Astron. Astrophys. 322, 576-590 (1997)

5. Discussion

5.1. Quiescent X-ray emission

5.1.1. The X-ray spectrum and luminosity

Previous analyses of X-ray emission of symbiotic stars have been interpreted either with blackbody (Kenyon and Webbink 1984) or NLTE atmosphere (Jordan et al. 1994) models representing the hot component, or with bremsstrahlung emission from a hot, gaseous nebula (Kwok and Leahy 1984). Our ROSAT PSPC spectra of AG Dra show no hint for any hard X-ray emission. The soft spectrum is well fitted with a blackbody model. A thermal bremsstrahlung fit is not acceptable to this soft energy distribution. There is also no need for a second component as proposed by Anderson et al. (1981): a fit of a blackbody and thermal bremsstrahlung model with the bremsstrahlung temperature limited to values above 0.1 keV results in an unabsorbed flux ratio between blackbody and bremsstrahlung of 85000:1 in the 0.1-2.4 keV band. A hard X-ray component could be expected from an accretion disk both for radial and disk accretion (Livio 1988) at the relatively low accretion rate implied for this system. This absence of any hard emission also rules out the possibility of interpreting the soft spectrum as arising in the boundary layer. The most extreme ratios for soft to hard emission observed sofar in magnetic cataclysmic variables are of the order of 1000:1. An example of the blackbody fit is given in Fig. 4 using the observation of April 1993 (having the largest number of photons).

The simple model of explaining the soft X-ray emission by an accretion disk is ruled out not only due to the soft spectrum, but also due to the observed luminosity which is much higher (see below) than can be produced by an accretion disk around a compact dwarf and the extremely low accretion rate. If the X-ray spectrum were to explained by a standard accretion disk model (Shakura & Sunyaev 1973) then the necessary accretion rate would be of the order of 1.7 10^-7 (or 4.5 10^-15 /yr) for a 1 compact object. This is unreasonably small, basically due to the low effective temperature ( = 24 eV) of the X-ray emission coupled with a small mass compact object.

Using the blackbody fit parameters while was fixed at its galactic value, i.e. kT = 14.5 eV and a normalization parameter of 206 photons/cm² /s corresponding to observation 201044 (from experience with soft ROSAT spectra we know that this procedure helps to avoid overpredicting the flux using blackbody models), the unabsorbed bolometric luminosity of the hot component in AG Dra during quiescence (1990-1993) is (9.5 1.5) 10³⁶ (D/2.5 kpc)² erg/s (or equivalently 2500 400 (D/2.5 kpc)² ) with an uncertainty of a factor of a few due to the errors in the absorbing column and the temperature (see Fig. 5 for a visualization of the fact that the ROSAT measurement actually covers only the tail of the spectral energy distribution). The blackbody radius is derived to be = (4.1 1.5) 10⁹ cm (D/2.5 kpc). (Previous estimates arrived at much lower values due to the overestimate of the temperature.)

Fig. 5. Energy distribution of AG Dra during quiescence and outburst: The full line shows the quiescent spectrum, combined from a UV spectrum of April 9, 1993, and a ROSAT observation of April 15, 1993 (best fit blackbody extrapolation is shown on top of the absorption corrected data points). The optical B measurement is also from April 15, while the U measurement is from April 10, 1993. The dashed/dotted lines are the measured UV spectra during outburst of July 28 / Sep 14, 1995 together with the modelled blackbody spectra of lower temperature (11/9.5 eV) which reproduce the observed HRI countrates as measured with ROSAT on July 31 / Sep. 14, 1995.

This high luminosity during quiescence and the size of the emitting region comparable to a white dwarf radius suggests that the primary is a white dwarf in the state of surface hydrogen burning. Assuming that the bolometric luminosity of the hot component equals the nuclear burning luminosity, the burning rate is (3.2 0.5) 10^-8 (D/2.5 kpc)² /yr. Under the assumption of a steady state the same amount of matter should be accreted onto the white dwarf from the companion (via wind or an accretion disk, see below). Indeed, this burning rate lies within the stable-burning regime for a white dwarf of less than 0.6 (Iben and Tutukov 1989) consistent with the core mass-luminosity relation L/ 4.6 10⁴ ( / - 0.26) (Yungelson et al. 1996). From the orbital parameters (Mikolajewska et al. 1995, Smith et al. 1996) and an inclination less than about (see paragraph 5.1.2) the donor mass is estimated to be lower than 2 , in agreement with the estimated surface gravity of Smith et al. (1996).

5.1.2. Orbital flux modulation?

The observational coverage of AG Dra over 1992/1993 is extensive enough to also investigate possible temporal and spectral changes of the X-ray flux along the orbital phase. Using the ephemeris from Skopal (1994) the minimum phases of AG Dra occurred in November 1991 and June 1993.

The coincidence of the U minimum and the drop in X-ray flux in mid-1993 is surprising, suggesting the possible discovery of orbital modulation of the X-ray flux. Unfortunately, we have no full coverage of the orbital period, and thus the duration of the flux depression cannot be determined. But even with the available data this duration is a matter of concern: The X-ray intensity decrease occurs very slowly over a time interval of two months, suggesting a total duration of at least 4 months if this modulation is symmetric in shape.

The radial velocity data demonstrate that during the U minima the cool companion lies in front of the hot component (cf. Mikolajewska et al. 1995). Depending on the luminosity class, the maximum duration for an eclipse of the WD by the giant (bright giant) companion is 10 (24) days, i.e. much shorter than the observed time scale at X-rays. Thus, the drop in X-ray intensity cannot be due to a WD eclipse. As evidenced by near-simultaneous IUE spectra between April and June 1993, the far-UV continuum associated to the tail of the hot component's continuum emission is not occulted, supporting the previous argument. A similar conclusion was reached when considering the nearly complete lack of orbital variations of the short wavelength UV continuum (Mikolajewska et al. 1995). By inverting the arguments, the lack of an X-ray eclipse implies that the inclination should be less than () for a giant (bright giant) with 20 (70 ) radius.

Even in the reflection model of Formiggini & Leibowitz (1990), in which the eclipse of the hot component is short and its depth is expected to be considerably larger in the UV (and possibly in the soft X-ray band) than in the optical, it is difficult to imagine that a possible UV eclipse has been gone undetected.

Alternatively to an eclipse interpretation of the X-ray intensity drop just before the 1993 U band minimum, one could think of a pre-outburst which was missed in the optical region except for the marginal flux increase in the B band after JD = 244 9000. In this case the interpretation would be similar to that of the major X-ray intensity drops during the optical outbursts in 1994 and 1995 mass loss (see below). Independent of the actual behaviour around JD = 9100 it is interesting to note that there is a secure detection of a 2 mag U brightness jump coincident with a nearly 1 mag B brightness increase shortly after JD = 9200. A slight brightening by about 0.2-0.3 mag is also present in the AAVSO light curve around JD 9205-9215 (Mattei 1995).

5.1.3. Wind mass loss of the donor and accretion

Accepting the high luminosity during quiescence and consequently assuming that the hot component in the AG Dra system is in (or near to) the surface hydrogen burning regime, the companion has to supply the matter at the high rate of consumption by the hot component.

It is generally assumed (and supported by three-dimensional gas dynamical calculations of the accretion from an inhomogeneous medium) that the compact objects in symbiotic systems accrete matter from the wind of the companion according to the classical Bondi-Hoyle formula. The wind mass loss rate of the companion depends on several parameters. One of the important ones is the luminosity which governs the location with respect to the Linsky and Haisch dividing line. Accordingly, a bright giant (luminosity class II) is expected to have a considerably larger mass loss rate than a giant (luminosity class III).

As mentioned already earlier, the spectral classification and the luminosity class of the companion in the AG Dra binary system is not yet securely determined. According to the formula of Reimers (1975):

where R, L and M are the radius, luminosity and mass of the star in solar units, and a constant between 0.1 and 1, depending primarily on the initial stellar mass, the commonly used K3III classification would imply a mass loss rate of (0.2-7) 10^-10 /yr, i.e. a factor of 100 lower than the rate necessary for steady state burning.

However, there are two additional observational hints which might help resolve the discrepancy between the expected mass loss of the cool donor and the necessary rate for a steady state burning white dwarf: (1) The cool component of AG Dra might be brighter than an average solar-metallicity giant. A comparison of the luminosity functions of K giants with 4000 4400 in a low-metallicity versus solar-metallicity population has shown convincingly that low-metallicity K giants are nearly 2 mag brighter than the solar-metallicity giants (Smith et al. 1996). (2) Low-metallicity giants might have larger mass-loss rates than usual giants. From observed larger 12µm excess in symbiotics as compared to that in normal giants a larger mass loss of giants in symbiotics has been deduced (Kenyon 1988). The recent comparison of IR excesses of d-type symbiotics with that of CH and barium stars supports this evidence (Smith et al. 1996). Both of these observational hints argue for a mass loss of the cool component in AG Dra which is higher than the value derived from Reimer's formula. At present it seems premature, however, to conclude that the wind mass loss of the donor in AG Dra is large enough to supply the matter for a steady state surface burning on the white dwarf.

An additional problem with spherical accretion at very high rates is the earlier recognized fact (Nussbaumer & Vogel 1987) that the density of matter in the vicinity of the accretor is 2 10⁷ ( /10^-7 /yr) (V/10 km/s)³ ( / )^-2 cm^-3 (Yungelson et al. 1996), i.e. the soft X-rays of the burning accretor will be heavily absorbed.

Alternatively, one might consider the possibility that the donor overfills its Roche lobe and that the matter supply to the white dwarf occurs via an accretion disk. Several issues have to be considered: (1) Roche-lobe filling: In order to fill its Roche lobe, the donor has to be either rather massive (as compared to a usual K giant) or rather luminous. However, a donor mass larger than 2 seems difficult due to the high galactic latitude and proper motion as well as on evolutionary grounds. Similarly, a binary system with a bright giant implies a larger distance as compared to a giant companion. (2) Evidence for the existence of an accretion disk: Indeed, Garcia (1986) has raised the suspicion that AG Dra contains an accretion disk. Circumstantial evidence for this comes from predictions of Roche lobe overflow, rapid flickering on timescales of the order of minutes in the optical band and the observation of double-peaked Balmer emission lines. Robinson et al. (1994) have investigated in detail the double-peaked line profile of the Balmer lines in several symbiotic stars. In the case of AG Dra they find double-peaked emission lines only in one out of three observations which were three and one year apart, respectively. Using an inclination of i= as proposed by Garcia (1986) the double-peaked profile gives an acceptable fit for an accretion disk with an inner radius of 1.1 10⁸ cm and an outer radius of 1.3 10¹⁰ cm though the asymmetry may be explained also by self-absorption (Tomov & Tomova 1997). Recent high-resolution spectroscopy of AG Dra in the red wavelength region using AURELIE at OHP performed in December 1990 and January 1995 (i.e. during very different phases in the AG Dra orbit) revealed only a single peaked H profile (Rossi et al. 1996) and Dobrzycka et al. (1996) failed to find evidence for flickering. Both of these new observational results argue against a steady accretion disk in the AG Dra system.

5.2. Quiescent UV emission

UV observations show that the hot components of symbiotic stars are located in the same quarters of the Hertzsprung-Russell diagram as the central stars of planetary nebula (Mürset et al. 1991). Due to the large binary separation in symbiotic systems the present hot component (or evolved component) should have evolved nearly undisturbed through the red giant phase. However, the outermost layers of the white dwarf might be enriched in hydrogen rich material accreted from the cool companion.

Presently the far-UV radiation of the WD is ionizing a circumstellar nebula, mostly formed (or filled in) by the cool star wind. The UV spectroscopy suggests that the CNO composition of the nebula is nitrogen enriched (C/N=0.63, and (C+N)/O=0.43), which is typical of the composition of a metal poor giant atmosphere after CN cycle burning and a first dredge up phase (Schmidt & Nussbaumer 1993). We recall that recently a low metal abundance of the K-star photosphere of [Fe/H]=-1.3 was derived (Smith et al. 1996) in agreement with it being a halo object. The UV spectrum in quiescence is the usual in symbiotic systems: a continuum increasing toward the shortest wavelengths with strong narrow emission lines superimposed. The most prominent lines in the quiescence UV spectrum of AG Dra are, in order of decreasing intensity: HeII 1640 Å, CIV 1550 Å, NV 1240 Å, the blend of SiIV and OIV] at 1400 Å, NIV] 1486 Å and OIII] 1663 Å. The continuum becomes flatter longward approximately 2600 Å due to the contribution of the recombination continuum originated in the nebula. Although both continuum and emission lines have been found to be variable during quiescence (e.g. Mikolajewska et al. 1995), there is no clear relation with the orbital period of the system. The ratio intensity of the recombination HeII 1640Å line to the far-UV continuum at 1340 Å suggests a Zanstra temperature of around 1.0-1.1 10⁵ K during quiescence (see Fig. 3).

The large X-ray luminosity together with its soft spectrum allows to understand the large flux from HeII, most notably 4686 Å and 1640 Å, at quiescence (which could not be explained in earlier models; see e.g. Kenyon & Webbink 1984) as being due to X-ray ionization.

5.3. X-ray emission during the optical outbursts

5.3.1. Previously proposed models

Three different basic mechanisms have been proposed to explain the drastic intensity changes of symbiotic stars during the outburst events: (1) Thermonuclear runaways on the surface of the hot component (WD) after the accreted envelope has reached a critical mass (Tutukov & Yungelson 1976, Paczynski & Rudak 1980). The characteristic features are considerable changes in effective temperature at constant bolometric luminosity, and the appearance of an A-F supergiant spectrum thought to be produced by the expanding WD shell. (2) Instabilities in an accretion disk after an increase of mass transfer from the companion (Bath & Pringle 1982, Duschl 1986). (3) Ionization changes of the HII region (from density-bounded to radiation-bounded) around the hot component caused by an abrupt change in the mass loss rate of the companion (Nussbaumer & Vogel 1987, Mikolajewska & Kenyon 1992). In the case of AG Dra, all these three scenarios have problems with some observational facts. The thermonuclear runaway is rejected by the fact that the quiescent luminosity is already at a level which strongly suggests burning before the outbursts. The disk instability scenario predicts substantial variation of the bolometric luminosity during the outbursts, for which no hints are available in AG Dra. The application of ionization changes to AG Dra is questionable because the nebula is apparently not in ionization equilibrium (Leibowitz & Formiggini 1992).

Specifically for AG Dra an additional scenario has been proposed, namely (4) the liberation of mechanical energy in the atmosphere of the companion (Leibowitz and Formiggini 1992).

5.3.2. Expanding and contracting white dwarf

Using our finding of an anticorrelation of the optical and X-ray intensity and the lack of considerable changes in the temperature of the hot component during the 1994/95 outburst of AG Dra, we propose the following rough scenario. (1) The white dwarf is already burning hydrogen stably on its surface before the optical outburst(s). (2) Increased mass transfer, possibly episodic, from the cool companion results in a slow expansion of the white dwarf. (3) The expansion is restricted either due to the finite excess mass accreted or by the wind driven mass loss from the expanding photosphere of the accretor. This wind possibly also suppresses further accretion onto the white dwarf. The photosphere is expected to get cooler with increasing radius. (4) The white dwarf is contracting back to its original state once the accretion rate drops to its pre-outburst level. Since the white dwarf is very sensitive to its boundary conditions, it is not expected to return into a steady state immediately (Paczynski & Rudak 1980). Instead, it might oscillate around the equilibrium state giving rise to secondary or even a sequence of smaller outbursts following the first one.

The scenario of an expanding white dwarf photosphere due to the increase in mass of the hydrogen-rich envelope has been proposed already by Sugimoto et al. (1979) on theoretical grounds without application to a specific source class. The expansion velocity was shown to be rather low (Fujimoto 1982)

where at was adopted, appropriate for a low mass white dwarf accreting at high rates (Fujimoto 1982). We note in passing that this value could be different at smaller radii, i.e. near white dwarf dimensions (see e.g. Neo et al. (1977)). In particular, depending on the location in the H-R diagram it would be around 2.3 on the high-luminosity plateau of a 0.6 white dwarf, but rising up to 7 around the high-temperature knee (Blöcker 1996), thus suggesting a non-linear expansion rate. If we assume that the luminosity remains constant during the expansion we can determine the expansion velocity simply by folding the corresponding temperature decrease by the response of the ROSAT detector. Fitting the countrate decrease of a factor of 3.5 within 23 days (from 0.14 cts/s on HJD 9929 to 0.04 cts/s on HJD 9952 corresponding to a concordant temperature decrease from 12.3 eV to 11.1 eV) we find that

for the input parameters (blackbody radius) corresponding to 2.5 kpc (note that the input parameters change with distance). The mass of the envelope is difficult to assess due to the non-stationarity in the AG Dra system. In thermal equilibrium at our derived mean temperature it should be larger than 5 10^-5 for a white dwarf with mass M 0.6 in the steady hydrogen-burning regime (Fujimoto 1982, Iben & Tutukov 1989). Inserting this in the above equation implies that the accretion rate necessary to trigger an expansion which is consistent with the X-ray measurements has to be (8.1/2.5/1.4) 10^-6 /yr for white dwarf masses of 0.4/0.5/0.6 . This is a factor of 40-250 larger than the quiescent burning/accretion rate. The corresponding mean expansion rate is 6.5 m/s. With a duration of the X-ray decline in 1995 of about 100 days (which is consistent with the 1994 rise time in the optical region) we find that during the 1995 outburst (and probably also during the 1994 outburst) the radius of the accretor roughly doubled while the temperature decreased by about 35% (from 14.5 eV to 9.5 eV for the two parameter fit). The countrate decrease modelled with the above parameters is shown as the dotted line in Fig. 3, and the extrapolation to the quiescent countrate level gives an expected onset of the expansion around HJD = 9902.

A completely different and independent estimate of the response time of a white dwarf gives a similar result, thus it seems quite reasonable that a white dwarf can indeed expand and contract on a timescale (see e.g. Kovetz & Prialnik 1994, Kato 1996) which is observed as optical outburst rise and fall time in AG Dra. The contraction timescale can be approximated by the duration of the mass-ejection phase (Livio 1992), similar to the application to the supersoft transient RX J0513.9-6951 (Southwell et al. 1996), where 51 / ( - ) days with being the ratio of white dwarf mass to the Chandrasekhar mass (Livio 1992). This relation gives a timescale of the order of 100 days (as observed) for a white dwarf mass of 0.5-0.6 , consistent with observations of the AG Dra optical outbursts.

Given the substantial accretion rate triggering the optical outbursts (which has to be supplied by the donor), mass loss via a wind from the cool companion seems to be too low to power the outbursts. Depending on the donor state in quiescence (where wind accretion is possible though we prefer Roche lobe filling; see paragraph 5.1.3.), two possibilities are conceivable: Either the companion fills its Roche lobe all the time, and the outbursts (i.e. the increased mass transfer) are triggered by fluctuations of the cool companion (e.g. radius), or a more massive and/or more luminous companion produces a strong enough wind to sustain the burning in the quiescent state without filling its Roche lobe and only occasionally overfills its Roche lobe thus triggering the outbursts. As noted above, a Roche lobe filling giant implies a distance larger than the adopted 2.5 kpc. While this imposes no problems with the interpretation of our UV and X-ray data (in fact a larger distance implies a larger intrinsic luminosity and thus shifts AG Dra even further into the stability burning region for even higher white dwarf masses), the distance dependent numbers derived here have to be adapted accordingly.

5.3.3. Wind from the accretor

If the accretion indeed is spherical (i.e. as wind from the donor), it may occasionally be suppressed by the wind from the accretor (Inaguchi et al. 1986). Hot stars with radiative envelopes are thought to suffer intense (radiation-driven) stellar winds at a rate of = L/( c) where L is the burning luminosity, is the escape velocity and c the speed of light. This wind becomes increasingly effective at larger radii. Thus, a radiatively driven wind from the WD (Prialnik, Shara & Shaviv 1978, Kato 1983a, b) might restrict the expansion of a RG-like envelope or an expelled shell or even might remove the envelope (Yungelson et al. 1995). An optically thick wind can be even more powerful by up to a factor of 10 (Kato and Hachisu 1994).

The wind of the hot white dwarf has typical velocities of a few hundreds km/s (Kato & Hachisu 1994). At these velocities it would take only several days until this white dwarf wind reaches the Roche lobe of the cool component, i.e. the wind of the cool component. This timescale is short enough to possibly cause the variations in the intensity of those lines which are thought to arise at the illuminated side of the wind zone of the cool component.

5.3.4. Alternative scenario

An alternative to increased mass transfer would be to invoke a mild He flash on the hydrogen burning white dwarf which again might cause the photosphere to expand. Previous investigations for mild flashes have yielded timescales of the order of 15-20 yrs even for the most massive white dwarfs (Paczynski 1975). Recent calculations of H and He flashes have shown that under certain circumstances flash ignition can be rather mild without leading to drastic explosive phenomena, and shorter than about 1 yr already for white dwarf masses below 1 (Kato 1996).

5.4. UV emission during the optical outbursts

There is an evident mismatch between the outburst UV spectra and the extrapolation toward shorter wavelengths of the blackbodies with temperatures inferred from the ROSAT data (Fig. 5). The most plausible explanation is the existence of an additional emission mechanism in the UV, namely recombination continuum in the nebula surrounding the hot star. While during the quiescent phase the contribution of the nebular continuum is not very large (although it is not negligible), it increases substantially during the outburst. As an example, the change in the strength of the Balmer jump from July 1994 to December 1995 can be seen in Fig. 4 of Viotti et al. (1996).

5.5. Comparison with other supersoft X-ray sources

Supersoft X-ray sources (SSS) are characterized by very soft X-ray radiation of high luminosity. The ROSAT spectra are well described by blackbody emission at a temperature of about kT 25-40 eV and a luminosity close to the Eddington limit (Greiner et al. 1991, Heise et al. 1994). After the discovery of supersoft X-ray sources with Einstein observations, the ROSAT satellite has discovered more than a dozen new SSS. Most of these have been observed in nearby galaxies (see Greiner (1996) for a recent compilation). Among the optically identified objects there are several different types of objects: close binaries like the prototype CAL 83 (Long et al. 1981), novae, planetary nebulae, and symbiotic systems.

In addition to AG Dra, two other X-ray luminous symbiotic systems with supersoft X-ray spectra are known (RR Tel, SMC 3), both being symbiotic novae. RR Tel was shown to exhibit a similar soft X-ray spectrum (Jordan et al. 1994) and luminosity (Mürset & Nussbaumer 1994). RR Tel is one of the only seven known symbiotic nova systems. It went into outburst in 1945 with a brightness increase of , and since then declined only slowly. SMC 3 (= RX J0048.4-7332) in the Small Magellanic Cloud was found to be supersoft in X-rays by Kahabka et al. (1994), and has been observed to be in optical outburst in 1981 (Morgan 1992). Simultaneous fitting of the ROSAT and UV data was possible only after inclusion of a wind mass loss from the hot component, and gives a temperature above 260 000 K (Jordan et al. 1996). Symbiotic novae are generally believed to be due to a thermonuclear outburst after the compact object has accreted enough material from the (wind of the) companion or an accretion disk.

The similarity in the quiescent X-ray properties of AG Dra to those of the symbiotic novae RR Tel and SMC 3 might support the speculation that AG Dra is a symbiotic nova in the post-outburst stage for which the turn-on is not documented. We have checked some early records such as the Bonner Durchmusterung, but always find AG Dra at the 10- intensity level. Thus, if AG Dra should be a symbiotic nova, its hypothetical turn on would have occurred before 1855. We note, however, that there are a number of observational differences between AG Dra and RR Tel which would be difficult to understand if AG Dra were a symbiotic nova.

If AG Dra is not a symbiotic nova, it would be the first wide-binary supersoft source (as opposed to the classical close-binary supersoft sources) the existence of which has been predicted recently (Di Stefano et al. 1997). These systems are believed to have donor companions more massive than the accreting white dwarf which makes the mass transfer unstable on a thermal timescale.

© European Southern Observatory (ESO) 1997

Online publication: June 5, 1998

helpdesk.link@springer.de