathrm{FUV}}/{E}_{{\rm{B}}-{\rm{V}}}$ and ${A}_{\mathrm{NUV}}/{E}_{{\rm{B}}-{\rm{V}}}$, we determined factors of 7.81 and 6.30, respectively, suggesting that the NUV–FUV color is not independent of extinction for cool stars. The selective extinctions for the optical and NIR bandpasses are listed in Table 3.

Table 3.  Correlations with Other Bandpasses

Band	$\tfrac{{A}_{\mathrm{BP}}}{E(B-V)}$	${N}_{\mathrm{NUV}}$	${r}_{\mathrm{NUV}}$ a	${\rho }_{\mathrm{NUV}}$ b	${N}_{\mathrm{FUV}}$	${r}_{\mathrm{FUV}}$ a	${\rho }_{\mathrm{FUV}}$ b
U	4.85	38	0.70	0%	12	0.24	46%
B	3.92	177	0.78	0%	37	0.18	28%
V	3.03	162	0.63	0%	38	0.16	33%
R	2.26	48	0.52	0%	13	0.40	18%
I	1.58	24	0.55	0%	9	0.17	65%
J	0.89	178	0.48	0%	37	0.25	14%
H	0.56	178	0.40	0%	37	0.28	10%
K	0.36	178	0.35	0%	37	0.28	9%

Notes.

^aPearson correlation coefficient for NUV or FUV sample versus given bandpass. In samples with more than 100 values, a Pearson correlation coefficient, r, above 0.2 is significant evidence of correlation. In samples with fewer than 100 values, the significance of r decreases rapidly. For example, for sample sizes of 20 and 50, r values of 0.3 and 0.45 are not significantly distinct from 0 (no correlation) and ρ becomes unreliable. ^bProbability that an uncorrelated sample can produce the given distribution. Note: ρ becomes unreliable for small sample sizes.

Download table as:  ASCIITypeset image

Next, we compared the detection fraction for the three chemical subtypes (M-, S-, and C-types) indicative of the C/O ratio in their stellar atmospheres and the dust and molecular chemistry in their circumstellar envelopes.⁹ Overall, the observed sample is skewed toward M-types (249), followed by C-types (47) and then S-types (21). The detection fractions for our sample of AGB stars are ∼60% ± 5% for M-types, ∼70% ± 20% for S-types, and ∼34% ± 9% for C-types. The disparity in M-type versus C-type detection fractions suggests that their different circumstellar environments might influence the UV absorption and, hence, detectability (Mathis & Cardelli 1992; Nagao et al. 2016). The lower detection rate among C-type AGB stars could be due to their carbonaceous dust, which has higher opacity for photons in the GALEX bandpasses (e.g., Suh 2000).

The relationship between the GALEX-detected UV emission and optical and near-infrared emission can also help us understand the nature of the UV emission. In Figures 6–7 we compare the dereddened optical and near-infrared fluxes to the dereddened NUV and FUV fluxes. There are apparent correlations between the optical/NIR fluxes and the NUV flux, while the FUV flux appears uncorrelated with any of the bandpasses. In Table 3 we have compiled tests for linear correlation (r, the Pearson correlation coefficient¹⁰ ) for the UBVRIJHK broadband photometric bandpasses. In all bands considered, we find evidence for correlation with the NUV fluxes and no strong correlation with FUV fluxes.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Since there is no tight correlation (i.e., no values of $| r| \sim 1$) in the samples shown in Figure 6, we considered how the known long-period variability of AGB stars might influence any correlation between NUV and optical/NIR emission. In Figure 8 we display AAVSO visible light curves (see footnote 6) of AGB stars with at least three separate measurements in the NUV. In each case, ∼10 yr of AAVSO measurements acquired during the GALEX mission lifetime were phased to their appropriate periods. We phased the dereddened GALEX NUV measurements to the same phase used for the AAVSO light curves. Finally, the NUV light curves of a given AGB star were scaled to best match the visible light curve or mean value of the visible light curve. As shown in the light curves, the visible light magnitudes for this selection of our sample can vary by up to 8 mag.

We note two important properties of the sample of visible and NUV light curves displayed in Figure 8. First, the arbitrary scaling used to shift NUV light curves to the visible light curves ranges from 6 to 12 mag. Such a large range suggests that multiple scaling relationships exist or that additional influences, such as extinction and possible binary companions, play a role in the level of NUV flux. Second, although very few of the examples in Figure 8 have adequate phase coverage to absolutely determine the variable nature of the NUV flux, we note that most of the NUV measurements are consistent with the general properties of the visible light curves. R Cet has the largest number of observations and largest phase coverage, and its NUV light curve clearly mimics the visible light behavior. On the other hand, the UV measurements of Y Aqr, which has the next largest phase coverage, appear anticorrelated with the visible light curve.

Given the range of V-band variability and suspected correlation between GALEX UV emission and the V band (see Figure 6), we attempted to quantify the scatter introduced by noncontemporaneous UV and V-band observations with Monte Carlo simulations of the observations. First, we created synthetic V magnitudes drawn from the distribution of V-band magnitudes given in Figure 6. Next, we used the visual light curves in Figure 8 to estimate the mean amplitude variation of the visible magnitudes (ΔVis. ∼ 4.6 mag) and their standard deviation (1.6 mag). Assuming similar variability in the NUV (fully correlated signals), we determine the synthetic NUV magnitudes using a range of power-law scaling relationships characterized by power-law index, α, plus a random offset based on the V-band mean amplitude variation. Given the well-behaved sinusoidal behavior of the variation seen in Figure 8, we model the variation as a simple uniform random variable in the range of 4.6 ± 1.6 mag. With these assumptions, we are able to generate a synthetic sample of NUV and V-band observations that have a 2D distribution similar to the observed sample seen in Figure 6. We find that a linear scaling relationship (α = 1) with the given amplitude variation reproduces the sample scatter. Scaling relationships with α < 1 can reproduce the scatter, but only if the amplitude variation is increased. However, for α < 0.5, increasing the amplitude variation fails to reproduce the observed scatter. Overall, these considerations suggest that the 2D distribution of dereddened NUV and V-band magnitudes is consistent with correlated fluxes observed noncontemporaneously. A more precise determination of the scaling relationship between NUV and V magnitudes requires contemporaneous multiwavelength observing campaigns.

Ten AGB stars in our sample were observed with the slitless GALEX grism (see Table 4 for information on grism observations). All objects with spectroscopic observations were detected in the NUV grism bandpass except for IRC +10216; only Mira is detected in both NUV and FUV grism bandpasses. Collectively, the observed spectra are fairly homogeneous, displaying some continuum emission with emission lines such as Mg ii (∼2800 Å) and C ii] (∼2325 Å). For some sources the continuum may not be detected throughout the NUV grism bandpass, leading to a "flat" appearance toward longer wavelengths. We note that the emission lines are multiplets but the grism spectroscopy has insufficient resolution to resolve the individual lines. In Figure 9 we display each grism spectrum along with the location of these two common UV emission lines. Only two stars, R UMa and VY UMa, do not display the Mg ii emission lines, while R UMa also displays a distinct broad feature in the vicinity of the Mg ii location.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 4.  AGB Stars with GALEX Grism Spectroscopy

Object	Obs. Date	${t}_{\mathrm{NUV}}$ (s)	${t}_{\mathrm{FUV}}$ (s)	Detected
EP Aqr	2007 Feb 07	2759	2759	NUV only
RW Boo	2006 Nov 01	5113	1721	NUV only
AA Cam	2005 Jan 30	3224	1693	NUV only
Mira (omi Cet)	2006 Nov 18	11325	11325	NUV and FUV
V Eri	2005 Nov 03	1704	1616	NUV only
TW Hor	2006 May 07	1088	1007	NUV only
V Hya	2005 Aug 17	2696	2696	NUV only
IRC +10216 (CW Leo)	2008 Jan 23	8761	8760	Nondetection
R UMa	2006 Jan 06	1703	1703	NUV only
VY UMa	2006 Jan 07	1704	1704	NUV only

Download table as:  ASCIITypeset image

5.1.1. Scattering of the Interstellar Radiation Field

5.1.2. Binary Companions

By analyzing a nearly complete, volume-limited sample of AGB stars and thereby avoiding the biases inherent in previous studies that employed smaller samples, we find that the brightest (closest and most well studied) AGB stars have higher detection fractions in the NUV and FUV than more distant AGB stars (Section 4.1). This suggests that the GALEX-detected UV emission from AGB stars is most likely inherent to the AGB star and not due to extrinsic processes. Indeed, IUE observations reveal the prevalence of chromospheric-like UV emission from cool giants and supergiants (Luttermoser 2000). The photospheric models adopted by Sahai et al. (2008) and Ortiz & Guerrero (2016) do not include radiation from such overlying chromospheric emission, and it would appear likely that such a process is responsible for AGB UV emission in many, if not most, cases.

The faint UV fluxes, apparent scaling relationships with other bandpasses, and phase correlations between visual and UV magnitudes suggest that the observed UV emission could arise from the Wein tail of the photospheric spectral energy distribution. However, we note that the behavior of the photosphere in these regimes is poorly understood and the gas and dust of the circumstellar envelope are strong sources of UV absorption. Hence, although photospheric radiation may contribute to the GALEX-detected UV emission, it is unlikely to account for all of the observed UV flux.

NUV to FUV spectroscopy from IUE and HST of AGB stars reveals that emission lines from Mg ii, C ii], and Fe ii are prevalent. Such emission lines are indicative of a chromospheric layer in the atmospheres of these stars that persists into these highly evolved stages, perhaps never dropping below a basal level (Schrijver 1995). The chromospheres appear to be much cooler than solar-like chromospheres. The heating mechanism is uncertain, but could be tied to pulsations, wave heating, or magnetic fields (see review by Schrijver 1995, and references therein). The apparent prevalence of UV emission from AGB stars (Figure 3), the apparent correlation with stellar fluxes (Figure 6), and the spectral characteristics of past IUE and recent GALEX grism spectroscopic observations (Figure 9) suggest that chromospheres are significant contributors to the GALEX-detected UV emission from AGB stars.

Since the photospheres and chromospheres reside within the circumstellar envelope, the properties of the UV emission should be influenced by absorption by the circumstellar material. To explore this notion, we compiled mass-loss rates, $\dot{M}$, and gas kinematics, ${v}_{\exp }$, for the Hipparcos sample of AGB stars (Groenewegen et al. 1999; Schöier & Olofsson 2001; Olofsson et al. 2002; González Delgado et al. 2003; Ramstedt et al. 2009; Danilovich et al. 2015) to compute the density proxy, $\dot{M}/{v}_{\exp }$. In Figure 10 we compare the dereddened and distance-corrected NUV and FUV emission with this mass-loss proxy for density. Despite the scatter, this comparison provides tentative evidence that the circumstellar density and the dereddened distance-corrected NUV and FUV fluxes are anticorrelated. Performing a similar comparison for other broadband measurements, we find that the anticorrelation increases for shorter wavelengths. Such behavior is expected if the circumstellar material resides between us and the source of UV emission and attenuates the flux. The evidence for multiple scaling laws suggested by the NUV correlations with optical and NIR emission could then be explained by varying degrees of absorption in the time-varying circumstellar shell.

Zoom In Zoom Out Reset image size