Astron. Astrophys. 345, 391-402 (1999)

4. Discussion

Before trying to interpret the trends shown above, it is useful to recall some basics on emission lines and photoionization models. This is done in the next subsection.

4.1. A reminder on emission line theory

Schematically, the (reddening corrected) intensity ratios of emission lines produced in photoionized regions are a function of the following parameters (see e.g. Stasinska 1998 and references therein):

1. the global metallicity, O/H;

2. abundance ratios, like N/O and S/O;

3. the mean effective temperature of the ionizing radiation field ;

4. the average ionization parameter , where is the number of ionizing photons of an HII region, is the Strömgren radius, and n is the gas density.

The effect of these parameters on the line ratios are the following.

1. Oxygen is the major coolant in HII regions. As O/H increases, the electron temperature decreases because of the increased cooling of the gas. The [OIII]/H ratio first increases (due to an abundance effect), then, for O/H greater than about half solar, decreases due to an electron temperature effect: the gas becomes so cool that the optical [OIII]5007 line gets difficult to excite. In this case, cooling occurs through the [OIII] far infrared lines at 52 µm and 88 µm. Qualitatively, the [OII]/H ratio behaves similarly to [OIII]/H, but it is not so reduced at high metallicities, because the cooling is less efficient in the region emitting [OII] than in the region emitting [OIII]. Line ratios like [NII]/H or [SII]/H are affected by O/H through the electron temperature: they get enhanced as O/H decreases.

2. An increase of N/O only enhances [NII]/[OII], but does not affect the other optical line ratios, since nitrogen contributes little to the cooling at the abundances expected for the general interstellar medium. Similarly, an increase of S/O enhances [SII]/[OII] and leaves the rest unchanged. However, one does not expected S/O to vary among galaxies, since sulfur and oxygen are produced in the same stars. Therefore, any change in [SII]/[OII] should rather be attributed to other causes.

3. The effects of are twofold. Firstly, acts on the ionization structure: the proportions of O⁺, N⁺ or S⁺ ions decrease with increasing . Secondly, it influences the thermal balance of the nebula: as increases, the energy gains become larger and the electron temperature rises, increasing the intensities of the forbidden lines with respect to H or H.

4. The effects of decreasing U are to reduce the average ionization, and to decrease ratios like O/O⁺. For very low U, such as found in the diffuse interstellar medium, a wide transition zone of low ionization develops, containing O⁰ and S⁺ ions and still hot enough to allow collisional excitation of optical forbidden lines. Therefore, the [SII]/H and the [SII]/[OII] ratios are higher than in HII regions having a large U.

If shocks are present, the emission line ratios are modified. The effect of shocks is to heat the gas to very high temperatures ( K). By recombination and free-free processes, this produces a hard radiation field which strongly heats the post shock region, producing an extended, warm, transition region, and enhancing the lines that are most sensitive to the temperature. As a result, [OII]/H, [NII]/H, and [SII]/H line ratios are enhanced with respect to pure photoionization nebulae.

4.2. Diagnostic diagrams

Line ratio diagrams (Baldwin et al. 1981, Veilleux & Osterbrock 1987) are helpful for the diagnostics of emission line objects. We now use such diagrams to compare the emission properties of galaxy integrated spectra with those of individual giant HII regions. Figs. 7 and 8 show the data from the integrated spectra of the normal spiral galaxies together with those of the giant HII regions (GHRs) observed by McCall et al. (1985) in several spiral galaxies and which compose the GHR sequence. This sequence is interpreted as being a sequence in metallicity (O/H) and also in mean effective temperature or in mean ionization parameter (or both) (McCall et al. 1985, Dopita & Evans 1986). Note that radial abundance gradients, which are a common feature in spiral galaxies (Zaritsky et al. 1994), complicate the interpretation of integrated spectra: the contribution of each annular region is weighted by the luminosities of the HII regions found there and by their spectral properties.

Fig. 7a-e. Classical emission line ratio diagnostic diagrams. The data for galaxies of our sample (circles) are plotted together with data for the giant HII regions observed by McCall et al. (1985) (dots). Upper values (of [OIII]) are indicated by arrows. The panels are: a  [OIII]/H versus [OII]/H, b  [OIII]/H versus [NII]/H, c  [NII]/H versus [OII]/H, d  [OIII]/H versus [SII]/H, and e  [SII]/H versus [OII]/H.

Fig. 8a-c. Forbidden line ratios as a function of the O/H indicator ([OII]+[OIII])/H. The data for galaxies of our sample (circles) are plotted together with data for the giant HII regions observed by McCall et al. (1985) (dots). Upper values (of [OIII]) are indicated by arrows. The panels are: a  [OIII]/[OII] versus ([OII]+[OIII])/H, b  [NII]/[OII] versus ([OII]+[OIII])/H, and c  [SII]/[OII] versus ([OII]+[OIII])/H.

Panel 7-a shows [OIII]/H versus [OII]/H. We see that the points corresponding to the integrated spectra of our standard galaxies that have [OIII] large enough to be observed (i.e. those with ST 0) are well inside the GHR sequence.

Panel 7-b displays [OIII]/H versus [NII]/H. Galaxies with measured [OIII] emission are disposed along the GHR sequence. The outliers are galaxies for which only upper values of the [OIII] emission are available. They are early type spirals, with ST 0, as can be deduced from Figs. 5c and 5d. These galaxies stand out conspicuously in the [NII]/H versus [OII]/H, diagram (Panel 7c) as well. As will be seen below, we interpret this behavior as being the combined effect of an overabundance of nitrogen in early-type spirals and an ionization source different from ordinary O stars.

Panel 7-d displays [OIII]/H versus [SII]/H. In this diagram, while the standard galaxies with measured [OIII]/H) 1 are within the HII region sequence, those with [OIII]/H) 1 (i.e. all the galaxies with ST 9) are slightly to the upper right. This may indicate an increasing contribution of a diffuse ionized medium with increasing galaxy spectral type. Note that Lehnert & Heckman (1994), plotting the spectra of K82a having EW(H) 30 Å in [OIII]/H versus [NII]/H and [OIII]/H versus [SII]/H planes also concluded that diffuse ionized gas contributes to the integrated spectra of galaxies. ¹

Panel 7-e shows [OII]/H versus [SII]/H. The same early-type spirals that show enhanced [NII] emission also show enhanced [SII] emission compared with the GHRs sample of McCall et al. 1985, although to a lesser extent. They are among the ones with the largest [OII]/H ratios. We will argue below that the bulk of the emission in these objects is not due to ordinary O stars.

Fig. 8 shows some line ratios as a function of ([OII] +[OIII])/H. Since the pioneering work of Pagel et al. (1979), the [OII]+[OIII])/H ratio has been widely used to derive the oxygen abundance. In principle, the relation between ([OII]+[OIII])/H and O/H is double sided (e.g. McGaugh 1991, Stasinska 1998). In integrated spectra of galaxies, regions within a few kiloparsecs from the galactic center have probably the greater weight, because most of the star formation is there (Rozas et al. 1996). Since these regions are the most metal rich, one expects that the integrated spectra correspond to the regime where ([OII]+[OIII])/H increases as O/H decreases (but this should be confirmed by simulations).

Panel 8-a shows [OIII]/[OII] versus ([OII]+[OIII])/H. [OIII]/[OII] roughly indicates the ionization state, which is linked to the ionization parameter U, thus to the gas density distribution in each HII region, and to the hardness of the ionizing radiation field. In this diagram, the data points for the integrated galaxy spectra lie well within the region occupied by GHRs except, maybe, for the early spectral type galaxies, with only upper values of [OIII]/H, that tend to locate in the lower envelope of the GHR sequence. The spread of ([OII]+[OIII])/H for the galaxies is smaller than for individual GHRs. This is reasonable, since the integrated spectra are averages over GHRs of various metallicities, ionization parameters and mean affective temperatures.

Panel 8-b displays [NII]/[OII] versus ([OII]+[OIII])/H. We see that all galaxies in the sample fall inside the GHR sequence in Fig. 8b, except three that have higher [NII]/[OII] than GHRs of same ([OII]+[OIII])/H. Note that the three exceptions also have high [NII]/H and are all the galaxies of our sample that have ST -3.

Panel 8-c displays [SII]/[OII] versus ([OII]+[OIII])/H. In this diagram, most of the standard galaxies fall inside the GHR sequence (although, as already noted by McCall et al. 1985, the GHR sequence is not so well defined when using [SII]/[OII] instead of [NII]/[OII]), the galaxies with ST -3 tending to be above that sequence.

4.3. A tentative interpretation of emission line trends with galaxy spectral types

While the trends between the emission line properties in the integrated spectra of normal spiral galaxies and spectral types are impressive, their interpretation is not necessarily straightforward, since it is dependent on the physical state and spatial distribution of the gas and of the ionization mechanisms.

It was shown by Zaritsky et al. (1994) that the characteristic oxygen abundance in a spiral galaxy derived from its individual giant HII regions increases towards earlier morphological types. The trend seen in Fig. 5f for ST 0 corresponds (at least qualitatively) to what is expected. Using the Zaritsky et al. (1994) calibration of ([OII]+[OIII])/H into O/H, it would translate into a decrease in the average O/H by 0.5 dex for ST going from 0 to 10. This is only an indicative value, because the presence of abundance gradients makes it impossible to be more informative without numerical simulations. The case of the galaxies with ST 0, will be discussed later.

Zaritsky et al. (1994) have also found a strong correlation between the characteristic oxygen abundance and galaxy luminosity (see also Vila-Costas & Edmunds 1992; Roberts & Haynes 1994). Due to the strong correlation between galaxy luminosity and Hubble type in their sample, it is impossible to know which of these two quantities is the primary cause of the correlation. It is interesting to point out that, in the sample of galaxies discussed here, there is no such correlation between O/H and luminosity (most of the galaxies in Table 2 have M_B in the narrow range between -19.9 and -20.7, see K92b) and that the correlation between luminosity and spectral type, if existent, is at most weak. This suggests that O/H is rather linked to the stellar populations, as measured by the spectral types, than to galaxy luminosity.

The increase of [OIII]/[OII] with spectral type (Fig. 5g) for ST 0 is probably due to an increase of (or an increase of U) when O/H decreases, similarly to what is invoked for individual GHRs (e.g., Garnett & Shields 1987).

The increase of [NII]/[OII] with decreasing spectral type (Fig. 5h) is probably due to an increase of N/O. Indeed, in GHRs, the fact that [NII]/[OII] increases with decreasing ([OII]+[OIII])/H is interpreted, using photoionization models, as being due to an increase of N/O with O/H (Thurston et al. 1996). It is likely then, that in the spectral sequence of galaxies there is an increase of the average N/O with decreasing spectral type, linked with the increase of O/H. This is in qualitative agreement with a secondary production of nitrogen. A priori , one cannot exclude the fact that the increase of [NII]/[OII] with decreasing spectral type could be simply attributed to the decrease of O/H (i.e., with N/O staying constant), due to the thermal properties of GHRs. Indeed, [NII]/[OII] = (N/O) ((NII)/(OII)), where (NII) and (OII) are the emissivities of the [NII] and [OII] lines, and (NII)/(OII) is a decreasing function of the electron temperature, thus an increasing function of O/H. Note, however, that the three early-type galaxies that are out of the GHR sequence in Figs. 7c, 7e, and 8b, must have particularly high N/O, because their large [NII]/[OII] cannot be due to a particularly low electron temperature, since [NII]/H is large as well.

The ratio [SII]/[OII] does not present any significant trend with spectral type (Fig. 5i). This is consistent with the assumption that the abundance ratio S/O is constant, since both S and O are primary elements. There is a hint of a decrease in that line ratio with increasing spectral type but this may be ascribed to the variations of the ionization parameter along the sequence more than to variations in the relative abundances of S and O.

The increase of [OII]/H as the galaxy spectral type decreases from 0 downwards (as well as that of [SII]/H and [NII]/H, Figs. 5b, d, e) is very interesting. Early spectral type galaxies appear as upper limits in the [OIII]/H versus [OII]/H diagnostic diagram, since their [OIII] emission is weak. They stand out completely from the giant HII region sequence in diagnostic diagrams with [NII] or [SII]. The integrated spectra of these galaxies are similar to those of LINERs.

Interestingly, 11 of the 15 galaxies of our sample have had their nuclear regions observed by Ho et al. (1997). Their classification of these nuclei into LINERs (L), Seyfert 2 (S2) and HII regions (H) is as follows (see Table 1): all the H type nuclei correspond to ST 0, while NGC3623 and NGC3368 (ST -3) correspond to LINERs. NGC3147 (ST = 0.0) has a S2 nuclear spectrum and NGC3627 (ST = 1.4) has a transition spectrum (T2/S2). The other galaxies in Table 1 have not been observed by Ho et al. (1997). So, the integrated spectra of the galaxies with earliest spectral types have characteristics of LINERs, and the nuclear regions of these galaxies too (we have checked that the contribution of the LINER nuclei to the integrated galaxy spectra is negligible). This seems to indicate that what gives rise to the LINER phenomenon, at least in the normal galaxies we are studying here, is not specifically related to the nucleus.

Given the evidence presented above that metallicity, as measured by O/H, decreases along the spectral sequence, it would be tempting to attribute the distinct behavior of the galaxies having ST 0 in Figs. 5, 7, and 8 to over-abundances in the gas. But, as stated in Sect. 4.1, over-abundances would produce low [OII]/H, not large ones as observed.

Another possibility is that the distinctive features presented by early type galaxies are due to the aging of the stellar populations associated with GHRs. Sodré & Cuevas (1997), using the spectral synthesis code GISSEL (Bruzual & Charlot 1995), have shown that the spectral sequence of galaxies is well reproduced if one assumes that the star formation rate has the form , where is an increasing function of the galaxy spectral type. Such a model explains, at least qualitatively, why the equivalent width of H emission increases with increasing galaxy spectral type, as found in Sect. 3. For galaxies with the earliest spectral types, most of the star formation has occurred long ago, and very few O stars are present. In this case, the radiation field in the Lyman continuum may be dominated by post-AGB stars (e.g. Bressan et al. 1994), because, in simple starburst models, these stars provide most of the ionizing photons for bursts older than yr. This radiation field is much harder than that produced by the massive O stars that power classical HII regions. It is available to ionize the gas bound in the primitive HII regions or the diffuse material produced by their eventual disruption. Binette et al. (1994) have shown that the emission line properties (EWs and line ratios) of early type galaxies can be accounted for by photoionization by such a stellar population. A similar model could well explain at the same time the high [OII]/H, [SII]/H, and [NII]/H found in the early spectral type galaxies of our sample.

The long lasting dispute about the ionization mechanism in LINERs - photoionization versus shocks - is not settled yet, precisely because under certain conditions both explanations can account for the observed emission line spectra. Models of evolving stellar populations that estimate the energy release in supernovae and stellar winds (e.g., Leitherer & Heckman 1995) will help in the solution of the problem.

We have seen that the effective extinction coefficient at H, , decreases steadily with increasing spectral type. The beautiful trend shown in Fig. 5a is somewhat surprising, since it is not commonly believed that early-type spirals have higher average extinction than galaxies of later types. Indeed, the related issue of the overall opacity in galaxies has been subject of significant debate, often with contradictory opinions (e.g., Valentjin 1990, Davies & Burstein 1995). Wang & Heckman (1996) have verified that the dust opacity, measured from the ultraviolet to far-infrared luminosity ratio, increases with increasing galaxy infrared luminosity. Taking into account that the far-infrared luminosity is approximately constant for early morphological type spirals and decreases slowly towards later morphological types (e.g., Roberts & Haynes 1994), the correlation noticed by Wang & Heckman might indicate that the dust opacity is indeed decreasing towards later morphological types, in qualitative agreement with the results reported here.

Now, assuming for simplicity that dust properties are invariant with galaxy spectral type, this trend suggests that the mean surface density of dust in galaxies decreases along the spectral sequence from early to late spectral types by a factor of 4. Since the mean surface density of neutral hydrogen in normal galaxies increases towards later types (Roberts & Haynes 1994), the dust-to-gas ratio must be decreasing as the spectral type increases. This is consistent with the idea of dust being formed less efficiently in low metallicity environements, as shown observationally by Lisenfeld & Ferrara (1998). Note that some of the scatter in Fig. 5a may be produced by the random inclination of the galaxies and that the far infrared properties of normal galaxies actually suggest some variations of the dust properties with galaxy type (Sauvage & Thuan 1994).

© European Southern Observatory (ESO) 1999

Online publication: April 19, 1999
helpdesk.link@springer.de