On The Very Bright Dropouts Selected Using the James Webb Space Telescope NIRCam Instrument

$\bullet$ PRIMER in COSMOS and UDS: two shallow but wide fields. The total areas covered by NIRCam are 137.13 and 185.83 arcmin², respectively, and those covered by MIRI are 100.58 and 112.51 arcmin², respectively. The PRIMER program executed the MIRI observations as the primary and the NIRCam ones as the coordinated parallel while maximizing the overlapping areas between the two. In the end, the overlapping areas between the NIRCam and MIRI observations are 67.80 and 76.91 arcmin² for these two fields, respectively. The NIRCam observations utilized eight passbands: F090W, F115W, F150W, and F200W in the short-wavelength channel (SW), and F277W, F356W, F410M, and F444W in the long-wavelength channel (LW). The MIRI observations were in F770W and F1800W. In these two fields, we selected dropouts in F090W, F115W, F150W, F200W, and F277W. To assist the F090W dropout selection, we integrated the HST ACS data from the CANDELS survey (Grogin et al., 2011; Koekemoer et al., 2011) in both COSMOS and UDS.

$\bullet$ CEERS in EGS: another shallow but wide field. It was observed in seven NIRCam bands: F115W, F150W, and F200W in the SW channel, and F277W, F356W, F410M, and F444W in the LW channel. The NIRCam coverage is 86.45 arcmin². The MIRI observations were made in smaller areas and were in seven bands: F560W, F770W, F1000W, F1280W, F1500W, F1800W, and F2100W. However, the MIRI footprints are non-overlapping for all these bands, and different parts of the field are covered only by certain sets of filters: (1) F560W and F770W (7.69 arcmin² overlapping with NIRCam), (2) F1000W, F1280W, F1500W, and F1800W (3.72 arcmin² overlaping with NIRCam), (3) F1500W only (5.15 arcmin² overlaping with NIRCam), and (4) F2100W only (no overlap with NIRCam). Due to the lack of the F090W data, we only searched for the F115W, F150W, F200W, and F277W dropouts. We utilized the ancillary HST ACS data from the CANDELS survey to assist the F115W dropout selection.

$\bullet$ GOODS-S: the main part of the data are from the JWST Advanced Deep Extragalactic Survey (JADES; Eisenstein et al., 2023a), which made the NIRCam observations in two areas within the GOODS-S field at two different depths, “deep” and “medium” (hereafter JADES-deep and JADES-medium). The data were taken in 9 NIRCam bands: F090W, F115W, F150W, and F200W in the SW channel, and F277W, F335M, F356W, F410M, and F444W in the LW channel. These data are all much deeper than PRIMER and CEERS. We also integrated the NIRCam data from the Pure Parallel Wide Area Legacy Imaging Survey (PANORAMIC, PID 2514; Williams et al., 2024) and those from PID 3215 (Eisenstein et al., 2023b) into the JADES-medium field. The final NIRCam images cover 25.50 and 37.94 arcmin² in JADES-deep and JADES-medium, respectively. For MIRI, we combined all the public observations overlapping the JADES coverage into one mosaic per band, which include those from PID 1207 (Rieke et al., 2017), 1283 (Norgaard-Nielsen & Perez-Gonzalez, 2017), and 1180 (PI Eisenstein). Eight MIRI bands were utilized by these programs: F560W, F770W, F1000W, F1280W, F1500W, F1800W, F2100W, and F2550W, and the final MIRI images cover 32.90 arcmin² in all eight bands. The overlapping areas between NIRCam and MIRI are 20.83 and 15.99 arcmin² in JADES-deep and JADES-medium, respectively.

We reduced all the data in this study on our own using the JWST pipeline (Bushouse et al., 2024). The NIRCam data reduction followed the procedures outlined in Yan et al. (2023b) and Yan et al. (2024). For MIRI, we ran through the similar procedures but with four changes: (1) in the calwebb_detector1 step, we set jump.find_showers to “True” to remove the large residuals on single MIRI exposures due to strong cosmic ray events; (2) for the products after the stage 2 pipeline process, we followed the recipe by Yang et al. (2023) to remove the stripe-like noise pattern in horizontal and/or vertical directions; (3) we then removed the remaining noise patterns by constructing templates on a per-observation and per-filter basis and subtracted the corresponding template from each single exposure; and (4) we excluded the coronagraph areas in each individual exposure by masking their pixels to DO_NOT_USE in the data quality array.

The dropout selection involves the NIRCam data but not the MIRI data. For each field, we ran SExtractor in the dual-image mode and adopted F356W-based match-aperture photometry. We chose the F356W band as the detection band for the following reasons: (1) the F356W images are usually the deepest across all bands; (2) the PSF in this band is sufficiently large, which ensures that the apertures properly determined in this band include all source fluxes in any bluer bands; (3) the F356W images present a notably cleaner background than those in the SW bands.

The source extraction was done by applying a $5\times 5$ Gaussian convolution filter with a full width at half maximum (FWHM) of 2 pixels. From the weight images produced by the JWST data reduction pipeline, we derived the “root mean square” (rms) maps using the astroRMS routine ¹¹1See https://github.com/mmechtley/astroRMS; we modified the routine slightly by adopting a better source masking functionality., which calculates the auto-correlation of the science image pixels due to the drizzling process that should be applied to scale the weight images. These rms maps were used for both the source detection and the estimate of photometric errors. The detection and analysis thresholds were set to 1.0 in SExtractor. We adopted the MAG_ISO magnitudes and only retained the sources with signal-to-noise ratio (S/N) of at least 5 and ISOAREA of at least 10 pixels in the F356W image.

When constructing SEDs, a common practice is to use photometry on PSF-matched images, i.e., the images in different bands are all convolved to have the same PSF size as in the band that has the largest PSF. Our SED analysis would involve both the NIRCam and the MIRI data, and PSF-matched photometry across all bands would not be appropriate in this case: smearing the NIRCam data (PSF FWHM 0$\farcs$030 to 0$\farcs$145 from F090W to F444W) to the coarsest MIRI resolution (F2550W PSF FWHM of $\sim$0$\farcs$803) would blend many unrelated neighbors in the NIRCam images and corrupt the NIRCam photometry.

As the best compromise, we took a hybrid approach. In the NIRCam wavelengths, we prepared another set of NIRCam photometry done on the PSF- matched NIRCam images. For each field, we convolved the images in the bluer bands to the PSF size of the F444W image. The PSFs were derived using isolated stars in the field, following the procedure in Ling & Yan (2022). SExtractor was again run in the dual-image mode, and this time the F444W image was used as the basis. We again adopted the MAG_ISO magnitudes to calculate the colors. A common practice would be to scale up the MAG_ISO-based SED to the “total flux” SED by adding the difference between the MAG_ISO and the MAG_AUTO magnitude in F444W band. As we will show in Section 4, however, a large fraction of our objects have neighbors that would contaminate the MAG_AUTO magnitude. Therefore, we chose not to apply such a correction. For isolated, bright ($m_{444}<26.5$ mag) sources, we found that the differences between the MAG_ISO and the MAG_AUTO magnitudes were under 0.05 mag, and the impact of omitting such a correction would only be marginal to the SED analysis.

The MIRI PSF size vary greatly in different bands, which makes the PSF-matched photometry also inappropriate among the MIRI images. Therefore, we ran SExtractor in the single-image mode for each MIRI band and adopted the MAG_ISO magnitudes. Similar to what was mentioned above for the NIRCam images, we also found that the difference between the MAG_ISO and the MAG_AUTO magnitudes were under 0.05 mag for bright, isolated sources in the MIRI images.

Lastly, we note that we treated any photometry with S/N $<2$ as non-detection and adopted the 2 $\sigma$ upper limit. Such limits were measured on the rms maps at the source locations using circular apertures of sizes equivalent to those of the MAG_ISO apertures.

For simplicity, hereafter we denote the magnitudes in the HST/ACS F435W, F606W, F775W, F814W, and F850LP bands as $m_{435}$, $m_{606}$, $m_{775}$, $m_{814}$, $m_{850}$, respectively, those in the JWST/NIRCam F090W, F115W, F150W, F200W, F277W, F335M, F356W, F410M, and F444W bands as $m_{090}$, $m_{115}$, $m_{150}$, $m_{200}$, $m_{277}$, $m_{335}$, $m_{356}$, $m_{410}$, and $m_{444}$, respectively, and those in the JWST/MIRI F560W, F770W, F1000W, F1280W, F1500W, F1800W, F2100W, and F2550W as $m_{560}$, $m_{770}$, $m_{1000}$, $m_{1280}$, $m_{1500}$, $m_{1800}$, $m_{2100}$, and $m_{2550}$, respectively.

We fitted the SEDs using three different tools, namely, Le Phare (version 2.2; Arnouts et al. 1999; Ilbert et al. 2006), EAZY (eazy-py version 0.6.4; Brammer et al. 2008), and CIGALE (version 2022.1; Boquien et al. 2019). A major goal was to obtain their photometric redshifts ($z_{\rm ph}$), which were allowed to vary between $z=0$ to 25 in the fitting process. The settings for each tool are as follows.

$\bullet$ Le Phare: we constructed the templates using the population synthesis models of Bruzual & Charlot (2003, hereafter “BC03”) and the initial mass function (IMF) of Chabrier (2003). We adopted an exponentially declining star formation history (SFH), i.e., SFR $\propto e^{-t/\tau}$, where $\tau$ ranges from 0 to 13 Gyr. We applied the Calzetti’s extinction law (Calzetti et al., 1994; Calzetti, 2001) with E(B-V) ranging from 0 to 1 mag at a step size of 0.1 mag. The option to include the contribution from emission lines was turned on.

$\bullet$ EAZY: we used the template set GALSEDATLAS of Brown et al. (2014), which was retrieved from MAST ²²2https://archive.stsci.edu/hlsp/galsedatlas. This set includes the spectra of 129 nearby galaxies of different types, which cover the UV-to-mid-IR wavelength range. Presumably, using this set of templates would optimize the fitting at low redshifts.

$\bullet$ CIGALE: we adopted a grid of CIGALE templates that include those using a delayed $\tau$ model with $0.01<\tau\leq 13$ Gyr, a recent starburst, and the simple stellar population (SSP) models (i.e., single bursts) from BC03 assuming Chabrier IMF; we fixed the metallicity to $Z=0.02$. We set the nebular emission contribution with $-4<\log U<-1$ and at a step size of 0.5. We also adopted a modified Calzetti’s extinction law under the dust_modified_starburst module with the color excess of nebular gas E(B-V)_g ranging from 0 to 3 mag at a step of 0.2 mag, and a fixed multiplication factor of 0.44 to apply on E(B-V)_g to calculate the stellar continuum attenuation E(B-V)_s. There are a lot of cases where the bright dropouts have enhanced emission in the central region increasing with wavelength, which could be caused by AGN. To investigate this probability, we included the skirtor2016 AGN models from Stalevski et al. (2012, 2016). We varied the AGN fraction (hereafter $f_{\rm AGN}$), i.e., the AGN contribution to $L_{IR}$, from 0 to 1 at a step size of 0.2. We set the viewing angles at either $30^{\circ}$ or $70^{\circ}$ for Type 1 or 2 AGNs, respectively. We noticed that the program would run into errors when $f_{\rm AGN}=1$, and we resolved this by setting the maximum $f_{\rm AGN}$ to 0.999 ³³3In the cases of CIGALE returning $f_{\rm AGN}=0.999$, they are reported as $f_{\rm AGN}=1.0$. For the objects that do not show a compact central region at any passbands (in other words, their optical-to-IR emission is not likely due to AGN), we fixed their $f_{\rm AGN}=0$.

Examples of SED fitting results from these three runs are provided in Figure 3. We note on the treatment of the upper limits. For Le Phare, we set the magnitude error to $-1$ when a given band has the upper limit imposed, and the routine rejected any fits that violated the upper limit. For EAZY, we used the modified code as described in Yan et al. (2023a, b) to achieve this functionality. For CIGALE, we set the fluxes to the upper limit and the corresponding flux errors to $-1$ times the upper limit.

We divided the bright dropouts in the main sample into three categories based on their $z_{\rm ph}$, namely, “High-$z$”, “Low-$z$” and “Undecided”. The “High-$z$” category consists of objects for which at least two SED fitting tools (among the three) consistently derive $z_{\rm ph}\geq 6.0$. Similarly, the “Low-$z$” category consists of objects that have consistent $z_{\rm ph}<6.0$ from at least two fitting tools. In both categories, we further ranked the objects into “Tier 1” (T1) and “Tier 2” (T2) depending on the goodness of fits and the consistency of the results. For practical purpose, we deemed the fits with raw $\chi^{2}\leq 100$ as “good” fits. For the “High-$z$” category, if an object has good fits and $z_{\rm ph}\geq 6.0$ from all three tools, it was put in T1; if it has good fits and $z_{\rm ph}\geq 6.0$ from only two tools, or if it has $z_{\rm ph}\geq 6.0$ from all three tools but the fits are not always good, it was placed in T2. The ranking for the “Low-$z$” category was done similarly, with a slightly more stringent requirement on the consistency of $z_{\rm ph}$: if an object has good fits and consistent $z_{\rm ph}$ from all three tools so that the differences $\Delta z_{\rm ph}<1.0$, it was put in T1; if it has good fits and consistent $z_{\rm ph}$ from only two tools, or if it has consistent $z_{\rm ph}$ from all three tools but the fits are not always good, it was placed in T2. Finally, all objects that were not in either the “High-$z$” or the “Low-$z$” category as ranked were assigned to the “Undecided” category.

$\bullet$ High-$z$: We identified 10 high-$z$ candidates from the sample, among which 3 are in T1 and 7 are in T2. They are listed in Table 3 and 4, respectively, along with their physical properties derived from the SED analysis.

$\bullet$ Low-$z$: There are 43 T1 and 50 T2 objects in this category, which are listed in Tables 5 and 6, respectively. Among the 93 objects, the vast majority of them have $z_{\rm ph}=1$–4.