4. The nuclear starburst of NGC 3593
To date there has been no study to estimate the age of the stellar population in disk II, especially relative to disk I . If disk II consists of stars formed later from the accreted gas, then the most recent star formation would be occurring in the CND. On the basis of the observed optical emission line ratios, Ho et al. (1997) classified the nucleus of NGC 3593 as a HII starburst. That NGC 3593 hosts a starburst is also suggested by the high (total) star formation rate (SFR) measured using far-infrared (FIR) data [SFR(IRAS) 3 yr-1; e.g., Wilson et al. 1991]. However due to the poor spatial resolution of IRAS (4´), this global estimate is only an upper limit for the CND alone. In order to explore the link between the starburst and the gaseous CND, we must focus on the high-resolution observations which trace directly the birth of massive stars.
Several authors have observed the emission of the H+NII nebular lines in the disk of NGC 3593 (Pogge & Eskridge 1993; Corsini et al. 1998). Most of the flux arises from a compact disk whose size is significantly smaller than the one characterizing the CND seen in CO. Two emission peaks lie along the major axis at 7-8 " (see Fig. 3a), which implies D() 15-16". An arc of emission connects these two maxima on the northern side. In contrast, there is no southern counterpart of this arc. Hardly any H emission is detected towards the central offset either. The marked asymmetry of the disk and the different values of the nuclear disk diameters seen in CO and H might result from dust obscuration of the H image. In this case of high extinction, the SFR inferred from H would be underestimated. Indeed, the SFR measured from H (Kennicutt 1983) is a factor of 7 lower [SFR(H0.4-0.5yr-1] than that measured from the IRAS data.
Contrary to the distribution of H shown in Fig. 3a, the Pa image (Fig. 4a) shows similar sizes for the molecular and ionized gas rings (D 20"). The image also reveals a previously unseen peak of emission towards the center. Although the bulk of massive star formation arises in the CND, some complexes of HII regions are found along the one-arm spiral. Pa emission peaks at (,)=(-22 ",-3 "), (1 ",-5 ") and (-14 ",-3 ") witness the birth of massive stars.
We have derived a visual extinction map from the hydrogen recombination lines ratios, after rebinning the Pa image to the lower ground-based resolution, correcting for the different point-spread functions, and registration. The derivation was performed in the usual way using a standard extinction curve (Savage & Mathis 1979), and assuming case B recombination (10000 K) which gives an intrinsic ratio of H/Pa (Osterbrock 1989). This recombination-derived extinction map, shown in Fig. 4b, gives a mean 0.36 mag for the entire CND and the arm, and 0.70 for the CND alone. Locally, however, this map gives AV values of 3 mag. With the commonly-used gas-to-dust ratio of N(H2)/AV = 1.0 1021 cm- 2 mag-1 (Bohlin et al. 1978), we obtain a molecular gas column density in the CND of 0.7 cm-2.
We have also estimated the extinction in the CND from the color map. Typical colors in the ring and CND are 1.3-1.4, which are rather redder than normal bulge colors. If we assume that is tracing extinction (i.e., that there is no hot dust in emission that would be detectable in the K band), and with a normal bulge color of (Moriondo et al. 1998a), the mean (assuming foreground screen) in the ring/CND is found to be 1.4 mag. This is roughly twice the value of that derived from the recombination lines.
The average extinction in the CND can also be derived from the CO integrated intensity map of Fig. 2a. Using a conversion factor /I(CO) = 2.3cm-2K-1km-1s (Strong et al. 1988), we obtain a mean 20-25 mag for the CND alone; this value is a factor of 15-20 higher than that derived from the recombination lines. Moreover, we can estimate an average extinction from the 100 µm emission measured by IRAS. With a conver- sion factor for low-latitude molecular clouds [I(100 µm)/AV = 6.3 MJy sr- 1 mag-1, Laureijs et al. 1987], we find a lower limit 1 to the extinction of 5.5 mag. At face value, such high values of derived from the CO map and from the 100 µm flux imply that the recombination lines are optically thick, and that we are detecting only a fraction of the emission from HII regions involved in the starburst. A scenario where recombination lines are optically thick can be discarded, however, because the measured P/H is close to the intrinsic value. Indeed, there are more plausible reasons which could explain the discrepancy between the extinction derived from the recombination lines and CO/IRAS measurements, and we investigate them in the following.
First, there is strong evidence that the CO-to- column density ratio X in central regions of galaxies may be lower than the standard Galactic value. The value of X in the Galactic center region, for example, is found to be a factor of 3-10 lower than that for molecular clouds in the inner disk (Sodroski et al. 1995). In the inner parts of M 51 and in NGC 891, X is also found to be between 3 and 4 times lower than the standard value, according to the calculations of line transfer models and measurements of the cold dust emission (Adler et al. 1992; García-Burillo et al. 1992; Guélin et al. 1993; Guélin et al. 1995). A similar scenario holds in the nuclear regions of some starburst and non-starburst galaxies (Mauersberger et al. 1996; Solomon et al. 1997). Possible underlying reasons for this discrepancy include: first, the presence of a non negligible mass percentage (1/2) of molecular gas contained in a diffuse phase (5 102-103cm-3) where 12CO lines are still optically thick; also, the influence of a strong central stellar potential (Mauersberger et al. 1996); and, finally, the influence of metallicity gradients (Dahmen et al 1998). Low optical depth effects in NGC 3593 may also play a role in lowering the conversion factor. However, such an effect should not reduce the factor by more than 5, because, judging from the 2-1/1-0 12CO ratio (0.7; see Wiklind & Henkel 1992), the CO emission must still be optically thick.
Second, the IRAS spectral energy distribution (SED) may not be amenable to straightforward extinction calculations. In later types, the "starburst" component, associated with thermal emission from large-grain dust, usually dominates the far-infrared (FIR) luminosity (Persson & Helou 1987). In early types, a large fraction (86% in the mean) of is typically due to quiescent cirrus (Sauvage & Thuan 1992), and for this reason such systems have a high FIR-to-H luminosity ratio. The observed FIR-to-H luminosity ratio is 375, roughly a 2 deviation higher than the mean of 162 in typical Sa's (Sauvage & Thuan 1992); when we correct for 0.7 mag of visual extinction (in the CND), the /H ratio becomes 260, still slightly higher than the 160+1 upper limit. Since cirrus clouds may have conversion factors X which are substantially lower than galactic molecular clouds (Heithausen & Mebold 1989), the extinction derived from the 100 µm flux may be in error. However, in addition to the significant contribution from cirrus clouds, the high /H suggests that NGC 3593 has an extra component. The 60/100 µm flux ratio and FIR-to-blue luminosity ratio is typical of infrared-selected galaxies which tend to have higher star-formation rates than optically-selected ones, and indeed the SFR inferred from IRAS data is relatively high. It is likely therefore that this extra FIR component in NGC 3593 is associated with the starburst in the CND.
We can check the consistency of the IRAS data, as well as provide an indirect confirmation of the nature of the FIR emission. While there is no significant continuum emission at 2.6 mm (see Sect. 2), NGC 3593 has been observed at 1.1 mm, 800, and 450 µm by Fich & Hodge (1993). They find a dust temperature of 40 K, consistent with the temperature derived from IRAS data alone, and with our non-detection at 2.6 mm. Following Thronson & Telesco (1986), we can calculate the dust mass responsible for this emission, and find that . Together with the total H2 mass derived from the CO map (Sect. 3), this gives a 750, in good agreement with the value of 700 found for giant molecular clouds in the Galaxy (e.g., Thronson & Telesco 1986), but much higher than the canonical value of 100. Since infrared measurements are very likely detecting all the dust, such a high ratio is probably yet another indication that X is a factor of 3-7 too high. Assuming a conservative correction factor of 1/3 for X, this implies AV5 mag from CO, consistent with the result from the IRAS data. This also leads to reduce molecular mass estimates made in Sect. 3 for the CND to Mgas1.5108 and for the entire molecular gas disk to Mgas3108.
To better assess what fraction of could be due to a starburst, we have calculated the bolometric luminosity of the massive stellar population (OB) responsible for the ionization of the hydrogen lines. Using the total Pa flux (3.15 erg s-1 cm-2), and the emission coefficients given in Osterbrock (1989), we can estimate the number of ionizing photons . With , and assuming a trend of luminosity with mass appropriate for high-mass stars (), we can then calculate the bolometric luminosity of the massive stars that ionize the gas. We have done this by integrating a massive stellar population from 10 to 63 , with a Salpeter Initial Mass Function (IMF), and using for massive stars given in Panagia (1973). The bolometric luminosity of the massive stars turns out to be 4.8, or about 1/11 of . This fraction remains unchanged even when the IMF is integrated down to 0.1 , since the luminosity is dominated by the high-mass stars. If we were to correct the optical emission-line luminosity for the 5 mag of visual extinction deduced from the IRAS 100 µm flux, the stellar bolometric luminosity would exceed ; if we were to correct the Pa emission, yet another gross inconsistency would arise since the observed line ratio H/Pa is very close to the intrinsic one. Therefore, the line emission that we observe in the CND cannot suffer from the same extinction that we infer globally from the IRAS data.
The last, and probably most convincing explanation for the "extinction discrepancy" is the probable invalidity of comparing circumnuclear data at different spatial resolutions. The IRAS data, with a resolution at 100 µm of 4´, are sampling the entire galactic disk in NGC 3593, while the CO map at 3-4 " is sampling the CND itself, but at a lower resolution than the recombination line images. The clumpy nature of the interstellar medium makes such comparisons difficult because dust and gas can be very localized even on small spatial scales. Higher-resolution data are needed to better understand the dust distribution and extinction in NGC 3593.
Although stellar luminosities inferred from recombination lines are similar to , it may be that much of the molecular gas is not involved in actively forming stars at the present epoch, but rather being channelled towards the ring. The mechanisms explaining the nuclear infall of gas and the onset of the starburst will be discussed in the following sections by studying the observed gas kinematics, with the additional input from numerical simulations.
© European Southern Observatory (ESO) 2000
Online publication: December 5, 2000