Table of contents

The dominant mode of star formation in galaxies


In the 90s, deep optical images from the Hubble space telescope allowed us to observe the redshifted ultraviolet light of distant galaxies, a direct probe of their star formation activity. From these images, astronomers first deduced that, 9 billion years ago (z=1.5), galaxies were forming stars an order of magnitude faster than today (Madau et al. 1998). The main uncertainty in this result comes from the absorption of the galaxies’ light by interstellar dust. In the present-day universe, only 30% of the light is absorbed by dust in a typical spiral galaxy (Saunders et al. 1990). The absorbed energy is re-emitted in the far-infrared (FIR), which is a domain where observations are more difficult, and typically much less sensitive. Only a minority of galaxies are strongly obscured by dust, and these also happen to be experiencing intense star formation bursts caused by major mergers, a violent encounter with another galaxy (Sanders et al. 1996). The impact of dust was therefore deemed marginal.

Only later did it become clear that this was a mistake. Deeper observations of the FIR sky were produced by the ISO telescope, allowing for the first time the direct detection of dusty galaxies much beyond the local Universe. Analyzing these data, it was found that a much larger fraction (>70%) of the star formation activity was hidden by dust at these epochs (Chary & Elbaz 2001). Most of this activity happened in IR-luminous galaxies, a priori analogues to the merging galaxies observed locally. Consequently, our initial estimate of the past star formation activity increased by an order of magnitude, and it was believed to be mostly triggered by galaxy interactions.

Some years later, a new satellite was launched to observe the mid and far-IR emission of distant galaxies: the Spitzer space telescope. Detecting fainter galaxies than that observed by ISO, this telescope allowed a first complete census of star-forming galaxies up to z=1.5. Surprisingly, all of them appeared to form a tight “Main Sequence”, where their star formation rates (SFRs) were correlated to the stellar masses (Elbaz et al. 2007). This was independently discovered with the more uncertain SFRs derived from the [OII] nebular line emission (Noeske et al. 2007). This Main Sequence suggests that star formation was happening continuously over several billion years, instead of the short bursts expected from galaxy mergers.

Therefore, an other mode of star formation had to be involved. Since galaxies only host moderate reservoirs of gas which would be entirely consumed over less than a billion years (Tacconi et al. 2010), they need to be continuously replenished from an external source. The most likely hypothesis is therefore that galaxies grow through the continuous accretion of gas from the inter-galactic medium, possibly through the yet unobserved cold flows (Dekel et al. 2009).

Like Spitzer succeeded to ISO, the Herschel space telescope was launched in 2009 with a much larger mirror and vastly improved sensitivities at wavelengths from 70 to 500 microns, while Spitzer was mostly limited to 24 microns. Thanks to this, Herschel was able not only to detect more distant galaxies, but also to measure the dust emission where it is the brightest — around 100 microns in the rest-frame — thereby reducing the uncertainties when determining the total infrared luminosities and star formation rates. Beyond 200 microns, the emitted light traces the mass of dust in the galaxy, which can be used to infer the mass of hydrogen gas when the metallicity is known.

My contributions

In Schreiber et al. (2015), I stacked the deepest Herschel images to characterize the correlation between the SFR and the stellar mass, expanding it towards more distant and otherwise fainter galaxies than previous studies. Because the concept of the Main Sequence relies on the tightness of this correlation, I developed a novel stacking technique to measure the scatter in SFR at fixed mass. Crucially, this study was based on the most complete sample of distant galaxies obtained with to the Hubble near-IR imaging in the CANDELS fields, allowing me to draw robust conclusions on the galaxy population as a whole, for the first time up to z=4.

Using this data set, I found that the Main Sequence was already in place at z=3 with a constant scatter, and deduced that more than 66% of the present day stars were formed in galaxies evolving on this sequence. This definitively showed that the secular mode of galaxy growth introduced above was the dominant mechanism for star formation in galaxies. In contrast, starbursting galaxies that do not belong to the sequence remain rare, and contribute only 10% of the star formation activity at all times, suggesting that mergers never play a significant role.

Beyond z=3, Herschel lacked in sensitivity to make any strong claim. For this reason we build a complementary survey with ALMA to observe a hundred massive galaxies at z=4 (PI: R. Leiton). In Schreiber et al. (2017a), I show that these galaxies also form a Main Sequence of tight scatter, therefore extending our conclusions up to 1.5 billion years after the Big Bang.

With this precise determination, the Main Sequence displays an interesting evolution through time, not only in overall normalization (distant galaxies were forming more stars at fixed stellar mass), but also its shape is found to vary. While the Main Sequence is linear at high redshift, starting from z=2 it becomes progressively sub-linear at the massive end. In other words, the most massive galaxies have a reduced specific SFR (the ratio of SFR to the stellar mass). This indicates that these galaxies progressively became less active, and was independently confirmed using Spitzer data (Whitaker et al. 2014).

In Schreiber et al. (2016), I analyze the stellar profiles of these galaxies and show that this is not caused by the growth of a quiescent stellar population in a bulge. Instead, thanks to the well sampled dust spectrum coming from the Herschel data, I measure the gas content in these galaxies and find that this reduced SFR is caused by a reduction of the efficiency with which the galaxy uses its gas to form stars. The mechanism responsible for this “downfall” of efficiency remains unknown, but suggests nonetheless that there exists a scenario in which the cessation of star formation in galaxies (which is discussed further below) happens on long timescales, in contrast to the quasi instantaneous quenching thought to be caused by super-massive black hole activity.

Related articles:

Dusty and massive galaxies in the early Universe


In the previous section I have discussed how galaxies grow with time. In parallel to this field of study, one of the most important question that remains unanswered is: how do galaxies stop growing?

By looking at galaxies close to our Milky Way, it is clear that they can be separated in two main populations: star-forming blue galaxies, and quiescent red galaxies (Baldry et al. 2004). The latter are called “red” because they are mostly composed of old stars, which have colder temperatures and thus have a redder spectrum. This imply that they have stopped forming stars up to several billion years ago, and why this happened is still debated.

There are several important differences between these two populations. Firstly, quiescent galaxies are mostly massive galaxies, with a typical mass close to that of the Milky Way. Star-forming galaxies, on the other hand, span a much broader range. They can have very low masses (down to that of dwarf galaxies, a thousand times less massive than the Milky Way), and can reach large masses too, but the most massive galaxies are typically quiescent (see Baldry et al. 2012).

This is in the local Universe though. As we look backwards in time, this picture changes. At higher redshifts, massive galaxies also had red colors (Franx et al. 2003), which could suggest this whole population has formed even earlier than expected. But observations with Spitzer showed that this was not always true: half of these red galaxies are red for a completely different reason: dust obscuration (Papovich et al. 2006). This means part of these red galaxies are actually forming stars vigorously. In the absence of mid- or far-IR observations, determining if a galaxy is red because it contains old stars or because it is obscured by dust is not easy. Selection criteria have been developed to this very end, and have proven mostly successful in a statistical sense (Williams et al. 2009), but this “age-dust” degeneracy makes it difficult to determine their star formation activity and star formation histories.

One thing is clear though: regardless of the reason why, massive galaxies are red. This means they are observationally challenging to detect: most of their light will be seen in the near-IR where instruments are less sensitive. Furthermore, the widely used Lyman-break selection of high-redshift galaxies (Steidel et al. 1996), which relies on observing the Lyman break in the far-UV regime, is strongly biased against these red galaxies. For this reason their numbers have often been underestimated (Glazebrook et al. 2004).

Yet these galaxies are crucial to understand the whole picture of galaxy evolution and the physics governing galaxy growth. This is true for two reasons. First, the maximum mass a galaxy can reach at a given epoch of the Universe places constraints on how efficient star-formation can be. Models with too strong or too weak feedback mechanisms (i.e., radiation from young stars and supernovae preventing future star formation) will produce too few or too many massive galaxies, for example (Wilkins et al. 2013). Second, because all the quiescent galaxies are massive, understanding massive galaxies is a necessary first step to understand why star-formation stops. For example, it has been found that most massive galaxies host super-massive black holes, which have the potential to wipe out or heat the gas in their host galaxy (Cattaneo et al. 2009) and may be the reason why galaxies “quench” their star formation.

The more backward in time we look, the more interesting things become. Indeed, at higher redshifts, the age of the Universe was smaller, and there was less time for everything to happen. If a massive galaxy is observed in the present day, it could have grown slowly and steadily since the Big Bang, or it could have formed rapidly at essentially any time and then stopped growing. In contrast, in the young Universe, the options are more limited. A massive galaxy observed at a redshift z=2 only had 3 billion years to grow instead of 13, meaning it must have formed relatively fast (Kriek et al. 2009).

My contributions

To be published!

Related articles: