Astronomy & Astrophysics manuscript no.

metal_jwst ©ESO 2024

August 2, 2024

The mass-metallicity relation as a ruler for galaxy evolution:

insights from the James Webb Space Telescope
A. Pallottini 1,⋆ , A. Ferrara 1, S. Gallerani 1,L. Sommovigo 2, S. Carniani 1, L. Vallini 3, M. Kohandel 1,

and G. Venturi 1

1
Scuola Normale Superiore, Piazza dei Cavalieri 7, 56126 Pisa, Italy
2
Center for Computational Astrophysics, Flatiron Institute, 162 5th Avenue, New York, NY 10010, USA
3
INAF-Osservatorio di Astrofisica e Scienza dello Spazio, via Gobetti 93/3, I-40129, Bologna, Italy

Received July 31, 2024; accepted XX XX, XXXX

arXiv:2408.00061v1 [astro-ph.GA] 31 Jul 2024

ABSTRACT

Context. Galaxy evolution emerges from the balance between cosmic gas accretion, fueling star formation, and supernova feedback
regulating metal enrichment of the interstellar medium. Hence, the relation between stellar mass (M⋆ ) and gas metallicity (Zg ) is
fundamental to understand the physics of galaxies. High-quality spectroscopic JWST data enable accurate measurements of both M⋆
and Zg up to redshift z ≃ 10.
Aims. Our aims are to understand (i) the nature of the observed mass-metallicity relation (MZR), (ii) its connection with the star
formation rate (SFR), (iii) the role played by SFR stochasticity (flickering), and (iv) how it is regulated by stellar feedback.
Methods. We compare the MZR obtained by the JADES, CEERS, and UNCOVER surveys, which comprise about 180 galaxies at
z =≃ 3 − 10 with 106 M⊙ ∼ < M < 1010 M , with ≃ 200 simulated galaxies in the same mass range from the serra high-resolution (≃
⋆ ∼ ⊙
20 pc) suite of cosmological radiation-hydrodynamic simulations. To interpret the MZR, we develop a minimal, physically motivated,
model for galaxy evolution that includes: cosmic accretion, possibly modulated with an amplitude A100 on 100 Myr timescales; a time
delay td between SFR and supernova feedback; SN-driven outflows with a varying mass loading factor ϵSN , that is normalized to the
fire simulations predictions for ϵSN = 1.
Results. Using our minimal model, we find the observed mean MZR is reproduced for relatively inefficient outflows (ϵSN = 1/4),
in line with findings from JADES. Matching the observed MZR dispersion across the full stellar mass range requires a delay time
td = 20 Myr, in addition to a significant modulation (A100 = 1/3) of the accretion rate. Successful models are characterized by a
relatively low flickering (σSFR ≃ 0.2), corresponding to a metallicity dispersion of σZ ≃ 0.2. Such values are close but slightly lower
than predicted from serra (σSFR ≃ 0.24, σZ ≃ 0.3), clarifying why simulations show no clear trend for the MZR, and show some
tension with the observations especially at M⋆ ≃ 1010 M⊙ .
Conclusions. As the MZR is very sensitive to SFR stochasticity, models predicting high r.m.s. values (σSFR ≃ 0.5) result in a
“chemical chaos” (i.e. σZ ≃ 1.4), virtually destroying the observed MZR. As a consequence, invoking a highly stochastic SFR
(σSFR ≃ 0.8) to explain the overabundance of bright, super-early galaxies leads to inconsistencies with the observed MZR.
Key words. Galaxies: star formation – evolution – high-redshift

1. Introduction cally, these observations have shown that Zg increases with stel-
lar mass up to M⋆ ≃ 1010 M⊙ and beyond that point it flattens.
The baryon cycle plays a pivotal role in regulating the galaxy This can be understood as due to the fact that low mass galax-
formation and evolution process through cosmic time. As cos- ies are prone to metal ejection due to their shallower potential
mic gas accretes in the dark matter halo potential well, it can well, while massive galaxies can retain most of the metals they
cool and eventually form stars; massive stars explode as super- have produced (Ferrara 2008). In this view, galaxy evolution is
novae, enriching the surrounding interstellar medium with met- guided by a “bathtub” equilibrium (Bouché et al. 2010; Dekel &
als, that can potentially be ejected from the galaxy. Therefore key Mandelker 2014) which determines the metallicity by balancing
information regarding the star formation and enrichment history infall and outflows (Lilly et al. 2013).
is encoded in the relation between stellar mass (M⋆ ) and gas
metallicity (Zg ), the so called mass-metallicity relation (MZR), Further, observations have highlighted that Zg and M⋆ are
making it effectively a standard ruler for galaxy evolution (see tightly connected with the star formation rate (SFR) in the so-
Maiolino & Mannucci 2019, for a review). called fundamental mass-metallicity relation (FMR, Mannucci
Over the past 20 years, it has become possible to measure et al. 2010), which seems to be redshift-independent up to z ≃ 3
the MZR from z ≃ 0 (Tremonti et al. 2004; Kirby et al. 2013; (Curti et al. 2020); other factors as galaxy size (Ellison et al.
Andrews & Martini 2013) up to the beginning of cosmic noon 2008) and molecular gas content (Bothwell et al. 2013) might
(z ≃ 3, Erb et al. 2006; Maiolino et al. 2008; Zahid et al. 2011; also play some secondary role in connecting Zg to M⋆ .
Sanders et al. 2018; Curti et al. 2020), mostly by using the oxy- The MZR evolution at z ∼ > 4 has been predicted via semi-
gen to hydrogen abundance ratio [O/H] as a proxy for Zg . Lo- analytical models (Dayal et al. 2013; Zahid et al. 2014; Ucci
et al. 2023) or cosmological simulations (Torrey et al. 2019;
⋆
[email protected] Liu & Bromm 2020; Langan et al. 2020; Wilkins et al. 2023).
Article number, page 1 of 13
 A&A proofs: manuscript no. metal_jwst

Most models predict almost no evolution or a very weak red- the possibility of detecting Population III stars (Vikaeus et al.
shift dependence for z > 3 (Ma et al. 2016; Kannan et al. 2022; 2022; Katz et al. 2023). In addition, the flickering can modify the
Marszewski et al. 2024); some others suggest a break down of UV spectral slopes of high-z galaxies and the escape fraction of
the relation at z ∼> 9 − 12 (Pallottini et al. 2014; Sarmento et al. ionizing photons (Gelli et al. 2024a), ultimately controlling the
2018). cosmic reionization history (Davies & Furlanetto 2016; Nikolić
Thanks to the exquisite spectroscopic capabilities of JWST, et al. 2024).
we can accurately infer Zg for galaxies well within the Epoch Analyzing the amplitude and temporal variation of SFR
of Reionization (EoR, z > 6). At present, surveys like JADES flickering provides unique insights into the various feedback pro-
(Bunker et al. 2023) and CEERS (Finkelstein et al. 2023), have cesses that regulate the formation and evolution of early galax-
provided metallicity measurements at z ≃ 3 − 10 (Curti et al. ies (Pallottini & Ferrara 2023). On the one hand, high-amplitude
2024a) and z ≃ 4 − 10 (Nakajima et al. 2023), for a combined and high-frequency flickering, on timescales shorter than the de-
sample of about 170 galaxies in the 5×106 M⊙ ∼ < M < 1010 M lay between SFR and SN explosions (≃ 20−40 Myr) are invoked
⋆ ∼ ⊙
stellar mass range. (Gelli et al. 2023, 2024b) to explain the observed abrupt quench-
These JWST observations suggest that the MZR is already ing of SFR in ≃ 109 M⊙ galaxies already at z ≃ 7 (Dome et al.
in place in the EoR, albeit downshifted by 0.5 dex with respect 2024; Looser et al. 2024). On the other hand, the presence of
to the local one. The evolution in the z ≃ 3 − 10 range is very disks observed in M⋆ ≃ 109−10 M⊙ galaxies up to z ≃ 7 (Row-
mild, as relatively low mass (M⋆ ≃ 109−10 M⊙ ) galaxies can be land et al. 2024; Fujimoto et al. 2024) implies a limited level of
chemically mature (Zg ≃ 0.3 Z⊙ ) already at z ≃ 6. The MZR stochasticity, as otherwise the disk would be disrupted (Kohan-
shows a r.m.s. dispersion σZ = 0.3 dex, and possibly a hint of del et al. 2024).
flattening at z ∼> 6 (Curti et al. 2024a). In the present work, we aim to clarify the effects of SFR
Furthermore, multiple line detections in the same target can flickering and feedback regulation of high-z galaxies by using
be used to constrain abundance patterns of individual elements the MZR as a standard ruler. The paper is structured as follows;
(Kobayashi & Ferrara 2024; Curti et al. 2024b). Resolved ob- in Sec. 2 we present the serra simulation suite and compare it
servations are beginning to probe the presence of metallicity with MZR observations from JADES, CEERS, and UNCOVER;
gradients up to z ≃ 8 (Venturi et al. 2024), which should pro- in Sec. 3 we develop a minimal model to give a physical interpre-
vide deeper insights into the formation process of the galaxy and tation to the MZR; in Sec. 4 we discuss the discrepancies found
the history of the stellar mass buildup (see Vallini et al. 2024, between simulations and data in the light of the analytical model;
for an ALMA perspective). Observations targeting lensed fields conclusions are given in Sec. 5.
(Bezanson et al. 2022; Chemerynska et al. 2024b) are uncover-
ing the metallicities of even the fainter galaxies (Chemerynska 2. The mass-metallicity relation at high redshift
et al. 2024a), which is key in order to explore the infall/outflow
interplay due to stellar feedback in the less massive objects. Im- To study the MZR at high-z, firstly we adopt the data from serra,
portantly, using observations of lensed targets at z ≃ 3 − 9, Mor- a suite of cosmological zoom-in simulations that follows the evo-
lution of z ∼> 6 galaxies (Pallottini et al. 2022). In serra we use
ishita et al. (2024) suggest that the 0.5 dex shift of normalization
and the higher scattering (0.3 dex) w.r.t. the local MZR can be the code ramses (Teyssier 2002) to evolve DM, gas, and stars.
caused by the high level of burstiness of the SFR expected in We enable the module ramses-rt (Rosdahl et al. 2013) to track
high-z galaxies (Pallottini & Ferrara 2023; Sun et al. 2023b). radiation on-the-fly by coupling it (Pallottini et al. 2019; De-
Indeed, as we move from the local Universe to high z cataldo et al. 2019) with krome (Grassi et al. 2014), in order to
(see Madau & Dickinson 2014; Dayal & Ferrara 2018; Förster account for non-equilibrium chemistry up to H2 formation (Pal-
Schreiber & Wuyts 2020, for reviews), galaxies are expected to lottini et al. 2017a). Stars are formed using a Schmidt (1959)-
become more bursty, as a consequence of the increase of specific Kennicutt (1998) like relation based on H2 . Once formed, stars
SFR (sSFR, González et al. 2010; Stark et al. 2013; Smit et al. produce UV radiation that ionizes the gas, photo-dissociates H2 ,
2014) combined with a decrease in size (Shibuya et al. 2015), and builds a radiation pressure in the gas. Massive stars gener-
which can make the stellar feedback more effective (Krumholz ate stellar winds and explode as SNe. Depending on the kind
& Burkhart 2016) in maintaining a higher level of turbulence (Si- of feedback, energy injection in the gas can be of thermal (sub-
mons et al. 2017; Genzel et al. 2017). With the recent availability ject to radiative losses) and/or turbulent (later dissipated) nature
of JWST observations (Roberts-Borsani et al. 2022; Castellano (Pallottini et al. 2017b). Adopting a ≃ 1.2 × 104 M⊙ (≃ 20 pc) gas
et al. 2022; Finkelstein & Bagley 2022; Naidu et al. 2022; Treu mass (spatial) resolution, with music (Hahn & Abel 2011) we
et al. 2022; Adams et al. 2023; Atek et al. 2023; Donnan et al. initialize cosmological initial conditions1 generated at z = 100,
2023; Harikane et al. 2023; Santini et al. 2023), the time vary- and follow the evolution of galaxies down to z = 6 (and z = 4
ing stochastic SFR behavior (in short, flickering or burstiness), for a sub-sample, see Kohandel et al. 2024).
has been invoked as a possible mechanism (Mason et al. 2023; For this work, we select ≃ 200 central serra galaxies at
z = 7.7 with stellar mass 106 M⊙ ∼ < M < 1010 M ; we compute
Mirocha & Furlanetto 2023; Shen et al. 2023; Sun et al. 2023b; ⋆ ∼ ⊙
Kravtsov & Belokurov 2024) to explain the overabundance of the gas metallicity as Zg = MgZ /Mg by considering the total gas
bright galaxies at z ∼ > 10 (see Dekel et al. 2023; Ferrara et al. (Mg ) and metal (MZ ) mass contained within 3 stellar effective
2023, for alternatives). However, some works predict that r.m.s. radii from the galaxy center, which is close to the gas half-mass
amplitude of the SFR variability falls short of explaining the said radius2 . Note that, for a more direct comparison with obser-
overabundance (Pallottini & Ferrara 2023). Thus, it is unclear if vation, we should compute line and continuum emission from
flickering alone can explain the phenomenon, particularly since 1
We adopt a ΛCDM model with vacuum, matter, and baryon densities
the required high level of variability (Muñoz et al. 2023) is ap- in units of the critical density ΩΛ = 0.692, Ωm = 0.308, Ωb = 0.0481,
parently not seen in the data (Ciesla et al. 2024). normalized Hubble constant h = H0 /(100 km s−1 Mpc−1 ) = 0.678, spec-
Studies of the SFR flickering have experienced a recent tral index n = 0.967, and σ8 = 0.826 (Planck Collaboration et al. 2014).
surge, as stochastic SFR variations can have an impact on our 2
Different assumptions on the integration radius produce only small
ability to observe distant galaxies (Sun et al. 2023a) and affect changes in the inferred metallicity.

Article number, page 2 of 13

 Pallottini et al.: The MZR as a ruler for galaxy evolution: insights from the JWST

the lensed galaxies from Chemerynska et al. (2024a) seem to in-

dicate lower values of Zg at M⋆ ∼ < 107 M ; the lensed data seems
⊙
consistent with serra, but only a handful of simulated galaxies
and observations are in that mass range.
As noted by Curti et al. (2024a), the lack of a clear trend and
the large scatter in the serra relation resemble the behavior of
their 6 < z < 10 sample. However clear differences are present,
such as the relatively metal rich galaxies at M⋆ ≃ 108 M⊙ and
low metallicity galaxies at M⋆ ≃ 1010 M⊙ . Further, the lack of
a clear trend might be partially due galaxy sample selection, as
most of the serra data has 108 M⊙ ∼ < M < 109 M and only a
⋆ ∼ ⊙
small fraction of galaxies has M⋆ ≃ 107 M⊙ . Thus, the observed
galaxy mass range is not uniformly covered by serra galaxies,
whose sample is drawn from a collection of zoom-in simulations
focusing on M⋆ ≃ 1010 M⊙ galaxies and their close environment.
Indeed, some of the M⋆ ∼ < 109 M mass galaxies form in a highly
⊙
pre-enriched environment, which explains their relatively high
metallicities (Zg ≃ 10−0.5 , see Gelli et al. 2020). Note that con-
Fig. 1: The mass-metallicity relation (MZR) at high-redshift (z). sidering different redshift intervals3 yields qualitatively similar
We plot the gas metallicity (Zg ) as a function of stellar mass (M⋆ ) results.
for the serra simulated galaxies (Pallottini et al. 2022, z ≃ 7) Thus, in summary, the normalization of the MZR of serra is
and the JWST observations from CEERS Nakajima et al. (2023, roughly similar to the observed galaxies, but no trend is clearly
z ≃ 4 − 10), JADES (Curti et al. 2024a, z ≃ 3 − 6 and 6 − visible and some discrepancies are present, especially in the
10), and UNCOVER (Chemerynska et al. 2024a, z ≃ 6 − 8). To M⋆ ≃ 1010 M⊙ range. Similar (or larger) tensions are present
guide the eye, the observational fits from Nakajima et al. (2023) also when a comparison with other models is performed. These
and Curti et al. (2024a) are also reported as lines with the same can be noted, e.g. from the fit obtained by the fire-2 simulations
colors of corresponding data. As a reference, the log(Zg /Z⊙ ) = (Marszewski et al. 2024). fire-2 data are close to the observed Zg
0.37 log(M⋆ /M⊙ ) − 4.3 fit from fire-2 simulations (Marszewski at M⋆ ≃ 109.5 M⊙ , but the predicted slope of the MZR is different,
et al. 2024, z = 5−12) is shown, along with their binned data and thus the tension with Nakajima et al. (2023); Curti et al. (2024a)
scatter at z = 8. The right axis reports the logarithmic oxygen increase with decreasing M⋆ . Finally, fire-2 galaxies present a
abundance [O/H] assuming a solar composition (Asplund et al. much higher scatter, that increases with decreasing stellar mass
2009). up to ≃ 1 dex at M⋆ ≃ 5 × 107 M⊙ .
An in depth comparison between observations and simula-
tions is presented in Curti et al. (2024a), which, in addition to
serra galaxies; then, we could infer M⋆ from synthetic SED fit- serra, considers the zoom-in simulations from firstlight (Lan-
ting and combine line emission to prepare metallicity calibra- gan et al. 2020, see Ceverino et al. 2017 for the main paper)
tors. Such an approach would reduce potential mismatches due and fire (Ma et al. 2016, see Hopkins et al. 2014), the Il-
to uncertainties in using different calibrators (Curti et al. 2017; lustrisTNG cosmological simulations Torrey et al. (2019, see
Chemerynska et al. 2024a), e.g. to relax the single zone model Pillepich et al. 2018), and the astraeus semi-analytical models
assumption (Marconi et al. 2024) and correct for biases in the (Ucci et al. 2023, see Hutter et al. 2021); as detailed in Curti
comparison (Cameron et al. 2023), also avoiding potential is- et al. (2024a), most simulations seem to reasonably match the
sues in the M⋆ determinations (Narayanan et al. 2024). While Zg data at ≃ 109 M⊙ but have steeper slopes and underpredict the
this kind of forward modeling approach is very powerful, also metallicities at lower M⋆ .
allowing to prepare novel observational strategies (Zanella et al. Interestingly, some of the simulations seem to be consistent
2021; Rizzo et al. 2022), here it hinders the possibility to com- with what observed by Chemerynska et al. (2024a), which how-
pare with most of the other models, as the MZR is usually com- ever have only eight targets, thus it is unclear if there is a change
of slope of the MZR for M⋆ ∼ < 107 M .
puted integrating metal, gas, and stellar masses; thus we avoid it ⊙
in the present work.
The prediction for serra are plotted in Fig. 1, along with 3. A minimal physical model to explain the MZR
JWST data from Curti et al. (2024a, JADES sample, Bunker
Given the very contrived match between simulations and obser-
et al. 2023), Nakajima et al. (2023, CEERS sample, Finkelstein
vations, it seems worthwhile to step back and try to understand
et al. 2023), and Chemerynska et al. (2024a, UNCOVER sam-
the trend and dispersion of the MZR using basic physical mod-
ple, Bezanson et al. 2022), for a total of about 180 galaxies at
< M < 1010 M . The bulk of serra els. The advantage of the approach consists in simplifying the
z ≃ 3 − 10 with 106 M⊙ ∼ ⋆ ∼ ⊙
7 < M < 109 M ; these galaxies have complexity of numerical simulation while retaining the most im-
galaxies is between 10 M⊙ ∼ ⋆ ∼ ⊙
< Z < 10−0.5 Z , which is broadly consistent with the portant physical processes shaping the MZR.
10−1.5 Z⊙ ∼ g ∼ ⊙
< M < 108 M To this aim we devise a minimal, physical model describing
observed JWST galaxies. However, at 107 M⊙ ∼ ⋆ ∼ ⊙ the evolution of DM (Mdm ), gas (Mg ), star (M⋆ ), and gas-phase
some serra galaxies show Zg ≃ 0.25 dex higher than the ob-
metal (MgZ ) mass of a galaxy, following Dayal et al. (2013). As
served vales; moreover, for M⋆ ≃ 1010 M⊙ , Zg is on average in Furlanetto & Mirocha (2022), we introduce an explicit feed-
lower than observed, up to extreme cases of 0.5 dex. In general, back delay resulting in a modulation of the star formation history
serra galaxies show no clear MZR trend and their dispersion ap- (SFH).
pears to be somewhat larger than the observed one. Interestingly,
with respect to Nakajima et al. (2023) and Curti et al. (2024a), 3
We have checked the serra data at z = 7 , 8, and 9.

Article number, page 3 of 13

 A&A proofs: manuscript no. metal_jwst

3.1. Model setup

We assume that the DM halo increases as a result of cosmologi-
cal accretion, which on average can be written as in Correa et al.
(2015, see eq. 23 therein)
! !
Mdm 0.7
Ṁdm = 71.6 E(z) M⊙ yr−1 , (1a)
1012 M⊙ h

where E(z) = [−0.24 + 0.75(1 + z)] Ωm (1 + z)3 + ΩΛ . We allow

p
stars to form (eq. 2b) and include stellar feedback via SNe, ex-
ploding after a delay time td . SN cause mass outflows with a rate
Ṁout (eq. 2a), enrich the gas of metals with a yield y, and have
a return fraction R. Thus, the stellar mass increases because of
SFR and decreases after td because of SN explosions:
Ṁ⋆ = SFR(t) − R SFR(t − td ) , (1b)
which for td = 0 (no-SN-delay) would give an instantaneous
recycling approximation as in Dayal et al. (2013). Fig. 2: Example of the impact of SFR burstiness on the evolution
The gas mass increases because of cosmic accretion and re- of a galaxies hosted in a Mdm = 3 × 1010 M⊙ DM halo at z = 6
turn from stars, and it decreases because of the SFR process and using the minimal physical model described in Sec. 3.1 (see in
SN driven outflows: particular eq.s 1). The time evolution of the dark matter (DM,
Ṁg = fb Ṁdm (t) + R SFR(t − td ) − SFR(t) − Ṁout (t − td ) , (1c) black), star (green), gas (brown), and metal mass (M) content is
shown with solid (dashed) lines for a SN delay time td = 10 Myr
where fb = Ωb /Ωm is the cosmological baryion fraction. Finally, (td = 0, i.e. no delay). The upper axis shows the redshift z corre-
the gas metal content increases because of the SN yields while it sponding to cosmic time t.
decreases because of astration and outflows:
ṀgZ = fb Zin Ṁdm (t)+y SFR(t−td )−Zg [SFR(t)− Ṁout (t−td )] , (1d) limits of the depletion times reported by Dessauges-Zavadsky
et al. (2020) for ALPINE galaxies, and it is discussed in more
where Zg = MgZ /Mg is the gas metallicity, and we include the detail in Sec. 4.2 (see also App. A).
possibility for the cosmic gas to be already enriched at Zin , which To complete the model, we assume that halos can form stars
we set to Zin = 10−7 Z⊙ , i.e. virtually metal free. after they reach a virial temperature for atomic cooling to be
To solve the system in eq.s 1, we need to specify a functional effective (104 K); similarly to serra, we assume return fraction
form for Ṁout and SFR. We take the outflow rate as in Muratov and yields appropriate for a Kroupa (2001) IMF and adopting
et al. (2015): Bertelli et al. (1994) tracks with a Z⋆ = Z⊙ stellar population5 ,
!−0.35 which gives y ≃ 0.0228 and R ≃ 0.3242.
M⋆
Ṁout = 3.6 ϵSN SFR M⊙ yr−1 , (2a) The system (eq.s 1 and 2) is solved along with z = z(t) us-
1010 M⊙ ing an embedded 5th/6th order Runge-Kutta method (Fehlberg
with ϵSN an efficiency parameter that we use to calibrate the mass 1968) with adaptive timestep, which is selected in order to have
loading factor; ϵSN = 1 correspond to the results from Mura- a fractional (absolute) precision of at least ≤ 10−5 (≤ 102 M⊙ ) for
tov et al. (fire simulations, Hopkins et al. 2014 2014). Note that Mdm , M⋆ , Mg , and MgZ . The delayed SFR and stellar mass used
ϵSN = 1 yields a loading factor consistent with what Herrera- to evaluate the stellar feedback are computed using a 7 points
Camus et al. (2021) observes for HZ4 (but see Parlanti et al. time stencil.
2024). Further, using ϵSN = 1 in eq. 2 would match the results
from Pizzati et al. (2023) for z ≃ 5 − 6 ALPINE galaxies (Le 3.2. Overview of the minimal physical model
Fèvre et al. 2020) which feature a spatially extended gas com-
ponent (Fujimoto et al. 2020). However, ϵSN ≃ 1/3 is needed to The delay between SFR and feedback td and the amplitude of the
match the loading factor inferred from z ≃ 3 − 9 JADES galaxies outflow (ϵSN ) regulate the burstiness of the galaxy in our single
(Carniani et al. 2024). zone model (cfr. Furlanetto & Mirocha 2022). To understand the
We assume that the SFR is proportional to the gas mass: impact of burstiness on galaxy evolution, it is instructive to con-
sider a halo with mass Mdm = 3 × 1010 M⊙ at z = 6 and set the
SFR =
Mg
, (2b) outflow efficiency to the standard value of ϵSN = 1.
t⋆ In Fig. 2 we compare the evolution of the stellar, gas, and
metal mass content for td = 0 (no-SN-delay, dashed lines) and
and we select a constant t⋆ = 1 Gyr as our fiducial star formation td = 10 Myr (solid lines). By construction, the halo mass growth
time scale4 . The choice of t⋆ is roughly consistent with the upper (eq. 1a) is unaffected by td , thus the two models follow the same
4
In the model, galaxy evolution is self-regulated by feedback, thus the DM evolution, which smoothly grows from Mdm ≃ 107 M⊙ at z ≃
sensitivity of the results from t⋆ is limited in our case, and null in the 18, when the universe was t ≃ 200 Myr old. At that time, for td =
td = 0 case with constant loading factors for infall and outflow (Lilly 0 the gas and stars has ≃ 105 M⊙ and ≃ 102 M⊙ , respectively, and
et al. 2013). A posteriori, we note that the fiducial t⋆ = 1 Gyr gives
specific star formation rates sSFR = SFR/M⋆ which are consistent with 5
Changing the metallicity of the stellar population does modifies the
serra galaxies, i.e. ≃ sSFR ≃ 100 − 1 Gyr−1 for galaxies with masses yield and return fraction, but gives only minor modifications to the up-
M⋆ = 106 − 1010 M⊙ . coming results.

Article number, page 4 of 13

 Pallottini et al.: The MZR as a ruler for galaxy evolution: insights from the JWST

Similarly to the flickering, we define the analog variable for

the gas metallicity
Zg
δZ ≡ log , (3b)
⟨Zg ⟩

where log⟨Zg /Z⊙ ⟩ is a second order polynomial fit in the variable

log M⋆ /M⊙ . For both δ and δZ we can define the typical varia-
tion as the root mean square (r.m.s.) deviation, i.e. σSFR and σZ ,
respectively.
Adopting eq.s 3, we find that the td = 10 Myr case has a
flickering σSFR ≃ 0.3 before the two models start to converge,
which is slightly higher than the σSFR ≃ 0.24 obtained for serra
galaxies (Pallottini & Ferrara 2023). For different td the quali-
tative behavior is very similar, as the model is off-balance and
oscillates around the no-SN-delay case, that is similar to a bath-
tub solution (Bouché et al. 2010; Dekel & Mandelker 2014). For
shorter (longer) td , σSFR can be smaller (higher), as galaxy with
Fig. 3: Example of the impact of SFR burstiness on the MZR a smaller (higher) M⋆ can be off-balanced.
for a system with Mdm = 3 × 1010 M⊙ at z = 6 evolved with our The balance between infall and outflow rates is critical, since
minimal physical model. Different lines indicates different delay the dynamical system (eq.s 1) modeling the galaxy evolution
times for the SN feedback, as indicated in the legend. Note that give rise to exponentially increasing/decreasing trends; introduc-
the model results have been resampled in 2 Myr steps and we do ing a td , 0 can efficiently break the balance, causing cycles of
not consider the evolutionary phases when the galaxies have a complete gas and metal depletion.
negligible amount of gas (Mg ≤ 10−3 Mdm ). To explore the MZR variations induced by the flickering, in
Fig. 3 we plot the tracks of the selected Mdm = 3 × 1010 M⊙ at
z = 6 with td from 5 to 25 Myr. Because of the longer td , the SFR
both grow reaching approximately ≃ 5 × 107 M⊙ at t ≃ 800 Myr, becomes increasingly bursty. Specifically, for the selected model
with M⋆ /Mdm ≃ 5 × 10−3 broadly consistent with what expected the flickering ranges from σSFR ≃ 0.1 to σSFR ≃ 0.6 for td = 5
from abundance matching models (e.g. Behroozi et al. 2013). Myr to td = 25 Myr, respectively.
In the case with td = 10 Myr, the secular trends are qual- The the MZR modulations induced by the flickering are
itatively similar, but the system experiences a series of bursts highly non linear. With td = 5 − 10 Myr, the MZR dispersion
on time scales of ≃ 20 Myr = 2 td . Let us focus on a single is negligible-small (σZ ≃ 0.05 − 0.2), as the flickering can affect
burst cycle around z ≃ 10, when the halo has Mdm ≃ 109 M⊙ galaxies only up to M⋆ ≃ 105−7 M⊙ . For td = 15 Myr, the full M⋆
and M⋆ ≃ 106 M⊙ . In the first half period of the cycle (0 − td ), range experience flickering with a σSFR ≃ 0.5, yielding an high
cosmic accretion quickly replenish the gas (gas infall rate is σZ ≃ 0.8. Cases with td > 15 Myr do not show a higher stochas-
≃ 0.5 M⊙ yr−1 from eq. 1a), reaching Mg ≃ 107 M⊙ in a few ticity, as an increasing delay only regulates which M⋆ can be
Myr. Stars form at a modest rate SFR ≃ 10−2 M⊙ yr−1 (eq. 2b) affected by the feedback unbalance6 . Qualitatively, the behavior
under feedback-free conditions (cfr. Dekel et al. 2023). After td , is similar with higher Mdm , but the convergence to the bathtub
the feedback kicks in, first enriching the galaxy and eventually solution is faster.
(≃ 2 td ) becoming powerful enough to expel both gas (gas out- In summary, a feedback delay induces a stochastic, bursty
flow is ≃ 1M⊙ yr−1 using eq. 2a) and metals (cfr. Ferrara et al. star formation behavior that becomes progressively more en-
2023). However, in the example the impact of SFR flickering is hanced as td is increased. Such SFR flickering corresponds to
limited already after z ≃ 15, when the galaxy has M⋆ ≃ 105 M⊙ analog Zg fluctuations which act to decrease the mean metallic-
or 10× lower if no delay is considered. At z ≃ 6, the final stellar ity as gas and metals are effectively ejected from the galaxy by
mass is M⋆ ≃ 107 M⊙ within a factor of 2 for the two models. For outflows.
the gas and metals, the model with SN delay starts to converge to
the no-SN-delay case at z ≃ 8, when M⋆ ≃ 107 M⊙ ; oscillations 3.3. The impact of a stochastic SFR on the MZR
for the gas and metals are still present up to z ≃ 7, but they are
milder. We run sets of models that at z = 6 have a Mdm from 109 M⊙ to
As in Pallottini & Ferrara (2023), it is convenient to quantify 1013 M⊙ with a logarithmic binning of 0.1 dex, in order to cover
the mass range 105 M⊙ ∼< M < 5×1010 M . We analyze different
the burstiness of the SFR by defining the variable δ expressing ⋆ ∼ ⊙
the log of the stochastic SFR variation with respect to its mean variations of the minimal models, reporting the results in differ-
(in short: flickering) ent panels of Fig. 4, where they are compared with data from
JWST observations (Nakajima et al. 2023; Curti et al. 2024a;
SFR Chemerynska et al. 2024a).
δ ≡ log , (3a)
⟨SFR⟩
Base models
where log⟨SFR/M⊙ yr−1 ⟩ is a second order polynomial fit in the Let us start by discussing the base models analyzed in Sec. 3, i.e.
time variable t. The low order fit is needed to factor out the with ϵSN = 1, consistent with fire simulations (Muratov et al.
smooth behavior of the functions without removing time mod-
ulations (see Leja et al. 2019; Chaves-Montero & Hearin 2021, 6
Note that an higher/lower level of flickering can be induced by chang-
for alternative methods). ing the infall/outflow rates, as shown later in Sec. 3.3.

Article number, page 5 of 13

 A&A proofs: manuscript no. metal_jwst

Fig. 4: Predicted MZR relation for different sets of minimal models with 109 M⊙ ≤ Mdm ≤ 1013 M⊙ halos at z = 6. In the upper left
panel we report the base models with ϵS N and td = 20 Myr (see Sec. 3.2). In the upper right panel we modify the base model by
reducing the efficiency of the feedback (ϵSN = 1/4 in eq. 2a). In the lower left panel we consider both weak feedback and a cosmic
accretion that is modulated on a time scales of 100 Myr (A100 = 1/3 in eq. 4). In the lower left panel we show the effect of turning
off the SN-delay for the weak feedback models with modulated cosmic accretion. In each panel, pastel lines are the individual tracks
sampled every 2 Myr; their median and 16% − 84% dispersion is plotted with violet dashed and solid lines, respectively; the dashed
black line is the control case with no-SN-delay (td = 0) and no modulated accretion (A100 = 0). For each model, we report the value
values of the SFR flickering (σSFR ) and the MZR dispersion (σZ , see Fig. 5). As in Fig. 1, we show JWST data and fits (Nakajima
et al. 2023; Curti et al. 2024a; Chemerynska et al. 2024a).

2015), and selecting td = 20 Myr, so that most of the explored extreme, as only part of the lensed galaxy data can be matched
stellar mass range is expected to have a stochastic SFR. (Chemerynska et al. 2024a).
In summary, the efficient feedback (ϵSN = 1) combined with
In the upper left panel of Fig. 4 we report both the indi- a td = 20 Myr generates a SFR flickering of σSFR ≃ 0.5, in-
vidual tracks of the base models and their M⋆ average trends. ducing MZR variations of σZ ≃ 1.4 resulting in a “chemical
For the base models, the median case falls significantly below chaos” that is not observed. Barring the option that observations
the observed average MZR. Also, the dispersion is much larger are tracing only the tip of the Zg -M⋆ relation, in order to explain
than seen in JWST data, that can be barely matched by the up- the observed MZR we modify the base models as follows.
per envelope of the predicted curves. The base models have a
σSFR ≃ 0.5 flickering, that induces a σZ ∼> 1 almost in the entire
< 9
Weak feedback
stellar mass range (M⋆ ∼ 10 M⊙ , Fig. 5). The flickering is due
to cycles of infall/outflow which determine an oscillatory behav- First, we reduce the impact of stellar feedback by decreasing the
ior of the evolution around the td = 0 control case, that is largely outflow rate from Muratov et al. (2015) (eq. 2a), i.e. setting ϵSN =
independent of the hosting Mdm . The Zg -M⋆ normalization is de- 1/4; such a loading factor is closer to what Carniani et al. (2024)
termined by the strength of the outflows, that completely devoids < M < 109 M JADES galaxies. As shown in the
infer for 107.5 ∼ ⋆ ∼ ⊙
the galaxy of gas and metals, not allowing the metallicity to rise upper right panel of Fig. 4, this modification increases the aver-
to the observed level. Even for td = 0 the outflow seems too age metallicity of the modeled galaxies, matching the observed
Article number, page 6 of 13
 Pallottini et al.: The MZR as a ruler for galaxy evolution: insights from the JWST

with Ṁdm being the average cosmic accretion from eq. 1a and
with A100 scaling the amplitude of the modulation7 .
As feedback determines the mean trend of the MZR, we keep
the reduced outflow efficiency (ϵSN = 1/4 in eq. 2a) and compute
the evolution of a minimal model by setting the amplitude of the
oscillation to A100 = 1/3, which should yield a scatter below the
0.3 dex dispersion expected for the distribution of growth rates
of DM halos (Rodríguez-Puebla et al. 2016; Ren et al. 2019;
Mirocha et al. 2021). The results from these modified models
are shown in the lower left panel of Fig. 4: both the mean trend
and scatter of the observed MZR are recovered (Fig. 5). Such
a matching set of modified models yields a roughly constant
σSFR ≃ 0.2 and σZ ≃ 0.3, satisfactorily matching the JWST
data.

Weak feedback, modulated accretion, no-SN-delay

As a final check, in the lower right panel of Fig. 4 we show the ef-
fect of removing the SN-delay from the best matching model just
discussed. While the mean MZR trend is recovered, the metal-
licity dispersion (σZ ≃ 0.05) is much smaller than observed
(σZ ≃ 0.25). Also, the median of the models fails to reproduce
the relation for most of the low mass lensed galaxies (Chemeryn-
ska et al. 2024a).
Fig. 5: SFR (σSFR , upper panel) and gas metallicity (σZ , lower With no-SN-delay, the SFR flickering is induced only by
panel) r.m.s. variations as a function of stellar mass for the sets modulation of the cosmic accretion, which fails to unbalance the
of models presented in Fig. 4 (see eq.s 3 for the definitions). As a system through the ejection of a substantial fraction of the con-
reference, we report the serra average σSFR (Pallottini & Ferrara tained gas. With the selected A100 = 1/3 and td = 0, the model
2023) and σZ (this work), the σSFR from Ciesla et al. (2024); Ma- reach σSFR ≃ 0.1 for M⋆ ∼ < 108 M , while the flickering gets to
⊙
son et al. (2023); Muñoz et al. (2023); Shen et al. (2023), and the σSFR ≃ 0.2 only for M⋆ ≃ 5 × 108 M⊙ galaxies.
average σZ for JWST data. Note that ⟨JWST⟩ is the r.m.s. of the Compared to the control case (A100 = 0), the average Zg is
observations after subtraction of the fit to the MZR in the differ- slightly higher; practically, SFR flickering induced by the accre-
ent data sets in Nakajima et al. (2023) and Curti et al. (2024a); tion modulation can increase Zg since a higher gas mass can be
the thickness of the shaded area encloses the min/max of the converted into stars without efficiently ejecting the gas via out-
r.m.s. of the different data sets. flows, i.e. the opposite outcome with respect to SFR flickering
induced via SN-delay. Interestingly, despite the relatively high
< M < 1010 M mass range. SFR flickering (average of σSFR ≃ 0.2), the metallicity variation
mean trends in the whole 106 M⊙ ∼ ⋆ ∼ ⊙
is the lowest of all considered cases (σZ ≃ 0.05): only a delayed
Comparing to the td = 0 control case, it is evident that the out-
SN feedback can efficiently off-balance the system and induce
flow efficiency is the main driver of the mean Zg -M⋆ trend.
modulations on both SFR and Zg .
Note that, reducing the outflow efficiency while maintaining
We note that the serra cosmological simulations show a rel-
a fixed td = 20 Myr results in σZ ≃ 0.1 modulation of the MZR,
atively modest SFR flickering (σSFR ≃ 0.24, Pallottini & Ferrara
i.e. too low with respect to the observed (σZ ≃ 0.25, see Fig. 5).
2023), and close to the expectations for z ≃ 6 galaxies (Muñoz
This effect is induced by the suppression of the amplitude of the
et al. 2023; Ciesla et al. 2024). Yet, given the sharp sensitivity of
SFR flickering, especially at the high mass end.
σZ to SFR variations, such time variability results in a σZ larger
Indeed the models show a small flickering, with σSFR ≃ 0.15
< 108 M , and even slightly lower values for higher stel- than observed.
for M⋆ ∼ ⊙
lar masses (Fig. 5). As discussed in Pallottini & Ferrara (2023), To summarize, our analysis shows that: (i) a weak feedback,
M⋆ ∼ < 109 M galaxies show SFR fluctuations with characteristic and (ii) a relatively mild stochasticity given by (iii) SN-delay
⊙
periods of ≃ 20 − 40 Myr, which correspond to the typical evo- combined with a long term accretion rate modulation are critical
lutionary times of SNe. Such flickering timescale increases to to simultaneously match both the mean and dispersion of the ob-
≃ 80 − 150 Myr in more massive galaxies which are more sensi- served MZR relation in the entire stellar mass range of galaxies
tive to mergers and accretion rate modulation. The latter effects sampled by JWST at z = 3 − 10.
are not accounted for in our minimal model (eq. 1a). The next Finally, we note that invoking stochasticity to explain the
step is then to heuristically incorporate them in our treatment. overabundance of bright JWST galaxies requires σSFR > 0.5
(more specifically, σSFR ≃ 0.8, σSFR ≃ 0.7, and σSFR ≃ 0.6 ac-
cording to Muñoz et al. 2023, Shen et al. 2023, and Mason et al.
Weak feedback, modulated accretion 2023, respectively; see Fig. 5). However, such extreme flickering
amplitudes are (i) higher than that recovered from SED fitting al-
To incorporate the extra stochastic behavior expected from vari-
ation in the cosmic accretion in a simple manner, we impose a 7
Note that small halos should be relatively less affected by modula-
sinusoidal modulation on a 100 Myr timescale tion on long timescales (≃ 100 Myr), since the hosted galaxies have
" !# shorter SFH. However, cosmological accretion variations can be coher-
t ently combined with the SN delay to give a higher σSFR also in the low
Ṁdm = Ṁdm 1 + A100 sin
mod
, (4) mass range.
100 Myr
Article number, page 7 of 13
 A&A proofs: manuscript no. metal_jwst

Violent outflows are present in fire-2 (Hopkins et al. 2018),

while the suppression of SFR due to the photo-dissociation of
molecular hydrogen in serra (Pallottini et al. 2019) can more
gently can modulate the SFR activity without drastically modi-
fying the galaxy dynamics.
Similar considerations hold when comparing successive im-
provements of simulations suites. For instance, the loading factor
from fire-2 (Pandya et al. 2021) are lower by a factor of ≃ 2 with
respect to fire (Muratov et al. 2015); this difference dictates that
the MZR from fire-2 (Marszewski et al. 2024) is about 0.3 dex
higher than the MZR from fire (Ma et al. 2016); in both cases the
slope is steeper (normalization is lower) than what observed by
Nakajima et al. (2023); Curti et al. (2024a) and is instead more
consistent with Chemerynska et al. (2024a). Despite the simplic-
ity of the minimal model presented, hints of such differences in
the simulations can be highlighted and explained.
However, it is difficult to directly quantify the differences
in the simulations with the presented models. For instance, the
Fig. 6: Deviation from the fundamental mass-metallicity relation loading factor in our simplified models (but also in relatively
(FMR) for high-z galaxies. We adopt the µ ≡ log(M⋆ /M⊙ ) − more complex ones as Thompson et al. 2016 and Pizzati et al.
0.65 log(SFR/M⊙ yr−1 ) FMR obtained by Curti et al. (2020) 2020) is a scalar quantity used as an input value for each galaxy,
for the high sSFR galaxies sub-sample from the SDSS (z ∼ < 2 while in simulations is usually defined as an output that has a
York et al. 2000). We report the results of the weak feedback spatial dependence (Gallerani et al. 2018).
(ϵSN = 1/4), modulated accretion (A100 = 1/3) models along Additionally, when dealing with simulation snapshots, quan-
with the data from JWST observations (Nakajima et al. 2023; tifying both the SFR event and the consequent outflows is
Curti et al. 2024a; Chemerynska et al. 2024a) and serra simula- tricky because of the time-delay between SFR and feedback
tions (Pallottini et al. 2022). (Pandya et al. 2021), particularly for galaxies far from a quasi-
equilibrium situation.
Further, the simplifications done in our minimal model im-
< 0.5, Ciesla et al. 2024), and
lowing for SFH variability (σSFR ∼ pose we can only capture part of the complexity present in sim-
> 1.4) that is not present
(ii) would induce a chemical chaos (σZ ∼ ulations. For instance, while the SN-feedback delay is uniquely
in the observed MZR at z = 3 − 10 (σZ ≃ 0.25 Nakajima et al. defined, simulations finds a distribution of td that depend on the
2023; Curti et al. 2024a; Chemerynska et al. 2024a). underlying physical models included (Pallottini & Ferrara 2023);
thus, trying to find the set of parameters of a minimal model that
gives the best match to a simulation will only give effective pa-
4. Discussion rameters, a sort of summary statistic that can be used to compare
different simulations.
The minimal models presented here can also be useful as a tool Apart from these difficulties, a systematic parameter fitting
to guide the development and compare complex cosmological procedure should give a better quantification of the comparison
simulations, particularly when focusing on relations that - like between different simulations, and can also be used to enable
the MZR - are very sensitive to small variations of the underly- an even better match between the data and the minimal models.
ing physics, that is included via sub-grid models in galaxy sim- However, such models are lacking a few ingredients that pre-
ulations (Agertz et al. 2020). vents reproducing the full connection between metal build up
and star formation history of galaxies.
4.1. Comparing numerical simulations
In galaxy simulations, ejective feedback, such as outflows 4.2. Caveats of the minimal model
caused by mechanical/kinetic prescriptions, can effectively sup- The limitations of the presented minimal models can be appre-
press the SFR activity as much as preventative feedback, such as ciated by considering the fundamental mass-metallicity relation
turbulence/delayed cooling prescriptions (Rosdahl et al. 2017), (FMR), i.e. the link between Zg , M⋆ , and SFR that is observed to
leading to similar SFR histories when the same initial condi- be redshift independent up to z ∼ 3 (Mannucci et al. 2010; Curti
tions are considered (Lupi et al. 2020). Despite similar SFR his- et al. 2020) and seems to break down at z ∼ > 5 (Nakajima et al.
tories, other properties can differ significantly (Rosdahl et al. 2023; Curti et al. 2024a). We adopt the parameterization from
2017), such as the dynamical state of a galaxy. For instance, Curti et al. (2020)
Fujimoto et al. (2024) show that fire-2 galaxies (Wetzel et al.
2023) have a much lower incidence of disks compared to serra µ ≡ log(M⋆ /M⊙ ) − 0.65 log(SFR/M⊙ yr−1 ) , (5)
galaxies (Kohandel et al. 2024). Additionally, SERRA galaxies
tend to have higher Zg and lower MZR scatter compared to fire- which is obtained by minimizing the Zg dispersion for the high
2 (Marszewski et al. 2024), especially in the M⋆ ≃ 109 M⊙ mass sSFR sub-sample from the Sloan Digital Sky Survey (SDSS,
range. < 2).
York et al. 2000, z ∼
Both effects are likely caused by the higher SFR flickering of In Fig. 6 we show the FMR for the minimal models matching
fire-2 (Sun et al. 2023a) with respect to serra (Pallottini & Fer- the mean MZR and its dispersion, i.e. the weak feedback (ϵSN =
rara 2023), which is ultimately regulated by the efficiency and 1/4) set with modulated accretion (A100 = 1/3). These models
kind (e.g. ejective vs preventative) of included stellar feedback. < 7, with the
are below the SDSS relation by about 0.5 dex at µ ∼
Article number, page 8 of 13
 Pallottini et al.: The MZR as a ruler for galaxy evolution: insights from the JWST

separation that decreases with increasing µ, down to 0.1 dex at (2023), Curti et al. (2024a), and Chemerynska et al. (2024a),
µ∼ > 9, almost connecting with the FMR trend observed at low and obtained from the CEERS (Finkelstein et al. 2023), JADES
z. As expected (Nakajima et al. 2023; Curti et al. 2024a), JWST (Bunker et al. 2023), and UNCOVER (Bezanson et al. 2022) sur-
galaxies are offset below the SDSS relation. The best matching veys, respectively.
models for the MZR only partially match the JWST data in the We compared the observations with the predictions from the
FMR plane; the minimal models are closer to the JWST data serra simulations (Pallottini et al. 2022), finding a broad agree-
when considering the lensed galaxies from Chemerynska et al. ment with the observed MZR. However, simulations show a lack
(2024a). of a clear metallicity trend, and some tension, particularly at
Further the minimal models only produce downward devi- M⋆ ≃ 1010 M⊙ .
ation with respect to the FMR from the SDSS. Instead, serra To better understand the observed high-z MZR and to clar-
galaxies manage to better recover the trend for JWST galax- ify the behaviour of the simulations, we have devised a minimal
ies around µ ≃ 8, reproducing both downward and upward physical model for galaxy evolution in which star formation is
deviations with respect to the FMR from the SDSS; however, fueled by cosmic gas accretion (Correa et al. 2015) and is regu-
serra galaxies show an overabundance of low Zg galaxies at lated by supernova driven outflows (eq. 2a), with a loading factor
µ ≃ 8.5 − 9.0, which is not seen in the JWST data. from Muratov et al. (2015) that is controlled by an efficiency ϵSN ;
In an equilibrium model (Lilly et al. 2013), the shape of the additionally, we incorporated an explicit delay (td ) between star
MZR is independent from the SFR, which uniquely determines formation and supernova feedback, inducing a stochastic SFR
the speed a galaxy can climb up the Zg -M⋆ curve. Note that the behavior. Some models also explore the possible modulation of
simplified model seems to deviate more from the data at low cosmic accretion (eq. 4). s are:
µ, i.e. a combination of relatively high SFR and low M⋆ ; the
adopted feedback increases in efficiency for lower M⋆ (eq. 2a). • To recover the average trend of the MZR observed at high-z,
Thus, the observed FMR deviation might be recovered by the outflows must be less efficient (ϵSN = 1/4) than predicted in
minimal model if we consider not only ejecting but also a pre- Muratov et al. (2015). However, this evidence is consistent
venting feedback, e.g. photodissociation of molecular hydrogen, with the mass loading factors inferred for JADES galaxies
which can suppress the star formation without affecting the gas (Carniani et al. 2024).
mass (e.g. see Alyssum, a serra galaxy described in Pallottini • To match the MZR dispersion in the full stellar mass range,
et al. 2022) and can play a role in explaining some of the bright both a delayed (td = 20 Myr) SN feedback, and a modulation
blue monsters seen by JWST (Ferrara 2024). of the cosmic accretion (A100 = 1/3) are required.
As an alternative to keep matching the MZR and recover- • The r.m.s. variation of the MZR (σZ ) is very sensitive to
ing the offset from the FMR, we could try to change the sim- the SFR flickering (σSFR ). The weak feedback necessary
plified SFR prescription (eq. 2b); this can be done for instance to reproduce the average MZR trend (ϵSN = 1/4) also re-
by adopting as the time scale for star formation a redshift depen- sults in a low-amplitude SFR flickering (σSFR ≃ 0.2). This
dent depletion time (Tacconi et al. 2010, 2020; Sommovigo et al. prescription correctly reproduces the moderate MZR scatter
2022), which are typically shorter than the selected t⋆ = 1 Gyr (σZ ≃ 0.25) observed by JWST.
by about a factor ≃ 10, similar to what Vallini et al. (2024) report • serra galaxies have slightly higher σSFR ≃ 0.24 (Pallottini &
for z ≃ 7 galaxies by using glam (Vallini et al. 2020, 2021) de- Ferrara 2023), resulting in a metallicity scatter of σZ ≃ 0.3,
terminations, thus promoting a faster galaxy evolution at high-z. higher than observed. Differently from the analytical model,
Indeed, by adopting the depletion time from Sommovigo serra galaxies also overpredict the number of low metallicity
et al. (2022) as the time scale for SFR (see App. A, in partic- (Zg ≃ 0.2Z⊙ ) galaxies at M⋆ ≃ 1010 M⊙ .
• In general, any model predicting σSFR ∼ > 0.5 is likely to over-
ular Fig. A.1), the stellar mass build up is faster at M⋆ ∼< 106 M ,
⊙
shoot the observed metallicity r.m.s. scatter, possibly lead-
while later saturates to values similar to those resulting from the > 1.4) that is it not present in
ing to a “chemical chaos” (σZ ∼
t⋆ = 1 Gyr case as both SFHs become feedback regulated, ap-
JWST data, for which σZ ≃ 0.25).
proaching a bathtub equilibrium.
However, on the one hand, the modifications affect the pre- The last point also entails that simultaneously explaining the
dictions of the minimal model mostly in the M⋆ ∼ < 106 M range;
⊙ observed MZR (Nakajima et al. 2023; Curti et al. 2024a) and the
on the other hand, a z average depletion time is sensitive to overabundance of bright galaxies observed by JWST (Finkel-
the population of the sample, e.g. for main sequence is ≃ ×5 stein et al. 2022; Naidu et al. 2022) through high SFR flicker-
lower than for starburst (Tacconi et al. 2020), and the results ing (Mason et al. 2023; Mirocha & Furlanetto 2023; Shen et al.
from Sommovigo et al. (2022) are mostly relying on the data 2023; Sun et al. 2023b; Kravtsov & Belokurov 2024) is very
> 108−10 M galaxies from the ALPINE (Le Fèvre et al.
of M⋆ ∼ ⊙ challenging, considering that a σSFR ≃ 0.8 is required to explain
2020) and REBELS (Bouwens et al. 2022) surveys. To summa- z > 10 observations (Muñoz et al. 2023). We note that while
rize, while trying to reconcile the FMR behavior is an interest- both the SN-feedback delay and cosmic accretion modulation
ing perspective and there are a few options, some care should be induce SFR flickering, the latter tends to give a gentler modula-
taken in extending/modifying the prescriptions for the minimal tion of Zg , since high feedback efficiency and a delay are needed
models; we leave such a possibility for a future work. in order to off-balance the MZR generated from a bathtub-like
equilibrium (Lilly et al. 2013). However, variation of cosmic ac-
5. Conclusions cretion can only go up to 0.3 dex (Rodríguez-Puebla et al. 2016;
Ren et al. 2019; Mirocha et al. 2021). Thus, alternative explana-
The exquisite spectroscopic data collected by JWST allow for tions (Dekel et al. 2023; Ferrara et al. 2023) seem more favored
the first time to derive the mass-metallicity relation (MZR) in to solve the overabundance problem.
galaxies up to redshift z ≃ 10. While we showed that the presented minimal model can also
In this work, we considered the MZR data for a com- be used to guide our intuition in comparing complex cosmolog-
bined sample of about 180 galaxies with stellar mass ical simulations, we recall that such a simplified model does not
106 M⊙ ∼ < M < 1010 M at z = 3 − 10 given Nakajima et al. fully capture the physical complexity of the connection between
⋆ ∼ ⊙

Article number, page 9 of 13

 A&A proofs: manuscript no. metal_jwst

galaxy formation and metal enrichment, as highlighted by the Ferrara, A. 2008, in Low-Metallicity Star Formation: From the First Stars to
poor match with the JWST data when analyzing the off-set from Dwarf Galaxies, ed. L. K. Hunt, S. C. Madden, & R. Schneider, Vol. 255,
the fundamental mass-metallicity relation. Combining insights 86–99
Ferrara, A. 2024, arXiv e-prints, arXiv:2405.20370
from such models with further analysis of ongoing observations, Ferrara, A., Pallottini, A., & Dayal, P. 2023, MNRAS, 522, 3986
and more complex numerical simulations is crucial for under- Finkelstein, S. L. & Bagley, M. B. 2022, ApJ, 938, 25
standing the galaxy formation and evolution process at high-z. Finkelstein, S. L., Bagley, M. B., Ferguson, H. C., et al. 2023, ApJL, 946, L13
Finkelstein, S. L., Bagley, M. B., Haro, P. A., et al. 2022, ApJL, 940, L55
Acknowledgements. We acknowledge the CINECA award under the ISCRA ini- Förster Schreiber, N. M. & Wuyts, S. 2020, ARA&A, 58, 661
tiative, for the availability of high performance computing resources and support Fujimoto, S., Ouchi, M., Kohno, K., et al. 2024, arXiv e-prints,
from the Class B project SERRA HP10BPUZ8F (PI: Pallottini). AF acknowl- arXiv:2402.18543
edges support from the ERC Advanced Grant INTERSTELLAR H2020/740120 Fujimoto, S., Silverman, J. D., Bethermin, M., et al. 2020, ApJ, 900, 1
(PI: Ferrara). SC and GV acknowledge support support from the ERC Starting Furlanetto, S. R. & Mirocha, J. 2022, MNRAS, 511, 3895
Grant WINGS H2020/101040227 (PI: Carniani). Any dissemination of results Gallerani, S., Pallottini, A., Feruglio, C., et al. 2018, MNRAS, 473, 1909
must indicate that it reflects only the author’s view and that the Commission Gelli, V., Mason, C., & Hayward, C. C. 2024a, arXiv e-prints, arXiv:2405.13108
is not responsible for any use that may be made of the information it contains. Gelli, V., Salvadori, S., Ferrara, A., & Pallottini, A. 2024b, ApJ, 964, 76
Partial support (AF) from the Carl Friedrich von Siemens-Forschungspreis der Gelli, V., Salvadori, S., Ferrara, A., Pallottini, A., & Carniani, S. 2023, ApJL,
Alexander von Humboldt-Stiftung Research Award is kindly acknowledged. We 954, L11
gratefully acknowledge computational resources of the Center for High Perfor- Gelli, V., Salvadori, S., Pallottini, A., & Ferrara, A. 2020, MNRAS, 498, 4134
mance Computing (CHPC) at SNS. We acknowledge usage of the Python pro- Genzel, R., Förster Schreiber, N. M., Übler, H., et al. 2017, Nature, 543, 397
gramming language (Van Rossum & de Boer 1991; Van Rossum & Drake 2009), González, V., Labbé, I., Bouwens, R. J., et al. 2010, ApJ, 713, 115
Astropy (Astropy Collaboration et al. 2013), Cython (Behnel et al. 2011), Mat- Grassi, T., Bovino, S., Schleicher, D. R. G., et al. 2014, MNRAS, 439, 2386
plotlib (Hunter 2007), Numba (Lam et al. 2015), NumPy (van der Walt et al. Hahn, O. & Abel, T. 2011, MNRAS, 415, 2101
2011), pynbody (Pontzen et al. 2013), and SciPy (Virtanen et al. 2020). Harikane, Y., Ouchi, M., Oguri, M., et al. 2023, ApJS, 265, 5
Herrera-Camus, R., Förster Schreiber, N., Genzel, R., et al. 2021, A&A, 649,
A31
Hopkins, P. F., Kereš, D., Oñorbe, J., et al. 2014, MNRAS, 445, 581
Hopkins, P. F., Wetzel, A., Kereš, D., et al. 2018, MNRAS, 480, 800
References Hunter, J. D. 2007, Computing in Science Engineering, 9, 90
Adams, N. J., Conselice, C. J., Ferreira, L., et al. 2023, MNRAS, 518, 4755 Hutter, A., Dayal, P., Yepes, G., et al. 2021, MNRAS, 503, 3698
Agertz, O., Pontzen, A., Read, J. I., et al. 2020, MNRAS, 491, 1656 Kannan, R., Garaldi, E., Smith, A., et al. 2022, MNRAS, 511, 4005
Andrews, B. H. & Martini, P. 2013, ApJ, 765, 140 Katz, H., Kimm, T., Ellis, R. S., Devriendt, J., & Slyz, A. 2023, MNRAS, 524,
Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009, ARA&A, 47, 481 351
Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, Kennicutt, Jr., R. C. 1998, ApJ, 498, 541
A33 Kirby, E. N., Cohen, J. G., Guhathakurta, P., et al. 2013, ApJ, 779, 102
Atek, H., Shuntov, M., Furtak, L. J., et al. 2023, MNRAS, 519, 1201 Kobayashi, C. & Ferrara, A. 2024, ApJL, 962, L6
Behnel, S., Bradshaw, R., Citro, C., et al. 2011, Computing in Science Engineer- Kohandel, M., Pallottini, A., Ferrara, A., et al. 2024, A&A, 685, A72
ing, 13, 31 Kravtsov, A. & Belokurov, V. 2024, arXiv e-prints, arXiv:2405.04578
Behroozi, P. S., Wechsler, R. H., & Wu, H.-Y. 2013, ApJ, 762, 109 Kroupa, P. 2001, MNRAS, 322, 231
Bertelli, G., Bressan, A., Chiosi, C., Fagotto, F., & Nasi, E. 1994, A&A Supp., Krumholz, M. R. & Burkhart, B. 2016, MNRAS, 458, 1671
106, 275 Lam, S. K., Pitrou, A., & Seibert, S. 2015, in Proc. Second Workshop on the
Bezanson, R., Labbe, I., Whitaker, K. E., et al. 2022, arXiv e-prints, LLVM Compiler Infrastructure in HPC, 1–6
arXiv:2212.04026 Langan, I., Ceverino, D., & Finlator, K. 2020, MNRAS, 494, 1988
Bothwell, M. S., Maiolino, R., Kennicutt, R., et al. 2013, MNRAS, 433, 1425 Le Fèvre, O., Béthermin, M., Faisst, A., et al. 2020, A&A, 643, A1
Bouché, N., Dekel, A., Genzel, R., et al. 2010, ApJ, 718, 1001 Leja, J., Carnall, A. C., Johnson, B. D., Conroy, C., & Speagle, J. S. 2019, ApJ,
Bouwens, R. J., Smit, R., Schouws, S., et al. 2022, ApJ, 931, 160 876, 3
Bunker, A. J., Cameron, A. J., Curtis-Lake, E., et al. 2023, arXiv e-prints, Lilly, S. J., Carollo, C. M., Pipino, A., Renzini, A., & Peng, Y. 2013, ApJ, 772,
arXiv:2306.02467 119
Cameron, A. J., Katz, H., & Rey, M. P. 2023, MNRAS, 522, L89 Liu, B. & Bromm, V. 2020, MNRAS, 497, 2839
Carniani, S., Venturi, G., Parlanti, E., et al. 2024, A&A, 685, A99 Looser, T. J., D’Eugenio, F., Maiolino, R., et al. 2024, Nature, 629, 53
Castellano, M., Fontana, A., Treu, T., et al. 2022, ApJL, 938, L15 Lupi, A., Pallottini, A., Ferrara, A., et al. 2020, MNRAS, 496, 5160
Ceverino, D., Glover, S. C. O., & Klessen, R. S. 2017, MNRAS, 470, 2791 Ma, X., Hopkins, P. F., Faucher-Giguère, C.-A., et al. 2016, MNRAS, 456, 2140
Chaves-Montero, J. & Hearin, A. 2021, MNRAS, 506, 2373 Madau, P. & Dickinson, M. 2014, ARA&A, 52, 415
Chemerynska, I., Atek, H., Dayal, P., et al. 2024a, arXiv e-prints, Maiolino, R. & Mannucci, F. 2019, A&A Rev., 27, 3
arXiv:2407.17110 Maiolino, R., Nagao, T., Grazian, A., et al. 2008, A&A, 488, 463
Chemerynska, I., Atek, H., Furtak, L. J., et al. 2024b, MNRAS, 531, 2615 Mannucci, F., Cresci, G., Maiolino, R., Marconi, A., & Gnerucci, A. 2010, MN-
Ciesla, L., Elbaz, D., Ilbert, O., et al. 2024, A&A, 686, A128 RAS, 408, 2115
Correa, C. A., Wyithe, J. S. B., Schaye, J., & Duffy, A. R. 2015, MNRAS, 450, Marconi, A., Amiri, A., Feltre, A., et al. 2024, arXiv e-prints, arXiv:2401.13028
1521 Marszewski, A., Sun, G., Faucher-Giguère, C.-A., Hayward, C. C., & Feldmann,
Curti, M., Cresci, G., Mannucci, F., et al. 2017, MNRAS, 465, 1384 R. 2024, ApJL, 967, L41
Curti, M., Maiolino, R., Curtis-Lake, E., et al. 2024a, A&A, 684, A75 Mason, C. A., Trenti, M., & Treu, T. 2023, MNRAS, 521, 497
Curti, M., Mannucci, F., Cresci, G., & Maiolino, R. 2020, MNRAS, 491, 944 Mirocha, J. & Furlanetto, S. R. 2023, MNRAS, 519, 843
Curti, M., Witstok, J., Jakobsen, P., et al. 2024b, arXiv e-prints, Mirocha, J., La Plante, P., & Liu, A. 2021, MNRAS, 507, 3872
arXiv:2407.02575 Morishita, T., Stiavelli, M., Grillo, C., et al. 2024, arXiv e-prints,
Davies, F. B. & Furlanetto, S. R. 2016, MNRAS, 460, 1328 arXiv:2402.14084
Dayal, P. & Ferrara, A. 2018, Phys. Rep., 780, 1 Muñoz, J. B., Mirocha, J., Furlanetto, S., & Sabti, N. 2023, MNRAS, 526, L47
Dayal, P., Ferrara, A., & Dunlop, J. S. 2013, MNRAS, 430, 2891 Muratov, A. L., Kereš, D., Faucher-Giguère, C.-A., et al. 2015, MNRAS, 454,
Decataldo, D., Pallottini, A., Ferrara, A., Vallini, L., & Gallerani, S. 2019, MN- 2691
RAS, 487, 3377 Naidu, R. P., Oesch, P. A., van Dokkum, P., et al. 2022, ApJL, 940, L14
Dekel, A. & Mandelker, N. 2014, MNRAS, 444, 2071 Nakajima, K., Ouchi, M., Isobe, Y., et al. 2023, ApJS, 269, 33
Dekel, A., Sarkar, K. C., Birnboim, Y., Mandelker, N., & Li, Z. 2023, MNRAS, Narayanan, D., Lower, S., Torrey, P., et al. 2024, ApJ, 961, 73
523, 3201 Nikolić, I., Mesinger, A., Davies, J. E., & Prelogović, D. 2024, arXiv e-prints,
Dessauges-Zavadsky, M., Ginolfi, M., Pozzi, F., et al. 2020, A&A, 643, A5 arXiv:2406.15237
Dome, T., Tacchella, S., Fialkov, A., et al. 2024, MNRAS, 527, 2139 Pallottini, A. & Ferrara, A. 2023, A&A, 677, L4
Donnan, C. T., McLeod, D. J., Dunlop, J. S., et al. 2023, MNRAS, 518, 6011 Pallottini, A., Ferrara, A., Bovino, S., et al. 2017a, MNRAS, 471, 4128
Ellison, S. L., Patton, D. R., Simard, L., & McConnachie, A. W. 2008, ApJL, Pallottini, A., Ferrara, A., Decataldo, D., et al. 2019, MNRAS, 487, 1689
672, L107 Pallottini, A., Ferrara, A., Gallerani, S., et al. 2022, MNRAS, 513, 5621
Erb, D. K., Shapley, A. E., Pettini, M., et al. 2006, ApJ, 644, 813 Pallottini, A., Ferrara, A., Gallerani, S., Salvadori, S., & D’Odorico, V. 2014,
Fehlberg, E. 1968, Classical Fifth-, Sixth-, Seventh-, and Eighth-Order Runge- MNRAS, 440, 2498
Kutta Formulas with Stepsize Control Pallottini, A., Ferrara, A., Gallerani, S., et al. 2017b, MNRAS, 465, 2540

Article number, page 10 of 13

 Pallottini et al.: The MZR as a ruler for galaxy evolution: insights from the JWST

Pandya, V., Fielding, D. B., Anglés-Alcázar, D., et al. 2021, MNRAS, 508, 2979
Parlanti, E., Carniani, S., Venturi, G., et al. 2024, arXiv e-prints,
arXiv:2407.19008
Pillepich, A., Springel, V., Nelson, D., et al. 2018, MNRAS, 473, 4077
Pizzati, E., Ferrara, A., Pallottini, A., et al. 2020, MNRAS, 495, 160
Pizzati, E., Ferrara, A., Pallottini, A., et al. 2023, MNRAS, 519, 4608
Planck Collaboration, Ade, P. A. R., Aghanim, N., et al. 2014, A&A, 571, A16
Pontzen, A., Rovskar, R., Stinson, G. S., et al. 2013, pynbody: Astrophysics Sim-
ulation Analysis for Python, astrophysics Source Code Library, ascl:1305.002
Ren, K., Trenti, M., & Mason, C. A. 2019, ApJ, 878, 114
Rizzo, F., Kohandel, M., Pallottini, A., et al. 2022, A&A, 667, A5
Roberts-Borsani, G., Morishita, T., Treu, T., et al. 2022, ApJL, 938, L13
Rodríguez-Puebla, A., Primack, J. R., Behroozi, P., & Faber, S. M. 2016, MN-
RAS, 455, 2592
Rosdahl, J., Blaizot, J., Aubert, D., Stranex, T., & Teyssier, R. 2013, MNRAS,
436, 2188
Rosdahl, J., Schaye, J., Dubois, Y., Kimm, T., & Teyssier, R. 2017, MNRAS,
466, 11
Rowland, L. E., Hodge, J., Bouwens, R., et al. 2024, arXiv e-prints,
arXiv:2405.06025
Sanders, R. L., Shapley, A. E., Kriek, M., et al. 2018, ApJ, 858, 99
Santini, P., Fontana, A., Castellano, M., et al. 2023, ApJL, 942, L27
Sarmento, R., Scannapieco, E., & Cohen, S. 2018, ApJ, 854, 75
Schmidt, M. 1959, ApJ, 129, 243
Shen, X., Vogelsberger, M., Boylan-Kolchin, M., Tacchella, S., & Kannan, R.
2023, MNRAS, 525, 3254
Shibuya, T., Ouchi, M., & Harikane, Y. 2015, ApJS, 219, 15
Simons, R. C., Kassin, S. A., Weiner, B. J., et al. 2017, ApJ, 843, 46
Smit, R., Bouwens, R. J., Labbé, I., et al. 2014, ApJ, 784, 58
Sommovigo, L., Ferrara, A., Carniani, S., et al. 2022, MNRAS, 517, 5930
Stark, D. P., Schenker, M. A., Ellis, R., et al. 2013, ApJ, 763, 129
Sun, G., Faucher-Giguère, C.-A., Hayward, C. C., & Shen, X. 2023a, MNRAS,
526, 2665
Sun, G., Faucher-Giguère, C.-A., Hayward, C. C., et al. 2023b, ApJL, 955, L35
Tacconi, L. J., Genzel, R., Neri, R., et al. 2010, Nature, 463, 781
Tacconi, L. J., Genzel, R., & Sternberg, A. 2020, ARA&A, 58, 157
Teyssier, R. 2002, A&A, 385, 337
Thompson, T. A., Quataert, E., Zhang, D., & Weinberg, D. H. 2016, MNRAS,
455, 1830
Torrey, P., Vogelsberger, M., Marinacci, F., et al. 2019, MNRAS, 484, 5587
Tremonti, C. A., Heckman, T. M., Kauffmann, G., et al. 2004, ApJ, 613, 898
Treu, T., Roberts-Borsani, G., Bradac, M., et al. 2022, ApJ, 935, 110
Ucci, G., Dayal, P., Hutter, A., et al. 2023, MNRAS, 518, 3557
Vallini, L., Ferrara, A., Pallottini, A., Carniani, S., & Gallerani, S. 2020, MN-
RAS, 495, L22
Vallini, L., Ferrara, A., Pallottini, A., Carniani, S., & Gallerani, S. 2021, MN-
RAS, 505, 5543
Vallini, L., Witstok, J., Sommovigo, L., et al. 2024, MNRAS, 527, 10
van der Walt, S., Colbert, S. C., & Varoquaux, G. 2011, Computing in Science
Engineering, 13, 22
Van Rossum, G. & de Boer, J. 1991, CWI Quarterly, 4, 283
Van Rossum, G. & Drake, F. L. 2009, Python 3 Reference Manual (Scotts Valley,
CA: CreateSpace)
Venturi, G., Carniani, S., Parlanti, E., et al. 2024, arXiv e-prints,
arXiv:2403.03977
Vikaeus, A., Whalen, D. J., & Zackrisson, E. 2022, ApJL, 933, L8
Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261
Wetzel, A., Hayward, C. C., Sanderson, R. E., et al. 2023, ApJS, 265, 44
Wilkins, S. M., Vijayan, A. P., Lovell, C. C., et al. 2023, MNRAS, 519, 3118
York, D. G., Adelman, J., Anderson, John E., J., et al. 2000, AJ, 120, 1579
Zahid, H. J., Dima, G. I., Kudritzki, R.-P., et al. 2014, ApJ, 791, 130
Zahid, H. J., Kewley, L. J., & Bresolin, F. 2011, ApJ, 730, 137
Zanella, A., Pallottini, A., Ferrara, A., et al. 2021, MNRAS, 500, 118

Article number, page 11 of 13

 A&A proofs: manuscript no. metal_jwst

Appendix A: Modifications to the SFR time scales

In bathtub models (Lilly et al. 2013), the shape of the MZR is
mostly independent from the functional form of the SFR, as the
equilibrium solution is determined by the balance between in-
fall and outflow, while the SFR uniquely determines the speed a
model climbs the MZR from low to high M⋆ . Instead, the SFR
prescription can be important if a stochastic behavior is present
in the system and when analyzing the FMR.
In the fiducial model presented in the paper we assumed that
the SFR is proportional to the gas mass and is regulated by a
fixed time scale for star formation, i.e. t⋆ = 1 Gyr. As a nat-
ural alternatively for a fixed t⋆ , a model can consider to adopt
t⋆ via empirical determinations (Tacconi et al. 2010) or model-
based predictions (Sommovigo et al. 2022) of the depletion time,
which is expected to decrease with redshift, thus giving a faster
star formation rate at higher z.
As a test, we modify our models by changing t⋆ in our SFR
prescription (eq. 2b) by adopting the scaling suggested by Som-
movigo et al. (2022, see eq. 16 therein)
8.95 Gyr
t⋆mod ≃ (A.1)
E(z)
8.95 Gyr
= ,
[−0.24 + 0.75(1 + z)] Ωm (1 + z)3 + ΩΛ
p

and setting the proportionality constant in order to match Tac-

coni et al. (2010) determinations at z = 0. As a reference,
t⋆ ≃ 20 t⋆mod at z ≃ 10, thus a faster stellar mass build is expected
for the modified model.
As a benchmark, we select the halo with Mdm = 3 × 1010 M⊙
at z = 6 analyzed in Sec. 3.2 (see Fig. 2), evolve it with t⋆ =
1 Gyr and t⋆mod from (eq. A.1), and plot the results in Fig. A.1 for
both td = 10 Myr and a reference no-SN-delay case.
The largest differences between the t⋆ = 1 Gyr and t⋆mod
models are present for the stellar mass build up at z ∼ > 10, when

the galaxy has M⋆ ∼ < 107 M , The tmod model promotes a faster
⊙ ⋆
build up, particularly when a SN-delay is present, as the lack
of feedback combined with the short SFR timescales make he
galaxy reach M⋆ ∼ < 106 M during the first burst. As the time
⊙
progresses, differences between the models tend to wash out,
as the galaxy approaches M⋆ ≃ 108 M⊙ , the SFH starts to be-
come feedback-regulated, and at z ≃ 6, stellar, gas, and metal
masses are roughly similar in all cases. Interestingly, while the
amplitude of the flickering is similar (σSFR ≃ 0.5), the modula-
tion acts on slightly shorter time scales for the t⋆mod model, as a
consequence of the faster SFR rise and the heavier off-balance
caused by the stellar feedback. This implies that the t⋆mod model
< 7.5 gives an even enhanced dispersion for the MZR (from
at z ∼
σZ ≃ 0.7 to σZ ≃ 1.7). Note that when no-SN-delay is present,
the SFH evolution is feedback regulated from the start, and the
main difference is an increased gas fraction for the t⋆ = 1 Gyr
model.
Note that adopting Sommovigo et al. (2022) implies a deple-
tion time with a ∝ z−2.5 redshift dependence, i.e. steeper with re-
spect to the ∝ z−1.5 dependence from Tacconi et al. (2010). Thus,
by assuming Tacconi et al. (2010) instead of Sommovigo et al.
(2022) to modify the SFR timescale, we expect smaller variation
with respect to the fiducial t⋆ = 1 Gyr model.

Article number, page 12 of 13

 Pallottini et al.: The MZR as a ruler for galaxy evolution: insights from the JWST

Fig. A.1: Difference in the evolution of a Mdm = 3 × 1010 M⊙ computed by prescribing the fiducial (t⋆ = 1 Gry, left panel) and faster
(t⋆mod , eq. A.1, right panel) star formation timescale (eq. 2b). For both panels, Dark matter, stellar, gas, and metal mass evolution is
reported for both the no-SN-delay and td = 10 Myr model with the same notation of as in Fig. 2. Note that the left panel is exactly
Fig. 2, it is reported here to have a clearer visual comparison.

Article number, page 13 of 13