Draft version January 19, 2024

Typeset using LATEX twocolumn style in AASTeX631

Lyα profile shape as an escape-fraction diagnostic at high redshift

1, 2, ∗
Anthony Pahl, Alice Shapley,2 Charles C. Steidel,3 Naveen A. Reddy,4 Yuguang Chen,5 and
Gwen C. Rudie1
1 The Observatories of the Carnegie Institution for Science, 813 Santa Barbara Street, Pasadena, CA 91101, USA
2 Department of Physics and Astronomy, University of California, Los Angeles, CA 90095, USA
3 Cahill Center for Astronomy and Astrophysics, California Institute of Technology, MC249-17, Pasadena, CA 91125, USA
4 Department of Physics and Astronomy, University of California Riverside, Riverside, CA 92521, USA
arXiv:2401.09526v1 [astro-ph.GA] 17 Jan 2024

5 Department of Physics and Astronomy, University of California Davis, 1 Shields Avenue, Davis, CA 95616, USA

ABSTRACT
While the shape of the Lyα profile is viewed as one of the best tracers of ionizing-photon escape
fraction (fesc ) within low redshift (z ∼ 0.3) surveys of the Lyman continuum, this connection remains
untested at high redshift. Here, we combine deep, rest-UV Keck/LRIS spectra of 80 objects from
the Keck Lyman Continuum Spectroscopic Survey with rest-optical Keck/MOSFIRE spectroscopy in
order to examine potential correlations between Lyα profile shape and the escape of ionizing radiation
within z ∼ 3 star-forming galaxies. We measure the velocity separation between double-peaked Lyα
emission structure (vsep ), between red-side Lyα emission peaks and systemic (vLyα,red ), and between
red-side emission peaks and low-ionization interstellar absorption lines (vLyα−LIS ). We find that the
IGM-corrected ratio of ionizing to non-ionizing flux density is significantly higher in KLCS objects
with lower vLyα,red . We find no significant trend between measures of ionizing-photon escape and
vLyα−LIS . We compare our results to measurements of z ∼ 0.3 “Green Peas” from the literature and
find that KLCS objects have larger vsep at fixed vLyα,red , larger fesc at fixed vLyα,red , and higher
vLyα,red overall than z ∼ 0.3 analogs. We conclude that the Lyα profile shapes of our high-redshift
sources are fundamentally different, and that measurements of profile shape such as vLyα,red map on
to fesc in different ways. We caution against building reionization-era fesc diagnostics based purely on
Lyα profiles of low-redshift dwarf galaxies. Tracing vsep , vLyα,red , and fesc in a larger sample of z ∼ 3
galaxies will reveal how these variables may be connected for galaxies at the epoch of reionization.

Keywords: Galaxy evolution (594), High-redshift galaxies (734), Lyman-alpha galaxies (978), Reion-
ization (1383), Optical astronomy (1776), Near infrared astronomy (1093)

1. INTRODUCTION tively small number densities during reionization (Inoue

Cosmic reionization represents a significant milestone et al. 2006; Jiang et al. 2016; Parsa et al. 2018), but
in the evolution of the Universe, in which Hydrogen in which population of star-forming galaxies are most crit-
the intergalactic medium (IGM) transitioned from neu- ical to reionization, and the detailed physics of ionizing
tral to ionized. Based on the emergent UV spectrum photon production and escape, remain unsettled ques-
of distant quasars, reionization is thought to end at tions (e.g., Finkelstein et al. 2019; Naidu et al. 2020).
z ∼ 5.5 − 6 (Fan et al. 2006; Becker et al. 2021). While In order to constrain the evolving ionizing emissivity
there are multiple constraints on the timeline of reion- in the Universe, we can parameterize the quantity us-
ization, the relative contribution of ionizing light from ing three variables: the cosmic star-formation rate den-
astronomical sources, and thus the chief drivers of this sity (commonly measured as the UV luminosity density
process, remain in debate. Some works have shown that ρUV ), the ionizing photon production efficiency ξion , and
star-forming galaxies provide the majority of the ioniz- the escape fraction of ionizing photons fesc (Robertson
ing emissivity at z ≳ 6, considering quasars have rela- et al. 2015). With the advent of the James Webb Space
Telescope (JWST ), high-quality near-IR observations
have been acquired for a large number of reionization-
∗ Carnegie Fellow era galaxies, which has led to improved constraints on
2 Pahl et al.

both ρUV (e.g., Finkelstein & Bagley 2022; Donnan et al. correlated with fesc . This correlation is the tightest ob-
2023; Harikane et al. 2023) and ξion (e.g., Tang et al. served between galaxy properties (e.g., O32 , M∗ ) and
2023) out to z ∼ 9 − 12. The gradual decline in ρUV and fesc (Izotov et al. 2018b, 2021). In models where neutral
the elevated ξion at higher redshifts seen in early JWST gas is distributed in a spherical shell around a galaxy,
surveys appear to indicate that sufficient ionizing pho- shells with lower column densities result in smaller peak
tons are produced in star-forming galaxies to drive reion- separation of Lyα, explaining the connection to fesc
ization, but only those that escape their source galaxies (Verhamme et al. 2015; Orlitová et al. 2018). The dis-
will ultimately affect the ionization state of the IGM. tribution of neutral gas around star-forming galaxies is
Determining fesc directly from photometric or spectro- likely more complex and clumpy, with some regions be-
scopic surveys is virtually impossible for reionization-era ing entirely opaque and some and transparent to LyC
sources, even with JWST, as the transmission of Ly- photons (Rudie et al. 2013; Reddy et al. 2016b; Stei-
man continuum (LyC) photons through the marginally- del et al. 2018). A complete theoretical and empirical
neutral IGM drops off precipitously past z ∼ 4 (Vanzella understanding of the neutral-phase ISM and CGM, par-
et al. 2012). Thus, significant observational effort has ticularly how they affect emergent Lyα and LyC across
been expended to understand the average ionizing sig- redshift, remains incomplete.
nal from different populations of galaxies at lower red- Using the promising results found at z ∼ 0.3, infer-
shifts, and uncovering indirect tracers of fesc for use ences have been made about escape fractions of galaxies
at z ≳ 6 when reionization is in progress. These ef- at z ≳ 2 with high-resolution spectroscopy of Lyα. In
forts include surveys at z ∼ 0.3 using the Cosmic Naidu et al. (2022), the authors used the Lyα profile
Origins Spectrograph on the Hubble Space Telescope shapes of a sample of 35 z ∼ 2 Lyα emitters (LAEs)
(HST /COS), where observations of LyC-leaking, com- observed with X-SHOOTER to infer their escape frac-
pact, star-forming galaxies (“Green Peas”) act as pur- tions. In Matthee et al. (2022), the authors extended
ported analogs to high-redshift sources (e.g., Izotov et al. their framework to construct a model of reionization
2016b, 2018b, 2021). Further success has been found by that progresses rapidly and ends late, and is primar-
including extreme optical emission-line ratios or blue UV ily driven by LAEs. These two works are built upon the
slopes as selection criteria, thus probing a wider range connection between Lyα profile shape and LyC escape,
of parameter space (Flury et al. 2022a). Closer to the which remains untested beyond z ∼ 0.3. The z ∼ 2
epoch of reionization at z ∼ 3 − 4, the transmission of galaxies within the X-SHOOTER survey did not have
the LyC through single IGM sightlines is inherently un- LyC measurements necessary to directly constrain fesc ,
certain, making fesc constraints for individual galaxies and thus definitively establish if fesc -vsep relationships
indeterminate. Thus, statistical samples must be uti- are invariant with redshift.
lized, as the average transmission across many sight- The Keck Lyman Continuum Survey (KLCS) repre-
lines can be precisely constrained. Both photometric sents an ideal opportunity for directly constraining the
surveys of LyC, such as those using HST /WFC3 (e.g., connection between fesc and Lyα profile shape at the
Fletcher et al. 2019; Begley et al. 2022), and those us- highest redshifts possible (Steidel et al. 2018). The spec-
ing deep, rest-UV spectroscopic measurements (Marchi troscopic sample, which covers the LyC wavelength re-
et al. 2017, 2018; Steidel et al. 2018; Pahl et al. 2021) gion, is both deep and has been cleaned from foreground
have presented a consensus: ∼ L∗ star-forming galaxies contamination (Pahl et al. 2021), which is a significant
have an average fesc of 5 − 10% at z ∼ 3 − 4. concern plaguing LyC surveys at z ∼ 2 − 4 (Vanzella
Within these surveys, the Lyα spectral feature ap- et al. 2012). Given the relatively low resolution of the
pears to be a remarkably successful indicator of escap- Keck/LRIS spectra in the KLCS (R ∼ 1300), double
ing ionizing light. Direct correlations have been found peaked profiles are not reliably recovered at low velocity
between fesc and the equivalent width of Lyα at z ∼ 3 separations. Other probes of Lyα kinematics are more
(Marchi et al. 2018; Steidel et al. 2018; Pahl et al. 2021). robust to lower resolution, such as the peak velocity off-
In “down the barrel” observations, both LyC and the res- set from systemic, and are measurable for a significant
onant Lyα photons will be attenuated by neutral-phase subset of the KLCS. This measurement is correlated
gas in the interstellar and circumgalactic medium (ISM with vsep in low-redshift samples and is more readily
and CGM, respectively), intertwining the fate of these observed at z ≳ 6 when the blue side of Lyα is entirely
photons. At lower redshift, the shape of the emergent attenuated by the neutral-phase IGM. Other spectral
Lyα profile has been demonstrated to strongly predict features that trace the velocity structure of the ISM and
LyC leakage. When Lyα features double-peaked struc- CGM of the galaxies, such as low-ionization metal ab-
ture, the velocity separation of these peaks is inversely sorption lines, are also recoverable given the depth of
 Lyα profile shape and fesc 3

the Keck/LRIS observations. A detailed analysis of the and describe our measurements that quantify kinematic
shape of Lyα profiles in the KLCS is therefore of great information contained within the Lyα profiles of KLCS
value to understanding whether the mechanisms of LyC Lyα emitters.
escape are consistent across redshift, and for testing how
the shape of Lyα correlates with fesc at high redshift be- 2.1. Uncontaminated KLCS
fore Lyα-based fesc diagnostics are applied to galaxies in The KLCS is a deep, spectroscopic exploration of
the epoch of reionization. This work serves as an exten- the escaping ionizing signal from a sample of 136 star-
sion to the discussion of Lyα emission-line kinematics in forming, Lyman break galaxies (LBGs) at z ∼ 3. The
the KLCS presented in Steidel et al. (2018), and features survey was conducted using the Low Resolution Imag-
analyses of Lyα profiles of individual galaxies, explicit ing Spectrometer (LRIS, Oke et al. 1995; Steidel et al.
stacks as a function of Lyα profile shape, and compari- 2004) on the Keck I telescope from 2006 to 2008. Each of
son to recent low- and high-redshift surveys of galaxies the objects in the nine survey fields were observed for an
with Lyα in emission. average of 9 hours in order to sensitively probe the LyC
The paper is organized as follows. In Section 2, we de- spectral region. After removing objects with clear spec-
scribe the KLCS sample and follow-up observations, lay troscopic blends and instrumental defects, the final sam-
out the methodology for making quantitative measure- ple of 124 galaxies was presented in Steidel et al. (2018).
ments of Lyα profile shapes, and review our spectral- Included in this parent sample were 15 galaxies individu-
stacking and fitting techniques. In Section 3, we present ally detected in LyC, defined as f900 > 3σ900 , where f900
the Lyα measurements of the KLCS and the modeled is the average flux density 880 ≤ λ0 /Å ≤ 910 and σ900 is
fesc of composite spectra binned across such measure- the f900 measurement uncertainty. Measurements of the
ments. In Section 4, we discuss the applicability of our LyC at high redshift require careful consideration of con-
results for reionization-era observations, and compare to tamination from foreground sources, which can signifi-
findings to those at low redshift and the fesc assumptions cantly bias average LyC fluxes. High-resolution imaging
that have been applied to z ∼ 2 and beyond. In Section is necessary for resolving and analyzing foreground ob-
5, we summarize the main results of this work. jects (Vanzella et al. 2012; Mostardi et al. 2015). The
Throughout this paper, we adopt a standard ΛCDM KLCS was the subject of a follow up by the Hubble Space
cosmology with Ωm = 0.3, ΩΛ = 0.7 and H0 = 70 Telescope (HST ) in order to explore this contamination.
km s−1 Mpc−1 . The fesc values reported in this paper Using V606 J125 H160 imaging, we selected contaminates
are absolute escape fractions, equivalent to fesc,abs in on the basis of their colors and removed them from the
Steidel et al. (2018), and defined as the fraction of all sample, reducing the sample size to 120 (with 13 individ-
H-ionizing photons produced within a galaxy that es- ually detected in LyC) and reporting a revised average
capes into the IGM. We also employ the AB magnitude fesc of 0.06 ± 0.01 for the KLCS (Pahl et al. 2021).
system (Oke & Gunn 1983). In addition to deep, rest-UV LRIS spectroscopy and
HST imaging, a wealth of ancillary data is available for
2. SAMPLE AND METHODOLOGY the objects in the KLCS. Both ground-based and space-
In order to understand the connection between Lyα based photometry were analyzed in tandem with spec-
profile shape and escaping ionizing radiation from galax- tral stacking in order to explore fesc and galaxy prop-
ies at high redshift, we require both spectroscopy of Lyα erties in Pahl et al. (2023). These properties, including
with sufficient resolution to discern its structure and stellar mass (M∗ ), star-formation rate (SFR), E(B-V),
measurements of the Lyman Continuum (LyC) region and stellar age, were measured using stellar-population
to constrain fesc . These analyses must be applied to synthesis (SPS) models for 96 KLCS objects. We recov-
statistical samples with adequate numbers of galaxies ered inverse trends between fesc and M∗ as well as fesc
that allow for binning as a function of a given mea- and E(B-V), and no significant trends between fesc and
surement of the Lyα profile, given the uncertain cor- stellar mass or specific star-formation rate (sSFR).
rection of the IGM on individual LyC flux density mea-
surements. The galaxies in the KLCS have deep rest- 2.2. Redshift measurements and velocity analysis
UV spectroscopy that covers both Lyα and the LyC, as Of the galaxies within the uncontaminated KLCS, 80
well as low-ionization interstellar absorption that probes have Lyα profiles such that a redshift can be estimated
outflowing gas in the ISM+CGM, and rest-optical spec- from at least one peak of emission. Here, we analyze
troscopy covering nebular emission lines useful for de- the number and position of peaks within the Lyα pro-
termining systemic redshifts. In this section, we sum- files of these 80 objects. We quantify the position of the
marize the observations available for the KLCS sample peaks relative to systemic redshifts, which are based on
4 Pahl et al.

rest-optical nebular emission lines, and low-ionization dashed line and zLyα,blue as a blue, dashed line. We note
interstellar absorption redshifts. The velocity separa- that we measured only one Lyα redshift for the major-
tions that we endeavor to measure, which include the ity (65/80) of the KLCS objects, considering these had
Lyα peak separation (vsep ), the separation between the no resolved double-peaked structure. We define these
red-side Lyα peak and the systemic redshift (vLyα,red ), singular redshifts as zLyα,red for simplicity, but add no
and the separation between the red-side Lyα peak and requirement that these peaks be necessarily redshifted.
the low-ionization interstellar redshift (vLyα−LIS ), are We estimated the errors on both zLyα,red and zLyα,blue
illustrated for a KLCS galaxy in Figure 1 as a demon- measurements in a similar fashion. After fitting a Gaus-
stration of our methodology. sian to a given profile to determine the window in which
the peak flux would be computed, we perturbed the ob-
2.2.1. Lyα emission peaks served spectrum according to the error spectrum. Each
The velocity separation of the blue- and red-side flux density value at a given wavelength was randomly
peaks of Lyα emission (vsep ) has been demonstrated to modulated according to a Gaussian with zero mean and
strongly correlate with fesc at low redshift (Verhamme width equal to the corresponding error spectrum value.
et al. 2015; Izotov et al. 2018b), indicating that vsep From this perturbed spectrum, we recorded the wave-
increases with optical depth in Lyα, H i column den- length corresponding to the maximum flux within 1σ of
sity, and/or neutral-gas covering fraction. Of the 80 the central wavelength of the initial best-fit Gaussian.
KLCS objects with Lyα emission, 24 have visual evi- Each spectrum was perturbed 100 times in this man-
dence of double-peak structure within their 1D spectra. ner, resulting in 100 zLyα,red (zLyα,blue ) measurements
These Lyα profiles were typically red-side dominated: for each object. The final zLyα,red (zLyα,blue ) estimate
we show the spectrum of a KLCS object (Q0100-C16) was taken as the mean of these 100 values, with the
with double-peaked structure as an example in Figure error on zLyα,red (zLyα,blue ) taken as the standard devi-
1. Of these 24 objects, we identified 15 with sufficiently- ation of the 100 values. For the four objects in which
resolved blue-side peaks, such that the wavelength of λLyα,red was estimated from the best-fit Gaussian alone,
peak emission could be estimated for both sides of the we performed a similar procedure, but instead re-fit the
bimodal emission profile. In order to make an estimate Gaussian using the perturbed spectrum, resulting in 100
the peak wavelength for the blue-side emission, we first values for zLyα,red estimated from the central wavelength
fit a Gaussian profile and defined a window that lies of each best-fit Gaussian.
within 1σ of the central wavelength. We then deter- 2.2.2. Systemic redshifts
mined λLyα,blue , which corresponded to the wavelength
in which maximum flux occurred within this window. We determined the systemic redshift (zsys ) from rest-
This procedure ensured that any asymmetries within the optical nebular emission lines. From our parent sample
blue-side profile were accounted for. We defined the red- of 80 objects with Lyα emission, 45 objects have either
shift of the peak of blue-side Lyα emission as: [O iii]λλ4959, 5007, [O ii]λλ3726, 3729 or Hβ emission in
existing Keck/MOSFIRE spectroscopy. We fit Gaussian
λLyα,blue profiles to these lines, and used the central wavelength
zLyα,blue = − 1. (1)
1215.67 of well-fit profiles to calculate the object’s systemic red-
shift. If multiple of these lines were recovered in the
We determined the redshift corresponding to the peak
rest-optical spectrum, we estimated zsys by averaging
of red-side Lyα emission in an identical manner to that
the redshifts implied by each line measurement. The
of the blue side, save for four objects. For these four pro-
nebular redshift for Q1009-C16 is presented as a dotted,
files, the Lyα emission was either broad or noisy, and did
purple line in Figure 1. Given the strength and nar-
not have a well-defined maximum flux density. We thus
row width of the nebular emission lines in these objects,
calculated zLyα,red using simply the central wavelength
in addition to the higher spectral resolution and SNR
of the fitted profile.
of the Keck/MOSFIRE spectra, we consider the error
We define zLyα,red as
on zsys negligible compared to redshifts measured from
λLyα,red Keck/LRIS.
zLyα,red = − 1, (2)
1215.67 2.2.3. Low-ionization absorption redshifts
where λLyα,red is either the wavelength corresponding to The wavelength position of interstellar metal absorp-
the maximum flux within 1σ of the central wavelength tion lines traces the velocity of low-ionization material
of the best-fit Gaussian, or the central wavelength itself. in the ISM/CGM, which tends to be negative (out-
In Figure 1, we display zLyα,red for Q0100-C16 as a red, flowing) at these redshifts (Shapley et al. 2003; Steidel
 Lyα profile shape and fesc 5

v (km s 1)
2000 1000 0 1000 2000 3000
6 Q1009-C16
zneb
5 zLy , red 1.5

OI+SiII
vLy , red zLy , blue

SiII
4

CII
vLy LIS zLIS
vsep 1.0
3
F (uJy)

2 0.5
1
Ly
0 0.0

1205 1210 1215 1220 1225 1230 1260 1280 1300 1320 1340
Rest Wavelength (Å)
Figure 1. Lyα velocity measurements for the KLCS galaxy Q1009-C16 overlaid on its Keck/LRIS spectrum. The velocities are
calculated as the difference between different redshift estimates: vLyα,red is the difference between zLyα,red and zsys , vLyα−LIS is
the difference between zLyα,red and zLIS , and vsep is the difference between zLyα,red and zLyα,blue . Left panel: The Lyα profile
of Q1009-C16. zLyα,red = 3.1662 and zLyα,blue = 3.1557 are determined from the wavelength corresponding to the maximum
flux density within the red- and blue-side Lyα profile, respectively. zsys = 3.1621 is determined from rest-optical nebular
emission lines. Right panel: Three low-ionization metal absorption lines detected in Q1009-C16, which are used to determine
zLIS = 3.1607.
et al. 2010; Jones et al. 2012; Du et al. 2018). We de- zLIS . For each object, we performed this perturbation
termined the low-ionization absorption redshift by us- 100 times, and measure 100 values for zLIS . The sub-
ing the same Keck/LRIS spectra used for Lyα anal- sequent zLIS measurement and error was determined as
ysis, as the spectral coverage extends to ∼ 1600Å in the mean and standard deviation of these 100 values.
the rest frame. The strongest low-ionization absorp-
2.2.4. Velocities
tion features in these spectra were typically Si iiλ1260,
O iλ1302+Si iiλ1304, and C iiλ1334. We estimated Different kinematic measurements are possible for
zLIS by fitting Gaussian profiles to these three lines, subsets of the KLCS, depending on which spectral fea-
and averaging the redshifts calculated from the cen- tures were detected. For the objects for which we mea-
tral wavelengths of well-constrained Gaussian fits. For sured zLyα,blue , zLyα,red , and zsys , we define the separa-
O iλ1302+Si iiλ1304, which is blended at the resolu- tion between the blue- and red-side Lyα peak as
tion of our spectra, we assumed a singular wavelength zLyα,red − zLyα,blue
of 1302.5Å. In the right panel of Figure 1, we display vsep = c × . (3)
1 + zsys
Si iiλ1260, O iλ1302+Si iiλ1304, and C iiλ1334 for
Q1009-C16. The zLIS measured from these three lines is For objects with only zLyα,blue and zLyα,red , we use an
displayed as a green, dash-dot line. The same redshift average of zLyα,blue and zLyα,red rather than zsys to de-
is also displayed in the left panel, showing its position fine relative velocities:
relative to Lyα. zLyα,red − zLyα,blue
We estimated the errors on zLIS empirically, again vsep = c × . (4)
1 + (zLyα,red + zLyα,blue )/2
by using the Keck/LRIS error spectra. We performed
Monte Carlo simulations by perturbed each flux density We defined the KLCSsep sample as the 15 objects with
value by a Gaussian with zero mean and width equal this measurement. Of these 15, only one is individually-
to the value of the error spectrum at the correspond- detected in LyC.
ing wavelength. We then re-fit Gaussian profiles to each While vsep has been demonstrated to be an fesc diag-
line using this perturbed spectrum, and re-calculated nostic in low-redshift (z ∼ 0.3) LyC surveys, its observ-
ability becomes increasingly difficult at higher redshifts
6 Pahl et al.

as the blue-side peak becomes attenuated by the neutral- of the signal from neutral hydrogen along the line of
phase IGM. We introduce vLyα,red as the velocity sepa- sight in the IGM and CGM. The transmission of LyC
ration between the red-side Lyα peak and systemic red- emission varies significantly from sightline to sightline
shift, as a vsep -like quantity more readily measured both at the redshifts of our sample, introducing large uncer-
in the KLCS and in reionization-era galaxies. We mea- tainties on individual LyC measurements (Rudie et al.
sured this quantity for the galaxies with zLyα,red and zsys 2013; Steidel et al. 2018). To circumvent this sightline
as to sightline variability, we used binned subsamples and
composite spectra that reflect average effects of IGM
zLyα,red − zsys and CGM attenuation on the LyC spectral region, as in
vLyα,red = c × , (5)
1 + zsys Steidel et al. (2018). Given the range in sample sizes of
and defined the KLCSLyα,red sample as the 45 ob- the three KLCS subsamples, we discuss the feasibility
jects with this measurement. Eight of the objects of recovering trends between fesc and Lyα profile shape
within KLCSLyα,red are LyC detections. We lack in Appendix A, considering the inherent uncertainties
Keck/MOSFIRE observations of the full KLCS, as ob- introduced by sample variance and IGM/CGM variabil-
jects at 2.75 ≲ z ≲ 2.95 lack strong nebular lines in ity. We excluded KLCSsep from any fesc analyses due
atmospheric windows. We therefore investigate the ve- to its small sample size and the large f900 uncertainties
locity separation of zLyα,red and zLIS , which contrasts associated with binned subsamples with size ≤ 8. We
the position of Lyα to the velocities of neutral ISM. We included KLCSLyα,red and KLCSLyα−LIS and considered
define this quantity as them appropriate for investigating fesc vs. vLyα,red and
fesc vs. vLyα−LIS , respectively.
zLyα,red − zLIS In order to understand how ionizing-spectral prop-
vLyα−LIS = c × , (6)
1 + (zLyα,red + zLIS )/2 erties vary with vLyα,red and vLyα−LIS , we bin the
and defined the KLCSLyα−LIS sample as the 67 objects KLCSLyα,red sample as a function of vLyα,red and the
with both Lyα emission and LIS absorption redshifts KLCSLyα−LIS sample as a function of vLyα−LIS . We
measured. This sample includes six of the 13 LyC de- create two bins for each property, with each vLyα,red
tections in the full KLCS. We overlay vsep , vLyα,red , bin containing 22 or 23 galaxies, and each vLyα−LIS bin
vLyα−LIS on the Lyα profile of Q1009-C16 in Figure 1. containing 33 or 34 galaxies1 . We additionally bin both
Due to its complete suite of redshifts measurements, this KLCSLyα,red and KLCSLyα−LIS by Wλ (Lyα).
object was included in all three KLCS subsamples. For each subsample, we generated composite spectra
We summarize the combination of detected emission representing the average spectral properties of the com-
lines required for each of the three KLCS subsam- ponent galaxies. We follow the methodology of Stei-
ples, including their resulting sample sizes, in Table del et al. (2018) (also see Pahl et al. 2021, 2022, 2023).
1. We additionally display sample statistics of the full Briefly, each individual spectrum is first normalized to
KLCS alongside the three KLCS subsamples in Figure 2. the average flux density in the non-ionizing UV spec-
We include the distributions of spectroscopic redshifts, tral region. We then computed the sigma-clipped mean
UV luminosities, and Lyα equivalent widths (Wλ (Lyα)) of the distribution of flux densities at each rest-frame
for each subsample, alongside their respective medians wavelength increment, with clipping applied at 3σ. We
and standard errors of the median. The KLCSLyα,red did not apply sigma clipping to the Lyα spectral region
and KLCSLyα−LIS samples have median Wλ (Lyα) and (1200 − 1230Å) to avoid censoring the actual line emis-
UV luminosity consistent with those of the full KLCS, sion in this region.
within one standard error. The KLCSsep sample has a We determined ⟨f900 /f1500 ⟩obs , a metric defined as the
significantly-higher median Wλ (Lyα) than that of the ratio between the mean flux densities in the Lyman con-
full KLCS, demonstrating that vsep measurements re- tinuum (LyC) region (880 − 910Å, f900 ) and the non-
quire larger equivalent widths, such that the flux blue- ionizing UV continuum (1475 − 1525Å, f1500 ), for each
ward of systemic was sufficiently detected. composite spectrum. While this ratio offers insights into
the average observed leakage of ionizing photons, it is
necessary to address the impact of reduced transmis-
sion from the IGM in the LyC region in order to under-
2.3. Sample binning and spectral modeling
1
While f900 can be measured for each object individ- We assign the extra galaxy to the bin containing objects with
ually, constraining the LyC leaking in the vicinity of larger velocity measurements.
a galaxy requires an understanding of the attenuation
 Lyα profile shape and fesc 7

Table 1. Emission line criteria for the KLCS subsamples.

N Required lines
KLCSsep 15 Lyα with double peaked structure
KLCSLyα,red 45 Lyα; [O iii]λλ4959, 5007, [O ii]λλ3726, 3729, or Hβ
KLCSLyα−LIS 69 Lyα; Si iiλ1260, O iλ1302+Si iiλ1304, or C iiλ1334

50 Full KLCS
KLCS (vLy LIS)
KLCS (vLy , red)
40 KLCS (vsep)
Sample median
30
N

20

10

0
2.8 3.0 3.2 3.4 1 2 3 0 50 100 150
zspec LUV/L*UV WLy (Å)

Figure 2. Distributions of zspec , LUV /L∗UV , and Wλ (Lyα) for the full KLCS as well as the KLCSLyα−LIS , KLCSLyα,red , and
KLCSsep subsamples. The full KLCS includes all 120 galaxies as defined in Pahl et al. (2021). The KLCSLyα−LIS sample
contains the 68 objects with zLyα,red and zLIS measurements, while KLCSLyα,red contains the 45 objects with zLyα,red and zsys
measurements. The KLCSsep sample includes the 15 objects with resolved double-peaked Lyα structure. The medians of the
four samples with respect to zspec , LUV , and Wλ (Lyα) are displayed as circles, with the respective standard errors on the
median displayed as horizontal error bars. In vertical descending order, the circles correspond to: the full KLCS, KLCSLyα−LIS ,
KLCSLyα,red , and KLCSsep . The median Wλ (Lyα) and LUV of the KLCSLyα−LIS and KLCSLyα,red subsamples are consistent
with those the full KLCS, while KLCSsep features a larger median Wλ (Lyα).

stand the emergent ionizing signal at the galaxy’s edge. does not rely on model choices, fesc retains widespread
We applied corrections to the spectra using the average use in reionization models. We used the Binary Popula-
”IGM+CGM” transmission functions sourced from Stei- tion and Spectral Synthesis stellar-population synthesis
del et al. (2018). These corrections were computed at the models (BPASS v2.2.1 Eldridge et al. 2017), combining
mean redshift specific to each composite subsample. The these models with a SMC extinction curve (Gordon et al.
basis for these corrections was grounded in the statistics 2003), and considered a range of E(B-V) values from 0.0
of H i absorption systems along QSO sightlines, as elu- to 0.6, along with a fixed metallicity of 0.07 times solar.
cidated by Rudie et al. (2012) and Rudie et al. (2013). To model the ISM, we adopt the “holes” model, assuming
After applying the correction, we recalculated the ratio that LyC light escapes through a patchy neutral-phase
of f900 to f1500 to derive ⟨f900 /f1500 ⟩out , introduced ear- gas (Zackrisson et al. 2013; Reddy et al. 2016b, 2022;
lier. This metric encapsulates the ratio that would be Steidel et al. 2018). The free parameters of the fit are
observed at 50 proper kpc from the galaxy center, as the neutral gas covering fraction fc , the column density
detailed in Steidel et al. (2018). of neutral hydrogen NHI , and the dust attenuation from
For each composite spectrum, we derive fesc by again the foreground gas E(B-V)cov (with the uncorrected por-
following the methodology originally presented in Stei- tion assumed to be dust-free). In the “holes” model, fesc
del et al. (2018). While ⟨f900 /f1500 ⟩out is a useful em- is derived from fc , where fesc = 1 − fc .
pirical metric for quantifying leaking LyC radiation and
8 Pahl et al.

In order to estimate the uncertainty in average es- ties. Meanwhile, vLyα,red,low and vLyα−LIS,low have a
cape parameters for a given set of galaxies, we must larger amount of Lyα flux observed at systemic, which
understand the level of variability induced from sample indicates objects with these properties have more re-
construction and from the errors on individual galaxy gions in their ISM and CGM with low-column-density
properties. We follow the procedure described in Pahl (≲ 1012 cm−2 ), neutral-phase gas. Double peaked struc-
et al. (2023). We randomly perturb each measurement ture is apparent in all composites. However, all com-
(vLyα,red , vLyα−LIS , Wλ (Lyα)) by a Gaussian with width posites save for vLyα,red,low have a blue-side profile that
equal to its error. We sorted objects into two bins us- is unresolved from the stronger, red-side profile. Fi-
ing the new, perturbed parameter histograms. In order nally, vLyα,red,low features stronger Lyα emission than
to incorporate sample variance, we then bootstrap re- vLyα,red,high , indicating that galaxies with Lyα profiles
sampled (with replacement) the galaxies in each binned less offset from systemic also have larger Lyα equivalent
sample. We performed this procedure 100 times, in or- widths, as both are modulated by the covering fraction
der to generate 100 sets of objects for each binned sub- of neutral gas. Consistent with this result, surveys of
sample. Following that, we generated composite spectra LAEs have been demonstrated to have lower vLyα,red
and assessed the escape of ionizing photons for each ran- (and higher Wλ (Lyα)) than typical LBGs (e.g., Trainor
dom draw, employing the previously outlined procedure et al. 2015). In contrast with the KLCS vLyα−LIS com-
in this section. The average and standard deviation of posites, surveys of LBGs have shown a direct correla-
the distributions of ⟨f900 /f1500 ⟩obs , ⟨f900 /f1500 ⟩out , and tion between vLyα−LIS and Wλ (Lyα), although across a
fesc derived from the 100 composite spectra served as larger dynamic range of vLyα−LIS than is probed by our
the fiducial value and error estimate for the respective bins (e.g., Shapley et al. 2003).
binned sample. From these composites, we conducted three assess-
ments of ionizing photon escape, as explained in Section
3. RESULTS 2.3. The first measurement, denoted as ⟨f900 /f1500 ⟩obs ,
represents the ratio of ionizing to non-ionizing flux
In order to understand how fesc depends on Lyα pro-
density observed directly in the composite derived
file shape and kinematics, we measured vsep , vLyα,red ,
from individual spectra. The second measurement,
and vLyα−LIS for different subsets of the KLCS. We dis-
⟨f900 /f1500 ⟩out , is the same ratio calculated from a com-
play these velocity measurements in the histograms in
posite spectrum that has been adjusted for the average
Figure 3. Within each histogram, the median value is
line-of-sight attenuation caused by the IGM and CGM.
displayed as a dashed, vertical line. This line represents
Lastly, fesc is a parameter estimated through stellar pop-
the bin edge from which two equally-sized samples were
ulation synthesis modeling of the entire rest-UV compos-
constructed for composite spectra generation. We also
ite. In Figure 5, we present these three assessments of
display the relationships between the velocity measure-
ionizing escape along with their respective uncertainties,
ments in the scatterplots in Figure 3. We see correlations
organized by galaxy property for each subsample. The
between vLyα,red and vsep , and no correlation between
numerical values corresponding to these results can be
vLyα−LIS and vsep although the sample of KLCS galax-
found in Table 2, which also includes the median prop-
ies with vsep measurements is too small to do formal
erties of the galaxies within each subsample.
analysis. The sample of objects with both vLyα,red and
In order to determine whether ionizing-photon escape
vLyα−LIS measurements is 36, large enough to use the
is correlated with a given kinematic measurement of
Spearman rank correlation test to quantify the relation-
Lyα, we determine significance as defined by Pahl et al.
ship between the two variables. With the null hypoth-
(2022). For two composites constructed via binning
esis being that vLyα,red and vLyα−LIS are uncorrelated,
across a galaxy property, fesc is significantly correlated
we find a correlation coefficient of 0.487 with a p-value
with this property only when
of 0.003. This result significantly rejects the null hy-
pothesis, and indicates that vLyα,red is correlated with q
vLyα−LIS . |fesc,high − fesc,low | > (σfesc,high )2 + (σfesc,low )2 , (7)
For each binned sample illustrated in Figure 3, we cre-
ated a composite in order to understand their average where fesc,highest is measured from the bin containing
ionizing and non-ionizing rest-UV spectra. To highlight objects with the largest of a binned value, and fesc,lowest
the Lyα profiles of the objects in each bin, we display is measured from the bin containing objects with the
composites generated from bins of vLyα,red and vLyα−LIS smallest values. Significance between ⟨f900 /f1500 ⟩out
in Figure 4. Both vLyα,red,high and vLyα−LIS,high fea- and ⟨f900 /f1500 ⟩obs and galaxy property are determined
ture Lyα profiles shifted to larger redshifted veloci- in an identical fashion.
 Lyα profile shape and fesc 9

vsep, med=891±136 km/s

KLCSsep
6

N
2

0 vLy , red, med=306±128 km/s

KLCSLy , red 15
600
vLy , red (km/s)

400 10

N
200 5

0 0vLy LIS, med=495±147 km/s

800 KLCSLy LIS

15
(km/s)

600
10

N
LIS

400
5
vLy

200
0
800 1000 1200 0 250 500 750 250 500 750
vsep (km/s) vLy , red (km/s) vLy LIS (km/s)

Figure 3. Corner plot of Lyα velocity measurements for the KLCS. Upper-right panels: Histograms of vsep in green, vLyα,red
in red, and vLyα−LIS in yellow. The median of each distribution is displayed as a dashed line. For vLyα,red and vLyα−LIS , this
median defines the edge of the two bins used for composite spectrum generation for vLyα,red . Medians and standard deviations
are also displayed above each histogram. Objects individually detected in LyC are overplotted as transparent, black histograms.
Lower-left panels: Relationships between vsep , vLyα,red , and vLyα−LIS . Only objects with both measurements are displayed.
Typical errors are shown in black in the lower right of each panel. We find a significant (p∼ 0.001) correlation between vLyα−LIS
and vLyα,red . We also see a correlation between vLyα,red and vsep , but the sample size is insufficient for formal analysis.

Both fesc and ⟨f900 /f1500 ⟩out appear to decrease at upper right panel. We see a significant difference in
greater vLyα,red in the upper-left panel of Figure 5. Ac- both ⟨f900 /f1500 ⟩out and fesc (1σ and 2σ, respectively)
cording to Equation 7, the difference in ⟨f900 /f1500 ⟩out between the two bins of Wλ (Lyα) in the KLCSLyα,red
measurements between the two bins of vLyα,red is signif- subsample, reflecting the correlation of the two param-
icant (1σ). The difference between fesc measurements, eters in the full KLCS (Steidel et al. 2018; Pahl et al.
however, does not meet our significance threshold. For a 2021). Despite any differences in the ionizing spectra
given composite, ⟨f900 /f1500 ⟩out may not precisely trace across bins of vLyα,red , it is clear that any anti correla-
fesc , as fesc depends on stellar population modeling and tion between fesc and vLyα,red is less significant than that
fits to the entire rest-UV spectrum, including Lyman- between fesc and Wλ (Lyα). Nonetheless, the fact that
series absorption lines. The trends with vLyα,red can be significant correlations are found when binning across
compared to correlations between ⟨f900 /f1500 ⟩out , fesc , vLyα,red and Wλ (Lyα) is not surprising considering both
and Wλ (Lyα) within the same subsample, shown in the lower vLyα,red and higher Wλ (Lyα) probe conditions in
10 Pahl et al.

12.5 terial. Additionally, the metal-enriched gas responsible

vLy , red, low for LIS absorption may have a different velocity distribu-
10.0 vLy , red, high
Relative Intensity

tion and/or covering fraction than that of the gas that is

responsible for the attenuation of Lyα and LyC photons.
7.5 Studies have shown that covering fractions derived from
5.0 metal lines are systematically lower than those deduced
from H i absorption lines (e.g., Reddy et al. 2016b; Gaza-
2.5 gnes et al. 2018; Reddy et al. 2022). We note that the
bins in the KLCSLyα−LIS sample (lower panels) contain

Ly
0.0 ∼ 34 galaxies each, while the bins in the KLCSLyα,red
1205 1210 1215 1220 1225 sample (upper panels) contain ∼ 23 galaxies. This dif-
Rest Wavelength (Å) ference introduces larger uncertainties in fesc for the
12.5 composites generated from bins of vLyα,red , both from
vLy LIS, low sample variance and errors on assumed IGM transmis-
10.0 vLy
Relative Intensity

LIS, high sion fractions. However, vLyα−LIS measurements have

typically-larger uncertainties than vLyα,red (79 km s−1
7.5
and 39 km s−1 , respectively). Thus, through our boot-
5.0 strapping procedure, more galaxies near the vLyα−LIS
bin edge will move across bins when generating per-
2.5 turbed samples, which will result in larger error bars
in fesc .
Ly

0.0
1205 1210 1215 1220 1225
Rest Wavelength (Å)
4. DISCUSSION
Figure 4. Composite KLCS spectra binned as a function While measurements of the profile shape of Lyα have
of different spectral parameters. The composite constructed
been demonstrated to correlate with fesc at low redshift
from objects with lower values is displayed as a black curve,
while the objects with the highest values are averaged into (e.g., Izotov et al. 2018b), these results must be validated
a composite shown with an orange curve. The upper panel at high redshift before extrapolating to the reionization
features two bins of vLyα,red , with the lower bin containing era, and must be considered in context of what is feasibly
22 galaxies while the upper bin contained 23. The lower observable at z ≳ 5.5 when the IGM is significantly more
panel is instead binned by vLyα−LIS , with the lower bin con- neutral. Here, we discuss our results in context of those
taining 33 galaxies and the upper bin containing 34. Both from low-redshift (z ∼ 0.3) surveys, and attempt to con-
vLyα,red,low and vLyα−LIS,low feature more flux at systemic
nect our results to observations of the earliest galaxies,
than their counterparts, and the vLyα,red,low composite has
a larger Wλ (Lyα) than the vLyα,red,high composite.
including recent results from JWST.
4.1. Comparison to low redshift surveys
the ISM and CGM that are conducive to ionizing-photon The connections between fesc and galaxy properties at
escape: lower covering fractions of neutral gas, which at- z ∼ 3 provide key insights into the physics of ionizing-
tenuates and resonantly-scatters Lyα. photon escape, and help ground assumptions of fesc for
In contrast to vLyα,red , in the the lower right panel of reionization-era galaxies. We found that galaxies with
Figure 5, we do not see a significant difference in fesc , red-side peaks more offset from systemic (thus, higher
⟨f900 /f1500 ⟩out or ⟨f900 /f1500 ⟩obs across the two bins vLyα,red ) did not have significantly-different fesc values
of vLyα−LIS . This lack of significant difference can be than those with lower vLyα,red . We do find a signif-
compared to the significant difference in fesc measured icant difference in the IGM-corrected ionizing to non-
across two bins of Wλ (Lyα) in the KLCSLyα−LIS sample, ionizing flux ratio ⟨f900 /f1500 ⟩out across bins of vLyα,red ,
shown in the lower right panel. This result indicates that which is less model dependent measurement of the es-
vLyα−LIS may not be a robust proxy for fesc at z ∼ 3. cape of ionizing radiation. Conversely, we find galax-
While vLyα,red represents the amount that the peak of ies with larger velocity differences between the peak
Lyα emission is offset from systemic due to resonant of Lyα emission and low-ionization absorption absorp-
scattering from neutral-phase gas, vLyα−LIS includes the tion (thus, higher vLyα−LIS ) do not have significantly-
velocity offset from low-ionization absorption, which is different fesc or ⟨f900 /f1500 ⟩out values than those with
sensitive to the velocity of the (typically) outflowing ma- lower vLyα−LIS .
 Lyα profile shape and fesc 11

0.175 0.175
< f900/f1500 > obs
0.150 < f900/f1500 > out 0.150
0.125 fesc 0.125
0.100 0.100

< f900/f1500 >

0.075 0.075

fesc
0.050 0.050
0.025 0.025
0.000 0.000
0.025 0.025
300 400 0 10 20 30
vLy , red (km/s) WLy (Å)
0.175 0.175
0.150 0.150
0.125 0.125
0.100 0.100
< f900/f1500 >

0.075 0.075

fesc
0.050 0.050
0.025 0.025
0.000 0.000
0.025 0.025
400 500 600 0 10
vLy (km/s)
LIS WLy (Å)

Figure 5. Different measurements of ionizing photon escape estimated for binned subsamples of the KLCS. Each measurement
is displayed with respect to the median property of the objects contained in each bin. The blue circles are ⟨f900 /f1500 ⟩obs
measurements, which are computed directly from the composite spectra. The purple stars are ⟨f900 /f1500 ⟩out measurements,
which include a correction to f900 from attenuation from the IGM and outer CGM. Estimates of fesc of composite spectra are
displayed as gold squares, which rely on stellar-population synthesis modeling and ISM geometry featuring “holes” free from
neutral gas and dust (Steidel et al. 2018; Reddy et al. 2016b, 2022). The blue circles and purple stars are slightly shifted left
and right, respectively, for visual clarity. The upper two panels are both constructed from the same parent sample, which are
all KLCS objects with both Lyα and systemic velocity measurements. The lower two panels are constructed from a sample of
galaxies with Lyα emission and LIS absorption velocity measurements. We measure a significant difference in ⟨f900 /f1500 ⟩out
across the two bins of vLyα,red , while we find no significant difference in either ⟨f900 /f1500 ⟩out or fesc across bins of vLyα−LIS .

In the z ∼ 0.3 Universe, LyC radiation from star- ies have larger vLyα,red and vsep values. We note that
forming galaxies is observable using the HST -COS spec- Lyα photons adopt the velocities of neutral Hydrogen in
trograph and is fully unobscured thanks fewer absorbers the CGM in order to escape their source galaxy (Stei-
in the IGM (e.g., Weymann et al. 1998). Searches for del et al. 2010), and the amplitudes of typical out-
LyC leakers have revealed an anti-correlation between flow velocities in the CGM are mass dependent, such
fesc and Lyα peak separation, vsep , often presented as that more massive galaxies tend to have faster outflows
the strongest tracer of fesc within local reionization-era (Weiner et al. 2009; Martin et al. 2012; Chisholm et al.
analogs (Izotov et al. 2016b, 2018b, 2021; Flury et al. 2015). Given that the Green Peas from the literature
2022a,b). Considering the apparent correlation observed are roughly an order of magnitude lower in M∗ than
between vsep and vLyα,red in the KLCS, it is reason- our KLCS LBGs, it is expected that our velocity mea-
able that we found a significant (1σ) correlation between surements are higher overall. We perform a orthogonal
⟨f900 /f1500 ⟩out and vLyα,red at z ∼ 3. We contrast the distance regression on the z ∼ 0.3 data using the python
vsep -vLyα,red relationship in the KLCS and that from package scipy and find the relation
z ∼ 0.3 Green Peas in literature in the left panel of
Figure 6. A correlation between vsep and vLyα,red is
found in both redshift samples, although KLCS galax- vsep = 1.133 × vLyα,red + 164.5 km s−1 . (8)
12 Pahl et al.

Table 2. Measurements of ionizing-photon escape for the KLCSLyα,red and KLCSLyα−LIS samples, binned according Lyα strength
and kinematics.

vLyα,red,med a,b vLyα−LIS,med a,b WLyα,med a LUV,med / < f900 / < f900 /
N fesc a
(km s−1 ) (km s−1 ) (Å) L∗UV a f1500 >obs a f1500 >out a

KLCSLyα,red

all 45 323 ± 13 — 9.8 ± 3.7 0.93 ± 0.09 0.022 ± 0.007 0.061 ± 0.021 0.06 ± 0.01
vLyα,red,low 22 242 ± 14 — 21.0 ± 6.8 0.84 ± 0.12 0.029 ± 0.014 0.090 ± 0.045 0.08 ± 0.03
vLyα,red,high 23 388 ± 19 — 4.4 ± 3.0 1.11 ± 0.16 0.014 ± 0.011 0.036 ± 0.028 0.05 ± 0.02
WLyα,low 22 362 ± 34 — 1.2 ± 1.2 1.29 ± 0.10 0.013 ± 0.011 0.035 ± 0.029 0.03 ± 0.01
WLyα,high 23 255 ± 29 — 27.7 ± 3.5 0.70 ± 0.06 0.033 ± 0.013 0.098 ± 0.042 0.12 ± 0.03

KLCSLyα−LIS

all 68 — 533 ± 23 4.1 ± 2.0 1.03 ± 0.06 0.015 ± 0.006 0.042 ± 0.017 0.06 ± 0.01
vLyα−LIS,low 33 — 407 ± 21 4.1 ± 4.2 1.07 ± 0.10 0.015 ± 0.011 0.041 ± 0.031 0.06 ± 0.02
vLyα−LIS,high 34 — 571 ± 16 3.6 ± 2.1 1.03 ± 0.13 0.010 ± 0.009 0.025 ± 0.024 0.05 ± 0.01
WLyα,low 33 — 498 ± 22 0.6 ± 0.9 1.25 ± 0.15 0.015 ± 0.011 0.039 ± 0.028 0.04 ± 0.01
WLyα,high 34 — 481 ± 44 15.9 ± 4.0 0.99 ± 0.08 0.010 ± 0.007 0.026 ± 0.020 0.08 ± 0.02
a The medians and standard error of the subsamples with respect to a given galaxy property, with the binned parameter
highlighted in bold.
b Medians were excluded if measurements weren’t available for all objects within a subsample.

We display this best-fit relation and the corresponding of sufficient size to precisely constrain fesc as a function
95% confidence region as a solid and dashed black lines, of vsep at z ∼ 3. In the left panel of Figure 7, we show
respectively, in the left panel of Figure 6. By comparing the fesc -vsep relation reported in Izotov et al. (2016b)
to the average trend at low redshift, we see a offset to alongside published z ∼ 0.3 leakers from the literature,
larger vsep values at fixed vLyα,red at z ∼ 3 when com- including z ∼ 0.3 Green Peas without LyC observations
pared to the lower-redshift data. Stated in a different from Yang et al. (2017). We translate this trendline into
way, this difference indicates that our sample has Lyα vLyα,red space by using Equation 8; the translated rela-
profiles with larger blue-side velocity offsets (i.e., more tion is displayed as a black, solid line in the middle panel
negative) at fixed red-side velocity offset when compared of Figure 7. At z ∼ 0.3, fesc is much more tightly corre-
to low-redshift Green Peas. This difference appears to lated with vsep than with vLyα,red . This result appears
indicate that the shape of Lyα profiles may not be struc- to indicate that vLyα,red is a substandard indicator for
tured in the same way across redshift. Unfortunately, fesc as compared to vsep . We also find that vLyα,red is a
this comparison can only be made for a highly limited substandard indicator of fesc as compared to Wλ (Lyα)
subsample within the KLCS: this figure contains the in Figure 5, although correlation is still shown between
seven galaxies with double-peaked Lyα emission profiles ⟨f900 /f1500 ⟩out and vLyα,red . At vLyα,red values where
sufficiently resolved in the R ∼ 800 − 1400 spectra and direct comparison can be made across redshift, we find
have follow up observations of rest-optical nebular lines. higher fesc values in the KLCS at fixed vLyα,red when
Given the difference in spectral resolution between the compared to z ∼ 0.3 leakers. Critically, it appears that
two samples, we discuss the effects of spectral resolution relationships between fesc and Lyα profile shape may
on our velocity measurements in Appendix B. A larger not be invariant across redshift. The KLCSLyα,red sam-
sample of resolved blue Lyα peaks for KLCS objects ple has a median vLyα,red = 323 km s−1 , which would
would allow for a more robust test of this evolution. correspond to vsep = 576 km s−1 according to Equation
We expand upon potential differences between low- 8. Using the fesc -vsep relation fit in Izotov et al. (2018b),
and high-redshift Lyα profiles by displaying low-redshift this vsep value would correspond to fesc ∼ 0.01, sig-
data for both fesc vs. vsep and fesc vs. vLyα,red in Figure nificantly different from the full-sample measurement of
7, alongside our results found in bins of vLyα,red . Con- fesc = 0.06±0.01. At z ∼ 3, it appears that fesc is higher
sidering our small sample size of objects with vsep mea- for a given vsep than at z ∼ 0.3. We also note that the
surements in the KLCS, we were unable to make bins while the z ∼ 0.3 samples are lower in stellar mass, they
 Lyα profile shape and fesc 13

KLCS
1600 z 0.3, known leakage 1600
z 0.3, unknown leakage
1400 1400
1200 1200
vsep (km/s)

vsep (km/s)
1000 1000
800 800
600 600
400 400
200 200

200 400 600 0 200 400 600

vLy , red (km/s) |vLy , blue|(km/s)

Figure 6. Comparison between the the velocity separation between the peaks of Lyα emission (vsep ) and the velocity offset of
different peaks of Lyα emission and systemic, both at low redshift and this work. Measurements of galaxies at z ∼ 0.3, primarily
Green Peas, are sourced from Yang et al. (2017); Gazagnes et al. (2020); and Flury et al. (2022a). Measurements presented
in Gazagnes et al. (2020) include objects originally published in Pettini & Pagel (2004); Izotov et al. (2011, 2016b,a, 2018b,a);
Leitherer et al. (2016); and Chisholm et al. (2017). Left panel: Comparison between vsep and the velocity offset of the red side
of Lyα emission and systemic (vLyα,red ). Both KLCS and z ∼ 0.3 samples show a positive correlation between vsep and vLyα,red ,
while KLCS objects show higher vsep and vLyα,red measurements on average than the z ∼ 0.3 objects. A linear relationship fit
to the z ∼ 0.3 data is shown as a black, dotted line. The 95% confidence region for this fit is shown with dashed, black lines.
Right panel: Comparison between vsep and the velocity offset of the blue side of Lyα emission and systemic (vLyα,blue ). A
linear relationship fit to the z ∼ 0.3 data is shown as a black, solid line. The 95% confidence region for this fit is shown with
dashed, black lines. The z ∼ 0.3 vLyα,blue -vsep relation is tighter than that of vLyα,red -vsep .

are also generally lower in UV luminosity and higher in includes prescriptions for binary stars, yields larger val-
SFR as compared to the KLCS, which would not neces- ues of intrinsic f900 than Starburst99, thus our KLCS
sarily suppress fesc in these samples (Pahl et al. 2023). galaxies will have comparatively-lower fesc values from
Furthermore, as discussed in Flury et al. (2022b), KLCS this model choice alone (Chisholm et al. 2019). Treat-
composites have higher fesc at fixed Wλ (Lyα) than the ment of dust within spectral fits similarly affects fesc
z ∼ 0.3 objects within the LzLCS survey. We therefore estimates. In our fesc modeling discussed in Section 2.3,
exercise caution in using fesc determinations at high red- we assumed an SMC dust attenuation curve, and that
shift (i.e., z > 2) built directly from Lyα properties of the ISM within KLCS galaxies include “holes” free of
low (z ∼ 0.3) redshift analogs, discussed further in Sec- both neutral gas and dust. The z ∼ 0.3 fesc estimates
tion 4.3. discussed in this section instead assume a Reddy et al.
When comparing fesc estimates across different works, (2016a) dust attenuation curve, and that the dust is sit-
it is necessary to examine any differences in underlying uated in a uniform screen. If these model choices were
model assumptions. All fesc estimates displayed in Fig- applied to our analysis of KLCS objects, our reported
ure 7 assume an intrinsic ionizing flux based on stellar fesc values would be elevated (Steidel et al. 2018). Con-
population fits to a galaxy’s spectrum, which is subse- sidering that our model choices only shift our fesc esti-
quently compared to the observed ionizing spectrum in mates to comparatively-lower values, we conclude that
order to compute fesc . Choice of model SED necessar- the increased fesc at fixed Lyα profile shape seen within
ily changes the estimation of fesc . In this work, we used the KLCS is not driven by inconsistent methodology.
BPASS SPS models, in contrast to the Starburst99 mod-
els assumed by the low-redshift LyC surveys discussed
in this section (Leitherer et al. 1999). BPASS, which
14 Pahl et al.

100

10 1
fesc

10 2

Izotov+2018b (translated)
10 3
Izotov+2018b Naidu+2022 (translated)
Naidu+2022 This work Izotov+2018b (translated)
z 0.3, known leakage z 0.3, known leakage z 0.3, known leakage
200 400 600 0 100 200 300 400 0 200 400
vsep vLy , red (km/s) -vLy , blue (km/s)
Figure 7. Left panel: Comparison between fesc and vsep within z ∼ 0.3 LyC leakers from the literature. These objects are the
same as those with known leakage shown in Figure 6. The best-fit inverse power law from Izotov et al. (2018b) is displayed as
a dashed, black line. We did not bin the KLCS as a function of vsep as there were too few objects with resolved, double-peaked
Lyα structure. The inferred range of fesc based on vsep from Naidu et al. (2022) is displayed as shaded, purple region. Middle
panel: Measured fesc values for bins of vLyα,red within the KLCS, alongside z ∼ 0.3 leakers from literature. We translate the
fesc -vsep trend of Izotov et al. (2018b) into a fesc -vLyα,red trend using Equation 8, which is derived from z ∼ 0.3 objects. The
95% confidence interval computed for Equation 8 is also translated and is displayed via a dashed line. We also translate the fesc
criteria of Naidu et al. (2022) into vLyα,red space using the same equation. We see a larger fesc at fixed vLyα,red within the KLCS
in comparison to literature z ∼ 0.3 LyC leakers. We also find a significant discrepancy between the fesc of the vLyα,red,low KLCS
subsample and the translated fesc criteria of Naidu et al. (2022). Right panel: fesc vs. vLyα,blue for z ∼ 0.3 LyC leakers from the
literature. The solid line is a translated version of the fesc -vsep relation in the left panel, using the linear relation displayed in
the right panel of Figure 6 and its corresponding confidence interval. vLyα,blue is a better predictor of fesc than vLyα,red within
the z ∼ 0.3 sample.

4.2. Red vs. blue Lyα peaks We show vsep vs. vLyα,blue for the KLCS and z ∼ 0.3
Considering the apparent relationship between vsep samples in the right panel of Figure 6, and see a tighter
and vLyα,red across redshift, one would expect each to correlation as compared to that of vsep and vLyα,red
modulate the escape fraction of ionizing photons in across both redshift samples. This tighter correlation
a similar way. The tenuous trend between fesc and was reported in previous analyses of z ∼ 0.3 Green Peas
vLyα,red within the z ∼ 0.3 high-redshift analogs moti- (Henry et al. 2015; Verhamme et al. 2017; Orlitová et al.
vates a closer examination of which physical properties 2018). We display the z ∼ 0.3 fesc -vLyα,blue correlation
of the neutral-phase ISM and CGM are probed by vsep from literature sources in the right panel of Figure 7,
and vLyα,red . Additionally, in the reionization era, any and see a similar correlation as that with vsep , in con-
Lyα emission bluewards of systemic will be resonantly trast to the relative lack of correlation with vLyα,red .
absorbed by optically thick neutral gas in the nearby Again, due to a lack of a large sample of KLCS galaxy
IGM (e.g., Mason & Gronke 2020), making the obser- measurements with vLyα,blue , these relationships remain
vation of double-peaked structure exceedingly difficult. untested at high redshift. However, at lower redshift it
Understanding any physical differences between vsep and is clear that the configuration of the neutral-phase ISM
vLyα,red , and analogously, vLyα,blue and vLyα,red , is of and CGM that concerns ionizing-photon escape are more
critical importance when interpreting Lyα profiles dur- tied to the observable parameters of vLyα,blue and vsep ,
ing the epoch of reionization in the context of escaping rather than vLyα,red .
ionizing radiation. Simple radiative transfer models of Lyα feature the
neutral-phase CGM distributed in a homogeneous shell
 Lyα profile shape and fesc 15

around a source. In these models, the escape fraction fer simulations of Lyα through a clumpy medium and
of ionizing radiation is related to the column density of find the physical parameters that affect the emergent
neutral hydrogen (NHI ), which is optically thin to LyC profile shape. They find that vLyα,red is sensitive to
radiation. In these models, both vsep and vLyα,red are clump covering fraction and clump radial velocity, while
modulated by NHI , although small vLyα,red is not a suf- vLyα,blue is more sensitive to the column density of the
ficient condition for identifying a LyC leaker. Lower static, hot inter-clump medium.
values of vLyα,red may also be driven by high outflow Under an assumption of a heterogeneous medium, the
velocities, rather than higher Lyα and LyC escape frac- red-side Lyα profile will not directly reflect properties
tions (Verhamme et al. 2015). Our measurements of of the foreground gas. Lyα photons redward of systemic
vLyα−LIS contrast the position of the red-side peak of primarily take paths that backscatter off of outflowing
Lyα to that of low-ionization metal absorption lines, material behind the galaxy, while those blueward of sys-
which are produced in the intervening material in the temic are scattered off resonance by gas (e.g., Steidel
ISM/CGM and directly encode their kinematics. We et al. 2010). This fact alone may explain why fesc is less
find a lack of correlation between fesc and vLyα−LIS sensitive to vLyα,red than vsep or vLyα,blue , as gas in the
within the KLCS, which supports the idea that greater background does not attenuate LyC photons traveling
outflow velocities do not significantly affect measured in the line-of-sight direction. Considering that viewing
fesc values. More systemic redshift measurements are angle is expected to strongly affect emergent Lyα pro-
needed to directly test the correlations between fesc and files (e.g., Blaizot et al. 2023), it stands to reason that
LIS velocities within the KLCS. Nonetheless, LIS lines sample-averaged statistics may smooth out these varia-
likely do encode information about the covering fraction tions, and vLyα,red is still predictive of average ionizing-
of neutral material and thus fesc : Saldana-Lopez et al. photon escape, as in the KLCS.
(2022) have presented empirical calibrations that use the
depth of LIS lines and dust attenuation measurements 4.3. The fesc assumptions of Naidu et al. (2022)
to predict the escape fractions of LzLCS galaxies (see
The primary goal of this work was to test the connec-
also Reddy et al. 2016b).
tion between fesc and Lyα profile shape in the z ∼ 3
It’s likely that the multiphase ISM in z ∼ 3 galaxies
Universe, and whether trends found at lower redshift
is more complex and dynamically perturbed than homo-
hold. While the KLCS does not currently have the
geneous shells. Our modeling of composite spectra as-
spectral resolution to discern double-peak structure for
sumes that clear channels in an optically-thick neutral-
the majority of the sample, detailed spectral analyses of
phase medium allow Lyα and LyC photons to escape
higher-resolution Lyα profiles at intermediate redshifts
unabated. This scenario is supported observationally
have been performed, like those of the XLS-z2 survey
by the connections between Lyα equivalent width, LIS
(Matthee et al. 2021). We can use the Lyα profiles of
absorption strength, and E(B-V) found in z ∼ 2 − 5
the KLCS, which have the added benefit of direct LyC
galaxy surveys (e.g. Shapley et al. 2003; Jones et al.
measurements, to add indirect fesc constraints to the
2012; Du et al. 2018; Pahl et al. 2021). Indeed, in Kaki-
objects in this survey.
ichi & Gronke (2021), radiation hydrodynamics simu-
The XLS-z2 survey features 35 z ∼ 2 LAEs, selected
lations indicate that galaxies with highly-asymmetric
primarily via narrow-band imaging and targeted with
Lyα profiles, such as those in the KLCS, feature both
the X-SHOOTER spectrograph on VLT with spectral
optically-thick and thin channels that govern the escape
resolution of R ∼ 4000. Systemic redshifts were also
of LyC photons. These types of profiles are ubiquitous
recovered for 33/35 of the sources. Using criteria empir-
at high redshift (Steidel et al. 2010; Kulas et al. 2012;
ically motivated from the Lyα profiles of LyC leakers,
Hashimoto et al. 2015; Trainor et al. 2015). Modeling
Naidu et al. (2022) presented a framework for predicting
pursued in Orlitová et al. (2018) notes the puzzling dif-
fesc using the Lyα profiles of XLS-z2 objects alone. The
ferences between vLyα,red and vLyα,blue in the context of
resulting criteria were based on measurements from LyC
poor fits when assuming a homogeneous, expanding shell
leakers at z ∼ 0.3, as well as a few individually-leaking
of neutral Hydrogen. They find that by including sys-
objects found at z ∼ 2 − 4 (although individual LyC
temic redshifts in their fits, the spherical-shell fit fails,
detections at high redshift have highly-uncertain fesc ).
as objects with weak blue-side Lyα peaks do not have
The criteria were two-fold: objects with low vsep or high
symmetric Lyα profiles across systemic. Other studies
flux densities at systemic (fcen ) were assumed to have
of the radiative transfer of Lyα explore the differences
> 20% fesc , while objects with high vsep and low fcen
between vLyα,red and vLyα,blue . In Gronke & Dijkstra
were assumed to have < 5% fesc . The vsep criteria are
(2016), the authors utilize Monte-Carlo radiative trans-
displayed in shaded regions in the left panel of Figure 7.
16 Pahl et al.

By introducing criteria based on vsep , the sample also that, at fixed vLyα,red (and implied vsep ), fesc is higher
demonstrated a connection between assumed fesc and at z ∼ 3 than at z ∼ 0.3. Nonetheless, vLyα,red may be
vLyα,red . Composite spectra of the low- and high-fesc the best proxy of vsep actually observable for galaxies in
XLS-z2 galaxies featured different vLyα,red : those with the reionization era. In simulations of Lyα transmission
assumed fesc > 20% had vLyα,red = 106±3 km s−1 , while fractions at z ∼ 6, typical sightlines through the neutral-
those with fesc < 5% had vLyα,red = 254 ± 4 km s−1 . phase IGM feature feature complete attenuation of Lyα
We attempt to test the fesc assumptions of Naidu flux below 1216Å (Laursen et al. 2011).
et al. (2022) directly with the KLCS by translating Despite the difficulty of observing blue-side Lyα flux
the vsep -based fesc criteria into one based on vLyα,red . in the epoch of reionization, a handful of z ≳ 6 galax-
Using Equation 8 and assuming a direct relationship ies have been observed with this signature (Hu et al.
between the two variables, the “high escape” objects 2016; Songaila et al. 2018; Bosman et al. 2020). The
would have vLyα,red < 89 km s−1 , while the “low es- mere observability of blue-side Lyα indicates that these
cape” objects would have vLyα,red > 164 km s−1 . The photons are traveling through highly-transparent and
translated criteria are displayed as shaded regions in unique sightlines through the IGM, and these objects
the middle panel of Figure 7. The KLCS objects span are residing in ionized bubbles (Mason & Gronke 2020).
a dynamic range of vLyα,red that allow a comparison While such signatures are useful for examining these rare
to the fesc < 5% region: the vLyα,red,low composite objects and probing the topology of reionization, the
has fesc = 0.08 ± 0.03, which lies outside of the re- vast majority of galaxies will not have observed blue-
gion. The “low escape” spectral stack of Naidu et al. side Lyα flux.
(2022) can also be directly compared to this vLyα,red,low In contrast, the red-side peak of Lyα and associated
KLCS stack. The X-SHOOTER composite spectrum nebular emission lines required to constrain zsys , and
has vLyα,red = 254 ± 4 km s−1 , Wλ (Lyα) = 61 ± 3Å, and thus vLyα,red , are observable at extreme redshifts. A lim-
an assumed fesc < 5%. In comparison, the vLyα,red,low ited number of vLyα,red measurements have been made
stack has a median vLyα,red = 242 ± 14 km s−1 and for z > 6 galaxies using C iii] as a probe of zsys (Stark
Wλ (Lyα) = 21 ± 7. The vLyα,red,low stack, despite hav- et al. 2015, 2017; Hutchison et al. 2019). More recently,
ing lower Wλ (Lyα) than that of the “low escape” stack of surveys targeting [C ii] and [O iii] with ALMA have
Naidu et al. (2022), has a significantly larger fesc value found further success (e.g., Cassata et al. 2020; Ends-
than that assumed when using the Lyα-based criteria in ley et al. 2022). With the advent of JWST/NIRSpec,
Naidu et al. (2022). It is likely that fesc measured for the rest-optical nebular emission lines are more readily re-
vLyα,red,low stack would be even higher if the Wλ (Lyα) coverable at z > 6, and samples with vLyα,red measure-
were comparable to the “low escape” stack of Naidu et al. ments are growing (Tang et al. 2023). These objects
(2022), considering the direct connection between fesc have vLyα,red between ∼ 100 km s−1 and ∼ 800 km s−1 ,
and Wλ (Lyα) ubiquitous in z ∼ 2 − 4 LyC samples with Wλ (Lyα) < 100Å and −23 <MUV < −19. Our fesc
(Steidel et al. 2018; Marchi et al. 2018; Begley et al. estimates as a function of both vLyα,red and Wλ (Lyα)
2022). While we cannot directly test the fesc assump- represent the first comparison sample to these objects
tions of Naidu et al. (2022) without higher-resolution drawn from a z ∼ 3 sample. Based on Lyα alone,
spectra of the KLCS, these discrepancies indicate that objects such as REBELS-23 (vLyα,red = 227 km s−1 ,
fesc assumptions based on Lyα profile shape may not be Wλ (Lyα) = 14Å; Endsley et al. 2022) may have fesc ∼
consistent across all samples, especially across redshift 0.08, considering its similarity to our vLyα,red,low sub-
and/or galaxy properties such as stellar mass, UV lumi- sample.
nosity and SFR. From the comparison presented in this Direct comparison to Lyα profiles at z > 6 has some
section, it appears that fesc values inferred from low-z caveats. At the neutral fractions of the IGM during
scaling relations will be underestimated. reionization, even the blue side of the red Lyα peak be-
gins to be attenuated, which is more prominent closer
4.4. Lyα profiles during reionization to systemic (Laursen et al. 2011). This attenuation will
bias vLyα,red to higher values as the Lyα profiles become
The connections between fesc and galaxy properties
asymmetrically modulated. Additionally, Lyα profiles
at z ∼ 3 can indicate the most appropriate assumptions
that have strong emission significantly redward of sys-
for fesc at even higher redshifts, during the epoch of
temic will be preferentially observed, preventing an un-
reionization. We find that galaxies with red-side peaks
biased view into the average velocity offsets of different
more offset from systemic (thus, higher vLyα,red ) tend
populations of reionization-era galaxies. Despite these
to have lower ⟨f900 /f1500 ⟩out values, although they did
caveats, there is promising evidence that the Lyα pro-
not have significantly-different fesc values. We also find
 Lyα profile shape and fesc 17

files of UV bright populations may not be fully atten- adopt to escape from a galaxy (e.g., Steidel et al.
uated by the neutral IGM at z > 6. The Wλ (Lyα) 2010). We contrast our fesc measurements in two
distribution in MUV < −21 galaxies appears not to de- bins of vLyα,red to that of literature LyC leakers
cline from 6 < z < 8 (Endsley et al. 2021; Jung et al. at z ∼ 0.3, and find larger fesc at fixed vLyα,red
2022; Roberts-Borsani et al. 2023), and Lyα appears to within our higher-redshift sample. From this evo-
be ubiquitous in this population at z ∼ 7.5 − 9 (e.g., lution, we infer that while not only are Lyα pro-
Stark et al. 2017; Larson et al. 2022). Thus, for lumi- files different across redshift and stellar mass, Lyα
nous galaxies in particular, the strength and shape of profile shape measurements such as vLyα,red ap-
Lyα may be a valid indicator of fesc at z ≳ 6. pear to trace escaping ionizing radiation in a dif-
ferent manner. Building on this point, we perform
5. SUMMARY a linear fit between vsep and vLyα,red , and we find
In this work, we investigate the connections between larger fesc at fixed vsep than those predicted by
Lyα profile shape and ionizing-photon escape at high z ∼ 0.3 data. We further conclude that these dif-
redshift, as a complement and comparison to existing ferences could not be caused by inconsistent model
analyses at z ∼ 0.3 (e.g., Izotov et al. 2016b, 2018b, assumptions.
2021; Flury et al. 2022a,b). To this end, we utilized
KLCS galaxies, for which R ∼ 800−1400, rest-UV spec- 3. We compare the correlations between fesc and
troscopy is available. These spectra cover both the Lyα vLyα,red , vsep , and vLyα,blue by compiling results
spectral feature and the LyC spectral region, with suf- for z ∼ 0.3 LyC leakers in literature. We find that
ficient depth to sensitively constrain fesc in composite fesc correlates much more strongly with vsep and
spectra. By making quantitative measurements of Lyα vLyα,blue than with vLyα,red at z ∼ 0.3, despite our
and binning the KLCS as a function of these proper- measured correlation between ⟨f900 /f1500 ⟩out and
ties, we explored the connections between the velocity vLyα,red within the KLCS. We infer that the het-
structure of Lyα and the escape of ionizing radiation in erogeneity of neutral-phase gas around LyC leak-
z ∼ 3 star-forming galaxies. Our main conclusions are ers, which manifests in a large viewing-angle de-
as follows: pendence on both Lyα profile shape and fesc (e.g.,
Verhamme et al. 2012; Behrens & Braun 2014;
1. We find positive correlations between the velocity Smith et al. 2019, 2022b,a; Blaizot et al. 2023), will
separation between the red-side peak of Lyα and scatter the relationship between fesc and vLyα,red
systemic redshift vLyα,red and the Lyα peak sepa- in individual objects, considering the red side of
ration vsep within the KLCS, but we lack the sam- Lyα is primarily concerned with gas in the back-
ple size to quantify the significance of this trend. ground of a galaxy. Within the KLCS, we effec-
We find a significant correlation between vLyα,red tively averaged across many viewing angles by cre-
and and velocity separation between the red-side ating composite spectra of subsamples binned as
peak of Lyα and the low-ionization interstellar a function of vLyα,red , thus probing the average
redshift vLyα−LIS . When comparing our measure- neutral-gas configuration of galaxies within each
ments to samples of z ∼ 0.3 Green Peas from the stack.
literature, we find an evolution in the vsep -vLyα,red
relationship with redshift. Galaxies at z ∼ 3 from
the KLCS have larger vsep at fixed vLyα,red , indi- 4. We compare our low-vLyα,red bin to that of the
cating that Lyα profile shapes potentially evolve “low escape” stack of the XLS-z2 survey presented
with redshift and/or stellar mass. in Naidu et al. (2022), in order to test fesc as-
sumptions primarily calibrated using local sam-
2. We find a significant (1σ), negative correlation ples. Our KLCS composite constructed from ob-
between the ionizing to non-ionizing flux ratio jects with the lowest vLyα,red measurements had
⟨f900 /f1500 ⟩out and vLyα,red within the KLCS, an estimated fesc = 0.08 ± 0.03, greater than
found across two bins of vLyα,red . We find no the assumed fesc < 0.05 in a XLS-z2 compos-
significant differences for any measure of ionizing- ite with comparable vLyα,red . While we lack the
photon escape across two bins of vLyα−LIS . The higher resolution spectra to directly test their vsep
connection between ⟨f900 /f1500 ⟩out and vLyα,red is and central-flux-based criteria, the discrepancy be-
supported by radiative transfer models that con- tween the assumed fesc at fixed vLyα,red indicates
nect increasing covering fractions of neutral gas that further investigation is needed into the Lyα
to larger velocity shifts that Lyα photons must profile shapes of LyC leakers across redshift, at
18 Pahl et al.

larger dynamic ranges of vLyα,red and vsep than is of diagnostics to predict fesc of reionization-era galaxies.
probed by the KLCS.
The authors acknowledge helpful conversations with
John Chisholm and Drew Newman during the draft-
In the reionization epoch, any Lyα flux blueward ing of this manuscript. AJP was generously supported
of systemic emitted from typical galaxies is entirely by a Carnegie Fellowship through the Carnegie Ob-
attenuated due to the neutral-phase IGM. These re- servatories while conducting part of this work. We
sults, presented as a function of red-side Lyα profile acknowledge support from NSF AAG grants 0606912,
shape, are critical for developing new Lyα-based cri- 0908805, 1313472, 2009313, 2009085, and 2009278. Sup-
teria usable for z > 6 galaxies. In addition, we’ve port for program HST-GO-15287.001 was provided by
demonstrated that aspects of the Lyα profile and it’s NASA through a grant from the Space Telescope Sci-
effects on ionizing-photon escape may evolve with red- ence Institute, which is operated by the Associations of
shift and stellar mass. These analyses, performed at Universities for Research in Astronomy, Incorporated,
the some of the highest redshift possible, provide im- under NASA contract NAS5-26555. CS was supported
portant new insights into the mechanics of ionizing- in part by the Caltech/JPL President’s and Director’s
photon escape during reionization. Larger samples program. We wish to extend special thanks to those
with high-resolution, rest-UV spectra covering Lyα, of Hawaiian ancestry on whose sacred mountain we are
coupled with direct LyC measurements, across larger privileged to be guests. Without their generous hospi-
dynamic ranges of galaxy properties are necessary to tality, most of the observations presented herein would
build on these results and construct a complete suite not have been possible.

APPENDIX

A. FEASIBILITY OF RECOVERING TRENDS BETWEEN fesc AND LYα PROFILE SHAPE FOR DIFFERENT
SAMPLE SIZES
In Pahl et al. (2022), we performed simulations to determine the minimum sample sizes required to confidently
recover trends between fesc and galaxy properties within the KLCS. We used the trend between ⟨f900 /f1500 ⟩out and
Wλ (Lyα) as a measure of the reliability of smaller samples, considering the ubiquity of trends between ionizing-photon
escape and Wλ (Lyα) found within z ∼ 3 LyC surveys (Marchi et al. 2017, 2018; Fletcher et al. 2019; Steidel et al. 2018).
⟨f900 /f1500 ⟩out is defined as the ratio between f900 and the average flux density in the non-ionizing UV continuum
(1475 − 1525Å, f1500 ) after performing an IGM+CGM transmission correction. It is a more model-independent
measurement of escaping ionizing light than fesc , since it is not based on any stellar-population-dependent assumption
of the intrinsic production rate of ionizing radiation. We performed simulations by drawing random samples from the
full KLCS of a fixed sample size, binning the objects into two Wλ (Lyα) subsamples, and determining whether the
resulting composite spectra feature a 1σ difference between their estimated ⟨f900 /f1500 ⟩out values. The Pahl et al.
(2022) simulations were performed in two different modes: one assuming preferential selection of LyC detections, and
one with entirely random selection. The preferential selection mode included all 13 KLCS galaxies individually detected
in LyC for each random draw. This mode was relevant for subset of galaxies in Pahl et al. (2022), which included
all LyC detections. For this work, we consider the preferential LyC detection mode comparable to the KLCSLyα,red
sample, considering that 8/45 objects are LyC detections, compared to 13/120 for the full KLCS. Both KLCSLyα−LIS
samples and KLCSsep samples are more comparable to the random selection mode (6/67 and 1/15 are LyC detections,
respectively).
We used the results from these simulations to determine if the sizes of each samples are adequate for investigation
of fesc and the respective binned property. In the LyC preferential-selection mode, a sample size of 45 resulted in
a recovery of the ⟨f900 /f1500 ⟩out -Wλ (Lyα) trend in 92% of realizations, indicating that KLCSLyα,red is appropriate
for determining trends between fesc and vLyα,red . In fully random mode, a sample size of 63 finds the same recovery
in 80% of realizations, while a sample size of 15 recovers the trend in 30%. We therefore disinclude KLCSsep from
analyses of ionizing photon escape.
 Lyα profile shape and fesc 19

B. EFFECTS OF SPECTRAL RESOLUTION ON VELOCITY MEASUREMENTS

Our conclusions herein are based on relatively low resolution (R ≃ 800 for LRIS-B, z < 3.11; R ≃ 1400 for LRIS-R,
z > 3.11) spectra, which results in a small subset of the KLCS with resolved blue-side Lyα peaks, and thus measurable
peak separations. We therefore relied on the velocity offset of the red-side peak of Lyα within this analysis, and in our
comparison to low-redshift data (HST-COS; R ≃ 16000 at Lyα) and the survey of LAEs of XLS-z2 discussed in Naidu
et al. (2022) (X-SHOOTER, R ≃ 4000 at Lyα). Considering the difference in spectral resolution between the KLCS
observations originally presented in Steidel et al. (2018) and these surveys, we discuss the effects of spectral resolution
on our velocity measurements.
In Verhamme et al. (2015), the authors perform simulations on theoretical Lyα profiles to understand the effects of
broadening from lower-resolution spectrographs on the shape of Lyα. In the lowest resolution (R ∼ 1000) simulation,
the blue peak of Lyα becomes a bump on the wing of the stronger, red-side component of the Lyα profile, or simply
presents an elongated blue-side tail. Similar profile shapes are observed within the KLCS. Across the different-resolution
simulations, the location of the red-side peak of Lyα is well conserved. The peak shifts only slightly to lower values
(by < 50 km s−1 ), indicating that any bias in vLyα,red within our sample is minimal.
We can directly quantify the effects of spectral resolution on the position of the red-side peak of Lyα for a subset
of the KLCS. In order to resolve additional double-peaked structure of KLCS galaxies with Lyα in emission, we
have obtained a limited amount of follow up Keck/LRIS spectroscopy with the 1200-line grating on the red side
of the dichroic, featuring a spectral resolution of R ∼ 2500. While the sample of KLCS objects with vsep remains
modest, we measured the Lyα red-side peak offsets from systemic, thus vLyα,red , for 11 KLCS objects with this higher-
resolution data following the procedures presented in Section 2.2. We calculate the velocity difference between vLyα,red
measured from the low-resolution LRIS data and that measured from the high-resolution data, and find a median
∆v= +56 km s−1 with a standard deviation of 63 km s−1 , where the high-resolution data typically features smaller
vLyα,red measurements. This difference represents a small shift compared to the median vLyα,red of our sample of
319 km s−1 , and is comparable to the typical errors on the vLyα,red measurements of 39 km s−1 . Additionally, our
results are presented using bins of vLyα,red all measured from spectra of similar resolution, thus, our result of ionizing-
photon escape correlations with vLyα,red will not be affected by any potential biases from low spectral resolution.
Nonetheless, it is worth noting potential discrepancies when comparing the vLyα,red values of this work and others
with higher spectral resolution.

REFERENCES
Becker, G. D., D’Aloisio, A., Christenson, H. M., et al. Chisholm, J., Orlitová, I., Schaerer, D., et al. 2017,
2021, Monthly Notices of the Royal Astronomical Society, Astronomy & Astrophysics, Volume 605, id.A67, 19,
508, 1853, doi: 10.1093/mnras/stab2696 doi: 10.1051/0004-6361/201730610
Begley, R., Cullen, F., McLure, R. J., et al. 2022, Monthly Chisholm, J., Rigby, J. R., Bayliss, M., et al. 2019, The
Notices of the Royal Astronomical Society, 513, 3510, Astrophysical Journal, 882, 182,
doi: 10.1093/mnras/stac1067 doi: 10.3847/1538-4357/ab3104
Behrens, C., & Braun, H. 2014, Astronomy and Chisholm, J., Tremonti, C. A., Leitherer, C., et al. 2015,
Astrophysics, 572, A74, The Astrophysical Journal, 811, 149,
doi: 10.1051/0004-6361/201424755 doi: 10.1088/0004-637X/811/2/149
Blaizot, J., Garel, T., Verhamme, A., et al. 2023, Monthly Donnan, C. T., McLeod, D. J., Dunlop, J. S., et al. 2023,
Notices of the Royal Astronomical Society, 523, 3749, Monthly Notices of the Royal Astronomical Society, 518,
doi: 10.1093/mnras/stad1523 6011, doi: 10.1093/mnras/stac3472
Bosman, S. E. I., Kakiichi, K., Meyer, R. A., et al. 2020, Du, X., Shapley, A. E., Reddy, N. A., et al. 2018, The
The Astrophysical Journal, 896, 49, Astrophysical Journal, 860, 75,
doi: 10.3847/1538-4357/ab85cd doi: 10.3847/1538-4357/aabfcf
Cassata, P., Morselli, L., Faisst, A., et al. 2020, Astronomy Eldridge, J. J., Stanway, E. R., Xiao, L., et al. 2017,
and Astrophysics, 643, A6, Publications of the Astronomical Society of Australia, 34,
doi: 10.1051/0004-6361/202037517 61, doi: 10.1017/pasa.2017.51
20 Pahl et al.

Endsley, R., Stark, D. P., Charlot, S., et al. 2021, Monthly Inoue, A. K., Iwata, I., & Deharveng, J. M. 2006, Monthly
Notices of the Royal Astronomical Society, 502, 6044, Notices of the Royal Astronomical Society: Letters, 371,
doi: 10.1093/mnras/stab432 doi: 10.1111/j.1745-3933.2006.00195.x
Endsley, R., Stark, D. P., Bouwens, R. J., et al. 2022, Izotov, Y. I., Guseva, N. G., & Thuan, T. X. 2011, The
Monthly Notices of the Royal Astronomical Society, 517, Astrophysical Journal, 728, 161,
5642, doi: 10.1093/mnras/stac3064 doi: 10.1088/0004-637X/728/2/161
Fan, X., Carilli, C. L., & Keating, B. 2006, Annual Review Izotov, Y. I., Orlitová, I., Schaerer, D., et al. 2016a, Nature,
of Astronomy & Astrophysics, 44, 415, 529, 178, doi: 10.1038/nature16456
doi: 10.1146/annurev.astro.44.051905.092514 Izotov, Y. I., Schaerer, D., Thuan, T. X., et al. 2016b,
Monthly Notices of the Royal Astronomical Society, 461,
Finkelstein, S. L., & Bagley, M. B. 2022, The Astrophysical
3683, doi: 10.1093/mnras/stw1205
Journal, 938, 25, doi: 10.3847/1538-4357/ac89eb
Izotov, Y. I., Schaerer, D., Worseck, G., et al. 2018a,
Finkelstein, S. L., D’Aloisio, A., Paardekooper, J.-P., et al.
Monthly Notices of the Royal Astronomical Society, 474,
2019, The Astrophysical Journal, 879, 36,
4514, doi: 10.1093/mnras/stx3115
doi: 10.3847/1538-4357/ab1ea8
Izotov, Y. I., Worseck, G., Schaerer, D., et al. 2021,
Fletcher, T. J., Tang, M., Robertson, B. E., et al. 2019, The
Monthly Notices of the Royal Astronomical Society, 503,
Astrophysical Journal, 878, 87,
1734, doi: 10.1093/mnras/stab612
doi: 10.3847/1538-4357/ab2045
—. 2018b, Monthly Notices of the Royal Astronomical
Flury, S. R., Jaskot, A. E., Ferguson, H. C., et al. 2022a, Society, 478, 4851, doi: 10.1093/mnras/sty1378
The Astrophysical Journal Supplement Series, 260, 1, Jiang, L., McGreer, I. D., Fan, X., et al. 2016, The
doi: 10.3847/1538-4365/ac5331 Astrophysical Journal, 833, 222,
—. 2022b, The Astrophysical Journal, 930, 126, doi: 10.3847/1538-4357/833/2/222
doi: 10.3847/1538-4357/ac61e4 Jones, T., Stark, D. P., & Ellis, R. S. 2012, The
Gazagnes, S., Chisholm, J., Schaerer, D., Verhamme, A., & Astrophysical Journal, 751,
Izotov, Y. 2020, Astronomy and Astrophysics, 639, A85, doi: 10.1088/0004-637X/751/1/51
doi: 10.1051/0004-6361/202038096 Jung, I., Papovich, C., Finkelstein, S. L., et al. 2022, The
Gazagnes, S., Chisholm, J., Schaerer, D., et al. 2018, Astrophysical Journal, 933, 87,
Astronomy and Astrophysics, 616, 29, doi: 10.3847/1538-4357/ac6fe7
doi: 10.1051/0004-6361/201832759 Kakiichi, K., & Gronke, M. 2021, The Astrophysical
Gordon, K. D., Clayton, G. C., Misselt, K. A., Landolt, Journal, 908, 30, doi: 10.3847/1538-4357/abc2d9
A. U., & Wolff, M. J. 2003, The Astrophysical Journal, Kulas, K. R., Shapley, A. E., Kollmeier, J. A., et al. 2012,
594, 279, doi: 10.1086/376774 The Astrophysical Journal, 745, 33,
Gronke, M., & Dijkstra, M. 2016, The Astrophysical doi: 10.1088/0004-637X/745/1/33
Journal, 826, 14, doi: 10.3847/0004-637X/826/1/14 Larson, R. L., Finkelstein, S. L., Hutchison, T. A., et al.
2022, The Astrophysical Journal, 930, 104,
Harikane, Y., Ouchi, M., Oguri, M., et al. 2023, The
doi: 10.3847/1538-4357/ac5dbd
Astrophysical Journal Supplement Series, 265, 5,
Laursen, P., Sommer-Larsen, J., & Razoumov, A. O. 2011,
doi: 10.3847/1538-4365/acaaa9
The Astrophysical Journal, 728,
Hashimoto, T., Verhamme, A., Ouchi, M., et al. 2015, The
doi: 10.1088/0004-637X/728/1/52
Astrophysical Journal, 812, 157,
Leitherer, C., Hernandez, S., Lee, J. C., & Oey, M. S. 2016,
doi: 10.1088/0004-637X/812/2/157
The Astrophysical Journal, 823, 64,
Henry, A., Scarlata, C., Martin, C. L., & Erb, D. 2015, The doi: 10.3847/0004-637X/823/1/64
Astrophysical Journal, 809, 19, Leitherer, C., Schaerer, D., Goldader, J. D., et al. 1999,
doi: 10.1088/0004-637X/809/1/19 The Astrophysical Journal Supplement Series, 123, 3,
Hu, E. M., Cowie, L. L., Songaila, A., et al. 2016, The doi: 10.1086/313233
Astrophysical Journal Letters, 825, L7, Marchi, F., Pentericci, L., Guaita, L., et al. 2017,
doi: 10.3847/2041-8205/825/1/L7 Astronomy and Astrophysics, 601, 73,
Hutchison, T. A., Papovich, C., Finkelstein, S. L., et al. doi: 10.1051/0004-6361/201630054
2019, The Astrophysical Journal, 879, 70, —. 2018, Astronomy and Astrophysics, 614, 11,
doi: 10.3847/1538-4357/ab22a2 doi: 10.1051/0004-6361/201732133
 Lyα profile shape and fesc 21

Martin, C. L., Shapley, A. E., Coil, A. L., et al. 2012, The Reddy, N. A., Steidel, C. C., Pettini, M., Bogosavljević, M.,
Astrophysical Journal, 760, 127, & Shapley, A. E. 2016b, The Astrophysical Journal, 828,
doi: 10.1088/0004-637X/760/2/127 108, doi: 10.3847/0004-637x/828/2/108
Mason, C. A., & Gronke, M. 2020, Monthly Notices of the Reddy, N. A., Topping, M. W., Shapley, A. E., et al. 2022,
Royal Astronomical Society, 499, 1395, The Astrophysical Journal, 926, 31,
doi: 10.1093/mnras/staa2910 doi: 10.3847/1538-4357/ac3b4c
Matthee, J., Sobral, D., Hayes, M., et al. 2021, Monthly Roberts-Borsani, G., Treu, T., Mason, C., et al. 2023, The
Notices of the Royal Astronomical Society, 505, 1382, Astrophysical Journal, 948, 54,
doi: 10.1093/mnras/stab1304 doi: 10.3847/1538-4357/acc798
Robertson, B. E., Ellis, R. S., Furlanetto, S. R., & Dunlop,
Matthee, J., Naidu, R. P., Pezzulli, G., et al. 2022, Monthly
J. S. 2015, The Astrophysical Journal Letters, 802, L19,
Notices of the Royal Astronomical Society, 512, 5960,
doi: 10.1088/2041-8205/802/2/L19
doi: 10.1093/mnras/stac801
Rudie, G. C., Steidel, C. C., Shapley, A. E., & Pettini, M.
Mostardi, R. E., Shapley, A. E., Steidel, C. C., et al. 2015,
2013, The Astrophysical Journal, 769, 146,
Astrophysical Journal, 810, 107,
doi: 10.1088/0004-637X/769/2/146
doi: 10.1088/0004-637X/810/2/107
Rudie, G. C., Steidel, C. C., Trainor, R. F., et al. 2012, The
Naidu, R. P., Tacchella, S., Mason, C. A., et al. 2020, The
Astrophysical Journal, 750, 67,
Astrophysical Journal, 892, 109, doi: 10.1088/0004-637X/750/1/67
doi: 10.3847/1538-4357/ab7cc9 Saldana-Lopez, A., Schaerer, D., Chisholm, J., et al. 2022,
Naidu, R. P., Matthee, J., Oesch, P. A., et al. 2022, Astronomy and Astrophysics, 663, A59,
Monthly Notices of the Royal Astronomical Society, 510, doi: 10.1051/0004-6361/202141864
4582, doi: 10.1093/mnras/stab3601 Shapley, A. E., Steidel, C. C., Pettini, M., & Adelberger,
Oke, J. B., & Gunn, J. E. 1983, The Astrophysical Journal, K. L. 2003, The Astrophysical Journal, 588, 65,
266, 713, doi: 10.1086/160817 doi: 10.1086/373922
Oke, J. B., Cohen, J. G., Carr, M., et al. 1995, Publications Smith, A., Kannan, R., Garaldi, E., et al. 2022a, Monthly
of the Astronomical Society of the Pacific, 107, 375, Notices of the Royal Astronomical Society, 512, 3243,
doi: 10.1086/133562 doi: 10.1093/mnras/stac713
Orlitová, I., Verhamme, A., Henry, A., et al. 2018, Smith, A., Ma, X., Bromm, V., et al. 2019, Monthly
Astronomy and Astrophysics, 616, A60, Notices of the Royal Astronomical Society, 484, 39,
doi: 10.1051/0004-6361/201732478 doi: 10.1093/mnras/sty3483
Pahl, A. J., Shapley, A., Steidel, C. C., Chen, Y., & Reddy, Smith, A., Kannan, R., Tacchella, S., et al. 2022b, Monthly
N. A. 2021, Monthly Notices of the Royal Astronomical Notices of the Royal Astronomical Society, 517, 1,
Society, 505, 2447, doi: 10.1093/mnras/stab1374 doi: 10.1093/mnras/stac2641
Songaila, A., Hu, E. M., Barger, A. J., et al. 2018, The
Pahl, A. J., Shapley, A., Steidel, C. C., Reddy, N. A., &
Astrophysical Journal, 859, 91,
Chen, Y. 2022, Monthly Notices of the Royal
doi: 10.3847/1538-4357/aac021
Astronomical Society, 516, 2062,
Stark, D. P., Walth, G., Charlot, S., et al. 2015, Monthly
doi: 10.1093/mnras/stac1767
Notices of the Royal Astronomical Society, 454, 1393,
Pahl, A. J., Shapley, A., Steidel, C. C., et al. 2023, Monthly
doi: 10.1093/mnras/stv1907
Notices of the Royal Astronomical Society, 521, 3247,
Stark, D. P., Ellis, R. S., Charlot, S., et al. 2017, Monthly
doi: 10.1093/mnras/stad774
Notices of the Royal Astronomical Society, 464, 469,
Parsa, S., Dunlop, J. S., & McLure, R. J. 2018, Monthly doi: 10.1093/mnras/stw2233
Notices of the Royal Astronomical Society, 474, 2904, Steidel, C. C., Bogosavljević, M., Shapley, A. E., et al.
doi: 10.1093/mnras/stx2887 2018, The Astrophysical Journal, 869, 123,
Pettini, M., & Pagel, B. E. J. 2004, Monthly Notices of the doi: 10.3847/1538-4357/aaed28
Royal Astronomical Society, 348, L59, Steidel, C. C., Erb, D. K., Shapley, A. E., et al. 2010, The
doi: 10.1111/j.1365-2966.2004.07591.x Astrophysical Journal, 717, 289,
Reddy, N. A., Steidel, C. C., Pettini, M., & Bogosavljević, doi: 10.1088/0004-637X/717/1/289
M. 2016a, The Astrophysical Journal, 828, 107, Steidel, C. C., Shapley, A. E., Pettini, M., et al. 2004, The
doi: 10.3847/0004-637x/828/2/107 Astrophysical Journal, 604, 534, doi: 10.1086/381960
22 Pahl et al.

Tang, M., Stark, D. P., Chen, Z., et al. 2023, Verhamme, A., Orlitová, I., Schaerer, D., et al. 2017,
JWST/NIRSpec Spectroscopy of $z=7-9$ Star Forming Astronomy &amp; Astrophysics, Volume 597, id.A13, 11,
Galaxies with CEERS: New Insight into Bright doi: 10.1051/0004-6361/201629264
Ly$\alpha$ Emitters in Ionized Bubbles,
Weiner, B. J., Coil, A. L., Prochaska, J. X., et al. 2009, The
doi: 10.48550/arXiv.2301.07072
Astrophysical Journal, 692, 187,
Trainor, R. F., Steidel, C. C., Strom, A. L., & Rudie, G. C.
2015, The Astrophysical Journal, 809, 89, doi: 10.1088/0004-637X/692/1/187
doi: 10.1088/0004-637X/809/1/89 Weymann, R. J., Jannuzi, B. T., Lu, L., et al. 1998, The
Vanzella, E., Guo, Y., Giavalisco, M., et al. 2012, The Astrophysical Journal, 506, 1, doi: 10.1086/306233
Astrophysical Journal, 751, 70, Yang, H., Malhotra, S., Gronke, M., et al. 2017, The
doi: 10.1088/0004-637X/751/1/70
Astrophysical Journal, 844, 171,
Verhamme, A., Dubois, Y., Blaizot, J., et al. 2012,
doi: 10.3847/1538-4357/aa7d4d
Astronomy and Astrophysics, 546, A111,
doi: 10.1051/0004-6361/201218783 Zackrisson, E., Inoue, A. K., & Jensen, H. 2013, The
Verhamme, A., Orlitová, I., Schaerer, D., & Hayes, M. Astrophysical Journal, 777, 39,
2015, Astronomy and Astrophysics, 578, A7, doi: 10.1088/0004-637X/777/1/39
doi: 10.1051/0004-6361/201423978