D RAFT VERSION S EPTEMBER 20, 2022

Typeset using LATEX twocolumn style in AASTeX631

CLASSY IV: Exploring UV diagnostics of the interstellar medium in local high-z analogs at the dawn of the JWST era∗

M ATILDE M INGOZZI , 1 B ETHAN L. JAMES , 2 K ARLA Z. A RELLANO -C ÓRDOVA , 3 DANIELLE A. B ERG , 3

P ETER S ENCHYNA , 4 J OHN C HISHOLM , 3 JARLE B RINCHMANN , 5 A LESSANDRA A LOISI , 1 R ICARDO O. A MORÍN , 6, 7
S TÉPHANE C HARLOT , 8 A NNA F ELTRE , 9 M ATTHEW H AYES , 10 T IMOTHY H ECKMAN , 11 A LAINA H ENRY , 1, 11
S VEA H ERNANDEZ , 2 N IMISHA K UMARI , 2 C LAUS L EITHERER , 1 M ARIO L LERENA , 6, 7 C RYSTAL L. M ARTIN , 12
T HEMIYA NANAYAKKARA , 13 S WARA R AVINDRANATH , 1 E VAN D. S KILLMAN , 14 Y UMA S UGAHARA , 15, 16, 17
arXiv:2209.09047v1 [astro-ph.GA] 19 Sep 2022

A IDA W OFFORD , 18 AND X INFENG X U 11

1 Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA
2 AURA for ESA, Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA
3 Department of Astronomy, The University of Texas at Austin, 2515 Speedway, Stop C1400, Austin, TX 78712, USA
4 Carnegie Observatories, 813 Santa Barbara Street, Pasadena, CA 91101, USA
5 Instituto de Astrofísica e Ciências do Espaço, Universidade do Porto, CAUP, Rua das Estrelas, PT4150-762 Porto, Portugal
6 Instituto de Investigación Multidisciplinar en Ciencia y Tecnología, Universidad de La Serena, Raul Bitrán 1305, La Serena 2204000, Chile
7 Departamento de Astronomía, Universidad de La Serena, Av. Juan Cisternas 1200 Norte, La Serena 1720236, Chile
8 Sorbonne Université, CNRS, UMR7095, Institut d’Astrophysique de Paris, F-75014, Paris, France
9 INAF - Osservatorio di Astrofisica e Scienza dello Spazio di Bologna, Via P. Gobetti 93/3, 40129 Bologna, Italy
10 Stockholm University, Department of Astronomy and Oskar Klein Centre for Cosmoparticle Physics, AlbaNova University Centre, SE-10691, Stockholm,

Sweden
11 Center for Astrophysical Sciences, Department of Physics & Astronomy, Johns Hopkins University, Baltimore, MD 21218, USA
12 Department of Physics, University of California, Santa Barbara, Santa Barbara, CA 93106, USA
13 Swinburne University of Technology, Melbourne, Victoria, AU
14 Minnesota Institute for Astrophysics, University of Minnesota, 116 Church Street SE, Minneapolis, MN 55455, USA
15 Institute for Cosmic Ray Research, The University of Tokyo, Kashiwa-no-ha, Kashiwa 277-8582, Japan
16 National Astronomical Observatory of Japan, 2-21-1 Osawa, Mitaka, Tokyo 181-8588, Japan
17 Waseda Research Institute for Science and Engineering, Faculty of Science and Engineering, Waseda University, 3-4-1, Okubo, Shinjuku, Tokyo 169-8555,

Japan
18 Instituto de Astronomía, Universidad Nacional Autónoma de México, Unidad Académica en Ensenada, Km 103 Carr. Tijuana-Ensenada, Ensenada 22860,

México

Submitted to AASJournal ApJ

ABSTRACT
The COS Legacy Archive Spectroscopic SurveY (CLASSY) HST/COS treasury program provides the first
high-resolution spectral catalogue of 45 local high-z analogues in the ultra-violet (UV; 1200 − 2000 Å) to in-
vestigate their stellar and gas properties. Here we present a toolkit of UV interstellar medium (ISM) diagnos-
tics, analyzing the main emission lines of CLASSY spectra (N IV] λλ1483,87, C IV λλ1548,51, He IIλ1640,
O III]λλ1661,6, Si III] λλ1883,92, C III] λ1907,9). Specifically, our aim is to provide accurate diagnostics for
reddening E(B − V ), electron density ne , electron temperature Te , metallicity 12+log(O/H) and ionization pa-
rameter log(U), taking into account the different ISM ionization zones. We calibrate our UV toolkit using well-
known optical diagnostics, analyzing archival optical spectra for all the CLASSY targets. We find that UV den-
sity diagnostics estimate ne values that are ∼ 1−2 dex higher (e.g., ne (C III]λλ1907,9) ∼ 104 cm−3 ) than those in-
ferred from their optical counterparts (e.g., ne ([S II]λλ6717,31) ∼ 102 cm−3 ; ne ([Ar IV]λλ4714,41) ∼ 103 cm−3 ).
Te derived from the hybrid ratio [O III] λ1666/λ5007 proves to be reliable, implying differences in determining
12+log(O/H) compared to the optical counterpart O III] λ4363/[O III] λ5007 within ∼ ±0.3 dex. We also inves-

Corresponding author: Matilde Mingozzi

[email protected]
∗ Based on observations made with the NASA/ESA Hubble Space Telescope,
obtained from the Data Archive at the Space Telescope Science Institute,
which is operated by the Association of Universities for Research in As-
tronomy, Inc., under NASA contract NAS 5-26555.
2 M INGOZZI ET AL .

tigate the relation between the stellar and gas E(B −V ), finding consistent values at high specific star formation
rates (log(sSFR) & −8 yr−1 ), while at low sSFRs we confirmed an excess of dust attenuation in the gas. Finally,
we investigate UV line ratios and equivalent widths to provide correlations with 12+log(O/H) and log(U), but
note there are degeneracies between the two. With this suite of UV-based diagnostics, we illustrate the pivotal
role CLASSY plays in understanding the chemical and physical properties of high-z systems that JWST can
observe in the rest-frame UV.

Keywords: Dwarf galaxies (416), Ultraviolet astronomy (1736), Galaxy chemical evolution (580), Galaxy spec-
troscopy (2171), High-redshift galaxies (734), Emission line galaxies (459)

1. INTRODUCTION the capacity to provide the community with a ‘diagnostic

The galaxies that host a substantial fraction of the star toolkit’, with which we can directly diagnose the ISM prop-
formation (SF) in the high-z universe (z & 6) and likely erties in star-forming galaxies.
play a key role in the reionization era tend to be com- Due to the intrinsic faintness of several UV emission lines,
pact, metal-poor, with a low-mass and large specific star an alternative form of ‘in-direct’ diagnostics, which evolve
formation rates (e.g., Wise et al. 2014; Robertson et al. from empirical calibrations between ISM conditions (e.g.,
2015; Madau & Haardt 2015; Stark 2016; Stanway et al. metallicity) and properties of the stronger emission line prop-
2016). Deep rest-frame UV spectra of several of these erties (e.g., equivalent widths of C III]), is also needed. To
high-redshift galaxies (z ∼ 5 − 7) already revealed prominent this end, several past studies have taken big steps forward in
high-ionization nebular emission lines, such as He II λ1640, the interpretation of UV emission in the local Universe. For
O III] λλ1661, 66, [C III] λ1907 and C III] λ1909 (C III] here- example, Rigby et al. (2015) showed that C III] can be used
after) and C IV λλ1548, 1551 (C IV hereafter; e.g., Stark to pick out low-metallicity galaxies with strong bursts of star-
et al. 2015; Mainali et al. 2017; 2018). In the upcoming era formation, whereas Senchyna et al. (2017) suggested that
of the James Webb Space Telescope (JWST) and extremely nebular He II and C IV emission has the potential to constrain
large telescopes (ELTs), the UV spectroscopic frontier will metallicity. Additionally, Senchyna et al. (2019) demon-
be pushed to higher redshifts than ever before, finally re- strated that C IV emission is ubiquitous in extremely metal
vealing detailed rest-frame UV observations of statistically- poor systems with very high specific star formation rates -
significant samples of galaxies in the distant Universe. As albeit with equivalent widths smaller than those measured at
such, the time to sharpen our understanding of UV nebular high-z. With regards to the strength of the ionizing radia-
emission and exploit its diagnostic power is upon us. tion, Ravindranath et al. (2020) found a strong correlation
Far- and Near-Ultraviolet (FUV, ∼ 1200 − 1700 Å; NUV, between C III] and O III] emission and the O32 ratio (a proxy
∼ 1700 − 2000 Å) spectra can foster our understanding of for the ionization parameter), confirming that a hard radia-
star-forming galaxies in terms of the stellar populations host- tion field is required to produce the high-ionization nebular
ing massive stars and their impact on interstellar medium lines. Using two nearby extreme UV emitting galaxies, Berg
(ISM) physical conditions, chemical evolution, feedback et al. (2019a) showed us that a combination of strong C IV
processes, and reionization. Due to the line production and He II emission may identify galaxies that not only pro-
mechanisms alone, nebular UV emission can be use to di- duce but also transmit a substantial number of high-energy
rectly calculate the physical and chemical conditions un- photons - i.e., potential contributors to cosmic reionization
der which they are produced. For instance, both C III] and (see also Schaerer et al. 2022). While each of these stud-
[Si III] λ1883, Si III] λ1892 (Si III] hereafter) doublets are di- ies provided a significant step-forward in understanding the
rect tracers of electron density (Jaskot & Ravindranath 2016; conditions required for UV emission, these works have been
Gutkin et al. 2016; Byler et al. 2018), the line intensity ratio limited to single peculiar objects or small nearby samples and
C III]/O III]λ1666 can be used to estimate the elemental car- lack the large statistics that we need to interpret the high vol-
bon abundances (Garnett et al. 1995; Berg et al. 2016; Pérez- ume of high-z UV spectroscopy that will arrive in the next
Montero & Amorín 2017; Berg et al. 2018), and He II λ1640 decade.
and the C IV/C III] ratio both have the potential to constrain Statistically larger rest-UV spectroscopic studies do exist,
the level of ionization (Feltre et al. 2016). Moreover, the typically targeting 2 < z < 4 star-forming galaxies. For in-
combination of all these UV lines can provide information stance, using deep VLT/MUSE spectroscopy, Maseda et al.
about the nature of the ionizing sources in general (Feltre (2017) collected a sample of 17 unlensed C III] emitters at
et al. 2016; Gutkin et al. 2016; Jaskot & Ravindranath 2016; 1.5 . z . 4 which provided an unbiased sample toward the
Nakajima et al. 2018). UV emission lines therefore have lowest mass, bluest galaxies. Stacked spectra of 15 gravi-
 CLASSY IV 3

tationally lensed galaxies at redshifts 1.68 < z < 3.6 from Spectroscopic SurveY (CLASSY) treasury (Berg et al. 2022;
project MEGaSaURA by Rigby et al. (2018), produced a new James et al. 2022, Paper I and Paper II hereafter) repre-
spectral composite of star-forming galaxies at redshift z ∼ 2 sents the first high-quality (S/N1500 & 5 per resolution ele-
which clearly revealed strong C III] and Mg II λλ2796,2803 ment, resel), high-resolution (R ∼ 15, 000) and broad wave-
as well as weaker lines, such as He II, and Si III]. Llerena length range (∼ 1200 − 2000 Å) UV database of 45 nearby
et al. (2022) exploited a broader representative sample of (0.002 < z < 0.182) star-forming galaxies. These objects
217 C III] emitters (∼ 30% of the total sample) from the were selected to include properties similar to reionization-
VANDELS survey (McLure et al. 2018), collecting main- era systems, in terms of specific star-formation rate, direct
sequence galaxies at z ∼ 2 − 4 to investigate their aver- gas-phase metallicity, ionization level, reddening, and nebu-
age properties using the spectral stacking technique. Fi- lar density (see Paper I for more details). Moreover, optical
nally, Schmidt et al. (2021) presented an even larger sam- observations are in-hand for all the galaxies of the sample,
ple, collecting 2052 spectroscopically confirmed emission allowing us to make detailed comparisons of UV and optical
line galaxies at 1.5 . z . 6.4, providing line properties of the diagnostics. As such, CLASSY provides the ideal UV atlas
main UV lines, and subsequently confirming the wealth of with which we can tailor our UV diagnostic toolkit.
information and physical properties that rest-frame UV emis- This is the first in a series of two CLASSY papers in which
sion features red-wards of Lyα can probe. These works cur- we present a FUV-based toolkit and show how this compares
rently represent our most comprehensive rest-FUV spectral to well-known optical diagnostics. Specifically, in this work
datasets at high redshift. However, the majority of them have we provide a detailed calculations of dust attenuation, elec-
focused mainly on C III] emitters, since C III] the strongest tron density ne , electron temperature Te , gas-phase metallic-
UV emission line after Lyα, and it is extremely challeng- ity 12+log(O/H) and ionization parameter log(U), using both
ing to obtain the required high signal-to-noise (S/N) to detect UV and optical direct diagnostics, taking into account the
fainter lines even employing the stacking technique. Mod- different ionization zones of the ISM. Then, from their com-
erate spectral resolution (R ∼ 18000) and broad wavelength parison, we provide a set of diagnostic equations to estimate
coverage are also necessary to fully investigate the potential ISM properties only from UV emission lines. In Sec. 2 we
of UV diagnostics. Also, the limited wavelength range avail- describe the CLASSY sample, covering both the UV and op-
able for each of these studies has prevented us from carrying tical data, while in Sec. 3 we present the spectroscopic anal-
out a comparison of multi-wavelength diagnostics for ISM ysis, including stellar continuum and emission line fitting.
properties within the same targets. In Sec. 4 we discuss the chemical and physical diagnostics
Indeed, in order for us to derive an accurate and detailed used in our analysis, showing and comparing the derived ISM
UV toolkit, we not only need to cover the full UV regime, but properties in Sec. 5. Then, in Sec. 6, we introduce and dis-
also optical wavelengths. Historically, ISM tracers have re- cuss our UV-based toolkit, providing also a comparison with
lied heavily on optical diagnostics, and as such they are very previous works. Finally, in Sec. 7 we summarize our main
well calibrated. A crucial step in understanding the condi- findings.
tions that produce UV emission would therefore be compar- The data presented in this paper were obtained from the
ing UV line strengths with ISM conditions derived from pre- Mikulski Archive for Space Telescopes (MAST) at the Space
existing optical diagnostics within the same targets, to effec- Telescope Science Institute. The specific observations ana-
tively calibrate a toolkit that depends solely on UV emission lyzed can be accessed via 10.17909/m3fq-jj25. All the prod-
lines. This aspect is particularly important because the en- ucts of this paper (UV and optical line fluxes; UV and opti-
tire optical wavelength range on which our current diagnostic cal z; UV-optical flux offsets; ISM properties, i.e., E(B −V ),
toolkit relies (from [O II]λλ3727,9 to [S III]λ9069]), that is ne , Te , optical and UV 12+log(O/H)) will be provided on
easily accessible in the local Universe, will not be available the CLASSY MAST webpage as downloadable tables. In
for sources in the Reionization epoch. Specifically, JWST App. B, C and D we show which information will be pro-
instruments such as NIRSpec will cover blueward of 7000 Å vided. Throughout this paper, we adopt a the solar metallicity
and 4500 Å only in objects between z ∼ 6 and z ∼ 10, respec- scale of Asplund et al. (2009), where 12+log(O/H) = 8.69.
tively. As such, a UV toolkit will be essential for characteriz-
ing and interpreting the spectroscopic observations of high-z 2. SAMPLE PRESENTATION
systems.
CLASSY is a sample of 45 star-forming UV-
The ideal framework from which a UV toolkit can be
bright (mFUV < 21AB arcsec−2 ), relatively compact
built would consist of high S/N spectra with the possibility
(GFW HMNUV < 2.5”) galaxies in the local Universe
of extensive wavelength coverage that spans UV to optical
(0.002 < z < 0.182), spanning a wide range of stel-
wavelengths. Each of these essential elements are offered
lar masses (6.22 < log(M? /M ) < 10.06), star formation
by local high-z analogs. In this context, the COS Legacy
rates (−2 < log(SFR/M yr−1 ) < +2), oxygen abundances
4 M INGOZZI ET AL .

(6.98 < 12+log(O/H) < 8.77), electron densities (10 < CLASSY combines 135 orbits of new HST data (PID:
ne /cm−3 < 1120), degree of ionization (0.54 < O3O2 < 38.0, 15840, PI: Berg) with 177 orbits of archival HST data, for
with O3O2 = [O III] λ5007/[O II] λλ3727,9), and reddening a total of 312 orbits. In order to achieve nearly-panchromatic
values (0.001 < E(B − V ) < 0.673). This broad sampling of FUV spectral coverage with the highest spectral resolution
parameter space makes the CLASSY sample representative possible, CLASSY combines the G130M, G160M, G185M,
of star-forming galaxies across all redshifts, with a bias to- G225M and G140L gratings, spanning from 1150 Å to 2100–
wards more extreme O3O2 values, low stellar masses, and 2500 Å to allow synergistic co-spatial studies of stars and gas
high SFRs, typical of high-z systems (see Paper I). In Paper I, within the same galaxy.
we presented our sample, explaining in detail the selection Each HST/COS grating has a different spectral resolu-
criteria and giving an extensive overview of the HST/COS tion that must be accounted for when combining data from
and archival optical spectra. To summarize, from the Hub- multiple gratings. This coaddition process is explained in
ble Spectral Legacy Archive (HSLA), 101 nearby (z < 0.2) Sec. 2.3 of Paper I and in Paper II, which presents all
galaxies were selected on the basis of the high signal-to-noise the details of this multi-stage technical process, concern-
(& 7 per 100 km/s resolution element) COS spectroscopy in ing extracting, reducing, aligning, and coadding the spec-
at least one medium resolution grating (i.e., G130M, G160M, tra from the different gratings. This paper focuses on the
or G185M), applying further selection criteria to assemble analysis of all the emission lines (except for Lyα) in the
a high-quality, comprehensive rest-frame set of FUV spec- range 1150 − 2000 Å. We used the so-called High Resolu-
tra for a large and diverse sample of star-forming galaxies. tion (HR: G130M+G160M; R ∼ 10000 − 24000) and Mod-
Specifically, any targets with secondary classifications or erate Resolution (MR: G130M+G160M+G185M+G225M;
visually confirmed spectra features of quasi-stellar object R ∼ 10000 − 20000) co-added spectra with a dispersion of
(QSO) or Seyfert were removed. The data reduction has 12.23 mÅ/pixel, and a resolution of 0.073 Å per resolution
been presented in detail in Paper II, including spectra extrac- element (Å/resel, where 1 resel equates to 6 native COS pix-
tion, co-addition, wavelength calibration, and vignetting. els), and 33 mÅ/pixel and 0.200 Å/resel, respectively.
In this work, we take into account the properties of the For the galaxies J1044+0353 and J1418+2102, in-
CLASSY galaxies in terms of redshift, stellar mass, SFR, stead of the MR co-added spectra, we used the so-
12+log(O/H), and galaxy half-light radius (r50 ), estimated called Low Resolution (LR: G130M+G160M+G140L or
from Paper I, which for completeness we also show in Ta- G130M+G160M+G185M+G225M+G140L; R ∼ 1500 −
ble 1. Specifically, the redshifts have been taken from SDSS 4000) coadds, with a nominal point source resolution of 80.3
where available, r50 was estimated from PanSTARRS imag- mÅ/pixel or 0.498 Å/resel. Additionally, the COS G185M
ing, while the stellar masses and SFRs have been estimated and G225M observations for the J1112+5503 galaxy were
from the spectra energy distribution (SED) fitting via the impacted by guide star failures and thus we excluded this
BayEsian Analysis of GaLaxy sEds (BEAGLE, Chevallard galaxy from the sample.
& Charlot 2016), as explained in Paper I (see Sec. 4.7). Fi- We performed the stellar continuum subtraction with the
nally, the calculation of 12+log(O/H) is explained in detail in method described in Sec. 3.1.1 from the HR spectra, binned
Paper I Sec. 4.5, and is based on the direct Te method, using by 15 native COS pixels. We also fit the MR coadded spectra,
[S II] λ6717/λ6731 and [O III] λ4363/λ5007 as electron den- after binning them by 6 native COS pixels, in order to fit all
sity and temperature tracers, respectively. The UV redshifts the main emission lines not covered by the HR coadds. Fi-
zUV instead are obtained from the analysis of UV emission nally, we doubly rebinned both configurations (i.e., binning
lines, and are part of the results of this paper. the HR and MR coadds by 30 and 12 native COS pixels, re-
As described in Paper I, both UV and optical CLASSY spectively), to improve the fit of the faintest emission lines
spectra have been corrected for the total Galactic foreground when possible.
reddening along the line of sight of their coordinates using
the PYTHON dustmaps (Green 2018) interface to query the 2.2. Optical data
Bayestar 3D dust maps of Green et al. (2015). The Green
High-quality optical spectra have been collected for the en-
et al. (2015) map was adopted over more recent versions due
tire CLASSY sample to ensure uniform determinations of
to its more optimal coverage of the CLASSY sample. The
galaxy properties and to allow comparisons between proper-
Galactic foreground reddening correction was then applied
ties derived from optical and UV diagnostics, thus enabling
using the Cardelli et al. (1989) reddening law.
an accurately calibrated suite of UV diagnostics.
In the following, we briefly summarize the UV and optical
DR7 APO/SDSS spectra with a 3.0" aperture exist for 38
data sample and properties.
of the CLASSY galaxies (Abazajian et al. 2009), while for
one galaxy, J1444+4237, there are DR13 BOSS spectrograph
2.1. UV data data with a 2.0" aperture (Albareti et al. 2017; Guseva et al.
 CLASSY IV 5

Table 1. CLASSY sample main properties.

Tot. log M? log SFR
Target Name zlit. zUV (M ) (M yr−1 ) 12+log(O/H) r50
0.18 0.14
1. J0021+0052 0.09839 ··· 9.09±0.38 +1.07±0.11 8.17±0.07 0.784
0.26 0.19
2. J0036-3333 Haro 11 knot 0.02060 ··· 9.14±0.23 +1.01±0.21 8.21±0.17 2.846
3. J0127-0619 Mrk 996 0.00540 0.00547 8.74±0.18
0.15
0.15
−0.75±0.13 7.68±0.02 2.374
4. J0144+0453 UM133 0.00520 0.00533 7.65±0.24
0.29 −0.81±0.29
0.46 7.76±0.02 2.851
5. J0337-0502 SBS0335-052 E 0.01352 0.01346 7.06±0.24
0.21 −0.32±0.07
0.11 7.46±0.04 1.433
6. J0405-3648 0.00280 ··· 6.61±0.28
0.28 −1.81±0.31
0.27 7.04±0.05 3.557
7. J0808+3948 0.09123 ··· 9.12±0.30
0.17 +1.26±0.18
0.25 8.77±0.12 1.114
8. J0823+2806 LARS9 0.04722 0.04741 9.38±0.33
0.19 +1.48±0.15
0.32 8.28±0.01 2.134
9. J0926+4427 LARS14 0.18067 0.18000 8.76±0.30
0.26 +1.03±0.13
0.13 8.08±0.02 0.889
10. J0934+5514 I zw 18 NW 0.00250 0.00264 6.27±0.15
0.20 −1.52±0.09
0.07 6.98±0.01 2.606
11. J0938+5428 0.10210 0.10210 9.15±0.18
0.29 +1.05±0.20
0.17 8.25±0.02 1.095
12. J0940+2935 0.00168 ··· 6.71±0.23
0.40 −2.01±0.42
0.37 7.66±0.07 5.151
13. J0942+3547 CG-274, SB 110 0.01486 0.01482 7.56±0.21
0.29 −0.76±0.19
0.12 8.13±0.03 1.328
14. J0944-0038 CGCG007-025, SB 2 0.00478 0.00487 6.83±0.44
0.25 −0.78±0.19
0.16 7.83±0.01 0.984
15. J0944+3442 0.02005 0.02005 8.19±0.40
0.23 −0.01±0.28
0.65 7.62±0.11 2.458
16. J1016+3754 1427-52996-221 0.00388 0.00390 6.72±0.27
0.22 −1.17±0.18
0.18 7.56±0.01 1.835
17. J1024+0524 SB 36 0.03319 0.03326 7.89±0.37
0.24 +0.21±0.14
0.12 7.84±0.03 1.325
18. J1025+3622 0.12650 0.12717 8.87±0.25
0.27 +1.04±0.14
0.18 8.13±0.01 0.843
19. J1044+0353 0.01287 0.01286 6.80±0.41
0.26 −0.59±0.11
0.14 7.45±0.03 1.204
20. J1105+4444 1363-53053-510 0.02154 0.02147 8.98±0.29
0.24 +0.69±0.28
0.22 8.23±0.01 2.646
21. J1112+5503 0.13164 ··· 9.59±0.33
0.19 +1.60±0.20
0.25 8.45±0.06 0.920
22. J1119+5130 0.00446 0.00444 6.77±0.15
0.28 −1.58±0.21
0.12 7.57±0.04 1.870
23. J1129+2034 SB 179 0.00470 0.00467 8.09±0.37
0.27 −0.37±0.38
0.56 8.28±0.04 3.098
24. J1132+5722 SBSG1129+576 0.00504 0.00504 7.31±0.23
0.26 −1.07±0.27
0.35 7.58±0.08 2.249
25. J1132+1411 SB 125 0.01764 0.01760 8.68±0.28
0.19 +0.44±0.24
0.27 8.25±0.01 7.289
26. J1144+4012 0.12695 0.12700 9.89±0.18
0.29 +1.51±0.20
0.29 8.43±0.20 1.158
27. J1148+2546 SB 182 0.04512 0.04522 8.14±0.34
0.24 +0.53±0.17
0.14 7.94±0.01 0.874
28. J1150+1501 SB 126, Mrk 0750 0.00245 0.00246 6.84±0.28
0.30 −1.33±0.29
0.23 8.14±0.01 1.760
29. J1157+3220 1991-53446-584 0.01097 0.01101 9.04±0.32
0.18 +0.97±0.21
0.42 8.43±0.02 2.894
30. J1200+1343 0.06675 0.06699 8.12±0.47
0.42 +0.75±0.20
0.16 8.26±0.02 0.908
0.34 0.26
31. J1225+6109 0955-52409-608 0.00234 0.00234 7.12±0.24 −1.08±0.26 7.97±0.01 2.596
32. J1253-0312 SHOC391 0.02272 0.02267 7.65±0.51
0.23 +0.56±0.15
0.15 8.06±0.01 1.079
33. J1314+3452 SB 153 0.00288 0.00282 7.56±0.30
0.21 −0.67±0.23
0.55 8.26±0.01 1.765
34. J1323-0132 0.02246 0.02246 6.31±0.26
0.10 −0.72±0.08
0.09 7.71±0.04 0.698
35. J1359+5726 Ly 52, Mrk 1486 0.03383 0.03381 8.41±0.31
0.26 +0.42±0.20
0.14 7.98±0.01 1.395
0.32 0.21
36. J1416+1223 0.12316 ··· 9.59±0.26 +1.57±0.25 8.53±0.11 0.985
37. J1418+2102 0.00855 0.00858 6.22±0.49
0.35 −1.13±0.15
0.16 7.75±0.02 1.130
38. J1428+1653 0.18167 ··· 9.56±0.15
0.23 +1.22±0.26
0.19 8.33±0.05 0.933
0.35 0.11
39. J1429+0643 0.17350 0.17340 8.80±0.21 +1.42±0.17 8.10±0.03 0.859
40. J1444+4237 HS1442+4250 0.00230 0.00220 6.48±0.17
0.17 −1.94±0.11
0.08 7.64±0.02 2.760
41. J1448-0110 SB 61 0.02741 0.02744 7.61±0.41
0.24 +0.39±0.13
0.14 8.13±0.01 1.070
42. J1521+0759 0.09426 ··· 9.00±0.29
0.30 +0.95±0.16
0.17 8.31±0.14 0.983
0.28
43. J1525+0757 0.07579 ··· 10.06±0.42 +1.00±0.69
0.24 8.33±0.04 1.319
44. J1545+0858 1725-54266-068 0.03772 0.03772 7.52±0.43
0.26 +0.37±0.13
0.17 7.75±0.03 1.075
45. J1612+0817 0.14914 ··· 9.78±0.28
0.26 +1.58±0.28
0.24 8.18±0.19 0.878
(1) (2) (3) (4) (5) (6) (7) (8)

N OTE— CLASSY sample properties derived from UV+optical photometry and spectra. Columns 1 and 2 indicate the target name used in this
work and alternative names, respectively. Columns 3 and 4 give the target redshift from the literature and FUV UV emission-lines, respectively.
Columns 5 and 6 list the total stellar masses and SFRs derived from Beagle SED fitting in Paper I. Column 7 gives the oxygen abundances
derived in Paper I. Column 8 lists the galaxy half-light radius (r50 ), estimated from PanSTARRS imaging, from Paper I.
6 M INGOZZI ET AL .

2017). These spectra are in the wavelength range of 3800 − Paper V compares the SDSS, LBT and MUSE integrated
9200Å (3600 − 10400Å for BOSS), with a spectral resolution spectra for the galaxies with multiple observations, demon-
of R ∼ 1500 − 2500 (Eisenstein et al. 2011). strating that flux calibration issues or aperture differences
For the remaining galaxies of the sample (J0036-3333, do not introduce significant discrepancies in the optical ISM
J0127-0619, J0337-0502, J0405-3648, J0934+5514, and properties in terms of gas attenuation, density, temperature,
J0144+0453), we used integral field spectroscopy data, when metallicity and SFRs. This result supports our comparison of
available, or long-slit spectroscopy, instead of SDSS. Specif- the physical properties obtained using these different sets of
ically, we used VLT/VIMOS integral field unit (IFU) from optical data. The UV and optical fluxes, highlighting which
James et al. (2009) for J0127-0619, MMT Blue Chan- telescope and instrument was considered for each galaxy, as
nel Spectrograph spectra from Senchyna et al. (2019) for well as the products of the analysis of this paper will be pro-
J0144+0453, Keck/KCWI IFU spectra from Rickards Vaught vided on the CLASSY MAST webpage as downloadable ta-
et al. (2021) for J0934+5514, and VLT/MUSE IFU spectra bles. In App. B, C and D we show which information will be
for the remaining three galaxies. MUSE spectra are also provided.
available for the galaxies J0021+0052 (PI: Göran Östlin),
J1044+0353 and J1418+2102 (PI: Dawn Erb). We used these 3. DATA ANALYSIS
data to retrieve emission line-ratios involving faint auroral The UV and optical spectra were analyzed making use of
lines, if undetected (S/N < 3) in SDSS spectra. Finally, for a set of customized python scripts in order to first fit and
the galaxies J0808+3948, J0944-0038, J1148+2546, J1323- subtract the stellar continuum and then fit the main emis-
0132 and J1545+0808, Multi-Object Double Spectographs sion lines with multiple Gaussian components where needed.
(MODS) data from the LBT telescope, presented in Arellano- This allowed us to estimate the stellar population properties
Córdova et al. 2022, are also available (Paper V hereafter). (i.e., age, metallicity and stellar dust attenuation), emission-
Information on each of the optical datasets is provided in the line properties (fluxes, velocities, velocity dispersions, and
following paragraphs. equivalent widths), the UV-optical flux offset (discussed in
Concerning the IFU data available for J0021+0052, J0036- Appendix A), and ISM gas properties. In the following, all
3333, J0127-0619, J0337-0502, J0405-3648, J0934+5514, the steps are explained in detail.
J1044+0353 and J1418+2102, we extracted a spectrum from
a 2.5" aperture centred at the same coordinates of COS ob- 3.1. Stellar continuum
servations to match the COS aperture (see also Paper I and 3.1.1. UV spectra
Paper V). Specifically, the integrated VIMOS spectrum of
The analysis of UV spectra relies on a robust stellar contin-
J0127-0619 is obtained combining the high-resolution blue
uum fitting procedure both for determining the properties of
and orange grisms, covering the wavelength range 4150 −
the stellar population and for accurately measuring nebular
7400 Å with a spectral resolution R ∼ 1150 − 2150 (see
UV emission lines, such as He II λ1640 or C IV λλ1548,51.
James et al. 2009 for more details). The KCWI spectrum
For the purposes of subtracting the UV stellar continuum in
of J0934+5514 is in the wavelength range 3500 − 5600 Å at
the HR spectra1 for this paper, we compare the results of two
a median spectral resolution of R ∼ 3600. Finally, MUSE
sets of fits which will be described in detail in Senchyna et al.
spectra are in the wavelength range 4300 − 9300 Å at a spec-
(2022a) (S22 hereafter). Both fits considered here are based
tral resolution of R ∼ 2000 − 3500.
upon a flexible linear combination of spectra of SSPs span-
Regarding the long-slit data, the MMT spectrum of
ning a wide range of metallicities and ages, as described by
J0144+0453 was taken with the 300 lines/mm grating with
Chisholm et al. (2019), and assume a Reddy et al. (2016) at-
a 10"×1" slit, oriented along the parallactic angle to mini-
tenuation law. The primary difference between the two sets
mize slit losses (see Senchyna et al. 2019 for more details).
of results is the stellar population synthesis framework used
The wavelength coverage is 3200 − 8000 Å with a resolu-
to generate the basis of SSP spectra. The first uses the S TAR -
tion R ∼ 740. For the galaxies J0808+3948, J0944-0038,
BURST 99 theoretical UV predictions described by Leitherer
J1148+2546, J1323-0132 and J1545+0808, instead of SDSS,
et al. (1999; 2010), while the second relies on the latest ver-
we took advantage of the MODS data from LBT obtained us-
sion of the Bruzual & Charlot (2003) models (S. Charlot & G.
ing the G400L and G670L. MODS long-slit data were taken
Bruzual, in-preparation, hereafter C&B; see also Gutkin et al.
with a 60"×1" slit, with an extraction aperture of 2.5"×1",
2016; Vidal-García et al. 2017; Plat et al. 2019). These pop-
and a slit orientation along the parallactic angle (see Paper V
for more details). The wavelength coverage extends from
3200 Å to 10000 Å with a moderate spectral resolution of 1 It is not necessary to fit the stellar continuum in the MR spectra, since
R ∼ 2000. the NUV range does not contribute significantly to the stellar population
analysis, and in that range there are no significant absorption or resonant
emission lines due to stars.
 CLASSY IV 7

ulation synthesis models adopt different prescriptions for the fully checking that the subtraction of either S TARBURST 99
evolution and atmospheres of massive stars, resulting in par- or C&B stellar continuum best-fit gave similar results for our
ticularly significant differences for lines such as He II λ1640 emission line fitting, we ultimately decided to use the C&B
that can be powered in the dense optically-thick winds of best-fit, since it takes into account the stellar He IIλ1640 con-
very luminous stars (e.g. Senchyna et al. 2021). tribution. Moreover, the C&B models can be extended to
In addition to stellar light and dust attenuation, the other wavelengths of ∼ 9000 Å in the optical, which allowed us
crucial constituent of the UV light of star-forming galaxies to perform accurate flux-scaling between the optical and UV
is the nebular continuum. Both sets of models include the spectra.
contribution of the nebular continuum computed in a self- The full flux-scaling analysis is described in App. A, where
consistent manner and assuming a closed geometry, as de- we explain our method to properly scale the flux of the op-
scribed in Leitherer et al. (1999) and Gutkin et al. (2016); Plat tical spectra to the UV. In summary, a flux offset between
et al. (2019), respectively. The S TARBURST 99 predictions do COS and the optical spectra is expected since they have
not include variable parameters describing this emission, but been obtained via different instruments with slightly differ-
the Cloudy-computed nebular continuum for the C&B mod- ent apertures and pointing position. The median value that
els are presented at varying log(U) which can have an impact we find for this UV-optical flux offset is 0.78 ± 0.03, and we
in the UV (see e.g., Senchyna et al. 2022b). Our fiducial as- report the value obtained for each galaxy on the CLASSY
sumed volume-averaged log(U) = −2.5 (defined as in Gutkin MAST webpage as shown in App. C. We multiplied the ob-
et al. 2016) represents the median value inferred for the full served optical spectra of each CLASSY galaxy by its corre-
CLASSY sample from fitting the UV continuum with differ- sponding UV-optical flux offset to correct them. We high-
ent log(U) in the range [−3; −1], and it is also typical of the light that this flux correction has no impact on the proper-
values inferred from fitting the nebular line emission of sim- ties derived from the flux-ratios within each galaxy, but only
ilar local star-forming galaxies (Plat et al. 2019; Senchyna when ratios between UV and optical emission lines (e.g.,
et al. 2022b). However, we stress that the choice of fixing O III]λ1666/[O III]λ5007) are considered.
this parameter has a minimal impact on the fidelity of the UV
continuum fits, with a negligible median difference in the re- 3.1.2. Optical spectra
duced chi-square (i.e., . 0.01; S22). Moreover, Chisholm
et al. (2019) explored variations with the log(U) and density, Since the UV stellar continuum models were optimized
finding no changes in the shape of the nebular continuum and for the young stellar population, it was not feasible to use
in the relative contribution of the nebular/stellar continuum the UV models to perform a self-consistent fit of the opti-
ratio over the expected log(U) range (see also Byler et al. cal wavelength portion of the spectrum, due to the dominant
2017). contribution from the older population of stars in this wave-
In most other respects, the fits proceed in a similar manner. length regime (e.g., Leitherer et al. 1999). Thus being, in
The observed HR spectra are fitted after first rebinning by 15 order to remove any stellar absorption components present in
pixels and after smoothing the models with a Gaussian ker- the Balmer emission lines, we model the optical stellar con-
nel to best represent the achieved resolution and S/N. In both tinuum using Starlight2 spectral synthesis code of Cid
cases, we adopt the maximum initial mass function (IMF) Fernandes et al. (2005) and the stellar models of Bruzual &
upper mass cutoff provided for the models (using Kroupa Charlot (2003) with the IMF of Chabrier (2003). The set
2001 and Chabrier 2003, for S TARBURST 99 and C&B, re- of the stellar models taken into account comprises 25 ages
spectively); this is 100 M for S TARBURST 99 and 600 M (1 Myr – 18 Gyr) and six metallicities (0.05 < Z?/Z < 2.5).
for C&B (see e.g. Plat et al. 2019; Senchyna et al. 2022b). It should be noted that while the Starlight models do not
The uncertainties in the fits were calculated via a Monte include a nebular continuum component, the nebular contin-
Carlo technique, modulating the observed flux with a Gaus- uum contribution in this wavelength regime is known to be
sian kernel centered on zero with a width equal to the formal negligible (<10%; Byler et al. 2017).
estimated error on the flux. To summarize, the two stellar As a preliminary step, we corrected the spectra for the
continuum fits provide independent estimates of the intrinsic Galactic foreground reddening correction (see Sec. 2.1), and
stellar E(B−V )UV reddening, and the light-weighted ages and uniformly sampled the rest-frame wavelength, the flux and
metallicities of the ionizing stellar populations, alongside full the error in steps of ∆λ = 1 Å. For reddening the models,
fits to the UV continuum. we used the attenuation law of Cardelli et al. (1989). The
For the purpose of this work, the stellar continuum fitting Starlight models are fitted over the wavelength range
is used to subtract the stellar contribution from the observed 3700 − 9100 Å. In Fig. 17, included in App. A, we show
UV spectra, thus allowing us to accurately measure the neb-
ular emission lines in the range 1150 − 2000 Å. After care- 2 www.starlight.ufsc.br
8 M INGOZZI ET AL .

our UV and optical stellar continuum best-fit for the galaxy where we consider the transition probability of the doublets
J0021+0052 and J1144+4012 as an example. and assumed a fixed line ratio of 0.333 between the fainter
and the brighter line (Osterbrock 1989). We did not fix the
3.2. Emission-lines
[O III] λλ4959,5007 fluxes because in the KCWI data of
The analysis of the emission lines in the UV and optical J0934+5514 and SDSS data of J1253-0312 the [O III] λ5007
spectra (after the subtraction of the best-fit stellar continuum, line is saturated. For these objects we obtained an estimate
described in Sec. 3.1) was performed separately, but with a of the [O III] λ5007 by applying the fixed line ratio of 3 with
similar approach. We simultaneously fit each spectrum tak- respect to [O III] λ4959 (Osterbrock 1989). We notice that
ing into account a set of UV and optical emission lines in the for J1253-0312 also the Hα line is clipped, so we discarded
wavelength range 1265 − 2000 Å and 3700 − 9100 Å, respec- its flux.
tively, with a linear baseline centered on zero and a single Concerning C III] λλ1907,9, we constrained the line ra-
Gaussian, making use of the code MPFIT (Markwardt 2009), tios [C III] λ1907/C III] λ1909 to vary up to 1.6 (Osterbrock
which performs a robust non-linear least squares curve fit- 1989), to avoid non-physical values. Our procedure allows us
ting. We list the final fluxes, corrected for dust reddening, to fit this doublet with two well-separated Gaussians, since
of all fitted UV and optical emission lines in the CLASSY the distance between the centroids of the two emission lines
MAST webpage, as shown in App. B and C. of the doublet is fixed. Conversely, the resolution of the
In our procedure the main Milky Way absorption lines COS/G185M (R ∼ 20000@2100Å) does allow us to resolve
were masked, and the fitting was performed only in windows the C III] doublet even after the 6-12 native pixel binning
of 3000 km s−1 centered around each emission line. We tied (i.e., ∼ 40 − 80 km/s), as long as the width of the fitted emis-
together the velocity (i.e., the line center) and in optical spec- sion lines is smaller than half of their wavelength separation
tra also the velocity dispersion (or line width) for all the emis- (i.e., ∼ 300 km/s). Among the galaxies with significant C III]
sion lines, to better constrain weak or blended features, while emission (S/N > 3), the latter condition is not satisfied in
allowing the line flux to vary freely (in general). An excep- J1044+0353 and J1418+2102, because of the lower resolu-
tion was made for the line center of the UV emission-lines tion of COS/G140L. Concerning [O II] λλ3727,29 doublet,
C III] and C IV, because they can be significantly shifted in blended in SDSS, MOD and MMT data, the distribution of
velocity with respect to the others due to their origin, as we flux between these two Gaussians is not reliable enough to
will discuss in a forthcoming paper focused on the kinemat- derive an accurate line ratio. Therefore, the [O II] ratio is de-
ics and ionization source of the gas in the CLASSY galaxies rived only for the KCWI data of J0934+5514 (the doublet is
(Mingozzi et al. 2022 in prep.; M22 hereafter). not covered by the wavelength range of MUSE).
In order to robustly determine uncertainties, we followed Another aspect we took into account in our fitting pro-
the Monte Carlo method where we perturbed N times (with cedure is that optical lines such as [O III] λ4363 and
N = 100) the observed spectra by adding to each spectral el- [Ar IV] λ4714 can suffer from contamination due to the
ement a random value drawn from a Gaussian distribution Fe II λ4360 and He I λ4714, respectively (see e.g., Curti et al.
centred on 0 with a standard deviation equal to the observed 2017; Arellano-Córdova et al. 2020). For instance, Arellano-
spectrum uncertainty. We then fitted each configuration with Córdova et al. (2020) demonstrated that the use of a contam-
MPFIT (Markwardt 2009), obtaining 100 estimates of the inated [O III] λ4363 could lead differences in metallicity of
free parameters of the fit, that are flux, velocity and velocity up to 0.08 dex. In order to mitigate this problem, we fitted
dispersion. Finally, we calculated the 50th (i.e., the median) these faint features simultaneously with the other emission
and the (50th –16th ) and (84th –50th ) percentiles of the distribu- lines, tying them to the brighter He I λ4471 and Fe II λ4288,
tions of the fitted perturbed spectra and of the free parameters assuming a ratio of 0.728 and 0.125 (valid at ne = 100 cm−3
of the fit. The median of each free parameter is considered and Te = 104 K, from PyNeb), respectively3 .
as best-fit value, with a lower and upper uncertainty given
by the sum in quadrature of the (50th –16th ) and (84th –50th ) 3.2.2. Multi-component fitting
percentiles, divided by the square root of N, and the MPFIT After careful inspection of the optical spectra, we noticed
error. The S/N associated to each line is then defined as the that the Hα profile (in particular) shows a broad compo-
ratio between the flux and the flux uncertainty. All emission nent in many CLASSY galaxies. We therefore performed
lines with signal-to-noise higher than 3 are considered to be two-component Gaussian fits to the main optical emission
reliable detections. lines (i.e., Hδ, Hγ, [O III] λ4363, Hβ, [O III] λλ4959,5007,
3.2.1. Special constraints [N II] λ5755, [O I] λλ6300,74, [N II] λλ6548,84, Hα,

In our fitting procedure the line flux of each emission

3
line must be non-negative, but it is left free to vary, apart We fitted the Fe II λ4288 only for galaxies in which this line is visible, that
from the doublets [N II] λλ6548, 6584 and [O I] λλ6300,64, is at 12+log(O/H) & 7.7.
 CLASSY IV 9

[S II] λλ6717,31). Specifically, we took into account one For these cases, we are confident that the possible contribu-
narrow component, with an observed velocity dispersion σ < tion from broad component emission to the single (narrow)
200 km/s, that is a representative cut-off for the galaxies of component fit is negligible and within the uncertainties of the
our sample, and a broad component (σ < 1000 km/s). Their emission lines. We will investigate possible differences of the
velocity and velocity dispersion are tied to be the same for conditions of the broad component in our next paper focused
all the emission lines. To understand if the addition of a sec- on the kinematics and ionization mechanisms (M22).
ond component is significant, we calculated the reduced chi-
square χ̃2 of the single and double-component fits in the rest- 3.2.3. UV emission line detections
frame wavelength range 6540–6590 Å covering Hα (4950–
While there is a plethora of strong optical emission lines
5010 Å for J1253-0312 and J0934+5514, for which Hα is un-
that are uniformly detected throughout the sample, UV emis-
available), and chose the model with more components only
sion lines can be mostly faint and sometimes not detected at
if the χ̃2 was at least 0.1 dex smaller. This condition is satis-
all. It is therefore important for us to highlight in how many
fied in 24 out of 44 galaxies of our sample.
galaxies the UV emission lines are clearly detected. As ex-
In our UV spectra, the S/N is usually not high enough to
plained in Sec. 2.1, we fitted the HR and MR/LR Coadds
detect faint broad components in the emission lines of inter-
spectra, after performing different levels of binning. We con-
est here. However, after a visual inspection we did notice a
sider UV emission lines with S/N > 3 to be detections. We
clear broad profile in the emission lines of J1044+0353 and
took into account the results from the doubly-rebinned spec-
J1418+2102 (see also Berg et al. 2021), J1016+3754, J0337-
tra (of 30 and 12 pixels for HR and MR/LR Coadds, respec-
0502, J1323-0132 and J1545+0858. For these objects we
tively) only for the emission lines with a S/N < 3. If the
fitted the He II λ1640 and [O III]λλ1661,6 with two com-
S/N is still lower than the chosen threshold, then we con-
ponents. We tested a two-Gaussian component fitting also
sider the flux as an upper limit, while if the line is not ob-
on the C III] doublet, without finding a significant improve-
served at all, as a non-detection. As an example, in Fig. 1 we
ment in our results. This is due to the very small wavelength
show the UV emission-lines fitted by our fitting routine for
separation of the C III] doublet lines, which results in de-
the galaxy J0337-0502 (i.e., SBS0335-052 E). The spectrum
generate line centroids that make it difficult to use multi-
and the fit are reported in black and red, while the spectrum
ple components. Interestingly, for J0337-0502, J1044+0353,
and fit uncertainties are shown in shaded gray and red, re-
J1418+2102, J1323-0132 (see Fig. 2) we also observed a
spectively. The id and signal-to-noise of each zoomed line is
doubled-peak profile in the C IV doublet. As discussed
indicated on top of each panel, whose margins are coloured
in Berg et al. 2021, such profiles are the result of resonant
according to the binning applied to the spectrum before the
scattering, whereas broadening can be due to radiation trans-
fitting (HR rebinned of 15 in blue, HR rebinned of 30 in cyan,
port/scattering. Due to the different line processes responsi-
MR rebinned of 6 in dark green, MR rebinned of 12 in light
ble for C IV emission, it should be noted that the properties
green). In Fig. 2 instead we show the fitted CLASSY COS
of the multi-component fits to this line were not constrained
spectra of the C IV λλ1548,51 and C III] λλ1907,9 emission
with the same kinematics as the nebular emission lines.
lines for all the galaxies in which the lines are detected with
For the purpose of this work, we chose to only consider
S/N > 3. In App. B, in Fig. 19–Fig. 23 we show analogous
the narrow (and dominant) component of our emission lines
figures for N IV] λλ1483,87, He II λ1640, [O III] λλ1661,6,
which on average constitutes > 70% of the total flux. This
[N III] λλ1747–54 and Si III] λλ1893,92 emission lines, re-
allows us to maintain the highest accuracy in the emission
spectively, with S/N > 3. In the following we describe the
line diagnostics derived here, since each emission line com-
detection of each of these lines within the CLASSY sample.
ponent originates in gas with different physical conditions
C IV λλ1548,51 is observed in pure emission with S/N >
(ionization degree, temperature, density, velocity etc, see e.g.
3 in only 9 CLASSY galaxies (see Fig. 2), while in the other
James et al. 2009). Indeed, broad emission indicates large
galaxies it shows a P-Cygni profile or is only in absorption.
velocities that can be driven by different mechanisms such as
Generally, the C IV doublet is dominated by a broad P-Cygni
stellar winds, galactic-outflows or turbulence, and possibly
profile due to winds of luminous O stars (e.g., Shapley et al.
linked to different ionization sources, such as photoioniza-
2003; Steidel et al. 2016; Rigby et al. 2018; Llerena et al.
tion and/or shocks (e.g., Izotov & Thuan 2007; James et al.
2022). Only high resolution spectra such as those of the
2009; Amorín et al. 2012; Bosch et al. 2019; Komarova et al.
CLASSY survey can allow to successfully separate the stel-
2021; Hogarth et al. 2020).
lar and the nebular components of C IV emission (see also
It should be noted that we were unable to fit a broad com-
Crowther 2007; Quider et al. 2009). Pure nebular emission
ponent emission in the UV nebular lines of all the galaxies
in C IV λλ1548,51 has been recently detected in z > 6 targets
that displayed broad component emission in the optical due
(Stark et al. 2015; Mainali et al. 2017; Schmidt et al. 2017)
to S/N limitations and the faintness of UV emission lines.
and, rarely, in local galaxies (Berg et al. 2016; Senchyna
10 M INGOZZI ET AL .

Figure 1. UV emission-lines fitted by our fitting routine for the galaxy J0337-0502 (i.e., SBS0335-052 E): the spectrum and the fit are reported
in black and red, while the spectrum and fit uncertainties are shown in shaded gray and red, respectively. The id and signal-to-noise of each
zoomed line is indicated on top of each panel, whose margins are coloured according to the binning applied to the spectrum before the fitting
(HR rebinned of 15 in blue, MR rebinned of 6 in dark green, MR rebinned of 12 in light green). The dashed black vertical lines indicate the
observed wavelength of each line according to the redshift of the galaxy zlit , reported in Tab. 1.
 CLASSY IV 11

Figure 2. Fit of the C IV λλ1548,51 and C III] λλ1907,9 emission line doublet with S/N > 3, visible in 9 and 28 galaxies, respectively. The
observed flux and the best-fit model are shown in black and red, respectively, while their uncertainties are given by the gray and red shades.
The black dashed vertical lines indicate the line positions, taking into account zlit. . The margins are coloured according to the binning applied
to the spectrum before the fitting (HR rebinned of 15 in blue, MR rebinned of 6 in dark green, MR rebinned of 12 in light green).
12 M INGOZZI ET AL .

et al. 2017; 2019; Berg et al. 2019b; Wofford et al. 2021; estingly, only one galaxy of the CLASSY sample, J0127-
Senchyna et al. 2022b). For this paper, we only take into ac- 0619 (i.e., Mrk 996) shows this multiplet in clear emission
count only the 9 CLASSY galaxies with C IV in pure emis- with S/N ∼ 9. WR features (mainly late type WN stars)
sion, without considering the galaxies that show a P-Cygni in this galaxy were discovered for the first time by Thuan
or pure absorption line profile. All these galaxies also show et al. (1996), while their distribution as well as the ISM
C III] λλ1907,9, He II λ1640 and [O III] λ1666, apart from abundances and kinematics were investigated by James et al.
J0934+5514 (i.e., Izw 18), where the [O III] λ1666 is unde- (2009). This could indicate a non-ISM origin of this emission
tected because of a MW line contamination. (see also M22). A hint of emission with S/N ∼ 3 is observed
C III] λλ1907,9, often the strongest UV nebular emission also in J1253-0312.
line, is observed with S/N > 3 in 28 CLASSY galaxies (see Si III] λλ1893,92 is observed with S/N > 3 in 6 CLASSY
Fig. 2). Among these, we can see this doublet de-blended galaxies. These lines are generally very faint, but also in
in 26 objects (excluding J1044+0353 and J1418+2102; see many targets one of the two or both fall out of COS ob-
Sec. 3.2.1). This doublet is a density diagnostic, as we will served wavelength range (both doublet lines are observed
discuss in Sec. 5.2 and Sec. 6.2. only in J1044+0353, J1253-0312 and J1448-0110). Similarly
N IV] λλ1483,87 is observed with S/N > 3 in only 6 to C III], this doublet is a density diagnostic.
CLASSY galaxies (both doublet lines are observed only in Along with these UV emission lines we also fitted the
J1044+0353, J1253-0312 and J1545+0858). This doublet other lines shown in Fig. 1, namely: Si IIλ1265, Si IVλ1394,
has rarely been seen in emission in star-forming galaxies O IVλ1401, Si IVλ1403, O IVSi IV, S IVλ1406, O IVλ1407,
(Fosbury et al. 2003; Raiter et al. 2010; Vanzella et al. 2010; S IVλ1410M S IVλ1417, S IVλ1424, Si II∗ λ1534. Many of
Stark et al. 2014). These lines are probably due to young these lines can be visible in emission in galaxies that show
stellar populations, and, if the source is not hosting an AGN, also C IV in pure emission, as shown in Fig. 1. Due to their
they could be a signature of massive and hot stars with an lack of detection throughout the sample with enough S/N,
associated nebular emission (Vanzella et al. 2010). This dou- we do not consider these lines any further.
blet is also a density diagnostic (Keenan et al. 1995), and it
traces higher-ionization regions with respect to the C III] and 4. DERIVING PHYSICAL PROPERTIES OF THE ISM
Si III] doublets (see Sec. 5.2 and Sec. 6.2).
H II regions are stratified, with higher ionization species,
He II λ1640 is detected with S/N > 3 in 19 CLASSY
such as [Ar IV] or [O III], closer to the ionization source
galaxies and generally shows a narrow profile (median ve-
and lower ionization species, such as [S II] or [O II], in the
locity dispersion of σ ∼ 55 km/s), indicating its nebular na-
outer parts. Typically H II regions are modeled by three
ture. However, in the spectra of J0942+3547, J1129+2034,
zones of different ionization: the low-, intermediate- and
J1200+1343, J1253-0312 and J1314+3452 the line profile
high-ionization zones. As pointed out in Berg et al. (2021;
looks broader (with σ up to 200 km/s), which suggests the
2022), high-z systems and their local analogues are charac-
presence of a stellar component residual despite the removal
terized by the presence of high-energy UV and optical emis-
of the C&B best-fit stellar continuum (see e.g., Nanayakkara
sion lines due to their low metallicity and thus extreme ra-
et al. 2019; Senchyna et al. 2021).
diation fields, revealing the presence of an additional ‘very-
O III] λ1661, 6, one of the strongest UV emission lines,
high-ionization’ zone. It is important to stress that radiation
has S/N > 3 in 22 CLASSY galaxies. In J0127-0619 and
fields and metallicity are tightly linked such that stellar popu-
J1225+6109, where one of the two lines of the O III] dou-
lations of lower metallicity have harder radiation fields. This
blet is contaminated by a MW absorption line, we estimated
led Berg et al. (2021) to extend the classical 3-zone model
the flux from the other line, using a line ratio measured from
to a 4-zone model, adding the He+2 species necessary to pro-
PyNeb of 0.4 (valid at ne = 100 cm−3 and Te = 104 K). These
duce the observed He II emission via recombination (ionisa-
are auroral lines, similar to the optical [O III] λ4363, and thus
tion potential E > 54.42 eV).
can be used as a temperature diagnostics in comparison with
Overall, an accurate determination of H II region proper-
the optical nebular [O III] λλ4959,5007, as we will discuss
ties requires reliable tracers for each zone. This is because
in Sec. 5.3 and Sec. 6.3.
different ions are tracing different conditions of nebulae in
[N III] λλ1747–54 is a multiplet (i.e., a blend of emission
terms of density, temperature and ionization, since they are
at 1746.8, 1748.6, 1749.7, 1750.4, and 1752.2 Å; Keenan
not co-spatial (e.g., Nicholls et al. 2020). The COS aperture
et al. 1994). These lines are suggested to have a nebular
on CLASSY targets is covering multiple H II regions or even
origin and may be used in the so-called UV-BPT diagrams
the entire galaxy for the most compact objects. Hence, we
(Feltre et al. 2016) to discriminate between SF and AGN ac-
can employ the use of multiple diagnostics both in the op-
tivity. However, the multiplet is usually revealed in spec-
tical and in the UV to trace the conditions in the different
tra of WN-type stars (e.g., Crowther & Smith 1997). Inter-
ionization regions and compare their properties, investigat-
 CLASSY IV 13

Table 2. Optical ISM diagnostics in the different ionization zones according to the literature and available in this work.
Property Ionization Zone
Low Intermediate High Very-high
E(B −V ) Hα/Hβ, Hβ/Hγ, Hβ/Hδ Hα/Hβ, Hβ/Hγ, Hβ/Hδ Hα/Hβ, Hβ/Hγ, Hβ/Hδ Hα/Hβ, Hβ/Hγ, Hβ/Hδ

ne [S II] λ6717/λ6731 [Cl III] λ5518/λ5538 [Ar IV] λ4714/λ4741 [Ar IV] λ4714/λ4741
[O II] λ3729/λ3727 [Fe III] λ4701/λ4659

Te [N II] λ5755/λ6584 [S III] λ6312/λ9069 [O III] λ4363/λ5007 [O III] λ4363/λ5007

[S II] λλ4069,72/λλ6717,31
[O II] λλ3727,29/λλ7320,30

logU — — [S III] λλ9069,9532 / [S II] λλ6717,31 — —

— — [O III] λ5007 / [O II] λλ3727,9 — —
— — [Ar IV] λλ4714,41 / [Ar III] λ7135 — —
N OTE—ISM diagnostics available in the optical for each ionization zone in a 4-zone model (see Berg et al. 2021). Specifically, we list
diagnostic line ratios for dust attenuation (E(B −V )), electron density (ne ), electron temperature (Te ) and ionization parameter (log(U)).

Table 3. ISM UV diagnostics in the different ionization zones according to the literature and available in this work.
Property Ionization Zone
Low Intermediate High Very-high
E(B −V ) β-slope β-slope β-slope β-slope

ne ··· C III] λ1907/λ1909 — — N IV λ1483/λ1487 — —

Si III] λ1883/λ1892

Te ··· ··· O III] λ1666/[O III] λ5007 O III] λ1666/[O III] λ5007

logU — — C IV λλ1548,51/C III] λλ1907,9 — —

—— EW(C IV λλ1548,51) ——
N OTE—Same as Table 2.

ing the ISM structure of our targets with the utmost detail. dian of values and the standard deviation for uncertainties.
Here we stress that a great advantage of the CLASSY sur- Once densities and temperatures are known in each zone of
vey is the simultaneous coverage of many optical and UV the nebula, it is then possible to calculate the corresponding
diagnostic lines. In particular, UV emission lines are com- ionic abundances, with a similar iterative procedure using
ing from higher ionization zones, which gives us access to a the getIonAbundance PyNeb task, and the same method to
wider range of ionization zone tracers than typically available estimate the uncertainties.
from the optical alone. Table 2 and Table 3 show the optical and UV diagnostics
Then, we employed iteratively the PyNeb task investigated for the different ionization zones in this work.
getCrossTemDen, that combines a density and a tempera- Unfortunately in this work we lack the [Ne III] λ3342/λ3868
ture diagnostic, and ultimately converges to a final value of ratio that Berg et al. (2021) used to estimate the temperature
ne and Te . First, we calculated the intrinsic Balmer line ratios of the very-high ionization zone. We note, however, that this
using PyNeb, assuming a Case-B Hydrogen recombination ratio provided results consistent to the values obtained for the
with a starting temperature of Te = 1 × 104 K and ne = 102 high ionization zone (Berg et al. 2021). Our set of UV lines
cm−3 , considered appropriate for typical star-forming regions is characteristic of the intermediate and high-ionization zone.
(Osterbrock 1989; Osterbrock & Ferland 2006). Then, we The comparison between the different properties calculated
iteratively calculated density and temperature, using the red- with optical and UV diagnostics are shown and discussed in
dening value to correct the line ratio used as temperature Sec. 5 and Sec. 6. In the following sections we provide the
tracer, and updating at each cycle the Hα and Hβ emissiv- details about each calculated quantity.
ities (and thus E(B − V )), ne and Te , only if the new value
obtained was finite. Our iterative approach stops once the 4.1. Dust attenuation
difference in temperature between two cycles becomes lower Before comparing ratios of emission lines separated in
than 20 K. To estimate the fiducial values and errors on ne wavelength throughout the UV-optical wavelength regime,
and Te , we run the getCrossTemDen task 500 times for each the emission lines were corrected for the intrinsic galaxy dust
different combination of ne and Te diagnostics, taking the me- attenuation in terms of E(B − V ). E(B − V ) was determined
14 M INGOZZI ET AL .

comparing the observed relative intensities of the strongest Overall, looking at Fig. 3, for the low-ionization zone, the
Balmer lines available in our optical spectra (i.e., Hα/Hβ, most typical density diagnostics are [S II] λ6717/λ6731 and
Hβ/Hγ, Hβ/Hδ) with their intrinsic values. These intrinsic [O II] λ3729/λ3727, sensitive in the range ∼ 40 − 5000 cm−3 .
values depend on the density and temperature of the gas, Moving towards higher ionization potential, other density
which we estimate with the corresponding diagnostics for tracers are [Cl III] λ5518/λ5538, sensitive in the range ∼
each ionization zone, as explained in Sec. 4.2 and Sec. 4.3. 102 − 2 × 104 cm−3 , or [Si III] λ1883/Si III] λ1892 and
The final reddening estimate is an error-weighted average of [C III] λ1907/C III] λ1909 in the UV, tracing values in the
the Hα/Hβ, Hγ/Hβ, and Hδ/Hβ reddening values. range ∼ 103 − 2 × 105 cm−3 . Moreover, Méndez-Delgado
To correct the optical emission lines, we applied the et al. (2021) proposed the use of [Fe III] λ4701/λ4659,
Cardelli et al. (1989) reddening law with Rv = 3.1, which is sensitive in the range 103 − 106 cm−3 . At the highest
appropriate for the CLASSY emission line fluxes (Berg et al. ionization levels (E > 40 eV), possible diagnostics are
2022). Indeed, Wild et al. (2011b) found that the nebular [Ar IV] λ4714/λ4741 in the optical and N IV] λ1483/λ1487
attenuation curve has a slope similar to the MW attenuation in the UV, sensitive up to ne ∼ 1 × 105 cm−3 and ne ∼
curve, rather than that of the SMC (Gordon & Clayton 1998; 1 × 106 cm−3 . respectively.
Gordon et al. 2003) or the one from Calzetti et al. (2000). In general, the predictions by PyNeb accurately represent
The UV emission lines, instead, were corrected assuming the line ratios observed within the uncertainties, which can
a Reddy et al. (2016) attenuation curve with Rv = 2.191, unfortunately be very large for some transitions. In these
which represents the first spectroscopic measurement of the cases, our measurements can be considered upper limits of
shape of the far-UV dust attenuation curve for galaxies at the density. Moreover, from Fig. 3, it is clear that the den-
high-redshift (z ∼ 3), i.e., systems that are analogous to our sity diagnostics have generally a very low dependence on
CLASSY sample. Te . The highest dependence on temperature is found for the
Dust attenuation can be also estimated from comparing the N IV] λ1483/λ1487 line ratio, whose derived densities can be
observed slope of the UV spectra in the range 1400 − 1800Å different up to ∼ 1 dex, with the highest values derived for
(the ‘β slope’; see e.g., Leitherer et al. 1999; Calzetti et al. the lowest temperatures and vice-versa.
1994) with the intrinsic slope of the models used in the best- The main aspect highlighted by Fig. 3 is that the density
fit of the stellar populations (Calzetti et al. 2000; Reddy range traced by the different diagnostics can vary consider-
et al. 2016). This quantity (E(B − V )UV hereafter) is given ably. This depends on the critical density, that is defined as
as an output of the UV stellar continuum fitting described in the density at which collisional transitions are equally prob-
Sec. 3.1.1. E(B − V )UV represents the stellar attenuation and able with radiative transitions (Osterbrock 1989; Osterbrock
its relation with the gas E(B − V ) derived from the Balmer & Ferland 2006). Hence, transitions with higher critical den-
decrement is not trivial, as it is discussed in Sec. 6.1. sities can be used as diagnostics in denser environments. In-
terestingly, we noticed that higher critical densities do not
4.2. Density automatically correspond to higher ionization (see upper pan-
els of Fig. 3), which means that the density structure could
The electron density ne can be derived from intensity ratios
be not directly related to the ionization structure. For in-
of lines emitted by a single ion from two levels with nearly
stance, Si III] and C III] transitions are characterized by a
the same energy, but different radiative-transition probabil-
lower ionization potential than [Ar IV] or N IV], but overall
ities or different collisional de-excitation rates (Osterbrock
they can probe higher densities than [Ar IV] and similar val-
1989). As a guide to the reader, in Fig. 3 we highlight the
ues to N IV]. We will further comment about this in Sec. 5.2
different combinations of diagnostics considered throughout
and Sec. 6.2. Also, [Fe III] has an ionization potential com-
this study and their characteristics, as introduced in Table 2
parable to [S II] or [O II], but it is probing electron densities
and Table 3. Specifically, in the eight main panels of Fig. 3,
between 103 and 106 cm−3 .
we report the measurements of each line ratio used as tracer
Finally, the shaded red regions in Fig. 3 show the pre-
of ne for the CLASSY galaxies and its corresponding density
dictions from the Cloudy 17.00 (Ferland et al. 2013) mod-
calculated with PyNeb. The dots are color-coded accord-
els from Berg et al. (2019b; 2021), that we used to estimate
ing to the ion species and are the same used in the top two
the ionization parameter, which we discuss in Sec. 4.4. The
panels, where the ionization potential and the traced density
Cloudy models we took into account are made using Bi-
range of each ion and line ratio, are reported. The differ-
nary Population and Spectral Synthesis (BPASSv2.14; El-
ent symbols indicate which line ratio we assumed to esti-
dridge & Stanway 2016; Stanway et al. 2016) burst models
mate the temperature (described in Fig. 4 and Sec. 4.3), as
for the input ionizing radiation field. The parameter space
reported in the legend. The black curves show the variation
covered is appropriate for our sample, including an age range
of each line ratio as a function of the temperature, consider-
of 1 − 10 Myr for young bursts and a range in ionization pa-
ing Te = 5 × 103 K, 1 × 104 K and 2.5 × 104 K.
 CLASSY IV 15

Ionization Potential [eV] n [cm 3]4

0 10 20 30 40 50 60 70 80 100 101 102 10e3 10 105 106 107

1.5 1.5 4.0 1.5

[Fe III] 4701/4659

Si III] 1882/1893
[O II] 3729/3727
[S II] 6716/6731

3.5
1.0 1.0 1.0
3.0
0.5 0.5 2.5 0.5 Te = 5000 K
Te([NII]) Te([NII])
Te([SII] 4069/6717,31) Te([SII] 4069/6717,31) Te = 1 × 104 K
Te([OII] 3727,9/7320,30) Te([OII] 3727,9/7320,30) Te([SIII]) Te = 2.5 × 104 K
0.0 0 0.0 0 2.0 0.0 0
10 102 104 106 10 102 104 106 100 102 104 106 10 102 104 106
1.5 1.5 1.5 1.5

[Ar IV] 4714/4741

[Cl III] 5518/5538

N IV] 1483/1487
C III] 1907/1909

1.0 1.0 1.0 1.0

0.5 0.5 0.5 0.5

Cloudy 17.00
Te([OIII] 4363/5007) Te([OIII] 4363/5007)
Te([SIII]) Te([SIII]) Te([OIII] 1666/5007) Te([OIII] 1666/5007)
0.0 0 0.0 0 0.0 0 0.0 0
10 102 104 106 10 102 104 106 10 102 104 106 10 102 104 106
ne [cm 3] ne [cm 3] ne [cm 3] ne [cm 3]

Figure 3. The eight square panels show the set of diagnostics to estimate the gas density ne in the different ionization zones (see also Tab. 2
and Tab. 3). The minor top left and right panels show the ionization potential and the range of traced densities of the diagnostics taken into
account, calculated with PyNeb, using the same color-coding. Specifically, in order of ionization potential, we have: [S II] λ6717/λ6731
(purple), [O II] λ3729/λ3727 (blue), [Fe III] λ4701/λ4659 (turquoise), [Si III] λ1883/Si III] λ1892 (dark green), [Cl III] λ5518/λ5538 (green),
[C III] λ1907/C III] λ1909 (gold), [Ar IV] λ4714/λ4741 (orange) and N IV] λ1483/λ1487 (red). The different symbols show the CLASSY
calculated values with PyNeb using the available temperature diagnostics described in Fig. 4. The solid, dashed and dotted black curves
represent PyNeb predictions at Te = 5 × 103 K, 1 × 104 K and 2.5 × 104 K, as reported in the legend, while the shaded red regions (very narrow
due to the low dependence of density on temperature) show the predictions from our Cloudy models of Berg et al. (2019b; 2021).

rameter of −4.0 < log(U) < 0, matching stellar and nebu- for more details). Finally, we note that our Cloudy models
lar metallicities (Z? = Zneb = 0.001, 0.002, 0.004, 0.008, cor- do not include [Fe III] lines.
responding to 0.05, 0.10, 0.20, 0.40 Z ). In particular, Berg
et al. (2019b) used the GASS10 solar abundance ratios (in- 4.3. Temperature
cluding dust) to initialize the relative gas-phase abundances,
The temperature Te can be determined via the intensity ra-
then scaling them to match the observed values for nearby
tios of particular emission line doublets, emitted by a single
metal-poor dwarf galaxies. Specifically, we calculated the
ion from two levels with considerably different excitation en-
median value of Cloudy predictions in the range of densities
ergies (Osterbrock 1989). To guide the reader, we show the
and temperatures taken into account, and the boundaries of
available temperature diagnostics used within this study in
the shaded red regions represent the ±3σ of the distribution.
the six panels of Fig. 4, as introduced in Table 2 and Table 3.
These regions appear narrow because of the very low depen-
In Fig. 4, we also report the measurements of each line ratio
dence of density on temperature (see Berg et al. 2018; 2019b;
for the CLASSY galaxies and its corresponding temperature,
2021 for all the details). Overall, from Fig. 3 we find good
using symbols and colors consistent with Fig. 3, to indicate
agreement between Cloudy and PyNeb, despite minor dif-
which density diagnostic we used while calculating the tem-
ferences in the default atomic data used by each code. The
perature using the getCrossTemDen task, as reported in the
main difference that we underline is that Cloudy models for
legend.
[Ar IV] are shifted towards higher densities. This discrep-
Overall, the most used optical Te tracers are given
ancy implies that Cloudy [Ar IV] densities could be . 1 dex
by [N II] λ5755/λ6584, [S III] λ6312/λ9069 and
higher than those measured with PyNeb. This discrepancy
[O III] λ4363/λ5007 for the low-, intermediate- and high-
could be due to the different atomic data used by Cloudy (see
ionization zones, respectively. Indeed, [N II] emission
Juan de Dios & Rodríguez 2017; 2021 and references within
is stronger in the outer parts of H II regions, where the
 16 M INGOZZI ET AL .
Low-ionization Low-ionization Low-ionization
ne([SII]); Te([SII] 4069/6717,31) ne([SII]); Te([OII] 3727,9/7320,30)
101 ne([OII]); Te([SII] 4069/6717,31) ne([OII]); Te([OII] 3727,9/7320,30)
10 1
103

[O II] 3727,9/7320,30
[S II] 4069/6717,31
[N II] 5755/6584

100 102
10 2
10 1 101

ne([SII]); Te([NII]) 100

ne([OII]); Te([NII])
10 3 10 2
Intermediate-ionization High-ionization High-ionization
ne([ArIV]); Te([OIII] 4363/5007)
10 1 ne(NIV]); Te([OIII] 4363/5007) 10 1

[O III] 4363/5007

[O III] 1666/5007
[S III] 6312/9069

10 1

ne = 100 cm 3
10 2 10 2
ne = 1000 cm 3
ne = 104 cm 3
ne([FeIII]); Te([OIII] 4363/5007) ne = 105 cm 3
10 2 ne([ClIII]); Te([OIII] 1666/5007)
ne(SiIII]); Te([SIII])
Cloudy 17.00
ne([ArIV]); Te([OIII] 1666/5007)
ne(CIII]); Te([SIII]) ne(NIV]); Te([OIII] 1666/5007)
10 3 10 3
5000 10000 15000 20000 25000 5000 10000 15000 20000 25000 5000 10000 15000 20000 25000
Te [K] Te [K] Te [K]

Figure 4. Set of diagnostics to estimate the gas temperature in the CLASSY galaxies in the different ionization zones: [N II] λ5755/[N II] λ6584,
[S II] λλ4069/[S II] λλ6717,31 and [O II] λλ3727,29/[O II] λλ7320,30 for the low-ionization zone, [S III] λ6312/[S III] λ9069,
[O III] λ4363/[O III] λ5007 and O III] λ1666/[O III] λ5007. The different filled symbols show the CLASSY calculated values with PyNeb, as
reported in the legend. The black curves represent PyNeb predictions at densities ne = 100 cm−3 , 5000 cm−3 , 104 cm−3 and 105 cm−3 , while the
shaded red regions show the predictions from out Cloudy models.

ionization is lower and O mostly exists as O+ (Oster- For high-redshift targets the O III] λ1666/λ5007 has been
brock 1989). With respect to the low ionization zone in explored as temperature diagnostics, since this ratio can be
particular, other available temperature indicators are the helpful for those cases where the optical auroral line is not
[S II] λλ4069/λλ6717,31 and [O II] λλ3727,29/λλ7320,30 available (weak, undetected lines, or outside of the observed
line ratios. Their main drawback is the wide separation in wavelength range; Villar-Martín et al. 2004; James et al.
wavelength, that introduces larger relative uncertainties via 2014; Steidel et al. 2014; Berg et al. 2016; Vanzella et al.
the dust attenuation correction. Also, [S II] λ4069 is usually 2016; Kojima et al. 2017; Pérez-Montero & Amorín 2017;
very faint, while [O II] λλ7320,30 lines can be affected by Patrício et al. 2018; Sanders et al. 2020).
telluric absorption depending on the redshift of the galaxy. Comparing the fifth and sixth panels of Fig. 4, we notice
Unfortunately, [N II] λ5755/λ6584 and [S III] λ6312/λ9069 that in general the O III] λ1666/λ5007 dependence on tem-
lines are not available for all the targets, given the very faint perature is steeper than [O III] λ4363/λ5007, suggesting that
nature of the [N II] λ5755 and [S III] λ6312 auroral lines, in principle O III] λ1666/λ5007 could represent a better Te
and the fact that [S III] λ9069 can fall out of the observed diagnostic (see also Kojima et al. 2017; Nicholls et al. 2020).
wavelength range, depending on the redshift of the source. However, in practice it is worth noting that several issues
In these cases, to estimate the temperature of the low- and arise in deriving ratios from optical and UV emission lines.
intermediate-ionization regions, in our iterative procedure Firstly, there can be flux matching issues and mismatched
we used the Garnett (1992) relations that link Te ([N II]) and aperture effects, if observations are taken with different in-
Te ([S III]) to Te ([O III]), : struments, as we discuss in detail in App. A. A second draw-
back to take into account is the large uncertainties resulting
Te ([N II])/[K] = 0.70 × Te ([O III]) + 3000 (1) from reddening estimates derived over such a large wave-
length window. Finally, there can be an intrinsic effects due
to the density, temperature and ionization structure of star-
Te ([S III])/[K] = 0.83 × Te ([O III]) + 1700 (2) forming regions. Indeed, if the ISM is patchy, the UV light is
These derived values are not reported in Fig. 4 for the sake of visible only through the less dense and/or less reddened re-
clarity, but do show good agreement with the PyNeb curves. gions along the line of sight, while the optical may be arising
 CLASSY IV 17

also from denser and/or more reddened regions. To further Sec. 4.3, where we showed their agreement both with our
discuss this, in Sec. 5.3 and Sec. 6.3, we show the compar- observed line ratios and PyNeb predictions. To calculate the
ison of temperatures derived with O III] λ1666/λ5007 and corresponding log(U) of each CLASSY galaxy, we assumed
[O III] λ4363/λ5007, and the resulting difference in deriving the gas-phase metallicity reported in Table 1.
12+log(O/H). Given that [S III]λλ9069, 9532 is outside the ob-
Finally, Fig. 4 highlights a very low dependence of the served wavelength range for many CLASSY galaxies, and
intermediate- and high-ionization temperature diagnostics on [Ar IV]λλ9069, 9532 and [Ar III]λ7138 lines can be faint and
ne . Concerning the low-ionization temperature diagnostics, thus below the required signal-to-noise threshold of 3, we can
the dependence on density is higher, but the comparison with measure the S3S2 and Ar4Ar3 line ratios only in 20 and 28
the observed line ratios used as diagnostics indicate that only CLASSY galaxies, respectively. On the other hand, we can
ne < 104 cm−3 are feasible. This is in-line with the fact that calculate the O3O2 line ratios for all the galaxies of our sam-
these line ratios are tracing the external and more diffuse re- ple, apart from J1444+4237. For the 18 CLASSY galaxies
gions of nebulae. with [O II]λλ3727,9 outside the observed wavelength range,
we estimated O3O2 using the emissivities of [O II]λλ3727,9
4.4. Ionization level and [O II]λλ7320,30 obtained with PyNeb, where calcu-
lations were performed using ne and Te associated to the
Another important parameter of the ISM is the ionization
low-ionization emitting zone. It is thus possible to recover
parameter, log(U), defined as the ratio of the number of ion-
[O II]λλ3727,9 by multiplying the calculated empirical ratio
izing photons to the density of hydrogen atoms. Empiri-
with the observed [O II]λλ7320,30 line fluxes (see also Pa-
cally, this property is best determined by ratios of emission
per V). In Sec. 5.4 we describe our results, while in Sec. 6.4
lines of the same element with a different ionization stage,
we discuss how the log(U) optical tracers relate to potential
such as O3O2 = [O III] λ5007/[O II] λλ3727,29 and S3S2 =
UV analogs.
[S III] λλ9069,9532/[S II] λλ6717,31 line ratios (e.g., Kew-
ley et al. 2019). Usually, O3O2 is the most widely used
proxy in the optical range because these oxygen lines lie in 5. RESULTS: COMPARING UV AND OPTICAL
a wavelength range accessible to many different instruments PHYSICAL PROPERTIES OF THE ISM
and are among the strongest line in the optical range. Also, 5.1. Dust attenuation diagnostics
they nicely span the entire energy range of an H II region
We calculated E(B − V ) for the CLASSY galaxies as de-
(Berg et al. 2021). However, O3O2 is strongly dependent
scribed in Sec. 4.1, with the iterative method explained in
on metallicity (Kewley & Dopita 2002; Kewley et al. 2019).
Sec. 4, finding values in the range E(B − V ) ∼ 0 − 0.5 mag.
S3S2 is less commonly used because the near-infrared (NIR)
To explore the systematic uncertainties on E(B − V ) due to
[S III] λλ9069,9532 lines are weaker than their oxygen coun-
the effect of density and temperature variations on the Balmer
terparts and lie at wavelengths that are less frequently cov-
decrement, we obtained a value for each combination of Te
ered together with [S II] lines. Moreover, the NIR wavelength
and ne estimates for each ionization zone (described in Fig. 3
range suffers more from telluric absorption and sky line con-
and Fig. 4). The top panels of Fig. 5 display the E(B − V )
tamination. Nevertheless, given the redder wavelengths of
values obtained for the low- (pink squares), intermediate-
the sulphur emission lines, and consequently their lower ex-
(red diamonds), high-ionization region (blue circles) densi-
citation energies with respect to oxygen, S3S2 is less affected
ties and temperatures as a function of 12+log(O/H) and stel-
by metallicity and also insensitive to ISM pressure (Dopita
lar mass, on the left and right, respectively. We also show
& Evans 1986; Kewley & Dopita 2002; Kewley et al. 2019;
E(B − V )0 , which we define assuming ne = 102 cm−3 and
Mingozzi et al. 2020). The lower excitation energies of [S II]
Te = 1 × 104 K (empty black pentagons), usually considered
and [S III] also imply that this ratio is tracing the ionization
appropriate conditions for star-forming regions (Osterbrock
parameter of the low-ionization regions of nebulae (see e.g.,
1989; Osterbrock & Ferland 2006). The dashed lines colored
Fig. 4 in Berg et al. 2021). Finally, Berg et al. (2021) intro-
accordingly indicate the median values. The Pearson corre-
duced the Ar4Ar3 = [Ar IV] λλ4714,71/[Ar III] λ7138 ratio
lation factors are R ∼ 0.50 for 12+log(O/H) and R ∼ 0.63 for
as a ionization parameter tracer of the very high-ionization
stellar mass, with pvalue of p ∼ 1 × 10−3 and p ∼ 5 × 10−6 ,
region.
respectively.
In our analysis, in order to calculate log(U), we used the
The lower panels show the difference between E(B − V )0
calibration of Berg et al. (2018) for O3O2 (see their table 3)
and the values calculated assuming the coherent density and
and the one of Berg et al. (2021) for S3S2 and Ar4Ar3 (see
temperature of the different ionization zones, expressed in
their table 4). These calibrations relate these line ratios and
[mag]:
log(U) as a function of the gas metallicity, and are obtained
using the set of Cloudy models described in Sec. 4.2 and ∆(E(B − V)) = E(B − V)0 − E(B − V)
18 M INGOZZI ET AL .
0.6 0.6
E(B V)0 (ne = 100 cm 3; Te = 104 K) E(B V)0 (ne = 100 cm 3; Te = 104 K)
Median E(B V)low Median E(B V)low
0.5 Median E(B V)int 0.5 Median E(B V)int
Median E(B V)high Median E(B V)high
0.4 0.4

E(B V) [mag]
E(B V) [mag]

0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
log(E(B V)0)-log(E(B V)low) Median E(B V)low
0.04 log(E(B V)0)-log(E(B V)int) 0.04 Median E(B V)int
log(E(B V)0)-log(E(B V)high) Median E(B V)high

(E(B V)) [mag]

(E(B V)) [mag]

0.02 0.02

0.00 0.00

0.02 0.02

0.04 0.04
7.00 7.25 7.50 7.75 8.00 8.25 8.50 8.75 6 7 8 9 10
12+log(O/H) log(M /M )

Figure 5. Top panels: comparison of the E(B − V ) estimated assuming Te = 1 × 104 K and ne = 102 cm−3 (E(B − V )0 ; empty black pentagon)
with the value obtained from low- (purple squares), intermediate- (green diamonds) and high-ionization regions (red circles), as a function of
12+log(O/H) and M?. The uncertainties of the displayed quantities are shown in both axis. Bottom panels: difference between E(B − V )0 and
the E(B − V ) measurements of each ionization zone, keeping the same symbols and colors as in the top panel. The colored dashed and dotted
lines indicate the median values of the shown distributions. The shaded regions (partially overlapped) in the bottom panel represent the 68%
intrinsic scatter of each distribution, color-coded accordingly. At 12+log(O/H) . 7.75 or stellar mass log(M?/M ) . 6.5, E(B −V ) is down to
∼ −0.04 mag lower than E(B −V )0 .

where E(B − V )0 is the value derived with ne = 102 cm−3 the emission lines without introducing a significant bias in
and Te = 1 × 104 K). The median values and the 68% intrin- the results. Hence, for our analysis we used the weighted av-
sic scatter of the distributions are shown by the dotted lines erage of these values (defined generically as E(B − V ) here-
and shaded regions (mostly overlapped), color-coded corre- after), to correct both UV and optical emission lines as ex-
spondingly. plained in Sec. 4.1.
Even though the difference of the colored dots with re-
spect to E(B −V )0 looks small, with an overall median value 5.2. Density diagnostics
around ∼ −0.01 mag and similar intrinsic scatter, there are
In the left and right upper panels of Fig. 6 we show the opti-
a few objects with discrepancies down to ∼ −0.04 mag, at
cal and UV density diagnostics as a function of 12+log(O/H).
12+log(O/H) . 7.75 or stellar masses log(M?/M ) . 6.5.
In both panels we also display the low-ionization density ob-
Specifically, E(B − V ) is generally larger than E(B − V )0 (of
tained through [S II] λ6717/λ6731 (purple squares) as a ref-
∼ −0.01 mag on average, or ∼ −0.05 dex), while at higher
erence. For the galaxy J0934+5514, we estimate the low-
12+log(O/H) their difference tends to 0. Therefore, assum-
ionization density from [O II] λ3729/λ3729 (blue square),
ing Te = 1 × 104 K and ne = 102 cm−3 can lead to underesti-
since KCWI data do not cover [S II] λλ6717,31 lines. For
mates in the dust attenuation, especially at low 12+log(O/H),
the optical, we show the intermediate-ionization density de-
where Te is higher (see Sec. 5.3). This is due to the slight de-
rived from [Cl III] λ5518/λ5538 (green left-pointing trian-
pendence of recombination lines on temperature and density,
gles) and [Fe III] λ4701/λ4659 (turquoise left-pointing and
that still causes the Balmer decrement to drop from ∼ 2.86 to
blue right-pointing triangles), and the high-ionization den-
∼ 2.70 (Osterbrock 1989; Osterbrock & Ferland 2006). The
sity from [Ar IV] λ4714/λ4741 (orange pentagons). For
fact that the low, intermediate and high-ionization E(B − V )
the UV, instead, we show the intermediate-ionization den-
values are consistent within their uncertainties indicates that
sity derived from [Si III] λ1883/Si III] λ1892 (darkgreen
a simpler approach, considering a single value of density and
up-pointing triangles) and [C III] λ1907/C III] λ1909 (gold
temperature that represents all zones, can be used to correct
down-pointing triangles), and the high-ionization density
 CLASSY IV 19
Optical ne diagnostics UV ne diagnostics
ne([SII]) ne(SiIII]; Te([SIII])
ne([OII]) ne(CIII]; Te([SIII])
5 ne([ClIII]; Te([SIII]) ne(NIV]; Te([OIII])
ne([FeIII]; Te([SIII])
ne([ArIV]; Te([OIII])
4
log(ne/cm 3)

1
1
(log(ne)[SII]) [dex]

0
1
2
3
4
7.00 7.25 7.50 7.75 8.00 8.25 8.50 8.75 7.00 7.25 7.50 7.75 8.00 8.25 8.50 8.75
12+log(O/H) 12+log(O/H)

Figure 6. Top panels: comparison of ne estimated from low- (ne ([S II]) or ne ([O II]) for J0934+5514; squares), intermediate- (ne ([Cl III]),
ne ([Fe III]), ne (Si III]), ne (C III]); triangles) and high-ionization regions (ne ([Ar IV], ne (N IV]); pentagons, hexagons), as a function of
12+log(O/H). For clarity reasons the optical and UV diagnostics are shown on the left and right, respectively, while the optical ne ([S II])
or ne ([O II]) are shown in both panels. The dots are color-coded consistently with Fig. 3, as indicated in the legend. The uncertainties of the
displayed quantities are shown in both axis. Bottom panels: difference in dex between the low-ionization density (purple squares) and the other
ne measurements, keeping the same symbols as in the top panel. The colored dashed and dotted lines indicate the median values. The shaded
regions (partially overlapped) in the bottom panel represent the 68% intrinsic scatter of each distribution, color-coded accordingly. Excluding
ne ([Fe III]) for which we have mainly upper limits, clearly the highest discrepancies with respect to ne ([S II]) are found for UV tracers, that
predict on average densities ∼ 1 − 2 dex higher.

from N IV] λ1483/λ1487 (red hexagons) values. Overall, the expected because [Ar IV] densities have a higher critical den-
values obtained span in ne ∼ 30 cm−3 to ne ∼ 105 cm−3 . The sity than [S II], which instead traces the low-ionization and
dashed lines represent the median value of density given by diffuse gas within nebulae (see Fig. 3). Interestingly, we find
each diagnostic, and are color-coded accordingly. that in the optical, the [Ar IV] densities that trace the high-
In the bottom panels of Fig. 6, we show the difference in ionization regions are somewhat lower than expected when
dex between the low-ionization zone density and the other compared to their UV counterparts N IV], and are instead
values (∆(log(ne )[SII] )), keeping the same color-coding of the consistent with [Cl III] values, which trace the intermediate-
main panel. The dotted lines show the median values of the ionization regions. Also, we note that when comparing the
offsets in dex, that are of ∼ −0.8, ∼ −2.3 and ∼ −0.9 dex, UV and optical diagnostics, both the [Ar IV] and [Cl III] den-
for [Cl III], [Fe III] and [Ar IV], and ∼ −1.3, ∼ −2.0, and sities are lower than those derived from C III] and Si III] in
∼ −2.1 dex for Si III], C III] and N IV], respectively, sorting the UV, which are both tracers of intermediate-ionization re-
the optical and UV diagnostics as a function of the increas- gions. On the other hand, [Fe III], which is characterized
ing ionization potential. The 68% intrinsic scatter of each by an ionization potential lower than [Cl III], traces similar
distribution is shown by the shaded regions, color-coded cor- densities to the UV diagnostics, but the values can only be
respondingly. The partial overlaps indicate a similar behavior considered upper limits due to the large error bars. Exclud-
of the distributions. ing [Fe III], clearly the highest discrepancies with respect to
We note that [Ar IV] densities are slightly larger than [S II] ne ([S II]) are found for UV tracers, that predict on average
densities with values on average around ∼ 1000 cm−3 , while densities around ∼ 104 cm−3 .
[S II] densities are always lower than ∼ 1000 cm−3 . This is
20 M INGOZZI ET AL .

We will further comment about this in Sec. 6.2. As ex- high-ionized regions, as observed. However, galaxies char-
pected, we find no correlation between log(ne ) (and also acterised by a very high excitation such as those covered by
∆(log(ne )[SII] )) and 12+log(O/H), as well as the other galaxy the CLASSY survey are expected to have minimal contri-
properties, such as stellar mass, stellar metallicity, stellar age, butions from the low-ionization lines, and thus little depen-
or SFR. dence on the low-ionization zone temperatures for the oxy-
gen abundance (Berg et al. 2021). In this respect, we feel
5.3. Temperature diagnostics confident that for those galaxies in our sample for which we
used the Garnett (1992) relation (i.e., 23 using Eq. 1 and 13
The left and right upper panels of Fig. 7 show the com-
using Eq. 2), the derived Te values are reliable estimates.
parison between the low- and intermediate-ionization re-
Concerning the high-ionization temperature, the main
gion temperatures inferred through the Garnett (1992) rela-
panel of Fig. 8 compares the values obtained iteratively with
tions (Eq. 1 and Eq. 2, respectively) as a function of the
either [O III] λ4363/λ5007 or the hybrid UV-optical ratio
high-ionization zone temperature Te ([O III] 4363/5007), to
O III] λ1666/λ5007, and the [S II], [Ar IV] and N IV] density
see to what extent they give consistent results. Specifi-
diagnostics, as reported in the legend. We note differences
cally, in the left panel of Fig. 7 we report the three different
lower than . 750 K using either [S II], [Ar IV] or N IV], con-
low-ionization Te estimates made with [N II] λ5755/λ6584
firming again the low dependence of temperature on density
(purple squares), [S II] λ4069/λλ6717,31 (pink dots)
diagnostics, even when the difference in log(ne ) can be as
and [O II] λλ3727,29/λλ7320,30 (blue diamonds), using
large as ∼ 2 dex. In general, the values are roughly in agree-
the [S II] λλ6717,31 doublet as density tracer (and the
ment with the dashed black line that indicates the 1:1 relation.
[O II] λλ3727,29 doublet for J0934+5514). On the right
To better evaluate this, the bottom panel shows the difference
panels there are no estimates of T([S III]([Cl III],[S III])), be-
between Te ([O III] λ4363/λ5007) and Te (O III] λ1666/λ5007)
cause we could not find finite values with PyNeb, and only
in dex (∆log(Te )opt−hybrid ), keeping the same symbols and
one galaxy for which we measured T([S III](Si III],[S III])).
colors of the main panel. ∆log(Te )opt−hybrid is in median
In general, the low-ionization temperatures obtained are
∼ −0.025 dex (∼ 1000 K), with the highest temperatures
in the range Te ∼ 8000 − 24000 K. The bottom panels
measured with [O III]λ1666/λ5007. This trend is consistent
of Fig. 7 report the differences in dex between the val-
with other works in the literature who made the same com-
ues obtained with the relations from Garnett (1992) and
parison in smaller samples (e.g., Berg et al. 2016). We com-
the Te inferred through the different temperature diag-
ment more on the offset ∆log(Te )opt−hybrid and its impact on
nostics. We note that the median values (dotted lines)
the estimate of 12+log(O/H) in Sec. 6.3.
in these subpanels are close to zero (the median values
are ∼ −0.1 dex), but systematically below, with a bet-
ter agreement with Eq. 1 for Te ([S II] λ4069/λλ6717,31) 5.4. Ionization parameter diagnostics
and Te ([O II] λλ3727,29/λλ7320,30) with respect to The upper panel of Fig. 9 shows the comparison of log(U)
Te ([N II] λ5755/λ6584). Also the 68% intrinsic scatter of estimated from low- (purple squares), intermediate- (green
the distributions, shown by the shaded regions color-coded diamonds) and high-ionization regions (orange circles) as a
correspondingly, show a similar behaviour (i.e., mostly over- function of 12+log(O/H), using the S3S2, O3O2 and Ar4Ar3
lapped). However, we note that overall, Eq. 1 tends to under- line ratios, respectively, as explained in Sec. 4.4. Fig. 9
estimate the temperature up to ∼ 0.2 dex. demonstrates that in our sample there is not an evident anti-
Fig. 7 shows that the intermediate-ionization Te derived correlation between log(U)(O3O2) and metallicity (r ∼ −0.3,
with [S III] λ6312/λ9069 are better in agreement with Gar- p ∼ 0.05), which is instead predicted by the theoretical rela-
nett (1992)’s relation with respect to the low-ionization Te tion presented in Dopita & Evans (1986) and Dopita et al.
discussed in the previous paragraph. To summarise, several (2006). This would be expected as a consequence of the en-
authors found significant differences with Garnett (1992)’s hanced opacity of stellar winds and higher level of scattering
relations using large samples of star-forming galaxies (e.g. at increasing metallicity. The former factor would lead to a
Kennicutt et al. 2003; Binette et al. 2012; Berg et al. 2015). decrease of the ionizing photons in the surrounding H II re-
A possible interpretation of the larger discrepancy that we gion, while the latter to a greater conversion efficiency from
find for Eq. 1 than for Eq. 2 could be explained by the luminous energy flux to mechanical energy flux, hence de-
large absorption cross section of low energy ionizing pho- creasing log(U). On the other hand, we notice a slight anti-
tons (Osterbrock 1989), which are thus preferentially ab- correlation with log(M?) (r ∼ −0.5, p ∼ 1 × 10−4 ), as sug-
sorbed in the H II regions with respect to higher energy ones, gested in some previous works on local star-forming galax-
leading to a hardened spectrum (e.g., Hoopes & Walterbos ies (e.g., Dopita et al. 2006; Brinchmann et al. 2008). We
2003). An implication would be that the low-ionized regions also find no correlation with SFR, which instead has been re-
of the H II regions could have higher temperatures than the vealed from some works based on spatially resolved optical
 CLASSY IV 21
Low-ionization region Te Intermediate-ionization region Te
24000 G92-N2 24000
ne([SII]; Te([NII] 5755/6584)
22000 ne([SII]; Te([SII] 4069/6717,31)
ne([SII]; Te([OII] 3727,9/7320,30)
22000
Te([NII];[SII];[OII]) [K]

20000 20000

Te([SIII]) [K]
18000 18000
16000 16000
14000 14000
12000 12000
G92-S3
ne([ClIII] 5518/5538; Te([SIII] 6312/9069)
10000 10000 ne(SiIII] 1882/1893; Te([SIII] 6312/9069)
ne(CIII] 1909/1907; Te([SIII] 6312/9069)
8000 8000
(log(Te)) [dex]

(log(Te)) [dex]
0.2 0.2

0.0 0.0

0.2 0.2
10000 12500 15000 17500 20000 22500 10000 12500 15000 17500 20000 22500
Te([OIII]([SII],[OIII])) [K] Te([OIII]([SII],[OIII])) [K]

Figure 7. Upper panels: Comparison of the three Te estimates of the low-ionization (on the left) and intermediate-ionization (on the right)
with the temperature diagnostics [N II] λ5755/λ6584, [S II] λλ4069/λλ6717,31, [O II] λλ3727,29/λλ7320,30 and [S III] λ6312/λ9071, as a
function of the high-ionization temperature T([O III]([S II],[O III])), taking into account the available density tracers. The uncertainties of the
displayed quantities are shown in both axis. The dot-dashed lines in the left and right panels indicate Eq. 1 and Eq. 2, respectively. On the right
panels there are no estimates of T([S III]([Cl III],[S III])), because we could not find finite values with PyNeb, and only one measurement for
T([S III](Si III],[S III])). Lower panels: difference in dex between the prediction of Eq. 1 and Eq. 2, and the calculated values. The shaded regions
(overlapped) in the bottom panel represent the 68% intrinsic scatter of each distribution, color-coded accordingly. Overall, the median values of
the computed differences are ∼ −0.1 dex and ∼ −0.05 dex (dotted lines in the minor panels) for the low-ionization and intermediate-ionization
Te , respectively.

spectra (e.g., Dopita et al. 2014; Kaplan et al. 2016). Finally, ture in the nebular environments of our targets, as described
in agreement with Kewley et al. (2015); Bian et al. (2016); in Sec. 4 and in Berg et al. (2021) for the CLASSY galaxies
Kaasinen et al. (2018); Mingozzi et al. (2020), we find J1044+0353 and J1418+2102.
a strong correlation between log(U)(O3O2) and log(sSFR) Among the available UV emission lines, only the
(r ∼ 0.7, p ∼ 1 × 10−8 ). We comment further about this in C IV/C III] line ratio involves emission originating from dif-
Sec. 6.4. We stress that similar results hold for log(U)(S3S2) ferent ionization states of the same element, and thus could
and log(U)(Ar4Ar3). constrain the ionization parameter. We will further discuss
Interestingly, from Fig. 9 we note that using these three this in Sec. 6.4.
different diagnostics we obtain a scatter of values of log(U),
6. THE UV TOOLKIT
in the range −3.5 − 0., with the lowest values derived from
S3S2 and the highest ones with Ar4Ar3. The lower panel of In this section we compare the optical and UV diagnos-
Fig. 9 illustrates better this scatter, showing the differences tic results detailed in Section 5 and discuss correlations be-
∆(logU)O3O2−S3S2 and ∆(logU)O3O2−Ar4Ar3 in purple and or- tween gas-phase properties and UV emission lines. Using
ange, respectively (we refer to this quantity as ∆(logU)O3O2 these relationships, we provide a set of diagnostic equations
hereafter), revealing significant discrepancies in the range that can be used to estimate gas-phase E(B-V), electron den-
±1 dex. The median values and the 68% intrinsic scatter of sity, electron temperature, metallicity and ionization param-
the distributions are shown by the dotted lines and shaded eter. Each of these equations relies only on emission lines
regions, color-coded correspondingly. Specifically, in our observed at rest-frame UV wavelengths and thus represents a
sample log(U(O3O2)) is in median ∼ 0.4 dex higher than UV Toolkit for deriving the chemical and physical conditions
log(U(S3S2)) and ∼ −0.6 dex lower than log(U(Ar4Ar3)). of star-forming galaxies.
This is in line with the presence of a clear ionization struc-
6.1. UV Diagnostics for E(B-V)
 22 M INGOZZI ET AL .

22000 High/Very-High ionization region Te 0.0

[SIII] 9069/[SII] 6717,31
ne([SII] 6717/6731) [OIII] 5007/[OII] 3727,29
ne([ArIV] 4714/4741) 0.5 [ArIV] 4714,41/[ArIII] 7138,7753
20000 ne(NIV 1483/1487)
1.0
Te([OIII] 1666/5007) [K]

18000
1.5
16000

log(U)
2.0
14000
2.5
12000
3.0
10000
3.5
2
[dex]

0.05

(log(U))O3O2 [dex]
1
hybrid

0.00
0
(log(Te))opt

0.05
1
0.10
8000 10000 12000 14000 16000 18000 20000 22000 2
Te([OIII] 4363/5007) [K] 7.00 7.25 7.50 7.75 8.00 8.25 8.50 8.75
12+log(O/H)
Figure 8. Top panels: comparison of the high-ionization re-
Figure 9. Top panel: comparison of the log(U) estimated from
gions Te (O III] λ1666/[O III] λ5007) estimated from the the hy-
low- (S2S3; purple squares), intermediate- (O3O2; green diamonds)
brid UV-optical ratio, using ne ([Ar IV]) (blue squares on the
and high-ionization regions (Ar4Ar3; orange circles), as a function
left) or ne (N IV]) (green crosses on the right), as a function of
of 12+log(O/H). The uncertainties of the displayed quantities are
Te ([O III] λ4363/[O III] λ5007). The uncertainties of the dis-
shown in both axis. Bottom panel: differences ∆(logU)O3O2−S3S2 , in
played quantities are shown in both axis. In general, the values
purple, and ∆(logU)O3O2−Ar4Ar3 , in orange (we refer to this quantity
are roughly in agreement with the dashed black line, which repre-
as ∆(logU)O3O2 ), revealing significant discrepancies in the range
sents the 1:1 relation. Bottom panels: difference in dex between
±1 dex. The colored dashed (in the main panel) and dotted (in the
Te ([O III] λ4363/[O III] λ5007) and Te (O III] λ1666/[O III] λ5007),
bottom panel) lines indicate the median values. The shaded regions
keeping same symbols and colors as in the top panel. The colored
in the bottom panel represent the 68% intrinsic scatter of each distri-
dashed and dotted horizontal lines in the main and minor panels
bution, color-coded accordingly. This scatter of log(U) values is in
respectively show the median values of the displayed quantities.
line with the presence of a clear ionization structure in the nebular
environments of our targets.
Here, we aim to explore the behaviour of the gas E(B −V )
(obtained with the Balmer decrement and the Cardelli et al. ∼ 0.43). We notice that majority of the CLASSY galaxies
1989 attenuation law, as described in Sec. 4.1), and the stel- lies along the 1:1 relation or are within that and the C97 line.
lar E(B − V )UV (obtained from the β slope of stellar con- The most significant correlation that we observe is between
tinuum fitting and the Reddy et al. 2016 attenuation law, the gas E(B − V ) and the difference (in mag) ∆(E(B − V )) =
as described in Sec. 3.1.1). The upper panels of Fig. 10 E(B − V ) − E(B − V )UV , characterized by a Pearson factor of
show the comparison between the stellar E(B − V )UV and ∼ +0.78, as shown in the bottom panel of Fig. 10. The linear
the gas E(B − V ), color-coded as a function of the SFR best-fit of the correlation between ∆(E(B −V )) and E(B −V )
and specific SFR (sSFR), calculated considering the stel- is indicated by the solid red lines and reported on the upper
lar mass and SFR within the COS aperture (see Table 6 right of the bottom panels, and translates into:
in Paper I). The black dashed line shows the 1:1 relation,
while the black dotted line shows the empirical relation be- E(B −V )UV = (0.33 ± 0.08) × E(B −V ) + (0.03 ± 0.01)
tween stellar and gas E(B − V ) according to Calzetti (1997) (3)
E(B −V ) = (3.00 ± 0.70) × E(B −V )UV − (0.09 ± 0.04)
(C97 hereafter): E(B − V )UV = (0.44 ± 0.03) × E(B − V ) (or
E(B −V ) = (2.27 ± 0.15) × E(B −V )UV ). with scatter of 0.05, while the dashed and dotted lines repre-
From Fig. 10, we note that there is just a mild correla- sent the 1σ and 2.6σ around the best-fit. To infer the best-fit
tion between E(B − V )UV and E(B − V ) (Pearson factor of (here and in the following sections) we used the LtsFit pack-
 CLASSY IV 23
0.5 1.43 0.5 7.00
Calzetti+1997b Calzetti+1997b

0.93 7.16
0.4 0.4
0.43 7.31
E(B V)UV [mag]

E(B V)UV [mag]

0.3 0.3
0.06 7.47

Aper. log(SFR/[M yr 1])

0.2 0.2

Aper. log(sSFR/yr 1)
0.56 7.62

0.1 1.06
0.1 7.78

0.0 E(B V) [mag] 1.56 0.0 E(B V) [mag] 7.93

0.3 0.3
(E(B V)) [mag]

(E(B V)) [mag]

0.2 2.05 0.2 8.09
0.1 0.1
0.0 2.55 0.0 8.24
0.1 best-fit: y = (0.69±0.08)x + (-0.03±0.01) 0.1 best-fit: y = (0.67±0.08)x + (-0.03±0.01)
= 0.05 = 0.05
0.2 3.05 0.2 8.40
0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
E(B V) [mag] E(B V) [mag]

Figure 10. Upper panels: comparison between the stellar E(B − V )UV with respect to the gas E(B − V ), color-coded as a function of log(SFR)
(on the left) and log(sSFR) (on the right) within the COS aperture. The dashed black line represents the 1:1 relation, while the dotted line is
the C97 empirical relation between stellar and gas E(B − V ) (i.e., E(B − V )UV = (0.44 ± 0.03) × E(B − V )). The uncertainties of the displayed
quantities are shown in both axis. Lower panels: ∆(E(B −V )) defined as E(B −V ) − E(B −V )UV . ∆(E(B −V )) increases linearly with E(B −V ),
and the linear best-fit is indicated by the red solid lines, reported on the upper right, while the dashed and dotted lines represent the 1σ and
2.6σ around the best-fit. The red horizontal dotted line represents the median ∆(E(B −V )) while the shaded red region shows the 68% intrinsic
scatter of the distribution. ∆(E(B −V )) increases at increasing log(SFR) and decreasing log(sSFR)
. Overall, E(B −V ) ∼ E(B −V )UV at log(SFR/[M yr−1 ]) . −1.5 and log(sSFR/yr−1 ) & −8, whereas at higher SFRs and lower sSFRs the stellar
vs. gas E(B-V) relation follows the C97 relation.
age developed by Cappellari et al. (2013) which allows us low ∼ 0.2 mag, while the gas E(B − V ) increases, reach-
to perform a robust linear fit taking into account the uncer- ing ∼ 0.5 mag. This confirms the presence of an excess of
tainties on both axis and the intrinsic scatter. This method dust attenuation in the gas with respect to the stars at high
also clips outliers, using the robust Least Trimmed Squares log(SFR) and low log(sSFR), in line with the empirical rela-
(LTS) technique by Rousseeuw & Van Driessen (2006). As tion found by C97. This relation implies that stars are on av-
explained in Section 3.2.2 of Cappellari et al. (2013), their erage a factor two less reddened than the ionized gas, which
algorithm adopts an initial guess and performs a first least- is related to the fact that the covering factor of the dust is
squares fit. Then it computes the standard deviation of the larger for the gas than for the stars (C97). This topic is quite
residuals, selecting all data point deviating no more than 2.6σ debated, with works suggesting a similar nebular and stellar
from the fitted relation, performing a final fit for the selected reddening (e.g., Erb et al. 2006; Reddy et al. 2010; Pannella
points. The errors on the coefficients are computed from the et al. 2015; Shivaei et al. 2015; Puglisi et al. 2016) and oth-
covariance matrix. ers finding a higher nebular reddening (e.g., Calzetti 1997;
Interestingly, Fig. 10 shows that E(B − V ) and E(B − V )UV Calzetti et al. 2000; Wild et al. 2011b; Kashino et al. 2013;
have similar values at low log(SFR) and high low log(sSFR) Price et al. 2014; Reddy et al. 2015; Qin et al. 2019; Theios
(and low gas-phase metallicity, since 12+log(O/H) corre- et al. 2019; Shivaei et al. 2020), as nicely summarised in Ta-
lates with E(B − V ); see Fig. 5), and are in fact close to ble 1 in Shivaei et al. (2020).
the 1:1 relation (dashed black line) until ∼ 0.2 mag. At As explained in C97, the nebular emission requires the
increasing log(SFR) and decreasing log(sSFR), the stellar presence of the ionizing stars, that remain relatively close to
E(B −V )UV remains slightly constant with values always be- their (dusty) place of birth during their short lifetime, while
24 M INGOZZI ET AL .

the UV stellar continuum is contributed also by non-ionizing 2020). However, this discrepancy has also been found in a
stars, that have the time to move to regions of lower dust handful of cases at z ∼ 1 − 3 where both the rest-frame opti-
density. Hence, ionizing and non-ionizing stars are not ex- cal and UV density diagnostics were available (Hainline et al.
pected to be co-spatial, implying that a correlation is not 2009; Quider et al. 2009; Christensen et al. 2012; Bayliss
expected between stellar continuum and nebular emission, et al. 2014; James et al. 2014; Maseda et al. 2017; James
and thus between their reddening, in contrast with the em- et al. 2018; Berg et al. 2018; Acharyya et al. 2019; Schmidt
pirical result found by C97. A possible interpretation to ex- et al. 2021), and in few local galaxies as well (e.g., Berg et al.
plain this differential attenuation is a two-component dust 2016). We also find a consistent offset between ne ([S II])
model, with diffuse dust attenuating light from all stars, and and the density values from [Si III] λ1883/Si III] λ1892 for
the birth cloud dust component only attenuating light origi- the three galaxies for which it is available (i.e., J1044+0353,
nating from the star-forming regions (Wild et al. 2011a). In J1253-0312 and J1448-0110) and an even larger offset for
line with this interpretation, Wild et al. (2011a) found a de- the two galaxies with N IV] densities (i.e., J1253-0312 and
pendence of this extra-attenuation and sSFR. As nicely il- J1545+0858).
lustrated and explained in Fig. 5 of Price et al. (2014), at A possible reason for this UV-optical ne discrepancy, as
high sSFR the continuum light is dominated by young, mas- explained in Sec. 4, is that the optical [S II] is tracing the
sive stars located in the birth clouds, so both the continuum low-ionization regions while the UV doublets originate in
and emission lines are attenuated by both dust components, intermediate-ionization and high-ionization regions, closer
and stellar and nebular E(B − V ) are similar; at decreasing to the ionizing source (Maseda et al. 2017; Berg et al. 2018;
sSFR, the discrepancy between stellar and nebular E(B − V ) James et al. 2018). Indeed, the ionization potential of these
increases, since less massive stars, generally residing out- UV diagnostics (25 − 75 eV) and the range of densities for
side the birth clouds, have a higher contribution to the con- which they are ne -sensitive (up to 105 cm−3 ) reach higher val-
tinuum emission, while the emission lines are still attenu- ues than [S II] doublet (E ∼ 10 − 20 eV and ne < 104 cm−3 ),
ated by both dust components. This is in line with what as shown in the upper panels of Fig. 3. In contrast to our find-
we find in Fig. 10. An important thing to underline is that ings, it should be noted that Patrício et al. (2016) measured
both Wild et al. (2011a) and Price et al. (2014) take into consistent densities using C III] and [O II] density diagnos-
account galaxies with log(sSFR/yr−1 ) in the range ∼ [-10,- tics for a z = 3.5 star-forming galaxy, finding ne ∼ 100 cm−3 ,
8.5], while CLASSY galaxies are systematically shifted at typical of local H II regions (Osterbrock 1989; Osterbrock &
higher log(sSFR/yr−1 ) in the range ∼ [-8.4,-7] (see Paper I). Ferland 2006). However, they did find higher values of ne
Indeed, CLASSY galaxies were selected to be compact and using the UV N IV] and Si III] diagnostics, suggesting that
UV bright, and thus have significantly enhanced sSFRs with there must be other ISM properties that are playing a role.
respect to their star-forming main-sequence counterparts at As introduced in Sec. 4, a more self-consistent comparison
z ∼ 0, and are more comparable to the z ∼ 2 galaxy popula- between UV and optical ne diagnostics would be achieved by
tion (Paper I). This could explain why we find no correlation comparing UV ne diagnostics with the [Cl III] and [Ar IV] op-
of the sSFR with E(B −V )UV , and just a mild anti-correlation tical line ratios, which similarly originate in the intermediate-
with the gas E(B − V ) (Pearson factor of ∼ −0.36), and with and high-ionization regions, respectively. As shown in Fig. 3,
their difference (Pearson factor of ∼ −0.40). Overall, we con- [Cl III] traces a slightly lower range of densities (with ne up
clude that E(B − V ) ∼ E(B − V )UV at log(SFR/[M yr−1 ]) . to ∼ 5 × 104 ), while [Ar IV] a similar range to C III]. If the
−1.5 and log(sSFR/yr−1 ) & −8, whereas at higher SFRs and density structure were following the ionization structure, go-
lower sSFRs the stellar vs. gas E(B-V) relation follows that ing from the diffuse to the densest gas, looking at the ion-
of C97. This relationship has the potential for us to derive ization potential of each density diagnostic tracer, the order
the gas E(B −V ) in galaxies for which rest-frame UV spectra would be: ne ([S II]) < ne ([O II]) < ne ([Fe III]) < ne (Si III]) <
are available but the optical wavelengths are not accessible, ne ([Cl III]) < ne (C III]) < ne ([Ar IV]) < ne (N IV]). This would
as it will happen for the reionization galaxies that JWST will be true assuming that nebulae are formed of uniform shells
reveal. or rings, with density decreasing inside-out according to 1/r2
(Stanghellini & Kaler 1989). However, this density structure
6.2. UV Diagnostics for Electron Density is not clearly confirmed by observations, with many works
showing contradicting results. For instance, Maseda et al.
As shown in Sec. 5.2, the electron density estimated from
(2017) found a consistent median density of log(ne ) ∼ 3.2
[C III] λ1907/C III] λ1909 is overall ∼ 2 dex higher than
derived from C III] for a sample of 17 z > 1.5 C III]-emitting
the one obtained from [S II] λ6717/[S II] λ6731, despite the
galaxies and densities obtained from [Ar IV] in a control-
large errors on C III] densities. Unfortunately, many previ-
sample of SDSS z ∼ 0 objects. On the contrary, as detailed
ous works have been limited in determining C III] densities
in Sec. 5.2, we found that in the optical, the [Ar IV] densi-
by the low resolution of the spectra (e.g., Ravindranath et al.
 CLASSY IV 25

ties are consistent with [Cl III] values, with a median value 3.5 ne(SiIII]); Te([SIII])
ne ∼ 103 cm−3 , whereas both are lower than the UV diagnos- ne(CIII]; Te([SIII])
ne(NIV); Te([OIII])
8
tics by ∼ 1 dex, despite the fact that [Ar IV] should trace a
higher ionization zone. UV densities are found to be con- 3.0
sistent with [Fe III] densities (and [Fe III] has a lower ion- 7

log(ne)([SII]) [cm 3]
ization potential than [Cl III]), but these are just upper limits
given the large error bars of the measured [Fe III] line ratios. 2.5

EW(CIII])
Indeed, the relation between the ionization and density struc- 6
ture is not trivial, since the internal structure of H II regions is
shaped both by winds and radiation pressure even when they 2.0
are not dynamically important (see e.g., Fig 2 Geen et al. 5
2020). Also, the apertures of our UV and optical observa-
tions are not allowing us to observe singular H II regions, and 1.5 4
because of the differential dust attenuation with wavelength,
UV and optical tracers may not be tracing the same regions
along the line of sight, with optical tracers coming also from 1.02 3 4 5 3
regions completely extincted in the UV.
A possible scenario that explains why ne ([Ar IV]) densities log(ne)(UV) [cm 3]
are not higher than ne (C III]) would be an ionization front
due to the central source that produces a rarefication at the Figure 11. The ne ([S II]) in logarithm as a function of ne (Si III]) (up-
inner boundary of the nebula, and thus high-ionization zone pointing triangles) and ne (N IV]) (hexagons), as well as ne (C III])
density tracers such as [Ar IV] would be produced in a cen- (down-pointing triangles), color-coded as function of EW(C III]).
tral hole (Giuliani 1981; Stanghellini & Kaler 1989). How- The uncertainties of the displayed quantities are shown in both axis.
The black dot-dashed line shows the 1:1 relation, while the dot-
ever, if this were the case, we should see a significant differ-
dashed black line represents the same relation shifted of 2 dex, that
ence in the kinematics of [Ar IV] lines with respect to lower- we added to highlight the level of the offset.
ionization region density tracers (which we do not, M22 in
prep.). Hence, we cannot find evidence of an hydrodynamic
of EW(C III]) with density would be expected according to
effect at the basis of the density structure from our data. In
Jaskot & Ravindranath (2016) models. Indeed, the higher
this context, very high spectral resolution optical data, es-
density nebula are characterized by a higher incident ioniz-
pecially integral-field spectroscopy data that allows to map
ing flux, because of the definition of the ionization parame-
the density structure within a galaxy, could help understand
ter (see Eq. 1 in Jaskot & Ravindranath 2016). Therefore, a
this offset. An alternative explanation for the ne ([Ar IV])-
higher density implies higher temperatures, as well as higher
ne (C III]) discrepancy could be related to the atomic parame-
collision rates, which, in combination with the abundance of
ters taken into account (Morisset et al. 2020; Juan de Dios &
C2+ ions, ultimately determine the C III] emission strength
Rodríguez 2021). Indeed, PyNeb and Cloudy adopt differ-
(Jaskot & Ravindranath 2016). However, the strength of
ent atomic data for the ions considered in this work, and in
C III] emission is also affected by the gas-phase metallicity,
Fig. 3, we showed that in general Cloudy predictions for den-
the ionization parameter or the stellar age, with increasing
sities were in agreement with PyNeb, apart from [Ar IV], for
equivalent widths predicted to reach values & 10 − 15 Å at
which Cloudy would predict densities up to ∼ 1 dex higher.
ages < 3 Myr and log(U) > −2 (Jaskot & Ravindranath 2016;
This corresponds to the discrepancy with the UV tracers that
Nakajima et al. 2018; Ravindranath et al. 2020). In CLASSY
we find in this work.
galaxies, we find an anti-correlation between log(EW(C III]))
To further investigate the relation between optical and UV
and 12+log(O/H) (Pearson: r = −0.43, p = 0.03), a slight cor-
densities, Fig. 11 shows ne ([S II]) as a function of ne (Si III])
relation with log(U) (Pearson: r = 0.3, p = 0.4), and an anti-
(up-pointing triangles), ne (C III]) (down-pointing triangles)
correlation with stellar age (Pearson: r = −0.4, p = 0.04), con-
and ne (N IV]) (hexagons), colour-coded as a function of the
firming that the C III] strength of the emission is determined
EW(C III]). The dashed black lines shows the 1:1 relation,
by many factors that are not yet fully understood (Ravin-
while the dot-dashed black line represents the 1:1 relation
dranath et al. 2020). We will comment more on the complex
shifted of 2 dex, that we added to highlight the level of
relation between C III], metallicity and ionization parameter
the offset. We tested several quantities, but a weak corre-
in Sec. 6.4.
lation was only observed between log(ne ([S II]λλ6717,31))
Overall, the CLASSY galaxies show that ne (C III]) is on
and EW(C III]) (Pearson: r ∼ 0.40, p ∼ 0.02), which can be
average ∼ 2 dex higher than ne ([S II]), and this offset can be
appreciated by the color-coding of Fig. 11. A slight increase
used to determine the low-ionization zone electron density in
26 M INGOZZI ET AL .

similar galaxies if only the intermediate-ionization zone elec- ne ([Ar IV]) and [O III] λ4363/λ5007 (Te,opt ; darkgreen di-
tron density is available from rest-frame UV spectra covering amonds) or ne ([S II]), ne ([Ar IV]) and O III] λ1666/λ5007
C III] lines. (Te,hybrid ; gray pentagons and blue squares), as a function of
the stellar mass. We also tested the combinations ne (N IV])
6.3. UV Diagnostics for Electron Temperature and [O III] λ4363/λ5007 (green hexagons), or ne (N IV]) and
O III] λ1666/λ5007 (turquoise crosses). For clarity reasons,
Here we have the important (and rare) opportunity to in-
we display the optical 12+log(O/H) on the left panels and
vestigate if the hybrid UV-optical ratio O III] λ1666/λ5007
the Hybrid/UV 12+log(O/H) on the right panels, showing as
(Te,hybrid ) is a reliable tracer of Te with respect to the
a reference the Paper I measurements. The lower panels of
most generally used optical ratio, [O III] λ4363/λ5007
Fig. 12 show the difference in dex between the various es-
(Te,opt ). In Sec. 5.3 we commented that the median differ-
timates and the values from Paper I. The dashed and dotted
ence in temperature between Te,opt and Te,hybrid is on aver-
horizontal lines in the main and minor panels show the me-
age ∆log(Te )opt−hybrid ∼ −1000 K, with higher values found
dian values of the plotted quantities, respectively, while the
for Te,hybrid . This difference could be due to several fac-
shaded regions in the bottom panels represent the 68% intrin-
tors. Indeed, it is not trivial to compare O III] λ1666 and
sic scatter of each distribution, color-coded accordingly.
[O III] λ5007, given that different instruments are used,
While there is an offset between 12+log(O/H) from Pa-
with different apertures and calibration accuracies (see also
per I and the different estimates of 12+log(O/H), it is mainly
App. A), as well as the large difference in wavelength, that
within ±0.3 dex. In general, we find that this offset (the me-
can lead to higher uncertainties in the dust attenuation correc-
dian value and its distribution, shown in the bottom panels of
tion. Moreover, as commented in Sec. 4.3, in a patchy ISM,
Fig. 12) is consistent between the value derived using Te,opt
the UV light could be visible only through the less dense
and the one using Te,hybrid , when measured with different den-
and/or less reddened regions along the line of sight, while
sity diagnostics. As commented in Sec. 4, if possible it is
the optical may trace also from denser and/or more reddened
important to take into account diagnostics of the same ion-
regions, leading to an intrisic difference between the line ra-
ization zone because the lack of co-spatiality in estimating
tios, and thus in the derived temperatures.
ISM properties could lead to incorrect estimates of physi-
Another reason to explain the discrepancy could be related
cal and chemical properties. Despite this fact, the method
to an effect of the gas pressure, as proposed by Nicholls et al.
based on the [S II] and the optical [O III] ratios used in Pa-
(2020). In their work, Nicholls et al. (2020) list this ef-
per I is the most common, even though [S II] emission is not
fect under the intrinsic caveats and limitations of the direct
co-spatial with [O III] emission (e.g., see Fig. 4 from Nicholls
method to derive temperatures. Indeed, they commented that
et al. 2020). Using for instance [Ar IV] or N IV] diagnostics
[O III] λ5007 is quenched more rapidly than [O III] λ4363 as
to measure the density would be more ideal, but in practice
the gas pressure increases, meaning that [O III] λ4363/λ5007
these alternative optical density diagnostic lines are faint, and
increases at increasing pressure, emulating a higher temper-
thus not available for all the CLASSY targets.
ature as it happens in Active Galactic Nuclei (AGN; e.g.,
Overall, we conclude that the hybrid UV/optical temper-
Nagao et al. 2001). This occurs because of [O III] λ4363
ature diagnostic, Te,hybrid , is not dominated by potential sys-
has a significantly higher critical density than [O III] λ5007
tematic uncertainties, and can be used as a reliable tracer of
(ncrit ∼ 3.3 × 107 cm−3 versus ncrit ∼ 7.0 × 105 cm−3 ; Os-
the electron temperature in the systems covered here. We
terbrock 1989; Osterbrock & Ferland 2006). They specify
also conclude that the hybrid UV/optical Te allows to esti-
that this effect can especially be a problem in unresolved
mate the gas-phase metallicity with a scatter of ±0.3 dex with
high luminosity H II regions. O III] λ1666 has an even
respect to the optical 12+log(O/H), again without showing
higher critical density (ncrit ∼ 4 × 1010 cm−3 ; Wei 1988) than
systematic discrepancies. This is particularly important as in
[O III] λ4363. Therefore, Nicholls et al. (2020) conclude that
our sample we find that the O III] λ1666 flux is on average
higher values of Te (O III] λ1666/[O III] λ5007) compared to
∼0.3 dex brighter than the [O III] λ4363, and thus it could be
Te ([O III] λ4363/λ5007) may be due to pressure bias, under
more easily observed in high-z galaxies than the weaker and
suitable conditions of pressure and temperature.
less detectable [O III] λ4363 emission.
All the factors discussed in the previous paragraphs could
While we have shown that O III] λ1666 can allow us
contribute to the median offset of ∼ −1000 K between Te,opt
to reliably and directly estimate the metallicity of the
and Te,hybrid that we observe. However, the fundamental as-
ionized gas, the Te,hybrid method still depends on us hav-
pect to understand is to what extent this offset impacts the
ing access to the optical [O III]λλ4959,5007 emission.
estimate of 12+log(O/H). We explore this in the top panels
To the best of our knowledge the only work in the lit-
of Fig. 12, where we compare the 12+log(O/H) values that
erature that proposed promising 12+log(O/H) diagnos-
Paper I calculated using ne ([S II]) and [O III] λ4363/λ5007
tics using exclusively UV emission lines is Byler et al.
(black pentagons), with the values we calculate using either
 CLASSY IV 27

10.0 Optical 12+log(O/H) Hybrid Optical-UV 12+log(O/H)

Berg+22 (ne([SII]); Te([OIII] 4363/5007)) ne([SII]); Te([OIII] 1666/5007)
9.5 ne([ArIV]; Te([OIII] 4363/5007) ne([ArIV]; Te([OIII] 1666/5007)
ne(NIV; Te([OIII] 1666/5007)
9.0 ne(NIV; Te([OIII] 4363/5007)
12 + log(O/H)

8.5
8.0
7.5
7.0
6.5
0.25
[dex]

0.00
0.25

6.5 7.0 7.5 8.0 8.5 9.0 9.5 10.0 6.5 7.0 7.5 8.0 8.5 9.0 9.5 10.0
log(M ) [M ] log(M ) [M ]

Figure 12. Top panels: comparison of 12+log(O/H) from Paper I, with the estimates obtained using the combination of with either ne ([Ar IV]) or
ne (N IV]), and either Te ([O III] λ4363/λ5007) or Te ([O III] λ1666/λ5007), as indicated in the legend, as a function of the stellar mass. For clarity
reasons, the optical and hybrid optical-UV measurements of 12+log(O/H) are shown on the left and right, respectively, while 12+log(O/H) from
Paper I is shown in both panels. The uncertainties of the displayed quantities are shown in both axis. Bottom panels: difference in dex between
Paper I 12+log(O/H) and the other estimates, keeping the same symbols and colors as in the top panel.

(2020) (B20 hereafter). Using Cloudy modeling, these au- In order to investigate this offset further, we tested whether
thors derived calibrations from He IIλ1640/C III]λλ1907,9 the optical and UV-diagnostic 12+log(O/H) offset correlates
versus O III]λ1666/C III]λλ1907,9 (He2-O3C3 here- with any optical or UV quantities that we measured here, or
after) and Si III]λλ1883,92/C III]λλ1907,9 versus with CLASSY galaxy properties in general. We only found
O III]λ1666/C III]λλ1907,9 (Si3-O3C3 hereafter). The typ- correlations with the optical 12+log(O/H), as it can be seen in
ical statistical errors for these relations are ±0.08 dex and Fig. 13. The dot-dashed blue and magenta best-fit relations
±0.14 dex, respectively (Byler et al. 2020). In Fig. 13, in Fig. 13 allow us to essentially re-calibrate the B20 equa-
we compare the optical-based 12+log(O/H) values derived tions and to obtain the optical 12+log(O/H) from UV-only
in Paper I with the values we derive here from the He2- quantities, which we show in the following expressions:
O3C3 and Si3-O3C3 calibrations, shown as blue stars and
magenta pluses, respectively. A clear disagreement can 12 + log(O/H) = (0.39 ± 0.06) × (12 + log(O/H))He2
(4)
be seen between the 12+log(O/H) values derived from the −(4.68 ± 0.47)
UV diagnostics and optical direct method. The compar-
ison with the 1:1 relation shows the difference between 12 + log(O/H) = (0.42 ± 0.20) × (12 + log(O/H))Si3
optical and UV 12+log(O/H). The Si3-O3C3 values are (5)
systematically higher with an offset up to ∼ 0.6 dex with −(4.81 ± 1.59)
respect to the optical 12+log(O/H) from Paper I, which de- with a scatter of 0.08 and 0.03 dex, respectively, while the
creases at increasing 12+log(O/H). However, we do not dashed and dotted lines represent the 1σ and 2.6σ around the
have enough Si III]λλ1883,92 measurements to robustly best-fits.
evaluate the level of this trend. The He2-O3C3 estimates
instead show a larger scatter of ∼ ±0.5 dex, over-predicting 6.4. UV diagnostics for metallicity and ionization
the metallicity at 12+log(O/H) < 8, and under-predicting parameter
it at 12+log(O/H) > 8. This is in agreement with what There have been many attempts to infer the gas-phase
found by Rigby et al. (2021), who compared UV and op- metallicity using a combination of photoionization models
tical 12+log(O/H) diagnostics for a single gravitationally and UV emission lines (e.g., Fosbury et al. 2003; Jaskot &
lensed source, finding that the UV estimates from B20 are Ravindranath 2016; Gutkin et al. 2016; Feltre et al. 2016;
∼ 0.5 − 0.8 dex lower than the optical estimates, found in the Chevallard & Charlot 2016; Pérez-Montero & Amorín 2017;
range 12+log(O/H) ∼ 8.20 − 8.60. Byler et al. 2018; Nakajima et al. 2018) and/or measuring
 28 M INGOZZI ET AL .

ies (e.g., Ravindranath et al. 2020), it is particularly useful,

8.6
if possible, to exploit them to find reliable calibrations to
8.4 infer metallicity and ionization parameter for galaxies with
properties comparable to those of the CLASSY sample.
8.2 In Fig. 14 we show how these quantities vary as a func-
tion of 12+log(O/H) (derived with the direct method, Pa-
12+log(O/H) (UV)

8.0 per I), color-coded as a function of log(U) derived from

O3O2 (top panel), and vice-versa (bottom panel). In par-
7.8 ticular, we show only the measurements for which the S/N of
the lines used to derive the quantities is higher than the cho-
7.6 sen threshold (S/N > 3). In each panel we also provide the
Pearson correlation factor and relative pvalue. Among these
7.4 best-fit: y = (0.39±0.06)x + (4.68±0.47) quantities, we find that log(C III]/O III]), log(EW(O III]))
= 0.08 and log(EW(C III])) show the most promising relation with
7.2 best-fit: y = (0.42±0.2)x + (4.81±1.59)
= 0.03 12+log(O/H), with the highest Pearson factor found for
Byler20 He2-O3C3 calibration C III]/O III] (Pearson correlation factor ∼ 0.68 and p-value ∼
7.0 Byler20 Si3-O3C3 calibration 0.001). O III] λ1666/He II λ1640 is clearly increasing as
7.0 7.2 7.4 7.6 7.8 8.0 8.2 8.4 8.6 metallicity decreases but with a turn-over at 12+log(O/H) ∼
12+log(O/H) (Optical) 7.7. We also notice that the presence of nebular C IV in
pure emission can be used as well as a metallicity indica-
Figure 13. Comparison of optical and UV metallicities, compar- tor, since it comes uniquely from 12+log(O/H). 8 (see also
ing Paper I 12+log(O/H) with the exclusively UV He2-O3C3 and Senchyna et al. 2019; Berg et al. 2021; Senchyna et al. 2022b;
Si3-O3C3 calibrations proposed by B20, shown as blue stars and Schaerer et al. 2022). A slight anti-correlation between
magenta pluses. The uncertainties of the displayed quantities are log(C IV/C III]), log(EW(C IV)) and log(EW(He II)) with
shown in both axis. The dashed black line indicates the 1:1 rela-
metallicity (Pearson: r ∼ −0.40, r ∼ −0.35 and r ∼ −0.20,
tion. The dot-dashed blue and magenta lines show the best-fits of
the 12+log(O/H)UV as a function of the optical 12+log(O/H), as re-
respectively) could be due instead to residuals in the subtrac-
ported in the legend. The blue and magenta dashed and dotted lines tion of stellar features, which would be most prominent at
represent the 1σ and 2.6σ around each best-fit. increasing metallicities.
In the left panel of Fig. 15, we highlight the C III]/O III] as
metallicity directly with UV O III] auroral line (e.g., Erb et al. a function of the 12+log(O/H), color-coded with respect to
2010; Berg et al. 2018; James et al. 2018). However, few log(U). C III]/O III] clearly increases at increasing metallic-
12+log(O/H) calibrations have been proposed using exclu- ities, following the C/O abundance ratio (e.g., Garnett 1990;
sively UV emission lines (e.g., C IVλλ1548,51, He II λ1640, Izotov & Thuan 1999; Garnett et al. 2004; Erb et al. 2010;
O III] λ1666 and C III]λλ1907,9; see Byler et al. 2020; Guseva et al. 2011). Indeed, C III]/O III] is considered a good
Pérez-Montero & Amorín 2017), because UV lines are also tracer of C/O, because of the similar excitation and ionization
strongly sensitive to other properties, such as the ionization potentials of C+2 and O+2 (i.e., 6 − 8 eV and 24.8 − 35.1 eV,
parameter and shape of the ionizing continuum (Gutkin et al. respectively), and its minimal uncertainty due to reddening,
2016; Feltre et al. 2016; Nakajima et al. 2018; Maiolino & given that the interstellar attenuation curve is nearly flat over
Mannucci 2019). The parameter space coverage of CLASSY the wavelength 1600−2000 Å (Garnett et al. 1995; Berg et al.
and the high frequency of UV line detections throughout 2016).
the sample allow us to explore how UV lines vary with the The metallicity dependence of C/O was also shown in
gas-phase metallicity 12+log(O/H) and ionization parameter, Jaskot & Ravindranath (2016), who attribute it either to
log(U), which we infer from reliable optical diagnostics. the weaker winds of low-metallicity massive stars or to the
In order to fully explore the utility of UV diag- longer timescales to enrich the ISM for lower mass stars (see
nostic lines as possible tracers for these ISM prop- also, Maeder 1992; Henry et al. 2000; Chiappini et al. 2003).
erties, in this work we investigated a suite of de- However, this relation is also characterized by a scatter that
rived quantities including: O III] λ1666/He II λ1640; largely increases at low metallicities (12+log(O/H) < 8; see
C III] λλ1907,9/He II λ1640; C III] λλ1907,9/O III] λ1666; Fig. 6a of Berg et al. 2016, Fig. 3 of Amorín et al. 2017
C IV λλ1548,51/C III] λλ1907,9; EW(He II λ1640); or Fig. 11 of Berg et al. 2019b) because at low metallic-
EW(O III] λ1666); EW(C III] λλ1907,9); EW(C IV λλ1548,51). ity, C/O can be sensitive to different physical conditions, in-
Since after Lyα, C III] λλ1907,9 and C IV λλ1548,51 are cluding star formation histories, star formation efficiency, in-
among the brighest UV emission lines in star-forming galax- flow rates, and variations in the IMF (Skillman 1998; Berg
 CLASSY IV 29

1.50
r =-0.41 , p =0.09 r =0.68 , p =0.0011 r =0.25 , p =0.32 r =-0.4 , p =0.29
8.5 1.75
2.00
12 + log(O/H)

8.0

log(U)
2.25
7.5 2.50
2.75
7.0
0.0 0.5 0.00 0.25 0.50 0.75 0.0 0.5 1.0 1.0 0.5 0.0 3.00
log(OIII] 1666/HeII 1640) log(CIII] 1907,9/OIII] 1666) log(CIII] 1907,9/HeII 1640) log(CIV 1448,1501/CIII] 1907,9)
1.50
r =-0.2 , p =0.42 r =-0.52 , p =0.02 r =-0.43 , p =0.03 r =-0.35 , p =0.4
8.5 1.75
2.00
12 + log(O/H)

8.0

log(U)
2.25
7.5 2.50
2.75
7.0
0.5 0.0 0.5 0.5 0.0 0.5 1.0 0.0 0.5 1.0 1.5 0.5 0.0 0.5 1.0 3.00
log(EW(HeII 1640)) log(EW(OIII] 1666)) log(EW(CIII] 1907,9)) log(EW(CIV 1448,1501))
0.5 r =0.43 , p =0.4 r =-0.72 , p =0.3964 r =-0.39 , p =0.4 r =0.49 , p =0.4
1.0 7.8
1.5

12+log(O/H)
7.6
log(U)

2.0
7.4
2.5
7.2
3.0
3.5 7.0
0.0 0.5 0.00 0.25 0.50 0.75 0.0 0.5 1.0 1.0 0.5 0.0
log(OIII] 1666/HeII 1640) log(CIII] 1907,9/OIII] 1666) log(CIII] 1907,9/HeII 1640) log(CIV 1448,1501/CIII] 1907,9)
0.5 r =0.51 , p =0.4 r =0.65 , p =0.4 r =0.28 , p =0.4 r =0.78 , p =0.4
1.0 7.8
1.5

12+log(O/H)
7.6
log(U)

2.0
7.4
2.5
7.2
3.0
3.5 7.0
0.5 0.0 0.5 0.5 0.0 0.5 1.0 0.0 0.5 1.0 1.5 0.5 0.0 0.5 1.0
log(EW(HeII 1640)) log(EW(OIII] 1666)) log(EW(CIII] 1907,9)) log(EW(CIV 1448,1501))

Figure 14. Top panel: 12+log(O/H) as a function of O III] λ1666/He II λ1640, C III]/O III] λ1666, C III]/He II λ1640 and C IV/C III] line
ratios (upper panels), and EW(He II λ1640), EW(O III] λ1666), EW(C III]) and EW(C IV) (lower panels), color-coded as a function of the
ionization parameter log(U) derived from O3O2. The uncertainties of the displayed quantities are shown in both axis. All the quantities are in
logarithm, and the Pearson factor and pvalue are shown in each panel. Overall the most promising relations are between C III]/[O III] λ1666,
EW(O III] λ1666) < 4 Å and EW(C III]) with 12+log(O/H), but there is scatter in the plots, partly due to a degeneracy with the ionization
parameter, as it can be appreciated from the color-coding. Bottom panel: Same for log(U), color-coding as a function of 12+log(O/H) from
Paper I.
30 M INGOZZI ET AL .

et al. 2016; Amorín et al. 2017; Berg et al. 2019b). In ever, at these low metallicities, the carbon abundance de-
agreement with Stark et al. (2014), and the predictions of creases as well, suppressing C III] emission (e.g., Rigby et al.
Jaskot & Ravindranath (2016), C III]/O III] > 1 for all the 2015). Moreover, as also discussed in Sec. 6.2, the strength
CLASSY galaxies (the lowest value is C III]/O III] ∼ 1.70 of C III] emission is not only affected by metallicity, but
for J1323-0132). Jaskot & Ravindranath (2016) predict that also by the ionization parameter, stellar age and gas density,
C III] should only become comparable to O III] in photoion- with increasing equivalent widths predicted to reach values
ization models with metallicity of Z < 0.004, only if both & 10 − 15 Å at ages < 3 Myr and log(U) > −2 (Jaskot &
C/O ratios are low (C/O ≤ 0.12) and the ionization parame- Ravindranath 2016; Nakajima et al. 2018; Ravindranath et al.
ter is high (log(U) ≥ −2), which is consistent with the values 2020). This explains why J0934+5514 (i.e., I Zw 18), charac-
of 12+log(O/H) and log(U) we measure in this work. terized by the lowest metallicity among the CLASSY targets
The best-fit relation between C III]/O III] and the direct- (12+log(O/H) ∼ 6.98), has EW(C III]) ∼ 1.10 Å despite a rel-
method metallicity 12+log(O/H), shown as the red solid line ative high ionization parameter (log(U) ∼ −2.20), as already
in the left-panel of Fig. 15, is represented by: acknowledged by Rigby et al. (2015).
Overall, the best-fit relation between log(EW(C III]λλ1907,9))
12 + log(O/H) = (0.80 ± 0.26) × log(CIII]1907, 9/OIII]1666)
and the direct-method metallicity, valid at 12+log(O/H) >
+ (7.34 ± 0.18) 7.5, is shown in the right panel of Fig. 15 by the red solid
(6)
line and is represented by:
with an intrinsic scatter of 0.18 dex. The dashed and dotted
black lines indicate the 1σ and 2.6σ scatter from the relation, 12 + log(O/H) = (−0.51 ± 0.13) × log(EW (CIII]1907, 9))
respectively. The scatter in this relation, which is expected as + (8.43 ± 0.11)
we commented in the previous paragraph, can also be partly (7)
explained by a secondary dependence to log(U), given the with an intrinsic scatter of 0.18 dex. The dashed and dotted
difference of the ionization potentials of the C and O lines black lines indicate the 1σ and 2.6σ scatter from the relation,
(Jaskot & Ravindranath 2016). We will investigate in detail respectively.
the behavior of C III]/O III] with the C/O ratio and its scatter Returning to Fig. 14, log(C III]/O III]), log(C IV/C III]),
with respect to O/H in CLASSY in a forthcoming paper, fo- log(EW(O III])) and log(EW(C IV)) show the most promis-
cused on detailed photoionization modeling of the CLASSY ing relations with log(U), with the highest Pearson factors
galaxies. for EW(C IV) (r ∼ 0.78). Overall, many of the shown quan-
The right panel of Fig. 15 highlights the anti-correlation tities in Fig. 14 show a dependence on both 12+log(O/H)
between EW(C III]) and 12+log(O/H). Here we also include and log(U), as typically happens for UV tracers (e.g.,
values found for local star-forming galaxies studied in re- Maiolino & Mannucci 2019; Ravindranath et al. 2020), with
cent works (Rigby et al. 2015; Berg et al. 2016; Senchyna C III]/O III] increasing with increasing 12+log(O/H) and de-
et al. 2017; 2019; Berg et al. 2019a; Schaerer et al. 2022), in creasing log(U), while EW(O III]) and EW(C III]) show the
which 12+log(O/H) was computed with the direct method, opposite behavior. We further investigate the most promis-
as in this work. As found by Rigby et al. (2015); Naka- ing correlations with ionization parameter in Fig. 16, where
jima et al. (2018); Ravindranath et al. (2020) low-metallicity we separate log(U) estimates for the low- (purple squares;
galaxies tend to have a stronger EW(C III]), showing a lin- from S3S2), intermediate- (green diamonds; from O3O2)
ear anti-correlation down to EW(C III]) > 5 Å for galax- and high-ionization regions (orange circles; from Ar4Ar3)
ies with 12+log(O/H) . 8.25. A similar transition at (as discussed in Sec. 5.4). The first three rows with three
EW(C III]) > 5 Å at 12+log(O/H) . 8.4 is found and dis- panels each show the relations of log(C IV/C III]) (C4C3),
cussed in Senchyna et al. (2017; 2019). However, as shown log(EW(C IV)) and log(C III]/O III]), with the S3S2, O3O2
in Fig. 15, for galaxies with EW(C III]) < 5 Å there is clearly and Ar4Ar3 log(U), respectively. The solid red lines show
a large scatter towards lower metallicities, due to the fact that the best-fit to each of the UV-based log(U) measurements
galaxies with 12+log(O/H) . 7.5 can show EW(C III]) < 5 Å from CLASSY, while the dashed and dotted lines represent
as well. As such, the EW(C III])-12+log(O/H) relation is the 1σ and 2.6σ around the best-fit. The equations for the
double-branched with a turn-over point at 12+log(O/H) ∼ best-fit correlations and their intrinsic scatter are reported in
7.5. the legend of each panel, alongside the corresponding Pear-
Indeed, as discussed in Jaskot & Ravindranath (2016), son factors. In general, the UV emission line properties are
metallicity sets the shape of the ionizing spectral energy more tightly correlated to O3O2 and Ar4Ar3 log(U), and
distribution (SED) and defines the nebular gas temperature, only slightly or mostly not related to S3S2 log(U). This could
which enhances C III] emission at low metallicity due to the be due to the fact that the three log(U) diagnostics are tracing
larger amount of ionizing photons and higher collisional ex- different ionizing regions (see Sec. 4.4), and the considered
citation rates (e.g., Erb et al. 2010; Stark et al. 2014). How- UV lines are coming from intermediate or high-ionization
 CLASSY IV 31

8.50 r = 0.68, p = 0.02 1.6

8.25 1.8

8.00 2.0
12 + log(O/H)

7.75 2.2

log(U)
7.50 r = -0.48, p = 0.02 2.4
best-fit: y = (-0.51±0.13)x + (8.43±0.11)
7.25 Rigby+15
= 0.18
2.6
Senchyna+17
7.00 best-fit: y = (0.8±0.26)x + (7.34±0.18)
Senchyna+19
Berg+16 2.8
= 0.18 Berg+19
6.75 CLASSY sample Schaerer+22
3.0
0.0 0.2 0.4 0.6 0.8 1 0 1
log(CIII] 1907,9/OIII] 1666) log(EW(CIII] 1907,9))

Figure 15. Gas phase metallicity, 12+log(O/H), as a function of C III]/O III] λ1666 (left-panel) and EW(C III]) (right-panel), color-coded as a
function of log(U). The uncertainties of the displayed quantities are shown in both axis. Our best-fit linear relations are shown by the solid red
lines, while the dashed and dotted black lines represent the 1σ and 2.6σ around the best-fit. The scatter in both panels can be partly explained
by a secondary dependence on the ionization parameter. In the right panel we also include values found for star-forming galaxies studied in
recent works (Rigby et al. 2015; Berg et al. 2016; Senchyna et al. 2017; 2019; Berg et al. 2019a; Schaerer et al. 2022), in which 12+log(O/H)
was computed with the direct method, comparably to this work. This figure shows the most promising UV tracers of metallicity.

regions (Berg et al. 2021). Overall, the most promising UV- Using log(C III]/O III]):
based diagnostics for log(U) are the following.
Using log(C IV/C III]): log(U(S3S2)) = (−1.25 ± 0.24) × log(CIII]1907, 9/OIII]1666)
− (2.47 ± 0.05)
log(U(O3O2)) = (0.41 ± 0.26) × log(C4C3)
− (1.97 ± 0.16) (8) (12)

log(U(O3O2)) = (−2.74 ± 0.44) × log(CIII]1907, 9/OIII]1666)

− (0.22 ± 0.29)
log(U(Ar4Ar3)) = (0.36 ± 0.2) × log(C4C3)
− (1.33 ± 0.11) (9) (13)

log(U(Ar4Ar3)) = (−2.83 ± 0.5) × log(CIII]1907, 9/OIII]1666)

Using log(EW(C IV)):
+ (0.09 ± 0.41)

log(U(O3O2)) = (0.31 ± 0.18) × log(EW (CIV 1548, 51)) (14)

In practice, Eq. 8-14 in combination with Eq. 6 and 7
− (2.2 ± 0.1)
for 12+log(O/H) (see Fig. 15) could guide photoioniza-
tion models to explore the relationship between log(U) and
(10) 12+log(O/H), with the hope of breaking its degeneracy.

6.4.1. A Note regarding C IV

log(U(Ar4Ar3)) = (1.36 ± 0.19) × log(EW (CIV 1548, 51))
As a ratio of two UV emission lines from consecutive ion-
− (1.74 ± 0.0.12) ization states, C4C3, could trace the ionization structure of
the nebula (Jaskot & Ravindranath 2016), similarly to S3S2,
(11) O3O2 and Ar4Ar3 in the optical. It is thought to be also a
32 M INGOZZI ET AL .
Low ionization zone Low/High ionization zone High/Very-High ionization zone
0.0 best-fit: y = (-0.01±0.01)x + (-2.72±0.01) best-fit: y = (0.41±0.26)x + (-1.97±0.16)
= 0.11 = 0.37
0.5 [SIII] 9069/[SII] 6717,31
r = 0.25, p = 0.6
[OIII] 5007/[OII] 3727,29
r = 0.59, p = 0.124 r = 0.55, p = 0.25
1.0
1.5
log(U)

2.0
2.5
best-fit: y = (0.36±0.2)x + (-1.33±0.11)
3.0 = 0.19
[ArIV] 4714,41/[ArIII] 7138
1.5 1.0 0.5 0.0 1.5 1.0 0.5 0.0 1.5 1.0 0.5 0.0
log(CIV 1548,51/CIII] 1907,09) log(CIV 1548,51/CIII] 1907,09) log(CIV 1548,51/CIII] 1907,09)
Low ionization zone Low/High ionization zone High/Very-High ionization zone
0.0 best-fit: y = (-0.01±0.01)x + (-2.71±0.01) best-fit: y = (0.31±0.18)x + (-2.2±0.1) r = 0.96, p = 0.01
= 0.11 = 0.2
0.5 [SIII] 9069/[SII] 6717,31
r = 0.35, p = 0.4
[OIII] 5007/[OII] 3727,29
r = 0.7, p = 0.08
1.0
1.5
log(U)

2.0
2.5
best-fit: y = (1.36±0.19)x + (-1.74±0.12)
3.0 = 0.18
[ArIV] 4714,41/[ArIII] 7138
0.5 0.0 0.5 1.0 0.5 0.0 0.5 1.0 0.5 0.0 0.5 1.0
log(EW(CIV 1548,51)) log(EW(CIV 1548,51)) log(EW(CIV 1548,51))
Low ionization zone Low/High ionization zone High/Very-High ionization zone
0.0 best-fit: y = (-1.25±0.24)x + (-1.88±0.12) best-fit: y = (-3.0±0.39)x + (-0.22±0.29) r = -0.6, p = 0.01
= 0.16 = 0.34
0.5 [SIII] 9069/[SII] 6717,31 [OIII] 5007/[OII] 3727,29
r = -0.56, p = 0.1 r = -0.73, p = 0.00037
1.0
1.5
log(U)

2.0
2.5
best-fit: y = (-2.76±0.56)x + (0.09±0.41)
3.0 = 0.3
[ArIV] 4714,41/[ArIII] 7138
0.2 0.4 0.6 0.8 1.0 1.2 0.2 0.4 0.6 0.8 1.0 1.2 0.2 0.4 0.6 0.8 1.0 1.2
log(CIII] 1907,09/OIII] 1666) log(CIII] 1907,09/OIII] 1666) log(CIII] 1907,09/OIII] 1666)
0.0 Low ionization zone Low/High ionization zone High/Very-High ionization zone
Mingozzi+2020 best-fit: y = (0.52±0.12)x + (-3.96±0.33) best-fit: y = (1.16±0.38)x + (-5.09±1.08)
best-fit: y = (0.48±0.17)x + (-4.11±0.48) = 0.19 = 0.42
0.5 = 0.17 [OIII] 5007/[OII] 3727,29 [ArIV] 4714,41/[ArIII] 7138
[SIII] 9069/[SII] 6717,31
1.0
1.5
log(U)

2.0
2.5
3.0
3.5 r = 0.75, p = 0.0002 r = 0.7, p = 1e-06 r = 0.46, p = 0.025
2.0 2.5 3.0 3.5 2.0 2.5 3.0 3.5 2.0 2.5 3.0 3.5
log(EW(H )) log(EW(H )) log(EW(H ))

Figure 16. log(U) estimated for low- (purple squares; S3S2 line ratio), intermediate- (green diamonds; O3O2 line ratio) and high-
ionization regions (orange circles; Ar4Ar3 line ratio) as a function of C IV λλ1548,51/C III] λλ1907,09, log(EW(C IV λλ1548,51)),
C III] λλ1907,09/O III] λ1666 and log(EW(Hα)). The uncertainties of the displayed quantities are shown in both axis. The red solid lines
represent the best-fit that we calculated between the observables shown and the three log(U) values, while the dashed and dotted lines show
the 1σ and 2.6σ scatter around the relation. The best-fit equations as well as the intrinsic scatter are reported in the legend in each panel. The
dash-dot purple line in the bottom figure, left panel, is the correlation found by Mingozzi et al. (2020) between log(U) and log(EW(Hα)). This
figure shows the most promising UV tracers of log(U).
 CLASSY IV 33

potential indicator of Lyman continuum (LyC) escape (Jaskot lar continuum modeling, but this may not always be the case -
& Oey 2013; Nakajima & Ouchi 2014; Schaerer et al. 2022). especially for high-z observations of faint galaxies. Another
Several authors have indeed proposed the use of C4C3 as a caveat to take into account for C4C3 is the different nature
ionization parameter tracer (e.g., Pérez-Montero & Amorín of C III] and C IV emission, while both are collisional neb-
2017; Kewley et al. 2019; Schaerer et al. 2022). The first row ular lines, C IV is also a resonant line affected by radiation
of Fig. 16 confirms a correlation between C4C3 and O3O2 transfer effects. This means that C IV absorption along the
and ArAr3 log(U), with a large intrinsic scatter, reported in line-of-sight, which depends on the gas kinematics and col-
the legend. Also EW(C IV) correlates well with log(U) de- umn density, could contribute in reducing the observed C IV
rived from both O3O2 and Ar4Ar3, as shown in the second emission (e.g., Steidel et al. 2016). We will investigate in de-
row of Fig. 16. tail any differences in the kinematics of these lines and the
We stress that the fitting method that we are using performs causes that contribute to their emission in M22 (in prep.),
a clipping of the points out from 2.6σ of the relation (black which is dedicated to the ionization mechanisms behind the
dotted lines), which means that the best-fit equations that we UV emission lines presented here.
find are not able to take into account all the available galaxies.
This limitation is probably related to the low statistics, given
7. CONCLUSIONS
that it is rare to observe pure nebular C IV emission, while
the remaining galaxies showing pure absorption or P-Cygni In this work we investigated UV and optical diagnos-
profiles. Indeed, nebular C IV emission is rarely observed in tics of ISM properties by exploiting the CLASSY sur-
the literature, and comes uniquely from studies of systems vey presented by Berg et al. (2022); James et al. (2022),
with 12+log(O/H). 8, indicative of a rapid hardening of the which represents the first high-quality, high-resolution and
ionizing spectrum at low metallities (see also Senchyna et al. broad wavelength range (∼ 1200 − 2000 Å) FUV spectral
2019; Berg et al. 2021; Senchyna et al. 2022b; Schaerer et al. database of 45 nearby (0.002 < z < 0.182) star-forming
2022). This holds also for the galaxies of the CLASSY sam- galaxies. Specifically we focused on the main UV emis-
ple, where C IV nebular emission is coming only from objects sion lines visible in the COS spectra apart from Lyα
with 12+log(O/H). 8 and generally high ionization parame- (i.e., N IV] λλ1483,87, C IV λλ1548,51, He II λ1640,
ter log(U) & −2.5, as highlighted in Fig. 16. O III]λλ1661,6, Si III] λλ1883,92, C III] λλ1907,9), and
Senchyna et al. (2019) suggested also a correlation be- combine them with emission lines comprised between
tween EW(C IV) and EW(Hβ) (with galaxies characterized [O II]λλ3727,9 and [S III]λ9069 from optical spectroscopy
by C IV nebular emission having EW(Hβ) ≥ 100 Å). In turn, of the same pointing. Our aim was to provide a UV toolkit of
the equivalent width of the strongest hydrogen recombina- ISM diagnostics, i.e., a set of equations to diagnose E(B −V ),
tion lines measures the ratio of the young, ionising stars ne , Te , 12+log(O/H) and log(U), that use only UV emis-
over the old, non-ionising population (e.g., Leitherer 2005), sion lines. By carefully assessing the stratified ionization
and thus correlates with the sSFR and the degree to which structure of our targets, we accurately calculated the physical
the youngest stars dominate the optical (Kewley et al. 2015; and chemical properties using diagnostic line ratios specific
Kaasinen et al. 2018; Mingozzi et al. 2020). This partly ex- to each ionization zone (constituting a set of ‘direct’ diag-
plains the correlation between log(EW(C IV)) and log(U). nostics), then subsequently comparing and calibrating each
Indeed, Mingozzi et al. (2020) found a tight correlation be- property with the well-known optical diagnostics to derive
tween log(EW(Hα)) and log(U) (derived from S3S2), ex- a series of ‘indirect’ diagnostic UV-based equations. In the
plained in terms of the good correlation between log(U) and following we summarize our main findings.
the age of the stellar population (Pellegrini et al. 2020), traced
by EW(Hα) (a similar correlation holds for EW(Hβ); e.g.,
• UV density diagnostics ([C III] λ1907/C III] λ1909,
Senchyna & Stark 2019). A comparable relation holds also
[Si III] λ1883/Si III] λ1892, N IV] λ1483/N IV] λ1487)
for the CLASSY galaxies, as shown in the bottom panel
give log(ne ) ∼ 2 dex higher than the optical counter-
of Fig. 16, that displays the optical log(U) as a function of
parts [S II] λ6717/[S II] λ6731 and [O II] λ3729/λ3727
log(EW(Hα)).
and ∼ 1 dex than [Cl III] λ5518/λ5538 and
It should be noted, however, that diagnostics using nebu-
[Ar IV] λ4714/λ4741, as summarized in Fig. 6 and
lar C IV emission may be difficult to employ due to strong
discussed in Sec. 6.2. UV ne are consistent with
stellar C IV P-Cygni profiles, as highlighted also in Jaskot
ne ([Fe III] λ4701/λ4659), for which we can derive only
& Ravindranath (2016). As commented in Sec. 3, CLASSY
upper limits. The UV-to-optical electron density off-
data benefits from a combination of relatively high-S/N and
set enables us to derive a low-ionization zone ne (e.g.,
high-spectral resolution which allowed us to separate the stel-
ne ([S II])) for high-z targets in case optical wavelengths
lar and the nebular components of C IV emission via UV stel-
are not available.
34 M INGOZZI ET AL .

• The hybrid UV/optical temperature diagnostic and EW(C III] λλ1907,9) as 12+log(O/H) diag-
O III] λ1666/λ5007 is not dominated by major sys- nostics (see Eq. 6 and 7, displayed in Fig. 15).
tematic uncertainties, as shown in Fig. 8. As such, we We also highlight that the presence of nebular
find this auroral line ratio to be as reliable a tracer of C IV λλ1548,51 can be used as a metallicity indi-
electron temperature as the most generally used opti- cator, since it is revealed uniquely in galaxies with
cal ratio, [O III] λ4363/λ5007. We also find that the 12+loG(O/H) . 8, in agreement with the few previ-
O III] λ1666 flux is on average ∼0.3 dex brighter than ous works that studied this rare spectral feature. In
the [O III] λ4363, and thus it could be more easily Fig. 16, we include the most promising log(U) diag-
observed in high-z galaxies than the weaker and less nostics, providing expressions to derive the ioniza-
detectable [O III] λ4363 emission. tion parameter from C IV λλ1548,51/C III] λλ1907,9,
EW(C IV λλ1548,51) and C III] λλ1907,9/O III] λ1666
• On average, O III] λ1666/λ5007 gives temperatures (see Eq. 8-14). Finally, in Sec. 6.4.1, we discuss about
higher than [O III] λ4363/λ5007 by ∼ 1000 K. the limitations in using nebular C IV diagnostics.
Despite this offset, we discuss in Sec. 6.3 that
Te (O III] λ1666/λ5007) allows us to estimate direct- In summary, we present an empirically calibrated UV
method gas-phase metallicities with only a scatter toolkit of ISM diagnostics that will be fundamental in char-
of ±0.3 dex with respect to the optically-derived acterizing and interpreting the spectroscopic observations of
12+log(O/H), as shown in Fig. 12. We also in- high-z systems in the upcoming JWST and ELT era, for
vestigated how the optical 12+log(O/H) compare which the optical wavelength range will not be available.
with metallicities derived from UV-based He2-O3C3 Specifically, NIRISS and NIRSpec JWST instruments will
and Si3-O3C3 metallicity calibrations by B20 (see provide spectra in the wavelength ranges of 0.8 − 2.2 and
Fig. 13). We find an offset correlating with the op- 0.6 − 5.3 µm, respectively, covering (in full or in part) the
tical 12+log(O/H), and we provide expressions to re- suite of optical emission lines (from [O II]λλ3727 to Hα)
calibrate the B20 equations (i.e., Eq. 4 and 5). together with the C III]λλ1907, 9 at 5 < z < 6. At higher
z, optical lines are progressively shifted out of the observed
• We derive ionization parameters in the range
range, while at z ≥ 10 only UV lines from Lyα to C III] will
−3.5 < log(U) < 0. using the optical diagnos-
be accessible. The low-resolution (R ∼ 150) spectroscopy
tics S3S2 ([S III] λλ9069,9532/[S II] λλ6717,31),
of JWST/NIRISS is not enough to resolve the C III] dou-
O3O2 ([O III] λ5007/[O II] λλ3727,29) and
blet, but is expected to detect the UV emission lines down
Ar4Ar3 ([Ar IV] λλ4714,71/[Ar III] λ7138), as
to ∼ 10−18 erg/s/cm−2 (Treu et al. 2022; JWST User Docu-
shown in Fig. 9. Specifically, we find that
mentation 2016). On the other hand, the higher spectral res-
log(U(S3S2)) < log(U(O3O2)) < log(U(Ar4Ar43)),
olution of JWST/NIRSpec (up to R ∼ 2700) will provide a
in line with the presence of a clear ionization structure
resolved C III] doublet (JWST User Documentation 2016).
in the nebular environments of our targets, as described
As demonstrated here, the CLASSY survey has provided the
in Sec. 4 and in Berg et al. (2021).
ideal ATLAS to create a powerful UV toolkit to explore these
• In Fig. 10, we investigated how the gas attenuation, high-z spectra in terms of specific star-formation rate, direct
E(B −V ) (obtained with the Balmer decrement and the gas-phase metallicity, ionization level, reddening, and nebu-
Cardelli et al. 1989 attenuation law; see Sec. 4.1), com- lar density, since it represents a high-quality UV and optical
pares with the stellar attenuation, E(B−V )UV (obtained database for local analogs to reionization-era systems. To
from the β slope of stellar continuum fitting and the complete our toolkit, in the next paper on UV-based diagnos-
Reddy et al. 2016 attenuation law; see Sec. 3.1.1). Our tics, we will focus on the ionized gas kinematics, both of UV
conclusion is that the gas and stellar attenuation factors and optical emission, and on the diagnostics to inspect their
are approximately the same at log(sSFR) & −8 yr−1 , main ionization mechanisms, to also understand the origin of
whereas at lower sSFRs the values tend towards the the main UV emission lines taken into account.
C97 relation, E(B −V ) ∼ 2.27 × E(B −V )UV . This rela-
tion subsequently allows us to use E(B−V )UV to derive
E(B −V ) in galaxies if optical wavelengths are not ac-
cessible.

• Possible UV diagnostics for 12+log(O/H) and log(U)

were explored in Sec. 6.4 and Fig. 14, where we also
confirm that UV lines are usually sensitive to both pa-
rameters. We proposed C III] λλ1907,9/O III] λ1666
 CLASSY IV 35

We thank the anonymous referee for their comments and trophysics, Max Planck Institute for Extraterrestrial Physics,
advice, which help improve the quality of the paper. MM, New Mexico State University, New York University, Ohio
DAB and XX are grateful for the support for this program, State University, Pennsylvania State University, University of
HST-GO-15840, that was provided by NASA through a Portsmouth, Princeton University, the Spanish Participation
grant from the Space Telescope Science Institute, which is Group, University of Tokyo, University of Utah, Vanderbilt
operated by the Associations of Universities for Research University, University of Virginia, University of Washington,
in Astronomy, Incorporated, under NASA contract NAS5- and Yale University.
26555. Also, MM is grateful to Carlo Cannarozzo for inspir- This work also uses the services of the ESO Science
ing conversations and advice. BLJ, SH and NK are thank- Archive Facility, observations collected at the European
ful for support from the European Space Agency (ESA). Southern Observatory under ESO programmes 096.B-0690,
AF acknowledges the support from grant PRIN MIUR2017- 0103.B-0531, 0103.D-0705, and 0104.D-0503, and observa-
20173ML3WW_001. AW acknowledges the support of tions obtained with the Large Binocular Telescope (LBT).
UNAM via grant agreement PAPIIT no. IN106922. RA ac- The LBT is an international collaboration among institu-
knowledges support from ANID Fondecyt Regular 1202007. tions in the United States, Italy and Germany. LBT Cor-
The CLASSY collaboration extends special gratitude to poration partners are: The University of Arizona on behalf
the Lorentz Center for useful discussions during the "Charac- of the Arizona Board of Regents; Istituto Nazionale di As-
terizing Galaxies with Spectroscopy with a view for JWST" trofisica, Italy; LBT Beteiligungsgesellschaft, Germany, rep-
2017 workshop that led to the formation of the CLASSY col- resenting the Max-Planck Society, The Leibniz Institute for
laboration and survey. The CLASSY collaboration thanks Astrophysics Potsdam, and Heidelberg University; The Ohio
the COS team for all their assistance and advice in the reduc- State University, University of Notre Dame, University of
tion of the COS data. Minnesota, and University of Virginia.
Funding for SDSS-III has been provided by the Alfred This research has made use of the HSLA database, devel-
P. Sloan Foundation, the Participating Institutions, the Na- oped and maintained at STScI, Baltimore, USA.
tional Science Foundation, and the U.S. Department of
Facilities: HST (COS), LBT (MODS), APO (SDSS),
Energy Office of Science. The SDSS-III web site is
KECK (ESI), VLT (MUSE, VIMOS)
http://www.sdss3.org/. SDSS-III is managed by the Astro-
physical Research Consortium for the Participating Institu- Software: astropy (The Astropy Collaboration 2013,
tions of the SDSS-III Collaboration including the University 2018) BEAGLE (Chevallard & Charlot 2006), CalCOS
of Arizona, the Brazilian Participation Group, Brookhaven (STScI), dustmaps (Green 2018), jupyter (Kluyver 2016),
National Laboratory, Carnegie Mellon University, Univer- LINMIX (Kelly 2007) MPFIT (Markwardt 2009), MODS re-
sity of Florida, the French Participation Group, the German duction Pipeline, Photutils (Bradley 2021), PYNEB (Lurid-
Participation Group, Harvard University, the Instituto de As- iana 2012; 2015), python, pysynphot (STScI Development
trofisica de Canarias, the Michigan State/Notre Dame/JINA Team), RASCAS (Michel-Dansac 2020), SALT (Scarlata
Participation Group, Johns Hopkins University, Lawrence & Panagia 2015), STARLIGHT (Fernandes 2005), TLAC
Berkeley National Laboratory, Max Planck Institute for As- (Gronke & Dijkstra 2014), XIDL

REFERENCES
Abazajian, K. N., Adelman-McCarthy, J. K., Agüeros, M. A., et al. Arellano-Córdova, K. Z., Mingozzi, M., Berg, D. A., et al. 2022,
2009, ApJS, 182, 543, doi: 10.1088/0067-0049/182/2/543 ApJ, 935, 74, doi: 10.3847/1538-4357/ac7854
Acharyya, A., Kewley, L. J., Rigby, J. R., et al. 2019, MNRAS, Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009,
488, 5862, doi: 10.1093/mnras/stz1987 ARA&A, 47, 481
Bayliss, M. B., Rigby, J. R., Sharon, K., et al. 2014, ApJ, 790, 144
Albareti, F. D., Allende Prieto, C., Almeida, A., et al. 2017, ApJS,
Berg, D. A., Chisholm, J., Erb, D. K., et al. 2019a, ApJL, 878, L3,
233, 25, doi: 10.3847/1538-4365/aa8992
doi: 10.3847/2041-8213/ab21dc
Amorín, R., Fontana, A., Pérez-Montero, E., et al. 2017, Nature
—. 2021, arXiv e-prints, arXiv:2105.12765.
Astronomy, 1, 0052
https://arxiv.org/abs/2105.12765
Amorín, R., Vílchez, J. M., Hägele, G. F., et al. 2012, ApJL, 754, Berg, D. A., Erb, D. K., Auger, M. W., Pettini, M., & Brammer,
L22, doi: 10.1088/2041-8205/754/2/L22 G. B. 2018, ApJ, 859, 164
Arellano-Córdova, K. Z., Esteban, C., García-Rojas, J., & Berg, D. A., Erb, D. K., Henry, R. B. C., Skillman, E. D., &
Méndez-Delgado, J. E. 2020, MNRAS, 496, 1051, McQuinn, K. B. W. 2019b, ApJ, 874, 93,
doi: 10.1093/mnras/staa1523 doi: 10.3847/1538-4357/ab020a
36 M INGOZZI ET AL .

Berg, D. A., Skillman, E. D., Croxall, K., et al. 2015, ApJ, 806, 16 Dopita, M. A., Fischera, J., Sutherland, R. S., et al. 2006, ApJ, 647,
Berg, D. A., Skillman, E. D., Henry, R. B. C., Erb, D. K., & Carigi, 244, doi: 10.1086/505418
L. 2016, ApJ, 827, 126 Eisenstein, D. J., Weinberg, D. H., Agol, E., et al. 2011, AJ, 142, 72
Berg, D. A., James, B. L., King, T., et al. 2022, ApJS, 261, 31, Eldridge, J. J., & Stanway, E. R. 2016, MNRAS, 462, 3302
doi: 10.3847/1538-4365/ac6c03 Erb, D. K., Pettini, M., Shapley, A. E., et al. 2010, ApJ, 719, 1168
Bian, F., Kewley, L. J., Dopita, M. A., & Juneau, S. 2016, ApJ, Erb, D. K., Steidel, C. C., Shapley, A. E., et al. 2006, ApJ, 646,
822, 62, doi: 10.3847/0004-637X/822/2/62 107, doi: 10.1086/504891
Binette, L., Matadamas, R., Hägele, G. F., et al. 2012, A&A, 547, Feltre, A., Charlot, S., & Gutkin, J. 2016, MNRAS, 456, 3354
A29 Ferland, G. J., Porter, R. L., van Hoof, P. A. M., & others. 2013,
Bosch, G., Hägele, G. F., Amorín, R., et al. 2019, MNRAS, 489, RMxAA, 49, 137
1787, doi: 10.1093/mnras/stz2230 Fosbury, R. A. E., Villar-Martín, M., & others. 2003, ApJ, 596, 797
Brinchmann, J., Kunth, D., & Durret, F. 2008, A&A, 485, 657. Garnett, D. R. 1990, ApJ, 363, 142
https://arxiv.org/abs/0805.1073 Garnett, D. R. 1992, AJ, 103, 1330
Bruzual, G., & Charlot, S. 2003, MNRAS, 344, 1000, Garnett, D. R., Edmunds, M. G., Henry, R. B. C., Pagel, B. E. J., &
doi: 10.1046/j.1365-8711.2003.06897.x Skillman, E. D. 2004, AJ, 128, 2772
Byler, N., Dalcanton, J. J., Conroy, C., & Johnson, B. D. 2017, Garnett, D. R., Skillman, E. D., Dufour, R. J., et al. 1995, ApJ,
ApJ, 840, 44 443, 64
Byler, N., Dalcanton, J. J., Conroy, C., et al. 2018, ApJ, 863, 14,
Geen, S., Pellegrini, E., Bieri, R., & Klessen, R. 2020, MNRAS,
doi: 10.3847/1538-4357/aacd50
492, 915, doi: 10.1093/mnras/stz3491
Byler, N., Kewley, L. J., Rigby, J. R., et al. 2020, ApJ, 893, 1,
Giuliani, J. L., J. 1981, ApJ, 245, 903, doi: 10.1086/158867
doi: 10.3847/1538-4357/ab7ea9
Gordon, K. D., & Clayton, G. C. 1998, ApJ, 500, 816,
Calzetti, D. 1997, in American Institute of Physics Conference
doi: 10.1086/305774
Series, Vol. 408, The ultraviolet universe at low and High
Gordon, K. D., Clayton, G. C., Misselt, K. A., Landolt, A. U., &
redshift, ed. W. H. Waller, 403–412
Wolff, M. J. 2003, ApJ, 594, 279, doi: 10.1086/376774
Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682
Green, G. 2018, The Journal of Open Source Software, 3, 695,
Calzetti, D., Kinney, A. L., & Storchi-Bergmann, T. 1994, ApJ,
doi: 10.21105/joss.00695
429, 582, doi: 10.1086/174346
Green, G. M., Schlafly, E. F., Finkbeiner, D. P., et al. 2015, ApJ,
Cappellari, M., Scott, N., Alatalo, K., et al. 2013, MNRAS, 432,
810, 25, doi: 10.1088/0004-637X/810/1/25
1709, doi: 10.1093/mnras/stt562
Guseva, N. G., Izotov, Y. I., Fricke, K. J., & Henkel, C. 2017,
Cardelli, J. A., Clayton, G. C., & Mathis, J. S. 1989, ApJ, 345, 245
A&A, 599, A65, doi: 10.1051/0004-6361/201629181
Chabrier, G. 2003, PASP, 115, 763, doi: 10.1086/376392
Guseva, N. G., Izotov, Y. I., Stasińska, G., et al. 2011, A&A, 529,
Chevallard, J., & Charlot, S. 2016, MNRAS, 462, 1415,
A149
doi: 10.1093/mnras/stw1756
Gutkin, J., Charlot, S., & Bruzual, G. 2016, MNRAS, 462, 1757
Chiappini, C., Romano, D., & Matteucci, F. 2003, MNRAS, 339,
63 Hainline, K. N., Shapley, A. E., Kornei, K. A., et al. 2009, ApJ,
Chisholm, J., Rigby, J. R., Bayliss, M., et al. 2019, ApJ, 882, 182, 701, 52
doi: 10.3847/1538-4357/ab3104 Henry, R. B. C., Edmunds, M. G., & Köppen, J. 2000, ApJ, 541,
Christensen, L., Laursen, P., Richard, J., et al. 2012, MNRAS, 427, 660
1973 Hogarth, L., Amorín, R., Vílchez, J. M., et al. 2020, MNRAS, 494,
Cid Fernandes, R., Mateus, A., Sodré, L., Stasińska, G., & Gomes, 3541, doi: 10.1093/mnras/staa851
J. M. 2005, MNRAS, 358, 363, Hoopes, C. G., & Walterbos, R. A. M. 2003, ApJ, 586, 902,
doi: 10.1111/j.1365-2966.2005.08752.x doi: 10.1086/367954
Crowther, P. A. 2007, 45, 177 Izotov, Y. I., & Thuan, T. X. 1999, ApJ, 511, 639
Crowther, P. A., & Smith, L. J. 1997, A&A, 320, 500 —. 2007, ApJ, 665, 1115
Curti, M., Cresci, G., Mannucci, F., et al. 2017, MNRAS, 465, James, B. L., Auger, M., Pettini, M., et al. 2018, MNRAS, 476,
1384, doi: 10.1093/mnras/stw2766 1726, doi: 10.1093/mnras/sty315
Dopita, M. A., & Evans, I. N. 1986, ApJ, 307, 431, James, B. L., Pettini, M., Christensen, L., et al. 2014, MNRAS,
doi: 10.1086/164432 440, 1794
Dopita, M. A., Rich, J., Vogt, F. P. A., et al. 2014, Ap&SS, 350, James, B. L., Tsamis, Y. G., Barlow, M. J., et al. 2009, MNRAS,
741, doi: 10.1007/s10509-013-1753-2 398, 2, doi: 10.1111/j.1365-2966.2009.15172.x
 CLASSY IV 37

James, B. L., Berg, D. A., King, T., et al. 2022, arXiv e-prints, Markwardt, C. B. 2009, in Astronomical Society of the Pacific
arXiv:2206.01224. https://arxiv.org/abs/2206.01224 Conference Series, Vol. 411, Astronomical Data Analysis
Jaskot, A. E., & Oey, M. S. 2013, ApJ, 766, 91 Software and Systems XVIII, ed. D. A. Bohlender, D. Durand,
Jaskot, A. E., & Ravindranath, S. 2016, ApJ, 833, 136 & P. Dowler, 251
Juan de Dios, L., & Rodríguez, M. 2017, MNRAS, 469, 1036, Maseda, M. V., Brinchmann, J., Franx, M., et al. 2017, A&A, 608,
doi: 10.1093/mnras/stx916 A4, doi: 10.1051/0004-6361/201730985
—. 2021, MNRAS, 507, 5331, doi: 10.1093/mnras/stab2488 McLure, R. J., Pentericci, L., Cimatti, A., et al. 2018, MNRAS,
479, 25, doi: 10.1093/mnras/sty1213
JWST User Documentation. 2016, JWST User Documentation
Méndez-Delgado, J. E., Esteban, C., García-Rojas, J., et al. 2021,
(JDox), JWST User Documentation Website
MNRAS, 502, 1703, doi: 10.1093/mnras/stab068
Kaasinen, M., Kewley, L., Bian, F., et al. 2018, MNRAS, 477,
Mingozzi, M., Berg, D. A., Chisholm, J., et al. 2022, ApJ
5568, doi: 10.1093/mnras/sty1012
Mingozzi, M., Belfiore, F., Cresci, G., et al. 2020, A&A, 636, A42,
Kaplan, K. F., Jogee, S., Kewley, L., et al. 2016, MNRAS, 462,
doi: 10.1051/0004-6361/201937203
1642, doi: 10.1093/mnras/stw1422
Morisset, C., Luridiana, V., García-Rojas, J., et al. 2020, Atoms, 8,
Kashino, D., Silverman, J. D., Rodighiero, G., et al. 2013, ApJL,
66, doi: 10.3390/atoms8040066
777, L8, doi: 10.1088/2041-8205/777/1/L8
Nagao, T., Murayama, T., & Taniguchi, Y. 2001, ApJ, 549, 155,
Keenan, F. P., Dufton, P. L., Feibelman, W. A., et al. 1994, ApJ, doi: 10.1086/319062
423, 882, doi: 10.1086/173866 Nakajima, K., & Ouchi, M. 2014, MNRAS, 442, 900,
Keenan, F. P., Ramsbottom, C. A., Bell, K. L., et al. 1995, ApJ, doi: 10.1093/mnras/stu902
438, 500, doi: 10.1086/175094 Nakajima, K., Schaerer, D., Le Fèvre, O., et al. 2018, A&A, 612,
Kennicutt, Jr., R. C., Bresolin, F., & Garnett, D. R. 2003, ApJ, 591, A94, doi: 10.1051/0004-6361/201731935
801 Nanayakkara, T., Brinchmann, J., Boogaard, L., et al. 2019, A&A,
Kewley, L. J., & Dopita, M. A. 2002, ApJS, 142, 35 624, A89, doi: 10.1051/0004-6361/201834565
Kewley, L. J., Nicholls, D. C., & Sutherland, R. S. 2019, ARA&A, Nicholls, D. C., Kewley, L. J., & Sutherland, R. S. 2020, PASP,
57, 511, doi: 10.1146/annurev-astro-081817-051832 132, 033001, doi: 10.1088/1538-3873/ab6818
Kewley, L. J., Zahid, H. J., Geller, M. J., et al. 2015, ApJL, 812, Osterbrock, D. E. 1989, Astrophysics of gaseous nebulae and
L20, doi: 10.1088/2041-8205/812/2/L20 active galactic nuclei
Kojima, T., Ouchi, M., Nakajima, K., et al. 2017, PASJ, 69, 44, Osterbrock, D. E., & Ferland, G. J. 2006, Astrophysics of gaseous
doi: 10.1093/pasj/psx017 nebulae and active galactic nuclei
Komarova, L., Oey, M. S., Krumholz, M. R., et al. 2021, ApJL, Pannella, M., Elbaz, D., Daddi, E., et al. 2015, ApJ, 807, 141,
920, L46, doi: 10.3847/2041-8213/ac2c09 doi: 10.1088/0004-637X/807/2/141
Kroupa, P. 2001, MNRAS, 322, 231, Patrício, V., Christensen, L., Rhodin, H., Cañameras, R., &
doi: 10.1046/j.1365-8711.2001.04022.x Lara-López, M. A. 2018, MNRAS, 481, 3520,
doi: 10.1093/mnras/sty2508
Leitherer, C. 2005, in American Institute of Physics Conference
Patrício, V., Richard, J., Verhamme, A., et al. 2016, MNRAS, 456,
Series, Vol. 783, The Evolution of Starbursts, ed. S. Hüttmeister,
4191
E. Manthey, D. Bomans, & K. Weis, 280–295
Pellegrini, E. W., Rahner, D., Reissl, S., et al. 2020, MNRAS, 496,
Leitherer, C., Ortiz Otálvaro, P. A., & Bresolin, F. o. 2010, ApJS,
339, doi: 10.1093/mnras/staa1473
189, 309, doi: 10.1088/0067-0049/189/2/309
Pérez-Montero, E., & Amorín, R. 2017, MNRAS, 467, 1287,
Leitherer, C., Schaerer, D., Goldader, J. D., et al. 1999, ApJS, 123,
doi: 10.1093/mnras/stx186
3
Plat, A., Charlot, S., Bruzual, G., et al. 2019, MNRAS, 490, 978,
Llerena, M., Amorín, R., Cullen, F., et al. 2022, A&A, 659, A16,
doi: 10.1093/mnras/stz2616
doi: 10.1051/0004-6361/202141651 Price, S. H., Kriek, M., Brammer, G. B., et al. 2014, ApJ, 788, 86,
Madau, P., & Haardt, F. 2015, ApJL, 813, L8 doi: 10.1088/0004-637X/788/1/86
Maeder, A. 1992, A&A, 264, 105 Puglisi, A., Rodighiero, G., Franceschini, A., et al. 2016, A&A,
Mainali, R., Kollmeier, J. A., Stark, D. P., & others. 2017, ApJL, 586, A83, doi: 10.1051/0004-6361/201526782
836, L14 Qin, J., Zheng, X. Z., Wuyts, S., Pan, Z., & Ren, J. 2019, ApJ, 886,
Mainali, R., Zitrin, A., Stark, D. P., et al. 2018, MNRAS, 479, 1180 28, doi: 10.3847/1538-4357/ab4a04
Maiolino, R., & Mannucci, F. 2019, A&A Rv, 27, 3, Quider, A. M., Pettini, M., Shapley, A. E., & Steidel, C. C. 2009,
doi: 10.1007/s00159-018-0112-2 MNRAS, 398, 1263
38 M INGOZZI ET AL .

Raiter, A., Fosbury, R. A. E., & Teimoorinia, H. 2010, A&A, 510, Shapley, A. E., Steidel, C. C., Pettini, M., & Adelberger, K. L.
A109, doi: 10.1051/0004-6361/200912429 2003, ApJ, 588, 65
Ravindranath, S., Monroe, T., Jaskot, A., Ferguson, H. C., & Shivaei, I., Reddy, N. A., Shapley, A. E., et al. 2015, ApJ, 815, 98,
Tumlinson, J. 2020, ApJ, 896, 170, doi: 10.1088/0004-637X/815/2/98
doi: 10.3847/1538-4357/ab91a5 Shivaei, I., Reddy, N., Rieke, G., et al. 2020, ApJ, 899, 117,
Reddy, N. A., Erb, D. K., Pettini, M., Steidel, C. C., & Shapley, doi: 10.3847/1538-4357/aba35e
A. E. 2010, ApJ, 712, 1070, Skillman, E. D. 1998, in Stellar astrophysics for the local group:
doi: 10.1088/0004-637X/712/2/1070 VIII Canary Islands Winter School of Astrophysics, ed.
Reddy, N. A., Steidel, C. C., Pettini, M., & Bogosavljević, M. A. Aparicio, A. Herrero, & F. Sánchez, 457
2016, ApJ, 828, 107, doi: 10.3847/0004-637X/828/2/107 Stanghellini, L., & Kaler, J. B. 1989, ApJ, 343, 811,
Reddy, N. A., Kriek, M., Shapley, A. E., et al. 2015, ApJ, 806, doi: 10.1086/167751
259, doi: 10.1088/0004-637X/806/2/259 Stanway, E. R., Eldridge, J. J., & Becker, G. D. 2016, MNRAS,
Rickards Vaught, R. J., Sandstrom, K. M., & Hunt, L. K. 2021, 456, 485
ApJL, 911, L17, doi: 10.3847/2041-8213/abf09b Stark, D. P. 2016, ARA&A, 54, 761
Rigby, J. R., Bayliss, M. B., Gladders, M. D., & others. 2015, Stark, D. P., Richard, J., Charlot, S., & others. 2015, MNRAS, 450,
ApJL, 814, L6 1846
Rigby, J. R., Bayliss, M. B., Sharon, K., et al. 2018, AJ, 155, 104, Stark, D. P., Richard, J., Siana, B., et al. 2014, MNRAS, 445, 3200
doi: 10.3847/1538-3881/aaa2ff Steidel, C. C., Rudie, G. C., Strom, A. L., et al. 2014, ApJ, 795,
Rigby, J. R., Florian, M., Acharyya, A., et al. 2021, ApJ, 908, 154, 165
doi: 10.3847/1538-4357/abcfc9 Steidel, C. C., Strom, A. L., Pettini, M., et al. 2016, ApJ, 826, 159
Robertson, B. E., Ellis, R. S., Furlanetto, S. R., & Dunlop, J. S. Theios, R. L., Steidel, C. C., Strom, A. L., et al. 2019, ApJ, 871,
2015, ApJL, 802, L19 128, doi: 10.3847/1538-4357/aaf386
Rousseeuw, P. J., & Van Driessen, K. 2006, Data Mining and Thuan, T. X., Izotov, Y. I., & Lipovetsky, V. A. 1996, ApJ, 463,
Knowledge Discovery, 12, 29, doi: 10.1007/s10618-005-0024-4 120, doi: 10.1086/177228
Sanders, R. L., Shapley, A. E., Reddy, N. A., et al. 2020, MNRAS, Treu, T., Roberts-Borsani, G., Bradac, M., et al. 2022, arXiv
491, 1427, doi: 10.1093/mnras/stz3032 e-prints, arXiv:2206.07978. https://arxiv.org/abs/2206.07978
Schaerer, D., Izotov, Y. I., Worseck, G., et al. 2022, arXiv e-prints, Vanzella, E., De Barros, S., Cupani, G., & others. 2016, ApJL, 821,
arXiv:2202.07768. https://arxiv.org/abs/2202.07768 L27
Schmidt, K. B., Huang, K.-H., Treu, T., et al. 2017, ApJ, 839, 17 Vanzella, E., Grazian, A., Hayes, M., et al. 2010, A&A, 513, A20,
Schmidt, K. B., Kerutt, J., Wisotzki, L., et al. 2021, arXiv e-prints, doi: 10.1051/0004-6361/200913042
arXiv:2108.01713. https://arxiv.org/abs/2108.01713 Vidal-García, A., Charlot, S., Bruzual, G., & Hubeny, I. 2017,
Senchyna, P., Chisholm, J., Berg, D. A., James, B. L., & others (in MNRAS, 470, 3532, doi: 10.1093/mnras/stx1324
Villar-Martín, M., Cerviño, M., & González Delgado, R. M. 2004,
preparation). 2022a, ApJ
MNRAS, 355, 1132, doi: 10.1111/j.1365-2966.2004.08395.x
Senchyna, P., & Stark, D. P. 2019, MNRAS, 484, 1270,
Wei, Z. 1988, Astrophysical Letters and Communications, 27, 275
doi: 10.1093/mnras/stz058
Wild, V., Charlot, S., Brinchmann, J., et al. 2011a, MNRAS, 417,
Senchyna, P., Stark, D. P., Charlot, S., et al. 2021, MNRAS, 503,
1760, doi: 10.1111/j.1365-2966.2011.19367.x
6112, doi: 10.1093/mnras/stab884
Wild, V., Groves, B., Heckman, T., et al. 2011b, MNRAS, 410,
Senchyna, P., Stark, D. P., Chevallard, J., et al. 2019, MNRAS,
1593, doi: 10.1111/j.1365-2966.2010.17536.x
488, 3492, doi: 10.1093/mnras/stz1907
Wise, J. H., Demchenko, V. G., Halicek, M. T., et al. 2014,
Senchyna, P., Stark, D. P., Vidal-García, A., et al. 2017, MNRAS,
MNRAS, 442, 2560
472, 2608
Wofford, A., Vidal-García, A., Feltre, A., Chevallard, J., et al.
Senchyna, P., Stark, D. P., Charlot, S., et al. 2022b, ApJ, 930, 105,
2021, MNRAS, 500, 2908, doi: 10.1093/mnras/staa3365
doi: 10.3847/1538-4357/ac5d38
 CLASSY IV 39

APPENDIX

A. UV-OPTICAL FLUX OFFSET

Before comparing COS and optical data, it was necessary to check and correct optical spectra for potential UV-optical flux
offsets. Below we explain in detail the procedure we carried out to check if there is alignment between the continuum flux
of the two wavelength regimes. Such an offset is expected since the UV and optical spectra have been obtained via different
instruments and different apertures. Specifically, we have to take into account: (i) the different aperture size of COS (2.5"), SDSS
(3"), MMT (10"×1" slit) and LBT (2.5"×1" slit), while the MUSE, VIMOS and KCWI data were extracted using an aperture of
2.5", identical to COS; (ii) a possible different orientation of the COS and SDSS apertures, and MMT and LBT slits, for extended
sources and/or targets with multi-component light profiles throughout the aperture; (iii) COS vignetting of non-compact objects;
(iv) flux calibration issues specific to the instrument. Having said that, we stress that this is an issue only for mixed UV-optical
diagnostics, such as the O III] λ1666/λ5007 line ratio, that can be used to measure the electron temperature of the gas (see
Sec. 4.3, 5.3 and 6.3). Indeed, Paper V analyzed the optical data of 12 CLASSY galaxies taking into account SDSS, VLT/MUSE,
and, for a subsample, LBT/MODS and Keck/ESI longslit spectra, to assess the impact of using different aperture combinations
on the determination of the ISM physical conditions (i.e., ne , Te , E(B − V ), log(U) and 12+log(O/H)). Overall, Paper V found
that these measurements remained roughly constant with aperture size, indicating that the optical gas conditions are relatively
uniform for this sample.
To explore the UV-optical flux offset, we extended the UV stellar continuum best-fit performed in the range 1200-2000 Å with
the method from S22 and C&B models as described in Sec. 3.1.1, to optical wavelengths up to ∼ 9000 Å, taking into account a
SMC attenuation law (Gordon et al. 2003). Here we stress that the choice of the attenuation law is not affecting the results in the
optical wavelength range, since the different known attenuation laws (e.g., Cardelli et al. 1989, Calzetti et al. 2000, Gordon et al.
2003) are all superimposed at λ & 3800 Å (e.g., see Fig. 3 in Shivaei et al. 2020). This is not the case in the UV wavelength range,
where we adopted the most suitable Reddy et al. (2016) attenuation law, as discussed in Sec. 3.1.1 and Sec. 4.1. The performed
extrapolation of the UV stellar continuum best-fit towards the optical wavelengths allows us to perform a direct comparison with
the optical spectrum fitted with Starlight as described in Sec. 3.1.2, by providing an overlapping wavelength window. This is
illustrated in Fig. 17, which shows the observed COS (black solid line) and optical SDSS and MUSE (gray and black dotted line,
respectively) spectra for the CLASSY galaxies J0021+0052 and J1144+4012. Overlaid are the extended-UV stellar-continuum
best-fit (dashed blue line) and the optical stellar-continuum best-fit (dashed red and magenta line for SDSS and MUSE spectra,
respectively). The UV-optical flux offset is then estimated by dividing the extended UV stellar continuum fit by the optical stellar
continuum best-fit in eight featureless wavelength windows (40 Å each in width) between 3700 and 5600 Å (highlighted in cyan),
avoiding the region of the 4000 Å break, and calculating their median value and using the standard deviation as corresponding
error. The final optical spectra, corrected for the UV-optical flux offset, are shown by the green and orange solid lines for SDSS
and MUSE, respectively.
The left panel of Fig. 18 shows the UV-optical flux offsets measured throughout the CLASSY sample, using different symbols
to indicate the SDSS, MUSE, LBT, KCWI, VIMOS and MMT observations, as reported in the legend. An important aspect
to explore is that there is no dependence of the UV-optical flux offset we measure with the galaxy extent, that for some targets
can be larger than that covered by the COS and optical apertures that we are taking into account. In these cases, COS is not
measuring the entire flux for these objects. For instance, in the hypothesis that the sources were filling uniformly the COS and
SDSS apertures centered in the same position, then the optical fluxes should be scaled by the ratio of their apertures (2.5”/3.0”)2
(i.e., dashed magenta line). The dash-dot blue blue line in Fig. 18 represents the median value of the SDSS UV-optical offset,
taking into account the 24 galaxies with an average half light radius (r50 ) larger than the optical extent, calculated by plotting the
fraction of the optical flux from PanSTARRS imaging (taken from Table 6 in Paper I and reported in Tab. 1). This value is close
to the magenta line, but offsets obtained for the SDSS galaxies (blue dots) are spread in the range ∼ 0.5 − 1.3, demonstrating that
there are other factors at play. Furthermore, in the right panel of Fig. 18 we show the UV-optical offset as a function of r50 , and
reassuringly we notice that there is no correlation between the two (Pearson factor of ∼ 0.02, with pvalue of ∼ 0.90). The dotted
red and dashed green horizontal lines in both the left and right panels of Fig. 18 show the median UV-optical flux offsets for
the MUSE and LBT spectra, respectively. In particular, the COS and MUSE-KCWI-VIMOS data are aperture-matched. Indeed,
these optical spectra were obtained by selecting the same aperture and position of COS (see Sec. 2), and thus the difference
in aperture size can be completely ruled out as the origin of the flux offset. Nevertheless, the flux offset calculated can be far
different than 1, down to ∼ 0.05 for the VIMOS data and up to ∼ 1.95 for the KCWI, and with values closer to ∼ 1 for MUSE.
40 M INGOZZI ET AL .
J0021+0052
COS spectrum
5 SDSS spectrum
MUSE spectrum
CB fit
Starlight SDSS fit
4 Starlight MUSE fit
SDSS spectrum - corr
MUSE spectrum - corr
3
erg cm 2 s 1 Å 1) 2
1
0
1000 2000 3000 4000 5000 6000 7000 8000 9000
15

0.6
F (×10

0.4
SDSS spectrum
0.2 MUSE spectrum
Starlight SDSS fit
Starlight MUSE fit
SDSS spectrum - corr
MUSE spectrum - corr
0.03500 4000 4500 5000 5500 6000
Rest Wavelength (Å)
J1144+4012
2.0 COS spectrum
SDSS spectrum
CB fit
Starlight SDSS fit
1.5 SDSS spectrum - corr

1.0
erg cm 2 s 1 Å 1)

0.5

0.0
1000 2000 3000 4000 5000 6000 7000 8000 9000
0.5
15

0.4
F (×10

0.3
0.2
SDSS spectrum
0.1 Starlight SDSS fit
Starlight MUSE fit
SDSS spectrum - corr
0.03500 4000 4500 5000 5500 6000
Rest Wavelength (Å)

Figure 17. Upper panels: COS, SDSS and MUSE spectra of the galaxy J0021+0052. The COS spectrum is shown with a black solid line, while
the original SDSS and MUSE optical spectra with a dotted gray and black line, respectively. The UV-optical flux offset is estimated comparing
the UV-extended stellar continuum fitting (dashed blue line) and optical stellar continuum fitting (dashed red line for SDSS and magenta for
MUSE) in several featureless wavelength ranges of 40 Å between 4000 and 5600 Å, highlighted in cyan. The UV-optical flux offset corrected
spectra are shown from the green and orange solid lines, for SDSS and MUSE, respectively. The bottom panel shows a zoom on the optical
region of the spectra. Lower panels: same for the galaxy J1144+4012.
 CLASSY IV 41

Median COS/SDSS
6 Median COS/MUSE Median COS/SDSS
Median COS/LBT 6 Median COS/MUSE
COS/SDSS aperture scale COS/SDSS aperture scale
COS/SDSS Median COS/LBT
5 COS/MUSE COS/SDSS
COS/LBT 5 COS/KCWI
COS/KCWI COS/VIMOS
COS/VIMOS COS/MMT
4 COS/MMT COS/MUSE
Stellar fit UV/Optical

4 COS/LBT

Stellar fit UV/Optical

3
3
2
2
1
1
0
0
J0021+0052
J0036-3333
J0127-0619
J0144+0453
J0337-0502
J0405-3648
J0808+3948
J0823+2806
J0926+4427
J0934+5514
J0938+5428
J0940+2935
J0942+3547
J0944+3424
J0944-0038
J1016+3754
J1024+0524
J1025+3622
J1044+0353
J1105+4444
J1112+5503
J1119+5130
J1129+2034
J1132+1411
J1132+5722
J1144+4012
J1148+2546
J1150+1501
J1157+3220
J1200+1343
J1225+6109
J1253-0312
J1314+3452
J1323-0132
J1359+5726
J1416+1223
J1418+2102
J1428+1653
J1429+0643
J1444+4237
J1448-0110
J1521+0759
J1525+0757
J1545+0858
J1612+0817

1 10
r50 ["]

Figure 18. Left panel: UV-optical flux offset for each CLASSY galaxy, obtained from the comparison of extended-UV stellar continuum
fitting and optical stellar continuum fitting of SDSS (blue circles), MUSE (red triangles), KCWI (gold pentagon), VIMOS (purple square), LBT
(green hexagons) and MMT (gray diamond) data. Right panel: UV-optical flux offset as a function of the average half light radius (r50 ) taken
from Paper I and calculated from PanSTARRS imaging. The solid magenta horizontal lines are centred at the ratio of COS/SDSS aperture (i.e.,
(2.5"/3")2 ), while the dash-dot blue, dotted red and dashed green horizontal lines show the median UV-optical flux offsets for the SDSS, MUSE
and LBT spectra, respectively. The uncertainties are shown for all the displayed quantities.

In these cases, apart from flux calibration issues, there are other effects within the COS aperture that may cause a flux offset here,
such as vignetting (see Paper II for details).
We stress that, as already reported in Sec. 3.1, the C&B models include the nebular continuum. Specifically, the nebular
continuum emits significantly in the UV blue-ward of the Balmer break, accounting for ∼ 5 − 10% of the total flux in the range
900–1800 Å and up to ∼ 20% in the range 1800–4000 Å (Byler et al. 2017), increasing the overall continuum level. Also, its
impact increases if the gas is characterized by high ionization parameter and low metallicity, typical of CLASSY objects (see e.g.,
Fig. 12 of Byler et al. 2017). Therefore, it is fundamental to take its contribution into account for our comparison. A caveat of our
approach is that in the UV stellar-continuum best-fit, we fixed the ionization parameter to log(U) = −2.5, which sets the overall
intensity of the nebular continuum spectrum, since recombination emission depends on the number of incident ionizing photons
(Byler et al. 2017). However, we tested to what extent the variation of the stellar population ionization parameter in the range
[-3; -1] could influence the nebular continuum shape to estimate the UV-optical offset, finding minimal variations (< 0.05 dex
compared to the case with log(U) = −2.5) in the 4000–6000 Å wavelength range considered for the comparison. Overall, we
conclude that the observed UV-optical flux offsets are primarily due to systematic effects of the instruments such as the COS
aperture vignetting and/or uncertainties in the flux calibration.
To summarize, the need to use hybrid line ratios such as O III] λ1666/[O III] λ4363 from UV and optical spectra, where each
line is measured using different instruments, makes it essential to align the continuum fluxes using scaling factors. Deriving
this scaling factor is not trivial, since it is not related to a single cause such as the aperture difference, but could instead be due
to a combination of instrumental systematics beyond our control (e.g., flux calibration, vignetting). Therefore, we consider the
method described above our best avenue in estimating this important flux offset. This offset is reported in Table 5 for the first 5
galaxies of the CLASSY sample, as an example. The complete table covering all CLASSY galaxies is available on the CLASSY
MAST Webpage.
42 M INGOZZI ET AL .

B. UV EMISSION LINES FITS

In Figures 19–21 we show each of the UV emission line detections (i.e., corresponding to S/N > 3) for N IV] λλ1483,87,
He II λ1640, O III] λλ1661,6, [N III] λ1750 and Si III] λλ1883,92. The analog figures for C IV λλ1548,51 and C III] λλ1907,09
are shown in Sec. 3.2.3 (Fig. 2). The observed flux and the best-fit model are shown in black and red, respectively, while their
uncertainties are given by the gray and red shades. The dashed vertical lines indicate the position of the emission features
according to the literature z, reported in Table 1. In each figure, the color of the frames indicates which dataset and binning are
used for the fit: HR rebinned of 15 pixels spectra in blue, HR rebinned of 30 in cyan, MR rebinned of 6 in dark green, MR
rebinned of 12 in light green. The dereddened fluxes of the fitted UV lines (expressed in 10−15 erg/s/cm2 ) and the equivalent
widths (in Å) shown in this paper are reported in Tab. 4 for the first 5 galaxies of the CLASSY sample, as an example. The
complete table covering all CLASSY galaxies is available on the CLASSY MAST Webpage. The E(B − V ) taken into account
(see Sec. 4.1 and Sec. 5.1) is also reported. The fluxes for undetected lines are given as less than their 3σ upper limits.

Figure 19. Fit of the N IV] λλ1483,87 emission line doublet, with one of the lines detected with a S/N > 3 in 6 out of 44 galaxies. The
observed flux and the best-fit model are shown in black and red, respectively, while their uncertainties are given by the gray and red shades.
The black dashed vertical lines indicate the line positions, taking into account zlit. .

Figure 20. Same as Fig. 19 for the [N III] λ1750 multiplet in the CLASSY survey detected with a S/N > 3 in 2 out of 44 galaxies.

Figure 21. Same as Fig. 19 for the Si III] λλ1883,92 doublet, with one of the lines detected with a S/N > 3 in 6 out of 44 galaxies.
 CLASSY IV 43

Figure 22. Same as Fig. 19 for the He II λ1640 emission line in emission, detected with a S/N > 3 in 19 out of 44 galaxies.

Figure 23. Same as Fig. 19 for [O III] λλ1661,6 emission line in the CLASSY survey detected with a S/N > 3 in 22 out of 44 galaxies.
44 M INGOZZI ET AL .

Table 4. FUV Emission line fluxes from HST/COS for the first 5 galaxies of the CLASSY survey. The
complete table covering all CLASSY galaxies is available on the CLASSY MAST Webpage.

Ion J0021+0052 J0036-3333 J0127-0619 J0144+0453 J0337-0502

Si II λ1265.00 5.84 ± 2.07 144.04 ± 18.59 ··· 0.98 ± 0.28 ···
Si IV λ1393.76 ··· ··· ··· ··· ···
O IV λ1401.16 ··· ··· ··· ··· < 1.31
Si IV λ1402.77 ··· ··· ··· ··· 1.23 ± 0.39
O IV λ1404.81 ··· ··· ··· < 0.42 2.18 ± 0.53
S IV λ1406.02 ··· ··· ··· ··· ···
O IV λ1407.38 ··· < 23.98 ··· ··· 0.98 ± 0.32
O IV λ1410.00 ··· 33.76 ± 12.23 ··· ··· ···
S IV λ1416.89 ··· ··· ··· ··· ···
S IV λ1423.85 ··· ··· ··· ··· < 0.60
N IV] λ1483.33 ··· ··· ··· ··· 0.89 ± 0.34
N IV] λ1486.50 ··· ··· ··· ··· ···
Si II* λ1533.43 3.53 ± 1.22 21.59 ± 6.90 33.66 ± 9.75 < 0.80 ···
C IV λ1548.19 ··· ··· ··· < 0.45 15.99 ± 2.20
C IV λ1550.77 ··· ··· ··· < 0.51 6.86 ± 1.26
He II λ1640.42 ··· ··· < 51.99 1.27 ± 0.37 18.30 ± 3.17
O III] λ1660.81 ··· ··· ··· < 1.03 5.53 ± 1.67
O III] λ1666.15 ··· < 24.27 161.28 ± 35.14 2.56 ± 0.61 13.21 ± 2.70
[Si III] λ1883.00 ··· ··· ··· ··· ···
Si III] λ1892.03 < 4.24 ··· < 182.77 ··· ···
C III] λ1906.68 ··· ··· 1185.31 ± 146.28 8.61 ± 2.66 14.73 ± 3.40
[C III] λ1908.72 < 4.04 < 105.54 773.70 ± 68.09 6.22 ± 1.69 10.46 ± 1.55
EW(C IV) ··· ··· ··· ··· 2.13 ± 0.24
EW(He II) ··· ··· ··· 0.63 ± 0.18 1.97 ± 0.34
EW(O III]λ1666) ··· ··· 1.23 ± 0.27 1.33 ± 0.31 1.49 ± 0.31
EW(C III) ··· ··· 20.58 ± 1.68 12.44 ± 2.61 4.02 ± 0.60
N OTE—The fluxes are expressed in 10−15 erg/s/cm2 and are corrected for the dust attenuation using the
Reddy et al. (2016) law. Fluxes for undetected lines are given as less than their 3σ upper limits. The
EWs are in Å.

C. OPTICAL EMISSION LINES FITS

In Table 5 we report the fluxes of the fitted optical lines (expressed in 10−15 erg/s/cm2 ) and the Hα equivalent widths (in Å)
for the first 5 galaxies of the CLASSY sample. The complete table covering all CLASSY galaxies is available on the CLASSY
MAST Webpage. The fluxes for undetected lines are given as less than their 3σ upper limits. We also report the Telescope and
instrument taken into account for each galaxy, and the UV-optical flux offsets calculated and used to correct the optical line fluxes
(see Sec. A). The E(B −V ) taken into account (see Sec. 4.1 and Sec. 5.1) is reported in Table 4.

D. INTERSTELLAR MEDIUM PROPERTIES

In Table 6 we report the densities, temperatures, ionization parameters of the low, high and intermediate ionization zones, as
well as the E(B −V ) used to correct the optical and UV emission lines, and optical and UV metallicities for the first 5 galaxies of
the CLASSY sample.
 CLASSY IV 45

Table 5. Optical emission line fluxes for the first 5 galaxies of the CLASSY survey. The complete table
covering all CLASSY galaxies is available on the CLASSY MAST Webpage.

Ion J0021+0052 J0036-3333 J0127-0619 J0144+0453 J0337-0502

[O II] λ3727.092 14.55 ± 0.45 ··· ··· ··· ···
[O II] λ3729.875 16.39 ± 0.05 ··· ··· ··· ···
[S II] λ4069.749 0.52 ± 0.05 ··· ··· ··· ···
Hδ λ4102.892 5.20 ± 0.16 ··· ··· ··· ···
Hγ λ4341.691 9.64 ± 0.19 ··· 155.11 ± 2.04 ··· ···
[O III] λ4364.436 0.50 ± 0.10 ··· ··· ··· ···
[Fe III] λ4659.35 0.32 ± 0.04 0.61 ± 0.00 9.78 ± 0.60 ··· 0.10 ± 0.00
He II λ4687.015 0.34 ± 0.04 0.25 ± 0.01 4.94 ± 0.54 < 0.40 1.42 ± 0.00
[Fe III] λ4702.85 < 0.12 0.12 ± 0.00 4.80 ± 0.48 ··· 0.03 ± 0.00
[Ar IV] λ4712.69 ··· 0.02 ± 0.01 ··· ··· 0.41 ± 0.00
[Ar IV] λ4741.49 ··· ··· ··· ··· 0.34 ± 0.00
Hβ λ4862.691 22.71 ± 0.30 30.04 ± 0.01 423.38 ± 3.33 < 5.22 38.64 ± 0.02
[O III] λ5008.240 110.54 ± 1.00 94.62 ± 0.02 1553.65 ± 4.44 26.42 ± 5.62 133.51 ± 0.04
[Cl III] λ5519.25 ··· 0.12 ± 0.00 1.86 ± 0.28 ··· 0.03 ± 0.00
[Cl III] λ5539.43 < 0.09 0.15 ± 0.00 < 1.39 ··· 0.01 ± 0.00
[N II] λ5756.240 ··· 0.69 ± 0.00 ··· ··· ···
[O I] λ6302.046 1.13 ± 0.06 2.70 ± 0.00 18.77 ± 0.67 ··· 0.28 ± 0.00
[S II] λ6313.8 0.29 ± 0.05 0.37 ± 0.00 9.66 ± 1.13 ··· 0.27 ± 0.00
Hα λ6564.632 71.82 ± 0.75 125.23 ± 0.03 1934.23 ± 19.79 50.76 ± 2.07 89.08 ± 0.03
[N II] λ6585.271 3.76 ± 0.11 21.81 ± 0.01 125.97 ± 3.15 ··· 0.44 ± 0.00
[S II] λ6718.294 4.13 ± 0.10 13.85 ± 0.01 108.59 ± 0.77 2.24 ± 0.15 0.79 ± 0.00
[S II] λ6732.674 2.98 ± 0.09 9.05 ± 0.01 99.22 ± 0.71 1.50 ± 0.32 0.64 ± 0.00
[O II] λ7322.01 0.58 ± 0.06 1.18 ± 0.00 24.29 ± 1.51 ··· 0.22 ± 0.00
[O II] λ7332.0 0.45 ± 0.07 0.90 ± 0.00 20.00 ± 1.48 ··· 0.17 ± 0.00
[Ar III] λ7137.8 1.96 ± 0.07 3.77 ± 0.00 61.35 ± 1.29 ··· 0.81 ± 0.00
[S III] λ9071.1 ··· 9.64 ± 0.01 ··· ··· 1.28 ± 0.00
EW(Hα) 385.94 ± 4.06 125.95 ± 0.03 512.35 ± 5.24 0.00 ± 0.00 611.40 ± 0.22
Instrument/Telescope APO/SDSS VLT/MUSE VLT/VIMOS MMT VLT/MUSE
UV-Opt offset 0.69 ± 0.05 0.71 ± 0.01 0.05 ± 0.01 0.73 ± 0.02 1.16 ± 0.01
N OTE—The fluxes are expressed in 10−15 erg/s/cm2 corrected for dust attenuation using the Cardelli et al.
(1989) law, but not multiplied yet by the UV-optical offset, which is also reported. Fluxes for undetected
lines are given as less than their 3σ upper limits. The EW(Hα) is expressed in Å. The last line of each
block indicates from which instrument and telescope the data are taken.
46 M INGOZZI ET AL .

Table 6. ISM properties for the first 5 CLASSY galaxies. The complete table covering all CLASSY galaxies is
available on the CLASSY MAST Webpage.

J0021+0052 J0036-3333 J0127-0619 J0144+0453 J0337-0502

Low Ionization Zone
ne ([S II],[N II]) 21 ± 2 ··· 488 ± 1 ··· 155 ± 1
ne ([O II],[N II]) ··· ··· ··· ··· ···
ne ([S II],[S II]) ··· ··· 394 ± 1 ··· 155 ± 1
ne ([O II],[S II]) ··· ··· ··· ··· ···
ne ([S II],[O II]) 39 ± 2 ··· 394 ± 1 ··· 155 ± 1
ne ([O II],[O II]) ··· ··· ··· ··· ···
Te ([S II],[N II]) 23763 ± 140 ··· ··· ··· ···
Te ([O II],[N II]) ··· ··· ··· ··· ···
Te ([S II],[S II]) ··· ··· 15854 ± 149 ··· ···
Te ([O II],[S II]) ··· ··· ··· ··· ···
Te ([S II],[O II]) 12123 ± 28 ··· 15854 ± 149 ··· ···
Te ([O II],[O II]) ··· ··· ··· ··· ···
Te,low−G92 11610 ± 11610 ··· 15854 ± 15854 ··· ···
log(U)(S3S2) ··· −3.22 ± 0.00 ··· ··· −2.65 ± 0.00
Intermediate Ionization Zone
ne ([Cl III],[S III]) ··· 4119 ± 12 272 ± 2019 ··· ···
ne ([Fe III],[S III]) 31819 ± 461332 ··· ··· ··· 2440 ± 507
ne ([Si III],[S III]) ··· ··· ··· ··· ···
ne ([C III],[S III]) ··· ··· ··· 3308 ± 4532 4258 ± 1067
Te ([Cl III],[S III]) ··· 11961 ± 1 15854 ± 149 ··· ···
Te ([Fe III],[S III]) 11610 ± 142 ··· ··· ··· 21466 ± 8
Te ([Si III],[S III]) ··· ··· ··· ··· ···
Te ([C III],[S III]) ··· ··· 15854 ± 149 ··· 21850 ± 9
Te,int−G92 11610 ± 142 ··· 15854 ± 149 ··· ···
log(U)(O3O2) −2.90 ± 0.01 −3.08 ± 0.00 −3.29 ± 0.02 −2.99 ± 0.08 −2.38 ± 0.00
High Ionization Zone
ne ([S II],[O III]) 39 ± 1 ··· 415 ± 1 ··· 155 ± 1
ne ([S II],[O III]UV ) ··· ··· 645 ± 2 ··· 180 ± 1
ne ([N IV],[O III]) ··· ··· ··· ··· ···
ne ([N IV],[O III]UV ) ··· ··· ··· ··· ···
ne ([Ar IV],[O III]) ··· ··· ··· ··· 1107 ± 5
ne ([Ar IV],[O III]UV ) ··· ··· ··· ··· 1003 ± 7
Te ([S II],[O III]) 12300 ± 12 ··· ··· ··· ···
Te ([S II],[O III]UV ) ··· ··· 55513 ± 849 23172 ± 148 19330 ± 63
Te ([N IV],[O III]) ··· ··· ··· ··· ···
Te ([N IV],[O III]UV ) ··· ··· ··· ··· 19330 ± 62
Te ([Ar IV],[O III]) ··· ··· ··· ··· ···
Te ([Ar IV],[O III]UV ) ··· ··· ··· ··· 19323 ± 63
log(U)(Ar4Ar3) −2.54 ± 0.03 −3.06 ± 0.12 ··· ··· −1.57 ± 0.01
E(B-V)(100 cm−3 ; 104 K) 0.12 ± 0.02 0.38 ± 0.00 0.47 ± 0.01 0.04 ± 0.03 0.05 ± 0.01
E(B-V) 0.14 ± 0.01 0.38 ± 0.00 0.50 ± 0.01 0.04 ± 0.01 0.05 ± 0.00
(12+log(O/H))([S II],[O III]UV ) ··· ··· ··· ··· 7.37 ± 0.01
(12+log(O/H))([Ar IV],[O III]) ··· ··· ··· ··· ···
(12+log(O/H))([Ar IV],[O III]UV ) ··· ··· ··· ··· 7.37 ± 0.01
(12+log(O/H))([N IV],[O III]) ··· ··· ··· ··· ···
(12+log(O/H))([N IV],[O III]UV ) ··· ··· ··· ··· ···
(12+log(O/H))He2 ··· 8.47 8.53 8.05 7.24
(12+log(O/H))Si3 ··· ··· ··· ··· ···
N OTE—ne , Te and log(U) derived for the different ionization zones, where the first and second species listed in paren-
thesis indicate the diagnostics used to calculate ne and Te , respectively (see Sec. 4 for details). Te,low−G92 and Te, int−G92
are the low and intermediate ionization temperatures derived with the Garnett (1992) relations, as explained in
Sec. 4.3. E(B − V ) derived assuming ne =100 cm−3 and Te = 104 K, E(B − V ) used to correct the optical and UV
emission lines, and the different estimates of 12+log(O/H).