The discovery of extragalactic objects with very low heavy element abundance was made by Searle and Sargent (1972) who reported the properties of two intriguing galaxies, IZw18 and IIZw40. They emphasised that they could be genuinely young galaxies in the process of formation, because of their extreme metal under-abundance, more than 10 times less than solar, and even more extreme than that of H II regions found in the outskirts of spiral galaxies. At the time of this discovery the general wisdom that most galaxies (in particular the ellipticals) had been formed over a short period during a dynamical free fall time of few 107 years (Eggen et al. 1962) started to be challenged (e.g. Searle and Zinn 1978). It is also during the 70s that the first hierarchical models of galaxy formation were constructed (Press and Schecter 1974). Because dwarf galaxies condense from smaller perturbations than giants, the Cold Dark Matter models (CDM) predict that low mass galaxies could still be forming at present epoch. The discovery by Searle and Sargent (1972) has been an impressive stroke, since one of these two galaxies (IZw18) is still in the book of records, as we shall later elaborate on. These two objects gave rise to many systematic searches for more objects in a quest for local genuine young galaxies or ``unevolved galaxies'', depending upon the alternative viewpoints that some galaxies could be caught in the process of formation or that they simply were the result of a very mild evolution over the Hubble time. These galaxies had the advantage of being gas-rich, with spectra dominated by strong emission lines (see Fig. 6) favouring their detection. Many techniques have been employed to find them, sometimes at large distances despite their intrinsic low luminosity.
The last nearly three decades have brought a wealth of data from numerous studies on dwarf galaxies, including information on their chemical composition. It became clear that ``metal-poor'' would be analogous of ``low mass'' galaxies (Lequeux et al. 1979). For this reason, our review will largely focus on dwarf galaxies, but we shall address the question of the existence of large and massive proto-galaxies - essentially devoid of metals - at large redshift in the last section. The dwarf galaxies are not only interesting for understanding the process of galaxy formation. For the gas-rich ones with active star formation, one motivation to study them has been the hope to better understand the processes of massive star formation in low metallicity gas. The fact that they are dwarfs means that spiral waves can not be sustained. They turn out to be test cases for chemical evolutionary models and offer the possibility to approach the primordial helium abundance with a minimum of extrapolation to early conditions. Many galaxies of different kinds can be identified as metal-poor and it is an interesting question to find out about the connections that bridge them together. Finally, a lot of new studies concentrate on the impact of massive star formation onto the ISM in star-bursting dwarf objects, that in turn can lead to constrain the supernovae rate, IMF, the metal dependence of the winds in Wolf-Rayet (WR) stars etc. In fact there are many issues in astronomy where it is essential to understand dwarf galaxies (actively forming stars or not).
A definition of what is a metal-poor galaxy is indeed necessary at this point. To define the metallicity (Z, i.e. the relative abundance of elements other than hydrogen and helium) of a galaxy requires some words of caution because in a given galaxy, depending where one looks, this quantity may vary substantially. For example in our Galaxy, the bulge, the solar neighbourhood and the halo differ in metallicity. The most metal-poor halo stars have heavy element abundances 10-4 times that of the Sun (Cayrel 1996) while stars in the Galactic centre may be three times more metal rich than the Sun. On the other hand, the large ionised complexes in the ISM show a narrower range down to only 1/10 the solar value. Thus metallicity depends on what one looks at: stars or gas, and if one considers gas - what phase: neutral atomic, molecular or ionised? Moreover the metallicity of stars is found to depend on their age, and depends on which elements are investigated; that a star or nebula is deficient of a certain element does not automatically mean that the overall chemical abundances are low.
Thus, one must be careful in defining what metallicity means for a given galaxy before comparing observations and looking for trends. Another complication stems from the different techniques in use for determining chemical abundances. Metallicities of Local Group dwarf spheroidals (dSph) have largely been investigated through photometry of their resolved stellar populations, which are dominated by old stars, and are found to be metal-poor, as measured by [Fe/H]. In general dSph galaxies contain little or no gas, and no H II regions. Dwarf irregular (dI) galaxies on the other hand usually have plenty of gas and ongoing star formation as witnessed by the presence of H II regions. It is relatively easy to derive metallicities, especially oxygen (O) of the ionised gas in H II regions explaining why most investigations are based on such. Hence the ``metallicity'' of dSphs and dIs measures quantities that are not directly comparable. Some galaxies have no H II regions and very low surface brightness, and the only available spectroscopic method hitherto, has been to study individual bright stars which are within reach in nearby galaxies only. In more distant galaxies, individual stars are not observable and the metallicity has to be inferred from either absorption line spectra of an integrated stellar population (not necessarily homogeneous in age or composition) or from the nebular emission lines, if any. Again, the different methods in general measure different elements.
So what useful definitions of ``metallicity'' do we have at hand? The possibility of using the metallicity of the H II regions has the advantage of providing an ``up to date'' metallicity, while stars reflect the metallicity of the cloud from where they were born, perhaps a Hubble time ago. Nebular abundances show large spatial abundance variations and gradients in some galaxies, e.g. giant spiral galaxies, like the Milky Way. Although measurable abundance inhomogeneities could be expected, most dwarf galaxies seem rather well mixed. A possible problem with metal-poor H II regions is self-pollution of fresh metals by winds from young massive stars, so that the abundance inferred from the nebular emission lines may not be representative of that in the local ISM (Kunth & Sargent 1986). On the other hand, the metallicity of the stars in a galaxy depends on which stellar population is studied. Thus it is not surprising, in particular for galaxies which have experienced continuous star formation, that stars have different abundances. Integrated spectra of galaxies provide a luminosity-weighted average of the metallicity. This average metallicity will change with time due to the photometric evolution of the stellar population, even if no new stars are formed.
A good, well defined, metallicity indicator would be the fraction of baryonic matter (presumably having initially primordial composition) that has been converted into heavier elements by means of stellar nucleosynthesis. This material may have been returned to the ISM or may still be locked up in stars. Such a definition would indicate that our main interest goes beyond the element abundances themselves from the fact that they provide information about the history of a galaxy. The relative abundances and gas mass fractions might unveil different histories among galaxies with the same metallicities. However, it is clear from he above discussion that such a definition of metallicity remains impractical, since not directly measurable. However, we would like to keep this ideal definition in mind for the rest of the discussion.
Now we wish to go back to the question - What is a metal-poor galaxy? Under the assumption that various tracers give a plausible picture about the ``ideal metallicity'' we can now try to compare different kinds of galaxies. One of the most fundamental parameters for a galaxy is of course its mass. This mass, which may consist of stars, gas, dust, baryonic- and non-baryonic dark matter, is more difficult to measure, than e.g. the luminosity, but to the first approximation, mass and luminosity correlate. Based on their luminosity and size, galaxies can be divided into dwarfs and giants. It has been found that the metallicity of a galaxy in the local Universe correlates positively with its luminosity (although with a large scatter), thus also reflecting a positive correlation with mass. The reason for this behaviour is a fundamental issue to understand. One explanation could be that dwarfs evolve more slowly because of smaller mass densities, which to the first order fits with the observation that dwarfs, except dwarf elliptical/spheroidals, are more gas-rich than giants. Another possible explanation is that dwarfs have weaker gravitational potential hence are more susceptible to loose metal enriched material from supernova driven winds. We will elaborate more on this later.
A natural reference for element abundances and the ratio between them, could be the Sun. Thus a starting point could be that ``metal-poor'' means anything which has sub-solar abundances and vice versa for ``metal rich'', which implies that basically all local galaxies fainter than our Galaxy are metal-poor. High redshift neutral gas clouds, which may be the building blocks of today's galaxies, are observed to have metallicities down to 0.001 Z. Thus, there is a large range of metallicity to explore, and it is meaningful to distinguish between metal-poor, very metal-poor and extremely metal-poor. Since this review is called ``the most metal-poor galaxies'' it is natural that we focus on the latter two subclasses. What do we see locally? Among dSph we find metallicities extending down to 1/100 Z, while the LMC and SMC are at roughly 1/3 Z and 1/8 Z respectively. Dwarf irregulars have sub-solar abundances, ranging down to 1/40 Z. In addition there are many blue compact galaxies (BCGs) in the range 1/10 to 1/50 Z, with, as we shall show later, IZw18 at the extreme.
A more workable definition could use the minimum enrichment one predicts for a single burst event, using the instantaneous recycling closed box model. Kunth and Sargent (1986) found that such minimal expected metallicity increment in a pristine H II region would be higher than or equal to the metallicity of IZw18. Similarly one finds that even converting only on the order of two percent of pristine gas in a galaxy to stars, results in a metallicity of 1/50 Z, i.e. the metallicity of IZw18. There are of course a lot of uncertainties that go into these calculations, but they can be a useful guide. Another, more practical guide, is to consider the O/H distribution of star forming dwarf galaxies studied over the last 30 years. This shows a peak around 1/10 Z and drops sharply below that value. Moreover, for most models: of stellar winds, evolutionary tracks, WR-stars, star formation etc., a critical dependence on metallicity is seen around 1/10 Z. This is why we have adopted throughout this review the working hypothesis that galaxies with metallicity below 1/10 Z will be considered as very metal deficient. Therefore galaxies like the Magellanic clouds will not be our main interest in this paper. Moreover, such a limit means that this review will be biased towards dwarf galaxies. In particular we will focus on blue compact galaxies (BCGs). The reason is partly due to selection effects: since blue compact galaxies have bright emission lines and high surface brightness, it is fairly easy to discover them and derive their metal content. Thus, there exists a lot of high quality data on BCGs, but we should keep in mind that very metal-poor galaxies may be as common among other types of dwarf galaxies.
Metallicities can be studied at great distances under special conditions. Observations of high redshift QSOs and radio galaxies (e.g. Dunlop et al. 1994), reveal the presence of dust and metal rich gas, suggesting that prior stellar nucleosynthesis has already taken place. High redshift QSO absorption line systems show a wide range of metallicities, from one thousandth solar up to 1/3 solar. Thus, while the average metallicity of the Universe certainly must have increased since the early epochs, the situation is more complex than a simple picture where high redshift means metal-poor, and low redshift metal-rich. Objects with high and low metallicities are found at all redshifts. Surely we expect objects that in the local Universe appear as metal deficient to be even more deficient at high redshift, if we could observe their precursors. Also the ancestors of the local metal rich galaxy population, i.e. the giant spirals and ellipticals, should have started out with very low abundances unless they were gradually built up by merging smaller galaxies. Currently, both the theoretical and observational pictures, tell that the latter is an important mechanism. Dwarf galaxies, the survivors who form the local metal-poor galaxy population, may thus be the principal building blocks of the Universe on large scales.
The structure of the rest of paper is as follows: In Section 2 we discuss how metallicities are measured and in Sect. 3 the physical mechanisms that control the metallicity of a galaxy. In Section 4 we review the physics of metal-poor galaxies in the local Universe, while in Sect. 5 turning to some key objects like IZw18. In Sect. 6 we discuss survey techniques, and the distribution in space and luminosity of metal-poor galaxies. In Sect. 7 we examine observed trends in the metal-poor galaxy population and various possible evolutionary links. In Sect. 8 we focus on cosmology and the high redshift Universe, and in Sect. 9 we conclude.
Throughout this paper we adopt 12 + log(O/H) = 8.91 as the solar oxygen abundance. As customary, element ratios given in square brackets represent logarithmic values with respect to solar values, e.g. [Fe/H] = log(Fe/H) - log(Fe/H). We use H0 = 75 km/s/Mpc, and rescale results from the litterature based on other values of H0 when necessary. A list of commonly used abbreviations and acronyms are given at the end of the paper.