The puzzling properties of the MACS1149-JD1 galaxy at z=9.11

We analyze new JWST NIRCam and NIRSpec data on the redshift 9.11 galaxy MACS1149-JD1. Our NIRCam imaging data reveal that JD1 comprises three spatially distinct components. Our spectroscopic data indicate that JD1 appears dust-free but is already enriched, $12 + \log {\rm (O/H) } = 7.90^{+0.04}_{-0.05}$. We also find that the Carbon and Neon abundances in JD1 are below the solar abundance ratio. Particularly the Carbon under-abundance is suggestive of recent star formation where Type~II supernovae have already enriched the ISM in Oxygen but intermediate mass stars have not yet enriched the ISM in Carbon. A recent burst of star formation is also revealed by the star formation history derived from NIRCam photometry. Our data do not reveal the presence of a significant amount of old populations, resulting in a factor of $\sim7\times$ smaller stellar mass than previous estimates. Thus, our data support the view that JD1 is a young galaxy.


INTRODUCTION
The high sensitivity of the NIRSpec spectrometer onboard JWST (Böker et al. 2023) has enabled us to study emission line diagnostics of galaxies at redshift z > 8 with unprecedented details. Already the ERO observations of SMACS J0723.3-7327 S04590 at z = 8.496 Heintz et al. 2023) enabled the first detection at high redshift of the auroral line [OIII]λ4363 and its use to derive an electron temperature T e ≥ 24, 000 K. The derived metallicity for this object is 12+log (O/H) ∼ 7. Comparison with strong emission line diagnostics showed that this object was outside the validity range of most calibrations. A broader look at a small sample of objects with auroral lines by Laseter et al. (2023); Sanders et al. (2023) finds that abundances derived from the T e method are generally not consistent with those from locally calibrated strong emission lines.
The galaxy MACS1149-JD1 (Zheng et al. 2012) (hereafter JD1) was one of a handful of confirmed ≥ 9 galaxies known before the launch of JWST and, being substantially lensed, it was a credible candidate for searching for a potentially very low metallicty proto-galaxy. The high-redshift nature of MACS1149-JD1 was confirmed thanks to ALMA detection of [OIII]λ88µm (Hashimoto et al. 2018;Tokuoka et al. 2022). The previous measurements by Spitzer indicate the presence of old stellar populations, making it a relatively massive system among other galaxies at similar redshifts. These exceptional properties of JD1 led us to make it as one of the prime candidates for deep spectroscopy with JWST.
It is worthwhile noticing that another bright, pre-JWST high-z galaxy, GN-z11 (Oesch et al. 2016), while extremely interesting (see e.g. Cameron et al. 2023a;Tacchella et al. 2023;Bunker et al. 2023;Charbonnel et al. 2023;Maiolino et al. 2023), is considered to possibly host an AGN and is at a redshift where [OIII]λ5007 is beyond the NIRSpec sensitivity range and therefore makes it less likely to be able to carry out a direct metallicity measurement using the auroral lines. As such, JD1 is currently one of a few luminous galaxies at z > 9 that allow a reliable auroral line analysis. 1 In this paper, we present a comprehensive analysis of the properties of JD1 based on new observations by JWST. We describe our data on JD1 in Section 2. Section 3 derives the Hβ based star formation rate. Section 4 is devoted to deriving an estimate for dust extinction and presence of dust as well as constraints on the electron density. Section 5 derives the metallicity of JD1 through the direct method based on auroral lines and discusses implications from the other lines. Section 6 discusses the non-solar abundance ratios for JD1. The star formation history and stellar mass are described in Section 7. Section 8 discusses our conclusions.

NIRCam photometry
NIRCam imaging observations (GTO #1199) were executed on June 6 -8, with six filters configured (F090W, F115W, F150W, F200W, F277W, F356W, F444W), with ∼ 1.1 hrs exposure each. We conduct photometry on the newly taken NIRCam images, along with the archival HST images that were originally taken in several HST programs (CLASH, HFF, GLASS, BUF-FALO; Postman et al. 2012;Treu et al. 2015;Lotz et al. 2017;Kelly et al. 2018;Steinhardt et al. 2020). We follow the same procedure presented in Morishita & Stiavelli (2023) for the image reduction and photometry. We hereby provide a high level description of our workflow: raw NIRCam images are reduced by using the official jwst pipeline (ver1.10.0, with context pmap #1069 Bushouse et al. 2023, with several customized steps included to effectively remove artifacts and improve cosmic ray rejection. The final drizzled images are aligned to the WCS of GAIA. Source fluxes are calculated by using SExtractor (Bertin & Arnouts 1996) in PSF-matched images to the psf of F444W, in an aperture of r = 0. ′′ 16, reaching to 5 σ limiting magnitude of ∼ 27.9-28.8 mag.
The NIRCam images of JD1 (see Figure 1) reveal the presence of three main components, two of which are included in our NIRSpec MSA slits (Sec. 2.2). This means that geometry might play a significant role in the interpretation of the observations. However, we do not observe significant color differences between these components and therefore in our following analysis we will not consider them separately.

NIRSpec MSA spectroscopy
Our NIRSpec observations were executed over four different visits. In the first two visits, #20 and #22 respectively for G235M and G395M, were taken at PA=257.766. In the second group of visits, the same grating pair was used, but with a slightly different PA, namely PA=259.660 for visit #21/G235M and PA=256.766 for visit #23/G395M. Slightly offset pointings ensure that roughly the same area of JD1 is covered by all exposures.
We reduce the MSA spectra using msaexp 2 (ver0.6.13), following the procedure presented in Morishita et al. (2023a, also, Morishita et al., in prep). The two-dimensional sky background is estimated by nodding the stacked spectrum for 6 pixels and then it is subtracted. The one-dimensional spectrum is optimally extracted (Horne 1986) by using the one-dimensional source profile derived from the 2-dimensional stacked spectrum as weight. For the extracted 1-dimensional spectrum, we fit each line of interest with a Gaussian, after subtracting the underlying continuum spectrum, estimated by scaling the best-fit SED template (see Sec. 7). We note that our flux measurements include a small correction for absorption in the Balmer series, which are inferred by the stellar template of the bestfit SED model. The correction is 1% for Hβ, 3% for Hγ, and 4% for Hδ. For each line, the total flux is estimated by integrating the flux over the wavelength range of 2×FWHM, the latter derived from the Gaussian fit. For [O iii]-doublet, we fix the ratio to 1:3 and adopt a single parameter for the widths of both lines. Lastly, we determined an aperture correction by scaling the difference between the continuum measured around individual spectral lines and the one inferred by the best-fit SED template. We find that this correction is well described by the slit-loss correction one derives for perfectly centered source (Jdox 2016, NIRSpec MOS Operations Slit Losses) rescaled by the factor ∼ 2.36 we find for Hβ and [O iii] . This correction is applied to measured line fluxes. However, most of our results are relatively unaffected by this wavelength dependent aperture correction. The aperture correction allows us to apply measurements such as the Star Formation Rate (SFR) to the whole galaxy. In what follows, when necessary we adopt the redshift z = 9.114 obtained for the We have checked that the line ratios obtained in all visits are consistent. Therefore, in the following we coadd the spectra and focus our analysis on the combined spectra. The measured line fluxes are reported in Table 1.

STAR FORMATION RATE
The measured flux in Hβ gives us an estimate of the star formation rate in JD1. Following the same approach as Heintz et al. (2023), we assume the Case B ratio for f Hα/Hβ = 2.80 for T e = 1.6 × 10 4 K as derived from the direct analysis described in Section 5. We derive the star formation rate as: From the measured Hβ flux of 3.7 +0.1 −0.1 × 10 −18 erg s −1 cm −2 and assuming a gravitational magnification factor µ = 10 (Hashimoto et al. 2018) 3 , we obtain SFR Hβ = 5.9 +0.2 −0.2 (10/µ) M ⊙ yr −1 . Note that this is the aperture-corrected value of the SFR (Sec. 2.2). The estimated Hβ-based SFR is consistent with the one derived with [O iii] 88 µm (4.2 +0.8 −1.1 (10/µ) M ⊙ yr −1 ; Hashimoto et al. 2018). Also, the estimate is roughly consistent with the one from our SED analysis, ∼ 3.2 (10/µ) M ⊙ yr −1 , derived from the rest-frame UV luminosity by following Morishita et al. (2023b).
Given the large uncertainties in the magnification values at the location of JD1 (e.g. Grillo et al. 2016;Finney et al. 2018) and the relative insensitivity of our results on the specific magnification we are expressing our results as a function of µ/10 rather than adopting a specific value.
We have attempted to determine the presence of dust by looking at the Balmer decrement. We measure Hγ/Hβ = 0.46 ± 0.03 as compared to the Case B recombination value of 0.47 for T e = 1.6 × 10 4 K. This value would suggest a small A V = 0.12 but is also compatible with no dust. We also find Hδ/Hβ = 0.33 ± 0.03 as opposed to the Case B value of 0.26. The measured value is incompatible at 2.3+σ with dust. These conclusions are supported also by analysing both pointings separately.
In Figure 3, we show the measured values of Hγ/Hβ and Hδ/Hβ (with their 1-σ error band as a function of the electron density n e for a number of Cloudy (Ferland et al. 2017) models that we have run as well as for Gutkin et al. (2016) models. The value we measure is incompatible with the presence of dust for gas temperatures below 2 × 10 4 K. The absence of significant amounts of dust is also in agreement with the non-detection of FIR continuum for this object (Hashimoto et al. 2018;Tokuoka et al. 2022).
An intriguing possibility arising from Figure 3 is that JD1 might be characterized by a higher value of the electron density than the typically assumed values in the range 100-1000. Unfortunately, our spectra cannot resolve the [OII] doublet and therefore do not allow us to directly probe the electron density. As an alternative, we can derive an estimate for the electron density from the [OIII]λ88µm/[OIII]λ5007 ratio. This approach has been adopted by, e.g., Fujimoto et al. (2022) for the galaxy ERO S04590 at z = 8.5. In order to compare the ALMA 88µm measurement with our [OIII]λλ4959, 5007 measurements, we need to apply a significant aperture correction which will entail a relatively large uncertainty. We have determined an aperture correction using two different methods. The first approach is to assume that the ALMA flux is distributed across an area similar to the NIRCam image of JD1, with which we have estimated the aperture correction factor in Sec. 2.2 (i.e. 2.36). Alternatively, we can derive an aperture correction by comparing the SFR derived from ALMA [OIII]λ88µm to that derived from Hβ (Sec. 3), to find an aperture correction of 1.69. 4 In the following we will adopt the average of these two values and adopt as error their semi-difference, i.e., 2.02 ± 0.33.
Armed with an estimate of the ALMA 88µm emission, we can derive the [OIII]λ88µm/[OIII]λλ4959,5007 ratio as 0.055 ± 0.012 ± 0.009, where the first error is the measurement one and the second the aperture  correction one. We adopt the same approach as Fujimoto et al. (2022) except by using Cloudy models instead of the PyNeb ones (which gives very similar results). Our Cloudy models (see Section 3.2 of Oesch et al. 2007) are based on a wide variety of electron densities, and use both a range of blackbody temperatures and a range of constant star formation rate of various metallicities modelled by GALAXEV (Bruzual & Charlot 2003). Our results are shown in Figure 4. For JD1 we find log n e = 2.60 +0.25 −0.27 , or n e ≃ 400. As a consistency check, for ERO S04590 we recover a value intermediate between 100 and 300 as found by Fujimoto et al. (2022).
On this basis, in the following we assume that JD1 has a density of n e ≃ 400 and no dust.

EMISSION LINE ANALYSIS
In the following, we will adopt the standard notation for line ratios, namely: With robust measurements of the auroral line [OIII]λ4363 and a value of the electron density, we can now apply the direct method to derive the electron temperature and the oxygen metallicity. The values derived at the two pointings are within the 1-σ error bar and thus we will use the value from the combined spectrum. Using Aller (1984) iterative method (see also Izotov et al. 2006) and n e = 400, we derive T e (O ++ ) = 1.6 × 10 4 K and infer 12 + log O ++ /H = 7.88 +0.04 −0.05 . The error bars quoted here are those related to the measurement error. Varying the density between 100 and 1000 would contribute a 0.002 error. We have estimated T e (O + ) using a variety of methods (Izotov et al. 2006;Campbell et al. 1986;Pilyugin & Grebel 2016;Laseter et al. 2023;Sanders et al. 2023) and find values similar to or lower than T e (O ++ ). Using these values we derive a contribution from O + to metallicity 12 + log O + /H = 6.39 − 6.55 which is in any case small compared to the O ++ contribution. For the total Oxygen metallicity, we adopt 12 + log (O/H) = 7.90 +0.04 −0.05 . We have also re-derived the direct metallicity for ERO S04590 and found values in agreement with the published ones.
Using the methods outlined in Izotov et al. (2006) we also derive an estimate for the Neon abundance log(Ne/H) = −5.03 ± 0.04. We correct the measurement with the Ionization Correction Factor (ICF) prescription by Amayo et al. (2021), to find log(Ne/H) = −5.02±0.04 or log(Ne/O) = −0.69. Our measurements suggest that JD1 is under-abundant in Neon compared to solar by ∆ log(Ne/O) ≃ −0.23.
We can also derive a Carbon over Oxygen ratio using the direct method (Aller 1984;Izotov & Thuan 1999). Here we find log(C ++ /O ++ ) = −1.09. We can compute an ionization correction factor following Berg et al. (2019). From the measured O32 = 1.535 ± 0.025 and using the Z = 0.1Z ⊙ fit, we find log U = −1.647 ± 0.026 and a Carbon ICF of 1.216 ± 0.013. This gives us log(C/O) = −1.08, implying ∆ log(C/O) = −0.78 or a C/O ratio of 0.16 of the solar value. Figure 5 compares the metallicity and the R23 values we derive for JD1 with those derived for other auro- ral line measurements from the literature (Laseter et al. 2023;Sanders et al. 2023).
In order to explore strong line diagnostics, we considered Gutkin et al. (2016) models and found that the best fitting models correspond to 0.1 solar metallicity with n e = 100 and upper mass function cutoff m up = 100M ⊙ and the highest values of the ionizing parameter explored in the models, U = −1.5/ − 1. This is in broad agreement with the direct measurement of the Oxygen metallicity. The high level of ionization is also hinted at by the very high [OIII]λ5007/Hβ = 10.03 ± 0.31 which would correspond to the AGN-star formation boundary in the BPT (Baldwin et al. 1981;Lamareille et al. 2009;Lamareille 2010) diagram. Indeed, this diagram appears insufficient to separate star formation from AGN photoionization at z ≥ 6 (Cameron et al. 2023b;Übler et al. 2023). However, we do not see any evidence of a broad component in the Balmer lines.
The measurement of (C/O) for JD1 is consistent with Gutkin's (C/O) ∼ 0.38 solar models. Thus, the broad conclusions of the direct measurements, namely that JD1 has sub-solar metallicity, high-ionization, and Carbon-underabundance, could be derived from a strong emission line analysis, even though the details might not be precisely the same.

ON THE NON-SOLAR ABUNDANCE RATIO
We have seen that the direct method indicated that JD1 has C/O ratio of 0.16 the solar value.
We should note the large difference in wavelength between CIII]λ1909 and Hβ, and the fact that they fall on different NIRSpec gratings. The measurement of (C/O) depends also on an estimation of an ionization correction based on the O32 ratio which is itself vulnerable to the presence of dust. This opens up the possibility of uncertainties due to dust (i.e. an additional correction that we have not applied) and uncertainties in the instrumental calibration. The latter, however, should be less than 5 %. Acknowledging this potential source of uncertainty, we now proceed to examine the implication of the derived under-abundance of Carbon.
A similar under-abundance has been found in a galaxy at z = 6.23 by Jones et al. (2023). However, that galaxy has an overall lower metallicity than JD1, which may make a very young age and enrichment dominated by core-collapse supernovae less surprising. Comparing JD1 with the results by Arellano-Córdova et al. (2022), looking at the top panels of their Figure 4, one notices that JD1's (C/O) is lower than what they find for their z > 7 galaxies but is not anomalous compared to z ∼ 2 galaxies. The same is true for the (Ne/O) value of JD1 is not observed locally for objects of its metallicity.
The under-abundance of Carbon exceeds the underabundance of Neon (see Section 5). It should be noted that while a low C/O is a common feature of corecollapse supernova yields (e.g., Nomoto et al. 2013), the same is not true for (Ne/O), even though there are models where this is the case (e.g., Rauscher et al. 2002, model S20). Massive, Population III, PISN supernovae can also lead to significantly under-abundant (Ne/O) (Heger & Woosley 2010).

STAR FORMATION HISTORY AND ABSENCE OF OLD STELLAR POPULATIONS
The sensitive photometry over 1-5 µm by NIRCam enables us to constrain the stellar component in JD1. The F356W and F444W data points are critical, as they cover, for the redshift of JD1, the rest-frame 3800-4200Å where the Balmer break, a characteristic break for relatively older (i.e. B, A, and F-type) stars, is located. Previous studies using Spitzer Ch1 and Ch2 reported red color (m ch1 − m ch2 ∼ 0.9 mag; Huang et al. 2016;Zheng et al. 2017;Hoag et al. 2018), speculating the presence of such old populations formed at z ∼ 15 (Hashimoto et al. 2018). Here, using our JWST data we report m 356 = 25.76 ± 0.02 and m 444 = 25.63 ± 0.02 i.e. the absence of such characteristic features in JD1. As the F356W magnitude is consistent with the previous IRAC Ch1 measurements, we suspect that image con-fusion might have affected the Spitzer IRAC Ch2 measurements.
The NIRCam photometry reveals a blue color, with m F356W − m F444W = 0.13 ± 0.01 mag. This is considerably smaller than the previous measurement. We note that the filter curves of the corresponding bands of the two telescopes are only slightly different. For the redshift of JD1, [O iii]λ5007 line is within the red edge of F444W, but a hair outside the Spitzer Ch2. As such, the absence of strong color in F356W/F444W (despite the strong [O iii] included in the redder filter) indicates the dominance of young (< 100 Myr) stellar populations.
Our spectrum is deep enough to reveal the continuum, and thus allows us to directly measure the Balmer break. Following the definition of Balogh et al. (1999), we measure D(4000) ∼ 0.5, meaning that the continuum at the blue side is brighter than the red side. We also follow the procedure presented in Curtis-Lake et al. (2023) and measure the strength of the Balmer break 0.5 ± 0.1, supporting the dominance of young populations.
Lastly, we perform a joint SED analysis combining photometric and spectroscopic data using gsf (Morishita et al. 2019). Briefly, we use fsps (Conroy et al. 2009), with the default MIST isochrone (Choi et al. 2016) and MILES stellar library (Falcón-Barroso et al. 2011). Both stellar-and gas-phase metallicity are fixed to 1/5 the solar, and dust is set to zero (A V = 0), based on the results from our ISM analyses above. We adopt a binned star formation history, which offers flexible inference on star formation history, with a set of ages [0.001,0.003,0.01,0.03,0.1,0.3,0.5] / Gyr. We find that JD1 experienced two distinct, major star formation phases in the past -one ∼ 300 Myr old (z ∼ 14) and a more intense one 10 Myr old. The presence of a more recent burst is consistent with the Carbon deficit that we have found in our spectral line analysis. We note that the absence of a significant amount of old stars results in a much smaller stellar mass in JD1, by a factor of ∼ ×6.8 from previous measurements, namely a stellar mass of 1.6 +0.3 −0.3 × 10 8 (10/µ)M ⊙ . Combined with the SFR from Section 3 we find a specific star formation rate of 37 +10 −7 Gyr −1 . To check for consistency among various assumptions in SED fitting, we perform another SED analysis with BAGPIPES (Carnall et al. 2018;Hsiao et al. 2023) using BPASS v2.2.1 stellar population models (Stanway & Eldridge 2018) and CLOUDY c17.03 photoionisation code (Ferland et al. 2017). We also fix the metallicities to 0.15 solar and dust to null. We assume a smooth, Gaussian process-based nonparametric SFH (Iyer et al. 2019). We set the SFH to be controlled by four parameters: stellar mass, SFR, and two shape parameters, which essentially divide the SFH into three lookback time intervals in which the galaxy formed equal mass. The derived SFH agrees reasonably well with the binned SFH from above. The resulting stellar mass is 1.42 +0.05 −0.04 × 10 8 (10/µ)M ⊙ . This stellar mass gives us a specific star formation rate of 40 +3 −2 Gyr −1 .

DISCUSSION AND SUMMARY
We find that JD1 is an actively star forming galaxy characterized by metallicity about 1/5 the solar value and with emission line ratios best explained by a very strong ionizing continuum suggestive of a very young stellar population. The youth of the object is also supported by the star formation history we derived, showing a major burst ∼ 10 Myrs ago, and by our finding of a low (C/O) ratio suggestive of Oxygen enrichment by supernovae, while intermediate mass stars have not yet had the time to increase the Carbon fraction. Such indications of a very active, recent and on-going star formation are somewhat surprising when coupled with the lack of evidence for dust as indicated by the measured Balmer decrement as well as the non-detection in the ALMA continuum reported in the literature. While geometry could conspire to give rise to grey dust (see e.g. Witt & Gordon 1996, 2000, it is interesting that this object seems also qualitatively compatible with the dust ejection scenario described by Ferrara et al. (2023, see also Tsuna et al. 2023. Indeed, the value of specific star formation rate we measure exceeds the value of ∼ 32Gyr −1 that we derive from Eq. 3 of Ferrara et al. (2023). It should be noted that the complexities of how supernova driven winds would affect the observed C/O ratio would need to be evaluated in more detail (see, e.g., Berg et al. 2019).
It is worth noticing that the (C/O) ratio of JD1, as well as its metallicity and high ionization (as inferred from the O32 ratio) are comparable to what is observed in some Green Peas galaxies at z < 0.3 (Ravindranath et al. 2020;Rhoads et al. 2023).
We have seen in Section 2.1 that JD1 is characterized by multiple components. This could be an indication of recent interaction or merger. However, we see no evidence of color gradients, and the line ratios at the two slightly offset slit positions are compatible. Given the absence of a color gradient, it is hard to assess how general our conclusions can be once one opens up geometry and substructure. For completeness we have also considered two component Cloudy models. The significant number of additional degrees of freedom prevents us from obtaining a unique solution but our best twocomponent fits share similar properties of hard ionizing flux, sometimes with AGN or an additional stellar population with 8 × 10 4 K of effective temperature, and oxygen enriched models. Given the high level of degeneracy, we do not consider it useful to discuss these solutions in greater detail except to say that they are broadly compatible with the conclusions of single component modeling.
While completing this paper we were sent a prepublication copy of a paper on JD1 by Bradač et al., we find their results broadly compatible with ours given the differences in JWST data for the two papers. Furthermore, JD1 has been observed using the NIRSpec IFU by another program (PID: 1262, PI: N. Luetzgendorf) but we have currently no access to these data. It will be interesting to see whether the spatial resolved spectroscopy modifies the picture presented here and how a combined analysis of both data and of our Cycle 3 observations (#4552) aiming at observing with high resolution the rest frame UV of JD1 will affect the picture.