The Astrophysical Journal, 860:105 (8pp), 2018 June 20 https://doi.org/10.3847/1538-4357/aac3d7
© 2018. The American Astronomical Society. All rights reserved.
Binarity and Accretion in AGB Stars: HST/STIS Observations of
UV Flickering in YGem
R. Sahai1 , C. Sánchez Contreras2 , A. S. Mangan1,3, J. Sanz-Forcada2 , C. Muthumariappan4, and M. J. Claussen5
1 Jet Propulsion Laboratory, MS 183-900, California Institute of Technology, Pasadena, CA 91109, USA
2 Astrobiology Center (CSIC-INTA), ESAC Campus, E-28691 Villanueva de la Cañada, Madrid, Spain
3 Iowa State University, Ames, IA 50011, USA
4 Indian Institute of Astrophysics, Bangalore 560034, India
5 National Radio Astronomy Observatory, 1003 Lopezville Road, Socorro, NM 87801, USA
Received 2018 February 22; revised 2018 May 6; accepted 2018 May 8; published 2018 June 18
Abstract
Binarity is believed to dramatically affect the history and geometry of mass loss in AGB and post-AGB stars, but
observational evidence of binarity is sorely lacking. As part of a project to search for hot binary companions to cool
AGB stars using the GALEX archive, we discovered a late-M star, Y Gem, to be a source of strong and variable UV
and X-ray emission. Here we report UV spectroscopic observations of Y Gem obtained with the Hubble Space
Telescope that show strong flickering in the UV continuum on timescales of 20 s, characteristic of an active
accretion disk. Several UV lines with P-Cygni-type profiles from species such as Si IV and C IV are also observed,
with emission and absorption features that are red- and blueshifted by velocities of ∼500 km s-1from the systemic
velocity. Our model for these (and previous) observations is that material from the primary star is gravitationally
captured by a companion, producing a hot accretion disk. The latter powers a fast outflow that produces blueshifted
features due to the absorption of UV continuum emitted by the disk, whereas the redshifted emission features arise
in heated infalling material from the primary. The outflow velocities support a previous inference by Sahai et al.
that Y Gem’s companion is a low-mass main-sequence star. Blackbody fitting of the UV continuum implies an
accretion luminosity of about 13 Le, and thus a mass-accretion rate >5×10
−7M −1e yr ; we infer that Roche-lobe
overflow is the most likely binary accretion mode for Y Gem.
Key words: binaries: close – circumstellar matter – stars: AGB and post-AGB – stars: individual (Y Gem) –
stars: mass-loss
1. Introduction However, there is a lack of observational evidence of
One of the biggest challenges for 21st century stellar widespread binarity in AGB stars. We have therefore been
astronomy is a comprehensive understanding of the impact of using UV and X-ray observations as new probes of accretion-
related phenomena in AGB stars—archival surveys in the UV
binary interactions on stellar evolution. Close binary interac-
using the GALEX database show that a large fraction of AGB
tions are expected to dominate a substantial fraction of stellar stars show FUV emission, with relatively high FUV/NUV flux
phenomenology—e.g., cataclysmic variables, SNe Ia progeni- ratios and significant variability, likely resulting from variable
tors, and low- and high-mass X-ray binaries. accretion associated with a companion (Sahai et al. 2008,
Binary star interactions are specifically believed to underlie 2011b, 2015, 2016). Roughly, about 40% of objects surveyed
the formation of the overwhelming majority of planetary show strong, variable, X-ray emission that likely arises in an
nebulae (PNe), which represent the bright end stage of most accretion disk around a compact companion (see Sahai et al.
stars in the universe. Such interactions are likely the key to the 2015, hereafter Setal15). In this paper, we report stochastic
resolution of this long-standing puzzle: although PNe (and their variations on 20 s timescales in the UV spectrum of our best-
progenitors, pre-PNe [PPNe]) evolve from slowly expanding studied UV-/X-ray-emitting star, Y Gem, similar to the
(Vexp∼5–15 km s-1), spherically symmetric circumstellar flickering phenomenon seen in other well-known classes of
envelopes of AGB stars, modern surveys reveal that the vast accreting binaries. These variations provide independent and
majority of the former deviate strongly from spherical robust evidence of binarity in this object.
symmetry, showing a dazzling variety of elliptical, bipolar, Until recently, Y Gem lacked a significant measurement of
and multipolar morphologies and fast, collimated outflows the parallax (its Hipparcos parallax/error is 1.30 mas/
(Vexp50–100 km s-1; e.g., Sahai & Trauger 1998; Balick & 1.38 mas: van Leeuwen 2007), and we have adopted, in our
Frank 2002; Sahai et al. 2011a). previous studies of this object, a distance of D=0.58 kpc, as
A close association between jets and binary interaction inferred from its K-band magnitude, following Kahane & Jura
involving a red giant or AGB primary is exemplified by (1994) who assume that late-M semi-regular stars have
symbiotic stars, a small class of objects in which the optical absolute magnitudes MK=−7.6 (Sahai et al. 2011b: Setal11).
spectrum shows features of TiO (showing the presence of a In the recent Gaia Data Release 2 (L. Lindegren et al. 2018, in
cool red giant primary), but also optical emission lines, e.g., preparation), its measured parallax is 1.4997±0.1561 mas,
H I, He II, and [O III]. In such objects, a compact star, usually a giving a distance of 0.67±0.07 kpc. Since this is roughly
white dwarf (WD), accretes matter from the giant primary. within 1σ of our adopted value, we have conservatively kept
Prime examples of symbiotic stars with jets are R Aqr and our original distance estimate for this paper, especially since
CH Cyg (e.g., Corradi et al. 1999). the DR2 results have yet to be subjected to scrutiny by the
1
The Astrophysical Journal, 860:105 (8pp), 2018 June 20 Sahai et al.
Table 1 order to obtain a good signal-to-noise ratio while still
Log of Observations maintaining a large number of subexposures for our timing
Data Set Exp. #a Grating Start Time Exp.Timeb analysis. We ran calstis in PyRAF to create FITS-format files of
the calibrated subexposure data.
OD9C01010 10 G140L 2016-10-11 21:56:34 347.020
OD9C01020 20 G140L 2016-10-11 22:05:47 347.020
OD9C01030 30 G230L 2016-10-11 23:02:57 312.126 4. Results and Analysis
OD9C01040 40 G230L 2016-10-11 23:12:10 304.769
OD9C01050 50 G140L 2016-10-11 23:26:06 517.020 The G140L and G230L spectra show the presence of
OD9C01060 60 G140L 2016-10-11 23:38:09 517.020 continuum emission and many strong emission lines, some
OD9C01070 70 G230L 2016-10-12 00:38:23 503.600 with blueshifted absorption features resulting in P-Cygni-type
OD9C01080 80 G230L 2016-10-12 00:50:26 486.315 profiles (e.g., Castor 1970), from ionized levels of abundant
metals such as C, O, N, Si, and Mg. No molecular H2 lines are
Notes. seen. Representative spectra and line lists are given in Figures 1
a Exposure No. (a) and (b) and Table 2, respectively. For both the subexposure
b Exposure Time (s). and full exposure spectra, we used small line-free wavelength
windows around each strong line feature to subtract an
astrophysical community. The slightly larger distance, if underlying linear continuum, followed by Gaussian line fitting
correct, would imply a ∼30% increase in estimated luminos- to determine the line center, the FWHM, and the flux of the
ities and does not affect our conclusions. emission and absorption features (e.g., Figure 1(c)). The
The plan of this paper is as follows. We describe the instrumental resolution, which varies from 1.7 (2.2) to 1.4
observational setup in Section 2, and the data reduction and (2.1)Åover the wavelength range of the G140L (G230L)
calibration procedures in Section 3. The observational results grating, has been deconvolved from the line-width values in
and their analysis are described in Section 4, with a focus on Table 2. Representative line fitting results are given for
the spectral time variability (Section 4.1), the UV continuum exposures 20 and 30 in Table 2.
and its modeling (Section 4.2), and the line profiles
(Section 4.3). We present a new geometrical model to explain 4.1. Time Variations
our results and discuss their implications for binarity and
accretion in Y Gem in Section 5. Our main conclusions are Our analysis of the subexposure spectra reveals fast
summarized in Section 6. variations in the continuum UV flux of Y Gem at all
wavelengths (Figures 2(a) and (d)). The continuum light
curves, extracted from line-free regions, and that underlying the
2. Observations different spectral lines, are quite similar, with significant
We obtained UV spectra of Y Gem with the Space Telescope variations seen across successive subexposures, implying
Imaging Spectrograph (STIS) on board the Hubble Space variability on timescales of 20 s or less. The maximum peak-
Telescope (HST) between UT 22:05 10-11-2016 and 00:47 to-peak amplitude of the continuum variations is 0.53 mag
10-12-2016, using the G140L and G230L gratings, and the (exposure 60). Using the ratio between the observed and
52″×0 2 aperture in Time-Tag mode. The slit width fully expected root-mean-square variability (s and sexp, respectively)
encompasses the entire source, which is expected to be to quantify the significance of stochastic variations in the light
unresolved (Setal15). The observing sequence consisted of curves (see e.g., Nuñez et al. 2016), we find s sexp ~ 6–8 for
two pairs of G140L exposures followed by two pairs of several line-free continuum windows in the G140L and G230L
G2300L exposures; this sequence was repeated a second time, spectra. Periodogram analysis reveals no specific period; the
providing four separate exposures each in the two wavelength variations appear to be stochastic in nature.
bands (∼1120–1700Åand ∼1600–3150Å). An observation Fluctuations in the spectral-line parameters are also seen, but
log is given in Table 1. We obtained similar data during a the uncertainties in the fitted parameters are larger and so in
following epoch (2017 April), but during these observations the general, the variability is less significant than in the continuum.
source was much brighter, causing the Time-tag buffer to However, we do find significant fractional variations in the
overflow, so the absolute flux levels are not reliable. These data Si IV(1) absorption line equivalent width (Figure 2(b)) and in
will be discussed in a subsequent paper. the Si IV(1) absorption and emission-line Doppler shifts
(Figure 2(c)). In addition, the amplitude of the blueshift for
3. Data Reduction the Si IV(1) absorption feature appears to be anticorrelated with
the redshift of the Si IV(1) emission feature. In contrast, the
In order to observe changes in the spectra over time intervals Si IV(1) and C IV(1,2) emission-line fluxes show much smaller
smaller than the full integration time of each exposure, we variations within each exposure.
analyzed our data in Time-Tag mode, following the procedure During the period covering exposures 10–60 (∼1.7 hr),
described in the STIS Time-Tag Analysis Guide (Dashevsky excluding that covered by exposure 50, we find a general
et al. 2000). We used IRAF to run the “odelaytime” command, decline in the continuum fluxes and the Si IV(1) absorption
which applied heliocentric and barycentric time corrections, equivalent width and blueshift; however, these, together with
followed by “inttag,” which created a raw data file, with the continuum fluxes, all deviate upward from the above trend
multiple subexposures of a chosen duration. Each subexposure during exposure 50. The most striking decline (by factors >5)
used data only in the good time intervals (GTIs). The is seen in the Si IV(1) equivalent width, from the end of
subexposures at the end of each GTI had shorter durations exposure 50 to the end of exposure 60. Correspondingly, the
resulting in relatively larger errors in the associated fluxes. In FUV continuum declines to its lowest level in exposure 60. The
our analysis below, we chose a subexposure time of 20 s in Si IV(1) absorption line equivalent width shows a broad peak at
2
The Astrophysical Journal, 860:105 (8pp), 2018 June 20 Sahai et al.
Figure 1. Representative STIS/UV spectra of Y Gem observed using gratings (a) G140L (exposure 20) and (b) G230L (exposure 30). Exposures 20 and 30 were used
to calculate the parameters for the spectral lines in Table 2. The green curves show a model fit consisting of two blackbody components, characterized by
Teff=35,500 K, L=6.3 Le, and Teff=9400 K, L=6.7 Le. (c) STIS/UV spectra of Y Gem in the vicinity of the Si IV(1) line for the first two 20 s subexposure in
exposure 20. The blue curve shows Gaussian line-profile fits (together with a linear baseline) to the absorption (top) and emission (bottom) features. Such fitting has
been used to calculate the parameters for the spectral lines in Table 2.
an epoch that lies between the two peak luminosity epochs in covered in the G230L spectra; a significantly cooler blackbody is
exposure 50. We interpret these correlations in the context of a needed to fit the latter. For example, single-blackbody fits to the
geometric model in Section 5. G230L spectra from exposures 30, 40, 70, and 80 give values for
We note that much smaller amplitude photometric variations the luminosities of L∼7.0, 7.0, 4.9, and 5.5 Leand effective
have been recently reported at optical wavelengths (e.g., 0.06 mag temperatures of Teff∼13,140, 13,800, 13,400, and 13,450K,
peak to peak in the u′ band with a typical timescale of 10minutes) respectively. Thus, whereas the effective temperatures do not
for YGem using ground-based differential photometry (Snaid change much with time, the luminosities derived for both the
et al. 2018). Although a comparison with differential photometry hotter and cooler blackbodies above show significant changes
of two field stars indicates that these variations are significant, the that are similar to those seen in the continuum level in the
effects of aperture photometry in variable seeing conditions can G140L and G230L exposures. We have not included dust
produce spurious fluctuations that are not easily quantified. For extinction in our modeling because there is no measurable dust
example, in the Snaid et al. data set, large spurious fluctuations are excess in Y Gem (Setal11).
clearly noticeable in both the differential photometry and the In order to constrain the values of L and Tefffor each of these
seeing ∼10–20minutes into the observation time stream, but blackbodies and their variations more accurately, we construct
smaller seeing fluctuations are present in the remainder of the time a model that combines two blackbodies to fit the G140L and
stream and may be contributing to the photometric variations G230L spectra. Since the G140L and G230L spectra are not
as well. co-eval, and the fits above indicate that both L and Teffcan vary
with time, we use exposures 40 (for G230L) and 50 (for
G140L) as these were taken closest to each other in time. We
4.2. Continuum Modeling
find that the G140L spectrum requires L(h)=6.8 Leand
Single-blackbody fits (using least-squares minimization) to Teff(h)∼36,600 K, and a cooler blackbody with L(c)∼6.3 Le
the G140L spectra from exposures 10, 20, 50, and 60 give values and Teff(c)∼9940 K. Thus, L(h) and Teff(h) respectively
for the luminosities of L∼6.9, 6.0, 6.4, and 4.5 Leand increase by about 6% and 24%, and L(c) and Teff(c)
effective temperatures of Teff∼31,440, 29,600, 29,540, and respectively decrease by about 29% and 24%, over the
29,980K, respectively. However, these blackbodies do not corresponding single-blackbody fits. The two-blackbody fits
provide sufficient flux at the longer wavelengths (i.e., 1800 Å) for the other pairs of successive G140L+G230L exposures
3
The Astrophysical Journal, 860:105 (8pp), 2018 June 20 Sahai et al.
Table 2
Properties of Selected Observed Lines from Exposures 20 (G140L) and 30 (G230L)
Name Rest λ Emiss./Abs.a Obs. λ Vhel FWHM
b Flux
(Å) (Å) (km s-1) (km s-1) (10−13 ergs cm−2 s−2)
Ly α 1215.670 Emiss. 1218.573±0.008 716±2 774±4 54.5±0.4
Si IV(1)c 1393.756 Abs. 1391.200±0.030 −549±7 565±17 −4.70±0.15
Si IV(1)c 1393.756 Emiss. 1395.657±0.023 409±5 436±11 6.84±0.19
Si IV(2)d 1402.770 Abs. 1400.366±0.037 −514±8 322±19 −2.43±0.13
Si IV(2)d 1402.770 Emiss. 1404.613±0.030 394±6 518±16 6.73±0.23
C IV(1,2)e 1549.490 Emiss. 1553.249±0.026 727±5 941±11 27.7±0.5
C IV(1,2)e 1549.490 Abs. 1545.309±0.060 −809±12 588±30 −3.42±0.21
Si III] 1892.030 Emiss. 1893.453±0.026 225±4 915±9 34.5±0.5
C III]f 1908.734 Emiss. 1910.320±0.038 249±6 1022±14 23.4±0.4
Mg II(1)c 2796.352 Emiss. 2798.318±0.013 211±1 537±3 49.8±0.4
Mg II(2)d 2803.531 Emiss. 2804.983±0.033 155±4 511±7 27.8±0.4
Notes.
a Emiss.—Emission, Abs.—Absorption.
b Corrected for instrumental resolution.
c First line in a resolved (or partially resolved) doublet.
d Second line in a resolved (or partially resolved) doublet.
e Mean wavelength of unresolved C IV doublet with λ0=1548.20 Åand 1550.78 Å.
f Unresolved blend with [C III] transition at λ0=1906.683 Å.
(20+30 or 60+70) produce very similar values of L(h), Teff(h), But the large values of the redshifts of the emission features
6
L(c), and Teff(c), but these are much more uncertain as the (Vred(e)) are discrepant from our expectation for a classical
G140L+G230L exposure pairs have very long time gaps P-Cygni profile that results from an outflow surrounding a
between them (about 1 hr). continuum source—in these profiles, the emission feature is
We investigate luminosity and temperature changes in both generally centered near or at the systemic velocity (Castor 1970).
blackbodies on the short variability timescales that we find for However, if the outflow velocity distribution covers values over a
the continuum, as follows. First, we determine L(h) and Teff(h) range from ∼0 to Vmax and there is substantial line broadening,
as a function of subexposure in exposure 50 using two- e.g., due to microturbulence, the absorption feature may be wide
blackbody fits, with L(c) and Teff(c) set to their (average) values enough to extend to redshifted velocities, thereby reducing the
for exposure 40 (as determined above). Then, in order to emission feature’s blue-wing intensity and thus shifting its
account for the variability of L(c) and Teff(c) during exposure centroid substantially redwards of the systemic velocity. But, as
50, we follow the following procedure. First, using a single- can be seen in the predicted spectra from detailed P-Cygni models
blackbody fit to the subexposure spectra in exposure 40, we by van Loon et al. (2001, hereafter vLKH01), it is only when the
find that the peak-to-peak variations in L(c) and Teff(c) are absorption feature is extremely saturated, and the microturbulence
about ±10% and ±3%, respectively. Using these to set upper is a substantial fraction of the maximum outflow velocity (V ),
and lower bounds on the average value of L(c) and Teff(c), we
max
that one gets values of Vred(e) comparable to the blueshift of thedetermine the upper and lower bounds on the variability of L(h) absorption feature, which is located at or near −V (e.g., see
and Teff(h). We find significant variations in L(h) (Figure 2(e)),
max
( ) Figure 9(a) of vLKH01, where τ=300 and the line-broadeningbut not in Teff h . A similar analysis for the cooler blackbody parameter σ =0.45).7shows that the variations in L(c) are marginally signi cant v When the lines are less optically thick,fi
(Figure 2(f)), and no signi cant variations are found for T (c). Vred(e) is about 0.3–0.5 for τ=10–100 (e.g., see Figure 6(a), (b)fi eff
The variations in L(h) and L(c) follow the variations in the of vLKH01, where σv=0.2). Since the observed absorption
continuum uxes (Figures 2(e) and (f)). features in our spectra are clearly not heavily saturated, wefl
conclude that a classical P-Cygni model can probably not explain
YGem’s UV lines.
4.3. Line Pro les Another model to explain P-Cygni profiles is one that wasfi
proposed for the PPN Hen 3–1475 (Sánchez Contreras & Sahai
All emission lines are redshifted relative to the systemic 2001) to explain its Hα profile (observed with HST/STIS), which
velocity (Vhel=13.5 km s-1); the velocities lie in the range also shows blueshifted (redshifted) absorption (emission) features.
Vhel∼400–700 (150–250) km s-1for lines observed with the This model features a fast, neutral, collimated outflow within
G140L (G230L) grating (Table 2, Figure 3). Prominent more slowly expanding bipolar lobes with dense, dusty walls. Hα
P-Cygni-type absorption features are seen in lines of N V, photons produced from a central source pass through gas in the
O I, Si IV, and C IV. The blueshifted absorption features in these collimated outflow producing the absorption feature and then are
profiles are consistent with the presence of a high-speed scattered from the dusty walls along the los, producing a
outflow (>500 km s-1) along the line of sight (los) to a hot redshifted emission feature. However, unlike Hen 3–1475, there is
continuum source. The lack of absorption features in semi-
forbidden lines such as Si III]λ 1892.03 and C III]λ 1908.73 is 6 As a fraction of the outflow velocity.
probably due to these being excited in a low-density region that 7 In these and other plots of model spectra in vLKH01, all velocities are given
is much larger than the continuum source. as a fraction of Vmax, which is set to 1.
4
The Astrophysical Journal, 860:105 (8pp), 2018 June 20 Sahai et al.
Figure 2. Short-term time variations in the continuum and lines observed in Y Gem: (a) line-free continuum in the bands 1346–1367 Å(blue) and 1571–1600 Å(red),
and continuum underlying the Si IV(1) line (green). (b) The equivalent width of the Si IV(1) absorption line (blue) and the flux of the Si IV(1) (red) and C IV(1, 2)
emission lines (pink). The emission-line uxes have been scaled up by a factor of 1.5×1012fl . (c) The Doppler shifts (absolute values) of the lines in panel (b): the
Si IV(1) absorption feature is blueshifted, whereas the Si IV(1) and C IV(1, 2) emission features are redshifted. (d) The line-free continuum in the bands
2000–2100 Å(blue) and 3013–3068 Å(red), and the continuum underlying the Mg II(1) line (green; scaled up by a factor of 1.1 for clarity). Error bars in
panels (a)–(d) are ±1σ. The light curves in panels (a)–(c) and (d) are, respectively, extracted from the G140L exposures 10, 20, 50, and 60 and G230L exposures 30,
40,70, and 80. (e) The hotter blackbody’s luminosity (black), derived using a two-blackbody fit to the continuum observed in the subexposures within exposure 50: the
dashed curves show the upper and lower bounds on the luminosity due to the estimated uncertainties in the cooler blackbody’s luminosity and temperature. The
1571–1600 Åcontinuum, scaled up by a factor of 1.1×107 (cyan), and the square root of the Si IV(1) absorption line equivalent width, scaled up by a factor of 5.2
(blue), are shown for comparison. (f) As in panel (e), but for the cooler blackbody’s luminosity, derived from fitting the subexposures within exposure 40 (the
2000–2100 Å continuum, scaled up by a factor of 2.6×1013 (cyan), is shown for comparison).
no measurable dust excess in Y Gem (Setal11), so the presence of 5. Discussion
dusty lobes in it is somewhat implausible. The short-term stochastic fluctuations in the UV continuum
We propose below a new geometric model to explain the of Y Gem appear to be similar to the photometric variations
large values of Vred(e). seen in other well-known classes of accreting binaries,
5
The Astrophysical Journal, 860:105 (8pp), 2018 June 20 Sahai et al.
value for a cool WD (Teff∼36,500 K, L<0.5 Le; Figure 8 of
Miller Bertolami 2016). The cooler blackbody is under-
luminous for a main-sequence star (Teff∼9400 K, i.e., spectral
type early-A, L∼30–80 Le). Note, of course, that neither of
these blackbodies is consistent with the temperature and
luminosity of a 0.35Memain-sequence companion. We
conclude that both the hot and cool UV components arise in
the accretion disk.
Assuming that the combined luminosity of the hot and cool
blackbodies results from accretion, i.e., Lacc=13 Leduring
exposures 40–50, then taking Lacc  GM˙accMc Rc, where Ṁacc
is the accretion rate, and Mc∼0.35Meand Rc=0.44 Re(as
above), we find that Ṁacc>5×10
−7M −1e yr . This relatively
large accretion rate makes wind-accretion mechanisms such as
Bondi–Hoyle or wind Roche-lobe overflow (e.g., Huarte-
Figure 3. Lyα (blue) and SI IV(1) (red) line profiles (continuum-subtracted) Espinosa et al. 2013; Chen et al. 2017) infeasible because
from exposure 20. The dashed vertical line shows the systemic velocity. Y Gem does not have a detectable wind—the very weak,
narrow CO J=2–1 emission line detected toward it likely
commonly labelled as “flickering”—e.g., cataclysmic variables arises in an extended disk (Setal11), and the infrared excess is
and recurrent novae (e.g., Bruch 2015), X-ray binaries with negligible or very low (McDonald et al. 2012), implying a
neutron stars and black holes (e.g., van der Klis 2006), mass-loss rate of 10−7Me yr−1(Snaid et al. 2018). There-
symbiotic stars (e.g., Zamanov et al. 2017), and active galactic fore, either the primary overflows its Roche lobe and transfers
nuclei (e.g., Pronik et al. 1999). The timescale of flickering material to the accreting star, or accretion occurs within a
depends on the dimensions of the region where the emission common-envelope (CE) configuration. For these two accretion
arises. modes, the binary separation must be small enough and
Using the light-travel time corresponding to 20 s, char- comparable to the primary star’s radius—1.5 au for Y Gem’s
acteristic of the variations that we have found in Y Gem, we luminosity and effective temperature of 5800 Leand 2800 K,
estimate that the continuum-emitting region is less than respectively (Setal11). However, since we observe the accre-
∼0.05 au (8.6 Re) in size. Our previous X-ray and UV tion luminosity in the UV, a CE scenario—in which the
observations of Y Gem also show variability, including both radiation would be trapped within the envelope—is less likely.
a quasi-periodic and a stochastic component (Setal15). The We conclude that Roche-lobe overflow is the most likely
periodic component seen most clearly in the X-ray data has accretion mode for Y Gem.
P∼1.3 hr, which Setal15 interpreted as being associated with The absence of molecular H2 lines (e.g., prominently
emission from the inner radius of an accretion disk around a detected in accreting T Tauri stars: Ingleby et al. 2011) in
subsolar mass (0.35Me) main-sequence (MS) companion, Y Gem’s UV spectrum is striking, especially since the
although a cold (Teff<30,000 K) WD could not be excluded.
8
detection of the 6.4 keV Fe I line in its X-ray spectrum implies
However, the bulk outflow speed implied by the velocity the presence of a neutral disk in it. The two main mechanisms
offset (from the systemic velocity) of the absorption features
- for exciting H2 lines are fluorescence due to Lyαpumping(∼500 km s 1) strongly argues for a main-sequence compa- (e.g., Yang et al. 2011) or collisional excitation by hot electrons
nion. On both theoretical and empirical grounds, the speed of (e.g., Ingleby et al. 2009). If the disk in Y Gem is composed
an outflow driven from an accretion disk is expected to be of primarily of molecular gas, then the absence of the H lines
the order of the escape velocity close to the central accreting 2
object (Livio 1997), V = 620 km s-1 0.5
suggests that there are large temporal variations in the total
esc (Mc Rc) , where Mc amount of neutral material in the disk. Alternatively, H2 lineand Rc are the companion mass and radius in solar units. Thus, emission could be present but blocked from the observer’s
the observed bulk expansion velocity of the high-speed outflow view by an optically thick continuum-emitting region of hot
is consistent with a solar or subsolar MS companion—e.g., if gas. Otherwise, the disk is primarily composed of atomic gas.
Mc∼0.35Me(thus, Rc=0.44 Re: inferred using Table 15.8 We propose a simple geometric model that can explain the
in Cox 2000), which is the largest companion mass allowed Doppler shifts of the absorption and emission features
by the 1.3 hr orbital period seen in Y Gem’s X-ray light
-1 (Figure 4). In this model, material from the primary AGB starcurve (Setal15), then Vesc=550 km s . For a cool WD, (either from an outflow or Roche-lobe overflow) is gravita-
with a typical mass of 0.6Me and radius of 0.01 Re,
-1( )0.5( )0.5 tionally focused toward the companion, producing a hotVesc = 4800 km s Mc 0.6 0.01 Rc is much larger than accretion disk. The emission features arise in the infalling
the observed outflow velocities. Any correction to the outflow material, which gets heated either from internal shocks (e.g.,
velocity due to projection effects is likely to be small, given our due to density and velocity variations within the stream) and/or
geometrical model for the continuum source and outflow friction due to passage through the accretion disk’s “atmos-
(described later in this section). phere.” The accretion disk produces an outflow that is seen in
Neither of the two blackbodies discussed in Section 4.2 fit absorption against the continuum emission from the accretion
the properties of a viable stellar companion to the primary. The disk. Sufficiently hot regions in the disk may also produce UV
hotter blackbody is overluminous compared to the expected
emission lines, which would be centered roughly around the
8 systemic velocity. Since we do not see a signature of suchA hot WD (as in symbiotic stars) is ruled out by Setal15 because no optical
forbidden line emission—characteristic of symbiotic star spectra—is seen in emission in our spectra, the disk emission lines must be
Y Gem’s M8 optical spectrum. relatively faint compared to those from the outflow, and the
6
The Astrophysical Journal, 860:105 (8pp), 2018 June 20 Sahai et al.
Figure 5. Ionization fractions of C IV (green) and Si IV (red) for a range of
temperatures for a cloud of density 103 cm−3.
Si IV (C IV) ionization state (Figure 5). We note that within this
Figure 4. Schematic geometry of a model (not to scale) to explain the Doppler temperature range, the Si IV and C IV ionization states are also
shifts of the absorption and emission features observed in the UV spectra of
Y Gem. A hot accretion disk (blue ellipse) around a compact companion (cyan well populated; they represent the second most populated5
circle) captures part of the material (red arrows) in an outflow or Roche-lobe states: their populations differ, at e.g., T∼0.9×10 K, by
overflow from the primary AGB star (red circle) via two possible infall streams factors of 2.4 and 3.3, respectively, from that of the most
(dashed red/pink arrows)—these produce redshifted emission features. The populated ones.
accretion disk powers a fast outflow (cyan arrows) that absorbs UV photons
from the disk, producing blueshifted absorption features. The anticorrelation between the blueshift for the Si IV(1)
absorption feature and the redshift of the Si IV(1) emission
feature may be explained as follows. In our model, the P-Cygni
size of the hot disk region must be small compared with that of feature is due to the sum of the features produced by (i) the
the outflow. absorption of continuum by the outflow and (ii) the emission
The large observed values of the blue and red Doppler shifts from the infalling stream (which sits on top of the underlying
of the absorption and emission features, respectively, are continuum). When the outflow velocity increases, the absorp-
naturally explained in this model because both features tion feature shifts bluewards as a whole, thus making the blue
are associated with material moving in the gravitational well side of the emission feature stronger by decreasing the
of the central star near the accretion disk’s inner regions. The underlying absorption—the net result is that the observed
emission features from high-excitation lines (C IV, Si IV) are redshift of the emission feature becomes smaller.
generated closer to the disk, in the infall’s hottest parts, In order to explain the observed short-term variations in the
whereas the low-excitation Mg II lines are generated in a region line widths, red- and blueshifts, and fluxes, we propose that the
farther away, hence the redshift of the Mg II lines is much infall and outflow streams experience short-term temporal
lower than that of the high-excitation lines. variations in the densities, temperatures, velocities, and velocity
The correlation between the variations seen in the FUV gradients in these streams that are likely stochastic. The
continuum and Si IV absorption line features suggests that both relatively large observed line widths—e.g., ∼335–515 km s-1
of these are associated with variations in the accretion rate. The (∼775–1000 km s-1) for the emission features in Si IV(1) (Lyα),
mismatch between the epoch during which the Si IV(1) depending on the epoch, are likely a consequence of velocity
absorption line equivalent width peaks and that at which L(h) gradients (due to both turbulent and systematic motion) and the
peaks (in exposure 50) confirms that these arise in spatially presence of a range of inclinations, relative to the los, within the
separated structures. In this model, the long-term UV and X-ray outflow and infall streams.
variations noted by Setal15 may result from orbital motion as
varying segments of the disk are eclipsed by the AGB star.
We also find an anticorrelation between the fluxes of the 6. Conclusions
C IV and Si IV emission lines—e.g., over the period covered by Using the Hubble Space Telescope, we have carried out UV
exposures 10–60, the Si IV flux shows a general increase, spectroscopic observations of the late-M star, Y Gem—the
whereas the C IV flux shows a general decrease (Figure 2(b)). A most prominent member of a class of AGB stars that are
plausible explanation for this anticorrelation is that it results sources of strong and variable UV and X-ray emission, likely
from changes in the ionization fractions of each of these species resulting from accretion activity due to the presence of a binary
as a function of temperature. Our simple CLOUDY (Ferland companion.
et al. 2013) modeling using uniform low-density spherical
plasma clouds, with a wide range of plausible densities, 1. Y Gem shows the presence of strong emission in
∼10 105 cm−3– , shows that the ionization fractions of C IV and the FUV (∼1120–1700 Å) and NUV (∼1600–3150 Å)
Si IV are anticorrelated for temperatures in the range bands, both in the continuum as well as in lines such as
(0.7–1.1)×105 K—i.e., a decrease in temperature results in Lyα, C IV λλ1548,1551, Si IV λλ1394,1403, and Mg II
an increase (decrease) in the fractional population of the λλ2796,2803.
7
The Astrophysical Journal, 860:105 (8pp), 2018 June 20 Sahai et al.
2. The UV continuum shows short-term stochastic time J.S.F. acknowledges support from the Spanish MINECO through
variations (on timescales of 20 s); this flickering grant AYA2014-54348-C3-2-R. The National Radio Astronomy
phenomenon is characteristic of the presence of an active Observatory is a facility of the National Science Foundation
accretion disk. We also find a long-term trend (overall operated under cooperative agreement by Associated Universi-
decline) in the FUV continuum over a period of ∼1.7 hr. ties, Inc.
3. The continuum can be modeled as a sum of a hotter and a
cooler blackbody component. The luminosities of these
components, as derived from the best-modeled pair of ORCID iDs
near-contemporaneous FUV and NUV spectra, are R. Sahai https://orcid.org/0000-0002-6858-5063
∼6.8 Leand 6.3 Le, and the temperatures are
4 4 C. Sánchez Contreras https://orcid.org/0000-0002-6341-592X∼3.7×10 K and ∼10 K, respectively. The changes J. Sanz-Forcada https://orcid.org/0000-0002-1600-7835
in the FUV continuum appear to be related to similar
changes in the luminosity of the hotter blackbody; the
temperature is significantly less variable. References
4. Neither of the two blackbodies fits the properties of a
viable stellar companion (WD or main sequence) to the Balick, B., & Frank, A. 2002, ARA&A, 40, 43
primary, suggesting that both the hot and cool UV Bruch, A. 2015, A&A, 579, A50
Castor, J. I. 1970, MNRAS, 149, 111
components arise in the accretion disk. Chen, Z., Frank, A., Blackman, E. G., Nordhaus, J., & Carroll-Nellenback, J.
5. Lines from species such as Si IV and C IV show 2017, MNRAS, 468, 4465
prominent P-Cygni-type profiles, with emission and Corradi, R. L. M., Brandi, E., Ferrer, O. E., & Schwarz, H. E. 1999, A&A,
absorption features that are red- and blueshifted by 343, 841
∼500 km s-1from the systemic velocity. The relatively Cox, A. N. 2000, Allen’s Astrophysical Quantities (4th ed; New York:AIP Press)
large redshifts of the emission features are not consistent Dashevsky, I., Sahu, K., & Smith, E. 2000, Space Telescope STIS Instrument
with a classical P-Cygni profile that results from an Science Report 2000-02
outflow surrounding a continuum source. Ferland, G. J., Porter, R. L., van Hoof, P. A. M., et al. 2013, RMxAA, 49, 137
6. Our model for these (and previous) observations is that Huarte-Espinosa, M., Carroll-Nellenback, J., Nordhaus, J., Frank, A., &
Blackman, E. G. 2013, MNRAS, 433, 295
material from the primary star is gravitationally captured Ingleby, L., Calvet, N., Bergin, E., et al. 2009, ApJL, 703, L137
by a companion, producing a hot accretion disk. The Ingleby, L., Calvet, N., Bergin, E., et al. 2011, ApJ, 743, 105
latter powers a fast outflow that produces blueshifted Kahane, C., & Jura, M. 1994, A&A, 290, 183
features due to the absorption of the UV continuum Livio, M. 1997, in ASP Conf. Ser. 121, Accretion Phenomena and Related
emitted by the disk, whereas the redshifted emission Outflows, ed. D. T. Wickramasinghe, G. V. Bicknell, & L. Ferrario (SanFrancisco, CA: ASP), 845
features arise in heated infalling material from the McDonald, I., Zijlstra, A. A., & Boyer, M. L. 2012, MNRAS, 427, 343
primary. Miller Bertolami, M. M. 2016, A&A, 588, A25
7. The outflow velocities support a previous inference that Nuñez, N. E., Nelson, T., Mukai, K., Sokoloski, J. L., & Luna, G. J. M. 2016,
Y Gem’s companion is a low-mass main-sequence star. ApJ, 824, 23
Pronik, I. I., Merkulova, N. I., & Metik, L. P. 1999, A&A, 351, 21
8. The combined luminosity of the hot and cool blackbodies Sahai, R., Findeisen, K., Gil de Paz, A., & Sánchez Contreras, C. 2008, ApJ,
require a relatively large accretion rate,>5×10−7M −1e yr ; 689, 1274
since the primary in YGem has negligible (or very weak) Sahai, R., Morris, M. R., & Villar, G. G. 2011a, AJ, 141, 134
mass loss, wind-accretion modes are infeasible, and between Sahai, R., Neill, J. D., Gil de Paz, A., & Sánchez Contreras, C. 2011b, ApJL,
CE and Roche-lobe overflow accretion modes, the latter is 740, L39Sahai, R., Sanz-Forcada, J., & Sánchez Contreras, C. 2016, JPhCS, 728,
preferred. 042003
Sahai, R., Sanz-Forcada, J., Sánchez Contreras, C., & Stute, M. 2015, ApJ,
We acknowledge helpful discussions with Eric Blackman 810, 77
related to binary accretion modes. We thank the referee, Dr. Jacco Sahai, R., & Trauger, J. T. 1998, AJ, 116, 1357
Sánchez Contreras, C., & Sahai, R. 2001, ApJL, 553, L173
Th. van Loon, for helpful comments that helped to improve an Snaid, S., Zijlstra, A. A., McDonald, I., et al. 2018, MNRAS, 477, 4200
earlier version of this paper. R.S.’s contribution to the research van der Klis, M. 2006, in Compact Stellar X-ray Sources (Cambridge:
described here was carried out at the Jet Propulsion Laboratory, Cambridge Univ. Press), 39
California Institute of Technology, under a contract with NASA. van Leeuwen, F. 2007, A&A, 474, 653
C.S.C. has been partially supported by the Spanish MINECO van Loon, J. T., Kaper, L., & Hammerschlag-Hensberge, G. 2001, A&A,375, 498
through grant AYA2016-75066-C2-1-P and by the European Yang, H., Linsky, J. L., & France, K. 2011, ApJL, 730, L10
Research Council through ERC grant 610256: NANOCOSMOS. Zamanov, R. K., Boeva, S., Nikolov, Y. M., et al. 2017, AN, 338, 680
8