FUSE Spectroscopic Analysis of the Slowest Symbiotic Nova AG Peg During Quiescence

AG Peg is the slowest known symbiotic nova (M3/4III giant + hot white dwarf; P_orb = 818.4 days) which underwent a nova explosion in 1850 followed by ∼165 yr of decline and quiescence. A number of studies (Kenyon et al. 2001; Yoo 2006, 2008; Kim & Hyung 2008; Sanad & Bobrowsky 2017) document the very slow decline, since the 1850 nova explosion, and the transition into quiescence (up to ∼1995). From roughly 1997 to 2015 June, the brightness of AG Peg appears to be fairly constant: its visual magnitude (e.g., AAVSO light curve—LC) shows variation between about ∼9.3 and ∼8.3. Its photometric U-LC clearly exhibits a modulation with the orbital phase (Skopal et al. 2017, attributed to the nebular component). This led Skopal et al. (2017, and references therein) to confirm that AG Peg had at last reached its quiescence. In 2015, AG Peg unexpectedly transitioned to a weaker, sharper outburst with an optical amplitude of ∼1.5 magnitudes and short duration characteristic of Z-And-type symbiotic variables. This new type of outburst may be due to gravitational energy release driven by (disk?) accretion and unlikely to be of thermonuclear origin like the 1850 outburst. This kind of transition of a symbiotic nova to Z-And-type outbursts is unprecedented.

In order to shed light on the nature of the hot component (the "central engine"), we present here an analysis of a Far-Ultraviolet Spectroscopic Explorer (FUSE) spectrum of AG Peg obtained in quiescence. In this far-ultraviolet wavelength range down to the Lyman limit, the contribution of the nebular continuum produced by the photoionization of the red giant wind (which flattens the slope) is totally negligible. In this region, we are probing the flux of the innermost disk and white dwarf photosphere.

The published orbital and physical parameters of AG Peg are listed in Table 1 with their references. Its Gaia (DR2, Prusti et al. 2016; Brown et al. 2018) parallax is ϖ ∼ 0.380298.. ± 0.081982.. mas, giving a distance is pc. However, the standard uncertainty is σ_ϖ ∼0.02 for a 9 mag star (e.g., Schaefer 2018) with such a parallax (ϖ ∼ 0.4 mas), or four times smaller. The uncertainty in distance (d = 1/ϖ) will have a subtantially non-Gaussian shape when σ_ϖ grows to a substantial fraction of the parallax ϖ. The problem becomes nontrivial for cases where σ_ϖ/ϖ ∼ 0.2 or larger (Bailer-Jones 2015). The parallax of AG Peg has a significance of 4.6σ, and from its quality flags (e.g., goodness of fit, number of visibility periods) it appears that the Gaia parallax for AG Peg is actually unreliable (see, e.g., Eyer et al. 2018; Luri et al. 2018; see also https://gea.esac.esa.int/archive/documentation/GDR2/Data_processing/). One of the explanations for the unreliability of AG Peg parallax is the fact that a long-period binary orbit will cause the center of light to wobble with a shift comparable to the parallaxe. As the binary separation is about 2 au with a period of almost 2 yr, over a 6 month period, the components move a distance of the order ∼1 au (depending on the exact mass ratio and inclination). Therefore, we mainly present results based on the widely accepted pre-Gaia distance to AG Peg of ∼800 pc.

Table 1.  Orbital and Physical Parameters of AG Peg

Parameter	Value	References
Period	818.4 days	Fekel et al. (2000)
Inclination	50°	Kenyon et al. (1993)
Distance	800 pc	Kenyon et al. (1993)
Separation	2.5 ± 0.1 au	Kenyon et al. (1993)
	0.10 ± 0.05	Kenyon et al. (1993)
RG Spectral Type	M3/4III	Kenyon et al. (1993)
WD Mass	0.65 ± 0.1 M⊙	Kenyon et al. (1993)
WD Radius (quiescence)	∼0.05–0.06 R⊙	Skopal et al. (2017)
WD Temperature (quiescence)	95k K, ∼160k K	Mürset et al. (1991), Skopal et al. (2017)
RG Mass	2.6 ± 0.4 M⊙	Kenyon et al. (1993)
RG Temperature	3787 ± 387 K	Akras et al. (2019)

Download table as:  ASCIITypeset image

In Section 2, we provide the details of the FUSE observation including the complexities encountered and how we corrected for them, particularly the molecular hydrogen absorption that is pervasive in the FUSE wavelength range along with other ISM absorption features (see Section 2.1). In Section 2.2, we describe our suite of modeling codes for both NLTE high-gravity model atmospheres and model accretion disks: the Wade & Hubeny (1998) grid of standard, steady-state disk models and accretion disk models we have generated from scratch. Our grid has nonstandard accretion disk models in the form of (inner and outer) disk truncation, and covers parameter space outside of the Wade and Hubeny grid. In Section 3, we present the results of our synthetic spectral analysis of the FUSE data using accretion disk models and white dwarf models. Finally, in Section 4 we summarize our conclusions.

2.1. The FUSE Spectrum

AG Peg was observed with FUSE on 2003 June 5.618 (exposure start time MJD 52795.61791667) and data from the AAVSO indicate that it had a magnitude m_v = 8.5. The orbital phase at the time of the FUSE observation was ϕ = 0.31, which we derived using the orbital ephemeris of Fekel et al. (2000) 2,447,165.3(±48)+ 818.2(±1.6) x E (itself based upon the inferior conjunction of the M3/4 giant). We retrieved the FUSE data (ID Q1110103) from the MAST archive. The instrument was set in time-tag and the data were obtained through the MDRS aperture with an exposure time of 2116 s (one single FUSE orbit). The data were calibrated through the pipeline with the final version of CalFUSE (Dixon et al. 2007) and further processed using our suite of FORTRAN programs, unix scripts, and IRAF procedures especially written for this purpose (Godon et al. 2012). The FUSE data come in the form of eight spectral segments (SiC1a, SiC1b, SiC2a, SiC2b, LiF1a, LiF1b, LiF2a, and LiF2b), which have to be combined together to give the final FUSE spectrum. These spectral segments do overlap and provide a way to renormalize the spectra in the SiC1, SiC2, and LiF2 channels to the flux in the LiF1 channel (which is the most reliable segment). In the present case, the two SiC channels experienced intermittent (and unexplained) drops in the count rates (which was more severe for the SiC2 channel than for the SiC1 channel). When combining the eight spectral segments together, we took care to renormalize the SiC spectral segments to the LiF spectral segments.

The FUSE spectrum with line identifications is displayed in Figure 1. Many ISM molecular hydrogen absorption lines are indicated with vertical tick marks. Numerous helium lines dominate the shortest wavelengths. Also seen are neon emission lines. FPN is a fixed pattern noise from the FUSE detector.

Zoom In Zoom Out Reset image size

The FUSE spectrum was previously analyzed by Eriksson et al. (2006) who studied the locations, origins, and excitation mechanisms of the emission lines, and by Skopal et al. (2017) who used the FUSE spectrum to analyze the O vi 1032 Å line as part of a multiwavelength analysis.

In preparation for the spectral analysis, we dereddened the FUSE spectrum assuming (see Table 1) and using our dereddening script based on the extinction curve from Fitzpatrick & Massa (2007) with R = 3.1.

2.2. Spectral Modeling and Analysis

In the following, we adopt a distance of 800 pc, an inclination of 50°, a WD mass of about 0.65 M_⊙ and we expect the WD temperature to be at least of the order of 100,000 K (see Table 1). We carry out all the disk and photosphere model fits for these values.

3.1. NLTE WD Photosphere Analysis

The input gravity of the models was varied from log(g) = 6.0 to log(g) = 8.0 in steps of 0.5, and the WD temperature was varied from ∼40,000 to ∼200,000 K (in steps of 5000 and 10,000 K). We find that the best-fit WD photosphere models are obtained for the lowest gravity (log(g) = 6.0, 6.5) in our grid of models and for a temperature T_wd = 100,000 to 180,000 K. The radius of the WD is then obtained by scaling the theoretical spectrum obtained for a distance of 800 pc to the observed spectrum. The exact mass of the WD is uknown but it is likely around ∼0.65 M_⊙ (see Table 1), and an output value of the gravity, log(g), can then be obtained and compared to the input value of log(g) for self-consistency.

Explicitly, for T_wd = 100,000 K, the WD radius scaled to a distance of 800 pc is R_wd = 55.5 × 10⁹ cm (∼0.08 R_⊙), giving log(g) = 6.44 for an assumed WD mass of M_wd = 0.65 M_⊙ (see Table 2). Values of log(g) are also listed for a 0.4 M_⊙ and 1 M_⊙ WD masses and are consistant with the best fits obtained for log(g) = 6.0 and log(g) = 6.5.

Table 2.  Single Hot Component Fitting Results

Th	Rh	Rh	Mh	Log(g)	L	L
(1000 K)	(1000 km)	(R⊙)	(M⊙)	(cgs)	(L⊙)	(erg s−1)
100	55.5	0.0798	0.4	6.23	572	2.19 × 1036
100	55.5	0.0798	0.65	6.44	572	2.19 × 1036
100	55.5	0.0798	1.0	6.63	572	2.19 × 1036
150	42.9	0.0615	0.4	6.46	1729	6.64 × 1036
150	42.9	0.0615	0.65	6.67	1729	6.64 × 1036
150	42.9	0.0615	1.0	6.86	1729	6.64 × 1036
180	39.6	0.0570	0.4	6.53	3056	1.17 × 1037
180	39.6	0.0570	0.65	6.74	3056	1.17 × 1037
180	39.6	0.0570	1.0	6.93	3056	1.17 × 1037

Note. The radius of the hot component R_h is derived by scaling the model flux to the observed flux assuming a distance of 800 pc. The output gravity is then obtained assuming a given mass for the hot component.

Download table as:  ASCIITypeset image

For the hottest model T_wd = 180,000 K, the WD radius scaled to a distance of 800 pc is R_wd = 39.6 × 10⁹ cm (0.057 R_⊙), giving log(g) = 6.74 for a 0.65 M_⊙ WD mass (Table 2). The 100,000 K model is slightly defficient in flux around 930–950 Å, while the 180,000 K model provides slightly too much flux in that region. At longer wavelengths, the models provide the same fit. Among these models, the best-fitting photosphere (namely in the 930–950 Å region) has T_wd = 150,000 K, and the WD radius scaled to a distance of 800 pc is R_wd = 42.9 × 10⁹ cm (0.0615 R_⊙), giving log(g) = 6.67 for a 0.65 M_⊙ WD mass. This WD photosphere model fit to the FUSE spectrum is displayed in Figure 2, and includes the addition of ISM absorption lines. We note that our derived value of T_wd agrees with the multiwavelength analysis of Skopal et al. (2017) who took T_wd = 160,000 K but did not model the FUSE spectrum explicitly. The low mass (and/or low gravity) of the white dwarf is consistent with the very small mass function derived from the full orbit solution of Fekel et al. (2000).

Zoom In Zoom Out Reset image size

To further check the self-consistency of these models, we compute the luminosity (see Table 2, last column). Our best-fit model has a luminosity of 1729 L_⊙, which agrees with Skopal et al.'s (2017) estimate, which is not surprising because the temperature is of the same order as Skopal et al.'s (2017). However, our hotest model has a luminosity about twice as large, which must be ruled out.

3.2. NLTE Accretion Disk Analysis

We extended our analysis of the FUSE spectrum to optically thick, steady-state accretion disk models. We first used the grid of UV disk models by Wade & Hubeny (1998), and found that AG Peg must have a high-mass transfer rate if one ascribes all of the FUV flux to accretion luminosity. We tried the highest mass accretion rate models of Wade & Hubeny (1998) with yr⁻¹, M_wd = 0.55 M_⊙, and M_wd = 0.80 M_⊙, all with an inclination i = 41°. However, all these disk models yielded derived distances much shorter, and did not provide enough flux in the short wavelength range of FUSE (λ < 960 Å). When linearly scaling these disk models to 800 pc, we found that the mass accretion rate should be of the order of yr⁻¹. Consequently, we generated accretion disk models for larger mass accretion rates using TLUSTY and SYNSPEC (Hubeny 1988; Hubeny & Lanz 1995) as described in the previous section.

We chose a WD mass M_wd = 0.7 M_⊙, with a radius R_wd = 8500 km, and an inclination i = 50°. The disk extends to where its temperature T_disk drops to ≈3500 K. For a given mass accretion rate the disk model scales to a given distance.

To match a distance of 800 pc we obtained that the mass accretion rate has to be yr⁻¹. This disk model has an inner radius of 8500 km and an outer radius of 1 million km, and is presented in Figure 3. The fit in the shorter wavelength of FUSE is not as good as for the single hot WD model fit (Figure 2), as the model is deficient in flux.

Zoom In Zoom Out Reset image size

We note that even if we increase the mass accretion to yr⁻¹, the fit in the short wavelength range is still not as good as the single WD component. This model (shown in Figure 4) leads to a distance of 1320 pc.

Zoom In Zoom Out Reset image size

We also tried combined WD + accretion disk models where the WD contributes a significant fraction of the flux, and the result was intermediate between the single WD models and the single disk models. Namely, the WD+disk models were improved over the single disk models, but did not provide a fit as good as the single WD models.

On the basis of fitting very hot NLTE white dwarf photosphere models to the FUSE spectrum of AG Peg in quiescence and comparing with the best-fitting high (10⁻⁷–10⁻⁶ M_⊙ yr⁻¹) disk models, we report evidence that a very hot white dwarf with a temperature T_wd = 150,000 K dominates the entire FUSE wavelength range while an accretion disk with high gives a poorer fit in the short wavelengths (λ < 960 Å). This hot WD model has a gravity log(g) ∼ 6.5, and the distance of 800 pc gives a scaled radius R_wd ∼ 43 × 10⁹ cm (∼0.06 R_⊙). The addition of an accretion disk only degrades the hot WD fit and implies that the WD is possibly overshining the disk.

It is interesting to compare our results on the hot component of AG Peg using FUSE spectra with a recent multiwavelength analysis by Skopal et al. (2017). They computed the peak luminosity of the June 2015 Z-And-like outburst of AG Peg to be (2–11) × 10³⁷ erg s⁻¹ (for a distance of 0.8 kpc). They estimated that the white dwarf had to be accreting at ∼3 × 10⁻⁷ M_⊙ yr⁻¹, which they claimed exceeds the stable burning limit by which one assumes is the steady-state limit in which the accreted material is burned at the same rate that it accretes.

Our best-fitting NLTE high-gravity photosphere models were fit best with T_wd = 150,000 K, which is in agreement with the modeling of the UV/Optical continuum in quiescence from 1993 to 1996 (Skopal et al. 2017), and considerably lower than the hot component of AG Peg in outburst. The luminosity of our best-fit model also agrees with the quiescent luminosity estimate of Skopal et al. (2017). During the outburst, the effective radius of the hot component, R_h, varies between ∼0.09 R_⊙ and ∼0.15 R_⊙. The accretion rate they found, on average, was a factor of 2–3 larger during the outburst than during quiescence.

Note that Mürset et al. (1991) obtained temperatures and luminosities of the hot component in symbiotic novae using the modified Zanstra method. For AG Peg they obtained a hot component temperature between 95,000 and 100,000 K during the decline phase. However, we note that Mürset et al. (1991) adopted a shorter distance (650 pc versus 800 pc) as well as a smaller E(B-V) = 0.05 (versus 0.10), which leads to a lower effective temperature for the hot component (because the UV continuum is less steep and has a much lower luminosity than that derived by Kenyon et al. 1993, 2001). Other attempts assumed black bodies to fit the SED of AG Peg encompassing the optical all the way to the IUE SWP range. These efforts obtained temperatures significantly cooler than the temperature of 150,000 K that we obtained from the best-fitting NLTE model atmosphere to the FUSE spectrum.

It is interesting to note that our method of deriving the temperature of the hot component gives a temperature () similar to the temperature obtained from an analysis of the optical H i and He ii lines (see Figure 5 in Kenyon et al. 2001), but much higher than the temperature derived from an analysis of the UV lines or the UV continuum. The luminosity we derive here is significantly larger than that derived by these three methods (Kenyon et al. 2001). In the present work we dereddened the FUSE spectrum using the extinction law given by Fitzpatrick & Massa (2007), while Kenyon et al. (2001) adopted Mathis' (1990) extinction curve (which itself is the Cardelli et al. 1989 extinction law). In the FUV region the difference between Cardelli et al. (1989) and Fitzpatrick & Massa (2007) extinction laws increases with decreasing wavelength and can gives different results (see, e.g., Selvelli & Gilmozzi 2013). It is possible that the difference in the extinction laws in the shortest wavelengths of FUSE increased the flux (in the dereddened spectrum) to better match a 150,000 K model, because the difference between the 100,000 K model and the 150,000 K model is very small. A temperature closer to 100,000 K gives a luminosity in better agreement with the quiescent luminosity.

It is also possible that the observed difference in temperature is due to a real change in the WD surface temperature. The short wavelength range of the FUSE spacecraft down to the Lyman limit probes the radiation from the innermost disk/boundary layer/white dwarf, which likely accounts for the higher temperature that we obtain. No previous attempts to derive a temperature for the hot component in AG Peg covered a wavelength range down to the Lyman Limit.

Overall, our result suggests the possibility that the system is dominated by a very hot, low gravity white dwarf. This is similar to our analysis of the FUSE spectrum of the S-Type symbiotic star RW Hydrae (Sion et al. 2017), which indicates that it also contains a very hot, bare white dwarf with no evidence of an accretion disk. An important question now is whether the Z-And-type outbursts are due to bursts of accretion from the M3/4III companion or if they are also thermonclear powered like the 1850 nova outburst.

On the other hand, the actual structure of an accretion disk surrounding a hot stellar component in symbiotic binary is poorly understood compared to catalcysmic variables. First, the scale of a symbiotic binary relative to a cataclysmic variable is vastly larger and second, the mass transfer occurs via a red giant donor that may or may not fill its Roche lobe. This points to the possibility that the disk models might be different than the standard disk model. 3D hydrodynamic model simulations of symbiotic binaries (de Val-Borro et al. 2009, 2017) reveal that a disk-like structure resembling a steady-state disk does appear to form, but the thin disk approximation may break down in symbiotics. Given the uncertainty in what a true symbiotic accretion disk would look like, and that a very hot low gravity degenerate star dominates the FUV flux, the presence of a steady-state (standard) accretion disk cannot be summarily ruled out or denied.

However, we note that for the hot WD to contribute a significant fraction of the total flux to the FUSE spectrum, in addition to that of an accretion disk, its emitting photosphere radius has to be inflated (Section 3.1). With a radius of the order of R_wd ∼ 0.05 R_⊙, rather than R_wd ∼ 0.01 R_⊙, the accretion disk temperature drops significantly (Godon et al. 2017). For example, at a mass accretion rate of yr⁻¹ the peak temperature in the disk reaches 103,000 K if R_wd = 8500 km (which is the radius of a ∼30,000 K WD of mass 0.7 M_⊙). However, for a WD radius of R_wd = 35,000 km ≈ 0.05 R_⊙, the peak temperature of such a disk drops to 36,500 K. At a mass accretion rate of yr⁻¹, the disk peak temperature drops from ∼171,000 to ∼65,000 K. In both cases ( yr⁻¹ and yr⁻¹) the disk will contribute little flux to the FUSE spectrum if R_wd = 0.05 R_⊙. As the accretion luminosity decreases like ∝ 1/R_wd, so does the disk luminosity, and as long as the WD radius remains inflated it will dominate the FUV spectrum.

This work is supported by funding from the National Aeronautics and Space Administration (NASA) under grant number NNX17AF36G issued through the Office of Astrophysics and DATA Analysis Program (ADAP) to Villanova University. This study has also been supported in part by the National Science Centre, Poland, grant OPUS 2017/27/B/ST9/01940. P.G. is pleased to thank William (Bill) P. Blair at the Henry Augustus Rowland Department of Physics and Astronomy at the Johns Hopkins University, Baltimore, Maryland, USA, for his kind hospitality. We made use of online data from the AAVSO and we are thankful to the AAVSO and its members worldwide for their constant monitoring of CVs and for making their data public.

Facility: FUSE. -

Software: IRAF (v2.16.1, Tody (1993)), Tlusty (v203) Synspec (v48) Rotin(v4) Disksyn (v7) (Hubeny & Lanz 2017a, 2017b, 2017c), PGPLOT (v5.2), Cygwin-X (Cygwin v1.7.16), xmgrace (Grace v2), XV (v3.10).