The JWST EXCELS Survey: A spectroscopic investigation of the ionizing properties of star-forming galaxies at 1<z<8

Abstract

Charting the Epoch of Reionization demands robust assessments of what drives the production of ionizing photons in high-redshift star-forming galaxies (SFGs), and requires better predictive capabilities from current observations. Using a sample of $N=159$ SFGs at $1<z<8$, observed with deep medium-resolution spectroscopy from the JWST/NIRSpec EXCELS survey, we perform a statistical analysis of their ionizing photon production efficiencies ($ξ_\rm{ion}$). We consider $ξ_\rm{ion}$, measured with Balmer line measurements, in relation to a number of key galaxy properties including; nebular emission line strengths ($W_λ(\rm{Hα})$ and $W_λ$( [OIII])), UV luminosity ($M_\rm{UV}$) and UV slope ($β_\rm{UV}$), as well as dust attenuation ($E(B-V)_\rm{neb}$) and redshift. Implementing a Bayesian linear regression methodology, we fit $ξ_\rm{ion}$ against the principal observables while fully marginalising over all measurement uncertainties, mitigating against the impact of outliers and determining the intrinsic scatter. Significant relations between $ξ_\rm{ion}$ and $ W_λ(\rm{Hα})$, $W_λ$([OIII]) and $β_\rm{UV}$ are recovered. Moreover, the weak trends with $M_\rm{UV}$ and redshift can be fully explained by the remaining property dependencies. Expanding our analysis to multivariate regression, we determine that $W_λ(\rm{Hα})$ or $W_λ$([OIII]), along with $β_\rm{UV}$ and $E(B-V)_\rm{neb}$, are the most important observables for accurately predicting $ξ_\rm{ion,0}$. The latter identifies the most common outliers as SFGs with relatively high $E(B-V)_\rm{neb}\gtrsim0.5$, possibly indicative of obscured star-formation or strong differential attenuation. Combining these properties enable $ξ_\rm{ion,0}$ to be inferred with an accuracy of $\sim0.15\,$dex, with a population intrinsic scatter of $σ_\rm{int}\sim0.035\,$dex.

Figures (19)

• Figure 1: Example spectroscopy available for a galaxy in the EXCELS sample (ID: EXCELS-59720). The G235M/F170LP and G395M/F290LP spectra from EXCELS for this SFG are shown (left and right, respectively), highlighting the plethora of rest-frame optical emission lines (including but not limited to; [O iii], $\mathrm{H\alpha}$, $\mathrm{H\beta}$, $\mathrm{H\gamma}$, $\mathrm{H\delta}$; see insets) potentially observable with the deep NIRSpec spectroscopy.
• Figure 2: The observed Balmer decrement (see Section \ref{['subsubsec:physicalproperties:dustcorrection']} for details) distributions for the EXCELS galaxy sample ($N=159$) subsets with $\mathrm{H\alpha/H\beta}$ ($N=105$; top panel), $\mathrm{H\gamma/H\beta}$ ($N=121$; centre panel) and $\mathrm{H\delta/H\beta}$ ($N=112$; bottom panel). The inset panels show the distributions of measurement errors ($\mathrm{\sigma}$) for each Balmer decrement pair $\mathrm{(H_X/H_Y)}_\mathrm{obs}$, with typical errors of $\simeq^{+0.24}_{-0.15}$ for $\mathrm{H\alpha/H\beta}$, and of $\simeq^{+0.06}_{-0.07}$ for both $\mathrm{H\gamma/H\beta}$ and $\mathrm{H\delta/H\beta}$. The dashed lines and preceding grey shaded regions illustrate the intrinsic Balmer line ratios under the Case B recombination assumptions employed here.
• Figure 3: A corner plot of the six key galaxy observables $\mathbf{X}$ (where $X_i=\{ W_\lambda(\mathrm{H\alpha}),\, W_\lambda([\mathrm{O\,III}]),\, \beta_{\mathrm{UV}},\, M_{1500},\, E(B-V),\, z_\mathrm{spec} \}$, shown in order as columns left to right, and as rows top to bottom), from which we derive relationships with $\xi_\mathrm{ion,0}$ in Section \ref{['sec:analysis']}. Along the diagonal axes, we show the 1D marginal histograms for each covariate, highlighting the large dynamic range of physical properties spanned by our spectroscopic sample. Markers throughout are coloured by the inferred ionizing photon production efficiency ($\xi_\mathrm{ion,0}$; here specifically derived from $\mathrm{H\alpha}$, where available), with the histogram bars along the diagonal axes indicating the median $\xi_\mathrm{ion,0}$ in the given bin. In each panel,the star symbol represents the property medians and $16^\mathrm{th}-84^{\rm th}$ percentile range. Additionally, we present the Spearman rank correlation coefficient ($\rho_\mathrm{S}$) and associated $p-$value between parameter pairs, alongside the $95$ per cent confidence intervals on $\rho_\mathrm{S}$, computed with Pingouinvallat+18. As shown in Section \ref{['sec:analysis']}, the apparent strength of the predictive power for a single parameter $X_i$ on $\xi_\mathrm{ion,0}$ can significantly weaken as a result of the collinearity between the observables in the multivariate case (see also Table \ref{['tab:correlations_table']}).
• Figure 4: The distribution of ionizing photon production efficiencies ($\mathrm{\xi_\mathrm{ion,0}}$) for the sample of $N=159$ EXCELS galaxies (light grey), calculated from the intrinsically brightest available Balmer line from $\mathrm{H\,\alpha}$, $\mathrm{H\,\beta}$, $\mathrm{H\,\gamma}$ (see Section \ref{['subsubsec:physicalproperties:ionizingphotonproduction']} for details). From the total sample, $N=112$ are computed from $\mathrm{H\,\alpha}$, with a remaining $N=34$ from $\mathrm{H\,\beta}$ and $N=13$ from $\mathrm{H\,\gamma}$ (with each plotted individually in purple, blue and green, respectively). The insets show the associated full $\xi_\mathrm{ion,0}$ distributions from each Balmer line (in galaxies where they are measurable), highlighting the excellent consistency across each. The coloured thick dotted lines indicate the distribution medians, and the black dashed vertical lines mark the canonical $\mathrm{log_{10}(\xi_\mathrm{ion,0}\,/\,erg^{-1}Hz})\simeq25.2$ value often cited as the benchmark "typical" limit for galaxies to have driven reionization completely robertson+13
• Figure 5: Heat map of the Spearman rank correlation ($\rho$) matrix for the six key predictors used in this work: $W_\lambda(\mathrm{H\alpha}),\: W_\lambda(\mathrm{[O\,III]}),\: \beta_\mathrm{UV},\: M_\mathrm{UV},\: E(B-V),\: z_\mathrm{spec}$. The lower panels (below the diagonal; denoted with a $*$) show the partial$-\rho$ which removes the affect of all other covariates, while the upper panels (above the diagonal; denoted with a $^\dagger$) display the ('Skipped Spearman') robust$-\rho$, which mitigates against the impact of outliers. All quantities, including the $p-$values displayed below the coefficient at the centre of each panel, are computed using Pingouinvallat+18. In particular, the heat map brings attention to the mild-to-moderate (and in some cases very strong, e.g., between $\mathrm{H\,\alpha}$ and [O iii] equivalent width) collinearity between the predictors, which may impact the regression parameters.