Astron. Astrophys. 326, 528-536 (1997)

4. Discussion

4.1. The nature of the resolved blue stars

The data, as displayed in the CMD of Fig. 2, have been compared with the isochrones of Bertelli et al. (1994) for the initial chemical composition [Z = 0.02, Y = 0.28]. The absolute visual magnitude and effective temperature at different steps along these isochrones have been transformed in the m₁₇₅ vs. m m₁₇₅m₃₄₂ plane, using the synphot synthetic photometry package and the stellar atmosphere models of Kurucz (1979). A distance modulus of 27.47 and a foreground extinction with E(BV) = 0.05 have been adopted (McMillan et al. 1994). Specifically, the isochrones for the ages = 6.6, 7.0, 7.2, 7.4, 7.6 and 7.8 years have been superposed on the data of Fig. 2. For a more familiar comparison, typical BOV and B5V stars would have m₁₇₅m₃₄₂ colors of 1.7 and 1.15 respectively (along the = 6.6 isochrone).

Most of the blue stars detected and plotted in the CMD lie below the isochrone = 7.2 and therefore appear as resulting from a recent star formation episode that, unless the IMF is truncated at high masses, would have ended 1.5 10⁷ years ago. By comparison with the masses given along the isochrones, we determine that a truncation at 15 M would be necessary to mimic the observed CMD with on-going star formation and this possibility is discarded. There is, however, a difficulty with the recent star formation interpretation in the sense that the stars are expected to be more concentrated on the blue part of the isochrones: for the ages under consideration, the red part of the isochrones (color ) correspond to stars that are both massive (typically 13 for = 7.2) and in a relatively narrow range of mass; two reasons for a smaller proportion of stars along the red part than along the blue part of the isochrones. There are several possible explanations, including an unusual IMF, the effects of dust extinction, photometric errors and the presence of another type of blue stars than young stars.

For one, a dust extinction with a color excess E(BV) = 0.1, as may be possible given the presence of dust patches, would move a star in the CMD by 0.23 mag to the red (the calculations were made with synphot assuming a galactic reddening law). Such a shift would increase the photometric dispersion that we may consider as illustrated by the data at the left of the main sequence but does not provide a satisfying explanation. We have explored the possibility of a systematic error which would affect the photometry in the F342W image. The crowding is more severe and the background higher in this latter image than in the F175W image. The trend in the CMD is not found to be changed by visually selecting only high quality data.

Fig. 7. Same as Fig. 2. The solid lines show the P-AGB evolutionary tracks of Vassiliadis & Wood for the core masses 0.900, 0.677, 0.569 M (from top to bottom). The discontinuity in two of these tracks is due to the use of black body curves in place of stellar atmosphere models at high effective temperatures.

A last possibility is that we have resolved Post-Asymptotic Giant Branch (P-AGB) stars. The P-AGB stars are known to be the most UV luminous of the various hot and low-mass stars that can be anticipated in an old population (e.g. Greggio & Renzini 1990, Dorman et al. 1995). The stellar population responsible for the bulk of the visible light of the bulge is old enough to include P-AGB stars since a number of planetary nebulae have indeed been identified by McMillan et al. (1994). We have therefore superposed the recent evolutionary tracks for P-AGB stars of Vassiliadis & Wood (1994) on the data in Fig. 7. Specifically we have used the H-burning PNN evolutionary models for core masses of 0.569, 0.677 and 0.900 M and metallicity Z = 0.016. For transporting the tracks into the m₁₇₅ vs. m₁₇₅m₃₄₂ plane, we have proceeded as previously, except that we have used bolometric corrections since we started from bolometric luminosities, and black-body curves when the effective temperatures are larger than encompassed by Kurucz's models. Fig. 7 shows that the blue stars may be P-AGB stars and not young stars. However, an interpretation exclusively in terms of P-AGB stars has difficulty in the sense that the distribution and relative proportion of the stars in the color-magnitude diagram do not match the time evolution pattern on the P-AGB tracks. First, the stars appear concentrated on the red part of the evolutionary tracks while time evolution is slower at high temperature; it has been verified that the lack of very blue objects does not result from a selection effect (the F342W frame is deep enough). Second, the time evolution strongly decreases when the stellar core-mass decreases (i.e. from upper to lower tracks) and would give a lower proportion of bright stars. A related argument is that a number of the resolved stars are too bright to be P-AGB stars.

It is also interesting to attempt a direct comparison with the number of P-AGB stars detected in an UV image of the bulge of M31 (King et al. 1992). A limit of m₁₇₅ = 22.5 in NGC 5102 is equivalent to a magnitude of 19.75 in M31 (adopting = 24.25 and a color excess E(BV) = 0.11 for the latter galaxy). The 30 stars brighter than this latter limit in M31 would imply 10 P-AGB stars in NGC 5102 after accounting for the respective bolometric luminosity of the two galaxies (i.e. the V luminosity assuming similar bolometric correction), in the fields of view observed in the ultraviolet. This number confirms the fact that most of the resolved stars are not P-AGB stars but it should be kept in mind that the comparison has many uncertainties since the origin of the UV flux from old population is not well understood and governed by metallicity effects.

In conclusion, and although the observed CMD is not fully understood, the resolved blue stars are most likely young stars and the last generation of a star formation episode that ended 1.5 10⁷ years ago. It is somewhat ironic to see that the resolution of stars and the building of a CMD do not significantly improve the reliability of the interpretation over the previous arguments based on the integrated light, such as blue color, color gradient in the center and A-type Balmer absorption-lines in the spectrum (Faber et al. unpublished quoted by Burstein et al. 1988).

In the context of the favoured interpretation in terms of recently formed stars, we have examined how the determination of 1.5 10⁷ years as the end of the star formation episode may be affected by the uncertainties of the measurements and the parameters used. A metallicity less than the solar abundance adopted, as suggested from the Mg₂ index of NGC 5102 (Burstein et al. 1988), would change the isochrones by a quantity smaller than the uncertainty of the determination itself ( 0.3 10⁷ years). This same uncertainty is also found larger than the difference between the two sets of isochrones (at specifically log t = 7.2) currently used for modeling of stellar populations (those of Bertelli et al. and those derived from the evolutionary tracks of Schaller et al. 1992). Last, the impact of a change of distance modulus (+0.16, 0.25 is the uncertainty quoted by McMillan et al. 1994) or of a systematic photometric error (about 0.2 mag) are readily seen in Fig. 2: again they stay within the accuracy of the determination of the end of the star formation episode.

A rough evaluation of the recent star formation rate is possible by counting the number of stars between two isochrones. We count 29 stars brighter than m₁₇₅ = 22 between the log t = 7.4 and 7.2 isochrones. After a rough correction for completeness in our central domain, we get a number of 32. According to the distribution of masses along these isochrones (Bertelli et al. 1994), these stars have masses larger than 8 M . According to a Salpeter IMF, 1 M formed corresponds to 7.4 10^-3 star more massive than 8 M . From our count of 32 stars over a period of about 9 Myr we therefore derive a star formation rate of M yr^-1 in the FOC field of view.

4.2. The star formation history

Because of the limit of detection and completeness effects, the resolved stars and the CMD cannot be used for determining when the star formation episode began? To address this question we have to rely on the integrated light, using specifically the IUE spectrum and the UVV color.

Fig. 8. Comparison of the IUE spectrum (thick solid line) (Burstein et al. 1988) with a synthetic spectrum (dashed line) resulting from the combination of an old population of 13 Gyr, an intermediate-age burst of 500 Myr, and a young burst of 30 Myr. Each of these components, listed in order of increasing UV flux at 2000 Å, is displayed as thin line.

We have a first qualitative indication that a large fraction of the star formation is not recent from the blueish m m₁₇₅m₃₄₂ integrated light in our FOC images (as discussed above Sect. 3.2) as compared to the color of resolved stars. It is also clear from evolutionary models (e.g. Fioc & Rocca-Volmerange 1997), that the star formation episode started a relatively long time ago. In these models, the combination of a relatively young burst to a conventional old population would never reproduce the relatively flat UV spectrum observed with IUE (Burstein et al. 1988). Unless we have special circumstances in NGC 5102 (extinction law, etc), we need to add the contribution of a spectrum declining in the UV as produced by an intermediate-age burst (500-1000 Myr). Fig. 8 shows a possible fit of the observed UV spectrum and the V magnitude in the IUE equivalent aperture with an intermediate-age burst of 800 Myr, an old population of 13 Gyr, and a young burst of 30 Myr (accounting for the young stars that we resolve). The spectral evolution models are from Fioc & Rocca-Volmerange (1997). In the example displayed in Fig. 8 the intermediate-age burst contributes to 60% of the light at 3200 Å but slightly different contributions and ages (in the range 500-800 Myr) are possible. The old component does not contribute significantly below 3000 Å  indicating that our conclusions do not depend on the assumptions adopted for the UV spectrum and the star formation history of the old population.

A model with two instantaneous bursts has been used for simplicity but the intermediate-age burst has not necessarily ended abruptly and may naturally be seen as the beginning of a continuous star formation episode of which we are resolving the last generation of stars. A continuous formation would have the advantage to reduce the UV excess produced by the young burst at the shortest UV wavelengths (Fig. 8). In a continuous formation scenario, the star formation rate would be declining as a function of time if one wants the first generation of stars to be able to balance the last generation in terms of UV emission. A continuous exponentially decaying burst would imply a time constant of 70 Myr. Our result is close to the conclusion reached by Bica (1988) that the bulk of the star burst was formed between the ages 300 to 500 Myr.

4.3. Comparison with NGC 3115

NGC 5102 has been compared with another S0 galaxy, NGC 3115, more typical of early-type galaxies than NGC 5102 in the sense that it is gas poor, exhibits a regular UV spectrum and fits well in the (1550V) vs. Mg₂ absorption-line index relation for elliptical galaxies (Burstein et al. 1988, Bertola et al. 1993). The UV spectra of these two galaxies were also compared by Rocca-Volmerange & Guiderdoni (1987). A FOC exposure of NGC 3115 with the F175W filter and the same exposure time as the NGC 5102 image does not resolve any star. With a distance modulus of 30.21 (Elson 1997) and a foreground extinction A_B = 0.10 (Burstein & Heiles 1984), the isochrones and evolutionary paths for NGC 3115 are basically those of Figs. 2 and 7 translated down by 2.5 mag along the m₁₇₅ axis (ignoring the small effect on the color due to the assumption of a lower extinction). The following results stand out for NGC 3115: (i) stars earlier than approximately B0V are ruled out. If we assume a conventional IMF any star formation more recent than 15 Myr is excluded. (ii) the most massive P-AGB stars are below the detection threshold.

In conclusion, the constraints that UV imaging with the FOC can place on residual star formation in elliptical-type population are weak at the distance of most elliptical galaxies.

4.4. Interpretation

The star formation phenomenon in the central region of the S0 galaxy NGC 5102 appears different from what we are used to in regular sites of star formation, in the sense that the young stars approximately follow the overall light distribution, do not seem confined to a disk and are not significantly clumped into clusters. These features as well as the existence of an UV gradient close to the center are naturally explained by the role of the galaxy potential well. Interestingly, the high angular resolution of the HST shows that, in contrast to the conclusion from ground-based observations (Pritchet 1979), the nucleus does not contribute to the UV gradient confirming that we have a purely stellar population phenomenon. The star formation also provides a natural explanation, at the expected location in the bulge, for the ring of ionized gas discovered by McMillan et al. (1994). According to the models of Leitherer & Heckman (1995), the total kinematic energy of 10⁵² ergs that McMillan et al. have calculated for the supershell would be produced over a time scale of 15 Myr by the last 2 Myr of the current star formation rate of M yr^-1 (except for an IMF poor in massive stars).

Pritchet (1979), van Woerden et al. (1993) have searched for external factors that may have been able to supply gas and/or to trigger star formation in NGC 5102; they found no evidence for any possible encounter with another galaxy within the last few 10⁹ years. With the mounting evidence for a multiplicity of star formation history in early-type galaxies (Bressan et al. 1996) and especially the S0 galaxies (Fisher et al. 1996), the need for an external cause is less compelling. NGC 5102 would then only appear as an extreme example of a current phenomenon with perhaps the youth of the burst and the rather short distance favouring the detection of the residual star formation. This is also consistent with the fact that the star formation takes place in the center where the gas is expected to concentrate. The central depression in the HI distribution would be the witness of the star burst started 500-800 Myr ago that would have stopped because of gas exhaustion. The HI surface density of 0.9 M pc^-2 (van Woerden et al. 1993) is indeed at the lowest extremity of the range of threshold values reported by Kennicutt (1989). The very central peak of HI may indicate a new replenishment of gas since the end of the burst. This is highly speculative since the origin of the gas itself (not to mention the time scale of the processes involved) is not understood. In addition, even if the gas shed by stellar evolution is a possibility that could be invoked for the origin of the gas, it is not clear why the gas should be removed and/or the star formation inhibited in a number of early-type galaxies. In this context it is interesting to note the evidence of multiple episodes of star formation in local group dwarf ellipticals (Ferguson & Binggeli 1994) and the larger fraction of color gradient (bluer toward the center) among early-type dwarf galaxies (Vader et al. 1988, Chaboyer 1994) indicating a possible increase of the occurrence of relatively recent star formation in galaxies of low-luminosity. Although not a dwarf galaxy itself, NGC 5102 has a low luminosity (M).

Finally our observations illustrate the general difficulty of understanding the nature of the stars responsible for the ultraviolet flux of elliptical galaxies, and, more specifically, of how any residual star formation may be distinguished from extreme horizontal branch stars and their evolutionary progeny. The comparison with NGC 3115 shows that only the most massive stars (implying only very recent bursts or on-going star formation) can be resolved by the HST provided the ellipticals are not too distant. Even though blue stars are resolved, it may be difficult to determine individually the nature of these stars with UV photometry alone as shown by the example of NGC 5102. It is worth noting that other features or parameters such as blue color gradient or integrated UVvisible color are not free of ambiguities as indicators of star formation: globular clusters also have central UV color gradient (Djorgovski & Piotto 1992) and NGC 5102, although distinct from elliptical galaxies in the (15V) vs. Mg₂ plane (Dorman et al. 1995), is at the location of the globular clusters in this plane.

© European Southern Observatory (ESO) 1997

Online publication: October 15, 1997
helpdesk.link@springer.de