JWST Identification of Extremely Low C/N Galaxies with [N/O]$\gtrsim 0.5$ at $z\sim 6-10$
Evidencing the Early CNO-Cycle Enrichment and a Connection with Globular Cluster Formation

We measure emission-line fluxes by fitting a Gaussian function convolved by line-spread functions that Isobe et al. (2023) have derived from in-flight NIRSpec data of a planetary nebula (PI: J. Muzerolle; PID: 1125). To measure weak emission lines accurately, we fix redshift and velocity dispersion of the Gaussian function to those of [O iii]$\lambda\lambda$4959,5007. To derive Ne/O, C/O, and N/O ratios, we explore the following line detections: [Ne iii]$\lambda$3869 (for neon abundance), C iii]$\lambda\lambda$1907,1909 (for carbon abundance), N iii]$\lambda$1750¹¹1We refer to the quintet of N iii] lines at restframe 1747, 1749, 1750, 1752, and 1754 Å as N iii]$\lambda$1750 in this paper. and N iv]$\lambda\lambda$1483,1486 (for nitrogen abundance), [O ii]$\lambda\lambda$3727,3729, [O iii]$\lambda$5007, and O iii]$\lambda\lambda$1661,1666 (for oxygen abundance), [O iii]$\lambda$4363 (for electron temperature $T_{\rm e}$), and Balmer lines of H$\alpha$, H$\beta$, and H$\gamma$ (for hydrogen abundance and dust correction). Hereafter, we just write [Ne iii], C iii], N iii], N iv], [O ii], [O iii], and O iii] as meaning [Ne iii]$\lambda$3869, C iii]$\lambda\lambda$1907,1909, N iii]$\lambda$1750, N iv]$\lambda\lambda$1483,1486, [O ii]$\lambda\lambda$3727,3729, [O iii]$\lambda$5007, and O iii]$\lambda\lambda$1661,1666, respectively, for simplicity. Thirty-five and 7 of our sample galaxies have $S/N>3$ detections of [Ne iii] and C iii], respectively. We find that one of our sample galaxies, GLASS_150008, has a detection of N iii] with $S/N=4.2$. We also identify the detection of N iv] ($S/N=4.3$) from one of our sample galaxies, CEERS_01019, which has also been reported by Larson et al. (2023).

Figure 1 shows JWST/NIRCam images²²2The imaging data of GLASS_150008 are drawn from the UNCOVER (Bezanson et al., 2022) website (https://jwst-uncover.github.io/), while the imaging data of CEERS_01019 are taken from the CEERS survey (Finkelstein et al., 2022) and reduced by Harikane et al. (2023b). and spectra around N iv], O iii], N iii], and C iii] of GLASS_150008 and CEERS_01019. The O iii], N iii], and C iii] of GLASS_150008 and the N iv] of CEERS_01019 have the $S/N$ ratios larger than 4. We note that GLASS_150008 has the ratio of each line flux in the N iii]$\lambda$1750 quintet to the total flux of the quintet in agreement within a 1$\sigma$ error level with that based on the electron density of $\lesssim 10^{4}$ cm^-3 derived from the observed high C iii]$\lambda$1907/C iii]$\lambda$1909 ratio.

The left panels of Figure 2 compare N iii]/O iii] and N iv]/O iii] ratios of GLASS_150008 (double red circle) and CEERS_01019 (red square) with those of GN-z11 (magenta square) as a function of N iv]/N iii]. The N iii] and N iv] fluxes from Maiolino et al. (2023) and the O iii] upper limit from Bunker et al. (2023)³³3Originally reported as $2\sigma$ upper limit in Bunker et al. (2023) but converted to $3\sigma$ upper limit for consistency with our flux measurements in Section 3.1. are scaled by C iii] lines. We find that GLASS_150008 and CEERS_01019 have high N iii]/O iii] and N iv]/O iii] ratios, respectively, both of which are comparable to those of GN-z11, which is reported to have a super-solar N/O ratio (Section 1). The right panels of Figure 2 also show that GLASS_150008 and CEERS_01019 have low C iii]/N iii] and C iii]/N iv] ratios, respectively, both of which are comparable to those of GN-z11.

To compare the observed emission-line ratios, we construct photoionization models using Cloudy (Ferland et al., 2013) for both stellar and AGN radiations. To model young star-forming galaxies, we use BPASS (Stanway & Eldridge, 2018) binary stellar radiations under the assumptions of the instantaneous star-formation history with the stellar age of 10 Myr, upper star mass cut of 100 $M_{\odot}$ with the Salpeter (1955) initial mass function (IMF), and the hydrogen density $n_{\rm H}$ of 300 cm^-3. The $n_{\rm H}$ value is inferred from a typical value of the electron density ($n_{\rm e}$) of $\gtrsim 300$ cm^-3 in the ISM of $z>4$ galaxies (Isobe et al., 2023). We also fix He/H and metal-to-oxygen ratios to be the solar abundances. We vary O/H and ionization parameter ($U$) values within the ranges of $-2\leq[$O/H$]\leq 1$ and $-3.5\leq\log(U)\leq-0.5$ in 0.25 and 0.25 increments, respectively. We set the stellar metallicity equal to the nebular metallicity defined by the O/H ratio. We refer to this model as the young stellar model, hereafter. We also construct the photoionization models for extremely young star-forming galaxies with very massive stars (extremely young massive stellar model, hereafter), assuming the stellar age of 1 Myr and upper star mass cut of 300 $M_{\odot}$. The other parameters are the same as those of the young stellar model. We also compute the stellar photoionization models with [N/O$]=1$, whose other parameters are the same as those of the young stellar model and extremely young massive stellar model.

We also construct AGN photoionization models whose incident radiation is parameterized by the following 4 parameters: the big-bump temperature $T_{\rm BB}$, the X-ray to UV ratio $\alpha_{\rm OX}$, the low-energy slope of the big-bump component $\alpha_{\rm UV}$, and the X-ray component slope $\alpha_{\rm X}$. Typical AGNs have $\alpha_{\rm OX}\sim-1.4$ (Zamorani et al., 1981) and $\alpha_{\rm UV}\sim-0.5$ (Francis, 1993; Elvis et al., 1994). We refer to the AGN model with the parameter set of $T_{\rm BB}=1.5\times 10^{5}$ K, $\alpha_{\rm OX}=-1.4$, $\alpha_{\rm UV}=-0.5$, and $\alpha_{\rm X}=-1.0$ as the default AGN model. The flux density per frequency $f_{\nu}$ is expressed by:

where $a$ is a coefficient corresponding to $\alpha_{\rm OX}$ and $T_{\rm IR}$ is an infrared cutoff temperature at $kT_{\rm IR}=0.01$ Ryd. The continuum above 100 keV is also assumed to cut off by $\nu^{-2}$. Using the AGN radiation, we calculate emission line ratios under the assumption of $n_{\rm H}=300$ cm^-3. We vary O/H, $U$, and N/O within the ranges of $-2\leq[$O/H$]\leq 1$, $-3.5\leq\log(U)\leq-0.5$, $0\leq[$N/O$]\leq 2$, and $-1\leq[$C/O$]\leq 0$ in 0.25, 0.25, 1, and 1 increments, respectively. We also fix He/H and metal-to-oxygen ratios other than N/O to be the solar abundances. Note that different assumptions of $n_{\rm H}$ from 10 to $10^{6}$ cm^-3 can change key emission-line ratios of N iii]/O iii], C iii]/O iii]. and C iii]/N iii] by only $\lesssim 0.1$ dex.

In Figure 2, we plot the young stellar models with [N/O$]=0$ and [C/N$]=0$ in blue, the defalut AGN models with [N/O$]=0$ and [C/N$]=0$ in cyan, the young stellar models with [N/O$]=1$ and [C/N$]=-1$ in green, and the default AGN models with [N/O$]=1$ and [C/N$]=-1$ in yellow. We find that the high N iii]/O iii] ratios of GLASS_150008 and GN-z11, as well as the high N iv]/O iii] ratios of CEERS_01019 and GN-z11, are larger than those predicted by both the young stellar model and the default AGN model with [N/O$]=0$ (blue and cyan). Similarly, the low C iii]/N iii] and C iii]/N iv] ratios in the 3 galaxies are comparable to those of the models with [C/N$]=-1$ (green and yellow). These results indicate that the 3 galaxies have very high N/O and low C/N ratios significantly above and below the solar abundance, respectively. These results are not likely affected by the metallicity changes from $\log(\rm Z/Z_{\odot})=-1$ (blue) to 0.25 (red) as shown in Figure 2, while for accuracy, we derive metallicities first in Sections 3.3 and 3.5.

We derive color excesses $E(B-V)$ from Balmer line ratios of H$\beta$/H$\alpha$, H$\gamma$/H$\alpha$, and H$\gamma$/H$\beta$ as many as possible by assuming the dust attenuation curve of Calzetti et al. (2000) and the case B recombination. Intrinsic values of the Balmer line ratios are calculated by PyNeb (Luridiana et al. 2015; v1.1.15). If none of the above Balmer line ratios are available, we use a median value of $E(B-V)=0.07$ for the other galaxies in our sample. Using the $E(B-V)$ values and Calzetti et al. (2000)’s curve, we correct emission-line ratios for dust attenuation.

For the 9 galaxies with the measurements of electron temperatures of O iii ($T_{\rm e}(\rm O\,\textsc{iii})$) by Nakajima et al. (2023), we use the $T_{\rm e}$ values and their errors. In addition, we derive $T_{\rm e}$(O iii) values of 5 galaxies with [O iii]$\lambda$4363 and/or O iii]$\lambda$1666. The other galaxies in our sample are assumed to have $T_{\rm e}$(O iii$)=15000$ K with 1$\sigma$ uncertainty of 5000 K. We calculate $T_{\rm e}$(O ii) from $T_{\rm e}$(O iii) and the empirical relation of Garnett (1992). We assume $n_{\rm e}$ values to be 300 cm^-3.

We derive O⁺/H⁺ from [O ii]/H$\beta$ and $T_{\rm e}$(O ii) and O²⁺/H⁺ from [O iii]$\lambda\lambda$4959,5007/H$\beta$ and $T_{\rm e}$(O iii). In our sample galaxies where only [O ii] upper limits are present, we presume the smallest values of [O ii] fluxes to be the ones obtained from [O iii] under the assumption of $\log(U)=-1$. For these galaxies, we also suppose the true and maximum values of [O ii] fluxes to be their 1$\sigma$ and 3$\sigma$ upper limits, respectively. For our sample galaxies where [O ii] falls outside the wavelength coverage of NIRSpec, we assume the true [O ii] fluxes of the galaxies to be [O iii] divided by a median [O iii]/[O ii] ratio of the other galaxies in our sample. The minimum and maximum values are assumed to be [O iii] divided by [O iii]/[O ii] at $\log(U)=-1$ and $-3$, respectively. We then obtain gas-phase metallicity $12+\log(\rm O/H)$ from the equation ${\rm O/H}={\rm O}^{+}/{\rm H}^{+}+{\rm O}^{2+}/{\rm H}^{+}$ by ignoring neutral oxygen and O³⁺ and higher-order oxygen ions as in e.g., Izotov et al. (2006).

We include errors of the used fluxes as well as uncertainties of the assumptions of $T_{\rm e}$ and [O ii] for some of our sample galaxies into $12+\log(\rm O/H)$ errors by Monte Carlo simulations. We derive 1000 values of $12+\log(\rm O/H)$ with $T_{\rm e}$ and flux values such as [O ii] randomly fluctuated by their errors under the assumption of the normal distribution. Then, we derive the 16th and 84th percentiles of the distribution of the 1000 $12+\log(\rm O/H)$ values as the $\pm 1\sigma$ confidence interval of $12+\log(\rm O/H)$. We note that we cannot measure $12+\log(\rm O/H)$ of 7 galaxies in our sample due to the lack of H$\beta$. The derived $T_{\rm e}$, $E(B-V)$, and $12+\log(\rm O/H)$ values are summarized in Table 3. We have confirmed that our derived $12+\log(\rm O/H)$ values of GLASS_150008 and CEERS_01019 are consistent with those measured by Jones et al. (2023) and Larson et al. (2023) based on the stellar radiation within a $1\sigma$ error level, respectively.

We also estimate $12+\log(\rm O/H)$ of GN-z11 in a similar way to our sample galaxies. However, we use [O iii]$\lambda$4363 instead of [O iii]$\lambda$5007, which is outside the NIRSpec wavelength coverage in the case of GN-z11. We assume $T_{\rm e}(\mathrm{O\,\textsc{iii}})=15000\pm 5000$ K and $n_{\rm e}=300$ cm^-3. We estimate $E(B-V)=0.00$, $\log(U)=-2.04$, O⁺/H${}^{+}=6.08\times 10^{-6}$, and O²⁺/H${}^{+}=9.39\times 10^{-5}$ from the observed H$\delta$/H$\gamma$, [O iii]$\lambda 4363$/[O ii], [O ii]/H$\delta$, and [O iii]$\lambda 4363$/H$\delta$ ratios, respectively. The H$\delta$/H$\gamma$ ratio is taken from Maiolino et al. (2023), while the other line ratios are taken from Bunker et al. (2023). We obtain $12+\log(\rm O/H)=8.00^{+0.76}_{-0.46}$, which is consistent with that derived by Cameron et al. (2023) and Senchyna et al. (2023) based on the stellar radiation within a $1\sigma$ error level.

We derive Ne/O, C/O, and N/O ratios from ionic abundance ratios with similar ionization energies of Ne²⁺/O²⁺, C²⁺/O²⁺, and N²⁺/O²⁺ to minimize systematics of ionization correction factors (ICFs). We calculate Ne²⁺/O²⁺ ratios from [Ne iii]/[O iii] ratios. For GLASS_150008 and GN-z11 with the O iii] detection, we use C iii]/O iii] and N iii]/O iii] ratios to obtain C²⁺/O²⁺ and N²⁺/O²⁺ ratios with less systematics. For the other galaxies in our sample, we calculate C²⁺/O²⁺ and N²⁺/O²⁺ ratios from [O iii] lines. We calculate these ionic abundance ratios using PyNeb with the same transition probabilities and collision strengths listed in Table 1.

To derive chemical abundance ratios (Ne/O, C/O, and N/O) from the ionic abundance ratios (Ne²⁺/O²⁺, C²⁺/O²⁺, and N²⁺/O²⁺), we extract ICFs from the young stellar model (Section 3.2). We have checked that our ICFs provide Ne/O and C/O ratios consistent with those based on Izotov et al. (2006) and Berg et al. (2019)’s ICFs, respectively. We have also confirmed that ICF(Ne²⁺/O²⁺), ICF(C²⁺/O²⁺), and ICF(N²⁺/O²⁺) of all our sample galaxies are $\sim 1$. In addition, the extremely young massive stellar model provides Ne/O, C/O, and N/O values similar to those of the young stellar model. We derive the errors of the Ne/O, C/O, and N/O ratios by Monte Carlo simulations in the same way as for $12+\log(\rm O/H)$ (Section 3.3). For our sample galaxies without the detections of [Ne iii], C iii], or N iii], we use $3\sigma$ upper limits of these lines to obtain upper limits of Ne/O, C/O, or N/O ratios. We list the derived abundance ratios in Table 3, while those of GLASS_150008, CEERS_01019, and GN-z11 are highlighted in Table 2. All the 3 galaxies have super-solar N/O values under the assumption of the stellar radiation.

The top left panel of Figure 6 shows N/O ratios as a function of metallicity. We find that GLASS_150008 and CEERS_01019 show [N/O$]\gtrsim 0.5$ as well as GN-z11. For CEERS_01019 and GN-z11, this conclusion does not change regardless of whether AGN or stellar radiation is assumed. CEERS_01019 and GLASS_150008 are the second and third examples of super-solar N/O galaxies at $z>6$ after the report of GN-z11. Hereafter we refer to these 3 galaxies (CEERS_01019, GLASS_150008, and GN-z11) as JWST N-rich galaxies. The super-solar N/O values are significantly higher than those of typical local dwarf galaxies, Galactic H ii regions (gray dots), and typical Galactic stars (gray curve) at a given $12+\log(\rm O/H)$, but comparable to that of a Wolf-Rayet galaxy, Mrk996.

Contrary to the high N/O ratios, the top right panel of Figure 6 illustrates that CEERS_01019 (together with many of our sample galaxies in the semi-transparent white circles) has a C/O ratio comparable to those of the typical local galaxies, local stars, and $z\sim 3$ dwarf star-forming galaxies (e.g., Llerena et al., 2022, 2023). GLASS_150008 shows a C/O ratio even lower than those of the typical local galaxies and stars but comparable to some of our sample galaxies.

This discrepancy between the high N/O and low C/O ratios is illustrated by the bottom left panel of Figure 6, which shows the C/N ratios as a function of metallicity. We find that all the JWST N-rich galaxies have [C/N] values less than $\sim-1$, which are significantly lower than those of the typical local galaxies and stars. This conclusion is also the case regardless of the assumed radiation.

Such low C/N ratios are expected to be observed in nitrogen-loud quasars (e.g., Batra & Baldwin, 2014), while those O/H ratios have not fully been investigated. Alternatively, we plot a chemical evolution model of Hamann & Ferland (1993) reproducing observed emission-line ratios of quasars. Nitrogen in the model is mainly enriched by AGB stars. Although the quasar model with AGB stars can produce low C/N ratios down to [C/N$]\sim-1$, it does not reproduce the (relatively-)low O/H ratios of the JWST N-rich galaxies simultaneously. This is because AGB stars require much delay time ($\gtrsim 1$ Gyr) to decrease the C/N ratio to [C/N$]\sim-1$, which increases O/H ratios too much. This result also suggests that the JWST N-rich galaxies are not likely to be nitrogen-loud quasars whose nitrogen is enriched by AGB stars.

The bottom right panel of Figure 6 shows the relations between N/O and C/O. We confirm that the JWST N-rich galaxies have N/O ratios higher than those of the typical local galaxies and stars at a given C/O ratios, which indicate that only nitrogen is selectively enriched in the JWST N-rich galaxies with respect to the typical local galaxies and stars.

We find that Mrk996, a local metal-poor dwarf star-forming galaxy with a high value of $\log(\rm N/O)\sim 0$ (Senchyna et al., 2023), has a low C/N value similar to the JWST N-rich galaxies (bottom left panel of Figure 6). The bottom right panel of Figure 6 also shows that Mrk996 coincides with the JWST N-rich galaxies on the N/O vs. C/O plane. Mrk996 is known to host WR stars evidenced by the presence of the WR bumps (e.g., Telles et al., 2014). Similar to Mrk996, the JWST N-rich galaxies may also host WR stars.

We also identify some of carbon-enhanced metal-poor (CEMP) stars (Norris et al., 2013) and nitrogen-enhanced metal-poor (NEMP) stars (Beveridge & Sneden, 1994) in the Galactic halo with the C/N vs. O/H and N/O vs. C/O relations comparable to those of the JWST N-rich galaxies as shown by the pink crosses in Figure 6. It should be noted that some of such stars are giants undergoing the conversion of C to N in their hydrogen layers.

Moreover, the purple circles in the bottom panels of Figure 6 indicate dwarf stars in a globular cluster (GC) NGC6752, which have been reported to have high N/O ratios similar to GN-z11 (Senchyna et al., 2023). The purple stars with the error bars also show median and 16th-84th percentile values of the abundance ratios of dwarf stars in a GC 47 Tuc. Carbon and nitrogen abundances of the dwarf stars are drawn from Briley et al. (2004), while the oxygen abundances are taken from D’Orazi et al. (2010). The metallicity of GLASS_150008 is similar to that of the dwarf stars in NGC6752, while CEERS_01019 (GN-z11) has metallicity similar to (or close to) that of the dwarf stars in 47 Tuc. We find that the GC dwarf stars also have low C/N ratios and N/O vs. C/O relations similar to those of the JWST N-rich galaxies. As surface abundance of dwarf stars are likely unevovled, the gas composition at the time of GC formation may also have similar abundance ratios to the JWST N-rich galaxies. This implies that the JWST N-rich galaxies may be progenitors of GCs.

In the bottom panels of Figure 6, we plot the abundance ratios processed by the CNO cycle at the equilibrium state (Maeder et al., 2015) by the brown shaded regions, while core-collapse supernova yields (Watanabe et al., 2023) are shown by the cyan shaded regions. In contrast to the typical local galaxies and stars, we find that the JWST N-rich galaxies have the abundance ratios between those of CCSNe and those of the equilibrium state of the CNO cycle. This suggests that not only CCSNe but also the CNO cycle significantly contribute to the metal enrichment of the JWST N-rich galaxies. Thus, we need scenarios that can enrich gaseous materials originating from the CNO cycle while suppressing oxygen supply by CCSNe in the relatively low-metallicity environments.

In fact, these requirements are met by the scenarios presented to explain nitrogen enrichment of GN-z11 such as Wolf-Rayet (WR) stars (Senchyna et al., 2023; Watanabe et al., 2023), supermassive stars (SMS; Charbonnel et al. 2023; Nagele & Umeda 2023), and tidal disruption events (TDE; Cameron et al. 2023). In all the 3 scenarios, CNO-cycle material can be ejected from the outermost hydrogen-burning layer of stars via stellar winds (for the WR star and SMS scenarios) or gravitational interactions with black holes (for the TDE scenario). Low-metallicity massive stars can also undergo direct collapse (i.e., no ejecta) due to the low opacity of the hydrogen envelope, resulting in a heavier helium core during collapse (Heger et al., 2003). The blue dashed, green solid, and yellow dashed-dotted lines in Figure 7 represent chemical evolution models of WR stars, SMSs, and TDEs, respectively, all of which are constructed by Watanabe et al. (2023) and Watanabe et al. (in prep.). The WR star and SMS models include the yields of WR stars with 25–120 $M_{\odot}$ (Limongi & Chieffi, 2018) and SMSs with $10^{5}\ M_{\odot}$ (Nagele & Umeda, 2023), respectively. The TDE model includes TDE yields calculated by Watanabe et al. (in prep.) with the nucleosynthesis code of Tominaga et al. (2007), assuming the destruction of stars with 9–40 $M_{\odot}$. Note that 100%, 99.99%, and 90% of massive stars are assumed to undergo direct collapse (i.e., no ejecta) in the WR star, SMS, and TDE models, respectively, and that the remaining massive stars explode as CCSNe whose yields are taken from Nomoto et al. (2013). We confirm that all the 3 models can reproduce the observed N/O and C/O ratios of the JWST N-rich galaxies. We have also checked that the N/O and C/O ratios of the wind yields from the $10^{3}\ M_{\odot}$ SMS (Nagele & Umeda, 2023) are also comparable to those of the JWST N-rich galaxies. These findings suggest the presence of high-$z$ galaxies affected by WR stars, SMSs, and/or TDEs with frequent direct collapses.

It should be noted that it is unclear that all high-$z$ galaxies follow these scenarios with the significant amount of CNO-cycle materials because N/O ratios of the other high-$z$ galaxies are not constrained well. Figure 8 shows N/O ratios of our sample galaxies as a function of $M_{\mathrm{UV}}$ and $S/N$ ratio of H$\delta$. The $M_{\mathrm{UV}}$ values of our sample galaxies and GN-z11 are taken from Nakajima et al. (2023) and Harikane et al. (2023a), respectively. We find that all of our sample galaxies (semi-transparent white circle) but the JWST N-rich galaxies have poor upper limits above the solar abundance. We also note that all the 3 JWST N-rich galaxies have large values of $M_{\mathrm{UV}}$ and/or $S/N$ ratio of H$\delta$ relative to the other galaxies in our sample. These results suggest that we cannot tell whether the other galaxies in our sample have high N/O ratios consistent with the scenarios with dominant CNO-cycle materials, or whether the galaxies have low N/O ratios comparable to those of the CCSN yields, but are not constrained simply because the galaxies are not bright enough and/or the exposure time is not long enough.

We also investigate the typical N/O ratio of high-$z$ galaxies by stacking the JWST spectra. We select JWST galaxy spectra with $R\sim 1000$ and at $z=5.0$–9.4 to cover O iii] and [O iii]. The total number of such spectra reaches $\sim 30$. We perform median stacking on the $\sim 30$ spectra and 100 times bootstrap resampling to estimate the error of the stacked spectrum. We simply fit the Gaussian function to each emission line in the stacked spectrum. We find that neither N iii] nor O iii] is detected in the stacked spectrum. We then estimate an upper limit of the N/O ratio from N iii]/[O iii]. The (opaque) white circle in Figure 8 shows the N/O upper limit, together with the median $M_{\mathrm{UV}}$ value of the used galaxies and the $S/N$ ratio of H$\delta$. We find that the N/O upper limit of the stacked spectrum is still higher than the solar abundance value, which means that we cannot conclude whether even the typical N/O ratio of high-$z$ galaxies exceeds the solar abundance.

The left panel of Figure 9 shows Ne/O ratios of our sample galaxies (semi-transparent red circle) as a function of $12+\log(\rm O/H)$. Although the majority of our sample galaxies have Ne/O ratios comparable to those of local dwarf galaxies (gray dot; Izotov et al. 2006; Kojima et al. 2021; Isobe et al. 2022), 4 of our sample galaxies (GLASS_100003, CEERS_00698, CEERS_01143, and CEERS_01149) have $\log(\rm Ne/O)$ values of $<-1.0$, which are significantly lower than those of the local dwarf galaxies and the other our sample galaxies. The right panel of Figure 9 illustrates the Ne/O ratios as a function of redshift. Interestingly, all the 4 galaxies with the low Ne/O ratios are located at $z>6$.

Here, we explore the origins of the low Ne/O. First, the CNO cycle cannot explain the low Ne/O ratios because the CNO cycle reduces the oxygen abundance and maintains the neon abundance. The 4 galaxies with the low Ne/O ratios have only upper limits of N/O, which means that there is also a possibility that these 4 galaxies have low N/O ratios comparable to those of the CCSN yields (see also the last paragraph of Section 4.3). Since CCSNe can produce large amounts of neon and oxygen, we consider below the contribution of CCSNe to the low Ne/O ratios.

We compare these data points with (Watanabe et al., 2023)’s chemical evolution models accumulating the CCSN yields calculated by the nucleosynthesis code of Tominaga et al. (2007) under the assumption of Kroupa (2001) IMF. We assume the metallicity evolution of the Milky Way (Suzuki & Maeda, 2018), and convert the maximum stellar ages of the models (Age_mod) to $12+\log(\rm O/H)$. The blue curves in Figure 9 correspond to the chemical evolution models that accumulate CCSN ejecta with fixed metallicities of $Z=0$, 0.001, and 0.004. We also illustrate isochrones of the models at $\log(\rm Age_{\rm mod}/yr)=6.7$, 6.8, 6.9, and 7.0, which correspond to lifetimes of stars with 40, 30, 25, and 20 $M_{\odot}$, respectively (Portinari et al., 1998). As shown in the top left panel of Figure 9, the models predict an increase in Ne/O with Age_mod. This increase can originate from the prediction that CCSNe with more massive progenitors have higher temperatures in the carbon-burning layer (e.g., Woosley & Janka, 2005), which reduces neon abundance.

As shown in the top left panel of Figure 9, we find that the models with $M_{\rm prog}\gtrsim 30\ M_{\odot}$ reproduce the low Ne/O ratios of the 4 galaxies. This finding agrees with the fact that the 4 galaxies are located at higher redshifts beyond 6, because galaxies at higher redshifts are expected to be younger and/or have top-heavy IMFs due to the metal-poorer environment.