COSMOS-Web: The Role of Galaxy Interactions and Disk Instabilities in Producing
Starbursts at
z
<
4
Andreas L. Faisst
1
, Lilan Yang
2
, M. Brinch
3
,
4
, C. M. Casey
3
,
5
, N. Chartab
6
, M. Dessauges-Zavadsky
7
,
N. E. Drakos
8
, S. Gillman
3
,
4
, G. Gozaliasl
9
,
10
, C. C. Hayward
11
, O. Ilbert
12
, P. Jablonka
13
, A. Kaminsky
14
,
J. S. Kartaltepe
2
, A. M. Koekemoer
15
, V. Kokorev
16
, E. Lambrides
17
,
42
, D. Liu
18
, C. Maraston
19
, C. L. Martin
20
,
A. Renzini
21
, B. E. Robertson
22
, D. B. Sanders
23
, Z. Sattari
6
,
24
, N. Scoville
25
, C. M. Urry
26
, A. P. Vijayan
3
,
4
,
J. R. Weaver
27
, H. B. Akins
5
, N. Allen
3
,
28
, R. C. Arango-Toro
12
, O. R. Cooper
5
,
43
, M. Franco
5
, F. Gentile
29
,
30
,
S. Harish
2
, M. Hirschmann
31
,
32
, A. A. Khostovan
2
,
33
, C. Laigle
34
, R. L. Larson
5
,
44
, M. Lee
3
,
4
, Z. Liu
35
,
36
,
37
,
A. S. Long
5
,
45
, G. Magdis
3
,
4
,
28
, R. Massey
38
, H. J. McCracken
34
, J. McKinney
5
, L. Paquereau
34
, J. Rhodes
39
,
R. M. Rich
40
, M. Shuntov
3
,
28
, J. D. Silverman
35
,
37
, M. Talia
29
,
30
, S. Toft
3
,
28
, and J. A. Zavala
41
1
Caltech
/
IPAC, 1200 E. California Blvd., Pasadena, CA 91125, USA;
afaisst@caltech.edu
2
Laboratory for Multiwavelength Astrophysics, School of Physics and Astronomy, Rochester Institute of Technology, 84 Lomb Memorial Dr., Rochester
,NY
14623, USA
3
Cosmic Dawn Center
(
DAWN
)
, Denmark
4
DTU Space, Technical University of Denmark, Elektrovej 327, 2800 Kgs. Lyngby, Denmark
5
The University of Texas at Austin, 2515 Speedway Blvd. Stop C1400, Austin, TX 78712, USA
6
The Observatories of the Carnegie Institution for Science, 813 Santa Barbara St., Pasadena, CA 91101, USA
7
Département d
’
Astronomie, Université de Genève, Chemin Pegasi 51, CH-1290 Versoix, Switzerland
8
Department of Physics and Astronomy, University of Hawaii, Hilo, 200 W. Kawili St., Hilo, HI 96720, USA
9
Department of Computer Science, Aalto University, P.O. Box 15400, Espoo, FI-00 076, Finland
10
Department of Physics, University of Helsinki, P.O. Box 64, FI-00014 Helsinki, Finland
11
Center for Computational Astrophysics, Flatiron Institute, 162 Fifth Ave., New York, NY 10010, USA
12
Aix Marseille Univ, CNRS, CNES, LAM, Marseille, France
13
Ecole Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland
14
Department of Physics, University of Miami, Coral Gables, FL 33124, USA
15
Space Telescope Science Institute, 3700 San Martin Dr., Baltimore, MD 21218, USA
16
Kapteyn Astronomical Institute, University of Groningen, PO Box 800, 9700 AV Groningen, The Netherlands
17
NASA-Goddard Space Flight Center, Code 662, Greenbelt, MD 20771, USA
18
Purple Mountain Observatory, Chinese Academy of Sciences, 10 Yuanhua Road, Nanjing 210023, People
’
s Republic of China
19
Institute of Cosmology and Gravitation, University of Portsmouth, Dennis Sciama Building, Burnaby Road, Portsmouth PO13FX, UK
20
Department of Physics, University of California, Santa Barbara, Santa Barbara, CA 93109, USA
21
INAF, Osservatorio Astronomico di Padova, Vicolo dell
’
Osservatorio 5, 35122, Padova, Italy
22
Department of Astronomy and Astrophysics, University of California, Santa Cruz, 1156 High St., Santa Cruz, CA 95064, USA
23
Institute for Astronomy, University of Hawai
’
i at Manoa, 2680 Woodlawn Dr., Honolulu, HI 96822, USA
24
Department of Physics and Astronomy, University of California, Riverside, 900 University Ave., Riverside, CA 92521, USA
25
Astronomy Dept., California Institute of Technology, 1200 E. California Blvd., Pasadena, CA, USA
26
Physics Department and Yale Center for Astronomy & Astrophysics, Yale University, P.O. Box 208120, New Haven, CT 06520-8120, USA
27
Department of Astronomy, University of Massachusetts, Amherst, MA 01003, USA
28
Niels Bohr Institute, University of Copenhagen, Jagtvej 128, DK-2200, Copenhagen, Denmark
29
University of Bologna
—
Department of Physics and Astronomy
“
Augusto Righi
”
(
DIFA
)
, Via Gobetti 93
/
2, I-40129 Bologna, Italy
30
INAF, Osservatorio di Astro
fi
sica e Scienza dello Spazio, Via Gobetti 93
/
3, I-40129, Bologna, Italy
31
Institute of Physics, GalSpec, Ecole Polytechnique Fédérale de Lausanne, Observatoire de Sauverny, Chemin Pegasi 51, 1290 Versoix, Switzerland
32
INAF, Astronomical Observatory of Trieste, Via Tiepolo 11, 34131 Trieste, Italy
33
Astrophysics Division, NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA
34
Institut d
’
Astrophysique de Paris, UMR 7095, CNRS, and Sorbonne Université, 98bis boulevard Arago, 75014 Paris, France
35
Kavli Institute for the Physics and Mathematics of the Universe
(
WPI
)
, The University of Tokyo, Kashiwa, Chiba 277-8583, Japan
36
Center for Data-Driven Discovery, Kavli IPMU
(
WPI
)
, UTIAS, The University of Tokyo, Kashiwa, Chiba 277-8583, Japan
37
Department of Astronomy, School of Science, The University of Tokyo, 7-3-1 Hongo, Bunkyo, Tokyo 113-0033, Japan
38
Department of Physics, Centre for Extragalactic Astronomy, Durham University, South Road, Durham DH1 3LE, UK
39
Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Dr., Pasadena, CA 91001, USA
40
Department of Physics and Astronomy, UCLA, PAB 430 Portola Plaza, Box 951547, Los Angeles, CA 90095-1547, USA
41
National Astronomical Observatory of Japan, 2-21-1 Osawa, Mitaka, Tokyo 181-8588, Japan
Received 2024 May 10; revised 2024 December 12; accepted 2024 December 18; published 2025 February 14
The Astrophysical Journal,
980:204
(
12pp
)
, 2025 February 20
https:
//
doi.org
/
10.3847
/
1538-4357
/
ada566
© 2025. The Author
(
s
)
. Published by the American Astronomical Society.
42
NPP Fellow.
43
NSF Graduate Research Fellow.
44
NSF Graduate Fellow.
45
NASA Hubble Fellow.
Original content from this work may be used under the terms
of the
Creative Commons Attribution 4.0 licence
. Any further
distribution of this work must maintain attribution to the author
(
s
)
and the title
of the work, journal citation and DOI.
1
Abstract
We study of the role of galaxy
–
galaxy interactions and disk instabilities in producing starburst activity in galaxies
out to
z
=
4. For this, we use a sample of 387 galaxies with robust total star formation rate measurements from
Herschel, gas masses from the Atacama Large Millimeter
/
submillimeter Array, stellar masses and redshifts from
multiband photometry, and JWST
/
NIRCam rest-frame optical imaging. Using mass-controlled samples, we
fi
nd
an increased fraction of interacting galaxies in the starburst regime at all redshifts out to
z
=
4. This increase
correlates with star formation ef
fi
ciency
(
SFE
)
but not with gas fraction. However, the correlation is weak
(
and
only signi
fi
cant out to
z
=
2
)
, which could be explained by the short duration of SFE increase during interaction. In
addition, we
fi
nd that isolated disk galaxies make up a signi
fi
cant fraction of the starburst population. The fraction
of such galaxies with star-forming clumps
(
“
clumpy disks
”
)
is signi
fi
cantly increased compared to the main-
sequence disk population. Furthermore, this fraction directly correlates with SFE. This is direct observational
evidence for a long-term increase of SFE maintained due to disk instabilities, contributing to the majority of
starburst galaxies in our sample and hence to substantial mass growth in these systems. This result could also be of
importance for explaining the growth of the most massive galaxies at
z
>
6.
Uni
fi
ed Astronomy Thesaurus concepts:
Starburst galaxies
(
1570
)
;
Galaxy interactions
(
600
)
;
Galaxy disks
(
589
)
1. Introduction
For most of their lives, star-forming galaxies follow the main
sequence
(
e.g., E. Daddi et al.
2007
; D. Elbaz et al.
2007
;
K. G. Noeske et al.
2007
)
, a tight relation between stellar mass
and star formation rate
(
SFR
)
set by an equilibrium state between
gas consumption, out
fl
ow, and in
fl
ow
(
R. Davé et al.
2012
;
S. J. Lilly et al.
2013
; R. Feldmann
2015
)
. The scatter of this
relation
(
∼
0.3 dex
)
is set by oscillations around that equilibrium,
driven by constant adjustment of the galaxies
’
SFR to the
available cold gas and replenishing gas in
fl
ows
(
e.g.,N.Bouché
et al.
2010
; A. Dekel et al.
2013
;S.Tacchellaetal.
2016
)
.
Stronger oscillations are observe
d mostly in low-mass galaxies at
low and local redshifts, while at higher redshifts, galaxies across
all masses seem to show such a behavior
(
e.g.,D.R.Weiszetal.
2012
; N. Emami et al.
2019
;A.L.Faisstetal.
2019
)
. In addition
to these oscillations, galaxies experience signi
fi
cant bursts of star
formation, elevating their SFRs by factors of 10 and more above
the main sequence
(
D. B. Sanders & I. F. Mirabel
1996
)
.The
fraction of starburst galaxies to th
e total population increases with
redshift from
∼
1% at
z
<
0.4
(
N. Bergvall et al.
2016
)
to
∼
2%
–
5% at
z
∼
0.5
–
1.0
(
G. Rodighiero et al.
2011
; L. Bisigello et al.
2018
)
and higher fractions at higher r
edshifts depending on stellar
mass
(
M.T.Sargentetal.
2012
;K.I.Caputietal.
2017
;L.Bis-
igello et al.
2018
)
. Interactions between gas-rich galaxies lead to
the in
fl
ow and compression of gas and are, therefore, usually the
prerequisites for starbursts
(
e.g.,D.B.Sanders&I.F.Mira-
bel
1996
;T.J.Coxetal.
2008
; R. Genzel et al.
2010
;J.S.Kart-
altepe et al.
2012
; C.-L. Hung et al.
2013
;P.F.Hopkinsetal.
2018
; A. Cibinel et al.
2019
;J.Morenoetal.
2021
;E.A.Shah
et al.
2022
)
. However, some simulations and observations suggest
that starbursts and galaxy major mergers do not necessarily have
to be causally connected
(
e.g., G. Rodighiero et al.
2011
;
S.Kavirajetal.
2013
; M. Sparre & V. Springel
2016
;J.Fensch
et al.
2017
; S. Wilkinson et al.
2022
;Y.A.Lietal.
2023
;Z.Liu
et al.
2024
)
. In addition to starbursts, disk instabilities in gas-rich
galaxies
(
A. Toomre
1964
;F.Bournaudetal.
2007
)
,whichmay
be tightly related to increased gas in
fl
ows or minor mergers
(
N. Scoville et al.
2023
)
, could increase the rate of star formation
without external inter
actions, hence contributing to the starburst
population
(
e.g.,B.Wang&J.Silk
1994
; A. Immeli et al.
2004
;
A. Dekel et al.
2009
;K.Tadakietal.
2018
)
and the buildup of
bulge-dominated galaxies later on
(
e.g.,R.Bouwensetal.
1999
)
.
The above shows that the origin of starbursts is, therefore,
complex, and the de
fi
nitive link between galaxy interactions, disk
structure, increased star formation ef
fi
ciency
(
SFE
)
, and gas
fraction out to high redshift has yet to be studied with robust
measurements and comprehensive samples.
The works by D. Liu et al.
(
2019
)
and N. Scoville et al.
(
2023
)
are likely the most comprehensive submillimeter dust-
based studies
(
in terms of sample size and redshift coverage
)
on
the relation between gas and star formation properties of main-
sequence and starburst galaxies to date. They combine robust
SFRs from Herschel, gas fraction and SFE measurements from
the Atacama Large Millimeter
/
submillimeter Array
(
ALMA
)
,
and other parameters
(
such as stellar masses
)
from compre-
hensive multiband photometry for
>
700 galaxies out to
z
=
6.
Speci
fi
cally, N. Scoville et al.
(
2023
)
found that starbursts at a
given redshift and stellar mass are primarily triggered due to
enhanced SFE rather than a high gas fraction. Similar results
have been found in independent studies by J. D. Silverman
et al.
(
2015
)
and L. J. Tacconi et al.
(
2018
)
using large-sample
CO data
(
for a review, see L. J. Tacconi et al.
2020
)
. On the
other hand, an increase in the gas fraction is likely responsible
for the maintained high speci
fi
c SFRs at high redshift
(
e.g.,
L. J. Tacconi et al.
2010
,
2018
; R. Genzel et al.
2015
; J. Fre-
undlich et al.
2019
; A. Gowardhan et al.
2019
; D. Liu et al.
2019
; M. Dessauges-Zavadsky et al.
2020
)
.
In this work, we explore the population of starburst galaxies
out to
z
=
4 in terms of their interactions, disk structure, SFE,
and gas fractions. Speci
fi
cally, we aim to study the role of
galaxy
–
galaxy interactions and disk instabilities in producing
starburst activity. To this end, we combine the N. Scoville et al.
(
2023
)
sample with high-resolution rest-frame optical
imaging from the COSMOS-Web JWST
/
NIRCam imaging
(
C. M. Casey et al.
2023
)
.
After summarizing the data
(
Section
2
)
, we detail the
morphological classi
fi
cation in Section
3
. In Section
4
,wepresent
the results, and we conclude in Section
5
. Throughout this work,
we assume a
Λ
CDM cosmology with
H
0
=
70 km s
−
1
Mpc
−
1
,
Ω
Λ
=
0.7, and
Ω
m
=
0.3, and magnitudes are given in the AB
system
(
J. B. Oke
1974
)
.WeuseaG.Chabrier
(
2003
)
initial mass
function for stellar masses and SFRs.
2. Sample and Basic Measurements
For this work, we use a sample of 387 galaxies out to
z
∼
4.
This sample is a subsample of the 704 galaxies from N. Scoville
et al.
(
2023
; see more details in earlier works; N. Scoville et al.
2014
,
2016
,
2017
)
requiring existing JWST imaging. The
2
The Astrophysical Journal,
980:204
(
12pp
)
, 2025 February 20
Faisst et al.
selected galaxy sample is best for carrying out this study, as it
includes
1. robust measurements of total SFR from UV
+
far-IR
Herschel and ALMA measurements,
2. measured molecular gas masses from far-IR dust
continuum observations with ALMA,
3. measured stellar masses and accurate photometric red-
shifts from multiband photometry, and
4. deep subkiloparsec-resolution JWST rest-frame optical
and near-IR imaging data.
The galaxies reside in the COSMOS
fi
eld
(
N. Scoville et al.
2007
)
, which provides a wealth of ancillary data including
X-ray and radio measurements, Hubble rest-frame UV and
optical imaging
(
ACS
/
F814W; A. M. Koekemoer et al.
2007
)
,
and spectroscopic redshifts for more than half of the sample. In
the following, we summarize the different data products, basic
measurements, and sample properties.
2.1. Photometry and Imaging
2.1.1. UV and Optical Photometry
All photometric redshifts, UV-b
ased SFRs, and stellar masses
are based on the multiband COSMOS2020 catalog photometry
(
J. R. Weaver et al.
2022
)
. Obvious active galactic nuclei
(
AGNs
)
are removed from the sample by a combination of spectral energy
distribution
(
SED
)
fi
tting
(
based on a comparison of the goodness
of
fi
t between star-forming and AGN templates; see J. R. Weaver
et al.
2022
)
and selection in X-ray and radio emission. The dust-
unobscured star formation is measured from the 1500
Å
emission
constrained by the best-
fi
t SED. Stellar masses are derived using
LePhare
(
S. Arnouts et al.
1999
;O.Ilbertetal.
2006
)
, assuming
a variety of templates, ages, met
allicities, and dust attenuations.
Uncertainties in SFRs and stella
r masses are generally less than
0.3 dex
(
J. R. Weaver et al.
2022
)
. Spectroscopic redshifts are
available for 58% of the galaxies in our sample from various
surveys
(
e.g., O. Le Fèvre et al.
2015
;G.Hasingeretal.
2018
;
A. Khostovan et al. 2025, in preparation
)
and suggest a
photometric redshift accuracy better than
σ
z
=
0.14. Note that
using the spectroscopic redshifts
for deriving stellar masses and
other properties would change their values within their uncertain-
ties, thus would not change the conclusions of this work.
2.1.2. JWST Imaging
(
for Morphology
)
JWST provides rest-frame optical
/
near-IR imaging at a
30 mas pixel scale from two cycle 1 GO programs: COSMOS-
Web
(
PID 1727; PIs: Kartaltepe & Casey; C. M. Casey et al.
2023
)
covering 0.54 deg
2
of the COSMOS
fi
eld and PRIMER-
COSMOS
(
PID 1837; PI: Dunlop
)
covering the CANDELS-
COSMOS
fi
eld. In this work, we make use of the full
COSMOS-Web data set, including the NIRCam F115W,
F150W, F277W, and F444W
fi
lters over the full area.
PRIMER-COSMOS uses F090W, F150W, F200W, F277W,
F356W, F410M, and F444W. The point-spread function
(
PSF
)
FWHM varies between 40 mas and 145 mas for the different
fi
lters. The physical scale varies from 6.2 kpc arcsec
–
1
at
z
=
0.5 to 7.4 kpc arcsec
–
1
at
z
=
3.5. MIRI imaging is not used
here due to the small area coverage and shallow depth. See
M. Franco et al.
(
2024
)
for the details of the JWST data
reduction. These rest-frame optical
/
near-IR JWST observa-
tions are crucial to trace the bulk of the stellar mass, as rest-
frame UV imaging would be biased to unobscured star-forming
regions in the galaxies and therefore could mislead the
identi
fi
cation of merging systems
(
see A. Cibinel et al.
2019
)
.
(
However, we note that rest-frame UV imaging is used to
identify galaxies with star-forming clumps.
)
We note that the
JWST photometry is not used for SED
fi
tting as it does not add
much in terms of wavelength coverage and depth to the
available ground-based data for these relatively bright galaxies.
However, we checked internally that the photometry between
JWST and ground-based observations is consistent, and we
therefore do not expect any biases in the photometry.
2.1.3. Herschel Photometry and Measurements
Infrared photometry is derived from various data available
for the COSMOS
fi
eld: at 24
μ
m by Spitzer
(
D. B. Sanders
et al.
2007
)
; at 100
μ
m and 160
μ
m by Herschel-PACS
(
A. Poglitsch et al.
2010
)
as part of the PACS Evolutionary
Probe program
(
D. Lutz et al.
2011
)
; and at 250
μ
m, 350
μ
m,
and 500
μ
m by Herschel-SPIRE
(
M. J. Grif
fi
n et al.
2010
)
as
part of the Herschel Multi-tiered Extragalactic Survey
(
S. J. Oliver et al.
2012
)
. For the
fl
ux extraction, a linear
inversion technique of cross-identi
fi
cation
(
I. G. Roseboom
et al.
2010
,
2012
)
is used based on positional priors from the
Spitzer 24
μ
m catalog and Very Large Array 1.4 GHz data
(
E. Le Floc
’
h et al.
2009
; E. Schinnerer et al.
2010
)
. All sources
are detected at
>
3
σ
in at least two of the
fi
ve Herschel bands.
For detailed information on deblending algorithms, we refer to
Appendix C2 in N. Scoville et al.
(
2023
)
as well as N. Lee et al.
(
2013
)
. The infrared SFRs are measured from the total infrared
luminosity via SFR
IR
[
M
e
yr
−
1
]
=
8.6
×
10
−
11
L
IR
[
L
e
]
. This
assumes that all stellar light is dust-obscured in the
fi
rst
100 Myr
(
and none thereafter
)
. The derived infrared SFRs
would increase by 50% for dust-enshrouded timescales of
10 Myr
(
N. Scoville & L. Murchikova
2013
)
. The infrared
luminosity is measured by integrating over a modi
fi
ed black-
body
(
C. M. Casey
2012
)
fi
t to the infrared photometry
(
see
N. Scoville et al.
2023
for more details
)
. The infrared SFRs are
combined with the UV SFRs to obtain the total SFRs.
2.1.4. ALMA Continuum Measurements
The molecular interstellar medium
(
ISM
)
gasmassesare
derived from the Rayleigh
–
Jeans
(
RJ
)
dust continuum
(
rest-frame
850
μ
m
)
as described in N. Scoville et al.
(
2014
)
.The
/
n
m
L
M
ga
s
850 m
ratio was calibrated over a range of galaxy types
(
main-sequence, starburst, an
d luminous infrared galaxies
)
out to
z
=
3. The RJ continuum is derived from archival ALMA band 6
and 7 observations
(
at
>
2
σ
signi
fi
cance
)
as of 2021 June. Only
data with
uv
-coverage resolving a source extent of
∼
1
′′
are used
for robust
fl
ux measurements. The
fl
uxes and their uncertainties
are derived from least-squares
fi
tting. More information on the
fl
ux measurements is provided in N. Scoville et al.
(
2023
)
.
2.2. De
fi
nitions and Sample Properties
We de
fi
ne the molecular gas fraction as
f
gas
=
M
mol,gas
/
(
M
*
+
M
mol,gas
)
and the SFE based on the total UV
+
IR SFR as SFE
=
M
mol,gas
/
SFR in units of Gyr
−
1
. The
depletion time
(
t
depl
)
is the inverse of the SFE. To calculate
Δ
MS
(
the logarithmic offset from the star-forming main
sequence at a given stellar mass
)
, we assume the main-
sequence parameterization from N. Lee et al.
(
2015
)
, which is
based on Herschel measurements of individual galaxies and
stacked samples on the COSMOS
fi
eld. We emphasize that
3
The Astrophysical Journal,
980:204
(
12pp
)
, 2025 February 20
Faisst et al.
Herschel observations are crucial for deriving a robust
Δ
MS
of
starburst galaxies, often dominated by dust-obscured star
formation. The use of other parameterizations
(
e.g., J. S. Speagle
et al.
2014
; C. Schreiber et al.
2015
)
leads to similar
Δ
MS
and
has a minimal impact on the
fi
nal results of this work.
Figure
1
shows the dependence of SFE and
f
gas
on stellar
mass and
Δ
MS
for our selected 387 galaxies in different
redshift bins. The underlying hex bins show the full sample
from N. Scoville et al.
(
2023
)
. This shows that our selected
subsample traces the full sample properties closely. In addition,
note that
Δ
MS
correlates well with SFE and less with
f
gas
,asit
has been found in various works including that of N. Scoville
et al.
(
2014
)
and J. D. Silverman et al.
(
2015
)
.
There are mainly two selection biases to be noted. These are
due to the requirement of Herschel and ALMA detections for the
computation of accurate far-IR SFRs and gas masses. First, the
high-redshift sample is biased toward high
Δ
MS
. Second, low
stellar masses are biased toward high
Δ
MS
. While the
fi
rst bias is
not a large concern
(
our analysis will be done in separate redshift
bins
)
, the second bias may introduce unphysical relations with
stellar mass. We therefore mitigate that bias by introducing mass-
controlled samples for the following analysis by adopting two
mass bins for each redshift range
. We found that two stellar mass
bins are suf
fi
cient to mitigate the bias, and more mass bins would
reduce the sample and decrea
se the statistical robustness.
However, due to the small sample at
z
>
3, we only adopt a
single mass bin for the highest redshifts. We de
fi
ne the mass cuts
in each of the redshift bins based on Figure
1
to remove any
signi
fi
cant dependence between stellar mass and
Δ
MS
within the
mass-binned subsamples. The adopted mass bins
(
in
/
()
MM
log
)
are
[
10.1, 10.8
]
and
[
10.8, 11.7
]
for
z
∼
0.5,
[
10.4, 11.1
]
and
[
11.1, 11.8
]
for
z
∼
1.5,
[
10.7, 11.1
]
and
[
11.1, 11.7
]
for
z
∼
2.5,
and
[
10.7, 11.2
]
for
z
∼
3.5
(
all redshift bins with
Δ
z
=
1
)
.
Changing these mass bins in a reasonable range
(
±
0.2 dex
)
does
not affect the
fi
nal conclusions of this work.
3. Structural Analysis
3.1. Visual Classi
fi
cation
We perform a visual classi
fi
cation of the 387 galaxies using
all of the COSMOS-Web JWST NIRCam
fi
lters to minimize
the effects of color-dependent morphology corrections. In the
following, we de
fi
ne four morphological groups.
Noninteracting.
Isolated galaxies of various kinds, such as
disk or compact galaxies, that are not in an interacting state.
Disk galaxies.
These are extended galaxies
(
to be distin-
guished from compact galaxies
)
, preferentially with a
semiminor-to-semimajor axis ratio of less than 0.8 and a
disk structure. The disk can either be face-on or edge-on.
Clumpy disks.
This is a subset of the disk galaxies category,
showing more than one star-forming clump. The clumps are
pronounced in the bluer bands but visible throughout redder
wavelengths. Due to limitations in resolution and confusion
with interacting systems, we have not classi
fi
ed clumpy disks
at redshifts
z
>
3 in this work.
Interacting.
Interacting or irregular galaxy systems are
identi
fi
ed by multiple nuclei, irregular structures, or tidal
features. This category should include mainly galaxy systems
in premerger and postmerger phases.
It is important to note that we must make use of all available
NIRCam
fi
lters for a reliable morphological classi
fi
cation. One
example to showcase the importance of multiple photometric
bands for a morphological classi
fi
cation is discussed in Z. Liu
et al.
(
2024
)
. As shown in their Figure 1, this starburst galaxy
could be classi
fi
ed as a clear merger in JWST
fi
lters F115W
and F150W. However, at longer rest-frame wavelengths
(
speci
fi
cally F227W
)
, this source shows a clear disk structure
—
a
fi
nding that is supported by kinematic measurement of the
CO gas.
For consistency across redshifts, we therefore visually
classi
fi
ed interacting galaxies primarily at rest-frame optical
wavelengths
(
0.8
–
1
μ
m
)
and redder bands where available.
This choice is motivated by the fact that those wavelengths
trace the stellar mass of galaxies and hence are more ideal to
discriminate between interacting systems and galaxies with a
signi
fi
cant disk disk component as shown in the case by Z. Liu
et al.
(
2024
)
. We use F150W for
z
<
1.0, F277W for 1
<
z
<
3,
and F444W for
z
>
3 but also consider redder bands where
available. In addition, red, green, and blue
(
RGB
)
color images
(
in addition to redshifts where available
)
are used to identify
Figure 1.
Dependency of SFE
(
color-coded from 0.1 Gyr
−
1
, dark, to 10 Gyr
−
1
, light
)
and molecular gas fraction
(
f
gas
; size of circles from 0, small, to 1, large
)
on
stellar mass and offset from the star-forming main sequence
(
Δ
MS
)
in our sample. Each panel shows a different redshift bin. The underlying hex bins show the SFE for
the full sample of 704 galaxies from N. Scoville et al.
(
2023
)
. Note the stronger correlation between
Δ
MS
and SFE compared to
f
gas
. Also visible is a selection effect
due to the requirement of Herschel detections, which leads to the lack of galaxies on the main sequence at lower stellar masses
(
see text for more detailed discussion
)
.
4
The Astrophysical Journal,
980:204
(
12pp
)
, 2025 February 20
Faisst et al.
companions
(
mergers
)
and reject background
/
foreground
galaxies. We note that the PSF varies signi
fi
cantly between
those bands
(
from FWHM 0
.
06 to 0
.
16 for F115W to F444W;
M.-Y. Zhuang & Y. Shen
2024
)
, which may be a concern for
high redshifts. We tested our classi
fi
cation by carefully
comparing PSF-matched images to rule out signi
fi
cant biases
in our visual classi
fi
cation due to such PSF differences.
Although we made use of all available bands in the
morphological classi
fi
cation, we additionally tested the
dependence of our morphological classi
fi
cation on single
bands. We
fi
nd, as expected, that using
fi
lters at bluer
wavelengths breaks up the galaxies in clumps
(
see also below
)
and biases the morphological classi
fi
cation against disk
galaxies. On the other hand, the use of
fi
lters redder than
F227W for
z
>
1 galaxies does not change the classi
fi
cation. In
addition,
∼
5% of galaxies are covered by the PRIMER-
COSMOS observations, which include four more
fi
lters in
addition to COSMOS-Web. The morphological classi
fi
cation
was also tested against these
fi
lters
(
speci
fi
cally F356W,
providing an intermediate band between F277W and F444W at
a better PSF resolution than F444W
)
, and we do not
fi
nd biases
using redder bands for the classi
fi
cation of interacting systems.
On the other hand, we use bluer bands
(
F115W and F150W
)
for the identi
fi
cation of clumpy disk galaxies. This is motivated
by the fact that such clumps are generally star-forming and rest-
UV-bright
(
and at the same time, the PSF of bluer bands is
smaller
)
. However, there may be a bias in clump identi
fi
cation
in high-redshift galaxies caused by their smaller physical sizes
and the large PSF at the same rest-frame
fi
lters. Furthermore, a
differentiation between clumpy disks and interacting galaxies
without kinematic information from spectroscopy is nearly
impossible in these cases
(
G. C. Jones et al.
2021
)
.We
therefore do not attempt to identify clumpy disk galaxies at
z
>
3. In addition to the above method, we use residual images
for identifying clumps. These are generated in two ways:
(
i
)
by
subtracting a smooth pro
fi
le
fi
t
(
Sérsic
)
performed using
statmorph
(
V. Rodriguez-Gomez et al.
2019
)
from the
galaxy image and
(
ii
)
by directly subtracting a long-wavelength
from a short-wavelength image
(
e.g., F150W
–
F444W
)
, which
highlights blue rest-UV structure and removes the smooth ISM.
We classify a galaxy as a
clumpy disk
if there are more than
three bright
(
S
/
N
>
3
)
clumps detected. We use both methods
to support the visual classi
fi
cation of clumpy disks from the
individual or RGB images.
Finally, we compare the above visual classi
fi
cation with
more statistical classi
fi
cations such as the
G
–
M
20
(
Gini versus
M
20
)
or
C
–
A
(
concentration versus asymmetry
)
diagnostics. We
fi
nd that these methods are most ef
fi
cient in selecting
interacting galaxies; however, they have a signi
fi
cantly lower
purity and completeness for selecting clumpy disks
(
see the
Appendix
)
.
In the following, we denote the fraction of interacting
galaxies in our sample with
f
int
. Note that, as mentioned above,
without spectroscopic con
fi
rmation of the detailed kinematics,
we are not able to identify close-pair mergers. The interacting
group therefore mostly focuses on the pre- and postmerger
phases. The fraction of clumpy disk galaxies compared to the
total disk galaxy population in a given selection bin is denoted
as
f
clumpy
. We derive the uncertainties on these fractions using
the formalism of
(
Bayesian
)
binomial con
fi
dence intervals
suggested by E. Cameron
(
2011
)
and shown in code form in
their Appendix A.
Figure
2
shows examples of our classi
fi
cation scheme in four
redshift bins. The redshifts and galaxy types are indicated on
each cutout, as well as
f
gas
and SFE. Note the clear distinction
of disk galaxies with and without clumps.
Figure 2.
Examples of galaxies imaged in JWST NIRCam
/
F277W at different redshifts organized in three different morphological groups
(
here disk galaxies are
combined in the noninteracting group
)
. Subcategories
(
such as disk vs. clumpy disk vs. compact
)
are labeled in the cutouts as well as
f
gas
, SFE, and redshifts. Note that
we do not classify clumpy disks at
z
>
3 due to a lack of resolution and signi
fi
cant confusion with interactions.
5
The Astrophysical Journal,
980:204
(
12pp
)
, 2025 February 20
Faisst et al.
4. Results
4.1. The Impact of Galaxy
–
Galaxy Interactions on Starburst
Activity
We
fi
rst compare the fraction of interacting systems to the
offset from the main sequence, the gas fraction, and the SFE.
The top panel in Figure
3
shows the fraction of interacting
galaxies as a function of
Δ
MS
in two different mass bins per
redshift. The fraction clearly in
creases off the main sequence
(
indicated by the gray area
)
toward higher
Δ
MS
and reaches 20%
–
40% at
z
<
3and80%at
z
=
3.5. However, we note that, due to
the lack of near-IR observations at
these redshifts, the merger
classi
fi
cation may not be as robust as at lower redshifts
(
for
example, due to dust obscuration for starburst galaxies
)
.Thiscan
lead to a classi
fi
cation bias
(
e.g., due to clumpy disks; see
discussion in Section
3.1
)
thatmayleadtoanoverpredictionof
the fraction of merging galaxies at
z
=
3.5.
Note that here we focus on the change in fractions across
different properties and do not attempt to compare their
absolute values to the literature. This is because relative
changes are more reliable than the absolute values of
f
int
, which
depend on sample selection and subjective visual selection
thresholds and therefore are dif
fi
cult to compare. The two mass
bins show very similar behaviors. Keeping this in mind, an
increase of the fraction of interacting systems toward higher
redshifts is found, which is in agreement with a global increase
in the merger fraction studied in other works
(
e.g.,
L. A. M. Tasca
2014
; M. Romano et al.
2021
)
. Overall, this
is consistent with the theory of galaxy
–
galaxy interactions
playing some role in inducing starburst activity. Here we show
that this may be the case out to
z
∼
4. We also note that this is
in line with and directly related to the increased fraction of
interacting galaxies found at higher infrared luminosities
(
see
the study at
z
<
1.5 by C.-L. Hung et al.
2013
)
.
It is suggested that the gas fraction is weakly correlated with
starburst activity
(
e.g., N. Scoville et al.
2023
)
, and we
therefore would not expect a signi
fi
cant correlation between
f
int
and
f
gas
. The middle panels of Figure
3
show that this is indeed
the case by displaying
f
int
as a function of the gas fraction of
galaxies relative to the population mean
(
i.e.
,
for a given stellar
mass and redshift
)
. No signi
fi
cant correlation between those
quantities is seen in either stellar mass bin.
Finally, the bottom panels in Figure
3
show
f
int
as a function
of SFE
(
relative to the mean of the population at a given
redshift and stellar mass
)
.Upto
z
=
2, we
fi
nd a clear trend of
galaxies with a higher-than-average SFE residing more
frequently in interacting systems, speci
fi
cally in the lower
mass bin. At higher redshifts, this trend is less signi
fi
cant. As
discussed in Section
5
, theoretical works indicate that the SFE
increase may be less correlated with interactions prior to
cosmic noon. Furthermore, as argued later, disk fragmentation
in gas-rich environments at high redshifts could play a more
signi
fi
cant role in increasing the galaxies
’
SFE.
Overall, we found that the fraction of interacting galaxies is
increased in the starburst regime out to
z
∼
4 and correlates
with SFE but not gas fraction
(
at least out to
z
=
2
)
. Galaxy
–
galaxy interactions therefore represent a veritable way to push
galaxies into the starburst region at redshifts beyond cosmic
noon. However, our analysis also shows that interacting
galaxies only make up at most 40% of the
z
<
3 starburst
galaxies in our sample. This implies that noninteracting
systems contribute signi
fi
cantly to the starburst population.
The increase of star formation
(
ef
fi
ciency
)
through disk instabilities
could provide another avenue for galaxies to reach the starburst
regime
(
see Section
1
)
. In this case, we would expect a higher
fraction of disk galaxies with pronounced star-forming clumps in
the starburst regime. This is studied in the next section.
4.2. The Impact of Disk Instabilities on Starburst Activity
A signi
fi
cant fraction
(
50%
–
80%
)
of galaxies in the starburst
regime in our sample are noninteracting
(
i.e.
,
isolated
)
disk
galaxies. These may have been interacting in the past; however,
taking the current morphological evidence at face value shows that
they are currently not in a merging state. A possible way to reach
the starburst regime is through an increased SFE due to the
instability in gas-rich disks. The idea
(
see, for example, A. Dekel
et al.
2009
;A.B.Romeo&K.Fathi
2016
)
here is that gas-rich
streams increase the gas density of the disk, which then becomes
unstable and starts to fragment into clumps
(
Toomre instability;
A. Toomre
1964
)
. These clumps can contribute to several percent
of the total disk mass, and star formation is maintained in dense
subclumps over timescales of several hundred Myr. Steady inbound
gas streams on the disk maintain the instability of the disk and
replenish gas over several Gyr. Eventually, the clumps might
migrate toward the center due to dynamical friction and may form
spheroid-dominated galaxies later on. Recent measurements with
JWST using multiple broad and medium bands suggest that the
majority of clumps are less than
∼
200 Myr old, suggesting a
similarly long survival time
(
e.g., A. Claeyssens et al.
2023
)
.
The occurrence of UV-bright star-forming clumps in
galaxies is ubiquitous at high redshifts
(
about 60% of
z
=
1
–
3 galaxies contain bright UV clumps; L. L. Cowie
et al.
1995
; C. J. Conselice et al.
2004
; B. G. Elmegreen et al.
2013
; Y. Guo et al.
2015
; E. Soto et al.
2017
; A. Zanella et al.
2019
)
. Several studies support the formation of these clumps
through in situ physical processes based on differences in the
evolution of the clump fraction and minor
/
major mergers; the
disk nature of their host galaxies and, similarly, the kinematic
properties of ordered disk rotation with high velocity disper-
sion; the distribution of clumps with scale height for edge-on
galaxies; and the stellar mass function being similar to local
star clusters and H
II
regions
(
see T. Shibuya et al.
2016
;
B. G. Elmegreen et al.
2017
; M. Dessauges-Zavadsky &
A. Adamo
2018
; M. Girard et al.
2020
)
. A recent study of
lensed clumpy galaxies at
z
=
1
(
M. Dessauges-Zavadsky et al.
2019
,
2023
)
corroborates the picture in which clumps are
formed in situ through disk instabilities.
Along these lines, B. Wang & J. Silk
(
1994
)
present a simple
recipe to link star formation to disk instability. In this simple
theoretical model, the SFE
(
de
fi
ned here as SFE
/
M
gas
)
is
proportional to
/
()
()
ºμ
-
M
Q
Q
SFE
SFR
1
,1
gas
21 2
where
=
k
p
S
Q
v
G
gas
gas
is the Toomre disk instability parameter,
with
κ
the epicyclic frequency,
v
g
the radial cloud velocity
dispersion,
S=
p
M
r
gas
gas
disk
2
the gas surface density, and
G
the
gravitational constant. A similar expression can be derived
from the analytical model presented in A. Dekel et al.
(
2009
;
see their Equation
(
47
))
. Equation
(
1
)
is valid for an unstable
disk, i.e.
,Q
<
1. According to this model, the more unstable
the disk, the more stars per surface are formed. Consequently,
6
The Astrophysical Journal,
980:204
(
12pp
)
, 2025 February 20
Faisst et al.
SFE rises as 1
/
Q
as
Q
→
0
(
see Equation
(
1
))
. We emphasize
the simplicity of this model
(
see, for example, the discussion in
A. B. Romeo & J. Wiegert
2011
; A. B. Romeo & N. Fals-
tad
2013
; S. E. Meidt et al.
2023
)
as stars can also be formed if
Q
>
1; however,
Q
=
1 may be a condition of exceptional star
formation. Note that by simply applying empirical correlations,
we get
()
m
μS μ μ μ
--
*
*
*
Q
r
M
M
M
M
,2
gas
1disk
2
gas
0.4
0.24
where we have used
m
=μ
-
-
*
M
f
f
1
0.36
gas
gas
(
L. J. Tacconi et al.
2018
)
and
μ
a
*
r
M
disk
with
α
∼
0.2
(
e.g., L. Yang et al.
2021
)
.
We therefore would expect only a weak stellar mass
dependence. In addition, we note that the dispersion
Q
∝
v
g
may be weakly positively correlated with stellar mass
(
H. Übler
et al.
2019
)
, making the dependence of
Q
on stellar mass even
weaker. If disk instabilities are at work to push disk galaxies
into the starburst regime, we would expect an increase of
galaxies with a number of dense star-forming clumps
Figure 3.
Relation between the fraction of interacting systems
(
f
int
)
and the offset from the main sequence
(
Δ
MS
; top
)
, gas fraction
(
f
gas
; middle
)
, and SFE
(
bottom
)
for different redshift ranges
(
Δ
z
=
1
)
and two stellar mass bins. The latter two quantities are normalized to the mean of the population
(
width indicated by the gray
region
)
at a given redshift and stellar mass.
7
The Astrophysical Journal,
980:204
(
12pp
)
, 2025 February 20
Faisst et al.