6.1. Comparison with previous results
Our finding that there is much more HI than in disc galaxies confirms on a much larger sample the previous results of S93, Horellou et al. (1995b), Casoli et al. 1996, and Boselli et al. 1997), but stands in strong contrast with that of YK. Apart from a slight difference in the X factor, since YK adopted a value of 2.8 cm-2 / (K kms-1), we see several reasons for this discrepancy. First, while our complete sample contains most of the YK one, since it contains the whole FCRAO survey (with the exception of the objects observed by Sage 1993a), we have excluded very HI-deficient objects. Conversely, the YK sample contains many HI-deficient objects belonging to the Virgo cluster. This can explain part of the discrepancy for the mean ratio, as well as the morphological type variation, since it appears that early-type objects are most affected by the HI-deficiency.
A part of the difference could also be related to the way the statistical analysis is made. Because of the wide range of the ratios, we have chosen to compute the mean of the logarithms of this ratio, which is different from (and lower than) the logarithm of the mean. There is of course no difference for the medians. In addition, using survival analysis decreases the molecular gas fraction, since about one-fourth of the galaxies are not detected in CO(1-0) . Finally, the fact that the FCRAO sample contains mostly FIR and optically bright objects could also be an explanation since it leads to observing CO-bright objects. In Figs. 7 and 8, the median values of and for the FCRAO sample are indicated as arrows, and are clearly located in the region where high values of are found. All these differences go in the same direction, and it is thus not surprising that we find lower values for the molecular gas contents.
6.2. Molecular gas fraction in the star-forming disc
Our previous discussion concerns galaxies as a whole. Our results do not imply that in the star-forming disc, HI is also predominant, because it is well known that H2 is found in the inner regions and HI in the outer regions of galaxies (e.g. Sofue 1997, Sage & Solomon 1991). For most of the sample galaxies there is no information on the actual distribution of the CO and HI emissions; we have then tried to take this into account in a statistical way. We have used the data from Warmels (1986) to estimate the ratio between the total HI mass and the HI mass inside half of the 25-th magnitude radius (this is roughly the overall extent of the CO emission, Young et al. 1995). For a sample of 57 Sa-Sd galaxies, the median value of this ratio is 2.1, that is, half of the HI mass is found inside . Within this radius, the ratio is then between 40 and 70 percent.
Another way to estimate this fraction is to consider the HI-deficient galaxies only. Indeed, observations suggest that moderately deficient objects have been stripped of their atomic gas in the outer regions, but that their inner gas is not affected. Very deficient galaxies should however be HI-deficient even in their central regions: this is the result found by Cayatte et al. (1994) for Virgo galaxies. From their data, it seems that the borderline between galaxies which are affected over their whole disc and those which are not lies at an HI deficiency around 0.6. We have thus computed the ratio for galaxies with 0.3 HIdef 0.6. There are 27 such galaxies, all of morphological type between Sa and Sc (as Cayatte et al. remark, late-type deficient objects are rare), 21 are detected in CO(1-0) , and the mean value of is 61 10 percent. Although this estimate is a very rough one, it gives the same order of magnitude as the previous one.
6.3. Variations of the conversion factor X
Our conclusions are clearly dependent on the actual value of the conversion factor for each galaxy. In classical spirals, there are several indications that the conversion factor between CO(1-0) emissivities and masses could be lower than our adopted value of . For our Galaxy, recent analysis of the Gamma-Ray Observatory data suggests a global value around (Digel et al. 1996). Other indications in the same direction come from observations of the continuum emission at 1mm of the central regions of NGC 891 and M51 by Guélin et al. (1993, 1995), and of the disk of NGC4565 by Neininger et al. (1996). This continuum emission traces the cold dust and, with an estimate of the dust temperature, allowed them to derive gas masses independantly of the CO(1-0) observations. The molecular gas masses that they find are several times lower than deduced from the CO observations, which also points towards a low value of X. This would mean that the actual masses are even lower than what we find, and that the atomic phase is even more dominant.
The determination of X in late-type spirals and irregular galaxies is a difficult problem, because of the lack of independent determinations of the mass. For the Small Magellanic Cloud, Rubio et al. (1993) have suggested a high value of X, about cm-2 / (K kms-1), by assuming that the virial masses of the clouds are equal to their molecular masses. However these clouds may not be in equilibrium and may still contain a large amount of atomic hydrogen. The uncertainties are not so high that they can change our main conclusion, that there is more HI than in all types from S0's to Irr.
The fact that high-mass late-type spirals have a higher molecular fraction than low-mass ones (Sect. 3.3) could be attributed to a variation of X. Indeed, the standard conversion factor almost certainly underestimates the true molecular mass in objects with a low metallicity, a low dust content and a high radiation field, which could the case for the CO-weak late-type objects. On the other hand, we have found no difference between the optical colors (U-B) and (B-V) of the high-mass and low-mass late-type objects; neither did we find a difference in their / ratio. Thus it seems that the difference between high-mass and low-mass late-type spirals is not due to a difference in X. One may speculate that massive galaxies form molecular clouds more easily through the effect of stronger density waves (see also Elmegreen 1993). The causes of their different behaviors remain to be explored.
6.4. Some consequences of the low values of the molecular gas fraction
The low content of spirals observed in this analysis raises an interesting question. With a median molecular gas content of , how will spiral galaxies (Sa-Sc) be able to sustain reasonably high star formation rates for more than a few billion years? If we compute roughly the molecular gas exhaustion time scale by estimating the star formation rate from the far-infrared luminosity using
(this does not take into account the return of gas to the interstellar medium, but we need only an order of magnitude), we find that half of the sample galaxies will have exhausted their molecular gas reservoir in less than 1.3 109 yr. To escape this problem, we can envisage that there are actually large exchanges between the atomic and molecular phase. If all the atomic hydrogen inside is converted eventually into molecular gas, can be enhanced by a factor of 2. Is this is not sufficient, the unescapable consequence is that there must be some inward motions of the atomic gas in order to refuel the inner regions. Such motions should be detectable by comparing the HI and CO(1-0) velocity fields in nearby spirals. This problem has been already raised by Blitz (1997) for the Milky Way, who reached similar conclusions.
If the atomic phase is more closely linked to the molecular one than usually thought of, it could also exhibit some link with the star formation tracers. Indeed, several authors (Buat 1992, Boselli 1994, Casoli et al. 1996, and references therein) have found good relationships between various star formation indicators and the HI content. This can be understood if the transformation of HI into is a slow process (indeed Heck et al. 1992 estimate that the characteristic conversion time is larger than 107 yr), while the molecular phase is relatively short-lived, either because it forms stars or it is disrupted by star formation. The atomic phase would then be the limiting factor in the cycle HI - - stars. Another interpretation has been proposed by Elmegreen (1993), in which the transition HI- is mainly governed by the external pressure and radiation field; if molecular clouds are not always self-gravitating, the molecular mass fraction in a galaxy could indeed be a poor indicator of its star-forming properties.
Finally let us point out that the fact that with present-day receivers and telescopes, the CO(1-0) emission of a normal spiral (that is, not especially selected for its far-infrared luminosity) is often undetectable at a redshift as small as a few thousand . This suggests that even in deep surveys, the CO emission of the large majority of distant spiral galaxies will not be easily detectable.
© European Southern Observatory (ESO) 1998
Online publication: February 16, 1998