Star Formation Histories of Ultra-Faint Dwarf Galaxies: environmental differences between Magellanic and non-Magellanic satellites?

We present the color-magnitude diagrams and star formation histories (SFHs) of seven ultra-faint dwarf galaxies: Horologium 1, Hydra 2, Phoenix 2, Reticulum 2, Sagittarius 2, Triangulum 2, and Tucana 2, derived from high-precision Hubble Space Telescope photometry. We find that the SFH of each galaxy is consistent with them having created at least 80% of the stellar mass by $z\sim6$. For all galaxies, we find quenching times older than 11.5 Gyr ago, compatible with the scenario in which reionization suppresses the star formation of small dark matter halos. However, our analysis also reveals some differences in the SFHs of candidate Magellanic Cloud satellites, i.e., galaxies that are likely satellites of the Large Magellanic Cloud and that entered the Milky Way potential only recently. Indeed, Magellanic satellites show quenching times about 600 Myr more recent with respect to those of other Milky Way satellites, on average, even though the respective timings are still compatible within the errors. This finding is consistent with theoretical models that suggest that satellites' SFHs may depend on their host environment at early times, although we caution that within the error bars all galaxies in our sample are consistent with being quenched at a single epoch.

1. INTRODUCTION Ultra-faint dwarf (UFD) galaxies are interesting and peculiar objects, representing many extremes in terms of galaxy properties. Their population includes the least luminous, least chemically-enriched, most dark matter (DM) dominated, and oldest satellite galaxies of the Milky Way (MW; Simon 2019). UFDs could be the relics of the first galaxies believed to form, and therefore provide us with a fossil record of the conditions for star formation in the era of reionization (Bovill & Ricotti 2009 and references therein). According to the pre-Based on observations obtained with the NASA/ESA Hubble Space Telescope at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy under NASA Contract NAS 5-26555.
vailing Lambda Cold Dark Matter (ΛCDM) cosmological scenario, an important channel for mass growth of DM halos is hierarchical accretion. Indeed, simulations show that even low-mass host halos have substructures down to their resolution limit (Wetzel et al. 2016;Dooley et al. 2017;Besla et al. 2018;Jahn et al. 2019;Wang et al. 2020). UFDs would also undergo hierarchical growth, and are thus precious tools to study the physics of galaxy assembly in the early Universe.
Recent simulations have illustrated in detail that star formation (SF) in UFDs is impacted by the local background UV ionizing field and stellar feedback, with both acting as effective quenching mechanisms at such lowmass scales (e.g., Jeon et al. 2017Jeon et al. , 2021Wheeler et al. 2019;Applebaum et al. 2021). The timescales in these simulations are such that shortly after the epoch of reion-ization ends (z ∼ 6), UFDs are effectively quenched, although the exact duration of SF can vary depending on the halo mass, even for a fixed ionization background (Jeon et al. 2017), and often some residual interstellar medium remains in the galaxies and can fuel SF for another 1 -2 Gyr (Wheeler et al. 2019). Analyses of zoom-in simulations of UFDs in MW-like environments have found that traditional environmental effects (e.g., tidal field, ram pressure) are not primary factors in the quenching timescale (Applebaum et al. 2021).
Observational data support a ubiquitous quenching timescale for UFDs around the time of reionization (Brown et al. 2014;Weisz et al. 2014). Given that the MW accreted these systems at different times (Fritz et al. 2018), this indicates a global rather than a local physical explanation for the common quenching timescale. However, other studies (e.g., Joshi et al. 2021) have shown a dependence of the quenching timescale on host mass, presumably because the strength of the background and local UV field changes depending on the environment (more massive hosts would have more SF and thus produce more ionizing radiation). However, this study explores the mass range 10 7−10 M , thus their results might not hold for the lower-mass galaxies we are interested in.
UFDs are hard to identify due to their low luminosities M V > −8, implying stellar masses M 10 4 M , and generally old (> 10 Gyr) stellar populations (Simon 2019), which lack bright young stars that would ease their discovery and identification. However, a great effort has been made in the past few years to increase the statistics of satellites around the MW, and many new UFDs were discovered thanks to wide-field surveys such as PAN-STARRS (Laevens et al. 2015), DES Drlica-Wagner et al. 2015;Koposov et al. 2015), and ATLAS (Torrealba et al. 2016).
Many of these new satellites were found in the proximity of the Magellanic Clouds (MC), a region targeted by several deep imaging surveys (e.g., Drlica-Wagner et al. 2015;Martin et al. 2015;Nidever et al. 2017;Koposov et al. 2018;Torrealba et al. 2018). These surveys provide a great opportunity to test the self-similarity of ΛCDM, which predicts that MW satellites, such as the Large Magellanic Cloud (LMC), should also have their own satellites, which fell into the MW potential with the whole Magellanic system (e.g., D'Onghia & Lake 2008; Sales et al. 2011;Deason et al. 2015;Jahn et al. 2019). One way to test association is to reconstruct their 3D kinematics and orbital history, which is now possible thanks to detailed proper motion (PM) measurements enabled by the Gaia mission. Kallivayalil et al. (2018) analyzed the PMs and radial velocities of 13 UFDs using Gaia data release (DR) 2, compared their kinematics to the tidal debris of a simulated analog of the LMC, and found four UFDs whose kinematics are compatible with the LMC debris, Carina 2, Carina 3, Horologium 1, and Hydrus 1. Using a different technique, Erkal & Belokurov (2019) used Gaia DR2 PMs to rewind the satellite orbits from their present day positions and determine which ones were originally bound to the LMC; of the 25 analyzed UFDs they concluded that six, Carina 2, Carina 3, Horologium 1, Hydrus 1, Reticulum 2, and Phoenix 2, are highly compati-ble with a Magellanic origin. Recently, Patel et al. (2020) calculated the orbital histories of 13 ultra-faint satellites including the combined potential of the MW, LMC, and Small Magellanic Cloud (SMC) for the first time; in this scenario, they identified Carina 2, Carina 3, Horologium 1, and Hydrus 1 as long-term Magellanic satellites, and Reticulum 2 and Phoenix 2 as recently captured Magellanic satellites.
Although these works provide fundamental contributions to our understanding of the MW satellites' dynamics, we need star formation histories (SFHs) to fully explore their properties and understand the impact of environment and reionization on such low mass galaxies. Within the Local Group, the Hubble Space Telescope (HST ) is able to resolve individual stars in galaxies down to several magnitudes below the oldest main sequence (MS) turnoff, allowing us to measure their ancient SFHs and explore differences in the SF quenching behavior (as in, e.g., Brown et al. 2014). A detailed analysis and comparison of UFDs residing in different environments at early times is also one way to discover variations in the ionization field over large scales.
Here we present an analysis of the optical colormagnitude diagrams (CMDs) and SFHs of seven UFD galaxies part of the HST Treasury Program 14734 (PI: Kallivayalil). They are listed in Table 1, together with their distances, in the range ∼ 30 − 150 kpc, V -band magnitudes, between −1.8 and −5.2, and possible association with the Magellanic Clouds. Our analysis is based on high-precision photometry from the Advanced Camera for Surveys (ACS) and literature spectroscopic measurements, and employs the synthetic CMD method to derive the SFH of each galaxy.

DATA AND PHOTOMETRY
Observations of a total of 30 UFDs were performed using the F606W and F814W filters of the HST ACS Wide Field Channel (Treasury Program 14734; PI: Kallivayalil). The basic observing strategy included collecting four dithered 1100 s exposures in both filters for each target.
The images were processed through the current ACS pipeline, which corrects for charge-transfer inefficiency (CTI), and the separate dithers were combined using the DRIZZLE package to create the drc files. The CTIcorrected separate dither images, or flc files, were also used to create the photometric catalogs. The photutils routines DAOStarFinder and aperture photometry were used to detect sources and to calculate the flux inside circular apertures of both four-and six-pixel radii. We imposed two separate criteria to create a flag differentiating real sources from artifacts. First, we tracked whether a given source's magnitude in the four-and sixpixel radii was within 1.5 standard deviations of the median difference across all sources. The second criterion was that the magnitude difference between the two radii must be positive. Sources meeting both conditions were flagged as real.We performed encircled energy corrections and converted the flux to STMAG. The four-pixel radius drc magnitudes for each matching source between filters were used in the final photometric catalogs. Lastly, magnitudes were converted to the Vegamag photometric system. Table 1 Properties of the seven UFD galaxies analyzed here. Columns 1 and 2 list the galaxy names, columns 3 and 4 the galactocentric coordinates, columns 5 and 6 the distance from the Sun (in kpc) and distance modulus (Fritz et al. 2018(Fritz et al. , 2019, and references therein), column 7 the number of stars used in this study, column 8 the estimated stellar mass (Sales et al. 2017 and references therein; for cases without estimates of M , it was derived from the listed V -band magnitude assuming a mass-to-light ratio Υ = 2 in solar units), column 9 the V -band magnitude, column 10 the V -band extinction, and column 11 the average metallicity (Fritz et al. 2018(Fritz et al. , 2019, and references therein). The last column indicates whether a galaxy is a potential Magellanic satellite, according to the analysis in Erkal & Belokurov (2019) and Patel et al. (2020).  The right y-axis shows the range of apparent magnitudes spanned by each galaxy. The foreground/background contamination will be accounted for in the CMD modeling phase.
Sources in the flc images underwent the same steps and were matched across the four separate dithers in each filter using a 6-parameter linear transformation. To derive an empirical error in the mean magnitude for the sources, we used the sigma-clipped standard deviation of the fourpixel radius flc magnitudes for each source in the separate filters. The flc sources were then matched between the filters before being matched to the drc sources using the same 6-parameter transformation method.
We analyze here the UFDs in the program with enough stars (at least 100) to perform a reliable CMD fit; for less populated CMDs, random uncertainties due to stochastic sampling of the CMD become critical and affect the reliability of the fitting technique.
We include Sag 2 in our sample, though there is conflicting evidence of it being either a UFD (Longeard et al. 2020) or a globular cluster (Mutlu-Pakdil et al. 2018;Longeard et al. 2021); spectroscopic follow-up observations would be the ultimate confirmation of its true nature.
3. COLOR-MAGNITUDE DIAGRAMS Figure 1 presents the M F606W vs. M F606W − M F814W CMDs of the UFD galaxies analyzed here. We show absolute magnitudes, derived according to the appropriate distance and extinction (as in Table 1), to ease the comparison among galaxies (apparent magnitudes are on the right y-axis). The gray box shows the approximate location of the horizontal branch (HB), the core He-burning evolutionary phase of stars with masses 2 M and low metallicity ([Fe/H] < −1.5), while the red line is a reference isochrone with age = 13.7 Gyr and [Fe/H] = −2.0 from the MESA/MIST library (Dotter 2016). These CMDs are dominated by an ancient metal-poor population, and we can identify the oldest MS turnoff at M F606W ∼ 3 and M F606W − M F814W ∼ 0.5, which is our most sensitive and reliable age constraint. Using the isochrone as a guide, we also notice how the color, color spread, and turnoff morphology vary slightly from galaxy to galaxy, indicating their different SF and chemical evolution histories.
A few galaxies, in particular Hya 2, and Sag 2, show an extended HB, reaching very blue colors, while Hor 1 and Phx 2 may have a single blue HB candidate. The current empirical and theoretical evidence indicates that the HB morphology is affected by the initial metal content, with more metal-poor populations showing bluer HBs. However, there might be other, more complicated parameters affecting the HB morphology, and the modeling of this phase is still subject to great uncertainties (see, e.g. Torelli et al. 2019).
There are a few blue straggler stars in some of the CMDs, extending at bluer colors and brighter magnitudes than the dominant MS turnoff. This is particularly evident in Phx 2 and Sag 2. These stars are very common in ancient populations (Santana et al. 2013), but they can mimic a much younger sub-population.
For example, the turnoff mass at 12 − 13 Gyr is ∼ 0.8 M , but blue stragglers can be even twice as massive, resembling a MS population of ∼ 2 Gyr. This is why we exclude them from the CMD fitting (see Section 4).

STAR FORMATION HISTORY DERIVATION
To date, the most powerful approach to recover a SFH from an observed CMD is the comparison with models, a method applied by many different groups (see, e.g., the review by Tolstoy et al. 2009). Synthetic CMDs are built from a set of stellar models (evolutionary tracks or isochrones) adequately treated to match the distance, extinction, and photometric properties of the galaxy under analysis. To add the right scatter to the models, we adopt a scattering function based on fitting the photometric errors generated by the photometry pipeline (see Section 2). Each synthetic CMD represents a simple stellar population of fixed age and metallicity, and a linear combination of these creates a composite population that can represent, with the appropriate weights, any SFH. The best-fitting weights are determined by using a minimization algorithm to compare data and models. To take into account the effect of field contamination, we include an additional component built from the Besançon Galaxy model (Robin et al. 2003) along the line of sight to each galaxy (as in, e.g., Brown et al. 2014).
We adopt Poisson maximum likelihood statistics to perform the minimization of the residuals between data and models, to accommodate the fact that some parts of the CMD might have a low number of stars. We implemented the construction of the synthetic CMDs and the comparison between models and data in the hybrid genetic code SFERA (Cignoni et al. 2015).
To take into account possible systematic effects due to the adopted stellar evolution models, we derive SFHs using two different sets of models, the Victoria-Regina library (VandenBerg et al. 2014) and the MESA/MIST library (Dotter 2016). While the MESA/MIST models are available only for scaled-solar abundances, we adopt the Victoria-Regina models with an enhancement of +0.4 for α-elements, which is more appropriate for old metalpoor populations. While this should represent a good approximation for UFDs, some variations are observed from galaxy to galaxy, which might introduce some uncertainty in the ages estimates; however, we expect them to affect the SFHs at a much smaller level than the statistical and systematic uncertainties already taken into account in our analysis. Moreover, while various systematic uncertainties certainly affect our absolute age estimates, one of the main goals of this paper is to do a relative comparison between ages, which is more robust as systematics will affect all SFHs in the same way.
We use an isochrone grid with [Fe/H] in the range [−1, −4] with 0.1 dex steps (using literature spectroscopic measurements as a further constraint), and ages in the range [8, 13.7] Gyr with 100 Myr steps. We assume a Kroupa (2001) IMF and 30% of binaries drawn from the same IMF (Spencer et al. 2017; we checked that very little differences are found when different fractions or mass ratios are adopted). Distance moduli and extinctions are taken from the literature (Table 1), but the code allows for small variations in these parameters to maximize the likelihood.
We fit the CMD sequence from the MS turnoff to the top of the sub-giant branch (see the box in Figure 2), to focus on the region most sensitive to age variations, while at the same time avoiding areas which are poorly constrained by the models. Also, excluding the lower MS minimizes the sensitivity of the fit to the assumed IMF. Moreover, we avoid blue stragglers, which might mimic a much younger population as discussed in Section 3. At the faintest end of the SFH derivation the observations are near 100% complete, and the typical magnitude error is 0.01 − 0.02 mag, depending on the galaxy.

RESULTS AND DISCUSSION
An example of the output from SFERA is shown in Figure 2 for Sag 2. Panels a) and b) show the Hess diagrams of the observed and recovered CMDs (the latter based on MIST models), while panel c) shows the residuals between the two; the black box corresponds to the MS turnoff area, used as a constraint for the SFH derivation. Panel d) shows the recovered cumulative SFHs (i.e., the fraction of total stellar mass formed prior to a given epoch) from both the MIST models (in red) and the Victoria-Regina models (in blue); the gray band indicates the mid-point reionization redshift with its uncertainty, z re = 7.7 ± 0.7 (Planck Collaboration et al. 2020).
Comparing observed and synthetic CMDs demonstrates how well we reproduce the data, even when fitting only the MS turnoff area. We can also reproduce the luminosity of the HB, even though the models do not reach the most extreme blue colors; this is one of the well-known complications in HB morphology modeling. Figure 3 (left panel) shows the cumulative SFHs for the galaxies we analyzed, with the same reference to reionization as in Figure 2. As expected, the SFHs are predominantly old, consistent with the galaxies having created at least 80% of their stellar mass by z ∼ 6 (in agreement    with, e.g., Bose et al. 2018). Table 2 summarizes the time at which 50% of the total stellar mass was formed (τ 50 ) and the time at which 90% of the total stellar mass was formed (τ 90 , quenching time, as in, e.g., Weisz et al. 2015Weisz et al. , 2019 by each galaxy. We adopt τ 90 as a proxy for the quenching time to minimize possible residual contamination from blue stragglers and to ease the comparison of our results with the literature. All galaxies in our sample have quenching times older than 11.5 Gyr ago. The middle and right panels of Figure 3 show the same results divided in two sub-samples: candidate Magellanic satellites and non-Magellanic satellites, according to the results by Kallivayalil et al. (2018), Erkal & Belokurov (2019), Fritz et al. (2019), and Patel et al. (2020. For each panel, the black solid curves also show the errorweighted average SFH of the two sub-samples, while the dashed gray line is the average of the other sub-sample. On average, we find τ 90 = 12.06 ± 0.72 Gyr ago for Magellanic satellites and τ 90 = 12.68 ± 0.23 Gyr ago for non-Magellanic satellites (this becomes 12.67 ± 0.41 if we exclude Sag 2); even though they are comparable within the errors, there is a marginal difference between these values (a two-sample KS test of the average distributions gives a p-value of 5.7×10 −10 ). Indeed, the candidate Magellanic satellites appear to have more prolonged SFHs compared to non-Magellanic ones, possibly due to the fact that they should have resided in low-density environments for most of their lifetimes; however, the uncertainties are too large to reach a definitive conclusion.
Another interesting consideration is that, among the non-Magellanic satellites, the one with the youngest SFH is Hya 2, i.e., the most distant one in the sample (∼ 150 kpc, whereas all others lie within ∼ 80 kpc), and the most massive one, which is also likely on first infall into the MW's halo (Patel et al. 2020). However, distance makes the SFH uncertainties quite large, so these results should be regarded with caution. The shape of the SFH is also different for the two sub-samples, with non-Magellanic satellites reaching 100% of the stellar mass assembly within 11.5 Gyr ago, i.e. z ∼ 3, while MC satellites show more prolonged star formation and reach the same point only 8.5 Gyr ago.
Similarly, in a study comparing the quenching times of M31 and MW satellites, Weisz et al. (2019) found that Table 2 Summary of the time at which 50% of the total stellar was mass formed (τ 50 ) and the time at which 90% of the total stellar mass was formed (τ 90 , quenching time) for each galaxy. The list is sorted by quenching time (from oldest to youngest). The last column indicates whether a galaxy is a potential Magellanic satellite (Erkal & Belokurov 2019;Patel et al. 2020 These results, together with the extended SFHs we find for MC satellites, might support the idea that satellites of low-mass hosts experience a weaker ionization field, resulting in a more prolonged SFH than if they were satellites of more massive hosts, as found by, e.g., Joshi et al. (2021). On the other hand, it is important to be mindful of our uncertainties, and be cautious about over-interpreting these results. Within the error bars, all galaxies in our sample are compatible with being quenched by reionization, as found in other works (e.g., Brown et al. 2014;Weisz et al. 2015;Tollerud & Peek 2018;Fillingham et al. 2019).
From the numerical point of view, simulations reaching the necessary resolution for UFDs have mainly examined field galaxies (e.g., Jeon et al. 2017;Wheeler et al. 2019), and find that reionization and SNe feedback are key factors to their quenching. More recently, Applebaum et al. (2021) presented a suite of cosmological simulations that contain zoom-ins of MW-like galaxies, with simulated satellites extending to the UFD regime. By comparing UFD satellites to UFDs in the field, the authors found that reionization and feedback were indeed the main quenching mechanisms, rather than environmental effects (it is worth noting that these simulations lack radiative feedback from the host galaxies and adopt a uniform reionization model). However, studies examining the SFHs of UFDs around LMC-like dwarfs are not available, yet.
Despite the challenges, studies of the kind presented here are fundamental to fully explore the properties of low-mass galaxies, and to understand the impact of environment and reionization on such systems. A detailed analysis and comparison of the SFHs of UFDs residing in different environments at early times is also a way to discover variations in the ionization field over large scales. The observations presented here will also be an important baseline for follow-up imaging with HST and the James Webb Space Telescope, to measure both bulk PMs and internal motions, while spectroscopic surveys will improve metallicity constraints (thus allowing also more precise age derivations) on these very interesting galaxies.