1. The background of the AGAPE search
1.1. Dark matter in galaxies
The presence of a large amount of unseen matter is a very old astrophysical problem (Oort 1932, Zwicky 1933) but its importance was widely recognised only in the seventies (Ostriker, Peebles & Yahil 1974, Faber & Gallagher 1979). Actually there are several "dark matter problems" on different scales: stellar systems, individual galaxies, clusters and superclusters of galaxies, up to cosmological scales. Dark matter appears also necessary to understand large structures formation. For a recent review on these subjects, see Dolgov (1995). Many observations suggest that spiral galaxies are embedded in massive dark haloes (Kormandy & Knapp 1987, Trimble 1987). The most conspicuous evidence for such haloes is the rotation curve of galactic disks, which does not decrease near the outskirts of galaxies. If the mass density and surface brightness profiles were similar, the rotation curve should fall according to Kepler's law.
From the rotation curves of spiral galaxies, one can estimate the amount of dark matter within 2 Holmberg radii to be larger by one order of magnitude than the amount of luminous matter, but the shape of dark haloes is unknown. Several lines of argument point towards a more or less spherical distribution, such as the existence of galaxies with a rapidly rotating polar ring, the stability of the disk of spiral galaxies against bar formation (Ostriker & Peebles 1973) or the distribution of the globular clusters (Harris & Racine 1979). The sphere seems often flattened in the direction of the rotation axis (for a recent review, see Sackett 1995and references therein).
The nature of dark haloes remains also unknown. Many candidates have been proposed, either baryonic or not, ranging from light neutrinos to very heavy black holes of 10 , but it is out of the scope of this paper to review them extensively (for recent reviews, see for instance Dolgov 1995and Griest 1995). Nevertheless, we can mention some unconventional views, such as the modified Newton dynamics (Bekenstein & Milgrom 1984), or cold molecular hydrogen as the constituent of dark haloes of spiral galaxies (Pfenniger, Combes & Martinet 1994).
1.2. Baryonic dark matter
Although the subject of primordial abundances has recently become rather confused, Big Bang Nucleosynthesis indicates that the density of baryonic matter in the universe is probably around 10 times larger than that seen as stars or interstellar gas (for a recent discussion, see for instance Cardall & Fuller 1996 and references therein). But the Cosmological Standard Model gives no hint as to the location of this baryonic matter and its relative distribution between galactic haloes and intergalactic medium in clusters of galaxies.
It has been suggested that galactic dark matter could be essentially made of compact baryonic objects such as low mass stars or brown dwarfs. Brown dwarfs are stars too light () for the gravitational pressure to fire nuclear reactions and are a natural candidate for the constituent of galactic haloes (Carr, Bond & Arnett 1984). It is considered that they should be heavier than 10 lest they would evaporate too quickly (De Rújula, Jetzer & Massó 1992). Such objects should most easily be seen in the red and infrared bands (Kerins & Carr 1991). A few may have been in fact observed, some orbiting brighter compagnons: GD 165B (Zuckerman & Becklin 1988) and Gl229B (Nakajima et al. 1995, Allard et al. 1996), as well as others free flying in the Pleiades cluster: PPl 15 (Stauffer, Hamilton & Probst 1994), Teide 1 (Rebolo, Zapaterio Osorio & Martín 1995) and Calar 3 (Zapaterio Osorio, Rebolo & Martín 1996). Both PPl 15 and Teide 1 have residual Lithium, and Calar 3 resembles Teide 1 like a twin. (Basri, Marcy & Graham 1996, Martín, Rebolo & Zapaterio Osorio 1996).
1.3. Gravitational microlensing
Direct searches for brown dwarfs can at best explore the solar neighbourhood. To detect them further out, it was proposed a few years ago by Paczyski (1986) to search for dark objects through gravitational lensing. When a compact object passes near the line of sight of a background star, the luminosity of this star will be temporarily increased in a characteristic way.
Several experiments have been implementing this idea since 1990 and have indeed seen microlensing events. Two groups have been looking towards the Magellanic Clouds: the EROS collaboration (Aubourg et al. 1993, Ansari et al. 1995a, Milsztajn 1996) and the MACHO collaboration (Alcock et al. 1993 , 1995a, Bennett 1996). Microlensings have also been searched for in the direction of the galactic bulge by three groups: OGLE (Udalski et al. 1993, 1994), MACHO (Alcock et al. 1995b, Sutherland 1996) and DUO (Alard, Mao & Guibert 1995, Alard 1996), who have observed a large number of events. The microlensing phenomenon can now be considered as established.
However, the number of events towards the Large Magellanic Cloud (LMC) is lower than expected, 50% or less of what one would expect with a standard spherical halo (Bennett 1996, Milsztajn 1996), but statistics remain very poor. Moreover, with only one line of sight, it is very difficult to disentangle the various parameters which enter in a galactic halo model: density, velocity distribution, mass distribution, flattening.
MACHO will continue for two more years and the upgrade of EROS (Couchot 1996) will start operation soon. However, the "classical" technique used in these experiments does not allow to explore other directions through the halo, because the two Magellanic Clouds are the only possible targets with enough resolved stars.
1.4. Going further, the "pixel method"
It is thus tempting to look at rich fields of stars further out, such as the M 31 galaxy. But most stars of M 31 are not resolved and a new technique must be developed. Such a technique, the "pixel method", has been proposed and implemented by us (Baillon et al. 1992, 1993, Ansari et al. 1995b). A similar idea, relying on image subtraction, has been independently proposed by (Crotts 1992), and implemented by the Columbia-VATT collaboration (Tomaney & Crotts 1994, Tomaney 1996).
The method we propose is the following: in a dense field of stars, many of them contribute to each pixel. However if one unresolved star is sufficiently magnified, the increase of the total flux of the pixel will be large enough to be detected. Therefore, instead of monitoring individual stars, we propose to follow the luminous intensity of the pixels of the image. Then all stars in the field, and not the only few resolved ones, are candidates for a micro-lensing, so that the event rate is potentially much larger. Of course, only the brightest stars will be amplified enough to become detectable above the fluctuations of the background, unless the amplification is very high and this occurs very seldom. In a galaxy like M 31, however, this is compensated for by the very high density of stars, and indeed various evaluations (Baillon et al. 1993, Jetzer 1994, Colley 1995, Han & Gould 1996) show that a fair number of events should be detectable. This paper is devoted to the description of AGAPE (Andromeda Gravitational Amplification Pixel Experiment), which implements this idea in the direction of M 31, on data taken in autumns 1994 and 1995 at the 2 metre telescope Bernard Lyot (TBL) at Pic du Midi Observatory in the French Pyrénées.
In Sect. 2, after recalling the principles of the method, (introduced in Baillon et al. 1992and 1993), we give analytic evaluations of the number of events expected. Although these analytic estimates can at best be very rough, they provide useful qualitative insights. To get reliable estimates in the true observational conditions, we resort to Monte-Carlo simulations.
In Sect. 3 we describe the telescope, the detector, the conditions and the course of the observations. Sect. 4 is devoted to the geometric and photometric alignments of successive images and to the absolute photometry. In Sect. 5 we show that the high level of stability reached on the average super-pixel (a group of elementary pixels) allows us to detect variable objects that would have been very difficult to see otherwise. The detailed analysis of the variations we detect will be the subject of separate publications.
The pixel method should also give interesting results in the bar of the LMC, and we have started to analyse the data of the EROS collaboration in this framework (Melchior 1995). The results will also be published elsewhere.
© European Southern Observatory (ESO) 1997
Online publication: May 5, 1998