What Are Those Tiny Things? A First Study of Compact Star Clusters in the SMACS0723 Field with JWST

In this work, we focus on compact stellar clusters in the galaxy cluster environment SMACS J0723.3–7327 at z = 0.39 (07^h23^m133, −73^d27^m25^s). SMACS0723 was initially discovered as part of the southern extension to the ROSAT All-Sky Survey (Voges et al. 1999) based Massive Cluster Survey (MACS; Ebeling et al. 2001) and also detected by Planck (Planck Collaboration et al. 2011) through the Sunyaev–Zel'dovich effect. It was then subsequently studied by the Reionization Lensing Cluster Survey (RELICS) using the Hubble Space Telescope (HST; Coe et al. 2019). A new study using JWST by Mahler et al. (2022) places the total mass of the cluster at M_{<400 kpc} = 3 × 10¹⁴ M_⊙ (compared to the measurement by Planck of M₅₀₀ = 8 × 10¹⁴ M_⊙). The BCG has a stellar mass of ∼2 × 10¹¹ M_⊙ (close to the "knee" of the galaxy stellar mass function; e.g., Davidzon et al. 2017) and resembles a relatively featureless elliptical galaxy with little to no ongoing star formation. Here, we use new JWST observations of the cluster (ERO program ID 2736; Pontoppidan et al. 2022) taken with the Near Infrared Camera (NIRCam; Rieke et al. 2005) broadband filters F090W (0.90 μm), F150W (1.50 μm), F200W (1.99 μm), F277W (2.76 μm), F356W (3.57 μm), and F444W (4.41 μm). These filters correspond to wavelengths in the cluster's rest frame of 0.65, 1.08, 1.43, 1.98, 2.57, and 3.17 μm, respectively. NIRCam consists of two modules, each observing a $2\buildrel{\,\prime}\over{.} 2\,\times \,2\buildrel{\,\prime}\over{.} 2$ FoV, one centered on the cluster's BCG and the other one centered $\sim 3^{\prime}$ (∼1 Mpc at the cluster's redshift) to the southwest. The data were downloaded from the Mikulski Archive of Space Telescopes (MAST) ⁴ and subsequently reduced with the JWST data reduction pipeline. The reduced image data was then combined using the grism redshift and line analysis software for space-based spectroscopy (Grizli; Brammer & Matharu 2021) package to a final pixel scale of 002 px for F090W, F150W, and F200W and 004 px for F277W, F356W, and F444W, respectively (Brammer 2022). Note that the recent jwst_0942.pmap photometric calibration reference file was used with modifications as of 2022 September 5 as described in the Grizli GitHub repository. ⁵ The details of the data reduction will be presented in detail in an upcoming paper (G. Brammer et al. 2022, in preparation). The field was also imaged by JWST/MIRI at mid-infrared wavelengths. However, the MIRI data are too shallow for this study and are therefore not being used. Figure 1 shows an overview of the data used.

Zoom In Zoom Out Reset image size

In this work, we study the faintest, compact stellar clusters in the SMACS0723 field, such as potential globular clusters around the cluster host galaxies. These sources are selected manually mainly on the F200W image; however, we require them also to be detected in at least two other bands in order to reject potential spurious sources. In addition, the sources are selected to be unresolved point sources as well as faint (to exclude Milky Way stars; see Section 3.5). For example, a good upper limit in size of globular cluster is ∼10 pc (e.g., Larsen et al. 2001), which, at z = 0.39, corresponds to an angular size of ∼1.4 mas (note that there are some larger globular clusters, for example NGC 2419 in the Milky Way and Lindsay 1 in the Small Magellanic Cloud). This is significantly smaller than the point-spread function (PSF) size (between 20 and 120 mas at 1–5 μm), thus motivating the point-source assumption. Milky Way stars are the most likely contaminants in this selection, as discussed in Section 3.5. We tried an automated selection of globular clusters; however, we found that this approach results in a severely incomplete selection due to blending with the BCG and other cluster galaxies, as well as the variation of the intracluster and scattered light across the field. In total, we identified 178 compact star clusters in the full FoV of NIRCam around the cluster host galaxies (see Figure 1), which we characterize and study in the following. The coordinates and NIRCam fluxes of the extracted star clusters are listed in Table 1. We note that our selection is not complete, but serves as a starting point to understand the properties and origin of these faint sources. We refer to the work by Lee et al. (2022) for a detailed study of the distribution and number density of these star clusters in the ICM of SMACS0723. If these sources are globular clusters, we would expect to only see the tip of the iceberg (i.e., the brightest ones) even at the unprecedented depths of these observations, assuming the commonly used Gaussian globular cluster luminosity function (e.g., Alamo-Martínez et al. 2013).

To measure the photometry of unresolved point sources, a robust determination of the PSF is crucial. Here, we measure the PSF in all six NIRCam filters by stacking stars on the full FoV of the available observations. For the selection of stars, we use an identical approach as in Faisst et al. (2022). Summarizing, the stars are selected based on the R_e versus magnitude diagram (produced by the flux_radius and flux_auto quantities derived by SExtractor, Bertin & Arnouts 1996, run the NIRCam images). We note that, based on the upturn in size measurements on that diagram, point sources at ∼25 mag already start to enter the nonlinear regime toward saturation at the depth of these NIRCam observations. We therefore require stars between 25 and 27 mag, to avoid the aforementioned saturation effect, as well as to exclude potential faint, spurious sources. In addition, a half-light radius of less than 3 pixels is required to select unresolved sources. For each of the 58 stars, we create a 2'' × 2'' cutout and subsequently center the stars before stacking them to obtain the final PSF. The final PSF for each filter (normalized and of the same pixel scale as the images; see Section 2.1) are available for download. ⁶

The relatively low number density of stars does not allow us to quantify variations of the PSF. However, the recent study by Nardiello et al. (2022) measured PSF variations across the NIRCam FoV based on stars in the globular cluster M92, finding variations in the FWHM of up to 15%–20%. Such variations have a negligible effect on the measured photometry in our case, but we will include this effect later when we constrain the sizes of our compact star clusters.

The photometry of the selected globular cluster is measured using the software Tractor ⁷ (Lang et al. 2016a, 2016b; J. Weaver et al. 2022, in preparation), which performs a prior-based forced photometry including the PSF of the observations. We first compared the one-dimensional projected light profiles from the stack of all compact star clusters to the one of the PSF (left panel of Figure 2). This test shows that the star clusters are indeed consistent with unresolved point sources, hence we use the point-source fitting option of Tractor. In the following, we fit the photometry of individual sources to study the variations in the stellar population across our sample. We also derive the photometry from the stacks of the sources in each of the six NIRCam filters to study the average properties. In both cases, we create cutouts of 08 size of all the sources. The stacks were created by first performing a local background subtraction (measured in a sky annulus with an inner radius of 5 px and an outer radius of 10 px) on the individual cutouts and then combining them via median stacking. The following steps are then identical for the fitting of individual sources and the stacks. Tractor is run on the 08 cutouts applying a point-source model, using the F200W position of the sources as prior (determined from the zeroth-order moment of light on the source). During the fitting, the position is free to vary within ±2 pixels of the prior position. This helps to mitigate potential misalignments between the different image products. In addition, we let Tractor fit and subtract the (slowly varying on 08 scales) local background at the position of the source to remove the contribution of the halo from nearby cluster galaxies and the intracluster light. Note that for that reason, simple aperture photometry will result in overestimated flux measurements. The inverse variance image (used for error estimation) is created as 1/σ² with σ derived from σ-clipping statistics on a large region of the image not contaminated by the intracluster light and other galaxies. The variance output by Tractor is used to obtain the uncertainties on the flux measurements.

Zoom In Zoom Out Reset image size

Figure 3 shows the fits to the stacked cutouts of the compact star clusters in the six different bands. The point-source assumption is also validated by the residuals, which are consistent with zero. The NIRCam six-band photometry derived for the stack is listed in Table 2.

Zoom In Zoom Out Reset image size

To study the properties of the compact star clusters, we fit simple stellar population (SSP) models to the photometry of the median stacks obtained in Section 2.4. We assume that the star clusters have been formed in a single burst, thus are well described with an SSP model. Given the low sampling in wavelength of these data, we think that this is a good approximation as multiple stellar populations would be difficult to tell apart if they exist. Before fitting, we need to correct the extracted flux densities for Milky Way dust extinction, which we obtain from the Schlegel et al. (1998) dust maps ⁸ recalibrated by Schlafly & Finkbeiner (2011). The dust extinction toward the SMACS0723 is significant at an E(B − V) ∼ 0.2 mag level, which results in a wavelength-dependent extinction of ∼0.25 mag at 0.9 μm, ∼0.07 mag at 2.0 μm, and less than 0.03 mag at 4.4 μm assuming the Fitzpatrick (1999) dust extinction curve. We use these values to correct the measured flux densities of the star clusters. For our fiducial SSP models, we chose "Padova 1994" stellar evolutionary tracks (Bertelli et al. 1994) together with the Basel Stellar Library (BaSeL, version 3.1; Lejeune et al. 1997, 1998; Westera & Buser 2003) complemented by the empirical STELIB library (Le Borgne et al. 2003) ⁹ at wavelengths blueward of 9500 Å (see Bruzual & Charlot 2003). In the following, we assume a Chabrier (2003) IMF. These fiducial SSP models were created using the software GALAXEV ¹⁰ in a grid of different ages and stellar metallicities. Specifically, the age grid ranges from 0.5 to 11 Gyr in steps of 0.5 Gyr and we use six different stellar metallicities of 0.005, 0.02, 0.2, 0.4, 1, and 2.5 Z_⊙. We compared the results using these fiducial models to other SSP models as well. Specifically, we compared to the MIST stellar tracks (Choi et al. 2016; Dotter 2016), based on the MESA isochrones (Paxton et al. 2011, 2013, 2015, 2018), ¹¹ assuming a BaSeL stellar library. These additional SSP models are created using the Flexible Stellar Population Synthesis code (FSPS; Conroy et al. 2009; Conroy & Gunn 2010) ¹² for a Chabrier IMF and using the same age and metallicity grid. Generally, we find that wavelengths blueward of ∼1 μm rest frame are most affected by the choice of these different models. We will comment on the differences in the results in Section 3. The fit to the various models was performed by χ² minimization using the SciPy least-squares Python package. We note that we fixed the redshift of the compact star clusters to the one of the galaxies in the SMACS0723 cluster (z = 0.39). In Section 3.1.1 we will show that this is indeed a good assumption based on the similarity in colors of the star clusters and galaxies at z = 0.39. Also, small changes in redshift (e.g., due to the kinematics in the cluster environment) do not have an impact on the fitted properties.

3.1.1. Photometry and Colors

Figure 4 shows the observed photometric properties of the extracted compact star clusters in SMACS0723. As shown on the left panel, the brightness of the star clusters ranges from 28 to 30 mag in F090W and 27.5 to 29.5 mag in F200W (left panel). The NIRCam filters cover the (redshifted) 2 μm peak; hence, the F090W and F444W brightness is similar. The middle left panel shows the color distribution of the compact star clusters (purple circles) compared to star-forming (gray) and quiescent (red) galaxies selected in the COSMOS2020 catalog (Weaver et al. 2022) at a redshift z = 0.39 with Δz = 0.2 ¹³ . Due to their SED shape, which is due to the redshifting of the 1.6 μm bump through the bandpasses, the globular clusters reside mostly at red [F090W]−[F150W] ≳ 0 colors and blue [F200W]−[F277W] ≲ 0 colors. This parameter space is similar to quiescent galaxies at the same redshift, suggesting similar stellar populations. However, the star clusters are significantly fainter, 2–4 mag below the detection limits of the COSMOS surveys (right panel).

Zoom In Zoom Out Reset image size

3.1.2. Population-averaged Ages and Metallicities

We note that these star clusters are faint even for JWST and most of the individual star clusters are detected at very low signal-to-noise ratio in blue and red bands where their SED flux drops. We therefore first fit SSP models (see Section 2.5) to the photometry derived from the stacked cutouts to obtain a population-averaged median age and stellar metallicity measurement. The following values for metallicity and age are therefore an average of the studied relatively massive cluster population (see also Section 3.3). Later in Section 3.2 we will fit individual star clusters to study the population variations of these properties.

Figure 5 shows the results from SED fitting using our fiducial SSP models (Padova stellar tracks) to the stacked photometry. The left panel shows a χ² map for the adopted age and metallicity grid with 3σ, 4σ, and 5σ contours indicated. We find a best-fit age of ${1.5}_{-0.5}^{+0.5}\,\mathrm{Gyr}$ and a best-fit metallicity of 0.2–0.3 Z_⊙ (3σ). Interestingly, the χ² map also indicates the presence of a younger (<1 Gyr) and more metal-rich (0.4–1.0 Z_⊙) solution. The right panel of Figure 5 shows the data (black symbols) with the best-fit SED (black line) as well as other models (see legend in figure) for comparison with the same age or metallicity as the best fit.

Zoom In Zoom Out Reset image size

The data exclude with high significance metallicities below 0.2 Z_⊙ as well as solar/supersolar. This is because changes in metallicity affect the colors significantly at a fixed age. For example, compare the 1.5 Gyr model at 0.02 Z_⊙ (gray dotted line) and at 2.5 Z_⊙ (gray dashed line) with the best fit (0.2 Z_⊙) in the right panel of Figure 5.

On the other hand, ages are not constrained well by our current data, especially with the JWST filters available. For example, ages up to 5 Gyr cannot be excluded at better than 4σ level and maximal ages of the universe at z = 0.39 (9.3 Gyr) are possible at 5σ level. This is expected by the lack of blue coverage of the JWST filters, and in addition the relation between color and age only changes weakly for SSPs with ages older than ∼3 Gyr. For example, the [F090W]–[F200W] color (corresponding to rest-frame (r − H) color) changes by 0.4–0.6 mag (depending on metallicity) for ages <3 Gyr (caused by hot massive stars), while it only changes by ∼0.2 mag thereafter. We note that a better sampling of the observed 0.5–1.5 μm regime could constrain the older ages better (see Figure 5).

In addition, we quantify the reliability of these results by comparing to fits using the MIST isochrones as discussed in Section 2.5. We find that the models differ significantly at wavelengths blueward of ∼1 μm rest frame for low metallicities and low ages. However, we find overall that our results are robust against the choice of different model parameterizations. Specifically, using the MIST isochrones, we find consistent ages of ${1.5}_{-0.5}^{+0.5}\,\mathrm{Gyr}$ and a best-fit metallicity of 0.4 Z_⊙ with a slightly larger 3σ range of 0.2–0.4 Z_⊙. Similarly to the other models, metallicities of <0.2 Z_⊙ as well as solar metallicities are robustly excluded.

3.1.3. Stellar Masses

From the best-fit SSP models obtained for each individual star cluster, we derive a V-band mass-to-light (M/L) ratio from which we can estimate the total stellar masses of the compact star clusters. For the best-fit SSP model fit to the stacked photometry, we derive a M/L ratio of $0.45\,{M}_{\odot }\,{L}_{\odot }^{-1}$. The M/L ratios of the individual star clusters range from 0.2 to 4.8 $\,{M}_{\odot }\,{L}_{\odot }^{-1}$. With the V-band luminosity derived from the NIRCam/F090W filter (which is close to the ∼5500 Å V-band in rest frame), we derive stellar masses for the individual star clusters and we find a median of ${2.4}_{-1.5}^{+3.0}\,\times \,{10}^{6}\,{M}_{\odot }$ (using the M/L ratio derived from the stacked SED would result in a stacked median stellar mass of ${3.9}_{-1.8}^{+3.2}\,\times \,{10}^{6}\,{M}_{\odot }$). As shown in the right panel in Figure 4, the star clusters line up well with quiescent galaxies at z = 0.39 on the K-band magnitude versus stellar mass diagram. This suggests that these compact star clusters have a similar M/L ratio as quiescent galaxies at the same redshift. Assuming the quiescent galaxy M/L relation, we would obtain an average stellar mass of ${3.3}_{-1.1}^{+1.6}\,\times \,{10}^{6}\,{M}_{\odot }$.

We can compare this obtained M/L ratio to what is measured for globular clusters in the Milky Way and local galaxies. The M/L ratios depend on the age (hence mass itself) and metallicity of the stellar population. Given the estimated age and metallicity of our star clusters, we find a possible range in M/L of 1–4 ${M}_{\odot }\,{L}_{\odot }^{-1}$ in rest-frame V-band according to the observations and models in Kruijssen (2008). Note that the M/L ratio increases with age; hence, the upper end of this range would be applicable to old globular clusters, while the lower end is more common in young star clusters. Our measured M/L is consistent with younger ages.

Summarizing, the masses derived for our compact star clusters are on the high end of what is expected for globular star clusters in local galaxies (see also Figure 6). Furthermore, the high-mass end of the derived mass range may overlap with stellar masses estimated for Milky Way dwarf galaxies. To assess this, we compare the size constraints on our compact star clusters with the sizes of local dwarf galaxies.

Zoom In Zoom Out Reset image size

3.1.4. Sizes

As shown in the middle panel of Figure 1, the selected compact star clusters span about 1.5–2 mag in color space, which could be indicative of variations in age and metal properties. In the following, we try to characterize these color variations in terms of variations in these physical parameters.

For this, we fit the same SSP models as described in Section 2.5 to each individual compact star cluster. We note that while the uncertainties of these measurements for individual compact star clusters are rather large due to their faintness, we still are able to investigate statistically trends of age and metallicity in our sample. Figure 7 shows the individual star clusters on the [F090W]−[F150W] versus [F200W]−[F277W] color−color diagram color-coded by the best-fit age (blue scale) and metallicity (red scale). Note that each symbol has two colors assigned (metallicity as edge-color and age as face-color). Light symbols correspond to star clusters with lower ages and metallicities, while darker symbols correspond to star clusters with older ages and higher metallicities. The background shows a Gaussian kernel density estimation (including the uncertainties of the data points) to visualize the most probable density of points. The lines indicate colors derived from the SSP models for different metallicities and ages (running from left to right, dots on lines indicate ages of 0.5, 1, 4, and 9 Gyr). Older models result in redder [F090W]−[F150W] colors, while more metal-rich models result in a reddening of both colors.

Zoom In Zoom Out Reset image size

The range in colors of the individual compact star clusters are coupled with variations in ages and metallicity. This suggests different formation times and possibly different formation processes (causing different metal contents) of the star clusters in the SMACS0723 environment. We also note that the observed color (hence metallicity) distribution may be expected to be affected by the "blue tilt," caused by a lack of the most massive, metal-poor star clusters (see Forbes & Bridges 2010; Usher et al. 2018).

In the following, we compare the BCG and the identified compact star clusters to systems in the local universe.

The BCG of the SMACS0723 cluster has a stellar mass of ∼2 × 10¹¹ M_⊙ and a V-band absolute magnitude of ${M}_{V}^{{Vega}}\sim -23.2$, ¹⁶ which is at the faint end of the brightness distribution of the BCGs studied in Harris et al. (2014; $-24.1\lesssim {M}_{V}^{{Vega}}\lesssim -23.0$). The V-band luminosities of our compact star clusters range between 2.5 × 10⁶ L_⊙ and 3.1 × 10⁷ L_⊙, which is about 1–2 orders of magnitudes brighter than the common turnover luminosity of the globular cluster luminosity function at ∼10⁵ L_⊙ (Larsen et al. 2001; Harris et al. 2014). (Note that the lower limit is due to the sensitivity limit of the JWST observations.) In this luminosity range, the luminosity functions of globular clusters around local BCGs with similar luminosity find about 100 globular clusters (Harris et al. 2014), which is 60% less than what we find in the SMACS0723 cluster field. Several reasons could contribute to this difference. First, the SMACS0723 galaxy cluster is more massive (8 × 10¹⁴ M_⊙ as measured by Planck) compared to the Abell clusters (1–5 × 10¹⁴ M_⊙), which affects the number of globular clusters. For example, as shown in Harris et al. (2014), the luminosity function of globular clusters around the BCG in A3558 (4.7 × 10¹⁴ M_⊙) peaks at 2 times higher number densities as compared to the one of the less massive cluster A1736 (2.8 × 10¹⁴ M_⊙). Second, our selection includes the full ICM (over several 100 kpc; see Figure 1), which may lead to the inclusion of globular clusters around other galaxies in the vicinity of the BCG. Finally, as pointed out in Section 3.1.4, we might also include the tidally stripped compact cores of dwarf galaxies in our sample. Given the above points, the number of compact star clusters is close to what is expected based on local observations.

Note that >50% of our compact star clusters have $\mathrm{log}(L/{L}_{\odot })\gt 6.7$ and thus would be considered as superluminous globular clusters according to Harris et al. (2014). Such superluminous objects could be related to ultracompact dwarf galaxies (or the stripped cores thereof) or bridge the gap between globular clusters and dwarf galaxies. These objects have been found in several rich local galaxy clusters (e.g., Misgeld et al. 2011; Mieske et al. 2012). SMACS0723 is a massive galaxy cluster at these redshifts (approximately a factor of 2 more massive than local galaxy clusters studied in Harris et al.); hence, the large occurrence of such superluminous globular clusters would not be surprising.

Finally, we compare the color distribution of our compact star clusters to the globular clusters around local BCGs from Harris et al. (2014). Figure 8 shows the (B − I)_Vega versus ${M}_{I}^{\mathrm{Vega}}$ color–magnitude diagram for our extracted compact star clusters at z = 0.39 (purple circles) ¹⁷ as well as globular clusters around the local galaxies NGC 7720 (gray) and ESO444-G046 (brown), which are the BCGs of the clusters A2634 and A3558, respectively. The former BCG is brighter (−23.8 versus −23.4 Vega mag) and resides in a more massive cluster (4.7 × 10¹⁴ versus 1.5 × 10¹⁴ M_⊙), which, however, is still 1.7 times less massive than SMACS0723. The compact star clusters studied in this work at z = 0.39 reside at the bright end of the distribution of local globular clusters. Their colors are consistent with the colors of globular clusters in A3558, the more massive one of the two local galaxy clusters shown in Figure 8. This is consistent with the idea that the brightness of the globular clusters scales with the mass of the galaxy clusters.

Zoom In Zoom Out Reset image size

Recent studies by Mowla et al. (2022) and Claeyssens et al. (2022) examined the properties of individual globular star clusters in SMACS0723 field of lensed galaxies at higher redshifts. The significant magnification (up to 100×) due to lensing allows the authors of these studies to probe fainter star clusters in the mass range of 10⁵–10⁷ M_⊙. However, the wavelength range probed by the JWST NIRCam filters is bluer due to the higher redshift of the sources, which may affect how well different properties of the clusters can be constrained with the given photometry. The Claeyssens et al. (2022) study finds relatively young ages of hundreds of Myr up to ∼1 Gyr for the studied star clusters around 18 galaxies in the redshift range 1.3 < z < 7.7. The star clusters are consistent local globular cluster in terms of their size and stellar mass. These ages are ∼1–2 Gyr younger than what is measured for our BCG star clusters at 3σ, which is in line with the formation of star clusters over cosmic time. On the other hand, Mowla et al. (2022) study in detail a system dubbed as the "Sparkler" at z = 1.378 and find ages of 3–5 Gyr for 6 out of the 12 star clusters. This would indicate that these star clusters have been formed during the early reionization phase of the intergalactic medium at z = 8–11, which would be an interesting but also extreme scenario for cluster formation as understood by current models and other observations of lensed galaxies at high redshifts (see the discussion in Claeyssens et al. 2022 and references therein as well as, for example, Vanzella et al. 2017). We note that Claeyssens et al. find much younger ages (∼100 Myr) for the same star clusters. Understanding this disagreement is beyond the scope of our Letter; however, there could be multiple reasons such as aperture versus PSF fitting photometer methods, the inclusion of galactic extinction, differences in SSP models and SED-fitting codes, as well as differences in the JWST photometric calibration references.

Due to their pointlike nature and faintness, cool dwarf stars in our Milky Way could be mistaken as globular clusters in the z = 0.39 host galaxies. A significant contamination of our sample by such stars is, however, ruled out by two reasons.

First, as shown in Figure 5, blackbody models (representing cool Milky Way dwarf stars from 3000–3400 K) are not consistent with the extracted photometry of our sources. Specifically, the mid-infrared peak is redshifted consistent with the redshift of the cluster galaxies (z = 0.39). A combination of blackbody models (i.e., stars of different temperatures) would be necessary to explain the observed photometry. We note that due to the stacking analysis, stars of different temperatures can be combined, leading to a similar SED as the stacked photometry of the star clusters in our sample. As an additional test, we therefore compared the photometry of randomly picked star clusters to point sources (i.e., stars) at comparable faint magnitudes selected from the parallel NIRCam module off the center of SMACS0723. We find that the mid-infrared peak emission between the two samples is significantly offset (as a results of the redshift), suggesting negligible contribution of stars to the stack.

Second, the predicted number of stars in the NIRCam FoV is low compared to the selected compact star clusters. Specifically, we predicted the number of stars and brown dwarfs from the Wainscoat et al. (1992) as well as the Kirkpatrick et al. (2021) models. The former model is in good agreement with the number of stars in the COSMOS field (Scoville et al. 2007) as shown in Fajardo-Acosta et al. (2022); however, it does not include consistently M, L, and T spectral types, which start to contribute significantly at fainter magnitudes. The expected number of stars from the Wainscoat and Kirkpatrick models are shown in the left panel of Figure 4 as a blue histogram and dark blue horizontal lines, respectively. We would expect on the order of ∼20 stars and dwarfs over the full FoV of NIRCam, and approximately 5 over the area where the compact star clusters are found (see Figure 1).

The negligible contribution of Milky Way stars to our sample is also suggested by comparing the number of selected star clusters in the FoVs of the two NIRCam modules. Specifically, the parallel module off-center of the SMACS0723 cluster should be dominated by Milky Way stars. Repeating the sample selection on that field results in only ∼10 point sources at similar magnitudes as the selected star clusters. Conservatively assuming that these are all stars and that the stellar density is the same in both FoVs, we would expect <10 out of the 178 selected compact star clusters to be Milky Way stars.

Vice versa, we note that globular cluster and dwarf galaxies in and around our Milky Way would be easily resolved by JWST and can therefore be ruled out.