Genesis-Metallicity: Universal Non-Parametric Gas-Phase Metallicity Estimation

TL;DR

This work tackles the challenge of universal, non-parametric gas-phase metallicity estimation when direct-temperature diagnostics are incomplete. It introduces genesis-metallicity, a KDE-based framework that learns the joint distribution of strong-line observables in a multi-dimensional space and infers metallicity without parametric forms, while also providing a Te(OII) estimator to enable direct-method metallicities in cases where lines permit. Leveraging a large calibration set of 1510 galaxies with direct-method metallicities—including 122 new $z>1$ measurements—the authors demonstrate a robust, redshift-independent metallicity accuracy of about $0.09$ dex for the strong-line estimator. The approach is validated on JWST/NIRSpec and ground-based data, and the authors publicly release the software and calibration data to enable widespread application in upcoming surveys. Overall, this work provides a versatile, data-driven path to reliable gas-phase metallicities across cosmic time, reducing biases introduced by parametric strong-line calibrations.

Abstract

We introduce genesis-metallicity, a gas-phase metallicity measurement python software employing the direct and strong-line methods depending on the available oxygen lines. The non-parametric strong-line estimator is calibrated based on a kernel density estimate in the 4-dimensional space of O2 = [O II]$λλ3727,29$/H$β$; O3 = [O III]$λ5007$/H$β$; H$β$ equivalent width EW(H$β$); and gas-phase metallicity $12 + \log$(O/H). We use a calibration sample of 1510 galaxies at $0 < z < 10$ with direct-method metallicity measurements, compiled from the JWST/NIRSpec and ground-based observations. In particular, we report 122 new NIRSpec direct-method metallicity measurements at $z > 1$. We show that the O2, O3, and EW(H$β$) measurements are sufficient for a gas-phase metallicity estimate that is more accurate than 0.09 dex. Our calibration is universal, meaning that its accuracy does not depend on the target redshift. Furthermore, the direct-method module employs a non-parametric ${\rm T}_{\rm e}$(O II) electron temperature estimator based on a kernel density estimate in the 5-dimensional space of O2, O3, EW(H$β$), ${\rm T}_{\rm e}$(O III), and ${\rm T}_{\rm e}$(O II). This ${\rm T}_{\rm e}$(O II) estimator is calibrated based on 1004 spectra with detections of both [O III]$λ4363$ and [O II]$λλ7320,30$, notably reporting 20 new NIRSpec detections of the [O II]$λλ7320,30$ doublet. We make genesis-metallicity and its calibration data publicly available and commit to keeping both up-to-date in light of the incoming data.

Figures (3)

• Figure 1: Overview of the calibration sample used in this work. This includes 1510 spectra with direct-method metallicity measurements, including 171 galaxies observed with the NIRSpec MSA (orange), 1213 galaxies observed with ground-based instruments (blue and green), and 126 spectra generated by stacking the SDSS spectra.
• Figure 2: Directly measured ${\rm T}_{\rm e}$(O$\;$) and ${\rm T}_{\rm e}$(O$\;$) for the 1004 spectra where both measurements are available. Each data point is color-coded with its corresponding O3 measurement. The ${\rm T}_{\rm e}$(O$\;$) and ${\rm T}_{\rm e}$(O$\;$) seem to be correlated for average galaxies, with a large scatter that is captured by the O3 value. The x- and y-axis limits are chosen to optimize data visualization; a small number of data points lie outside the displayed range. The full dataset is available in machine-readable format (see Table \ref{['table: overview']}).
• Figure 3: Evaluating the accuracy of the ${\rm T}_{\rm e}$(O$\;$) estimator. Here, we show the ${\rm T}_{\rm e}$(O$\;$) estimates vs. the values measured directly from the [O$\;$]$\lambda\lambda3727,29$/[O$\;$]$\lambda\lambda7320,30$ ratios. The ${\rm T}_{\rm e}$(O$\;$) is estimated employing a kernel density estimation of the probability density function in the 5-dimensional space of O2, O3, EW(H$\beta$), ${\rm T}_{\rm e}$(O$\;$), and ${\rm T}_{\rm e}$(O$\;$). The ${\rm T}_{\rm e}$(O$\;$) estimator is more accurate than 0.04 dex, defined as the absolute estimate vs. directly measured ${\rm T}_{\rm e}$(O$\;$) offset that contains $68\%$ of the estimates. The accuracy of the ${\rm T}_{\rm e}$(O$\;$) estimator declines to 0.1 dex at ${\rm T}_{\rm e}$(O$\;$) $> 14000$ K, where the parameter space is sparsely sampled by the calibration data (see Figure \ref{['fig: temperatures']}).