1.4.1 The UV and LAE LF
As mentioned in the beginning of this Introduction, SFGs are the most likely candidates as main driver of reionization, as all previous candidates were dismissed one by one (see Sect. 1.1.3). But a definitive observational proof of this fact has yet to be made. To investigate the SFGs as potential sources of reionization, the method often used is to derive the LF and to infer the ionizing flux density of the selected population. As explained in Sect. 1.2, two methods are used to select SFGs, the UV selection (LBGs) or selection from the Lyα line (LAEs).
The underlying biases introduced by these selection methods are not completely clear and whether there are intrinsic physical differences between galaxies selected as LAEs or LBG (or both) remains to be determined. Up to now the study of the LF of SFGs has been divided in two: the UV LF for the LBG selected galaxies, and the LAE LF for the Lyα selected galaxies. Progresses on the LF of SFGs have been made in parallel as it is not known how to combine/reconcile the results obtained with LAEs and LBGs.
Based on LBG studies, the LF is well represented by the Schechter function and evolves sig- nificantly for z > 4 with a depletion of bright galaxies and an a steepening of faint end slope as illustrated in Fig. 1.12 taken from Bouwens et al. (2015b). This indicates that the proportion of faint UV galaxies increases at higher redshift, as it is expected and predicted by the hierarchical scenario of galaxy formation. The same trend had already been observed prior to this work and is now considered a well established fact (see e.g. Bouwens et al., 2007; Oesch et al., 2014; Finkelstein et al., 2015, and the references therein).
Figure 1.12: Left panel: observed UV LFs in different redshift bins. Right panel: observed UV LFs normalised to have the same density at M1600,AB ∼ −21.
Both figures are taken form Bouwens et al. (2015b)
sented by the Schechter function for extremely faint galaxies, or whether a turn around is observed. As for now, no sign of such turn around has been observed, but this conclusion remains limited by the depth that can be achieved in current surveys.
Regarding reionization, it was estimated in Bouwens et al. (2015a) using the LFs derived in Bouwens et al. (2015b), that the faint end slope would have to be extrapolated down to M1600,AB ∼ −13 to allow LBGs to match the ionizing cosmic emissivity and therefore drive reion-
ization. However, the corresponding observations only extend down to M1600,AB ∼ −16, and only
for the lowest redshift bin at z = 4, which leaves a consequent gap of 4 dex with no reliable information.
The study of the LAE LF is not as "straightforward” as for the LBG LF. The main reason behind that is that the LAE population is affected by more observational biases:
◦ The difficulty to scan large volumes. Using large band photometry combined with photomet- ric redshift for LBG selection, allows to probe a continuous redshift range within the area surveyed. This is not the case with imaging, since the narrow filters only allow LAEs within very restricted redshift ranges. Furthermore, spectroscopic confirmations are (telescope-)time consuming making it difficult to achieve large samples of LAEs over large volumes. When using serendipitous detections in slit spectroscopy, the volume is even lower and introduce additional biases regarding the measured flux (see Sect. 2.3.2).
◦ The more complex selection process can lead to higher incompleteness. For example, using NB imaging allows only to detect LAEs with EWLyα above a certain threshold.
◦ Potential systematic loss of flux which can happen when using NB imaging (if the line profile is larger than the filter) or slit spectroscopy (if the source is extended).
◦ Escape fraction of Lyα photons and apparent clustering, especially when probing redshift approaching zeor (see previous sections of this introduction).
Historically the first LAE samples were selected using NB imaging (see Cowie & Hu, 1998; Ouchi et al., 2003; Hu et al., 2004; Dawson et al., 2007; Ouchi et al., 2008, and more). With progresses made in observational techniques and instrumentation, these techniques became more and more efficient over the past years to survey larger areas of the sky (see e.g. SILVERRUSH survey, Ouchi et al. (2018); Shibuya et al. (2018b,a) which covers 14 and 21 square degrees at respectively z = 6 − 7). Narrow-band imaging is therefore becoming essential to survey large areas, which is required to study the brightest and rarest LAEs. Nonetheless, recent NB surveys remain affected by
the biases described above. To avoid these, it is possible to use a blind spectroscopic selection. Such selection can be achieved with serendipitous detections in slit spectroscopy observations (Cassata et al., 2011), but the ideal way to go is to use IFUs, which produce complete spectra for each spatial pixel of their FoV (see Chpt. 2). The advantage of this type of instrument, in addition to the large quantity of information, is that they allow an as unbiased as possible selection of all LAEs within a given volume.
The very first attempt of LAE selection with IFU was made in van Breukelen et al. (2005) and a few years later in Adams et al. (2011). However, these first pioneer studies were severely limited by either a small FoV, low spectral resolution or low sensitivity.
More recently the VLT/MUSE instrument Bacon et al. (2014) was commissioned. It is a large field of view (1′× 1′) IFU allowing a continuous redshift selection between 2.9 < z < 6.7 for LAEs
and a spectral resolution varying between R = 2000 and R = 4000 (see Chpt. 2 dedicated to the MUSE instrument). The primary science goal of MUSE is to detect LAEs taking advantage of its increased sensitivity, high spectral resolution and large FoV. The first results of the LAE LF with MUSE in blank fields are presented in Drake et al. (2017b,a); Herenz et al. (2019) and proved the efficiency of this new generation instrument to detect extremely faint LAEs with minimal biases. Studies of the LAE population find a deficit of bright Lyα emitting galaxies at z ≥ 6 which can be attributed to either an increase in the fraction of neutral hydrogen in the IGM or an evolution of the parent population (or both). For lower redshift, while some studies report an evolution similar to the one described for the LBG population, this evolution seems to be less significant and is not always seen (Kashikawa et al., 2006; Pentericci et al., 2014; Tilvi et al., 2014; Drake et al., 2017b; Herenz et al., 2019). In the same way as for the UV LF, the study of the LAE LF has been limited by available observations and depth. And the uncertainties are much larger since the LAE population is intrinsically more biased. In order to get around the depth limitation that is currently affecting both the study of the UV and LAE LF, recent work have turned towards strong lensing clusters to increase their reach.
1.4.2 Using strong lensing clusters
Since the first observation of a lensed galaxy in Soucail et al. (1988), strong lensing clusters have been used for various purposes. Specific and highly magnified objects can be used to perform detailed analysis taking advantage of the increased signal-to-noise (S/N) ratio and/or the increased spatial resolution (see e.g. Patricio et al., 2018). Alternatively, galaxy clusters can be used as gravitational telescopes while surveying the entire background of the cluster looking for fainter – but magnified – galaxies (Pelló et al., 1998; Richard et al., 2006). Such technique can be used to either select intrinsically fainter galaxies or galaxies at higher redshift. Recently, the Hubble Space Telescope (HST) Frontier Fields program (HFF) (Lotz et al., 2017) observed six of the most massive galaxy clusters. Each of the six clusters were observed for a total of 103 hours, pushing the observations to the limits of what can be achieved with HST.
Encouraged by the unprecedented depth of these new observations, many studies started work- ing on the UV LF of the galaxies detected behind these clusters (see e.g. Livermore et al., 2017; Bouwens et al., 2017; Atek et al., 2018). At the cost of increased uncertainties, smaller volumes probed and more complex data processing, these studies managed to further constrain the shape of the UV LF. As an illustration, Bouwens et al. (2015b) only reached down to MUV,AB∼ −17 at
z = 6 using a combination of deep HST blank fields, whereas using similar methods and the HFF observation, the three studies mentioned above managed to set reasonable constraints down to to M1600,AB ∼ −15 at the same redshift (see Fig. 1.14). For fainter magnitudes, the LFs derived start
Figure 1.13: Lensed galaxy in the A370 galaxy cluster observed with HST. This arc was historically the first confirmed observation of a galaxy lensed by a galaxy cluster (Soucail et al., 1988).
credits: NASA, ESA, Jennifer Lotz and the HFF Team (STScI)
by the uncertainty on both the magnification of individual galaxies and the choice of specific mass model(s). A more detailed presentation of all lensing effects and of the methods used in the frame of lensing fields is provided in Chpt. 3. Even though significant progress was achieved, all three studies concluded that no turn over was visible within the magnitudes probed by the observations. And the large uncertainties on the faint end transpose into larger uncertainties on the integrated ionizing flux.
To date, strong lensing fields have only been used once in Bina et al. (2016) as a proof of concept using MUSE, to set constraints on the LAE LF. The small sample (17 LAEs only) and the extremely small volume probed seriously limited the conclusion of this work, leaving a large margin of progression on the determination of the faint end shape of the LAE LF.
By essence, lensing fields surveys are exploring a different population than blank fields. The increased depth in these fields comes at the cost of a smaller volume explored, and since faint galaxies are much more numerous than bright ones (see LFs in Fig. 1.12, it is extremely unlikely to find bright galaxies in the background of lensing fields. For that reason, it is challenging to derive a single and coherent LF that would be well constrained on both the faint and the bright end. This is clearly visible in Fig. 1.14, where the LFs derived in both Bouwens et al. (2017) and Livermore et al. (2017) could almost be adjusted by a single straight line showing that they are very inefficient to select even moderately bright sources. On the contrary, it can be seen in Fig. 1.12 that LFs computed in blank fields are obviously less sensitive to the shape of the faint end. Lensing fields are therefore very important to probe the faint end of the LF, whereas larger blank-fields surveys are needed to probe the bright end. Regarding the Schechter parameterization, this means that it is very challenging to have a simultaneous and consistent determination of the three parameters, unless combinning very large volumes with extreme depth.
1.4.3 Total SFGs contribution to reionization
As explained all along this Introduction, LAEs and LBG are just reflecting a selection method, and the separation of the SFGs in these two populations is not necessarily representative of any
Figure 1.14: UV LFs derived at z = 6 using HFF observations.
Figure taken from Atek et al. (2018)
physical differences between them. However, regarding reionization and the study of the LF, these two populations have most of the time been studied in parallel with no attempt to unify them or to further characterize their interrelation. This raises the questions: how much of the SFGs are we missing when looking at only the LAEs or the LBGs ? Do we need to unify these two population to see SFGs producing enough ionizing flux to drive and maintain reionization ?
The interrelation between these two populations is studied in Arrabal Haro et al. (2018), where a simultaneous search for LAEs and LBGs is undertaken using the narrow band photometry of the SHARDS survey (Pérez-González et al., 2013) and public HST data. The large field of view (130 arcmin2) combined with the use of narrow band filters allows for a good statistic and very secure
photometric redshift in the range 3.35 ≤ z ≤ 6.8. However, and as stated by the authors these observations are not suited to detect sources with wither faint Lyα emission or faint continuum (unfortunately the range explored in both MUV and LLyα are not discussed in this work). The
interesting point of this work, is that for once, both LAE and LBGs are detected using the exact same set of observation, making the comparison between these populations extra relevant. As seen in Fig. 1.15 (taken from Arrabal Haro et al. (2018)), it appears that within the volume explored, the fraction of observed galaxies showing Lyα emission increases with redshift. And more precisely, the fraction of LAE without continuum detection also increases with redshift.
Following the same line of inquiry, Maseda et al. (2018) combined the Hubble Ultra-Deep Field (HUDF) observations with the MUSE HUDF and UDF-10 observations (see Bacon et al., 2017) to investigate the LAEs detected with MUSE without optical counterparts on HST images. This type of sources appears to be numerous (see e.g., Bacon et al., 2015, 2017; Herenz et al., 2017) and are expected to play a significant role in the reionization, yet they remain invisible in the deep photometric observations of the HUDF. The same type of population has been observed in Mahler et al. (2018), by comparing again MUSE observations to the even deeper HFF images of the A2744 lensing field. These LAEs have intrinsically high equivalent widths, but since no UV continua can be measured, only (high) lower limits of EWLyα can be derived. However these limits are sufficient
to see that some LAEs are detected with EWLyα > 240Å with no sign of AGN activity, therefore
reinforcing the claim that this can be a sign of either a young age or very low metallicity (see Sect. 1.2.1). As suggested in Hashimoto et al. (2017) such high EWLyα can be explained without
Figure 1.15: Interrelation between the LAE and LBG population in the SHARDS survey.
Figure taken from Arrabal Haro et al. (2018)
invoking population III stars or irregular top-heavy IMFs and is therefore not necessarily indicating galaxies of first generation, strictly speaking.
By stacking HST images in different redshift bins, Maseda et al. (2018) showed that LAEs detected in the MUSE HUDF with no optical continuum detection have a typical UV magnitude of MUV ∼ −15, which makes them the faintest "UV-objects” with spectroscopic confirmation. The
large discrepancy between the expected number of these UV-faint galaxies, obtained by extrapola- tion of the current UV LFs, and the actual number of detection suggests that only those with the highest EWLyα are detected with HST. In other words, only LAEs with an emission bright enough
show on observations as their signal is not completely drown in the large width of the HST filters. In this regards, using IFUs such as MUSE allows to indirectly probe a new population of UV-faint galaxies.
From the work of Arrabal Haro et al. (2018) and Maseda et al. (2018) it appears that LAEs and LBGs cannot be treated as two completely independent populations, and that the LAE selection is probably more suited to select intrinsically UV-faint galaxies out of the reach of deep large band photometry surveys. This picture has emerged during the last years, and it is presently urgent to adress the intersection between the different populations of SFGs. When this thesis started, no specific study had been made to compare the LBG-UV and LAE populations within the same volume of the Universe, in particular regarding the faintest galaxies responsible for cosmic reionization.