O32 Ratio in Nebular Spectroscopy

• O32 Ratio is defined as the ratio of high-ionization [O III] emission to low-ionization [O II] emission, with varying measurement conventions.
• It serves as a diagnostic for the ionization state in H II regions and is used to infer properties like ionization parameter and density-bounded conditions.
• Variations in dust correction, spectral lines used, and geometry highlight its role as a high-information yet non-unique tool in nebular and metallicity diagnostics.

The O32 ratio is a nebular oxygen excitation diagnostic that compares emission from doubly ionized oxygen to emission from singly ionized oxygen. In extragalactic spectroscopy it is used to characterize the ionization state of H II regions and narrow-line gas, and it is frequently employed as a proxy for the ionization parameter, for density-bounded versus ionization-bounded nebular structure, and for the likelihood of Lyman-continuum (LyC) escape. The quantity is not defined with a single universal convention: some studies use only [O III] $\lambda5007$ in the numerator, others use the sum of [O III] $\lambda4959+\lambda5007$; some express O32 linearly, others as $\log_{10}$ of the ratio. Across these conventions, the underlying observable is the same comparison between high- and low-ionization oxygen emission, and the major interpretive issue is likewise the same: O32 is informative but not one-to-one with any single physical parameter (Izotov et al., 2017, Morelli et al., 2012, Cleri et al., 28 May 2026).

1. Definition and measurement conventions

O32 is defined in several closely related ways in the literature, reflecting differences in observational setup, spectral coverage, and calibration strategy.

Convention	Expression	Representative usage
Linear, single [O III] line	$$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$	LyC-leaker studies and MUSE surveys
Linear, summed [O III] lines	$$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$	Strong-line metallicity work, Green Pea studies, CEERS
Logarithmic form	$$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$	$z\sim1$ dwarf-galaxy and AGN analyses

The single-line convention appears explicitly in studies of compact star-forming galaxies and LyC leakers, for example $$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$ (Izotov et al., 2017). A summed-[O III] convention is used in metallicity and ionization analyses, for example ${\it O}_{32}=([\ion{O}{iii}]\lambda4959+[\ion{O}{iii}]\lambda5007)/[\ion{O}{ii}]\lambda3727$ (Morelli et al., 2012), and in Green Pea work as $$O32\equiv [\mathrm{O\,III}]\lambda5007+\lambda4959/[\mathrm{O\,II}]\lambda3727,3729$$ (Khasnovis et al., 4 Sep 2025). High-redshift surveys also use a logarithmic version, such as $\lambda4959+\lambda5007$0 (Pharo et al., 2023).

Measurement practice varies correspondingly. Some studies form O32 from extinction-corrected line intensities (Izotov et al., 2017, Izotov et al., 7 May 2026), some from dust-corrected fluxes using Balmer decrements (Reddy et al., 2023), and some from rest-frame equivalent widths when spectra are not flux-calibrated and H$\lambda4959+\lambda5007$1 is unavailable (Morelli et al., 2012). The [O II] term may denote an unresolved blended feature or the summed $\lambda4959+\lambda5007$2 doublet, and some authors note explicitly that alternative [O III]-numerator conventions differ only by a fixed rescaling because $\lambda4959+\lambda5007$3 (Cleri et al., 28 May 2026).

2. Physical interpretation as an ionization diagnostic

In the standard interpretation, O32 is mostly sensitive to the ionization because it tracks the balance between O$\lambda4959+\lambda5007$4 and O$\lambda4959+\lambda5007$5 emission (Morelli et al., 2012). Larger O32 means that the nebula is dominated more strongly by high-ionization gas, which is usually associated with a harder radiation field, a higher ionization parameter, low metallicity, or some combination of these conditions (Izotov et al., 2017, Izotov et al., 7 May 2026).

A central physical idea recurring across the literature is the distinction between ionization-bounded and density-bounded H II regions. If the outer low-ionization zone is truncated because the gas runs out before all ionizing photons are absorbed, [O II] is reduced relative to [O III], and O32 rises. This makes high O32 a plausible signature of porous or density-bounded nebulae, although not a unique one (Izotov et al., 2017, Khasnovis et al., 4 Sep 2025).

In high-redshift statistical work, O32 is often treated as an empirical proxy for the ionization parameter $\lambda4959+\lambda5007$6. CEERS defines

$\lambda4959+\lambda5007$7

and for a radiation-bounded spherical nebula writes

$\lambda4959+\lambda5007$8

after relating the Strömgren radius to the ionizing photon production rate $\lambda4959+\lambda5007$9, electron density $\log_{10}$0, and filling factor $\log_{10}$1 (Reddy et al., 2023). In that framework, O32 is not merely a descriptive ratio; it is an observational handle on the coupled effects of photon production, gas density, and geometry.

The inverse case is also instructive. In the circumgalactic nebula “Oxyster,” O32 remains below 1 throughout the nebula, with line luminosities implying an integrated O32 of about 0.16. There the low ratio is interpreted as evidence for a low-ionization nebula, a relatively soft and/or weak ionizing field, and $\log_{10}$2 in the AGN CLOUDY grids considered by the authors (Lu et al., 15 Apr 2025).

3. O32 and Lyman-continuum escape

O32 became especially prominent as a LyC-leakage diagnostic in studies of compact, low-mass star-forming galaxies. In J1154+2443, the extinction-corrected ratio is $\log_{10}$3, the escape fraction is $\log_{10}$4, and the authors report that, when combined with previous low-redshift LyC-leaker measurements, $\log_{10}$5 increases with increasing O32. They give the regression

$\log_{10}$6

and interpret the result as support for the view that compact low-mass star-forming galaxies with high O32 ratios can lose a substantial fraction of their ionizing radiation to the IGM (Izotov et al., 2017).

A closely related line of argument appears in Green Pea work that adopts O32 = 10 as a practical dividing line. In a sample of 60 Green Peas, the H I 21 cm detection rate is $\log_{10}$7 for the 32 galaxies with O32 $\log_{10}$8 but only $\log_{10}$9 for the 28 galaxies with O32 $$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$0. The high-O32 subsample also has lower H I mass, lower H I-to-stellar mass ratio, and shorter H I depletion timescale, which is interpreted as evidence that the strongest expected LyC leakers are H I-poor because much of their neutral gas has been consumed in the starburst (Khasnovis et al., 4 Sep 2025).

At the same time, the literature repeatedly emphasizes that O32 is not a sufficient diagnostic by itself. A sample of five compact star-forming galaxies with extremely high O32 values from 23 to 43 was selected precisely because such galaxies were thought to be promising LyC leakers, yet the conclusion was that high O32 may not be a sufficient condition for LyC leakage; the authors proposed He I $$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$1 and $$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$2 ratios as more discriminating optical diagnostics (Izotov et al., 2017). A pilot MOSFIRE study likewise argued for a lack of correlation between [O III]/[O II] and LyC escape fraction: the sample has $$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$3, but the data show cases of high O32 with $$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$4 consistent with zero and low O32 with nonzero tentative escape, with clumpy geometry, mergers, and shocks highlighted as possible explanations (Bassett et al., 2018).

This caution is strengthened by the HETDEX Mg II-selected sample. Its 14 galaxies span O32 $$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$5–8.51, and the strongest inferred LyC emitter has an O32 about an order of magnitude below previously known strong leakers. The conclusion there is explicit: low O32 does not rule out strong LyC escape, and using O32 alone may miss bona fide strong leakers (Salazar et al., 2023).

4. Relation to metallicity diagnostics and abundance calibration

Long before O32 became closely associated with LyC-leaker selection, it was already a standard strong-line diagnostic used together with $$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$6 in nebular abundance work. In intermediate-redshift galaxies, $$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$7 is defined as

$$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$8

while

$$\mathrm{O}_{32} = [\mathrm{O\,III}]\,\lambda5007 / [\mathrm{O\,II}]\,\lambda3727$$9

In this framework, $$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$0 is the metallicity-sensitive index and O32 is the ionization-sensitive index that helps check branch selection in the degenerate $$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$1–metallicity relation (Morelli et al., 2012).

Extreme O32 systems later became important calibration anchors at the very metal-poor end. J2229+2725 has $$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$2, and J1046+4047 has O32 $$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$3; both are described as among the most extreme low-metallicity dwarf star-forming galaxies known (Izotov et al., 2021, Izotov et al., 2023). Their importance lies not only in extreme excitation but also in their leverage on O32-dependent abundance calibrations.

For J2229+2725, the earlier strong-line relation

$$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$4

was modified to

$$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$5

with

$$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$6

so that the extreme O32 regime could be accommodated in abundance work (Izotov et al., 2021). J1046+4047 then extended the calibration further by introducing

$$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$7

where

$$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$8

The revised fit is summarized as appropriate for galaxies with O32 $$\mathrm{O}_{32} = ([\mathrm{O\,III}]\,\lambda4959+\lambda5007)/[\mathrm{O\,II}]\,\lambda3727,3729$$9 and extremely low metallicity (Izotov et al., 2023).

These studies also highlight an observational caveat that matters directly for O32: extinction treatment can dominate the inferred value. In J1046+4047, H$$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$0 is found to be enhanced by non-recombination processes and therefore unsuitable for deriving the extinction coefficient; including it yields an anomalously high $$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$1 and biases the derived oxygen abundance low by about 0.1 dex (Izotov et al., 2023).

5. Empirical regimes, sample statistics, and cosmic evolution

O32 has been measured across a wide range of galaxy populations, from nearby compact dwarfs to representative high-redshift star-forming samples. In a MUSE survey at $$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$2, O32 is defined as $$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$3, and among 406 star-forming galaxies above the redshift-dependent star-forming main sequence, 104 have O32 $$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$4 and 15 have O32 $$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$5. The study reports no significant correlation between $$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$6, SFR, or distance from the SFR–$$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$7 relation with O32, and no evidence for redshift evolution in the fraction of high-O32 emitters across that interval once metallicity effects are accounted for (Paalvast et al., 2018).

At $$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$8, the HALO7D and DEEPWinds dwarf-galaxy analysis uses O32 jointly with Ne3O2. The dwarf composite stacks span O32 values from about 0.16 down to $$\mathrm{O32} = \log_{10}([\mathrm{OIII}]/[\mathrm{OII}])$$9 in logarithmic units, and the principal empirical result is that the $z\sim1$0 dwarf composites have higher O32 at fixed Ne3O2 than local galaxies. The authors interpret this as evidence that the decline in ionization parameter seen from $z\sim1$1 to today had largely ceased by $z\sim1$2 (Pharo et al., 2023).

JWST/NIRSpec studies extend the O32-based ionization analysis to the reionization era. In CEERS, 48 galaxies at $z\sim1$3–6.3 were split into equal-number low- and high-O32 bins with mean values $z\sim1$4 and $z\sim1$5. The high-O32 composite has $z\sim1$6, at least a factor of $z\sim1$7 higher than the low-O32 sample, while $z\sim1$8 is statistically indistinguishable between the bins. The same study finds a highly significant correlation between $z\sim1$9 and $$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$0, with

$$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$1

for 22 galaxies with size measurements (Reddy et al., 2023).

The RUBIES survey generalizes this to a much larger high-redshift population. Using O32-based Cloudy inference for 434 galaxies at $$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$2, the study finds that $$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$3 increases with redshift and sSFR, decreases with stellar mass, and still increases by a factor of $$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$4 from $$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$5 to $$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$6 at fixed stellar mass and fixed sSFR. Equally important, it reports a systematic uncertainty floor of $$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$7 dex in $$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$8 at zero measurement uncertainty because many photoionization models predict the same O32 without informative priors (Cleri et al., 28 May 2026).

6. Use in AGN and other ionized environments, and the main caveats

Although O32 is often discussed in the context of compact star-forming galaxies, it is also used as an ionization-sensitive parameter in AGN and circumgalactic environments. In X-ray selected zCOSMOS AGNs, O32 is defined as $$\mathrm{O}_{32}=[\mathrm{O\,III}]\,\lambda5007/[\mathrm{O\,II}]\,\lambda3727$$9 and is used to compare obscured versus unobscured and high- versus low-excitation sources. The reported mean values are ${\it O}_{32}=([\ion{O}{iii}]\lambda4959+[\ion{O}{iii}]\lambda5007)/[\ion{O}{ii}]\lambda3727$0 for low-excitation obscured AGNs, ${\it O}_{32}=([\ion{O}{iii}]\lambda4959+[\ion{O}{iii}]\lambda5007)/[\ion{O}{ii}]\lambda3727$1 for high-excitation obscured AGNs, ${\it O}_{32}=([\ion{O}{iii}]\lambda4959+[\ion{O}{iii}]\lambda5007)/[\ion{O}{ii}]\lambda3727$2 for low-excitation unobscured AGNs, and ${\it O}_{32}=([\ion{O}{iii}]\lambda4959+[\ion{O}{iii}]\lambda5007)/[\ion{O}{ii}]\lambda3727$3 for high-excitation unobscured AGNs. The clearest trend is that unobscured AGNs show a steeper and tighter O32–${\it O}_{32}=([\ion{O}{iii}]\lambda4959+[\ion{O}{iii}]\lambda5007)/[\ion{O}{ii}]\lambda3727$4 correlation than obscured AGNs (Bornancini et al., 11 Jul 2025).

In He II-selected galaxy samples, O32 separates star-forming and AGN-like excitation in a different way. For purely star-forming galaxies, O32 is tightly correlated with ${\it O}_{32}=([\ion{O}{iii}]\lambda4959+[\ion{O}{iii}]\lambda5007)/[\ion{O}{ii}]\lambda3727$5,

${\it O}_{32}=([\ion{O}{iii}]\lambda4959+[\ion{O}{iii}]\lambda5007)/[\ion{O}{ii}]\lambda3727$6

with a standard deviation of 0.18 dex, while O32 does not correlate with He II/H${\it O}_{32}=([\ion{O}{iii}]\lambda4959+[\ion{O}{iii}]\lambda5007)/[\ion{O}{ii}]\lambda3727$7 in those star-forming galaxies. By contrast, AGNs occupy a separate high-O32 region and show He II/H${\it O}_{32}=([\ion{O}{iii}]\lambda4959+[\ion{O}{iii}]\lambda5007)/[\ion{O}{ii}]\lambda3727$8 increasing monotonically with O32 (Kouroumpatzakis et al., 5 Mar 2025).

The principal caveat that unifies these diverse applications is that O32 is not controlled by ionization parameter alone. The papers surveyed here repeatedly identify metallicity, stellar metallicity, hardness of the ionizing spectrum, gas density, dust attenuation, internal dust absorption of ionizing photons, geometry, filling factor, shocks, mergers, and LyC escape itself as confounding influences (Bassett et al., 2018, Pharo et al., 2023, Reddy et al., 2023, Cleri et al., 28 May 2026). This suggests that O32 is best regarded as a high-information but non-unique strong-line diagnostic. It remains exceptionally useful for selecting unusual systems and for population studies, but robust inference for individual objects generally requires additional lines, external priors, or explicit photoionization modeling.